Upgrade kernel to support dynamic local work sizes

[hashcat.git] / OpenCL / m08000_a3.cl

diff --git a/OpenCL/m08000_a3.cl b/OpenCL/m08000_a3.cl
index 0fb81bf..a5eb3fc 100644 (file)
--- a/OpenCL/m08000_a3.cl
+++ b/OpenCL/m08000_a3.cl
@@ -232,22 +232,7 @@ static void m08000m (__local u32 w_s1[64], __local u32 w_s2[64], u32 w[16], cons
 
   const u32 gid = get_global_id (0);
   const u32 lid = get_local_id (0);
-
-  w[ 1] = w[ 1] >> 8;
-  w[ 2] = w[ 2] >> 8;
-  w[ 3] = w[ 3] >> 8;
-  w[ 4] = w[ 4] >> 8;
-  w[ 5] = w[ 5] >> 8;
-  w[ 6] = w[ 6] >> 8;
-  w[ 7] = w[ 7] >> 8;
-  w[ 8] = w[ 8] >> 8;
-  w[ 9] = w[ 9] >> 8;
-  w[10] = w[10] >> 8;
-  w[11] = w[11] >> 8;
-  w[12] = w[12] >> 8;
-  w[13] = w[13] >> 8;
-  w[14] = w[14] >> 8;
-  w[15] = w[15] >> 8;
+  const u32 lsz = get_local_size (0);
 
   /**
    * salt
@@ -261,8 +246,11 @@ static void m08000m (__local u32 w_s1[64], __local u32 w_s2[64], u32 w[16], cons
    * precompute final msg blocks
    */
 
-  w_s1[lid] = 0;
-  w_s2[lid] = 0;
+  for (u32 i = lid; i < 64; i += lsz)
+  {
+    w_s1[i] = 0;
+    w_s2[i] = 0;
+  }
 
   barrier (CLK_LOCAL_MEM_FENCE);
 
@@ -275,9 +263,7 @@ static void m08000m (__local u32 w_s1[64], __local u32 w_s2[64], u32 w[16], cons
     {
       w_s1[i] = SHA256_EXPAND (w_s1[i - 2], w_s1[i - 7], w_s1[i - 15], w_s1[i - 16]);
     }
-  }
-  else if (lid == 1)
-  {
+
     w_s2[ 0] = salt_buf0 << 16 | salt_buf1 >> 16;
     w_s2[ 1] = salt_buf1 << 16 | salt_buf2 >> 16;
     w_s2[ 2] = salt_buf2 << 16 | 0;
@@ -294,6 +280,26 @@ static void m08000m (__local u32 w_s1[64], __local u32 w_s2[64], u32 w[16], cons
 
   if (gid >= gid_max) return;
 
+  /**
+   * modifier
+   */
+
+  w[ 1] = w[ 1] >> 8;
+  w[ 2] = w[ 2] >> 8;
+  w[ 3] = w[ 3] >> 8;
+  w[ 4] = w[ 4] >> 8;
+  w[ 5] = w[ 5] >> 8;
+  w[ 6] = w[ 6] >> 8;
+  w[ 7] = w[ 7] >> 8;
+  w[ 8] = w[ 8] >> 8;
+  w[ 9] = w[ 9] >> 8;
+  w[10] = w[10] >> 8;
+  w[11] = w[11] >> 8;
+  w[12] = w[12] >> 8;
+  w[13] = w[13] >> 8;
+  w[14] = w[14] >> 8;
+  w[15] = w[15] >> 8;
+
   /**
    * loop
    */
@@ -346,22 +352,7 @@ static void m08000s (__local u32 w_s1[64], __local u32 w_s2[64], u32 w[16], cons
 
   const u32 gid = get_global_id (0);
   const u32 lid = get_local_id (0);
-
-  w[ 1] = w[ 1] >> 8;
-  w[ 2] = w[ 2] >> 8;
-  w[ 3] = w[ 3] >> 8;
-  w[ 4] = w[ 4] >> 8;
-  w[ 5] = w[ 5] >> 8;
-  w[ 6] = w[ 6] >> 8;
-  w[ 7] = w[ 7] >> 8;
-  w[ 8] = w[ 8] >> 8;
-  w[ 9] = w[ 9] >> 8;
-  w[10] = w[10] >> 8;
-  w[11] = w[11] >> 8;
-  w[12] = w[12] >> 8;
-  w[13] = w[13] >> 8;
-  w[14] = w[14] >> 8;
-  w[15] = w[15] >> 8;
+  const u32 lsz = get_local_size (0);
 
   /**
    * salt
@@ -375,8 +366,11 @@ static void m08000s (__local u32 w_s1[64], __local u32 w_s2[64], u32 w[16], cons
    * precompute final msg blocks
    */
 
-  w_s1[lid] = 0;
-  w_s2[lid] = 0;
+  for (u32 i = lid; i < 64; i += lsz)
+  {
+    w_s1[i] = 0;
+    w_s2[i] = 0;
+  }
 
   barrier (CLK_LOCAL_MEM_FENCE);
 
@@ -389,9 +383,7 @@ static void m08000s (__local u32 w_s1[64], __local u32 w_s2[64], u32 w[16], cons
     {
       w_s1[i] = SHA256_EXPAND (w_s1[i - 2], w_s1[i - 7], w_s1[i - 15], w_s1[i - 16]);
     }
-  }
-  else if (lid == 1)
-  {
+
     w_s2[ 0] = salt_buf0 << 16 | salt_buf1 >> 16;
     w_s2[ 1] = salt_buf1 << 16 | salt_buf2 >> 16;
     w_s2[ 2] = salt_buf2 << 16 | 0;
@@ -408,6 +400,26 @@ static void m08000s (__local u32 w_s1[64], __local u32 w_s2[64], u32 w[16], cons
 
   if (gid >= gid_max) return;
 
+  /**
+   * modifier
+   */
+
+  w[ 1] = w[ 1] >> 8;
+  w[ 2] = w[ 2] >> 8;
+  w[ 3] = w[ 3] >> 8;
+  w[ 4] = w[ 4] >> 8;
+  w[ 5] = w[ 5] >> 8;
+  w[ 6] = w[ 6] >> 8;
+  w[ 7] = w[ 7] >> 8;
+  w[ 8] = w[ 8] >> 8;
+  w[ 9] = w[ 9] >> 8;
+  w[10] = w[10] >> 8;
+  w[11] = w[11] >> 8;
+  w[12] = w[12] >> 8;
+  w[13] = w[13] >> 8;
+  w[14] = w[14] >> 8;
+  w[15] = w[15] >> 8;
+
   /**
    * digest
    */
@@ -464,7 +476,7 @@ static void m08000s (__local u32 w_s1[64], __local u32 w_s2[64], u32 w[16], cons
   }
 }
 
-__kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m08000_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32 * 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 bfs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void m08000_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32 * 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 bfs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
 {
   /**
    * base
@@ -503,7 +515,7 @@ __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m08000_m04 (__glo
   m08000m (w_s1, w_s2, 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, bfs_cnt, digests_cnt, digests_offset, gid_max);
 }
 
-__kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m08000_m08 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32 * 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 bfs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void m08000_m08 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32 * 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 bfs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
 {
   /**
    * base
@@ -542,7 +554,7 @@ __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m08000_m08 (__glo
   m08000m (w_s1, w_s2, 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, bfs_cnt, digests_cnt, digests_offset, gid_max);
 }
 
-__kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m08000_m16 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32 * 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 bfs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void m08000_m16 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32 * 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 bfs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
 {
   /**
    * base
@@ -581,7 +593,7 @@ __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m08000_m16 (__glo
   m08000m (w_s1, w_s2, 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, bfs_cnt, digests_cnt, digests_offset, gid_max);
 }
 
-__kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m08000_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32 * 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 bfs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void m08000_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32 * 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 bfs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
 {
   /**
    * base
@@ -620,7 +632,7 @@ __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m08000_s04 (__glo
   m08000s (w_s1, w_s2, 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, bfs_cnt, digests_cnt, digests_offset, gid_max);
 }
 
-__kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m08000_s08 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32 * 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 bfs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void m08000_s08 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32 * 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 bfs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
 {
   /**
    * base
@@ -659,7 +671,7 @@ __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m08000_s08 (__glo
   m08000s (w_s1, w_s2, 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, bfs_cnt, digests_cnt, digests_offset, gid_max);
 }
 
-__kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m08000_s16 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32 * 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 bfs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void m08000_s16 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32 * 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 bfs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
 {
   /**
    * base