2 * Author......: Jens Steube <jens.steube@gmail.com>
8 #include "include/constants.h"
9 #include "include/kernel_vendor.h"
16 #include "include/kernel_functions.c"
17 #include "types_ocl.c"
19 #include "include/rp_gpu.h"
22 #define COMPARE_S "check_single_comp4.c"
23 #define COMPARE_M "check_multi_comp4.c"
26 #define SIPROUND(v0,v1,v2,v3) \
28 (v1) = rotl64 ((v1), 13); \
30 (v0) = as_ulong (as_uint2 ((v0)).s10); \
32 (v3) = rotl64 ((v3), 16); \
35 (v3) = rotl64 ((v3), 21); \
38 (v1) = rotl64 ((v1), 17); \
40 (v2) = as_ulong (as_uint2 ((v2)).s10);
42 #define SIPROUND(v0,v1,v2,v3) \
44 (v1) = rotl64 ((v1), 13); \
46 (v0) = rotl64 ((v0), 32); \
48 (v3) = rotl64 ((v3), 16); \
51 (v3) = rotl64 ((v3), 21); \
54 (v1) = rotl64 ((v1), 17); \
56 (v2) = rotl64 ((v2), 32);
59 __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)
65 const u32 lid = get_local_id (0);
71 const u32 gid = get_global_id (0);
73 if (gid >= gid_max) return;
77 pw_buf0[0] = pws[gid].i[ 0];
78 pw_buf0[1] = pws[gid].i[ 1];
79 pw_buf0[2] = pws[gid].i[ 2];
80 pw_buf0[3] = pws[gid].i[ 3];
84 pw_buf1[0] = pws[gid].i[ 4];
85 pw_buf1[1] = pws[gid].i[ 5];
86 pw_buf1[2] = pws[gid].i[ 6];
87 pw_buf1[3] = pws[gid].i[ 7];
89 const u32 pw_len = pws[gid].pw_len;
100 v0p ^= hl32_to_64 (salt_bufs[salt_pos].salt_buf[1], salt_bufs[salt_pos].salt_buf[0]);
101 v1p ^= hl32_to_64 (salt_bufs[salt_pos].salt_buf[3], salt_bufs[salt_pos].salt_buf[2]);
102 v2p ^= hl32_to_64 (salt_bufs[salt_pos].salt_buf[1], salt_bufs[salt_pos].salt_buf[0]);
103 v3p ^= hl32_to_64 (salt_bufs[salt_pos].salt_buf[3], salt_bufs[salt_pos].salt_buf[2]);
109 for (u32 il_pos = 0; il_pos < rules_cnt; il_pos++)
130 const u32 out_len = apply_rules (rules_buf[il_pos].cmds, &w[0], &w[4], pw_len);
132 u64 *w_ptr = (u64 *) w;
134 w_ptr[out_len / 8] |= (u64) out_len << 56;
144 for (i = 0, j = 0; i <= pw_len; i += 8, j += 2)
146 u64 m = hl32_to_64 (w[j + 1], w[j + 0]);
150 SIPROUND (v0, v1, v2, v3);
151 SIPROUND (v0, v1, v2, v3);
158 SIPROUND (v0, v1, v2, v3);
159 SIPROUND (v0, v1, v2, v3);
160 SIPROUND (v0, v1, v2, v3);
161 SIPROUND (v0, v1, v2, v3);
163 const u64 v = v0 ^ v1 ^ v2 ^ v3;
165 const u32 a = l32_from_64 (v);
166 const u32 b = h32_from_64 (v);
177 __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)
181 __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)
185 __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)
191 const u32 lid = get_local_id (0);
197 const u32 gid = get_global_id (0);
199 if (gid >= gid_max) return;
203 pw_buf0[0] = pws[gid].i[ 0];
204 pw_buf0[1] = pws[gid].i[ 1];
205 pw_buf0[2] = pws[gid].i[ 2];
206 pw_buf0[3] = pws[gid].i[ 3];
210 pw_buf1[0] = pws[gid].i[ 4];
211 pw_buf1[1] = pws[gid].i[ 5];
212 pw_buf1[2] = pws[gid].i[ 6];
213 pw_buf1[3] = pws[gid].i[ 7];
215 const u32 pw_len = pws[gid].pw_len;
221 u64 v0p = SIPHASHM_0;
222 u64 v1p = SIPHASHM_1;
223 u64 v2p = SIPHASHM_2;
224 u64 v3p = SIPHASHM_3;
226 v0p ^= hl32_to_64 (salt_bufs[salt_pos].salt_buf[1], salt_bufs[salt_pos].salt_buf[0]);
227 v1p ^= hl32_to_64 (salt_bufs[salt_pos].salt_buf[3], salt_bufs[salt_pos].salt_buf[2]);
228 v2p ^= hl32_to_64 (salt_bufs[salt_pos].salt_buf[1], salt_bufs[salt_pos].salt_buf[0]);
229 v3p ^= hl32_to_64 (salt_bufs[salt_pos].salt_buf[3], salt_bufs[salt_pos].salt_buf[2]);
235 const u32 search[4] =
237 digests_buf[digests_offset].digest_buf[DGST_R0],
238 digests_buf[digests_offset].digest_buf[DGST_R1],
239 digests_buf[digests_offset].digest_buf[DGST_R2],
240 digests_buf[digests_offset].digest_buf[DGST_R3]
247 for (u32 il_pos = 0; il_pos < rules_cnt; il_pos++)
268 const u32 out_len = apply_rules (rules_buf[il_pos].cmds, &w[0], &w[4], pw_len);
270 u64 *w_ptr = (u64 *) w;
272 w_ptr[out_len / 8] |= (u64) out_len << 56;
282 for (i = 0, j = 0; i <= pw_len; i += 8, j += 2)
284 u64 m = hl32_to_64 (w[j + 1], w[j + 0]);
288 SIPROUND (v0, v1, v2, v3);
289 SIPROUND (v0, v1, v2, v3);
296 SIPROUND (v0, v1, v2, v3);
297 SIPROUND (v0, v1, v2, v3);
298 SIPROUND (v0, v1, v2, v3);
299 SIPROUND (v0, v1, v2, v3);
301 const u64 v = v0 ^ v1 ^ v2 ^ v3;
303 const u32 a = l32_from_64 (v);
304 const u32 b = h32_from_64 (v);
315 __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)
319 __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)