2 * Author......: Jens Steube <jens.steube@gmail.com>
10 #include "inc_vendor.cl"
11 #include "inc_hash_constants.h"
12 #include "inc_hash_functions.cl"
13 #include "inc_types.cl"
14 #include "inc_common.cl"
17 #include "inc_simd.cl"
19 __kernel void m00040_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_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_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)
25 const u32 lid = get_local_id (0);
31 const u32 gid = get_global_id (0);
33 if (gid >= gid_max) return;
38 pw_buf0[0] = pws[gid].i[0];
39 pw_buf0[1] = pws[gid].i[1];
40 pw_buf0[2] = pws[gid].i[2];
41 pw_buf0[3] = pws[gid].i[3];
42 pw_buf1[0] = pws[gid].i[4];
43 pw_buf1[1] = pws[gid].i[5];
44 pw_buf1[2] = pws[gid].i[6];
45 pw_buf1[3] = pws[gid].i[7];
47 const u32 pw_len = pws[gid].pw_len;
58 salt_buf0[0] = salt_bufs[salt_pos].salt_buf[ 0];
59 salt_buf0[1] = salt_bufs[salt_pos].salt_buf[ 1];
60 salt_buf0[2] = salt_bufs[salt_pos].salt_buf[ 2];
61 salt_buf0[3] = salt_bufs[salt_pos].salt_buf[ 3];
62 salt_buf1[0] = salt_bufs[salt_pos].salt_buf[ 4];
63 salt_buf1[1] = salt_bufs[salt_pos].salt_buf[ 5];
64 salt_buf1[2] = salt_bufs[salt_pos].salt_buf[ 6];
65 salt_buf1[3] = salt_bufs[salt_pos].salt_buf[ 7];
66 salt_buf2[0] = salt_bufs[salt_pos].salt_buf[ 8];
67 salt_buf2[1] = salt_bufs[salt_pos].salt_buf[ 9];
68 salt_buf2[2] = salt_bufs[salt_pos].salt_buf[10];
69 salt_buf2[3] = salt_bufs[salt_pos].salt_buf[11];
70 salt_buf3[0] = salt_bufs[salt_pos].salt_buf[12];
71 salt_buf3[1] = salt_bufs[salt_pos].salt_buf[13];
72 salt_buf3[2] = salt_bufs[salt_pos].salt_buf[14];
73 salt_buf3[3] = salt_bufs[salt_pos].salt_buf[15];
75 const u32 salt_len = salt_bufs[salt_pos].salt_len;
81 for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
88 const u32x out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
90 make_unicode (w1, w2, w3);
91 make_unicode (w0, w0, w1);
93 const u32x out_len2 = out_len * 2;
99 const u32x out_salt_len = out_len2 + salt_len;
101 switch_buffer_by_offset_le_VV (w0, w1, w2, w3, salt_len);
103 w0[0] |= salt_buf0[0];
104 w0[1] |= salt_buf0[1];
105 w0[2] |= salt_buf0[2];
106 w0[3] |= salt_buf0[3];
107 w1[0] |= salt_buf1[0];
108 w1[1] |= salt_buf1[1];
109 w1[2] |= salt_buf1[2];
110 w1[3] |= salt_buf1[3];
111 w2[0] |= salt_buf2[0];
112 w2[1] |= salt_buf2[1];
113 w2[2] |= salt_buf2[2];
114 w2[3] |= salt_buf2[3];
115 w3[0] |= salt_buf3[0];
116 w3[1] |= salt_buf3[1];
117 w3[2] = out_salt_len * 8;
120 append_0x80_4x4_VV (w0, w1, w2, w3, out_salt_len);
131 MD5_STEP (MD5_Fo, a, b, c, d, w0[0], MD5C00, MD5S00);
132 MD5_STEP (MD5_Fo, d, a, b, c, w0[1], MD5C01, MD5S01);
133 MD5_STEP (MD5_Fo, c, d, a, b, w0[2], MD5C02, MD5S02);
134 MD5_STEP (MD5_Fo, b, c, d, a, w0[3], MD5C03, MD5S03);
135 MD5_STEP (MD5_Fo, a, b, c, d, w1[0], MD5C04, MD5S00);
136 MD5_STEP (MD5_Fo, d, a, b, c, w1[1], MD5C05, MD5S01);
137 MD5_STEP (MD5_Fo, c, d, a, b, w1[2], MD5C06, MD5S02);
138 MD5_STEP (MD5_Fo, b, c, d, a, w1[3], MD5C07, MD5S03);
139 MD5_STEP (MD5_Fo, a, b, c, d, w2[0], MD5C08, MD5S00);
140 MD5_STEP (MD5_Fo, d, a, b, c, w2[1], MD5C09, MD5S01);
141 MD5_STEP (MD5_Fo, c, d, a, b, w2[2], MD5C0a, MD5S02);
142 MD5_STEP (MD5_Fo, b, c, d, a, w2[3], MD5C0b, MD5S03);
143 MD5_STEP (MD5_Fo, a, b, c, d, w3[0], MD5C0c, MD5S00);
144 MD5_STEP (MD5_Fo, d, a, b, c, w3[1], MD5C0d, MD5S01);
145 MD5_STEP (MD5_Fo, c, d, a, b, w3[2], MD5C0e, MD5S02);
146 MD5_STEP (MD5_Fo, b, c, d, a, w3[3], MD5C0f, MD5S03);
148 MD5_STEP (MD5_Go, a, b, c, d, w0[1], MD5C10, MD5S10);
149 MD5_STEP (MD5_Go, d, a, b, c, w1[2], MD5C11, MD5S11);
150 MD5_STEP (MD5_Go, c, d, a, b, w2[3], MD5C12, MD5S12);
151 MD5_STEP (MD5_Go, b, c, d, a, w0[0], MD5C13, MD5S13);
152 MD5_STEP (MD5_Go, a, b, c, d, w1[1], MD5C14, MD5S10);
153 MD5_STEP (MD5_Go, d, a, b, c, w2[2], MD5C15, MD5S11);
154 MD5_STEP (MD5_Go, c, d, a, b, w3[3], MD5C16, MD5S12);
155 MD5_STEP (MD5_Go, b, c, d, a, w1[0], MD5C17, MD5S13);
156 MD5_STEP (MD5_Go, a, b, c, d, w2[1], MD5C18, MD5S10);
157 MD5_STEP (MD5_Go, d, a, b, c, w3[2], MD5C19, MD5S11);
158 MD5_STEP (MD5_Go, c, d, a, b, w0[3], MD5C1a, MD5S12);
159 MD5_STEP (MD5_Go, b, c, d, a, w2[0], MD5C1b, MD5S13);
160 MD5_STEP (MD5_Go, a, b, c, d, w3[1], MD5C1c, MD5S10);
161 MD5_STEP (MD5_Go, d, a, b, c, w0[2], MD5C1d, MD5S11);
162 MD5_STEP (MD5_Go, c, d, a, b, w1[3], MD5C1e, MD5S12);
163 MD5_STEP (MD5_Go, b, c, d, a, w3[0], MD5C1f, MD5S13);
165 MD5_STEP (MD5_H , a, b, c, d, w1[1], MD5C20, MD5S20);
166 MD5_STEP (MD5_H , d, a, b, c, w2[0], MD5C21, MD5S21);
167 MD5_STEP (MD5_H , c, d, a, b, w2[3], MD5C22, MD5S22);
168 MD5_STEP (MD5_H , b, c, d, a, w3[2], MD5C23, MD5S23);
169 MD5_STEP (MD5_H , a, b, c, d, w0[1], MD5C24, MD5S20);
170 MD5_STEP (MD5_H , d, a, b, c, w1[0], MD5C25, MD5S21);
171 MD5_STEP (MD5_H , c, d, a, b, w1[3], MD5C26, MD5S22);
172 MD5_STEP (MD5_H , b, c, d, a, w2[2], MD5C27, MD5S23);
173 MD5_STEP (MD5_H , a, b, c, d, w3[1], MD5C28, MD5S20);
174 MD5_STEP (MD5_H , d, a, b, c, w0[0], MD5C29, MD5S21);
175 MD5_STEP (MD5_H , c, d, a, b, w0[3], MD5C2a, MD5S22);
176 MD5_STEP (MD5_H , b, c, d, a, w1[2], MD5C2b, MD5S23);
177 MD5_STEP (MD5_H , a, b, c, d, w2[1], MD5C2c, MD5S20);
178 MD5_STEP (MD5_H , d, a, b, c, w3[0], MD5C2d, MD5S21);
179 MD5_STEP (MD5_H , c, d, a, b, w3[3], MD5C2e, MD5S22);
180 MD5_STEP (MD5_H , b, c, d, a, w0[2], MD5C2f, MD5S23);
182 MD5_STEP (MD5_I , a, b, c, d, w0[0], MD5C30, MD5S30);
183 MD5_STEP (MD5_I , d, a, b, c, w1[3], MD5C31, MD5S31);
184 MD5_STEP (MD5_I , c, d, a, b, w3[2], MD5C32, MD5S32);
185 MD5_STEP (MD5_I , b, c, d, a, w1[1], MD5C33, MD5S33);
186 MD5_STEP (MD5_I , a, b, c, d, w3[0], MD5C34, MD5S30);
187 MD5_STEP (MD5_I , d, a, b, c, w0[3], MD5C35, MD5S31);
188 MD5_STEP (MD5_I , c, d, a, b, w2[2], MD5C36, MD5S32);
189 MD5_STEP (MD5_I , b, c, d, a, w0[1], MD5C37, MD5S33);
190 MD5_STEP (MD5_I , a, b, c, d, w2[0], MD5C38, MD5S30);
191 MD5_STEP (MD5_I , d, a, b, c, w3[3], MD5C39, MD5S31);
192 MD5_STEP (MD5_I , c, d, a, b, w1[2], MD5C3a, MD5S32);
193 MD5_STEP (MD5_I , b, c, d, a, w3[1], MD5C3b, MD5S33);
194 MD5_STEP (MD5_I , a, b, c, d, w1[0], MD5C3c, MD5S30);
195 MD5_STEP (MD5_I , d, a, b, c, w2[3], MD5C3d, MD5S31);
196 MD5_STEP (MD5_I , c, d, a, b, w0[2], MD5C3e, MD5S32);
197 MD5_STEP (MD5_I , b, c, d, a, w2[1], MD5C3f, MD5S33);
199 COMPARE_M_SIMD (a, d, c, b);
203 __kernel void m00040_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_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_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)
207 __kernel void m00040_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_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_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)
211 __kernel void m00040_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_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_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)
217 const u32 lid = get_local_id (0);
223 const u32 gid = get_global_id (0);
225 if (gid >= gid_max) return;
230 pw_buf0[0] = pws[gid].i[0];
231 pw_buf0[1] = pws[gid].i[1];
232 pw_buf0[2] = pws[gid].i[2];
233 pw_buf0[3] = pws[gid].i[3];
234 pw_buf1[0] = pws[gid].i[4];
235 pw_buf1[1] = pws[gid].i[5];
236 pw_buf1[2] = pws[gid].i[6];
237 pw_buf1[3] = pws[gid].i[7];
239 const u32 pw_len = pws[gid].pw_len;
250 salt_buf0[0] = salt_bufs[salt_pos].salt_buf[ 0];
251 salt_buf0[1] = salt_bufs[salt_pos].salt_buf[ 1];
252 salt_buf0[2] = salt_bufs[salt_pos].salt_buf[ 2];
253 salt_buf0[3] = salt_bufs[salt_pos].salt_buf[ 3];
254 salt_buf1[0] = salt_bufs[salt_pos].salt_buf[ 4];
255 salt_buf1[1] = salt_bufs[salt_pos].salt_buf[ 5];
256 salt_buf1[2] = salt_bufs[salt_pos].salt_buf[ 6];
257 salt_buf1[3] = salt_bufs[salt_pos].salt_buf[ 7];
258 salt_buf2[0] = salt_bufs[salt_pos].salt_buf[ 8];
259 salt_buf2[1] = salt_bufs[salt_pos].salt_buf[ 9];
260 salt_buf2[2] = salt_bufs[salt_pos].salt_buf[10];
261 salt_buf2[3] = salt_bufs[salt_pos].salt_buf[11];
262 salt_buf3[0] = salt_bufs[salt_pos].salt_buf[12];
263 salt_buf3[1] = salt_bufs[salt_pos].salt_buf[13];
264 salt_buf3[2] = salt_bufs[salt_pos].salt_buf[14];
265 salt_buf3[3] = salt_bufs[salt_pos].salt_buf[15];
267 const u32 salt_len = salt_bufs[salt_pos].salt_len;
273 const u32 search[4] =
275 digests_buf[digests_offset].digest_buf[DGST_R0],
276 digests_buf[digests_offset].digest_buf[DGST_R1],
277 digests_buf[digests_offset].digest_buf[DGST_R2],
278 digests_buf[digests_offset].digest_buf[DGST_R3]
285 for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
292 const u32x out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
294 make_unicode (w1, w2, w3);
295 make_unicode (w0, w0, w1);
297 const u32x out_len2 = out_len * 2;
303 const u32x out_salt_len = out_len2 + salt_len;
305 switch_buffer_by_offset_le_VV (w0, w1, w2, w3, salt_len);
307 w0[0] |= salt_buf0[0];
308 w0[1] |= salt_buf0[1];
309 w0[2] |= salt_buf0[2];
310 w0[3] |= salt_buf0[3];
311 w1[0] |= salt_buf1[0];
312 w1[1] |= salt_buf1[1];
313 w1[2] |= salt_buf1[2];
314 w1[3] |= salt_buf1[3];
315 w2[0] |= salt_buf2[0];
316 w2[1] |= salt_buf2[1];
317 w2[2] |= salt_buf2[2];
318 w2[3] |= salt_buf2[3];
319 w3[0] |= salt_buf3[0];
320 w3[1] |= salt_buf3[1];
321 w3[2] = out_salt_len * 8;
324 append_0x80_4x4_VV (w0, w1, w2, w3, out_salt_len);
335 MD5_STEP (MD5_Fo, a, b, c, d, w0[0], MD5C00, MD5S00);
336 MD5_STEP (MD5_Fo, d, a, b, c, w0[1], MD5C01, MD5S01);
337 MD5_STEP (MD5_Fo, c, d, a, b, w0[2], MD5C02, MD5S02);
338 MD5_STEP (MD5_Fo, b, c, d, a, w0[3], MD5C03, MD5S03);
339 MD5_STEP (MD5_Fo, a, b, c, d, w1[0], MD5C04, MD5S00);
340 MD5_STEP (MD5_Fo, d, a, b, c, w1[1], MD5C05, MD5S01);
341 MD5_STEP (MD5_Fo, c, d, a, b, w1[2], MD5C06, MD5S02);
342 MD5_STEP (MD5_Fo, b, c, d, a, w1[3], MD5C07, MD5S03);
343 MD5_STEP (MD5_Fo, a, b, c, d, w2[0], MD5C08, MD5S00);
344 MD5_STEP (MD5_Fo, d, a, b, c, w2[1], MD5C09, MD5S01);
345 MD5_STEP (MD5_Fo, c, d, a, b, w2[2], MD5C0a, MD5S02);
346 MD5_STEP (MD5_Fo, b, c, d, a, w2[3], MD5C0b, MD5S03);
347 MD5_STEP (MD5_Fo, a, b, c, d, w3[0], MD5C0c, MD5S00);
348 MD5_STEP (MD5_Fo, d, a, b, c, w3[1], MD5C0d, MD5S01);
349 MD5_STEP (MD5_Fo, c, d, a, b, w3[2], MD5C0e, MD5S02);
350 MD5_STEP (MD5_Fo, b, c, d, a, w3[3], MD5C0f, MD5S03);
352 MD5_STEP (MD5_Go, a, b, c, d, w0[1], MD5C10, MD5S10);
353 MD5_STEP (MD5_Go, d, a, b, c, w1[2], MD5C11, MD5S11);
354 MD5_STEP (MD5_Go, c, d, a, b, w2[3], MD5C12, MD5S12);
355 MD5_STEP (MD5_Go, b, c, d, a, w0[0], MD5C13, MD5S13);
356 MD5_STEP (MD5_Go, a, b, c, d, w1[1], MD5C14, MD5S10);
357 MD5_STEP (MD5_Go, d, a, b, c, w2[2], MD5C15, MD5S11);
358 MD5_STEP (MD5_Go, c, d, a, b, w3[3], MD5C16, MD5S12);
359 MD5_STEP (MD5_Go, b, c, d, a, w1[0], MD5C17, MD5S13);
360 MD5_STEP (MD5_Go, a, b, c, d, w2[1], MD5C18, MD5S10);
361 MD5_STEP (MD5_Go, d, a, b, c, w3[2], MD5C19, MD5S11);
362 MD5_STEP (MD5_Go, c, d, a, b, w0[3], MD5C1a, MD5S12);
363 MD5_STEP (MD5_Go, b, c, d, a, w2[0], MD5C1b, MD5S13);
364 MD5_STEP (MD5_Go, a, b, c, d, w3[1], MD5C1c, MD5S10);
365 MD5_STEP (MD5_Go, d, a, b, c, w0[2], MD5C1d, MD5S11);
366 MD5_STEP (MD5_Go, c, d, a, b, w1[3], MD5C1e, MD5S12);
367 MD5_STEP (MD5_Go, b, c, d, a, w3[0], MD5C1f, MD5S13);
369 MD5_STEP (MD5_H , a, b, c, d, w1[1], MD5C20, MD5S20);
370 MD5_STEP (MD5_H , d, a, b, c, w2[0], MD5C21, MD5S21);
371 MD5_STEP (MD5_H , c, d, a, b, w2[3], MD5C22, MD5S22);
372 MD5_STEP (MD5_H , b, c, d, a, w3[2], MD5C23, MD5S23);
373 MD5_STEP (MD5_H , a, b, c, d, w0[1], MD5C24, MD5S20);
374 MD5_STEP (MD5_H , d, a, b, c, w1[0], MD5C25, MD5S21);
375 MD5_STEP (MD5_H , c, d, a, b, w1[3], MD5C26, MD5S22);
376 MD5_STEP (MD5_H , b, c, d, a, w2[2], MD5C27, MD5S23);
377 MD5_STEP (MD5_H , a, b, c, d, w3[1], MD5C28, MD5S20);
378 MD5_STEP (MD5_H , d, a, b, c, w0[0], MD5C29, MD5S21);
379 MD5_STEP (MD5_H , c, d, a, b, w0[3], MD5C2a, MD5S22);
380 MD5_STEP (MD5_H , b, c, d, a, w1[2], MD5C2b, MD5S23);
381 MD5_STEP (MD5_H , a, b, c, d, w2[1], MD5C2c, MD5S20);
382 MD5_STEP (MD5_H , d, a, b, c, w3[0], MD5C2d, MD5S21);
383 MD5_STEP (MD5_H , c, d, a, b, w3[3], MD5C2e, MD5S22);
384 MD5_STEP (MD5_H , b, c, d, a, w0[2], MD5C2f, MD5S23);
386 MD5_STEP (MD5_I , a, b, c, d, w0[0], MD5C30, MD5S30);
387 MD5_STEP (MD5_I , d, a, b, c, w1[3], MD5C31, MD5S31);
388 MD5_STEP (MD5_I , c, d, a, b, w3[2], MD5C32, MD5S32);
389 MD5_STEP (MD5_I , b, c, d, a, w1[1], MD5C33, MD5S33);
390 MD5_STEP (MD5_I , a, b, c, d, w3[0], MD5C34, MD5S30);
391 MD5_STEP (MD5_I , d, a, b, c, w0[3], MD5C35, MD5S31);
392 MD5_STEP (MD5_I , c, d, a, b, w2[2], MD5C36, MD5S32);
393 MD5_STEP (MD5_I , b, c, d, a, w0[1], MD5C37, MD5S33);
394 MD5_STEP (MD5_I , a, b, c, d, w2[0], MD5C38, MD5S30);
395 MD5_STEP (MD5_I , d, a, b, c, w3[3], MD5C39, MD5S31);
396 MD5_STEP (MD5_I , c, d, a, b, w1[2], MD5C3a, MD5S32);
397 MD5_STEP (MD5_I , b, c, d, a, w3[1], MD5C3b, MD5S33);
398 MD5_STEP (MD5_I , a, b, c, d, w1[0], MD5C3c, MD5S30);
400 if (MATCHES_NONE_VS (a, search[0])) continue;
402 MD5_STEP (MD5_I , d, a, b, c, w2[3], MD5C3d, MD5S31);
403 MD5_STEP (MD5_I , c, d, a, b, w0[2], MD5C3e, MD5S32);
404 MD5_STEP (MD5_I , b, c, d, a, w2[1], MD5C3f, MD5S33);
406 COMPARE_S_SIMD (a, d, c, b);
410 __kernel void m00040_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_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_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)
414 __kernel void m00040_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_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_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)