X-Git-Url: https://www.flypig.org.uk/git/?a=blobdiff_plain;f=OpenCL%2Fm05000_a1.cl;h=a6192843fad01e6c8252702f466aaeef001b463b;hb=161a6eb4bc643d8e636e96eda613f5137d30da59;hp=37d2c81dd40834eff7af53a5a4796ea1831412f1;hpb=b409e5e9e162927b25af88e672326b246f17ec5f;p=hashcat.git

diff --git a/OpenCL/m05000_a1.cl b/OpenCL/m05000_a1.cl
index 37d2c81..a619284 100644
--- a/OpenCL/m05000_a1.cl
+++ b/OpenCL/m05000_a1.cl
@@ -7,18 +7,12 @@
 
 #define NEW_SIMD_CODE
 
-#include "include/constants.h"
-#include "include/kernel_vendor.h"
-
-#define DGST_R0 2
-#define DGST_R1 3
-#define DGST_R2 4
-#define DGST_R3 5
-
-#include "include/kernel_functions.c"
-#include "OpenCL/types_ocl.c"
-#include "OpenCL/common.c"
-#include "OpenCL/simd.c"
+#include "inc_vendor.cl"
+#include "inc_hash_constants.h"
+#include "inc_hash_functions.cl"
+#include "inc_types.cl"
+#include "inc_common.cl"
+#include "inc_simd.cl"
 
 __constant u64 keccakf_rndc[24] =
 {
@@ -49,8 +43,8 @@ __constant u64 keccakf_rndc[24] =
 
 #define Rho_Pi(s)               \
 {                               \
-  u32 j = keccakf_piln[s];     \
-  u32 k = keccakf_rotc[s];     \
+  u32 j = keccakf_piln[s];      \
+  u32 k = keccakf_rotc[s];      \
   bc0 = st[j];                  \
   st[j] = rotl64 (t, k);        \
   t = bc0;                      \
@@ -70,7 +64,7 @@ __constant u64 keccakf_rndc[24] =
   st[4 + s] ^= ~bc0 & bc1;      \
 }
 
-__kernel void m05000_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_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 combs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void m05000_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_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 u32 gid_max)
 {
   /**
    * modifier
@@ -79,7 +73,29 @@ __kernel void m05000_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
   const u32 lid = get_local_id (0);
 
   /**
-   * const
+   * base
+   */
+
+  const u32 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  u32 pw_buf0[4];
+  u32 pw_buf1[4];
+
+  pw_buf0[0] = pws[gid].i[0];
+  pw_buf0[1] = pws[gid].i[1];
+  pw_buf0[2] = pws[gid].i[2];
+  pw_buf0[3] = pws[gid].i[3];
+  pw_buf1[0] = pws[gid].i[4];
+  pw_buf1[1] = pws[gid].i[5];
+  pw_buf1[2] = pws[gid].i[6];
+  pw_buf1[3] = pws[gid].i[7];
+
+  const u32 pw_l_len = pws[gid].pw_len;
+
+  /**
+   * constants
    */
 
   const u8 keccakf_rotc[24] =
@@ -94,28 +110,6 @@ __kernel void m05000_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
     15, 23, 19, 13, 12,  2, 20, 14, 22,  9,  6,  1
   };
 
-  /**
-   * base
-   */
-
-  const u32 gid = get_global_id (0);
-
-  if (gid >= gid_max) return;
-
-  u32 pws0[4] = { 0 };
-  u32 pws1[4] = { 0 };
-
-  pws0[0] = pws[gid].i[0];
-  pws0[1] = pws[gid].i[1];
-  pws0[2] = pws[gid].i[2];
-  pws0[3] = pws[gid].i[3];
-  pws1[0] = pws[gid].i[4];
-  pws1[1] = pws[gid].i[5];
-  pws1[2] = pws[gid].i[6];
-  pws1[3] = pws[gid].i[7];
-
-  const u32 pw_l_len = pws[gid].pw_len;
-
   /**
    * 0x80 keccak, very special
    */
@@ -130,12 +124,30 @@ __kernel void m05000_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
    * loop
    */
 
-  for (u32 il_pos = 0; il_pos < combs_cnt; il_pos += VECT_SIZE)
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
   {
     const u32x pw_r_len = pwlenx_create_combt (combs_buf, il_pos);
 
     const u32x pw_len = pw_l_len + pw_r_len;
 
+    /**
+     * concat password candidate
+     */
+
+    u32x wordl0[4] = { 0 };
+    u32x wordl1[4] = { 0 };
+    u32x wordl2[4] = { 0 };
+    u32x wordl3[4] = { 0 };
+
+    wordl0[0] = pw_buf0[0];
+    wordl0[1] = pw_buf0[1];
+    wordl0[2] = pw_buf0[2];
+    wordl0[3] = pw_buf0[3];
+    wordl1[0] = pw_buf1[0];
+    wordl1[1] = pw_buf1[1];
+    wordl1[2] = pw_buf1[2];
+    wordl1[3] = pw_buf1[3];
+
     u32x wordr0[4] = { 0 };
     u32x wordr1[4] = { 0 };
     u32x wordr2[4] = { 0 };
@@ -152,49 +164,49 @@ __kernel void m05000_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
 
     if (combs_mode == COMBINATOR_MODE_BASE_LEFT)
     {
-      append_0x01_2x4 (wordr0, wordr1, pw_r_len);
-
-      switch_buffer_by_offset_le (wordr0, wordr1, wordr2, wordr3, pw_l_len);
+      switch_buffer_by_offset_le_VV (wordr0, wordr1, wordr2, wordr3, pw_l_len);
+    }
+    else
+    {
+      switch_buffer_by_offset_le_VV (wordl0, wordl1, wordl2, wordl3, pw_r_len);
     }
 
     u32x w0[4];
+    u32x w1[4];
+    u32x w2[4];
+    u32x w3[4];
 
     w0[0] = wordl0[0] | wordr0[0];
     w0[1] = wordl0[1] | wordr0[1];
     w0[2] = wordl0[2] | wordr0[2];
     w0[3] = wordl0[3] | wordr0[3];
-
-    u32x w1[4];
-
     w1[0] = wordl1[0] | wordr1[0];
     w1[1] = wordl1[1] | wordr1[1];
     w1[2] = wordl1[2] | wordr1[2];
     w1[3] = wordl1[3] | wordr1[3];
-
-    u32x w2[4];
-
     w2[0] = wordl2[0] | wordr2[0];
     w2[1] = wordl2[1] | wordr2[1];
     w2[2] = wordl2[2] | wordr2[2];
     w2[3] = wordl2[3] | wordr2[3];
-
-    u32x w3[4];
-
     w3[0] = wordl3[0] | wordr3[0];
     w3[1] = wordl3[1] | wordr3[1];
-    w3[2] = pw_len * 8;
-    w3[3] = 0;
-
-    u64 st[25];
-
-    st[ 0] = (u64) (w0[0]) | (u64) (w0[1]) << 32;
-    st[ 1] = (u64) (w0[2]) | (u64) (w0[3]) << 32;
-    st[ 2] = (u64) (w1[0]) | (u64) (w1[1]) << 32;
-    st[ 3] = (u64) (w1[2]) | (u64) (w1[3]) << 32;
-    st[ 4] = 0;
-    st[ 5] = 0;
-    st[ 6] = 0;
-    st[ 7] = 0;
+    w3[2] = wordl3[2] | wordr3[2];
+    w3[3] = wordl3[3] | wordr3[3];
+
+    /**
+     * Keccak
+     */
+
+    u64x st[25];
+
+    st[ 0] = hl32_to_64 (w0[1], w0[0]);
+    st[ 1] = hl32_to_64 (w0[3], w0[2]);
+    st[ 2] = hl32_to_64 (w1[1], w1[0]);
+    st[ 3] = hl32_to_64 (w1[3], w1[2]);
+    st[ 4] = hl32_to_64 (w2[1], w2[0]);
+    st[ 5] = hl32_to_64 (w2[3], w2[2]);
+    st[ 6] = hl32_to_64 (w3[1], w3[0]);
+    st[ 7] = hl32_to_64 (w3[3], w3[2]);
     st[ 8] = 0;
     st[ 9] = 0;
     st[10] = 0;
@@ -221,13 +233,13 @@ __kernel void m05000_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
     {
       // Theta
 
-      u64 bc0 = Theta1 (0);
-      u64 bc1 = Theta1 (1);
-      u64 bc2 = Theta1 (2);
-      u64 bc3 = Theta1 (3);
-      u64 bc4 = Theta1 (4);
+      u64x bc0 = Theta1 (0);
+      u64x bc1 = Theta1 (1);
+      u64x bc2 = Theta1 (2);
+      u64x bc3 = Theta1 (3);
+      u64x bc4 = Theta1 (4);
 
-      u64 t;
+      u64x t;
 
       t = bc4 ^ rotl64 (bc1, 1); Theta2 (0);
       t = bc0 ^ rotl64 (bc2, 1); Theta2 (1);
@@ -277,24 +289,24 @@ __kernel void m05000_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
       st[0] ^= keccakf_rndc[round];
     }
 
-    const u32 r0 = l32_from_64 (st[1]);
-    const u32 r1 = h32_from_64 (st[1]);
-    const u32 r2 = l32_from_64 (st[2]);
-    const u32 r3 = h32_from_64 (st[2]);
+    const u32x r0 = l32_from_64 (st[1]);
+    const u32x r1 = h32_from_64 (st[1]);
+    const u32x r2 = l32_from_64 (st[2]);
+    const u32x r3 = h32_from_64 (st[2]);
 
-    #include COMPARE_M
+    COMPARE_M_SIMD (r0, r1, r2, r3);
   }
 }
 
-__kernel void m05000_m08 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_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 combs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void m05000_m08 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_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 u32 gid_max)
 {
 }
 
-__kernel void m05000_m16 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_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 combs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void m05000_m16 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_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 u32 gid_max)
 {
 }
 
-__kernel void m05000_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_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 combs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void m05000_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_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 u32 gid_max)
 {
   /**
    * modifier
@@ -303,7 +315,29 @@ __kernel void m05000_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
   const u32 lid = get_local_id (0);
 
   /**
-   * const
+   * base
+   */
+
+  const u32 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  u32 pw_buf0[4];
+  u32 pw_buf1[4];
+
+  pw_buf0[0] = pws[gid].i[0];
+  pw_buf0[1] = pws[gid].i[1];
+  pw_buf0[2] = pws[gid].i[2];
+  pw_buf0[3] = pws[gid].i[3];
+  pw_buf1[0] = pws[gid].i[4];
+  pw_buf1[1] = pws[gid].i[5];
+  pw_buf1[2] = pws[gid].i[6];
+  pw_buf1[3] = pws[gid].i[7];
+
+  const u32 pw_l_len = pws[gid].pw_len;
+
+  /**
+   * constants
    */
 
   const u8 keccakf_rotc[24] =
@@ -319,26 +353,14 @@ __kernel void m05000_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
   };
 
   /**
-   * base
+   * 0x80 keccak, very special
    */
 
-  const u32 gid = get_global_id (0);
-
-  if (gid >= gid_max) return;
-
-  u32 pws0[4] = { 0 };
-  u32 pws1[4] = { 0 };
+  const u32 mdlen = salt_bufs[salt_pos].keccak_mdlen;
 
-  pws0[0] = pws[gid].i[0];
-  pws0[1] = pws[gid].i[1];
-  pws0[2] = pws[gid].i[2];
-  pws0[3] = pws[gid].i[3];
-  pws1[0] = pws[gid].i[4];
-  pws1[1] = pws[gid].i[5];
-  pws1[2] = pws[gid].i[6];
-  pws1[3] = pws[gid].i[7];
+  const u32 rsiz = 200 - (2 * mdlen);
 
-  const u32 pw_l_len = pws[gid].pw_len;
+  const u32 add80w = (rsiz - 1) / 8;
 
   /**
    * digest
@@ -352,26 +374,34 @@ __kernel void m05000_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
     digests_buf[digests_offset].digest_buf[DGST_R3]
   };
 
-  /**
-   * 0x80 keccak, very special
-   */
-
-  const u32 mdlen = salt_bufs[salt_pos].keccak_mdlen;
-
-  const u32 rsiz = 200 - (2 * mdlen);
-
-  const u32 add80w = (rsiz - 1) / 8;
-
   /**
    * loop
    */
 
-  for (u32 il_pos = 0; il_pos < combs_cnt; il_pos += VECT_SIZE)
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
   {
     const u32x pw_r_len = pwlenx_create_combt (combs_buf, il_pos);
 
     const u32x pw_len = pw_l_len + pw_r_len;
 
+    /**
+     * concat password candidate
+     */
+
+    u32x wordl0[4] = { 0 };
+    u32x wordl1[4] = { 0 };
+    u32x wordl2[4] = { 0 };
+    u32x wordl3[4] = { 0 };
+
+    wordl0[0] = pw_buf0[0];
+    wordl0[1] = pw_buf0[1];
+    wordl0[2] = pw_buf0[2];
+    wordl0[3] = pw_buf0[3];
+    wordl1[0] = pw_buf1[0];
+    wordl1[1] = pw_buf1[1];
+    wordl1[2] = pw_buf1[2];
+    wordl1[3] = pw_buf1[3];
+
     u32x wordr0[4] = { 0 };
     u32x wordr1[4] = { 0 };
     u32x wordr2[4] = { 0 };
@@ -388,49 +418,49 @@ __kernel void m05000_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
 
     if (combs_mode == COMBINATOR_MODE_BASE_LEFT)
     {
-      append_0x01_2x4 (wordr0, wordr1, pw_r_len);
-
-      switch_buffer_by_offset_le (wordr0, wordr1, wordr2, wordr3, pw_l_len);
+      switch_buffer_by_offset_le_VV (wordr0, wordr1, wordr2, wordr3, pw_l_len);
+    }
+    else
+    {
+      switch_buffer_by_offset_le_VV (wordl0, wordl1, wordl2, wordl3, pw_r_len);
     }
 
     u32x w0[4];
+    u32x w1[4];
+    u32x w2[4];
+    u32x w3[4];
 
     w0[0] = wordl0[0] | wordr0[0];
     w0[1] = wordl0[1] | wordr0[1];
     w0[2] = wordl0[2] | wordr0[2];
     w0[3] = wordl0[3] | wordr0[3];
-
-    u32x w1[4];
-
     w1[0] = wordl1[0] | wordr1[0];
     w1[1] = wordl1[1] | wordr1[1];
     w1[2] = wordl1[2] | wordr1[2];
     w1[3] = wordl1[3] | wordr1[3];
-
-    u32x w2[4];
-
     w2[0] = wordl2[0] | wordr2[0];
     w2[1] = wordl2[1] | wordr2[1];
     w2[2] = wordl2[2] | wordr2[2];
     w2[3] = wordl2[3] | wordr2[3];
-
-    u32x w3[4];
-
     w3[0] = wordl3[0] | wordr3[0];
     w3[1] = wordl3[1] | wordr3[1];
-    w3[2] = pw_len * 8;
-    w3[3] = 0;
-
-    u64 st[25];
-
-    st[ 0] = (u64) (w0[0]) | (u64) (w0[1]) << 32;
-    st[ 1] = (u64) (w0[2]) | (u64) (w0[3]) << 32;
-    st[ 2] = (u64) (w1[0]) | (u64) (w1[1]) << 32;
-    st[ 3] = (u64) (w1[2]) | (u64) (w1[3]) << 32;
-    st[ 4] = 0;
-    st[ 5] = 0;
-    st[ 6] = 0;
-    st[ 7] = 0;
+    w3[2] = wordl3[2] | wordr3[2];
+    w3[3] = wordl3[3] | wordr3[3];
+
+    /**
+     * Keccak
+     */
+
+    u64x st[25];
+
+    st[ 0] = hl32_to_64 (w0[1], w0[0]);
+    st[ 1] = hl32_to_64 (w0[3], w0[2]);
+    st[ 2] = hl32_to_64 (w1[1], w1[0]);
+    st[ 3] = hl32_to_64 (w1[3], w1[2]);
+    st[ 4] = hl32_to_64 (w2[1], w2[0]);
+    st[ 5] = hl32_to_64 (w2[3], w2[2]);
+    st[ 6] = hl32_to_64 (w3[1], w3[0]);
+    st[ 7] = hl32_to_64 (w3[3], w3[2]);
     st[ 8] = 0;
     st[ 9] = 0;
     st[10] = 0;
@@ -457,13 +487,13 @@ __kernel void m05000_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
     {
       // Theta
 
-      u64 bc0 = Theta1 (0);
-      u64 bc1 = Theta1 (1);
-      u64 bc2 = Theta1 (2);
-      u64 bc3 = Theta1 (3);
-      u64 bc4 = Theta1 (4);
+      u64x bc0 = Theta1 (0);
+      u64x bc1 = Theta1 (1);
+      u64x bc2 = Theta1 (2);
+      u64x bc3 = Theta1 (3);
+      u64x bc4 = Theta1 (4);
 
-      u64 t;
+      u64x t;
 
       t = bc4 ^ rotl64 (bc1, 1); Theta2 (0);
       t = bc0 ^ rotl64 (bc2, 1); Theta2 (1);
@@ -513,19 +543,19 @@ __kernel void m05000_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
       st[0] ^= keccakf_rndc[round];
     }
 
-    const u32 r0 = l32_from_64 (st[1]);
-    const u32 r1 = h32_from_64 (st[1]);
-    const u32 r2 = l32_from_64 (st[2]);
-    const u32 r3 = h32_from_64 (st[2]);
+    const u32x r0 = l32_from_64 (st[1]);
+    const u32x r1 = h32_from_64 (st[1]);
+    const u32x r2 = l32_from_64 (st[2]);
+    const u32x r3 = h32_from_64 (st[2]);
 
-    #include COMPARE_S
+    COMPARE_S_SIMD (r0, r1, r2, r3);
   }
 }
 
-__kernel void m05000_s08 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_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 combs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void m05000_s08 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_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 u32 gid_max)
 {
 }
 
-__kernel void m05000_s16 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_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 combs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void m05000_s16 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_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 u32 gid_max)
 {
 }