OpenCL Kernels: Use three separate comparison kernels (depending on keyver) for WPA instead of one

OpenCL/inc_common.cl

@@ -3,7 +3,7 @@
  * License.....: MIT
  */
 
-__constant static u32a c_append_helper[64][16] =
+__constant u32a c_append_helper[64][16] =
 {
   { 0x000000ff, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000 },
   { 0x0000ff00, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000 },

OpenCL/m02500.cl

@@ -243,6 +243,395 @@ __kernel void m02500_loop (__global pw_t *pws, __global const kernel_rule_t *rul
 }
 
 __kernel void m02500_comp (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global wpa_tmp_t *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const wpa_t *wpa_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u64 gid_max)
+{
+  // not in use here, special case...
+}
+
+__kernel void m02500_aux1 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global wpa_tmp_t *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const wpa_t *wpa_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u64 gid_max)
+{
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+
+  if (gid >= gid_max) return;
+
+  u32 out[8];
+
+  out[0] = tmps[gid].out[0];
+  out[1] = tmps[gid].out[1];
+  out[2] = tmps[gid].out[2];
+  out[3] = tmps[gid].out[3];
+  out[4] = tmps[gid].out[4];
+  out[5] = tmps[gid].out[5];
+  out[6] = tmps[gid].out[6];
+  out[7] = tmps[gid].out[7];
+
+  const u32 digest_pos = loop_pos;
+
+  const u32 digest_cur = digests_offset + digest_pos;
+
+  __global const wpa_t *wpa = &wpa_bufs[digest_cur];
+
+  u32 pke[32];
+
+  pke[ 0] = wpa->pke[ 0];
+  pke[ 1] = wpa->pke[ 1];
+  pke[ 2] = wpa->pke[ 2];
+  pke[ 3] = wpa->pke[ 3];
+  pke[ 4] = wpa->pke[ 4];
+  pke[ 5] = wpa->pke[ 5];
+  pke[ 6] = wpa->pke[ 6];
+  pke[ 7] = wpa->pke[ 7];
+  pke[ 8] = wpa->pke[ 8];
+  pke[ 9] = wpa->pke[ 9];
+  pke[10] = wpa->pke[10];
+  pke[11] = wpa->pke[11];
+  pke[12] = wpa->pke[12];
+  pke[13] = wpa->pke[13];
+  pke[14] = wpa->pke[14];
+  pke[15] = wpa->pke[15];
+  pke[16] = wpa->pke[16];
+  pke[17] = wpa->pke[17];
+  pke[18] = wpa->pke[18];
+  pke[19] = wpa->pke[19];
+  pke[20] = wpa->pke[20];
+  pke[21] = wpa->pke[21];
+  pke[22] = wpa->pke[22];
+  pke[23] = wpa->pke[23];
+  pke[24] = wpa->pke[24];
+  pke[25] = wpa->pke[25];
+  pke[26] = wpa->pke[26];
+  pke[27] = wpa->pke[27];
+  pke[28] = wpa->pke[28];
+  pke[29] = wpa->pke[29];
+  pke[30] = wpa->pke[30];
+  pke[31] = wpa->pke[31];
+
+  u32 to;
+
+  if (wpa->nonce_compare < 0)
+  {
+    to = pke[15] << 24
+       | pke[16] >>  8;
+  }
+  else
+  {
+    to = pke[23] << 24
+       | pke[24] >>  8;
+  }
+
+  const u32 nonce_error_corrections = wpa->nonce_error_corrections;
+
+  for (u32 nonce_error_correction = 0; nonce_error_correction <= nonce_error_corrections; nonce_error_correction++)
+  {
+    u32 t = to;
+
+    t = swap32_S (t);
+
+    t -= nonce_error_corrections / 2;
+    t += nonce_error_correction;
+
+    t = swap32_S (t);
+
+    if (wpa->nonce_compare < 0)
+    {
+      pke[15] = (pke[15] & ~0x000000ff) | (t >> 24);
+      pke[16] = (pke[16] & ~0xffffff00) | (t <<  8);
+    }
+    else
+    {
+      pke[23] = (pke[23] & ~0x000000ff) | (t >> 24);
+      pke[24] = (pke[24] & ~0xffffff00) | (t <<  8);
+    }
+
+    u32 w0[4];
+    u32 w1[4];
+    u32 w2[4];
+    u32 w3[4];
+
+    w0[0] = out[0];
+    w0[1] = out[1];
+    w0[2] = out[2];
+    w0[3] = out[3];
+    w1[0] = out[4];
+    w1[1] = out[5];
+    w1[2] = out[6];
+    w1[3] = out[7];
+    w2[0] = 0;
+    w2[1] = 0;
+    w2[2] = 0;
+    w2[3] = 0;
+    w3[0] = 0;
+    w3[1] = 0;
+    w3[2] = 0;
+    w3[3] = 0;
+
+    u32 keymic[4];
+
+    keymic[0] = 0;
+    keymic[1] = 0;
+    keymic[2] = 0;
+    keymic[3] = 0;
+
+    sha1_hmac_ctx_t ctx1;
+
+    sha1_hmac_init_64 (&ctx1, w0, w1, w2, w3);
+
+    sha1_hmac_update (&ctx1, pke, 100);
+
+    sha1_hmac_final (&ctx1);
+
+    u32 digest[4];
+
+    digest[0] = ctx1.opad.h[0];
+    digest[1] = ctx1.opad.h[1];
+    digest[2] = ctx1.opad.h[2];
+    digest[3] = ctx1.opad.h[3];
+
+    u32 t0[4];
+    u32 t1[4];
+    u32 t2[4];
+    u32 t3[4];
+
+    t0[0] = swap32_S (digest[0]);
+    t0[1] = swap32_S (digest[1]);
+    t0[2] = swap32_S (digest[2]);
+    t0[3] = swap32_S (digest[3]);
+    t1[0] = 0;
+    t1[1] = 0;
+    t1[2] = 0;
+    t1[3] = 0;
+    t2[0] = 0;
+    t2[1] = 0;
+    t2[2] = 0;
+    t2[3] = 0;
+    t3[0] = 0;
+    t3[1] = 0;
+    t3[2] = 0;
+    t3[3] = 0;
+
+    md5_hmac_ctx_t ctx2;
+
+    md5_hmac_init_64 (&ctx2, t0, t1, t2, t3);
+
+    md5_hmac_update_global (&ctx2, wpa->eapol, wpa->eapol_len);
+
+    md5_hmac_final (&ctx2);
+
+    keymic[0] = ctx2.opad.h[0];
+    keymic[1] = ctx2.opad.h[1];
+    keymic[2] = ctx2.opad.h[2];
+    keymic[3] = ctx2.opad.h[3];
+
+    /**
+     * final compare
+     */
+
+    if ((keymic[0] == wpa->keymic[0])
+     && (keymic[1] == wpa->keymic[1])
+     && (keymic[2] == wpa->keymic[2])
+     && (keymic[3] == wpa->keymic[3]))
+    {
+      if (atomic_inc (&hashes_shown[digest_cur]) == 0)
+      {
+        mark_hash (plains_buf, d_return_buf, salt_pos, digests_cnt, digest_pos, digest_cur, gid, 0);
+      }
+    }
+  }
+}
+
+__kernel void m02500_aux2 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global wpa_tmp_t *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const wpa_t *wpa_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u64 gid_max)
+{
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+
+  if (gid >= gid_max) return;
+
+  u32 out[8];
+
+  out[0] = tmps[gid].out[0];
+  out[1] = tmps[gid].out[1];
+  out[2] = tmps[gid].out[2];
+  out[3] = tmps[gid].out[3];
+  out[4] = tmps[gid].out[4];
+  out[5] = tmps[gid].out[5];
+  out[6] = tmps[gid].out[6];
+  out[7] = tmps[gid].out[7];
+
+  const u32 digest_pos = loop_pos;
+
+  const u32 digest_cur = digests_offset + digest_pos;
+
+  __global const wpa_t *wpa = &wpa_bufs[digest_cur];
+
+  u32 pke[32];
+
+  pke[ 0] = wpa->pke[ 0];
+  pke[ 1] = wpa->pke[ 1];
+  pke[ 2] = wpa->pke[ 2];
+  pke[ 3] = wpa->pke[ 3];
+  pke[ 4] = wpa->pke[ 4];
+  pke[ 5] = wpa->pke[ 5];
+  pke[ 6] = wpa->pke[ 6];
+  pke[ 7] = wpa->pke[ 7];
+  pke[ 8] = wpa->pke[ 8];
+  pke[ 9] = wpa->pke[ 9];
+  pke[10] = wpa->pke[10];
+  pke[11] = wpa->pke[11];
+  pke[12] = wpa->pke[12];
+  pke[13] = wpa->pke[13];
+  pke[14] = wpa->pke[14];
+  pke[15] = wpa->pke[15];
+  pke[16] = wpa->pke[16];
+  pke[17] = wpa->pke[17];
+  pke[18] = wpa->pke[18];
+  pke[19] = wpa->pke[19];
+  pke[20] = wpa->pke[20];
+  pke[21] = wpa->pke[21];
+  pke[22] = wpa->pke[22];
+  pke[23] = wpa->pke[23];
+  pke[24] = wpa->pke[24];
+  pke[25] = wpa->pke[25];
+  pke[26] = wpa->pke[26];
+  pke[27] = wpa->pke[27];
+  pke[28] = wpa->pke[28];
+  pke[29] = wpa->pke[29];
+  pke[30] = wpa->pke[30];
+  pke[31] = wpa->pke[31];
+
+  u32 to;
+
+  if (wpa->nonce_compare < 0)
+  {
+    to = pke[15] << 24
+       | pke[16] >>  8;
+  }
+  else
+  {
+    to = pke[23] << 24
+       | pke[24] >>  8;
+  }
+
+  const u32 nonce_error_corrections = wpa->nonce_error_corrections;
+
+  for (u32 nonce_error_correction = 0; nonce_error_correction <= nonce_error_corrections; nonce_error_correction++)
+  {
+    u32 t = to;
+
+    t = swap32_S (t);
+
+    t -= nonce_error_corrections / 2;
+    t += nonce_error_correction;
+
+    t = swap32_S (t);
+
+    if (wpa->nonce_compare < 0)
+    {
+      pke[15] = (pke[15] & ~0x000000ff) | (t >> 24);
+      pke[16] = (pke[16] & ~0xffffff00) | (t <<  8);
+    }
+    else
+    {
+      pke[23] = (pke[23] & ~0x000000ff) | (t >> 24);
+      pke[24] = (pke[24] & ~0xffffff00) | (t <<  8);
+    }
+
+    u32 w0[4];
+    u32 w1[4];
+    u32 w2[4];
+    u32 w3[4];
+
+    w0[0] = out[0];
+    w0[1] = out[1];
+    w0[2] = out[2];
+    w0[3] = out[3];
+    w1[0] = out[4];
+    w1[1] = out[5];
+    w1[2] = out[6];
+    w1[3] = out[7];
+    w2[0] = 0;
+    w2[1] = 0;
+    w2[2] = 0;
+    w2[3] = 0;
+    w3[0] = 0;
+    w3[1] = 0;
+    w3[2] = 0;
+    w3[3] = 0;
+
+    u32 keymic[4];
+
+    keymic[0] = 0;
+    keymic[1] = 0;
+    keymic[2] = 0;
+    keymic[3] = 0;
+
+    sha1_hmac_ctx_t ctx1;
+
+    sha1_hmac_init_64 (&ctx1, w0, w1, w2, w3);
+
+    sha1_hmac_update (&ctx1, pke, 100);
+
+    sha1_hmac_final (&ctx1);
+
+    u32 digest[4];
+
+    digest[0] = ctx1.opad.h[0];
+    digest[1] = ctx1.opad.h[1];
+    digest[2] = ctx1.opad.h[2];
+    digest[3] = ctx1.opad.h[3];
+
+    u32 t0[4];
+    u32 t1[4];
+    u32 t2[4];
+    u32 t3[4];
+
+    t0[0] = digest[0];
+    t0[1] = digest[1];
+    t0[2] = digest[2];
+    t0[3] = digest[3];
+    t1[0] = 0;
+    t1[1] = 0;
+    t1[2] = 0;
+    t1[3] = 0;
+    t2[0] = 0;
+    t2[1] = 0;
+    t2[2] = 0;
+    t2[3] = 0;
+    t3[0] = 0;
+    t3[1] = 0;
+    t3[2] = 0;
+    t3[3] = 0;
+
+    sha1_hmac_ctx_t ctx2;
+
+    sha1_hmac_init_64 (&ctx2, t0, t1, t2, t3);
+
+    sha1_hmac_update_global (&ctx2, wpa->eapol, wpa->eapol_len);
+
+    sha1_hmac_final (&ctx2);
+
+    keymic[0] = ctx2.opad.h[0];
+    keymic[1] = ctx2.opad.h[1];
+    keymic[2] = ctx2.opad.h[2];
+    keymic[3] = ctx2.opad.h[3];
+
+    /**
+     * final compare
+     */
+
+    if ((keymic[0] == wpa->keymic[0])
+     && (keymic[1] == wpa->keymic[1])
+     && (keymic[2] == wpa->keymic[2])
+     && (keymic[3] == wpa->keymic[3]))
+    {
+      if (atomic_inc (&hashes_shown[digest_cur]) == 0)
+      {
+        mark_hash (plains_buf, d_return_buf, salt_pos, digests_cnt, digest_pos, digest_cur, gid, 0);
+      }
+    }
+  }
+}
+
+__kernel void m02500_aux3 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global wpa_tmp_t *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const wpa_t *wpa_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u64 gid_max)
 {
   const u64 gid = get_global_id (0);
   const u64 lid = get_local_id (0);
@@ -419,112 +808,6 @@ __kernel void m02500_comp (__global pw_t *pws, __global const kernel_rule_t *rul
     keymic[2] = 0;
     keymic[3] = 0;
 
-    if (wpa->keyver == 1)
-    {
-      sha1_hmac_ctx_t ctx1;
-
-      sha1_hmac_init_64 (&ctx1, w0, w1, w2, w3);
-
-      sha1_hmac_update (&ctx1, pke, 100);
-
-      sha1_hmac_final (&ctx1);
-
-      u32 digest[4];
-
-      digest[0] = ctx1.opad.h[0];
-      digest[1] = ctx1.opad.h[1];
-      digest[2] = ctx1.opad.h[2];
-      digest[3] = ctx1.opad.h[3];
-
-      u32 t0[4];
-      u32 t1[4];
-      u32 t2[4];
-      u32 t3[4];
-
-      t0[0] = swap32_S (digest[0]);
-      t0[1] = swap32_S (digest[1]);
-      t0[2] = swap32_S (digest[2]);
-      t0[3] = swap32_S (digest[3]);
-      t1[0] = 0;
-      t1[1] = 0;
-      t1[2] = 0;
-      t1[3] = 0;
-      t2[0] = 0;
-      t2[1] = 0;
-      t2[2] = 0;
-      t2[3] = 0;
-      t3[0] = 0;
-      t3[1] = 0;
-      t3[2] = 0;
-      t3[3] = 0;
-
-      md5_hmac_ctx_t ctx2;
-
-      md5_hmac_init_64 (&ctx2, t0, t1, t2, t3);
-
-      md5_hmac_update_global (&ctx2, wpa->eapol, wpa->eapol_len);
-
-      md5_hmac_final (&ctx2);
-
-      keymic[0] = ctx2.opad.h[0];
-      keymic[1] = ctx2.opad.h[1];
-      keymic[2] = ctx2.opad.h[2];
-      keymic[3] = ctx2.opad.h[3];
-    }
-    else if (wpa->keyver == 2)
-    {
-      sha1_hmac_ctx_t ctx1;
-
-      sha1_hmac_init_64 (&ctx1, w0, w1, w2, w3);
-
-      sha1_hmac_update (&ctx1, pke, 100);
-
-      sha1_hmac_final (&ctx1);
-
-      u32 digest[4];
-
-      digest[0] = ctx1.opad.h[0];
-      digest[1] = ctx1.opad.h[1];
-      digest[2] = ctx1.opad.h[2];
-      digest[3] = ctx1.opad.h[3];
-
-      u32 t0[4];
-      u32 t1[4];
-      u32 t2[4];
-      u32 t3[4];
-
-      t0[0] = digest[0];
-      t0[1] = digest[1];
-      t0[2] = digest[2];
-      t0[3] = digest[3];
-      t1[0] = 0;
-      t1[1] = 0;
-      t1[2] = 0;
-      t1[3] = 0;
-      t2[0] = 0;
-      t2[1] = 0;
-      t2[2] = 0;
-      t2[3] = 0;
-      t3[0] = 0;
-      t3[1] = 0;
-      t3[2] = 0;
-      t3[3] = 0;
-
-      sha1_hmac_ctx_t ctx2;
-
-      sha1_hmac_init_64 (&ctx2, t0, t1, t2, t3);
-
-      sha1_hmac_update_global (&ctx2, wpa->eapol, wpa->eapol_len);
-
-      sha1_hmac_final (&ctx2);
-
-      keymic[0] = ctx2.opad.h[0];
-      keymic[1] = ctx2.opad.h[1];
-      keymic[2] = ctx2.opad.h[2];
-      keymic[3] = ctx2.opad.h[3];
-    }
-    else if (wpa->keyver == 3)
-    {
     sha256_hmac_ctx_t ctx1;
 
     sha256_hmac_init_64 (&ctx1, w0, w1, w2, w3);
@@ -605,262 +888,6 @@ __kernel void m02500_comp (__global pw_t *pws, __global const kernel_rule_t *rul
     m[3] ^= iv[3];
 
     aes128_encrypt (ks, m, keymic, s_te0, s_te1, s_te2, s_te3, s_te4);
-    }
-
-    /**
-     * final compare
-     */
-
-    if ((keymic[0] == wpa->keymic[0])
-     && (keymic[1] == wpa->keymic[1])
-     && (keymic[2] == wpa->keymic[2])
-     && (keymic[3] == wpa->keymic[3]))
-    {
-      if (atomic_inc (&hashes_shown[digest_cur]) == 0)
-      {
-        mark_hash (plains_buf, d_return_buf, salt_pos, digests_cnt, digest_pos, digest_cur, gid, 0);
-      }
-    }
-  }
-
-  // the same code again, but with BE order for the t++
-
-  for (u32 nonce_error_correction = 0; nonce_error_correction <= nonce_error_corrections; nonce_error_correction++)
-  {
-    u32 t = to;
-
-    t -= nonce_error_corrections / 2;
-    t += nonce_error_correction;
-
-    if (t == to) continue; // we already had this checked in the LE loop
-
-    if (wpa->nonce_compare < 0)
-    {
-      pke[15] = (pke[15] & ~0x000000ff) | (t >> 24);
-      pke[16] = (pke[16] & ~0xffffff00) | (t <<  8);
-    }
-    else
-    {
-      pke[23] = (pke[23] & ~0x000000ff) | (t >> 24);
-      pke[24] = (pke[24] & ~0xffffff00) | (t <<  8);
-    }
-
-    u32 w0[4];
-    u32 w1[4];
-    u32 w2[4];
-    u32 w3[4];
-
-    w0[0] = out[0];
-    w0[1] = out[1];
-    w0[2] = out[2];
-    w0[3] = out[3];
-    w1[0] = out[4];
-    w1[1] = out[5];
-    w1[2] = out[6];
-    w1[3] = out[7];
-    w2[0] = 0;
-    w2[1] = 0;
-    w2[2] = 0;
-    w2[3] = 0;
-    w3[0] = 0;
-    w3[1] = 0;
-    w3[2] = 0;
-    w3[3] = 0;
-
-    u32 keymic[4];
-
-    keymic[0] = 0;
-    keymic[1] = 0;
-    keymic[2] = 0;
-    keymic[3] = 0;
-
-    if (wpa->keyver == 1)
-    {
-      sha1_hmac_ctx_t ctx1;
-
-      sha1_hmac_init_64 (&ctx1, w0, w1, w2, w3);
-
-      sha1_hmac_update (&ctx1, pke, 100);
-
-      sha1_hmac_final (&ctx1);
-
-      u32 digest[4];
-
-      digest[0] = ctx1.opad.h[0];
-      digest[1] = ctx1.opad.h[1];
-      digest[2] = ctx1.opad.h[2];
-      digest[3] = ctx1.opad.h[3];
-
-      u32 t0[4];
-      u32 t1[4];
-      u32 t2[4];
-      u32 t3[4];
-
-      t0[0] = swap32_S (digest[0]);
-      t0[1] = swap32_S (digest[1]);
-      t0[2] = swap32_S (digest[2]);
-      t0[3] = swap32_S (digest[3]);
-      t1[0] = 0;
-      t1[1] = 0;
-      t1[2] = 0;
-      t1[3] = 0;
-      t2[0] = 0;
-      t2[1] = 0;
-      t2[2] = 0;
-      t2[3] = 0;
-      t3[0] = 0;
-      t3[1] = 0;
-      t3[2] = 0;
-      t3[3] = 0;
-
-      md5_hmac_ctx_t ctx2;
-
-      md5_hmac_init_64 (&ctx2, t0, t1, t2, t3);
-
-      md5_hmac_update_global (&ctx2, wpa->eapol, wpa->eapol_len);
-
-      md5_hmac_final (&ctx2);
-
-      keymic[0] = ctx2.opad.h[0];
-      keymic[1] = ctx2.opad.h[1];
-      keymic[2] = ctx2.opad.h[2];
-      keymic[3] = ctx2.opad.h[3];
-    }
-    else if (wpa->keyver == 2)
-    {
-      sha1_hmac_ctx_t ctx1;
-
-      sha1_hmac_init_64 (&ctx1, w0, w1, w2, w3);
-
-      sha1_hmac_update (&ctx1, pke, 100);
-
-      sha1_hmac_final (&ctx1);
-
-      u32 digest[4];
-
-      digest[0] = ctx1.opad.h[0];
-      digest[1] = ctx1.opad.h[1];
-      digest[2] = ctx1.opad.h[2];
-      digest[3] = ctx1.opad.h[3];
-
-      u32 t0[4];
-      u32 t1[4];
-      u32 t2[4];
-      u32 t3[4];
-
-      t0[0] = digest[0];
-      t0[1] = digest[1];
-      t0[2] = digest[2];
-      t0[3] = digest[3];
-      t1[0] = 0;
-      t1[1] = 0;
-      t1[2] = 0;
-      t1[3] = 0;
-      t2[0] = 0;
-      t2[1] = 0;
-      t2[2] = 0;
-      t2[3] = 0;
-      t3[0] = 0;
-      t3[1] = 0;
-      t3[2] = 0;
-      t3[3] = 0;
-
-      sha1_hmac_ctx_t ctx2;
-
-      sha1_hmac_init_64 (&ctx2, t0, t1, t2, t3);
-
-      sha1_hmac_update_global (&ctx2, wpa->eapol, wpa->eapol_len);
-
-      sha1_hmac_final (&ctx2);
-
-      keymic[0] = ctx2.opad.h[0];
-      keymic[1] = ctx2.opad.h[1];
-      keymic[2] = ctx2.opad.h[2];
-      keymic[3] = ctx2.opad.h[3];
-    }
-    else if (wpa->keyver == 3)
-    {
-      sha256_hmac_ctx_t ctx1;
-
-      sha256_hmac_init_64 (&ctx1, w0, w1, w2, w3);
-
-      sha256_hmac_update (&ctx1, pke, 102);
-
-      sha256_hmac_final (&ctx1);
-
-      u32 digest[4];
-
-      digest[0] = swap32_S (ctx1.opad.h[0]);
-      digest[1] = swap32_S (ctx1.opad.h[1]);
-      digest[2] = swap32_S (ctx1.opad.h[2]);
-      digest[3] = swap32_S (ctx1.opad.h[3]);
-
-      // AES CMAC
-
-      u32 ks[44];
-
-      aes128_set_encrypt_key (ks, digest, s_te0, s_te1, s_te2, s_te3, s_te4);
-
-      u32 m[4];
-
-      m[0] = 0;
-      m[1] = 0;
-      m[2] = 0;
-      m[3] = 0;
-
-      u32 iv[4];
-
-      iv[0] = 0;
-      iv[1] = 0;
-      iv[2] = 0;
-      iv[3] = 0;
-
-      int eapol_left;
-      int eapol_idx;
-
-      for (eapol_left = wpa->eapol_len, eapol_idx = 0; eapol_left > 16; eapol_left -= 16, eapol_idx += 4)
-      {
-        m[0] = wpa->eapol[eapol_idx + 0] ^ iv[0];
-        m[1] = wpa->eapol[eapol_idx + 1] ^ iv[1];
-        m[2] = wpa->eapol[eapol_idx + 2] ^ iv[2];
-        m[3] = wpa->eapol[eapol_idx + 3] ^ iv[3];
-
-        aes128_encrypt (ks, m, iv, s_te0, s_te1, s_te2, s_te3, s_te4);
-      }
-
-      m[0] = wpa->eapol[eapol_idx + 0];
-      m[1] = wpa->eapol[eapol_idx + 1];
-      m[2] = wpa->eapol[eapol_idx + 2];
-      m[3] = wpa->eapol[eapol_idx + 3];
-
-      u32 k[4];
-
-      k[0] = 0;
-      k[1] = 0;
-      k[2] = 0;
-      k[3] = 0;
-
-      aes128_encrypt (ks, k, k, s_te0, s_te1, s_te2, s_te3, s_te4);
-
-      make_kn (k);
-
-      if (eapol_left < 16)
-      {
-        make_kn (k);
-      }
-
-      m[0] ^= k[0];
-      m[1] ^= k[1];
-      m[2] ^= k[2];
-      m[3] ^= k[3];
-
-      m[0] ^= iv[0];
-      m[1] ^= iv[1];
-      m[2] ^= iv[2];
-      m[3] ^= iv[3];
-
-      aes128_encrypt (ks, m, keymic, s_te0, s_te1, s_te2, s_te3, s_te4);
-    }
 
     /**
      * final compare

OpenCL/m02501.cl

@@ -118,6 +118,395 @@ __kernel void m02501_loop (__global pw_t *pws, __global const kernel_rule_t *rul
 }
 
 __kernel void m02501_comp (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global wpapmk_tmp_t *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const wpa_t *wpa_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u64 gid_max)
+{
+  // not in use here, special case...
+}
+
+__kernel void m02501_aux1 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global wpa_tmp_t *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const wpa_t *wpa_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u64 gid_max)
+{
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+
+  if (gid >= gid_max) return;
+
+  u32 out[8];
+
+  out[0] = tmps[gid].out[0];
+  out[1] = tmps[gid].out[1];
+  out[2] = tmps[gid].out[2];
+  out[3] = tmps[gid].out[3];
+  out[4] = tmps[gid].out[4];
+  out[5] = tmps[gid].out[5];
+  out[6] = tmps[gid].out[6];
+  out[7] = tmps[gid].out[7];
+
+  const u32 digest_pos = loop_pos;
+
+  const u32 digest_cur = digests_offset + digest_pos;
+
+  __global const wpa_t *wpa = &wpa_bufs[digest_cur];
+
+  u32 pke[32];
+
+  pke[ 0] = wpa->pke[ 0];
+  pke[ 1] = wpa->pke[ 1];
+  pke[ 2] = wpa->pke[ 2];
+  pke[ 3] = wpa->pke[ 3];
+  pke[ 4] = wpa->pke[ 4];
+  pke[ 5] = wpa->pke[ 5];
+  pke[ 6] = wpa->pke[ 6];
+  pke[ 7] = wpa->pke[ 7];
+  pke[ 8] = wpa->pke[ 8];
+  pke[ 9] = wpa->pke[ 9];
+  pke[10] = wpa->pke[10];
+  pke[11] = wpa->pke[11];
+  pke[12] = wpa->pke[12];
+  pke[13] = wpa->pke[13];
+  pke[14] = wpa->pke[14];
+  pke[15] = wpa->pke[15];
+  pke[16] = wpa->pke[16];
+  pke[17] = wpa->pke[17];
+  pke[18] = wpa->pke[18];
+  pke[19] = wpa->pke[19];
+  pke[20] = wpa->pke[20];
+  pke[21] = wpa->pke[21];
+  pke[22] = wpa->pke[22];
+  pke[23] = wpa->pke[23];
+  pke[24] = wpa->pke[24];
+  pke[25] = wpa->pke[25];
+  pke[26] = wpa->pke[26];
+  pke[27] = wpa->pke[27];
+  pke[28] = wpa->pke[28];
+  pke[29] = wpa->pke[29];
+  pke[30] = wpa->pke[30];
+  pke[31] = wpa->pke[31];
+
+  u32 to;
+
+  if (wpa->nonce_compare < 0)
+  {
+    to = pke[15] << 24
+       | pke[16] >>  8;
+  }
+  else
+  {
+    to = pke[23] << 24
+       | pke[24] >>  8;
+  }
+
+  const u32 nonce_error_corrections = wpa->nonce_error_corrections;
+
+  for (u32 nonce_error_correction = 0; nonce_error_correction <= nonce_error_corrections; nonce_error_correction++)
+  {
+    u32 t = to;
+
+    t = swap32_S (t);
+
+    t -= nonce_error_corrections / 2;
+    t += nonce_error_correction;
+
+    t = swap32_S (t);
+
+    if (wpa->nonce_compare < 0)
+    {
+      pke[15] = (pke[15] & ~0x000000ff) | (t >> 24);
+      pke[16] = (pke[16] & ~0xffffff00) | (t <<  8);
+    }
+    else
+    {
+      pke[23] = (pke[23] & ~0x000000ff) | (t >> 24);
+      pke[24] = (pke[24] & ~0xffffff00) | (t <<  8);
+    }
+
+    u32 w0[4];
+    u32 w1[4];
+    u32 w2[4];
+    u32 w3[4];
+
+    w0[0] = out[0];
+    w0[1] = out[1];
+    w0[2] = out[2];
+    w0[3] = out[3];
+    w1[0] = out[4];
+    w1[1] = out[5];
+    w1[2] = out[6];
+    w1[3] = out[7];
+    w2[0] = 0;
+    w2[1] = 0;
+    w2[2] = 0;
+    w2[3] = 0;
+    w3[0] = 0;
+    w3[1] = 0;
+    w3[2] = 0;
+    w3[3] = 0;
+
+    u32 keymic[4];
+
+    keymic[0] = 0;
+    keymic[1] = 0;
+    keymic[2] = 0;
+    keymic[3] = 0;
+
+    sha1_hmac_ctx_t ctx1;
+
+    sha1_hmac_init_64 (&ctx1, w0, w1, w2, w3);
+
+    sha1_hmac_update (&ctx1, pke, 100);
+
+    sha1_hmac_final (&ctx1);
+
+    u32 digest[4];
+
+    digest[0] = ctx1.opad.h[0];
+    digest[1] = ctx1.opad.h[1];
+    digest[2] = ctx1.opad.h[2];
+    digest[3] = ctx1.opad.h[3];
+
+    u32 t0[4];
+    u32 t1[4];
+    u32 t2[4];
+    u32 t3[4];
+
+    t0[0] = swap32_S (digest[0]);
+    t0[1] = swap32_S (digest[1]);
+    t0[2] = swap32_S (digest[2]);
+    t0[3] = swap32_S (digest[3]);
+    t1[0] = 0;
+    t1[1] = 0;
+    t1[2] = 0;
+    t1[3] = 0;
+    t2[0] = 0;
+    t2[1] = 0;
+    t2[2] = 0;
+    t2[3] = 0;
+    t3[0] = 0;
+    t3[1] = 0;
+    t3[2] = 0;
+    t3[3] = 0;
+
+    md5_hmac_ctx_t ctx2;
+
+    md5_hmac_init_64 (&ctx2, t0, t1, t2, t3);
+
+    md5_hmac_update_global (&ctx2, wpa->eapol, wpa->eapol_len);
+
+    md5_hmac_final (&ctx2);
+
+    keymic[0] = ctx2.opad.h[0];
+    keymic[1] = ctx2.opad.h[1];
+    keymic[2] = ctx2.opad.h[2];
+    keymic[3] = ctx2.opad.h[3];
+
+    /**
+     * final compare
+     */
+
+    if ((keymic[0] == wpa->keymic[0])
+     && (keymic[1] == wpa->keymic[1])
+     && (keymic[2] == wpa->keymic[2])
+     && (keymic[3] == wpa->keymic[3]))
+    {
+      if (atomic_inc (&hashes_shown[digest_cur]) == 0)
+      {
+        mark_hash (plains_buf, d_return_buf, salt_pos, digests_cnt, digest_pos, digest_cur, gid, 0);
+      }
+    }
+  }
+}
+
+__kernel void m02501_aux2 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global wpa_tmp_t *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const wpa_t *wpa_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u64 gid_max)
+{
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+
+  if (gid >= gid_max) return;
+
+  u32 out[8];
+
+  out[0] = tmps[gid].out[0];
+  out[1] = tmps[gid].out[1];
+  out[2] = tmps[gid].out[2];
+  out[3] = tmps[gid].out[3];
+  out[4] = tmps[gid].out[4];
+  out[5] = tmps[gid].out[5];
+  out[6] = tmps[gid].out[6];
+  out[7] = tmps[gid].out[7];
+
+  const u32 digest_pos = loop_pos;
+
+  const u32 digest_cur = digests_offset + digest_pos;
+
+  __global const wpa_t *wpa = &wpa_bufs[digest_cur];
+
+  u32 pke[32];
+
+  pke[ 0] = wpa->pke[ 0];
+  pke[ 1] = wpa->pke[ 1];
+  pke[ 2] = wpa->pke[ 2];
+  pke[ 3] = wpa->pke[ 3];
+  pke[ 4] = wpa->pke[ 4];
+  pke[ 5] = wpa->pke[ 5];
+  pke[ 6] = wpa->pke[ 6];
+  pke[ 7] = wpa->pke[ 7];
+  pke[ 8] = wpa->pke[ 8];
+  pke[ 9] = wpa->pke[ 9];
+  pke[10] = wpa->pke[10];
+  pke[11] = wpa->pke[11];
+  pke[12] = wpa->pke[12];
+  pke[13] = wpa->pke[13];
+  pke[14] = wpa->pke[14];
+  pke[15] = wpa->pke[15];
+  pke[16] = wpa->pke[16];
+  pke[17] = wpa->pke[17];
+  pke[18] = wpa->pke[18];
+  pke[19] = wpa->pke[19];
+  pke[20] = wpa->pke[20];
+  pke[21] = wpa->pke[21];
+  pke[22] = wpa->pke[22];
+  pke[23] = wpa->pke[23];
+  pke[24] = wpa->pke[24];
+  pke[25] = wpa->pke[25];
+  pke[26] = wpa->pke[26];
+  pke[27] = wpa->pke[27];
+  pke[28] = wpa->pke[28];
+  pke[29] = wpa->pke[29];
+  pke[30] = wpa->pke[30];
+  pke[31] = wpa->pke[31];
+
+  u32 to;
+
+  if (wpa->nonce_compare < 0)
+  {
+    to = pke[15] << 24
+       | pke[16] >>  8;
+  }
+  else
+  {
+    to = pke[23] << 24
+       | pke[24] >>  8;
+  }
+
+  const u32 nonce_error_corrections = wpa->nonce_error_corrections;
+
+  for (u32 nonce_error_correction = 0; nonce_error_correction <= nonce_error_corrections; nonce_error_correction++)
+  {
+    u32 t = to;
+
+    t = swap32_S (t);
+
+    t -= nonce_error_corrections / 2;
+    t += nonce_error_correction;
+
+    t = swap32_S (t);
+
+    if (wpa->nonce_compare < 0)
+    {
+      pke[15] = (pke[15] & ~0x000000ff) | (t >> 24);
+      pke[16] = (pke[16] & ~0xffffff00) | (t <<  8);
+    }
+    else
+    {
+      pke[23] = (pke[23] & ~0x000000ff) | (t >> 24);
+      pke[24] = (pke[24] & ~0xffffff00) | (t <<  8);
+    }
+
+    u32 w0[4];
+    u32 w1[4];
+    u32 w2[4];
+    u32 w3[4];
+
+    w0[0] = out[0];
+    w0[1] = out[1];
+    w0[2] = out[2];
+    w0[3] = out[3];
+    w1[0] = out[4];
+    w1[1] = out[5];
+    w1[2] = out[6];
+    w1[3] = out[7];
+    w2[0] = 0;
+    w2[1] = 0;
+    w2[2] = 0;
+    w2[3] = 0;
+    w3[0] = 0;
+    w3[1] = 0;
+    w3[2] = 0;
+    w3[3] = 0;
+
+    u32 keymic[4];
+
+    keymic[0] = 0;
+    keymic[1] = 0;
+    keymic[2] = 0;
+    keymic[3] = 0;
+
+    sha1_hmac_ctx_t ctx1;
+
+    sha1_hmac_init_64 (&ctx1, w0, w1, w2, w3);
+
+    sha1_hmac_update (&ctx1, pke, 100);
+
+    sha1_hmac_final (&ctx1);
+
+    u32 digest[4];
+
+    digest[0] = ctx1.opad.h[0];
+    digest[1] = ctx1.opad.h[1];
+    digest[2] = ctx1.opad.h[2];
+    digest[3] = ctx1.opad.h[3];
+
+    u32 t0[4];
+    u32 t1[4];
+    u32 t2[4];
+    u32 t3[4];
+
+    t0[0] = digest[0];
+    t0[1] = digest[1];
+    t0[2] = digest[2];
+    t0[3] = digest[3];
+    t1[0] = 0;
+    t1[1] = 0;
+    t1[2] = 0;
+    t1[3] = 0;
+    t2[0] = 0;
+    t2[1] = 0;
+    t2[2] = 0;
+    t2[3] = 0;
+    t3[0] = 0;
+    t3[1] = 0;
+    t3[2] = 0;
+    t3[3] = 0;
+
+    sha1_hmac_ctx_t ctx2;
+
+    sha1_hmac_init_64 (&ctx2, t0, t1, t2, t3);
+
+    sha1_hmac_update_global (&ctx2, wpa->eapol, wpa->eapol_len);
+
+    sha1_hmac_final (&ctx2);
+
+    keymic[0] = ctx2.opad.h[0];
+    keymic[1] = ctx2.opad.h[1];
+    keymic[2] = ctx2.opad.h[2];
+    keymic[3] = ctx2.opad.h[3];
+
+    /**
+     * final compare
+     */
+
+    if ((keymic[0] == wpa->keymic[0])
+     && (keymic[1] == wpa->keymic[1])
+     && (keymic[2] == wpa->keymic[2])
+     && (keymic[3] == wpa->keymic[3]))
+    {
+      if (atomic_inc (&hashes_shown[digest_cur]) == 0)
+      {
+        mark_hash (plains_buf, d_return_buf, salt_pos, digests_cnt, digest_pos, digest_cur, gid, 0);
+      }
+    }
+  }
+}
+
+__kernel void m02501_aux3 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global wpa_tmp_t *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const wpa_t *wpa_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u64 gid_max)
 {
   const u64 gid = get_global_id (0);
   const u64 lid = get_local_id (0);
@@ -294,112 +683,6 @@ __kernel void m02501_comp (__global pw_t *pws, __global const kernel_rule_t *rul
     keymic[2] = 0;
     keymic[3] = 0;
 
-    if (wpa->keyver == 1)
-    {
-      sha1_hmac_ctx_t ctx1;
-
-      sha1_hmac_init_64 (&ctx1, w0, w1, w2, w3);
-
-      sha1_hmac_update (&ctx1, pke, 100);
-
-      sha1_hmac_final (&ctx1);
-
-      u32 digest[4];
-
-      digest[0] = ctx1.opad.h[0];
-      digest[1] = ctx1.opad.h[1];
-      digest[2] = ctx1.opad.h[2];
-      digest[3] = ctx1.opad.h[3];
-
-      u32 t0[4];
-      u32 t1[4];
-      u32 t2[4];
-      u32 t3[4];
-
-      t0[0] = swap32_S (digest[0]);
-      t0[1] = swap32_S (digest[1]);
-      t0[2] = swap32_S (digest[2]);
-      t0[3] = swap32_S (digest[3]);
-      t1[0] = 0;
-      t1[1] = 0;
-      t1[2] = 0;
-      t1[3] = 0;
-      t2[0] = 0;
-      t2[1] = 0;
-      t2[2] = 0;
-      t2[3] = 0;
-      t3[0] = 0;
-      t3[1] = 0;
-      t3[2] = 0;
-      t3[3] = 0;
-
-      md5_hmac_ctx_t ctx2;
-
-      md5_hmac_init_64 (&ctx2, t0, t1, t2, t3);
-
-      md5_hmac_update_global (&ctx2, wpa->eapol, wpa->eapol_len);
-
-      md5_hmac_final (&ctx2);
-
-      keymic[0] = ctx2.opad.h[0];
-      keymic[1] = ctx2.opad.h[1];
-      keymic[2] = ctx2.opad.h[2];
-      keymic[3] = ctx2.opad.h[3];
-    }
-    else if (wpa->keyver == 2)
-    {
-      sha1_hmac_ctx_t ctx1;
-
-      sha1_hmac_init_64 (&ctx1, w0, w1, w2, w3);
-
-      sha1_hmac_update (&ctx1, pke, 100);
-
-      sha1_hmac_final (&ctx1);
-
-      u32 digest[4];
-
-      digest[0] = ctx1.opad.h[0];
-      digest[1] = ctx1.opad.h[1];
-      digest[2] = ctx1.opad.h[2];
-      digest[3] = ctx1.opad.h[3];
-
-      u32 t0[4];
-      u32 t1[4];
-      u32 t2[4];
-      u32 t3[4];
-
-      t0[0] = digest[0];
-      t0[1] = digest[1];
-      t0[2] = digest[2];
-      t0[3] = digest[3];
-      t1[0] = 0;
-      t1[1] = 0;
-      t1[2] = 0;
-      t1[3] = 0;
-      t2[0] = 0;
-      t2[1] = 0;
-      t2[2] = 0;
-      t2[3] = 0;
-      t3[0] = 0;
-      t3[1] = 0;
-      t3[2] = 0;
-      t3[3] = 0;
-
-      sha1_hmac_ctx_t ctx2;
-
-      sha1_hmac_init_64 (&ctx2, t0, t1, t2, t3);
-
-      sha1_hmac_update_global (&ctx2, wpa->eapol, wpa->eapol_len);
-
-      sha1_hmac_final (&ctx2);
-
-      keymic[0] = ctx2.opad.h[0];
-      keymic[1] = ctx2.opad.h[1];
-      keymic[2] = ctx2.opad.h[2];
-      keymic[3] = ctx2.opad.h[3];
-    }
-    else if (wpa->keyver == 3)
-    {
     sha256_hmac_ctx_t ctx1;
 
     sha256_hmac_init_64 (&ctx1, w0, w1, w2, w3);
@@ -480,262 +763,6 @@ __kernel void m02501_comp (__global pw_t *pws, __global const kernel_rule_t *rul
     m[3] ^= iv[3];
 
     aes128_encrypt (ks, m, keymic, s_te0, s_te1, s_te2, s_te3, s_te4);
-    }
-
-    /**
-     * final compare
-     */
-
-    if ((keymic[0] == wpa->keymic[0])
-     && (keymic[1] == wpa->keymic[1])
-     && (keymic[2] == wpa->keymic[2])
-     && (keymic[3] == wpa->keymic[3]))
-    {
-      if (atomic_inc (&hashes_shown[digest_cur]) == 0)
-      {
-        mark_hash (plains_buf, d_return_buf, salt_pos, digests_cnt, digest_pos, digest_cur, gid, 0);
-      }
-    }
-  }
-
-  // the same code again, but with BE order for the t++
-
-  for (u32 nonce_error_correction = 0; nonce_error_correction <= nonce_error_corrections; nonce_error_correction++)
-  {
-    u32 t = to;
-
-    t -= nonce_error_corrections / 2;
-    t += nonce_error_correction;
-
-    if (t == to) continue; // we already had this checked in the LE loop
-
-    if (wpa->nonce_compare < 0)
-    {
-      pke[15] = (pke[15] & ~0x000000ff) | (t >> 24);
-      pke[16] = (pke[16] & ~0xffffff00) | (t <<  8);
-    }
-    else
-    {
-      pke[23] = (pke[23] & ~0x000000ff) | (t >> 24);
-      pke[24] = (pke[24] & ~0xffffff00) | (t <<  8);
-    }
-
-    u32 w0[4];
-    u32 w1[4];
-    u32 w2[4];
-    u32 w3[4];
-
-    w0[0] = out[0];
-    w0[1] = out[1];
-    w0[2] = out[2];
-    w0[3] = out[3];
-    w1[0] = out[4];
-    w1[1] = out[5];
-    w1[2] = out[6];
-    w1[3] = out[7];
-    w2[0] = 0;
-    w2[1] = 0;
-    w2[2] = 0;
-    w2[3] = 0;
-    w3[0] = 0;
-    w3[1] = 0;
-    w3[2] = 0;
-    w3[3] = 0;
-
-    u32 keymic[4];
-
-    keymic[0] = 0;
-    keymic[1] = 0;
-    keymic[2] = 0;
-    keymic[3] = 0;
-
-    if (wpa->keyver == 1)
-    {
-      sha1_hmac_ctx_t ctx1;
-
-      sha1_hmac_init_64 (&ctx1, w0, w1, w2, w3);
-
-      sha1_hmac_update (&ctx1, pke, 100);
-
-      sha1_hmac_final (&ctx1);
-
-      u32 digest[4];
-
-      digest[0] = ctx1.opad.h[0];
-      digest[1] = ctx1.opad.h[1];
-      digest[2] = ctx1.opad.h[2];
-      digest[3] = ctx1.opad.h[3];
-
-      u32 t0[4];
-      u32 t1[4];
-      u32 t2[4];
-      u32 t3[4];
-
-      t0[0] = swap32_S (digest[0]);
-      t0[1] = swap32_S (digest[1]);
-      t0[2] = swap32_S (digest[2]);
-      t0[3] = swap32_S (digest[3]);
-      t1[0] = 0;
-      t1[1] = 0;
-      t1[2] = 0;
-      t1[3] = 0;
-      t2[0] = 0;
-      t2[1] = 0;
-      t2[2] = 0;
-      t2[3] = 0;
-      t3[0] = 0;
-      t3[1] = 0;
-      t3[2] = 0;
-      t3[3] = 0;
-
-      md5_hmac_ctx_t ctx2;
-
-      md5_hmac_init_64 (&ctx2, t0, t1, t2, t3);
-
-      md5_hmac_update_global (&ctx2, wpa->eapol, wpa->eapol_len);
-
-      md5_hmac_final (&ctx2);
-
-      keymic[0] = ctx2.opad.h[0];
-      keymic[1] = ctx2.opad.h[1];
-      keymic[2] = ctx2.opad.h[2];
-      keymic[3] = ctx2.opad.h[3];
-    }
-    else if (wpa->keyver == 2)
-    {
-      sha1_hmac_ctx_t ctx1;
-
-      sha1_hmac_init_64 (&ctx1, w0, w1, w2, w3);
-
-      sha1_hmac_update (&ctx1, pke, 100);
-
-      sha1_hmac_final (&ctx1);
-
-      u32 digest[4];
-
-      digest[0] = ctx1.opad.h[0];
-      digest[1] = ctx1.opad.h[1];
-      digest[2] = ctx1.opad.h[2];
-      digest[3] = ctx1.opad.h[3];
-
-      u32 t0[4];
-      u32 t1[4];
-      u32 t2[4];
-      u32 t3[4];
-
-      t0[0] = digest[0];
-      t0[1] = digest[1];
-      t0[2] = digest[2];
-      t0[3] = digest[3];
-      t1[0] = 0;
-      t1[1] = 0;
-      t1[2] = 0;
-      t1[3] = 0;
-      t2[0] = 0;
-      t2[1] = 0;
-      t2[2] = 0;
-      t2[3] = 0;
-      t3[0] = 0;
-      t3[1] = 0;
-      t3[2] = 0;
-      t3[3] = 0;
-
-      sha1_hmac_ctx_t ctx2;
-
-      sha1_hmac_init_64 (&ctx2, t0, t1, t2, t3);
-
-      sha1_hmac_update_global (&ctx2, wpa->eapol, wpa->eapol_len);
-
-      sha1_hmac_final (&ctx2);
-
-      keymic[0] = ctx2.opad.h[0];
-      keymic[1] = ctx2.opad.h[1];
-      keymic[2] = ctx2.opad.h[2];
-      keymic[3] = ctx2.opad.h[3];
-    }
-    else if (wpa->keyver == 3)
-    {
-      sha256_hmac_ctx_t ctx1;
-
-      sha256_hmac_init_64 (&ctx1, w0, w1, w2, w3);
-
-      sha256_hmac_update (&ctx1, pke, 102);
-
-      sha256_hmac_final (&ctx1);
-
-      u32 digest[4];
-
-      digest[0] = swap32_S (ctx1.opad.h[0]);
-      digest[1] = swap32_S (ctx1.opad.h[1]);
-      digest[2] = swap32_S (ctx1.opad.h[2]);
-      digest[3] = swap32_S (ctx1.opad.h[3]);
-
-      // AES CMAC
-
-      u32 ks[44];
-
-      aes128_set_encrypt_key (ks, digest, s_te0, s_te1, s_te2, s_te3, s_te4);
-
-      u32 m[4];
-
-      m[0] = 0;
-      m[1] = 0;
-      m[2] = 0;
-      m[3] = 0;
-
-      u32 iv[4];
-
-      iv[0] = 0;
-      iv[1] = 0;
-      iv[2] = 0;
-      iv[3] = 0;
-
-      int eapol_left;
-      int eapol_idx;
-
-      for (eapol_left = wpa->eapol_len, eapol_idx = 0; eapol_left > 16; eapol_left -= 16, eapol_idx += 4)
-      {
-        m[0] = wpa->eapol[eapol_idx + 0] ^ iv[0];
-        m[1] = wpa->eapol[eapol_idx + 1] ^ iv[1];
-        m[2] = wpa->eapol[eapol_idx + 2] ^ iv[2];
-        m[3] = wpa->eapol[eapol_idx + 3] ^ iv[3];
-
-        aes128_encrypt (ks, m, iv, s_te0, s_te1, s_te2, s_te3, s_te4);
-      }
-
-      m[0] = wpa->eapol[eapol_idx + 0];
-      m[1] = wpa->eapol[eapol_idx + 1];
-      m[2] = wpa->eapol[eapol_idx + 2];
-      m[3] = wpa->eapol[eapol_idx + 3];
-
-      u32 k[4];
-
-      k[0] = 0;
-      k[1] = 0;
-      k[2] = 0;
-      k[3] = 0;
-
-      aes128_encrypt (ks, k, k, s_te0, s_te1, s_te2, s_te3, s_te4);
-
-      make_kn (k);
-
-      if (eapol_left < 16)
-      {
-        make_kn (k);
-      }
-
-      m[0] ^= k[0];
-      m[1] ^= k[1];
-      m[2] ^= k[2];
-      m[3] ^= k[3];
-
-      m[0] ^= iv[0];
-      m[1] ^= iv[1];
-      m[2] ^= iv[2];
-      m[3] ^= iv[3];
-
-      aes128_encrypt (ks, m, keymic, s_te0, s_te1, s_te2, s_te3, s_te4);
-    }
 
     /**
      * final compare

docs/changes.txt

@@ -64,6 +64,7 @@
 - OpenCL Kernels: Replace variables from uXX to uXXa if used in __constant space
 - OpenCL Kernels: Use a special kernel to initialize the password buffer used during autotune measurements, to reduce startup time
 - OpenCL Kernels: Refactored kernel thread management from native to maximum per kernel
+- OpenCL Kernels: Use three separate comparison kernels (depending on keyver) for WPA instead of one
 - OpenCL Runtime: Add current timestamp to OpenCL kernel source in order to force OpenCL JiT compiler to recompile and not use the cache
 - OpenCL Runtime: Enforce to use OpenCL version 1.2 to restrain OpenCL runtimes to make use of the __generic address space qualifier
 - OpenCL Runtime: Updated rocm detection

include/types.h

@@ -227,7 +227,11 @@ typedef enum kern_run
   KERN_RUN_3      = 3000,
   KERN_RUN_4      = 4000,
   KERN_RUN_INIT2  = 5000,
-  KERN_RUN_LOOP2 = 6000
+  KERN_RUN_LOOP2  = 6000,
+  KERN_RUN_AUX1   = 7001,
+  KERN_RUN_AUX2   = 7002,
+  KERN_RUN_AUX3   = 7003,
+  KERN_RUN_AUX4   = 7004,
 
 } kern_run_t;
 
@@ -376,7 +380,11 @@ typedef enum opts_type
   OPTS_TYPE_HOOK23            = (1ULL << 31),
   OPTS_TYPE_INIT2             = (1ULL << 32),
   OPTS_TYPE_LOOP2             = (1ULL << 33),
-  OPTS_TYPE_BINARY_HASHFILE   = (1ULL << 34),
+  OPTS_TYPE_AUX1              = (1ULL << 34),
+  OPTS_TYPE_AUX2              = (1ULL << 35),
+  OPTS_TYPE_AUX3              = (1ULL << 36),
+  OPTS_TYPE_AUX4              = (1ULL << 37),
+  OPTS_TYPE_BINARY_HASHFILE   = (1ULL << 38),
 
 } opts_type_t;
 
@@ -938,6 +946,10 @@ typedef struct hc_device_param
   u32     kernel_wgs_memset;
   u32     kernel_wgs_atinit;
   u32     kernel_wgs_decompress;
+  u32     kernel_wgs_aux1;
+  u32     kernel_wgs_aux2;
+  u32     kernel_wgs_aux3;
+  u32     kernel_wgs_aux4;
 
   u32     kernel_preferred_wgs_multiple1;
   u32     kernel_preferred_wgs_multiple12;
@@ -955,6 +967,10 @@ typedef struct hc_device_param
   u32     kernel_preferred_wgs_multiple_memset;
   u32     kernel_preferred_wgs_multiple_atinit;
   u32     kernel_preferred_wgs_multiple_decompress;
+  u32     kernel_preferred_wgs_multiple_aux1;
+  u32     kernel_preferred_wgs_multiple_aux2;
+  u32     kernel_preferred_wgs_multiple_aux3;
+  u32     kernel_preferred_wgs_multiple_aux4;
 
   u64     kernel_local_mem_size1;
   u64     kernel_local_mem_size12;
@@ -972,6 +988,10 @@ typedef struct hc_device_param
   u64     kernel_local_mem_size_memset;
   u64     kernel_local_mem_size_atinit;
   u64     kernel_local_mem_size_decompress;
+  u64     kernel_local_mem_size_aux1;
+  u64     kernel_local_mem_size_aux2;
+  u64     kernel_local_mem_size_aux3;
+  u64     kernel_local_mem_size_aux4;
 
   u32     kernel_threads;
 
@@ -1042,6 +1062,10 @@ typedef struct hc_device_param
   double  exec_us_prev4[EXPECTED_ITERATIONS];
   double  exec_us_prev_init2[EXPECTED_ITERATIONS];
   double  exec_us_prev_loop2[EXPECTED_ITERATIONS];
+  double  exec_us_prev_aux1[EXPECTED_ITERATIONS];
+  double  exec_us_prev_aux2[EXPECTED_ITERATIONS];
+  double  exec_us_prev_aux3[EXPECTED_ITERATIONS];
+  double  exec_us_prev_aux4[EXPECTED_ITERATIONS];
 
   // this is "current" speed
 
@@ -1085,6 +1109,10 @@ typedef struct hc_device_param
   cl_kernel  kernel_memset;
   cl_kernel  kernel_atinit;
   cl_kernel  kernel_decompress;
+  cl_kernel  kernel_aux1;
+  cl_kernel  kernel_aux2;
+  cl_kernel  kernel_aux3;
+  cl_kernel  kernel_aux4;
 
   cl_context context;
 

src/autotune.c

@@ -248,6 +248,10 @@ static int autotune (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
   memset (device_param->exec_us_prev4,      0, EXPECTED_ITERATIONS * sizeof (double));
   memset (device_param->exec_us_prev_init2, 0, EXPECTED_ITERATIONS * sizeof (double));
   memset (device_param->exec_us_prev_loop2, 0, EXPECTED_ITERATIONS * sizeof (double));
+  memset (device_param->exec_us_prev_aux1,  0, EXPECTED_ITERATIONS * sizeof (double));
+  memset (device_param->exec_us_prev_aux2,  0, EXPECTED_ITERATIONS * sizeof (double));
+  memset (device_param->exec_us_prev_aux3,  0, EXPECTED_ITERATIONS * sizeof (double));
+  memset (device_param->exec_us_prev_aux4,  0, EXPECTED_ITERATIONS * sizeof (double));
 
   // store
 

src/interface.c

@@ -22371,6 +22371,9 @@ int hashconfig_init (hashcat_ctx_t *hashcat_ctx)
                  hashconfig->salt_type      = SALT_TYPE_EMBEDDED;
                  hashconfig->attack_exec    = ATTACK_EXEC_OUTSIDE_KERNEL;
                  hashconfig->opts_type      = OPTS_TYPE_PT_GENERATE_LE
+                                            | OPTS_TYPE_AUX1
+                                            | OPTS_TYPE_AUX2
+                                            | OPTS_TYPE_AUX3
                                             | OPTS_TYPE_BINARY_HASHFILE;
                  hashconfig->kern_type      = KERN_TYPE_WPA;
                  hashconfig->dgst_size      = DGST_SIZE_4_4;
@@ -22389,6 +22392,9 @@ int hashconfig_init (hashcat_ctx_t *hashcat_ctx)
                  hashconfig->salt_type      = SALT_TYPE_EMBEDDED;
                  hashconfig->attack_exec    = ATTACK_EXEC_OUTSIDE_KERNEL;
                  hashconfig->opts_type      = OPTS_TYPE_PT_GENERATE_LE
+                                            | OPTS_TYPE_AUX1
+                                            | OPTS_TYPE_AUX2
+                                            | OPTS_TYPE_AUX3
                                             | OPTS_TYPE_BINARY_HASHFILE;
                  hashconfig->kern_type      = KERN_TYPE_WPAPMK;
                  hashconfig->dgst_size      = DGST_SIZE_4_4;

src/opencl.c

@@ -1401,23 +1401,50 @@ int choose_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param,
 
     if (run_comp == true)
     {
-      u32 loops_cnt = 1;
-
       if ((hashconfig->hash_mode == 2500) || (hashconfig->hash_mode == 2501))
       {
-        loops_cnt = hashes->salts_buf[salt_pos].digests_cnt;
+        const u32 loops_cnt = hashes->salts_buf[salt_pos].digests_cnt;
-      }
 
         for (u32 loops_pos = 0; loops_pos < loops_cnt; loops_pos++)
         {
           device_param->kernel_params_buf32[28] = loops_pos;
           device_param->kernel_params_buf32[29] = loops_cnt;
 
+          const u32 digests_offset = hashes->salts_buf[salt_pos].digests_offset;
+
+          wpa_t *wpas = (wpa_t *) hashes->esalts_buf;
+
+          wpa_t *wpa = &wpas[digests_offset + loops_pos];
+
+printf ("%u\n", wpa->keyver);
+
+          if (wpa->keyver == 1)
+          {
+            CL_rc = run_kernel (hashcat_ctx, device_param, KERN_RUN_AUX1, pws_cnt, false, 0);
+
+            if (CL_rc == -1) return -1;
+          }
+          else if (wpa->keyver == 2)
+          {
+            CL_rc = run_kernel (hashcat_ctx, device_param, KERN_RUN_AUX2, pws_cnt, false, 0);
+
+            if (CL_rc == -1) return -1;
+          }
+          else if (wpa->keyver == 3)
+          {
+            CL_rc = run_kernel (hashcat_ctx, device_param, KERN_RUN_AUX3, pws_cnt, false, 0);
+
+            if (CL_rc == -1) return -1;
+          }
+
+          if (status_ctx->run_thread_level2 == false) break;
+        }
+      }
+      else
+      {
         CL_rc = run_kernel (hashcat_ctx, device_param, KERN_RUN_3, pws_cnt, false, 0);
 
         if (CL_rc == -1) return -1;
-
-        if (status_ctx->run_thread_level2 == false) break;
       }
     }
   }
@@ -1523,6 +1550,22 @@ int run_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, con
       kernel          = device_param->kernel_loop2;
       kernel_threads  = device_param->kernel_wgs_loop2;
       break;
+    case KERN_RUN_AUX1:
+      kernel          = device_param->kernel_aux1;
+      kernel_threads  = device_param->kernel_wgs_aux1;
+      break;
+    case KERN_RUN_AUX2:
+      kernel          = device_param->kernel_aux2;
+      kernel_threads  = device_param->kernel_wgs_aux2;
+      break;
+    case KERN_RUN_AUX3:
+      kernel          = device_param->kernel_aux3;
+      kernel_threads  = device_param->kernel_wgs_aux3;
+      break;
+    case KERN_RUN_AUX4:
+      kernel          = device_param->kernel_aux4;
+      kernel_threads  = device_param->kernel_wgs_aux4;
+      break;
     default:
       event_log_error (hashcat_ctx, "Invalid kernel specified.");
       return -1;
@@ -1620,6 +1663,10 @@ int run_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, con
           case KERN_RUN_4:      if (device_param->exec_us_prev4[iteration]      > 0) usleep ((useconds_t)(device_param->exec_us_prev4[iteration]      * device_param->nvidia_spin_damp)); break;
           case KERN_RUN_INIT2:  if (device_param->exec_us_prev_init2[iteration] > 0) usleep ((useconds_t)(device_param->exec_us_prev_init2[iteration] * device_param->nvidia_spin_damp)); break;
           case KERN_RUN_LOOP2:  if (device_param->exec_us_prev_loop2[iteration] > 0) usleep ((useconds_t)(device_param->exec_us_prev_loop2[iteration] * device_param->nvidia_spin_damp)); break;
+          case KERN_RUN_AUX1:   if (device_param->exec_us_prev_aux1[iteration]  > 0) usleep ((useconds_t)(device_param->exec_us_prev_aux1[iteration]  * device_param->nvidia_spin_damp)); break;
+          case KERN_RUN_AUX2:   if (device_param->exec_us_prev_aux2[iteration]  > 0) usleep ((useconds_t)(device_param->exec_us_prev_aux2[iteration]  * device_param->nvidia_spin_damp)); break;
+          case KERN_RUN_AUX3:   if (device_param->exec_us_prev_aux3[iteration]  > 0) usleep ((useconds_t)(device_param->exec_us_prev_aux3[iteration]  * device_param->nvidia_spin_damp)); break;
+          case KERN_RUN_AUX4:   if (device_param->exec_us_prev_aux4[iteration]  > 0) usleep ((useconds_t)(device_param->exec_us_prev_aux4[iteration]  * device_param->nvidia_spin_damp)); break;
         }
       }
     }
@@ -1649,6 +1696,10 @@ int run_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, con
         case KERN_RUN_4:      device_param->exec_us_prev4[iteration]      = exec_us; break;
         case KERN_RUN_INIT2:  device_param->exec_us_prev_init2[iteration] = exec_us; break;
         case KERN_RUN_LOOP2:  device_param->exec_us_prev_loop2[iteration] = exec_us; break;
+        case KERN_RUN_AUX1:   device_param->exec_us_prev_aux1[iteration]  = exec_us; break;
+        case KERN_RUN_AUX2:   device_param->exec_us_prev_aux2[iteration]  = exec_us; break;
+        case KERN_RUN_AUX3:   device_param->exec_us_prev_aux3[iteration]  = exec_us; break;
+        case KERN_RUN_AUX4:   device_param->exec_us_prev_aux4[iteration]  = exec_us; break;
       }
     }
   }
@@ -5624,6 +5675,98 @@ int opencl_session_begin (hashcat_ctx_t *hashcat_ctx)
 
         if (CL_rc == -1) return -1;
       }
+
+      // aux1
+
+      if (hashconfig->opts_type & OPTS_TYPE_AUX1)
+      {
+        snprintf (kernel_name, sizeof (kernel_name) - 1, "m%05u_aux1", hashconfig->kern_type);
+
+        CL_rc = hc_clCreateKernel (hashcat_ctx, device_param->program, kernel_name, &device_param->kernel_aux1);
+
+        if (CL_rc == -1) return -1;
+
+        CL_rc = get_kernel_wgs (hashcat_ctx, device_param, device_param->kernel_aux1, &device_param->kernel_wgs_aux1);
+
+        if (CL_rc == -1) return -1;
+
+        CL_rc = get_kernel_local_mem_size (hashcat_ctx, device_param, device_param->kernel_aux1, &device_param->kernel_local_mem_size_aux1);
+
+        if (CL_rc == -1) return -1;
+
+        CL_rc = get_kernel_preferred_wgs_multiple (hashcat_ctx, device_param, device_param->kernel_aux1, &device_param->kernel_preferred_wgs_multiple_aux1);
+
+        if (CL_rc == -1) return -1;
+      }
+
+      // aux2
+
+      if (hashconfig->opts_type & OPTS_TYPE_AUX2)
+      {
+        snprintf (kernel_name, sizeof (kernel_name) - 1, "m%05u_aux2", hashconfig->kern_type);
+
+        CL_rc = hc_clCreateKernel (hashcat_ctx, device_param->program, kernel_name, &device_param->kernel_aux2);
+
+        if (CL_rc == -1) return -1;
+
+        CL_rc = get_kernel_wgs (hashcat_ctx, device_param, device_param->kernel_aux2, &device_param->kernel_wgs_aux2);
+
+        if (CL_rc == -1) return -1;
+
+        CL_rc = get_kernel_local_mem_size (hashcat_ctx, device_param, device_param->kernel_aux2, &device_param->kernel_local_mem_size_aux2);
+
+        if (CL_rc == -1) return -1;
+
+        CL_rc = get_kernel_preferred_wgs_multiple (hashcat_ctx, device_param, device_param->kernel_aux2, &device_param->kernel_preferred_wgs_multiple_aux2);
+
+        if (CL_rc == -1) return -1;
+      }
+
+      // aux3
+
+      if (hashconfig->opts_type & OPTS_TYPE_AUX3)
+      {
+        snprintf (kernel_name, sizeof (kernel_name) - 1, "m%05u_aux3", hashconfig->kern_type);
+
+        CL_rc = hc_clCreateKernel (hashcat_ctx, device_param->program, kernel_name, &device_param->kernel_aux3);
+
+        if (CL_rc == -1) return -1;
+
+        CL_rc = get_kernel_wgs (hashcat_ctx, device_param, device_param->kernel_aux3, &device_param->kernel_wgs_aux3);
+
+        if (CL_rc == -1) return -1;
+
+        CL_rc = get_kernel_local_mem_size (hashcat_ctx, device_param, device_param->kernel_aux3, &device_param->kernel_local_mem_size_aux3);
+
+        if (CL_rc == -1) return -1;
+
+        CL_rc = get_kernel_preferred_wgs_multiple (hashcat_ctx, device_param, device_param->kernel_aux3, &device_param->kernel_preferred_wgs_multiple_aux3);
+
+        if (CL_rc == -1) return -1;
+      }
+
+      // aux4
+
+      if (hashconfig->opts_type & OPTS_TYPE_AUX4)
+      {
+        snprintf (kernel_name, sizeof (kernel_name) - 1, "m%05u_aux4", hashconfig->kern_type);
+
+        CL_rc = hc_clCreateKernel (hashcat_ctx, device_param->program, kernel_name, &device_param->kernel_aux4);
+
+        if (CL_rc == -1) return -1;
+
+        CL_rc = get_kernel_wgs (hashcat_ctx, device_param, device_param->kernel_aux4, &device_param->kernel_wgs_aux4);
+
+        if (CL_rc == -1) return -1;
+
+        CL_rc = get_kernel_local_mem_size (hashcat_ctx, device_param, device_param->kernel_aux4, &device_param->kernel_local_mem_size_aux4);
+
+        if (CL_rc == -1) return -1;
+
+        CL_rc = get_kernel_preferred_wgs_multiple (hashcat_ctx, device_param, device_param->kernel_aux4, &device_param->kernel_preferred_wgs_multiple_aux4);
+
+        if (CL_rc == -1) return -1;
+      }
     }
 
     // GPU memset
@@ -6224,6 +6367,10 @@ void opencl_session_destroy (hashcat_ctx_t *hashcat_ctx)
     if (device_param->kernel_memset)    hc_clReleaseKernel (hashcat_ctx, device_param->kernel_memset);
     if (device_param->kernel_atinit)    hc_clReleaseKernel (hashcat_ctx, device_param->kernel_atinit);
     if (device_param->kernel_decompress)hc_clReleaseKernel (hashcat_ctx, device_param->kernel_decompress);
+    if (device_param->kernel_aux1)      hc_clReleaseKernel (hashcat_ctx, device_param->kernel_aux1);
+    if (device_param->kernel_aux2)      hc_clReleaseKernel (hashcat_ctx, device_param->kernel_aux2);
+    if (device_param->kernel_aux3)      hc_clReleaseKernel (hashcat_ctx, device_param->kernel_aux3);
+    if (device_param->kernel_aux4)      hc_clReleaseKernel (hashcat_ctx, device_param->kernel_aux4);
 
     if (device_param->program)          hc_clReleaseProgram (hashcat_ctx, device_param->program);
     if (device_param->program_mp)       hc_clReleaseProgram (hashcat_ctx, device_param->program_mp);

src/selftest.c

@@ -412,12 +412,18 @@ static int selftest (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
     {
       device_param->kernel_params_buf32[28] = 0;
       device_param->kernel_params_buf32[29] = 1;
-    }
 
+      CL_rc = run_kernel (hashcat_ctx, device_param, KERN_RUN_AUX1, 1, false, 0);
+
+      if (CL_rc == -1) return -1;
+    }
+    else
+    {
       CL_rc = run_kernel (hashcat_ctx, device_param, KERN_RUN_3, 1, false, 0);
 
       if (CL_rc == -1) return -1;
     }
+  }
 
   // check : check if cracked