Upgrade kernel to support dynamic local work sizes

[hashcat.git] / OpenCL / m05600_a0.cl

diff --git a/OpenCL/m05600_a0.cl b/OpenCL/m05600_a0.cl
index c35fe3b..cc3f5a5 100644 (file)
--- a/OpenCL/m05600_a0.cl
+++ b/OpenCL/m05600_a0.cl
@@ -286,19 +286,47 @@ static void hmac_md5_run (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], u32 ipad[4
   md5_transform (w0, w1, w2, w3, digest);
 }
 
-__kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m05600_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 netntlm_t *netntlm_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 rules_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void m05600_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 netntlm_t *netntlm_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 rules_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
 {
   /**
    * modifier
    */
 
+  const u32 gid = get_global_id (0);
   const u32 lid = get_local_id (0);
+  const u32 lsz = get_local_size (0);
 
   /**
-   * base
+   * salt
    */
 
-  const u32 gid = get_global_id (0);
+  __local u32 s_userdomain_buf[64];
+
+  for (u32 i = lid; i < 64; i += lsz)
+  {
+    s_userdomain_buf[i] = netntlm_bufs[salt_pos].userdomain_buf[i];
+  }
+
+  __local u32 s_chall_buf[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    s_chall_buf[i] = netntlm_bufs[salt_pos].chall_buf[i];
+  }
+
+  barrier (CLK_LOCAL_MEM_FENCE);
+
+  if (gid >= gid_max) return;
+
+  const u32 userdomain_len = netntlm_bufs[salt_pos].user_len
+                           + netntlm_bufs[salt_pos].domain_len;
+
+  const u32 chall_len = netntlm_bufs[salt_pos].srvchall_len
+                      + netntlm_bufs[salt_pos].clichall_len;
+
+  /**
+   * base
+   */
 
   u32 pw_buf0[4];
 
@@ -316,35 +344,6 @@ __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m05600_m04 (__glo
 
   const u32 pw_len = pws[gid].pw_len;
 
-  /**
-   * salt
-   */
-
-  __local u32 s_userdomain_buf[64];
-  __local u32 s_chall_buf[256];
-
-  const u32 userdomain_len = netntlm_bufs[salt_pos].user_len
-                            + netntlm_bufs[salt_pos].domain_len;
-
-  const u32 chall_len = netntlm_bufs[salt_pos].srvchall_len
-                       + netntlm_bufs[salt_pos].clichall_len;
-
-  if (lid < 64)
-  {
-    s_userdomain_buf[lid] = netntlm_bufs[salt_pos].userdomain_buf[lid];
-  }
-
-  const u32 lid4 = lid * 4;
-
-  s_chall_buf[lid4 + 0] = netntlm_bufs[salt_pos].chall_buf[lid4 + 0];
-  s_chall_buf[lid4 + 1] = netntlm_bufs[salt_pos].chall_buf[lid4 + 1];
-  s_chall_buf[lid4 + 2] = netntlm_bufs[salt_pos].chall_buf[lid4 + 2];
-  s_chall_buf[lid4 + 3] = netntlm_bufs[salt_pos].chall_buf[lid4 + 3];
-
-  barrier (CLK_LOCAL_MEM_FENCE);
-
-  if (gid >= gid_max) return;
-
   /**
    * loop
    */
@@ -547,27 +546,55 @@ __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m05600_m04 (__glo
   }
 }
 
-__kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m05600_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 netntlm_t *netntlm_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 combs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void m05600_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 netntlm_t *netntlm_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 combs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
 {
 }
 
-__kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m05600_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 netntlm_t *netntlm_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 combs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void m05600_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 netntlm_t *netntlm_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 combs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
 {
 }
 
-__kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m05600_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 netntlm_t *netntlm_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 rules_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void m05600_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 netntlm_t *netntlm_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 rules_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
 {
   /**
    * modifier
    */
 
+  const u32 gid = get_global_id (0);
   const u32 lid = get_local_id (0);
+  const u32 lsz = get_local_size (0);
 
   /**
-   * base
+   * salt
    */
 
-  const u32 gid = get_global_id (0);
+  __local u32 s_userdomain_buf[64];
+
+  for (u32 i = lid; i < 64; i += lsz)
+  {
+    s_userdomain_buf[i] = netntlm_bufs[salt_pos].userdomain_buf[i];
+  }
+
+  __local u32 s_chall_buf[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    s_chall_buf[i] = netntlm_bufs[salt_pos].chall_buf[i];
+  }
+
+  barrier (CLK_LOCAL_MEM_FENCE);
+
+  if (gid >= gid_max) return;
+
+  const u32 userdomain_len = netntlm_bufs[salt_pos].user_len
+                           + netntlm_bufs[salt_pos].domain_len;
+
+  const u32 chall_len = netntlm_bufs[salt_pos].srvchall_len
+                      + netntlm_bufs[salt_pos].clichall_len;
+
+  /**
+   * base
+   */
 
   u32 pw_buf0[4];
 
@@ -585,35 +612,6 @@ __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m05600_s04 (__glo
 
   const u32 pw_len = pws[gid].pw_len;
 
-  /**
-   * salt
-   */
-
-  __local u32 s_userdomain_buf[64];
-  __local u32 s_chall_buf[256];
-
-  const u32 userdomain_len = netntlm_bufs[salt_pos].user_len
-                            + netntlm_bufs[salt_pos].domain_len;
-
-  const u32 chall_len = netntlm_bufs[salt_pos].srvchall_len
-                       + netntlm_bufs[salt_pos].clichall_len;
-
-  if (lid < 64)
-  {
-    s_userdomain_buf[lid] = netntlm_bufs[salt_pos].userdomain_buf[lid];
-  }
-
-  const u32 lid4 = lid * 4;
-
-  s_chall_buf[lid4 + 0] = netntlm_bufs[salt_pos].chall_buf[lid4 + 0];
-  s_chall_buf[lid4 + 1] = netntlm_bufs[salt_pos].chall_buf[lid4 + 1];
-  s_chall_buf[lid4 + 2] = netntlm_bufs[salt_pos].chall_buf[lid4 + 2];
-  s_chall_buf[lid4 + 3] = netntlm_bufs[salt_pos].chall_buf[lid4 + 3];
-
-  barrier (CLK_LOCAL_MEM_FENCE);
-
-  if (gid >= gid_max) return;
-
   /**
    * digest
    */
@@ -828,10 +826,10 @@ __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m05600_s04 (__glo
   }
 }
 
-__kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m05600_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 netntlm_t *netntlm_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 combs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void m05600_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 netntlm_t *netntlm_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 combs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
 {
 }
 
-__kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m05600_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 netntlm_t *netntlm_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 combs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void m05600_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 netntlm_t *netntlm_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 combs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
 {
 }