X-Git-Url: https://www.flypig.org.uk/git/?a=blobdiff_plain;f=OpenCL%2Fm01720_a3.cl;h=9b74a9a254eea11ef652c16888e6d9c8e01d0c19;hb=161a6eb4bc643d8e636e96eda613f5137d30da59;hp=8d0eb13d9c8358359ff175d06b5df8fcaea770a6;hpb=7190dcf855355440502559dddf59038081ff6350;p=hashcat.git

diff --git a/OpenCL/m01720_a3.cl b/OpenCL/m01720_a3.cl
index 8d0eb13..9b74a9a 100644
--- a/OpenCL/m01720_a3.cl
+++ b/OpenCL/m01720_a3.cl
@@ -7,18 +7,12 @@
 
 #define NEW_SIMD_CODE
 
-#include "include/constants.h"
-#include "include/kernel_vendor.h"
-
-#define DGST_R0 14
-#define DGST_R1 15
-#define DGST_R2 6
-#define DGST_R3 7
-
-#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 k_sha512[80] =
 {
@@ -44,7 +38,7 @@ __constant u64 k_sha512[80] =
   SHA512C4c, SHA512C4d, SHA512C4e, SHA512C4f,
 };
 
-static void sha512_transform (const u32x w0[4], const u32x w1[4], const u32x w2[4], const u32x w3[4], u64x digest[8])
+void sha512_transform (const u32x w0[4], const u32x w1[4], const u32x w2[4], const u32x w3[4], u64x digest[8])
 {
   u64x w0_t = hl32_to_64 (w0[0], w0[1]);
   u64x w1_t = hl32_to_64 (w0[2], w0[3]);
@@ -114,7 +108,9 @@ static void sha512_transform (const u32x w0[4], const u32x w1[4], const u32x w2[
 
   ROUND_STEP (0);
 
+  #ifdef _unroll
   #pragma unroll
+  #endif
   for (int i = 16; i < 80; i += 16)
   {
     ROUND_EXPAND (); ROUND_STEP (i);
@@ -141,7 +137,7 @@ static void sha512_transform (const u32x w0[4], const u32x w1[4], const u32x w2[
   digest[7] = h;
 }
 
-static void m01720m (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], const u32 pw_len, __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 bfs_cnt, const u32 digests_cnt, const u32 digests_offset)
+void m01720m (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], const u32 pw_len, __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)
 {
   /**
    * modifier
@@ -155,159 +151,116 @@ static void m01720m (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], const u32 pw_le
    */
 
   u32 salt_buf0[4];
-
-  salt_buf0[0] = salt_bufs[salt_pos].salt_buf[ 0];
-  salt_buf0[1] = salt_bufs[salt_pos].salt_buf[ 1];
-  salt_buf0[2] = salt_bufs[salt_pos].salt_buf[ 2];
-  salt_buf0[3] = salt_bufs[salt_pos].salt_buf[ 3];
-
   u32 salt_buf1[4];
-
-  salt_buf1[0] = salt_bufs[salt_pos].salt_buf[ 4];
-  salt_buf1[1] = salt_bufs[salt_pos].salt_buf[ 5];
-  salt_buf1[2] = salt_bufs[salt_pos].salt_buf[ 6];
-  salt_buf1[3] = salt_bufs[salt_pos].salt_buf[ 7];
-
   u32 salt_buf2[4];
-
-  salt_buf2[0] = 0;
-  salt_buf2[1] = 0;
-  salt_buf2[2] = 0;
-  salt_buf2[3] = 0;
-
   u32 salt_buf3[4];
 
-  salt_buf3[0] = 0;
-  salt_buf3[1] = 0;
-  salt_buf3[2] = 0;
-  salt_buf3[3] = 0;
+  salt_buf0[0] = swap32_S (salt_bufs[salt_pos].salt_buf[ 0]);
+  salt_buf0[1] = swap32_S (salt_bufs[salt_pos].salt_buf[ 1]);
+  salt_buf0[2] = swap32_S (salt_bufs[salt_pos].salt_buf[ 2]);
+  salt_buf0[3] = swap32_S (salt_bufs[salt_pos].salt_buf[ 3]);
+  salt_buf1[0] = swap32_S (salt_bufs[salt_pos].salt_buf[ 4]);
+  salt_buf1[1] = swap32_S (salt_bufs[salt_pos].salt_buf[ 5]);
+  salt_buf1[2] = swap32_S (salt_bufs[salt_pos].salt_buf[ 6]);
+  salt_buf1[3] = swap32_S (salt_bufs[salt_pos].salt_buf[ 7]);
+  salt_buf2[0] = swap32_S (salt_bufs[salt_pos].salt_buf[ 8]);
+  salt_buf2[1] = swap32_S (salt_bufs[salt_pos].salt_buf[ 9]);
+  salt_buf2[2] = swap32_S (salt_bufs[salt_pos].salt_buf[10]);
+  salt_buf2[3] = swap32_S (salt_bufs[salt_pos].salt_buf[11]);
+  salt_buf3[0] = swap32_S (salt_bufs[salt_pos].salt_buf[12]);
+  salt_buf3[1] = swap32_S (salt_bufs[salt_pos].salt_buf[13]);
+  salt_buf3[2] = swap32_S (salt_bufs[salt_pos].salt_buf[14]);
+  salt_buf3[3] = swap32_S (salt_bufs[salt_pos].salt_buf[15]);
 
   const u32 salt_len = salt_bufs[salt_pos].salt_len;
 
   const u32 pw_salt_len = pw_len + salt_len;
 
-
   /**
    * prepend salt
    */
 
-  u32 w0_t[4];
-  u32 w1_t[4];
-  u32 w2_t[4];
-  u32 w3_t[4];
-
-  w0_t[0] = swap32_S (w0[0]);
-  w0_t[1] = swap32_S (w0[1]);
-  w0_t[2] = swap32_S (w0[2]);
-  w0_t[3] = swap32_S (w0[3]);
-  w1_t[0] = swap32_S (w1[0]);
-  w1_t[1] = swap32_S (w1[1]);
-  w1_t[2] = swap32_S (w1[2]);
-  w1_t[3] = swap32_S (w1[3]);
-  w2_t[0] = swap32_S (w2[0]);
-  w2_t[1] = swap32_S (w2[1]);
-  w2_t[2] = swap32_S (w2[2]);
-  w2_t[3] = swap32_S (w2[3]);
-  w3_t[0] = swap32_S (w3[0]);
-  w3_t[1] = swap32_S (w3[1]);
-  w3_t[2] = swap32_S (w3[2]);
-  w3_t[3] = swap32_S (w3[3]);
-
-  switch_buffer_by_offset_le_S (w0_t, w1_t, w2_t, w3_t, salt_len);
-
-  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];
-
-  w0_t[0] = swap32_S (w0_t[0]);
-  w0_t[1] = swap32_S (w0_t[1]);
-  w0_t[2] = swap32_S (w0_t[2]);
-  w0_t[3] = swap32_S (w0_t[3]);
-  w1_t[0] = swap32_S (w1_t[0]);
-  w1_t[1] = swap32_S (w1_t[1]);
-  w1_t[2] = swap32_S (w1_t[2]);
-  w1_t[3] = swap32_S (w1_t[3]);
-  w2_t[0] = swap32_S (w2_t[0]);
-  w2_t[1] = swap32_S (w2_t[1]);
-  w2_t[2] = swap32_S (w2_t[2]);
-  w2_t[3] = swap32_S (w2_t[3]);
-  w3_t[0] = swap32_S (w3_t[0]);
-  w3_t[1] = swap32_S (w3_t[1]);
-  w3_t[2] = swap32_S (w3_t[2]);
-  w3_t[3] = swap32_S (w3_t[3]);
+  const u32 w0l = w0[0];
+
+  switch_buffer_by_offset_be_S (w0, w1, w2, w3, salt_len);
+
+  w0[0] |= salt_buf0[0];
+  w0[1] |= salt_buf0[1];
+  w0[2] |= salt_buf0[2];
+  w0[3] |= salt_buf0[3];
+  w1[0] |= salt_buf1[0];
+  w1[1] |= salt_buf1[1];
+  w1[2] |= salt_buf1[2];
+  w1[3] |= salt_buf1[3];
+  w2[0] |= salt_buf2[0];
+  w2[1] |= salt_buf2[1];
+  w2[2] |= salt_buf2[2];
+  w2[3] |= salt_buf2[3];
+  w3[0] |= salt_buf3[0];
+  w3[1] |= salt_buf3[1];
+  w3[2] |= salt_buf3[2];
+  w3[3] |= salt_buf3[3];
+
+  u32x t0[4];
+  u32x t1[4];
+  u32x t2[4];
+  u32x t3[4];
+
+  t0[0] = w0[0];
+  t0[1] = w0[1];
+  t0[2] = w0[2];
+  t0[3] = w0[3];
+  t1[0] = w1[0];
+  t1[1] = w1[1];
+  t1[2] = w1[2];
+  t1[3] = w1[3];
+  t2[0] = w2[0];
+  t2[1] = w2[1];
+  t2[2] = w2[2];
+  t2[3] = w2[3];
+  t3[0] = w3[0];
+  t3[1] = w3[1];
+  t3[2] = w3[2];
+  t3[3] = w3[3];
 
   /**
    * loop
    */
 
-  u32 w0l = w0[0];
-
-  for (u32 il_pos = 0; il_pos < bfs_cnt; il_pos += VECT_SIZE)
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
   {
     const u32x w0r = ix_create_bft (bfs_buf, il_pos);
 
     const u32x w0lr = w0l | w0r;
 
-    u32x wx[16];
-
-    wx[ 0] = w0_t[0];
-    wx[ 1] = w0_t[1];
-    wx[ 2] = w0_t[2];
-    wx[ 3] = w0_t[3];
-    wx[ 4] = w1_t[0];
-    wx[ 5] = w1_t[1];
-    wx[ 6] = w1_t[2];
-    wx[ 7] = w1_t[3];
-    wx[ 8] = w2_t[0];
-    wx[ 9] = w2_t[1];
-    wx[10] = w2_t[2];
-    wx[11] = w2_t[3];
-    wx[12] = w3_t[0];
-    wx[13] = w3_t[1];
-    wx[14] = w3_t[2];
-    wx[15] = w3_t[3];
-
-    overwrite_at_be (wx, w0lr, salt_len);
+    overwrite_at_be_4x4 (t0, t1, t2, t3, w0lr, salt_len);
+
+    /**
+     * sha512
+     */
 
     u32x w0_t[4];
     u32x w1_t[4];
     u32x w2_t[4];
     u32x w3_t[4];
 
-    w0_t[0] = wx[ 0];
-    w0_t[1] = wx[ 1];
-    w0_t[2] = wx[ 2];
-    w0_t[3] = wx[ 3];
-    w1_t[0] = wx[ 4];
-    w1_t[1] = wx[ 5];
-    w1_t[2] = wx[ 6];
-    w1_t[3] = wx[ 7];
-    w2_t[0] = wx[ 8];
-    w2_t[1] = wx[ 9];
-    w2_t[2] = wx[10];
-    w2_t[3] = wx[11];
-    w3_t[0] = wx[12];
-    w3_t[1] = wx[13];
+    w0_t[0] = t0[0];
+    w0_t[1] = t0[1];
+    w0_t[2] = t0[2];
+    w0_t[3] = t0[3];
+    w1_t[0] = t1[0];
+    w1_t[1] = t1[1];
+    w1_t[2] = t1[2];
+    w1_t[3] = t1[3];
+    w2_t[0] = t2[0];
+    w2_t[1] = t2[1];
+    w2_t[2] = t2[2];
+    w2_t[3] = t2[3];
+    w3_t[0] = t3[0];
+    w3_t[1] = t3[1];
     w3_t[2] = 0;
     w3_t[3] = pw_salt_len * 8;
 
-    /**
-     * sha512
-     */
-
     u64x digest[8];
 
     digest[0] = SHA512M_A;
@@ -330,7 +283,7 @@ static void m01720m (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], const u32 pw_le
   }
 }
 
-static void m01720s (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], const u32 pw_len, __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 bfs_cnt, const u32 digests_cnt, const u32 digests_offset)
+void m01720s (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], const u32 pw_len, __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)
 {
   /**
    * modifier
@@ -356,32 +309,26 @@ static void m01720s (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], const u32 pw_le
    */
 
   u32 salt_buf0[4];
-
-  salt_buf0[0] = salt_bufs[salt_pos].salt_buf[ 0];
-  salt_buf0[1] = salt_bufs[salt_pos].salt_buf[ 1];
-  salt_buf0[2] = salt_bufs[salt_pos].salt_buf[ 2];
-  salt_buf0[3] = salt_bufs[salt_pos].salt_buf[ 3];
-
   u32 salt_buf1[4];
-
-  salt_buf1[0] = salt_bufs[salt_pos].salt_buf[ 4];
-  salt_buf1[1] = salt_bufs[salt_pos].salt_buf[ 5];
-  salt_buf1[2] = salt_bufs[salt_pos].salt_buf[ 6];
-  salt_buf1[3] = salt_bufs[salt_pos].salt_buf[ 7];
-
   u32 salt_buf2[4];
-
-  salt_buf2[0] = 0;
-  salt_buf2[1] = 0;
-  salt_buf2[2] = 0;
-  salt_buf2[3] = 0;
-
   u32 salt_buf3[4];
 
-  salt_buf3[0] = 0;
-  salt_buf3[1] = 0;
-  salt_buf3[2] = 0;
-  salt_buf3[3] = 0;
+  salt_buf0[0] = swap32_S (salt_bufs[salt_pos].salt_buf[ 0]);
+  salt_buf0[1] = swap32_S (salt_bufs[salt_pos].salt_buf[ 1]);
+  salt_buf0[2] = swap32_S (salt_bufs[salt_pos].salt_buf[ 2]);
+  salt_buf0[3] = swap32_S (salt_bufs[salt_pos].salt_buf[ 3]);
+  salt_buf1[0] = swap32_S (salt_bufs[salt_pos].salt_buf[ 4]);
+  salt_buf1[1] = swap32_S (salt_bufs[salt_pos].salt_buf[ 5]);
+  salt_buf1[2] = swap32_S (salt_bufs[salt_pos].salt_buf[ 6]);
+  salt_buf1[3] = swap32_S (salt_bufs[salt_pos].salt_buf[ 7]);
+  salt_buf2[0] = swap32_S (salt_bufs[salt_pos].salt_buf[ 8]);
+  salt_buf2[1] = swap32_S (salt_bufs[salt_pos].salt_buf[ 9]);
+  salt_buf2[2] = swap32_S (salt_bufs[salt_pos].salt_buf[10]);
+  salt_buf2[3] = swap32_S (salt_bufs[salt_pos].salt_buf[11]);
+  salt_buf3[0] = swap32_S (salt_bufs[salt_pos].salt_buf[12]);
+  salt_buf3[1] = swap32_S (salt_bufs[salt_pos].salt_buf[13]);
+  salt_buf3[2] = swap32_S (salt_bufs[salt_pos].salt_buf[14]);
+  salt_buf3[3] = swap32_S (salt_bufs[salt_pos].salt_buf[15]);
 
   const u32 salt_len = salt_bufs[salt_pos].salt_len;
 
@@ -391,120 +338,84 @@ static void m01720s (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], const u32 pw_le
    * prepend salt
    */
 
-  u32 w0_t[4];
-  u32 w1_t[4];
-  u32 w2_t[4];
-  u32 w3_t[4];
-
-  w0_t[0] = swap32_S (w0[0]);
-  w0_t[1] = swap32_S (w0[1]);
-  w0_t[2] = swap32_S (w0[2]);
-  w0_t[3] = swap32_S (w0[3]);
-  w1_t[0] = swap32_S (w1[0]);
-  w1_t[1] = swap32_S (w1[1]);
-  w1_t[2] = swap32_S (w1[2]);
-  w1_t[3] = swap32_S (w1[3]);
-  w2_t[0] = swap32_S (w2[0]);
-  w2_t[1] = swap32_S (w2[1]);
-  w2_t[2] = swap32_S (w2[2]);
-  w2_t[3] = swap32_S (w2[3]);
-  w3_t[0] = swap32_S (w3[0]);
-  w3_t[1] = swap32_S (w3[1]);
-  w3_t[2] = swap32_S (w3[2]);
-  w3_t[3] = swap32_S (w3[3]);
-
-  switch_buffer_by_offset_le_S (w0_t, w1_t, w2_t, w3_t, salt_len);
-
-  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];
-
-  w0_t[0] = swap32_S (w0_t[0]);
-  w0_t[1] = swap32_S (w0_t[1]);
-  w0_t[2] = swap32_S (w0_t[2]);
-  w0_t[3] = swap32_S (w0_t[3]);
-  w1_t[0] = swap32_S (w1_t[0]);
-  w1_t[1] = swap32_S (w1_t[1]);
-  w1_t[2] = swap32_S (w1_t[2]);
-  w1_t[3] = swap32_S (w1_t[3]);
-  w2_t[0] = swap32_S (w2_t[0]);
-  w2_t[1] = swap32_S (w2_t[1]);
-  w2_t[2] = swap32_S (w2_t[2]);
-  w2_t[3] = swap32_S (w2_t[3]);
-  w3_t[0] = swap32_S (w3_t[0]);
-  w3_t[1] = swap32_S (w3_t[1]);
-  w3_t[2] = swap32_S (w3_t[2]);
-  w3_t[3] = swap32_S (w3_t[3]);
+  const u32 w0l = w0[0];
+
+  switch_buffer_by_offset_be_S (w0, w1, w2, w3, salt_len);
+
+  w0[0] |= salt_buf0[0];
+  w0[1] |= salt_buf0[1];
+  w0[2] |= salt_buf0[2];
+  w0[3] |= salt_buf0[3];
+  w1[0] |= salt_buf1[0];
+  w1[1] |= salt_buf1[1];
+  w1[2] |= salt_buf1[2];
+  w1[3] |= salt_buf1[3];
+  w2[0] |= salt_buf2[0];
+  w2[1] |= salt_buf2[1];
+  w2[2] |= salt_buf2[2];
+  w2[3] |= salt_buf2[3];
+  w3[0] |= salt_buf3[0];
+  w3[1] |= salt_buf3[1];
+  w3[2] |= salt_buf3[2];
+  w3[3] |= salt_buf3[3];
+
+  u32x t0[4];
+  u32x t1[4];
+  u32x t2[4];
+  u32x t3[4];
+
+  t0[0] = w0[0];
+  t0[1] = w0[1];
+  t0[2] = w0[2];
+  t0[3] = w0[3];
+  t1[0] = w1[0];
+  t1[1] = w1[1];
+  t1[2] = w1[2];
+  t1[3] = w1[3];
+  t2[0] = w2[0];
+  t2[1] = w2[1];
+  t2[2] = w2[2];
+  t2[3] = w2[3];
+  t3[0] = w3[0];
+  t3[1] = w3[1];
+  t3[2] = w3[2];
+  t3[3] = w3[3];
 
   /**
    * loop
    */
 
-  u32 w0l = w0[0];
-
-  for (u32 il_pos = 0; il_pos < bfs_cnt; il_pos += VECT_SIZE)
+  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;
 
+    overwrite_at_be_4x4 (t0, t1, t2, t3, w0lr, salt_len);
+
     /**
-     * prepend salt
+     * sha512
      */
 
-    u32x wx[16];
-
-    wx[ 0] = w0_t[0];
-    wx[ 1] = w0_t[1];
-    wx[ 2] = w0_t[2];
-    wx[ 3] = w0_t[3];
-    wx[ 4] = w1_t[0];
-    wx[ 5] = w1_t[1];
-    wx[ 6] = w1_t[2];
-    wx[ 7] = w1_t[3];
-    wx[ 8] = w2_t[0];
-    wx[ 9] = w2_t[1];
-    wx[10] = w2_t[2];
-    wx[11] = w2_t[3];
-    wx[12] = w3_t[0];
-    wx[13] = w3_t[1];
-    wx[14] = w3_t[2];
-    wx[15] = w3_t[3];
-
-    overwrite_at_be (wx, w0lr, salt_len);
-
     u32x w0_t[4];
     u32x w1_t[4];
     u32x w2_t[4];
     u32x w3_t[4];
 
-    w0_t[0] = wx[ 0];
-    w0_t[1] = wx[ 1];
-    w0_t[2] = wx[ 2];
-    w0_t[3] = wx[ 3];
-    w1_t[0] = wx[ 4];
-    w1_t[1] = wx[ 5];
-    w1_t[2] = wx[ 6];
-    w1_t[3] = wx[ 7];
-    w2_t[0] = wx[ 8];
-    w2_t[1] = wx[ 9];
-    w2_t[2] = wx[10];
-    w2_t[3] = wx[11];
-    w3_t[0] = wx[12];
-    w3_t[1] = wx[13];
+    w0_t[0] = t0[0];
+    w0_t[1] = t0[1];
+    w0_t[2] = t0[2];
+    w0_t[3] = t0[3];
+    w1_t[0] = t1[0];
+    w1_t[1] = t1[1];
+    w1_t[2] = t1[2];
+    w1_t[3] = t1[3];
+    w2_t[0] = t2[0];
+    w2_t[1] = t2[1];
+    w2_t[2] = t2[2];
+    w2_t[3] = t2[3];
+    w3_t[0] = t3[0];
+    w3_t[1] = t3[1];
     w3_t[2] = 0;
     w3_t[3] = pw_salt_len * 8;
 
@@ -530,7 +441,7 @@ static void m01720s (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], const u32 pw_le
   }
 }
 
-__kernel void m01720_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 bfs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void m01720_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)
 {
   /**
    * base
@@ -574,10 +485,10 @@ __kernel void m01720_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
    * main
    */
 
-  m01720m (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_scryptV_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, bfs_cnt, digests_cnt, digests_offset);
+  m01720m (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
 }
 
-__kernel void m01720_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 bfs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void m01720_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)
 {
   /**
    * base
@@ -621,10 +532,10 @@ __kernel void m01720_m08 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
    * main
    */
 
-  m01720m (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_scryptV_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, bfs_cnt, digests_cnt, digests_offset);
+  m01720m (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
 }
 
-__kernel void m01720_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 bfs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void m01720_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)
 {
   /**
    * base
@@ -668,10 +579,10 @@ __kernel void m01720_m16 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
    * main
    */
 
-  m01720m (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_scryptV_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, bfs_cnt, digests_cnt, digests_offset);
+  m01720m (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
 }
 
-__kernel void m01720_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 bfs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void m01720_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)
 {
   /**
    * base
@@ -715,10 +626,10 @@ __kernel void m01720_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
    * main
    */
 
-  m01720s (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_scryptV_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, bfs_cnt, digests_cnt, digests_offset);
+  m01720s (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
 }
 
-__kernel void m01720_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 bfs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void m01720_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)
 {
   /**
    * base
@@ -762,10 +673,10 @@ __kernel void m01720_s08 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
    * main
    */
 
-  m01720s (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_scryptV_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, bfs_cnt, digests_cnt, digests_offset);
+  m01720s (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
 }
 
-__kernel void m01720_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 bfs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void m01720_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)
 {
   /**
    * base
@@ -809,5 +720,5 @@ __kernel void m01720_s16 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
    * main
    */
 
-  m01720s (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_scryptV_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, bfs_cnt, digests_cnt, digests_offset);
+  m01720s (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
 }