initial rakp-hmac-md5 optmized kernels - pure kernels fall back on copes of m00050 (hmac md5)

OpenCL/m07350_a0-optimized.cl

@@ -0,0 +1,426 @@
+/**
+ * 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
+
+DECLSPEC void hmac_md5_pad (PRIVATE_AS u32x *w0, PRIVATE_AS u32x *w1, PRIVATE_AS u32x *w2, PRIVATE_AS u32x *w3, PRIVATE_AS u32x *ipad, PRIVATE_AS u32x *opad)
+{
+  w0[0] = w0[0] ^ 0x36363636;
+  w0[1] = w0[1] ^ 0x36363636;
+  w0[2] = w0[2] ^ 0x36363636;
+  w0[3] = w0[3] ^ 0x36363636;
+  w1[0] = w1[0] ^ 0x36363636;
+  w1[1] = w1[1] ^ 0x36363636;
+  w1[2] = w1[2] ^ 0x36363636;
+  w1[3] = w1[3] ^ 0x36363636;
+  w2[0] = w2[0] ^ 0x36363636;
+  w2[1] = w2[1] ^ 0x36363636;
+  w2[2] = w2[2] ^ 0x36363636;
+  w2[3] = w2[3] ^ 0x36363636;
+  w3[0] = w3[0] ^ 0x36363636;
+  w3[1] = w3[1] ^ 0x36363636;
+  w3[2] = w3[2] ^ 0x36363636;
+  w3[3] = w3[3] ^ 0x36363636;
+
+  ipad[0] = MD5M_A;
+  ipad[1] = MD5M_B;
+  ipad[2] = MD5M_C;
+  ipad[3] = MD5M_D;
+
+  md5_transform_vector (w0, w1, w2, w3, ipad);
+
+  w0[0] = w0[0] ^ 0x6a6a6a6a;
+  w0[1] = w0[1] ^ 0x6a6a6a6a;
+  w0[2] = w0[2] ^ 0x6a6a6a6a;
+  w0[3] = w0[3] ^ 0x6a6a6a6a;
+  w1[0] = w1[0] ^ 0x6a6a6a6a;
+  w1[1] = w1[1] ^ 0x6a6a6a6a;
+  w1[2] = w1[2] ^ 0x6a6a6a6a;
+  w1[3] = w1[3] ^ 0x6a6a6a6a;
+  w2[0] = w2[0] ^ 0x6a6a6a6a;
+  w2[1] = w2[1] ^ 0x6a6a6a6a;
+  w2[2] = w2[2] ^ 0x6a6a6a6a;
+  w2[3] = w2[3] ^ 0x6a6a6a6a;
+  w3[0] = w3[0] ^ 0x6a6a6a6a;
+  w3[1] = w3[1] ^ 0x6a6a6a6a;
+  w3[2] = w3[2] ^ 0x6a6a6a6a;
+  w3[3] = w3[3] ^ 0x6a6a6a6a;
+
+  opad[0] = MD5M_A;
+  opad[1] = MD5M_B;
+  opad[2] = MD5M_C;
+  opad[3] = MD5M_D;
+
+  md5_transform_vector (w0, w1, w2, w3, opad);
+}
+
+DECLSPEC void hmac_md5_run (PRIVATE_AS u32x *w0, PRIVATE_AS u32x *w1, PRIVATE_AS u32x *w2, PRIVATE_AS u32x *w3, PRIVATE_AS u32x *ipad, PRIVATE_AS u32x *opad, PRIVATE_AS u32x *digest)
+{
+  digest[0] = ipad[0];
+  digest[1] = ipad[1];
+  digest[2] = ipad[2];
+  digest[3] = ipad[3];
+
+  md5_transform_vector (w0, w1, w2, w3, digest);
+
+  w0[0] = digest[0];
+  w0[1] = digest[1];
+  w0[2] = digest[2];
+  w0[3] = digest[3];
+  w1[0] = 0x80;
+  w1[1] = 0;
+  w1[2] = 0;
+  w1[3] = 0;
+  w2[0] = 0;
+  w2[1] = 0;
+  w2[2] = 0;
+  w2[3] = 0;
+  w3[0] = 0;
+  w3[1] = 0;
+  w3[2] = (64 + 16) * 8;
+  w3[3] = 0;
+
+  digest[0] = opad[0];
+  digest[1] = opad[1];
+  digest[2] = opad[2];
+  digest[3] = opad[3];
+
+  md5_transform_vector (w0, w1, w2, w3, digest);
+}
+
+KERNEL_FQ void m07350_m04 (KERN_ATTR_RULES ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+
+  /**
+   * base
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= GID_CNT) return;
+
+  u32 pw_buf0[4];
+  u32 pw_buf1[4];
+
+  pw_buf0[0] = pws[gid].i[0];
+  pw_buf0[1] = pws[gid].i[1];
+  pw_buf0[2] = pws[gid].i[2];
+  pw_buf0[3] = pws[gid].i[3];
+  pw_buf1[0] = pws[gid].i[4];
+  pw_buf1[1] = pws[gid].i[5];
+  pw_buf1[2] = pws[gid].i[6];
+  pw_buf1[3] = pws[gid].i[7];
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * salt
+   */
+
+  u32 salt_buf0[4];
+  u32 salt_buf1[4];
+  u32 salt_buf2[4];
+  u32 salt_buf3[4];
+  u32 salt_buf4[4];
+
+  salt_buf0[0] = salt_bufs[SALT_POS_HOST].salt_buf[ 0];
+  salt_buf0[1] = salt_bufs[SALT_POS_HOST].salt_buf[ 1];
+  salt_buf0[2] = salt_bufs[SALT_POS_HOST].salt_buf[ 2];
+  salt_buf0[3] = salt_bufs[SALT_POS_HOST].salt_buf[ 3];
+  salt_buf1[0] = salt_bufs[SALT_POS_HOST].salt_buf[ 4];
+  salt_buf1[1] = salt_bufs[SALT_POS_HOST].salt_buf[ 5];
+  salt_buf1[2] = salt_bufs[SALT_POS_HOST].salt_buf[ 6];
+  salt_buf1[3] = salt_bufs[SALT_POS_HOST].salt_buf[ 7];
+  salt_buf2[0] = salt_bufs[SALT_POS_HOST].salt_buf[ 8];
+  salt_buf2[1] = salt_bufs[SALT_POS_HOST].salt_buf[ 9];
+  salt_buf2[2] = salt_bufs[SALT_POS_HOST].salt_buf[10];
+  salt_buf2[3] = salt_bufs[SALT_POS_HOST].salt_buf[11];
+  salt_buf3[0] = salt_bufs[SALT_POS_HOST].salt_buf[12];
+  salt_buf3[1] = salt_bufs[SALT_POS_HOST].salt_buf[13];
+  salt_buf3[2] = salt_bufs[SALT_POS_HOST].salt_buf[14];
+  salt_buf3[3] = salt_bufs[SALT_POS_HOST].salt_buf[15];
+
+  salt_buf4[0] = salt_bufs[SALT_POS_HOST].salt_buf[16];
+  salt_buf4[1] = salt_bufs[SALT_POS_HOST].salt_buf[17];
+  salt_buf4[2] = salt_bufs[SALT_POS_HOST].salt_buf[18];
+  salt_buf4[3] = salt_bufs[SALT_POS_HOST].salt_buf[19];
+
+  const u32 salt_len = salt_bufs[SALT_POS_HOST].salt_len;
+
+  /**
+   * 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 };
+
+    apply_rules_vect_optimized (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
+
+    /**
+     * pads
+     */
+
+    u32x ipad[4];
+    u32x opad[4];
+
+    hmac_md5_pad (w0, w1, w2, w3, ipad, opad);
+
+    w0[0] = salt_buf0[0];
+    w0[1] = salt_buf0[1];
+    w0[2] = salt_buf0[2];
+    w0[3] = salt_buf0[3];
+    w1[0] = salt_buf1[0];
+    w1[1] = salt_buf1[1];
+    w1[2] = salt_buf1[2];
+    w1[3] = salt_buf1[3];
+    w2[0] = salt_buf2[0];
+    w2[1] = salt_buf2[1];
+    w2[2] = salt_buf2[2];
+    w2[3] = salt_buf2[3];
+    w3[0] = salt_buf3[0];
+    w3[1] = salt_buf3[1];
+    w3[2] = salt_buf3[2];
+    w3[3] = salt_buf3[3];
+
+    md5_transform_vector(w0, w1, w2, w3, ipad);
+
+    w0[0] = salt_buf4[0];
+    w0[1] = salt_buf4[1];
+    w0[2] = salt_buf4[2];
+    w0[3] = salt_buf4[3];
+    w1[0] = 0;
+    w1[1] = 0;
+    w1[2] = 0;
+    w1[3] = 0;
+    w2[0] = 0;
+    w2[1] = 0;
+    w2[2] = 0;
+    w2[3] = 0;
+    w3[0] = 0;
+    w3[1] = 0;
+    w3[2] = (64 + salt_len) * 8;
+    w3[3] = 0;
+
+    md5_transform_vector(w0, w1, w2, w3, ipad);
+
+    w0[0] = ipad[0];
+    w0[1] = ipad[1];
+    w0[2] = ipad[2];
+    w0[3] = ipad[3];
+    w1[0] = 0x80;
+    w1[1] = 0;
+    w1[2] = 0;
+    w1[3] = 0;
+    w2[0] = 0;
+    w2[1] = 0;
+    w2[2] = 0;
+    w2[3] = 0;
+    w3[0] = 0;
+    w3[1] = 0;
+    w3[2] = (64 + 16) * 8;
+    w3[3] = 0;
+
+    md5_transform_vector (w0, w1, w2, w3, opad);
+
+    COMPARE_M_SIMD (opad[0], opad[3], opad[2], opad[1]);
+  }
+}
+
+KERNEL_FQ void m07350_m08 (KERN_ATTR_RULES ())
+{
+}
+
+KERNEL_FQ void m07350_m16 (KERN_ATTR_RULES ())
+{
+}
+
+KERNEL_FQ void m07350_s04 (KERN_ATTR_RULES ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+
+  /**
+   * base
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= GID_CNT) return;
+
+  u32 pw_buf0[4];
+  u32 pw_buf1[4];
+
+  pw_buf0[0] = pws[gid].i[0];
+  pw_buf0[1] = pws[gid].i[1];
+  pw_buf0[2] = pws[gid].i[2];
+  pw_buf0[3] = pws[gid].i[3];
+  pw_buf1[0] = pws[gid].i[4];
+  pw_buf1[1] = pws[gid].i[5];
+  pw_buf1[2] = pws[gid].i[6];
+  pw_buf1[3] = pws[gid].i[7];
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * salt
+   */
+
+  u32 salt_buf0[4];
+  u32 salt_buf1[4];
+  u32 salt_buf2[4];
+  u32 salt_buf3[4];
+  u32 salt_buf4[4];
+
+  salt_buf0[0] = salt_bufs[SALT_POS_HOST].salt_buf[ 0];
+  salt_buf0[1] = salt_bufs[SALT_POS_HOST].salt_buf[ 1];
+  salt_buf0[2] = salt_bufs[SALT_POS_HOST].salt_buf[ 2];
+  salt_buf0[3] = salt_bufs[SALT_POS_HOST].salt_buf[ 3];
+  salt_buf1[0] = salt_bufs[SALT_POS_HOST].salt_buf[ 4];
+  salt_buf1[1] = salt_bufs[SALT_POS_HOST].salt_buf[ 5];
+  salt_buf1[2] = salt_bufs[SALT_POS_HOST].salt_buf[ 6];
+  salt_buf1[3] = salt_bufs[SALT_POS_HOST].salt_buf[ 7];
+  salt_buf2[0] = salt_bufs[SALT_POS_HOST].salt_buf[ 8];
+  salt_buf2[1] = salt_bufs[SALT_POS_HOST].salt_buf[ 9];
+  salt_buf2[2] = salt_bufs[SALT_POS_HOST].salt_buf[10];
+  salt_buf2[3] = salt_bufs[SALT_POS_HOST].salt_buf[11];
+  salt_buf3[0] = salt_bufs[SALT_POS_HOST].salt_buf[12];
+  salt_buf3[1] = salt_bufs[SALT_POS_HOST].salt_buf[13];
+  salt_buf3[2] = salt_bufs[SALT_POS_HOST].salt_buf[14];
+  salt_buf3[3] = salt_bufs[SALT_POS_HOST].salt_buf[15];
+
+  salt_buf4[0] = salt_bufs[SALT_POS_HOST].salt_buf[16];
+  salt_buf4[1] = salt_bufs[SALT_POS_HOST].salt_buf[17];
+  salt_buf4[2] = salt_bufs[SALT_POS_HOST].salt_buf[18];
+  salt_buf4[3] = salt_bufs[SALT_POS_HOST].salt_buf[19];
+
+  const u32 salt_len = salt_bufs[SALT_POS_HOST].salt_len;
+
+  /**
+   * 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 };
+
+    apply_rules_vect_optimized (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
+
+    /**
+     * pads
+     */
+
+    u32x ipad[4];
+    u32x opad[4];
+
+    hmac_md5_pad (w0, w1, w2, w3, ipad, opad);
+
+    w0[0] = salt_buf0[0];
+    w0[1] = salt_buf0[1];
+    w0[2] = salt_buf0[2];
+    w0[3] = salt_buf0[3];
+    w1[0] = salt_buf1[0];
+    w1[1] = salt_buf1[1];
+    w1[2] = salt_buf1[2];
+    w1[3] = salt_buf1[3];
+    w2[0] = salt_buf2[0];
+    w2[1] = salt_buf2[1];
+    w2[2] = salt_buf2[2];
+    w2[3] = salt_buf2[3];
+    w3[0] = salt_buf3[0];
+    w3[1] = salt_buf3[1];
+    w3[2] = salt_buf3[2];
+    w3[3] = salt_buf3[3];
+
+    md5_transform_vector(w0, w1, w2, w3, ipad);
+
+    w0[0] = salt_buf4[0];
+    w0[1] = salt_buf4[1];
+    w0[2] = salt_buf4[2];
+    w0[3] = salt_buf4[3];
+    w1[0] = 0;
+    w1[1] = 0;
+    w1[2] = 0;
+    w1[3] = 0;
+    w2[0] = 0;
+    w2[1] = 0;
+    w2[2] = 0;
+    w2[3] = 0;
+    w3[0] = 0;
+    w3[1] = 0;
+    w3[2] = (64 + salt_len) * 8;
+    w3[3] = 0;
+
+    md5_transform_vector(w0, w1, w2, w3, ipad);
+
+    w0[0] = ipad[0];
+    w0[1] = ipad[1];
+    w0[2] = ipad[2];
+    w0[3] = ipad[3];
+    w1[0] = 0x80;
+    w1[1] = 0;
+    w1[2] = 0;
+    w1[3] = 0;
+    w2[0] = 0;
+    w2[1] = 0;
+    w2[2] = 0;
+    w2[3] = 0;
+    w3[0] = 0;
+    w3[1] = 0;
+    w3[2] = (64 + 16) * 8;
+    w3[3] = 0;
+
+    md5_transform_vector (w0, w1, w2, w3, opad);
+
+    // hmac_md5_run (w0, w1, w2, w3, ipad, opad, digest);
+
+    COMPARE_S_SIMD (opad[0], opad[3], opad[2], opad[1]);
+  }
+}
+
+KERNEL_FQ void m07350_s08 (KERN_ATTR_RULES ())
+{
+}
+
+KERNEL_FQ void m07350_s16 (KERN_ATTR_RULES ())
+{
+}

OpenCL/m07350_a0-pure.cl

@@ -0,0 +1,135 @@
+/**
+ * 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
+
+KERNEL_FQ void m07350_mxx (KERN_ATTR_RULES ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+
+  if (gid >= GID_CNT) return;
+
+  /**
+   * base
+   */
+
+  COPY_PW (pws[gid]);
+
+  const u32 salt_len = salt_bufs[SALT_POS_HOST].salt_len;
+
+  u32 s[64] = { 0 };
+
+  for (u32 i = 0, idx = 0; i < salt_len; i += 4, idx += 1)
+  {
+    s[idx] = salt_bufs[SALT_POS_HOST].salt_buf[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_hmac_ctx_t ctx;
+
+    md5_hmac_init (&ctx, tmp.i, tmp.pw_len);
+
+    md5_hmac_update (&ctx, s, salt_len);
+
+    md5_hmac_final (&ctx);
+
+    const u32 r0 = ctx.opad.h[DGST_R0];
+    const u32 r1 = ctx.opad.h[DGST_R1];
+    const u32 r2 = ctx.opad.h[DGST_R2];
+    const u32 r3 = ctx.opad.h[DGST_R3];
+
+    COMPARE_M_SCALAR (r0, r1, r2, r3);
+  }
+}
+
+KERNEL_FQ void m07350_sxx (KERN_ATTR_RULES ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+
+  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 = salt_bufs[SALT_POS_HOST].salt_len;
+
+  u32 s[64] = { 0 };
+
+  for (u32 i = 0, idx = 0; i < salt_len; i += 4, idx += 1)
+  {
+    s[idx] = salt_bufs[SALT_POS_HOST].salt_buf[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_hmac_ctx_t ctx;
+
+    md5_hmac_init (&ctx, tmp.i, tmp.pw_len);
+
+    md5_hmac_update (&ctx, s, salt_len);
+
+    md5_hmac_final (&ctx);
+
+    const u32 r0 = ctx.opad.h[DGST_R0];
+    const u32 r1 = ctx.opad.h[DGST_R1];
+    const u32 r2 = ctx.opad.h[DGST_R2];
+    const u32 r3 = ctx.opad.h[DGST_R3];
+
+    COMPARE_S_SCALAR (r0, r1, r2, r3);
+  }
+}

OpenCL/m07350_a3-optimized.cl

@@ -0,0 +1,660 @@
+/**
+ * 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
+
+DECLSPEC void hmac_md5_pad (PRIVATE_AS u32x *w0, PRIVATE_AS u32x *w1, PRIVATE_AS u32x *w2, PRIVATE_AS u32x *w3, PRIVATE_AS u32x *ipad, PRIVATE_AS u32x *opad)
+{
+  w0[0] = w0[0] ^ 0x36363636;
+  w0[1] = w0[1] ^ 0x36363636;
+  w0[2] = w0[2] ^ 0x36363636;
+  w0[3] = w0[3] ^ 0x36363636;
+  w1[0] = w1[0] ^ 0x36363636;
+  w1[1] = w1[1] ^ 0x36363636;
+  w1[2] = w1[2] ^ 0x36363636;
+  w1[3] = w1[3] ^ 0x36363636;
+  w2[0] = w2[0] ^ 0x36363636;
+  w2[1] = w2[1] ^ 0x36363636;
+  w2[2] = w2[2] ^ 0x36363636;
+  w2[3] = w2[3] ^ 0x36363636;
+  w3[0] = w3[0] ^ 0x36363636;
+  w3[1] = w3[1] ^ 0x36363636;
+  w3[2] = w3[2] ^ 0x36363636;
+  w3[3] = w3[3] ^ 0x36363636;
+
+  ipad[0] = MD5M_A;
+  ipad[1] = MD5M_B;
+  ipad[2] = MD5M_C;
+  ipad[3] = MD5M_D;
+
+  md5_transform_vector (w0, w1, w2, w3, ipad);
+
+  w0[0] = w0[0] ^ 0x6a6a6a6a;
+  w0[1] = w0[1] ^ 0x6a6a6a6a;
+  w0[2] = w0[2] ^ 0x6a6a6a6a;
+  w0[3] = w0[3] ^ 0x6a6a6a6a;
+  w1[0] = w1[0] ^ 0x6a6a6a6a;
+  w1[1] = w1[1] ^ 0x6a6a6a6a;
+  w1[2] = w1[2] ^ 0x6a6a6a6a;
+  w1[3] = w1[3] ^ 0x6a6a6a6a;
+  w2[0] = w2[0] ^ 0x6a6a6a6a;
+  w2[1] = w2[1] ^ 0x6a6a6a6a;
+  w2[2] = w2[2] ^ 0x6a6a6a6a;
+  w2[3] = w2[3] ^ 0x6a6a6a6a;
+  w3[0] = w3[0] ^ 0x6a6a6a6a;
+  w3[1] = w3[1] ^ 0x6a6a6a6a;
+  w3[2] = w3[2] ^ 0x6a6a6a6a;
+  w3[3] = w3[3] ^ 0x6a6a6a6a;
+
+  opad[0] = MD5M_A;
+  opad[1] = MD5M_B;
+  opad[2] = MD5M_C;
+  opad[3] = MD5M_D;
+
+  md5_transform_vector (w0, w1, w2, w3, opad);
+}
+
+DECLSPEC void m07350m (PRIVATE_AS u32 *w0, PRIVATE_AS u32 *w1, PRIVATE_AS u32 *w2, PRIVATE_AS u32 *w3, const u32 pw_len, KERN_ATTR_FUNC_BASIC ())
+{
+  /**
+   * modifiers are taken from args
+   */
+
+  /**
+   * salt
+   */
+
+  u32 salt_buf0[4];
+  u32 salt_buf1[4];
+  u32 salt_buf2[4];
+  u32 salt_buf3[4];
+  u32 salt_buf4[4];
+
+  salt_buf0[0] = salt_bufs[SALT_POS_HOST].salt_buf[ 0];
+  salt_buf0[1] = salt_bufs[SALT_POS_HOST].salt_buf[ 1];
+  salt_buf0[2] = salt_bufs[SALT_POS_HOST].salt_buf[ 2];
+  salt_buf0[3] = salt_bufs[SALT_POS_HOST].salt_buf[ 3];
+  salt_buf1[0] = salt_bufs[SALT_POS_HOST].salt_buf[ 4];
+  salt_buf1[1] = salt_bufs[SALT_POS_HOST].salt_buf[ 5];
+  salt_buf1[2] = salt_bufs[SALT_POS_HOST].salt_buf[ 6];
+  salt_buf1[3] = salt_bufs[SALT_POS_HOST].salt_buf[ 7];
+  salt_buf2[0] = salt_bufs[SALT_POS_HOST].salt_buf[ 8];
+  salt_buf2[1] = salt_bufs[SALT_POS_HOST].salt_buf[ 9];
+  salt_buf2[2] = salt_bufs[SALT_POS_HOST].salt_buf[10];
+  salt_buf2[3] = salt_bufs[SALT_POS_HOST].salt_buf[11];
+  salt_buf3[0] = salt_bufs[SALT_POS_HOST].salt_buf[12];
+  salt_buf3[1] = salt_bufs[SALT_POS_HOST].salt_buf[13];
+  salt_buf3[2] = salt_bufs[SALT_POS_HOST].salt_buf[14];
+  salt_buf3[3] = salt_bufs[SALT_POS_HOST].salt_buf[15];
+
+  salt_buf4[0] = salt_bufs[SALT_POS_HOST].salt_buf[16];
+  salt_buf4[1] = salt_bufs[SALT_POS_HOST].salt_buf[17];
+  salt_buf4[2] = salt_bufs[SALT_POS_HOST].salt_buf[18];
+  salt_buf4[3] = salt_bufs[SALT_POS_HOST].salt_buf[19];
+
+  const u32 salt_len = salt_bufs[SALT_POS_HOST].salt_len;
+
+  /**
+   * 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;
+
+    /**
+     * pads
+     */
+
+    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];
+
+    u32x ipad[4];
+    u32x opad[4];
+
+    hmac_md5_pad (w0_t, w1_t, w2_t, w3_t, ipad, opad);
+
+    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] = salt_buf2[0];
+    w2_t[1] = salt_buf2[1];
+    w2_t[2] = salt_buf2[2];
+    w2_t[3] = salt_buf2[3];
+    w3_t[0] = salt_buf3[0];
+    w3_t[1] = salt_buf3[1];
+    w3_t[2] = salt_buf3[2];
+    w3_t[3] = salt_buf3[3];
+
+    md5_transform_vector(w0_t, w1_t, w2_t, w3_t, ipad);
+
+    w0_t[0] = salt_buf4[0];
+    w0_t[1] = salt_buf4[1];
+    w0_t[2] = salt_buf4[2];
+    w0_t[3] = salt_buf4[3];
+    w1_t[0] = 0;
+    w1_t[1] = 0;
+    w1_t[2] = 0;
+    w1_t[3] = 0;
+    w2_t[0] = 0;
+    w2_t[1] = 0;
+    w2_t[2] = 0;
+    w2_t[3] = 0;
+    w3_t[0] = 0;
+    w3_t[1] = 0;
+    w3_t[2] = (64 + salt_len) * 8;
+    w3_t[3] = 0;
+
+    md5_transform_vector(w0_t, w1_t, w2_t, w3_t, ipad);
+
+    w0_t[0] = ipad[0];
+    w0_t[1] = ipad[1];
+    w0_t[2] = ipad[2];
+    w0_t[3] = ipad[3];
+    w1_t[0] = 0x80;
+    w1_t[1] = 0;
+    w1_t[2] = 0;
+    w1_t[3] = 0;
+    w2_t[0] = 0;
+    w2_t[1] = 0;
+    w2_t[2] = 0;
+    w2_t[3] = 0;
+    w3_t[0] = 0;
+    w3_t[1] = 0;
+    w3_t[2] = (64 + 16) * 8;
+    w3_t[3] = 0;
+
+    md5_transform_vector (w0_t, w1_t, w2_t, w3_t, opad);
+
+    COMPARE_M_SIMD (opad[0], opad[3], opad[2], opad[1]);
+  }
+}
+
+DECLSPEC void m07350s (PRIVATE_AS u32 *w0, PRIVATE_AS u32 *w1, PRIVATE_AS u32 *w2, PRIVATE_AS u32 *w3, const u32 pw_len, KERN_ATTR_FUNC_BASIC ())
+{
+  /**
+   * modifiers are taken from args
+   */
+
+  /**
+   * salt
+   */
+
+  u32 salt_buf0[4];
+  u32 salt_buf1[4];
+  u32 salt_buf2[4];
+  u32 salt_buf3[4];
+  u32 salt_buf4[4];
+
+  salt_buf0[0] = salt_bufs[SALT_POS_HOST].salt_buf[ 0];
+  salt_buf0[1] = salt_bufs[SALT_POS_HOST].salt_buf[ 1];
+  salt_buf0[2] = salt_bufs[SALT_POS_HOST].salt_buf[ 2];
+  salt_buf0[3] = salt_bufs[SALT_POS_HOST].salt_buf[ 3];
+  salt_buf1[0] = salt_bufs[SALT_POS_HOST].salt_buf[ 4];
+  salt_buf1[1] = salt_bufs[SALT_POS_HOST].salt_buf[ 5];
+  salt_buf1[2] = salt_bufs[SALT_POS_HOST].salt_buf[ 6];
+  salt_buf1[3] = salt_bufs[SALT_POS_HOST].salt_buf[ 7];
+  salt_buf2[0] = salt_bufs[SALT_POS_HOST].salt_buf[ 8];
+  salt_buf2[1] = salt_bufs[SALT_POS_HOST].salt_buf[ 9];
+  salt_buf2[2] = salt_bufs[SALT_POS_HOST].salt_buf[10];
+  salt_buf2[3] = salt_bufs[SALT_POS_HOST].salt_buf[11];
+  salt_buf3[0] = salt_bufs[SALT_POS_HOST].salt_buf[12];
+  salt_buf3[1] = salt_bufs[SALT_POS_HOST].salt_buf[13];
+  salt_buf3[2] = salt_bufs[SALT_POS_HOST].salt_buf[14];
+  salt_buf3[3] = salt_bufs[SALT_POS_HOST].salt_buf[15];
+
+  salt_buf4[0] = salt_bufs[SALT_POS_HOST].salt_buf[16];
+  salt_buf4[1] = salt_bufs[SALT_POS_HOST].salt_buf[17];
+  salt_buf4[2] = salt_bufs[SALT_POS_HOST].salt_buf[18];
+  salt_buf4[3] = salt_bufs[SALT_POS_HOST].salt_buf[19];
+
+  const u32 salt_len = salt_bufs[SALT_POS_HOST].salt_len;
+
+  /**
+   * 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;
+
+    /**
+     * pads
+     */
+
+    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];
+
+    u32x ipad[4];
+    u32x opad[4];
+
+    hmac_md5_pad (w0_t, w1_t, w2_t, w3_t, ipad, opad);
+
+    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] = salt_buf2[0];
+    w2_t[1] = salt_buf2[1];
+    w2_t[2] = salt_buf2[2];
+    w2_t[3] = salt_buf2[3];
+    w3_t[0] = salt_buf3[0];
+    w3_t[1] = salt_buf3[1];
+    w3_t[2] = salt_buf3[2];
+    w3_t[3] = salt_buf3[3];
+
+    md5_transform_vector(w0_t, w1_t, w2_t, w3_t, ipad);
+
+    w0_t[0] = salt_buf4[0];
+    w0_t[1] = salt_buf4[1];
+    w0_t[2] = salt_buf4[2];
+    w0_t[3] = salt_buf4[3];
+    w1_t[0] = 0;
+    w1_t[1] = 0;
+    w1_t[2] = 0;
+    w1_t[3] = 0;
+    w2_t[0] = 0;
+    w2_t[1] = 0;
+    w2_t[2] = 0;
+    w2_t[3] = 0;
+    w3_t[0] = 0;
+    w3_t[1] = 0;
+    w3_t[2] = (64 + salt_len) * 8;
+    w3_t[3] = 0;
+
+    md5_transform_vector(w0_t, w1_t, w2_t, w3_t, ipad);
+
+    w0_t[0] = ipad[0];
+    w0_t[1] = ipad[1];
+    w0_t[2] = ipad[2];
+    w0_t[3] = ipad[3];
+    w1_t[0] = 0x80;
+    w1_t[1] = 0;
+    w1_t[2] = 0;
+    w1_t[3] = 0;
+    w2_t[0] = 0;
+    w2_t[1] = 0;
+    w2_t[2] = 0;
+    w2_t[3] = 0;
+    w3_t[0] = 0;
+    w3_t[1] = 0;
+    w3_t[2] = (64 + 16) * 8;
+    w3_t[3] = 0;
+
+    md5_transform_vector (w0_t, w1_t, w2_t, w3_t, opad);
+
+    COMPARE_M_SIMD (opad[0], opad[3], opad[2], opad[1]);
+  }
+}
+
+KERNEL_FQ void m07350_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);
+
+  if (gid >= GID_CNT) return;
+
+  u32 w0[4];
+
+  w0[0] = pws[gid].i[ 0];
+  w0[1] = pws[gid].i[ 1];
+  w0[2] = pws[gid].i[ 2];
+  w0[3] = pws[gid].i[ 3];
+
+  u32 w1[4];
+
+  w1[0] = 0;
+  w1[1] = 0;
+  w1[2] = 0;
+  w1[3] = 0;
+
+  u32 w2[4];
+
+  w2[0] = 0;
+  w2[1] = 0;
+  w2[2] = 0;
+  w2[3] = 0;
+
+  u32 w3[4];
+
+  w3[0] = 0;
+  w3[1] = 0;
+  w3[2] = 0;
+  w3[3] = 0;
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * main
+   */
+
+  m07350m (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);
+}
+
+KERNEL_FQ void m07350_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);
+
+  if (gid >= GID_CNT) return;
+
+  u32 w0[4];
+
+  w0[0] = pws[gid].i[ 0];
+  w0[1] = pws[gid].i[ 1];
+  w0[2] = pws[gid].i[ 2];
+  w0[3] = pws[gid].i[ 3];
+
+  u32 w1[4];
+
+  w1[0] = pws[gid].i[ 4];
+  w1[1] = pws[gid].i[ 5];
+  w1[2] = pws[gid].i[ 6];
+  w1[3] = pws[gid].i[ 7];
+
+  u32 w2[4];
+
+  w2[0] = 0;
+  w2[1] = 0;
+  w2[2] = 0;
+  w2[3] = 0;
+
+  u32 w3[4];
+
+  w3[0] = 0;
+  w3[1] = 0;
+  w3[2] = 0;
+  w3[3] = 0;
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * main
+   */
+
+  m07350m (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);
+}
+
+KERNEL_FQ void m07350_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);
+
+  if (gid >= GID_CNT) return;
+
+  u32 w0[4];
+
+  w0[0] = pws[gid].i[ 0];
+  w0[1] = pws[gid].i[ 1];
+  w0[2] = pws[gid].i[ 2];
+  w0[3] = pws[gid].i[ 3];
+
+  u32 w1[4];
+
+  w1[0] = pws[gid].i[ 4];
+  w1[1] = pws[gid].i[ 5];
+  w1[2] = pws[gid].i[ 6];
+  w1[3] = pws[gid].i[ 7];
+
+  u32 w2[4];
+
+  w2[0] = pws[gid].i[ 8];
+  w2[1] = pws[gid].i[ 9];
+  w2[2] = pws[gid].i[10];
+  w2[3] = pws[gid].i[11];
+
+  u32 w3[4];
+
+  w3[0] = pws[gid].i[12];
+  w3[1] = pws[gid].i[13];
+  w3[2] = 0;
+  w3[3] = 0;
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * main
+   */
+
+  m07350m (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);
+}
+
+KERNEL_FQ void m07350_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);
+
+  if (gid >= GID_CNT) return;
+
+  u32 w0[4];
+
+  w0[0] = pws[gid].i[ 0];
+  w0[1] = pws[gid].i[ 1];
+  w0[2] = pws[gid].i[ 2];
+  w0[3] = pws[gid].i[ 3];
+
+  u32 w1[4];
+
+  w1[0] = 0;
+  w1[1] = 0;
+  w1[2] = 0;
+  w1[3] = 0;
+
+  u32 w2[4];
+
+  w2[0] = 0;
+  w2[1] = 0;
+  w2[2] = 0;
+  w2[3] = 0;
+
+  u32 w3[4];
+
+  w3[0] = 0;
+  w3[1] = 0;
+  w3[2] = 0;
+  w3[3] = 0;
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * main
+   */
+
+  m07350s (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);
+}
+
+KERNEL_FQ void m07350_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);
+
+  if (gid >= GID_CNT) return;
+
+  u32 w0[4];
+
+  w0[0] = pws[gid].i[ 0];
+  w0[1] = pws[gid].i[ 1];
+  w0[2] = pws[gid].i[ 2];
+  w0[3] = pws[gid].i[ 3];
+
+  u32 w1[4];
+
+  w1[0] = pws[gid].i[ 4];
+  w1[1] = pws[gid].i[ 5];
+  w1[2] = pws[gid].i[ 6];
+  w1[3] = pws[gid].i[ 7];
+
+  u32 w2[4];
+
+  w2[0] = 0;
+  w2[1] = 0;
+  w2[2] = 0;
+  w2[3] = 0;
+
+  u32 w3[4];
+
+  w3[0] = 0;
+  w3[1] = 0;
+  w3[2] = 0;
+  w3[3] = 0;
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * main
+   */
+
+  m07350s (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);
+}
+
+KERNEL_FQ void m07350_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);
+
+  if (gid >= GID_CNT) return;
+
+  u32 w0[4];
+
+  w0[0] = pws[gid].i[ 0];
+  w0[1] = pws[gid].i[ 1];
+  w0[2] = pws[gid].i[ 2];
+  w0[3] = pws[gid].i[ 3];
+
+  u32 w1[4];
+
+  w1[0] = pws[gid].i[ 4];
+  w1[1] = pws[gid].i[ 5];
+  w1[2] = pws[gid].i[ 6];
+  w1[3] = pws[gid].i[ 7];
+
+  u32 w2[4];
+
+  w2[0] = pws[gid].i[ 8];
+  w2[1] = pws[gid].i[ 9];
+  w2[2] = pws[gid].i[10];
+  w2[3] = pws[gid].i[11];
+
+  u32 w3[4];
+
+  w3[0] = pws[gid].i[12];
+  w3[1] = pws[gid].i[13];
+  w3[2] = 0;
+  w3[3] = 0;
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * main
+   */
+
+  m07350s (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);
+}

OpenCL/m07350_a3-pure.cl

@@ -0,0 +1,155 @@
+/**
+ * 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
+
+KERNEL_FQ void m00050_mxx (KERN_ATTR_VECTOR ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+
+  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 = salt_bufs[SALT_POS_HOST].salt_len;
+
+  u32x s[64] = { 0 };
+
+  for (u32 i = 0, idx = 0; i < salt_len; i += 4, idx += 1)
+  {
+    s[idx] = salt_bufs[SALT_POS_HOST].salt_buf[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 w0 = w0l | w0r;
+
+    w[0] = w0;
+
+    md5_hmac_ctx_vector_t ctx;
+
+    md5_hmac_init_vector (&ctx, w, pw_len);
+
+    md5_hmac_update_vector (&ctx, s, salt_len);
+
+    md5_hmac_final_vector (&ctx);
+
+    const u32x r0 = ctx.opad.h[DGST_R0];
+    const u32x r1 = ctx.opad.h[DGST_R1];
+    const u32x r2 = ctx.opad.h[DGST_R2];
+    const u32x r3 = ctx.opad.h[DGST_R3];
+
+    COMPARE_M_SIMD (r0, r1, r2, r3);
+  }
+}
+
+KERNEL_FQ void m00050_sxx (KERN_ATTR_VECTOR ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+
+  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 = salt_bufs[SALT_POS_HOST].salt_len;
+
+  u32x s[64] = { 0 };
+
+  for (u32 i = 0, idx = 0; i < salt_len; i += 4, idx += 1)
+  {
+    s[idx] = salt_bufs[SALT_POS_HOST].salt_buf[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 w0 = w0l | w0r;
+
+    w[0] = w0;
+
+    md5_hmac_ctx_vector_t ctx;
+
+    md5_hmac_init_vector (&ctx, w, pw_len);
+
+    md5_hmac_update_vector (&ctx, s, salt_len);
+
+    md5_hmac_final_vector (&ctx);
+
+    const u32x r0 = ctx.opad.h[DGST_R0];
+    const u32x r1 = ctx.opad.h[DGST_R1];
+    const u32x r2 = ctx.opad.h[DGST_R2];
+    const u32x r3 = ctx.opad.h[DGST_R3];
+
+    COMPARE_S_SIMD (r0, r1, r2, r3);
+  }
+}

src/modules/module_07350.c

@@ -0,0 +1,198 @@
+/**
+ * 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"
+
+#define RAKP_HEX_SALT_MIN 116
+#define RAKP_HEX_SALT_MAX 148
+
+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_NETWORK_PROTOCOL;
+static const char *HASH_NAME      = "IPMI2 RAKP HMAC-MD5";
+static const u64   KERN_TYPE      = 7350;
+static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE
+                                  | OPTI_TYPE_NOT_ITERATED;
+static const u64   OPTS_TYPE      = OPTS_TYPE_STOCK_MODULE
+                                  | OPTS_TYPE_PT_GENERATE_LE
+                                  | OPTS_TYPE_ST_ADD80
+                                  | OPTS_TYPE_ST_ADDBITS14
+                                  | OPTS_TYPE_ST_HEX;
+static const u32   SALT_TYPE      = SALT_TYPE_GENERIC;
+static const char *ST_PASS        = "admin";
+static const char *ST_HASH        = "08b017f3628b9835c748521e412429c9:f3450000df540000cdd981b0b3441be8774a61e69321291891a29a0c5fdac3f06194bd2c29fa5246000000000000000000000000000000001400";
+
+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;         }
+
+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] = RAKP_HEX_SALT_MIN;
+  token.len_max[1] = RAKP_HEX_SALT_MAX;
+  token.attr[1]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  const int rc_tokenizer = input_tokenizer ((const u8 *) line_buf, line_len, &token);
+
+  if (rc_tokenizer != PARSER_OK) return (rc_tokenizer);
+
+  const u8 *hash_pos = token.buf[0];
+
+  digest[0] = hex_to_u32 (hash_pos +  0);
+  digest[1] = hex_to_u32 (hash_pos +  8);
+  digest[2] = hex_to_u32 (hash_pos + 16);
+  digest[3] = hex_to_u32 (hash_pos + 24);
+
+  const u8 *salt_pos = token.buf[1];
+  const int salt_len = token.len[1];
+
+  const bool parse_rc = generic_salt_decode (hashconfig, salt_pos, salt_len, (u8 *) salt->salt_buf, (int *) &salt->salt_len);
+
+  if (parse_rc == false) return (PARSER_SALT_LENGTH);
+
+  return (PARSER_OK);
+}
+
+int module_hash_encode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const void *digest_buf, MAYBE_UNUSED const salt_t *salt, MAYBE_UNUSED const void *esalt_buf, MAYBE_UNUSED const void *hook_salt_buf, MAYBE_UNUSED const hashinfo_t *hash_info, char *line_buf, MAYBE_UNUSED const int line_size)
+{
+  const u32 *digest = (const u32 *) digest_buf;
+
+  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;
+}