2 * Author......: Jens Steube <jens.steube@gmail.com>
8 #include "include/constants.h"
9 #include "include/kernel_vendor.h"
28 #include "include/kernel_functions.c"
29 #include "types_amd.c"
30 #include "common_amd.c"
31 #include "include/rp_gpu.h"
35 #define VECT_COMPARE_S "check_single_vect1_comp4.c"
36 #define VECT_COMPARE_M "check_multi_vect1_comp4.c"
40 #define VECT_COMPARE_S "check_single_vect2_comp4.c"
41 #define VECT_COMPARE_M "check_multi_vect2_comp4.c"
45 #define VECT_COMPARE_S "check_single_vect4_comp4.c"
46 #define VECT_COMPARE_M "check_multi_vect4_comp4.c"
50 #define SIPROUND(v0,v1,v2,v3) \
52 (v1) = rotl64 ((v1), 13); \
54 (v0) = as_ulong (as_uint2 ((v0)).s10); \
56 (v3) = rotl64 ((v3), 16); \
59 (v3) = rotl64 ((v3), 21); \
62 (v1) = rotl64 ((v1), 17); \
64 (v2) = as_ulong (as_uint2 ((v2)).s10);
66 #define SIPROUND(v0,v1,v2,v3) \
68 (v1) = rotl64 ((v1), 13); \
70 (v0) = rotl64 ((v0), 32); \
72 (v3) = rotl64 ((v3), 16); \
75 (v3) = rotl64 ((v3), 21); \
78 (v1) = rotl64 ((v1), 17); \
80 (v2) = rotl64 ((v2), 32);
83 __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m10100_m04 (__global pw_t *pws, __global gpu_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 rules_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
89 const u32 lid = get_local_id (0);
95 const u32 gid = get_global_id (0);
97 if (gid >= gid_max) return;
101 pw_buf0[0] = pws[gid].i[ 0];
102 pw_buf0[1] = pws[gid].i[ 1];
103 pw_buf0[2] = pws[gid].i[ 2];
104 pw_buf0[3] = pws[gid].i[ 3];
108 pw_buf1[0] = pws[gid].i[ 4];
109 pw_buf1[1] = pws[gid].i[ 5];
110 pw_buf1[2] = pws[gid].i[ 6];
111 pw_buf1[3] = pws[gid].i[ 7];
113 const u32 pw_len = pws[gid].pw_len;
119 u64 v0p = SIPHASHM_0;
120 u64 v1p = SIPHASHM_1;
121 u64 v2p = SIPHASHM_2;
122 u64 v3p = SIPHASHM_3;
124 v0p ^= hl32_to_64 (salt_bufs[salt_pos].salt_buf[1], salt_bufs[salt_pos].salt_buf[0]);
125 v1p ^= hl32_to_64 (salt_bufs[salt_pos].salt_buf[3], salt_bufs[salt_pos].salt_buf[2]);
126 v2p ^= hl32_to_64 (salt_bufs[salt_pos].salt_buf[1], salt_bufs[salt_pos].salt_buf[0]);
127 v3p ^= hl32_to_64 (salt_bufs[salt_pos].salt_buf[3], salt_bufs[salt_pos].salt_buf[2]);
133 for (u32 il_pos = 0; il_pos < rules_cnt; il_pos++)
154 const u32x out_len = apply_rules (rules_buf[il_pos].cmds, &w[0], &w[4], pw_len);
156 u64 *w_ptr = (u64 *) w;
158 w_ptr[out_len / 8] |= (u64) out_len << 56;
168 for (i = 0, j = 0; i <= pw_len; i += 8, j += 2)
170 u64x m = hl32_to_64 (w[j + 1], w[j + 0]);
174 SIPROUND (v0, v1, v2, v3);
175 SIPROUND (v0, v1, v2, v3);
182 SIPROUND (v0, v1, v2, v3);
183 SIPROUND (v0, v1, v2, v3);
184 SIPROUND (v0, v1, v2, v3);
185 SIPROUND (v0, v1, v2, v3);
187 const u64x v = v0 ^ v1 ^ v2 ^ v3;
189 const u32x a = l32_from_64 (v);
190 const u32x b = h32_from_64 (v);
197 #include VECT_COMPARE_M
201 __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m10100_m08 (__global pw_t *pws, __global gpu_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 rules_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
205 __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m10100_m16 (__global pw_t *pws, __global gpu_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 rules_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
209 __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m10100_s04 (__global pw_t *pws, __global gpu_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 rules_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
215 const u32 lid = get_local_id (0);
221 const u32 gid = get_global_id (0);
223 if (gid >= gid_max) return;
227 pw_buf0[0] = pws[gid].i[ 0];
228 pw_buf0[1] = pws[gid].i[ 1];
229 pw_buf0[2] = pws[gid].i[ 2];
230 pw_buf0[3] = pws[gid].i[ 3];
234 pw_buf1[0] = pws[gid].i[ 4];
235 pw_buf1[1] = pws[gid].i[ 5];
236 pw_buf1[2] = pws[gid].i[ 6];
237 pw_buf1[3] = pws[gid].i[ 7];
239 const u32 pw_len = pws[gid].pw_len;
245 u64 v0p = SIPHASHM_0;
246 u64 v1p = SIPHASHM_1;
247 u64 v2p = SIPHASHM_2;
248 u64 v3p = SIPHASHM_3;
250 v0p ^= hl32_to_64 (salt_bufs[salt_pos].salt_buf[1], salt_bufs[salt_pos].salt_buf[0]);
251 v1p ^= hl32_to_64 (salt_bufs[salt_pos].salt_buf[3], salt_bufs[salt_pos].salt_buf[2]);
252 v2p ^= hl32_to_64 (salt_bufs[salt_pos].salt_buf[1], salt_bufs[salt_pos].salt_buf[0]);
253 v3p ^= hl32_to_64 (salt_bufs[salt_pos].salt_buf[3], salt_bufs[salt_pos].salt_buf[2]);
259 const u32 search[4] =
261 digests_buf[digests_offset].digest_buf[DGST_R0],
262 digests_buf[digests_offset].digest_buf[DGST_R1],
263 digests_buf[digests_offset].digest_buf[DGST_R2],
264 digests_buf[digests_offset].digest_buf[DGST_R3]
271 for (u32 il_pos = 0; il_pos < rules_cnt; il_pos++)
292 const u32x out_len = apply_rules (rules_buf[il_pos].cmds, &w[0], &w[4], pw_len);
294 u64 *w_ptr = (u64 *) w;
296 w_ptr[out_len / 8] |= (u64) out_len << 56;
306 for (i = 0, j = 0; i <= pw_len; i += 8, j += 2)
308 u64x m = hl32_to_64 (w[j + 1], w[j + 0]);
312 SIPROUND (v0, v1, v2, v3);
313 SIPROUND (v0, v1, v2, v3);
320 SIPROUND (v0, v1, v2, v3);
321 SIPROUND (v0, v1, v2, v3);
322 SIPROUND (v0, v1, v2, v3);
323 SIPROUND (v0, v1, v2, v3);
325 const u64x v = v0 ^ v1 ^ v2 ^ v3;
327 const u32x a = l32_from_64 (v);
328 const u32x b = h32_from_64 (v);
335 #include VECT_COMPARE_S
339 __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m10100_s08 (__global pw_t *pws, __global gpu_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 rules_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
343 __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m10100_s16 (__global pw_t *pws, __global gpu_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 rules_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)