#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"
-#ifdef VECT_SIZE1
-#define COMPARE_M "check_multi_vect1_comp4.c"
-#endif
+#define COMPARE_S "OpenCL/check_single_comp4.c"
+#define COMPARE_M "OpenCL/check_multi_comp4.c"
__constant u64 k_sha512[80] =
{
SHA512C4c, SHA512C4d, SHA512C4e, SHA512C4f,
};
-static void sha512_transform (const u64 w[16], u64 dgst[8])
+void sha512_transform (const u64 w[16], u64 dgst[8])
{
u64 a = dgst[0];
u64 b = dgst[1];
ROUND_STEP (0);
- //#pragma unroll // kernel fails if used
+ //#pragma unroll
for (int i = 16; i < 80; i += 16)
{
ROUND_EXPAND (); ROUND_STEP (i);
dgst[7] += h;
}
-static void hmac_run (const u64 w1[16], const u64 ipad[8], const u64 opad[8], u64 dgst[8])
+void hmac_run (const u64 w1[16], const u64 ipad[8], const u64 opad[8], u64 dgst[8])
{
dgst[0] = ipad[0];
dgst[1] = ipad[1];
sha512_transform (w, dgst);
}
-static void hmac_init (u64 w[16], u64 ipad[8], u64 opad[8])
+void hmac_init (u64 w[16], u64 ipad[8], u64 opad[8])
{
w[ 0] ^= 0x3636363636363636;
w[ 1] ^= 0x3636363636363636;
sha512_transform (w, opad);
}
-__kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m07100_init (__global pw_t *pws, __global gpu_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global pbkdf2_sha512_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 pbkdf2_sha512_t *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 m07100_init (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global pbkdf2_sha512_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 pbkdf2_sha512_t *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 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
{
/**
* base
u32 w0[4];
- w0[0] = swap_workaround (pws[gid].i[ 0]);
- w0[1] = swap_workaround (pws[gid].i[ 1]);
- w0[2] = swap_workaround (pws[gid].i[ 2]);
- w0[3] = swap_workaround (pws[gid].i[ 3]);
+ w0[0] = swap32 (pws[gid].i[ 0]);
+ w0[1] = swap32 (pws[gid].i[ 1]);
+ w0[2] = swap32 (pws[gid].i[ 2]);
+ w0[3] = swap32 (pws[gid].i[ 3]);
u32 w1[4];
- w1[0] = swap_workaround (pws[gid].i[ 4]);
- w1[1] = swap_workaround (pws[gid].i[ 5]);
- w1[2] = swap_workaround (pws[gid].i[ 6]);
- w1[3] = swap_workaround (pws[gid].i[ 7]);
+ w1[0] = swap32 (pws[gid].i[ 4]);
+ w1[1] = swap32 (pws[gid].i[ 5]);
+ w1[2] = swap32 (pws[gid].i[ 6]);
+ w1[3] = swap32 (pws[gid].i[ 7]);
u32 w2[4];
- w2[0] = swap_workaround (pws[gid].i[ 8]);
- w2[1] = swap_workaround (pws[gid].i[ 9]);
- w2[2] = swap_workaround (pws[gid].i[10]);
- w2[3] = swap_workaround (pws[gid].i[11]);
+ w2[0] = swap32 (pws[gid].i[ 8]);
+ w2[1] = swap32 (pws[gid].i[ 9]);
+ w2[2] = swap32 (pws[gid].i[10]);
+ w2[3] = swap32 (pws[gid].i[11]);
u32 w3[4];
- w3[0] = swap_workaround (pws[gid].i[12]);
- w3[1] = swap_workaround (pws[gid].i[13]);
- w3[2] = swap_workaround (pws[gid].i[14]);
- w3[3] = swap_workaround (pws[gid].i[15]);
+ w3[0] = swap32 (pws[gid].i[12]);
+ w3[1] = swap32 (pws[gid].i[13]);
+ w3[2] = swap32 (pws[gid].i[14]);
+ w3[3] = swap32 (pws[gid].i[15]);
/**
* salt
u32 salt_len = salt_bufs[salt_pos].salt_len;
- esalt_buf[ 0] = hl32_to_64 (swap_workaround (esalt_bufs[salt_pos].salt_buf[ 0]), swap_workaround (esalt_bufs[salt_pos].salt_buf[ 1]));
- esalt_buf[ 1] = hl32_to_64 (swap_workaround (esalt_bufs[salt_pos].salt_buf[ 2]), swap_workaround (esalt_bufs[salt_pos].salt_buf[ 3]));
- esalt_buf[ 2] = hl32_to_64 (swap_workaround (esalt_bufs[salt_pos].salt_buf[ 4]), swap_workaround (esalt_bufs[salt_pos].salt_buf[ 5]));
- esalt_buf[ 3] = hl32_to_64 (swap_workaround (esalt_bufs[salt_pos].salt_buf[ 6]), swap_workaround (esalt_bufs[salt_pos].salt_buf[ 7]));
- esalt_buf[ 4] = hl32_to_64 (swap_workaround (esalt_bufs[salt_pos].salt_buf[ 8]), swap_workaround (esalt_bufs[salt_pos].salt_buf[ 9]));
- esalt_buf[ 5] = hl32_to_64 (swap_workaround (esalt_bufs[salt_pos].salt_buf[10]), swap_workaround (esalt_bufs[salt_pos].salt_buf[11]));
- esalt_buf[ 6] = hl32_to_64 (swap_workaround (esalt_bufs[salt_pos].salt_buf[12]), swap_workaround (esalt_bufs[salt_pos].salt_buf[13]));
- esalt_buf[ 7] = hl32_to_64 (swap_workaround (esalt_bufs[salt_pos].salt_buf[14]), swap_workaround (esalt_bufs[salt_pos].salt_buf[15]));
- esalt_buf[ 8] = hl32_to_64 (swap_workaround (esalt_bufs[salt_pos].salt_buf[16]), swap_workaround (esalt_bufs[salt_pos].salt_buf[17]));
- esalt_buf[ 9] = hl32_to_64 (swap_workaround (esalt_bufs[salt_pos].salt_buf[18]), swap_workaround (esalt_bufs[salt_pos].salt_buf[19]));
- esalt_buf[10] = hl32_to_64 (swap_workaround (esalt_bufs[salt_pos].salt_buf[20]), swap_workaround (esalt_bufs[salt_pos].salt_buf[21]));
- esalt_buf[11] = hl32_to_64 (swap_workaround (esalt_bufs[salt_pos].salt_buf[22]), swap_workaround (esalt_bufs[salt_pos].salt_buf[23]));
- esalt_buf[12] = hl32_to_64 (swap_workaround (esalt_bufs[salt_pos].salt_buf[24]), swap_workaround (esalt_bufs[salt_pos].salt_buf[25]));
- esalt_buf[13] = hl32_to_64 (swap_workaround (esalt_bufs[salt_pos].salt_buf[26]), swap_workaround (esalt_bufs[salt_pos].salt_buf[27]));
+ esalt_buf[ 0] = hl32_to_64 (swap32 (esalt_bufs[salt_pos].salt_buf[ 0]), swap32 (esalt_bufs[salt_pos].salt_buf[ 1]));
+ esalt_buf[ 1] = hl32_to_64 (swap32 (esalt_bufs[salt_pos].salt_buf[ 2]), swap32 (esalt_bufs[salt_pos].salt_buf[ 3]));
+ esalt_buf[ 2] = hl32_to_64 (swap32 (esalt_bufs[salt_pos].salt_buf[ 4]), swap32 (esalt_bufs[salt_pos].salt_buf[ 5]));
+ esalt_buf[ 3] = hl32_to_64 (swap32 (esalt_bufs[salt_pos].salt_buf[ 6]), swap32 (esalt_bufs[salt_pos].salt_buf[ 7]));
+ esalt_buf[ 4] = hl32_to_64 (swap32 (esalt_bufs[salt_pos].salt_buf[ 8]), swap32 (esalt_bufs[salt_pos].salt_buf[ 9]));
+ esalt_buf[ 5] = hl32_to_64 (swap32 (esalt_bufs[salt_pos].salt_buf[10]), swap32 (esalt_bufs[salt_pos].salt_buf[11]));
+ esalt_buf[ 6] = hl32_to_64 (swap32 (esalt_bufs[salt_pos].salt_buf[12]), swap32 (esalt_bufs[salt_pos].salt_buf[13]));
+ esalt_buf[ 7] = hl32_to_64 (swap32 (esalt_bufs[salt_pos].salt_buf[14]), swap32 (esalt_bufs[salt_pos].salt_buf[15]));
+ esalt_buf[ 8] = hl32_to_64 (swap32 (esalt_bufs[salt_pos].salt_buf[16]), swap32 (esalt_bufs[salt_pos].salt_buf[17]));
+ esalt_buf[ 9] = hl32_to_64 (swap32 (esalt_bufs[salt_pos].salt_buf[18]), swap32 (esalt_bufs[salt_pos].salt_buf[19]));
+ esalt_buf[10] = hl32_to_64 (swap32 (esalt_bufs[salt_pos].salt_buf[20]), swap32 (esalt_bufs[salt_pos].salt_buf[21]));
+ esalt_buf[11] = hl32_to_64 (swap32 (esalt_bufs[salt_pos].salt_buf[22]), swap32 (esalt_bufs[salt_pos].salt_buf[23]));
+ esalt_buf[12] = hl32_to_64 (swap32 (esalt_bufs[salt_pos].salt_buf[24]), swap32 (esalt_bufs[salt_pos].salt_buf[25]));
+ esalt_buf[13] = hl32_to_64 (swap32 (esalt_bufs[salt_pos].salt_buf[26]), swap32 (esalt_bufs[salt_pos].salt_buf[27]));
esalt_buf[14] = 0;
esalt_buf[15] = (128 + salt_len + 4) * 8;
}
}
-__kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m07100_loop (__global pw_t *pws, __global gpu_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global pbkdf2_sha512_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 pbkdf2_sha512_t *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 m07100_loop (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global pbkdf2_sha512_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 pbkdf2_sha512_t *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 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);
}
}
-__kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m07100_comp (__global pw_t *pws, __global gpu_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global pbkdf2_sha512_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 pbkdf2_sha512_t *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 m07100_comp (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global pbkdf2_sha512_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 pbkdf2_sha512_t *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 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
{
/**
* base
projects / hashcat.git / blobdiff