Remove some unroll to reduce register pressure

[hashcat.git] / OpenCL / m06500.cl

diff --git a/OpenCL/m06500.cl b/OpenCL/m06500.cl
index 2c556cb..44e4d7f 100644 (file)
--- a/OpenCL/m06500.cl
+++ b/OpenCL/m06500.cl
@@ -14,11 +14,11 @@
 #define DGST_R3 3
 
 #include "include/kernel_functions.c"
-#include "types_ocl.c"
-#include "common.c"
+#include "OpenCL/types_ocl.c"
+#include "OpenCL/common.c"
 
-#define COMPARE_S "check_single_comp4.c"
-#define COMPARE_M "check_multi_comp4.c"
+#define COMPARE_S "OpenCL/check_single_comp4.c"
+#define COMPARE_M "OpenCL/check_multi_comp4.c"
 
 __constant u64 k_sha512[80] =
 {
@@ -44,7 +44,7 @@ __constant u64 k_sha512[80] =
   SHA512C4c, SHA512C4d, SHA512C4e, SHA512C4f,
 };
 
-static void sha512_transform (const u64 w0[4], const u64 w1[4], const u64 w2[4], const u64 w3[4], u64 digest[8])
+void sha512_transform (const u64 w0[4], const u64 w1[4], const u64 w2[4], const u64 w3[4], u64 digest[8])
 {
   u64 a = digest[0];
   u64 b = digest[1];
@@ -114,7 +114,7 @@ static void sha512_transform (const u64 w0[4], const u64 w1[4], const u64 w2[4],
 
   ROUND_STEP (0);
 
-  //#pragma unroll // kernel fails if used
+  //#pragma unroll
   for (int i = 16; i < 80; i += 16)
   {
     ROUND_EXPAND (); ROUND_STEP (i);
@@ -130,7 +130,7 @@ static void sha512_transform (const u64 w0[4], const u64 w1[4], const u64 w2[4],
   digest[7] += h;
 }
 
-static void hmac_sha512_pad (u64 w0[4], u64 w1[4], u64 w2[4], u64 w3[4], u64 ipad[8], u64 opad[8])
+void hmac_sha512_pad (u64 w0[4], u64 w1[4], u64 w2[4], u64 w3[4], u64 ipad[8], u64 opad[8])
 {
   w0[0] = w0[0] ^ 0x3636363636363636;
   w0[1] = w0[1] ^ 0x3636363636363636;
@@ -189,7 +189,7 @@ static void hmac_sha512_pad (u64 w0[4], u64 w1[4], u64 w2[4], u64 w3[4], u64 ipa
   sha512_transform (w0, w1, w2, w3, opad);
 }
 
-static void hmac_sha512_run (u64 w0[4], u64 w1[4], u64 w2[4], u64 w3[4], u64 ipad[8], u64 opad[8], u64 digest[8])
+void hmac_sha512_run (u64 w0[4], u64 w1[4], u64 w2[4], u64 w3[4], u64 ipad[8], u64 opad[8], u64 digest[8])
 {
   digest[0] = ipad[0];
   digest[1] = ipad[1];
@@ -231,7 +231,7 @@ static void hmac_sha512_run (u64 w0[4], u64 w1[4], u64 w2[4], u64 w3[4], u64 ipa
   sha512_transform (w0, w1, w2, w3, digest);
 }
 
-__kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m06500_init (__global pw_t *pws, __global gpu_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global sha512aix_tmp_t *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 wpa_t *wpa_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 m06500_init (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global sha512aix_tmp_t *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 wpa_t *wpa_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)
 {
   /**
    * base
@@ -390,14 +390,14 @@ __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m06500_init (__gl
   w3l[2] = 0;
   w3l[3] = 0;
 
-  w0l[0] = swap32 (w0l[0]);
-  w0l[1] = swap32 (w0l[1]);
-  w0l[2] = swap32 (w0l[2]);
-  w0l[3] = swap32 (w0l[3]);
-  w1l[0] = swap32 (w1l[0]);
-  w1l[1] = swap32 (w1l[1]);
-  w1l[2] = swap32 (w1l[2]);
-  w1l[3] = swap32 (w1l[3]);
+  w0l[0] = swap64 (w0l[0]);
+  w0l[1] = swap64 (w0l[1]);
+  w0l[2] = swap64 (w0l[2]);
+  w0l[3] = swap64 (w0l[3]);
+  w1l[0] = swap64 (w1l[0]);
+  w1l[1] = swap64 (w1l[1]);
+  w1l[2] = swap64 (w1l[2]);
+  w1l[3] = swap64 (w1l[3]);
   w2l[0] = 0;
   w2l[1] = 0;
   w2l[2] = 0;
@@ -430,7 +430,7 @@ __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m06500_init (__gl
   tmps[gid].out[7] = dgst[7];
 }
 
-__kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m06500_loop (__global pw_t *pws, __global gpu_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global sha512aix_tmp_t *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 wpa_t *wpa_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 m06500_loop (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global sha512aix_tmp_t *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 wpa_t *wpa_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)
 {
   const u32 gid = get_global_id (0);
 
@@ -533,7 +533,7 @@ __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m06500_loop (__gl
   tmps[gid].out[7] = out[7];
 }
 
-__kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m06500_comp (__global pw_t *pws, __global gpu_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global sha512aix_tmp_t *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 wpa_t *wpa_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 m06500_comp (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global sha512aix_tmp_t *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 wpa_t *wpa_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)
 {
   /**
    * base