#define _BCRYPT_
-#include "include/constants.h"
-#include "include/kernel_vendor.h"
+#include "inc_vendor.cl"
+#include "inc_hash_constants.h"
+#include "inc_hash_functions.cl"
+#include "inc_types.cl"
+#include "inc_common.cl"
-#define DGST_R0 0
-#define DGST_R1 1
-#define DGST_R2 2
-#define DGST_R3 3
-
-#include "include/kernel_functions.c"
-#include "types_ocl.c"
-#include "common.c"
-
-#ifdef VECT_SIZE1
-#define COMPARE_M "check_multi_vect1_comp4.c"
-#endif
+#define COMPARE_S "inc_comp_single.cl"
+#define COMPARE_M "inc_comp_multi.cl"
// http://www.schneier.com/code/constants.txt
0xb74e6132, 0xce77e25b, 0x578fdfe3, 0x3ac372e6
};
-__constant u32 c_pbox[18] =
-{
- 0x243f6a88, 0x85a308d3, 0x13198a2e, 0x03707344,
- 0xa4093822, 0x299f31d0, 0x082efa98, 0xec4e6c89,
- 0x452821e6, 0x38d01377, 0xbe5466cf, 0x34e90c6c,
- 0xc0ac29b7, 0xc97c50dd, 0x3f84d5b5, 0xb5470917,
- 0x9216d5d9, 0x8979fb1b
-};
+#ifdef IS_AMD
+#define BF_ROUND(L,R,N) \
+{ \
+ uchar4 c = as_uchar4 ((L)); \
+ \
+ u32 tmp; \
+ \
+ tmp = S0[c.s3]; \
+ tmp += S1[c.s2]; \
+ tmp ^= S2[c.s1]; \
+ tmp += S3[c.s0]; \
+ \
+ (R) ^= tmp ^ P[(N)]; \
+}
+#endif
+#ifdef IS_NV
+#define BF_ROUND(L,R,N) \
+{ \
+ u32 tmp; \
+ \
+ tmp = S0[__bfe ((L), 24, 8)]; \
+ tmp += S1[__bfe ((L), 16, 8)]; \
+ tmp ^= S2[__bfe ((L), 8, 8)]; \
+ tmp += S3[__bfe ((L), 0, 8)]; \
+ \
+ (R) ^= tmp ^ P[(N)]; \
+}
+#endif
+
+#ifdef IS_GENERIC
#define BF_ROUND(L,R,N) \
{ \
uchar4 c = as_uchar4 ((L)); \
\
- u32 tmp; \
+ u32 tmp; \
\
tmp = S0[c.s3]; \
tmp += S1[c.s2]; \
\
(R) ^= tmp ^ P[(N)]; \
}
+#endif
#define BF_ENCRYPT(L,R) \
{ \
L ^= P[17]; \
}
-static void expand_key (u32 E[34], const u32 W[16], const u32 len)
+void expand_key (u32 E[18], const u32 W[16], const u32 len)
{
u8 *E_cur = (u8 *) E;
u8 *E_stop = E_cur + 72;
}
}
-__kernel void __attribute__((reqd_work_group_size (8, 1, 1))) m03200_init (__global pw_t *pws, __global gpu_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global bcrypt_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 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 rules_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void __attribute__((reqd_work_group_size (8, 1, 1))) m03200_init (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global bcrypt_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 void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_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
*/
const u32 gid = get_global_id (0);
+ const u32 lid = get_local_id (0);
if (gid >= gid_max) return;
- const u32 lid = get_local_id (0);
-
const u32 pw_len = pws[gid].pw_len;
u32 w[16];
w[14] = pws[gid].i[14];
w[15] = pws[gid].i[15];
- u32 E[34];
+ u32 E[18];
expand_key (E, w, pw_len);
E[16] = swap32 (E[16]);
E[17] = swap32 (E[17]);
+ for (u32 i = 0; i < 18; i++)
+ {
+ tmps[gid].E[i] = E[i];
+ }
+
/**
* salt
*/
// initstate
- u32 P[18];
-
- for (u32 i = 0; i < 18; i++)
+ u32 P[18] =
{
- P[i] = c_pbox[i];
- }
+ 0x243f6a88, 0x85a308d3, 0x13198a2e, 0x03707344,
+ 0xa4093822, 0x299f31d0, 0x082efa98, 0xec4e6c89,
+ 0x452821e6, 0x38d01377, 0xbe5466cf, 0x34e90c6c,
+ 0xc0ac29b7, 0xc97c50dd, 0x3f84d5b5, 0xb5470917,
+ 0x9216d5d9, 0x8979fb1b
+ };
for (u32 i = 0; i < 256; i++)
{
}
}
-__kernel void __attribute__((reqd_work_group_size (8, 1, 1))) m03200_loop (__global pw_t *pws, __global gpu_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global bcrypt_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 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 rules_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void __attribute__((reqd_work_group_size (8, 1, 1))) m03200_loop (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global bcrypt_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 void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_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
*/
const u32 gid = get_global_id (0);
-
- if (gid >= gid_max) return;
-
const u32 lid = get_local_id (0);
- const u32 pw_len = pws[gid].pw_len;
-
- u32 w[16];
-
- w[ 0] = pws[gid].i[ 0];
- w[ 1] = pws[gid].i[ 1];
- w[ 2] = pws[gid].i[ 2];
- w[ 3] = pws[gid].i[ 3];
- w[ 4] = pws[gid].i[ 4];
- w[ 5] = pws[gid].i[ 5];
- w[ 6] = pws[gid].i[ 6];
- w[ 7] = pws[gid].i[ 7];
- w[ 8] = pws[gid].i[ 8];
- w[ 9] = pws[gid].i[ 9];
- w[10] = pws[gid].i[10];
- w[11] = pws[gid].i[11];
- w[12] = pws[gid].i[12];
- w[13] = pws[gid].i[13];
- w[14] = pws[gid].i[14];
- w[15] = pws[gid].i[15];
-
- u32 E[34];
+ if (gid >= gid_max) return;
- expand_key (E, w, pw_len);
+ // load
- E[ 0] = swap32 (E[ 0]);
- E[ 1] = swap32 (E[ 1]);
- E[ 2] = swap32 (E[ 2]);
- E[ 3] = swap32 (E[ 3]);
- E[ 4] = swap32 (E[ 4]);
- E[ 5] = swap32 (E[ 5]);
- E[ 6] = swap32 (E[ 6]);
- E[ 7] = swap32 (E[ 7]);
- E[ 8] = swap32 (E[ 8]);
- E[ 9] = swap32 (E[ 9]);
- E[10] = swap32 (E[10]);
- E[11] = swap32 (E[11]);
- E[12] = swap32 (E[12]);
- E[13] = swap32 (E[13]);
- E[14] = swap32 (E[14]);
- E[15] = swap32 (E[15]);
- E[16] = swap32 (E[16]);
- E[17] = swap32 (E[17]);
+ u32 E[18];
- // load
+ for (u32 i = 0; i < 18; i++)
+ {
+ E[i] = tmps[gid].E[i];
+ }
u32 P[18];
L0 = 0;
R0 = 0;
+ #ifdef _unroll
+ #pragma unroll
+ #endif
for (u32 i = 0; i < 9; i++)
{
BF_ENCRYPT (L0, R0);
L0 = 0;
R0 = 0;
+ #ifdef _unroll
+ #pragma unroll
+ #endif
for (u32 i = 0; i < 9; i++)
{
BF_ENCRYPT (L0, R0);
}
}
-__kernel void __attribute__((reqd_work_group_size (8, 1, 1))) m03200_comp (__global pw_t *pws, __global gpu_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global bcrypt_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 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 rules_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void __attribute__((reqd_work_group_size (8, 1, 1))) m03200_comp (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global bcrypt_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 void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_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
*/
const u32 gid = get_global_id (0);
+ const u32 lid = get_local_id (0);
if (gid >= gid_max) return;
- const u32 lid = get_local_id (0);
-
// load
u32 P[18];
projects / hashcat.git / blobdiff