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"
15 #include "inc_simd.cl"
17 #define COMPARE_S "inc_comp_single.cl"
18 #define COMPARE_M "inc_comp_multi.cl"
20 void md5_transform_S (const u32 w0[4], const u32 w1[4], const u32 w2[4], const u32 w3[4], u32 digest[4])
44 MD5_STEP_S (MD5_Fo, a, b, c, d, w0_t, MD5C00, MD5S00);
45 MD5_STEP_S (MD5_Fo, d, a, b, c, w1_t, MD5C01, MD5S01);
46 MD5_STEP_S (MD5_Fo, c, d, a, b, w2_t, MD5C02, MD5S02);
47 MD5_STEP_S (MD5_Fo, b, c, d, a, w3_t, MD5C03, MD5S03);
48 MD5_STEP_S (MD5_Fo, a, b, c, d, w4_t, MD5C04, MD5S00);
49 MD5_STEP_S (MD5_Fo, d, a, b, c, w5_t, MD5C05, MD5S01);
50 MD5_STEP_S (MD5_Fo, c, d, a, b, w6_t, MD5C06, MD5S02);
51 MD5_STEP_S (MD5_Fo, b, c, d, a, w7_t, MD5C07, MD5S03);
52 MD5_STEP_S (MD5_Fo, a, b, c, d, w8_t, MD5C08, MD5S00);
53 MD5_STEP_S (MD5_Fo, d, a, b, c, w9_t, MD5C09, MD5S01);
54 MD5_STEP_S (MD5_Fo, c, d, a, b, wa_t, MD5C0a, MD5S02);
55 MD5_STEP_S (MD5_Fo, b, c, d, a, wb_t, MD5C0b, MD5S03);
56 MD5_STEP_S (MD5_Fo, a, b, c, d, wc_t, MD5C0c, MD5S00);
57 MD5_STEP_S (MD5_Fo, d, a, b, c, wd_t, MD5C0d, MD5S01);
58 MD5_STEP_S (MD5_Fo, c, d, a, b, we_t, MD5C0e, MD5S02);
59 MD5_STEP_S (MD5_Fo, b, c, d, a, wf_t, MD5C0f, MD5S03);
61 MD5_STEP_S (MD5_Go, a, b, c, d, w1_t, MD5C10, MD5S10);
62 MD5_STEP_S (MD5_Go, d, a, b, c, w6_t, MD5C11, MD5S11);
63 MD5_STEP_S (MD5_Go, c, d, a, b, wb_t, MD5C12, MD5S12);
64 MD5_STEP_S (MD5_Go, b, c, d, a, w0_t, MD5C13, MD5S13);
65 MD5_STEP_S (MD5_Go, a, b, c, d, w5_t, MD5C14, MD5S10);
66 MD5_STEP_S (MD5_Go, d, a, b, c, wa_t, MD5C15, MD5S11);
67 MD5_STEP_S (MD5_Go, c, d, a, b, wf_t, MD5C16, MD5S12);
68 MD5_STEP_S (MD5_Go, b, c, d, a, w4_t, MD5C17, MD5S13);
69 MD5_STEP_S (MD5_Go, a, b, c, d, w9_t, MD5C18, MD5S10);
70 MD5_STEP_S (MD5_Go, d, a, b, c, we_t, MD5C19, MD5S11);
71 MD5_STEP_S (MD5_Go, c, d, a, b, w3_t, MD5C1a, MD5S12);
72 MD5_STEP_S (MD5_Go, b, c, d, a, w8_t, MD5C1b, MD5S13);
73 MD5_STEP_S (MD5_Go, a, b, c, d, wd_t, MD5C1c, MD5S10);
74 MD5_STEP_S (MD5_Go, d, a, b, c, w2_t, MD5C1d, MD5S11);
75 MD5_STEP_S (MD5_Go, c, d, a, b, w7_t, MD5C1e, MD5S12);
76 MD5_STEP_S (MD5_Go, b, c, d, a, wc_t, MD5C1f, MD5S13);
78 MD5_STEP_S (MD5_H , a, b, c, d, w5_t, MD5C20, MD5S20);
79 MD5_STEP_S (MD5_H , d, a, b, c, w8_t, MD5C21, MD5S21);
80 MD5_STEP_S (MD5_H , c, d, a, b, wb_t, MD5C22, MD5S22);
81 MD5_STEP_S (MD5_H , b, c, d, a, we_t, MD5C23, MD5S23);
82 MD5_STEP_S (MD5_H , a, b, c, d, w1_t, MD5C24, MD5S20);
83 MD5_STEP_S (MD5_H , d, a, b, c, w4_t, MD5C25, MD5S21);
84 MD5_STEP_S (MD5_H , c, d, a, b, w7_t, MD5C26, MD5S22);
85 MD5_STEP_S (MD5_H , b, c, d, a, wa_t, MD5C27, MD5S23);
86 MD5_STEP_S (MD5_H , a, b, c, d, wd_t, MD5C28, MD5S20);
87 MD5_STEP_S (MD5_H , d, a, b, c, w0_t, MD5C29, MD5S21);
88 MD5_STEP_S (MD5_H , c, d, a, b, w3_t, MD5C2a, MD5S22);
89 MD5_STEP_S (MD5_H , b, c, d, a, w6_t, MD5C2b, MD5S23);
90 MD5_STEP_S (MD5_H , a, b, c, d, w9_t, MD5C2c, MD5S20);
91 MD5_STEP_S (MD5_H , d, a, b, c, wc_t, MD5C2d, MD5S21);
92 MD5_STEP_S (MD5_H , c, d, a, b, wf_t, MD5C2e, MD5S22);
93 MD5_STEP_S (MD5_H , b, c, d, a, w2_t, MD5C2f, MD5S23);
95 MD5_STEP_S (MD5_I , a, b, c, d, w0_t, MD5C30, MD5S30);
96 MD5_STEP_S (MD5_I , d, a, b, c, w7_t, MD5C31, MD5S31);
97 MD5_STEP_S (MD5_I , c, d, a, b, we_t, MD5C32, MD5S32);
98 MD5_STEP_S (MD5_I , b, c, d, a, w5_t, MD5C33, MD5S33);
99 MD5_STEP_S (MD5_I , a, b, c, d, wc_t, MD5C34, MD5S30);
100 MD5_STEP_S (MD5_I , d, a, b, c, w3_t, MD5C35, MD5S31);
101 MD5_STEP_S (MD5_I , c, d, a, b, wa_t, MD5C36, MD5S32);
102 MD5_STEP_S (MD5_I , b, c, d, a, w1_t, MD5C37, MD5S33);
103 MD5_STEP_S (MD5_I , a, b, c, d, w8_t, MD5C38, MD5S30);
104 MD5_STEP_S (MD5_I , d, a, b, c, wf_t, MD5C39, MD5S31);
105 MD5_STEP_S (MD5_I , c, d, a, b, w6_t, MD5C3a, MD5S32);
106 MD5_STEP_S (MD5_I , b, c, d, a, wd_t, MD5C3b, MD5S33);
107 MD5_STEP_S (MD5_I , a, b, c, d, w4_t, MD5C3c, MD5S30);
108 MD5_STEP_S (MD5_I , d, a, b, c, wb_t, MD5C3d, MD5S31);
109 MD5_STEP_S (MD5_I , c, d, a, b, w2_t, MD5C3e, MD5S32);
110 MD5_STEP_S (MD5_I , b, c, d, a, w9_t, MD5C3f, MD5S33);
118 void md5_transform_V (const u32x w0[4], const u32x w1[4], const u32x w2[4], const u32x w3[4], u32x digest[4])
142 MD5_STEP (MD5_Fo, a, b, c, d, w0_t, MD5C00, MD5S00);
143 MD5_STEP (MD5_Fo, d, a, b, c, w1_t, MD5C01, MD5S01);
144 MD5_STEP (MD5_Fo, c, d, a, b, w2_t, MD5C02, MD5S02);
145 MD5_STEP (MD5_Fo, b, c, d, a, w3_t, MD5C03, MD5S03);
146 MD5_STEP (MD5_Fo, a, b, c, d, w4_t, MD5C04, MD5S00);
147 MD5_STEP (MD5_Fo, d, a, b, c, w5_t, MD5C05, MD5S01);
148 MD5_STEP (MD5_Fo, c, d, a, b, w6_t, MD5C06, MD5S02);
149 MD5_STEP (MD5_Fo, b, c, d, a, w7_t, MD5C07, MD5S03);
150 MD5_STEP (MD5_Fo, a, b, c, d, w8_t, MD5C08, MD5S00);
151 MD5_STEP (MD5_Fo, d, a, b, c, w9_t, MD5C09, MD5S01);
152 MD5_STEP (MD5_Fo, c, d, a, b, wa_t, MD5C0a, MD5S02);
153 MD5_STEP (MD5_Fo, b, c, d, a, wb_t, MD5C0b, MD5S03);
154 MD5_STEP (MD5_Fo, a, b, c, d, wc_t, MD5C0c, MD5S00);
155 MD5_STEP (MD5_Fo, d, a, b, c, wd_t, MD5C0d, MD5S01);
156 MD5_STEP (MD5_Fo, c, d, a, b, we_t, MD5C0e, MD5S02);
157 MD5_STEP (MD5_Fo, b, c, d, a, wf_t, MD5C0f, MD5S03);
159 MD5_STEP (MD5_Go, a, b, c, d, w1_t, MD5C10, MD5S10);
160 MD5_STEP (MD5_Go, d, a, b, c, w6_t, MD5C11, MD5S11);
161 MD5_STEP (MD5_Go, c, d, a, b, wb_t, MD5C12, MD5S12);
162 MD5_STEP (MD5_Go, b, c, d, a, w0_t, MD5C13, MD5S13);
163 MD5_STEP (MD5_Go, a, b, c, d, w5_t, MD5C14, MD5S10);
164 MD5_STEP (MD5_Go, d, a, b, c, wa_t, MD5C15, MD5S11);
165 MD5_STEP (MD5_Go, c, d, a, b, wf_t, MD5C16, MD5S12);
166 MD5_STEP (MD5_Go, b, c, d, a, w4_t, MD5C17, MD5S13);
167 MD5_STEP (MD5_Go, a, b, c, d, w9_t, MD5C18, MD5S10);
168 MD5_STEP (MD5_Go, d, a, b, c, we_t, MD5C19, MD5S11);
169 MD5_STEP (MD5_Go, c, d, a, b, w3_t, MD5C1a, MD5S12);
170 MD5_STEP (MD5_Go, b, c, d, a, w8_t, MD5C1b, MD5S13);
171 MD5_STEP (MD5_Go, a, b, c, d, wd_t, MD5C1c, MD5S10);
172 MD5_STEP (MD5_Go, d, a, b, c, w2_t, MD5C1d, MD5S11);
173 MD5_STEP (MD5_Go, c, d, a, b, w7_t, MD5C1e, MD5S12);
174 MD5_STEP (MD5_Go, b, c, d, a, wc_t, MD5C1f, MD5S13);
176 MD5_STEP (MD5_H , a, b, c, d, w5_t, MD5C20, MD5S20);
177 MD5_STEP (MD5_H , d, a, b, c, w8_t, MD5C21, MD5S21);
178 MD5_STEP (MD5_H , c, d, a, b, wb_t, MD5C22, MD5S22);
179 MD5_STEP (MD5_H , b, c, d, a, we_t, MD5C23, MD5S23);
180 MD5_STEP (MD5_H , a, b, c, d, w1_t, MD5C24, MD5S20);
181 MD5_STEP (MD5_H , d, a, b, c, w4_t, MD5C25, MD5S21);
182 MD5_STEP (MD5_H , c, d, a, b, w7_t, MD5C26, MD5S22);
183 MD5_STEP (MD5_H , b, c, d, a, wa_t, MD5C27, MD5S23);
184 MD5_STEP (MD5_H , a, b, c, d, wd_t, MD5C28, MD5S20);
185 MD5_STEP (MD5_H , d, a, b, c, w0_t, MD5C29, MD5S21);
186 MD5_STEP (MD5_H , c, d, a, b, w3_t, MD5C2a, MD5S22);
187 MD5_STEP (MD5_H , b, c, d, a, w6_t, MD5C2b, MD5S23);
188 MD5_STEP (MD5_H , a, b, c, d, w9_t, MD5C2c, MD5S20);
189 MD5_STEP (MD5_H , d, a, b, c, wc_t, MD5C2d, MD5S21);
190 MD5_STEP (MD5_H , c, d, a, b, wf_t, MD5C2e, MD5S22);
191 MD5_STEP (MD5_H , b, c, d, a, w2_t, MD5C2f, MD5S23);
193 MD5_STEP (MD5_I , a, b, c, d, w0_t, MD5C30, MD5S30);
194 MD5_STEP (MD5_I , d, a, b, c, w7_t, MD5C31, MD5S31);
195 MD5_STEP (MD5_I , c, d, a, b, we_t, MD5C32, MD5S32);
196 MD5_STEP (MD5_I , b, c, d, a, w5_t, MD5C33, MD5S33);
197 MD5_STEP (MD5_I , a, b, c, d, wc_t, MD5C34, MD5S30);
198 MD5_STEP (MD5_I , d, a, b, c, w3_t, MD5C35, MD5S31);
199 MD5_STEP (MD5_I , c, d, a, b, wa_t, MD5C36, MD5S32);
200 MD5_STEP (MD5_I , b, c, d, a, w1_t, MD5C37, MD5S33);
201 MD5_STEP (MD5_I , a, b, c, d, w8_t, MD5C38, MD5S30);
202 MD5_STEP (MD5_I , d, a, b, c, wf_t, MD5C39, MD5S31);
203 MD5_STEP (MD5_I , c, d, a, b, w6_t, MD5C3a, MD5S32);
204 MD5_STEP (MD5_I , b, c, d, a, wd_t, MD5C3b, MD5S33);
205 MD5_STEP (MD5_I , a, b, c, d, w4_t, MD5C3c, MD5S30);
206 MD5_STEP (MD5_I , d, a, b, c, wb_t, MD5C3d, MD5S31);
207 MD5_STEP (MD5_I , c, d, a, b, w2_t, MD5C3e, MD5S32);
208 MD5_STEP (MD5_I , b, c, d, a, w9_t, MD5C3f, MD5S33);
216 __kernel void m00400_init (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global phpass_tmp_t *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)
222 const u32 gid = get_global_id (0);
224 if (gid >= gid_max) return;
228 w0[0] = pws[gid].i[ 0];
229 w0[1] = pws[gid].i[ 1];
230 w0[2] = pws[gid].i[ 2];
231 w0[3] = pws[gid].i[ 3];
235 w1[0] = pws[gid].i[ 4];
236 w1[1] = pws[gid].i[ 5];
237 w1[2] = pws[gid].i[ 6];
238 w1[3] = pws[gid].i[ 7];
242 w2[0] = pws[gid].i[ 8];
243 w2[1] = pws[gid].i[ 9];
247 const u32 pw_len = pws[gid].pw_len;
255 salt_buf[0] = salt_bufs[salt_pos].salt_buf[0];
256 salt_buf[1] = salt_bufs[salt_pos].salt_buf[1];
262 u32 block_len = 8 + pw_len;
266 block0[0] = salt_buf[0];
267 block0[1] = salt_buf[1];
289 block3[2] = block_len * 8;
292 append_0x80_4x4_S (block0, block1, block2, block3, block_len);
305 md5_transform_S (block0, block1, block2, block3, digest);
307 tmps[gid].digest_buf[0] = digest[0];
308 tmps[gid].digest_buf[1] = digest[1];
309 tmps[gid].digest_buf[2] = digest[2];
310 tmps[gid].digest_buf[3] = digest[3];
313 __kernel void m00400_loop (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global phpass_tmp_t *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)
319 const u32 gid = get_global_id (0);
321 if (gid >= gid_max) return;
327 w0[0] = packv (pws, i, gid, 0);
328 w0[1] = packv (pws, i, gid, 1);
329 w0[2] = packv (pws, i, gid, 2);
330 w0[3] = packv (pws, i, gid, 3);
331 w1[0] = packv (pws, i, gid, 4);
332 w1[1] = packv (pws, i, gid, 5);
333 w1[2] = packv (pws, i, gid, 6);
334 w1[3] = packv (pws, i, gid, 7);
335 w2[0] = packv (pws, i, gid, 8);
336 w2[1] = packv (pws, i, gid, 9);
340 u32x pw_len = packvf (pws, pw_len, gid);
344 digest[0] = packv (tmps, digest_buf, gid, 0);
345 digest[1] = packv (tmps, digest_buf, gid, 1);
346 digest[2] = packv (tmps, digest_buf, gid, 2);
347 digest[3] = packv (tmps, digest_buf, gid, 3);
353 u32x block_len = (16 + pw_len);
374 block3[2] = block_len * 8;
377 append_0x80_4x4_VV (block0, block1, block2, block3, block_len);
383 for (u32 i = 0; i < loop_cnt; i++)
385 block0[0] = digest[0];
386 block0[1] = digest[1];
387 block0[2] = digest[2];
388 block0[3] = digest[3];
395 md5_transform_V (block0, block1, block2, block3, digest);
398 unpackv (tmps, digest_buf, gid, 0, digest[0]);
399 unpackv (tmps, digest_buf, gid, 1, digest[1]);
400 unpackv (tmps, digest_buf, gid, 2, digest[2]);
401 unpackv (tmps, digest_buf, gid, 3, digest[3]);
404 __kernel void m00400_comp (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global phpass_tmp_t *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)
410 const u32 gid = get_global_id (0);
411 const u32 lid = get_local_id (0);
413 if (gid >= gid_max) return;
419 const u32 r0 = tmps[gid].digest_buf[DGST_R0];
420 const u32 r1 = tmps[gid].digest_buf[DGST_R1];
421 const u32 r2 = tmps[gid].digest_buf[DGST_R2];
422 const u32 r3 = tmps[gid].digest_buf[DGST_R3];