From f5c53a7e7736df04bc1b01e3ce6ffd06160f380e Mon Sep 17 00:00:00 2001
From: piwvvo <pivvogoodpivvo@protonmail.com>
Date: Mon, 14 Nov 2022 18:19:41 +0100
Subject: [PATCH 1/6] added mode 30500

---
 OpenCL/m30500_a0-optimized.cl |  988 ++++++++++++++++++++++++
 OpenCL/m30500_a0-pure.cl      |  353 +++++++++
 OpenCL/m30500_a1-optimized.cl | 1104 +++++++++++++++++++++++++++
 OpenCL/m30500_a1-pure.cl      |  347 +++++++++
 OpenCL/m30500_a3-optimized.cl | 1350 +++++++++++++++++++++++++++++++++
 OpenCL/m30500_a3-pure.cl      |  373 +++++++++
 docs/changes.txt              |    1 +
 docs/readme.txt               |    1 +
 src/modules/module_30500.c    |  236 ++++++
 tools/test_modules/m30500.pm  |   44 ++
 10 files changed, 4797 insertions(+)
 create mode 100644 OpenCL/m30500_a0-optimized.cl
 create mode 100644 OpenCL/m30500_a0-pure.cl
 create mode 100644 OpenCL/m30500_a1-optimized.cl
 create mode 100644 OpenCL/m30500_a1-pure.cl
 create mode 100644 OpenCL/m30500_a3-optimized.cl
 create mode 100644 OpenCL/m30500_a3-pure.cl
 create mode 100644 src/modules/module_30500.c
 create mode 100644 tools/test_modules/m30500.pm

diff --git a/OpenCL/m30500_a0-optimized.cl b/OpenCL/m30500_a0-optimized.cl
new file mode 100644
index 000000000..08e8194de
--- /dev/null
+++ b/OpenCL/m30500_a0-optimized.cl
@@ -0,0 +1,988 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include M2S(INCLUDE_PATH/inc_vendor.h)
+#include M2S(INCLUDE_PATH/inc_types.h)
+#include M2S(INCLUDE_PATH/inc_platform.cl)
+#include M2S(INCLUDE_PATH/inc_common.cl)
+#include M2S(INCLUDE_PATH/inc_rp_optimized.h)
+#include M2S(INCLUDE_PATH/inc_rp_optimized.cl)
+#include M2S(INCLUDE_PATH/inc_simd.cl)
+#include M2S(INCLUDE_PATH/inc_hash_md5.cl)
+#endif
+
+#if   VECT_SIZE == 1
+#define uint_to_hex_lower8(i) make_u32x (l_bin2asc[(i)])
+#elif VECT_SIZE == 2
+#define uint_to_hex_lower8(i) make_u32x (l_bin2asc[(i).s0], l_bin2asc[(i).s1])
+#elif VECT_SIZE == 4
+#define uint_to_hex_lower8(i) make_u32x (l_bin2asc[(i).s0], l_bin2asc[(i).s1], l_bin2asc[(i).s2], l_bin2asc[(i).s3])
+#elif VECT_SIZE == 8
+#define uint_to_hex_lower8(i) make_u32x (l_bin2asc[(i).s0], l_bin2asc[(i).s1], l_bin2asc[(i).s2], l_bin2asc[(i).s3], l_bin2asc[(i).s4], l_bin2asc[(i).s5], l_bin2asc[(i).s6], l_bin2asc[(i).s7])
+#elif VECT_SIZE == 16
+#define uint_to_hex_lower8(i) make_u32x (l_bin2asc[(i).s0], l_bin2asc[(i).s1], l_bin2asc[(i).s2], l_bin2asc[(i).s3], l_bin2asc[(i).s4], l_bin2asc[(i).s5], l_bin2asc[(i).s6], l_bin2asc[(i).s7], l_bin2asc[(i).s8], l_bin2asc[(i).s9], l_bin2asc[(i).sa], l_bin2asc[(i).sb], l_bin2asc[(i).sc], l_bin2asc[(i).sd], l_bin2asc[(i).se], l_bin2asc[(i).sf])
+#endif
+
+KERNEL_FQ void m30500_m04 (KERN_ATTR_RULES ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * bin2asc table
+   */
+
+  LOCAL_VK u32 l_bin2asc[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    const u32 i0 = (i >> 0) & 15;
+    const u32 i1 = (i >> 4) & 15;
+
+    l_bin2asc[i] = ((i0 < 10) ? '0' + i0 : 'a' - 10 + i0) << 8
+                 | ((i1 < 10) ? '0' + i1 : 'a' - 10 + i1) << 0;
+  }
+
+  SYNC_THREADS ();
+
+  if (gid >= GID_CNT) return;
+
+  /**
+   * base
+   */
+
+  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];
+  u32 salt_buf1[4];
+  u32 salt_buf2[4];
+  u32 salt_buf3[4];
+
+  salt_buf0[0] = salt_bufs[SALT_POS_HOST].salt_buf_pc[0];
+  salt_buf0[1] = salt_bufs[SALT_POS_HOST].salt_buf_pc[1];
+  salt_buf0[2] = salt_bufs[SALT_POS_HOST].salt_buf_pc[2];
+  salt_buf0[3] = salt_bufs[SALT_POS_HOST].salt_buf_pc[3];
+  salt_buf1[0] = salt_bufs[SALT_POS_HOST].salt_buf_pc[4];
+  salt_buf1[1] = salt_bufs[SALT_POS_HOST].salt_buf_pc[5];
+  salt_buf1[2] = salt_bufs[SALT_POS_HOST].salt_buf_pc[6];
+  salt_buf1[3] = salt_bufs[SALT_POS_HOST].salt_buf_pc[7];
+  salt_buf2[0] = 0;
+  salt_buf2[1] = 0;
+  salt_buf2[2] = 0;
+  salt_buf2[3] = 0;
+  salt_buf3[0] = 0;
+  salt_buf3[1] = 0;
+  salt_buf3[2] = 0;
+  salt_buf3[3] = 0;
+
+  /**
+   * 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);
+
+    append_0x80_2x4_VV (w0, w1, out_len);
+
+    w3[2] = out_len * 8;
+    w3[3] = 0;
+
+    u32x a = MD5M_A;
+    u32x b = MD5M_B;
+    u32x c = MD5M_C;
+    u32x d = MD5M_D;
+
+    MD5_STEP (MD5_Fo, a, b, c, d, w0[0], MD5C00, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w0[1], MD5C01, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w0[2], MD5C02, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w0[3], MD5C03, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w1[0], MD5C04, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w1[1], MD5C05, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w1[2], MD5C06, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w1[3], MD5C07, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w2[0], MD5C08, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w2[1], MD5C09, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w2[2], MD5C0a, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w2[3], MD5C0b, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w3[0], MD5C0c, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w3[1], MD5C0d, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w3[2], MD5C0e, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w3[3], MD5C0f, MD5S03);
+
+    MD5_STEP (MD5_Go, a, b, c, d, w0[1], MD5C10, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w1[2], MD5C11, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w2[3], MD5C12, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w0[0], MD5C13, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w1[1], MD5C14, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w2[2], MD5C15, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w3[3], MD5C16, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w1[0], MD5C17, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w2[1], MD5C18, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w3[2], MD5C19, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w0[3], MD5C1a, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w2[0], MD5C1b, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w3[1], MD5C1c, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w0[2], MD5C1d, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w1[3], MD5C1e, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w3[0], MD5C1f, MD5S13);
+
+    u32x t;
+
+    MD5_STEP (MD5_H1, a, b, c, d, w1[1], MD5C20, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w2[0], MD5C21, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w2[3], MD5C22, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w3[2], MD5C23, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w0[1], MD5C24, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w1[0], MD5C25, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w1[3], MD5C26, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w2[2], MD5C27, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w3[1], MD5C28, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w0[0], MD5C29, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w0[3], MD5C2a, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w1[2], MD5C2b, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w2[1], MD5C2c, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w3[0], MD5C2d, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w3[3], MD5C2e, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w0[2], MD5C2f, MD5S23);
+
+    MD5_STEP (MD5_I , a, b, c, d, w0[0], MD5C30, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w1[3], MD5C31, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w3[2], MD5C32, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w1[1], MD5C33, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w3[0], MD5C34, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w0[3], MD5C35, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w2[2], MD5C36, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w0[1], MD5C37, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w2[0], MD5C38, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w3[3], MD5C39, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w1[2], MD5C3a, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w3[1], MD5C3b, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w1[0], MD5C3c, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w2[3], MD5C3d, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w0[2], MD5C3e, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w2[1], MD5C3f, MD5S33);
+
+    a += make_u32x (MD5M_A);
+    b += make_u32x (MD5M_B);
+    c += make_u32x (MD5M_C);
+    d += make_u32x (MD5M_D);
+
+    w0[0] = uint_to_hex_lower8 ((a >>  0) & 255) <<  0
+          | uint_to_hex_lower8 ((a >>  8) & 255) << 16;
+    w0[1] = uint_to_hex_lower8 ((a >> 16) & 255) <<  0
+          | uint_to_hex_lower8 ((a >> 24) & 255) << 16;
+    w0[2] = uint_to_hex_lower8 ((b >>  0) & 255) <<  0
+          | uint_to_hex_lower8 ((b >>  8) & 255) << 16;
+    w0[3] = uint_to_hex_lower8 ((b >> 16) & 255) <<  0
+          | uint_to_hex_lower8 ((b >> 24) & 255) << 16;
+    w1[0] = uint_to_hex_lower8 ((c >>  0) & 255) <<  0
+          | uint_to_hex_lower8 ((c >>  8) & 255) << 16;
+    w1[1] = uint_to_hex_lower8 ((c >> 16) & 255) <<  0
+          | uint_to_hex_lower8 ((c >> 24) & 255) << 16;
+    w1[2] = uint_to_hex_lower8 ((d >>  0) & 255) <<  0
+          | uint_to_hex_lower8 ((d >>  8) & 255) << 16;
+    w1[3] = uint_to_hex_lower8 ((d >> 16) & 255) <<  0
+          | uint_to_hex_lower8 ((d >> 24) & 255) << 16;
+		  
+    w2[0] = 0x80;
+    w2[1] = 0;
+    w2[2] = 0;
+    w2[3] = 0;
+    w3[0] = 0;
+    w3[1] = 0;
+    w3[2] = 32 * 8;
+    w3[3] = 0;
+	
+    a = MD5M_A;
+    b = MD5M_B;
+    c = MD5M_C;
+    d = MD5M_D;
+
+    MD5_STEP (MD5_Fo, a, b, c, d, w0[0], MD5C00, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w0[1], MD5C01, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w0[2], MD5C02, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w0[3], MD5C03, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w1[0], MD5C04, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w1[1], MD5C05, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w1[2], MD5C06, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w1[3], MD5C07, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w2[0], MD5C08, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w2[1], MD5C09, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w2[2], MD5C0a, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w2[3], MD5C0b, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w3[0], MD5C0c, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w3[1], MD5C0d, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w3[2], MD5C0e, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w3[3], MD5C0f, MD5S03);
+
+    MD5_STEP (MD5_Go, a, b, c, d, w0[1], MD5C10, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w1[2], MD5C11, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w2[3], MD5C12, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w0[0], MD5C13, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w1[1], MD5C14, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w2[2], MD5C15, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w3[3], MD5C16, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w1[0], MD5C17, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w2[1], MD5C18, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w3[2], MD5C19, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w0[3], MD5C1a, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w2[0], MD5C1b, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w3[1], MD5C1c, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w0[2], MD5C1d, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w1[3], MD5C1e, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w3[0], MD5C1f, MD5S13);
+
+    MD5_STEP (MD5_H1, a, b, c, d, w1[1], MD5C20, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w2[0], MD5C21, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w2[3], MD5C22, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w3[2], MD5C23, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w0[1], MD5C24, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w1[0], MD5C25, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w1[3], MD5C26, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w2[2], MD5C27, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w3[1], MD5C28, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w0[0], MD5C29, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w0[3], MD5C2a, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w1[2], MD5C2b, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w2[1], MD5C2c, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w3[0], MD5C2d, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w3[3], MD5C2e, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w0[2], MD5C2f, MD5S23);
+
+    MD5_STEP (MD5_I , a, b, c, d, w0[0], MD5C30, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w1[3], MD5C31, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w3[2], MD5C32, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w1[1], MD5C33, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w3[0], MD5C34, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w0[3], MD5C35, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w2[2], MD5C36, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w0[1], MD5C37, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w2[0], MD5C38, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w3[3], MD5C39, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w1[2], MD5C3a, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w3[1], MD5C3b, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w1[0], MD5C3c, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w2[3], MD5C3d, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w0[2], MD5C3e, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w2[1], MD5C3f, MD5S33);
+
+    a += make_u32x (MD5M_A);
+    b += make_u32x (MD5M_B);
+    c += make_u32x (MD5M_C);
+    d += make_u32x (MD5M_D);
+
+    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] = uint_to_hex_lower8 ((a >>  0) & 255) <<  0
+          | uint_to_hex_lower8 ((a >>  8) & 255) << 16;
+    w2[1] = uint_to_hex_lower8 ((a >> 16) & 255) <<  0
+          | uint_to_hex_lower8 ((a >> 24) & 255) << 16;
+    w2[2] = uint_to_hex_lower8 ((b >>  0) & 255) <<  0
+          | uint_to_hex_lower8 ((b >>  8) & 255) << 16;
+    w2[3] = uint_to_hex_lower8 ((b >> 16) & 255) <<  0
+          | uint_to_hex_lower8 ((b >> 24) & 255) << 16;
+    w3[0] = uint_to_hex_lower8 ((c >>  0) & 255) <<  0
+          | uint_to_hex_lower8 ((c >>  8) & 255) << 16;
+    w3[1] = uint_to_hex_lower8 ((c >> 16) & 255) <<  0
+          | uint_to_hex_lower8 ((c >> 24) & 255) << 16;
+    w3[2] = uint_to_hex_lower8 ((d >>  0) & 255) <<  0
+          | uint_to_hex_lower8 ((d >>  8) & 255) << 16;
+    w3[3] = uint_to_hex_lower8 ((d >> 16) & 255) <<  0
+          | uint_to_hex_lower8 ((d >> 24) & 255) << 16;
+
+    a = MD5M_A;
+    b = MD5M_B;
+    c = MD5M_C;
+    d = MD5M_D;
+
+    MD5_STEP (MD5_Fo, a, b, c, d, w0[0], MD5C00, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w0[1], MD5C01, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w0[2], MD5C02, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w0[3], MD5C03, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w1[0], MD5C04, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w1[1], MD5C05, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w1[2], MD5C06, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w1[3], MD5C07, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w2[0], MD5C08, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w2[1], MD5C09, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w2[2], MD5C0a, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w2[3], MD5C0b, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w3[0], MD5C0c, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w3[1], MD5C0d, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w3[2], MD5C0e, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w3[3], MD5C0f, MD5S03);
+
+    MD5_STEP (MD5_Go, a, b, c, d, w0[1], MD5C10, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w1[2], MD5C11, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w2[3], MD5C12, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w0[0], MD5C13, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w1[1], MD5C14, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w2[2], MD5C15, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w3[3], MD5C16, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w1[0], MD5C17, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w2[1], MD5C18, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w3[2], MD5C19, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w0[3], MD5C1a, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w2[0], MD5C1b, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w3[1], MD5C1c, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w0[2], MD5C1d, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w1[3], MD5C1e, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w3[0], MD5C1f, MD5S13);
+
+    MD5_STEP (MD5_H1, a, b, c, d, w1[1], MD5C20, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w2[0], MD5C21, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w2[3], MD5C22, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w3[2], MD5C23, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w0[1], MD5C24, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w1[0], MD5C25, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w1[3], MD5C26, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w2[2], MD5C27, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w3[1], MD5C28, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w0[0], MD5C29, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w0[3], MD5C2a, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w1[2], MD5C2b, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w2[1], MD5C2c, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w3[0], MD5C2d, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w3[3], MD5C2e, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w0[2], MD5C2f, MD5S23);
+
+    MD5_STEP (MD5_I , a, b, c, d, w0[0], MD5C30, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w1[3], MD5C31, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w3[2], MD5C32, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w1[1], MD5C33, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w3[0], MD5C34, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w0[3], MD5C35, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w2[2], MD5C36, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w0[1], MD5C37, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w2[0], MD5C38, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w3[3], MD5C39, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w1[2], MD5C3a, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w3[1], MD5C3b, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w1[0], MD5C3c, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w2[3], MD5C3d, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w0[2], MD5C3e, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w2[1], MD5C3f, MD5S33);
+
+    const u32x r_a = a + make_u32x (MD5M_A);
+    const u32x r_b = b + make_u32x (MD5M_B);
+    const u32x r_c = c + make_u32x (MD5M_C);
+    const u32x r_d = d + make_u32x (MD5M_D);
+
+    const u32x r_00 = 0x80;
+    const u32x r_14 = 64 * 8;
+
+    a = r_a;
+    b = r_b;
+    c = r_c;
+    d = r_d;
+
+    MD5_STEP (MD5_Fo, a, b, c, d, r_00, MD5C00, MD5S00);
+    MD5_STEP0(MD5_Fo, d, a, b, c,       MD5C01, MD5S01);
+    MD5_STEP0(MD5_Fo, c, d, a, b,       MD5C02, MD5S02);
+    MD5_STEP0(MD5_Fo, b, c, d, a,       MD5C03, MD5S03);
+    MD5_STEP0(MD5_Fo, a, b, c, d,       MD5C04, MD5S00);
+    MD5_STEP0(MD5_Fo, d, a, b, c,       MD5C05, MD5S01);
+    MD5_STEP0(MD5_Fo, c, d, a, b,       MD5C06, MD5S02);
+    MD5_STEP0(MD5_Fo, b, c, d, a,       MD5C07, MD5S03);
+    MD5_STEP0(MD5_Fo, a, b, c, d,       MD5C08, MD5S00);
+    MD5_STEP0(MD5_Fo, d, a, b, c,       MD5C09, MD5S01);
+    MD5_STEP0(MD5_Fo, c, d, a, b,       MD5C0a, MD5S02);
+    MD5_STEP0(MD5_Fo, b, c, d, a,       MD5C0b, MD5S03);
+    MD5_STEP0(MD5_Fo, a, b, c, d,       MD5C0c, MD5S00);
+    MD5_STEP0(MD5_Fo, d, a, b, c,       MD5C0d, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, r_14, MD5C0e, MD5S02);
+    MD5_STEP0(MD5_Fo, b, c, d, a,       MD5C0f, MD5S03);
+
+    MD5_STEP0(MD5_Go, a, b, c, d,       MD5C10, MD5S10);
+    MD5_STEP0(MD5_Go, d, a, b, c,       MD5C11, MD5S11);
+    MD5_STEP0(MD5_Go, c, d, a, b,       MD5C12, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, r_00, MD5C13, MD5S13);
+    MD5_STEP0(MD5_Go, a, b, c, d,       MD5C14, MD5S10);
+    MD5_STEP0(MD5_Go, d, a, b, c,       MD5C15, MD5S11);
+    MD5_STEP0(MD5_Go, c, d, a, b,       MD5C16, MD5S12);
+    MD5_STEP0(MD5_Go, b, c, d, a,       MD5C17, MD5S13);
+    MD5_STEP0(MD5_Go, a, b, c, d,       MD5C18, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, r_14, MD5C19, MD5S11);
+    MD5_STEP0(MD5_Go, c, d, a, b,       MD5C1a, MD5S12);
+    MD5_STEP0(MD5_Go, b, c, d, a,       MD5C1b, MD5S13);
+    MD5_STEP0(MD5_Go, a, b, c, d,       MD5C1c, MD5S10);
+    MD5_STEP0(MD5_Go, d, a, b, c,       MD5C1d, MD5S11);
+    MD5_STEP0(MD5_Go, c, d, a, b,       MD5C1e, MD5S12);
+    MD5_STEP0(MD5_Go, b, c, d, a,       MD5C1f, MD5S13);
+
+    MD5_STEP0(MD5_H1, a, b, c, d,       MD5C20, MD5S20);
+    MD5_STEP0(MD5_H2, d, a, b, c,       MD5C21, MD5S21);
+    MD5_STEP0(MD5_H1, c, d, a, b,       MD5C22, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, r_14, MD5C23, MD5S23);
+    MD5_STEP0(MD5_H1, a, b, c, d,       MD5C24, MD5S20);
+    MD5_STEP0(MD5_H2, d, a, b, c,       MD5C25, MD5S21);
+    MD5_STEP0(MD5_H1, c, d, a, b,       MD5C26, MD5S22);
+    MD5_STEP0(MD5_H2, b, c, d, a,       MD5C27, MD5S23);
+    MD5_STEP0(MD5_H1, a, b, c, d,       MD5C28, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, r_00, MD5C29, MD5S21);
+    MD5_STEP0(MD5_H1, c, d, a, b,       MD5C2a, MD5S22);
+    MD5_STEP0(MD5_H2, b, c, d, a,       MD5C2b, MD5S23);
+    MD5_STEP0(MD5_H1, a, b, c, d,       MD5C2c, MD5S20);
+    MD5_STEP0(MD5_H2, d, a, b, c,       MD5C2d, MD5S21);
+    MD5_STEP0(MD5_H1, c, d, a, b,       MD5C2e, MD5S22);
+    MD5_STEP0(MD5_H2, b, c, d, a,       MD5C2f, MD5S23);
+
+    MD5_STEP (MD5_I , a, b, c, d, r_00, MD5C30, MD5S30);
+    MD5_STEP0(MD5_I , d, a, b, c,       MD5C31, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, r_14, MD5C32, MD5S32);
+    MD5_STEP0(MD5_I , b, c, d, a,       MD5C33, MD5S33);
+    MD5_STEP0(MD5_I , a, b, c, d,       MD5C34, MD5S30);
+    MD5_STEP0(MD5_I , d, a, b, c,       MD5C35, MD5S31);
+    MD5_STEP0(MD5_I , c, d, a, b,       MD5C36, MD5S32);
+    MD5_STEP0(MD5_I , b, c, d, a,       MD5C37, MD5S33);
+    MD5_STEP0(MD5_I , a, b, c, d,       MD5C38, MD5S30);
+    MD5_STEP0(MD5_I , d, a, b, c,       MD5C39, MD5S31);
+    MD5_STEP0(MD5_I , c, d, a, b,       MD5C3a, MD5S32);
+    MD5_STEP0(MD5_I , b, c, d, a,       MD5C3b, MD5S33);
+    MD5_STEP0(MD5_I , a, b, c, d,       MD5C3c, MD5S30);
+    MD5_STEP0(MD5_I , d, a, b, c,       MD5C3d, MD5S31);
+    MD5_STEP0(MD5_I , c, d, a, b,       MD5C3e, MD5S32);
+    MD5_STEP0(MD5_I , b, c, d, a,       MD5C3f, MD5S33);
+
+    a += r_a;
+    b += r_b;
+    c += r_c;
+    d += r_d;
+
+    COMPARE_M_SIMD (a, d, c, b);
+  }
+}
+
+KERNEL_FQ void m30500_m08 (KERN_ATTR_RULES ())
+{
+}
+
+KERNEL_FQ void m30500_m16 (KERN_ATTR_RULES ())
+{
+}
+
+KERNEL_FQ void m30500_s04 (KERN_ATTR_RULES ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * bin2asc table
+   */
+
+  LOCAL_VK u32 l_bin2asc[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    const u32 i0 = (i >> 0) & 15;
+    const u32 i1 = (i >> 4) & 15;
+
+    l_bin2asc[i] = ((i0 < 10) ? '0' + i0 : 'a' - 10 + i0) << 8
+                 | ((i1 < 10) ? '0' + i1 : 'a' - 10 + i1) << 0;
+  }
+
+  SYNC_THREADS ();
+
+  if (gid >= GID_CNT) return;
+
+  /**
+   * base
+   */
+
+  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];
+  u32 salt_buf1[4];
+  u32 salt_buf2[4];
+  u32 salt_buf3[4];
+
+  salt_buf0[0] = salt_bufs[SALT_POS_HOST].salt_buf_pc[0];
+  salt_buf0[1] = salt_bufs[SALT_POS_HOST].salt_buf_pc[1];
+  salt_buf0[2] = salt_bufs[SALT_POS_HOST].salt_buf_pc[2];
+  salt_buf0[3] = salt_bufs[SALT_POS_HOST].salt_buf_pc[3];
+  salt_buf1[0] = salt_bufs[SALT_POS_HOST].salt_buf_pc[4];
+  salt_buf1[1] = salt_bufs[SALT_POS_HOST].salt_buf_pc[5];
+  salt_buf1[2] = salt_bufs[SALT_POS_HOST].salt_buf_pc[6];
+  salt_buf1[3] = salt_bufs[SALT_POS_HOST].salt_buf_pc[7];
+  salt_buf2[0] = 0;
+  salt_buf2[1] = 0;
+  salt_buf2[2] = 0;
+  salt_buf2[3] = 0;
+  salt_buf3[0] = 0;
+  salt_buf3[1] = 0;
+  salt_buf3[2] = 0;
+  salt_buf3[3] = 0;
+
+  /**
+   * digest
+   */
+
+  const u32 search[4] =
+  {
+    digests_buf[DIGESTS_OFFSET_HOST].digest_buf[DGST_R0],
+    digests_buf[DIGESTS_OFFSET_HOST].digest_buf[DGST_R1],
+    digests_buf[DIGESTS_OFFSET_HOST].digest_buf[DGST_R2],
+    digests_buf[DIGESTS_OFFSET_HOST].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);
+
+    append_0x80_2x4_VV (w0, w1, out_len);
+
+    w3[2] = out_len * 8;
+    w3[3] = 0;
+
+    u32x a = MD5M_A;
+    u32x b = MD5M_B;
+    u32x c = MD5M_C;
+    u32x d = MD5M_D;
+
+    MD5_STEP (MD5_Fo, a, b, c, d, w0[0], MD5C00, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w0[1], MD5C01, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w0[2], MD5C02, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w0[3], MD5C03, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w1[0], MD5C04, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w1[1], MD5C05, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w1[2], MD5C06, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w1[3], MD5C07, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w2[0], MD5C08, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w2[1], MD5C09, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w2[2], MD5C0a, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w2[3], MD5C0b, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w3[0], MD5C0c, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w3[1], MD5C0d, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w3[2], MD5C0e, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w3[3], MD5C0f, MD5S03);
+
+    MD5_STEP (MD5_Go, a, b, c, d, w0[1], MD5C10, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w1[2], MD5C11, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w2[3], MD5C12, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w0[0], MD5C13, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w1[1], MD5C14, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w2[2], MD5C15, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w3[3], MD5C16, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w1[0], MD5C17, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w2[1], MD5C18, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w3[2], MD5C19, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w0[3], MD5C1a, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w2[0], MD5C1b, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w3[1], MD5C1c, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w0[2], MD5C1d, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w1[3], MD5C1e, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w3[0], MD5C1f, MD5S13);
+
+    u32x t;
+
+    MD5_STEP (MD5_H1, a, b, c, d, w1[1], MD5C20, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w2[0], MD5C21, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w2[3], MD5C22, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w3[2], MD5C23, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w0[1], MD5C24, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w1[0], MD5C25, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w1[3], MD5C26, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w2[2], MD5C27, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w3[1], MD5C28, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w0[0], MD5C29, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w0[3], MD5C2a, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w1[2], MD5C2b, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w2[1], MD5C2c, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w3[0], MD5C2d, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w3[3], MD5C2e, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w0[2], MD5C2f, MD5S23);
+
+    MD5_STEP (MD5_I , a, b, c, d, w0[0], MD5C30, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w1[3], MD5C31, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w3[2], MD5C32, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w1[1], MD5C33, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w3[0], MD5C34, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w0[3], MD5C35, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w2[2], MD5C36, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w0[1], MD5C37, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w2[0], MD5C38, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w3[3], MD5C39, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w1[2], MD5C3a, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w3[1], MD5C3b, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w1[0], MD5C3c, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w2[3], MD5C3d, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w0[2], MD5C3e, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w2[1], MD5C3f, MD5S33);
+
+    a += make_u32x (MD5M_A);
+    b += make_u32x (MD5M_B);
+    c += make_u32x (MD5M_C);
+    d += make_u32x (MD5M_D);
+
+    w0[0] = uint_to_hex_lower8 ((a >>  0) & 255) <<  0
+          | uint_to_hex_lower8 ((a >>  8) & 255) << 16;
+    w0[1] = uint_to_hex_lower8 ((a >> 16) & 255) <<  0
+          | uint_to_hex_lower8 ((a >> 24) & 255) << 16;
+    w0[2] = uint_to_hex_lower8 ((b >>  0) & 255) <<  0
+          | uint_to_hex_lower8 ((b >>  8) & 255) << 16;
+    w0[3] = uint_to_hex_lower8 ((b >> 16) & 255) <<  0
+          | uint_to_hex_lower8 ((b >> 24) & 255) << 16;
+    w1[0] = uint_to_hex_lower8 ((c >>  0) & 255) <<  0
+          | uint_to_hex_lower8 ((c >>  8) & 255) << 16;
+    w1[1] = uint_to_hex_lower8 ((c >> 16) & 255) <<  0
+          | uint_to_hex_lower8 ((c >> 24) & 255) << 16;
+    w1[2] = uint_to_hex_lower8 ((d >>  0) & 255) <<  0
+          | uint_to_hex_lower8 ((d >>  8) & 255) << 16;
+    w1[3] = uint_to_hex_lower8 ((d >> 16) & 255) <<  0
+          | uint_to_hex_lower8 ((d >> 24) & 255) << 16;
+		  
+    w2[0] = 0x80;
+    w2[1] = 0;
+    w2[2] = 0;
+    w2[3] = 0;
+    w3[0] = 0;
+    w3[1] = 0;
+    w3[2] = 32 * 8;
+    w3[3] = 0;
+	
+    a = MD5M_A;
+    b = MD5M_B;
+    c = MD5M_C;
+    d = MD5M_D;
+
+    MD5_STEP (MD5_Fo, a, b, c, d, w0[0], MD5C00, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w0[1], MD5C01, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w0[2], MD5C02, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w0[3], MD5C03, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w1[0], MD5C04, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w1[1], MD5C05, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w1[2], MD5C06, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w1[3], MD5C07, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w2[0], MD5C08, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w2[1], MD5C09, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w2[2], MD5C0a, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w2[3], MD5C0b, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w3[0], MD5C0c, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w3[1], MD5C0d, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w3[2], MD5C0e, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w3[3], MD5C0f, MD5S03);
+
+    MD5_STEP (MD5_Go, a, b, c, d, w0[1], MD5C10, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w1[2], MD5C11, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w2[3], MD5C12, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w0[0], MD5C13, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w1[1], MD5C14, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w2[2], MD5C15, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w3[3], MD5C16, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w1[0], MD5C17, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w2[1], MD5C18, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w3[2], MD5C19, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w0[3], MD5C1a, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w2[0], MD5C1b, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w3[1], MD5C1c, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w0[2], MD5C1d, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w1[3], MD5C1e, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w3[0], MD5C1f, MD5S13);
+
+    MD5_STEP (MD5_H1, a, b, c, d, w1[1], MD5C20, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w2[0], MD5C21, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w2[3], MD5C22, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w3[2], MD5C23, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w0[1], MD5C24, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w1[0], MD5C25, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w1[3], MD5C26, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w2[2], MD5C27, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w3[1], MD5C28, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w0[0], MD5C29, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w0[3], MD5C2a, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w1[2], MD5C2b, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w2[1], MD5C2c, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w3[0], MD5C2d, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w3[3], MD5C2e, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w0[2], MD5C2f, MD5S23);
+
+    MD5_STEP (MD5_I , a, b, c, d, w0[0], MD5C30, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w1[3], MD5C31, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w3[2], MD5C32, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w1[1], MD5C33, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w3[0], MD5C34, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w0[3], MD5C35, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w2[2], MD5C36, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w0[1], MD5C37, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w2[0], MD5C38, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w3[3], MD5C39, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w1[2], MD5C3a, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w3[1], MD5C3b, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w1[0], MD5C3c, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w2[3], MD5C3d, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w0[2], MD5C3e, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w2[1], MD5C3f, MD5S33);
+
+    a += make_u32x (MD5M_A);
+    b += make_u32x (MD5M_B);
+    c += make_u32x (MD5M_C);
+    d += make_u32x (MD5M_D);
+
+    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] = uint_to_hex_lower8 ((a >>  0) & 255) <<  0
+          | uint_to_hex_lower8 ((a >>  8) & 255) << 16;
+    w2[1] = uint_to_hex_lower8 ((a >> 16) & 255) <<  0
+          | uint_to_hex_lower8 ((a >> 24) & 255) << 16;
+    w2[2] = uint_to_hex_lower8 ((b >>  0) & 255) <<  0
+          | uint_to_hex_lower8 ((b >>  8) & 255) << 16;
+    w2[3] = uint_to_hex_lower8 ((b >> 16) & 255) <<  0
+          | uint_to_hex_lower8 ((b >> 24) & 255) << 16;
+    w3[0] = uint_to_hex_lower8 ((c >>  0) & 255) <<  0
+          | uint_to_hex_lower8 ((c >>  8) & 255) << 16;
+    w3[1] = uint_to_hex_lower8 ((c >> 16) & 255) <<  0
+          | uint_to_hex_lower8 ((c >> 24) & 255) << 16;
+    w3[2] = uint_to_hex_lower8 ((d >>  0) & 255) <<  0
+          | uint_to_hex_lower8 ((d >>  8) & 255) << 16;
+    w3[3] = uint_to_hex_lower8 ((d >> 16) & 255) <<  0
+          | uint_to_hex_lower8 ((d >> 24) & 255) << 16;
+
+    a = MD5M_A;
+    b = MD5M_B;
+    c = MD5M_C;
+    d = MD5M_D;
+
+    MD5_STEP (MD5_Fo, a, b, c, d, w0[0], MD5C00, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w0[1], MD5C01, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w0[2], MD5C02, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w0[3], MD5C03, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w1[0], MD5C04, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w1[1], MD5C05, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w1[2], MD5C06, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w1[3], MD5C07, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w2[0], MD5C08, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w2[1], MD5C09, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w2[2], MD5C0a, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w2[3], MD5C0b, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w3[0], MD5C0c, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w3[1], MD5C0d, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w3[2], MD5C0e, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w3[3], MD5C0f, MD5S03);
+
+    MD5_STEP (MD5_Go, a, b, c, d, w0[1], MD5C10, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w1[2], MD5C11, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w2[3], MD5C12, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w0[0], MD5C13, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w1[1], MD5C14, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w2[2], MD5C15, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w3[3], MD5C16, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w1[0], MD5C17, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w2[1], MD5C18, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w3[2], MD5C19, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w0[3], MD5C1a, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w2[0], MD5C1b, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w3[1], MD5C1c, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w0[2], MD5C1d, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w1[3], MD5C1e, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w3[0], MD5C1f, MD5S13);
+
+    MD5_STEP (MD5_H1, a, b, c, d, w1[1], MD5C20, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w2[0], MD5C21, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w2[3], MD5C22, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w3[2], MD5C23, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w0[1], MD5C24, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w1[0], MD5C25, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w1[3], MD5C26, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w2[2], MD5C27, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w3[1], MD5C28, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w0[0], MD5C29, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w0[3], MD5C2a, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w1[2], MD5C2b, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w2[1], MD5C2c, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w3[0], MD5C2d, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w3[3], MD5C2e, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w0[2], MD5C2f, MD5S23);
+
+    MD5_STEP (MD5_I , a, b, c, d, w0[0], MD5C30, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w1[3], MD5C31, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w3[2], MD5C32, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w1[1], MD5C33, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w3[0], MD5C34, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w0[3], MD5C35, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w2[2], MD5C36, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w0[1], MD5C37, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w2[0], MD5C38, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w3[3], MD5C39, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w1[2], MD5C3a, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w3[1], MD5C3b, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w1[0], MD5C3c, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w2[3], MD5C3d, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w0[2], MD5C3e, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w2[1], MD5C3f, MD5S33);
+
+    const u32x r_a = a + make_u32x (MD5M_A);
+    const u32x r_b = b + make_u32x (MD5M_B);
+    const u32x r_c = c + make_u32x (MD5M_C);
+    const u32x r_d = d + make_u32x (MD5M_D);
+
+    const u32x r_00 = 0x80;
+    const u32x r_14 = 64 * 8;
+
+    a = r_a;
+    b = r_b;
+    c = r_c;
+    d = r_d;
+
+    MD5_STEP (MD5_Fo, a, b, c, d, r_00, MD5C00, MD5S00);
+    MD5_STEP0(MD5_Fo, d, a, b, c,       MD5C01, MD5S01);
+    MD5_STEP0(MD5_Fo, c, d, a, b,       MD5C02, MD5S02);
+    MD5_STEP0(MD5_Fo, b, c, d, a,       MD5C03, MD5S03);
+    MD5_STEP0(MD5_Fo, a, b, c, d,       MD5C04, MD5S00);
+    MD5_STEP0(MD5_Fo, d, a, b, c,       MD5C05, MD5S01);
+    MD5_STEP0(MD5_Fo, c, d, a, b,       MD5C06, MD5S02);
+    MD5_STEP0(MD5_Fo, b, c, d, a,       MD5C07, MD5S03);
+    MD5_STEP0(MD5_Fo, a, b, c, d,       MD5C08, MD5S00);
+    MD5_STEP0(MD5_Fo, d, a, b, c,       MD5C09, MD5S01);
+    MD5_STEP0(MD5_Fo, c, d, a, b,       MD5C0a, MD5S02);
+    MD5_STEP0(MD5_Fo, b, c, d, a,       MD5C0b, MD5S03);
+    MD5_STEP0(MD5_Fo, a, b, c, d,       MD5C0c, MD5S00);
+    MD5_STEP0(MD5_Fo, d, a, b, c,       MD5C0d, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, r_14, MD5C0e, MD5S02);
+    MD5_STEP0(MD5_Fo, b, c, d, a,       MD5C0f, MD5S03);
+
+    MD5_STEP0(MD5_Go, a, b, c, d,       MD5C10, MD5S10);
+    MD5_STEP0(MD5_Go, d, a, b, c,       MD5C11, MD5S11);
+    MD5_STEP0(MD5_Go, c, d, a, b,       MD5C12, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, r_00, MD5C13, MD5S13);
+    MD5_STEP0(MD5_Go, a, b, c, d,       MD5C14, MD5S10);
+    MD5_STEP0(MD5_Go, d, a, b, c,       MD5C15, MD5S11);
+    MD5_STEP0(MD5_Go, c, d, a, b,       MD5C16, MD5S12);
+    MD5_STEP0(MD5_Go, b, c, d, a,       MD5C17, MD5S13);
+    MD5_STEP0(MD5_Go, a, b, c, d,       MD5C18, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, r_14, MD5C19, MD5S11);
+    MD5_STEP0(MD5_Go, c, d, a, b,       MD5C1a, MD5S12);
+    MD5_STEP0(MD5_Go, b, c, d, a,       MD5C1b, MD5S13);
+    MD5_STEP0(MD5_Go, a, b, c, d,       MD5C1c, MD5S10);
+    MD5_STEP0(MD5_Go, d, a, b, c,       MD5C1d, MD5S11);
+    MD5_STEP0(MD5_Go, c, d, a, b,       MD5C1e, MD5S12);
+    MD5_STEP0(MD5_Go, b, c, d, a,       MD5C1f, MD5S13);
+
+    MD5_STEP0(MD5_H1, a, b, c, d,       MD5C20, MD5S20);
+    MD5_STEP0(MD5_H2, d, a, b, c,       MD5C21, MD5S21);
+    MD5_STEP0(MD5_H1, c, d, a, b,       MD5C22, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, r_14, MD5C23, MD5S23);
+    MD5_STEP0(MD5_H1, a, b, c, d,       MD5C24, MD5S20);
+    MD5_STEP0(MD5_H2, d, a, b, c,       MD5C25, MD5S21);
+    MD5_STEP0(MD5_H1, c, d, a, b,       MD5C26, MD5S22);
+    MD5_STEP0(MD5_H2, b, c, d, a,       MD5C27, MD5S23);
+    MD5_STEP0(MD5_H1, a, b, c, d,       MD5C28, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, r_00, MD5C29, MD5S21);
+    MD5_STEP0(MD5_H1, c, d, a, b,       MD5C2a, MD5S22);
+    MD5_STEP0(MD5_H2, b, c, d, a,       MD5C2b, MD5S23);
+    MD5_STEP0(MD5_H1, a, b, c, d,       MD5C2c, MD5S20);
+    MD5_STEP0(MD5_H2, d, a, b, c,       MD5C2d, MD5S21);
+    MD5_STEP0(MD5_H1, c, d, a, b,       MD5C2e, MD5S22);
+    MD5_STEP0(MD5_H2, b, c, d, a,       MD5C2f, MD5S23);
+
+    MD5_STEP (MD5_I , a, b, c, d, r_00, MD5C30, MD5S30);
+    MD5_STEP0(MD5_I , d, a, b, c,       MD5C31, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, r_14, MD5C32, MD5S32);
+    MD5_STEP0(MD5_I , b, c, d, a,       MD5C33, MD5S33);
+    MD5_STEP0(MD5_I , a, b, c, d,       MD5C34, MD5S30);
+    MD5_STEP0(MD5_I , d, a, b, c,       MD5C35, MD5S31);
+    MD5_STEP0(MD5_I , c, d, a, b,       MD5C36, MD5S32);
+    MD5_STEP0(MD5_I , b, c, d, a,       MD5C37, MD5S33);
+    MD5_STEP0(MD5_I , a, b, c, d,       MD5C38, MD5S30);
+    MD5_STEP0(MD5_I , d, a, b, c,       MD5C39, MD5S31);
+    MD5_STEP0(MD5_I , c, d, a, b,       MD5C3a, MD5S32);
+    MD5_STEP0(MD5_I , b, c, d, a,       MD5C3b, MD5S33);
+    MD5_STEP0(MD5_I , a, b, c, d,       MD5C3c, MD5S30);
+
+    if (MATCHES_NONE_VS ((a + r_a), search[0])) continue;
+
+    MD5_STEP0(MD5_I , d, a, b, c,       MD5C3d, MD5S31);
+    MD5_STEP0(MD5_I , c, d, a, b,       MD5C3e, MD5S32);
+    MD5_STEP0(MD5_I , b, c, d, a,       MD5C3f, MD5S33);
+
+    a += r_a;
+    b += r_b;
+    c += r_c;
+    d += r_d;
+
+    COMPARE_S_SIMD (a, d, c, b);
+  }
+}
+
+KERNEL_FQ void m30500_s08 (KERN_ATTR_RULES ())
+{
+}
+
+KERNEL_FQ void m30500_s16 (KERN_ATTR_RULES ())
+{
+}
diff --git a/OpenCL/m30500_a0-pure.cl b/OpenCL/m30500_a0-pure.cl
new file mode 100644
index 000000000..e78e6c708
--- /dev/null
+++ b/OpenCL/m30500_a0-pure.cl
@@ -0,0 +1,353 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+//#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include M2S(INCLUDE_PATH/inc_vendor.h)
+#include M2S(INCLUDE_PATH/inc_types.h)
+#include M2S(INCLUDE_PATH/inc_platform.cl)
+#include M2S(INCLUDE_PATH/inc_common.cl)
+#include M2S(INCLUDE_PATH/inc_rp.h)
+#include M2S(INCLUDE_PATH/inc_rp.cl)
+#include M2S(INCLUDE_PATH/inc_scalar.cl)
+#include M2S(INCLUDE_PATH/inc_hash_md5.cl)
+#endif
+
+#if   VECT_SIZE == 1
+#define uint_to_hex_lower8(i) make_u32x (l_bin2asc[(i)])
+#elif VECT_SIZE == 2
+#define uint_to_hex_lower8(i) make_u32x (l_bin2asc[(i).s0], l_bin2asc[(i).s1])
+#elif VECT_SIZE == 4
+#define uint_to_hex_lower8(i) make_u32x (l_bin2asc[(i).s0], l_bin2asc[(i).s1], l_bin2asc[(i).s2], l_bin2asc[(i).s3])
+#elif VECT_SIZE == 8
+#define uint_to_hex_lower8(i) make_u32x (l_bin2asc[(i).s0], l_bin2asc[(i).s1], l_bin2asc[(i).s2], l_bin2asc[(i).s3], l_bin2asc[(i).s4], l_bin2asc[(i).s5], l_bin2asc[(i).s6], l_bin2asc[(i).s7])
+#elif VECT_SIZE == 16
+#define uint_to_hex_lower8(i) make_u32x (l_bin2asc[(i).s0], l_bin2asc[(i).s1], l_bin2asc[(i).s2], l_bin2asc[(i).s3], l_bin2asc[(i).s4], l_bin2asc[(i).s5], l_bin2asc[(i).s6], l_bin2asc[(i).s7], l_bin2asc[(i).s8], l_bin2asc[(i).s9], l_bin2asc[(i).sa], l_bin2asc[(i).sb], l_bin2asc[(i).sc], l_bin2asc[(i).sd], l_bin2asc[(i).se], l_bin2asc[(i).sf])
+#endif
+
+KERNEL_FQ void m30500_mxx (KERN_ATTR_RULES ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * bin2asc table
+   */
+
+  LOCAL_VK u32 l_bin2asc[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    const u32 i0 = (i >> 0) & 15;
+    const u32 i1 = (i >> 4) & 15;
+
+    l_bin2asc[i] = ((i0 < 10) ? '0' + i0 : 'a' - 10 + i0) << 8
+                 | ((i1 < 10) ? '0' + i1 : 'a' - 10 + i1) << 0;
+  }
+
+  SYNC_THREADS ();
+
+  if (gid >= GID_CNT) return;
+
+  /**
+   * base
+   */
+
+  COPY_PW (pws[gid]);
+
+  const u32 salt_len = 32;
+
+  u32 s[8] = { 0 };
+
+  for (u32 i = 0, idx = 0; i < salt_len; i += 4, idx += 1)
+  {
+    s[idx] = salt_bufs[SALT_POS_HOST].salt_buf_pc[idx];
+  }
+
+  /**
+   * 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);
+
+    md5_ctx_t ctx0;
+
+    md5_init (&ctx0);
+
+    md5_update (&ctx0, tmp.i, tmp.pw_len);
+
+    md5_final (&ctx0);
+
+    u32 a = ctx0.h[0];
+    u32 b = ctx0.h[1];
+    u32 c = ctx0.h[2];
+    u32 d = ctx0.h[3];
+
+    md5_ctx_t ctx;
+
+    md5_init (&ctx);
+
+    ctx.w0[0] = uint_to_hex_lower8 ((a >>  0) & 255) <<  0
+              | uint_to_hex_lower8 ((a >>  8) & 255) << 16;
+    ctx.w0[1] = uint_to_hex_lower8 ((a >> 16) & 255) <<  0
+              | uint_to_hex_lower8 ((a >> 24) & 255) << 16;
+    ctx.w0[2] = uint_to_hex_lower8 ((b >>  0) & 255) <<  0
+              | uint_to_hex_lower8 ((b >>  8) & 255) << 16;
+    ctx.w0[3] = uint_to_hex_lower8 ((b >> 16) & 255) <<  0
+              | uint_to_hex_lower8 ((b >> 24) & 255) << 16;
+    ctx.w1[0] = uint_to_hex_lower8 ((c >>  0) & 255) <<  0
+              | uint_to_hex_lower8 ((c >>  8) & 255) << 16;
+    ctx.w1[1] = uint_to_hex_lower8 ((c >> 16) & 255) <<  0
+              | uint_to_hex_lower8 ((c >> 24) & 255) << 16;
+    ctx.w1[2] = uint_to_hex_lower8 ((d >>  0) & 255) <<  0
+              | uint_to_hex_lower8 ((d >>  8) & 255) << 16;
+    ctx.w1[3] = uint_to_hex_lower8 ((d >> 16) & 255) <<  0
+              | uint_to_hex_lower8 ((d >> 24) & 255) << 16;
+
+    ctx.len = 32;
+
+    md5_final (&ctx);
+
+    a = ctx.h[0];
+    b = ctx.h[1];
+    c = ctx.h[2];
+    d = ctx.h[3];
+
+    md5_ctx_t ctx1;
+
+    md5_init (&ctx1);
+
+    ctx1.w0[0] = s[0];
+    ctx1.w0[1] = s[1];
+    ctx1.w0[2] = s[2];
+    ctx1.w0[3] = s[3];
+    ctx1.w1[0] = s[4];
+    ctx1.w1[1] = s[5];
+    ctx1.w1[2] = s[6];
+    ctx1.w1[3] = s[7];
+    ctx1.w2[0] = uint_to_hex_lower8 ((a >>  0) & 255) <<  0
+              | uint_to_hex_lower8 ((a >>  8) & 255) << 16;
+    ctx1.w2[1] = uint_to_hex_lower8 ((a >> 16) & 255) <<  0
+              | uint_to_hex_lower8 ((a >> 24) & 255) << 16;
+    ctx1.w2[2] = uint_to_hex_lower8 ((b >>  0) & 255) <<  0
+              | uint_to_hex_lower8 ((b >>  8) & 255) << 16;
+    ctx1.w2[3] = uint_to_hex_lower8 ((b >> 16) & 255) <<  0
+              | uint_to_hex_lower8 ((b >> 24) & 255) << 16;
+    ctx1.w3[0] = uint_to_hex_lower8 ((c >>  0) & 255) <<  0
+              | uint_to_hex_lower8 ((c >>  8) & 255) << 16;
+    ctx1.w3[1] = uint_to_hex_lower8 ((c >> 16) & 255) <<  0
+              | uint_to_hex_lower8 ((c >> 24) & 255) << 16;
+    ctx1.w3[2] = uint_to_hex_lower8 ((d >>  0) & 255) <<  0
+              | uint_to_hex_lower8 ((d >>  8) & 255) << 16;
+    ctx1.w3[3] = uint_to_hex_lower8 ((d >> 16) & 255) <<  0
+              | uint_to_hex_lower8 ((d >> 24) & 255) << 16;
+
+    md5_transform (ctx1.w0, ctx1.w1, ctx1.w2, ctx1.w3, ctx1.h);
+
+    ctx1.w0[0] = 0x80;
+    ctx1.w0[1] = 0;
+    ctx1.w0[2] = 0;
+    ctx1.w0[3] = 0;
+    ctx1.w1[0] = 0;
+    ctx1.w1[1] = 0;
+    ctx1.w1[2] = 0;
+    ctx1.w1[3] = 0;
+    ctx1.w2[0] = 0;
+    ctx1.w2[1] = 0;
+    ctx1.w2[2] = 0;
+    ctx1.w2[3] = 0;
+    ctx1.w3[0] = 0;
+    ctx1.w3[1] = 0;
+    ctx1.w3[2] = 64 * 8;
+    ctx1.w3[3] = 0;
+
+    md5_transform (ctx1.w0, ctx1.w1, ctx1.w2, ctx1.w3, ctx1.h);
+
+    const u32 r0 = ctx1.h[DGST_R0];
+    const u32 r1 = ctx1.h[DGST_R1];
+    const u32 r2 = ctx1.h[DGST_R2];
+    const u32 r3 = ctx1.h[DGST_R3];
+
+    COMPARE_M_SCALAR (r0, r1, r2, r3);
+  }
+}
+
+KERNEL_FQ void m30500_sxx (KERN_ATTR_RULES ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * bin2asc table
+   */
+
+  LOCAL_VK u32 l_bin2asc[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    const u32 i0 = (i >> 0) & 15;
+    const u32 i1 = (i >> 4) & 15;
+
+    l_bin2asc[i] = ((i0 < 10) ? '0' + i0 : 'a' - 10 + i0) << 8
+                 | ((i1 < 10) ? '0' + i1 : 'a' - 10 + i1) << 0;
+  }
+
+  SYNC_THREADS ();
+
+  if (gid >= GID_CNT) return;
+
+  /**
+   * digest
+   */
+
+  const u32 search[4] =
+  {
+    digests_buf[DIGESTS_OFFSET_HOST].digest_buf[DGST_R0],
+    digests_buf[DIGESTS_OFFSET_HOST].digest_buf[DGST_R1],
+    digests_buf[DIGESTS_OFFSET_HOST].digest_buf[DGST_R2],
+    digests_buf[DIGESTS_OFFSET_HOST].digest_buf[DGST_R3]
+  };
+
+  /**
+   * base
+   */
+
+  COPY_PW (pws[gid]);
+
+  const u32 salt_len = 32;
+
+  u32 s[8] = { 0 };
+
+  for (u32 i = 0, idx = 0; i < salt_len; i += 4, idx += 1)
+  {
+    s[idx] = salt_bufs[SALT_POS_HOST].salt_buf_pc[idx];
+  }
+
+  /**
+   * 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);
+
+    md5_ctx_t ctx0;
+
+    md5_init (&ctx0);
+
+    md5_update (&ctx0, tmp.i, tmp.pw_len);
+
+    md5_final (&ctx0);
+
+    u32 a = ctx0.h[0];
+    u32 b = ctx0.h[1];
+    u32 c = ctx0.h[2];
+    u32 d = ctx0.h[3];
+
+    md5_ctx_t ctx;
+
+    md5_init (&ctx);
+
+    ctx.w0[0] = uint_to_hex_lower8 ((a >>  0) & 255) <<  0
+              | uint_to_hex_lower8 ((a >>  8) & 255) << 16;
+    ctx.w0[1] = uint_to_hex_lower8 ((a >> 16) & 255) <<  0
+              | uint_to_hex_lower8 ((a >> 24) & 255) << 16;
+    ctx.w0[2] = uint_to_hex_lower8 ((b >>  0) & 255) <<  0
+              | uint_to_hex_lower8 ((b >>  8) & 255) << 16;
+    ctx.w0[3] = uint_to_hex_lower8 ((b >> 16) & 255) <<  0
+              | uint_to_hex_lower8 ((b >> 24) & 255) << 16;
+    ctx.w1[0] = uint_to_hex_lower8 ((c >>  0) & 255) <<  0
+              | uint_to_hex_lower8 ((c >>  8) & 255) << 16;
+    ctx.w1[1] = uint_to_hex_lower8 ((c >> 16) & 255) <<  0
+              | uint_to_hex_lower8 ((c >> 24) & 255) << 16;
+    ctx.w1[2] = uint_to_hex_lower8 ((d >>  0) & 255) <<  0
+              | uint_to_hex_lower8 ((d >>  8) & 255) << 16;
+    ctx.w1[3] = uint_to_hex_lower8 ((d >> 16) & 255) <<  0
+              | uint_to_hex_lower8 ((d >> 24) & 255) << 16;
+
+    ctx.len = 32;
+
+    md5_final (&ctx);
+
+    a = ctx.h[0];
+    b = ctx.h[1];
+    c = ctx.h[2];
+    d = ctx.h[3];
+
+    md5_ctx_t ctx1;
+
+    md5_init (&ctx1);
+
+    ctx1.w0[0] = s[0];
+    ctx1.w0[1] = s[1];
+    ctx1.w0[2] = s[2];
+    ctx1.w0[3] = s[3];
+    ctx1.w1[0] = s[4];
+    ctx1.w1[1] = s[5];
+    ctx1.w1[2] = s[6];
+    ctx1.w1[3] = s[7];
+    ctx1.w2[0] = uint_to_hex_lower8 ((a >>  0) & 255) <<  0
+              | uint_to_hex_lower8 ((a >>  8) & 255) << 16;
+    ctx1.w2[1] = uint_to_hex_lower8 ((a >> 16) & 255) <<  0
+              | uint_to_hex_lower8 ((a >> 24) & 255) << 16;
+    ctx1.w2[2] = uint_to_hex_lower8 ((b >>  0) & 255) <<  0
+              | uint_to_hex_lower8 ((b >>  8) & 255) << 16;
+    ctx1.w2[3] = uint_to_hex_lower8 ((b >> 16) & 255) <<  0
+              | uint_to_hex_lower8 ((b >> 24) & 255) << 16;
+    ctx1.w3[0] = uint_to_hex_lower8 ((c >>  0) & 255) <<  0
+              | uint_to_hex_lower8 ((c >>  8) & 255) << 16;
+    ctx1.w3[1] = uint_to_hex_lower8 ((c >> 16) & 255) <<  0
+              | uint_to_hex_lower8 ((c >> 24) & 255) << 16;
+    ctx1.w3[2] = uint_to_hex_lower8 ((d >>  0) & 255) <<  0
+              | uint_to_hex_lower8 ((d >>  8) & 255) << 16;
+    ctx1.w3[3] = uint_to_hex_lower8 ((d >> 16) & 255) <<  0
+              | uint_to_hex_lower8 ((d >> 24) & 255) << 16;
+
+    md5_transform (ctx1.w0, ctx1.w1, ctx1.w2, ctx1.w3, ctx1.h);
+
+    ctx1.w0[0] = 0x80;
+    ctx1.w0[1] = 0;
+    ctx1.w0[2] = 0;
+    ctx1.w0[3] = 0;
+    ctx1.w1[0] = 0;
+    ctx1.w1[1] = 0;
+    ctx1.w1[2] = 0;
+    ctx1.w1[3] = 0;
+    ctx1.w2[0] = 0;
+    ctx1.w2[1] = 0;
+    ctx1.w2[2] = 0;
+    ctx1.w2[3] = 0;
+    ctx1.w3[0] = 0;
+    ctx1.w3[1] = 0;
+    ctx1.w3[2] = 64 * 8;
+    ctx1.w3[3] = 0;
+
+    md5_transform (ctx1.w0, ctx1.w1, ctx1.w2, ctx1.w3, ctx1.h);
+
+    const u32 r0 = ctx1.h[DGST_R0];
+    const u32 r1 = ctx1.h[DGST_R1];
+    const u32 r2 = ctx1.h[DGST_R2];
+    const u32 r3 = ctx1.h[DGST_R3];
+
+    COMPARE_S_SCALAR (r0, r1, r2, r3);
+  }
+}
diff --git a/OpenCL/m30500_a1-optimized.cl b/OpenCL/m30500_a1-optimized.cl
new file mode 100644
index 000000000..d4ff6f1cb
--- /dev/null
+++ b/OpenCL/m30500_a1-optimized.cl
@@ -0,0 +1,1104 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include M2S(INCLUDE_PATH/inc_vendor.h)
+#include M2S(INCLUDE_PATH/inc_types.h)
+#include M2S(INCLUDE_PATH/inc_platform.cl)
+#include M2S(INCLUDE_PATH/inc_common.cl)
+#include M2S(INCLUDE_PATH/inc_simd.cl)
+#include M2S(INCLUDE_PATH/inc_hash_md5.cl)
+#endif
+
+#if   VECT_SIZE == 1
+#define uint_to_hex_lower8(i) make_u32x (l_bin2asc[(i)])
+#elif VECT_SIZE == 2
+#define uint_to_hex_lower8(i) make_u32x (l_bin2asc[(i).s0], l_bin2asc[(i).s1])
+#elif VECT_SIZE == 4
+#define uint_to_hex_lower8(i) make_u32x (l_bin2asc[(i).s0], l_bin2asc[(i).s1], l_bin2asc[(i).s2], l_bin2asc[(i).s3])
+#elif VECT_SIZE == 8
+#define uint_to_hex_lower8(i) make_u32x (l_bin2asc[(i).s0], l_bin2asc[(i).s1], l_bin2asc[(i).s2], l_bin2asc[(i).s3], l_bin2asc[(i).s4], l_bin2asc[(i).s5], l_bin2asc[(i).s6], l_bin2asc[(i).s7])
+#elif VECT_SIZE == 16
+#define uint_to_hex_lower8(i) make_u32x (l_bin2asc[(i).s0], l_bin2asc[(i).s1], l_bin2asc[(i).s2], l_bin2asc[(i).s3], l_bin2asc[(i).s4], l_bin2asc[(i).s5], l_bin2asc[(i).s6], l_bin2asc[(i).s7], l_bin2asc[(i).s8], l_bin2asc[(i).s9], l_bin2asc[(i).sa], l_bin2asc[(i).sb], l_bin2asc[(i).sc], l_bin2asc[(i).sd], l_bin2asc[(i).se], l_bin2asc[(i).sf])
+#endif
+
+KERNEL_FQ void m30500_m04 (KERN_ATTR_BASIC ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * bin2asc table
+   */
+
+  LOCAL_VK u32 l_bin2asc[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    const u32 i0 = (i >> 0) & 15;
+    const u32 i1 = (i >> 4) & 15;
+
+    l_bin2asc[i] = ((i0 < 10) ? '0' + i0 : 'a' - 10 + i0) << 8
+                 | ((i1 < 10) ? '0' + i1 : 'a' - 10 + i1) << 0;
+  }
+
+  SYNC_THREADS ();
+
+  if (gid >= GID_CNT) return;
+
+  /**
+   * base
+   */
+
+  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];
+  u32 salt_buf1[4];
+  u32 salt_buf2[4];
+  u32 salt_buf3[4];
+
+  salt_buf0[0] = salt_bufs[SALT_POS_HOST].salt_buf_pc[0];
+  salt_buf0[1] = salt_bufs[SALT_POS_HOST].salt_buf_pc[1];
+  salt_buf0[2] = salt_bufs[SALT_POS_HOST].salt_buf_pc[2];
+  salt_buf0[3] = salt_bufs[SALT_POS_HOST].salt_buf_pc[3];
+  salt_buf1[0] = salt_bufs[SALT_POS_HOST].salt_buf_pc[4];
+  salt_buf1[1] = salt_bufs[SALT_POS_HOST].salt_buf_pc[5];
+  salt_buf1[2] = salt_bufs[SALT_POS_HOST].salt_buf_pc[6];
+  salt_buf1[3] = salt_bufs[SALT_POS_HOST].salt_buf_pc[7];
+  salt_buf2[0] = 0;
+  salt_buf2[1] = 0;
+  salt_buf2[2] = 0;
+  salt_buf2[3] = 0;
+  salt_buf3[0] = 0;
+  salt_buf3[1] = 0;
+  salt_buf3[2] = 0;
+  salt_buf3[3] = 0;
+
+  /**
+   * 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] = pw_len * 8;
+    w3[3] = 0;
+
+    /**
+     * md5
+     */
+
+    u32x a = MD5M_A;
+    u32x b = MD5M_B;
+    u32x c = MD5M_C;
+    u32x d = MD5M_D;
+
+    MD5_STEP (MD5_Fo, a, b, c, d, w0[0], MD5C00, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w0[1], MD5C01, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w0[2], MD5C02, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w0[3], MD5C03, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w1[0], MD5C04, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w1[1], MD5C05, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w1[2], MD5C06, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w1[3], MD5C07, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w2[0], MD5C08, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w2[1], MD5C09, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w2[2], MD5C0a, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w2[3], MD5C0b, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w3[0], MD5C0c, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w3[1], MD5C0d, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w3[2], MD5C0e, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w3[3], MD5C0f, MD5S03);
+
+    MD5_STEP (MD5_Go, a, b, c, d, w0[1], MD5C10, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w1[2], MD5C11, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w2[3], MD5C12, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w0[0], MD5C13, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w1[1], MD5C14, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w2[2], MD5C15, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w3[3], MD5C16, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w1[0], MD5C17, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w2[1], MD5C18, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w3[2], MD5C19, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w0[3], MD5C1a, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w2[0], MD5C1b, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w3[1], MD5C1c, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w0[2], MD5C1d, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w1[3], MD5C1e, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w3[0], MD5C1f, MD5S13);
+
+    u32x t;
+
+    MD5_STEP (MD5_H1, a, b, c, d, w1[1], MD5C20, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w2[0], MD5C21, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w2[3], MD5C22, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w3[2], MD5C23, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w0[1], MD5C24, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w1[0], MD5C25, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w1[3], MD5C26, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w2[2], MD5C27, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w3[1], MD5C28, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w0[0], MD5C29, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w0[3], MD5C2a, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w1[2], MD5C2b, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w2[1], MD5C2c, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w3[0], MD5C2d, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w3[3], MD5C2e, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w0[2], MD5C2f, MD5S23);
+
+    MD5_STEP (MD5_I , a, b, c, d, w0[0], MD5C30, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w1[3], MD5C31, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w3[2], MD5C32, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w1[1], MD5C33, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w3[0], MD5C34, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w0[3], MD5C35, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w2[2], MD5C36, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w0[1], MD5C37, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w2[0], MD5C38, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w3[3], MD5C39, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w1[2], MD5C3a, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w3[1], MD5C3b, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w1[0], MD5C3c, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w2[3], MD5C3d, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w0[2], MD5C3e, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w2[1], MD5C3f, MD5S33);
+
+    a += make_u32x (MD5M_A);
+    b += make_u32x (MD5M_B);
+    c += make_u32x (MD5M_C);
+    d += make_u32x (MD5M_D);
+
+    w0[0] = uint_to_hex_lower8 ((a >>  0) & 255) <<  0
+          | uint_to_hex_lower8 ((a >>  8) & 255) << 16;
+    w0[1] = uint_to_hex_lower8 ((a >> 16) & 255) <<  0
+          | uint_to_hex_lower8 ((a >> 24) & 255) << 16;
+    w0[2] = uint_to_hex_lower8 ((b >>  0) & 255) <<  0
+          | uint_to_hex_lower8 ((b >>  8) & 255) << 16;
+    w0[3] = uint_to_hex_lower8 ((b >> 16) & 255) <<  0
+          | uint_to_hex_lower8 ((b >> 24) & 255) << 16;
+    w1[0] = uint_to_hex_lower8 ((c >>  0) & 255) <<  0
+          | uint_to_hex_lower8 ((c >>  8) & 255) << 16;
+    w1[1] = uint_to_hex_lower8 ((c >> 16) & 255) <<  0
+          | uint_to_hex_lower8 ((c >> 24) & 255) << 16;
+    w1[2] = uint_to_hex_lower8 ((d >>  0) & 255) <<  0
+          | uint_to_hex_lower8 ((d >>  8) & 255) << 16;
+    w1[3] = uint_to_hex_lower8 ((d >> 16) & 255) <<  0
+          | uint_to_hex_lower8 ((d >> 24) & 255) << 16;
+		  
+    w2[0] = 0x80;
+    w2[1] = 0;
+    w2[2] = 0;
+    w2[3] = 0;
+    w3[0] = 0;
+    w3[1] = 0;
+    w3[2] = 32 * 8;
+    w3[3] = 0;
+	
+    a = MD5M_A;
+    b = MD5M_B;
+    c = MD5M_C;
+    d = MD5M_D;
+
+    MD5_STEP (MD5_Fo, a, b, c, d, w0[0], MD5C00, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w0[1], MD5C01, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w0[2], MD5C02, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w0[3], MD5C03, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w1[0], MD5C04, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w1[1], MD5C05, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w1[2], MD5C06, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w1[3], MD5C07, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w2[0], MD5C08, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w2[1], MD5C09, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w2[2], MD5C0a, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w2[3], MD5C0b, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w3[0], MD5C0c, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w3[1], MD5C0d, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w3[2], MD5C0e, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w3[3], MD5C0f, MD5S03);
+
+    MD5_STEP (MD5_Go, a, b, c, d, w0[1], MD5C10, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w1[2], MD5C11, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w2[3], MD5C12, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w0[0], MD5C13, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w1[1], MD5C14, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w2[2], MD5C15, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w3[3], MD5C16, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w1[0], MD5C17, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w2[1], MD5C18, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w3[2], MD5C19, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w0[3], MD5C1a, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w2[0], MD5C1b, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w3[1], MD5C1c, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w0[2], MD5C1d, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w1[3], MD5C1e, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w3[0], MD5C1f, MD5S13);
+
+    MD5_STEP (MD5_H1, a, b, c, d, w1[1], MD5C20, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w2[0], MD5C21, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w2[3], MD5C22, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w3[2], MD5C23, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w0[1], MD5C24, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w1[0], MD5C25, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w1[3], MD5C26, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w2[2], MD5C27, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w3[1], MD5C28, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w0[0], MD5C29, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w0[3], MD5C2a, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w1[2], MD5C2b, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w2[1], MD5C2c, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w3[0], MD5C2d, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w3[3], MD5C2e, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w0[2], MD5C2f, MD5S23);
+
+    MD5_STEP (MD5_I , a, b, c, d, w0[0], MD5C30, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w1[3], MD5C31, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w3[2], MD5C32, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w1[1], MD5C33, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w3[0], MD5C34, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w0[3], MD5C35, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w2[2], MD5C36, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w0[1], MD5C37, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w2[0], MD5C38, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w3[3], MD5C39, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w1[2], MD5C3a, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w3[1], MD5C3b, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w1[0], MD5C3c, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w2[3], MD5C3d, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w0[2], MD5C3e, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w2[1], MD5C3f, MD5S33);
+
+    a += make_u32x (MD5M_A);
+    b += make_u32x (MD5M_B);
+    c += make_u32x (MD5M_C);
+    d += make_u32x (MD5M_D);
+
+    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] = uint_to_hex_lower8 ((a >>  0) & 255) <<  0
+          | uint_to_hex_lower8 ((a >>  8) & 255) << 16;
+    w2[1] = uint_to_hex_lower8 ((a >> 16) & 255) <<  0
+          | uint_to_hex_lower8 ((a >> 24) & 255) << 16;
+    w2[2] = uint_to_hex_lower8 ((b >>  0) & 255) <<  0
+          | uint_to_hex_lower8 ((b >>  8) & 255) << 16;
+    w2[3] = uint_to_hex_lower8 ((b >> 16) & 255) <<  0
+          | uint_to_hex_lower8 ((b >> 24) & 255) << 16;
+    w3[0] = uint_to_hex_lower8 ((c >>  0) & 255) <<  0
+          | uint_to_hex_lower8 ((c >>  8) & 255) << 16;
+    w3[1] = uint_to_hex_lower8 ((c >> 16) & 255) <<  0
+          | uint_to_hex_lower8 ((c >> 24) & 255) << 16;
+    w3[2] = uint_to_hex_lower8 ((d >>  0) & 255) <<  0
+          | uint_to_hex_lower8 ((d >>  8) & 255) << 16;
+    w3[3] = uint_to_hex_lower8 ((d >> 16) & 255) <<  0
+          | uint_to_hex_lower8 ((d >> 24) & 255) << 16;
+
+    a = MD5M_A;
+    b = MD5M_B;
+    c = MD5M_C;
+    d = MD5M_D;
+
+    MD5_STEP (MD5_Fo, a, b, c, d, w0[0], MD5C00, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w0[1], MD5C01, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w0[2], MD5C02, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w0[3], MD5C03, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w1[0], MD5C04, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w1[1], MD5C05, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w1[2], MD5C06, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w1[3], MD5C07, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w2[0], MD5C08, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w2[1], MD5C09, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w2[2], MD5C0a, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w2[3], MD5C0b, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w3[0], MD5C0c, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w3[1], MD5C0d, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w3[2], MD5C0e, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w3[3], MD5C0f, MD5S03);
+
+    MD5_STEP (MD5_Go, a, b, c, d, w0[1], MD5C10, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w1[2], MD5C11, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w2[3], MD5C12, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w0[0], MD5C13, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w1[1], MD5C14, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w2[2], MD5C15, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w3[3], MD5C16, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w1[0], MD5C17, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w2[1], MD5C18, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w3[2], MD5C19, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w0[3], MD5C1a, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w2[0], MD5C1b, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w3[1], MD5C1c, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w0[2], MD5C1d, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w1[3], MD5C1e, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w3[0], MD5C1f, MD5S13);
+
+    MD5_STEP (MD5_H1, a, b, c, d, w1[1], MD5C20, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w2[0], MD5C21, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w2[3], MD5C22, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w3[2], MD5C23, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w0[1], MD5C24, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w1[0], MD5C25, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w1[3], MD5C26, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w2[2], MD5C27, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w3[1], MD5C28, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w0[0], MD5C29, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w0[3], MD5C2a, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w1[2], MD5C2b, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w2[1], MD5C2c, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w3[0], MD5C2d, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w3[3], MD5C2e, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w0[2], MD5C2f, MD5S23);
+
+    MD5_STEP (MD5_I , a, b, c, d, w0[0], MD5C30, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w1[3], MD5C31, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w3[2], MD5C32, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w1[1], MD5C33, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w3[0], MD5C34, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w0[3], MD5C35, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w2[2], MD5C36, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w0[1], MD5C37, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w2[0], MD5C38, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w3[3], MD5C39, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w1[2], MD5C3a, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w3[1], MD5C3b, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w1[0], MD5C3c, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w2[3], MD5C3d, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w0[2], MD5C3e, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w2[1], MD5C3f, MD5S33);
+
+    const u32x r_a = a + make_u32x (MD5M_A);
+    const u32x r_b = b + make_u32x (MD5M_B);
+    const u32x r_c = c + make_u32x (MD5M_C);
+    const u32x r_d = d + make_u32x (MD5M_D);
+
+    const u32x r_00 = 0x80;
+    const u32x r_14 = 64 * 8;
+
+    a = r_a;
+    b = r_b;
+    c = r_c;
+    d = r_d;
+
+    MD5_STEP (MD5_Fo, a, b, c, d, r_00, MD5C00, MD5S00);
+    MD5_STEP0(MD5_Fo, d, a, b, c,       MD5C01, MD5S01);
+    MD5_STEP0(MD5_Fo, c, d, a, b,       MD5C02, MD5S02);
+    MD5_STEP0(MD5_Fo, b, c, d, a,       MD5C03, MD5S03);
+    MD5_STEP0(MD5_Fo, a, b, c, d,       MD5C04, MD5S00);
+    MD5_STEP0(MD5_Fo, d, a, b, c,       MD5C05, MD5S01);
+    MD5_STEP0(MD5_Fo, c, d, a, b,       MD5C06, MD5S02);
+    MD5_STEP0(MD5_Fo, b, c, d, a,       MD5C07, MD5S03);
+    MD5_STEP0(MD5_Fo, a, b, c, d,       MD5C08, MD5S00);
+    MD5_STEP0(MD5_Fo, d, a, b, c,       MD5C09, MD5S01);
+    MD5_STEP0(MD5_Fo, c, d, a, b,       MD5C0a, MD5S02);
+    MD5_STEP0(MD5_Fo, b, c, d, a,       MD5C0b, MD5S03);
+    MD5_STEP0(MD5_Fo, a, b, c, d,       MD5C0c, MD5S00);
+    MD5_STEP0(MD5_Fo, d, a, b, c,       MD5C0d, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, r_14, MD5C0e, MD5S02);
+    MD5_STEP0(MD5_Fo, b, c, d, a,       MD5C0f, MD5S03);
+
+    MD5_STEP0(MD5_Go, a, b, c, d,       MD5C10, MD5S10);
+    MD5_STEP0(MD5_Go, d, a, b, c,       MD5C11, MD5S11);
+    MD5_STEP0(MD5_Go, c, d, a, b,       MD5C12, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, r_00, MD5C13, MD5S13);
+    MD5_STEP0(MD5_Go, a, b, c, d,       MD5C14, MD5S10);
+    MD5_STEP0(MD5_Go, d, a, b, c,       MD5C15, MD5S11);
+    MD5_STEP0(MD5_Go, c, d, a, b,       MD5C16, MD5S12);
+    MD5_STEP0(MD5_Go, b, c, d, a,       MD5C17, MD5S13);
+    MD5_STEP0(MD5_Go, a, b, c, d,       MD5C18, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, r_14, MD5C19, MD5S11);
+    MD5_STEP0(MD5_Go, c, d, a, b,       MD5C1a, MD5S12);
+    MD5_STEP0(MD5_Go, b, c, d, a,       MD5C1b, MD5S13);
+    MD5_STEP0(MD5_Go, a, b, c, d,       MD5C1c, MD5S10);
+    MD5_STEP0(MD5_Go, d, a, b, c,       MD5C1d, MD5S11);
+    MD5_STEP0(MD5_Go, c, d, a, b,       MD5C1e, MD5S12);
+    MD5_STEP0(MD5_Go, b, c, d, a,       MD5C1f, MD5S13);
+
+    MD5_STEP0(MD5_H1, a, b, c, d,       MD5C20, MD5S20);
+    MD5_STEP0(MD5_H2, d, a, b, c,       MD5C21, MD5S21);
+    MD5_STEP0(MD5_H1, c, d, a, b,       MD5C22, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, r_14, MD5C23, MD5S23);
+    MD5_STEP0(MD5_H1, a, b, c, d,       MD5C24, MD5S20);
+    MD5_STEP0(MD5_H2, d, a, b, c,       MD5C25, MD5S21);
+    MD5_STEP0(MD5_H1, c, d, a, b,       MD5C26, MD5S22);
+    MD5_STEP0(MD5_H2, b, c, d, a,       MD5C27, MD5S23);
+    MD5_STEP0(MD5_H1, a, b, c, d,       MD5C28, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, r_00, MD5C29, MD5S21);
+    MD5_STEP0(MD5_H1, c, d, a, b,       MD5C2a, MD5S22);
+    MD5_STEP0(MD5_H2, b, c, d, a,       MD5C2b, MD5S23);
+    MD5_STEP0(MD5_H1, a, b, c, d,       MD5C2c, MD5S20);
+    MD5_STEP0(MD5_H2, d, a, b, c,       MD5C2d, MD5S21);
+    MD5_STEP0(MD5_H1, c, d, a, b,       MD5C2e, MD5S22);
+    MD5_STEP0(MD5_H2, b, c, d, a,       MD5C2f, MD5S23);
+
+    MD5_STEP (MD5_I , a, b, c, d, r_00, MD5C30, MD5S30);
+    MD5_STEP0(MD5_I , d, a, b, c,       MD5C31, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, r_14, MD5C32, MD5S32);
+    MD5_STEP0(MD5_I , b, c, d, a,       MD5C33, MD5S33);
+    MD5_STEP0(MD5_I , a, b, c, d,       MD5C34, MD5S30);
+    MD5_STEP0(MD5_I , d, a, b, c,       MD5C35, MD5S31);
+    MD5_STEP0(MD5_I , c, d, a, b,       MD5C36, MD5S32);
+    MD5_STEP0(MD5_I , b, c, d, a,       MD5C37, MD5S33);
+    MD5_STEP0(MD5_I , a, b, c, d,       MD5C38, MD5S30);
+    MD5_STEP0(MD5_I , d, a, b, c,       MD5C39, MD5S31);
+    MD5_STEP0(MD5_I , c, d, a, b,       MD5C3a, MD5S32);
+    MD5_STEP0(MD5_I , b, c, d, a,       MD5C3b, MD5S33);
+    MD5_STEP0(MD5_I , a, b, c, d,       MD5C3c, MD5S30);
+    MD5_STEP0(MD5_I , d, a, b, c,       MD5C3d, MD5S31);
+    MD5_STEP0(MD5_I , c, d, a, b,       MD5C3e, MD5S32);
+    MD5_STEP0(MD5_I , b, c, d, a,       MD5C3f, MD5S33);
+
+    a += r_a;
+    b += r_b;
+    c += r_c;
+    d += r_d;
+
+    COMPARE_M_SIMD (a, d, c, b);
+  }
+}
+
+KERNEL_FQ void m30500_m08 (KERN_ATTR_BASIC ())
+{
+}
+
+KERNEL_FQ void m30500_m16 (KERN_ATTR_BASIC ())
+{
+}
+
+KERNEL_FQ void m30500_s04 (KERN_ATTR_BASIC ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * bin2asc table
+   */
+
+  LOCAL_VK u32 l_bin2asc[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    const u32 i0 = (i >> 0) & 15;
+    const u32 i1 = (i >> 4) & 15;
+
+    l_bin2asc[i] = ((i0 < 10) ? '0' + i0 : 'a' - 10 + i0) << 8
+                 | ((i1 < 10) ? '0' + i1 : 'a' - 10 + i1) << 0;
+  }
+
+  SYNC_THREADS ();
+
+  if (gid >= GID_CNT) return;
+
+  /**
+   * base
+   */
+
+  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];
+  u32 salt_buf1[4];
+  u32 salt_buf2[4];
+  u32 salt_buf3[4];
+
+  salt_buf0[0] = salt_bufs[SALT_POS_HOST].salt_buf_pc[0];
+  salt_buf0[1] = salt_bufs[SALT_POS_HOST].salt_buf_pc[1];
+  salt_buf0[2] = salt_bufs[SALT_POS_HOST].salt_buf_pc[2];
+  salt_buf0[3] = salt_bufs[SALT_POS_HOST].salt_buf_pc[3];
+  salt_buf1[0] = salt_bufs[SALT_POS_HOST].salt_buf_pc[4];
+  salt_buf1[1] = salt_bufs[SALT_POS_HOST].salt_buf_pc[5];
+  salt_buf1[2] = salt_bufs[SALT_POS_HOST].salt_buf_pc[6];
+  salt_buf1[3] = salt_bufs[SALT_POS_HOST].salt_buf_pc[7];
+  salt_buf2[0] = 0;
+  salt_buf2[1] = 0;
+  salt_buf2[2] = 0;
+  salt_buf2[3] = 0;
+  salt_buf3[0] = 0;
+  salt_buf3[1] = 0;
+  salt_buf3[2] = 0;
+  salt_buf3[3] = 0;
+
+  /**
+   * digest
+   */
+
+  const u32 search[4] =
+  {
+    digests_buf[DIGESTS_OFFSET_HOST].digest_buf[DGST_R0],
+    digests_buf[DIGESTS_OFFSET_HOST].digest_buf[DGST_R1],
+    digests_buf[DIGESTS_OFFSET_HOST].digest_buf[DGST_R2],
+    digests_buf[DIGESTS_OFFSET_HOST].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] = pw_len * 8;
+    w3[3] = 0;
+
+    /**
+     * md5
+     */
+
+    u32x a = MD5M_A;
+    u32x b = MD5M_B;
+    u32x c = MD5M_C;
+    u32x d = MD5M_D;
+
+    MD5_STEP (MD5_Fo, a, b, c, d, w0[0], MD5C00, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w0[1], MD5C01, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w0[2], MD5C02, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w0[3], MD5C03, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w1[0], MD5C04, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w1[1], MD5C05, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w1[2], MD5C06, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w1[3], MD5C07, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w2[0], MD5C08, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w2[1], MD5C09, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w2[2], MD5C0a, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w2[3], MD5C0b, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w3[0], MD5C0c, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w3[1], MD5C0d, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w3[2], MD5C0e, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w3[3], MD5C0f, MD5S03);
+
+    MD5_STEP (MD5_Go, a, b, c, d, w0[1], MD5C10, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w1[2], MD5C11, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w2[3], MD5C12, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w0[0], MD5C13, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w1[1], MD5C14, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w2[2], MD5C15, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w3[3], MD5C16, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w1[0], MD5C17, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w2[1], MD5C18, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w3[2], MD5C19, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w0[3], MD5C1a, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w2[0], MD5C1b, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w3[1], MD5C1c, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w0[2], MD5C1d, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w1[3], MD5C1e, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w3[0], MD5C1f, MD5S13);
+
+    u32x t;
+
+    MD5_STEP (MD5_H1, a, b, c, d, w1[1], MD5C20, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w2[0], MD5C21, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w2[3], MD5C22, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w3[2], MD5C23, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w0[1], MD5C24, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w1[0], MD5C25, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w1[3], MD5C26, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w2[2], MD5C27, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w3[1], MD5C28, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w0[0], MD5C29, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w0[3], MD5C2a, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w1[2], MD5C2b, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w2[1], MD5C2c, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w3[0], MD5C2d, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w3[3], MD5C2e, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w0[2], MD5C2f, MD5S23);
+
+    MD5_STEP (MD5_I , a, b, c, d, w0[0], MD5C30, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w1[3], MD5C31, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w3[2], MD5C32, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w1[1], MD5C33, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w3[0], MD5C34, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w0[3], MD5C35, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w2[2], MD5C36, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w0[1], MD5C37, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w2[0], MD5C38, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w3[3], MD5C39, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w1[2], MD5C3a, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w3[1], MD5C3b, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w1[0], MD5C3c, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w2[3], MD5C3d, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w0[2], MD5C3e, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w2[1], MD5C3f, MD5S33);
+
+    a += make_u32x (MD5M_A);
+    b += make_u32x (MD5M_B);
+    c += make_u32x (MD5M_C);
+    d += make_u32x (MD5M_D);
+
+    w0[0] = uint_to_hex_lower8 ((a >>  0) & 255) <<  0
+          | uint_to_hex_lower8 ((a >>  8) & 255) << 16;
+    w0[1] = uint_to_hex_lower8 ((a >> 16) & 255) <<  0
+          | uint_to_hex_lower8 ((a >> 24) & 255) << 16;
+    w0[2] = uint_to_hex_lower8 ((b >>  0) & 255) <<  0
+          | uint_to_hex_lower8 ((b >>  8) & 255) << 16;
+    w0[3] = uint_to_hex_lower8 ((b >> 16) & 255) <<  0
+          | uint_to_hex_lower8 ((b >> 24) & 255) << 16;
+    w1[0] = uint_to_hex_lower8 ((c >>  0) & 255) <<  0
+          | uint_to_hex_lower8 ((c >>  8) & 255) << 16;
+    w1[1] = uint_to_hex_lower8 ((c >> 16) & 255) <<  0
+          | uint_to_hex_lower8 ((c >> 24) & 255) << 16;
+    w1[2] = uint_to_hex_lower8 ((d >>  0) & 255) <<  0
+          | uint_to_hex_lower8 ((d >>  8) & 255) << 16;
+    w1[3] = uint_to_hex_lower8 ((d >> 16) & 255) <<  0
+          | uint_to_hex_lower8 ((d >> 24) & 255) << 16;
+		  
+    w2[0] = 0x80;
+    w2[1] = 0;
+    w2[2] = 0;
+    w2[3] = 0;
+    w3[0] = 0;
+    w3[1] = 0;
+    w3[2] = 32 * 8;
+    w3[3] = 0;
+	
+    a = MD5M_A;
+    b = MD5M_B;
+    c = MD5M_C;
+    d = MD5M_D;
+
+    MD5_STEP (MD5_Fo, a, b, c, d, w0[0], MD5C00, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w0[1], MD5C01, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w0[2], MD5C02, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w0[3], MD5C03, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w1[0], MD5C04, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w1[1], MD5C05, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w1[2], MD5C06, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w1[3], MD5C07, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w2[0], MD5C08, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w2[1], MD5C09, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w2[2], MD5C0a, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w2[3], MD5C0b, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w3[0], MD5C0c, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w3[1], MD5C0d, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w3[2], MD5C0e, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w3[3], MD5C0f, MD5S03);
+
+    MD5_STEP (MD5_Go, a, b, c, d, w0[1], MD5C10, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w1[2], MD5C11, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w2[3], MD5C12, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w0[0], MD5C13, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w1[1], MD5C14, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w2[2], MD5C15, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w3[3], MD5C16, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w1[0], MD5C17, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w2[1], MD5C18, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w3[2], MD5C19, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w0[3], MD5C1a, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w2[0], MD5C1b, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w3[1], MD5C1c, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w0[2], MD5C1d, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w1[3], MD5C1e, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w3[0], MD5C1f, MD5S13);
+
+    MD5_STEP (MD5_H1, a, b, c, d, w1[1], MD5C20, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w2[0], MD5C21, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w2[3], MD5C22, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w3[2], MD5C23, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w0[1], MD5C24, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w1[0], MD5C25, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w1[3], MD5C26, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w2[2], MD5C27, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w3[1], MD5C28, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w0[0], MD5C29, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w0[3], MD5C2a, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w1[2], MD5C2b, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w2[1], MD5C2c, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w3[0], MD5C2d, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w3[3], MD5C2e, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w0[2], MD5C2f, MD5S23);
+
+    MD5_STEP (MD5_I , a, b, c, d, w0[0], MD5C30, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w1[3], MD5C31, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w3[2], MD5C32, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w1[1], MD5C33, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w3[0], MD5C34, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w0[3], MD5C35, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w2[2], MD5C36, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w0[1], MD5C37, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w2[0], MD5C38, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w3[3], MD5C39, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w1[2], MD5C3a, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w3[1], MD5C3b, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w1[0], MD5C3c, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w2[3], MD5C3d, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w0[2], MD5C3e, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w2[1], MD5C3f, MD5S33);
+
+    a += make_u32x (MD5M_A);
+    b += make_u32x (MD5M_B);
+    c += make_u32x (MD5M_C);
+    d += make_u32x (MD5M_D);
+
+    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] = uint_to_hex_lower8 ((a >>  0) & 255) <<  0
+          | uint_to_hex_lower8 ((a >>  8) & 255) << 16;
+    w2[1] = uint_to_hex_lower8 ((a >> 16) & 255) <<  0
+          | uint_to_hex_lower8 ((a >> 24) & 255) << 16;
+    w2[2] = uint_to_hex_lower8 ((b >>  0) & 255) <<  0
+          | uint_to_hex_lower8 ((b >>  8) & 255) << 16;
+    w2[3] = uint_to_hex_lower8 ((b >> 16) & 255) <<  0
+          | uint_to_hex_lower8 ((b >> 24) & 255) << 16;
+    w3[0] = uint_to_hex_lower8 ((c >>  0) & 255) <<  0
+          | uint_to_hex_lower8 ((c >>  8) & 255) << 16;
+    w3[1] = uint_to_hex_lower8 ((c >> 16) & 255) <<  0
+          | uint_to_hex_lower8 ((c >> 24) & 255) << 16;
+    w3[2] = uint_to_hex_lower8 ((d >>  0) & 255) <<  0
+          | uint_to_hex_lower8 ((d >>  8) & 255) << 16;
+    w3[3] = uint_to_hex_lower8 ((d >> 16) & 255) <<  0
+          | uint_to_hex_lower8 ((d >> 24) & 255) << 16;
+
+    a = MD5M_A;
+    b = MD5M_B;
+    c = MD5M_C;
+    d = MD5M_D;
+
+    MD5_STEP (MD5_Fo, a, b, c, d, w0[0], MD5C00, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w0[1], MD5C01, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w0[2], MD5C02, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w0[3], MD5C03, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w1[0], MD5C04, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w1[1], MD5C05, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w1[2], MD5C06, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w1[3], MD5C07, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w2[0], MD5C08, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w2[1], MD5C09, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w2[2], MD5C0a, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w2[3], MD5C0b, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w3[0], MD5C0c, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w3[1], MD5C0d, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w3[2], MD5C0e, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w3[3], MD5C0f, MD5S03);
+
+    MD5_STEP (MD5_Go, a, b, c, d, w0[1], MD5C10, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w1[2], MD5C11, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w2[3], MD5C12, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w0[0], MD5C13, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w1[1], MD5C14, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w2[2], MD5C15, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w3[3], MD5C16, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w1[0], MD5C17, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w2[1], MD5C18, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w3[2], MD5C19, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w0[3], MD5C1a, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w2[0], MD5C1b, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w3[1], MD5C1c, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w0[2], MD5C1d, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w1[3], MD5C1e, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w3[0], MD5C1f, MD5S13);
+
+    MD5_STEP (MD5_H1, a, b, c, d, w1[1], MD5C20, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w2[0], MD5C21, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w2[3], MD5C22, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w3[2], MD5C23, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w0[1], MD5C24, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w1[0], MD5C25, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w1[3], MD5C26, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w2[2], MD5C27, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w3[1], MD5C28, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w0[0], MD5C29, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w0[3], MD5C2a, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w1[2], MD5C2b, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w2[1], MD5C2c, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w3[0], MD5C2d, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w3[3], MD5C2e, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w0[2], MD5C2f, MD5S23);
+
+    MD5_STEP (MD5_I , a, b, c, d, w0[0], MD5C30, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w1[3], MD5C31, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w3[2], MD5C32, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w1[1], MD5C33, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w3[0], MD5C34, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w0[3], MD5C35, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w2[2], MD5C36, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w0[1], MD5C37, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w2[0], MD5C38, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w3[3], MD5C39, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w1[2], MD5C3a, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w3[1], MD5C3b, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w1[0], MD5C3c, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w2[3], MD5C3d, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w0[2], MD5C3e, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w2[1], MD5C3f, MD5S33);
+
+    const u32x r_a = a + make_u32x (MD5M_A);
+    const u32x r_b = b + make_u32x (MD5M_B);
+    const u32x r_c = c + make_u32x (MD5M_C);
+    const u32x r_d = d + make_u32x (MD5M_D);
+
+    const u32x r_00 = 0x80;
+    const u32x r_14 = 64 * 8;
+
+    a = r_a;
+    b = r_b;
+    c = r_c;
+    d = r_d;
+
+    MD5_STEP (MD5_Fo, a, b, c, d, r_00, MD5C00, MD5S00);
+    MD5_STEP0(MD5_Fo, d, a, b, c,       MD5C01, MD5S01);
+    MD5_STEP0(MD5_Fo, c, d, a, b,       MD5C02, MD5S02);
+    MD5_STEP0(MD5_Fo, b, c, d, a,       MD5C03, MD5S03);
+    MD5_STEP0(MD5_Fo, a, b, c, d,       MD5C04, MD5S00);
+    MD5_STEP0(MD5_Fo, d, a, b, c,       MD5C05, MD5S01);
+    MD5_STEP0(MD5_Fo, c, d, a, b,       MD5C06, MD5S02);
+    MD5_STEP0(MD5_Fo, b, c, d, a,       MD5C07, MD5S03);
+    MD5_STEP0(MD5_Fo, a, b, c, d,       MD5C08, MD5S00);
+    MD5_STEP0(MD5_Fo, d, a, b, c,       MD5C09, MD5S01);
+    MD5_STEP0(MD5_Fo, c, d, a, b,       MD5C0a, MD5S02);
+    MD5_STEP0(MD5_Fo, b, c, d, a,       MD5C0b, MD5S03);
+    MD5_STEP0(MD5_Fo, a, b, c, d,       MD5C0c, MD5S00);
+    MD5_STEP0(MD5_Fo, d, a, b, c,       MD5C0d, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, r_14, MD5C0e, MD5S02);
+    MD5_STEP0(MD5_Fo, b, c, d, a,       MD5C0f, MD5S03);
+
+    MD5_STEP0(MD5_Go, a, b, c, d,       MD5C10, MD5S10);
+    MD5_STEP0(MD5_Go, d, a, b, c,       MD5C11, MD5S11);
+    MD5_STEP0(MD5_Go, c, d, a, b,       MD5C12, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, r_00, MD5C13, MD5S13);
+    MD5_STEP0(MD5_Go, a, b, c, d,       MD5C14, MD5S10);
+    MD5_STEP0(MD5_Go, d, a, b, c,       MD5C15, MD5S11);
+    MD5_STEP0(MD5_Go, c, d, a, b,       MD5C16, MD5S12);
+    MD5_STEP0(MD5_Go, b, c, d, a,       MD5C17, MD5S13);
+    MD5_STEP0(MD5_Go, a, b, c, d,       MD5C18, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, r_14, MD5C19, MD5S11);
+    MD5_STEP0(MD5_Go, c, d, a, b,       MD5C1a, MD5S12);
+    MD5_STEP0(MD5_Go, b, c, d, a,       MD5C1b, MD5S13);
+    MD5_STEP0(MD5_Go, a, b, c, d,       MD5C1c, MD5S10);
+    MD5_STEP0(MD5_Go, d, a, b, c,       MD5C1d, MD5S11);
+    MD5_STEP0(MD5_Go, c, d, a, b,       MD5C1e, MD5S12);
+    MD5_STEP0(MD5_Go, b, c, d, a,       MD5C1f, MD5S13);
+
+    MD5_STEP0(MD5_H1, a, b, c, d,       MD5C20, MD5S20);
+    MD5_STEP0(MD5_H2, d, a, b, c,       MD5C21, MD5S21);
+    MD5_STEP0(MD5_H1, c, d, a, b,       MD5C22, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, r_14, MD5C23, MD5S23);
+    MD5_STEP0(MD5_H1, a, b, c, d,       MD5C24, MD5S20);
+    MD5_STEP0(MD5_H2, d, a, b, c,       MD5C25, MD5S21);
+    MD5_STEP0(MD5_H1, c, d, a, b,       MD5C26, MD5S22);
+    MD5_STEP0(MD5_H2, b, c, d, a,       MD5C27, MD5S23);
+    MD5_STEP0(MD5_H1, a, b, c, d,       MD5C28, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, r_00, MD5C29, MD5S21);
+    MD5_STEP0(MD5_H1, c, d, a, b,       MD5C2a, MD5S22);
+    MD5_STEP0(MD5_H2, b, c, d, a,       MD5C2b, MD5S23);
+    MD5_STEP0(MD5_H1, a, b, c, d,       MD5C2c, MD5S20);
+    MD5_STEP0(MD5_H2, d, a, b, c,       MD5C2d, MD5S21);
+    MD5_STEP0(MD5_H1, c, d, a, b,       MD5C2e, MD5S22);
+    MD5_STEP0(MD5_H2, b, c, d, a,       MD5C2f, MD5S23);
+
+    MD5_STEP (MD5_I , a, b, c, d, r_00, MD5C30, MD5S30);
+    MD5_STEP0(MD5_I , d, a, b, c,       MD5C31, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, r_14, MD5C32, MD5S32);
+    MD5_STEP0(MD5_I , b, c, d, a,       MD5C33, MD5S33);
+    MD5_STEP0(MD5_I , a, b, c, d,       MD5C34, MD5S30);
+    MD5_STEP0(MD5_I , d, a, b, c,       MD5C35, MD5S31);
+    MD5_STEP0(MD5_I , c, d, a, b,       MD5C36, MD5S32);
+    MD5_STEP0(MD5_I , b, c, d, a,       MD5C37, MD5S33);
+    MD5_STEP0(MD5_I , a, b, c, d,       MD5C38, MD5S30);
+    MD5_STEP0(MD5_I , d, a, b, c,       MD5C39, MD5S31);
+    MD5_STEP0(MD5_I , c, d, a, b,       MD5C3a, MD5S32);
+    MD5_STEP0(MD5_I , b, c, d, a,       MD5C3b, MD5S33);
+    MD5_STEP0(MD5_I , a, b, c, d,       MD5C3c, MD5S30);
+
+    if (MATCHES_NONE_VS ((a + r_a), search[0])) continue;
+
+    MD5_STEP0(MD5_I , d, a, b, c,       MD5C3d, MD5S31);
+    MD5_STEP0(MD5_I , c, d, a, b,       MD5C3e, MD5S32);
+    MD5_STEP0(MD5_I , b, c, d, a,       MD5C3f, MD5S33);
+
+    a += r_a;
+    b += r_b;
+    c += r_c;
+    d += r_d;
+
+    COMPARE_S_SIMD (a, d, c, b);
+  }
+}
+
+KERNEL_FQ void m30500_s08 (KERN_ATTR_BASIC ())
+{
+}
+
+KERNEL_FQ void m30500_s16 (KERN_ATTR_BASIC ())
+{
+}
diff --git a/OpenCL/m30500_a1-pure.cl b/OpenCL/m30500_a1-pure.cl
new file mode 100644
index 000000000..18bed013f
--- /dev/null
+++ b/OpenCL/m30500_a1-pure.cl
@@ -0,0 +1,347 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+//#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include M2S(INCLUDE_PATH/inc_vendor.h)
+#include M2S(INCLUDE_PATH/inc_types.h)
+#include M2S(INCLUDE_PATH/inc_platform.cl)
+#include M2S(INCLUDE_PATH/inc_common.cl)
+#include M2S(INCLUDE_PATH/inc_scalar.cl)
+#include M2S(INCLUDE_PATH/inc_hash_md5.cl)
+#endif
+
+#if   VECT_SIZE == 1
+#define uint_to_hex_lower8(i) make_u32x (l_bin2asc[(i)])
+#elif VECT_SIZE == 2
+#define uint_to_hex_lower8(i) make_u32x (l_bin2asc[(i).s0], l_bin2asc[(i).s1])
+#elif VECT_SIZE == 4
+#define uint_to_hex_lower8(i) make_u32x (l_bin2asc[(i).s0], l_bin2asc[(i).s1], l_bin2asc[(i).s2], l_bin2asc[(i).s3])
+#elif VECT_SIZE == 8
+#define uint_to_hex_lower8(i) make_u32x (l_bin2asc[(i).s0], l_bin2asc[(i).s1], l_bin2asc[(i).s2], l_bin2asc[(i).s3], l_bin2asc[(i).s4], l_bin2asc[(i).s5], l_bin2asc[(i).s6], l_bin2asc[(i).s7])
+#elif VECT_SIZE == 16
+#define uint_to_hex_lower8(i) make_u32x (l_bin2asc[(i).s0], l_bin2asc[(i).s1], l_bin2asc[(i).s2], l_bin2asc[(i).s3], l_bin2asc[(i).s4], l_bin2asc[(i).s5], l_bin2asc[(i).s6], l_bin2asc[(i).s7], l_bin2asc[(i).s8], l_bin2asc[(i).s9], l_bin2asc[(i).sa], l_bin2asc[(i).sb], l_bin2asc[(i).sc], l_bin2asc[(i).sd], l_bin2asc[(i).se], l_bin2asc[(i).sf])
+#endif
+
+KERNEL_FQ void m30500_mxx (KERN_ATTR_BASIC ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * bin2asc table
+   */
+
+  LOCAL_VK u32 l_bin2asc[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    const u32 i0 = (i >> 0) & 15;
+    const u32 i1 = (i >> 4) & 15;
+
+    l_bin2asc[i] = ((i0 < 10) ? '0' + i0 : 'a' - 10 + i0) << 8
+                 | ((i1 < 10) ? '0' + i1 : 'a' - 10 + i1) << 0;
+  }
+
+  SYNC_THREADS ();
+
+  if (gid >= GID_CNT) return;
+
+  /**
+   * base
+   */
+
+  const u32 salt_len = 32;
+
+  u32 s[8] = { 0 };
+
+  for (u32 i = 0, idx = 0; i < salt_len; i += 4, idx += 1)
+  {
+    s[idx] = salt_bufs[SALT_POS_HOST].salt_buf_pc[idx];
+  }
+
+  md5_ctx_t ctx0;
+
+  md5_init (&ctx0);
+
+  md5_update_global (&ctx0, pws[gid].i, pws[gid].pw_len);
+
+  /**
+   * loop
+   */
+
+  for (u32 il_pos = 0; il_pos < IL_CNT; il_pos++)
+  {
+    md5_ctx_t ctx1 = ctx0;
+
+    md5_update_global (&ctx1, combs_buf[il_pos].i, combs_buf[il_pos].pw_len);
+
+    md5_final (&ctx1);
+
+    u32 a = ctx1.h[0];
+    u32 b = ctx1.h[1];
+    u32 c = ctx1.h[2];
+    u32 d = ctx1.h[3];
+
+    md5_ctx_t ctx;
+
+    md5_init (&ctx);
+
+    ctx.w0[0] = uint_to_hex_lower8 ((a >>  0) & 255) <<  0
+              | uint_to_hex_lower8 ((a >>  8) & 255) << 16;
+    ctx.w0[1] = uint_to_hex_lower8 ((a >> 16) & 255) <<  0
+              | uint_to_hex_lower8 ((a >> 24) & 255) << 16;
+    ctx.w0[2] = uint_to_hex_lower8 ((b >>  0) & 255) <<  0
+              | uint_to_hex_lower8 ((b >>  8) & 255) << 16;
+    ctx.w0[3] = uint_to_hex_lower8 ((b >> 16) & 255) <<  0
+              | uint_to_hex_lower8 ((b >> 24) & 255) << 16;
+    ctx.w1[0] = uint_to_hex_lower8 ((c >>  0) & 255) <<  0
+              | uint_to_hex_lower8 ((c >>  8) & 255) << 16;
+    ctx.w1[1] = uint_to_hex_lower8 ((c >> 16) & 255) <<  0
+              | uint_to_hex_lower8 ((c >> 24) & 255) << 16;
+    ctx.w1[2] = uint_to_hex_lower8 ((d >>  0) & 255) <<  0
+              | uint_to_hex_lower8 ((d >>  8) & 255) << 16;
+    ctx.w1[3] = uint_to_hex_lower8 ((d >> 16) & 255) <<  0
+              | uint_to_hex_lower8 ((d >> 24) & 255) << 16;
+
+    ctx.len = 32;
+
+    md5_final (&ctx);
+
+    a = ctx.h[0];
+    b = ctx.h[1];
+    c = ctx.h[2];
+    d = ctx.h[3];
+
+    md5_ctx_t ctx2;
+
+    md5_init (&ctx2);
+
+    ctx2.w0[0] = s[0];
+    ctx2.w0[1] = s[1];
+    ctx2.w0[2] = s[2];
+    ctx2.w0[3] = s[3];
+    ctx2.w1[0] = s[4];
+    ctx2.w1[1] = s[5];
+    ctx2.w1[2] = s[6];
+    ctx2.w1[3] = s[7];
+    ctx2.w2[0] = uint_to_hex_lower8 ((a >>  0) & 255) <<  0
+               | uint_to_hex_lower8 ((a >>  8) & 255) << 16;
+    ctx2.w2[1] = uint_to_hex_lower8 ((a >> 16) & 255) <<  0
+               | uint_to_hex_lower8 ((a >> 24) & 255) << 16;
+    ctx2.w2[2] = uint_to_hex_lower8 ((b >>  0) & 255) <<  0
+               | uint_to_hex_lower8 ((b >>  8) & 255) << 16;
+    ctx2.w2[3] = uint_to_hex_lower8 ((b >> 16) & 255) <<  0
+               | uint_to_hex_lower8 ((b >> 24) & 255) << 16;
+    ctx2.w3[0] = uint_to_hex_lower8 ((c >>  0) & 255) <<  0
+               | uint_to_hex_lower8 ((c >>  8) & 255) << 16;
+    ctx2.w3[1] = uint_to_hex_lower8 ((c >> 16) & 255) <<  0
+               | uint_to_hex_lower8 ((c >> 24) & 255) << 16;
+    ctx2.w3[2] = uint_to_hex_lower8 ((d >>  0) & 255) <<  0
+               | uint_to_hex_lower8 ((d >>  8) & 255) << 16;
+    ctx2.w3[3] = uint_to_hex_lower8 ((d >> 16) & 255) <<  0
+               | uint_to_hex_lower8 ((d >> 24) & 255) << 16;
+
+    md5_transform (ctx2.w0, ctx2.w1, ctx2.w2, ctx2.w3, ctx2.h);
+
+    ctx2.w0[0] = 0x80;
+    ctx2.w0[1] = 0;
+    ctx2.w0[2] = 0;
+    ctx2.w0[3] = 0;
+    ctx2.w1[0] = 0;
+    ctx2.w1[1] = 0;
+    ctx2.w1[2] = 0;
+    ctx2.w1[3] = 0;
+    ctx2.w2[0] = 0;
+    ctx2.w2[1] = 0;
+    ctx2.w2[2] = 0;
+    ctx2.w2[3] = 0;
+    ctx2.w3[0] = 0;
+    ctx2.w3[1] = 0;
+    ctx2.w3[2] = 64 * 8;
+    ctx2.w3[3] = 0;
+
+    md5_transform (ctx2.w0, ctx2.w1, ctx2.w2, ctx2.w3, ctx2.h);
+
+    const u32 r0 = ctx2.h[DGST_R0];
+    const u32 r1 = ctx2.h[DGST_R1];
+    const u32 r2 = ctx2.h[DGST_R2];
+    const u32 r3 = ctx2.h[DGST_R3];
+
+    COMPARE_M_SCALAR (r0, r1, r2, r3);
+  }
+}
+
+KERNEL_FQ void m30500_sxx (KERN_ATTR_BASIC ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * bin2asc table
+   */
+
+  LOCAL_VK u32 l_bin2asc[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    const u32 i0 = (i >> 0) & 15;
+    const u32 i1 = (i >> 4) & 15;
+
+    l_bin2asc[i] = ((i0 < 10) ? '0' + i0 : 'a' - 10 + i0) << 8
+                 | ((i1 < 10) ? '0' + i1 : 'a' - 10 + i1) << 0;
+  }
+
+  SYNC_THREADS ();
+
+  if (gid >= GID_CNT) return;
+
+  /**
+   * digest
+   */
+
+  const u32 search[4] =
+  {
+    digests_buf[DIGESTS_OFFSET_HOST].digest_buf[DGST_R0],
+    digests_buf[DIGESTS_OFFSET_HOST].digest_buf[DGST_R1],
+    digests_buf[DIGESTS_OFFSET_HOST].digest_buf[DGST_R2],
+    digests_buf[DIGESTS_OFFSET_HOST].digest_buf[DGST_R3]
+  };
+
+  /**
+   * base
+   */
+
+  const u32 salt_len = 32;
+
+  u32 s[8] = { 0 };
+
+  for (u32 i = 0, idx = 0; i < salt_len; i += 4, idx += 1)
+  {
+    s[idx] = salt_bufs[SALT_POS_HOST].salt_buf_pc[idx];
+  }
+
+  md5_ctx_t ctx0;
+
+  md5_init (&ctx0);
+
+  md5_update_global (&ctx0, pws[gid].i, pws[gid].pw_len);
+
+  /**
+   * loop
+   */
+
+  for (u32 il_pos = 0; il_pos < IL_CNT; il_pos++)
+  {
+    md5_ctx_t ctx1 = ctx0;
+
+    md5_update_global (&ctx1, combs_buf[il_pos].i, combs_buf[il_pos].pw_len);
+
+    md5_final (&ctx1);
+
+    u32 a = ctx1.h[0];
+    u32 b = ctx1.h[1];
+    u32 c = ctx1.h[2];
+    u32 d = ctx1.h[3];
+
+    md5_ctx_t ctx;
+
+    md5_init (&ctx);
+
+    ctx.w0[0] = uint_to_hex_lower8 ((a >>  0) & 255) <<  0
+              | uint_to_hex_lower8 ((a >>  8) & 255) << 16;
+    ctx.w0[1] = uint_to_hex_lower8 ((a >> 16) & 255) <<  0
+              | uint_to_hex_lower8 ((a >> 24) & 255) << 16;
+    ctx.w0[2] = uint_to_hex_lower8 ((b >>  0) & 255) <<  0
+              | uint_to_hex_lower8 ((b >>  8) & 255) << 16;
+    ctx.w0[3] = uint_to_hex_lower8 ((b >> 16) & 255) <<  0
+              | uint_to_hex_lower8 ((b >> 24) & 255) << 16;
+    ctx.w1[0] = uint_to_hex_lower8 ((c >>  0) & 255) <<  0
+              | uint_to_hex_lower8 ((c >>  8) & 255) << 16;
+    ctx.w1[1] = uint_to_hex_lower8 ((c >> 16) & 255) <<  0
+              | uint_to_hex_lower8 ((c >> 24) & 255) << 16;
+    ctx.w1[2] = uint_to_hex_lower8 ((d >>  0) & 255) <<  0
+              | uint_to_hex_lower8 ((d >>  8) & 255) << 16;
+    ctx.w1[3] = uint_to_hex_lower8 ((d >> 16) & 255) <<  0
+              | uint_to_hex_lower8 ((d >> 24) & 255) << 16;
+
+    ctx.len = 32;
+
+    md5_final (&ctx);
+
+    a = ctx.h[0];
+    b = ctx.h[1];
+    c = ctx.h[2];
+    d = ctx.h[3];
+
+    md5_ctx_t ctx2;
+
+    md5_init (&ctx2);
+
+    ctx2.w0[0] = s[0];
+    ctx2.w0[1] = s[1];
+    ctx2.w0[2] = s[2];
+    ctx2.w0[3] = s[3];
+    ctx2.w1[0] = s[4];
+    ctx2.w1[1] = s[5];
+    ctx2.w1[2] = s[6];
+    ctx2.w1[3] = s[7];
+    ctx2.w2[0] = uint_to_hex_lower8 ((a >>  0) & 255) <<  0
+               | uint_to_hex_lower8 ((a >>  8) & 255) << 16;
+    ctx2.w2[1] = uint_to_hex_lower8 ((a >> 16) & 255) <<  0
+               | uint_to_hex_lower8 ((a >> 24) & 255) << 16;
+    ctx2.w2[2] = uint_to_hex_lower8 ((b >>  0) & 255) <<  0
+               | uint_to_hex_lower8 ((b >>  8) & 255) << 16;
+    ctx2.w2[3] = uint_to_hex_lower8 ((b >> 16) & 255) <<  0
+               | uint_to_hex_lower8 ((b >> 24) & 255) << 16;
+    ctx2.w3[0] = uint_to_hex_lower8 ((c >>  0) & 255) <<  0
+               | uint_to_hex_lower8 ((c >>  8) & 255) << 16;
+    ctx2.w3[1] = uint_to_hex_lower8 ((c >> 16) & 255) <<  0
+               | uint_to_hex_lower8 ((c >> 24) & 255) << 16;
+    ctx2.w3[2] = uint_to_hex_lower8 ((d >>  0) & 255) <<  0
+               | uint_to_hex_lower8 ((d >>  8) & 255) << 16;
+    ctx2.w3[3] = uint_to_hex_lower8 ((d >> 16) & 255) <<  0
+               | uint_to_hex_lower8 ((d >> 24) & 255) << 16;
+
+    md5_transform (ctx2.w0, ctx2.w1, ctx2.w2, ctx2.w3, ctx2.h);
+
+    ctx2.w0[0] = 0x80;
+    ctx2.w0[1] = 0;
+    ctx2.w0[2] = 0;
+    ctx2.w0[3] = 0;
+    ctx2.w1[0] = 0;
+    ctx2.w1[1] = 0;
+    ctx2.w1[2] = 0;
+    ctx2.w1[3] = 0;
+    ctx2.w2[0] = 0;
+    ctx2.w2[1] = 0;
+    ctx2.w2[2] = 0;
+    ctx2.w2[3] = 0;
+    ctx2.w3[0] = 0;
+    ctx2.w3[1] = 0;
+    ctx2.w3[2] = 64 * 8;
+    ctx2.w3[3] = 0;
+
+    md5_transform (ctx2.w0, ctx2.w1, ctx2.w2, ctx2.w3, ctx2.h);
+
+    const u32 r0 = ctx2.h[DGST_R0];
+    const u32 r1 = ctx2.h[DGST_R1];
+    const u32 r2 = ctx2.h[DGST_R2];
+    const u32 r3 = ctx2.h[DGST_R3];
+
+    COMPARE_S_SCALAR (r0, r1, r2, r3);
+  }
+}
diff --git a/OpenCL/m30500_a3-optimized.cl b/OpenCL/m30500_a3-optimized.cl
new file mode 100644
index 000000000..2326eb1c3
--- /dev/null
+++ b/OpenCL/m30500_a3-optimized.cl
@@ -0,0 +1,1350 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include M2S(INCLUDE_PATH/inc_vendor.h)
+#include M2S(INCLUDE_PATH/inc_types.h)
+#include M2S(INCLUDE_PATH/inc_platform.cl)
+#include M2S(INCLUDE_PATH/inc_common.cl)
+#include M2S(INCLUDE_PATH/inc_simd.cl)
+#include M2S(INCLUDE_PATH/inc_hash_md5.cl)
+#endif
+
+#if   VECT_SIZE == 1
+#define uint_to_hex_lower8(i) make_u32x (l_bin2asc[(i)])
+#elif VECT_SIZE == 2
+#define uint_to_hex_lower8(i) make_u32x (l_bin2asc[(i).s0], l_bin2asc[(i).s1])
+#elif VECT_SIZE == 4
+#define uint_to_hex_lower8(i) make_u32x (l_bin2asc[(i).s0], l_bin2asc[(i).s1], l_bin2asc[(i).s2], l_bin2asc[(i).s3])
+#elif VECT_SIZE == 8
+#define uint_to_hex_lower8(i) make_u32x (l_bin2asc[(i).s0], l_bin2asc[(i).s1], l_bin2asc[(i).s2], l_bin2asc[(i).s3], l_bin2asc[(i).s4], l_bin2asc[(i).s5], l_bin2asc[(i).s6], l_bin2asc[(i).s7])
+#elif VECT_SIZE == 16
+#define uint_to_hex_lower8(i) make_u32x (l_bin2asc[(i).s0], l_bin2asc[(i).s1], l_bin2asc[(i).s2], l_bin2asc[(i).s3], l_bin2asc[(i).s4], l_bin2asc[(i).s5], l_bin2asc[(i).s6], l_bin2asc[(i).s7], l_bin2asc[(i).s8], l_bin2asc[(i).s9], l_bin2asc[(i).sa], l_bin2asc[(i).sb], l_bin2asc[(i).sc], l_bin2asc[(i).sd], l_bin2asc[(i).se], l_bin2asc[(i).sf])
+#endif
+
+DECLSPEC void m30500m (PRIVATE_AS u32 *w0, PRIVATE_AS u32 *w1, PRIVATE_AS u32 *w2, PRIVATE_AS u32 *w3, const u32 pw_len, KERN_ATTR_FUNC_BASIC (), LOCAL_AS u32 *l_bin2asc)
+{
+  /**
+   * modifiers are taken from args
+   */
+
+  /**
+   * salt
+   */
+
+  u32 salt_buf0[4];
+  u32 salt_buf1[4];
+  u32 salt_buf2[4];
+  u32 salt_buf3[4];
+
+  salt_buf0[0] = salt_bufs[SALT_POS_HOST].salt_buf_pc[0];
+  salt_buf0[1] = salt_bufs[SALT_POS_HOST].salt_buf_pc[1];
+  salt_buf0[2] = salt_bufs[SALT_POS_HOST].salt_buf_pc[2];
+  salt_buf0[3] = salt_bufs[SALT_POS_HOST].salt_buf_pc[3];
+  salt_buf1[0] = salt_bufs[SALT_POS_HOST].salt_buf_pc[4];
+  salt_buf1[1] = salt_bufs[SALT_POS_HOST].salt_buf_pc[5];
+  salt_buf1[2] = salt_bufs[SALT_POS_HOST].salt_buf_pc[6];
+  salt_buf1[3] = salt_bufs[SALT_POS_HOST].salt_buf_pc[7];
+  salt_buf2[0] = 0;
+  salt_buf2[1] = 0;
+  salt_buf2[2] = 0;
+  salt_buf2[3] = 0;
+  salt_buf3[0] = 0;
+  salt_buf3[1] = 0;
+  salt_buf3[2] = 0;
+  salt_buf3[3] = 0;
+
+  /**
+   * loop
+   */
+
+  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 w0_t[4];
+    u32x w1_t[4];
+    u32x w2_t[4];
+    u32x w3_t[4];
+
+    w0_t[0] = w0lr;
+    w0_t[1] = w0[1];
+    w0_t[2] = w0[2];
+    w0_t[3] = w0[3];
+    w1_t[0] = w1[0];
+    w1_t[1] = w1[1];
+    w1_t[2] = w1[2];
+    w1_t[3] = w1[3];
+    w2_t[0] = w2[0];
+    w2_t[1] = w2[1];
+    w2_t[2] = w2[2];
+    w2_t[3] = w2[3];
+    w3_t[0] = w3[0];
+    w3_t[1] = w3[1];
+    w3_t[2] = w3[2];
+    w3_t[3] = w3[3];
+
+    /**
+     * md5
+     */
+
+    u32x a = MD5M_A;
+    u32x b = MD5M_B;
+    u32x c = MD5M_C;
+    u32x d = MD5M_D;
+
+    MD5_STEP (MD5_Fo, a, b, c, d, w0_t[0], MD5C00, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w0_t[1], MD5C01, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w0_t[2], MD5C02, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w0_t[3], MD5C03, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w1_t[0], MD5C04, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w1_t[1], MD5C05, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w1_t[2], MD5C06, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w1_t[3], MD5C07, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w2_t[0], MD5C08, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w2_t[1], MD5C09, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w2_t[2], MD5C0a, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w2_t[3], MD5C0b, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w3_t[0], MD5C0c, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w3_t[1], MD5C0d, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w3_t[2], MD5C0e, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w3_t[3], MD5C0f, MD5S03);
+
+    MD5_STEP (MD5_Go, a, b, c, d, w0_t[1], MD5C10, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w1_t[2], MD5C11, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w2_t[3], MD5C12, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w0_t[0], MD5C13, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w1_t[1], MD5C14, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w2_t[2], MD5C15, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w3_t[3], MD5C16, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w1_t[0], MD5C17, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w2_t[1], MD5C18, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w3_t[2], MD5C19, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w0_t[3], MD5C1a, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w2_t[0], MD5C1b, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w3_t[1], MD5C1c, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w0_t[2], MD5C1d, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w1_t[3], MD5C1e, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w3_t[0], MD5C1f, MD5S13);
+
+    u32x t;
+
+    MD5_STEP (MD5_H1, a, b, c, d, w1_t[1], MD5C20, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w2_t[0], MD5C21, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w2_t[3], MD5C22, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w3_t[2], MD5C23, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w0_t[1], MD5C24, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w1_t[0], MD5C25, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w1_t[3], MD5C26, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w2_t[2], MD5C27, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w3_t[1], MD5C28, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w0_t[0], MD5C29, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w0_t[3], MD5C2a, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w1_t[2], MD5C2b, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w2_t[1], MD5C2c, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w3_t[0], MD5C2d, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w3_t[3], MD5C2e, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w0_t[2], MD5C2f, MD5S23);
+
+    MD5_STEP (MD5_I , a, b, c, d, w0_t[0], MD5C30, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w1_t[3], MD5C31, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w3_t[2], MD5C32, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w1_t[1], MD5C33, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w3_t[0], MD5C34, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w0_t[3], MD5C35, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w2_t[2], MD5C36, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w0_t[1], MD5C37, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w2_t[0], MD5C38, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w3_t[3], MD5C39, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w1_t[2], MD5C3a, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w3_t[1], MD5C3b, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w1_t[0], MD5C3c, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w2_t[3], MD5C3d, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w0_t[2], MD5C3e, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w2_t[1], MD5C3f, MD5S33);
+
+    a += make_u32x (MD5M_A);
+    b += make_u32x (MD5M_B);
+    c += make_u32x (MD5M_C);
+    d += make_u32x (MD5M_D);
+
+    w0_t[0] = uint_to_hex_lower8 ((a >>  0) & 255) <<  0
+            | uint_to_hex_lower8 ((a >>  8) & 255) << 16;
+    w0_t[1] = uint_to_hex_lower8 ((a >> 16) & 255) <<  0
+            | uint_to_hex_lower8 ((a >> 24) & 255) << 16;
+    w0_t[2] = uint_to_hex_lower8 ((b >>  0) & 255) <<  0
+            | uint_to_hex_lower8 ((b >>  8) & 255) << 16;
+    w0_t[3] = uint_to_hex_lower8 ((b >> 16) & 255) <<  0
+            | uint_to_hex_lower8 ((b >> 24) & 255) << 16;
+    w1_t[0] = uint_to_hex_lower8 ((c >>  0) & 255) <<  0
+            | uint_to_hex_lower8 ((c >>  8) & 255) << 16;
+    w1_t[1] = uint_to_hex_lower8 ((c >> 16) & 255) <<  0
+            | uint_to_hex_lower8 ((c >> 24) & 255) << 16;
+    w1_t[2] = uint_to_hex_lower8 ((d >>  0) & 255) <<  0
+            | uint_to_hex_lower8 ((d >>  8) & 255) << 16;
+    w1_t[3] = uint_to_hex_lower8 ((d >> 16) & 255) <<  0
+            | uint_to_hex_lower8 ((d >> 24) & 255) << 16;
+
+    w2_t[0] = 0x80;
+    w2_t[1] = 0;
+    w2_t[2] = 0;
+    w2_t[3] = 0;
+    w3_t[0] = 0;
+    w3_t[1] = 0;
+    w3_t[2] = 32 * 8;
+    w3_t[3] = 0;
+
+    a = MD5M_A;
+    b = MD5M_B;
+    c = MD5M_C;
+    d = MD5M_D;
+
+    MD5_STEP (MD5_Fo, a, b, c, d, w0_t[0], MD5C00, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w0_t[1], MD5C01, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w0_t[2], MD5C02, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w0_t[3], MD5C03, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w1_t[0], MD5C04, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w1_t[1], MD5C05, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w1_t[2], MD5C06, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w1_t[3], MD5C07, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w2_t[0], MD5C08, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w2_t[1], MD5C09, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w2_t[2], MD5C0a, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w2_t[3], MD5C0b, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w3_t[0], MD5C0c, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w3_t[1], MD5C0d, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w3_t[2], MD5C0e, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w3_t[3], MD5C0f, MD5S03);
+
+    MD5_STEP (MD5_Go, a, b, c, d, w0_t[1], MD5C10, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w1_t[2], MD5C11, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w2_t[3], MD5C12, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w0_t[0], MD5C13, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w1_t[1], MD5C14, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w2_t[2], MD5C15, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w3_t[3], MD5C16, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w1_t[0], MD5C17, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w2_t[1], MD5C18, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w3_t[2], MD5C19, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w0_t[3], MD5C1a, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w2_t[0], MD5C1b, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w3_t[1], MD5C1c, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w0_t[2], MD5C1d, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w1_t[3], MD5C1e, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w3_t[0], MD5C1f, MD5S13);
+
+    MD5_STEP (MD5_H1, a, b, c, d, w1_t[1], MD5C20, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w2_t[0], MD5C21, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w2_t[3], MD5C22, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w3_t[2], MD5C23, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w0_t[1], MD5C24, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w1_t[0], MD5C25, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w1_t[3], MD5C26, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w2_t[2], MD5C27, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w3_t[1], MD5C28, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w0_t[0], MD5C29, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w0_t[3], MD5C2a, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w1_t[2], MD5C2b, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w2_t[1], MD5C2c, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w3_t[0], MD5C2d, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w3_t[3], MD5C2e, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w0_t[2], MD5C2f, MD5S23);
+
+    MD5_STEP (MD5_I , a, b, c, d, w0_t[0], MD5C30, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w1_t[3], MD5C31, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w3_t[2], MD5C32, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w1_t[1], MD5C33, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w3_t[0], MD5C34, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w0_t[3], MD5C35, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w2_t[2], MD5C36, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w0_t[1], MD5C37, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w2_t[0], MD5C38, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w3_t[3], MD5C39, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w1_t[2], MD5C3a, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w3_t[1], MD5C3b, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w1_t[0], MD5C3c, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w2_t[3], MD5C3d, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w0_t[2], MD5C3e, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w2_t[1], MD5C3f, MD5S33);
+
+
+    a += make_u32x (MD5M_A);
+    b += make_u32x (MD5M_B);
+    c += make_u32x (MD5M_C);
+    d += make_u32x (MD5M_D);
+
+    w0_t[0] = salt_buf0[0];
+    w0_t[1] = salt_buf0[1];
+    w0_t[2] = salt_buf0[2];
+    w0_t[3] = salt_buf0[3];
+    w1_t[0] = salt_buf1[0];
+    w1_t[1] = salt_buf1[1];
+    w1_t[2] = salt_buf1[2];
+    w1_t[3] = salt_buf1[3];
+
+    w2_t[0] = uint_to_hex_lower8 ((a >>  0) & 255) <<  0
+            | uint_to_hex_lower8 ((a >>  8) & 255) << 16;
+    w2_t[1] = uint_to_hex_lower8 ((a >> 16) & 255) <<  0
+            | uint_to_hex_lower8 ((a >> 24) & 255) << 16;
+    w2_t[2] = uint_to_hex_lower8 ((b >>  0) & 255) <<  0
+            | uint_to_hex_lower8 ((b >>  8) & 255) << 16;
+    w2_t[3] = uint_to_hex_lower8 ((b >> 16) & 255) <<  0
+            | uint_to_hex_lower8 ((b >> 24) & 255) << 16;
+    w3_t[0] = uint_to_hex_lower8 ((c >>  0) & 255) <<  0
+            | uint_to_hex_lower8 ((c >>  8) & 255) << 16;
+    w3_t[1] = uint_to_hex_lower8 ((c >> 16) & 255) <<  0
+            | uint_to_hex_lower8 ((c >> 24) & 255) << 16;
+    w3_t[2] = uint_to_hex_lower8 ((d >>  0) & 255) <<  0
+            | uint_to_hex_lower8 ((d >>  8) & 255) << 16;
+    w3_t[3] = uint_to_hex_lower8 ((d >> 16) & 255) <<  0
+            | uint_to_hex_lower8 ((d >> 24) & 255) << 16;
+
+    a = MD5M_A;
+    b = MD5M_B;
+    c = MD5M_C;
+    d = MD5M_D;
+
+    MD5_STEP (MD5_Fo, a, b, c, d, w0_t[0], MD5C00, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w0_t[1], MD5C01, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w0_t[2], MD5C02, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w0_t[3], MD5C03, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w1_t[0], MD5C04, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w1_t[1], MD5C05, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w1_t[2], MD5C06, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w1_t[3], MD5C07, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w2_t[0], MD5C08, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w2_t[1], MD5C09, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w2_t[2], MD5C0a, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w2_t[3], MD5C0b, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w3_t[0], MD5C0c, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w3_t[1], MD5C0d, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w3_t[2], MD5C0e, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w3_t[3], MD5C0f, MD5S03);
+
+    MD5_STEP (MD5_Go, a, b, c, d, w0_t[1], MD5C10, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w1_t[2], MD5C11, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w2_t[3], MD5C12, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w0_t[0], MD5C13, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w1_t[1], MD5C14, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w2_t[2], MD5C15, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w3_t[3], MD5C16, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w1_t[0], MD5C17, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w2_t[1], MD5C18, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w3_t[2], MD5C19, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w0_t[3], MD5C1a, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w2_t[0], MD5C1b, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w3_t[1], MD5C1c, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w0_t[2], MD5C1d, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w1_t[3], MD5C1e, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w3_t[0], MD5C1f, MD5S13);
+
+    MD5_STEP (MD5_H1, a, b, c, d, w1_t[1], MD5C20, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w2_t[0], MD5C21, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w2_t[3], MD5C22, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w3_t[2], MD5C23, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w0_t[1], MD5C24, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w1_t[0], MD5C25, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w1_t[3], MD5C26, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w2_t[2], MD5C27, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w3_t[1], MD5C28, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w0_t[0], MD5C29, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w0_t[3], MD5C2a, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w1_t[2], MD5C2b, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w2_t[1], MD5C2c, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w3_t[0], MD5C2d, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w3_t[3], MD5C2e, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w0_t[2], MD5C2f, MD5S23);
+
+    MD5_STEP (MD5_I , a, b, c, d, w0_t[0], MD5C30, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w1_t[3], MD5C31, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w3_t[2], MD5C32, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w1_t[1], MD5C33, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w3_t[0], MD5C34, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w0_t[3], MD5C35, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w2_t[2], MD5C36, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w0_t[1], MD5C37, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w2_t[0], MD5C38, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w3_t[3], MD5C39, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w1_t[2], MD5C3a, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w3_t[1], MD5C3b, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w1_t[0], MD5C3c, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w2_t[3], MD5C3d, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w0_t[2], MD5C3e, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w2_t[1], MD5C3f, MD5S33);
+
+    const u32x r_a = a + make_u32x (MD5M_A);
+    const u32x r_b = b + make_u32x (MD5M_B);
+    const u32x r_c = c + make_u32x (MD5M_C);
+    const u32x r_d = d + make_u32x (MD5M_D);
+
+    const u32x r_00 = 0x80;
+    const u32x r_14 = 64 * 8;
+
+    a = r_a;
+    b = r_b;
+    c = r_c;
+    d = r_d;
+
+    MD5_STEP (MD5_Fo, a, b, c, d, r_00, MD5C00, MD5S00);
+    MD5_STEP0(MD5_Fo, d, a, b, c,       MD5C01, MD5S01);
+    MD5_STEP0(MD5_Fo, c, d, a, b,       MD5C02, MD5S02);
+    MD5_STEP0(MD5_Fo, b, c, d, a,       MD5C03, MD5S03);
+    MD5_STEP0(MD5_Fo, a, b, c, d,       MD5C04, MD5S00);
+    MD5_STEP0(MD5_Fo, d, a, b, c,       MD5C05, MD5S01);
+    MD5_STEP0(MD5_Fo, c, d, a, b,       MD5C06, MD5S02);
+    MD5_STEP0(MD5_Fo, b, c, d, a,       MD5C07, MD5S03);
+    MD5_STEP0(MD5_Fo, a, b, c, d,       MD5C08, MD5S00);
+    MD5_STEP0(MD5_Fo, d, a, b, c,       MD5C09, MD5S01);
+    MD5_STEP0(MD5_Fo, c, d, a, b,       MD5C0a, MD5S02);
+    MD5_STEP0(MD5_Fo, b, c, d, a,       MD5C0b, MD5S03);
+    MD5_STEP0(MD5_Fo, a, b, c, d,       MD5C0c, MD5S00);
+    MD5_STEP0(MD5_Fo, d, a, b, c,       MD5C0d, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, r_14, MD5C0e, MD5S02);
+    MD5_STEP0(MD5_Fo, b, c, d, a,       MD5C0f, MD5S03);
+
+    MD5_STEP0(MD5_Go, a, b, c, d,       MD5C10, MD5S10);
+    MD5_STEP0(MD5_Go, d, a, b, c,       MD5C11, MD5S11);
+    MD5_STEP0(MD5_Go, c, d, a, b,       MD5C12, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, r_00, MD5C13, MD5S13);
+    MD5_STEP0(MD5_Go, a, b, c, d,       MD5C14, MD5S10);
+    MD5_STEP0(MD5_Go, d, a, b, c,       MD5C15, MD5S11);
+    MD5_STEP0(MD5_Go, c, d, a, b,       MD5C16, MD5S12);
+    MD5_STEP0(MD5_Go, b, c, d, a,       MD5C17, MD5S13);
+    MD5_STEP0(MD5_Go, a, b, c, d,       MD5C18, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, r_14, MD5C19, MD5S11);
+    MD5_STEP0(MD5_Go, c, d, a, b,       MD5C1a, MD5S12);
+    MD5_STEP0(MD5_Go, b, c, d, a,       MD5C1b, MD5S13);
+    MD5_STEP0(MD5_Go, a, b, c, d,       MD5C1c, MD5S10);
+    MD5_STEP0(MD5_Go, d, a, b, c,       MD5C1d, MD5S11);
+    MD5_STEP0(MD5_Go, c, d, a, b,       MD5C1e, MD5S12);
+    MD5_STEP0(MD5_Go, b, c, d, a,       MD5C1f, MD5S13);
+
+    MD5_STEP0(MD5_H1, a, b, c, d,       MD5C20, MD5S20);
+    MD5_STEP0(MD5_H2, d, a, b, c,       MD5C21, MD5S21);
+    MD5_STEP0(MD5_H1, c, d, a, b,       MD5C22, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, r_14, MD5C23, MD5S23);
+    MD5_STEP0(MD5_H1, a, b, c, d,       MD5C24, MD5S20);
+    MD5_STEP0(MD5_H2, d, a, b, c,       MD5C25, MD5S21);
+    MD5_STEP0(MD5_H1, c, d, a, b,       MD5C26, MD5S22);
+    MD5_STEP0(MD5_H2, b, c, d, a,       MD5C27, MD5S23);
+    MD5_STEP0(MD5_H1, a, b, c, d,       MD5C28, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, r_00, MD5C29, MD5S21);
+    MD5_STEP0(MD5_H1, c, d, a, b,       MD5C2a, MD5S22);
+    MD5_STEP0(MD5_H2, b, c, d, a,       MD5C2b, MD5S23);
+    MD5_STEP0(MD5_H1, a, b, c, d,       MD5C2c, MD5S20);
+    MD5_STEP0(MD5_H2, d, a, b, c,       MD5C2d, MD5S21);
+    MD5_STEP0(MD5_H1, c, d, a, b,       MD5C2e, MD5S22);
+    MD5_STEP0(MD5_H2, b, c, d, a,       MD5C2f, MD5S23);
+
+    MD5_STEP (MD5_I , a, b, c, d, r_00, MD5C30, MD5S30);
+    MD5_STEP0(MD5_I , d, a, b, c,       MD5C31, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, r_14, MD5C32, MD5S32);
+    MD5_STEP0(MD5_I , b, c, d, a,       MD5C33, MD5S33);
+    MD5_STEP0(MD5_I , a, b, c, d,       MD5C34, MD5S30);
+    MD5_STEP0(MD5_I , d, a, b, c,       MD5C35, MD5S31);
+    MD5_STEP0(MD5_I , c, d, a, b,       MD5C36, MD5S32);
+    MD5_STEP0(MD5_I , b, c, d, a,       MD5C37, MD5S33);
+    MD5_STEP0(MD5_I , a, b, c, d,       MD5C38, MD5S30);
+    MD5_STEP0(MD5_I , d, a, b, c,       MD5C39, MD5S31);
+    MD5_STEP0(MD5_I , c, d, a, b,       MD5C3a, MD5S32);
+    MD5_STEP0(MD5_I , b, c, d, a,       MD5C3b, MD5S33);
+    MD5_STEP0(MD5_I , a, b, c, d,       MD5C3c, MD5S30);
+    MD5_STEP0(MD5_I , d, a, b, c,       MD5C3d, MD5S31);
+    MD5_STEP0(MD5_I , c, d, a, b,       MD5C3e, MD5S32);
+    MD5_STEP0(MD5_I , b, c, d, a,       MD5C3f, MD5S33);
+
+    a += r_a;
+    b += r_b;
+    c += r_c;
+    d += r_d;
+
+    COMPARE_M_SIMD (a, d, c, b);
+  }
+}
+
+DECLSPEC void m30500s (PRIVATE_AS u32 *w0, PRIVATE_AS u32 *w1, PRIVATE_AS u32 *w2, PRIVATE_AS u32 *w3, const u32 pw_len, KERN_ATTR_FUNC_BASIC (), LOCAL_AS u32 *l_bin2asc)
+{
+  /**
+   * modifiers are taken from args
+   */
+
+  /**
+   * salt
+   */
+
+  u32 salt_buf0[4];
+  u32 salt_buf1[4];
+  u32 salt_buf2[4];
+  u32 salt_buf3[4];
+
+  salt_buf0[0] = salt_bufs[SALT_POS_HOST].salt_buf_pc[0];
+  salt_buf0[1] = salt_bufs[SALT_POS_HOST].salt_buf_pc[1];
+  salt_buf0[2] = salt_bufs[SALT_POS_HOST].salt_buf_pc[2];
+  salt_buf0[3] = salt_bufs[SALT_POS_HOST].salt_buf_pc[3];
+  salt_buf1[0] = salt_bufs[SALT_POS_HOST].salt_buf_pc[4];
+  salt_buf1[1] = salt_bufs[SALT_POS_HOST].salt_buf_pc[5];
+  salt_buf1[2] = salt_bufs[SALT_POS_HOST].salt_buf_pc[6];
+  salt_buf1[3] = salt_bufs[SALT_POS_HOST].salt_buf_pc[7];
+  salt_buf2[0] = 0;
+  salt_buf2[1] = 0;
+  salt_buf2[2] = 0;
+  salt_buf2[3] = 0;
+  salt_buf3[0] = 0;
+  salt_buf3[1] = 0;
+  salt_buf3[2] = 0;
+  salt_buf3[3] = 0;
+
+  /**
+   * digest
+   */
+
+  const u32 search[4] =
+  {
+    digests_buf[DIGESTS_OFFSET_HOST].digest_buf[DGST_R0],
+    digests_buf[DIGESTS_OFFSET_HOST].digest_buf[DGST_R1],
+    digests_buf[DIGESTS_OFFSET_HOST].digest_buf[DGST_R2],
+    digests_buf[DIGESTS_OFFSET_HOST].digest_buf[DGST_R3]
+  };
+
+  /**
+   * loop
+   */
+
+  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 w0_t[4];
+    u32x w1_t[4];
+    u32x w2_t[4];
+    u32x w3_t[4];
+
+    w0_t[0] = w0lr;
+    w0_t[1] = w0[1];
+    w0_t[2] = w0[2];
+    w0_t[3] = w0[3];
+    w1_t[0] = w1[0];
+    w1_t[1] = w1[1];
+    w1_t[2] = w1[2];
+    w1_t[3] = w1[3];
+    w2_t[0] = w2[0];
+    w2_t[1] = w2[1];
+    w2_t[2] = w2[2];
+    w2_t[3] = w2[3];
+    w3_t[0] = w3[0];
+    w3_t[1] = w3[1];
+    w3_t[2] = w3[2];
+    w3_t[3] = w3[3];
+
+    /**
+     * md5
+     */
+
+    u32x a = MD5M_A;
+    u32x b = MD5M_B;
+    u32x c = MD5M_C;
+    u32x d = MD5M_D;
+
+    MD5_STEP (MD5_Fo, a, b, c, d, w0_t[0], MD5C00, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w0_t[1], MD5C01, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w0_t[2], MD5C02, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w0_t[3], MD5C03, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w1_t[0], MD5C04, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w1_t[1], MD5C05, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w1_t[2], MD5C06, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w1_t[3], MD5C07, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w2_t[0], MD5C08, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w2_t[1], MD5C09, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w2_t[2], MD5C0a, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w2_t[3], MD5C0b, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w3_t[0], MD5C0c, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w3_t[1], MD5C0d, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w3_t[2], MD5C0e, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w3_t[3], MD5C0f, MD5S03);
+
+    MD5_STEP (MD5_Go, a, b, c, d, w0_t[1], MD5C10, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w1_t[2], MD5C11, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w2_t[3], MD5C12, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w0_t[0], MD5C13, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w1_t[1], MD5C14, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w2_t[2], MD5C15, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w3_t[3], MD5C16, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w1_t[0], MD5C17, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w2_t[1], MD5C18, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w3_t[2], MD5C19, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w0_t[3], MD5C1a, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w2_t[0], MD5C1b, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w3_t[1], MD5C1c, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w0_t[2], MD5C1d, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w1_t[3], MD5C1e, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w3_t[0], MD5C1f, MD5S13);
+
+    u32x t;
+
+    MD5_STEP (MD5_H1, a, b, c, d, w1_t[1], MD5C20, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w2_t[0], MD5C21, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w2_t[3], MD5C22, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w3_t[2], MD5C23, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w0_t[1], MD5C24, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w1_t[0], MD5C25, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w1_t[3], MD5C26, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w2_t[2], MD5C27, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w3_t[1], MD5C28, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w0_t[0], MD5C29, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w0_t[3], MD5C2a, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w1_t[2], MD5C2b, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w2_t[1], MD5C2c, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w3_t[0], MD5C2d, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w3_t[3], MD5C2e, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w0_t[2], MD5C2f, MD5S23);
+
+    MD5_STEP (MD5_I , a, b, c, d, w0_t[0], MD5C30, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w1_t[3], MD5C31, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w3_t[2], MD5C32, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w1_t[1], MD5C33, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w3_t[0], MD5C34, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w0_t[3], MD5C35, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w2_t[2], MD5C36, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w0_t[1], MD5C37, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w2_t[0], MD5C38, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w3_t[3], MD5C39, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w1_t[2], MD5C3a, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w3_t[1], MD5C3b, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w1_t[0], MD5C3c, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w2_t[3], MD5C3d, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w0_t[2], MD5C3e, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w2_t[1], MD5C3f, MD5S33);
+
+    a += make_u32x (MD5M_A);
+    b += make_u32x (MD5M_B);
+    c += make_u32x (MD5M_C);
+    d += make_u32x (MD5M_D);
+
+    w0_t[0] = uint_to_hex_lower8 ((a >>  0) & 255) <<  0
+            | uint_to_hex_lower8 ((a >>  8) & 255) << 16;
+    w0_t[1] = uint_to_hex_lower8 ((a >> 16) & 255) <<  0
+            | uint_to_hex_lower8 ((a >> 24) & 255) << 16;
+    w0_t[2] = uint_to_hex_lower8 ((b >>  0) & 255) <<  0
+            | uint_to_hex_lower8 ((b >>  8) & 255) << 16;
+    w0_t[3] = uint_to_hex_lower8 ((b >> 16) & 255) <<  0
+            | uint_to_hex_lower8 ((b >> 24) & 255) << 16;
+    w1_t[0] = uint_to_hex_lower8 ((c >>  0) & 255) <<  0
+            | uint_to_hex_lower8 ((c >>  8) & 255) << 16;
+    w1_t[1] = uint_to_hex_lower8 ((c >> 16) & 255) <<  0
+            | uint_to_hex_lower8 ((c >> 24) & 255) << 16;
+    w1_t[2] = uint_to_hex_lower8 ((d >>  0) & 255) <<  0
+            | uint_to_hex_lower8 ((d >>  8) & 255) << 16;
+    w1_t[3] = uint_to_hex_lower8 ((d >> 16) & 255) <<  0
+            | uint_to_hex_lower8 ((d >> 24) & 255) << 16;
+
+    w2_t[0] = 0x80;
+    w2_t[1] = 0;
+    w2_t[2] = 0;
+    w2_t[3] = 0;
+    w3_t[0] = 0;
+    w3_t[1] = 0;
+    w3_t[2] = 32 * 8;
+    w3_t[3] = 0;
+
+    a = MD5M_A;
+    b = MD5M_B;
+    c = MD5M_C;
+    d = MD5M_D;
+
+    MD5_STEP (MD5_Fo, a, b, c, d, w0_t[0], MD5C00, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w0_t[1], MD5C01, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w0_t[2], MD5C02, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w0_t[3], MD5C03, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w1_t[0], MD5C04, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w1_t[1], MD5C05, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w1_t[2], MD5C06, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w1_t[3], MD5C07, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w2_t[0], MD5C08, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w2_t[1], MD5C09, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w2_t[2], MD5C0a, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w2_t[3], MD5C0b, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w3_t[0], MD5C0c, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w3_t[1], MD5C0d, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w3_t[2], MD5C0e, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w3_t[3], MD5C0f, MD5S03);
+
+    MD5_STEP (MD5_Go, a, b, c, d, w0_t[1], MD5C10, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w1_t[2], MD5C11, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w2_t[3], MD5C12, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w0_t[0], MD5C13, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w1_t[1], MD5C14, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w2_t[2], MD5C15, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w3_t[3], MD5C16, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w1_t[0], MD5C17, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w2_t[1], MD5C18, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w3_t[2], MD5C19, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w0_t[3], MD5C1a, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w2_t[0], MD5C1b, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w3_t[1], MD5C1c, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w0_t[2], MD5C1d, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w1_t[3], MD5C1e, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w3_t[0], MD5C1f, MD5S13);
+
+    MD5_STEP (MD5_H1, a, b, c, d, w1_t[1], MD5C20, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w2_t[0], MD5C21, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w2_t[3], MD5C22, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w3_t[2], MD5C23, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w0_t[1], MD5C24, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w1_t[0], MD5C25, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w1_t[3], MD5C26, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w2_t[2], MD5C27, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w3_t[1], MD5C28, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w0_t[0], MD5C29, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w0_t[3], MD5C2a, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w1_t[2], MD5C2b, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w2_t[1], MD5C2c, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w3_t[0], MD5C2d, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w3_t[3], MD5C2e, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w0_t[2], MD5C2f, MD5S23);
+
+    MD5_STEP (MD5_I , a, b, c, d, w0_t[0], MD5C30, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w1_t[3], MD5C31, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w3_t[2], MD5C32, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w1_t[1], MD5C33, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w3_t[0], MD5C34, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w0_t[3], MD5C35, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w2_t[2], MD5C36, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w0_t[1], MD5C37, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w2_t[0], MD5C38, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w3_t[3], MD5C39, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w1_t[2], MD5C3a, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w3_t[1], MD5C3b, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w1_t[0], MD5C3c, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w2_t[3], MD5C3d, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w0_t[2], MD5C3e, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w2_t[1], MD5C3f, MD5S33);
+
+
+    a += make_u32x (MD5M_A);
+    b += make_u32x (MD5M_B);
+    c += make_u32x (MD5M_C);
+    d += make_u32x (MD5M_D);
+
+    w0_t[0] = salt_buf0[0];
+    w0_t[1] = salt_buf0[1];
+    w0_t[2] = salt_buf0[2];
+    w0_t[3] = salt_buf0[3];
+    w1_t[0] = salt_buf1[0];
+    w1_t[1] = salt_buf1[1];
+    w1_t[2] = salt_buf1[2];
+    w1_t[3] = salt_buf1[3];
+
+    w2_t[0] = uint_to_hex_lower8 ((a >>  0) & 255) <<  0
+            | uint_to_hex_lower8 ((a >>  8) & 255) << 16;
+    w2_t[1] = uint_to_hex_lower8 ((a >> 16) & 255) <<  0
+            | uint_to_hex_lower8 ((a >> 24) & 255) << 16;
+    w2_t[2] = uint_to_hex_lower8 ((b >>  0) & 255) <<  0
+            | uint_to_hex_lower8 ((b >>  8) & 255) << 16;
+    w2_t[3] = uint_to_hex_lower8 ((b >> 16) & 255) <<  0
+            | uint_to_hex_lower8 ((b >> 24) & 255) << 16;
+    w3_t[0] = uint_to_hex_lower8 ((c >>  0) & 255) <<  0
+            | uint_to_hex_lower8 ((c >>  8) & 255) << 16;
+    w3_t[1] = uint_to_hex_lower8 ((c >> 16) & 255) <<  0
+            | uint_to_hex_lower8 ((c >> 24) & 255) << 16;
+    w3_t[2] = uint_to_hex_lower8 ((d >>  0) & 255) <<  0
+            | uint_to_hex_lower8 ((d >>  8) & 255) << 16;
+    w3_t[3] = uint_to_hex_lower8 ((d >> 16) & 255) <<  0
+            | uint_to_hex_lower8 ((d >> 24) & 255) << 16;
+
+    a = MD5M_A;
+    b = MD5M_B;
+    c = MD5M_C;
+    d = MD5M_D;
+
+    MD5_STEP (MD5_Fo, a, b, c, d, w0_t[0], MD5C00, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w0_t[1], MD5C01, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w0_t[2], MD5C02, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w0_t[3], MD5C03, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w1_t[0], MD5C04, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w1_t[1], MD5C05, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w1_t[2], MD5C06, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w1_t[3], MD5C07, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w2_t[0], MD5C08, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w2_t[1], MD5C09, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w2_t[2], MD5C0a, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w2_t[3], MD5C0b, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w3_t[0], MD5C0c, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w3_t[1], MD5C0d, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w3_t[2], MD5C0e, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w3_t[3], MD5C0f, MD5S03);
+
+    MD5_STEP (MD5_Go, a, b, c, d, w0_t[1], MD5C10, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w1_t[2], MD5C11, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w2_t[3], MD5C12, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w0_t[0], MD5C13, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w1_t[1], MD5C14, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w2_t[2], MD5C15, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w3_t[3], MD5C16, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w1_t[0], MD5C17, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w2_t[1], MD5C18, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w3_t[2], MD5C19, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w0_t[3], MD5C1a, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w2_t[0], MD5C1b, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w3_t[1], MD5C1c, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w0_t[2], MD5C1d, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w1_t[3], MD5C1e, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w3_t[0], MD5C1f, MD5S13);
+
+    MD5_STEP (MD5_H1, a, b, c, d, w1_t[1], MD5C20, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w2_t[0], MD5C21, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w2_t[3], MD5C22, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w3_t[2], MD5C23, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w0_t[1], MD5C24, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w1_t[0], MD5C25, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w1_t[3], MD5C26, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w2_t[2], MD5C27, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w3_t[1], MD5C28, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w0_t[0], MD5C29, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w0_t[3], MD5C2a, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w1_t[2], MD5C2b, MD5S23);
+    MD5_STEP (MD5_H1, a, b, c, d, w2_t[1], MD5C2c, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, w3_t[0], MD5C2d, MD5S21);
+    MD5_STEP (MD5_H1, c, d, a, b, w3_t[3], MD5C2e, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, w0_t[2], MD5C2f, MD5S23);
+
+    MD5_STEP (MD5_I , a, b, c, d, w0_t[0], MD5C30, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w1_t[3], MD5C31, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w3_t[2], MD5C32, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w1_t[1], MD5C33, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w3_t[0], MD5C34, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w0_t[3], MD5C35, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w2_t[2], MD5C36, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w0_t[1], MD5C37, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w2_t[0], MD5C38, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w3_t[3], MD5C39, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w1_t[2], MD5C3a, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w3_t[1], MD5C3b, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w1_t[0], MD5C3c, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w2_t[3], MD5C3d, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w0_t[2], MD5C3e, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w2_t[1], MD5C3f, MD5S33);
+
+    const u32x r_a = a + make_u32x (MD5M_A);
+    const u32x r_b = b + make_u32x (MD5M_B);
+    const u32x r_c = c + make_u32x (MD5M_C);
+    const u32x r_d = d + make_u32x (MD5M_D);
+
+    const u32x r_00 = 0x80;
+    const u32x r_14 = 64 * 8;
+
+    a = r_a;
+    b = r_b;
+    c = r_c;
+    d = r_d;
+
+    MD5_STEP (MD5_Fo, a, b, c, d, r_00, MD5C00, MD5S00);
+    MD5_STEP0(MD5_Fo, d, a, b, c,       MD5C01, MD5S01);
+    MD5_STEP0(MD5_Fo, c, d, a, b,       MD5C02, MD5S02);
+    MD5_STEP0(MD5_Fo, b, c, d, a,       MD5C03, MD5S03);
+    MD5_STEP0(MD5_Fo, a, b, c, d,       MD5C04, MD5S00);
+    MD5_STEP0(MD5_Fo, d, a, b, c,       MD5C05, MD5S01);
+    MD5_STEP0(MD5_Fo, c, d, a, b,       MD5C06, MD5S02);
+    MD5_STEP0(MD5_Fo, b, c, d, a,       MD5C07, MD5S03);
+    MD5_STEP0(MD5_Fo, a, b, c, d,       MD5C08, MD5S00);
+    MD5_STEP0(MD5_Fo, d, a, b, c,       MD5C09, MD5S01);
+    MD5_STEP0(MD5_Fo, c, d, a, b,       MD5C0a, MD5S02);
+    MD5_STEP0(MD5_Fo, b, c, d, a,       MD5C0b, MD5S03);
+    MD5_STEP0(MD5_Fo, a, b, c, d,       MD5C0c, MD5S00);
+    MD5_STEP0(MD5_Fo, d, a, b, c,       MD5C0d, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, r_14, MD5C0e, MD5S02);
+    MD5_STEP0(MD5_Fo, b, c, d, a,       MD5C0f, MD5S03);
+
+    MD5_STEP0(MD5_Go, a, b, c, d,       MD5C10, MD5S10);
+    MD5_STEP0(MD5_Go, d, a, b, c,       MD5C11, MD5S11);
+    MD5_STEP0(MD5_Go, c, d, a, b,       MD5C12, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, r_00, MD5C13, MD5S13);
+    MD5_STEP0(MD5_Go, a, b, c, d,       MD5C14, MD5S10);
+    MD5_STEP0(MD5_Go, d, a, b, c,       MD5C15, MD5S11);
+    MD5_STEP0(MD5_Go, c, d, a, b,       MD5C16, MD5S12);
+    MD5_STEP0(MD5_Go, b, c, d, a,       MD5C17, MD5S13);
+    MD5_STEP0(MD5_Go, a, b, c, d,       MD5C18, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, r_14, MD5C19, MD5S11);
+    MD5_STEP0(MD5_Go, c, d, a, b,       MD5C1a, MD5S12);
+    MD5_STEP0(MD5_Go, b, c, d, a,       MD5C1b, MD5S13);
+    MD5_STEP0(MD5_Go, a, b, c, d,       MD5C1c, MD5S10);
+    MD5_STEP0(MD5_Go, d, a, b, c,       MD5C1d, MD5S11);
+    MD5_STEP0(MD5_Go, c, d, a, b,       MD5C1e, MD5S12);
+    MD5_STEP0(MD5_Go, b, c, d, a,       MD5C1f, MD5S13);
+
+    MD5_STEP0(MD5_H1, a, b, c, d,       MD5C20, MD5S20);
+    MD5_STEP0(MD5_H2, d, a, b, c,       MD5C21, MD5S21);
+    MD5_STEP0(MD5_H1, c, d, a, b,       MD5C22, MD5S22);
+    MD5_STEP (MD5_H2, b, c, d, a, r_14, MD5C23, MD5S23);
+    MD5_STEP0(MD5_H1, a, b, c, d,       MD5C24, MD5S20);
+    MD5_STEP0(MD5_H2, d, a, b, c,       MD5C25, MD5S21);
+    MD5_STEP0(MD5_H1, c, d, a, b,       MD5C26, MD5S22);
+    MD5_STEP0(MD5_H2, b, c, d, a,       MD5C27, MD5S23);
+    MD5_STEP0(MD5_H1, a, b, c, d,       MD5C28, MD5S20);
+    MD5_STEP (MD5_H2, d, a, b, c, r_00, MD5C29, MD5S21);
+    MD5_STEP0(MD5_H1, c, d, a, b,       MD5C2a, MD5S22);
+    MD5_STEP0(MD5_H2, b, c, d, a,       MD5C2b, MD5S23);
+    MD5_STEP0(MD5_H1, a, b, c, d,       MD5C2c, MD5S20);
+    MD5_STEP0(MD5_H2, d, a, b, c,       MD5C2d, MD5S21);
+    MD5_STEP0(MD5_H1, c, d, a, b,       MD5C2e, MD5S22);
+    MD5_STEP0(MD5_H2, b, c, d, a,       MD5C2f, MD5S23);
+
+    MD5_STEP (MD5_I , a, b, c, d, r_00, MD5C30, MD5S30);
+    MD5_STEP0(MD5_I , d, a, b, c,       MD5C31, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, r_14, MD5C32, MD5S32);
+    MD5_STEP0(MD5_I , b, c, d, a,       MD5C33, MD5S33);
+    MD5_STEP0(MD5_I , a, b, c, d,       MD5C34, MD5S30);
+    MD5_STEP0(MD5_I , d, a, b, c,       MD5C35, MD5S31);
+    MD5_STEP0(MD5_I , c, d, a, b,       MD5C36, MD5S32);
+    MD5_STEP0(MD5_I , b, c, d, a,       MD5C37, MD5S33);
+    MD5_STEP0(MD5_I , a, b, c, d,       MD5C38, MD5S30);
+    MD5_STEP0(MD5_I , d, a, b, c,       MD5C39, MD5S31);
+    MD5_STEP0(MD5_I , c, d, a, b,       MD5C3a, MD5S32);
+    MD5_STEP0(MD5_I , b, c, d, a,       MD5C3b, MD5S33);
+    MD5_STEP0(MD5_I , a, b, c, d,       MD5C3c, MD5S30);
+
+    if (MATCHES_NONE_VS ((a + r_a), search[0])) continue;
+
+    MD5_STEP0(MD5_I , d, a, b, c,       MD5C3d, MD5S31);
+    MD5_STEP0(MD5_I , c, d, a, b,       MD5C3e, MD5S32);
+    MD5_STEP0(MD5_I , b, c, d, a,       MD5C3f, MD5S33);
+
+    a += r_a;
+    b += r_b;
+    c += r_c;
+    d += r_d;
+
+    COMPARE_S_SIMD (a, d, c, b);
+  }
+}
+
+KERNEL_FQ void m30500_m04 (KERN_ATTR_BASIC ())
+{
+  /**
+   * base
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * modifier
+   */
+
+  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] = pws[gid].i[14];
+  w3[3] = 0;
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * bin2asc table
+   */
+
+  LOCAL_VK u32 l_bin2asc[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    const u32 i0 = (i >> 0) & 15;
+    const u32 i1 = (i >> 4) & 15;
+
+    l_bin2asc[i] = ((i0 < 10) ? '0' + i0 : 'a' - 10 + i0) << 8
+                 | ((i1 < 10) ? '0' + i1 : 'a' - 10 + i1) << 0;
+  }
+
+  SYNC_THREADS ();
+
+  if (gid >= GID_CNT) return;
+
+  /**
+   * main
+   */
+
+  m30500m (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, kernel_param, gid, lid, lsz, l_bin2asc);
+}
+
+KERNEL_FQ void m30500_m08 (KERN_ATTR_BASIC ())
+{
+  /**
+   * base
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * modifier
+   */
+
+  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] = pws[gid].i[14];
+  w3[3] = 0;
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * bin2asc table
+   */
+
+  LOCAL_VK u32 l_bin2asc[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    const u32 i0 = (i >> 0) & 15;
+    const u32 i1 = (i >> 4) & 15;
+
+    l_bin2asc[i] = ((i0 < 10) ? '0' + i0 : 'a' - 10 + i0) << 8
+                 | ((i1 < 10) ? '0' + i1 : 'a' - 10 + i1) << 0;
+  }
+
+  SYNC_THREADS ();
+
+  if (gid >= GID_CNT) return;
+
+  /**
+   * main
+   */
+
+  m30500m (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, kernel_param, gid, lid, lsz, l_bin2asc);
+}
+
+KERNEL_FQ void m30500_m16 (KERN_ATTR_BASIC ())
+{
+  /**
+   * base
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * modifier
+   */
+
+  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] = pws[gid].i[14];
+  w3[3] = pws[gid].i[15];
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * bin2asc table
+   */
+
+  LOCAL_VK u32 l_bin2asc[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    const u32 i0 = (i >> 0) & 15;
+    const u32 i1 = (i >> 4) & 15;
+
+    l_bin2asc[i] = ((i0 < 10) ? '0' + i0 : 'a' - 10 + i0) << 8
+                 | ((i1 < 10) ? '0' + i1 : 'a' - 10 + i1) << 0;
+  }
+
+  SYNC_THREADS ();
+
+  if (gid >= GID_CNT) return;
+
+  /**
+   * main
+   */
+
+  m30500m (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, kernel_param, gid, lid, lsz, l_bin2asc);
+}
+
+KERNEL_FQ void m30500_s04 (KERN_ATTR_BASIC ())
+{
+  /**
+   * base
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * modifier
+   */
+
+  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] = pws[gid].i[14];
+  w3[3] = 0;
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * bin2asc table
+   */
+
+  LOCAL_VK u32 l_bin2asc[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    const u32 i0 = (i >> 0) & 15;
+    const u32 i1 = (i >> 4) & 15;
+
+    l_bin2asc[i] = ((i0 < 10) ? '0' + i0 : 'a' - 10 + i0) << 8
+                 | ((i1 < 10) ? '0' + i1 : 'a' - 10 + i1) << 0;
+  }
+
+  SYNC_THREADS ();
+
+  if (gid >= GID_CNT) return;
+
+  /**
+   * main
+   */
+
+  m30500s (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, kernel_param, gid, lid, lsz, l_bin2asc);
+}
+
+KERNEL_FQ void m30500_s08 (KERN_ATTR_BASIC ())
+{
+  /**
+   * base
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * modifier
+   */
+
+  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] = pws[gid].i[14];
+  w3[3] = 0;
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * bin2asc table
+   */
+
+  LOCAL_VK u32 l_bin2asc[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    const u32 i0 = (i >> 0) & 15;
+    const u32 i1 = (i >> 4) & 15;
+
+    l_bin2asc[i] = ((i0 < 10) ? '0' + i0 : 'a' - 10 + i0) << 8
+                 | ((i1 < 10) ? '0' + i1 : 'a' - 10 + i1) << 0;
+  }
+
+  SYNC_THREADS ();
+
+  if (gid >= GID_CNT) return;
+
+  /**
+   * main
+   */
+
+  m30500s (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, kernel_param, gid, lid, lsz, l_bin2asc);
+}
+
+KERNEL_FQ void m30500_s16 (KERN_ATTR_BASIC ())
+{
+  /**
+   * base
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * modifier
+   */
+
+  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] = pws[gid].i[14];
+  w3[3] = pws[gid].i[15];
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * bin2asc table
+   */
+
+  LOCAL_VK u32 l_bin2asc[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    const u32 i0 = (i >> 0) & 15;
+    const u32 i1 = (i >> 4) & 15;
+
+    l_bin2asc[i] = ((i0 < 10) ? '0' + i0 : 'a' - 10 + i0) << 8
+                 | ((i1 < 10) ? '0' + i1 : 'a' - 10 + i1) << 0;
+  }
+
+  SYNC_THREADS ();
+
+  if (gid >= GID_CNT) return;
+
+  /**
+   * main
+   */
+
+  m30500s (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, kernel_param, gid, lid, lsz, l_bin2asc);
+}
diff --git a/OpenCL/m30500_a3-pure.cl b/OpenCL/m30500_a3-pure.cl
new file mode 100644
index 000000000..ad3e5da9f
--- /dev/null
+++ b/OpenCL/m30500_a3-pure.cl
@@ -0,0 +1,373 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include M2S(INCLUDE_PATH/inc_vendor.h)
+#include M2S(INCLUDE_PATH/inc_types.h)
+#include M2S(INCLUDE_PATH/inc_platform.cl)
+#include M2S(INCLUDE_PATH/inc_common.cl)
+#include M2S(INCLUDE_PATH/inc_simd.cl)
+#include M2S(INCLUDE_PATH/inc_hash_md5.cl)
+#endif
+
+#if   VECT_SIZE == 1
+#define uint_to_hex_lower8(i) make_u32x (l_bin2asc[(i)])
+#elif VECT_SIZE == 2
+#define uint_to_hex_lower8(i) make_u32x (l_bin2asc[(i).s0], l_bin2asc[(i).s1])
+#elif VECT_SIZE == 4
+#define uint_to_hex_lower8(i) make_u32x (l_bin2asc[(i).s0], l_bin2asc[(i).s1], l_bin2asc[(i).s2], l_bin2asc[(i).s3])
+#elif VECT_SIZE == 8
+#define uint_to_hex_lower8(i) make_u32x (l_bin2asc[(i).s0], l_bin2asc[(i).s1], l_bin2asc[(i).s2], l_bin2asc[(i).s3], l_bin2asc[(i).s4], l_bin2asc[(i).s5], l_bin2asc[(i).s6], l_bin2asc[(i).s7])
+#elif VECT_SIZE == 16
+#define uint_to_hex_lower8(i) make_u32x (l_bin2asc[(i).s0], l_bin2asc[(i).s1], l_bin2asc[(i).s2], l_bin2asc[(i).s3], l_bin2asc[(i).s4], l_bin2asc[(i).s5], l_bin2asc[(i).s6], l_bin2asc[(i).s7], l_bin2asc[(i).s8], l_bin2asc[(i).s9], l_bin2asc[(i).sa], l_bin2asc[(i).sb], l_bin2asc[(i).sc], l_bin2asc[(i).sd], l_bin2asc[(i).se], l_bin2asc[(i).sf])
+#endif
+
+KERNEL_FQ void m30500_mxx (KERN_ATTR_VECTOR ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * bin2asc table
+   */
+
+  LOCAL_VK u32 l_bin2asc[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    const u32 i0 = (i >> 0) & 15;
+    const u32 i1 = (i >> 4) & 15;
+
+    l_bin2asc[i] = ((i0 < 10) ? '0' + i0 : 'a' - 10 + i0) << 8
+                 | ((i1 < 10) ? '0' + i1 : 'a' - 10 + i1) << 0;
+  }
+
+  SYNC_THREADS ();
+
+  if (gid >= GID_CNT) return;
+
+  /**
+   * 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];
+  }
+
+  const u32 salt_len = 32;
+
+  u32 s[8] = { 0 };
+
+  for (u32 i = 0, idx = 0; i < salt_len; i += 4, idx += 1)
+  {
+    s[idx] = salt_bufs[SALT_POS_HOST].salt_buf_pc[idx];
+  }
+
+  /**
+   * 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 w0lr = w0l | w0r;
+
+    w[0] = w0lr;
+
+    md5_ctx_vector_t ctx0;
+
+    md5_init_vector (&ctx0);
+
+    md5_update_vector (&ctx0, w, pw_len);
+
+    md5_final_vector (&ctx0);
+
+    u32x a = ctx0.h[0];
+    u32x b = ctx0.h[1];
+    u32x c = ctx0.h[2];
+    u32x d = ctx0.h[3];
+
+    md5_ctx_vector_t ctx;
+
+    md5_init_vector (&ctx);
+
+    ctx.w0[0] = uint_to_hex_lower8 ((a >>  0) & 255) <<  0
+              | uint_to_hex_lower8 ((a >>  8) & 255) << 16;
+    ctx.w0[1] = uint_to_hex_lower8 ((a >> 16) & 255) <<  0
+              | uint_to_hex_lower8 ((a >> 24) & 255) << 16;
+    ctx.w0[2] = uint_to_hex_lower8 ((b >>  0) & 255) <<  0
+              | uint_to_hex_lower8 ((b >>  8) & 255) << 16;
+    ctx.w0[3] = uint_to_hex_lower8 ((b >> 16) & 255) <<  0
+              | uint_to_hex_lower8 ((b >> 24) & 255) << 16;
+    ctx.w1[0] = uint_to_hex_lower8 ((c >>  0) & 255) <<  0
+              | uint_to_hex_lower8 ((c >>  8) & 255) << 16;
+    ctx.w1[1] = uint_to_hex_lower8 ((c >> 16) & 255) <<  0
+              | uint_to_hex_lower8 ((c >> 24) & 255) << 16;
+    ctx.w1[2] = uint_to_hex_lower8 ((d >>  0) & 255) <<  0
+              | uint_to_hex_lower8 ((d >>  8) & 255) << 16;
+    ctx.w1[3] = uint_to_hex_lower8 ((d >> 16) & 255) <<  0
+              | uint_to_hex_lower8 ((d >> 24) & 255) << 16;
+
+    ctx.len = 32;
+
+    md5_final_vector (&ctx);
+
+    a = ctx.h[0];
+    b = ctx.h[1];
+    c = ctx.h[2];
+    d = ctx.h[3];
+
+    md5_ctx_vector_t ctx1;
+
+    md5_init_vector (&ctx1);
+
+    ctx1.w0[0] = s[0];
+    ctx1.w0[1] = s[1];
+    ctx1.w0[2] = s[2];
+    ctx1.w0[3] = s[3];
+    ctx1.w1[0] = s[4];
+    ctx1.w1[1] = s[5];
+    ctx1.w1[2] = s[6];
+    ctx1.w1[3] = s[7];
+    ctx1.w2[0] = uint_to_hex_lower8 ((a >>  0) & 255) <<  0
+               | uint_to_hex_lower8 ((a >>  8) & 255) << 16;
+    ctx1.w2[1] = uint_to_hex_lower8 ((a >> 16) & 255) <<  0
+               | uint_to_hex_lower8 ((a >> 24) & 255) << 16;
+    ctx1.w2[2] = uint_to_hex_lower8 ((b >>  0) & 255) <<  0
+               | uint_to_hex_lower8 ((b >>  8) & 255) << 16;
+    ctx1.w2[3] = uint_to_hex_lower8 ((b >> 16) & 255) <<  0
+               | uint_to_hex_lower8 ((b >> 24) & 255) << 16;
+    ctx1.w3[0] = uint_to_hex_lower8 ((c >>  0) & 255) <<  0
+               | uint_to_hex_lower8 ((c >>  8) & 255) << 16;
+    ctx1.w3[1] = uint_to_hex_lower8 ((c >> 16) & 255) <<  0
+               | uint_to_hex_lower8 ((c >> 24) & 255) << 16;
+    ctx1.w3[2] = uint_to_hex_lower8 ((d >>  0) & 255) <<  0
+               | uint_to_hex_lower8 ((d >>  8) & 255) << 16;
+    ctx1.w3[3] = uint_to_hex_lower8 ((d >> 16) & 255) <<  0
+               | uint_to_hex_lower8 ((d >> 24) & 255) << 16;
+
+    md5_transform_vector (ctx1.w0, ctx1.w1, ctx1.w2, ctx1.w3, ctx1.h);
+
+    ctx1.w0[0] = 0x80;
+    ctx1.w0[1] = 0;
+    ctx1.w0[2] = 0;
+    ctx1.w0[3] = 0;
+    ctx1.w1[0] = 0;
+    ctx1.w1[1] = 0;
+    ctx1.w1[2] = 0;
+    ctx1.w1[3] = 0;
+    ctx1.w2[0] = 0;
+    ctx1.w2[1] = 0;
+    ctx1.w2[2] = 0;
+    ctx1.w2[3] = 0;
+    ctx1.w3[0] = 0;
+    ctx1.w3[1] = 0;
+    ctx1.w3[2] = 64 * 8;
+    ctx1.w3[3] = 0;
+
+    md5_transform_vector (ctx1.w0, ctx1.w1, ctx1.w2, ctx1.w3, ctx1.h);
+
+    const u32x r0 = ctx1.h[DGST_R0];
+    const u32x r1 = ctx1.h[DGST_R1];
+    const u32x r2 = ctx1.h[DGST_R2];
+    const u32x r3 = ctx1.h[DGST_R3];
+
+    COMPARE_M_SIMD (r0, r1, r2, r3);
+  }
+}
+
+KERNEL_FQ void m30500_sxx (KERN_ATTR_VECTOR ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * bin2asc table
+   */
+
+  LOCAL_VK u32 l_bin2asc[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    const u32 i0 = (i >> 0) & 15;
+    const u32 i1 = (i >> 4) & 15;
+
+    l_bin2asc[i] = ((i0 < 10) ? '0' + i0 : 'a' - 10 + i0) << 8
+                 | ((i1 < 10) ? '0' + i1 : 'a' - 10 + i1) << 0;
+  }
+
+  SYNC_THREADS ();
+
+  if (gid >= GID_CNT) return;
+
+  /**
+   * digest
+   */
+
+  const u32 search[4] =
+  {
+    digests_buf[DIGESTS_OFFSET_HOST].digest_buf[DGST_R0],
+    digests_buf[DIGESTS_OFFSET_HOST].digest_buf[DGST_R1],
+    digests_buf[DIGESTS_OFFSET_HOST].digest_buf[DGST_R2],
+    digests_buf[DIGESTS_OFFSET_HOST].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];
+  }
+
+  const u32 salt_len = 32;
+
+  u32 s[8] = { 0 };
+
+  for (u32 i = 0, idx = 0; i < salt_len; i += 4, idx += 1)
+  {
+    s[idx] = salt_bufs[SALT_POS_HOST].salt_buf_pc[idx];
+  }
+
+  /**
+   * 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 w0lr = w0l | w0r;
+
+    w[0] = w0lr;
+
+    md5_ctx_vector_t ctx0;
+
+    md5_init_vector (&ctx0);
+
+    md5_update_vector (&ctx0, w, pw_len);
+
+    md5_final_vector (&ctx0);
+
+    u32x a = ctx0.h[0];
+    u32x b = ctx0.h[1];
+    u32x c = ctx0.h[2];
+    u32x d = ctx0.h[3];
+
+    md5_ctx_vector_t ctx;
+
+    md5_init_vector (&ctx);
+
+    ctx.w0[0] = uint_to_hex_lower8 ((a >>  0) & 255) <<  0
+              | uint_to_hex_lower8 ((a >>  8) & 255) << 16;
+    ctx.w0[1] = uint_to_hex_lower8 ((a >> 16) & 255) <<  0
+              | uint_to_hex_lower8 ((a >> 24) & 255) << 16;
+    ctx.w0[2] = uint_to_hex_lower8 ((b >>  0) & 255) <<  0
+              | uint_to_hex_lower8 ((b >>  8) & 255) << 16;
+    ctx.w0[3] = uint_to_hex_lower8 ((b >> 16) & 255) <<  0
+              | uint_to_hex_lower8 ((b >> 24) & 255) << 16;
+    ctx.w1[0] = uint_to_hex_lower8 ((c >>  0) & 255) <<  0
+              | uint_to_hex_lower8 ((c >>  8) & 255) << 16;
+    ctx.w1[1] = uint_to_hex_lower8 ((c >> 16) & 255) <<  0
+              | uint_to_hex_lower8 ((c >> 24) & 255) << 16;
+    ctx.w1[2] = uint_to_hex_lower8 ((d >>  0) & 255) <<  0
+              | uint_to_hex_lower8 ((d >>  8) & 255) << 16;
+    ctx.w1[3] = uint_to_hex_lower8 ((d >> 16) & 255) <<  0
+              | uint_to_hex_lower8 ((d >> 24) & 255) << 16;
+
+    ctx.len = 32;
+
+    md5_final_vector (&ctx);
+
+    a = ctx.h[0];
+    b = ctx.h[1];
+    c = ctx.h[2];
+    d = ctx.h[3];
+
+    md5_ctx_vector_t ctx1;
+
+    md5_init_vector (&ctx1);
+
+    ctx1.w0[0] = s[0];
+    ctx1.w0[1] = s[1];
+    ctx1.w0[2] = s[2];
+    ctx1.w0[3] = s[3];
+    ctx1.w1[0] = s[4];
+    ctx1.w1[1] = s[5];
+    ctx1.w1[2] = s[6];
+    ctx1.w1[3] = s[7];
+    ctx1.w2[0] = uint_to_hex_lower8 ((a >>  0) & 255) <<  0
+               | uint_to_hex_lower8 ((a >>  8) & 255) << 16;
+    ctx1.w2[1] = uint_to_hex_lower8 ((a >> 16) & 255) <<  0
+               | uint_to_hex_lower8 ((a >> 24) & 255) << 16;
+    ctx1.w2[2] = uint_to_hex_lower8 ((b >>  0) & 255) <<  0
+               | uint_to_hex_lower8 ((b >>  8) & 255) << 16;
+    ctx1.w2[3] = uint_to_hex_lower8 ((b >> 16) & 255) <<  0
+               | uint_to_hex_lower8 ((b >> 24) & 255) << 16;
+    ctx1.w3[0] = uint_to_hex_lower8 ((c >>  0) & 255) <<  0
+               | uint_to_hex_lower8 ((c >>  8) & 255) << 16;
+    ctx1.w3[1] = uint_to_hex_lower8 ((c >> 16) & 255) <<  0
+               | uint_to_hex_lower8 ((c >> 24) & 255) << 16;
+    ctx1.w3[2] = uint_to_hex_lower8 ((d >>  0) & 255) <<  0
+               | uint_to_hex_lower8 ((d >>  8) & 255) << 16;
+    ctx1.w3[3] = uint_to_hex_lower8 ((d >> 16) & 255) <<  0
+               | uint_to_hex_lower8 ((d >> 24) & 255) << 16;
+
+    md5_transform_vector (ctx1.w0, ctx1.w1, ctx1.w2, ctx1.w3, ctx1.h);
+
+    ctx1.w0[0] = 0x80;
+    ctx1.w0[1] = 0;
+    ctx1.w0[2] = 0;
+    ctx1.w0[3] = 0;
+    ctx1.w1[0] = 0;
+    ctx1.w1[1] = 0;
+    ctx1.w1[2] = 0;
+    ctx1.w1[3] = 0;
+    ctx1.w2[0] = 0;
+    ctx1.w2[1] = 0;
+    ctx1.w2[2] = 0;
+    ctx1.w2[3] = 0;
+    ctx1.w3[0] = 0;
+    ctx1.w3[1] = 0;
+    ctx1.w3[2] = 64 * 8;
+    ctx1.w3[3] = 0;
+
+    md5_transform_vector (ctx1.w0, ctx1.w1, ctx1.w2, ctx1.w3, ctx1.h);
+
+    const u32x r0 = ctx1.h[DGST_R0];
+    const u32x r1 = ctx1.h[DGST_R1];
+    const u32x r2 = ctx1.h[DGST_R2];
+    const u32x r3 = ctx1.h[DGST_R3];
+
+    COMPARE_S_SIMD (r0, r1, r2, r3);
+  }
+}
diff --git a/docs/changes.txt b/docs/changes.txt
index 8050700eb..83bedab49 100644
--- a/docs/changes.txt
+++ b/docs/changes.txt
@@ -11,6 +11,7 @@
 - Added hash-mode: ENCsecurity Datavault (MD5/keychain)
 - Added hash-mode: DANE RFC7929/RFC8162 SHA2-256
 - Added hash-mode: Dahua NVR/DVR/HVR (md5($salt1.strtoupper(md5($salt2.$pass))))
+- Added hash-mode: md5(md5($salt).md5(md5($pass)))
 
 ##
 ## Bugs
diff --git a/docs/readme.txt b/docs/readme.txt
index 47f04246c..a101d296e 100644
--- a/docs/readme.txt
+++ b/docs/readme.txt
@@ -91,6 +91,7 @@ NVIDIA GPUs require "NVIDIA Driver" (440.64 or later) and "CUDA Toolkit" (9.0 or
 - md5(sha1($salt).md5($pass))
 - md5(strtoupper(md5($pass)))
 - md5(utf16le($pass).$salt)
+- md5(md5($salt).md5(md5($pass)))
 - sha1($pass.$salt)
 - sha1($salt.$pass)
 - sha1($salt.$pass.$salt)
diff --git a/src/modules/module_30500.c b/src/modules/module_30500.c
new file mode 100644
index 000000000..f9d0a6342
--- /dev/null
+++ b/src/modules/module_30500.c
@@ -0,0 +1,236 @@
+/**
+ * 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_hash_md5.h"
+
+static const u32   ATTACK_EXEC    = ATTACK_EXEC_INSIDE_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_FORUM_SOFTWARE;
+static const char *HASH_NAME      = "md5(md5($salt).md5(md5($pass)))";
+static const u64   KERN_TYPE      = 30500;
+static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
+                                  | OPTI_TYPE_PRECOMPUTE_INIT
+                                  | OPTI_TYPE_EARLY_SKIP;
+static const u64   OPTS_TYPE      = OPTS_TYPE_STOCK_MODULE
+                                  | OPTS_TYPE_PT_GENERATE_LE
+                                  | OPTS_TYPE_PT_ADD80
+                                  | OPTS_TYPE_PT_ADDBITS14;
+static const u32   SALT_TYPE      = SALT_TYPE_GENERIC;
+static const char *ST_PASS        = "hashcat";
+static const char *ST_HASH        = "e13bb4b8e5a98db7277df344aa3363cf:28945624531";
+
+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 void md5_complete_no_limit (u32 digest[4], const u32 *plain, const u32 plain_len)
+{
+  // plain = u32 tmp_md5_buf[64] so this is compatible
+
+  md5_ctx_t md5_ctx;
+
+  md5_init (&md5_ctx);
+  md5_update (&md5_ctx, plain, plain_len);
+  md5_final (&md5_ctx);
+
+  digest[0] = md5_ctx.h[0];
+  digest[1] = md5_ctx.h[1];
+  digest[2] = md5_ctx.h[2];
+  digest[3] = md5_ctx.h[3];
+}
+
+static void precompute_salt_md5 (const u32 *salt_buf, const u32 salt_len, u8 *salt_pc)
+{
+  u32 digest[4] = { 0 };
+
+  md5_complete_no_limit (digest, salt_buf, salt_len);
+
+  u32_to_hex (digest[0], salt_pc +  0);
+  u32_to_hex (digest[1], salt_pc +  8);
+  u32_to_hex (digest[2], salt_pc + 16);
+  u32_to_hex (digest[3], salt_pc + 24);
+}
+
+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;
+
+  hc_token_t token;
+
+  token.token_cnt  = 2;
+
+  token.sep[0]     = hashconfig->separator;
+  token.len_min[0] = 32;
+  token.len_max[0] = 32;
+  token.attr[0]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  token.len_min[1] = SALT_MIN;
+  token.len_max[1] = SALT_MAX;
+  token.attr[1]    = TOKEN_ATTR_VERIFY_LENGTH;
+
+  if (hashconfig->opts_type & OPTS_TYPE_ST_HEX)
+  {
+    token.len_min[1] *= 2;
+    token.len_max[1] *= 2;
+
+    token.attr[1] |= 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);
+
+  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);
+
+  precompute_salt_md5 (salt->salt_buf, salt->salt_len, (u8 *) salt->salt_buf_pc);
+
+  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[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;
+
+  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] = hashconfig->separator;
+
+  out_len += 1;
+
+  out_len += generic_salt_encode (hashconfig, (const u8 *) salt->salt_buf, (const int) 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_charset        = 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_postprocess  = 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_DEFAULT;
+  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_DEFAULT;
+  module_ctx->module_unstable_warning         = MODULE_DEFAULT;
+  module_ctx->module_warmup_disable           = MODULE_DEFAULT;
+}
diff --git a/tools/test_modules/m30500.pm b/tools/test_modules/m30500.pm
new file mode 100644
index 000000000..5ecee4c6c
--- /dev/null
+++ b/tools/test_modules/m30500.pm
@@ -0,0 +1,44 @@
+#!/usr/bin/env perl
+
+##
+## Author......: See docs/credits.txt
+## License.....: MIT
+##
+
+use strict;
+use warnings;
+
+use Digest::MD5 qw (md5_hex);
+
+sub module_constraints { [[0, 256], [0, 256], [0, 55], [0, 55], [-1, -1]] }
+
+sub module_generate_hash
+{
+  my $word = shift;
+  my $salt = shift;
+
+  my $digest = md5_hex (md5_hex ($salt) . md5_hex (md5_hex ($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;

From 66b22fa64472b4d809743c35fb05fc3c993a5cd2 Mon Sep 17 00:00:00 2001
From: Jens Steube <jens.steube@gmail.com>
Date: Wed, 16 Nov 2022 14:26:54 +0100
Subject: [PATCH 2/6] Add support for Metal > 300 and reject support for older
 version

---
 OpenCL/inc_vendor.h |  2 +-
 src/backend.c       | 21 +++++++++++++++++++--
 2 files changed, 20 insertions(+), 3 deletions(-)

diff --git a/OpenCL/inc_vendor.h b/OpenCL/inc_vendor.h
index c39fce952..0916a30b3 100644
--- a/OpenCL/inc_vendor.h
+++ b/OpenCL/inc_vendor.h
@@ -12,7 +12,7 @@
 #define IS_CUDA
 #elif defined __HIPCC__
 #define IS_HIP
-#elif defined __METAL_MACOS__
+#elif defined __METAL__
 #define IS_METAL
 #else
 #define IS_OPENCL
diff --git a/src/backend.c b/src/backend.c
index 3c63882e1..8aa8c587c 100644
--- a/src/backend.c
+++ b/src/backend.c
@@ -4503,7 +4503,19 @@ int backend_ctx_init (hashcat_ctx_t *hashcat_ctx)
 
       if (hc_mtlRuntimeGetVersionString (hashcat_ctx, backend_ctx->metal_runtimeVersionStr, &version_len) == -1) return -1;
 
-      // TODO: needs version check
+      if (atoi (backend_ctx->metal_runtimeVersionStr) < 300)
+      {
+        event_log_warning (hashcat_ctx, "Unsupported Apple Metal runtime version '%s' detected! Falling back to OpenCL...", backend_ctx->metal_runtimeVersionStr);
+        event_log_warning (hashcat_ctx, NULL);
+
+        rc_metal_init = -1;
+
+        backend_ctx->rc_metal_init = rc_metal_init;
+
+        backend_ctx->mtl = NULL;
+
+        mtl_close (hashcat_ctx);
+      }
     }
     else
     {
@@ -4511,6 +4523,8 @@ int backend_ctx_init (hashcat_ctx_t *hashcat_ctx)
 
       backend_ctx->rc_metal_init = rc_metal_init;
 
+      backend_ctx->mtl = NULL;
+
       mtl_close (hashcat_ctx);
     }
   }
@@ -9265,8 +9279,11 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
     {
       // set some limits with Metal
 
-      device_param->kernel_threads_max = 128;
+      device_param->kernel_threads_max = 64;
+      device_param->kernel_threads_min = MIN (device_param->kernel_threads_min, device_param->kernel_threads_max);
+
       device_param->kernel_loops_max = 1024;  // autotune go over ...
+      device_param->kernel_loops_min = MIN (device_param->kernel_loops_min, device_param->kernel_loops_max);
     }
     #endif
 

From 1ed5d7d6d8ae213d84c15626f331ce94ab593a4e Mon Sep 17 00:00:00 2001
From: Mr Dromedary <dromedary@dromedary.camel.org>
Date: Wed, 16 Nov 2022 15:49:06 +0100
Subject: [PATCH 3/6] Support for GPG keys with SHA-512 as s2k digest algorithm

---
 OpenCL/m17020-pure.cl      | 550 +++++++++++++++++++++++++++++++++++++
 src/modules/module_17020.c | 414 ++++++++++++++++++++++++++++
 2 files changed, 964 insertions(+)
 create mode 100644 OpenCL/m17020-pure.cl
 create mode 100644 src/modules/module_17020.c

diff --git a/OpenCL/m17020-pure.cl b/OpenCL/m17020-pure.cl
new file mode 100644
index 000000000..b0440a986
--- /dev/null
+++ b/OpenCL/m17020-pure.cl
@@ -0,0 +1,550 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+//#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include M2S(INCLUDE_PATH/inc_vendor.h)
+#include M2S(INCLUDE_PATH/inc_types.h)
+#include M2S(INCLUDE_PATH/inc_platform.cl)
+#include M2S(INCLUDE_PATH/inc_common.cl)
+#include M2S(INCLUDE_PATH/inc_hash_sha1.cl)
+#include M2S(INCLUDE_PATH/inc_cipher_aes.cl)
+#include M2S(INCLUDE_PATH/inc_hash_sha512.cl)
+#endif
+
+typedef struct gpg
+{
+  u32 cipher_algo;
+  u32 iv[4];
+  u32 modulus_size;
+  u32 encrypted_data[384];
+  u32 encrypted_data_size;
+
+} gpg_t;
+
+typedef struct gpg_tmp
+{
+  // buffer for a maximum of 256 + 8 octets, we extend it to 384 octets so it's always 128 byte aligned (1024 bits)
+  u32 salted_pw_block[96];
+  // actual number of bytes in 'salted_pwd' that are used since salt and password are copied multiple times into the buffer
+  u32 salted_pw_block_len;
+
+  u64 h[8];
+
+  u32 w0[4];
+  u32 w1[4];
+  u32 w2[4];
+  u32 w3[4];
+  u32 w4[4];
+  u32 w5[4];
+  u32 w6[4];
+  u32 w7[4];
+
+  int len;
+
+} gpg_tmp_t;
+
+
+DECLSPEC u32 hc_bytealign_le_S (const u32 a, const u32 b, const int c)
+{
+  const int c_mod_4 = c & 3;
+
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 0) || defined IS_GENERIC
+  const u32 r = l32_from_64_S ((v64_from_v32ab_S (b, a) >> (c_mod_4 * 8)));
+  #endif
+
+  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1) || defined IS_NV
+
+  #if defined IS_NV
+  const int selector = (0x76543210 >> (c_mod_4 * 4)) & 0xffff;
+  #endif
+
+  #if (defined IS_AMD || defined IS_HIP)
+  const int selector = l32_from_64_S (0x0706050403020100UL >> (c_mod_4 * 8));
+  #endif
+
+  const u32 r = hc_byte_perm (b, a, selector);
+  #endif
+
+  return r;
+}
+
+DECLSPEC void memcat_le_S (PRIVATE_AS u32 *block, const u32 offset, PRIVATE_AS const u32 *append, u32 len)
+{
+  const u32 start_index = (offset - 1) >> 2;
+  const u32 count = ((offset + len + 3) >> 2) - start_index;
+  const int off_mod_4 = offset & 3;
+  const int off_minus_4 = 4 - off_mod_4;
+
+  block[start_index] |= hc_bytealign_le_S (append[0], 0, off_minus_4);
+
+  for (u32 idx = 1; idx < count; idx++)
+  {
+    block[start_index + idx] = hc_bytealign_le_S (append[idx], append[idx - 1], off_minus_4);
+  }
+}
+
+DECLSPEC void memzero_le_S (PRIVATE_AS u32 *block, const u32 start_offset, const u32 end_offset)
+{
+  const u32 start_idx = start_offset / 4;
+
+  // zero out bytes in the first u32 starting from 'start_offset'
+  // math is a bit complex to avoid shifting by 32 bits, which is not possible on some architectures
+  block[start_idx] &= ~(0xffffffff << ((start_offset & 3) * 8));
+
+  const u32 end_idx = (end_offset + 3) / 4;
+
+  // zero out bytes in u32 units -- note that the last u32 is completely zeroed!
+  for (u32 i = start_idx + 1; i < end_idx; i++)
+  {
+    block[i] = 0;
+  }
+}
+
+DECLSPEC void memzero_be_S (PRIVATE_AS u32 *block, const u32 start_offset, const u32 end_offset)
+{
+  const u32 start_idx = start_offset / 4;
+
+  // zero out bytes in the first u32 starting from 'start_offset'
+  // math is a bit complex to avoid shifting by 32 bits, which is not possible on some architectures
+  block[start_idx] &= ~(0xffffffff >> ((start_offset & 3) * 8));
+
+  const u32 end_idx = (end_offset + 3) / 4;
+
+  // zero out bytes in u32 units -- note that the last u32 is completely zeroed!
+  for (u32 i = start_idx + 1; i < end_idx; i++)
+  {
+    block[i] = 0;
+  }
+}
+
+DECLSPEC void aes128_decrypt_cfb (GLOBAL_AS const u32 *encrypted_data, int data_len, PRIVATE_AS const u32 *iv, PRIVATE_AS const u32 *key, PRIVATE_AS u32 *decrypted_data,
+                                  SHM_TYPE u32 *s_te0, SHM_TYPE u32 *s_te1, SHM_TYPE u32 *s_te2, SHM_TYPE u32 *s_te3, SHM_TYPE u32 *s_te4)
+{
+  u32 ks[44];
+  u32 essiv[4];
+
+  // Copy the IV, since this will be modified
+  essiv[0] = iv[0];
+  essiv[1] = iv[1];
+  essiv[2] = iv[2];
+  essiv[3] = iv[3];
+
+  aes128_set_encrypt_key (ks, key, s_te0, s_te1, s_te2, s_te3);
+
+  // Decrypt an AES-128 encrypted block
+  for (u32 i = 0; i < (data_len + 3) / 4; i += 4)
+  {
+    aes128_encrypt (ks, essiv, decrypted_data + i, s_te0, s_te1, s_te2, s_te3, s_te4);
+
+    decrypted_data[i + 0] ^= encrypted_data[i + 0];
+    decrypted_data[i + 1] ^= encrypted_data[i + 1];
+    decrypted_data[i + 2] ^= encrypted_data[i + 2];
+    decrypted_data[i + 3] ^= encrypted_data[i + 3];
+
+    // Note: Not necessary if you are only decrypting a single block!
+    essiv[0] = encrypted_data[i + 0];
+    essiv[1] = encrypted_data[i + 1];
+    essiv[2] = encrypted_data[i + 2];
+    essiv[3] = encrypted_data[i + 3];
+  }
+}
+
+DECLSPEC void aes256_decrypt_cfb (GLOBAL_AS const u32 *encrypted_data, int data_len, PRIVATE_AS const u32 *iv, PRIVATE_AS const u32 *key, PRIVATE_AS u32 *decrypted_data,
+                                  SHM_TYPE u32 *s_te0, SHM_TYPE u32 *s_te1, SHM_TYPE u32 *s_te2, SHM_TYPE u32 *s_te3, SHM_TYPE u32 *s_te4)
+{
+  u32 ks[60];
+  u32 essiv[4];
+
+  // Copy the IV, since this will be modified
+  essiv[0] = iv[0];
+  essiv[1] = iv[1];
+  essiv[2] = iv[2];
+  essiv[3] = iv[3];
+
+  aes256_set_encrypt_key (ks, key, s_te0, s_te1, s_te2, s_te3);
+
+  // Decrypt an AES-256 encrypted block
+  for (u32 i = 0; i < (data_len + 3) / 4; i += 4)
+  {
+    aes256_encrypt (ks, essiv, decrypted_data + i, s_te0, s_te1, s_te2, s_te3, s_te4);
+
+    decrypted_data[i + 0] ^= encrypted_data[i + 0];
+    decrypted_data[i + 1] ^= encrypted_data[i + 1];
+    decrypted_data[i + 2] ^= encrypted_data[i + 2];
+    decrypted_data[i + 3] ^= encrypted_data[i + 3];
+
+    // Note: Not necessary if you are only decrypting a single block!
+    essiv[0] = encrypted_data[i + 0];
+    essiv[1] = encrypted_data[i + 1];
+    essiv[2] = encrypted_data[i + 2];
+    essiv[3] = encrypted_data[i + 3];
+  }
+}
+
+DECLSPEC int check_decoded_data (PRIVATE_AS u32 *decoded_data, const u32 decoded_data_size)
+{
+  // Check the SHA-1 of the decrypted data which is stored at the end of the decrypted data
+  const u32 sha1_byte_off = (decoded_data_size - 20);
+  const u32 sha1_u32_off = sha1_byte_off / 4;
+
+  u32 expected_sha1[5];
+
+  expected_sha1[0] = hc_bytealign_le_S (decoded_data[sha1_u32_off + 1], decoded_data[sha1_u32_off + 0], sha1_byte_off);
+  expected_sha1[1] = hc_bytealign_le_S (decoded_data[sha1_u32_off + 2], decoded_data[sha1_u32_off + 1], sha1_byte_off);
+  expected_sha1[2] = hc_bytealign_le_S (decoded_data[sha1_u32_off + 3], decoded_data[sha1_u32_off + 2], sha1_byte_off);
+  expected_sha1[3] = hc_bytealign_le_S (decoded_data[sha1_u32_off + 4], decoded_data[sha1_u32_off + 3], sha1_byte_off);
+  expected_sha1[4] = hc_bytealign_le_S (decoded_data[sha1_u32_off + 5], decoded_data[sha1_u32_off + 4], sha1_byte_off);
+
+  memzero_le_S (decoded_data, sha1_byte_off, 384 * sizeof(u32));
+
+  sha1_ctx_t ctx;
+
+  sha1_init (&ctx);
+
+  sha1_update_swap (&ctx, decoded_data, sha1_byte_off);
+
+  sha1_final (&ctx);
+
+  return (expected_sha1[0] == hc_swap32_S (ctx.h[0]))
+      && (expected_sha1[1] == hc_swap32_S (ctx.h[1]))
+      && (expected_sha1[2] == hc_swap32_S (ctx.h[2]))
+      && (expected_sha1[3] == hc_swap32_S (ctx.h[3]))
+      && (expected_sha1[4] == hc_swap32_S (ctx.h[4]));
+}
+
+KERNEL_FQ void m17020_init (KERN_ATTR_TMPS_ESALT (gpg_tmp_t, gpg_t))
+{
+  const u64 gid = get_global_id (0);
+
+  if (gid >= GID_CNT) return;
+
+  const u32 pw_len = pws[gid].pw_len;
+  const u32 salted_pw_len = (salt_bufs[SALT_POS_HOST].salt_len + pw_len);
+
+  u32 salted_pw_block[96];
+
+  // concatenate salt and password -- the salt is always 8 bytes (2 * u32)
+  salted_pw_block[0] = salt_bufs[SALT_POS_HOST].salt_buf[0];
+  salted_pw_block[1] = salt_bufs[SALT_POS_HOST].salt_buf[1];
+
+  for (u32 idx = 0; idx < 64; idx++) salted_pw_block[idx + 2] = pws[gid].i[idx];
+
+  // zero remainder of buffer, the buffer will now be 96 words (3072 bits) containing:
+  //    0 -  1: salt
+  //    2 - 65: zero-padded password (max pwd len: 64 words = 256 bytes)
+  //   66 - 95: zeros
+  for (u32 idx = 66; idx < 96; idx++) salted_pw_block[idx] = 0;
+
+  // create a number of copies for efficiency
+  const u32 copies = 96 * sizeof(u32) / salted_pw_len;
+
+  for (u32 idx = 1; idx < copies; idx++)
+  {
+    memcat_le_S (salted_pw_block, idx * salted_pw_len, salted_pw_block, salted_pw_len);
+  }
+
+  for (u32 idx = 0; idx < 96; idx++)
+  {
+    tmps[gid].salted_pw_block[idx] = hc_swap32_S (salted_pw_block[idx]);
+  }
+
+  tmps[gid].salted_pw_block_len = (copies * salted_pw_len);
+
+  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;
+  tmps[gid].h[ 8] = SHA512M_A;
+  tmps[gid].h[ 9] = SHA512M_B;
+  tmps[gid].h[10] = SHA512M_C;
+  tmps[gid].h[11] = SHA512M_D;
+  tmps[gid].h[12] = SHA512M_E;
+  tmps[gid].h[13] = SHA512M_F;
+  tmps[gid].h[14] = SHA512M_G;
+  tmps[gid].h[15] = SHA512M_H;
+
+  tmps[gid].len = 0;
+}
+
+KERNEL_FQ void m17020_loop_prepare (KERN_ATTR_TMPS_ESALT (gpg_tmp_t, gpg_t))
+{
+  const u64 gid = get_global_id (0);
+
+  if (gid >= GID_CNT) return;
+
+  tmps[gid].w0[0] = 0;
+  tmps[gid].w0[1] = 0;
+  tmps[gid].w0[2] = 0;
+  tmps[gid].w0[3] = 0;
+  tmps[gid].w1[0] = 0;
+  tmps[gid].w1[1] = 0;
+  tmps[gid].w1[2] = 0;
+  tmps[gid].w1[3] = 0;
+  tmps[gid].w2[0] = 0;
+  tmps[gid].w2[1] = 0;
+  tmps[gid].w2[2] = 0;
+  tmps[gid].w2[3] = 0;
+  tmps[gid].w3[0] = 0;
+  tmps[gid].w3[1] = 0;
+  tmps[gid].w3[2] = 0;
+  tmps[gid].w3[3] = 0;
+  tmps[gid].w4[0] = 0;
+  tmps[gid].w4[1] = 0;
+  tmps[gid].w4[2] = 0;
+  tmps[gid].w4[3] = 0;
+  tmps[gid].w5[0] = 0;
+  tmps[gid].w5[1] = 0;
+  tmps[gid].w5[2] = 0;
+  tmps[gid].w5[3] = 0;
+  tmps[gid].w6[0] = 0;
+  tmps[gid].w6[1] = 0;
+  tmps[gid].w6[2] = 0;
+  tmps[gid].w6[3] = 0;
+  tmps[gid].w7[0] = 0;
+  tmps[gid].w7[1] = 0;
+  tmps[gid].w7[2] = 0;
+  tmps[gid].w7[3] = 0;
+
+  tmps[gid].len = 0;
+}
+
+KERNEL_FQ void m17020_loop (KERN_ATTR_TMPS_ESALT (gpg_tmp_t, gpg_t))
+{
+  const u64 gid = get_global_id (0);
+
+  if (gid >= GID_CNT) return;
+
+  // get the prepared buffer from the gpg_tmp_t struct into a local buffer
+  u32 salted_pw_block[96];
+
+  for (int i = 0; i < 96; i++) salted_pw_block[i] = tmps[gid].salted_pw_block[i];
+
+  const u32 salted_pw_block_len = tmps[gid].salted_pw_block_len;
+
+  // do we really need this, since the salt is always length 8?
+  if (salted_pw_block_len == 0) return;
+
+  /**
+   * context load
+   */
+
+  sha512_ctx_t ctx;
+
+  for (int i = 0; i < 8; i++) ctx.h[i] = tmps[gid].h[i];
+
+  for (int i = 0; i < 4; i++) ctx.w0[i] = tmps[gid].w0[i];
+  for (int i = 0; i < 4; i++) ctx.w1[i] = tmps[gid].w1[i];
+  for (int i = 0; i < 4; i++) ctx.w2[i] = tmps[gid].w2[i];
+  for (int i = 0; i < 4; i++) ctx.w3[i] = tmps[gid].w3[i];
+  for (int i = 0; i < 4; i++) ctx.w4[i] = tmps[gid].w4[i];
+  for (int i = 0; i < 4; i++) ctx.w5[i] = tmps[gid].w5[i];
+  for (int i = 0; i < 4; i++) ctx.w6[i] = tmps[gid].w6[i];
+  for (int i = 0; i < 4; i++) ctx.w7[i] = tmps[gid].w7[i];
+
+  ctx.len = tmps[gid].len;
+
+  // sha-512 of salt and password, up to 'salt_iter' bytes
+  const u32 salt_iter = salt_bufs[SALT_POS_HOST].salt_iter;
+
+  const u32 salted_pw_block_pos = LOOP_POS % salted_pw_block_len;
+  const u32 rounds = (LOOP_CNT + salted_pw_block_pos) / salted_pw_block_len;
+
+  for (u32 i = 0; i < rounds; i++)
+  {
+    sha512_update (&ctx, salted_pw_block, salted_pw_block_len);
+  }
+
+  if ((LOOP_POS + LOOP_CNT) == salt_iter)
+  {
+    const u32 remaining_bytes = salt_iter % salted_pw_block_len;
+
+    if (remaining_bytes)
+    {
+      memzero_be_S (salted_pw_block, remaining_bytes, salted_pw_block_len);
+
+      sha512_update (&ctx, salted_pw_block, remaining_bytes);
+    }
+
+    sha512_final (&ctx);
+  }
+
+  /**
+   * context save
+   */
+
+  for (int i = 0; i < 8; i++) tmps[gid].h[i] = ctx.h[i];
+
+  for (int i = 0; i < 4; i++) tmps[gid].w0[i] = ctx.w0[i];
+  for (int i = 0; i < 4; i++) tmps[gid].w1[i] = ctx.w1[i];
+  for (int i = 0; i < 4; i++) tmps[gid].w2[i] = ctx.w2[i];
+  for (int i = 0; i < 4; i++) tmps[gid].w3[i] = ctx.w3[i];
+  for (int i = 0; i < 4; i++) tmps[gid].w4[i] = ctx.w4[i];
+  for (int i = 0; i < 4; i++) tmps[gid].w5[i] = ctx.w5[i];
+  for (int i = 0; i < 4; i++) tmps[gid].w6[i] = ctx.w6[i];
+  for (int i = 0; i < 4; i++) tmps[gid].w7[i] = ctx.w7[i];
+
+  tmps[gid].len = ctx.len;
+}
+
+KERNEL_FQ void m17020_comp (KERN_ATTR_TMPS_ESALT (gpg_tmp_t, gpg_t))
+{
+  // not in use here, special case...
+}
+
+KERNEL_FQ void m17020_aux1 (KERN_ATTR_TMPS_ESALT (gpg_tmp_t, gpg_t))
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * aes shared
+   */
+
+  #ifdef REAL_SHM
+
+  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_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_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_CNT) return;
+
+  // retrieve and use the SHA-512 as the key for AES
+
+  u32 aes_key[4];
+
+  aes_key[0] = hc_swap32_S (h32_from_64 (tmps[gid].h[0]));
+  aes_key[1] = hc_swap32_S (l32_from_64 (tmps[gid].h[0]));
+  aes_key[2] = hc_swap32_S (h32_from_64 (tmps[gid].h[1]));
+  aes_key[3] = hc_swap32_S (l32_from_64 (tmps[gid].h[1]));
+
+  u32 iv[4] = {0};
+
+  for (int idx = 0; idx < 4; idx++) iv[idx] = esalt_bufs[DIGESTS_OFFSET_HOST].iv[idx];
+
+  u32 decoded_data[384];
+
+  const u32 enc_data_size = esalt_bufs[DIGESTS_OFFSET_HOST].encrypted_data_size;
+
+  aes128_decrypt_cfb (esalt_bufs[DIGESTS_OFFSET_HOST].encrypted_data, enc_data_size, iv, aes_key, decoded_data, s_te0, s_te1, s_te2, s_te3, s_te4);
+
+  if (check_decoded_data (decoded_data, enc_data_size))
+  {
+    if (hc_atomic_inc (&hashes_shown[DIGESTS_OFFSET_HOST]) == 0)
+    {
+      mark_hash (plains_buf, d_return_buf, SALT_POS_HOST, DIGESTS_CNT, 0, DIGESTS_OFFSET_HOST + 0, gid, 0, 0, 0);
+    }
+  }
+}
+
+KERNEL_FQ void m17020_aux2 (KERN_ATTR_TMPS_ESALT (gpg_tmp_t, gpg_t))
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * aes shared
+   */
+
+  #ifdef REAL_SHM
+
+  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_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_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_CNT) return;
+
+  // retrieve and use the SHA-512 as the key for AES
+
+  u32 aes_key[8];
+
+  aes_key[0] = hc_swap32_S (h32_from_64 (tmps[gid].h[0]));
+  aes_key[1] = hc_swap32_S (l32_from_64 (tmps[gid].h[0]));
+  aes_key[2] = hc_swap32_S (h32_from_64 (tmps[gid].h[1]));
+  aes_key[3] = hc_swap32_S (l32_from_64 (tmps[gid].h[1]));
+  aes_key[4] = hc_swap32_S (h32_from_64 (tmps[gid].h[2]));
+  aes_key[5] = hc_swap32_S (l32_from_64 (tmps[gid].h[2]));
+  aes_key[6] = hc_swap32_S (h32_from_64 (tmps[gid].h[3]));
+  aes_key[7] = hc_swap32_S (l32_from_64 (tmps[gid].h[3]));
+
+  u32 iv[4] = {0};
+
+  for (int idx = 0; idx < 4; idx++) iv[idx] = esalt_bufs[DIGESTS_OFFSET_HOST].iv[idx];
+
+  u32 decoded_data[384];
+
+  const u32 enc_data_size = esalt_bufs[DIGESTS_OFFSET_HOST].encrypted_data_size;
+
+  aes256_decrypt_cfb (esalt_bufs[DIGESTS_OFFSET_HOST].encrypted_data, enc_data_size, iv, aes_key, decoded_data, s_te0, s_te1, s_te2, s_te3, s_te4);
+
+  if (check_decoded_data (decoded_data, enc_data_size))
+  {
+    if (hc_atomic_inc (&hashes_shown[DIGESTS_OFFSET_HOST]) == 0)
+    {
+      mark_hash (plains_buf, d_return_buf, SALT_POS_HOST, DIGESTS_CNT, 0, DIGESTS_OFFSET_HOST + 0, gid, 0, 0, 0);
+    }
+  }
+}
diff --git a/src/modules/module_17020.c b/src/modules/module_17020.c
new file mode 100644
index 000000000..340cf5183
--- /dev/null
+++ b/src/modules/module_17020.c
@@ -0,0 +1,414 @@
+/**
+ * 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_RAW_HASH;
+static const char *HASH_NAME      = "GPG (AES-128/AES-256 (SHA-512($pass)))";
+static const u64   KERN_TYPE      = 17020;
+static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE;
+static const u64   OPTS_TYPE      = OPTS_TYPE_STOCK_MODULE
+                                  | OPTS_TYPE_PT_GENERATE_LE
+                                  | OPTS_TYPE_LOOP_PREPARE
+                                  | OPTS_TYPE_AUX1
+                                  | OPTS_TYPE_AUX2
+                                  | OPTS_TYPE_DEEP_COMP_KERNEL;
+static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
+static const char *ST_PASS        = "hashcat";
+static const char *ST_HASH        = "$gpg$*1*668*2048*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*3*254*10*9*16*d1547688c9cc944482d16dff17df0858*20971520*1fef4e57e302d34e";
+
+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 gpg
+{
+  u32 cipher_algo;
+  u32 iv[4];
+  u32 modulus_size;
+  u32 encrypted_data[384];
+  u32 encrypted_data_size;
+
+} gpg_t;
+
+typedef struct gpg_tmp
+{
+  u32 salted_pw_block[96];
+
+  u32 salted_pw_block_len;
+
+  u64 h[8];
+
+  u32 w0[4];
+  u32 w1[4];
+  u32 w2[4];
+  u32 w3[4];
+  u32 w4[4];
+  u32 w5[4];
+  u32 w6[4];
+  u32 w7[4];
+
+  int len;
+
+} gpg_tmp_t;
+
+static const char *SIGNATURE_GPG = "$gpg$";
+
+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 (gpg_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 (gpg_tmp_t);
+
+  return tmp_size;
+}
+
+bool module_hlfmt_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)
+{
+  const bool hlfmt_disable = true;
+
+  return hlfmt_disable;
+}
+
+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)
+{
+  const u32 kernel_loops_min = 1024;
+
+  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 = 65536;
+
+  return kernel_loops_max;
+}
+
+u32 module_deep_comp_kernel (MAYBE_UNUSED const hashes_t *hashes, MAYBE_UNUSED const u32 salt_pos, MAYBE_UNUSED const u32 digest_pos)
+{
+  const u32 digests_offset = hashes->salts_buf[salt_pos].digests_offset;
+
+  gpg_t *gpgs = (gpg_t *) hashes->esalts_buf;
+
+  gpg_t *gpg = &gpgs[digests_offset + digest_pos];
+
+  if (gpg->cipher_algo == 7)
+  {
+    return KERN_RUN_AUX1;
+  }
+  else if (gpg->cipher_algo == 9)
+  {
+    return KERN_RUN_AUX2;
+  }
+
+  return 0;
+}
+
+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;
+
+  gpg_t *gpg = (gpg_t *) esalt_buf;
+
+  hc_token_t token;
+
+  token.token_cnt = 13;
+
+  // signature $gpg$
+  token.signatures_cnt = 1;
+  token.signatures_buf[0] = SIGNATURE_GPG;
+
+  // signature $gpg$
+  token.len_min[0]  = 5;
+  token.len_max[0]  = 5;
+  token.sep[0]      = '*';
+  token.attr[0]     = TOKEN_ATTR_VERIFY_LENGTH
+                    | TOKEN_ATTR_VERIFY_SIGNATURE;
+
+  // "1" -- unknown option
+  token.len_min[1]  = 1;
+  token.len_max[1]  = 1;
+  token.sep[1]      = '*';
+  token.attr[1]     = TOKEN_ATTR_VERIFY_LENGTH
+                    | TOKEN_ATTR_VERIFY_DIGIT;
+
+  // size of the encrypted data in bytes
+  token.len_min[2]  = 3;
+  token.len_max[2]  = 4;
+  token.sep[2]      = '*';
+  token.attr[2]     = TOKEN_ATTR_VERIFY_LENGTH
+                    | TOKEN_ATTR_VERIFY_DIGIT;
+
+  // size of the key: 1024, 2048, 4096, etc.
+  token.len_min[3]  = 3;
+  token.len_max[3]  = 4;
+  token.sep[3]      = '*';
+  token.attr[3]     = TOKEN_ATTR_VERIFY_LENGTH
+                    | TOKEN_ATTR_VERIFY_DIGIT;
+
+  // encrypted key -- twice the amount of byte because its interpreted as characters
+  token.len_min[4]  = 256;
+  token.len_max[4]  = 3072;
+  token.sep[4]      = '*';
+  token.attr[4]     = TOKEN_ATTR_VERIFY_LENGTH
+                    | TOKEN_ATTR_VERIFY_HEX;
+
+  // "3" - String2Key parameter
+  token.len_min[5]  = 1;
+  token.len_max[5]  = 1;
+  token.sep[5]      = '*';
+  token.attr[5]     = TOKEN_ATTR_VERIFY_LENGTH
+                    | TOKEN_ATTR_VERIFY_DIGIT;
+
+  // "254" - String2Key parameters
+  token.len_min[6]  = 3;
+  token.len_max[6]  = 3;
+  token.sep[6]      = '*';
+  token.attr[6]     = TOKEN_ATTR_VERIFY_LENGTH
+                    | TOKEN_ATTR_VERIFY_DIGIT;
+
+  // "10" - String2Key parameters
+  token.len_min[7]  = 2;
+  token.len_max[7]  = 2;
+  token.sep[7]      = '*';
+  token.attr[7]     = TOKEN_ATTR_VERIFY_LENGTH
+                    | TOKEN_ATTR_VERIFY_DIGIT;
+
+  // cipher mode: 7 or 9
+  token.len_min[8]  = 1;
+  token.len_max[8]  = 1;
+  token.sep[8]      = '*';
+  token.attr[8]     = TOKEN_ATTR_VERIFY_LENGTH
+                    | TOKEN_ATTR_VERIFY_DIGIT;
+
+  // size of initial vector in bytes: 16
+  token.len_min[9]  = 2;
+  token.len_max[9]  = 2;
+  token.sep[9]      = '*';
+  token.attr[9]     = TOKEN_ATTR_VERIFY_LENGTH
+                    | TOKEN_ATTR_VERIFY_DIGIT;
+
+  // initial vector - twice the amount of bytes because its interpreted as characters
+  token.len_min[10] = 32;
+  token.len_max[10] = 32;
+  token.sep[10]     = '*';
+  token.attr[10]    = TOKEN_ATTR_VERIFY_LENGTH
+                    | TOKEN_ATTR_VERIFY_HEX;
+
+  // iteration count
+  token.len_min[11] = 1;
+  token.len_max[11] = 8;
+  token.sep[11]     = '*';
+  token.attr[11]    = TOKEN_ATTR_VERIFY_LENGTH
+                    | TOKEN_ATTR_VERIFY_DIGIT;
+
+  // salt - 8 bytes / 16 characters
+  token.len_min[12] = 16;
+  token.len_max[12] = 16;
+  token.attr[12]    = 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);
+
+  // Modulus size
+
+  const int modulus_size = hc_strtoul ((const char *) token.buf[3], NULL, 10);
+
+  if ((modulus_size < 256) || (modulus_size > 16384)) return (PARSER_SALT_LENGTH);
+
+  gpg->modulus_size = modulus_size;
+
+  // Encrypted data
+
+  const int enc_data_size = hc_strtoul ((const char *) token.buf[2], NULL, 10);
+
+  const int encrypted_data_size = hex_decode ((const u8 *) token.buf[4], token.len[4], (u8 *) gpg->encrypted_data);
+
+  if (enc_data_size != encrypted_data_size) return (PARSER_CT_LENGTH);
+
+  gpg->encrypted_data_size = encrypted_data_size;
+
+  // Check String2Key parameters
+
+  if (hc_strtoul ((const char *) token.buf[5], NULL, 10) !=   3) return (PARSER_HASH_VALUE);
+  if (hc_strtoul ((const char *) token.buf[6], NULL, 10) != 254) return (PARSER_HASH_VALUE);
+  if (hc_strtoul ((const char *) token.buf[7], NULL, 10) !=  10) return (PARSER_HASH_VALUE);
+
+  // Cipher algo
+
+  const int cipher_algo = hc_strtoul ((const char *) token.buf[8], NULL, 10);
+
+  if ((cipher_algo != 7) && (cipher_algo != 9)) return (PARSER_CIPHER);
+
+  gpg->cipher_algo = cipher_algo;
+
+  // IV (size)
+
+  if (hc_strtoul ((const char *) token.buf[9], NULL, 10) != sizeof (gpg->iv)) return (PARSER_IV_LENGTH);
+
+  const int iv_size = hex_decode ((const u8 *) token.buf[10], token.len[10], (u8 *) gpg->iv);
+
+  if (iv_size != sizeof (gpg->iv)) return (PARSER_IV_LENGTH);
+
+  // Salt Iter
+
+  const u32 salt_iter = hc_strtoul ((const char *) token.buf[11], NULL, 10);
+
+  if (salt_iter < 8 || salt_iter > 65011712) return (PARSER_SALT_ITERATION);
+
+  salt->salt_iter = salt_iter;
+
+  // Salt Value
+
+  salt->salt_len = hex_decode ((const u8 *) token.buf[12], token.len[12], (u8 *) salt->salt_buf);
+
+  if (salt->salt_len != 8) return (PARSER_SALT_LENGTH);
+
+  // hash fake
+  digest[0] = gpg->iv[0];
+  digest[1] = gpg->iv[1];
+  digest[2] = gpg->iv[2];
+  digest[3] = gpg->iv[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 gpg_t *gpg = (const gpg_t *) esalt_buf;
+
+  u8 encrypted_data[(384 * 8) + 1];
+
+  hex_encode ((const u8 *) gpg->encrypted_data, gpg->encrypted_data_size, (u8 *) encrypted_data);
+
+  const int line_len = snprintf (line_buf, line_size, "%s*%d*%d*%d*%s*%d*%d*%d*%d*%d*%08x%08x%08x%08x*%d*%08x%08x",
+    SIGNATURE_GPG,
+    1, /* unknown field */
+    gpg->encrypted_data_size,
+    gpg->modulus_size,
+    encrypted_data,
+    3, /* version (major?) */
+    254, /* version (minor?) */
+    10, /* key hash (sha-512) */
+    gpg->cipher_algo,
+    16, /*iv_size*/
+    byte_swap_32 (gpg->iv[0]),
+    byte_swap_32 (gpg->iv[1]),
+    byte_swap_32 (gpg->iv[2]),
+    byte_swap_32 (gpg->iv[3]),
+    salt->salt_iter,
+    byte_swap_32 (salt->salt_buf[0]),
+    byte_swap_32 (salt->salt_buf[1]));
+
+  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_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;
+  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_postprocess  = 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_hlfmt_disable;
+  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_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;
+}

From c3e852b28797b06c0576af1b5fc52830342fb757 Mon Sep 17 00:00:00 2001
From: Mr Dromedary <dromedary@dromedary.camel.org>
Date: Wed, 16 Nov 2022 16:06:44 +0100
Subject: [PATCH 4/6] Interface changed, add module_benchmark_charset

---
 src/modules/module_17020.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/src/modules/module_17020.c b/src/modules/module_17020.c
index 340cf5183..05bcccbcf 100644
--- a/src/modules/module_17020.c
+++ b/src/modules/module_17020.c
@@ -343,6 +343,7 @@ void module_init (module_ctx_t *module_ctx)
   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_charset        = MODULE_DEFAULT;
   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;

From c35d8ecfd441d21cdea620d15f55f6e2c0b7f5d4 Mon Sep 17 00:00:00 2001
From: jsteube <jens.steube@gmail.com>
Date: Thu, 17 Nov 2022 11:31:35 +0000
Subject: [PATCH 5/6] Update hash category for hash-mode 170xx

---
 src/modules/module_17010.c | 2 +-
 src/modules/module_17020.c | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/src/modules/module_17010.c b/src/modules/module_17010.c
index 1af405223..ede464576 100644
--- a/src/modules/module_17010.c
+++ b/src/modules/module_17010.c
@@ -16,7 +16,7 @@ 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_RAW_HASH;
+static const u32   HASH_CATEGORY  = HASH_CATEGORY_PRIVATE_KEY;
 static const char *HASH_NAME      = "GPG (AES-128/AES-256 (SHA-1($pass)))";
 static const u64   KERN_TYPE      = 17010;
 static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE;
diff --git a/src/modules/module_17020.c b/src/modules/module_17020.c
index 05bcccbcf..4ce8e21ea 100644
--- a/src/modules/module_17020.c
+++ b/src/modules/module_17020.c
@@ -16,7 +16,7 @@ static const u32   DGST_POS1      = 1;
 static const u32   DGST_POS2      = 2;
 static const u32   DGST_POS3      = 3;
 static const u32   DGST_SIZE      = DGST_SIZE_4_4;
-static const u32   HASH_CATEGORY  = HASH_CATEGORY_RAW_HASH;
+static const u32   HASH_CATEGORY  = HASH_CATEGORY_PRIVATE_KEY;
 static const char *HASH_NAME      = "GPG (AES-128/AES-256 (SHA-512($pass)))";
 static const u64   KERN_TYPE      = 17020;
 static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE;

From 9a52ee6d01e11465a36333c20cf6b835551975a9 Mon Sep 17 00:00:00 2001
From: jsteube <jens.steube@gmail.com>
Date: Fri, 18 Nov 2022 09:47:17 +0000
Subject: [PATCH 6/6] Fix error message in backend.c and add missing new
 hash-mode entries in changes.txt

---
 docs/changes.txt | 4 +++-
 src/backend.c    | 2 +-
 2 files changed, 4 insertions(+), 2 deletions(-)

diff --git a/docs/changes.txt b/docs/changes.txt
index 83bedab49..a6138d86a 100644
--- a/docs/changes.txt
+++ b/docs/changes.txt
@@ -11,6 +11,8 @@
 - Added hash-mode: ENCsecurity Datavault (MD5/keychain)
 - Added hash-mode: DANE RFC7929/RFC8162 SHA2-256
 - Added hash-mode: Dahua NVR/DVR/HVR (md5($salt1.strtoupper(md5($salt2.$pass))))
+- Added hash-mode: GPG (AES-128/AES-256 (SHA-1($pass)))
+- Added hash-mode: GPG (AES-128/AES-256 (SHA-512($pass)))
 - Added hash-mode: md5(md5($salt).md5(md5($pass)))
 
 ##
@@ -1896,4 +1898,4 @@ It combines all features of all hashcat projects in one project.
 - Created environment variable to inform pocl OpenCL runtime to not create its own kernel cache
 - Dropped special 64-bit rotate() handling for NV, it seems that they've added it to their OpenCL runtime
 - Completely get rid of HAVE_ADL, HAVE_NVML and HAVE_NVAPI in sources
-- Replaced NVAPI with NVML on windows
\ No newline at end of file
+- Replaced NVAPI with NVML on windows
diff --git a/src/backend.c b/src/backend.c
index 8aa8c587c..d803b9841 100644
--- a/src/backend.c
+++ b/src/backend.c
@@ -14703,7 +14703,7 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
     if (kernel_accel_min > kernel_accel_max)
     {
       event_log_error (hashcat_ctx, "* Device #%u: Too many compute units to keep minimum kernel accel limit.", device_id + 1);
-      event_log_error (hashcat_ctx, "             Retry with lower --backend-kernel-threads value.");
+      event_log_error (hashcat_ctx, "             Retry with lower --kernel-threads value.");
 
       backend_kernel_accel_warnings++;