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"
20 #define COMPARE_S "OpenCL/check_single_comp4.c"
21 #define COMPARE_M "OpenCL/check_multi_comp4.c"
23 __kernel void m00200_m04 (__global pw_t *pws, __global kernel_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)
29 const u32 lid = get_local_id (0);
35 const u32 gid = get_global_id (0);
37 if (gid >= gid_max) return;
41 wordl0[0] = pws[gid].i[ 0];
42 wordl0[1] = pws[gid].i[ 1];
43 wordl0[2] = pws[gid].i[ 2];
44 wordl0[3] = pws[gid].i[ 3];
48 wordl1[0] = pws[gid].i[ 4];
49 wordl1[1] = pws[gid].i[ 5];
50 wordl1[2] = pws[gid].i[ 6];
51 wordl1[3] = pws[gid].i[ 7];
67 const u32 pw_l_len = pws[gid].pw_len;
69 if (combs_mode == COMBINATOR_MODE_BASE_RIGHT)
71 switch_buffer_by_offset_le (wordl0, wordl1, wordl2, wordl3, combs_buf[0].pw_len);
78 for (u32 il_pos = 0; il_pos < combs_cnt; il_pos++)
80 const u32 pw_r_len = combs_buf[il_pos].pw_len;
82 const u32 pw_len = pw_l_len + pw_r_len;
86 wordr0[0] = combs_buf[il_pos].i[0];
87 wordr0[1] = combs_buf[il_pos].i[1];
88 wordr0[2] = combs_buf[il_pos].i[2];
89 wordr0[3] = combs_buf[il_pos].i[3];
93 wordr1[0] = combs_buf[il_pos].i[4];
94 wordr1[1] = combs_buf[il_pos].i[5];
95 wordr1[2] = combs_buf[il_pos].i[6];
96 wordr1[3] = combs_buf[il_pos].i[7];
112 if (combs_mode == COMBINATOR_MODE_BASE_LEFT)
114 switch_buffer_by_offset_le (wordr0, wordr1, wordr2, wordr3, pw_l_len);
119 w_t[ 0] = wordl0[0] | wordr0[0];
120 w_t[ 1] = wordl0[1] | wordr0[1];
121 w_t[ 2] = wordl0[2] | wordr0[2];
122 w_t[ 3] = wordl0[3] | wordr0[3];
123 w_t[ 4] = wordl1[0] | wordr1[0];
124 w_t[ 5] = wordl1[1] | wordr1[1];
125 w_t[ 6] = wordl1[2] | wordr1[2];
126 w_t[ 7] = wordl1[3] | wordr1[3];
127 w_t[ 8] = wordl2[0] | wordr2[0];
128 w_t[ 9] = wordl2[1] | wordr2[1];
129 w_t[10] = wordl2[2] | wordr2[2];
130 w_t[11] = wordl2[3] | wordr2[3];
131 w_t[12] = wordl3[0] | wordr3[0];
132 w_t[13] = wordl3[1] | wordr3[1];
133 w_t[14] = wordl3[2] | wordr3[2];
143 a ^= (((a & 0x3f) + add) * (v)) + (a << 8); \
151 for (i = 0, j = 0; i <= (int) pw_len - 4; i += 4, j += 1)
153 const u32 wj = w_t[j];
155 ROUND ((wj >> 0) & 0xff);
156 ROUND ((wj >> 8) & 0xff);
157 ROUND ((wj >> 16) & 0xff);
158 ROUND ((wj >> 24) & 0xff);
161 const u32 wj = w_t[j];
163 const u32 left = pw_len - i;
167 ROUND ((wj >> 0) & 0xff);
168 ROUND ((wj >> 8) & 0xff);
169 ROUND ((wj >> 16) & 0xff);
173 ROUND ((wj >> 0) & 0xff);
174 ROUND ((wj >> 8) & 0xff);
178 ROUND ((wj >> 0) & 0xff);
193 __kernel void m00200_m08 (__global pw_t *pws, __global kernel_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)
197 __kernel void m00200_m16 (__global pw_t *pws, __global kernel_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)
201 __kernel void m00200_s04 (__global pw_t *pws, __global kernel_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)
207 const u32 lid = get_local_id (0);
213 const u32 gid = get_global_id (0);
215 if (gid >= gid_max) return;
219 wordl0[0] = pws[gid].i[ 0];
220 wordl0[1] = pws[gid].i[ 1];
221 wordl0[2] = pws[gid].i[ 2];
222 wordl0[3] = pws[gid].i[ 3];
226 wordl1[0] = pws[gid].i[ 4];
227 wordl1[1] = pws[gid].i[ 5];
228 wordl1[2] = pws[gid].i[ 6];
229 wordl1[3] = pws[gid].i[ 7];
245 const u32 pw_l_len = pws[gid].pw_len;
247 if (combs_mode == COMBINATOR_MODE_BASE_RIGHT)
249 switch_buffer_by_offset_le (wordl0, wordl1, wordl2, wordl3, combs_buf[0].pw_len);
256 const u32 search[4] =
258 digests_buf[digests_offset].digest_buf[DGST_R0],
259 digests_buf[digests_offset].digest_buf[DGST_R1],
260 digests_buf[digests_offset].digest_buf[DGST_R2],
261 digests_buf[digests_offset].digest_buf[DGST_R3]
268 for (u32 il_pos = 0; il_pos < combs_cnt; il_pos++)
270 const u32 pw_r_len = combs_buf[il_pos].pw_len;
272 const u32 pw_len = pw_l_len + pw_r_len;
276 wordr0[0] = combs_buf[il_pos].i[0];
277 wordr0[1] = combs_buf[il_pos].i[1];
278 wordr0[2] = combs_buf[il_pos].i[2];
279 wordr0[3] = combs_buf[il_pos].i[3];
283 wordr1[0] = combs_buf[il_pos].i[4];
284 wordr1[1] = combs_buf[il_pos].i[5];
285 wordr1[2] = combs_buf[il_pos].i[6];
286 wordr1[3] = combs_buf[il_pos].i[7];
302 if (combs_mode == COMBINATOR_MODE_BASE_LEFT)
304 switch_buffer_by_offset_le (wordr0, wordr1, wordr2, wordr3, pw_l_len);
309 w_t[ 0] = wordl0[0] | wordr0[0];
310 w_t[ 1] = wordl0[1] | wordr0[1];
311 w_t[ 2] = wordl0[2] | wordr0[2];
312 w_t[ 3] = wordl0[3] | wordr0[3];
313 w_t[ 4] = wordl1[0] | wordr1[0];
314 w_t[ 5] = wordl1[1] | wordr1[1];
315 w_t[ 6] = wordl1[2] | wordr1[2];
316 w_t[ 7] = wordl1[3] | wordr1[3];
317 w_t[ 8] = wordl2[0] | wordr2[0];
318 w_t[ 9] = wordl2[1] | wordr2[1];
319 w_t[10] = wordl2[2] | wordr2[2];
320 w_t[11] = wordl2[3] | wordr2[3];
321 w_t[12] = wordl3[0] | wordr3[0];
322 w_t[13] = wordl3[1] | wordr3[1];
323 w_t[14] = wordl3[2] | wordr3[2];
333 a ^= (((a & 0x3f) + add) * (v)) + (a << 8); \
341 for (i = 0, j = 0; i <= (int) pw_len - 4; i += 4, j += 1)
343 const u32 wj = w_t[j];
345 ROUND ((wj >> 0) & 0xff);
346 ROUND ((wj >> 8) & 0xff);
347 ROUND ((wj >> 16) & 0xff);
348 ROUND ((wj >> 24) & 0xff);
351 const u32 wj = w_t[j];
353 const u32 left = pw_len - i;
357 ROUND ((wj >> 0) & 0xff);
358 ROUND ((wj >> 8) & 0xff);
359 ROUND ((wj >> 16) & 0xff);
363 ROUND ((wj >> 0) & 0xff);
364 ROUND ((wj >> 8) & 0xff);
368 ROUND ((wj >> 0) & 0xff);
383 __kernel void m00200_s08 (__global pw_t *pws, __global kernel_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)
387 __kernel void m00200_s16 (__global pw_t *pws, __global kernel_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)