X-Git-Url: https://www.flypig.org.uk/git/?a=blobdiff_plain;f=OpenCL%2Fm01100_a1.cl;h=2e3e7eca5bc9544021c1aaea75d58247418a2348;hb=ed1863c2630bc8194d38587d753dde71a315402a;hp=ebde5819aa14ff0743fc2c808a397edf505a4433;hpb=9390be8594a603f5298c8aefc5bf2ee831d5e539;p=hashcat.git

diff --git a/OpenCL/m01100_a1.cl b/OpenCL/m01100_a1.cl
index ebde581..2e3e7ec 100644
--- a/OpenCL/m01100_a1.cl
+++ b/OpenCL/m01100_a1.cl
@@ -7,34 +7,21 @@
 
 #define NEW_SIMD_CODE
 
-#include "include/constants.h"
-#include "include/kernel_vendor.h"
-
-#define DGST_R0 0
-#define DGST_R1 3
-#define DGST_R2 2
-#define DGST_R3 1
-
-#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"
 
 __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 +41,30 @@ __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 +214,17 @@ __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 +297,12 @@ __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 +322,30 @@ __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 +507,17 @@ __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;