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"
20 #define COMPARE_S "check_single_comp4.c"
21 #define COMPARE_M "check_multi_comp4.c"
23 __constant u32 crc32tab[0x100] =
25 0x00000000, 0x77073096, 0xee0e612c, 0x990951ba,
26 0x076dc419, 0x706af48f, 0xe963a535, 0x9e6495a3,
27 0x0edb8832, 0x79dcb8a4, 0xe0d5e91e, 0x97d2d988,
28 0x09b64c2b, 0x7eb17cbd, 0xe7b82d07, 0x90bf1d91,
29 0x1db71064, 0x6ab020f2, 0xf3b97148, 0x84be41de,
30 0x1adad47d, 0x6ddde4eb, 0xf4d4b551, 0x83d385c7,
31 0x136c9856, 0x646ba8c0, 0xfd62f97a, 0x8a65c9ec,
32 0x14015c4f, 0x63066cd9, 0xfa0f3d63, 0x8d080df5,
33 0x3b6e20c8, 0x4c69105e, 0xd56041e4, 0xa2677172,
34 0x3c03e4d1, 0x4b04d447, 0xd20d85fd, 0xa50ab56b,
35 0x35b5a8fa, 0x42b2986c, 0xdbbbc9d6, 0xacbcf940,
36 0x32d86ce3, 0x45df5c75, 0xdcd60dcf, 0xabd13d59,
37 0x26d930ac, 0x51de003a, 0xc8d75180, 0xbfd06116,
38 0x21b4f4b5, 0x56b3c423, 0xcfba9599, 0xb8bda50f,
39 0x2802b89e, 0x5f058808, 0xc60cd9b2, 0xb10be924,
40 0x2f6f7c87, 0x58684c11, 0xc1611dab, 0xb6662d3d,
41 0x76dc4190, 0x01db7106, 0x98d220bc, 0xefd5102a,
42 0x71b18589, 0x06b6b51f, 0x9fbfe4a5, 0xe8b8d433,
43 0x7807c9a2, 0x0f00f934, 0x9609a88e, 0xe10e9818,
44 0x7f6a0dbb, 0x086d3d2d, 0x91646c97, 0xe6635c01,
45 0x6b6b51f4, 0x1c6c6162, 0x856530d8, 0xf262004e,
46 0x6c0695ed, 0x1b01a57b, 0x8208f4c1, 0xf50fc457,
47 0x65b0d9c6, 0x12b7e950, 0x8bbeb8ea, 0xfcb9887c,
48 0x62dd1ddf, 0x15da2d49, 0x8cd37cf3, 0xfbd44c65,
49 0x4db26158, 0x3ab551ce, 0xa3bc0074, 0xd4bb30e2,
50 0x4adfa541, 0x3dd895d7, 0xa4d1c46d, 0xd3d6f4fb,
51 0x4369e96a, 0x346ed9fc, 0xad678846, 0xda60b8d0,
52 0x44042d73, 0x33031de5, 0xaa0a4c5f, 0xdd0d7cc9,
53 0x5005713c, 0x270241aa, 0xbe0b1010, 0xc90c2086,
54 0x5768b525, 0x206f85b3, 0xb966d409, 0xce61e49f,
55 0x5edef90e, 0x29d9c998, 0xb0d09822, 0xc7d7a8b4,
56 0x59b33d17, 0x2eb40d81, 0xb7bd5c3b, 0xc0ba6cad,
57 0xedb88320, 0x9abfb3b6, 0x03b6e20c, 0x74b1d29a,
58 0xead54739, 0x9dd277af, 0x04db2615, 0x73dc1683,
59 0xe3630b12, 0x94643b84, 0x0d6d6a3e, 0x7a6a5aa8,
60 0xe40ecf0b, 0x9309ff9d, 0x0a00ae27, 0x7d079eb1,
61 0xf00f9344, 0x8708a3d2, 0x1e01f268, 0x6906c2fe,
62 0xf762575d, 0x806567cb, 0x196c3671, 0x6e6b06e7,
63 0xfed41b76, 0x89d32be0, 0x10da7a5a, 0x67dd4acc,
64 0xf9b9df6f, 0x8ebeeff9, 0x17b7be43, 0x60b08ed5,
65 0xd6d6a3e8, 0xa1d1937e, 0x38d8c2c4, 0x4fdff252,
66 0xd1bb67f1, 0xa6bc5767, 0x3fb506dd, 0x48b2364b,
67 0xd80d2bda, 0xaf0a1b4c, 0x36034af6, 0x41047a60,
68 0xdf60efc3, 0xa867df55, 0x316e8eef, 0x4669be79,
69 0xcb61b38c, 0xbc66831a, 0x256fd2a0, 0x5268e236,
70 0xcc0c7795, 0xbb0b4703, 0x220216b9, 0x5505262f,
71 0xc5ba3bbe, 0xb2bd0b28, 0x2bb45a92, 0x5cb36a04,
72 0xc2d7ffa7, 0xb5d0cf31, 0x2cd99e8b, 0x5bdeae1d,
73 0x9b64c2b0, 0xec63f226, 0x756aa39c, 0x026d930a,
74 0x9c0906a9, 0xeb0e363f, 0x72076785, 0x05005713,
75 0x95bf4a82, 0xe2b87a14, 0x7bb12bae, 0x0cb61b38,
76 0x92d28e9b, 0xe5d5be0d, 0x7cdcefb7, 0x0bdbdf21,
77 0x86d3d2d4, 0xf1d4e242, 0x68ddb3f8, 0x1fda836e,
78 0x81be16cd, 0xf6b9265b, 0x6fb077e1, 0x18b74777,
79 0x88085ae6, 0xff0f6a70, 0x66063bca, 0x11010b5c,
80 0x8f659eff, 0xf862ae69, 0x616bffd3, 0x166ccf45,
81 0xa00ae278, 0xd70dd2ee, 0x4e048354, 0x3903b3c2,
82 0xa7672661, 0xd06016f7, 0x4969474d, 0x3e6e77db,
83 0xaed16a4a, 0xd9d65adc, 0x40df0b66, 0x37d83bf0,
84 0xa9bcae53, 0xdebb9ec5, 0x47b2cf7f, 0x30b5ffe9,
85 0xbdbdf21c, 0xcabac28a, 0x53b39330, 0x24b4a3a6,
86 0xbad03605, 0xcdd70693, 0x54de5729, 0x23d967bf,
87 0xb3667a2e, 0xc4614ab8, 0x5d681b02, 0x2a6f2b94,
88 0xb40bbe37, 0xc30c8ea1, 0x5a05df1b, 0x2d02ef8d
91 static u32 round_crc32 (u32 a, const u32 v)
93 const u32 k = (a ^ v) & 0xff;
102 a.s0 = crc32tab[k.s0];
103 a.s1 = crc32tab[k.s1];
107 a.s0 = crc32tab[k.s0];
108 a.s1 = crc32tab[k.s1];
109 a.s2 = crc32tab[k.s2];
110 a.s3 = crc32tab[k.s3];
118 static u32 crc32 (const u32 w[16], const u32 pw_len, const u32 iv)
122 if (pw_len >= 1) a = round_crc32 (a, w[0] >> 0);
123 if (pw_len >= 2) a = round_crc32 (a, w[0] >> 8);
124 if (pw_len >= 3) a = round_crc32 (a, w[0] >> 16);
125 if (pw_len >= 4) a = round_crc32 (a, w[0] >> 24);
127 for (u32 i = 4, j = 1; i < pw_len; i += 4, j += 1)
129 if (pw_len >= (i + 1)) a = round_crc32 (a, w[j] >> 0);
130 if (pw_len >= (i + 2)) a = round_crc32 (a, w[j] >> 8);
131 if (pw_len >= (i + 3)) a = round_crc32 (a, w[j] >> 16);
132 if (pw_len >= (i + 4)) a = round_crc32 (a, w[j] >> 24);
138 __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m11500_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 combs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
144 const u32 lid = get_local_id (0);
150 const u32 iv = salt_bufs[salt_pos].salt_buf[0];
156 const u32 gid = get_global_id (0);
158 if (gid >= gid_max) return;
162 wordl0[0] = pws[gid].i[ 0];
163 wordl0[1] = pws[gid].i[ 1];
164 wordl0[2] = pws[gid].i[ 2];
165 wordl0[3] = pws[gid].i[ 3];
169 wordl1[0] = pws[gid].i[ 4];
170 wordl1[1] = pws[gid].i[ 5];
171 wordl1[2] = pws[gid].i[ 6];
172 wordl1[3] = pws[gid].i[ 7];
188 const u32 pw_l_len = pws[gid].pw_len;
190 if (combs_mode == COMBINATOR_MODE_BASE_RIGHT)
192 switch_buffer_by_offset (wordl0, wordl1, wordl2, wordl3, combs_buf[0].pw_len);
199 for (u32 il_pos = 0; il_pos < combs_cnt; il_pos++)
201 const u32 pw_r_len = combs_buf[il_pos].pw_len;
203 const u32 pw_len = pw_l_len + pw_r_len;
207 wordr0[0] = combs_buf[il_pos].i[0];
208 wordr0[1] = combs_buf[il_pos].i[1];
209 wordr0[2] = combs_buf[il_pos].i[2];
210 wordr0[3] = combs_buf[il_pos].i[3];
214 wordr1[0] = combs_buf[il_pos].i[4];
215 wordr1[1] = combs_buf[il_pos].i[5];
216 wordr1[2] = combs_buf[il_pos].i[6];
217 wordr1[3] = combs_buf[il_pos].i[7];
233 if (combs_mode == COMBINATOR_MODE_BASE_LEFT)
235 switch_buffer_by_offset (wordr0, wordr1, wordr2, wordr3, pw_l_len);
240 w_t[ 0] = wordl0[0] | wordr0[0];
241 w_t[ 1] = wordl0[1] | wordr0[1];
242 w_t[ 2] = wordl0[2] | wordr0[2];
243 w_t[ 3] = wordl0[3] | wordr0[3];
244 w_t[ 4] = wordl1[0] | wordr1[0];
245 w_t[ 5] = wordl1[1] | wordr1[1];
246 w_t[ 6] = wordl1[2] | wordr1[2];
247 w_t[ 7] = wordl1[3] | wordr1[3];
248 w_t[ 8] = wordl2[0] | wordr2[0];
249 w_t[ 9] = wordl2[1] | wordr2[1];
250 w_t[10] = wordl2[2] | wordr2[2];
251 w_t[11] = wordl2[3] | wordr2[3];
252 w_t[12] = wordl3[0] | wordr3[0];
253 w_t[13] = wordl3[1] | wordr3[1];
254 w_t[14] = wordl3[2] | wordr3[2];
257 u32 a = crc32 (w_t, pw_len, iv);
269 __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m11500_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 combs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
273 __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m11500_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 combs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
277 __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m11500_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 combs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
283 const u32 lid = get_local_id (0);
289 const u32 gid = get_global_id (0);
291 if (gid >= gid_max) return;
295 wordl0[0] = pws[gid].i[ 0];
296 wordl0[1] = pws[gid].i[ 1];
297 wordl0[2] = pws[gid].i[ 2];
298 wordl0[3] = pws[gid].i[ 3];
302 wordl1[0] = pws[gid].i[ 4];
303 wordl1[1] = pws[gid].i[ 5];
304 wordl1[2] = pws[gid].i[ 6];
305 wordl1[3] = pws[gid].i[ 7];
321 const u32 pw_l_len = pws[gid].pw_len;
323 if (combs_mode == COMBINATOR_MODE_BASE_RIGHT)
325 switch_buffer_by_offset (wordl0, wordl1, wordl2, wordl3, combs_buf[0].pw_len);
332 const u32 iv = salt_bufs[salt_pos].salt_buf[0];
334 const u32 search[4] =
336 digests_buf[digests_offset].digest_buf[DGST_R0],
337 digests_buf[digests_offset].digest_buf[DGST_R1],
338 digests_buf[digests_offset].digest_buf[DGST_R2],
339 digests_buf[digests_offset].digest_buf[DGST_R3]
346 for (u32 il_pos = 0; il_pos < combs_cnt; il_pos++)
348 const u32 pw_r_len = combs_buf[il_pos].pw_len;
350 const u32 pw_len = pw_l_len + pw_r_len;
354 wordr0[0] = combs_buf[il_pos].i[0];
355 wordr0[1] = combs_buf[il_pos].i[1];
356 wordr0[2] = combs_buf[il_pos].i[2];
357 wordr0[3] = combs_buf[il_pos].i[3];
361 wordr1[0] = combs_buf[il_pos].i[4];
362 wordr1[1] = combs_buf[il_pos].i[5];
363 wordr1[2] = combs_buf[il_pos].i[6];
364 wordr1[3] = combs_buf[il_pos].i[7];
380 if (combs_mode == COMBINATOR_MODE_BASE_LEFT)
382 switch_buffer_by_offset (wordr0, wordr1, wordr2, wordr3, pw_l_len);
387 w_t[ 0] = wordl0[0] | wordr0[0];
388 w_t[ 1] = wordl0[1] | wordr0[1];
389 w_t[ 2] = wordl0[2] | wordr0[2];
390 w_t[ 3] = wordl0[3] | wordr0[3];
391 w_t[ 4] = wordl1[0] | wordr1[0];
392 w_t[ 5] = wordl1[1] | wordr1[1];
393 w_t[ 6] = wordl1[2] | wordr1[2];
394 w_t[ 7] = wordl1[3] | wordr1[3];
395 w_t[ 8] = wordl2[0] | wordr2[0];
396 w_t[ 9] = wordl2[1] | wordr2[1];
397 w_t[10] = wordl2[2] | wordr2[2];
398 w_t[11] = wordl2[3] | wordr2[3];
399 w_t[12] = wordl3[0] | wordr3[0];
400 w_t[13] = wordl3[1] | wordr3[1];
401 w_t[14] = wordl3[2] | wordr3[2];
404 u32 a = crc32 (w_t, pw_len, iv);
416 __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m11500_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 combs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
420 __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m11500_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 combs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)