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"
33 #define VECT_COMPARE_S "check_single_vect1_comp4.c"
34 #define VECT_COMPARE_M "check_multi_vect1_comp4.c"
38 #define VECT_COMPARE_S "check_single_vect2_comp4.c"
39 #define VECT_COMPARE_M "check_multi_vect2_comp4.c"
43 #define VECT_COMPARE_S "check_single_vect4_comp4.c"
44 #define VECT_COMPARE_M "check_multi_vect4_comp4.c"
47 __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m00200_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)
53 const u32 lid = get_local_id (0);
59 const u32 gid = get_global_id (0);
61 if (gid >= gid_max) return;
65 wordl0[0] = pws[gid].i[ 0];
66 wordl0[1] = pws[gid].i[ 1];
67 wordl0[2] = pws[gid].i[ 2];
68 wordl0[3] = pws[gid].i[ 3];
72 wordl1[0] = pws[gid].i[ 4];
73 wordl1[1] = pws[gid].i[ 5];
74 wordl1[2] = pws[gid].i[ 6];
75 wordl1[3] = pws[gid].i[ 7];
91 const u32 pw_l_len = pws[gid].pw_len;
93 if (combs_mode == COMBINATOR_MODE_BASE_RIGHT)
95 switch_buffer_by_offset (wordl0, wordl1, wordl2, wordl3, combs_buf[0].pw_len);
102 for (u32 il_pos = 0; il_pos < combs_cnt; il_pos++)
104 const u32 pw_r_len = combs_buf[il_pos].pw_len;
106 const u32 pw_len = pw_l_len + pw_r_len;
110 wordr0[0] = combs_buf[il_pos].i[0];
111 wordr0[1] = combs_buf[il_pos].i[1];
112 wordr0[2] = combs_buf[il_pos].i[2];
113 wordr0[3] = combs_buf[il_pos].i[3];
117 wordr1[0] = combs_buf[il_pos].i[4];
118 wordr1[1] = combs_buf[il_pos].i[5];
119 wordr1[2] = combs_buf[il_pos].i[6];
120 wordr1[3] = combs_buf[il_pos].i[7];
136 if (combs_mode == COMBINATOR_MODE_BASE_LEFT)
138 switch_buffer_by_offset (wordr0, wordr1, wordr2, wordr3, pw_l_len);
143 w_t[ 0] = wordl0[0] | wordr0[0];
144 w_t[ 1] = wordl0[1] | wordr0[1];
145 w_t[ 2] = wordl0[2] | wordr0[2];
146 w_t[ 3] = wordl0[3] | wordr0[3];
147 w_t[ 4] = wordl1[0] | wordr1[0];
148 w_t[ 5] = wordl1[1] | wordr1[1];
149 w_t[ 6] = wordl1[2] | wordr1[2];
150 w_t[ 7] = wordl1[3] | wordr1[3];
151 w_t[ 8] = wordl2[0] | wordr2[0];
152 w_t[ 9] = wordl2[1] | wordr2[1];
153 w_t[10] = wordl2[2] | wordr2[2];
154 w_t[11] = wordl2[3] | wordr2[3];
155 w_t[12] = wordl3[0] | wordr3[0];
156 w_t[13] = wordl3[1] | wordr3[1];
157 w_t[14] = wordl3[2] | wordr3[2];
167 a ^= (((a & 0x3f) + add) * (v)) + (a << 8); \
175 for (i = 0, j = 0; i <= (int) pw_len - 4; i += 4, j += 1)
177 const u32x wj = w_t[j];
179 ROUND ((wj >> 0) & 0xff);
180 ROUND ((wj >> 8) & 0xff);
181 ROUND ((wj >> 16) & 0xff);
182 ROUND ((wj >> 24) & 0xff);
185 const u32x wj = w_t[j];
187 const u32 left = pw_len - i;
191 ROUND ((wj >> 0) & 0xff);
192 ROUND ((wj >> 8) & 0xff);
193 ROUND ((wj >> 16) & 0xff);
197 ROUND ((wj >> 0) & 0xff);
198 ROUND ((wj >> 8) & 0xff);
202 ROUND ((wj >> 0) & 0xff);
213 #include VECT_COMPARE_M
217 __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m00200_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)
221 __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m00200_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)
225 __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m00200_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)
231 const u32 lid = get_local_id (0);
237 const u32 gid = get_global_id (0);
239 if (gid >= gid_max) return;
243 wordl0[0] = pws[gid].i[ 0];
244 wordl0[1] = pws[gid].i[ 1];
245 wordl0[2] = pws[gid].i[ 2];
246 wordl0[3] = pws[gid].i[ 3];
250 wordl1[0] = pws[gid].i[ 4];
251 wordl1[1] = pws[gid].i[ 5];
252 wordl1[2] = pws[gid].i[ 6];
253 wordl1[3] = pws[gid].i[ 7];
269 const u32 pw_l_len = pws[gid].pw_len;
271 if (combs_mode == COMBINATOR_MODE_BASE_RIGHT)
273 switch_buffer_by_offset (wordl0, wordl1, wordl2, wordl3, combs_buf[0].pw_len);
280 const u32 search[4] =
282 digests_buf[digests_offset].digest_buf[DGST_R0],
283 digests_buf[digests_offset].digest_buf[DGST_R1],
284 digests_buf[digests_offset].digest_buf[DGST_R2],
285 digests_buf[digests_offset].digest_buf[DGST_R3]
292 for (u32 il_pos = 0; il_pos < combs_cnt; il_pos++)
294 const u32 pw_r_len = combs_buf[il_pos].pw_len;
296 const u32 pw_len = pw_l_len + pw_r_len;
300 wordr0[0] = combs_buf[il_pos].i[0];
301 wordr0[1] = combs_buf[il_pos].i[1];
302 wordr0[2] = combs_buf[il_pos].i[2];
303 wordr0[3] = combs_buf[il_pos].i[3];
307 wordr1[0] = combs_buf[il_pos].i[4];
308 wordr1[1] = combs_buf[il_pos].i[5];
309 wordr1[2] = combs_buf[il_pos].i[6];
310 wordr1[3] = combs_buf[il_pos].i[7];
326 if (combs_mode == COMBINATOR_MODE_BASE_LEFT)
328 switch_buffer_by_offset (wordr0, wordr1, wordr2, wordr3, pw_l_len);
333 w_t[ 0] = wordl0[0] | wordr0[0];
334 w_t[ 1] = wordl0[1] | wordr0[1];
335 w_t[ 2] = wordl0[2] | wordr0[2];
336 w_t[ 3] = wordl0[3] | wordr0[3];
337 w_t[ 4] = wordl1[0] | wordr1[0];
338 w_t[ 5] = wordl1[1] | wordr1[1];
339 w_t[ 6] = wordl1[2] | wordr1[2];
340 w_t[ 7] = wordl1[3] | wordr1[3];
341 w_t[ 8] = wordl2[0] | wordr2[0];
342 w_t[ 9] = wordl2[1] | wordr2[1];
343 w_t[10] = wordl2[2] | wordr2[2];
344 w_t[11] = wordl2[3] | wordr2[3];
345 w_t[12] = wordl3[0] | wordr3[0];
346 w_t[13] = wordl3[1] | wordr3[1];
347 w_t[14] = wordl3[2] | wordr3[2];
357 a ^= (((a & 0x3f) + add) * (v)) + (a << 8); \
365 for (i = 0, j = 0; i <= (int) pw_len - 4; i += 4, j += 1)
367 const u32x wj = w_t[j];
369 ROUND ((wj >> 0) & 0xff);
370 ROUND ((wj >> 8) & 0xff);
371 ROUND ((wj >> 16) & 0xff);
372 ROUND ((wj >> 24) & 0xff);
375 const u32x wj = w_t[j];
377 const u32 left = pw_len - i;
381 ROUND ((wj >> 0) & 0xff);
382 ROUND ((wj >> 8) & 0xff);
383 ROUND ((wj >> 16) & 0xff);
387 ROUND ((wj >> 0) & 0xff);
388 ROUND ((wj >> 8) & 0xff);
392 ROUND ((wj >> 0) & 0xff);
403 #include VECT_COMPARE_S
407 __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m00200_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)
411 __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m00200_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)