From 22137866186fe88f0d32756b06f21eeff4da7f64 Mon Sep 17 00:00:00 2001
From: jsteube <jens.steube@gmail.com>
Date: Sat, 28 May 2016 15:36:04 +0200
Subject: [PATCH] Increase Domain Cached Credentials (DCC), MS Cache cracking
 performance

---
 OpenCL/m01100_a0.cl | 142 ++++++++++++++++++----------------
 OpenCL/m01100_a1.cl | 130 ++++++++++++++++---------------
 OpenCL/m01100_a3.cl | 182 +++++++++++++++++++++++++++++---------------
 3 files changed, 268 insertions(+), 186 deletions(-)

diff --git a/OpenCL/m01100_a0.cl b/OpenCL/m01100_a0.cl
index e13d20f..5408b3a 100644
--- a/OpenCL/m01100_a0.cl
+++ b/OpenCL/m01100_a0.cl
@@ -24,19 +24,12 @@
 
 __kernel void m01100_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 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
 {
-  /**
-   * modifier
-   */
-
-  const u32 lid = get_local_id (0);
-
   /**
    * base
    */
 
   const u32 gid = get_global_id (0);
-
-  if (gid >= gid_max) return;
+  const u32 lid = get_local_id (0);
 
   u32 pw_buf0[4];
   u32 pw_buf1[4];
@@ -56,18 +49,30 @@ __kernel void m01100_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
    * salt
    */
 
-  const u32 salt_buf0 = salt_bufs[salt_pos].salt_buf[0];
-  const u32 salt_buf1 = salt_bufs[salt_pos].salt_buf[1];
-  const u32 salt_buf2 = salt_bufs[salt_pos].salt_buf[2];
-  const u32 salt_buf3 = salt_bufs[salt_pos].salt_buf[3];
-  const u32 salt_buf4 = salt_bufs[salt_pos].salt_buf[4];
-  const u32 salt_buf5 = salt_bufs[salt_pos].salt_buf[5];
-  const u32 salt_buf6 = salt_bufs[salt_pos].salt_buf[6];
-  const u32 salt_buf7 = salt_bufs[salt_pos].salt_buf[7];
-  const u32 salt_buf8 = salt_bufs[salt_pos].salt_buf[8];
-  const u32 salt_buf9 = salt_bufs[salt_pos].salt_buf[9];
+  __local salt_t s_salt_buf[1];
 
-  const u32 salt_len  = salt_bufs[salt_pos].salt_len;
+  if (lid == 0)
+  {
+    s_salt_buf[0] = salt_bufs[salt_pos];
+
+    s_salt_buf[0].salt_buf[10] = (16 + s_salt_buf[0].salt_len) * 8;
+  }
+
+  barrier (CLK_LOCAL_MEM_FENCE);
+
+  if (gid >= gid_max) return;
+
+  #define salt_buf00 s_salt_buf[0].salt_buf[ 0]
+  #define salt_buf01 s_salt_buf[0].salt_buf[ 1]
+  #define salt_buf02 s_salt_buf[0].salt_buf[ 2]
+  #define salt_buf03 s_salt_buf[0].salt_buf[ 3]
+  #define salt_buf04 s_salt_buf[0].salt_buf[ 4]
+  #define salt_buf05 s_salt_buf[0].salt_buf[ 5]
+  #define salt_buf06 s_salt_buf[0].salt_buf[ 6]
+  #define salt_buf07 s_salt_buf[0].salt_buf[ 7]
+  #define salt_buf08 s_salt_buf[0].salt_buf[ 8]
+  #define salt_buf09 s_salt_buf[0].salt_buf[ 9]
+  #define salt_buf10 s_salt_buf[0].salt_buf[10]
 
   /**
    * loop
@@ -155,17 +160,17 @@ __kernel void m01100_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
     w0[1] = b;
     w0[2] = c;
     w0[3] = d;
-    w1[0] = salt_buf0;
-    w1[1] = salt_buf1;
-    w1[2] = salt_buf2;
-    w1[3] = salt_buf3;
-    w2[0] = salt_buf4;
-    w2[1] = salt_buf5;
-    w2[2] = salt_buf6;
-    w2[3] = salt_buf7;
-    w3[0] = salt_buf8;
-    w3[1] = salt_buf9;
-    w3[2] = (16 + salt_len) * 8;
+    w1[0] = salt_buf00;
+    w1[1] = salt_buf01;
+    w1[2] = salt_buf02;
+    w1[3] = salt_buf03;
+    w2[0] = salt_buf04;
+    w2[1] = salt_buf05;
+    w2[2] = salt_buf06;
+    w2[3] = salt_buf07;
+    w3[0] = salt_buf08;
+    w3[1] = salt_buf09;
+    w3[2] = salt_buf10;
     w3[3] = 0;
 
     a = MD4M_A;
@@ -238,19 +243,12 @@ __kernel void m01100_m16 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
 
 __kernel void m01100_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 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
 {
-  /**
-   * modifier
-   */
-
-  const u32 lid = get_local_id (0);
-
   /**
    * base
    */
 
   const u32 gid = get_global_id (0);
-
-  if (gid >= gid_max) return;
+  const u32 lid = get_local_id (0);
 
   u32 pw_buf0[4];
   u32 pw_buf1[4];
@@ -266,6 +264,35 @@ __kernel void m01100_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
 
   const u32 pw_len = pws[gid].pw_len;
 
+  /**
+   * salt
+   */
+
+  __local salt_t s_salt_buf[1];
+
+  if (lid == 0)
+  {
+    s_salt_buf[0] = salt_bufs[salt_pos];
+
+    s_salt_buf[0].salt_buf[10] = (16 + s_salt_buf[0].salt_len) * 8;
+  }
+
+  barrier (CLK_LOCAL_MEM_FENCE);
+
+  if (gid >= gid_max) return;
+
+  #define salt_buf00 s_salt_buf[0].salt_buf[ 0]
+  #define salt_buf01 s_salt_buf[0].salt_buf[ 1]
+  #define salt_buf02 s_salt_buf[0].salt_buf[ 2]
+  #define salt_buf03 s_salt_buf[0].salt_buf[ 3]
+  #define salt_buf04 s_salt_buf[0].salt_buf[ 4]
+  #define salt_buf05 s_salt_buf[0].salt_buf[ 5]
+  #define salt_buf06 s_salt_buf[0].salt_buf[ 6]
+  #define salt_buf07 s_salt_buf[0].salt_buf[ 7]
+  #define salt_buf08 s_salt_buf[0].salt_buf[ 8]
+  #define salt_buf09 s_salt_buf[0].salt_buf[ 9]
+  #define salt_buf10 s_salt_buf[0].salt_buf[10]
+
   /**
    * digest
    */
@@ -278,23 +305,6 @@ __kernel void m01100_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
     digests_buf[digests_offset].digest_buf[DGST_R3]
   };
 
-  /**
-   * salt
-   */
-
-  const u32 salt_buf0 = salt_bufs[salt_pos].salt_buf[0];
-  const u32 salt_buf1 = salt_bufs[salt_pos].salt_buf[1];
-  const u32 salt_buf2 = salt_bufs[salt_pos].salt_buf[2];
-  const u32 salt_buf3 = salt_bufs[salt_pos].salt_buf[3];
-  const u32 salt_buf4 = salt_bufs[salt_pos].salt_buf[4];
-  const u32 salt_buf5 = salt_bufs[salt_pos].salt_buf[5];
-  const u32 salt_buf6 = salt_bufs[salt_pos].salt_buf[6];
-  const u32 salt_buf7 = salt_bufs[salt_pos].salt_buf[7];
-  const u32 salt_buf8 = salt_bufs[salt_pos].salt_buf[8];
-  const u32 salt_buf9 = salt_bufs[salt_pos].salt_buf[9];
-
-  const u32 salt_len  = salt_bufs[salt_pos].salt_len;
-
   /**
    * loop
    */
@@ -381,17 +391,17 @@ __kernel void m01100_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
     w0[1] = b;
     w0[2] = c;
     w0[3] = d;
-    w1[0] = salt_buf0;
-    w1[1] = salt_buf1;
-    w1[2] = salt_buf2;
-    w1[3] = salt_buf3;
-    w2[0] = salt_buf4;
-    w2[1] = salt_buf5;
-    w2[2] = salt_buf6;
-    w2[3] = salt_buf7;
-    w3[0] = salt_buf8;
-    w3[1] = salt_buf9;
-    w3[2] = (16 + salt_len) * 8;
+    w1[0] = salt_buf00;
+    w1[1] = salt_buf01;
+    w1[2] = salt_buf02;
+    w1[3] = salt_buf03;
+    w2[0] = salt_buf04;
+    w2[1] = salt_buf05;
+    w2[2] = salt_buf06;
+    w2[3] = salt_buf07;
+    w3[0] = salt_buf08;
+    w3[1] = salt_buf09;
+    w3[2] = salt_buf10;
     w3[3] = 0;
 
     a = MD4M_A;
diff --git a/OpenCL/m01100_a1.cl b/OpenCL/m01100_a1.cl
index fbaba9b..3e3fd65 100644
--- a/OpenCL/m01100_a1.cl
+++ b/OpenCL/m01100_a1.cl
@@ -22,19 +22,12 @@
 
 __kernel void m01100_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 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
 {
-  /**
-   * modifier
-   */
-
-  const u32 lid = get_local_id (0);
-
   /**
    * base
    */
 
   const u32 gid = get_global_id (0);
-
-  if (gid >= gid_max) return;
+  const u32 lid = get_local_id (0);
 
   u32 pw_buf0[4];
   u32 pw_buf1[4];
@@ -54,18 +47,30 @@ __kernel void m01100_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
    * salt
    */
 
-  const u32 salt_buf0 = salt_bufs[salt_pos].salt_buf[0];
-  const u32 salt_buf1 = salt_bufs[salt_pos].salt_buf[1];
-  const u32 salt_buf2 = salt_bufs[salt_pos].salt_buf[2];
-  const u32 salt_buf3 = salt_bufs[salt_pos].salt_buf[3];
-  const u32 salt_buf4 = salt_bufs[salt_pos].salt_buf[4];
-  const u32 salt_buf5 = salt_bufs[salt_pos].salt_buf[5];
-  const u32 salt_buf6 = salt_bufs[salt_pos].salt_buf[6];
-  const u32 salt_buf7 = salt_bufs[salt_pos].salt_buf[7];
-  const u32 salt_buf8 = salt_bufs[salt_pos].salt_buf[8];
-  const u32 salt_buf9 = salt_bufs[salt_pos].salt_buf[9];
+  __local salt_t s_salt_buf[1];
+
+  if (lid == 0)
+  {
+    s_salt_buf[0] = salt_bufs[salt_pos];
+
+    s_salt_buf[0].salt_buf[10] = (16 + s_salt_buf[0].salt_len) * 8;
+  }
+
+  barrier (CLK_LOCAL_MEM_FENCE);
+
+  if (gid >= gid_max) return;
 
-  const u32 salt_len  = salt_bufs[salt_pos].salt_len;
+  #define salt_buf00 s_salt_buf[0].salt_buf[ 0]
+  #define salt_buf01 s_salt_buf[0].salt_buf[ 1]
+  #define salt_buf02 s_salt_buf[0].salt_buf[ 2]
+  #define salt_buf03 s_salt_buf[0].salt_buf[ 3]
+  #define salt_buf04 s_salt_buf[0].salt_buf[ 4]
+  #define salt_buf05 s_salt_buf[0].salt_buf[ 5]
+  #define salt_buf06 s_salt_buf[0].salt_buf[ 6]
+  #define salt_buf07 s_salt_buf[0].salt_buf[ 7]
+  #define salt_buf08 s_salt_buf[0].salt_buf[ 8]
+  #define salt_buf09 s_salt_buf[0].salt_buf[ 9]
+  #define salt_buf10 s_salt_buf[0].salt_buf[10]
 
   /**
    * loop
@@ -215,17 +220,17 @@ __kernel void m01100_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
     w0[1] = b;
     w0[2] = c;
     w0[3] = d;
-    w1[0] = salt_buf0;
-    w1[1] = salt_buf1;
-    w1[2] = salt_buf2;
-    w1[3] = salt_buf3;
-    w2[0] = salt_buf4;
-    w2[1] = salt_buf5;
-    w2[2] = salt_buf6;
-    w2[3] = salt_buf7;
-    w3[0] = salt_buf8;
-    w3[1] = salt_buf9;
-    w3[2] = (16 + salt_len) * 8;
+    w1[0] = salt_buf00;
+    w1[1] = salt_buf01;
+    w1[2] = salt_buf02;
+    w1[3] = salt_buf03;
+    w2[0] = salt_buf04;
+    w2[1] = salt_buf05;
+    w2[2] = salt_buf06;
+    w2[3] = salt_buf07;
+    w3[0] = salt_buf08;
+    w3[1] = salt_buf09;
+    w3[2] = salt_buf10;
     w3[3] = 0;
 
     a = MD4M_A;
@@ -298,19 +303,12 @@ __kernel void m01100_m16 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
 
 __kernel void m01100_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 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
 {
-  /**
-   * modifier
-   */
-
-  const u32 lid = get_local_id (0);
-
   /**
    * base
    */
 
   const u32 gid = get_global_id (0);
-
-  if (gid >= gid_max) return;
+  const u32 lid = get_local_id (0);
 
   u32 pw_buf0[4];
   u32 pw_buf1[4];
@@ -330,18 +328,30 @@ __kernel void m01100_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
    * salt
    */
 
-  const u32 salt_buf0 = salt_bufs[salt_pos].salt_buf[0];
-  const u32 salt_buf1 = salt_bufs[salt_pos].salt_buf[1];
-  const u32 salt_buf2 = salt_bufs[salt_pos].salt_buf[2];
-  const u32 salt_buf3 = salt_bufs[salt_pos].salt_buf[3];
-  const u32 salt_buf4 = salt_bufs[salt_pos].salt_buf[4];
-  const u32 salt_buf5 = salt_bufs[salt_pos].salt_buf[5];
-  const u32 salt_buf6 = salt_bufs[salt_pos].salt_buf[6];
-  const u32 salt_buf7 = salt_bufs[salt_pos].salt_buf[7];
-  const u32 salt_buf8 = salt_bufs[salt_pos].salt_buf[8];
-  const u32 salt_buf9 = salt_bufs[salt_pos].salt_buf[9];
+  __local salt_t s_salt_buf[1];
+
+  if (lid == 0)
+  {
+    s_salt_buf[0] = salt_bufs[salt_pos];
+
+    s_salt_buf[0].salt_buf[10] = (16 + s_salt_buf[0].salt_len) * 8;
+  }
+
+  barrier (CLK_LOCAL_MEM_FENCE);
+
+  if (gid >= gid_max) return;
 
-  const u32 salt_len  = salt_bufs[salt_pos].salt_len;
+  #define salt_buf00 s_salt_buf[0].salt_buf[ 0]
+  #define salt_buf01 s_salt_buf[0].salt_buf[ 1]
+  #define salt_buf02 s_salt_buf[0].salt_buf[ 2]
+  #define salt_buf03 s_salt_buf[0].salt_buf[ 3]
+  #define salt_buf04 s_salt_buf[0].salt_buf[ 4]
+  #define salt_buf05 s_salt_buf[0].salt_buf[ 5]
+  #define salt_buf06 s_salt_buf[0].salt_buf[ 6]
+  #define salt_buf07 s_salt_buf[0].salt_buf[ 7]
+  #define salt_buf08 s_salt_buf[0].salt_buf[ 8]
+  #define salt_buf09 s_salt_buf[0].salt_buf[ 9]
+  #define salt_buf10 s_salt_buf[0].salt_buf[10]
 
   /**
    * digest
@@ -503,17 +513,17 @@ __kernel void m01100_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
     w0[1] = b;
     w0[2] = c;
     w0[3] = d;
-    w1[0] = salt_buf0;
-    w1[1] = salt_buf1;
-    w1[2] = salt_buf2;
-    w1[3] = salt_buf3;
-    w2[0] = salt_buf4;
-    w2[1] = salt_buf5;
-    w2[2] = salt_buf6;
-    w2[3] = salt_buf7;
-    w3[0] = salt_buf8;
-    w3[1] = salt_buf9;
-    w3[2] = (16 + salt_len) * 8;
+    w1[0] = salt_buf00;
+    w1[1] = salt_buf01;
+    w1[2] = salt_buf02;
+    w1[3] = salt_buf03;
+    w2[0] = salt_buf04;
+    w2[1] = salt_buf05;
+    w2[2] = salt_buf06;
+    w2[3] = salt_buf07;
+    w3[0] = salt_buf08;
+    w3[1] = salt_buf09;
+    w3[2] = salt_buf10;
     w3[3] = 0;
 
     a = MD4M_A;
diff --git a/OpenCL/m01100_a3.cl b/OpenCL/m01100_a3.cl
index 53d62d1..0a11010 100644
--- a/OpenCL/m01100_a3.cl
+++ b/OpenCL/m01100_a3.cl
@@ -20,7 +20,7 @@
 #include "inc_common.cl"
 #include "inc_simd.cl"
 
-void m01100m (u32 w[16], const u32 pw_len, __global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32x * words_buf_r, __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 il_cnt, const u32 digests_cnt, const u32 digests_offset)
+void m01100m (__local salt_t s_salt_buf[1], u32 w[16], const u32 pw_len, __global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32x * words_buf_r, __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 il_cnt, const u32 digests_cnt, const u32 digests_offset)
 {
   /**
    * modifier
@@ -33,18 +33,17 @@ void m01100m (u32 w[16], const u32 pw_len, __global pw_t *pws, __global kernel_r
    * salt
    */
 
-  const u32 salt_buf0 = salt_bufs[salt_pos].salt_buf[0];
-  const u32 salt_buf1 = salt_bufs[salt_pos].salt_buf[1];
-  const u32 salt_buf2 = salt_bufs[salt_pos].salt_buf[2];
-  const u32 salt_buf3 = salt_bufs[salt_pos].salt_buf[3];
-  const u32 salt_buf4 = salt_bufs[salt_pos].salt_buf[4];
-  const u32 salt_buf5 = salt_bufs[salt_pos].salt_buf[5];
-  const u32 salt_buf6 = salt_bufs[salt_pos].salt_buf[6];
-  const u32 salt_buf7 = salt_bufs[salt_pos].salt_buf[7];
-  const u32 salt_buf8 = salt_bufs[salt_pos].salt_buf[8];
-  const u32 salt_buf9 = salt_bufs[salt_pos].salt_buf[9];
-
-  const u32 salt_len  = salt_bufs[salt_pos].salt_len;
+  #define salt_buf00 s_salt_buf[0].salt_buf[ 0]
+  #define salt_buf01 s_salt_buf[0].salt_buf[ 1]
+  #define salt_buf02 s_salt_buf[0].salt_buf[ 2]
+  #define salt_buf03 s_salt_buf[0].salt_buf[ 3]
+  #define salt_buf04 s_salt_buf[0].salt_buf[ 4]
+  #define salt_buf05 s_salt_buf[0].salt_buf[ 5]
+  #define salt_buf06 s_salt_buf[0].salt_buf[ 6]
+  #define salt_buf07 s_salt_buf[0].salt_buf[ 7]
+  #define salt_buf08 s_salt_buf[0].salt_buf[ 8]
+  #define salt_buf09 s_salt_buf[0].salt_buf[ 9]
+  #define salt_buf10 s_salt_buf[0].salt_buf[10]
 
   /**
    * base
@@ -183,17 +182,17 @@ void m01100m (u32 w[16], const u32 pw_len, __global pw_t *pws, __global kernel_r
     w0_t[1] = b;
     w0_t[2] = c;
     w0_t[3] = d;
-    w1_t[0] = salt_buf0;
-    w1_t[1] = salt_buf1;
-    w1_t[2] = salt_buf2;
-    w1_t[3] = salt_buf3;
-    w2_t[0] = salt_buf4;
-    w2_t[1] = salt_buf5;
-    w2_t[2] = salt_buf6;
-    w2_t[3] = salt_buf7;
-    w3_t[0] = salt_buf8;
-    w3_t[1] = salt_buf9;
-    w3_t[2] = (16 + salt_len) * 8;
+    w1_t[0] = salt_buf00;
+    w1_t[1] = salt_buf01;
+    w1_t[2] = salt_buf02;
+    w1_t[3] = salt_buf03;
+    w2_t[0] = salt_buf04;
+    w2_t[1] = salt_buf05;
+    w2_t[2] = salt_buf06;
+    w2_t[3] = salt_buf07;
+    w3_t[0] = salt_buf08;
+    w3_t[1] = salt_buf09;
+    w3_t[2] = salt_buf10;
     w3_t[3] = 0;
 
     a = MD4M_A;
@@ -256,7 +255,7 @@ void m01100m (u32 w[16], const u32 pw_len, __global pw_t *pws, __global kernel_r
   }
 }
 
-void m01100s (u32 w[16], const u32 pw_len, __global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32x * words_buf_r, __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 il_cnt, const u32 digests_cnt, const u32 digests_offset)
+void m01100s (__local salt_t s_salt_buf[1], u32 w[16], const u32 pw_len, __global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32x * words_buf_r, __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 il_cnt, const u32 digests_cnt, const u32 digests_offset)
 {
   /**
    * modifier
@@ -269,18 +268,17 @@ void m01100s (u32 w[16], const u32 pw_len, __global pw_t *pws, __global kernel_r
    * salt
    */
 
-  const u32 salt_buf0 = salt_bufs[salt_pos].salt_buf[0];
-  const u32 salt_buf1 = salt_bufs[salt_pos].salt_buf[1];
-  const u32 salt_buf2 = salt_bufs[salt_pos].salt_buf[2];
-  const u32 salt_buf3 = salt_bufs[salt_pos].salt_buf[3];
-  const u32 salt_buf4 = salt_bufs[salt_pos].salt_buf[4];
-  const u32 salt_buf5 = salt_bufs[salt_pos].salt_buf[5];
-  const u32 salt_buf6 = salt_bufs[salt_pos].salt_buf[6];
-  const u32 salt_buf7 = salt_bufs[salt_pos].salt_buf[7];
-  const u32 salt_buf8 = salt_bufs[salt_pos].salt_buf[8];
-  const u32 salt_buf9 = salt_bufs[salt_pos].salt_buf[9];
-
-  const u32 salt_len  = salt_bufs[salt_pos].salt_len;
+  #define salt_buf00 s_salt_buf[0].salt_buf[ 0]
+  #define salt_buf01 s_salt_buf[0].salt_buf[ 1]
+  #define salt_buf02 s_salt_buf[0].salt_buf[ 2]
+  #define salt_buf03 s_salt_buf[0].salt_buf[ 3]
+  #define salt_buf04 s_salt_buf[0].salt_buf[ 4]
+  #define salt_buf05 s_salt_buf[0].salt_buf[ 5]
+  #define salt_buf06 s_salt_buf[0].salt_buf[ 6]
+  #define salt_buf07 s_salt_buf[0].salt_buf[ 7]
+  #define salt_buf08 s_salt_buf[0].salt_buf[ 8]
+  #define salt_buf09 s_salt_buf[0].salt_buf[ 9]
+  #define salt_buf10 s_salt_buf[0].salt_buf[10]
 
   /**
    * base
@@ -431,17 +429,17 @@ void m01100s (u32 w[16], const u32 pw_len, __global pw_t *pws, __global kernel_r
     w0_t[1] = b;
     w0_t[2] = c;
     w0_t[3] = d;
-    w1_t[0] = salt_buf0;
-    w1_t[1] = salt_buf1;
-    w1_t[2] = salt_buf2;
-    w1_t[3] = salt_buf3;
-    w2_t[0] = salt_buf4;
-    w2_t[1] = salt_buf5;
-    w2_t[2] = salt_buf6;
-    w2_t[3] = salt_buf7;
-    w3_t[0] = salt_buf8;
-    w3_t[1] = salt_buf9;
-    w3_t[2] = (16 + salt_len) * 8;
+    w1_t[0] = salt_buf00;
+    w1_t[1] = salt_buf01;
+    w1_t[2] = salt_buf02;
+    w1_t[3] = salt_buf03;
+    w2_t[0] = salt_buf04;
+    w2_t[1] = salt_buf05;
+    w2_t[2] = salt_buf06;
+    w2_t[3] = salt_buf07;
+    w3_t[0] = salt_buf08;
+    w3_t[1] = salt_buf09;
+    w3_t[2] = salt_buf10;
     w3_t[3] = 0;
 
     a = MD4M_A;
@@ -514,8 +512,7 @@ __kernel void m01100_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
    */
 
   const u32 gid = get_global_id (0);
-
-  if (gid >= gid_max) return;
+  const u32 lid = get_local_id (0);
 
   u32 w[16];
 
@@ -538,11 +535,28 @@ __kernel void m01100_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
 
   const u32 pw_len = pws[gid].pw_len;
 
+  /**
+   * salt
+   */
+
+  __local salt_t s_salt_buf[1];
+
+  if (lid == 0)
+  {
+    s_salt_buf[0] = salt_bufs[salt_pos];
+
+    s_salt_buf[0].salt_buf[10] = (16 + s_salt_buf[0].salt_len) * 8;
+  }
+
+  barrier (CLK_LOCAL_MEM_FENCE);
+
+  if (gid >= gid_max) return;
+
   /**
    * main
    */
 
-  m01100m (w, pw_len, pws, rules_buf, combs_buf, words_buf_r, 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, il_cnt, digests_cnt, digests_offset);
+  m01100m (s_salt_buf, w, pw_len, pws, rules_buf, combs_buf, words_buf_r, 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, il_cnt, digests_cnt, digests_offset);
 }
 
 __kernel void m01100_m08 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32x * words_buf_r, __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 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
@@ -552,8 +566,7 @@ __kernel void m01100_m08 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
    */
 
   const u32 gid = get_global_id (0);
-
-  if (gid >= gid_max) return;
+  const u32 lid = get_local_id (0);
 
   u32 w[16];
 
@@ -576,11 +589,28 @@ __kernel void m01100_m08 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
 
   const u32 pw_len = pws[gid].pw_len;
 
+  /**
+   * salt
+   */
+
+  __local salt_t s_salt_buf[1];
+
+  if (lid == 0)
+  {
+    s_salt_buf[0] = salt_bufs[salt_pos];
+
+    s_salt_buf[0].salt_buf[10] = (16 + s_salt_buf[0].salt_len) * 8;
+  }
+
+  barrier (CLK_LOCAL_MEM_FENCE);
+
+  if (gid >= gid_max) return;
+
   /**
    * main
    */
 
-  m01100m (w, pw_len, pws, rules_buf, combs_buf, words_buf_r, 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, il_cnt, digests_cnt, digests_offset);
+  m01100m (s_salt_buf, w, pw_len, pws, rules_buf, combs_buf, words_buf_r, 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, il_cnt, digests_cnt, digests_offset);
 }
 
 __kernel void m01100_m16 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32x * words_buf_r, __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 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
@@ -594,8 +624,7 @@ __kernel void m01100_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
    */
 
   const u32 gid = get_global_id (0);
-
-  if (gid >= gid_max) return;
+  const u32 lid = get_local_id (0);
 
   u32 w[16];
 
@@ -618,11 +647,28 @@ __kernel void m01100_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
 
   const u32 pw_len = pws[gid].pw_len;
 
+  /**
+   * salt
+   */
+
+  __local salt_t s_salt_buf[1];
+
+  if (lid == 0)
+  {
+    s_salt_buf[0] = salt_bufs[salt_pos];
+
+    s_salt_buf[0].salt_buf[10] = (16 + s_salt_buf[0].salt_len) * 8;
+  }
+
+  barrier (CLK_LOCAL_MEM_FENCE);
+
+  if (gid >= gid_max) return;
+
   /**
    * main
    */
 
-  m01100s (w, pw_len, pws, rules_buf, combs_buf, words_buf_r, 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, il_cnt, digests_cnt, digests_offset);
+  m01100s (s_salt_buf, w, pw_len, pws, rules_buf, combs_buf, words_buf_r, 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, il_cnt, digests_cnt, digests_offset);
 }
 
 __kernel void m01100_s08 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32x * words_buf_r, __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 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
@@ -632,8 +678,7 @@ __kernel void m01100_s08 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
    */
 
   const u32 gid = get_global_id (0);
-
-  if (gid >= gid_max) return;
+  const u32 lid = get_local_id (0);
 
   u32 w[16];
 
@@ -656,11 +701,28 @@ __kernel void m01100_s08 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
 
   const u32 pw_len = pws[gid].pw_len;
 
+  /**
+   * salt
+   */
+
+  __local salt_t s_salt_buf[1];
+
+  if (lid == 0)
+  {
+    s_salt_buf[0] = salt_bufs[salt_pos];
+
+    s_salt_buf[0].salt_buf[10] = (16 + s_salt_buf[0].salt_len) * 8;
+  }
+
+  barrier (CLK_LOCAL_MEM_FENCE);
+
+  if (gid >= gid_max) return;
+
   /**
    * main
    */
 
-  m01100s (w, pw_len, pws, rules_buf, combs_buf, words_buf_r, 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, il_cnt, digests_cnt, digests_offset);
+  m01100s (s_salt_buf, w, pw_len, pws, rules_buf, combs_buf, words_buf_r, 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, il_cnt, digests_cnt, digests_offset);
 }
 
 __kernel void m01100_s16 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32x * words_buf_r, __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 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
-- 
2.25.1