Prepare NEW_SIMD_MODE for -a 1 kernels

[hashcat.git] / OpenCL / m06100_a3.cl

diff --git a/OpenCL/m06100_a3.cl b/OpenCL/m06100_a3.cl
index ef1face..fbacce5 100644 (file)
--- a/OpenCL/m06100_a3.cl
+++ b/OpenCL/m06100_a3.cl
@@ -1,5 +1,8 @@
 /**
- * Author......: Jens Steube <jens.steube@gmail.com>
+ * Authors.....: Jens Steube <jens.steube@gmail.com>
+ *               Gabriele Gristina <matrix@hashcat.net>
+ *               magnum <john.magnum@hushmail.com>
+ *
  * License.....: MIT
  */
 
@@ -30,6 +33,8 @@
 #define BOX(S,n,i) (u32x) ((S)[(n)][(i).s0], (S)[(n)][(i).s1], (S)[(n)][(i).s2], (S)[(n)][(i).s3])
 #elif VECT_SIZE == 8
 #define BOX(S,n,i) (u32x) ((S)[(n)][(i).s0], (S)[(n)][(i).s1], (S)[(n)][(i).s2], (S)[(n)][(i).s3], (S)[(n)][(i).s4], (S)[(n)][(i).s5], (S)[(n)][(i).s6], (S)[(n)][(i).s7])
+#elif VECT_SIZE == 16
+#define BOX(S,n,i) (u32x) ((S)[(n)][(i).s0], (S)[(n)][(i).s1], (S)[(n)][(i).s2], (S)[(n)][(i).s3], (S)[(n)][(i).s4], (S)[(n)][(i).s5], (S)[(n)][(i).s6], (S)[(n)][(i).s7], (S)[(n)][(i).s8], (S)[(n)][(i).s9], (S)[(n)][(i).sa], (S)[(n)][(i).sb], (S)[(n)][(i).sc], (S)[(n)][(i).sd], (S)[(n)][(i).se], (S)[(n)][(i).sf])
 #endif
 
 __constant u32 Ch[8][256] =
@@ -1128,7 +1133,7 @@ __constant u32 rcl[R + 1] =
 
 // this is a highly optimized that assumes dgst[16] = { 0 }; only reuse of no 2nd transform is needed
 
-static void whirlpool_transform (const u32x w[16], u32x dgst[16], __local u32 s_Ch[8][256], __local u32 s_Cl[8][256])
+static void whirlpool_transform (const u32x w[16], u32x dgst[16], __local u32 (*s_Ch)[256], __local u32 (*s_Cl)[256])
 {
   u32x Kh[8];
   u32x Kl[8];
@@ -1341,7 +1346,7 @@ static void whirlpool_transform (const u32x w[16], u32x dgst[16], __local u32 s_
   dgst[15] = statel[7] ^ w[15];
 }
 
-static void m06100m (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, __local u32 s_Cl[8][256], __local u32 s_Ch[8][256])
+static void m06100m (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, __local u32 (*s_Cl)[256], __local u32 (*s_Ch)[256])
 {
   /**
    * modifier
@@ -1358,7 +1363,7 @@ static void m06100m (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], const u32 pw_le
 
   for (u32 il_pos = 0; il_pos < bfs_cnt; il_pos += VECT_SIZE)
   {
-    const u32x w0r = w0r_create_bft (bfs_buf, il_pos);
+    const u32x w0r = ix_create_bft (bfs_buf, il_pos);
 
     const u32x w0lr = w0l | w0r;
 
@@ -1389,7 +1394,7 @@ static void m06100m (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], const u32 pw_le
   }
 }
 
-static void m06100s (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, __local u32 s_Cl[8][256], __local u32 s_Ch[8][256])
+static void m06100s (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, __local u32 (*s_Cl)[256], __local u32 (*s_Ch)[256])
 {
   /**
    * modifier
@@ -1418,7 +1423,7 @@ static void m06100s (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], const u32 pw_le
 
   for (u32 il_pos = 0; il_pos < bfs_cnt; il_pos += VECT_SIZE)
   {
-    const u32x w0r = w0r_create_bft (bfs_buf, il_pos);
+    const u32x w0r = ix_create_bft (bfs_buf, il_pos);
 
     const u32x w0lr = w0l | w0r;