2 * Author......: Jens Steube <jens.steube@gmail.com>
10 #include "inc_hash_constants.h"
11 #include "inc_vendor.cl"
18 #include "inc_hash_functions.cl"
19 #include "inc_types.cl"
20 #include "inc_common.cl"
23 #include "inc_simd.cl"
25 __kernel void m02410_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 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
31 const u32 lid = get_local_id (0);
37 const u32 gid = get_global_id (0);
39 if (gid >= gid_max) return;
44 pw_buf0[0] = pws[gid].i[0];
45 pw_buf0[1] = pws[gid].i[1];
46 pw_buf0[2] = pws[gid].i[2];
47 pw_buf0[3] = pws[gid].i[3];
48 pw_buf1[0] = pws[gid].i[4];
49 pw_buf1[1] = pws[gid].i[5];
50 pw_buf1[2] = pws[gid].i[6];
51 pw_buf1[3] = pws[gid].i[7];
53 const u32 pw_len = pws[gid].pw_len;
64 salt_buf0[0] = salt_bufs[salt_pos].salt_buf[ 0];
85 for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
92 const u32x out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
103 s0[0] = salt_buf0[0];
104 s0[1] = salt_buf0[1];
105 s0[2] = salt_buf0[2];
106 s0[3] = salt_buf0[3];
107 s1[0] = salt_buf1[0];
108 s1[1] = salt_buf1[1];
109 s1[2] = salt_buf1[2];
110 s1[3] = salt_buf1[3];
111 s2[0] = salt_buf2[0];
112 s2[1] = salt_buf2[1];
113 s2[2] = salt_buf2[2];
114 s2[3] = salt_buf2[3];
115 s3[0] = salt_buf3[0];
116 s3[1] = salt_buf3[1];
117 s3[2] = salt_buf3[2];
118 s3[3] = salt_buf3[3];
120 switch_buffer_by_offset_le_VV (s0, s1, s2, s3, out_len);
161 MD5_STEP (MD5_Fo, a, b, c, d, w0[0], MD5C00, MD5S00);
162 MD5_STEP (MD5_Fo, d, a, b, c, w0[1], MD5C01, MD5S01);
163 MD5_STEP (MD5_Fo, c, d, a, b, w0[2], MD5C02, MD5S02);
164 MD5_STEP (MD5_Fo, b, c, d, a, w0[3], MD5C03, MD5S03);
165 MD5_STEP (MD5_Fo, a, b, c, d, w1[0], MD5C04, MD5S00);
166 MD5_STEP (MD5_Fo, d, a, b, c, w1[1], MD5C05, MD5S01);
167 MD5_STEP (MD5_Fo, c, d, a, b, w1[2], MD5C06, MD5S02);
168 MD5_STEP (MD5_Fo, b, c, d, a, w1[3], MD5C07, MD5S03);
169 MD5_STEP (MD5_Fo, a, b, c, d, w2[0], MD5C08, MD5S00);
170 MD5_STEP (MD5_Fo, d, a, b, c, w2[1], MD5C09, MD5S01);
171 MD5_STEP (MD5_Fo, c, d, a, b, w2[2], MD5C0a, MD5S02);
172 MD5_STEP (MD5_Fo, b, c, d, a, w2[3], MD5C0b, MD5S03);
173 MD5_STEP (MD5_Fo, a, b, c, d, w3[0], MD5C0c, MD5S00);
174 MD5_STEP (MD5_Fo, d, a, b, c, w3[1], MD5C0d, MD5S01);
175 MD5_STEP (MD5_Fo, c, d, a, b, w3[2], MD5C0e, MD5S02);
176 MD5_STEP (MD5_Fo, b, c, d, a, w3[3], MD5C0f, MD5S03);
178 MD5_STEP (MD5_Go, a, b, c, d, w0[1], MD5C10, MD5S10);
179 MD5_STEP (MD5_Go, d, a, b, c, w1[2], MD5C11, MD5S11);
180 MD5_STEP (MD5_Go, c, d, a, b, w2[3], MD5C12, MD5S12);
181 MD5_STEP (MD5_Go, b, c, d, a, w0[0], MD5C13, MD5S13);
182 MD5_STEP (MD5_Go, a, b, c, d, w1[1], MD5C14, MD5S10);
183 MD5_STEP (MD5_Go, d, a, b, c, w2[2], MD5C15, MD5S11);
184 MD5_STEP (MD5_Go, c, d, a, b, w3[3], MD5C16, MD5S12);
185 MD5_STEP (MD5_Go, b, c, d, a, w1[0], MD5C17, MD5S13);
186 MD5_STEP (MD5_Go, a, b, c, d, w2[1], MD5C18, MD5S10);
187 MD5_STEP (MD5_Go, d, a, b, c, w3[2], MD5C19, MD5S11);
188 MD5_STEP (MD5_Go, c, d, a, b, w0[3], MD5C1a, MD5S12);
189 MD5_STEP (MD5_Go, b, c, d, a, w2[0], MD5C1b, MD5S13);
190 MD5_STEP (MD5_Go, a, b, c, d, w3[1], MD5C1c, MD5S10);
191 MD5_STEP (MD5_Go, d, a, b, c, w0[2], MD5C1d, MD5S11);
192 MD5_STEP (MD5_Go, c, d, a, b, w1[3], MD5C1e, MD5S12);
193 MD5_STEP (MD5_Go, b, c, d, a, w3[0], MD5C1f, MD5S13);
195 MD5_STEP (MD5_H , a, b, c, d, w1[1], MD5C20, MD5S20);
196 MD5_STEP (MD5_H , d, a, b, c, w2[0], MD5C21, MD5S21);
197 MD5_STEP (MD5_H , c, d, a, b, w2[3], MD5C22, MD5S22);
198 MD5_STEP (MD5_H , b, c, d, a, w3[2], MD5C23, MD5S23);
199 MD5_STEP (MD5_H , a, b, c, d, w0[1], MD5C24, MD5S20);
200 MD5_STEP (MD5_H , d, a, b, c, w1[0], MD5C25, MD5S21);
201 MD5_STEP (MD5_H , c, d, a, b, w1[3], MD5C26, MD5S22);
202 MD5_STEP (MD5_H , b, c, d, a, w2[2], MD5C27, MD5S23);
203 MD5_STEP (MD5_H , a, b, c, d, w3[1], MD5C28, MD5S20);
204 MD5_STEP (MD5_H , d, a, b, c, w0[0], MD5C29, MD5S21);
205 MD5_STEP (MD5_H , c, d, a, b, w0[3], MD5C2a, MD5S22);
206 MD5_STEP (MD5_H , b, c, d, a, w1[2], MD5C2b, MD5S23);
207 MD5_STEP (MD5_H , a, b, c, d, w2[1], MD5C2c, MD5S20);
208 MD5_STEP (MD5_H , d, a, b, c, w3[0], MD5C2d, MD5S21);
209 MD5_STEP (MD5_H , c, d, a, b, w3[3], MD5C2e, MD5S22);
210 MD5_STEP (MD5_H , b, c, d, a, w0[2], MD5C2f, MD5S23);
212 MD5_STEP (MD5_I , a, b, c, d, w0[0], MD5C30, MD5S30);
213 MD5_STEP (MD5_I , d, a, b, c, w1[3], MD5C31, MD5S31);
214 MD5_STEP (MD5_I , c, d, a, b, w3[2], MD5C32, MD5S32);
215 MD5_STEP (MD5_I , b, c, d, a, w1[1], MD5C33, MD5S33);
216 MD5_STEP (MD5_I , a, b, c, d, w3[0], MD5C34, MD5S30);
217 MD5_STEP (MD5_I , d, a, b, c, w0[3], MD5C35, MD5S31);
218 MD5_STEP (MD5_I , c, d, a, b, w2[2], MD5C36, MD5S32);
219 MD5_STEP (MD5_I , b, c, d, a, w0[1], MD5C37, MD5S33);
220 MD5_STEP (MD5_I , a, b, c, d, w2[0], MD5C38, MD5S30);
221 MD5_STEP (MD5_I , d, a, b, c, w3[3], MD5C39, MD5S31);
222 MD5_STEP (MD5_I , c, d, a, b, w1[2], MD5C3a, MD5S32);
223 MD5_STEP (MD5_I , b, c, d, a, w3[1], MD5C3b, MD5S33);
224 MD5_STEP (MD5_I , a, b, c, d, w1[0], MD5C3c, MD5S30);
225 MD5_STEP (MD5_I , d, a, b, c, w2[3], MD5C3d, MD5S31);
226 MD5_STEP (MD5_I , c, d, a, b, w0[2], MD5C3e, MD5S32);
227 MD5_STEP (MD5_I , b, c, d, a, w2[1], MD5C3f, MD5S33);
234 COMPARE_M_SIMD (a, d, c, b);
238 __kernel void m02410_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 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
242 __kernel void m02410_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 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
246 __kernel void m02410_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 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
252 const u32 lid = get_local_id (0);
258 const u32 gid = get_global_id (0);
260 if (gid >= gid_max) return;
265 pw_buf0[0] = pws[gid].i[0];
266 pw_buf0[1] = pws[gid].i[1];
267 pw_buf0[2] = pws[gid].i[2];
268 pw_buf0[3] = pws[gid].i[3];
269 pw_buf1[0] = pws[gid].i[4];
270 pw_buf1[1] = pws[gid].i[5];
271 pw_buf1[2] = pws[gid].i[6];
272 pw_buf1[3] = pws[gid].i[7];
274 const u32 pw_len = pws[gid].pw_len;
285 salt_buf0[0] = salt_bufs[salt_pos].salt_buf[ 0];
306 const u32 search[4] =
308 digests_buf[digests_offset].digest_buf[DGST_R0],
309 digests_buf[digests_offset].digest_buf[DGST_R1],
310 digests_buf[digests_offset].digest_buf[DGST_R2],
311 digests_buf[digests_offset].digest_buf[DGST_R3]
318 for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
325 const u32x out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
336 s0[0] = salt_buf0[0];
337 s0[1] = salt_buf0[1];
338 s0[2] = salt_buf0[2];
339 s0[3] = salt_buf0[3];
340 s1[0] = salt_buf1[0];
341 s1[1] = salt_buf1[1];
342 s1[2] = salt_buf1[2];
343 s1[3] = salt_buf1[3];
344 s2[0] = salt_buf2[0];
345 s2[1] = salt_buf2[1];
346 s2[2] = salt_buf2[2];
347 s2[3] = salt_buf2[3];
348 s3[0] = salt_buf3[0];
349 s3[1] = salt_buf3[1];
350 s3[2] = salt_buf3[2];
351 s3[3] = salt_buf3[3];
353 switch_buffer_by_offset_le_VV (s0, s1, s2, s3, out_len);
394 MD5_STEP (MD5_Fo, a, b, c, d, w0[0], MD5C00, MD5S00);
395 MD5_STEP (MD5_Fo, d, a, b, c, w0[1], MD5C01, MD5S01);
396 MD5_STEP (MD5_Fo, c, d, a, b, w0[2], MD5C02, MD5S02);
397 MD5_STEP (MD5_Fo, b, c, d, a, w0[3], MD5C03, MD5S03);
398 MD5_STEP (MD5_Fo, a, b, c, d, w1[0], MD5C04, MD5S00);
399 MD5_STEP (MD5_Fo, d, a, b, c, w1[1], MD5C05, MD5S01);
400 MD5_STEP (MD5_Fo, c, d, a, b, w1[2], MD5C06, MD5S02);
401 MD5_STEP (MD5_Fo, b, c, d, a, w1[3], MD5C07, MD5S03);
402 MD5_STEP (MD5_Fo, a, b, c, d, w2[0], MD5C08, MD5S00);
403 MD5_STEP (MD5_Fo, d, a, b, c, w2[1], MD5C09, MD5S01);
404 MD5_STEP (MD5_Fo, c, d, a, b, w2[2], MD5C0a, MD5S02);
405 MD5_STEP (MD5_Fo, b, c, d, a, w2[3], MD5C0b, MD5S03);
406 MD5_STEP (MD5_Fo, a, b, c, d, w3[0], MD5C0c, MD5S00);
407 MD5_STEP (MD5_Fo, d, a, b, c, w3[1], MD5C0d, MD5S01);
408 MD5_STEP (MD5_Fo, c, d, a, b, w3[2], MD5C0e, MD5S02);
409 MD5_STEP (MD5_Fo, b, c, d, a, w3[3], MD5C0f, MD5S03);
411 MD5_STEP (MD5_Go, a, b, c, d, w0[1], MD5C10, MD5S10);
412 MD5_STEP (MD5_Go, d, a, b, c, w1[2], MD5C11, MD5S11);
413 MD5_STEP (MD5_Go, c, d, a, b, w2[3], MD5C12, MD5S12);
414 MD5_STEP (MD5_Go, b, c, d, a, w0[0], MD5C13, MD5S13);
415 MD5_STEP (MD5_Go, a, b, c, d, w1[1], MD5C14, MD5S10);
416 MD5_STEP (MD5_Go, d, a, b, c, w2[2], MD5C15, MD5S11);
417 MD5_STEP (MD5_Go, c, d, a, b, w3[3], MD5C16, MD5S12);
418 MD5_STEP (MD5_Go, b, c, d, a, w1[0], MD5C17, MD5S13);
419 MD5_STEP (MD5_Go, a, b, c, d, w2[1], MD5C18, MD5S10);
420 MD5_STEP (MD5_Go, d, a, b, c, w3[2], MD5C19, MD5S11);
421 MD5_STEP (MD5_Go, c, d, a, b, w0[3], MD5C1a, MD5S12);
422 MD5_STEP (MD5_Go, b, c, d, a, w2[0], MD5C1b, MD5S13);
423 MD5_STEP (MD5_Go, a, b, c, d, w3[1], MD5C1c, MD5S10);
424 MD5_STEP (MD5_Go, d, a, b, c, w0[2], MD5C1d, MD5S11);
425 MD5_STEP (MD5_Go, c, d, a, b, w1[3], MD5C1e, MD5S12);
426 MD5_STEP (MD5_Go, b, c, d, a, w3[0], MD5C1f, MD5S13);
428 MD5_STEP (MD5_H , a, b, c, d, w1[1], MD5C20, MD5S20);
429 MD5_STEP (MD5_H , d, a, b, c, w2[0], MD5C21, MD5S21);
430 MD5_STEP (MD5_H , c, d, a, b, w2[3], MD5C22, MD5S22);
431 MD5_STEP (MD5_H , b, c, d, a, w3[2], MD5C23, MD5S23);
432 MD5_STEP (MD5_H , a, b, c, d, w0[1], MD5C24, MD5S20);
433 MD5_STEP (MD5_H , d, a, b, c, w1[0], MD5C25, MD5S21);
434 MD5_STEP (MD5_H , c, d, a, b, w1[3], MD5C26, MD5S22);
435 MD5_STEP (MD5_H , b, c, d, a, w2[2], MD5C27, MD5S23);
436 MD5_STEP (MD5_H , a, b, c, d, w3[1], MD5C28, MD5S20);
437 MD5_STEP (MD5_H , d, a, b, c, w0[0], MD5C29, MD5S21);
438 MD5_STEP (MD5_H , c, d, a, b, w0[3], MD5C2a, MD5S22);
439 MD5_STEP (MD5_H , b, c, d, a, w1[2], MD5C2b, MD5S23);
440 MD5_STEP (MD5_H , a, b, c, d, w2[1], MD5C2c, MD5S20);
441 MD5_STEP (MD5_H , d, a, b, c, w3[0], MD5C2d, MD5S21);
442 MD5_STEP (MD5_H , c, d, a, b, w3[3], MD5C2e, MD5S22);
443 MD5_STEP (MD5_H , b, c, d, a, w0[2], MD5C2f, MD5S23);
445 MD5_STEP (MD5_I , a, b, c, d, w0[0], MD5C30, MD5S30);
446 MD5_STEP (MD5_I , d, a, b, c, w1[3], MD5C31, MD5S31);
447 MD5_STEP (MD5_I , c, d, a, b, w3[2], MD5C32, MD5S32);
448 MD5_STEP (MD5_I , b, c, d, a, w1[1], MD5C33, MD5S33);
449 MD5_STEP (MD5_I , a, b, c, d, w3[0], MD5C34, MD5S30);
450 MD5_STEP (MD5_I , d, a, b, c, w0[3], MD5C35, MD5S31);
451 MD5_STEP (MD5_I , c, d, a, b, w2[2], MD5C36, MD5S32);
452 MD5_STEP (MD5_I , b, c, d, a, w0[1], MD5C37, MD5S33);
453 MD5_STEP (MD5_I , a, b, c, d, w2[0], MD5C38, MD5S30);
454 MD5_STEP (MD5_I , d, a, b, c, w3[3], MD5C39, MD5S31);
455 MD5_STEP (MD5_I , c, d, a, b, w1[2], MD5C3a, MD5S32);
456 MD5_STEP (MD5_I , b, c, d, a, w3[1], MD5C3b, MD5S33);
457 MD5_STEP (MD5_I , a, b, c, d, w1[0], MD5C3c, MD5S30);
459 if (MATCHES_NONE_VS ((a & 0x00ffffff), search[0])) continue;
461 MD5_STEP (MD5_I , d, a, b, c, w2[3], MD5C3d, MD5S31);
462 MD5_STEP (MD5_I , c, d, a, b, w0[2], MD5C3e, MD5S32);
463 MD5_STEP (MD5_I , b, c, d, a, w2[1], MD5C3f, MD5S33);
470 COMPARE_S_SIMD (a, d, c, b);
474 __kernel void m02410_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 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
478 __kernel void m02410_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 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)