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 "OpenCL/types_ocl.c"
18 #include "OpenCL/common.c"
19 #include "include/rp_gpu.h"
22 #define COMPARE_S "OpenCL/check_single_comp4.c"
23 #define COMPARE_M "OpenCL/check_multi_comp4.c"
25 #define SIPROUND(v0,v1,v2,v3) \
27 (v1) = rotl64 ((v1), 13); \
29 (v0) = as_ulong (as_uint2 ((v0)).s10); \
31 (v3) = rotl64 ((v3), 16); \
34 (v3) = rotl64 ((v3), 21); \
37 (v1) = rotl64 ((v1), 17); \
39 (v2) = as_ulong (as_uint2 ((v2)).s10);
41 __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)
47 const u32 lid = get_local_id (0);
53 const u32 gid = get_global_id (0);
55 if (gid >= gid_max) return;
59 pw_buf0[0] = pws[gid].i[ 0];
60 pw_buf0[1] = pws[gid].i[ 1];
61 pw_buf0[2] = pws[gid].i[ 2];
62 pw_buf0[3] = pws[gid].i[ 3];
66 pw_buf1[0] = pws[gid].i[ 4];
67 pw_buf1[1] = pws[gid].i[ 5];
68 pw_buf1[2] = pws[gid].i[ 6];
69 pw_buf1[3] = pws[gid].i[ 7];
71 const u32 pw_len = pws[gid].pw_len;
82 v0p ^= hl32_to_64 (salt_bufs[salt_pos].salt_buf[1], salt_bufs[salt_pos].salt_buf[0]);
83 v1p ^= hl32_to_64 (salt_bufs[salt_pos].salt_buf[3], salt_bufs[salt_pos].salt_buf[2]);
84 v2p ^= hl32_to_64 (salt_bufs[salt_pos].salt_buf[1], salt_bufs[salt_pos].salt_buf[0]);
85 v3p ^= hl32_to_64 (salt_bufs[salt_pos].salt_buf[3], salt_bufs[salt_pos].salt_buf[2]);
91 for (u32 il_pos = 0; il_pos < rules_cnt; il_pos++)
112 const u32 out_len = apply_rules (rules_buf[il_pos].cmds, &w[0], &w[4], pw_len);
114 u64 *w_ptr = (u64 *) w;
116 w_ptr[out_len / 8] |= (u64) out_len << 56;
126 for (i = 0, j = 0; i <= pw_len; i += 8, j += 2)
128 u64 m = hl32_to_64 (w[j + 1], w[j + 0]);
132 SIPROUND (v0, v1, v2, v3);
133 SIPROUND (v0, v1, v2, v3);
140 SIPROUND (v0, v1, v2, v3);
141 SIPROUND (v0, v1, v2, v3);
142 SIPROUND (v0, v1, v2, v3);
143 SIPROUND (v0, v1, v2, v3);
145 const u64 v = v0 ^ v1 ^ v2 ^ v3;
147 const u32 a = l32_from_64 (v);
148 const u32 b = h32_from_64 (v);
159 __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)
163 __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)
167 __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)
173 const u32 lid = get_local_id (0);
179 const u32 gid = get_global_id (0);
181 if (gid >= gid_max) return;
185 pw_buf0[0] = pws[gid].i[ 0];
186 pw_buf0[1] = pws[gid].i[ 1];
187 pw_buf0[2] = pws[gid].i[ 2];
188 pw_buf0[3] = pws[gid].i[ 3];
192 pw_buf1[0] = pws[gid].i[ 4];
193 pw_buf1[1] = pws[gid].i[ 5];
194 pw_buf1[2] = pws[gid].i[ 6];
195 pw_buf1[3] = pws[gid].i[ 7];
197 const u32 pw_len = pws[gid].pw_len;
203 u64 v0p = SIPHASHM_0;
204 u64 v1p = SIPHASHM_1;
205 u64 v2p = SIPHASHM_2;
206 u64 v3p = SIPHASHM_3;
208 v0p ^= hl32_to_64 (salt_bufs[salt_pos].salt_buf[1], salt_bufs[salt_pos].salt_buf[0]);
209 v1p ^= hl32_to_64 (salt_bufs[salt_pos].salt_buf[3], salt_bufs[salt_pos].salt_buf[2]);
210 v2p ^= hl32_to_64 (salt_bufs[salt_pos].salt_buf[1], salt_bufs[salt_pos].salt_buf[0]);
211 v3p ^= hl32_to_64 (salt_bufs[salt_pos].salt_buf[3], salt_bufs[salt_pos].salt_buf[2]);
217 const u32 search[4] =
219 digests_buf[digests_offset].digest_buf[DGST_R0],
220 digests_buf[digests_offset].digest_buf[DGST_R1],
221 digests_buf[digests_offset].digest_buf[DGST_R2],
222 digests_buf[digests_offset].digest_buf[DGST_R3]
229 for (u32 il_pos = 0; il_pos < rules_cnt; il_pos++)
250 const u32 out_len = apply_rules (rules_buf[il_pos].cmds, &w[0], &w[4], pw_len);
252 u64 *w_ptr = (u64 *) w;
254 w_ptr[out_len / 8] |= (u64) out_len << 56;
264 for (i = 0, j = 0; i <= pw_len; i += 8, j += 2)
266 u64 m = hl32_to_64 (w[j + 1], w[j + 0]);
270 SIPROUND (v0, v1, v2, v3);
271 SIPROUND (v0, v1, v2, v3);
278 SIPROUND (v0, v1, v2, v3);
279 SIPROUND (v0, v1, v2, v3);
280 SIPROUND (v0, v1, v2, v3);
281 SIPROUND (v0, v1, v2, v3);
283 const u64 v = v0 ^ v1 ^ v2 ^ v3;
285 const u32 a = l32_from_64 (v);
286 const u32 b = h32_from_64 (v);
297 __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)
301 __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)