diff --git a/OpenCL/inc_common.cl b/OpenCL/inc_common.cl
index 82b50b7c8..95d82caa4 100644
--- a/OpenCL/inc_common.cl
+++ b/OpenCL/inc_common.cl
@@ -1483,7 +1483,7 @@ DECLSPEC u32 hc_bytealign_S (const u32 a, const u32 b, const int c)
   return r;
 }
 
-#if HAS_VPERM
+#if HAS_VPERM == 1
 DECLSPEC u32x hc_byte_perm (const u32x a, const u32x b, const int c)
 {
   u32x r = 0;
@@ -1533,7 +1533,7 @@ DECLSPEC u32 hc_byte_perm_S (const u32 a, const u32 b, const int c)
 }
 #endif
 
-#if HAS_VADD3
+#if HAS_VADD3 == 1
 DECLSPEC u32x hc_add3 (const u32x a, const u32x b, const u32x c)
 {
   /*
@@ -1711,6 +1711,15 @@ DECLSPEC u32 hc_bfe_S (const u32 a, const u32 b, const u32 c)
   return r;
 }
 
+DECLSPEC u32 hc_bytealign_be_S (const u32 a, const u32 b, const int c)
+{
+  const int c_mod_4 = c & 3;
+
+  const u32 r = hc_byte_perm_S (b, a, (0x76543210 >> (c_mod_4 * 4)) & 0xffff);
+
+  return r;
+}
+
 DECLSPEC u32x hc_bytealign (const u32x a, const u32x b, const int c)
 {
   const int c_mod_4 = c & 3;
@@ -2056,6 +2065,152 @@ DECLSPEC int hc_enc_has_next (hc_enc_t *hc_enc, const int sz)
   return 0;
 }
 
+DECLSPEC int hc_enc_validate_utf8 (const u32 *src_buf, const int src_pos, const int extraBytesToRead)
+{
+  const u8 *src_ptr = (const u8 *) src_buf;
+
+  if (extraBytesToRead == 0)
+  {
+    const u8 c0 = src_ptr[src_pos + 0]; if (c0 >= 0x80) return 0;
+  }
+  else if (extraBytesToRead == 1)
+  {
+    const u8 c0 = src_ptr[src_pos + 0]; if ((c0 < 0xc2) || (c0 > 0xdf)) return 0;
+    const u8 c1 = src_ptr[src_pos + 1]; if ((c1 < 0x80) || (c1 > 0xbf)) return 0;
+  }
+  else if (extraBytesToRead == 2)
+  {
+    const u8 c0 = src_ptr[src_pos + 0];
+
+    if ((c0 >= 0xe0) && (c0 <= 0xe0))
+    {
+      const u8 c1 = src_ptr[src_pos + 1]; if ((c1 < 0xa0) || (c1 > 0xbf)) return 0;
+      const u8 c2 = src_ptr[src_pos + 2]; if ((c2 < 0x80) || (c2 > 0xbf)) return 0;
+    }
+    else if ((c0 >= 0xe1) && (c0 <= 0xec))
+    {
+      const u8 c1 = src_ptr[src_pos + 1]; if ((c1 < 0x80) || (c1 > 0xbf)) return 0;
+      const u8 c2 = src_ptr[src_pos + 2]; if ((c2 < 0x80) || (c2 > 0xbf)) return 0;
+    }
+    else if ((c0 >= 0xed) && (c0 <= 0xed))
+    {
+      const u8 c1 = src_ptr[src_pos + 1]; if ((c1 < 0x80) || (c1 > 0x9f)) return 0;
+      const u8 c2 = src_ptr[src_pos + 2]; if ((c2 < 0x80) || (c2 > 0xbf)) return 0;
+    }
+    else if ((c0 >= 0xee) && (c0 <= 0xef))
+    {
+      const u8 c1 = src_ptr[src_pos + 1]; if ((c1 < 0x80) || (c1 > 0xbf)) return 0;
+      const u8 c2 = src_ptr[src_pos + 2]; if ((c2 < 0x80) || (c2 > 0xbf)) return 0;
+    }
+    else
+    {
+      return 0;
+    }
+  }
+  else if (extraBytesToRead == 3)
+  {
+    const u8 c0 = src_ptr[src_pos + 0];
+
+    if ((c0 >= 0xf0) && (c0 <= 0xf0))
+    {
+      const u8 c1 = src_ptr[src_pos + 1]; if ((c1 < 0x90) || (c1 > 0xbf)) return 0;
+      const u8 c2 = src_ptr[src_pos + 2]; if ((c2 < 0x80) || (c2 > 0xbf)) return 0;
+      const u8 c3 = src_ptr[src_pos + 3]; if ((c3 < 0x80) || (c3 > 0xbf)) return 0;
+    }
+    else if ((c0 >= 0xf1) && (c0 <= 0xf3))
+    {
+      const u8 c1 = src_ptr[src_pos + 1]; if ((c1 < 0x80) || (c1 > 0xbf)) return 0;
+      const u8 c2 = src_ptr[src_pos + 2]; if ((c2 < 0x80) || (c2 > 0xbf)) return 0;
+      const u8 c3 = src_ptr[src_pos + 3]; if ((c3 < 0x80) || (c3 > 0xbf)) return 0;
+    }
+    else if ((c0 >= 0xf4) && (c0 <= 0xf4))
+    {
+      const u8 c1 = src_ptr[src_pos + 1]; if ((c1 < 0x80) || (c1 > 0xbf)) return 0;
+      const u8 c2 = src_ptr[src_pos + 2]; if ((c2 < 0x80) || (c2 > 0xbf)) return 0;
+      const u8 c3 = src_ptr[src_pos + 3]; if ((c3 < 0x80) || (c3 > 0xbf)) return 0;
+    }
+    else
+    {
+      return 0;
+    }
+  }
+
+  return 1;
+}
+
+DECLSPEC int hc_enc_validate_utf8_global (GLOBAL_AS const u32 *src_buf, const int src_pos, const int extraBytesToRead)
+{
+  GLOBAL_AS const u8 *src_ptr = (GLOBAL_AS const u8 *) src_buf;
+
+  if (extraBytesToRead == 0)
+  {
+    const u8 c0 = src_ptr[src_pos + 0]; if (c0 >= 0x80) return 0;
+  }
+  else if (extraBytesToRead == 1)
+  {
+    const u8 c0 = src_ptr[src_pos + 0]; if ((c0 < 0xc2) || (c0 > 0xdf)) return 0;
+    const u8 c1 = src_ptr[src_pos + 1]; if ((c1 < 0x80) || (c1 > 0xbf)) return 0;
+  }
+  else if (extraBytesToRead == 2)
+  {
+    const u8 c0 = src_ptr[src_pos + 0];
+
+    if ((c0 >= 0xe0) && (c0 <= 0xe0))
+    {
+      const u8 c1 = src_ptr[src_pos + 1]; if ((c1 < 0xa0) || (c1 > 0xbf)) return 0;
+      const u8 c2 = src_ptr[src_pos + 2]; if ((c2 < 0x80) || (c2 > 0xbf)) return 0;
+    }
+    else if ((c0 >= 0xe1) && (c0 <= 0xec))
+    {
+      const u8 c1 = src_ptr[src_pos + 1]; if ((c1 < 0x80) || (c1 > 0xbf)) return 0;
+      const u8 c2 = src_ptr[src_pos + 2]; if ((c2 < 0x80) || (c2 > 0xbf)) return 0;
+    }
+    else if ((c0 >= 0xed) && (c0 <= 0xed))
+    {
+      const u8 c1 = src_ptr[src_pos + 1]; if ((c1 < 0x80) || (c1 > 0x9f)) return 0;
+      const u8 c2 = src_ptr[src_pos + 2]; if ((c2 < 0x80) || (c2 > 0xbf)) return 0;
+    }
+    else if ((c0 >= 0xee) && (c0 <= 0xef))
+    {
+      const u8 c1 = src_ptr[src_pos + 1]; if ((c1 < 0x80) || (c1 > 0xbf)) return 0;
+      const u8 c2 = src_ptr[src_pos + 2]; if ((c2 < 0x80) || (c2 > 0xbf)) return 0;
+    }
+    else
+    {
+      return 0;
+    }
+  }
+  else if (extraBytesToRead == 3)
+  {
+    const u8 c0 = src_ptr[src_pos + 0];
+
+    if ((c0 >= 0xf0) && (c0 <= 0xf0))
+    {
+      const u8 c1 = src_ptr[src_pos + 1]; if ((c1 < 0x90) || (c1 > 0xbf)) return 0;
+      const u8 c2 = src_ptr[src_pos + 2]; if ((c2 < 0x80) || (c2 > 0xbf)) return 0;
+      const u8 c3 = src_ptr[src_pos + 3]; if ((c3 < 0x80) || (c3 > 0xbf)) return 0;
+    }
+    else if ((c0 >= 0xf1) && (c0 <= 0xf3))
+    {
+      const u8 c1 = src_ptr[src_pos + 1]; if ((c1 < 0x80) || (c1 > 0xbf)) return 0;
+      const u8 c2 = src_ptr[src_pos + 2]; if ((c2 < 0x80) || (c2 > 0xbf)) return 0;
+      const u8 c3 = src_ptr[src_pos + 3]; if ((c3 < 0x80) || (c3 > 0xbf)) return 0;
+    }
+    else if ((c0 >= 0xf4) && (c0 <= 0xf4))
+    {
+      const u8 c1 = src_ptr[src_pos + 1]; if ((c1 < 0x80) || (c1 > 0xbf)) return 0;
+      const u8 c2 = src_ptr[src_pos + 2]; if ((c2 < 0x80) || (c2 > 0xbf)) return 0;
+      const u8 c3 = src_ptr[src_pos + 3]; if ((c3 < 0x80) || (c3 > 0xbf)) return 0;
+    }
+    else
+    {
+      return 0;
+    }
+  }
+
+  return 1;
+}
+
 // Input buffer and Output buffer size has to be multiple of 4 and at least of size 4.
 // The output buffer is not zero padded, so entire buffer has to be set all zero before entering this function or truncated afterwards.
 
@@ -2066,6 +2221,12 @@ DECLSPEC int hc_enc_next (hc_enc_t *hc_enc, const u32 *src_buf, const int src_le
 
   int src_pos = hc_enc->pos;
 
+  #if VENDOR_ID == 8
+  // Work around segmentation fault in Intel JiT
+  // Tested with 2021.12.6.0.19_160000
+  volatile
+  #endif
+
   int dst_pos = hc_enc->clen;
 
   dst_buf[0] = hc_enc->cbuf;
@@ -2109,6 +2270,15 @@ DECLSPEC int hc_enc_next (hc_enc_t *hc_enc, const u32 *src_buf, const int src_le
       return dst_pos;
     }
 
+    if (hc_enc_validate_utf8 (src_buf, src_pos, extraBytesToRead) == 0)
+    {
+      // broken input
+
+      hc_enc->pos = src_len;
+
+      return dst_pos;
+    }
+
     u32 ch = 0;
 
     switch (extraBytesToRead)
@@ -2197,6 +2367,12 @@ DECLSPEC int hc_enc_next_global (hc_enc_t *hc_enc, GLOBAL_AS const u32 *src_buf,
 
   int src_pos = hc_enc->pos;
 
+  #if VENDOR_ID == 8
+  // Work around segmentation fault in Intel JiT
+  // Tested with 2021.12.6.0.19_160000
+  volatile
+  #endif
+
   int dst_pos = hc_enc->clen;
 
   dst_buf[0] = hc_enc->cbuf;
@@ -2240,6 +2416,15 @@ DECLSPEC int hc_enc_next_global (hc_enc_t *hc_enc, GLOBAL_AS const u32 *src_buf,
       return dst_pos;
     }
 
+    if (hc_enc_validate_utf8_global (src_buf, src_pos, extraBytesToRead) == 0)
+    {
+      // broken input
+
+      hc_enc->pos = src_len;
+
+      return dst_pos;
+    }
+
     u32 ch = 0;
 
     switch (extraBytesToRead)
@@ -2300,6 +2485,8 @@ DECLSPEC int hc_enc_next_global (hc_enc_t *hc_enc, GLOBAL_AS const u32 *src_buf,
 
       if ((dst_pos + 2) == dst_sz)
       {
+        // this section seems to break intel opencl runtime but is unknown why
+
         dst_ptr[dst_pos++] = (a >> 0) & 0xff;
         dst_ptr[dst_pos++] = (a >> 8) & 0xff;
 
@@ -2767,7 +2954,7 @@ DECLSPEC void make_utf16be (const u32x *in, u32x *out1, u32x *out2)
   out1[1] = hc_byte_perm (in[0], 0, 0x3727);
   out1[0] = hc_byte_perm (in[0], 0, 0x1707);
 
-  #elif (defined IS_AMD || defined IS_HIP) && defined HAS_VPERM
+  #elif (defined IS_AMD || defined IS_HIP) && defined HAS_VPERM == 1
 
   out2[3] = hc_byte_perm (in[3], 0, 0x03070207);
   out2[2] = hc_byte_perm (in[3], 0, 0x01070007);
@@ -2805,7 +2992,7 @@ DECLSPEC void make_utf16beN (const u32x *in, u32x *out1, u32x *out2)
   out1[1] = hc_byte_perm (in[0], 0, 0x1707);
   out1[0] = hc_byte_perm (in[0], 0, 0x3727);
 
-  #elif (defined IS_AMD || defined IS_HIP) && defined HAS_VPERM
+  #elif (defined IS_AMD || defined IS_HIP) && defined HAS_VPERM == 1
 
   out2[3] = hc_byte_perm (in[3], 0, 0x01070007);
   out2[2] = hc_byte_perm (in[3], 0, 0x03070207);
@@ -2843,7 +3030,7 @@ DECLSPEC void make_utf16le (const u32x *in, u32x *out1, u32x *out2)
   out1[1] = hc_byte_perm (in[0], 0, 0x7372);
   out1[0] = hc_byte_perm (in[0], 0, 0x7170);
 
-  #elif (defined IS_AMD || defined IS_HIP) && defined HAS_VPERM
+  #elif (defined IS_AMD || defined IS_HIP) && defined HAS_VPERM == 1
 
   out2[3] = hc_byte_perm (in[3], 0, 0x07030702);
   out2[2] = hc_byte_perm (in[3], 0, 0x07010700);
@@ -2881,7 +3068,7 @@ DECLSPEC void make_utf16leN (const u32x *in, u32x *out1, u32x *out2)
   out1[1] = hc_byte_perm (in[0], 0, 0x7170);
   out1[0] = hc_byte_perm (in[0], 0, 0x7372);
 
-  #elif (defined IS_AMD || defined IS_HIP) && defined HAS_VPERM
+  #elif (defined IS_AMD || defined IS_HIP) && defined HAS_VPERM == 1
 
   out2[3] = hc_byte_perm (in[3], 0, 0x07010700);
   out2[2] = hc_byte_perm (in[3], 0, 0x07030702);
@@ -2915,7 +3102,7 @@ DECLSPEC void undo_utf16be (const u32x *in1, const u32x *in2, u32x *out)
   out[2] = hc_byte_perm (in2[0], in2[1], 0x4602);
   out[3] = hc_byte_perm (in2[2], in2[3], 0x4602);
 
-  #elif (defined IS_AMD || defined IS_HIP) && defined HAS_VPERM
+  #elif (defined IS_AMD || defined IS_HIP) && defined HAS_VPERM == 1
 
   out[0] = hc_byte_perm (in1[0], in1[1], 0x04060002);
   out[1] = hc_byte_perm (in1[2], in1[3], 0x04060002);
@@ -2945,7 +3132,7 @@ DECLSPEC void undo_utf16le (const u32x *in1, const u32x *in2, u32x *out)
   out[2] = hc_byte_perm (in2[0], in2[1], 0x6420);
   out[3] = hc_byte_perm (in2[2], in2[3], 0x6420);
 
-  #elif (defined IS_AMD || defined IS_HIP) && defined HAS_VPERM
+  #elif (defined IS_AMD || defined IS_HIP) && defined HAS_VPERM == 1
 
   out[0] = hc_byte_perm (in1[0], in1[1], 0x06040200);
   out[1] = hc_byte_perm (in1[2], in1[3], 0x06040200);
@@ -36454,6 +36641,20 @@ DECLSPEC void append_0x01_4x4_S (u32 *w0, u32 *w1, u32 *w2, u32 *w3, const u32 o
   append_helper_1x4_S (w3, ((offset16 == 3) ? 0x01010101 : 0), v);
 }
 
+DECLSPEC void append_0x2d_4x4_S (u32 *w0, u32 *w1, u32 *w2, u32 *w3, const u32 offset)
+{
+  u32 v[4];
+
+  set_mark_1x4_S (v, offset);
+
+  const u32 offset16 = offset / 16;
+
+  append_helper_1x4_S (w0, ((offset16 == 0) ? 0x2d2d2d2d : 0), v);
+  append_helper_1x4_S (w1, ((offset16 == 1) ? 0x2d2d2d2d : 0), v);
+  append_helper_1x4_S (w2, ((offset16 == 2) ? 0x2d2d2d2d : 0), v);
+  append_helper_1x4_S (w3, ((offset16 == 3) ? 0x2d2d2d2d : 0), v);
+}
+
 DECLSPEC void append_0x80_1x4_S (u32 *w0, const u32 offset)
 {
   u32 v[4];
@@ -36533,7 +36734,7 @@ DECLSPEC void make_utf16be_S (const u32 *in, u32 *out1, u32 *out2)
   out1[1] = hc_byte_perm_S (in[0], 0, 0x3727);
   out1[0] = hc_byte_perm_S (in[0], 0, 0x1707);
 
-  #elif (defined IS_AMD || defined IS_HIP) && HAS_VPERM
+  #elif (defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1
 
   out2[3] = hc_byte_perm_S (in[3], 0, 0x03070207);
   out2[2] = hc_byte_perm_S (in[3], 0, 0x01070007);
@@ -36571,7 +36772,7 @@ DECLSPEC void make_utf16le_S (const u32 *in, u32 *out1, u32 *out2)
   out1[1] = hc_byte_perm_S (in[0], 0, 0x7372);
   out1[0] = hc_byte_perm_S (in[0], 0, 0x7170);
 
-  #elif (defined IS_AMD || defined IS_HIP) && HAS_VPERM
+  #elif (defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1
 
   out2[3] = hc_byte_perm_S (in[3], 0, 0x07030702);
   out2[2] = hc_byte_perm_S (in[3], 0, 0x07010700);
@@ -36605,7 +36806,7 @@ DECLSPEC void undo_utf16be_S (const u32 *in1, const u32 *in2, u32 *out)
   out[2] = hc_byte_perm_S (in2[0], in2[1], 0x4602);
   out[3] = hc_byte_perm_S (in2[2], in2[3], 0x4602);
 
-  #elif (defined IS_AMD || defined IS_HIP) && HAS_VPERM
+  #elif (defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1
 
   out[0] = hc_byte_perm_S (in1[0], in1[1], 0x04060002);
   out[1] = hc_byte_perm_S (in1[2], in1[3], 0x04060002);
@@ -36635,7 +36836,7 @@ DECLSPEC void undo_utf16le_S (const u32 *in1, const u32 *in2, u32 *out)
   out[2] = hc_byte_perm_S (in2[0], in2[1], 0x6420);
   out[3] = hc_byte_perm_S (in2[2], in2[3], 0x6420);
 
-  #elif (defined IS_AMD || defined IS_HIP) && HAS_VPERM
+  #elif (defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1
 
   out[0] = hc_byte_perm_S (in1[0], in1[1], 0x06040200);
   out[1] = hc_byte_perm_S (in1[2], in1[3], 0x06040200);
@@ -68749,3 +68950,63 @@ DECLSPEC void append_0x80_4x4_VV (u32x *w0, u32x *w1, u32x *w2, u32x *w3, const
 
   #endif
 }
+
+DECLSPEC void append_0x2d_4x4_VV (u32x *w0, u32x *w1, u32x *w2, u32x *w3, const u32x offset)
+{
+  #if VECT_SIZE == 1
+
+  append_0x2d_4x4_S (w0, w1, w2, w3, offset);
+
+  #else
+
+  u32 t0[4];
+  u32 t1[4];
+  u32 t2[4];
+  u32 t3[4];
+
+  #endif
+
+  #if   VECT_SIZE == 2
+
+  PACKVS44 (t0, t1, t2, t3, w0, w1, w2, w3, 0); append_0x2d_4x4_S (t0, t1, t2, t3, offset.s0); PACKSV44 (t0, t1, t2, t3, w0, w1, w2, w3, 0);
+  PACKVS44 (t0, t1, t2, t3, w0, w1, w2, w3, 1); append_0x2d_4x4_S (t0, t1, t2, t3, offset.s1); PACKSV44 (t0, t1, t2, t3, w0, w1, w2, w3, 1);
+
+  #elif VECT_SIZE == 4
+
+  PACKVS44 (t0, t1, t2, t3, w0, w1, w2, w3, 0); append_0x2d_4x4_S (t0, t1, t2, t3, offset.s0); PACKSV44 (t0, t1, t2, t3, w0, w1, w2, w3, 0);
+  PACKVS44 (t0, t1, t2, t3, w0, w1, w2, w3, 1); append_0x2d_4x4_S (t0, t1, t2, t3, offset.s1); PACKSV44 (t0, t1, t2, t3, w0, w1, w2, w3, 1);
+  PACKVS44 (t0, t1, t2, t3, w0, w1, w2, w3, 2); append_0x2d_4x4_S (t0, t1, t2, t3, offset.s2); PACKSV44 (t0, t1, t2, t3, w0, w1, w2, w3, 2);
+  PACKVS44 (t0, t1, t2, t3, w0, w1, w2, w3, 3); append_0x2d_4x4_S (t0, t1, t2, t3, offset.s3); PACKSV44 (t0, t1, t2, t3, w0, w1, w2, w3, 3);
+
+  #elif VECT_SIZE == 8
+
+  PACKVS44 (t0, t1, t2, t3, w0, w1, w2, w3, 0); append_0x2d_4x4_S (t0, t1, t2, t3, offset.s0); PACKSV44 (t0, t1, t2, t3, w0, w1, w2, w3, 0);
+  PACKVS44 (t0, t1, t2, t3, w0, w1, w2, w3, 1); append_0x2d_4x4_S (t0, t1, t2, t3, offset.s1); PACKSV44 (t0, t1, t2, t3, w0, w1, w2, w3, 1);
+  PACKVS44 (t0, t1, t2, t3, w0, w1, w2, w3, 2); append_0x2d_4x4_S (t0, t1, t2, t3, offset.s2); PACKSV44 (t0, t1, t2, t3, w0, w1, w2, w3, 2);
+  PACKVS44 (t0, t1, t2, t3, w0, w1, w2, w3, 3); append_0x2d_4x4_S (t0, t1, t2, t3, offset.s3); PACKSV44 (t0, t1, t2, t3, w0, w1, w2, w3, 3);
+  PACKVS44 (t0, t1, t2, t3, w0, w1, w2, w3, 4); append_0x2d_4x4_S (t0, t1, t2, t3, offset.s4); PACKSV44 (t0, t1, t2, t3, w0, w1, w2, w3, 4);
+  PACKVS44 (t0, t1, t2, t3, w0, w1, w2, w3, 5); append_0x2d_4x4_S (t0, t1, t2, t3, offset.s5); PACKSV44 (t0, t1, t2, t3, w0, w1, w2, w3, 5);
+  PACKVS44 (t0, t1, t2, t3, w0, w1, w2, w3, 6); append_0x2d_4x4_S (t0, t1, t2, t3, offset.s6); PACKSV44 (t0, t1, t2, t3, w0, w1, w2, w3, 6);
+  PACKVS44 (t0, t1, t2, t3, w0, w1, w2, w3, 7); append_0x2d_4x4_S (t0, t1, t2, t3, offset.s7); PACKSV44 (t0, t1, t2, t3, w0, w1, w2, w3, 7);
+
+  #elif VECT_SIZE == 16
+
+  PACKVS44 (t0, t1, t2, t3, w0, w1, w2, w3, 0); append_0x2d_4x4_S (t0, t1, t2, t3, offset.s0); PACKSV44 (t0, t1, t2, t3, w0, w1, w2, w3, 0);
+  PACKVS44 (t0, t1, t2, t3, w0, w1, w2, w3, 1); append_0x2d_4x4_S (t0, t1, t2, t3, offset.s1); PACKSV44 (t0, t1, t2, t3, w0, w1, w2, w3, 1);
+  PACKVS44 (t0, t1, t2, t3, w0, w1, w2, w3, 2); append_0x2d_4x4_S (t0, t1, t2, t3, offset.s2); PACKSV44 (t0, t1, t2, t3, w0, w1, w2, w3, 2);
+  PACKVS44 (t0, t1, t2, t3, w0, w1, w2, w3, 3); append_0x2d_4x4_S (t0, t1, t2, t3, offset.s3); PACKSV44 (t0, t1, t2, t3, w0, w1, w2, w3, 3);
+  PACKVS44 (t0, t1, t2, t3, w0, w1, w2, w3, 4); append_0x2d_4x4_S (t0, t1, t2, t3, offset.s4); PACKSV44 (t0, t1, t2, t3, w0, w1, w2, w3, 4);
+  PACKVS44 (t0, t1, t2, t3, w0, w1, w2, w3, 5); append_0x2d_4x4_S (t0, t1, t2, t3, offset.s5); PACKSV44 (t0, t1, t2, t3, w0, w1, w2, w3, 5);
+  PACKVS44 (t0, t1, t2, t3, w0, w1, w2, w3, 6); append_0x2d_4x4_S (t0, t1, t2, t3, offset.s6); PACKSV44 (t0, t1, t2, t3, w0, w1, w2, w3, 6);
+  PACKVS44 (t0, t1, t2, t3, w0, w1, w2, w3, 7); append_0x2d_4x4_S (t0, t1, t2, t3, offset.s7); PACKSV44 (t0, t1, t2, t3, w0, w1, w2, w3, 7);
+  PACKVS44 (t0, t1, t2, t3, w0, w1, w2, w3, 8); append_0x2d_4x4_S (t0, t1, t2, t3, offset.s8); PACKSV44 (t0, t1, t2, t3, w0, w1, w2, w3, 8);
+  PACKVS44 (t0, t1, t2, t3, w0, w1, w2, w3, 9); append_0x2d_4x4_S (t0, t1, t2, t3, offset.s9); PACKSV44 (t0, t1, t2, t3, w0, w1, w2, w3, 9);
+  PACKVS44 (t0, t1, t2, t3, w0, w1, w2, w3, a); append_0x2d_4x4_S (t0, t1, t2, t3, offset.sa); PACKSV44 (t0, t1, t2, t3, w0, w1, w2, w3, a);
+  PACKVS44 (t0, t1, t2, t3, w0, w1, w2, w3, b); append_0x2d_4x4_S (t0, t1, t2, t3, offset.sb); PACKSV44 (t0, t1, t2, t3, w0, w1, w2, w3, b);
+  PACKVS44 (t0, t1, t2, t3, w0, w1, w2, w3, c); append_0x2d_4x4_S (t0, t1, t2, t3, offset.sc); PACKSV44 (t0, t1, t2, t3, w0, w1, w2, w3, c);
+  PACKVS44 (t0, t1, t2, t3, w0, w1, w2, w3, d); append_0x2d_4x4_S (t0, t1, t2, t3, offset.sd); PACKSV44 (t0, t1, t2, t3, w0, w1, w2, w3, d);
+  PACKVS44 (t0, t1, t2, t3, w0, w1, w2, w3, e); append_0x2d_4x4_S (t0, t1, t2, t3, offset.se); PACKSV44 (t0, t1, t2, t3, w0, w1, w2, w3, e);
+  PACKVS44 (t0, t1, t2, t3, w0, w1, w2, w3, f); append_0x2d_4x4_S (t0, t1, t2, t3, offset.sf); PACKSV44 (t0, t1, t2, t3, w0, w1, w2, w3, f);
+
+  #endif
+}
\ No newline at end of file
diff --git a/OpenCL/inc_common.h b/OpenCL/inc_common.h
index ebd0107c1..dd0f3560c 100644
--- a/OpenCL/inc_common.h
+++ b/OpenCL/inc_common.h
@@ -290,6 +290,7 @@ DECLSPEC void append_helper_1x4_S (u32 *r, const u32 v, const u32 *m);
 DECLSPEC void append_0x01_2x4_S (u32 *w0, u32 *w1, const u32 offset);
 DECLSPEC void append_0x06_2x4_S (u32 *w0, u32 *w1, const u32 offset);
 DECLSPEC void append_0x01_4x4_S (u32 *w0, u32 *w1, u32 *w2, u32 *w3, const u32 offset);
+DECLSPEC void append_0x2d_4x4_S (u32 *w0, u32 *w1, u32 *w2, u32 *w3, const u32 offset);
 DECLSPEC void append_0x80_1x4_S (u32 *w0, const u32 offset);
 DECLSPEC void append_0x80_2x4_S (u32 *w0, u32 *w1, const u32 offset);
 DECLSPEC void append_0x80_3x4_S (u32 *w0, u32 *w1, u32 *w2, const u32 offset);
@@ -316,5 +317,6 @@ DECLSPEC void append_0x01_4x4_VV (u32x *w0, u32x *w1, u32x *w2, u32x *w3, const
 DECLSPEC void append_0x06_2x4_VV (u32x *w0, u32x *w1, const u32x offset);
 DECLSPEC void append_0x80_2x4_VV (u32x *w0, u32x *w1, const u32x offset);
 DECLSPEC void append_0x80_4x4_VV (u32x *w0, u32x *w1, u32x *w2, u32x *w3, const u32x offset);
+DECLSPEC void append_0x2d_4x4_VV (u32x *w0, u32x *w1, u32x *w2, u32x *w3, const u32x offset);
 
 #endif
diff --git a/OpenCL/inc_ecc_secp256k1.cl b/OpenCL/inc_ecc_secp256k1.cl
index b3a70df78..a487152ec 100644
--- a/OpenCL/inc_ecc_secp256k1.cl
+++ b/OpenCL/inc_ecc_secp256k1.cl
@@ -124,7 +124,9 @@ DECLSPEC u32 sub (u32 *r, const u32 *a, const u32 *b)
     :  "r"(a[0]),  "r"(a[1]),  "r"(a[2]),  "r"(a[3]),  "r"(a[4]),  "r"(a[5]),  "r"(a[6]),  "r"(a[7]),
        "r"(b[0]),  "r"(b[1]),  "r"(b[2]),  "r"(b[3]),  "r"(b[4]),  "r"(b[5]),  "r"(b[6]),  "r"(b[7])
   );
-  #elif (defined IS_AMD || defined IS_HIP) && HAS_VSUB == 1 && HAS_VSUBB == 1
+  // HIP doesnt support these so we stick to OpenCL (aka IS_AMD) - is also faster without asm
+  //#elif (defined IS_AMD || defined IS_HIP) && HAS_VSUB == 1 && HAS_VSUBB == 1
+  #elif 0
   __asm__ __volatile__
   (
     "V_SUB_U32   %0,  %9, %17;"
@@ -176,7 +178,9 @@ DECLSPEC u32 add (u32 *r, const u32 *a, const u32 *b)
     :  "r"(a[0]),  "r"(a[1]),  "r"(a[2]),  "r"(a[3]),  "r"(a[4]),  "r"(a[5]),  "r"(a[6]),  "r"(a[7]),
        "r"(b[0]),  "r"(b[1]),  "r"(b[2]),  "r"(b[3]),  "r"(b[4]),  "r"(b[5]),  "r"(b[6]),  "r"(b[7])
   );
-  #elif (defined IS_AMD || defined IS_HIP) && HAS_VADD == 1 && HAS_VADDC == 1
+  // HIP doesnt support these so we stick to OpenCL (aka IS_AMD) - is also faster without asm
+  //#elif (defined IS_AMD || defined IS_HIP) && HAS_VSUB == 1 && HAS_VSUBB == 1
+  #elif 0
   __asm__ __volatile__
   (
     "V_ADD_U32   %0,  %9, %17;"
diff --git a/OpenCL/inc_hash_blake2b.cl b/OpenCL/inc_hash_blake2b.cl
index 33467c933..8bbe0511f 100644
--- a/OpenCL/inc_hash_blake2b.cl
+++ b/OpenCL/inc_hash_blake2b.cl
@@ -24,7 +24,7 @@ DECLSPEC u64 blake2b_rot16_S (const u64 a)
 
   return out.v64;
 
-  #elif (defined IS_AMD || defined IS_HIP) && HAS_VPERM
+  #elif (defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1
 
   vconv64_t in;
 
@@ -98,7 +98,7 @@ DECLSPEC u64 blake2b_rot24_S (const u64 a)
 
   return out.v64;
 
-  #elif (defined IS_AMD || defined IS_HIP) && HAS_VPERM
+  #elif (defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1
 
   vconv64_t in;
 
@@ -113,7 +113,7 @@ DECLSPEC u64 blake2b_rot24_S (const u64 a)
 
   #else
 
-  return hc_rotr64_S (a, 16);
+  return hc_rotr64_S (a, 24);
 
   #endif
 }
diff --git a/OpenCL/inc_hash_sha384.cl b/OpenCL/inc_hash_sha384.cl
index 3d433bb33..625908503 100644
--- a/OpenCL/inc_hash_sha384.cl
+++ b/OpenCL/inc_hash_sha384.cl
@@ -1502,14 +1502,14 @@ DECLSPEC void sha384_hmac_init_128 (sha384_hmac_ctx_t *ctx, const u32 *w0, const
 
   // opad
 
-  u32x b0[4];
-  u32x b1[4];
-  u32x b2[4];
-  u32x b3[4];
-  u32x b4[4];
-  u32x b5[4];
-  u32x b6[4];
-  u32x b7[4];
+  u32 b0[4];
+  u32 b1[4];
+  u32 b2[4];
+  u32 b3[4];
+  u32 b4[4];
+  u32 b5[4];
+  u32 b6[4];
+  u32 b7[4];
 
   b0[0] = w0[0] ^ 0x5c5c5c5c;
   b0[1] = w0[1] ^ 0x5c5c5c5c;
@@ -3009,14 +3009,14 @@ DECLSPEC void sha384_hmac_init_vector_128 (sha384_hmac_ctx_vector_t *ctx, const
 
   // opad
 
-  u32 b0[4];
-  u32 b1[4];
-  u32 b2[4];
-  u32 b3[4];
-  u32 b4[4];
-  u32 b5[4];
-  u32 b6[4];
-  u32 b7[4];
+  u32x b0[4];
+  u32x b1[4];
+  u32x b2[4];
+  u32x b3[4];
+  u32x b4[4];
+  u32x b5[4];
+  u32x b6[4];
+  u32x b7[4];
 
   b0[0] = w0[0] ^ 0x5c5c5c5c;
   b0[1] = w0[1] ^ 0x5c5c5c5c;
diff --git a/OpenCL/inc_platform.h b/OpenCL/inc_platform.h
index c65891a74..916af055a 100644
--- a/OpenCL/inc_platform.h
+++ b/OpenCL/inc_platform.h
@@ -48,9 +48,9 @@ DECLSPEC u32 hc_atomic_dec (volatile GLOBAL_AS u32 *p);
 DECLSPEC u32 hc_atomic_inc (volatile GLOBAL_AS u32 *p);
 DECLSPEC u32 hc_atomic_or  (volatile GLOBAL_AS u32 *p, volatile const u32 val);
 
-DECLSPEC size_t get_global_id   (const u32 dimindx __attribute__((unused)));
-DECLSPEC size_t get_local_id    (const u32 dimindx __attribute__((unused)));
-DECLSPEC size_t get_local_size  (const u32 dimindx __attribute__((unused)));
+DECLSPEC size_t get_global_id   (const u32 dimindx);
+DECLSPEC size_t get_local_id    (const u32 dimindx);
+DECLSPEC size_t get_local_size  (const u32 dimindx);
 
 DECLSPEC u32x rotl32   (const u32x a, const int n);
 DECLSPEC u32x rotr32   (const u32x a, const int n);
diff --git a/OpenCL/inc_rp.cl b/OpenCL/inc_rp.cl
index 8cecc661d..8ad61b27f 100644
--- a/OpenCL/inc_rp.cl
+++ b/OpenCL/inc_rp.cl
@@ -269,6 +269,35 @@ DECLSPEC int mangle_toggle_at (MAYBE_UNUSED const u8 p0, MAYBE_UNUSED const u8 p
   return (len);
 }
 
+DECLSPEC int mangle_toggle_at_sep (MAYBE_UNUSED const u8 p0, MAYBE_UNUSED const u8 p1, u32 *buf, const int len)
+{
+  if (len >= RP_PASSWORD_SIZE) return (len);
+
+  u8 occurence = 0;
+
+  u32 rem = 0;
+
+  for (int i = 0, idx = 0; i < len; i += 4, idx += 1)
+  {
+    const u32 t = buf[idx];
+
+    buf[idx] = t | generate_cmask (t);
+
+    u32 out = rem;
+
+    rem = 0;
+
+    if (((t >>  0) & 0xff) == p1) { if (occurence == p0) out = 0x0000ff00; occurence++; }
+    if (((t >>  8) & 0xff) == p1) { if (occurence == p0) out = 0x00ff0000; occurence++; }
+    if (((t >> 16) & 0xff) == p1) { if (occurence == p0) out = 0xff000000; occurence++; }
+    if (((t >> 24) & 0xff) == p1) { if (occurence == p0) rem = 0x000000ff; occurence++; }
+
+    buf[idx] = t ^ (generate_cmask (t) & out);
+  }
+
+  return (len);
+}
+
 DECLSPEC int mangle_reverse (MAYBE_UNUSED const u8 p0, MAYBE_UNUSED const u8 p1, u32 *buf, const int len)
 {
   for (int l = 0; l < len / 2; l++)
@@ -725,6 +754,7 @@ DECLSPEC int apply_rule (const u32 name, MAYBE_UNUSED const u8 p0, MAYBE_UNUSED
     case RULE_OP_MANGLE_UREST_LFIRST:     out_len = mangle_urest_lfirst     (p0, p1,        buf, out_len); break;
     case RULE_OP_MANGLE_TREST:            out_len = mangle_trest            (p0, p1,        buf, out_len); break;
     case RULE_OP_MANGLE_TOGGLE_AT:        out_len = mangle_toggle_at        (p0, p1,        buf, out_len); break;
+    case RULE_OP_MANGLE_TOGGLE_AT_SEP:    out_len = mangle_toggle_at_sep    (p0, p1,        buf, out_len); break;
     case RULE_OP_MANGLE_REVERSE:          out_len = mangle_reverse          (p0, p1,        buf, out_len); break;
     case RULE_OP_MANGLE_DUPEWORD:         out_len = mangle_dupeword         (p0, p1,        buf, out_len); break;
     case RULE_OP_MANGLE_DUPEWORD_TIMES:   out_len = mangle_dupeword_times   (p0, p1, (u8 *) buf, out_len); break;
diff --git a/OpenCL/inc_rp.h b/OpenCL/inc_rp.h
index c13d06e1f..3b91ff9da 100644
--- a/OpenCL/inc_rp.h
+++ b/OpenCL/inc_rp.h
@@ -21,6 +21,7 @@
 #define RULE_OP_MANGLE_UREST_LFIRST     'C'
 #define RULE_OP_MANGLE_TREST            't'
 #define RULE_OP_MANGLE_TOGGLE_AT        'T'
+#define RULE_OP_MANGLE_TOGGLE_AT_SEP    '3'
 #define RULE_OP_MANGLE_REVERSE          'r'
 #define RULE_OP_MANGLE_DUPEWORD         'd'
 #define RULE_OP_MANGLE_DUPEWORD_TIMES   'p'
@@ -83,6 +84,7 @@ DECLSPEC int mangle_urest (MAYBE_UNUSED const u8 p0, MAYBE_UNUSED const u8 p1, u
 DECLSPEC int mangle_urest_lfirst (MAYBE_UNUSED const u8 p0, MAYBE_UNUSED const u8 p1, u32 *buf, const int len);
 DECLSPEC int mangle_trest (MAYBE_UNUSED const u8 p0, MAYBE_UNUSED const u8 p1, u32 *buf, const int len);
 DECLSPEC int mangle_toggle_at (MAYBE_UNUSED const u8 p0, MAYBE_UNUSED const u8 p1, u32 *buf, const int len);
+DECLSPEC int mangle_toggle_at_sep (MAYBE_UNUSED const u8 p0, MAYBE_UNUSED const u8 p1, u32 *buf, const int len);
 DECLSPEC int mangle_reverse (MAYBE_UNUSED const u8 p0, MAYBE_UNUSED const u8 p1, u32 *buf, const int len);
 DECLSPEC int mangle_dupeword (MAYBE_UNUSED const u8 p0, MAYBE_UNUSED const u8 p1, u32 *buf, const int len);
 DECLSPEC int mangle_dupeword_times (MAYBE_UNUSED const u8 p0, MAYBE_UNUSED const u8 p1, u8 *buf, const int len);
diff --git a/OpenCL/inc_rp_optimized.cl b/OpenCL/inc_rp_optimized.cl
index 026198f09..38239c0ec 100644
--- a/OpenCL/inc_rp_optimized.cl
+++ b/OpenCL/inc_rp_optimized.cl
@@ -1163,6 +1163,82 @@ DECLSPEC u32 rule_op_mangle_toggle_at (MAYBE_UNUSED const u32 p0, MAYBE_UNUSED c
   return (in_len);
 }
 
+DECLSPEC u32 rule_op_mangle_toggle_at_sep (MAYBE_UNUSED const u32 p0, MAYBE_UNUSED const u32 p1, MAYBE_UNUSED u32 *buf0, MAYBE_UNUSED u32 *buf1, const u32 in_len)
+{
+  if (in_len == 0) return in_len;
+
+  u32 r0 = search_on_register (buf0[0], p1);
+  u32 r1 = search_on_register (buf0[1], p1);
+  u32 r2 = search_on_register (buf0[2], p1);
+  u32 r3 = search_on_register (buf0[3], p1);
+  u32 r4 = search_on_register (buf1[0], p1);
+  u32 r5 = search_on_register (buf1[1], p1);
+  u32 r6 = search_on_register (buf1[2], p1);
+  u32 r7 = search_on_register (buf1[3], p1);
+
+  const u32 rn = (r0 <<  0)
+               | (r1 <<  4)
+               | (r2 <<  8)
+               | (r3 << 12)
+               | (r4 << 16)
+               | (r5 << 20)
+               | (r6 << 24)
+               | (r7 << 28);
+
+  if (rn == 0) return in_len;
+
+  u32 occurence = 0;
+
+  u32 ro = 0;
+
+  #ifdef _unroll
+  #pragma unroll
+  #endif
+  for (int i = 0; i < 32; i++)
+  {
+    if ((rn >> i) & 1)
+    {
+      if (occurence == p0)
+      {
+        ro = 1 << i;
+
+        break;
+      }
+
+      occurence++;
+    }
+  }
+
+  r0 = (ro >>  0) & 15;
+  r1 = (ro >>  4) & 15;
+  r2 = (ro >>  8) & 15;
+  r3 = (ro >> 12) & 15;
+  r4 = (ro >> 16) & 15;
+  r5 = (ro >> 20) & 15;
+  r6 = (ro >> 24) & 15;
+  r7 = (ro >> 28) & 15;
+
+  r0 <<= 1;
+  r1 <<= 1; r1 |= r0 >> 4;
+  r2 <<= 1; r2 |= r1 >> 4;
+  r3 <<= 1; r3 |= r2 >> 4;
+  r4 <<= 1; r4 |= r3 >> 4;
+  r5 <<= 1; r5 |= r4 >> 4;
+  r6 <<= 1; r6 |= r5 >> 4;
+  r7 <<= 1; r7 |= r6 >> 4;
+
+  buf0[0] = toggle_on_register (buf0[0], r0);
+  buf0[1] = toggle_on_register (buf0[1], r1);
+  buf0[2] = toggle_on_register (buf0[2], r2);
+  buf0[3] = toggle_on_register (buf0[3], r3);
+  buf1[0] = toggle_on_register (buf1[0], r4);
+  buf1[1] = toggle_on_register (buf1[1], r5);
+  buf1[2] = toggle_on_register (buf1[2], r6);
+  buf1[3] = toggle_on_register (buf1[3], r7);
+
+  return in_len;
+}
+
 DECLSPEC u32 rule_op_mangle_reverse (MAYBE_UNUSED const u32 p0, MAYBE_UNUSED const u32 p1, MAYBE_UNUSED u32 *buf0, MAYBE_UNUSED u32 *buf1, const u32 in_len)
 {
   reverse_block_optimized (buf0, buf1, buf0, buf1, in_len);
@@ -2285,6 +2361,7 @@ DECLSPEC u32 apply_rule_optimized (const u32 name, const u32 p0, const u32 p1, u
     case RULE_OP_MANGLE_UREST_LFIRST:     out_len = rule_op_mangle_urest_lfirst     (p0, p1, buf0, buf1, out_len); break;
     case RULE_OP_MANGLE_TREST:            out_len = rule_op_mangle_trest            (p0, p1, buf0, buf1, out_len); break;
     case RULE_OP_MANGLE_TOGGLE_AT:        out_len = rule_op_mangle_toggle_at        (p0, p1, buf0, buf1, out_len); break;
+    case RULE_OP_MANGLE_TOGGLE_AT_SEP:    out_len = rule_op_mangle_toggle_at_sep    (p0, p1, buf0, buf1, out_len); break;
     case RULE_OP_MANGLE_REVERSE:          out_len = rule_op_mangle_reverse          (p0, p1, buf0, buf1, out_len); break;
     case RULE_OP_MANGLE_DUPEWORD:         out_len = rule_op_mangle_dupeword         (p0, p1, buf0, buf1, out_len); break;
     case RULE_OP_MANGLE_DUPEWORD_TIMES:   out_len = rule_op_mangle_dupeword_times   (p0, p1, buf0, buf1, out_len); break;
diff --git a/OpenCL/inc_rp_optimized.h b/OpenCL/inc_rp_optimized.h
index b6a133086..d391d1cfe 100644
--- a/OpenCL/inc_rp_optimized.h
+++ b/OpenCL/inc_rp_optimized.h
@@ -21,6 +21,7 @@
 #define RULE_OP_MANGLE_UREST_LFIRST     'C'
 #define RULE_OP_MANGLE_TREST            't'
 #define RULE_OP_MANGLE_TOGGLE_AT        'T'
+#define RULE_OP_MANGLE_TOGGLE_AT_SEP    '3'
 #define RULE_OP_MANGLE_REVERSE          'r'
 #define RULE_OP_MANGLE_DUPEWORD         'd'
 #define RULE_OP_MANGLE_DUPEWORD_TIMES   'p'
@@ -85,6 +86,7 @@ DECLSPEC u32 rule_op_mangle_lrest_ufirst (MAYBE_UNUSED const u32 p0, MAYBE_UNUSE
 DECLSPEC u32 rule_op_mangle_urest_lfirst (MAYBE_UNUSED const u32 p0, MAYBE_UNUSED const u32 p1, MAYBE_UNUSED u32 *buf0, MAYBE_UNUSED u32 *buf1, const u32 in_len);
 DECLSPEC u32 rule_op_mangle_trest (MAYBE_UNUSED const u32 p0, MAYBE_UNUSED const u32 p1, MAYBE_UNUSED u32 *buf0, MAYBE_UNUSED u32 *buf1, const u32 in_len);
 DECLSPEC u32 rule_op_mangle_toggle_at (MAYBE_UNUSED const u32 p0, MAYBE_UNUSED const u32 p1, MAYBE_UNUSED u32 *buf0, MAYBE_UNUSED u32 *buf1, const u32 in_len);
+DECLSPEC u32 rule_op_mangle_toggle_at_sep (MAYBE_UNUSED const u32 p0, MAYBE_UNUSED const u32 p1, MAYBE_UNUSED u32 *buf0, MAYBE_UNUSED u32 *buf1, const u32 in_len);
 DECLSPEC u32 rule_op_mangle_reverse (MAYBE_UNUSED const u32 p0, MAYBE_UNUSED const u32 p1, MAYBE_UNUSED u32 *buf0, MAYBE_UNUSED u32 *buf1, const u32 in_len);
 DECLSPEC u32 rule_op_mangle_dupeword (MAYBE_UNUSED const u32 p0, MAYBE_UNUSED const u32 p1, MAYBE_UNUSED u32 *buf0, MAYBE_UNUSED u32 *buf1, const u32 in_len);
 DECLSPEC u32 rule_op_mangle_dupeword_times (MAYBE_UNUSED const u32 p0, MAYBE_UNUSED const u32 p1, MAYBE_UNUSED u32 *buf0, MAYBE_UNUSED u32 *buf1, const u32 in_len);
diff --git a/OpenCL/inc_simd.h b/OpenCL/inc_simd.h
index f30f07dc3..c3577b297 100644
--- a/OpenCL/inc_simd.h
+++ b/OpenCL/inc_simd.h
@@ -1049,6 +1049,18 @@
 #define MATCHES_NONE_VV(a,b) !(MATCHES_ONE_VV ((a), (b)))
 #define MATCHES_NONE_VS(a,b) !(MATCHES_ONE_VS ((a), (b)))
 
+#if   VECT_SIZE == 1
+#define pack(arr,var,gid,idx) make_u32x ((arr)[((gid) *  1) + 0].var)
+#elif VECT_SIZE == 2
+#define pack(arr,var,gid,idx) make_u32x ((arr)[((gid) *  2) + 0].var, (arr)[((gid) *  2) + 1].var)
+#elif VECT_SIZE == 4
+#define pack(arr,var,gid,idx) make_u32x ((arr)[((gid) *  4) + 0].var, (arr)[((gid) *  4) + 1].var, (arr)[((gid) *  4) + 2].var, (arr)[((gid) *  4) + 3].var)
+#elif VECT_SIZE == 8
+#define pack(arr,var,gid,idx) make_u32x ((arr)[((gid) *  8) + 0].var, (arr)[((gid) *  8) + 1].var, (arr)[((gid) *  8) + 2].var, (arr)[((gid) *  8) + 3].var, (arr)[((gid) *  8) + 4].var, (arr)[((gid) *  8) + 5].var, (arr)[((gid) *  8) + 6].var, (arr)[((gid) *  8) + 7].var)
+#elif VECT_SIZE == 16
+#define pack(arr,var,gid,idx) make_u32x ((arr)[((gid) * 16) + 0].var, (arr)[((gid) * 16) + 1].var, (arr)[((gid) * 16) + 2].var, (arr)[((gid) * 16) + 3].var, (arr)[((gid) * 16) + 4].var, (arr)[((gid) * 16) + 5].var, (arr)[((gid) * 16) + 6].var, (arr)[((gid) * 16) + 7].var, (arr)[((gid) * 16) + 8].var, (arr)[((gid) * 16) + 9].var, (arr)[((gid) * 16) + 10].var, (arr)[((gid) * 16) + 11].var, (arr)[((gid) * 16) + 12].var, (arr)[((gid) * 16) + 13].var, (arr)[((gid) * 16) + 14].var, (arr)[((gid) * 16) + 15].var)
+#endif
+
 #if   VECT_SIZE == 1
 #define packv(arr,var,gid,idx) make_u32x ((arr)[((gid) *  1) + 0].var[(idx)])
 #elif VECT_SIZE == 2
@@ -1097,6 +1109,18 @@
 #define pack64vf(arr,var,gid) make_u64x ((arr)[((gid) * 16) + 0].var, (arr)[((gid) * 16) + 1].var, (arr)[((gid) * 16) + 2].var, (arr)[((gid) * 16) + 3].var, (arr)[((gid) * 16) + 4].var, (arr)[((gid) * 16) + 5].var, (arr)[((gid) * 16) + 6].var, (arr)[((gid) * 16) + 7].var, (arr)[((gid) * 16) + 8].var, (arr)[((gid) * 16) + 9].var, (arr)[((gid) * 16) + 10].var, (arr)[((gid) * 16) + 11].var, (arr)[((gid) * 16) + 12].var, (arr)[((gid) * 16) + 13].var, (arr)[((gid) * 16) + 14].var, (arr)[((gid) * 16) + 15].var)
 #endif
 
+#if   VECT_SIZE == 1
+#define unpack(arr,var,gid,val) (arr)[((gid) *  1) + 0].var = val;
+#elif VECT_SIZE == 2
+#define unpack(arr,var,gid,val) (arr)[((gid) *  2) + 0].var = val.s0; (arr)[((gid) *  2) + 1].var = val.s1;
+#elif VECT_SIZE == 4
+#define unpack(arr,var,gid,val) (arr)[((gid) *  4) + 0].var = val.s0; (arr)[((gid) *  4) + 1].var = val.s1; (arr)[((gid) *  4) + 2].var = val.s2; (arr)[((gid) *  4) + 3].var = val.s3;
+#elif VECT_SIZE == 8
+#define unpack(arr,var,gid,val) (arr)[((gid) *  8) + 0].var = val.s0; (arr)[((gid) *  8) + 1].var = val.s1; (arr)[((gid) *  8) + 2].var = val.s2; (arr)[((gid) *  8) + 3].var = val.s3; (arr)[((gid) *  8) + 4].var = val.s4; (arr)[((gid) *  8) + 5].var = val.s5; (arr)[((gid) *  8) + 6].var = val.s6; (arr)[((gid) *  8) + 7].var = val.s7;
+#elif VECT_SIZE == 16
+#define unpack(arr,var,gid,val) (arr)[((gid) * 16) + 0].var = val.s0; (arr)[((gid) * 16) + 1].var = val.s1; (arr)[((gid) * 16) + 2].var = val.s2; (arr)[((gid) * 16) + 3].var = val.s3; (arr)[((gid) * 16) + 4].var = val.s4; (arr)[((gid) * 16) + 5].var = val.s5; (arr)[((gid) * 16) + 6].var = val.s6; (arr)[((gid) * 16) + 7].var = val.s7; (arr)[((gid) * 16) + 8].var = val.s8; (arr)[((gid) * 16) + 9].var = val.s9; (arr)[((gid) * 16) + 10].var = val.sa; (arr)[((gid) * 16) + 11].var = val.sb; (arr)[((gid) * 16) + 12].var = val.sc; (arr)[((gid) * 16) + 13].var = val.sd; (arr)[((gid) * 16) + 14].var = val.se; (arr)[((gid) * 16) + 15].var = val.sf;
+#endif
+
 #if   VECT_SIZE == 1
 #define unpackv(arr,var,gid,idx,val) (arr)[((gid) *  1) + 0].var[(idx)] = val;
 #elif VECT_SIZE == 2
diff --git a/OpenCL/inc_types.h b/OpenCL/inc_types.h
index f8708e38f..2cff4e0c8 100644
--- a/OpenCL/inc_types.h
+++ b/OpenCL/inc_types.h
@@ -17,18 +17,27 @@
 #endif
 
 #ifdef IS_CUDA
-//https://docs.nvidia.com/cuda/nvrtc/index.html#integer-size
-typedef unsigned char      uchar;
-typedef unsigned short     ushort;
-typedef unsigned int       uint;
-typedef unsigned long long ulong;
+// https://docs.nvidia.com/cuda/nvrtc/index.html#integer-size
+typedef unsigned char       uchar;
+typedef unsigned short      ushort;
+typedef unsigned int        uint;
+typedef unsigned long       ulong;
+typedef unsigned long long  ullong;
+#endif
+
+#ifdef IS_OPENCL
+typedef ulong   ullong;
+typedef ulong2  ullong2;
+typedef ulong4  ullong4;
+typedef ulong8  ullong8;
+typedef ulong16 ullong16;
 #endif
 
 #ifdef KERNEL_STATIC
 typedef uchar  u8;
 typedef ushort u16;
 typedef uint   u32;
-typedef ulong  u64;
+typedef ullong u64;
 #else
 typedef uint8_t  u8;
 typedef uint16_t u16;
@@ -68,7 +77,7 @@ typedef u64  u64x;
 #define make_u64x (u64)
 
 #else
-#if defined IS_CUDA
+#if defined IS_CUDA || defined IS_HIP
 
 #if VECT_SIZE == 2
 
@@ -824,767 +833,11 @@ typedef __device_builtin__ struct u64x u64x;
 #define make_u32x u32x
 #define make_u64x u64x
 
-#elif defined IS_HIP
-
-#if VECT_SIZE == 2
-
-struct u8x
-{
-  u8 s0;
-  u8 s1;
-
-  inline __device__  u8x (const u8 a, const u8 b) : s0(a), s1(b) { }
-  inline __device__  u8x (const u8 a)             : s0(a), s1(a) { }
-
-  inline __device__  u8x (void) : s0(0), s1(0) { }
-  inline __device__ ~u8x (void) { }
-};
-
-struct u16x
-{
-  u16 s0;
-  u16 s1;
-
-  inline __device__  u16x (const u16 a, const u16 b) : s0(a), s1(b) { }
-  inline __device__  u16x (const u16 a)              : s0(a), s1(a) { }
-
-  inline __device__  u16x (void) : s0(0), s1(0) { }
-  inline __device__ ~u16x (void) { }
-};
-
-struct u32x
-{
-  u32 s0;
-  u32 s1;
-
-  inline __device__  u32x (const u32 a, const u32 b) : s0(a), s1(b) { }
-  inline __device__  u32x (const u32 a)              : s0(a), s1(a) { }
-
-  inline __device__  u32x (void) : s0(0), s1(0) { }
-  inline __device__ ~u32x (void) { }
-};
-
-struct u64x
-{
-  u64 s0;
-  u64 s1;
-
-  inline __device__  u64x (const u64 a, const u64 b) : s0(a), s1(b) { }
-  inline __device__  u64x (const u64 a)              : s0(a), s1(a) { }
-
-  inline __device__  u64x (void) : s0(0), s1(0) { }
-  inline __device__ ~u64x (void) { }
-};
-
-inline __device__ bool operator != (const u32x a, const u32  b) { return ((a.s0 != b)    && (a.s1 != b));    }
-inline __device__ bool operator != (const u32x a, const u32x b) { return ((a.s0 != b.s0) && (a.s1 != b.s1)); }
-
-inline __device__ void operator ^= (u32x &a, const u32  b) { a.s0 ^= b;    a.s1 ^= b;     }
-inline __device__ void operator ^= (u32x &a, const u32x b) { a.s0 ^= b.s0; a.s1 ^= b.s1;  }
-
-inline __device__ void operator |= (u32x &a, const u32  b) { a.s0 |= b;    a.s1 |= b;     }
-inline __device__ void operator |= (u32x &a, const u32x b) { a.s0 |= b.s0; a.s1 |= b.s1;  }
-
-inline __device__ void operator &= (u32x &a, const u32  b) { a.s0 &= b;    a.s1 &= b;     }
-inline __device__ void operator &= (u32x &a, const u32x b) { a.s0 &= b.s0; a.s1 &= b.s1;  }
-
-inline __device__ void operator += (u32x &a, const u32  b) { a.s0 += b;    a.s1 += b;     }
-inline __device__ void operator += (u32x &a, const u32x b) { a.s0 += b.s0; a.s1 += b.s1;  }
-
-inline __device__ void operator -= (u32x &a, const u32  b) { a.s0 -= b;    a.s1 -= b;     }
-inline __device__ void operator -= (u32x &a, const u32x b) { a.s0 -= b.s0; a.s1 -= b.s1;  }
-
-inline __device__ void operator *= (u32x &a, const u32  b) { a.s0 *= b;    a.s1 *= b;     }
-inline __device__ void operator *= (u32x &a, const u32x b) { a.s0 *= b.s0; a.s1 *= b.s1;  }
-
-inline __device__ void operator >>= (u32x &a, const u32  b) { a.s0 >>= b;    a.s1 >>= b;     }
-inline __device__ void operator >>= (u32x &a, const u32x b) { a.s0 >>= b.s0; a.s1 >>= b.s1;  }
-
-inline __device__ void operator <<= (u32x &a, const u32  b) { a.s0 <<= b;    a.s1 <<= b;     }
-inline __device__ void operator <<= (u32x &a, const u32x b) { a.s0 <<= b.s0; a.s1 <<= b.s1;  }
-
-inline __device__ u32x operator << (const u32x a, const u32  b) { return u32x ((a.s0 << b),    (a.s1 << b)   );  }
-inline __device__ u32x operator << (const u32x a, const u32x b) { return u32x ((a.s0 << b.s0), (a.s1 << b.s1));  }
-
-inline __device__ u32x operator >> (const u32x a, const u32  b) { return u32x ((a.s0 >> b),    (a.s1 >> b)   );  }
-inline __device__ u32x operator >> (const u32x a, const u32x b) { return u32x ((a.s0 >> b.s0), (a.s1 >> b.s1));  }
-
-inline __device__ u32x operator ^  (const u32x a, const u32  b) { return u32x ((a.s0 ^  b),    (a.s1 ^  b)   );  }
-inline __device__ u32x operator ^  (const u32x a, const u32x b) { return u32x ((a.s0 ^  b.s0), (a.s1 ^  b.s1));  }
-
-inline __device__ u32x operator |  (const u32x a, const u32  b) { return u32x ((a.s0 |  b),    (a.s1 |  b)   );  }
-inline __device__ u32x operator |  (const u32x a, const u32x b) { return u32x ((a.s0 |  b.s0), (a.s1 |  b.s1));  }
-
-inline __device__ u32x operator &  (const u32x a, const u32  b) { return u32x ((a.s0 &  b),    (a.s1 &  b)   );  }
-inline __device__ u32x operator &  (const u32x a, const u32x b) { return u32x ((a.s0 &  b.s0), (a.s1 &  b.s1));  }
-
-inline __device__ u32x operator +  (const u32x a, const u32  b) { return u32x ((a.s0 +  b),    (a.s1 +  b)   );  }
-inline __device__ u32x operator +  (const u32x a, const u32x b) { return u32x ((a.s0 +  b.s0), (a.s1 +  b.s1));  }
-
-inline __device__ u32x operator -  (const u32x a, const u32  b) { return u32x ((a.s0 -  b),    (a.s1 -  b)   );  }
-inline __device__ u32x operator -  (const u32x a, const u32x b) { return u32x ((a.s0 -  b.s0), (a.s1 -  b.s1));  }
-
-inline __device__ u32x operator *  (const u32x a, const u32  b) { return u32x ((a.s0 *  b),    (a.s1 *  b)   );  }
-inline __device__ u32x operator *  (const u32x a, const u32x b) { return u32x ((a.s0 *  b.s0), (a.s1 *  b.s1));  }
-
-inline __device__ u32x operator %  (const u32x a, const u32  b) { return u32x ((a.s0 %  b),    (a.s1 %  b)   );  }
-inline __device__ u32x operator %  (const u32x a, const u32x b) { return u32x ((a.s0 %  b.s0), (a.s1 %  b.s1));  }
-
-inline __device__ u32x operator ~  (const u32x a) { return u32x (~a.s0, ~a.s1); }
-
-inline __device__ bool operator != (const u64x a, const u64  b) { return ((a.s0 != b)    && (a.s1 != b));    }
-inline __device__ bool operator != (const u64x a, const u64x b) { return ((a.s0 != b.s0) && (a.s1 != b.s1)); }
-
-inline __device__ void operator ^= (u64x &a, const u64  b) { a.s0 ^= b;    a.s1 ^= b;     }
-inline __device__ void operator ^= (u64x &a, const u64x b) { a.s0 ^= b.s0; a.s1 ^= b.s1;  }
-
-inline __device__ void operator |= (u64x &a, const u64  b) { a.s0 |= b;    a.s1 |= b;     }
-inline __device__ void operator |= (u64x &a, const u64x b) { a.s0 |= b.s0; a.s1 |= b.s1;  }
-
-inline __device__ void operator &= (u64x &a, const u64  b) { a.s0 &= b;    a.s1 &= b;     }
-inline __device__ void operator &= (u64x &a, const u64x b) { a.s0 &= b.s0; a.s1 &= b.s1;  }
-
-inline __device__ void operator += (u64x &a, const u64  b) { a.s0 += b;    a.s1 += b;     }
-inline __device__ void operator += (u64x &a, const u64x b) { a.s0 += b.s0; a.s1 += b.s1;  }
-
-inline __device__ void operator -= (u64x &a, const u64  b) { a.s0 -= b;    a.s1 -= b;     }
-inline __device__ void operator -= (u64x &a, const u64x b) { a.s0 -= b.s0; a.s1 -= b.s1;  }
-
-inline __device__ void operator *= (u64x &a, const u64  b) { a.s0 *= b;    a.s1 *= b;     }
-inline __device__ void operator *= (u64x &a, const u64x b) { a.s0 *= b.s0; a.s1 *= b.s1;  }
-
-inline __device__ void operator >>= (u64x &a, const u64  b) { a.s0 >>= b;    a.s1 >>= b;     }
-inline __device__ void operator >>= (u64x &a, const u64x b) { a.s0 >>= b.s0; a.s1 >>= b.s1;  }
-
-inline __device__ void operator <<= (u64x &a, const u64  b) { a.s0 <<= b;    a.s1 <<= b;     }
-inline __device__ void operator <<= (u64x &a, const u64x b) { a.s0 <<= b.s0; a.s1 <<= b.s1;  }
-
-inline __device__ u64x operator << (const u64x a, const u64  b) { return u64x ((a.s0 << b),    (a.s1 << b)   );  }
-inline __device__ u64x operator << (const u64x a, const u64x b) { return u64x ((a.s0 << b.s0), (a.s1 << b.s1));  }
-
-inline __device__ u64x operator >> (const u64x a, const u64  b) { return u64x ((a.s0 >> b),    (a.s1 >> b)   );  }
-inline __device__ u64x operator >> (const u64x a, const u64x b) { return u64x ((a.s0 >> b.s0), (a.s1 >> b.s1));  }
-
-inline __device__ u64x operator ^  (const u64x a, const u64  b) { return u64x ((a.s0 ^  b),    (a.s1 ^  b)   );  }
-inline __device__ u64x operator ^  (const u64x a, const u64x b) { return u64x ((a.s0 ^  b.s0), (a.s1 ^  b.s1));  }
-
-inline __device__ u64x operator |  (const u64x a, const u64  b) { return u64x ((a.s0 |  b),    (a.s1 |  b)   );  }
-inline __device__ u64x operator |  (const u64x a, const u64x b) { return u64x ((a.s0 |  b.s0), (a.s1 |  b.s1));  }
-
-inline __device__ u64x operator &  (const u64x a, const u64  b) { return u64x ((a.s0 &  b),    (a.s1 &  b)   );  }
-inline __device__ u64x operator &  (const u64x a, const u64x b) { return u64x ((a.s0 &  b.s0), (a.s1 &  b.s1));  }
-
-inline __device__ u64x operator +  (const u64x a, const u64  b) { return u64x ((a.s0 +  b),    (a.s1 +  b)   );  }
-inline __device__ u64x operator +  (const u64x a, const u64x b) { return u64x ((a.s0 +  b.s0), (a.s1 +  b.s1));  }
-
-inline __device__ u64x operator -  (const u64x a, const u64  b) { return u64x ((a.s0 -  b),    (a.s1 -  b)   );  }
-inline __device__ u64x operator -  (const u64x a, const u64x b) { return u64x ((a.s0 -  b.s0), (a.s1 -  b.s1));  }
-
-inline __device__ u64x operator *  (const u64x a, const u64  b) { return u64x ((a.s0 *  b),    (a.s1 *  b)   );  }
-inline __device__ u64x operator *  (const u64x a, const u64x b) { return u64x ((a.s0 *  b.s0), (a.s1 *  b.s1));  }
-
-inline __device__ u64x operator %  (const u64x a, const u64  b) { return u64x ((a.s0 %  b),    (a.s1 %  b)   );  }
-inline __device__ u64x operator %  (const u64x a, const u64x b) { return u64x ((a.s0 %  b.s0), (a.s1 %  b.s1));  }
-
-inline __device__ u64x operator ~  (const u64x a) { return u64x (~a.s0, ~a.s1); }
-
-#endif
-
-#if VECT_SIZE == 4
-
-struct u8x
-{
-  u8 s0;
-  u8 s1;
-  u8 s2;
-  u8 s3;
-
-  inline __device__  u8x (const u8 a, const u8 b, const u8 c, const u8 d) : s0(a), s1(b), s2(c), s3(d) { }
-  inline __device__  u8x (const u8 a)                                     : s0(a), s1(a), s2(a), s3(a) { }
-
-  inline __device__  u8x (void) : s0(0), s1(0), s2(0), s3(0) { }
-  inline __device__ ~u8x (void) { }
-};
-
-struct u16x
-{
-  u16 s0;
-  u16 s1;
-  u16 s2;
-  u16 s3;
-
-  inline __device__  u16x (const u16 a, const u16 b, const u16 c, const u16 d) : s0(a), s1(b), s2(c), s3(d) { }
-  inline __device__  u16x (const u16 a)                                        : s0(a), s1(a), s2(a), s3(a) { }
-
-  inline __device__  u16x (void) : s0(0), s1(0), s2(0), s3(0) { }
-  inline __device__ ~u16x (void) { }
-};
-
-struct u32x
-{
-  u32 s0;
-  u32 s1;
-  u32 s2;
-  u32 s3;
-
-  inline __device__  u32x (const u32 a, const u32 b, const u32 c, const u32 d) : s0(a), s1(b), s2(c), s3(d) { }
-  inline __device__  u32x (const u32 a)                                        : s0(a), s1(a), s2(a), s3(a) { }
-
-  inline __device__  u32x (void) : s0(0), s1(0), s2(0), s3(0) { }
-  inline __device__ ~u32x (void) { }
-};
-
-struct u64x
-{
-  u64 s0;
-  u64 s1;
-  u64 s2;
-  u64 s3;
-
-  inline __device__  u64x (const u64 a, const u64 b, const u64 c, const u64 d) : s0(a), s1(b), s2(c), s3(d) { }
-  inline __device__  u64x (const u64 a)                                        : s0(a), s1(a), s2(a), s3(a) { }
-
-  inline __device__  u64x (void) : s0(0), s1(0), s2(0), s3(0) { }
-  inline __device__ ~u64x (void) { }
-};
-
-inline __device__ bool operator != (const u32x a, const u32  b) { return ((a.s0 != b)    && (a.s1 != b)    && (a.s2 != b)    && (a.s3 != b)   ); }
-inline __device__ bool operator != (const u32x a, const u32x b) { return ((a.s0 != b.s0) && (a.s1 != b.s1) && (a.s2 != b.s2) && (a.s3 != b.s3)); }
-
-inline __device__ void operator ^= (u32x &a, const u32  b) { a.s0 ^= b;    a.s1 ^= b;    a.s2 ^= b;    a.s3 ^= b;     }
-inline __device__ void operator ^= (u32x &a, const u32x b) { a.s0 ^= b.s0; a.s1 ^= b.s1; a.s2 ^= b.s2; a.s3 ^= b.s3;  }
-
-inline __device__ void operator |= (u32x &a, const u32  b) { a.s0 |= b;    a.s1 |= b;    a.s2 |= b;    a.s3 |= b;     }
-inline __device__ void operator |= (u32x &a, const u32x b) { a.s0 |= b.s0; a.s1 |= b.s1; a.s2 |= b.s2; a.s3 |= b.s3;  }
-
-inline __device__ void operator &= (u32x &a, const u32  b) { a.s0 &= b;    a.s1 &= b;    a.s2 &= b;    a.s3 &= b;     }
-inline __device__ void operator &= (u32x &a, const u32x b) { a.s0 &= b.s0; a.s1 &= b.s1; a.s2 &= b.s2; a.s3 &= b.s3;  }
-
-inline __device__ void operator += (u32x &a, const u32  b) { a.s0 += b;    a.s1 += b;    a.s2 += b;    a.s3 += b;     }
-inline __device__ void operator += (u32x &a, const u32x b) { a.s0 += b.s0; a.s1 += b.s1; a.s2 += b.s2; a.s3 += b.s3;  }
-
-inline __device__ void operator -= (u32x &a, const u32  b) { a.s0 -= b;    a.s1 -= b;    a.s2 -= b;    a.s3 -= b;     }
-inline __device__ void operator -= (u32x &a, const u32x b) { a.s0 -= b.s0; a.s1 -= b.s1; a.s2 -= b.s2; a.s3 -= b.s3;  }
-
-inline __device__ void operator *= (u32x &a, const u32  b) { a.s0 *= b;    a.s1 *= b;    a.s2 *= b;    a.s3 *= b;     }
-inline __device__ void operator *= (u32x &a, const u32x b) { a.s0 *= b.s0; a.s1 *= b.s1; a.s2 *= b.s2; a.s3 *= b.s3;  }
-
-inline __device__ void operator >>= (u32x &a, const u32  b) { a.s0 >>= b;    a.s1 >>= b;    a.s2 >>= b;    a.s3 >>= b;     }
-inline __device__ void operator >>= (u32x &a, const u32x b) { a.s0 >>= b.s0; a.s1 >>= b.s1; a.s2 >>= b.s2; a.s3 >>= b.s3;  }
-
-inline __device__ void operator <<= (u32x &a, const u32  b) { a.s0 <<= b;    a.s1 <<= b;    a.s2 <<= b;    a.s3 <<= b;     }
-inline __device__ void operator <<= (u32x &a, const u32x b) { a.s0 <<= b.s0; a.s1 <<= b.s1; a.s2 <<= b.s2; a.s3 <<= b.s3;  }
-
-inline __device__ u32x operator << (const u32x a, const u32  b) { return u32x ((a.s0 << b),    (a.s1 << b)   , (a.s2 << b),    (a.s3 << b)   );  }
-inline __device__ u32x operator << (const u32x a, const u32x b) { return u32x ((a.s0 << b.s0), (a.s1 << b.s1), (a.s2 << b.s2), (a.s3 << b.s3));  }
-
-inline __device__ u32x operator >> (const u32x a, const u32  b) { return u32x ((a.s0 >> b),    (a.s1 >> b)   , (a.s2 >> b),    (a.s3 >> b)   );  }
-inline __device__ u32x operator >> (const u32x a, const u32x b) { return u32x ((a.s0 >> b.s0), (a.s1 >> b.s1), (a.s2 >> b.s2), (a.s3 >> b.s3));  }
-
-inline __device__ u32x operator ^  (const u32x a, const u32  b) { return u32x ((a.s0 ^  b),    (a.s1 ^  b)   , (a.s2 ^  b),    (a.s3 ^  b)   );  }
-inline __device__ u32x operator ^  (const u32x a, const u32x b) { return u32x ((a.s0 ^  b.s0), (a.s1 ^  b.s1), (a.s2 ^  b.s2), (a.s3 ^  b.s3));  }
-
-inline __device__ u32x operator |  (const u32x a, const u32  b) { return u32x ((a.s0 |  b),    (a.s1 |  b)   , (a.s2 |  b),    (a.s3 |  b)   );  }
-inline __device__ u32x operator |  (const u32x a, const u32x b) { return u32x ((a.s0 |  b.s0), (a.s1 |  b.s1), (a.s2 |  b.s2), (a.s3 |  b.s3));  }
-
-inline __device__ u32x operator &  (const u32x a, const u32  b) { return u32x ((a.s0 &  b),    (a.s1 &  b)   , (a.s2 &  b),    (a.s3 &  b)   );  }
-inline __device__ u32x operator &  (const u32x a, const u32x b) { return u32x ((a.s0 &  b.s0), (a.s1 &  b.s1), (a.s2 &  b.s2), (a.s3 &  b.s3));  }
-
-inline __device__ u32x operator +  (const u32x a, const u32  b) { return u32x ((a.s0 +  b),    (a.s1 +  b)   , (a.s2 +  b),    (a.s3 +  b)   );  }
-inline __device__ u32x operator +  (const u32x a, const u32x b) { return u32x ((a.s0 +  b.s0), (a.s1 +  b.s1), (a.s2 +  b.s2), (a.s3 +  b.s3));  }
-
-inline __device__ u32x operator -  (const u32x a, const u32  b) { return u32x ((a.s0 -  b),    (a.s1 -  b)   , (a.s2 -  b),    (a.s3 -  b)   );  }
-inline __device__ u32x operator -  (const u32x a, const u32x b) { return u32x ((a.s0 -  b.s0), (a.s1 -  b.s1), (a.s2 -  b.s2), (a.s3 -  b.s3));  }
-
-inline __device__ u32x operator *  (const u32x a, const u32  b) { return u32x ((a.s0 *  b),    (a.s1 *  b)   , (a.s2 *  b),    (a.s3 *  b)   );  }
-inline __device__ u32x operator *  (const u32x a, const u32x b) { return u32x ((a.s0 *  b.s0), (a.s1 *  b.s1), (a.s2 *  b.s2), (a.s3 *  b.s3));  }
-
-inline __device__ u32x operator %  (const u32x a, const u32  b) { return u32x ((a.s0 %  b),    (a.s1 %  b)   , (a.s2 %  b),    (a.s3 %  b)   );  }
-inline __device__ u32x operator %  (const u32x a, const u32x b) { return u32x ((a.s0 %  b.s0), (a.s1 %  b.s1), (a.s2 %  b.s2), (a.s3 %  b.s3));  }
-
-inline __device__ u32x operator ~  (const u32x a) { return u32x (~a.s0, ~a.s1, ~a.s2, ~a.s3); }
-
-inline __device__ bool operator != (const u64x a, const u64  b) { return ((a.s0 != b)    && (a.s1 != b)    && (a.s2 != b)    && (a.s3 != b)   ); }
-inline __device__ bool operator != (const u64x a, const u64x b) { return ((a.s0 != b.s0) && (a.s1 != b.s1) && (a.s2 != b.s2) && (a.s3 != b.s3)); }
-
-inline __device__ void operator ^= (u64x &a, const u64  b) { a.s0 ^= b;    a.s1 ^= b;    a.s2 ^= b;    a.s3 ^= b;     }
-inline __device__ void operator ^= (u64x &a, const u64x b) { a.s0 ^= b.s0; a.s1 ^= b.s1; a.s2 ^= b.s2; a.s3 ^= b.s3;  }
-
-inline __device__ void operator |= (u64x &a, const u64  b) { a.s0 |= b;    a.s1 |= b;    a.s2 |= b;    a.s3 |= b;     }
-inline __device__ void operator |= (u64x &a, const u64x b) { a.s0 |= b.s0; a.s1 |= b.s1; a.s2 |= b.s2; a.s3 |= b.s3;  }
-
-inline __device__ void operator &= (u64x &a, const u64  b) { a.s0 &= b;    a.s1 &= b;    a.s2 &= b;    a.s3 &= b;     }
-inline __device__ void operator &= (u64x &a, const u64x b) { a.s0 &= b.s0; a.s1 &= b.s1; a.s2 &= b.s2; a.s3 &= b.s3;  }
-
-inline __device__ void operator += (u64x &a, const u64  b) { a.s0 += b;    a.s1 += b;    a.s2 += b;    a.s3 += b;     }
-inline __device__ void operator += (u64x &a, const u64x b) { a.s0 += b.s0; a.s1 += b.s1; a.s2 += b.s2; a.s3 += b.s3;  }
-
-inline __device__ void operator -= (u64x &a, const u64  b) { a.s0 -= b;    a.s1 -= b;    a.s2 -= b;    a.s3 -= b;     }
-inline __device__ void operator -= (u64x &a, const u64x b) { a.s0 -= b.s0; a.s1 -= b.s1; a.s2 -= b.s2; a.s3 -= b.s3;  }
-
-inline __device__ void operator *= (u64x &a, const u64  b) { a.s0 *= b;    a.s1 *= b;    a.s2 *= b;    a.s3 *= b;     }
-inline __device__ void operator *= (u64x &a, const u64x b) { a.s0 *= b.s0; a.s1 *= b.s1; a.s2 *= b.s2; a.s3 *= b.s3;  }
-
-inline __device__ void operator >>= (u64x &a, const u64  b) { a.s0 >>= b;    a.s1 >>= b;    a.s2 >>= b;    a.s3 >>= b;     }
-inline __device__ void operator >>= (u64x &a, const u64x b) { a.s0 >>= b.s0; a.s1 >>= b.s1; a.s2 >>= b.s2; a.s3 >>= b.s3;  }
-
-inline __device__ void operator <<= (u64x &a, const u64  b) { a.s0 <<= b;    a.s1 <<= b;    a.s2 <<= b;    a.s3 <<= b;     }
-inline __device__ void operator <<= (u64x &a, const u64x b) { a.s0 <<= b.s0; a.s1 <<= b.s1; a.s2 <<= b.s2; a.s3 <<= b.s3;  }
-
-inline __device__ u64x operator << (const u64x a, const u64  b) { return u64x ((a.s0 << b),    (a.s1 << b)   , (a.s2 << b),    (a.s3 << b)   );  }
-inline __device__ u64x operator << (const u64x a, const u64x b) { return u64x ((a.s0 << b.s0), (a.s1 << b.s1), (a.s2 << b.s2), (a.s3 << b.s3));  }
-
-inline __device__ u64x operator >> (const u64x a, const u64  b) { return u64x ((a.s0 >> b),    (a.s1 >> b)   , (a.s2 >> b),    (a.s3 >> b)   );  }
-inline __device__ u64x operator >> (const u64x a, const u64x b) { return u64x ((a.s0 >> b.s0), (a.s1 >> b.s1), (a.s2 >> b.s2), (a.s3 >> b.s3));  }
-
-inline __device__ u64x operator ^  (const u64x a, const u64  b) { return u64x ((a.s0 ^  b),    (a.s1 ^  b)   , (a.s2 ^  b),    (a.s3 ^  b)   );  }
-inline __device__ u64x operator ^  (const u64x a, const u64x b) { return u64x ((a.s0 ^  b.s0), (a.s1 ^  b.s1), (a.s2 ^  b.s2), (a.s3 ^  b.s3));  }
-
-inline __device__ u64x operator |  (const u64x a, const u64  b) { return u64x ((a.s0 |  b),    (a.s1 |  b)   , (a.s2 |  b),    (a.s3 |  b)   );  }
-inline __device__ u64x operator |  (const u64x a, const u64x b) { return u64x ((a.s0 |  b.s0), (a.s1 |  b.s1), (a.s2 |  b.s2), (a.s3 |  b.s3));  }
-
-inline __device__ u64x operator &  (const u64x a, const u64  b) { return u64x ((a.s0 &  b),    (a.s1 &  b)   , (a.s2 &  b),    (a.s3 &  b)   );  }
-inline __device__ u64x operator &  (const u64x a, const u64x b) { return u64x ((a.s0 &  b.s0), (a.s1 &  b.s1), (a.s2 &  b.s2), (a.s3 &  b.s3));  }
-
-inline __device__ u64x operator +  (const u64x a, const u64  b) { return u64x ((a.s0 +  b),    (a.s1 +  b)   , (a.s2 +  b),    (a.s3 +  b)   );  }
-inline __device__ u64x operator +  (const u64x a, const u64x b) { return u64x ((a.s0 +  b.s0), (a.s1 +  b.s1), (a.s2 +  b.s2), (a.s3 +  b.s3));  }
-
-inline __device__ u64x operator -  (const u64x a, const u64  b) { return u64x ((a.s0 -  b),    (a.s1 -  b)   , (a.s2 -  b),    (a.s3 -  b)   );  }
-inline __device__ u64x operator -  (const u64x a, const u64x b) { return u64x ((a.s0 -  b.s0), (a.s1 -  b.s1), (a.s2 -  b.s2), (a.s3 -  b.s3));  }
-
-inline __device__ u64x operator *  (const u64x a, const u64  b) { return u64x ((a.s0 *  b),    (a.s1 *  b)   , (a.s2 *  b),    (a.s3 *  b)   );  }
-inline __device__ u64x operator *  (const u64x a, const u64x b) { return u64x ((a.s0 *  b.s0), (a.s1 *  b.s1), (a.s2 *  b.s2), (a.s3 *  b.s3));  }
-
-inline __device__ u64x operator %  (const u64x a, const u32  b) { return u64x ((a.s0 %  b),    (a.s1 %  b)   , (a.s2 %  b),    (a.s3 %  b)   );  }
-inline __device__ u64x operator %  (const u64x a, const u64x b) { return u64x ((a.s0 %  b.s0), (a.s1 %  b.s1), (a.s2 %  b.s2), (a.s3 %  b.s3));  }
-
-inline __device__ u64x operator ~  (const u64x a) { return u64x (~a.s0, ~a.s1, ~a.s2, ~a.s3); }
-
-#endif
-
-#if VECT_SIZE == 8
-
-struct u8x
-{
-  u8 s0;
-  u8 s1;
-  u8 s2;
-  u8 s3;
-  u8 s4;
-  u8 s5;
-  u8 s6;
-  u8 s7;
-
-  inline __device__  u8x (const u8 a, const u8 b, const u8 c, const u8 d, const u8 e, const u8 f, const u8 g, const u8 h) : s0(a), s1(b), s2(c), s3(d), s4(e), s5(f), s6(g), s7(h) { }
-  inline __device__  u8x (const u8 a)                                                                                     : s0(a), s1(a), s2(a), s3(a), s4(a), s5(a), s6(a), s7(a) { }
-
-  inline __device__  u8x (void) : s0(0), s1(0), s2(0), s3(0), s4(0), s5(0), s6(0), s7(0) { }
-  inline __device__ ~u8x (void) { }
-};
-
-struct u16x
-{
-  u16 s0;
-  u16 s1;
-  u16 s2;
-  u16 s3;
-  u16 s4;
-  u16 s5;
-  u16 s6;
-  u16 s7;
-
-  inline __device__  u16x (const u16 a, const u16 b, const u16 c, const u16 d, const u16 e, const u16 f, const u16 g, const u16 h) : s0(a), s1(b), s2(c), s3(d), s4(e), s5(f), s6(g), s7(h) { }
-  inline __device__  u16x (const u16 a)                                                                                            : s0(a), s1(a), s2(a), s3(a), s4(a), s5(a), s6(a), s7(a) { }
-
-  inline __device__  u16x (void) : s0(0), s1(0), s2(0), s3(0), s4(0), s5(0), s6(0), s7(0) { }
-  inline __device__ ~u16x (void) { }
-};
-
-struct u32x
-{
-  u32 s0;
-  u32 s1;
-  u32 s2;
-  u32 s3;
-  u32 s4;
-  u32 s5;
-  u32 s6;
-  u32 s7;
-
-  inline __device__  u32x (const u32 a, const u32 b, const u32 c, const u32 d, const u32 e, const u32 f, const u32 g, const u32 h) : s0(a), s1(b), s2(c), s3(d), s4(e), s5(f), s6(g), s7(h) { }
-  inline __device__  u32x (const u32 a)                                                                                            : s0(a), s1(a), s2(a), s3(a), s4(a), s5(a), s6(a), s7(a) { }
-
-  inline __device__  u32x (void) : s0(0), s1(0), s2(0), s3(0), s4(0), s5(0), s6(0), s7(0) { }
-  inline __device__ ~u32x (void) { }
-};
-
-struct u64x
-{
-  u64 s0;
-  u64 s1;
-  u64 s2;
-  u64 s3;
-  u64 s4;
-  u64 s5;
-  u64 s6;
-  u64 s7;
-
-  inline __device__  u64x (const u64 a, const u64 b, const u64 c, const u64 d, const u64 e, const u64 f, const u64 g, const u64 h) : s0(a), s1(b), s2(c), s3(d), s4(e), s5(f), s6(g), s7(h) { }
-  inline __device__  u64x (const u64 a)                                                                                            : s0(a), s1(a), s2(a), s3(a), s4(a), s5(a), s6(a), s7(a) { }
-
-  inline __device__  u64x (void) : s0(0), s1(0), s2(0), s3(0), s4(0), s5(0), s6(0), s7(0) { }
-  inline __device__ ~u64x (void) { }
-};
-
-inline __device__ bool operator != (const u32x a, const u32  b) { return ((a.s0 != b)    && (a.s1 != b)    && (a.s2 != b)    && (a.s3 != b)    && (a.s4 != b)    && (a.s5 != b)    && (a.s6 != b)    && (a.s7 != b)   ); }
-inline __device__ bool operator != (const u32x a, const u32x b) { return ((a.s0 != b.s0) && (a.s1 != b.s1) && (a.s2 != b.s2) && (a.s3 != b.s3) && (a.s4 != b.s4) && (a.s5 != b.s5) && (a.s6 != b.s6) && (a.s7 != b.s7)); }
-
-inline __device__ void operator ^= (u32x &a, const u32  b) { a.s0 ^= b;    a.s1 ^= b;    a.s2 ^= b;    a.s3 ^= b;    a.s4 ^= b;    a.s5 ^= b;    a.s6 ^= b;    a.s7 ^= b;     }
-inline __device__ void operator ^= (u32x &a, const u32x b) { a.s0 ^= b.s0; a.s1 ^= b.s1; a.s2 ^= b.s2; a.s3 ^= b.s3; a.s4 ^= b.s4; a.s5 ^= b.s5; a.s6 ^= b.s6; a.s7 ^= b.s7;  }
-
-inline __device__ void operator |= (u32x &a, const u32  b) { a.s0 |= b;    a.s1 |= b;    a.s2 |= b;    a.s3 |= b;    a.s4 |= b;    a.s5 |= b;    a.s6 |= b;    a.s7 |= b;     }
-inline __device__ void operator |= (u32x &a, const u32x b) { a.s0 |= b.s0; a.s1 |= b.s1; a.s2 |= b.s2; a.s3 |= b.s3; a.s4 |= b.s4; a.s5 |= b.s5; a.s6 |= b.s6; a.s7 |= b.s7;  }
-
-inline __device__ void operator &= (u32x &a, const u32  b) { a.s0 &= b;    a.s1 &= b;    a.s2 &= b;    a.s3 &= b;    a.s4 &= b;    a.s5 &= b;    a.s6 &= b;    a.s7 &= b;     }
-inline __device__ void operator &= (u32x &a, const u32x b) { a.s0 &= b.s0; a.s1 &= b.s1; a.s2 &= b.s2; a.s3 &= b.s3; a.s4 &= b.s4; a.s5 &= b.s5; a.s6 &= b.s6; a.s7 &= b.s7;  }
-
-inline __device__ void operator += (u32x &a, const u32  b) { a.s0 += b;    a.s1 += b;    a.s2 += b;    a.s3 += b;    a.s4 += b;    a.s5 += b;    a.s6 += b;    a.s7 += b;     }
-inline __device__ void operator += (u32x &a, const u32x b) { a.s0 += b.s0; a.s1 += b.s1; a.s2 += b.s2; a.s3 += b.s3; a.s4 += b.s4; a.s5 += b.s5; a.s6 += b.s6; a.s7 += b.s7;  }
-
-inline __device__ void operator -= (u32x &a, const u32  b) { a.s0 -= b;    a.s1 -= b;    a.s2 -= b;    a.s3 -= b;    a.s4 -= b;    a.s5 -= b;    a.s6 -= b;    a.s7 -= b;     }
-inline __device__ void operator -= (u32x &a, const u32x b) { a.s0 -= b.s0; a.s1 -= b.s1; a.s2 -= b.s2; a.s3 -= b.s3; a.s4 -= b.s4; a.s5 -= b.s5; a.s6 -= b.s6; a.s7 -= b.s7;  }
-
-inline __device__ void operator *= (u32x &a, const u32  b) { a.s0 *= b;    a.s1 *= b;    a.s2 *= b;    a.s3 *= b;    a.s4 *= b;    a.s5 *= b;    a.s6 *= b;    a.s7 *= b;     }
-inline __device__ void operator *= (u32x &a, const u32x b) { a.s0 *= b.s0; a.s1 *= b.s1; a.s2 *= b.s2; a.s3 *= b.s3; a.s4 *= b.s4; a.s5 *= b.s5; a.s6 *= b.s6; a.s7 *= b.s7;  }
-
-inline __device__ void operator >>= (u32x &a, const u32  b) { a.s0 >>= b;    a.s1 >>= b;    a.s2 >>= b;    a.s3 >>= b;    a.s4 >>= b;    a.s5 >>= b;    a.s6 >>= b;    a.s7 >>= b;     }
-inline __device__ void operator >>= (u32x &a, const u32x b) { a.s0 >>= b.s0; a.s1 >>= b.s1; a.s2 >>= b.s2; a.s3 >>= b.s3; a.s4 >>= b.s4; a.s5 >>= b.s5; a.s6 >>= b.s6; a.s7 >>= b.s7;  }
-
-inline __device__ void operator <<= (u32x &a, const u32  b) { a.s0 <<= b;    a.s1 <<= b;    a.s2 <<= b;    a.s3 <<= b;    a.s4 <<= b;    a.s5 <<= b;    a.s6 <<= b;    a.s7 <<= b;     }
-inline __device__ void operator <<= (u32x &a, const u32x b) { a.s0 <<= b.s0; a.s1 <<= b.s1; a.s2 <<= b.s2; a.s3 <<= b.s3; a.s4 <<= b.s4; a.s5 <<= b.s5; a.s6 <<= b.s6; a.s7 <<= b.s7;  }
-
-inline __device__ u32x operator << (const u32x a, const u32  b) { return u32x ((a.s0 << b),    (a.s1 << b)   , (a.s2 << b),    (a.s3 << b)   , (a.s4 << b),    (a.s5 << b)   , (a.s6 << b),    (a.s7 << b)   );  }
-inline __device__ u32x operator << (const u32x a, const u32x b) { return u32x ((a.s0 << b.s0), (a.s1 << b.s1), (a.s2 << b.s2), (a.s3 << b.s3), (a.s4 << b.s4), (a.s5 << b.s5), (a.s6 << b.s6), (a.s7 << b.s7));  }
-
-inline __device__ u32x operator >> (const u32x a, const u32  b) { return u32x ((a.s0 >> b),    (a.s1 >> b)   , (a.s2 >> b),    (a.s3 >> b)   , (a.s4 >> b),    (a.s5 >> b)   , (a.s6 >> b),    (a.s7 >> b)   );  }
-inline __device__ u32x operator >> (const u32x a, const u32x b) { return u32x ((a.s0 >> b.s0), (a.s1 >> b.s1), (a.s2 >> b.s2), (a.s3 >> b.s3), (a.s4 >> b.s4), (a.s5 >> b.s5), (a.s6 >> b.s6), (a.s7 >> b.s7));  }
-
-inline __device__ u32x operator ^  (const u32x a, const u32  b) { return u32x ((a.s0 ^  b),    (a.s1 ^  b)   , (a.s2 ^  b),    (a.s3 ^  b)   , (a.s4 ^  b),    (a.s5 ^  b)   , (a.s6 ^  b),    (a.s7 ^  b)   );  }
-inline __device__ u32x operator ^  (const u32x a, const u32x b) { return u32x ((a.s0 ^  b.s0), (a.s1 ^  b.s1), (a.s2 ^  b.s2), (a.s3 ^  b.s3), (a.s4 ^  b.s4), (a.s5 ^  b.s5), (a.s6 ^  b.s6), (a.s7 ^  b.s7));  }
-
-inline __device__ u32x operator |  (const u32x a, const u32  b) { return u32x ((a.s0 |  b),    (a.s1 |  b)   , (a.s2 |  b),    (a.s3 |  b)   , (a.s4 |  b),    (a.s5 |  b)   , (a.s6 |  b),    (a.s7 |  b)   );  }
-inline __device__ u32x operator |  (const u32x a, const u32x b) { return u32x ((a.s0 |  b.s0), (a.s1 |  b.s1), (a.s2 |  b.s2), (a.s3 |  b.s3), (a.s4 |  b.s4), (a.s5 |  b.s5), (a.s6 |  b.s6), (a.s7 |  b.s7));  }
-
-inline __device__ u32x operator &  (const u32x a, const u32  b) { return u32x ((a.s0 &  b),    (a.s1 &  b)   , (a.s2 &  b),    (a.s3 &  b)   , (a.s4 &  b),    (a.s5 &  b)   , (a.s6 &  b),    (a.s7 &  b)   );  }
-inline __device__ u32x operator &  (const u32x a, const u32x b) { return u32x ((a.s0 &  b.s0), (a.s1 &  b.s1), (a.s2 &  b.s2), (a.s3 &  b.s3), (a.s4 &  b.s4), (a.s5 &  b.s5), (a.s6 &  b.s6), (a.s7 &  b.s7));  }
-
-inline __device__ u32x operator +  (const u32x a, const u32  b) { return u32x ((a.s0 +  b),    (a.s1 +  b)   , (a.s2 +  b),    (a.s3 +  b)   , (a.s4 +  b),    (a.s5 +  b)   , (a.s6 +  b),    (a.s7 +  b)   );  }
-inline __device__ u32x operator +  (const u32x a, const u32x b) { return u32x ((a.s0 +  b.s0), (a.s1 +  b.s1), (a.s2 +  b.s2), (a.s3 +  b.s3), (a.s4 +  b.s4), (a.s5 +  b.s5), (a.s6 +  b.s6), (a.s7 +  b.s7));  }
-
-inline __device__ u32x operator -  (const u32x a, const u32  b) { return u32x ((a.s0 -  b),    (a.s1 -  b)   , (a.s2 -  b),    (a.s3 -  b)   , (a.s4 -  b),    (a.s5 -  b)   , (a.s6 -  b),    (a.s7 -  b)   );  }
-inline __device__ u32x operator -  (const u32x a, const u32x b) { return u32x ((a.s0 -  b.s0), (a.s1 -  b.s1), (a.s2 -  b.s2), (a.s3 -  b.s3), (a.s4 -  b.s4), (a.s5 -  b.s5), (a.s6 -  b.s6), (a.s7 -  b.s7));  }
-
-inline __device__ u32x operator *  (const u32x a, const u32  b) { return u32x ((a.s0 *  b),    (a.s1 *  b)   , (a.s2 *  b),    (a.s3 *  b)   , (a.s4 *  b),    (a.s5 *  b)   , (a.s6 *  b),    (a.s7 *  b)   );  }
-inline __device__ u32x operator *  (const u32x a, const u32x b) { return u32x ((a.s0 *  b.s0), (a.s1 *  b.s1), (a.s2 *  b.s2), (a.s3 *  b.s3), (a.s4 *  b.s4), (a.s5 *  b.s5), (a.s6 *  b.s6), (a.s7 *  b.s7));  }
-
-inline __device__ u32x operator %  (const u32x a, const u32  b) { return u32x ((a.s0 %  b),    (a.s1 %  b)   , (a.s2 %  b),    (a.s3 %  b)   , (a.s4 %  b),    (a.s5 %  b)   , (a.s6 %  b),    (a.s7 %  b)   );  }
-inline __device__ u32x operator %  (const u32x a, const u32x b) { return u32x ((a.s0 %  b.s0), (a.s1 %  b.s1), (a.s2 %  b.s2), (a.s3 %  b.s3), (a.s4 %  b.s4), (a.s5 %  b.s5), (a.s6 %  b.s6), (a.s7 %  b.s7));  }
-
-inline __device__ u32x operator ~  (const u32x a) { return u32x (~a.s0, ~a.s1, ~a.s2, ~a.s3, ~a.s4, ~a.s5, ~a.s6, ~a.s7); }
-
-inline __device__ bool operator != (const u64x a, const u64  b) { return ((a.s0 != b)    && (a.s1 != b)    && (a.s2 != b)    && (a.s3 != b)    && (a.s4 != b)    && (a.s5 != b)    && (a.s6 != b)    && (a.s7 != b)   ); }
-inline __device__ bool operator != (const u64x a, const u64x b) { return ((a.s0 != b.s0) && (a.s1 != b.s1) && (a.s2 != b.s2) && (a.s3 != b.s3) && (a.s4 != b.s4) && (a.s5 != b.s5) && (a.s6 != b.s6) && (a.s7 != b.s7)); }
-
-inline __device__ void operator ^= (u64x &a, const u64  b) { a.s0 ^= b;    a.s1 ^= b;    a.s2 ^= b;    a.s3 ^= b;    a.s4 ^= b;    a.s5 ^= b;    a.s6 ^= b;    a.s7 ^= b;     }
-inline __device__ void operator ^= (u64x &a, const u64x b) { a.s0 ^= b.s0; a.s1 ^= b.s1; a.s2 ^= b.s2; a.s3 ^= b.s3; a.s4 ^= b.s4; a.s5 ^= b.s5; a.s6 ^= b.s6; a.s7 ^= b.s7;  }
-
-inline __device__ void operator |= (u64x &a, const u64  b) { a.s0 |= b;    a.s1 |= b;    a.s2 |= b;    a.s3 |= b;    a.s4 |= b;    a.s5 |= b;    a.s6 |= b;    a.s7 |= b;     }
-inline __device__ void operator |= (u64x &a, const u64x b) { a.s0 |= b.s0; a.s1 |= b.s1; a.s2 |= b.s2; a.s3 |= b.s3; a.s4 |= b.s4; a.s5 |= b.s5; a.s6 |= b.s6; a.s7 |= b.s7;  }
-
-inline __device__ void operator &= (u64x &a, const u64  b) { a.s0 &= b;    a.s1 &= b;    a.s2 &= b;    a.s3 &= b;    a.s4 &= b;    a.s5 &= b;    a.s6 &= b;    a.s7 &= b;     }
-inline __device__ void operator &= (u64x &a, const u64x b) { a.s0 &= b.s0; a.s1 &= b.s1; a.s2 &= b.s2; a.s3 &= b.s3; a.s4 &= b.s4; a.s5 &= b.s5; a.s6 &= b.s6; a.s7 &= b.s7;  }
-
-inline __device__ void operator += (u64x &a, const u64  b) { a.s0 += b;    a.s1 += b;    a.s2 += b;    a.s3 += b;    a.s4 += b;    a.s5 += b;    a.s6 += b;    a.s7 += b;     }
-inline __device__ void operator += (u64x &a, const u64x b) { a.s0 += b.s0; a.s1 += b.s1; a.s2 += b.s2; a.s3 += b.s3; a.s4 += b.s4; a.s5 += b.s5; a.s6 += b.s6; a.s7 += b.s7;  }
-
-inline __device__ void operator -= (u64x &a, const u64  b) { a.s0 -= b;    a.s1 -= b;    a.s2 -= b;    a.s3 -= b;    a.s4 -= b;    a.s5 -= b;    a.s6 -= b;    a.s7 -= b;     }
-inline __device__ void operator -= (u64x &a, const u64x b) { a.s0 -= b.s0; a.s1 -= b.s1; a.s2 -= b.s2; a.s3 -= b.s3; a.s4 -= b.s4; a.s5 -= b.s5; a.s6 -= b.s6; a.s7 -= b.s7;  }
-
-inline __device__ void operator *= (u64x &a, const u64  b) { a.s0 *= b;    a.s1 *= b;    a.s2 *= b;    a.s3 *= b;    a.s4 *= b;    a.s5 *= b;    a.s6 *= b;    a.s7 *= b;     }
-inline __device__ void operator *= (u64x &a, const u64x b) { a.s0 *= b.s0; a.s1 *= b.s1; a.s2 *= b.s2; a.s3 *= b.s3; a.s4 *= b.s4; a.s5 *= b.s5; a.s6 *= b.s6; a.s7 *= b.s7;  }
-
-inline __device__ void operator >>= (u64x &a, const u64  b) { a.s0 >>= b;    a.s1 >>= b;    a.s2 >>= b;    a.s3 >>= b;    a.s4 >>= b;    a.s5 >>= b;    a.s6 >>= b;    a.s7 >>= b;     }
-inline __device__ void operator >>= (u64x &a, const u64x b) { a.s0 >>= b.s0; a.s1 >>= b.s1; a.s2 >>= b.s2; a.s3 >>= b.s3; a.s4 >>= b.s4; a.s5 >>= b.s5; a.s6 >>= b.s6; a.s7 >>= b.s7;  }
-
-inline __device__ void operator <<= (u64x &a, const u64  b) { a.s0 <<= b;    a.s1 <<= b;    a.s2 <<= b;    a.s3 <<= b;    a.s4 <<= b;    a.s5 <<= b;    a.s6 <<= b;    a.s7 <<= b;     }
-inline __device__ void operator <<= (u64x &a, const u64x b) { a.s0 <<= b.s0; a.s1 <<= b.s1; a.s2 <<= b.s2; a.s3 <<= b.s3; a.s4 <<= b.s4; a.s5 <<= b.s5; a.s6 <<= b.s6; a.s7 <<= b.s7;  }
-
-inline __device__ u64x operator << (const u64x a, const u64  b) { return u64x ((a.s0 << b),    (a.s1 << b)   , (a.s2 << b),    (a.s3 << b)   , (a.s4 << b),    (a.s5 << b)   , (a.s6 << b),    (a.s7 << b)   );  }
-inline __device__ u64x operator << (const u64x a, const u64x b) { return u64x ((a.s0 << b.s0), (a.s1 << b.s1), (a.s2 << b.s2), (a.s3 << b.s3), (a.s4 << b.s4), (a.s5 << b.s5), (a.s6 << b.s6), (a.s7 << b.s7));  }
-
-inline __device__ u64x operator >> (const u64x a, const u64  b) { return u64x ((a.s0 >> b),    (a.s1 >> b)   , (a.s2 >> b),    (a.s3 >> b)   , (a.s4 >> b),    (a.s5 >> b)   , (a.s6 >> b),    (a.s7 >> b)   );  }
-inline __device__ u64x operator >> (const u64x a, const u64x b) { return u64x ((a.s0 >> b.s0), (a.s1 >> b.s1), (a.s2 >> b.s2), (a.s3 >> b.s3), (a.s4 >> b.s4), (a.s5 >> b.s5), (a.s6 >> b.s6), (a.s7 >> b.s7));  }
-
-inline __device__ u64x operator ^  (const u64x a, const u64  b) { return u64x ((a.s0 ^  b),    (a.s1 ^  b)   , (a.s2 ^  b),    (a.s3 ^  b)   , (a.s4 ^  b),    (a.s5 ^  b)   , (a.s6 ^  b),    (a.s7 ^  b)   );  }
-inline __device__ u64x operator ^  (const u64x a, const u64x b) { return u64x ((a.s0 ^  b.s0), (a.s1 ^  b.s1), (a.s2 ^  b.s2), (a.s3 ^  b.s3), (a.s4 ^  b.s4), (a.s5 ^  b.s5), (a.s6 ^  b.s6), (a.s7 ^  b.s7));  }
-
-inline __device__ u64x operator |  (const u64x a, const u64  b) { return u64x ((a.s0 |  b),    (a.s1 |  b)   , (a.s2 |  b),    (a.s3 |  b)   , (a.s4 |  b),    (a.s5 |  b)   , (a.s6 |  b),    (a.s7 |  b)   );  }
-inline __device__ u64x operator |  (const u64x a, const u64x b) { return u64x ((a.s0 |  b.s0), (a.s1 |  b.s1), (a.s2 |  b.s2), (a.s3 |  b.s3), (a.s4 |  b.s4), (a.s5 |  b.s5), (a.s6 |  b.s6), (a.s7 |  b.s7));  }
-
-inline __device__ u64x operator &  (const u64x a, const u64  b) { return u64x ((a.s0 &  b),    (a.s1 &  b)   , (a.s2 &  b),    (a.s3 &  b)   , (a.s4 &  b),    (a.s5 &  b)   , (a.s6 &  b),    (a.s7 &  b)   );  }
-inline __device__ u64x operator &  (const u64x a, const u64x b) { return u64x ((a.s0 &  b.s0), (a.s1 &  b.s1), (a.s2 &  b.s2), (a.s3 &  b.s3), (a.s4 &  b.s4), (a.s5 &  b.s5), (a.s6 &  b.s6), (a.s7 &  b.s7));  }
-
-inline __device__ u64x operator +  (const u64x a, const u64  b) { return u64x ((a.s0 +  b),    (a.s1 +  b)   , (a.s2 +  b),    (a.s3 +  b)   , (a.s4 +  b),    (a.s5 +  b)   , (a.s6 +  b),    (a.s7 +  b)   );  }
-inline __device__ u64x operator +  (const u64x a, const u64x b) { return u64x ((a.s0 +  b.s0), (a.s1 +  b.s1), (a.s2 +  b.s2), (a.s3 +  b.s3), (a.s4 +  b.s4), (a.s5 +  b.s5), (a.s6 +  b.s6), (a.s7 +  b.s7));  }
-
-inline __device__ u64x operator -  (const u64x a, const u64  b) { return u64x ((a.s0 -  b),    (a.s1 -  b)   , (a.s2 -  b),    (a.s3 -  b)   , (a.s4 -  b),    (a.s5 -  b)   , (a.s6 -  b),    (a.s7 -  b)   );  }
-inline __device__ u64x operator -  (const u64x a, const u64x b) { return u64x ((a.s0 -  b.s0), (a.s1 -  b.s1), (a.s2 -  b.s2), (a.s3 -  b.s3), (a.s4 -  b.s4), (a.s5 -  b.s5), (a.s6 -  b.s6), (a.s7 -  b.s7));  }
-
-inline __device__ u64x operator *  (const u64x a, const u64  b) { return u64x ((a.s0 *  b),    (a.s1 *  b)   , (a.s2 *  b),    (a.s3 *  b)   , (a.s4 *  b),    (a.s5 *  b)   , (a.s6 *  b),    (a.s7 *  b)   );  }
-inline __device__ u64x operator *  (const u64x a, const u64x b) { return u64x ((a.s0 *  b.s0), (a.s1 *  b.s1), (a.s2 *  b.s2), (a.s3 *  b.s3), (a.s4 *  b.s4), (a.s5 *  b.s5), (a.s6 *  b.s6), (a.s7 *  b.s7));  }
-
-inline __device__ u64x operator %  (const u64x a, const u64  b) { return u64x ((a.s0 %  b),    (a.s1 %  b)   , (a.s2 %  b),    (a.s3 %  b)   , (a.s4 %  b),    (a.s5 %  b)   , (a.s6 %  b),    (a.s7 %  b)   );  }
-inline __device__ u64x operator %  (const u64x a, const u64x b) { return u64x ((a.s0 %  b.s0), (a.s1 %  b.s1), (a.s2 %  b.s2), (a.s3 %  b.s3), (a.s4 %  b.s4), (a.s5 %  b.s5), (a.s6 %  b.s6), (a.s7 %  b.s7));  }
-
-inline __device__ u64x operator ~  (const u64x a) { return u64x (~a.s0, ~a.s1, ~a.s2, ~a.s3, ~a.s4, ~a.s5, ~a.s6, ~a.s7); }
-
-#endif
-
-#if VECT_SIZE == 16
-
-struct u8x
-{
-  u8 s0;
-  u8 s1;
-  u8 s2;
-  u8 s3;
-  u8 s4;
-  u8 s5;
-  u8 s6;
-  u8 s7;
-  u8 s8;
-  u8 s9;
-  u8 sa;
-  u8 sb;
-  u8 sc;
-  u8 sd;
-  u8 se;
-  u8 sf;
-
-  inline __device__  u8x (const u8 a, const u8 b, const u8 c, const u8 d, const u8 e, const u8 f, const u8 g, const u8 h, const u8 i, const u8 j, const u8 k, const u8 l, const u8 m, const u8 n, const u8 o, const u8 p) : s0(a), s1(b), s2(c), s3(d), s4(e), s5(f), s6(g), s7(h), s8(i), s9(j), sa(k), sb(l), sc(m), sd(n), se(o), sf(p) { }
-  inline __device__  u8x (const u8 a)                                                                                                                                                                                     : s0(a), s1(a), s2(a), s3(a), s4(a), s5(a), s6(a), s7(a), s8(a), s9(a), sa(a), sb(a), sc(a), sd(a), se(a), sf(a) { }
-
-  inline __device__  u8x (void) : s0(0), s1(0), s2(0), s3(0), s4(0), s5(0), s6(0), s7(0), s8(0), s9(0), sa(0), sb(0), sc(0), sd(0), se(0), sf(0) { }
-  inline __device__ ~u8x (void) { }
-};
-
-struct u16x
-{
-  u16 s0;
-  u16 s1;
-  u16 s2;
-  u16 s3;
-  u16 s4;
-  u16 s5;
-  u16 s6;
-  u16 s7;
-  u16 s8;
-  u16 s9;
-  u16 sa;
-  u16 sb;
-  u16 sc;
-  u16 sd;
-  u16 se;
-  u16 sf;
-
-  inline __device__  u16x (const u16 a, const u16 b, const u16 c, const u16 d, const u16 e, const u16 f, const u16 g, const u16 h, const u16 i, const u16 j, const u16 k, const u16 l, const u16 m, const u16 n, const u16 o, const u16 p) : s0(a), s1(b), s2(c), s3(d), s4(e), s5(f), s6(g), s7(h), s8(i), s9(j), sa(k), sb(l), sc(m), sd(n), se(o), sf(p) { }
-  inline __device__  u16x (const u16 a)                                                                                                                                                                                     : s0(a), s1(a), s2(a), s3(a), s4(a), s5(a), s6(a), s7(a), s8(a), s9(a), sa(a), sb(a), sc(a), sd(a), se(a), sf(a) { }
-
-  inline __device__  u16x (void) : s0(0), s1(0), s2(0), s3(0), s4(0), s5(0), s6(0), s7(0), s8(0), s9(0), sa(0), sb(0), sc(0), sd(0), se(0), sf(0){ }
-  inline __device__ ~u16x (void) { }
-};
-
-struct u32x
-{
-  u32 s0;
-  u32 s1;
-  u32 s2;
-  u32 s3;
-  u32 s4;
-  u32 s5;
-  u32 s6;
-  u32 s7;
-  u32 s8;
-  u32 s9;
-  u32 sa;
-  u32 sb;
-  u32 sc;
-  u32 sd;
-  u32 se;
-  u32 sf;
-
-  inline __device__  u32x (const u32 a, const u32 b, const u32 c, const u32 d, const u32 e, const u32 f, const u32 g, const u32 h, const u32 i, const u32 j, const u32 k, const u32 l, const u32 m, const u32 n, const u32 o, const u32 p) : s0(a), s1(b), s2(c), s3(d), s4(e), s5(f), s6(g), s7(h), s8(i), s9(j), sa(k), sb(l), sc(m), sd(n), se(o), sf(p) { }
-  inline __device__  u32x (const u32 a)                                                                                                                                                                                     : s0(a), s1(a), s2(a), s3(a), s4(a), s5(a), s6(a), s7(a), s8(a), s9(a), sa(a), sb(a), sc(a), sd(a), se(a), sf(a) { }
-
-  inline __device__  u32x (void) : s0(0), s1(0), s2(0), s3(0), s4(0), s5(0), s6(0), s7(0), s8(0), s9(0), sa(0), sb(0), sc(0), sd(0), se(0), sf(0){ }
-  inline __device__ ~u32x (void) { }
-};
-
-struct u64x
-{
-  u64 s0;
-  u64 s1;
-  u64 s2;
-  u64 s3;
-  u64 s4;
-  u64 s5;
-  u64 s6;
-  u64 s7;
-  u64 s8;
-  u64 s9;
-  u64 sa;
-  u64 sb;
-  u64 sc;
-  u64 sd;
-  u64 se;
-  u64 sf;
-
-  inline __device__  u64x (const u64 a, const u64 b, const u64 c, const u64 d, const u64 e, const u64 f, const u64 g, const u64 h, const u64 i, const u64 j, const u64 k, const u64 l, const u64 m, const u64 n, const u64 o, const u64 p) : s0(a), s1(b), s2(c), s3(d), s4(e), s5(f), s6(g), s7(h), s8(i), s9(j), sa(k), sb(l), sc(m), sd(n), se(o), sf(p) { }
-  inline __device__  u64x (const u64 a)                                                                                                                                                                                     : s0(a), s1(a), s2(a), s3(a), s4(a), s5(a), s6(a), s7(a), s8(a), s9(a), sa(a), sb(a), sc(a), sd(a), se(a), sf(a) { }
-
-  inline __device__  u64x (void) : s0(0), s1(0), s2(0), s3(0), s4(0), s5(0), s6(0), s7(0), s8(0), s9(0), sa(0), sb(0), sc(0), sd(0), se(0), sf(0) { }
-  inline __device__ ~u64x (void) { }
-};
-
-inline __device__ bool operator != (const u32x a, const u32  b) { return ((a.s0 != b)    && (a.s1 != b)    && (a.s2 != b)    && (a.s3 != b)    && (a.s4 != b)    && (a.s5 != b)    && (a.s6 != b)    && (a.s7 != b)    && (a.s8 != b)    && (a.s9 != b)    && (a.sa != b)    && (a.sb != b)    && (a.sc != b)    && (a.sd != b)    && (a.se != b)    && (a.sf != b)   ); }
-inline __device__ bool operator != (const u32x a, const u32x b) { return ((a.s0 != b.s0) && (a.s1 != b.s1) && (a.s2 != b.s2) && (a.s3 != b.s3) && (a.s4 != b.s4) && (a.s5 != b.s5) && (a.s6 != b.s6) && (a.s7 != b.s7) && (a.s8 != b.s8) && (a.s9 != b.s9) && (a.sa != b.sa) && (a.sb != b.sb) && (a.sc != b.sc) && (a.sd != b.sd) && (a.se != b.se) && (a.sf != b.sf)); }
-
-inline __device__ void operator ^= (u32x &a, const u32  b) { a.s0 ^= b;    a.s1 ^= b;    a.s2 ^= b;    a.s3 ^= b;    a.s4 ^= b;    a.s5 ^= b;    a.s6 ^= b;    a.s7 ^= b;    a.s8 ^= b;    a.s9 ^= b;    a.sa ^= b;    a.sb ^= b;    a.sc ^= b;    a.sd ^= b;    a.se ^= b;    a.sf ^= b;    }
-inline __device__ void operator ^= (u32x &a, const u32x b) { a.s0 ^= b.s0; a.s1 ^= b.s1; a.s2 ^= b.s2; a.s3 ^= b.s3; a.s4 ^= b.s4; a.s5 ^= b.s5; a.s6 ^= b.s6; a.s7 ^= b.s7; a.s8 ^= b.s8; a.s9 ^= b.s9; a.sa ^= b.sa; a.sb ^= b.sb; a.sc ^= b.sc; a.sd ^= b.sd; a.se ^= b.se; a.sf ^= b.sf; }
-
-inline __device__ void operator |= (u32x &a, const u32  b) { a.s0 |= b;    a.s1 |= b;    a.s2 |= b;    a.s3 |= b;    a.s4 |= b;    a.s5 |= b;    a.s6 |= b;    a.s7 |= b;    a.s8 |= b;    a.s9 |= b;    a.sa |= b;    a.sb |= b;    a.sc |= b;    a.sd |= b;    a.se |= b;    a.sf |= b;    }
-inline __device__ void operator |= (u32x &a, const u32x b) { a.s0 |= b.s0; a.s1 |= b.s1; a.s2 |= b.s2; a.s3 |= b.s3; a.s4 |= b.s4; a.s5 |= b.s5; a.s6 |= b.s6; a.s7 |= b.s7; a.s8 |= b.s8; a.s9 |= b.s9; a.sa |= b.sa; a.sb |= b.sb; a.sc |= b.sc; a.sd |= b.sd; a.se |= b.se; a.sf |= b.sf; }
-
-inline __device__ void operator &= (u32x &a, const u32  b) { a.s0 &= b;    a.s1 &= b;    a.s2 &= b;    a.s3 &= b;    a.s4 &= b;    a.s5 &= b;    a.s6 &= b;    a.s7 &= b;    a.s8 &= b;    a.s9 &= b;    a.sa &= b;    a.sb &= b;    a.sc &= b;    a.sd &= b;    a.se &= b;    a.sf &= b;    }
-inline __device__ void operator &= (u32x &a, const u32x b) { a.s0 &= b.s0; a.s1 &= b.s1; a.s2 &= b.s2; a.s3 &= b.s3; a.s4 &= b.s4; a.s5 &= b.s5; a.s6 &= b.s6; a.s7 &= b.s7; a.s8 &= b.s8; a.s9 &= b.s9; a.sa &= b.sa; a.sb &= b.sb; a.sc &= b.sc; a.sd &= b.sd; a.se &= b.se; a.sf &= b.sf; }
-
-inline __device__ void operator += (u32x &a, const u32  b) { a.s0 += b;    a.s1 += b;    a.s2 += b;    a.s3 += b;    a.s4 += b;    a.s5 += b;    a.s6 += b;    a.s7 += b;    a.s8 += b;    a.s9 += b;    a.sa += b;    a.sb += b;    a.sc += b;    a.sd += b;    a.se += b;    a.sf += b;    }
-inline __device__ void operator += (u32x &a, const u32x b) { a.s0 += b.s0; a.s1 += b.s1; a.s2 += b.s2; a.s3 += b.s3; a.s4 += b.s4; a.s5 += b.s5; a.s6 += b.s6; a.s7 += b.s7; a.s8 += b.s8; a.s9 += b.s9; a.sa += b.sa; a.sb += b.sb; a.sc += b.sc; a.sd += b.sd; a.se += b.se; a.sf += b.sf; }
-
-inline __device__ void operator -= (u32x &a, const u32  b) { a.s0 -= b;    a.s1 -= b;    a.s2 -= b;    a.s3 -= b;    a.s4 -= b;    a.s5 -= b;    a.s6 -= b;    a.s7 -= b;    a.s8 -= b;    a.s9 -= b;    a.sa -= b;    a.sb -= b;    a.sc -= b;    a.sd -= b;    a.se -= b;    a.sf -= b;    }
-inline __device__ void operator -= (u32x &a, const u32x b) { a.s0 -= b.s0; a.s1 -= b.s1; a.s2 -= b.s2; a.s3 -= b.s3; a.s4 -= b.s4; a.s5 -= b.s5; a.s6 -= b.s6; a.s7 -= b.s7; a.s8 -= b.s8; a.s9 -= b.s9; a.sa -= b.sa; a.sb -= b.sb; a.sc -= b.sc; a.sd -= b.sd; a.se -= b.se; a.sf -= b.sf; }
-
-inline __device__ void operator *= (u32x &a, const u32  b) { a.s0 *= b;    a.s1 *= b;    a.s2 *= b;    a.s3 *= b;    a.s4 *= b;    a.s5 *= b;    a.s6 *= b;    a.s7 *= b;    a.s8 *= b;    a.s9 *= b;    a.sa *= b;    a.sb *= b;    a.sc *= b;    a.sd *= b;    a.se *= b;    a.sf *= b;    }
-inline __device__ void operator *= (u32x &a, const u32x b) { a.s0 *= b.s0; a.s1 *= b.s1; a.s2 *= b.s2; a.s3 *= b.s3; a.s4 *= b.s4; a.s5 *= b.s5; a.s6 *= b.s6; a.s7 *= b.s7; a.s8 *= b.s8; a.s9 *= b.s9; a.sa *= b.sa; a.sb *= b.sb; a.sc *= b.sc; a.sd *= b.sd; a.se *= b.se; a.sf *= b.sf; }
-
-inline __device__ void operator >>= (u32x &a, const u32  b) { a.s0 >>= b;    a.s1 >>= b;    a.s2 >>= b;    a.s3 >>= b;    a.s4 >>= b;    a.s5 >>= b;    a.s6 >>= b;    a.s7 >>= b;    a.s8 >>= b;    a.s9 >>= b;    a.sa >>= b;    a.sb >>= b;    a.sc >>= b;    a.sd >>= b;    a.se >>= b;    a.sf >>= b;    }
-inline __device__ void operator >>= (u32x &a, const u32x b) { a.s0 >>= b.s0; a.s1 >>= b.s1; a.s2 >>= b.s2; a.s3 >>= b.s3; a.s4 >>= b.s4; a.s5 >>= b.s5; a.s6 >>= b.s6; a.s7 >>= b.s7; a.s8 >>= b.s8; a.s9 >>= b.s9; a.sa >>= b.sa; a.sb >>= b.sb; a.sc >>= b.sc; a.sd >>= b.sd; a.se >>= b.se; a.sf >>= b.sf; }
-
-inline __device__ void operator <<= (u32x &a, const u32  b) { a.s0 <<= b;    a.s1 <<= b;    a.s2 <<= b;    a.s3 <<= b;    a.s4 <<= b;    a.s5 <<= b;    a.s6 <<= b;    a.s7 <<= b;    a.s8 <<= b;    a.s9 <<= b;    a.sa <<= b;    a.sb <<= b;    a.sc <<= b;    a.sd <<= b;    a.se <<= b;    a.sf <<= b;    }
-inline __device__ void operator <<= (u32x &a, const u32x b) { a.s0 <<= b.s0; a.s1 <<= b.s1; a.s2 <<= b.s2; a.s3 <<= b.s3; a.s4 <<= b.s4; a.s5 <<= b.s5; a.s6 <<= b.s6; a.s7 <<= b.s7; a.s8 <<= b.s8; a.s9 <<= b.s9; a.sa <<= b.sa; a.sb <<= b.sb; a.sc <<= b.sc; a.sd <<= b.sd; a.se <<= b.se; a.sf <<= b.sf; }
-
-inline __device__ u32x operator << (const u32x a, const u32  b) { return u32x ((a.s0 << b),    (a.s1 << b)   , (a.s2 << b),    (a.s3 << b)   , (a.s4 << b),    (a.s5 << b)   , (a.s6 << b),    (a.s7 << b),    (a.s8 << b),    (a.s9 << b)   , (a.sa << b),    (a.sb << b)   , (a.sc << b),    (a.sd << b)   , (a.se << b),    (a.sf << b)   );  }
-inline __device__ u32x operator << (const u32x a, const u32x b) { return u32x ((a.s0 << b.s0), (a.s1 << b.s1), (a.s2 << b.s2), (a.s3 << b.s3), (a.s4 << b.s4), (a.s5 << b.s5), (a.s6 << b.s6), (a.s7 << b.s7), (a.s8 << b.s8), (a.s9 << b.s9), (a.sa << b.sa), (a.sb << b.sb), (a.sc << b.sc), (a.sd << b.sd), (a.se << b.se), (a.sf << b.sf));  }
-
-inline __device__ u32x operator >> (const u32x a, const u32  b) { return u32x ((a.s0 >> b),    (a.s1 >> b)   , (a.s2 >> b),    (a.s3 >> b)   , (a.s4 >> b),    (a.s5 >> b)   , (a.s6 >> b),    (a.s7 >> b),    (a.s8 >> b),    (a.s9 >> b)   , (a.sa >> b),    (a.sb >> b)   , (a.sc >> b),    (a.sd >> b)   , (a.se >> b),    (a.sf >> b)   );  }
-inline __device__ u32x operator >> (const u32x a, const u32x b) { return u32x ((a.s0 >> b.s0), (a.s1 >> b.s1), (a.s2 >> b.s2), (a.s3 >> b.s3), (a.s4 >> b.s4), (a.s5 >> b.s5), (a.s6 >> b.s6), (a.s7 >> b.s7), (a.s8 >> b.s8), (a.s9 >> b.s9), (a.sa >> b.sa), (a.sb >> b.sb), (a.sc >> b.sc), (a.sd >> b.sd), (a.se >> b.se), (a.sf >> b.sf));  }
-
-inline __device__ u32x operator ^  (const u32x a, const u32  b) { return u32x ((a.s0 ^  b),    (a.s1 ^  b)   , (a.s2 ^  b),    (a.s3 ^  b)   , (a.s4 ^  b),    (a.s5 ^  b)   , (a.s6 ^  b),    (a.s7 ^  b),    (a.s8 ^  b),    (a.s9 ^  b)   , (a.sa ^  b),    (a.sb ^  b)   , (a.sc ^  b),    (a.sd ^  b)   , (a.se ^  b),    (a.sf ^  b)   );  }
-inline __device__ u32x operator ^  (const u32x a, const u32x b) { return u32x ((a.s0 ^  b.s0), (a.s1 ^  b.s1), (a.s2 ^  b.s2), (a.s3 ^  b.s3), (a.s4 ^  b.s4), (a.s5 ^  b.s5), (a.s6 ^  b.s6), (a.s7 ^  b.s7), (a.s8 ^  b.s8), (a.s9 ^  b.s9), (a.sa ^  b.sa), (a.sb ^  b.sb), (a.sc ^  b.sc), (a.sd ^  b.sd), (a.se ^  b.se), (a.sf ^  b.sf));  }
-
-inline __device__ u32x operator |  (const u32x a, const u32  b) { return u32x ((a.s0 |  b),    (a.s1 |  b)   , (a.s2 |  b),    (a.s3 |  b)   , (a.s4 |  b),    (a.s5 |  b)   , (a.s6 |  b),    (a.s7 |  b),    (a.s8 |  b),    (a.s9 |  b)   , (a.sa |  b),    (a.sb |  b)   , (a.sc |  b),    (a.sd |  b)   , (a.se |  b),    (a.sf |  b)   );  }
-inline __device__ u32x operator |  (const u32x a, const u32x b) { return u32x ((a.s0 |  b.s0), (a.s1 |  b.s1), (a.s2 |  b.s2), (a.s3 |  b.s3), (a.s4 |  b.s4), (a.s5 |  b.s5), (a.s6 |  b.s6), (a.s7 |  b.s7), (a.s8 |  b.s8), (a.s9 |  b.s9), (a.sa |  b.sa), (a.sb |  b.sb), (a.sc |  b.sc), (a.sd |  b.sd), (a.se |  b.se), (a.sf |  b.sf));  }
-
-inline __device__ u32x operator &  (const u32x a, const u32  b) { return u32x ((a.s0 &  b),    (a.s1 &  b)   , (a.s2 &  b),    (a.s3 &  b)   , (a.s4 &  b),    (a.s5 &  b)   , (a.s6 &  b),    (a.s7 &  b),    (a.s8 &  b),    (a.s9 &  b)   , (a.sa &  b),    (a.sb &  b)   , (a.sc &  b),    (a.sd &  b)   , (a.se &  b),    (a.sf &  b)   );  }
-inline __device__ u32x operator &  (const u32x a, const u32x b) { return u32x ((a.s0 &  b.s0), (a.s1 &  b.s1), (a.s2 &  b.s2), (a.s3 &  b.s3), (a.s4 &  b.s4), (a.s5 &  b.s5), (a.s6 &  b.s6), (a.s7 &  b.s7), (a.s8 &  b.s8), (a.s9 &  b.s9), (a.sa &  b.sa), (a.sb &  b.sb), (a.sc &  b.sc), (a.sd &  b.sd), (a.se &  b.se), (a.sf &  b.sf));  }
-
-inline __device__ u32x operator +  (const u32x a, const u32  b) { return u32x ((a.s0 +  b),    (a.s1 +  b)   , (a.s2 +  b),    (a.s3 +  b)   , (a.s4 +  b),    (a.s5 +  b)   , (a.s6 +  b),    (a.s7 +  b),    (a.s8 +  b),    (a.s9 +  b)   , (a.sa +  b),    (a.sb +  b)   , (a.sc +  b),    (a.sd +  b)   , (a.se +  b),    (a.sf +  b)   );  }
-inline __device__ u32x operator +  (const u32x a, const u32x b) { return u32x ((a.s0 +  b.s0), (a.s1 +  b.s1), (a.s2 +  b.s2), (a.s3 +  b.s3), (a.s4 +  b.s4), (a.s5 +  b.s5), (a.s6 +  b.s6), (a.s7 +  b.s7), (a.s8 +  b.s8), (a.s9 +  b.s9), (a.sa +  b.sa), (a.sb +  b.sb), (a.sc +  b.sc), (a.sd +  b.sd), (a.se +  b.se), (a.sf +  b.sf));  }
-
-inline __device__ u32x operator -  (const u32x a, const u32  b) { return u32x ((a.s0 -  b),    (a.s1 -  b)   , (a.s2 -  b),    (a.s3 -  b)   , (a.s4 -  b),    (a.s5 -  b)   , (a.s6 -  b),    (a.s7 -  b),    (a.s8 -  b),    (a.s9 -  b)   , (a.sa -  b),    (a.sb -  b)   , (a.sc -  b),    (a.sd -  b)   , (a.se -  b),    (a.sf -  b)   );  }
-inline __device__ u32x operator -  (const u32x a, const u32x b) { return u32x ((a.s0 -  b.s0), (a.s1 -  b.s1), (a.s2 -  b.s2), (a.s3 -  b.s3), (a.s4 -  b.s4), (a.s5 -  b.s5), (a.s6 -  b.s6), (a.s7 -  b.s7), (a.s8 -  b.s8), (a.s9 -  b.s9), (a.sa -  b.sa), (a.sb -  b.sb), (a.sc -  b.sc), (a.sd -  b.sd), (a.se -  b.se), (a.sf -  b.sf));  }
-
-inline __device__ u32x operator *  (const u32x a, const u32  b) { return u32x ((a.s0 *  b),    (a.s1 *  b)   , (a.s2 *  b),    (a.s3 *  b)   , (a.s4 *  b),    (a.s5 *  b)   , (a.s6 *  b),    (a.s7 *  b),    (a.s8 *  b),    (a.s9 *  b)   , (a.sa *  b),    (a.sb *  b)   , (a.sc *  b),    (a.sd *  b)   , (a.se *  b),    (a.sf *  b)   );  }
-inline __device__ u32x operator *  (const u32x a, const u32x b) { return u32x ((a.s0 *  b.s0), (a.s1 *  b.s1), (a.s2 *  b.s2), (a.s3 *  b.s3), (a.s4 *  b.s4), (a.s5 *  b.s5), (a.s6 *  b.s6), (a.s7 *  b.s7), (a.s8 *  b.s8), (a.s9 *  b.s9), (a.sa *  b.sa), (a.sb *  b.sb), (a.sc *  b.sc), (a.sd *  b.sd), (a.se *  b.se), (a.sf *  b.sf));  }
-
-inline __device__ u32x operator %  (const u32x a, const u32  b) { return u32x ((a.s0 %  b),    (a.s1 %  b)   , (a.s2 %  b),    (a.s3 %  b)   , (a.s4 %  b),    (a.s5 %  b)   , (a.s6 %  b),    (a.s7 %  b),    (a.s8 %  b),    (a.s9 %  b)   , (a.sa %  b),    (a.sb %  b)   , (a.sc %  b),    (a.sd %  b)   , (a.se %  b),    (a.sf %  b)   );  }
-inline __device__ u32x operator %  (const u32x a, const u32x b) { return u32x ((a.s0 %  b.s0), (a.s1 %  b.s1), (a.s2 %  b.s2), (a.s3 %  b.s3), (a.s4 %  b.s4), (a.s5 %  b.s5), (a.s6 %  b.s6), (a.s7 %  b.s7), (a.s8 %  b.s8), (a.s9 %  b.s9), (a.sa %  b.sa), (a.sb %  b.sb), (a.sc %  b.sc), (a.sd %  b.sd), (a.se %  b.se), (a.sf %  b.sf));  }
-
-inline __device__ u32x operator ~  (const u32x a) { return u32x (~a.s0, ~a.s1, ~a.s2, ~a.s3, ~a.s4, ~a.s5, ~a.s6, ~a.s7, ~a.s8, ~a.s9, ~a.sa, ~a.sb, ~a.sc, ~a.sd, ~a.se, ~a.sf); }
-
-inline __device__ bool operator != (const u64x a, const u64  b) { return ((a.s0 != b)    && (a.s1 != b)    && (a.s2 != b)    && (a.s3 != b)    && (a.s4 != b)    && (a.s5 != b)    && (a.s6 != b)    && (a.s7 != b)    && (a.s8 != b)    && (a.s9 != b)    && (a.sa != b)    && (a.sb != b)    && (a.sc != b)    && (a.sd != b)    && (a.se != b)    && (a.sf != b)   ); }
-inline __device__ bool operator != (const u64x a, const u64x b) { return ((a.s0 != b.s0) && (a.s1 != b.s1) && (a.s2 != b.s2) && (a.s3 != b.s3) && (a.s4 != b.s4) && (a.s5 != b.s5) && (a.s6 != b.s6) && (a.s7 != b.s7) && (a.s8 != b.s8) && (a.s9 != b.s9) && (a.sa != b.sa) && (a.sb != b.sb) && (a.sc != b.sc) && (a.sd != b.sd) && (a.se != b.se) && (a.sf != b.sf)); }
-
-inline __device__ void operator ^= (u64x &a, const u64  b) { a.s0 ^= b;    a.s1 ^= b;    a.s2 ^= b;    a.s3 ^= b;    a.s4 ^= b;    a.s5 ^= b;    a.s6 ^= b;    a.s7 ^= b;    a.s8 ^= b;    a.s9 ^= b;    a.sa ^= b;    a.sb ^= b;    a.sc ^= b;    a.sd ^= b;    a.se ^= b;    a.sf ^= b;    }
-inline __device__ void operator ^= (u64x &a, const u64x b) { a.s0 ^= b.s0; a.s1 ^= b.s1; a.s2 ^= b.s2; a.s3 ^= b.s3; a.s4 ^= b.s4; a.s5 ^= b.s5; a.s6 ^= b.s6; a.s7 ^= b.s7; a.s8 ^= b.s8; a.s9 ^= b.s9; a.sa ^= b.sa; a.sb ^= b.sb; a.sc ^= b.sc; a.sd ^= b.sd; a.se ^= b.se; a.sf ^= b.sf; }
-
-inline __device__ void operator |= (u64x &a, const u64  b) { a.s0 |= b;    a.s1 |= b;    a.s2 |= b;    a.s3 |= b;    a.s4 |= b;    a.s5 |= b;    a.s6 |= b;    a.s7 |= b;    a.s8 |= b;    a.s9 |= b;    a.sa |= b;    a.sb |= b;    a.sc |= b;    a.sd |= b;    a.se |= b;    a.sf |= b;    }
-inline __device__ void operator |= (u64x &a, const u64x b) { a.s0 |= b.s0; a.s1 |= b.s1; a.s2 |= b.s2; a.s3 |= b.s3; a.s4 |= b.s4; a.s5 |= b.s5; a.s6 |= b.s6; a.s7 |= b.s7; a.s8 |= b.s8; a.s9 |= b.s9; a.sa |= b.sa; a.sb |= b.sb; a.sc |= b.sc; a.sd |= b.sd; a.se |= b.se; a.sf |= b.sf; }
-
-inline __device__ void operator &= (u64x &a, const u64  b) { a.s0 &= b;    a.s1 &= b;    a.s2 &= b;    a.s3 &= b;    a.s4 &= b;    a.s5 &= b;    a.s6 &= b;    a.s7 &= b;    a.s8 &= b;    a.s9 &= b;    a.sa &= b;    a.sb &= b;    a.sc &= b;    a.sd &= b;    a.se &= b;    a.sf &= b;    }
-inline __device__ void operator &= (u64x &a, const u64x b) { a.s0 &= b.s0; a.s1 &= b.s1; a.s2 &= b.s2; a.s3 &= b.s3; a.s4 &= b.s4; a.s5 &= b.s5; a.s6 &= b.s6; a.s7 &= b.s7; a.s8 &= b.s8; a.s9 &= b.s9; a.sa &= b.sa; a.sb &= b.sb; a.sc &= b.sc; a.sd &= b.sd; a.se &= b.se; a.sf &= b.sf; }
-
-inline __device__ void operator += (u64x &a, const u64  b) { a.s0 += b;    a.s1 += b;    a.s2 += b;    a.s3 += b;    a.s4 += b;    a.s5 += b;    a.s6 += b;    a.s7 += b;    a.s8 += b;    a.s9 += b;    a.sa += b;    a.sb += b;    a.sc += b;    a.sd += b;    a.se += b;    a.sf += b;    }
-inline __device__ void operator += (u64x &a, const u64x b) { a.s0 += b.s0; a.s1 += b.s1; a.s2 += b.s2; a.s3 += b.s3; a.s4 += b.s4; a.s5 += b.s5; a.s6 += b.s6; a.s7 += b.s7; a.s8 += b.s8; a.s9 += b.s9; a.sa += b.sa; a.sb += b.sb; a.sc += b.sc; a.sd += b.sd; a.se += b.se; a.sf += b.sf; }
-
-inline __device__ void operator -= (u64x &a, const u64  b) { a.s0 -= b;    a.s1 -= b;    a.s2 -= b;    a.s3 -= b;    a.s4 -= b;    a.s5 -= b;    a.s6 -= b;    a.s7 -= b;    a.s8 -= b;    a.s9 -= b;    a.sa -= b;    a.sb -= b;    a.sc -= b;    a.sd -= b;    a.se -= b;    a.sf -= b;    }
-inline __device__ void operator -= (u64x &a, const u64x b) { a.s0 -= b.s0; a.s1 -= b.s1; a.s2 -= b.s2; a.s3 -= b.s3; a.s4 -= b.s4; a.s5 -= b.s5; a.s6 -= b.s6; a.s7 -= b.s7; a.s8 -= b.s8; a.s9 -= b.s9; a.sa -= b.sa; a.sb -= b.sb; a.sc -= b.sc; a.sd -= b.sd; a.se -= b.se; a.sf -= b.sf; }
-
-inline __device__ void operator *= (u64x &a, const u64  b) { a.s0 *= b;    a.s1 *= b;    a.s2 *= b;    a.s3 *= b;    a.s4 *= b;    a.s5 *= b;    a.s6 *= b;    a.s7 *= b;    a.s8 *= b;    a.s9 *= b;    a.sa *= b;    a.sb *= b;    a.sc *= b;    a.sd *= b;    a.se *= b;    a.sf *= b;    }
-inline __device__ void operator *= (u64x &a, const u64x b) { a.s0 *= b.s0; a.s1 *= b.s1; a.s2 *= b.s2; a.s3 *= b.s3; a.s4 *= b.s4; a.s5 *= b.s5; a.s6 *= b.s6; a.s7 *= b.s7; a.s8 *= b.s8; a.s9 *= b.s9; a.sa *= b.sa; a.sb *= b.sb; a.sc *= b.sc; a.sd *= b.sd; a.se *= b.se; a.sf *= b.sf; }
-
-inline __device__ void operator >>= (u64x &a, const u64  b) { a.s0 >>= b;    a.s1 >>= b;    a.s2 >>= b;    a.s3 >>= b;    a.s4 >>= b;    a.s5 >>= b;    a.s6 >>= b;    a.s7 >>= b;    a.s8 >>= b;    a.s9 >>= b;    a.sa >>= b;    a.sb >>= b;    a.sc >>= b;    a.sd >>= b;    a.se >>= b;    a.sf >>= b;    }
-inline __device__ void operator >>= (u64x &a, const u64x b) { a.s0 >>= b.s0; a.s1 >>= b.s1; a.s2 >>= b.s2; a.s3 >>= b.s3; a.s4 >>= b.s4; a.s5 >>= b.s5; a.s6 >>= b.s6; a.s7 >>= b.s7; a.s8 >>= b.s8; a.s9 >>= b.s9; a.sa >>= b.sa; a.sb >>= b.sb; a.sc >>= b.sc; a.sd >>= b.sd; a.se >>= b.se; a.sf >>= b.sf; }
-
-inline __device__ void operator <<= (u64x &a, const u64  b) { a.s0 <<= b;    a.s1 <<= b;    a.s2 <<= b;    a.s3 <<= b;    a.s4 <<= b;    a.s5 <<= b;    a.s6 <<= b;    a.s7 <<= b;    a.s8 <<= b;    a.s9 <<= b;    a.sa <<= b;    a.sb <<= b;    a.sc <<= b;    a.sd <<= b;    a.se <<= b;    a.sf <<= b;    }
-inline __device__ void operator <<= (u64x &a, const u64x b) { a.s0 <<= b.s0; a.s1 <<= b.s1; a.s2 <<= b.s2; a.s3 <<= b.s3; a.s4 <<= b.s4; a.s5 <<= b.s5; a.s6 <<= b.s6; a.s7 <<= b.s7; a.s8 <<= b.s8; a.s9 <<= b.s9; a.sa <<= b.sa; a.sb <<= b.sb; a.sc <<= b.sc; a.sd <<= b.sd; a.se <<= b.se; a.sf <<= b.sf; }
-
-inline __device__ u64x operator << (const u64x a, const u64  b) { return u64x ((a.s0 << b),    (a.s1 << b)   , (a.s2 << b),    (a.s3 << b)   , (a.s4 << b),    (a.s5 << b)   , (a.s6 << b),    (a.s7 << b),    (a.s8 << b),    (a.s9 << b)   , (a.sa << b),    (a.sb << b)   , (a.sc << b),    (a.sd << b)   , (a.se << b),    (a.sf << b)   );  }
-inline __device__ u64x operator << (const u64x a, const u64x b) { return u64x ((a.s0 << b.s0), (a.s1 << b.s1), (a.s2 << b.s2), (a.s3 << b.s3), (a.s4 << b.s4), (a.s5 << b.s5), (a.s6 << b.s6), (a.s7 << b.s7), (a.s8 << b.s8), (a.s9 << b.s9), (a.sa << b.sa), (a.sb << b.sb), (a.sc << b.sc), (a.sd << b.sd), (a.se << b.se), (a.sf << b.sf));  }
-
-inline __device__ u64x operator >> (const u64x a, const u64  b) { return u64x ((a.s0 >> b),    (a.s1 >> b)   , (a.s2 >> b),    (a.s3 >> b)   , (a.s4 >> b),    (a.s5 >> b)   , (a.s6 >> b),    (a.s7 >> b),    (a.s8 >> b),    (a.s9 >> b)   , (a.sa >> b),    (a.sb >> b)   , (a.sc >> b),    (a.sd >> b)   , (a.se >> b),    (a.sf >> b)   );  }
-inline __device__ u64x operator >> (const u64x a, const u64x b) { return u64x ((a.s0 >> b.s0), (a.s1 >> b.s1), (a.s2 >> b.s2), (a.s3 >> b.s3), (a.s4 >> b.s4), (a.s5 >> b.s5), (a.s6 >> b.s6), (a.s7 >> b.s7), (a.s8 >> b.s8), (a.s9 >> b.s9), (a.sa >> b.sa), (a.sb >> b.sb), (a.sc >> b.sc), (a.sd >> b.sd), (a.se >> b.se), (a.sf >> b.sf));  }
-
-inline __device__ u64x operator ^  (const u64x a, const u64  b) { return u64x ((a.s0 ^  b),    (a.s1 ^  b)   , (a.s2 ^  b),    (a.s3 ^  b)   , (a.s4 ^  b),    (a.s5 ^  b)   , (a.s6 ^  b),    (a.s7 ^  b),    (a.s8 ^  b),    (a.s9 ^  b)   , (a.sa ^  b),    (a.sb ^  b)   , (a.sc ^  b),    (a.sd ^  b)   , (a.se ^  b),    (a.sf ^  b)   );  }
-inline __device__ u64x operator ^  (const u64x a, const u64x b) { return u64x ((a.s0 ^  b.s0), (a.s1 ^  b.s1), (a.s2 ^  b.s2), (a.s3 ^  b.s3), (a.s4 ^  b.s4), (a.s5 ^  b.s5), (a.s6 ^  b.s6), (a.s7 ^  b.s7), (a.s8 ^  b.s8), (a.s9 ^  b.s9), (a.sa ^  b.sa), (a.sb ^  b.sb), (a.sc ^  b.sc), (a.sd ^  b.sd), (a.se ^  b.se), (a.sf ^  b.sf));  }
-
-inline __device__ u64x operator |  (const u64x a, const u64  b) { return u64x ((a.s0 |  b),    (a.s1 |  b)   , (a.s2 |  b),    (a.s3 |  b)   , (a.s4 |  b),    (a.s5 |  b)   , (a.s6 |  b),    (a.s7 |  b),    (a.s8 |  b),    (a.s9 |  b)   , (a.sa |  b),    (a.sb |  b)   , (a.sc |  b),    (a.sd |  b)   , (a.se |  b),    (a.sf |  b)   );  }
-inline __device__ u64x operator |  (const u64x a, const u64x b) { return u64x ((a.s0 |  b.s0), (a.s1 |  b.s1), (a.s2 |  b.s2), (a.s3 |  b.s3), (a.s4 |  b.s4), (a.s5 |  b.s5), (a.s6 |  b.s6), (a.s7 |  b.s7), (a.s8 |  b.s8), (a.s9 |  b.s9), (a.sa |  b.sa), (a.sb |  b.sb), (a.sc |  b.sc), (a.sd |  b.sd), (a.se |  b.se), (a.sf |  b.sf));  }
-
-inline __device__ u64x operator &  (const u64x a, const u64  b) { return u64x ((a.s0 &  b),    (a.s1 &  b)   , (a.s2 &  b),    (a.s3 &  b)   , (a.s4 &  b),    (a.s5 &  b)   , (a.s6 &  b),    (a.s7 &  b),    (a.s8 &  b),    (a.s9 &  b)   , (a.sa &  b),    (a.sb &  b)   , (a.sc &  b),    (a.sd &  b)   , (a.se &  b),    (a.sf &  b)   );  }
-inline __device__ u64x operator &  (const u64x a, const u64x b) { return u64x ((a.s0 &  b.s0), (a.s1 &  b.s1), (a.s2 &  b.s2), (a.s3 &  b.s3), (a.s4 &  b.s4), (a.s5 &  b.s5), (a.s6 &  b.s6), (a.s7 &  b.s7), (a.s8 &  b.s8), (a.s9 &  b.s9), (a.sa &  b.sa), (a.sb &  b.sb), (a.sc &  b.sc), (a.sd &  b.sd), (a.se &  b.se), (a.sf &  b.sf));  }
-
-inline __device__ u64x operator +  (const u64x a, const u64  b) { return u64x ((a.s0 +  b),    (a.s1 +  b)   , (a.s2 +  b),    (a.s3 +  b)   , (a.s4 +  b),    (a.s5 +  b)   , (a.s6 +  b),    (a.s7 +  b),    (a.s8 +  b),    (a.s9 +  b)   , (a.sa +  b),    (a.sb +  b)   , (a.sc +  b),    (a.sd +  b)   , (a.se +  b),    (a.sf +  b)   );  }
-inline __device__ u64x operator +  (const u64x a, const u64x b) { return u64x ((a.s0 +  b.s0), (a.s1 +  b.s1), (a.s2 +  b.s2), (a.s3 +  b.s3), (a.s4 +  b.s4), (a.s5 +  b.s5), (a.s6 +  b.s6), (a.s7 +  b.s7), (a.s8 +  b.s8), (a.s9 +  b.s9), (a.sa +  b.sa), (a.sb +  b.sb), (a.sc +  b.sc), (a.sd +  b.sd), (a.se +  b.se), (a.sf +  b.sf));  }
-
-inline __device__ u64x operator -  (const u64x a, const u64  b) { return u64x ((a.s0 -  b),    (a.s1 -  b)   , (a.s2 -  b),    (a.s3 -  b)   , (a.s4 -  b),    (a.s5 -  b)   , (a.s6 -  b),    (a.s7 -  b),    (a.s8 -  b),    (a.s9 -  b)   , (a.sa -  b),    (a.sb -  b)   , (a.sc -  b),    (a.sd -  b)   , (a.se -  b),    (a.sf -  b)   );  }
-inline __device__ u64x operator -  (const u64x a, const u64x b) { return u64x ((a.s0 -  b.s0), (a.s1 -  b.s1), (a.s2 -  b.s2), (a.s3 -  b.s3), (a.s4 -  b.s4), (a.s5 -  b.s5), (a.s6 -  b.s6), (a.s7 -  b.s7), (a.s8 -  b.s8), (a.s9 -  b.s9), (a.sa -  b.sa), (a.sb -  b.sb), (a.sc -  b.sc), (a.sd -  b.sd), (a.se -  b.se), (a.sf -  b.sf));  }
-
-inline __device__ u64x operator *  (const u64x a, const u64  b) { return u64x ((a.s0 *  b),    (a.s1 *  b)   , (a.s2 *  b),    (a.s3 *  b)   , (a.s4 *  b),    (a.s5 *  b)   , (a.s6 *  b),    (a.s7 *  b),    (a.s8 *  b),    (a.s9 *  b)   , (a.sa *  b),    (a.sb *  b)   , (a.sc *  b),    (a.sd *  b)   , (a.se *  b),    (a.sf *  b)   );  }
-inline __device__ u64x operator *  (const u64x a, const u64x b) { return u64x ((a.s0 *  b.s0), (a.s1 *  b.s1), (a.s2 *  b.s2), (a.s3 *  b.s3), (a.s4 *  b.s4), (a.s5 *  b.s5), (a.s6 *  b.s6), (a.s7 *  b.s7), (a.s8 *  b.s8), (a.s9 *  b.s9), (a.sa *  b.sa), (a.sb *  b.sb), (a.sc *  b.sc), (a.sd *  b.sd), (a.se *  b.se), (a.sf *  b.sf));  }
-
-inline __device__ u64x operator %  (const u64x a, const u64  b) { return u64x ((a.s0 %  b),    (a.s1 %  b)   , (a.s2 %  b),    (a.s3 %  b)   , (a.s4 %  b),    (a.s5 %  b)   , (a.s6 %  b),    (a.s7 %  b),    (a.s8 %  b),    (a.s9 %  b)   , (a.sa %  b),    (a.sb %  b)   , (a.sc %  b),    (a.sd %  b)   , (a.se %  b),    (a.sf %  b)   );  }
-inline __device__ u64x operator %  (const u64x a, const u64x b) { return u64x ((a.s0 %  b.s0), (a.s1 %  b.s1), (a.s2 %  b.s2), (a.s3 %  b.s3), (a.s4 %  b.s4), (a.s5 %  b.s5), (a.s6 %  b.s6), (a.s7 %  b.s7), (a.s8 %  b.s8), (a.s9 %  b.s9), (a.sa %  b.sa), (a.sb %  b.sb), (a.sc %  b.sc), (a.sd %  b.sd), (a.se %  b.se), (a.sf %  b.sf));  }
-
-inline __device__ u64x operator ~  (const u64x a) { return u64x (~a.s0, ~a.s1, ~a.s2, ~a.s3, ~a.s4, ~a.s5, ~a.s6, ~a.s7, ~a.s8, ~a.s9, ~a.sa, ~a.sb, ~a.sc, ~a.sd, ~a.se, ~a.sf); }
-
-#endif
-
-typedef struct u8x  u8x;
-typedef struct u16x u16x;
-typedef struct u32x u32x;
-typedef struct u64x u64x;
-
-#define make_u8x  u8x
-#define make_u16x u16x
-#define make_u32x u32x
-#define make_u64x u64x
-
 #else
-typedef VTYPE(uchar,  VECT_SIZE)  u8x;
+typedef VTYPE(uchar,  VECT_SIZE) u8x;
 typedef VTYPE(ushort, VECT_SIZE) u16x;
 typedef VTYPE(uint,   VECT_SIZE) u32x;
-typedef VTYPE(ulong,  VECT_SIZE) u64x;
+typedef VTYPE(ullong, VECT_SIZE) u64x;
 
 #define make_u8x  (u8x)
 #define make_u16x (u16x)
diff --git a/OpenCL/inc_vendor.h b/OpenCL/inc_vendor.h
index f4c31f59a..43f5d8fb8 100644
--- a/OpenCL/inc_vendor.h
+++ b/OpenCL/inc_vendor.h
@@ -16,10 +16,6 @@
 #define IS_OPENCL
 #endif
 
-#ifdef IS_HIP
-#include <hip/hip_runtime.h>
-#endif
-
 #if defined IS_NATIVE
 #define CONSTANT_VK
 #define CONSTANT_AS
diff --git a/OpenCL/inc_zip_inflate.cl b/OpenCL/inc_zip_inflate.cl
index d05f6a792..00f762d81 100644
--- a/OpenCL/inc_zip_inflate.cl
+++ b/OpenCL/inc_zip_inflate.cl
@@ -73,18 +73,16 @@ enum{
     MZ_VERSION_ERROR = -6,
     MZ_PARAM_ERROR = -10000
 };
-typedef unsigned long mz_ulong;
+typedef ullong mz_ulong;
 
 #ifndef MINIZ_NO_ZLIB_COMPATIBLE_NAMES
 typedef unsigned char Byte;
 typedef unsigned int uInt;
-typedef mz_ulong uLong;
 typedef Byte Bytef;
 typedef uInt uIntf;
 typedef char charf;
 typedef int intf;
 typedef void *voidpf;
-typedef uLong uLongf;
 typedef void *voidp;
 typedef void *const voidpc;
 #define Z_NULL 0
@@ -204,10 +202,6 @@ DECLSPEC void *memset(u8 *s, int c, u32 len){
 #define MZ_MIN(a, b) (((a) < (b)) ? (a) : (b))
 #define MZ_DEFAULT_WINDOW_BITS 15
 #define TINFL_LZ_DICT_SIZE 32768
-#define TINFL_MEMCPY(d, s, l) memcpy(d, s, l)
-#define TINFL_MEMCPY_G(d, s, l, p) memcpy_g(d, s, l, p)
-#define TINFL_MEMSET(p, c, l) memset(p, c, (u32)l)
-#define MZ_CLEAR_OBJ(obj) memset(&(obj), 0, sizeof(obj))
 
 // hashcat-patched/hashcat-specific:
 #ifdef CRC32_IN_INFLATE
@@ -583,7 +577,7 @@ DECLSPEC tinfl_status tinfl_decompress(tinfl_decompressor *r, MAYBE_GLOBAL const
                     TINFL_CR_RETURN(38, (decomp_flags & TINFL_FLAG_HAS_MORE_INPUT) ? TINFL_STATUS_NEEDS_MORE_INPUT : TINFL_STATUS_FAILED_CANNOT_MAKE_PROGRESS);
                 }
                 n = MZ_MIN(MZ_MIN((size_t)(pOut_buf_end - pOut_buf_cur), (size_t)(pIn_buf_end - pIn_buf_cur)), counter);
-                TINFL_MEMCPY_G(pOut_buf_cur, pIn_buf_cur, n, pStream);
+                memcpy_g(pOut_buf_cur, pIn_buf_cur, n, pStream);
                 pIn_buf_cur += n;
                 pOut_buf_cur += n;
                 counter -= (mz_uint)n;
@@ -601,7 +595,7 @@ DECLSPEC tinfl_status tinfl_decompress(tinfl_decompressor *r, MAYBE_GLOBAL const
                 mz_uint i;
                 r->m_table_sizes[0] = 288;
                 r->m_table_sizes[1] = 32;
-                TINFL_MEMSET(r->m_tables[1].m_code_size, 5, 32);
+                memset(r->m_tables[1].m_code_size, 5, 32);
                 for (i = 0; i <= 143; ++i)
                     *p++ = 8;
                 for (; i <= 255; ++i)
@@ -618,7 +612,8 @@ DECLSPEC tinfl_status tinfl_decompress(tinfl_decompressor *r, MAYBE_GLOBAL const
                     TINFL_GET_BITS(11, r->m_table_sizes[counter], "\05\05\04"[counter]);
                     r->m_table_sizes[counter] += s_min_table_sizes[counter];
                 }
-                MZ_CLEAR_OBJ(r->m_tables[2].m_code_size);
+                memset(r->m_tables[2].m_code_size, 0, TINFL_MAX_HUFF_SYMBOLS_0);
+
                 for (counter = 0; counter < r->m_table_sizes[2]; counter++)
                 {
                     mz_uint s;
@@ -633,9 +628,11 @@ DECLSPEC tinfl_status tinfl_decompress(tinfl_decompressor *r, MAYBE_GLOBAL const
                 tinfl_huff_table *pTable;
                 mz_uint i, j, used_syms, total, sym_index, next_code[17], total_syms[16];
                 pTable = &r->m_tables[r->m_type];
-                MZ_CLEAR_OBJ(total_syms);
-                MZ_CLEAR_OBJ(pTable->m_look_up);
-                MZ_CLEAR_OBJ(pTable->m_tree);
+
+                memset((u8 *) total_syms, 0, 64);
+                memset((u8 *) pTable->m_look_up, 0, TINFL_FAST_LOOKUP_SIZE * 2);
+                memset((u8 *) pTable->m_tree, 0, TINFL_MAX_HUFF_SYMBOLS_0 * 2 * 2);
+
                 for (i = 0; i < r->m_table_sizes[r->m_type]; ++i)
                     total_syms[pTable->m_code_size[i]]++;
                 used_syms = 0, total = 0;
@@ -707,15 +704,18 @@ DECLSPEC tinfl_status tinfl_decompress(tinfl_decompressor *r, MAYBE_GLOBAL const
                         num_extra = "\02\03\07"[dist - 16];
                         TINFL_GET_BITS(18, s, num_extra);
                         s += "\03\03\013"[dist - 16];
-                        TINFL_MEMSET(r->m_len_codes + counter, (dist == 16) ? r->m_len_codes[counter - 1] : 0, s);
+
+                        memset(r->m_len_codes + counter, (dist == 16) ? r->m_len_codes[counter - 1] : 0, s);
+
+
                         counter += s;
                     }
                     if ((r->m_table_sizes[0] + r->m_table_sizes[1]) != counter)
                     {
                         TINFL_CR_RETURN_FOREVER(21, TINFL_STATUS_FAILED);
                     }
-                    TINFL_MEMCPY(r->m_tables[0].m_code_size, r->m_len_codes, r->m_table_sizes[0]);
-                    TINFL_MEMCPY(r->m_tables[1].m_code_size, r->m_len_codes + r->m_table_sizes[0], r->m_table_sizes[1]);
+                    memcpy(r->m_tables[0].m_code_size, r->m_len_codes, r->m_table_sizes[0]);
+                    memcpy(r->m_tables[1].m_code_size, r->m_len_codes + r->m_table_sizes[0], r->m_table_sizes[1]);
                 }
             }
             for (;;)
diff --git a/OpenCL/m00500-optimized.cl b/OpenCL/m00500-optimized.cl
index 19f7153ff..38a361b96 100644
--- a/OpenCL/m00500-optimized.cl
+++ b/OpenCL/m00500-optimized.cl
@@ -32,7 +32,7 @@ DECLSPEC void memcat16 (u32 *block0, u32 *block1, u32 *block2, u32 *block3, cons
   u32 tmp3;
   u32 tmp4;
 
-  #if (defined IS_AMD || defined IS_HIP) || defined IS_GENERIC
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 0) || defined IS_GENERIC
   u32 in0 = append[0];
   u32 in1 = append[1];
   u32 in2 = append[2];
@@ -45,12 +45,18 @@ DECLSPEC void memcat16 (u32 *block0, u32 *block1, u32 *block2, u32 *block3, cons
   tmp4 = hc_bytealign (in3,   0, offset);
   #endif
 
-  #ifdef IS_NV
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1) || defined IS_NV
   const int offset_mod_4 = offset & 3;
 
   const int offset_minus_4 = 4 - offset_mod_4;
 
+  #if defined IS_NV
   const int selector = (0x76543210 >> (offset_minus_4 * 4)) & 0xffff;
+  #endif
+
+  #if (defined IS_AMD || defined IS_HIP)
+  const int selector = l32_from_64_S (0x0706050403020100UL >> (offset_minus_4 * 8));
+  #endif
 
   u32 in0 = append[0];
   u32 in1 = append[1];
@@ -139,7 +145,7 @@ DECLSPEC void memcat16_x80 (u32 *block0, u32 *block1, u32 *block2, u32 *block3,
   u32 tmp3;
   u32 tmp4;
 
-  #if (defined IS_AMD || defined IS_HIP) || defined IS_GENERIC
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 0) || defined IS_GENERIC
   u32 in0 = append[0];
   u32 in1 = append[1];
   u32 in2 = append[2];
@@ -153,12 +159,18 @@ DECLSPEC void memcat16_x80 (u32 *block0, u32 *block1, u32 *block2, u32 *block3,
   tmp4 = hc_bytealign (in3, in4, offset);
   #endif
 
-  #ifdef IS_NV
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1) || defined IS_NV
   const int offset_mod_4 = offset & 3;
 
   const int offset_minus_4 = 4 - offset_mod_4;
 
+  #if defined IS_NV
   const int selector = (0x76543210 >> (offset_minus_4 * 4)) & 0xffff;
+  #endif
+
+  #if (defined IS_AMD || defined IS_HIP)
+  const int selector = l32_from_64_S (0x0706050403020100UL >> (offset_minus_4 * 8));
+  #endif
 
   u32 in0 = append[0];
   u32 in1 = append[1];
@@ -246,7 +258,7 @@ DECLSPEC void memcat8 (u32 *block0, u32 *block1, u32 *block2, u32 *block3, const
   u32 tmp1;
   u32 tmp2;
 
-  #if (defined IS_AMD || defined IS_HIP) || defined IS_GENERIC
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 0) || defined IS_GENERIC
   u32 in0 = append[0];
   u32 in1 = append[1];
 
@@ -255,12 +267,18 @@ DECLSPEC void memcat8 (u32 *block0, u32 *block1, u32 *block2, u32 *block3, const
   tmp2 = hc_bytealign (in1,   0, offset);
   #endif
 
-  #ifdef IS_NV
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1) || defined IS_NV
   const int offset_mod_4 = offset & 3;
 
   const int offset_minus_4 = 4 - offset_mod_4;
 
+  #if defined IS_NV
   const int selector = (0x76543210 >> (offset_minus_4 * 4)) & 0xffff;
+  #endif
+
+  #if (defined IS_AMD || defined IS_HIP)
+  const int selector = l32_from_64_S (0x0706050403020100UL >> (offset_minus_4 * 8));
+  #endif
 
   u32 in0 = append[0];
   u32 in1 = append[1];
diff --git a/OpenCL/m01500_a3-pure.cl b/OpenCL/m01500_a3-pure.cl
index c2c4245e1..7a5adf017 100644
--- a/OpenCL/m01500_a3-pure.cl
+++ b/OpenCL/m01500_a3-pure.cl
@@ -1664,18 +1664,18 @@ DECLSPEC void DESCrypt (const u32 SALT, const u32 K00, const u32 K01, const u32
 
 DECLSPEC void DESCrypt (const u32 SALT, const u32 K00, const u32 K01, const u32 K02, const u32 K03, const u32 K04, const u32 K05, const u32 K06, const u32 K07, const u32 K08, const u32 K09, const u32 K10, const u32 K11, const u32 K12, const u32 K13, const u32 K14, const u32 K15, const u32 K16, const u32 K17, const u32 K18, const u32 K19, const u32 K20, const u32 K21, const u32 K22, const u32 K23, const u32 K24, const u32 K25, const u32 K26, const u32 K27, const u32 K28, const u32 K29, const u32 K30, const u32 K31, const u32 K32, const u32 K33, const u32 K34, const u32 K35, const u32 K36, const u32 K37, const u32 K38, const u32 K39, const u32 K40, const u32 K41, const u32 K42, const u32 K43, const u32 K44, const u32 K45, const u32 K46, const u32 K47, const u32 K48, const u32 K49, const u32 K50, const u32 K51, const u32 K52, const u32 K53, const u32 K54, const u32 K55, u32 *D00, u32 *D01, u32 *D02, u32 *D03, u32 *D04, u32 *D05, u32 *D06, u32 *D07, u32 *D08, u32 *D09, u32 *D10, u32 *D11, u32 *D12, u32 *D13, u32 *D14, u32 *D15, u32 *D16, u32 *D17, u32 *D18, u32 *D19, u32 *D20, u32 *D21, u32 *D22, u32 *D23, u32 *D24, u32 *D25, u32 *D26, u32 *D27, u32 *D28, u32 *D29, u32 *D30, u32 *D31, u32 *D32, u32 *D33, u32 *D34, u32 *D35, u32 *D36, u32 *D37, u32 *D38, u32 *D39, u32 *D40, u32 *D41, u32 *D42, u32 *D43, u32 *D44, u32 *D45, u32 *D46, u32 *D47, u32 *D48, u32 *D49, u32 *D50, u32 *D51, u32 *D52, u32 *D53, u32 *D54, u32 *D55, u32 *D56, u32 *D57, u32 *D58, u32 *D59, u32 *D60, u32 *D61, u32 *D62, u32 *D63)
 {
-  const u32 s001 = (0x001 & SALT) ? 0xffffffff : 0;
-  const u32 s002 = (0x002 & SALT) ? 0xffffffff : 0;
-  const u32 s004 = (0x004 & SALT) ? 0xffffffff : 0;
-  const u32 s008 = (0x008 & SALT) ? 0xffffffff : 0;
-  const u32 s010 = (0x010 & SALT) ? 0xffffffff : 0;
-  const u32 s020 = (0x020 & SALT) ? 0xffffffff : 0;
-  const u32 s040 = (0x040 & SALT) ? 0xffffffff : 0;
-  const u32 s080 = (0x080 & SALT) ? 0xffffffff : 0;
-  const u32 s100 = (0x100 & SALT) ? 0xffffffff : 0;
-  const u32 s200 = (0x200 & SALT) ? 0xffffffff : 0;
-  const u32 s400 = (0x400 & SALT) ? 0xffffffff : 0;
-  const u32 s800 = (0x800 & SALT) ? 0xffffffff : 0;
+  const u32 s001 = (0x001 & SALT) ? 1 : 0;
+  const u32 s002 = (0x002 & SALT) ? 1 : 0;
+  const u32 s004 = (0x004 & SALT) ? 1 : 0;
+  const u32 s008 = (0x008 & SALT) ? 1 : 0;
+  const u32 s010 = (0x010 & SALT) ? 1 : 0;
+  const u32 s020 = (0x020 & SALT) ? 1 : 0;
+  const u32 s040 = (0x040 & SALT) ? 1 : 0;
+  const u32 s080 = (0x080 & SALT) ? 1 : 0;
+  const u32 s100 = (0x100 & SALT) ? 1 : 0;
+  const u32 s200 = (0x200 & SALT) ? 1 : 0;
+  const u32 s400 = (0x400 & SALT) ? 1 : 0;
+  const u32 s800 = (0x800 & SALT) ? 1 : 0;
 
   KXX_DECL u32 k00, k01, k02, k03, k04, k05;
   KXX_DECL u32 k06, k07, k08, k09, k10, k11;
diff --git a/OpenCL/m01600-optimized.cl b/OpenCL/m01600-optimized.cl
index cfaad44cc..62194e973 100644
--- a/OpenCL/m01600-optimized.cl
+++ b/OpenCL/m01600-optimized.cl
@@ -31,7 +31,7 @@ DECLSPEC void memcat16 (u32 *block0, u32 *block1, u32 *block2, u32 *block3, cons
   u32 tmp3;
   u32 tmp4;
 
-  #if (defined IS_AMD || defined IS_HIP) || defined IS_GENERIC
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 0) || defined IS_GENERIC
   u32 in0 = append[0];
   u32 in1 = append[1];
   u32 in2 = append[2];
@@ -44,12 +44,18 @@ DECLSPEC void memcat16 (u32 *block0, u32 *block1, u32 *block2, u32 *block3, cons
   tmp4 = hc_bytealign (in3,   0, offset);
   #endif
 
-  #ifdef IS_NV
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1) || defined IS_NV
   const int offset_mod_4 = offset & 3;
 
   const int offset_minus_4 = 4 - offset_mod_4;
 
+  #if defined IS_NV
   const int selector = (0x76543210 >> (offset_minus_4 * 4)) & 0xffff;
+  #endif
+
+  #if (defined IS_AMD || defined IS_HIP)
+  const int selector = l32_from_64_S (0x0706050403020100UL >> (offset_minus_4 * 8));
+  #endif
 
   u32 in0 = append[0];
   u32 in1 = append[1];
@@ -138,7 +144,7 @@ DECLSPEC void memcat16_x80 (u32 *block0, u32 *block1, u32 *block2, u32 *block3,
   u32 tmp3;
   u32 tmp4;
 
-  #if (defined IS_AMD || defined IS_HIP) || defined IS_GENERIC
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 0) || defined IS_GENERIC
   u32 in0 = append[0];
   u32 in1 = append[1];
   u32 in2 = append[2];
@@ -152,12 +158,18 @@ DECLSPEC void memcat16_x80 (u32 *block0, u32 *block1, u32 *block2, u32 *block3,
   tmp4 = hc_bytealign (in3, in4, offset);
   #endif
 
-  #ifdef IS_NV
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1) || defined IS_NV
   const int offset_mod_4 = offset & 3;
 
   const int offset_minus_4 = 4 - offset_mod_4;
 
+  #if defined IS_NV
   const int selector = (0x76543210 >> (offset_minus_4 * 4)) & 0xffff;
+  #endif
+
+  #if (defined IS_AMD || defined IS_HIP)
+  const int selector = l32_from_64_S (0x0706050403020100UL >> (offset_minus_4 * 8));
+  #endif
 
   u32 in0 = append[0];
   u32 in1 = append[1];
@@ -245,7 +257,7 @@ DECLSPEC void memcat8 (u32 *block0, u32 *block1, u32 *block2, u32 *block3, const
   u32 tmp1;
   u32 tmp2;
 
-  #if (defined IS_AMD || defined IS_HIP) || defined IS_GENERIC
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 0) || defined IS_GENERIC
   u32 in0 = append[0];
   u32 in1 = append[1];
 
@@ -254,12 +266,18 @@ DECLSPEC void memcat8 (u32 *block0, u32 *block1, u32 *block2, u32 *block3, const
   tmp2 = hc_bytealign (in1,   0, offset);
   #endif
 
-  #ifdef IS_NV
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1) || defined IS_NV
   const int offset_mod_4 = offset & 3;
 
   const int offset_minus_4 = 4 - offset_mod_4;
 
+  #if defined IS_NV
   const int selector = (0x76543210 >> (offset_minus_4 * 4)) & 0xffff;
+  #endif
+
+  #if (defined IS_AMD || defined IS_HIP)
+  const int selector = l32_from_64_S (0x0706050403020100UL >> (offset_minus_4 * 8));
+  #endif
 
   u32 in0 = append[0];
   u32 in1 = append[1];
diff --git a/OpenCL/m02500-pure.cl b/OpenCL/m02500-pure.cl
index 5bcd35d63..f6e20fa5b 100644
--- a/OpenCL/m02500-pure.cl
+++ b/OpenCL/m02500-pure.cl
@@ -775,11 +775,7 @@ KERNEL_FQ void m02500_aux3 (KERN_ATTR_TMPS_ESALT (wpa_pbkdf2_tmp_t, wpa_eapol_t)
     s_te4[i] = te4[i];
   }
 
-  #if defined IS_CUDA || defined IS_HIP
-  __syncthreads();
-  #else
   SYNC_THREADS ();
-  #endif
 
   #else
 
diff --git a/OpenCL/m05800-optimized.cl b/OpenCL/m05800-optimized.cl
index 38099159f..9f8f5a3cc 100644
--- a/OpenCL/m05800-optimized.cl
+++ b/OpenCL/m05800-optimized.cl
@@ -2119,7 +2119,7 @@ DECLSPEC void append_salt (u32 *w0, u32 *w1, u32 *w2, const u32 *append, const u
   u32 tmp4;
   u32 tmp5;
 
-  #if (defined IS_AMD || defined IS_HIP) || defined IS_GENERIC
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 0) || defined IS_GENERIC
   u32 in0 = append[0];
   u32 in1 = append[1];
   u32 in2 = append[2];
@@ -2134,12 +2134,18 @@ DECLSPEC void append_salt (u32 *w0, u32 *w1, u32 *w2, const u32 *append, const u
   tmp5 = hc_bytealign (in4,   0, offset);
   #endif
 
-  #ifdef IS_NV
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1) || defined IS_NV
   const int offset_mod_4 = offset & 3;
 
   const int offset_minus_4 = 4 - offset_mod_4;
 
+  #if defined IS_NV
   const int selector = (0x76543210 >> (offset_minus_4 * 4)) & 0xffff;
+  #endif
+
+  #if (defined IS_AMD || defined IS_HIP)
+  const int selector = l32_from_64_S (0x0706050403020100UL >> (offset_minus_4 * 8));
+  #endif
 
   u32 in0 = append[0];
   u32 in1 = append[1];
diff --git a/OpenCL/m06300-optimized.cl b/OpenCL/m06300-optimized.cl
index b7c9ddddd..f242259da 100644
--- a/OpenCL/m06300-optimized.cl
+++ b/OpenCL/m06300-optimized.cl
@@ -28,7 +28,7 @@ DECLSPEC void memcat16 (u32 *block0, u32 *block1, u32 *block2, u32 *block3, cons
   u32 tmp3;
   u32 tmp4;
 
-  #if (defined IS_AMD || defined IS_HIP) || defined IS_GENERIC
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 0) || defined IS_GENERIC
   u32 in0 = append[0];
   u32 in1 = append[1];
   u32 in2 = append[2];
@@ -41,12 +41,18 @@ DECLSPEC void memcat16 (u32 *block0, u32 *block1, u32 *block2, u32 *block3, cons
   tmp4 = hc_bytealign (in3,   0, offset);
   #endif
 
-  #ifdef IS_NV
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1) || defined IS_NV
   const int offset_mod_4 = offset & 3;
 
   const int offset_minus_4 = 4 - offset_mod_4;
 
+  #if defined IS_NV
   const int selector = (0x76543210 >> (offset_minus_4 * 4)) & 0xffff;
+  #endif
+
+  #if (defined IS_AMD || defined IS_HIP)
+  const int selector = l32_from_64_S (0x0706050403020100UL >> (offset_minus_4 * 8));
+  #endif
 
   u32 in0 = append[0];
   u32 in1 = append[1];
@@ -135,7 +141,7 @@ DECLSPEC void memcat16_x80 (u32 *block0, u32 *block1, u32 *block2, u32 *block3,
   u32 tmp3;
   u32 tmp4;
 
-  #if (defined IS_AMD || defined IS_HIP) || defined IS_GENERIC
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 0) || defined IS_GENERIC
   u32 in0 = append[0];
   u32 in1 = append[1];
   u32 in2 = append[2];
@@ -149,12 +155,18 @@ DECLSPEC void memcat16_x80 (u32 *block0, u32 *block1, u32 *block2, u32 *block3,
   tmp4 = hc_bytealign (in3, in4, offset);
   #endif
 
-  #ifdef IS_NV
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1) || defined IS_NV
   const int offset_mod_4 = offset & 3;
 
   const int offset_minus_4 = 4 - offset_mod_4;
 
+  #if defined IS_NV
   const int selector = (0x76543210 >> (offset_minus_4 * 4)) & 0xffff;
+  #endif
+
+  #if (defined IS_AMD || defined IS_HIP)
+  const int selector = l32_from_64_S (0x0706050403020100UL >> (offset_minus_4 * 8));
+  #endif
 
   u32 in0 = append[0];
   u32 in1 = append[1];
@@ -242,7 +254,7 @@ DECLSPEC void memcat8 (u32 *block0, u32 *block1, u32 *block2, u32 *block3, const
   u32 tmp1;
   u32 tmp2;
 
-  #if (defined IS_AMD || defined IS_HIP) || defined IS_GENERIC
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 0) || defined IS_GENERIC
   u32 in0 = append[0];
   u32 in1 = append[1];
 
@@ -251,12 +263,18 @@ DECLSPEC void memcat8 (u32 *block0, u32 *block1, u32 *block2, u32 *block3, const
   tmp2 = hc_bytealign (in1,   0, offset);
   #endif
 
-  #ifdef IS_NV
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1) || defined IS_NV
   const int offset_mod_4 = offset & 3;
 
   const int offset_minus_4 = 4 - offset_mod_4;
 
+  #if defined IS_NV
   const int selector = (0x76543210 >> (offset_minus_4 * 4)) & 0xffff;
+  #endif
+
+  #if (defined IS_AMD || defined IS_HIP)
+  const int selector = l32_from_64_S (0x0706050403020100UL >> (offset_minus_4 * 8));
+  #endif
 
   u32 in0 = append[0];
   u32 in1 = append[1];
diff --git a/OpenCL/m07400-optimized.cl b/OpenCL/m07400-optimized.cl
index 7efa5c94e..5fb83a2ad 100644
--- a/OpenCL/m07400-optimized.cl
+++ b/OpenCL/m07400-optimized.cl
@@ -45,7 +45,7 @@ DECLSPEC u32 memcat16 (u32 *block, const u32 offset, const u32 *append, const u3
   u32 in2 = append[2];
   u32 in3 = append[3];
 
-  #if (defined IS_AMD || defined IS_HIP) || defined IS_GENERIC
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 0) || defined IS_GENERIC
   const u32 tmp0 = hc_bytealign_be (  0, in0, offset);
   const u32 tmp1 = hc_bytealign_be (in0, in1, offset);
   const u32 tmp2 = hc_bytealign_be (in1, in2, offset);
@@ -53,8 +53,15 @@ DECLSPEC u32 memcat16 (u32 *block, const u32 offset, const u32 *append, const u3
   const u32 tmp4 = hc_bytealign_be (in3,   0, offset);
   #endif
 
-  #ifdef IS_NV
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1) || defined IS_NV
+
+  #if defined IS_NV
   const int selector = (0x76543210 >> ((offset & 3) * 4)) & 0xffff;
+  #endif
+
+  #if (defined IS_AMD || defined IS_HIP)
+  const int selector = l32_from_64_S (0x0706050403020100UL >> ((offset & 3) * 8));
+  #endif
 
   const u32 tmp0 = hc_byte_perm_S (in0,   0, selector);
   const u32 tmp1 = hc_byte_perm_S (in1, in0, selector);
@@ -165,7 +172,7 @@ DECLSPEC u32 memcat16c (u32 *block, const u32 offset, const u32 *append, const u
   u32 in2 = append[2];
   u32 in3 = append[3];
 
-  #if (defined IS_AMD || defined IS_HIP) || defined IS_GENERIC
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 0) || defined IS_GENERIC
   const u32 tmp0 = hc_bytealign_be (  0, in0, offset);
   const u32 tmp1 = hc_bytealign_be (in0, in1, offset);
   const u32 tmp2 = hc_bytealign_be (in1, in2, offset);
@@ -173,8 +180,15 @@ DECLSPEC u32 memcat16c (u32 *block, const u32 offset, const u32 *append, const u
   const u32 tmp4 = hc_bytealign_be (in3,   0, offset);
   #endif
 
-  #ifdef IS_NV
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1) || defined IS_NV
+
+  #if defined IS_NV
   const int selector = (0x76543210 >> ((offset & 3) * 4)) & 0xffff;
+  #endif
+
+  #if (defined IS_AMD || defined IS_HIP)
+  const int selector = l32_from_64_S (0x0706050403020100UL >> ((offset & 3) * 8));
+  #endif
 
   const u32 tmp0 = hc_byte_perm_S (in0,   0, selector);
   const u32 tmp1 = hc_byte_perm_S (in1, in0, selector);
@@ -322,7 +336,7 @@ DECLSPEC u32 memcat16s (u32 *block, const u32 offset, const u32 *append, const u
   u32 in3 = append[3];
   u32 in4 = append[4];
 
-  #if (defined IS_AMD || defined IS_HIP) || defined IS_GENERIC
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 0) || defined IS_GENERIC
   const u32 tmp0 = hc_bytealign_be (  0, in0, offset);
   const u32 tmp1 = hc_bytealign_be (in0, in1, offset);
   const u32 tmp2 = hc_bytealign_be (in1, in2, offset);
@@ -331,8 +345,15 @@ DECLSPEC u32 memcat16s (u32 *block, const u32 offset, const u32 *append, const u
   const u32 tmp5 = hc_bytealign_be (in4,   0, offset);
   #endif
 
-  #ifdef IS_NV
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1) || defined IS_NV
+
+  #if defined IS_NV
   const int selector = (0x76543210 >> ((offset & 3) * 4)) & 0xffff;
+  #endif
+
+  #if (defined IS_AMD || defined IS_HIP)
+  const int selector = l32_from_64_S (0x0706050403020100UL >> ((offset & 3) * 8));
+  #endif
 
   const u32 tmp0 = hc_byte_perm_S (in0,   0, selector);
   const u32 tmp1 = hc_byte_perm_S (in1, in0, selector);
@@ -456,7 +477,7 @@ DECLSPEC u32 memcat16sc (u32 *block, const u32 offset, const u32 *append, const
   u32 in3 = append[3];
   u32 in4 = append[4];
 
-  #if (defined IS_AMD || defined IS_HIP) || defined IS_GENERIC
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 0) || defined IS_GENERIC
   const u32 tmp0 = hc_bytealign_be (  0, in0, offset);
   const u32 tmp1 = hc_bytealign_be (in0, in1, offset);
   const u32 tmp2 = hc_bytealign_be (in1, in2, offset);
@@ -465,8 +486,15 @@ DECLSPEC u32 memcat16sc (u32 *block, const u32 offset, const u32 *append, const
   const u32 tmp5 = hc_bytealign_be (in4,   0, offset);
   #endif
 
-  #ifdef IS_NV
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1) || defined IS_NV
+
+  #if defined IS_NV
   const int selector = (0x76543210 >> ((offset & 3) * 4)) & 0xffff;
+  #endif
+
+  #if (defined IS_AMD || defined IS_HIP)
+  const int selector = l32_from_64_S (0x0706050403020100UL >> ((offset & 3) * 8));
+  #endif
 
   const u32 tmp0 = hc_byte_perm_S (in0,   0, selector);
   const u32 tmp1 = hc_byte_perm_S (in1, in0, selector);
@@ -756,7 +784,7 @@ DECLSPEC u32 memcat20 (u32 *block, const u32 offset, const u32 *append, const u3
   u32 in2 = append[2];
   u32 in3 = append[3];
 
-  #if (defined IS_AMD || defined IS_HIP) || defined IS_GENERIC
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 0) || defined IS_GENERIC
   const u32 tmp0 = hc_bytealign_be_S (  0, in0, offset);
   const u32 tmp1 = hc_bytealign_be_S (in0, in1, offset);
   const u32 tmp2 = hc_bytealign_be_S (in1, in2, offset);
@@ -764,8 +792,15 @@ DECLSPEC u32 memcat20 (u32 *block, const u32 offset, const u32 *append, const u3
   const u32 tmp4 = hc_bytealign_be_S (in3,   0, offset);
   #endif
 
-  #ifdef IS_NV
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1) || defined IS_NV
+
+  #if defined IS_NV
   const int selector = (0x76543210 >> ((offset & 3) * 4)) & 0xffff;
+  #endif
+
+  #if (defined IS_AMD || defined IS_HIP)
+  const int selector = l32_from_64_S (0x0706050403020100UL >> ((offset & 3) * 8));
+  #endif
 
   const u32 tmp0 = hc_byte_perm_S (in0,   0, selector);
   const u32 tmp1 = hc_byte_perm_S (in1, in0, selector);
@@ -915,7 +950,7 @@ DECLSPEC u32 memcat20_x80 (u32 *block, const u32 offset, const u32 *append, cons
   u32 in3 = append[3];
   u32 in4 = 0x80000000;
 
-  #if (defined IS_AMD || defined IS_HIP) || defined IS_GENERIC
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 0) || defined IS_GENERIC
   const u32 tmp0 = hc_bytealign_be_S (  0, in0, offset);
   const u32 tmp1 = hc_bytealign_be_S (in0, in1, offset);
   const u32 tmp2 = hc_bytealign_be_S (in1, in2, offset);
@@ -923,8 +958,15 @@ DECLSPEC u32 memcat20_x80 (u32 *block, const u32 offset, const u32 *append, cons
   const u32 tmp4 = hc_bytealign_be_S (in3, in4, offset);
   #endif
 
-  #ifdef IS_NV
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1) || defined IS_NV
+
+  #if defined IS_NV
   const int selector = (0x76543210 >> ((offset & 3) * 4)) & 0xffff;
+  #endif
+
+  #if (defined IS_AMD || defined IS_HIP)
+  const int selector = l32_from_64_S (0x0706050403020100UL >> ((offset & 3) * 8));
+  #endif
 
   const u32 tmp0 = hc_byte_perm_S (in0,   0, selector);
   const u32 tmp1 = hc_byte_perm_S (in1, in0, selector);
@@ -1074,7 +1116,7 @@ DECLSPEC u32 memcat24 (u32 *block, const u32 offset, const u32 *append, const u3
   u32 in3 = append[3];
   u32 in4 = append[4];
 
-  #if (defined IS_AMD || defined IS_HIP) || defined IS_GENERIC
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 0) || defined IS_GENERIC
   const u32 tmp0 = hc_bytealign_be_S (  0, in0, offset);
   const u32 tmp1 = hc_bytealign_be_S (in0, in1, offset);
   const u32 tmp2 = hc_bytealign_be_S (in1, in2, offset);
@@ -1083,8 +1125,15 @@ DECLSPEC u32 memcat24 (u32 *block, const u32 offset, const u32 *append, const u3
   const u32 tmp5 = hc_bytealign_be_S (in4,   0, offset);
   #endif
 
-  #ifdef IS_NV
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1) || defined IS_NV
+
+  #if defined IS_NV
   const int selector = (0x76543210 >> ((offset & 3) * 4)) & 0xffff;
+  #endif
+
+  #if (defined IS_AMD || defined IS_HIP)
+  const int selector = l32_from_64_S (0x0706050403020100UL >> ((offset & 3) * 8));
+  #endif
 
   const u32 tmp0 = hc_byte_perm_S (in0,   0, selector);
   const u32 tmp1 = hc_byte_perm_S (in1, in0, selector);
diff --git a/OpenCL/m08900-pure.cl b/OpenCL/m08900-pure.cl
index 0f282b509..74f8a0e66 100644
--- a/OpenCL/m08900-pure.cl
+++ b/OpenCL/m08900-pure.cl
@@ -303,7 +303,7 @@ KERNEL_FQ void m08900_init (KERN_ATTR_TMPS (scrypt_tmp_t))
     digest[6] = sha256_hmac_ctx2.opad.h[6];
     digest[7] = sha256_hmac_ctx2.opad.h[7];
 
-    #if defined IS_CUDA || defined IS_HIP
+    #if defined IS_CUDA
     const uint4 tmp0 = make_uint4 (digest[0], digest[1], digest[2], digest[3]);
     const uint4 tmp1 = make_uint4 (digest[4], digest[5], digest[6], digest[7]);
     #else
@@ -331,7 +331,7 @@ KERNEL_FQ void m08900_init (KERN_ATTR_TMPS (scrypt_tmp_t))
 
     uint4 X[4];
 
-    #if defined IS_CUDA || defined IS_HIP
+    #if defined IS_CUDA
     X[0] = make_uint4 (T[0].x, T[1].y, T[2].z, T[3].w);
     X[1] = make_uint4 (T[1].x, T[2].y, T[3].z, T[0].w);
     X[2] = make_uint4 (T[2].x, T[3].y, T[0].z, T[1].w);
@@ -441,7 +441,7 @@ KERNEL_FQ void m08900_comp (KERN_ATTR_TMPS (scrypt_tmp_t))
 
     uint4 T[4];
 
-    #if defined IS_CUDA || defined IS_HIP
+    #if defined IS_CUDA
     T[0] = make_uint4 (X[0].x, X[3].y, X[2].z, X[1].w);
     T[1] = make_uint4 (X[1].x, X[0].y, X[3].z, X[2].w);
     T[2] = make_uint4 (X[2].x, X[1].y, X[0].z, X[3].w);
diff --git a/OpenCL/m09000-pure.cl b/OpenCL/m09000-pure.cl
index 737adde4e..323cf8387 100644
--- a/OpenCL/m09000-pure.cl
+++ b/OpenCL/m09000-pure.cl
@@ -310,6 +310,51 @@ CONSTANT_VK u32a c_pbox[18] =
   0x9216d5d9, 0x8979fb1b
 };
 
+// Yes, works only with CUDA atm
+
+#ifdef DYNAMIC_LOCAL
+#define BCRYPT_AVOID_BANK_CONFLICTS
+#endif
+
+#ifdef BCRYPT_AVOID_BANK_CONFLICTS
+
+// access pattern: minimize bank ID based on thread ID but thread ID is not saved from computation
+
+#define KEY32(lid,key) (((key) * FIXED_LOCAL_SIZE) + (lid))
+
+DECLSPEC u32 GET_KEY32 (LOCAL_AS u32 *S, const u64 key)
+{
+  const u64 lid = get_local_id (0);
+
+  return S[KEY32 (lid, key)];
+}
+
+DECLSPEC void SET_KEY32 (LOCAL_AS u32 *S, const u64 key, const u32 val)
+{
+  const u64 lid = get_local_id (0);
+
+  S[KEY32 (lid, key)] = val;
+}
+
+#undef KEY32
+
+#else
+
+// access pattern: linear access with S offset already set to right offset based on thread ID saving it from compuation
+//                 makes sense if there are not thread ID's (for instance on CPU)
+
+DECLSPEC inline u32 GET_KEY32 (LOCAL_AS u32 *S, const u64 key)
+{
+  return S[key];
+}
+
+DECLSPEC inline void SET_KEY32 (LOCAL_AS u32 *S, const u64 key, const u32 val)
+{
+  S[key] = val;
+}
+
+#endif
+
 #define BF_ROUND(L,R,N)                       \
 {                                             \
   u32 tmp;                                    \
@@ -319,10 +364,10 @@ CONSTANT_VK u32a c_pbox[18] =
   const u32 r2 = unpack_v8b_from_v32_S ((L)); \
   const u32 r3 = unpack_v8a_from_v32_S ((L)); \
                                               \
-  tmp  = S0[r0];                              \
-  tmp += S1[r1];                              \
-  tmp ^= S2[r2];                              \
-  tmp += S3[r3];                              \
+  tmp  = GET_KEY32 (S0, r0);                  \
+  tmp += GET_KEY32 (S1, r1);                  \
+  tmp ^= GET_KEY32 (S2, r2);                  \
+  tmp += GET_KEY32 (S3, r3);                  \
                                               \
   (R) ^= tmp ^ P[(N)];                        \
 }
@@ -357,6 +402,10 @@ CONSTANT_VK u32a c_pbox[18] =
   L ^= P[17];           \
 }
 
+#ifdef DYNAMIC_LOCAL
+extern __shared__ u32 S[];
+#endif
+
 KERNEL_FQ void FIXED_THREAD_COUNT(FIXED_LOCAL_SIZE) m09000_init (KERN_ATTR_TMPS (pwsafe2_tmp_t))
 {
   /**
@@ -471,22 +520,33 @@ KERNEL_FQ void FIXED_THREAD_COUNT(FIXED_LOCAL_SIZE) m09000_init (KERN_ATTR_TMPS
     P[i] = c_pbox[i];
   }
 
+  #ifdef DYNAMIC_LOCAL
+  // from host
+  #else
   LOCAL_VK u32 S0_all[FIXED_LOCAL_SIZE][256];
   LOCAL_VK u32 S1_all[FIXED_LOCAL_SIZE][256];
   LOCAL_VK u32 S2_all[FIXED_LOCAL_SIZE][256];
   LOCAL_VK u32 S3_all[FIXED_LOCAL_SIZE][256];
+  #endif
 
+  #ifdef BCRYPT_AVOID_BANK_CONFLICTS
+  LOCAL_AS u32 *S0 = S + (FIXED_LOCAL_SIZE * 256 * 0);
+  LOCAL_AS u32 *S1 = S + (FIXED_LOCAL_SIZE * 256 * 1);
+  LOCAL_AS u32 *S2 = S + (FIXED_LOCAL_SIZE * 256 * 2);
+  LOCAL_AS u32 *S3 = S + (FIXED_LOCAL_SIZE * 256 * 3);
+  #else
   LOCAL_AS u32 *S0 = S0_all[lid];
   LOCAL_AS u32 *S1 = S1_all[lid];
   LOCAL_AS u32 *S2 = S2_all[lid];
   LOCAL_AS u32 *S3 = S3_all[lid];
+  #endif
 
   for (u32 i = 0; i < 256; i++)
   {
-    S0[i] = c_sbox0[i];
-    S1[i] = c_sbox1[i];
-    S2[i] = c_sbox2[i];
-    S3[i] = c_sbox3[i];
+    SET_KEY32 (S0, i, c_sbox0[i]);
+    SET_KEY32 (S1, i, c_sbox1[i]);
+    SET_KEY32 (S2, i, c_sbox2[i]);
+    SET_KEY32 (S3, i, c_sbox3[i]);
   }
 
   for (u32 i = 0; i < 18; i++)
@@ -509,59 +569,59 @@ KERNEL_FQ void FIXED_THREAD_COUNT(FIXED_LOCAL_SIZE) m09000_init (KERN_ATTR_TMPS
   {
     BF_ENCRYPT (L0, R0);
 
-    S0[i + 0] = L0;
-    S0[i + 1] = R0;
+    SET_KEY32 (S0, i + 0, L0);
+    SET_KEY32 (S0, i + 1, R0);
 
     BF_ENCRYPT (L0, R0);
 
-    S0[i + 2] = L0;
-    S0[i + 3] = R0;
+    SET_KEY32 (S0, i + 2, L0);
+    SET_KEY32 (S0, i + 3, R0);
   }
 
   for (u32 i = 0; i < 256; i += 4)
   {
     BF_ENCRYPT (L0, R0);
 
-    S1[i + 0] = L0;
-    S1[i + 1] = R0;
+    SET_KEY32 (S1, i + 0, L0);
+    SET_KEY32 (S1, i + 1, R0);
 
     BF_ENCRYPT (L0, R0);
 
-    S1[i + 2] = L0;
-    S1[i + 3] = R0;
+    SET_KEY32 (S1, i + 2, L0);
+    SET_KEY32 (S1, i + 3, R0);
   }
 
   for (u32 i = 0; i < 256; i += 4)
   {
     BF_ENCRYPT (L0, R0);
 
-    S2[i + 0] = L0;
-    S2[i + 1] = R0;
+    SET_KEY32 (S2, i + 0, L0);
+    SET_KEY32 (S2, i + 1, R0);
 
     BF_ENCRYPT (L0, R0);
 
-    S2[i + 2] = L0;
-    S2[i + 3] = R0;
+    SET_KEY32 (S2, i + 2, L0);
+    SET_KEY32 (S2, i + 3, R0);
   }
 
   for (u32 i = 0; i < 256; i += 4)
   {
     BF_ENCRYPT (L0, R0);
 
-    S3[i + 0] = L0;
-    S3[i + 1] = R0;
+    SET_KEY32 (S3, i + 0, L0);
+    SET_KEY32 (S3, i + 1, R0);
 
     BF_ENCRYPT (L0, R0);
 
-    S3[i + 2] = L0;
-    S3[i + 3] = R0;
+    SET_KEY32 (S3, i + 2, L0);
+    SET_KEY32 (S3, i + 3, R0);
   }
 
-  // store
-
   tmps[gid].digest[0] = salt_buf[0];
   tmps[gid].digest[1] = salt_buf[1];
 
+  // store
+
   for (u32 i = 0; i < 18; i++)
   {
     tmps[gid].P[i] = P[i];
@@ -569,10 +629,10 @@ KERNEL_FQ void FIXED_THREAD_COUNT(FIXED_LOCAL_SIZE) m09000_init (KERN_ATTR_TMPS
 
   for (u32 i = 0; i < 256; i++)
   {
-    tmps[gid].S0[i] = S0[i];
-    tmps[gid].S1[i] = S1[i];
-    tmps[gid].S2[i] = S2[i];
-    tmps[gid].S3[i] = S3[i];
+    tmps[gid].S0[i] = GET_KEY32 (S0, i);
+    tmps[gid].S1[i] = GET_KEY32 (S1, i);
+    tmps[gid].S2[i] = GET_KEY32 (S2, i);
+    tmps[gid].S3[i] = GET_KEY32 (S3, i);
   }
 }
 
@@ -602,22 +662,33 @@ KERNEL_FQ void FIXED_THREAD_COUNT(FIXED_LOCAL_SIZE) m09000_loop (KERN_ATTR_TMPS
     P[i] = tmps[gid].P[i];
   }
 
+  #ifdef DYNAMIC_LOCAL
+  // from host
+  #else
   LOCAL_VK u32 S0_all[FIXED_LOCAL_SIZE][256];
   LOCAL_VK u32 S1_all[FIXED_LOCAL_SIZE][256];
   LOCAL_VK u32 S2_all[FIXED_LOCAL_SIZE][256];
   LOCAL_VK u32 S3_all[FIXED_LOCAL_SIZE][256];
+  #endif
 
+  #ifdef BCRYPT_AVOID_BANK_CONFLICTS
+  LOCAL_AS u32 *S0 = S + (FIXED_LOCAL_SIZE * 256 * 0);
+  LOCAL_AS u32 *S1 = S + (FIXED_LOCAL_SIZE * 256 * 1);
+  LOCAL_AS u32 *S2 = S + (FIXED_LOCAL_SIZE * 256 * 2);
+  LOCAL_AS u32 *S3 = S + (FIXED_LOCAL_SIZE * 256 * 3);
+  #else
   LOCAL_AS u32 *S0 = S0_all[lid];
   LOCAL_AS u32 *S1 = S1_all[lid];
   LOCAL_AS u32 *S2 = S2_all[lid];
   LOCAL_AS u32 *S3 = S3_all[lid];
+  #endif
 
   for (u32 i = 0; i < 256; i++)
   {
-    S0[i] = tmps[gid].S0[i];
-    S1[i] = tmps[gid].S1[i];
-    S2[i] = tmps[gid].S2[i];
-    S3[i] = tmps[gid].S3[i];
+    SET_KEY32 (S0, i, tmps[gid].S0[i]);
+    SET_KEY32 (S1, i, tmps[gid].S1[i]);
+    SET_KEY32 (S2, i, tmps[gid].S2[i]);
+    SET_KEY32 (S3, i, tmps[gid].S3[i]);
   }
 
   // loop
@@ -630,8 +701,6 @@ KERNEL_FQ void FIXED_THREAD_COUNT(FIXED_LOCAL_SIZE) m09000_loop (KERN_ATTR_TMPS
     BF_ENCRYPT (L0, R0);
   }
 
-  // store
-
   tmps[gid].digest[0] = L0;
   tmps[gid].digest[1] = R0;
 }
diff --git a/OpenCL/m10700-optimized.cl b/OpenCL/m10700-optimized.cl
index a9b50a6ac..9779c8fe6 100644
--- a/OpenCL/m10700-optimized.cl
+++ b/OpenCL/m10700-optimized.cl
@@ -232,7 +232,7 @@ DECLSPEC void make_sc (u32 *sc, const u32 *pw, const u32 pw_len, const u32 *bl,
 
     u32 i;
 
-    #if (defined IS_AMD || defined IS_HIP) || defined IS_GENERIC
+    #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 0) || defined IS_GENERIC
     for (i = 0; i < pd; i++) sc[idx++] = pw[i];
                              sc[idx++] = pw[i]
                                        | hc_bytealign_be (bl[0],         0, pm4);
@@ -242,8 +242,15 @@ DECLSPEC void make_sc (u32 *sc, const u32 *pw, const u32 pw_len, const u32 *bl,
                              sc[idx++] = hc_bytealign_be (    0, sc[i - 1], pm4);
     #endif
 
-    #ifdef IS_NV
-    int selector = (0x76543210 >> (pm4 * 4)) & 0xffff;
+    #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1) || defined IS_NV
+
+    #if defined IS_NV
+    const int selector = (0x76543210 >> ((pm4 & 3) * 4)) & 0xffff;
+    #endif
+
+    #if (defined IS_AMD || defined IS_HIP)
+    const int selector = l32_from_64_S (0x0706050403020100UL >> ((pm4 & 3) * 8));
+    #endif
 
     for (i = 0; i < pd; i++) sc[idx++] = pw[i];
                              sc[idx++] = pw[i]
@@ -263,16 +270,22 @@ DECLSPEC void make_pt_with_offset (u32 *pt, const u32 offset, const u32 *sc, con
   const u32 om = m % 4;
   const u32 od = m / 4;
 
-  #if (defined IS_AMD || defined IS_HIP) || defined IS_GENERIC
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 0) || defined IS_GENERIC
   pt[0] = hc_bytealign_be (sc[od + 1], sc[od + 0], om);
   pt[1] = hc_bytealign_be (sc[od + 2], sc[od + 1], om);
   pt[2] = hc_bytealign_be (sc[od + 3], sc[od + 2], om);
   pt[3] = hc_bytealign_be (sc[od + 4], sc[od + 3], om);
   #endif
 
-  #ifdef IS_NV
-  int selector = (0x76543210 >> (om * 4)) & 0xffff;
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1) || defined IS_NV
 
+  #if defined IS_NV
+  const int selector = (0x76543210 >> ((om & 3) * 4)) & 0xffff;
+  #endif
+
+  #if (defined IS_AMD || defined IS_HIP)
+  const int selector = l32_from_64_S (0x0706050403020100UL >> ((om & 3) * 8));
+  #endif
   pt[0] = hc_byte_perm (sc[od + 0], sc[od + 1], selector);
   pt[1] = hc_byte_perm (sc[od + 1], sc[od + 2], selector);
   pt[2] = hc_byte_perm (sc[od + 2], sc[od + 3], selector);
diff --git a/OpenCL/m11600-pure.cl b/OpenCL/m11600-pure.cl
index be42e185b..d321aee3a 100644
--- a/OpenCL/m11600-pure.cl
+++ b/OpenCL/m11600-pure.cl
@@ -42,13 +42,20 @@ DECLSPEC void memcat8c_be (u32 *w0, u32 *w1, u32 *w2, u32 *w3, const u32 len, co
   u32 tmp0;
   u32 tmp1;
 
-  #if (defined IS_AMD || defined IS_HIP) || defined IS_GENERIC
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 0) || defined IS_GENERIC
   tmp0 = hc_bytealign_be (0, append, func_len);
   tmp1 = hc_bytealign_be (append, 0, func_len);
   #endif
 
-  #ifdef IS_NV
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1) || defined IS_NV
+
+  #if defined IS_NV
   const int selector = (0x76543210 >> ((func_len & 3) * 4)) & 0xffff;
+  #endif
+
+  #if (defined IS_AMD || defined IS_HIP)
+  const int selector = l32_from_64_S (0x0706050403020100UL >> ((func_len & 3) * 8));
+  #endif
 
   tmp0 = hc_byte_perm (append, 0, selector);
   tmp1 = hc_byte_perm (0, append, selector);
diff --git a/OpenCL/m12500-pure.cl b/OpenCL/m12500-pure.cl
index f8ed47771..020ac1619 100644
--- a/OpenCL/m12500-pure.cl
+++ b/OpenCL/m12500-pure.cl
@@ -37,13 +37,20 @@ DECLSPEC void memcat8c_be (u32 *w0, u32 *w1, u32 *w2, u32 *w3, const u32 len, co
   u32 tmp0;
   u32 tmp1;
 
-  #if (defined IS_AMD || defined IS_HIP) || defined IS_GENERIC
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 0) || defined IS_GENERIC
   tmp0 = hc_bytealign_be (0, append, func_len);
   tmp1 = hc_bytealign_be (append, 0, func_len);
   #endif
 
-  #ifdef IS_NV
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1) || defined IS_NV
+
+  #if defined IS_NV
   const int selector = (0x76543210 >> ((func_len & 3) * 4)) & 0xffff;
+  #endif
+
+  #if (defined IS_AMD || defined IS_HIP)
+  const int selector = l32_from_64_S (0x0706050403020100UL >> ((func_len & 3) * 8));
+  #endif
 
   tmp0 = hc_byte_perm (append, 0, selector);
   tmp1 = hc_byte_perm (0, append, selector);
@@ -797,7 +804,7 @@ KERNEL_FQ void m12500_loop (KERN_ATTR_TMPS (rar3_tmp_t))
    * base
    */
 
-  const u32 pw_len = pws[gid].pw_len & 255;
+  const u32 pw_len = pws[gid].pw_len;
 
   const u32 salt_len = 8;
 
@@ -963,7 +970,7 @@ KERNEL_FQ void m12500_comp (KERN_ATTR_TMPS (rar3_tmp_t))
    * base
    */
 
-  const u32 pw_len = pws[gid].pw_len & 255;
+  const u32 pw_len = pws[gid].pw_len;
 
   const u32 salt_len = 8;
 
diff --git a/OpenCL/m13600-pure.cl b/OpenCL/m13600-pure.cl
index 0202cf0bf..34356b3a9 100644
--- a/OpenCL/m13600-pure.cl
+++ b/OpenCL/m13600-pure.cl
@@ -37,7 +37,7 @@ typedef struct zip2
   u32 verify_bytes;
   u32 compress_length;
   u32 data_len;
-  u32 data_buf[0x4000000];
+  u32 data_buf[0x200000];
   u32 auth_len;
   u32 auth_buf[4];
 
diff --git a/OpenCL/m13711-pure.cl b/OpenCL/m13711-pure.cl
index 812a9ecfc..b8f62904f 100644
--- a/OpenCL/m13711-pure.cl
+++ b/OpenCL/m13711-pure.cl
@@ -366,7 +366,9 @@ KERNEL_FQ void m13711_loop (KERN_ATTR_TMPS_ESALT (vc_tmp_t, vc_t))
         unpackv (tmps, pim_key, gid, i + 3, out[3]);
         unpackv (tmps, pim_key, gid, i + 4, out[4]);
 
-        tmps[gid].pim_check = pim;
+        const u32x pimx = make_u32x (pim);
+
+        unpack (tmps, pim_check, gid, pimx);
       }
     }
 
diff --git a/OpenCL/m13712-pure.cl b/OpenCL/m13712-pure.cl
index f60004897..c00d2144f 100644
--- a/OpenCL/m13712-pure.cl
+++ b/OpenCL/m13712-pure.cl
@@ -417,7 +417,9 @@ KERNEL_FQ void m13712_loop (KERN_ATTR_TMPS_ESALT (vc_tmp_t, vc_t))
         unpackv (tmps, pim_key, gid, i + 3, out[3]);
         unpackv (tmps, pim_key, gid, i + 4, out[4]);
 
-        tmps[gid].pim_check = pim;
+        const u32x pimx = make_u32x (pim);
+
+        unpack (tmps, pim_check, gid, pimx);
       }
     }
 
diff --git a/OpenCL/m13713-pure.cl b/OpenCL/m13713-pure.cl
index 4a18a3885..ffd193ff8 100644
--- a/OpenCL/m13713-pure.cl
+++ b/OpenCL/m13713-pure.cl
@@ -482,7 +482,9 @@ KERNEL_FQ void m13713_loop (KERN_ATTR_TMPS_ESALT (vc_tmp_t, vc_t))
         unpackv (tmps, pim_key, gid, i + 3, out[3]);
         unpackv (tmps, pim_key, gid, i + 4, out[4]);
 
-        tmps[gid].pim_check = pim;
+        const u32x pimx = make_u32x (pim);
+
+        unpack (tmps, pim_check, gid, pimx);
       }
     }
 
diff --git a/OpenCL/m13721-pure.cl b/OpenCL/m13721-pure.cl
index dd1b8f752..743cd3f30 100644
--- a/OpenCL/m13721-pure.cl
+++ b/OpenCL/m13721-pure.cl
@@ -458,7 +458,9 @@ KERNEL_FQ void m13721_loop (KERN_ATTR_TMPS_ESALT (vc64_tmp_t, vc_t))
         unpack64v (tmps, pim_key, gid, i + 6, out[6]);
         unpack64v (tmps, pim_key, gid, i + 7, out[7]);
 
-        tmps[gid].pim_check = pim;
+        const u32x pimx = make_u32x (pim);
+
+        unpack (tmps, pim_check, gid, pimx);
       }
     }
 
diff --git a/OpenCL/m13722-pure.cl b/OpenCL/m13722-pure.cl
index 80a9fb1b6..0425af528 100644
--- a/OpenCL/m13722-pure.cl
+++ b/OpenCL/m13722-pure.cl
@@ -509,7 +509,9 @@ KERNEL_FQ void m13722_loop (KERN_ATTR_TMPS_ESALT (vc64_tmp_t, vc_t))
         unpack64v (tmps, pim_key, gid, i + 6, out[6]);
         unpack64v (tmps, pim_key, gid, i + 7, out[7]);
 
-        tmps[gid].pim_check = pim;
+        const u32x pimx = make_u32x (pim);
+
+        unpack (tmps, pim_check, gid, pimx);
       }
     }
 
diff --git a/OpenCL/m13723-pure.cl b/OpenCL/m13723-pure.cl
index 465c5000d..8d2c51293 100644
--- a/OpenCL/m13723-pure.cl
+++ b/OpenCL/m13723-pure.cl
@@ -574,7 +574,9 @@ KERNEL_FQ void m13723_loop (KERN_ATTR_TMPS_ESALT (vc64_tmp_t, vc_t))
         unpack64v (tmps, pim_key, gid, i + 6, out[6]);
         unpack64v (tmps, pim_key, gid, i + 7, out[7]);
 
-        tmps[gid].pim_check = pim;
+        const u32x pimx = make_u32x (pim);
+
+        unpack (tmps, pim_check, gid, pimx);
       }
     }
 
diff --git a/OpenCL/m13731-pure.cl b/OpenCL/m13731-pure.cl
index 53c0aee63..d29bbe121 100644
--- a/OpenCL/m13731-pure.cl
+++ b/OpenCL/m13731-pure.cl
@@ -622,7 +622,9 @@ KERNEL_FQ void m13731_loop (KERN_ATTR_TMPS_ESALT (vc_tmp_t, vc_t))
         unpackv (tmps, pim_key, gid, i + 14, out[14]);
         unpackv (tmps, pim_key, gid, i + 15, out[15]);
 
-        tmps[gid].pim_check = pim;
+        const u32x pimx = make_u32x (pim);
+
+        unpack (tmps, pim_check, gid, pimx);
       }
     }
 
diff --git a/OpenCL/m13732-pure.cl b/OpenCL/m13732-pure.cl
index fde64f5cd..57e153479 100644
--- a/OpenCL/m13732-pure.cl
+++ b/OpenCL/m13732-pure.cl
@@ -673,7 +673,9 @@ KERNEL_FQ void m13732_loop (KERN_ATTR_TMPS_ESALT (vc_tmp_t, vc_t))
         unpackv (tmps, pim_key, gid, i + 14, out[14]);
         unpackv (tmps, pim_key, gid, i + 15, out[15]);
 
-        tmps[gid].pim_check = pim;
+        const u32x pimx = make_u32x (pim);
+
+        unpack (tmps, pim_check, gid, pimx);
       }
     }
 
diff --git a/OpenCL/m13733-pure.cl b/OpenCL/m13733-pure.cl
index 63d5b1a4c..82994a66a 100644
--- a/OpenCL/m13733-pure.cl
+++ b/OpenCL/m13733-pure.cl
@@ -738,7 +738,9 @@ KERNEL_FQ void m13733_loop (KERN_ATTR_TMPS_ESALT (vc_tmp_t, vc_t))
         unpackv (tmps, pim_key, gid, i + 14, out[14]);
         unpackv (tmps, pim_key, gid, i + 15, out[15]);
 
-        tmps[gid].pim_check = pim;
+        const u32x pimx = make_u32x (pim);
+
+        unpack (tmps, pim_check, gid, pimx);
       }
     }
 
diff --git a/OpenCL/m13751-pure.cl b/OpenCL/m13751-pure.cl
index e795fe1e6..9fc0f8d2a 100644
--- a/OpenCL/m13751-pure.cl
+++ b/OpenCL/m13751-pure.cl
@@ -435,7 +435,9 @@ KERNEL_FQ void m13751_loop (KERN_ATTR_TMPS_ESALT (vc_tmp_t, vc_t))
         unpackv (tmps, pim_key, gid, i + 6, out[6]);
         unpackv (tmps, pim_key, gid, i + 7, out[7]);
 
-        tmps[gid].pim_check = pim;
+        const u32x pimx = make_u32x (pim);
+
+        unpack (tmps, pim_check, gid, pimx);
       }
     }
 
diff --git a/OpenCL/m13752-pure.cl b/OpenCL/m13752-pure.cl
index 5ae1b2468..f5490bcc9 100644
--- a/OpenCL/m13752-pure.cl
+++ b/OpenCL/m13752-pure.cl
@@ -457,7 +457,9 @@ KERNEL_FQ void m13752_loop (KERN_ATTR_TMPS_ESALT (vc_tmp_t, vc_t))
         unpackv (tmps, pim_key, gid, i + 6, out[6]);
         unpackv (tmps, pim_key, gid, i + 7, out[7]);
 
-        tmps[gid].pim_check = pim;
+        const u32x pimx = make_u32x (pim);
+
+        unpack (tmps, pim_check, gid, pimx);
       }
     }
 
diff --git a/OpenCL/m13753-pure.cl b/OpenCL/m13753-pure.cl
index 0d33af281..5d76f0637 100644
--- a/OpenCL/m13753-pure.cl
+++ b/OpenCL/m13753-pure.cl
@@ -522,7 +522,9 @@ KERNEL_FQ void m13753_loop (KERN_ATTR_TMPS_ESALT (vc_tmp_t, vc_t))
         unpackv (tmps, pim_key, gid, i + 6, out[6]);
         unpackv (tmps, pim_key, gid, i + 7, out[7]);
 
-        tmps[gid].pim_check = pim;
+        const u32x pimx = make_u32x (pim);
+
+        unpack (tmps, pim_check, gid, pimx);
       }
     }
 
diff --git a/OpenCL/m13771-pure.cl b/OpenCL/m13771-pure.cl
index 2c5e24985..b22738aa1 100644
--- a/OpenCL/m13771-pure.cl
+++ b/OpenCL/m13771-pure.cl
@@ -537,7 +537,9 @@ KERNEL_FQ void m13771_loop (KERN_ATTR_TMPS_ESALT (vc64_sbog_tmp_t, vc_t))
         unpack64v (tmps, pim_key, gid, i + 6, out[6]);
         unpack64v (tmps, pim_key, gid, i + 7, out[7]);
 
-        tmps[gid].pim_check = pim;
+        const u32x pimx = make_u32x (pim);
+
+        unpack (tmps, pim_check, gid, pimx);
       }
     }
 
diff --git a/OpenCL/m13772-pure.cl b/OpenCL/m13772-pure.cl
index 449a2c2b9..7d189e4e3 100644
--- a/OpenCL/m13772-pure.cl
+++ b/OpenCL/m13772-pure.cl
@@ -588,7 +588,9 @@ KERNEL_FQ void m13772_loop (KERN_ATTR_TMPS_ESALT (vc64_sbog_tmp_t, vc_t))
         unpack64v (tmps, pim_key, gid, i + 6, out[6]);
         unpack64v (tmps, pim_key, gid, i + 7, out[7]);
 
-        tmps[gid].pim_check = pim;
+        const u32x pimx = make_u32x (pim);
+
+        unpack (tmps, pim_check, gid, pimx);
       }
     }
 
diff --git a/OpenCL/m13773-pure.cl b/OpenCL/m13773-pure.cl
index 7299593f8..f257ec5eb 100644
--- a/OpenCL/m13773-pure.cl
+++ b/OpenCL/m13773-pure.cl
@@ -653,7 +653,9 @@ KERNEL_FQ void m13773_loop (KERN_ATTR_TMPS_ESALT (vc64_sbog_tmp_t, vc_t))
         unpack64v (tmps, pim_key, gid, i + 6, out[6]);
         unpack64v (tmps, pim_key, gid, i + 7, out[7]);
 
-        tmps[gid].pim_check = pim;
+        const u32x pimx = make_u32x (pim);
+
+        unpack (tmps, pim_check, gid, pimx);
       }
     }
 
diff --git a/OpenCL/m13800_a0-optimized.cl b/OpenCL/m13800_a0-optimized.cl
index 6758ffbd4..043ed0d13 100644
--- a/OpenCL/m13800_a0-optimized.cl
+++ b/OpenCL/m13800_a0-optimized.cl
@@ -51,7 +51,7 @@ DECLSPEC void memcat64c_be (u32x *block, const u32 offset, u32x *carry)
   u32x tmp15;
   u32x tmp16;
 
-  #if (defined IS_AMD || defined IS_HIP) || defined IS_GENERIC
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 0) || defined IS_GENERIC
   tmp00 = hc_bytealign_be (        0, carry[ 0], offset);
   tmp01 = hc_bytealign_be (carry[ 0], carry[ 1], offset);
   tmp02 = hc_bytealign_be (carry[ 1], carry[ 2], offset);
@@ -71,8 +71,15 @@ DECLSPEC void memcat64c_be (u32x *block, const u32 offset, u32x *carry)
   tmp16 = hc_bytealign_be (carry[15],         0, offset);
   #endif
 
-  #ifdef IS_NV
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1) || defined IS_NV
+
+  #if defined IS_NV
   const int selector = (0x76543210 >> ((offset & 3) * 4)) & 0xffff;
+  #endif
+
+  #if (defined IS_AMD || defined IS_HIP)
+  const int selector = l32_from_64_S (0x0706050403020100UL >> ((offset & 3) * 8));
+  #endif
 
   tmp00 = hc_byte_perm (carry[ 0],         0, selector);
   tmp01 = hc_byte_perm (carry[ 1], carry[ 0], selector);
diff --git a/OpenCL/m13800_a1-optimized.cl b/OpenCL/m13800_a1-optimized.cl
index 85e711b94..4227e48d5 100644
--- a/OpenCL/m13800_a1-optimized.cl
+++ b/OpenCL/m13800_a1-optimized.cl
@@ -49,7 +49,7 @@ DECLSPEC void memcat64c_be (u32x *block, const u32 offset, u32x *carry)
   u32x tmp15;
   u32x tmp16;
 
-  #if (defined IS_AMD || defined IS_HIP) || defined IS_GENERIC
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 0) || defined IS_GENERIC
   tmp00 = hc_bytealign_be (        0, carry[ 0], offset);
   tmp01 = hc_bytealign_be (carry[ 0], carry[ 1], offset);
   tmp02 = hc_bytealign_be (carry[ 1], carry[ 2], offset);
@@ -69,8 +69,15 @@ DECLSPEC void memcat64c_be (u32x *block, const u32 offset, u32x *carry)
   tmp16 = hc_bytealign_be (carry[15],         0, offset);
   #endif
 
-  #ifdef IS_NV
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1) || defined IS_NV
+
+  #if defined IS_NV
   const int selector = (0x76543210 >> ((offset & 3) * 4)) & 0xffff;
+  #endif
+
+  #if (defined IS_AMD || defined IS_HIP)
+  const int selector = l32_from_64_S (0x0706050403020100UL >> ((offset & 3) * 8));
+  #endif
 
   tmp00 = hc_byte_perm (carry[ 0],         0, selector);
   tmp01 = hc_byte_perm (carry[ 1], carry[ 0], selector);
diff --git a/OpenCL/m13800_a3-optimized.cl b/OpenCL/m13800_a3-optimized.cl
index 65b759de0..895d4378c 100644
--- a/OpenCL/m13800_a3-optimized.cl
+++ b/OpenCL/m13800_a3-optimized.cl
@@ -48,7 +48,7 @@ DECLSPEC void memcat64c_be (u32x *block, const u32 offset, u32x *carry)
   u32x tmp15;
   u32x tmp16;
 
-  #if (defined IS_AMD || defined IS_HIP) || defined IS_GENERIC
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 0) || defined IS_GENERIC
   tmp00 = hc_bytealign_be (        0, carry[ 0], offset);
   tmp01 = hc_bytealign_be (carry[ 0], carry[ 1], offset);
   tmp02 = hc_bytealign_be (carry[ 1], carry[ 2], offset);
@@ -68,8 +68,15 @@ DECLSPEC void memcat64c_be (u32x *block, const u32 offset, u32x *carry)
   tmp16 = hc_bytealign_be (carry[15],         0, offset);
   #endif
 
-  #ifdef IS_NV
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1) || defined IS_NV
+
+  #if defined IS_NV
   const int selector = (0x76543210 >> ((offset & 3) * 4)) & 0xffff;
+  #endif
+
+  #if (defined IS_AMD || defined IS_HIP)
+  const int selector = l32_from_64_S (0x0706050403020100UL >> ((offset & 3) * 8));
+  #endif
 
   tmp00 = hc_byte_perm (carry[ 0],         0, selector);
   tmp01 = hc_byte_perm (carry[ 1], carry[ 0], selector);
diff --git a/OpenCL/m15700-pure.cl b/OpenCL/m15700-pure.cl
index 70b4ed4fd..09819b085 100644
--- a/OpenCL/m15700-pure.cl
+++ b/OpenCL/m15700-pure.cl
@@ -439,7 +439,7 @@ KERNEL_FQ void m15700_init (KERN_ATTR_TMPS_ESALT (scrypt_tmp_t, ethereum_scrypt_
     digest[6] = sha256_hmac_ctx2.opad.h[6];
     digest[7] = sha256_hmac_ctx2.opad.h[7];
 
-    #if defined IS_CUDA || defined IS_HIP
+    #if defined IS_CUDA
     const uint4 tmp0 = make_uint4 (digest[0], digest[1], digest[2], digest[3]);
     const uint4 tmp1 = make_uint4 (digest[4], digest[5], digest[6], digest[7]);
     #else
@@ -467,7 +467,7 @@ KERNEL_FQ void m15700_init (KERN_ATTR_TMPS_ESALT (scrypt_tmp_t, ethereum_scrypt_
 
     uint4 X[4];
 
-    #if defined IS_CUDA || defined IS_HIP
+    #if defined IS_CUDA
     X[0] = make_uint4 (T[0].x, T[1].y, T[2].z, T[3].w);
     X[1] = make_uint4 (T[1].x, T[2].y, T[3].z, T[0].w);
     X[2] = make_uint4 (T[2].x, T[3].y, T[0].z, T[1].w);
@@ -577,7 +577,7 @@ KERNEL_FQ void m15700_comp (KERN_ATTR_TMPS_ESALT (scrypt_tmp_t, ethereum_scrypt_
 
     uint4 T[4];
 
-    #if defined IS_CUDA || defined IS_HIP
+    #if defined IS_CUDA
     T[0] = make_uint4 (X[0].x, X[3].y, X[2].z, X[1].w);
     T[1] = make_uint4 (X[1].x, X[0].y, X[3].z, X[2].w);
     T[2] = make_uint4 (X[2].x, X[1].y, X[0].z, X[3].w);
diff --git a/OpenCL/m18600-pure.cl b/OpenCL/m18600-pure.cl
index 061c61bd8..f98aca9c0 100644
--- a/OpenCL/m18600-pure.cl
+++ b/OpenCL/m18600-pure.cl
@@ -319,6 +319,51 @@ CONSTANT_VK u32a c_pbox[18] =
   0x9216d5d9, 0x8979fb1b
 };
 
+// Yes, works only with CUDA atm
+
+#ifdef DYNAMIC_LOCAL
+#define BCRYPT_AVOID_BANK_CONFLICTS
+#endif
+
+#ifdef BCRYPT_AVOID_BANK_CONFLICTS
+
+// access pattern: minimize bank ID based on thread ID but thread ID is not saved from computation
+
+#define KEY32(lid,key) (((key) * FIXED_LOCAL_SIZE_COMP) + (lid))
+
+DECLSPEC u32 GET_KEY32 (LOCAL_AS u32 *S, const u64 key)
+{
+  const u64 lid = get_local_id (0);
+
+  return S[KEY32 (lid, key)];
+}
+
+DECLSPEC void SET_KEY32 (LOCAL_AS u32 *S, const u64 key, const u32 val)
+{
+  const u64 lid = get_local_id (0);
+
+  S[KEY32 (lid, key)] = val;
+}
+
+#undef KEY32
+
+#else
+
+// access pattern: linear access with S offset already set to right offset based on thread ID saving it from compuation
+//                 makes sense if there are not thread ID's (for instance on CPU)
+
+DECLSPEC inline u32 GET_KEY32 (LOCAL_AS u32 *S, const u64 key)
+{
+  return S[key];
+}
+
+DECLSPEC inline void SET_KEY32 (LOCAL_AS u32 *S, const u64 key, const u32 val)
+{
+  S[key] = val;
+}
+
+#endif
+
 #define BF_ROUND(L,R,N)                       \
 {                                             \
   u32 tmp;                                    \
@@ -328,10 +373,10 @@ CONSTANT_VK u32a c_pbox[18] =
   const u32 r2 = unpack_v8b_from_v32_S ((L)); \
   const u32 r3 = unpack_v8a_from_v32_S ((L)); \
                                               \
-  tmp  = S0[r0];                              \
-  tmp += S1[r1];                              \
-  tmp ^= S2[r2];                              \
-  tmp += S3[r3];                              \
+  tmp  = GET_KEY32 (S0, r0);                  \
+  tmp += GET_KEY32 (S1, r1);                  \
+  tmp ^= GET_KEY32 (S2, r2);                  \
+  tmp += GET_KEY32 (S3, r3);                  \
                                               \
   (R) ^= tmp ^ P[(N)];                        \
 }
@@ -366,6 +411,10 @@ CONSTANT_VK u32a c_pbox[18] =
   L ^= P[17];           \
 }
 
+#ifdef DYNAMIC_LOCAL
+extern __shared__ u32 S[];
+#endif
+
 DECLSPEC void hmac_sha1_run_V (u32x *w0, u32x *w1, u32x *w2, u32x *w3, u32x *ipad, u32x *opad, u32x *digest)
 {
   digest[0] = ipad[0];
@@ -586,7 +635,7 @@ KERNEL_FQ void m18600_loop (KERN_ATTR_TMPS_ESALT (odf11_tmp_t, odf11_t))
   }
 }
 
-KERNEL_FQ void FIXED_THREAD_COUNT(FIXED_LOCAL_SIZE) m18600_comp (KERN_ATTR_TMPS_ESALT (odf11_tmp_t, odf11_t))
+KERNEL_FQ void FIXED_THREAD_COUNT(FIXED_LOCAL_SIZE_COMP) m18600_comp (KERN_ATTR_TMPS_ESALT (odf11_tmp_t, odf11_t))
 {
   const u64 gid = get_global_id (0);
   const u64 lid = get_local_id (0);
@@ -616,22 +665,33 @@ KERNEL_FQ void FIXED_THREAD_COUNT(FIXED_LOCAL_SIZE) m18600_comp (KERN_ATTR_TMPS_
     P[i] = c_pbox[i] ^ ukey[i % 4];
   }
 
-  LOCAL_VK u32 S0_all[FIXED_LOCAL_SIZE][256];
-  LOCAL_VK u32 S1_all[FIXED_LOCAL_SIZE][256];
-  LOCAL_VK u32 S2_all[FIXED_LOCAL_SIZE][256];
-  LOCAL_VK u32 S3_all[FIXED_LOCAL_SIZE][256];
+  #ifdef DYNAMIC_LOCAL
+  // from host
+  #else
+  LOCAL_VK u32 S0_all[FIXED_LOCAL_SIZE_COMP][256];
+  LOCAL_VK u32 S1_all[FIXED_LOCAL_SIZE_COMP][256];
+  LOCAL_VK u32 S2_all[FIXED_LOCAL_SIZE_COMP][256];
+  LOCAL_VK u32 S3_all[FIXED_LOCAL_SIZE_COMP][256];
+  #endif
 
+  #ifdef BCRYPT_AVOID_BANK_CONFLICTS
+  LOCAL_AS u32 *S0 = S + (FIXED_LOCAL_SIZE_COMP * 256 * 0);
+  LOCAL_AS u32 *S1 = S + (FIXED_LOCAL_SIZE_COMP * 256 * 1);
+  LOCAL_AS u32 *S2 = S + (FIXED_LOCAL_SIZE_COMP * 256 * 2);
+  LOCAL_AS u32 *S3 = S + (FIXED_LOCAL_SIZE_COMP * 256 * 3);
+  #else
   LOCAL_AS u32 *S0 = S0_all[lid];
   LOCAL_AS u32 *S1 = S1_all[lid];
   LOCAL_AS u32 *S2 = S2_all[lid];
   LOCAL_AS u32 *S3 = S3_all[lid];
+  #endif
 
   for (u32 i = 0; i < 256; i++)
   {
-    S0[i] = c_sbox0[i];
-    S1[i] = c_sbox1[i];
-    S2[i] = c_sbox2[i];
-    S3[i] = c_sbox3[i];
+    SET_KEY32 (S0, i, c_sbox0[i]);
+    SET_KEY32 (S1, i, c_sbox1[i]);
+    SET_KEY32 (S2, i, c_sbox2[i]);
+    SET_KEY32 (S3, i, c_sbox3[i]);
   }
 
   u32 L0 = 0;
@@ -649,52 +709,52 @@ KERNEL_FQ void FIXED_THREAD_COUNT(FIXED_LOCAL_SIZE) m18600_comp (KERN_ATTR_TMPS_
   {
     BF_ENCRYPT (L0, R0);
 
-    S0[i + 0] = L0;
-    S0[i + 1] = R0;
+    SET_KEY32 (S0, i + 0, L0);
+    SET_KEY32 (S0, i + 1, R0);
 
     BF_ENCRYPT (L0, R0);
 
-    S0[i + 2] = L0;
-    S0[i + 3] = R0;
+    SET_KEY32 (S0, i + 2, L0);
+    SET_KEY32 (S0, i + 3, R0);
   }
 
   for (u32 i = 0; i < 256; i += 4)
   {
     BF_ENCRYPT (L0, R0);
 
-    S1[i + 0] = L0;
-    S1[i + 1] = R0;
+    SET_KEY32 (S1, i + 0, L0);
+    SET_KEY32 (S1, i + 1, R0);
 
     BF_ENCRYPT (L0, R0);
 
-    S1[i + 2] = L0;
-    S1[i + 3] = R0;
+    SET_KEY32 (S1, i + 2, L0);
+    SET_KEY32 (S1, i + 3, R0);
   }
 
   for (u32 i = 0; i < 256; i += 4)
   {
     BF_ENCRYPT (L0, R0);
 
-    S2[i + 0] = L0;
-    S2[i + 1] = R0;
+    SET_KEY32 (S2, i + 0, L0);
+    SET_KEY32 (S2, i + 1, R0);
 
     BF_ENCRYPT (L0, R0);
 
-    S2[i + 2] = L0;
-    S2[i + 3] = R0;
+    SET_KEY32 (S2, i + 2, L0);
+    SET_KEY32 (S2, i + 3, R0);
   }
 
   for (u32 i = 0; i < 256; i += 4)
   {
     BF_ENCRYPT (L0, R0);
 
-    S3[i + 0] = L0;
-    S3[i + 1] = R0;
+    SET_KEY32 (S3, i + 0, L0);
+    SET_KEY32 (S3, i + 1, R0);
 
     BF_ENCRYPT (L0, R0);
 
-    S3[i + 2] = L0;
-    S3[i + 3] = R0;
+    SET_KEY32 (S3, i + 2, L0);
+    SET_KEY32 (S3, i + 3, R0);
   }
 
   GLOBAL_AS const odf11_t *es = &esalt_bufs[DIGESTS_OFFSET];
diff --git a/OpenCL/m22000-pure.cl b/OpenCL/m22000-pure.cl
index cfe645bc7..6117d2e31 100644
--- a/OpenCL/m22000-pure.cl
+++ b/OpenCL/m22000-pure.cl
@@ -797,11 +797,7 @@ KERNEL_FQ void m22000_aux3 (KERN_ATTR_TMPS_ESALT (wpa_pbkdf2_tmp_t, wpa_t))
     s_te4[i] = te4[i];
   }
 
-  #if defined IS_CUDA || defined IS_HIP
-  __syncthreads();
-  #else
   SYNC_THREADS ();
-  #endif
 
   #else
 
diff --git a/OpenCL/m22001-pure.cl b/OpenCL/m22001-pure.cl
index 95431e08f..23015b863 100644
--- a/OpenCL/m22001-pure.cl
+++ b/OpenCL/m22001-pure.cl
@@ -610,11 +610,7 @@ KERNEL_FQ void m22001_aux3 (KERN_ATTR_TMPS_ESALT (wpa_pmk_tmp_t, wpa_t))
     s_te4[i] = te4[i];
   }
 
-  #if defined IS_CUDA || defined IS_HIP
-  __syncthreads();
-  #else
   SYNC_THREADS ();
-  #endif
 
   #else
 
diff --git a/OpenCL/m22700-pure.cl b/OpenCL/m22700-pure.cl
index 4660843a2..a28b458c2 100644
--- a/OpenCL/m22700-pure.cl
+++ b/OpenCL/m22700-pure.cl
@@ -374,7 +374,7 @@ KERNEL_FQ void m22700_init (KERN_ATTR_TMPS (scrypt_tmp_t))
     digest[6] = sha256_hmac_ctx2.opad.h[6];
     digest[7] = sha256_hmac_ctx2.opad.h[7];
 
-    #if defined IS_CUDA || defined IS_HIP
+    #if defined IS_CUDA
     const uint4 tmp0 = make_uint4 (digest[0], digest[1], digest[2], digest[3]);
     const uint4 tmp1 = make_uint4 (digest[4], digest[5], digest[6], digest[7]);
     #else
@@ -402,7 +402,7 @@ KERNEL_FQ void m22700_init (KERN_ATTR_TMPS (scrypt_tmp_t))
 
     uint4 X[4];
 
-    #if defined IS_CUDA || defined IS_HIP
+    #if defined IS_CUDA
     X[0] = make_uint4 (T[0].x, T[1].y, T[2].z, T[3].w);
     X[1] = make_uint4 (T[1].x, T[2].y, T[3].z, T[0].w);
     X[2] = make_uint4 (T[2].x, T[3].y, T[0].z, T[1].w);
@@ -575,7 +575,7 @@ KERNEL_FQ void m22700_comp (KERN_ATTR_TMPS (scrypt_tmp_t))
 
     uint4 T[4];
 
-    #if defined IS_CUDA || defined IS_HIP
+    #if defined IS_CUDA
     T[0] = make_uint4 (X[0].x, X[3].y, X[2].z, X[1].w);
     T[1] = make_uint4 (X[1].x, X[0].y, X[3].z, X[2].w);
     T[2] = make_uint4 (X[2].x, X[1].y, X[0].z, X[3].w);
diff --git a/OpenCL/m23700-pure.cl b/OpenCL/m23700-pure.cl
index af287574e..c33c5f123 100644
--- a/OpenCL/m23700-pure.cl
+++ b/OpenCL/m23700-pure.cl
@@ -145,13 +145,20 @@ DECLSPEC void memcat8c_be (u32 *w0, u32 *w1, u32 *w2, u32 *w3, const u32 len, co
   u32 tmp0;
   u32 tmp1;
 
-  #if (defined IS_AMD || defined IS_HIP) || defined IS_GENERIC
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 0) || defined IS_GENERIC
   tmp0 = hc_bytealign_be (0, append, func_len);
   tmp1 = hc_bytealign_be (append, 0, func_len);
   #endif
 
-  #ifdef IS_NV
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1) || defined IS_NV
+
+  #if defined IS_NV
   const int selector = (0x76543210 >> ((func_len & 3) * 4)) & 0xffff;
+  #endif
+
+  #if (defined IS_AMD || defined IS_HIP)
+  const int selector = l32_from_64_S (0x0706050403020100UL >> ((func_len & 3) * 8));
+  #endif
 
   tmp0 = hc_byte_perm (append, 0, selector);
   tmp1 = hc_byte_perm (0, append, selector);
@@ -905,7 +912,7 @@ KERNEL_FQ void m23700_loop (KERN_ATTR_TMPS_ESALT (rar3_tmp_t, rar3_t))
    * base
    */
 
-  const u32 pw_len = pws[gid].pw_len & 255;
+  const u32 pw_len = pws[gid].pw_len;
 
   const u32 salt_len = 8;
 
@@ -1079,7 +1086,7 @@ KERNEL_FQ void m23700_comp (KERN_ATTR_TMPS_ESALT (rar3_tmp_t, rar3_t))
    * base
    */
 
-  const u32 pw_len = pws[gid].pw_len & 255;
+  const u32 pw_len = pws[gid].pw_len;
 
   const u32 salt_len = 8;
 
diff --git a/OpenCL/m23800-pure.cl b/OpenCL/m23800-pure.cl
index f6d345677..fb38fab7d 100644
--- a/OpenCL/m23800-pure.cl
+++ b/OpenCL/m23800-pure.cl
@@ -56,13 +56,20 @@ DECLSPEC void memcat8c_be (u32 *w0, u32 *w1, u32 *w2, u32 *w3, const u32 len, co
   u32 tmp0;
   u32 tmp1;
 
-  #if (defined IS_AMD || defined IS_HIP) || defined IS_GENERIC
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 0) || defined IS_GENERIC
   tmp0 = hc_bytealign_be (0, append, func_len);
   tmp1 = hc_bytealign_be (append, 0, func_len);
   #endif
 
-  #ifdef IS_NV
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1) || defined IS_NV
+
+  #if defined IS_NV
   const int selector = (0x76543210 >> ((func_len & 3) * 4)) & 0xffff;
+  #endif
+
+  #if (defined IS_AMD || defined IS_HIP)
+  const int selector = l32_from_64_S (0x0706050403020100UL >> ((func_len & 3) * 8));
+  #endif
 
   tmp0 = hc_byte_perm (append, 0, selector);
   tmp1 = hc_byte_perm (0, append, selector);
@@ -816,7 +823,7 @@ KERNEL_FQ void m23800_loop (KERN_ATTR_TMPS_HOOKS_ESALT (rar3_tmp_t, rar3_hook_t,
    * base
    */
 
-  const u32 pw_len = pws[gid].pw_len & 255;
+  const u32 pw_len = pws[gid].pw_len;
 
   const u32 salt_len = 8;
 
@@ -983,7 +990,7 @@ KERNEL_FQ void m23800_hook23 (KERN_ATTR_TMPS_HOOKS_ESALT (rar3_tmp_t, rar3_hook_
    * base
    */
 
-  const u32 pw_len = pws[gid].pw_len & 255;
+  const u32 pw_len = pws[gid].pw_len;
 
   const u32 salt_len = 8;
 
diff --git a/OpenCL/m25000-pure.cl b/OpenCL/m25000-pure.cl
new file mode 100644
index 000000000..491f9a2b6
--- /dev/null
+++ b/OpenCL/m25000-pure.cl
@@ -0,0 +1,592 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+//#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include "inc_vendor.h"
+#include "inc_types.h"
+#include "inc_platform.cl"
+#include "inc_common.cl"
+#include "inc_simd.cl"
+#include "inc_hash_md5.cl"
+#include "inc_hash_sha1.cl"
+#endif
+
+#define COMPARE_S "inc_comp_single.cl"
+#define COMPARE_M "inc_comp_multi.cl"
+
+#define SNMPV3_SALT_MAX             1500
+#define SNMPV3_ENGINEID_MAX         34
+#define SNMPV3_MSG_AUTH_PARAMS_LEN  12
+#define SNMPV3_ROUNDS               1048576
+#define SNMPV3_MAX_PW_LENGTH        64
+
+#define SNMPV3_TMP_ELEMS            4096 // 4096 = (256 (max pw length) * 64) / sizeof (u32)
+#define SNMPV3_HASH_ELEMS_MD5       4
+#define SNMPV3_HASH_ELEMS_SHA1      8 // 8 = aligned 5
+
+#define SNMPV3_MAX_SALT_ELEMS       512 // 512 * 4 = 2048 > 1500, also has to be multiple of 64
+#define SNMPV3_MAX_ENGINE_ELEMS     16  // 16 * 4 = 64 > 32, also has to be multiple of 64
+#define SNMPV3_MAX_PNUM_ELEMS       4   // 4 * 4 = 16 > 9
+
+#define SNMPV3_MAX_PW_LENGTH_OPT    64
+#define SNMPV3_TMP_ELEMS_OPT        ((SNMPV3_MAX_PW_LENGTH_OPT * SNMPV3_MAX_PW_LENGTH) / 4)
+                                    // (64 * 64) / 4 = 1024
+                                    // for pw length > 64 we use global memory reads
+
+typedef struct hmac_md5_tmp
+{
+  u32 tmp_md5[SNMPV3_TMP_ELEMS];
+  u32 tmp_sha1[SNMPV3_TMP_ELEMS];
+
+  u32 h_md5[SNMPV3_HASH_ELEMS_MD5];
+  u32 h_sha1[SNMPV3_HASH_ELEMS_SHA1];
+
+} hmac_md5_tmp_t;
+
+typedef struct snmpv3
+{
+  u32 salt_buf[SNMPV3_MAX_SALT_ELEMS];
+  u32 salt_len;
+
+  u32 engineID_buf[SNMPV3_MAX_ENGINE_ELEMS];
+  u32 engineID_len;
+
+  u32 packet_number[SNMPV3_MAX_PNUM_ELEMS];
+
+} snmpv3_t;
+
+KERNEL_FQ void m25000_init (KERN_ATTR_TMPS_ESALT (hmac_md5_tmp_t, snmpv3_t))
+{
+  /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  const u32 pw_len = pws[gid].pw_len;
+
+  u32 w[64] = { 0 };
+
+  for (u32 i = 0, idx = 0; i < pw_len; i += 4, idx += 1)
+  {
+    w[idx] = pws[gid].i[idx];
+  }
+
+  u8 *src_ptr = (u8 *) w;
+
+  // password 64 times, also swapped
+
+  u32 dst_buf[16];
+
+  u8 *dst_ptr = (u8 *) dst_buf;
+
+  int tmp_idx = 0;
+
+  for (int i = 0; i < 64; i++)
+  {
+    for (int j = 0; j < pw_len; j++)
+    {
+      const int dst_idx = tmp_idx & 63;
+
+      dst_ptr[dst_idx] = src_ptr[j];
+
+      // write to global memory every time 64 byte are written into cache
+
+      if (dst_idx == 63)
+      {
+        const int tmp_idx4 = (tmp_idx - 63) / 4;
+
+        // md5
+
+        tmps[gid].tmp_md5[tmp_idx4 +  0] = dst_buf[ 0];
+        tmps[gid].tmp_md5[tmp_idx4 +  1] = dst_buf[ 1];
+        tmps[gid].tmp_md5[tmp_idx4 +  2] = dst_buf[ 2];
+        tmps[gid].tmp_md5[tmp_idx4 +  3] = dst_buf[ 3];
+        tmps[gid].tmp_md5[tmp_idx4 +  4] = dst_buf[ 4];
+        tmps[gid].tmp_md5[tmp_idx4 +  5] = dst_buf[ 5];
+        tmps[gid].tmp_md5[tmp_idx4 +  6] = dst_buf[ 6];
+        tmps[gid].tmp_md5[tmp_idx4 +  7] = dst_buf[ 7];
+        tmps[gid].tmp_md5[tmp_idx4 +  8] = dst_buf[ 8];
+        tmps[gid].tmp_md5[tmp_idx4 +  9] = dst_buf[ 9];
+        tmps[gid].tmp_md5[tmp_idx4 + 10] = dst_buf[10];
+        tmps[gid].tmp_md5[tmp_idx4 + 11] = dst_buf[11];
+        tmps[gid].tmp_md5[tmp_idx4 + 12] = dst_buf[12];
+        tmps[gid].tmp_md5[tmp_idx4 + 13] = dst_buf[13];
+        tmps[gid].tmp_md5[tmp_idx4 + 14] = dst_buf[14];
+        tmps[gid].tmp_md5[tmp_idx4 + 15] = dst_buf[15];
+
+        // sha1
+
+        tmps[gid].tmp_sha1[tmp_idx4 +  0] = hc_swap32_S (dst_buf[ 0]);
+        tmps[gid].tmp_sha1[tmp_idx4 +  1] = hc_swap32_S (dst_buf[ 1]);
+        tmps[gid].tmp_sha1[tmp_idx4 +  2] = hc_swap32_S (dst_buf[ 2]);
+        tmps[gid].tmp_sha1[tmp_idx4 +  3] = hc_swap32_S (dst_buf[ 3]);
+        tmps[gid].tmp_sha1[tmp_idx4 +  4] = hc_swap32_S (dst_buf[ 4]);
+        tmps[gid].tmp_sha1[tmp_idx4 +  5] = hc_swap32_S (dst_buf[ 5]);
+        tmps[gid].tmp_sha1[tmp_idx4 +  6] = hc_swap32_S (dst_buf[ 6]);
+        tmps[gid].tmp_sha1[tmp_idx4 +  7] = hc_swap32_S (dst_buf[ 7]);
+        tmps[gid].tmp_sha1[tmp_idx4 +  8] = hc_swap32_S (dst_buf[ 8]);
+        tmps[gid].tmp_sha1[tmp_idx4 +  9] = hc_swap32_S (dst_buf[ 9]);
+        tmps[gid].tmp_sha1[tmp_idx4 + 10] = hc_swap32_S (dst_buf[10]);
+        tmps[gid].tmp_sha1[tmp_idx4 + 11] = hc_swap32_S (dst_buf[11]);
+        tmps[gid].tmp_sha1[tmp_idx4 + 12] = hc_swap32_S (dst_buf[12]);
+        tmps[gid].tmp_sha1[tmp_idx4 + 13] = hc_swap32_S (dst_buf[13]);
+        tmps[gid].tmp_sha1[tmp_idx4 + 14] = hc_swap32_S (dst_buf[14]);
+        tmps[gid].tmp_sha1[tmp_idx4 + 15] = hc_swap32_S (dst_buf[15]);
+     }
+
+      tmp_idx++;
+    }
+  }
+
+  // hash md5
+
+  tmps[gid].h_md5[0] = MD5M_A;
+  tmps[gid].h_md5[1] = MD5M_B;
+  tmps[gid].h_md5[2] = MD5M_C;
+  tmps[gid].h_md5[3] = MD5M_D;
+
+  // hash sha1
+
+  tmps[gid].h_sha1[0] = SHA1M_A;
+  tmps[gid].h_sha1[1] = SHA1M_B;
+  tmps[gid].h_sha1[2] = SHA1M_C;
+  tmps[gid].h_sha1[3] = SHA1M_D;
+  tmps[gid].h_sha1[4] = SHA1M_E;
+}
+
+KERNEL_FQ void m25000_loop (KERN_ATTR_TMPS_ESALT (hmac_md5_tmp_t, snmpv3_t))
+{
+  /**
+   * base
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  u32 h_md5[4];
+
+  h_md5[0] = tmps[gid].h_md5[0];
+  h_md5[1] = tmps[gid].h_md5[1];
+  h_md5[2] = tmps[gid].h_md5[2];
+  h_md5[3] = tmps[gid].h_md5[3];
+
+  u32 h_sha1[5];
+
+  h_sha1[0] = tmps[gid].h_sha1[0];
+  h_sha1[1] = tmps[gid].h_sha1[1];
+  h_sha1[2] = tmps[gid].h_sha1[2];
+  h_sha1[3] = tmps[gid].h_sha1[3];
+  h_sha1[4] = tmps[gid].h_sha1[4];
+
+  const u32 pw_len = pws[gid].pw_len;
+
+  const int pw_len64 = pw_len * 64;
+
+  if (pw_len <= SNMPV3_MAX_PW_LENGTH_OPT)
+  {
+    u32 tmp_shared[SNMPV3_TMP_ELEMS_OPT];
+
+    // md5
+
+    for (int i = 0; i < pw_len64 / 4; i++)
+    {
+      tmp_shared[i] = tmps[gid].tmp_md5[i];
+    }
+
+    for (int i = 0, j = loop_pos; i < loop_cnt; i += 64, j += 64)
+    {
+      const int idx = (j % pw_len64) / 4; // the optimization trick is to be able to do this
+
+      u32 w0[4];
+      u32 w1[4];
+      u32 w2[4];
+      u32 w3[4];
+
+      w0[0] = tmp_shared[idx +  0];
+      w0[1] = tmp_shared[idx +  1];
+      w0[2] = tmp_shared[idx +  2];
+      w0[3] = tmp_shared[idx +  3];
+      w1[0] = tmp_shared[idx +  4];
+      w1[1] = tmp_shared[idx +  5];
+      w1[2] = tmp_shared[idx +  6];
+      w1[3] = tmp_shared[idx +  7];
+      w2[0] = tmp_shared[idx +  8];
+      w2[1] = tmp_shared[idx +  9];
+      w2[2] = tmp_shared[idx + 10];
+      w2[3] = tmp_shared[idx + 11];
+      w3[0] = tmp_shared[idx + 12];
+      w3[1] = tmp_shared[idx + 13];
+      w3[2] = tmp_shared[idx + 14];
+      w3[3] = tmp_shared[idx + 15];
+
+      md5_transform (w0, w1, w2, w3, h_md5);
+    }
+
+    // sha1
+
+    for (int i = 0; i < pw_len64 / 4; i++)
+    {
+      tmp_shared[i] = tmps[gid].tmp_sha1[i];
+    }
+
+    for (int i = 0, j = loop_pos; i < loop_cnt; i += 64, j += 64)
+    {
+      const int idx = (j % pw_len64) / 4; // the optimization trick is to be able to do this
+
+      u32 w0[4];
+      u32 w1[4];
+      u32 w2[4];
+      u32 w3[4];
+
+      w0[0] = tmp_shared[idx +  0];
+      w0[1] = tmp_shared[idx +  1];
+      w0[2] = tmp_shared[idx +  2];
+      w0[3] = tmp_shared[idx +  3];
+      w1[0] = tmp_shared[idx +  4];
+      w1[1] = tmp_shared[idx +  5];
+      w1[2] = tmp_shared[idx +  6];
+      w1[3] = tmp_shared[idx +  7];
+      w2[0] = tmp_shared[idx +  8];
+      w2[1] = tmp_shared[idx +  9];
+      w2[2] = tmp_shared[idx + 10];
+      w2[3] = tmp_shared[idx + 11];
+      w3[0] = tmp_shared[idx + 12];
+      w3[1] = tmp_shared[idx + 13];
+      w3[2] = tmp_shared[idx + 14];
+      w3[3] = tmp_shared[idx + 15];
+
+      sha1_transform (w0, w1, w2, w3, h_sha1);
+    }
+  }
+  else
+  {
+    for (int i = 0, j = loop_pos; i < loop_cnt; i += 64, j += 64)
+    {
+      const int idx = (j % pw_len64) / 4; // the optimization trick is to be able to do this
+
+      u32 w0[4];
+      u32 w1[4];
+      u32 w2[4];
+      u32 w3[4];
+
+      // md5
+
+      w0[0] = tmps[gid].tmp_md5[idx +  0];
+      w0[1] = tmps[gid].tmp_md5[idx +  1];
+      w0[2] = tmps[gid].tmp_md5[idx +  2];
+      w0[3] = tmps[gid].tmp_md5[idx +  3];
+      w1[0] = tmps[gid].tmp_md5[idx +  4];
+      w1[1] = tmps[gid].tmp_md5[idx +  5];
+      w1[2] = tmps[gid].tmp_md5[idx +  6];
+      w1[3] = tmps[gid].tmp_md5[idx +  7];
+      w2[0] = tmps[gid].tmp_md5[idx +  8];
+      w2[1] = tmps[gid].tmp_md5[idx +  9];
+      w2[2] = tmps[gid].tmp_md5[idx + 10];
+      w2[3] = tmps[gid].tmp_md5[idx + 11];
+      w3[0] = tmps[gid].tmp_md5[idx + 12];
+      w3[1] = tmps[gid].tmp_md5[idx + 13];
+      w3[2] = tmps[gid].tmp_md5[idx + 14];
+      w3[3] = tmps[gid].tmp_md5[idx + 15];
+
+      md5_transform (w0, w1, w2, w3, h_md5);
+
+      // sha1
+
+      w0[0] = tmps[gid].tmp_sha1[idx +  0];
+      w0[1] = tmps[gid].tmp_sha1[idx +  1];
+      w0[2] = tmps[gid].tmp_sha1[idx +  2];
+      w0[3] = tmps[gid].tmp_sha1[idx +  3];
+      w1[0] = tmps[gid].tmp_sha1[idx +  4];
+      w1[1] = tmps[gid].tmp_sha1[idx +  5];
+      w1[2] = tmps[gid].tmp_sha1[idx +  6];
+      w1[3] = tmps[gid].tmp_sha1[idx +  7];
+      w2[0] = tmps[gid].tmp_sha1[idx +  8];
+      w2[1] = tmps[gid].tmp_sha1[idx +  9];
+      w2[2] = tmps[gid].tmp_sha1[idx + 10];
+      w2[3] = tmps[gid].tmp_sha1[idx + 11];
+      w3[0] = tmps[gid].tmp_sha1[idx + 12];
+      w3[1] = tmps[gid].tmp_sha1[idx + 13];
+      w3[2] = tmps[gid].tmp_sha1[idx + 14];
+      w3[3] = tmps[gid].tmp_sha1[idx + 15];
+
+      sha1_transform (w0, w1, w2, w3, h_sha1);
+    }
+  }
+
+  tmps[gid].h_md5[0] = h_md5[0];
+  tmps[gid].h_md5[1] = h_md5[1];
+  tmps[gid].h_md5[2] = h_md5[2];
+  tmps[gid].h_md5[3] = h_md5[3];
+
+  tmps[gid].h_sha1[0] = h_sha1[0];
+  tmps[gid].h_sha1[1] = h_sha1[1];
+  tmps[gid].h_sha1[2] = h_sha1[2];
+  tmps[gid].h_sha1[3] = h_sha1[3];
+  tmps[gid].h_sha1[4] = h_sha1[4];
+}
+
+KERNEL_FQ void m25000_comp (KERN_ATTR_TMPS_ESALT (hmac_md5_tmp_t, snmpv3_t))
+{
+  /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  u32 w0[4];
+  u32 w1[4];
+  u32 w2[4];
+  u32 w3[4];
+
+  // md5
+
+  w0[0] = 0x00000080;
+  w0[1] = 0;
+  w0[2] = 0;
+  w0[3] = 0;
+  w1[0] = 0;
+  w1[1] = 0;
+  w1[2] = 0;
+  w1[3] = 0;
+  w2[0] = 0;
+  w2[1] = 0;
+  w2[2] = 0;
+  w2[3] = 0;
+  w3[0] = 0;
+  w3[1] = 0;
+  w3[2] = 1048576 * 8;
+  w3[3] = 0;
+
+  u32 h_md5[4];
+
+  h_md5[0] = tmps[gid].h_md5[0];
+  h_md5[1] = tmps[gid].h_md5[1];
+  h_md5[2] = tmps[gid].h_md5[2];
+  h_md5[3] = tmps[gid].h_md5[3];
+
+  md5_transform (w0, w1, w2, w3, h_md5);
+
+  // sha1
+
+  w0[0] = 0x80000000;
+  w0[1] = 0;
+  w0[2] = 0;
+  w0[3] = 0;
+  w1[0] = 0;
+  w1[1] = 0;
+  w1[2] = 0;
+  w1[3] = 0;
+  w2[0] = 0;
+  w2[1] = 0;
+  w2[2] = 0;
+  w2[3] = 0;
+  w3[0] = 0;
+  w3[1] = 0;
+  w3[2] = 0;
+  w3[3] = 1048576 * 8;
+
+  u32 h_sha1[5];
+
+  h_sha1[0] = tmps[gid].h_sha1[0];
+  h_sha1[1] = tmps[gid].h_sha1[1];
+  h_sha1[2] = tmps[gid].h_sha1[2];
+  h_sha1[3] = tmps[gid].h_sha1[3];
+  h_sha1[4] = tmps[gid].h_sha1[4];
+
+  sha1_transform (w0, w1, w2, w3, h_sha1);
+
+  md5_ctx_t md5_ctx;
+  sha1_ctx_t sha1_ctx;
+
+  md5_init (&md5_ctx);
+  sha1_init (&sha1_ctx);
+
+  u32 w[16];
+
+  // md5
+
+  w[ 0] = h_md5[0];
+  w[ 1] = h_md5[1];
+  w[ 2] = h_md5[2];
+  w[ 3] = h_md5[3];
+  w[ 4] = 0;
+  w[ 5] = 0;
+  w[ 6] = 0;
+  w[ 7] = 0;
+  w[ 8] = 0;
+  w[ 9] = 0;
+  w[10] = 0;
+  w[11] = 0;
+  w[12] = 0;
+  w[13] = 0;
+  w[14] = 0;
+  w[15] = 0;
+
+  md5_update (&md5_ctx, w, 16);
+
+  // sha1
+
+  w[ 0] = h_sha1[0];
+  w[ 1] = h_sha1[1];
+  w[ 2] = h_sha1[2];
+  w[ 3] = h_sha1[3];
+  w[ 4] = h_sha1[4];
+  w[ 5] = 0;
+  w[ 6] = 0;
+  w[ 7] = 0;
+  w[ 8] = 0;
+  w[ 9] = 0;
+  w[10] = 0;
+  w[11] = 0;
+  w[12] = 0;
+  w[13] = 0;
+  w[14] = 0;
+  w[15] = 0;
+
+  sha1_update (&sha1_ctx, w, 20);
+
+  // engineID
+
+  md5_update_global (&md5_ctx, esalt_bufs[DIGESTS_OFFSET].engineID_buf, esalt_bufs[DIGESTS_OFFSET].engineID_len);
+
+  sha1_update_global_swap (&sha1_ctx, esalt_bufs[DIGESTS_OFFSET].engineID_buf, esalt_bufs[DIGESTS_OFFSET].engineID_len);
+
+  // md5
+
+  w[ 0] = h_md5[0];
+  w[ 1] = h_md5[1];
+  w[ 2] = h_md5[2];
+  w[ 3] = h_md5[3];
+  w[ 4] = 0;
+  w[ 5] = 0;
+  w[ 6] = 0;
+  w[ 7] = 0;
+  w[ 8] = 0;
+  w[ 9] = 0;
+  w[10] = 0;
+  w[11] = 0;
+  w[12] = 0;
+  w[13] = 0;
+  w[14] = 0;
+  w[15] = 0;
+
+  md5_update (&md5_ctx, w, 16);
+
+  // sha1
+
+  w[ 0] = h_sha1[0];
+  w[ 1] = h_sha1[1];
+  w[ 2] = h_sha1[2];
+  w[ 3] = h_sha1[3];
+  w[ 4] = h_sha1[4];
+  w[ 5] = 0;
+  w[ 6] = 0;
+  w[ 7] = 0;
+  w[ 8] = 0;
+  w[ 9] = 0;
+  w[10] = 0;
+  w[11] = 0;
+  w[12] = 0;
+  w[13] = 0;
+  w[14] = 0;
+  w[15] = 0;
+
+  sha1_update (&sha1_ctx, w, 20);
+
+  md5_final (&md5_ctx);
+  sha1_final (&sha1_ctx);
+
+  // md5
+
+  w[ 0] = md5_ctx.h[0];
+  w[ 1] = md5_ctx.h[1];
+  w[ 2] = md5_ctx.h[2];
+  w[ 3] = md5_ctx.h[3];
+  w[ 4] = 0;
+  w[ 5] = 0;
+  w[ 6] = 0;
+  w[ 7] = 0;
+  w[ 8] = 0;
+  w[ 9] = 0;
+  w[10] = 0;
+  w[11] = 0;
+  w[12] = 0;
+  w[13] = 0;
+  w[14] = 0;
+  w[15] = 0;
+
+  md5_hmac_ctx_t md5_hmac_ctx;
+
+  md5_hmac_init (&md5_hmac_ctx, w, 16);
+
+  md5_hmac_update_global (&md5_hmac_ctx, esalt_bufs[DIGESTS_OFFSET].salt_buf, esalt_bufs[DIGESTS_OFFSET].salt_len);
+
+  md5_hmac_final (&md5_hmac_ctx);
+
+  {
+    const u32 r0 = hc_swap32_S (md5_hmac_ctx.opad.h[DGST_R0]);
+    const u32 r1 = hc_swap32_S (md5_hmac_ctx.opad.h[DGST_R1]);
+    const u32 r2 = hc_swap32_S (md5_hmac_ctx.opad.h[DGST_R2]);
+    const u32 r3 = 0;
+
+    #define il_pos 0
+
+    #ifdef KERNEL_STATIC
+    #include COMPARE_M
+    #endif
+  }
+
+  // sha1
+
+  w[ 0] = sha1_ctx.h[0];
+  w[ 1] = sha1_ctx.h[1];
+  w[ 2] = sha1_ctx.h[2];
+  w[ 3] = sha1_ctx.h[3];
+  w[ 4] = sha1_ctx.h[4];
+  w[ 5] = 0;
+  w[ 6] = 0;
+  w[ 7] = 0;
+  w[ 8] = 0;
+  w[ 9] = 0;
+  w[10] = 0;
+  w[11] = 0;
+  w[12] = 0;
+  w[13] = 0;
+  w[14] = 0;
+  w[15] = 0;
+
+  sha1_hmac_ctx_t sha1_hmac_ctx;
+
+  sha1_hmac_init (&sha1_hmac_ctx, w, 20);
+
+  sha1_hmac_update_global_swap (&sha1_hmac_ctx, esalt_bufs[DIGESTS_OFFSET].salt_buf, esalt_bufs[DIGESTS_OFFSET].salt_len);
+
+  sha1_hmac_final (&sha1_hmac_ctx);
+
+  {
+    const u32 r0 = sha1_hmac_ctx.opad.h[DGST_R0];
+    const u32 r1 = sha1_hmac_ctx.opad.h[DGST_R1];
+    const u32 r2 = sha1_hmac_ctx.opad.h[DGST_R2];
+    const u32 r3 = 0;
+
+    #define il_pos 0
+
+    #ifdef KERNEL_STATIC
+    #include COMPARE_M
+    #endif
+  }
+}
diff --git a/OpenCL/m25100-pure.cl b/OpenCL/m25100-pure.cl
new file mode 100644
index 000000000..dba63622e
--- /dev/null
+++ b/OpenCL/m25100-pure.cl
@@ -0,0 +1,355 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+//#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include "inc_vendor.h"
+#include "inc_types.h"
+#include "inc_platform.cl"
+#include "inc_common.cl"
+#include "inc_simd.cl"
+#include "inc_hash_md5.cl"
+#endif
+
+#define COMPARE_S "inc_comp_single.cl"
+#define COMPARE_M "inc_comp_multi.cl"
+
+#define SNMPV3_SALT_MAX             1500
+#define SNMPV3_ENGINEID_MAX         34
+#define SNMPV3_MSG_AUTH_PARAMS_LEN  12
+#define SNMPV3_ROUNDS               1048576
+#define SNMPV3_MAX_PW_LENGTH        64
+
+#define SNMPV3_TMP_ELEMS            4096 // 4096 = (256 (max pw length) * 64) / sizeof (u32)
+#define SNMPV3_HASH_ELEMS           4
+
+#define SNMPV3_MAX_SALT_ELEMS       512 // 512 * 4 = 2048 > 1500, also has to be multiple of 64
+#define SNMPV3_MAX_ENGINE_ELEMS     16  // 16 * 4 = 64 > 32, also has to be multiple of 64
+#define SNMPV3_MAX_PNUM_ELEMS       4   // 4 * 4 = 16 > 9
+
+#define SNMPV3_MAX_PW_LENGTH_OPT    64
+#define SNMPV3_TMP_ELEMS_OPT        ((SNMPV3_MAX_PW_LENGTH_OPT * SNMPV3_MAX_PW_LENGTH) / 4)
+                                    // (64 * 64) / 4 = 1024
+                                    // for pw length > 64 we use global memory reads
+
+typedef struct hmac_md5_tmp
+{
+  u32 tmp[SNMPV3_TMP_ELEMS];
+  u32 h[SNMPV3_HASH_ELEMS];
+
+} hmac_md5_tmp_t;
+
+typedef struct snmpv3
+{
+  u32 salt_buf[SNMPV3_MAX_SALT_ELEMS];
+  u32 salt_len;
+
+  u32 engineID_buf[SNMPV3_MAX_ENGINE_ELEMS];
+  u32 engineID_len;
+
+  u32 packet_number[SNMPV3_MAX_PNUM_ELEMS];
+
+} snmpv3_t;
+
+KERNEL_FQ void m25100_init (KERN_ATTR_TMPS_ESALT (hmac_md5_tmp_t, snmpv3_t))
+{
+  /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  const u32 pw_len = pws[gid].pw_len;
+
+  u32 w[64] = { 0 };
+
+  for (u32 i = 0, idx = 0; i < pw_len; i += 4, idx += 1)
+  {
+    w[idx] = pws[gid].i[idx];
+  }
+
+  u8 *src_ptr = (u8 *) w;
+
+  // password 64 times, also swapped
+
+  u32 dst_buf[16];
+
+  u8 *dst_ptr = (u8 *) dst_buf;
+
+  int tmp_idx = 0;
+
+  for (int i = 0; i < 64; i++)
+  {
+    for (int j = 0; j < pw_len; j++)
+    {
+      const int dst_idx = tmp_idx & 63;
+
+      dst_ptr[dst_idx] = src_ptr[j];
+
+      // write to global memory every time 64 byte are written into cache
+
+      if (dst_idx == 63)
+      {
+        const int tmp_idx4 = (tmp_idx - 63) / 4;
+
+        tmps[gid].tmp[tmp_idx4 +  0] = dst_buf[ 0];
+        tmps[gid].tmp[tmp_idx4 +  1] = dst_buf[ 1];
+        tmps[gid].tmp[tmp_idx4 +  2] = dst_buf[ 2];
+        tmps[gid].tmp[tmp_idx4 +  3] = dst_buf[ 3];
+        tmps[gid].tmp[tmp_idx4 +  4] = dst_buf[ 4];
+        tmps[gid].tmp[tmp_idx4 +  5] = dst_buf[ 5];
+        tmps[gid].tmp[tmp_idx4 +  6] = dst_buf[ 6];
+        tmps[gid].tmp[tmp_idx4 +  7] = dst_buf[ 7];
+        tmps[gid].tmp[tmp_idx4 +  8] = dst_buf[ 8];
+        tmps[gid].tmp[tmp_idx4 +  9] = dst_buf[ 9];
+        tmps[gid].tmp[tmp_idx4 + 10] = dst_buf[10];
+        tmps[gid].tmp[tmp_idx4 + 11] = dst_buf[11];
+        tmps[gid].tmp[tmp_idx4 + 12] = dst_buf[12];
+        tmps[gid].tmp[tmp_idx4 + 13] = dst_buf[13];
+        tmps[gid].tmp[tmp_idx4 + 14] = dst_buf[14];
+        tmps[gid].tmp[tmp_idx4 + 15] = dst_buf[15];
+     }
+
+      tmp_idx++;
+    }
+  }
+
+  // hash
+
+  tmps[gid].h[0] = MD5M_A;
+  tmps[gid].h[1] = MD5M_B;
+  tmps[gid].h[2] = MD5M_C;
+  tmps[gid].h[3] = MD5M_D;
+}
+
+KERNEL_FQ void m25100_loop (KERN_ATTR_TMPS_ESALT (hmac_md5_tmp_t, snmpv3_t))
+{
+  /**
+   * base
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  u32 h[4];
+
+  h[0] = tmps[gid].h[0];
+  h[1] = tmps[gid].h[1];
+  h[2] = tmps[gid].h[2];
+  h[3] = tmps[gid].h[3];
+
+  const u32 pw_len = pws[gid].pw_len;
+
+  const int pw_len64 = pw_len * 64;
+
+  if (pw_len <= SNMPV3_MAX_PW_LENGTH_OPT)
+  {
+    u32 tmp[SNMPV3_TMP_ELEMS_OPT];
+
+    for (int i = 0; i < pw_len64 / 4; i++)
+    {
+      tmp[i] = tmps[gid].tmp[i];
+    }
+
+    for (int i = 0, j = loop_pos; i < loop_cnt; i += 64, j += 64)
+    {
+      const int idx = (j % pw_len64) / 4; // the optimization trick is to be able to do this
+
+      u32 w0[4];
+      u32 w1[4];
+      u32 w2[4];
+      u32 w3[4];
+
+      w0[0] = tmp[idx +  0];
+      w0[1] = tmp[idx +  1];
+      w0[2] = tmp[idx +  2];
+      w0[3] = tmp[idx +  3];
+      w1[0] = tmp[idx +  4];
+      w1[1] = tmp[idx +  5];
+      w1[2] = tmp[idx +  6];
+      w1[3] = tmp[idx +  7];
+      w2[0] = tmp[idx +  8];
+      w2[1] = tmp[idx +  9];
+      w2[2] = tmp[idx + 10];
+      w2[3] = tmp[idx + 11];
+      w3[0] = tmp[idx + 12];
+      w3[1] = tmp[idx + 13];
+      w3[2] = tmp[idx + 14];
+      w3[3] = tmp[idx + 15];
+
+      md5_transform (w0, w1, w2, w3, h);
+    }
+  }
+  else
+  {
+    for (int i = 0, j = loop_pos; i < loop_cnt; i += 64, j += 64)
+    {
+      const int idx = (j % pw_len64) / 4; // the optimization trick is to be able to do this
+
+      u32 w0[4];
+      u32 w1[4];
+      u32 w2[4];
+      u32 w3[4];
+
+      w0[0] = tmps[gid].tmp[idx +  0];
+      w0[1] = tmps[gid].tmp[idx +  1];
+      w0[2] = tmps[gid].tmp[idx +  2];
+      w0[3] = tmps[gid].tmp[idx +  3];
+      w1[0] = tmps[gid].tmp[idx +  4];
+      w1[1] = tmps[gid].tmp[idx +  5];
+      w1[2] = tmps[gid].tmp[idx +  6];
+      w1[3] = tmps[gid].tmp[idx +  7];
+      w2[0] = tmps[gid].tmp[idx +  8];
+      w2[1] = tmps[gid].tmp[idx +  9];
+      w2[2] = tmps[gid].tmp[idx + 10];
+      w2[3] = tmps[gid].tmp[idx + 11];
+      w3[0] = tmps[gid].tmp[idx + 12];
+      w3[1] = tmps[gid].tmp[idx + 13];
+      w3[2] = tmps[gid].tmp[idx + 14];
+      w3[3] = tmps[gid].tmp[idx + 15];
+
+      md5_transform (w0, w1, w2, w3, h);
+    }
+  }
+
+  tmps[gid].h[0] = h[0];
+  tmps[gid].h[1] = h[1];
+  tmps[gid].h[2] = h[2];
+  tmps[gid].h[3] = h[3];
+}
+
+KERNEL_FQ void m25100_comp (KERN_ATTR_TMPS_ESALT (hmac_md5_tmp_t, snmpv3_t))
+{
+  /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  u32 w0[4];
+  u32 w1[4];
+  u32 w2[4];
+  u32 w3[4];
+
+  w0[0] = 0x00000080;
+  w0[1] = 0;
+  w0[2] = 0;
+  w0[3] = 0;
+  w1[0] = 0;
+  w1[1] = 0;
+  w1[2] = 0;
+  w1[3] = 0;
+  w2[0] = 0;
+  w2[1] = 0;
+  w2[2] = 0;
+  w2[3] = 0;
+  w3[0] = 0;
+  w3[1] = 0;
+  w3[2] = 1048576 * 8;
+  w3[3] = 0;
+
+  u32 h[4];
+
+  h[0] = tmps[gid].h[0];
+  h[1] = tmps[gid].h[1];
+  h[2] = tmps[gid].h[2];
+  h[3] = tmps[gid].h[3];
+
+  md5_transform (w0, w1, w2, w3, h);
+
+  md5_ctx_t ctx;
+
+  md5_init (&ctx);
+
+  u32 w[16];
+
+  w[ 0] = h[0];
+  w[ 1] = h[1];
+  w[ 2] = h[2];
+  w[ 3] = h[3];
+  w[ 4] = 0;
+  w[ 5] = 0;
+  w[ 6] = 0;
+  w[ 7] = 0;
+  w[ 8] = 0;
+  w[ 9] = 0;
+  w[10] = 0;
+  w[11] = 0;
+  w[12] = 0;
+  w[13] = 0;
+  w[14] = 0;
+  w[15] = 0;
+
+  md5_update (&ctx, w, 16);
+
+  md5_update_global (&ctx, esalt_bufs[DIGESTS_OFFSET].engineID_buf, esalt_bufs[DIGESTS_OFFSET].engineID_len);
+
+  w[ 0] = h[0];
+  w[ 1] = h[1];
+  w[ 2] = h[2];
+  w[ 3] = h[3];
+  w[ 4] = 0;
+  w[ 5] = 0;
+  w[ 6] = 0;
+  w[ 7] = 0;
+  w[ 8] = 0;
+  w[ 9] = 0;
+  w[10] = 0;
+  w[11] = 0;
+  w[12] = 0;
+  w[13] = 0;
+  w[14] = 0;
+  w[15] = 0;
+
+  md5_update (&ctx, w, 16);
+
+  md5_final (&ctx);
+
+  w[ 0] = ctx.h[0];
+  w[ 1] = ctx.h[1];
+  w[ 2] = ctx.h[2];
+  w[ 3] = ctx.h[3];
+  w[ 4] = 0;
+  w[ 5] = 0;
+  w[ 6] = 0;
+  w[ 7] = 0;
+  w[ 8] = 0;
+  w[ 9] = 0;
+  w[10] = 0;
+  w[11] = 0;
+  w[12] = 0;
+  w[13] = 0;
+  w[14] = 0;
+  w[15] = 0;
+
+  md5_hmac_ctx_t hmac_ctx;
+
+  md5_hmac_init (&hmac_ctx, w, 16);
+
+  md5_hmac_update_global (&hmac_ctx, esalt_bufs[DIGESTS_OFFSET].salt_buf, esalt_bufs[DIGESTS_OFFSET].salt_len);
+
+  md5_hmac_final (&hmac_ctx);
+
+  const u32 r0 = hmac_ctx.opad.h[DGST_R0];
+  const u32 r1 = hmac_ctx.opad.h[DGST_R1];
+  const u32 r2 = hmac_ctx.opad.h[DGST_R2];
+  const u32 r3 = 0;
+
+  #define il_pos 0
+
+  #ifdef KERNEL_STATIC
+  #include COMPARE_M
+  #endif
+}
diff --git a/OpenCL/m25200-pure.cl b/OpenCL/m25200-pure.cl
new file mode 100644
index 000000000..3acb85572
--- /dev/null
+++ b/OpenCL/m25200-pure.cl
@@ -0,0 +1,360 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+//#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include "inc_vendor.h"
+#include "inc_types.h"
+#include "inc_platform.cl"
+#include "inc_common.cl"
+#include "inc_simd.cl"
+#include "inc_hash_sha1.cl"
+#endif
+
+#define COMPARE_S "inc_comp_single.cl"
+#define COMPARE_M "inc_comp_multi.cl"
+
+#define SNMPV3_SALT_MAX             1500
+#define SNMPV3_ENGINEID_MAX         34
+#define SNMPV3_MSG_AUTH_PARAMS_LEN  12
+#define SNMPV3_ROUNDS               1048576
+#define SNMPV3_MAX_PW_LENGTH        64
+
+#define SNMPV3_TMP_ELEMS            4096 // 4096 = (256 (max pw length) * 64) / sizeof (u32)
+#define SNMPV3_HASH_ELEMS           8    // 8 = aligned 5
+
+#define SNMPV3_MAX_SALT_ELEMS       512 // 512 * 4 = 2048 > 1500, also has to be multiple of 64
+#define SNMPV3_MAX_ENGINE_ELEMS     16  // 16 * 4 = 64 > 32, also has to be multiple of 64
+#define SNMPV3_MAX_PNUM_ELEMS       4   // 4 * 4 = 16 > 9
+
+#define SNMPV3_MAX_PW_LENGTH_OPT    64
+#define SNMPV3_TMP_ELEMS_OPT        ((SNMPV3_MAX_PW_LENGTH_OPT * SNMPV3_MAX_PW_LENGTH) / 4)
+                                    // (64 * 64) / 4 = 1024
+                                    // for pw length > 64 we use global memory reads
+
+typedef struct hmac_sha1_tmp
+{
+  u32 tmp[SNMPV3_TMP_ELEMS];
+  u32 h[SNMPV3_HASH_ELEMS];
+
+} hmac_sha1_tmp_t;
+
+typedef struct snmpv3
+{
+  u32 salt_buf[SNMPV3_MAX_SALT_ELEMS];
+  u32 salt_len;
+
+  u32 engineID_buf[SNMPV3_MAX_ENGINE_ELEMS];
+  u32 engineID_len;
+
+  u32 packet_number[SNMPV3_MAX_PNUM_ELEMS];
+
+} snmpv3_t;
+
+KERNEL_FQ void m25200_init (KERN_ATTR_TMPS_ESALT (hmac_sha1_tmp_t, snmpv3_t))
+{
+  /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  const u32 pw_len = pws[gid].pw_len;
+
+  u32 w[64] = { 0 };
+
+  for (u32 i = 0, idx = 0; i < pw_len; i += 4, idx += 1)
+  {
+    w[idx] = pws[gid].i[idx];
+  }
+
+  u8 *src_ptr = (u8 *) w;
+
+  // password 64 times, also swapped
+
+  u32 dst_buf[16];
+
+  u8 *dst_ptr = (u8 *) dst_buf;
+
+  int tmp_idx = 0;
+
+  for (int i = 0; i < 64; i++)
+  {
+    for (int j = 0; j < pw_len; j++)
+    {
+      const int dst_idx = tmp_idx & 63;
+
+      dst_ptr[dst_idx] = src_ptr[j];
+
+      // write to global memory every time 64 byte are written into cache
+
+      if (dst_idx == 63)
+      {
+        const int tmp_idx4 = (tmp_idx - 63) / 4;
+
+        tmps[gid].tmp[tmp_idx4 +  0] = hc_swap32_S (dst_buf[ 0]);
+        tmps[gid].tmp[tmp_idx4 +  1] = hc_swap32_S (dst_buf[ 1]);
+        tmps[gid].tmp[tmp_idx4 +  2] = hc_swap32_S (dst_buf[ 2]);
+        tmps[gid].tmp[tmp_idx4 +  3] = hc_swap32_S (dst_buf[ 3]);
+        tmps[gid].tmp[tmp_idx4 +  4] = hc_swap32_S (dst_buf[ 4]);
+        tmps[gid].tmp[tmp_idx4 +  5] = hc_swap32_S (dst_buf[ 5]);
+        tmps[gid].tmp[tmp_idx4 +  6] = hc_swap32_S (dst_buf[ 6]);
+        tmps[gid].tmp[tmp_idx4 +  7] = hc_swap32_S (dst_buf[ 7]);
+        tmps[gid].tmp[tmp_idx4 +  8] = hc_swap32_S (dst_buf[ 8]);
+        tmps[gid].tmp[tmp_idx4 +  9] = hc_swap32_S (dst_buf[ 9]);
+        tmps[gid].tmp[tmp_idx4 + 10] = hc_swap32_S (dst_buf[10]);
+        tmps[gid].tmp[tmp_idx4 + 11] = hc_swap32_S (dst_buf[11]);
+        tmps[gid].tmp[tmp_idx4 + 12] = hc_swap32_S (dst_buf[12]);
+        tmps[gid].tmp[tmp_idx4 + 13] = hc_swap32_S (dst_buf[13]);
+        tmps[gid].tmp[tmp_idx4 + 14] = hc_swap32_S (dst_buf[14]);
+        tmps[gid].tmp[tmp_idx4 + 15] = hc_swap32_S (dst_buf[15]);
+      }
+
+      tmp_idx++;
+    }
+  }
+
+  // hash
+
+  tmps[gid].h[0] = SHA1M_A;
+  tmps[gid].h[1] = SHA1M_B;
+  tmps[gid].h[2] = SHA1M_C;
+  tmps[gid].h[3] = SHA1M_D;
+  tmps[gid].h[4] = SHA1M_E;
+}
+
+KERNEL_FQ void m25200_loop (KERN_ATTR_TMPS_ESALT (hmac_sha1_tmp_t, snmpv3_t))
+{
+  /**
+   * base
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  u32 h[5];
+
+  h[0] = tmps[gid].h[0];
+  h[1] = tmps[gid].h[1];
+  h[2] = tmps[gid].h[2];
+  h[3] = tmps[gid].h[3];
+  h[4] = tmps[gid].h[4];
+
+  const u32 pw_len = pws[gid].pw_len;
+
+  const int pw_len64 = pw_len * 64;
+
+  if (pw_len <= SNMPV3_MAX_PW_LENGTH_OPT)
+  {
+    u32 tmp[SNMPV3_TMP_ELEMS_OPT];
+
+    for (int i = 0; i < pw_len64 / 4; i++)
+    {
+      tmp[i] = tmps[gid].tmp[i];
+    }
+
+    for (int i = 0, j = loop_pos; i < loop_cnt; i += 64, j += 64)
+    {
+      const int idx = (j % pw_len64) / 4; // the optimization trick is to be able to do this
+
+      u32 w0[4];
+      u32 w1[4];
+      u32 w2[4];
+      u32 w3[4];
+
+      w0[0] = tmp[idx +  0];
+      w0[1] = tmp[idx +  1];
+      w0[2] = tmp[idx +  2];
+      w0[3] = tmp[idx +  3];
+      w1[0] = tmp[idx +  4];
+      w1[1] = tmp[idx +  5];
+      w1[2] = tmp[idx +  6];
+      w1[3] = tmp[idx +  7];
+      w2[0] = tmp[idx +  8];
+      w2[1] = tmp[idx +  9];
+      w2[2] = tmp[idx + 10];
+      w2[3] = tmp[idx + 11];
+      w3[0] = tmp[idx + 12];
+      w3[1] = tmp[idx + 13];
+      w3[2] = tmp[idx + 14];
+      w3[3] = tmp[idx + 15];
+
+      sha1_transform (w0, w1, w2, w3, h);
+    }
+  }
+  else
+  {
+    for (int i = 0, j = loop_pos; i < loop_cnt; i += 64, j += 64)
+    {
+      const int idx = (j % pw_len64) / 4; // the optimization trick is to be able to do this
+
+      u32 w0[4];
+      u32 w1[4];
+      u32 w2[4];
+      u32 w3[4];
+
+      w0[0] = tmps[gid].tmp[idx +  0];
+      w0[1] = tmps[gid].tmp[idx +  1];
+      w0[2] = tmps[gid].tmp[idx +  2];
+      w0[3] = tmps[gid].tmp[idx +  3];
+      w1[0] = tmps[gid].tmp[idx +  4];
+      w1[1] = tmps[gid].tmp[idx +  5];
+      w1[2] = tmps[gid].tmp[idx +  6];
+      w1[3] = tmps[gid].tmp[idx +  7];
+      w2[0] = tmps[gid].tmp[idx +  8];
+      w2[1] = tmps[gid].tmp[idx +  9];
+      w2[2] = tmps[gid].tmp[idx + 10];
+      w2[3] = tmps[gid].tmp[idx + 11];
+      w3[0] = tmps[gid].tmp[idx + 12];
+      w3[1] = tmps[gid].tmp[idx + 13];
+      w3[2] = tmps[gid].tmp[idx + 14];
+      w3[3] = tmps[gid].tmp[idx + 15];
+
+      sha1_transform (w0, w1, w2, w3, h);
+    }
+  }
+
+  tmps[gid].h[0] = h[0];
+  tmps[gid].h[1] = h[1];
+  tmps[gid].h[2] = h[2];
+  tmps[gid].h[3] = h[3];
+  tmps[gid].h[4] = h[4];
+}
+
+KERNEL_FQ void m25200_comp (KERN_ATTR_TMPS_ESALT (hmac_sha1_tmp_t, snmpv3_t))
+{
+  /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  u32 w0[4];
+  u32 w1[4];
+  u32 w2[4];
+  u32 w3[4];
+
+  w0[0] = 0x80000000;
+  w0[1] = 0;
+  w0[2] = 0;
+  w0[3] = 0;
+  w1[0] = 0;
+  w1[1] = 0;
+  w1[2] = 0;
+  w1[3] = 0;
+  w2[0] = 0;
+  w2[1] = 0;
+  w2[2] = 0;
+  w2[3] = 0;
+  w3[0] = 0;
+  w3[1] = 0;
+  w3[2] = 0;
+  w3[3] = 1048576 * 8;
+
+  u32 h[5];
+
+  h[0] = tmps[gid].h[0];
+  h[1] = tmps[gid].h[1];
+  h[2] = tmps[gid].h[2];
+  h[3] = tmps[gid].h[3];
+  h[4] = tmps[gid].h[4];
+
+  sha1_transform (w0, w1, w2, w3, h);
+
+  sha1_ctx_t ctx;
+
+  sha1_init (&ctx);
+
+  u32 w[16];
+
+  w[ 0] = h[0];
+  w[ 1] = h[1];
+  w[ 2] = h[2];
+  w[ 3] = h[3];
+  w[ 4] = h[4];
+  w[ 5] = 0;
+  w[ 6] = 0;
+  w[ 7] = 0;
+  w[ 8] = 0;
+  w[ 9] = 0;
+  w[10] = 0;
+  w[11] = 0;
+  w[12] = 0;
+  w[13] = 0;
+  w[14] = 0;
+  w[15] = 0;
+
+  sha1_update (&ctx, w, 20);
+
+  sha1_update_global_swap (&ctx, esalt_bufs[DIGESTS_OFFSET].engineID_buf, esalt_bufs[DIGESTS_OFFSET].engineID_len);
+
+  w[ 0] = h[0];
+  w[ 1] = h[1];
+  w[ 2] = h[2];
+  w[ 3] = h[3];
+  w[ 4] = h[4];
+  w[ 5] = 0;
+  w[ 6] = 0;
+  w[ 7] = 0;
+  w[ 8] = 0;
+  w[ 9] = 0;
+  w[10] = 0;
+  w[11] = 0;
+  w[12] = 0;
+  w[13] = 0;
+  w[14] = 0;
+  w[15] = 0;
+
+  sha1_update (&ctx, w, 20);
+
+  sha1_final (&ctx);
+
+  w[ 0] = ctx.h[0];
+  w[ 1] = ctx.h[1];
+  w[ 2] = ctx.h[2];
+  w[ 3] = ctx.h[3];
+  w[ 4] = ctx.h[4];
+  w[ 5] = 0;
+  w[ 6] = 0;
+  w[ 7] = 0;
+  w[ 8] = 0;
+  w[ 9] = 0;
+  w[10] = 0;
+  w[11] = 0;
+  w[12] = 0;
+  w[13] = 0;
+  w[14] = 0;
+  w[15] = 0;
+
+  sha1_hmac_ctx_t hmac_ctx;
+
+  sha1_hmac_init (&hmac_ctx, w, 20);
+
+  sha1_hmac_update_global_swap (&hmac_ctx, esalt_bufs[DIGESTS_OFFSET].salt_buf, esalt_bufs[DIGESTS_OFFSET].salt_len);
+
+  sha1_hmac_final (&hmac_ctx);
+
+  const u32 r0 = hmac_ctx.opad.h[DGST_R0];
+  const u32 r1 = hmac_ctx.opad.h[DGST_R1];
+  const u32 r2 = hmac_ctx.opad.h[DGST_R2];
+  const u32 r3 = 0;
+
+  #define il_pos 0
+
+  #ifdef KERNEL_STATIC
+  #include COMPARE_M
+  #endif
+}
+
diff --git a/OpenCL/m26700-pure.cl b/OpenCL/m26700-pure.cl
new file mode 100644
index 000000000..5c7718c1f
--- /dev/null
+++ b/OpenCL/m26700-pure.cl
@@ -0,0 +1,371 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+//#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include "inc_vendor.h"
+#include "inc_types.h"
+#include "inc_platform.cl"
+#include "inc_common.cl"
+#include "inc_simd.cl"
+#include "inc_hash_sha224.cl"
+#endif
+
+#define COMPARE_S "inc_comp_single.cl"
+#define COMPARE_M "inc_comp_multi.cl"
+
+#define SNMPV3_SALT_MAX             1500
+#define SNMPV3_ENGINEID_MAX         34
+#define SNMPV3_MSG_AUTH_PARAMS_MAX  16
+#define SNMPV3_ROUNDS               1048576
+#define SNMPV3_MAX_PW_LENGTH        64
+
+#define SNMPV3_TMP_ELEMS            4096 // 4096 = (256 (max pw length) * 64) / sizeof (u32)
+#define SNMPV3_HASH_ELEMS           8
+
+#define SNMPV3_MAX_SALT_ELEMS       512 // 512 * 4 = 2048 > 1500, also has to be multiple of 64
+#define SNMPV3_MAX_ENGINE_ELEMS     16  // 16 * 4 = 64 > 32, also has to be multiple of 64
+#define SNMPV3_MAX_PNUM_ELEMS       4   // 4 * 4 = 16 > 9
+
+#define SNMPV3_MAX_PW_LENGTH_OPT    64
+#define SNMPV3_TMP_ELEMS_OPT        ((SNMPV3_MAX_PW_LENGTH_OPT * SNMPV3_MAX_PW_LENGTH) / 4)
+                                    // (64 * 64) / 4 = 1024
+                                    // for pw length > 64 we use global memory reads
+
+typedef struct hmac_sha224_tmp
+{
+  u32 tmp[SNMPV3_TMP_ELEMS];
+  u32 h[SNMPV3_HASH_ELEMS];
+
+} hmac_sha224_tmp_t;
+
+typedef struct snmpv3
+{
+  u32 salt_buf[SNMPV3_MAX_SALT_ELEMS];
+  u32 salt_len;
+
+  u32 engineID_buf[SNMPV3_MAX_ENGINE_ELEMS];
+  u32 engineID_len;
+
+  u32 packet_number[SNMPV3_MAX_PNUM_ELEMS];
+
+} snmpv3_t;
+
+KERNEL_FQ void m26700_init (KERN_ATTR_TMPS_ESALT (hmac_sha224_tmp_t, snmpv3_t))
+{
+  /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  const u32 pw_len = pws[gid].pw_len;
+
+  u32 w[64] = { 0 };
+
+  for (u32 i = 0, idx = 0; i < pw_len; i += 4, idx += 1)
+  {
+    w[idx] = pws[gid].i[idx];
+  }
+
+  u8 *src_ptr = (u8 *) w;
+
+  // password 64 times, also swapped
+
+  u32 dst_buf[16];
+
+  u8 *dst_ptr = (u8 *) dst_buf;
+
+  int tmp_idx = 0;
+
+  for (int i = 0; i < 64; i++)
+  {
+    for (int j = 0; j < pw_len; j++)
+    {
+      const int dst_idx = tmp_idx & 63;
+
+      dst_ptr[dst_idx] = src_ptr[j];
+
+      // write to global memory every time 64 byte are written into cache
+
+      if (dst_idx == 63)
+      {
+        const int tmp_idx4 = (tmp_idx - 63) / 4;
+
+        tmps[gid].tmp[tmp_idx4 +  0] = hc_swap32_S (dst_buf[ 0]);
+        tmps[gid].tmp[tmp_idx4 +  1] = hc_swap32_S (dst_buf[ 1]);
+        tmps[gid].tmp[tmp_idx4 +  2] = hc_swap32_S (dst_buf[ 2]);
+        tmps[gid].tmp[tmp_idx4 +  3] = hc_swap32_S (dst_buf[ 3]);
+        tmps[gid].tmp[tmp_idx4 +  4] = hc_swap32_S (dst_buf[ 4]);
+        tmps[gid].tmp[tmp_idx4 +  5] = hc_swap32_S (dst_buf[ 5]);
+        tmps[gid].tmp[tmp_idx4 +  6] = hc_swap32_S (dst_buf[ 6]);
+        tmps[gid].tmp[tmp_idx4 +  7] = hc_swap32_S (dst_buf[ 7]);
+        tmps[gid].tmp[tmp_idx4 +  8] = hc_swap32_S (dst_buf[ 8]);
+        tmps[gid].tmp[tmp_idx4 +  9] = hc_swap32_S (dst_buf[ 9]);
+        tmps[gid].tmp[tmp_idx4 + 10] = hc_swap32_S (dst_buf[10]);
+        tmps[gid].tmp[tmp_idx4 + 11] = hc_swap32_S (dst_buf[11]);
+        tmps[gid].tmp[tmp_idx4 + 12] = hc_swap32_S (dst_buf[12]);
+        tmps[gid].tmp[tmp_idx4 + 13] = hc_swap32_S (dst_buf[13]);
+        tmps[gid].tmp[tmp_idx4 + 14] = hc_swap32_S (dst_buf[14]);
+        tmps[gid].tmp[tmp_idx4 + 15] = hc_swap32_S (dst_buf[15]);
+      }
+
+      tmp_idx++;
+    }
+  }
+
+  // hash
+
+  tmps[gid].h[0] = SHA224M_A;
+  tmps[gid].h[1] = SHA224M_B;
+  tmps[gid].h[2] = SHA224M_C;
+  tmps[gid].h[3] = SHA224M_D;
+  tmps[gid].h[4] = SHA224M_E;
+  tmps[gid].h[5] = SHA224M_F;
+  tmps[gid].h[6] = SHA224M_G;
+  tmps[gid].h[7] = SHA224M_H;
+}
+
+KERNEL_FQ void m26700_loop (KERN_ATTR_TMPS_ESALT (hmac_sha224_tmp_t, snmpv3_t))
+{
+  /**
+   * base
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  u32 h[8];
+
+  h[0] = tmps[gid].h[0];
+  h[1] = tmps[gid].h[1];
+  h[2] = tmps[gid].h[2];
+  h[3] = tmps[gid].h[3];
+  h[4] = tmps[gid].h[4];
+  h[5] = tmps[gid].h[5];
+  h[6] = tmps[gid].h[6];
+  h[7] = tmps[gid].h[7];
+
+  const u32 pw_len = pws[gid].pw_len;
+
+  const int pw_len64 = pw_len * 64;
+
+  if (pw_len <= SNMPV3_MAX_PW_LENGTH_OPT)
+  {
+    u32 tmp[SNMPV3_TMP_ELEMS_OPT];
+
+    for (int i = 0; i < pw_len64 / 4; i++)
+    {
+      tmp[i] = tmps[gid].tmp[i];
+    }
+
+    for (int i = 0, j = loop_pos; i < loop_cnt; i += 64, j += 64)
+    {
+      const int idx = (j % pw_len64) / 4; // the optimization trick is to be able to do this
+
+      u32 w0[4];
+      u32 w1[4];
+      u32 w2[4];
+      u32 w3[4];
+
+      w0[0] = tmp[idx +  0];
+      w0[1] = tmp[idx +  1];
+      w0[2] = tmp[idx +  2];
+      w0[3] = tmp[idx +  3];
+      w1[0] = tmp[idx +  4];
+      w1[1] = tmp[idx +  5];
+      w1[2] = tmp[idx +  6];
+      w1[3] = tmp[idx +  7];
+      w2[0] = tmp[idx +  8];
+      w2[1] = tmp[idx +  9];
+      w2[2] = tmp[idx + 10];
+      w2[3] = tmp[idx + 11];
+      w3[0] = tmp[idx + 12];
+      w3[1] = tmp[idx + 13];
+      w3[2] = tmp[idx + 14];
+      w3[3] = tmp[idx + 15];
+
+      sha224_transform (w0, w1, w2, w3, h);
+    }
+  }
+  else
+  {
+    for (int i = 0, j = loop_pos; i < loop_cnt; i += 64, j += 64)
+    {
+      const int idx = (j % pw_len64) / 4; // the optimization trick is to be able to do this
+
+      u32 w0[4];
+      u32 w1[4];
+      u32 w2[4];
+      u32 w3[4];
+
+      w0[0] = tmps[gid].tmp[idx +  0];
+      w0[1] = tmps[gid].tmp[idx +  1];
+      w0[2] = tmps[gid].tmp[idx +  2];
+      w0[3] = tmps[gid].tmp[idx +  3];
+      w1[0] = tmps[gid].tmp[idx +  4];
+      w1[1] = tmps[gid].tmp[idx +  5];
+      w1[2] = tmps[gid].tmp[idx +  6];
+      w1[3] = tmps[gid].tmp[idx +  7];
+      w2[0] = tmps[gid].tmp[idx +  8];
+      w2[1] = tmps[gid].tmp[idx +  9];
+      w2[2] = tmps[gid].tmp[idx + 10];
+      w2[3] = tmps[gid].tmp[idx + 11];
+      w3[0] = tmps[gid].tmp[idx + 12];
+      w3[1] = tmps[gid].tmp[idx + 13];
+      w3[2] = tmps[gid].tmp[idx + 14];
+      w3[3] = tmps[gid].tmp[idx + 15];
+
+      sha224_transform (w0, w1, w2, w3, h);
+    }
+  }
+
+  tmps[gid].h[0] = h[0];
+  tmps[gid].h[1] = h[1];
+  tmps[gid].h[2] = h[2];
+  tmps[gid].h[3] = h[3];
+  tmps[gid].h[4] = h[4];
+  tmps[gid].h[5] = h[5];
+  tmps[gid].h[6] = h[6];
+  tmps[gid].h[7] = h[7];
+}
+
+KERNEL_FQ void m26700_comp (KERN_ATTR_TMPS_ESALT (hmac_sha224_tmp_t, snmpv3_t))
+{
+  /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  u32 w0[4];
+  u32 w1[4];
+  u32 w2[4];
+  u32 w3[4];
+
+  w0[0] = 0x80000000;
+  w0[1] = 0;
+  w0[2] = 0;
+  w0[3] = 0;
+  w1[0] = 0;
+  w1[1] = 0;
+  w1[2] = 0;
+  w1[3] = 0;
+  w2[0] = 0;
+  w2[1] = 0;
+  w2[2] = 0;
+  w2[3] = 0;
+  w3[0] = 0;
+  w3[1] = 0;
+  w3[2] = 0;
+  w3[3] = 1048576 * 8;
+
+  u32 h[8];
+
+  h[0] = tmps[gid].h[0];
+  h[1] = tmps[gid].h[1];
+  h[2] = tmps[gid].h[2];
+  h[3] = tmps[gid].h[3];
+  h[4] = tmps[gid].h[4];
+  h[5] = tmps[gid].h[5];
+  h[6] = tmps[gid].h[6];
+  h[7] = tmps[gid].h[7];
+
+  sha224_transform (w0, w1, w2, w3, h);
+
+  sha224_ctx_t ctx;
+
+  sha224_init (&ctx);
+
+  u32 w[16];
+
+  w[ 0] = h[0];
+  w[ 1] = h[1];
+  w[ 2] = h[2];
+  w[ 3] = h[3];
+  w[ 4] = h[4];
+  w[ 5] = h[5];
+  w[ 6] = h[6];
+  w[ 7] = 0;
+  w[ 8] = 0;
+  w[ 9] = 0;
+  w[10] = 0;
+  w[11] = 0;
+  w[12] = 0;
+  w[13] = 0;
+  w[14] = 0;
+  w[15] = 0;
+
+  sha224_update (&ctx, w, 28);
+
+  sha224_update_global_swap (&ctx, esalt_bufs[DIGESTS_OFFSET].engineID_buf, esalt_bufs[DIGESTS_OFFSET].engineID_len);
+
+  w[ 0] = h[0];
+  w[ 1] = h[1];
+  w[ 2] = h[2];
+  w[ 3] = h[3];
+  w[ 4] = h[4];
+  w[ 5] = h[5];
+  w[ 6] = h[6];
+  w[ 7] = 0;
+  w[ 8] = 0;
+  w[ 9] = 0;
+  w[10] = 0;
+  w[11] = 0;
+  w[12] = 0;
+  w[13] = 0;
+  w[14] = 0;
+  w[15] = 0;
+
+  sha224_update (&ctx, w, 28);
+
+  sha224_final (&ctx);
+
+  w[ 0] = ctx.h[0];
+  w[ 1] = ctx.h[1];
+  w[ 2] = ctx.h[2];
+  w[ 3] = ctx.h[3];
+  w[ 4] = ctx.h[4];
+  w[ 5] = ctx.h[5];
+  w[ 6] = ctx.h[6];
+  w[ 7] = 0;
+  w[ 8] = 0;
+  w[ 9] = 0;
+  w[10] = 0;
+  w[11] = 0;
+  w[12] = 0;
+  w[13] = 0;
+  w[14] = 0;
+  w[15] = 0;
+
+  sha224_hmac_ctx_t hmac_ctx;
+
+  sha224_hmac_init (&hmac_ctx, w, 28);
+
+  sha224_hmac_update_global_swap (&hmac_ctx, esalt_bufs[DIGESTS_OFFSET].salt_buf, esalt_bufs[DIGESTS_OFFSET].salt_len);
+
+  sha224_hmac_final (&hmac_ctx);
+
+  const u32 r0 = hmac_ctx.opad.h[DGST_R0];
+  const u32 r1 = hmac_ctx.opad.h[DGST_R1];
+  const u32 r2 = hmac_ctx.opad.h[DGST_R2];
+  const u32 r3 = hmac_ctx.opad.h[DGST_R3];
+
+  #define il_pos 0
+
+  #ifdef KERNEL_STATIC
+  #include COMPARE_M
+  #endif
+}
diff --git a/OpenCL/m26800-pure.cl b/OpenCL/m26800-pure.cl
new file mode 100644
index 000000000..679233a38
--- /dev/null
+++ b/OpenCL/m26800-pure.cl
@@ -0,0 +1,371 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+//#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include "inc_vendor.h"
+#include "inc_types.h"
+#include "inc_platform.cl"
+#include "inc_common.cl"
+#include "inc_simd.cl"
+#include "inc_hash_sha256.cl"
+#endif
+
+#define COMPARE_S "inc_comp_single.cl"
+#define COMPARE_M "inc_comp_multi.cl"
+
+#define SNMPV3_SALT_MAX             1500
+#define SNMPV3_ENGINEID_MAX         34
+#define SNMPV3_MSG_AUTH_PARAMS_MAX  24
+#define SNMPV3_ROUNDS               1048576
+#define SNMPV3_MAX_PW_LENGTH        64
+
+#define SNMPV3_TMP_ELEMS            4096 // 4096 = (256 (max pw length) * 64) / sizeof (u32)
+#define SNMPV3_HASH_ELEMS           8
+
+#define SNMPV3_MAX_SALT_ELEMS       512 // 512 * 4 = 2048 > 1500, also has to be multiple of 64
+#define SNMPV3_MAX_ENGINE_ELEMS     16  // 16 * 4 = 64 > 32, also has to be multiple of 64
+#define SNMPV3_MAX_PNUM_ELEMS       4   // 4 * 4 = 16 > 9
+
+#define SNMPV3_MAX_PW_LENGTH_OPT    64
+#define SNMPV3_TMP_ELEMS_OPT        ((SNMPV3_MAX_PW_LENGTH_OPT * SNMPV3_MAX_PW_LENGTH) / 4)
+                                    // (64 * 64) / 4 = 1024
+                                    // for pw length > 64 we use global memory reads
+
+typedef struct hmac_sha256_tmp
+{
+  u32 tmp[SNMPV3_TMP_ELEMS];
+  u32 h[SNMPV3_HASH_ELEMS];
+
+} hmac_sha256_tmp_t;
+
+typedef struct snmpv3
+{
+  u32 salt_buf[SNMPV3_MAX_SALT_ELEMS];
+  u32 salt_len;
+
+  u32 engineID_buf[SNMPV3_MAX_ENGINE_ELEMS];
+  u32 engineID_len;
+
+  u32 packet_number[SNMPV3_MAX_PNUM_ELEMS];
+
+} snmpv3_t;
+
+KERNEL_FQ void m26800_init (KERN_ATTR_TMPS_ESALT (hmac_sha256_tmp_t, snmpv3_t))
+{
+  /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  const u32 pw_len = pws[gid].pw_len;
+
+  u32 w[64] = { 0 };
+
+  for (u32 i = 0, idx = 0; i < pw_len; i += 4, idx += 1)
+  {
+    w[idx] = pws[gid].i[idx];
+  }
+
+  u8 *src_ptr = (u8 *) w;
+
+  // password 64 times, also swapped
+
+  u32 dst_buf[16];
+
+  u8 *dst_ptr = (u8 *) dst_buf;
+
+  int tmp_idx = 0;
+
+  for (int i = 0; i < 64; i++)
+  {
+    for (int j = 0; j < pw_len; j++)
+    {
+      const int dst_idx = tmp_idx & 63;
+
+      dst_ptr[dst_idx] = src_ptr[j];
+
+      // write to global memory every time 64 byte are written into cache
+
+      if (dst_idx == 63)
+      {
+        const int tmp_idx4 = (tmp_idx - 63) / 4;
+
+        tmps[gid].tmp[tmp_idx4 +  0] = hc_swap32_S (dst_buf[ 0]);
+        tmps[gid].tmp[tmp_idx4 +  1] = hc_swap32_S (dst_buf[ 1]);
+        tmps[gid].tmp[tmp_idx4 +  2] = hc_swap32_S (dst_buf[ 2]);
+        tmps[gid].tmp[tmp_idx4 +  3] = hc_swap32_S (dst_buf[ 3]);
+        tmps[gid].tmp[tmp_idx4 +  4] = hc_swap32_S (dst_buf[ 4]);
+        tmps[gid].tmp[tmp_idx4 +  5] = hc_swap32_S (dst_buf[ 5]);
+        tmps[gid].tmp[tmp_idx4 +  6] = hc_swap32_S (dst_buf[ 6]);
+        tmps[gid].tmp[tmp_idx4 +  7] = hc_swap32_S (dst_buf[ 7]);
+        tmps[gid].tmp[tmp_idx4 +  8] = hc_swap32_S (dst_buf[ 8]);
+        tmps[gid].tmp[tmp_idx4 +  9] = hc_swap32_S (dst_buf[ 9]);
+        tmps[gid].tmp[tmp_idx4 + 10] = hc_swap32_S (dst_buf[10]);
+        tmps[gid].tmp[tmp_idx4 + 11] = hc_swap32_S (dst_buf[11]);
+        tmps[gid].tmp[tmp_idx4 + 12] = hc_swap32_S (dst_buf[12]);
+        tmps[gid].tmp[tmp_idx4 + 13] = hc_swap32_S (dst_buf[13]);
+        tmps[gid].tmp[tmp_idx4 + 14] = hc_swap32_S (dst_buf[14]);
+        tmps[gid].tmp[tmp_idx4 + 15] = hc_swap32_S (dst_buf[15]);
+      }
+
+      tmp_idx++;
+    }
+  }
+
+  // hash
+
+  tmps[gid].h[0] = SHA256M_A;
+  tmps[gid].h[1] = SHA256M_B;
+  tmps[gid].h[2] = SHA256M_C;
+  tmps[gid].h[3] = SHA256M_D;
+  tmps[gid].h[4] = SHA256M_E;
+  tmps[gid].h[5] = SHA256M_F;
+  tmps[gid].h[6] = SHA256M_G;
+  tmps[gid].h[7] = SHA256M_H;
+}
+
+KERNEL_FQ void m26800_loop (KERN_ATTR_TMPS_ESALT (hmac_sha256_tmp_t, snmpv3_t))
+{
+  /**
+   * base
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  u32 h[8];
+
+  h[0] = tmps[gid].h[0];
+  h[1] = tmps[gid].h[1];
+  h[2] = tmps[gid].h[2];
+  h[3] = tmps[gid].h[3];
+  h[4] = tmps[gid].h[4];
+  h[5] = tmps[gid].h[5];
+  h[6] = tmps[gid].h[6];
+  h[7] = tmps[gid].h[7];
+
+  const u32 pw_len = pws[gid].pw_len;
+
+  const int pw_len64 = pw_len * 64;
+
+  if (pw_len <= SNMPV3_MAX_PW_LENGTH_OPT)
+  {
+    u32 tmp[SNMPV3_TMP_ELEMS_OPT];
+
+    for (int i = 0; i < pw_len64 / 4; i++)
+    {
+      tmp[i] = tmps[gid].tmp[i];
+    }
+
+    for (int i = 0, j = loop_pos; i < loop_cnt; i += 64, j += 64)
+    {
+      const int idx = (j % pw_len64) / 4; // the optimization trick is to be able to do this
+
+      u32 w0[4];
+      u32 w1[4];
+      u32 w2[4];
+      u32 w3[4];
+
+      w0[0] = tmp[idx +  0];
+      w0[1] = tmp[idx +  1];
+      w0[2] = tmp[idx +  2];
+      w0[3] = tmp[idx +  3];
+      w1[0] = tmp[idx +  4];
+      w1[1] = tmp[idx +  5];
+      w1[2] = tmp[idx +  6];
+      w1[3] = tmp[idx +  7];
+      w2[0] = tmp[idx +  8];
+      w2[1] = tmp[idx +  9];
+      w2[2] = tmp[idx + 10];
+      w2[3] = tmp[idx + 11];
+      w3[0] = tmp[idx + 12];
+      w3[1] = tmp[idx + 13];
+      w3[2] = tmp[idx + 14];
+      w3[3] = tmp[idx + 15];
+
+      sha256_transform (w0, w1, w2, w3, h);
+    }
+  }
+  else
+  {
+    for (int i = 0, j = loop_pos; i < loop_cnt; i += 64, j += 64)
+    {
+      const int idx = (j % pw_len64) / 4; // the optimization trick is to be able to do this
+
+      u32 w0[4];
+      u32 w1[4];
+      u32 w2[4];
+      u32 w3[4];
+
+      w0[0] = tmps[gid].tmp[idx +  0];
+      w0[1] = tmps[gid].tmp[idx +  1];
+      w0[2] = tmps[gid].tmp[idx +  2];
+      w0[3] = tmps[gid].tmp[idx +  3];
+      w1[0] = tmps[gid].tmp[idx +  4];
+      w1[1] = tmps[gid].tmp[idx +  5];
+      w1[2] = tmps[gid].tmp[idx +  6];
+      w1[3] = tmps[gid].tmp[idx +  7];
+      w2[0] = tmps[gid].tmp[idx +  8];
+      w2[1] = tmps[gid].tmp[idx +  9];
+      w2[2] = tmps[gid].tmp[idx + 10];
+      w2[3] = tmps[gid].tmp[idx + 11];
+      w3[0] = tmps[gid].tmp[idx + 12];
+      w3[1] = tmps[gid].tmp[idx + 13];
+      w3[2] = tmps[gid].tmp[idx + 14];
+      w3[3] = tmps[gid].tmp[idx + 15];
+
+      sha256_transform (w0, w1, w2, w3, h);
+    }
+  }
+
+  tmps[gid].h[0] = h[0];
+  tmps[gid].h[1] = h[1];
+  tmps[gid].h[2] = h[2];
+  tmps[gid].h[3] = h[3];
+  tmps[gid].h[4] = h[4];
+  tmps[gid].h[5] = h[5];
+  tmps[gid].h[6] = h[6];
+  tmps[gid].h[7] = h[7];
+}
+
+KERNEL_FQ void m26800_comp (KERN_ATTR_TMPS_ESALT (hmac_sha256_tmp_t, snmpv3_t))
+{
+  /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  u32 w0[4];
+  u32 w1[4];
+  u32 w2[4];
+  u32 w3[4];
+
+  w0[0] = 0x80000000;
+  w0[1] = 0;
+  w0[2] = 0;
+  w0[3] = 0;
+  w1[0] = 0;
+  w1[1] = 0;
+  w1[2] = 0;
+  w1[3] = 0;
+  w2[0] = 0;
+  w2[1] = 0;
+  w2[2] = 0;
+  w2[3] = 0;
+  w3[0] = 0;
+  w3[1] = 0;
+  w3[2] = 0;
+  w3[3] = 1048576 * 8;
+
+  u32 h[8];
+
+  h[0] = tmps[gid].h[0];
+  h[1] = tmps[gid].h[1];
+  h[2] = tmps[gid].h[2];
+  h[3] = tmps[gid].h[3];
+  h[4] = tmps[gid].h[4];
+  h[5] = tmps[gid].h[5];
+  h[6] = tmps[gid].h[6];
+  h[7] = tmps[gid].h[7];
+
+  sha256_transform (w0, w1, w2, w3, h);
+
+  sha256_ctx_t ctx;
+
+  sha256_init (&ctx);
+
+  u32 w[16];
+
+  w[ 0] = h[0];
+  w[ 1] = h[1];
+  w[ 2] = h[2];
+  w[ 3] = h[3];
+  w[ 4] = h[4];
+  w[ 5] = h[5];
+  w[ 6] = h[6];
+  w[ 7] = h[7];
+  w[ 8] = 0;
+  w[ 9] = 0;
+  w[10] = 0;
+  w[11] = 0;
+  w[12] = 0;
+  w[13] = 0;
+  w[14] = 0;
+  w[15] = 0;
+
+  sha256_update (&ctx, w, 32);
+
+  sha256_update_global_swap (&ctx, esalt_bufs[DIGESTS_OFFSET].engineID_buf, esalt_bufs[DIGESTS_OFFSET].engineID_len);
+
+  w[ 0] = h[0];
+  w[ 1] = h[1];
+  w[ 2] = h[2];
+  w[ 3] = h[3];
+  w[ 4] = h[4];
+  w[ 5] = h[5];
+  w[ 6] = h[6];
+  w[ 7] = h[7];
+  w[ 8] = 0;
+  w[ 9] = 0;
+  w[10] = 0;
+  w[11] = 0;
+  w[12] = 0;
+  w[13] = 0;
+  w[14] = 0;
+  w[15] = 0;
+
+  sha256_update (&ctx, w, 32);
+
+  sha256_final (&ctx);
+
+  w[ 0] = ctx.h[0];
+  w[ 1] = ctx.h[1];
+  w[ 2] = ctx.h[2];
+  w[ 3] = ctx.h[3];
+  w[ 4] = ctx.h[4];
+  w[ 5] = ctx.h[5];
+  w[ 6] = ctx.h[6];
+  w[ 7] = ctx.h[7];
+  w[ 8] = 0;
+  w[ 9] = 0;
+  w[10] = 0;
+  w[11] = 0;
+  w[12] = 0;
+  w[13] = 0;
+  w[14] = 0;
+  w[15] = 0;
+
+  sha256_hmac_ctx_t hmac_ctx;
+
+  sha256_hmac_init (&hmac_ctx, w, 32);
+
+  sha256_hmac_update_global_swap (&hmac_ctx, esalt_bufs[DIGESTS_OFFSET].salt_buf, esalt_bufs[DIGESTS_OFFSET].salt_len);
+
+  sha256_hmac_final (&hmac_ctx);
+
+  const u32 r0 = hmac_ctx.opad.h[DGST_R0];
+  const u32 r1 = hmac_ctx.opad.h[DGST_R1];
+  const u32 r2 = hmac_ctx.opad.h[DGST_R2];
+  const u32 r3 = hmac_ctx.opad.h[DGST_R3];
+
+  #define il_pos 0
+
+  #ifdef KERNEL_STATIC
+  #include COMPARE_M
+  #endif
+}
diff --git a/OpenCL/m26900-pure.cl b/OpenCL/m26900-pure.cl
new file mode 100644
index 000000000..3b63483fa
--- /dev/null
+++ b/OpenCL/m26900-pure.cl
@@ -0,0 +1,495 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+//#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include "inc_vendor.h"
+#include "inc_types.h"
+#include "inc_platform.cl"
+#include "inc_common.cl"
+#include "inc_simd.cl"
+#include "inc_hash_sha384.cl"
+#endif
+
+#define COMPARE_S "inc_comp_single.cl"
+#define COMPARE_M "inc_comp_multi.cl"
+
+#define SNMPV3_SALT_MAX             1500
+#define SNMPV3_ENGINEID_MAX         34
+#define SNMPV3_MSG_AUTH_PARAMS_MAX  32
+#define SNMPV3_ROUNDS               1048576
+#define SNMPV3_MAX_PW_LENGTH        128
+
+#define SNMPV3_TMP_ELEMS            8192 // 8192 = (256 (max pw length) * SNMPV3_MAX_PW_LENGTH) / sizeof (u32)
+#define SNMPV3_HASH_ELEMS           8
+
+#define SNMPV3_MAX_SALT_ELEMS       512 // 512 * 4 = 2048 > 1500, also has to be multiple of SNMPV3_MAX_PW_LENGTH
+#define SNMPV3_MAX_ENGINE_ELEMS     32  // 32 * 4 = 128 > 34, also has to be multiple of SNMPV3_MAX_PW_LENGTH
+#define SNMPV3_MAX_PNUM_ELEMS       4   // 4 * 4 = 16 > 9
+
+#define SNMPV3_MAX_PW_LENGTH_OPT    32
+#define SNMPV3_TMP_ELEMS_OPT        ((SNMPV3_MAX_PW_LENGTH_OPT * SNMPV3_MAX_PW_LENGTH) / 4)
+                                    // (32 * 128) / 4 = 1024
+                                    // for pw length > 32 we use global memory reads
+
+typedef struct hmac_sha384_tmp
+{
+  u32 tmp[SNMPV3_TMP_ELEMS];
+  u64 h[SNMPV3_HASH_ELEMS];
+
+} hmac_sha384_tmp_t;
+
+typedef struct snmpv3
+{
+  u32 salt_buf[SNMPV3_MAX_SALT_ELEMS];
+  u32 salt_len;
+
+  u32 engineID_buf[SNMPV3_MAX_ENGINE_ELEMS];
+  u32 engineID_len;
+
+  u32 packet_number[SNMPV3_MAX_PNUM_ELEMS];
+
+} snmpv3_t;
+
+KERNEL_FQ void m26900_init (KERN_ATTR_TMPS_ESALT (hmac_sha384_tmp_t, snmpv3_t))
+{
+  /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  const u32 pw_len = pws[gid].pw_len;
+
+  u32 w[128] = { 0 };
+
+  for (u32 i = 0, idx = 0; i < pw_len; i += 4, idx += 1)
+  {
+    w[idx] = pws[gid].i[idx];
+  }
+
+  u8 *src_ptr = (u8 *) w;
+
+  // password 128 times, also swapped
+
+  u32 dst_buf[32];
+
+  u8 *dst_ptr = (u8 *) dst_buf;
+
+  int tmp_idx = 0;
+
+  for (int i = 0; i < 128; i++)
+  {
+    for (u32 j = 0; j < pw_len; j++)
+    {
+      const int dst_idx = tmp_idx & 127;
+
+      dst_ptr[dst_idx] = src_ptr[j];
+
+      // write to global memory every time 64 byte are written into cache
+
+      if (dst_idx == 127)
+      {
+        const int tmp_idx4 = (tmp_idx - 127) / 4;
+
+        tmps[gid].tmp[tmp_idx4 +  0] = hc_swap32_S (dst_buf[ 0]);
+        tmps[gid].tmp[tmp_idx4 +  1] = hc_swap32_S (dst_buf[ 1]);
+        tmps[gid].tmp[tmp_idx4 +  2] = hc_swap32_S (dst_buf[ 2]);
+        tmps[gid].tmp[tmp_idx4 +  3] = hc_swap32_S (dst_buf[ 3]);
+        tmps[gid].tmp[tmp_idx4 +  4] = hc_swap32_S (dst_buf[ 4]);
+        tmps[gid].tmp[tmp_idx4 +  5] = hc_swap32_S (dst_buf[ 5]);
+        tmps[gid].tmp[tmp_idx4 +  6] = hc_swap32_S (dst_buf[ 6]);
+        tmps[gid].tmp[tmp_idx4 +  7] = hc_swap32_S (dst_buf[ 7]);
+        tmps[gid].tmp[tmp_idx4 +  8] = hc_swap32_S (dst_buf[ 8]);
+        tmps[gid].tmp[tmp_idx4 +  9] = hc_swap32_S (dst_buf[ 9]);
+        tmps[gid].tmp[tmp_idx4 + 10] = hc_swap32_S (dst_buf[10]);
+        tmps[gid].tmp[tmp_idx4 + 11] = hc_swap32_S (dst_buf[11]);
+        tmps[gid].tmp[tmp_idx4 + 12] = hc_swap32_S (dst_buf[12]);
+        tmps[gid].tmp[tmp_idx4 + 13] = hc_swap32_S (dst_buf[13]);
+        tmps[gid].tmp[tmp_idx4 + 14] = hc_swap32_S (dst_buf[14]);
+        tmps[gid].tmp[tmp_idx4 + 15] = hc_swap32_S (dst_buf[15]);
+        tmps[gid].tmp[tmp_idx4 + 16] = hc_swap32_S (dst_buf[16]);
+        tmps[gid].tmp[tmp_idx4 + 17] = hc_swap32_S (dst_buf[17]);
+        tmps[gid].tmp[tmp_idx4 + 18] = hc_swap32_S (dst_buf[18]);
+        tmps[gid].tmp[tmp_idx4 + 19] = hc_swap32_S (dst_buf[19]);
+        tmps[gid].tmp[tmp_idx4 + 20] = hc_swap32_S (dst_buf[20]);
+        tmps[gid].tmp[tmp_idx4 + 21] = hc_swap32_S (dst_buf[21]);
+        tmps[gid].tmp[tmp_idx4 + 22] = hc_swap32_S (dst_buf[22]);
+        tmps[gid].tmp[tmp_idx4 + 23] = hc_swap32_S (dst_buf[23]);
+        tmps[gid].tmp[tmp_idx4 + 24] = hc_swap32_S (dst_buf[24]);
+        tmps[gid].tmp[tmp_idx4 + 25] = hc_swap32_S (dst_buf[25]);
+        tmps[gid].tmp[tmp_idx4 + 26] = hc_swap32_S (dst_buf[26]);
+        tmps[gid].tmp[tmp_idx4 + 27] = hc_swap32_S (dst_buf[27]);
+        tmps[gid].tmp[tmp_idx4 + 28] = hc_swap32_S (dst_buf[28]);
+        tmps[gid].tmp[tmp_idx4 + 29] = hc_swap32_S (dst_buf[29]);
+        tmps[gid].tmp[tmp_idx4 + 30] = hc_swap32_S (dst_buf[30]);
+        tmps[gid].tmp[tmp_idx4 + 31] = hc_swap32_S (dst_buf[31]);
+      }
+
+      tmp_idx++;
+    }
+  }
+
+  // hash
+
+  tmps[gid].h[0] = SHA384M_A;
+  tmps[gid].h[1] = SHA384M_B;
+  tmps[gid].h[2] = SHA384M_C;
+  tmps[gid].h[3] = SHA384M_D;
+  tmps[gid].h[4] = SHA384M_E;
+  tmps[gid].h[5] = SHA384M_F;
+  tmps[gid].h[6] = SHA384M_G;
+  tmps[gid].h[7] = SHA384M_H;
+}
+
+KERNEL_FQ void m26900_loop (KERN_ATTR_TMPS_ESALT (hmac_sha384_tmp_t, snmpv3_t))
+{
+  /**
+   * base
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  u64 h[8];
+
+  h[0] = tmps[gid].h[0];
+  h[1] = tmps[gid].h[1];
+  h[2] = tmps[gid].h[2];
+  h[3] = tmps[gid].h[3];
+  h[4] = tmps[gid].h[4];
+  h[5] = tmps[gid].h[5];
+  h[6] = tmps[gid].h[6];
+  h[7] = tmps[gid].h[7];
+
+  const u32 pw_len = pws[gid].pw_len;
+
+  const int pw_len128 = pw_len * 128;
+
+  if (pw_len <= SNMPV3_MAX_PW_LENGTH_OPT)
+  {
+    u32 tmp[SNMPV3_TMP_ELEMS_OPT];
+
+    for (int i = 0; i < pw_len128 / 4; i++)
+    {
+      tmp[i] = tmps[gid].tmp[i];
+    }
+
+    for (u32 i = 0, j = loop_pos; i < loop_cnt; i += 128, j += 128)
+    {
+      const int idx = (j % pw_len128) / 4; // the optimization trick is to be able to do this
+
+      u32 w0[4];
+      u32 w1[4];
+      u32 w2[4];
+      u32 w3[4];
+      u32 w4[4];
+      u32 w5[4];
+      u32 w6[4];
+      u32 w7[4];
+
+      w0[0] = tmp[idx +  0];
+      w0[1] = tmp[idx +  1];
+      w0[2] = tmp[idx +  2];
+      w0[3] = tmp[idx +  3];
+      w1[0] = tmp[idx +  4];
+      w1[1] = tmp[idx +  5];
+      w1[2] = tmp[idx +  6];
+      w1[3] = tmp[idx +  7];
+      w2[0] = tmp[idx +  8];
+      w2[1] = tmp[idx +  9];
+      w2[2] = tmp[idx + 10];
+      w2[3] = tmp[idx + 11];
+      w3[0] = tmp[idx + 12];
+      w3[1] = tmp[idx + 13];
+      w3[2] = tmp[idx + 14];
+      w3[3] = tmp[idx + 15];
+      w4[0] = tmp[idx + 16];
+      w4[1] = tmp[idx + 17];
+      w4[2] = tmp[idx + 18];
+      w4[3] = tmp[idx + 19];
+      w5[0] = tmp[idx + 20];
+      w5[1] = tmp[idx + 21];
+      w5[2] = tmp[idx + 22];
+      w5[3] = tmp[idx + 23];
+      w6[0] = tmp[idx + 24];
+      w6[1] = tmp[idx + 25];
+      w6[2] = tmp[idx + 26];
+      w6[3] = tmp[idx + 27];
+      w7[0] = tmp[idx + 28];
+      w7[1] = tmp[idx + 29];
+      w7[2] = tmp[idx + 30];
+      w7[3] = tmp[idx + 31];
+
+      sha384_transform (w0, w1, w2, w3, w4, w5, w6, w7, h);
+    }
+  }
+  else
+  {
+    for (u32 i = 0, j = loop_pos; i < loop_cnt; i += 128, j += 128)
+    {
+      const int idx = (j % pw_len128) / 4; // the optimization trick is to be able to do this
+
+      u32 w0[4];
+      u32 w1[4];
+      u32 w2[4];
+      u32 w3[4];
+      u32 w4[4];
+      u32 w5[4];
+      u32 w6[4];
+      u32 w7[4];
+
+      w0[0] = tmps[gid].tmp[idx +  0];
+      w0[1] = tmps[gid].tmp[idx +  1];
+      w0[2] = tmps[gid].tmp[idx +  2];
+      w0[3] = tmps[gid].tmp[idx +  3];
+      w1[0] = tmps[gid].tmp[idx +  4];
+      w1[1] = tmps[gid].tmp[idx +  5];
+      w1[2] = tmps[gid].tmp[idx +  6];
+      w1[3] = tmps[gid].tmp[idx +  7];
+      w2[0] = tmps[gid].tmp[idx +  8];
+      w2[1] = tmps[gid].tmp[idx +  9];
+      w2[2] = tmps[gid].tmp[idx + 10];
+      w2[3] = tmps[gid].tmp[idx + 11];
+      w3[0] = tmps[gid].tmp[idx + 12];
+      w3[1] = tmps[gid].tmp[idx + 13];
+      w3[2] = tmps[gid].tmp[idx + 14];
+      w3[3] = tmps[gid].tmp[idx + 15];
+      w4[0] = tmps[gid].tmp[idx + 16];
+      w4[1] = tmps[gid].tmp[idx + 17];
+      w4[2] = tmps[gid].tmp[idx + 18];
+      w4[3] = tmps[gid].tmp[idx + 19];
+      w5[0] = tmps[gid].tmp[idx + 20];
+      w5[1] = tmps[gid].tmp[idx + 21];
+      w5[2] = tmps[gid].tmp[idx + 22];
+      w5[3] = tmps[gid].tmp[idx + 23];
+      w6[0] = tmps[gid].tmp[idx + 24];
+      w6[1] = tmps[gid].tmp[idx + 25];
+      w6[2] = tmps[gid].tmp[idx + 26];
+      w6[3] = tmps[gid].tmp[idx + 27];
+      w7[0] = tmps[gid].tmp[idx + 28];
+      w7[1] = tmps[gid].tmp[idx + 29];
+      w7[2] = tmps[gid].tmp[idx + 30];
+      w7[3] = tmps[gid].tmp[idx + 31];
+
+      sha384_transform (w0, w1, w2, w3, w4, w5, w6, w7, h);
+    }
+  }
+
+  tmps[gid].h[0] = h[0];
+  tmps[gid].h[1] = h[1];
+  tmps[gid].h[2] = h[2];
+  tmps[gid].h[3] = h[3];
+  tmps[gid].h[4] = h[4];
+  tmps[gid].h[5] = h[5];
+  tmps[gid].h[6] = h[6];
+  tmps[gid].h[7] = h[7];
+}
+
+KERNEL_FQ void m26900_comp (KERN_ATTR_TMPS_ESALT (hmac_sha384_tmp_t, snmpv3_t))
+{
+  /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  u32 w0[4];
+  u32 w1[4];
+  u32 w2[4];
+  u32 w3[4];
+  u32 w4[4];
+  u32 w5[4];
+  u32 w6[4];
+  u32 w7[4];
+
+  w0[0] = 0x80000000;
+  w0[1] = 0;
+  w0[2] = 0;
+  w0[3] = 0;
+  w1[0] = 0;
+  w1[1] = 0;
+  w1[2] = 0;
+  w1[3] = 0;
+  w2[0] = 0;
+  w2[1] = 0;
+  w2[2] = 0;
+  w2[3] = 0;
+  w3[0] = 0;
+  w3[1] = 0;
+  w3[2] = 0;
+  w3[3] = 0;
+  w4[0] = 0;
+  w4[1] = 0;
+  w4[2] = 0;
+  w4[3] = 0;
+  w5[0] = 0;
+  w5[1] = 0;
+  w5[2] = 0;
+  w5[3] = 0;
+  w6[0] = 0;
+  w6[1] = 0;
+  w6[2] = 0;
+  w6[3] = 0;
+  w7[0] = 0;
+  w7[1] = 0;
+  w7[2] = 0;
+  w7[3] = 1048576 * 8;
+
+  u64 h[8];
+
+  h[0] = tmps[gid].h[0];
+  h[1] = tmps[gid].h[1];
+  h[2] = tmps[gid].h[2];
+  h[3] = tmps[gid].h[3];
+  h[4] = tmps[gid].h[4];
+  h[5] = tmps[gid].h[5];
+  h[6] = tmps[gid].h[6];
+  h[7] = tmps[gid].h[7];
+
+  sha384_transform (w0, w1, w2, w3, w4, w5, w6, w7, h);
+
+  sha384_ctx_t ctx;
+
+  sha384_init (&ctx);
+
+  u32 w[32];
+
+  w[ 0] = h32_from_64_S (h[0]);
+  w[ 1] = l32_from_64_S (h[0]);
+  w[ 2] = h32_from_64_S (h[1]);
+  w[ 3] = l32_from_64_S (h[1]);
+  w[ 4] = h32_from_64_S (h[2]);
+  w[ 5] = l32_from_64_S (h[2]);
+  w[ 6] = h32_from_64_S (h[3]);
+  w[ 7] = l32_from_64_S (h[3]);
+  w[ 8] = h32_from_64_S (h[4]);
+  w[ 9] = l32_from_64_S (h[4]);
+  w[10] = h32_from_64_S (h[5]);
+  w[11] = l32_from_64_S (h[5]);
+  w[12] = 0;
+  w[13] = 0;
+  w[14] = 0;
+  w[15] = 0;
+  w[16] = 0;
+  w[17] = 0;
+  w[18] = 0;
+  w[19] = 0;
+  w[20] = 0;
+  w[21] = 0;
+  w[22] = 0;
+  w[23] = 0;
+  w[24] = 0;
+  w[25] = 0;
+  w[26] = 0;
+  w[27] = 0;
+  w[28] = 0;
+  w[29] = 0;
+  w[30] = 0;
+  w[31] = 0;
+
+  sha384_update (&ctx, w, 48);
+
+  sha384_update_global_swap (&ctx, esalt_bufs[DIGESTS_OFFSET].engineID_buf, esalt_bufs[DIGESTS_OFFSET].engineID_len);
+
+  w[ 0] = h32_from_64_S (h[0]);
+  w[ 1] = l32_from_64_S (h[0]);
+  w[ 2] = h32_from_64_S (h[1]);
+  w[ 3] = l32_from_64_S (h[1]);
+  w[ 4] = h32_from_64_S (h[2]);
+  w[ 5] = l32_from_64_S (h[2]);
+  w[ 6] = h32_from_64_S (h[3]);
+  w[ 7] = l32_from_64_S (h[3]);
+  w[ 8] = h32_from_64_S (h[4]);
+  w[ 9] = l32_from_64_S (h[4]);
+  w[10] = h32_from_64_S (h[5]);
+  w[11] = l32_from_64_S (h[5]);
+  w[12] = 0;
+  w[13] = 0;
+  w[14] = 0;
+  w[15] = 0;
+  w[16] = 0;
+  w[17] = 0;
+  w[18] = 0;
+  w[19] = 0;
+  w[20] = 0;
+  w[21] = 0;
+  w[22] = 0;
+  w[23] = 0;
+  w[24] = 0;
+  w[25] = 0;
+  w[26] = 0;
+  w[27] = 0;
+  w[28] = 0;
+  w[29] = 0;
+  w[30] = 0;
+  w[31] = 0;
+
+  sha384_update (&ctx, w, 48);
+
+  sha384_final (&ctx);
+
+  w[ 0] = h32_from_64_S (ctx.h[0]);
+  w[ 1] = l32_from_64_S (ctx.h[0]);
+  w[ 2] = h32_from_64_S (ctx.h[1]);
+  w[ 3] = l32_from_64_S (ctx.h[1]);
+  w[ 4] = h32_from_64_S (ctx.h[2]);
+  w[ 5] = l32_from_64_S (ctx.h[2]);
+  w[ 6] = h32_from_64_S (ctx.h[3]);
+  w[ 7] = l32_from_64_S (ctx.h[3]);
+  w[ 8] = h32_from_64_S (ctx.h[4]);
+  w[ 9] = l32_from_64_S (ctx.h[4]);
+  w[10] = h32_from_64_S (ctx.h[5]);
+  w[11] = l32_from_64_S (ctx.h[5]);
+  w[12] = 0;
+  w[13] = 0;
+  w[14] = 0;
+  w[15] = 0;
+  w[16] = 0;
+  w[17] = 0;
+  w[18] = 0;
+  w[19] = 0;
+  w[20] = 0;
+  w[21] = 0;
+  w[22] = 0;
+  w[23] = 0;
+  w[24] = 0;
+  w[25] = 0;
+  w[26] = 0;
+  w[27] = 0;
+  w[28] = 0;
+  w[29] = 0;
+  w[30] = 0;
+  w[31] = 0;
+
+  sha384_hmac_ctx_t hmac_ctx;
+
+  sha384_hmac_init (&hmac_ctx, w, 48);
+
+  sha384_hmac_update_global_swap (&hmac_ctx, esalt_bufs[DIGESTS_OFFSET].salt_buf, esalt_bufs[DIGESTS_OFFSET].salt_len);
+
+  sha384_hmac_final (&hmac_ctx);
+
+  const u32 r0 = l32_from_64 (hmac_ctx.opad.h[1]);
+  const u32 r1 = h32_from_64 (hmac_ctx.opad.h[1]);
+  const u32 r2 = l32_from_64 (hmac_ctx.opad.h[0]);
+  const u32 r3 = h32_from_64 (hmac_ctx.opad.h[0]);
+
+  #define il_pos 0
+
+  #ifdef KERNEL_STATIC
+  #include COMPARE_M
+  #endif
+}
diff --git a/OpenCL/m27000-pure.cl b/OpenCL/m27000-pure.cl
new file mode 100644
index 000000000..6f5e8955a
--- /dev/null
+++ b/OpenCL/m27000-pure.cl
@@ -0,0 +1,697 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+//#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include "inc_vendor.h"
+#include "inc_types.h"
+#include "inc_platform.cl"
+#include "inc_common.cl"
+#include "inc_rp.h"
+#include "inc_rp.cl"
+#include "inc_scalar.cl"
+#include "inc_hash_md4.cl"
+#endif
+
+#define COMPARE_S "inc_comp_single.cl"
+#define COMPARE_M "inc_comp_multi.cl"
+
+#define PERM_OP(a,b,tt,n,m) \
+{                           \
+  tt = a >> n;              \
+  tt = tt ^ b;              \
+  tt = tt & m;              \
+  b = b ^ tt;               \
+  tt = tt << n;             \
+  a = a ^ tt;               \
+}
+
+#define HPERM_OP(a,tt,n,m)  \
+{                           \
+  tt = a << (16 + n);       \
+  tt = tt ^ a;              \
+  tt = tt & m;              \
+  a  = a ^ tt;              \
+  tt = tt >> (16 + n);      \
+  a  = a ^ tt;              \
+}
+
+CONSTANT_VK u32a c_SPtrans[8][64] =
+{
+  {
+    0x02080800, 0x00080000, 0x02000002, 0x02080802,
+    0x02000000, 0x00080802, 0x00080002, 0x02000002,
+    0x00080802, 0x02080800, 0x02080000, 0x00000802,
+    0x02000802, 0x02000000, 0x00000000, 0x00080002,
+    0x00080000, 0x00000002, 0x02000800, 0x00080800,
+    0x02080802, 0x02080000, 0x00000802, 0x02000800,
+    0x00000002, 0x00000800, 0x00080800, 0x02080002,
+    0x00000800, 0x02000802, 0x02080002, 0x00000000,
+    0x00000000, 0x02080802, 0x02000800, 0x00080002,
+    0x02080800, 0x00080000, 0x00000802, 0x02000800,
+    0x02080002, 0x00000800, 0x00080800, 0x02000002,
+    0x00080802, 0x00000002, 0x02000002, 0x02080000,
+    0x02080802, 0x00080800, 0x02080000, 0x02000802,
+    0x02000000, 0x00000802, 0x00080002, 0x00000000,
+    0x00080000, 0x02000000, 0x02000802, 0x02080800,
+    0x00000002, 0x02080002, 0x00000800, 0x00080802,
+  },
+  {
+    0x40108010, 0x00000000, 0x00108000, 0x40100000,
+    0x40000010, 0x00008010, 0x40008000, 0x00108000,
+    0x00008000, 0x40100010, 0x00000010, 0x40008000,
+    0x00100010, 0x40108000, 0x40100000, 0x00000010,
+    0x00100000, 0x40008010, 0x40100010, 0x00008000,
+    0x00108010, 0x40000000, 0x00000000, 0x00100010,
+    0x40008010, 0x00108010, 0x40108000, 0x40000010,
+    0x40000000, 0x00100000, 0x00008010, 0x40108010,
+    0x00100010, 0x40108000, 0x40008000, 0x00108010,
+    0x40108010, 0x00100010, 0x40000010, 0x00000000,
+    0x40000000, 0x00008010, 0x00100000, 0x40100010,
+    0x00008000, 0x40000000, 0x00108010, 0x40008010,
+    0x40108000, 0x00008000, 0x00000000, 0x40000010,
+    0x00000010, 0x40108010, 0x00108000, 0x40100000,
+    0x40100010, 0x00100000, 0x00008010, 0x40008000,
+    0x40008010, 0x00000010, 0x40100000, 0x00108000,
+  },
+  {
+    0x04000001, 0x04040100, 0x00000100, 0x04000101,
+    0x00040001, 0x04000000, 0x04000101, 0x00040100,
+    0x04000100, 0x00040000, 0x04040000, 0x00000001,
+    0x04040101, 0x00000101, 0x00000001, 0x04040001,
+    0x00000000, 0x00040001, 0x04040100, 0x00000100,
+    0x00000101, 0x04040101, 0x00040000, 0x04000001,
+    0x04040001, 0x04000100, 0x00040101, 0x04040000,
+    0x00040100, 0x00000000, 0x04000000, 0x00040101,
+    0x04040100, 0x00000100, 0x00000001, 0x00040000,
+    0x00000101, 0x00040001, 0x04040000, 0x04000101,
+    0x00000000, 0x04040100, 0x00040100, 0x04040001,
+    0x00040001, 0x04000000, 0x04040101, 0x00000001,
+    0x00040101, 0x04000001, 0x04000000, 0x04040101,
+    0x00040000, 0x04000100, 0x04000101, 0x00040100,
+    0x04000100, 0x00000000, 0x04040001, 0x00000101,
+    0x04000001, 0x00040101, 0x00000100, 0x04040000,
+  },
+  {
+    0x00401008, 0x10001000, 0x00000008, 0x10401008,
+    0x00000000, 0x10400000, 0x10001008, 0x00400008,
+    0x10401000, 0x10000008, 0x10000000, 0x00001008,
+    0x10000008, 0x00401008, 0x00400000, 0x10000000,
+    0x10400008, 0x00401000, 0x00001000, 0x00000008,
+    0x00401000, 0x10001008, 0x10400000, 0x00001000,
+    0x00001008, 0x00000000, 0x00400008, 0x10401000,
+    0x10001000, 0x10400008, 0x10401008, 0x00400000,
+    0x10400008, 0x00001008, 0x00400000, 0x10000008,
+    0x00401000, 0x10001000, 0x00000008, 0x10400000,
+    0x10001008, 0x00000000, 0x00001000, 0x00400008,
+    0x00000000, 0x10400008, 0x10401000, 0x00001000,
+    0x10000000, 0x10401008, 0x00401008, 0x00400000,
+    0x10401008, 0x00000008, 0x10001000, 0x00401008,
+    0x00400008, 0x00401000, 0x10400000, 0x10001008,
+    0x00001008, 0x10000000, 0x10000008, 0x10401000,
+  },
+  {
+    0x08000000, 0x00010000, 0x00000400, 0x08010420,
+    0x08010020, 0x08000400, 0x00010420, 0x08010000,
+    0x00010000, 0x00000020, 0x08000020, 0x00010400,
+    0x08000420, 0x08010020, 0x08010400, 0x00000000,
+    0x00010400, 0x08000000, 0x00010020, 0x00000420,
+    0x08000400, 0x00010420, 0x00000000, 0x08000020,
+    0x00000020, 0x08000420, 0x08010420, 0x00010020,
+    0x08010000, 0x00000400, 0x00000420, 0x08010400,
+    0x08010400, 0x08000420, 0x00010020, 0x08010000,
+    0x00010000, 0x00000020, 0x08000020, 0x08000400,
+    0x08000000, 0x00010400, 0x08010420, 0x00000000,
+    0x00010420, 0x08000000, 0x00000400, 0x00010020,
+    0x08000420, 0x00000400, 0x00000000, 0x08010420,
+    0x08010020, 0x08010400, 0x00000420, 0x00010000,
+    0x00010400, 0x08010020, 0x08000400, 0x00000420,
+    0x00000020, 0x00010420, 0x08010000, 0x08000020,
+  },
+  {
+    0x80000040, 0x00200040, 0x00000000, 0x80202000,
+    0x00200040, 0x00002000, 0x80002040, 0x00200000,
+    0x00002040, 0x80202040, 0x00202000, 0x80000000,
+    0x80002000, 0x80000040, 0x80200000, 0x00202040,
+    0x00200000, 0x80002040, 0x80200040, 0x00000000,
+    0x00002000, 0x00000040, 0x80202000, 0x80200040,
+    0x80202040, 0x80200000, 0x80000000, 0x00002040,
+    0x00000040, 0x00202000, 0x00202040, 0x80002000,
+    0x00002040, 0x80000000, 0x80002000, 0x00202040,
+    0x80202000, 0x00200040, 0x00000000, 0x80002000,
+    0x80000000, 0x00002000, 0x80200040, 0x00200000,
+    0x00200040, 0x80202040, 0x00202000, 0x00000040,
+    0x80202040, 0x00202000, 0x00200000, 0x80002040,
+    0x80000040, 0x80200000, 0x00202040, 0x00000000,
+    0x00002000, 0x80000040, 0x80002040, 0x80202000,
+    0x80200000, 0x00002040, 0x00000040, 0x80200040,
+  },
+  {
+    0x00004000, 0x00000200, 0x01000200, 0x01000004,
+    0x01004204, 0x00004004, 0x00004200, 0x00000000,
+    0x01000000, 0x01000204, 0x00000204, 0x01004000,
+    0x00000004, 0x01004200, 0x01004000, 0x00000204,
+    0x01000204, 0x00004000, 0x00004004, 0x01004204,
+    0x00000000, 0x01000200, 0x01000004, 0x00004200,
+    0x01004004, 0x00004204, 0x01004200, 0x00000004,
+    0x00004204, 0x01004004, 0x00000200, 0x01000000,
+    0x00004204, 0x01004000, 0x01004004, 0x00000204,
+    0x00004000, 0x00000200, 0x01000000, 0x01004004,
+    0x01000204, 0x00004204, 0x00004200, 0x00000000,
+    0x00000200, 0x01000004, 0x00000004, 0x01000200,
+    0x00000000, 0x01000204, 0x01000200, 0x00004200,
+    0x00000204, 0x00004000, 0x01004204, 0x01000000,
+    0x01004200, 0x00000004, 0x00004004, 0x01004204,
+    0x01000004, 0x01004200, 0x01004000, 0x00004004,
+  },
+  {
+    0x20800080, 0x20820000, 0x00020080, 0x00000000,
+    0x20020000, 0x00800080, 0x20800000, 0x20820080,
+    0x00000080, 0x20000000, 0x00820000, 0x00020080,
+    0x00820080, 0x20020080, 0x20000080, 0x20800000,
+    0x00020000, 0x00820080, 0x00800080, 0x20020000,
+    0x20820080, 0x20000080, 0x00000000, 0x00820000,
+    0x20000000, 0x00800000, 0x20020080, 0x20800080,
+    0x00800000, 0x00020000, 0x20820000, 0x00000080,
+    0x00800000, 0x00020000, 0x20000080, 0x20820080,
+    0x00020080, 0x20000000, 0x00000000, 0x00820000,
+    0x20800080, 0x20020080, 0x20020000, 0x00800080,
+    0x20820000, 0x00000080, 0x00800080, 0x20020000,
+    0x20820080, 0x00800000, 0x20800000, 0x20000080,
+    0x00820000, 0x00020080, 0x20020080, 0x20800000,
+    0x00000080, 0x20820000, 0x00820080, 0x00000000,
+    0x20000000, 0x20800080, 0x00020000, 0x00820080,
+  }
+};
+
+CONSTANT_VK u32a c_skb[8][64] =
+{
+  {
+    0x00000000, 0x00000010, 0x20000000, 0x20000010,
+    0x00010000, 0x00010010, 0x20010000, 0x20010010,
+    0x00000800, 0x00000810, 0x20000800, 0x20000810,
+    0x00010800, 0x00010810, 0x20010800, 0x20010810,
+    0x00000020, 0x00000030, 0x20000020, 0x20000030,
+    0x00010020, 0x00010030, 0x20010020, 0x20010030,
+    0x00000820, 0x00000830, 0x20000820, 0x20000830,
+    0x00010820, 0x00010830, 0x20010820, 0x20010830,
+    0x00080000, 0x00080010, 0x20080000, 0x20080010,
+    0x00090000, 0x00090010, 0x20090000, 0x20090010,
+    0x00080800, 0x00080810, 0x20080800, 0x20080810,
+    0x00090800, 0x00090810, 0x20090800, 0x20090810,
+    0x00080020, 0x00080030, 0x20080020, 0x20080030,
+    0x00090020, 0x00090030, 0x20090020, 0x20090030,
+    0x00080820, 0x00080830, 0x20080820, 0x20080830,
+    0x00090820, 0x00090830, 0x20090820, 0x20090830,
+  },
+  {
+    0x00000000, 0x02000000, 0x00002000, 0x02002000,
+    0x00200000, 0x02200000, 0x00202000, 0x02202000,
+    0x00000004, 0x02000004, 0x00002004, 0x02002004,
+    0x00200004, 0x02200004, 0x00202004, 0x02202004,
+    0x00000400, 0x02000400, 0x00002400, 0x02002400,
+    0x00200400, 0x02200400, 0x00202400, 0x02202400,
+    0x00000404, 0x02000404, 0x00002404, 0x02002404,
+    0x00200404, 0x02200404, 0x00202404, 0x02202404,
+    0x10000000, 0x12000000, 0x10002000, 0x12002000,
+    0x10200000, 0x12200000, 0x10202000, 0x12202000,
+    0x10000004, 0x12000004, 0x10002004, 0x12002004,
+    0x10200004, 0x12200004, 0x10202004, 0x12202004,
+    0x10000400, 0x12000400, 0x10002400, 0x12002400,
+    0x10200400, 0x12200400, 0x10202400, 0x12202400,
+    0x10000404, 0x12000404, 0x10002404, 0x12002404,
+    0x10200404, 0x12200404, 0x10202404, 0x12202404,
+  },
+  {
+    0x00000000, 0x00000001, 0x00040000, 0x00040001,
+    0x01000000, 0x01000001, 0x01040000, 0x01040001,
+    0x00000002, 0x00000003, 0x00040002, 0x00040003,
+    0x01000002, 0x01000003, 0x01040002, 0x01040003,
+    0x00000200, 0x00000201, 0x00040200, 0x00040201,
+    0x01000200, 0x01000201, 0x01040200, 0x01040201,
+    0x00000202, 0x00000203, 0x00040202, 0x00040203,
+    0x01000202, 0x01000203, 0x01040202, 0x01040203,
+    0x08000000, 0x08000001, 0x08040000, 0x08040001,
+    0x09000000, 0x09000001, 0x09040000, 0x09040001,
+    0x08000002, 0x08000003, 0x08040002, 0x08040003,
+    0x09000002, 0x09000003, 0x09040002, 0x09040003,
+    0x08000200, 0x08000201, 0x08040200, 0x08040201,
+    0x09000200, 0x09000201, 0x09040200, 0x09040201,
+    0x08000202, 0x08000203, 0x08040202, 0x08040203,
+    0x09000202, 0x09000203, 0x09040202, 0x09040203,
+  },
+  {
+    0x00000000, 0x00100000, 0x00000100, 0x00100100,
+    0x00000008, 0x00100008, 0x00000108, 0x00100108,
+    0x00001000, 0x00101000, 0x00001100, 0x00101100,
+    0x00001008, 0x00101008, 0x00001108, 0x00101108,
+    0x04000000, 0x04100000, 0x04000100, 0x04100100,
+    0x04000008, 0x04100008, 0x04000108, 0x04100108,
+    0x04001000, 0x04101000, 0x04001100, 0x04101100,
+    0x04001008, 0x04101008, 0x04001108, 0x04101108,
+    0x00020000, 0x00120000, 0x00020100, 0x00120100,
+    0x00020008, 0x00120008, 0x00020108, 0x00120108,
+    0x00021000, 0x00121000, 0x00021100, 0x00121100,
+    0x00021008, 0x00121008, 0x00021108, 0x00121108,
+    0x04020000, 0x04120000, 0x04020100, 0x04120100,
+    0x04020008, 0x04120008, 0x04020108, 0x04120108,
+    0x04021000, 0x04121000, 0x04021100, 0x04121100,
+    0x04021008, 0x04121008, 0x04021108, 0x04121108,
+  },
+  {
+    0x00000000, 0x10000000, 0x00010000, 0x10010000,
+    0x00000004, 0x10000004, 0x00010004, 0x10010004,
+    0x20000000, 0x30000000, 0x20010000, 0x30010000,
+    0x20000004, 0x30000004, 0x20010004, 0x30010004,
+    0x00100000, 0x10100000, 0x00110000, 0x10110000,
+    0x00100004, 0x10100004, 0x00110004, 0x10110004,
+    0x20100000, 0x30100000, 0x20110000, 0x30110000,
+    0x20100004, 0x30100004, 0x20110004, 0x30110004,
+    0x00001000, 0x10001000, 0x00011000, 0x10011000,
+    0x00001004, 0x10001004, 0x00011004, 0x10011004,
+    0x20001000, 0x30001000, 0x20011000, 0x30011000,
+    0x20001004, 0x30001004, 0x20011004, 0x30011004,
+    0x00101000, 0x10101000, 0x00111000, 0x10111000,
+    0x00101004, 0x10101004, 0x00111004, 0x10111004,
+    0x20101000, 0x30101000, 0x20111000, 0x30111000,
+    0x20101004, 0x30101004, 0x20111004, 0x30111004,
+  },
+  {
+    0x00000000, 0x08000000, 0x00000008, 0x08000008,
+    0x00000400, 0x08000400, 0x00000408, 0x08000408,
+    0x00020000, 0x08020000, 0x00020008, 0x08020008,
+    0x00020400, 0x08020400, 0x00020408, 0x08020408,
+    0x00000001, 0x08000001, 0x00000009, 0x08000009,
+    0x00000401, 0x08000401, 0x00000409, 0x08000409,
+    0x00020001, 0x08020001, 0x00020009, 0x08020009,
+    0x00020401, 0x08020401, 0x00020409, 0x08020409,
+    0x02000000, 0x0A000000, 0x02000008, 0x0A000008,
+    0x02000400, 0x0A000400, 0x02000408, 0x0A000408,
+    0x02020000, 0x0A020000, 0x02020008, 0x0A020008,
+    0x02020400, 0x0A020400, 0x02020408, 0x0A020408,
+    0x02000001, 0x0A000001, 0x02000009, 0x0A000009,
+    0x02000401, 0x0A000401, 0x02000409, 0x0A000409,
+    0x02020001, 0x0A020001, 0x02020009, 0x0A020009,
+    0x02020401, 0x0A020401, 0x02020409, 0x0A020409,
+  },
+  {
+    0x00000000, 0x00000100, 0x00080000, 0x00080100,
+    0x01000000, 0x01000100, 0x01080000, 0x01080100,
+    0x00000010, 0x00000110, 0x00080010, 0x00080110,
+    0x01000010, 0x01000110, 0x01080010, 0x01080110,
+    0x00200000, 0x00200100, 0x00280000, 0x00280100,
+    0x01200000, 0x01200100, 0x01280000, 0x01280100,
+    0x00200010, 0x00200110, 0x00280010, 0x00280110,
+    0x01200010, 0x01200110, 0x01280010, 0x01280110,
+    0x00000200, 0x00000300, 0x00080200, 0x00080300,
+    0x01000200, 0x01000300, 0x01080200, 0x01080300,
+    0x00000210, 0x00000310, 0x00080210, 0x00080310,
+    0x01000210, 0x01000310, 0x01080210, 0x01080310,
+    0x00200200, 0x00200300, 0x00280200, 0x00280300,
+    0x01200200, 0x01200300, 0x01280200, 0x01280300,
+    0x00200210, 0x00200310, 0x00280210, 0x00280310,
+    0x01200210, 0x01200310, 0x01280210, 0x01280310,
+  },
+  {
+    0x00000000, 0x04000000, 0x00040000, 0x04040000,
+    0x00000002, 0x04000002, 0x00040002, 0x04040002,
+    0x00002000, 0x04002000, 0x00042000, 0x04042000,
+    0x00002002, 0x04002002, 0x00042002, 0x04042002,
+    0x00000020, 0x04000020, 0x00040020, 0x04040020,
+    0x00000022, 0x04000022, 0x00040022, 0x04040022,
+    0x00002020, 0x04002020, 0x00042020, 0x04042020,
+    0x00002022, 0x04002022, 0x00042022, 0x04042022,
+    0x00000800, 0x04000800, 0x00040800, 0x04040800,
+    0x00000802, 0x04000802, 0x00040802, 0x04040802,
+    0x00002800, 0x04002800, 0x00042800, 0x04042800,
+    0x00002802, 0x04002802, 0x00042802, 0x04042802,
+    0x00000820, 0x04000820, 0x00040820, 0x04040820,
+    0x00000822, 0x04000822, 0x00040822, 0x04040822,
+    0x00002820, 0x04002820, 0x00042820, 0x04042820,
+    0x00002822, 0x04002822, 0x00042822, 0x04042822
+  }
+};
+
+#if   VECT_SIZE == 1
+#define BOX(i,n,S) (S)[(n)][(i)]
+#elif VECT_SIZE == 2
+#define BOX(i,n,S) make_u32x ((S)[(n)][(i).s0], (S)[(n)][(i).s1])
+#elif VECT_SIZE == 4
+#define BOX(i,n,S) make_u32x ((S)[(n)][(i).s0], (S)[(n)][(i).s1], (S)[(n)][(i).s2], (S)[(n)][(i).s3])
+#elif VECT_SIZE == 8
+#define BOX(i,n,S) make_u32x ((S)[(n)][(i).s0], (S)[(n)][(i).s1], (S)[(n)][(i).s2], (S)[(n)][(i).s3], (S)[(n)][(i).s4], (S)[(n)][(i).s5], (S)[(n)][(i).s6], (S)[(n)][(i).s7])
+#elif VECT_SIZE == 16
+#define BOX(i,n,S) make_u32x ((S)[(n)][(i).s0], (S)[(n)][(i).s1], (S)[(n)][(i).s2], (S)[(n)][(i).s3], (S)[(n)][(i).s4], (S)[(n)][(i).s5], (S)[(n)][(i).s6], (S)[(n)][(i).s7], (S)[(n)][(i).s8], (S)[(n)][(i).s9], (S)[(n)][(i).sa], (S)[(n)][(i).sb], (S)[(n)][(i).sc], (S)[(n)][(i).sd], (S)[(n)][(i).se], (S)[(n)][(i).sf])
+#endif
+
+DECLSPEC void _des_crypt_encrypt (u32 *iv, u32 *data, u32 *Kc, u32 *Kd, SHM_TYPE u32 (*s_SPtrans)[64])
+{
+  u32 r = data[0];
+  u32 l = data[1];
+
+  #ifdef _unroll
+  #pragma unroll
+  #endif
+  for (u32 i = 0; i < 16; i += 2)
+  {
+    u32 u;
+    u32 t;
+
+    u = Kc[i + 0] ^ hc_rotl32 (r, 30u);
+    t = Kd[i + 0] ^ hc_rotl32 (r, 26u);
+
+    l ^= BOX (((u >>  0) & 0x3f), 0, s_SPtrans)
+       | BOX (((u >>  8) & 0x3f), 2, s_SPtrans)
+       | BOX (((u >> 16) & 0x3f), 4, s_SPtrans)
+       | BOX (((u >> 24) & 0x3f), 6, s_SPtrans)
+       | BOX (((t >>  0) & 0x3f), 1, s_SPtrans)
+       | BOX (((t >>  8) & 0x3f), 3, s_SPtrans)
+       | BOX (((t >> 16) & 0x3f), 5, s_SPtrans)
+       | BOX (((t >> 24) & 0x3f), 7, s_SPtrans);
+
+    u = Kc[i + 1] ^ hc_rotl32 (l, 30u);
+    t = Kd[i + 1] ^ hc_rotl32 (l, 26u);
+
+    r ^= BOX (((u >>  0) & 0x3f), 0, s_SPtrans)
+       | BOX (((u >>  8) & 0x3f), 2, s_SPtrans)
+       | BOX (((u >> 16) & 0x3f), 4, s_SPtrans)
+       | BOX (((u >> 24) & 0x3f), 6, s_SPtrans)
+       | BOX (((t >>  0) & 0x3f), 1, s_SPtrans)
+       | BOX (((t >>  8) & 0x3f), 3, s_SPtrans)
+       | BOX (((t >> 16) & 0x3f), 5, s_SPtrans)
+       | BOX (((t >> 24) & 0x3f), 7, s_SPtrans);
+  }
+
+  iv[0] = l;
+  iv[1] = r;
+}
+
+DECLSPEC void _des_crypt_keysetup (u32 c, u32 d, u32 *Kc, u32 *Kd, SHM_TYPE u32 (*s_skb)[64])
+{
+  u32 tt;
+
+  PERM_OP  (d, c, tt, 4, 0x0f0f0f0f);
+  HPERM_OP (c,    tt, 2, 0xcccc0000);
+  HPERM_OP (d,    tt, 2, 0xcccc0000);
+  PERM_OP  (d, c, tt, 1, 0x55555555);
+  PERM_OP  (c, d, tt, 8, 0x00ff00ff);
+  PERM_OP  (d, c, tt, 1, 0x55555555);
+
+  d = ((d & 0x000000ff) << 16)
+    | ((d & 0x0000ff00) <<  0)
+    | ((d & 0x00ff0000) >> 16)
+    | ((c & 0xf0000000) >>  4);
+
+  c = c & 0x0fffffff;
+
+  #ifdef _unroll
+  #pragma unroll
+  #endif
+  for (u32 i = 0; i < 16; i++)
+  {
+    if ((i < 2) || (i == 8) || (i == 15))
+    {
+      c = ((c >> 1) | (c << 27));
+      d = ((d >> 1) | (d << 27));
+    }
+    else
+    {
+      c = ((c >> 2) | (c << 26));
+      d = ((d >> 2) | (d << 26));
+    }
+
+    c = c & 0x0fffffff;
+    d = d & 0x0fffffff;
+
+    const u32 c00 = (c >>  0) & 0x0000003f;
+    const u32 c06 = (c >>  6) & 0x00383003;
+    const u32 c07 = (c >>  7) & 0x0000003c;
+    const u32 c13 = (c >> 13) & 0x0000060f;
+    const u32 c20 = (c >> 20) & 0x00000001;
+
+    u32 s = BOX (((c00 >>  0) & 0xff), 0, s_skb)
+           | BOX (((c06 >>  0) & 0xff)
+                 |((c07 >>  0) & 0xff), 1, s_skb)
+           | BOX (((c13 >>  0) & 0xff)
+                 |((c06 >>  8) & 0xff), 2, s_skb)
+           | BOX (((c20 >>  0) & 0xff)
+                 |((c13 >>  8) & 0xff)
+                 |((c06 >> 16) & 0xff), 3, s_skb);
+
+    const u32 d00 = (d >>  0) & 0x00003c3f;
+    const u32 d07 = (d >>  7) & 0x00003f03;
+    const u32 d21 = (d >> 21) & 0x0000000f;
+    const u32 d22 = (d >> 22) & 0x00000030;
+
+    u32 t = BOX (((d00 >>  0) & 0xff), 4, s_skb)
+           | BOX (((d07 >>  0) & 0xff)
+                 |((d00 >>  8) & 0xff), 5, s_skb)
+           | BOX (((d07 >>  8) & 0xff), 6, s_skb)
+           | BOX (((d21 >>  0) & 0xff)
+                 |((d22 >>  0) & 0xff), 7, s_skb);
+
+    Kc[i] = ((t << 16) | (s & 0x0000ffff));
+    Kd[i] = ((s >> 16) | (t & 0xffff0000));
+  }
+}
+
+DECLSPEC void transform_netntlmv1_key (const u32 w0, const u32 w1, u32 *out)
+{
+  u32 t[8];
+
+  t[0] = (w0 >>  0) & 0xff;
+  t[1] = (w0 >>  8) & 0xff;
+  t[2] = (w0 >> 16) & 0xff;
+  t[3] = (w0 >> 24) & 0xff;
+  t[4] = (w1 >>  0) & 0xff;
+  t[5] = (w1 >>  8) & 0xff;
+  t[6] = (w1 >> 16) & 0xff;
+  t[7] = (w1 >> 24) & 0xff;
+
+  u32 k[8];
+
+  k[0] =               (t[0] >> 0);
+  k[1] = (t[0] << 7) | (t[1] >> 1);
+  k[2] = (t[1] << 6) | (t[2] >> 2);
+  k[3] = (t[2] << 5) | (t[3] >> 3);
+  k[4] = (t[3] << 4) | (t[4] >> 4);
+  k[5] = (t[4] << 3) | (t[5] >> 5);
+  k[6] = (t[5] << 2) | (t[6] >> 6);
+  k[7] = (t[6] << 1);
+
+  out[0] = ((k[0] & 0xff) <<  0)
+         | ((k[1] & 0xff) <<  8)
+         | ((k[2] & 0xff) << 16)
+         | ((k[3] & 0xff) << 24);
+
+  out[1] = ((k[4] & 0xff) <<  0)
+         | ((k[5] & 0xff) <<  8)
+         | ((k[6] & 0xff) << 16)
+         | ((k[7] & 0xff) << 24);
+}
+
+#ifdef KERNEL_STATIC
+DECLSPEC u8 hex_convert (const u8 c)
+{
+  return (c & 15) + (c >> 6) * 9;
+}
+
+DECLSPEC u8 hex_to_u8 (const u8 *hex)
+{
+  u8 v = 0;
+
+  v |= ((u8) hex_convert (hex[1]) << 0);
+  v |= ((u8) hex_convert (hex[0]) << 4);
+
+  return (v);
+}
+#endif
+
+typedef struct netntlm
+{
+  u32 user_len;
+  u32 domain_len;
+  u32 srvchall_len;
+  u32 clichall_len;
+
+  u32 userdomain_buf[64];
+  u32 chall_buf[256];
+
+} netntlm_t;
+
+typedef struct netntlm_tmp
+{
+  u32 digest_buf[4];
+
+} netntlm_tmp_t;
+
+KERNEL_FQ void m27000_init (KERN_ATTR_TMPS_ESALT (netntlm_tmp_t, netntlm_t))
+{
+  /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  if (gid >= gid_max) return;
+
+  /**
+   * salt
+   */
+
+  u32 in[16];
+
+  in[ 0] = pws[gid].i[ 0];
+  in[ 1] = pws[gid].i[ 1];
+  in[ 2] = pws[gid].i[ 2];
+  in[ 3] = pws[gid].i[ 3];
+  in[ 4] = pws[gid].i[ 4];
+  in[ 5] = pws[gid].i[ 5];
+  in[ 6] = pws[gid].i[ 6];
+  in[ 7] = pws[gid].i[ 7];
+
+  u8 *in_ptr = (u8 *) in;
+
+  u32 out[4];
+
+  u8 *out_ptr = (u8 *) out;
+
+  for (int i = 0, j = 0; i < 16; i += 1, j += 2)
+  {
+    out_ptr[i] = hex_to_u8 (in_ptr + j);
+  }
+
+  tmps[gid].digest_buf[0] = out[ 0];
+  tmps[gid].digest_buf[1] = out[ 1];
+  tmps[gid].digest_buf[2] = out[ 2];
+  tmps[gid].digest_buf[3] = out[ 3];
+
+}
+
+KERNEL_FQ void m27000_loop (KERN_ATTR_TMPS_ESALT (netntlm_tmp_t, netntlm_t))
+{
+
+}
+
+KERNEL_FQ void m27000_comp (KERN_ATTR_TMPS_ESALT (netntlm_tmp_t, netntlm_t))
+{
+   /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * sbox, kbox
+   */
+
+  #ifdef REAL_SHM
+
+  LOCAL_VK u32 s_SPtrans[8][64];
+  LOCAL_VK u32 s_skb[8][64];
+
+  for (u32 i = lid; i < 64; i += lsz)
+  {
+    s_SPtrans[0][i] = c_SPtrans[0][i];
+    s_SPtrans[1][i] = c_SPtrans[1][i];
+    s_SPtrans[2][i] = c_SPtrans[2][i];
+    s_SPtrans[3][i] = c_SPtrans[3][i];
+    s_SPtrans[4][i] = c_SPtrans[4][i];
+    s_SPtrans[5][i] = c_SPtrans[5][i];
+    s_SPtrans[6][i] = c_SPtrans[6][i];
+    s_SPtrans[7][i] = c_SPtrans[7][i];
+
+    s_skb[0][i] = c_skb[0][i];
+    s_skb[1][i] = c_skb[1][i];
+    s_skb[2][i] = c_skb[2][i];
+    s_skb[3][i] = c_skb[3][i];
+    s_skb[4][i] = c_skb[4][i];
+    s_skb[5][i] = c_skb[5][i];
+    s_skb[6][i] = c_skb[6][i];
+    s_skb[7][i] = c_skb[7][i];
+  }
+
+  SYNC_THREADS ();
+
+  #else
+
+  CONSTANT_AS u32a (*s_SPtrans)[64] = c_SPtrans;
+  CONSTANT_AS u32a (*s_skb)[64]     = c_skb;
+
+  #endif
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  const u32 s0 = salt_bufs[SALT_POS].salt_buf[0];
+  const u32 s1 = salt_bufs[SALT_POS].salt_buf[1];
+  const u32 s2 = salt_bufs[SALT_POS].salt_buf[2];
+
+  const u32 a = tmps[gid].digest_buf[0];
+  const u32 b = tmps[gid].digest_buf[1];
+  const u32 c = tmps[gid].digest_buf[2];
+  const u32 d = tmps[gid].digest_buf[3];
+
+  // I believe this matches the last 2 bytes and throws away.
+  // Taken from 5500. 
+  if ((d >> 16) != s2) return;
+
+  /**
+    * DES1
+    */
+
+  u32 key[2];
+
+  transform_netntlmv1_key (a, b, key);
+
+  u32 Kc[16];
+  u32 Kd[16];
+
+  _des_crypt_keysetup (key[0], key[1], Kc, Kd, s_skb);
+
+  u32 data[2];
+
+  data[0] = s0;
+  data[1] = s1;
+
+  u32 out1[2];
+
+  _des_crypt_encrypt (out1, data, Kc, Kd, s_SPtrans);
+
+  /**
+    * DES2
+    */
+
+  transform_netntlmv1_key (((b >> 24) | (c << 8)), ((c >> 24) | (d << 8)), key);
+
+  _des_crypt_keysetup (key[0], key[1], Kc, Kd, s_skb);
+
+  u32 out2[2];
+
+  _des_crypt_encrypt (out2, data, Kc, Kd, s_SPtrans);
+
+  /**
+   * digest
+   */
+
+  const u32 r0 = out1[0];
+  const u32 r1 = out1[1];
+  const u32 r2 = out2[0];
+  const u32 r3 = out2[1];
+
+  #define il_pos 0
+
+  #ifdef KERNEL_STATIC
+  #include COMPARE_M
+  #endif
+}
\ No newline at end of file
diff --git a/OpenCL/m27100-pure.cl b/OpenCL/m27100-pure.cl
new file mode 100644
index 000000000..d34355187
--- /dev/null
+++ b/OpenCL/m27100-pure.cl
@@ -0,0 +1,197 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+// #define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include "inc_vendor.h"
+#include "inc_types.h"
+#include "inc_platform.cl"
+#include "inc_common.cl"
+#include "inc_rp.h"
+#include "inc_rp.cl"
+#include "inc_scalar.cl"
+#include "inc_hash_md4.cl"
+#include "inc_hash_md5.cl"
+#endif
+
+#define COMPARE_S "inc_comp_single.cl"
+#define COMPARE_M "inc_comp_multi.cl"
+
+#ifdef KERNEL_STATIC
+DECLSPEC u8 hex_convert (const u8 c)
+{
+  return (c & 15) + (c >> 6) * 9;
+}
+
+DECLSPEC u8 hex_to_u8 (const u8 *hex)
+{
+  u8 v = 0;
+
+  v |= ((u8) hex_convert (hex[1]) << 0);
+  v |= ((u8) hex_convert (hex[0]) << 4);
+
+  return (v);
+}
+#endif
+
+typedef struct netntlm
+{
+  u32 user_len;
+  u32 domain_len;
+  u32 srvchall_len;
+  u32 clichall_len;
+
+  u32 userdomain_buf[64];
+  u32 chall_buf[256];
+
+} netntlm_t;
+
+typedef struct netntlmv2_tmp
+{
+  u32 digest_buf[4];
+
+} netntlm_tmp_t;
+
+
+KERNEL_FQ void m27100_init (KERN_ATTR_TMPS_ESALT (netntlm_tmp_t, netntlm_t))
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  u32 in[16];
+
+  in[ 0] = pws[gid].i[ 0];
+  in[ 1] = pws[gid].i[ 1];
+  in[ 2] = pws[gid].i[ 2];
+  in[ 3] = pws[gid].i[ 3];
+  in[ 4] = pws[gid].i[ 4];
+  in[ 5] = pws[gid].i[ 5];
+  in[ 6] = pws[gid].i[ 6];
+  in[ 7] = pws[gid].i[ 7];
+
+  u8 *in_ptr = (u8 *) in;
+
+  u32 out[4];
+
+  u8 *out_ptr = (u8 *) out;
+
+  for (int i = 0, j = 0; i < 16; i += 1, j += 2)
+  {
+    out_ptr[i] = hex_to_u8 (in_ptr + j);
+  }
+
+  tmps[gid].digest_buf[0] = out[ 0];
+  tmps[gid].digest_buf[1] = out[ 1];
+  tmps[gid].digest_buf[2] = out[ 2];
+  tmps[gid].digest_buf[3] = out[ 3];
+
+}
+
+
+KERNEL_FQ void m27100_loop (KERN_ATTR_TMPS_ESALT (netntlm_tmp_t, netntlm_t))
+{
+
+
+}
+
+KERNEL_FQ void m27100_comp (KERN_ATTR_TMPS_ESALT (netntlm_tmp_t, netntlm_t))
+{
+   /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  const u64 lid = get_local_id (0);
+
+  u32 w0[4];
+  u32 w1[4];
+  u32 w2[4];
+  u32 w3[4];
+
+  w0[0] = tmps[gid].digest_buf[0];
+  w0[1] = tmps[gid].digest_buf[1];
+  w0[2] = tmps[gid].digest_buf[2];
+  w0[3] = tmps[gid].digest_buf[3];
+  w1[0] = 0;
+  w1[1] = 0;
+  w1[2] = 0;
+  w1[3] = 0;
+  w2[0] = 0;
+  w2[1] = 0;
+  w2[2] = 0;
+  w2[3] = 0;
+  w3[0] = 0;
+  w3[1] = 0;
+  w3[2] = 0;
+  w3[3] = 0;
+
+  md5_hmac_ctx_t ctx0;
+
+  md5_hmac_init_64 (&ctx0, w0, w1, w2, w3);
+
+  md5_hmac_update_global (&ctx0, esalt_bufs[DIGESTS_OFFSET].userdomain_buf, esalt_bufs[DIGESTS_OFFSET].user_len + esalt_bufs[DIGESTS_OFFSET].domain_len);
+
+  md5_hmac_final (&ctx0);
+
+  w0[0] = ctx0.opad.h[0];
+  w0[1] = ctx0.opad.h[1];
+  w0[2] = ctx0.opad.h[2];
+  w0[3] = ctx0.opad.h[3];
+  w1[0] = 0;
+  w1[1] = 0;
+  w1[2] = 0;
+  w1[3] = 0;
+  w2[0] = 0;
+  w2[1] = 0;
+  w2[2] = 0;
+  w2[3] = 0;
+  w3[0] = 0;
+  w3[1] = 0;
+  w3[2] = 0;
+  w3[3] = 0;
+
+  md5_hmac_ctx_t ctx;
+
+  md5_hmac_init_64 (&ctx, w0, w1, w2, w3);
+
+  md5_hmac_update_global (&ctx, esalt_bufs[DIGESTS_OFFSET].chall_buf, esalt_bufs[DIGESTS_OFFSET].srvchall_len + esalt_bufs[DIGESTS_OFFSET].clichall_len);
+
+  md5_hmac_final (&ctx);
+
+  tmps[gid].digest_buf[0] = ctx.opad.h[0];
+  tmps[gid].digest_buf[1] = ctx.opad.h[1];
+  tmps[gid].digest_buf[2] = ctx.opad.h[2];
+  tmps[gid].digest_buf[3] = ctx.opad.h[3];
+
+
+  /**
+   * digest
+   */
+
+  const u32 r0 = ctx.opad.h[DGST_R0];
+  const u32 r1 = ctx.opad.h[DGST_R1];
+  const u32 r2 = ctx.opad.h[DGST_R2];
+  const u32 r3 = ctx.opad.h[DGST_R3];
+
+  #define il_pos 0
+
+  #ifdef KERNEL_STATIC
+  #include COMPARE_M
+  #endif
+}
\ No newline at end of file
diff --git a/OpenCL/m27200_a0-optimized.cl b/OpenCL/m27200_a0-optimized.cl
new file mode 100644
index 000000000..a611b25c9
--- /dev/null
+++ b/OpenCL/m27200_a0-optimized.cl
@@ -0,0 +1,513 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include "inc_vendor.h"
+#include "inc_types.h"
+#include "inc_platform.cl"
+#include "inc_common.cl"
+#include "inc_rp_optimized.h"
+#include "inc_rp_optimized.cl"
+#include "inc_simd.cl"
+#include "inc_hash_sha1.cl"
+#endif
+
+KERNEL_FQ void m27200_m04 (KERN_ATTR_RULES ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+
+  /**
+   * base
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  u32 pw_buf0[4];
+  u32 pw_buf1[4];
+
+  pw_buf0[0] = pws[gid].i[0];
+  pw_buf0[1] = pws[gid].i[1];
+  pw_buf0[2] = pws[gid].i[2];
+  pw_buf0[3] = pws[gid].i[3];
+  pw_buf1[0] = pws[gid].i[4];
+  pw_buf1[1] = pws[gid].i[5];
+  pw_buf1[2] = pws[gid].i[6];
+  pw_buf1[3] = pws[gid].i[7];
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * salt
+   */
+
+  u32 salt_buf0[4] = { 0x2d2d, 0, 0, 0 };
+  u32 salt_buf1[4];
+  u32 salt_buf2[4];
+  u32 salt_buf3[4] = { 0 };
+
+  salt_buf0[0] |= salt_bufs[SALT_POS].salt_buf[ 0] << 16;
+  salt_buf0[1] = salt_bufs[SALT_POS].salt_buf[ 0] >> 16 | salt_bufs[SALT_POS].salt_buf[ 1] << 16;
+  salt_buf0[2] = salt_bufs[SALT_POS].salt_buf[ 1] >> 16 | salt_bufs[SALT_POS].salt_buf[ 2] << 16;
+  salt_buf0[3] = salt_bufs[SALT_POS].salt_buf[ 2] >> 16 | salt_bufs[SALT_POS].salt_buf[ 3] << 16;
+  salt_buf1[0] = salt_bufs[SALT_POS].salt_buf[ 3] >> 16 | salt_bufs[SALT_POS].salt_buf[ 4] << 16;
+  salt_buf1[1] = salt_bufs[SALT_POS].salt_buf[ 4] >> 16 | salt_bufs[SALT_POS].salt_buf[ 5] << 16;
+  salt_buf1[2] = salt_bufs[SALT_POS].salt_buf[ 5] >> 16 | salt_bufs[SALT_POS].salt_buf[ 6] << 16;
+  salt_buf1[3] = salt_bufs[SALT_POS].salt_buf[ 6] >> 16 | salt_bufs[SALT_POS].salt_buf[ 7] << 16;
+  salt_buf2[0] = salt_bufs[SALT_POS].salt_buf[ 7] >> 16 | salt_bufs[SALT_POS].salt_buf[ 8] << 16;
+  salt_buf2[1] = salt_bufs[SALT_POS].salt_buf[ 8] >> 16 | salt_bufs[SALT_POS].salt_buf[ 9] << 16;
+  salt_buf2[2] = salt_bufs[SALT_POS].salt_buf[ 9] >> 16 | 0x2d2d0000;
+  salt_buf2[3] = 0;
+
+  const u32 salt_len = 44; //salt_bufs[SALT_POS].salt_len;
+
+  /**
+   * digest
+   */
+
+  const u32 search[4] =
+  {
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R0],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R1],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R2],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R3]
+  };
+
+  /**
+   * loop
+   */
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
+  {
+    u32x w0[4] = { 0 };
+    u32x w1[4] = { 0 };
+    u32x w2[4] = { 0 };
+    u32x w3[4] = { 0 };
+
+    const u32x out_len = apply_rules_vect_optimized (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
+
+    /**
+     * prepend salt
+     */
+
+    const u32x out_salt_len = out_len + salt_len;
+
+    switch_buffer_by_offset_le_VV (w0, w1, w2, w3, salt_len);
+
+    w0[0] |= salt_buf0[0];
+    w0[1] |= salt_buf0[1];
+    w0[2] |= salt_buf0[2];
+    w0[3] |= salt_buf0[3];
+    w1[0] |= salt_buf1[0];
+    w1[1] |= salt_buf1[1];
+    w1[2] |= salt_buf1[2];
+    w1[3] |= salt_buf1[3];
+    w2[0] |= salt_buf2[0];
+    w2[1] |= salt_buf2[1];
+    w2[2] |= salt_buf2[2];
+    w2[3] |= salt_buf2[3];
+    w3[0] |= salt_buf3[0];
+    w3[1] |= salt_buf3[1];
+    w3[2] |= salt_buf3[2];
+    w3[3] |= salt_buf3[3];
+
+    append_0x2d_4x4_VV (w0, w1, w2, w3, out_salt_len);
+    append_0x2d_4x4_VV (w0, w1, w2, w3, out_salt_len + 1);
+    append_0x80_4x4_VV (w0, w1, w2, w3, out_salt_len + 2);
+
+    /**
+     * sha1
+     */
+
+    u32x w0_t = hc_swap32 (w0[0]);
+    u32x w1_t = hc_swap32 (w0[1]);
+    u32x w2_t = hc_swap32 (w0[2]);
+    u32x w3_t = hc_swap32 (w0[3]);
+    u32x w4_t = hc_swap32 (w1[0]);
+    u32x w5_t = hc_swap32 (w1[1]);
+    u32x w6_t = hc_swap32 (w1[2]);
+    u32x w7_t = hc_swap32 (w1[3]);
+    u32x w8_t = hc_swap32 (w2[0]);
+    u32x w9_t = hc_swap32 (w2[1]);
+    u32x wa_t = hc_swap32 (w2[2]);
+    u32x wb_t = hc_swap32 (w2[3]);
+    u32x wc_t = hc_swap32 (w3[0]);
+    u32x wd_t = hc_swap32 (w3[1]);
+    u32x we_t = 0;
+    u32x wf_t = (out_salt_len + 2) * 8;
+
+    u32x a = SHA1M_A;
+    u32x b = SHA1M_B;
+    u32x c = SHA1M_C;
+    u32x d = SHA1M_D;
+    u32x e = SHA1M_E;
+
+    #undef K
+    #define K SHA1C00
+
+    SHA1_STEP (SHA1_F0o, a, b, c, d, e, w0_t);
+    SHA1_STEP (SHA1_F0o, e, a, b, c, d, w1_t);
+    SHA1_STEP (SHA1_F0o, d, e, a, b, c, w2_t);
+    SHA1_STEP (SHA1_F0o, c, d, e, a, b, w3_t);
+    SHA1_STEP (SHA1_F0o, b, c, d, e, a, w4_t);
+    SHA1_STEP (SHA1_F0o, a, b, c, d, e, w5_t);
+    SHA1_STEP (SHA1_F0o, e, a, b, c, d, w6_t);
+    SHA1_STEP (SHA1_F0o, d, e, a, b, c, w7_t);
+    SHA1_STEP (SHA1_F0o, c, d, e, a, b, w8_t);
+    SHA1_STEP (SHA1_F0o, b, c, d, e, a, w9_t);
+    SHA1_STEP (SHA1_F0o, a, b, c, d, e, wa_t);
+    SHA1_STEP (SHA1_F0o, e, a, b, c, d, wb_t);
+    SHA1_STEP (SHA1_F0o, d, e, a, b, c, wc_t);
+    SHA1_STEP (SHA1_F0o, c, d, e, a, b, wd_t);
+    SHA1_STEP (SHA1_F0o, b, c, d, e, a, we_t);
+    SHA1_STEP (SHA1_F0o, a, b, c, d, e, wf_t);
+    w0_t = hc_rotl32 ((wd_t ^ w8_t ^ w2_t ^ w0_t), 1u); SHA1_STEP (SHA1_F0o, e, a, b, c, d, w0_t);
+    w1_t = hc_rotl32 ((we_t ^ w9_t ^ w3_t ^ w1_t), 1u); SHA1_STEP (SHA1_F0o, d, e, a, b, c, w1_t);
+    w2_t = hc_rotl32 ((wf_t ^ wa_t ^ w4_t ^ w2_t), 1u); SHA1_STEP (SHA1_F0o, c, d, e, a, b, w2_t);
+    w3_t = hc_rotl32 ((w0_t ^ wb_t ^ w5_t ^ w3_t), 1u); SHA1_STEP (SHA1_F0o, b, c, d, e, a, w3_t);
+
+    #undef K
+    #define K SHA1C01
+
+    w4_t = hc_rotl32 ((w1_t ^ wc_t ^ w6_t ^ w4_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w4_t);
+    w5_t = hc_rotl32 ((w2_t ^ wd_t ^ w7_t ^ w5_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w5_t);
+    w6_t = hc_rotl32 ((w3_t ^ we_t ^ w8_t ^ w6_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w6_t);
+    w7_t = hc_rotl32 ((w4_t ^ wf_t ^ w9_t ^ w7_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w7_t);
+    w8_t = hc_rotl32 ((w5_t ^ w0_t ^ wa_t ^ w8_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w8_t);
+    w9_t = hc_rotl32 ((w6_t ^ w1_t ^ wb_t ^ w9_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w9_t);
+    wa_t = hc_rotl32 ((w7_t ^ w2_t ^ wc_t ^ wa_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, wa_t);
+    wb_t = hc_rotl32 ((w8_t ^ w3_t ^ wd_t ^ wb_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, wb_t);
+    wc_t = hc_rotl32 ((w9_t ^ w4_t ^ we_t ^ wc_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, wc_t);
+    wd_t = hc_rotl32 ((wa_t ^ w5_t ^ wf_t ^ wd_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, wd_t);
+    we_t = hc_rotl32 ((wb_t ^ w6_t ^ w0_t ^ we_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, we_t);
+    wf_t = hc_rotl32 ((wc_t ^ w7_t ^ w1_t ^ wf_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, wf_t);
+    w0_t = hc_rotl32 ((wd_t ^ w8_t ^ w2_t ^ w0_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w0_t);
+    w1_t = hc_rotl32 ((we_t ^ w9_t ^ w3_t ^ w1_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w1_t);
+    w2_t = hc_rotl32 ((wf_t ^ wa_t ^ w4_t ^ w2_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w2_t);
+    w3_t = hc_rotl32 ((w0_t ^ wb_t ^ w5_t ^ w3_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w3_t);
+    w4_t = hc_rotl32 ((w1_t ^ wc_t ^ w6_t ^ w4_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w4_t);
+    w5_t = hc_rotl32 ((w2_t ^ wd_t ^ w7_t ^ w5_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w5_t);
+    w6_t = hc_rotl32 ((w3_t ^ we_t ^ w8_t ^ w6_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w6_t);
+    w7_t = hc_rotl32 ((w4_t ^ wf_t ^ w9_t ^ w7_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w7_t);
+
+    #undef K
+    #define K SHA1C02
+
+    w8_t = hc_rotl32 ((w5_t ^ w0_t ^ wa_t ^ w8_t), 1u); SHA1_STEP (SHA1_F2o, a, b, c, d, e, w8_t);
+    w9_t = hc_rotl32 ((w6_t ^ w1_t ^ wb_t ^ w9_t), 1u); SHA1_STEP (SHA1_F2o, e, a, b, c, d, w9_t);
+    wa_t = hc_rotl32 ((w7_t ^ w2_t ^ wc_t ^ wa_t), 1u); SHA1_STEP (SHA1_F2o, d, e, a, b, c, wa_t);
+    wb_t = hc_rotl32 ((w8_t ^ w3_t ^ wd_t ^ wb_t), 1u); SHA1_STEP (SHA1_F2o, c, d, e, a, b, wb_t);
+    wc_t = hc_rotl32 ((w9_t ^ w4_t ^ we_t ^ wc_t), 1u); SHA1_STEP (SHA1_F2o, b, c, d, e, a, wc_t);
+    wd_t = hc_rotl32 ((wa_t ^ w5_t ^ wf_t ^ wd_t), 1u); SHA1_STEP (SHA1_F2o, a, b, c, d, e, wd_t);
+    we_t = hc_rotl32 ((wb_t ^ w6_t ^ w0_t ^ we_t), 1u); SHA1_STEP (SHA1_F2o, e, a, b, c, d, we_t);
+    wf_t = hc_rotl32 ((wc_t ^ w7_t ^ w1_t ^ wf_t), 1u); SHA1_STEP (SHA1_F2o, d, e, a, b, c, wf_t);
+    w0_t = hc_rotl32 ((wd_t ^ w8_t ^ w2_t ^ w0_t), 1u); SHA1_STEP (SHA1_F2o, c, d, e, a, b, w0_t);
+    w1_t = hc_rotl32 ((we_t ^ w9_t ^ w3_t ^ w1_t), 1u); SHA1_STEP (SHA1_F2o, b, c, d, e, a, w1_t);
+    w2_t = hc_rotl32 ((wf_t ^ wa_t ^ w4_t ^ w2_t), 1u); SHA1_STEP (SHA1_F2o, a, b, c, d, e, w2_t);
+    w3_t = hc_rotl32 ((w0_t ^ wb_t ^ w5_t ^ w3_t), 1u); SHA1_STEP (SHA1_F2o, e, a, b, c, d, w3_t);
+    w4_t = hc_rotl32 ((w1_t ^ wc_t ^ w6_t ^ w4_t), 1u); SHA1_STEP (SHA1_F2o, d, e, a, b, c, w4_t);
+    w5_t = hc_rotl32 ((w2_t ^ wd_t ^ w7_t ^ w5_t), 1u); SHA1_STEP (SHA1_F2o, c, d, e, a, b, w5_t);
+    w6_t = hc_rotl32 ((w3_t ^ we_t ^ w8_t ^ w6_t), 1u); SHA1_STEP (SHA1_F2o, b, c, d, e, a, w6_t);
+    w7_t = hc_rotl32 ((w4_t ^ wf_t ^ w9_t ^ w7_t), 1u); SHA1_STEP (SHA1_F2o, a, b, c, d, e, w7_t);
+    w8_t = hc_rotl32 ((w5_t ^ w0_t ^ wa_t ^ w8_t), 1u); SHA1_STEP (SHA1_F2o, e, a, b, c, d, w8_t);
+    w9_t = hc_rotl32 ((w6_t ^ w1_t ^ wb_t ^ w9_t), 1u); SHA1_STEP (SHA1_F2o, d, e, a, b, c, w9_t);
+    wa_t = hc_rotl32 ((w7_t ^ w2_t ^ wc_t ^ wa_t), 1u); SHA1_STEP (SHA1_F2o, c, d, e, a, b, wa_t);
+    wb_t = hc_rotl32 ((w8_t ^ w3_t ^ wd_t ^ wb_t), 1u); SHA1_STEP (SHA1_F2o, b, c, d, e, a, wb_t);
+
+    #undef K
+    #define K SHA1C03
+
+    wc_t = hc_rotl32 ((w9_t ^ w4_t ^ we_t ^ wc_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, wc_t);
+    wd_t = hc_rotl32 ((wa_t ^ w5_t ^ wf_t ^ wd_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, wd_t);
+    we_t = hc_rotl32 ((wb_t ^ w6_t ^ w0_t ^ we_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, we_t);
+    wf_t = hc_rotl32 ((wc_t ^ w7_t ^ w1_t ^ wf_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, wf_t);
+    w0_t = hc_rotl32 ((wd_t ^ w8_t ^ w2_t ^ w0_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w0_t);
+    w1_t = hc_rotl32 ((we_t ^ w9_t ^ w3_t ^ w1_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w1_t);
+    w2_t = hc_rotl32 ((wf_t ^ wa_t ^ w4_t ^ w2_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w2_t);
+    w3_t = hc_rotl32 ((w0_t ^ wb_t ^ w5_t ^ w3_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w3_t);
+    w4_t = hc_rotl32 ((w1_t ^ wc_t ^ w6_t ^ w4_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w4_t);
+    w5_t = hc_rotl32 ((w2_t ^ wd_t ^ w7_t ^ w5_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w5_t);
+    w6_t = hc_rotl32 ((w3_t ^ we_t ^ w8_t ^ w6_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w6_t);
+    w7_t = hc_rotl32 ((w4_t ^ wf_t ^ w9_t ^ w7_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w7_t);
+    w8_t = hc_rotl32 ((w5_t ^ w0_t ^ wa_t ^ w8_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w8_t);
+    w9_t = hc_rotl32 ((w6_t ^ w1_t ^ wb_t ^ w9_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w9_t);
+    wa_t = hc_rotl32 ((w7_t ^ w2_t ^ wc_t ^ wa_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, wa_t);
+    wb_t = hc_rotl32 ((w8_t ^ w3_t ^ wd_t ^ wb_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, wb_t);
+    wc_t = hc_rotl32 ((w9_t ^ w4_t ^ we_t ^ wc_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, wc_t);
+    wd_t = hc_rotl32 ((wa_t ^ w5_t ^ wf_t ^ wd_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, wd_t);
+    we_t = hc_rotl32 ((wb_t ^ w6_t ^ w0_t ^ we_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, we_t);
+    wf_t = hc_rotl32 ((wc_t ^ w7_t ^ w1_t ^ wf_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, wf_t);
+
+    COMPARE_M_SIMD (d, e, c, b);
+  }
+}
+
+KERNEL_FQ void m27200_m08 (KERN_ATTR_RULES ())
+{
+}
+
+KERNEL_FQ void m27200_m16 (KERN_ATTR_RULES ())
+{
+}
+
+KERNEL_FQ void m27200_s04 (KERN_ATTR_RULES ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+
+  /**
+   * base
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  u32 pw_buf0[4];
+  u32 pw_buf1[4];
+
+  pw_buf0[0] = pws[gid].i[0];
+  pw_buf0[1] = pws[gid].i[1];
+  pw_buf0[2] = pws[gid].i[2];
+  pw_buf0[3] = pws[gid].i[3];
+  pw_buf1[0] = pws[gid].i[4];
+  pw_buf1[1] = pws[gid].i[5];
+  pw_buf1[2] = pws[gid].i[6];
+  pw_buf1[3] = pws[gid].i[7];
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * salt
+   */
+
+  u32 salt_buf0[4] = { 0x2d2d, 0, 0, 0 };
+  u32 salt_buf1[4];
+  u32 salt_buf2[4];
+  u32 salt_buf3[4] = { 0 };
+
+  salt_buf0[0] |= salt_bufs[SALT_POS].salt_buf[ 0] << 16;
+  salt_buf0[1] = salt_bufs[SALT_POS].salt_buf[ 0] >> 16 | salt_bufs[SALT_POS].salt_buf[ 1] << 16;
+  salt_buf0[2] = salt_bufs[SALT_POS].salt_buf[ 1] >> 16 | salt_bufs[SALT_POS].salt_buf[ 2] << 16;
+  salt_buf0[3] = salt_bufs[SALT_POS].salt_buf[ 2] >> 16 | salt_bufs[SALT_POS].salt_buf[ 3] << 16;
+  salt_buf1[0] = salt_bufs[SALT_POS].salt_buf[ 3] >> 16 | salt_bufs[SALT_POS].salt_buf[ 4] << 16;
+  salt_buf1[1] = salt_bufs[SALT_POS].salt_buf[ 4] >> 16 | salt_bufs[SALT_POS].salt_buf[ 5] << 16;
+  salt_buf1[2] = salt_bufs[SALT_POS].salt_buf[ 5] >> 16 | salt_bufs[SALT_POS].salt_buf[ 6] << 16;
+  salt_buf1[3] = salt_bufs[SALT_POS].salt_buf[ 6] >> 16 | salt_bufs[SALT_POS].salt_buf[ 7] << 16;
+  salt_buf2[0] = salt_bufs[SALT_POS].salt_buf[ 7] >> 16 | salt_bufs[SALT_POS].salt_buf[ 8] << 16;
+  salt_buf2[1] = salt_bufs[SALT_POS].salt_buf[ 8] >> 16 | salt_bufs[SALT_POS].salt_buf[ 9] << 16;
+  salt_buf2[2] = salt_bufs[SALT_POS].salt_buf[ 9] >> 16 | 0x2d2d0000;
+  salt_buf2[3] = 0;
+
+  const u32 salt_len = 44;
+
+  /**
+   * digest
+   */
+
+  const u32 search[4] =
+  {
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R0],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R1],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R2],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R3]
+  };
+
+  /**
+   * reverse
+   */
+
+  const u32 e_rev = hc_rotl32_S (search[1], 2u);
+
+  /**
+   * loop
+   */
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
+  {
+    u32x w0[4] = { 0 };
+    u32x w1[4] = { 0 };
+    u32x w2[4] = { 0 };
+    u32x w3[4] = { 0 };
+
+    const u32x out_len = apply_rules_vect_optimized (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
+
+    /**
+     * prepend salt
+     */
+
+    const u32x out_salt_len = out_len + salt_len;
+
+    switch_buffer_by_offset_le_VV (w0, w1, w2, w3, salt_len);
+
+    w0[0] |= salt_buf0[0];
+    w0[1] |= salt_buf0[1];
+    w0[2] |= salt_buf0[2];
+    w0[3] |= salt_buf0[3];
+    w1[0] |= salt_buf1[0];
+    w1[1] |= salt_buf1[1];
+    w1[2] |= salt_buf1[2];
+    w1[3] |= salt_buf1[3];
+    w2[0] |= salt_buf2[0];
+    w2[1] |= salt_buf2[1];
+    w2[2] |= salt_buf2[2];
+    w2[3] |= salt_buf2[3];
+    w3[0] |= salt_buf3[0];
+    w3[1] |= salt_buf3[1];
+    w3[2] |= salt_buf3[2];
+    w3[3] |= salt_buf3[3];
+
+    append_0x2d_4x4_VV (w0, w1, w2, w3, out_salt_len);
+    append_0x2d_4x4_VV (w0, w1, w2, w3, out_salt_len + 1);
+    append_0x80_4x4_VV (w0, w1, w2, w3, out_salt_len + 2);
+
+    /**
+     * sha1
+     */
+
+    u32x w0_t = hc_swap32 (w0[0]);
+    u32x w1_t = hc_swap32 (w0[1]);
+    u32x w2_t = hc_swap32 (w0[2]);
+    u32x w3_t = hc_swap32 (w0[3]);
+    u32x w4_t = hc_swap32 (w1[0]);
+    u32x w5_t = hc_swap32 (w1[1]);
+    u32x w6_t = hc_swap32 (w1[2]);
+    u32x w7_t = hc_swap32 (w1[3]);
+    u32x w8_t = hc_swap32 (w2[0]);
+    u32x w9_t = hc_swap32 (w2[1]);
+    u32x wa_t = hc_swap32 (w2[2]);
+    u32x wb_t = hc_swap32 (w2[3]);
+    u32x wc_t = hc_swap32 (w3[0]);
+    u32x wd_t = hc_swap32 (w3[1]);
+    u32x we_t = 0;
+    u32x wf_t = (out_salt_len + 2) * 8;
+
+
+    u32x a = SHA1M_A;
+    u32x b = SHA1M_B;
+    u32x c = SHA1M_C;
+    u32x d = SHA1M_D;
+    u32x e = SHA1M_E;
+
+    #undef K
+    #define K SHA1C00
+
+    SHA1_STEP (SHA1_F0o, a, b, c, d, e, w0_t);
+    SHA1_STEP (SHA1_F0o, e, a, b, c, d, w1_t);
+    SHA1_STEP (SHA1_F0o, d, e, a, b, c, w2_t);
+    SHA1_STEP (SHA1_F0o, c, d, e, a, b, w3_t);
+    SHA1_STEP (SHA1_F0o, b, c, d, e, a, w4_t);
+    SHA1_STEP (SHA1_F0o, a, b, c, d, e, w5_t);
+    SHA1_STEP (SHA1_F0o, e, a, b, c, d, w6_t);
+    SHA1_STEP (SHA1_F0o, d, e, a, b, c, w7_t);
+    SHA1_STEP (SHA1_F0o, c, d, e, a, b, w8_t);
+    SHA1_STEP (SHA1_F0o, b, c, d, e, a, w9_t);
+    SHA1_STEP (SHA1_F0o, a, b, c, d, e, wa_t);
+    SHA1_STEP (SHA1_F0o, e, a, b, c, d, wb_t);
+    SHA1_STEP (SHA1_F0o, d, e, a, b, c, wc_t);
+    SHA1_STEP (SHA1_F0o, c, d, e, a, b, wd_t);
+    SHA1_STEP (SHA1_F0o, b, c, d, e, a, we_t);
+    SHA1_STEP (SHA1_F0o, a, b, c, d, e, wf_t);
+    w0_t = hc_rotl32 ((wd_t ^ w8_t ^ w2_t ^ w0_t), 1u); SHA1_STEP (SHA1_F0o, e, a, b, c, d, w0_t);
+    w1_t = hc_rotl32 ((we_t ^ w9_t ^ w3_t ^ w1_t), 1u); SHA1_STEP (SHA1_F0o, d, e, a, b, c, w1_t);
+    w2_t = hc_rotl32 ((wf_t ^ wa_t ^ w4_t ^ w2_t), 1u); SHA1_STEP (SHA1_F0o, c, d, e, a, b, w2_t);
+    w3_t = hc_rotl32 ((w0_t ^ wb_t ^ w5_t ^ w3_t), 1u); SHA1_STEP (SHA1_F0o, b, c, d, e, a, w3_t);
+
+    #undef K
+    #define K SHA1C01
+
+    w4_t = hc_rotl32 ((w1_t ^ wc_t ^ w6_t ^ w4_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w4_t);
+    w5_t = hc_rotl32 ((w2_t ^ wd_t ^ w7_t ^ w5_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w5_t);
+    w6_t = hc_rotl32 ((w3_t ^ we_t ^ w8_t ^ w6_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w6_t);
+    w7_t = hc_rotl32 ((w4_t ^ wf_t ^ w9_t ^ w7_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w7_t);
+    w8_t = hc_rotl32 ((w5_t ^ w0_t ^ wa_t ^ w8_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w8_t);
+    w9_t = hc_rotl32 ((w6_t ^ w1_t ^ wb_t ^ w9_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w9_t);
+    wa_t = hc_rotl32 ((w7_t ^ w2_t ^ wc_t ^ wa_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, wa_t);
+    wb_t = hc_rotl32 ((w8_t ^ w3_t ^ wd_t ^ wb_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, wb_t);
+    wc_t = hc_rotl32 ((w9_t ^ w4_t ^ we_t ^ wc_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, wc_t);
+    wd_t = hc_rotl32 ((wa_t ^ w5_t ^ wf_t ^ wd_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, wd_t);
+    we_t = hc_rotl32 ((wb_t ^ w6_t ^ w0_t ^ we_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, we_t);
+    wf_t = hc_rotl32 ((wc_t ^ w7_t ^ w1_t ^ wf_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, wf_t);
+    w0_t = hc_rotl32 ((wd_t ^ w8_t ^ w2_t ^ w0_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w0_t);
+    w1_t = hc_rotl32 ((we_t ^ w9_t ^ w3_t ^ w1_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w1_t);
+    w2_t = hc_rotl32 ((wf_t ^ wa_t ^ w4_t ^ w2_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w2_t);
+    w3_t = hc_rotl32 ((w0_t ^ wb_t ^ w5_t ^ w3_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w3_t);
+    w4_t = hc_rotl32 ((w1_t ^ wc_t ^ w6_t ^ w4_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w4_t);
+    w5_t = hc_rotl32 ((w2_t ^ wd_t ^ w7_t ^ w5_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w5_t);
+    w6_t = hc_rotl32 ((w3_t ^ we_t ^ w8_t ^ w6_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w6_t);
+    w7_t = hc_rotl32 ((w4_t ^ wf_t ^ w9_t ^ w7_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w7_t);
+
+    #undef K
+    #define K SHA1C02
+
+    w8_t = hc_rotl32 ((w5_t ^ w0_t ^ wa_t ^ w8_t), 1u); SHA1_STEP (SHA1_F2o, a, b, c, d, e, w8_t);
+    w9_t = hc_rotl32 ((w6_t ^ w1_t ^ wb_t ^ w9_t), 1u); SHA1_STEP (SHA1_F2o, e, a, b, c, d, w9_t);
+    wa_t = hc_rotl32 ((w7_t ^ w2_t ^ wc_t ^ wa_t), 1u); SHA1_STEP (SHA1_F2o, d, e, a, b, c, wa_t);
+    wb_t = hc_rotl32 ((w8_t ^ w3_t ^ wd_t ^ wb_t), 1u); SHA1_STEP (SHA1_F2o, c, d, e, a, b, wb_t);
+    wc_t = hc_rotl32 ((w9_t ^ w4_t ^ we_t ^ wc_t), 1u); SHA1_STEP (SHA1_F2o, b, c, d, e, a, wc_t);
+    wd_t = hc_rotl32 ((wa_t ^ w5_t ^ wf_t ^ wd_t), 1u); SHA1_STEP (SHA1_F2o, a, b, c, d, e, wd_t);
+    we_t = hc_rotl32 ((wb_t ^ w6_t ^ w0_t ^ we_t), 1u); SHA1_STEP (SHA1_F2o, e, a, b, c, d, we_t);
+    wf_t = hc_rotl32 ((wc_t ^ w7_t ^ w1_t ^ wf_t), 1u); SHA1_STEP (SHA1_F2o, d, e, a, b, c, wf_t);
+    w0_t = hc_rotl32 ((wd_t ^ w8_t ^ w2_t ^ w0_t), 1u); SHA1_STEP (SHA1_F2o, c, d, e, a, b, w0_t);
+    w1_t = hc_rotl32 ((we_t ^ w9_t ^ w3_t ^ w1_t), 1u); SHA1_STEP (SHA1_F2o, b, c, d, e, a, w1_t);
+    w2_t = hc_rotl32 ((wf_t ^ wa_t ^ w4_t ^ w2_t), 1u); SHA1_STEP (SHA1_F2o, a, b, c, d, e, w2_t);
+    w3_t = hc_rotl32 ((w0_t ^ wb_t ^ w5_t ^ w3_t), 1u); SHA1_STEP (SHA1_F2o, e, a, b, c, d, w3_t);
+    w4_t = hc_rotl32 ((w1_t ^ wc_t ^ w6_t ^ w4_t), 1u); SHA1_STEP (SHA1_F2o, d, e, a, b, c, w4_t);
+    w5_t = hc_rotl32 ((w2_t ^ wd_t ^ w7_t ^ w5_t), 1u); SHA1_STEP (SHA1_F2o, c, d, e, a, b, w5_t);
+    w6_t = hc_rotl32 ((w3_t ^ we_t ^ w8_t ^ w6_t), 1u); SHA1_STEP (SHA1_F2o, b, c, d, e, a, w6_t);
+    w7_t = hc_rotl32 ((w4_t ^ wf_t ^ w9_t ^ w7_t), 1u); SHA1_STEP (SHA1_F2o, a, b, c, d, e, w7_t);
+    w8_t = hc_rotl32 ((w5_t ^ w0_t ^ wa_t ^ w8_t), 1u); SHA1_STEP (SHA1_F2o, e, a, b, c, d, w8_t);
+    w9_t = hc_rotl32 ((w6_t ^ w1_t ^ wb_t ^ w9_t), 1u); SHA1_STEP (SHA1_F2o, d, e, a, b, c, w9_t);
+    wa_t = hc_rotl32 ((w7_t ^ w2_t ^ wc_t ^ wa_t), 1u); SHA1_STEP (SHA1_F2o, c, d, e, a, b, wa_t);
+    wb_t = hc_rotl32 ((w8_t ^ w3_t ^ wd_t ^ wb_t), 1u); SHA1_STEP (SHA1_F2o, b, c, d, e, a, wb_t);
+
+    #undef K
+    #define K SHA1C03
+
+    wc_t = hc_rotl32 ((w9_t ^ w4_t ^ we_t ^ wc_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, wc_t);
+    wd_t = hc_rotl32 ((wa_t ^ w5_t ^ wf_t ^ wd_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, wd_t);
+    we_t = hc_rotl32 ((wb_t ^ w6_t ^ w0_t ^ we_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, we_t);
+    wf_t = hc_rotl32 ((wc_t ^ w7_t ^ w1_t ^ wf_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, wf_t);
+    w0_t = hc_rotl32 ((wd_t ^ w8_t ^ w2_t ^ w0_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w0_t);
+    w1_t = hc_rotl32 ((we_t ^ w9_t ^ w3_t ^ w1_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w1_t);
+    w2_t = hc_rotl32 ((wf_t ^ wa_t ^ w4_t ^ w2_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w2_t);
+    w3_t = hc_rotl32 ((w0_t ^ wb_t ^ w5_t ^ w3_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w3_t);
+    w4_t = hc_rotl32 ((w1_t ^ wc_t ^ w6_t ^ w4_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w4_t);
+    w5_t = hc_rotl32 ((w2_t ^ wd_t ^ w7_t ^ w5_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w5_t);
+    w6_t = hc_rotl32 ((w3_t ^ we_t ^ w8_t ^ w6_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w6_t);
+    w7_t = hc_rotl32 ((w4_t ^ wf_t ^ w9_t ^ w7_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w7_t);
+    w8_t = hc_rotl32 ((w5_t ^ w0_t ^ wa_t ^ w8_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w8_t);
+    w9_t = hc_rotl32 ((w6_t ^ w1_t ^ wb_t ^ w9_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w9_t);
+    wa_t = hc_rotl32 ((w7_t ^ w2_t ^ wc_t ^ wa_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, wa_t);
+    wb_t = hc_rotl32 ((w8_t ^ w3_t ^ wd_t ^ wb_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, wb_t);
+
+    if (MATCHES_NONE_VS (e, e_rev)) continue;
+
+    wc_t = hc_rotl32 ((w9_t ^ w4_t ^ we_t ^ wc_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, wc_t);
+    wd_t = hc_rotl32 ((wa_t ^ w5_t ^ wf_t ^ wd_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, wd_t);
+    we_t = hc_rotl32 ((wb_t ^ w6_t ^ w0_t ^ we_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, we_t);
+    wf_t = hc_rotl32 ((wc_t ^ w7_t ^ w1_t ^ wf_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, wf_t);
+
+    COMPARE_S_SIMD (d, e, c, b);
+  }
+}
+
+KERNEL_FQ void m27200_s08 (KERN_ATTR_RULES ())
+{
+}
+
+KERNEL_FQ void m27200_s16 (KERN_ATTR_RULES ())
+{
+}
diff --git a/OpenCL/m27200_a0-pure.cl b/OpenCL/m27200_a0-pure.cl
new file mode 100644
index 000000000..7b60e4440
--- /dev/null
+++ b/OpenCL/m27200_a0-pure.cl
@@ -0,0 +1,142 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+//#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include "inc_vendor.h"
+#include "inc_types.h"
+#include "inc_platform.cl"
+#include "inc_common.cl"
+#include "inc_rp.h"
+#include "inc_rp.cl"
+#include "inc_scalar.cl"
+#include "inc_hash_sha1.cl"
+#endif
+
+KERNEL_FQ void m27200_mxx (KERN_ATTR_RULES ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  const u32 dash[16] = { 0x2d2d0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
+
+  /**
+   * base
+   */
+
+  COPY_PW (pws[gid]);
+
+  sha1_ctx_t ctx0;
+
+  sha1_init (&ctx0);
+
+  ctx0.w0[0] = dash[0];
+  ctx0.w0[1] = dash[1];
+
+  ctx0.len = 2;
+  
+  sha1_update_global_swap (&ctx0, salt_bufs[SALT_POS].salt_buf, salt_bufs[SALT_POS].salt_len);
+  sha1_update (&ctx0, dash, 2);
+
+  /**
+   * loop
+   */
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos++)
+  {
+    pw_t tmp = PASTE_PW;
+
+    tmp.pw_len = apply_rules (rules_buf[il_pos].cmds, tmp.i, tmp.pw_len);
+
+    sha1_ctx_t ctx = ctx0;
+
+    sha1_update_swap (&ctx, tmp.i, tmp.pw_len);
+    sha1_update (&ctx, dash, 2);
+    sha1_final (&ctx);
+
+    const u32 r0 = ctx.h[DGST_R0];
+    const u32 r1 = ctx.h[DGST_R1];
+    const u32 r2 = ctx.h[DGST_R2];
+    const u32 r3 = ctx.h[DGST_R3];
+
+    COMPARE_M_SCALAR (r0, r1, r2, r3);
+  }
+}
+
+KERNEL_FQ void m27200_sxx (KERN_ATTR_RULES ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  const u32 dash[16] = { 0x2d2d0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
+
+  /**
+   * digest
+   */
+
+  const u32 search[4] =
+  {
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R0],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R1],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R2],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R3]
+  };
+
+  /**
+   * base
+   */
+
+  COPY_PW (pws[gid]);
+
+  sha1_ctx_t ctx0;
+
+  sha1_init (&ctx0);
+  
+  ctx0.w0[0] = dash[0];
+  ctx0.w0[1] = dash[1];
+
+  ctx0.len = 2;
+  
+  sha1_update_global_swap (&ctx0, salt_bufs[SALT_POS].salt_buf, salt_bufs[SALT_POS].salt_len);
+  sha1_update (&ctx0, dash, 2);
+
+  /**
+   * loop
+   */
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos++)
+  {
+    pw_t tmp = PASTE_PW;
+
+    tmp.pw_len = apply_rules (rules_buf[il_pos].cmds, tmp.i, tmp.pw_len);
+
+    sha1_ctx_t ctx = ctx0;
+
+    sha1_update_swap (&ctx, tmp.i, tmp.pw_len);
+    sha1_update (&ctx, dash, 2);
+
+    sha1_final (&ctx);
+
+    const u32 r0 = ctx.h[DGST_R0];
+    const u32 r1 = ctx.h[DGST_R1];
+    const u32 r2 = ctx.h[DGST_R2];
+    const u32 r3 = ctx.h[DGST_R3];
+
+    COMPARE_S_SCALAR (r0, r1, r2, r3);
+  }
+}
diff --git a/OpenCL/m27200_a1-optimized.cl b/OpenCL/m27200_a1-optimized.cl
new file mode 100644
index 000000000..6df369682
--- /dev/null
+++ b/OpenCL/m27200_a1-optimized.cl
@@ -0,0 +1,630 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include "inc_vendor.h"
+#include "inc_types.h"
+#include "inc_platform.cl"
+#include "inc_common.cl"
+#include "inc_simd.cl"
+#include "inc_hash_sha1.cl"
+#endif
+
+KERNEL_FQ void m27200_m04 (KERN_ATTR_BASIC ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+
+  /**
+   * base
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  u32 pw_buf0[4];
+  u32 pw_buf1[4];
+
+  pw_buf0[0] = pws[gid].i[0];
+  pw_buf0[1] = pws[gid].i[1];
+  pw_buf0[2] = pws[gid].i[2];
+  pw_buf0[3] = pws[gid].i[3];
+  pw_buf1[0] = pws[gid].i[4];
+  pw_buf1[1] = pws[gid].i[5];
+  pw_buf1[2] = pws[gid].i[6];
+  pw_buf1[3] = pws[gid].i[7];
+
+  const u32 pw_l_len = pws[gid].pw_len & 63;
+
+  /**
+   * salt
+   */
+
+  u32 salt_buf0[4] = { 0x2d2d, 0, 0, 0 };
+  u32 salt_buf1[4];
+  u32 salt_buf2[4];
+  u32 salt_buf3[4] = { 0 };
+
+  salt_buf0[0] |= salt_bufs[SALT_POS].salt_buf[ 0] << 16;
+  salt_buf0[1] = salt_bufs[SALT_POS].salt_buf[ 0] >> 16 | salt_bufs[SALT_POS].salt_buf[ 1] << 16;
+  salt_buf0[2] = salt_bufs[SALT_POS].salt_buf[ 1] >> 16 | salt_bufs[SALT_POS].salt_buf[ 2] << 16;
+  salt_buf0[3] = salt_bufs[SALT_POS].salt_buf[ 2] >> 16 | salt_bufs[SALT_POS].salt_buf[ 3] << 16;
+  salt_buf1[0] = salt_bufs[SALT_POS].salt_buf[ 3] >> 16 | salt_bufs[SALT_POS].salt_buf[ 4] << 16;
+  salt_buf1[1] = salt_bufs[SALT_POS].salt_buf[ 4] >> 16 | salt_bufs[SALT_POS].salt_buf[ 5] << 16;
+  salt_buf1[2] = salt_bufs[SALT_POS].salt_buf[ 5] >> 16 | salt_bufs[SALT_POS].salt_buf[ 6] << 16;
+  salt_buf1[3] = salt_bufs[SALT_POS].salt_buf[ 6] >> 16 | salt_bufs[SALT_POS].salt_buf[ 7] << 16;
+  salt_buf2[0] = salt_bufs[SALT_POS].salt_buf[ 7] >> 16 | salt_bufs[SALT_POS].salt_buf[ 8] << 16;
+  salt_buf2[1] = salt_bufs[SALT_POS].salt_buf[ 8] >> 16 | salt_bufs[SALT_POS].salt_buf[ 9] << 16;
+  salt_buf2[2] = salt_bufs[SALT_POS].salt_buf[ 9] >> 16 | 0x2d2d0000;
+  salt_buf2[3] = 0;
+
+  const u32 salt_len = 44;
+
+  /**
+   * digest
+   */
+
+  const u32 search[4] =
+  {
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R0],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R1],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R2],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R3]
+  };
+
+  /**
+   * loop
+   */
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
+  {
+    const u32x pw_r_len = pwlenx_create_combt (combs_buf, il_pos) & 63;
+
+    const u32x pw_len = (pw_l_len + pw_r_len) & 63;
+
+    /**
+     * concat password candidate
+     */
+
+    u32x wordl0[4] = { 0 };
+    u32x wordl1[4] = { 0 };
+    u32x wordl2[4] = { 0 };
+    u32x wordl3[4] = { 0 };
+
+    wordl0[0] = pw_buf0[0];
+    wordl0[1] = pw_buf0[1];
+    wordl0[2] = pw_buf0[2];
+    wordl0[3] = pw_buf0[3];
+    wordl1[0] = pw_buf1[0];
+    wordl1[1] = pw_buf1[1];
+    wordl1[2] = pw_buf1[2];
+    wordl1[3] = pw_buf1[3];
+
+    u32x wordr0[4] = { 0 };
+    u32x wordr1[4] = { 0 };
+    u32x wordr2[4] = { 0 };
+    u32x wordr3[4] = { 0 };
+
+    wordr0[0] = ix_create_combt (combs_buf, il_pos, 0);
+    wordr0[1] = ix_create_combt (combs_buf, il_pos, 1);
+    wordr0[2] = ix_create_combt (combs_buf, il_pos, 2);
+    wordr0[3] = ix_create_combt (combs_buf, il_pos, 3);
+    wordr1[0] = ix_create_combt (combs_buf, il_pos, 4);
+    wordr1[1] = ix_create_combt (combs_buf, il_pos, 5);
+    wordr1[2] = ix_create_combt (combs_buf, il_pos, 6);
+    wordr1[3] = ix_create_combt (combs_buf, il_pos, 7);
+
+    if (combs_mode == COMBINATOR_MODE_BASE_LEFT)
+    {
+      switch_buffer_by_offset_le_VV (wordr0, wordr1, wordr2, wordr3, pw_l_len);
+    }
+    else
+    {
+      switch_buffer_by_offset_le_VV (wordl0, wordl1, wordl2, wordl3, pw_r_len);
+    }
+
+    u32x w0[4];
+    u32x w1[4];
+    u32x w2[4];
+    u32x w3[4];
+
+    w0[0] = wordl0[0] | wordr0[0];
+    w0[1] = wordl0[1] | wordr0[1];
+    w0[2] = wordl0[2] | wordr0[2];
+    w0[3] = wordl0[3] | wordr0[3];
+    w1[0] = wordl1[0] | wordr1[0];
+    w1[1] = wordl1[1] | wordr1[1];
+    w1[2] = wordl1[2] | wordr1[2];
+    w1[3] = wordl1[3] | wordr1[3];
+    w2[0] = wordl2[0] | wordr2[0];
+    w2[1] = wordl2[1] | wordr2[1];
+    w2[2] = wordl2[2] | wordr2[2];
+    w2[3] = wordl2[3] | wordr2[3];
+    w3[0] = wordl3[0] | wordr3[0];
+    w3[1] = wordl3[1] | wordr3[1];
+    w3[2] = wordl3[2] | wordr3[2];
+    w3[3] = wordl3[3] | wordr3[3];
+
+    /**
+     * prepend salt
+     */
+
+    switch_buffer_by_offset_le (w0, w1, w2, w3, salt_len);
+
+    const u32x pw_salt_len = pw_len + salt_len;
+
+    w0[0] |= salt_buf0[0];
+    w0[1] |= salt_buf0[1];
+    w0[2] |= salt_buf0[2];
+    w0[3] |= salt_buf0[3];
+    w1[0] |= salt_buf1[0];
+    w1[1] |= salt_buf1[1];
+    w1[2] |= salt_buf1[2];
+    w1[3] |= salt_buf1[3];
+    w2[0] |= salt_buf2[0];
+    w2[1] |= salt_buf2[1];
+    w2[2] |= salt_buf2[2];
+    w2[3] |= salt_buf2[3];
+    w3[0] |= salt_buf3[0];
+    w3[1] |= salt_buf3[1];
+    w3[2] |= salt_buf3[2];
+    w3[3] |= salt_buf3[3];
+
+    append_0x2d_4x4_VV (w0, w1, w2, w3, pw_salt_len);
+    append_0x2d_4x4_VV (w0, w1, w2, w3, pw_salt_len + 1);
+    append_0x80_4x4_VV (w0, w1, w2, w3, pw_salt_len + 2);
+
+    /**
+     * sha1
+     */
+
+    u32x w0_t = hc_swap32 (w0[0]);
+    u32x w1_t = hc_swap32 (w0[1]);
+    u32x w2_t = hc_swap32 (w0[2]);
+    u32x w3_t = hc_swap32 (w0[3]);
+    u32x w4_t = hc_swap32 (w1[0]);
+    u32x w5_t = hc_swap32 (w1[1]);
+    u32x w6_t = hc_swap32 (w1[2]);
+    u32x w7_t = hc_swap32 (w1[3]);
+    u32x w8_t = hc_swap32 (w2[0]);
+    u32x w9_t = hc_swap32 (w2[1]);
+    u32x wa_t = hc_swap32 (w2[2]);
+    u32x wb_t = hc_swap32 (w2[3]);
+    u32x wc_t = hc_swap32 (w3[0]);
+    u32x wd_t = hc_swap32 (w3[1]);
+    u32x we_t = 0;
+    u32x wf_t = (pw_salt_len + 2) * 8;
+
+    u32x a = SHA1M_A;
+    u32x b = SHA1M_B;
+    u32x c = SHA1M_C;
+    u32x d = SHA1M_D;
+    u32x e = SHA1M_E;
+
+    #undef K
+    #define K SHA1C00
+
+    SHA1_STEP (SHA1_F0o, a, b, c, d, e, w0_t);
+    SHA1_STEP (SHA1_F0o, e, a, b, c, d, w1_t);
+    SHA1_STEP (SHA1_F0o, d, e, a, b, c, w2_t);
+    SHA1_STEP (SHA1_F0o, c, d, e, a, b, w3_t);
+    SHA1_STEP (SHA1_F0o, b, c, d, e, a, w4_t);
+    SHA1_STEP (SHA1_F0o, a, b, c, d, e, w5_t);
+    SHA1_STEP (SHA1_F0o, e, a, b, c, d, w6_t);
+    SHA1_STEP (SHA1_F0o, d, e, a, b, c, w7_t);
+    SHA1_STEP (SHA1_F0o, c, d, e, a, b, w8_t);
+    SHA1_STEP (SHA1_F0o, b, c, d, e, a, w9_t);
+    SHA1_STEP (SHA1_F0o, a, b, c, d, e, wa_t);
+    SHA1_STEP (SHA1_F0o, e, a, b, c, d, wb_t);
+    SHA1_STEP (SHA1_F0o, d, e, a, b, c, wc_t);
+    SHA1_STEP (SHA1_F0o, c, d, e, a, b, wd_t);
+    SHA1_STEP (SHA1_F0o, b, c, d, e, a, we_t);
+    SHA1_STEP (SHA1_F0o, a, b, c, d, e, wf_t);
+    w0_t = hc_rotl32 ((wd_t ^ w8_t ^ w2_t ^ w0_t), 1u); SHA1_STEP (SHA1_F0o, e, a, b, c, d, w0_t);
+    w1_t = hc_rotl32 ((we_t ^ w9_t ^ w3_t ^ w1_t), 1u); SHA1_STEP (SHA1_F0o, d, e, a, b, c, w1_t);
+    w2_t = hc_rotl32 ((wf_t ^ wa_t ^ w4_t ^ w2_t), 1u); SHA1_STEP (SHA1_F0o, c, d, e, a, b, w2_t);
+    w3_t = hc_rotl32 ((w0_t ^ wb_t ^ w5_t ^ w3_t), 1u); SHA1_STEP (SHA1_F0o, b, c, d, e, a, w3_t);
+
+    #undef K
+    #define K SHA1C01
+
+    w4_t = hc_rotl32 ((w1_t ^ wc_t ^ w6_t ^ w4_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w4_t);
+    w5_t = hc_rotl32 ((w2_t ^ wd_t ^ w7_t ^ w5_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w5_t);
+    w6_t = hc_rotl32 ((w3_t ^ we_t ^ w8_t ^ w6_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w6_t);
+    w7_t = hc_rotl32 ((w4_t ^ wf_t ^ w9_t ^ w7_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w7_t);
+    w8_t = hc_rotl32 ((w5_t ^ w0_t ^ wa_t ^ w8_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w8_t);
+    w9_t = hc_rotl32 ((w6_t ^ w1_t ^ wb_t ^ w9_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w9_t);
+    wa_t = hc_rotl32 ((w7_t ^ w2_t ^ wc_t ^ wa_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, wa_t);
+    wb_t = hc_rotl32 ((w8_t ^ w3_t ^ wd_t ^ wb_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, wb_t);
+    wc_t = hc_rotl32 ((w9_t ^ w4_t ^ we_t ^ wc_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, wc_t);
+    wd_t = hc_rotl32 ((wa_t ^ w5_t ^ wf_t ^ wd_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, wd_t);
+    we_t = hc_rotl32 ((wb_t ^ w6_t ^ w0_t ^ we_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, we_t);
+    wf_t = hc_rotl32 ((wc_t ^ w7_t ^ w1_t ^ wf_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, wf_t);
+    w0_t = hc_rotl32 ((wd_t ^ w8_t ^ w2_t ^ w0_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w0_t);
+    w1_t = hc_rotl32 ((we_t ^ w9_t ^ w3_t ^ w1_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w1_t);
+    w2_t = hc_rotl32 ((wf_t ^ wa_t ^ w4_t ^ w2_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w2_t);
+    w3_t = hc_rotl32 ((w0_t ^ wb_t ^ w5_t ^ w3_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w3_t);
+    w4_t = hc_rotl32 ((w1_t ^ wc_t ^ w6_t ^ w4_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w4_t);
+    w5_t = hc_rotl32 ((w2_t ^ wd_t ^ w7_t ^ w5_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w5_t);
+    w6_t = hc_rotl32 ((w3_t ^ we_t ^ w8_t ^ w6_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w6_t);
+    w7_t = hc_rotl32 ((w4_t ^ wf_t ^ w9_t ^ w7_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w7_t);
+
+    #undef K
+    #define K SHA1C02
+
+    w8_t = hc_rotl32 ((w5_t ^ w0_t ^ wa_t ^ w8_t), 1u); SHA1_STEP (SHA1_F2o, a, b, c, d, e, w8_t);
+    w9_t = hc_rotl32 ((w6_t ^ w1_t ^ wb_t ^ w9_t), 1u); SHA1_STEP (SHA1_F2o, e, a, b, c, d, w9_t);
+    wa_t = hc_rotl32 ((w7_t ^ w2_t ^ wc_t ^ wa_t), 1u); SHA1_STEP (SHA1_F2o, d, e, a, b, c, wa_t);
+    wb_t = hc_rotl32 ((w8_t ^ w3_t ^ wd_t ^ wb_t), 1u); SHA1_STEP (SHA1_F2o, c, d, e, a, b, wb_t);
+    wc_t = hc_rotl32 ((w9_t ^ w4_t ^ we_t ^ wc_t), 1u); SHA1_STEP (SHA1_F2o, b, c, d, e, a, wc_t);
+    wd_t = hc_rotl32 ((wa_t ^ w5_t ^ wf_t ^ wd_t), 1u); SHA1_STEP (SHA1_F2o, a, b, c, d, e, wd_t);
+    we_t = hc_rotl32 ((wb_t ^ w6_t ^ w0_t ^ we_t), 1u); SHA1_STEP (SHA1_F2o, e, a, b, c, d, we_t);
+    wf_t = hc_rotl32 ((wc_t ^ w7_t ^ w1_t ^ wf_t), 1u); SHA1_STEP (SHA1_F2o, d, e, a, b, c, wf_t);
+    w0_t = hc_rotl32 ((wd_t ^ w8_t ^ w2_t ^ w0_t), 1u); SHA1_STEP (SHA1_F2o, c, d, e, a, b, w0_t);
+    w1_t = hc_rotl32 ((we_t ^ w9_t ^ w3_t ^ w1_t), 1u); SHA1_STEP (SHA1_F2o, b, c, d, e, a, w1_t);
+    w2_t = hc_rotl32 ((wf_t ^ wa_t ^ w4_t ^ w2_t), 1u); SHA1_STEP (SHA1_F2o, a, b, c, d, e, w2_t);
+    w3_t = hc_rotl32 ((w0_t ^ wb_t ^ w5_t ^ w3_t), 1u); SHA1_STEP (SHA1_F2o, e, a, b, c, d, w3_t);
+    w4_t = hc_rotl32 ((w1_t ^ wc_t ^ w6_t ^ w4_t), 1u); SHA1_STEP (SHA1_F2o, d, e, a, b, c, w4_t);
+    w5_t = hc_rotl32 ((w2_t ^ wd_t ^ w7_t ^ w5_t), 1u); SHA1_STEP (SHA1_F2o, c, d, e, a, b, w5_t);
+    w6_t = hc_rotl32 ((w3_t ^ we_t ^ w8_t ^ w6_t), 1u); SHA1_STEP (SHA1_F2o, b, c, d, e, a, w6_t);
+    w7_t = hc_rotl32 ((w4_t ^ wf_t ^ w9_t ^ w7_t), 1u); SHA1_STEP (SHA1_F2o, a, b, c, d, e, w7_t);
+    w8_t = hc_rotl32 ((w5_t ^ w0_t ^ wa_t ^ w8_t), 1u); SHA1_STEP (SHA1_F2o, e, a, b, c, d, w8_t);
+    w9_t = hc_rotl32 ((w6_t ^ w1_t ^ wb_t ^ w9_t), 1u); SHA1_STEP (SHA1_F2o, d, e, a, b, c, w9_t);
+    wa_t = hc_rotl32 ((w7_t ^ w2_t ^ wc_t ^ wa_t), 1u); SHA1_STEP (SHA1_F2o, c, d, e, a, b, wa_t);
+    wb_t = hc_rotl32 ((w8_t ^ w3_t ^ wd_t ^ wb_t), 1u); SHA1_STEP (SHA1_F2o, b, c, d, e, a, wb_t);
+
+    #undef K
+    #define K SHA1C03
+
+    wc_t = hc_rotl32 ((w9_t ^ w4_t ^ we_t ^ wc_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, wc_t);
+    wd_t = hc_rotl32 ((wa_t ^ w5_t ^ wf_t ^ wd_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, wd_t);
+    we_t = hc_rotl32 ((wb_t ^ w6_t ^ w0_t ^ we_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, we_t);
+    wf_t = hc_rotl32 ((wc_t ^ w7_t ^ w1_t ^ wf_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, wf_t);
+    w0_t = hc_rotl32 ((wd_t ^ w8_t ^ w2_t ^ w0_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w0_t);
+    w1_t = hc_rotl32 ((we_t ^ w9_t ^ w3_t ^ w1_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w1_t);
+    w2_t = hc_rotl32 ((wf_t ^ wa_t ^ w4_t ^ w2_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w2_t);
+    w3_t = hc_rotl32 ((w0_t ^ wb_t ^ w5_t ^ w3_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w3_t);
+    w4_t = hc_rotl32 ((w1_t ^ wc_t ^ w6_t ^ w4_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w4_t);
+    w5_t = hc_rotl32 ((w2_t ^ wd_t ^ w7_t ^ w5_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w5_t);
+    w6_t = hc_rotl32 ((w3_t ^ we_t ^ w8_t ^ w6_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w6_t);
+    w7_t = hc_rotl32 ((w4_t ^ wf_t ^ w9_t ^ w7_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w7_t);
+    w8_t = hc_rotl32 ((w5_t ^ w0_t ^ wa_t ^ w8_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w8_t);
+    w9_t = hc_rotl32 ((w6_t ^ w1_t ^ wb_t ^ w9_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w9_t);
+    wa_t = hc_rotl32 ((w7_t ^ w2_t ^ wc_t ^ wa_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, wa_t);
+    wb_t = hc_rotl32 ((w8_t ^ w3_t ^ wd_t ^ wb_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, wb_t);
+    wc_t = hc_rotl32 ((w9_t ^ w4_t ^ we_t ^ wc_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, wc_t);
+    wd_t = hc_rotl32 ((wa_t ^ w5_t ^ wf_t ^ wd_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, wd_t);
+    we_t = hc_rotl32 ((wb_t ^ w6_t ^ w0_t ^ we_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, we_t);
+    wf_t = hc_rotl32 ((wc_t ^ w7_t ^ w1_t ^ wf_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, wf_t);
+
+    COMPARE_M_SIMD (d, e, c, b);
+  }
+}
+
+KERNEL_FQ void m27200_m08 (KERN_ATTR_BASIC ())
+{
+}
+
+KERNEL_FQ void m27200_m16 (KERN_ATTR_BASIC ())
+{
+}
+
+KERNEL_FQ void m27200_s04 (KERN_ATTR_BASIC ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+
+  /**
+   * base
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  u32 pw_buf0[4];
+  u32 pw_buf1[4];
+
+  pw_buf0[0] = pws[gid].i[0];
+  pw_buf0[1] = pws[gid].i[1];
+  pw_buf0[2] = pws[gid].i[2];
+  pw_buf0[3] = pws[gid].i[3];
+  pw_buf1[0] = pws[gid].i[4];
+  pw_buf1[1] = pws[gid].i[5];
+  pw_buf1[2] = pws[gid].i[6];
+  pw_buf1[3] = pws[gid].i[7];
+
+  const u32 pw_l_len = pws[gid].pw_len & 63;
+
+  /**
+   * salt
+   */
+
+  u32 salt_buf0[4] = { 0x2d2d, 0, 0, 0 };
+  u32 salt_buf1[4];
+  u32 salt_buf2[4];
+  u32 salt_buf3[4] = { 0 };
+
+  salt_buf0[0] |= salt_bufs[SALT_POS].salt_buf[ 0] << 16;
+  salt_buf0[1] = salt_bufs[SALT_POS].salt_buf[ 0] >> 16 | salt_bufs[SALT_POS].salt_buf[ 1] << 16;
+  salt_buf0[2] = salt_bufs[SALT_POS].salt_buf[ 1] >> 16 | salt_bufs[SALT_POS].salt_buf[ 2] << 16;
+  salt_buf0[3] = salt_bufs[SALT_POS].salt_buf[ 2] >> 16 | salt_bufs[SALT_POS].salt_buf[ 3] << 16;
+  salt_buf1[0] = salt_bufs[SALT_POS].salt_buf[ 3] >> 16 | salt_bufs[SALT_POS].salt_buf[ 4] << 16;
+  salt_buf1[1] = salt_bufs[SALT_POS].salt_buf[ 4] >> 16 | salt_bufs[SALT_POS].salt_buf[ 5] << 16;
+  salt_buf1[2] = salt_bufs[SALT_POS].salt_buf[ 5] >> 16 | salt_bufs[SALT_POS].salt_buf[ 6] << 16;
+  salt_buf1[3] = salt_bufs[SALT_POS].salt_buf[ 6] >> 16 | salt_bufs[SALT_POS].salt_buf[ 7] << 16;
+  salt_buf2[0] = salt_bufs[SALT_POS].salt_buf[ 7] >> 16 | salt_bufs[SALT_POS].salt_buf[ 8] << 16;
+  salt_buf2[1] = salt_bufs[SALT_POS].salt_buf[ 8] >> 16 | salt_bufs[SALT_POS].salt_buf[ 9] << 16;
+  salt_buf2[2] = salt_bufs[SALT_POS].salt_buf[ 9] >> 16 | 0x2d2d0000;
+  salt_buf2[3] = 0;
+
+  const u32 salt_len = 44;
+
+  /**
+   * digest
+   */
+
+  const u32 search[4] =
+  {
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R0],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R1],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R2],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R3]
+  };
+
+  /**
+   * reverse
+   */
+
+  const u32 e_rev = hc_rotl32_S (search[1], 2u);
+
+  /**
+   * loop
+   */
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
+  {
+    const u32x pw_r_len = pwlenx_create_combt (combs_buf, il_pos) & 63;
+
+    const u32x pw_len = (pw_l_len + pw_r_len) & 63;
+
+    /**
+     * concat password candidate
+     */
+
+    u32x wordl0[4] = { 0 };
+    u32x wordl1[4] = { 0 };
+    u32x wordl2[4] = { 0 };
+    u32x wordl3[4] = { 0 };
+
+    wordl0[0] = pw_buf0[0];
+    wordl0[1] = pw_buf0[1];
+    wordl0[2] = pw_buf0[2];
+    wordl0[3] = pw_buf0[3];
+    wordl1[0] = pw_buf1[0];
+    wordl1[1] = pw_buf1[1];
+    wordl1[2] = pw_buf1[2];
+    wordl1[3] = pw_buf1[3];
+
+    u32x wordr0[4] = { 0 };
+    u32x wordr1[4] = { 0 };
+    u32x wordr2[4] = { 0 };
+    u32x wordr3[4] = { 0 };
+
+    wordr0[0] = ix_create_combt (combs_buf, il_pos, 0);
+    wordr0[1] = ix_create_combt (combs_buf, il_pos, 1);
+    wordr0[2] = ix_create_combt (combs_buf, il_pos, 2);
+    wordr0[3] = ix_create_combt (combs_buf, il_pos, 3);
+    wordr1[0] = ix_create_combt (combs_buf, il_pos, 4);
+    wordr1[1] = ix_create_combt (combs_buf, il_pos, 5);
+    wordr1[2] = ix_create_combt (combs_buf, il_pos, 6);
+    wordr1[3] = ix_create_combt (combs_buf, il_pos, 7);
+
+    if (combs_mode == COMBINATOR_MODE_BASE_LEFT)
+    {
+      switch_buffer_by_offset_le_VV (wordr0, wordr1, wordr2, wordr3, pw_l_len);
+    }
+    else
+    {
+      switch_buffer_by_offset_le_VV (wordl0, wordl1, wordl2, wordl3, pw_r_len);
+    }
+
+    u32x w0[4];
+    u32x w1[4];
+    u32x w2[4];
+    u32x w3[4];
+
+    w0[0] = wordl0[0] | wordr0[0];
+    w0[1] = wordl0[1] | wordr0[1];
+    w0[2] = wordl0[2] | wordr0[2];
+    w0[3] = wordl0[3] | wordr0[3];
+    w1[0] = wordl1[0] | wordr1[0];
+    w1[1] = wordl1[1] | wordr1[1];
+    w1[2] = wordl1[2] | wordr1[2];
+    w1[3] = wordl1[3] | wordr1[3];
+    w2[0] = wordl2[0] | wordr2[0];
+    w2[1] = wordl2[1] | wordr2[1];
+    w2[2] = wordl2[2] | wordr2[2];
+    w2[3] = wordl2[3] | wordr2[3];
+    w3[0] = wordl3[0] | wordr3[0];
+    w3[1] = wordl3[1] | wordr3[1];
+    w3[2] = wordl3[2] | wordr3[2];
+    w3[3] = wordl3[3] | wordr3[3];
+
+    /**
+     * prepend salt
+     */
+
+    switch_buffer_by_offset_le (w0, w1, w2, w3, salt_len);
+
+    const u32x pw_salt_len = pw_len + salt_len;
+
+    w0[0] |= salt_buf0[0];
+    w0[1] |= salt_buf0[1];
+    w0[2] |= salt_buf0[2];
+    w0[3] |= salt_buf0[3];
+    w1[0] |= salt_buf1[0];
+    w1[1] |= salt_buf1[1];
+    w1[2] |= salt_buf1[2];
+    w1[3] |= salt_buf1[3];
+    w2[0] |= salt_buf2[0];
+    w2[1] |= salt_buf2[1];
+    w2[2] |= salt_buf2[2];
+    w2[3] |= salt_buf2[3];
+    w3[0] |= salt_buf3[0];
+    w3[1] |= salt_buf3[1];
+    w3[2] |= salt_buf3[2];
+    w3[3] |= salt_buf3[3];
+
+    append_0x2d_4x4_VV (w0, w1, w2, w3, pw_salt_len);
+    append_0x2d_4x4_VV (w0, w1, w2, w3, pw_salt_len + 1);
+    append_0x80_4x4_VV (w0, w1, w2, w3, pw_salt_len + 2);
+
+    /**
+     * sha1
+     */
+
+    u32x w0_t = hc_swap32 (w0[0]);
+    u32x w1_t = hc_swap32 (w0[1]);
+    u32x w2_t = hc_swap32 (w0[2]);
+    u32x w3_t = hc_swap32 (w0[3]);
+    u32x w4_t = hc_swap32 (w1[0]);
+    u32x w5_t = hc_swap32 (w1[1]);
+    u32x w6_t = hc_swap32 (w1[2]);
+    u32x w7_t = hc_swap32 (w1[3]);
+    u32x w8_t = hc_swap32 (w2[0]);
+    u32x w9_t = hc_swap32 (w2[1]);
+    u32x wa_t = hc_swap32 (w2[2]);
+    u32x wb_t = hc_swap32 (w2[3]);
+    u32x wc_t = hc_swap32 (w3[0]);
+    u32x wd_t = hc_swap32 (w3[1]);
+    u32x we_t = 0;
+    u32x wf_t = (pw_salt_len + 2) * 8;
+
+    u32x a = SHA1M_A;
+    u32x b = SHA1M_B;
+    u32x c = SHA1M_C;
+    u32x d = SHA1M_D;
+    u32x e = SHA1M_E;
+
+    #undef K
+    #define K SHA1C00
+
+    SHA1_STEP (SHA1_F0o, a, b, c, d, e, w0_t);
+    SHA1_STEP (SHA1_F0o, e, a, b, c, d, w1_t);
+    SHA1_STEP (SHA1_F0o, d, e, a, b, c, w2_t);
+    SHA1_STEP (SHA1_F0o, c, d, e, a, b, w3_t);
+    SHA1_STEP (SHA1_F0o, b, c, d, e, a, w4_t);
+    SHA1_STEP (SHA1_F0o, a, b, c, d, e, w5_t);
+    SHA1_STEP (SHA1_F0o, e, a, b, c, d, w6_t);
+    SHA1_STEP (SHA1_F0o, d, e, a, b, c, w7_t);
+    SHA1_STEP (SHA1_F0o, c, d, e, a, b, w8_t);
+    SHA1_STEP (SHA1_F0o, b, c, d, e, a, w9_t);
+    SHA1_STEP (SHA1_F0o, a, b, c, d, e, wa_t);
+    SHA1_STEP (SHA1_F0o, e, a, b, c, d, wb_t);
+    SHA1_STEP (SHA1_F0o, d, e, a, b, c, wc_t);
+    SHA1_STEP (SHA1_F0o, c, d, e, a, b, wd_t);
+    SHA1_STEP (SHA1_F0o, b, c, d, e, a, we_t);
+    SHA1_STEP (SHA1_F0o, a, b, c, d, e, wf_t);
+    w0_t = hc_rotl32 ((wd_t ^ w8_t ^ w2_t ^ w0_t), 1u); SHA1_STEP (SHA1_F0o, e, a, b, c, d, w0_t);
+    w1_t = hc_rotl32 ((we_t ^ w9_t ^ w3_t ^ w1_t), 1u); SHA1_STEP (SHA1_F0o, d, e, a, b, c, w1_t);
+    w2_t = hc_rotl32 ((wf_t ^ wa_t ^ w4_t ^ w2_t), 1u); SHA1_STEP (SHA1_F0o, c, d, e, a, b, w2_t);
+    w3_t = hc_rotl32 ((w0_t ^ wb_t ^ w5_t ^ w3_t), 1u); SHA1_STEP (SHA1_F0o, b, c, d, e, a, w3_t);
+
+    #undef K
+    #define K SHA1C01
+
+    w4_t = hc_rotl32 ((w1_t ^ wc_t ^ w6_t ^ w4_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w4_t);
+    w5_t = hc_rotl32 ((w2_t ^ wd_t ^ w7_t ^ w5_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w5_t);
+    w6_t = hc_rotl32 ((w3_t ^ we_t ^ w8_t ^ w6_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w6_t);
+    w7_t = hc_rotl32 ((w4_t ^ wf_t ^ w9_t ^ w7_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w7_t);
+    w8_t = hc_rotl32 ((w5_t ^ w0_t ^ wa_t ^ w8_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w8_t);
+    w9_t = hc_rotl32 ((w6_t ^ w1_t ^ wb_t ^ w9_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w9_t);
+    wa_t = hc_rotl32 ((w7_t ^ w2_t ^ wc_t ^ wa_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, wa_t);
+    wb_t = hc_rotl32 ((w8_t ^ w3_t ^ wd_t ^ wb_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, wb_t);
+    wc_t = hc_rotl32 ((w9_t ^ w4_t ^ we_t ^ wc_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, wc_t);
+    wd_t = hc_rotl32 ((wa_t ^ w5_t ^ wf_t ^ wd_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, wd_t);
+    we_t = hc_rotl32 ((wb_t ^ w6_t ^ w0_t ^ we_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, we_t);
+    wf_t = hc_rotl32 ((wc_t ^ w7_t ^ w1_t ^ wf_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, wf_t);
+    w0_t = hc_rotl32 ((wd_t ^ w8_t ^ w2_t ^ w0_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w0_t);
+    w1_t = hc_rotl32 ((we_t ^ w9_t ^ w3_t ^ w1_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w1_t);
+    w2_t = hc_rotl32 ((wf_t ^ wa_t ^ w4_t ^ w2_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w2_t);
+    w3_t = hc_rotl32 ((w0_t ^ wb_t ^ w5_t ^ w3_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w3_t);
+    w4_t = hc_rotl32 ((w1_t ^ wc_t ^ w6_t ^ w4_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w4_t);
+    w5_t = hc_rotl32 ((w2_t ^ wd_t ^ w7_t ^ w5_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w5_t);
+    w6_t = hc_rotl32 ((w3_t ^ we_t ^ w8_t ^ w6_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w6_t);
+    w7_t = hc_rotl32 ((w4_t ^ wf_t ^ w9_t ^ w7_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w7_t);
+
+    #undef K
+    #define K SHA1C02
+
+    w8_t = hc_rotl32 ((w5_t ^ w0_t ^ wa_t ^ w8_t), 1u); SHA1_STEP (SHA1_F2o, a, b, c, d, e, w8_t);
+    w9_t = hc_rotl32 ((w6_t ^ w1_t ^ wb_t ^ w9_t), 1u); SHA1_STEP (SHA1_F2o, e, a, b, c, d, w9_t);
+    wa_t = hc_rotl32 ((w7_t ^ w2_t ^ wc_t ^ wa_t), 1u); SHA1_STEP (SHA1_F2o, d, e, a, b, c, wa_t);
+    wb_t = hc_rotl32 ((w8_t ^ w3_t ^ wd_t ^ wb_t), 1u); SHA1_STEP (SHA1_F2o, c, d, e, a, b, wb_t);
+    wc_t = hc_rotl32 ((w9_t ^ w4_t ^ we_t ^ wc_t), 1u); SHA1_STEP (SHA1_F2o, b, c, d, e, a, wc_t);
+    wd_t = hc_rotl32 ((wa_t ^ w5_t ^ wf_t ^ wd_t), 1u); SHA1_STEP (SHA1_F2o, a, b, c, d, e, wd_t);
+    we_t = hc_rotl32 ((wb_t ^ w6_t ^ w0_t ^ we_t), 1u); SHA1_STEP (SHA1_F2o, e, a, b, c, d, we_t);
+    wf_t = hc_rotl32 ((wc_t ^ w7_t ^ w1_t ^ wf_t), 1u); SHA1_STEP (SHA1_F2o, d, e, a, b, c, wf_t);
+    w0_t = hc_rotl32 ((wd_t ^ w8_t ^ w2_t ^ w0_t), 1u); SHA1_STEP (SHA1_F2o, c, d, e, a, b, w0_t);
+    w1_t = hc_rotl32 ((we_t ^ w9_t ^ w3_t ^ w1_t), 1u); SHA1_STEP (SHA1_F2o, b, c, d, e, a, w1_t);
+    w2_t = hc_rotl32 ((wf_t ^ wa_t ^ w4_t ^ w2_t), 1u); SHA1_STEP (SHA1_F2o, a, b, c, d, e, w2_t);
+    w3_t = hc_rotl32 ((w0_t ^ wb_t ^ w5_t ^ w3_t), 1u); SHA1_STEP (SHA1_F2o, e, a, b, c, d, w3_t);
+    w4_t = hc_rotl32 ((w1_t ^ wc_t ^ w6_t ^ w4_t), 1u); SHA1_STEP (SHA1_F2o, d, e, a, b, c, w4_t);
+    w5_t = hc_rotl32 ((w2_t ^ wd_t ^ w7_t ^ w5_t), 1u); SHA1_STEP (SHA1_F2o, c, d, e, a, b, w5_t);
+    w6_t = hc_rotl32 ((w3_t ^ we_t ^ w8_t ^ w6_t), 1u); SHA1_STEP (SHA1_F2o, b, c, d, e, a, w6_t);
+    w7_t = hc_rotl32 ((w4_t ^ wf_t ^ w9_t ^ w7_t), 1u); SHA1_STEP (SHA1_F2o, a, b, c, d, e, w7_t);
+    w8_t = hc_rotl32 ((w5_t ^ w0_t ^ wa_t ^ w8_t), 1u); SHA1_STEP (SHA1_F2o, e, a, b, c, d, w8_t);
+    w9_t = hc_rotl32 ((w6_t ^ w1_t ^ wb_t ^ w9_t), 1u); SHA1_STEP (SHA1_F2o, d, e, a, b, c, w9_t);
+    wa_t = hc_rotl32 ((w7_t ^ w2_t ^ wc_t ^ wa_t), 1u); SHA1_STEP (SHA1_F2o, c, d, e, a, b, wa_t);
+    wb_t = hc_rotl32 ((w8_t ^ w3_t ^ wd_t ^ wb_t), 1u); SHA1_STEP (SHA1_F2o, b, c, d, e, a, wb_t);
+
+    #undef K
+    #define K SHA1C03
+
+    wc_t = hc_rotl32 ((w9_t ^ w4_t ^ we_t ^ wc_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, wc_t);
+    wd_t = hc_rotl32 ((wa_t ^ w5_t ^ wf_t ^ wd_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, wd_t);
+    we_t = hc_rotl32 ((wb_t ^ w6_t ^ w0_t ^ we_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, we_t);
+    wf_t = hc_rotl32 ((wc_t ^ w7_t ^ w1_t ^ wf_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, wf_t);
+    w0_t = hc_rotl32 ((wd_t ^ w8_t ^ w2_t ^ w0_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w0_t);
+    w1_t = hc_rotl32 ((we_t ^ w9_t ^ w3_t ^ w1_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w1_t);
+    w2_t = hc_rotl32 ((wf_t ^ wa_t ^ w4_t ^ w2_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w2_t);
+    w3_t = hc_rotl32 ((w0_t ^ wb_t ^ w5_t ^ w3_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w3_t);
+    w4_t = hc_rotl32 ((w1_t ^ wc_t ^ w6_t ^ w4_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w4_t);
+    w5_t = hc_rotl32 ((w2_t ^ wd_t ^ w7_t ^ w5_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w5_t);
+    w6_t = hc_rotl32 ((w3_t ^ we_t ^ w8_t ^ w6_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w6_t);
+    w7_t = hc_rotl32 ((w4_t ^ wf_t ^ w9_t ^ w7_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w7_t);
+    w8_t = hc_rotl32 ((w5_t ^ w0_t ^ wa_t ^ w8_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w8_t);
+    w9_t = hc_rotl32 ((w6_t ^ w1_t ^ wb_t ^ w9_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w9_t);
+    wa_t = hc_rotl32 ((w7_t ^ w2_t ^ wc_t ^ wa_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, wa_t);
+    wb_t = hc_rotl32 ((w8_t ^ w3_t ^ wd_t ^ wb_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, wb_t);
+
+    if (MATCHES_NONE_VS (e, e_rev)) continue;
+
+    wc_t = hc_rotl32 ((w9_t ^ w4_t ^ we_t ^ wc_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, wc_t);
+    wd_t = hc_rotl32 ((wa_t ^ w5_t ^ wf_t ^ wd_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, wd_t);
+    we_t = hc_rotl32 ((wb_t ^ w6_t ^ w0_t ^ we_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, we_t);
+    wf_t = hc_rotl32 ((wc_t ^ w7_t ^ w1_t ^ wf_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, wf_t);
+
+    COMPARE_S_SIMD (d, e, c, b);
+  }
+}
+
+KERNEL_FQ void m27200_s08 (KERN_ATTR_BASIC ())
+{
+}
+
+KERNEL_FQ void m27200_s16 (KERN_ATTR_BASIC ())
+{
+}
diff --git a/OpenCL/m27200_a1-pure.cl b/OpenCL/m27200_a1-pure.cl
new file mode 100644
index 000000000..dca979941
--- /dev/null
+++ b/OpenCL/m27200_a1-pure.cl
@@ -0,0 +1,131 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+//#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include "inc_vendor.h"
+#include "inc_types.h"
+#include "inc_platform.cl"
+#include "inc_common.cl"
+#include "inc_scalar.cl"
+#include "inc_hash_sha1.cl"
+#endif
+
+KERNEL_FQ void m27200_mxx (KERN_ATTR_BASIC ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  const u32 dash[16] = { 0x2d2d0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
+
+  /**
+   * base
+   */
+
+  sha1_ctx_t ctx0;
+
+  sha1_init (&ctx0);
+
+  ctx0.w0[0] = dash[0];
+  ctx0.w0[1] = dash[1];
+
+  ctx0.len = 2;
+
+  sha1_update_global_swap (&ctx0, salt_bufs[SALT_POS].salt_buf, salt_bufs[SALT_POS].salt_len);
+  sha1_update (&ctx0, dash, 2);
+
+  sha1_update_global_swap (&ctx0, pws[gid].i, pws[gid].pw_len);
+
+  /**
+   * loop
+   */
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos++)
+  {
+    sha1_ctx_t ctx = ctx0;
+
+    sha1_update_global_swap (&ctx, combs_buf[il_pos].i, combs_buf[il_pos].pw_len);
+    sha1_update (&ctx, dash, 2);
+    sha1_final (&ctx);
+
+    const u32 r0 = ctx.h[DGST_R0];
+    const u32 r1 = ctx.h[DGST_R1];
+    const u32 r2 = ctx.h[DGST_R2];
+    const u32 r3 = ctx.h[DGST_R3];
+
+    COMPARE_M_SCALAR (r0, r1, r2, r3);
+  }
+}
+
+KERNEL_FQ void m27200_sxx (KERN_ATTR_BASIC ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  const u32 dash[16] = { 0x2d2d0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
+  
+  /**
+   * digest
+   */
+
+  const u32 search[4] =
+  {
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R0],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R1],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R2],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R3]
+  };
+
+  /**
+   * base
+   */
+
+  sha1_ctx_t ctx0;
+
+  sha1_init (&ctx0);
+
+  ctx0.w0[0] = dash[0];
+  ctx0.w0[1] = dash[1];
+
+  ctx0.len = 2;
+  
+  sha1_update_global_swap (&ctx0, salt_bufs[SALT_POS].salt_buf, salt_bufs[SALT_POS].salt_len);
+  sha1_update (&ctx0, dash, 2);
+
+  sha1_update_global_swap (&ctx0, pws[gid].i, pws[gid].pw_len);
+
+  /**
+   * loop
+   */
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos++)
+  {
+    sha1_ctx_t ctx = ctx0;
+
+    sha1_update_global_swap (&ctx, combs_buf[il_pos].i, combs_buf[il_pos].pw_len);
+    sha1_update (&ctx, dash, 2);
+    sha1_final (&ctx);
+
+    const u32 r0 = ctx.h[DGST_R0];
+    const u32 r1 = ctx.h[DGST_R1];
+    const u32 r2 = ctx.h[DGST_R2];
+    const u32 r3 = ctx.h[DGST_R3];
+
+    COMPARE_S_SCALAR (r0, r1, r2, r3);
+  }
+}
diff --git a/OpenCL/m27200_a3-optimized.cl b/OpenCL/m27200_a3-optimized.cl
new file mode 100644
index 000000000..a7cf11877
--- /dev/null
+++ b/OpenCL/m27200_a3-optimized.cl
@@ -0,0 +1,828 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include "inc_vendor.h"
+#include "inc_types.h"
+#include "inc_platform.cl"
+#include "inc_common.cl"
+#include "inc_simd.cl"
+#include "inc_hash_sha1.cl"
+#endif
+
+DECLSPEC void append_4 (const u32 offset, u32 *w0, u32 *w1, u32 *w2, u32 *w3, const u32 src_r0)
+{
+  u32 tmp[2];
+
+  switch (offset & 3)
+  {
+    case  0:  tmp[0] = src_r0;
+              tmp[1] = 0;
+              break;
+    case  1:  tmp[0] = src_r0 >>  8;
+              tmp[1] = src_r0 << 24;
+              break;
+    case  2:  tmp[0] = src_r0 >> 16;
+              tmp[1] = src_r0 << 16;
+              break;
+    case  3:  tmp[0] = src_r0 >> 24;
+              tmp[1] = src_r0 <<  8;
+              break;
+  }
+
+  switch (offset / 4)
+  {
+    case  0:  w0[0] |= tmp[0];
+              w0[1]  = tmp[1];
+              break;
+    case  1:  w0[1] |= tmp[0];
+              w0[2]  = tmp[1];
+              break;
+    case  2:  w0[2] |= tmp[0];
+              w0[3]  = tmp[1];
+              break;
+    case  3:  w0[3] |= tmp[0];
+              w1[0]  = tmp[1];
+              break;
+    case  4:  w1[0] |= tmp[0];
+              w1[1]  = tmp[1];
+              break;
+    case  5:  w1[1] |= tmp[0];
+              w1[2]  = tmp[1];
+              break;
+    case  6:  w1[2] |= tmp[0];
+              w1[3]  = tmp[1];
+              break;
+    case  7:  w1[3] |= tmp[0];
+              w2[0]  = tmp[1];
+              break;
+    case  8:  w2[0] |= tmp[0];
+              w2[1]  = tmp[1];
+              break;
+    case  9:  w2[1] |= tmp[0];
+              w2[2]  = tmp[1];
+              break;
+    case 10:  w2[2] |= tmp[0];
+              w2[3]  = tmp[1];
+              break;
+    case 11:  w2[3] |= tmp[0];
+              w3[0]  = tmp[1];
+              break;
+    case 12:  w3[0] |= tmp[0];
+              w3[1]  = tmp[1];
+              break;
+    case 13:  w3[1] |= tmp[0];
+              w3[2]  = tmp[1];
+              break;
+    case 14:  w3[2] |= tmp[0];
+              w3[3]  = tmp[1];
+              break;
+    case 15:  w3[3] |= tmp[0];
+              break;
+  }
+}
+
+DECLSPEC void m27200m (u32 *w0, u32 *w1, u32 *w2, u32 *w3, const u32 pw_len, KERN_ATTR_BASIC ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+
+  const u32 dash_stop = 0x2d2d8000;
+
+  /**
+   * salt
+   */
+
+  u32 salt_buf0[4] = { 0x2d2d0000, 0, 0, 0 };
+  u32 salt_buf1[4];
+  u32 salt_buf2[4];
+  u32 salt_buf3[4] = { 0 };
+
+  salt_buf0[0] |= hc_swap32_S (salt_bufs[SALT_POS].salt_buf[ 0] << 16);
+  salt_buf0[1] = hc_swap32_S (salt_bufs[SALT_POS].salt_buf[ 0] >> 16 | salt_bufs[SALT_POS].salt_buf[ 1] << 16);
+  salt_buf0[2] = hc_swap32_S (salt_bufs[SALT_POS].salt_buf[ 1] >> 16 | salt_bufs[SALT_POS].salt_buf[ 2] << 16);
+  salt_buf0[3] = hc_swap32_S (salt_bufs[SALT_POS].salt_buf[ 2] >> 16 | salt_bufs[SALT_POS].salt_buf[ 3] << 16);
+  salt_buf1[0] = hc_swap32_S (salt_bufs[SALT_POS].salt_buf[ 3] >> 16 | salt_bufs[SALT_POS].salt_buf[ 4] << 16);
+  salt_buf1[1] = hc_swap32_S (salt_bufs[SALT_POS].salt_buf[ 4] >> 16 | salt_bufs[SALT_POS].salt_buf[ 5] << 16);
+  salt_buf1[2] = hc_swap32_S (salt_bufs[SALT_POS].salt_buf[ 5] >> 16 | salt_bufs[SALT_POS].salt_buf[ 6] << 16);
+  salt_buf1[3] = hc_swap32_S (salt_bufs[SALT_POS].salt_buf[ 6] >> 16 | salt_bufs[SALT_POS].salt_buf[ 7] << 16);
+  salt_buf2[0] = hc_swap32_S (salt_bufs[SALT_POS].salt_buf[ 7] >> 16 | salt_bufs[SALT_POS].salt_buf[ 8] << 16);
+  salt_buf2[1] = hc_swap32_S (salt_bufs[SALT_POS].salt_buf[ 8] >> 16 | salt_bufs[SALT_POS].salt_buf[ 9] << 16);
+  salt_buf2[2] = hc_swap32_S (salt_bufs[SALT_POS].salt_buf[ 9] >> 16 | 0x2d2d0000);
+  salt_buf2[3] = 0;
+
+  const u32 salt_len = 44;
+
+  append_4 (pw_len, w0, w1, w2, w3, dash_stop);
+
+  const u32 pw_salt_len = pw_len + 2 + salt_len;
+
+  /**
+   * loop
+   */
+
+  const u32 w0l = w0[0];
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
+  {
+    const u32x w0r = ix_create_bft (bfs_buf, il_pos);
+
+    const u32x w0lr = w0l | w0r;
+
+    u32x t0[4];
+    u32x t1[4];
+    u32x t2[4];
+    u32x t3[4];
+
+    t0[0] = w0lr;
+    t0[1] = w0[1];
+    t0[2] = w0[2];
+    t0[3] = w0[3];
+    t1[0] = w1[0];
+    t1[1] = w1[1];
+    t1[2] = w1[2];
+    t1[3] = w1[3];
+    t2[0] = w2[0];
+    t2[1] = w2[1];
+    t2[2] = w2[2];
+    t2[3] = w2[3];
+    t3[0] = w3[0];
+    t3[1] = w3[1];
+    t3[2] = w3[2];
+    t3[3] = w3[3];
+    
+    switch_buffer_by_offset_be (t0, t1, t2, t3, salt_len);
+
+    t0[0] |= salt_buf0[0];
+    t0[1] |= salt_buf0[1];
+    t0[2] |= salt_buf0[2];
+    t0[3] |= salt_buf0[3];
+    t1[0] |= salt_buf1[0];
+    t1[1] |= salt_buf1[1];
+    t1[2] |= salt_buf1[2];
+    t1[3] |= salt_buf1[3];
+    t2[0] |= salt_buf2[0];
+    t2[1] |= salt_buf2[1];
+    t2[2] |= salt_buf2[2];
+    t2[3] |= salt_buf2[3];
+    t3[0] |= salt_buf3[0];
+    t3[1] |= salt_buf3[1];
+    t3[2]  = 0;
+    t3[3]  = (pw_salt_len) * 8;
+
+    /**
+     * sha1
+     */
+
+    u32x w0_t = t0[0];
+    u32x w1_t = t0[1];
+    u32x w2_t = t0[2];
+    u32x w3_t = t0[3];
+    u32x w4_t = t1[0];
+    u32x w5_t = t1[1];
+    u32x w6_t = t1[2];
+    u32x w7_t = t1[3];
+    u32x w8_t = t2[0];
+    u32x w9_t = t2[1];
+    u32x wa_t = t2[2];
+    u32x wb_t = t2[3];
+    u32x wc_t = t3[0];
+    u32x wd_t = t3[1];
+    u32x we_t = 0;
+    u32x wf_t = (pw_salt_len) * 8;
+
+    u32x a = SHA1M_A;
+    u32x b = SHA1M_B;
+    u32x c = SHA1M_C;
+    u32x d = SHA1M_D;
+    u32x e = SHA1M_E;
+
+    #undef K
+    #define K SHA1C00
+
+    SHA1_STEP (SHA1_F0o, a, b, c, d, e, w0_t);
+    SHA1_STEP (SHA1_F0o, e, a, b, c, d, w1_t);
+    SHA1_STEP (SHA1_F0o, d, e, a, b, c, w2_t);
+    SHA1_STEP (SHA1_F0o, c, d, e, a, b, w3_t);
+    SHA1_STEP (SHA1_F0o, b, c, d, e, a, w4_t);
+    SHA1_STEP (SHA1_F0o, a, b, c, d, e, w5_t);
+    SHA1_STEP (SHA1_F0o, e, a, b, c, d, w6_t);
+    SHA1_STEP (SHA1_F0o, d, e, a, b, c, w7_t);
+    SHA1_STEP (SHA1_F0o, c, d, e, a, b, w8_t);
+    SHA1_STEP (SHA1_F0o, b, c, d, e, a, w9_t);
+    SHA1_STEP (SHA1_F0o, a, b, c, d, e, wa_t);
+    SHA1_STEP (SHA1_F0o, e, a, b, c, d, wb_t);
+    SHA1_STEP (SHA1_F0o, d, e, a, b, c, wc_t);
+    SHA1_STEP (SHA1_F0o, c, d, e, a, b, wd_t);
+    SHA1_STEP (SHA1_F0o, b, c, d, e, a, we_t);
+    SHA1_STEP (SHA1_F0o, a, b, c, d, e, wf_t);
+    w0_t = hc_rotl32 ((wd_t ^ w8_t ^ w2_t ^ w0_t), 1u); SHA1_STEP (SHA1_F0o, e, a, b, c, d, w0_t);
+    w1_t = hc_rotl32 ((we_t ^ w9_t ^ w3_t ^ w1_t), 1u); SHA1_STEP (SHA1_F0o, d, e, a, b, c, w1_t);
+    w2_t = hc_rotl32 ((wf_t ^ wa_t ^ w4_t ^ w2_t), 1u); SHA1_STEP (SHA1_F0o, c, d, e, a, b, w2_t);
+    w3_t = hc_rotl32 ((w0_t ^ wb_t ^ w5_t ^ w3_t), 1u); SHA1_STEP (SHA1_F0o, b, c, d, e, a, w3_t);
+
+    #undef K
+    #define K SHA1C01
+
+    w4_t = hc_rotl32 ((w1_t ^ wc_t ^ w6_t ^ w4_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w4_t);
+    w5_t = hc_rotl32 ((w2_t ^ wd_t ^ w7_t ^ w5_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w5_t);
+    w6_t = hc_rotl32 ((w3_t ^ we_t ^ w8_t ^ w6_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w6_t);
+    w7_t = hc_rotl32 ((w4_t ^ wf_t ^ w9_t ^ w7_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w7_t);
+    w8_t = hc_rotl32 ((w5_t ^ w0_t ^ wa_t ^ w8_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w8_t);
+    w9_t = hc_rotl32 ((w6_t ^ w1_t ^ wb_t ^ w9_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w9_t);
+    wa_t = hc_rotl32 ((w7_t ^ w2_t ^ wc_t ^ wa_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, wa_t);
+    wb_t = hc_rotl32 ((w8_t ^ w3_t ^ wd_t ^ wb_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, wb_t);
+    wc_t = hc_rotl32 ((w9_t ^ w4_t ^ we_t ^ wc_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, wc_t);
+    wd_t = hc_rotl32 ((wa_t ^ w5_t ^ wf_t ^ wd_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, wd_t);
+    we_t = hc_rotl32 ((wb_t ^ w6_t ^ w0_t ^ we_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, we_t);
+    wf_t = hc_rotl32 ((wc_t ^ w7_t ^ w1_t ^ wf_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, wf_t);
+    w0_t = hc_rotl32 ((wd_t ^ w8_t ^ w2_t ^ w0_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w0_t);
+    w1_t = hc_rotl32 ((we_t ^ w9_t ^ w3_t ^ w1_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w1_t);
+    w2_t = hc_rotl32 ((wf_t ^ wa_t ^ w4_t ^ w2_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w2_t);
+    w3_t = hc_rotl32 ((w0_t ^ wb_t ^ w5_t ^ w3_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w3_t);
+    w4_t = hc_rotl32 ((w1_t ^ wc_t ^ w6_t ^ w4_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w4_t);
+    w5_t = hc_rotl32 ((w2_t ^ wd_t ^ w7_t ^ w5_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w5_t);
+    w6_t = hc_rotl32 ((w3_t ^ we_t ^ w8_t ^ w6_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w6_t);
+    w7_t = hc_rotl32 ((w4_t ^ wf_t ^ w9_t ^ w7_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w7_t);
+
+    #undef K
+    #define K SHA1C02
+
+    w8_t = hc_rotl32 ((w5_t ^ w0_t ^ wa_t ^ w8_t), 1u); SHA1_STEP (SHA1_F2o, a, b, c, d, e, w8_t);
+    w9_t = hc_rotl32 ((w6_t ^ w1_t ^ wb_t ^ w9_t), 1u); SHA1_STEP (SHA1_F2o, e, a, b, c, d, w9_t);
+    wa_t = hc_rotl32 ((w7_t ^ w2_t ^ wc_t ^ wa_t), 1u); SHA1_STEP (SHA1_F2o, d, e, a, b, c, wa_t);
+    wb_t = hc_rotl32 ((w8_t ^ w3_t ^ wd_t ^ wb_t), 1u); SHA1_STEP (SHA1_F2o, c, d, e, a, b, wb_t);
+    wc_t = hc_rotl32 ((w9_t ^ w4_t ^ we_t ^ wc_t), 1u); SHA1_STEP (SHA1_F2o, b, c, d, e, a, wc_t);
+    wd_t = hc_rotl32 ((wa_t ^ w5_t ^ wf_t ^ wd_t), 1u); SHA1_STEP (SHA1_F2o, a, b, c, d, e, wd_t);
+    we_t = hc_rotl32 ((wb_t ^ w6_t ^ w0_t ^ we_t), 1u); SHA1_STEP (SHA1_F2o, e, a, b, c, d, we_t);
+    wf_t = hc_rotl32 ((wc_t ^ w7_t ^ w1_t ^ wf_t), 1u); SHA1_STEP (SHA1_F2o, d, e, a, b, c, wf_t);
+    w0_t = hc_rotl32 ((wd_t ^ w8_t ^ w2_t ^ w0_t), 1u); SHA1_STEP (SHA1_F2o, c, d, e, a, b, w0_t);
+    w1_t = hc_rotl32 ((we_t ^ w9_t ^ w3_t ^ w1_t), 1u); SHA1_STEP (SHA1_F2o, b, c, d, e, a, w1_t);
+    w2_t = hc_rotl32 ((wf_t ^ wa_t ^ w4_t ^ w2_t), 1u); SHA1_STEP (SHA1_F2o, a, b, c, d, e, w2_t);
+    w3_t = hc_rotl32 ((w0_t ^ wb_t ^ w5_t ^ w3_t), 1u); SHA1_STEP (SHA1_F2o, e, a, b, c, d, w3_t);
+    w4_t = hc_rotl32 ((w1_t ^ wc_t ^ w6_t ^ w4_t), 1u); SHA1_STEP (SHA1_F2o, d, e, a, b, c, w4_t);
+    w5_t = hc_rotl32 ((w2_t ^ wd_t ^ w7_t ^ w5_t), 1u); SHA1_STEP (SHA1_F2o, c, d, e, a, b, w5_t);
+    w6_t = hc_rotl32 ((w3_t ^ we_t ^ w8_t ^ w6_t), 1u); SHA1_STEP (SHA1_F2o, b, c, d, e, a, w6_t);
+    w7_t = hc_rotl32 ((w4_t ^ wf_t ^ w9_t ^ w7_t), 1u); SHA1_STEP (SHA1_F2o, a, b, c, d, e, w7_t);
+    w8_t = hc_rotl32 ((w5_t ^ w0_t ^ wa_t ^ w8_t), 1u); SHA1_STEP (SHA1_F2o, e, a, b, c, d, w8_t);
+    w9_t = hc_rotl32 ((w6_t ^ w1_t ^ wb_t ^ w9_t), 1u); SHA1_STEP (SHA1_F2o, d, e, a, b, c, w9_t);
+    wa_t = hc_rotl32 ((w7_t ^ w2_t ^ wc_t ^ wa_t), 1u); SHA1_STEP (SHA1_F2o, c, d, e, a, b, wa_t);
+    wb_t = hc_rotl32 ((w8_t ^ w3_t ^ wd_t ^ wb_t), 1u); SHA1_STEP (SHA1_F2o, b, c, d, e, a, wb_t);
+
+    #undef K
+    #define K SHA1C03
+
+    wc_t = hc_rotl32 ((w9_t ^ w4_t ^ we_t ^ wc_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, wc_t);
+    wd_t = hc_rotl32 ((wa_t ^ w5_t ^ wf_t ^ wd_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, wd_t);
+    we_t = hc_rotl32 ((wb_t ^ w6_t ^ w0_t ^ we_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, we_t);
+    wf_t = hc_rotl32 ((wc_t ^ w7_t ^ w1_t ^ wf_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, wf_t);
+    w0_t = hc_rotl32 ((wd_t ^ w8_t ^ w2_t ^ w0_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w0_t);
+    w1_t = hc_rotl32 ((we_t ^ w9_t ^ w3_t ^ w1_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w1_t);
+    w2_t = hc_rotl32 ((wf_t ^ wa_t ^ w4_t ^ w2_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w2_t);
+    w3_t = hc_rotl32 ((w0_t ^ wb_t ^ w5_t ^ w3_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w3_t);
+    w4_t = hc_rotl32 ((w1_t ^ wc_t ^ w6_t ^ w4_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w4_t);
+    w5_t = hc_rotl32 ((w2_t ^ wd_t ^ w7_t ^ w5_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w5_t);
+    w6_t = hc_rotl32 ((w3_t ^ we_t ^ w8_t ^ w6_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w6_t);
+    w7_t = hc_rotl32 ((w4_t ^ wf_t ^ w9_t ^ w7_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w7_t);
+    w8_t = hc_rotl32 ((w5_t ^ w0_t ^ wa_t ^ w8_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w8_t);
+    w9_t = hc_rotl32 ((w6_t ^ w1_t ^ wb_t ^ w9_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w9_t);
+    wa_t = hc_rotl32 ((w7_t ^ w2_t ^ wc_t ^ wa_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, wa_t);
+    wb_t = hc_rotl32 ((w8_t ^ w3_t ^ wd_t ^ wb_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, wb_t);
+    wc_t = hc_rotl32 ((w9_t ^ w4_t ^ we_t ^ wc_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, wc_t);
+    wd_t = hc_rotl32 ((wa_t ^ w5_t ^ wf_t ^ wd_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, wd_t);
+    we_t = hc_rotl32 ((wb_t ^ w6_t ^ w0_t ^ we_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, we_t);
+    wf_t = hc_rotl32 ((wc_t ^ w7_t ^ w1_t ^ wf_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, wf_t);
+
+    COMPARE_M_SIMD (d, e, c, b);
+  }
+}
+
+DECLSPEC void m27200s (u32 *w0, u32 *w1, u32 *w2, u32 *w3, const u32 pw_len, KERN_ATTR_BASIC ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+
+  const u32 dash_stop = 0x2d2d8000;
+
+  /**
+   * digest
+   */
+
+  const u32 search[4] =
+  {
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R0],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R1],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R2],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R3]
+  };
+
+  /**
+   * reverse
+   */
+
+  const u32 e_rev = hc_rotl32_S (search[1], 2u);
+
+  /**
+   * salt
+   */
+
+  u32 salt_buf0[4] = { 0x2d2d0000, 0, 0, 0 };
+  u32 salt_buf1[4];
+  u32 salt_buf2[4];
+  u32 salt_buf3[4] = { 0 };
+
+  salt_buf0[0] |= hc_swap32_S (salt_bufs[SALT_POS].salt_buf[ 0] << 16);
+  salt_buf0[1] = hc_swap32_S (salt_bufs[SALT_POS].salt_buf[ 0] >> 16 | salt_bufs[SALT_POS].salt_buf[ 1] << 16);
+  salt_buf0[2] = hc_swap32_S (salt_bufs[SALT_POS].salt_buf[ 1] >> 16 | salt_bufs[SALT_POS].salt_buf[ 2] << 16);
+  salt_buf0[3] = hc_swap32_S (salt_bufs[SALT_POS].salt_buf[ 2] >> 16 | salt_bufs[SALT_POS].salt_buf[ 3] << 16);
+  salt_buf1[0] = hc_swap32_S (salt_bufs[SALT_POS].salt_buf[ 3] >> 16 | salt_bufs[SALT_POS].salt_buf[ 4] << 16);
+  salt_buf1[1] = hc_swap32_S (salt_bufs[SALT_POS].salt_buf[ 4] >> 16 | salt_bufs[SALT_POS].salt_buf[ 5] << 16);
+  salt_buf1[2] = hc_swap32_S (salt_bufs[SALT_POS].salt_buf[ 5] >> 16 | salt_bufs[SALT_POS].salt_buf[ 6] << 16);
+  salt_buf1[3] = hc_swap32_S (salt_bufs[SALT_POS].salt_buf[ 6] >> 16 | salt_bufs[SALT_POS].salt_buf[ 7] << 16);
+  salt_buf2[0] = hc_swap32_S (salt_bufs[SALT_POS].salt_buf[ 7] >> 16 | salt_bufs[SALT_POS].salt_buf[ 8] << 16);
+  salt_buf2[1] = hc_swap32_S (salt_bufs[SALT_POS].salt_buf[ 8] >> 16 | salt_bufs[SALT_POS].salt_buf[ 9] << 16);
+  salt_buf2[2] = hc_swap32_S (salt_bufs[SALT_POS].salt_buf[ 9] >> 16 | 0x2d2d0000);
+  salt_buf2[3] = 0;
+
+  const u32 salt_len = 44;
+
+  append_4 (pw_len, w0, w1, w2, w3, dash_stop);
+
+  const u32 pw_salt_len = pw_len + 2 + salt_len;
+
+  /**
+   * loop
+   */
+
+  const u32 w0l = w0[0];
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
+  {
+    const u32x w0r = ix_create_bft (bfs_buf, il_pos);
+
+    const u32x w0lr = w0l | w0r;
+
+    u32x t0[4];
+    u32x t1[4];
+    u32x t2[4];
+    u32x t3[4];
+
+    t0[0] = w0lr;
+    t0[1] = w0[1];
+    t0[2] = w0[2];
+    t0[3] = w0[3];
+    t1[0] = w1[0];
+    t1[1] = w1[1];
+    t1[2] = w1[2];
+    t1[3] = w1[3];
+    t2[0] = w2[0];
+    t2[1] = w2[1];
+    t2[2] = w2[2];
+    t2[3] = w2[3];
+    t3[0] = w3[0];
+    t3[1] = w3[1];
+    t3[2] = w3[2];
+    t3[3] = w3[3];
+
+    switch_buffer_by_offset_be (t0, t1, t2, t3, salt_len);
+
+    t0[0] |= salt_buf0[0];
+    t0[1] |= salt_buf0[1];
+    t0[2] |= salt_buf0[2];
+    t0[3] |= salt_buf0[3];
+    t1[0] |= salt_buf1[0];
+    t1[1] |= salt_buf1[1];
+    t1[2] |= salt_buf1[2];
+    t1[3] |= salt_buf1[3];
+    t2[0] |= salt_buf2[0];
+    t2[1] |= salt_buf2[1];
+    t2[2] |= salt_buf2[2];
+    t2[3] |= salt_buf2[3];
+    t3[0] |= salt_buf3[0];
+    t3[1] |= salt_buf3[1];
+    t3[2]  = 0;
+    t3[3]  = pw_salt_len * 8;
+
+    /**
+     * sha1
+     */
+
+    u32x w0_t = t0[0];
+    u32x w1_t = t0[1];
+    u32x w2_t = t0[2];
+    u32x w3_t = t0[3];
+    u32x w4_t = t1[0];
+    u32x w5_t = t1[1];
+    u32x w6_t = t1[2];
+    u32x w7_t = t1[3];
+    u32x w8_t = t2[0];
+    u32x w9_t = t2[1];
+    u32x wa_t = t2[2];
+    u32x wb_t = t2[3];
+    u32x wc_t = t3[0];
+    u32x wd_t = t3[1];
+    u32x we_t = 0;
+    u32x wf_t = pw_salt_len * 8;
+
+    u32x a = SHA1M_A;
+    u32x b = SHA1M_B;
+    u32x c = SHA1M_C;
+    u32x d = SHA1M_D;
+    u32x e = SHA1M_E;
+
+    #undef K
+    #define K SHA1C00
+
+    SHA1_STEP (SHA1_F0o, a, b, c, d, e, w0_t);
+    SHA1_STEP (SHA1_F0o, e, a, b, c, d, w1_t);
+    SHA1_STEP (SHA1_F0o, d, e, a, b, c, w2_t);
+    SHA1_STEP (SHA1_F0o, c, d, e, a, b, w3_t);
+    SHA1_STEP (SHA1_F0o, b, c, d, e, a, w4_t);
+    SHA1_STEP (SHA1_F0o, a, b, c, d, e, w5_t);
+    SHA1_STEP (SHA1_F0o, e, a, b, c, d, w6_t);
+    SHA1_STEP (SHA1_F0o, d, e, a, b, c, w7_t);
+    SHA1_STEP (SHA1_F0o, c, d, e, a, b, w8_t);
+    SHA1_STEP (SHA1_F0o, b, c, d, e, a, w9_t);
+    SHA1_STEP (SHA1_F0o, a, b, c, d, e, wa_t);
+    SHA1_STEP (SHA1_F0o, e, a, b, c, d, wb_t);
+    SHA1_STEP (SHA1_F0o, d, e, a, b, c, wc_t);
+    SHA1_STEP (SHA1_F0o, c, d, e, a, b, wd_t);
+    SHA1_STEP (SHA1_F0o, b, c, d, e, a, we_t);
+    SHA1_STEP (SHA1_F0o, a, b, c, d, e, wf_t);
+    w0_t = hc_rotl32 ((wd_t ^ w8_t ^ w2_t ^ w0_t), 1u); SHA1_STEP (SHA1_F0o, e, a, b, c, d, w0_t);
+    w1_t = hc_rotl32 ((we_t ^ w9_t ^ w3_t ^ w1_t), 1u); SHA1_STEP (SHA1_F0o, d, e, a, b, c, w1_t);
+    w2_t = hc_rotl32 ((wf_t ^ wa_t ^ w4_t ^ w2_t), 1u); SHA1_STEP (SHA1_F0o, c, d, e, a, b, w2_t);
+    w3_t = hc_rotl32 ((w0_t ^ wb_t ^ w5_t ^ w3_t), 1u); SHA1_STEP (SHA1_F0o, b, c, d, e, a, w3_t);
+
+    #undef K
+    #define K SHA1C01
+
+    w4_t = hc_rotl32 ((w1_t ^ wc_t ^ w6_t ^ w4_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w4_t);
+    w5_t = hc_rotl32 ((w2_t ^ wd_t ^ w7_t ^ w5_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w5_t);
+    w6_t = hc_rotl32 ((w3_t ^ we_t ^ w8_t ^ w6_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w6_t);
+    w7_t = hc_rotl32 ((w4_t ^ wf_t ^ w9_t ^ w7_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w7_t);
+    w8_t = hc_rotl32 ((w5_t ^ w0_t ^ wa_t ^ w8_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w8_t);
+    w9_t = hc_rotl32 ((w6_t ^ w1_t ^ wb_t ^ w9_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w9_t);
+    wa_t = hc_rotl32 ((w7_t ^ w2_t ^ wc_t ^ wa_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, wa_t);
+    wb_t = hc_rotl32 ((w8_t ^ w3_t ^ wd_t ^ wb_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, wb_t);
+    wc_t = hc_rotl32 ((w9_t ^ w4_t ^ we_t ^ wc_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, wc_t);
+    wd_t = hc_rotl32 ((wa_t ^ w5_t ^ wf_t ^ wd_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, wd_t);
+    we_t = hc_rotl32 ((wb_t ^ w6_t ^ w0_t ^ we_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, we_t);
+    wf_t = hc_rotl32 ((wc_t ^ w7_t ^ w1_t ^ wf_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, wf_t);
+    w0_t = hc_rotl32 ((wd_t ^ w8_t ^ w2_t ^ w0_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w0_t);
+    w1_t = hc_rotl32 ((we_t ^ w9_t ^ w3_t ^ w1_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w1_t);
+    w2_t = hc_rotl32 ((wf_t ^ wa_t ^ w4_t ^ w2_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w2_t);
+    w3_t = hc_rotl32 ((w0_t ^ wb_t ^ w5_t ^ w3_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w3_t);
+    w4_t = hc_rotl32 ((w1_t ^ wc_t ^ w6_t ^ w4_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w4_t);
+    w5_t = hc_rotl32 ((w2_t ^ wd_t ^ w7_t ^ w5_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w5_t);
+    w6_t = hc_rotl32 ((w3_t ^ we_t ^ w8_t ^ w6_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w6_t);
+    w7_t = hc_rotl32 ((w4_t ^ wf_t ^ w9_t ^ w7_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w7_t);
+
+    #undef K
+    #define K SHA1C02
+
+    w8_t = hc_rotl32 ((w5_t ^ w0_t ^ wa_t ^ w8_t), 1u); SHA1_STEP (SHA1_F2o, a, b, c, d, e, w8_t);
+    w9_t = hc_rotl32 ((w6_t ^ w1_t ^ wb_t ^ w9_t), 1u); SHA1_STEP (SHA1_F2o, e, a, b, c, d, w9_t);
+    wa_t = hc_rotl32 ((w7_t ^ w2_t ^ wc_t ^ wa_t), 1u); SHA1_STEP (SHA1_F2o, d, e, a, b, c, wa_t);
+    wb_t = hc_rotl32 ((w8_t ^ w3_t ^ wd_t ^ wb_t), 1u); SHA1_STEP (SHA1_F2o, c, d, e, a, b, wb_t);
+    wc_t = hc_rotl32 ((w9_t ^ w4_t ^ we_t ^ wc_t), 1u); SHA1_STEP (SHA1_F2o, b, c, d, e, a, wc_t);
+    wd_t = hc_rotl32 ((wa_t ^ w5_t ^ wf_t ^ wd_t), 1u); SHA1_STEP (SHA1_F2o, a, b, c, d, e, wd_t);
+    we_t = hc_rotl32 ((wb_t ^ w6_t ^ w0_t ^ we_t), 1u); SHA1_STEP (SHA1_F2o, e, a, b, c, d, we_t);
+    wf_t = hc_rotl32 ((wc_t ^ w7_t ^ w1_t ^ wf_t), 1u); SHA1_STEP (SHA1_F2o, d, e, a, b, c, wf_t);
+    w0_t = hc_rotl32 ((wd_t ^ w8_t ^ w2_t ^ w0_t), 1u); SHA1_STEP (SHA1_F2o, c, d, e, a, b, w0_t);
+    w1_t = hc_rotl32 ((we_t ^ w9_t ^ w3_t ^ w1_t), 1u); SHA1_STEP (SHA1_F2o, b, c, d, e, a, w1_t);
+    w2_t = hc_rotl32 ((wf_t ^ wa_t ^ w4_t ^ w2_t), 1u); SHA1_STEP (SHA1_F2o, a, b, c, d, e, w2_t);
+    w3_t = hc_rotl32 ((w0_t ^ wb_t ^ w5_t ^ w3_t), 1u); SHA1_STEP (SHA1_F2o, e, a, b, c, d, w3_t);
+    w4_t = hc_rotl32 ((w1_t ^ wc_t ^ w6_t ^ w4_t), 1u); SHA1_STEP (SHA1_F2o, d, e, a, b, c, w4_t);
+    w5_t = hc_rotl32 ((w2_t ^ wd_t ^ w7_t ^ w5_t), 1u); SHA1_STEP (SHA1_F2o, c, d, e, a, b, w5_t);
+    w6_t = hc_rotl32 ((w3_t ^ we_t ^ w8_t ^ w6_t), 1u); SHA1_STEP (SHA1_F2o, b, c, d, e, a, w6_t);
+    w7_t = hc_rotl32 ((w4_t ^ wf_t ^ w9_t ^ w7_t), 1u); SHA1_STEP (SHA1_F2o, a, b, c, d, e, w7_t);
+    w8_t = hc_rotl32 ((w5_t ^ w0_t ^ wa_t ^ w8_t), 1u); SHA1_STEP (SHA1_F2o, e, a, b, c, d, w8_t);
+    w9_t = hc_rotl32 ((w6_t ^ w1_t ^ wb_t ^ w9_t), 1u); SHA1_STEP (SHA1_F2o, d, e, a, b, c, w9_t);
+    wa_t = hc_rotl32 ((w7_t ^ w2_t ^ wc_t ^ wa_t), 1u); SHA1_STEP (SHA1_F2o, c, d, e, a, b, wa_t);
+    wb_t = hc_rotl32 ((w8_t ^ w3_t ^ wd_t ^ wb_t), 1u); SHA1_STEP (SHA1_F2o, b, c, d, e, a, wb_t);
+
+    #undef K
+    #define K SHA1C03
+
+    wc_t = hc_rotl32 ((w9_t ^ w4_t ^ we_t ^ wc_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, wc_t);
+    wd_t = hc_rotl32 ((wa_t ^ w5_t ^ wf_t ^ wd_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, wd_t);
+    we_t = hc_rotl32 ((wb_t ^ w6_t ^ w0_t ^ we_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, we_t);
+    wf_t = hc_rotl32 ((wc_t ^ w7_t ^ w1_t ^ wf_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, wf_t);
+    w0_t = hc_rotl32 ((wd_t ^ w8_t ^ w2_t ^ w0_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w0_t);
+    w1_t = hc_rotl32 ((we_t ^ w9_t ^ w3_t ^ w1_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w1_t);
+    w2_t = hc_rotl32 ((wf_t ^ wa_t ^ w4_t ^ w2_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w2_t);
+    w3_t = hc_rotl32 ((w0_t ^ wb_t ^ w5_t ^ w3_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w3_t);
+    w4_t = hc_rotl32 ((w1_t ^ wc_t ^ w6_t ^ w4_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w4_t);
+    w5_t = hc_rotl32 ((w2_t ^ wd_t ^ w7_t ^ w5_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w5_t);
+    w6_t = hc_rotl32 ((w3_t ^ we_t ^ w8_t ^ w6_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w6_t);
+    w7_t = hc_rotl32 ((w4_t ^ wf_t ^ w9_t ^ w7_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w7_t);
+    w8_t = hc_rotl32 ((w5_t ^ w0_t ^ wa_t ^ w8_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w8_t);
+    w9_t = hc_rotl32 ((w6_t ^ w1_t ^ wb_t ^ w9_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w9_t);
+    wa_t = hc_rotl32 ((w7_t ^ w2_t ^ wc_t ^ wa_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, wa_t);
+    wb_t = hc_rotl32 ((w8_t ^ w3_t ^ wd_t ^ wb_t), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, wb_t);
+
+    if (MATCHES_NONE_VS (e, e_rev)) continue;
+
+    wc_t = hc_rotl32 ((w9_t ^ w4_t ^ we_t ^ wc_t), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, wc_t);
+    wd_t = hc_rotl32 ((wa_t ^ w5_t ^ wf_t ^ wd_t), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, wd_t);
+    we_t = hc_rotl32 ((wb_t ^ w6_t ^ w0_t ^ we_t), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, we_t);
+    wf_t = hc_rotl32 ((wc_t ^ w7_t ^ w1_t ^ wf_t), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, wf_t);
+
+    COMPARE_S_SIMD (d, e, c, b);
+  }
+}
+
+KERNEL_FQ void m27200_m04 (KERN_ATTR_BASIC ())
+{
+  /**
+   * base
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  u32 w0[4];
+
+  w0[0] = pws[gid].i[ 0];
+  w0[1] = pws[gid].i[ 1];
+  w0[2] = pws[gid].i[ 2];
+  w0[3] = pws[gid].i[ 3];
+
+  u32 w1[4];
+
+  w1[0] = 0;
+  w1[1] = 0;
+  w1[2] = 0;
+  w1[3] = 0;
+
+  u32 w2[4];
+
+  w2[0] = 0;
+  w2[1] = 0;
+  w2[2] = 0;
+  w2[3] = 0;
+
+  u32 w3[4];
+
+  w3[0] = 0;
+  w3[1] = 0;
+  w3[2] = 0;
+  w3[3] = 0;
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * main
+   */
+
+  m27200m (w0, w1, w2, w3, pw_len, pws, rules_buf, combs_buf, bfs_buf, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_extra0_buf, d_extra1_buf, d_extra2_buf, d_extra3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, SALT_POS, loop_pos, loop_cnt, il_cnt, digests_cnt, DIGESTS_OFFSET, combs_mode, salt_repeat, pws_pos, gid_max);
+}
+
+KERNEL_FQ void m27200_m08 (KERN_ATTR_BASIC ())
+{
+  /**
+   * base
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  u32 w0[4];
+
+  w0[0] = pws[gid].i[ 0];
+  w0[1] = pws[gid].i[ 1];
+  w0[2] = pws[gid].i[ 2];
+  w0[3] = pws[gid].i[ 3];
+
+  u32 w1[4];
+
+  w1[0] = pws[gid].i[ 4];
+  w1[1] = pws[gid].i[ 5];
+  w1[2] = pws[gid].i[ 6];
+  w1[3] = pws[gid].i[ 7];
+
+  u32 w2[4];
+
+  w2[0] = 0;
+  w2[1] = 0;
+  w2[2] = 0;
+  w2[3] = 0;
+
+  u32 w3[4];
+
+  w3[0] = 0;
+  w3[1] = 0;
+  w3[2] = 0;
+  w3[3] = 0;
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * main
+   */
+
+  m27200m (w0, w1, w2, w3, pw_len, pws, rules_buf, combs_buf, bfs_buf, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_extra0_buf, d_extra1_buf, d_extra2_buf, d_extra3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, SALT_POS, loop_pos, loop_cnt, il_cnt, digests_cnt, DIGESTS_OFFSET, combs_mode, salt_repeat, pws_pos, gid_max);
+}
+
+KERNEL_FQ void m27200_m16 (KERN_ATTR_BASIC ())
+{
+  /**
+   * base
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  u32 w0[4];
+
+  w0[0] = pws[gid].i[ 0];
+  w0[1] = pws[gid].i[ 1];
+  w0[2] = pws[gid].i[ 2];
+  w0[3] = pws[gid].i[ 3];
+
+  u32 w1[4];
+
+  w1[0] = pws[gid].i[ 4];
+  w1[1] = pws[gid].i[ 5];
+  w1[2] = pws[gid].i[ 6];
+  w1[3] = pws[gid].i[ 7];
+
+  u32 w2[4];
+
+  w2[0] = pws[gid].i[ 8];
+  w2[1] = pws[gid].i[ 9];
+  w2[2] = pws[gid].i[10];
+  w2[3] = pws[gid].i[11];
+
+  u32 w3[4];
+
+  w3[0] = pws[gid].i[12];
+  w3[1] = pws[gid].i[13];
+  w3[2] = 0;
+  w3[3] = 0;
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * main
+   */
+
+  m27200m (w0, w1, w2, w3, pw_len, pws, rules_buf, combs_buf, bfs_buf, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_extra0_buf, d_extra1_buf, d_extra2_buf, d_extra3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, SALT_POS, loop_pos, loop_cnt, il_cnt, digests_cnt, DIGESTS_OFFSET, combs_mode, salt_repeat, pws_pos, gid_max);
+}
+
+KERNEL_FQ void m27200_s04 (KERN_ATTR_BASIC ())
+{
+  /**
+   * base
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  u32 w0[4];
+
+  w0[0] = pws[gid].i[ 0];
+  w0[1] = pws[gid].i[ 1];
+  w0[2] = pws[gid].i[ 2];
+  w0[3] = pws[gid].i[ 3];
+
+  u32 w1[4];
+
+  w1[0] = 0;
+  w1[1] = 0;
+  w1[2] = 0;
+  w1[3] = 0;
+
+  u32 w2[4];
+
+  w2[0] = 0;
+  w2[1] = 0;
+  w2[2] = 0;
+  w2[3] = 0;
+
+  u32 w3[4];
+
+  w3[0] = 0;
+  w3[1] = 0;
+  w3[2] = 0;
+  w3[3] = 0;
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * main
+   */
+
+  m27200s (w0, w1, w2, w3, pw_len, pws, rules_buf, combs_buf, bfs_buf, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_extra0_buf, d_extra1_buf, d_extra2_buf, d_extra3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, SALT_POS, loop_pos, loop_cnt, il_cnt, digests_cnt, DIGESTS_OFFSET, combs_mode, salt_repeat, pws_pos, gid_max);
+}
+
+KERNEL_FQ void m27200_s08 (KERN_ATTR_BASIC ())
+{
+  /**
+   * base
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  u32 w0[4];
+
+  w0[0] = pws[gid].i[ 0];
+  w0[1] = pws[gid].i[ 1];
+  w0[2] = pws[gid].i[ 2];
+  w0[3] = pws[gid].i[ 3];
+
+  u32 w1[4];
+
+  w1[0] = pws[gid].i[ 4];
+  w1[1] = pws[gid].i[ 5];
+  w1[2] = pws[gid].i[ 6];
+  w1[3] = pws[gid].i[ 7];
+
+  u32 w2[4];
+
+  w2[0] = 0;
+  w2[1] = 0;
+  w2[2] = 0;
+  w2[3] = 0;
+
+  u32 w3[4];
+
+  w3[0] = 0;
+  w3[1] = 0;
+  w3[2] = 0;
+  w3[3] = 0;
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * main
+   */
+
+  m27200s (w0, w1, w2, w3, pw_len, pws, rules_buf, combs_buf, bfs_buf, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_extra0_buf, d_extra1_buf, d_extra2_buf, d_extra3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, SALT_POS, loop_pos, loop_cnt, il_cnt, digests_cnt, DIGESTS_OFFSET, combs_mode, salt_repeat, pws_pos, gid_max);
+}
+
+KERNEL_FQ void m27200_s16 (KERN_ATTR_BASIC ())
+{
+  /**
+   * base
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  u32 w0[4];
+
+  w0[0] = pws[gid].i[ 0];
+  w0[1] = pws[gid].i[ 1];
+  w0[2] = pws[gid].i[ 2];
+  w0[3] = pws[gid].i[ 3];
+
+  u32 w1[4];
+
+  w1[0] = pws[gid].i[ 4];
+  w1[1] = pws[gid].i[ 5];
+  w1[2] = pws[gid].i[ 6];
+  w1[3] = pws[gid].i[ 7];
+
+  u32 w2[4];
+
+  w2[0] = pws[gid].i[ 8];
+  w2[1] = pws[gid].i[ 9];
+  w2[2] = pws[gid].i[10];
+  w2[3] = pws[gid].i[11];
+
+  u32 w3[4];
+
+  w3[0] = pws[gid].i[12];
+  w3[1] = pws[gid].i[13];
+  w3[2] = 0;
+  w3[3] = 0;
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * main
+   */
+
+  m27200s (w0, w1, w2, w3, pw_len, pws, rules_buf, combs_buf, bfs_buf, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_extra0_buf, d_extra1_buf, d_extra2_buf, d_extra3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, SALT_POS, loop_pos, loop_cnt, il_cnt, digests_cnt, DIGESTS_OFFSET, combs_mode, salt_repeat, pws_pos, gid_max);
+}
diff --git a/OpenCL/m27200_a3-pure.cl b/OpenCL/m27200_a3-pure.cl
new file mode 100644
index 000000000..4d7b9215f
--- /dev/null
+++ b/OpenCL/m27200_a3-pure.cl
@@ -0,0 +1,169 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include "inc_vendor.h"
+#include "inc_types.h"
+#include "inc_platform.cl"
+#include "inc_common.cl"
+#include "inc_simd.cl"
+#include "inc_hash_sha1.cl"
+#endif
+
+KERNEL_FQ void m27200_mxx (KERN_ATTR_VECTOR ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  const u32x dash_vector[16] = { 0x2d2d0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
+  const u32 dash_scalar[16] = { 0x2d2d0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
+
+  /**
+   * base
+   */
+
+  const u32 pw_len = pws[gid].pw_len;
+
+  u32x w[64] = { 0 };
+
+  for (u32 i = 0, idx = 0; i < pw_len; i += 4, idx += 1)
+  {
+    w[idx] = pws[gid].i[idx];
+  }
+
+  sha1_ctx_t ctx0;
+
+  sha1_init (&ctx0);
+
+  ctx0.w0[0] = dash_scalar[0];
+  ctx0.w0[1] = dash_scalar[1];
+
+  ctx0.len = 2;
+
+  sha1_update_global_swap (&ctx0, salt_bufs[SALT_POS].salt_buf, salt_bufs[SALT_POS].salt_len);
+  sha1_update (&ctx0, dash_scalar, 2);
+
+  /**
+   * loop
+   */
+
+  u32x w0l = w[0];
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
+  {
+    const u32x w0r = words_buf_r[il_pos / VECT_SIZE];
+
+    const u32x w0 = w0l | w0r;
+
+    w[0] = w0;
+
+    sha1_ctx_vector_t ctx;
+
+    sha1_init_vector_from_scalar (&ctx, &ctx0);
+
+    sha1_update_vector (&ctx, w, pw_len);
+    sha1_update_vector (&ctx, dash_vector, 2);
+
+    sha1_final_vector (&ctx);
+
+    const u32x r0 = ctx.h[DGST_R0];
+    const u32x r1 = ctx.h[DGST_R1];
+    const u32x r2 = ctx.h[DGST_R2];
+    const u32x r3 = ctx.h[DGST_R3];
+
+    COMPARE_M_SIMD (r0, r1, r2, r3);
+  }
+}
+
+KERNEL_FQ void m27200_sxx (KERN_ATTR_VECTOR ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  const u32x dash_vector[16] = { 0x2d2d0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
+  const u32 dash_scalar[16] = { 0x2d2d0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
+  
+  /**
+   * digest
+   */
+
+  const u32 search[4] =
+  {
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R0],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R1],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R2],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R3]
+  };
+
+  /**
+   * base
+   */
+
+  const u32 pw_len = pws[gid].pw_len;
+
+  u32x w[64] = { 0 };
+
+  for (u32 i = 0, idx = 0; i < pw_len; i += 4, idx += 1)
+  {
+    w[idx] = pws[gid].i[idx];
+  }
+
+  sha1_ctx_t ctx0;
+
+  sha1_init (&ctx0);
+
+  ctx0.w0[0] = dash_scalar[0];
+  ctx0.w0[1] = dash_scalar[1];
+
+  ctx0.len = 2;
+  
+  sha1_update_global_swap (&ctx0, salt_bufs[SALT_POS].salt_buf, salt_bufs[SALT_POS].salt_len);
+  sha1_update (&ctx0, dash_scalar, 2);
+
+  /**
+   * loop
+   */
+
+  u32x w0l = w[0];
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
+  {
+    const u32x w0r = words_buf_r[il_pos / VECT_SIZE];
+
+    const u32x w0 = w0l | w0r;
+
+    w[0] = w0;
+
+    sha1_ctx_vector_t ctx;
+
+    sha1_init_vector_from_scalar (&ctx, &ctx0);
+
+    sha1_update_vector (&ctx, w, pw_len);
+    sha1_update_vector (&ctx, dash_vector, 2);
+
+    sha1_final_vector (&ctx);
+
+    const u32x r0 = ctx.h[DGST_R0];
+    const u32x r1 = ctx.h[DGST_R1];
+    const u32x r2 = ctx.h[DGST_R2];
+    const u32x r3 = ctx.h[DGST_R3];
+
+    COMPARE_S_SIMD (r0, r1, r2, r3);
+  }
+}
diff --git a/OpenCL/m27300-pure.cl b/OpenCL/m27300-pure.cl
new file mode 100644
index 000000000..308c46483
--- /dev/null
+++ b/OpenCL/m27300-pure.cl
@@ -0,0 +1,495 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+//#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include "inc_vendor.h"
+#include "inc_types.h"
+#include "inc_platform.cl"
+#include "inc_common.cl"
+#include "inc_simd.cl"
+#include "inc_hash_sha512.cl"
+#endif
+
+#define COMPARE_S "inc_comp_single.cl"
+#define COMPARE_M "inc_comp_multi.cl"
+
+#define SNMPV3_SALT_MAX             1500
+#define SNMPV3_ENGINEID_MAX         34
+#define SNMPV3_MSG_AUTH_PARAMS_MAX  48
+#define SNMPV3_ROUNDS               1048576
+#define SNMPV3_MAX_PW_LENGTH        128
+
+#define SNMPV3_TMP_ELEMS            8192 // 8192 = (256 (max pw length) * SNMPV3_MAX_PW_LENGTH) / sizeof (u32)
+#define SNMPV3_HASH_ELEMS           8
+
+#define SNMPV3_MAX_SALT_ELEMS       512 // 512 * 4 = 2048 > 1500, also has to be multiple of SNMPV3_MAX_PW_LENGTH
+#define SNMPV3_MAX_ENGINE_ELEMS     32  // 32 * 4 = 128 > 34, also has to be multiple of SNMPV3_MAX_PW_LENGTH
+#define SNMPV3_MAX_PNUM_ELEMS       4   // 4 * 4 = 16 > 9
+
+#define SNMPV3_MAX_PW_LENGTH_OPT    32
+#define SNMPV3_TMP_ELEMS_OPT        ((SNMPV3_MAX_PW_LENGTH_OPT * SNMPV3_MAX_PW_LENGTH) / 4)
+                                    // (32 * 128) / 4 = 1024
+                                    // for pw length > 32 we use global memory reads
+
+typedef struct hmac_sha512_tmp
+{
+  u32 tmp[SNMPV3_TMP_ELEMS];
+  u64 h[SNMPV3_HASH_ELEMS];
+
+} hmac_sha512_tmp_t;
+
+typedef struct snmpv3
+{
+  u32 salt_buf[SNMPV3_MAX_SALT_ELEMS];
+  u32 salt_len;
+
+  u32 engineID_buf[SNMPV3_MAX_ENGINE_ELEMS];
+  u32 engineID_len;
+
+  u32 packet_number[SNMPV3_MAX_PNUM_ELEMS];
+
+} snmpv3_t;
+
+KERNEL_FQ void m27300_init (KERN_ATTR_TMPS_ESALT (hmac_sha512_tmp_t, snmpv3_t))
+{
+  /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  const u32 pw_len = pws[gid].pw_len;
+
+  u32 w[128] = { 0 };
+
+  for (u32 i = 0, idx = 0; i < pw_len; i += 4, idx += 1)
+  {
+    w[idx] = pws[gid].i[idx];
+  }
+
+  u8 *src_ptr = (u8 *) w;
+
+  // password 128 times, also swapped
+
+  u32 dst_buf[32];
+
+  u8 *dst_ptr = (u8 *) dst_buf;
+
+  int tmp_idx = 0;
+
+  for (int i = 0; i < 128; i++)
+  {
+    for (u32 j = 0; j < pw_len; j++)
+    {
+      const int dst_idx = tmp_idx & 127;
+
+      dst_ptr[dst_idx] = src_ptr[j];
+
+      // write to global memory every time 64 byte are written into cache
+
+      if (dst_idx == 127)
+      {
+        const int tmp_idx4 = (tmp_idx - 127) / 4;
+
+        tmps[gid].tmp[tmp_idx4 +  0] = hc_swap32_S (dst_buf[ 0]);
+        tmps[gid].tmp[tmp_idx4 +  1] = hc_swap32_S (dst_buf[ 1]);
+        tmps[gid].tmp[tmp_idx4 +  2] = hc_swap32_S (dst_buf[ 2]);
+        tmps[gid].tmp[tmp_idx4 +  3] = hc_swap32_S (dst_buf[ 3]);
+        tmps[gid].tmp[tmp_idx4 +  4] = hc_swap32_S (dst_buf[ 4]);
+        tmps[gid].tmp[tmp_idx4 +  5] = hc_swap32_S (dst_buf[ 5]);
+        tmps[gid].tmp[tmp_idx4 +  6] = hc_swap32_S (dst_buf[ 6]);
+        tmps[gid].tmp[tmp_idx4 +  7] = hc_swap32_S (dst_buf[ 7]);
+        tmps[gid].tmp[tmp_idx4 +  8] = hc_swap32_S (dst_buf[ 8]);
+        tmps[gid].tmp[tmp_idx4 +  9] = hc_swap32_S (dst_buf[ 9]);
+        tmps[gid].tmp[tmp_idx4 + 10] = hc_swap32_S (dst_buf[10]);
+        tmps[gid].tmp[tmp_idx4 + 11] = hc_swap32_S (dst_buf[11]);
+        tmps[gid].tmp[tmp_idx4 + 12] = hc_swap32_S (dst_buf[12]);
+        tmps[gid].tmp[tmp_idx4 + 13] = hc_swap32_S (dst_buf[13]);
+        tmps[gid].tmp[tmp_idx4 + 14] = hc_swap32_S (dst_buf[14]);
+        tmps[gid].tmp[tmp_idx4 + 15] = hc_swap32_S (dst_buf[15]);
+        tmps[gid].tmp[tmp_idx4 + 16] = hc_swap32_S (dst_buf[16]);
+        tmps[gid].tmp[tmp_idx4 + 17] = hc_swap32_S (dst_buf[17]);
+        tmps[gid].tmp[tmp_idx4 + 18] = hc_swap32_S (dst_buf[18]);
+        tmps[gid].tmp[tmp_idx4 + 19] = hc_swap32_S (dst_buf[19]);
+        tmps[gid].tmp[tmp_idx4 + 20] = hc_swap32_S (dst_buf[20]);
+        tmps[gid].tmp[tmp_idx4 + 21] = hc_swap32_S (dst_buf[21]);
+        tmps[gid].tmp[tmp_idx4 + 22] = hc_swap32_S (dst_buf[22]);
+        tmps[gid].tmp[tmp_idx4 + 23] = hc_swap32_S (dst_buf[23]);
+        tmps[gid].tmp[tmp_idx4 + 24] = hc_swap32_S (dst_buf[24]);
+        tmps[gid].tmp[tmp_idx4 + 25] = hc_swap32_S (dst_buf[25]);
+        tmps[gid].tmp[tmp_idx4 + 26] = hc_swap32_S (dst_buf[26]);
+        tmps[gid].tmp[tmp_idx4 + 27] = hc_swap32_S (dst_buf[27]);
+        tmps[gid].tmp[tmp_idx4 + 28] = hc_swap32_S (dst_buf[28]);
+        tmps[gid].tmp[tmp_idx4 + 29] = hc_swap32_S (dst_buf[29]);
+        tmps[gid].tmp[tmp_idx4 + 30] = hc_swap32_S (dst_buf[30]);
+        tmps[gid].tmp[tmp_idx4 + 31] = hc_swap32_S (dst_buf[31]);
+      }
+
+      tmp_idx++;
+    }
+  }
+
+  // hash
+
+  tmps[gid].h[0] = SHA512M_A;
+  tmps[gid].h[1] = SHA512M_B;
+  tmps[gid].h[2] = SHA512M_C;
+  tmps[gid].h[3] = SHA512M_D;
+  tmps[gid].h[4] = SHA512M_E;
+  tmps[gid].h[5] = SHA512M_F;
+  tmps[gid].h[6] = SHA512M_G;
+  tmps[gid].h[7] = SHA512M_H;
+}
+
+KERNEL_FQ void m27300_loop (KERN_ATTR_TMPS_ESALT (hmac_sha512_tmp_t, snmpv3_t))
+{
+  /**
+   * base
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  u64 h[8];
+
+  h[0] = tmps[gid].h[0];
+  h[1] = tmps[gid].h[1];
+  h[2] = tmps[gid].h[2];
+  h[3] = tmps[gid].h[3];
+  h[4] = tmps[gid].h[4];
+  h[5] = tmps[gid].h[5];
+  h[6] = tmps[gid].h[6];
+  h[7] = tmps[gid].h[7];
+
+  const u32 pw_len = pws[gid].pw_len;
+
+  const int pw_len128 = pw_len * 128;
+
+  if (pw_len <= SNMPV3_MAX_PW_LENGTH_OPT)
+  {
+    u32 tmp[SNMPV3_TMP_ELEMS_OPT];
+
+    for (int i = 0; i < pw_len128 / 4; i++)
+    {
+      tmp[i] = tmps[gid].tmp[i];
+    }
+
+    for (u32 i = 0, j = loop_pos; i < loop_cnt; i += 128, j += 128)
+    {
+      const int idx = (j % pw_len128) / 4; // the optimization trick is to be able to do this
+
+      u32 w0[4];
+      u32 w1[4];
+      u32 w2[4];
+      u32 w3[4];
+      u32 w4[4];
+      u32 w5[4];
+      u32 w6[4];
+      u32 w7[4];
+
+      w0[0] = tmp[idx +  0];
+      w0[1] = tmp[idx +  1];
+      w0[2] = tmp[idx +  2];
+      w0[3] = tmp[idx +  3];
+      w1[0] = tmp[idx +  4];
+      w1[1] = tmp[idx +  5];
+      w1[2] = tmp[idx +  6];
+      w1[3] = tmp[idx +  7];
+      w2[0] = tmp[idx +  8];
+      w2[1] = tmp[idx +  9];
+      w2[2] = tmp[idx + 10];
+      w2[3] = tmp[idx + 11];
+      w3[0] = tmp[idx + 12];
+      w3[1] = tmp[idx + 13];
+      w3[2] = tmp[idx + 14];
+      w3[3] = tmp[idx + 15];
+      w4[0] = tmp[idx + 16];
+      w4[1] = tmp[idx + 17];
+      w4[2] = tmp[idx + 18];
+      w4[3] = tmp[idx + 19];
+      w5[0] = tmp[idx + 20];
+      w5[1] = tmp[idx + 21];
+      w5[2] = tmp[idx + 22];
+      w5[3] = tmp[idx + 23];
+      w6[0] = tmp[idx + 24];
+      w6[1] = tmp[idx + 25];
+      w6[2] = tmp[idx + 26];
+      w6[3] = tmp[idx + 27];
+      w7[0] = tmp[idx + 28];
+      w7[1] = tmp[idx + 29];
+      w7[2] = tmp[idx + 30];
+      w7[3] = tmp[idx + 31];
+
+      sha512_transform (w0, w1, w2, w3, w4, w5, w6, w7, h);
+    }
+  }
+  else
+  {
+    for (u32 i = 0, j = loop_pos; i < loop_cnt; i += 128, j += 128)
+    {
+      const int idx = (j % pw_len128) / 4; // the optimization trick is to be able to do this
+
+      u32 w0[4];
+      u32 w1[4];
+      u32 w2[4];
+      u32 w3[4];
+      u32 w4[4];
+      u32 w5[4];
+      u32 w6[4];
+      u32 w7[4];
+
+      w0[0] = tmps[gid].tmp[idx +  0];
+      w0[1] = tmps[gid].tmp[idx +  1];
+      w0[2] = tmps[gid].tmp[idx +  2];
+      w0[3] = tmps[gid].tmp[idx +  3];
+      w1[0] = tmps[gid].tmp[idx +  4];
+      w1[1] = tmps[gid].tmp[idx +  5];
+      w1[2] = tmps[gid].tmp[idx +  6];
+      w1[3] = tmps[gid].tmp[idx +  7];
+      w2[0] = tmps[gid].tmp[idx +  8];
+      w2[1] = tmps[gid].tmp[idx +  9];
+      w2[2] = tmps[gid].tmp[idx + 10];
+      w2[3] = tmps[gid].tmp[idx + 11];
+      w3[0] = tmps[gid].tmp[idx + 12];
+      w3[1] = tmps[gid].tmp[idx + 13];
+      w3[2] = tmps[gid].tmp[idx + 14];
+      w3[3] = tmps[gid].tmp[idx + 15];
+      w4[0] = tmps[gid].tmp[idx + 16];
+      w4[1] = tmps[gid].tmp[idx + 17];
+      w4[2] = tmps[gid].tmp[idx + 18];
+      w4[3] = tmps[gid].tmp[idx + 19];
+      w5[0] = tmps[gid].tmp[idx + 20];
+      w5[1] = tmps[gid].tmp[idx + 21];
+      w5[2] = tmps[gid].tmp[idx + 22];
+      w5[3] = tmps[gid].tmp[idx + 23];
+      w6[0] = tmps[gid].tmp[idx + 24];
+      w6[1] = tmps[gid].tmp[idx + 25];
+      w6[2] = tmps[gid].tmp[idx + 26];
+      w6[3] = tmps[gid].tmp[idx + 27];
+      w7[0] = tmps[gid].tmp[idx + 28];
+      w7[1] = tmps[gid].tmp[idx + 29];
+      w7[2] = tmps[gid].tmp[idx + 30];
+      w7[3] = tmps[gid].tmp[idx + 31];
+
+      sha512_transform (w0, w1, w2, w3, w4, w5, w6, w7, h);
+    }
+  }
+
+  tmps[gid].h[0] = h[0];
+  tmps[gid].h[1] = h[1];
+  tmps[gid].h[2] = h[2];
+  tmps[gid].h[3] = h[3];
+  tmps[gid].h[4] = h[4];
+  tmps[gid].h[5] = h[5];
+  tmps[gid].h[6] = h[6];
+  tmps[gid].h[7] = h[7];
+}
+
+KERNEL_FQ void m27300_comp (KERN_ATTR_TMPS_ESALT (hmac_sha512_tmp_t, snmpv3_t))
+{
+  /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  u32 w0[4];
+  u32 w1[4];
+  u32 w2[4];
+  u32 w3[4];
+  u32 w4[4];
+  u32 w5[4];
+  u32 w6[4];
+  u32 w7[4];
+
+  w0[0] = 0x80000000;
+  w0[1] = 0;
+  w0[2] = 0;
+  w0[3] = 0;
+  w1[0] = 0;
+  w1[1] = 0;
+  w1[2] = 0;
+  w1[3] = 0;
+  w2[0] = 0;
+  w2[1] = 0;
+  w2[2] = 0;
+  w2[3] = 0;
+  w3[0] = 0;
+  w3[1] = 0;
+  w3[2] = 0;
+  w3[3] = 0;
+  w4[0] = 0;
+  w4[1] = 0;
+  w4[2] = 0;
+  w4[3] = 0;
+  w5[0] = 0;
+  w5[1] = 0;
+  w5[2] = 0;
+  w5[3] = 0;
+  w6[0] = 0;
+  w6[1] = 0;
+  w6[2] = 0;
+  w6[3] = 0;
+  w7[0] = 0;
+  w7[1] = 0;
+  w7[2] = 0;
+  w7[3] = 1048576 * 8;
+
+  u64 h[8];
+
+  h[0] = tmps[gid].h[0];
+  h[1] = tmps[gid].h[1];
+  h[2] = tmps[gid].h[2];
+  h[3] = tmps[gid].h[3];
+  h[4] = tmps[gid].h[4];
+  h[5] = tmps[gid].h[5];
+  h[6] = tmps[gid].h[6];
+  h[7] = tmps[gid].h[7];
+
+  sha512_transform (w0, w1, w2, w3, w4, w5, w6, w7, h);
+
+  sha512_ctx_t ctx;
+
+  sha512_init (&ctx);
+
+  u32 w[32];
+
+  w[ 0] = h32_from_64_S (h[0]);
+  w[ 1] = l32_from_64_S (h[0]);
+  w[ 2] = h32_from_64_S (h[1]);
+  w[ 3] = l32_from_64_S (h[1]);
+  w[ 4] = h32_from_64_S (h[2]);
+  w[ 5] = l32_from_64_S (h[2]);
+  w[ 6] = h32_from_64_S (h[3]);
+  w[ 7] = l32_from_64_S (h[3]);
+  w[ 8] = h32_from_64_S (h[4]);
+  w[ 9] = l32_from_64_S (h[4]);
+  w[10] = h32_from_64_S (h[5]);
+  w[11] = l32_from_64_S (h[5]);
+  w[12] = h32_from_64_S (h[6]);
+  w[13] = l32_from_64_S (h[6]);
+  w[14] = h32_from_64_S (h[7]);
+  w[15] = l32_from_64_S (h[7]);
+  w[16] = 0;
+  w[17] = 0;
+  w[18] = 0;
+  w[19] = 0;
+  w[20] = 0;
+  w[21] = 0;
+  w[22] = 0;
+  w[23] = 0;
+  w[24] = 0;
+  w[25] = 0;
+  w[26] = 0;
+  w[27] = 0;
+  w[28] = 0;
+  w[29] = 0;
+  w[30] = 0;
+  w[31] = 0;
+
+  sha512_update (&ctx, w, 64);
+
+  sha512_update_global_swap (&ctx, esalt_bufs[DIGESTS_OFFSET].engineID_buf, esalt_bufs[DIGESTS_OFFSET].engineID_len);
+
+  w[ 0] = h32_from_64_S (h[0]);
+  w[ 1] = l32_from_64_S (h[0]);
+  w[ 2] = h32_from_64_S (h[1]);
+  w[ 3] = l32_from_64_S (h[1]);
+  w[ 4] = h32_from_64_S (h[2]);
+  w[ 5] = l32_from_64_S (h[2]);
+  w[ 6] = h32_from_64_S (h[3]);
+  w[ 7] = l32_from_64_S (h[3]);
+  w[ 8] = h32_from_64_S (h[4]);
+  w[ 9] = l32_from_64_S (h[4]);
+  w[10] = h32_from_64_S (h[5]);
+  w[11] = l32_from_64_S (h[5]);
+  w[12] = h32_from_64_S (h[6]);
+  w[13] = l32_from_64_S (h[6]);
+  w[14] = h32_from_64_S (h[7]);
+  w[15] = l32_from_64_S (h[7]);
+  w[16] = 0;
+  w[17] = 0;
+  w[18] = 0;
+  w[19] = 0;
+  w[20] = 0;
+  w[21] = 0;
+  w[22] = 0;
+  w[23] = 0;
+  w[24] = 0;
+  w[25] = 0;
+  w[26] = 0;
+  w[27] = 0;
+  w[28] = 0;
+  w[29] = 0;
+  w[30] = 0;
+  w[31] = 0;
+
+  sha512_update (&ctx, w, 64);
+
+  sha512_final (&ctx);
+
+  w[ 0] = h32_from_64_S (ctx.h[0]);
+  w[ 1] = l32_from_64_S (ctx.h[0]);
+  w[ 2] = h32_from_64_S (ctx.h[1]);
+  w[ 3] = l32_from_64_S (ctx.h[1]);
+  w[ 4] = h32_from_64_S (ctx.h[2]);
+  w[ 5] = l32_from_64_S (ctx.h[2]);
+  w[ 6] = h32_from_64_S (ctx.h[3]);
+  w[ 7] = l32_from_64_S (ctx.h[3]);
+  w[ 8] = h32_from_64_S (ctx.h[4]);
+  w[ 9] = l32_from_64_S (ctx.h[4]);
+  w[10] = h32_from_64_S (ctx.h[5]);
+  w[11] = l32_from_64_S (ctx.h[5]);
+  w[12] = h32_from_64_S (ctx.h[6]);
+  w[13] = l32_from_64_S (ctx.h[6]);
+  w[14] = h32_from_64_S (ctx.h[7]);
+  w[15] = l32_from_64_S (ctx.h[7]);
+  w[16] = 0;
+  w[17] = 0;
+  w[18] = 0;
+  w[19] = 0;
+  w[20] = 0;
+  w[21] = 0;
+  w[22] = 0;
+  w[23] = 0;
+  w[24] = 0;
+  w[25] = 0;
+  w[26] = 0;
+  w[27] = 0;
+  w[28] = 0;
+  w[29] = 0;
+  w[30] = 0;
+  w[31] = 0;
+
+  sha512_hmac_ctx_t hmac_ctx;
+
+  sha512_hmac_init (&hmac_ctx, w, 64);
+
+  sha512_hmac_update_global_swap (&hmac_ctx, esalt_bufs[DIGESTS_OFFSET].salt_buf, esalt_bufs[DIGESTS_OFFSET].salt_len);
+
+  sha512_hmac_final (&hmac_ctx);
+
+  const u32 r0 = l32_from_64 (hmac_ctx.opad.h[1]);
+  const u32 r1 = h32_from_64 (hmac_ctx.opad.h[1]);
+  const u32 r2 = l32_from_64 (hmac_ctx.opad.h[0]);
+  const u32 r3 = h32_from_64 (hmac_ctx.opad.h[0]);
+
+  #define il_pos 0
+
+  #ifdef KERNEL_STATIC
+  #include COMPARE_M
+  #endif
+}
diff --git a/OpenCL/m27400-pure.cl b/OpenCL/m27400-pure.cl
new file mode 100644
index 000000000..9373d2b68
--- /dev/null
+++ b/OpenCL/m27400-pure.cl
@@ -0,0 +1,351 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include "inc_vendor.h"
+#include "inc_types.h"
+#include "inc_platform.cl"
+#include "inc_common.cl"
+#include "inc_simd.cl"
+#include "inc_hash_sha1.cl"
+#include "inc_cipher_aes.cl"
+#endif
+
+#define COMPARE_S "inc_comp_single.cl"
+#define COMPARE_M "inc_comp_multi.cl"
+
+typedef struct pbkdf2_sha1_tmp
+{
+  u32  ipad[5];
+  u32  opad[5];
+
+  u32  dgst[32];
+  u32  out[32];
+
+} pbkdf2_sha1_tmp_t;
+
+typedef struct vmware_vmx
+{
+  u32 salt_buf[64];
+  u32 iv_buf[4];
+  u32 ct_buf[4];
+
+} vmware_vmx_t;
+
+DECLSPEC void hmac_sha1_run_V (u32x *w0, u32x *w1, u32x *w2, u32x *w3, u32x *ipad, u32x *opad, u32x *digest)
+{
+  digest[0] = ipad[0];
+  digest[1] = ipad[1];
+  digest[2] = ipad[2];
+  digest[3] = ipad[3];
+  digest[4] = ipad[4];
+
+  sha1_transform_vector (w0, w1, w2, w3, digest);
+
+  w0[0] = digest[0];
+  w0[1] = digest[1];
+  w0[2] = digest[2];
+  w0[3] = digest[3];
+  w1[0] = digest[4];
+  w1[1] = 0x80000000;
+  w1[2] = 0;
+  w1[3] = 0;
+  w2[0] = 0;
+  w2[1] = 0;
+  w2[2] = 0;
+  w2[3] = 0;
+  w3[0] = 0;
+  w3[1] = 0;
+  w3[2] = 0;
+  w3[3] = (64 + 20) * 8;
+
+  digest[0] = opad[0];
+  digest[1] = opad[1];
+  digest[2] = opad[2];
+  digest[3] = opad[3];
+  digest[4] = opad[4];
+
+  sha1_transform_vector (w0, w1, w2, w3, digest);
+}
+
+KERNEL_FQ void m27400_init (KERN_ATTR_TMPS_ESALT (pbkdf2_sha1_tmp_t, vmware_vmx_t))
+{
+  /**
+   * base
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  sha1_hmac_ctx_t sha1_hmac_ctx;
+
+  sha1_hmac_init_global_swap (&sha1_hmac_ctx, pws[gid].i, pws[gid].pw_len);
+
+  tmps[gid].ipad[0] = sha1_hmac_ctx.ipad.h[0];
+  tmps[gid].ipad[1] = sha1_hmac_ctx.ipad.h[1];
+  tmps[gid].ipad[2] = sha1_hmac_ctx.ipad.h[2];
+  tmps[gid].ipad[3] = sha1_hmac_ctx.ipad.h[3];
+  tmps[gid].ipad[4] = sha1_hmac_ctx.ipad.h[4];
+
+  tmps[gid].opad[0] = sha1_hmac_ctx.opad.h[0];
+  tmps[gid].opad[1] = sha1_hmac_ctx.opad.h[1];
+  tmps[gid].opad[2] = sha1_hmac_ctx.opad.h[2];
+  tmps[gid].opad[3] = sha1_hmac_ctx.opad.h[3];
+  tmps[gid].opad[4] = sha1_hmac_ctx.opad.h[4];
+
+  sha1_hmac_update_global_swap (&sha1_hmac_ctx, esalt_bufs[DIGESTS_OFFSET].salt_buf, salt_bufs[SALT_POS].salt_len);
+
+  for (u32 i = 0, j = 1; i < 8; i += 5, j += 1)
+  {
+    sha1_hmac_ctx_t sha1_hmac_ctx2 = sha1_hmac_ctx;
+
+    u32 w0[4];
+    u32 w1[4];
+    u32 w2[4];
+    u32 w3[4];
+
+    w0[0] = j;
+    w0[1] = 0;
+    w0[2] = 0;
+    w0[3] = 0;
+    w1[0] = 0;
+    w1[1] = 0;
+    w1[2] = 0;
+    w1[3] = 0;
+    w2[0] = 0;
+    w2[1] = 0;
+    w2[2] = 0;
+    w2[3] = 0;
+    w3[0] = 0;
+    w3[1] = 0;
+    w3[2] = 0;
+    w3[3] = 0;
+
+    sha1_hmac_update_64 (&sha1_hmac_ctx2, w0, w1, w2, w3, 4);
+
+    sha1_hmac_final (&sha1_hmac_ctx2);
+
+    tmps[gid].dgst[i + 0] = sha1_hmac_ctx2.opad.h[0];
+    tmps[gid].dgst[i + 1] = sha1_hmac_ctx2.opad.h[1];
+    tmps[gid].dgst[i + 2] = sha1_hmac_ctx2.opad.h[2];
+    tmps[gid].dgst[i + 3] = sha1_hmac_ctx2.opad.h[3];
+    tmps[gid].dgst[i + 4] = sha1_hmac_ctx2.opad.h[4];
+
+    tmps[gid].out[i + 0] = tmps[gid].dgst[i + 0];
+    tmps[gid].out[i + 1] = tmps[gid].dgst[i + 1];
+    tmps[gid].out[i + 2] = tmps[gid].dgst[i + 2];
+    tmps[gid].out[i + 3] = tmps[gid].dgst[i + 3];
+    tmps[gid].out[i + 4] = tmps[gid].dgst[i + 4];
+  }
+}
+
+KERNEL_FQ void m27400_loop (KERN_ATTR_TMPS_ESALT (pbkdf2_sha1_tmp_t, vmware_vmx_t))
+{
+  const u64 gid = get_global_id (0);
+
+  if ((gid * VECT_SIZE) >= gid_max) return;
+
+  u32x ipad[5];
+  u32x opad[5];
+
+  ipad[0] = packv (tmps, ipad, gid, 0);
+  ipad[1] = packv (tmps, ipad, gid, 1);
+  ipad[2] = packv (tmps, ipad, gid, 2);
+  ipad[3] = packv (tmps, ipad, gid, 3);
+  ipad[4] = packv (tmps, ipad, gid, 4);
+
+  opad[0] = packv (tmps, opad, gid, 0);
+  opad[1] = packv (tmps, opad, gid, 1);
+  opad[2] = packv (tmps, opad, gid, 2);
+  opad[3] = packv (tmps, opad, gid, 3);
+  opad[4] = packv (tmps, opad, gid, 4);
+
+  for (u32 i = 0; i < 8; i += 5)
+  {
+    u32x dgst[5];
+    u32x out[5];
+
+    dgst[0] = packv (tmps, dgst, gid, i + 0);
+    dgst[1] = packv (tmps, dgst, gid, i + 1);
+    dgst[2] = packv (tmps, dgst, gid, i + 2);
+    dgst[3] = packv (tmps, dgst, gid, i + 3);
+    dgst[4] = packv (tmps, dgst, gid, i + 4);
+
+    out[0] = packv (tmps, out, gid, i + 0);
+    out[1] = packv (tmps, out, gid, i + 1);
+    out[2] = packv (tmps, out, gid, i + 2);
+    out[3] = packv (tmps, out, gid, i + 3);
+    out[4] = packv (tmps, out, gid, i + 4);
+
+    for (u32 j = 0; j < loop_cnt; j++)
+    {
+      u32x w0[4];
+      u32x w1[4];
+      u32x w2[4];
+      u32x w3[4];
+
+      w0[0] = dgst[0];
+      w0[1] = dgst[1];
+      w0[2] = dgst[2];
+      w0[3] = dgst[3];
+      w1[0] = dgst[4];
+      w1[1] = 0x80000000;
+      w1[2] = 0;
+      w1[3] = 0;
+      w2[0] = 0;
+      w2[1] = 0;
+      w2[2] = 0;
+      w2[3] = 0;
+      w3[0] = 0;
+      w3[1] = 0;
+      w3[2] = 0;
+      w3[3] = (64 + 20) * 8;
+
+      hmac_sha1_run_V (w0, w1, w2, w3, ipad, opad, dgst);
+
+      out[0] ^= dgst[0];
+      out[1] ^= dgst[1];
+      out[2] ^= dgst[2];
+      out[3] ^= dgst[3];
+      out[4] ^= dgst[4];
+    }
+
+    unpackv (tmps, dgst, gid, i + 0, dgst[0]);
+    unpackv (tmps, dgst, gid, i + 1, dgst[1]);
+    unpackv (tmps, dgst, gid, i + 2, dgst[2]);
+    unpackv (tmps, dgst, gid, i + 3, dgst[3]);
+    unpackv (tmps, dgst, gid, i + 4, dgst[4]);
+
+    unpackv (tmps, out, gid, i + 0, out[0]);
+    unpackv (tmps, out, gid, i + 1, out[1]);
+    unpackv (tmps, out, gid, i + 2, out[2]);
+    unpackv (tmps, out, gid, i + 3, out[3]);
+    unpackv (tmps, out, gid, i + 4, out[4]);
+  }
+}
+
+KERNEL_FQ void m27400_comp (KERN_ATTR_TMPS_ESALT (pbkdf2_sha1_tmp_t, vmware_vmx_t))
+{
+  /**
+   * base
+   */
+
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id(0);
+  const u64 lsz = get_local_size(0);
+
+  /**
+   * aes shared
+   */
+
+  #ifdef REAL_SHM
+
+  LOCAL_VK u32 s_td0[256];
+  LOCAL_VK u32 s_td1[256];
+  LOCAL_VK u32 s_td2[256];
+  LOCAL_VK u32 s_td3[256];
+  LOCAL_VK u32 s_td4[256];
+
+  LOCAL_VK u32 s_te0[256];
+  LOCAL_VK u32 s_te1[256];
+  LOCAL_VK u32 s_te2[256];
+  LOCAL_VK u32 s_te3[256];
+  LOCAL_VK u32 s_te4[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    s_td0[i] = td0[i];
+    s_td1[i] = td1[i];
+    s_td2[i] = td2[i];
+    s_td3[i] = td3[i];
+    s_td4[i] = td4[i];
+
+    s_te0[i] = te0[i];
+    s_te1[i] = te1[i];
+    s_te2[i] = te2[i];
+    s_te3[i] = te3[i];
+    s_te4[i] = te4[i];
+  }
+
+  SYNC_THREADS();
+
+  #else
+
+  CONSTANT_AS u32a* s_td0 = td0;
+  CONSTANT_AS u32a* s_td1 = td1;
+  CONSTANT_AS u32a* s_td2 = td2;
+  CONSTANT_AS u32a* s_td3 = td3;
+  CONSTANT_AS u32a* s_td4 = td4;
+
+  CONSTANT_AS u32a* s_te0 = te0;
+  CONSTANT_AS u32a* s_te1 = te1;
+  CONSTANT_AS u32a* s_te2 = te2;
+  CONSTANT_AS u32a* s_te3 = te3;
+  CONSTANT_AS u32a* s_te4 = te4;
+
+  #endif
+
+  if (gid >= gid_max) return;
+
+  u32 ukey[8];
+
+  ukey[0] = tmps[gid].out[0];
+  ukey[1] = tmps[gid].out[1];
+  ukey[2] = tmps[gid].out[2];
+  ukey[3] = tmps[gid].out[3];
+  ukey[4] = tmps[gid].out[4];
+  ukey[5] = tmps[gid].out[5];
+  ukey[6] = tmps[gid].out[6];
+  ukey[7] = tmps[gid].out[7];
+
+  u32 ks[60];
+
+  AES256_set_decrypt_key (ks, ukey, s_te0, s_te1, s_te2, s_te3, s_td0, s_td1, s_td2, s_td3);
+
+  // iv
+
+  u32 iv_buf[4];
+
+  iv_buf[0] = esalt_bufs[DIGESTS_OFFSET].iv_buf[0];
+  iv_buf[1] = esalt_bufs[DIGESTS_OFFSET].iv_buf[1];
+  iv_buf[2] = esalt_bufs[DIGESTS_OFFSET].iv_buf[2];
+  iv_buf[3] = esalt_bufs[DIGESTS_OFFSET].iv_buf[3];
+
+  // ct
+
+  u32 ct_buf[4];
+
+  ct_buf[0] = esalt_bufs[DIGESTS_OFFSET].ct_buf[0];
+  ct_buf[1] = esalt_bufs[DIGESTS_OFFSET].ct_buf[1];
+  ct_buf[2] = esalt_bufs[DIGESTS_OFFSET].ct_buf[2];
+  ct_buf[3] = esalt_bufs[DIGESTS_OFFSET].ct_buf[3];
+
+  // decrypt first block
+
+  u32 pt_buf[4];
+
+  AES256_decrypt (ks, ct_buf, pt_buf, s_td0, s_td1, s_td2, s_td3, s_td4);
+
+  pt_buf[0] ^= iv_buf[0];
+  pt_buf[1] ^= iv_buf[1];
+  pt_buf[2] ^= iv_buf[2];
+  pt_buf[3] ^= iv_buf[3];
+
+  // check
+
+  const u32 r0 = pt_buf[0];
+  const u32 r1 = pt_buf[1];
+  const u32 r2 = pt_buf[2];
+  const u32 r3 = pt_buf[3];
+
+  #define il_pos 0
+
+  #ifdef KERNEL_STATIC
+  #include COMPARE_M
+  #endif
+}
diff --git a/OpenCL/shared.cl b/OpenCL/shared.cl
index 3cc96e79f..fe4ff4087 100644
--- a/OpenCL/shared.cl
+++ b/OpenCL/shared.cl
@@ -117,10 +117,7 @@ KERNEL_FQ void gpu_memset (GLOBAL_AS uint4 *buf, const u32 value, const u64 gid_
   #if   defined IS_NATIVE
   r = value;
   #elif defined IS_OPENCL
-  r.s0 = value;
-  r.s1 = value;
-  r.s2 = value;
-  r.s3 = value;
+  r = (uint4) (value);
   #elif defined IS_CUDA
   r.x = value;
   r.y = value;
@@ -136,6 +133,33 @@ KERNEL_FQ void gpu_memset (GLOBAL_AS uint4 *buf, const u32 value, const u64 gid_
   buf[gid] = r;
 }
 
+KERNEL_FQ void gpu_bzero(GLOBAL_AS uint4* buf, const u64 gid_max)
+{
+  const u64 gid = get_global_id(0);
+
+  if (gid >= gid_max) return;
+
+  uint4 r;
+
+  #if   defined IS_NATIVE
+  r = 0;
+  #elif defined IS_OPENCL
+  r = (uint4) (0);
+  #elif defined IS_CUDA
+  r.x = 0;
+  r.y = 0;
+  r.z = 0;
+  r.w = 0;
+  #elif defined IS_HIP
+  r.x = 0;
+  r.y = 0;
+  r.z = 0;
+  r.w = 0;
+  #endif
+
+  buf[gid] = r;
+}
+
 KERNEL_FQ void gpu_atinit (GLOBAL_AS pw_t *buf, const u64 gid_max)
 {
   const u64 gid = get_global_id (0);
diff --git a/README.md b/README.md
index 438dd77f3..d102d505c 100644
--- a/README.md
+++ b/README.md
@@ -12,7 +12,7 @@ Download the [latest release](https://hashcat.net/hashcat/) and unpack it in the
 
 ### Usage/Help ###
 
-Please refer to the [Hashcat Wiki](https://hashcat.net/wiki/) and the output of `--help` for usage information and general help. A list of frequently asked questions may also be found [here](https://hashcat.net/wiki/doku.php?id=frequently_asked_questions). The [Hashcat Forum](https://hashcat.net/forum/) also contains a plethora of information.
+Please refer to the [Hashcat Wiki](https://hashcat.net/wiki/) and the output of `--help` for usage information and general help. A list of frequently asked questions may also be found [here](https://hashcat.net/wiki/doku.php?id=frequently_asked_questions). The [Hashcat Forum](https://hashcat.net/forum/) also contains a plethora of information. If you still think you need help by a real human come to [Discord](https://discord.gg/HFS523HGBT).
 
 ### Building ###
 
diff --git a/docs/changes.txt b/docs/changes.txt
index c7b8b35ca..828eec5d3 100644
--- a/docs/changes.txt
+++ b/docs/changes.txt
@@ -1,9 +1,83 @@
 * changes v6.2.3 -> v6.2.x
 
+##
+## Features
+##
+
+- Added option --deprecated-check-disable to enable deprecated plugins
+- Added option --multiply-accel-disable (short: -M) to disable multiplying of the kernel accel with the multiprocessor count
+- Added rule function '3' to change the case of the first letter after the occurrence of N of character X
+- Added support for auto tuning --kernel-threads (-T) at startup
+- Added support for HIP version 4.3 or later and removed support for older HIP versions as they are not compatible
+
+##
+## Bugs
+##
+
+- Fixed broken support for --generate-rules-func-min which was ignored under certain conditions
+- Fixed buffer overflow in DPAPI masterkey file v1/v2 module in hash_encode() and hash_decode()
+- Fixed buffer overflow in Stargazer Stellar Wallet XLM module in hash_encode() when a hash was cracked
+- Fixed false negative in all VeraCrypt hash modes if both conditions are met: 1. Use CPU for cracking and 2. PIM area was used
+- Fixed invalid data type in the sha384_hmac_init_vector_128() function that take effect if the vector data type was specified manually
+- Fixed out-of-boundary read in input_tokenizer() if the signature in the hash is longer than the length of the plugin's signature constant
+- Fixed out-of-boundary read in the Stuffit5 module in hash_decode()
+- Fixed random rule generator option --generate-rules-func-min by fixing switch() case to not select a not existing option group type
+- Fixed syntax check of HAS_VPERM macro in several kernel includes causing invalid error message for AMD GPUs on Windows
+- Fixed uninitialized tmps variable in autotune for slow hashes by calling _init and _prepare kernel before calling _loop kernel
+
+##
+## Performance
+##
+
+- AMD GPUs: Add inline assembly code for md5crypt, sha256crypt, PDF 1.7, 7-Zip, RAR3, Samsung Android and Windows Phone 8+
+- AMD GPUs: On the Apple OpenCL platform, we ask for the preferred kernel thread size rather than hard-coding 32
+- Backend Interface: Replace most of the blocking Compute API functions with asynchronous ones to improve GPU utilization
+- Blake Kernels: Optimize 3/4 BLAKE2B_ROUND() 64-bit rotations with inline assembly hc_byte_perm_S() calls
+- Blowfish Kernels: Backport optimizations to reduce bank conflicts from bcrypt to Password Safe v2 and Open Document Format (ODF) 1.1
+- ECC secp256k1: The inline assembly code for AMD GPUs has been removed as the latest JIT compilers optimize it with the same efficiency
+- HIP Kernels: Enable vector data types for HIP kernels for functionality and performance
+- Kernel threads: Use warp size / wavefront size query instead of hard-coded values as the basis for kernel threads
+- SCRYPT Kernels: Improve Hashcat.hctune entries for many NV and AMD GPUs for hash mode 8900, 9300, 15700 and 22700
+- Tuning Database: Add new module function module_extra_tuningdb_block() to extend hashcat.hctune content from a module
+
+##
+## Technical
+##
+
+- 7-Zip Hook: Increase the supported data length from 320kb to 8mb
+- ADL: Updated support for AMD Display Library to 15.0, updated data types
+- AMD Driver: Updated requirements for AMD Linux drivers to ROCm 4.3 or later due to new HIP interface
+- AMD Driver: Updated requirements for AMD Windows drivers to Adrenalin 21.2.1 or later due to new ADL library
+- Backend Interface: Implement gpu_bzero() as a gpu_memset() replacement, since all gpu_memset() operations used 0 as the value
+- Backend Interface: Improve the query kernel's dynamic memory size based on DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK_OPTIN instead of BF
+- Brain Session: Adds hashconfig-specific opti_type and opts_type parameters to the session calculation to enable cover functions like -O
+- Commandline: Throw an error if the separator specified by the user with the -p option is not exactly 1 byte
+- Constants: Make const char * pointers actually const char * const pointers
+- Deprecated Plugins: Add new module function module_deprecated_notice() to mark a plugin as deprecated and to return a free text user notice
+- Deprecated Plugins: Marked plugins 2500/2501 and 16800/16801 as deprecated
+- Encoding: Truncate password candidates in UTF8 -> UTF16 conversion if it contains an invalid UTF8 byte sequence
+- Filehandling: Use const char for fopen mode to fix -Wwrite-strings warnings
+- Hardware Monitor: Added support for OverDrive 7 and 8 based GPUs
+- HIP Kernels: Dependency on hip/hip runtime.h has been removed to enable easier integration of the HIP backend under Windows
+- Kernel cache: Add kernel threads for hash calculation, which will later be used in the file name of the kernel cache
+- Memory Management: Refactored the code responsible for limiting kernel accel with the goal to avoid low host memory situations
+- OpenCL Runtime: Workaround for Intel OpenCL runtime: segmentation fault when compiling hc_enc_next() / hc_enc_next_global()
+- RC4 Kernels: Use improved native thread derivation for RC4-based hash modes 7500, 13100, 18200, 25400
+- Shared Memory: Calculate the dynamic memory size of the kernel based on CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK_OPTIN
+- Slow kernels: Set some of the slowest kernels to OPTS_TYPE_MP_MULTI_DISABLE to make it easier to handle small word lists
+- Vendor Discovery: Add "Intel" as a valid vendor name for GPUs on macOS
+
 ##
 ## Algorithms
 ##
 
+- Added hash-mode: SNMPv3 HMAC-MD5-96/HMAC-SHA1-96
+- Added hash-mode: SNMPv3 HMAC-MD5-96
+- Added hash-mode: SNMPv3 HMAC-SHA1-96
+- Added hash-mode: SNMPv3 HMAC-SHA224-128
+- Added hash-mode: SNMPv3 HMAC-SHA256-192
+- Added hash-mode: SNMPv3 HMAC-SHA384-256
+- Added hash-mode: SNMPv3 HMAC-SHA512-384
 - Added hash-mode: VirtualBox (PBKDF2-HMAC-SHA256 & AES-128-XTS)
 - Added hash-mode: VirtualBox (PBKDF2-HMAC-SHA256 & AES-256-XTS)
 
diff --git a/docs/contact.txt b/docs/contact.txt
index 1ffcbce30..edfe3e1f7 100644
--- a/docs/contact.txt
+++ b/docs/contact.txt
@@ -1,2 +1,3 @@
 web: https://hashcat.net
-irc: freenode #hashcat
+discord: https://discord.gg/HFS523HGBT
+irc: Libera Chat #hashcat
diff --git a/docs/readme.txt b/docs/readme.txt
index 107393cde..eef13b8fa 100644
--- a/docs/readme.txt
+++ b/docs/readme.txt
@@ -10,8 +10,8 @@
 hashcat v6.2.3
 ==============
 
-AMD GPUs on Linux require "RadeonOpenCompute (ROCm)" Software Platform (3.1 or later)
-AMD GPUs on Windows require "AMD Radeon Adrenalin 2020 Edition" (20.2.2 or later)
+AMD GPUs on Linux require "AMD ROCm" (4.3 or later)
+AMD GPUs on Windows require "AMD Radeon Adrenalin 2020 Edition" (21.2.1 or later)
 Intel CPUs require "OpenCL Runtime for Intel Core and Intel Xeon Processors" (16.1.1 or later)
 NVIDIA GPUs require "NVIDIA Driver" (440.64 or later) and "CUDA Toolkit" (9.0 or later)
 
@@ -155,6 +155,13 @@ NVIDIA GPUs require "NVIDIA Driver" (440.64 or later) and "CUDA Toolkit" (9.0 or
 - SIP digest authentication (MD5)
 - IKE-PSK MD5
 - IKE-PSK SHA1
+- SNMPv3 HMAC-MD5-96/HMAC-SHA1-96
+- SNMPv3 HMAC-MD5-96
+- SNMPv3 HMAC-SHA1-96
+- SNMPv3 HMAC-SHA224-128
+- SNMPv3 HMAC-SHA256-192
+- SNMPv3 HMAC-SHA384-256
+- SNMPv3 HMAC-SHA512-384
 - WPA-EAPOL-PBKDF2
 - WPA-EAPOL-PMK
 - WPA-PBKDF2-PMKID+EAPOL
@@ -265,6 +272,7 @@ NVIDIA GPUs require "NVIDIA Driver" (440.64 or later) and "CUDA Toolkit" (9.0 or
 - Huawei sha1(md5($pass).$salt)
 - AuthMe sha256
 - AES Crypt (SHA256)
+- VMware VMX (PBKDF2-HMAC-SHA1 + AES-256-CBC)
 - LUKS
 - VeraCrypt
 - BestCrypt v3 Volume Encryption
diff --git a/docs/rules.txt b/docs/rules.txt
index b2621bba4..59233d1ad 100644
--- a/docs/rules.txt
+++ b/docs/rules.txt
@@ -5,6 +5,7 @@
 #define RULE_OP_MANGLE_UREST_LFIRST     'C' // upper case all chars, lower case 1st
 #define RULE_OP_MANGLE_TREST            't' // switch the case of each char
 #define RULE_OP_MANGLE_TOGGLE_AT        'T' // switch the case of each char on pos N
+#define RULE_OP_MANGLE_TOGGLE_AT_SEP    '3' // switch the case of the first letter after occurrence N of char X
 #define RULE_OP_MANGLE_REVERSE          'r' // reverse word
 #define RULE_OP_MANGLE_DUPEWORD         'd' // append word to itself
 #define RULE_OP_MANGLE_DUPEWORD_TIMES   'p' // append word to itself N times
diff --git a/docs/team.txt b/docs/team.txt
index bc243d6d9..eca7c93a4 100644
--- a/docs/team.txt
+++ b/docs/team.txt
@@ -24,6 +24,7 @@ We're a group of people participating in the yearly repeating password cracking
 | CracktheCon         | Cyphercon, Milwaukee       | 2019 | 1st    |
 | Crack Me If You Can | DEF CON, Las Vegas         | 2019 | 1st    |
 | Crack Me If You Can | DEF CON, Remote            | 2020 | 1st    |
+| Crack Me If You Can | DEF CON, Las Vegas         | 2021 | 1st    |
 
 * Special recognition for team hashcat goes to:
 
diff --git a/extra/tab_completion/hashcat.sh b/extra/tab_completion/hashcat.sh
index 54440e636..dac5d8adf 100644
--- a/extra/tab_completion/hashcat.sh
+++ b/extra/tab_completion/hashcat.sh
@@ -426,7 +426,7 @@ _hashcat ()
   local BUILD_IN_CHARSETS='?l ?u ?d ?a ?b ?s ?h ?H'
 
   local SHORT_OPTS="-m -a -V -h -b -t -T -o -p -c -d -D -w -n -u -j -k -r -g -1 -2 -3 -4 -i -I -s -l -O -S -z"
-  local LONG_OPTS="--hash-type --attack-mode --version --help --quiet --benchmark --benchmark-all --hex-salt --hex-wordlist --hex-charset --force --status --status-json --status-timer --stdin-timeout-abort --machine-readable --loopback --markov-hcstat2 --markov-disable --markov-inverse --markov-classic --markov-threshold --runtime --session --speed-only --progress-only --restore --restore-file-path --restore-disable --outfile --outfile-format --outfile-autohex-disable --outfile-check-timer --outfile-check-dir --wordlist-autohex-disable --separator --show --left --username --remove --remove-timer --potfile-disable --potfile-path --debug-mode --debug-file --induction-dir --segment-size --bitmap-min --bitmap-max --cpu-affinity --example-hashes --hash-info --backend-ignore-cuda --backend-ignore-opencl --backend-info --backend-devices --opencl-device-types --backend-vector-width --workload-profile --kernel-accel --kernel-loops --kernel-threads --spin-damp --hwmon-disable --hwmon-temp-abort --skip --limit --keyspace --rule-left --rule-right --rules-file --generate-rules --generate-rules-func-min --generate-rules-func-max --generate-rules-seed --custom-charset1 --custom-charset2 --custom-charset3 --custom-charset4 --hook-threads --increment --increment-min --increment-max --logfile-disable --scrypt-tmto --keyboard-layout-mapping --truecrypt-keyfiles --veracrypt-keyfiles --veracrypt-pim-start --veracrypt-pim-stop --stdout --keep-guessing --hccapx-message-pair --nonce-error-corrections --encoding-from --encoding-to --optimized-kernel-enable --self-test-disable  --slow-candidates --brain-server --brain-server-timer --brain-client --brain-client-features --brain-host --brain-port --brain-session --brain-session-whitelist --brain-password --identify"
+  local LONG_OPTS="--hash-type --attack-mode --version --help --quiet --benchmark --benchmark-all --hex-salt --hex-wordlist --hex-charset --force --status --status-json --status-timer --stdin-timeout-abort --machine-readable --loopback --markov-hcstat2 --markov-disable --markov-inverse --markov-classic --markov-threshold --runtime --session --speed-only --progress-only --restore --restore-file-path --restore-disable --outfile --outfile-format --outfile-autohex-disable --outfile-check-timer --outfile-check-dir --wordlist-autohex-disable --separator --show --deprecated-check-disable --left --username --remove --remove-timer --potfile-disable --potfile-path --debug-mode --debug-file --induction-dir --segment-size --bitmap-min --bitmap-max --cpu-affinity --example-hashes --hash-info --backend-ignore-cuda --backend-ignore-opencl --backend-info --backend-devices --opencl-device-types --backend-vector-width --workload-profile --kernel-accel --kernel-loops --kernel-threads --spin-damp --hwmon-disable --hwmon-temp-abort --skip --limit --keyspace --rule-left --rule-right --rules-file --generate-rules --generate-rules-func-min --generate-rules-func-max --generate-rules-seed --custom-charset1 --custom-charset2 --custom-charset3 --custom-charset4 --hook-threads --increment --increment-min --increment-max --logfile-disable --scrypt-tmto --keyboard-layout-mapping --truecrypt-keyfiles --veracrypt-keyfiles --veracrypt-pim-start --veracrypt-pim-stop --stdout --keep-guessing --hccapx-message-pair --nonce-error-corrections --encoding-from --encoding-to --optimized-kernel-enable --multiply-accel-disable --self-test-disable  --slow-candidates --brain-server --brain-server-timer --brain-client --brain-client-features --brain-host --brain-port --brain-session --brain-session-whitelist --brain-password --identify"
   local OPTIONS="-m -a -t -o -p -c -d -w -n -u -j -k -r -g -1 -2 -3 -4 -s -l --hash-type --attack-mode --status-timer --stdin-timeout-abort --markov-hcstat2 --markov-threshold --runtime --session --timer --outfile --outfile-format --outfile-check-timer --outfile-check-dir --separator --remove-timer --potfile-path --restore-file-path --debug-mode --debug-file --induction-dir --segment-size --bitmap-min --bitmap-max --cpu-affinity --backend-devices --opencl-device-types --backend-vector-width --workload-profile --kernel-accel --kernel-loops --kernel-threads --spin-damp --hwmon-temp-abort --skip --limit --rule-left --rule-right --rules-file --generate-rules --generate-rules-func-min --generate-rules-func-max --generate-rules-seed --custom-charset1 --custom-charset2 --custom-charset3 --custom-charset4 --hook-threads --increment-min --increment-max --scrypt-tmto --keyboard-layout-mapping --truecrypt-keyfiles --veracrypt-keyfiles --veracrypt-pim-start --veracrypt-pim-stop --hccapx-message-pair --nonce-error-corrections --encoding-from --encoding-to --brain-server-timer --brain-client-features --brain-host --brain-password --brain-port --brain-session --brain-session-whitelist"
 
   COMPREPLY=()
diff --git a/hashcat.hctune b/hashcat.hctune
index b11441ba2..5b991c35b 100644
--- a/hashcat.hctune
+++ b/hashcat.hctune
@@ -280,6 +280,14 @@ GeForce_RTX_3090                                ALIAS_nv_sm50_or_higher
 
 Device_738c                                     ALIAS_AMD_MI100
 
+AMD_Radeon_(TM)_RX_480_Graphics                 ALIAS_AMD_RX480
+
+Vega_10_XL/XT_[Radeon_RX_Vega_56/64]            ALIAS_AMD_Vega64
+AMD_Radeon_Vega_64                              ALIAS_AMD_Vega64
+
+Device_73bf                                     ALIAS_AMD_RX6900XT
+AMD_Radeon_RX_6900_XT                           ALIAS_AMD_RX6900XT
+
 #############
 ## ENTRIES ##
 #############
@@ -378,146 +386,13 @@ GeForce_GTX_TITAN                               3       9900    2       A
 DEVICE_TYPE_CPU                                 *       3200    1       N       A
 
 ##
-## SCRYPT
+## SCRYPT: Tunings for SCRYPT based hash-modes can be found inside the plugin source
+##         See function module_extra_tuningdb_block()
 ##
 
-DEVICE_TYPE_CPU                                 *       8900    1       N       A
-DEVICE_TYPE_CPU                                 *       9300    1       N       A
-DEVICE_TYPE_CPU                                 *       15700   1       N       A
-DEVICE_TYPE_CPU                                 *       22700   1       N       A
-
-DEVICE_TYPE_GPU                                 *       8900    1       N       A
-DEVICE_TYPE_GPU                                 *       9300    1       N       A
-DEVICE_TYPE_GPU                                 *       15700   1       1       A
-DEVICE_TYPE_GPU                                 *       22700   1       N       A
-
 ##
 ## CryptoAPI
 ##
 
 DEVICE_TYPE_CPU                                 *       14500   1       A       A
 DEVICE_TYPE_GPU                                 *       14500   1       A       A
-
-## Here's an example of how to manually tune SCRYPT algorithm kernels for your hardware.
-## Manually tuning the GPU will yield increased performance. There is typically no noticeable change to CPU performance.
-##
-## First, you need to know the parameters of your SCRYPT hash: N, r and p.
-##
-## The reference SCRYPT parameter values are N=14, r=8 and p=1, but these will likely not match the parameters used by real-world applications.
-## For reference, the N value represents an exponent (2^N, which we calculate by bit shifting 1 left by N bits).
-## Hashcat expects this N value in decimal format: 1 << 14 = 16384
-##
-## Now that you have the 3 configuration items in decimal format, multiply them by 128 (underlaying crypto primitive block size).
-## For example: 128 * 16384 * 8 * 1 = 16777216 = 16MB
-## This is the amount of memory required for the GPU to compute the hash of one password candidate.
-##
-## Hashcat computes multiple password candidates in parallel - this is what allows for full utilization of the device.
-## The number of password candidates that Hashcat can run in parallel is VRAM limited and depends on:
-##
-## 1. Compute devices' native compute units
-## 2. Compute devices' native thread count
-## 3. Artificial multiplier (--kernel-accel aka -n)
-##
-## In order to find these values:
-##
-## 1. On startup Hashcat will show: * Device #1: GeForce GTX 980, 3963/4043 MB, 16MCU. The 16 MCU is the number of compute units on that device.
-## 2. Native thread counts are fixed values: CPU=1, GPU-Intel=8, GPU-AMD=64 (wavefronts), GPU-NVIDIA=32 (warps)
-##
-## Now multiply them together. For my GTX980: 16 * 32 * 16777216 = 8589934592 = 8GB
-##
-## If we want to actually make use of all computing resources, this GPU would require 8GB of GPU RAM.
-## However, it doesn't have that:
-##
-## Device #1: GeForce GTX 980, 3963/4043 MB, 16MCU. We only have 4043 MB (4GB minus some overhead from the OS).
-##
-## How do we deal with this? This is where SCRYPT TMTO(time-memory trde off) kicks in. The SCRYPT algorithm is designed in such a way that we
-## can pre-compute that 16MB buffer from a self-choosen offset. Details on how this actually works are not important for this process.
-##
-## What's relevant to us is that we can halve the buffer size, but we pay with twice the computation time.
-## We can repeat this as often as we want. That's why it's a trade-off.
-##
-## This mechanic can be manually set using --scrypt-tmto on the commandline, but this is not the best way.
-##
-## Back to our problem. We need 8GB of memory but have only ~4GB.
-## It's not a full 4GB. The OS needs some of it and Hashcat needs some of it to store password candidates and other things.
-## If you run a headless server it should be safe to subtract a fixed value of 200MB from whatever you have in your GPU.
-##
-## So lets divide our required memory(8GB) by 2 until it fits in our VRAM - 200MB.
-##
-## (8GB >> 0) = 8GB < 3.8GB = No, Does not fit
-## (8GB >> 1) = 4GB < 3.8GB = No, Does not fit
-## (8GB >> 2) = 2GB < 3.8GB = Yes!
-##
-## This process is automated in Hashcat, but it is important to understand what's happening here.
-## Because of the light overhead from the OS and Hashcat, we pay a very high price.
-## Even though it is just 200MB, it forces us to increase the TMTO by another step.
-## In terms of speed, the speed is now only 1/4 of what we could archieve on that same GPU if it had only 8.2GB ram.
-## But now we end up in a situation that we waste 1.8GB RAM which costs us ((1.8GB/16MB)>>1) candidates/second.
-##
-## This is where manual tuning can come into play.
-## If we know that the resources we need are close to what we have (in this case 3.8GB <-> 4.0GB)
-## We could decide to throw away some of our compute units so that we will no longer need 4.0GB but only 3.8GB.
-## Therefore, we do not need to increase the TMTO by another step to fit in VRAM.
-##
-## If we cut down our 16 MCU to only 15 MCU or 14 MCU using --kernel-accel(-n), we end up with:
-##
-## 16 * 32 * 16777216 = 8589934592 / 2 = 4294967296 = 4.00GB < 3.80GB = Nope, next
-## 15 * 32 * 16777216 = 8053063680 / 2 = 4026531840 = 3.84GB < 3.80GB = Nope, next
-## 14 * 32 * 16777216 = 7516192768 / 2 = 3758096384 = 3.58GB < 3.80GB = Yes!
-##
-## So we can throw away 2/16 compute units, but save half of the computation trade-off on the rest of the compute device.
-## On my GTX980, this improves the performance from 163 H/s to 201 H/s.
-## You don't need to control --scrypt-tmto manually because now that the multiplier (-n) is smaller than the native value
-## Hashcat will automatically realize it can decrease the TMTO by one.
-##
-## At this point, you found the optimal base value for your compute device. In this case: 14.
-##
-## Depending on our hardware, especially hardware with very slow memory access like a GPU
-## there's a good chance that it's cheaper (faster) to compute an extra step on the GPU register.
-## So if we increase the TMTO again by one, this gives an extra speed boost.
-##
-## On my GTX980, this improves the performance from 201 H/s to 255 H/s.
-## Again, there's no need to control this with --scrypt-tmto. Hashcat will realize it has to increase the TMTO again.
-##
-## All together, you can control all of this by using the -n parameter in the command line.
-## This is not ideal in a production environment because you must use the --force flag.
-## The best way to set this is by using this Hashcat.hctune file to store it. This avoids the need to bypass any warnings.
-##
-## Find the ideal -n value, then store it here along with the proper compute device name.
-## Formatting guidelines are availabe at the top of this document.
-
-## 4GB
-GeForce_GTX_980                                 *       8900    1      28       A
-GeForce_GTX_980                                 *       9300    1     128       A
-GeForce_GTX_980                                 *       15700   1      28       A
-GeForce_GTX_980                                 *       22700   1      28       A
-
-## 8GB
-GeForce_GTX_1080                                *       8900    1      14       A
-GeForce_GTX_1080                                *       9300    1     256       A
-GeForce_GTX_1080                                *       15700   1      14       A
-GeForce_GTX_1080                                *       22700   1      14       A
-
-## 11GB
-GeForce_RTX_2080_Ti                             *       8900    1      68       A
-GeForce_RTX_2080_Ti                             *       9300    1     532       A
-GeForce_RTX_2080_Ti                             *       15700   1      68       A
-GeForce_RTX_2080_Ti                             *       22700   1      68       A
-
-## 4GB
-AMD_Radeon_(TM)_RX_480_Graphics                 *       8900    1      14       A
-AMD_Radeon_(TM)_RX_480_Graphics                 *       9300    1     126       A
-AMD_Radeon_(TM)_RX_480_Graphics                 *       15700   1      14       A
-AMD_Radeon_(TM)_RX_480_Graphics                 *       22700   1      14       A
-
-## 8GB
-Vega_10_XL/XT_[Radeon_RX_Vega_56/64]            *       8900    1      28       A
-Vega_10_XL/XT_[Radeon_RX_Vega_56/64]            *       9300    1     442       A
-Vega_10_XL/XT_[Radeon_RX_Vega_56/64]            *       15700   1      28       A
-Vega_10_XL/XT_[Radeon_RX_Vega_56/64]            *       22700   1      28       A
-
-## 32GB
-ALIAS_AMD_MI100                                 *       8900    1      76       A
-ALIAS_AMD_MI100                                 *       9300    1      63       A
-ALIAS_AMD_MI100                                 *       15700   1      76       A
-ALIAS_AMD_MI100                                 *       22700   1      76       A
diff --git a/include/backend.h b/include/backend.h
index a3d997ca8..de74cb511 100644
--- a/include/backend.h
+++ b/include/backend.h
@@ -9,18 +9,19 @@
 #include <stdio.h>
 #include <errno.h>
 
-static const char CL_VENDOR_AMD1[]            = "Advanced Micro Devices, Inc.";
-static const char CL_VENDOR_AMD2[]            = "AuthenticAMD";
-static const char CL_VENDOR_AMD_USE_INTEL[]   = "GenuineIntel";
-static const char CL_VENDOR_APPLE[]           = "Apple";
-static const char CL_VENDOR_APPLE_USE_AMD[]   = "AMD";
-static const char CL_VENDOR_APPLE_USE_NV[]    = "NVIDIA";
-static const char CL_VENDOR_APPLE_USE_INTEL[] = "Intel Inc.";
-static const char CL_VENDOR_INTEL_BEIGNET[]   = "Intel";
-static const char CL_VENDOR_INTEL_SDK[]       = "Intel(R) Corporation";
-static const char CL_VENDOR_MESA[]            = "Mesa";
-static const char CL_VENDOR_NV[]              = "NVIDIA Corporation";
-static const char CL_VENDOR_POCL[]            = "The pocl project";
+static const char CL_VENDOR_AMD1[]              = "Advanced Micro Devices, Inc.";
+static const char CL_VENDOR_AMD2[]              = "AuthenticAMD";
+static const char CL_VENDOR_AMD_USE_INTEL[]     = "GenuineIntel";
+static const char CL_VENDOR_APPLE[]             = "Apple";
+static const char CL_VENDOR_APPLE_USE_AMD[]     = "AMD";
+static const char CL_VENDOR_APPLE_USE_NV[]      = "NVIDIA";
+static const char CL_VENDOR_APPLE_USE_INTEL[]   = "Intel";
+static const char CL_VENDOR_APPLE_USE_INTEL2[]  = "Intel Inc.";
+static const char CL_VENDOR_INTEL_BEIGNET[]     = "Intel";
+static const char CL_VENDOR_INTEL_SDK[]         = "Intel(R) Corporation";
+static const char CL_VENDOR_MESA[]              = "Mesa";
+static const char CL_VENDOR_NV[]                = "NVIDIA Corporation";
+static const char CL_VENDOR_POCL[]              = "The pocl project";
 
 int  cuda_init    (hashcat_ctx_t *hashcat_ctx);
 void cuda_close   (hashcat_ctx_t *hashcat_ctx);
@@ -68,10 +69,12 @@ int hc_cuFuncSetAttribute        (hashcat_ctx_t *hashcat_ctx, CUfunction hfunc,
 int hc_cuInit                    (hashcat_ctx_t *hashcat_ctx, unsigned int Flags);
 int hc_cuLaunchKernel            (hashcat_ctx_t *hashcat_ctx, CUfunction f, unsigned int gridDimX, unsigned int gridDimY, unsigned int gridDimZ, unsigned int blockDimX, unsigned int blockDimY, unsigned int blockDimZ, unsigned int sharedMemBytes, CUstream hStream, void **kernelParams, void **extra);
 int hc_cuMemAlloc                (hashcat_ctx_t *hashcat_ctx, CUdeviceptr *dptr, size_t bytesize);
-int hc_cuMemcpyDtoD              (hashcat_ctx_t *hashcat_ctx, CUdeviceptr dstDevice, CUdeviceptr srcDevice, size_t ByteCount);
-int hc_cuMemcpyDtoH              (hashcat_ctx_t *hashcat_ctx, void *dstHost, CUdeviceptr srcDevice, size_t ByteCount);
-int hc_cuMemcpyHtoD              (hashcat_ctx_t *hashcat_ctx, CUdeviceptr dstDevice, const void *srcHost, size_t ByteCount);
+int hc_cuMemcpyDtoDAsync         (hashcat_ctx_t *hashcat_ctx, CUdeviceptr dstDevice, CUdeviceptr srcDevice, size_t ByteCount, CUstream hStream);
+int hc_cuMemcpyDtoHAsync         (hashcat_ctx_t *hashcat_ctx, void *dstHost, CUdeviceptr srcDevice, size_t ByteCount, CUstream hStream);
+int hc_cuMemcpyHtoDAsync         (hashcat_ctx_t *hashcat_ctx, CUdeviceptr dstDevice, const void *srcHost, size_t ByteCount, CUstream hStream);
 int hc_cuMemFree                 (hashcat_ctx_t *hashcat_ctx, CUdeviceptr dptr);
+int hc_cuMemsetD32Async          (hashcat_ctx_t *hashcat_ctx, CUdeviceptr dstDevice, unsigned int ui, size_t N, CUstream hStream);
+int hc_cuMemsetD8Async           (hashcat_ctx_t *hashcat_ctx, CUdeviceptr dstDevice, unsigned char uc, size_t N, CUstream hStream);
 int hc_cuModuleGetFunction       (hashcat_ctx_t *hashcat_ctx, CUfunction *hfunc, CUmodule hmod, const char *name);
 int hc_cuModuleLoadDataEx        (hashcat_ctx_t *hashcat_ctx, CUmodule *module, const void *image, unsigned int numOptions, CUjit_option *options, void **optionValues);
 int hc_cuModuleUnload            (hashcat_ctx_t *hashcat_ctx, CUmodule hmod);
@@ -85,53 +88,50 @@ int hc_cuLinkAddData             (hashcat_ctx_t *hashcat_ctx, CUlinkState state,
 int hc_cuLinkDestroy             (hashcat_ctx_t *hashcat_ctx, CUlinkState state);
 int hc_cuLinkComplete            (hashcat_ctx_t *hashcat_ctx, CUlinkState state, void **cubinOut, size_t *sizeOut);
 
-int hc_nvrtcCreateProgram        (hashcat_ctx_t *hashcat_ctx, nvrtcProgram *prog, const char *src, const char *name, int numHeaders, const char * const *headers, const char * const *includeNames);
-int hc_nvrtcDestroyProgram       (hashcat_ctx_t *hashcat_ctx, nvrtcProgram *prog);
-int hc_nvrtcCompileProgram       (hashcat_ctx_t *hashcat_ctx, nvrtcProgram prog, int numOptions, const char * const *options);
-int hc_nvrtcGetProgramLogSize    (hashcat_ctx_t *hashcat_ctx, nvrtcProgram prog, size_t *logSizeRet);
-int hc_nvrtcGetProgramLog        (hashcat_ctx_t *hashcat_ctx, nvrtcProgram prog, char *log);
-int hc_nvrtcGetPTXSize           (hashcat_ctx_t *hashcat_ctx, nvrtcProgram prog, size_t *ptxSizeRet);
-int hc_nvrtcGetPTX               (hashcat_ctx_t *hashcat_ctx, nvrtcProgram prog, char *ptx);
-int hc_nvrtcVersion              (hashcat_ctx_t *hashcat_ctx, int *major, int *minor);
+int hc_hipCreateProgram          (hashcat_ctx_t *hashcat_ctx, hiprtcProgram *prog, const char *src, const char *name, int numHeaders, const char * const *headers, const char * const *includeNames);
+int hc_hipDestroyProgram         (hashcat_ctx_t *hashcat_ctx, hiprtcProgram *prog);
+int hc_hipCompileProgram         (hashcat_ctx_t *hashcat_ctx, hiprtcProgram prog, int numOptions, const char * const *options);
+int hc_hipGetProgramLogSize      (hashcat_ctx_t *hashcat_ctx, hiprtcProgram prog, size_t *logSizeRet);
+int hc_hipGetProgramLog          (hashcat_ctx_t *hashcat_ctx, hiprtcProgram prog, char *log);
+int hc_hipGetCodeSize            (hashcat_ctx_t *hashcat_ctx, hiprtcProgram prog, size_t *codeSizeRet);
+int hc_hipGetCode                (hashcat_ctx_t *hashcat_ctx, hiprtcProgram prog, char *code);
 
-int hc_hipCtxCreate              (hashcat_ctx_t *hashcat_ctx, HIPcontext *pctx, unsigned int flags, HIPdevice dev);
-int hc_hipCtxDestroy             (hashcat_ctx_t *hashcat_ctx, HIPcontext ctx);
-int hc_hipCtxSetCurrent          (hashcat_ctx_t *hashcat_ctx, HIPcontext ctx);
-int hc_hipCtxSetCacheConfig      (hashcat_ctx_t *hashcat_ctx, HIPfunc_cache config);
+int hc_hipCtxCreate              (hashcat_ctx_t *hashcat_ctx, hipCtx_t *pctx, unsigned int flags, hipDevice_t dev);
+int hc_hipCtxDestroy             (hashcat_ctx_t *hashcat_ctx, hipCtx_t ctx);
+int hc_hipCtxPopCurrent          (hashcat_ctx_t *hashcat_ctx, hipCtx_t *pctx);
+int hc_hipCtxPushCurrent         (hashcat_ctx_t *hashcat_ctx, hipCtx_t ctx);
+int hc_hipCtxSetCurrent          (hashcat_ctx_t *hashcat_ctx, hipCtx_t ctx);
 int hc_hipCtxSynchronize         (hashcat_ctx_t *hashcat_ctx);
-int hc_hipDeviceGetAttribute     (hashcat_ctx_t *hashcat_ctx, int *pi, HIPdevice_attribute attrib, HIPdevice dev);
+int hc_hipDeviceGet              (hashcat_ctx_t *hashcat_ctx, hipDevice_t *device, int ordinal);
+int hc_hipDeviceGetAttribute     (hashcat_ctx_t *hashcat_ctx, int *pi, hipDeviceAttribute_t attrib, hipDevice_t dev);
 int hc_hipDeviceGetCount         (hashcat_ctx_t *hashcat_ctx, int *count);
-int hc_hipDeviceGet              (hashcat_ctx_t *hashcat_ctx, HIPdevice *device, int ordinal);
-int hc_hipDeviceGetName          (hashcat_ctx_t *hashcat_ctx, char *name, int len, HIPdevice dev);
-int hc_hipDeviceTotalMem         (hashcat_ctx_t *hashcat_ctx, size_t *bytes, HIPdevice dev);
+int hc_hipDeviceGetName          (hashcat_ctx_t *hashcat_ctx, char *name, int len, hipDevice_t dev);
+int hc_hipDeviceTotalMem         (hashcat_ctx_t *hashcat_ctx, size_t *bytes, hipDevice_t dev);
 int hc_hipDriverGetVersion       (hashcat_ctx_t *hashcat_ctx, int *driverVersion);
-int hc_hipEventCreate            (hashcat_ctx_t *hashcat_ctx, HIPevent *phEvent, unsigned int Flags);
-int hc_hipEventDestroy           (hashcat_ctx_t *hashcat_ctx, HIPevent hEvent);
-int hc_hipEventElapsedTime       (hashcat_ctx_t *hashcat_ctx, float *pMilliseconds, HIPevent hStart, HIPevent hEnd);
-int hc_hipEventQuery             (hashcat_ctx_t *hashcat_ctx, HIPevent hEvent);
-int hc_hipEventRecord            (hashcat_ctx_t *hashcat_ctx, HIPevent hEvent, HIPstream hStream);
-int hc_hipEventSynchronize       (hashcat_ctx_t *hashcat_ctx, HIPevent hEvent);
-int hc_hipFuncGetAttribute       (hashcat_ctx_t *hashcat_ctx, int *pi, HIPfunction_attribute attrib, HIPfunction hfunc);
-//int hc_hipFuncSetAttribute       (hashcat_ctx_t *hashcat_ctx, HIPfunction hfunc, HIPfunction_attribute attrib, int value);
+int hc_hipEventCreate            (hashcat_ctx_t *hashcat_ctx, hipEvent_t *phEvent, unsigned int Flags);
+int hc_hipEventDestroy           (hashcat_ctx_t *hashcat_ctx, hipEvent_t hEvent);
+int hc_hipEventElapsedTime       (hashcat_ctx_t *hashcat_ctx, float *pMilliseconds, hipEvent_t hStart, hipEvent_t hEnd);
+int hc_hipEventQuery             (hashcat_ctx_t *hashcat_ctx, hipEvent_t hEvent);
+int hc_hipEventRecord            (hashcat_ctx_t *hashcat_ctx, hipEvent_t hEvent, hipStream_t hStream);
+int hc_hipEventSynchronize       (hashcat_ctx_t *hashcat_ctx, hipEvent_t hEvent);
+int hc_hipFuncGetAttribute       (hashcat_ctx_t *hashcat_ctx, int *pi, hipFunction_attribute attrib, hipFunction_t hfunc);
 int hc_hipInit                   (hashcat_ctx_t *hashcat_ctx, unsigned int Flags);
-int hc_hipLaunchKernel           (hashcat_ctx_t *hashcat_ctx, HIPfunction f, unsigned int gridDimX, unsigned int gridDimY, unsigned int gridDimZ, unsigned int blockDimX, unsigned int blockDimY, unsigned int blockDimZ, unsigned int sharedMemBytes, HIPstream hStream, void **kernelParams, void **extra);
-int hc_hipMemAlloc               (hashcat_ctx_t *hashcat_ctx, HIPdeviceptr *dptr, size_t bytesize);
-int hc_hipMemcpyDtoD             (hashcat_ctx_t *hashcat_ctx, HIPdeviceptr dstDevice, HIPdeviceptr srcDevice, size_t ByteCount);
-int hc_hipMemcpyDtoH             (hashcat_ctx_t *hashcat_ctx, void *dstHost, HIPdeviceptr srcDevice, size_t ByteCount);
-int hc_hipMemcpyHtoD             (hashcat_ctx_t *hashcat_ctx, HIPdeviceptr dstDevice, const void *srcHost, size_t ByteCount);
-int hc_hipMemFree                (hashcat_ctx_t *hashcat_ctx, HIPdeviceptr dptr);
-int hc_hipModuleGetFunction      (hashcat_ctx_t *hashcat_ctx, HIPfunction *hfunc, HIPmodule hmod, const char *name);
-int hc_hipModuleLoadDataEx       (hashcat_ctx_t *hashcat_ctx, HIPmodule *module, const void *image, unsigned int numOptions, HIPjit_option *options, void **optionValues);
-int hc_hipModuleUnload           (hashcat_ctx_t *hashcat_ctx, HIPmodule hmod);
-int hc_hipStreamCreate           (hashcat_ctx_t *hashcat_ctx, HIPstream *phStream, unsigned int Flags);
-int hc_hipStreamDestroy          (hashcat_ctx_t *hashcat_ctx, HIPstream hStream);
-int hc_hipStreamSynchronize      (hashcat_ctx_t *hashcat_ctx, HIPstream hStream);
-int hc_hipCtxPushCurrent         (hashcat_ctx_t *hashcat_ctx, HIPcontext ctx);
-int hc_hipCtxPopCurrent          (hashcat_ctx_t *hashcat_ctx, HIPcontext *pctx);
-int hc_hipLinkCreate             (hashcat_ctx_t *hashcat_ctx, unsigned int numOptions, HIPjit_option *options, void **optionValues, HIPlinkState *stateOut);
-int hc_hipLinkAddData            (hashcat_ctx_t *hashcat_ctx, HIPlinkState state, HIPjitInputType type, void *data, size_t size, const char *name, unsigned int numOptions, HIPjit_option *options, void **optionValues);
-int hc_hipLinkDestroy            (hashcat_ctx_t *hashcat_ctx, HIPlinkState state);
-int hc_hipLinkComplete           (hashcat_ctx_t *hashcat_ctx, HIPlinkState state, void **cubinOut, size_t *sizeOut);
+int hc_hipLaunchKernel           (hashcat_ctx_t *hashcat_ctx, hipFunction_t f, unsigned int gridDimX, unsigned int gridDimY, unsigned int gridDimZ, unsigned int blockDimX, unsigned int blockDimY, unsigned int blockDimZ, unsigned int sharedMemBytes, hipStream_t hStream, void **kernelParams, void **extra);
+int hc_hipMemAlloc               (hashcat_ctx_t *hashcat_ctx, hipDeviceptr_t *dptr, size_t bytesize);
+int hc_hipMemFree                (hashcat_ctx_t *hashcat_ctx, hipDeviceptr_t dptr);
+int hc_hipMemcpyDtoDAsync        (hashcat_ctx_t *hashcat_ctx, hipDeviceptr_t dstDevice, hipDeviceptr_t srcDevice, size_t ByteCount, hipStream_t hStream);
+int hc_hipMemcpyDtoHAsync        (hashcat_ctx_t *hashcat_ctx, void *dstHost, hipDeviceptr_t srcDevice, size_t ByteCount, hipStream_t hStream);
+int hc_hipMemcpyHtoDAsync        (hashcat_ctx_t *hashcat_ctx, hipDeviceptr_t dstDevice, const void *srcHost, size_t ByteCount, hipStream_t hStream);
+int hc_hipMemsetD32Async         (hashcat_ctx_t *hashcat_ctx, hipDeviceptr_t dstDevice, unsigned int ui, size_t N, hipStream_t hStream);
+int hc_hipMemsetD8Async          (hashcat_ctx_t *hashcat_ctx, hipDeviceptr_t dstDevice, unsigned char uc, size_t N, hipStream_t hStream);
+int hc_hipModuleGetFunction      (hashcat_ctx_t *hashcat_ctx, hipFunction_t *hfunc, hipModule_t hmod, const char *name);
+int hc_hipModuleGetGlobal        (hashcat_ctx_t *hashcat_ctx, hipDeviceptr_t *dptr, size_t *bytes, hipModule_t hmod, const char *name);
+int hc_hipModuleLoadDataEx       (hashcat_ctx_t *hashcat_ctx, hipModule_t *module, const void *image, unsigned int numOptions, hipJitOption *options, void **optionValues);
+int hc_hipModuleUnload           (hashcat_ctx_t *hashcat_ctx, hipModule_t hmod);
+int hc_hipRuntimeGetVersion      (hashcat_ctx_t *hashcat_ctx, int *runtimeVersion);
+int hc_hipStreamCreate           (hashcat_ctx_t *hashcat_ctx, hipStream_t *phStream, unsigned int Flags);
+int hc_hipStreamDestroy          (hashcat_ctx_t *hashcat_ctx, hipStream_t hStream);
+int hc_hipStreamSynchronize      (hashcat_ctx_t *hashcat_ctx, hipStream_t hStream);
 
 int hc_clBuildProgram            (hashcat_ctx_t *hashcat_ctx, cl_program program, cl_uint num_devices, const cl_device_id *device_list, const char *options, void (CL_CALLBACK *pfn_notify) (cl_program program, void *user_data), void *user_data);
 int hc_clCompileProgram          (hashcat_ctx_t *hashcat_ctx, cl_program program, cl_uint num_devices, const cl_device_id *device_list, const char *options, cl_uint num_input_headers, const cl_program *input_headers, const char **header_include_names, void (CL_CALLBACK *pfn_notify) (cl_program program, void *user_data), void *user_data);
@@ -142,6 +142,7 @@ int hc_clCreateKernel            (hashcat_ctx_t *hashcat_ctx, cl_program program
 int hc_clCreateProgramWithBinary (hashcat_ctx_t *hashcat_ctx, cl_context context, cl_uint num_devices, const cl_device_id *device_list, const size_t *lengths, const unsigned char **binaries, cl_int *binary_status, cl_program *program);
 int hc_clCreateProgramWithSource (hashcat_ctx_t *hashcat_ctx, cl_context context, cl_uint count, const char **strings, const size_t *lengths, cl_program *program);
 int hc_clEnqueueCopyBuffer       (hashcat_ctx_t *hashcat_ctx, cl_command_queue command_queue, cl_mem src_buffer, cl_mem dst_buffer, size_t src_offset, size_t dst_offset, size_t size, cl_uint num_events_in_wait_list, const cl_event *event_wait_list, cl_event *event);
+int hc_clEnqueueFillBuffer       (hashcat_ctx_t *hashcat_ctx, cl_command_queue command_queue, cl_mem buffer, const void *pattern, size_t pattern_size, size_t offset, size_t size, cl_uint num_events_in_wait_list, const cl_event *event_wait_list, cl_event *event);
 int hc_clEnqueueMapBuffer        (hashcat_ctx_t *hashcat_ctx, cl_command_queue command_queue, cl_mem buffer, cl_bool blocking_map, cl_map_flags map_flags, size_t offset, size_t size, cl_uint num_events_in_wait_list, const cl_event *event_wait_list, cl_event *event, void **buf);
 int hc_clEnqueueNDRangeKernel    (hashcat_ctx_t *hashcat_ctx, cl_command_queue command_queue, cl_kernel kernel, cl_uint work_dim, const size_t *global_work_offset, const size_t *global_work_size, const size_t *local_work_size, cl_uint num_events_in_wait_list, const cl_event *event_wait_list, cl_event *event);
 int hc_clEnqueueReadBuffer       (hashcat_ctx_t *hashcat_ctx, cl_command_queue command_queue, cl_mem buffer, cl_bool blocking_read, size_t offset, size_t size, void *ptr, cl_uint num_events_in_wait_list, const cl_event *event_wait_list, cl_event *event);
@@ -177,17 +178,20 @@ void rebuild_pws_compressed_append (hc_device_param_t *device_param, const u64 p
 
 int run_cuda_kernel_atinit          (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, CUdeviceptr buf, const u64 num);
 int run_cuda_kernel_utf8toutf16le   (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, CUdeviceptr buf, const u64 num);
-int run_cuda_kernel_memset          (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, CUdeviceptr buf, const u32 value, const u64 size);
+int run_cuda_kernel_memset          (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, CUdeviceptr buf, const u64 offset, const u8  value, const u64 size);
+int run_cuda_kernel_memset32        (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, CUdeviceptr buf, const u64 offset, const u32 value, const u64 size);
 int run_cuda_kernel_bzero           (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, CUdeviceptr buf, const u64 size);
 
-int run_hip_kernel_atinit           (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, HIPdeviceptr buf, const u64 num);
-int run_hip_kernel_utf8toutf16le    (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, HIPdeviceptr buf, const u64 num);
-int run_hip_kernel_memset           (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, HIPdeviceptr buf, const u32 value, const u64 size);
-int run_hip_kernel_bzero            (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, HIPdeviceptr buf, const u64 size);
+int run_hip_kernel_atinit           (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, hipDeviceptr_t buf, const u64 num);
+int run_hip_kernel_utf8toutf16le    (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, hipDeviceptr_t buf, const u64 num);
+int run_hip_kernel_memset           (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, hipDeviceptr_t buf, const u64 offset, const u8  value, const u64 size);
+int run_hip_kernel_memset32         (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, hipDeviceptr_t buf, const u64 offset, const u32 value, const u64 size);
+int run_hip_kernel_bzero            (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, hipDeviceptr_t buf, const u64 size);
 
 int run_opencl_kernel_atinit        (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, cl_mem buf, const u64 num);
 int run_opencl_kernel_utf8toutf16le (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, cl_mem buf, const u64 num);
-int run_opencl_kernel_memset        (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, cl_mem buf, const u32 value, const u64 size);
+int run_opencl_kernel_memset        (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, cl_mem buf, const u64 offset, const u8  value, const u64 size);
+int run_opencl_kernel_memset32      (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, cl_mem buf, const u64 offset, const u32 value, const u64 size);
 int run_opencl_kernel_bzero         (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, cl_mem buf, const u64 size);
 
 int run_kernel                (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, const u32 kern_run, const u64 pws_pos, const u64 num, const u32 event_update, const u32 iteration);
diff --git a/include/common.h b/include/common.h
index 445ebc405..3da61bc6c 100644
--- a/include/common.h
+++ b/include/common.h
@@ -63,6 +63,14 @@
 #define HC_API_CALL
 #endif
 
+#if defined (__GNUC__)
+#define HC_ALIGN(x) __attribute__((aligned(x)))
+#elif defined (_MSC_VER)
+#define HC_ALIGN(x) __declspec(align(x))
+#else
+#define HC_ALIGN(x)
+#endif
+
 #if defined (_WIN)
 #define WIN32_LEAN_AND_MEAN
 #endif
diff --git a/include/ext_ADL.h b/include/ext_ADL.h
index fd8438c8e..369a8eb1a 100644
--- a/include/ext_ADL.h
+++ b/include/ext_ADL.h
@@ -13,228 +13,569 @@
 #include <windows.h>
 #endif // _WIN
 
+// Declarations from:
+// https://github.com/GPUOpen-LibrariesAndSDKs/display-library/blob/209538e1dc7273f7459411a3a5044ffe2437ed95/include/adl_defines.h
+// https://github.com/GPUOpen-LibrariesAndSDKs/display-library/blob/209538e1dc7273f7459411a3a5044ffe2437ed95/include/adl_structures.h
+
+
+/// Defines ADL_TRUE
+#define ADL_TRUE    1
+/// Defines ADL_FALSE
+#define ADL_FALSE        0
+
+//Define Performance Metrics Log max sensors number
+#define ADL_PMLOG_MAX_SENSORS  256
+
+typedef enum ADLSensorType
+{
+	SENSOR_MAXTYPES = 0,
+	PMLOG_CLK_GFXCLK = 1,
+	PMLOG_CLK_MEMCLK = 2,
+	PMLOG_CLK_SOCCLK = 3,
+	PMLOG_CLK_UVDCLK1 = 4,
+	PMLOG_CLK_UVDCLK2 = 5,
+	PMLOG_CLK_VCECLK = 6,
+	PMLOG_CLK_VCNCLK = 7,
+	PMLOG_TEMPERATURE_EDGE = 8,
+	PMLOG_TEMPERATURE_MEM = 9,
+	PMLOG_TEMPERATURE_VRVDDC = 10,
+	PMLOG_TEMPERATURE_VRMVDD = 11,
+	PMLOG_TEMPERATURE_LIQUID = 12,
+	PMLOG_TEMPERATURE_PLX = 13,
+	PMLOG_FAN_RPM = 14,
+	PMLOG_FAN_PERCENTAGE = 15,
+	PMLOG_SOC_VOLTAGE = 16,
+	PMLOG_SOC_POWER = 17,
+	PMLOG_SOC_CURRENT = 18,
+	PMLOG_INFO_ACTIVITY_GFX = 19,
+	PMLOG_INFO_ACTIVITY_MEM = 20,
+	PMLOG_GFX_VOLTAGE = 21,
+	PMLOG_MEM_VOLTAGE = 22,
+	PMLOG_ASIC_POWER = 23,
+	PMLOG_TEMPERATURE_VRSOC = 24,
+	PMLOG_TEMPERATURE_VRMVDD0 = 25,
+	PMLOG_TEMPERATURE_VRMVDD1 = 26,
+	PMLOG_TEMPERATURE_HOTSPOT = 27,
+        PMLOG_TEMPERATURE_GFX = 28,
+        PMLOG_TEMPERATURE_SOC = 29,
+        PMLOG_GFX_POWER = 30,
+        PMLOG_GFX_CURRENT = 31,
+        PMLOG_TEMPERATURE_CPU = 32,
+        PMLOG_CPU_POWER = 33,
+        PMLOG_CLK_CPUCLK = 34,
+        PMLOG_THROTTLER_STATUS = 35,
+        PMLOG_CLK_VCN1CLK1 = 36,
+        PMLOG_CLK_VCN1CLK2 = 37,
+        PMLOG_SMART_POWERSHIFT_CPU = 38,
+        PMLOG_SMART_POWERSHIFT_DGPU = 39,
+        PMLOG_BUS_SPEED = 40,
+        PMLOG_BUS_LANES = 41,
+	PMLOG_MAX_SENSORS_REAL
+} ADLSensorType;
+
+/// Defines the maximum string length
+#define ADL_MAX_CHAR                                    4096
+/// Defines the maximum string length
+#define ADL_MAX_PATH                                    256
+/// Defines the maximum number of supported adapters
+#define ADL_MAX_ADAPTERS                               250
+/// Defines the maxumum number of supported displays
+#define ADL_MAX_DISPLAYS                                150
+/// Defines the maxumum string length for device name
+#define ADL_MAX_DEVICENAME                                32
+/// Defines for all adapters
+#define ADL_ADAPTER_INDEX_ALL                            -1
+
+/// \defgroup define_adl_results Result Codes
+/// This group of definitions are the various results returned by all ADL functions \n
+/// @{
+/// All OK, but need to wait
+#define ADL_OK_WAIT                4
+/// All OK, but need restart
+#define ADL_OK_RESTART                3
+/// All OK but need mode change
+#define ADL_OK_MODE_CHANGE            2
+/// All OK, but with warning
+#define ADL_OK_WARNING                1
+/// ADL function completed successfully
+#define ADL_OK                    0
+/// Generic Error. Most likely one or more of the Escape calls to the driver failed!
+#define ADL_ERR                    -1
+/// ADL not initialized
+#define ADL_ERR_NOT_INIT            -2
+/// One of the parameter passed is invalid
+#define ADL_ERR_INVALID_PARAM            -3
+/// One of the parameter size is invalid
+#define ADL_ERR_INVALID_PARAM_SIZE        -4
+/// Invalid ADL index passed
+#define ADL_ERR_INVALID_ADL_IDX            -5
+/// Invalid controller index passed
+#define ADL_ERR_INVALID_CONTROLLER_IDX        -6
+/// Invalid display index passed
+#define ADL_ERR_INVALID_DIPLAY_IDX        -7
+/// Function  not supported by the driver
+#define ADL_ERR_NOT_SUPPORTED            -8
+/// Null Pointer error
+#define ADL_ERR_NULL_POINTER            -9
+/// Call can't be made due to disabled adapter
+#define ADL_ERR_DISABLED_ADAPTER        -10
+/// Invalid Callback
+#define ADL_ERR_INVALID_CALLBACK            -11
+/// Display Resource conflict
+#define ADL_ERR_RESOURCE_CONFLICT                -12
+//Failed to update some of the values. Can be returned by set request that include multiple values if not all values were successfully committed.
+#define ADL_ERR_SET_INCOMPLETE                 -20
+/// There's no Linux XDisplay in Linux Console environment
+#define ADL_ERR_NO_XDISPLAY                    -21
+
+//values for ADLFanSpeedValue.iSpeedType
+#define ADL_DL_FANCTRL_SPEED_TYPE_PERCENT    1
+#define ADL_DL_FANCTRL_SPEED_TYPE_RPM        2
+
+//values for ADLFanSpeedValue.iFlags
+#define ADL_DL_FANCTRL_FLAG_USER_DEFINED_SPEED   1
+
 /**
- * Declarations from adl_sdk.h and subheaders
+ * Declarations from adl_structures.h
  */
 
-#define ADL_OK                                  0
-#define ADL_ERR                                -1
-#define ADL_ERR_NOT_SUPPORTED                  -8
-
-#define ADL_MAX_PATH                            256
-
-#define ADL_DL_FANCTRL_SPEED_TYPE_PERCENT       1
-#define ADL_DL_FANCTRL_FLAG_USER_DEFINED_SPEED  1
-
+/////////////////////////////////////////////////////////////////////////////////////////////
+///\brief Structure containing information about the graphics adapter.
+///
+/// This structure is used to store various information about the graphics adapter.  This
+/// information can be returned to the user. Alternatively, it can be used to access various driver calls to set
+/// or fetch various settings upon the user's request.
+/// \nosubgrouping
+////////////////////////////////////////////////////////////////////////////////////////////
 typedef struct AdapterInfo
 {
-  int  iSize;
-  int  iAdapterIndex;
-  char strUDID[ADL_MAX_PATH];
-  int  iBusNumber;
-  int  iDeviceNumber;
-  int  iFunctionNumber;
-  int  iVendorID;
-  char strAdapterName[ADL_MAX_PATH];
-  char strDisplayName[ADL_MAX_PATH];
-  int  iPresent;
+/// \ALL_STRUCT_MEM
 
-  #if defined (_WIN32) || defined (_WIN64) || defined (__CYGWIN__)
-  int  iExist;
-  char strDriverPath[ADL_MAX_PATH];
-  char strDriverPathExt[ADL_MAX_PATH];
-  char strPNPString[ADL_MAX_PATH];
-  int  iOSDisplayIndex;
-  #endif /* (_WIN32) || (_WIN64) || (__CYGWIN__) */
+/// Size of the structure.
+    int iSize;
+/// The ADL index handle. One GPU may be associated with one or two index handles
+    int iAdapterIndex;
+/// The unique device ID associated with this adapter.
+    char strUDID[ADL_MAX_PATH];
+/// The BUS number associated with this adapter.
+    int iBusNumber;
+/// The driver number associated with this adapter.
+    int iDeviceNumber;
+/// The function number.
+    int iFunctionNumber;
+/// The vendor ID associated with this adapter.
+    int iVendorID;
+/// Adapter name.
+    char strAdapterName[ADL_MAX_PATH];
+/// Display name. For example, "\\\\Display0" for Windows or ":0:0" for Linux.
+    char strDisplayName[ADL_MAX_PATH];
+/// Present or not; 1 if present and 0 if not present.It the logical adapter is present, the display name such as \\\\.\\Display1 can be found from OS
+    int iPresent;
 
-  #if defined (__linux__)
-  int  iXScreenNum;
-  int  iDrvIndex;
-  char strXScreenConfigName[ADL_MAX_PATH];
-  #endif /* (__linux__) */
+#if defined (_WIN32) || defined (_WIN64)
+/// \WIN_STRUCT_MEM
+
+/// Exist or not; 1 is exist and 0 is not present.
+    int iExist;
+/// Driver registry path.
+    char strDriverPath[ADL_MAX_PATH];
+/// Driver registry path Ext for.
+    char strDriverPathExt[ADL_MAX_PATH];
+/// PNP string from Windows.
+    char strPNPString[ADL_MAX_PATH];
+/// It is generated from EnumDisplayDevices.
+    int iOSDisplayIndex;
+
+#endif /* (_WIN32) || (_WIN64) */
+
+#if defined (LINUX)
+/// \LNX_STRUCT_MEM
+
+/// Internal X screen number from GPUMapInfo (DEPRICATED use XScreenInfo)
+    int iXScreenNum;
+/// Internal driver index from GPUMapInfo
+    int iDrvIndex;
+/// \deprecated Internal x config file screen identifier name. Use XScreenInfo instead.
+    char strXScreenConfigName[ADL_MAX_PATH];
+
+#endif /* (LINUX) */
 } AdapterInfo, *LPAdapterInfo;
 
+/////////////////////////////////////////////////////////////////////////////////////////////
+///\brief Structure containing information about thermal controller.
+///
+/// This structure is used to store information about thermal controller.
+/// This structure is used by ADL_PM_ThermalDevices_Enum.
+/// \nosubgrouping
+////////////////////////////////////////////////////////////////////////////////////////////
 typedef struct ADLThermalControllerInfo
 {
+/// Must be set to the size of the structure
   int iSize;
+/// Possible valies: \ref ADL_DL_THERMAL_DOMAIN_OTHER or \ref ADL_DL_THERMAL_DOMAIN_GPU.
   int iThermalDomain;
+///    GPU 0, 1, etc.
   int iDomainIndex;
+/// Possible valies: \ref ADL_DL_THERMAL_FLAG_INTERRUPT or \ref ADL_DL_THERMAL_FLAG_FANCONTROL
   int iFlags;
 } ADLThermalControllerInfo;
 
+/////////////////////////////////////////////////////////////////////////////////////////////
+///\brief Structure containing information about thermal controller temperature.
+///
+/// This structure is used to store information about thermal controller temperature.
+/// This structure is used by the ADL_PM_Temperature_Get() function.
+/// \nosubgrouping
+////////////////////////////////////////////////////////////////////////////////////////////
 typedef struct ADLTemperature
 {
+/// Must be set to the size of the structure
   int iSize;
+/// Temperature in millidegrees Celsius.
   int iTemperature;
 } ADLTemperature;
 
+/////////////////////////////////////////////////////////////////////////////////////////////
+///\brief Structure containing information about thermal controller fan speed.
+///
+/// This structure is used to store information about thermal controller fan speed.
+/// This structure is used by the ADL_PM_FanSpeedInfo_Get() function.
+/// \nosubgrouping
+////////////////////////////////////////////////////////////////////////////////////////////
 typedef struct ADLFanSpeedInfo
 {
+/// Must be set to the size of the structure
   int iSize;
+/// \ref define_fanctrl
   int iFlags;
+/// Minimum possible fan speed value in percents.
   int iMinPercent;
+/// Maximum possible fan speed value in percents.
   int iMaxPercent;
+/// Minimum possible fan speed value in RPM.
   int iMinRPM;
+/// Maximum possible fan speed value in RPM.
   int iMaxRPM;
 } ADLFanSpeedInfo;
 
+/////////////////////////////////////////////////////////////////////////////////////////////
+///\brief Structure containing information about fan speed reported by thermal controller.
+///
+/// This structure is used to store information about fan speed reported by thermal controller.
+/// This structure is used by the ADL_Overdrive5_FanSpeed_Get() and ADL_Overdrive5_FanSpeed_Set() functions.
+/// \nosubgrouping
+////////////////////////////////////////////////////////////////////////////////////////////
 typedef struct ADLFanSpeedValue
 {
+/// Must be set to the size of the structure
   int iSize;
+/// Possible valies: \ref ADL_DL_FANCTRL_SPEED_TYPE_PERCENT or \ref ADL_DL_FANCTRL_SPEED_TYPE_RPM
   int iSpeedType;
+/// Fan speed value
   int iFanSpeed;
+/// The only flag for now is: \ref ADL_DL_FANCTRL_FLAG_USER_DEFINED_SPEED
   int iFlags;
 } ADLFanSpeedValue;
 
-typedef struct ADLDisplayID
-{
-  int iDisplayLogicalIndex;
-  int iDisplayPhysicalIndex;
-  int iDisplayLogicalAdapterIndex;
-  int iDisplayPhysicalAdapterIndex;
-} ADLDisplayID, *LPADLDisplayID;
-
-typedef struct ADLDisplayInfo
-{
-  ADLDisplayID displayID;
-  int  iDisplayControllerIndex;
-  char strDisplayName[ADL_MAX_PATH];
-  char strDisplayManufacturerName[ADL_MAX_PATH];
-  int  iDisplayType;
-  int  iDisplayOutputType;
-  int  iDisplayConnector;
-  int  iDisplayInfoMask;
-  int  iDisplayInfoValue;
-} ADLDisplayInfo, *LPADLDisplayInfo;
-
-typedef struct ADLBiosInfo
-{
-  char strPartNumber[ADL_MAX_PATH];
-  char strVersion[ADL_MAX_PATH];
-  char strDate[ADL_MAX_PATH];
-} ADLBiosInfo, *LPADLBiosInfo;
-
+/////////////////////////////////////////////////////////////////////////////////////////////
+///\brief Structure containing information about current power management related activity.
+///
+/// This structure is used to store information about current power management related activity.
+/// This structure (Overdrive 5 interfaces) is used by the ADL_PM_CurrentActivity_Get() function.
+/// \nosubgrouping
+////////////////////////////////////////////////////////////////////////////////////////////
 typedef struct ADLPMActivity
 {
-  int iSize;
-  int iEngineClock;
-  int iMemoryClock;
-  int iVddc;
-  int iActivityPercent;
-  int iCurrentPerformanceLevel;
-  int iCurrentBusSpeed;
-  int iCurrentBusLanes;
-  int iMaximumBusLanes;
-  int iReserved;
+/// Must be set to the size of the structure
+    int iSize;
+/// Current engine clock.
+    int iEngineClock;
+/// Current memory clock.
+    int iMemoryClock;
+/// Current core voltage.
+    int iVddc;
+/// GPU utilization.
+    int iActivityPercent;
+/// Performance level index.
+    int iCurrentPerformanceLevel;
+/// Current PCIE bus speed.
+    int iCurrentBusSpeed;
+/// Number of PCIE bus lanes.
+    int iCurrentBusLanes;
+/// Maximum number of PCIE bus lanes.
+    int iMaximumBusLanes;
+/// Reserved for future purposes.
+    int iReserved;
 } ADLPMActivity;
 
+////////////////////////////////////////////////////////////////////////////////////////////
+///\brief Structure containing the range of Overdrive parameter.
+///
+/// This structure is used to store information about the range of Overdrive parameter.
+/// This structure is used by ADLODParameters.
+/// \nosubgrouping
+////////////////////////////////////////////////////////////////////////////////////////////
 typedef struct ADLODParameterRange
 {
+/// Minimum parameter value.
   int iMin;
+/// Maximum parameter value.
   int iMax;
+/// Parameter step value.
   int iStep;
 } ADLODParameterRange;
 
+/////////////////////////////////////////////////////////////////////////////////////////////
+///\brief Structure containing information about Overdrive parameters.
+///
+/// This structure is used to store information about Overdrive parameters.
+/// This structure is used by the ADL_Overdrive5_ODParameters_Get() function.
+/// \nosubgrouping
+////////////////////////////////////////////////////////////////////////////////////////////
 typedef struct ADLODParameters
 {
+/// Must be set to the size of the structure
   int iSize;
+/// Number of standard performance states.
   int iNumberOfPerformanceLevels;
+/// Indicates whether the GPU is capable to measure its activity.
   int iActivityReportingSupported;
+/// Indicates whether the GPU supports discrete performance levels or performance range.
   int iDiscretePerformanceLevels;
+/// Reserved for future use.
   int iReserved;
+/// Engine clock range.
   ADLODParameterRange sEngineClock;
+/// Memory clock range.
   ADLODParameterRange sMemoryClock;
+/// Core voltage range.
   ADLODParameterRange sVddc;
 } ADLODParameters;
 
+/////////////////////////////////////////////////////////////////////////////////////////////
+///\brief Structure containing information about Overdrive 6 fan speed information
+///
+/// This structure is used to store information about Overdrive 6 fan speed information
+/// \nosubgrouping
+////////////////////////////////////////////////////////////////////////////////////////////
+typedef struct ADLOD6FanSpeedInfo
+{
+    /// Contains a bitmap of the valid fan speed type flags.  Possible values: \ref ADL_OD6_FANSPEED_TYPE_PERCENT, \ref ADL_OD6_FANSPEED_TYPE_RPM, \ref ADL_OD6_FANSPEED_USER_DEFINED
+    int     iSpeedType;
+    /// Contains current fan speed in percent (if valid flag exists in iSpeedType)
+    int     iFanSpeedPercent;
+    /// Contains current fan speed in RPM (if valid flag exists in iSpeedType)
+    int        iFanSpeedRPM;
+
+    /// Value for future extension
+    int     iExtValue;
+    /// Mask for future extension
+    int     iExtMask;
+
+} ADLOD6FanSpeedInfo;
+
+/////////////////////////////////////////////////////////////////////////////////////////////
+///\brief Structure containing information about Overdrive 6 fan speed value
+///
+/// This structure is used to store information about Overdrive 6 fan speed value
+/// \nosubgrouping
+////////////////////////////////////////////////////////////////////////////////////////////
+typedef struct ADLOD6FanSpeedValue
+{
+    /// Indicates the units of the fan speed.  Possible values: \ref ADL_OD6_FANSPEED_TYPE_PERCENT, \ref ADL_OD6_FANSPEED_TYPE_RPM
+    int     iSpeedType;
+    /// Fan speed value (units as indicated above)
+    int     iFanSpeed;
+
+    /// Value for future extension
+    int     iExtValue;
+    /// Mask for future extension
+    int     iExtMask;
+
+} ADLOD6FanSpeedValue;
+
+/////////////////////////////////////////////////////////////////////////////////////////////
+///\brief Structure containing information about current Overdrive 6 performance status.
+///
+/// This structure is used to store information about current Overdrive 6 performance status.
+/// \nosubgrouping
+////////////////////////////////////////////////////////////////////////////////////////////
+typedef struct ADLOD6CurrentStatus
+{
+    /// Current engine clock in 10 KHz.
+    int     iEngineClock;
+    /// Current memory clock in 10 KHz.
+    int     iMemoryClock;
+    /// Current GPU activity in percent.  This
+    /// indicates how "busy" the GPU is.
+    int     iActivityPercent;
+    /// Not used.  Reserved for future use.
+    int     iCurrentPerformanceLevel;
+    /// Current PCI-E bus speed
+    int     iCurrentBusSpeed;
+    /// Current PCI-E bus # of lanes
+    int     iCurrentBusLanes;
+    /// Maximum possible PCI-E bus # of lanes
+    int     iMaximumBusLanes;
+
+    /// Value for future extension
+    int     iExtValue;
+    /// Mask for future extension
+    int     iExtMask;
+
+} ADLOD6CurrentStatus;
+
+/////////////////////////////////////////////////////////////////////////////////////////////
+///\brief Structure containing information about Overdrive 6 clock range
+///
+/// This structure is used to store information about Overdrive 6 clock range
+/// \nosubgrouping
+////////////////////////////////////////////////////////////////////////////////////////////
+typedef struct ADLOD6ParameterRange
+{
+    /// The starting value of the clock range
+    int     iMin;
+    /// The ending value of the clock range
+    int     iMax;
+    /// The minimum increment between clock values
+    int     iStep;
+
+} ADLOD6ParameterRange;
+
+/////////////////////////////////////////////////////////////////////////////////////////////
+///\brief Structure containing information about Overdrive 6 capabilities
+///
+/// This structure is used to store information about Overdrive 6 capabilities
+/// \nosubgrouping
+////////////////////////////////////////////////////////////////////////////////////////////
+typedef struct ADLOD6Capabilities
+{
+    /// Contains a bitmap of the OD6 capability flags.  Possible values: \ref ADL_OD6_CAPABILITY_SCLK_CUSTOMIZATION,
+    /// \ref ADL_OD6_CAPABILITY_MCLK_CUSTOMIZATION, \ref ADL_OD6_CAPABILITY_GPU_ACTIVITY_MONITOR
+    int     iCapabilities;
+    /// Contains a bitmap indicating the power states
+    /// supported by OD6.  Currently only the performance state
+    /// is supported. Possible Values: \ref ADL_OD6_SUPPORTEDSTATE_PERFORMANCE
+    int     iSupportedStates;
+    /// Number of levels. OD6 will always use 2 levels, which describe
+    /// the minimum to maximum clock ranges.
+    /// The 1st level indicates the minimum clocks, and the 2nd level
+    /// indicates the maximum clocks.
+    int     iNumberOfPerformanceLevels;
+    /// Contains the hard limits of the sclk range.  Overdrive
+    /// clocks cannot be set outside this range.
+    ADLOD6ParameterRange     sEngineClockRange;
+    /// Contains the hard limits of the mclk range.  Overdrive
+    /// clocks cannot be set outside this range.
+    ADLOD6ParameterRange     sMemoryClockRange;
+
+    /// Value for future extension
+    int     iExtValue;
+    /// Mask for future extension
+    int     iExtMask;
+
+} ADLOD6Capabilities;
+
+/////////////////////////////////////////////////////////////////////////////////////////////
+///\brief Structure containing information about Overdrive level.
+///
+/// This structure is used to store information about Overdrive level.
+/// This structure is used by ADLODPerformanceLevels.
+/// \nosubgrouping
+////////////////////////////////////////////////////////////////////////////////////////////
 typedef struct ADLODPerformanceLevel
 {
+/// Engine clock.
   int iEngineClock;
+/// Memory clock.
   int iMemoryClock;
+/// Core voltage.
   int iVddc;
 } ADLODPerformanceLevel;
 
-/*
- * Attention: we had to change this struct due to an out-of-bound problem mentioned here:
- * https://github.com/hashcat/hashcat/issues/244
- * the change: ADLODPerformanceLevel aLevels [1] -> ADLODPerformanceLevel aLevels [2]
- */
-
-typedef struct ADLODPerformanceLevels
-{
-  int iSize;
-  int iReserved;
-  ADLODPerformanceLevel aLevels [2];
-} ADLODPerformanceLevels;
-
-typedef struct ADLOD6FanSpeedInfo
-{
-  int iSpeedType;
-  int iFanSpeedPercent;
-  int iFanSpeedRPM;
-  int iExtValue;
-  int iExtMask;
-} ADLOD6FanSpeedInfo;
-
-typedef struct ADLOD6FanSpeedValue
-{
-  int iSpeedType;
-  int iFanSpeed;
-  int iExtValue;
-  int iExtMask;
-} ADLOD6FanSpeedValue;
-
-typedef struct ADLOD6CurrentStatus
-{
-  int iEngineClock;
-  int iMemoryClock;
-  int iActivityPercent;
-  int iCurrentPerformanceLevel;
-  int iCurrentBusSpeed;
-  int iCurrentBusLanes;
-  int iMaximumBusLanes;
-  int iExtValue;
-  int iExtMask;
-} ADLOD6CurrentStatus;
-
-typedef struct ADLOD6ParameterRange
-{
-  int iMin;
-  int iMax;
-  int iStep;
-} ADLOD6ParameterRange;
-
-typedef struct ADLOD6Capabilities
-{
-  int iCapabilities;
-  int iSupportedStates;
-  int iNumberOfPerformanceLevels;
-  ADLOD6ParameterRange sEngineClockRange;
-  ADLOD6ParameterRange sMemoryClockRange;
-  int iExtValue;
-  int iExtMask;
-} ADLOD6Capabilities;
-
+/////////////////////////////////////////////////////////////////////////////////////////////
+///\brief Structure containing information about Overdrive 6 clock values.
+///
+/// This structure is used to store information about Overdrive 6 clock values.
+/// \nosubgrouping
+////////////////////////////////////////////////////////////////////////////////////////////
 typedef struct ADLOD6PerformanceLevel
 {
-  int iEngineClock;
-  int iMemoryClock;
+    /// Engine (core) clock.
+    int iEngineClock;
+    /// Memory clock.
+    int iMemoryClock;
+
 } ADLOD6PerformanceLevel;
 
-/*
- * Attention: we had to change this struct due to an out-of-bound problem mentioned here:
- * https://github.com/hashcat/hashcat/issues/244
- * the change: ADLOD6PerformanceLevel aLevels [1] -> ADLOD6PerformanceLevel aLevels [2]
- */
-
+/////////////////////////////////////////////////////////////////////////////////////////////
+///\brief Structure containing information about Overdrive 6 clocks.
+///
+/// This structure is used to store information about Overdrive 6 clocks.  This is a
+/// variable-sized structure.  iNumberOfPerformanceLevels indicate how many elements
+/// are contained in the aLevels array.
+/// \nosubgrouping
+////////////////////////////////////////////////////////////////////////////////////////////
 typedef struct ADLOD6StateInfo
 {
-  int iNumberOfPerformanceLevels;
-  int iExtValue;
-  int iExtMask;
-  ADLOD6PerformanceLevel aLevels [2];
+    /// Number of levels.  OD6 uses clock ranges instead of discrete performance levels.
+    /// iNumberOfPerformanceLevels is always 2.  The 1st level indicates the minimum clocks
+    /// in the range.  The 2nd level indicates the maximum clocks in the range.
+    int     iNumberOfPerformanceLevels;
+
+    /// Value for future extension
+    int     iExtValue;
+    /// Mask for future extension
+    int     iExtMask;
+
+    /// Variable-sized array of levels.
+    /// The number of elements in the array is specified by iNumberofPerformanceLevels.
+    ADLOD6PerformanceLevel aLevels [1];
+
 } ADLOD6StateInfo;
 
+/////////////////////////////////////////////////////////////////////////////////////////////
+///\brief Structure containing information about Overdrive performance levels.
+///
+/// This structure is used to store information about Overdrive performance levels.
+/// This structure is used by the ADL_Overdrive5_ODPerformanceLevels_Get() and ADL_Overdrive5_ODPerformanceLevels_Set() functions.
+/// \nosubgrouping
+////////////////////////////////////////////////////////////////////////////////////////////
+typedef struct ADLODPerformanceLevels
+{
+/// Must be set to sizeof( \ref ADLODPerformanceLevels ) + sizeof( \ref ADLODPerformanceLevel ) * (ADLODParameters.iNumberOfPerformanceLevels - 1)
+  int iSize;
+  int iReserved;
+/// Array of performance state descriptors. Must have ADLODParameters.iNumberOfPerformanceLevels elements.
+  ADLODPerformanceLevel aLevels [1];
+} ADLODPerformanceLevels;
+
+/////////////////////////////////////////////////////////////////////////////////////////////
+///\brief Structure containing information about Performance Metrics data
+///
+/// This structure is used to store information about Performance Metrics data output
+/// \nosubgrouping
+////////////////////////////////////////////////////////////////////////////////////////////
+typedef struct ADLSingleSensorData
+{
+    int supported;
+    int  value;
+} ADLSingleSensorData;
+
+typedef struct ADLPMLogDataOutput
+{
+    int size;
+    ADLSingleSensorData sensors[ADL_PMLOG_MAX_SENSORS];
+}ADLPMLogDataOutput;
+
+/// \brief Handle to ADL client context.
+///
+///  ADL clients obtain context handle from initial call to \ref ADL2_Main_Control_Create.
+///  Clients have to pass the handle to each subsequent ADL call and finally destroy
+///  the context with call to \ref ADL2_Main_Control_Destroy
+/// \nosubgrouping
+typedef void *ADL_CONTEXT_HANDLE;
+
 #if defined (__MSC_VER)
 #define ADL_API_CALL __cdecl
 #elif defined (_WIN32) || defined (__WIN32__)
@@ -251,62 +592,51 @@ typedef void* (ADL_API_CALL *ADL_MAIN_MALLOC_CALLBACK )( int );
 
 typedef int HM_ADAPTER_ADL;
 
-typedef struct struct_ADLOD6MemClockState
-{
-  ADLOD6StateInfo state;
-  ADLOD6PerformanceLevel level;
-
-} ADLOD6MemClockState;
-
-typedef int (ADL_API_CALL *ADL_MAIN_CONTROL_DESTROY) (void);
-typedef int (ADL_API_CALL *ADL_MAIN_CONTROL_CREATE) (ADL_MAIN_MALLOC_CALLBACK, int);
-typedef int (ADL_API_CALL *ADL_ADAPTER_NUMBEROFADAPTERS_GET) (int *);
-typedef int (ADL_API_CALL *ADL_ADAPTER_ADAPTERINFO_GET) (LPAdapterInfo, int);
-typedef int (ADL_API_CALL *ADL_DISPLAY_DISPLAYINFO_GET) (int, int *, ADLDisplayInfo **, int);
-typedef int (ADL_API_CALL *ADL_OVERDRIVE5_TEMPERATURE_GET) (int, int, ADLTemperature *);
-typedef int (ADL_API_CALL *ADL_OVERDRIVE6_TEMPERATURE_GET) (int, int *);
-typedef int (ADL_API_CALL *ADL_OVERDRIVE5_CURRENTACTIVITY_GET) (int, ADLPMActivity *);
-typedef int (ADL_API_CALL *ADL_OVERDRIVE5_THERMALDEVICES_ENUM) (int, int, ADLThermalControllerInfo *);
-typedef int (ADL_API_CALL *ADL_ADAPTER_ID_GET) (int, int *);
-typedef int (ADL_API_CALL *ADL_ADAPTER_VIDEOBIOSINFO_GET) (int, ADLBiosInfo *);
-typedef int (ADL_API_CALL *ADL_OVERDRIVE5_FANSPEEDINFO_GET) (int, int, ADLFanSpeedInfo *);
-typedef int (ADL_API_CALL *ADL_OVERDRIVE5_FANSPEED_GET) (int, int, ADLFanSpeedValue *);
-typedef int (ADL_API_CALL *ADL_OVERDRIVE6_FANSPEED_GET) (int, ADLOD6FanSpeedInfo *);
-typedef int (ADL_API_CALL *ADL_OVERDRIVE5_ODPARAMETERS_GET) (int, ADLODParameters *);
-typedef int (ADL_API_CALL *ADL_OVERDRIVE5_ODPERFORMANCELEVELS_GET) (int, int, ADLODPerformanceLevels *);
-typedef int (ADL_API_CALL *ADL_ADAPTER_ACTIVE_GET) (int, int *);
-typedef int (ADL_API_CALL *ADL_OVERDRIVE_CAPS) (int, int *, int *, int *);
-typedef int (ADL_API_CALL *ADL_OVERDRIVE6_CURRENTSTATUS_GET) (int, ADLOD6CurrentStatus *);
-typedef int (ADL_API_CALL *ADL_OVERDRIVE6_STATEINFO_GET) (int, int, ADLOD6MemClockState *);
-typedef int (ADL_API_CALL *ADL_OVERDRIVE6_CAPABILITIES_GET) (int, ADLOD6Capabilities *);
-typedef int (ADL_API_CALL *ADL_OVERDRIVE6_TARGETTEMPERATUREDATA_GET) (int, int *, int *);
-typedef int (ADL_API_CALL *ADL_OVERDRIVE6_TARGETTEMPERATURERANGEINFO_GET) (int, ADLOD6ParameterRange *);
+typedef int (ADL_API_CALL *ADL_ADAPTER_ACTIVE_GET ) ( int, int* );
+typedef int (ADL_API_CALL *ADL_ADAPTER_ADAPTERINFO_GET ) ( LPAdapterInfo, int );
+typedef int (ADL_API_CALL *ADL_ADAPTER_NUMBEROFADAPTERS_GET ) ( int* );
+typedef int (ADL_API_CALL *ADL_MAIN_CONTROL_CREATE )(ADL_MAIN_MALLOC_CALLBACK, int );
+typedef int (ADL_API_CALL *ADL_MAIN_CONTROL_DESTROY )();
+typedef int (ADL_API_CALL *ADL_OVERDRIVE5_CURRENTACTIVITY_GET ) (int iAdapterIndex, ADLPMActivity *lpActivity);
+typedef int (ADL_API_CALL *ADL_OVERDRIVE5_FANSPEEDINFO_GET ) (int iAdapterIndex, int iThermalControllerIndex, ADLFanSpeedInfo *lpFanSpeedInfo);
+typedef int (ADL_API_CALL *ADL_OVERDRIVE5_FANSPEED_GET ) (int iAdapterIndex, int iThermalControllerIndex, ADLFanSpeedValue *lpFanSpeedValue);
+typedef int (ADL_API_CALL *ADL_OVERDRIVE5_ODPARAMETERS_GET ) (int iAdapterIndex, ADLODParameters *lpOdParameters);
+typedef int (ADL_API_CALL *ADL_OVERDRIVE5_ODPERFORMANCELEVELS_GET ) (int iAdapterIndex, int iDefault, ADLODPerformanceLevels *lpOdPerformanceLevels);
+typedef int (ADL_API_CALL *ADL_OVERDRIVE5_TEMPERATURE_GET ) (int iAdapterIndex, int iThermalControllerIndex, ADLTemperature *lpTemperature);
+typedef int (ADL_API_CALL *ADL_OVERDRIVE5_THERMALDEVICES_ENUM ) (int iAdapterIndex, int iThermalControllerIndex, ADLThermalControllerInfo *lpThermalControllerInfo);
+typedef int (ADL_API_CALL *ADL_OVERDRIVE6_CAPABILITIES_GET ) (int iAdapterIndex, ADLOD6Capabilities *lpODCapabilities);
+typedef int (ADL_API_CALL *ADL_OVERDRIVE6_CURRENTSTATUS_GET )(int iAdapterIndex, ADLOD6CurrentStatus *lpCurrentStatus);
+typedef int (ADL_API_CALL *ADL_OVERDRIVE6_FANSPEED_GET )(int iAdapterIndex, ADLOD6FanSpeedInfo *lpFanSpeedInfo);
+typedef int (ADL_API_CALL *ADL_OVERDRIVE6_STATEINFO_GET )(int iAdapterIndex, int iStateType, ADLOD6StateInfo *lpStateInfo);
+typedef int (ADL_API_CALL *ADL_OVERDRIVE6_TEMPERATURE_GET )(int iAdapterIndex, int *lpTemperature);
+typedef int (ADL_API_CALL *ADL_OVERDRIVE_CAPS ) (int iAdapterIndex, int *iSupported, int *iEnabled, int *iVersion);
+typedef int (ADL_API_CALL *ADL2_OVERDRIVE_CAPS) (ADL_CONTEXT_HANDLE context, int iAdapterIndex, int * iSupported, int * iEnabled, int * iVersion);
+typedef int (ADL_API_CALL *ADL2_NEW_QUERYPMLOGDATA_GET) (ADL_CONTEXT_HANDLE, int, ADLPMLogDataOutput*);
 
 typedef struct hm_adl_lib
 {
   hc_dynlib_t lib;
 
-  ADL_MAIN_CONTROL_DESTROY ADL_Main_Control_Destroy;
-  ADL_MAIN_CONTROL_CREATE ADL_Main_Control_Create;
-  ADL_ADAPTER_NUMBEROFADAPTERS_GET ADL_Adapter_NumberOfAdapters_Get;
+  ADL_ADAPTER_ACTIVE_GET ADL_Adapter_Active_Get;
   ADL_ADAPTER_ADAPTERINFO_GET ADL_Adapter_AdapterInfo_Get;
-  ADL_DISPLAY_DISPLAYINFO_GET ADL_Display_DisplayInfo_Get;
-  ADL_ADAPTER_ID_GET ADL_Adapter_ID_Get;
-  ADL_ADAPTER_VIDEOBIOSINFO_GET ADL_Adapter_VideoBiosInfo_Get;
-  ADL_OVERDRIVE5_THERMALDEVICES_ENUM ADL_Overdrive5_ThermalDevices_Enum;
-  ADL_OVERDRIVE5_TEMPERATURE_GET ADL_Overdrive5_Temperature_Get;
-  ADL_OVERDRIVE6_TEMPERATURE_GET ADL_Overdrive6_Temperature_Get;
+  ADL_ADAPTER_NUMBEROFADAPTERS_GET ADL_Adapter_NumberOfAdapters_Get;
+  ADL_MAIN_CONTROL_CREATE ADL_Main_Control_Create;
+  ADL_MAIN_CONTROL_DESTROY ADL_Main_Control_Destroy;
   ADL_OVERDRIVE5_CURRENTACTIVITY_GET ADL_Overdrive5_CurrentActivity_Get;
   ADL_OVERDRIVE5_FANSPEEDINFO_GET ADL_Overdrive5_FanSpeedInfo_Get;
   ADL_OVERDRIVE5_FANSPEED_GET ADL_Overdrive5_FanSpeed_Get;
-  ADL_OVERDRIVE6_FANSPEED_GET ADL_Overdrive6_FanSpeed_Get;
-  ADL_ADAPTER_ACTIVE_GET ADL_Adapter_Active_Get;
-  ADL_OVERDRIVE_CAPS ADL_Overdrive_Caps;
+  ADL_OVERDRIVE5_ODPARAMETERS_GET ADL_Overdrive5_ODParameters_Get;
+  ADL_OVERDRIVE5_ODPERFORMANCELEVELS_GET ADL_Overdrive5_ODPerformanceLevels_Get;
+  ADL_OVERDRIVE5_TEMPERATURE_GET ADL_Overdrive5_Temperature_Get;
+  ADL_OVERDRIVE5_THERMALDEVICES_ENUM ADL_Overdrive5_ThermalDevices_Enum;
   ADL_OVERDRIVE6_CAPABILITIES_GET ADL_Overdrive6_Capabilities_Get;
-  ADL_OVERDRIVE6_STATEINFO_GET  ADL_Overdrive6_StateInfo_Get;
   ADL_OVERDRIVE6_CURRENTSTATUS_GET ADL_Overdrive6_CurrentStatus_Get;
-  ADL_OVERDRIVE6_TARGETTEMPERATUREDATA_GET ADL_Overdrive6_TargetTemperatureData_Get;
-  ADL_OVERDRIVE6_TARGETTEMPERATURERANGEINFO_GET ADL_Overdrive6_TargetTemperatureRangeInfo_Get;
+  ADL_OVERDRIVE6_FANSPEED_GET ADL_Overdrive6_FanSpeed_Get;
+  ADL_OVERDRIVE6_STATEINFO_GET  ADL_Overdrive6_StateInfo_Get;
+  ADL_OVERDRIVE6_TEMPERATURE_GET ADL_Overdrive6_Temperature_Get;
+  ADL_OVERDRIVE_CAPS ADL_Overdrive_Caps;
+  ADL2_OVERDRIVE_CAPS ADL2_Overdrive_Caps;
+  ADL2_NEW_QUERYPMLOGDATA_GET ADL2_New_QueryPMLogData_Get;
 
 } hm_adl_lib_t;
 
@@ -326,6 +656,8 @@ int hm_ADL_Overdrive_CurrentActivity_Get (void *hashcat_ctx, int iAdapterIndex,
 int hm_ADL_Overdrive5_FanSpeed_Get (void *hashcat_ctx, int iAdapterIndex, int iThermalControllerIndex, ADLFanSpeedValue *lpFanSpeedValue);
 int hm_ADL_Overdrive6_FanSpeed_Get (void *hashcat_ctx, int iAdapterIndex, ADLOD6FanSpeedInfo *lpFanSpeedInfo);
 int hm_ADL_Overdrive_Caps (void *hashcat_ctx, int iAdapterIndex, int *od_supported, int *od_enabled, int *od_version);
-int hm_ADL_Overdrive6_TargetTemperatureData_Get (void *hashcat_ctx, int iAdapterIndex, int *cur_temp, int *default_temp);
+int hm_ADL2_Overdrive_Caps (void *hashcat_ctx, int iAdapterIndex, int *od_supported, int *od_enabled, int *od_version);
+int hm_ADL2_New_QueryPMLogData_Get (void *hashcat_ctx, int iAdapterIndex, ADLPMLogDataOutput *lpDataOutput);
+
 
 #endif // _EXT_ADL_H
diff --git a/include/ext_OpenCL.h b/include/ext_OpenCL.h
index 42e37deed..40f3515ed 100644
--- a/include/ext_OpenCL.h
+++ b/include/ext_OpenCL.h
@@ -46,6 +46,7 @@ typedef cl_context       (CL_API_CALL *OCL_CLCREATECONTEXT)           (const cl_
 typedef cl_kernel        (CL_API_CALL *OCL_CLCREATEKERNEL)            (cl_program, const char *, cl_int *);
 typedef cl_program       (CL_API_CALL *OCL_CLCREATEPROGRAMWITHBINARY) (cl_context, cl_uint, const cl_device_id *, const size_t *, const unsigned char **, cl_int *, cl_int *);
 typedef cl_program       (CL_API_CALL *OCL_CLCREATEPROGRAMWITHSOURCE) (cl_context, cl_uint, const char **, const size_t *, cl_int *);
+typedef cl_int           (CL_API_CALL *OCL_CLENQUEUEFILLBUFFER)       (cl_command_queue, cl_mem, const void *, size_t, size_t, size_t, cl_uint, const cl_event *, cl_event *);
 typedef cl_int           (CL_API_CALL *OCL_CLENQUEUECOPYBUFFER)       (cl_command_queue, cl_mem, cl_mem, size_t, size_t, size_t, cl_uint, const cl_event *, cl_event *);
 typedef void *           (CL_API_CALL *OCL_CLENQUEUEMAPBUFFER)        (cl_command_queue, cl_mem, cl_bool, cl_map_flags, size_t, size_t, cl_uint, const cl_event *, cl_event *, cl_int *);
 typedef cl_int           (CL_API_CALL *OCL_CLENQUEUENDRANGEKERNEL)    (cl_command_queue, cl_kernel, cl_uint, const size_t *, const size_t *, const size_t *, cl_uint, const cl_event *, cl_event *);
@@ -87,6 +88,7 @@ typedef struct hc_opencl_lib
   OCL_CLCREATEPROGRAMWITHBINARY clCreateProgramWithBinary;
   OCL_CLCREATEPROGRAMWITHSOURCE clCreateProgramWithSource;
   OCL_CLENQUEUECOPYBUFFER       clEnqueueCopyBuffer;
+  OCL_CLENQUEUEFILLBUFFER       clEnqueueFillBuffer;
   OCL_CLENQUEUEMAPBUFFER        clEnqueueMapBuffer;
   OCL_CLENQUEUENDRANGEKERNEL    clEnqueueNDRangeKernel;
   OCL_CLENQUEUEREADBUFFER       clEnqueueReadBuffer;
diff --git a/include/ext_cuda.h b/include/ext_cuda.h
index 0e3619fc4..11ed86048 100644
--- a/include/ext_cuda.h
+++ b/include/ext_cuda.h
@@ -1028,14 +1028,14 @@ typedef CUresult (CUDA_API_CALL *CUDA_CUINIT)                   (unsigned int);
 typedef CUresult (CUDA_API_CALL *CUDA_CULAUNCHKERNEL)           (CUfunction, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, CUstream, void **, void **);
 typedef CUresult (CUDA_API_CALL *CUDA_CUMEMALLOC)               (CUdeviceptr *, size_t);
 typedef CUresult (CUDA_API_CALL *CUDA_CUMEMALLOCHOST)           (void **, size_t);
-typedef CUresult (CUDA_API_CALL *CUDA_CUMEMCPYDTOD)             (CUdeviceptr, CUdeviceptr, size_t);
-typedef CUresult (CUDA_API_CALL *CUDA_CUMEMCPYDTOH)             (void *, CUdeviceptr, size_t);
-typedef CUresult (CUDA_API_CALL *CUDA_CUMEMCPYHTOD)             (CUdeviceptr, const void *, size_t);
+typedef CUresult (CUDA_API_CALL *CUDA_CUMEMCPYDTODASYNC)        (CUdeviceptr, CUdeviceptr, size_t, CUstream);
+typedef CUresult (CUDA_API_CALL *CUDA_CUMEMCPYDTOHASYNC)        (void *, CUdeviceptr, size_t, CUstream);
+typedef CUresult (CUDA_API_CALL *CUDA_CUMEMCPYHTODASYNC)        (CUdeviceptr, const void *, size_t, CUstream);
 typedef CUresult (CUDA_API_CALL *CUDA_CUMEMFREE)                (CUdeviceptr);
 typedef CUresult (CUDA_API_CALL *CUDA_CUMEMFREEHOST)            (void *);
 typedef CUresult (CUDA_API_CALL *CUDA_CUMEMGETINFO)             (size_t *, size_t *);
-typedef CUresult (CUDA_API_CALL *CUDA_CUMEMSETD32)              (CUdeviceptr, unsigned int, size_t);
-typedef CUresult (CUDA_API_CALL *CUDA_CUMEMSETD8)               (CUdeviceptr, unsigned char, size_t);
+typedef CUresult (CUDA_API_CALL *CUDA_CUMEMSETD32ASYNC)         (CUdeviceptr, unsigned int, size_t, CUstream);
+typedef CUresult (CUDA_API_CALL *CUDA_CUMEMSETD8ASYNC)          (CUdeviceptr, unsigned char, size_t, CUstream);
 typedef CUresult (CUDA_API_CALL *CUDA_CUMODULEGETFUNCTION)      (CUfunction *, CUmodule, const char *);
 typedef CUresult (CUDA_API_CALL *CUDA_CUMODULEGETGLOBAL)        (CUdeviceptr *, size_t *, CUmodule, const char *);
 typedef CUresult (CUDA_API_CALL *CUDA_CUMODULELOAD)             (CUmodule *, const char *);
@@ -1090,14 +1090,14 @@ typedef struct hc_cuda_lib
   CUDA_CULAUNCHKERNEL           cuLaunchKernel;
   CUDA_CUMEMALLOC               cuMemAlloc;
   CUDA_CUMEMALLOCHOST           cuMemAllocHost;
-  CUDA_CUMEMCPYDTOD             cuMemcpyDtoD;
-  CUDA_CUMEMCPYDTOH             cuMemcpyDtoH;
-  CUDA_CUMEMCPYHTOD             cuMemcpyHtoD;
+  CUDA_CUMEMCPYDTODASYNC        cuMemcpyDtoDAsync;
+  CUDA_CUMEMCPYDTOHASYNC        cuMemcpyDtoHAsync;
+  CUDA_CUMEMCPYHTODASYNC        cuMemcpyHtoDAsync;
   CUDA_CUMEMFREE                cuMemFree;
   CUDA_CUMEMFREEHOST            cuMemFreeHost;
   CUDA_CUMEMGETINFO             cuMemGetInfo;
-  CUDA_CUMEMSETD32              cuMemsetD32;
-  CUDA_CUMEMSETD8               cuMemsetD8;
+  CUDA_CUMEMSETD32ASYNC         cuMemsetD32Async;
+  CUDA_CUMEMSETD8ASYNC          cuMemsetD8Async;
   CUDA_CUMODULEGETFUNCTION      cuModuleGetFunction;
   CUDA_CUMODULEGETGLOBAL        cuModuleGetGlobal;
   CUDA_CUMODULELOAD             cuModuleLoad;
diff --git a/include/ext_hip.h b/include/ext_hip.h
index 15840d671..c29309ef3 100644
--- a/include/ext_hip.h
+++ b/include/ext_hip.h
@@ -6,995 +6,343 @@
 #ifndef _EXT_HIP_H
 #define _EXT_HIP_H
 
-/**
- * TODO: FIX ME
- */
+// The general Idea with HIP is to use it for AMD GPU since we use CUDA for NV
+// Therefore, we need to take certain items, such as hipDeviceptr_t from driver specific paths like amd_driver_types.h
+// We just need to keep this in mind in case we need to update these constants from future SDK versions
 
-#define __HIP_API_VERSION 4221131
+// start: amd_driver_types.h
 
-/**
- * HIP device pointer
- * HIPdeviceptr is defined as an unsigned integer type whose size matches the size of a pointer on the target platform.
- */
-#if __HIP_API_VERSION >= 3020
+typedef void* hipDeviceptr_t;
 
-#if defined(_WIN64) || defined(__LP64__)
-typedef unsigned long long HIPdeviceptr;
+typedef enum hipFunction_attribute {
+    HIP_FUNC_ATTRIBUTE_MAX_THREADS_PER_BLOCK,
+    HIP_FUNC_ATTRIBUTE_SHARED_SIZE_BYTES,
+    HIP_FUNC_ATTRIBUTE_CONST_SIZE_BYTES,
+    HIP_FUNC_ATTRIBUTE_LOCAL_SIZE_BYTES,
+    HIP_FUNC_ATTRIBUTE_NUM_REGS,
+    HIP_FUNC_ATTRIBUTE_PTX_VERSION,
+    HIP_FUNC_ATTRIBUTE_BINARY_VERSION,
+    HIP_FUNC_ATTRIBUTE_CACHE_MODE_CA,
+    HIP_FUNC_ATTRIBUTE_MAX_DYNAMIC_SHARED_SIZE_BYTES,
+    HIP_FUNC_ATTRIBUTE_PREFERRED_SHARED_MEMORY_CARVEOUT,
+    HIP_FUNC_ATTRIBUTE_MAX
+}hipFunction_attribute;
+
+// stop: amd_driver_types.h
+
+// start: hip_runtime_api.h
+
+typedef int hipDevice_t;
+typedef struct ihipCtx_t* hipCtx_t;
+typedef struct ihipEvent_t* hipEvent_t;
+typedef struct ihipStream_t* hipStream_t;
+typedef struct ihipModule_t* hipModule_t;
+typedef struct ihipModuleSymbol_t* hipFunction_t;
+
+// Ignoring error-code return values from hip APIs is discouraged. On C++17,
+// we can make that yield a warning
+#if __cplusplus >= 201703L
+#define __HIP_NODISCARD [[nodiscard]]
 #else
-typedef unsigned int HIPdeviceptr;
+#define __HIP_NODISCARD
 #endif
 
-#endif /* __HIP_API_VERSION >= 3020 */
-
-typedef int HIPdevice;                                     /**< HIP device */
-typedef struct HIPctx_st *HIPcontext;                       /**< HIP context */
-typedef struct HIPevent_st *HIPevent;                       /**< HIP event */
-typedef struct HIPfunc_st *HIPfunction;                     /**< HIP function */
-typedef struct HIPmod_st *HIPmodule;                        /**< HIP module */
-typedef struct HIPstream_st *HIPstream;                     /**< HIP stream */
-typedef struct HIPlinkState_st *HIPlinkState;
-
-
-typedef enum hipError_enum {
-    /**
-     * The API call returned with no errors. In the case of query calls, this
-     * also means that the operation being queried is complete (see
-     * ::hipEventQuery() and ::hipStreamQuery()).
-     */
-    HIP_SUCCESS                              = 0,
-
-    /**
-     * This indicates that one or more of the parameters passed to the API call
-     * is not within an acceptable range of values.
-     */
-    HIP_ERROR_INVALID_VALUE                  = 1,
-
-    /**
-     * The API call failed because it was unable to allocate enough memory to
-     * perform the requested operation.
-     */
-    HIP_ERROR_OUT_OF_MEMORY                  = 2,
-
-    /**
-     * This indicates that the HIP driver has not been initialized with
-     * ::hipInit() or that initialization has failed.
-     */
-    HIP_ERROR_NOT_INITIALIZED                = 3,
-
-    /**
-     * This indicates that the HIP driver is in the process of shutting down.
-     */
-    HIP_ERROR_DEINITIALIZED                  = 4,
-
-    /**
-     * This indicates profiler is not initialized for this run. This can
-     * happen when the application is running with external profiling tools
-     * like visual profiler.
-     */
-    HIP_ERROR_PROFILER_DISABLED              = 5,
-
-    /**
-     * \deprecated
-     * This error return is deprecated as of HIP 5.0. It is no longer an error
-     * to attempt to enable/disable the profiling via ::hipProfilerStart or
-     * ::hipProfilerStop without initialization.
-     */
-    HIP_ERROR_PROFILER_NOT_INITIALIZED       = 6,
-
-    /**
-     * \deprecated
-     * This error return is deprecated as of HIP 5.0. It is no longer an error
-     * to call hipProfilerStart() when profiling is already enabled.
-     */
-    HIP_ERROR_PROFILER_ALREADY_STARTED       = 7,
-
-    /**
-     * \deprecated
-     * This error return is deprecated as of HIP 5.0. It is no longer an error
-     * to call hipProfilerStop() when profiling is already disabled.
-     */
-    HIP_ERROR_PROFILER_ALREADY_STOPPED       = 8,
-
-    /**
-     * This indicates that no HIP-capable devices were detected by the installed
-     * HIP driver.
-     */
-    HIP_ERROR_NO_DEVICE                      = 100,
-
-    /**
-     * This indicates that the device ordinal supplied by the user does not
-     * correspond to a valid HIP device.
-     */
-    HIP_ERROR_INVALID_DEVICE                 = 101,
-
-
-    /**
-     * This indicates that the device kernel image is invalid. This can also
-     * indicate an invalid HIP module.
-     */
-    HIP_ERROR_INVALID_IMAGE                  = 200,
-
-    /**
-     * This most frequently indicates that there is no context bound to the
-     * hiprrent thread. This can also be returned if the context passed to an
-     * API call is not a valid handle (such as a context that has had
-     * ::hipCtxDestroy() invoked on it). This can also be returned if a user
-     * mixes different API versions (i.e. 3010 context with 3020 API calls).
-     * See ::hipCtxGetApiVersion() for more details.
-     */
-    HIP_ERROR_INVALID_CONTEXT                = 201,
-
-    /**
-     * This indicated that the context being supplied as a parameter to the
-     * API call was already the active context.
-     * \deprecated
-     * This error return is deprecated as of HIP 3.2. It is no longer an
-     * error to attempt to push the active context via ::hipCtxPushCurrent().
-     */
-    HIP_ERROR_CONTEXT_ALREADY_CURRENT        = 202,
-
-    /**
-     * This indicates that a map or register operation has failed.
-     */
-    HIP_ERROR_MAP_FAILED                     = 205,
-
-    /**
-     * This indicates that an unmap or unregister operation has failed.
-     */
-    HIP_ERROR_UNMAP_FAILED                   = 206,
-
-    /**
-     * This indicates that the specified array is currently mapped and thus
-     * cannot be destroyed.
-     */
-    HIP_ERROR_ARRAY_IS_MAPPED                = 207,
-
-    /**
-     * This indicates that the resource is already mapped.
-     */
-    HIP_ERROR_ALREADY_MAPPED                 = 208,
-
-    /**
-     * This indicates that there is no kernel image available that is suitable
-     * for the device. This can occur when a user specifies code generation
-     * options for a particular HIP source file that do not include the
-     * corresponding device configuration.
-     */
-    HIP_ERROR_NO_BINARY_FOR_GPU              = 209,
-
-    /**
-     * This indicates that a resource has already been acquired.
-     */
-    HIP_ERROR_ALREADY_ACQUIRED               = 210,
-
-    /**
-     * This indicates that a resource is not mapped.
-     */
-    HIP_ERROR_NOT_MAPPED                     = 211,
-
-    /**
-     * This indicates that a mapped resource is not available for access as an
-     * array.
-     */
-    HIP_ERROR_NOT_MAPPED_AS_ARRAY            = 212,
-
-    /**
-     * This indicates that a mapped resource is not available for access as a
-     * pointer.
-     */
-    HIP_ERROR_NOT_MAPPED_AS_POINTER          = 213,
-
-    /**
-     * This indicates that an uncorrectable ECC error was detected during
-     * execution.
-     */
-    HIP_ERROR_ECC_UNCORRECTABLE              = 214,
-
-    /**
-     * This indicates that the ::HIPlimit passed to the API call is not
-     * supported by the active device.
-     */
-    HIP_ERROR_UNSUPPORTED_LIMIT              = 215,
-
-    /**
-     * This indicates that the ::HIPcontext passed to the API call can
-     * only be bound to a single CPU thread at a time but is already
-     * bound to a CPU thread.
-     */
-    HIP_ERROR_CONTEXT_ALREADY_IN_USE         = 216,
-
-    /**
-     * This indicates that peer access is not supported across the given
-     * devices.
-     */
-    HIP_ERROR_PEER_ACCESS_UNSUPPORTED        = 217,
-
-    /**
-     * This indicates that a PTX JIT compilation failed.
-     */
-    HIP_ERROR_INVALID_PTX                    = 218,
-
-    /**
-     * This indicates an error with OpenGL or DirectX context.
-     */
-    HIP_ERROR_INVALID_GRAPHICS_CONTEXT       = 219,
-
-    /**
-    * This indicates that an uncorrectable NVLink error was detected during the
-    * execution.
-    */
-    HIP_ERROR_NVLINK_UNCORRECTABLE           = 220,
-
-    /**
-    * This indicates that the PTX JIT compiler library was not found.
-    */
-    HIP_ERROR_JIT_COMPILER_NOT_FOUND         = 221,
-
-    /**
-     * This indicates that the device kernel source is invalid.
-     */
-    HIP_ERROR_INVALID_SOURCE                 = 300,
-
-    /**
-     * This indicates that the file specified was not found.
-     */
-    HIP_ERROR_FILE_NOT_FOUND                 = 301,
-
-    /**
-     * This indicates that a link to a shared object failed to resolve.
-     */
-    HIP_ERROR_SHARED_OBJECT_SYMBOL_NOT_FOUND = 302,
-
-    /**
-     * This indicates that initialization of a shared object failed.
-     */
-    HIP_ERROR_SHARED_OBJECT_INIT_FAILED      = 303,
-
-    /**
-     * This indicates that an OS call failed.
-     */
-    HIP_ERROR_OPERATING_SYSTEM               = 304,
-
-    /**
-     * This indicates that a resource handle passed to the API call was not
-     * valid. Resource handles are opaque types like ::HIPstream and ::HIPevent.
-     */
-    HIP_ERROR_INVALID_HANDLE                 = 400,
-
-    /**
-     * This indicates that a resource required by the API call is not in a
-     * valid state to perform the requested operation.
-     */
-    HIP_ERROR_ILLEGAL_STATE                  = 401,
-
-    /**
-     * This indicates that a named symbol was not found. Examples of symbols
-     * are global/constant variable names, texture names, and surface names.
-     */
-    HIP_ERROR_NOT_FOUND                      = 500,
-
-    /**
-     * This indicates that asynchronous operations issued previously have not
-     * completed yet. This result is not actually an error, but must be indicated
-     * differently than ::HIP_SUCCESS (which indicates completion). Calls that
-     * may return this value include ::hipEventQuery() and ::hipStreamQuery().
-     */
-    HIP_ERROR_NOT_READY                      = 600,
-
-    /**
-     * While executing a kernel, the device encountered a
-     * load or store instruction on an invalid memory address.
-     * This leaves the process in an inconsistent state and any further HIP work
-     * will return the same error. To continue using HIP, the process must be terminated
-     * and relaunched.
-     */
-    HIP_ERROR_ILLEGAL_ADDRESS                = 700,
-
-    /**
-     * This indicates that a launch did not occur because it did not have
-     * appropriate resources. This error usually indicates that the user has
-     * attempted to pass too many arguments to the device kernel, or the
-     * kernel launch specifies too many threads for the kernel's register
-     * count. Passing arguments of the wrong size (i.e. a 64-bit pointer
-     * when a 32-bit int is expected) is equivalent to passing too many
-     * arguments and can also result in this error.
-     */
-    HIP_ERROR_LAUNCH_OUT_OF_RESOURCES        = 701,
-
-    /**
-     * This indicates that the device kernel took too long to execute. This can
-     * only occur if timeouts are enabled - see the device attribute
-     * ::HIP_DEVICE_ATTRIBUTE_KERNEL_EXEC_TIMEOUT for more information.
-     * This leaves the process in an inconsistent state and any further HIP work
-     * will return the same error. To continue using HIP, the process must be terminated
-     * and relaunched.
-     */
-    HIP_ERROR_LAUNCH_TIMEOUT                 = 702,
-
-    /**
-     * This error indicates a kernel launch that uses an incompatible texturing
-     * mode.
-     */
-    HIP_ERROR_LAUNCH_INCOMPATIBLE_TEXTURING  = 703,
-
-    /**
-     * This error indicates that a call to ::hipCtxEnablePeerAccess() is
-     * trying to re-enable peer access to a context which has already
-     * had peer access to it enabled.
-     */
-    HIP_ERROR_PEER_ACCESS_ALREADY_ENABLED    = 704,
-
-    /**
-     * This error indicates that ::hipCtxDisablePeerAccess() is
-     * trying to disable peer access which has not been enabled yet
-     * via ::hipCtxEnablePeerAccess().
-     */
-    HIP_ERROR_PEER_ACCESS_NOT_ENABLED        = 705,
-
-    /**
-     * This error indicates that the primary context for the specified device
-     * has already been initialized.
-     */
-    HIP_ERROR_PRIMARY_CONTEXT_ACTIVE         = 708,
-
-    /**
-     * This error indicates that the context hiprrent to the calling thread
-     * has been destroyed using ::hipCtxDestroy, or is a primary context which
-     * has not yet been initialized.
-     */
-    HIP_ERROR_CONTEXT_IS_DESTROYED           = 709,
-
-    /**
-     * A device-side assert triggered during kernel execution. The context
-     * cannot be used anymore, and must be destroyed. All existing device
-     * memory allocations from this context are invalid and must be
-     * reconstructed if the program is to continue using HIP.
-     */
-    HIP_ERROR_ASSERT                         = 710,
-
-    /**
-     * This error indicates that the hardware resources required to enable
-     * peer access have been exhausted for one or more of the devices
-     * passed to ::hipCtxEnablePeerAccess().
-     */
-    HIP_ERROR_TOO_MANY_PEERS                 = 711,
-
-    /**
-     * This error indicates that the memory range passed to ::hipMemHostRegister()
-     * has already been registered.
-     */
-    HIP_ERROR_HOST_MEMORY_ALREADY_REGISTERED = 712,
-
-    /**
-     * This error indicates that the pointer passed to ::hipMemHostUnregister()
-     * does not correspond to any currently registered memory region.
-     */
-    HIP_ERROR_HOST_MEMORY_NOT_REGISTERED     = 713,
-
-    /**
-     * While executing a kernel, the device encountered a stack error.
-     * This can be due to stack corruption or exceeding the stack size limit.
-     * This leaves the process in an inconsistent state and any further HIP work
-     * will return the same error. To continue using HIP, the process must be terminated
-     * and relaunched.
-     */
-    HIP_ERROR_HARDWARE_STACK_ERROR           = 714,
-
-    /**
-     * While executing a kernel, the device encountered an illegal instruction.
-     * This leaves the process in an inconsistent state and any further HIP work
-     * will return the same error. To continue using HIP, the process must be terminated
-     * and relaunched.
-     */
-    HIP_ERROR_ILLEGAL_INSTRUCTION            = 715,
-
-    /**
-     * While executing a kernel, the device encountered a load or store instruction
-     * on a memory address which is not aligned.
-     * This leaves the process in an inconsistent state and any further HIP work
-     * will return the same error. To continue using HIP, the process must be terminated
-     * and relaunched.
-     */
-    HIP_ERROR_MISALIGNED_ADDRESS             = 716,
-
-    /**
-     * While executing a kernel, the device encountered an instruction
-     * which can only operate on memory locations in certain address spaces
-     * (global, shared, or local), but was supplied a memory address not
-     * belonging to an allowed address space.
-     * This leaves the process in an inconsistent state and any further HIP work
-     * will return the same error. To continue using HIP, the process must be terminated
-     * and relaunched.
-     */
-    HIP_ERROR_INVALID_ADDRESS_SPACE          = 717,
-
-    /**
-     * While executing a kernel, the device program counter wrapped its address space.
-     * This leaves the process in an inconsistent state and any further HIP work
-     * will return the same error. To continue using HIP, the process must be terminated
-     * and relaunched.
-     */
-    HIP_ERROR_INVALID_PC                     = 718,
-
-    /**
-     * An exception occurred on the device while executing a kernel. Common
-     * causes include dereferencing an invalid device pointer and accessing
-     * out of bounds shared memory. Less common cases can be system specific - more
-     * information about these cases can be found in the system specific user guide.
-     * This leaves the process in an inconsistent state and any further HIP work
-     * will return the same error. To continue using HIP, the process must be terminated
-     * and relaunched.
-     */
-    HIP_ERROR_LAUNCH_FAILED                  = 719,
-
-    /**
-     * This error indicates that the number of blocks launched per grid for a kernel that was
-     * launched via either ::hipLaunchCooperativeKernel or ::hipLaunchCooperativeKernelMultiDevice
-     * exceeds the maximum number of blocks as allowed by ::hipOccupancyMaxActiveBlocksPerMultiprocessor
-     * or ::hipOccupancyMaxActiveBlocksPerMultiprocessorWithFlags times the number of multiprocessors
-     * as specified by the device attribute ::HIP_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT.
-     */
-    HIP_ERROR_COOPERATIVE_LAUNCH_TOO_LARGE   = 720,
-
-    /**
-     * This error indicates that the attempted operation is not permitted.
-     */
-    HIP_ERROR_NOT_PERMITTED                  = 800,
-
-    /**
-     * This error indicates that the attempted operation is not supported
-     * on the current system or device.
-     */
-    HIP_ERROR_NOT_SUPPORTED                  = 801,
-
-    /**
-     * This error indicates that the system is not yet ready to start any HIP
-     * work.  To continue using HIP, verify the system configuration is in a
-     * valid state and all required driver daemons are actively running.
-     * More information about this error can be found in the system specific
-     * user guide.
-     */
-    HIP_ERROR_SYSTEM_NOT_READY               = 802,
-
-    /**
-     * This error indicates that there is a mismatch between the versions of
-     * the display driver and the HIP driver. Refer to the compatibility documentation
-     * for supported versions.
-     */
-    HIP_ERROR_SYSTEM_DRIVER_MISMATCH         = 803,
-
-    /**
-     * This error indicates that the system was upgraded to run with forward compatibility
-     * but the visible hardware detected by HIP does not support this configuration.
-     * Refer to the compatibility documentation for the supported hardware matrix or ensure
-     * that only supported hardware is visible during initialization via the HIP_VISIBLE_DEVICES
-     * environment variable.
-     */
-    HIP_ERROR_COMPAT_NOT_SUPPORTED_ON_DEVICE = 804,
-
-    /**
-     * This error indicates that the operation is not permitted when
-     * the stream is capturing.
-     */
-    HIP_ERROR_STREAM_CAPTURE_UNSUPPORTED     = 900,
-
-    /**
-     * This error indicates that the current capture sequence on the stream
-     * has been invalidated due to a previous error.
-     */
-    HIP_ERROR_STREAM_CAPTURE_INVALIDATED     = 901,
-
-    /**
-     * This error indicates that the operation would have resulted in a merge
-     * of two independent capture sequences.
-     */
-    HIP_ERROR_STREAM_CAPTURE_MERGE           = 902,
-
-    /**
-     * This error indicates that the capture was not initiated in this stream.
-     */
-    HIP_ERROR_STREAM_CAPTURE_UNMATCHED       = 903,
-
-    /**
-     * This error indicates that the capture sequence contains a fork that was
-     * not joined to the primary stream.
-     */
-    HIP_ERROR_STREAM_CAPTURE_UNJOINED        = 904,
-
-    /**
-     * This error indicates that a dependency would have been created which
-     * crosses the capture sequence boundary. Only implicit in-stream ordering
-     * dependencies are allowed to cross the boundary.
-     */
-    HIP_ERROR_STREAM_CAPTURE_ISOLATION       = 905,
-
-    /**
-     * This error indicates a disallowed implicit dependency on a current capture
-     * sequence from HIPStreamLegacy.
-     */
-    HIP_ERROR_STREAM_CAPTURE_IMPLICIT        = 906,
-
-    /**
-     * This error indicates that the operation is not permitted on an event which
-     * was last recorded in a capturing stream.
-     */
-    HIP_ERROR_CAPTURED_EVENT                 = 907,
-
-    /**
-     * A stream capture sequence not initiated with the ::HIP_STREAM_CAPTURE_MODE_RELAXED
-     * argument to ::HIPStreamBeginCapture was passed to ::hipStreamEndCapture in a
-     * different thread.
-     */
-    HIP_ERROR_STREAM_CAPTURE_WRONG_THREAD    = 908,
-
-    /**
-     * This indicates that an unknown internal error has occurred.
-     */
-    HIP_ERROR_UNKNOWN                        = 999
-} HIPresult;
-
-/**
- * Online compiler and linker options
- */
-typedef enum HIPjit_option_enum
-{
-    /**
-     * Max number of registers that a thread may use.\n
-     * Option type: unsigned int\n
-     * Applies to: compiler only
-     */
-    HIP_JIT_MAX_REGISTERS = 0,
-
-    /**
-     * IN: Specifies minimum number of threads per block to target compilation
-     * for\n
-     * OUT: Returns the number of threads the compiler actually targeted.
-     * This restricts the resource utilization fo the compiler (e.g. max
-     * registers) such that a block with the given number of threads should be
-     * able to launch based on register limitations. Note, this option does not
-     * currently take into account any other resource limitations, such as
-     * shared memory utilization.\n
-     * Cannot be combined with ::HIP_JIT_TARGET.\n
-     * Option type: unsigned int\n
-     * Applies to: compiler only
-     */
-    HIP_JIT_THREADS_PER_BLOCK,
-
-    /**
-     * Overwrites the option value with the total wall clock time, in
-     * milliseconds, spent in the compiler and linker\n
-     * Option type: float\n
-     * Applies to: compiler and linker
-     */
-    HIP_JIT_WALL_TIME,
-
-    /**
-     * Pointer to a buffer in which to print any log messages
-     * that are informational in nature (the buffer size is specified via
-     * option ::HIP_JIT_INFO_LOG_BUFFER_SIZE_BYTES)\n
-     * Option type: char *\n
-     * Applies to: compiler and linker
-     */
-    HIP_JIT_INFO_LOG_BUFFER,
-
-    /**
-     * IN: Log buffer size in bytes.  Log messages will be capped at this size
-     * (including null terminator)\n
-     * OUT: Amount of log buffer filled with messages\n
-     * Option type: unsigned int\n
-     * Applies to: compiler and linker
-     */
-    HIP_JIT_INFO_LOG_BUFFER_SIZE_BYTES,
-
-    /**
-     * Pointer to a buffer in which to print any log messages that
-     * reflect errors (the buffer size is specified via option
-     * ::HIP_JIT_ERROR_LOG_BUFFER_SIZE_BYTES)\n
-     * Option type: char *\n
-     * Applies to: compiler and linker
-     */
-    HIP_JIT_ERROR_LOG_BUFFER,
-
-    /**
-     * IN: Log buffer size in bytes.  Log messages will be capped at this size
-     * (including null terminator)\n
-     * OUT: Amount of log buffer filled with messages\n
-     * Option type: unsigned int\n
-     * Applies to: compiler and linker
-     */
-    HIP_JIT_ERROR_LOG_BUFFER_SIZE_BYTES,
-
-    /**
-     * Level of optimizations to apply to generated code (0 - 4), with 4
-     * being the default and highest level of optimizations.\n
-     * Option type: unsigned int\n
-     * Applies to: compiler only
-     */
-    HIP_JIT_OPTIMIZATION_LEVEL,
-
-    /**
-     * No option value required. Determines the target based on the current
-     * attached context (default)\n
-     * Option type: No option value needed\n
-     * Applies to: compiler and linker
-     */
-    HIP_JIT_TARGET_FROM_HIPCONTEXT,
-
-    /**
-     * Target is chosen based on supplied ::HIPjit_target.  Cannot be
-     * combined with ::HIP_JIT_THREADS_PER_BLOCK.\n
-     * Option type: unsigned int for enumerated type ::HIPjit_target\n
-     * Applies to: compiler and linker
-     */
-    HIP_JIT_TARGET,
-
-    /**
-     * Specifies choice of fallback strategy if matching HIPbin is not found.
-     * Choice is based on supplied ::HIPjit_fallback.  This option cannot be
-     * used with HIPLink* APIs as the linker requires exact matches.\n
-     * Option type: unsigned int for enumerated type ::HIPjit_fallback\n
-     * Applies to: compiler only
-     */
-    HIP_JIT_FALLBACK_STRATEGY,
-
-    /**
-     * Specifies whether to create debug information in output (-g)
-     * (0: false, default)\n
-     * Option type: int\n
-     * Applies to: compiler and linker
-     */
-    HIP_JIT_GENERATE_DEBUG_INFO,
-
-    /**
-     * Generate verbose log messages (0: false, default)\n
-     * Option type: int\n
-     * Applies to: compiler and linker
-     */
-    HIP_JIT_LOG_VERBOSE,
-
-    /**
-     * Generate line number information (-lineinfo) (0: false, default)\n
-     * Option type: int\n
-     * Applies to: compiler only
-     */
-    HIP_JIT_GENERATE_LINE_INFO,
-
-    /**
-     * Specifies whether to enable caching explicitly (-dlcm) \n
-     * Choice is based on supplied ::HIPjit_cacheMode_enum.\n
-     * Option type: unsigned int for enumerated type ::HIPjit_cacheMode_enum\n
-     * Applies to: compiler only
-     */
-    HIP_JIT_CACHE_MODE,
-
-    /**
-     * The below jit options are used for internal purposes only, in this version of HIP
-     */
-    HIP_JIT_NEW_SM3X_OPT,
-    HIP_JIT_FAST_COMPILE,
-
-    /**
-     * Array of device symbol names that will be relocated to the corresponing
-     * host addresses stored in ::HIP_JIT_GLOBAL_SYMBOL_ADDRESSES.\n
-     * Must contain ::HIP_JIT_GLOBAL_SYMBOL_COUNT entries.\n
-     * When loding a device module, driver will relocate all encountered
-     * unresolved symbols to the host addresses.\n
-     * It is only allowed to register symbols that correspond to unresolved
-     * global variables.\n
-     * It is illegal to register the same device symbol at multiple addresses.\n
-     * Option type: const char **\n
-     * Applies to: dynamic linker only
-     */
-    HIP_JIT_GLOBAL_SYMBOL_NAMES,
-
-    /**
-     * Array of host addresses that will be used to relocate corresponding
-     * device symbols stored in ::HIP_JIT_GLOBAL_SYMBOL_NAMES.\n
-     * Must contain ::HIP_JIT_GLOBAL_SYMBOL_COUNT entries.\n
-     * Option type: void **\n
-     * Applies to: dynamic linker only
-     */
-    HIP_JIT_GLOBAL_SYMBOL_ADDRESSES,
-
-    /**
-     * Number of entries in ::HIP_JIT_GLOBAL_SYMBOL_NAMES and
-     * ::HIP_JIT_GLOBAL_SYMBOL_ADDRESSES arrays.\n
-     * Option type: unsigned int\n
-     * Applies to: dynamic linker only
-     */
-    HIP_JIT_GLOBAL_SYMBOL_COUNT,
-
-    HIP_JIT_NUM_OPTIONS
-
-} HIPjit_option;
-
-
-/**
- * Device properties
- */
-typedef enum HIPdevice_attribute_enum {
-    
-    HIP_DEVICE_ATTRIBUTE_MAX_THREADS_PER_BLOCK = 0,              /**< Maximum number of threads per block */
-    HIP_DEVICE_ATTRIBUTE_MAX_BLOCK_DIM_X = 1,                    /**< Maximum block dimension X */
-    HIP_DEVICE_ATTRIBUTE_MAX_BLOCK_DIM_Y = 2,                    /**< Maximum block dimension Y */
-    HIP_DEVICE_ATTRIBUTE_MAX_BLOCK_DIM_Z = 3,                    /**< Maximum block dimension Z */
-    HIP_DEVICE_ATTRIBUTE_MAX_GRID_DIM_X = 4,                     /**< Maximum grid dimension X */
-    HIP_DEVICE_ATTRIBUTE_MAX_GRID_DIM_Y = 5,                     /**< Maximum grid dimension Y */
-    HIP_DEVICE_ATTRIBUTE_MAX_GRID_DIM_Z = 6,                     /**< Maximum grid dimension Z */
-    HIP_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK = 7,        /**< Maximum shared memory available per block in bytes */
-    HIP_DEVICE_ATTRIBUTE_SHARED_MEMORY_PER_BLOCK = 7,            /**< Deprecated, use HIP_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK */
-    HIP_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK_OPTIN = 7, /**< Maximum optin shared memory per block */
-    HIP_DEVICE_ATTRIBUTE_TOTAL_CONSTANT_MEMORY = 8,              /**< Memory available on device for __constant__ variables in a HIP C kernel in bytes */
-    HIP_DEVICE_ATTRIBUTE_WARP_SIZE = 9,                         /**< Warp size in threads */
-    HIP_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_BLOCK = 10,           /**< Maximum number of 32-bit registers available per block */
-    HIP_DEVICE_ATTRIBUTE_REGISTERS_PER_BLOCK = 10,               /**< Deprecated, use HIP_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_BLOCK */
-    HIP_DEVICE_ATTRIBUTE_CLOCK_RATE = 11,                        /**< Typical clock frequency in kilohertz */
-    HIP_DEVICE_ATTRIBUTE_MEMORY_CLOCK_RATE = 12,                 /**< Peak memory clock frequency in kilohertz */
-    HIP_DEVICE_ATTRIBUTE_GLOBAL_MEMORY_BUS_WIDTH = 13,           /**< Global memory bus width in bits */
-    HIP_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT = 14,              /**< Number of multiprocessors on device */
-    HIP_DEVICE_ATTRIBUTE_COMPUTE_MODE = 15,                      /**< Compute mode (See ::HIPcomputemode for details) */
-    HIP_DEVICE_ATTRIBUTE_L2_CACHE_SIZE = 16,                     /**< Size of L2 cache in bytes */
-    HIP_DEVICE_ATTRIBUTE_MAX_THREADS_PER_MULTIPROCESSOR = 17,    /**< Maximum resident threads per multiprocessor */
-    HIP_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR = 18,          /**< Major compute capability version number */
-    HIP_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR = 19,          /**< Minor compute capability version number */
-    HIP_DEVICE_ATTRIBUTE_CONCURRENT_KERNELS = 20,                /**< Device can possibly execute multiple kernels concurrently */
-    HIP_DEVICE_ATTRIBUTE_PCI_BUS_ID = 21,                        /**< PCI bus ID of the device */
-    HIP_DEVICE_ATTRIBUTE_PCI_DEVICE_ID = 22,                     /**< PCI device ID of the device */
-    HIP_DEVICE_ATTRIBUTE_PCI_DOMAIN_ID = 22,                     /**< PCI domain ID of the device */
-    HIP_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_MULTIPROCESSOR = 23,  /**< Maximum shared memory available per multiprocessor in bytes */
-    HIP_DEVICE_ATTRIBUTE_MULTI_GPU_BOARD = 24,                    /**< Device is on a multi-GPU board */
-    HIP_DEVICE_ATTRIBUTE_INTEGRATED = 25,                        /**< Device is integrated with host memory */
-    HIP_DEVICE_ATTRIBUTE_COOPERATIVE_LAUNCH = 26,                /**< Device supports launching cooperative kernels via ::hipLaunchCooperativeKernel */
-    HIP_DEVICE_ATTRIBUTE_COOPERATIVE_MULTI_DEVICE_LAUNCH = 27,   /**< Device can participate in cooperative kernels launched via ::hipLaunchCooperativeKernelMultiDevice */
-    HIP_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_WIDTH = 28,           /**< Maximum 1D texture width */
-    HIP_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_WIDTH = 29,           /**< Maximum 2D texture width */
-    HIP_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_HEIGHT = 30,          /**< Maximum 2D texture height */
-    HIP_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_WIDTH = 31,           /**< Maximum 3D texture width */
-    HIP_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_HEIGHT = 32,          /**< Maximum 3D texture height */
-    HIP_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_DEPTH = 33,           /**< Maximum 3D texture depth */
-    
-    HIP_DEVICE_ATTRIBUTE_TEXTURE_ALIGNMENT = 37,                 /**< Alignment requirement for textures */
-    HIP_DEVICE_ATTRIBUTE_TEXTURE_PITCH_ALIGNMENT = 38,           /**< Pitch alignment requirement for textures */
-    HIP_DEVICE_ATTRIBUTE_KERNEL_EXEC_TIMEOUT = 39,               /**< Specifies whether there is a run time limit on kernels */
-    HIP_DEVICE_ATTRIBUTE_CAN_MAP_HOST_MEMORY = 40,               /**< Device can map host memory into HIP address space */
-    HIP_DEVICE_ATTRIBUTE_ECC_ENABLED = 41,                       /**< Device has ECC support enabled */
-    
-    HIP_DEVICE_ATTRIBUTE_MANAGED_MEMORY = 47,                    /**< Device can allocate managed memory on this system */
-    HIP_DEVICE_ATTRIBUTE_DIRECT_MANAGED_MEM_ACCESS_FROM_HOST = 48, /**< The host can directly access managed memory on the device without migration. */
-    HIP_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS = 49,         /**< Device can coherently access managed memory concurrently with the CPU */
-    HIP_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS = 50,            /**< Device supports coherently accessing pageable memory without calling HIPHostRegister on it */
-    HIP_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS_USES_HOST_PAGE_TABLES = 51, /**< Device accesses pageable memory via the host's page tables. */
-    HIP_DEVICE_ATTRIBUTE_CAN_USE_STREAM_WAIT_VALUE_NOR = 52,     /**< ::HIP_STREAM_WAIT_VALUE_NOR is supported. */
-    
-    
-    // HIP_DEVICE_ATTRIBUTE_MAX_PITCH = ,                         /**< Maximum pitch in bytes allowed by memory copies */
-    // HIP_DEVICE_ATTRIBUTE_GPU_OVERLAP = ,                       /**< Device can possibly copy memory and execute a kernel concurrently. Deprecated. Use instead HIP_DEVICE_ATTRIBUTE_ASYNC_ENGINE_COUNT. */
-    // 
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_WIDTH = ,   /**< Maximum 2D layered texture width */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_HEIGHT = ,  /**< Maximum 2D layered texture height */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_LAYERS = ,  /**< Maximum layers in a 2D layered texture */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_ARRAY_WIDTH = ,     /**< Deprecated, use HIP_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_WIDTH */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_ARRAY_HEIGHT = ,    /**< Deprecated, use HIP_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_HEIGHT */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_ARRAY_NUMSLICES = , /**< Deprecated, use HIP_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_LAYERS */
-    // HIP_DEVICE_ATTRIBUTE_SURFACE_ALIGNMENT =,                 /**< Alignment requirement for surfaces */
-    // HIP_DEVICE_ATTRIBUTE_TCC_DRIVER = ,                        /**< Device is using TCC driver model */
-    // HIP_DEVICE_ATTRIBUTE_ASYNC_ENGINE_COUNT = ,                /**< Number of asynchronous engines */
-    // HIP_DEVICE_ATTRIBUTE_UNIFIED_ADDRESSING = ,                /**< Device shares a unified address space with the host */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_LAYERED_WIDTH = ,   /**< Maximum 1D layered texture width */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_LAYERED_LAYERS = ,  /**< Maximum layers in a 1D layered texture */
-    // HIP_DEVICE_ATTRIBUTE_CAN_TEX2D_GATHER = ,                  /**< Deprecated, do not use. */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_GATHER_WIDTH = ,    /**< Maximum 2D texture width if HIP_ARRAY3D_TEXTURE_GATHER is set */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_GATHER_HEIGHT = ,   /**< Maximum 2D texture height if HIP_ARRAY3D_TEXTURE_GATHER is set */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_WIDTH_ALTERNATE = , /**< Alternate maximum 3D texture width */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_HEIGHT_ALTERNATE = ,/**< Alternate maximum 3D texture height */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_DEPTH_ALTERNATE = , /**< Alternate maximum 3D texture depth */
-    // 
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURECUBEMAP_WIDTH = ,      /**< Maximum cubemap texture width/height */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURECUBEMAP_LAYERED_WIDTH = ,  /**< Maximum cubemap layered texture width/height */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURECUBEMAP_LAYERED_LAYERS = , /**< Maximum layers in a cubemap layered texture */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE1D_WIDTH = ,           /**< Maximum 1D surface width */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE2D_WIDTH = ,           /**< Maximum 2D surface width */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE2D_HEIGHT = ,          /**< Maximum 2D surface height */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE3D_WIDTH = ,           /**< Maximum 3D surface width */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE3D_HEIGHT = ,          /**< Maximum 3D surface height */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE3D_DEPTH = ,           /**< Maximum 3D surface depth */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE1D_LAYERED_WIDTH = ,   /**< Maximum 1D layered surface width */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE1D_LAYERED_LAYERS = ,  /**< Maximum layers in a 1D layered surface */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE2D_LAYERED_WIDTH = ,   /**< Maximum 2D layered surface width */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE2D_LAYERED_HEIGHT = ,  /**< Maximum 2D layered surface height */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE2D_LAYERED_LAYERS = ,  /**< Maximum layers in a 2D layered surface */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_SURFACECUBEMAP_WIDTH = ,      /**< Maximum cubemap surface width */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_SURFACECUBEMAP_LAYERED_WIDTH = ,  /**< Maximum cubemap layered surface width */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_SURFACECUBEMAP_LAYERED_LAYERS = , /**< Maximum layers in a cubemap layered surface */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_LINEAR_WIDTH = ,    /**< Maximum 1D linear texture width */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LINEAR_WIDTH = ,    /**< Maximum 2D linear texture width */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LINEAR_HEIGHT = ,   /**< Maximum 2D linear texture height */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LINEAR_PITCH = ,    /**< Maximum 2D linear texture pitch in bytes */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_MIPMAPPED_WIDTH = , /**< Maximum mipmapped 2D texture width */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_MIPMAPPED_HEIGHT = ,/**< Maximum mipmapped 2D texture height */
-    // HIP_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_MIPMAPPED_WIDTH = , /**< Maximum mipmapped 1D texture width */
-    // HIP_DEVICE_ATTRIBUTE_STREAM_PRIORITIES_SUPPORTED = ,       /**< Device supports stream priorities */
-    // HIP_DEVICE_ATTRIBUTE_GLOBAL_L1_CACHE_SUPPORTED = ,         /**< Device supports caching globals in L1 */
-    // HIP_DEVICE_ATTRIBUTE_LOCAL_L1_CACHE_SUPPORTED = ,          /**< Device supports caching locals in L1 */
-    // HIP_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_MULTIPROCESSOR = ,  /**< Maximum number of 32-bit registers available per multiprocessor */
-    // HIP_DEVICE_ATTRIBUTE_MULTI_GPU_BOARD_GROUP_ID = ,           /**< Unique id for a group of devices on the same multi-GPU board */
-    // HIP_DEVICE_ATTRIBUTE_HOST_NATIVE_ATOMIC_SUPPORTED = ,       /**< Link between the device and the host supports native atomic operations (this is a placeholder attribute, and is not supported on any current hardware)*/
-    // HIP_DEVICE_ATTRIBUTE_SINGLE_TO_DOUBLE_PRECISION_PERF_RATIO = ,  /**< Ratio of single precision performance (in floating-point operations per second) to double precision performance */
-    // HIP_DEVICE_ATTRIBUTE_COMPUTE_PREEMPTION_SUPPORTED = ,      /**< Device supports compute preemption. */
-    // HIP_DEVICE_ATTRIBUTE_CAN_USE_HOST_POINTER_FOR_REGISTERED_MEM = , /**< Device can access host registered memory at the same virtual address as the CPU */
-    // HIP_DEVICE_ATTRIBUTE_CAN_USE_STREAM_MEM_OPS = ,            /**< ::hipStreamBatchMemOp and related APIs are supported. */
-    // HIP_DEVICE_ATTRIBUTE_CAN_USE_64_BIT_STREAM_MEM_OPS = ,     /**< 64-bit operations are supported in ::hipStreamBatchMemOp and related APIs. */
-    // HIP_DEVICE_ATTRIBUTE_CAN_FLUSH_REMOTE_WRITES = ,           /**< Both the ::HIP_STREAM_WAIT_VALUE_FLUSH flag and the ::HIP_STREAM_MEM_OP_FLUSH_REMOTE_WRITES MemOp are supported on the device. See \ref HIP_MEMOP for additional details. */
-    // HIP_DEVICE_ATTRIBUTE_HOST_REGISTER_SUPPORTED = ,           /**< Device supports host memory registration via ::HIPHostRegister. */
-    // HIP_DEVICE_ATTRIBUTE_MAX
-} HIPdevice_attribute;
-
-/**
- * Function cache configurations
- */
-typedef enum HIPfunc_cache_enum {
-    HIP_FUNC_CACHE_PREFER_NONE    = 0x00, /**< no preference for shared memory or L1 (default) */
-    HIP_FUNC_CACHE_PREFER_SHARED  = 0x01, /**< prefer larger shared memory and smaller L1 cache */
-    HIP_FUNC_CACHE_PREFER_L1      = 0x02, /**< prefer larger L1 cache and smaller shared memory */
-    HIP_FUNC_CACHE_PREFER_EQUAL   = 0x03  /**< prefer equal sized L1 cache and shared memory */
-} HIPfunc_cache;
-
-/**
- * Shared memory configurations
- */
-typedef enum HIPsharedconfig_enum {
-    HIP_SHARED_MEM_CONFIG_DEFAULT_BANK_SIZE    = 0x00, /**< set default shared memory bank size */
-    HIP_SHARED_MEM_CONFIG_FOUR_BYTE_BANK_SIZE  = 0x01, /**< set shared memory bank width to four bytes */
-    HIP_SHARED_MEM_CONFIG_EIGHT_BYTE_BANK_SIZE = 0x02  /**< set shared memory bank width to eight bytes */
-} HIPsharedconfig;
-
-/**
- * Function properties
- */
-typedef enum HIPfunction_attribute_enum {
-    /**
-     * The maximum number of threads per block, beyond which a launch of the
-     * function would fail. This number depends on both the function and the
-     * device on which the function is currently loaded.
-     */
-    HIP_FUNC_ATTRIBUTE_MAX_THREADS_PER_BLOCK = 0,
-
-    /**
-     * The size in bytes of statically-allocated shared memory required by
-     * this function. This does not include dynamically-allocated shared
-     * memory requested by the user at runtime.
-     */
-    HIP_FUNC_ATTRIBUTE_SHARED_SIZE_BYTES = 1,
-
-    /**
-     * The size in bytes of user-allocated constant memory required by this
-     * function.
-     */
-    HIP_FUNC_ATTRIBUTE_CONST_SIZE_BYTES = 2,
-
-    /**
-     * The size in bytes of local memory used by each thread of this function.
-     */
-    HIP_FUNC_ATTRIBUTE_LOCAL_SIZE_BYTES = 3,
-
-    /**
-     * The number of registers used by each thread of this function.
-     */
-    HIP_FUNC_ATTRIBUTE_NUM_REGS = 4,
-
-    /**
-     * The PTX virtual architecture version for which the function was
-     * compiled. This value is the major PTX version * 10 + the minor PTX
-     * version, so a PTX version 1.3 function would return the value 13.
-     * Note that this may return the undefined value of 0 for cubins
-     * compiled prior to HIP 3.0.
-     */
-    HIP_FUNC_ATTRIBUTE_PTX_VERSION = 5,
-
-    /**
-     * The binary architecture version for which the function was compiled.
-     * This value is the major binary version * 10 + the minor binary version,
-     * so a binary version 1.3 function would return the value 13. Note that
-     * this will return a value of 10 for legacy cubins that do not have a
-     * properly-encoded binary architecture version.
-     */
-    HIP_FUNC_ATTRIBUTE_BINARY_VERSION = 6,
-
-    /**
-     * The attribute to indicate whether the function has been compiled with
-     * user specified option "-Xptxas --dlcm=ca" set .
-     */
-    HIP_FUNC_ATTRIBUTE_CACHE_MODE_CA = 7,
-
-    /**
-     * The maximum size in bytes of dynamically-allocated shared memory that can be used by
-     * this function. If the user-specified dynamic shared memory size is larger than this
-     * value, the launch will fail.
-     * See ::hipFuncSetAttribute
-     */
-    HIP_FUNC_ATTRIBUTE_MAX_DYNAMIC_SHARED_SIZE_BYTES = 8,
-
-    /**
-     * On devices where the L1 cache and shared memory use the same hardware resources,
-     * this sets the shared memory carveout preference, in percent of the total shared memory.
-     * Refer to ::HIP_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_MULTIPROCESSOR.
-     * This is only a hint, and the driver can choose a different ratio if required to execute the function.
-     * See ::hipFuncSetAttribute
-     */
-    HIP_FUNC_ATTRIBUTE_PREFERRED_SHARED_MEMORY_CARVEOUT = 9,
-
-    HIP_FUNC_ATTRIBUTE_MAX
-} HIPfunction_attribute;
-
-/**
- * Context creation flags
- */
-typedef enum HIPctx_flags_enum {
-    HIP_CTX_SCHED_AUTO          = 0x00, /**< Automatic scheduling */
-    HIP_CTX_SCHED_SPIN          = 0x01, /**< Set spin as default scheduling */
-    HIP_CTX_SCHED_YIELD         = 0x02, /**< Set yield as default scheduling */
-    HIP_CTX_SCHED_BLOCKING_SYNC = 0x04, /**< Set blocking synchronization as default scheduling */
-    HIP_CTX_BLOCKING_SYNC       = 0x04, /**< Set blocking synchronization as default scheduling
-                                         *  \deprecated This flag was deprecated as of HIP 4.0
-                                         *  and was replaced with ::HIP_CTX_SCHED_BLOCKING_SYNC. */
-    HIP_CTX_SCHED_MASK          = 0x07,
-    HIP_CTX_MAP_HOST            = 0x08, /**< Support mapped pinned allocations */
-    HIP_CTX_LMEM_RESIZE_TO_MAX  = 0x10, /**< Keep local memory allocation after launch */
-    HIP_CTX_FLAGS_MASK          = 0x1f
-} HIPctx_flags;
-
-/**
- * Stream creation flags
- */
-typedef enum HIPstream_flags_enum {
-    HIP_STREAM_DEFAULT      = 0x0, /**< Default stream flag */
-    HIP_STREAM_NON_BLOCKING = 0x1  /**< Stream does not synchronize with stream 0 (the NULL stream) */
-} HIPstream_flags;
-
-/**
- * Event creation flags
- */
-typedef enum HIPevent_flags_enum {
-    HIP_EVENT_DEFAULT        = 0x0, /**< Default event flag */
-    HIP_EVENT_BLOCKING_SYNC  = 0x1, /**< Event uses blocking synchronization */
-    HIP_EVENT_DISABLE_TIMING = 0x2, /**< Event will not record timing data */
-    HIP_EVENT_INTERPROCESS   = 0x4  /**< Event is suitable for interprocess use. HIP_EVENT_DISABLE_TIMING must be set */
-} HIPevent_flags;
-
-typedef enum HIPjitInputType_enum
-{
-    /**
-     * Compiled device-class-specific device code\n
-     * Applicable options: none
-     */
-    HIP_JIT_INPUT_HIPBIN = 0,
-
-    /**
-     * PTX source code\n
-     * Applicable options: PTX compiler options
-     */
-    HIP_JIT_INPUT_PTX,
-
-    /**
-     * Bundle of multiple cubins and/or PTX of some device code\n
-     * Applicable options: PTX compiler options, ::HIP_JIT_FALLBACK_STRATEGY
-     */
-    HIP_JIT_INPUT_FATBINARY,
-
-    /**
-     * Host object with embedded device code\n
-     * Applicable options: PTX compiler options, ::HIP_JIT_FALLBACK_STRATEGY
-     */
-    HIP_JIT_INPUT_OBJECT,
-
-    /**
-     * Archive of host objects with embedded device code\n
-     * Applicable options: PTX compiler options, ::HIP_JIT_FALLBACK_STRATEGY
-     */
-    HIP_JIT_INPUT_LIBRARY,
-
-    HIP_JIT_NUM_INPUT_TYPES
-} HIPjitInputType;
+typedef enum __HIP_NODISCARD hipError_t {
+    hipSuccess = 0,  ///< Successful completion.
+    hipErrorInvalidValue = 1,  ///< One or more of the parameters passed to the API call is NULL
+                               ///< or not in an acceptable range.
+    hipErrorOutOfMemory = 2,
+    // Deprecated
+    hipErrorMemoryAllocation = 2,  ///< Memory allocation error.
+    hipErrorNotInitialized = 3,
+    // Deprecated
+    hipErrorInitializationError = 3,
+    hipErrorDeinitialized = 4,
+    hipErrorProfilerDisabled = 5,
+    hipErrorProfilerNotInitialized = 6,
+    hipErrorProfilerAlreadyStarted = 7,
+    hipErrorProfilerAlreadyStopped = 8,
+    hipErrorInvalidConfiguration = 9,
+    hipErrorInvalidPitchValue = 12,
+    hipErrorInvalidSymbol = 13,
+    hipErrorInvalidDevicePointer = 17,  ///< Invalid Device Pointer
+    hipErrorInvalidMemcpyDirection = 21,  ///< Invalid memory copy direction
+    hipErrorInsufficientDriver = 35,
+    hipErrorMissingConfiguration = 52,
+    hipErrorPriorLaunchFailure = 53,
+    hipErrorInvalidDeviceFunction = 98,
+    hipErrorNoDevice = 100,  ///< Call to hipGetDeviceCount returned 0 devices
+    hipErrorInvalidDevice = 101,  ///< DeviceID must be in range 0...#compute-devices.
+    hipErrorInvalidImage = 200,
+    hipErrorInvalidContext = 201,  ///< Produced when input context is invalid.
+    hipErrorContextAlreadyCurrent = 202,
+    hipErrorMapFailed = 205,
+    // Deprecated
+    hipErrorMapBufferObjectFailed = 205,  ///< Produced when the IPC memory attach failed from ROCr.
+    hipErrorUnmapFailed = 206,
+    hipErrorArrayIsMapped = 207,
+    hipErrorAlreadyMapped = 208,
+    hipErrorNoBinaryForGpu = 209,
+    hipErrorAlreadyAcquired = 210,
+    hipErrorNotMapped = 211,
+    hipErrorNotMappedAsArray = 212,
+    hipErrorNotMappedAsPointer = 213,
+    hipErrorECCNotCorrectable = 214,
+    hipErrorUnsupportedLimit = 215,
+    hipErrorContextAlreadyInUse = 216,
+    hipErrorPeerAccessUnsupported = 217,
+    hipErrorInvalidKernelFile = 218,  ///< In CUDA DRV, it is CUDA_ERROR_INVALID_PTX
+    hipErrorInvalidGraphicsContext = 219,
+    hipErrorInvalidSource = 300,
+    hipErrorFileNotFound = 301,
+    hipErrorSharedObjectSymbolNotFound = 302,
+    hipErrorSharedObjectInitFailed = 303,
+    hipErrorOperatingSystem = 304,
+    hipErrorInvalidHandle = 400,
+    // Deprecated
+    hipErrorInvalidResourceHandle = 400,  ///< Resource handle (hipEvent_t or hipStream_t) invalid.
+    hipErrorNotFound = 500,
+    hipErrorNotReady = 600,  ///< Indicates that asynchronous operations enqueued earlier are not
+                             ///< ready.  This is not actually an error, but is used to distinguish
+                             ///< from hipSuccess (which indicates completion).  APIs that return
+                             ///< this error include hipEventQuery and hipStreamQuery.
+    hipErrorIllegalAddress = 700,
+    hipErrorLaunchOutOfResources = 701,  ///< Out of resources error.
+    hipErrorLaunchTimeOut = 702,
+    hipErrorPeerAccessAlreadyEnabled =
+        704,  ///< Peer access was already enabled from the current device.
+    hipErrorPeerAccessNotEnabled =
+        705,  ///< Peer access was never enabled from the current device.
+    hipErrorSetOnActiveProcess = 708,
+    hipErrorContextIsDestroyed = 709,
+    hipErrorAssert = 710,  ///< Produced when the kernel calls assert.
+    hipErrorHostMemoryAlreadyRegistered =
+        712,  ///< Produced when trying to lock a page-locked memory.
+    hipErrorHostMemoryNotRegistered =
+        713,  ///< Produced when trying to unlock a non-page-locked memory.
+    hipErrorLaunchFailure =
+        719,  ///< An exception occurred on the device while executing a kernel.
+    hipErrorCooperativeLaunchTooLarge =
+        720,  ///< This error indicates that the number of blocks launched per grid for a kernel
+              ///< that was launched via cooperative launch APIs exceeds the maximum number of
+              ///< allowed blocks for the current device
+    hipErrorNotSupported = 801,  ///< Produced when the hip API is not supported/implemented
+    hipErrorStreamCaptureUnsupported = 900,  ///< The operation is not permitted when the stream
+                                             ///< is capturing.
+    hipErrorStreamCaptureInvalidated = 901,  ///< The current capture sequence on the stream
+                                             ///< has been invalidated due to a previous error.
+    hipErrorStreamCaptureMerge = 902,  ///< The operation would have resulted in a merge of
+                                       ///< two independent capture sequences.
+    hipErrorStreamCaptureUnmatched = 903,  ///< The capture was not initiated in this stream.
+    hipErrorStreamCaptureUnjoined = 904,  ///< The capture sequence contains a fork that was not
+                                          ///< joined to the primary stream.
+    hipErrorStreamCaptureIsolation = 905,  ///< A dependency would have been created which crosses
+                                           ///< the capture sequence boundary. Only implicit
+                                           ///< in-stream ordering dependencies  are allowed
+                                           ///< to cross the boundary
+    hipErrorStreamCaptureImplicit = 906,  ///< The operation would have resulted in a disallowed
+                                          ///< implicit dependency on a current capture sequence
+                                          ///< from hipStreamLegacy.
+    hipErrorCapturedEvent = 907,  ///< The operation is not permitted on an event which was last
+                                  ///< recorded in a capturing stream.
+    hipErrorStreamCaptureWrongThread = 908,  ///< A stream capture sequence not initiated with
+                                             ///< the hipStreamCaptureModeRelaxed argument to
+                                             ///< hipStreamBeginCapture was passed to
+                                             ///< hipStreamEndCapture in a different thread.
+    hipErrorUnknown = 999,  //< Unknown error.
+    // HSA Runtime Error Codes start here.
+    hipErrorRuntimeMemory = 1052,  ///< HSA runtime memory call returned error.  Typically not seen
+                                   ///< in production systems.
+    hipErrorRuntimeOther = 1053,  ///< HSA runtime call other than memory returned error.  Typically
+                                  ///< not seen in production systems.
+    hipErrorTbd  ///< Marker that more error codes are needed.
+} hipError_t;
+
+#undef __HIP_NODISCARD
+
+typedef enum hipDeviceAttribute_t {
+    hipDeviceAttributeCudaCompatibleBegin = 0,
+
+    hipDeviceAttributeEccEnabled = hipDeviceAttributeCudaCompatibleBegin, ///< Whether ECC support is enabled.
+    hipDeviceAttributeAccessPolicyMaxWindowSize,        ///< Cuda only. The maximum size of the window policy in bytes.
+    hipDeviceAttributeAsyncEngineCount,                 ///< Cuda only. Asynchronous engines number.
+    hipDeviceAttributeCanMapHostMemory,                 ///< Whether host memory can be mapped into device address space
+    hipDeviceAttributeCanUseHostPointerForRegisteredMem,///< Cuda only. Device can access host registered memory
+                                                        ///< at the same virtual address as the CPU
+    hipDeviceAttributeClockRate,                        ///< Peak clock frequency in kilohertz.
+    hipDeviceAttributeComputeMode,                      ///< Compute mode that device is currently in.
+    hipDeviceAttributeComputePreemptionSupported,       ///< Cuda only. Device supports Compute Preemption.
+    hipDeviceAttributeConcurrentKernels,                ///< Device can possibly execute multiple kernels concurrently.
+    hipDeviceAttributeConcurrentManagedAccess,          ///< Device can coherently access managed memory concurrently with the CPU
+    hipDeviceAttributeCooperativeLaunch,                ///< Support cooperative launch
+    hipDeviceAttributeCooperativeMultiDeviceLaunch,     ///< Support cooperative launch on multiple devices
+    hipDeviceAttributeDeviceOverlap,                    ///< Cuda only. Device can concurrently copy memory and execute a kernel.
+                                                        ///< Deprecated. Use instead asyncEngineCount.
+    hipDeviceAttributeDirectManagedMemAccessFromHost,   ///< Host can directly access managed memory on
+                                                        ///< the device without migration
+    hipDeviceAttributeGlobalL1CacheSupported,           ///< Cuda only. Device supports caching globals in L1
+    hipDeviceAttributeHostNativeAtomicSupported,        ///< Cuda only. Link between the device and the host supports native atomic operations
+    hipDeviceAttributeIntegrated,                       ///< Device is integrated GPU
+    hipDeviceAttributeIsMultiGpuBoard,                  ///< Multiple GPU devices.
+    hipDeviceAttributeKernelExecTimeout,                ///< Run time limit for kernels executed on the device
+    hipDeviceAttributeL2CacheSize,                      ///< Size of L2 cache in bytes. 0 if the device doesn't have L2 cache.
+    hipDeviceAttributeLocalL1CacheSupported,            ///< caching locals in L1 is supported
+    hipDeviceAttributeLuid,                             ///< Cuda only. 8-byte locally unique identifier in 8 bytes. Undefined on TCC and non-Windows platforms
+    hipDeviceAttributeLuidDeviceNodeMask,               ///< Cuda only. Luid device node mask. Undefined on TCC and non-Windows platforms
+    hipDeviceAttributeComputeCapabilityMajor,           ///< Major compute capability version number.
+    hipDeviceAttributeManagedMemory,                    ///< Device supports allocating managed memory on this system
+    hipDeviceAttributeMaxBlocksPerMultiProcessor,       ///< Cuda only. Max block size per multiprocessor
+    hipDeviceAttributeMaxBlockDimX,                     ///< Max block size in width.
+    hipDeviceAttributeMaxBlockDimY,                     ///< Max block size in height.
+    hipDeviceAttributeMaxBlockDimZ,                     ///< Max block size in depth.
+    hipDeviceAttributeMaxGridDimX,                      ///< Max grid size  in width.
+    hipDeviceAttributeMaxGridDimY,                      ///< Max grid size  in height.
+    hipDeviceAttributeMaxGridDimZ,                      ///< Max grid size  in depth.
+    hipDeviceAttributeMaxSurface1D,                     ///< Maximum size of 1D surface.
+    hipDeviceAttributeMaxSurface1DLayered,              ///< Cuda only. Maximum dimensions of 1D layered surface.
+    hipDeviceAttributeMaxSurface2D,                     ///< Maximum dimension (width, height) of 2D surface.
+    hipDeviceAttributeMaxSurface2DLayered,              ///< Cuda only. Maximum dimensions of 2D layered surface.
+    hipDeviceAttributeMaxSurface3D,                     ///< Maximum dimension (width, height, depth) of 3D surface.
+    hipDeviceAttributeMaxSurfaceCubemap,                ///< Cuda only. Maximum dimensions of Cubemap surface.
+    hipDeviceAttributeMaxSurfaceCubemapLayered,         ///< Cuda only. Maximum dimension of Cubemap layered surface.
+    hipDeviceAttributeMaxTexture1DWidth,                ///< Maximum size of 1D texture.
+    hipDeviceAttributeMaxTexture1DLayered,              ///< Cuda only. Maximum dimensions of 1D layered texture.
+    hipDeviceAttributeMaxTexture1DLinear,               ///< Maximum number of elements allocatable in a 1D linear texture.
+                                                        ///< Use cudaDeviceGetTexture1DLinearMaxWidth() instead on Cuda.
+    hipDeviceAttributeMaxTexture1DMipmap,               ///< Cuda only. Maximum size of 1D mipmapped texture.
+    hipDeviceAttributeMaxTexture2DWidth,                ///< Maximum dimension width of 2D texture.
+    hipDeviceAttributeMaxTexture2DHeight,               ///< Maximum dimension hight of 2D texture.
+    hipDeviceAttributeMaxTexture2DGather,               ///< Cuda only. Maximum dimensions of 2D texture if gather operations  performed.
+    hipDeviceAttributeMaxTexture2DLayered,              ///< Cuda only. Maximum dimensions of 2D layered texture.
+    hipDeviceAttributeMaxTexture2DLinear,               ///< Cuda only. Maximum dimensions (width, height, pitch) of 2D textures bound to pitched memory.
+    hipDeviceAttributeMaxTexture2DMipmap,               ///< Cuda only. Maximum dimensions of 2D mipmapped texture.
+    hipDeviceAttributeMaxTexture3DWidth,                ///< Maximum dimension width of 3D texture.
+    hipDeviceAttributeMaxTexture3DHeight,               ///< Maximum dimension height of 3D texture.
+    hipDeviceAttributeMaxTexture3DDepth,                ///< Maximum dimension depth of 3D texture.
+    hipDeviceAttributeMaxTexture3DAlt,                  ///< Cuda only. Maximum dimensions of alternate 3D texture.
+    hipDeviceAttributeMaxTextureCubemap,                ///< Cuda only. Maximum dimensions of Cubemap texture
+    hipDeviceAttributeMaxTextureCubemapLayered,         ///< Cuda only. Maximum dimensions of Cubemap layered texture.
+    hipDeviceAttributeMaxThreadsDim,                    ///< Maximum dimension of a block
+    hipDeviceAttributeMaxThreadsPerBlock,               ///< Maximum number of threads per block.
+    hipDeviceAttributeMaxThreadsPerMultiProcessor,      ///< Maximum resident threads per multiprocessor.
+    hipDeviceAttributeMaxPitch,                         ///< Maximum pitch in bytes allowed by memory copies
+    hipDeviceAttributeMemoryBusWidth,                   ///< Global memory bus width in bits.
+    hipDeviceAttributeMemoryClockRate,                  ///< Peak memory clock frequency in kilohertz.
+    hipDeviceAttributeComputeCapabilityMinor,           ///< Minor compute capability version number.
+    hipDeviceAttributeMultiGpuBoardGroupID,             ///< Cuda only. Unique ID of device group on the same multi-GPU board
+    hipDeviceAttributeMultiprocessorCount,              ///< Number of multiprocessors on the device.
+    hipDeviceAttributeName,                             ///< Device name.
+    hipDeviceAttributePageableMemoryAccess,             ///< Device supports coherently accessing pageable memory
+                                                        ///< without calling hipHostRegister on it
+    hipDeviceAttributePageableMemoryAccessUsesHostPageTables, ///< Device accesses pageable memory via the host's page tables
+    hipDeviceAttributePciBusId,                         ///< PCI Bus ID.
+    hipDeviceAttributePciDeviceId,                      ///< PCI Device ID.
+    hipDeviceAttributePciDomainID,                      ///< PCI Domain ID.
+    hipDeviceAttributePersistingL2CacheMaxSize,         ///< Cuda11 only. Maximum l2 persisting lines capacity in bytes
+    hipDeviceAttributeMaxRegistersPerBlock,             ///< 32-bit registers available to a thread block. This number is shared
+                                                        ///< by all thread blocks simultaneously resident on a multiprocessor.
+    hipDeviceAttributeMaxRegistersPerMultiprocessor,    ///< 32-bit registers available per block.
+    hipDeviceAttributeReservedSharedMemPerBlock,        ///< Cuda11 only. Shared memory reserved by CUDA driver per block.
+    hipDeviceAttributeMaxSharedMemoryPerBlock,          ///< Maximum shared memory available per block in bytes.
+    hipDeviceAttributeSharedMemPerBlockOptin,           ///< Cuda only. Maximum shared memory per block usable by special opt in.
+    hipDeviceAttributeSharedMemPerMultiprocessor,       ///< Cuda only. Shared memory available per multiprocessor.
+    hipDeviceAttributeSingleToDoublePrecisionPerfRatio, ///< Cuda only. Performance ratio of single precision to double precision.
+    hipDeviceAttributeStreamPrioritiesSupported,        ///< Cuda only. Whether to support stream priorities.
+    hipDeviceAttributeSurfaceAlignment,                 ///< Cuda only. Alignment requirement for surfaces
+    hipDeviceAttributeTccDriver,                        ///< Cuda only. Whether device is a Tesla device using TCC driver
+    hipDeviceAttributeTextureAlignment,                 ///< Alignment requirement for textures
+    hipDeviceAttributeTexturePitchAlignment,            ///< Pitch alignment requirement for 2D texture references bound to pitched memory;
+    hipDeviceAttributeTotalConstantMemory,              ///< Constant memory size in bytes.
+    hipDeviceAttributeTotalGlobalMem,                   ///< Global memory available on devicice.
+    hipDeviceAttributeUnifiedAddressing,                ///< Cuda only. An unified address space shared with the host.
+    hipDeviceAttributeUuid,                             ///< Cuda only. Unique ID in 16 byte.
+    hipDeviceAttributeWarpSize,                         ///< Warp size in threads.
+
+    hipDeviceAttributeCudaCompatibleEnd = 9999,
+    hipDeviceAttributeAmdSpecificBegin = 10000,
+
+    hipDeviceAttributeClockInstructionRate = hipDeviceAttributeAmdSpecificBegin,  ///< Frequency in khz of the timer used by the device-side "clock*"
+    hipDeviceAttributeArch,                                     ///< Device architecture
+    hipDeviceAttributeMaxSharedMemoryPerMultiprocessor,         ///< Maximum Shared Memory PerMultiprocessor.
+    hipDeviceAttributeGcnArch,                                  ///< Device gcn architecture
+    hipDeviceAttributeGcnArchName,                              ///< Device gcnArch name in 256 bytes
+    hipDeviceAttributeHdpMemFlushCntl,                          ///< Address of the HDP_MEM_COHERENCY_FLUSH_CNTL register
+    hipDeviceAttributeHdpRegFlushCntl,                          ///< Address of the HDP_REG_COHERENCY_FLUSH_CNTL register
+    hipDeviceAttributeCooperativeMultiDeviceUnmatchedFunc,      ///< Supports cooperative launch on multiple
+                                                                ///< devices with unmatched functions
+    hipDeviceAttributeCooperativeMultiDeviceUnmatchedGridDim,   ///< Supports cooperative launch on multiple
+                                                                ///< devices with unmatched grid dimensions
+    hipDeviceAttributeCooperativeMultiDeviceUnmatchedBlockDim,  ///< Supports cooperative launch on multiple
+                                                                ///< devices with unmatched block dimensions
+    hipDeviceAttributeCooperativeMultiDeviceUnmatchedSharedMem, ///< Supports cooperative launch on multiple
+                                                                ///< devices with unmatched shared memories
+    hipDeviceAttributeIsLargeBar,                               ///< Whether it is LargeBar
+    hipDeviceAttributeAsicRevision,                             ///< Revision of the GPU in this device
+    hipDeviceAttributeCanUseStreamWaitValue,                    ///< '1' if Device supports hipStreamWaitValue32() and
+                                                                ///< hipStreamWaitValue64() , '0' otherwise.
+
+    hipDeviceAttributeAmdSpecificEnd = 19999,
+    hipDeviceAttributeVendorSpecificBegin = 20000,
+    // Extended attributes for vendors
+} hipDeviceAttribute_t;
+
+//! Flags that can be used with hipStreamCreateWithFlags
+#define hipStreamDefault                                                                           \
+    0x00  ///< Default stream creation flags. These are used with hipStreamCreate().
+#define hipStreamNonBlocking 0x01  ///< Stream does not implicitly synchronize with null stream
+
+
+//! Flags that can be used with hipEventCreateWithFlags:
+#define hipEventDefault 0x0  ///< Default flags
+#define hipEventBlockingSync                                                                       \
+    0x1  ///< Waiting will yield CPU.  Power-friendly and usage-friendly but may increase latency.
+#define hipEventDisableTiming                                                                      \
+    0x2  ///< Disable event's capability to record timing information.  May improve performance.
+#define hipEventInterprocess 0x4  ///< Event can support IPC.  @warning - not supported in HIP.
+#define hipEventReleaseToDevice                                                                    \
+    0x40000000  /// < Use a device-scope release when recording this event.  This flag is useful to
+                /// obtain more precise timings of commands between events.  The flag is a no-op on
+                /// CUDA platforms.
+#define hipEventReleaseToSystem                                                                    \
+    0x80000000  /// < Use a system-scope release when recording this event.  This flag is
+                /// useful to make non-coherent host memory visible to the host.  The flag is a
+                /// no-op on CUDA platforms.
+
+
+#define hipDeviceScheduleAuto 0x0  ///< Automatically select between Spin and Yield
+#define hipDeviceScheduleSpin                                                                      \
+    0x1  ///< Dedicate a CPU core to spin-wait.  Provides lowest latency, but burns a CPU core and
+         ///< may consume more power.
+#define hipDeviceScheduleYield                                                                     \
+    0x2  ///< Yield the CPU to the operating system when waiting.  May increase latency, but lowers
+         ///< power and is friendlier to other threads in the system.
+#define hipDeviceScheduleBlockingSync 0x4
+#define hipDeviceScheduleMask 0x7
+#define hipDeviceMapHost 0x8
+#define hipDeviceLmemResizeToMax 0x16
+
+typedef enum hipJitOption {
+    hipJitOptionMaxRegisters = 0,
+    hipJitOptionThreadsPerBlock,
+    hipJitOptionWallTime,
+    hipJitOptionInfoLogBuffer,
+    hipJitOptionInfoLogBufferSizeBytes,
+    hipJitOptionErrorLogBuffer,
+    hipJitOptionErrorLogBufferSizeBytes,
+    hipJitOptionOptimizationLevel,
+    hipJitOptionTargetFromContext,
+    hipJitOptionTarget,
+    hipJitOptionFallbackStrategy,
+    hipJitOptionGenerateDebugInfo,
+    hipJitOptionLogVerbose,
+    hipJitOptionGenerateLineInfo,
+    hipJitOptionCacheMode,
+    hipJitOptionSm3xOpt,
+    hipJitOptionFastCompile,
+    hipJitOptionNumOptions
+} hipJitOption;
+
+// stop: hip_runtime_api.h
 
 #ifdef _WIN32
 #define HIPAPI __stdcall
@@ -1004,63 +352,44 @@ typedef enum HIPjitInputType_enum
 
 #define HIP_API_CALL HIPAPI
 
-typedef HIPresult (HIP_API_CALL *HIP_HIPCTXCREATE)              (HIPcontext *, unsigned int, HIPdevice);
-typedef HIPresult (HIP_API_CALL *HIP_HIPCTXDESTROY)             (HIPcontext);
-typedef HIPresult (HIP_API_CALL *HIP_HIPCTXGETCACHECONFIG)      (HIPfunc_cache *);
-typedef HIPresult (HIP_API_CALL *HIP_HIPCTXGETCURRENT)          (HIPcontext *);
-typedef HIPresult (HIP_API_CALL *HIP_HIPCTXGETSHAREDMEMCONFIG)  (HIPsharedconfig *);
-typedef HIPresult (HIP_API_CALL *HIP_HIPCTXPOPCURRENT)          (HIPcontext *);
-typedef HIPresult (HIP_API_CALL *HIP_HIPCTXPUSHCURRENT)         (HIPcontext);
-typedef HIPresult (HIP_API_CALL *HIP_HIPCTXSETCACHECONFIG)      (HIPfunc_cache);
-typedef HIPresult (HIP_API_CALL *HIP_HIPCTXSETCURRENT)          (HIPcontext);
-typedef HIPresult (HIP_API_CALL *HIP_HIPCTXSETSHAREDMEMCONFIG)  (HIPsharedconfig);
-typedef HIPresult (HIP_API_CALL *HIP_HIPCTXSYNCHRONIZE)         ();
-typedef HIPresult (HIP_API_CALL *HIP_HIPDEVICEGETATTRIBUTE)     (int *, HIPdevice_attribute, HIPdevice);
-typedef HIPresult (HIP_API_CALL *HIP_HIPDEVICEGETCOUNT)         (int *);
-typedef HIPresult (HIP_API_CALL *HIP_HIPDEVICEGET)              (HIPdevice *, int);
-typedef HIPresult (HIP_API_CALL *HIP_HIPDEVICEGETNAME)          (char *, int, HIPdevice);
-typedef HIPresult (HIP_API_CALL *HIP_HIPDEVICETOTALMEM)         (size_t *, HIPdevice);
-typedef HIPresult (HIP_API_CALL *HIP_HIPDRIVERGETVERSION)       (int *);
-typedef HIPresult (HIP_API_CALL *HIP_HIPEVENTCREATE)            (HIPevent *, unsigned int);
-typedef HIPresult (HIP_API_CALL *HIP_HIPEVENTDESTROY)           (HIPevent);
-typedef HIPresult (HIP_API_CALL *HIP_HIPEVENTELAPSEDTIME)       (float *, HIPevent, HIPevent);
-typedef HIPresult (HIP_API_CALL *HIP_HIPEVENTQUERY)             (HIPevent);
-typedef HIPresult (HIP_API_CALL *HIP_HIPEVENTRECORD)            (HIPevent, HIPstream);
-typedef HIPresult (HIP_API_CALL *HIP_HIPEVENTSYNCHRONIZE)       (HIPevent);
-typedef HIPresult (HIP_API_CALL *HIP_HIPFUNCGETATTRIBUTE)       (int *, HIPfunction_attribute, HIPfunction);
-typedef HIPresult (HIP_API_CALL *HIP_HIPFUNCSETATTRIBUTE)       (HIPfunction, HIPfunction_attribute, int);
-typedef HIPresult (HIP_API_CALL *HIP_HIPFUNCSETCACHECONFIG)     (HIPfunction, HIPfunc_cache);
-typedef HIPresult (HIP_API_CALL *HIP_HIPFUNCSETSHAREDMEMCONFIG) (HIPfunction, HIPsharedconfig);
-typedef HIPresult (HIP_API_CALL *HIP_HIPGETERRORNAME)           (HIPresult, const char **);
-typedef HIPresult (HIP_API_CALL *HIP_HIPGETERRORSTRING)         (HIPresult, const char **);
-typedef HIPresult (HIP_API_CALL *HIP_HIPINIT)                   (unsigned int);
-typedef HIPresult (HIP_API_CALL *HIP_HIPLAUNCHKERNEL)           (HIPfunction, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, HIPstream, void **, void **);
-typedef HIPresult (HIP_API_CALL *HIP_HIPMEMALLOC)               (HIPdeviceptr *, size_t);
-typedef HIPresult (HIP_API_CALL *HIP_HIPMEMALLOCHOST)           (void **, size_t);
-typedef HIPresult (HIP_API_CALL *HIP_HIPMEMCPYDTOD)             (HIPdeviceptr, HIPdeviceptr, size_t);
-typedef HIPresult (HIP_API_CALL *HIP_HIPMEMCPYDTOH)             (void *, HIPdeviceptr, size_t);
-typedef HIPresult (HIP_API_CALL *HIP_HIPMEMCPYHTOD)             (HIPdeviceptr, const void *, size_t);
-typedef HIPresult (HIP_API_CALL *HIP_HIPMEMFREE)                (HIPdeviceptr);
-typedef HIPresult (HIP_API_CALL *HIP_HIPMEMFREEHOST)            (void *);
-typedef HIPresult (HIP_API_CALL *HIP_HIPMEMGETINFO)             (size_t *, size_t *);
-typedef HIPresult (HIP_API_CALL *HIP_HIPMEMSETD32)              (HIPdeviceptr, unsigned int, size_t);
-typedef HIPresult (HIP_API_CALL *HIP_HIPMEMSETD8)               (HIPdeviceptr, unsigned char, size_t);
-typedef HIPresult (HIP_API_CALL *HIP_HIPMODULEGETFUNCTION)      (HIPfunction *, HIPmodule, const char *);
-typedef HIPresult (HIP_API_CALL *HIP_HIPMODULEGETGLOBAL)        (HIPdeviceptr *, size_t *, HIPmodule, const char *);
-typedef HIPresult (HIP_API_CALL *HIP_HIPMODULELOAD)             (HIPmodule *, const char *);
-typedef HIPresult (HIP_API_CALL *HIP_HIPMODULELOADDATA)         (HIPmodule *, const void *);
-typedef HIPresult (HIP_API_CALL *HIP_HIPMODULELOADDATAEX)       (HIPmodule *, const void *, unsigned int, HIPjit_option *, void **);
-typedef HIPresult (HIP_API_CALL *HIP_HIPMODULEUNLOAD)           (HIPmodule);
-typedef HIPresult (HIP_API_CALL *HIP_HIPPROFILERSTART)          ();
-typedef HIPresult (HIP_API_CALL *HIP_HIPPROFILERSTOP)           ();
-typedef HIPresult (HIP_API_CALL *HIP_HIPSTREAMCREATE)           (HIPstream *, unsigned int);
-typedef HIPresult (HIP_API_CALL *HIP_HIPSTREAMDESTROY)          (HIPstream);
-typedef HIPresult (HIP_API_CALL *HIP_HIPSTREAMSYNCHRONIZE)      (HIPstream);
-typedef HIPresult (HIP_API_CALL *HIP_HIPSTREAMWAITEVENT)        (HIPstream, HIPevent, unsigned int);
-typedef HIPresult (HIP_API_CALL *HIP_HIPLINKCREATE)             (unsigned int, HIPjit_option *, void **, HIPlinkState *);
-typedef HIPresult (HIP_API_CALL *HIP_HIPLINKADDDATA)            (HIPlinkState, HIPjitInputType, void *, size_t, const char *, unsigned int, HIPjit_option *, void **);
-typedef HIPresult (HIP_API_CALL *HIP_HIPLINKDESTROY)            (HIPlinkState);
-typedef HIPresult (HIP_API_CALL *HIP_HIPLINKCOMPLETE)           (HIPlinkState, void **, size_t *);
+typedef hipError_t (HIP_API_CALL *HIP_HIPCTXCREATE)              (hipCtx_t *, unsigned int, hipDevice_t);
+typedef hipError_t (HIP_API_CALL *HIP_HIPCTXDESTROY)             (hipCtx_t);
+typedef hipError_t (HIP_API_CALL *HIP_HIPCTXPOPCURRENT)          (hipCtx_t *);
+typedef hipError_t (HIP_API_CALL *HIP_HIPCTXPUSHCURRENT)         (hipCtx_t);
+typedef hipError_t (HIP_API_CALL *HIP_HIPCTXSETCURRENT)          (hipCtx_t);
+typedef hipError_t (HIP_API_CALL *HIP_HIPCTXSYNCHRONIZE)         ();
+typedef hipError_t (HIP_API_CALL *HIP_HIPDEVICEGETATTRIBUTE)     (int *, hipDeviceAttribute_t, hipDevice_t);
+typedef hipError_t (HIP_API_CALL *HIP_HIPDEVICEGETCOUNT)         (int *);
+typedef hipError_t (HIP_API_CALL *HIP_HIPDEVICEGET)              (hipDevice_t *, int);
+typedef hipError_t (HIP_API_CALL *HIP_HIPDEVICEGETNAME)          (char *, int, hipDevice_t);
+typedef hipError_t (HIP_API_CALL *HIP_HIPDEVICETOTALMEM)         (size_t *, hipDevice_t);
+typedef hipError_t (HIP_API_CALL *HIP_HIPDRIVERGETVERSION)       (int *);
+typedef hipError_t (HIP_API_CALL *HIP_HIPEVENTCREATE)            (hipEvent_t *, unsigned int);
+typedef hipError_t (HIP_API_CALL *HIP_HIPEVENTDESTROY)           (hipEvent_t);
+typedef hipError_t (HIP_API_CALL *HIP_HIPEVENTELAPSEDTIME)       (float *, hipEvent_t, hipEvent_t);
+typedef hipError_t (HIP_API_CALL *HIP_HIPEVENTRECORD)            (hipEvent_t, hipStream_t);
+typedef hipError_t (HIP_API_CALL *HIP_HIPEVENTSYNCHRONIZE)       (hipEvent_t);
+typedef hipError_t (HIP_API_CALL *HIP_HIPFUNCGETATTRIBUTE)       (int *, hipFunction_attribute, hipFunction_t);
+typedef hipError_t (HIP_API_CALL *HIP_HIPGETERRORNAME)           (hipError_t, const char **);
+typedef hipError_t (HIP_API_CALL *HIP_HIPGETERRORSTRING)         (hipError_t, const char **);
+typedef hipError_t (HIP_API_CALL *HIP_HIPINIT)                   (unsigned int);
+typedef hipError_t (HIP_API_CALL *HIP_HIPLAUNCHKERNEL)           (hipFunction_t, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, hipStream_t, void **, void **);
+typedef hipError_t (HIP_API_CALL *HIP_HIPMEMALLOC)               (hipDeviceptr_t *, size_t);
+typedef hipError_t (HIP_API_CALL *HIP_HIPMEMFREE)                (hipDeviceptr_t);
+typedef hipError_t (HIP_API_CALL *HIP_HIPMEMGETINFO)             (size_t *, size_t *);
+typedef hipError_t (HIP_API_CALL *HIP_HIPMEMCPYDTODASYNC)        (hipDeviceptr_t, hipDeviceptr_t, size_t, hipStream_t);
+typedef hipError_t (HIP_API_CALL *HIP_HIPMEMCPYDTOHASYNC)        (void *, hipDeviceptr_t, size_t, hipStream_t);
+typedef hipError_t (HIP_API_CALL *HIP_HIPMEMCPYHTODASYNC)        (hipDeviceptr_t, const void *, size_t, hipStream_t);
+typedef hipError_t (HIP_API_CALL *HIP_HIPMEMSETD32ASYNC)         (hipDeviceptr_t, unsigned int, size_t, hipStream_t);
+typedef hipError_t (HIP_API_CALL *HIP_HIPMEMSETD8ASYNC)          (hipDeviceptr_t, unsigned char, size_t, hipStream_t);
+typedef hipError_t (HIP_API_CALL *HIP_HIPMODULEGETFUNCTION)      (hipFunction_t *, hipModule_t, const char *);
+typedef hipError_t (HIP_API_CALL *HIP_HIPMODULEGETGLOBAL)        (hipDeviceptr_t *, size_t *, hipModule_t, const char *);
+typedef hipError_t (HIP_API_CALL *HIP_HIPMODULELOADDATAEX)       (hipModule_t *, const void *, unsigned int, hipJitOption *, void **);
+typedef hipError_t (HIP_API_CALL *HIP_HIPMODULEUNLOAD)           (hipModule_t);
+typedef hipError_t (HIP_API_CALL *HIP_HIPRUNTIMEGETVERSION)      (int *);
+typedef hipError_t (HIP_API_CALL *HIP_HIPSTREAMCREATE)           (hipStream_t *, unsigned int);
+typedef hipError_t (HIP_API_CALL *HIP_HIPSTREAMDESTROY)          (hipStream_t);
+typedef hipError_t (HIP_API_CALL *HIP_HIPSTREAMSYNCHRONIZE)      (hipStream_t);
 
 typedef struct hc_hip_lib
 {
@@ -1068,14 +397,9 @@ typedef struct hc_hip_lib
 
   HIP_HIPCTXCREATE              hipCtxCreate;
   HIP_HIPCTXDESTROY             hipCtxDestroy;
-  HIP_HIPCTXGETCACHECONFIG      hipCtxGetCacheConfig;
-  HIP_HIPCTXGETCURRENT          hipCtxGetCurrent;
-  HIP_HIPCTXGETSHAREDMEMCONFIG  hipCtxGetSharedMemConfig;
   HIP_HIPCTXPOPCURRENT          hipCtxPopCurrent;
   HIP_HIPCTXPUSHCURRENT         hipCtxPushCurrent;
-  HIP_HIPCTXSETCACHECONFIG      hipCtxSetCacheConfig;
   HIP_HIPCTXSETCURRENT          hipCtxSetCurrent;
-  HIP_HIPCTXSETSHAREDMEMCONFIG  hipCtxSetSharedMemConfig;
   HIP_HIPCTXSYNCHRONIZE         hipCtxSynchronize;
   HIP_HIPDEVICEGETATTRIBUTE     hipDeviceGetAttribute;
   HIP_HIPDEVICEGETCOUNT         hipDeviceGetCount;
@@ -1086,46 +410,32 @@ typedef struct hc_hip_lib
   HIP_HIPEVENTCREATE            hipEventCreate;
   HIP_HIPEVENTDESTROY           hipEventDestroy;
   HIP_HIPEVENTELAPSEDTIME       hipEventElapsedTime;
-  HIP_HIPEVENTQUERY             hipEventQuery;
   HIP_HIPEVENTRECORD            hipEventRecord;
   HIP_HIPEVENTSYNCHRONIZE       hipEventSynchronize;
   HIP_HIPFUNCGETATTRIBUTE       hipFuncGetAttribute;
-  HIP_HIPFUNCSETATTRIBUTE       hipFuncSetAttribute;
-  HIP_HIPFUNCSETCACHECONFIG     hipFuncSetCacheConfig;
-  HIP_HIPFUNCSETSHAREDMEMCONFIG hipFuncSetSharedMemConfig;
   HIP_HIPGETERRORNAME           hipGetErrorName;
   HIP_HIPGETERRORSTRING         hipGetErrorString;
   HIP_HIPINIT                   hipInit;
   HIP_HIPLAUNCHKERNEL           hipLaunchKernel;
   HIP_HIPMEMALLOC               hipMemAlloc;
-  HIP_HIPMEMALLOCHOST           hipMemAllocHost;
-  HIP_HIPMEMCPYDTOD             hipMemcpyDtoD;
-  HIP_HIPMEMCPYDTOH             hipMemcpyDtoH;
-  HIP_HIPMEMCPYHTOD             hipMemcpyHtoD;
   HIP_HIPMEMFREE                hipMemFree;
-  HIP_HIPMEMFREEHOST            hipMemFreeHost;
   HIP_HIPMEMGETINFO             hipMemGetInfo;
-  HIP_HIPMEMSETD32              hipMemsetD32;
-  HIP_HIPMEMSETD8               hipMemsetD8;
+  HIP_HIPMEMCPYDTODASYNC        hipMemcpyDtoDAsync;
+  HIP_HIPMEMCPYDTOHASYNC        hipMemcpyDtoHAsync;
+  HIP_HIPMEMCPYHTODASYNC        hipMemcpyHtoDAsync;
+  HIP_HIPMEMSETD32ASYNC         hipMemsetD32Async;
+  HIP_HIPMEMSETD8ASYNC          hipMemsetD8Async;
   HIP_HIPMODULEGETFUNCTION      hipModuleGetFunction;
   HIP_HIPMODULEGETGLOBAL        hipModuleGetGlobal;
-  HIP_HIPMODULELOAD             hipModuleLoad;
-  HIP_HIPMODULELOADDATA         hipModuleLoadData;
   HIP_HIPMODULELOADDATAEX       hipModuleLoadDataEx;
   HIP_HIPMODULEUNLOAD           hipModuleUnload;
-  HIP_HIPPROFILERSTART          hipProfilerStart;
-  HIP_HIPPROFILERSTOP           hipProfilerStop;
+  HIP_HIPRUNTIMEGETVERSION      hipRuntimeGetVersion;
   HIP_HIPSTREAMCREATE           hipStreamCreate;
   HIP_HIPSTREAMDESTROY          hipStreamDestroy;
   HIP_HIPSTREAMSYNCHRONIZE      hipStreamSynchronize;
-  HIP_HIPSTREAMWAITEVENT        hipStreamWaitEvent;
-  HIP_HIPLINKCREATE             hipLinkCreate;
-  HIP_HIPLINKADDDATA            hipLinkAddData;
-  HIP_HIPLINKDESTROY            hipLinkDestroy;
-  HIP_HIPLINKCOMPLETE           hipLinkComplete;
 
 } hc_hip_lib_t;
 
 typedef hc_hip_lib_t HIP_PTR;
 
-#endif // _EXT_HIP_H
\ No newline at end of file
+#endif // _EXT_HIP_H
diff --git a/include/ext_hiprtc.h b/include/ext_hiprtc.h
index cd1be6c4b..347239c38 100644
--- a/include/ext_hiprtc.h
+++ b/include/ext_hiprtc.h
@@ -6,41 +6,26 @@
 #ifndef _EXT_HIPRTC_H
 #define _EXT_HIPRTC_H
 
-/**
- * from hip_runtime.h (/opt/rocm/hip/include/hip/amd_detail/hiprtc.h)
- */
+// start: amd_detail/hiprtc.h
 
-/**
- * \ingroup error
- * \brief   The enumerated type hiprtcResult defines API call result codes.
- *          HIPRTC API functions return hiprtcResult to indicate the call
- *          result.
- */
-typedef enum {
-  HIPRTC_SUCCESS = 0,
-  HIPRTC_ERROR_OUT_OF_MEMORY = 1,
-  HIPRTC_ERROR_PROGRAM_CREATION_FAILURE = 2,
-  HIPRTC_ERROR_INVALID_INPUT = 3,
-  HIPRTC_ERROR_INVALID_PROGRAM = 4,
-  HIPRTC_ERROR_INVALID_OPTION = 5,
-  HIPRTC_ERROR_COMPILATION = 6,
-  HIPRTC_ERROR_BUILTIN_OPERATION_FAILURE = 7,
-  HIPRTC_ERROR_NO_NAME_EXPRESSIONS_AFTER_COMPILATION = 8,
-  HIPRTC_ERROR_NO_LOWERED_NAMES_BEFORE_COMPILATION = 9,
-  HIPRTC_ERROR_NAME_EXPRESSION_NOT_VALID = 10,
-  HIPRTC_ERROR_INTERNAL_ERROR = 11
+typedef enum hiprtcResult {
+    HIPRTC_SUCCESS = 0,
+    HIPRTC_ERROR_OUT_OF_MEMORY = 1,
+    HIPRTC_ERROR_PROGRAM_CREATION_FAILURE = 2,
+    HIPRTC_ERROR_INVALID_INPUT = 3,
+    HIPRTC_ERROR_INVALID_PROGRAM = 4,
+    HIPRTC_ERROR_INVALID_OPTION = 5,
+    HIPRTC_ERROR_COMPILATION = 6,
+    HIPRTC_ERROR_BUILTIN_OPERATION_FAILURE = 7,
+    HIPRTC_ERROR_NO_NAME_EXPRESSIONS_AFTER_COMPILATION = 8,
+    HIPRTC_ERROR_NO_LOWERED_NAMES_BEFORE_COMPILATION = 9,
+    HIPRTC_ERROR_NAME_EXPRESSION_NOT_VALID = 10,
+    HIPRTC_ERROR_INTERNAL_ERROR = 11
 } hiprtcResult;
 
-/**
- * \ingroup compilation
- * \brief   hiprtcProgram is the unit of compilation, and an opaque handle for
- *          a program.
- *
- * To compile a CUDA program string, an instance of hiprtcProgram must be
- * created first with ::hiprtcCreateProgram, then compiled with
- * ::hiprtcCompileProgram.
- */
-typedef struct _hiprtcProgram *hiprtcProgram;
+typedef struct _hiprtcProgram* hiprtcProgram;
+
+// stop: amd_detail/hiprtc.h
 
 #ifdef _WIN32
 #define HIPRTCAPI __stdcall
@@ -54,13 +39,12 @@ typedef hiprtcResult  (HIPRTC_API_CALL *HIPRTC_HIPRTCADDNAMEEXPRESSION)  (hiprtc
 typedef hiprtcResult  (HIPRTC_API_CALL *HIPRTC_HIPRTCCOMPILEPROGRAM)     (hiprtcProgram, int, const char * const *);
 typedef hiprtcResult  (HIPRTC_API_CALL *HIPRTC_HIPRTCCREATEPROGRAM)      (hiprtcProgram *, const char *, const char *, int, const char * const *, const char * const *);
 typedef hiprtcResult  (HIPRTC_API_CALL *HIPRTC_HIPRTCDESTROYPROGRAM)     (hiprtcProgram *);
+typedef hiprtcResult  (HIPRTC_API_CALL *HIPRTC_HIPRTCGETCODE)            (hiprtcProgram, char *);
+typedef hiprtcResult  (HIPRTC_API_CALL *HIPRTC_HIPRTCGETCODESIZE)        (hiprtcProgram, size_t *);
 typedef hiprtcResult  (HIPRTC_API_CALL *HIPRTC_HIPRTCGETLOWEREDNAME)     (hiprtcProgram, const char * const, const char **);
-typedef hiprtcResult  (HIPRTC_API_CALL *HIPRTC_HIPRTCGETPTX)             (hiprtcProgram, char *);
-typedef hiprtcResult  (HIPRTC_API_CALL *HIPRTC_HIPRTCGETPTXSIZE)         (hiprtcProgram, size_t *);
 typedef hiprtcResult  (HIPRTC_API_CALL *HIPRTC_HIPRTCGETPROGRAMLOG)      (hiprtcProgram, char *);
 typedef hiprtcResult  (HIPRTC_API_CALL *HIPRTC_HIPRTCGETPROGRAMLOGSIZE)  (hiprtcProgram, size_t *);
-typedef const char * (HIPRTC_API_CALL *HIPRTC_HIPRTCGETERRORSTRING)      (hiprtcResult);
-typedef hiprtcResult  (HIPRTC_API_CALL *HIPRTC_HIPRTCVERSION)            (int *, int *);
+typedef const char *  (HIPRTC_API_CALL *HIPRTC_HIPRTCGETERRORSTRING)     (hiprtcResult);
 
 typedef struct hc_hiprtc_lib
 {
@@ -70,13 +54,12 @@ typedef struct hc_hiprtc_lib
   HIPRTC_HIPRTCCOMPILEPROGRAM     hiprtcCompileProgram;
   HIPRTC_HIPRTCCREATEPROGRAM      hiprtcCreateProgram;
   HIPRTC_HIPRTCDESTROYPROGRAM     hiprtcDestroyProgram;
+  HIPRTC_HIPRTCGETCODE            hiprtcGetCode;
+  HIPRTC_HIPRTCGETCODESIZE        hiprtcGetCodeSize;
   HIPRTC_HIPRTCGETLOWEREDNAME     hiprtcGetLoweredName;
-  HIPRTC_HIPRTCGETPTX             hiprtcGetCode;
-  HIPRTC_HIPRTCGETPTXSIZE         hiprtcGetCodeSize;
   HIPRTC_HIPRTCGETPROGRAMLOG      hiprtcGetProgramLog;
   HIPRTC_HIPRTCGETPROGRAMLOGSIZE  hiprtcGetProgramLogSize;
   HIPRTC_HIPRTCGETERRORSTRING     hiprtcGetErrorString;
-  HIPRTC_HIPRTCVERSION            hiprtcVersion;
 
 } hc_hiprtc_lib_t;
 
diff --git a/include/filehandling.h b/include/filehandling.h
index 1d13097e4..db03ac456 100644
--- a/include/filehandling.h
+++ b/include/filehandling.h
@@ -15,8 +15,8 @@
 int    _wopen       (const char *path, int oflag, ...);
 #endif
 
-bool   hc_fopen     (HCFILE *fp, const char *path, char *mode);
-bool   hc_fopen_raw (HCFILE *fp, const char *path, char *mode);
+bool   hc_fopen     (HCFILE *fp, const char *path, const char *mode);
+bool   hc_fopen_raw (HCFILE *fp, const char *path, const char *mode);
 int    hc_fscanf    (HCFILE *fp, const char *format, void *ptr);
 int    hc_fprintf   (HCFILE *fp, const char *format, ...);
 int    hc_vfprintf  (HCFILE *fp, const char *format, va_list ap);
diff --git a/include/hashes.h b/include/hashes.h
index ca88cab13..1e043aad4 100644
--- a/include/hashes.h
+++ b/include/hashes.h
@@ -15,7 +15,7 @@ int hash_encode (const hashconfig_t *hashconfig, const hashes_t *hashes, const m
 
 int save_hash (hashcat_ctx_t *hashcat_ctx);
 
-void check_hash (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, plain_t *plain);
+int check_hash (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, plain_t *plain);
 
 //int check_cracked (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, const u32 salt_pos);
 int check_cracked (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param);
diff --git a/include/interface.h b/include/interface.h
index 2035b6063..4e3485ced 100644
--- a/include/interface.h
+++ b/include/interface.h
@@ -13,7 +13,7 @@
 #include <limits.h>
 #include <inttypes.h>
 
-static const int MODULE_INTERFACE_VERSION_MINIMUM = 520;
+static const int MODULE_INTERFACE_VERSION_MINIMUM = 630;
 
 static const int MODULE_HASH_MODES_MAXIMUM = 100000;
 
diff --git a/include/modules.h b/include/modules.h
index 7f24c3c48..b32fb5580 100644
--- a/include/modules.h
+++ b/include/modules.h
@@ -11,6 +11,7 @@ void       *module_benchmark_esalt          (MAYBE_UNUSED const hashconfig_t *ha
 void       *module_benchmark_hook_salt      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra);
 const char *module_benchmark_mask           (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra);
 salt_t     *module_benchmark_salt           (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra);
+const char *module_deprecated_notice        (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra);
 bool        module_dictstat_disable         (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra);
 u32         module_dgst_pos0                (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra);
 u32         module_dgst_pos1                (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra);
@@ -18,6 +19,7 @@ u32         module_dgst_pos2                (MAYBE_UNUSED const hashconfig_t *ha
 u32         module_dgst_pos3                (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra);
 u32         module_dgst_size                (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra);
 u64         module_esalt_size               (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra);
+const char *module_extra_tuningdb_block     (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra);
 u32         module_forced_outfile_format    (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra);
 u32         module_hash_category            (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra);
 const char *module_hash_name                (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra);
diff --git a/include/tuningdb.h b/include/tuningdb.h
index 60fb60e68..1f306a3f4 100644
--- a/include/tuningdb.h
+++ b/include/tuningdb.h
@@ -11,9 +11,13 @@
 
 #define TUNING_DB_FILE "hashcat.hctune"
 
+int sort_by_tuning_db_alias (const void *v1, const void *v2);
+int sort_by_tuning_db_entry (const void *v1, const void *v2);
+
 int  tuning_db_init    (hashcat_ctx_t *hashcat_ctx);
 void tuning_db_destroy (hashcat_ctx_t *hashcat_ctx);
 
+bool tuning_db_process_line (hashcat_ctx_t *hashcat_ctx, const char *line_buf, const int line_num);
 tuning_db_entry_t *tuning_db_search (hashcat_ctx_t *hashcat_ctx, const char *device_name, const cl_device_type device_type, int attack_mode, const int hash_mode);
 
 #endif // _TUNINGDB_H
diff --git a/include/types.h b/include/types.h
index f68d3050c..0f0061a4f 100644
--- a/include/types.h
+++ b/include/types.h
@@ -295,6 +295,7 @@ typedef enum rule_functions
   RULE_OP_MANGLE_UREST_LFIRST    = 'C',
   RULE_OP_MANGLE_TREST           = 't',
   RULE_OP_MANGLE_TOGGLE_AT       = 'T',
+  RULE_OP_MANGLE_TOGGLE_AT_SEP   = '3',
   RULE_OP_MANGLE_REVERSE         = 'r',
   RULE_OP_MANGLE_DUPEWORD        = 'd',
   RULE_OP_MANGLE_DUPEWORD_TIMES  = 'p',
@@ -463,6 +464,8 @@ typedef enum dgst_size
   DGST_SIZE_4_32 = (32 * sizeof (u32)), // 128 !!!
   DGST_SIZE_4_64 = (64 * sizeof (u32)), // 256
   DGST_SIZE_8_2  = (2  * sizeof (u64)), // 16 !!!
+  DGST_SIZE_8_4  = (4  * sizeof (u64)), // 32 !!!
+  DGST_SIZE_8_6  = (6  * sizeof (u64)), // 48 !!!
   DGST_SIZE_8_8  = (8  * sizeof (u64)), // 64 !!!
   DGST_SIZE_8_16 = (16 * sizeof (u64)), // 128 !!!
   DGST_SIZE_8_25 = (25 * sizeof (u64))  // 200
@@ -612,6 +615,7 @@ typedef enum user_options_defaults
   BRAIN_SESSION            = 0,
   #endif
   DEBUG_MODE               = 0,
+  DEPRECATED_CHECK_DISABLE = false,
   FORCE                    = false,
   HWMON_DISABLE            = false,
   #if defined (__APPLE__)
@@ -651,6 +655,7 @@ typedef enum user_options_defaults
   BACKEND_INFO             = false,
   BACKEND_VECTOR_WIDTH     = 0,
   OPTIMIZED_KERNEL_ENABLE  = false,
+  MULTIPLY_ACCEL_DISABLE   = false,
   OUTFILE_AUTOHEX          = true,
   OUTFILE_CHECK_TIMER      = 5,
   OUTFILE_FORMAT           = 3,
@@ -670,7 +675,6 @@ typedef enum user_options_defaults
   SCRYPT_TMTO              = 0,
   SEGMENT_SIZE             = 33554432,
   SELF_TEST_DISABLE        = false,
-  SEPARATOR                = ':',
   SHOW                     = false,
   SKIP                     = 0,
   SLOW_CANDIDATES          = false,
@@ -723,88 +727,90 @@ typedef enum user_options_map
   IDX_CUSTOM_CHARSET_4          = '4',
   IDX_DEBUG_FILE                = 0xff11,
   IDX_DEBUG_MODE                = 0xff12,
-  IDX_ENCODING_FROM             = 0xff13,
-  IDX_ENCODING_TO               = 0xff14,
-  IDX_HASH_INFO                 = 0xff15,
-  IDX_FORCE                     = 0xff16,
-  IDX_HWMON_DISABLE             = 0xff17,
-  IDX_HWMON_TEMP_ABORT          = 0xff18,
+  IDX_DEPRECATED_CHECK_DISABLE  = 0xff13,
+  IDX_ENCODING_FROM             = 0xff14,
+  IDX_ENCODING_TO               = 0xff15,
+  IDX_HASH_INFO                 = 0xff16,
+  IDX_FORCE                     = 0xff17,
+  IDX_HWMON_DISABLE             = 0xff18,
+  IDX_HWMON_TEMP_ABORT          = 0xff19,
   IDX_HASH_MODE                 = 'm',
-  IDX_HCCAPX_MESSAGE_PAIR       = 0xff19,
+  IDX_HCCAPX_MESSAGE_PAIR       = 0xff1a,
   IDX_HELP                      = 'h',
-  IDX_HEX_CHARSET               = 0xff1a,
-  IDX_HEX_SALT                  = 0xff1b,
-  IDX_HEX_WORDLIST              = 0xff1c,
-  IDX_HOOK_THREADS              = 0xff1d,
-  IDX_IDENTIFY                  = 0xff1e,
+  IDX_HEX_CHARSET               = 0xff1b,
+  IDX_HEX_SALT                  = 0xff1c,
+  IDX_HEX_WORDLIST              = 0xff1d,
+  IDX_HOOK_THREADS              = 0xff1e,
+  IDX_IDENTIFY                  = 0xff1f,
   IDX_INCREMENT                 = 'i',
-  IDX_INCREMENT_MAX             = 0xff1f,
-  IDX_INCREMENT_MIN             = 0xff20,
-  IDX_INDUCTION_DIR             = 0xff21,
-  IDX_KEEP_GUESSING             = 0xff22,
+  IDX_INCREMENT_MAX             = 0xff20,
+  IDX_INCREMENT_MIN             = 0xff21,
+  IDX_INDUCTION_DIR             = 0xff22,
+  IDX_KEEP_GUESSING             = 0xff23,
   IDX_KERNEL_ACCEL              = 'n',
   IDX_KERNEL_LOOPS              = 'u',
   IDX_KERNEL_THREADS            = 'T',
-  IDX_KEYBOARD_LAYOUT_MAPPING   = 0xff23,
-  IDX_KEYSPACE                  = 0xff24,
-  IDX_LEFT                      = 0xff25,
+  IDX_KEYBOARD_LAYOUT_MAPPING   = 0xff24,
+  IDX_KEYSPACE                  = 0xff25,
+  IDX_LEFT                      = 0xff26,
   IDX_LIMIT                     = 'l',
-  IDX_LOGFILE_DISABLE           = 0xff26,
-  IDX_LOOPBACK                  = 0xff27,
-  IDX_MACHINE_READABLE          = 0xff28,
-  IDX_MARKOV_CLASSIC            = 0xff29,
-  IDX_MARKOV_DISABLE            = 0xff2a,
-  IDX_MARKOV_HCSTAT2            = 0xff2b,
-  IDX_MARKOV_INVERSE            = 0xff2c,
+  IDX_LOGFILE_DISABLE           = 0xff27,
+  IDX_LOOPBACK                  = 0xff28,
+  IDX_MACHINE_READABLE          = 0xff29,
+  IDX_MARKOV_CLASSIC            = 0xff2a,
+  IDX_MARKOV_DISABLE            = 0xff2b,
+  IDX_MARKOV_HCSTAT2            = 0xff2c,
+  IDX_MARKOV_INVERSE            = 0xff2d,
   IDX_MARKOV_THRESHOLD          = 't',
-  IDX_NONCE_ERROR_CORRECTIONS   = 0xff2d,
+  IDX_NONCE_ERROR_CORRECTIONS   = 0xff2e,
   IDX_OPENCL_DEVICE_TYPES       = 'D',
   IDX_OPTIMIZED_KERNEL_ENABLE   = 'O',
-  IDX_OUTFILE_AUTOHEX_DISABLE   = 0xff2e,
-  IDX_OUTFILE_CHECK_DIR         = 0xff2f,
-  IDX_OUTFILE_CHECK_TIMER       = 0xff30,
-  IDX_OUTFILE_FORMAT            = 0xff31,
+  IDX_MULTIPLY_ACCEL_DISABLE    = 'M',
+  IDX_OUTFILE_AUTOHEX_DISABLE   = 0xff2f,
+  IDX_OUTFILE_CHECK_DIR         = 0xff30,
+  IDX_OUTFILE_CHECK_TIMER       = 0xff31,
+  IDX_OUTFILE_FORMAT            = 0xff32,
   IDX_OUTFILE                   = 'o',
-  IDX_POTFILE_DISABLE           = 0xff32,
-  IDX_POTFILE_PATH              = 0xff33,
-  IDX_PROGRESS_ONLY             = 0xff34,
-  IDX_QUIET                     = 0xff35,
-  IDX_REMOVE                    = 0xff36,
-  IDX_REMOVE_TIMER              = 0xff37,
-  IDX_RESTORE                   = 0xff38,
-  IDX_RESTORE_DISABLE           = 0xff39,
-  IDX_RESTORE_FILE_PATH         = 0xff3a,
+  IDX_POTFILE_DISABLE           = 0xff33,
+  IDX_POTFILE_PATH              = 0xff34,
+  IDX_PROGRESS_ONLY             = 0xff35,
+  IDX_QUIET                     = 0xff36,
+  IDX_REMOVE                    = 0xff37,
+  IDX_REMOVE_TIMER              = 0xff38,
+  IDX_RESTORE                   = 0xff39,
+  IDX_RESTORE_DISABLE           = 0xff3a,
+  IDX_RESTORE_FILE_PATH         = 0xff3b,
   IDX_RP_FILE                   = 'r',
-  IDX_RP_GEN_FUNC_MAX           = 0xff3b,
-  IDX_RP_GEN_FUNC_MIN           = 0xff3c,
+  IDX_RP_GEN_FUNC_MAX           = 0xff3c,
+  IDX_RP_GEN_FUNC_MIN           = 0xff3d,
   IDX_RP_GEN                    = 'g',
-  IDX_RP_GEN_SEED               = 0xff3d,
+  IDX_RP_GEN_SEED               = 0xff3e,
   IDX_RULE_BUF_L                = 'j',
   IDX_RULE_BUF_R                = 'k',
-  IDX_RUNTIME                   = 0xff3e,
-  IDX_SCRYPT_TMTO               = 0xff3f,
+  IDX_RUNTIME                   = 0xff3f,
+  IDX_SCRYPT_TMTO               = 0xff40,
   IDX_SEGMENT_SIZE              = 'c',
-  IDX_SELF_TEST_DISABLE         = 0xff40,
+  IDX_SELF_TEST_DISABLE         = 0xff41,
   IDX_SEPARATOR                 = 'p',
-  IDX_SESSION                   = 0xff41,
-  IDX_SHOW                      = 0xff42,
+  IDX_SESSION                   = 0xff42,
+  IDX_SHOW                      = 0xff43,
   IDX_SKIP                      = 's',
   IDX_SLOW_CANDIDATES           = 'S',
-  IDX_SPEED_ONLY                = 0xff43,
-  IDX_SPIN_DAMP                 = 0xff44,
-  IDX_STATUS                    = 0xff45,
-  IDX_STATUS_JSON               = 0xff46,
-  IDX_STATUS_TIMER              = 0xff47,
-  IDX_STDOUT_FLAG               = 0xff48,
-  IDX_STDIN_TIMEOUT_ABORT       = 0xff49,
-  IDX_TRUECRYPT_KEYFILES        = 0xff4a,
-  IDX_USERNAME                  = 0xff4b,
-  IDX_VERACRYPT_KEYFILES        = 0xff4c,
-  IDX_VERACRYPT_PIM_START       = 0xff4d,
-  IDX_VERACRYPT_PIM_STOP        = 0xff4e,
+  IDX_SPEED_ONLY                = 0xff44,
+  IDX_SPIN_DAMP                 = 0xff45,
+  IDX_STATUS                    = 0xff46,
+  IDX_STATUS_JSON               = 0xff47,
+  IDX_STATUS_TIMER              = 0xff48,
+  IDX_STDOUT_FLAG               = 0xff49,
+  IDX_STDIN_TIMEOUT_ABORT       = 0xff4a,
+  IDX_TRUECRYPT_KEYFILES        = 0xff4b,
+  IDX_USERNAME                  = 0xff4c,
+  IDX_VERACRYPT_KEYFILES        = 0xff4d,
+  IDX_VERACRYPT_PIM_START       = 0xff4e,
+  IDX_VERACRYPT_PIM_STOP        = 0xff4f,
   IDX_VERSION_LOWER             = 'v',
   IDX_VERSION                   = 'V',
-  IDX_WORDLIST_AUTOHEX_DISABLE  = 0xff4f,
+  IDX_WORDLIST_AUTOHEX_DISABLE  = 0xff50,
   IDX_WORKLOAD_PROFILE          = 'w',
 
 } user_options_map_t;
@@ -925,7 +931,6 @@ typedef struct hashes
 
   void        *digests_buf;
   u32         *digests_shown;
-  u32         *digests_shown_tmp;
 
   u32          salts_cnt;
   u32          salts_done;
@@ -1074,7 +1079,7 @@ typedef struct hc_fp
   bool        is_zip;
   int         bom_size;
 
-  char       *mode;
+  const char *mode;
   const char *path;
 
 } HCFILE;
@@ -1116,6 +1121,8 @@ typedef struct hc_device_param
   int     sm_minor;
   u32     kernel_exec_timeout;
 
+  u32     kernel_preferred_wgs_multiple;
+
   st_status_t st_status;
 
   int     vector_width;
@@ -1137,6 +1144,7 @@ typedef struct hc_device_param
   u32     kernel_wgs_amp;
   u32     kernel_wgs_tm;
   u32     kernel_wgs_memset;
+  u32     kernel_wgs_bzero;
   u32     kernel_wgs_atinit;
   u32     kernel_wgs_utf8toutf16le;
   u32     kernel_wgs_decompress;
@@ -1162,6 +1170,7 @@ typedef struct hc_device_param
   u32     kernel_preferred_wgs_multiple_amp;
   u32     kernel_preferred_wgs_multiple_tm;
   u32     kernel_preferred_wgs_multiple_memset;
+  u32     kernel_preferred_wgs_multiple_bzero;
   u32     kernel_preferred_wgs_multiple_atinit;
   u32     kernel_preferred_wgs_multiple_utf8toutf16le;
   u32     kernel_preferred_wgs_multiple_decompress;
@@ -1187,6 +1196,7 @@ typedef struct hc_device_param
   u64     kernel_local_mem_size_amp;
   u64     kernel_local_mem_size_tm;
   u64     kernel_local_mem_size_memset;
+  u64     kernel_local_mem_size_bzero;
   u64     kernel_local_mem_size_atinit;
   u64     kernel_local_mem_size_utf8toutf16le;
   u64     kernel_local_mem_size_decompress;
@@ -1212,6 +1222,7 @@ typedef struct hc_device_param
   u64     kernel_dynamic_local_mem_size_amp;
   u64     kernel_dynamic_local_mem_size_tm;
   u64     kernel_dynamic_local_mem_size_memset;
+  u64     kernel_dynamic_local_mem_size_bzero;
   u64     kernel_dynamic_local_mem_size_atinit;
   u64     kernel_dynamic_local_mem_size_utf8toutf16le;
   u64     kernel_dynamic_local_mem_size_decompress;
@@ -1373,6 +1384,7 @@ typedef struct hc_device_param
   void   *kernel_params_amp[PARAMCNT];
   void   *kernel_params_tm[PARAMCNT];
   void   *kernel_params_memset[PARAMCNT];
+  void   *kernel_params_bzero[PARAMCNT];
   void   *kernel_params_atinit[PARAMCNT];
   void   *kernel_params_utf8toutf16le[PARAMCNT];
   void   *kernel_params_decompress[PARAMCNT];
@@ -1395,6 +1407,9 @@ typedef struct hc_device_param
   u32     kernel_params_memset_buf32[PARAMCNT];
   u64     kernel_params_memset_buf64[PARAMCNT];
 
+  u32     kernel_params_bzero_buf32[PARAMCNT];
+  u64     kernel_params_bzero_buf64[PARAMCNT];
+
   u32     kernel_params_atinit_buf32[PARAMCNT];
   u64     kernel_params_atinit_buf64[PARAMCNT];
 
@@ -1416,6 +1431,7 @@ typedef struct hc_device_param
 
   CUevent           cuda_event1;
   CUevent           cuda_event2;
+  CUevent           cuda_event3;
 
   CUmodule          cuda_module;
   CUmodule          cuda_module_shared;
@@ -1439,6 +1455,7 @@ typedef struct hc_device_param
   CUfunction        cuda_function_amp;
   CUfunction        cuda_function_tm;
   CUfunction        cuda_function_memset;
+  CUfunction        cuda_function_bzero;
   CUfunction        cuda_function_atinit;
   CUfunction        cuda_function_utf8toutf16le;
   CUfunction        cuda_function_decompress;
@@ -1490,79 +1507,81 @@ typedef struct hc_device_param
 
   int               hip_warp_size;
 
-  HIPdevice         hip_device;
-  HIPcontext        hip_context;
-  HIPstream         hip_stream;
+  hipDevice_t       hip_device;
+  hipCtx_t          hip_context;
+  hipStream_t       hip_stream;
 
-  HIPevent          hip_event1;
-  HIPevent          hip_event2;
+  hipEvent_t        hip_event1;
+  hipEvent_t        hip_event2;
+  hipEvent_t        hip_event3;
 
-  HIPmodule         hip_module;
-  HIPmodule         hip_module_shared;
-  HIPmodule         hip_module_mp;
-  HIPmodule         hip_module_amp;
+  hipModule_t       hip_module;
+  hipModule_t       hip_module_shared;
+  hipModule_t       hip_module_mp;
+  hipModule_t       hip_module_amp;
 
-  HIPfunction       hip_function1;
-  HIPfunction       hip_function12;
-  HIPfunction       hip_function2p;
-  HIPfunction       hip_function2;
-  HIPfunction       hip_function2e;
-  HIPfunction       hip_function23;
-  HIPfunction       hip_function3;
-  HIPfunction       hip_function4;
-  HIPfunction       hip_function_init2;
-  HIPfunction       hip_function_loop2p;
-  HIPfunction       hip_function_loop2;
-  HIPfunction       hip_function_mp;
-  HIPfunction       hip_function_mp_l;
-  HIPfunction       hip_function_mp_r;
-  HIPfunction       hip_function_amp;
-  HIPfunction       hip_function_tm;
-  HIPfunction       hip_function_memset;
-  HIPfunction       hip_function_atinit;
-  HIPfunction       hip_function_utf8toutf16le;
-  HIPfunction       hip_function_decompress;
-  HIPfunction       hip_function_aux1;
-  HIPfunction       hip_function_aux2;
-  HIPfunction       hip_function_aux3;
-  HIPfunction       hip_function_aux4;
+  hipFunction_t     hip_function1;
+  hipFunction_t     hip_function12;
+  hipFunction_t     hip_function2p;
+  hipFunction_t     hip_function2;
+  hipFunction_t     hip_function2e;
+  hipFunction_t     hip_function23;
+  hipFunction_t     hip_function3;
+  hipFunction_t     hip_function4;
+  hipFunction_t     hip_function_init2;
+  hipFunction_t     hip_function_loop2p;
+  hipFunction_t     hip_function_loop2;
+  hipFunction_t     hip_function_mp;
+  hipFunction_t     hip_function_mp_l;
+  hipFunction_t     hip_function_mp_r;
+  hipFunction_t     hip_function_amp;
+  hipFunction_t     hip_function_tm;
+  hipFunction_t     hip_function_memset;
+  hipFunction_t     hip_function_bzero;
+  hipFunction_t     hip_function_atinit;
+  hipFunction_t     hip_function_utf8toutf16le;
+  hipFunction_t     hip_function_decompress;
+  hipFunction_t     hip_function_aux1;
+  hipFunction_t     hip_function_aux2;
+  hipFunction_t     hip_function_aux3;
+  hipFunction_t     hip_function_aux4;
 
-  HIPdeviceptr      hip_d_pws_buf;
-  HIPdeviceptr      hip_d_pws_amp_buf;
-  HIPdeviceptr      hip_d_pws_comp_buf;
-  HIPdeviceptr      hip_d_pws_idx;
-  HIPdeviceptr      hip_d_rules;
-  HIPdeviceptr      hip_d_rules_c;
-  HIPdeviceptr      hip_d_combs;
-  HIPdeviceptr      hip_d_combs_c;
-  HIPdeviceptr      hip_d_bfs;
-  HIPdeviceptr      hip_d_bfs_c;
-  HIPdeviceptr      hip_d_tm_c;
-  HIPdeviceptr      hip_d_bitmap_s1_a;
-  HIPdeviceptr      hip_d_bitmap_s1_b;
-  HIPdeviceptr      hip_d_bitmap_s1_c;
-  HIPdeviceptr      hip_d_bitmap_s1_d;
-  HIPdeviceptr      hip_d_bitmap_s2_a;
-  HIPdeviceptr      hip_d_bitmap_s2_b;
-  HIPdeviceptr      hip_d_bitmap_s2_c;
-  HIPdeviceptr      hip_d_bitmap_s2_d;
-  HIPdeviceptr      hip_d_plain_bufs;
-  HIPdeviceptr      hip_d_digests_buf;
-  HIPdeviceptr      hip_d_digests_shown;
-  HIPdeviceptr      hip_d_salt_bufs;
-  HIPdeviceptr      hip_d_esalt_bufs;
-  HIPdeviceptr      hip_d_tmps;
-  HIPdeviceptr      hip_d_hooks;
-  HIPdeviceptr      hip_d_result;
-  HIPdeviceptr      hip_d_extra0_buf;
-  HIPdeviceptr      hip_d_extra1_buf;
-  HIPdeviceptr      hip_d_extra2_buf;
-  HIPdeviceptr      hip_d_extra3_buf;
-  HIPdeviceptr      hip_d_root_css_buf;
-  HIPdeviceptr      hip_d_markov_css_buf;
-  HIPdeviceptr      hip_d_st_digests_buf;
-  HIPdeviceptr      hip_d_st_salts_buf;
-  HIPdeviceptr      hip_d_st_esalts_buf;
+  hipDeviceptr_t    hip_d_pws_buf;
+  hipDeviceptr_t    hip_d_pws_amp_buf;
+  hipDeviceptr_t    hip_d_pws_comp_buf;
+  hipDeviceptr_t    hip_d_pws_idx;
+  hipDeviceptr_t    hip_d_rules;
+  hipDeviceptr_t    hip_d_rules_c;
+  hipDeviceptr_t    hip_d_combs;
+  hipDeviceptr_t    hip_d_combs_c;
+  hipDeviceptr_t    hip_d_bfs;
+  hipDeviceptr_t    hip_d_bfs_c;
+  hipDeviceptr_t    hip_d_tm_c;
+  hipDeviceptr_t    hip_d_bitmap_s1_a;
+  hipDeviceptr_t    hip_d_bitmap_s1_b;
+  hipDeviceptr_t    hip_d_bitmap_s1_c;
+  hipDeviceptr_t    hip_d_bitmap_s1_d;
+  hipDeviceptr_t    hip_d_bitmap_s2_a;
+  hipDeviceptr_t    hip_d_bitmap_s2_b;
+  hipDeviceptr_t    hip_d_bitmap_s2_c;
+  hipDeviceptr_t    hip_d_bitmap_s2_d;
+  hipDeviceptr_t    hip_d_plain_bufs;
+  hipDeviceptr_t    hip_d_digests_buf;
+  hipDeviceptr_t    hip_d_digests_shown;
+  hipDeviceptr_t    hip_d_salt_bufs;
+  hipDeviceptr_t    hip_d_esalt_bufs;
+  hipDeviceptr_t    hip_d_tmps;
+  hipDeviceptr_t    hip_d_hooks;
+  hipDeviceptr_t    hip_d_result;
+  hipDeviceptr_t    hip_d_extra0_buf;
+  hipDeviceptr_t    hip_d_extra1_buf;
+  hipDeviceptr_t    hip_d_extra2_buf;
+  hipDeviceptr_t    hip_d_extra3_buf;
+  hipDeviceptr_t    hip_d_root_css_buf;
+  hipDeviceptr_t    hip_d_markov_css_buf;
+  hipDeviceptr_t    hip_d_st_digests_buf;
+  hipDeviceptr_t    hip_d_st_salts_buf;
+  hipDeviceptr_t    hip_d_st_esalts_buf;
 
   // API: opencl
 
@@ -1604,6 +1623,7 @@ typedef struct hc_device_param
   cl_kernel         opencl_kernel_amp;
   cl_kernel         opencl_kernel_tm;
   cl_kernel         opencl_kernel_memset;
+  cl_kernel         opencl_kernel_bzero;
   cl_kernel         opencl_kernel_atinit;
   cl_kernel         opencl_kernel_utf8toutf16le;
   cl_kernel         opencl_kernel_decompress;
@@ -1712,8 +1732,8 @@ typedef struct backend_ctx
   int                 rc_hip_init;
   int                 rc_hiprtc_init;
 
-  int                 hiprtc_driver_version;
-  int                 hip_driver_version;
+  int                 hip_runtimeVersion;
+  int                 hip_driverVersion;
 
   // opencl
 
@@ -1785,8 +1805,6 @@ typedef struct hwmon_ctx
 
   hm_attrs_t *hm_device;
 
-  ADLOD6MemClockState *od_clock_mem_status;
-
 } hwmon_ctx_t;
 
 #if defined (__APPLE__)
@@ -2031,9 +2049,11 @@ typedef struct tuning_db
 
   tuning_db_alias_t *alias_buf;
   int                alias_cnt;
+  int                alias_alloc;
 
   tuning_db_entry_t *entry_buf;
   int                entry_cnt;
+  int                entry_alloc;
 
 } tuning_db_t;
 
@@ -2090,6 +2110,7 @@ typedef struct user_options
   bool         skip_chgd;
   bool         limit_chgd;
   bool         scrypt_tmto_chgd;
+  bool         separator_chgd;
 
   bool         advice_disable;
   bool         benchmark;
@@ -2099,6 +2120,7 @@ typedef struct user_options
   bool         brain_server;
   #endif
   bool         force;
+  bool         deprecated_check_disable;
   bool         hwmon_disable;
   bool         hash_info;
   bool         hex_charset;
@@ -2119,6 +2141,7 @@ typedef struct user_options
   bool         backend_ignore_opencl;
   bool         backend_info;
   bool         optimized_kernel_enable;
+  bool         multiply_accel_disable;
   bool         outfile_autohex;
   bool         potfile_disable;
   bool         progress_only;
@@ -2158,7 +2181,7 @@ typedef struct user_options
   char        *potfile_path;
   char        *restore_file_path;
   char       **rp_files;
-  char         separator;
+  char        *separator;
   char        *truecrypt_keyfiles;
   char        *veracrypt_keyfiles;
   const char  *custom_charset_1;
@@ -2224,6 +2247,8 @@ typedef struct user_options_extra
 
   u32 wordlist_mode;
 
+  char   separator;
+
   char  *hc_hash;   // can be filename or string
 
   int    hc_workc;  // can be 0 in bf-mode = default mask
@@ -2626,13 +2651,15 @@ typedef struct module_ctx
   void       *(*module_benchmark_hook_salt)     (const hashconfig_t *, const user_options_t *, const user_options_extra_t *);
   const char *(*module_benchmark_mask)          (const hashconfig_t *, const user_options_t *, const user_options_extra_t *);
   salt_t     *(*module_benchmark_salt)          (const hashconfig_t *, const user_options_t *, const user_options_extra_t *);
-  bool        (*module_dictstat_disable)        (const hashconfig_t *, const user_options_t *, const user_options_extra_t *);
+  const char *(*module_deprecated_notice)       (const hashconfig_t *, const user_options_t *, const user_options_extra_t *);
   u32         (*module_dgst_pos0)               (const hashconfig_t *, const user_options_t *, const user_options_extra_t *);
   u32         (*module_dgst_pos1)               (const hashconfig_t *, const user_options_t *, const user_options_extra_t *);
   u32         (*module_dgst_pos2)               (const hashconfig_t *, const user_options_t *, const user_options_extra_t *);
   u32         (*module_dgst_pos3)               (const hashconfig_t *, const user_options_t *, const user_options_extra_t *);
   u32         (*module_dgst_size)               (const hashconfig_t *, const user_options_t *, const user_options_extra_t *);
+  bool        (*module_dictstat_disable)        (const hashconfig_t *, const user_options_t *, const user_options_extra_t *);
   u64         (*module_esalt_size)              (const hashconfig_t *, const user_options_t *, const user_options_extra_t *);
+  const char *(*module_extra_tuningdb_block)    (const hashconfig_t *, const user_options_t *, const user_options_extra_t *);
   u32         (*module_forced_outfile_format)   (const hashconfig_t *, const user_options_t *, const user_options_extra_t *);
   u32         (*module_hash_category)           (const hashconfig_t *, const user_options_t *, const user_options_extra_t *);
   const char *(*module_hash_name)               (const hashconfig_t *, const user_options_t *, const user_options_extra_t *);
diff --git a/src/Makefile b/src/Makefile
index e1ff6a534..066c898af 100644
--- a/src/Makefile
+++ b/src/Makefile
@@ -53,7 +53,7 @@ endif
 ## Do not modify
 ##
 
-MODULE_INTERFACE_VERSION := 620
+MODULE_INTERFACE_VERSION := 630
 
 ##
 ## Native compiler paths
@@ -203,12 +203,11 @@ endif
 ## because UNRAR
 ifeq ($(ENABLE_UNRAR),1)
 ifeq ($(USE_SYSTEM_UNRAR),0)
-ifneq ($(UNAME),Darwin)
-CFLAGS_UNRAR            += -Wno-misleading-indentation
+ifneq ($(CC),clang)
 CFLAGS_UNRAR            += -Wno-class-memaccess
-else
-CFLAGS_UNRAR            += -Wno-missing-braces
+CFLAGS_UNRAR            += -Wno-misleading-indentation
 endif
+CFLAGS_UNRAR            += -Wno-missing-braces
 CFLAGS_UNRAR            += -Wno-unused-variable
 CFLAGS_UNRAR            += -Wno-unused-parameter
 CFLAGS_UNRAR            += -Wno-unused-function
diff --git a/src/autotune.c b/src/autotune.c
index f9aeddd83..9f2ac312c 100644
--- a/src/autotune.c
+++ b/src/autotune.c
@@ -10,7 +10,7 @@
 #include "status.h"
 #include "autotune.h"
 
-static double try_run (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, const u32 kernel_accel, const u32 kernel_loops)
+static double try_run (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, const u32 kernel_accel, const u32 kernel_loops, const u32 kernel_threads)
 {
   hashconfig_t   *hashconfig   = hashcat_ctx->hashconfig;
   user_options_t *user_options = hashcat_ctx->user_options;
@@ -19,7 +19,9 @@ static double try_run (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_par
   device_param->kernel_params_buf32[29] = kernel_loops; // not a bug, both need to be set
   device_param->kernel_params_buf32[30] = kernel_loops; // because there's two variables for inner iters for slow and fast hashes
 
-  u32 kernel_power_try = device_param->hardware_power * kernel_accel;
+  const u32 hardware_power = ((hashconfig->opts_type & OPTS_TYPE_MP_MULTI_DISABLE) ? 1 : device_param->device_processors) * kernel_threads;
+
+  u32 kernel_power_try = hardware_power * kernel_accel;
 
   if (user_options->attack_mode == ATTACK_MODE_ASSOCIATION)
   {
@@ -33,6 +35,10 @@ static double try_run (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_par
     }
   }
 
+  const u32 kernel_threads_sav = device_param->kernel_threads;
+
+  device_param->kernel_threads = kernel_threads;
+
   const double spin_damp_sav = device_param->spin_damp;
 
   device_param->spin_damp = 0;
@@ -55,59 +61,46 @@ static double try_run (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_par
 
   device_param->spin_damp = spin_damp_sav;
 
-  const double exec_msec_prev = get_avg_exec_time (device_param, 1);
-
-  return exec_msec_prev;
-}
-
-/*
-static double try_run_preferred (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, const u32 kernel_accel, const u32 kernel_loops)
-{
-  hashconfig_t *hashconfig = hashcat_ctx->hashconfig;
-
-  device_param->kernel_params_buf32[28] = 0;
-  device_param->kernel_params_buf32[29] = kernel_loops; // not a bug, both need to be set
-  device_param->kernel_params_buf32[30] = kernel_loops; // because there's two variables for inner iters for slow and fast hashes
-
-  const u32 kernel_power_try = device_param->hardware_power * kernel_accel;
-
-  const u32 kernel_threads_sav = device_param->kernel_threads;
-
-  const double spin_damp_sav = device_param->spin_damp;
-
-  device_param->spin_damp = 0;
-
-  if (hashconfig->attack_exec == ATTACK_EXEC_INSIDE_KERNEL)
-  {
-    if (hashconfig->opti_type & OPTI_TYPE_OPTIMIZED_KERNEL)
-    {
-      device_param->kernel_threads = device_param->kernel_preferred_wgs_multiple1;
-
-      run_kernel (hashcat_ctx, device_param, KERN_RUN_1, 0, kernel_power_try, true, 0);
-    }
-    else
-    {
-      device_param->kernel_threads = device_param->kernel_preferred_wgs_multiple4;
-
-      run_kernel (hashcat_ctx, device_param, KERN_RUN_4, 0, kernel_power_try, true, 0);
-    }
-  }
-  else
-  {
-    device_param->kernel_threads = device_param->kernel_preferred_wgs_multiple2;
-
-    run_kernel (hashcat_ctx, device_param, KERN_RUN_2, 0, kernel_power_try, true, 0);
-  }
-
   device_param->kernel_threads = kernel_threads_sav;
 
-  device_param->spin_damp = spin_damp_sav;
-
   const double exec_msec_prev = get_avg_exec_time (device_param, 1);
 
   return exec_msec_prev;
 }
-*/
+
+static double try_run_times (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, const u32 kernel_accel, const u32 kernel_loops, const u32 kernel_threads, const int times)
+{
+  double exec_msec_best = try_run (hashcat_ctx, device_param, kernel_accel, kernel_loops, kernel_threads);
+
+  for (int i = 1; i < times; i++)
+  {
+    double exec_msec = try_run (hashcat_ctx, device_param, kernel_accel, kernel_loops, kernel_threads);
+
+    if (exec_msec > exec_msec_best) continue;
+
+    exec_msec_best = exec_msec;
+  }
+
+  return exec_msec_best;
+}
+
+static u32 previous_power_of_two (const u32 x)
+{
+  // https://stackoverflow.com/questions/2679815/previous-power-of-2
+  // really cool!
+
+  if (x == 0) return 0;
+
+  u32 r = x;
+
+  r |= (r >>  1);
+  r |= (r >>  2);
+  r |= (r >>  4);
+  r |= (r >>  8);
+  r |= (r >> 16);
+
+  return r - (r >> 1);
+}
 
 static int autotune (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param)
 {
@@ -124,9 +117,57 @@ static int autotune (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
   const u32 kernel_loops_min = device_param->kernel_loops_min;
   const u32 kernel_loops_max = device_param->kernel_loops_max;
 
+  const u32 kernel_threads_min = device_param->kernel_threads_min;
+  const u32 kernel_threads_max = device_param->kernel_threads_max;
+
   u32 kernel_accel = kernel_accel_min;
   u32 kernel_loops = kernel_loops_min;
 
+  // for the threads we take as initial value what we receive from the runtime
+  // but is only to start with something, we will fine tune this value as soon as we have our workload specified
+  // this thread limiting is also performed insinde run_kernel() so we need to redo it here, too
+
+  u32 kernel_wgs = 0;
+  u32 kernel_wgs_multiple = 0;
+
+  if (hashconfig->attack_exec == ATTACK_EXEC_INSIDE_KERNEL)
+  {
+    if (hashconfig->opti_type & OPTI_TYPE_OPTIMIZED_KERNEL)
+    {
+      kernel_wgs = device_param->kernel_wgs1;
+
+      kernel_wgs_multiple = device_param->kernel_preferred_wgs_multiple1;
+    }
+    else
+    {
+      kernel_wgs = device_param->kernel_wgs4;
+
+      kernel_wgs_multiple = device_param->kernel_preferred_wgs_multiple4;
+    }
+  }
+  else
+  {
+    kernel_wgs = device_param->kernel_wgs2;
+
+    kernel_wgs_multiple = device_param->kernel_preferred_wgs_multiple2;
+  }
+
+  u32 kernel_threads = kernel_threads_max;
+
+  if ((kernel_wgs >= kernel_threads_min) && (kernel_wgs <= kernel_threads_max))
+  {
+    kernel_threads = kernel_wgs;
+  }
+
+  // having a value power of 2 makes it easier to divide
+
+  const u32 kernel_threads_p2 = previous_power_of_two (kernel_threads);
+
+  if ((kernel_threads_p2 >= kernel_threads_min) && (kernel_threads_p2 <= kernel_threads_max))
+  {
+    kernel_threads = kernel_threads_p2;
+  }
+
   // in this case the user specified a fixed -n and -u on the commandline
   // no way to tune anything
   // but we need to run a few caching rounds
@@ -142,10 +183,10 @@ static int autotune (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
 
     if (hashconfig->warmup_disable == false)
     {
-      try_run (hashcat_ctx, device_param, kernel_accel, kernel_loops);
-      try_run (hashcat_ctx, device_param, kernel_accel, kernel_loops);
-      try_run (hashcat_ctx, device_param, kernel_accel, kernel_loops);
-      try_run (hashcat_ctx, device_param, kernel_accel, kernel_loops);
+      try_run (hashcat_ctx, device_param, kernel_accel, kernel_loops, kernel_threads);
+      try_run (hashcat_ctx, device_param, kernel_accel, kernel_loops, kernel_threads);
+      try_run (hashcat_ctx, device_param, kernel_accel, kernel_loops, kernel_threads);
+      try_run (hashcat_ctx, device_param, kernel_accel, kernel_loops, kernel_threads);
     }
 
     #endif
@@ -157,29 +198,19 @@ static int autotune (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
 
     const u32 kernel_power_max = device_param->hardware_power * kernel_accel_max;
 
-    int CU_rc;
-    int HIP_rc;
-    int CL_rc;
-
     if (device_param->is_cuda == true)
     {
-      CU_rc = run_cuda_kernel_atinit (hashcat_ctx, device_param, device_param->cuda_d_pws_buf, kernel_power_max);
-
-      if (CU_rc == -1) return -1;
+      if (run_cuda_kernel_atinit (hashcat_ctx, device_param, device_param->cuda_d_pws_buf, kernel_power_max) == -1) return -1;
     }
 
     if (device_param->is_hip == true)
     {
-      HIP_rc = run_hip_kernel_atinit (hashcat_ctx, device_param, device_param->hip_d_pws_buf, kernel_power_max);
-
-      if (HIP_rc == -1) return -1;
+      if (run_hip_kernel_atinit (hashcat_ctx, device_param, device_param->hip_d_pws_buf, kernel_power_max) == -1) return -1;
     }
 
     if (device_param->is_opencl == true)
     {
-      CL_rc = run_opencl_kernel_atinit (hashcat_ctx, device_param, device_param->opencl_d_pws_buf, kernel_power_max);
-
-      if (CL_rc == -1) return -1;
+      if (run_opencl_kernel_atinit (hashcat_ctx, device_param, device_param->opencl_d_pws_buf, kernel_power_max) == -1) return -1;
     }
 
     if (user_options->slow_candidates == true)
@@ -193,35 +224,53 @@ static int autotune (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
         {
           if (device_param->is_cuda == true)
           {
-            CU_rc = hc_cuMemcpyDtoD (hashcat_ctx, device_param->cuda_d_rules_c, device_param->cuda_d_rules, MIN (kernel_loops_max, KERNEL_RULES) * sizeof (kernel_rule_t));
-
-            if (CU_rc == -1) return -1;
+            if (hc_cuMemcpyDtoDAsync (hashcat_ctx, device_param->cuda_d_rules_c, device_param->cuda_d_rules, MIN (kernel_loops_max, KERNEL_RULES) * sizeof (kernel_rule_t), device_param->cuda_stream) == -1) return -1;
           }
 
           if (device_param->is_hip == true)
           {
-            HIP_rc = hc_hipMemcpyDtoD (hashcat_ctx, device_param->hip_d_rules_c, device_param->hip_d_rules, MIN (kernel_loops_max, KERNEL_RULES) * sizeof (kernel_rule_t));
-
-            if (HIP_rc == -1) return -1;
+            if (hc_hipMemcpyDtoDAsync (hashcat_ctx, device_param->hip_d_rules_c, device_param->hip_d_rules, MIN (kernel_loops_max, KERNEL_RULES) * sizeof (kernel_rule_t), device_param->hip_stream) == -1) return -1;
           }
 
           if (device_param->is_opencl == true)
           {
-            CL_rc = hc_clEnqueueCopyBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_rules, device_param->opencl_d_rules_c, 0, 0, MIN (kernel_loops_max, KERNEL_RULES) * sizeof (kernel_rule_t), 0, NULL, NULL);
-
-            if (CL_rc == -1) return -1;
+            if (hc_clEnqueueCopyBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_rules, device_param->opencl_d_rules_c, 0, 0, MIN (kernel_loops_max, KERNEL_RULES) * sizeof (kernel_rule_t), 0, NULL, NULL) == -1) return -1;
           }
         }
       }
     }
 
+    // we also need to initialize some values using kernels
+
+    if (hashconfig->attack_exec == ATTACK_EXEC_INSIDE_KERNEL)
+    {
+      // nothing to do
+    }
+    else
+    {
+      const u32 kernel_threads_sav = device_param->kernel_threads;
+
+      device_param->kernel_threads = device_param->kernel_wgs1;
+
+      run_kernel (hashcat_ctx, device_param, KERN_RUN_1, 0, kernel_power_max, false, 0);
+
+      if (hashconfig->opts_type & OPTS_TYPE_LOOP_PREPARE)
+      {
+        device_param->kernel_threads = device_param->kernel_wgs2p;
+
+        run_kernel (hashcat_ctx, device_param, KERN_RUN_2P, 0, kernel_power_max, false, 0);
+      }
+
+      device_param->kernel_threads = kernel_threads_sav;
+    }
+
     // Do a pre-autotune test run to find out if kernel runtime is above some TDR limit
 
     u32 kernel_loops_max_reduced = kernel_loops_max;
 
     if (true)
     {
-      double exec_msec = try_run (hashcat_ctx, device_param, kernel_accel_min, kernel_loops_min);
+      double exec_msec = try_run (hashcat_ctx, device_param, kernel_accel_min, kernel_loops_min, kernel_threads);
 
       if (exec_msec > 2000)
       {
@@ -230,7 +279,7 @@ static int autotune (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
         return -1;
       }
 
-      exec_msec = try_run (hashcat_ctx, device_param, kernel_accel_min, kernel_loops_min);
+      exec_msec = try_run (hashcat_ctx, device_param, kernel_accel_min, kernel_loops_min, kernel_threads);
 
       const u32 mm = kernel_loops_max / kernel_loops_min;
 
@@ -250,16 +299,16 @@ static int autotune (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
       {
         if (kernel_loops > kernel_loops_max_reduced) continue;
 
-        double exec_msec = try_run (hashcat_ctx, device_param, kernel_accel_min, kernel_loops);
+        double exec_msec = try_run_times (hashcat_ctx, device_param, kernel_accel_min, kernel_loops, kernel_threads, 1);
 
         if (exec_msec < target_msec) break;
       }
     }
 
-    // now the same for kernel-accel but with the new kernel-loops from previous loop set
-
     #define STEPS_CNT 16
 
+    // now the same for kernel-accel but with the new kernel-loops from previous loop set
+
     if (kernel_accel_min < kernel_accel_max)
     {
       for (int i = 0; i < STEPS_CNT; i++)
@@ -269,7 +318,7 @@ static int autotune (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
         if (kernel_accel_try < kernel_accel_min) continue;
         if (kernel_accel_try > kernel_accel_max) break;
 
-        double exec_msec = try_run (hashcat_ctx, device_param, kernel_accel_try, kernel_loops);
+        double exec_msec = try_run_times (hashcat_ctx, device_param, kernel_accel_try, kernel_loops, kernel_threads, 1);
 
         if (exec_msec > target_msec) break;
 
@@ -285,7 +334,7 @@ static int autotune (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
       const u32 kernel_accel_orig = kernel_accel;
       const u32 kernel_loops_orig = kernel_loops;
 
-      double exec_msec_prev = try_run (hashcat_ctx, device_param, kernel_accel, kernel_loops);
+      double exec_msec_prev = try_run_times (hashcat_ctx, device_param, kernel_accel, kernel_loops, kernel_threads, 1);
 
       for (int i = 1; i < STEPS_CNT; i++)
       {
@@ -300,7 +349,7 @@ static int autotune (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
 
         // do a real test
 
-        const double exec_msec = try_run (hashcat_ctx, device_param, kernel_accel_try, kernel_loops_try);
+        const double exec_msec = try_run_times (hashcat_ctx, device_param, kernel_accel_try, kernel_loops_try, kernel_threads, 1);
 
         if (exec_msec_prev < exec_msec) break;
 
@@ -317,7 +366,7 @@ static int autotune (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
       }
     }
 
-    double exec_msec_pre_final = try_run (hashcat_ctx, device_param, kernel_accel, kernel_loops);
+    double exec_msec_pre_final = try_run_times (hashcat_ctx, device_param, kernel_accel, kernel_loops, kernel_threads, 1);
 
     const u32 exec_left = (const u32) (target_msec / exec_msec_pre_final);
 
@@ -332,46 +381,43 @@ static int autotune (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
       kernel_accel *= exec_accel_min;
     }
 
-    // start finding best thread count is easier.
-    // it's either the preferred or the maximum thread count
+    // v6.2.4 new section: find thread count
+    // This is not as effective as it could be because of inaccurate kernel return timers
+    // But is better than fixed values
+    // Timers in this section are critical, so we rerun meassurements 3 times
 
-    /*
-    const u32 kernel_threads_min = device_param->kernel_threads_min;
-    const u32 kernel_threads_max = device_param->kernel_threads_max;
-
-    if (kernel_threads_min < kernel_threads_max)
+    if (kernel_threads_max > kernel_threads_min)
     {
-      const double exec_msec_max = try_run (hashcat_ctx, device_param, kernel_accel, kernel_loops);
+      const u32 kernel_accel_orig   = kernel_accel;
+      const u32 kernel_threads_orig = kernel_threads;
 
-      u32 preferred_threads = 0;
+      double exec_msec_prev = try_run_times (hashcat_ctx, device_param, kernel_accel, kernel_loops, kernel_threads, 3);
 
-      if (hashconfig->attack_exec == ATTACK_EXEC_INSIDE_KERNEL)
+      for (int i = 1; i < STEPS_CNT; i++)
       {
-        if (hashconfig->opti_type & OPTI_TYPE_OPTIMIZED_KERNEL)
-        {
-          preferred_threads = device_param->kernel_preferred_wgs_multiple1;
-        }
-        else
-        {
-          preferred_threads = device_param->kernel_preferred_wgs_multiple4;
-        }
-      }
-      else
-      {
-        preferred_threads = device_param->kernel_preferred_wgs_multiple2;
-      }
+        const u32 kernel_accel_try   = kernel_accel_orig   * (1U << i);
+        const u32 kernel_threads_try = kernel_threads_orig / (1U << i);
 
-      if ((preferred_threads >= kernel_threads_min) && (preferred_threads <= kernel_threads_max))
-      {
-        const double exec_msec_preferred = try_run_preferred (hashcat_ctx, device_param, kernel_accel, kernel_loops);
+        // since we do not modify total amount of workitems, we can (and need) to do increase kernel_accel_max
 
-        if (exec_msec_preferred < exec_msec_max)
-        {
-          device_param->kernel_threads = preferred_threads;
-        }
+        const u32 kernel_accel_max_try = kernel_accel_max * (1U << i);
+
+        if (kernel_accel_try > kernel_accel_max_try) break;
+
+        if (kernel_threads_try < kernel_threads_min) break;
+
+        if (kernel_threads_try % kernel_wgs_multiple) break; // this would just be waste of time
+
+        double exec_msec = try_run_times (hashcat_ctx, device_param, kernel_accel_try, kernel_loops, kernel_threads_try, 3);
+
+        if (exec_msec > exec_msec_prev) continue;
+
+        exec_msec_prev = exec_msec;
+
+        kernel_accel   = kernel_accel_try;
+        kernel_threads = kernel_threads_try;
       }
     }
-    */
   }
 
   // reset them fake words
@@ -379,65 +425,43 @@ static int autotune (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
 
   if (device_param->is_cuda == true)
   {
-    int CU_rc;
+    if (run_cuda_kernel_bzero (hashcat_ctx, device_param, device_param->cuda_d_pws_buf, device_param->size_pws) == -1) return -1;
 
-    CU_rc = run_cuda_kernel_memset (hashcat_ctx, device_param, device_param->cuda_d_pws_buf, 0, device_param->size_pws);
+    if (run_cuda_kernel_bzero (hashcat_ctx, device_param, device_param->cuda_d_plain_bufs, device_param->size_plains) == -1) return -1;
 
-    if (CU_rc == -1) return -1;
+    if (run_cuda_kernel_bzero (hashcat_ctx, device_param, device_param->cuda_d_digests_shown, device_param->size_shown) == -1) return -1;
 
-    CU_rc = run_cuda_kernel_memset (hashcat_ctx, device_param, device_param->cuda_d_plain_bufs, 0, device_param->size_plains);
+    if (run_cuda_kernel_bzero (hashcat_ctx, device_param, device_param->cuda_d_result, device_param->size_results) == -1) return -1;
 
-    if (CU_rc == -1) return -1;
-
-    CU_rc = run_cuda_kernel_memset (hashcat_ctx, device_param, device_param->cuda_d_digests_shown, 0, device_param->size_shown);
-
-    if (CU_rc == -1) return -1;
-
-    CU_rc = run_cuda_kernel_memset (hashcat_ctx, device_param, device_param->cuda_d_result, 0, device_param->size_results);
-
-    if (CU_rc == -1) return -1;
+    if (run_cuda_kernel_bzero (hashcat_ctx, device_param, device_param->cuda_d_tmps, device_param->size_tmps) == -1) return -1;
   }
 
   if (device_param->is_hip == true)
   {
-    int HIP_rc;
+    if (run_hip_kernel_bzero (hashcat_ctx, device_param, device_param->hip_d_pws_buf, device_param->size_pws) == -1) return -1;
 
-    HIP_rc = run_hip_kernel_memset (hashcat_ctx, device_param, device_param->hip_d_pws_buf, 0, device_param->size_pws);
+    if (run_hip_kernel_bzero (hashcat_ctx, device_param, device_param->hip_d_plain_bufs, device_param->size_plains) == -1) return -1;
 
-    if (HIP_rc == -1) return -1;
+    if (run_hip_kernel_bzero (hashcat_ctx, device_param, device_param->hip_d_digests_shown, device_param->size_shown) == -1) return -1;
 
-    HIP_rc = run_hip_kernel_memset (hashcat_ctx, device_param, device_param->hip_d_plain_bufs, 0, device_param->size_plains);
+    if (run_hip_kernel_bzero (hashcat_ctx, device_param, device_param->hip_d_result, device_param->size_results) == -1) return -1;
 
-    if (HIP_rc == -1) return -1;
-
-    HIP_rc = run_hip_kernel_memset (hashcat_ctx, device_param, device_param->hip_d_digests_shown, 0, device_param->size_shown);
-
-    if (HIP_rc == -1) return -1;
-
-    HIP_rc = run_hip_kernel_memset (hashcat_ctx, device_param, device_param->hip_d_result, 0, device_param->size_results);
-
-    if (HIP_rc == -1) return -1;
+    if (run_hip_kernel_bzero (hashcat_ctx, device_param, device_param->hip_d_tmps, device_param->size_tmps) == -1) return -1;
   }
 
   if (device_param->is_opencl == true)
   {
-    int CL_rc;
+    if (run_opencl_kernel_bzero (hashcat_ctx, device_param, device_param->opencl_d_pws_buf, device_param->size_pws) == -1) return -1;
 
-    CL_rc = run_opencl_kernel_memset (hashcat_ctx, device_param, device_param->opencl_d_pws_buf, 0, device_param->size_pws);
+    if (run_opencl_kernel_bzero (hashcat_ctx, device_param, device_param->opencl_d_plain_bufs, device_param->size_plains) == -1) return -1;
 
-    if (CL_rc == -1) return -1;
+    if (run_opencl_kernel_bzero (hashcat_ctx, device_param, device_param->opencl_d_digests_shown, device_param->size_shown) == -1) return -1;
 
-    CL_rc = run_opencl_kernel_memset (hashcat_ctx, device_param, device_param->opencl_d_plain_bufs, 0, device_param->size_plains);
+    if (run_opencl_kernel_bzero (hashcat_ctx, device_param, device_param->opencl_d_result, device_param->size_results) == -1) return -1;
 
-    if (CL_rc == -1) return -1;
+    if (run_opencl_kernel_bzero (hashcat_ctx, device_param, device_param->opencl_d_tmps, device_param->size_tmps) == -1) return -1;
 
-    CL_rc = run_opencl_kernel_memset (hashcat_ctx, device_param, device_param->opencl_d_digests_shown, 0, device_param->size_shown);
-
-    if (CL_rc == -1) return -1;
-
-    CL_rc = run_opencl_kernel_memset (hashcat_ctx, device_param, device_param->opencl_d_result, 0, device_param->size_results);
-
-    if (CL_rc == -1) return -1;
+    if (hc_clFlush (hashcat_ctx, device_param->opencl_command_queue) == -1) return -1;
   }
 
   // reset timer
@@ -459,8 +483,13 @@ static int autotune (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
 
   // store
 
-  device_param->kernel_accel = kernel_accel;
-  device_param->kernel_loops = kernel_loops;
+  device_param->kernel_accel   = kernel_accel;
+  device_param->kernel_loops   = kernel_loops;
+  device_param->kernel_threads = kernel_threads;
+
+  const u32 hardware_power = ((hashconfig->opts_type & OPTS_TYPE_MP_MULTI_DISABLE) ? 1 : device_param->device_processors) * device_param->kernel_threads;
+
+  device_param->hardware_power = hardware_power;
 
   const u32 kernel_power = device_param->hardware_power * device_param->kernel_accel;
 
diff --git a/src/backend.c b/src/backend.c
index 4caff74ce..3f1a91362 100644
--- a/src/backend.c
+++ b/src/backend.c
@@ -26,10 +26,10 @@
 #include "terminal.h"
 
 #if defined (__linux__)
-static const char *dri_card0_path = "/dev/dri/card0";
+static const char *const  dri_card0_path = "/dev/dri/card0";
 
-static const char *drm_card0_vendor_path = "/sys/class/drm/card0/device/vendor";
-static const char *drm_card0_driver_path = "/sys/class/drm/card0/device/driver";
+static const char *const  drm_card0_vendor_path = "/sys/class/drm/card0/device/vendor";
+static const char *const  drm_card0_driver_path = "/sys/class/drm/card0/device/driver";
 #endif
 
 static const u32 full01 = 0x01010101;
@@ -38,6 +38,9 @@ static const u32 full80 = 0x80808080;
 
 static double TARGET_MSEC_PROFILE[4] = { 2, 12, 96, 480 };
 
+HC_ALIGN(16)
+static const u32 bzeros[4] = { 0, 0, 0, 0 };
+
 static bool is_same_device (const hc_device_param_t *src, const hc_device_param_t *dst)
 {
   // First check by PCI address
@@ -980,11 +983,11 @@ int hiprtc_init (hashcat_ctx_t *hashcat_ctx)
   memset (hiprtc, 0, sizeof (HIPRTC_PTR));
 
   #if   defined (_WIN)
-  hiprtc->lib = hc_dlopen ("fixme.dll");
+  hiprtc->lib = hc_dlopen ("amdhip64.dll");
   #elif defined (__APPLE__)
   hiprtc->lib = hc_dlopen ("fixme.dylib");
   #elif defined (__CYGWIN__)
-  hiprtc->lib = hc_dlopen ("fixme.dll");
+  hiprtc->lib = hc_dlopen ("amdhip64.dll");
   #else
   hiprtc->lib = hc_dlopen ("libamdhip64.so");
 
@@ -998,12 +1001,11 @@ int hiprtc_init (hashcat_ctx_t *hashcat_ctx)
   HC_LOAD_FUNC (hiprtc, hiprtcCreateProgram,      HIPRTC_HIPRTCCREATEPROGRAM,     HIPRTC, 1);
   HC_LOAD_FUNC (hiprtc, hiprtcDestroyProgram,     HIPRTC_HIPRTCDESTROYPROGRAM,    HIPRTC, 1);
   HC_LOAD_FUNC (hiprtc, hiprtcGetLoweredName,     HIPRTC_HIPRTCGETLOWEREDNAME,    HIPRTC, 1);
-  HC_LOAD_FUNC (hiprtc, hiprtcGetCode,            HIPRTC_HIPRTCGETPTX,            HIPRTC, 1);
-  HC_LOAD_FUNC (hiprtc, hiprtcGetCodeSize,        HIPRTC_HIPRTCGETPTXSIZE,        HIPRTC, 1);
+  HC_LOAD_FUNC (hiprtc, hiprtcGetCode,            HIPRTC_HIPRTCGETCODE,           HIPRTC, 1);
+  HC_LOAD_FUNC (hiprtc, hiprtcGetCodeSize,        HIPRTC_HIPRTCGETCODESIZE,       HIPRTC, 1);
   HC_LOAD_FUNC (hiprtc, hiprtcGetProgramLog,      HIPRTC_HIPRTCGETPROGRAMLOG,     HIPRTC, 1);
   HC_LOAD_FUNC (hiprtc, hiprtcGetProgramLogSize,  HIPRTC_HIPRTCGETPROGRAMLOGSIZE, HIPRTC, 1);
   HC_LOAD_FUNC (hiprtc, hiprtcGetErrorString,     HIPRTC_HIPRTCGETERRORSTRING,    HIPRTC, 1);
-  HC_LOAD_FUNC (hiprtc, hiprtcVersion,            HIPRTC_HIPRTCVERSION,           HIPRTC, 1);
 
   return 0;
 }
@@ -1069,11 +1071,6 @@ int hc_hiprtcCompileProgram (hashcat_ctx_t *hashcat_ctx, hiprtcProgram prog, int
 
   HIPRTC_PTR *hiprtc = (HIPRTC_PTR *) backend_ctx->hiprtc;
 
-  #if 0
-  for(int i =0; i< numOptions; i++)
-    printf("Option_%d = %s\n", i, options[i]);
-  #endif
-
   const hiprtcResult HIPRTC_err = hiprtc->hiprtcCompileProgram (prog, numOptions, options);
 
   if (HIPRTC_err != HIPRTC_SUCCESS)
@@ -1122,13 +1119,13 @@ int hc_hiprtcGetProgramLog (hashcat_ctx_t *hashcat_ctx, hiprtcProgram prog, char
   return 0;
 }
 
-int hc_hiprtcGetCodeSize (hashcat_ctx_t *hashcat_ctx, hiprtcProgram prog, size_t *ptxSizeRet)
+int hc_hiprtcGetCodeSize (hashcat_ctx_t *hashcat_ctx, hiprtcProgram prog, size_t *codeSizeRet)
 {
   backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
 
   HIPRTC_PTR *hiprtc = (HIPRTC_PTR *) backend_ctx->hiprtc;
 
-  const hiprtcResult HIPRTC_err = hiprtc->hiprtcGetCodeSize (prog, ptxSizeRet);
+  const hiprtcResult HIPRTC_err = hiprtc->hiprtcGetCodeSize (prog, codeSizeRet);
 
   if (HIPRTC_err != HIPRTC_SUCCESS)
   {
@@ -1140,13 +1137,13 @@ int hc_hiprtcGetCodeSize (hashcat_ctx_t *hashcat_ctx, hiprtcProgram prog, size_t
   return 0;
 }
 
-int hc_hiprtcGetCode (hashcat_ctx_t *hashcat_ctx, hiprtcProgram prog, char *ptx)
+int hc_hiprtcGetCode (hashcat_ctx_t *hashcat_ctx, hiprtcProgram prog, char *code)
 {
   backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
 
   HIPRTC_PTR *hiprtc = (HIPRTC_PTR *) backend_ctx->hiprtc;
 
-  const hiprtcResult HIPRTC_err = hiprtc->hiprtcGetCode (prog, ptx);
+  const hiprtcResult HIPRTC_err = hiprtc->hiprtcGetCode (prog, code);
 
   if (HIPRTC_err != HIPRTC_SUCCESS)
   {
@@ -1158,24 +1155,6 @@ int hc_hiprtcGetCode (hashcat_ctx_t *hashcat_ctx, hiprtcProgram prog, char *ptx)
   return 0;
 }
 
-int hc_hiprtcVersion (hashcat_ctx_t *hashcat_ctx, int *major, int *minor)
-{
-  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
-
-  HIPRTC_PTR *hiprtc = (HIPRTC_PTR *) backend_ctx->hiprtc;
-
-  const hiprtcResult HIPRTC_err = hiprtc->hiprtcVersion (major, minor);
-
-  if (HIPRTC_err != HIPRTC_SUCCESS)
-  {
-    event_log_error (hashcat_ctx, "hiprtcVersion(): %s", hiprtc->hiprtcGetErrorString (HIPRTC_err));
-
-    return -1;
-  }
-
-  return 0;
-}
-
 // CUDA
 
 int cuda_init (hashcat_ctx_t *hashcat_ctx)
@@ -1254,14 +1233,14 @@ int cuda_init (hashcat_ctx_t *hashcat_ctx)
   HC_LOAD_FUNC_CUDA (cuda, cuLaunchKernel,           cuLaunchKernel,            CUDA_CULAUNCHKERNEL,            CUDA, 1);
   HC_LOAD_FUNC_CUDA (cuda, cuMemAlloc,               cuMemAlloc_v2,             CUDA_CUMEMALLOC,                CUDA, 1);
   HC_LOAD_FUNC_CUDA (cuda, cuMemAllocHost,           cuMemAllocHost_v2,         CUDA_CUMEMALLOCHOST,            CUDA, 1);
-  HC_LOAD_FUNC_CUDA (cuda, cuMemcpyDtoD,             cuMemcpyDtoD_v2,           CUDA_CUMEMCPYDTOD,              CUDA, 1);
-  HC_LOAD_FUNC_CUDA (cuda, cuMemcpyDtoH,             cuMemcpyDtoH_v2,           CUDA_CUMEMCPYDTOH,              CUDA, 1);
-  HC_LOAD_FUNC_CUDA (cuda, cuMemcpyHtoD,             cuMemcpyHtoD_v2,           CUDA_CUMEMCPYHTOD,              CUDA, 1);
+  HC_LOAD_FUNC_CUDA (cuda, cuMemcpyDtoDAsync,        cuMemcpyDtoDAsync_v2,      CUDA_CUMEMCPYDTODASYNC,         CUDA, 1);
+  HC_LOAD_FUNC_CUDA (cuda, cuMemcpyDtoHAsync,        cuMemcpyDtoHAsync_v2,      CUDA_CUMEMCPYDTOHASYNC,         CUDA, 1);
+  HC_LOAD_FUNC_CUDA (cuda, cuMemcpyHtoDAsync,        cuMemcpyHtoDAsync_v2,      CUDA_CUMEMCPYHTODASYNC,         CUDA, 1);
   HC_LOAD_FUNC_CUDA (cuda, cuMemFree,                cuMemFree_v2,              CUDA_CUMEMFREE,                 CUDA, 1);
   HC_LOAD_FUNC_CUDA (cuda, cuMemFreeHost,            cuMemFreeHost,             CUDA_CUMEMFREEHOST,             CUDA, 1);
   HC_LOAD_FUNC_CUDA (cuda, cuMemGetInfo,             cuMemGetInfo_v2,           CUDA_CUMEMGETINFO,              CUDA, 1);
-  HC_LOAD_FUNC_CUDA (cuda, cuMemsetD32,              cuMemsetD32_v2,            CUDA_CUMEMSETD32,               CUDA, 1);
-  HC_LOAD_FUNC_CUDA (cuda, cuMemsetD8,               cuMemsetD8_v2,             CUDA_CUMEMSETD8,                CUDA, 1);
+  HC_LOAD_FUNC_CUDA (cuda, cuMemsetD32Async,         cuMemsetD32Async,          CUDA_CUMEMSETD32ASYNC,          CUDA, 1);
+  HC_LOAD_FUNC_CUDA (cuda, cuMemsetD8Async,          cuMemsetD8Async,           CUDA_CUMEMSETD8ASYNC,           CUDA, 1);
   HC_LOAD_FUNC_CUDA (cuda, cuModuleGetFunction,      cuModuleGetFunction,       CUDA_CUMODULEGETFUNCTION,       CUDA, 1);
   HC_LOAD_FUNC_CUDA (cuda, cuModuleGetGlobal,        cuModuleGetGlobal_v2,      CUDA_CUMODULEGETGLOBAL,         CUDA, 1);
   HC_LOAD_FUNC_CUDA (cuda, cuModuleLoad,             cuModuleLoad,              CUDA_CUMODULELOAD,              CUDA, 1);
@@ -1681,13 +1660,13 @@ int hc_cuMemFree (hashcat_ctx_t *hashcat_ctx, CUdeviceptr dptr)
   return 0;
 }
 
-int hc_cuMemcpyDtoH (hashcat_ctx_t *hashcat_ctx, void *dstHost, CUdeviceptr srcDevice, size_t ByteCount)
+int hc_cuMemcpyDtoHAsync (hashcat_ctx_t *hashcat_ctx, void *dstHost, CUdeviceptr srcDevice, size_t ByteCount, CUstream hStream)
 {
   backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
 
   CUDA_PTR *cuda = (CUDA_PTR *) backend_ctx->cuda;
 
-  const CUresult CU_err = cuda->cuMemcpyDtoH (dstHost, srcDevice, ByteCount);
+  const CUresult CU_err = cuda->cuMemcpyDtoHAsync (dstHost, srcDevice, ByteCount, hStream);
 
   if (CU_err != CUDA_SUCCESS)
   {
@@ -1695,11 +1674,11 @@ int hc_cuMemcpyDtoH (hashcat_ctx_t *hashcat_ctx, void *dstHost, CUdeviceptr srcD
 
     if (cuda->cuGetErrorString (CU_err, &pStr) == CUDA_SUCCESS)
     {
-      event_log_error (hashcat_ctx, "cuMemcpyDtoH(): %s", pStr);
+      event_log_error (hashcat_ctx, "cuMemcpyDtoHAsync(): %s", pStr);
     }
     else
     {
-      event_log_error (hashcat_ctx, "cuMemcpyDtoH(): %d", CU_err);
+      event_log_error (hashcat_ctx, "cuMemcpyDtoHAsync(): %d", CU_err);
     }
 
     return -1;
@@ -1708,13 +1687,13 @@ int hc_cuMemcpyDtoH (hashcat_ctx_t *hashcat_ctx, void *dstHost, CUdeviceptr srcD
   return 0;
 }
 
-int hc_cuMemcpyDtoD (hashcat_ctx_t *hashcat_ctx, CUdeviceptr dstDevice, CUdeviceptr srcDevice, size_t ByteCount)
+int hc_cuMemcpyDtoDAsync (hashcat_ctx_t *hashcat_ctx, CUdeviceptr dstDevice, CUdeviceptr srcDevice, size_t ByteCount, CUstream hStream)
 {
   backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
 
   CUDA_PTR *cuda = (CUDA_PTR *) backend_ctx->cuda;
 
-  const CUresult CU_err = cuda->cuMemcpyDtoD (dstDevice, srcDevice, ByteCount);
+  const CUresult CU_err = cuda->cuMemcpyDtoDAsync (dstDevice, srcDevice, ByteCount, hStream);
 
   if (CU_err != CUDA_SUCCESS)
   {
@@ -1722,11 +1701,11 @@ int hc_cuMemcpyDtoD (hashcat_ctx_t *hashcat_ctx, CUdeviceptr dstDevice, CUdevice
 
     if (cuda->cuGetErrorString (CU_err, &pStr) == CUDA_SUCCESS)
     {
-      event_log_error (hashcat_ctx, "cuMemcpyDtoD(): %s", pStr);
+      event_log_error (hashcat_ctx, "cuMemcpyDtoDAsync(): %s", pStr);
     }
     else
     {
-      event_log_error (hashcat_ctx, "cuMemcpyDtoD(): %d", CU_err);
+      event_log_error (hashcat_ctx, "cuMemcpyDtoDAsync(): %d", CU_err);
     }
 
     return -1;
@@ -1735,13 +1714,13 @@ int hc_cuMemcpyDtoD (hashcat_ctx_t *hashcat_ctx, CUdeviceptr dstDevice, CUdevice
   return 0;
 }
 
-int hc_cuMemcpyHtoD (hashcat_ctx_t *hashcat_ctx, CUdeviceptr dstDevice, const void *srcHost, size_t ByteCount)
+int hc_cuMemcpyHtoDAsync (hashcat_ctx_t *hashcat_ctx, CUdeviceptr dstDevice, const void *srcHost, size_t ByteCount, CUstream hStream)
 {
   backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
 
   CUDA_PTR *cuda = (CUDA_PTR *) backend_ctx->cuda;
 
-  const CUresult CU_err = cuda->cuMemcpyHtoD (dstDevice, srcHost, ByteCount);
+  const CUresult CU_err = cuda->cuMemcpyHtoDAsync (dstDevice, srcHost, ByteCount, hStream);
 
   if (CU_err != CUDA_SUCCESS)
   {
@@ -1749,11 +1728,65 @@ int hc_cuMemcpyHtoD (hashcat_ctx_t *hashcat_ctx, CUdeviceptr dstDevice, const vo
 
     if (cuda->cuGetErrorString (CU_err, &pStr) == CUDA_SUCCESS)
     {
-      event_log_error (hashcat_ctx, "cuMemcpyHtoD(): %s", pStr);
+      event_log_error (hashcat_ctx, "cuMemcpyHtoDAsync(): %s", pStr);
     }
     else
     {
-      event_log_error (hashcat_ctx, "cuMemcpyHtoD(): %d", CU_err);
+      event_log_error (hashcat_ctx, "cuMemcpyHtoDAsync(): %d", CU_err);
+    }
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hc_cuMemsetD32Async (hashcat_ctx_t *hashcat_ctx, CUdeviceptr dstDevice, unsigned int ui, size_t N, CUstream hStream)
+{
+  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
+
+  CUDA_PTR *cuda = (CUDA_PTR *) backend_ctx->cuda;
+
+  const CUresult CU_err = cuda->cuMemsetD32Async (dstDevice, ui, N, hStream);
+
+  if (CU_err != CUDA_SUCCESS)
+  {
+    const char *pStr = NULL;
+
+    if (cuda->cuGetErrorString (CU_err, &pStr) == CUDA_SUCCESS)
+    {
+      event_log_error (hashcat_ctx, "cuMemsetD32Async(): %s", pStr);
+    }
+    else
+    {
+      event_log_error (hashcat_ctx, "cuMemsetD32Async(): %d", CU_err);
+    }
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hc_cuMemsetD8Async (hashcat_ctx_t *hashcat_ctx, CUdeviceptr dstDevice, unsigned char uc, size_t N, CUstream hStream)
+{
+  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
+
+  CUDA_PTR *cuda = (CUDA_PTR *) backend_ctx->cuda;
+
+  const CUresult CU_err = cuda->cuMemsetD8Async (dstDevice, uc, N, hStream);
+
+  if (CU_err != CUDA_SUCCESS)
+  {
+    const char *pStr = NULL;
+
+    if (cuda->cuGetErrorString (CU_err, &pStr) == CUDA_SUCCESS)
+    {
+      event_log_error (hashcat_ctx, "cuMemsetD8Async(): %s", pStr);
+    }
+    else
+    {
+      event_log_error (hashcat_ctx, "cuMemsetD8Async(): %d", CU_err);
     }
 
     return -1;
@@ -2394,16 +2427,13 @@ int hip_init (hashcat_ctx_t *hashcat_ctx)
   memset (hip, 0, sizeof (HIP_PTR));
 
   #if   defined (_WIN)
-  hip->lib = hc_dlopen ("fixme.dll");
+  hip->lib = hc_dlopen ("amdhip64.dll");
   #elif defined (__APPLE__)
   hip->lib = hc_dlopen ("fixme.dylib");
   #elif defined (__CYGWIN__)
-  hip->lib = hc_dlopen ("fixme.dll");
+  hip->lib = hc_dlopen ("amdhip64.dll");
   #else
   hip->lib = hc_dlopen ("libamdhip64.so");
-
-  //TODO: grab the 4 from the major RT version
-  if (hip->lib == NULL) hip->lib = hc_dlopen ("libamdhip64.so.4.2.40200");
   #endif
 
   if (hip->lib == NULL) return -1;
@@ -2432,63 +2462,43 @@ int hip_init (hashcat_ctx_t *hashcat_ctx)
 
   HC_LOAD_FUNC_HIP (hip, hipCtxCreate,              hipCtxCreate,               HIP_HIPCTXCREATE,               HIP, 1);
   HC_LOAD_FUNC_HIP (hip, hipCtxDestroy,             hipCtxDestroy,              HIP_HIPCTXDESTROY,              HIP, 1);
-  HC_LOAD_FUNC_HIP (hip, hipCtxGetCacheConfig,      hipCtxGetCacheConfig,       HIP_HIPCTXGETCACHECONFIG,       HIP, 1);
-  HC_LOAD_FUNC_HIP (hip, hipCtxGetCurrent,          hipCtxGetCurrent,           HIP_HIPCTXGETCURRENT,           HIP, 1);
-  HC_LOAD_FUNC_HIP (hip, hipCtxGetSharedMemConfig,  hipCtxGetSharedMemConfig,   HIP_HIPCTXGETSHAREDMEMCONFIG,   HIP, 1);
   HC_LOAD_FUNC_HIP (hip, hipCtxPopCurrent,          hipCtxPopCurrent,           HIP_HIPCTXPOPCURRENT,           HIP, 1);
   HC_LOAD_FUNC_HIP (hip, hipCtxPushCurrent,         hipCtxPushCurrent,          HIP_HIPCTXPUSHCURRENT,          HIP, 1);
-  HC_LOAD_FUNC_HIP (hip, hipCtxSetCacheConfig,      hipCtxSetCacheConfig,       HIP_HIPCTXSETCACHECONFIG,       HIP, 1);
   HC_LOAD_FUNC_HIP (hip, hipCtxSetCurrent,          hipCtxSetCurrent,           HIP_HIPCTXSETCURRENT,           HIP, 1);
-  HC_LOAD_FUNC_HIP (hip, hipCtxSetSharedMemConfig,  hipCtxSetSharedMemConfig,   HIP_HIPCTXSETSHAREDMEMCONFIG,   HIP, 1);
   HC_LOAD_FUNC_HIP (hip, hipCtxSynchronize,         hipCtxSynchronize,          HIP_HIPCTXSYNCHRONIZE,          HIP, 1);
+  HC_LOAD_FUNC_HIP (hip, hipDeviceGet,              hipDeviceGet,               HIP_HIPDEVICEGET,               HIP, 1);
   HC_LOAD_FUNC_HIP (hip, hipDeviceGetAttribute,     hipDeviceGetAttribute,      HIP_HIPDEVICEGETATTRIBUTE,      HIP, 1);
   HC_LOAD_FUNC_HIP (hip, hipDeviceGetCount,         hipGetDeviceCount,          HIP_HIPDEVICEGETCOUNT,          HIP, 1);
-  HC_LOAD_FUNC_HIP (hip, hipDeviceGet,              hipDeviceGet,               HIP_HIPDEVICEGET,               HIP, 1);
   HC_LOAD_FUNC_HIP (hip, hipDeviceGetName,          hipDeviceGetName,           HIP_HIPDEVICEGETNAME,           HIP, 1);
   HC_LOAD_FUNC_HIP (hip, hipDeviceTotalMem,         hipDeviceTotalMem,          HIP_HIPDEVICETOTALMEM,          HIP, 1);
   HC_LOAD_FUNC_HIP (hip, hipDriverGetVersion,       hipDriverGetVersion,        HIP_HIPDRIVERGETVERSION,        HIP, 1);
   HC_LOAD_FUNC_HIP (hip, hipEventCreate,            hipEventCreateWithFlags,    HIP_HIPEVENTCREATE,             HIP, 1);
   HC_LOAD_FUNC_HIP (hip, hipEventDestroy,           hipEventDestroy,            HIP_HIPEVENTDESTROY,            HIP, 1);
   HC_LOAD_FUNC_HIP (hip, hipEventElapsedTime,       hipEventElapsedTime,        HIP_HIPEVENTELAPSEDTIME,        HIP, 1);
-  HC_LOAD_FUNC_HIP (hip, hipEventQuery,             hipEventQuery,              HIP_HIPEVENTQUERY,              HIP, 1);
   HC_LOAD_FUNC_HIP (hip, hipEventRecord,            hipEventRecord,             HIP_HIPEVENTRECORD,             HIP, 1);
   HC_LOAD_FUNC_HIP (hip, hipEventSynchronize,       hipEventSynchronize,        HIP_HIPEVENTSYNCHRONIZE,        HIP, 1);
   HC_LOAD_FUNC_HIP (hip, hipFuncGetAttribute,       hipFuncGetAttribute,        HIP_HIPFUNCGETATTRIBUTE,        HIP, 1);
-  //HC_LOAD_FUNC_HIP (hip, hipFuncSetAttribute,       hipFuncSetAttribute,        HIP_HIPFUNCSETATTRIBUTE,        HIP, 1);
-  HC_LOAD_FUNC_HIP (hip, hipFuncSetCacheConfig,     hipFuncSetCacheConfig,      HIP_HIPFUNCSETCACHECONFIG,      HIP, 1);
-  HC_LOAD_FUNC_HIP (hip, hipFuncSetSharedMemConfig, hipFuncSetSharedMemConfig,  HIP_HIPFUNCSETSHAREDMEMCONFIG,  HIP, 1);
   HC_LOAD_FUNC_HIP (hip, hipGetErrorName,           hipGetErrorName,            HIP_HIPGETERRORNAME,            HIP, 1);
   HC_LOAD_FUNC_HIP (hip, hipGetErrorString,         hipGetErrorString,          HIP_HIPGETERRORSTRING,          HIP, 1);
   HC_LOAD_FUNC_HIP (hip, hipInit,                   hipInit,                    HIP_HIPINIT,                    HIP, 1);
   HC_LOAD_FUNC_HIP (hip, hipLaunchKernel,           hipModuleLaunchKernel,      HIP_HIPLAUNCHKERNEL,            HIP, 1);
   HC_LOAD_FUNC_HIP (hip, hipMemAlloc,               hipMalloc,                  HIP_HIPMEMALLOC,                HIP, 1);
-  HC_LOAD_FUNC_HIP (hip, hipMemAllocHost,           hipMemAllocHost,            HIP_HIPMEMALLOCHOST,            HIP, 1);
-  HC_LOAD_FUNC_HIP (hip, hipMemcpyDtoD,             hipMemcpyDtoD,              HIP_HIPMEMCPYDTOD,              HIP, 1);
-  HC_LOAD_FUNC_HIP (hip, hipMemcpyDtoH,             hipMemcpyDtoH,              HIP_HIPMEMCPYDTOH,              HIP, 1);
-  HC_LOAD_FUNC_HIP (hip, hipMemcpyHtoD,             hipMemcpyHtoD,              HIP_HIPMEMCPYHTOD,              HIP, 1);
   HC_LOAD_FUNC_HIP (hip, hipMemFree,                hipFree,                    HIP_HIPMEMFREE,                 HIP, 1);
-  HC_LOAD_FUNC_HIP (hip, hipMemFreeHost,            hipFreeHost,                HIP_HIPMEMFREEHOST,             HIP, 1);
   HC_LOAD_FUNC_HIP (hip, hipMemGetInfo,             hipMemGetInfo,              HIP_HIPMEMGETINFO,              HIP, 1);
-  HC_LOAD_FUNC_HIP (hip, hipMemsetD32,              hipMemsetD32,               HIP_HIPMEMSETD32,               HIP, 1);
-  HC_LOAD_FUNC_HIP (hip, hipMemsetD8,               hipMemsetD8,                HIP_HIPMEMSETD8,                HIP, 1);
+  HC_LOAD_FUNC_HIP (hip, hipMemcpyDtoDAsync,        hipMemcpyDtoDAsync,         HIP_HIPMEMCPYDTODASYNC,         HIP, 1);
+  HC_LOAD_FUNC_HIP (hip, hipMemcpyDtoHAsync,        hipMemcpyDtoHAsync,         HIP_HIPMEMCPYDTOHASYNC,         HIP, 1);
+  HC_LOAD_FUNC_HIP (hip, hipMemcpyHtoDAsync,        hipMemcpyHtoDAsync,         HIP_HIPMEMCPYHTODASYNC,         HIP, 1);
+  HC_LOAD_FUNC_HIP (hip, hipMemsetD32Async,         hipMemsetD32Async,          HIP_HIPMEMSETD32ASYNC,          HIP, 1);
+  HC_LOAD_FUNC_HIP (hip, hipMemsetD8Async,          hipMemsetD8Async,           HIP_HIPMEMSETD8ASYNC,           HIP, 1);
+  HC_LOAD_FUNC_HIP (hip, hipMemcpyHtoDAsync,        hipMemcpyHtoDAsync,         HIP_HIPMEMCPYHTODASYNC,         HIP, 1);
   HC_LOAD_FUNC_HIP (hip, hipModuleGetFunction,      hipModuleGetFunction,       HIP_HIPMODULEGETFUNCTION,       HIP, 1);
   HC_LOAD_FUNC_HIP (hip, hipModuleGetGlobal,        hipModuleGetGlobal,         HIP_HIPMODULEGETGLOBAL,         HIP, 1);
-  HC_LOAD_FUNC_HIP (hip, hipModuleLoad,             hipModuleLoad,              HIP_HIPMODULELOAD,              HIP, 1);
-  HC_LOAD_FUNC_HIP (hip, hipModuleLoadData,         hipModuleLoadData,          HIP_HIPMODULELOADDATA,          HIP, 1);
   HC_LOAD_FUNC_HIP (hip, hipModuleLoadDataEx,       hipModuleLoadDataEx,        HIP_HIPMODULELOADDATAEX,        HIP, 1);
   HC_LOAD_FUNC_HIP (hip, hipModuleUnload,           hipModuleUnload,            HIP_HIPMODULEUNLOAD,            HIP, 1);
-  HC_LOAD_FUNC_HIP (hip, hipProfilerStart,          hipProfilerStart,           HIP_HIPPROFILERSTART,           HIP, 1);
-  HC_LOAD_FUNC_HIP (hip, hipProfilerStop,           hipProfilerStop,            HIP_HIPPROFILERSTOP,            HIP, 1);
+  HC_LOAD_FUNC_HIP (hip, hipRuntimeGetVersion,      hipRuntimeGetVersion,       HIP_HIPRUNTIMEGETVERSION,       HIP, 1);
   HC_LOAD_FUNC_HIP (hip, hipStreamCreate,           hipStreamCreate,            HIP_HIPSTREAMCREATE,            HIP, 1);
   HC_LOAD_FUNC_HIP (hip, hipStreamDestroy,          hipStreamDestroy,           HIP_HIPSTREAMDESTROY,           HIP, 1);
   HC_LOAD_FUNC_HIP (hip, hipStreamSynchronize,      hipStreamSynchronize,       HIP_HIPSTREAMSYNCHRONIZE,       HIP, 1);
-  HC_LOAD_FUNC_HIP (hip, hipStreamWaitEvent,        hipStreamWaitEvent,         HIP_HIPSTREAMWAITEVENT,         HIP, 1);
-  #if defined (WITH_CUBINX)
-  HC_LOAD_FUNC_HIP (hip, hipLinkCreate,             hipLinkCreate,              HIP_HIPLINKCREATE,              HIP, 1);
-  HC_LOAD_FUNC_HIP (hip, hipLinkAddData,            hipLinkAddData,             HIP_HIPLINKADDDATA,             HIP, 1);
-  HC_LOAD_FUNC_HIP (hip, hipLinkDestroy,            hipLinkDestroy,             HIP_HIPLINKDESTROY,             HIP, 1);
-  HC_LOAD_FUNC_HIP (hip, hipLinkComplete,           hipLinkComplete,            HIP_HIPLINKCOMPLETE,            HIP, 1);
-  #endif
 
   return 0;
 }
@@ -2512,25 +2522,25 @@ void hip_close (hashcat_ctx_t *hashcat_ctx)
   }
 }
 
-int hc_hipInit (hashcat_ctx_t *hashcat_ctx, unsigned int Flags)
+int hc_hipCtxCreate (hashcat_ctx_t *hashcat_ctx, hipCtx_t *pctx, unsigned int flags, hipDevice_t dev)
 {
   backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
 
   HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
 
-  const HIPresult HIP_err = hip->hipInit (Flags);
+  const hipError_t HIP_err = hip->hipCtxCreate (pctx, flags, dev);
 
-  if (HIP_err != HIP_SUCCESS)
+  if (HIP_err != hipSuccess)
   {
     const char *pStr = NULL;
 
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
     {
-      event_log_error (hashcat_ctx, "hipInit(): %s", pStr);
+      event_log_error (hashcat_ctx, "hipCtxCreate(): %s", pStr);
     }
     else
     {
-      event_log_error (hashcat_ctx, "hipInit(): %d", HIP_err);
+      event_log_error (hashcat_ctx, "hipCtxCreate(): %d", HIP_err);
     }
 
     return -1;
@@ -2539,19 +2549,181 @@ int hc_hipInit (hashcat_ctx_t *hashcat_ctx, unsigned int Flags)
   return 0;
 }
 
-int hc_hipDeviceGetAttribute (hashcat_ctx_t *hashcat_ctx, int *pi, HIPdevice_attribute attrib, HIPdevice dev)
+int hc_hipCtxDestroy (hashcat_ctx_t *hashcat_ctx, hipCtx_t ctx)
 {
   backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
 
   HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
 
-  const HIPresult HIP_err = hip->hipDeviceGetAttribute (pi, attrib, dev);
+  const hipError_t HIP_err = hip->hipCtxDestroy (ctx);
 
-  if (HIP_err != HIP_SUCCESS)
+  if (HIP_err != hipSuccess)
   {
     const char *pStr = NULL;
 
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
+    {
+      event_log_error (hashcat_ctx, "hipCtxDestroy(): %s", pStr);
+    }
+    else
+    {
+      event_log_error (hashcat_ctx, "hipCtxDestroy(): %d", HIP_err);
+    }
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hc_hipCtxPopCurrent (hashcat_ctx_t *hashcat_ctx, hipCtx_t *pctx)
+{
+  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
+
+  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
+
+  const hipError_t HIP_err = hip->hipCtxPopCurrent (pctx);
+
+  if (HIP_err != hipSuccess)
+  {
+    const char *pStr = NULL;
+
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
+    {
+      event_log_error (hashcat_ctx, "hipCtxPopCurrent(): %s", pStr);
+    }
+    else
+    {
+      event_log_error (hashcat_ctx, "hipCtxPopCurrent(): %d", HIP_err);
+    }
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hc_hipCtxPushCurrent (hashcat_ctx_t *hashcat_ctx, hipCtx_t ctx)
+{
+  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
+
+  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
+
+  const hipError_t HIP_err = hip->hipCtxPushCurrent (ctx);
+
+  if (HIP_err != hipSuccess)
+  {
+    const char *pStr = NULL;
+
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
+    {
+      event_log_error (hashcat_ctx, "hipCtxPushCurrent(): %s", pStr);
+    }
+    else
+    {
+      event_log_error (hashcat_ctx, "hipCtxPushCurrent(): %d", HIP_err);
+    }
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hc_hipCtxSetCurrent (hashcat_ctx_t *hashcat_ctx, hipCtx_t ctx)
+{
+  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
+
+  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
+
+  const hipError_t HIP_err = hip->hipCtxSetCurrent (ctx);
+
+  if (HIP_err != hipSuccess)
+  {
+    const char *pStr = NULL;
+
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
+    {
+      event_log_error (hashcat_ctx, "hipCtxSetCurrent(): %s", pStr);
+    }
+    else
+    {
+      event_log_error (hashcat_ctx, "hipCtxSetCurrent(): %d", HIP_err);
+    }
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hc_hipCtxSynchronize (hashcat_ctx_t *hashcat_ctx)
+{
+  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
+
+  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
+
+  const hipError_t HIP_err = hip->hipCtxSynchronize ();
+
+  if (HIP_err != hipSuccess)
+  {
+    const char *pStr = NULL;
+
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
+    {
+      event_log_error (hashcat_ctx, "hipCtxSynchronize(): %s", pStr);
+    }
+    else
+    {
+      event_log_error (hashcat_ctx, "hipCtxSynchronize(): %d", HIP_err);
+    }
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hc_hipDeviceGet (hashcat_ctx_t *hashcat_ctx, hipDevice_t* device, int ordinal)
+{
+  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
+
+  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
+
+  const hipError_t HIP_err = hip->hipDeviceGet (device, ordinal);
+
+  if (HIP_err != hipSuccess)
+  {
+    const char *pStr = NULL;
+
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
+    {
+      event_log_error (hashcat_ctx, "hipDeviceGet(): %s", pStr);
+    }
+    else
+    {
+      event_log_error (hashcat_ctx, "hipDeviceGet(): %d", HIP_err);
+    }
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hc_hipDeviceGetAttribute (hashcat_ctx_t *hashcat_ctx, int *pi, hipDeviceAttribute_t attrib, hipDevice_t dev)
+{
+  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
+
+  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
+
+  const hipError_t HIP_err = hip->hipDeviceGetAttribute (pi, attrib, dev);
+
+  if (HIP_err != hipSuccess)
+  {
+    const char *pStr = NULL;
+
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
     {
       event_log_error (hashcat_ctx, "hipDeviceGetAttribute(): %s", pStr);
     }
@@ -2572,13 +2744,13 @@ int hc_hipDeviceGetCount (hashcat_ctx_t *hashcat_ctx, int *count)
 
   HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
 
-  const HIPresult HIP_err = hip->hipDeviceGetCount (count);
+  const hipError_t HIP_err = hip->hipDeviceGetCount (count);
 
-  if (HIP_err != HIP_SUCCESS)
+  if (HIP_err != hipSuccess)
   {
     const char *pStr = NULL;
 
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
     {
       event_log_error (hashcat_ctx, "hipDeviceGetCount(): %s", pStr);
     }
@@ -2593,46 +2765,19 @@ int hc_hipDeviceGetCount (hashcat_ctx_t *hashcat_ctx, int *count)
   return 0;
 }
 
-int hc_hipDeviceGet (hashcat_ctx_t *hashcat_ctx, HIPdevice* device, int ordinal)
+int hc_hipDeviceGetName (hashcat_ctx_t *hashcat_ctx, char *name, int len, hipDevice_t dev)
 {
   backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
 
   HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
 
-  const HIPresult HIP_err = hip->hipDeviceGet (device, ordinal);
+  const hipError_t HIP_err = hip->hipDeviceGetName (name, len, dev);
 
-  if (HIP_err != HIP_SUCCESS)
+  if (HIP_err != hipSuccess)
   {
     const char *pStr = NULL;
 
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
-    {
-      event_log_error (hashcat_ctx, "hipDeviceGet(): %s", pStr);
-    }
-    else
-    {
-      event_log_error (hashcat_ctx, "hipDeviceGet(): %d", HIP_err);
-    }
-
-    return -1;
-  }
-
-  return 0;
-}
-
-int hc_hipDeviceGetName (hashcat_ctx_t *hashcat_ctx, char *name, int len, HIPdevice dev)
-{
-  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
-
-  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
-
-  const HIPresult HIP_err = hip->hipDeviceGetName (name, len, dev);
-
-  if (HIP_err != HIP_SUCCESS)
-  {
-    const char *pStr = NULL;
-
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
     {
       event_log_error (hashcat_ctx, "hipDeviceGetName(): %s", pStr);
     }
@@ -2647,19 +2792,19 @@ int hc_hipDeviceGetName (hashcat_ctx_t *hashcat_ctx, char *name, int len, HIPdev
   return 0;
 }
 
-int hc_hipDeviceTotalMem (hashcat_ctx_t *hashcat_ctx, size_t *bytes, HIPdevice dev)
+int hc_hipDeviceTotalMem (hashcat_ctx_t *hashcat_ctx, size_t *bytes, hipDevice_t dev)
 {
   backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
 
   HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
 
-  const HIPresult HIP_err = hip->hipDeviceTotalMem (bytes, dev);
+  const hipError_t HIP_err = hip->hipDeviceTotalMem (bytes, dev);
 
-  if (HIP_err != HIP_SUCCESS)
+  if (HIP_err != hipSuccess)
   {
     const char *pStr = NULL;
 
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
     {
       event_log_error (hashcat_ctx, "hipDeviceTotalMem(): %s", pStr);
     }
@@ -2680,13 +2825,13 @@ int hc_hipDriverGetVersion (hashcat_ctx_t *hashcat_ctx, int *driverVersion)
 
   HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
 
-  const HIPresult HIP_err = hip->hipDriverGetVersion (driverVersion);
+  const hipError_t HIP_err = hip->hipDriverGetVersion (driverVersion);
 
-  if (HIP_err != HIP_SUCCESS)
+  if (HIP_err != hipSuccess)
   {
     const char *pStr = NULL;
 
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
     {
       event_log_error (hashcat_ctx, "hipDriverGetVersion(): %s", pStr);
     }
@@ -2701,564 +2846,19 @@ int hc_hipDriverGetVersion (hashcat_ctx_t *hashcat_ctx, int *driverVersion)
   return 0;
 }
 
-int hc_hipCtxCreate (hashcat_ctx_t *hashcat_ctx, HIPcontext *pctx, unsigned int flags, HIPdevice dev)
+int hc_hipEventCreate (hashcat_ctx_t *hashcat_ctx, hipEvent_t *phEvent, unsigned int Flags)
 {
   backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
 
   HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
 
-  const HIPresult HIP_err = hip->hipCtxCreate (pctx, flags, dev);
+  const hipError_t HIP_err = hip->hipEventCreate (phEvent, Flags);
 
-  if (HIP_err != HIP_SUCCESS)
+  if (HIP_err != hipSuccess)
   {
     const char *pStr = NULL;
 
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
-    {
-      event_log_error (hashcat_ctx, "hipCtxCreate(): %s", pStr);
-    }
-    else
-    {
-      event_log_error (hashcat_ctx, "hipCtxCreate(): %d", HIP_err);
-    }
-
-    return -1;
-  }
-
-  return 0;
-}
-
-int hc_hipCtxDestroy (hashcat_ctx_t *hashcat_ctx, HIPcontext ctx)
-{
-  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
-
-  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
-
-  const HIPresult HIP_err = hip->hipCtxDestroy (ctx);
-
-  if (HIP_err != HIP_SUCCESS)
-  {
-    const char *pStr = NULL;
-
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
-    {
-      event_log_error (hashcat_ctx, "hipCtxDestroy(): %s", pStr);
-    }
-    else
-    {
-      event_log_error (hashcat_ctx, "hipCtxDestroy(): %d", HIP_err);
-    }
-
-    return -1;
-  }
-
-  return 0;
-}
-
-int hc_hipModuleLoadDataEx (hashcat_ctx_t *hashcat_ctx, HIPmodule *module, const void *image, unsigned int numOptions, HIPjit_option *options, void **optionValues)
-{
-  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
-
-  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
-
-  const HIPresult HIP_err = hip->hipModuleLoadDataEx (module, image, numOptions, options, optionValues);
-
-  if (HIP_err != HIP_SUCCESS)
-  {
-    const char *pStr = NULL;
-
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
-    {
-      event_log_error (hashcat_ctx, "hipModuleLoadDataEx(): %s", pStr);
-    }
-    else
-    {
-      event_log_error (hashcat_ctx, "hipModuleLoadDataEx(): %d", HIP_err);
-    }
-
-    return -1;
-  }
-
-  return 0;
-}
-
-int hc_hipModuleUnload (hashcat_ctx_t *hashcat_ctx, HIPmodule hmod)
-{
-  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
-
-  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
-
-  const HIPresult HIP_err = hip->hipModuleUnload (hmod);
-
-  if (HIP_err != HIP_SUCCESS)
-  {
-    const char *pStr = NULL;
-
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
-    {
-      event_log_error (hashcat_ctx, "hipModuleUnload(): %s", pStr);
-    }
-    else
-    {
-      event_log_error (hashcat_ctx, "hipModuleUnload(): %d", HIP_err);
-    }
-
-    return -1;
-  }
-
-  return 0;
-}
-
-int hc_hipCtxSetCurrent (hashcat_ctx_t *hashcat_ctx, HIPcontext ctx)
-{
-  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
-
-  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
-
-  const HIPresult HIP_err = hip->hipCtxSetCurrent (ctx);
-
-  if (HIP_err != HIP_SUCCESS)
-  {
-    const char *pStr = NULL;
-
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
-    {
-      event_log_error (hashcat_ctx, "hipCtxSetCurrent(): %s", pStr);
-    }
-    else
-    {
-      event_log_error (hashcat_ctx, "hipCtxSetCurrent(): %d", HIP_err);
-    }
-
-    return -1;
-  }
-
-  return 0;
-}
-
-int hc_hipMemAlloc (hashcat_ctx_t *hashcat_ctx, HIPdeviceptr *dptr, size_t bytesize)
-{
-  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
-
-  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
-
-  const HIPresult HIP_err = hip->hipMemAlloc (dptr, bytesize);
-
-  if (HIP_err != HIP_SUCCESS)
-  {
-    const char *pStr = NULL;
-
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
-    {
-      event_log_error (hashcat_ctx, "hipMemAlloc(): %s", pStr);
-    }
-    else
-    {
-      event_log_error (hashcat_ctx, "hipMemAlloc(): %d", HIP_err);
-    }
-
-    return -1;
-  }
-
-  return 0;
-}
-
-int hc_hipMemFree (hashcat_ctx_t *hashcat_ctx, HIPdeviceptr dptr)
-{
-  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
-
-  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
-
-  const HIPresult HIP_err = hip->hipMemFree (dptr);
-
-  if (HIP_err != HIP_SUCCESS)
-  {
-    const char *pStr = NULL;
-
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
-    {
-      event_log_error (hashcat_ctx, "hipMemFree(): %s", pStr);
-    }
-    else
-    {
-      event_log_error (hashcat_ctx, "hipMemFree(): %d", HIP_err);
-    }
-
-    return -1;
-  }
-
-  return 0;
-}
-
-int hc_hipMemcpyDtoH (hashcat_ctx_t *hashcat_ctx, void *dstHost, HIPdeviceptr srcDevice, size_t ByteCount)
-{
-  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
-
-  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
-
-  const HIPresult HIP_err = hip->hipMemcpyDtoH (dstHost, srcDevice, ByteCount);
-
-  if (HIP_err != HIP_SUCCESS)
-  {
-    const char *pStr = NULL;
-
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
-    {
-      event_log_error (hashcat_ctx, "hipMemcpyDtoH(): %s", pStr);
-    }
-    else
-    {
-      event_log_error (hashcat_ctx, "hipMemcpyDtoH(): %d", HIP_err);
-    }
-
-    return -1;
-  }
-
-  return 0;
-}
-
-int hc_hipMemcpyDtoD (hashcat_ctx_t *hashcat_ctx, HIPdeviceptr dstDevice, HIPdeviceptr srcDevice, size_t ByteCount)
-{
-  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
-
-  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
-
-  const HIPresult HIP_err = hip->hipMemcpyDtoD (dstDevice, srcDevice, ByteCount);
-
-  if (HIP_err != HIP_SUCCESS)
-  {
-    const char *pStr = NULL;
-
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
-    {
-      event_log_error (hashcat_ctx, "hipMemcpyDtoD(): %s", pStr);
-    }
-    else
-    {
-      event_log_error (hashcat_ctx, "hipMemcpyDtoD(): %d", HIP_err);
-    }
-
-    return -1;
-  }
-
-  return 0;
-}
-
-int hc_hipMemcpyHtoD (hashcat_ctx_t *hashcat_ctx, HIPdeviceptr dstDevice, const void *srcHost, size_t ByteCount)
-{
-  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
-
-  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
-
-  const HIPresult HIP_err = hip->hipMemcpyHtoD (dstDevice, srcHost, ByteCount);
-
-  if (HIP_err != HIP_SUCCESS)
-  {
-    const char *pStr = NULL;
-
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
-    {
-      event_log_error (hashcat_ctx, "hipMemcpyHtoD(): %s", pStr);
-    }
-    else
-    {
-      event_log_error (hashcat_ctx, "hipMemcpyHtoD(): %d", HIP_err);
-    }
-
-    return -1;
-  }
-
-  return 0;
-}
-
-int hc_hipModuleGetFunction (hashcat_ctx_t *hashcat_ctx, HIPfunction *hfunc, HIPmodule hmod, const char *name)
-{
-  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
-
-  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
-
-  const HIPresult HIP_err = hip->hipModuleGetFunction (hfunc, hmod, name);
-
-  if (HIP_err != HIP_SUCCESS)
-  {
-    const char *pStr = NULL;
-
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
-    {
-      event_log_error (hashcat_ctx, "hipModuleGetFunction(): %s", pStr);
-    }
-    else
-    {
-      event_log_error (hashcat_ctx, "hipModuleGetFunction(): %d", HIP_err);
-    }
-
-    return -1;
-  }
-
-  return 0;
-}
-
-int hc_hipModuleGetGlobal (hashcat_ctx_t *hashcat_ctx, HIPdeviceptr *dptr, size_t *bytes, HIPmodule hmod, const char *name)
-{
-  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
-
-  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
-
-  const HIPresult HIP_err = hip->hipModuleGetGlobal (dptr, bytes, hmod, name);
-
-  if (HIP_err != HIP_SUCCESS)
-  {
-    const char *pStr = NULL;
-
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
-    {
-      event_log_error (hashcat_ctx, "hipModuleGetGlobal(): %s", pStr);
-    }
-    else
-    {
-      event_log_error (hashcat_ctx, "hipModuleGetGlobal(): %d", HIP_err);
-    }
-
-    return -1;
-  }
-
-  return 0;
-}
-
-int hc_hipMemGetInfo (hashcat_ctx_t *hashcat_ctx, size_t *free, size_t *total)
-{
-  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
-
-  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
-
-  const HIPresult HIP_err = hip->hipMemGetInfo (free, total);
-
-  if (HIP_err != HIP_SUCCESS)
-  {
-    const char *pStr = NULL;
-
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
-    {
-      event_log_error (hashcat_ctx, "hipMemGetInfo(): %s", pStr);
-    }
-    else
-    {
-      event_log_error (hashcat_ctx, "hipMemGetInfo(): %d", HIP_err);
-    }
-
-    return -1;
-  }
-
-  return 0;
-}
-
-int hc_hipFuncGetAttribute (hashcat_ctx_t *hashcat_ctx, int *pi, HIPfunction_attribute attrib, HIPfunction hfunc)
-{
-  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
-
-  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
-
-  const HIPresult HIP_err = hip->hipFuncGetAttribute (pi, attrib, hfunc);
-
-  if (HIP_err != HIP_SUCCESS)
-  {
-    const char *pStr = NULL;
-
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
-    {
-      event_log_error (hashcat_ctx, "hipFuncGetAttribute(): %s", pStr);
-    }
-    else
-    {
-      event_log_error (hashcat_ctx, "hipFuncGetAttribute(): %d", HIP_err);
-    }
-
-    return -1;
-  }
-
-  return 0;
-}
-
-/*
-
-// ATTENTION, this one maps to cudaFuncSetAttribute not cuFuncSetAttribute !!!
-
-int hc_hipFuncSetAttribute (hashcat_ctx_t *hashcat_ctx, HIPfunction hfunc, HIPfunction_attribute attrib, int value)
-{
-  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
-
-  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
-
-  const HIPresult HIP_err = hip->hipFuncSetAttribute (hfunc, attrib, value);
-
-  if (HIP_err != HIP_SUCCESS)
-  {
-    const char *pStr = NULL;
-
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
-    {
-      event_log_error (hashcat_ctx, "hipFuncSetAttribute(): %s", pStr);
-    }
-    else
-    {
-      event_log_error (hashcat_ctx, "hipFuncSetAttribute(): %d", HIP_err);
-    }
-
-    return -1;
-  }
-
-  return 0;
-}
-*/
-
-int hc_hipStreamCreate (hashcat_ctx_t *hashcat_ctx, HIPstream *phStream, unsigned int Flags)
-{
-  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
-
-  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
-
-  const HIPresult HIP_err = hip->hipStreamCreate (phStream, Flags);
-
-  if (HIP_err != HIP_SUCCESS)
-  {
-    const char *pStr = NULL;
-
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
-    {
-      event_log_error (hashcat_ctx, "hipStreamCreate(): %s", pStr);
-    }
-    else
-    {
-      event_log_error (hashcat_ctx, "hipStreamCreate(): %d", HIP_err);
-    }
-
-    return -1;
-  }
-
-  return 0;
-}
-
-int hc_hipStreamDestroy (hashcat_ctx_t *hashcat_ctx, HIPstream hStream)
-{
-  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
-
-  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
-
-  const HIPresult HIP_err = hip->hipStreamDestroy (hStream);
-
-  if (HIP_err != HIP_SUCCESS)
-  {
-    const char *pStr = NULL;
-
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
-    {
-      event_log_error (hashcat_ctx, "hipStreamDestroy(): %s", pStr);
-    }
-    else
-    {
-      event_log_error (hashcat_ctx, "hipStreamDestroy(): %d", HIP_err);
-    }
-
-    return -1;
-  }
-
-  return 0;
-}
-
-int hc_hipStreamSynchronize (hashcat_ctx_t *hashcat_ctx, HIPstream hStream)
-{
-  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
-
-  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
-
-  const HIPresult HIP_err = hip->hipStreamSynchronize (hStream);
-
-  if (HIP_err != HIP_SUCCESS)
-  {
-    const char *pStr = NULL;
-
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
-    {
-      event_log_error (hashcat_ctx, "hipStreamSynchronize(): %s", pStr);
-    }
-    else
-    {
-      event_log_error (hashcat_ctx, "hipStreamSynchronize(): %d", HIP_err);
-    }
-
-    return -1;
-  }
-
-  return 0;
-}
-
-int hc_hipLaunchKernel (hashcat_ctx_t *hashcat_ctx, HIPfunction f, unsigned int gridDimX, unsigned int gridDimY, unsigned int gridDimZ, unsigned int blockDimX, unsigned int blockDimY, unsigned int blockDimZ, unsigned int sharedMemBytes, HIPstream hStream, void **kernelParams, void **extra)
-{
-  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
-
-  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
-
-  const HIPresult HIP_err = hip->hipLaunchKernel (f, gridDimX, gridDimY, gridDimZ, blockDimX, blockDimY, blockDimZ, sharedMemBytes, hStream, kernelParams, extra);
-
-  if (HIP_err != HIP_SUCCESS)
-  {
-    const char *pStr = NULL;
-
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
-    {
-      event_log_error (hashcat_ctx, "hipLaunchKernel(): %s", pStr);
-    }
-    else
-    {
-      event_log_error (hashcat_ctx, "hipLaunchKernel(): %d", HIP_err);
-    }
-
-    return -1;
-  }
-
-  return 0;
-}
-
-int hc_hipCtxSynchronize (hashcat_ctx_t *hashcat_ctx)
-{
-  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
-
-  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
-
-  const HIPresult HIP_err = hip->hipCtxSynchronize ();
-
-  if (HIP_err != HIP_SUCCESS)
-  {
-    const char *pStr = NULL;
-
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
-    {
-      event_log_error (hashcat_ctx, "hipCtxSynchronize(): %s", pStr);
-    }
-    else
-    {
-      event_log_error (hashcat_ctx, "hipCtxSynchronize(): %d", HIP_err);
-    }
-
-    return -1;
-  }
-
-  return 0;
-}
-
-int hc_hipEventCreate (hashcat_ctx_t *hashcat_ctx, HIPevent *phEvent, unsigned int Flags)
-{
-  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
-
-  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
-
-  const HIPresult HIP_err = hip->hipEventCreate (phEvent, Flags);
-
-  if (HIP_err != HIP_SUCCESS)
-  {
-    const char *pStr = NULL;
-
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
     {
       event_log_error (hashcat_ctx, "hipEventCreate(): %s", pStr);
     }
@@ -3273,19 +2873,19 @@ int hc_hipEventCreate (hashcat_ctx_t *hashcat_ctx, HIPevent *phEvent, unsigned i
   return 0;
 }
 
-int hc_hipEventDestroy (hashcat_ctx_t *hashcat_ctx, HIPevent hEvent)
+int hc_hipEventDestroy (hashcat_ctx_t *hashcat_ctx, hipEvent_t hEvent)
 {
   backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
 
   HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
 
-  const HIPresult HIP_err = hip->hipEventDestroy (hEvent);
+  const hipError_t HIP_err = hip->hipEventDestroy (hEvent);
 
-  if (HIP_err != HIP_SUCCESS)
+  if (HIP_err != hipSuccess)
   {
     const char *pStr = NULL;
 
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
     {
       event_log_error (hashcat_ctx, "hipEventDestroy(): %s", pStr);
     }
@@ -3300,19 +2900,19 @@ int hc_hipEventDestroy (hashcat_ctx_t *hashcat_ctx, HIPevent hEvent)
   return 0;
 }
 
-int hc_hipEventElapsedTime (hashcat_ctx_t *hashcat_ctx, float *pMilliseconds, HIPevent hStart, HIPevent hEnd)
+int hc_hipEventElapsedTime (hashcat_ctx_t *hashcat_ctx, float *pMilliseconds, hipEvent_t hStart, hipEvent_t hEnd)
 {
   backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
 
   HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
 
-  const HIPresult HIP_err = hip->hipEventElapsedTime (pMilliseconds, hStart, hEnd);
+  const hipError_t HIP_err = hip->hipEventElapsedTime (pMilliseconds, hStart, hEnd);
 
-  if (HIP_err != HIP_SUCCESS)
+  if (HIP_err != hipSuccess)
   {
     const char *pStr = NULL;
 
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
     {
       event_log_error (hashcat_ctx, "hipEventElapsedTime(): %s", pStr);
     }
@@ -3327,46 +2927,19 @@ int hc_hipEventElapsedTime (hashcat_ctx_t *hashcat_ctx, float *pMilliseconds, HI
   return 0;
 }
 
-int hc_hipEventQuery (hashcat_ctx_t *hashcat_ctx, HIPevent hEvent)
+int hc_hipEventRecord (hashcat_ctx_t *hashcat_ctx, hipEvent_t hEvent, hipStream_t hStream)
 {
   backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
 
   HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
 
-  const HIPresult HIP_err = hip->hipEventQuery (hEvent);
+  const hipError_t HIP_err = hip->hipEventRecord (hEvent, hStream);
 
-  if (HIP_err != HIP_SUCCESS)
+  if (HIP_err != hipSuccess)
   {
     const char *pStr = NULL;
 
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
-    {
-      event_log_error (hashcat_ctx, "hipEventQuery(): %s", pStr);
-    }
-    else
-    {
-      event_log_error (hashcat_ctx, "hipEventQuery(): %d", HIP_err);
-    }
-
-    return -1;
-  }
-
-  return 0;
-}
-
-int hc_hipEventRecord (hashcat_ctx_t *hashcat_ctx, HIPevent hEvent, HIPstream hStream)
-{
-  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
-
-  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
-
-  const HIPresult HIP_err = hip->hipEventRecord (hEvent, hStream);
-
-  if (HIP_err != HIP_SUCCESS)
-  {
-    const char *pStr = NULL;
-
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
     {
       event_log_error (hashcat_ctx, "hipEventRecord(): %s", pStr);
     }
@@ -3381,19 +2954,19 @@ int hc_hipEventRecord (hashcat_ctx_t *hashcat_ctx, HIPevent hEvent, HIPstream hS
   return 0;
 }
 
-int hc_hipEventSynchronize (hashcat_ctx_t *hashcat_ctx, HIPevent hEvent)
+int hc_hipEventSynchronize (hashcat_ctx_t *hashcat_ctx, hipEvent_t hEvent)
 {
   backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
 
   HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
 
-  const HIPresult HIP_err = hip->hipEventSynchronize (hEvent);
+  const hipError_t HIP_err = hip->hipEventSynchronize (hEvent);
 
-  if (HIP_err != HIP_SUCCESS)
+  if (HIP_err != hipSuccess)
   {
     const char *pStr = NULL;
 
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
     {
       event_log_error (hashcat_ctx, "hipEventSynchronize(): %s", pStr);
     }
@@ -3408,25 +2981,25 @@ int hc_hipEventSynchronize (hashcat_ctx_t *hashcat_ctx, HIPevent hEvent)
   return 0;
 }
 
-int hc_hipCtxSetCacheConfig (hashcat_ctx_t *hashcat_ctx, HIPfunc_cache config)
+int hc_hipFuncGetAttribute (hashcat_ctx_t *hashcat_ctx, int *pi, hipFunction_attribute attrib, hipFunction_t hfunc)
 {
   backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
 
   HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
 
-  const HIPresult HIP_err = hip->hipCtxSetCacheConfig (config);
+  const hipError_t HIP_err = hip->hipFuncGetAttribute (pi, attrib, hfunc);
 
-  if (HIP_err != HIP_SUCCESS)
+  if (HIP_err != hipSuccess)
   {
     const char *pStr = NULL;
 
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
     {
-      event_log_error (hashcat_ctx, "hipCtxSetCacheConfig(): %s", pStr);
+      event_log_error (hashcat_ctx, "hipFuncGetAttribute(): %s", pStr);
     }
     else
     {
-      event_log_error (hashcat_ctx, "hipCtxSetCacheConfig(): %d", HIP_err);
+      event_log_error (hashcat_ctx, "hipFuncGetAttribute(): %d", HIP_err);
     }
 
     return -1;
@@ -3435,25 +3008,25 @@ int hc_hipCtxSetCacheConfig (hashcat_ctx_t *hashcat_ctx, HIPfunc_cache config)
   return 0;
 }
 
-int hc_hipCtxPushCurrent (hashcat_ctx_t *hashcat_ctx, HIPcontext ctx)
+int hc_hipLaunchKernel (hashcat_ctx_t *hashcat_ctx, hipFunction_t f, unsigned int gridDimX, unsigned int gridDimY, unsigned int gridDimZ, unsigned int blockDimX, unsigned int blockDimY, unsigned int blockDimZ, unsigned int sharedMemBytes, hipStream_t hStream, void **kernelParams, void **extra)
 {
   backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
 
   HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
 
-  const HIPresult HIP_err = hip->hipCtxPushCurrent (ctx);
+  const hipError_t HIP_err = hip->hipLaunchKernel (f, gridDimX, gridDimY, gridDimZ, blockDimX, blockDimY, blockDimZ, sharedMemBytes, hStream, kernelParams, extra);
 
-  if (HIP_err != HIP_SUCCESS)
+  if (HIP_err != hipSuccess)
   {
     const char *pStr = NULL;
 
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
     {
-      event_log_error (hashcat_ctx, "hipCtxPushCurrent(): %s", pStr);
+      event_log_error (hashcat_ctx, "hipLaunchKernel(): %s", pStr);
     }
     else
     {
-      event_log_error (hashcat_ctx, "hipCtxPushCurrent(): %d", HIP_err);
+      event_log_error (hashcat_ctx, "hipLaunchKernel(): %d", HIP_err);
     }
 
     return -1;
@@ -3462,25 +3035,25 @@ int hc_hipCtxPushCurrent (hashcat_ctx_t *hashcat_ctx, HIPcontext ctx)
   return 0;
 }
 
-int hc_hipCtxPopCurrent (hashcat_ctx_t *hashcat_ctx, HIPcontext *pctx)
+int hc_hipInit (hashcat_ctx_t *hashcat_ctx, unsigned int Flags)
 {
   backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
 
   HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
 
-  const HIPresult HIP_err = hip->hipCtxPopCurrent (pctx);
+  const hipError_t HIP_err = hip->hipInit (Flags);
 
-  if (HIP_err != HIP_SUCCESS)
+  if (HIP_err != hipSuccess)
   {
     const char *pStr = NULL;
 
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
     {
-      event_log_error (hashcat_ctx, "hipCtxPopCurrent(): %s", pStr);
+      event_log_error (hashcat_ctx, "hipInit(): %s", pStr);
     }
     else
     {
-      event_log_error (hashcat_ctx, "hipCtxPopCurrent(): %d", HIP_err);
+      event_log_error (hashcat_ctx, "hipInit(): %d", HIP_err);
     }
 
     return -1;
@@ -3489,25 +3062,25 @@ int hc_hipCtxPopCurrent (hashcat_ctx_t *hashcat_ctx, HIPcontext *pctx)
   return 0;
 }
 
-int hc_hipLinkCreate (hashcat_ctx_t *hashcat_ctx, unsigned int numOptions, HIPjit_option *options, void **optionValues, HIPlinkState *stateOut)
+int hc_hipMemAlloc (hashcat_ctx_t *hashcat_ctx, hipDeviceptr_t *dptr, size_t bytesize)
 {
   backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
 
   HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
 
-  const HIPresult HIP_err = hip->hipLinkCreate (numOptions, options, optionValues, stateOut);
+  const hipError_t HIP_err = hip->hipMemAlloc (dptr, bytesize);
 
-  if (HIP_err != HIP_SUCCESS)
+  if (HIP_err != hipSuccess)
   {
     const char *pStr = NULL;
 
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
     {
-      event_log_error (hashcat_ctx, "hipLinkCreate(): %s", pStr);
+      event_log_error (hashcat_ctx, "hipMemAlloc(): %s", pStr);
     }
     else
     {
-      event_log_error (hashcat_ctx, "hipLinkCreate(): %d", HIP_err);
+      event_log_error (hashcat_ctx, "hipMemAlloc(): %d", HIP_err);
     }
 
     return -1;
@@ -3516,25 +3089,25 @@ int hc_hipLinkCreate (hashcat_ctx_t *hashcat_ctx, unsigned int numOptions, HIPji
   return 0;
 }
 
-int hc_hipLinkAddData (hashcat_ctx_t *hashcat_ctx, HIPlinkState state, HIPjitInputType type, void *data, size_t size, const char *name, unsigned int numOptions, HIPjit_option *options, void **optionValues)
+int hc_hipMemFree (hashcat_ctx_t *hashcat_ctx, hipDeviceptr_t dptr)
 {
   backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
 
   HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
 
-  const HIPresult HIP_err = hip->hipLinkAddData (state, type, data, size, name, numOptions, options, optionValues);
+  const hipError_t HIP_err = hip->hipMemFree (dptr);
 
-  if (HIP_err != HIP_SUCCESS)
+  if (HIP_err != hipSuccess)
   {
     const char *pStr = NULL;
 
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
     {
-      event_log_error (hashcat_ctx, "hipLinkAddData(): %s", pStr);
+      event_log_error (hashcat_ctx, "hipMemFree(): %s", pStr);
     }
     else
     {
-      event_log_error (hashcat_ctx, "hipLinkAddData(): %d", HIP_err);
+      event_log_error (hashcat_ctx, "hipMemFree(): %d", HIP_err);
     }
 
     return -1;
@@ -3543,25 +3116,25 @@ int hc_hipLinkAddData (hashcat_ctx_t *hashcat_ctx, HIPlinkState state, HIPjitInp
   return 0;
 }
 
-int hc_hipLinkDestroy (hashcat_ctx_t *hashcat_ctx, HIPlinkState state)
+int hc_hipMemGetInfo (hashcat_ctx_t *hashcat_ctx, size_t *free, size_t *total)
 {
   backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
 
   HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
 
-  const HIPresult HIP_err = hip->hipLinkDestroy (state);
+  const hipError_t HIP_err = hip->hipMemGetInfo (free, total);
 
-  if (HIP_err != HIP_SUCCESS)
+  if (HIP_err != hipSuccess)
   {
     const char *pStr = NULL;
 
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
     {
-      event_log_error (hashcat_ctx, "hipLinkDestroy(): %s", pStr);
+      event_log_error (hashcat_ctx, "hipMemGetInfo(): %s", pStr);
     }
     else
     {
-      event_log_error (hashcat_ctx, "hipLinkDestroy(): %d", HIP_err);
+      event_log_error (hashcat_ctx, "hipMemGetInfo(): %d", HIP_err);
     }
 
     return -1;
@@ -3570,25 +3143,349 @@ int hc_hipLinkDestroy (hashcat_ctx_t *hashcat_ctx, HIPlinkState state)
   return 0;
 }
 
-int hc_hipLinkComplete (hashcat_ctx_t *hashcat_ctx, HIPlinkState state, void **hipbinOut, size_t *sizeOut)
+int hc_hipMemcpyDtoHAsync (hashcat_ctx_t *hashcat_ctx, void *dstHost, hipDeviceptr_t srcDevice, size_t ByteCount, hipStream_t hStream)
 {
   backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
 
   HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
 
-  const HIPresult HIP_err = hip->hipLinkComplete (state, hipbinOut, sizeOut);
+  const hipError_t HIP_err = hip->hipMemcpyDtoHAsync (dstHost, srcDevice, ByteCount, hStream);
 
-  if (HIP_err != HIP_SUCCESS)
+  if (HIP_err != hipSuccess)
   {
     const char *pStr = NULL;
 
-    if (hip->hipGetErrorString (HIP_err, &pStr) == HIP_SUCCESS)
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
     {
-      event_log_error (hashcat_ctx, "hipLinkComplete(): %s", pStr);
+      event_log_error (hashcat_ctx, "hipMemcpyDtoHAsync(): %s", pStr);
     }
     else
     {
-      event_log_error (hashcat_ctx, "hipLinkComplete(): %d", HIP_err);
+      event_log_error (hashcat_ctx, "hipMemcpyDtoHAsync(): %d", HIP_err);
+    }
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hc_hipMemcpyDtoDAsync (hashcat_ctx_t *hashcat_ctx, hipDeviceptr_t dstDevice, hipDeviceptr_t srcDevice, size_t ByteCount, hipStream_t hStream)
+{
+  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
+
+  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
+
+  const hipError_t HIP_err = hip->hipMemcpyDtoDAsync (dstDevice, srcDevice, ByteCount, hStream);
+
+  if (HIP_err != hipSuccess)
+  {
+    const char *pStr = NULL;
+
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
+    {
+      event_log_error (hashcat_ctx, "hipMemcpyDtoDAsync(): %s", pStr);
+    }
+    else
+    {
+      event_log_error (hashcat_ctx, "hipMemcpyDtoDAsync(): %d", HIP_err);
+    }
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hc_hipMemcpyHtoDAsync (hashcat_ctx_t *hashcat_ctx, hipDeviceptr_t dstDevice, const void *srcHost, size_t ByteCount, hipStream_t hStream)
+{
+  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
+
+  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
+
+  const hipError_t HIP_err = hip->hipMemcpyHtoDAsync (dstDevice, srcHost, ByteCount, hStream);
+
+  if (HIP_err != hipSuccess)
+  {
+    const char *pStr = NULL;
+
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
+    {
+      event_log_error (hashcat_ctx, "hipMemcpyHtoDAsync(): %s", pStr);
+    }
+    else
+    {
+      event_log_error (hashcat_ctx, "hipMemcpyHtoDAsync(): %d", HIP_err);
+    }
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hc_hipMemsetD32Async (hashcat_ctx_t *hashcat_ctx, hipDeviceptr_t dstDevice, unsigned int ui, size_t N, hipStream_t hStream)
+{
+  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
+
+  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
+
+  const hipError_t HIP_err = hip->hipMemsetD32Async (dstDevice, ui, N, hStream);
+
+  if (HIP_err != hipSuccess)
+  {
+    const char *pStr = NULL;
+
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
+    {
+      event_log_error (hashcat_ctx, "hipMemsetD32Async(): %s", pStr);
+    }
+    else
+    {
+      event_log_error (hashcat_ctx, "hipMemsetD32Async(): %d", HIP_err);
+    }
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hc_hipMemsetD8Async (hashcat_ctx_t *hashcat_ctx, hipDeviceptr_t dstDevice, unsigned char uc, size_t N, hipStream_t hStream)
+{
+  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
+
+  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
+
+  const hipError_t HIP_err = hip->hipMemsetD8Async (dstDevice, uc, N, hStream);
+
+  if (HIP_err != hipSuccess)
+  {
+    const char *pStr = NULL;
+
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
+    {
+      event_log_error (hashcat_ctx, "hipMemsetD8Async(): %s", pStr);
+    }
+    else
+    {
+      event_log_error (hashcat_ctx, "hipMemsetD8Async(): %d", HIP_err);
+    }
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hc_hipModuleGetFunction (hashcat_ctx_t *hashcat_ctx, hipFunction_t *hfunc, hipModule_t hmod, const char *name)
+{
+  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
+
+  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
+
+  const hipError_t HIP_err = hip->hipModuleGetFunction (hfunc, hmod, name);
+
+  if (HIP_err != hipSuccess)
+  {
+    const char *pStr = NULL;
+
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
+    {
+      event_log_error (hashcat_ctx, "hipModuleGetFunction(): %s", pStr);
+    }
+    else
+    {
+      event_log_error (hashcat_ctx, "hipModuleGetFunction(): %d", HIP_err);
+    }
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hc_hipModuleGetGlobal (hashcat_ctx_t *hashcat_ctx, hipDeviceptr_t *dptr, size_t *bytes, hipModule_t hmod, const char *name)
+{
+  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
+
+  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
+
+  const hipError_t HIP_err = hip->hipModuleGetGlobal (dptr, bytes, hmod, name);
+
+  if (HIP_err != hipSuccess)
+  {
+    const char *pStr = NULL;
+
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
+    {
+      event_log_error (hashcat_ctx, "hipModuleGetGlobal(): %s", pStr);
+    }
+    else
+    {
+      event_log_error (hashcat_ctx, "hipModuleGetGlobal(): %d", HIP_err);
+    }
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hc_hipModuleLoadDataEx (hashcat_ctx_t *hashcat_ctx, hipModule_t *module, const void *image, unsigned int numOptions, hipJitOption *options, void **optionValues)
+{
+  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
+
+  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
+
+  const hipError_t HIP_err = hip->hipModuleLoadDataEx (module, image, numOptions, options, optionValues);
+
+  if (HIP_err != hipSuccess)
+  {
+    const char *pStr = NULL;
+
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
+    {
+      event_log_error (hashcat_ctx, "hipModuleLoadDataEx(): %s", pStr);
+    }
+    else
+    {
+      event_log_error (hashcat_ctx, "hipModuleLoadDataEx(): %d", HIP_err);
+    }
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hc_hipModuleUnload (hashcat_ctx_t *hashcat_ctx, hipModule_t hmod)
+{
+  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
+
+  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
+
+  const hipError_t HIP_err = hip->hipModuleUnload (hmod);
+
+  if (HIP_err != hipSuccess)
+  {
+    const char *pStr = NULL;
+
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
+    {
+      event_log_error (hashcat_ctx, "hipModuleUnload(): %s", pStr);
+    }
+    else
+    {
+      event_log_error (hashcat_ctx, "hipModuleUnload(): %d", HIP_err);
+    }
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hc_hipRuntimeGetVersion (hashcat_ctx_t *hashcat_ctx, int *runtimeVersion)
+{
+  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
+
+  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
+
+  const hipError_t HIP_err = hip->hipRuntimeGetVersion (runtimeVersion);
+
+  if (HIP_err != hipSuccess)
+  {
+    const char *pStr = NULL;
+
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
+    {
+      event_log_error (hashcat_ctx, "hipRuntimeGetVersion(): %s", pStr);
+    }
+    else
+    {
+      event_log_error (hashcat_ctx, "hipRuntimeGetVersion(): %d", HIP_err);
+    }
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hc_hipStreamCreate (hashcat_ctx_t *hashcat_ctx, hipStream_t *phStream, unsigned int Flags)
+{
+  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
+
+  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
+
+  const hipError_t HIP_err = hip->hipStreamCreate (phStream, Flags);
+
+  if (HIP_err != hipSuccess)
+  {
+    const char *pStr = NULL;
+
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
+    {
+      event_log_error (hashcat_ctx, "hipStreamCreate(): %s", pStr);
+    }
+    else
+    {
+      event_log_error (hashcat_ctx, "hipStreamCreate(): %d", HIP_err);
+    }
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hc_hipStreamDestroy (hashcat_ctx_t *hashcat_ctx, hipStream_t hStream)
+{
+  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
+
+  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
+
+  const hipError_t HIP_err = hip->hipStreamDestroy (hStream);
+
+  if (HIP_err != hipSuccess)
+  {
+    const char *pStr = NULL;
+
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
+    {
+      event_log_error (hashcat_ctx, "hipStreamDestroy(): %s", pStr);
+    }
+    else
+    {
+      event_log_error (hashcat_ctx, "hipStreamDestroy(): %d", HIP_err);
+    }
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hc_hipStreamSynchronize (hashcat_ctx_t *hashcat_ctx, hipStream_t hStream)
+{
+  backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
+
+  HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
+
+  const hipError_t HIP_err = hip->hipStreamSynchronize (hStream);
+
+  if (HIP_err != hipSuccess)
+  {
+    const char *pStr = NULL;
+
+    if (hip->hipGetErrorString (HIP_err, &pStr) == hipSuccess)
+    {
+      event_log_error (hashcat_ctx, "hipStreamSynchronize(): %s", pStr);
+    }
+    else
+    {
+      event_log_error (hashcat_ctx, "hipStreamSynchronize(): %d", HIP_err);
     }
 
     return -1;
@@ -3632,6 +3529,7 @@ int ocl_init (hashcat_ctx_t *hashcat_ctx)
   HC_LOAD_FUNC (ocl, clCreateProgramWithBinary, OCL_CLCREATEPROGRAMWITHBINARY,  OpenCL, 1);
   HC_LOAD_FUNC (ocl, clCreateProgramWithSource, OCL_CLCREATEPROGRAMWITHSOURCE,  OpenCL, 1);
   HC_LOAD_FUNC (ocl, clEnqueueCopyBuffer,       OCL_CLENQUEUECOPYBUFFER,        OpenCL, 1);
+  HC_LOAD_FUNC (ocl, clEnqueueFillBuffer,       OCL_CLENQUEUEFILLBUFFER,        OpenCL, -1);
   HC_LOAD_FUNC (ocl, clEnqueueMapBuffer,        OCL_CLENQUEUEMAPBUFFER,         OpenCL, 1);
   HC_LOAD_FUNC (ocl, clEnqueueNDRangeKernel,    OCL_CLENQUEUENDRANGEKERNEL,     OpenCL, 1);
   HC_LOAD_FUNC (ocl, clEnqueueReadBuffer,       OCL_CLENQUEUEREADBUFFER,        OpenCL, 1);
@@ -3807,6 +3705,23 @@ int hc_clEnqueueCopyBuffer (hashcat_ctx_t *hashcat_ctx, cl_command_queue command
   return 0;
 }
 
+int hc_clEnqueueFillBuffer (hashcat_ctx_t *hashcat_ctx, cl_command_queue command_queue, cl_mem buffer, const void *pattern, size_t pattern_size, size_t offset, size_t size, cl_uint num_events_in_wait_list, const cl_event *event_wait_list, cl_event *event)
+{
+  const backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
+  const OCL_PTR       *ocl         = backend_ctx->ocl;
+
+  cl_int CL_err = ocl->clEnqueueFillBuffer (command_queue, buffer, pattern, pattern_size, offset, size, num_events_in_wait_list, event_wait_list, event);
+
+  if (CL_err != CL_SUCCESS)
+  {
+    event_log_error (hashcat_ctx, "clEnqueueFillBuffer(): %s", val2cstr_cl (CL_err));
+
+    return -1;
+  }
+
+  return 0;
+}
+
 int hc_clEnqueueReadBuffer (hashcat_ctx_t *hashcat_ctx, cl_command_queue command_queue, cl_mem buffer, cl_bool blocking_read, size_t offset, size_t size, void *ptr, cl_uint num_events_in_wait_list, const cl_event *event_wait_list, cl_event *event)
 {
   backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
@@ -4341,22 +4256,23 @@ int gidd_to_pw_t (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, c
   {
     if (hc_cuCtxPushCurrent (hashcat_ctx, device_param->cuda_context) == -1) return -1;
 
-    if (hc_cuMemcpyDtoH (hashcat_ctx, &pw_idx, device_param->cuda_d_pws_idx + (gidd * sizeof (pw_idx_t)), sizeof (pw_idx_t)) == -1) return -1;
+    if (hc_cuMemcpyDtoHAsync (hashcat_ctx, &pw_idx, device_param->cuda_d_pws_idx + (gidd * sizeof (pw_idx_t)), sizeof (pw_idx_t), device_param->cuda_stream) == -1) return -1;
 
-    if (hc_cuCtxPopCurrent (hashcat_ctx, &device_param->cuda_context) == -1) return -1;
+    if (hc_cuStreamSynchronize (hashcat_ctx, device_param->cuda_stream) == -1) return -1;
   }
 
   if (device_param->is_hip == true)
   {
     if (hc_hipCtxPushCurrent (hashcat_ctx, device_param->hip_context) == -1) return -1;
 
-    if (hc_hipMemcpyDtoH (hashcat_ctx, &pw_idx, device_param->hip_d_pws_idx + (gidd * sizeof (pw_idx_t)), sizeof (pw_idx_t)) == -1) return -1;
+    if (hc_hipMemcpyDtoHAsync (hashcat_ctx, &pw_idx, device_param->hip_d_pws_idx + (gidd * sizeof (pw_idx_t)), sizeof (pw_idx_t), device_param->hip_stream) == -1) return -1;
 
-    if (hc_hipCtxPopCurrent (hashcat_ctx, &device_param->hip_context) == -1) return -1;
+    if (hc_hipStreamSynchronize (hashcat_ctx, device_param->hip_stream) == -1) return -1;
   }
 
   if (device_param->is_opencl == true)
   {
+    /* blocking */
     if (hc_clEnqueueReadBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_idx, CL_TRUE, gidd * sizeof (pw_idx_t), sizeof (pw_idx_t), &pw_idx, 0, NULL, NULL) == -1) return -1;
   }
 
@@ -4364,34 +4280,25 @@ int gidd_to_pw_t (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, c
   const u32 cnt = pw_idx.cnt;
   const u32 len = pw_idx.len;
 
-  if (device_param->is_cuda == true)
+  if (cnt > 0)
   {
-    if (cnt > 0)
+    if (device_param->is_cuda == true)
     {
-      if (hc_cuCtxPushCurrent (hashcat_ctx, device_param->cuda_context) == -1) return -1;
+      if (hc_cuMemcpyDtoHAsync (hashcat_ctx, pw->i, device_param->cuda_d_pws_comp_buf + (off * sizeof (u32)), cnt * sizeof (u32), device_param->cuda_stream) == -1) return -1;
 
-      if (hc_cuMemcpyDtoH (hashcat_ctx,pw->i, device_param->cuda_d_pws_comp_buf + (off * sizeof (u32)), cnt * sizeof (u32)) == -1) return -1;
-
-      if (hc_cuCtxPopCurrent (hashcat_ctx, &device_param->cuda_context) == -1) return -1;
+      if (hc_cuStreamSynchronize (hashcat_ctx, device_param->cuda_stream) == -1) return -1;
     }
-  }
 
-  if (device_param->is_hip == true)
-  {
-    if (cnt > 0)
+    if (device_param->is_hip == true)
     {
-      if (hc_hipCtxPushCurrent (hashcat_ctx, device_param->hip_context) == -1) return -1;
+      if (hc_hipMemcpyDtoHAsync (hashcat_ctx, pw->i, device_param->hip_d_pws_comp_buf + (off * sizeof (u32)), cnt * sizeof (u32), device_param->hip_stream) == -1) return -1;
 
-      if (hc_hipMemcpyDtoH (hashcat_ctx,pw->i, device_param->hip_d_pws_comp_buf + (off * sizeof (u32)), cnt * sizeof (u32)) == -1) return -1;
-
-      if (hc_hipCtxPopCurrent (hashcat_ctx, &device_param->hip_context) == -1) return -1;
+      if (hc_hipStreamSynchronize (hashcat_ctx, device_param->hip_stream) == -1) return -1;
     }
-  }
 
-  if (device_param->is_opencl == true)
-  {
-    if (cnt > 0)
+    if (device_param->is_opencl == true)
     {
+      /* blocking */
       if (hc_clEnqueueReadBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_comp_buf, CL_TRUE, off * sizeof (u32), cnt * sizeof (u32), pw->i, 0, NULL, NULL) == -1) return -1;
     }
   }
@@ -4403,6 +4310,16 @@ int gidd_to_pw_t (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, c
 
   pw->pw_len = len;
 
+  if (device_param->is_cuda == true)
+  {
+    if (hc_cuCtxPopCurrent (hashcat_ctx, &device_param->cuda_context) == -1) return -1;
+  }
+
+  if (device_param->is_hip == true)
+  {
+    if (hc_hipCtxPopCurrent (hashcat_ctx, &device_param->hip_context) == -1) return -1;
+  }
+
   return 0;
 }
 
@@ -4451,17 +4368,19 @@ int choose_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param,
 
           if (device_param->is_cuda == true)
           {
-            if (hc_cuMemcpyDtoD (hashcat_ctx, device_param->cuda_d_bfs_c, device_param->cuda_d_tm_c, size_tm) == -1) return -1;
+            if (hc_cuMemcpyDtoDAsync (hashcat_ctx, device_param->cuda_d_bfs_c, device_param->cuda_d_tm_c, size_tm, device_param->cuda_stream) == -1) return -1;
           }
 
           if (device_param->is_hip == true)
           {
-            if (hc_hipMemcpyDtoD (hashcat_ctx, device_param->hip_d_bfs_c, device_param->hip_d_tm_c, size_tm) == -1) return -1;
+            if (hc_hipMemcpyDtoDAsync (hashcat_ctx, device_param->hip_d_bfs_c, device_param->hip_d_tm_c, size_tm, device_param->hip_stream) == -1) return -1;
           }
 
           if (device_param->is_opencl == true)
           {
             if (hc_clEnqueueCopyBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_tm_c, device_param->opencl_d_bfs_c, 0, 0, size_tm, 0, NULL, NULL) == -1) return -1;
+
+            if (hc_clFlush (hashcat_ctx, device_param->opencl_command_queue) == -1) return -1;
           }
         }
       }
@@ -4517,12 +4436,12 @@ int choose_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param,
     {
       if (device_param->is_cuda == true)
       {
-        if (hc_cuMemcpyDtoD (hashcat_ctx, device_param->cuda_d_pws_buf, device_param->cuda_d_pws_amp_buf, pws_cnt * sizeof (pw_t)) == -1) return -1;
+        if (hc_cuMemcpyDtoDAsync (hashcat_ctx, device_param->cuda_d_pws_buf, device_param->cuda_d_pws_amp_buf, pws_cnt * sizeof (pw_t), device_param->cuda_stream) == -1) return -1;
       }
 
       if (device_param->is_hip == true)
       {
-        if (hc_hipMemcpyDtoD (hashcat_ctx, device_param->hip_d_pws_buf, device_param->hip_d_pws_amp_buf, pws_cnt * sizeof (pw_t)) == -1) return -1;
+        if (hc_hipMemcpyDtoDAsync (hashcat_ctx, device_param->hip_d_pws_buf, device_param->hip_d_pws_amp_buf, pws_cnt * sizeof (pw_t), device_param->hip_stream) == -1) return -1;
       }
 
       if (device_param->is_opencl == true)
@@ -4564,22 +4483,28 @@ int choose_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param,
 
         if (device_param->is_cuda == true)
         {
-          if (hc_cuMemcpyDtoH (hashcat_ctx, device_param->hooks_buf, device_param->cuda_d_hooks, pws_cnt * hashconfig->hook_size) == -1) return -1;
+          if (hc_cuMemcpyDtoHAsync (hashcat_ctx, device_param->hooks_buf, device_param->cuda_d_hooks, pws_cnt * hashconfig->hook_size, device_param->cuda_stream) == -1) return -1;
+
+          if (hc_cuStreamSynchronize (hashcat_ctx, device_param->cuda_stream) == -1) return -1;
         }
 
         if (device_param->is_hip == true)
         {
-          if (hc_hipMemcpyDtoH (hashcat_ctx, device_param->hooks_buf, device_param->hip_d_hooks, pws_cnt * hashconfig->hook_size) == -1) return -1;
+          if (hc_hipMemcpyDtoHAsync (hashcat_ctx, device_param->hooks_buf, device_param->hip_d_hooks, pws_cnt * hashconfig->hook_size, device_param->hip_stream) == -1) return -1;
+
+          if (hc_hipStreamSynchronize (hashcat_ctx, device_param->hip_stream) == -1) return -1;
         }
 
         if (device_param->is_opencl == true)
         {
+          /* blocking */
           if (hc_clEnqueueReadBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_hooks, CL_TRUE, 0, pws_cnt * hashconfig->hook_size, device_param->hooks_buf, 0, NULL, NULL) == -1) return -1;
         }
 
         const int hook_threads = (int) user_options->hook_threads;
 
-        hook_thread_param_t *hook_threads_param = (hook_thread_param_t *) hccalloc (hook_threads, sizeof (hook_thread_param_t));
+        hook_thread_param_t *hook_threads_param = (hook_thread_param_t *) hcmalloc (hook_threads * sizeof (hook_thread_param_t));
+        hc_thread_t         *c_threads          = (hc_thread_t *)         hcmalloc (hook_threads * sizeof (hc_thread_t));
 
         for (int i = 0; i < hook_threads; i++)
         {
@@ -4599,13 +4524,6 @@ int choose_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param,
           hook_thread_param->salt_pos = salt_pos;
 
           hook_thread_param->pws_cnt = pws_cnt;
-        }
-
-        hc_thread_t *c_threads = (hc_thread_t *) hccalloc (hook_threads, sizeof (hc_thread_t));
-
-        for (int i = 0; i < hook_threads; i++)
-        {
-          hook_thread_param_t *hook_thread_param = hook_threads_param + i;
 
           hc_thread_create (c_threads[i], hook12_thread, hook_thread_param);
         }
@@ -4613,22 +4531,21 @@ int choose_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param,
         hc_thread_wait (hook_threads, c_threads);
 
         hcfree (c_threads);
-
         hcfree (hook_threads_param);
 
         if (device_param->is_cuda == true)
         {
-          if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_hooks, device_param->hooks_buf, pws_cnt * hashconfig->hook_size) == -1) return -1;
+          if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_hooks, device_param->hooks_buf, pws_cnt * hashconfig->hook_size, device_param->cuda_stream) == -1) return -1;
         }
 
         if (device_param->is_hip == true)
         {
-          if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_hooks, device_param->hooks_buf, pws_cnt * hashconfig->hook_size) == -1) return -1;
+          if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_hooks, device_param->hooks_buf, pws_cnt * hashconfig->hook_size, device_param->hip_stream) == -1) return -1;
         }
 
         if (device_param->is_opencl == true)
         {
-          if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_hooks, CL_TRUE, 0, pws_cnt * hashconfig->hook_size, device_param->hooks_buf, 0, NULL, NULL) == -1) return -1;
+          if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_hooks, CL_FALSE, 0, pws_cnt * hashconfig->hook_size, device_param->hooks_buf, 0, NULL, NULL) == -1) return -1;
         }
       }
     }
@@ -4712,22 +4629,28 @@ int choose_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param,
 
             if (device_param->is_cuda == true)
             {
-              if (hc_cuMemcpyDtoH (hashcat_ctx, device_param->hooks_buf, device_param->cuda_d_hooks, pws_cnt * hashconfig->hook_size) == -1) return -1;
+              if (hc_cuMemcpyDtoHAsync (hashcat_ctx, device_param->hooks_buf, device_param->cuda_d_hooks, pws_cnt * hashconfig->hook_size, device_param->cuda_stream) == -1) return -1;
+
+              if (hc_cuStreamSynchronize (hashcat_ctx, device_param->cuda_stream) == -1) return -1;
             }
 
             if (device_param->is_hip == true)
             {
-              if (hc_hipMemcpyDtoH (hashcat_ctx, device_param->hooks_buf, device_param->hip_d_hooks, pws_cnt * hashconfig->hook_size) == -1) return -1;
+              if (hc_hipMemcpyDtoHAsync (hashcat_ctx, device_param->hooks_buf, device_param->hip_d_hooks, pws_cnt * hashconfig->hook_size, device_param->hip_stream) == -1) return -1;
+
+              if (hc_hipStreamSynchronize (hashcat_ctx, device_param->hip_stream) == -1) return -1;
             }
 
             if (device_param->is_opencl == true)
             {
+              /* blocking */
               if (hc_clEnqueueReadBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_hooks, CL_TRUE, 0, pws_cnt * hashconfig->hook_size, device_param->hooks_buf, 0, NULL, NULL) == -1) return -1;
             }
 
             const int hook_threads = (int) user_options->hook_threads;
 
-            hook_thread_param_t *hook_threads_param = (hook_thread_param_t *) hccalloc (hook_threads, sizeof (hook_thread_param_t));
+            hook_thread_param_t *hook_threads_param = (hook_thread_param_t *) hcmalloc (hook_threads * sizeof (hook_thread_param_t));
+            hc_thread_t         *c_threads          = (hc_thread_t *)         hcmalloc (hook_threads * sizeof (hc_thread_t));
 
             for (int i = 0; i < hook_threads; i++)
             {
@@ -4747,13 +4670,6 @@ int choose_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param,
               hook_thread_param->salt_pos = salt_pos;
 
               hook_thread_param->pws_cnt = pws_cnt;
-            }
-
-            hc_thread_t *c_threads = (hc_thread_t *) hccalloc (hook_threads, sizeof (hc_thread_t));
-
-            for (int i = 0; i < hook_threads; i++)
-            {
-              hook_thread_param_t *hook_thread_param = hook_threads_param + i;
 
               hc_thread_create (c_threads[i], hook23_thread, hook_thread_param);
             }
@@ -4761,22 +4677,21 @@ int choose_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param,
             hc_thread_wait (hook_threads, c_threads);
 
             hcfree (c_threads);
-
             hcfree (hook_threads_param);
 
             if (device_param->is_cuda == true)
             {
-              if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_hooks, device_param->hooks_buf, pws_cnt * hashconfig->hook_size) == -1) return -1;
+              if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_hooks, device_param->hooks_buf, pws_cnt * hashconfig->hook_size, device_param->cuda_stream) == -1) return -1;
             }
 
             if (device_param->is_hip == true)
             {
-              if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_hooks, device_param->hooks_buf, pws_cnt * hashconfig->hook_size) == -1) return -1;
+              if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_hooks, device_param->hooks_buf, pws_cnt * hashconfig->hook_size, device_param->hip_stream) == -1) return -1;
             }
 
             if (device_param->is_opencl == true)
             {
-              if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_hooks, CL_TRUE, 0, pws_cnt * hashconfig->hook_size, device_param->hooks_buf, 0, NULL, NULL) == -1) return -1;
+              if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_hooks, CL_FALSE, 0, pws_cnt * hashconfig->hook_size, device_param->hooks_buf, 0, NULL, NULL) == -1) return -1;
             }
           }
         }
@@ -4878,12 +4793,12 @@ int choose_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param,
 
     if (device_param->is_cuda == true)
     {
-      if (run_cuda_kernel_bzero   (hashcat_ctx, device_param, device_param->cuda_d_tmps,   device_param->size_tmps) == -1) return -1;
+      if (run_cuda_kernel_bzero (hashcat_ctx, device_param, device_param->cuda_d_tmps,   device_param->size_tmps) == -1) return -1;
     }
 
     if (device_param->is_hip == true)
     {
-      if (run_hip_kernel_bzero    (hashcat_ctx, device_param, device_param->hip_d_tmps,    device_param->size_tmps) == -1) return -1;
+      if (run_hip_kernel_bzero (hashcat_ctx, device_param, device_param->hip_d_tmps,    device_param->size_tmps) == -1) return -1;
     }
 
     if (device_param->is_opencl == true)
@@ -4896,12 +4811,12 @@ int choose_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param,
     {
       if (device_param->is_cuda == true)
       {
-        if (run_cuda_kernel_bzero   (hashcat_ctx, device_param, device_param->cuda_d_hooks,   pws_cnt * hashconfig->hook_size) == -1) return -1;
+        if (run_cuda_kernel_bzero (hashcat_ctx, device_param, device_param->cuda_d_hooks, pws_cnt * hashconfig->hook_size) == -1) return -1;
       }
 
       if (device_param->is_hip == true)
       {
-        if (run_hip_kernel_bzero    (hashcat_ctx, device_param, device_param->hip_d_hooks,    pws_cnt * hashconfig->hook_size) == -1) return -1;
+        if (run_hip_kernel_bzero (hashcat_ctx, device_param, device_param->hip_d_hooks, pws_cnt * hashconfig->hook_size) == -1) return -1;
       }
 
       if (device_param->is_opencl == true)
@@ -4980,8 +4895,6 @@ int run_cuda_kernel_atinit (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *devic
 
   if (hc_cuLaunchKernel (hashcat_ctx, function, num_elements, 1, 1, kernel_threads, 1, 1, 0, device_param->cuda_stream, device_param->kernel_params_atinit, NULL) == -1) return -1;
 
-  if (hc_cuStreamSynchronize (hashcat_ctx, device_param->cuda_stream) == -1) return -1;
-
   return 0;
 }
 
@@ -5000,65 +4913,53 @@ int run_cuda_kernel_utf8toutf16le (hashcat_ctx_t *hashcat_ctx, hc_device_param_t
 
   if (hc_cuLaunchKernel (hashcat_ctx, function, num_elements, 1, 1, kernel_threads, 1, 1, 0, device_param->cuda_stream, device_param->kernel_params_utf8toutf16le, NULL) == -1) return -1;
 
-  if (hc_cuStreamSynchronize (hashcat_ctx, device_param->cuda_stream) == -1) return -1;
-
   return 0;
 }
 
-int run_cuda_kernel_memset (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, CUdeviceptr buf, const u32 value, const u64 size)
+int run_cuda_kernel_memset (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, CUdeviceptr buf, const u64 offset, const u8 value, const u64 size)
+{
+  return hc_cuMemsetD8Async (hashcat_ctx, buf + offset, value, size, device_param->cuda_stream);
+}
+
+int run_cuda_kernel_memset32 (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, CUdeviceptr buf, const u64 offset, const u32 value, const u64 size)
+{
+  /* check that the size is multiple of element size */
+  if (size % 4 != 0)
+  {
+    return CUDA_ERROR_INVALID_VALUE;
+  }
+
+  return hc_cuMemsetD32Async (hashcat_ctx, buf + offset, value, size / 4, device_param->cuda_stream);
+}
+
+int run_cuda_kernel_bzero (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, CUdeviceptr buf, const u64 size)
 {
   const u64 num16d = size / 16;
   const u64 num16m = size % 16;
 
   if (num16d)
   {
-    device_param->kernel_params_memset[0]       = (void *) &buf;
-    device_param->kernel_params_memset_buf32[1] = value;
-    device_param->kernel_params_memset_buf64[2] = num16d;
+    device_param->kernel_params_bzero[0]       = (void *) &buf;
+    device_param->kernel_params_bzero_buf64[1] = num16d;
 
-    const u64 kernel_threads = device_param->kernel_wgs_memset;
+    const u64 kernel_threads = device_param->kernel_wgs_bzero;
 
-    u64 num_elements = num16d;
+    u64 num_elements = CEILDIV (num16d, kernel_threads);
 
-    num_elements = CEILDIV (num_elements, kernel_threads);
+    CUfunction function = device_param->cuda_function_bzero;
 
-    CUfunction function = device_param->cuda_function_memset;
-
-    //CU_rc = hc_clSetKernelArg (hashcat_ctx, kernel, 0, sizeof (cl_mem),   (void *) &buf);                         if (CU_rc == -1) return -1;
-    //CU_rc = hc_clSetKernelArg (hashcat_ctx, kernel, 1, sizeof (cl_uint),  device_param->kernel_params_memset[1]); if (CU_rc == -1) return -1;
-    //CU_rc = hc_clSetKernelArg (hashcat_ctx, kernel, 2, sizeof (cl_ulong), device_param->kernel_params_memset[2]); if (CU_rc == -1) return -1;
-
-    //const size_t global_work_size[3] = { num_elements,   1, 1 };
-    //const size_t local_work_size[3]  = { kernel_threads, 1, 1 };
-
-    if (hc_cuLaunchKernel (hashcat_ctx, function, num_elements, 1, 1, kernel_threads, 1, 1, 0, device_param->cuda_stream, device_param->kernel_params_memset, NULL) == -1) return -1;
-
-    if (hc_cuStreamSynchronize (hashcat_ctx, device_param->cuda_stream) == -1) return -1;
+    if (hc_cuLaunchKernel (hashcat_ctx, function, num_elements, 1, 1, kernel_threads, 1, 1, 0, device_param->cuda_stream, device_param->kernel_params_bzero, NULL) == -1) return -1;
   }
 
   if (num16m)
   {
-    u32 tmp[4];
-
-    tmp[0] = value;
-    tmp[1] = value;
-    tmp[2] = value;
-    tmp[3] = value;
-
-    // Apparently are allowed to do this: https://devtalk.nvidia.com/default/topic/761515/how-to-copy-to-device-memory-with-offset-/
-
-    if (hc_cuMemcpyHtoD (hashcat_ctx, buf + (num16d * 16), tmp, num16m) == -1) return -1;
+    if (hc_cuMemcpyHtoDAsync (hashcat_ctx, buf + (num16d * 16), bzeros, num16m, device_param->cuda_stream) == -1) return -1;
   }
 
   return 0;
 }
 
-int run_cuda_kernel_bzero (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, CUdeviceptr buf, const u64 size)
-{
-  return run_cuda_kernel_memset (hashcat_ctx, device_param, buf, 0, size);
-}
-
-int run_hip_kernel_atinit (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, HIPdeviceptr buf, const u64 num)
+int run_hip_kernel_atinit (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, hipDeviceptr_t buf, const u64 num)
 {
   u64 num_elements = num;
 
@@ -5069,16 +4970,14 @@ int run_hip_kernel_atinit (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device
 
   num_elements = CEILDIV (num_elements, kernel_threads);
 
-  HIPfunction function = device_param->hip_function_atinit;
+  hipFunction_t function = device_param->hip_function_atinit;
 
   if (hc_hipLaunchKernel (hashcat_ctx, function, num_elements, 1, 1, kernel_threads, 1, 1, 0, device_param->hip_stream, device_param->kernel_params_atinit, NULL) == -1) return -1;
 
-  if (hc_hipStreamSynchronize (hashcat_ctx, device_param->hip_stream) == -1) return -1;
-
   return 0;
 }
 
-int run_hip_kernel_utf8toutf16le (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, HIPdeviceptr buf, const u64 num)
+int run_hip_kernel_utf8toutf16le (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, hipDeviceptr_t buf, const u64 num)
 {
   u64 num_elements = num;
 
@@ -5089,68 +4988,56 @@ int run_hip_kernel_utf8toutf16le (hashcat_ctx_t *hashcat_ctx, hc_device_param_t
 
   num_elements = CEILDIV (num_elements, kernel_threads);
 
-  HIPfunction function = device_param->hip_function_utf8toutf16le;
+  hipFunction_t function = device_param->hip_function_utf8toutf16le;
 
   if (hc_hipLaunchKernel (hashcat_ctx, function, num_elements, 1, 1, kernel_threads, 1, 1, 0, device_param->hip_stream, device_param->kernel_params_utf8toutf16le, NULL) == -1) return -1;
 
-  if (hc_hipStreamSynchronize (hashcat_ctx, device_param->hip_stream) == -1) return -1;
-
   return 0;
 }
 
-int run_hip_kernel_memset (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, HIPdeviceptr buf, const u32 value, const u64 size)
+int run_hip_kernel_memset (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, hipDeviceptr_t buf, const u64 offset, const u8  value, const u64 size)
+{
+  return hc_hipMemsetD8Async (hashcat_ctx, buf + offset, value, size, device_param->hip_stream);
+}
+
+int run_hip_kernel_memset32 (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, hipDeviceptr_t buf, const u64 offset, const u32 value, const u64 size)
+{
+  /* check that the size is multiple of element size */
+  if (size % 4 != 0)
+  {
+    return hipErrorInvalidValue;
+  }
+
+  return hc_hipMemsetD32Async (hashcat_ctx, buf + offset, value, size / 4, device_param->hip_stream);
+}
+
+int run_hip_kernel_bzero (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, hipDeviceptr_t buf, const u64 size)
 {
   const u64 num16d = size / 16;
   const u64 num16m = size % 16;
 
   if (num16d)
   {
-    device_param->kernel_params_memset[0]       = (void *) &buf;
-    device_param->kernel_params_memset_buf32[1] = value;
-    device_param->kernel_params_memset_buf64[2] = num16d;
+    device_param->kernel_params_bzero[0]       = (void *) &buf;
+    device_param->kernel_params_bzero_buf64[1] = num16d;
 
-    const u64 kernel_threads = device_param->kernel_wgs_memset;
+    const u64 kernel_threads = device_param->kernel_wgs_bzero;
 
-    u64 num_elements = num16d;
+    u64 num_elements = CEILDIV(num16d, kernel_threads);
 
-    num_elements = CEILDIV (num_elements, kernel_threads);
+    hipFunction_t function = device_param->hip_function_bzero;
 
-    HIPfunction function = device_param->hip_function_memset;
-
-    //HIP_rc = hc_clSetKernelArg (hashcat_ctx, kernel, 0, sizeof (cl_mem),   (void *) &buf);                         if (HIP_rc == -1) return -1;
-    //HIP_rc = hc_clSetKernelArg (hashcat_ctx, kernel, 1, sizeof (cl_uint),  device_param->kernel_params_memset[1]); if (HIP_rc == -1) return -1;
-    //HIP_rc = hc_clSetKernelArg (hashcat_ctx, kernel, 2, sizeof (cl_ulong), device_param->kernel_params_memset[2]); if (HIP_rc == -1) return -1;
-
-    //const size_t global_work_size[3] = { num_elements,   1, 1 };
-    //const size_t local_work_size[3]  = { kernel_threads, 1, 1 };
-
-    if (hc_hipLaunchKernel (hashcat_ctx, function, num_elements, 1, 1, kernel_threads, 1, 1, 0, device_param->hip_stream, device_param->kernel_params_memset, NULL) == -1) return -1;
-
-    if (hc_hipStreamSynchronize (hashcat_ctx, device_param->hip_stream) == -1) return -1;
+    if (hc_hipLaunchKernel (hashcat_ctx, function, num_elements, 1, 1, kernel_threads, 1, 1, 0, device_param->hip_stream, device_param->kernel_params_bzero, NULL) == -1) return -1;
   }
 
   if (num16m)
   {
-    u32 tmp[4];
-
-    tmp[0] = value;
-    tmp[1] = value;
-    tmp[2] = value;
-    tmp[3] = value;
-
-    // Apparently are allowed to do this: https://devtalk.nvidia.com/default/topic/761515/how-to-copy-to-device-memory-with-offset-/
-
-    if (hc_hipMemcpyHtoD (hashcat_ctx, buf + (num16d * 16), tmp, num16m) == -1) return -1;
+    if (hc_hipMemcpyHtoDAsync (hashcat_ctx, buf + (num16d * 16), bzeros, num16m, device_param->hip_stream) == -1) return -1;
   }
 
   return 0;
 }
 
-int run_hip_kernel_bzero (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, HIPdeviceptr buf, const u64 size)
-{
-  return run_hip_kernel_memset (hashcat_ctx, device_param, buf, 0, size);
-}
-
 int run_opencl_kernel_atinit (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, cl_mem buf, const u64 num)
 {
   u64 num_elements = num;
@@ -5172,10 +5059,6 @@ int run_opencl_kernel_atinit (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *dev
 
   if (hc_clEnqueueNDRangeKernel (hashcat_ctx, device_param->opencl_command_queue, kernel, 1, NULL, global_work_size, local_work_size, 0, NULL, NULL) == -1) return -1;
 
-  if (hc_clFlush (hashcat_ctx, device_param->opencl_command_queue) == -1) return -1;
-
-  if (hc_clFinish (hashcat_ctx, device_param->opencl_command_queue) == -1) return -1;
-
   return 0;
 }
 
@@ -5200,65 +5083,104 @@ int run_opencl_kernel_utf8toutf16le (hashcat_ctx_t *hashcat_ctx, hc_device_param
 
   if (hc_clEnqueueNDRangeKernel (hashcat_ctx, device_param->opencl_command_queue, kernel, 1, NULL, global_work_size, local_work_size, 0, NULL, NULL) == -1) return -1;
 
-  if (hc_clFlush (hashcat_ctx, device_param->opencl_command_queue) == -1) return -1;
-
-  if (hc_clFinish (hashcat_ctx, device_param->opencl_command_queue) == -1) return -1;
-
   return 0;
 }
 
-int run_opencl_kernel_memset (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, cl_mem buf, const u32 value, const u64 size)
+int run_opencl_kernel_memset (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, cl_mem buf, const u64 offset, const u8 value, const u64 size)
+{
+  const backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
+  const OCL_PTR       *ocl         = backend_ctx->ocl;
+
+  int rc;
+
+  /* workaround if missing clEnqueueFillBuffer() */
+  if (ocl->clEnqueueFillBuffer == NULL)
+  {
+    char *tmp = hcmalloc (size * sizeof (u8));
+
+    memset (tmp, value, size);
+
+    /* blocking */
+    rc = hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, buf, CL_TRUE, offset, size, tmp, 0, NULL, NULL);
+
+    hcfree (tmp);
+  }
+  else
+  {
+    rc = hc_clEnqueueFillBuffer (hashcat_ctx, device_param->opencl_command_queue, buf, &value, sizeof (u8), offset, size, 0, NULL, NULL);
+  }
+
+  return rc;
+}
+
+int run_opencl_kernel_memset32 (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, cl_mem buf, const u64 offset, const u32 value, const u64 size)
+{
+  const backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
+  const OCL_PTR       *ocl         = backend_ctx->ocl;
+
+  int rc;
+
+  /* workaround if missing clEnqueueFillBuffer() */
+  if (ocl->clEnqueueFillBuffer == NULL)
+  {
+    const u64 N = size / 4;
+
+    /* check that the size is multiple of element size */
+    if (size % 4 != 0)
+    {
+      return CL_INVALID_VALUE;
+    }
+
+    u32 *tmp = (u32 *) hcmalloc (size);
+
+    for (u64 i = 0; i < N; i++)
+    {
+      tmp[i] = value;
+    }
+
+    /* blocking */
+    rc = hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, buf, CL_TRUE, offset, size, tmp, 0, NULL, NULL);
+
+    hcfree (tmp);
+  }
+  else
+  {
+    rc = hc_clEnqueueFillBuffer (hashcat_ctx, device_param->opencl_command_queue, buf, &value, sizeof (u32), offset, size, 0, NULL, NULL);
+  }
+
+  return rc;
+}
+
+int run_opencl_kernel_bzero (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, cl_mem buf, const u64 size)
 {
   const u64 num16d = size / 16;
   const u64 num16m = size % 16;
 
   if (num16d)
   {
-    device_param->kernel_params_memset_buf32[1] = value;
-    device_param->kernel_params_memset_buf64[2] = num16d;
+    const u64 kernel_threads = device_param->kernel_wgs_bzero;
 
-    const u64 kernel_threads = device_param->kernel_wgs_memset;
+    u64 num_elements = round_up_multiple_64(num16d, kernel_threads);
 
-    u64 num_elements = num16d;
+    cl_kernel kernel = device_param->opencl_kernel_bzero;
 
-    num_elements = round_up_multiple_64 (num_elements, kernel_threads);
-
-    cl_kernel kernel = device_param->opencl_kernel_memset;
-
-    if (hc_clSetKernelArg (hashcat_ctx, kernel, 0, sizeof (cl_mem),   (void *) &buf) == -1)                         return -1;
-    if (hc_clSetKernelArg (hashcat_ctx, kernel, 1, sizeof (cl_uint),  device_param->kernel_params_memset[1]) == -1) return -1;
-    if (hc_clSetKernelArg (hashcat_ctx, kernel, 2, sizeof (cl_ulong), device_param->kernel_params_memset[2]) == -1) return -1;
+    if (hc_clSetKernelArg (hashcat_ctx, kernel, 0, sizeof(cl_mem),   (void *) &buf)    == -1) return -1;
+    if (hc_clSetKernelArg (hashcat_ctx, kernel, 1, sizeof(cl_ulong), (void *) &num16d) == -1) return -1;
 
     const size_t global_work_size[3] = { num_elements,   1, 1 };
     const size_t local_work_size[3]  = { kernel_threads, 1, 1 };
 
     if (hc_clEnqueueNDRangeKernel (hashcat_ctx, device_param->opencl_command_queue, kernel, 1, NULL, global_work_size, local_work_size, 0, NULL, NULL) == -1) return -1;
-
-    if (hc_clFlush (hashcat_ctx, device_param->opencl_command_queue) == -1) return -1;
-
-    if (hc_clFinish (hashcat_ctx, device_param->opencl_command_queue) == -1) return -1;
   }
 
   if (num16m)
   {
-    u32 tmp[4];
-
-    tmp[0] = value;
-    tmp[1] = value;
-    tmp[2] = value;
-    tmp[3] = value;
-
-    if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, buf, CL_TRUE, num16d * 16, num16m, tmp, 0, NULL, NULL) == -1) return -1;
+    if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, buf, CL_FALSE, num16d * 16, num16m, bzeros, 0, NULL, NULL) == -1) return -1;
   }
 
   return 0;
 }
 
-int run_opencl_kernel_bzero (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, cl_mem buf, const u64 size)
-{
-  return run_opencl_kernel_memset (hashcat_ctx, device_param, buf, 0, size);
-}
-
 int run_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, const u32 kern_run, const u64 pws_pos, const u64 num, const u32 event_update, const u32 iteration)
 {
   const hashconfig_t   *hashconfig   = hashcat_ctx->hashconfig;
@@ -5429,10 +5351,10 @@ int run_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, con
 
     if (hc_cuEventRecord (hashcat_ctx, device_param->cuda_event2, device_param->cuda_stream) == -1) return -1;
 
-    if (hc_cuStreamSynchronize (hashcat_ctx, device_param->cuda_stream) == -1) return -1;
-
     if (hc_cuEventSynchronize (hashcat_ctx, device_param->cuda_event2) == -1) return -1;
 
+    if (hc_cuEventSynchronize (hashcat_ctx, device_param->cuda_event1) == -1) return -1;
+
     float exec_ms;
 
     if (hc_cuEventElapsedTime (hashcat_ctx, &exec_ms, device_param->cuda_event1, device_param->cuda_event2) == -1) return -1;
@@ -5456,7 +5378,7 @@ int run_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, con
 
   if (device_param->is_hip == true)
   {
-    HIPfunction hip_function = NULL;
+    hipFunction_t hip_function = NULL;
 
     if (device_param->is_hip == true)
     {
@@ -5528,8 +5450,6 @@ int run_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, con
 
     if (hc_hipEventRecord (hashcat_ctx, device_param->hip_event2, device_param->hip_stream) == -1) return -1;
 
-    if (hc_hipStreamSynchronize (hashcat_ctx, device_param->hip_stream) == -1) return -1;
-
     if (hc_hipEventSynchronize (hashcat_ctx, device_param->hip_event2) == -1) return -1;
 
     float exec_ms;
@@ -5627,8 +5547,6 @@ int run_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, con
 
     if (hc_clEnqueueNDRangeKernel (hashcat_ctx, device_param->opencl_command_queue, opencl_kernel, 1, NULL, global_work_size, local_work_size, 0, NULL, &opencl_event) == -1) return -1;
 
-    if (hc_clFlush (hashcat_ctx, device_param->opencl_command_queue) == -1) return -1;
-
     // spin damper section
 
     const u32 iterationm = iteration % EXPECTED_ITERATIONS;
@@ -5639,6 +5557,8 @@ int run_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, con
 
       size_t param_value_size_ret;
 
+      if (hc_clFlush (hashcat_ctx, device_param->opencl_command_queue) == -1) return -1;
+
       if (hc_clGetEventInfo (hashcat_ctx, opencl_event, CL_EVENT_COMMAND_EXECUTION_STATUS, sizeof (opencl_event_status), &opencl_event_status, &param_value_size_ret) == -1) return -1;
 
       double spin_total = device_param->spin_damp;
@@ -5675,11 +5595,18 @@ int run_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, con
 
         spin_total += device_param->spin_damp;
 
-        if (spin_total > 1) break;
+        if (spin_total > 1)
+        {
+          if (hc_clWaitForEvents (hashcat_ctx, 1, &opencl_event) == -1) return -1;
+
+          break;
+        }
       }
     }
-
-    if (hc_clWaitForEvents (hashcat_ctx, 1, &opencl_event) == -1) return -1;
+    else
+    {
+      if (hc_clWaitForEvents (hashcat_ctx, 1, &opencl_event) == -1) return -1;
+    }
 
     cl_ulong time_start;
     cl_ulong time_end;
@@ -5729,8 +5656,6 @@ int run_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, con
     }
 
     if (hc_clReleaseEvent (hashcat_ctx, opencl_event) == -1) return -1;
-
-    if (hc_clFinish (hashcat_ctx, device_param->opencl_command_queue) == -1) return -1;
   }
 
   return 0;
@@ -5778,13 +5703,11 @@ int run_kernel_mp (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param,
     num_elements = CEILDIV (num_elements, kernel_threads);
 
     if (hc_cuLaunchKernel (hashcat_ctx, cuda_function, num_elements, 1, 1, kernel_threads, 1, 1, 0, device_param->cuda_stream, cuda_args, NULL) == -1) return -1;
-
-    if (hc_cuStreamSynchronize (hashcat_ctx, device_param->cuda_stream) == -1) return -1;
   }
 
   if (device_param->is_hip == true)
   {
-    HIPfunction hip_function = NULL;
+    hipFunction_t hip_function = NULL;
 
     void **hip_args = NULL;
 
@@ -5804,8 +5727,6 @@ int run_kernel_mp (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param,
     num_elements = CEILDIV (num_elements, kernel_threads);
 
     if (hc_hipLaunchKernel (hashcat_ctx, hip_function, num_elements, 1, 1, kernel_threads, 1, 1, 0, device_param->hip_stream, hip_args, NULL) == -1) return -1;
-
-    if (hc_hipStreamSynchronize (hashcat_ctx, device_param->hip_stream) == -1) return -1;
   }
 
   if (device_param->is_opencl == true)
@@ -5851,10 +5772,6 @@ int run_kernel_mp (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param,
     const size_t local_work_size[3]  = { kernel_threads, 1, 1 };
 
     if (hc_clEnqueueNDRangeKernel (hashcat_ctx, device_param->opencl_command_queue, opencl_kernel, 1, NULL, global_work_size, local_work_size, 0, NULL, NULL) == -1) return -1;
-
-    if (hc_clFlush (hashcat_ctx, device_param->opencl_command_queue)  == -1) return -1;
-
-    if (hc_clFinish (hashcat_ctx, device_param->opencl_command_queue) == -1) return -1;
   }
 
   return 0;
@@ -5871,17 +5788,13 @@ int run_kernel_tm (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param)
     CUfunction cuda_function = device_param->cuda_function_tm;
 
     if (hc_cuLaunchKernel (hashcat_ctx, cuda_function, num_elements / kernel_threads, 1, 1, kernel_threads, 1, 1, 0, device_param->cuda_stream, device_param->kernel_params_tm, NULL) == -1) return -1;
-
-    if (hc_cuStreamSynchronize (hashcat_ctx, device_param->cuda_stream) == -1) return -1;
   }
 
   if (device_param->is_hip == true)
   {
-    HIPfunction hip_function = device_param->hip_function_tm;
+    hipFunction_t hip_function = device_param->hip_function_tm;
 
     if (hc_hipLaunchKernel (hashcat_ctx, hip_function, num_elements / kernel_threads, 1, 1, kernel_threads, 1, 1, 0, device_param->hip_stream, device_param->kernel_params_tm, NULL) == -1) return -1;
-
-    if (hc_hipStreamSynchronize (hashcat_ctx, device_param->hip_stream) == -1) return -1;
   }
 
   if (device_param->is_opencl == true)
@@ -5892,10 +5805,6 @@ int run_kernel_tm (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param)
     const size_t local_work_size[3]  = { kernel_threads,  1, 1 };
 
     if (hc_clEnqueueNDRangeKernel (hashcat_ctx, device_param->opencl_command_queue, cuda_kernel, 1, NULL, global_work_size, local_work_size, 0, NULL, NULL) == -1) return -1;
-
-    if (hc_clFlush (hashcat_ctx, device_param->opencl_command_queue) == -1) return -1;
-
-    if (hc_clFinish (hashcat_ctx, device_param->opencl_command_queue) == -1) return -1;
   }
 
   return 0;
@@ -5916,19 +5825,15 @@ int run_kernel_amp (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param,
     CUfunction cuda_function = device_param->cuda_function_amp;
 
     if (hc_cuLaunchKernel (hashcat_ctx, cuda_function, num_elements, 1, 1, kernel_threads, 1, 1, 0, device_param->cuda_stream, device_param->kernel_params_amp, NULL) == -1) return -1;
-
-    if (hc_cuStreamSynchronize (hashcat_ctx, device_param->cuda_stream) == -1) return -1;
   }
 
   if (device_param->is_hip == true)
   {
     num_elements = CEILDIV (num_elements, kernel_threads);
 
-    HIPfunction hip_function = device_param->hip_function_amp;
+    hipFunction_t hip_function = device_param->hip_function_amp;
 
     if (hc_hipLaunchKernel (hashcat_ctx, hip_function, num_elements, 1, 1, kernel_threads, 1, 1, 0, device_param->hip_stream, device_param->kernel_params_amp, NULL) == -1) return -1;
-
-    if (hc_hipStreamSynchronize (hashcat_ctx, device_param->hip_stream) == -1) return -1;
   }
 
   if (device_param->is_opencl == true)
@@ -5943,10 +5848,6 @@ int run_kernel_amp (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param,
     const size_t local_work_size[3]  = { kernel_threads,  1, 1 };
 
     if (hc_clEnqueueNDRangeKernel (hashcat_ctx, device_param->opencl_command_queue, opencl_kernel, 1, NULL, global_work_size, local_work_size, 0, NULL, NULL) == -1) return -1;
-
-    if (hc_clFlush (hashcat_ctx, device_param->opencl_command_queue)  == -1) return -1;
-
-    if (hc_clFinish (hashcat_ctx, device_param->opencl_command_queue) == -1) return -1;
   }
 
   return 0;
@@ -5967,19 +5868,15 @@ int run_kernel_decompress (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device
     CUfunction cuda_function = device_param->cuda_function_decompress;
 
     if (hc_cuLaunchKernel (hashcat_ctx, cuda_function, num_elements, 1, 1, kernel_threads, 1, 1, 0, device_param->cuda_stream, device_param->kernel_params_decompress, NULL) == -1) return -1;
-
-    if (hc_cuStreamSynchronize (hashcat_ctx, device_param->cuda_stream) == -1) return -1;
   }
 
   if (device_param->is_hip == true)
   {
     num_elements = CEILDIV (num_elements, kernel_threads);
 
-    HIPfunction hip_function = device_param->hip_function_decompress;
+    hipFunction_t hip_function = device_param->hip_function_decompress;
 
     if (hc_hipLaunchKernel (hashcat_ctx, hip_function, num_elements, 1, 1, kernel_threads, 1, 1, 0, device_param->hip_stream, device_param->kernel_params_decompress, NULL) == -1) return -1;
-
-    if (hc_hipStreamSynchronize (hashcat_ctx, device_param->hip_stream) == -1) return -1;
   }
 
   if (device_param->is_opencl == true)
@@ -5994,10 +5891,6 @@ int run_kernel_decompress (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device
     if (hc_clSetKernelArg (hashcat_ctx, opencl_kernel, 3, sizeof (cl_ulong), device_param->kernel_params_decompress[3]) == -1) return -1;
 
     if (hc_clEnqueueNDRangeKernel (hashcat_ctx, device_param->opencl_command_queue, opencl_kernel, 1, NULL, global_work_size, local_work_size, 0, NULL, NULL) == -1) return -1;
-
-    if (hc_clFlush (hashcat_ctx, device_param->opencl_command_queue) == -1) return -1;
-
-    if (hc_clFinish (hashcat_ctx, device_param->opencl_command_queue) == -1) return -1;
   }
 
   return 0;
@@ -6028,7 +5921,7 @@ int run_copy (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, const
   {
     if (device_param->is_cuda == true)
     {
-      if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_pws_idx, device_param->pws_idx, pws_cnt * sizeof (pw_idx_t)) == -1) return -1;
+      if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_pws_idx, device_param->pws_idx, pws_cnt * sizeof (pw_idx_t), device_param->cuda_stream) == -1) return -1;
 
       const pw_idx_t *pw_idx = device_param->pws_idx + pws_cnt;
 
@@ -6036,13 +5929,13 @@ int run_copy (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, const
 
       if (off)
       {
-        if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_pws_comp_buf, device_param->pws_comp, off * sizeof (u32)) == -1) return -1;
+        if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_pws_comp_buf, device_param->pws_comp, off * sizeof (u32), device_param->cuda_stream) == -1) return -1;
       }
     }
 
     if (device_param->is_hip == true)
     {
-      if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_pws_idx, device_param->pws_idx, pws_cnt * sizeof (pw_idx_t)) == -1) return -1;
+      if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_pws_idx, device_param->pws_idx, pws_cnt * sizeof (pw_idx_t), device_param->hip_stream) == -1) return -1;
 
       const pw_idx_t *pw_idx = device_param->pws_idx + pws_cnt;
 
@@ -6050,13 +5943,13 @@ int run_copy (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, const
 
       if (off)
       {
-        if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_pws_comp_buf, device_param->pws_comp, off * sizeof (u32)) == -1) return -1;
+        if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_pws_comp_buf, device_param->pws_comp, off * sizeof (u32), device_param->hip_stream) == -1) return -1;
       }
     }
 
     if (device_param->is_opencl == true)
     {
-      if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_idx, CL_TRUE, 0, pws_cnt * sizeof (pw_idx_t), device_param->pws_idx, 0, NULL, NULL) == -1) return -1;
+      if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_idx, CL_FALSE, 0, pws_cnt * sizeof (pw_idx_t), device_param->pws_idx, 0, NULL, NULL) == -1) return -1;
 
       const pw_idx_t *pw_idx = device_param->pws_idx + pws_cnt;
 
@@ -6064,7 +5957,7 @@ int run_copy (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, const
 
       if (off)
       {
-        if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_comp_buf, CL_TRUE, 0, off * sizeof (u32), device_param->pws_comp, 0, NULL, NULL) == -1) return -1;
+        if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_comp_buf, CL_FALSE, 0, off * sizeof (u32), device_param->pws_comp, 0, NULL, NULL) == -1) return -1;
       }
     }
 
@@ -6076,7 +5969,7 @@ int run_copy (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, const
     {
       if (device_param->is_cuda == true)
       {
-        if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_pws_idx, device_param->pws_idx, pws_cnt * sizeof (pw_idx_t)) == -1) return -1;
+        if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_pws_idx, device_param->pws_idx, pws_cnt * sizeof (pw_idx_t), device_param->cuda_stream) == -1) return -1;
 
         const pw_idx_t *pw_idx = device_param->pws_idx + pws_cnt;
 
@@ -6084,13 +5977,13 @@ int run_copy (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, const
 
         if (off)
         {
-          if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_pws_comp_buf, device_param->pws_comp, off * sizeof (u32)) == -1) return -1;
+          if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_pws_comp_buf, device_param->pws_comp, off * sizeof (u32), device_param->cuda_stream) == -1) return -1;
         }
       }
 
       if (device_param->is_hip == true)
       {
-        if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_pws_idx, device_param->pws_idx, pws_cnt * sizeof (pw_idx_t)) == -1) return -1;
+        if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_pws_idx, device_param->pws_idx, pws_cnt * sizeof (pw_idx_t), device_param->hip_stream) == -1) return -1;
 
         const pw_idx_t *pw_idx = device_param->pws_idx + pws_cnt;
 
@@ -6098,13 +5991,13 @@ int run_copy (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, const
 
         if (off)
         {
-          if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_pws_comp_buf, device_param->pws_comp, off * sizeof (u32)) == -1) return -1;
+          if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_pws_comp_buf, device_param->pws_comp, off * sizeof (u32), device_param->hip_stream) == -1) return -1;
         }
       }
 
       if (device_param->is_opencl == true)
       {
-        if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_idx, CL_TRUE, 0, pws_cnt * sizeof (pw_idx_t), device_param->pws_idx, 0, NULL, NULL) == -1) return -1;
+        if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_idx, CL_FALSE, 0, pws_cnt * sizeof (pw_idx_t), device_param->pws_idx, 0, NULL, NULL) == -1) return -1;
 
         const pw_idx_t *pw_idx = device_param->pws_idx + pws_cnt;
 
@@ -6112,7 +6005,7 @@ int run_copy (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, const
 
         if (off)
         {
-          if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_comp_buf, CL_TRUE, 0, off * sizeof (u32), device_param->pws_comp, 0, NULL, NULL) == -1) return -1;
+          if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_comp_buf, CL_FALSE, 0, off * sizeof (u32), device_param->pws_comp, 0, NULL, NULL) == -1) return -1;
         }
       }
 
@@ -6158,7 +6051,7 @@ int run_copy (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, const
 
         if (device_param->is_cuda == true)
         {
-          if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_pws_idx, device_param->pws_idx, pws_cnt * sizeof (pw_idx_t)) == -1) return -1;
+          if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_pws_idx, device_param->pws_idx, pws_cnt * sizeof (pw_idx_t), device_param->cuda_stream) == -1) return -1;
 
           const pw_idx_t *pw_idx = device_param->pws_idx + pws_cnt;
 
@@ -6166,13 +6059,13 @@ int run_copy (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, const
 
           if (off)
           {
-            if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_pws_comp_buf, device_param->pws_comp, off * sizeof (u32)) == -1) return -1;
+            if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_pws_comp_buf, device_param->pws_comp, off * sizeof (u32), device_param->cuda_stream) == -1) return -1;
           }
         }
 
         if (device_param->is_hip == true)
         {
-          if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_pws_idx, device_param->pws_idx, pws_cnt * sizeof (pw_idx_t)) == -1) return -1;
+          if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_pws_idx, device_param->pws_idx, pws_cnt * sizeof (pw_idx_t), device_param->hip_stream) == -1) return -1;
 
           const pw_idx_t *pw_idx = device_param->pws_idx + pws_cnt;
 
@@ -6180,13 +6073,13 @@ int run_copy (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, const
 
           if (off)
           {
-            if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_pws_comp_buf, device_param->pws_comp, off * sizeof (u32)) == -1) return -1;
+            if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_pws_comp_buf, device_param->pws_comp, off * sizeof (u32), device_param->hip_stream) == -1) return -1;
           }
         }
 
         if (device_param->is_opencl == true)
         {
-          if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_idx, CL_TRUE, 0, pws_cnt * sizeof (pw_idx_t), device_param->pws_idx, 0, NULL, NULL) == -1) return -1;
+          if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_idx, CL_FALSE, 0, pws_cnt * sizeof (pw_idx_t), device_param->pws_idx, 0, NULL, NULL) == -1) return -1;
 
           const pw_idx_t *pw_idx = device_param->pws_idx + pws_cnt;
 
@@ -6194,7 +6087,7 @@ int run_copy (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, const
 
           if (off)
           {
-            if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_comp_buf, CL_TRUE, 0, off * sizeof (u32), device_param->pws_comp, 0, NULL, NULL) == -1) return -1;
+            if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_comp_buf, CL_FALSE, 0, off * sizeof (u32), device_param->pws_comp, 0, NULL, NULL) == -1) return -1;
           }
         }
 
@@ -6206,7 +6099,7 @@ int run_copy (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, const
         {
           if (device_param->is_cuda == true)
           {
-            if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_pws_idx, device_param->pws_idx, pws_cnt * sizeof (pw_idx_t)) == -1) return -1;
+            if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_pws_idx, device_param->pws_idx, pws_cnt * sizeof (pw_idx_t), device_param->cuda_stream) == -1) return -1;
 
             const pw_idx_t *pw_idx = device_param->pws_idx + pws_cnt;
 
@@ -6214,13 +6107,13 @@ int run_copy (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, const
 
             if (off)
             {
-              if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_pws_comp_buf, device_param->pws_comp, off * sizeof (u32)) == -1) return -1;
+              if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_pws_comp_buf, device_param->pws_comp, off * sizeof (u32), device_param->cuda_stream) == -1) return -1;
             }
           }
 
           if (device_param->is_hip == true)
           {
-            if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_pws_idx, device_param->pws_idx, pws_cnt * sizeof (pw_idx_t)) == -1) return -1;
+            if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_pws_idx, device_param->pws_idx, pws_cnt * sizeof (pw_idx_t), device_param->hip_stream) == -1) return -1;
 
             const pw_idx_t *pw_idx = device_param->pws_idx + pws_cnt;
 
@@ -6228,13 +6121,13 @@ int run_copy (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, const
 
             if (off)
             {
-              if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_pws_comp_buf, device_param->pws_comp, off * sizeof (u32)) == -1) return -1;
+              if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_pws_comp_buf, device_param->pws_comp, off * sizeof (u32), device_param->hip_stream) == -1) return -1;
             }
           }
 
           if (device_param->is_opencl == true)
           {
-            if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_idx, CL_TRUE, 0, pws_cnt * sizeof (pw_idx_t), device_param->pws_idx, 0, NULL, NULL) == -1) return -1;
+            if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_idx, CL_FALSE, 0, pws_cnt * sizeof (pw_idx_t), device_param->pws_idx, 0, NULL, NULL) == -1) return -1;
 
             const pw_idx_t *pw_idx = device_param->pws_idx + pws_cnt;
 
@@ -6242,7 +6135,7 @@ int run_copy (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, const
 
             if (off)
             {
-              if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_comp_buf, CL_TRUE, 0, off * sizeof (u32), device_param->pws_comp, 0, NULL, NULL) == -1) return -1;
+              if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_comp_buf, CL_FALSE, 0, off * sizeof (u32), device_param->pws_comp, 0, NULL, NULL) == -1) return -1;
             }
           }
 
@@ -6252,7 +6145,7 @@ int run_copy (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, const
         {
           if (device_param->is_cuda == true)
           {
-            if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_pws_idx, device_param->pws_idx, pws_cnt * sizeof (pw_idx_t)) == -1) return -1;
+            if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_pws_idx, device_param->pws_idx, pws_cnt * sizeof (pw_idx_t), device_param->cuda_stream) == -1) return -1;
 
             const pw_idx_t *pw_idx = device_param->pws_idx + pws_cnt;
 
@@ -6260,13 +6153,13 @@ int run_copy (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, const
 
             if (off)
             {
-              if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_pws_comp_buf, device_param->pws_comp, off * sizeof (u32)) == -1) return -1;
+              if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_pws_comp_buf, device_param->pws_comp, off * sizeof (u32), device_param->cuda_stream) == -1) return -1;
             }
           }
 
           if (device_param->is_hip == true)
           {
-            if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_pws_idx, device_param->pws_idx, pws_cnt * sizeof (pw_idx_t)) == -1) return -1;
+            if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_pws_idx, device_param->pws_idx, pws_cnt * sizeof (pw_idx_t), device_param->hip_stream) == -1) return -1;
 
             const pw_idx_t *pw_idx = device_param->pws_idx + pws_cnt;
 
@@ -6274,13 +6167,13 @@ int run_copy (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, const
 
             if (off)
             {
-              if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_pws_comp_buf, device_param->pws_comp, off * sizeof (u32)) == -1) return -1;
+              if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_pws_comp_buf, device_param->pws_comp, off * sizeof (u32), device_param->hip_stream) == -1) return -1;
             }
           }
 
           if (device_param->is_opencl == true)
           {
-            if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_idx, CL_TRUE, 0, pws_cnt * sizeof (pw_idx_t), device_param->pws_idx, 0, NULL, NULL) == -1) return -1;
+            if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_idx, CL_FALSE, 0, pws_cnt * sizeof (pw_idx_t), device_param->pws_idx, 0, NULL, NULL) == -1) return -1;
 
             const pw_idx_t *pw_idx = device_param->pws_idx + pws_cnt;
 
@@ -6288,7 +6181,7 @@ int run_copy (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, const
 
             if (off)
             {
-              if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_comp_buf, CL_TRUE, 0, off * sizeof (u32), device_param->pws_comp, 0, NULL, NULL) == -1) return -1;
+              if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_comp_buf, CL_FALSE, 0, off * sizeof (u32), device_param->pws_comp, 0, NULL, NULL) == -1) return -1;
             }
           }
 
@@ -6314,6 +6207,11 @@ int run_copy (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, const
     }
   }
 
+  if (device_param->is_opencl == true)
+  {
+    if (hc_clFlush (hashcat_ctx, device_param->opencl_command_queue) == -1) return -1;
+  }
+
   return 0;
 }
 
@@ -6496,12 +6394,12 @@ int run_cracker (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, co
         {
           if (device_param->is_cuda == true)
           {
-            if (hc_cuMemcpyDtoD (hashcat_ctx, device_param->cuda_d_rules_c, device_param->cuda_d_rules + (innerloop_pos * sizeof (kernel_rule_t)), innerloop_left * sizeof (kernel_rule_t)) == -1) return -1;
+            if (hc_cuMemcpyDtoDAsync (hashcat_ctx, device_param->cuda_d_rules_c, device_param->cuda_d_rules + (innerloop_pos * sizeof (kernel_rule_t)), innerloop_left * sizeof (kernel_rule_t), device_param->cuda_stream) == -1) return -1;
           }
 
           if (device_param->is_hip == true)
           {
-            if (hc_hipMemcpyDtoD (hashcat_ctx, device_param->hip_d_rules_c, device_param->hip_d_rules + (innerloop_pos * sizeof (kernel_rule_t)), innerloop_left * sizeof (kernel_rule_t)) == -1) return -1;
+            if (hc_hipMemcpyDtoDAsync (hashcat_ctx, device_param->hip_d_rules_c, device_param->hip_d_rules + (innerloop_pos * sizeof (kernel_rule_t)), innerloop_left * sizeof (kernel_rule_t), device_param->hip_stream) == -1) return -1;
           }
 
           if (device_param->is_opencl == true)
@@ -6621,12 +6519,12 @@ int run_cracker (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, co
 
               if (device_param->is_cuda == true)
               {
-                if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_combs_c, device_param->combs_buf, innerloop_left * sizeof (pw_t)) == -1) return -1;
+                if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_combs_c, device_param->combs_buf, innerloop_left * sizeof (pw_t), device_param->cuda_stream) == -1) return -1;
               }
 
               if (device_param->is_hip == true)
               {
-                if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_combs_c, device_param->combs_buf, innerloop_left * sizeof (pw_t)) == -1) return -1;
+                if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_combs_c, device_param->combs_buf, innerloop_left * sizeof (pw_t), device_param->hip_stream) == -1) return -1;
               }
 
               if (device_param->is_opencl == true)
@@ -6644,12 +6542,12 @@ int run_cracker (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, co
 
               if (device_param->is_cuda == true)
               {
-                if (hc_cuMemcpyDtoD (hashcat_ctx, device_param->cuda_d_combs_c, device_param->cuda_d_combs, innerloop_left * sizeof (pw_t)) == -1) return -1;
+                if (hc_cuMemcpyDtoDAsync (hashcat_ctx, device_param->cuda_d_combs_c, device_param->cuda_d_combs, innerloop_left * sizeof (pw_t), device_param->cuda_stream) == -1) return -1;
               }
 
               if (device_param->is_hip == true)
               {
-                if (hc_hipMemcpyDtoD (hashcat_ctx, device_param->hip_d_combs_c, device_param->hip_d_combs, innerloop_left * sizeof (pw_t)) == -1) return -1;
+                if (hc_hipMemcpyDtoDAsync (hashcat_ctx, device_param->hip_d_combs_c, device_param->hip_d_combs, innerloop_left * sizeof (pw_t), device_param->hip_stream) == -1) return -1;
               }
 
               if (device_param->is_opencl == true)
@@ -6667,12 +6565,12 @@ int run_cracker (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, co
 
               if (device_param->is_cuda == true)
               {
-                if (hc_cuMemcpyDtoD (hashcat_ctx, device_param->cuda_d_combs_c, device_param->cuda_d_combs, innerloop_left * sizeof (pw_t)) == -1) return -1;
+                if (hc_cuMemcpyDtoDAsync (hashcat_ctx, device_param->cuda_d_combs_c, device_param->cuda_d_combs, innerloop_left * sizeof (pw_t), device_param->cuda_stream) == -1) return -1;
               }
 
               if (device_param->is_hip == true)
               {
-                if (hc_hipMemcpyDtoD (hashcat_ctx, device_param->hip_d_combs_c, device_param->hip_d_combs, innerloop_left * sizeof (pw_t)) == -1) return -1;
+                if (hc_hipMemcpyDtoDAsync (hashcat_ctx, device_param->hip_d_combs_c, device_param->hip_d_combs, innerloop_left * sizeof (pw_t), device_param->hip_stream) == -1) return -1;
               }
 
               if (device_param->is_opencl == true)
@@ -6793,17 +6691,17 @@ int run_cracker (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, co
 
               if (device_param->is_cuda == true)
               {
-                if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_combs_c, device_param->combs_buf, innerloop_left * sizeof (pw_t)) == -1) return -1;
+                if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_combs_c, device_param->combs_buf, innerloop_left * sizeof (pw_t), device_param->cuda_stream) == -1) return -1;
               }
 
               if (device_param->is_hip == true)
               {
-                if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_combs_c, device_param->combs_buf, innerloop_left * sizeof (pw_t)) == -1) return -1;
+                if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_combs_c, device_param->combs_buf, innerloop_left * sizeof (pw_t), device_param->hip_stream) == -1) return -1;
               }
 
               if (device_param->is_opencl == true)
               {
-                if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_combs_c, CL_TRUE, 0, innerloop_left * sizeof (pw_t), device_param->combs_buf, 0, NULL, NULL) == -1) return -1;
+                if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_combs_c, CL_FALSE, 0, innerloop_left * sizeof (pw_t), device_param->combs_buf, 0, NULL, NULL) == -1) return -1;
               }
             }
             else if (user_options->attack_mode == ATTACK_MODE_HYBRID1)
@@ -6816,12 +6714,12 @@ int run_cracker (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, co
 
               if (device_param->is_cuda == true)
               {
-                if (hc_cuMemcpyDtoD (hashcat_ctx, device_param->cuda_d_combs_c, device_param->cuda_d_combs, innerloop_left * sizeof (pw_t)) == -1) return -1;
+                if (hc_cuMemcpyDtoDAsync (hashcat_ctx, device_param->cuda_d_combs_c, device_param->cuda_d_combs, innerloop_left * sizeof (pw_t), device_param->cuda_stream) == -1) return -1;
               }
 
               if (device_param->is_hip == true)
               {
-                if (hc_hipMemcpyDtoD (hashcat_ctx, device_param->hip_d_combs_c, device_param->hip_d_combs, innerloop_left * sizeof (pw_t)) == -1) return -1;
+                if (hc_hipMemcpyDtoDAsync (hashcat_ctx, device_param->hip_d_combs_c, device_param->hip_d_combs, innerloop_left * sizeof (pw_t), device_param->hip_stream) == -1) return -1;
               }
 
               if (device_param->is_opencl == true)
@@ -6841,12 +6739,12 @@ int run_cracker (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, co
 
           if (device_param->is_cuda == true)
           {
-            if (hc_cuMemcpyDtoD (hashcat_ctx, device_param->cuda_d_bfs_c, device_param->cuda_d_bfs, innerloop_left * sizeof (bf_t)) == -1) return -1;
+            if (hc_cuMemcpyDtoDAsync (hashcat_ctx, device_param->cuda_d_bfs_c, device_param->cuda_d_bfs, innerloop_left * sizeof (bf_t), device_param->cuda_stream) == -1) return -1;
           }
 
           if (device_param->is_hip == true)
           {
-            if (hc_hipMemcpyDtoD (hashcat_ctx, device_param->hip_d_bfs_c, device_param->hip_d_bfs, innerloop_left * sizeof (bf_t)) == -1) return -1;
+            if (hc_hipMemcpyDtoDAsync (hashcat_ctx, device_param->hip_d_bfs_c, device_param->hip_d_bfs, innerloop_left * sizeof (bf_t), device_param->hip_stream) == -1) return -1;
           }
 
           if (device_param->is_opencl == true)
@@ -7177,52 +7075,82 @@ int backend_ctx_init (hashcat_ctx_t *hashcat_ctx)
 
     if ((rc_hip_init == 0) && (rc_hiprtc_init == 0))
     {
-      // hiprtc version
-
-      int hiprtc_major = 0;
-      int hiprtc_minor = 0;
-
-      if (hc_hiprtcVersion (hashcat_ctx, &hiprtc_major, &hiprtc_minor) == -1) return -1;
-
-      int hiprtc_driver_version = (hiprtc_major * 1000) + (hiprtc_minor * 10);
-
-      backend_ctx->hiprtc_driver_version = hiprtc_driver_version;
-
-      if (hiprtc_driver_version < 9000)
-      {
-        event_log_error (hashcat_ctx, "Outdated AMD HIPRTC driver version '%d' detected!", hiprtc_driver_version);
-
-        event_log_warning (hashcat_ctx, "See hashcat.net for officially supported AMD HIP versions.");
-        event_log_warning (hashcat_ctx, NULL);
-
-        return -1;
-      }
-
       // hip version
 
-      int hip_driver_version = 10000;
+      int hip_driverVersion;
 
-      //if (hc_hipDriverGetVersion (hashcat_ctx, &hip_driver_version) == -1) return -1;
+      if (hc_hipDriverGetVersion (hashcat_ctx, &hip_driverVersion) == -1) return -1;
 
-      backend_ctx->hip_driver_version = hip_driver_version;
+      backend_ctx->hip_driverVersion = hip_driverVersion;
 
-      if (hip_driver_version < 9000)
+      int hip_runtimeVersion;
+
+      if (hc_hipRuntimeGetVersion (hashcat_ctx, &hip_runtimeVersion) == -1) return -1;
+
+      backend_ctx->hip_runtimeVersion = hip_runtimeVersion;
+
+      if (hip_runtimeVersion < 1000)
       {
-        event_log_error (hashcat_ctx, "Outdated AMD HIP driver version '%d' detected!", hip_driver_version);
+        if (hip_runtimeVersion < 404)
+        {
+          event_log_warning (hashcat_ctx, "Unsupported AMD HIP runtime version '%d.%d' detected! Falling back to OpenCL...", hip_runtimeVersion / 100, hip_runtimeVersion % 10);
+          event_log_warning (hashcat_ctx, NULL);
 
-        event_log_warning (hashcat_ctx, "See hashcat.net for officially supported AMD HIP versions.");
+          rc_hip_init    = -1;
+          rc_hiprtc_init = -1;
+
+          backend_ctx->rc_hip_init    = rc_hip_init;
+          backend_ctx->rc_hiprtc_init = rc_hiprtc_init;
+
+          backend_ctx->hip    = NULL;
+          backend_ctx->hiprtc = NULL;
+
+          // if we call this, opencl stops working?! so we just zero the pointer
+          // this causes a memleak and an open filehandle but what can we do?
+          // hip_close    (hashcat_ctx);
+          // hiprtc_close (hashcat_ctx);
+        }
+      }
+      else
+      {
+        // we need to wait for 4.4 to be released to continue here
+        // ignore this backend
+
+        int hip_version_major = (hip_runtimeVersion - 0) / 10000000;
+        int hip_version_minor = (hip_runtimeVersion - (hip_version_major * 10000000)) / 100000;
+        int hip_version_patch = (hip_runtimeVersion - (hip_version_major * 10000000) - (hip_version_minor * 100000));
+
+        event_log_warning (hashcat_ctx, "Unsupported AMD HIP runtime version '%d.%d.%d' detected! Falling back to OpenCL...", hip_version_major, hip_version_minor, hip_version_patch);
         event_log_warning (hashcat_ctx, NULL);
 
-        return -1;
+        rc_hip_init    = -1;
+        rc_hiprtc_init = -1;
+
+        backend_ctx->rc_hip_init    = rc_hip_init;
+        backend_ctx->rc_hiprtc_init = rc_hiprtc_init;
+
+        backend_ctx->hip = NULL;
+
+        // if we call this, opencl stops working?! so we just zero the pointer
+        // this causes a memleak and an open filehandle but what can we do?
+        // hip_close    (hashcat_ctx);
+        // hiprtc_close (hashcat_ctx);
       }
     }
     else
     {
-      rc_hip_init  = -1;
+      rc_hip_init    = -1;
       rc_hiprtc_init = -1;
 
-      hip_close  (hashcat_ctx);
-      hiprtc_close (hashcat_ctx);
+      backend_ctx->rc_hip_init    = rc_hip_init;
+      backend_ctx->rc_hiprtc_init = rc_hiprtc_init;
+
+      backend_ctx->hip = NULL;
+
+      // if we call this, opencl stops working?! so we just zero the pointer
+      // this causes a memleak and an open filehandle but what can we do?
+      // hip_close    (hashcat_ctx);
+      // hiprtc_close (hashcat_ctx);
     }
   }
 
@@ -7258,10 +7186,10 @@ int backend_ctx_init (hashcat_ctx_t *hashcat_ctx)
 
       #if defined (__linux__)
       event_log_warning (hashcat_ctx, "* AMD GPUs on Linux require this driver:");
-      event_log_warning (hashcat_ctx, "  \"RadeonOpenCompute (ROCm)\" Software Platform (3.1 or later)");
+      event_log_warning (hashcat_ctx, "  \"AMD ROCm\" (4.3 or later)");
       #elif defined (_WIN)
       event_log_warning (hashcat_ctx, "* AMD GPUs on Windows require this driver:");
-      event_log_warning (hashcat_ctx, "  \"AMD Radeon Adrenalin 2020 Edition\" (20.2.2 or later)");
+      event_log_warning (hashcat_ctx, "  \"AMD Radeon Adrenalin 2020 Edition\" (21.2.1 or later)");
       #endif
 
       event_log_warning (hashcat_ctx, "* Intel CPUs require this runtime:");
@@ -7583,10 +7511,10 @@ int backend_ctx_init (hashcat_ctx_t *hashcat_ctx)
 
     #if defined (__linux__)
     event_log_warning (hashcat_ctx, "* AMD GPUs on Linux require this driver:");
-    event_log_warning (hashcat_ctx, "  \"RadeonOpenCompute (ROCm)\" Software Platform (3.1 or later)");
+    event_log_warning (hashcat_ctx, "  \"AMD ROCm\" (4.3 or later)");
     #elif defined (_WIN)
     event_log_warning (hashcat_ctx, "* AMD GPUs on Windows require this driver:");
-    event_log_warning (hashcat_ctx, "  \"AMD Radeon Adrenalin 2020 Edition\" (20.2.2 or later)");
+    event_log_warning (hashcat_ctx, "  \"AMD Radeon Adrenalin 2020 Edition\" (21.2.1 or later)");
     #endif
 
     event_log_warning (hashcat_ctx, "* Intel CPUs require this runtime:");
@@ -7841,11 +7769,23 @@ int backend_ctx_devices_init (hashcat_ctx_t *hashcat_ctx, const int comptime)
 
       device_param->kernel_exec_timeout = kernel_exec_timeout;
 
+      // warp size
+
+      int warp_size = 0;
+
+      if (hc_cuDeviceGetAttribute (hashcat_ctx, &warp_size, CU_DEVICE_ATTRIBUTE_WARP_SIZE, cuda_device) == -1)
+      {
+        device_param->skipped = true;
+        continue;
+      }
+
+      device_param->kernel_preferred_wgs_multiple = warp_size;
+
       // max_shared_memory_per_block
 
       int max_shared_memory_per_block = 0;
 
-      if (hc_cuDeviceGetAttribute (hashcat_ctx, &max_shared_memory_per_block, CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK, cuda_device) == -1)
+      if (hc_cuDeviceGetAttribute (hashcat_ctx, &max_shared_memory_per_block, CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK_OPTIN, cuda_device) == -1)
       {
         device_param->skipped = true;
         continue;
@@ -8047,7 +7987,7 @@ int backend_ctx_devices_init (hashcat_ctx_t *hashcat_ctx, const int comptime)
 
       backend_ctx->backend_device_from_hip[hip_devices_idx] = backend_devices_idx;
 
-      HIPdevice hip_device;
+      hipDevice_t hip_device;
 
       if (hc_hipDeviceGet (hashcat_ctx, &hip_device, hip_devices_idx) == -1)
       {
@@ -8086,7 +8026,7 @@ int backend_ctx_devices_init (hashcat_ctx_t *hashcat_ctx, const int comptime)
 
       int device_processors = 0;
 
-      if (hc_hipDeviceGetAttribute (hashcat_ctx, &device_processors, HIP_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT, hip_device) == -1)
+      if (hc_hipDeviceGetAttribute (hashcat_ctx, &device_processors, hipDeviceAttributeMultiprocessorCount, hip_device) == -1)
       {
         device_param->skipped = true;
         continue;
@@ -8114,7 +8054,7 @@ int backend_ctx_devices_init (hashcat_ctx_t *hashcat_ctx, const int comptime)
 
       int hip_warp_size = 0;
 
-      if (hc_hipDeviceGetAttribute (hashcat_ctx, &hip_warp_size, HIP_DEVICE_ATTRIBUTE_WARP_SIZE, hip_device) == -1)
+      if (hc_hipDeviceGetAttribute (hashcat_ctx, &hip_warp_size, hipDeviceAttributeWarpSize, hip_device) == -1)
       {
         device_param->skipped = true;
         continue;
@@ -8127,13 +8067,13 @@ int backend_ctx_devices_init (hashcat_ctx_t *hashcat_ctx, const int comptime)
       int sm_major = 0;
       int sm_minor = 0;
 
-      if (hc_hipDeviceGetAttribute (hashcat_ctx, &sm_major, HIP_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, hip_device) == -1)
+      if (hc_hipDeviceGetAttribute (hashcat_ctx, &sm_major, hipDeviceAttributeComputeCapabilityMajor, hip_device) == -1)
       {
         device_param->skipped = true;
         continue;
       }
 
-      if (hc_hipDeviceGetAttribute (hashcat_ctx, &sm_minor, HIP_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, hip_device) == -1)
+      if (hc_hipDeviceGetAttribute (hashcat_ctx, &sm_minor, hipDeviceAttributeComputeCapabilityMinor, hip_device) == -1)
       {
         device_param->skipped = true;
         continue;
@@ -8146,7 +8086,7 @@ int backend_ctx_devices_init (hashcat_ctx_t *hashcat_ctx, const int comptime)
 
       int device_maxworkgroup_size = 0;
 
-      if (hc_hipDeviceGetAttribute (hashcat_ctx, &device_maxworkgroup_size, HIP_DEVICE_ATTRIBUTE_MAX_THREADS_PER_BLOCK, hip_device) == -1)
+      if (hc_hipDeviceGetAttribute (hashcat_ctx, &device_maxworkgroup_size, hipDeviceAttributeMaxThreadsPerBlock, hip_device) == -1)
       {
         device_param->skipped = true;
         continue;
@@ -8158,7 +8098,7 @@ int backend_ctx_devices_init (hashcat_ctx_t *hashcat_ctx, const int comptime)
 
       int device_maxclock_frequency = 0;
 
-      if (hc_hipDeviceGetAttribute (hashcat_ctx, &device_maxclock_frequency, HIP_DEVICE_ATTRIBUTE_CLOCK_RATE, hip_device) == -1)
+      if (hc_hipDeviceGetAttribute (hashcat_ctx, &device_maxclock_frequency, hipDeviceAttributeClockRate, hip_device) == -1)
       {
         device_param->skipped = true;
         continue;
@@ -8172,19 +8112,20 @@ int backend_ctx_devices_init (hashcat_ctx_t *hashcat_ctx, const int comptime)
       int pci_bus_id_nv     = 0;
       int pci_slot_id_nv    = 0;
 
-      if (hc_hipDeviceGetAttribute (hashcat_ctx, &pci_domain_id_nv, HIP_DEVICE_ATTRIBUTE_PCI_DOMAIN_ID, hip_device) == -1)
+      // Not supported by HIP
+      //if (hc_hipDeviceGetAttribute (hashcat_ctx, &pci_domain_id_nv, hipDeviceAttributePciDomainID, hip_device) == -1)
+      //{
+      //  device_param->skipped = true;
+      //  continue;
+      //}
+
+      if (hc_hipDeviceGetAttribute (hashcat_ctx, &pci_bus_id_nv, hipDeviceAttributePciBusId, hip_device) == -1)
       {
         device_param->skipped = true;
         continue;
       }
 
-      if (hc_hipDeviceGetAttribute (hashcat_ctx, &pci_bus_id_nv, HIP_DEVICE_ATTRIBUTE_PCI_BUS_ID, hip_device) == -1)
-      {
-        device_param->skipped = true;
-        continue;
-      }
-
-      if (hc_hipDeviceGetAttribute (hashcat_ctx, &pci_slot_id_nv, HIP_DEVICE_ATTRIBUTE_PCI_DEVICE_ID, hip_device) == -1)
+      if (hc_hipDeviceGetAttribute (hashcat_ctx, &pci_slot_id_nv, hipDeviceAttributePciDeviceId, hip_device) == -1)
       {
         device_param->skipped = true;
         continue;
@@ -8192,6 +8133,7 @@ int backend_ctx_devices_init (hashcat_ctx_t *hashcat_ctx, const int comptime)
 
       device_param->pcie_domain   = (u8) (pci_domain_id_nv);
       device_param->pcie_bus      = (u8) (pci_bus_id_nv);
+
       device_param->pcie_device   = (u8) (pci_slot_id_nv >> 3);
       device_param->pcie_function = (u8) (pci_slot_id_nv & 7);
 
@@ -8199,7 +8141,7 @@ int backend_ctx_devices_init (hashcat_ctx_t *hashcat_ctx, const int comptime)
 
       int kernel_exec_timeout = 0;
 
-      if (hc_hipDeviceGetAttribute (hashcat_ctx, &kernel_exec_timeout, HIP_DEVICE_ATTRIBUTE_KERNEL_EXEC_TIMEOUT, hip_device) == -1)
+      if (hc_hipDeviceGetAttribute (hashcat_ctx, &kernel_exec_timeout, hipDeviceAttributeKernelExecTimeout, hip_device) == -1)
       {
         device_param->skipped = true;
         continue;
@@ -8207,11 +8149,23 @@ int backend_ctx_devices_init (hashcat_ctx_t *hashcat_ctx, const int comptime)
 
       device_param->kernel_exec_timeout = kernel_exec_timeout;
 
+      // warp size
+
+      int warp_size = 0;
+
+      if (hc_hipDeviceGetAttribute (hashcat_ctx, &warp_size, hipDeviceAttributeWarpSize, hip_device) == -1)
+      {
+        device_param->skipped = true;
+        continue;
+      }
+
+      device_param->kernel_preferred_wgs_multiple = warp_size;
+
       // max_shared_memory_per_block
 
       int max_shared_memory_per_block = 0;
 
-      if (hc_hipDeviceGetAttribute (hashcat_ctx, &max_shared_memory_per_block, HIP_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK, hip_device) == -1)
+      if (hc_hipDeviceGetAttribute (hashcat_ctx, &max_shared_memory_per_block, hipDeviceAttributeMaxSharedMemoryPerBlock, hip_device) == -1)
       {
         device_param->skipped = true;
         continue;
@@ -8230,7 +8184,7 @@ int backend_ctx_devices_init (hashcat_ctx_t *hashcat_ctx, const int comptime)
 
       int device_max_constant_buffer_size = 0;
 
-      if (hc_hipDeviceGetAttribute (hashcat_ctx, &device_max_constant_buffer_size, HIP_DEVICE_ATTRIBUTE_TOTAL_CONSTANT_MEMORY, hip_device) == -1)
+      if (hc_hipDeviceGetAttribute (hashcat_ctx, &device_max_constant_buffer_size, hipDeviceAttributeTotalConstantMemory, hip_device) == -1)
       {
         device_param->skipped = true;
         continue;
@@ -8300,14 +8254,14 @@ int backend_ctx_devices_init (hashcat_ctx_t *hashcat_ctx, const int comptime)
       {
         if ((user_options->force == false) && (user_options->backend_info == false))
         {
-          // HIPDA does not support query nvidia driver version, therefore no driver checks here
+          // CUDA does not support query nvidia driver version, therefore no driver checks here
           // IF needed, could be retrieved using nvmlSystemGetDriverVersion()
 
           if (device_param->sm_major < 5)
           {
-            if (user_options->quiet == false) event_log_warning (hashcat_ctx, "* Device #%u: This hardware has outdated HIPDA compute capability (%u.%u).", device_id + 1, device_param->sm_major, device_param->sm_minor);
+            if (user_options->quiet == false) event_log_warning (hashcat_ctx, "* Device #%u: This hardware has outdated CUDA compute capability (%u.%u).", device_id + 1, device_param->sm_major, device_param->sm_minor);
             if (user_options->quiet == false) event_log_warning (hashcat_ctx, "             For modern OpenCL performance, upgrade to hardware that supports");
-            if (user_options->quiet == false) event_log_warning (hashcat_ctx, "             HIPDA compute capability version 5.0 (Maxwell) or higher.");
+            if (user_options->quiet == false) event_log_warning (hashcat_ctx, "             CUDA compute capability version 5.0 (Maxwell) or higher.");
           }
 
           if (device_param->kernel_exec_timeout != 0)
@@ -8323,13 +8277,6 @@ int backend_ctx_devices_init (hashcat_ctx_t *hashcat_ctx, const int comptime)
 
       // instruction set
 
-      // bcrypt optimization?
-      //const int rc_cuCtxSetCacheConfig = hc_hipCtxSetCacheConfig (hashcat_ctx, HIP_FUNC_CACHE_PREFER_SHARED);
-      //
-      //if (rc_cuCtxSetCacheConfig == -1) return -1;
-
-      // const int sm = (device_param->sm_major * 10) + device_param->sm_minor;
-
       device_param->has_add   = false;
       device_param->has_addc  = false;
       device_param->has_sub   = false;
@@ -8341,9 +8288,9 @@ int backend_ctx_devices_init (hashcat_ctx_t *hashcat_ctx, const int comptime)
 
       // device_available_mem
 
-      HIPcontext hip_context;
+      hipCtx_t hip_context;
 
-      if (hc_hipCtxCreate (hashcat_ctx, &hip_context, HIP_CTX_SCHED_BLOCKING_SYNC, device_param->hip_device) == -1)
+      if (hc_hipCtxCreate (hashcat_ctx, &hip_context, hipDeviceScheduleBlockingSync, device_param->hip_device) == -1)
       {
         device_param->skipped = true;
         continue;
@@ -8590,6 +8537,10 @@ int backend_ctx_devices_init (hashcat_ctx_t *hashcat_ctx, const int comptime)
         {
           opencl_device_vendor_id = VENDOR_ID_INTEL_SDK;
         }
+        else if (strcmp (opencl_device_vendor, CL_VENDOR_APPLE_USE_INTEL2) == 0)
+        {
+          opencl_device_vendor_id = VENDOR_ID_INTEL_SDK;
+        }
         else if (strcmp (opencl_device_vendor, CL_VENDOR_INTEL_BEIGNET) == 0)
         {
           opencl_device_vendor_id = VENDOR_ID_INTEL_BEIGNET;
@@ -8824,6 +8775,13 @@ int backend_ctx_devices_init (hashcat_ctx_t *hashcat_ctx, const int comptime)
 
         hcfree (device_extensions);
 
+        // kernel_preferred_wgs_multiple
+
+        // There is global query for this attribute on OpenCL that is not linked to a specific kernel, so we set it to a fixed value
+        // Later in the code, we add vendor specific extensions to query it
+
+        device_param->kernel_preferred_wgs_multiple = 8;
+
         // device_local_mem_type
 
         cl_device_local_mem_type device_local_mem_type;
@@ -9099,10 +9057,43 @@ int backend_ctx_devices_init (hashcat_ctx_t *hashcat_ctx, const int comptime)
           }
         }
 
+        if (device_param->opencl_device_type & CL_DEVICE_TYPE_CPU)
+        {
+          // they like this
+
+          device_param->kernel_preferred_wgs_multiple = 1;
+        }
+
         if (device_param->opencl_device_type & CL_DEVICE_TYPE_GPU)
         {
+          if ((device_param->opencl_platform_vendor_id == VENDOR_ID_APPLE) && (device_param->opencl_device_vendor_id == VENDOR_ID_AMD))
+          {
+            // from https://www.khronos.org/registry/OpenCL/extensions/amd/cl_amd_device_attribute_query.txt
+            #define CL_DEVICE_WAVEFRONT_WIDTH_AMD                   0x4043
+
+            // crazy, but apple does not support this query!
+            // the best alternative is "Preferred work group size multiple (kernel)", but requires to specify a kernel.
+            // so we will set kernel_preferred_wgs_multiple intentionally to 0 because otherwise it it set to 8 by default.
+            // we then assign the value kernel_preferred_wgs_multiple a small kernel like bzero after test if this was set to 0.
+
+            device_param->kernel_preferred_wgs_multiple = 0;
+          }
+
           if ((device_param->opencl_platform_vendor_id == VENDOR_ID_AMD) && (device_param->opencl_device_vendor_id == VENDOR_ID_AMD))
           {
+            cl_uint device_wavefront_width_amd;
+
+            // from https://www.khronos.org/registry/OpenCL/extensions/amd/cl_amd_device_attribute_query.txt
+            #define CL_DEVICE_WAVEFRONT_WIDTH_AMD                   0x4043
+
+            if (hc_clGetDeviceInfo (hashcat_ctx, device_param->opencl_device, CL_DEVICE_WAVEFRONT_WIDTH_AMD, sizeof (device_wavefront_width_amd), &device_wavefront_width_amd, NULL) == -1)
+            {
+              device_param->skipped = true;
+              continue;
+            }
+
+            device_param->kernel_preferred_wgs_multiple = device_wavefront_width_amd;
+
             cl_device_topology_amd amdtopo;
 
             if (hc_clGetDeviceInfo (hashcat_ctx, device_param->opencl_device, CL_DEVICE_TOPOLOGY_AMD, sizeof (amdtopo), &amdtopo, NULL) == -1)
@@ -9119,6 +9110,19 @@ int backend_ctx_devices_init (hashcat_ctx_t *hashcat_ctx, const int comptime)
 
           if ((device_param->opencl_platform_vendor_id == VENDOR_ID_NV) && (device_param->opencl_device_vendor_id == VENDOR_ID_NV))
           {
+            cl_uint device_warp_size_nv;
+
+            // from deps/OpenCL-Headers/CL/cl_ext.h
+            #define CL_DEVICE_WARP_SIZE_NV                      0x4003
+
+            if (hc_clGetDeviceInfo (hashcat_ctx, device_param->opencl_device, CL_DEVICE_WARP_SIZE_NV, sizeof (device_warp_size_nv), &device_warp_size_nv, NULL) == -1)
+            {
+              device_param->skipped = true;
+              continue;
+            }
+
+            device_param->kernel_preferred_wgs_multiple = device_warp_size_nv;
+
             cl_uint pci_bus_id_nv;  // is cl_uint the right type for them??
             cl_uint pci_slot_id_nv;
 
@@ -9993,41 +9997,7 @@ static int get_cuda_kernel_local_mem_size (hashcat_ctx_t *hashcat_ctx, CUfunctio
   return 0;
 }
 
-static int get_cuda_kernel_dynamic_local_mem_size (hashcat_ctx_t *hashcat_ctx, CUfunction function, u64 *result)
-{
-  // AFAIK there's no way to query the maximum value for dynamic shared memory available (because it depends on kernel code).
-  // let's brute force it, therefore workaround the hashcat wrapper of cuFuncSetAttribute()
-
-  #define MAX_ASSUMED_SHARED (1024 * 1024)
-
-  u64 dynamic_shared_size_bytes = 0;
-
-  for (int i = 1; i <= MAX_ASSUMED_SHARED; i++)
-  {
-    backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
-
-    CUDA_PTR *cuda = (CUDA_PTR *) backend_ctx->cuda;
-
-    const CUresult CU_err = cuda->cuFuncSetAttribute (function, CU_FUNC_ATTRIBUTE_MAX_DYNAMIC_SHARED_SIZE_BYTES, i);
-
-    if (CU_err == CUDA_SUCCESS)
-    {
-      dynamic_shared_size_bytes = i;
-
-      continue;
-    }
-
-    break;
-  }
-
-  *result = dynamic_shared_size_bytes;
-
-  if (hc_cuFuncSetAttribute (hashcat_ctx, function, CU_FUNC_ATTRIBUTE_MAX_DYNAMIC_SHARED_SIZE_BYTES, 0) == -1) return -1;
-
-  return 0;
-}
-
-static int get_hip_kernel_wgs (hashcat_ctx_t *hashcat_ctx, HIPfunction function, u32 *result)
+static int get_hip_kernel_wgs (hashcat_ctx_t *hashcat_ctx, hipFunction_t function, u32 *result)
 {
   int max_threads_per_block;
 
@@ -10038,7 +10008,7 @@ static int get_hip_kernel_wgs (hashcat_ctx_t *hashcat_ctx, HIPfunction function,
   return 0;
 }
 
-static int get_hip_kernel_local_mem_size (hashcat_ctx_t *hashcat_ctx, HIPfunction function, u64 *result)
+static int get_hip_kernel_local_mem_size (hashcat_ctx_t *hashcat_ctx, hipFunction_t function, u64 *result)
 {
   int shared_size_bytes;
 
@@ -10049,44 +10019,6 @@ static int get_hip_kernel_local_mem_size (hashcat_ctx_t *hashcat_ctx, HIPfunctio
   return 0;
 }
 
-/*
-not supported because there's no cuFuncSetAttribute equivalent
-
-static int get_hip_kernel_dynamic_local_mem_size (hashcat_ctx_t *hashcat_ctx, HIPfunction function, u64 *result)
-{
-  // AFAIK there's no way to query the maximum value for dynamic shared memory available (because it depends on kernel code).
-  // let's brute force it, therefore workaround the hashcat wrapper of cuFuncSetAttribute()
-
-  #define MAX_ASSUMED_SHARED (1024 * 1024)
-
-  u64 dynamic_shared_size_bytes = 0;
-
-  for (int i = 1; i <= MAX_ASSUMED_SHARED; i++)
-  {
-    backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
-
-    HIP_PTR *hip = (HIP_PTR *) backend_ctx->hip;
-
-    const HIPresult HIP_err = hip->hipFuncSetAttribute (function, HIP_FUNC_ATTRIBUTE_MAX_DYNAMIC_SHARED_SIZE_BYTES, i);
-
-    if (HIP_err == HIP_SUCCESS)
-    {
-      dynamic_shared_size_bytes = i;
-
-      continue;
-    }
-
-    break;
-  }
-
-  *result = dynamic_shared_size_bytes;
-
-  if (hc_hipFuncSetAttribute (hashcat_ctx, function, HIP_FUNC_ATTRIBUTE_MAX_DYNAMIC_SHARED_SIZE_BYTES, 0) == -1) return -1;
-
-  return 0;
-}
-*/
-
 static int get_opencl_kernel_wgs (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, cl_kernel kernel, u32 *result)
 {
   size_t work_group_size = 0;
@@ -10150,70 +10082,11 @@ static int get_opencl_kernel_dynamic_local_mem_size (hashcat_ctx_t *hashcat_ctx,
   return 0;
 }
 
-static u32 get_kernel_threads (const hc_device_param_t *device_param)
-{
-  // this is an upper limit, a good start, since our strategy is to reduce thread counts only.
-
-  u32 kernel_threads_min = device_param->kernel_threads_min;
-  u32 kernel_threads_max = device_param->kernel_threads_max;
-
-  // the changes we do here are just optimizations, since the module always has priority.
-
-  const u32 device_maxworkgroup_size = (const u32) device_param->device_maxworkgroup_size;
-
-  kernel_threads_max = MIN (kernel_threads_max, device_maxworkgroup_size);
-
-  if (device_param->opencl_device_type & CL_DEVICE_TYPE_CPU)
-  {
-    // for all CPU we just do 1 ...
-
-    const u32 cpu_prefered_thread_count = 1;
-
-    kernel_threads_max = MIN (kernel_threads_max, cpu_prefered_thread_count);
-  }
-  else if (device_param->opencl_device_type & CL_DEVICE_TYPE_GPU)
-  {
-    // for GPU we need to distinguish by vendor
-
-    if (device_param->opencl_device_vendor_id == VENDOR_ID_INTEL_SDK)
-    {
-      const u32 gpu_prefered_thread_count = 8;
-
-      kernel_threads_max = MIN (kernel_threads_max, gpu_prefered_thread_count);
-    }
-    else if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD)
-    {
-      u32 gpu_prefered_thread_count = 64;
-
-      if (device_param->opencl_platform_vendor_id == VENDOR_ID_APPLE)
-      {
-        // based on clinfo output: Preferred work group size multiple (kernel)
-
-        gpu_prefered_thread_count = 32;
-      }
-
-      kernel_threads_max = MIN (kernel_threads_max, gpu_prefered_thread_count);
-    }
-    else if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD_USE_HIP)
-    {
-      u32 gpu_prefered_thread_count = 64;
-
-      kernel_threads_max = MIN (kernel_threads_max, gpu_prefered_thread_count);
-    }
-  }
-
-  // this is intenionally! at this point, kernel_threads_min can be higher than kernel_threads_max.
-  // in this case we actually want kernel_threads_min selected.
-
-  const u32 kernel_threads = MAX (kernel_threads_min, kernel_threads_max);
-
-  return kernel_threads;
-}
-
-static bool load_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, const char *kernel_name, char *source_file, char *cached_file, const char *build_options_buf, const bool cache_disable, cl_program *opencl_program, CUmodule *cuda_module, HIPmodule *hip_module)
+static bool load_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, const char *kernel_name, char *source_file, char *cached_file, const char *build_options_buf, const bool cache_disable, cl_program *opencl_program, CUmodule *cuda_module, hipModule_t *hip_module)
 {
   const hashconfig_t    *hashconfig    = hashcat_ctx->hashconfig;
   const folder_config_t *folder_config = hashcat_ctx->folder_config;
+  const user_options_t  *user_options  = hashcat_ctx->user_options;
 
   bool cached = true;
 
@@ -10247,8 +10120,6 @@ static bool load_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_p
   if (cached == false)
   {
     #if defined (DEBUG)
-    const user_options_t *user_options = hashcat_ctx->user_options;
-
     if (user_options->quiet == false) event_log_warning (hashcat_ctx, "* Device #%u: Kernel %s not found in cache. Please be patient...", device_param->device_id + 1, filename_from_filepath (cached_file));
     #endif
 
@@ -10495,25 +10366,32 @@ static bool load_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_p
 
       if (hc_hiprtcCreateProgram (hashcat_ctx, &program, kernel_sources[0], kernel_name, 0, NULL, NULL) == -1) return false;
 
-      char **hiprtc_options = (char **) hccalloc (6 + strlen (build_options_buf) + 1, sizeof (char *)); // ...
+      char **hiprtc_options = (char **) hccalloc (7 + strlen (build_options_buf) + 1, sizeof (char *)); // ...
 
       //hiprtc_options[0] = "--restrict";
       //hiprtc_options[1] = "--device-as-default-execution-space";
       //hiprtc_options[2] = "--gpu-architecture";
 
-      //hc_asprintf (&hiprtc_options[3], "compute_%d%d", device_param->sm_major, device_param->sm_minor);
+      hc_asprintf (&hiprtc_options[0], "--gpu-max-threads-per-block=%d", (user_options->kernel_threads_chgd == true) ? user_options->kernel_threads : device_param->kernel_threads_max);
 
-      hiprtc_options[0] = "--gpu-max-threads-per-block=64";
-      hiprtc_options[1] = "";
-      hiprtc_options[2] = "";
+      /* 4.3 linux
+      hiprtc_options[1] = "-I";
+      hiprtc_options[2] = "/opt/rocm/hip/bin/include";
+      hiprtc_options[3] = "-I";
+      hiprtc_options[4] = "/opt/rocm/include";
+      hiprtc_options[5] = "-I";
+      */
+
+      hiprtc_options[1] = "-nocudainc";
+      hiprtc_options[2] = "-nocudalib";
       hiprtc_options[3] = "";
-
-      hiprtc_options[4] = "-I";
-      hiprtc_options[5] = folder_config->cpath_real;
+      hiprtc_options[4] = "";
+      hiprtc_options[5] = "-I";
+      hiprtc_options[6] = folder_config->cpath_real;
 
       char *hiprtc_options_string = hcstrdup (build_options_buf);
 
-      const int num_options = 6 + hiprtc_make_options_array_from_string (hiprtc_options_string, hiprtc_options + 6);
+      const int num_options = 7 + hiprtc_make_options_array_from_string (hiprtc_options_string, hiprtc_options + 7);
 
       const int rc_hiprtcCompileProgram = hc_hiprtcCompileProgram (hashcat_ctx, program, num_options, (const char * const *) hiprtc_options);
 
@@ -10570,137 +10448,27 @@ static bool load_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_p
 
       int mod_cnt = 6;
 
-      HIPjit_option mod_opts[7];
-      void *mod_vals[7];
+      hipJitOption mod_opts[6];
+      void *mod_vals[6];
 
-      mod_opts[0] = HIP_JIT_TARGET_FROM_HIPCONTEXT;
+      mod_opts[0] = hipJitOptionTargetFromContext;
       mod_vals[0] = (void *) 0;
 
-      mod_opts[1] = HIP_JIT_LOG_VERBOSE;
+      mod_opts[1] = hipJitOptionLogVerbose;
       mod_vals[1] = (void *) 1;
 
-      mod_opts[2] = HIP_JIT_INFO_LOG_BUFFER;
+      mod_opts[2] = hipJitOptionInfoLogBuffer;
       mod_vals[2] = (void *) mod_info_log;
 
-      mod_opts[3] = HIP_JIT_INFO_LOG_BUFFER_SIZE_BYTES;
+      mod_opts[3] = hipJitOptionInfoLogBufferSizeBytes;
       mod_vals[3] = (void *) LOG_SIZE;
 
-      mod_opts[4] = HIP_JIT_ERROR_LOG_BUFFER;
+      mod_opts[4] = hipJitOptionErrorLogBuffer;
       mod_vals[4] = (void *) mod_error_log;
 
-      mod_opts[5] = HIP_JIT_ERROR_LOG_BUFFER_SIZE_BYTES;
+      mod_opts[5] = hipJitOptionErrorLogBufferSizeBytes;
       mod_vals[5] = (void *) LOG_SIZE;
 
-      if (hashconfig->opti_type & OPTI_TYPE_REGISTER_LIMIT)
-      {
-        mod_opts[6] = HIP_JIT_MAX_REGISTERS;
-        mod_vals[6] = (void *) 128;
-
-        mod_cnt++;
-      }
-
-      #if defined (WITH_HIPBIN)
-
-      char *jit_info_log  = (char *) hcmalloc (LOG_SIZE + 1);
-      char *jit_error_log = (char *) hcmalloc (LOG_SIZE + 1);
-
-      int jit_cnt = 6;
-
-      HIPjit_option jit_opts[7];
-      void *jit_vals[7];
-
-      jit_opts[0] = HIP_JIT_TARGET_FROM_HIPCONTEXT;
-      jit_vals[0] = (void *) 0;
-
-      jit_opts[1] = HIP_JIT_LOG_VERBOSE;
-      jit_vals[1] = (void *) 1;
-
-      jit_opts[2] = HIP_JIT_INFO_LOG_BUFFER;
-      jit_vals[2] = (void *) jit_info_log;
-
-      jit_opts[3] = HIP_JIT_INFO_LOG_BUFFER_SIZE_BYTES;
-      jit_vals[3] = (void *) LOG_SIZE;
-
-      jit_opts[4] = HIP_JIT_ERROR_LOG_BUFFER;
-      jit_vals[4] = (void *) jit_error_log;
-
-      jit_opts[5] = HIP_JIT_ERROR_LOG_BUFFER_SIZE_BYTES;
-      jit_vals[5] = (void *) LOG_SIZE;
-
-      if (hashconfig->opti_type & OPTI_TYPE_REGISTER_LIMIT)
-      {
-        jit_opts[6] = HIP_JIT_MAX_REGISTERS;
-        jit_vals[6] = (void *) 128;
-
-        jit_cnt++;
-      }
-
-      HIPlinkState state;
-
-      if (hc_cuLinkCreate (hashcat_ctx, jit_cnt, jit_opts, jit_vals, &state) == -1)
-      {
-        event_log_error (hashcat_ctx, "* Device #%u: Kernel %s link failed. Error Log:", device_param->device_id + 1, source_file);
-        event_log_error (hashcat_ctx, "%s", jit_error_log);
-        event_log_error (hashcat_ctx, NULL);
-
-        return false;
-      }
-
-      if (hc_cuLinkAddData (hashcat_ctx, state, HIP_JIT_INPUT_CODE, binary, binary_size, kernel_name, 0, NULL, NULL) == -1)
-      {
-        event_log_error (hashcat_ctx, "* Device #%u: Kernel %s link failed. Error Log:", device_param->device_id + 1, source_file);
-        event_log_error (hashcat_ctx, "%s", jit_error_log);
-        event_log_error (hashcat_ctx, NULL);
-
-        return false;
-      }
-
-      void *cubin = NULL;
-
-      size_t cubin_size = 0;
-
-      if (hc_cuLinkComplete (hashcat_ctx, state, &cubin, &cubin_size) == -1)
-      {
-        event_log_error (hashcat_ctx, "* Device #%u: Kernel %s link failed. Error Log:", device_param->device_id + 1, source_file);
-        event_log_error (hashcat_ctx, "%s", jit_error_log);
-        event_log_error (hashcat_ctx, NULL);
-
-        return false;
-      }
-
-      #if defined (DEBUG)
-      event_log_info (hashcat_ctx, "* Device #%u: Kernel %s link successful. Info Log:", device_param->device_id + 1, source_file);
-      event_log_info (hashcat_ctx, "%s", jit_info_log);
-      event_log_info (hashcat_ctx, NULL);
-      #endif
-
-      if (hc_cuModuleLoadDataEx (hashcat_ctx, hip_module, cubin, mod_cnt, mod_opts, mod_vals) == -1)
-      {
-        event_log_error (hashcat_ctx, "* Device #%u: Kernel %s load failed. Error Log:", device_param->device_id + 1, source_file);
-        event_log_error (hashcat_ctx, "%s", mod_error_log);
-        event_log_error (hashcat_ctx, NULL);
-
-        return false;
-      }
-
-      #if defined (DEBUG)
-      event_log_info (hashcat_ctx, "* Device #%u: Kernel %s load successful. Info Log:", device_param->device_id + 1, source_file);
-      event_log_info (hashcat_ctx, "%s", mod_info_log);
-      event_log_info (hashcat_ctx, NULL);
-      #endif
-
-      if (cache_disable == false)
-      {
-        if (write_kernel_binary (hashcat_ctx, cached_file, cubin, cubin_size) == false) return false;
-      }
-
-      if (hc_hipLinkDestroy (hashcat_ctx, state) == -1) return false;
-
-      hcfree (jit_info_log);
-      hcfree (jit_error_log);
-
-      #else
-
       if (hc_hipModuleLoadDataEx (hashcat_ctx, hip_module, binary, mod_cnt, mod_opts, mod_vals) == -1)
       {
         event_log_error (hashcat_ctx, "* Device #%u: Kernel %s load failed. Error Log:", device_param->device_id + 1, source_file);
@@ -10721,8 +10489,6 @@ static bool load_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_p
         if (write_kernel_binary (hashcat_ctx, cached_file, binary, binary_size) == false) return false;
       }
 
-      #endif
-
       hcfree (mod_info_log);
       hcfree (mod_error_log);
 
@@ -10870,35 +10636,27 @@ static bool load_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_p
 
       int mod_cnt = 6;
 
-      HIPjit_option mod_opts[7];
-      void *mod_vals[7];
+      hipJitOption mod_opts[6];
+      void *mod_vals[6];
 
-      mod_opts[0] = HIP_JIT_TARGET_FROM_HIPCONTEXT;
+      mod_opts[0] = hipJitOptionTargetFromContext;
       mod_vals[0] = (void *) 0;
 
-      mod_opts[1] = HIP_JIT_LOG_VERBOSE;
+      mod_opts[1] = hipJitOptionLogVerbose;
       mod_vals[1] = (void *) 1;
 
-      mod_opts[2] = HIP_JIT_INFO_LOG_BUFFER;
+      mod_opts[2] = hipJitOptionInfoLogBuffer;
       mod_vals[2] = (void *) mod_info_log;
 
-      mod_opts[3] = HIP_JIT_INFO_LOG_BUFFER_SIZE_BYTES;
+      mod_opts[3] = hipJitOptionInfoLogBufferSizeBytes;
       mod_vals[3] = (void *) LOG_SIZE;
 
-      mod_opts[4] = HIP_JIT_ERROR_LOG_BUFFER;
+      mod_opts[4] = hipJitOptionErrorLogBuffer;
       mod_vals[4] = (void *) mod_error_log;
 
-      mod_opts[5] = HIP_JIT_ERROR_LOG_BUFFER_SIZE_BYTES;
+      mod_opts[5] = hipJitOptionErrorLogBufferSizeBytes;
       mod_vals[5] = (void *) LOG_SIZE;
 
-      if (hashconfig->opti_type & OPTI_TYPE_REGISTER_LIMIT)
-      {
-        mod_opts[6] = HIP_JIT_MAX_REGISTERS;
-        mod_vals[6] = (void *) 128;
-
-        mod_cnt++;
-      }
-
       if (hc_hipModuleLoadDataEx (hashcat_ctx, hip_module, kernel_sources[0], mod_cnt, mod_opts, mod_vals) == -1)
       {
         event_log_error (hashcat_ctx, "* Device #%u: Kernel %s load failed. Error Log:", device_param->device_id + 1, source_file);
@@ -10985,6 +10743,48 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
       }
     }
 
+    /**
+     * tuning db
+     */
+
+    if (module_ctx->module_extra_tuningdb_block != MODULE_DEFAULT)
+    {
+      const char *extra_tuningdb_block = module_ctx->module_extra_tuningdb_block (hashconfig, user_options, user_options_extra);
+
+      char *lines_buf = hcstrdup (extra_tuningdb_block);
+
+      char *saveptr = NULL;
+
+      char *next = strtok_r (lines_buf, "\n", &saveptr);
+
+      int line_num = 0;
+
+      do
+      {
+        line_num++;
+
+        const size_t line_len = strlen (next);
+
+        if (line_len == 0) continue;
+
+        if (next[0] == '#') continue;
+
+        tuning_db_process_line (hashcat_ctx, next, line_num);
+
+      } while ((next = strtok_r ((char *) NULL, "\n", &saveptr)) != NULL);
+
+      hcfree (lines_buf);
+
+      // todo: print loaded 'cnt' message
+
+      // sort the database
+
+      tuning_db_t *tuning_db = hashcat_ctx->tuning_db;
+
+      qsort (tuning_db->alias_buf, tuning_db->alias_cnt, sizeof (tuning_db_alias_t), sort_by_tuning_db_alias);
+      qsort (tuning_db->entry_buf, tuning_db->entry_cnt, sizeof (tuning_db_entry_t), sort_by_tuning_db_entry);
+    }
+
     // vector_width
 
     int vector_width = 0;
@@ -11250,24 +11050,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
       }
       else if (device_param->opencl_device_type & CL_DEVICE_TYPE_GPU)
       {
-        // for GPU we need to distinguish by vendor
-
-        if (device_param->opencl_device_vendor_id == VENDOR_ID_INTEL_SDK)
-        {
-          native_threads = 8;
-        }
-        else if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD)
-        {
-          native_threads = 64;
-        }
-        else if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD_USE_HIP)
-        {
-          native_threads = 64;
-        }
-        else
-        {
-          native_threads = 32;
-        }
+        native_threads = device_param->kernel_preferred_wgs_multiple;
       }
       else
       {
@@ -11285,6 +11068,19 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
       }
     }
 
+    // this seems to work always
+
+    if (device_param->opencl_device_type & CL_DEVICE_TYPE_CPU)
+    {
+      u32 native_threads = 1;
+
+      if ((native_threads >= device_param->kernel_threads_min) && (native_threads <= device_param->kernel_threads_max))
+      {
+        device_param->kernel_threads_min = native_threads;
+        device_param->kernel_threads_max = native_threads;
+      }
+    }
+
     /**
      * create context for each device
      */
@@ -11306,7 +11102,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
     if (device_param->is_hip == true)
     {
-      if (hc_hipCtxCreate (hashcat_ctx, &device_param->hip_context, HIP_CTX_SCHED_BLOCKING_SYNC, device_param->hip_device) == -1)
+      if (hc_hipCtxCreate (hashcat_ctx, &device_param->hip_context, hipDeviceScheduleBlockingSync, device_param->hip_device) == -1)
       {
         device_param->skipped = true;
         continue;
@@ -11370,7 +11166,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
     if (device_param->is_hip == true)
     {
-      if (hc_hipStreamCreate (hashcat_ctx, &device_param->hip_stream, HIP_STREAM_DEFAULT) == -1)
+      if (hc_hipStreamCreate (hashcat_ctx, &device_param->hip_stream, hipStreamDefault) == -1)
       {
         device_param->skipped = true;
         continue;
@@ -11394,6 +11190,12 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
         device_param->skipped = true;
         continue;
       }
+
+      if (hc_cuEventCreate (hashcat_ctx, &device_param->cuda_event3, CU_EVENT_DISABLE_TIMING) == -1)
+      {
+        device_param->skipped = true;
+        continue;
+      }
     }
 
     /**
@@ -11402,13 +11204,19 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
     if (device_param->is_hip == true)
     {
-      if (hc_hipEventCreate (hashcat_ctx, &device_param->hip_event1, HIP_EVENT_BLOCKING_SYNC) == -1)
+      if (hc_hipEventCreate (hashcat_ctx, &device_param->hip_event1, hipEventBlockingSync) == -1)
       {
         device_param->skipped = true;
         continue;
       }
 
-      if (hc_hipEventCreate (hashcat_ctx, &device_param->hip_event2, HIP_EVENT_BLOCKING_SYNC) == -1)
+      if (hc_hipEventCreate (hashcat_ctx, &device_param->hip_event2, hipEventBlockingSync) == -1)
+      {
+        device_param->skipped = true;
+        continue;
+      }
+
+      if (hc_hipEventCreate (hashcat_ctx, &device_param->hip_event3, hipEventDisableTiming) == -1)
       {
         device_param->skipped = true;
         continue;
@@ -11593,15 +11401,16 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
     char device_name_chksum_amp_mp[HCBUFSIZ_TINY] = { 0 };
 
-    const size_t dnclen_amp_mp = snprintf (device_name_chksum_amp_mp, HCBUFSIZ_TINY, "%d-%d-%d-%d-%u-%s-%s-%s",
+    const size_t dnclen_amp_mp = snprintf (device_name_chksum_amp_mp, HCBUFSIZ_TINY, "%d-%d-%d-%d-%u-%s-%s-%s-%u",
       backend_ctx->comptime,
       backend_ctx->cuda_driver_version,
-      backend_ctx->hip_driver_version,
+      backend_ctx->hip_runtimeVersion,
       device_param->is_opencl,
       device_param->opencl_platform_vendor_id,
       device_param->device_name,
       device_param->opencl_device_version,
-      device_param->opencl_driver_version);
+      device_param->opencl_driver_version,
+      (user_options->kernel_threads_chgd == true) ? user_options->kernel_threads : device_param->kernel_threads_max);
 
     md5_ctx_t md5_ctx;
 
@@ -11685,13 +11494,21 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
         if (get_cuda_kernel_local_mem_size (hashcat_ctx, device_param->cuda_function_memset, &device_param->kernel_local_mem_size_memset) == -1) return -1;
 
-        if (get_cuda_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->cuda_function_memset, &device_param->kernel_dynamic_local_mem_size_memset) == -1) return -1;
+        device_param->kernel_dynamic_local_mem_size_memset = device_param->device_local_mem_size - device_param->kernel_local_mem_size_memset;
 
         device_param->kernel_preferred_wgs_multiple_memset = device_param->cuda_warp_size;
 
-        //CL_rc = hc_clSetKernelArg (hashcat_ctx, device_param->opencl_kernel_memset, 0, sizeof (cl_mem),   device_param->kernel_params_memset[0]); if (CL_rc == -1) return -1;
-        //CL_rc = hc_clSetKernelArg (hashcat_ctx, device_param->opencl_kernel_memset, 1, sizeof (cl_uint),  device_param->kernel_params_memset[1]); if (CL_rc == -1) return -1;
-        //CL_rc = hc_clSetKernelArg (hashcat_ctx, device_param->opencl_kernel_memset, 2, sizeof (cl_ulong), device_param->kernel_params_memset[2]); if (CL_rc == -1) return -1;
+        // GPU bzero
+
+        if (hc_cuModuleGetFunction (hashcat_ctx, &device_param->cuda_function_bzero, device_param->cuda_module_shared, "gpu_bzero") == -1) return -1;
+
+        if (get_cuda_kernel_wgs (hashcat_ctx, device_param->cuda_function_bzero, &device_param->kernel_wgs_bzero) == -1) return -1;
+
+        if (get_cuda_kernel_local_mem_size (hashcat_ctx, device_param->cuda_function_bzero, &device_param->kernel_local_mem_size_bzero) == -1) return -1;
+
+        device_param->kernel_dynamic_local_mem_size_bzero = device_param->device_local_mem_size - device_param->kernel_local_mem_size_bzero;
+
+        device_param->kernel_preferred_wgs_multiple_bzero = device_param->cuda_warp_size;
 
         // GPU autotune init
 
@@ -11701,7 +11518,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
         if (get_cuda_kernel_local_mem_size (hashcat_ctx, device_param->cuda_function_atinit, &device_param->kernel_local_mem_size_atinit) == -1) return -1;
 
-        if (get_cuda_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->cuda_function_atinit, &device_param->kernel_dynamic_local_mem_size_atinit) == -1) return -1;
+        device_param->kernel_dynamic_local_mem_size_atinit = device_param->device_local_mem_size - device_param->kernel_local_mem_size_atinit;
 
         device_param->kernel_preferred_wgs_multiple_atinit = device_param->cuda_warp_size;
 
@@ -11716,7 +11533,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
         if (get_cuda_kernel_local_mem_size (hashcat_ctx, device_param->cuda_function_decompress, &device_param->kernel_local_mem_size_decompress) == -1) return -1;
 
-        if (get_cuda_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->cuda_function_decompress, &device_param->kernel_dynamic_local_mem_size_decompress) == -1) return -1;
+        device_param->kernel_dynamic_local_mem_size_decompress = device_param->device_local_mem_size - device_param->kernel_local_mem_size_decompress;
 
         device_param->kernel_preferred_wgs_multiple_decompress = device_param->cuda_warp_size;
 
@@ -11728,7 +11545,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
         if (get_cuda_kernel_local_mem_size (hashcat_ctx, device_param->cuda_function_utf8toutf16le, &device_param->kernel_local_mem_size_utf8toutf16le) == -1) return -1;
 
-        if (get_cuda_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->cuda_function_utf8toutf16le, &device_param->kernel_dynamic_local_mem_size_utf8toutf16le) == -1) return -1;
+        device_param->kernel_dynamic_local_mem_size_utf8toutf16le = device_param->device_local_mem_size - device_param->kernel_local_mem_size_utf8toutf16le;
 
         device_param->kernel_preferred_wgs_multiple_utf8toutf16le = device_param->cuda_warp_size;
       }
@@ -11743,13 +11560,21 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
         if (get_hip_kernel_local_mem_size (hashcat_ctx, device_param->hip_function_memset, &device_param->kernel_local_mem_size_memset) == -1) return -1;
 
-        //if (get_hip_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->hip_function_memset, &device_param->kernel_dynamic_local_mem_size_memset) == -1) return -1;
+        device_param->kernel_dynamic_local_mem_size_memset = device_param->device_local_mem_size - device_param->kernel_local_mem_size_memset;
 
         device_param->kernel_preferred_wgs_multiple_memset = device_param->hip_warp_size;
 
-        //CL_rc = hc_clSetKernelArg (hashcat_ctx, device_param->opencl_kernel_memset, 0, sizeof (cl_mem),   device_param->kernel_params_memset[0]); if (CL_rc == -1) return -1;
-        //CL_rc = hc_clSetKernelArg (hashcat_ctx, device_param->opencl_kernel_memset, 1, sizeof (cl_uint),  device_param->kernel_params_memset[1]); if (CL_rc == -1) return -1;
-        //CL_rc = hc_clSetKernelArg (hashcat_ctx, device_param->opencl_kernel_memset, 2, sizeof (cl_ulong), device_param->kernel_params_memset[2]); if (CL_rc == -1) return -1;
+        // GPU bzero
+
+        if (hc_hipModuleGetFunction (hashcat_ctx, &device_param->hip_function_bzero, device_param->hip_module_shared, "gpu_bzero") == -1) return -1;
+
+        if (get_hip_kernel_wgs (hashcat_ctx, device_param->hip_function_bzero, &device_param->kernel_wgs_bzero) == -1) return -1;
+
+        if (get_hip_kernel_local_mem_size (hashcat_ctx, device_param->hip_function_bzero, &device_param->kernel_local_mem_size_bzero) == -1) return -1;
+
+        device_param->kernel_dynamic_local_mem_size_bzero = device_param->device_local_mem_size - device_param->kernel_local_mem_size_bzero;
+
+        device_param->kernel_preferred_wgs_multiple_bzero = device_param->hip_warp_size;
 
         // GPU autotune init
 
@@ -11759,7 +11584,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
         if (get_hip_kernel_local_mem_size (hashcat_ctx, device_param->hip_function_atinit, &device_param->kernel_local_mem_size_atinit) == -1) return -1;
 
-        //if (get_hip_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->hip_function_atinit, &device_param->kernel_dynamic_local_mem_size_atinit) == -1) return -1;
+        device_param->kernel_dynamic_local_mem_size_atinit = device_param->device_local_mem_size - device_param->kernel_local_mem_size_atinit;
 
         device_param->kernel_preferred_wgs_multiple_atinit = device_param->hip_warp_size;
 
@@ -11774,7 +11599,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
         if (get_hip_kernel_local_mem_size (hashcat_ctx, device_param->hip_function_decompress, &device_param->kernel_local_mem_size_decompress) == -1) return -1;
 
-        //if (get_hip_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->hip_function_decompress, &device_param->kernel_dynamic_local_mem_size_decompress) == -1) return -1;
+        device_param->kernel_dynamic_local_mem_size_decompress = device_param->device_local_mem_size - device_param->kernel_local_mem_size_decompress;
 
         device_param->kernel_preferred_wgs_multiple_decompress = device_param->hip_warp_size;
 
@@ -11786,7 +11611,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
         if (get_hip_kernel_local_mem_size (hashcat_ctx, device_param->hip_function_utf8toutf16le, &device_param->kernel_local_mem_size_utf8toutf16le) == -1) return -1;
 
-        //if (get_hip_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->hip_function_utf8toutf16le, &device_param->kernel_dynamic_local_mem_size_utf8toutf16le) == -1) return -1;
+        device_param->kernel_dynamic_local_mem_size_utf8toutf16le = device_param->device_local_mem_size - device_param->kernel_local_mem_size_utf8toutf16le;
 
         device_param->kernel_preferred_wgs_multiple_utf8toutf16le = device_param->hip_warp_size;
       }
@@ -11805,6 +11630,22 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
         if (get_opencl_kernel_preferred_wgs_multiple (hashcat_ctx, device_param, device_param->opencl_kernel_memset, &device_param->kernel_preferred_wgs_multiple_memset) == -1) return -1;
 
+        // GPU bzero
+
+        if (hc_clCreateKernel (hashcat_ctx, device_param->opencl_program_shared, "gpu_bzero", &device_param->opencl_kernel_bzero) == -1) return -1;
+
+        if (get_opencl_kernel_wgs (hashcat_ctx, device_param, device_param->opencl_kernel_bzero, &device_param->kernel_wgs_bzero) == -1) return -1;
+
+        if (get_opencl_kernel_local_mem_size (hashcat_ctx, device_param, device_param->opencl_kernel_bzero, &device_param->kernel_local_mem_size_bzero) == -1) return -1;
+
+        if (get_opencl_kernel_dynamic_local_mem_size (hashcat_ctx, device_param, device_param->opencl_kernel_bzero, &device_param->kernel_dynamic_local_mem_size_bzero) == -1) return -1;
+
+        if (get_opencl_kernel_preferred_wgs_multiple (hashcat_ctx, device_param, device_param->opencl_kernel_bzero, &device_param->kernel_preferred_wgs_multiple_bzero) == -1) return -1;
+
+        // apple hack, but perhaps also an alternative for other vendors
+
+        if (device_param->kernel_preferred_wgs_multiple == 0) device_param->kernel_preferred_wgs_multiple = device_param->kernel_preferred_wgs_multiple_bzero;
+
         // GPU autotune init
 
         if (hc_clCreateKernel (hashcat_ctx, device_param->opencl_program_shared, "gpu_atinit", &device_param->opencl_kernel_atinit) == -1) return -1;
@@ -11872,6 +11713,16 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
             device_param->kernel_threads_min = fixed_local_size;
             device_param->kernel_threads_max = fixed_local_size;
           }
+          else
+          {
+            // kernels specific minimum needs to be set so that self-test wont fail
+
+            if (sscanf (jit_build_options, "-D FIXED_LOCAL_SIZE_COMP=%u", &fixed_local_size) == 1)
+            {
+              device_param->kernel_threads_min = fixed_local_size;
+              // device_param->kernel_threads_max = fixed_local_size;
+            }
+          }
         }
       }
 
@@ -11892,10 +11743,10 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
       const u32 extra_value = (user_options->attack_mode == ATTACK_MODE_ASSOCIATION) ? ATTACK_MODE_ASSOCIATION : ATTACK_MODE_NONE;
 
-      const size_t dnclen = snprintf (device_name_chksum, HCBUFSIZ_TINY, "%d-%d-%d-%d-%u-%s-%s-%s-%d-%u-%u-%s",
+      const size_t dnclen = snprintf (device_name_chksum, HCBUFSIZ_TINY, "%d-%d-%d-%d-%u-%s-%s-%s-%d-%u-%u-%u-%s",
         backend_ctx->comptime,
         backend_ctx->cuda_driver_version,
-        backend_ctx->hip_driver_version,
+        backend_ctx->hip_runtimeVersion,
         device_param->is_opencl,
         device_param->opencl_platform_vendor_id,
         device_param->device_name,
@@ -11904,6 +11755,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
         device_param->vector_width,
         hashconfig->kern_type,
         extra_value,
+        (user_options->kernel_threads_chgd == true) ? user_options->kernel_threads : device_param->kernel_threads_max,
         build_options_module_buf);
 
       md5_ctx_t md5_ctx;
@@ -12130,16 +11982,16 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
       if (hc_cuMemAlloc (hashcat_ctx, &device_param->cuda_d_st_digests_buf, size_st_digests)         == -1) return -1;
       if (hc_cuMemAlloc (hashcat_ctx, &device_param->cuda_d_st_salts_buf,   size_st_salts)           == -1) return -1;
 
-      if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_bitmap_s1_a, bitmap_ctx->bitmap_s1_a, bitmap_ctx->bitmap_size) == -1) return -1;
-      if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_bitmap_s1_b, bitmap_ctx->bitmap_s1_b, bitmap_ctx->bitmap_size) == -1) return -1;
-      if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_bitmap_s1_c, bitmap_ctx->bitmap_s1_c, bitmap_ctx->bitmap_size) == -1) return -1;
-      if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_bitmap_s1_d, bitmap_ctx->bitmap_s1_d, bitmap_ctx->bitmap_size) == -1) return -1;
-      if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_bitmap_s2_a, bitmap_ctx->bitmap_s2_a, bitmap_ctx->bitmap_size) == -1) return -1;
-      if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_bitmap_s2_b, bitmap_ctx->bitmap_s2_b, bitmap_ctx->bitmap_size) == -1) return -1;
-      if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_bitmap_s2_c, bitmap_ctx->bitmap_s2_c, bitmap_ctx->bitmap_size) == -1) return -1;
-      if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_bitmap_s2_d, bitmap_ctx->bitmap_s2_d, bitmap_ctx->bitmap_size) == -1) return -1;
-      if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_digests_buf, hashes->digests_buf,     size_digests)            == -1) return -1;
-      if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_salt_bufs,   hashes->salts_buf,       size_salts)              == -1) return -1;
+      if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_bitmap_s1_a, bitmap_ctx->bitmap_s1_a, bitmap_ctx->bitmap_size, device_param->cuda_stream) == -1) return -1;
+      if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_bitmap_s1_b, bitmap_ctx->bitmap_s1_b, bitmap_ctx->bitmap_size, device_param->cuda_stream) == -1) return -1;
+      if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_bitmap_s1_c, bitmap_ctx->bitmap_s1_c, bitmap_ctx->bitmap_size, device_param->cuda_stream) == -1) return -1;
+      if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_bitmap_s1_d, bitmap_ctx->bitmap_s1_d, bitmap_ctx->bitmap_size, device_param->cuda_stream) == -1) return -1;
+      if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_bitmap_s2_a, bitmap_ctx->bitmap_s2_a, bitmap_ctx->bitmap_size, device_param->cuda_stream) == -1) return -1;
+      if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_bitmap_s2_b, bitmap_ctx->bitmap_s2_b, bitmap_ctx->bitmap_size, device_param->cuda_stream) == -1) return -1;
+      if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_bitmap_s2_c, bitmap_ctx->bitmap_s2_c, bitmap_ctx->bitmap_size, device_param->cuda_stream) == -1) return -1;
+      if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_bitmap_s2_d, bitmap_ctx->bitmap_s2_d, bitmap_ctx->bitmap_size, device_param->cuda_stream) == -1) return -1;
+      if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_digests_buf, hashes->digests_buf,     size_digests,            device_param->cuda_stream) == -1) return -1;
+      if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_salt_bufs,   hashes->salts_buf,       size_salts,              device_param->cuda_stream) == -1) return -1;
 
       /**
        * special buffers
@@ -12166,7 +12018,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
             if (hc_cuMemAlloc (hashcat_ctx, &device_param->cuda_d_rules_c, size_rules_c) == -1) return -1;
           }
 
-          if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_rules, straight_ctx->kernel_rules_buf, size_rules) == -1) return -1;
+          if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_rules, straight_ctx->kernel_rules_buf, size_rules, device_param->cuda_stream) == -1) return -1;
         }
         else if (user_options_extra->attack_kern == ATTACK_KERN_COMBI)
         {
@@ -12201,19 +12053,19 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
       {
         if (hc_cuMemAlloc (hashcat_ctx, &device_param->cuda_d_esalt_bufs, size_esalts) == -1) return -1;
 
-        if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_esalt_bufs, hashes->esalts_buf, size_esalts) == -1) return -1;
+        if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_esalt_bufs, hashes->esalts_buf, size_esalts, device_param->cuda_stream) == -1) return -1;
       }
 
       if (hashconfig->st_hash != NULL)
       {
-        if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_st_digests_buf, hashes->st_digests_buf, size_st_digests) == -1) return -1;
-        if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_st_salts_buf,   hashes->st_salts_buf,   size_st_salts)   == -1) return -1;
+        if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_st_digests_buf, hashes->st_digests_buf, size_st_digests, device_param->cuda_stream) == -1) return -1;
+        if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_st_salts_buf,   hashes->st_salts_buf,   size_st_salts,   device_param->cuda_stream)   == -1) return -1;
 
         if (size_esalts)
         {
           if (hc_cuMemAlloc (hashcat_ctx, &device_param->cuda_d_st_esalts_buf, size_st_esalts) == -1) return -1;
 
-          if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_st_esalts_buf, hashes->st_esalts_buf, size_st_esalts) == -1) return -1;
+          if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_st_esalts_buf, hashes->st_esalts_buf, size_st_esalts, device_param->cuda_stream) == -1) return -1;
         }
       }
     }
@@ -12240,16 +12092,16 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
       if (hc_hipMemAlloc (hashcat_ctx, &device_param->hip_d_st_digests_buf, size_st_digests)         == -1) return -1;
       if (hc_hipMemAlloc (hashcat_ctx, &device_param->hip_d_st_salts_buf,   size_st_salts)           == -1) return -1;
 
-      if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_bitmap_s1_a, bitmap_ctx->bitmap_s1_a, bitmap_ctx->bitmap_size) == -1) return -1;
-      if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_bitmap_s1_b, bitmap_ctx->bitmap_s1_b, bitmap_ctx->bitmap_size) == -1) return -1;
-      if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_bitmap_s1_c, bitmap_ctx->bitmap_s1_c, bitmap_ctx->bitmap_size) == -1) return -1;
-      if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_bitmap_s1_d, bitmap_ctx->bitmap_s1_d, bitmap_ctx->bitmap_size) == -1) return -1;
-      if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_bitmap_s2_a, bitmap_ctx->bitmap_s2_a, bitmap_ctx->bitmap_size) == -1) return -1;
-      if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_bitmap_s2_b, bitmap_ctx->bitmap_s2_b, bitmap_ctx->bitmap_size) == -1) return -1;
-      if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_bitmap_s2_c, bitmap_ctx->bitmap_s2_c, bitmap_ctx->bitmap_size) == -1) return -1;
-      if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_bitmap_s2_d, bitmap_ctx->bitmap_s2_d, bitmap_ctx->bitmap_size) == -1) return -1;
-      if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_digests_buf, hashes->digests_buf,     size_digests)            == -1) return -1;
-      if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_salt_bufs,   hashes->salts_buf,       size_salts)              == -1) return -1;
+      if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_bitmap_s1_a, bitmap_ctx->bitmap_s1_a, bitmap_ctx->bitmap_size, device_param->hip_stream) == -1) return -1;
+      if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_bitmap_s1_b, bitmap_ctx->bitmap_s1_b, bitmap_ctx->bitmap_size, device_param->hip_stream) == -1) return -1;
+      if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_bitmap_s1_c, bitmap_ctx->bitmap_s1_c, bitmap_ctx->bitmap_size, device_param->hip_stream) == -1) return -1;
+      if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_bitmap_s1_d, bitmap_ctx->bitmap_s1_d, bitmap_ctx->bitmap_size, device_param->hip_stream) == -1) return -1;
+      if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_bitmap_s2_a, bitmap_ctx->bitmap_s2_a, bitmap_ctx->bitmap_size, device_param->hip_stream) == -1) return -1;
+      if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_bitmap_s2_b, bitmap_ctx->bitmap_s2_b, bitmap_ctx->bitmap_size, device_param->hip_stream) == -1) return -1;
+      if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_bitmap_s2_c, bitmap_ctx->bitmap_s2_c, bitmap_ctx->bitmap_size, device_param->hip_stream) == -1) return -1;
+      if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_bitmap_s2_d, bitmap_ctx->bitmap_s2_d, bitmap_ctx->bitmap_size, device_param->hip_stream) == -1) return -1;
+      if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_digests_buf, hashes->digests_buf,     size_digests,            device_param->hip_stream) == -1) return -1;
+      if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_salt_bufs,   hashes->salts_buf,       size_salts,              device_param->hip_stream) == -1) return -1;
 
       /**
        * special buffers
@@ -12276,7 +12128,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
             if (hc_hipMemAlloc (hashcat_ctx, &device_param->hip_d_rules_c, size_rules_c) == -1) return -1;
           }
 
-          if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_rules, straight_ctx->kernel_rules_buf, size_rules) == -1) return -1;
+          if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_rules, straight_ctx->kernel_rules_buf, size_rules, device_param->hip_stream) == -1) return -1;
         }
         else if (user_options_extra->attack_kern == ATTACK_KERN_COMBI)
         {
@@ -12311,19 +12163,19 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
       {
         if (hc_hipMemAlloc (hashcat_ctx, &device_param->hip_d_esalt_bufs, size_esalts) == -1) return -1;
 
-        if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_esalt_bufs, hashes->esalts_buf, size_esalts) == -1) return -1;
+        if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_esalt_bufs, hashes->esalts_buf, size_esalts, device_param->hip_stream) == -1) return -1;
       }
 
       if (hashconfig->st_hash != NULL)
       {
-        if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_st_digests_buf, hashes->st_digests_buf, size_st_digests) == -1) return -1;
-        if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_st_salts_buf,   hashes->st_salts_buf,   size_st_salts)   == -1) return -1;
+        if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_st_digests_buf, hashes->st_digests_buf, size_st_digests, device_param->hip_stream) == -1) return -1;
+        if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_st_salts_buf,   hashes->st_salts_buf,   size_st_salts,   device_param->hip_stream) == -1) return -1;
 
         if (size_esalts)
         {
           if (hc_hipMemAlloc (hashcat_ctx, &device_param->hip_d_st_esalts_buf, size_st_esalts) == -1) return -1;
 
-          if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_st_esalts_buf, hashes->st_esalts_buf, size_st_esalts) == -1) return -1;
+          if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_st_esalts_buf, hashes->st_esalts_buf, size_st_esalts, device_param->hip_stream) == -1) return -1;
         }
       }
     }
@@ -12350,16 +12202,16 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
       if (hc_clCreateBuffer (hashcat_ctx, device_param->opencl_context, CL_MEM_READ_ONLY,   size_st_digests,         NULL, &device_param->opencl_d_st_digests_buf) == -1) return -1;
       if (hc_clCreateBuffer (hashcat_ctx, device_param->opencl_context, CL_MEM_READ_ONLY,   size_st_salts,           NULL, &device_param->opencl_d_st_salts_buf)   == -1) return -1;
 
-      if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_bitmap_s1_a, CL_TRUE, 0, bitmap_ctx->bitmap_size, bitmap_ctx->bitmap_s1_a, 0, NULL, NULL) == -1) return -1;
-      if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_bitmap_s1_b, CL_TRUE, 0, bitmap_ctx->bitmap_size, bitmap_ctx->bitmap_s1_b, 0, NULL, NULL) == -1) return -1;
-      if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_bitmap_s1_c, CL_TRUE, 0, bitmap_ctx->bitmap_size, bitmap_ctx->bitmap_s1_c, 0, NULL, NULL) == -1) return -1;
-      if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_bitmap_s1_d, CL_TRUE, 0, bitmap_ctx->bitmap_size, bitmap_ctx->bitmap_s1_d, 0, NULL, NULL) == -1) return -1;
-      if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_bitmap_s2_a, CL_TRUE, 0, bitmap_ctx->bitmap_size, bitmap_ctx->bitmap_s2_a, 0, NULL, NULL) == -1) return -1;
-      if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_bitmap_s2_b, CL_TRUE, 0, bitmap_ctx->bitmap_size, bitmap_ctx->bitmap_s2_b, 0, NULL, NULL) == -1) return -1;
-      if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_bitmap_s2_c, CL_TRUE, 0, bitmap_ctx->bitmap_size, bitmap_ctx->bitmap_s2_c, 0, NULL, NULL) == -1) return -1;
-      if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_bitmap_s2_d, CL_TRUE, 0, bitmap_ctx->bitmap_size, bitmap_ctx->bitmap_s2_d, 0, NULL, NULL) == -1) return -1;
-      if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_digests_buf, CL_TRUE, 0, size_digests,            hashes->digests_buf,     0, NULL, NULL) == -1) return -1;
-      if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_salt_bufs,   CL_TRUE, 0, size_salts,              hashes->salts_buf,       0, NULL, NULL) == -1) return -1;
+      if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_bitmap_s1_a, CL_FALSE, 0, bitmap_ctx->bitmap_size, bitmap_ctx->bitmap_s1_a, 0, NULL, NULL) == -1) return -1;
+      if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_bitmap_s1_b, CL_FALSE, 0, bitmap_ctx->bitmap_size, bitmap_ctx->bitmap_s1_b, 0, NULL, NULL) == -1) return -1;
+      if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_bitmap_s1_c, CL_FALSE, 0, bitmap_ctx->bitmap_size, bitmap_ctx->bitmap_s1_c, 0, NULL, NULL) == -1) return -1;
+      if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_bitmap_s1_d, CL_FALSE, 0, bitmap_ctx->bitmap_size, bitmap_ctx->bitmap_s1_d, 0, NULL, NULL) == -1) return -1;
+      if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_bitmap_s2_a, CL_FALSE, 0, bitmap_ctx->bitmap_size, bitmap_ctx->bitmap_s2_a, 0, NULL, NULL) == -1) return -1;
+      if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_bitmap_s2_b, CL_FALSE, 0, bitmap_ctx->bitmap_size, bitmap_ctx->bitmap_s2_b, 0, NULL, NULL) == -1) return -1;
+      if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_bitmap_s2_c, CL_FALSE, 0, bitmap_ctx->bitmap_size, bitmap_ctx->bitmap_s2_c, 0, NULL, NULL) == -1) return -1;
+      if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_bitmap_s2_d, CL_FALSE, 0, bitmap_ctx->bitmap_size, bitmap_ctx->bitmap_s2_d, 0, NULL, NULL) == -1) return -1;
+      if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_digests_buf, CL_FALSE, 0, size_digests,            hashes->digests_buf,     0, NULL, NULL) == -1) return -1;
+      if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_salt_bufs,   CL_FALSE, 0, size_salts,              hashes->salts_buf,       0, NULL, NULL) == -1) return -1;
 
       /**
        * special buffers
@@ -12376,7 +12228,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
           if (hc_clCreateBuffer (hashcat_ctx, device_param->opencl_context, CL_MEM_READ_ONLY, size_rules,   NULL, &device_param->opencl_d_rules)   == -1) return -1;
           if (hc_clCreateBuffer (hashcat_ctx, device_param->opencl_context, CL_MEM_READ_ONLY, size_rules_c, NULL, &device_param->opencl_d_rules_c) == -1) return -1;
 
-          if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_rules, CL_TRUE, 0, size_rules, straight_ctx->kernel_rules_buf, 0, NULL, NULL) == -1) return -1;
+          if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_rules, CL_FALSE, 0, size_rules, straight_ctx->kernel_rules_buf, 0, NULL, NULL) == -1) return -1;
         }
         else if (user_options_extra->attack_kern == ATTACK_KERN_COMBI)
         {
@@ -12399,21 +12251,23 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
       {
         if (hc_clCreateBuffer (hashcat_ctx, device_param->opencl_context, CL_MEM_READ_ONLY, size_esalts, NULL, &device_param->opencl_d_esalt_bufs) == -1) return -1;
 
-        if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_esalt_bufs, CL_TRUE, 0, size_esalts, hashes->esalts_buf, 0, NULL, NULL) == -1) return -1;
+        if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_esalt_bufs, CL_FALSE, 0, size_esalts, hashes->esalts_buf, 0, NULL, NULL) == -1) return -1;
       }
 
       if (hashconfig->st_hash != NULL)
       {
-        if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_st_digests_buf,  CL_TRUE, 0, size_st_digests,         hashes->st_digests_buf,  0, NULL, NULL) == -1) return -1;
-        if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_st_salts_buf,    CL_TRUE, 0, size_st_salts,           hashes->st_salts_buf,    0, NULL, NULL) == -1) return -1;
+        if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_st_digests_buf,  CL_FALSE, 0, size_st_digests,         hashes->st_digests_buf,  0, NULL, NULL) == -1) return -1;
+        if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_st_salts_buf,    CL_FALSE, 0, size_st_salts,           hashes->st_salts_buf,    0, NULL, NULL) == -1) return -1;
 
         if (size_esalts)
         {
           if (hc_clCreateBuffer (hashcat_ctx, device_param->opencl_context, CL_MEM_READ_ONLY, size_st_esalts, NULL, &device_param->opencl_d_st_esalts_buf) == -1) return -1;
 
-          if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_st_esalts_buf, CL_TRUE, 0, size_st_esalts, hashes->st_esalts_buf, 0, NULL, NULL) == -1) return -1;
+          if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_st_esalts_buf, CL_FALSE, 0, size_st_esalts, hashes->st_esalts_buf, 0, NULL, NULL) == -1) return -1;
         }
       }
+
+      if (hc_clFlush (hashcat_ctx, device_param->opencl_command_queue) == -1) return -1;
     }
 
     /**
@@ -12745,6 +12599,11 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
     device_param->kernel_params_memset[1] = &device_param->kernel_params_memset_buf32[1];
     device_param->kernel_params_memset[2] = &device_param->kernel_params_memset_buf64[2];
 
+    device_param->kernel_params_bzero_buf64[1] = 0; // gid_max
+
+    device_param->kernel_params_bzero[0] = NULL;
+    device_param->kernel_params_bzero[1] = &device_param->kernel_params_bzero_buf64[1];
+
     device_param->kernel_params_atinit_buf64[1] = 0; // gid_max
 
     device_param->kernel_params_atinit[0] = NULL;
@@ -12810,7 +12669,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
             if (get_cuda_kernel_local_mem_size (hashcat_ctx, device_param->cuda_function1, &device_param->kernel_local_mem_size1) == -1) return -1;
 
-            if (get_cuda_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->cuda_function1, &device_param->kernel_dynamic_local_mem_size1) == -1) return -1;
+            device_param->kernel_dynamic_local_mem_size1 = device_param->device_local_mem_size - device_param->kernel_local_mem_size1;
 
             device_param->kernel_preferred_wgs_multiple1 = device_param->cuda_warp_size;
 
@@ -12824,7 +12683,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
             if (get_cuda_kernel_local_mem_size (hashcat_ctx, device_param->cuda_function2, &device_param->kernel_local_mem_size2) == -1) return -1;
 
-            if (get_cuda_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->cuda_function2, &device_param->kernel_dynamic_local_mem_size2) == -1) return -1;
+            device_param->kernel_dynamic_local_mem_size2 = device_param->device_local_mem_size - device_param->kernel_local_mem_size2;
 
             device_param->kernel_preferred_wgs_multiple2 = device_param->cuda_warp_size;
 
@@ -12838,7 +12697,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
             if (get_cuda_kernel_local_mem_size (hashcat_ctx, device_param->cuda_function3, &device_param->kernel_local_mem_size3) == -1) return -1;
 
-            if (get_cuda_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->cuda_function3, &device_param->kernel_dynamic_local_mem_size3) == -1) return -1;
+            device_param->kernel_dynamic_local_mem_size3 = device_param->device_local_mem_size - device_param->kernel_local_mem_size3;
 
             device_param->kernel_preferred_wgs_multiple3 = device_param->cuda_warp_size;
           }
@@ -12852,7 +12711,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
             if (get_cuda_kernel_local_mem_size (hashcat_ctx, device_param->cuda_function4, &device_param->kernel_local_mem_size4) == -1) return -1;
 
-            if (get_cuda_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->cuda_function4, &device_param->kernel_dynamic_local_mem_size4) == -1) return -1;
+            device_param->kernel_dynamic_local_mem_size4 = device_param->device_local_mem_size - device_param->kernel_local_mem_size4;
 
             device_param->kernel_preferred_wgs_multiple4 = device_param->cuda_warp_size;
           }
@@ -12871,7 +12730,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
             if (get_cuda_kernel_local_mem_size (hashcat_ctx, device_param->cuda_function1, &device_param->kernel_local_mem_size1) == -1) return -1;
 
-            if (get_cuda_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->cuda_function1, &device_param->kernel_dynamic_local_mem_size1) == -1) return -1;
+            device_param->kernel_dynamic_local_mem_size1 = device_param->device_local_mem_size - device_param->kernel_local_mem_size1;
 
             device_param->kernel_preferred_wgs_multiple1 = device_param->cuda_warp_size;
 
@@ -12885,7 +12744,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
             if (get_cuda_kernel_local_mem_size (hashcat_ctx, device_param->cuda_function2, &device_param->kernel_local_mem_size2) == -1) return -1;
 
-            if (get_cuda_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->cuda_function2, &device_param->kernel_dynamic_local_mem_size2) == -1) return -1;
+            device_param->kernel_dynamic_local_mem_size2 = device_param->device_local_mem_size - device_param->kernel_local_mem_size2;
 
             device_param->kernel_preferred_wgs_multiple2 = device_param->cuda_warp_size;
 
@@ -12899,7 +12758,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
             if (get_cuda_kernel_local_mem_size (hashcat_ctx, device_param->cuda_function3, &device_param->kernel_local_mem_size3) == -1) return -1;
 
-            if (get_cuda_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->cuda_function3, &device_param->kernel_dynamic_local_mem_size3) == -1) return -1;
+            device_param->kernel_dynamic_local_mem_size3 = device_param->device_local_mem_size - device_param->kernel_local_mem_size3;
 
             device_param->kernel_preferred_wgs_multiple3 = device_param->cuda_warp_size;
           }
@@ -12913,7 +12772,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
             if (get_cuda_kernel_local_mem_size (hashcat_ctx, device_param->cuda_function4, &device_param->kernel_local_mem_size4) == -1) return -1;
 
-            if (get_cuda_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->cuda_function4, &device_param->kernel_dynamic_local_mem_size4) == -1) return -1;
+            device_param->kernel_dynamic_local_mem_size4 = device_param->device_local_mem_size - device_param->kernel_local_mem_size4;
 
             device_param->kernel_preferred_wgs_multiple4 = device_param->cuda_warp_size;
           }
@@ -12936,7 +12795,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
               if (get_cuda_kernel_local_mem_size (hashcat_ctx, device_param->cuda_function_tm, &device_param->kernel_local_mem_size_tm) == -1) return -1;
 
-              if (get_cuda_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->cuda_function_tm, &device_param->kernel_dynamic_local_mem_size_tm) == -1) return -1;
+              device_param->kernel_dynamic_local_mem_size_tm = device_param->device_local_mem_size - device_param->kernel_local_mem_size_tm;
 
               device_param->kernel_preferred_wgs_multiple_tm = device_param->cuda_warp_size;
             }
@@ -12955,7 +12814,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
         if (get_cuda_kernel_local_mem_size (hashcat_ctx, device_param->cuda_function1, &device_param->kernel_local_mem_size1) == -1) return -1;
 
-        if (get_cuda_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->cuda_function1, &device_param->kernel_dynamic_local_mem_size1) == -1) return -1;
+        device_param->kernel_dynamic_local_mem_size1 = device_param->device_local_mem_size - device_param->kernel_local_mem_size1;
 
         device_param->kernel_preferred_wgs_multiple1 = device_param->cuda_warp_size;
 
@@ -12969,7 +12828,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
         if (get_cuda_kernel_local_mem_size (hashcat_ctx, device_param->cuda_function2, &device_param->kernel_local_mem_size2) == -1) return -1;
 
-        if (get_cuda_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->cuda_function2, &device_param->kernel_dynamic_local_mem_size2) == -1) return -1;
+        device_param->kernel_dynamic_local_mem_size2 = device_param->device_local_mem_size - device_param->kernel_local_mem_size2;
 
         device_param->kernel_preferred_wgs_multiple2 = device_param->cuda_warp_size;
 
@@ -12983,7 +12842,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
         if (get_cuda_kernel_local_mem_size (hashcat_ctx, device_param->cuda_function3, &device_param->kernel_local_mem_size3) == -1) return -1;
 
-        if (get_cuda_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->cuda_function3, &device_param->kernel_dynamic_local_mem_size3) == -1) return -1;
+        device_param->kernel_dynamic_local_mem_size3 = device_param->device_local_mem_size - device_param->kernel_local_mem_size3;
 
         device_param->kernel_preferred_wgs_multiple3 = device_param->cuda_warp_size;
 
@@ -12999,7 +12858,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
           if (get_cuda_kernel_local_mem_size (hashcat_ctx, device_param->cuda_function2p, &device_param->kernel_local_mem_size2p) == -1) return -1;
 
-          if (get_cuda_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->cuda_function2p, &device_param->kernel_dynamic_local_mem_size2p) == -1) return -1;
+          device_param->kernel_dynamic_local_mem_size2p = device_param->device_local_mem_size - device_param->kernel_local_mem_size2p;
 
           device_param->kernel_preferred_wgs_multiple2p = device_param->cuda_warp_size;
         }
@@ -13016,7 +12875,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
           if (get_cuda_kernel_local_mem_size (hashcat_ctx, device_param->cuda_function2e, &device_param->kernel_local_mem_size2e) == -1) return -1;
 
-          if (get_cuda_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->cuda_function2e, &device_param->kernel_dynamic_local_mem_size2e) == -1) return -1;
+          device_param->kernel_dynamic_local_mem_size2e = device_param->device_local_mem_size - device_param->kernel_local_mem_size2e;
 
           device_param->kernel_preferred_wgs_multiple2e = device_param->cuda_warp_size;
         }
@@ -13033,7 +12892,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
           if (get_cuda_kernel_local_mem_size (hashcat_ctx, device_param->cuda_function12, &device_param->kernel_local_mem_size12) == -1) return -1;
 
-          if (get_cuda_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->cuda_function12, &device_param->kernel_dynamic_local_mem_size12) == -1) return -1;
+          device_param->kernel_dynamic_local_mem_size12 = device_param->device_local_mem_size - device_param->kernel_local_mem_size12;
 
           device_param->kernel_preferred_wgs_multiple12 = device_param->cuda_warp_size;
         }
@@ -13050,7 +12909,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
           if (get_cuda_kernel_local_mem_size (hashcat_ctx, device_param->cuda_function23, &device_param->kernel_local_mem_size23) == -1) return -1;
 
-          if (get_cuda_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->cuda_function23, &device_param->kernel_dynamic_local_mem_size23) == -1) return -1;
+          device_param->kernel_dynamic_local_mem_size23 = device_param->device_local_mem_size - device_param->kernel_local_mem_size23;
 
           device_param->kernel_preferred_wgs_multiple23 = device_param->cuda_warp_size;
         }
@@ -13067,7 +12926,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
           if (get_cuda_kernel_local_mem_size (hashcat_ctx, device_param->cuda_function_init2, &device_param->kernel_local_mem_size_init2) == -1) return -1;
 
-          if (get_cuda_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->cuda_function_init2, &device_param->kernel_dynamic_local_mem_size_init2) == -1) return -1;
+          device_param->kernel_dynamic_local_mem_size_init2 = device_param->device_local_mem_size - device_param->kernel_local_mem_size_init2;
 
           device_param->kernel_preferred_wgs_multiple_init2 = device_param->cuda_warp_size;
         }
@@ -13084,7 +12943,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
           if (get_cuda_kernel_local_mem_size (hashcat_ctx, device_param->cuda_function_loop2p, &device_param->kernel_local_mem_size_loop2p) == -1) return -1;
 
-          if (get_cuda_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->cuda_function_loop2p, &device_param->kernel_dynamic_local_mem_size_loop2p) == -1) return -1;
+          device_param->kernel_dynamic_local_mem_size_loop2p = device_param->device_local_mem_size - device_param->kernel_local_mem_size_loop2p;
 
           device_param->kernel_preferred_wgs_multiple_loop2p = device_param->cuda_warp_size;
         }
@@ -13101,7 +12960,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
           if (get_cuda_kernel_local_mem_size (hashcat_ctx, device_param->cuda_function_loop2, &device_param->kernel_local_mem_size_loop2) == -1) return -1;
 
-          if (get_cuda_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->cuda_function_loop2, &device_param->kernel_dynamic_local_mem_size_loop2) == -1) return -1;
+          device_param->kernel_dynamic_local_mem_size_loop2 = device_param->device_local_mem_size - device_param->kernel_local_mem_size_loop2;
 
           device_param->kernel_preferred_wgs_multiple_loop2 = device_param->cuda_warp_size;
         }
@@ -13118,7 +12977,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
           if (get_cuda_kernel_local_mem_size (hashcat_ctx, device_param->cuda_function_aux1, &device_param->kernel_local_mem_size_aux1) == -1) return -1;
 
-          if (get_cuda_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->cuda_function_aux1, &device_param->kernel_dynamic_local_mem_size_aux1) == -1) return -1;
+          device_param->kernel_dynamic_local_mem_size_aux1 = device_param->device_local_mem_size - device_param->kernel_local_mem_size_aux1;
 
           device_param->kernel_preferred_wgs_multiple_aux1 = device_param->cuda_warp_size;
         }
@@ -13135,7 +12994,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
           if (get_cuda_kernel_local_mem_size (hashcat_ctx, device_param->cuda_function_aux2, &device_param->kernel_local_mem_size_aux2) == -1) return -1;
 
-          if (get_cuda_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->cuda_function_aux2, &device_param->kernel_dynamic_local_mem_size_aux2) == -1) return -1;
+          device_param->kernel_dynamic_local_mem_size_aux2 = device_param->device_local_mem_size - device_param->kernel_local_mem_size_aux2;
 
           device_param->kernel_preferred_wgs_multiple_aux2 = device_param->cuda_warp_size;
         }
@@ -13152,7 +13011,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
           if (get_cuda_kernel_local_mem_size (hashcat_ctx, device_param->cuda_function_aux3, &device_param->kernel_local_mem_size_aux3) == -1) return -1;
 
-          if (get_cuda_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->cuda_function_aux3, &device_param->kernel_dynamic_local_mem_size_aux3) == -1) return -1;
+          device_param->kernel_dynamic_local_mem_size_aux3 = device_param->device_local_mem_size - device_param->kernel_local_mem_size_aux3;
 
           device_param->kernel_preferred_wgs_multiple_aux3 = device_param->cuda_warp_size;
         }
@@ -13169,7 +13028,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
           if (get_cuda_kernel_local_mem_size (hashcat_ctx, device_param->cuda_function_aux4, &device_param->kernel_local_mem_size_aux4) == -1) return -1;
 
-          if (get_cuda_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->cuda_function_aux4, &device_param->kernel_dynamic_local_mem_size_aux4) == -1) return -1;
+          device_param->kernel_dynamic_local_mem_size_aux4 = device_param->device_local_mem_size - device_param->kernel_local_mem_size_aux4;
 
           device_param->kernel_preferred_wgs_multiple_aux4 = device_param->cuda_warp_size;
         }
@@ -13197,7 +13056,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
           if (get_cuda_kernel_local_mem_size (hashcat_ctx, device_param->cuda_function_mp_l, &device_param->kernel_local_mem_size_mp_l) == -1) return -1;
 
-          if (get_cuda_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->cuda_function_mp_l, &device_param->kernel_dynamic_local_mem_size_mp_l) == -1) return -1;
+          device_param->kernel_dynamic_local_mem_size_mp_l = device_param->device_local_mem_size - device_param->kernel_local_mem_size_mp_l;
 
           device_param->kernel_preferred_wgs_multiple_mp_l = device_param->cuda_warp_size;
 
@@ -13209,7 +13068,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
           if (get_cuda_kernel_local_mem_size (hashcat_ctx, device_param->cuda_function_mp_r, &device_param->kernel_local_mem_size_mp_r) == -1) return -1;
 
-          if (get_cuda_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->cuda_function_mp_r, &device_param->kernel_dynamic_local_mem_size_mp_r) == -1) return -1;
+          device_param->kernel_dynamic_local_mem_size_mp_r = device_param->device_local_mem_size - device_param->kernel_local_mem_size_mp_r;
 
           device_param->kernel_preferred_wgs_multiple_mp_r = device_param->cuda_warp_size;
 
@@ -13230,7 +13089,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
           if (get_cuda_kernel_local_mem_size (hashcat_ctx, device_param->cuda_function_mp, &device_param->kernel_local_mem_size_mp) == -1) return -1;
 
-          if (get_cuda_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->cuda_function_mp, &device_param->kernel_dynamic_local_mem_size_mp) == -1) return -1;
+          device_param->kernel_dynamic_local_mem_size_mp = device_param->device_local_mem_size - device_param->kernel_local_mem_size_mp;
 
           device_param->kernel_preferred_wgs_multiple_mp = device_param->cuda_warp_size;
         }
@@ -13242,7 +13101,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
           if (get_cuda_kernel_local_mem_size (hashcat_ctx, device_param->cuda_function_mp, &device_param->kernel_local_mem_size_mp) == -1) return -1;
 
-          if (get_cuda_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->cuda_function_mp, &device_param->kernel_dynamic_local_mem_size_mp) == -1) return -1;
+          device_param->kernel_dynamic_local_mem_size_mp = device_param->device_local_mem_size - device_param->kernel_local_mem_size_mp;
 
           device_param->kernel_preferred_wgs_multiple_mp = device_param->cuda_warp_size;
         }
@@ -13265,7 +13124,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
           if (get_cuda_kernel_local_mem_size (hashcat_ctx, device_param->cuda_function_amp, &device_param->kernel_local_mem_size_amp) == -1) return -1;
 
-          if (get_cuda_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->cuda_function_amp, &device_param->kernel_dynamic_local_mem_size_amp) == -1) return -1;
+          device_param->kernel_dynamic_local_mem_size_amp = device_param->device_local_mem_size - device_param->kernel_local_mem_size_amp;
 
           device_param->kernel_preferred_wgs_multiple_amp = device_param->cuda_warp_size;
         }
@@ -13412,7 +13271,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
             if (get_hip_kernel_local_mem_size (hashcat_ctx, device_param->hip_function1, &device_param->kernel_local_mem_size1) == -1) return -1;
 
-            //if (get_hip_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->hip_function1, &device_param->kernel_dynamic_local_mem_size1) == -1) return -1;
+            device_param->kernel_dynamic_local_mem_size1 = device_param->device_local_mem_size - device_param->kernel_local_mem_size1;
 
             device_param->kernel_preferred_wgs_multiple1 = device_param->hip_warp_size;
 
@@ -13426,7 +13285,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
             if (get_hip_kernel_local_mem_size (hashcat_ctx, device_param->hip_function2, &device_param->kernel_local_mem_size2) == -1) return -1;
 
-            //if (get_hip_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->hip_function2, &device_param->kernel_dynamic_local_mem_size2) == -1) return -1;
+            device_param->kernel_dynamic_local_mem_size2 = device_param->device_local_mem_size - device_param->kernel_local_mem_size2;
 
             device_param->kernel_preferred_wgs_multiple2 = device_param->hip_warp_size;
 
@@ -13440,7 +13299,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
             if (get_hip_kernel_local_mem_size (hashcat_ctx, device_param->hip_function3, &device_param->kernel_local_mem_size3) == -1) return -1;
 
-            //if (get_hip_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->hip_function3, &device_param->kernel_dynamic_local_mem_size3) == -1) return -1;
+            device_param->kernel_dynamic_local_mem_size3 = device_param->device_local_mem_size - device_param->kernel_local_mem_size3;
 
             device_param->kernel_preferred_wgs_multiple3 = device_param->hip_warp_size;
           }
@@ -13454,7 +13313,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
             if (get_hip_kernel_local_mem_size (hashcat_ctx, device_param->hip_function4, &device_param->kernel_local_mem_size4) == -1) return -1;
 
-            //if (get_hip_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->hip_function4, &device_param->kernel_dynamic_local_mem_size4) == -1) return -1;
+            device_param->kernel_dynamic_local_mem_size4 = device_param->device_local_mem_size - device_param->kernel_local_mem_size4;
 
             device_param->kernel_preferred_wgs_multiple4 = device_param->hip_warp_size;
           }
@@ -13473,7 +13332,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
             if (get_hip_kernel_local_mem_size (hashcat_ctx, device_param->hip_function1, &device_param->kernel_local_mem_size1) == -1) return -1;
 
-            //if (get_hip_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->hip_function1, &device_param->kernel_dynamic_local_mem_size1) == -1) return -1;
+            device_param->kernel_dynamic_local_mem_size1 = device_param->device_local_mem_size - device_param->kernel_local_mem_size1;
 
             device_param->kernel_preferred_wgs_multiple1 = device_param->hip_warp_size;
 
@@ -13487,7 +13346,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
             if (get_hip_kernel_local_mem_size (hashcat_ctx, device_param->hip_function2, &device_param->kernel_local_mem_size2) == -1) return -1;
 
-            //if (get_hip_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->hip_function2, &device_param->kernel_dynamic_local_mem_size2) == -1) return -1;
+            device_param->kernel_dynamic_local_mem_size2 = device_param->device_local_mem_size - device_param->kernel_local_mem_size2;
 
             device_param->kernel_preferred_wgs_multiple2 = device_param->hip_warp_size;
 
@@ -13501,7 +13360,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
             if (get_hip_kernel_local_mem_size (hashcat_ctx, device_param->hip_function3, &device_param->kernel_local_mem_size3) == -1) return -1;
 
-            //if (get_hip_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->hip_function3, &device_param->kernel_dynamic_local_mem_size3) == -1) return -1;
+            device_param->kernel_dynamic_local_mem_size3 = device_param->device_local_mem_size - device_param->kernel_local_mem_size3;
 
             device_param->kernel_preferred_wgs_multiple3 = device_param->hip_warp_size;
           }
@@ -13515,7 +13374,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
             if (get_hip_kernel_local_mem_size (hashcat_ctx, device_param->hip_function4, &device_param->kernel_local_mem_size4) == -1) return -1;
 
-            //if (get_hip_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->hip_function4, &device_param->kernel_dynamic_local_mem_size4) == -1) return -1;
+            device_param->kernel_dynamic_local_mem_size4 = device_param->device_local_mem_size - device_param->kernel_local_mem_size4;
 
             device_param->kernel_preferred_wgs_multiple4 = device_param->hip_warp_size;
           }
@@ -13538,7 +13397,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
               if (get_hip_kernel_local_mem_size (hashcat_ctx, device_param->hip_function_tm, &device_param->kernel_local_mem_size_tm) == -1) return -1;
 
-              //if (get_hip_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->hip_function_tm, &device_param->kernel_dynamic_local_mem_size_tm) == -1) return -1;
+              device_param->kernel_dynamic_local_mem_size_tm = device_param->device_local_mem_size - device_param->kernel_local_mem_size_tm;
 
               device_param->kernel_preferred_wgs_multiple_tm = device_param->hip_warp_size;
             }
@@ -13557,7 +13416,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
         if (get_hip_kernel_local_mem_size (hashcat_ctx, device_param->hip_function1, &device_param->kernel_local_mem_size1) == -1) return -1;
 
-        //if (get_hip_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->hip_function1, &device_param->kernel_dynamic_local_mem_size1) == -1) return -1;
+        device_param->kernel_dynamic_local_mem_size1 = device_param->device_local_mem_size - device_param->kernel_local_mem_size1;
 
         device_param->kernel_preferred_wgs_multiple1 = device_param->hip_warp_size;
 
@@ -13571,7 +13430,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
         if (get_hip_kernel_local_mem_size (hashcat_ctx, device_param->hip_function2, &device_param->kernel_local_mem_size2) == -1) return -1;
 
-        //if (get_hip_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->hip_function2, &device_param->kernel_dynamic_local_mem_size2) == -1) return -1;
+        device_param->kernel_dynamic_local_mem_size2 = device_param->device_local_mem_size - device_param->kernel_local_mem_size2;
 
         device_param->kernel_preferred_wgs_multiple2 = device_param->hip_warp_size;
 
@@ -13585,7 +13444,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
         if (get_hip_kernel_local_mem_size (hashcat_ctx, device_param->hip_function3, &device_param->kernel_local_mem_size3) == -1) return -1;
 
-        //if (get_hip_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->hip_function3, &device_param->kernel_dynamic_local_mem_size3) == -1) return -1;
+        device_param->kernel_dynamic_local_mem_size3 = device_param->device_local_mem_size - device_param->kernel_local_mem_size3;
 
         device_param->kernel_preferred_wgs_multiple3 = device_param->hip_warp_size;
 
@@ -13601,7 +13460,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
           if (get_hip_kernel_local_mem_size (hashcat_ctx, device_param->hip_function2p, &device_param->kernel_local_mem_size2p) == -1) return -1;
 
-          //if (get_hip_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->hip_function2p, &device_param->kernel_dynamic_local_mem_size2p) == -1) return -1;
+          device_param->kernel_dynamic_local_mem_size2p = device_param->device_local_mem_size - device_param->kernel_local_mem_size2p;
 
           device_param->kernel_preferred_wgs_multiple2p = device_param->hip_warp_size;
         }
@@ -13618,7 +13477,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
           if (get_hip_kernel_local_mem_size (hashcat_ctx, device_param->hip_function2e, &device_param->kernel_local_mem_size2e) == -1) return -1;
 
-          //if (get_hip_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->hip_function2e, &device_param->kernel_dynamic_local_mem_size2e) == -1) return -1;
+          device_param->kernel_dynamic_local_mem_size2e = device_param->device_local_mem_size - device_param->kernel_local_mem_size2e;
 
           device_param->kernel_preferred_wgs_multiple2e = device_param->hip_warp_size;
         }
@@ -13635,7 +13494,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
           if (get_hip_kernel_local_mem_size (hashcat_ctx, device_param->hip_function12, &device_param->kernel_local_mem_size12) == -1) return -1;
 
-          //if (get_hip_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->hip_function12, &device_param->kernel_dynamic_local_mem_size12) == -1) return -1;
+          device_param->kernel_dynamic_local_mem_size12 = device_param->device_local_mem_size - device_param->kernel_local_mem_size12;
 
           device_param->kernel_preferred_wgs_multiple12 = device_param->hip_warp_size;
         }
@@ -13652,7 +13511,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
           if (get_hip_kernel_local_mem_size (hashcat_ctx, device_param->hip_function23, &device_param->kernel_local_mem_size23) == -1) return -1;
 
-          //if (get_hip_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->hip_function23, &device_param->kernel_dynamic_local_mem_size23) == -1) return -1;
+          device_param->kernel_dynamic_local_mem_size23 = device_param->device_local_mem_size - device_param->kernel_local_mem_size23;
 
           device_param->kernel_preferred_wgs_multiple23 = device_param->hip_warp_size;
         }
@@ -13669,7 +13528,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
           if (get_hip_kernel_local_mem_size (hashcat_ctx, device_param->hip_function_init2, &device_param->kernel_local_mem_size_init2) == -1) return -1;
 
-          //if (get_hip_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->hip_function_init2, &device_param->kernel_dynamic_local_mem_size_init2) == -1) return -1;
+          device_param->kernel_dynamic_local_mem_size_init2 = device_param->device_local_mem_size - device_param->kernel_local_mem_size_init2;
 
           device_param->kernel_preferred_wgs_multiple_init2 = device_param->hip_warp_size;
         }
@@ -13686,7 +13545,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
           if (get_hip_kernel_local_mem_size (hashcat_ctx, device_param->hip_function_loop2p, &device_param->kernel_local_mem_size_loop2p) == -1) return -1;
 
-          //if (get_hip_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->hip_function_loop2p, &device_param->kernel_dynamic_local_mem_size_loop2p) == -1) return -1;
+          device_param->kernel_dynamic_local_mem_size_loop2p = device_param->device_local_mem_size - device_param->kernel_local_mem_size_loop2p;
 
           device_param->kernel_preferred_wgs_multiple_loop2p = device_param->hip_warp_size;
         }
@@ -13703,7 +13562,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
           if (get_hip_kernel_local_mem_size (hashcat_ctx, device_param->hip_function_loop2, &device_param->kernel_local_mem_size_loop2) == -1) return -1;
 
-          //if (get_hip_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->hip_function_loop2, &device_param->kernel_dynamic_local_mem_size_loop2) == -1) return -1;
+          device_param->kernel_dynamic_local_mem_size_loop2 = device_param->device_local_mem_size - device_param->kernel_local_mem_size_loop2;
 
           device_param->kernel_preferred_wgs_multiple_loop2 = device_param->hip_warp_size;
         }
@@ -13720,7 +13579,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
           if (get_hip_kernel_local_mem_size (hashcat_ctx, device_param->hip_function_aux1, &device_param->kernel_local_mem_size_aux1) == -1) return -1;
 
-          //if (get_hip_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->hip_function_aux1, &device_param->kernel_dynamic_local_mem_size_aux1) == -1) return -1;
+          device_param->kernel_dynamic_local_mem_size_aux1 = device_param->device_local_mem_size - device_param->kernel_local_mem_size_aux1;
 
           device_param->kernel_preferred_wgs_multiple_aux1 = device_param->hip_warp_size;
         }
@@ -13737,7 +13596,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
           if (get_hip_kernel_local_mem_size (hashcat_ctx, device_param->hip_function_aux2, &device_param->kernel_local_mem_size_aux2) == -1) return -1;
 
-          //if (get_hip_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->hip_function_aux2, &device_param->kernel_dynamic_local_mem_size_aux2) == -1) return -1;
+          device_param->kernel_dynamic_local_mem_size_aux2 = device_param->device_local_mem_size - device_param->kernel_local_mem_size_aux2;
 
           device_param->kernel_preferred_wgs_multiple_aux2 = device_param->hip_warp_size;
         }
@@ -13754,7 +13613,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
           if (get_hip_kernel_local_mem_size (hashcat_ctx, device_param->hip_function_aux3, &device_param->kernel_local_mem_size_aux3) == -1) return -1;
 
-          //if (get_hip_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->hip_function_aux3, &device_param->kernel_dynamic_local_mem_size_aux3) == -1) return -1;
+          device_param->kernel_dynamic_local_mem_size_aux3 = device_param->device_local_mem_size - device_param->kernel_local_mem_size_aux3;
 
           device_param->kernel_preferred_wgs_multiple_aux3 = device_param->hip_warp_size;
         }
@@ -13771,7 +13630,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
           if (get_hip_kernel_local_mem_size (hashcat_ctx, device_param->hip_function_aux4, &device_param->kernel_local_mem_size_aux4) == -1) return -1;
 
-          //if (get_hip_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->hip_function_aux4, &device_param->kernel_dynamic_local_mem_size_aux4) == -1) return -1;
+          device_param->kernel_dynamic_local_mem_size_aux4 = device_param->device_local_mem_size - device_param->kernel_local_mem_size_aux4;
 
           device_param->kernel_preferred_wgs_multiple_aux4 = device_param->hip_warp_size;
         }
@@ -13799,7 +13658,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
           if (get_hip_kernel_local_mem_size (hashcat_ctx, device_param->hip_function_mp_l, &device_param->kernel_local_mem_size_mp_l) == -1) return -1;
 
-          //if (get_hip_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->hip_function_mp_l, &device_param->kernel_dynamic_local_mem_size_mp_l) == -1) return -1;
+          device_param->kernel_dynamic_local_mem_size_mp_l = device_param->device_local_mem_size - device_param->kernel_local_mem_size_mp_l;
 
           device_param->kernel_preferred_wgs_multiple_mp_l = device_param->hip_warp_size;
 
@@ -13811,7 +13670,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
           if (get_hip_kernel_local_mem_size (hashcat_ctx, device_param->hip_function_mp_r, &device_param->kernel_local_mem_size_mp_r) == -1) return -1;
 
-          //if (get_hip_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->hip_function_mp_r, &device_param->kernel_dynamic_local_mem_size_mp_r) == -1) return -1;
+          device_param->kernel_dynamic_local_mem_size_mp_r = device_param->device_local_mem_size - device_param->kernel_local_mem_size_mp_r;
 
           device_param->kernel_preferred_wgs_multiple_mp_r = device_param->hip_warp_size;
 
@@ -13832,7 +13691,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
           if (get_hip_kernel_local_mem_size (hashcat_ctx, device_param->hip_function_mp, &device_param->kernel_local_mem_size_mp) == -1) return -1;
 
-          //if (get_hip_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->hip_function_mp, &device_param->kernel_dynamic_local_mem_size_mp) == -1) return -1;
+          device_param->kernel_dynamic_local_mem_size_mp = device_param->device_local_mem_size - device_param->kernel_local_mem_size_mp;
 
           device_param->kernel_preferred_wgs_multiple_mp = device_param->hip_warp_size;
         }
@@ -13844,7 +13703,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
           if (get_hip_kernel_local_mem_size (hashcat_ctx, device_param->hip_function_mp, &device_param->kernel_local_mem_size_mp) == -1) return -1;
 
-          //if (get_hip_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->hip_function_mp, &device_param->kernel_dynamic_local_mem_size_mp) == -1) return -1;
+          device_param->kernel_dynamic_local_mem_size_mp = device_param->device_local_mem_size - device_param->kernel_local_mem_size_mp;
 
           device_param->kernel_preferred_wgs_multiple_mp = device_param->hip_warp_size;
         }
@@ -13867,7 +13726,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
           if (get_hip_kernel_local_mem_size (hashcat_ctx, device_param->hip_function_amp, &device_param->kernel_local_mem_size_amp) == -1) return -1;
 
-          //if (get_hip_kernel_dynamic_local_mem_size (hashcat_ctx, device_param->hip_function_amp, &device_param->kernel_dynamic_local_mem_size_amp) == -1) return -1;
+          device_param->kernel_dynamic_local_mem_size_amp = device_param->device_local_mem_size - device_param->kernel_local_mem_size_amp;
 
           device_param->kernel_preferred_wgs_multiple_amp = device_param->hip_warp_size;
         }
@@ -13996,12 +13855,6 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
     if (device_param->is_opencl == true)
     {
-      // GPU memset
-
-      if (hc_clSetKernelArg (hashcat_ctx, device_param->opencl_kernel_memset, 0, sizeof (cl_mem),   device_param->kernel_params_memset[0]) == -1) return -1;
-      if (hc_clSetKernelArg (hashcat_ctx, device_param->opencl_kernel_memset, 1, sizeof (cl_uint),  device_param->kernel_params_memset[1]) == -1) return -1;
-      if (hc_clSetKernelArg (hashcat_ctx, device_param->opencl_kernel_memset, 2, sizeof (cl_ulong), device_param->kernel_params_memset[2]) == -1) return -1;
-
       // GPU autotune init
 
       if (hc_clSetKernelArg (hashcat_ctx, device_param->opencl_kernel_atinit, 0, sizeof (cl_mem),   device_param->kernel_params_atinit[0]) == -1) return -1;
@@ -14625,19 +14478,11 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
       }
     }
 
-    // we
-
-    if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD_USE_HIP)
-    {
-      device_param->kernel_threads_min = MIN (device_param->kernel_threads_min, 64);
-      device_param->kernel_threads_max = MIN (device_param->kernel_threads_max, 64);
-    }
-
     /**
      * now everything that depends on threads and accel, basically dynamic workload
      */
 
-    u32 kernel_threads = get_kernel_threads (device_param);
+    //    u32 kernel_threads = get_kernel_threads (device_param);
 
     if (user_options->attack_mode == ATTACK_MODE_ASSOCIATION)
     {
@@ -14645,22 +14490,51 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
       // in autotune. in this attack mode kernel_power is limited by salts_cnt so we
       // do not have a lot of options left.
 
-      kernel_threads = MIN (kernel_threads, 64);
+      device_param->kernel_threads_min = MIN (device_param->kernel_threads_min, 64);
+      device_param->kernel_threads_max = MIN (device_param->kernel_threads_max, 64);
     }
 
-    device_param->kernel_threads = kernel_threads;
+    //    device_param->kernel_threads = kernel_threads;
+    device_param->kernel_threads = 0;
 
-    device_param->hardware_power = ((hashconfig->opts_type & OPTS_TYPE_MP_MULTI_DISABLE) ? 1 : device_processors) * kernel_threads;
+    device_param->hardware_power = ((hashconfig->opts_type & OPTS_TYPE_MP_MULTI_DISABLE) ? 1 : device_processors) * device_param->kernel_threads_max;
 
     u32 kernel_accel_min = device_param->kernel_accel_min;
     u32 kernel_accel_max = device_param->kernel_accel_max;
 
-    /**
-     * We need a kernel accel limiter otherwise we will allocate too much memory (Example 4* GTX1080):
-     * 4 (gpus) * 260 (sizeof pw_t) * 3 (pws, pws_comp, pw_pre) * 20 (MCU) * 1024 (threads) * 1024 (accel) = 65,431,142,400 bytes RAM!!
-     */
+    // We need to deal with the situation that the total video RAM > total host RAM.
+    // For the opposite direction, we do that in the loop section below.
+    // Especially in multi-GPU setups that is very likely.
+    // The buffers which actually take a lot of memory (except for SCRYPT) are the ones for the password candidates.
+    // They are stored in an aligned order for better performance, but this increases the memory pressure.
+    // The best way to keep these buffers to a reasonable size is by controlling the kernel_accel parameter.
+    //
+    // In theory this check could be disabled if we check if total video RAM < total host RAM,
+    // but at this point of initialization phase we don't have this information available.
 
-    const u32 accel_limit = CEILDIV ((64 * 1024), kernel_threads); // this should result in less than 4GB per GPU, but allow higher accel in case user reduces the threads manually using -T
+    // We need to hard-code some value, let's assume that (in 2021) the host has at least 8GB ram per active GPU
+
+    const u64 SIZE_8GB = 8ULL * 1024 * 1024 * 1024;
+
+    u64 accel_limit = SIZE_8GB;
+
+    // this is device_processors * kernel_threads
+
+    accel_limit /= device_param->hardware_power;
+
+    // single password candidate size
+
+    accel_limit /= sizeof (pw_t);
+
+    // pws[], pws_comp[] and pw_pre[] are some large blocks with password candidates
+
+    accel_limit /= 3;
+
+    // Is possible that the GPU simply has too much hardware resources and 8GB per GPU is not enough, but OTOH we can't get lower than 1
+
+    accel_limit = MAX (accel_limit, 1);
+
+    // I think vector size is not required because vector_size is dividing the pws_cnt in run_kernel()
 
     kernel_accel_max = MIN (kernel_accel_max, accel_limit);
 
@@ -14672,7 +14546,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
       return -1;
     }
 
-    // find out if we would request too much memory on memory blocks which are based on kernel_accel
+    // Opposite direction check: find out if we would request too much memory on memory blocks which are based on kernel_accel
 
     u64 size_pws      = 4;
     u64 size_pws_amp  = 4;
@@ -14687,12 +14561,6 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
     u64 size_brain_link_out = 4;
     #endif
 
-    // instead of a thread limit we can also use a memory limit.
-    // this value should represent a reasonable amount of memory a host system has per GPU.
-    // note we're allocating 3 blocks of that size.
-
-    const u64 PWS_SPACE = 1024ULL * 1024ULL * 1024ULL;
-
     while (kernel_accel_max >= kernel_accel_min)
     {
       const u64 kernel_power_max = device_param->hardware_power * kernel_accel_max;
@@ -14742,8 +14610,6 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
       int memory_limit_hit = 0;
 
-      if (size_pws > PWS_SPACE) memory_limit_hit = 1;
-
       // sometimes device_available_mem and device_maxmem_alloc reported back from the opencl runtime are a bit inaccurate.
       // let's add some extra space just to be sure.
       // now depends on the kernel-accel value (where scrypt and similar benefits), but also hard minimum 64mb and maximum 1024mb limit
@@ -14757,6 +14623,47 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
       if ((size_tmps  + EXTRA_SPACE) > device_param->device_maxmem_alloc) memory_limit_hit = 1;
       if ((size_hooks + EXTRA_SPACE) > device_param->device_maxmem_alloc) memory_limit_hit = 1;
 
+      // work around, for some reason apple opencl can't have buffers larger 2^31
+      // typically runs into trap 6
+      // maybe 32/64 bit problem affecting size_t?
+
+      if (device_param->opencl_platform_vendor_id == VENDOR_ID_APPLE)
+      {
+        const size_t undocumented_single_allocation_apple = 0x7fffffff;
+
+        if (bitmap_ctx->bitmap_size > undocumented_single_allocation_apple) memory_limit_hit = 1;
+        if (bitmap_ctx->bitmap_size > undocumented_single_allocation_apple) memory_limit_hit = 1;
+        if (bitmap_ctx->bitmap_size > undocumented_single_allocation_apple) memory_limit_hit = 1;
+        if (bitmap_ctx->bitmap_size > undocumented_single_allocation_apple) memory_limit_hit = 1;
+        if (bitmap_ctx->bitmap_size > undocumented_single_allocation_apple) memory_limit_hit = 1;
+        if (bitmap_ctx->bitmap_size > undocumented_single_allocation_apple) memory_limit_hit = 1;
+        if (bitmap_ctx->bitmap_size > undocumented_single_allocation_apple) memory_limit_hit = 1;
+        if (bitmap_ctx->bitmap_size > undocumented_single_allocation_apple) memory_limit_hit = 1;
+        if (size_bfs                > undocumented_single_allocation_apple) memory_limit_hit = 1;
+        if (size_combs              > undocumented_single_allocation_apple) memory_limit_hit = 1;
+        if (size_digests            > undocumented_single_allocation_apple) memory_limit_hit = 1;
+        if (size_esalts             > undocumented_single_allocation_apple) memory_limit_hit = 1;
+        if (size_hooks              > undocumented_single_allocation_apple) memory_limit_hit = 1;
+        if (size_markov_css         > undocumented_single_allocation_apple) memory_limit_hit = 1;
+        if (size_plains             > undocumented_single_allocation_apple) memory_limit_hit = 1;
+        if (size_pws                > undocumented_single_allocation_apple) memory_limit_hit = 1;
+        if (size_pws_amp            > undocumented_single_allocation_apple) memory_limit_hit = 1;
+        if (size_pws_comp           > undocumented_single_allocation_apple) memory_limit_hit = 1;
+        if (size_pws_idx            > undocumented_single_allocation_apple) memory_limit_hit = 1;
+        if (size_results            > undocumented_single_allocation_apple) memory_limit_hit = 1;
+        if (size_root_css           > undocumented_single_allocation_apple) memory_limit_hit = 1;
+        if (size_rules              > undocumented_single_allocation_apple) memory_limit_hit = 1;
+        if (size_rules_c            > undocumented_single_allocation_apple) memory_limit_hit = 1;
+        if (size_salts              > undocumented_single_allocation_apple) memory_limit_hit = 1;
+        if (size_extra_buffer       > undocumented_single_allocation_apple) memory_limit_hit = 1;
+        if (size_shown              > undocumented_single_allocation_apple) memory_limit_hit = 1;
+        if (size_tm                 > undocumented_single_allocation_apple) memory_limit_hit = 1;
+        if (size_tmps               > undocumented_single_allocation_apple) memory_limit_hit = 1;
+        if (size_st_digests         > undocumented_single_allocation_apple) memory_limit_hit = 1;
+        if (size_st_salts           > undocumented_single_allocation_apple) memory_limit_hit = 1;
+        if (size_st_esalts          > undocumented_single_allocation_apple) memory_limit_hit = 1;
+      }
+
       const u64 size_total
         = bitmap_ctx->bitmap_size
         + bitmap_ctx->bitmap_size
@@ -15225,6 +15132,7 @@ void backend_session_destroy (hashcat_ctx_t *hashcat_ctx)
 
       if (device_param->cuda_event1)           hc_cuEventDestroy (hashcat_ctx, device_param->cuda_event1);
       if (device_param->cuda_event2)           hc_cuEventDestroy (hashcat_ctx, device_param->cuda_event2);
+      if (device_param->cuda_event3)           hc_cuEventDestroy (hashcat_ctx, device_param->cuda_event3);
 
       if (device_param->cuda_stream)           hc_cuStreamDestroy (hashcat_ctx, device_param->cuda_stream);
 
@@ -15289,6 +15197,7 @@ void backend_session_destroy (hashcat_ctx_t *hashcat_ctx)
       device_param->cuda_function_tm          = NULL;
       device_param->cuda_function_amp         = NULL;
       device_param->cuda_function_memset      = NULL;
+      device_param->cuda_function_bzero       = NULL;
       device_param->cuda_function_atinit      = NULL;
       device_param->cuda_function_utf8toutf16le = NULL;
       device_param->cuda_function_decompress  = NULL;
@@ -15299,6 +15208,7 @@ void backend_session_destroy (hashcat_ctx_t *hashcat_ctx)
 
       device_param->cuda_event1               = NULL;
       device_param->cuda_event2               = NULL;
+      device_param->cuda_event3               = NULL;
 
       device_param->cuda_stream               = NULL;
 
@@ -15351,6 +15261,7 @@ void backend_session_destroy (hashcat_ctx_t *hashcat_ctx)
 
       if (device_param->hip_event1)           hc_hipEventDestroy (hashcat_ctx, device_param->hip_event1);
       if (device_param->hip_event2)           hc_hipEventDestroy (hashcat_ctx, device_param->hip_event2);
+      if (device_param->hip_event3)           hc_hipEventDestroy (hashcat_ctx, device_param->hip_event3);
 
       if (device_param->hip_stream)           hc_hipStreamDestroy (hashcat_ctx, device_param->hip_stream);
 
@@ -15415,6 +15326,7 @@ void backend_session_destroy (hashcat_ctx_t *hashcat_ctx)
       device_param->hip_function_tm          = NULL;
       device_param->hip_function_amp         = NULL;
       device_param->hip_function_memset      = NULL;
+      device_param->hip_function_bzero       = NULL;
       device_param->hip_function_atinit      = NULL;
       device_param->hip_function_utf8toutf16le = NULL;
       device_param->hip_function_decompress  = NULL;
@@ -15425,6 +15337,7 @@ void backend_session_destroy (hashcat_ctx_t *hashcat_ctx)
 
       device_param->hip_event1               = NULL;
       device_param->hip_event2               = NULL;
+      device_param->hip_event3               = NULL;
 
       device_param->hip_stream               = NULL;
 
@@ -15492,6 +15405,7 @@ void backend_session_destroy (hashcat_ctx_t *hashcat_ctx)
       if (device_param->opencl_kernel_tm)        hc_clReleaseKernel (hashcat_ctx, device_param->opencl_kernel_tm);
       if (device_param->opencl_kernel_amp)       hc_clReleaseKernel (hashcat_ctx, device_param->opencl_kernel_amp);
       if (device_param->opencl_kernel_memset)    hc_clReleaseKernel (hashcat_ctx, device_param->opencl_kernel_memset);
+      if (device_param->opencl_kernel_bzero)     hc_clReleaseKernel (hashcat_ctx, device_param->opencl_kernel_bzero);
       if (device_param->opencl_kernel_atinit)    hc_clReleaseKernel (hashcat_ctx, device_param->opencl_kernel_atinit);
       if (device_param->opencl_kernel_utf8toutf16le) hc_clReleaseKernel (hashcat_ctx, device_param->opencl_kernel_utf8toutf16le);
       if (device_param->opencl_kernel_decompress)hc_clReleaseKernel (hashcat_ctx, device_param->opencl_kernel_decompress);
@@ -15562,6 +15476,7 @@ void backend_session_destroy (hashcat_ctx_t *hashcat_ctx)
       device_param->opencl_kernel_tm           = NULL;
       device_param->opencl_kernel_amp          = NULL;
       device_param->opencl_kernel_memset       = NULL;
+      device_param->opencl_kernel_bzero        = NULL;
       device_param->opencl_kernel_atinit       = NULL;
       device_param->opencl_kernel_utf8toutf16le = NULL;
       device_param->opencl_kernel_decompress   = NULL;
@@ -15727,8 +15642,8 @@ int backend_session_update_mp (hashcat_ctx_t *hashcat_ctx)
       //for (u32 i = 3; i < 4; i++) { CL_rc = hc_clSetKernelArg (hashcat_ctx, device_param->opencl_kernel_mp, i, sizeof (cl_ulong), device_param->kernel_params_mp[i]); if (CL_rc == -1) return -1; }
       //for (u32 i = 4; i < 8; i++) { CL_rc = hc_clSetKernelArg (hashcat_ctx, device_param->opencl_kernel_mp, i, sizeof (cl_uint),  device_param->kernel_params_mp[i]); if (CL_rc == -1) return -1; }
 
-      if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_root_css_buf,   mask_ctx->root_css_buf,   device_param->size_root_css)   == -1) return -1;
-      if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_markov_css_buf, mask_ctx->markov_css_buf, device_param->size_markov_css) == -1) return -1;
+      if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_root_css_buf,   mask_ctx->root_css_buf,   device_param->size_root_css,   device_param->cuda_stream)   == -1) return -1;
+      if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_markov_css_buf, mask_ctx->markov_css_buf, device_param->size_markov_css, device_param->cuda_stream) == -1) return -1;
     }
 
     if (device_param->is_hip == true)
@@ -15736,8 +15651,8 @@ int backend_session_update_mp (hashcat_ctx_t *hashcat_ctx)
       //for (u32 i = 3; i < 4; i++) { CL_rc = hc_clSetKernelArg (hashcat_ctx, device_param->opencl_kernel_mp, i, sizeof (cl_ulong), device_param->kernel_params_mp[i]); if (CL_rc == -1) return -1; }
       //for (u32 i = 4; i < 8; i++) { CL_rc = hc_clSetKernelArg (hashcat_ctx, device_param->opencl_kernel_mp, i, sizeof (cl_uint),  device_param->kernel_params_mp[i]); if (CL_rc == -1) return -1; }
 
-      if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_root_css_buf,   mask_ctx->root_css_buf,   device_param->size_root_css)   == -1) return -1;
-      if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_markov_css_buf, mask_ctx->markov_css_buf, device_param->size_markov_css) == -1) return -1;
+      if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_root_css_buf,   mask_ctx->root_css_buf,   device_param->size_root_css,   device_param->hip_stream)   == -1) return -1;
+      if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_markov_css_buf, mask_ctx->markov_css_buf, device_param->size_markov_css, device_param->hip_stream) == -1) return -1;
     }
 
     if (device_param->is_opencl == true)
@@ -15745,8 +15660,8 @@ int backend_session_update_mp (hashcat_ctx_t *hashcat_ctx)
       for (u32 i = 3; i < 4; i++) { if (hc_clSetKernelArg (hashcat_ctx, device_param->opencl_kernel_mp, i, sizeof (cl_ulong), device_param->kernel_params_mp[i]) == -1) return -1; }
       for (u32 i = 4; i < 8; i++) { if (hc_clSetKernelArg (hashcat_ctx, device_param->opencl_kernel_mp, i, sizeof (cl_uint),  device_param->kernel_params_mp[i]) == -1) return -1; }
 
-      if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_root_css_buf,   CL_TRUE, 0, device_param->size_root_css,   mask_ctx->root_css_buf,   0, NULL, NULL) == -1) return -1;
-      if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_markov_css_buf, CL_TRUE, 0, device_param->size_markov_css, mask_ctx->markov_css_buf, 0, NULL, NULL) == -1) return -1;
+      if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_root_css_buf,   CL_FALSE, 0, device_param->size_root_css,   mask_ctx->root_css_buf,   0, NULL, NULL) == -1) return -1;
+      if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_markov_css_buf, CL_FALSE, 0, device_param->size_markov_css, mask_ctx->markov_css_buf, 0, NULL, NULL) == -1) return -1;
     }
   }
 
@@ -15788,8 +15703,8 @@ int backend_session_update_mp_rl (hashcat_ctx_t *hashcat_ctx, const u32 css_cnt_
       //for (u32 i = 4; i < 7; i++) { CL_rc = hc_clSetKernelArg (hashcat_ctx, device_param->opencl_kernel_mp_r, i, sizeof (cl_uint),  device_param->kernel_params_mp_r[i]); if (CL_rc == -1) return -1; }
       //for (u32 i = 8; i < 8; i++) { CL_rc = hc_clSetKernelArg (hashcat_ctx, device_param->opencl_kernel_mp_r, i, sizeof (cl_ulong), device_param->kernel_params_mp_r[i]); if (CL_rc == -1) return -1; }
 
-      if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_root_css_buf,   mask_ctx->root_css_buf,   device_param->size_root_css)   == -1) return -1;
-      if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_markov_css_buf, mask_ctx->markov_css_buf, device_param->size_markov_css) == -1) return -1;
+      if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_root_css_buf,   mask_ctx->root_css_buf,   device_param->size_root_css,   device_param->cuda_stream) == -1) return -1;
+      if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_markov_css_buf, mask_ctx->markov_css_buf, device_param->size_markov_css, device_param->cuda_stream) == -1) return -1;
     }
 
     if (device_param->is_hip == true)
@@ -15802,8 +15717,8 @@ int backend_session_update_mp_rl (hashcat_ctx_t *hashcat_ctx, const u32 css_cnt_
       //for (u32 i = 4; i < 7; i++) { CL_rc = hc_clSetKernelArg (hashcat_ctx, device_param->opencl_kernel_mp_r, i, sizeof (cl_uint),  device_param->kernel_params_mp_r[i]); if (CL_rc == -1) return -1; }
       //for (u32 i = 8; i < 8; i++) { CL_rc = hc_clSetKernelArg (hashcat_ctx, device_param->opencl_kernel_mp_r, i, sizeof (cl_ulong), device_param->kernel_params_mp_r[i]); if (CL_rc == -1) return -1; }
 
-      if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_root_css_buf,   mask_ctx->root_css_buf,   device_param->size_root_css)   == -1) return -1;
-      if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_markov_css_buf, mask_ctx->markov_css_buf, device_param->size_markov_css) == -1) return -1;
+      if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_root_css_buf,   mask_ctx->root_css_buf,   device_param->size_root_css,   device_param->hip_stream) == -1) return -1;
+      if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_markov_css_buf, mask_ctx->markov_css_buf, device_param->size_markov_css, device_param->hip_stream) == -1) return -1;
     }
 
     if (device_param->is_opencl == true)
@@ -15816,8 +15731,8 @@ int backend_session_update_mp_rl (hashcat_ctx_t *hashcat_ctx, const u32 css_cnt_
       for (u32 i = 4; i < 7; i++) { if (hc_clSetKernelArg (hashcat_ctx, device_param->opencl_kernel_mp_r, i, sizeof (cl_uint),  device_param->kernel_params_mp_r[i]) == -1) return -1; }
       for (u32 i = 8; i < 8; i++) { if (hc_clSetKernelArg (hashcat_ctx, device_param->opencl_kernel_mp_r, i, sizeof (cl_ulong), device_param->kernel_params_mp_r[i]) == -1) return -1; }
 
-      if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_root_css_buf,   CL_TRUE, 0, device_param->size_root_css,   mask_ctx->root_css_buf,   0, NULL, NULL) == -1) return -1;
-      if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_markov_css_buf, CL_TRUE, 0, device_param->size_markov_css, mask_ctx->markov_css_buf, 0, NULL, NULL) == -1) return -1;
+      if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_root_css_buf,   CL_FALSE, 0, device_param->size_root_css,   mask_ctx->root_css_buf,   0, NULL, NULL) == -1) return -1;
+      if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_markov_css_buf, CL_FALSE, 0, device_param->size_markov_css, mask_ctx->markov_css_buf, 0, NULL, NULL) == -1) return -1;
     }
   }
 
diff --git a/src/brain.c b/src/brain.c
index 64e3578ae..a12a1375b 100644
--- a/src/brain.c
+++ b/src/brain.c
@@ -172,6 +172,14 @@ u32 brain_compute_attack (hashcat_ctx_t *hashcat_ctx)
 
   XXH64_update (state, &hex_salt, sizeof (hex_salt));
 
+  const u32 opti_type = hashconfig->opti_type;
+
+  XXH64_update (state, &opti_type, sizeof (opti_type));
+
+  const u64 opts_type = hashconfig->opts_type;
+
+  XXH64_update (state, &opts_type, sizeof (opts_type));
+
   const int hccapx_message_pair = user_options->hccapx_message_pair;
 
   XXH64_update (state, &hccapx_message_pair, sizeof (hccapx_message_pair));
@@ -658,7 +666,7 @@ u32 brain_auth_challenge (void)
 
   #else
 
-  static const char *urandom = "/dev/urandom";
+  static const char *const urandom = "/dev/urandom";
 
   HCFILE fp;
 
diff --git a/src/ext_ADL.c b/src/ext_ADL.c
index e54420713..9676c916f 100644
--- a/src/ext_ADL.c
+++ b/src/ext_ADL.c
@@ -50,27 +50,26 @@ int adl_init (void *hashcat_ctx)
     return -1;
   }
 
-  HC_LOAD_FUNC(adl, ADL_Main_Control_Destroy, ADL_MAIN_CONTROL_DESTROY, ADL, 0);
-  HC_LOAD_FUNC(adl, ADL_Main_Control_Create, ADL_MAIN_CONTROL_CREATE, ADL, 0);
-  HC_LOAD_FUNC(adl, ADL_Adapter_NumberOfAdapters_Get, ADL_ADAPTER_NUMBEROFADAPTERS_GET, ADL, 0);
+  HC_LOAD_FUNC(adl, ADL_Adapter_Active_Get, ADL_ADAPTER_ACTIVE_GET, ADL, 0);
   HC_LOAD_FUNC(adl, ADL_Adapter_AdapterInfo_Get, ADL_ADAPTER_ADAPTERINFO_GET, ADL, 0);
-  HC_LOAD_FUNC(adl, ADL_Display_DisplayInfo_Get, ADL_DISPLAY_DISPLAYINFO_GET, ADL, 0);
-  HC_LOAD_FUNC(adl, ADL_Adapter_ID_Get, ADL_ADAPTER_ID_GET, ADL, 0);
-  HC_LOAD_FUNC(adl, ADL_Adapter_VideoBiosInfo_Get, ADL_ADAPTER_VIDEOBIOSINFO_GET, ADL, 0);
-  HC_LOAD_FUNC(adl, ADL_Overdrive5_ThermalDevices_Enum, ADL_OVERDRIVE5_THERMALDEVICES_ENUM, ADL, 0);
-  HC_LOAD_FUNC(adl, ADL_Overdrive5_Temperature_Get, ADL_OVERDRIVE5_TEMPERATURE_GET, ADL, 0);
-  HC_LOAD_FUNC(adl, ADL_Overdrive6_Temperature_Get, ADL_OVERDRIVE6_TEMPERATURE_GET, ADL, 0);
+  HC_LOAD_FUNC(adl, ADL_Adapter_NumberOfAdapters_Get, ADL_ADAPTER_NUMBEROFADAPTERS_GET, ADL, 0);
+  HC_LOAD_FUNC(adl, ADL_Main_Control_Create, ADL_MAIN_CONTROL_CREATE, ADL, 0);
+  HC_LOAD_FUNC(adl, ADL_Main_Control_Destroy, ADL_MAIN_CONTROL_DESTROY, ADL, 0);
   HC_LOAD_FUNC(adl, ADL_Overdrive5_CurrentActivity_Get, ADL_OVERDRIVE5_CURRENTACTIVITY_GET, ADL, 0);
   HC_LOAD_FUNC(adl, ADL_Overdrive5_FanSpeedInfo_Get, ADL_OVERDRIVE5_FANSPEEDINFO_GET, ADL, 0);
   HC_LOAD_FUNC(adl, ADL_Overdrive5_FanSpeed_Get, ADL_OVERDRIVE5_FANSPEED_GET, ADL, 0);
-  HC_LOAD_FUNC(adl, ADL_Overdrive6_FanSpeed_Get, ADL_OVERDRIVE6_FANSPEED_GET, ADL, 0);
-  HC_LOAD_FUNC(adl, ADL_Adapter_Active_Get, ADL_ADAPTER_ACTIVE_GET, ADL, 0);
-  HC_LOAD_FUNC(adl, ADL_Overdrive_Caps, ADL_OVERDRIVE_CAPS, ADL, 0);
+  HC_LOAD_FUNC(adl, ADL_Overdrive5_ODParameters_Get, ADL_OVERDRIVE5_ODPARAMETERS_GET, ADL, 0);
+  HC_LOAD_FUNC(adl, ADL_Overdrive5_ODPerformanceLevels_Get, ADL_OVERDRIVE5_ODPERFORMANCELEVELS_GET, ADL, 0);
+  HC_LOAD_FUNC(adl, ADL_Overdrive5_Temperature_Get, ADL_OVERDRIVE5_TEMPERATURE_GET, ADL, 0);
+  HC_LOAD_FUNC(adl, ADL_Overdrive5_ThermalDevices_Enum, ADL_OVERDRIVE5_THERMALDEVICES_ENUM, ADL, 0);
   HC_LOAD_FUNC(adl, ADL_Overdrive6_Capabilities_Get, ADL_OVERDRIVE6_CAPABILITIES_GET, ADL, 0);
-  HC_LOAD_FUNC(adl, ADL_Overdrive6_StateInfo_Get, ADL_OVERDRIVE6_STATEINFO_GET, ADL, 0);
   HC_LOAD_FUNC(adl, ADL_Overdrive6_CurrentStatus_Get, ADL_OVERDRIVE6_CURRENTSTATUS_GET, ADL, 0);
-  HC_LOAD_FUNC(adl, ADL_Overdrive6_TargetTemperatureData_Get, ADL_OVERDRIVE6_TARGETTEMPERATUREDATA_GET, ADL, 0);
-  HC_LOAD_FUNC(adl, ADL_Overdrive6_TargetTemperatureRangeInfo_Get, ADL_OVERDRIVE6_TARGETTEMPERATURERANGEINFO_GET, ADL, 0);
+  HC_LOAD_FUNC(adl, ADL_Overdrive6_FanSpeed_Get, ADL_OVERDRIVE6_FANSPEED_GET, ADL, 0);
+  HC_LOAD_FUNC(adl, ADL_Overdrive6_StateInfo_Get, ADL_OVERDRIVE6_STATEINFO_GET, ADL, 0);
+  HC_LOAD_FUNC(adl, ADL_Overdrive6_Temperature_Get, ADL_OVERDRIVE6_TEMPERATURE_GET, ADL, 0);
+  HC_LOAD_FUNC(adl, ADL_Overdrive_Caps, ADL_OVERDRIVE_CAPS, ADL, 0);
+  HC_LOAD_FUNC(adl, ADL2_Overdrive_Caps, ADL2_OVERDRIVE_CAPS, ADL, 1);
+  HC_LOAD_FUNC(adl, ADL2_New_QueryPMLogData_Get, ADL2_NEW_QUERYPMLOGDATA_GET, ADL, 1);
 
   return 0;
 }
@@ -270,17 +269,42 @@ int hm_ADL_Overdrive_Caps (void *hashcat_ctx, int iAdapterIndex, int *od_support
   return 0;
 }
 
-int hm_ADL_Overdrive6_TargetTemperatureData_Get (void *hashcat_ctx, int iAdapterIndex, int *cur_temp, int *default_temp)
+int hm_ADL2_Overdrive_Caps (void *hashcat_ctx, int iAdapterIndex, int *iSupported, int *iEnabled, int *iVersion)
 {
   hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
 
   ADL_PTR *adl = (ADL_PTR *) hwmon_ctx->hm_adl;
 
-  const int ADL_rc = adl->ADL_Overdrive6_TargetTemperatureData_Get (iAdapterIndex, cur_temp, default_temp);
+  // Not sure if that makes any sense...
+
+  if (adl->ADL2_Overdrive_Caps == NULL)
+  {
+    return hm_ADL_Overdrive_Caps (hashcat_ctx, iAdapterIndex, iSupported, iEnabled, iVersion);
+  }
+
+  const int ADL_rc = adl->ADL2_Overdrive_Caps (NULL, iAdapterIndex, iSupported, iEnabled, iVersion);
 
   if (ADL_rc != ADL_OK)
   {
-    event_log_error (hashcat_ctx, "ADL_Overdrive6_TargetTemperatureData_Get(): %d", ADL_rc);
+    event_log_error (hashcat_ctx, "ADL2_Overdrive_Caps(): %d", ADL_rc);
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hm_ADL2_New_QueryPMLogData_Get (void *hashcat_ctx, int iAdapterIndex, ADLPMLogDataOutput *lpDataOutput)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  ADL_PTR *adl = (ADL_PTR *) hwmon_ctx->hm_adl;
+
+  const int ADL_rc = adl->ADL2_New_QueryPMLogData_Get (NULL, iAdapterIndex, lpDataOutput);
+
+  if (ADL_rc != ADL_OK)
+  {
+    event_log_error (hashcat_ctx, "ADL2_New_QueryPMLogData_Get(): %d", ADL_rc);
 
     return -1;
   }
diff --git a/src/filehandling.c b/src/filehandling.c
index 54ba73369..9edd1cb35 100644
--- a/src/filehandling.c
+++ b/src/filehandling.c
@@ -21,7 +21,7 @@ int _wopen (const char *path, int oflag, ...)
 }
 #endif
 
-bool hc_fopen (HCFILE *fp, const char *path, char *mode)
+bool hc_fopen (HCFILE *fp, const char *path, const char *mode)
 {
   if (path == NULL || mode == NULL) return false;
 
@@ -130,7 +130,7 @@ bool hc_fopen (HCFILE *fp, const char *path, char *mode)
   return true;
 }
 
-bool hc_fopen_raw (HCFILE *fp, const char *path, char *mode)
+bool hc_fopen_raw (HCFILE *fp, const char *path, const char *mode)
 {
   if (path == NULL || mode == NULL) return false;
 
diff --git a/src/hashcat.c b/src/hashcat.c
index e4ac1a549..1beaf9a16 100644
--- a/src/hashcat.c
+++ b/src/hashcat.c
@@ -430,15 +430,17 @@ static int inner1_loop (hashcat_ctx_t *hashcat_ctx)
 
 static int outer_loop (hashcat_ctx_t *hashcat_ctx)
 {
-  hashconfig_t   *hashconfig    = hashcat_ctx->hashconfig;
-  hashes_t       *hashes        = hashcat_ctx->hashes;
-  mask_ctx_t     *mask_ctx      = hashcat_ctx->mask_ctx;
-  backend_ctx_t  *backend_ctx   = hashcat_ctx->backend_ctx;
-  outcheck_ctx_t *outcheck_ctx  = hashcat_ctx->outcheck_ctx;
-  restore_ctx_t  *restore_ctx   = hashcat_ctx->restore_ctx;
-  status_ctx_t   *status_ctx    = hashcat_ctx->status_ctx;
-  straight_ctx_t *straight_ctx  = hashcat_ctx->straight_ctx;
-  user_options_t *user_options  = hashcat_ctx->user_options;
+  hashconfig_t         *hashconfig          = hashcat_ctx->hashconfig;
+  hashes_t             *hashes              = hashcat_ctx->hashes;
+  mask_ctx_t           *mask_ctx            = hashcat_ctx->mask_ctx;
+  module_ctx_t         *module_ctx          = hashcat_ctx->module_ctx;
+  backend_ctx_t        *backend_ctx         = hashcat_ctx->backend_ctx;
+  outcheck_ctx_t       *outcheck_ctx        = hashcat_ctx->outcheck_ctx;
+  restore_ctx_t        *restore_ctx         = hashcat_ctx->restore_ctx;
+  status_ctx_t         *status_ctx          = hashcat_ctx->status_ctx;
+  straight_ctx_t       *straight_ctx        = hashcat_ctx->straight_ctx;
+  user_options_t       *user_options        = hashcat_ctx->user_options;
+  user_options_extra_t *user_options_extra  = hashcat_ctx->user_options_extra;
 
   status_ctx->devices_status = STATUS_INIT;
 
@@ -463,6 +465,46 @@ static int outer_loop (hashcat_ctx_t *hashcat_ctx)
 
   EVENT (EVENT_HASHCONFIG_POST);
 
+  /**
+   * deprecated notice
+   */
+
+  if (module_ctx->module_deprecated_notice != MODULE_DEFAULT)
+  {
+    if (user_options->deprecated_check_disable == false)
+    {
+      if ((user_options->show == true) || (user_options->left == true))
+      {
+        const char *module_deprecated_notice = module_ctx->module_deprecated_notice (hashconfig, user_options, user_options_extra);
+
+        event_log_warning (hashcat_ctx, "%s", module_deprecated_notice);
+        event_log_warning (hashcat_ctx, NULL);
+      }
+      else if (user_options->benchmark == true)
+      {
+        if (user_options->hash_mode_chgd == true)
+        {
+          const char *module_deprecated_notice = module_ctx->module_deprecated_notice (hashconfig, user_options, user_options_extra);
+
+          event_log_warning (hashcat_ctx, "%s", module_deprecated_notice);
+          event_log_warning (hashcat_ctx, NULL);
+        }
+        else
+        {
+          return 0;
+        }
+      }
+      else
+      {
+        const char *module_deprecated_notice = module_ctx->module_deprecated_notice (hashconfig, user_options, user_options_extra);
+
+        event_log_error (hashcat_ctx, "%s", module_deprecated_notice);
+
+        return 0;
+      }
+    }
+  }
+
   /**
    * generate hashlist filename for later use
    */
diff --git a/src/hashes.c b/src/hashes.c
index 27aa94370..371724e75 100644
--- a/src/hashes.c
+++ b/src/hashes.c
@@ -300,7 +300,7 @@ int save_hash (hashcat_ctx_t *hashcat_ctx)
   return 0;
 }
 
-void check_hash (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, plain_t *plain)
+int check_hash (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, plain_t *plain)
 {
   const debugfile_ctx_t *debugfile_ctx = hashcat_ctx->debugfile_ctx;
   const hashes_t        *hashes        = hashcat_ctx->hashes;
@@ -313,23 +313,63 @@ void check_hash (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, pl
 
   void *tmps = NULL;
 
+  cl_event opencl_event;
+
+  int rc = -1;
+
   if (hashconfig->opts_type & OPTS_TYPE_COPY_TMPS)
   {
     tmps = hcmalloc (hashconfig->tmp_size);
 
     if (device_param->is_cuda == true)
     {
-      hc_cuMemcpyDtoH (hashcat_ctx, tmps, device_param->cuda_d_tmps + (plain->gidvid * hashconfig->tmp_size), hashconfig->tmp_size);
+      rc = hc_cuMemcpyDtoHAsync (hashcat_ctx, tmps, device_param->cuda_d_tmps + (plain->gidvid * hashconfig->tmp_size), hashconfig->tmp_size, device_param->cuda_stream);
+
+      if (rc == 0)
+      {
+        rc = hc_cuEventRecord (hashcat_ctx, device_param->cuda_event3, device_param->cuda_stream);
+      }
+
+      if (rc == -1)
+      {
+        hcfree (tmps);
+
+        return -1;
+      }
     }
 
     if (device_param->is_hip == true)
     {
-      hc_hipMemcpyDtoH (hashcat_ctx, tmps, device_param->hip_d_tmps + (plain->gidvid * hashconfig->tmp_size), hashconfig->tmp_size);
+      rc = hc_hipMemcpyDtoHAsync (hashcat_ctx, tmps, device_param->hip_d_tmps + (plain->gidvid * hashconfig->tmp_size), hashconfig->tmp_size, device_param->hip_stream);
+
+      if (rc == 0)
+      {
+        rc = hc_hipEventRecord (hashcat_ctx, device_param->hip_event3, device_param->hip_stream);
+      }
+
+      if (rc == -1)
+      {
+        hcfree (tmps);
+
+        return -1;
+      }
     }
 
     if (device_param->is_opencl == true)
     {
-      hc_clEnqueueReadBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_tmps, CL_TRUE, plain->gidvid * hashconfig->tmp_size, hashconfig->tmp_size, tmps, 0, NULL, NULL);
+      rc = hc_clEnqueueReadBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_tmps, CL_FALSE, plain->gidvid * hashconfig->tmp_size, hashconfig->tmp_size, tmps, 0, NULL, &opencl_event);
+
+      if (rc == 0)
+      {
+        rc = hc_clFlush (hashcat_ctx, device_param->opencl_command_queue);
+      }
+
+      if (rc == -1)
+      {
+        hcfree (tmps);
+
+        return -1;
+      }
     }
   }
 
@@ -337,15 +377,14 @@ void check_hash (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, pl
 
   u8 *out_buf = hashes->out_buf;
 
-  int out_len = hash_encode (hashcat_ctx->hashconfig, hashcat_ctx->hashes, hashcat_ctx->module_ctx, (char *) out_buf, HCBUFSIZ_LARGE, salt_pos, digest_pos);
+  int out_len = hash_encode (hashconfig, hashes, module_ctx, (char *) out_buf, HCBUFSIZ_LARGE, salt_pos, digest_pos);
 
   out_buf[out_len] = 0;
 
   // plain
 
-  u8 plain_buf[0x1000]; // while the password itself can have only length 256, the module could encode it with something like base64 which inflates the requires buffer size
-
-  memset (plain_buf, 0, sizeof (plain_buf));
+  u8 plain_buf[HCBUFSIZ_TINY] = { 0 }; // while the password itself can have only length 256, the module could encode it with something like base64 which inflates the requires buffer size
+  u8 postprocess_buf[HCBUFSIZ_TINY] = { 0 };
 
   u8 *plain_ptr = plain_buf;
 
@@ -355,18 +394,27 @@ void check_hash (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, pl
 
   if (module_ctx->module_build_plain_postprocess != MODULE_DEFAULT)
   {
-    u8 temp_buf[0x1000];
-
-    memset (temp_buf, 0, sizeof (temp_buf));
-
-    const int temp_len = module_ctx->module_build_plain_postprocess (hashcat_ctx->hashconfig, hashcat_ctx->hashes, tmps, (u32 *) plain_buf, sizeof (plain_buf), plain_len, (u32 *)temp_buf, sizeof (temp_buf));
-
-    if (temp_len < (int) sizeof (plain_buf))
+    if (hashconfig->opts_type & OPTS_TYPE_COPY_TMPS)
     {
-      memcpy (plain_buf, temp_buf, temp_len);
+      if (device_param->is_cuda == true)
+      {
+        if (hc_cuEventSynchronize (hashcat_ctx, device_param->cuda_event3) == -1) return -1;
+      }
 
-      plain_len = temp_len;
+      if (device_param->is_hip == true)
+      {
+        if (hc_hipEventSynchronize (hashcat_ctx, device_param->hip_event3) == -1) return -1;
+      }
+
+      if (device_param->is_opencl == true)
+      {
+        if (hc_clWaitForEvents (hashcat_ctx, 1, &opencl_event) == -1) return -1;
+      }
     }
+
+    plain_len = module_ctx->module_build_plain_postprocess (hashconfig, hashes, tmps, (u32 *) plain_buf, sizeof (plain_buf), plain_len, (u32 *) postprocess_buf, sizeof (postprocess_buf));
+
+    plain_ptr = postprocess_buf;
   }
 
   // crackpos
@@ -407,6 +455,24 @@ void check_hash (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, pl
 
   if (module_ctx->module_hash_encode_potfile != MODULE_DEFAULT)
   {
+    if (hashconfig->opts_type & OPTS_TYPE_COPY_TMPS)
+    {
+      if (device_param->is_cuda == true)
+      {
+        if (hc_cuEventSynchronize (hashcat_ctx, device_param->cuda_event3) == -1) return -1;
+      }
+
+      if (device_param->is_hip == true)
+      {
+        if (hc_hipEventSynchronize (hashcat_ctx, device_param->hip_event3) == -1) return -1;
+      }
+
+      if (device_param->is_opencl == true)
+      {
+        if (hc_clWaitForEvents (hashcat_ctx, 1, &opencl_event) == -1) return -1;
+      }
+    }
+
     salt_t *salts_buf = hashes->salts_buf;
 
     salts_buf += salt_pos;
@@ -471,7 +537,14 @@ void check_hash (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, pl
   if (hashconfig->opts_type & OPTS_TYPE_COPY_TMPS)
   {
     hcfree (tmps);
+
+    if (device_param->is_opencl == true)
+    {
+      if (hc_clReleaseEvent (hashcat_ctx, opencl_event) == -1) return -1;
+    }
   }
+
+  return 0;
 }
 
 //int check_cracked (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, const u32 salt_pos)
@@ -485,172 +558,204 @@ int check_cracked (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param)
 
   u32 num_cracked = 0;
 
-  int CU_rc;
-  int HIP_rc;
-  int CL_rc;
+  int rc = -1;
 
   if (device_param->is_cuda == true)
   {
-    CU_rc = hc_cuMemcpyDtoH (hashcat_ctx, &num_cracked, device_param->cuda_d_result, sizeof (u32));
+    if (hc_cuMemcpyDtoHAsync (hashcat_ctx, &num_cracked, device_param->cuda_d_result, sizeof (u32), device_param->cuda_stream) == -1) return -1;
 
-    if (CU_rc == -1) return -1;
+    if (hc_cuStreamSynchronize (hashcat_ctx, device_param->cuda_stream) == -1) return -1;
   }
 
   if (device_param->is_hip == true)
   {
-    HIP_rc = hc_hipMemcpyDtoH (hashcat_ctx, &num_cracked, device_param->hip_d_result, sizeof (u32));
+    if (hc_hipMemcpyDtoHAsync (hashcat_ctx, &num_cracked, device_param->hip_d_result, sizeof (u32), device_param->hip_stream) == -1) return -1;
 
-    if (HIP_rc == -1) return -1;
+    if (hc_hipStreamSynchronize (hashcat_ctx, device_param->hip_stream) == -1) return -1;
   }
 
   if (device_param->is_opencl == true)
   {
-    CL_rc = hc_clEnqueueReadBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_result, CL_TRUE, 0, sizeof (u32), &num_cracked, 0, NULL, NULL);
-
-    if (CL_rc == -1) return -1;
+    /* blocking */
+    if (hc_clEnqueueReadBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_result, CL_TRUE, 0, sizeof (u32), &num_cracked, 0, NULL, NULL) == -1) return -1;
   }
 
-  if (user_options->speed_only == true)
+  if (num_cracked == 0 || user_options->speed_only == true)
   {
-    // we want the hc_clEnqueueReadBuffer to run in benchmark mode because it has an influence in performance
+    // we want to get the num_cracked in benchmark mode because it has an influence in performance
     // however if the benchmark cracks the artificial hash used for benchmarks we don't want to see that!
 
     return 0;
   }
 
-  if (num_cracked)
+  plain_t *cracked = (plain_t *) hcmalloc (num_cracked * sizeof (plain_t));
+
+  if (device_param->is_cuda == true)
   {
-    plain_t *cracked = (plain_t *) hccalloc (num_cracked, sizeof (plain_t));
+    rc = hc_cuMemcpyDtoHAsync (hashcat_ctx, cracked, device_param->cuda_d_plain_bufs, num_cracked * sizeof (plain_t), device_param->cuda_stream);
 
-    if (device_param->is_cuda == true)
+    if (rc == 0)
     {
-      CU_rc = hc_cuMemcpyDtoH (hashcat_ctx, cracked, device_param->cuda_d_plain_bufs, num_cracked * sizeof (plain_t));
-
-      if (CU_rc == -1) return -1;
+      rc = hc_cuStreamSynchronize (hashcat_ctx, device_param->cuda_stream);
     }
 
-    if (device_param->is_hip == true)
+    if (rc == -1)
     {
-      HIP_rc = hc_hipMemcpyDtoH (hashcat_ctx, cracked, device_param->hip_d_plain_bufs, num_cracked * sizeof (plain_t));
+      hcfree (cracked);
 
-      if (HIP_rc == -1) return -1;
+      return -1;
+    }
+  }
+
+  if (device_param->is_hip == true)
+  {
+    rc = hc_hipMemcpyDtoHAsync (hashcat_ctx, cracked, device_param->hip_d_plain_bufs, num_cracked * sizeof (plain_t), device_param->hip_stream);
+
+    if (rc == 0)
+    {
+      rc = hc_hipStreamSynchronize (hashcat_ctx, device_param->hip_stream);
     }
 
-    if (device_param->is_opencl == true)
+    if (rc == -1)
     {
-      CL_rc = hc_clEnqueueReadBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_plain_bufs, CL_TRUE, 0, num_cracked * sizeof (plain_t), cracked, 0, NULL, NULL);
+      hcfree (cracked);
 
-      if (CL_rc == -1) return -1;
+      return -1;
+    }
+  }
+
+  if (device_param->is_opencl == true)
+  {
+    /* blocking */
+    rc = hc_clEnqueueReadBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_plain_bufs, CL_TRUE, 0, num_cracked * sizeof (plain_t), cracked, 0, NULL, NULL);
+
+    if (rc == -1)
+    {
+      hcfree (cracked);
+
+      return -1;
+    }
+  }
+
+  u32 cpt_cracked = 0;
+
+  hc_thread_mutex_lock (status_ctx->mux_display);
+
+  for (u32 i = 0; i < num_cracked; i++)
+  {
+    const u32 hash_pos = cracked[i].hash_pos;
+
+    if (hashes->digests_shown[hash_pos] == 1) continue;
+
+    const u32 salt_pos = cracked[i].salt_pos;
+    salt_t *salt_buf = &hashes->salts_buf[salt_pos];
+
+    if ((hashconfig->opts_type & OPTS_TYPE_PT_NEVERCRACK) == 0)
+    {
+      hashes->digests_shown[hash_pos] = 1;
+
+      hashes->digests_done++;
+
+      cpt_cracked++;
+
+      salt_buf->digests_done++;
+
+      if (salt_buf->digests_done == salt_buf->digests_cnt)
+      {
+        hashes->salts_shown[salt_pos] = 1;
+
+        hashes->salts_done++;
+      }
     }
 
-    u32 cpt_cracked = 0;
+    if (hashes->salts_done == hashes->salts_cnt) mycracked (hashcat_ctx);
 
+    rc = check_hash (hashcat_ctx, device_param, &cracked[i]);
+
+    if (rc == -1)
+    {
+      break;
+    }
+
+    if (hashconfig->opts_type & OPTS_TYPE_PT_NEVERCRACK)
+    {
+      // we need to reset cracked state on the device
+      // otherwise host thinks again and again the hash was cracked
+      // and returns invalid password each time
+
+      if (device_param->is_cuda == true)
+      {
+        rc = run_cuda_kernel_bzero (hashcat_ctx, device_param, device_param->cuda_d_digests_shown + (salt_buf->digests_offset * sizeof (u32)), salt_buf->digests_cnt * sizeof (u32));
+
+        if (rc == -1)
+        {
+          break;
+        }
+      }
+
+      if (device_param->is_hip == true)
+      {
+        rc = run_hip_kernel_bzero (hashcat_ctx, device_param, device_param->hip_d_digests_shown + (salt_buf->digests_offset * sizeof (u32)), salt_buf->digests_cnt * sizeof (u32));
+
+        if (rc == -1)
+        {
+          break;
+        }
+      }
+
+      if (device_param->is_opencl == true)
+      {
+        /* NOTE: run_opencl_kernel_bzero() does not handle buffer offset */
+        rc = run_opencl_kernel_memset32 (hashcat_ctx, device_param, device_param->opencl_d_digests_shown, salt_buf->digests_offset * sizeof (u32), 0, salt_buf->digests_cnt * sizeof (u32));
+
+        if (rc == -1)
+        {
+          break;
+        }
+      }
+    }
+  }
+
+  hc_thread_mutex_unlock (status_ctx->mux_display);
+
+  hcfree (cracked);
+
+  if (rc == -1)
+  {
+    return -1;
+  }
+
+  if (cpt_cracked > 0)
+  {
     hc_thread_mutex_lock (status_ctx->mux_display);
 
-    for (u32 i = 0; i < num_cracked; i++)
-    {
-      const u32 hash_pos = cracked[i].hash_pos;
+    cpt_ctx->cpt_buf[cpt_ctx->cpt_pos].timestamp = time (NULL);
+    cpt_ctx->cpt_buf[cpt_ctx->cpt_pos].cracked   = cpt_cracked;
 
-      if (hashes->digests_shown[hash_pos] == 1) continue;
+    cpt_ctx->cpt_pos++;
 
-      const u32 salt_pos = cracked[i].salt_pos;
-      salt_t *salt_buf = &hashes->salts_buf[salt_pos];
+    cpt_ctx->cpt_total += cpt_cracked;
 
-      if ((hashconfig->opts_type & OPTS_TYPE_PT_NEVERCRACK) == 0)
-      {
-        hashes->digests_shown[hash_pos] = 1;
-
-        hashes->digests_done++;
-
-        cpt_cracked++;
-
-        salt_buf->digests_done++;
-
-        if (salt_buf->digests_done == salt_buf->digests_cnt)
-        {
-          hashes->salts_shown[salt_pos] = 1;
-
-          hashes->salts_done++;
-        }
-      }
-
-      if (hashes->salts_done == hashes->salts_cnt) mycracked (hashcat_ctx);
-
-      check_hash (hashcat_ctx, device_param, &cracked[i]);
-
-      if (hashconfig->opts_type & OPTS_TYPE_PT_NEVERCRACK)
-      {
-        // we need to reset cracked state on the device
-        // otherwise host thinks again and again the hash was cracked
-        // and returns invalid password each time
-
-        memset (hashes->digests_shown_tmp, 0, salt_buf->digests_cnt * sizeof (u32));
-
-        if (device_param->is_cuda == true)
-        {
-          CU_rc = hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_digests_shown + (salt_buf->digests_offset * sizeof (u32)), &hashes->digests_shown_tmp[salt_buf->digests_offset], salt_buf->digests_cnt * sizeof (u32));
-
-          if (CU_rc == -1) return -1;
-        }
-
-        if (device_param->is_hip == true)
-        {
-          HIP_rc = hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_digests_shown + (salt_buf->digests_offset * sizeof (u32)), &hashes->digests_shown_tmp[salt_buf->digests_offset], salt_buf->digests_cnt * sizeof (u32));
-
-          if (HIP_rc == -1) return -1;
-        }
-
-        if (device_param->is_opencl == true)
-        {
-          CL_rc = hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_digests_shown, CL_TRUE, salt_buf->digests_offset * sizeof (u32), salt_buf->digests_cnt * sizeof (u32), &hashes->digests_shown_tmp[salt_buf->digests_offset], 0, NULL, NULL);
-
-          if (CL_rc == -1) return -1;
-        }
-      }
-    }
+    if (cpt_ctx->cpt_pos == CPT_CACHE) cpt_ctx->cpt_pos = 0;
 
     hc_thread_mutex_unlock (status_ctx->mux_display);
+  }
 
-    hcfree (cracked);
+  if (device_param->is_cuda == true)
+  {
+    if (run_cuda_kernel_bzero (hashcat_ctx, device_param, device_param->cuda_d_result, sizeof (u32)) == -1) return -1;
+  }
 
-    if (cpt_cracked > 0)
-    {
-      hc_thread_mutex_lock (status_ctx->mux_display);
+  if (device_param->is_hip == true)
+  {
+    if (run_hip_kernel_bzero (hashcat_ctx, device_param, device_param->hip_d_result, sizeof (u32)) == -1) return -1;
+  }
 
-      cpt_ctx->cpt_buf[cpt_ctx->cpt_pos].timestamp = time (NULL);
-      cpt_ctx->cpt_buf[cpt_ctx->cpt_pos].cracked   = cpt_cracked;
+  if (device_param->is_opencl == true)
+  {
+    if (run_opencl_kernel_bzero (hashcat_ctx, device_param, device_param->opencl_d_result, sizeof (u32)) == -1) return -1;
 
-      cpt_ctx->cpt_pos++;
-
-      cpt_ctx->cpt_total += cpt_cracked;
-
-      if (cpt_ctx->cpt_pos == CPT_CACHE) cpt_ctx->cpt_pos = 0;
-
-      hc_thread_mutex_unlock (status_ctx->mux_display);
-    }
-
-    num_cracked = 0;
-
-    if (device_param->is_cuda == true)
-    {
-      CU_rc = hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_result, &num_cracked, sizeof (u32));
-
-      if (CU_rc == -1) return -1;
-    }
-
-    if (device_param->is_hip == true)
-    {
-      HIP_rc = hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_result, &num_cracked, sizeof (u32));
-
-      if (HIP_rc == -1) return -1;
-    }
-
-    if (device_param->is_opencl == true)
-    {
-      CL_rc = hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_result, CL_TRUE, 0, sizeof (u32), &num_cracked, 0, NULL, NULL);
-
-      if (CL_rc == -1) return -1;
-    }
+    if (hc_clFlush (hashcat_ctx, device_param->opencl_command_queue) == -1) return -1;
   }
 
   return 0;
@@ -1569,7 +1674,6 @@ int hashes_init_stage2 (hashcat_ctx_t *hashcat_ctx)
   u32 digests_done = 0;
 
   u32 *digests_shown     = (u32 *) hccalloc (digests_cnt, sizeof (u32));
-  u32 *digests_shown_tmp = (u32 *) hccalloc (digests_cnt, sizeof (u32));
 
   u32 salts_cnt   = 0;
   u32 salts_done  = 0;
@@ -1706,7 +1810,6 @@ int hashes_init_stage2 (hashcat_ctx_t *hashcat_ctx)
   hashes->digests_done      = digests_done;
   hashes->digests_buf       = digests_buf_new;
   hashes->digests_shown     = digests_shown;
-  hashes->digests_shown_tmp = digests_shown_tmp;
 
   hashes->salts_cnt         = salts_cnt;
   hashes->salts_done        = salts_done;
@@ -1977,7 +2080,7 @@ int hashes_init_selftest (hashcat_ctx_t *hashcat_ctx)
 
       memcpy (hashconfig_st, hashconfig, sizeof (hashconfig_t));
 
-      hashconfig_st->separator = SEPARATOR;
+      hashconfig_st->separator = ':';
 
       if (user_options->hex_salt)
       {
@@ -2193,7 +2296,6 @@ void hashes_destroy (hashcat_ctx_t *hashcat_ctx)
 
   hcfree (hashes->digests_buf);
   hcfree (hashes->digests_shown);
-  hcfree (hashes->digests_shown_tmp);
 
   hcfree (hashes->salts_buf);
   hcfree (hashes->salts_shown);
diff --git a/src/hlfmt.c b/src/hlfmt.c
index 0e803445b..be71742e2 100644
--- a/src/hlfmt.c
+++ b/src/hlfmt.c
@@ -10,16 +10,16 @@
 #include "hlfmt.h"
 #include "shared.h"
 
-static const char *HLFMT_TEXT_HASHCAT  = "native hashcat";
-static const char *HLFMT_TEXT_PWDUMP   = "pwdump";
-static const char *HLFMT_TEXT_PASSWD   = "passwd";
-static const char *HLFMT_TEXT_SHADOW   = "shadow";
-static const char *HLFMT_TEXT_DCC      = "DCC";
-static const char *HLFMT_TEXT_DCC2     = "DCC 2";
-static const char *HLFMT_TEXT_NETNTLM1 = "NetNTLMv1";
-static const char *HLFMT_TEXT_NETNTLM2 = "NetNTLMv2";
-static const char *HLFMT_TEXT_NSLDAP   = "nsldap";
-static const char *HLFMT_TEXT_NSLDAPS  = "nsldaps";
+static const char *const HLFMT_TEXT_HASHCAT  = "native hashcat";
+static const char *const HLFMT_TEXT_PWDUMP   = "pwdump";
+static const char *const HLFMT_TEXT_PASSWD   = "passwd";
+static const char *const HLFMT_TEXT_SHADOW   = "shadow";
+static const char *const HLFMT_TEXT_DCC      = "DCC";
+static const char *const HLFMT_TEXT_DCC2     = "DCC 2";
+static const char *const HLFMT_TEXT_NETNTLM1 = "NetNTLMv1";
+static const char *const HLFMT_TEXT_NETNTLM2 = "NetNTLMv2";
+static const char *const HLFMT_TEXT_NSLDAP   = "nsldap";
+static const char *const HLFMT_TEXT_NSLDAPS  = "nsldaps";
 
 // hlfmt hashcat
 
diff --git a/src/hwmon.c b/src/hwmon.c
index 374056d07..4e6a4f986 100644
--- a/src/hwmon.c
+++ b/src/hwmon.c
@@ -109,19 +109,7 @@ int hm_get_threshold_slowdown_with_devices_idx (hashcat_ctx_t *hashcat_ctx, cons
           }
           else if (hwmon_ctx->hm_device[backend_device_idx].od_version == 6)
           {
-            int CurrentValue = 0;
-            int DefaultValue = 0;
 
-            if (hm_ADL_Overdrive6_TargetTemperatureData_Get (hashcat_ctx, hwmon_ctx->hm_device[backend_device_idx].adl, &CurrentValue, &DefaultValue) == -1)
-            {
-              hwmon_ctx->hm_device[backend_device_idx].threshold_slowdown_get_supported = false;
-
-              return -1;
-            }
-
-            // the return value has never been tested since hm_ADL_Overdrive6_TargetTemperatureData_Get() never worked on any system. expect problems.
-
-            return DefaultValue;
           }
         }
       }
@@ -346,6 +334,22 @@ int hm_get_temperature_with_devices_idx (hashcat_ctx_t *hashcat_ctx, const int b
 
             return Temperature / 1000;
           }
+
+          if (hwmon_ctx->hm_device[backend_device_idx].od_version == 8)
+          {
+            ADLPMLogDataOutput odlpDataOutput;
+
+            memset (&odlpDataOutput, 0, sizeof (ADLPMLogDataOutput));
+
+            if (hm_ADL2_New_QueryPMLogData_Get (hashcat_ctx, hwmon_ctx->hm_device[backend_device_idx].adl, &odlpDataOutput) == -1)
+            {
+              hwmon_ctx->hm_device[backend_device_idx].temperature_get_supported = false;
+
+              return -1;
+            }
+
+            return odlpDataOutput.sensors[PMLOG_TEMPERATURE_EDGE].value;
+          }
         }
 
         if (hwmon_ctx->hm_sysfs_amdgpu)
@@ -431,8 +435,37 @@ int hm_get_fanpolicy_with_devices_idx (hashcat_ctx_t *hashcat_ctx, const int bac
 
           if (hwmon_ctx->hm_device[backend_device_idx].od_version == 6)
           {
+            ADLOD6FanSpeedInfo lpFanSpeedInfo;
+
+            memset (&lpFanSpeedInfo, 0, sizeof (lpFanSpeedInfo));
+
+            if (hm_ADL_Overdrive6_FanSpeed_Get (hashcat_ctx, hwmon_ctx->hm_device[backend_device_idx].adl, &lpFanSpeedInfo) == -1)
+            {
+              hwmon_ctx->hm_device[backend_device_idx].fanpolicy_get_supported = false;
+              hwmon_ctx->hm_device[backend_device_idx].fanspeed_get_supported  = false;
+
+              return -1;
+            }
+
             return 1;
           }
+
+          if (hwmon_ctx->hm_device[backend_device_idx].od_version == 8)
+          {
+            ADLPMLogDataOutput odlpDataOutput;
+
+            memset (&odlpDataOutput, 0, sizeof (ADLPMLogDataOutput));
+
+            if (hm_ADL2_New_QueryPMLogData_Get (hashcat_ctx, hwmon_ctx->hm_device[backend_device_idx].adl, &odlpDataOutput) == -1)
+            {
+              hwmon_ctx->hm_device[backend_device_idx].fanpolicy_get_supported = false;
+              hwmon_ctx->hm_device[backend_device_idx].fanspeed_get_supported  = false;
+
+              return -1;
+            }
+
+            return odlpDataOutput.sensors[PMLOG_FAN_PERCENTAGE].supported;
+          }
         }
 
         if (hwmon_ctx->hm_sysfs_amdgpu)
@@ -542,6 +575,22 @@ int hm_get_fanspeed_with_devices_idx (hashcat_ctx_t *hashcat_ctx, const int back
 
             return faninfo.iFanSpeedPercent;
           }
+
+          if (hwmon_ctx->hm_device[backend_device_idx].od_version == 8)
+          {
+            ADLPMLogDataOutput odlpDataOutput;
+
+            memset (&odlpDataOutput, 0, sizeof (ADLPMLogDataOutput));
+
+            if (hm_ADL2_New_QueryPMLogData_Get (hashcat_ctx, hwmon_ctx->hm_device[backend_device_idx].adl, &odlpDataOutput) == -1)
+            {
+              hwmon_ctx->hm_device[backend_device_idx].fanspeed_get_supported = false;
+
+              return -1;
+            }
+
+            return odlpDataOutput.sensors[PMLOG_FAN_PERCENTAGE].value;
+          }
         }
 
         if (hwmon_ctx->hm_sysfs_amdgpu)
@@ -617,18 +666,37 @@ int hm_get_buslanes_with_devices_idx (hashcat_ctx_t *hashcat_ctx, const int back
       {
         if (hwmon_ctx->hm_adl)
         {
-          ADLPMActivity PMActivity;
-
-          PMActivity.iSize = sizeof (ADLPMActivity);
-
-          if (hm_ADL_Overdrive_CurrentActivity_Get (hashcat_ctx, hwmon_ctx->hm_device[backend_device_idx].adl, &PMActivity) == -1)
+          if (hwmon_ctx->hm_device[backend_device_idx].od_version == 5)
           {
-            hwmon_ctx->hm_device[backend_device_idx].buslanes_get_supported = false;
+            ADLPMActivity PMActivity;
 
-            return -1;
+            PMActivity.iSize = sizeof (ADLPMActivity);
+
+            if (hm_ADL_Overdrive_CurrentActivity_Get (hashcat_ctx, hwmon_ctx->hm_device[backend_device_idx].adl, &PMActivity) == -1)
+            {
+              hwmon_ctx->hm_device[backend_device_idx].buslanes_get_supported = false;
+
+              return -1;
+            }
+
+            return PMActivity.iCurrentBusLanes;
           }
 
-          return PMActivity.iCurrentBusLanes;
+          if (hwmon_ctx->hm_device[backend_device_idx].od_version == 8)
+          {
+            ADLPMLogDataOutput odlpDataOutput;
+
+            memset (&odlpDataOutput, 0, sizeof (ADLPMLogDataOutput));
+
+            if (hm_ADL2_New_QueryPMLogData_Get (hashcat_ctx, hwmon_ctx->hm_device[backend_device_idx].adl, &odlpDataOutput) == -1)
+            {
+              hwmon_ctx->hm_device[backend_device_idx].buslanes_get_supported = false;
+
+              return -1;
+            }
+
+            return odlpDataOutput.sensors[PMLOG_BUS_LANES].value;
+          }
         }
 
         if (hwmon_ctx->hm_sysfs_amdgpu)
@@ -704,18 +772,37 @@ int hm_get_utilization_with_devices_idx (hashcat_ctx_t *hashcat_ctx, const int b
       {
         if (hwmon_ctx->hm_adl)
         {
-          ADLPMActivity PMActivity;
-
-          PMActivity.iSize = sizeof (ADLPMActivity);
-
-          if (hm_ADL_Overdrive_CurrentActivity_Get (hashcat_ctx, hwmon_ctx->hm_device[backend_device_idx].adl, &PMActivity) == -1)
+          if (hwmon_ctx->hm_device[backend_device_idx].od_version == 5)
           {
-            hwmon_ctx->hm_device[backend_device_idx].utilization_get_supported = false;
+            ADLPMActivity PMActivity;
 
-            return -1;
+            PMActivity.iSize = sizeof (ADLPMActivity);
+
+            if (hm_ADL_Overdrive_CurrentActivity_Get (hashcat_ctx, hwmon_ctx->hm_device[backend_device_idx].adl, &PMActivity) == -1)
+            {
+              hwmon_ctx->hm_device[backend_device_idx].utilization_get_supported = false;
+
+              return -1;
+            }
+
+            return PMActivity.iActivityPercent;
           }
 
-          return PMActivity.iActivityPercent;
+          if (hwmon_ctx->hm_device[backend_device_idx].od_version == 8)
+          {
+            ADLPMLogDataOutput odlpDataOutput;
+
+            memset (&odlpDataOutput, 0, sizeof (ADLPMLogDataOutput));
+
+            if (hm_ADL2_New_QueryPMLogData_Get (hashcat_ctx, hwmon_ctx->hm_device[backend_device_idx].adl, &odlpDataOutput) == -1)
+            {
+              hwmon_ctx->hm_device[backend_device_idx].utilization_get_supported = false;
+
+              return -1;
+            }
+
+            return odlpDataOutput.sensors[PMLOG_INFO_ACTIVITY_GFX].value;
+          }
         }
 
         if (hwmon_ctx->hm_sysfs_amdgpu)
@@ -808,18 +895,37 @@ int hm_get_memoryspeed_with_devices_idx (hashcat_ctx_t *hashcat_ctx, const int b
       {
         if (hwmon_ctx->hm_adl)
         {
-          ADLPMActivity PMActivity;
-
-          PMActivity.iSize = sizeof (ADLPMActivity);
-
-          if (hm_ADL_Overdrive_CurrentActivity_Get (hashcat_ctx, hwmon_ctx->hm_device[backend_device_idx].adl, &PMActivity) == -1)
+          if (hwmon_ctx->hm_device[backend_device_idx].od_version == 5)
           {
-            hwmon_ctx->hm_device[backend_device_idx].memoryspeed_get_supported = false;
+            ADLPMActivity PMActivity;
 
-            return -1;
+            PMActivity.iSize = sizeof (ADLPMActivity);
+
+            if (hm_ADL_Overdrive_CurrentActivity_Get (hashcat_ctx, hwmon_ctx->hm_device[backend_device_idx].adl, &PMActivity) == -1)
+            {
+              hwmon_ctx->hm_device[backend_device_idx].memoryspeed_get_supported = false;
+
+              return -1;
+            }
+
+            return PMActivity.iMemoryClock / 100;
           }
 
-          return PMActivity.iMemoryClock / 100;
+          if (hwmon_ctx->hm_device[backend_device_idx].od_version == 8)
+          {
+            ADLPMLogDataOutput odlpDataOutput;
+
+            memset (&odlpDataOutput, 0, sizeof (ADLPMLogDataOutput));
+
+            if (hm_ADL2_New_QueryPMLogData_Get (hashcat_ctx, hwmon_ctx->hm_device[backend_device_idx].adl, &odlpDataOutput) == -1)
+            {
+              hwmon_ctx->hm_device[backend_device_idx].memoryspeed_get_supported = false;
+
+              return -1;
+            }
+
+            return odlpDataOutput.sensors[PMLOG_CLK_MEMCLK].value;
+          }
         }
 
         if (hwmon_ctx->hm_sysfs_amdgpu)
@@ -895,18 +1001,37 @@ int hm_get_corespeed_with_devices_idx (hashcat_ctx_t *hashcat_ctx, const int bac
       {
         if (hwmon_ctx->hm_adl)
         {
-          ADLPMActivity PMActivity;
-
-          PMActivity.iSize = sizeof (ADLPMActivity);
-
-          if (hm_ADL_Overdrive_CurrentActivity_Get (hashcat_ctx, hwmon_ctx->hm_device[backend_device_idx].adl, &PMActivity) == -1)
+          if (hwmon_ctx->hm_device[backend_device_idx].od_version == 5)
           {
-            hwmon_ctx->hm_device[backend_device_idx].corespeed_get_supported = false;
+            ADLPMActivity PMActivity;
 
-            return -1;
+            PMActivity.iSize = sizeof (ADLPMActivity);
+
+            if (hm_ADL_Overdrive_CurrentActivity_Get (hashcat_ctx, hwmon_ctx->hm_device[backend_device_idx].adl, &PMActivity) == -1)
+            {
+              hwmon_ctx->hm_device[backend_device_idx].corespeed_get_supported = false;
+
+              return -1;
+            }
+
+            return PMActivity.iEngineClock / 100;
           }
 
-          return PMActivity.iEngineClock / 100;
+          if (hwmon_ctx->hm_device[backend_device_idx].od_version == 8)
+          {
+            ADLPMLogDataOutput odlpDataOutput;
+
+            memset (&odlpDataOutput, 0, sizeof (ADLPMLogDataOutput));
+
+            if (hm_ADL2_New_QueryPMLogData_Get (hashcat_ctx, hwmon_ctx->hm_device[backend_device_idx].adl, &odlpDataOutput) == -1)
+            {
+              hwmon_ctx->hm_device[backend_device_idx].corespeed_get_supported = false;
+
+              return -1;
+            }
+
+            return odlpDataOutput.sensors[PMLOG_CLK_GFXCLK].value;
+          }
         }
 
         if (hwmon_ctx->hm_sysfs_amdgpu)
@@ -1400,7 +1525,9 @@ int hwmon_ctx_init (hashcat_ctx_t *hashcat_ctx)
               int od_enabled   = 0;
               int od_version   = 0;
 
-              hm_ADL_Overdrive_Caps (hashcat_ctx, lpAdapterInfo[i].iAdapterIndex, &od_supported, &od_enabled, &od_version);
+              hm_ADL2_Overdrive_Caps (hashcat_ctx, lpAdapterInfo[i].iAdapterIndex, &od_supported, &od_enabled, &od_version);
+
+              if (od_version < 8) od_version = 5;
 
               hm_adapters_adl[device_id].od_version = od_version;
 
@@ -1534,12 +1661,6 @@ int hwmon_ctx_init (hashcat_ctx_t *hashcat_ctx)
 
   hwmon_ctx->enabled = true;
 
-  /**
-   * save buffer required for later restores
-   */
-
-  hwmon_ctx->od_clock_mem_status = (ADLOD6MemClockState *) hccalloc (backend_ctx->backend_devices_cnt, sizeof (ADLOD6MemClockState));
-
   /**
    * HM devices: copy
    */
@@ -1795,8 +1916,6 @@ void hwmon_ctx_destroy (hashcat_ctx_t *hashcat_ctx)
 
   // free memory
 
-  hcfree (hwmon_ctx->od_clock_mem_status);
-
   hcfree (hwmon_ctx->hm_device);
 
   memset (hwmon_ctx, 0, sizeof (hwmon_ctx_t));
diff --git a/src/interface.c b/src/interface.c
index 1d2eb8538..49fe712e7 100644
--- a/src/interface.c
+++ b/src/interface.c
@@ -129,7 +129,7 @@ int hashconfig_init (hashcat_ctx_t *hashcat_ctx)
   #define CHECK_DEFINED(func)                                                     \
     if ((func) == NULL)                                                           \
     {                                                                             \
-      event_log_error (hashcat_ctx, "Missing symbol definitions. Old template?"); \
+      event_log_error (hashcat_ctx, "Missing symbol definitions module for in hash-mode '%d'. Old template?", user_options->hash_mode); \
                                                                                   \
       return -1;                                                                  \
     }
@@ -141,6 +141,7 @@ int hashconfig_init (hashcat_ctx_t *hashcat_ctx)
   CHECK_DEFINED (module_ctx->module_benchmark_salt);
   CHECK_DEFINED (module_ctx->module_build_plain_postprocess);
   CHECK_DEFINED (module_ctx->module_deep_comp_kernel);
+  CHECK_DEFINED (module_ctx->module_deprecated_notice);
   CHECK_DEFINED (module_ctx->module_dgst_pos0);
   CHECK_DEFINED (module_ctx->module_dgst_pos1);
   CHECK_DEFINED (module_ctx->module_dgst_pos2);
@@ -150,6 +151,7 @@ int hashconfig_init (hashcat_ctx_t *hashcat_ctx)
   CHECK_DEFINED (module_ctx->module_esalt_size);
   CHECK_DEFINED (module_ctx->module_extra_buffer_size);
   CHECK_DEFINED (module_ctx->module_extra_tmp_size);
+  CHECK_DEFINED (module_ctx->module_extra_tuningdb_block);
   CHECK_DEFINED (module_ctx->module_forced_outfile_format);
   CHECK_DEFINED (module_ctx->module_hash_binary_count);
   CHECK_DEFINED (module_ctx->module_hash_binary_parse);
@@ -265,6 +267,11 @@ int hashconfig_init (hashcat_ctx_t *hashcat_ctx)
     }
   }
 
+  if (user_options->multiply_accel_disable == true)
+  {
+    hashconfig->opts_type |= OPTS_TYPE_MP_MULTI_DISABLE;
+  }
+
   if (user_options->self_test_disable == true)
   {
     hashconfig->opts_type |= OPTS_TYPE_SELF_TEST_DISABLE;
@@ -616,7 +623,7 @@ u64 default_hook_size (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED
 
 char default_separator (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
 {
-  return user_options->separator;
+  return user_options_extra->separator;
 }
 
 bool default_dictstat_disable (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
diff --git a/src/main.c b/src/main.c
index d3f7985af..2a2b680c7 100644
--- a/src/main.c
+++ b/src/main.c
@@ -1059,10 +1059,30 @@ static void main_selftest_finished (MAYBE_UNUSED hashcat_ctx_t *hashcat_ctx, MAY
   event_log_info_nn (hashcat_ctx, "Finished self-test");
 }
 
+static void main_autotune_starting (MAYBE_UNUSED hashcat_ctx_t *hashcat_ctx, MAYBE_UNUSED const void *buf, MAYBE_UNUSED const size_t len)
+{
+  const user_options_t *user_options = hashcat_ctx->user_options;
+
+  if (user_options->quiet == true) return;
+
+  event_log_info_nn (hashcat_ctx, "Starting autotune. Please be patient...");
+}
+
+static void main_autotune_finished (MAYBE_UNUSED hashcat_ctx_t *hashcat_ctx, MAYBE_UNUSED const void *buf, MAYBE_UNUSED const size_t len)
+{
+  const user_options_t *user_options = hashcat_ctx->user_options;
+
+  if (user_options->quiet == true) return;
+
+  event_log_info_nn (hashcat_ctx, "Finished autotune");
+}
+
 static void event (const u32 id, hashcat_ctx_t *hashcat_ctx, const void *buf, const size_t len)
 {
   switch (id)
   {
+    case EVENT_AUTOTUNE_FINISHED:         main_autotune_finished         (hashcat_ctx, buf, len); break;
+    case EVENT_AUTOTUNE_STARTING:         main_autotune_starting         (hashcat_ctx, buf, len); break;
     case EVENT_SELFTEST_FINISHED:         main_selftest_finished         (hashcat_ctx, buf, len); break;
     case EVENT_SELFTEST_STARTING:         main_selftest_starting         (hashcat_ctx, buf, len); break;
     case EVENT_AUTODETECT_FINISHED:       main_autodetect_finished       (hashcat_ctx, buf, len); break;
@@ -1153,6 +1173,8 @@ int main (int argc, char **argv)
 
   if (user_options_init (hashcat_ctx) == -1)
   {
+    hashcat_destroy (hashcat_ctx);
+
     hcfree (hashcat_ctx);
 
     return -1;
@@ -1162,6 +1184,10 @@ int main (int argc, char **argv)
 
   if (user_options_getopt (hashcat_ctx, argc, argv) == -1)
   {
+    user_options_destroy (hashcat_ctx);
+
+    hashcat_destroy (hashcat_ctx);
+
     hcfree (hashcat_ctx);
 
     return -1;
@@ -1169,6 +1195,10 @@ int main (int argc, char **argv)
 
   if (user_options_sanity (hashcat_ctx) == -1)
   {
+    user_options_destroy (hashcat_ctx);
+
+    hashcat_destroy (hashcat_ctx);
+
     hcfree (hashcat_ctx);
 
     return -1;
@@ -1193,6 +1223,10 @@ int main (int argc, char **argv)
   {
     printf ("%s\n", VERSION_TAG);
 
+    user_options_destroy (hashcat_ctx);
+
+    hashcat_destroy (hashcat_ctx);
+
     hcfree (hashcat_ctx);
 
     return 0;
diff --git a/src/modules/module_00000.c b/src/modules/module_00000.c
index 136aba0a4..bfd95d436 100644
--- a/src/modules/module_00000.c
+++ b/src/modules/module_00000.c
@@ -129,6 +129,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -138,6 +139,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00010.c b/src/modules/module_00010.c
index f5c31ad00..05b2a35c2 100644
--- a/src/modules/module_00010.c
+++ b/src/modules/module_00010.c
@@ -152,6 +152,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -161,6 +162,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00011.c b/src/modules/module_00011.c
index fcff6b47c..047332caa 100644
--- a/src/modules/module_00011.c
+++ b/src/modules/module_00011.c
@@ -155,6 +155,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -164,6 +165,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00012.c b/src/modules/module_00012.c
index 9c6e90829..1b9a6d467 100644
--- a/src/modules/module_00012.c
+++ b/src/modules/module_00012.c
@@ -147,6 +147,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -156,6 +157,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00020.c b/src/modules/module_00020.c
index e3ee404dd..db6005705 100644
--- a/src/modules/module_00020.c
+++ b/src/modules/module_00020.c
@@ -151,6 +151,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -160,6 +161,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00021.c b/src/modules/module_00021.c
index ce4686961..bba9d51a0 100644
--- a/src/modules/module_00021.c
+++ b/src/modules/module_00021.c
@@ -146,6 +146,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -155,6 +156,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00022.c b/src/modules/module_00022.c
index 42a50dbd0..3614927fa 100644
--- a/src/modules/module_00022.c
+++ b/src/modules/module_00022.c
@@ -259,6 +259,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -268,6 +269,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00023.c b/src/modules/module_00023.c
index 71b06a618..7fabddacc 100644
--- a/src/modules/module_00023.c
+++ b/src/modules/module_00023.c
@@ -161,6 +161,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -170,6 +171,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00024.c b/src/modules/module_00024.c
index 5248b46de..00ea5dd0b 100644
--- a/src/modules/module_00024.c
+++ b/src/modules/module_00024.c
@@ -145,6 +145,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -154,6 +155,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00030.c b/src/modules/module_00030.c
index 28d42da03..676dee542 100644
--- a/src/modules/module_00030.c
+++ b/src/modules/module_00030.c
@@ -153,6 +153,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -162,6 +163,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00040.c b/src/modules/module_00040.c
index 9ee815472..7b06e375f 100644
--- a/src/modules/module_00040.c
+++ b/src/modules/module_00040.c
@@ -152,6 +152,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -161,6 +162,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00050.c b/src/modules/module_00050.c
index 8eaa36679..6954a520b 100644
--- a/src/modules/module_00050.c
+++ b/src/modules/module_00050.c
@@ -131,6 +131,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -140,6 +141,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00060.c b/src/modules/module_00060.c
index 5bcaecfbf..32fb20fd6 100644
--- a/src/modules/module_00060.c
+++ b/src/modules/module_00060.c
@@ -131,6 +131,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -140,6 +141,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00070.c b/src/modules/module_00070.c
index c7df389de..622f658f7 100644
--- a/src/modules/module_00070.c
+++ b/src/modules/module_00070.c
@@ -127,6 +127,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -136,6 +137,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00100.c b/src/modules/module_00100.c
index b23531e46..d4b95d84e 100644
--- a/src/modules/module_00100.c
+++ b/src/modules/module_00100.c
@@ -145,6 +145,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -154,6 +155,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00101.c b/src/modules/module_00101.c
index e58114f4b..5aef6fd13 100644
--- a/src/modules/module_00101.c
+++ b/src/modules/module_00101.c
@@ -149,6 +149,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -158,6 +159,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00110.c b/src/modules/module_00110.c
index 8299e5900..c38b02d2f 100644
--- a/src/modules/module_00110.c
+++ b/src/modules/module_00110.c
@@ -171,6 +171,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -180,6 +181,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00111.c b/src/modules/module_00111.c
index 07335b5b8..088bcc6f8 100644
--- a/src/modules/module_00111.c
+++ b/src/modules/module_00111.c
@@ -175,6 +175,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -184,6 +185,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00112.c b/src/modules/module_00112.c
index d059d621d..accacadd0 100644
--- a/src/modules/module_00112.c
+++ b/src/modules/module_00112.c
@@ -172,6 +172,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -181,6 +182,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00120.c b/src/modules/module_00120.c
index 6e347a2a9..26b40e830 100644
--- a/src/modules/module_00120.c
+++ b/src/modules/module_00120.c
@@ -171,6 +171,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -180,6 +181,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00121.c b/src/modules/module_00121.c
index dab88f693..bdeb3d465 100644
--- a/src/modules/module_00121.c
+++ b/src/modules/module_00121.c
@@ -172,6 +172,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -181,6 +182,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00122.c b/src/modules/module_00122.c
index d5e46c318..9baee77e2 100644
--- a/src/modules/module_00122.c
+++ b/src/modules/module_00122.c
@@ -151,6 +151,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -160,6 +161,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00124.c b/src/modules/module_00124.c
index 17cd6ef8f..f5cd8652c 100644
--- a/src/modules/module_00124.c
+++ b/src/modules/module_00124.c
@@ -170,6 +170,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -179,6 +180,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00125.c b/src/modules/module_00125.c
index 29545c2a5..bbe69035a 100644
--- a/src/modules/module_00125.c
+++ b/src/modules/module_00125.c
@@ -153,6 +153,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -162,6 +163,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00130.c b/src/modules/module_00130.c
index 3ba62c5db..b79289ab9 100644
--- a/src/modules/module_00130.c
+++ b/src/modules/module_00130.c
@@ -172,6 +172,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -181,6 +182,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00131.c b/src/modules/module_00131.c
index 1a37a5876..c69a7688a 100644
--- a/src/modules/module_00131.c
+++ b/src/modules/module_00131.c
@@ -167,6 +167,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -176,6 +177,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00132.c b/src/modules/module_00132.c
index d8aed073d..9344b58b2 100644
--- a/src/modules/module_00132.c
+++ b/src/modules/module_00132.c
@@ -161,6 +161,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -170,6 +171,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00133.c b/src/modules/module_00133.c
index 1177017b0..93df96c69 100644
--- a/src/modules/module_00133.c
+++ b/src/modules/module_00133.c
@@ -144,6 +144,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -153,6 +154,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00140.c b/src/modules/module_00140.c
index f0ed02252..c57e4dfa7 100644
--- a/src/modules/module_00140.c
+++ b/src/modules/module_00140.c
@@ -172,6 +172,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -181,6 +182,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00141.c b/src/modules/module_00141.c
index e4d369381..cb798cd6c 100644
--- a/src/modules/module_00141.c
+++ b/src/modules/module_00141.c
@@ -179,6 +179,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -188,6 +189,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00150.c b/src/modules/module_00150.c
index 3673e511e..0df30dfd8 100644
--- a/src/modules/module_00150.c
+++ b/src/modules/module_00150.c
@@ -149,6 +149,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -158,6 +159,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00160.c b/src/modules/module_00160.c
index 81f4c9f41..f2b36bfdb 100644
--- a/src/modules/module_00160.c
+++ b/src/modules/module_00160.c
@@ -149,6 +149,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -158,6 +159,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00170.c b/src/modules/module_00170.c
index 0533ade86..28f89c6a3 100644
--- a/src/modules/module_00170.c
+++ b/src/modules/module_00170.c
@@ -146,6 +146,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -155,6 +156,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00200.c b/src/modules/module_00200.c
index 1cf90dda4..352f622ab 100644
--- a/src/modules/module_00200.c
+++ b/src/modules/module_00200.c
@@ -113,6 +113,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -122,6 +123,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00300.c b/src/modules/module_00300.c
index eabe88871..7b55f02b1 100644
--- a/src/modules/module_00300.c
+++ b/src/modules/module_00300.c
@@ -144,6 +144,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -153,6 +154,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00400.c b/src/modules/module_00400.c
index def472e8c..c20c8d417 100644
--- a/src/modules/module_00400.c
+++ b/src/modules/module_00400.c
@@ -240,6 +240,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -249,6 +250,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00500.c b/src/modules/module_00500.c
index da1492e45..a7a4e7b72 100644
--- a/src/modules/module_00500.c
+++ b/src/modules/module_00500.c
@@ -253,6 +253,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -262,6 +263,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00501.c b/src/modules/module_00501.c
index 5cb8776f2..599dd48d0 100644
--- a/src/modules/module_00501.c
+++ b/src/modules/module_00501.c
@@ -332,6 +332,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -341,6 +342,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00600.c b/src/modules/module_00600.c
index a6c4f2a8d..4c688430c 100644
--- a/src/modules/module_00600.c
+++ b/src/modules/module_00600.c
@@ -118,6 +118,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -127,6 +128,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_00900.c b/src/modules/module_00900.c
index 0e5215275..10e63079a 100644
--- a/src/modules/module_00900.c
+++ b/src/modules/module_00900.c
@@ -129,6 +129,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -138,6 +139,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_01000.c b/src/modules/module_01000.c
index 1d84c5edc..73c2946a8 100644
--- a/src/modules/module_01000.c
+++ b/src/modules/module_01000.c
@@ -132,6 +132,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -141,6 +142,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_01100.c b/src/modules/module_01100.c
index 8b7383399..973723597 100644
--- a/src/modules/module_01100.c
+++ b/src/modules/module_01100.c
@@ -156,6 +156,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -165,6 +166,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_01300.c b/src/modules/module_01300.c
index 82dbc90ef..e62c34673 100644
--- a/src/modules/module_01300.c
+++ b/src/modules/module_01300.c
@@ -159,6 +159,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -168,6 +169,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_01400.c b/src/modules/module_01400.c
index b7a33ff43..5807afe96 100644
--- a/src/modules/module_01400.c
+++ b/src/modules/module_01400.c
@@ -166,6 +166,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -175,6 +176,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_01410.c b/src/modules/module_01410.c
index f6fda5840..c730bc04c 100644
--- a/src/modules/module_01410.c
+++ b/src/modules/module_01410.c
@@ -192,6 +192,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -201,6 +202,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_01411.c b/src/modules/module_01411.c
index 684366b6b..3da7fdf3c 100644
--- a/src/modules/module_01411.c
+++ b/src/modules/module_01411.c
@@ -188,6 +188,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -197,6 +198,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_01420.c b/src/modules/module_01420.c
index 3df3a8316..7aa82fa68 100644
--- a/src/modules/module_01420.c
+++ b/src/modules/module_01420.c
@@ -192,6 +192,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -201,6 +202,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_01421.c b/src/modules/module_01421.c
index 811363a50..591c6854c 100644
--- a/src/modules/module_01421.c
+++ b/src/modules/module_01421.c
@@ -168,6 +168,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -177,6 +178,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_01430.c b/src/modules/module_01430.c
index 48bd7e713..c9fcbd511 100644
--- a/src/modules/module_01430.c
+++ b/src/modules/module_01430.c
@@ -193,6 +193,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -202,6 +203,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_01440.c b/src/modules/module_01440.c
index f45bd58b6..3d308f851 100644
--- a/src/modules/module_01440.c
+++ b/src/modules/module_01440.c
@@ -193,6 +193,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -202,6 +203,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_01441.c b/src/modules/module_01441.c
index 67f290425..787fb74bb 100644
--- a/src/modules/module_01441.c
+++ b/src/modules/module_01441.c
@@ -194,6 +194,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -203,6 +204,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_01450.c b/src/modules/module_01450.c
index 5a5bf97d8..12acafbc4 100644
--- a/src/modules/module_01450.c
+++ b/src/modules/module_01450.c
@@ -163,6 +163,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -172,6 +173,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_01460.c b/src/modules/module_01460.c
index f2952aa36..dc2254080 100644
--- a/src/modules/module_01460.c
+++ b/src/modules/module_01460.c
@@ -195,6 +195,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -204,6 +205,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_01470.c b/src/modules/module_01470.c
index ec2e26ae6..d65cebeb5 100644
--- a/src/modules/module_01470.c
+++ b/src/modules/module_01470.c
@@ -167,6 +167,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -176,6 +177,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_01500.c b/src/modules/module_01500.c
index ea01dab96..d954295b8 100644
--- a/src/modules/module_01500.c
+++ b/src/modules/module_01500.c
@@ -95,13 +95,6 @@ int module_build_plain_postprocess (MAYBE_UNUSED const hashconfig_t *hashconfig,
   return src_len;
 }
 
-u32 module_kernel_threads_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
-{
-  const u32 kernel_threads_max = 64; // performance only optimization
-
-  return kernel_threads_max;
-}
-
 u32 module_kernel_loops_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
 {
   u32 kernel_loops_max = KERNEL_LOOPS_MAX;
@@ -184,7 +177,11 @@ char *module_jit_build_options (MAYBE_UNUSED const hashconfig_t *hashconfig, MAY
   {
     if ((user_options->attack_mode == ATTACK_MODE_BF) && (hashes->salts_cnt == 1) && (user_options->slow_candidates == false))
     {
-      hc_asprintf (&jit_build_options, "-DDESCRYPT_SALT=%u -D _unroll", hashes->salts_buf[0].salt_buf[0] & 0xfff);
+      hc_asprintf (&jit_build_options, "-DDESCRYPT_SALT=%u -D _unroll -fno-experimental-new-pass-manager", hashes->salts_buf[0].salt_buf[0] & 0xfff);
+    }
+    else
+    {
+      hc_asprintf (&jit_build_options, "-D _unroll -fno-experimental-new-pass-manager");
     }
   }
   else
@@ -302,6 +299,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = module_build_plain_postprocess;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -311,6 +309,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
@@ -341,7 +340,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_kernel_accel_min         = MODULE_DEFAULT;
   module_ctx->module_kernel_loops_max         = module_kernel_loops_max;
   module_ctx->module_kernel_loops_min         = module_kernel_loops_min;
-  module_ctx->module_kernel_threads_max       = module_kernel_threads_max;
+  module_ctx->module_kernel_threads_max       = MODULE_DEFAULT;
   module_ctx->module_kernel_threads_min       = MODULE_DEFAULT;
   module_ctx->module_kern_type                = module_kern_type;
   module_ctx->module_kern_type_dynamic        = MODULE_DEFAULT;
diff --git a/src/modules/module_01600.c b/src/modules/module_01600.c
index f3d4d3e13..4aa724496 100644
--- a/src/modules/module_01600.c
+++ b/src/modules/module_01600.c
@@ -253,6 +253,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -262,6 +263,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_01700.c b/src/modules/module_01700.c
index 04f2762c1..49a9ac1f2 100644
--- a/src/modules/module_01700.c
+++ b/src/modules/module_01700.c
@@ -193,6 +193,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -202,6 +203,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_01710.c b/src/modules/module_01710.c
index a36c9492a..25a4a6ee0 100644
--- a/src/modules/module_01710.c
+++ b/src/modules/module_01710.c
@@ -193,6 +193,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -202,6 +203,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_01711.c b/src/modules/module_01711.c
index b037abe83..0dac035b3 100644
--- a/src/modules/module_01711.c
+++ b/src/modules/module_01711.c
@@ -189,6 +189,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -198,6 +199,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_01720.c b/src/modules/module_01720.c
index 3fdc77653..b1703e371 100644
--- a/src/modules/module_01720.c
+++ b/src/modules/module_01720.c
@@ -218,6 +218,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -227,6 +228,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_01722.c b/src/modules/module_01722.c
index 3264c5f46..10bc70b14 100644
--- a/src/modules/module_01722.c
+++ b/src/modules/module_01722.c
@@ -195,6 +195,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -204,6 +205,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_01730.c b/src/modules/module_01730.c
index f0b579fcc..7eece59cc 100644
--- a/src/modules/module_01730.c
+++ b/src/modules/module_01730.c
@@ -194,6 +194,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -203,6 +204,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_01731.c b/src/modules/module_01731.c
index c26465d77..77216d79d 100644
--- a/src/modules/module_01731.c
+++ b/src/modules/module_01731.c
@@ -180,6 +180,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -189,6 +190,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_01740.c b/src/modules/module_01740.c
index 9df594f94..a7b573cb3 100644
--- a/src/modules/module_01740.c
+++ b/src/modules/module_01740.c
@@ -194,6 +194,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -203,6 +204,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_01750.c b/src/modules/module_01750.c
index eb5ce8558..36b664d08 100644
--- a/src/modules/module_01750.c
+++ b/src/modules/module_01750.c
@@ -164,6 +164,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -173,6 +174,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_01760.c b/src/modules/module_01760.c
index 6811199af..d3b80ed11 100644
--- a/src/modules/module_01760.c
+++ b/src/modules/module_01760.c
@@ -165,6 +165,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -174,6 +175,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_01770.c b/src/modules/module_01770.c
index 50e802556..d5e833d41 100644
--- a/src/modules/module_01770.c
+++ b/src/modules/module_01770.c
@@ -182,6 +182,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -191,6 +192,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_01800.c b/src/modules/module_01800.c
index aefab6e3b..1c86ff9f5 100644
--- a/src/modules/module_01800.c
+++ b/src/modules/module_01800.c
@@ -21,7 +21,8 @@ static const char *HASH_NAME      = "sha512crypt $6$, SHA512 (Unix)";
 static const u64   KERN_TYPE      = 1800;
 static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
                                   | OPTI_TYPE_USES_BITS_64;
-static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE;
+static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
+                                  | OPTS_TYPE_MP_MULTI_DISABLE;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
 static const char *ST_PASS        = "hashcat";
 static const char *ST_HASH        = "$6$72820166$U4DVzpcYxgw7MVVDGGvB2/H5lRistD5.Ah4upwENR5UtffLR4X4SxSzfREv8z6wVl0jRFX40/KnYVvK4829kD1";
@@ -557,6 +558,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -566,6 +568,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_02000.c b/src/modules/module_02000.c
index 19fb60394..790a0f9ef 100644
--- a/src/modules/module_02000.c
+++ b/src/modules/module_02000.c
@@ -70,6 +70,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -79,6 +80,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_02100.c b/src/modules/module_02100.c
index 371d4a14e..c0bd77b77 100644
--- a/src/modules/module_02100.c
+++ b/src/modules/module_02100.c
@@ -166,6 +166,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -175,6 +176,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_02400.c b/src/modules/module_02400.c
index 5d1d18d02..b10a42e0a 100644
--- a/src/modules/module_02400.c
+++ b/src/modules/module_02400.c
@@ -160,6 +160,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -169,6 +170,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_02410.c b/src/modules/module_02410.c
index 502b48474..7f7f91914 100644
--- a/src/modules/module_02410.c
+++ b/src/modules/module_02410.c
@@ -176,6 +176,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -185,6 +186,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_02500.c b/src/modules/module_02500.c
index aede25a2f..ba534ae6d 100644
--- a/src/modules/module_02500.c
+++ b/src/modules/module_02500.c
@@ -889,6 +889,13 @@ int module_hash_encode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSE
   return line_len;
 }
 
+const char *module_deprecated_notice (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const char *deprecated_notice = "The plugin 2500 is deprecated and was replaced with plugin 22000. For more details, please read: https://hashcat.net/forum/thread-10253.html";
+
+  return deprecated_notice;
+}
+
 void module_init (module_ctx_t *module_ctx)
 {
   module_ctx->module_context_size             = MODULE_CONTEXT_SIZE_CURRENT;
@@ -901,6 +908,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = module_deep_comp_kernel;
+  module_ctx->module_deprecated_notice        = module_deprecated_notice;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -910,6 +918,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = module_hash_binary_count;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_02501.c b/src/modules/module_02501.c
index e29a64065..c789a037e 100644
--- a/src/modules/module_02501.c
+++ b/src/modules/module_02501.c
@@ -886,6 +886,13 @@ int module_hash_encode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSE
   return line_len;
 }
 
+const char *module_deprecated_notice (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const char *deprecated_notice = "The plugin 2501 is deprecated and was replaced with plugin 22001. For more details, please read: https://hashcat.net/forum/thread-10253.html";
+
+  return deprecated_notice;
+}
+
 void module_init (module_ctx_t *module_ctx)
 {
   module_ctx->module_context_size             = MODULE_CONTEXT_SIZE_CURRENT;
@@ -898,6 +905,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = module_deep_comp_kernel;
+  module_ctx->module_deprecated_notice        = module_deprecated_notice;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -907,6 +915,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = module_hash_binary_count;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_02600.c b/src/modules/module_02600.c
index 9cc503202..1f1852718 100644
--- a/src/modules/module_02600.c
+++ b/src/modules/module_02600.c
@@ -138,6 +138,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -147,6 +148,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_02611.c b/src/modules/module_02611.c
index 20980c526..b4faebd16 100644
--- a/src/modules/module_02611.c
+++ b/src/modules/module_02611.c
@@ -152,6 +152,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -161,6 +162,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_02612.c b/src/modules/module_02612.c
index 6629f9123..e579c255a 100644
--- a/src/modules/module_02612.c
+++ b/src/modules/module_02612.c
@@ -154,6 +154,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -163,6 +164,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_02711.c b/src/modules/module_02711.c
index 50f58dea4..616997319 100644
--- a/src/modules/module_02711.c
+++ b/src/modules/module_02711.c
@@ -136,6 +136,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -145,6 +146,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_02811.c b/src/modules/module_02811.c
index 71eaaef7b..0f5ef00cc 100644
--- a/src/modules/module_02811.c
+++ b/src/modules/module_02811.c
@@ -166,6 +166,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -175,6 +176,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_03000.c b/src/modules/module_03000.c
index c9b616ab5..e7c054e99 100644
--- a/src/modules/module_03000.c
+++ b/src/modules/module_03000.c
@@ -96,13 +96,6 @@ char *module_jit_build_options (MAYBE_UNUSED const hashconfig_t *hashconfig, MAY
   return jit_build_options;
 }
 
-u32 module_kernel_threads_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
-{
-  const u32 kernel_threads_max = 64; // performance only optimization
-
-  return kernel_threads_max;
-}
-
 u32 module_kernel_loops_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
 {
   u32 kernel_loops_max = KERNEL_LOOPS_MAX;
@@ -230,6 +223,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -239,6 +233,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
@@ -269,7 +264,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_kernel_accel_min         = MODULE_DEFAULT;
   module_ctx->module_kernel_loops_max         = module_kernel_loops_max;
   module_ctx->module_kernel_loops_min         = module_kernel_loops_min;
-  module_ctx->module_kernel_threads_max       = module_kernel_threads_max;
+  module_ctx->module_kernel_threads_max       = MODULE_DEFAULT;
   module_ctx->module_kernel_threads_min       = MODULE_DEFAULT;
   module_ctx->module_kern_type                = module_kern_type;
   module_ctx->module_kern_type_dynamic        = MODULE_DEFAULT;
diff --git a/src/modules/module_03100.c b/src/modules/module_03100.c
index 53a7dd9e3..5a7cf57a8 100644
--- a/src/modules/module_03100.c
+++ b/src/modules/module_03100.c
@@ -134,6 +134,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -143,6 +144,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_03200.c b/src/modules/module_03200.c
index 9f30edadf..820775550 100644
--- a/src/modules/module_03200.c
+++ b/src/modules/module_03200.c
@@ -21,6 +21,7 @@ static const char *HASH_NAME      = "bcrypt $2*$, Blowfish (Unix)";
 static const u64   KERN_TYPE      = 3200;
 static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE;
 static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
+                                  | OPTS_TYPE_MP_MULTI_DISABLE
                                   | OPTS_TYPE_DYNAMIC_SHARED;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
 static const char *ST_PASS        = "hashcat";
@@ -294,6 +295,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -303,6 +305,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_03500.c b/src/modules/module_03500.c
index 10182be02..103ce86a1 100644
--- a/src/modules/module_03500.c
+++ b/src/modules/module_03500.c
@@ -138,6 +138,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -147,6 +148,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_03710.c b/src/modules/module_03710.c
index 3d30567cb..e6eaf57b9 100644
--- a/src/modules/module_03710.c
+++ b/src/modules/module_03710.c
@@ -163,6 +163,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -172,6 +173,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_03711.c b/src/modules/module_03711.c
index ee5e1d3f6..0d1d755df 100644
--- a/src/modules/module_03711.c
+++ b/src/modules/module_03711.c
@@ -166,6 +166,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -175,6 +176,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_03800.c b/src/modules/module_03800.c
index 5656c411f..14940432c 100644
--- a/src/modules/module_03800.c
+++ b/src/modules/module_03800.c
@@ -152,6 +152,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -161,6 +162,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_03910.c b/src/modules/module_03910.c
index c4b81d6e6..6ca24fa08 100644
--- a/src/modules/module_03910.c
+++ b/src/modules/module_03910.c
@@ -168,6 +168,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -177,6 +178,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_04010.c b/src/modules/module_04010.c
index 79d38481f..074eb0bf6 100644
--- a/src/modules/module_04010.c
+++ b/src/modules/module_04010.c
@@ -166,6 +166,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -175,6 +176,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_04110.c b/src/modules/module_04110.c
index 653f27164..00c48acaf 100644
--- a/src/modules/module_04110.c
+++ b/src/modules/module_04110.c
@@ -164,6 +164,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -173,6 +174,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_04300.c b/src/modules/module_04300.c
index 0e348b89f..b98e8aa3f 100644
--- a/src/modules/module_04300.c
+++ b/src/modules/module_04300.c
@@ -138,6 +138,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -147,6 +148,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_04400.c b/src/modules/module_04400.c
index 478e4e2cf..cf917627e 100644
--- a/src/modules/module_04400.c
+++ b/src/modules/module_04400.c
@@ -128,6 +128,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -137,6 +138,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_04500.c b/src/modules/module_04500.c
index c4943a7d4..c09641cc1 100644
--- a/src/modules/module_04500.c
+++ b/src/modules/module_04500.c
@@ -143,6 +143,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -152,6 +153,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_04510.c b/src/modules/module_04510.c
index f6047bc90..7bb55f131 100644
--- a/src/modules/module_04510.c
+++ b/src/modules/module_04510.c
@@ -170,6 +170,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -179,6 +180,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_04520.c b/src/modules/module_04520.c
index 0ecafefa4..e940b01f1 100644
--- a/src/modules/module_04520.c
+++ b/src/modules/module_04520.c
@@ -151,6 +151,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -160,6 +161,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_04521.c b/src/modules/module_04521.c
index d326795d6..774ab3075 100644
--- a/src/modules/module_04521.c
+++ b/src/modules/module_04521.c
@@ -143,6 +143,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -152,6 +153,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_04522.c b/src/modules/module_04522.c
index 8fcc1b4f0..ddab2bea8 100644
--- a/src/modules/module_04522.c
+++ b/src/modules/module_04522.c
@@ -143,6 +143,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -152,6 +153,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_04700.c b/src/modules/module_04700.c
index 959ccd1a8..b3bb57e43 100644
--- a/src/modules/module_04700.c
+++ b/src/modules/module_04700.c
@@ -145,6 +145,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -154,6 +155,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_04710.c b/src/modules/module_04710.c
index 4c7361f3c..4e4dea1b5 100644
--- a/src/modules/module_04710.c
+++ b/src/modules/module_04710.c
@@ -170,6 +170,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -179,6 +180,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_04711.c b/src/modules/module_04711.c
index f4961f306..9e583d179 100644
--- a/src/modules/module_04711.c
+++ b/src/modules/module_04711.c
@@ -184,6 +184,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -193,6 +194,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_04800.c b/src/modules/module_04800.c
index baecc963c..b64e7ea22 100644
--- a/src/modules/module_04800.c
+++ b/src/modules/module_04800.c
@@ -151,6 +151,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -160,6 +161,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_04900.c b/src/modules/module_04900.c
index 151537fa8..07f453ef8 100644
--- a/src/modules/module_04900.c
+++ b/src/modules/module_04900.c
@@ -166,6 +166,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -175,6 +176,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_05000.c b/src/modules/module_05000.c
index 67ef974ce..c53cf5f24 100644
--- a/src/modules/module_05000.c
+++ b/src/modules/module_05000.c
@@ -166,6 +166,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -175,6 +176,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_05100.c b/src/modules/module_05100.c
index d1b07b76c..d1f7ab70f 100644
--- a/src/modules/module_05100.c
+++ b/src/modules/module_05100.c
@@ -106,6 +106,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -115,6 +116,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_05200.c b/src/modules/module_05200.c
index 6fb3f08d8..84087ad9c 100644
--- a/src/modules/module_05200.c
+++ b/src/modules/module_05200.c
@@ -192,6 +192,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -201,6 +202,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_05300.c b/src/modules/module_05300.c
index a1cd04076..05d4125b3 100644
--- a/src/modules/module_05300.c
+++ b/src/modules/module_05300.c
@@ -247,6 +247,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -256,6 +257,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_05400.c b/src/modules/module_05400.c
index d9472c480..961070d97 100644
--- a/src/modules/module_05400.c
+++ b/src/modules/module_05400.c
@@ -254,6 +254,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -263,6 +264,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_05500.c b/src/modules/module_05500.c
index 447a64d83..0e6951cd6 100644
--- a/src/modules/module_05500.c
+++ b/src/modules/module_05500.c
@@ -413,6 +413,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -422,6 +423,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_05600.c b/src/modules/module_05600.c
index 3c96edf1e..74b74aecb 100644
--- a/src/modules/module_05600.c
+++ b/src/modules/module_05600.c
@@ -308,6 +308,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -317,6 +318,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_05700.c b/src/modules/module_05700.c
index 9d3b86a93..a88ea1a5c 100644
--- a/src/modules/module_05700.c
+++ b/src/modules/module_05700.c
@@ -154,6 +154,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -163,6 +164,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_05800.c b/src/modules/module_05800.c
index 4c8f6d881..38221f6be 100644
--- a/src/modules/module_05800.c
+++ b/src/modules/module_05800.c
@@ -167,6 +167,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -176,6 +177,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_06000.c b/src/modules/module_06000.c
index 0d348f0e3..75ad6afeb 100644
--- a/src/modules/module_06000.c
+++ b/src/modules/module_06000.c
@@ -109,6 +109,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -118,6 +119,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_06100.c b/src/modules/module_06100.c
index 88d80f3ad..ed6eb5707 100644
--- a/src/modules/module_06100.c
+++ b/src/modules/module_06100.c
@@ -189,6 +189,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -198,6 +199,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_06211.c b/src/modules/module_06211.c
index 5cb417d26..a03f9eb61 100644
--- a/src/modules/module_06211.c
+++ b/src/modules/module_06211.c
@@ -251,6 +251,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -260,6 +261,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_06212.c b/src/modules/module_06212.c
index ceb18f192..4d312ca83 100644
--- a/src/modules/module_06212.c
+++ b/src/modules/module_06212.c
@@ -251,6 +251,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -260,6 +261,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_06213.c b/src/modules/module_06213.c
index 20323fc62..58f92827c 100644
--- a/src/modules/module_06213.c
+++ b/src/modules/module_06213.c
@@ -249,6 +249,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -258,6 +259,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_06221.c b/src/modules/module_06221.c
index 38cb70ed8..ef07759f7 100644
--- a/src/modules/module_06221.c
+++ b/src/modules/module_06221.c
@@ -253,6 +253,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -262,6 +263,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_06222.c b/src/modules/module_06222.c
index 4873b8fab..6848d39fe 100644
--- a/src/modules/module_06222.c
+++ b/src/modules/module_06222.c
@@ -253,6 +253,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -262,6 +263,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_06223.c b/src/modules/module_06223.c
index 8d4261420..0cecd824f 100644
--- a/src/modules/module_06223.c
+++ b/src/modules/module_06223.c
@@ -253,6 +253,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -262,6 +263,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_06231.c b/src/modules/module_06231.c
index c9699c033..0cd97afe7 100644
--- a/src/modules/module_06231.c
+++ b/src/modules/module_06231.c
@@ -263,6 +263,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -272,6 +273,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_06232.c b/src/modules/module_06232.c
index 496b9c575..e58c22059 100644
--- a/src/modules/module_06232.c
+++ b/src/modules/module_06232.c
@@ -263,6 +263,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -272,6 +273,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_06233.c b/src/modules/module_06233.c
index e6ba28e2d..cffbf12e0 100644
--- a/src/modules/module_06233.c
+++ b/src/modules/module_06233.c
@@ -259,6 +259,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -268,6 +269,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_06241.c b/src/modules/module_06241.c
index 47da47ffb..661f978cd 100644
--- a/src/modules/module_06241.c
+++ b/src/modules/module_06241.c
@@ -264,6 +264,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = module_build_plain_postprocess;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -273,6 +274,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_06242.c b/src/modules/module_06242.c
index 014b5d161..f6ad6359b 100644
--- a/src/modules/module_06242.c
+++ b/src/modules/module_06242.c
@@ -264,6 +264,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = module_build_plain_postprocess;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -273,6 +274,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_06243.c b/src/modules/module_06243.c
index a84abdd7c..851072fa4 100644
--- a/src/modules/module_06243.c
+++ b/src/modules/module_06243.c
@@ -264,6 +264,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = module_build_plain_postprocess;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -273,6 +274,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_06300.c b/src/modules/module_06300.c
index 0b5dbdc5c..2a7681ccb 100644
--- a/src/modules/module_06300.c
+++ b/src/modules/module_06300.c
@@ -244,6 +244,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -253,6 +254,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_06400.c b/src/modules/module_06400.c
index 2b032f4f5..b13ec72f2 100644
--- a/src/modules/module_06400.c
+++ b/src/modules/module_06400.c
@@ -369,6 +369,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -378,6 +379,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_06500.c b/src/modules/module_06500.c
index b95a4bd0b..6b74824e9 100644
--- a/src/modules/module_06500.c
+++ b/src/modules/module_06500.c
@@ -542,6 +542,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -551,6 +552,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_06600.c b/src/modules/module_06600.c
index be83e5567..c93ec971b 100644
--- a/src/modules/module_06600.c
+++ b/src/modules/module_06600.c
@@ -220,6 +220,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -229,6 +230,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_06700.c b/src/modules/module_06700.c
index f63248ca7..cd6f3d70d 100644
--- a/src/modules/module_06700.c
+++ b/src/modules/module_06700.c
@@ -295,6 +295,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -304,6 +305,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_06800.c b/src/modules/module_06800.c
index 0f25fa29d..57c953b2e 100644
--- a/src/modules/module_06800.c
+++ b/src/modules/module_06800.c
@@ -195,6 +195,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -204,6 +205,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_06900.c b/src/modules/module_06900.c
index cf9ea46c0..7ec00149e 100644
--- a/src/modules/module_06900.c
+++ b/src/modules/module_06900.c
@@ -120,6 +120,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -129,6 +130,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_07000.c b/src/modules/module_07000.c
index aa19f0677..0a8de9d40 100644
--- a/src/modules/module_07000.c
+++ b/src/modules/module_07000.c
@@ -192,6 +192,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -201,6 +202,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_07100.c b/src/modules/module_07100.c
index 5f81ba616..3af3cca3f 100644
--- a/src/modules/module_07100.c
+++ b/src/modules/module_07100.c
@@ -243,6 +243,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -252,6 +253,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_07200.c b/src/modules/module_07200.c
index 5ed129893..1ca0d6398 100644
--- a/src/modules/module_07200.c
+++ b/src/modules/module_07200.c
@@ -214,6 +214,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -223,6 +224,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_07300.c b/src/modules/module_07300.c
index f4844669f..49725d3d4 100644
--- a/src/modules/module_07300.c
+++ b/src/modules/module_07300.c
@@ -170,6 +170,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -179,6 +180,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_07400.c b/src/modules/module_07400.c
index 3be47f898..7a860cc3a 100644
--- a/src/modules/module_07400.c
+++ b/src/modules/module_07400.c
@@ -368,6 +368,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -377,6 +378,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_07401.c b/src/modules/module_07401.c
index ba71bf179..fa573c143 100644
--- a/src/modules/module_07401.c
+++ b/src/modules/module_07401.c
@@ -413,6 +413,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -422,6 +423,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_07500.c b/src/modules/module_07500.c
index 1681fb4a8..9215a853b 100644
--- a/src/modules/module_07500.c
+++ b/src/modules/module_07500.c
@@ -65,35 +65,13 @@ char *module_jit_build_options (MAYBE_UNUSED const hashconfig_t *hashconfig, MAY
   }
   else if (device_param->opencl_device_type & CL_DEVICE_TYPE_GPU)
   {
-    if (device_param->opencl_device_vendor_id == VENDOR_ID_INTEL_SDK)
+    if (device_param->device_local_mem_size < 49152)
     {
-      native_threads = 8;
-    }
-    else if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD)
-    {
-      if (device_param->device_local_mem_size < 49152)
-      {
-        native_threads = 32;
-      }
-      else
-      {
-        native_threads = 64;
-      }
-    }
-    else if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD_USE_HIP)
-    {
-      if (device_param->device_local_mem_size < 49152)
-      {
-        native_threads = 32;
-      }
-      else
-      {
-        native_threads = 64;
-      }
+      native_threads = MIN (device_param->kernel_preferred_wgs_multiple, 32); // We can't just set 32, because Intel GPU need 8
     }
     else
     {
-      native_threads = 32;
+      native_threads = device_param->kernel_preferred_wgs_multiple;
     }
   }
 
@@ -268,6 +246,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -277,6 +256,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_07700.c b/src/modules/module_07700.c
index 59d6337f9..3775ca786 100644
--- a/src/modules/module_07700.c
+++ b/src/modules/module_07700.c
@@ -152,6 +152,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -161,6 +162,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_07701.c b/src/modules/module_07701.c
index 5dbffa59c..474dd0e66 100644
--- a/src/modules/module_07701.c
+++ b/src/modules/module_07701.c
@@ -145,6 +145,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -154,6 +155,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_07800.c b/src/modules/module_07800.c
index 878847c7a..b36b67ed5 100644
--- a/src/modules/module_07800.c
+++ b/src/modules/module_07800.c
@@ -168,6 +168,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -177,6 +178,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_07801.c b/src/modules/module_07801.c
index e1d936421..755c612a5 100644
--- a/src/modules/module_07801.c
+++ b/src/modules/module_07801.c
@@ -168,6 +168,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -177,6 +178,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_07900.c b/src/modules/module_07900.c
index a51efecd9..82e803b75 100644
--- a/src/modules/module_07900.c
+++ b/src/modules/module_07900.c
@@ -452,6 +452,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -461,6 +462,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_08000.c b/src/modules/module_08000.c
index 65aa287f8..25f7f0a68 100644
--- a/src/modules/module_08000.c
+++ b/src/modules/module_08000.c
@@ -173,6 +173,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -182,6 +183,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_08100.c b/src/modules/module_08100.c
index 1318fdfae..0390bb73c 100644
--- a/src/modules/module_08100.c
+++ b/src/modules/module_08100.c
@@ -163,6 +163,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -172,6 +173,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_08200.c b/src/modules/module_08200.c
index 617b80972..86164410a 100644
--- a/src/modules/module_08200.c
+++ b/src/modules/module_08200.c
@@ -261,6 +261,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -270,6 +271,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_08300.c b/src/modules/module_08300.c
index 26439a0f6..e1014fddb 100644
--- a/src/modules/module_08300.c
+++ b/src/modules/module_08300.c
@@ -211,6 +211,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -220,6 +221,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_08400.c b/src/modules/module_08400.c
index 11ef09579..13ed2b3e5 100644
--- a/src/modules/module_08400.c
+++ b/src/modules/module_08400.c
@@ -142,6 +142,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -151,6 +152,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_08500.c b/src/modules/module_08500.c
index 00243f12b..23be2df26 100644
--- a/src/modules/module_08500.c
+++ b/src/modules/module_08500.c
@@ -208,6 +208,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -217,6 +218,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_08600.c b/src/modules/module_08600.c
index 7b46885e8..ca4108dd1 100644
--- a/src/modules/module_08600.c
+++ b/src/modules/module_08600.c
@@ -105,6 +105,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -114,6 +115,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_08700.c b/src/modules/module_08700.c
index be902d527..c466be802 100644
--- a/src/modules/module_08700.c
+++ b/src/modules/module_08700.c
@@ -159,6 +159,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -168,6 +169,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_08800.c b/src/modules/module_08800.c
index 3c2b8667e..b5ff11f76 100644
--- a/src/modules/module_08800.c
+++ b/src/modules/module_08800.c
@@ -217,6 +217,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -226,6 +227,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_08900.c b/src/modules/module_08900.c
index de2a936d6..b8eafedfc 100644
--- a/src/modules/module_08900.c
+++ b/src/modules/module_08900.c
@@ -387,6 +387,47 @@ int module_hash_encode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSE
   return line_len;
 }
 
+/*
+
+Find the right -n value for your GPU:
+=====================================
+
+1. For example, to find the value for 8900, first create a valid hash for 8900 as follows:
+
+$ ./hashcat --example-hashes -m 8900 | grep Example.Hash | grep -v Format | cut -b 25- > tmp.hash.8900
+
+2. Now let it iterate through all -n values to a certain point. In this case, I'm using 200, but in general it's a value that is at least twice that of the multiprocessor. If you don't mind you can just leave it as it is, it just runs a little longer.
+
+$ export i=1; while [ $i -ne 201 ]; do echo $i; ./hashcat --quiet tmp.hash.8900 --keep-guessing --self-test-disable --markov-disable --restore-disable --outfile-autohex-disable --wordlist-autohex-disable --potfile-disable --logfile-disable --hwmon-disable --status --status-timer 1 --runtime 28 --machine-readable --optimized-kernel-enable --workload-profile 3 --hash-type 8900 --attack-mode 3 ?b?b?b?b?b?b?b --backend-devices 1 --force -n $i; i=$(($i+1)); done | tee x
+
+3. Determine the highest measured H/s speed. But don't just use the highest value. Instead, use the number that seems most stable, usually at the beginning.
+
+$ grep "$(printf 'STATUS\t3')" x | cut -f4 -d$'\t' | sort -n | tail
+
+4. To match the speed you have chosen to the correct value in the 'x' file, simply search for it in it. Then go up a little on the block where you found him. The value -n is the single value that begins before the block start. If you have multiple blocks at the same speed, choose the lowest value for -n
+
+*/
+
+const char *module_extra_tuningdb_block (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const char *extra_tuningdb_block =
+    "DEVICE_TYPE_CPU                                 *       8900    1       N       A\n"
+    "DEVICE_TYPE_GPU                                 *       8900    1       N       A\n"
+    "GeForce_GTX_980                                 *       8900    1      29       A\n"
+    "GeForce_GTX_1080                                *       8900    1      15       A\n"
+    "GeForce_RTX_2080_Ti                             *       8900    1      68       A\n"
+    "GeForce_RTX_3060_Ti                             *       8900    1      51       A\n"
+    "GeForce_RTX_3070                                *       8900    1      46       A\n"
+    "GeForce_RTX_3090                                *       8900    1      82       A\n"
+    "ALIAS_AMD_RX480                                 *       8900    1      15       A\n"
+    "ALIAS_AMD_Vega64                                *       8900    1      31       A\n"
+    "ALIAS_AMD_MI100                                 *       8900    1      79       A\n"
+    "ALIAS_AMD_RX6900XT                              *       8900    1      59       A\n"
+  ;
+
+  return extra_tuningdb_block;
+}
+
 void module_init (module_ctx_t *module_ctx)
 {
   module_ctx->module_context_size             = MODULE_CONTEXT_SIZE_CURRENT;
@@ -399,6 +440,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -408,6 +450,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = module_extra_buffer_size;
   module_ctx->module_extra_tmp_size           = module_extra_tmp_size;
+  module_ctx->module_extra_tuningdb_block     = module_extra_tuningdb_block;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_09000.c b/src/modules/module_09000.c
index 3e3158c0a..1531765a2 100644
--- a/src/modules/module_09000.c
+++ b/src/modules/module_09000.c
@@ -22,7 +22,8 @@ static const u64   KERN_TYPE      = 9000;
 static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE;
 static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
                                   | OPTS_TYPE_BINARY_HASHFILE
-                                  | OPTS_TYPE_AUTODETECT_DISABLE;
+                                  | OPTS_TYPE_AUTODETECT_DISABLE
+                                  | OPTS_TYPE_DYNAMIC_SHARED;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
 static const char *ST_PASS        = "hashcat";
 static const char *ST_HASH        = "0a3f352686e5eb5be173e668a4fff5cd5df420927e1da2d5d4052340160637e3e6a5a92841a188ed240e13b919f3d91694bd4c0acba79271e9c08a83ea5ad387cbb74d5884066a1cb5a8caa80d847079168f84823847c631dbe3a834f1bc496acfebac3bff1608bf1c857717f8f428e07b5e2cb12aaeddfa83d7dcb6d840234d08b84f8ca6c6e562af73eea13148f7902bcaf0220d3e36eeeff1d37283dc421483a2791182614ebb";
@@ -75,16 +76,25 @@ char *module_jit_build_options (MAYBE_UNUSED const hashconfig_t *hashconfig, MAY
 {
   char *jit_build_options = NULL;
 
+  // this mode heavily depends on the available shared memory size
+  // note the kernel need to have some special code changes in order to make use to use post-48k memory region
+  // we need to set some macros
+
+  bool use_dynamic = false;
+
+  if (device_param->is_cuda == true)
+  {
+    use_dynamic = true;
+  }
+
   // this uses some nice feedback effect.
   // based on the device_local_mem_size the reqd_work_group_size in the kernel is set to some value
   // which is then is read from the opencl host in the kernel_preferred_wgs_multiple1/2/3 result.
   // therefore we do not need to set module_kernel_threads_min/max except for CPU, where the threads are set to fixed 1.
 
-  u32 fixed_local_size = 0;
-
   if (device_param->opencl_device_type & CL_DEVICE_TYPE_CPU)
   {
-    fixed_local_size = 1;
+    hc_asprintf (&jit_build_options, "-D FIXED_LOCAL_SIZE=%u", 1);
   }
   else
   {
@@ -100,29 +110,58 @@ char *module_jit_build_options (MAYBE_UNUSED const hashconfig_t *hashconfig, MAY
 
       if (device_param->is_opencl == true)
       {
-        overhead = 4;
+        overhead = 1;
       }
     }
 
     if (user_options->kernel_threads_chgd == true)
     {
-      fixed_local_size = user_options->kernel_threads;
+      u32 fixed_local_size = user_options->kernel_threads;
 
-      // otherwise out-of-bound reads
-
-      if ((fixed_local_size * 4096) > (device_param->device_local_mem_size - overhead))
+      if (use_dynamic == true)
       {
-        fixed_local_size = (device_param->device_local_mem_size - overhead) / 4096;
+        if ((fixed_local_size * 4096) > device_param->kernel_dynamic_local_mem_size_memset)
+        {
+          // otherwise out-of-bound reads
+
+          fixed_local_size = device_param->kernel_dynamic_local_mem_size_memset / 4096;
+        }
+
+        hc_asprintf (&jit_build_options, "-D FIXED_LOCAL_SIZE=%u -D DYNAMIC_LOCAL", fixed_local_size);
+      }
+      else
+      {
+        if ((fixed_local_size * 4096) > (device_param->device_local_mem_size - overhead))
+        {
+          // otherwise out-of-bound reads
+
+          fixed_local_size = (device_param->device_local_mem_size - overhead) / 4096;
+        }
+
+        hc_asprintf (&jit_build_options, "-D FIXED_LOCAL_SIZE=%u", fixed_local_size);
       }
     }
     else
     {
-      fixed_local_size = (device_param->device_local_mem_size - overhead) / 4096;
+      if (use_dynamic == true)
+      {
+        // using kernel_dynamic_local_mem_size_memset is a bit hackish.
+        // we had to brute-force this value out of an already loaded CUDA function.
+        // there's no official way to query for this value.
+
+        const u32 fixed_local_size = device_param->kernel_dynamic_local_mem_size_memset / 4096;
+
+        hc_asprintf (&jit_build_options, "-D FIXED_LOCAL_SIZE=%u -D DYNAMIC_LOCAL", fixed_local_size);
+      }
+      else
+      {
+        const u32 fixed_local_size = (device_param->device_local_mem_size - overhead) / 4096;
+
+        hc_asprintf (&jit_build_options, "-D FIXED_LOCAL_SIZE=%u", fixed_local_size);
+      }
     }
   }
 
-  hc_asprintf (&jit_build_options, "-D FIXED_LOCAL_SIZE=%u", fixed_local_size);
-
   return jit_build_options;
 }
 
@@ -187,6 +226,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -196,6 +236,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_09100.c b/src/modules/module_09100.c
index 092ec71fe..6081f027f 100644
--- a/src/modules/module_09100.c
+++ b/src/modules/module_09100.c
@@ -198,6 +198,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -207,6 +208,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_09200.c b/src/modules/module_09200.c
index e1fd7ece2..3459876fc 100644
--- a/src/modules/module_09200.c
+++ b/src/modules/module_09200.c
@@ -212,6 +212,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -221,6 +222,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_09300.c b/src/modules/module_09300.c
index 05dcbc585..e90dfcd93 100644
--- a/src/modules/module_09300.c
+++ b/src/modules/module_09300.c
@@ -344,6 +344,47 @@ int module_hash_encode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSE
   return line_len;
 }
 
+/*
+
+Find the right -n value for your GPU:
+=====================================
+
+1. For example, to find the value for 9300, first create a valid hash for 9300 as follows:
+
+$ ./hashcat --example-hashes -m 9300 | grep Example.Hash | grep -v Format | cut -b 25- > tmp.hash.9300
+
+2. Now let it iterate through all -n values to a certain point. In this case, I'm using 1032, but in general it's a value that is at least twice that of the multiprocessor. If you don't mind you can just leave it as it is, it just runs a little longer.
+
+$ export i=8; while [ $i -ne 1032 ]; do echo $i; ./hashcat --quiet tmp.hash.9300 --keep-guessing --self-test-disable --markov-disable --restore-disable --outfile-autohex-disable --wordlist-autohex-disable --potfile-disable --logfile-disable --hwmon-disable --status --status-timer 1 --runtime 28 --machine-readable --optimized-kernel-enable --workload-profile 3 --hash-type 9300 --attack-mode 3 ?b?b?b?b?b?b?b --backend-devices 1 --force -n $i; i=$(($i+8)); done | tee x
+
+3. Determine the highest measured H/s speed. But don't just use the highest value. Instead, use the number that seems most stable, usually at the beginning.
+
+$ grep "$(printf 'STATUS\t3')" x | cut -f4 -d$'\t' | sort -n | tail
+
+4. To match the speed you have chosen to the correct value in the 'x' file, simply search for it in it. Then go up a little on the block where you found him. The value -n is the single value that begins before the block start. If you have multiple blocks at the same speed, choose the lowest value for -n
+
+*/
+
+const char *module_extra_tuningdb_block (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const char *extra_tuningdb_block =
+    "DEVICE_TYPE_CPU                                 *       9300    1       N       A\n"
+    "DEVICE_TYPE_GPU                                 *       9300    1       N       A\n"
+    "GeForce_GTX_980                                 *       9300    1     128       A\n"
+    "GeForce_GTX_1080                                *       9300    1     256       A\n"
+    "GeForce_RTX_2080_Ti                             *       9300    1     528       A\n"
+    "GeForce_RTX_3060_Ti                             *       9300    1     256       A\n"
+    "GeForce_RTX_3070                                *       9300    1     368       A\n"
+    "GeForce_RTX_3090                                *       9300    1     984       A\n"
+    "ALIAS_AMD_RX480                                 *       9300    1     232       A\n"
+    "ALIAS_AMD_Vega64                                *       9300    1     440       A\n"
+    "ALIAS_AMD_MI100                                 *       9300    1    1000       A\n"
+    "ALIAS_AMD_RX6900XT                              *       9300    1     720       A\n"
+  ;
+
+  return extra_tuningdb_block;
+}
+
 void module_init (module_ctx_t *module_ctx)
 {
   module_ctx->module_context_size             = MODULE_CONTEXT_SIZE_CURRENT;
@@ -356,6 +397,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -365,6 +407,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = module_extra_buffer_size;
   module_ctx->module_extra_tmp_size           = module_extra_tmp_size;
+  module_ctx->module_extra_tuningdb_block     = module_extra_tuningdb_block;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_09400.c b/src/modules/module_09400.c
index 4870920a6..57af217e2 100644
--- a/src/modules/module_09400.c
+++ b/src/modules/module_09400.c
@@ -262,6 +262,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -271,6 +272,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_09500.c b/src/modules/module_09500.c
index 49a8e7546..151ac6e8b 100644
--- a/src/modules/module_09500.c
+++ b/src/modules/module_09500.c
@@ -283,6 +283,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = module_deep_comp_kernel;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -292,6 +293,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_09600.c b/src/modules/module_09600.c
index 8ed0888cb..1c3a47a54 100644
--- a/src/modules/module_09600.c
+++ b/src/modules/module_09600.c
@@ -23,6 +23,7 @@ static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
                                   | OPTI_TYPE_USES_BITS_64
                                   | OPTI_TYPE_SLOW_HASH_SIMD_LOOP;
 static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
+                                  | OPTS_TYPE_MP_MULTI_DISABLE
                                   | OPTS_TYPE_DEEP_COMP_KERNEL;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
 static const char *ST_PASS        = "hashcat";
@@ -315,6 +316,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = module_deep_comp_kernel;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -324,6 +326,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_09700.c b/src/modules/module_09700.c
index cad911186..241d21704 100644
--- a/src/modules/module_09700.c
+++ b/src/modules/module_09700.c
@@ -69,21 +69,13 @@ char *module_jit_build_options (MAYBE_UNUSED const hashconfig_t *hashconfig, MAY
   }
   else if (device_param->opencl_device_type & CL_DEVICE_TYPE_GPU)
   {
-    if (device_param->opencl_device_vendor_id == VENDOR_ID_INTEL_SDK)
+    if (device_param->device_local_mem_size < 49152)
     {
-      native_threads = 8;
-    }
-    else if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD)
-    {
-      native_threads = 64;
-    }
-    else if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD_USE_HIP)
-    {
-      native_threads = 64;
+      native_threads = MIN (device_param->kernel_preferred_wgs_multiple, 32); // We can't just set 32, because Intel GPU need 8
     }
     else
     {
-      native_threads = 32;
+      native_threads = device_param->kernel_preferred_wgs_multiple;
     }
   }
 
@@ -243,6 +235,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -252,6 +245,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_09710.c b/src/modules/module_09710.c
index 033f77ea0..673bf38cf 100644
--- a/src/modules/module_09710.c
+++ b/src/modules/module_09710.c
@@ -69,21 +69,13 @@ char *module_jit_build_options (MAYBE_UNUSED const hashconfig_t *hashconfig, MAY
   }
   else if (device_param->opencl_device_type & CL_DEVICE_TYPE_GPU)
   {
-    if (device_param->opencl_device_vendor_id == VENDOR_ID_INTEL_SDK)
+    if (device_param->device_local_mem_size < 49152)
     {
-      native_threads = 8;
-    }
-    else if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD)
-    {
-      native_threads = 64;
-    }
-    else if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD_USE_HIP)
-    {
-      native_threads = 64;
+      native_threads = MIN (device_param->kernel_preferred_wgs_multiple, 32); // We can't just set 32, because Intel GPU need 8
     }
     else
     {
-      native_threads = 32;
+      native_threads = device_param->kernel_preferred_wgs_multiple;
     }
   }
 
@@ -262,6 +254,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -271,6 +264,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = module_forced_outfile_format;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_09720.c b/src/modules/module_09720.c
index 04e99201f..196b2ba8d 100644
--- a/src/modules/module_09720.c
+++ b/src/modules/module_09720.c
@@ -70,21 +70,13 @@ char *module_jit_build_options (MAYBE_UNUSED const hashconfig_t *hashconfig, MAY
   }
   else if (device_param->opencl_device_type & CL_DEVICE_TYPE_GPU)
   {
-    if (device_param->opencl_device_vendor_id == VENDOR_ID_INTEL_SDK)
+    if (device_param->device_local_mem_size < 49152)
     {
-      native_threads = 8;
-    }
-    else if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD)
-    {
-      native_threads = 64;
-    }
-    else if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD_USE_HIP)
-    {
-      native_threads = 64;
+      native_threads = MIN (device_param->kernel_preferred_wgs_multiple, 32); // We can't just set 32, because Intel GPU need 8
     }
     else
     {
-      native_threads = 32;
+      native_threads = device_param->kernel_preferred_wgs_multiple;
     }
   }
 
@@ -275,6 +267,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -284,6 +277,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_09800.c b/src/modules/module_09800.c
index 2eb7fab05..7768a09ac 100644
--- a/src/modules/module_09800.c
+++ b/src/modules/module_09800.c
@@ -71,21 +71,13 @@ char *module_jit_build_options (MAYBE_UNUSED const hashconfig_t *hashconfig, MAY
   }
   else if (device_param->opencl_device_type & CL_DEVICE_TYPE_GPU)
   {
-    if (device_param->opencl_device_vendor_id == VENDOR_ID_INTEL_SDK)
+    if (device_param->device_local_mem_size < 49152)
     {
-      native_threads = 8;
-    }
-    else if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD)
-    {
-      native_threads = 64;
-    }
-    else if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD_USE_HIP)
-    {
-      native_threads = 64;
+      native_threads = MIN (device_param->kernel_preferred_wgs_multiple, 32); // We can't just set 32, because Intel GPU need 8
     }
     else
     {
-      native_threads = 32;
+      native_threads = device_param->kernel_preferred_wgs_multiple;
     }
   }
 
@@ -302,6 +294,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -311,6 +304,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_09810.c b/src/modules/module_09810.c
index 2a1074b2c..3b6dbf2ab 100644
--- a/src/modules/module_09810.c
+++ b/src/modules/module_09810.c
@@ -70,21 +70,13 @@ char *module_jit_build_options (MAYBE_UNUSED const hashconfig_t *hashconfig, MAY
   }
   else if (device_param->opencl_device_type & CL_DEVICE_TYPE_GPU)
   {
-    if (device_param->opencl_device_vendor_id == VENDOR_ID_INTEL_SDK)
+    if (device_param->device_local_mem_size < 49152)
     {
-      native_threads = 8;
-    }
-    else if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD)
-    {
-      native_threads = 64;
-    }
-    else if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD_USE_HIP)
-    {
-      native_threads = 64;
+      native_threads = MIN (device_param->kernel_preferred_wgs_multiple, 32); // We can't just set 32, because Intel GPU need 8
     }
     else
     {
-      native_threads = 32;
+      native_threads = device_param->kernel_preferred_wgs_multiple;
     }
   }
 
@@ -320,6 +312,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -329,6 +322,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = module_forced_outfile_format;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_09820.c b/src/modules/module_09820.c
index ea3dfe22b..56f0c464e 100644
--- a/src/modules/module_09820.c
+++ b/src/modules/module_09820.c
@@ -72,21 +72,13 @@ char *module_jit_build_options (MAYBE_UNUSED const hashconfig_t *hashconfig, MAY
   }
   else if (device_param->opencl_device_type & CL_DEVICE_TYPE_GPU)
   {
-    if (device_param->opencl_device_vendor_id == VENDOR_ID_INTEL_SDK)
+    if (device_param->device_local_mem_size < 49152)
     {
-      native_threads = 8;
-    }
-    else if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD)
-    {
-      native_threads = 64;
-    }
-    else if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD_USE_HIP)
-    {
-      native_threads = 64;
+      native_threads = MIN (device_param->kernel_preferred_wgs_multiple, 32); // We can't just set 32, because Intel GPU need 8
     }
     else
     {
-      native_threads = 32;
+      native_threads = device_param->kernel_preferred_wgs_multiple;
     }
   }
 
@@ -342,6 +334,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -351,6 +344,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_09900.c b/src/modules/module_09900.c
index 1cf0bf948..73c9b3b4a 100644
--- a/src/modules/module_09900.c
+++ b/src/modules/module_09900.c
@@ -118,6 +118,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -127,6 +128,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_10000.c b/src/modules/module_10000.c
index c2e4db394..8c733d451 100644
--- a/src/modules/module_10000.c
+++ b/src/modules/module_10000.c
@@ -222,6 +222,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -231,6 +232,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_10100.c b/src/modules/module_10100.c
index c5cafc8c9..4b6dfb637 100644
--- a/src/modules/module_10100.c
+++ b/src/modules/module_10100.c
@@ -138,6 +138,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -147,6 +148,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_10200.c b/src/modules/module_10200.c
index cd74cc1f1..25b11916e 100644
--- a/src/modules/module_10200.c
+++ b/src/modules/module_10200.c
@@ -181,6 +181,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -190,6 +191,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_10300.c b/src/modules/module_10300.c
index 723a3ed01..c69ea741c 100644
--- a/src/modules/module_10300.c
+++ b/src/modules/module_10300.c
@@ -183,6 +183,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -192,6 +193,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_10400.c b/src/modules/module_10400.c
index 77416f5ce..1c8144605 100644
--- a/src/modules/module_10400.c
+++ b/src/modules/module_10400.c
@@ -76,21 +76,13 @@ char *module_jit_build_options (MAYBE_UNUSED const hashconfig_t *hashconfig, MAY
   }
   else if (device_param->opencl_device_type & CL_DEVICE_TYPE_GPU)
   {
-    if (device_param->opencl_device_vendor_id == VENDOR_ID_INTEL_SDK)
+    if (device_param->device_local_mem_size < 49152)
     {
-      native_threads = 8;
-    }
-    else if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD)
-    {
-      native_threads = 64;
-    }
-    else if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD_USE_HIP)
-    {
-      native_threads = 64;
+      native_threads = MIN (device_param->kernel_preferred_wgs_multiple, 32); // We can't just set 32, because Intel GPU need 8
     }
     else
     {
-      native_threads = 32;
+      native_threads = device_param->kernel_preferred_wgs_multiple;
     }
   }
 
@@ -336,6 +328,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -345,6 +338,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_10410.c b/src/modules/module_10410.c
index b2c98363f..643f7a0dd 100644
--- a/src/modules/module_10410.c
+++ b/src/modules/module_10410.c
@@ -77,21 +77,13 @@ char *module_jit_build_options (MAYBE_UNUSED const hashconfig_t *hashconfig, MAY
   }
   else if (device_param->opencl_device_type & CL_DEVICE_TYPE_GPU)
   {
-    if (device_param->opencl_device_vendor_id == VENDOR_ID_INTEL_SDK)
+    if (device_param->device_local_mem_size < 49152)
     {
-      native_threads = 8;
-    }
-    else if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD)
-    {
-      native_threads = 64;
-    }
-    else if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD_USE_HIP)
-    {
-      native_threads = 64;
+      native_threads = MIN (device_param->kernel_preferred_wgs_multiple, 32); // We can't just set 32, because Intel GPU need 8
     }
     else
     {
-      native_threads = 32;
+      native_threads = device_param->kernel_preferred_wgs_multiple;
     }
   }
 
@@ -356,6 +348,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -365,6 +358,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = module_forced_outfile_format;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_10420.c b/src/modules/module_10420.c
index 23e537bf5..084b33334 100644
--- a/src/modules/module_10420.c
+++ b/src/modules/module_10420.c
@@ -76,21 +76,13 @@ char *module_jit_build_options (MAYBE_UNUSED const hashconfig_t *hashconfig, MAY
   }
   else if (device_param->opencl_device_type & CL_DEVICE_TYPE_GPU)
   {
-    if (device_param->opencl_device_vendor_id == VENDOR_ID_INTEL_SDK)
+    if (device_param->device_local_mem_size < 49152)
     {
-      native_threads = 8;
-    }
-    else if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD)
-    {
-      native_threads = 64;
-    }
-    else if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD_USE_HIP)
-    {
-      native_threads = 64;
+      native_threads = MIN (device_param->kernel_preferred_wgs_multiple, 32); // We can't just set 32, because Intel GPU need 8
     }
     else
     {
-      native_threads = 32;
+      native_threads = device_param->kernel_preferred_wgs_multiple;
     }
   }
 
@@ -371,6 +363,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -380,6 +373,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_10500.c b/src/modules/module_10500.c
index 80a8478ef..b1678608e 100644
--- a/src/modules/module_10500.c
+++ b/src/modules/module_10500.c
@@ -100,21 +100,13 @@ char *module_jit_build_options (MAYBE_UNUSED const hashconfig_t *hashconfig, MAY
   }
   else if (device_param->opencl_device_type & CL_DEVICE_TYPE_GPU)
   {
-    if (device_param->opencl_device_vendor_id == VENDOR_ID_INTEL_SDK)
+    if (device_param->device_local_mem_size < 49152)
     {
-      native_threads = 8;
-    }
-    else if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD)
-    {
-      native_threads = 64;
-    }
-    else if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD_USE_HIP)
-    {
-      native_threads = 64;
+      native_threads = MIN (device_param->kernel_preferred_wgs_multiple, 32); // We can't just set 32, because Intel GPU need 8
     }
     else
     {
-      native_threads = 32;
+      native_threads = device_param->kernel_preferred_wgs_multiple;
     }
   }
 
@@ -462,6 +454,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -471,6 +464,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_10600.c b/src/modules/module_10600.c
index 77d0ff1d7..5fd68c99a 100644
--- a/src/modules/module_10600.c
+++ b/src/modules/module_10600.c
@@ -293,6 +293,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -302,6 +303,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_10700.c b/src/modules/module_10700.c
index 5c801b3c9..a8c2a9fbe 100644
--- a/src/modules/module_10700.c
+++ b/src/modules/module_10700.c
@@ -319,6 +319,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -328,6 +329,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_10800.c b/src/modules/module_10800.c
index 1765bddac..7878b920c 100644
--- a/src/modules/module_10800.c
+++ b/src/modules/module_10800.c
@@ -204,6 +204,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -213,6 +214,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_10810.c b/src/modules/module_10810.c
index 10d1443f4..ed1bf9d05 100644
--- a/src/modules/module_10810.c
+++ b/src/modules/module_10810.c
@@ -230,6 +230,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -239,6 +240,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_10820.c b/src/modules/module_10820.c
index 82987fe39..af8ebfe7c 100644
--- a/src/modules/module_10820.c
+++ b/src/modules/module_10820.c
@@ -230,6 +230,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -239,6 +240,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_10830.c b/src/modules/module_10830.c
index f431762f8..d31ab900f 100644
--- a/src/modules/module_10830.c
+++ b/src/modules/module_10830.c
@@ -231,6 +231,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -240,6 +241,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_10840.c b/src/modules/module_10840.c
index f60d3ea13..582dbe00b 100644
--- a/src/modules/module_10840.c
+++ b/src/modules/module_10840.c
@@ -231,6 +231,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -240,6 +241,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_10870.c b/src/modules/module_10870.c
index 047c67242..9b5be2b09 100644
--- a/src/modules/module_10870.c
+++ b/src/modules/module_10870.c
@@ -205,6 +205,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -214,6 +215,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_10900.c b/src/modules/module_10900.c
index efde01301..eaacfe289 100644
--- a/src/modules/module_10900.c
+++ b/src/modules/module_10900.c
@@ -227,6 +227,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -236,6 +237,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_10901.c b/src/modules/module_10901.c
index 47049576a..b61ba697d 100644
--- a/src/modules/module_10901.c
+++ b/src/modules/module_10901.c
@@ -202,6 +202,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -211,6 +212,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_11000.c b/src/modules/module_11000.c
index 7b9398299..597efb400 100644
--- a/src/modules/module_11000.c
+++ b/src/modules/module_11000.c
@@ -141,6 +141,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -150,6 +151,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_11100.c b/src/modules/module_11100.c
index 7d76f5eff..063e5a899 100644
--- a/src/modules/module_11100.c
+++ b/src/modules/module_11100.c
@@ -176,6 +176,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -185,6 +186,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_11200.c b/src/modules/module_11200.c
index bcfd8a2ba..4acadcba0 100644
--- a/src/modules/module_11200.c
+++ b/src/modules/module_11200.c
@@ -140,6 +140,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -149,6 +150,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_11300.c b/src/modules/module_11300.c
index 981a0b471..db169998e 100644
--- a/src/modules/module_11300.c
+++ b/src/modules/module_11300.c
@@ -27,6 +27,7 @@ static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
                                   | OPTS_TYPE_ST_HEX
                                   | OPTS_TYPE_ST_ADD80
                                   | OPTS_TYPE_HASH_COPY
+                                  | OPTS_TYPE_MP_MULTI_DISABLE
                                   | OPTS_TYPE_DEEP_COMP_KERNEL;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
 static const char *ST_PASS        = "hashcat";
@@ -285,6 +286,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = module_deep_comp_kernel;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -294,6 +296,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_11400.c b/src/modules/module_11400.c
index f5ff1eb64..0aa1fb6b9 100644
--- a/src/modules/module_11400.c
+++ b/src/modules/module_11400.c
@@ -433,6 +433,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -442,6 +443,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_11500.c b/src/modules/module_11500.c
index 7b012979d..0e25a0a1d 100644
--- a/src/modules/module_11500.c
+++ b/src/modules/module_11500.c
@@ -114,6 +114,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -123,6 +124,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_11600.c b/src/modules/module_11600.c
index 25fe732a3..4d42afade 100644
--- a/src/modules/module_11600.c
+++ b/src/modules/module_11600.c
@@ -80,7 +80,7 @@ typedef struct seven_zip_hook_salt
 
   u8  data_type;
 
-  u32 data_buf[81882];
+  u32 data_buf[0x200000];
   u32 data_len;
 
   u32 unpack_size;
@@ -130,8 +130,8 @@ bool module_hook_extra_param_init (MAYBE_UNUSED const hashconfig_t *hashconfig,
 {
   seven_zip_hook_extra_t *seven_zip_hook_extra = (seven_zip_hook_extra_t *) hook_extra_param;
 
-  #define AESSIZE  320 * 1024
-  #define UNPSIZE 9766 * 1024 // or actually maximum is 9999999
+  #define AESSIZE 8 * 1024 * 1024
+  #define UNPSIZE 9999999
 
   seven_zip_hook_extra->aes = hccalloc (backend_ctx->backend_devices_cnt, sizeof (void *));
 
@@ -500,7 +500,7 @@ int module_hash_decode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSE
 
   token.sep[10]     = '$';
   token.len_min[10] = 2;
-  token.len_max[10] = 655056;
+  token.len_max[10] = 0x200000 * 4 * 2;
   token.attr[10]    = TOKEN_ATTR_VERIFY_LENGTH;
 
   const int rc_tokenizer = input_tokenizer ((const u8 *) line_buf, line_len, &token);
@@ -592,7 +592,7 @@ int module_hash_decode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSE
 
   if ((data_len * 2) != data_buf_len) return (PARSER_SALT_VALUE);
 
-  if (data_len > 327528) return (PARSER_SALT_VALUE);
+  if (data_len > 0x200000 * 4) return (PARSER_SALT_VALUE);
 
   if (unpack_size > data_len) return (PARSER_SALT_VALUE);
 
@@ -791,6 +791,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -800,6 +801,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_11700.c b/src/modules/module_11700.c
index ebf903a55..8b1b7c714 100644
--- a/src/modules/module_11700.c
+++ b/src/modules/module_11700.c
@@ -139,6 +139,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -148,6 +149,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_11750.c b/src/modules/module_11750.c
index f2302db35..7106f82a4 100644
--- a/src/modules/module_11750.c
+++ b/src/modules/module_11750.c
@@ -166,6 +166,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -175,6 +176,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_11760.c b/src/modules/module_11760.c
index 5c574e469..a3f5bc0f2 100644
--- a/src/modules/module_11760.c
+++ b/src/modules/module_11760.c
@@ -166,6 +166,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -175,6 +176,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_11800.c b/src/modules/module_11800.c
index 3c83bc0c0..819ad144b 100644
--- a/src/modules/module_11800.c
+++ b/src/modules/module_11800.c
@@ -163,6 +163,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -172,6 +173,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_11850.c b/src/modules/module_11850.c
index 6bf1853b4..1be2fcb41 100644
--- a/src/modules/module_11850.c
+++ b/src/modules/module_11850.c
@@ -190,6 +190,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -199,6 +200,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_11860.c b/src/modules/module_11860.c
index 67a5ff618..7a2e10d81 100644
--- a/src/modules/module_11860.c
+++ b/src/modules/module_11860.c
@@ -190,6 +190,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -199,6 +200,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_11900.c b/src/modules/module_11900.c
index d75ffd168..240a0496f 100644
--- a/src/modules/module_11900.c
+++ b/src/modules/module_11900.c
@@ -191,6 +191,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -200,6 +201,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_12000.c b/src/modules/module_12000.c
index c878760d3..26e3cf586 100644
--- a/src/modules/module_12000.c
+++ b/src/modules/module_12000.c
@@ -196,6 +196,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -205,6 +206,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_12001.c b/src/modules/module_12001.c
index 07c622b23..b92a7ab29 100644
--- a/src/modules/module_12001.c
+++ b/src/modules/module_12001.c
@@ -169,6 +169,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -178,6 +179,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_12100.c b/src/modules/module_12100.c
index 9407b2c23..daf6df9ff 100644
--- a/src/modules/module_12100.c
+++ b/src/modules/module_12100.c
@@ -201,6 +201,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -210,6 +211,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_12200.c b/src/modules/module_12200.c
index 3f6b57821..6d6720585 100644
--- a/src/modules/module_12200.c
+++ b/src/modules/module_12200.c
@@ -222,6 +222,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -231,6 +232,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_12300.c b/src/modules/module_12300.c
index a7b7e7fde..b25ed9c1e 100644
--- a/src/modules/module_12300.c
+++ b/src/modules/module_12300.c
@@ -178,6 +178,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -187,6 +188,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_12400.c b/src/modules/module_12400.c
index 425a2252b..cae2256cb 100644
--- a/src/modules/module_12400.c
+++ b/src/modules/module_12400.c
@@ -220,6 +220,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -229,6 +230,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_12500.c b/src/modules/module_12500.c
index fda848e0a..d526fe5c4 100644
--- a/src/modules/module_12500.c
+++ b/src/modules/module_12500.c
@@ -88,15 +88,6 @@ u32 module_kernel_loops_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_
   return kernel_loops_max;
 }
 
-u32 module_kernel_threads_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
-{
-  // -T 128 works slightly faster but it's free for the user to change
-
-  const u32 kernel_threads_max = (user_options->kernel_threads_chgd == true) ? user_options->kernel_threads : 128;
-
-  return kernel_threads_max;
-}
-
 u32 module_pw_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
 {
   const bool optimized_kernel = (hashconfig->opti_type & OPTI_TYPE_OPTIMIZED_KERNEL);
@@ -217,6 +208,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -226,6 +218,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
@@ -256,7 +249,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_kernel_accel_min         = MODULE_DEFAULT;
   module_ctx->module_kernel_loops_max         = module_kernel_loops_max;
   module_ctx->module_kernel_loops_min         = module_kernel_loops_min;
-  module_ctx->module_kernel_threads_max       = module_kernel_threads_max;
+  module_ctx->module_kernel_threads_max       = MODULE_DEFAULT;
   module_ctx->module_kernel_threads_min       = MODULE_DEFAULT;
   module_ctx->module_kern_type                = module_kern_type;
   module_ctx->module_kern_type_dynamic        = MODULE_DEFAULT;
diff --git a/src/modules/module_12600.c b/src/modules/module_12600.c
index bf689dfdb..e0dcd2835 100644
--- a/src/modules/module_12600.c
+++ b/src/modules/module_12600.c
@@ -222,6 +222,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -231,6 +232,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_12700.c b/src/modules/module_12700.c
index a29502fdf..7ce7e04dc 100644
--- a/src/modules/module_12700.c
+++ b/src/modules/module_12700.c
@@ -162,6 +162,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -171,6 +172,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_12800.c b/src/modules/module_12800.c
index cd2099cdc..e788f3e2d 100644
--- a/src/modules/module_12800.c
+++ b/src/modules/module_12800.c
@@ -226,6 +226,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -235,6 +236,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_12900.c b/src/modules/module_12900.c
index 4c5a9892b..20672fd9c 100644
--- a/src/modules/module_12900.c
+++ b/src/modules/module_12900.c
@@ -231,6 +231,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -240,6 +241,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_13000.c b/src/modules/module_13000.c
index ab389431d..9402de218 100644
--- a/src/modules/module_13000.c
+++ b/src/modules/module_13000.c
@@ -265,6 +265,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -274,6 +275,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_13100.c b/src/modules/module_13100.c
index fac5cb24c..4e42abd29 100644
--- a/src/modules/module_13100.c
+++ b/src/modules/module_13100.c
@@ -64,35 +64,13 @@ char *module_jit_build_options (MAYBE_UNUSED const hashconfig_t *hashconfig, MAY
   }
   else if (device_param->opencl_device_type & CL_DEVICE_TYPE_GPU)
   {
-    if (device_param->opencl_device_vendor_id == VENDOR_ID_INTEL_SDK)
+    if (device_param->device_local_mem_size < 49152)
     {
-      native_threads = 8;
-    }
-    else if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD)
-    {
-      if (device_param->device_local_mem_size < 49152)
-      {
-        native_threads = 32;
-      }
-      else
-      {
-        native_threads = 64;
-      }
-    }
-    else if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD_USE_HIP)
-    {
-      if (device_param->device_local_mem_size < 49152)
-      {
-        native_threads = 32;
-      }
-      else
-      {
-        native_threads = 64;
-      }
+      native_threads = MIN (device_param->kernel_preferred_wgs_multiple, 32); // We can't just set 32, because Intel GPU need 8
     }
     else
     {
-      native_threads = 32;
+      native_threads = device_param->kernel_preferred_wgs_multiple;
     }
   }
 
@@ -289,6 +267,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -298,6 +277,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_13200.c b/src/modules/module_13200.c
index e86917e99..fe2d73076 100644
--- a/src/modules/module_13200.c
+++ b/src/modules/module_13200.c
@@ -191,6 +191,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -200,6 +201,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_13300.c b/src/modules/module_13300.c
index 98782b227..cbbce7fb7 100644
--- a/src/modules/module_13300.c
+++ b/src/modules/module_13300.c
@@ -115,6 +115,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -124,6 +125,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_13400.c b/src/modules/module_13400.c
index 6fce19d15..c15384239 100644
--- a/src/modules/module_13400.c
+++ b/src/modules/module_13400.c
@@ -655,6 +655,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -664,6 +665,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_13500.c b/src/modules/module_13500.c
index 0ca00a2e8..e458d0e8f 100644
--- a/src/modules/module_13500.c
+++ b/src/modules/module_13500.c
@@ -209,6 +209,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -218,6 +219,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_13600.c b/src/modules/module_13600.c
index ec1c4e6ef..d2cba95bc 100644
--- a/src/modules/module_13600.c
+++ b/src/modules/module_13600.c
@@ -9,6 +9,7 @@
 #include "bitops.h"
 #include "convert.h"
 #include "shared.h"
+#include "memory.h"
 
 static const u32   ATTACK_EXEC    = ATTACK_EXEC_OUTSIDE_KERNEL;
 static const u32   DGST_POS0      = 0;
@@ -61,7 +62,7 @@ typedef struct zip2
   u32 verify_bytes;
   u32 compress_length;
   u32 data_len;
-  u32 data_buf[0x4000000];
+  u32 data_buf[0x200000];
   u32 auth_len;
   u32 auth_buf[4];
 
@@ -166,7 +167,7 @@ int module_hash_decode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSE
                    | TOKEN_ATTR_VERIFY_HEX;
 
   token.len_min[7] = 0;
-  token.len_max[7] = 0x4000000 * 4 * 2;
+  token.len_max[7] = 0x200000 * 4 * 2;
   token.sep[7]     = '*';
   token.attr[7]    = TOKEN_ATTR_VERIFY_LENGTH
                    | TOKEN_ATTR_VERIFY_HEX;
@@ -365,7 +366,7 @@ int module_hash_encode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSE
 
   const u32 data_len = zip2->data_len;
 
-  char data_tmp[16384 + 1] = { 0 };
+  char *data_tmp = (char *) hccalloc (1, (data_len * 2) + 1);
 
   for (u32 i = 0, j = 0; i < data_len; i += 1, j += 2)
   {
@@ -397,6 +398,8 @@ int module_hash_encode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSE
     auth_tmp,
     SIGNATURE_ZIP2_STOP);
 
+  hcfree (data_tmp);
+
   return line_len;
 }
 
@@ -412,6 +415,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -421,6 +425,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_13711.c b/src/modules/module_13711.c
index ffab1e130..314d0c5d3 100644
--- a/src/modules/module_13711.c
+++ b/src/modules/module_13711.c
@@ -27,6 +27,7 @@ static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
 static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
                                   | OPTS_TYPE_BINARY_HASHFILE
                                   | OPTS_TYPE_LOOP_EXTENDED
+                                  | OPTS_TYPE_MP_MULTI_DISABLE
                                   | OPTS_TYPE_COPY_TMPS;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
 static const char *ST_PASS        = "hashcat";
@@ -298,6 +299,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = module_build_plain_postprocess;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -307,6 +309,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_13712.c b/src/modules/module_13712.c
index 91cf21029..1e53d8945 100644
--- a/src/modules/module_13712.c
+++ b/src/modules/module_13712.c
@@ -27,6 +27,7 @@ static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
 static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
                                   | OPTS_TYPE_BINARY_HASHFILE
                                   | OPTS_TYPE_LOOP_EXTENDED
+                                  | OPTS_TYPE_MP_MULTI_DISABLE
                                   | OPTS_TYPE_COPY_TMPS;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
 static const char *ST_PASS        = "hashcat";
@@ -298,6 +299,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = module_build_plain_postprocess;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -307,6 +309,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_13713.c b/src/modules/module_13713.c
index ca15c4a01..3efdbf0a5 100644
--- a/src/modules/module_13713.c
+++ b/src/modules/module_13713.c
@@ -27,6 +27,7 @@ static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
 static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
                                   | OPTS_TYPE_BINARY_HASHFILE
                                   | OPTS_TYPE_LOOP_EXTENDED
+                                  | OPTS_TYPE_MP_MULTI_DISABLE
                                   | OPTS_TYPE_COPY_TMPS;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
 static const char *ST_PASS        = "hashcat";
@@ -298,6 +299,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = module_build_plain_postprocess;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -307,6 +309,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_13721.c b/src/modules/module_13721.c
index 72162ba85..3d2d450ca 100644
--- a/src/modules/module_13721.c
+++ b/src/modules/module_13721.c
@@ -28,6 +28,7 @@ static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
 static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
                                   | OPTS_TYPE_BINARY_HASHFILE
                                   | OPTS_TYPE_LOOP_EXTENDED
+                                  | OPTS_TYPE_MP_MULTI_DISABLE
                                   | OPTS_TYPE_COPY_TMPS;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
 static const char *ST_PASS        = "hashcat";
@@ -308,6 +309,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = module_build_plain_postprocess;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -317,6 +319,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_13722.c b/src/modules/module_13722.c
index d595a69d8..1e729ff85 100644
--- a/src/modules/module_13722.c
+++ b/src/modules/module_13722.c
@@ -28,6 +28,7 @@ static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
 static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
                                   | OPTS_TYPE_BINARY_HASHFILE
                                   | OPTS_TYPE_LOOP_EXTENDED
+                                  | OPTS_TYPE_MP_MULTI_DISABLE
                                   | OPTS_TYPE_COPY_TMPS;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
 static const char *ST_PASS        = "hashcat";
@@ -308,6 +309,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = module_build_plain_postprocess;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -317,6 +319,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_13723.c b/src/modules/module_13723.c
index 65ed55545..bf91db634 100644
--- a/src/modules/module_13723.c
+++ b/src/modules/module_13723.c
@@ -28,6 +28,7 @@ static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
 static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
                                   | OPTS_TYPE_BINARY_HASHFILE
                                   | OPTS_TYPE_LOOP_EXTENDED
+                                  | OPTS_TYPE_MP_MULTI_DISABLE
                                   | OPTS_TYPE_COPY_TMPS;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
 static const char *ST_PASS        = "hashcat";
@@ -308,6 +309,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = module_build_plain_postprocess;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -317,6 +319,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_13731.c b/src/modules/module_13731.c
index da13dd99f..8f5745f86 100644
--- a/src/modules/module_13731.c
+++ b/src/modules/module_13731.c
@@ -27,6 +27,7 @@ static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
 static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
                                   | OPTS_TYPE_BINARY_HASHFILE
                                   | OPTS_TYPE_LOOP_EXTENDED
+                                  | OPTS_TYPE_MP_MULTI_DISABLE
                                   | OPTS_TYPE_COPY_TMPS;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
 static const char *ST_PASS        = "hashcat";
@@ -287,6 +288,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = module_build_plain_postprocess;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -296,6 +298,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_13732.c b/src/modules/module_13732.c
index 1945e3b9a..bdb47391a 100644
--- a/src/modules/module_13732.c
+++ b/src/modules/module_13732.c
@@ -27,6 +27,7 @@ static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
 static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
                                   | OPTS_TYPE_BINARY_HASHFILE
                                   | OPTS_TYPE_LOOP_EXTENDED
+                                  | OPTS_TYPE_MP_MULTI_DISABLE
                                   | OPTS_TYPE_COPY_TMPS;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
 static const char *ST_PASS        = "hashcat";
@@ -287,6 +288,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = module_build_plain_postprocess;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -296,6 +298,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_13733.c b/src/modules/module_13733.c
index 72443b7d3..c4361ac5d 100644
--- a/src/modules/module_13733.c
+++ b/src/modules/module_13733.c
@@ -27,6 +27,7 @@ static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
 static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
                                   | OPTS_TYPE_BINARY_HASHFILE
                                   | OPTS_TYPE_LOOP_EXTENDED
+                                  | OPTS_TYPE_MP_MULTI_DISABLE
                                   | OPTS_TYPE_COPY_TMPS;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
 static const char *ST_PASS        = "hashcat";
@@ -287,6 +288,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = module_build_plain_postprocess;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -296,6 +298,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_13741.c b/src/modules/module_13741.c
index 7ad01f942..0469a99c5 100644
--- a/src/modules/module_13741.c
+++ b/src/modules/module_13741.c
@@ -27,6 +27,7 @@ static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
 static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
                                   | OPTS_TYPE_BINARY_HASHFILE
                                   | OPTS_TYPE_LOOP_EXTENDED
+                                  | OPTS_TYPE_MP_MULTI_DISABLE
                                   | OPTS_TYPE_KEYBOARD_MAPPING
                                   | OPTS_TYPE_COPY_TMPS;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
@@ -300,6 +301,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = module_build_plain_postprocess;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -309,6 +311,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_13742.c b/src/modules/module_13742.c
index 308fa9084..98ce513be 100644
--- a/src/modules/module_13742.c
+++ b/src/modules/module_13742.c
@@ -27,6 +27,7 @@ static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
 static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
                                   | OPTS_TYPE_BINARY_HASHFILE
                                   | OPTS_TYPE_LOOP_EXTENDED
+                                  | OPTS_TYPE_MP_MULTI_DISABLE
                                   | OPTS_TYPE_KEYBOARD_MAPPING
                                   | OPTS_TYPE_COPY_TMPS;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
@@ -300,6 +301,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = module_build_plain_postprocess;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -309,6 +311,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_13743.c b/src/modules/module_13743.c
index 10f808d9e..5a1b3b5d4 100644
--- a/src/modules/module_13743.c
+++ b/src/modules/module_13743.c
@@ -27,6 +27,7 @@ static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
 static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
                                   | OPTS_TYPE_BINARY_HASHFILE
                                   | OPTS_TYPE_LOOP_EXTENDED
+                                  | OPTS_TYPE_MP_MULTI_DISABLE
                                   | OPTS_TYPE_KEYBOARD_MAPPING
                                   | OPTS_TYPE_COPY_TMPS;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
@@ -300,6 +301,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = module_build_plain_postprocess;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -309,6 +311,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_13751.c b/src/modules/module_13751.c
index 7ea17284f..a1c30ea77 100644
--- a/src/modules/module_13751.c
+++ b/src/modules/module_13751.c
@@ -27,6 +27,7 @@ static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
 static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
                                   | OPTS_TYPE_BINARY_HASHFILE
                                   | OPTS_TYPE_LOOP_EXTENDED
+                                  | OPTS_TYPE_MP_MULTI_DISABLE
                                   | OPTS_TYPE_COPY_TMPS;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
 static const char *ST_PASS        = "hashcat";
@@ -317,6 +318,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = module_build_plain_postprocess;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -326,6 +328,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_13752.c b/src/modules/module_13752.c
index 8dcae3693..578295528 100644
--- a/src/modules/module_13752.c
+++ b/src/modules/module_13752.c
@@ -27,6 +27,7 @@ static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
 static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
                                   | OPTS_TYPE_BINARY_HASHFILE
                                   | OPTS_TYPE_LOOP_EXTENDED
+                                  | OPTS_TYPE_MP_MULTI_DISABLE
                                   | OPTS_TYPE_COPY_TMPS;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
 static const char *ST_PASS        = "hashcat";
@@ -317,6 +318,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = module_build_plain_postprocess;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -326,6 +328,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_13753.c b/src/modules/module_13753.c
index 630969373..aee841431 100644
--- a/src/modules/module_13753.c
+++ b/src/modules/module_13753.c
@@ -27,6 +27,7 @@ static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
 static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
                                   | OPTS_TYPE_BINARY_HASHFILE
                                   | OPTS_TYPE_LOOP_EXTENDED
+                                  | OPTS_TYPE_MP_MULTI_DISABLE
                                   | OPTS_TYPE_COPY_TMPS;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
 static const char *ST_PASS        = "hashcat";
@@ -317,6 +318,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = module_build_plain_postprocess;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -326,6 +328,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_13761.c b/src/modules/module_13761.c
index 5193f0a29..71d156536 100644
--- a/src/modules/module_13761.c
+++ b/src/modules/module_13761.c
@@ -27,6 +27,7 @@ static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
 static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
                                   | OPTS_TYPE_BINARY_HASHFILE
                                   | OPTS_TYPE_LOOP_EXTENDED
+                                  | OPTS_TYPE_MP_MULTI_DISABLE
                                   | OPTS_TYPE_KEYBOARD_MAPPING
                                   | OPTS_TYPE_COPY_TMPS;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
@@ -319,6 +320,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = module_build_plain_postprocess;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -328,6 +330,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_13762.c b/src/modules/module_13762.c
index 4780398ef..c627cf79d 100644
--- a/src/modules/module_13762.c
+++ b/src/modules/module_13762.c
@@ -27,6 +27,7 @@ static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
 static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
                                   | OPTS_TYPE_BINARY_HASHFILE
                                   | OPTS_TYPE_LOOP_EXTENDED
+                                  | OPTS_TYPE_MP_MULTI_DISABLE
                                   | OPTS_TYPE_KEYBOARD_MAPPING
                                   | OPTS_TYPE_COPY_TMPS;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
@@ -319,6 +320,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = module_build_plain_postprocess;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -328,6 +330,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_13763.c b/src/modules/module_13763.c
index 594b63519..696cdc9f7 100644
--- a/src/modules/module_13763.c
+++ b/src/modules/module_13763.c
@@ -27,6 +27,7 @@ static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
 static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
                                   | OPTS_TYPE_BINARY_HASHFILE
                                   | OPTS_TYPE_LOOP_EXTENDED
+                                  | OPTS_TYPE_MP_MULTI_DISABLE
                                   | OPTS_TYPE_KEYBOARD_MAPPING
                                   | OPTS_TYPE_COPY_TMPS;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
@@ -319,6 +320,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = module_build_plain_postprocess;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -328,6 +330,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_13771.c b/src/modules/module_13771.c
index 2afe73f00..a3ce7908b 100644
--- a/src/modules/module_13771.c
+++ b/src/modules/module_13771.c
@@ -28,6 +28,7 @@ static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
 static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
                                   | OPTS_TYPE_BINARY_HASHFILE
                                   | OPTS_TYPE_LOOP_EXTENDED
+                                  | OPTS_TYPE_MP_MULTI_DISABLE
                                   | OPTS_TYPE_COPY_TMPS;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
 static const char *ST_PASS        = "hashcat";
@@ -311,6 +312,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = module_build_plain_postprocess;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -320,6 +322,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_13772.c b/src/modules/module_13772.c
index ab274128d..1973d6263 100644
--- a/src/modules/module_13772.c
+++ b/src/modules/module_13772.c
@@ -28,6 +28,7 @@ static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
 static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
                                   | OPTS_TYPE_BINARY_HASHFILE
                                   | OPTS_TYPE_LOOP_EXTENDED
+                                  | OPTS_TYPE_MP_MULTI_DISABLE
                                   | OPTS_TYPE_COPY_TMPS;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
 static const char *ST_PASS        = "hashcat";
@@ -311,6 +312,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = module_build_plain_postprocess;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -320,6 +322,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_13773.c b/src/modules/module_13773.c
index 157d2ea88..f350973a9 100644
--- a/src/modules/module_13773.c
+++ b/src/modules/module_13773.c
@@ -28,6 +28,7 @@ static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
 static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
                                   | OPTS_TYPE_BINARY_HASHFILE
                                   | OPTS_TYPE_LOOP_EXTENDED
+                                  | OPTS_TYPE_MP_MULTI_DISABLE
                                   | OPTS_TYPE_COPY_TMPS;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
 static const char *ST_PASS        = "hashcat";
@@ -311,6 +312,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = module_build_plain_postprocess;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -320,6 +322,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_13781.c b/src/modules/module_13781.c
index aaa4ad96f..6a89281e6 100644
--- a/src/modules/module_13781.c
+++ b/src/modules/module_13781.c
@@ -28,6 +28,7 @@ static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
 static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
                                   | OPTS_TYPE_BINARY_HASHFILE
                                   | OPTS_TYPE_LOOP_EXTENDED
+                                  | OPTS_TYPE_MP_MULTI_DISABLE
                                   | OPTS_TYPE_KEYBOARD_MAPPING
                                   | OPTS_TYPE_COPY_TMPS;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
@@ -313,6 +314,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = module_build_plain_postprocess;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -322,6 +324,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_13782.c b/src/modules/module_13782.c
index 550b64ed0..3bb5e5bc9 100644
--- a/src/modules/module_13782.c
+++ b/src/modules/module_13782.c
@@ -28,6 +28,7 @@ static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
 static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
                                   | OPTS_TYPE_BINARY_HASHFILE
                                   | OPTS_TYPE_LOOP_EXTENDED
+                                  | OPTS_TYPE_MP_MULTI_DISABLE
                                   | OPTS_TYPE_KEYBOARD_MAPPING
                                   | OPTS_TYPE_COPY_TMPS;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
@@ -313,6 +314,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = module_build_plain_postprocess;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -322,6 +324,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_13783.c b/src/modules/module_13783.c
index 155aa0cd0..d184f847f 100644
--- a/src/modules/module_13783.c
+++ b/src/modules/module_13783.c
@@ -28,6 +28,7 @@ static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
 static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
                                   | OPTS_TYPE_BINARY_HASHFILE
                                   | OPTS_TYPE_LOOP_EXTENDED
+                                  | OPTS_TYPE_MP_MULTI_DISABLE
                                   | OPTS_TYPE_KEYBOARD_MAPPING
                                   | OPTS_TYPE_COPY_TMPS;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
@@ -313,6 +314,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = module_build_plain_postprocess;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -322,6 +324,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_13800.c b/src/modules/module_13800.c
index 07f0bdcdd..165eeee09 100644
--- a/src/modules/module_13800.c
+++ b/src/modules/module_13800.c
@@ -192,6 +192,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -201,6 +202,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_13900.c b/src/modules/module_13900.c
index fd7a06131..3437914b8 100644
--- a/src/modules/module_13900.c
+++ b/src/modules/module_13900.c
@@ -142,6 +142,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -151,6 +152,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_14000.c b/src/modules/module_14000.c
index 013888bcc..16ced9e3a 100644
--- a/src/modules/module_14000.c
+++ b/src/modules/module_14000.c
@@ -91,13 +91,6 @@ char *module_jit_build_options (MAYBE_UNUSED const hashconfig_t *hashconfig, MAY
   return jit_build_options;
 }
 
-u32 module_kernel_threads_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
-{
-  const u32 kernel_threads_max = 64; // performance only optimization
-
-  return kernel_threads_max;
-}
-
 u32 module_kernel_loops_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
 {
   u32 kernel_loops_max = KERNEL_LOOPS_MAX;
@@ -261,6 +254,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -270,6 +264,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
@@ -300,7 +295,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_kernel_accel_min         = MODULE_DEFAULT;
   module_ctx->module_kernel_loops_max         = module_kernel_loops_max;
   module_ctx->module_kernel_loops_min         = module_kernel_loops_min;
-  module_ctx->module_kernel_threads_max       = module_kernel_threads_max;
+  module_ctx->module_kernel_threads_max       = MODULE_DEFAULT;
   module_ctx->module_kernel_threads_min       = MODULE_DEFAULT;
   module_ctx->module_kern_type                = module_kern_type;
   module_ctx->module_kern_type_dynamic        = MODULE_DEFAULT;
diff --git a/src/modules/module_14100.c b/src/modules/module_14100.c
index cd8d907da..04760d232 100644
--- a/src/modules/module_14100.c
+++ b/src/modules/module_14100.c
@@ -205,6 +205,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -214,6 +215,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_14400.c b/src/modules/module_14400.c
index 925fe8a37..82459db59 100644
--- a/src/modules/module_14400.c
+++ b/src/modules/module_14400.c
@@ -160,6 +160,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -169,6 +170,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_14500.c b/src/modules/module_14500.c
index ba4edab17..5eaac012a 100644
--- a/src/modules/module_14500.c
+++ b/src/modules/module_14500.c
@@ -349,6 +349,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -358,6 +359,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_14600.c b/src/modules/module_14600.c
index 0ff26ac67..3fad04c55 100644
--- a/src/modules/module_14600.c
+++ b/src/modules/module_14600.c
@@ -179,17 +179,6 @@ typedef struct luks_tmp
 
 } luks_tmp_t;
 
-u32 module_kernel_threads_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
-{
-  // the module requires a lot of registers for key schedulers on _comp kernel.
-  // it's possible, if using too many threads, there's not enough registers available, typically ending with misleading error message:
-  // cuLaunchKernel(): out of memory
-
-  const u32 kernel_threads_max = 64;
-
-  return kernel_threads_max;
-}
-
 void *module_benchmark_esalt (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
 {
   luks_t *luks = (luks_t *) hcmalloc (sizeof (luks_t));
@@ -616,6 +605,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = module_benchmark_salt;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -625,6 +615,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = module_hash_binary_count;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
@@ -655,7 +646,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_kernel_accel_min         = MODULE_DEFAULT;
   module_ctx->module_kernel_loops_max         = MODULE_DEFAULT;
   module_ctx->module_kernel_loops_min         = MODULE_DEFAULT;
-  module_ctx->module_kernel_threads_max       = module_kernel_threads_max;
+  module_ctx->module_kernel_threads_max       = MODULE_DEFAULT;
   module_ctx->module_kernel_threads_min       = MODULE_DEFAULT;
   module_ctx->module_kern_type                = module_kern_type;
   module_ctx->module_kern_type_dynamic        = module_kern_type_dynamic;
diff --git a/src/modules/module_14700.c b/src/modules/module_14700.c
index b4292ab9a..4c2773115 100644
--- a/src/modules/module_14700.c
+++ b/src/modules/module_14700.c
@@ -357,6 +357,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -366,6 +367,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_14800.c b/src/modules/module_14800.c
index a5119ee6b..db71c4ed4 100644
--- a/src/modules/module_14800.c
+++ b/src/modules/module_14800.c
@@ -25,6 +25,7 @@ static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
                                   | OPTI_TYPE_SLOW_HASH_SIMD_LOOP2;
 static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
                                   | OPTS_TYPE_ST_HEX
+                                  | OPTS_TYPE_MP_MULTI_DISABLE
                                   | OPTS_TYPE_INIT2
                                   | OPTS_TYPE_LOOP2;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
@@ -383,6 +384,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = module_benchmark_salt;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -392,6 +394,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_14900.c b/src/modules/module_14900.c
index 84a5678ee..307ab2433 100644
--- a/src/modules/module_14900.c
+++ b/src/modules/module_14900.c
@@ -126,6 +126,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -135,6 +136,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_15000.c b/src/modules/module_15000.c
index 9fae0e769..5388a34ea 100644
--- a/src/modules/module_15000.c
+++ b/src/modules/module_15000.c
@@ -235,6 +235,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -244,6 +245,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_15100.c b/src/modules/module_15100.c
index aa2cde6f8..c90b342a6 100644
--- a/src/modules/module_15100.c
+++ b/src/modules/module_15100.c
@@ -321,6 +321,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -330,6 +331,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_15200.c b/src/modules/module_15200.c
index 0985e7923..9bdb1664b 100644
--- a/src/modules/module_15200.c
+++ b/src/modules/module_15200.c
@@ -168,6 +168,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -177,6 +178,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_15300.c b/src/modules/module_15300.c
index 7f85d4471..299f13365 100644
--- a/src/modules/module_15300.c
+++ b/src/modules/module_15300.c
@@ -224,9 +224,7 @@ int module_hash_decode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSE
 
   dpapimk->context = hc_strtoul ((const char *) context_pos, NULL, 10);
 
-  // division by 4 should be fine because contents_len is either 208 or 288
-
-  for (u32 i = 0; i < dpapimk->contents_len / 4; i++)
+  for (u32 i = 0; i < dpapimk->contents_len / 8; i++)
   {
     dpapimk->contents[i] = hex_to_u32 ((const u8 *) &contents_pos[i * 8]);
 
@@ -317,7 +315,7 @@ int module_hash_encode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSE
 
   SID_tmp = (u8 *) hcmalloc ((SID_len + 1) * sizeof(u8));
 
-  for (u32 i = 0; i < (SID_len / 4) + 1; i++)
+  for (u32 i = 0; i < (SID_len / 4); i++)
   {
     u8 hex[8] = { 0 };
 
@@ -395,6 +393,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -404,6 +403,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_15400.c b/src/modules/module_15400.c
index 64edfce9f..e90b4a398 100644
--- a/src/modules/module_15400.c
+++ b/src/modules/module_15400.c
@@ -219,6 +219,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -228,6 +229,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_15500.c b/src/modules/module_15500.c
index 3964d618b..4d5adf5e3 100644
--- a/src/modules/module_15500.c
+++ b/src/modules/module_15500.c
@@ -285,6 +285,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -294,6 +295,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_15600.c b/src/modules/module_15600.c
index c7acdb8d3..e4de98c93 100644
--- a/src/modules/module_15600.c
+++ b/src/modules/module_15600.c
@@ -274,6 +274,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -283,6 +284,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_15700.c b/src/modules/module_15700.c
index cb2d939ab..09652f515 100644
--- a/src/modules/module_15700.c
+++ b/src/modules/module_15700.c
@@ -453,6 +453,47 @@ int module_hash_encode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSE
   return line_len;
 }
 
+/*
+
+Find the right -n value for your GPU:
+=====================================
+
+1. For example, to find the value for 15700, first create a valid hash for 15700 as follows:
+
+$ ./hashcat --example-hashes -m 15700 | grep Example.Hash | grep -v Format | cut -b 25- > tmp.hash.15700
+
+2. Now let it iterate through all -n values to a certain point. In this case, I'm using 200, but in general it's a value that is at least twice that of the multiprocessor. If you don't mind you can just leave it as it is, it just runs a little longer.
+
+$ export i=1; while [ $i -ne 201 ]; do echo $i; ./hashcat --quiet tmp.hash.15700 --keep-guessing --self-test-disable --markov-disable --restore-disable --outfile-autohex-disable --wordlist-autohex-disable --potfile-disable --logfile-disable --hwmon-disable --status --status-timer 1 --runtime 28 --machine-readable --optimized-kernel-enable --workload-profile 3 --hash-type 15700 --attack-mode 3 ?b?b?b?b?b?b?b --backend-devices 1 --force -n $i; i=$(($i+1)); done | tee x
+
+3. Determine the highest measured H/s speed. But don't just use the highest value. Instead, use the number that seems most stable, usually at the beginning.
+
+$ grep "$(printf 'STATUS\t3')" x | cut -f4 -d$'\t' | sort -n | tail
+
+4. To match the speed you have chosen to the correct value in the 'x' file, simply search for it in it. Then go up a little on the block where you found him. The value -n is the single value that begins before the block start. If you have multiple blocks at the same speed, choose the lowest value for -n
+
+*/
+
+const char *module_extra_tuningdb_block (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const char *extra_tuningdb_block =
+    "DEVICE_TYPE_CPU                                 *       15700   1       N       A\n"
+    "DEVICE_TYPE_GPU                                 *       15700   1       1       A\n"
+    "GeForce_GTX_980                                 *       15700   1      24       A\n"
+    "GeForce_GTX_1080                                *       15700   1      28       A\n"
+    "GeForce_RTX_2080_Ti                             *       15700   1      68       A\n"
+    "GeForce_RTX_3060_Ti                             *       15700   1      11       A\n"
+    "GeForce_RTX_3070                                *       15700   1      22       A\n"
+    "GeForce_RTX_3090                                *       15700   1      82       A\n"
+    "ALIAS_AMD_RX480                                 *       15700   1      58       A\n"
+    "ALIAS_AMD_Vega64                                *       15700   1      53       A\n"
+    "ALIAS_AMD_MI100                                 *       15700   1     120       A\n"
+    "ALIAS_AMD_RX6900XT                              *       15700   1      56       A\n"
+  ;
+
+  return extra_tuningdb_block;
+}
+
 void module_init (module_ctx_t *module_ctx)
 {
   module_ctx->module_context_size             = MODULE_CONTEXT_SIZE_CURRENT;
@@ -465,6 +506,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -474,6 +516,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = module_extra_buffer_size;
   module_ctx->module_extra_tmp_size           = module_extra_tmp_size;
+  module_ctx->module_extra_tuningdb_block     = module_extra_tuningdb_block;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_15900.c b/src/modules/module_15900.c
index b07809107..84950912a 100644
--- a/src/modules/module_15900.c
+++ b/src/modules/module_15900.c
@@ -212,9 +212,7 @@ int module_hash_decode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSE
 
   dpapimk->context = hc_strtoul ((const char *) context_pos, NULL, 10);
 
-  // division by 4 should be fine because contents_len is either 208 or 288
-
-  for (u32 i = 0; i < dpapimk->contents_len / 4; i++)
+  for (u32 i = 0; i < dpapimk->contents_len / 8; i++)
   {
     dpapimk->contents[i] = hex_to_u32 ((const u8 *) &contents_pos[i * 8]);
 
@@ -303,7 +301,7 @@ int module_hash_encode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSE
 
   SID_tmp = (u8 *) hcmalloc ((SID_len + 1) * sizeof (u8));
 
-  for (u32 i = 0; i < (SID_len / 4) + 1; i++)
+  for (u32 i = 0; i < (SID_len / 4); i++)
   {
     u8 hex[8] = { 0 };
 
@@ -381,6 +379,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -390,6 +389,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_16000.c b/src/modules/module_16000.c
index 03ca3aad9..bba9c8e7d 100644
--- a/src/modules/module_16000.c
+++ b/src/modules/module_16000.c
@@ -139,6 +139,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -148,6 +149,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_16100.c b/src/modules/module_16100.c
index a1b1f801c..8bd6e8e59 100644
--- a/src/modules/module_16100.c
+++ b/src/modules/module_16100.c
@@ -196,6 +196,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -205,6 +206,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_16200.c b/src/modules/module_16200.c
index 19cbbac8d..573556c93 100644
--- a/src/modules/module_16200.c
+++ b/src/modules/module_16200.c
@@ -261,6 +261,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -270,6 +271,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_16300.c b/src/modules/module_16300.c
index 33997b1ed..bb9933b8d 100644
--- a/src/modules/module_16300.c
+++ b/src/modules/module_16300.c
@@ -274,6 +274,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -283,6 +284,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_16400.c b/src/modules/module_16400.c
index d76eba9f8..aed0bbf2d 100644
--- a/src/modules/module_16400.c
+++ b/src/modules/module_16400.c
@@ -121,6 +121,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -130,6 +131,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_16500.c b/src/modules/module_16500.c
index bad1e2172..41f889b34 100644
--- a/src/modules/module_16500.c
+++ b/src/modules/module_16500.c
@@ -341,6 +341,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = module_benchmark_salt;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -350,6 +351,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_16600.c b/src/modules/module_16600.c
index 9c92c690e..b44083651 100644
--- a/src/modules/module_16600.c
+++ b/src/modules/module_16600.c
@@ -193,6 +193,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -202,6 +203,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_16700.c b/src/modules/module_16700.c
index 5d5ddd234..ffbc9aa79 100644
--- a/src/modules/module_16700.c
+++ b/src/modules/module_16700.c
@@ -232,6 +232,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -241,6 +242,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_16800.c b/src/modules/module_16800.c
index 625d2fbf0..f6d53ed22 100644
--- a/src/modules/module_16800.c
+++ b/src/modules/module_16800.c
@@ -497,6 +497,13 @@ int module_hash_encode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSE
   return line_len;
 }
 
+const char *module_deprecated_notice (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const char *deprecated_notice = "The plugin 16800 is deprecated and was replaced with plugin 22000. For more details, please read: https://hashcat.net/forum/thread-10253.html";
+
+  return deprecated_notice;
+}
+
 void module_init (module_ctx_t *module_ctx)
 {
   module_ctx->module_context_size             = MODULE_CONTEXT_SIZE_CURRENT;
@@ -509,6 +516,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = module_deep_comp_kernel;
+  module_ctx->module_deprecated_notice        = module_deprecated_notice;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -518,6 +526,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_16801.c b/src/modules/module_16801.c
index 6d237ebf5..070d4b976 100644
--- a/src/modules/module_16801.c
+++ b/src/modules/module_16801.c
@@ -573,6 +573,13 @@ int module_hash_encode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSE
   return line_len;
 }
 
+const char *module_deprecated_notice (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const char *deprecated_notice = "The plugin 16801 is deprecated and was replaced with plugin 22001. For more details, please read: https://hashcat.net/forum/thread-10253.html";
+
+  return deprecated_notice;
+}
+
 void module_init (module_ctx_t *module_ctx)
 {
   module_ctx->module_context_size             = MODULE_CONTEXT_SIZE_CURRENT;
@@ -585,6 +592,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = module_deep_comp_kernel;
+  module_ctx->module_deprecated_notice        = module_deprecated_notice;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -594,6 +602,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_16900.c b/src/modules/module_16900.c
index 93915b592..df2900833 100644
--- a/src/modules/module_16900.c
+++ b/src/modules/module_16900.c
@@ -297,6 +297,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -306,6 +307,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_17200.c b/src/modules/module_17200.c
index 44798110d..8d29087ba 100644
--- a/src/modules/module_17200.c
+++ b/src/modules/module_17200.c
@@ -102,7 +102,7 @@ static const u32   HASH_CATEGORY  = HASH_CATEGORY_ARCHIVE;
 static const char *HASH_NAME      = "PKZIP (Compressed)";
 static const u64   KERN_TYPE      = 17200;
 static const u32   OPTI_TYPE      = 0;
-static const u64   OPTS_TYPE      = 0;
+static const u64   OPTS_TYPE      = OPTS_TYPE_NATIVE_THREADS;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
 static const char *ST_PASS        = "hashcat";
 static const char *ST_HASH        = "$pkzip2$1*1*2*0*e3*1c5*eda7a8de*0*28*8*e3*eda7*5096*a9fc1f4e951c8fb3031a6f903e5f4e3211c8fdc4671547bf77f6f682afbfcc7475d83898985621a7af9bccd1349d1976500a68c48f630b7f22d7a0955524d768e34868880461335417ddd149c65a917c0eb0a4bf7224e24a1e04cf4ace5eef52205f4452e66ded937db9545f843a68b1e84a2e933cc05fb36d3db90e6c5faf1bee2249fdd06a7307849902a8bb24ec7e8a0886a4544ca47979a9dfeefe034bdfc5bd593904cfe9a5309dd199d337d3183f307c2cb39622549a5b9b8b485b7949a4803f63f67ca427a0640ad3793a519b2476c52198488e3e2e04cac202d624fb7d13c2*$/pkzip2$";
@@ -170,11 +170,6 @@ bool module_unstable_warning (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE
   // it leads to CL_KERNEL_WORK_GROUP_SIZE to return 0 and later we will divide with 0
   // workaround would be to rewrite kernel to use global memory
 
-  if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD_USE_HIP)
-  {
-    return true;
-  }
-
   if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD)
   {
     return true;
@@ -387,6 +382,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -396,6 +392,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_17210.c b/src/modules/module_17210.c
index 326159fed..23fdb4126 100644
--- a/src/modules/module_17210.c
+++ b/src/modules/module_17210.c
@@ -375,6 +375,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -384,6 +385,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_17220.c b/src/modules/module_17220.c
index 9028040d9..ed94c32e2 100644
--- a/src/modules/module_17220.c
+++ b/src/modules/module_17220.c
@@ -102,7 +102,7 @@ static const u32   HASH_CATEGORY  = HASH_CATEGORY_ARCHIVE;
 static const char *HASH_NAME      = "PKZIP (Compressed Multi-File)";
 static const u64   KERN_TYPE      = 17220;
 static const u32   OPTI_TYPE      = 0;
-static const u64   OPTS_TYPE      = 0;
+static const u64   OPTS_TYPE      = OPTS_TYPE_NATIVE_THREADS;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
 static const char *ST_PASS        = "hashcat";
 static const char *ST_HASH        = "$pkzip2$3*1*1*0*8*24*a425*8827*d1730095cd829e245df04ebba6c52c0573d49d3bbeab6cb385b7fa8a28dcccd3098bfdd7*1*0*8*24*2a74*882a*51281ac874a60baedc375ca645888d29780e20d4076edd1e7154a99bde982152a736311f*2*0*e3*1c5*eda7a8de*0*29*8*e3*eda7*5096*1455781b59707f5151139e018bdcfeebfc89bc37e372883a7ec0670a5eafc622feb338f9b021b6601a674094898a91beac70e41e675f77702834ca6156111a1bf7361bc9f3715d77dfcdd626634c68354c6f2e5e0a7b1e1ce84a44e632d0f6e36019feeab92fb7eac9dda8df436e287aafece95d042059a1b27d533c5eab62c1c559af220dc432f2eb1a38a70f29e8f3cb5a207704274d1e305d7402180fd47e026522792f5113c52a116d5bb25b67074ffd6f4926b221555234aabddc69775335d592d5c7d22462b75de1259e8342a9ba71cb06223d13c7f51f13be2ad76352c3b8ed*$/pkzip2$";
@@ -170,11 +170,6 @@ bool module_unstable_warning (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE
   // it leads to CL_KERNEL_WORK_GROUP_SIZE to return 0 and later we will divide with 0
   // workaround would be to rewrite kernel to use global memory
 
-  if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD_USE_HIP)
-  {
-    return true;
-  }
-
   if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD)
   {
     return true;
@@ -390,6 +385,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -399,6 +395,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_17225.c b/src/modules/module_17225.c
index 75c376c9e..873b5e9c3 100644
--- a/src/modules/module_17225.c
+++ b/src/modules/module_17225.c
@@ -102,7 +102,7 @@ static const u32   HASH_CATEGORY  = HASH_CATEGORY_ARCHIVE;
 static const char *HASH_NAME      = "PKZIP (Mixed Multi-File)";
 static const u64   KERN_TYPE      = 17225;
 static const u32   OPTI_TYPE      = 0;
-static const u64   OPTS_TYPE      = 0;
+static const u64   OPTS_TYPE      = OPTS_TYPE_NATIVE_THREADS;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
 static const char *ST_PASS        = "hashcat";
 static const char *ST_HASH        = "$pkzip2$3*1*1*0*0*24*3e2c*3ef8*0619e9d17ff3f994065b99b1fa8aef41c056edf9fa4540919c109742dcb32f797fc90ce0*1*0*8*24*431a*3f26*18e2461c0dbad89bd9cc763067a020c89b5e16195b1ac5fa7fb13bd246d000b6833a2988*2*0*23*17*1e3c1a16*2e4*2f*0*23*1e3c*3f2d*54ea4dbc711026561485bbd191bf300ae24fa0997f3779b688cdad323985f8d3bb8b0c*$/pkzip2$";
@@ -170,11 +170,6 @@ bool module_unstable_warning (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE
   // it leads to CL_KERNEL_WORK_GROUP_SIZE to return 0 and later we will divide with 0
   // workaround would be to rewrite kernel to use global memory
 
-  if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD_USE_HIP)
-  {
-    return true;
-  }
-
   if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD)
   {
     return true;
@@ -391,6 +386,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -400,6 +396,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_17230.c b/src/modules/module_17230.c
index c6e720465..7cba32093 100644
--- a/src/modules/module_17230.c
+++ b/src/modules/module_17230.c
@@ -380,6 +380,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -389,6 +390,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_17300.c b/src/modules/module_17300.c
index fbd89af0f..0cf9cae26 100644
--- a/src/modules/module_17300.c
+++ b/src/modules/module_17300.c
@@ -116,6 +116,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -125,6 +126,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_17400.c b/src/modules/module_17400.c
index 63f5b8598..94f9585a4 100644
--- a/src/modules/module_17400.c
+++ b/src/modules/module_17400.c
@@ -107,6 +107,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -116,6 +117,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_17500.c b/src/modules/module_17500.c
index 6e6c1fa98..59586396c 100644
--- a/src/modules/module_17500.c
+++ b/src/modules/module_17500.c
@@ -113,6 +113,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -122,6 +123,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_17600.c b/src/modules/module_17600.c
index 1974550e7..e8a3128a3 100644
--- a/src/modules/module_17600.c
+++ b/src/modules/module_17600.c
@@ -119,6 +119,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -128,6 +129,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_17700.c b/src/modules/module_17700.c
index dd750e526..3dd4f0c50 100644
--- a/src/modules/module_17700.c
+++ b/src/modules/module_17700.c
@@ -116,6 +116,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -125,6 +126,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_17800.c b/src/modules/module_17800.c
index 7c516f071..d96407e6b 100644
--- a/src/modules/module_17800.c
+++ b/src/modules/module_17800.c
@@ -107,6 +107,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -116,6 +117,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_17900.c b/src/modules/module_17900.c
index c527ee657..8a08fd72b 100644
--- a/src/modules/module_17900.c
+++ b/src/modules/module_17900.c
@@ -113,6 +113,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -122,6 +123,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_18000.c b/src/modules/module_18000.c
index 9f1803dc4..aefeaf1c2 100644
--- a/src/modules/module_18000.c
+++ b/src/modules/module_18000.c
@@ -119,6 +119,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -128,6 +129,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_18100.c b/src/modules/module_18100.c
index 41c620bdc..fdbe2d6d3 100644
--- a/src/modules/module_18100.c
+++ b/src/modules/module_18100.c
@@ -149,6 +149,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = module_build_plain_postprocess;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -158,6 +159,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_18200.c b/src/modules/module_18200.c
index e6596306b..a15ac5063 100644
--- a/src/modules/module_18200.c
+++ b/src/modules/module_18200.c
@@ -64,35 +64,13 @@ char *module_jit_build_options (MAYBE_UNUSED const hashconfig_t *hashconfig, MAY
   }
   else if (device_param->opencl_device_type & CL_DEVICE_TYPE_GPU)
   {
-    if (device_param->opencl_device_vendor_id == VENDOR_ID_INTEL_SDK)
+    if (device_param->device_local_mem_size < 49152)
     {
-      native_threads = 8;
-    }
-    else if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD)
-    {
-      if (device_param->device_local_mem_size < 49152)
-      {
-        native_threads = 32;
-      }
-      else
-      {
-        native_threads = 64;
-      }
-    }
-    else if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD_USE_HIP)
-    {
-      if (device_param->device_local_mem_size < 49152)
-      {
-        native_threads = 32;
-      }
-      else
-      {
-        native_threads = 64;
-      }
+      native_threads = MIN (device_param->kernel_preferred_wgs_multiple, 32); // We can't just set 32, because Intel GPU need 8
     }
     else
     {
-      native_threads = 32;
+      native_threads = device_param->kernel_preferred_wgs_multiple;
     }
   }
 
@@ -251,6 +229,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -260,6 +239,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_18300.c b/src/modules/module_18300.c
index b58ef35f5..70c3753f5 100644
--- a/src/modules/module_18300.c
+++ b/src/modules/module_18300.c
@@ -271,6 +271,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -280,6 +281,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_18400.c b/src/modules/module_18400.c
index 35670017f..68b893502 100644
--- a/src/modules/module_18400.c
+++ b/src/modules/module_18400.c
@@ -289,6 +289,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -298,6 +299,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_18500.c b/src/modules/module_18500.c
index 7a3ab51be..953d3e1bd 100644
--- a/src/modules/module_18500.c
+++ b/src/modules/module_18500.c
@@ -145,6 +145,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -154,6 +155,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_18600.c b/src/modules/module_18600.c
index c24ea37fe..adf04fef3 100644
--- a/src/modules/module_18600.c
+++ b/src/modules/module_18600.c
@@ -21,7 +21,8 @@ static const char *HASH_NAME      = "Open Document Format (ODF) 1.1 (SHA-1, Blow
 static const u64   KERN_TYPE      = 18600;
 static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
                                   | OPTI_TYPE_SLOW_HASH_SIMD_LOOP;
-static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE;
+static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
+                                  | OPTS_TYPE_DYNAMIC_SHARED;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
 static const char *ST_PASS        = "hashcat";
 static const char *ST_HASH        = "$odf$*0*0*1024*16*bff753835f4ea15644b8a2f8e4b5be3d147b9576*8*ee371da34333b69d*16*a902eff54a4d782a26a899a31f97bef4*0*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";
@@ -66,16 +67,25 @@ char *module_jit_build_options (MAYBE_UNUSED const hashconfig_t *hashconfig, MAY
 {
   char *jit_build_options = NULL;
 
+  // this mode heavily depends on the available shared memory size
+  // note the kernel need to have some special code changes in order to make use to use post-48k memory region
+  // we need to set some macros
+
+  bool use_dynamic = false;
+
+  if (device_param->is_cuda == true)
+  {
+    use_dynamic = true;
+  }
+
   // this uses some nice feedback effect.
   // based on the device_local_mem_size the reqd_work_group_size in the kernel is set to some value
   // which is then is read from the opencl host in the kernel_preferred_wgs_multiple1/2/3 result.
   // therefore we do not need to set module_kernel_threads_min/max except for CPU, where the threads are set to fixed 1.
 
-  u32 fixed_local_size = 0;
-
   if (device_param->opencl_device_type & CL_DEVICE_TYPE_CPU)
   {
-    fixed_local_size = 1;
+    hc_asprintf (&jit_build_options, "-D FIXED_LOCAL_SIZE_COMP=%u", 1);
   }
   else
   {
@@ -91,29 +101,58 @@ char *module_jit_build_options (MAYBE_UNUSED const hashconfig_t *hashconfig, MAY
 
       if (device_param->is_opencl == true)
       {
-        overhead = 4;
+        overhead = 1;
       }
     }
 
     if (user_options->kernel_threads_chgd == true)
     {
-      fixed_local_size = user_options->kernel_threads;
+      u32 fixed_local_size = user_options->kernel_threads;
 
-      // otherwise out-of-bound reads
-
-      if ((fixed_local_size * 4096) > (device_param->device_local_mem_size - overhead))
+      if (use_dynamic == true)
       {
-        fixed_local_size = (device_param->device_local_mem_size - overhead) / 4096;
+        if ((fixed_local_size * 4096) > device_param->kernel_dynamic_local_mem_size_memset)
+        {
+          // otherwise out-of-bound reads
+
+          fixed_local_size = device_param->kernel_dynamic_local_mem_size_memset / 4096;
+        }
+
+        hc_asprintf (&jit_build_options, "-D FIXED_LOCAL_SIZE_COMP=%u -D DYNAMIC_LOCAL", fixed_local_size);
+      }
+      else
+      {
+        if ((fixed_local_size * 4096) > (device_param->device_local_mem_size - overhead))
+        {
+          // otherwise out-of-bound reads
+
+          fixed_local_size = (device_param->device_local_mem_size - overhead) / 4096;
+        }
+
+        hc_asprintf (&jit_build_options, "-D FIXED_LOCAL_SIZE_COMP=%u", fixed_local_size);
       }
     }
     else
     {
-      fixed_local_size = (device_param->device_local_mem_size - overhead) / 4096;
+      if (use_dynamic == true)
+      {
+        // using kernel_dynamic_local_mem_size_memset is a bit hackish.
+        // we had to brute-force this value out of an already loaded CUDA function.
+        // there's no official way to query for this value.
+
+        const u32 fixed_local_size = device_param->kernel_dynamic_local_mem_size_memset / 4096;
+
+        hc_asprintf (&jit_build_options, "-D FIXED_LOCAL_SIZE_COMP=%u -D DYNAMIC_LOCAL", fixed_local_size);
+      }
+      else
+      {
+        const u32 fixed_local_size = (device_param->device_local_mem_size - overhead) / 4096;
+
+        hc_asprintf (&jit_build_options, "-D FIXED_LOCAL_SIZE_COMP=%u", fixed_local_size);
+      }
     }
   }
 
-  hc_asprintf (&jit_build_options, "-D FIXED_LOCAL_SIZE=%u", fixed_local_size);
-
   return jit_build_options;
 }
 
@@ -338,6 +377,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -347,6 +387,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_18700.c b/src/modules/module_18700.c
index c23cc2efc..f23112699 100644
--- a/src/modules/module_18700.c
+++ b/src/modules/module_18700.c
@@ -94,6 +94,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -103,6 +104,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_18800.c b/src/modules/module_18800.c
index 6847edde2..431a9beaa 100644
--- a/src/modules/module_18800.c
+++ b/src/modules/module_18800.c
@@ -233,6 +233,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -242,6 +243,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_18900.c b/src/modules/module_18900.c
index 55e05f605..bc3368704 100644
--- a/src/modules/module_18900.c
+++ b/src/modules/module_18900.c
@@ -317,6 +317,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -326,6 +327,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_19000.c b/src/modules/module_19000.c
index a3735c2ed..c88ce3fab 100644
--- a/src/modules/module_19000.c
+++ b/src/modules/module_19000.c
@@ -187,6 +187,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -196,6 +197,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_19100.c b/src/modules/module_19100.c
index 7c663c4bb..04c58fe47 100644
--- a/src/modules/module_19100.c
+++ b/src/modules/module_19100.c
@@ -195,6 +195,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -204,6 +205,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_19200.c b/src/modules/module_19200.c
index 3dd7b46f6..84a84b7fe 100644
--- a/src/modules/module_19200.c
+++ b/src/modules/module_19200.c
@@ -212,6 +212,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -221,6 +222,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_19300.c b/src/modules/module_19300.c
index 1b0665ac5..673b14c5f 100644
--- a/src/modules/module_19300.c
+++ b/src/modules/module_19300.c
@@ -203,6 +203,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -212,6 +213,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_19500.c b/src/modules/module_19500.c
index c0f2ced88..563f68560 100644
--- a/src/modules/module_19500.c
+++ b/src/modules/module_19500.c
@@ -218,6 +218,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -227,6 +228,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_19600.c b/src/modules/module_19600.c
index 343b09637..63c458c0b 100644
--- a/src/modules/module_19600.c
+++ b/src/modules/module_19600.c
@@ -304,6 +304,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -313,6 +314,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_19700.c b/src/modules/module_19700.c
index 7d353e598..ffef6c791 100644
--- a/src/modules/module_19700.c
+++ b/src/modules/module_19700.c
@@ -304,6 +304,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -313,6 +314,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_19800.c b/src/modules/module_19800.c
index 0ac45ee57..e9eed9977 100644
--- a/src/modules/module_19800.c
+++ b/src/modules/module_19800.c
@@ -250,6 +250,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -259,6 +260,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_19900.c b/src/modules/module_19900.c
index 644968c04..c9108bfa1 100644
--- a/src/modules/module_19900.c
+++ b/src/modules/module_19900.c
@@ -250,6 +250,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -259,6 +260,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_20011.c b/src/modules/module_20011.c
index 7bd94888a..8b6007f63 100644
--- a/src/modules/module_20011.c
+++ b/src/modules/module_20011.c
@@ -205,6 +205,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -214,6 +215,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_20012.c b/src/modules/module_20012.c
index 58f88b2c6..c2d969387 100644
--- a/src/modules/module_20012.c
+++ b/src/modules/module_20012.c
@@ -205,6 +205,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -214,6 +215,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_20013.c b/src/modules/module_20013.c
index 0ecd72c8a..821c6237c 100644
--- a/src/modules/module_20013.c
+++ b/src/modules/module_20013.c
@@ -205,6 +205,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -214,6 +215,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_20200.c b/src/modules/module_20200.c
index 9f18dc7d9..23fb2002b 100644
--- a/src/modules/module_20200.c
+++ b/src/modules/module_20200.c
@@ -232,6 +232,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -241,6 +242,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_20300.c b/src/modules/module_20300.c
index 0624b8dce..d3756c28f 100644
--- a/src/modules/module_20300.c
+++ b/src/modules/module_20300.c
@@ -231,6 +231,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -240,6 +241,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_20400.c b/src/modules/module_20400.c
index 5df332861..2f600e94e 100644
--- a/src/modules/module_20400.c
+++ b/src/modules/module_20400.c
@@ -224,6 +224,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -233,6 +234,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_20500.c b/src/modules/module_20500.c
index f568d2dc3..d34da762e 100644
--- a/src/modules/module_20500.c
+++ b/src/modules/module_20500.c
@@ -161,6 +161,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -170,6 +171,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_20510.c b/src/modules/module_20510.c
index d3a34a961..7369415a9 100644
--- a/src/modules/module_20510.c
+++ b/src/modules/module_20510.c
@@ -200,6 +200,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = module_build_plain_postprocess;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -209,6 +210,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_20600.c b/src/modules/module_20600.c
index e270fde50..38dbeb19a 100644
--- a/src/modules/module_20600.c
+++ b/src/modules/module_20600.c
@@ -192,6 +192,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -201,6 +202,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_20710.c b/src/modules/module_20710.c
index 08f407ef4..46fa3cd0f 100644
--- a/src/modules/module_20710.c
+++ b/src/modules/module_20710.c
@@ -189,6 +189,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -198,6 +199,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_20711.c b/src/modules/module_20711.c
index 2fb69ce2c..07a0e1ab3 100644
--- a/src/modules/module_20711.c
+++ b/src/modules/module_20711.c
@@ -224,6 +224,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -233,6 +234,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_20720.c b/src/modules/module_20720.c
index 72992bc1f..9e47efb9c 100644
--- a/src/modules/module_20720.c
+++ b/src/modules/module_20720.c
@@ -189,6 +189,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -198,6 +199,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_20800.c b/src/modules/module_20800.c
index 65f4acb82..3ea10c651 100644
--- a/src/modules/module_20800.c
+++ b/src/modules/module_20800.c
@@ -166,6 +166,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -175,6 +176,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_20900.c b/src/modules/module_20900.c
index 6e269034a..76a0a69e1 100644
--- a/src/modules/module_20900.c
+++ b/src/modules/module_20900.c
@@ -128,6 +128,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -137,6 +138,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_21000.c b/src/modules/module_21000.c
index 7df98beb4..d2b331c34 100644
--- a/src/modules/module_21000.c
+++ b/src/modules/module_21000.c
@@ -207,6 +207,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -216,6 +217,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_21100.c b/src/modules/module_21100.c
index c1fc210ec..5d5a0454c 100644
--- a/src/modules/module_21100.c
+++ b/src/modules/module_21100.c
@@ -171,6 +171,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -180,6 +181,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_21200.c b/src/modules/module_21200.c
index 3a99a2cbd..6d22729f6 100644
--- a/src/modules/module_21200.c
+++ b/src/modules/module_21200.c
@@ -185,6 +185,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -194,6 +195,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_21300.c b/src/modules/module_21300.c
index e35e0160a..fa551496c 100644
--- a/src/modules/module_21300.c
+++ b/src/modules/module_21300.c
@@ -153,6 +153,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -162,6 +163,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_21400.c b/src/modules/module_21400.c
index d941c58de..19df994af 100644
--- a/src/modules/module_21400.c
+++ b/src/modules/module_21400.c
@@ -166,6 +166,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -175,6 +176,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_21500.c b/src/modules/module_21500.c
index f6c68655b..d4579421a 100644
--- a/src/modules/module_21500.c
+++ b/src/modules/module_21500.c
@@ -234,6 +234,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -243,6 +244,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_21501.c b/src/modules/module_21501.c
index 33285c18b..5d953dc3b 100644
--- a/src/modules/module_21501.c
+++ b/src/modules/module_21501.c
@@ -242,6 +242,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -251,6 +252,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_21600.c b/src/modules/module_21600.c
index 8112d0813..5841d8766 100644
--- a/src/modules/module_21600.c
+++ b/src/modules/module_21600.c
@@ -165,6 +165,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -174,6 +175,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_21700.c b/src/modules/module_21700.c
index 8152d57e2..95a9898c6 100644
--- a/src/modules/module_21700.c
+++ b/src/modules/module_21700.c
@@ -258,6 +258,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -267,6 +268,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_21800.c b/src/modules/module_21800.c
index a3f3a00c2..dd5a7104e 100644
--- a/src/modules/module_21800.c
+++ b/src/modules/module_21800.c
@@ -24,7 +24,8 @@ static const u64   KERN_TYPE      = 21800;
 static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
                                   | OPTI_TYPE_USES_BITS_64
                                   | OPTI_TYPE_SLOW_HASH_SIMD_LOOP;
-static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE;
+static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
+                                  | OPTS_TYPE_NATIVE_THREADS;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
 static const char *ST_PASS        = "hashcat";
 static const char *ST_HASH        = "$electrum$5*02170fee7c35f1ef3b229edc90fbd0793b688a0d6f41137a97aab2343d315cce16*94cf72d8f5d774932b414a3344984859e43721268d2eb35fa531de5a2fc7024b463c730a54f4f46229dd9fede5034b19ac415c2916e9c16b02094f845795df0c397ff76d597886b1f9e014ad1a8f64a3f617d9900aa645b3ba86f16ce542251fc22c41d93fa6bc118be96d9582917e19d2a299743331804cfc7ce2c035367b4cbcfb70adfb1e10a0f2795769f2165d8fd13daa8b45eeac495b5b63e91a87f63b42e483f84a881e49adecacf6519cb564694b42dd9fe80fcbc6cdb63cf5ae33f35255266f5c2524dd93d3cc15eba0f2ccdc3c109cc2d7e8f711b8b440f168caf8b005e8bcdfe694148e94a04d2a738f09349a96600bd8e8edae793b26ebae231022f24e96cb158db141ac40400a9e9ef099e673cfe017281537c57f82fb45c62bdb64462235a6eefb594961d5eb2c46537958e4d04250804c6e9f343ab7a0db07af6b8a9d1a6c5cfcd311b8fb8383ac9ed9d98d427d526c2f517fc97473bd87cb59899bd0e8fb8c57fa0f7e0d53daa57c972cf92764af4b1725a5fb8f504b663ec519731929b3caaa793d8ee74293eee27d0e208a60e26290bc546e6fa9ed865076e13febfea249729218c1b5752e912055fbf993fbac5df2cca2b37c5e0f9c30789858ceeb3c482a8db123966775aeed2eee2fc34efb160d164929f51589bff748ca773f38978bff3508d5a7591fb2d2795df983504a788071f469d78c88fd7899cabbc5804f458653d0206b82771a59522e1fa794d7de1536c51a437f5d6df5efd6654678e5794ca429b5752e1103340ed80786f1e9da7f5b39af628b2212e4d88cd36b8a7136d50a6b6e275ab406ba7c57cc70d77d01c4c16e9363901164fa92dc9e9b99219d5376f24862e775968605001e71b000e2c7123b4b43f3ca40db17efd729388782e46e64d43ccb947db4eb1473ff1a3836b74fe312cd1a33b73b8b8d80c087088932277773c329f2f66a01d6b3fc1e651c56959ebbed7b14a21b977f3acdedf1a0d98d519a74b50c39b3052d840106da4145345d86ec0461cddafacc2a4f0dd646457ad05bf04dcbcc80516a5c5ed14d2d639a70e77b686f19cbfb63f546d81ae19cc8ba35cce3f3b5b9602df25b678e14411fecec87b8347f5047513df415c6b1a3d39871a6bcb0f67d9cf8311596deae45fd1d84a04fd58f1fd55c5156b7309af09094c99a53674809cb87a45f95a2d69f9997a38085519cb4e056f9efd56672a2c1fe927d5ea8eec25b8aff6e56f9a2310f1a481daf407b8adf16201da267c59973920fd21bb087b88123ef98709839d6a3ee34efb8ccd5c15ed0e46cff3172682769531164b66c8689c35a26299dd26d09233d1f64f9667474141cf9c6a6de7f2bc52c3bb44cfe679ff4b912c06df406283836b3581773cb76d375304f46239da5996594a8d03b14c02f1b35a432dc44a96331242ae31174*33a7ee59d6d17ed1ee99dc0a71771227e6f3734b17ba36eb589bdced56244135";
@@ -284,6 +285,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -293,6 +295,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_22000.c b/src/modules/module_22000.c
index 2b5f60cfb..b450b3765 100644
--- a/src/modules/module_22000.c
+++ b/src/modules/module_22000.c
@@ -1283,6 +1283,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = module_deep_comp_kernel;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -1292,6 +1293,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = module_hash_binary_count;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_22001.c b/src/modules/module_22001.c
index dd45f3bd2..2c7e1d1aa 100644
--- a/src/modules/module_22001.c
+++ b/src/modules/module_22001.c
@@ -1284,6 +1284,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = module_deep_comp_kernel;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -1293,6 +1294,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = module_hash_binary_count;
   module_ctx->module_hash_binary_parse        = module_hash_binary_parse;
diff --git a/src/modules/module_22100.c b/src/modules/module_22100.c
index 47b72d7df..2f5cb510e 100644
--- a/src/modules/module_22100.c
+++ b/src/modules/module_22100.c
@@ -21,7 +21,8 @@ static const u32   HASH_CATEGORY  = HASH_CATEGORY_FDE;
 static const char *HASH_NAME      = "BitLocker";
 static const u64   KERN_TYPE      = 22100;
 static const u32   OPTI_TYPE      = OPTI_TYPE_SLOW_HASH_SIMD_LOOP;
-static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE;
+static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
+                                  | OPTS_TYPE_MP_MULTI_DISABLE;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
 static const char *ST_PASS        = "hashcat";
 static const char *ST_HASH        = "$bitlocker$1$16$6f972989ddc209f1eccf07313a7266a2$1048576$12$3a33a8eaff5e6f81d907b591$60$316b0f6d4cb445fb056f0e3e0633c413526ff4481bbf588917b70a4e8f8075f5ceb45958a800b42cb7ff9b7f5e17c6145bf8561ea86f52d3592059fb";
@@ -434,6 +435,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -443,6 +445,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_22200.c b/src/modules/module_22200.c
index 5ea525fbe..77d266050 100644
--- a/src/modules/module_22200.c
+++ b/src/modules/module_22200.c
@@ -238,6 +238,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -247,6 +248,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_22300.c b/src/modules/module_22300.c
index a5fde60de..95afbb9aa 100644
--- a/src/modules/module_22300.c
+++ b/src/modules/module_22300.c
@@ -189,6 +189,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -198,6 +199,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_22301.c b/src/modules/module_22301.c
index 9f96f4d0b..60e2c0084 100644
--- a/src/modules/module_22301.c
+++ b/src/modules/module_22301.c
@@ -191,6 +191,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -200,6 +201,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_22400.c b/src/modules/module_22400.c
index a3ab81101..7a8d6008e 100644
--- a/src/modules/module_22400.c
+++ b/src/modules/module_22400.c
@@ -312,6 +312,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -321,6 +322,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_22500.c b/src/modules/module_22500.c
index e30543f30..0775dcfdb 100644
--- a/src/modules/module_22500.c
+++ b/src/modules/module_22500.c
@@ -179,6 +179,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -188,6 +189,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_22600.c b/src/modules/module_22600.c
index 798bf88ab..32f7f159b 100644
--- a/src/modules/module_22600.c
+++ b/src/modules/module_22600.c
@@ -234,6 +234,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -243,6 +244,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_22700.c b/src/modules/module_22700.c
index 599144173..f4eae84da 100644
--- a/src/modules/module_22700.c
+++ b/src/modules/module_22700.c
@@ -392,6 +392,47 @@ int module_hash_encode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSE
   return line_len;
 }
 
+/*
+
+Find the right -n value for your GPU:
+=====================================
+
+1. For example, to find the value for 22700, first create a valid hash for 22700 as follows:
+
+$ ./hashcat --example-hashes -m 22700 | grep Example.Hash | grep -v Format | cut -b 25- > tmp.hash.22700
+
+2. Now let it iterate through all -n values to a certain point. In this case, I'm using 200, but in general it's a value that is at least twice that of the multiprocessor. If you don't mind you can just leave it as it is, it just runs a little longer.
+
+$ export i=1; while [ $i -ne 201 ]; do echo $i; ./hashcat --quiet tmp.hash.22700 --keep-guessing --self-test-disable --markov-disable --restore-disable --outfile-autohex-disable --wordlist-autohex-disable --potfile-disable --logfile-disable --hwmon-disable --status --status-timer 1 --runtime 28 --machine-readable --optimized-kernel-enable --workload-profile 3 --hash-type 22700 --attack-mode 3 ?b?b?b?b?b?b?b --backend-devices 1 --force -n $i; i=$(($i+1)); done | tee x
+
+3. Determine the highest measured H/s speed. But don't just use the highest value. Instead, use the number that seems most stable, usually at the beginning.
+
+$ grep "$(printf 'STATUS\t3')" x | cut -f4 -d$'\t' | sort -n | tail
+
+4. To match the speed you have chosen to the correct value in the 'x' file, simply search for it in it. Then go up a little on the block where you found him. The value -n is the single value that begins before the block start. If you have multiple blocks at the same speed, choose the lowest value for -n
+
+*/
+
+const char *module_extra_tuningdb_block (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const char *extra_tuningdb_block =
+    "DEVICE_TYPE_CPU                                 *       22700   1       N       A\n"
+    "DEVICE_TYPE_GPU                                 *       22700   1       N       A\n"
+    "GeForce_GTX_980                                 *       22700   1      29       A\n"
+    "GeForce_GTX_1080                                *       22700   1      15       A\n"
+    "GeForce_RTX_2080_Ti                             *       22700   1      68       A\n"
+    "GeForce_RTX_3060_Ti                             *       22700   1      51       A\n"
+    "GeForce_RTX_3070                                *       22700   1      46       A\n"
+    "GeForce_RTX_3090                                *       22700   1      82       A\n"
+    "ALIAS_AMD_RX480                                 *       22700   1      15       A\n"
+    "ALIAS_AMD_Vega64                                *       22700   1      31       A\n"
+    "ALIAS_AMD_MI100                                 *       22700   1      79       A\n"
+    "ALIAS_AMD_RX6900XT                              *       22700   1      59       A\n"
+  ;
+
+  return extra_tuningdb_block;
+}
+
 void module_init (module_ctx_t *module_ctx)
 {
   module_ctx->module_context_size             = MODULE_CONTEXT_SIZE_CURRENT;
@@ -404,6 +445,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -413,6 +455,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = module_extra_buffer_size;
   module_ctx->module_extra_tmp_size           = module_extra_tmp_size;
+  module_ctx->module_extra_tuningdb_block     = module_extra_tuningdb_block;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_22911.c b/src/modules/module_22911.c
index 26e11c5bd..aa910dc12 100644
--- a/src/modules/module_22911.c
+++ b/src/modules/module_22911.c
@@ -201,6 +201,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -210,6 +211,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_22921.c b/src/modules/module_22921.c
index b6dc3c425..5f32b0baa 100644
--- a/src/modules/module_22921.c
+++ b/src/modules/module_22921.c
@@ -201,6 +201,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -210,6 +211,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_22931.c b/src/modules/module_22931.c
index bb6d1a641..e3b439e02 100644
--- a/src/modules/module_22931.c
+++ b/src/modules/module_22931.c
@@ -205,6 +205,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -214,6 +215,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_22941.c b/src/modules/module_22941.c
index 71743996c..463e9fe1a 100644
--- a/src/modules/module_22941.c
+++ b/src/modules/module_22941.c
@@ -205,6 +205,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -214,6 +215,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_22951.c b/src/modules/module_22951.c
index 64e6a1fce..722bde55c 100644
--- a/src/modules/module_22951.c
+++ b/src/modules/module_22951.c
@@ -205,6 +205,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -214,6 +215,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_23001.c b/src/modules/module_23001.c
index 40f6d76bd..7e4187e7a 100644
--- a/src/modules/module_23001.c
+++ b/src/modules/module_23001.c
@@ -253,6 +253,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -262,6 +263,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_23002.c b/src/modules/module_23002.c
index b151974e0..7dd0b05d3 100644
--- a/src/modules/module_23002.c
+++ b/src/modules/module_23002.c
@@ -253,6 +253,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -262,6 +263,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_23003.c b/src/modules/module_23003.c
index 5a033c1fa..1f72bbf73 100644
--- a/src/modules/module_23003.c
+++ b/src/modules/module_23003.c
@@ -253,6 +253,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -262,6 +263,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_23100.c b/src/modules/module_23100.c
index e5d9db05e..2a18d9f5b 100644
--- a/src/modules/module_23100.c
+++ b/src/modules/module_23100.c
@@ -212,6 +212,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -221,6 +222,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_23200.c b/src/modules/module_23200.c
index 72f8325a5..88d54d94a 100644
--- a/src/modules/module_23200.c
+++ b/src/modules/module_23200.c
@@ -197,6 +197,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -206,6 +207,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_23300.c b/src/modules/module_23300.c
index b5be729eb..f6b2bb7c7 100644
--- a/src/modules/module_23300.c
+++ b/src/modules/module_23300.c
@@ -265,6 +265,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -274,6 +275,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_23400.c b/src/modules/module_23400.c
index 5921ca9db..c2e4345c7 100644
--- a/src/modules/module_23400.c
+++ b/src/modules/module_23400.c
@@ -251,6 +251,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -260,6 +261,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_23500.c b/src/modules/module_23500.c
index e0607a8d4..b0d56e8b2 100644
--- a/src/modules/module_23500.c
+++ b/src/modules/module_23500.c
@@ -290,6 +290,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -299,6 +300,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_23600.c b/src/modules/module_23600.c
index a0dde6ca8..5f7fc76d9 100644
--- a/src/modules/module_23600.c
+++ b/src/modules/module_23600.c
@@ -290,6 +290,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -299,6 +300,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_23700.c b/src/modules/module_23700.c
index d3803b0eb..fc9318049 100644
--- a/src/modules/module_23700.c
+++ b/src/modules/module_23700.c
@@ -105,15 +105,6 @@ u32 module_kernel_loops_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_
   return kernel_loops_max;
 }
 
-u32 module_kernel_threads_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
-{
-  // -T 128 works slightly faster but it's free for the user to change
-
-  const u32 kernel_threads_max = (user_options->kernel_threads_chgd == true) ? user_options->kernel_threads : 128;
-
-  return kernel_threads_max;
-}
-
 u32 module_pw_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
 {
   const bool optimized_kernel = (hashconfig->opti_type & OPTI_TYPE_OPTIMIZED_KERNEL);
@@ -338,6 +329,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -347,6 +339,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
@@ -377,7 +370,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_kernel_accel_min         = MODULE_DEFAULT;
   module_ctx->module_kernel_loops_max         = module_kernel_loops_max;
   module_ctx->module_kernel_loops_min         = module_kernel_loops_min;
-  module_ctx->module_kernel_threads_max       = module_kernel_threads_max;
+  module_ctx->module_kernel_threads_max       = MODULE_DEFAULT;
   module_ctx->module_kernel_threads_min       = MODULE_DEFAULT;
   module_ctx->module_kern_type                = module_kern_type;
   module_ctx->module_kern_type_dynamic        = MODULE_DEFAULT;
diff --git a/src/modules/module_23800.c b/src/modules/module_23800.c
index f910b361f..95dd3703a 100644
--- a/src/modules/module_23800.c
+++ b/src/modules/module_23800.c
@@ -388,15 +388,6 @@ u32 module_kernel_loops_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_
   return kernel_loops_max;
 }
 
-u32 module_kernel_threads_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
-{
-  // -T 128 works slightly faster but it's free for the user to change
-
-  const u32 kernel_threads_max = (user_options->kernel_threads_chgd == true) ? user_options->kernel_threads : 128;
-
-  return kernel_threads_max;
-}
-
 u32 module_pw_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
 {
   const bool optimized_kernel = (hashconfig->opti_type & OPTI_TYPE_OPTIMIZED_KERNEL);
@@ -615,6 +606,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -624,6 +616,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
@@ -654,7 +647,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_kernel_accel_min         = MODULE_DEFAULT;
   module_ctx->module_kernel_loops_max         = module_kernel_loops_max;
   module_ctx->module_kernel_loops_min         = module_kernel_loops_min;
-  module_ctx->module_kernel_threads_max       = module_kernel_threads_max;
+  module_ctx->module_kernel_threads_max       = MODULE_DEFAULT;
   module_ctx->module_kernel_threads_min       = MODULE_DEFAULT;
   module_ctx->module_kern_type                = module_kern_type;
   module_ctx->module_kern_type_dynamic        = MODULE_DEFAULT;
diff --git a/src/modules/module_23900.c b/src/modules/module_23900.c
index 35ef8df35..d4f260720 100644
--- a/src/modules/module_23900.c
+++ b/src/modules/module_23900.c
@@ -221,6 +221,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -230,6 +231,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_24100.c b/src/modules/module_24100.c
index cd9b4d9c9..f6f77be3d 100644
--- a/src/modules/module_24100.c
+++ b/src/modules/module_24100.c
@@ -285,6 +285,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -294,6 +295,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_24200.c b/src/modules/module_24200.c
index bd93b36e0..c9ec74ee2 100644
--- a/src/modules/module_24200.c
+++ b/src/modules/module_24200.c
@@ -325,6 +325,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -334,6 +335,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_24300.c b/src/modules/module_24300.c
index 14770ff33..fe479621a 100644
--- a/src/modules/module_24300.c
+++ b/src/modules/module_24300.c
@@ -149,6 +149,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -158,6 +159,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_24410.c b/src/modules/module_24410.c
index 1a6b4274c..cc1f35fbd 100644
--- a/src/modules/module_24410.c
+++ b/src/modules/module_24410.c
@@ -298,6 +298,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -307,6 +308,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_24420.c b/src/modules/module_24420.c
index c45de065e..a065e36d1 100644
--- a/src/modules/module_24420.c
+++ b/src/modules/module_24420.c
@@ -298,6 +298,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -307,6 +308,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_24500.c b/src/modules/module_24500.c
index 5539d0fe2..67efc995e 100644
--- a/src/modules/module_24500.c
+++ b/src/modules/module_24500.c
@@ -22,7 +22,8 @@ static const u64   KERN_TYPE      = 24500;
 static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
                                   | OPTI_TYPE_USES_BITS_64
                                   | OPTI_TYPE_SLOW_HASH_SIMD_LOOP;
-static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE;
+static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
+                                  | OPTS_TYPE_MP_MULTI_DISABLE;
 static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
 static const char *ST_PASS        = "hashcat";
 static const char *ST_HASH        = "$telegram$2*100000*77461dcb457ce9539f8e4235d33bd12455b4a38446e63b52ecdf2e7b65af4476*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";
@@ -255,6 +256,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -264,6 +266,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_24600.c b/src/modules/module_24600.c
index 7a2ef5dc6..86c876d23 100644
--- a/src/modules/module_24600.c
+++ b/src/modules/module_24600.c
@@ -290,6 +290,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -299,6 +300,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_24700.c b/src/modules/module_24700.c
index 04163dc37..0cc56a1d8 100644
--- a/src/modules/module_24700.c
+++ b/src/modules/module_24700.c
@@ -52,9 +52,8 @@ int module_hash_decode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSE
 
   token.token_cnt  = 1;
 
-  token.len_min[0] = 10;
-  token.len_max[0] = 10;
-  token.attr[0]    = TOKEN_ATTR_VERIFY_LENGTH
+  token.len[0]     = 10;
+  token.attr[0]    = TOKEN_ATTR_FIXED_LENGTH
                    | TOKEN_ATTR_VERIFY_HEX;
 
   const int rc_tokenizer = input_tokenizer ((const u8 *) line_buf, line_len, &token);
@@ -62,9 +61,14 @@ int module_hash_decode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSE
   if (rc_tokenizer != PARSER_OK) return (rc_tokenizer);
 
   const u8 *hash_pos = token.buf[0];
+  const u32 hash_len = token.len[0];
 
-  digest[0] = hex_to_u32 (hash_pos +  0);
-  digest[1] = hex_to_u32 (hash_pos +  8);
+  u8 digest_tmp[16] = { 0 };
+
+  memcpy (digest_tmp, hash_pos, hash_len);
+
+  digest[0] = hex_to_u32 (digest_tmp +  0);
+  digest[1] = hex_to_u32 (digest_tmp +  8);
 
   if (hashconfig->opti_type & OPTI_TYPE_OPTIMIZED_KERNEL)
   {
@@ -117,6 +121,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -126,6 +131,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_24800.c b/src/modules/module_24800.c
index c978cb358..3da8c8533 100644
--- a/src/modules/module_24800.c
+++ b/src/modules/module_24800.c
@@ -121,6 +121,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -130,6 +131,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_24900.c b/src/modules/module_24900.c
index b68b1e639..383b5ef8d 100644
--- a/src/modules/module_24900.c
+++ b/src/modules/module_24900.c
@@ -149,6 +149,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -158,6 +159,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_25000.c b/src/modules/module_25000.c
new file mode 100644
index 000000000..1b5051ba1
--- /dev/null
+++ b/src/modules/module_25000.c
@@ -0,0 +1,344 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+#include "common.h"
+#include "types.h"
+#include "modules.h"
+#include "bitops.h"
+#include "convert.h"
+#include "shared.h"
+#include "memory.h"
+#include "emu_inc_hash_md5.h"
+
+static const u32   ATTACK_EXEC    = ATTACK_EXEC_OUTSIDE_KERNEL;
+static const u32   DGST_POS0      = 0;
+static const u32   DGST_POS1      = 1;
+static const u32   DGST_POS2      = 2;
+static const u32   DGST_POS3      = 3;
+static const u32   DGST_SIZE      = DGST_SIZE_4_4; // 4_3
+static const u32   HASH_CATEGORY  = HASH_CATEGORY_NETWORK_PROTOCOL;
+static const char *HASH_NAME      = "SNMPv3 HMAC-MD5-96/HMAC-SHA1-96";
+static const u64   KERN_TYPE      = 25000;
+static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE;
+static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE;
+static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
+static const char *ST_PASS        = "hashcat1";
+static const char *ST_HASH        = "$SNMPv3$0$45889431$30818f0201033011020409242fc0020300ffe304010102010304383036041180001f88808106d566db57fd600000000002011002020118040a6d61747269785f4d4435040c0000000000000000000000000400303d041180001f88808106d566db57fd60000000000400a226020411f319300201000201003018301606082b06010201010200060a2b06010401bf0803020a$80001f88808106d566db57fd6000000000$1b37c3ea872731f922959e90";
+
+u32         module_attack_exec    (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ATTACK_EXEC;     }
+u32         module_dgst_pos0      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS0;       }
+u32         module_dgst_pos1      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS1;       }
+u32         module_dgst_pos2      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS2;       }
+u32         module_dgst_pos3      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS3;       }
+u32         module_dgst_size      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_SIZE;       }
+u32         module_hash_category  (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return HASH_CATEGORY;   }
+const char *module_hash_name      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return HASH_NAME;       }
+u64         module_kern_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return KERN_TYPE;       }
+u32         module_opti_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return OPTI_TYPE;       }
+u64         module_opts_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return OPTS_TYPE;       }
+u32         module_salt_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return SALT_TYPE;       }
+const char *module_st_hash        (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ST_HASH;         }
+const char *module_st_pass        (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ST_PASS;         }
+
+static const char *SIGNATURE_SNMPV3 = "$SNMPv3$0$";
+
+#define SNMPV3_SALT_MAX             1500
+#define SNMPV3_ENGINEID_MAX         34
+#define SNMPV3_MSG_AUTH_PARAMS_LEN  12
+#define SNMPV3_ROUNDS               1048576
+#define SNMPV3_MAX_PW_LENGTH        64
+
+#define SNMPV3_TMP_ELEMS            4096 // 4096 = (256 (max pw length) * 64) / sizeof (u32)
+#define SNMPV3_HASH_ELEMS_MD5       4
+#define SNMPV3_HASH_ELEMS_SHA1      8
+
+#define SNMPV3_MAX_SALT_ELEMS       512 // 512 * 4 = 2048 > 1500, also has to be multiple of 64
+#define SNMPV3_MAX_ENGINE_ELEMS     16  // 16 * 4 = 64 > 32, also has to be multiple of 64
+#define SNMPV3_MAX_PNUM_ELEMS       4   // 4 * 4 = 16 > 9
+
+typedef struct hmac_md5_tmp
+{
+  u32 tmp_md5[SNMPV3_TMP_ELEMS];
+  u32 tmp_sha1[SNMPV3_TMP_ELEMS];
+
+  u32 h_md5[SNMPV3_HASH_ELEMS_MD5];
+  u32 h_sha1[SNMPV3_HASH_ELEMS_SHA1];
+
+} hmac_md5_tmp_t;
+
+typedef struct snmpv3
+{
+  u32 salt_buf[SNMPV3_MAX_SALT_ELEMS];
+  u32 salt_len;
+
+  u32 engineID_buf[SNMPV3_MAX_ENGINE_ELEMS];
+  u32 engineID_len;
+
+  u32 packet_number[SNMPV3_MAX_PNUM_ELEMS];
+
+} snmpv3_t;
+
+u32 module_pw_min (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u32 pw_min = 8;
+
+  return pw_min;
+}
+
+u64 module_esalt_size (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u64 esalt_size = (const u64) sizeof (snmpv3_t);
+
+  return esalt_size;
+}
+
+u64 module_tmp_size (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u64 tmp_size = (const u64) sizeof (hmac_md5_tmp_t);
+
+  return tmp_size;
+}
+
+u32 module_kernel_loops_min (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  // we need to fix iteration count to guarantee the loop count is a multiple of 64
+  // 2k calls to md5_transform/sha1_transform typically is enough to overtime pcie bottleneck
+
+  const u32 kernel_loops_min = 2048 * 64;
+
+  return kernel_loops_min;
+}
+
+u32 module_kernel_loops_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u32 kernel_loops_max = 2048 * 64;
+
+  return kernel_loops_max;
+}
+
+int module_hash_decode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED void *digest_buf, MAYBE_UNUSED salt_t *salt, MAYBE_UNUSED void *esalt_buf, MAYBE_UNUSED void *hook_salt_buf, MAYBE_UNUSED hashinfo_t *hash_info, const char *line_buf, MAYBE_UNUSED const int line_len)
+{
+  u32 *digest = (u32 *) digest_buf;
+
+  snmpv3_t *snmpv3 = (snmpv3_t *) esalt_buf;
+
+  token_t token;
+
+  token.token_cnt  = 5;
+  token.signatures_cnt    = 1;
+  token.signatures_buf[0] = SIGNATURE_SNMPV3;
+
+  token.len[0]     = 10;
+  token.attr[0]    = TOKEN_ATTR_FIXED_LENGTH
+                   | TOKEN_ATTR_VERIFY_SIGNATURE;
+
+  // packet number
+  token.len_min[1] = 1;
+  token.len_max[1] = 8;
+  token.sep[1]     = '$';
+  token.attr[1]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_DIGIT;
+  // salt
+  token.len_min[2] = SNMPV3_MSG_AUTH_PARAMS_LEN * 2;
+  token.len_max[2] = SNMPV3_SALT_MAX * 2;
+  token.sep[2]     = '$';
+  token.attr[2]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  // engineid
+  token.len_min[3] = 26;
+  token.len_max[3] = SNMPV3_ENGINEID_MAX;
+  token.sep[3]     = '$';
+  token.attr[3]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  // digest
+  token.len[4]     = SNMPV3_MSG_AUTH_PARAMS_LEN * 2;
+  token.sep[4]     = '$';
+  token.attr[4]    = TOKEN_ATTR_FIXED_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  const int rc_tokenizer = input_tokenizer ((const u8 *) line_buf, line_len, &token);
+
+  if (rc_tokenizer != PARSER_OK) return (rc_tokenizer);
+
+  // packet number
+
+  const u8 *packet_number_pos = token.buf[1];
+  const int packet_number_len = token.len[1];
+
+  memset (snmpv3->packet_number, 0, sizeof (snmpv3->packet_number));
+
+  strncpy ((char *) snmpv3->packet_number, (char *) packet_number_pos, packet_number_len);
+
+  // salt
+
+  const u8 *salt_pos = token.buf[2];
+  const int salt_len = token.len[2];
+
+  u8 *salt_ptr = (u8 *) snmpv3->salt_buf;
+
+  snmpv3->salt_len = hex_decode (salt_pos, salt_len, salt_ptr);
+
+  salt->salt_iter = SNMPV3_ROUNDS;
+
+  // handle unique salts detection
+
+  md5_ctx_t md5_ctx;
+
+  md5_init   (&md5_ctx);
+  md5_update (&md5_ctx, snmpv3->salt_buf, snmpv3->salt_len);
+  md5_final  (&md5_ctx);
+
+  // store md5(snmpv3->salt_buf) in salt_buf
+
+  salt->salt_len = 16;
+
+  memcpy (salt->salt_buf, md5_ctx.h, salt->salt_len);
+
+  // engineid
+
+  const u8 *engineID_pos = token.buf[3];
+  const int engineID_len = token.len[3];
+
+  u8 *engineID_ptr = (u8 *) snmpv3->engineID_buf;
+
+  snmpv3->engineID_len = hex_decode (engineID_pos, engineID_len, engineID_ptr);
+
+  // digest
+
+  const u8 *hash_pos = token.buf[4];
+
+  digest[0] = hex_to_u32 (hash_pos +  0);
+  digest[1] = hex_to_u32 (hash_pos +  8);
+  digest[2] = hex_to_u32 (hash_pos + 16);
+
+  // prefer sha1 due to speed
+
+  digest[0] = byte_swap_32 (digest[0]);
+  digest[1] = byte_swap_32 (digest[1]);
+  digest[2] = byte_swap_32 (digest[2]);
+
+  digest[3] = 0;
+
+  return (PARSER_OK);
+}
+
+int module_hash_encode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const void *digest_buf, MAYBE_UNUSED const salt_t *salt, MAYBE_UNUSED const void *esalt_buf, MAYBE_UNUSED const void *hook_salt_buf, MAYBE_UNUSED const hashinfo_t *hash_info, char *line_buf, MAYBE_UNUSED const int line_size)
+{
+  const u32 *digest = (const u32 *) digest_buf;
+
+  snmpv3_t *snmpv3 = (snmpv3_t *) esalt_buf;
+
+  u8 *out_buf = (u8 *) line_buf;
+
+  int out_len = snprintf (line_buf, line_size, "%s%s$", SIGNATURE_SNMPV3, (char *) snmpv3->packet_number);
+
+  out_len += hex_encode ((u8 *) snmpv3->salt_buf, snmpv3->salt_len, out_buf + out_len);
+
+  out_buf[out_len] = '$';
+
+  out_len++;
+
+  out_len += hex_encode ((u8 *) snmpv3->engineID_buf, snmpv3->engineID_len, out_buf + out_len);
+
+  out_buf[out_len] = '$';
+
+  out_len++;
+
+  // prefer sha1 due to speed
+
+  u32 digest_tmp[3];
+
+  digest_tmp[0] = byte_swap_32 (digest[0]);
+  digest_tmp[1] = byte_swap_32 (digest[1]);
+  digest_tmp[2] = byte_swap_32 (digest[2]);
+
+  u32_to_hex (digest_tmp[0], out_buf + out_len); out_len += 8;
+  u32_to_hex (digest_tmp[1], out_buf + out_len); out_len += 8;
+  u32_to_hex (digest_tmp[2], out_buf + out_len); out_len += 8;
+
+  out_buf[out_len] = 0;
+
+  return out_len;
+}
+
+void module_init (module_ctx_t *module_ctx)
+{
+  module_ctx->module_context_size             = MODULE_CONTEXT_SIZE_CURRENT;
+  module_ctx->module_interface_version        = MODULE_INTERFACE_VERSION_CURRENT;
+
+  module_ctx->module_attack_exec              = module_attack_exec;
+  module_ctx->module_benchmark_esalt          = MODULE_DEFAULT;
+  module_ctx->module_benchmark_hook_salt      = MODULE_DEFAULT;
+  module_ctx->module_benchmark_mask           = MODULE_DEFAULT;
+  module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
+  module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
+  module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
+  module_ctx->module_dgst_pos0                = module_dgst_pos0;
+  module_ctx->module_dgst_pos1                = module_dgst_pos1;
+  module_ctx->module_dgst_pos2                = module_dgst_pos2;
+  module_ctx->module_dgst_pos3                = module_dgst_pos3;
+  module_ctx->module_dgst_size                = module_dgst_size;
+  module_ctx->module_dictstat_disable         = MODULE_DEFAULT;
+  module_ctx->module_esalt_size               = module_esalt_size;
+  module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
+  module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
+  module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_save         = MODULE_DEFAULT;
+  module_ctx->module_hash_decode_potfile      = MODULE_DEFAULT;
+  module_ctx->module_hash_decode_zero_hash    = MODULE_DEFAULT;
+  module_ctx->module_hash_decode              = module_hash_decode;
+  module_ctx->module_hash_encode_status       = MODULE_DEFAULT;
+  module_ctx->module_hash_encode_potfile      = MODULE_DEFAULT;
+  module_ctx->module_hash_encode              = module_hash_encode;
+  module_ctx->module_hash_init_selftest       = MODULE_DEFAULT;
+  module_ctx->module_hash_mode                = MODULE_DEFAULT;
+  module_ctx->module_hash_category            = module_hash_category;
+  module_ctx->module_hash_name                = module_hash_name;
+  module_ctx->module_hashes_count_min         = MODULE_DEFAULT;
+  module_ctx->module_hashes_count_max         = MODULE_DEFAULT;
+  module_ctx->module_hlfmt_disable            = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_size    = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_init    = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_term    = MODULE_DEFAULT;
+  module_ctx->module_hook12                   = MODULE_DEFAULT;
+  module_ctx->module_hook23                   = MODULE_DEFAULT;
+  module_ctx->module_hook_salt_size           = MODULE_DEFAULT;
+  module_ctx->module_hook_size                = MODULE_DEFAULT;
+  module_ctx->module_jit_build_options        = MODULE_DEFAULT;
+  module_ctx->module_jit_cache_disable        = MODULE_DEFAULT;
+  module_ctx->module_kernel_accel_max         = MODULE_DEFAULT;
+  module_ctx->module_kernel_accel_min         = MODULE_DEFAULT;
+  module_ctx->module_kernel_loops_max         = module_kernel_loops_max;
+  module_ctx->module_kernel_loops_min         = module_kernel_loops_min;
+  module_ctx->module_kernel_threads_max       = MODULE_DEFAULT;
+  module_ctx->module_kernel_threads_min       = MODULE_DEFAULT;
+  module_ctx->module_kern_type                = module_kern_type;
+  module_ctx->module_kern_type_dynamic        = MODULE_DEFAULT;
+  module_ctx->module_opti_type                = module_opti_type;
+  module_ctx->module_opts_type                = module_opts_type;
+  module_ctx->module_outfile_check_disable    = MODULE_DEFAULT;
+  module_ctx->module_outfile_check_nocomp     = MODULE_DEFAULT;
+  module_ctx->module_potfile_custom_check     = MODULE_DEFAULT;
+  module_ctx->module_potfile_disable          = MODULE_DEFAULT;
+  module_ctx->module_potfile_keep_all_hashes  = MODULE_DEFAULT;
+  module_ctx->module_pwdump_column            = MODULE_DEFAULT;
+  module_ctx->module_pw_max                   = MODULE_DEFAULT;
+  module_ctx->module_pw_min                   = module_pw_min;
+  module_ctx->module_salt_max                 = MODULE_DEFAULT;
+  module_ctx->module_salt_min                 = MODULE_DEFAULT;
+  module_ctx->module_salt_type                = module_salt_type;
+  module_ctx->module_separator                = MODULE_DEFAULT;
+  module_ctx->module_st_hash                  = module_st_hash;
+  module_ctx->module_st_pass                  = module_st_pass;
+  module_ctx->module_tmp_size                 = module_tmp_size;
+  module_ctx->module_unstable_warning         = MODULE_DEFAULT;
+  module_ctx->module_warmup_disable           = MODULE_DEFAULT;
+}
diff --git a/src/modules/module_25100.c b/src/modules/module_25100.c
new file mode 100644
index 000000000..adb0ffa48
--- /dev/null
+++ b/src/modules/module_25100.c
@@ -0,0 +1,325 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+#include "common.h"
+#include "types.h"
+#include "modules.h"
+#include "bitops.h"
+#include "convert.h"
+#include "shared.h"
+#include "memory.h"
+#include "emu_inc_hash_md5.h"
+
+static const u32   ATTACK_EXEC    = ATTACK_EXEC_OUTSIDE_KERNEL;
+static const u32   DGST_POS0      = 0;
+static const u32   DGST_POS1      = 1;
+static const u32   DGST_POS2      = 2;
+static const u32   DGST_POS3      = 3;
+static const u32   DGST_SIZE      = DGST_SIZE_4_4; // 4_3
+static const u32   HASH_CATEGORY  = HASH_CATEGORY_NETWORK_PROTOCOL;
+static const char *HASH_NAME      = "SNMPv3 HMAC-MD5-96";
+static const u64   KERN_TYPE      = 25100;
+static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE;
+static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE;
+static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
+static const char *ST_PASS        = "hashcat1";
+static const char *ST_HASH        = "$SNMPv3$1$45889431$30818f0201033011020409242fc0020300ffe304010102010304383036041180001f88808106d566db57fd600000000002011002020118040a6d61747269785f4d4435040c0000000000000000000000000400303d041180001f88808106d566db57fd60000000000400a226020411f319300201000201003018301606082b06010201010200060a2b06010401bf0803020a$80001f88808106d566db57fd6000000000$1b37c3ea872731f922959e90";
+
+u32         module_attack_exec    (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ATTACK_EXEC;     }
+u32         module_dgst_pos0      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS0;       }
+u32         module_dgst_pos1      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS1;       }
+u32         module_dgst_pos2      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS2;       }
+u32         module_dgst_pos3      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS3;       }
+u32         module_dgst_size      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_SIZE;       }
+u32         module_hash_category  (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return HASH_CATEGORY;   }
+const char *module_hash_name      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return HASH_NAME;       }
+u64         module_kern_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return KERN_TYPE;       }
+u32         module_opti_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return OPTI_TYPE;       }
+u64         module_opts_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return OPTS_TYPE;       }
+u32         module_salt_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return SALT_TYPE;       }
+const char *module_st_hash        (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ST_HASH;         }
+const char *module_st_pass        (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ST_PASS;         }
+
+static const char *SIGNATURE_SNMPV3 = "$SNMPv3$1$";
+
+#define SNMPV3_SALT_MAX             1500
+#define SNMPV3_ENGINEID_MAX         34
+#define SNMPV3_MSG_AUTH_PARAMS_LEN  12
+#define SNMPV3_ROUNDS               1048576
+#define SNMPV3_MAX_PW_LENGTH        64
+
+#define SNMPV3_TMP_ELEMS            4096 // 4096 = (256 (max pw length) * 64) / sizeof (u32)
+#define SNMPV3_HASH_ELEMS           4
+
+#define SNMPV3_MAX_SALT_ELEMS       512 // 512 * 4 = 2048 > 1500, also has to be multiple of 64
+#define SNMPV3_MAX_ENGINE_ELEMS     16  // 16 * 4 = 64 > 32, also has to be multiple of 64
+#define SNMPV3_MAX_PNUM_ELEMS       4   // 4 * 4 = 16 > 9
+
+typedef struct hmac_md5_tmp
+{
+  u32 tmp[SNMPV3_TMP_ELEMS];
+  u32 h[SNMPV3_HASH_ELEMS];
+
+} hmac_md5_tmp_t;
+
+typedef struct snmpv3
+{
+  u32 salt_buf[SNMPV3_MAX_SALT_ELEMS];
+  u32 salt_len;
+
+  u32 engineID_buf[SNMPV3_MAX_ENGINE_ELEMS];
+  u32 engineID_len;
+
+  u32 packet_number[SNMPV3_MAX_PNUM_ELEMS];
+
+} snmpv3_t;
+
+u32 module_pw_min (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u32 pw_min = 8;
+
+  return pw_min;
+}
+
+u64 module_esalt_size (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u64 esalt_size = (const u64) sizeof (snmpv3_t);
+
+  return esalt_size;
+}
+
+u64 module_tmp_size (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u64 tmp_size = (const u64) sizeof (hmac_md5_tmp_t);
+
+  return tmp_size;
+}
+
+u32 module_kernel_loops_min (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  // we need to fix iteration count to guarantee the loop count is a multiple of 64
+  // 2k calls to md5_transform typically is enough to overtime pcie bottleneck
+
+  const u32 kernel_loops_min = 2048 * 64;
+
+  return kernel_loops_min;
+}
+
+u32 module_kernel_loops_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u32 kernel_loops_max = 2048 * 64;
+
+  return kernel_loops_max;
+}
+
+int module_hash_decode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED void *digest_buf, MAYBE_UNUSED salt_t *salt, MAYBE_UNUSED void *esalt_buf, MAYBE_UNUSED void *hook_salt_buf, MAYBE_UNUSED hashinfo_t *hash_info, const char *line_buf, MAYBE_UNUSED const int line_len)
+{
+  u32 *digest = (u32 *) digest_buf;
+
+  snmpv3_t *snmpv3 = (snmpv3_t *) esalt_buf;
+
+  token_t token;
+
+  token.token_cnt  = 5;
+  token.signatures_cnt    = 1;
+  token.signatures_buf[0] = SIGNATURE_SNMPV3;
+
+  token.len[0]     = 10;
+  token.attr[0]    = TOKEN_ATTR_FIXED_LENGTH
+                   | TOKEN_ATTR_VERIFY_SIGNATURE;
+
+  // packet number
+  token.len_min[1] = 1;
+  token.len_max[1] = 8;
+  token.sep[1]     = '$';
+  token.attr[1]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_DIGIT;
+  // salt
+  token.len_min[2] = SNMPV3_MSG_AUTH_PARAMS_LEN * 2;
+  token.len_max[2] = SNMPV3_SALT_MAX * 2;
+  token.sep[2]     = '$';
+  token.attr[2]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  // engineid
+  token.len_min[3] = 26;
+  token.len_max[3] = SNMPV3_ENGINEID_MAX;
+  token.sep[3]     = '$';
+  token.attr[3]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  // digest
+  token.len[4]     = SNMPV3_MSG_AUTH_PARAMS_LEN * 2;
+  token.sep[4]     = '$';
+  token.attr[4]    = TOKEN_ATTR_FIXED_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  const int rc_tokenizer = input_tokenizer ((const u8 *) line_buf, line_len, &token);
+
+  if (rc_tokenizer != PARSER_OK) return (rc_tokenizer);
+
+  // packet number
+
+  const u8 *packet_number_pos = token.buf[1];
+  const int packet_number_len = token.len[1];
+
+  memset (snmpv3->packet_number, 0, sizeof (snmpv3->packet_number));
+
+  strncpy ((char *) snmpv3->packet_number, (char *) packet_number_pos, packet_number_len);
+
+  // salt
+
+  const u8 *salt_pos = token.buf[2];
+  const int salt_len = token.len[2];
+
+  u8 *salt_ptr = (u8 *) snmpv3->salt_buf;
+
+  snmpv3->salt_len = hex_decode (salt_pos, salt_len, salt_ptr);
+
+  salt->salt_iter = SNMPV3_ROUNDS;
+
+  // handle unique salts detection
+
+  md5_ctx_t md5_ctx;
+
+  md5_init   (&md5_ctx);
+  md5_update (&md5_ctx, snmpv3->salt_buf, snmpv3->salt_len);
+  md5_final  (&md5_ctx);
+
+  // store md5(snmpv3->salt_buf) in salt_buf
+
+  salt->salt_len = 16;
+
+  memcpy (salt->salt_buf, md5_ctx.h, salt->salt_len);
+
+  // engineid
+
+  const u8 *engineID_pos = token.buf[3];
+  const int engineID_len = token.len[3];
+
+  u8 *engineID_ptr = (u8 *) snmpv3->engineID_buf;
+
+  snmpv3->engineID_len = hex_decode (engineID_pos, engineID_len, engineID_ptr);
+
+  // digest
+
+  const u8 *hash_pos = token.buf[4];
+
+  digest[0] = hex_to_u32 (hash_pos +  0);
+  digest[1] = hex_to_u32 (hash_pos +  8);
+  digest[2] = hex_to_u32 (hash_pos + 16);
+  digest[3] = 0;
+
+  return (PARSER_OK);
+}
+
+int module_hash_encode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const void *digest_buf, MAYBE_UNUSED const salt_t *salt, MAYBE_UNUSED const void *esalt_buf, MAYBE_UNUSED const void *hook_salt_buf, MAYBE_UNUSED const hashinfo_t *hash_info, char *line_buf, MAYBE_UNUSED const int line_size)
+{
+  const u32 *digest = (const u32 *) digest_buf;
+
+  snmpv3_t *snmpv3 = (snmpv3_t *) esalt_buf;
+
+  u8 *out_buf = (u8 *) line_buf;
+
+  int out_len = snprintf (line_buf, line_size, "%s%s$", SIGNATURE_SNMPV3, (char *) snmpv3->packet_number);
+
+  out_len += hex_encode ((u8 *) snmpv3->salt_buf, snmpv3->salt_len, out_buf + out_len);
+
+  out_buf[out_len] = '$';
+
+  out_len++;
+
+  out_len += hex_encode ((u8 *) snmpv3->engineID_buf, snmpv3->engineID_len, out_buf + out_len);
+
+  out_buf[out_len] = '$';
+
+  out_len++;
+
+  u32_to_hex (digest[0], out_buf + out_len); out_len += 8;
+  u32_to_hex (digest[1], out_buf + out_len); out_len += 8;
+  u32_to_hex (digest[2], out_buf + out_len); out_len += 8;
+
+  out_buf[out_len] = 0;
+
+  return out_len;
+}
+
+void module_init (module_ctx_t *module_ctx)
+{
+  module_ctx->module_context_size             = MODULE_CONTEXT_SIZE_CURRENT;
+  module_ctx->module_interface_version        = MODULE_INTERFACE_VERSION_CURRENT;
+
+  module_ctx->module_attack_exec              = module_attack_exec;
+  module_ctx->module_benchmark_esalt          = MODULE_DEFAULT;
+  module_ctx->module_benchmark_hook_salt      = MODULE_DEFAULT;
+  module_ctx->module_benchmark_mask           = MODULE_DEFAULT;
+  module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
+  module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
+  module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
+  module_ctx->module_dgst_pos0                = module_dgst_pos0;
+  module_ctx->module_dgst_pos1                = module_dgst_pos1;
+  module_ctx->module_dgst_pos2                = module_dgst_pos2;
+  module_ctx->module_dgst_pos3                = module_dgst_pos3;
+  module_ctx->module_dgst_size                = module_dgst_size;
+  module_ctx->module_dictstat_disable         = MODULE_DEFAULT;
+  module_ctx->module_esalt_size               = module_esalt_size;
+  module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
+  module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
+  module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_save         = MODULE_DEFAULT;
+  module_ctx->module_hash_decode_potfile      = MODULE_DEFAULT;
+  module_ctx->module_hash_decode_zero_hash    = MODULE_DEFAULT;
+  module_ctx->module_hash_decode              = module_hash_decode;
+  module_ctx->module_hash_encode_status       = MODULE_DEFAULT;
+  module_ctx->module_hash_encode_potfile      = MODULE_DEFAULT;
+  module_ctx->module_hash_encode              = module_hash_encode;
+  module_ctx->module_hash_init_selftest       = MODULE_DEFAULT;
+  module_ctx->module_hash_mode                = MODULE_DEFAULT;
+  module_ctx->module_hash_category            = module_hash_category;
+  module_ctx->module_hash_name                = module_hash_name;
+  module_ctx->module_hashes_count_min         = MODULE_DEFAULT;
+  module_ctx->module_hashes_count_max         = MODULE_DEFAULT;
+  module_ctx->module_hlfmt_disable            = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_size    = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_init    = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_term    = MODULE_DEFAULT;
+  module_ctx->module_hook12                   = MODULE_DEFAULT;
+  module_ctx->module_hook23                   = MODULE_DEFAULT;
+  module_ctx->module_hook_salt_size           = MODULE_DEFAULT;
+  module_ctx->module_hook_size                = MODULE_DEFAULT;
+  module_ctx->module_jit_build_options        = MODULE_DEFAULT;
+  module_ctx->module_jit_cache_disable        = MODULE_DEFAULT;
+  module_ctx->module_kernel_accel_max         = MODULE_DEFAULT;
+  module_ctx->module_kernel_accel_min         = MODULE_DEFAULT;
+  module_ctx->module_kernel_loops_max         = module_kernel_loops_max;
+  module_ctx->module_kernel_loops_min         = module_kernel_loops_min;
+  module_ctx->module_kernel_threads_max       = MODULE_DEFAULT;
+  module_ctx->module_kernel_threads_min       = MODULE_DEFAULT;
+  module_ctx->module_kern_type                = module_kern_type;
+  module_ctx->module_kern_type_dynamic        = MODULE_DEFAULT;
+  module_ctx->module_opti_type                = module_opti_type;
+  module_ctx->module_opts_type                = module_opts_type;
+  module_ctx->module_outfile_check_disable    = MODULE_DEFAULT;
+  module_ctx->module_outfile_check_nocomp     = MODULE_DEFAULT;
+  module_ctx->module_potfile_custom_check     = MODULE_DEFAULT;
+  module_ctx->module_potfile_disable          = MODULE_DEFAULT;
+  module_ctx->module_potfile_keep_all_hashes  = MODULE_DEFAULT;
+  module_ctx->module_pwdump_column            = MODULE_DEFAULT;
+  module_ctx->module_pw_max                   = MODULE_DEFAULT;
+  module_ctx->module_pw_min                   = module_pw_min;
+  module_ctx->module_salt_max                 = MODULE_DEFAULT;
+  module_ctx->module_salt_min                 = MODULE_DEFAULT;
+  module_ctx->module_salt_type                = module_salt_type;
+  module_ctx->module_separator                = MODULE_DEFAULT;
+  module_ctx->module_st_hash                  = module_st_hash;
+  module_ctx->module_st_pass                  = module_st_pass;
+  module_ctx->module_tmp_size                 = module_tmp_size;
+  module_ctx->module_unstable_warning         = MODULE_DEFAULT;
+  module_ctx->module_warmup_disable           = MODULE_DEFAULT;
+}
diff --git a/src/modules/module_25200.c b/src/modules/module_25200.c
new file mode 100644
index 000000000..985b9dce4
--- /dev/null
+++ b/src/modules/module_25200.c
@@ -0,0 +1,336 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+#include "common.h"
+#include "types.h"
+#include "modules.h"
+#include "bitops.h"
+#include "convert.h"
+#include "shared.h"
+#include "memory.h"
+#include "emu_inc_hash_sha1.h"
+
+static const u32   ATTACK_EXEC    = ATTACK_EXEC_OUTSIDE_KERNEL;
+static const u32   DGST_POS0      = 0;
+static const u32   DGST_POS1      = 1;
+static const u32   DGST_POS2      = 2;
+static const u32   DGST_POS3      = 3;
+static const u32   DGST_SIZE      = DGST_SIZE_4_4; // 4_3
+static const u32   HASH_CATEGORY  = HASH_CATEGORY_NETWORK_PROTOCOL;
+static const char *HASH_NAME      = "SNMPv3 HMAC-SHA1-96";
+static const u64   KERN_TYPE      = 25200;
+static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE;
+static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE;
+static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
+static const char *ST_PASS        = "hashcat1";
+static const char *ST_HASH        = "$SNMPv3$2$45889431$30818f02010330110204371780f3020300ffe304010102010304383036041180001f88808106d566db57fd600000000002011002020118040a6d61747269785f534841040c0000000000000000000000000400303d041180001f88808106d566db57fd60000000000400a2260204073557d50201000201003018301606082b06010201010200060a2b06010401bf0803020a$80001f88808106d566db57fd6000000000$81f14f1930589f26f6755f6b";
+
+u32         module_attack_exec    (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ATTACK_EXEC;     }
+u32         module_dgst_pos0      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS0;       }
+u32         module_dgst_pos1      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS1;       }
+u32         module_dgst_pos2      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS2;       }
+u32         module_dgst_pos3      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS3;       }
+u32         module_dgst_size      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_SIZE;       }
+u32         module_hash_category  (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return HASH_CATEGORY;   }
+const char *module_hash_name      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return HASH_NAME;       }
+u64         module_kern_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return KERN_TYPE;       }
+u32         module_opti_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return OPTI_TYPE;       }
+u64         module_opts_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return OPTS_TYPE;       }
+u32         module_salt_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return SALT_TYPE;       }
+const char *module_st_hash        (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ST_HASH;         }
+const char *module_st_pass        (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ST_PASS;         }
+
+static const char *SIGNATURE_SNMPV3 = "$SNMPv3$2$";
+
+#define SNMPV3_SALT_MAX             1500
+#define SNMPV3_ENGINEID_MAX         34
+#define SNMPV3_MSG_AUTH_PARAMS_LEN  12
+#define SNMPV3_ROUNDS               1048576
+#define SNMPV3_MAX_PW_LENGTH        64
+
+#define SNMPV3_TMP_ELEMS  4096 // 4096 = (256 (max pw length) * 64) / sizeof (u32)
+#define SNMPV3_HASH_ELEMS 8    // 8 = aligned 5
+
+typedef struct hmac_sha1_tmp
+{
+  u32 tmp[SNMPV3_TMP_ELEMS];
+  u32 h[SNMPV3_HASH_ELEMS];
+
+} hmac_sha1_tmp_t;
+
+#define SNMPV3_MAX_SALT_ELEMS    512 // 512 * 4 = 2048 > 1500, also has to be multiple of 64
+#define SNMPV3_MAX_ENGINE_ELEMS  16  // 16 * 4 = 64 > 32, also has to be multiple of 64
+#define SNMPV3_MAX_PNUM_ELEMS    4   // 4 * 4 = 16 > 9
+
+typedef struct snmpv3
+{
+  u32 salt_buf[SNMPV3_MAX_SALT_ELEMS];
+  u32 salt_len;
+
+  u32 engineID_buf[SNMPV3_MAX_ENGINE_ELEMS];
+  u32 engineID_len;
+
+  u32 packet_number[SNMPV3_MAX_PNUM_ELEMS];
+
+} snmpv3_t;
+
+u32 module_pw_min (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u32 pw_min = 8;
+
+  return pw_min;
+}
+
+u64 module_esalt_size (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u64 esalt_size = (const u64) sizeof (snmpv3_t);
+
+  return esalt_size;
+}
+
+u64 module_tmp_size (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u64 tmp_size = (const u64) sizeof (hmac_sha1_tmp_t);
+
+  return tmp_size;
+}
+
+u32 module_kernel_loops_min (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  // we need to fix iteration count to guarantee the loop count is a multiple of 64
+  // 2k calls to sha1_transform typically is enough to overtime pcie bottleneck
+
+  const u32 kernel_loops_min = 2048 * 64;
+
+  return kernel_loops_min;
+}
+
+u32 module_kernel_loops_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u32 kernel_loops_max = 2048 * 64;
+
+  return kernel_loops_max;
+}
+
+int module_hash_decode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED void *digest_buf, MAYBE_UNUSED salt_t *salt, MAYBE_UNUSED void *esalt_buf, MAYBE_UNUSED void *hook_salt_buf, MAYBE_UNUSED hashinfo_t *hash_info, const char *line_buf, MAYBE_UNUSED const int line_len)
+{
+  u32 *digest = (u32 *) digest_buf;
+
+  snmpv3_t *snmpv3 = (snmpv3_t *) esalt_buf;
+
+  token_t token;
+
+  token.token_cnt  = 5;
+  token.signatures_cnt    = 1;
+  token.signatures_buf[0] = SIGNATURE_SNMPV3;
+
+  token.len[0]     = 10;
+  token.attr[0]    = TOKEN_ATTR_FIXED_LENGTH
+                   | TOKEN_ATTR_VERIFY_SIGNATURE;
+
+  // packet number
+  token.len_min[1] = 1;
+  token.len_max[1] = 8;
+  token.sep[1]     = '$';
+  token.attr[1]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_DIGIT;
+  // salt
+  token.len_min[2] = SNMPV3_MSG_AUTH_PARAMS_LEN * 2;
+  token.len_max[2] = SNMPV3_SALT_MAX * 2;
+  token.sep[2]     = '$';
+  token.attr[2]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  // engineid
+  token.len_min[3] = 26;
+  token.len_max[3] = SNMPV3_ENGINEID_MAX;
+  token.sep[3]     = '$';
+  token.attr[3]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  // digest
+  token.len[4]     = SNMPV3_MSG_AUTH_PARAMS_LEN * 2;
+  token.sep[4]     = '$';
+  token.attr[4]    = TOKEN_ATTR_FIXED_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  const int rc_tokenizer = input_tokenizer ((const u8 *) line_buf, line_len, &token);
+
+  if (rc_tokenizer != PARSER_OK) return (rc_tokenizer);
+
+  // packet number
+
+  const u8 *packet_number_pos = token.buf[1];
+  const int packet_number_len = token.len[1];
+
+  memset (snmpv3->packet_number, 0, sizeof (snmpv3->packet_number));
+
+  strncpy ((char *) snmpv3->packet_number, (char *) packet_number_pos, packet_number_len);
+
+  // salt
+
+  const u8 *salt_pos = token.buf[2];
+  const int salt_len = token.len[2];
+
+  u8 *salt_ptr = (u8 *) snmpv3->salt_buf;
+
+  snmpv3->salt_len = hex_decode (salt_pos, salt_len, salt_ptr);
+
+  salt->salt_iter = SNMPV3_ROUNDS;
+
+  // handle unique salts detection
+
+  sha1_ctx_t sha1_ctx;
+
+  sha1_init   (&sha1_ctx);
+  sha1_update (&sha1_ctx, snmpv3->salt_buf, snmpv3->salt_len);
+  sha1_final  (&sha1_ctx);
+
+  // store sha1(snmpv3->salt_buf) in salt_buf
+
+  salt->salt_len = 20;
+
+  memcpy (salt->salt_buf, sha1_ctx.h, salt->salt_len);
+
+  // engineid
+
+  const u8 *engineID_pos = token.buf[3];
+  const int engineID_len = token.len[3];
+
+  u8 *engineID_ptr = (u8 *) snmpv3->engineID_buf;
+
+  snmpv3->engineID_len = hex_decode (engineID_pos, engineID_len, engineID_ptr);
+
+  // digest
+
+  const u8 *hash_pos = token.buf[4];
+
+  digest[0] = hex_to_u32 (hash_pos +  0);
+  digest[1] = hex_to_u32 (hash_pos +  8);
+  digest[2] = hex_to_u32 (hash_pos + 16);
+
+  digest[0] = byte_swap_32 (digest[0]);
+  digest[1] = byte_swap_32 (digest[1]);
+  digest[2] = byte_swap_32 (digest[2]);
+
+  digest[3] = 0;
+
+  return (PARSER_OK);
+}
+
+int module_hash_encode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const void *digest_buf, MAYBE_UNUSED const salt_t *salt, MAYBE_UNUSED const void *esalt_buf, MAYBE_UNUSED const void *hook_salt_buf, MAYBE_UNUSED const hashinfo_t *hash_info, char *line_buf, MAYBE_UNUSED const int line_size)
+{
+  const u32 *digest = (const u32 *) digest_buf;
+
+  snmpv3_t *snmpv3 = (snmpv3_t *) esalt_buf;
+
+  u8 *out_buf = (u8 *) line_buf;
+
+  int out_len = snprintf (line_buf, line_size, "%s%s$", SIGNATURE_SNMPV3, (char *) snmpv3->packet_number);
+
+  out_len += hex_encode ((u8 *) snmpv3->salt_buf, snmpv3->salt_len, out_buf + out_len);
+
+  out_buf[out_len] = '$';
+
+  out_len++;
+
+  out_len += hex_encode ((u8 *) snmpv3->engineID_buf, snmpv3->engineID_len, out_buf + out_len);
+
+  out_buf[out_len] = '$';
+
+  out_len++;
+
+  u32 digest_tmp[3];
+
+  digest_tmp[0] = byte_swap_32 (digest[0]);
+  digest_tmp[1] = byte_swap_32 (digest[1]);
+  digest_tmp[2] = byte_swap_32 (digest[2]);
+
+  u32_to_hex (digest_tmp[0], out_buf + out_len); out_len += 8;
+  u32_to_hex (digest_tmp[1], out_buf + out_len); out_len += 8;
+  u32_to_hex (digest_tmp[2], out_buf + out_len); out_len += 8;
+
+  out_buf[out_len] = 0;
+
+  return out_len;
+}
+
+void module_init (module_ctx_t *module_ctx)
+{
+  module_ctx->module_context_size             = MODULE_CONTEXT_SIZE_CURRENT;
+  module_ctx->module_interface_version        = MODULE_INTERFACE_VERSION_CURRENT;
+
+  module_ctx->module_attack_exec              = module_attack_exec;
+  module_ctx->module_benchmark_esalt          = MODULE_DEFAULT;
+  module_ctx->module_benchmark_hook_salt      = MODULE_DEFAULT;
+  module_ctx->module_benchmark_mask           = MODULE_DEFAULT;
+  module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
+  module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
+  module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
+  module_ctx->module_dgst_pos0                = module_dgst_pos0;
+  module_ctx->module_dgst_pos1                = module_dgst_pos1;
+  module_ctx->module_dgst_pos2                = module_dgst_pos2;
+  module_ctx->module_dgst_pos3                = module_dgst_pos3;
+  module_ctx->module_dgst_size                = module_dgst_size;
+  module_ctx->module_dictstat_disable         = MODULE_DEFAULT;
+  module_ctx->module_esalt_size               = module_esalt_size;
+  module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
+  module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
+  module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_save         = MODULE_DEFAULT;
+  module_ctx->module_hash_decode_potfile      = MODULE_DEFAULT;
+  module_ctx->module_hash_decode_zero_hash    = MODULE_DEFAULT;
+  module_ctx->module_hash_decode              = module_hash_decode;
+  module_ctx->module_hash_encode_status       = MODULE_DEFAULT;
+  module_ctx->module_hash_encode_potfile      = MODULE_DEFAULT;
+  module_ctx->module_hash_encode              = module_hash_encode;
+  module_ctx->module_hash_init_selftest       = MODULE_DEFAULT;
+  module_ctx->module_hash_mode                = MODULE_DEFAULT;
+  module_ctx->module_hash_category            = module_hash_category;
+  module_ctx->module_hash_name                = module_hash_name;
+  module_ctx->module_hashes_count_min         = MODULE_DEFAULT;
+  module_ctx->module_hashes_count_max         = MODULE_DEFAULT;
+  module_ctx->module_hlfmt_disable            = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_size    = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_init    = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_term    = MODULE_DEFAULT;
+  module_ctx->module_hook12                   = MODULE_DEFAULT;
+  module_ctx->module_hook23                   = MODULE_DEFAULT;
+  module_ctx->module_hook_salt_size           = MODULE_DEFAULT;
+  module_ctx->module_hook_size                = MODULE_DEFAULT;
+  module_ctx->module_jit_build_options        = MODULE_DEFAULT;
+  module_ctx->module_jit_cache_disable        = MODULE_DEFAULT;
+  module_ctx->module_kernel_accel_max         = MODULE_DEFAULT;
+  module_ctx->module_kernel_accel_min         = MODULE_DEFAULT;
+  module_ctx->module_kernel_loops_max         = module_kernel_loops_max;
+  module_ctx->module_kernel_loops_min         = module_kernel_loops_min;
+  module_ctx->module_kernel_threads_max       = MODULE_DEFAULT;
+  module_ctx->module_kernel_threads_min       = MODULE_DEFAULT;
+  module_ctx->module_kern_type                = module_kern_type;
+  module_ctx->module_kern_type_dynamic        = MODULE_DEFAULT;
+  module_ctx->module_opti_type                = module_opti_type;
+  module_ctx->module_opts_type                = module_opts_type;
+  module_ctx->module_outfile_check_disable    = MODULE_DEFAULT;
+  module_ctx->module_outfile_check_nocomp     = MODULE_DEFAULT;
+  module_ctx->module_potfile_custom_check     = MODULE_DEFAULT;
+  module_ctx->module_potfile_disable          = MODULE_DEFAULT;
+  module_ctx->module_potfile_keep_all_hashes  = MODULE_DEFAULT;
+  module_ctx->module_pwdump_column            = MODULE_DEFAULT;
+  module_ctx->module_pw_max                   = MODULE_DEFAULT;
+  module_ctx->module_pw_min                   = module_pw_min;
+  module_ctx->module_salt_max                 = MODULE_DEFAULT;
+  module_ctx->module_salt_min                 = MODULE_DEFAULT;
+  module_ctx->module_salt_type                = module_salt_type;
+  module_ctx->module_separator                = MODULE_DEFAULT;
+  module_ctx->module_st_hash                  = module_st_hash;
+  module_ctx->module_st_pass                  = module_st_pass;
+  module_ctx->module_tmp_size                 = module_tmp_size;
+  module_ctx->module_unstable_warning         = MODULE_DEFAULT;
+  module_ctx->module_warmup_disable           = MODULE_DEFAULT;
+}
diff --git a/src/modules/module_25300.c b/src/modules/module_25300.c
index e21d55881..fc3b3c084 100644
--- a/src/modules/module_25300.c
+++ b/src/modules/module_25300.c
@@ -251,6 +251,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -260,6 +261,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_25400.c b/src/modules/module_25400.c
index 341837786..21d3ccfc5 100644
--- a/src/modules/module_25400.c
+++ b/src/modules/module_25400.c
@@ -103,35 +103,13 @@ char *module_jit_build_options (MAYBE_UNUSED const hashconfig_t *hashconfig, MAY
   }
   else if (device_param->opencl_device_type & CL_DEVICE_TYPE_GPU)
   {
-    if (device_param->opencl_device_vendor_id == VENDOR_ID_INTEL_SDK)
+    if (device_param->device_local_mem_size < 49152)
     {
-      native_threads = 8;
-    }
-    else if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD)
-    {
-      if (device_param->device_local_mem_size < 49152)
-      {
-        native_threads = 32;
-      }
-      else
-      {
-        native_threads = 64;
-      }
-    }
-    else if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD_USE_HIP)
-    {
-      if (device_param->device_local_mem_size < 49152)
-      {
-        native_threads = 32;
-      }
-      else
-      {
-        native_threads = 64;
-      }
+      native_threads = MIN (device_param->kernel_preferred_wgs_multiple, 32); // We can't just set 32, because Intel GPU need 8
     }
     else
     {
-      native_threads = 32;
+      native_threads = device_param->kernel_preferred_wgs_multiple;
     }
   }
 
@@ -479,6 +457,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -488,6 +467,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_25500.c b/src/modules/module_25500.c
index e6853e951..535f176d1 100644
--- a/src/modules/module_25500.c
+++ b/src/modules/module_25500.c
@@ -247,9 +247,9 @@ int module_hash_encode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSE
 
   // salt
 
-  #define SALT_LEN_BASE64 ((16 * 8) / 6) + 3
-  #define IV_LEN_BASE64   ((12 * 8) / 6) + 3
-  #define CT_LEN_BASE64   ((72 * 8) / 6) + 3
+  #define SALT_LEN_BASE64 ((16 * 8) / 6) + 3 + 1 // 25 vs 24
+  #define IV_LEN_BASE64   ((12 * 8) / 6) + 1     // 17 vs 16
+  #define CT_LEN_BASE64   ((72 * 8) / 6) + 1     // 97 vs 96
 
   u8 salt_buf[SALT_LEN_BASE64] = { 0 };
 
@@ -305,6 +305,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -314,6 +315,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_25600.c b/src/modules/module_25600.c
index 561c04d77..202ec3d67 100644
--- a/src/modules/module_25600.c
+++ b/src/modules/module_25600.c
@@ -287,6 +287,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -296,6 +297,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_25700.c b/src/modules/module_25700.c
index a142ea26c..7573e81ce 100644
--- a/src/modules/module_25700.c
+++ b/src/modules/module_25700.c
@@ -125,6 +125,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -134,6 +135,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_25800.c b/src/modules/module_25800.c
index 154e9bc5c..26bc3ceca 100644
--- a/src/modules/module_25800.c
+++ b/src/modules/module_25800.c
@@ -287,6 +287,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -296,6 +297,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_25900.c b/src/modules/module_25900.c
index c44a1642b..237910685 100644
--- a/src/modules/module_25900.c
+++ b/src/modules/module_25900.c
@@ -273,6 +273,7 @@ void module_init(module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = module_deep_comp_kernel;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -282,6 +283,7 @@ void module_init(module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_26000.c b/src/modules/module_26000.c
index 585a92fdb..1ed16cdb2 100644
--- a/src/modules/module_26000.c
+++ b/src/modules/module_26000.c
@@ -198,6 +198,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -207,6 +208,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_26100.c b/src/modules/module_26100.c
index 2a6478828..d00d85f44 100644
--- a/src/modules/module_26100.c
+++ b/src/modules/module_26100.c
@@ -254,6 +254,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -263,6 +264,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_26200.c b/src/modules/module_26200.c
index 8b082f2d5..c089eb594 100644
--- a/src/modules/module_26200.c
+++ b/src/modules/module_26200.c
@@ -132,6 +132,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -141,6 +142,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_26300.c b/src/modules/module_26300.c
index a07167582..086b9b631 100644
--- a/src/modules/module_26300.c
+++ b/src/modules/module_26300.c
@@ -198,6 +198,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -207,6 +208,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_26401.c b/src/modules/module_26401.c
index d8341b6b6..3bbbea1ec 100644
--- a/src/modules/module_26401.c
+++ b/src/modules/module_26401.c
@@ -139,6 +139,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -148,6 +149,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_26402.c b/src/modules/module_26402.c
index ecf1f03a1..6e2b65075 100644
--- a/src/modules/module_26402.c
+++ b/src/modules/module_26402.c
@@ -139,6 +139,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -148,6 +149,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_26403.c b/src/modules/module_26403.c
index c7c5fd11e..a376dfd09 100644
--- a/src/modules/module_26403.c
+++ b/src/modules/module_26403.c
@@ -139,6 +139,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -148,6 +149,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_26500.c b/src/modules/module_26500.c
index b473bd0fb..018bc7d91 100644
--- a/src/modules/module_26500.c
+++ b/src/modules/module_26500.c
@@ -250,6 +250,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -259,6 +260,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_26600.c b/src/modules/module_26600.c
index c9e04958c..0f2a216d7 100644
--- a/src/modules/module_26600.c
+++ b/src/modules/module_26600.c
@@ -325,6 +325,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -334,6 +335,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = module_esalt_size;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/modules/module_26700.c b/src/modules/module_26700.c
new file mode 100644
index 000000000..14cceb0aa
--- /dev/null
+++ b/src/modules/module_26700.c
@@ -0,0 +1,338 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+#include "common.h"
+#include "types.h"
+#include "modules.h"
+#include "bitops.h"
+#include "convert.h"
+#include "shared.h"
+#include "memory.h"
+#include "emu_inc_hash_sha1.h"
+
+static const u32   ATTACK_EXEC    = ATTACK_EXEC_OUTSIDE_KERNEL;
+static const u32   DGST_POS0      = 0;
+static const u32   DGST_POS1      = 1;
+static const u32   DGST_POS2      = 2;
+static const u32   DGST_POS3      = 3;
+static const u32   DGST_SIZE      = DGST_SIZE_4_4;
+static const u32   HASH_CATEGORY  = HASH_CATEGORY_NETWORK_PROTOCOL;
+static const char *HASH_NAME      = "SNMPv3 HMAC-SHA224-128";
+static const u64   KERN_TYPE      = 26700;
+static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE;
+static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE;
+static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
+static const char *ST_PASS        = "hashcat1";
+static const char *ST_HASH        = "$SNMPv3$3$45889431$308197020103301102047aa1a79e020300ffe30401010201030440303e041180001f88808106d566db57fd600000000002011002020118040e6d61747269785f5348412d3232340410000000000000000000000000000000000400303d041180001f88808106d566db57fd60000000000400a2260204272f76620201000201003018301606082b06010201010200060a2b06010401bf0803020a$80001f88808106d566db57fd6000000000$2f7a3891dd2e27d3f567e4d6d0257962";
+
+u32         module_attack_exec    (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ATTACK_EXEC;     }
+u32         module_dgst_pos0      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS0;       }
+u32         module_dgst_pos1      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS1;       }
+u32         module_dgst_pos2      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS2;       }
+u32         module_dgst_pos3      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS3;       }
+u32         module_dgst_size      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_SIZE;       }
+u32         module_hash_category  (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return HASH_CATEGORY;   }
+const char *module_hash_name      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return HASH_NAME;       }
+u64         module_kern_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return KERN_TYPE;       }
+u32         module_opti_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return OPTI_TYPE;       }
+u64         module_opts_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return OPTS_TYPE;       }
+u32         module_salt_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return SALT_TYPE;       }
+const char *module_st_hash        (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ST_HASH;         }
+const char *module_st_pass        (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ST_PASS;         }
+
+static const char *SIGNATURE_SNMPV3 = "$SNMPv3$3$";
+
+#define SNMPV3_SALT_MAX             1500
+#define SNMPV3_ENGINEID_MAX         34
+#define SNMPV3_MSG_AUTH_PARAMS_LEN  16
+#define SNMPV3_ROUNDS               1048576
+#define SNMPV3_MAX_PW_LENGTH        64
+
+#define SNMPV3_TMP_ELEMS            4096 // 4096 = (256 (max pw length) * 64) / sizeof (u32)
+#define SNMPV3_HASH_ELEMS           8
+
+#define SNMPV3_MAX_SALT_ELEMS       512 // 512 * 4 = 2048 > 1500, also has to be multiple of 64
+#define SNMPV3_MAX_ENGINE_ELEMS     16  // 16 * 4 = 64 > 32, also has to be multiple of 64
+#define SNMPV3_MAX_PNUM_ELEMS       4   // 4 * 4 = 16 > 9
+
+typedef struct hmac_sha224_tmp
+{
+  u32 tmp[SNMPV3_TMP_ELEMS];
+  u32 h[SNMPV3_HASH_ELEMS];
+
+} hmac_sha224_tmp_t;
+
+typedef struct snmpv3
+{
+  u32 salt_buf[SNMPV3_MAX_SALT_ELEMS];
+  u32 salt_len;
+
+  u32 engineID_buf[SNMPV3_MAX_ENGINE_ELEMS];
+  u32 engineID_len;
+
+  u32 packet_number[SNMPV3_MAX_PNUM_ELEMS];
+
+} snmpv3_t;
+
+u32 module_pw_min (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u32 pw_min = 8;
+
+  return pw_min;
+}
+
+u64 module_esalt_size (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u64 esalt_size = (const u64) sizeof (snmpv3_t);
+
+  return esalt_size;
+}
+
+u64 module_tmp_size (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u64 tmp_size = (const u64) sizeof (hmac_sha224_tmp_t);
+
+  return tmp_size;
+}
+
+u32 module_kernel_loops_min (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  // we need to fix iteration count to guarantee the loop count is a multiple of 64
+  // 2k calls to sha224_transform typically is enough to overtime pcie bottleneck
+
+  const u32 kernel_loops_min = 2048 * 64;
+
+  return kernel_loops_min;
+}
+
+u32 module_kernel_loops_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u32 kernel_loops_max = 2048 * 64;
+
+  return kernel_loops_max;
+}
+
+int module_hash_decode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED void *digest_buf, MAYBE_UNUSED salt_t *salt, MAYBE_UNUSED void *esalt_buf, MAYBE_UNUSED void *hook_salt_buf, MAYBE_UNUSED hashinfo_t *hash_info, const char *line_buf, MAYBE_UNUSED const int line_len)
+{
+  u32 *digest = (u32 *) digest_buf;
+
+  snmpv3_t *snmpv3 = (snmpv3_t *) esalt_buf;
+
+  token_t token;
+
+  token.token_cnt  = 5;
+  token.signatures_cnt    = 1;
+  token.signatures_buf[0] = SIGNATURE_SNMPV3;
+
+  token.len[0]     = 10;
+  token.attr[0]    = TOKEN_ATTR_FIXED_LENGTH
+                   | TOKEN_ATTR_VERIFY_SIGNATURE;
+
+  // packet number
+  token.len_min[1] = 1;
+  token.len_max[1] = 8;
+  token.sep[1]     = '$';
+  token.attr[1]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_DIGIT;
+  // salt
+  token.len_min[2] = SNMPV3_MSG_AUTH_PARAMS_LEN * 2;
+  token.len_max[2] = SNMPV3_SALT_MAX * 2;
+  token.sep[2]     = '$';
+  token.attr[2]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  // engineid
+  token.len_min[3] = 26;
+  token.len_max[3] = SNMPV3_ENGINEID_MAX;
+  token.sep[3]     = '$';
+  token.attr[3]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  // digest
+  token.len[4]     = SNMPV3_MSG_AUTH_PARAMS_LEN * 2;
+  token.sep[4]     = '$';
+  token.attr[4]    = TOKEN_ATTR_FIXED_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  const int rc_tokenizer = input_tokenizer ((const u8 *) line_buf, line_len, &token);
+
+  if (rc_tokenizer != PARSER_OK) return (rc_tokenizer);
+
+  // packet number
+
+  const u8 *packet_number_pos = token.buf[1];
+  const int packet_number_len = token.len[1];
+
+  memset (snmpv3->packet_number, 0, sizeof (snmpv3->packet_number));
+
+  strncpy ((char *) snmpv3->packet_number, (char *) packet_number_pos, packet_number_len);
+
+  // salt
+
+  const u8 *salt_pos = token.buf[2];
+  const int salt_len = token.len[2];
+
+  u8 *salt_ptr = (u8 *) snmpv3->salt_buf;
+
+  snmpv3->salt_len = hex_decode (salt_pos, salt_len, salt_ptr);
+
+  salt->salt_iter = SNMPV3_ROUNDS;
+
+  // handle unique salts detection
+
+  sha1_ctx_t sha1_ctx;
+
+  sha1_init   (&sha1_ctx);
+  sha1_update (&sha1_ctx, snmpv3->salt_buf, snmpv3->salt_len);
+  sha1_final  (&sha1_ctx);
+
+  // store sha1(snmpv3->salt_buf) in salt_buf
+
+  salt->salt_len = 20;
+
+  memcpy (salt->salt_buf, sha1_ctx.h, salt->salt_len);
+
+  // engineid
+
+  const u8 *engineID_pos = token.buf[3];
+  const int engineID_len = token.len[3];
+
+  u8 *engineID_ptr = (u8 *) snmpv3->engineID_buf;
+
+  snmpv3->engineID_len = hex_decode (engineID_pos, engineID_len, engineID_ptr);
+
+  // digest
+
+  const u8 *hash_pos = token.buf[4];
+
+  digest[0] = hex_to_u32 (hash_pos +  0);
+  digest[1] = hex_to_u32 (hash_pos +  8);
+  digest[2] = hex_to_u32 (hash_pos + 16);
+  digest[3] = hex_to_u32 (hash_pos + 24);
+
+  digest[0] = byte_swap_32 (digest[0]);
+  digest[1] = byte_swap_32 (digest[1]);
+  digest[2] = byte_swap_32 (digest[2]);
+  digest[3] = byte_swap_32 (digest[3]);
+
+  return (PARSER_OK);
+}
+
+int module_hash_encode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const void *digest_buf, MAYBE_UNUSED const salt_t *salt, MAYBE_UNUSED const void *esalt_buf, MAYBE_UNUSED const void *hook_salt_buf, MAYBE_UNUSED const hashinfo_t *hash_info, char *line_buf, MAYBE_UNUSED const int line_size)
+{
+  const u32 *digest = (const u32 *) digest_buf;
+
+  snmpv3_t *snmpv3 = (snmpv3_t *) esalt_buf;
+
+  u8 *out_buf = (u8 *) line_buf;
+
+  int out_len = snprintf (line_buf, line_size, "%s%s$", SIGNATURE_SNMPV3, (char *) snmpv3->packet_number);
+
+  out_len += hex_encode ((u8 *) snmpv3->salt_buf, snmpv3->salt_len, out_buf + out_len);
+
+  out_buf[out_len] = '$';
+
+  out_len++;
+
+  out_len += hex_encode ((u8 *) snmpv3->engineID_buf, snmpv3->engineID_len, out_buf + out_len);
+
+  out_buf[out_len] = '$';
+
+  out_len++;
+
+  u32 digest_tmp[4];
+
+  digest_tmp[0] = byte_swap_32 (digest[0]);
+  digest_tmp[1] = byte_swap_32 (digest[1]);
+  digest_tmp[2] = byte_swap_32 (digest[2]);
+  digest_tmp[3] = byte_swap_32 (digest[3]);
+
+  u32_to_hex (digest_tmp[0], out_buf + out_len); out_len += 8;
+  u32_to_hex (digest_tmp[1], out_buf + out_len); out_len += 8;
+  u32_to_hex (digest_tmp[2], out_buf + out_len); out_len += 8;
+  u32_to_hex (digest_tmp[3], out_buf + out_len); out_len += 8;
+
+  out_buf[out_len] = 0;
+
+  return out_len;
+}
+
+void module_init (module_ctx_t *module_ctx)
+{
+  module_ctx->module_context_size             = MODULE_CONTEXT_SIZE_CURRENT;
+  module_ctx->module_interface_version        = MODULE_INTERFACE_VERSION_CURRENT;
+
+  module_ctx->module_attack_exec              = module_attack_exec;
+  module_ctx->module_benchmark_esalt          = MODULE_DEFAULT;
+  module_ctx->module_benchmark_hook_salt      = MODULE_DEFAULT;
+  module_ctx->module_benchmark_mask           = MODULE_DEFAULT;
+  module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
+  module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
+  module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
+  module_ctx->module_dgst_pos0                = module_dgst_pos0;
+  module_ctx->module_dgst_pos1                = module_dgst_pos1;
+  module_ctx->module_dgst_pos2                = module_dgst_pos2;
+  module_ctx->module_dgst_pos3                = module_dgst_pos3;
+  module_ctx->module_dgst_size                = module_dgst_size;
+  module_ctx->module_dictstat_disable         = MODULE_DEFAULT;
+  module_ctx->module_esalt_size               = module_esalt_size;
+  module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
+  module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
+  module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_save         = MODULE_DEFAULT;
+  module_ctx->module_hash_decode_potfile      = MODULE_DEFAULT;
+  module_ctx->module_hash_decode_zero_hash    = MODULE_DEFAULT;
+  module_ctx->module_hash_decode              = module_hash_decode;
+  module_ctx->module_hash_encode_status       = MODULE_DEFAULT;
+  module_ctx->module_hash_encode_potfile      = MODULE_DEFAULT;
+  module_ctx->module_hash_encode              = module_hash_encode;
+  module_ctx->module_hash_init_selftest       = MODULE_DEFAULT;
+  module_ctx->module_hash_mode                = MODULE_DEFAULT;
+  module_ctx->module_hash_category            = module_hash_category;
+  module_ctx->module_hash_name                = module_hash_name;
+  module_ctx->module_hashes_count_min         = MODULE_DEFAULT;
+  module_ctx->module_hashes_count_max         = MODULE_DEFAULT;
+  module_ctx->module_hlfmt_disable            = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_size    = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_init    = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_term    = MODULE_DEFAULT;
+  module_ctx->module_hook12                   = MODULE_DEFAULT;
+  module_ctx->module_hook23                   = MODULE_DEFAULT;
+  module_ctx->module_hook_salt_size           = MODULE_DEFAULT;
+  module_ctx->module_hook_size                = MODULE_DEFAULT;
+  module_ctx->module_jit_build_options        = MODULE_DEFAULT;
+  module_ctx->module_jit_cache_disable        = MODULE_DEFAULT;
+  module_ctx->module_kernel_accel_max         = MODULE_DEFAULT;
+  module_ctx->module_kernel_accel_min         = MODULE_DEFAULT;
+  module_ctx->module_kernel_loops_max         = module_kernel_loops_max;
+  module_ctx->module_kernel_loops_min         = module_kernel_loops_min;
+  module_ctx->module_kernel_threads_max       = MODULE_DEFAULT;
+  module_ctx->module_kernel_threads_min       = MODULE_DEFAULT;
+  module_ctx->module_kern_type                = module_kern_type;
+  module_ctx->module_kern_type_dynamic        = MODULE_DEFAULT;
+  module_ctx->module_opti_type                = module_opti_type;
+  module_ctx->module_opts_type                = module_opts_type;
+  module_ctx->module_outfile_check_disable    = MODULE_DEFAULT;
+  module_ctx->module_outfile_check_nocomp     = MODULE_DEFAULT;
+  module_ctx->module_potfile_custom_check     = MODULE_DEFAULT;
+  module_ctx->module_potfile_disable          = MODULE_DEFAULT;
+  module_ctx->module_potfile_keep_all_hashes  = MODULE_DEFAULT;
+  module_ctx->module_pwdump_column            = MODULE_DEFAULT;
+  module_ctx->module_pw_max                   = MODULE_DEFAULT;
+  module_ctx->module_pw_min                   = module_pw_min;
+  module_ctx->module_salt_max                 = MODULE_DEFAULT;
+  module_ctx->module_salt_min                 = MODULE_DEFAULT;
+  module_ctx->module_salt_type                = module_salt_type;
+  module_ctx->module_separator                = MODULE_DEFAULT;
+  module_ctx->module_st_hash                  = module_st_hash;
+  module_ctx->module_st_pass                  = module_st_pass;
+  module_ctx->module_tmp_size                 = module_tmp_size;
+  module_ctx->module_unstable_warning         = MODULE_DEFAULT;
+  module_ctx->module_warmup_disable           = MODULE_DEFAULT;
+}
diff --git a/src/modules/module_26800.c b/src/modules/module_26800.c
new file mode 100644
index 000000000..6b565b6da
--- /dev/null
+++ b/src/modules/module_26800.c
@@ -0,0 +1,346 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+#include "common.h"
+#include "types.h"
+#include "modules.h"
+#include "bitops.h"
+#include "convert.h"
+#include "shared.h"
+#include "memory.h"
+#include "emu_inc_hash_sha1.h"
+
+static const u32   ATTACK_EXEC    = ATTACK_EXEC_OUTSIDE_KERNEL;
+static const u32   DGST_POS0      = 0;
+static const u32   DGST_POS1      = 1;
+static const u32   DGST_POS2      = 2;
+static const u32   DGST_POS3      = 3;
+static const u32   DGST_SIZE      = DGST_SIZE_4_6;
+static const u32   HASH_CATEGORY  = HASH_CATEGORY_NETWORK_PROTOCOL;
+static const char *HASH_NAME      = "SNMPv3 HMAC-SHA256-192";
+static const u64   KERN_TYPE      = 26800;
+static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE;
+static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE;
+static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
+static const char *ST_PASS        = "hashcat1";
+static const char *ST_HASH        = "$SNMPv3$4$45889431$30819f020103301102047fc51818020300ffe304010102010304483046041180001f88808106d566db57fd600000000002011002020118040e6d61747269785f5348412d32353604180000000000000000000000000000000000000000000000000400303d041180001f88808106d566db57fd60000000000400a22602040efec2600201000201003018301606082b06010201010200060a2b06010401bf0803020a$80001f88808106d566db57fd6000000000$36d655bfeb59e933845db47d719b68ac7bc59ec087eb89a0";
+
+u32         module_attack_exec    (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ATTACK_EXEC;     }
+u32         module_dgst_pos0      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS0;       }
+u32         module_dgst_pos1      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS1;       }
+u32         module_dgst_pos2      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS2;       }
+u32         module_dgst_pos3      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS3;       }
+u32         module_dgst_size      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_SIZE;       }
+u32         module_hash_category  (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return HASH_CATEGORY;   }
+const char *module_hash_name      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return HASH_NAME;       }
+u64         module_kern_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return KERN_TYPE;       }
+u32         module_opti_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return OPTI_TYPE;       }
+u64         module_opts_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return OPTS_TYPE;       }
+u32         module_salt_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return SALT_TYPE;       }
+const char *module_st_hash        (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ST_HASH;         }
+const char *module_st_pass        (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ST_PASS;         }
+
+static const char *SIGNATURE_SNMPV3 = "$SNMPv3$4$";
+
+#define SNMPV3_SALT_MAX             1500
+#define SNMPV3_ENGINEID_MAX         34
+#define SNMPV3_MSG_AUTH_PARAMS_LEN  24
+#define SNMPV3_ROUNDS               1048576
+#define SNMPV3_MAX_PW_LENGTH        64
+
+#define SNMPV3_TMP_ELEMS            4096 // 4096 = (256 (max pw length) * 64) / sizeof (u32)
+#define SNMPV3_HASH_ELEMS           8
+
+#define SNMPV3_MAX_SALT_ELEMS       512 // 512 * 4 = 2048 > 1500, also has to be multiple of 64
+#define SNMPV3_MAX_ENGINE_ELEMS     16  // 16 * 4 = 64 > 32, also has to be multiple of 64
+#define SNMPV3_MAX_PNUM_ELEMS       4   // 4 * 4 = 16 > 9
+
+typedef struct hmac_sha256_tmp
+{
+  u32 tmp[SNMPV3_TMP_ELEMS];
+  u32 h[SNMPV3_HASH_ELEMS];
+
+} hmac_sha256_tmp_t;
+
+typedef struct snmpv3
+{
+  u32 salt_buf[SNMPV3_MAX_SALT_ELEMS];
+  u32 salt_len;
+
+  u32 engineID_buf[SNMPV3_MAX_ENGINE_ELEMS];
+  u32 engineID_len;
+
+  u32 packet_number[SNMPV3_MAX_PNUM_ELEMS];
+
+} snmpv3_t;
+
+u32 module_pw_min (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u32 pw_min = 8;
+
+  return pw_min;
+}
+
+u64 module_esalt_size (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u64 esalt_size = (const u64) sizeof (snmpv3_t);
+
+  return esalt_size;
+}
+
+u64 module_tmp_size (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u64 tmp_size = (const u64) sizeof (hmac_sha256_tmp_t);
+
+  return tmp_size;
+}
+
+u32 module_kernel_loops_min (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  // we need to fix iteration count to guarantee the loop count is a multiple of 64
+  // 2k calls to sha256_transform typically is enough to overtime pcie bottleneck
+
+  const u32 kernel_loops_min = 2048 * 64;
+
+  return kernel_loops_min;
+}
+
+u32 module_kernel_loops_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u32 kernel_loops_max = 2048 * 64;
+
+  return kernel_loops_max;
+}
+
+int module_hash_decode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED void *digest_buf, MAYBE_UNUSED salt_t *salt, MAYBE_UNUSED void *esalt_buf, MAYBE_UNUSED void *hook_salt_buf, MAYBE_UNUSED hashinfo_t *hash_info, const char *line_buf, MAYBE_UNUSED const int line_len)
+{
+  u32 *digest = (u32 *) digest_buf;
+
+  snmpv3_t *snmpv3 = (snmpv3_t *) esalt_buf;
+
+  token_t token;
+
+  token.token_cnt  = 5;
+  token.signatures_cnt    = 1;
+  token.signatures_buf[0] = SIGNATURE_SNMPV3;
+
+  token.len[0]     = 10;
+  token.attr[0]    = TOKEN_ATTR_FIXED_LENGTH
+                   | TOKEN_ATTR_VERIFY_SIGNATURE;
+
+  // packet number
+  token.len_min[1] = 1;
+  token.len_max[1] = 8;
+  token.sep[1]     = '$';
+  token.attr[1]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_DIGIT;
+  // salt
+  token.len_min[2] = SNMPV3_MSG_AUTH_PARAMS_LEN * 2;
+  token.len_max[2] = SNMPV3_SALT_MAX * 2;
+  token.sep[2]     = '$';
+  token.attr[2]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  // engineid
+  token.len_min[3] = 26;
+  token.len_max[3] = SNMPV3_ENGINEID_MAX;
+  token.sep[3]     = '$';
+  token.attr[3]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  // digest
+  token.len[4]     = SNMPV3_MSG_AUTH_PARAMS_LEN * 2;
+  token.sep[4]     = '$';
+  token.attr[4]    = TOKEN_ATTR_FIXED_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  const int rc_tokenizer = input_tokenizer ((const u8 *) line_buf, line_len, &token);
+
+  if (rc_tokenizer != PARSER_OK) return (rc_tokenizer);
+
+  // packet number
+
+  const u8 *packet_number_pos = token.buf[1];
+  const int packet_number_len = token.len[1];
+
+  memset (snmpv3->packet_number, 0, sizeof (snmpv3->packet_number));
+
+  strncpy ((char *) snmpv3->packet_number, (char *) packet_number_pos, packet_number_len);
+
+  // salt
+
+  const u8 *salt_pos = token.buf[2];
+  const int salt_len = token.len[2];
+
+  u8 *salt_ptr = (u8 *) snmpv3->salt_buf;
+
+  snmpv3->salt_len = hex_decode (salt_pos, salt_len, salt_ptr);
+
+  salt->salt_iter = SNMPV3_ROUNDS;
+
+  // handle unique salts detection
+
+  sha1_ctx_t sha1_ctx;
+
+  sha1_init   (&sha1_ctx);
+  sha1_update (&sha1_ctx, snmpv3->salt_buf, snmpv3->salt_len);
+  sha1_final  (&sha1_ctx);
+
+  // store sha1(snmpv3->salt_buf) in salt_buf
+
+  salt->salt_len = 20;
+
+  memcpy (salt->salt_buf, sha1_ctx.h, salt->salt_len);
+
+  // engineid
+
+  const u8 *engineID_pos = token.buf[3];
+  const int engineID_len = token.len[3];
+
+  u8 *engineID_ptr = (u8 *) snmpv3->engineID_buf;
+
+  snmpv3->engineID_len = hex_decode (engineID_pos, engineID_len, engineID_ptr);
+
+  // digest
+
+  const u8 *hash_pos = token.buf[4];
+
+  digest[0] = hex_to_u32 (hash_pos +  0);
+  digest[1] = hex_to_u32 (hash_pos +  8);
+  digest[2] = hex_to_u32 (hash_pos + 16);
+  digest[3] = hex_to_u32 (hash_pos + 24);
+  digest[4] = hex_to_u32 (hash_pos + 32);
+  digest[5] = hex_to_u32 (hash_pos + 40);
+
+  digest[0] = byte_swap_32 (digest[0]);
+  digest[1] = byte_swap_32 (digest[1]);
+  digest[2] = byte_swap_32 (digest[2]);
+  digest[3] = byte_swap_32 (digest[3]);
+  digest[4] = byte_swap_32 (digest[4]);
+  digest[5] = byte_swap_32 (digest[5]);
+
+  return (PARSER_OK);
+}
+
+int module_hash_encode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const void *digest_buf, MAYBE_UNUSED const salt_t *salt, MAYBE_UNUSED const void *esalt_buf, MAYBE_UNUSED const void *hook_salt_buf, MAYBE_UNUSED const hashinfo_t *hash_info, char *line_buf, MAYBE_UNUSED const int line_size)
+{
+  const u32 *digest = (const u32 *) digest_buf;
+
+  snmpv3_t *snmpv3 = (snmpv3_t *) esalt_buf;
+
+  u8 *out_buf = (u8 *) line_buf;
+
+  int out_len = snprintf (line_buf, line_size, "%s%s$", SIGNATURE_SNMPV3, (char *) snmpv3->packet_number);
+
+  out_len += hex_encode ((u8 *) snmpv3->salt_buf, snmpv3->salt_len, out_buf + out_len);
+
+  out_buf[out_len] = '$';
+
+  out_len++;
+
+  out_len += hex_encode ((u8 *) snmpv3->engineID_buf, snmpv3->engineID_len, out_buf + out_len);
+
+  out_buf[out_len] = '$';
+
+  out_len++;
+
+  u32 digest_tmp[6];
+
+  digest_tmp[0] = byte_swap_32 (digest[0]);
+  digest_tmp[1] = byte_swap_32 (digest[1]);
+  digest_tmp[2] = byte_swap_32 (digest[2]);
+  digest_tmp[3] = byte_swap_32 (digest[3]);
+  digest_tmp[4] = byte_swap_32 (digest[4]);
+  digest_tmp[5] = byte_swap_32 (digest[5]);
+
+  u32_to_hex (digest_tmp[0], out_buf + out_len); out_len += 8;
+  u32_to_hex (digest_tmp[1], out_buf + out_len); out_len += 8;
+  u32_to_hex (digest_tmp[2], out_buf + out_len); out_len += 8;
+  u32_to_hex (digest_tmp[3], out_buf + out_len); out_len += 8;
+  u32_to_hex (digest_tmp[4], out_buf + out_len); out_len += 8;
+  u32_to_hex (digest_tmp[5], out_buf + out_len); out_len += 8;
+
+  out_buf[out_len] = 0;
+
+  return out_len;
+}
+
+void module_init (module_ctx_t *module_ctx)
+{
+  module_ctx->module_context_size             = MODULE_CONTEXT_SIZE_CURRENT;
+  module_ctx->module_interface_version        = MODULE_INTERFACE_VERSION_CURRENT;
+
+  module_ctx->module_attack_exec              = module_attack_exec;
+  module_ctx->module_benchmark_esalt          = MODULE_DEFAULT;
+  module_ctx->module_benchmark_hook_salt      = MODULE_DEFAULT;
+  module_ctx->module_benchmark_mask           = MODULE_DEFAULT;
+  module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
+  module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
+  module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
+  module_ctx->module_dgst_pos0                = module_dgst_pos0;
+  module_ctx->module_dgst_pos1                = module_dgst_pos1;
+  module_ctx->module_dgst_pos2                = module_dgst_pos2;
+  module_ctx->module_dgst_pos3                = module_dgst_pos3;
+  module_ctx->module_dgst_size                = module_dgst_size;
+  module_ctx->module_dictstat_disable         = MODULE_DEFAULT;
+  module_ctx->module_esalt_size               = module_esalt_size;
+  module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
+  module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
+  module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_save         = MODULE_DEFAULT;
+  module_ctx->module_hash_decode_potfile      = MODULE_DEFAULT;
+  module_ctx->module_hash_decode_zero_hash    = MODULE_DEFAULT;
+  module_ctx->module_hash_decode              = module_hash_decode;
+  module_ctx->module_hash_encode_status       = MODULE_DEFAULT;
+  module_ctx->module_hash_encode_potfile      = MODULE_DEFAULT;
+  module_ctx->module_hash_encode              = module_hash_encode;
+  module_ctx->module_hash_init_selftest       = MODULE_DEFAULT;
+  module_ctx->module_hash_mode                = MODULE_DEFAULT;
+  module_ctx->module_hash_category            = module_hash_category;
+  module_ctx->module_hash_name                = module_hash_name;
+  module_ctx->module_hashes_count_min         = MODULE_DEFAULT;
+  module_ctx->module_hashes_count_max         = MODULE_DEFAULT;
+  module_ctx->module_hlfmt_disable            = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_size    = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_init    = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_term    = MODULE_DEFAULT;
+  module_ctx->module_hook12                   = MODULE_DEFAULT;
+  module_ctx->module_hook23                   = MODULE_DEFAULT;
+  module_ctx->module_hook_salt_size           = MODULE_DEFAULT;
+  module_ctx->module_hook_size                = MODULE_DEFAULT;
+  module_ctx->module_jit_build_options        = MODULE_DEFAULT;
+  module_ctx->module_jit_cache_disable        = MODULE_DEFAULT;
+  module_ctx->module_kernel_accel_max         = MODULE_DEFAULT;
+  module_ctx->module_kernel_accel_min         = MODULE_DEFAULT;
+  module_ctx->module_kernel_loops_max         = module_kernel_loops_max;
+  module_ctx->module_kernel_loops_min         = module_kernel_loops_min;
+  module_ctx->module_kernel_threads_max       = MODULE_DEFAULT;
+  module_ctx->module_kernel_threads_min       = MODULE_DEFAULT;
+  module_ctx->module_kern_type                = module_kern_type;
+  module_ctx->module_kern_type_dynamic        = MODULE_DEFAULT;
+  module_ctx->module_opti_type                = module_opti_type;
+  module_ctx->module_opts_type                = module_opts_type;
+  module_ctx->module_outfile_check_disable    = MODULE_DEFAULT;
+  module_ctx->module_outfile_check_nocomp     = MODULE_DEFAULT;
+  module_ctx->module_potfile_custom_check     = MODULE_DEFAULT;
+  module_ctx->module_potfile_disable          = MODULE_DEFAULT;
+  module_ctx->module_potfile_keep_all_hashes  = MODULE_DEFAULT;
+  module_ctx->module_pwdump_column            = MODULE_DEFAULT;
+  module_ctx->module_pw_max                   = MODULE_DEFAULT;
+  module_ctx->module_pw_min                   = module_pw_min;
+  module_ctx->module_salt_max                 = MODULE_DEFAULT;
+  module_ctx->module_salt_min                 = MODULE_DEFAULT;
+  module_ctx->module_salt_type                = module_salt_type;
+  module_ctx->module_separator                = MODULE_DEFAULT;
+  module_ctx->module_st_hash                  = module_st_hash;
+  module_ctx->module_st_pass                  = module_st_pass;
+  module_ctx->module_tmp_size                 = module_tmp_size;
+  module_ctx->module_unstable_warning         = MODULE_DEFAULT;
+  module_ctx->module_warmup_disable           = MODULE_DEFAULT;
+}
diff --git a/src/modules/module_26900.c b/src/modules/module_26900.c
new file mode 100644
index 000000000..24473ce1c
--- /dev/null
+++ b/src/modules/module_26900.c
@@ -0,0 +1,383 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+#include "common.h"
+#include "types.h"
+#include "modules.h"
+#include "bitops.h"
+#include "convert.h"
+#include "shared.h"
+#include "memory.h"
+#include "emu_inc_hash_sha1.h"
+
+static const u32   ATTACK_EXEC    = ATTACK_EXEC_OUTSIDE_KERNEL;
+static const u32   DGST_POS0      = 2;
+static const u32   DGST_POS1      = 3;
+static const u32   DGST_POS2      = 0;
+static const u32   DGST_POS3      = 1;
+static const u32   DGST_SIZE      = DGST_SIZE_8_4;
+static const u32   HASH_CATEGORY  = HASH_CATEGORY_NETWORK_PROTOCOL;
+static const char *HASH_NAME      = "SNMPv3 HMAC-SHA384-256";
+static const u64   KERN_TYPE      = 26900;
+static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
+                                  | OPTI_TYPE_USES_BITS_64;
+static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_BE;
+static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
+static const char *ST_PASS        = "hashcat1";
+static const char *ST_HASH        = "$SNMPv3$5$45889431$3081a70201033011020455c0c85c020300ffe30401010201030450304e041180001f88808106d566db57fd600000000002011002020118040e6d61747269785f5348412d333834042000000000000000000000000000000000000000000000000000000000000000000400303d041180001f88808106d566db57fd60000000000400a226020411b3c3590201000201003018301606082b06010201010200060a2b06010401bf0803020a$80001f88808106d566db57fd60$89424907553231aaa27055f4b3b0a97c626ed4cdc4b660d903765b607af792a5";
+
+u32         module_attack_exec    (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ATTACK_EXEC;     }
+u32         module_dgst_pos0      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS0;       }
+u32         module_dgst_pos1      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS1;       }
+u32         module_dgst_pos2      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS2;       }
+u32         module_dgst_pos3      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS3;       }
+u32         module_dgst_size      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_SIZE;       }
+u32         module_hash_category  (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return HASH_CATEGORY;   }
+const char *module_hash_name      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return HASH_NAME;       }
+u64         module_kern_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return KERN_TYPE;       }
+u32         module_opti_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return OPTI_TYPE;       }
+u64         module_opts_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return OPTS_TYPE;       }
+u32         module_salt_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return SALT_TYPE;       }
+const char *module_st_hash        (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ST_HASH;         }
+const char *module_st_pass        (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ST_PASS;         }
+
+static const char *SIGNATURE_SNMPV3 = "$SNMPv3$5$";
+
+#define SNMPV3_SALT_MAX             1500
+#define SNMPV3_ENGINEID_MAX         34
+#define SNMPV3_MSG_AUTH_PARAMS_LEN  32
+#define SNMPV3_ROUNDS               1048576
+#define SNMPV3_MAX_PW_LENGTH        128
+
+#define SNMPV3_TMP_ELEMS            8192 // 8192 = (256 (max pw length) * SNMPV3_MAX_PW_LENGTH) / sizeof (u32)
+#define SNMPV3_HASH_ELEMS           8
+
+#define SNMPV3_MAX_SALT_ELEMS       512 // 512 * 4 = 2048 > 1500, also has to be multiple of SNMPV3_MAX_PW_LENGTH
+#define SNMPV3_MAX_ENGINE_ELEMS     32  // 32 * 4 = 128 > 34, also has to be multiple of SNMPV3_MAX_PW_LENGTH
+#define SNMPV3_MAX_PNUM_ELEMS       4   // 4 * 4 = 16 > 9
+
+typedef struct hmac_sha384_tmp
+{
+  u32 tmp[SNMPV3_TMP_ELEMS];
+  u64 h[SNMPV3_HASH_ELEMS];
+
+} hmac_sha384_tmp_t;
+
+typedef struct snmpv3
+{
+  u32 salt_buf[SNMPV3_MAX_SALT_ELEMS];
+  u32 salt_len;
+
+  u32 engineID_buf[SNMPV3_MAX_ENGINE_ELEMS];
+  u32 engineID_len;
+
+  u32 packet_number[SNMPV3_MAX_PNUM_ELEMS];
+
+} snmpv3_t;
+
+u32 module_pw_min (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u32 pw_min = 8;
+
+  return pw_min;
+}
+
+bool module_unstable_warning (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra, MAYBE_UNUSED const hc_device_param_t *device_param)
+{
+  if (device_param->opencl_platform_vendor_id == VENDOR_ID_APPLE)
+  {
+    // self-test failed
+    if ((device_param->opencl_device_vendor_id == VENDOR_ID_AMD) && (device_param->opencl_device_type & CL_DEVICE_TYPE_GPU))
+    {
+      return true;
+    }
+  }
+
+  return false;
+}
+
+char *module_jit_build_options (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra, MAYBE_UNUSED const hashes_t *hashes, MAYBE_UNUSED const hc_device_param_t *device_param)
+{
+  char *jit_build_options = NULL;
+
+  // Extra treatment for Apple systems
+  if (device_param->opencl_platform_vendor_id == VENDOR_ID_APPLE)
+  {
+    return jit_build_options;
+  }
+
+  // HIP
+  if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD_USE_HIP)
+  {
+    hc_asprintf (&jit_build_options, "-D _unroll");
+  }
+
+  // ROCM
+  if ((device_param->opencl_device_vendor_id == VENDOR_ID_AMD) && (device_param->has_vperm == true))
+  {
+    hc_asprintf (&jit_build_options, "-D _unroll");
+  }
+
+  return jit_build_options;
+}
+
+u64 module_esalt_size (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u64 esalt_size = (const u64) sizeof (snmpv3_t);
+
+  return esalt_size;
+}
+
+u64 module_tmp_size (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u64 tmp_size = (const u64) sizeof (hmac_sha384_tmp_t);
+
+  return tmp_size;
+}
+
+u32 module_kernel_loops_min (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  // we need to fix iteration count to guarantee the loop count is a multiple of SNMPV3_MAX_PW_LENGTH
+  // 2k calls to sha384_transform typically is enough to overtime pcie bottleneck
+
+  const u32 kernel_loops_min = 2048 * SNMPV3_MAX_PW_LENGTH;
+
+  return kernel_loops_min;
+}
+
+u32 module_kernel_loops_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u32 kernel_loops_max = 2048 * SNMPV3_MAX_PW_LENGTH;
+
+  return kernel_loops_max;
+}
+
+int module_hash_decode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED void *digest_buf, MAYBE_UNUSED salt_t *salt, MAYBE_UNUSED void *esalt_buf, MAYBE_UNUSED void *hook_salt_buf, MAYBE_UNUSED hashinfo_t *hash_info, const char *line_buf, MAYBE_UNUSED const int line_len)
+{
+  u64 *digest = (u64 *) digest_buf;
+
+  snmpv3_t *snmpv3 = (snmpv3_t *) esalt_buf;
+
+  token_t token;
+
+  token.token_cnt  = 5;
+  token.signatures_cnt    = 1;
+  token.signatures_buf[0] = SIGNATURE_SNMPV3;
+
+  token.len[0]     = 10;
+  token.attr[0]    = TOKEN_ATTR_FIXED_LENGTH
+                   | TOKEN_ATTR_VERIFY_SIGNATURE;
+
+  // packet number
+  token.len_min[1] = 1;
+  token.len_max[1] = 8;
+  token.sep[1]     = '$';
+  token.attr[1]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_DIGIT;
+  // salt
+  token.len_min[2] = SNMPV3_MSG_AUTH_PARAMS_LEN * 2;
+  token.len_max[2] = SNMPV3_SALT_MAX * 2;
+  token.sep[2]     = '$';
+  token.attr[2]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  // engineid
+  token.len_min[3] = 26;
+  token.len_max[3] = SNMPV3_ENGINEID_MAX;
+  token.sep[3]     = '$';
+  token.attr[3]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  // digest
+  token.len[4]     = SNMPV3_MSG_AUTH_PARAMS_LEN * 2;
+  token.sep[4]     = '$';
+  token.attr[4]    = TOKEN_ATTR_FIXED_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  const int rc_tokenizer = input_tokenizer ((const u8 *) line_buf, line_len, &token);
+
+  if (rc_tokenizer != PARSER_OK) return (rc_tokenizer);
+
+  // packet number
+
+  const u8 *packet_number_pos = token.buf[1];
+  const int packet_number_len = token.len[1];
+
+  memset (snmpv3->packet_number, 0, sizeof (snmpv3->packet_number));
+
+  strncpy ((char *) snmpv3->packet_number, (char *) packet_number_pos, packet_number_len);
+
+  // salt
+
+  const u8 *salt_pos = token.buf[2];
+  const int salt_len = token.len[2];
+
+  u8 *salt_ptr = (u8 *) snmpv3->salt_buf;
+
+  snmpv3->salt_len = hex_decode (salt_pos, salt_len, salt_ptr);
+
+  salt->salt_iter = SNMPV3_ROUNDS;
+
+  // handle unique salts detection
+
+  sha1_ctx_t sha1_ctx;
+
+  sha1_init   (&sha1_ctx);
+  sha1_update (&sha1_ctx, snmpv3->salt_buf, snmpv3->salt_len);
+  sha1_final  (&sha1_ctx);
+
+  // store sha1(snmpv3->salt_buf) in salt_buf
+
+  salt->salt_len = 20;
+
+  memcpy (salt->salt_buf, sha1_ctx.h, salt->salt_len);
+
+  // engineid
+
+  const u8 *engineID_pos = token.buf[3];
+  const int engineID_len = token.len[3];
+
+  memset (snmpv3->engineID_buf, 0, sizeof (snmpv3->engineID_buf));
+
+  u8 *engineID_ptr = (u8 *) snmpv3->engineID_buf;
+
+  hex_decode (engineID_pos, engineID_len, engineID_ptr);
+
+  // force len to 17, zero padding
+  snmpv3->engineID_len = SNMPV3_ENGINEID_MAX / 2;
+
+  // digest
+
+  const u8 *hash_pos = token.buf[4];
+
+  digest[0] = hex_to_u64 (hash_pos +  0);
+  digest[1] = hex_to_u64 (hash_pos + 16);
+  digest[2] = hex_to_u64 (hash_pos + 32);
+  digest[3] = hex_to_u64 (hash_pos + 48);
+
+  digest[0] = byte_swap_64 (digest[0]);
+  digest[1] = byte_swap_64 (digest[1]);
+  digest[2] = byte_swap_64 (digest[2]);
+  digest[3] = byte_swap_64 (digest[3]);
+
+  return (PARSER_OK);
+}
+
+int module_hash_encode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const void *digest_buf, MAYBE_UNUSED const salt_t *salt, MAYBE_UNUSED const void *esalt_buf, MAYBE_UNUSED const void *hook_salt_buf, MAYBE_UNUSED const hashinfo_t *hash_info, char *line_buf, MAYBE_UNUSED const int line_size)
+{
+  const u64 *digest = (const u64 *) digest_buf;
+
+  snmpv3_t *snmpv3 = (snmpv3_t *) esalt_buf;
+
+  u8 *out_buf = (u8 *) line_buf;
+
+  int out_len = snprintf (line_buf, line_size, "%s%s$", SIGNATURE_SNMPV3, (char *) snmpv3->packet_number);
+
+  out_len += hex_encode ((u8 *) snmpv3->salt_buf, snmpv3->salt_len, out_buf + out_len);
+
+  out_buf[out_len] = '$';
+
+  out_len++;
+
+  out_len += hex_encode ((u8 *) snmpv3->engineID_buf, snmpv3->engineID_len, out_buf + out_len);
+
+  out_buf[out_len] = '$';
+
+  out_len++;
+
+  u64 digest_tmp[4];
+
+  digest_tmp[0] = byte_swap_64 (digest[0]);
+  digest_tmp[1] = byte_swap_64 (digest[1]);
+  digest_tmp[2] = byte_swap_64 (digest[2]);
+  digest_tmp[3] = byte_swap_64 (digest[3]);
+
+  u64_to_hex (digest_tmp[0], out_buf + out_len); out_len += 16;
+  u64_to_hex (digest_tmp[1], out_buf + out_len); out_len += 16;
+  u64_to_hex (digest_tmp[2], out_buf + out_len); out_len += 16;
+  u64_to_hex (digest_tmp[3], out_buf + out_len); out_len += 16;
+
+  out_buf[out_len] = 0;
+
+  return out_len;
+}
+
+void module_init (module_ctx_t *module_ctx)
+{
+  module_ctx->module_context_size             = MODULE_CONTEXT_SIZE_CURRENT;
+  module_ctx->module_interface_version        = MODULE_INTERFACE_VERSION_CURRENT;
+
+  module_ctx->module_attack_exec              = module_attack_exec;
+  module_ctx->module_benchmark_esalt          = MODULE_DEFAULT;
+  module_ctx->module_benchmark_hook_salt      = MODULE_DEFAULT;
+  module_ctx->module_benchmark_mask           = MODULE_DEFAULT;
+  module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
+  module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
+  module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
+  module_ctx->module_dgst_pos0                = module_dgst_pos0;
+  module_ctx->module_dgst_pos1                = module_dgst_pos1;
+  module_ctx->module_dgst_pos2                = module_dgst_pos2;
+  module_ctx->module_dgst_pos3                = module_dgst_pos3;
+  module_ctx->module_dgst_size                = module_dgst_size;
+  module_ctx->module_dictstat_disable         = MODULE_DEFAULT;
+  module_ctx->module_esalt_size               = module_esalt_size;
+  module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
+  module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
+  module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_save         = MODULE_DEFAULT;
+  module_ctx->module_hash_decode_potfile      = MODULE_DEFAULT;
+  module_ctx->module_hash_decode_zero_hash    = MODULE_DEFAULT;
+  module_ctx->module_hash_decode              = module_hash_decode;
+  module_ctx->module_hash_encode_status       = MODULE_DEFAULT;
+  module_ctx->module_hash_encode_potfile      = MODULE_DEFAULT;
+  module_ctx->module_hash_encode              = module_hash_encode;
+  module_ctx->module_hash_init_selftest       = MODULE_DEFAULT;
+  module_ctx->module_hash_mode                = MODULE_DEFAULT;
+  module_ctx->module_hash_category            = module_hash_category;
+  module_ctx->module_hash_name                = module_hash_name;
+  module_ctx->module_hashes_count_min         = MODULE_DEFAULT;
+  module_ctx->module_hashes_count_max         = MODULE_DEFAULT;
+  module_ctx->module_hlfmt_disable            = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_size    = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_init    = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_term    = MODULE_DEFAULT;
+  module_ctx->module_hook12                   = MODULE_DEFAULT;
+  module_ctx->module_hook23                   = MODULE_DEFAULT;
+  module_ctx->module_hook_salt_size           = MODULE_DEFAULT;
+  module_ctx->module_hook_size                = MODULE_DEFAULT;
+  module_ctx->module_jit_build_options        = module_jit_build_options;
+  module_ctx->module_jit_cache_disable        = MODULE_DEFAULT;
+  module_ctx->module_kernel_accel_max         = MODULE_DEFAULT;
+  module_ctx->module_kernel_accel_min         = MODULE_DEFAULT;
+  module_ctx->module_kernel_loops_max         = module_kernel_loops_max;
+  module_ctx->module_kernel_loops_min         = module_kernel_loops_min;
+  module_ctx->module_kernel_threads_max       = MODULE_DEFAULT;
+  module_ctx->module_kernel_threads_min       = MODULE_DEFAULT;
+  module_ctx->module_kern_type                = module_kern_type;
+  module_ctx->module_kern_type_dynamic        = MODULE_DEFAULT;
+  module_ctx->module_opti_type                = module_opti_type;
+  module_ctx->module_opts_type                = module_opts_type;
+  module_ctx->module_outfile_check_disable    = MODULE_DEFAULT;
+  module_ctx->module_outfile_check_nocomp     = MODULE_DEFAULT;
+  module_ctx->module_potfile_custom_check     = MODULE_DEFAULT;
+  module_ctx->module_potfile_disable          = MODULE_DEFAULT;
+  module_ctx->module_potfile_keep_all_hashes  = MODULE_DEFAULT;
+  module_ctx->module_pwdump_column            = MODULE_DEFAULT;
+  module_ctx->module_pw_max                   = MODULE_DEFAULT;
+  module_ctx->module_pw_min                   = module_pw_min;
+  module_ctx->module_salt_max                 = MODULE_DEFAULT;
+  module_ctx->module_salt_min                 = MODULE_DEFAULT;
+  module_ctx->module_salt_type                = module_salt_type;
+  module_ctx->module_separator                = MODULE_DEFAULT;
+  module_ctx->module_st_hash                  = module_st_hash;
+  module_ctx->module_st_pass                  = module_st_pass;
+  module_ctx->module_tmp_size                 = module_tmp_size;
+  module_ctx->module_unstable_warning         = module_unstable_warning;
+  module_ctx->module_warmup_disable           = MODULE_DEFAULT;
+}
diff --git a/src/modules/module_27000.c b/src/modules/module_27000.c
new file mode 100644
index 000000000..cdbf71fe2
--- /dev/null
+++ b/src/modules/module_27000.c
@@ -0,0 +1,513 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+#include "common.h"
+#include "types.h"
+#include "modules.h"
+#include "bitops.h"
+#include "convert.h"
+#include "shared.h"
+#include "emu_inc_cipher_des.h"
+#include "emu_inc_hash_md5.h"
+
+static const u32   ATTACK_EXEC    = ATTACK_EXEC_OUTSIDE_KERNEL;
+static const u32   DGST_POS0      = 0;
+static const u32   DGST_POS1      = 1;
+static const u32   DGST_POS2      = 2;
+static const u32   DGST_POS3      = 3;
+static const u32   DGST_SIZE      = DGST_SIZE_4_4;
+static const u32   HASH_CATEGORY  = HASH_CATEGORY_NETWORK_PROTOCOL;
+static const char *HASH_NAME      = "NetNTLMv1 / NetNTLMv1+ESS (NT)";
+static const u64   KERN_TYPE      = 27000;
+static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE;
+static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
+                                  | OPTS_TYPE_PT_ADD80
+                                  | OPTS_TYPE_PT_ADDBITS14
+                                  | OPTS_TYPE_PT_UTF16LE
+                                  | OPTS_TYPE_ST_HEX;
+static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
+static const char *ST_PASS        = "b4b9b02e6f09a9bd760f388b67351e2b";
+static const char *ST_HASH        = "::5V4T:ada06359242920a500000000000000000000000000000000:0556d5297b5daa70eaffde82ef99293a3f3bb59b7c9704ea:9c23f6c094853920";
+
+typedef struct netntlm
+{
+  int user_len;
+  int domain_len;
+  int srvchall_len;
+  int clichall_len;
+
+  u32 userdomain_buf[64];
+  u32 chall_buf[256];
+
+} netntlm_t;
+
+typedef struct netntlm_tmp
+{
+  u32 digest_buf[4];
+
+} netntlm_tmp_t;
+
+u32         module_attack_exec    (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ATTACK_EXEC;     }
+u32         module_dgst_pos0      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS0;       }
+u32         module_dgst_pos1      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS1;       }
+u32         module_dgst_pos2      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS2;       }
+u32         module_dgst_pos3      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS3;       }
+u32         module_dgst_size      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_SIZE;       }
+u32         module_hash_category  (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return HASH_CATEGORY;   }
+const char *module_hash_name      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return HASH_NAME;       }
+u64         module_kern_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return KERN_TYPE;       }
+u32         module_opti_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return OPTI_TYPE;       }
+u64         module_opts_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return OPTS_TYPE;       }
+u32         module_salt_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return SALT_TYPE;       }
+const char *module_st_hash        (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ST_HASH;         }
+const char *module_st_pass        (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ST_PASS;         }
+
+u64 module_tmp_size (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u64 tmp_size = (const u64) sizeof (netntlm_tmp_t);
+
+  return tmp_size;
+}
+
+static void transform_netntlmv1_key (const u8 *nthash, u8 *key)
+{
+  key[0] =                    (nthash[0] >> 0);
+  key[1] = (nthash[0] << 7) | (nthash[1] >> 1);
+  key[2] = (nthash[1] << 6) | (nthash[2] >> 2);
+  key[3] = (nthash[2] << 5) | (nthash[3] >> 3);
+  key[4] = (nthash[3] << 4) | (nthash[4] >> 4);
+  key[5] = (nthash[4] << 3) | (nthash[5] >> 5);
+  key[6] = (nthash[5] << 2) | (nthash[6] >> 6);
+  key[7] = (nthash[6] << 1);
+
+  key[0] |= 0x01;
+  key[1] |= 0x01;
+  key[2] |= 0x01;
+  key[3] |= 0x01;
+  key[4] |= 0x01;
+  key[5] |= 0x01;
+  key[6] |= 0x01;
+  key[7] |= 0x01;
+}
+
+char *module_jit_build_options (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra, MAYBE_UNUSED const hashes_t *hashes, MAYBE_UNUSED const hc_device_param_t *device_param)
+{
+  char *jit_build_options = NULL;
+
+  // Extra treatment for Apple systems
+  if (device_param->opencl_platform_vendor_id == VENDOR_ID_APPLE)
+  {
+    return jit_build_options;
+  }
+
+  // Intel CPU
+  if ((device_param->opencl_device_vendor_id == VENDOR_ID_INTEL_SDK) && (device_param->opencl_device_type & CL_DEVICE_TYPE_CPU))
+  {
+    hc_asprintf (&jit_build_options, "-D _unroll");
+  }
+
+  return jit_build_options;
+}
+
+u64 module_esalt_size (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u64 esalt_size = (const u64) sizeof (netntlm_t);
+
+  return esalt_size;
+}
+
+int module_hash_decode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED void *digest_buf, MAYBE_UNUSED salt_t *salt, MAYBE_UNUSED void *esalt_buf, MAYBE_UNUSED void *hook_salt_buf, MAYBE_UNUSED hashinfo_t *hash_info, const char *line_buf, MAYBE_UNUSED const int line_len)
+{
+  u32 *digest = (u32 *) digest_buf;
+
+  netntlm_t *netntlm = (netntlm_t *) esalt_buf;
+
+  token_t token;
+
+  token.token_cnt  = 6;
+
+  // username
+  token.len_min[0] = 0;
+  token.len_max[0] = 60;
+  token.sep[0]     = ':';
+  token.attr[0]    = TOKEN_ATTR_VERIFY_LENGTH;
+
+  // unused
+  token.len_min[1] = 0;
+  token.len_max[1] = 0;
+  token.sep[1]     = ':';
+  token.attr[1]    = TOKEN_ATTR_VERIFY_LENGTH;
+
+  // domain
+  token.len_min[2] = 0;
+  token.len_max[2] = 45;
+  token.sep[2]     = ':';
+  token.attr[2]    = TOKEN_ATTR_VERIFY_LENGTH;
+
+  // lm response
+  token.len_min[3] = 0;
+  token.len_max[3] = 48;
+  token.sep[3]     = ':';
+  token.attr[3]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  // ntlm response
+  token.len_min[4] = 48;
+  token.len_max[4] = 48;
+  token.sep[4]     = ':';
+  token.attr[4]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  // challenge
+  token.len_min[5] = 16;
+  token.len_max[5] = 16;
+  token.sep[5]     = ':';
+  token.attr[5]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  const int rc_tokenizer = input_tokenizer ((const u8 *) line_buf, line_len, &token);
+
+  if (rc_tokenizer != PARSER_OK) return (rc_tokenizer);
+
+  const u8 *user_pos     = token.buf[0];
+  const u8 *domain_pos   = token.buf[2];
+  const u8 *srvchall_pos = token.buf[3];
+  const u8 *hash_pos     = token.buf[4];
+  const u8 *clichall_pos = token.buf[5];
+
+  const int user_len     = token.len[0];
+  const int domain_len   = token.len[2];
+  const int srvchall_len = token.len[3];
+  const int clichall_len = token.len[5];
+
+  /**
+   * store some data for later use
+   */
+
+  netntlm->user_len     = user_len     * 2;
+  netntlm->domain_len   = domain_len   * 2;
+  netntlm->srvchall_len = srvchall_len / 2;
+  netntlm->clichall_len = clichall_len / 2;
+
+  u8 *userdomain_ptr = (u8 *) netntlm->userdomain_buf;
+  u8 *chall_ptr      = (u8 *) netntlm->chall_buf;
+
+  /**
+   * handle username and domainname
+   */
+
+  for (int i = 0; i < user_len; i++)
+  {
+    *userdomain_ptr++ = user_pos[i];
+    *userdomain_ptr++ = 0;
+  }
+
+  for (int i = 0; i < domain_len; i++)
+  {
+    *userdomain_ptr++ = domain_pos[i];
+    *userdomain_ptr++ = 0;
+  }
+
+  /**
+   * handle server challenge encoding
+   */
+
+  for (int i = 0; i < srvchall_len; i += 2)
+  {
+    const u8 p0 = srvchall_pos[i + 0];
+    const u8 p1 = srvchall_pos[i + 1];
+
+    *chall_ptr++ = hex_convert (p1) << 0
+                 | hex_convert (p0) << 4;
+  }
+
+  /**
+   * handle client challenge encoding
+   */
+
+  for (int i = 0; i < clichall_len; i += 2)
+  {
+    const u8 p0 = clichall_pos[i + 0];
+    const u8 p1 = clichall_pos[i + 1];
+
+    *chall_ptr++ = hex_convert (p1) << 0
+                 | hex_convert (p0) << 4;
+  }
+
+  /**
+   * store data
+   */
+
+  const bool parse_rc = generic_salt_decode (hashconfig, clichall_pos, clichall_len, (u8 *) salt->salt_buf, (int *) &salt->salt_len);
+
+  if (parse_rc == false) return (PARSER_SALT_LENGTH);
+
+  digest[0] = hex_to_u32 (hash_pos +  0);
+  digest[1] = hex_to_u32 (hash_pos +  8);
+  digest[2] = hex_to_u32 (hash_pos + 16);
+  digest[3] = hex_to_u32 (hash_pos + 24);
+
+  /* special case, last 8 byte do not need to be checked since they are brute-forced next */
+
+  u32 digest_tmp[2];
+
+  digest_tmp[0] = hex_to_u32 (hash_pos + 32);
+  digest_tmp[1] = hex_to_u32 (hash_pos + 40);
+
+  /* special case 2: ESS */
+
+  if (srvchall_len == 48)
+  {
+    if ((netntlm->chall_buf[2] == 0) && (netntlm->chall_buf[3] == 0) && (netntlm->chall_buf[4] == 0) && (netntlm->chall_buf[5] == 0))
+    {
+      u32 w[16] = { 0 };
+
+      w[ 0] = salt->salt_buf[0];
+      w[ 1] = salt->salt_buf[1];
+      w[ 2] = netntlm->chall_buf[0];
+      w[ 3] = netntlm->chall_buf[1];
+      w[ 4] = 0x80;
+      w[14] = 16 * 8;
+
+      u32 dgst[4] = { 0 };
+
+      dgst[0] = MD5M_A;
+      dgst[1] = MD5M_B;
+      dgst[2] = MD5M_C;
+      dgst[3] = MD5M_D;
+
+      md5_transform (w + 0, w + 4, w + 8, w + 12, dgst);
+
+      salt->salt_buf[0] = dgst[0];
+      salt->salt_buf[1] = dgst[1];
+    }
+  }
+
+  /* precompute netntlmv1 exploit start */
+
+  for (u32 i = 0; i < 0x10000; i++)
+  {
+    u32 key_md4[2] = { 0 };
+    u32 key_des[2] = { 0 };
+
+    key_md4[0] = i;
+
+    transform_netntlmv1_key ((const u8 *) key_md4, (u8 *) key_des);
+
+    u32 Kc[16] = { 0 };
+    u32 Kd[16] = { 0 };
+
+    _des_crypt_keysetup (key_des[0], key_des[1], Kc, Kd, (u32 (*)[64]) c_skb);
+
+    u32 data3[2] = { salt->salt_buf[0], salt->salt_buf[1] };
+
+    _des_crypt_encrypt (data3, data3, Kc, Kd, (u32 (*)[64]) c_SPtrans);
+
+    if (data3[0] != digest_tmp[0]) continue;
+    if (data3[1] != digest_tmp[1]) continue;
+
+    salt->salt_buf[2] = i;
+
+    salt->salt_len = 24;
+
+    break;
+  }
+
+  salt->salt_buf_pc[0] = digest_tmp[0];
+  salt->salt_buf_pc[1] = digest_tmp[1];
+
+  /* precompute netntlmv1 exploit stop */
+
+  DES_IP (digest[0], digest[1]);
+  DES_IP (digest[2], digest[3]);
+
+  digest[0] = rotr32 (digest[0], 29);
+  digest[1] = rotr32 (digest[1], 29);
+  digest[2] = rotr32 (digest[2], 29);
+  digest[3] = rotr32 (digest[3], 29);
+
+  DES_IP (salt->salt_buf[0], salt->salt_buf[1]);
+
+  salt->salt_buf[0] = rotl32 (salt->salt_buf[0], 3);
+  salt->salt_buf[1] = rotl32 (salt->salt_buf[1], 3);
+
+  return (PARSER_OK);
+}
+
+int module_hash_encode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const void *digest_buf, MAYBE_UNUSED const salt_t *salt, MAYBE_UNUSED const void *esalt_buf, MAYBE_UNUSED const void *hook_salt_buf, MAYBE_UNUSED const hashinfo_t *hash_info, char *line_buf, MAYBE_UNUSED const int line_size)
+{
+  const u32 *digest = (const u32 *) digest_buf;
+
+  const netntlm_t *netntlm = (const netntlm_t *) esalt_buf;
+
+  // we can not change anything in the original buffer, otherwise destroying sorting
+  // therefore create some local buffer
+
+  u32 tmp[4];
+
+  tmp[0] = digest[0];
+  tmp[1] = digest[1];
+  tmp[2] = digest[2];
+  tmp[3] = digest[3];
+
+  tmp[0] = rotl32 (tmp[0], 29);
+  tmp[1] = rotl32 (tmp[1], 29);
+  tmp[2] = rotl32 (tmp[2], 29);
+  tmp[3] = rotl32 (tmp[3], 29);
+
+  DES_FP (tmp[1], tmp[0]);
+  DES_FP (tmp[3], tmp[2]);
+
+  u8 *out_buf = (u8 *) line_buf;
+
+  int out_len = 0;
+
+  u8 *ptr;
+
+  ptr = (u8 *) netntlm->userdomain_buf;
+
+  for (int i = 0; i < netntlm->user_len; i += 2)
+  {
+    out_buf[out_len++] = ptr[i];
+  }
+
+  out_buf[out_len++] = ':';
+  out_buf[out_len++] = ':';
+
+  ptr += netntlm->user_len;
+
+  for (int i = 0; i < netntlm->domain_len; i += 2)
+  {
+    out_buf[out_len++] = ptr[i];
+  }
+
+  out_buf[out_len++] = ':';
+
+  ptr = (u8 *) netntlm->chall_buf;
+
+  for (int i = 0; i < netntlm->srvchall_len; i++)
+  {
+    u8_to_hex (ptr[i], out_buf + out_len); out_len += 2;
+  }
+
+  out_buf[out_len++] = ':';
+
+  u32_to_hex (tmp[0], out_buf + out_len); out_len += 8;
+  u32_to_hex (tmp[1], out_buf + out_len); out_len += 8;
+  u32_to_hex (tmp[2], out_buf + out_len); out_len += 8;
+  u32_to_hex (tmp[3], out_buf + out_len); out_len += 8;
+
+  u32_to_hex (salt->salt_buf_pc[0], out_buf + out_len); out_len += 8;
+  u32_to_hex (salt->salt_buf_pc[1], out_buf + out_len); out_len += 8;
+
+  out_buf[out_len++] = ':';
+
+  ptr += netntlm->srvchall_len;
+
+  for (int i = 0; i < netntlm->clichall_len; i++)
+  {
+    u8_to_hex (ptr[i], out_buf + out_len); out_len += 2;
+  }
+
+  return out_len;
+}
+
+u32 module_pw_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u32 pw_max = 32; // Length of a NT hash
+
+  return pw_max;
+}
+
+u32 module_pw_min (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u32 pw_min = 32; // Length of a NT hash
+
+  return pw_min;
+}
+
+const char *module_benchmark_mask (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+ const char *mask = "?a?a?a?a?a?a?a?axxxxxxxxxxxxxxxx";
+ return mask;
+}
+
+void module_init (module_ctx_t *module_ctx)
+{
+  module_ctx->module_context_size             = MODULE_CONTEXT_SIZE_CURRENT;
+  module_ctx->module_interface_version        = MODULE_INTERFACE_VERSION_CURRENT;
+
+  module_ctx->module_attack_exec              = module_attack_exec;
+  module_ctx->module_benchmark_esalt          = MODULE_DEFAULT;
+  module_ctx->module_benchmark_hook_salt      = MODULE_DEFAULT;
+  module_ctx->module_benchmark_mask           = module_benchmark_mask;
+  module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
+  module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
+  module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
+  module_ctx->module_dgst_pos0                = module_dgst_pos0;
+  module_ctx->module_dgst_pos1                = module_dgst_pos1;
+  module_ctx->module_dgst_pos2                = module_dgst_pos2;
+  module_ctx->module_dgst_pos3                = module_dgst_pos3;
+  module_ctx->module_dgst_size                = module_dgst_size;
+  module_ctx->module_dictstat_disable         = MODULE_DEFAULT;
+  module_ctx->module_esalt_size               = module_esalt_size;
+  module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
+  module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
+  module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_save         = MODULE_DEFAULT;
+  module_ctx->module_hash_decode_potfile      = MODULE_DEFAULT;
+  module_ctx->module_hash_decode_zero_hash    = MODULE_DEFAULT;
+  module_ctx->module_hash_decode              = module_hash_decode;
+  module_ctx->module_hash_encode_status       = MODULE_DEFAULT;
+  module_ctx->module_hash_encode_potfile      = MODULE_DEFAULT;
+  module_ctx->module_hash_encode              = module_hash_encode;
+  module_ctx->module_hash_init_selftest       = MODULE_DEFAULT;
+  module_ctx->module_hash_mode                = MODULE_DEFAULT;
+  module_ctx->module_hash_category            = module_hash_category;
+  module_ctx->module_hash_name                = module_hash_name;
+  module_ctx->module_hashes_count_min         = MODULE_DEFAULT;
+  module_ctx->module_hashes_count_max         = MODULE_DEFAULT;
+  module_ctx->module_hlfmt_disable            = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_size    = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_init    = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_term    = MODULE_DEFAULT;
+  module_ctx->module_hook12                   = MODULE_DEFAULT;
+  module_ctx->module_hook23                   = MODULE_DEFAULT;
+  module_ctx->module_hook_salt_size           = MODULE_DEFAULT;
+  module_ctx->module_hook_size                = MODULE_DEFAULT;
+  module_ctx->module_jit_build_options        = module_jit_build_options;
+  module_ctx->module_jit_cache_disable        = MODULE_DEFAULT;
+  module_ctx->module_kernel_accel_max         = MODULE_DEFAULT;
+  module_ctx->module_kernel_accel_min         = MODULE_DEFAULT;
+  module_ctx->module_kernel_loops_max         = MODULE_DEFAULT;
+  module_ctx->module_kernel_loops_min         = MODULE_DEFAULT;
+  module_ctx->module_kernel_threads_max       = MODULE_DEFAULT;
+  module_ctx->module_kernel_threads_min       = MODULE_DEFAULT;
+  module_ctx->module_kern_type                = module_kern_type;
+  module_ctx->module_kern_type_dynamic        = MODULE_DEFAULT;
+  module_ctx->module_opti_type                = module_opti_type;
+  module_ctx->module_opts_type                = module_opts_type;
+  module_ctx->module_outfile_check_disable    = MODULE_DEFAULT;
+  module_ctx->module_outfile_check_nocomp     = MODULE_DEFAULT;
+  module_ctx->module_potfile_custom_check     = MODULE_DEFAULT;
+  module_ctx->module_potfile_disable          = MODULE_DEFAULT;
+  module_ctx->module_potfile_keep_all_hashes  = MODULE_DEFAULT;
+  module_ctx->module_pwdump_column            = MODULE_DEFAULT;
+  module_ctx->module_pw_max                   = module_pw_max;
+  module_ctx->module_pw_min                   = module_pw_min;
+  module_ctx->module_salt_max                 = MODULE_DEFAULT;
+  module_ctx->module_salt_min                 = MODULE_DEFAULT;
+  module_ctx->module_salt_type                = module_salt_type;
+  module_ctx->module_separator                = MODULE_DEFAULT;
+  module_ctx->module_st_hash                  = module_st_hash;
+  module_ctx->module_st_pass                  = module_st_pass;
+  module_ctx->module_tmp_size                 = module_tmp_size;
+  module_ctx->module_unstable_warning         = MODULE_DEFAULT;
+  module_ctx->module_warmup_disable           = MODULE_DEFAULT;
+}
diff --git a/src/modules/module_27100.c b/src/modules/module_27100.c
new file mode 100644
index 000000000..0b95b06ab
--- /dev/null
+++ b/src/modules/module_27100.c
@@ -0,0 +1,409 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+#include "common.h"
+#include "types.h"
+#include "modules.h"
+#include "bitops.h"
+#include "convert.h"
+#include "shared.h"
+#include "emu_inc_cipher_des.h"
+#include "emu_inc_hash_md5.h"
+
+static const u32   ATTACK_EXEC    = ATTACK_EXEC_OUTSIDE_KERNEL;
+static const u32   DGST_POS0      = 0;
+static const u32   DGST_POS1      = 3;
+static const u32   DGST_POS2      = 2;
+static const u32   DGST_POS3      = 1;
+static const u32   DGST_SIZE      = DGST_SIZE_4_4;
+static const u32   HASH_CATEGORY  = HASH_CATEGORY_NETWORK_PROTOCOL;
+static const char *HASH_NAME      = "NetNTLMv2 (NT)";
+static const u64   KERN_TYPE      = 27100;
+static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE;
+static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
+                                  | OPTS_TYPE_PT_ADD80
+                                  | OPTS_TYPE_PT_ADDBITS14
+                                  | OPTS_TYPE_PT_UTF16LE
+                                  | OPTS_TYPE_ST_HEX;
+static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
+static const char *ST_PASS        = "b4b9b02e6f09a9bd760f388b67351e2b";
+static const char *ST_HASH        = "0UL5G37JOI0SX::6VB1IS0KA74:ebe1afa18b7fbfa6:aab8bf8675658dd2a939458a1077ba08:010100000000000031c8aa092510945398b9f7b7dde1a9fb00000000f7876f2b04b700";
+
+typedef struct netntlm
+{
+  int user_len;
+  int domain_len;
+  int srvchall_len;
+  int clichall_len;
+
+  u32 userdomain_buf[64];
+  u32 chall_buf[256];
+
+} netntlm_t;
+
+typedef struct netntlmv2_tmp
+{
+  u32 digest_buf[4];
+
+} netntlm_tmp_t;
+
+u32         module_attack_exec    (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ATTACK_EXEC;     }
+u32         module_dgst_pos0      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS0;       }
+u32         module_dgst_pos1      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS1;       }
+u32         module_dgst_pos2      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS2;       }
+u32         module_dgst_pos3      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS3;       }
+u32         module_dgst_size      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_SIZE;       }
+u32         module_hash_category  (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return HASH_CATEGORY;   }
+const char *module_hash_name      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return HASH_NAME;       }
+u64         module_kern_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return KERN_TYPE;       }
+u32         module_opti_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return OPTI_TYPE;       }
+u64         module_opts_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return OPTS_TYPE;       }
+u32         module_salt_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return SALT_TYPE;       }
+const char *module_st_hash        (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ST_HASH;         }
+const char *module_st_pass        (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ST_PASS;         }
+
+u64 module_tmp_size (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u64 tmp_size = (const u64) sizeof (netntlm_tmp_t);
+
+  return tmp_size;
+}
+
+u64 module_esalt_size (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u64 esalt_size = (const u64) sizeof (netntlm_t);
+
+  return esalt_size;
+}
+
+int module_hash_decode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED void *digest_buf, MAYBE_UNUSED salt_t *salt, MAYBE_UNUSED void *esalt_buf, MAYBE_UNUSED void *hook_salt_buf, MAYBE_UNUSED hashinfo_t *hash_info, const char *line_buf, MAYBE_UNUSED const int line_len)
+{
+  u32 *digest = (u32 *) digest_buf;
+
+  netntlm_t *netntlm = (netntlm_t *) esalt_buf;
+
+  token_t token;
+
+  token.token_cnt  = 6;
+
+  // username
+  token.len_min[0] = 0;
+  token.len_max[0] = 60;
+  token.sep[0]     = ':';
+  token.attr[0]    = TOKEN_ATTR_VERIFY_LENGTH;
+
+  // unused
+  token.len_min[1] = 0;
+  token.len_max[1] = 0;
+  token.sep[1]     = ':';
+  token.attr[1]    = TOKEN_ATTR_VERIFY_LENGTH;
+
+  // domain
+  token.len_min[2] = 0;
+  token.len_max[2] = 45;
+  token.sep[2]     = ':';
+  token.attr[2]    = TOKEN_ATTR_VERIFY_LENGTH;
+
+  // lm response
+  token.len_min[3] = 16;
+  token.len_max[3] = 16;
+  token.sep[3]     = ':';
+  token.attr[3]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  // ntlm response
+  token.len_min[4] = 32;
+  token.len_max[4] = 32;
+  token.sep[4]     = ':';
+  token.attr[4]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  // challenge
+  token.len_min[5] = 2;
+  token.len_max[5] = 1024;
+  token.sep[5]     = ':';
+  token.attr[5]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  const int rc_tokenizer = input_tokenizer ((const u8 *) line_buf, line_len, &token);
+
+  if (rc_tokenizer != PARSER_OK) return (rc_tokenizer);
+
+  const u8 *user_pos     = token.buf[0];
+  const u8 *domain_pos   = token.buf[2];
+  const u8 *srvchall_pos = token.buf[3];
+  const u8 *hash_pos     = token.buf[4];
+  const u8 *clichall_pos = token.buf[5];
+
+  const int user_len     = token.len[0];
+  const int domain_len   = token.len[2];
+  const int srvchall_len = token.len[3];
+  const int clichall_len = token.len[5];
+
+  /**
+   * store some data for later use
+   */
+
+  netntlm->user_len     = user_len     * 2;
+  netntlm->domain_len   = domain_len   * 2;
+  netntlm->srvchall_len = srvchall_len / 2;
+  netntlm->clichall_len = clichall_len / 2;
+
+  u8 *userdomain_ptr = (u8 *) netntlm->userdomain_buf;
+  u8 *chall_ptr      = (u8 *) netntlm->chall_buf;
+
+  /**
+   * handle username and domainname
+   */
+
+  for (int i = 0; i < user_len; i++)
+  {
+    *userdomain_ptr++ = toupper (user_pos[i]);
+    *userdomain_ptr++ = 0;
+  }
+
+  for (int i = 0; i < domain_len; i++)
+  {
+    *userdomain_ptr++ = domain_pos[i];
+    *userdomain_ptr++ = 0;
+  }
+
+  *userdomain_ptr++ = 0x80;
+
+  /**
+   * handle server challenge encoding
+   */
+
+  for (int i = 0; i < srvchall_len; i += 2)
+  {
+    const u8 p0 = srvchall_pos[i + 0];
+    const u8 p1 = srvchall_pos[i + 1];
+
+    *chall_ptr++ = hex_convert (p1) << 0
+                 | hex_convert (p0) << 4;
+  }
+
+  /**
+   * handle client challenge encoding
+   */
+
+  for (int i = 0; i < clichall_len; i += 2)
+  {
+    const u8 p0 = clichall_pos[i + 0];
+    const u8 p1 = clichall_pos[i + 1];
+
+    *chall_ptr++ = hex_convert (p1) << 0
+                 | hex_convert (p0) << 4;
+  }
+
+  *chall_ptr++ = 0x80;
+
+  /**
+   * handle hash itself
+   */
+
+  digest[0] = hex_to_u32 (hash_pos +  0);
+  digest[1] = hex_to_u32 (hash_pos +  8);
+  digest[2] = hex_to_u32 (hash_pos + 16);
+  digest[3] = hex_to_u32 (hash_pos + 24);
+
+  /**
+   * reuse challange data as salt_buf, its the buffer that is most likely unique
+   */
+
+  salt->salt_buf[0] = 0;
+  salt->salt_buf[1] = 0;
+  salt->salt_buf[2] = 0;
+  salt->salt_buf[3] = 0;
+  salt->salt_buf[4] = 0;
+  salt->salt_buf[5] = 0;
+  salt->salt_buf[6] = 0;
+  salt->salt_buf[7] = 0;
+
+  u32 *uptr;
+
+  uptr = (u32 *) netntlm->userdomain_buf;
+
+  for (u32 i = 0; i < 64; i += 16, uptr += 16)
+  {
+    md5_transform (uptr + 0, uptr + 4, uptr + 8, uptr + 12, salt->salt_buf);
+  }
+
+  uptr = (u32 *) netntlm->chall_buf;
+
+  for (u32 i = 0; i < 256; i += 16, uptr += 16)
+  {
+    md5_transform (uptr + 0, uptr + 4, uptr + 8, uptr + 12, salt->salt_buf);
+  }
+
+  salt->salt_len = 16;
+
+  return (PARSER_OK);
+}
+
+int module_hash_encode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const void *digest_buf, MAYBE_UNUSED const salt_t *salt, MAYBE_UNUSED const void *esalt_buf, MAYBE_UNUSED const void *hook_salt_buf, MAYBE_UNUSED const hashinfo_t *hash_info, char *line_buf, MAYBE_UNUSED const int line_size)
+{
+  const u32 *digest = (const u32 *) digest_buf;
+
+  const netntlm_t *netntlm = (const netntlm_t *) esalt_buf;
+
+  // we can not change anything in the original buffer, otherwise destroying sorting
+  // therefore create some local buffer
+
+  u32 tmp[4];
+
+  tmp[0] = digest[0];
+  tmp[1] = digest[1];
+  tmp[2] = digest[2];
+  tmp[3] = digest[3];
+
+  u8 *out_buf = (u8 *) line_buf;
+
+  int out_len = 0;
+
+  u8 *ptr;
+
+  ptr = (u8 *) netntlm->userdomain_buf;
+
+  for (int i = 0; i < netntlm->user_len; i += 2)
+  {
+    out_buf[out_len++] = ptr[i];
+  }
+
+  out_buf[out_len++] = ':';
+  out_buf[out_len++] = ':';
+
+  ptr += netntlm->user_len;
+
+  for (int i = 0; i < netntlm->domain_len; i += 2)
+  {
+    out_buf[out_len++] = ptr[i];
+  }
+
+  out_buf[out_len++] = ':';
+
+  ptr = (u8 *) netntlm->chall_buf;
+
+  for (int i = 0; i < netntlm->srvchall_len; i++)
+  {
+    u8_to_hex (ptr[i], out_buf + out_len); out_len += 2;
+  }
+
+  out_buf[out_len++] = ':';
+
+  u32_to_hex (tmp[0], out_buf + out_len); out_len += 8;
+  u32_to_hex (tmp[1], out_buf + out_len); out_len += 8;
+  u32_to_hex (tmp[2], out_buf + out_len); out_len += 8;
+  u32_to_hex (tmp[3], out_buf + out_len); out_len += 8;
+
+  out_buf[out_len++] = ':';
+
+  ptr += netntlm->srvchall_len;
+
+  for (int i = 0; i < netntlm->clichall_len; i++)
+  {
+    u8_to_hex (ptr[i], out_buf + out_len); out_len += 2;
+  }
+
+  return out_len;
+}
+
+u32 module_pw_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u32 pw_max = 32; // Length of a NT hash
+
+  return pw_max;
+}
+
+u32 module_pw_min (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u32 pw_min = 32; // Length of a NT hash
+
+  return pw_min;
+}
+
+const char *module_benchmark_mask (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+ const char *mask = "?a?a?a?a?a?a?a?axxxxxxxxxxxxxxxx";
+ return mask;
+}
+
+void module_init (module_ctx_t *module_ctx)
+{
+  module_ctx->module_context_size             = MODULE_CONTEXT_SIZE_CURRENT;
+  module_ctx->module_interface_version        = MODULE_INTERFACE_VERSION_CURRENT;
+
+  module_ctx->module_attack_exec              = module_attack_exec;
+  module_ctx->module_benchmark_esalt          = MODULE_DEFAULT;
+  module_ctx->module_benchmark_hook_salt      = MODULE_DEFAULT;
+  module_ctx->module_benchmark_mask           = module_benchmark_mask;
+  module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
+  module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
+  module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
+  module_ctx->module_dgst_pos0                = module_dgst_pos0;
+  module_ctx->module_dgst_pos1                = module_dgst_pos1;
+  module_ctx->module_dgst_pos2                = module_dgst_pos2;
+  module_ctx->module_dgst_pos3                = module_dgst_pos3;
+  module_ctx->module_dgst_size                = module_dgst_size;
+  module_ctx->module_dictstat_disable         = MODULE_DEFAULT;
+  module_ctx->module_esalt_size               = module_esalt_size;
+  module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
+  module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
+  module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_save         = MODULE_DEFAULT;
+  module_ctx->module_hash_decode_potfile      = MODULE_DEFAULT;
+  module_ctx->module_hash_decode_zero_hash    = MODULE_DEFAULT;
+  module_ctx->module_hash_decode              = module_hash_decode;
+  module_ctx->module_hash_encode_status       = MODULE_DEFAULT;
+  module_ctx->module_hash_encode_potfile      = MODULE_DEFAULT;
+  module_ctx->module_hash_encode              = module_hash_encode;
+  module_ctx->module_hash_init_selftest       = MODULE_DEFAULT;
+  module_ctx->module_hash_mode                = MODULE_DEFAULT;
+  module_ctx->module_hash_category            = module_hash_category;
+  module_ctx->module_hash_name                = module_hash_name;
+  module_ctx->module_hashes_count_min         = MODULE_DEFAULT;
+  module_ctx->module_hashes_count_max         = MODULE_DEFAULT;
+  module_ctx->module_hlfmt_disable            = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_size    = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_init    = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_term    = MODULE_DEFAULT;
+  module_ctx->module_hook12                   = MODULE_DEFAULT;
+  module_ctx->module_hook23                   = MODULE_DEFAULT;
+  module_ctx->module_hook_salt_size           = MODULE_DEFAULT;
+  module_ctx->module_hook_size                = MODULE_DEFAULT;
+  module_ctx->module_jit_build_options        = MODULE_DEFAULT;
+  module_ctx->module_jit_cache_disable        = MODULE_DEFAULT;
+  module_ctx->module_kernel_accel_max         = MODULE_DEFAULT;
+  module_ctx->module_kernel_accel_min         = MODULE_DEFAULT;
+  module_ctx->module_kernel_loops_max         = MODULE_DEFAULT;
+  module_ctx->module_kernel_loops_min         = MODULE_DEFAULT;
+  module_ctx->module_kernel_threads_max       = MODULE_DEFAULT;
+  module_ctx->module_kernel_threads_min       = MODULE_DEFAULT;
+  module_ctx->module_kern_type                = module_kern_type;
+  module_ctx->module_kern_type_dynamic        = MODULE_DEFAULT;
+  module_ctx->module_opti_type                = module_opti_type;
+  module_ctx->module_opts_type                = module_opts_type;
+  module_ctx->module_outfile_check_disable    = MODULE_DEFAULT;
+  module_ctx->module_outfile_check_nocomp     = MODULE_DEFAULT;
+  module_ctx->module_potfile_custom_check     = MODULE_DEFAULT;
+  module_ctx->module_potfile_disable          = MODULE_DEFAULT;
+  module_ctx->module_potfile_keep_all_hashes  = MODULE_DEFAULT;
+  module_ctx->module_pwdump_column            = MODULE_DEFAULT;
+  module_ctx->module_pw_max                   = module_pw_max;
+  module_ctx->module_pw_min                   = module_pw_min;
+  module_ctx->module_salt_max                 = MODULE_DEFAULT;
+  module_ctx->module_salt_min                 = MODULE_DEFAULT;
+  module_ctx->module_salt_type                = module_salt_type;
+  module_ctx->module_separator                = MODULE_DEFAULT;
+  module_ctx->module_st_hash                  = module_st_hash;
+  module_ctx->module_st_pass                  = module_st_pass;
+  module_ctx->module_tmp_size                 = module_tmp_size;
+  module_ctx->module_unstable_warning         = MODULE_DEFAULT;
+  module_ctx->module_warmup_disable           = MODULE_DEFAULT;
+}
diff --git a/src/modules/module_27200.c b/src/modules/module_27200.c
new file mode 100644
index 000000000..fa7881bd3
--- /dev/null
+++ b/src/modules/module_27200.c
@@ -0,0 +1,257 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+#include "common.h"
+#include "types.h"
+#include "modules.h"
+#include "bitops.h"
+#include "convert.h"
+#include "shared.h"
+
+static const u32   ATTACK_EXEC    = ATTACK_EXEC_INSIDE_KERNEL;
+static const u32   DGST_POS0      = 3;
+static const u32   DGST_POS1      = 4;
+static const u32   DGST_POS2      = 2;
+static const u32   DGST_POS3      = 1;
+static const u32   DGST_SIZE      = DGST_SIZE_4_5;
+static const u32   HASH_CATEGORY  = HASH_CATEGORY_RAW_HASH_SALTED;
+static const char *HASH_NAME      = "Ruby on Rails Restful Auth (single round, no sitekey)";
+static const u64   KERN_TYPE      = 27200;
+static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
+                                  | OPTI_TYPE_PRECOMPUTE_INIT
+                                  | OPTI_TYPE_EARLY_SKIP
+                                  | OPTI_TYPE_NOT_ITERATED
+                                  | OPTI_TYPE_PREPENDED_SALT
+                                  | OPTI_TYPE_RAW_HASH;
+static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_BE;
+static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
+static const char *ST_PASS        = "hashcat";
+static const char *ST_HASH        = "3999d08db95797891ec77f07223ca81bf43e1be2:5dcc47b04c49d3c8e1b9e4ec367fddeed21b7b85";
+
+u32         module_attack_exec    (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ATTACK_EXEC;     }
+u32         module_dgst_pos0      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS0;       }
+u32         module_dgst_pos1      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS1;       }
+u32         module_dgst_pos2      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS2;       }
+u32         module_dgst_pos3      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS3;       }
+u32         module_dgst_size      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_SIZE;       }
+u32         module_hash_category  (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return HASH_CATEGORY;   }
+const char *module_hash_name      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return HASH_NAME;       }
+u64         module_kern_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return KERN_TYPE;       }
+u32         module_opti_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return OPTI_TYPE;       }
+u64         module_opts_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return OPTS_TYPE;       }
+u32         module_salt_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return SALT_TYPE;       }
+const char *module_st_hash        (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ST_HASH;         }
+const char *module_st_pass        (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ST_PASS;         }
+
+u32 module_pw_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const bool optimized_kernel = (hashconfig->opti_type & OPTI_TYPE_OPTIMIZED_KERNEL);
+
+  u32 pw_max = PW_MAX;
+
+  if (optimized_kernel == true)
+  {
+    pw_max = 9;
+  }
+
+  return pw_max;
+}
+
+u32 module_salt_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u32 salt_max = 40;
+
+  return salt_max;
+}
+
+u32 module_salt_min (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u32 salt_min = 40;
+
+  return salt_min;
+}
+
+int module_hash_decode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED void *digest_buf, MAYBE_UNUSED salt_t *salt, MAYBE_UNUSED void *esalt_buf, MAYBE_UNUSED void *hook_salt_buf, MAYBE_UNUSED hashinfo_t *hash_info, const char *line_buf, MAYBE_UNUSED const int line_len)
+{
+  u32 *digest = (u32 *) digest_buf;
+
+  token_t token;
+
+  token.token_cnt  = 2;
+
+  token.sep[0]     = hashconfig->separator;
+  token.len_min[0] = 40;
+  token.len_max[0] = 40;
+  token.attr[0]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  token.len_min[1] = 40;
+  token.len_max[1] = 40;
+  token.attr[1]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  const int rc_tokenizer = input_tokenizer ((const u8 *) line_buf, line_len, &token);
+
+  if (rc_tokenizer != PARSER_OK) return (rc_tokenizer);
+
+  const u8 *hash_pos = token.buf[0];
+
+  digest[0] = hex_to_u32 (hash_pos +  0);
+  digest[1] = hex_to_u32 (hash_pos +  8);
+  digest[2] = hex_to_u32 (hash_pos + 16);
+  digest[3] = hex_to_u32 (hash_pos + 24);
+  digest[4] = hex_to_u32 (hash_pos + 32);
+
+  digest[0] = byte_swap_32 (digest[0]);
+  digest[1] = byte_swap_32 (digest[1]);
+  digest[2] = byte_swap_32 (digest[2]);
+  digest[3] = byte_swap_32 (digest[3]);
+  digest[4] = byte_swap_32 (digest[4]);
+
+  if (hashconfig->opti_type & OPTI_TYPE_OPTIMIZED_KERNEL)
+  {
+    digest[0] -= SHA1M_A;
+    digest[1] -= SHA1M_B;
+    digest[2] -= SHA1M_C;
+    digest[3] -= SHA1M_D;
+    digest[4] -= SHA1M_E;
+  }
+
+  const u8 *salt_pos = token.buf[1];
+  const int salt_len = token.len[1];
+
+  const bool parse_rc = generic_salt_decode (hashconfig, salt_pos, salt_len, (u8 *) salt->salt_buf, (int *) &salt->salt_len);
+
+  if (parse_rc == false) return (PARSER_SALT_LENGTH);
+
+  return (PARSER_OK);
+}
+
+int module_hash_encode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const void *digest_buf, MAYBE_UNUSED const salt_t *salt, MAYBE_UNUSED const void *esalt_buf, MAYBE_UNUSED const void *hook_salt_buf, MAYBE_UNUSED const hashinfo_t *hash_info, char *line_buf, MAYBE_UNUSED const int line_size)
+{
+  const u32 *digest = (const u32 *) digest_buf;
+
+  // we can not change anything in the original buffer, otherwise destroying sorting
+  // therefore create some local buffer
+
+  u32 tmp[5];
+
+  tmp[0] = digest[0];
+  tmp[1] = digest[1];
+  tmp[2] = digest[2];
+  tmp[3] = digest[3];
+  tmp[4] = digest[4];
+
+  if (hashconfig->opti_type & OPTI_TYPE_OPTIMIZED_KERNEL)
+  {
+    tmp[0] += SHA1M_A;
+    tmp[1] += SHA1M_B;
+    tmp[2] += SHA1M_C;
+    tmp[3] += SHA1M_D;
+    tmp[4] += SHA1M_E;
+  }
+
+  tmp[0] = byte_swap_32 (tmp[0]);
+  tmp[1] = byte_swap_32 (tmp[1]);
+  tmp[2] = byte_swap_32 (tmp[2]);
+  tmp[3] = byte_swap_32 (tmp[3]);
+  tmp[4] = byte_swap_32 (tmp[4]);
+
+  u8 *out_buf = (u8 *) line_buf;
+
+  int out_len = 0;
+
+  u32_to_hex (tmp[0], out_buf + out_len); out_len += 8;
+  u32_to_hex (tmp[1], out_buf + out_len); out_len += 8;
+  u32_to_hex (tmp[2], out_buf + out_len); out_len += 8;
+  u32_to_hex (tmp[3], out_buf + out_len); out_len += 8;
+  u32_to_hex (tmp[4], out_buf + out_len); out_len += 8;
+
+  out_buf[out_len] = hashconfig->separator;
+
+  out_len += 1;
+
+  out_len += generic_salt_encode (hashconfig, (const u8 *) salt->salt_buf, salt->salt_len, out_buf + out_len);
+  
+  return out_len;
+}
+
+void module_init (module_ctx_t *module_ctx)
+{
+  module_ctx->module_context_size             = MODULE_CONTEXT_SIZE_CURRENT;
+  module_ctx->module_interface_version        = MODULE_INTERFACE_VERSION_CURRENT;
+
+  module_ctx->module_attack_exec              = module_attack_exec;
+  module_ctx->module_benchmark_esalt          = MODULE_DEFAULT;
+  module_ctx->module_benchmark_hook_salt      = MODULE_DEFAULT;
+  module_ctx->module_benchmark_mask           = MODULE_DEFAULT;
+  module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
+  module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
+  module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
+  module_ctx->module_dgst_pos0                = module_dgst_pos0;
+  module_ctx->module_dgst_pos1                = module_dgst_pos1;
+  module_ctx->module_dgst_pos2                = module_dgst_pos2;
+  module_ctx->module_dgst_pos3                = module_dgst_pos3;
+  module_ctx->module_dgst_size                = module_dgst_size;
+  module_ctx->module_dictstat_disable         = MODULE_DEFAULT;
+  module_ctx->module_esalt_size               = MODULE_DEFAULT;
+  module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
+  module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
+  module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_save         = MODULE_DEFAULT;
+  module_ctx->module_hash_decode_potfile      = MODULE_DEFAULT;
+  module_ctx->module_hash_decode_zero_hash    = MODULE_DEFAULT;
+  module_ctx->module_hash_decode              = module_hash_decode;
+  module_ctx->module_hash_encode_status       = MODULE_DEFAULT;
+  module_ctx->module_hash_encode_potfile      = MODULE_DEFAULT;
+  module_ctx->module_hash_encode              = module_hash_encode;
+  module_ctx->module_hash_init_selftest       = MODULE_DEFAULT;
+  module_ctx->module_hash_mode                = MODULE_DEFAULT;
+  module_ctx->module_hash_category            = module_hash_category;
+  module_ctx->module_hash_name                = module_hash_name;
+  module_ctx->module_hashes_count_min         = MODULE_DEFAULT;
+  module_ctx->module_hashes_count_max         = MODULE_DEFAULT;
+  module_ctx->module_hlfmt_disable            = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_size    = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_init    = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_term    = MODULE_DEFAULT;
+  module_ctx->module_hook12                   = MODULE_DEFAULT;
+  module_ctx->module_hook23                   = MODULE_DEFAULT;
+  module_ctx->module_hook_salt_size           = MODULE_DEFAULT;
+  module_ctx->module_hook_size                = MODULE_DEFAULT;
+  module_ctx->module_jit_build_options        = MODULE_DEFAULT;
+  module_ctx->module_jit_cache_disable        = MODULE_DEFAULT;
+  module_ctx->module_kernel_accel_max         = MODULE_DEFAULT;
+  module_ctx->module_kernel_accel_min         = MODULE_DEFAULT;
+  module_ctx->module_kernel_loops_max         = MODULE_DEFAULT;
+  module_ctx->module_kernel_loops_min         = MODULE_DEFAULT;
+  module_ctx->module_kernel_threads_max       = MODULE_DEFAULT;
+  module_ctx->module_kernel_threads_min       = MODULE_DEFAULT;
+  module_ctx->module_kern_type                = module_kern_type;
+  module_ctx->module_kern_type_dynamic        = MODULE_DEFAULT;
+  module_ctx->module_opti_type                = module_opti_type;
+  module_ctx->module_opts_type                = module_opts_type;
+  module_ctx->module_outfile_check_disable    = MODULE_DEFAULT;
+  module_ctx->module_outfile_check_nocomp     = MODULE_DEFAULT;
+  module_ctx->module_potfile_custom_check     = MODULE_DEFAULT;
+  module_ctx->module_potfile_disable          = MODULE_DEFAULT;
+  module_ctx->module_potfile_keep_all_hashes  = MODULE_DEFAULT;
+  module_ctx->module_pwdump_column            = MODULE_DEFAULT;
+  module_ctx->module_pw_max                   = module_pw_max;
+  module_ctx->module_pw_min                   = MODULE_DEFAULT;
+  module_ctx->module_salt_max                 = module_salt_max;
+  module_ctx->module_salt_min                 = module_salt_min;
+  module_ctx->module_salt_type                = module_salt_type;
+  module_ctx->module_separator                = MODULE_DEFAULT;
+  module_ctx->module_st_hash                  = module_st_hash;
+  module_ctx->module_st_pass                  = module_st_pass;
+  module_ctx->module_tmp_size                 = MODULE_DEFAULT;
+  module_ctx->module_unstable_warning         = MODULE_DEFAULT;
+  module_ctx->module_warmup_disable           = MODULE_DEFAULT;
+}
diff --git a/src/modules/module_27300.c b/src/modules/module_27300.c
new file mode 100644
index 000000000..1615dfad4
--- /dev/null
+++ b/src/modules/module_27300.c
@@ -0,0 +1,391 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+#include "common.h"
+#include "types.h"
+#include "modules.h"
+#include "bitops.h"
+#include "convert.h"
+#include "shared.h"
+#include "memory.h"
+#include "emu_inc_hash_sha1.h"
+
+static const u32   ATTACK_EXEC    = ATTACK_EXEC_OUTSIDE_KERNEL;
+static const u32   DGST_POS0      = 2;
+static const u32   DGST_POS1      = 3;
+static const u32   DGST_POS2      = 0;
+static const u32   DGST_POS3      = 1;
+static const u32   DGST_SIZE      = DGST_SIZE_8_6;
+static const u32   HASH_CATEGORY  = HASH_CATEGORY_NETWORK_PROTOCOL;
+static const char *HASH_NAME      = "SNMPv3 HMAC-SHA512-384";
+static const u64   KERN_TYPE      = 27300;
+static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
+                                  | OPTI_TYPE_USES_BITS_64;
+static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_BE;
+static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
+static const char *ST_PASS        = "hashcat1";
+static const char *ST_HASH        = "$SNMPv3$6$45889431$3081b702010330110204367c80d4020300ffe30401010201030460305e041180001f88808106d566db57fd600000000002011002020118040e6d61747269785f5348412d35313204300000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000400303d041180001f88808106d566db57fd60000000000400a22602046ea3546f0201000201003018301606082b06010201010200060a2b06010401bf0803020a$80001f88808106d566db57fd6000000000$9e4681768d5dee9e2d0ca7380dfa19f0a0f805c550142b889af548f5506c2c3587df980707600b58d97ed1beaa9feaf9";
+
+u32         module_attack_exec    (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ATTACK_EXEC;     }
+u32         module_dgst_pos0      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS0;       }
+u32         module_dgst_pos1      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS1;       }
+u32         module_dgst_pos2      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS2;       }
+u32         module_dgst_pos3      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS3;       }
+u32         module_dgst_size      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_SIZE;       }
+u32         module_hash_category  (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return HASH_CATEGORY;   }
+const char *module_hash_name      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return HASH_NAME;       }
+u64         module_kern_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return KERN_TYPE;       }
+u32         module_opti_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return OPTI_TYPE;       }
+u64         module_opts_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return OPTS_TYPE;       }
+u32         module_salt_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return SALT_TYPE;       }
+const char *module_st_hash        (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ST_HASH;         }
+const char *module_st_pass        (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ST_PASS;         }
+
+static const char *SIGNATURE_SNMPV3 = "$SNMPv3$6$";
+
+#define SNMPV3_SALT_MAX             1500
+#define SNMPV3_ENGINEID_MAX         34
+#define SNMPV3_MSG_AUTH_PARAMS_LEN  48
+#define SNMPV3_ROUNDS               1048576
+#define SNMPV3_MAX_PW_LENGTH        128
+
+#define SNMPV3_TMP_ELEMS            8192 // 8192 = (256 (max pw length) * SNMPV3_MAX_PW_LENGTH) / sizeof (u32)
+#define SNMPV3_HASH_ELEMS           8
+
+#define SNMPV3_MAX_SALT_ELEMS       512 // 512 * 4 = 2048 > 1500, also has to be multiple of SNMPV3_MAX_PW_LENGTH
+#define SNMPV3_MAX_ENGINE_ELEMS     32  // 32 * 4 = 128 > 34, also has to be multiple of SNMPV3_MAX_PW_LENGTH
+#define SNMPV3_MAX_PNUM_ELEMS       4   // 4 * 4 = 16 > 9
+
+typedef struct hmac_sha512_tmp
+{
+  u32 tmp[SNMPV3_TMP_ELEMS];
+  u64 h[SNMPV3_HASH_ELEMS];
+
+} hmac_sha512_tmp_t;
+
+typedef struct snmpv3
+{
+  u32 salt_buf[SNMPV3_MAX_SALT_ELEMS];
+  u32 salt_len;
+
+  u32 engineID_buf[SNMPV3_MAX_ENGINE_ELEMS];
+  u32 engineID_len;
+
+  u32 packet_number[SNMPV3_MAX_PNUM_ELEMS];
+
+} snmpv3_t;
+
+u32 module_pw_min (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u32 pw_min = 8;
+
+  return pw_min;
+}
+
+bool module_unstable_warning (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra, MAYBE_UNUSED const hc_device_param_t *device_param)
+{
+  if (device_param->opencl_platform_vendor_id == VENDOR_ID_APPLE)
+  {
+    // self-test failed
+    if ((device_param->opencl_device_vendor_id == VENDOR_ID_AMD) && (device_param->opencl_device_type & CL_DEVICE_TYPE_GPU))
+    {
+      return true;
+    }
+  }
+
+  return false;
+}
+
+char *module_jit_build_options (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra, MAYBE_UNUSED const hashes_t *hashes, MAYBE_UNUSED const hc_device_param_t *device_param)
+{
+  char *jit_build_options = NULL;
+
+  // Extra treatment for Apple systems
+  if (device_param->opencl_platform_vendor_id == VENDOR_ID_APPLE)
+  {
+    return jit_build_options;
+  }
+
+  // HIP
+  if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD_USE_HIP)
+  {
+    hc_asprintf (&jit_build_options, "-D _unroll");
+  }
+
+  // ROCM
+  if ((device_param->opencl_device_vendor_id == VENDOR_ID_AMD) && (device_param->has_vperm == true))
+  {
+    hc_asprintf (&jit_build_options, "-D _unroll");
+  }
+
+  return jit_build_options;
+}
+
+u64 module_esalt_size (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u64 esalt_size = (const u64) sizeof (snmpv3_t);
+
+  return esalt_size;
+}
+
+u64 module_tmp_size (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u64 tmp_size = (const u64) sizeof (hmac_sha512_tmp_t);
+
+  return tmp_size;
+}
+
+u32 module_kernel_loops_min (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  // we need to fix iteration count to guarantee the loop count is a multiple of SNMPV3_MAX_PW_LENGTH
+  // 2k calls to sha512_transform typically is enough to overtime pcie bottleneck
+
+  const u32 kernel_loops_min = 2048 * SNMPV3_MAX_PW_LENGTH;
+
+  return kernel_loops_min;
+}
+
+u32 module_kernel_loops_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u32 kernel_loops_max = 2048 * SNMPV3_MAX_PW_LENGTH;
+
+  return kernel_loops_max;
+}
+
+int module_hash_decode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED void *digest_buf, MAYBE_UNUSED salt_t *salt, MAYBE_UNUSED void *esalt_buf, MAYBE_UNUSED void *hook_salt_buf, MAYBE_UNUSED hashinfo_t *hash_info, const char *line_buf, MAYBE_UNUSED const int line_len)
+{
+  u64 *digest = (u64 *) digest_buf;
+
+  snmpv3_t *snmpv3 = (snmpv3_t *) esalt_buf;
+
+  token_t token;
+
+  token.token_cnt  = 5;
+  token.signatures_cnt    = 1;
+  token.signatures_buf[0] = SIGNATURE_SNMPV3;
+
+  token.len[0]     = 10;
+  token.attr[0]    = TOKEN_ATTR_FIXED_LENGTH
+                   | TOKEN_ATTR_VERIFY_SIGNATURE;
+
+  // packet number
+  token.len_min[1] = 1;
+  token.len_max[1] = 8;
+  token.sep[1]     = '$';
+  token.attr[1]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_DIGIT;
+  // salt
+  token.len_min[2] = SNMPV3_MSG_AUTH_PARAMS_LEN * 2;
+  token.len_max[2] = SNMPV3_SALT_MAX * 2;
+  token.sep[2]     = '$';
+  token.attr[2]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  // engineid
+  token.len_min[3] = 26;
+  token.len_max[3] = SNMPV3_ENGINEID_MAX;
+  token.sep[3]     = '$';
+  token.attr[3]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  // digest
+  token.len[4]     = SNMPV3_MSG_AUTH_PARAMS_LEN * 2;
+  token.sep[4]     = '$';
+  token.attr[4]    = TOKEN_ATTR_FIXED_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  const int rc_tokenizer = input_tokenizer ((const u8 *) line_buf, line_len, &token);
+
+  if (rc_tokenizer != PARSER_OK) return (rc_tokenizer);
+
+  // packet number
+
+  const u8 *packet_number_pos = token.buf[1];
+  const int packet_number_len = token.len[1];
+
+  memset (snmpv3->packet_number, 0, sizeof (snmpv3->packet_number));
+
+  strncpy ((char *) snmpv3->packet_number, (char *) packet_number_pos, packet_number_len);
+
+  // salt
+
+  const u8 *salt_pos = token.buf[2];
+  const int salt_len = token.len[2];
+
+  u8 *salt_ptr = (u8 *) snmpv3->salt_buf;
+
+  snmpv3->salt_len = hex_decode (salt_pos, salt_len, salt_ptr);
+
+  salt->salt_iter = SNMPV3_ROUNDS;
+
+  // handle unique salts detection
+
+  sha1_ctx_t sha1_ctx;
+
+  sha1_init   (&sha1_ctx);
+  sha1_update (&sha1_ctx, snmpv3->salt_buf, snmpv3->salt_len);
+  sha1_final  (&sha1_ctx);
+
+  // store sha1(snmpv3->salt_buf) in salt_buf
+
+  salt->salt_len = 20;
+
+  memcpy (salt->salt_buf, sha1_ctx.h, salt->salt_len);
+
+  // engineid
+
+  const u8 *engineID_pos = token.buf[3];
+  const int engineID_len = token.len[3];
+
+  memset (snmpv3->engineID_buf, 0, sizeof (snmpv3->engineID_buf));
+
+  u8 *engineID_ptr = (u8 *) snmpv3->engineID_buf;
+
+  hex_decode (engineID_pos, engineID_len, engineID_ptr);
+
+  // force len to 17, zero padding
+  snmpv3->engineID_len = SNMPV3_ENGINEID_MAX / 2;
+
+  // digest
+
+  const u8 *hash_pos = token.buf[4];
+
+  digest[0] = hex_to_u64 (hash_pos +  0);
+  digest[1] = hex_to_u64 (hash_pos + 16);
+  digest[2] = hex_to_u64 (hash_pos + 32);
+  digest[3] = hex_to_u64 (hash_pos + 48);
+  digest[4] = hex_to_u64 (hash_pos + 64);
+  digest[5] = hex_to_u64 (hash_pos + 80);
+
+  digest[0] = byte_swap_64 (digest[0]);
+  digest[1] = byte_swap_64 (digest[1]);
+  digest[2] = byte_swap_64 (digest[2]);
+  digest[3] = byte_swap_64 (digest[3]);
+  digest[4] = byte_swap_64 (digest[4]);
+  digest[5] = byte_swap_64 (digest[5]);
+
+  return (PARSER_OK);
+}
+
+int module_hash_encode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const void *digest_buf, MAYBE_UNUSED const salt_t *salt, MAYBE_UNUSED const void *esalt_buf, MAYBE_UNUSED const void *hook_salt_buf, MAYBE_UNUSED const hashinfo_t *hash_info, char *line_buf, MAYBE_UNUSED const int line_size)
+{
+  const u64 *digest = (const u64 *) digest_buf;
+
+  snmpv3_t *snmpv3 = (snmpv3_t *) esalt_buf;
+
+  u8 *out_buf = (u8 *) line_buf;
+
+  int out_len = snprintf (line_buf, line_size, "%s%s$", SIGNATURE_SNMPV3, (char *) snmpv3->packet_number);
+
+  out_len += hex_encode ((u8 *) snmpv3->salt_buf, snmpv3->salt_len, out_buf + out_len);
+
+  out_buf[out_len] = '$';
+
+  out_len++;
+
+  out_len += hex_encode ((u8 *) snmpv3->engineID_buf, snmpv3->engineID_len, out_buf + out_len);
+
+  out_buf[out_len] = '$';
+
+  out_len++;
+
+  u64 digest_tmp[6];
+
+  digest_tmp[0] = byte_swap_64 (digest[0]);
+  digest_tmp[1] = byte_swap_64 (digest[1]);
+  digest_tmp[2] = byte_swap_64 (digest[2]);
+  digest_tmp[3] = byte_swap_64 (digest[3]);
+  digest_tmp[4] = byte_swap_64 (digest[4]);
+  digest_tmp[5] = byte_swap_64 (digest[5]);
+
+  u64_to_hex (digest_tmp[0], out_buf + out_len); out_len += 16;
+  u64_to_hex (digest_tmp[1], out_buf + out_len); out_len += 16;
+  u64_to_hex (digest_tmp[2], out_buf + out_len); out_len += 16;
+  u64_to_hex (digest_tmp[3], out_buf + out_len); out_len += 16;
+  u64_to_hex (digest_tmp[4], out_buf + out_len); out_len += 16;
+  u64_to_hex (digest_tmp[5], out_buf + out_len); out_len += 16;
+
+  out_buf[out_len] = 0;
+
+  return out_len;
+}
+
+void module_init (module_ctx_t *module_ctx)
+{
+  module_ctx->module_context_size             = MODULE_CONTEXT_SIZE_CURRENT;
+  module_ctx->module_interface_version        = MODULE_INTERFACE_VERSION_CURRENT;
+
+  module_ctx->module_attack_exec              = module_attack_exec;
+  module_ctx->module_benchmark_esalt          = MODULE_DEFAULT;
+  module_ctx->module_benchmark_hook_salt      = MODULE_DEFAULT;
+  module_ctx->module_benchmark_mask           = MODULE_DEFAULT;
+  module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
+  module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
+  module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
+  module_ctx->module_dgst_pos0                = module_dgst_pos0;
+  module_ctx->module_dgst_pos1                = module_dgst_pos1;
+  module_ctx->module_dgst_pos2                = module_dgst_pos2;
+  module_ctx->module_dgst_pos3                = module_dgst_pos3;
+  module_ctx->module_dgst_size                = module_dgst_size;
+  module_ctx->module_dictstat_disable         = MODULE_DEFAULT;
+  module_ctx->module_esalt_size               = module_esalt_size;
+  module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
+  module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
+  module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_save         = MODULE_DEFAULT;
+  module_ctx->module_hash_decode_potfile      = MODULE_DEFAULT;
+  module_ctx->module_hash_decode_zero_hash    = MODULE_DEFAULT;
+  module_ctx->module_hash_decode              = module_hash_decode;
+  module_ctx->module_hash_encode_status       = MODULE_DEFAULT;
+  module_ctx->module_hash_encode_potfile      = MODULE_DEFAULT;
+  module_ctx->module_hash_encode              = module_hash_encode;
+  module_ctx->module_hash_init_selftest       = MODULE_DEFAULT;
+  module_ctx->module_hash_mode                = MODULE_DEFAULT;
+  module_ctx->module_hash_category            = module_hash_category;
+  module_ctx->module_hash_name                = module_hash_name;
+  module_ctx->module_hashes_count_min         = MODULE_DEFAULT;
+  module_ctx->module_hashes_count_max         = MODULE_DEFAULT;
+  module_ctx->module_hlfmt_disable            = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_size    = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_init    = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_term    = MODULE_DEFAULT;
+  module_ctx->module_hook12                   = MODULE_DEFAULT;
+  module_ctx->module_hook23                   = MODULE_DEFAULT;
+  module_ctx->module_hook_salt_size           = MODULE_DEFAULT;
+  module_ctx->module_hook_size                = MODULE_DEFAULT;
+  module_ctx->module_jit_build_options        = module_jit_build_options;
+  module_ctx->module_jit_cache_disable        = MODULE_DEFAULT;
+  module_ctx->module_kernel_accel_max         = MODULE_DEFAULT;
+  module_ctx->module_kernel_accel_min         = MODULE_DEFAULT;
+  module_ctx->module_kernel_loops_max         = module_kernel_loops_max;
+  module_ctx->module_kernel_loops_min         = module_kernel_loops_min;
+  module_ctx->module_kernel_threads_max       = MODULE_DEFAULT;
+  module_ctx->module_kernel_threads_min       = MODULE_DEFAULT;
+  module_ctx->module_kern_type                = module_kern_type;
+  module_ctx->module_kern_type_dynamic        = MODULE_DEFAULT;
+  module_ctx->module_opti_type                = module_opti_type;
+  module_ctx->module_opts_type                = module_opts_type;
+  module_ctx->module_outfile_check_disable    = MODULE_DEFAULT;
+  module_ctx->module_outfile_check_nocomp     = MODULE_DEFAULT;
+  module_ctx->module_potfile_custom_check     = MODULE_DEFAULT;
+  module_ctx->module_potfile_disable          = MODULE_DEFAULT;
+  module_ctx->module_potfile_keep_all_hashes  = MODULE_DEFAULT;
+  module_ctx->module_pwdump_column            = MODULE_DEFAULT;
+  module_ctx->module_pw_max                   = MODULE_DEFAULT;
+  module_ctx->module_pw_min                   = module_pw_min;
+  module_ctx->module_salt_max                 = MODULE_DEFAULT;
+  module_ctx->module_salt_min                 = MODULE_DEFAULT;
+  module_ctx->module_salt_type                = module_salt_type;
+  module_ctx->module_separator                = MODULE_DEFAULT;
+  module_ctx->module_st_hash                  = module_st_hash;
+  module_ctx->module_st_pass                  = module_st_pass;
+  module_ctx->module_tmp_size                 = module_tmp_size;
+  module_ctx->module_unstable_warning         = module_unstable_warning;
+  module_ctx->module_warmup_disable           = MODULE_DEFAULT;
+}
diff --git a/src/modules/module_27400.c b/src/modules/module_27400.c
new file mode 100644
index 000000000..86efb742e
--- /dev/null
+++ b/src/modules/module_27400.c
@@ -0,0 +1,307 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+#include "common.h"
+#include "types.h"
+#include "modules.h"
+#include "bitops.h"
+#include "convert.h"
+#include "shared.h"
+
+static const u32   ATTACK_EXEC    = ATTACK_EXEC_OUTSIDE_KERNEL;
+static const u32   DGST_POS0      = 0;
+static const u32   DGST_POS1      = 1;
+static const u32   DGST_POS2      = 2;
+static const u32   DGST_POS3      = 3;
+static const u32   DGST_SIZE      = DGST_SIZE_4_4;
+static const u32   HASH_CATEGORY  = HASH_CATEGORY_FDE;
+static const char *HASH_NAME      = "VMware VMX (PBKDF2-HMAC-SHA1 + AES-256-CBC)";
+static const u64   KERN_TYPE      = 27400;
+static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
+                                  | OPTI_TYPE_SLOW_HASH_SIMD_LOOP;
+static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
+                                  | OPTS_TYPE_ST_HEX;
+static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
+static const char *ST_PASS        = "hashcat";
+static const char *ST_HASH        = "$vmx$0$10000$264bbab02fdf7c1a793651120bec3723$cbb368564d8dfb99f509d4922f4693413f3816af713f0e76bc2409ff9336935d";
+
+u32         module_attack_exec    (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ATTACK_EXEC;     }
+u32         module_dgst_pos0      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS0;       }
+u32         module_dgst_pos1      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS1;       }
+u32         module_dgst_pos2      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS2;       }
+u32         module_dgst_pos3      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS3;       }
+u32         module_dgst_size      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_SIZE;       }
+u32         module_hash_category  (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return HASH_CATEGORY;   }
+const char *module_hash_name      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return HASH_NAME;       }
+u64         module_kern_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return KERN_TYPE;       }
+u32         module_opti_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return OPTI_TYPE;       }
+u64         module_opts_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return OPTS_TYPE;       }
+u32         module_salt_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return SALT_TYPE;       }
+const char *module_st_hash        (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ST_HASH;         }
+const char *module_st_pass        (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ST_PASS;         }
+
+typedef struct pbkdf2_sha1_tmp
+{
+  u32  ipad[5];
+  u32  opad[5];
+
+  u32  dgst[32];
+  u32  out[32];
+
+} pbkdf2_sha1_tmp_t;
+
+typedef struct vmware_vmx
+{
+  u32 salt_buf[64];
+  u32 iv_buf[4];
+  u32 ct_buf[4];
+
+} vmware_vmx_t;
+
+static const char *SIGNATURE_VMWARE_VMX = "$vmx$0$";
+
+u64 module_esalt_size (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u64 esalt_size = (const u64) sizeof (vmware_vmx_t);
+
+  return esalt_size;
+}
+
+u64 module_tmp_size (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u64 tmp_size = (const u64) sizeof (pbkdf2_sha1_tmp_t);
+
+  return tmp_size;
+}
+
+u32 module_pw_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  // this overrides the reductions of PW_MAX in case optimized kernel is selected
+  // IOW, even in optimized kernel mode it support length 256
+
+  const u32 pw_max = PW_MAX;
+
+  return pw_max;
+}
+
+int module_hash_decode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED void *digest_buf, MAYBE_UNUSED salt_t *salt, MAYBE_UNUSED void *esalt_buf, MAYBE_UNUSED void *hook_salt_buf, MAYBE_UNUSED hashinfo_t *hash_info, const char *line_buf, MAYBE_UNUSED const int line_len)
+{
+  u32 *digest = (u32 *) digest_buf;
+
+  vmware_vmx_t *vmware_vmx = (vmware_vmx_t *) esalt_buf;
+
+  token_t token;
+
+  token.token_cnt  = 4;
+
+  token.signatures_cnt    = 1;
+  token.signatures_buf[0] = SIGNATURE_VMWARE_VMX;
+
+  token.sep[0]     = '$';
+  token.len[0]     = 7;
+  token.attr[0]    = TOKEN_ATTR_FIXED_LENGTH
+                   | TOKEN_ATTR_VERIFY_SIGNATURE;
+
+  token.sep[1]     = '$';
+  token.len_min[1] = 1;
+  token.len_max[1] = 9;
+  token.attr[1]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_DIGIT;
+
+  token.sep[2]     = '$';
+  token.len_min[2] = 32;
+  token.len_max[2] = 32;
+  token.attr[2]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  token.sep[3]     = '$';
+  token.len_min[3] = 64;
+  token.len_max[3] = 64;
+  token.attr[3]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  const int rc_tokenizer = input_tokenizer ((const u8 *) line_buf, line_len, &token);
+
+  if (rc_tokenizer != PARSER_OK) return (rc_tokenizer);
+
+  // iter
+
+  const u8 *iter_pos = token.buf[1];
+
+  const u32 iter = hc_strtoul ((const char *) iter_pos, NULL, 10);
+
+  salt->salt_iter = iter - 1;
+
+  // salt
+
+  u8 *salt_pos = (u8 *) token.buf[2];
+  int salt_len = token.len[2];
+
+  salt->salt_buf[0] = hex_to_u32 (salt_pos +  0);
+  salt->salt_buf[1] = hex_to_u32 (salt_pos +  8);
+  salt->salt_buf[2] = hex_to_u32 (salt_pos + 16);
+  salt->salt_buf[3] = hex_to_u32 (salt_pos + 24);
+
+  salt->salt_len = salt_len / 2;
+
+  vmware_vmx->salt_buf[0] = salt->salt_buf[0];
+  vmware_vmx->salt_buf[1] = salt->salt_buf[1];
+  vmware_vmx->salt_buf[2] = salt->salt_buf[2];
+  vmware_vmx->salt_buf[3] = salt->salt_buf[3];
+
+  const u8 *hash_pos = token.buf[3];
+
+  // iv
+
+  vmware_vmx->iv_buf[0] = hex_to_u32 (hash_pos +  0);
+  vmware_vmx->iv_buf[1] = hex_to_u32 (hash_pos +  8);
+  vmware_vmx->iv_buf[2] = hex_to_u32 (hash_pos + 16);
+  vmware_vmx->iv_buf[3] = hex_to_u32 (hash_pos + 24);
+
+  vmware_vmx->iv_buf[0] = byte_swap_32 (vmware_vmx->iv_buf[0]);
+  vmware_vmx->iv_buf[1] = byte_swap_32 (vmware_vmx->iv_buf[1]);
+  vmware_vmx->iv_buf[2] = byte_swap_32 (vmware_vmx->iv_buf[2]);
+  vmware_vmx->iv_buf[3] = byte_swap_32 (vmware_vmx->iv_buf[3]);
+
+  // ct
+
+  vmware_vmx->ct_buf[0] = hex_to_u32 (hash_pos + 32);
+  vmware_vmx->ct_buf[1] = hex_to_u32 (hash_pos + 40);
+  vmware_vmx->ct_buf[2] = hex_to_u32 (hash_pos + 48);
+  vmware_vmx->ct_buf[3] = hex_to_u32 (hash_pos + 56);
+
+  vmware_vmx->ct_buf[0] = byte_swap_32 (vmware_vmx->ct_buf[0]);
+  vmware_vmx->ct_buf[1] = byte_swap_32 (vmware_vmx->ct_buf[1]);
+  vmware_vmx->ct_buf[2] = byte_swap_32 (vmware_vmx->ct_buf[2]);
+  vmware_vmx->ct_buf[3] = byte_swap_32 (vmware_vmx->ct_buf[3]);
+
+  // known pt => 'type=key:cipher='
+
+  digest[0] = 0x74797065;
+  digest[1] = 0x3d6b6579;
+  digest[2] = 0x3a636970;
+  digest[3] = 0x6865723d;
+
+  return (PARSER_OK);
+}
+
+int module_hash_encode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const void *digest_buf, MAYBE_UNUSED const salt_t *salt, MAYBE_UNUSED const void *esalt_buf, MAYBE_UNUSED const void *hook_salt_buf, MAYBE_UNUSED const hashinfo_t *hash_info, char *line_buf, MAYBE_UNUSED const int line_size)
+{
+  vmware_vmx_t *vmware_vmx = (vmware_vmx_t *) esalt_buf;
+
+  // salt
+
+  u8 salt_buf[32+1] = { 0 };
+
+  u32_to_hex (vmware_vmx->salt_buf[0], salt_buf +  0);
+  u32_to_hex (vmware_vmx->salt_buf[1], salt_buf +  8);
+  u32_to_hex (vmware_vmx->salt_buf[2], salt_buf + 16);
+  u32_to_hex (vmware_vmx->salt_buf[3], salt_buf + 24);
+
+  // iv
+
+  u8 iv_buf[32+1] = { 0 };
+
+  u32_to_hex (byte_swap_32 (vmware_vmx->iv_buf[0]), iv_buf +  0);
+  u32_to_hex (byte_swap_32 (vmware_vmx->iv_buf[1]), iv_buf +  8);
+  u32_to_hex (byte_swap_32 (vmware_vmx->iv_buf[2]), iv_buf + 16);
+  u32_to_hex (byte_swap_32 (vmware_vmx->iv_buf[3]), iv_buf + 24);
+
+  // ct
+
+  u8 ct_buf[32+1] = { 0 };
+
+  u32_to_hex (byte_swap_32 (vmware_vmx->ct_buf[0]), ct_buf +  0);
+  u32_to_hex (byte_swap_32 (vmware_vmx->ct_buf[1]), ct_buf +  8);
+  u32_to_hex (byte_swap_32 (vmware_vmx->ct_buf[2]), ct_buf + 16);
+  u32_to_hex (byte_swap_32 (vmware_vmx->ct_buf[3]), ct_buf + 24);
+
+  const int line_len = snprintf (line_buf, line_size, "%s%u$%s$%s%s",
+    SIGNATURE_VMWARE_VMX,
+    salt->salt_iter + 1,
+    salt_buf,
+    iv_buf,
+    ct_buf);
+
+  return line_len;
+}
+
+void module_init (module_ctx_t *module_ctx)
+{
+  module_ctx->module_context_size             = MODULE_CONTEXT_SIZE_CURRENT;
+  module_ctx->module_interface_version        = MODULE_INTERFACE_VERSION_CURRENT;
+
+  module_ctx->module_attack_exec              = module_attack_exec;
+  module_ctx->module_benchmark_esalt          = MODULE_DEFAULT;
+  module_ctx->module_benchmark_hook_salt      = MODULE_DEFAULT;
+  module_ctx->module_benchmark_mask           = MODULE_DEFAULT;
+  module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
+  module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
+  module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
+  module_ctx->module_dgst_pos0                = module_dgst_pos0;
+  module_ctx->module_dgst_pos1                = module_dgst_pos1;
+  module_ctx->module_dgst_pos2                = module_dgst_pos2;
+  module_ctx->module_dgst_pos3                = module_dgst_pos3;
+  module_ctx->module_dgst_size                = module_dgst_size;
+  module_ctx->module_dictstat_disable         = MODULE_DEFAULT;
+  module_ctx->module_esalt_size               = module_esalt_size;
+  module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
+  module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
+  module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_save         = MODULE_DEFAULT;
+  module_ctx->module_hash_decode_potfile      = MODULE_DEFAULT;
+  module_ctx->module_hash_decode_zero_hash    = MODULE_DEFAULT;
+  module_ctx->module_hash_decode              = module_hash_decode;
+  module_ctx->module_hash_encode_status       = MODULE_DEFAULT;
+  module_ctx->module_hash_encode_potfile      = MODULE_DEFAULT;
+  module_ctx->module_hash_encode              = module_hash_encode;
+  module_ctx->module_hash_init_selftest       = MODULE_DEFAULT;
+  module_ctx->module_hash_mode                = MODULE_DEFAULT;
+  module_ctx->module_hash_category            = module_hash_category;
+  module_ctx->module_hash_name                = module_hash_name;
+  module_ctx->module_hashes_count_min         = MODULE_DEFAULT;
+  module_ctx->module_hashes_count_max         = MODULE_DEFAULT;
+  module_ctx->module_hlfmt_disable            = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_size    = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_init    = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_term    = MODULE_DEFAULT;
+  module_ctx->module_hook12                   = MODULE_DEFAULT;
+  module_ctx->module_hook23                   = MODULE_DEFAULT;
+  module_ctx->module_hook_salt_size           = MODULE_DEFAULT;
+  module_ctx->module_hook_size                = MODULE_DEFAULT;
+  module_ctx->module_jit_build_options        = MODULE_DEFAULT;
+  module_ctx->module_jit_cache_disable        = MODULE_DEFAULT;
+  module_ctx->module_kernel_accel_max         = MODULE_DEFAULT;
+  module_ctx->module_kernel_accel_min         = MODULE_DEFAULT;
+  module_ctx->module_kernel_loops_max         = MODULE_DEFAULT;
+  module_ctx->module_kernel_loops_min         = MODULE_DEFAULT;
+  module_ctx->module_kernel_threads_max       = MODULE_DEFAULT;
+  module_ctx->module_kernel_threads_min       = MODULE_DEFAULT;
+  module_ctx->module_kern_type                = module_kern_type;
+  module_ctx->module_kern_type_dynamic        = MODULE_DEFAULT;
+  module_ctx->module_opti_type                = module_opti_type;
+  module_ctx->module_opts_type                = module_opts_type;
+  module_ctx->module_outfile_check_disable    = MODULE_DEFAULT;
+  module_ctx->module_outfile_check_nocomp     = MODULE_DEFAULT;
+  module_ctx->module_potfile_custom_check     = MODULE_DEFAULT;
+  module_ctx->module_potfile_disable          = MODULE_DEFAULT;
+  module_ctx->module_potfile_keep_all_hashes  = MODULE_DEFAULT;
+  module_ctx->module_pwdump_column            = MODULE_DEFAULT;
+  module_ctx->module_pw_max                   = module_pw_max;
+  module_ctx->module_pw_min                   = MODULE_DEFAULT;
+  module_ctx->module_salt_max                 = MODULE_DEFAULT;
+  module_ctx->module_salt_min                 = MODULE_DEFAULT;
+  module_ctx->module_salt_type                = module_salt_type;
+  module_ctx->module_separator                = MODULE_DEFAULT;
+  module_ctx->module_st_hash                  = module_st_hash;
+  module_ctx->module_st_pass                  = module_st_pass;
+  module_ctx->module_tmp_size                 = module_tmp_size;
+  module_ctx->module_unstable_warning         = MODULE_DEFAULT;
+  module_ctx->module_warmup_disable           = MODULE_DEFAULT;
+}
diff --git a/src/modules/module_99999.c b/src/modules/module_99999.c
index 0611c7ef6..2759b9ccb 100644
--- a/src/modules/module_99999.c
+++ b/src/modules/module_99999.c
@@ -130,6 +130,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
   module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
   module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_deprecated_notice        = MODULE_DEFAULT;
   module_ctx->module_dgst_pos0                = module_dgst_pos0;
   module_ctx->module_dgst_pos1                = module_dgst_pos1;
   module_ctx->module_dgst_pos2                = module_dgst_pos2;
@@ -139,6 +140,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_esalt_size               = MODULE_DEFAULT;
   module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
   module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_extra_tuningdb_block     = MODULE_DEFAULT;
   module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
   module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
   module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
diff --git a/src/mpsp.c b/src/mpsp.c
index 67168e189..d096b97b5 100644
--- a/src/mpsp.c
+++ b/src/mpsp.c
@@ -16,7 +16,7 @@
 #include "ext_lzma.h"
 #include "mpsp.h"
 
-static const char *DEF_MASK = "?1?2?2?2?2?2?2?3?3?3?3?d?d?d?d";
+static const char *const DEF_MASK = "?1?2?2?2?2?2?2?3?3?3?3?d?d?d?d";
 
 #define MAX_MFS 5 // 4*charset, 1*mask
 
diff --git a/src/rp.c b/src/rp.c
index 019b6333c..a2954acbf 100644
--- a/src/rp.c
+++ b/src/rp.c
@@ -71,7 +71,8 @@ static const char grp_op_chr_chr[] =
 static const char grp_op_pos_chr[] =
 {
   RULE_OP_MANGLE_INSERT,
-  RULE_OP_MANGLE_OVERSTRIKE
+  RULE_OP_MANGLE_OVERSTRIKE,
+  RULE_OP_MANGLE_TOGGLE_AT_SEP
 };
 
 static const char grp_op_pos_pos0[] =
@@ -140,7 +141,7 @@ int generate_random_rule (char rule_buf[RP_RULE_SIZE], const u32 rp_gen_func_min
     u32 p1 = 0;
     u32 p2 = 0;
 
-    switch ((char) get_random_num (0, 9))
+    switch ((char) get_random_num (0, 8))
     {
       case 0:
         r = get_random_num (0, sizeof (grp_op_nop));
@@ -444,12 +445,18 @@ int cpu_rule_to_kernel_rule (char *rule_buf, u32 rule_len, kernel_rule_t *rule)
         break;
 
       case RULE_OP_MANGLE_TITLE:
-        SET_NAME (rule, rule_buf[rule_pos]);
+        SET_NAME    (rule, rule_buf[rule_pos]);
         break;
 
       case RULE_OP_MANGLE_TITLE_SEP:
-        SET_NAME (rule, rule_buf[rule_pos]);
-        SET_P0   (rule, rule_buf[rule_pos]);
+        SET_NAME    (rule, rule_buf[rule_pos]);
+        SET_P0      (rule, rule_buf[rule_pos]);
+        break;
+
+      case RULE_OP_MANGLE_TOGGLE_AT_SEP:
+        SET_NAME    (rule, rule_buf[rule_pos]);
+        SET_P0_CONV (rule, rule_buf[rule_pos]);
+        SET_P1      (rule, rule_buf[rule_pos]);
         break;
 
       default:
@@ -675,6 +682,12 @@ int kernel_rule_to_cpu_rule (char *rule_buf, kernel_rule_t *rule)
         GET_P0 (rule);
         break;
 
+      case RULE_OP_MANGLE_TOGGLE_AT_SEP:
+        rule_buf[rule_pos] = rule_cmd;
+        GET_P0_CONV (rule);
+        GET_P1      (rule);
+        break;
+
       case 0:
         if (rule_pos == 0) return -1;
         return rule_pos - 1;
diff --git a/src/rp_cpu.c b/src/rp_cpu.c
index b9dc23e69..0fd6265af 100644
--- a/src/rp_cpu.c
+++ b/src/rp_cpu.c
@@ -45,6 +45,41 @@ static void MANGLE_SWITCH (char *arr, const int l, const int r)
   arr[l] = c;
 }
 
+static int mangle_toggle_at_sep (char arr[RP_PASSWORD_SIZE], int arr_len, char c, int upos)
+{
+  int toggle_next = 0;
+
+  int occurrence = 0;
+
+  int pos;
+
+  for (pos = 0; pos < arr_len; pos++)
+  {
+    if (arr[pos] == c)
+    {
+      if (occurrence == upos)
+      {
+        toggle_next = 1;
+      }
+      else
+      {
+        occurrence++;
+      }
+
+      continue;
+    }
+
+    if (toggle_next == 1)
+    {
+      MANGLE_TOGGLE_AT (arr, pos);
+
+      break;
+    }
+  }
+
+  return (arr_len);
+}
+
 static int mangle_lrest (char arr[RP_PASSWORD_SIZE], int arr_len)
 {
   int pos;
@@ -561,6 +596,13 @@ int _old_apply_rule (const char *rule, int rule_len, char in[RP_PASSWORD_SIZE],
         if (upos < out_len) MANGLE_TOGGLE_AT (out, upos);
         break;
 
+      case RULE_OP_MANGLE_TOGGLE_AT_SEP:
+        NEXT_RULEPOS (rule_pos);
+        NEXT_RPTOI (rule_new, rule_pos, upos);
+        NEXT_RULEPOS (rule_pos);
+        out_len = mangle_toggle_at_sep (out, out_len, rule_new[rule_pos], upos);
+        break;
+
       case RULE_OP_MANGLE_REVERSE:
         out_len = mangle_reverse (out, out_len);
         break;
diff --git a/src/selftest.c b/src/selftest.c
index 4f8dc1092..71b4e68b2 100644
--- a/src/selftest.c
+++ b/src/selftest.c
@@ -72,6 +72,10 @@ static int selftest (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
     tmp.pw_len = (u32) tmp_len;
   }
 
+  pw_t pw;
+  pw_t comb;
+  bf_t bf;
+
   u32 highest_pw_len = 0;
 
   if (user_options->slow_candidates == true)
@@ -81,8 +85,6 @@ static int selftest (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
       device_param->kernel_params_buf32[30] = 1;
     }
 
-    pw_t pw;
-
     memset (&pw, 0, sizeof (pw));
 
     char *pw_ptr = (char *) &pw.i;
@@ -95,17 +97,17 @@ static int selftest (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
 
     if (device_param->is_cuda == true)
     {
-      if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_pws_buf, &pw, 1 * sizeof (pw_t)) == -1) return -1;
+      if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_pws_buf, &pw, 1 * sizeof (pw_t), device_param->cuda_stream) == -1) return -1;
     }
 
     if (device_param->is_hip == true)
     {
-      if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_pws_buf, &pw, 1 * sizeof (pw_t)) == -1) return -1;
+      if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_pws_buf, &pw, 1 * sizeof (pw_t), device_param->hip_stream) == -1) return -1;
     }
 
     if (device_param->is_opencl == true)
     {
-      if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_buf, CL_TRUE, 0, 1 * sizeof (pw_t), &pw, 0, NULL, NULL) == -1) return -1;
+      if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_buf, CL_FALSE, 0, 1 * sizeof (pw_t), &pw, 0, NULL, NULL) == -1) return -1;
     }
   }
   else
@@ -116,8 +118,6 @@ static int selftest (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
       {
         device_param->kernel_params_buf32[30] = 1;
 
-        pw_t pw;
-
         memset (&pw, 0, sizeof (pw));
 
         char *pw_ptr = (char *) &pw.i;
@@ -135,17 +135,17 @@ static int selftest (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
 
         if (device_param->is_cuda == true)
         {
-          if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_pws_buf, &pw, 1 * sizeof (pw_t)) == -1) return -1;
+          if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_pws_buf, &pw, 1 * sizeof (pw_t), device_param->cuda_stream) == -1) return -1;
         }
 
         if (device_param->is_hip == true)
         {
-          if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_pws_buf, &pw, 1 * sizeof (pw_t)) == -1) return -1;
+          if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_pws_buf, &pw, 1 * sizeof (pw_t), device_param->hip_stream) == -1) return -1;
         }
 
         if (device_param->is_opencl == true)
         {
-          if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_buf, CL_TRUE, 0, 1 * sizeof (pw_t), &pw, 0, NULL, NULL) == -1) return -1;
+          if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_buf, CL_FALSE, 0, 1 * sizeof (pw_t), &pw, 0, NULL, NULL) == -1) return -1;
         }
       }
       else if (user_options_extra->attack_kern == ATTACK_KERN_COMBI)
@@ -153,8 +153,6 @@ static int selftest (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
         device_param->kernel_params_buf32[30] = 1;
         device_param->kernel_params_buf32[33] = COMBINATOR_MODE_BASE_LEFT;
 
-        pw_t pw;
-
         memset (&pw, 0, sizeof (pw));
 
         char *pw_ptr = (char *) &pw.i;
@@ -170,8 +168,6 @@ static int selftest (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
           uppercase ((u8 *) pw_ptr, pw.pw_len);
         }
 
-        pw_t comb;
-
         memset (&comb, 0, sizeof (comb));
 
         char *comb_ptr = (char *) &comb.i;
@@ -202,23 +198,23 @@ static int selftest (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
 
         if (device_param->is_cuda == true)
         {
-          if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_combs_c, &comb, 1 * sizeof (pw_t)) == -1) return -1;
+          if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_combs_c, &comb, 1 * sizeof (pw_t), device_param->cuda_stream) == -1) return -1;
 
-          if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_pws_buf, &pw, 1 * sizeof (pw_t)) == -1) return -1;
+          if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_pws_buf, &pw, 1 * sizeof (pw_t), device_param->cuda_stream) == -1) return -1;
         }
 
         if (device_param->is_hip == true)
         {
-          if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_combs_c, &comb, 1 * sizeof (pw_t)) == -1) return -1;
+          if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_combs_c, &comb, 1 * sizeof (pw_t), device_param->hip_stream) == -1) return -1;
 
-          if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_pws_buf, &pw, 1 * sizeof (pw_t)) == -1) return -1;
+          if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_pws_buf, &pw, 1 * sizeof (pw_t), device_param->hip_stream) == -1) return -1;
         }
 
         if (device_param->is_opencl == true)
         {
-          if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_combs_c, CL_TRUE, 0, 1 * sizeof (pw_t), &comb, 0, NULL, NULL) == -1) return -1;
+          if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_combs_c, CL_FALSE, 0, 1 * sizeof (pw_t), &comb, 0, NULL, NULL) == -1) return -1;
 
-          if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_buf, CL_TRUE, 0, 1 * sizeof (pw_t), &pw, 0, NULL, NULL) == -1) return -1;
+          if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_buf, CL_FALSE, 0, 1 * sizeof (pw_t), &pw, 0, NULL, NULL) == -1) return -1;
         }
       }
       else if (user_options_extra->attack_kern == ATTACK_KERN_BF)
@@ -227,8 +223,6 @@ static int selftest (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
 
         if (hashconfig->opts_type & OPTS_TYPE_TM_KERNEL)
         {
-          pw_t pw;
-
           memset (&pw, 0, sizeof (pw));
 
           char *pw_ptr = (char *) &pw.i;
@@ -246,23 +240,21 @@ static int selftest (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
 
           if (device_param->is_cuda == true)
           {
-            if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_pws_buf, &pw, 1 * sizeof (pw_t)) == -1) return -1;
+            if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_pws_buf, &pw, 1 * sizeof (pw_t), device_param->cuda_stream) == -1) return -1;
           }
 
           if (device_param->is_hip == true)
           {
-            if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_pws_buf, &pw, 1 * sizeof (pw_t)) == -1) return -1;
+            if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_pws_buf, &pw, 1 * sizeof (pw_t), device_param->hip_stream) == -1) return -1;
           }
 
           if (device_param->is_opencl == true)
           {
-            if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_buf, CL_TRUE, 0, 1 * sizeof (pw_t), &pw, 0, NULL, NULL) == -1) return -1;
+            if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_buf, CL_FALSE, 0, 1 * sizeof (pw_t), &pw, 0, NULL, NULL) == -1) return -1;
           }
         }
         else
         {
-          bf_t bf;
-
           memset (&bf, 0, sizeof (bf));
 
           char *bf_ptr = (char *) &bf.i;
@@ -302,21 +294,19 @@ static int selftest (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
 
           if (device_param->is_cuda == true)
           {
-            if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_bfs_c, &bf, 1 * sizeof (bf_t)) == -1) return -1;
+            if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_bfs_c, &bf, 1 * sizeof (bf_t), device_param->cuda_stream) == -1) return -1;
           }
 
           if (device_param->is_hip == true)
           {
-            if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_bfs_c, &bf, 1 * sizeof (bf_t)) == -1) return -1;
+            if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_bfs_c, &bf, 1 * sizeof (bf_t), device_param->hip_stream) == -1) return -1;
           }
 
           if (device_param->is_opencl == true)
           {
-            if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_bfs_c, CL_TRUE, 0, 1 * sizeof (bf_t), &bf, 0, NULL, NULL) == -1) return -1;
+            if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_bfs_c, CL_FALSE, 0, 1 * sizeof (bf_t), &bf, 0, NULL, NULL) == -1) return -1;
           }
 
-          pw_t pw;
-
           memset (&pw, 0, sizeof (pw));
 
           char *pw_ptr = (char *) &pw.i;
@@ -403,17 +393,17 @@ static int selftest (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
 
           if (device_param->is_cuda == true)
           {
-            if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_pws_buf, &pw, 1 * sizeof (pw_t)) == -1) return -1;
+            if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_pws_buf, &pw, 1 * sizeof (pw_t), device_param->cuda_stream) == -1) return -1;
           }
 
           if (device_param->is_hip == true)
           {
-            if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_pws_buf, &pw, 1 * sizeof (pw_t)) == -1) return -1;
+            if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_pws_buf, &pw, 1 * sizeof (pw_t), device_param->hip_stream) == -1) return -1;
           }
 
           if (device_param->is_opencl == true)
           {
-            if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_buf, CL_TRUE, 0, 1 * sizeof (pw_t), &pw, 0, NULL, NULL) == -1) return -1;
+            if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_buf, CL_FALSE, 0, 1 * sizeof (pw_t), &pw, 0, NULL, NULL) == -1) return -1;
           }
 
           highest_pw_len = pw.pw_len;
@@ -422,8 +412,6 @@ static int selftest (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
     }
     else
     {
-      pw_t pw;
-
       memset (&pw, 0, sizeof (pw));
 
       char *pw_ptr = (char *) &pw.i;
@@ -436,23 +424,27 @@ static int selftest (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
 
       if (device_param->is_cuda == true)
       {
-        if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_pws_buf, &pw, 1 * sizeof (pw_t)) == -1) return -1;
+        if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_pws_buf, &pw, 1 * sizeof (pw_t), device_param->cuda_stream) == -1) return -1;
       }
 
       if (device_param->is_hip == true)
       {
-        if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_pws_buf, &pw, 1 * sizeof (pw_t)) == -1) return -1;
+        if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_pws_buf, &pw, 1 * sizeof (pw_t), device_param->hip_stream) == -1) return -1;
       }
 
       if (device_param->is_opencl == true)
       {
-        if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_buf, CL_TRUE, 0, 1 * sizeof (pw_t), &pw, 0, NULL, NULL) == -1) return -1;
+        if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_pws_buf, CL_FALSE, 0, 1 * sizeof (pw_t), &pw, 0, NULL, NULL) == -1) return -1;
       }
     }
   }
 
   // main : run the kernel
 
+  const u32 kernel_threads_sav = device_param->kernel_threads;
+
+  device_param->kernel_threads = device_param->kernel_threads_min;
+
   const double spin_damp_sav = device_param->spin_damp;
 
   device_param->spin_damp = 0;
@@ -509,16 +501,21 @@ static int selftest (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
 
       if (device_param->is_cuda == true)
       {
-        if (hc_cuMemcpyDtoH (hashcat_ctx, device_param->hooks_buf, device_param->cuda_d_hooks, device_param->size_hooks) == -1) return -1;
+        if (hc_cuMemcpyDtoHAsync (hashcat_ctx, device_param->hooks_buf, device_param->cuda_d_hooks, device_param->size_hooks, device_param->cuda_stream) == -1) return -1;
+
+        if (hc_cuStreamSynchronize (hashcat_ctx, device_param->cuda_stream) == -1) return -1;
       }
 
       if (device_param->is_hip == true)
       {
-        if (hc_hipMemcpyDtoH (hashcat_ctx, device_param->hooks_buf, device_param->hip_d_hooks, device_param->size_hooks) == -1) return -1;
+        if (hc_hipMemcpyDtoHAsync (hashcat_ctx, device_param->hooks_buf, device_param->hip_d_hooks, device_param->size_hooks, device_param->hip_stream) == -1) return -1;
+
+        if (hc_hipStreamSynchronize (hashcat_ctx, device_param->hip_stream) == -1) return -1;
       }
 
       if (device_param->is_opencl == true)
       {
+        /* blocking */
         if (hc_clEnqueueReadBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_hooks, CL_TRUE, 0, device_param->size_hooks, device_param->hooks_buf, 0, NULL, NULL) == -1) return -1;
       }
 
@@ -526,17 +523,17 @@ static int selftest (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
 
       if (device_param->is_cuda == true)
       {
-        if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_hooks, device_param->hooks_buf, device_param->size_hooks) == -1) return -1;
+        if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_hooks, device_param->hooks_buf, device_param->size_hooks, device_param->cuda_stream) == -1) return -1;
       }
 
       if (device_param->is_hip == true)
       {
-        if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_hooks, device_param->hooks_buf, device_param->size_hooks) == -1) return -1;
+        if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_hooks, device_param->hooks_buf, device_param->size_hooks, device_param->hip_stream) == -1) return -1;
       }
 
       if (device_param->is_opencl == true)
       {
-        if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_hooks, CL_TRUE, 0, device_param->size_hooks, device_param->hooks_buf, 0, NULL, NULL) == -1) return -1;
+        if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_hooks, CL_FALSE, 0, device_param->size_hooks, device_param->hooks_buf, 0, NULL, NULL) == -1) return -1;
       }
     }
 
@@ -582,16 +579,21 @@ static int selftest (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
 
         if (device_param->is_cuda == true)
         {
-          if (hc_cuMemcpyDtoH (hashcat_ctx, device_param->hooks_buf, device_param->cuda_d_hooks, device_param->size_hooks) == -1) return -1;
+          if (hc_cuMemcpyDtoHAsync (hashcat_ctx, device_param->hooks_buf, device_param->cuda_d_hooks, device_param->size_hooks, device_param->cuda_stream) == -1) return -1;
+
+          if (hc_cuStreamSynchronize (hashcat_ctx, device_param->cuda_stream) == -1) return -1;
         }
 
         if (device_param->is_hip == true)
         {
-          if (hc_hipMemcpyDtoH (hashcat_ctx, device_param->hooks_buf, device_param->hip_d_hooks, device_param->size_hooks) == -1) return -1;
+          if (hc_hipMemcpyDtoHAsync (hashcat_ctx, device_param->hooks_buf, device_param->hip_d_hooks, device_param->size_hooks, device_param->hip_stream) == -1) return -1;
+
+          if (hc_hipStreamSynchronize (hashcat_ctx, device_param->hip_stream) == -1) return -1;
         }
 
         if (device_param->is_opencl == true)
         {
+          /* blocking */
           if (hc_clEnqueueReadBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_hooks, CL_TRUE, 0, device_param->size_hooks, device_param->hooks_buf, 0, NULL, NULL) == -1) return -1;
         }
 
@@ -599,17 +601,17 @@ static int selftest (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
 
         if (device_param->is_cuda == true)
         {
-          if (hc_cuMemcpyHtoD (hashcat_ctx, device_param->cuda_d_hooks, device_param->hooks_buf, device_param->size_hooks) == -1) return -1;
+          if (hc_cuMemcpyHtoDAsync (hashcat_ctx, device_param->cuda_d_hooks, device_param->hooks_buf, device_param->size_hooks, device_param->cuda_stream) == -1) return -1;
         }
 
         if (device_param->is_hip == true)
         {
-          if (hc_hipMemcpyHtoD (hashcat_ctx, device_param->hip_d_hooks, device_param->hooks_buf, device_param->size_hooks) == -1) return -1;
+          if (hc_hipMemcpyHtoDAsync (hashcat_ctx, device_param->hip_d_hooks, device_param->hooks_buf, device_param->size_hooks, device_param->hip_stream) == -1) return -1;
         }
 
         if (device_param->is_opencl == true)
         {
-          if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_hooks, CL_TRUE, 0, device_param->size_hooks, device_param->hooks_buf, 0, NULL, NULL) == -1) return -1;
+          if (hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_hooks, CL_FALSE, 0, device_param->size_hooks, device_param->hooks_buf, 0, NULL, NULL) == -1) return -1;
         }
       }
     }
@@ -677,23 +679,33 @@ static int selftest (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
 
   device_param->spin_damp = spin_damp_sav;
 
+  device_param->kernel_threads = kernel_threads_sav;
+
   // check : check if cracked
 
   u32 num_cracked = 0;
 
+  cl_event opencl_event;
+
   if (device_param->is_cuda == true)
   {
-    if (hc_cuMemcpyDtoH (hashcat_ctx, &num_cracked, device_param->cuda_d_result, sizeof (u32)) == -1) return -1;
+    if (hc_cuMemcpyDtoHAsync (hashcat_ctx, &num_cracked, device_param->cuda_d_result, sizeof (u32), device_param->cuda_stream) == -1) return -1;
+
+    if (hc_cuEventRecord (hashcat_ctx, device_param->cuda_event3, device_param->cuda_stream) == -1) return -1;
   }
 
   if (device_param->is_hip == true)
   {
-    if (hc_hipMemcpyDtoH (hashcat_ctx, &num_cracked, device_param->hip_d_result, sizeof (u32)) == -1) return -1;
+    if (hc_hipMemcpyDtoHAsync (hashcat_ctx, &num_cracked, device_param->hip_d_result, sizeof (u32), device_param->hip_stream) == -1) return -1;
+
+    if (hc_hipEventRecord (hashcat_ctx, device_param->hip_event3, device_param->hip_stream) == -1) return -1;
   }
 
   if (device_param->is_opencl == true)
   {
-    if (hc_clEnqueueReadBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_result, CL_TRUE, 0, sizeof (u32), &num_cracked, 0, NULL, NULL) == -1) return -1;
+    if (hc_clEnqueueReadBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_result, CL_FALSE, 0, sizeof (u32), &num_cracked, 0, NULL, &opencl_event) == -1) return -1;
+
+    if (hc_clFlush (hashcat_ctx, device_param->opencl_command_queue) == -1) return -1;
   }
 
   // finish : cleanup and restore
@@ -712,12 +724,12 @@ static int selftest (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
     device_param->kernel_params[17] = &device_param->cuda_d_salt_bufs;
     device_param->kernel_params[18] = &device_param->cuda_d_esalt_bufs;
 
-    if (run_cuda_kernel_bzero   (hashcat_ctx, device_param, device_param->cuda_d_pws_buf,         device_param->size_pws)     == -1) return -1;
-    if (run_cuda_kernel_bzero   (hashcat_ctx, device_param, device_param->cuda_d_tmps,            device_param->size_tmps)    == -1) return -1;
-    if (run_cuda_kernel_bzero   (hashcat_ctx, device_param, device_param->cuda_d_hooks,           device_param->size_hooks)   == -1) return -1;
-    if (run_cuda_kernel_bzero   (hashcat_ctx, device_param, device_param->cuda_d_plain_bufs,      device_param->size_plains)  == -1) return -1;
-    if (run_cuda_kernel_bzero   (hashcat_ctx, device_param, device_param->cuda_d_digests_shown,   device_param->size_shown)   == -1) return -1;
-    if (run_cuda_kernel_bzero   (hashcat_ctx, device_param, device_param->cuda_d_result,          device_param->size_results) == -1) return -1;
+    if (run_cuda_kernel_bzero (hashcat_ctx, device_param, device_param->cuda_d_pws_buf,       device_param->size_pws)     == -1) return -1;
+    if (run_cuda_kernel_bzero (hashcat_ctx, device_param, device_param->cuda_d_tmps,          device_param->size_tmps)    == -1) return -1;
+    if (run_cuda_kernel_bzero (hashcat_ctx, device_param, device_param->cuda_d_hooks,         device_param->size_hooks)   == -1) return -1;
+    if (run_cuda_kernel_bzero (hashcat_ctx, device_param, device_param->cuda_d_plain_bufs,    device_param->size_plains)  == -1) return -1;
+    if (run_cuda_kernel_bzero (hashcat_ctx, device_param, device_param->cuda_d_digests_shown, device_param->size_shown)   == -1) return -1;
+    if (run_cuda_kernel_bzero (hashcat_ctx, device_param, device_param->cuda_d_result,        device_param->size_results) == -1) return -1;
   }
 
   if (device_param->is_hip == true)
@@ -726,12 +738,12 @@ static int selftest (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
     device_param->kernel_params[17] = &device_param->hip_d_salt_bufs;
     device_param->kernel_params[18] = &device_param->hip_d_esalt_bufs;
 
-    if (run_hip_kernel_bzero   (hashcat_ctx, device_param, device_param->hip_d_pws_buf,         device_param->size_pws)     == -1) return -1;
-    if (run_hip_kernel_bzero   (hashcat_ctx, device_param, device_param->hip_d_tmps,            device_param->size_tmps)    == -1) return -1;
-    if (run_hip_kernel_bzero   (hashcat_ctx, device_param, device_param->hip_d_hooks,           device_param->size_hooks)   == -1) return -1;
-    if (run_hip_kernel_bzero   (hashcat_ctx, device_param, device_param->hip_d_plain_bufs,      device_param->size_plains)  == -1) return -1;
-    if (run_hip_kernel_bzero   (hashcat_ctx, device_param, device_param->hip_d_digests_shown,   device_param->size_shown)   == -1) return -1;
-    if (run_hip_kernel_bzero   (hashcat_ctx, device_param, device_param->hip_d_result,          device_param->size_results) == -1) return -1;
+    if (run_hip_kernel_bzero (hashcat_ctx, device_param, device_param->hip_d_pws_buf,       device_param->size_pws)     == -1) return -1;
+    if (run_hip_kernel_bzero (hashcat_ctx, device_param, device_param->hip_d_tmps,          device_param->size_tmps)    == -1) return -1;
+    if (run_hip_kernel_bzero (hashcat_ctx, device_param, device_param->hip_d_hooks,         device_param->size_hooks)   == -1) return -1;
+    if (run_hip_kernel_bzero (hashcat_ctx, device_param, device_param->hip_d_plain_bufs,    device_param->size_plains)  == -1) return -1;
+    if (run_hip_kernel_bzero (hashcat_ctx, device_param, device_param->hip_d_digests_shown, device_param->size_shown)   == -1) return -1;
+    if (run_hip_kernel_bzero (hashcat_ctx, device_param, device_param->hip_d_result,        device_param->size_results) == -1) return -1;
   }
 
   if (device_param->is_opencl == true)
@@ -820,8 +832,25 @@ static int selftest (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
     }
   }
 
-  // check return
+  // synchronize and ..
+  if (device_param->is_cuda == true)
+  {
+    if (hc_cuEventSynchronize (hashcat_ctx, device_param->cuda_event3) == -1) return -1;
+  }
 
+  if (device_param->is_hip == true)
+  {
+    if (hc_hipEventSynchronize (hashcat_ctx, device_param->hip_event3) == -1) return -1;
+  }
+
+  if (device_param->is_opencl == true)
+  {
+    if (hc_clWaitForEvents (hashcat_ctx, 1, &opencl_event) == -1) return -1;
+
+    if (hc_clReleaseEvent (hashcat_ctx, opencl_event) == -1) return -1;
+  }
+
+  // check return
   if (num_cracked == 0)
   {
     hc_thread_mutex_lock (status_ctx->mux_display);
@@ -903,13 +932,22 @@ HC_API_CALL void *thread_selftest (void *p)
 
   if (device_param->is_cuda == true)
   {
+    if (hc_cuStreamSynchronize (hashcat_ctx, device_param->cuda_stream) == -1) return NULL;
+
     if (hc_cuCtxPopCurrent (hashcat_ctx, &device_param->cuda_context) == -1) return NULL;
   }
 
   if (device_param->is_hip == true)
   {
+    if (hc_hipStreamSynchronize (hashcat_ctx, device_param->hip_stream) == -1) return NULL;
+
     if (hc_hipCtxPopCurrent (hashcat_ctx, &device_param->hip_context) == -1) return NULL;
   }
+  
+  if (device_param->is_opencl == true)
+  {
+    if (hc_clFinish (hashcat_ctx, device_param->opencl_command_queue) == -1) return NULL;
+  }
 
   return NULL;
 }
diff --git a/src/shared.c b/src/shared.c
index b101cc01f..919948232 100644
--- a/src/shared.c
+++ b/src/shared.c
@@ -15,97 +15,97 @@
 #include <sys/cygwin.h>
 #endif
 
-static const char *PA_000 = "OK";
-static const char *PA_001 = "Ignored due to comment";
-static const char *PA_002 = "Ignored due to zero length";
-static const char *PA_003 = "Line-length exception";
-static const char *PA_004 = "Hash-length exception";
-static const char *PA_005 = "Hash-value exception";
-static const char *PA_006 = "Salt-length exception";
-static const char *PA_007 = "Salt-value exception";
-static const char *PA_008 = "Salt-iteration count exception";
-static const char *PA_009 = "Separator unmatched";
-static const char *PA_010 = "Signature unmatched";
-static const char *PA_011 = "Invalid hccapx file size";
-static const char *PA_012 = "Invalid hccapx eapol size";
-static const char *PA_013 = "Invalid psafe2 filesize";
-static const char *PA_014 = "Invalid psafe3 filesize";
-static const char *PA_015 = "Invalid truecrypt filesize";
-static const char *PA_016 = "Invalid veracrypt filesize";
-static const char *PA_017 = "Invalid SIP directive, only MD5 is supported";
-static const char *PA_018 = "Hash-file exception";
-static const char *PA_019 = "Hash-encoding exception";
-static const char *PA_020 = "Salt-encoding exception";
-static const char *PA_021 = "Invalid LUKS filesize";
-static const char *PA_022 = "Invalid LUKS identifier";
-static const char *PA_023 = "Invalid LUKS version";
-static const char *PA_024 = "Invalid or unsupported LUKS cipher type";
-static const char *PA_025 = "Invalid or unsupported LUKS cipher mode";
-static const char *PA_026 = "Invalid or unsupported LUKS hash type";
-static const char *PA_027 = "Invalid LUKS key size";
-static const char *PA_028 = "Disabled LUKS key detected";
-static const char *PA_029 = "Invalid LUKS key AF stripes count";
-static const char *PA_030 = "Invalid combination of LUKS hash type and cipher type";
-static const char *PA_031 = "Invalid hccapx signature";
-static const char *PA_032 = "Invalid hccapx version";
-static const char *PA_033 = "Invalid hccapx message pair";
-static const char *PA_034 = "Token encoding exception";
-static const char *PA_035 = "Token length exception";
-static const char *PA_036 = "Insufficient entropy exception";
-static const char *PA_037 = "Hash contains unsupported compression type for current mode";
-static const char *PA_038 = "Invalid key size";
-static const char *PA_039 = "Invalid block size";
-static const char *PA_040 = "Invalid or unsupported cipher";
-static const char *PA_041 = "Invalid filesize";
-static const char *PA_042 = "IV length exception";
-static const char *PA_043 = "CT length exception";
-static const char *PA_255 = "Unknown error";
+static const char *const PA_000 = "OK";
+static const char *const PA_001 = "Ignored due to comment";
+static const char *const PA_002 = "Ignored due to zero length";
+static const char *const PA_003 = "Line-length exception";
+static const char *const PA_004 = "Hash-length exception";
+static const char *const PA_005 = "Hash-value exception";
+static const char *const PA_006 = "Salt-length exception";
+static const char *const PA_007 = "Salt-value exception";
+static const char *const PA_008 = "Salt-iteration count exception";
+static const char *const PA_009 = "Separator unmatched";
+static const char *const PA_010 = "Signature unmatched";
+static const char *const PA_011 = "Invalid hccapx file size";
+static const char *const PA_012 = "Invalid hccapx eapol size";
+static const char *const PA_013 = "Invalid psafe2 filesize";
+static const char *const PA_014 = "Invalid psafe3 filesize";
+static const char *const PA_015 = "Invalid truecrypt filesize";
+static const char *const PA_016 = "Invalid veracrypt filesize";
+static const char *const PA_017 = "Invalid SIP directive, only MD5 is supported";
+static const char *const PA_018 = "Hash-file exception";
+static const char *const PA_019 = "Hash-encoding exception";
+static const char *const PA_020 = "Salt-encoding exception";
+static const char *const PA_021 = "Invalid LUKS filesize";
+static const char *const PA_022 = "Invalid LUKS identifier";
+static const char *const PA_023 = "Invalid LUKS version";
+static const char *const PA_024 = "Invalid or unsupported LUKS cipher type";
+static const char *const PA_025 = "Invalid or unsupported LUKS cipher mode";
+static const char *const PA_026 = "Invalid or unsupported LUKS hash type";
+static const char *const PA_027 = "Invalid LUKS key size";
+static const char *const PA_028 = "Disabled LUKS key detected";
+static const char *const PA_029 = "Invalid LUKS key AF stripes count";
+static const char *const PA_030 = "Invalid combination of LUKS hash type and cipher type";
+static const char *const PA_031 = "Invalid hccapx signature";
+static const char *const PA_032 = "Invalid hccapx version";
+static const char *const PA_033 = "Invalid hccapx message pair";
+static const char *const PA_034 = "Token encoding exception";
+static const char *const PA_035 = "Token length exception";
+static const char *const PA_036 = "Insufficient entropy exception";
+static const char *const PA_037 = "Hash contains unsupported compression type for current mode";
+static const char *const PA_038 = "Invalid key size";
+static const char *const PA_039 = "Invalid block size";
+static const char *const PA_040 = "Invalid or unsupported cipher";
+static const char *const PA_041 = "Invalid filesize";
+static const char *const PA_042 = "IV length exception";
+static const char *const PA_043 = "CT length exception";
+static const char *const PA_255 = "Unknown error";
 
-static const char *OPTI_STR_OPTIMIZED_KERNEL     = "Optimized-Kernel";
-static const char *OPTI_STR_ZERO_BYTE            = "Zero-Byte";
-static const char *OPTI_STR_PRECOMPUTE_INIT      = "Precompute-Init";
-static const char *OPTI_STR_MEET_IN_MIDDLE       = "Meet-In-The-Middle";
-static const char *OPTI_STR_EARLY_SKIP           = "Early-Skip";
-static const char *OPTI_STR_NOT_SALTED           = "Not-Salted";
-static const char *OPTI_STR_NOT_ITERATED         = "Not-Iterated";
-static const char *OPTI_STR_PREPENDED_SALT       = "Prepended-Salt";
-static const char *OPTI_STR_APPENDED_SALT        = "Appended-Salt";
-static const char *OPTI_STR_SINGLE_HASH          = "Single-Hash";
-static const char *OPTI_STR_SINGLE_SALT          = "Single-Salt";
-static const char *OPTI_STR_BRUTE_FORCE          = "Brute-Force";
-static const char *OPTI_STR_RAW_HASH             = "Raw-Hash";
-static const char *OPTI_STR_SLOW_HASH_SIMD_INIT  = "Slow-Hash-SIMD-INIT";
-static const char *OPTI_STR_SLOW_HASH_SIMD_LOOP  = "Slow-Hash-SIMD-LOOP";
-static const char *OPTI_STR_SLOW_HASH_SIMD_COMP  = "Slow-Hash-SIMD-COMP";
-static const char *OPTI_STR_USES_BITS_8          = "Uses-8-Bit";
-static const char *OPTI_STR_USES_BITS_16         = "Uses-16-Bit";
-static const char *OPTI_STR_USES_BITS_32         = "Uses-32-Bit";
-static const char *OPTI_STR_USES_BITS_64         = "Uses-64-Bit";
+static const char *const OPTI_STR_OPTIMIZED_KERNEL     = "Optimized-Kernel";
+static const char *const OPTI_STR_ZERO_BYTE            = "Zero-Byte";
+static const char *const OPTI_STR_PRECOMPUTE_INIT      = "Precompute-Init";
+static const char *const OPTI_STR_MEET_IN_MIDDLE       = "Meet-In-The-Middle";
+static const char *const OPTI_STR_EARLY_SKIP           = "Early-Skip";
+static const char *const OPTI_STR_NOT_SALTED           = "Not-Salted";
+static const char *const OPTI_STR_NOT_ITERATED         = "Not-Iterated";
+static const char *const OPTI_STR_PREPENDED_SALT       = "Prepended-Salt";
+static const char *const OPTI_STR_APPENDED_SALT        = "Appended-Salt";
+static const char *const OPTI_STR_SINGLE_HASH          = "Single-Hash";
+static const char *const OPTI_STR_SINGLE_SALT          = "Single-Salt";
+static const char *const OPTI_STR_BRUTE_FORCE          = "Brute-Force";
+static const char *const OPTI_STR_RAW_HASH             = "Raw-Hash";
+static const char *const OPTI_STR_SLOW_HASH_SIMD_INIT  = "Slow-Hash-SIMD-INIT";
+static const char *const OPTI_STR_SLOW_HASH_SIMD_LOOP  = "Slow-Hash-SIMD-LOOP";
+static const char *const OPTI_STR_SLOW_HASH_SIMD_COMP  = "Slow-Hash-SIMD-COMP";
+static const char *const OPTI_STR_USES_BITS_8          = "Uses-8-Bit";
+static const char *const OPTI_STR_USES_BITS_16         = "Uses-16-Bit";
+static const char *const OPTI_STR_USES_BITS_32         = "Uses-32-Bit";
+static const char *const OPTI_STR_USES_BITS_64         = "Uses-64-Bit";
 
-static const char *HASH_CATEGORY_UNDEFINED_STR              = "Undefined";
-static const char *HASH_CATEGORY_RAW_HASH_STR               = "Raw Hash";
-static const char *HASH_CATEGORY_RAW_HASH_SALTED_STR        = "Raw Hash, Salted and/or Iterated";
-static const char *HASH_CATEGORY_RAW_HASH_AUTHENTICATED_STR = "Raw Hash, Authenticated";
-static const char *HASH_CATEGORY_RAW_CIPHER_KPA_STR         = "Raw Cipher, Known-Plaintext attack";
-static const char *HASH_CATEGORY_GENERIC_KDF_STR            = "Generic KDF";
-static const char *HASH_CATEGORY_NETWORK_PROTOCOL_STR       = "Network Protocols";
-static const char *HASH_CATEGORY_FORUM_SOFTWARE_STR         = "Forums, CMS, E-Commerce";
-static const char *HASH_CATEGORY_DATABASE_SERVER_STR        = "Database Server";
-static const char *HASH_CATEGORY_NETWORK_SERVER_STR         = "FTP, HTTP, SMTP, LDAP Server";
-static const char *HASH_CATEGORY_RAW_CHECKSUM_STR           = "Raw Checksum";
-static const char *HASH_CATEGORY_OS_STR                     = "Operating System";
-static const char *HASH_CATEGORY_EAS_STR                    = "Enterprise Application Software (EAS)";
-static const char *HASH_CATEGORY_ARCHIVE_STR                = "Archives";
-static const char *HASH_CATEGORY_FDE_STR                    = "Full-Disk Encryption (FDE)";
-static const char *HASH_CATEGORY_FBE_STR                    = "File-Based Encryption (FBE)";
-static const char *HASH_CATEGORY_DOCUMENTS_STR              = "Documents";
-static const char *HASH_CATEGORY_PASSWORD_MANAGER_STR       = "Password Managers";
-static const char *HASH_CATEGORY_OTP_STR                    = "One-Time Passwords";
-static const char *HASH_CATEGORY_PLAIN_STR                  = "Plaintext";
-static const char *HASH_CATEGORY_FRAMEWORK_STR              = "Framework";
-static const char *HASH_CATEGORY_PRIVATE_KEY_STR            = "Private Key";
-static const char *HASH_CATEGORY_IMS_STR                    = "Instant Messaging Service";
-static const char *HASH_CATEGORY_CRYPTOCURRENCY_WALLET_STR  = "Cryptocurrency Wallet";
+static const char *const HASH_CATEGORY_UNDEFINED_STR              = "Undefined";
+static const char *const HASH_CATEGORY_RAW_HASH_STR               = "Raw Hash";
+static const char *const HASH_CATEGORY_RAW_HASH_SALTED_STR        = "Raw Hash, Salted and/or Iterated";
+static const char *const HASH_CATEGORY_RAW_HASH_AUTHENTICATED_STR = "Raw Hash, Authenticated";
+static const char *const HASH_CATEGORY_RAW_CIPHER_KPA_STR         = "Raw Cipher, Known-Plaintext attack";
+static const char *const HASH_CATEGORY_GENERIC_KDF_STR            = "Generic KDF";
+static const char *const HASH_CATEGORY_NETWORK_PROTOCOL_STR       = "Network Protocols";
+static const char *const HASH_CATEGORY_FORUM_SOFTWARE_STR         = "Forums, CMS, E-Commerce";
+static const char *const HASH_CATEGORY_DATABASE_SERVER_STR        = "Database Server";
+static const char *const HASH_CATEGORY_NETWORK_SERVER_STR         = "FTP, HTTP, SMTP, LDAP Server";
+static const char *const HASH_CATEGORY_RAW_CHECKSUM_STR           = "Raw Checksum";
+static const char *const HASH_CATEGORY_OS_STR                     = "Operating System";
+static const char *const HASH_CATEGORY_EAS_STR                    = "Enterprise Application Software (EAS)";
+static const char *const HASH_CATEGORY_ARCHIVE_STR                = "Archives";
+static const char *const HASH_CATEGORY_FDE_STR                    = "Full-Disk Encryption (FDE)";
+static const char *const HASH_CATEGORY_FBE_STR                    = "File-Based Encryption (FBE)";
+static const char *const HASH_CATEGORY_DOCUMENTS_STR              = "Documents";
+static const char *const HASH_CATEGORY_PASSWORD_MANAGER_STR       = "Password Managers";
+static const char *const HASH_CATEGORY_OTP_STR                    = "One-Time Passwords";
+static const char *const HASH_CATEGORY_PLAIN_STR                  = "Plaintext";
+static const char *const HASH_CATEGORY_FRAMEWORK_STR              = "Framework";
+static const char *const HASH_CATEGORY_PRIVATE_KEY_STR            = "Private Key";
+static const char *const HASH_CATEGORY_IMS_STR                    = "Instant Messaging Service";
+static const char *const HASH_CATEGORY_CRYPTOCURRENCY_WALLET_STR  = "Cryptocurrency Wallet";
 
 int sort_by_string_sized (const void *p1, const void *p2)
 {
@@ -313,6 +313,8 @@ bool hc_path_is_file (const char *path)
 {
   struct stat s;
 
+  memset (&s, 0, sizeof (s));
+
   if (stat (path, &s) == -1) return false;
 
   if (S_ISREG (s.st_mode)) return true;
@@ -324,6 +326,8 @@ bool hc_path_is_directory (const char *path)
 {
   struct stat s;
 
+  memset (&s, 0, sizeof (s));
+
   if (stat (path, &s) == -1) return false;
 
   if (S_ISDIR (s.st_mode)) return true;
@@ -335,6 +339,8 @@ bool hc_path_is_empty (const char *path)
 {
   struct stat s;
 
+  memset (&s, 0, sizeof (s));
+
   if (stat (path, &s) == -1) return false;
 
   if (s.st_size == 0) return true;
@@ -678,6 +684,9 @@ bool hc_same_files (char *file1, char *file2)
     struct stat tmpstat_file1;
     struct stat tmpstat_file2;
 
+    memset (&tmpstat_file1, 0, sizeof (tmpstat_file1));
+    memset (&tmpstat_file2, 0, sizeof (tmpstat_file2));
+
     int do_check = 0;
 
     HCFILE fp;
@@ -1062,7 +1071,7 @@ static int rounds_count_length (const char *input_buf, const int input_len)
 {
   if (input_len >= 9) // 9 is minimum because of "rounds=X$"
   {
-    static const char *rounds = "rounds=";
+    static const char *const rounds = "rounds=";
 
     if (memcmp (input_buf, rounds, 7) == 0)
     {
@@ -1181,7 +1190,7 @@ int input_tokenizer (const u8 *input_buf, const int input_len, token_t *token)
 
       for (int signature_idx = 0; signature_idx < token->signatures_cnt; signature_idx++)
       {
-        if (memcmp (token->buf[token_idx], token->signatures_buf[signature_idx], token->len[token_idx]) == 0) matched = true;
+        if (strncmp ((char *) token->buf[token_idx], token->signatures_buf[signature_idx], token->len[token_idx]) == 0) matched = true;
       }
 
       if (matched == false) return (PARSER_SIGNATURE_UNMATCHED);
diff --git a/src/status.c b/src/status.c
index 6b4084b59..c26e43e51 100644
--- a/src/status.c
+++ b/src/status.c
@@ -18,29 +18,29 @@
 #include "shared.h"
 #include "status.h"
 
-static const char *ST_0000 = "Initializing";
-static const char *ST_0001 = "Autotuning";
-static const char *ST_0002 = "Selftest";
-static const char *ST_0003 = "Running";
-static const char *ST_0004 = "Paused";
-static const char *ST_0005 = "Exhausted";
-static const char *ST_0006 = "Cracked";
-static const char *ST_0007 = "Aborted";
-static const char *ST_0008 = "Quit";
-static const char *ST_0009 = "Bypass";
-static const char *ST_0010 = "Aborted (Checkpoint)";
-static const char *ST_0011 = "Aborted (Runtime)";
-static const char *ST_0012 = "Running (Checkpoint Quit requested)";
-static const char *ST_0013 = "Error";
-static const char *ST_0014 = "Aborted (Finish)";
-static const char *ST_0015 = "Running (Quit after attack requested)";
-static const char *ST_0016 = "Autodetect";
-static const char *ST_9999 = "Unknown! Bug!";
+static const char *const  ST_0000 = "Initializing";
+static const char *const  ST_0001 = "Autotuning";
+static const char *const  ST_0002 = "Selftest";
+static const char *const  ST_0003 = "Running";
+static const char *const  ST_0004 = "Paused";
+static const char *const  ST_0005 = "Exhausted";
+static const char *const  ST_0006 = "Cracked";
+static const char *const  ST_0007 = "Aborted";
+static const char *const  ST_0008 = "Quit";
+static const char *const  ST_0009 = "Bypass";
+static const char *const  ST_0010 = "Aborted (Checkpoint)";
+static const char *const  ST_0011 = "Aborted (Runtime)";
+static const char *const  ST_0012 = "Running (Checkpoint Quit requested)";
+static const char *const  ST_0013 = "Error";
+static const char *const  ST_0014 = "Aborted (Finish)";
+static const char *const  ST_0015 = "Running (Quit after attack requested)";
+static const char *const  ST_0016 = "Autodetect";
+static const char *const  ST_9999 = "Unknown! Bug!";
 
 static const char UNITS[7] = { ' ', 'k', 'M', 'G', 'T', 'P', 'E' };
 
-static const char *ETA_ABSOLUTE_MAX_EXCEEDED = "Next Big Bang"; // in honor of ighashgpu
-static const char *ETA_RELATIVE_MAX_EXCEEDED = "> 10 years";
+static const char *const  ETA_ABSOLUTE_MAX_EXCEEDED = "Next Big Bang"; // in honor of ighashgpu
+static const char *const  ETA_RELATIVE_MAX_EXCEEDED = "> 10 years";
 
 static char *status_get_rules_file (const hashcat_ctx_t *hashcat_ctx)
 {
diff --git a/src/terminal.c b/src/terminal.c
index 54bffed10..1d3b2dd6e 100644
--- a/src/terminal.c
+++ b/src/terminal.c
@@ -19,8 +19,8 @@
 
 static const size_t TERMINAL_LINE_LENGTH = 79;
 
-static const char *PROMPT_ACTIVE = "[s]tatus [p]ause [b]ypass [c]heckpoint [f]inish [q]uit => ";
-static const char *PROMPT_PAUSED = "[s]tatus [r]esume [b]ypass [c]heckpoint [f]inish [q]uit => ";
+static const char *const PROMPT_ACTIVE = "[s]tatus [p]ause [b]ypass [c]heckpoint [f]inish [q]uit => ";
+static const char *const PROMPT_PAUSED = "[s]tatus [r]esume [b]ypass [c]heckpoint [f]inish [q]uit => ";
 
 void welcome_screen (hashcat_ctx_t *hashcat_ctx, const char *version_tag)
 {
@@ -117,7 +117,7 @@ void goodbye_screen (hashcat_ctx_t *hashcat_ctx, const time_t proc_start, const
   if (user_options->identify    == true) return;
 
   char start_buf[32]; memset (start_buf, 0, sizeof (start_buf));
-  char stop_buf[32];  memset (start_buf, 0, sizeof (stop_buf));
+  char stop_buf[32];  memset (stop_buf,  0, sizeof (stop_buf));
 
   event_log_info_nn (hashcat_ctx, "Started: %s", ctime_r (&proc_start, start_buf));
   event_log_info_nn (hashcat_ctx, "Stopped: %s", ctime_r (&proc_stop,  stop_buf));
@@ -623,7 +623,7 @@ void compress_terminal_line_length (char *out_buf, const size_t keep_from_beginn
   *ptr1 = 0;
 }
 
-void hash_info_single (hashcat_ctx_t *hashcat_ctx, user_options_t *user_options)
+void hash_info_single (hashcat_ctx_t *hashcat_ctx, user_options_extra_t *user_options_extra)
 {
   if (hashconfig_init (hashcat_ctx) == 0)
   {
@@ -640,7 +640,7 @@ void hash_info_single (hashcat_ctx_t *hashcat_ctx, user_options_t *user_options)
     if (hashconfig->is_salted == true)
     {
       u32 t = hashconfig->salt_type;
-      char *t_desc = (t == SALT_TYPE_EMBEDDED) ? "Embedded\0" : (t == SALT_TYPE_GENERIC) ? "Generic\0" : "Virtual\0";
+      const char *t_desc = (t == SALT_TYPE_EMBEDDED) ? "Embedded\0" : (t == SALT_TYPE_GENERIC) ? "Generic\0" : "Virtual\0";
       event_log_info (hashcat_ctx, "  Salt.Type...........: %s", t_desc);
       event_log_info (hashcat_ctx, "  Salt.Len.Min........: %d", hashconfig->salt_min);
       event_log_info (hashcat_ctx, "  Salt.Len.Max........: %d", hashconfig->salt_max);
@@ -676,7 +676,7 @@ void hash_info_single (hashcat_ctx_t *hashcat_ctx, user_options_t *user_options)
         event_log_info (hashcat_ctx, "  Example.Hash........: %s", hashconfig->st_hash);
       }
 
-      if (need_hexify ((const u8 *) hashconfig->st_pass, strlen (hashconfig->st_pass), user_options->separator, false))
+      if (need_hexify ((const u8 *) hashconfig->st_pass, strlen (hashconfig->st_pass), user_options_extra->separator, false))
       {
         char *tmp_buf = (char *) hcmalloc (HCBUFSIZ_LARGE);
 
@@ -728,8 +728,9 @@ void hash_info_single (hashcat_ctx_t *hashcat_ctx, user_options_t *user_options)
 
 void hash_info (hashcat_ctx_t *hashcat_ctx)
 {
-  folder_config_t *folder_config = hashcat_ctx->folder_config;
-  user_options_t  *user_options  = hashcat_ctx->user_options;
+  folder_config_t      *folder_config      = hashcat_ctx->folder_config;
+  user_options_t       *user_options       = hashcat_ctx->user_options;
+  user_options_extra_t *user_options_extra = hashcat_ctx->user_options_extra;
 
   event_log_info (hashcat_ctx, "Hash Info:");
   event_log_info (hashcat_ctx, "==========");
@@ -737,7 +738,7 @@ void hash_info (hashcat_ctx_t *hashcat_ctx)
 
   if (user_options->hash_mode_chgd == true)
   {
-    hash_info_single (hashcat_ctx, user_options);
+    hash_info_single (hashcat_ctx, user_options_extra);
   }
   else
   {
@@ -751,7 +752,7 @@ void hash_info (hashcat_ctx_t *hashcat_ctx)
 
       if (hc_path_exist (modulefile) == false) continue;
 
-      hash_info_single (hashcat_ctx, user_options);
+      hash_info_single (hashcat_ctx, user_options_extra);
     }
 
     hcfree (modulefile);
@@ -817,10 +818,22 @@ void backend_info (hashcat_ctx_t *hashcat_ctx)
     event_log_info (hashcat_ctx, NULL);
 
     int hip_devices_cnt    = backend_ctx->hip_devices_cnt;
-    int hip_driver_version = backend_ctx->hip_driver_version;
+    int hip_runtimeVersion = backend_ctx->hip_runtimeVersion;
 
-    event_log_info (hashcat_ctx, "HIP.Version.: %d.%d", hip_driver_version / 1000, (hip_driver_version % 100) / 10);
-    event_log_info (hashcat_ctx, NULL);
+    if (hip_runtimeVersion > 1000)
+    {
+      int hip_version_major = (hip_runtimeVersion - 0) / 10000000;
+      int hip_version_minor = (hip_runtimeVersion - (hip_version_major * 10000000)) / 100000;
+      int hip_version_patch = (hip_runtimeVersion - (hip_version_major * 10000000) - (hip_version_minor * 100000));
+
+      event_log_info (hashcat_ctx, "HIP.Version.: %d.%d.%d", hip_version_major, hip_version_minor, hip_version_patch);
+      event_log_info (hashcat_ctx, NULL);
+    }
+    else
+    {
+      event_log_info (hashcat_ctx, "HIP.Version.: %d.%d", hip_runtimeVersion / 100, hip_runtimeVersion % 10);
+      event_log_info (hashcat_ctx, NULL);
+    }
 
     for (int hip_devices_idx = 0; hip_devices_idx < hip_devices_cnt; hip_devices_idx++)
     {
@@ -1014,9 +1027,22 @@ void backend_info_compact (hashcat_ctx_t *hashcat_ctx)
   if (backend_ctx->hip)
   {
     int hip_devices_cnt    = backend_ctx->hip_devices_cnt;
-    int hip_driver_version = backend_ctx->hip_driver_version;
+    int hip_runtimeVersion = backend_ctx->hip_runtimeVersion;
 
-    const size_t len = event_log_info (hashcat_ctx, "HIP API (HIP %d.%d)", hip_driver_version / 1000, (hip_driver_version % 100) / 10);
+    size_t len;
+
+    if (hip_runtimeVersion > 1000)
+    {
+      int hip_version_major = (hip_runtimeVersion - 0) / 10000000;
+      int hip_version_minor = (hip_runtimeVersion - (hip_version_major * 10000000)) / 100000;
+      int hip_version_patch = (hip_runtimeVersion - (hip_version_major * 10000000) - (hip_version_minor * 100000));
+
+      len = event_log_info (hashcat_ctx, "HIP API (HIP %d.%d.%d)", hip_version_major, hip_version_minor, hip_version_patch);
+    }
+    else
+    {
+      len = event_log_info (hashcat_ctx, "HIP API (HIP %d.%d)", hip_runtimeVersion / 100, hip_runtimeVersion % 10);
+    }
 
     char line[HCBUFSIZ_TINY] = { 0 };
 
diff --git a/src/tuningdb.c b/src/tuningdb.c
index f487c128c..d16e7517a 100644
--- a/src/tuningdb.c
+++ b/src/tuningdb.c
@@ -11,7 +11,7 @@
 #include "shared.h"
 #include "tuningdb.h"
 
-static int sort_by_tuning_db_alias (const void *v1, const void *v2)
+int sort_by_tuning_db_alias (const void *v1, const void *v2)
 {
   const tuning_db_alias_t *t1 = (const tuning_db_alias_t *) v1;
   const tuning_db_alias_t *t2 = (const tuning_db_alias_t *) v2;
@@ -23,7 +23,7 @@ static int sort_by_tuning_db_alias (const void *v1, const void *v2)
   return 0;
 }
 
-static int sort_by_tuning_db_entry (const void *v1, const void *v2)
+int sort_by_tuning_db_entry (const void *v1, const void *v2)
 {
   const tuning_db_entry_t *t1 = (const tuning_db_entry_t *) v1;
   const tuning_db_entry_t *t2 = (const tuning_db_entry_t *) v2;
@@ -47,10 +47,9 @@ static int sort_by_tuning_db_entry (const void *v1, const void *v2)
 
 int tuning_db_init (hashcat_ctx_t *hashcat_ctx)
 {
-  folder_config_t       *folder_config      = hashcat_ctx->folder_config;
-  tuning_db_t           *tuning_db          = hashcat_ctx->tuning_db;
-  user_options_t        *user_options       = hashcat_ctx->user_options;
-  user_options_extra_t  *user_options_extra = hashcat_ctx->user_options_extra;
+  folder_config_t *folder_config  = hashcat_ctx->folder_config;
+  tuning_db_t     *tuning_db      = hashcat_ctx->tuning_db;
+  user_options_t  *user_options   = hashcat_ctx->user_options;
 
   tuning_db->enabled = false;
 
@@ -80,18 +79,6 @@ int tuning_db_init (hashcat_ctx_t *hashcat_ctx)
 
   hcfree (tuning_db_file);
 
-  const size_t num_lines = count_lines (&fp);
-
-  // a bit over-allocated
-
-  tuning_db->alias_buf = (tuning_db_alias_t *) hccalloc (num_lines + 1, sizeof (tuning_db_alias_t));
-  tuning_db->alias_cnt = 0;
-
-  tuning_db->entry_buf = (tuning_db_entry_t *) hccalloc (num_lines + 1, sizeof (tuning_db_entry_t));
-  tuning_db->entry_cnt = 0;
-
-  hc_rewind (&fp);
-
   int line_num = 0;
 
   char *buf = (char *) hcmalloc (HCBUFSIZ_LARGE);
@@ -110,168 +97,7 @@ int tuning_db_init (hashcat_ctx_t *hashcat_ctx)
 
     if (line_buf[0] == '#') continue;
 
-    // start processing
-
-    char *token_ptr[7] = { NULL };
-
-    int token_cnt = 0;
-
-    char *saveptr = NULL;
-
-    char *next = strtok_r (line_buf, "\t ", &saveptr);
-
-    token_ptr[token_cnt] = next;
-
-    token_cnt++;
-
-    while ((next = strtok_r ((char *) NULL, "\t ", &saveptr)) != NULL)
-    {
-      token_ptr[token_cnt] = next;
-
-      token_cnt++;
-    }
-
-    if (token_cnt == 2)
-    {
-      char *device_name = token_ptr[0];
-      char *alias_name  = token_ptr[1];
-
-      tuning_db_alias_t *alias = &tuning_db->alias_buf[tuning_db->alias_cnt];
-
-      alias->device_name = hcstrdup (device_name);
-      alias->alias_name  = hcstrdup (alias_name);
-
-      tuning_db->alias_cnt++;
-    }
-    else if (token_cnt == 6)
-    {
-      if ((token_ptr[1][0] != '0') &&
-          (token_ptr[1][0] != '1') &&
-          (token_ptr[1][0] != '3') &&
-          (token_ptr[1][0] != '9') &&
-          (token_ptr[1][0] != '*'))
-      {
-        event_log_warning (hashcat_ctx, "Tuning-db: Invalid attack_mode '%c' in Line '%d'", token_ptr[1][0], line_num);
-
-        continue;
-      }
-
-      if ((token_ptr[3][0] != '1') &&
-          (token_ptr[3][0] != '2') &&
-          (token_ptr[3][0] != '4') &&
-          (token_ptr[3][0] != '8') &&
-          (token_ptr[3][0] != 'N'))
-      {
-        event_log_warning (hashcat_ctx, "Tuning-db: Invalid vector_width '%c' in Line '%d'", token_ptr[3][0], line_num);
-
-        continue;
-      }
-
-      char *device_name = token_ptr[0];
-
-      int attack_mode      = -1;
-      int hash_mode        = -1;
-      int vector_width     = -1;
-      int kernel_accel     = -1;
-      int kernel_loops     = -1;
-
-      if (token_ptr[1][0] != '*') attack_mode   = (int) strtol (token_ptr[1], NULL, 10);
-      if (token_ptr[2][0] != '*') hash_mode     = (int) strtol (token_ptr[2], NULL, 10);
-      if (token_ptr[3][0] != 'N') vector_width  = (int) strtol (token_ptr[3], NULL, 10);
-
-      if (token_ptr[4][0] == 'A')
-      {
-        kernel_accel = 0;
-      }
-      else if (token_ptr[4][0] == 'M')
-      {
-        kernel_accel = 1024;
-      }
-      else if (token_ptr[4][0] == 'N')
-      {
-        kernel_accel = -1;
-      }
-      else
-      {
-        kernel_accel = (int) strtol (token_ptr[4], NULL, 10);
-
-        if ((kernel_accel < 1) || (kernel_accel > 1024))
-        {
-          event_log_warning (hashcat_ctx, "Tuning-db: Invalid kernel_accel '%d' in Line '%d'", kernel_accel, line_num);
-
-          continue;
-        }
-      }
-
-      if (token_ptr[5][0] == 'A')
-      {
-        kernel_loops = 0;
-      }
-      else if (token_ptr[5][0] == 'M')
-      {
-        if (user_options_extra->attack_kern == ATTACK_KERN_STRAIGHT)
-        {
-          kernel_loops = KERNEL_RULES;
-        }
-        else if (user_options_extra->attack_kern == ATTACK_KERN_COMBI)
-        {
-          kernel_loops = KERNEL_COMBS;
-        }
-        else if (user_options_extra->attack_kern == ATTACK_KERN_BF)
-        {
-          kernel_loops = KERNEL_BFS;
-        }
-      }
-      else
-      {
-        kernel_loops = (int) strtol (token_ptr[5], NULL, 10);
-
-        if (kernel_loops < 1)
-        {
-          event_log_warning (hashcat_ctx, "Tuning-db: Invalid kernel_loops '%d' in Line '%d'", kernel_loops, line_num);
-
-          continue;
-        }
-
-        if ((user_options_extra->attack_kern == ATTACK_KERN_STRAIGHT) && (kernel_loops > KERNEL_RULES))
-        {
-          event_log_warning (hashcat_ctx, "Tuning-db: Invalid kernel_loops '%d' in Line '%d'", kernel_loops, line_num);
-
-          continue;
-        }
-
-        if ((user_options_extra->attack_kern == ATTACK_KERN_COMBI) && (kernel_loops > KERNEL_COMBS))
-        {
-          event_log_warning (hashcat_ctx, "Tuning-db: Invalid kernel_loops '%d' in Line '%d'", kernel_loops, line_num);
-
-          continue;
-        }
-
-        if ((user_options_extra->attack_kern == ATTACK_KERN_BF) && (kernel_loops > KERNEL_BFS))
-        {
-          event_log_warning (hashcat_ctx, "Tuning-db: Invalid kernel_loops '%d' in Line '%d'", kernel_loops, line_num);
-
-          continue;
-        }
-      }
-
-      tuning_db_entry_t *entry = &tuning_db->entry_buf[tuning_db->entry_cnt];
-
-      entry->device_name  = hcstrdup (device_name);
-      entry->attack_mode  = attack_mode;
-      entry->hash_mode    = hash_mode;
-      entry->vector_width = vector_width;
-      entry->kernel_accel = kernel_accel;
-      entry->kernel_loops = kernel_loops;
-
-      tuning_db->entry_cnt++;
-    }
-    else
-    {
-      event_log_warning (hashcat_ctx, "Tuning-db: Invalid number of token in Line '%d'", line_num);
-
-      continue;
-    }
+    tuning_db_process_line (hashcat_ctx, line_buf, line_num);
   }
 
   hcfree (buf);
@@ -317,6 +143,209 @@ void tuning_db_destroy (hashcat_ctx_t *hashcat_ctx)
   memset (tuning_db, 0, sizeof (tuning_db_t));
 }
 
+bool tuning_db_process_line (hashcat_ctx_t *hashcat_ctx, const char *line_buf, const int line_num)
+{
+  tuning_db_t           *tuning_db          = hashcat_ctx->tuning_db;
+  user_options_extra_t  *user_options_extra = hashcat_ctx->user_options_extra;
+
+  #define ADD_DB_ENTRIES 1
+
+  if (tuning_db->alias_cnt == tuning_db->alias_alloc)
+  {
+    tuning_db->alias_buf    = (tuning_db_alias_t *) hcrealloc (tuning_db->alias_buf, tuning_db->alias_alloc * sizeof (tuning_db_alias_t), ADD_DB_ENTRIES * sizeof (tuning_db_alias_t));
+    tuning_db->alias_alloc += ADD_DB_ENTRIES;
+  }
+
+  if (tuning_db->entry_cnt == tuning_db->entry_alloc)
+  {
+    tuning_db->entry_buf    = (tuning_db_entry_t *) hcrealloc (tuning_db->entry_buf, tuning_db->entry_alloc * sizeof (tuning_db_entry_t), ADD_DB_ENTRIES * sizeof (tuning_db_entry_t));
+    tuning_db->entry_alloc += ADD_DB_ENTRIES;
+  }
+
+  char *buf = hcstrdup (line_buf);
+
+  char *token_ptr[7] = { NULL };
+
+  int token_cnt = 0;
+
+  char *saveptr = NULL;
+
+  char *next = strtok_r (buf, "\t ", &saveptr);
+
+  token_ptr[token_cnt] = next;
+
+  token_cnt++;
+
+  while ((next = strtok_r ((char *) NULL, "\t ", &saveptr)) != NULL)
+  {
+    token_ptr[token_cnt] = next;
+
+    token_cnt++;
+  }
+
+  if (token_cnt == 2)
+  {
+    char *device_name = token_ptr[0];
+    char *alias_name  = token_ptr[1];
+
+    tuning_db_alias_t *alias = &tuning_db->alias_buf[tuning_db->alias_cnt];
+
+    alias->device_name = hcstrdup (device_name);
+    alias->alias_name  = hcstrdup (alias_name);
+
+    tuning_db->alias_cnt++;
+  }
+  else if (token_cnt == 6)
+  {
+    if ((token_ptr[1][0] != '0') &&
+        (token_ptr[1][0] != '1') &&
+        (token_ptr[1][0] != '3') &&
+        (token_ptr[1][0] != '9') &&
+        (token_ptr[1][0] != '*'))
+    {
+      event_log_warning (hashcat_ctx, "Tuning-db: Invalid attack_mode '%c' in Line '%d'", token_ptr[1][0], line_num);
+
+      hcfree (buf);
+
+      return false;
+    }
+
+    if ((token_ptr[3][0] != '1') &&
+        (token_ptr[3][0] != '2') &&
+        (token_ptr[3][0] != '4') &&
+        (token_ptr[3][0] != '8') &&
+        (token_ptr[3][0] != 'N'))
+    {
+      event_log_warning (hashcat_ctx, "Tuning-db: Invalid vector_width '%c' in Line '%d'", token_ptr[3][0], line_num);
+
+      hcfree (buf);
+
+      return false;
+    }
+
+    char *device_name = token_ptr[0];
+
+    int attack_mode      = -1;
+    int hash_mode        = -1;
+    int vector_width     = -1;
+    int kernel_accel     = -1;
+    int kernel_loops     = -1;
+
+    if (token_ptr[1][0] != '*') attack_mode   = (int) strtol (token_ptr[1], NULL, 10);
+    if (token_ptr[2][0] != '*') hash_mode     = (int) strtol (token_ptr[2], NULL, 10);
+    if (token_ptr[3][0] != 'N') vector_width  = (int) strtol (token_ptr[3], NULL, 10);
+
+    if (token_ptr[4][0] == 'A')
+    {
+      kernel_accel = 0;
+    }
+    else if (token_ptr[4][0] == 'M')
+    {
+      kernel_accel = 1024;
+    }
+    else if (token_ptr[4][0] == 'N')
+    {
+      kernel_accel = -1;
+    }
+    else
+    {
+      kernel_accel = (int) strtol (token_ptr[4], NULL, 10);
+
+      if ((kernel_accel < 1) || (kernel_accel > 1024))
+      {
+        event_log_warning (hashcat_ctx, "Tuning-db: Invalid kernel_accel '%d' in Line '%d'", kernel_accel, line_num);
+
+        hcfree (buf);
+
+        return false;
+      }
+    }
+
+    if (token_ptr[5][0] == 'A')
+    {
+      kernel_loops = 0;
+    }
+    else if (token_ptr[5][0] == 'M')
+    {
+      if (user_options_extra->attack_kern == ATTACK_KERN_STRAIGHT)
+      {
+        kernel_loops = KERNEL_RULES;
+      }
+      else if (user_options_extra->attack_kern == ATTACK_KERN_COMBI)
+      {
+        kernel_loops = KERNEL_COMBS;
+      }
+      else if (user_options_extra->attack_kern == ATTACK_KERN_BF)
+      {
+        kernel_loops = KERNEL_BFS;
+      }
+    }
+    else
+    {
+      kernel_loops = (int) strtol (token_ptr[5], NULL, 10);
+
+      if (kernel_loops < 1)
+      {
+        event_log_warning (hashcat_ctx, "Tuning-db: Invalid kernel_loops '%d' in Line '%d'", kernel_loops, line_num);
+
+        hcfree (buf);
+
+        return false;
+      }
+
+      if ((user_options_extra->attack_kern == ATTACK_KERN_STRAIGHT) && (kernel_loops > KERNEL_RULES))
+      {
+        event_log_warning (hashcat_ctx, "Tuning-db: Invalid kernel_loops '%d' in Line '%d'", kernel_loops, line_num);
+
+        hcfree (buf);
+
+        return false;
+      }
+
+      if ((user_options_extra->attack_kern == ATTACK_KERN_COMBI) && (kernel_loops > KERNEL_COMBS))
+      {
+        event_log_warning (hashcat_ctx, "Tuning-db: Invalid kernel_loops '%d' in Line '%d'", kernel_loops, line_num);
+
+        hcfree (buf);
+
+        return false;
+      }
+
+      if ((user_options_extra->attack_kern == ATTACK_KERN_BF) && (kernel_loops > KERNEL_BFS))
+      {
+        event_log_warning (hashcat_ctx, "Tuning-db: Invalid kernel_loops '%d' in Line '%d'", kernel_loops, line_num);
+
+        hcfree (buf);
+
+        return false;
+      }
+    }
+
+    tuning_db_entry_t *entry = &tuning_db->entry_buf[tuning_db->entry_cnt];
+
+    entry->device_name  = hcstrdup (device_name);
+    entry->attack_mode  = attack_mode;
+    entry->hash_mode    = hash_mode;
+    entry->vector_width = vector_width;
+    entry->kernel_accel = kernel_accel;
+    entry->kernel_loops = kernel_loops;
+
+    tuning_db->entry_cnt++;
+  }
+  else
+  {
+    event_log_warning (hashcat_ctx, "Tuning-db: Invalid number of token in Line '%d'", line_num);
+
+    hcfree (buf);
+
+    return false;
+  }
+
+  hcfree (buf);
+
+  return true;
+}
+
 tuning_db_entry_t *tuning_db_search_real (hashcat_ctx_t *hashcat_ctx, const char *device_name, const cl_device_type device_type, int attack_mode, const int hash_mode)
 {
   tuning_db_t *tuning_db = hashcat_ctx->tuning_db;
@@ -435,13 +464,22 @@ tuning_db_entry_t *tuning_db_search (hashcat_ctx_t *hashcat_ctx, const char *dev
 
   const char *NV_prefix = (const char *) "NVIDIA ";
 
-  if (strlen(device_name) >= strlen(NV_prefix) && memcmp (device_name, NV_prefix, strlen (NV_prefix)) == 0)
+  if (strncmp (device_name, NV_prefix, strlen (NV_prefix)) == 0)
   {
     entry = tuning_db_search_real (hashcat_ctx, device_name + strlen (NV_prefix), device_type, attack_mode, hash_mode);
 
     if (entry) return entry;
   }
 
+  const char *AMD_prefix = (const char *) "AMD ";
+
+  if (strncmp (device_name, AMD_prefix, strlen (AMD_prefix)) == 0)
+  {
+    entry = tuning_db_search_real (hashcat_ctx, device_name + strlen (AMD_prefix), device_type, attack_mode, hash_mode);
+
+    if (entry) return entry;
+  }
+
   entry = tuning_db_search_real (hashcat_ctx, device_name, device_type, attack_mode, hash_mode);
 
   return entry;
diff --git a/src/usage.c b/src/usage.c
index d8d57ebcd..55b61ce51 100644
--- a/src/usage.c
+++ b/src/usage.c
@@ -35,6 +35,7 @@ static const char *const USAGE_BIG_PRE_HASHMODES[] =
   "     --hex-salt                 |      | Assume salt is given in hex                          |",
   "     --hex-wordlist             |      | Assume words in wordlist are given in hex            |",
   "     --force                    |      | Ignore warnings                                      |",
+  "     --deprecated-check-disable |      | Enable deprecated plugins                            |",
   "     --status                   |      | Enable automatic update of the status screen         |",
   "     --status-json              |      | Enable JSON format for status output                 |",
   "     --status-timer             | Num  | Sets seconds between status screen updates to X      | --status-timer=1",
@@ -97,6 +98,7 @@ static const char *const USAGE_BIG_PRE_HASHMODES[] =
   " -d, --backend-devices          | Str  | Backend devices to use, separated with commas        | -d 1",
   " -D, --opencl-device-types      | Str  | OpenCL device-types to use, separated with commas    | -D 1",
   " -O, --optimized-kernel-enable  |      | Enable optimized kernels (limits password length)    |",
+  " -M, --multiply-accel-disable   |      | Disable multiply kernel-accel with processor count   |",
   " -w, --workload-profile         | Num  | Enable a specific workload profile, see pool below   | -w 3",
   " -n, --kernel-accel             | Num  | Manual workload tuning, set outerloop step size to X | -n 64",
   " -u, --kernel-loops             | Num  | Manual workload tuning, set innerloop step size to X | -u 256",
@@ -237,6 +239,10 @@ static const char *const USAGE_BIG_POST_HASHMODES[] =
   "",
   "* https://hashcat.net/wiki/#howtos_videos_papers_articles_etc_in_the_wild",
   "* https://hashcat.net/faq/",
+  "",
+  "If you think you need help by a real human come to the hashcat Discord:",
+  "",
+  "* https://discord.gg/HFS523HGBT",
   NULL
 };
 
diff --git a/src/user_options.c b/src/user_options.c
index b7e810b14..cca5ab344 100644
--- a/src/user_options.c
+++ b/src/user_options.c
@@ -20,11 +20,13 @@
 #endif
 
 #ifdef WITH_BRAIN
-static const char *short_options = "hVvm:a:r:j:k:g:o:t:d:D:n:u:T:c:p:s:l:1:2:3:4:iIbw:OSz";
+static const char *const short_options = "hVvm:a:r:j:k:g:o:t:d:D:n:u:T:c:p:s:l:1:2:3:4:iIbw:OMSz";
 #else
-static const char *short_options = "hVvm:a:r:j:k:g:o:t:d:D:n:u:T:c:p:s:l:1:2:3:4:iIbw:OS";
+static const char *const short_options = "hVvm:a:r:j:k:g:o:t:d:D:n:u:T:c:p:s:l:1:2:3:4:iIbw:OMS";
 #endif
 
+static char *const SEPARATOR = ":";
+
 static const struct option long_options[] =
 {
   {"advice-disable",            no_argument,       NULL, IDX_ADVICE_DISABLE},
@@ -46,6 +48,7 @@ static const struct option long_options[] =
   {"custom-charset4",           required_argument, NULL, IDX_CUSTOM_CHARSET_4},
   {"debug-file",                required_argument, NULL, IDX_DEBUG_FILE},
   {"debug-mode",                required_argument, NULL, IDX_DEBUG_MODE},
+  {"deprecated-check-disable",  no_argument,       NULL, IDX_DEPRECATED_CHECK_DISABLE},
   {"encoding-from",             required_argument, NULL, IDX_ENCODING_FROM},
   {"encoding-to",               required_argument, NULL, IDX_ENCODING_TO},
   {"example-hashes",            no_argument,       NULL, IDX_HASH_INFO}, // alias of hash-info
@@ -88,6 +91,7 @@ static const struct option long_options[] =
   {"nonce-error-corrections",   required_argument, NULL, IDX_NONCE_ERROR_CORRECTIONS},
   {"opencl-device-types",       required_argument, NULL, IDX_OPENCL_DEVICE_TYPES},
   {"optimized-kernel-enable",   no_argument,       NULL, IDX_OPTIMIZED_KERNEL_ENABLE},
+  {"multiply-accel-disable",    no_argument,       NULL, IDX_MULTIPLY_ACCEL_DISABLE},
   {"outfile-autohex-disable",   no_argument,       NULL, IDX_OUTFILE_AUTOHEX_DISABLE},
   {"outfile-check-dir",         required_argument, NULL, IDX_OUTFILE_CHECK_DIR},
   {"outfile-check-timer",       required_argument, NULL, IDX_OUTFILE_CHECK_TIMER},
@@ -144,15 +148,15 @@ static const struct option long_options[] =
   {NULL,                        0,                 NULL, 0 }
 };
 
-static const char *ENCODING_FROM = "utf-8";
-static const char *ENCODING_TO   = "utf-8";
+static const char *const ENCODING_FROM = "utf-8";
+static const char *const ENCODING_TO   = "utf-8";
 
-static const char *RULE_BUF_R = ":";
-static const char *RULE_BUF_L = ":";
+static const char *const RULE_BUF_R = ":";
+static const char *const RULE_BUF_L = ":";
 
-static const char *DEF_MASK_CS_1 = "?l?d?u";
-static const char *DEF_MASK_CS_2 = "?l?d";
-static const char *DEF_MASK_CS_3 = "?l?d*!$@_";
+static const char *const DEF_MASK_CS_1 = "?l?d?u";
+static const char *const DEF_MASK_CS_2 = "?l?d";
+static const char *const DEF_MASK_CS_3 = "?l?d*!$@_";
 
 int user_options_init (hashcat_ctx_t *hashcat_ctx)
 {
@@ -188,6 +192,7 @@ int user_options_init (hashcat_ctx_t *hashcat_ctx)
   user_options->custom_charset_4          = NULL;
   user_options->debug_file                = NULL;
   user_options->debug_mode                = DEBUG_MODE;
+  user_options->deprecated_check_disable  = DEPRECATED_CHECK_DISABLE;
   user_options->encoding_from             = ENCODING_FROM;
   user_options->encoding_to               = ENCODING_TO;
   user_options->force                     = FORCE;
@@ -224,6 +229,7 @@ int user_options_init (hashcat_ctx_t *hashcat_ctx)
   user_options->nonce_error_corrections   = NONCE_ERROR_CORRECTIONS;
   user_options->opencl_device_types       = NULL;
   user_options->optimized_kernel_enable   = OPTIMIZED_KERNEL_ENABLE;
+  user_options->multiply_accel_disable    = MULTIPLY_ACCEL_DISABLE;
   user_options->outfile_autohex           = OUTFILE_AUTOHEX;
   user_options->outfile_check_dir         = NULL;
   user_options->outfile_check_timer       = OUTFILE_CHECK_TIMER;
@@ -374,6 +380,7 @@ int user_options_getopt (hashcat_ctx_t *hashcat_ctx, int argc, char **argv)
       case IDX_RESTORE:                   user_options->restore                   = true;                            break;
       case IDX_QUIET:                     user_options->quiet                     = true;                            break;
       case IDX_SHOW:                      user_options->show                      = true;                            break;
+      case IDX_DEPRECATED_CHECK_DISABLE:  user_options->deprecated_check_disable  = true;                            break;
       case IDX_LEFT:                      user_options->left                      = true;                            break;
       case IDX_ADVICE_DISABLE:            user_options->advice_disable            = true;                            break;
       case IDX_USERNAME:                  user_options->username                  = true;                            break;
@@ -451,6 +458,7 @@ int user_options_getopt (hashcat_ctx_t *hashcat_ctx, int argc, char **argv)
                                           user_options->backend_vector_width_chgd = true;                            break;
       case IDX_OPENCL_DEVICE_TYPES:       user_options->opencl_device_types       = optarg;                          break;
       case IDX_OPTIMIZED_KERNEL_ENABLE:   user_options->optimized_kernel_enable   = true;                            break;
+      case IDX_MULTIPLY_ACCEL_DISABLE:    user_options->multiply_accel_disable    = true;                            break;
       case IDX_WORKLOAD_PROFILE:          user_options->workload_profile          = hc_strtoul (optarg, NULL, 10);
                                           user_options->workload_profile_chgd     = true;                            break;
       case IDX_KERNEL_ACCEL:              user_options->kernel_accel              = hc_strtoul (optarg, NULL, 10);
@@ -479,7 +487,8 @@ int user_options_getopt (hashcat_ctx_t *hashcat_ctx, int argc, char **argv)
                                           user_options->segment_size_chgd         = true;                            break;
       case IDX_SCRYPT_TMTO:               user_options->scrypt_tmto               = hc_strtoul (optarg, NULL, 10);
                                           user_options->scrypt_tmto_chgd          = true;                            break;
-      case IDX_SEPARATOR:                 user_options->separator                 = optarg[0];                       break;
+      case IDX_SEPARATOR:                 user_options->separator                 = optarg;
+                                          user_options->separator_chgd            = true;                            break;
       case IDX_BITMAP_MIN:                user_options->bitmap_min                = hc_strtoul (optarg, NULL, 10);   break;
       case IDX_BITMAP_MAX:                user_options->bitmap_max                = hc_strtoul (optarg, NULL, 10);   break;
       case IDX_HOOK_THREADS:              user_options->hook_threads              = hc_strtoul (optarg, NULL, 10);   break;
@@ -580,6 +589,16 @@ int user_options_sanity (hashcat_ctx_t *hashcat_ctx)
   }
   #endif
 
+  if (user_options->separator_chgd == true)
+  {
+    if (strlen (user_options->separator) != 1)
+    {
+      event_log_error (hashcat_ctx, "Separator length has to be exactly 1 byte.");
+
+      return -1;
+    }
+  }
+
   if (user_options->slow_candidates == true)
   {
     if ((user_options->attack_mode != ATTACK_MODE_STRAIGHT)
@@ -1986,6 +2005,11 @@ void user_options_info (hashcat_ctx_t *hashcat_ctx)
       event_log_info (hashcat_ctx, "* --optimized-kernel-enable");
     }
 
+    if (user_options->multiply_accel_disable == true)
+    {
+      event_log_info (hashcat_ctx, "* --multiply-accel-disable");
+    }
+
     if (user_options->backend_vector_width_chgd == true)
     {
       event_log_info (hashcat_ctx, "* --backend-vector-width=%u", user_options->backend_vector_width);
@@ -2040,6 +2064,11 @@ void user_options_info (hashcat_ctx_t *hashcat_ctx)
       event_log_info (hashcat_ctx, "# option: --optimized-kernel-enable");
     }
 
+    if (user_options->multiply_accel_disable == true)
+    {
+      event_log_info (hashcat_ctx, "# option: --multiply-accel-disable");
+    }
+
     if (user_options->backend_vector_width_chgd == true)
     {
       event_log_info (hashcat_ctx, "# option: --backend-vector-width=%u", user_options->backend_vector_width);
@@ -2072,18 +2101,25 @@ void user_options_extra_init (hashcat_ctx_t *hashcat_ctx)
   user_options_t       *user_options       = hashcat_ctx->user_options;
   user_options_extra_t *user_options_extra = hashcat_ctx->user_options_extra;
 
+  // separator
+
+  if (user_options->separator)
+  {
+    user_options_extra->separator = user_options->separator[0];
+  }
+
   // attack-kern
 
   user_options_extra->attack_kern = ATTACK_KERN_NONE;
 
   switch (user_options->attack_mode)
   {
-    case ATTACK_MODE_STRAIGHT: user_options_extra->attack_kern = ATTACK_KERN_STRAIGHT; break;
-    case ATTACK_MODE_COMBI:    user_options_extra->attack_kern = ATTACK_KERN_COMBI;    break;
-    case ATTACK_MODE_BF:       user_options_extra->attack_kern = ATTACK_KERN_BF;       break;
-    case ATTACK_MODE_HYBRID1:  user_options_extra->attack_kern = ATTACK_KERN_COMBI;    break;
-    case ATTACK_MODE_HYBRID2:  user_options_extra->attack_kern = ATTACK_KERN_COMBI;    break;
-    case ATTACK_MODE_ASSOCIATION:  user_options_extra->attack_kern = ATTACK_KERN_STRAIGHT; break;
+    case ATTACK_MODE_STRAIGHT:      user_options_extra->attack_kern = ATTACK_KERN_STRAIGHT; break;
+    case ATTACK_MODE_COMBI:         user_options_extra->attack_kern = ATTACK_KERN_COMBI;    break;
+    case ATTACK_MODE_BF:            user_options_extra->attack_kern = ATTACK_KERN_BF;       break;
+    case ATTACK_MODE_HYBRID1:       user_options_extra->attack_kern = ATTACK_KERN_COMBI;    break;
+    case ATTACK_MODE_HYBRID2:       user_options_extra->attack_kern = ATTACK_KERN_COMBI;    break;
+    case ATTACK_MODE_ASSOCIATION:   user_options_extra->attack_kern = ATTACK_KERN_STRAIGHT; break;
   }
 
   // rules
@@ -3014,7 +3050,6 @@ void user_options_logger (hashcat_ctx_t *hashcat_ctx)
   user_options_t *user_options = hashcat_ctx->user_options;
   logfile_ctx_t  *logfile_ctx  = hashcat_ctx->logfile_ctx;
 
-  logfile_top_char   (user_options->separator);
   #ifdef WITH_BRAIN
   logfile_top_string (user_options->brain_session_whitelist);
   #endif
@@ -3039,6 +3074,7 @@ void user_options_logger (hashcat_ctx_t *hashcat_ctx)
   logfile_top_string (user_options->rule_buf_l);
   logfile_top_string (user_options->rule_buf_r);
   logfile_top_string (user_options->session);
+  logfile_top_string (user_options->separator);
   logfile_top_string (user_options->truecrypt_keyfiles);
   logfile_top_string (user_options->veracrypt_keyfiles);
   #ifdef WITH_BRAIN
@@ -3078,6 +3114,7 @@ void user_options_logger (hashcat_ctx_t *hashcat_ctx)
   logfile_top_uint   (user_options->markov_disable);
   logfile_top_uint   (user_options->markov_inverse);
   logfile_top_uint   (user_options->markov_threshold);
+  logfile_top_uint   (user_options->multiply_accel_disable);
   logfile_top_uint   (user_options->backend_info);
   logfile_top_uint   (user_options->backend_vector_width);
   logfile_top_uint   (user_options->optimized_kernel_enable);
diff --git a/tools/benchmark_deep.pl b/tools/benchmark_deep.pl
index df6777441..a481a87e9 100755
--- a/tools/benchmark_deep.pl
+++ b/tools/benchmark_deep.pl
@@ -230,16 +230,19 @@ my @hash_types =
   13751,
   13761,
   13771,
+  13781,
   13800,
   13900,
   14000,
   14100,
   14400,
+  14500,
   14700,
   14800,
   14900,
   15000,
   15100,
+  15200,
   15300,
   15400,
   15500,
@@ -250,10 +253,13 @@ my @hash_types =
   16200,
   16300,
   16400,
+  16500,
   16600,
+  16700,
   16800,
   16801,
   16900,
+  17210,
   17300,
   17400,
   17500,
@@ -333,12 +339,23 @@ my @hash_types =
   24700,
   24800,
   24900,
+  25000,
+  25100,
+  25200,
   25300,
   25400,
   25500,
+  25700,
   25900,
   26000,
   26100,
+  26200,
+  26300,
+  26401,
+  26402,
+  26403,
+  26500,
+  26600,
 );
 
 if (scalar @ARGV)
@@ -392,7 +409,6 @@ for my $hash_type (@hash_types)
     "--wordlist-autohex-disable",
     "--potfile-disable",
     "--logfile-disable",
-    "--hwmon-disable",
     "--status",
     "--status-timer", 1,
     "--runtime", $runtime,
diff --git a/tools/cryptoloop2hashcat.py b/tools/cryptoloop2hashcat.py
old mode 100644
new mode 100755
diff --git a/tools/test_modules/m25000.pm b/tools/test_modules/m25000.pm
new file mode 100644
index 000000000..71cc1c512
--- /dev/null
+++ b/tools/test_modules/m25000.pm
@@ -0,0 +1,116 @@
+#!/usr/bin/env perl
+
+##
+## Author......: See docs/credits.txt
+## License.....: MIT
+##
+
+use strict;
+use warnings;
+
+use Digest::MD5 qw (md5 md5_hex);
+use Digest::SHA qw (sha1 sha1_hex);
+use Digest::HMAC qw (hmac hmac_hex);
+
+sub module_constraints { [[8, 256], [24, 3000], [-1, -1], [-1, -1], [-1, -1]] }
+
+sub module_generate_hash
+{
+  my $word = shift;
+  my $salt = shift;
+  my $pkt_num = shift // int(rand(100000000));
+  my $engineID = shift // random_hex_string(26, 34);
+  my $mode = shift // int(rand(1)) + 1;
+
+  # make even if needed
+
+  if (length($salt) %2 == 1)
+  {
+    $salt = $salt . "8";
+  }
+
+  my $string1 = $word x 1048576;
+
+  $string1 = substr ($string1, 0, 1048576);
+
+  my $digest1 = '';
+
+  if ($mode eq 2)
+  {
+    $digest1 = sha1_hex ($string1);
+  }
+  elsif ($mode eq 1)
+  {
+    $digest1 = md5_hex ($string1);
+  }
+
+  my $buf = join '', $digest1, $engineID, $digest1;
+
+  my $digest = '';
+
+  if ($mode eq 2)
+  {
+    my $digest2 = sha1(pack("H*", $buf));
+
+    $digest = hmac_hex (pack("H*", $salt), $digest2, \&sha1);
+  }
+  elsif ($mode eq 1)
+  {
+    my $digest2 = md5(pack("H*", $buf));
+
+    $digest = hmac_hex (pack("H*", $salt), $digest2, \&md5);
+  }
+
+  $digest = substr ($digest, 0, 24);
+
+  my $hash = sprintf ("\$SNMPv3\$0\$%s\$%s\$%s\$%s", $pkt_num, $salt, $engineID, $digest);
+
+  return $hash;
+}
+
+sub module_verify_hash
+{
+  my $line = shift;
+
+  my $idx = index ($line, ':');
+
+  return unless $idx >= 0;
+
+  my $hash = substr ($line, 0, $idx);
+  my $word = substr ($line, $idx + 1);
+
+  return unless length ($word) gt 0;
+  return unless substr ($hash, 0, 10) eq '$SNMPv3$0$';
+
+  my (undef, $signature, $version, $pkt_num, $salt, $engineID, $digest) = split '\$', $hash;
+
+  return unless defined $signature;
+  return unless defined $version;
+  return unless defined $pkt_num;
+  return unless defined $salt;
+  return unless defined $engineID;
+  return unless defined $digest;
+
+  my $word_packed = pack_if_HEX_notation ($word);
+
+  # gen md5 & sha1 hashes
+
+  my $new_hash_md5 = module_generate_hash ($word_packed, $salt, $pkt_num, $engineID, 1);
+  my $new_hash_sha1 = module_generate_hash ($word_packed, $salt, $pkt_num, $engineID, 2);
+
+  # parse digests
+
+  my (undef, undef, undef, undef, undef, undef, $digest_md5) = split '\$', $new_hash_md5;
+  my (undef, undef, undef, undef, undef, undef, $digest_sha1) = split '\$', $new_hash_sha1;
+
+  if ($digest eq $digest_md5)
+  {
+    return ($new_hash_md5, $word);
+  }
+  else
+  {
+    return ($new_hash_sha1, $word);
+  }
+}
+
+1;
diff --git a/tools/test_modules/m25100.pm b/tools/test_modules/m25100.pm
new file mode 100644
index 000000000..2335f7f2f
--- /dev/null
+++ b/tools/test_modules/m25100.pm
@@ -0,0 +1,79 @@
+#!/usr/bin/env perl
+
+##
+## Author......: See docs/credits.txt
+## License.....: MIT
+##
+
+use strict;
+use warnings;
+
+use Digest::MD5 qw (md5 md5_hex);
+use Digest::HMAC qw (hmac hmac_hex);
+
+sub module_constraints { [[8, 256], [24, 3000], [-1, -1], [-1, -1], [-1, -1]] }
+
+sub module_generate_hash
+{
+  my $word = shift;
+  my $salt = shift;
+  my $pkt_num = shift // int(rand(100000000));
+  my $engineID = shift // random_hex_string(26, 34);
+
+  # make even if needed
+
+  if (length($salt) %2 == 1)
+  {
+    $salt = $salt . "8";
+  }
+
+  my $string1 = $word x 1048576;
+
+  $string1 = substr ($string1, 0, 1048576);
+
+  my $md5_digest1 = md5_hex ($string1);
+
+  my $buf = join '', $md5_digest1, $engineID, $md5_digest1;
+
+  my $md5_digest2 = md5(pack("H*", $buf));
+
+  my $digest = hmac_hex (pack("H*", $salt), $md5_digest2, \&md5);
+
+  $digest = substr ($digest, 0, 24);
+
+  my $hash = sprintf ("\$SNMPv3\$1\$%s\$%s\$%s\$%s", $pkt_num, $salt, $engineID, $digest);
+
+  return $hash;
+}
+
+sub module_verify_hash
+{
+  my $line = shift;
+
+  my $idx = index ($line, ':');
+
+  return unless $idx >= 0;
+
+  my $hash = substr ($line, 0, $idx);
+  my $word = substr ($line, $idx + 1);
+
+  return unless length ($word) gt 0;
+  return unless substr ($hash, 0, 10) eq '$SNMPv3$1$';
+
+  my (undef, $signature, $version, $pkt_num, $salt, $engineID, $digest) = split '\$', $hash;
+
+  return unless defined $signature;
+  return unless defined $version;
+  return unless defined $pkt_num;
+  return unless defined $salt;
+  return unless defined $engineID;
+  return unless defined $digest;
+
+  my $word_packed = pack_if_HEX_notation ($word);
+
+  my $new_hash = module_generate_hash ($word_packed, $salt, $pkt_num, $engineID);
+
+  return ($new_hash, $word);
+}
+
+1;
diff --git a/tools/test_modules/m25200.pm b/tools/test_modules/m25200.pm
new file mode 100644
index 000000000..d27908255
--- /dev/null
+++ b/tools/test_modules/m25200.pm
@@ -0,0 +1,79 @@
+#!/usr/bin/env perl
+
+##
+## Author......: See docs/credits.txt
+## License.....: MIT
+##
+
+use strict;
+use warnings;
+
+use Digest::SHA qw (sha1 sha1_hex);
+use Digest::HMAC qw (hmac hmac_hex);
+
+sub module_constraints { [[8, 256], [24, 3000], [-1, -1], [-1, -1], [-1, -1]] }
+
+sub module_generate_hash
+{
+  my $word = shift;
+  my $salt = shift;
+  my $pkt_num = shift // int(rand(100000000));
+  my $engineID = shift // random_hex_string(26, 34);
+
+  # make even if needed
+
+  if (length($salt) %2 == 1)
+  {
+    $salt = $salt . "8";
+  }
+
+  my $string1 = $word x 1048576;
+
+  $string1 = substr ($string1, 0, 1048576);
+
+  my $sha1_digest1 = sha1_hex ($string1);
+
+  my $buf = join '', $sha1_digest1, $engineID, $sha1_digest1;
+
+  my $sha1_digest2 = sha1(pack("H*", $buf));
+
+  my $digest = hmac_hex (pack("H*", $salt), $sha1_digest2, \&sha1);
+
+  $digest = substr ($digest, 0, 24);
+
+  my $hash = sprintf ("\$SNMPv3\$2\$%s\$%s\$%s\$%s", $pkt_num, $salt, $engineID, $digest);
+
+  return $hash;
+}
+
+sub module_verify_hash
+{
+  my $line = shift;
+
+  my $idx = index ($line, ':');
+
+  return unless $idx >= 0;
+
+  my $hash = substr ($line, 0, $idx);
+  my $word = substr ($line, $idx + 1);
+
+  return unless length ($word) gt 0;
+  return unless substr ($hash, 0, 10) eq '$SNMPv3$2$';
+
+  my (undef, $signature, $version, $pkt_num, $salt, $engineID, $digest) = split '\$', $hash;
+
+  return unless defined $signature;
+  return unless defined $version;
+  return unless defined $pkt_num;
+  return unless defined $salt;
+  return unless defined $engineID;
+  return unless defined $digest;
+
+  my $word_packed = pack_if_HEX_notation ($word);
+
+  my $new_hash = module_generate_hash ($word_packed, $salt, $pkt_num, $engineID); #, $digest);
+
+  return ($new_hash, $word);
+}
+
+1;
diff --git a/tools/test_modules/m25700.pm b/tools/test_modules/m25700.pm
index d5c2e2424..56f6c042f 100644
--- a/tools/test_modules/m25700.pm
+++ b/tools/test_modules/m25700.pm
@@ -8,7 +8,7 @@
 use strict;
 use warnings;
 
-sub module_constraints { [[0, 256], [8, 8], [0, 55], [8, 8], [-1, -1]] }
+sub module_constraints { [[-1, -1], [-1, -1], [0, 55], [8, 8], [-1, -1]] }
 
 sub MurmurHash
 {
diff --git a/tools/test_modules/m26700.pm b/tools/test_modules/m26700.pm
new file mode 100644
index 000000000..a97bc246e
--- /dev/null
+++ b/tools/test_modules/m26700.pm
@@ -0,0 +1,79 @@
+#!/usr/bin/env perl
+
+##
+## Author......: See docs/credits.txt
+## License.....: MIT
+##
+
+use strict;
+use warnings;
+
+use Digest::SHA qw (sha224 sha224_hex);
+use Digest::HMAC qw (hmac hmac_hex);
+
+sub module_constraints { [[8, 256], [32, 3000], [-1, -1], [-1, -1], [-1, -1]] }
+
+sub module_generate_hash
+{
+  my $word = shift;
+  my $salt = shift;
+  my $pkt_num = shift // int(rand(100000000));
+  my $engineID = shift // random_hex_string(26, 34);
+
+  # make even if needed
+
+  if (length($salt) %2 == 1)
+  {
+    $salt = $salt . "8";
+  }
+
+  my $string1 = $word x 1048576;
+
+  $string1 = substr ($string1, 0, 1048576);
+
+  my $sha224_digest1 = sha224_hex ($string1);
+
+  my $buf = join '', $sha224_digest1, $engineID, $sha224_digest1;
+
+  my $sha224_digest2 = sha224(pack("H*", $buf));
+
+  my $digest = hmac_hex (pack("H*", $salt), $sha224_digest2, \&sha224);
+
+  $digest = substr ($digest, 0, 32);
+
+  my $hash = sprintf ("\$SNMPv3\$3\$%s\$%s\$%s\$%s", $pkt_num, $salt, $engineID, $digest);
+
+  return $hash;
+}
+
+sub module_verify_hash
+{
+  my $line = shift;
+
+  my $idx = index ($line, ':');
+
+  return unless $idx >= 0;
+
+  my $hash = substr ($line, 0, $idx);
+  my $word = substr ($line, $idx + 1);
+
+  return unless length ($word) gt 0;
+  return unless substr ($hash, 0, 10) eq '$SNMPv3$3$';
+
+  my (undef, $signature, $version, $pkt_num, $salt, $engineID, $digest) = split '\$', $hash;
+
+  return unless defined $signature;
+  return unless defined $version;
+  return unless defined $pkt_num;
+  return unless defined $salt;
+  return unless defined $engineID;
+  return unless defined $digest;
+
+  my $word_packed = pack_if_HEX_notation ($word);
+
+  my $new_hash = module_generate_hash ($word_packed, $salt, $pkt_num, $engineID);
+
+  return ($new_hash, $word);
+}
+
+1;
diff --git a/tools/test_modules/m26800.pm b/tools/test_modules/m26800.pm
new file mode 100644
index 000000000..3039afb6b
--- /dev/null
+++ b/tools/test_modules/m26800.pm
@@ -0,0 +1,79 @@
+#!/usr/bin/env perl
+
+##
+## Author......: See docs/credits.txt
+## License.....: MIT
+##
+
+use strict;
+use warnings;
+
+use Digest::SHA qw (sha256 sha256_hex);
+use Digest::HMAC qw (hmac hmac_hex);
+
+sub module_constraints { [[8, 256], [48, 3000], [-1, -1], [-1, -1], [-1, -1]] }
+
+sub module_generate_hash
+{
+  my $word = shift;
+  my $salt = shift;
+  my $pkt_num = shift // int(rand(100000000));
+  my $engineID = shift // random_hex_string(26, 34);
+
+  # make even if needed
+
+  if (length($salt) %2 == 1)
+  {
+    $salt = $salt . "8";
+  }
+
+  my $string1 = $word x 1048576;
+
+  $string1 = substr ($string1, 0, 1048576);
+
+  my $sha256_digest1 = sha256_hex ($string1);
+
+  my $buf = join '', $sha256_digest1, $engineID, $sha256_digest1;
+
+  my $sha256_digest2 = sha256(pack("H*", $buf));
+
+  my $digest = hmac_hex (pack("H*", $salt), $sha256_digest2, \&sha256);
+
+  $digest = substr ($digest, 0, 48);
+
+  my $hash = sprintf ("\$SNMPv3\$4\$%s\$%s\$%s\$%s", $pkt_num, $salt, $engineID, $digest);
+
+  return $hash;
+}
+
+sub module_verify_hash
+{
+  my $line = shift;
+
+  my $idx = index ($line, ':');
+
+  return unless $idx >= 0;
+
+  my $hash = substr ($line, 0, $idx);
+  my $word = substr ($line, $idx + 1);
+
+  return unless length ($word) gt 0;
+  return unless substr ($hash, 0, 10) eq '$SNMPv3$4$';
+
+  my (undef, $signature, $version, $pkt_num, $salt, $engineID, $digest) = split '\$', $hash;
+
+  return unless defined $signature;
+  return unless defined $version;
+  return unless defined $pkt_num;
+  return unless defined $salt;
+  return unless defined $engineID;
+  return unless defined $digest;
+
+  my $word_packed = pack_if_HEX_notation ($word);
+
+  my $new_hash = module_generate_hash ($word_packed, $salt, $pkt_num, $engineID);
+
+  return ($new_hash, $word);
+}
+
+1;
diff --git a/tools/test_modules/m26900.pm b/tools/test_modules/m26900.pm
new file mode 100644
index 000000000..89dbb4446
--- /dev/null
+++ b/tools/test_modules/m26900.pm
@@ -0,0 +1,85 @@
+#!/usr/bin/env perl
+
+##
+## Author......: See docs/credits.txt
+## License.....: MIT
+##
+
+use strict;
+use warnings;
+
+use Digest::SHA qw (sha384 sha384_hex hmac_sha384_hex);
+#use Digest::HMAC qw (hmac hmac_hex);
+
+sub module_constraints { [[8, 256], [64, 3000], [-1, -1], [-1, -1], [-1, -1]] }
+
+sub module_generate_hash
+{
+  my $word = shift;
+  my $salt = shift;
+  my $pkt_num = shift // int(rand(100000000));
+  my $engineID = shift // random_hex_string(26, 34);
+
+  # padding engineID: fill with zero
+
+  my $pad_len = 34 - length ($engineID);
+
+  $engineID = join '', $engineID, "0" x $pad_len;
+
+  # make salt even if needed
+
+  if (length($salt) %2 == 1)
+  {
+    $salt = $salt . "8";
+  }
+
+  my $string1 = $word x 1048576;
+
+  $string1 = substr ($string1, 0, 1048576);
+
+  my $sha384_digest1 = sha384_hex ($string1);
+
+  my $buf = join '', $sha384_digest1, $engineID, $sha384_digest1;
+
+  my $sha384_digest2 = sha384(pack("H*", $buf));
+
+  my $digest = hmac_sha384_hex (pack("H*", $salt), $sha384_digest2);
+
+  $digest = substr ($digest, 0, 64);
+
+  my $hash = "\$SNMPv3\$5\$" . $pkt_num . "\$" . $salt . "\$" . $engineID . "\$" . $digest;
+
+  return $hash;
+}
+
+sub module_verify_hash
+{
+  my $line = shift;
+
+  my $idx = index ($line, ':');
+
+  return unless $idx >= 0;
+
+  my $hash = substr ($line, 0, $idx);
+  my $word = substr ($line, $idx + 1);
+
+  return unless length ($word) gt 0;
+  return unless substr ($hash, 0, 10) eq '$SNMPv3$5$';
+
+  my (undef, $signature, $version, $pkt_num, $salt, $engineID, $digest) = split '\$', $hash;
+
+  return unless defined $signature;
+  return unless defined $version;
+  return unless defined $pkt_num;
+  return unless defined $salt;
+  return unless defined $engineID;
+  return unless defined $digest;
+
+  my $word_packed = pack_if_HEX_notation ($word);
+
+  my $new_hash = module_generate_hash ($word_packed, $salt, $pkt_num, $engineID);
+
+  return ($new_hash, $word);
+}
+
+1;
diff --git a/tools/test_modules/m27000.pm b/tools/test_modules/m27000.pm
new file mode 100644
index 000000000..2ae2ea163
--- /dev/null
+++ b/tools/test_modules/m27000.pm
@@ -0,0 +1,176 @@
+#!/usr/bin/env perl
+
+##
+## Author......: See docs/credits.txt
+## License.....: MIT
+##
+
+use strict;
+use warnings;
+
+use Digest::HMAC qw (hmac hmac_hex);
+use Digest::MD5  qw (md5);
+use Encode       qw (encode);
+use Crypt::ECB   qw (encrypt);
+
+sub setup_des_key
+{
+  my @key_56 = split (//, shift);
+
+  my $key = "";
+
+  $key = $key_56[0];
+
+  $key .= chr (((ord ($key_56[0]) << 7) | (ord ($key_56[1]) >> 1)) & 255);
+  $key .= chr (((ord ($key_56[1]) << 6) | (ord ($key_56[2]) >> 2)) & 255);
+  $key .= chr (((ord ($key_56[2]) << 5) | (ord ($key_56[3]) >> 3)) & 255);
+  $key .= chr (((ord ($key_56[3]) << 4) | (ord ($key_56[4]) >> 4)) & 255);
+  $key .= chr (((ord ($key_56[4]) << 3) | (ord ($key_56[5]) >> 5)) & 255);
+  $key .= chr (((ord ($key_56[5]) << 2) | (ord ($key_56[6]) >> 6)) & 255);
+  $key .= chr (( ord ($key_56[6]) << 1) & 255);
+
+  return $key;
+}
+
+sub get_random_netntlmv1_salt
+{
+  my $len_user   = shift;
+  my $len_domain = shift;
+
+  my $char;
+  my $type;
+  my $user = "";
+
+  for (my $i = 0; $i < $len_user; $i++)
+  {
+    $type = random_number (1, 3);
+
+    if ($type == 1)
+    {
+      $char = random_numeric_string (1);
+    }
+    elsif ($type == 2)
+    {
+      $char = random_uppercase_string (1);
+    }
+    else
+    {
+      $char = random_lowercase_string (1);
+    }
+
+    $user .= $char;
+  }
+
+  my $domain = "";
+
+  for (my $i = 0; $i < $len_domain; $i++)
+  {
+    $type = random_number (1, 3);
+
+    if ($type == 1)
+    {
+      $char = random_numeric_string (1);
+    }
+    elsif ($type == 2)
+    {
+      $char = random_uppercase_string (1);
+    }
+    else
+    {
+      $char = random_lowercase_string (1);
+    }
+
+    $domain .= $char;
+  }
+
+  my $c_challenge = random_bytes (8);
+  my $s_challenge = random_bytes (8);
+
+  my $salt_buf = $user . "::" . $domain . ":" . unpack ("H*", $c_challenge) . unpack ("H*", $s_challenge);
+
+  return $salt_buf;
+}
+
+sub module_constraints { [[32, 32], [-1, -1], [-1, -1], [-1, -1], [-1, -1]] } # room for improvement in pure kernel mode
+
+sub module_generate_hash
+{
+  my $word   = shift;
+  my $unused = shift;
+  my $salt   = shift // get_random_netntlmv1_salt (random_number (0, 15), random_number (0, 15));
+
+  my $index1 = index  ($salt, "::");
+  my $user   = substr ($salt, 0, $index1);
+
+  my $index2 = index  ($salt, ":", $index1 + 2);
+  my $domain = substr ($salt, $index1 + 2, $index2 - $index1 - 2);
+
+  my $len = length (substr ($salt, $index2 + 1));
+
+  my $c_challenge_hex;
+
+  if ($len > 32)
+  {
+    $c_challenge_hex = substr ($salt, $index2 +  1, 48);
+    $index2 += 32;
+  }
+  else
+  {
+    $c_challenge_hex  = substr ($salt, $index2 +  1, 16);
+    $c_challenge_hex .= 00 x 32;
+  }
+
+  my $c_challenge     = pack   ("H*", substr ($c_challenge_hex, 0, 16));
+  my $s_challenge_hex = substr ($salt, $index2 + 17, 16);
+  my $s_challenge     = pack   ("H*", $s_challenge_hex);
+
+  my $challenge = substr (md5 ($s_challenge . $c_challenge), 0, 8);
+
+  my $ntresp;
+
+  my $nthash = pack ("H*", $word);
+  my $nthashpadded = $nthash . "\x00" x 5;
+
+  $ntresp .= Crypt::ECB::encrypt (setup_des_key (substr ($nthashpadded,  0, 7)), "DES", $challenge, "none");
+  $ntresp .= Crypt::ECB::encrypt (setup_des_key (substr ($nthashpadded,  7, 7)), "DES", $challenge, "none");
+  $ntresp .= Crypt::ECB::encrypt (setup_des_key (substr ($nthashpadded, 14, 7)), "DES", $challenge, "none");
+
+  my $tmp_hash = sprintf ("%s::%s:%s:%s:%s", $user, $domain, $c_challenge_hex, unpack ("H*", $ntresp), $s_challenge_hex);
+
+  return $tmp_hash;
+}
+
+sub module_verify_hash
+{
+  my $line = shift;
+
+  my $index1 = index ($line, "::");
+
+  return if $index1 < 1;
+
+  my $index2 = index ($line, ":", $index1 + 2);
+
+  return if $index2 < 1;
+
+  $index2 = index ($line, ":", $index2 + 1);
+
+  return if $index2 < 1;
+
+  my $salt = substr ($line, 0, $index2 - 32);
+
+  $index2 = index ($line, ":", $index2 + 1);
+
+  return if $index2 < 1;
+
+  $salt .= substr ($line, $index2 + 1, 16);
+
+  my $word = substr ($line, $index2 + 1 + 16 + 1);
+
+  # my $word_packed = pack ("H*", $word);
+
+  my $new_hash = module_generate_hash ($word, undef, $salt);
+
+  return ($new_hash, $word);
+}
+
+1;
diff --git a/tools/test_modules/m27100.pm b/tools/test_modules/m27100.pm
new file mode 100644
index 000000000..fb52146c5
--- /dev/null
+++ b/tools/test_modules/m27100.pm
@@ -0,0 +1,90 @@
+#!/usr/bin/env perl
+
+##
+## Author......: See docs/credits.txt
+## License.....: MIT
+##
+
+use strict;
+use warnings;
+
+use Digest::HMAC qw (hmac hmac_hex);
+use Digest::MD5  qw (md5);
+use Encode       qw (encode);
+
+sub module_constraints { [[32, 32], [-1, -1], [-1, -1], [-1, -1], [-1, -1]] } 
+
+sub module_generate_hash
+{
+  my $word = shift;
+  my $user = shift;
+
+  my $user_len   = length $user;
+  my $domain_len = 27 - $user_len;
+
+  my $domain = shift // random_string ($domain_len);
+  my $srv_ch = shift // random_hex_string (2 * 8);
+  my $cli_ch = shift // random_client_challenge ();
+
+  my $b_srv_ch = pack ('H*', $srv_ch);
+  my $b_cli_ch = pack ('H*', $cli_ch);
+
+  my $nthash = pack ("H*", $word);
+  my $identity = encode ('UTF-16LE', uc ($user) . $domain);
+  my $digest   = hmac_hex ($b_srv_ch . $b_cli_ch, hmac ($identity, $nthash, \&md5, 64), \&md5, 64);
+
+  my $hash = sprintf ("%s::%s:%s:%s:%s", $user, $domain, $srv_ch, $digest, $cli_ch);
+
+  return $hash;
+}
+
+sub module_verify_hash
+{
+  my $line = shift;
+
+  my $user;
+  my $domain;
+  my $srv_ch;
+  my $cli_ch;
+  my $word;
+
+  my $hash;
+
+  my $index1 = index ($line, '::');
+  my $index2 = index ($line, ':', $index1 + 2);
+  my $index3 = index ($line, ':', $index2 + 3 + 16 + 32);
+
+  return if $index1 eq -1;
+  return if $index2 eq -1;
+  return if $index3 eq -1;
+
+  $hash = substr ($line, 0, $index3);
+
+  $user   = substr ($line, 0, $index1);
+  $domain = substr ($line, $index1 + 2, $index2 - $index1 - 2);
+  $srv_ch = substr ($line, $index2 + 1, 16);
+  $cli_ch = substr ($line, $index2 + 3 + 16 + 32, $index3 - $index2 - 3 - 16 - 32);
+  $word   = substr ($line, $index3 + 1);
+
+  my $word_packed = pack ("H*", $word);
+
+  my $new_hash = module_generate_hash ($word_packed, $user, $domain, $srv_ch, $cli_ch);
+
+  return ($new_hash, $word);
+}
+
+sub random_client_challenge
+{
+  my $ch;
+
+  $ch .= '0101000000000000';
+  $ch .= random_hex_string (2 * 16);
+  $ch .= '00000000';
+  $ch .= random_hex_string (2 * random_count (20));
+  $ch .= '00';
+
+  return $ch;
+}
+
+1;
+
diff --git a/tools/test_modules/m27200.pm b/tools/test_modules/m27200.pm
new file mode 100644
index 000000000..d0966d6ee
--- /dev/null
+++ b/tools/test_modules/m27200.pm
@@ -0,0 +1,44 @@
+#!/usr/bin/env perl
+
+##
+## Author......: See docs/credits.txt
+## License.....: MIT
+##
+
+use strict;
+use warnings;
+
+use Digest::SHA qw (sha1_hex);
+
+sub module_constraints { [[0, 256], [40, 40], [0, 9], [40, 40], [-1, -1]] }
+
+sub module_generate_hash
+{
+  my $word = shift;
+  my $salt = shift || random_hex_string (40);
+
+  my $digest = sha1_hex ('--' . $salt . '--' . $word . '--');
+
+  my $hash = sprintf ("%s:%s", $digest, $salt);
+
+  return $hash;
+}
+
+sub module_verify_hash
+{
+  my $line = shift;
+
+  my ($hash, $salt, $word) = split (':', $line);
+
+  return unless defined $hash;
+  return unless defined $salt;
+  return unless defined $word;
+
+  my $word_packed = pack_if_HEX_notation ($word);
+
+  my $new_hash = module_generate_hash ($word_packed, $salt);
+
+  return ($new_hash, $word);
+}
+
+1;
diff --git a/tools/test_modules/m27300.pm b/tools/test_modules/m27300.pm
new file mode 100644
index 000000000..a2cec5372
--- /dev/null
+++ b/tools/test_modules/m27300.pm
@@ -0,0 +1,84 @@
+#!/usr/bin/env perl
+
+##
+## Author......: See docs/credits.txt
+## License.....: MIT
+##
+
+use strict;
+use warnings;
+
+use Digest::SHA qw (sha512 sha512_hex hmac_sha512_hex);
+
+sub module_constraints { [[8, 256], [96, 3000], [-1, -1], [-1, -1], [-1, -1]] }
+
+sub module_generate_hash
+{
+  my $word = shift;
+  my $salt = shift;
+  my $pkt_num = shift // int(rand(100000000));
+  my $engineID = shift // random_hex_string(26, 34);
+
+  # padding engineID: fill with zero
+
+  my $pad_len = 34 - length ($engineID);
+
+  $engineID = join '', $engineID, "0" x $pad_len;
+
+  # make salt even if needed
+
+  if (length($salt) %2 == 1)
+  {
+    $salt = $salt . "8";
+  }
+
+  my $string1 = $word x 1048576;
+
+  $string1 = substr ($string1, 0, 1048576);
+
+  my $sha512_digest1 = sha512_hex ($string1);
+
+  my $buf = join '', $sha512_digest1, $engineID, $sha512_digest1;
+
+  my $sha512_digest2 = sha512(pack("H*", $buf));
+
+  my $digest = hmac_sha512_hex (pack("H*", $salt), $sha512_digest2);
+
+  $digest = substr ($digest, 0, 96);
+
+  my $hash = "\$SNMPv3\$6\$" . $pkt_num . "\$" . $salt . "\$" . $engineID . "\$" . $digest;
+
+  return $hash;
+}
+
+sub module_verify_hash
+{
+  my $line = shift;
+
+  my $idx = index ($line, ':');
+
+  return unless $idx >= 0;
+
+  my $hash = substr ($line, 0, $idx);
+  my $word = substr ($line, $idx + 1);
+
+  return unless length ($word) gt 0;
+  return unless substr ($hash, 0, 10) eq '$SNMPv3$6$';
+
+  my (undef, $signature, $version, $pkt_num, $salt, $engineID, $digest) = split '\$', $hash;
+
+  return unless defined $signature;
+  return unless defined $version;
+  return unless defined $pkt_num;
+  return unless defined $salt;
+  return unless defined $engineID;
+  return unless defined $digest;
+
+  my $word_packed = pack_if_HEX_notation ($word);
+
+  my $new_hash = module_generate_hash ($word_packed, $salt, $pkt_num, $engineID);
+
+  return ($new_hash, $word);
+}
+
+1;
diff --git a/tools/test_modules/m27400.pm b/tools/test_modules/m27400.pm
new file mode 100644
index 000000000..ad5941104
--- /dev/null
+++ b/tools/test_modules/m27400.pm
@@ -0,0 +1,102 @@
+#!/usr/bin/env perl
+
+##
+## Author......: See docs/credits.txt
+## License.....: MIT
+##
+
+use strict;
+use warnings;
+
+use Crypt::PBKDF2;
+use Crypt::CBC;
+
+sub module_constraints { [[0, 256], [32, 32], [-1, -1], [-1, -1], [-1, -1]] }
+
+sub module_generate_hash
+{
+  my $word = shift;
+  my $salt_str = shift;
+  my $ct_str = shift;
+
+  my $iv_str = "";
+
+  if (defined $ct_str)
+  {
+    $iv_str = substr($ct_str, 0, 32);
+  }
+  else
+  {
+    $iv_str = random_hex_string (32);
+  }
+
+  my $salt = pack ("H*", $salt_str);
+
+  my $iv = pack ("H*", $iv_str);
+
+  my $iterations = 10000;
+
+  my $hasher = Crypt::PBKDF2->hasher_from_algorithm ('HMACSHA1');
+
+  my $pbkdf2 = Crypt::PBKDF2->new (
+    hasher       => $hasher,
+    iterations   => $iterations,
+    output_len   => 32
+  );
+
+  my $key = $pbkdf2->PBKDF2 ($salt, $word);
+
+  my $cipher = Crypt::CBC->new ({
+    key         => $key,
+    cipher      => "Crypt::Rijndael",
+    iv          => $iv,
+    literal_key => 1,
+    header      => "none",
+    keysize     => 32
+  });
+
+  my $hash = "";
+
+  my $data = 'type=key:cipher=';
+
+  my $encrypted = unpack ("H*", $cipher->encrypt ($data));
+
+  if (defined $ct_str)
+  {
+    my $ct_bin = pack ("H*", $ct_str);
+    my $iv_bin = substr ($ct_bin, 0, 16);
+
+    $hash = sprintf ("\$vmx\$0\$%s\$%s\$%s%s", $iterations, unpack ("H*", $salt), unpack ("H*", $iv_bin), substr ($encrypted, 0, 32));
+  }
+  else
+  {
+    $hash = sprintf ("\$vmx\$0\$%s\$%s\$%s%s", $iterations, unpack ("H*", $salt), unpack ("H*", $iv), substr ($encrypted, 0, 32));
+  }
+
+  return $hash;
+}
+
+sub module_verify_hash
+{
+  my $line = shift;
+
+  my ($hash, $word) = split (':', $line);
+
+  return unless defined $hash;
+  return unless defined $word;
+
+  my (undef, $signature, $version, $rounds, $salt, $ct) = split '\$', $hash;
+
+  return unless ($signature eq "vmx");
+  return unless ($version eq 0);
+  return unless ($rounds eq 10000);
+  return unless (length $ct eq 64);
+
+  my $word_packed = pack_if_HEX_notation ($word);
+
+  my $new_hash = module_generate_hash ($word_packed, $salt, $ct);
+
+  return ($new_hash, $word);
+}
+
+1;
diff --git a/tools/vmwarevmx2hashcat.py b/tools/vmwarevmx2hashcat.py
new file mode 100644
index 000000000..015b2a614
--- /dev/null
+++ b/tools/vmwarevmx2hashcat.py
@@ -0,0 +1,76 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+
+# Based on "pyvmx-cracker" (https://github.com/axcheron/pyvmx-cracker) (MIT license)
+
+# Author: Gabriele 'matrix' Gristina
+# Version: 1.0
+# Date: Tue 13 Jul 2021 01:29:23 PM CEST
+# License: MIT
+
+import argparse
+from urllib.parse import unquote
+from binascii import hexlify
+import re
+import base64
+
+ks_re = '.+phrase/(.*?)/pass2key=(.*?):cipher=(.*?):rounds=(.*?):salt=(.*?),(.*?),(.*?)\)'
+
+ks_struct = {
+    'password_hash': None,
+    'password_cipher': None,
+    'hash_round': None,
+    'salt': None,
+    'dict': None
+}
+
+def parse_keysafe(file):
+    try:
+        with open(file, 'r') as data:
+            lines = data.readlines()
+    except (OSError, IOError):
+        sys.exit('[-] Cannot read from file ' + data)
+
+    for line in lines:
+        if 'encryption.keySafe' in line:
+            keysafe = line
+
+    keysafe = unquote(keysafe)
+
+    match = re.match(ks_re, keysafe)
+    if not match:
+        msg = 'Unsupported format of the encryption.keySafe line:\n' + keysafe
+        raise ValueError(msg)
+
+    vmx_ks = ks_struct
+
+    vmx_ks['password_hash'] = match.group(2)
+    if vmx_ks['password_hash'] != 'PBKDF2-HMAC-SHA-1':
+        msg = 'Unsupported password hash format: ' + vmx_ks['password_hash']
+        raise ValueError(msg)
+
+    vmx_ks['password_cipher'] = match.group(3)
+    if vmx_ks['password_cipher'] != 'AES-256':
+        msg = 'Unsupported cypher format: ' + vmx_ks['password_cypher']
+        raise ValueError(msg)
+
+    vmx_ks['hash_round'] = int(match.group(4))
+    vmx_ks['salt'] = base64.b64decode(unquote(match.group(5)))
+    vmx_ks['dict'] = base64.b64decode(match.group(7))[0:32]
+
+    return vmx_ks
+
+def pyvmx(vmx):
+    keysafe = parse_keysafe(vmx)
+    print("$vmx$0$" + str(keysafe['hash_round']) + "$" + hexlify(keysafe['salt']).decode() + "$" + hexlify(keysafe['dict']).decode())
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description="vmwarevmx2hashcat extraction tool")
+    parser.add_argument('--vmx', required=True, help='set vmware vmx file from path', type=str)
+
+    args = parser.parse_args()
+    if args.vmx:
+        pyvmx(args.vmx)
+    else:
+        parser.print_help()
+        exit(1)