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 __constant u32 padding[8] =
31 void md5_transform (const u32x w0[4], const u32x w1[4], const u32x w2[4], const u32x w3[4], u32x digest[4])
55 MD5_STEP (MD5_Fo, a, b, c, d, w0_t, MD5C00, MD5S00);
56 MD5_STEP (MD5_Fo, d, a, b, c, w1_t, MD5C01, MD5S01);
57 MD5_STEP (MD5_Fo, c, d, a, b, w2_t, MD5C02, MD5S02);
58 MD5_STEP (MD5_Fo, b, c, d, a, w3_t, MD5C03, MD5S03);
59 MD5_STEP (MD5_Fo, a, b, c, d, w4_t, MD5C04, MD5S00);
60 MD5_STEP (MD5_Fo, d, a, b, c, w5_t, MD5C05, MD5S01);
61 MD5_STEP (MD5_Fo, c, d, a, b, w6_t, MD5C06, MD5S02);
62 MD5_STEP (MD5_Fo, b, c, d, a, w7_t, MD5C07, MD5S03);
63 MD5_STEP (MD5_Fo, a, b, c, d, w8_t, MD5C08, MD5S00);
64 MD5_STEP (MD5_Fo, d, a, b, c, w9_t, MD5C09, MD5S01);
65 MD5_STEP (MD5_Fo, c, d, a, b, wa_t, MD5C0a, MD5S02);
66 MD5_STEP (MD5_Fo, b, c, d, a, wb_t, MD5C0b, MD5S03);
67 MD5_STEP (MD5_Fo, a, b, c, d, wc_t, MD5C0c, MD5S00);
68 MD5_STEP (MD5_Fo, d, a, b, c, wd_t, MD5C0d, MD5S01);
69 MD5_STEP (MD5_Fo, c, d, a, b, we_t, MD5C0e, MD5S02);
70 MD5_STEP (MD5_Fo, b, c, d, a, wf_t, MD5C0f, MD5S03);
72 MD5_STEP (MD5_Go, a, b, c, d, w1_t, MD5C10, MD5S10);
73 MD5_STEP (MD5_Go, d, a, b, c, w6_t, MD5C11, MD5S11);
74 MD5_STEP (MD5_Go, c, d, a, b, wb_t, MD5C12, MD5S12);
75 MD5_STEP (MD5_Go, b, c, d, a, w0_t, MD5C13, MD5S13);
76 MD5_STEP (MD5_Go, a, b, c, d, w5_t, MD5C14, MD5S10);
77 MD5_STEP (MD5_Go, d, a, b, c, wa_t, MD5C15, MD5S11);
78 MD5_STEP (MD5_Go, c, d, a, b, wf_t, MD5C16, MD5S12);
79 MD5_STEP (MD5_Go, b, c, d, a, w4_t, MD5C17, MD5S13);
80 MD5_STEP (MD5_Go, a, b, c, d, w9_t, MD5C18, MD5S10);
81 MD5_STEP (MD5_Go, d, a, b, c, we_t, MD5C19, MD5S11);
82 MD5_STEP (MD5_Go, c, d, a, b, w3_t, MD5C1a, MD5S12);
83 MD5_STEP (MD5_Go, b, c, d, a, w8_t, MD5C1b, MD5S13);
84 MD5_STEP (MD5_Go, a, b, c, d, wd_t, MD5C1c, MD5S10);
85 MD5_STEP (MD5_Go, d, a, b, c, w2_t, MD5C1d, MD5S11);
86 MD5_STEP (MD5_Go, c, d, a, b, w7_t, MD5C1e, MD5S12);
87 MD5_STEP (MD5_Go, b, c, d, a, wc_t, MD5C1f, MD5S13);
89 MD5_STEP (MD5_H , a, b, c, d, w5_t, MD5C20, MD5S20);
90 MD5_STEP (MD5_H , d, a, b, c, w8_t, MD5C21, MD5S21);
91 MD5_STEP (MD5_H , c, d, a, b, wb_t, MD5C22, MD5S22);
92 MD5_STEP (MD5_H , b, c, d, a, we_t, MD5C23, MD5S23);
93 MD5_STEP (MD5_H , a, b, c, d, w1_t, MD5C24, MD5S20);
94 MD5_STEP (MD5_H , d, a, b, c, w4_t, MD5C25, MD5S21);
95 MD5_STEP (MD5_H , c, d, a, b, w7_t, MD5C26, MD5S22);
96 MD5_STEP (MD5_H , b, c, d, a, wa_t, MD5C27, MD5S23);
97 MD5_STEP (MD5_H , a, b, c, d, wd_t, MD5C28, MD5S20);
98 MD5_STEP (MD5_H , d, a, b, c, w0_t, MD5C29, MD5S21);
99 MD5_STEP (MD5_H , c, d, a, b, w3_t, MD5C2a, MD5S22);
100 MD5_STEP (MD5_H , b, c, d, a, w6_t, MD5C2b, MD5S23);
101 MD5_STEP (MD5_H , a, b, c, d, w9_t, MD5C2c, MD5S20);
102 MD5_STEP (MD5_H , d, a, b, c, wc_t, MD5C2d, MD5S21);
103 MD5_STEP (MD5_H , c, d, a, b, wf_t, MD5C2e, MD5S22);
104 MD5_STEP (MD5_H , b, c, d, a, w2_t, MD5C2f, MD5S23);
106 MD5_STEP (MD5_I , a, b, c, d, w0_t, MD5C30, MD5S30);
107 MD5_STEP (MD5_I , d, a, b, c, w7_t, MD5C31, MD5S31);
108 MD5_STEP (MD5_I , c, d, a, b, we_t, MD5C32, MD5S32);
109 MD5_STEP (MD5_I , b, c, d, a, w5_t, MD5C33, MD5S33);
110 MD5_STEP (MD5_I , a, b, c, d, wc_t, MD5C34, MD5S30);
111 MD5_STEP (MD5_I , d, a, b, c, w3_t, MD5C35, MD5S31);
112 MD5_STEP (MD5_I , c, d, a, b, wa_t, MD5C36, MD5S32);
113 MD5_STEP (MD5_I , b, c, d, a, w1_t, MD5C37, MD5S33);
114 MD5_STEP (MD5_I , a, b, c, d, w8_t, MD5C38, MD5S30);
115 MD5_STEP (MD5_I , d, a, b, c, wf_t, MD5C39, MD5S31);
116 MD5_STEP (MD5_I , c, d, a, b, w6_t, MD5C3a, MD5S32);
117 MD5_STEP (MD5_I , b, c, d, a, wd_t, MD5C3b, MD5S33);
118 MD5_STEP (MD5_I , a, b, c, d, w4_t, MD5C3c, MD5S30);
119 MD5_STEP (MD5_I , d, a, b, c, wb_t, MD5C3d, MD5S31);
120 MD5_STEP (MD5_I , c, d, a, b, w2_t, MD5C3e, MD5S32);
121 MD5_STEP (MD5_I , b, c, d, a, w9_t, MD5C3f, MD5S33);
129 __kernel void m10420_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 pdf_t *pdf_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)
135 const u32 lid = get_local_id (0);
141 const u32 gid = get_global_id (0);
143 if (gid >= gid_max) return;
148 pw_buf0[0] = pws[gid].i[ 0];
149 pw_buf0[1] = pws[gid].i[ 1];
150 pw_buf0[2] = pws[gid].i[ 2];
151 pw_buf0[3] = pws[gid].i[ 3];
152 pw_buf1[0] = pws[gid].i[ 4];
153 pw_buf1[1] = pws[gid].i[ 5];
154 pw_buf1[2] = pws[gid].i[ 6];
155 pw_buf1[3] = pws[gid].i[ 7];
157 const u32 pw_len = pws[gid].pw_len;
165 o_buf[0] = pdf_bufs[salt_pos].o_buf[0];
166 o_buf[1] = pdf_bufs[salt_pos].o_buf[1];
167 o_buf[2] = pdf_bufs[salt_pos].o_buf[2];
168 o_buf[3] = pdf_bufs[salt_pos].o_buf[3];
169 o_buf[4] = pdf_bufs[salt_pos].o_buf[4];
170 o_buf[5] = pdf_bufs[salt_pos].o_buf[5];
171 o_buf[6] = pdf_bufs[salt_pos].o_buf[6];
172 o_buf[7] = pdf_bufs[salt_pos].o_buf[7];
174 u32 P = pdf_bufs[salt_pos].P;
178 id_buf[0] = pdf_bufs[salt_pos].id_buf[0];
179 id_buf[1] = pdf_bufs[salt_pos].id_buf[1];
180 id_buf[2] = pdf_bufs[salt_pos].id_buf[2];
181 id_buf[3] = pdf_bufs[salt_pos].id_buf[3];
187 for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
194 const u32x out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
222 switch_buffer_by_offset_le (p0, p1, p2, p3, out_len);
225 // truncate at 32 is wanted, not a bug!
252 md5_transform (w0, w1, w2, w3, digest);
271 md5_transform (w0, w1, w2, w3, digest);
274 u32x b = digest[1] & 0xff;
278 COMPARE_M_SIMD (a, b, c, d);
282 __kernel void m10420_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 pdf_t *pdf_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 rules__cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
286 __kernel void m10420_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 pdf_t *pdf_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)
290 __kernel void m10420_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 pdf_t *pdf_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)
296 const u32 lid = get_local_id (0);
302 const u32 gid = get_global_id (0);
304 if (gid >= gid_max) return;
309 pw_buf0[0] = pws[gid].i[ 0];
310 pw_buf0[1] = pws[gid].i[ 1];
311 pw_buf0[2] = pws[gid].i[ 2];
312 pw_buf0[3] = pws[gid].i[ 3];
313 pw_buf1[0] = pws[gid].i[ 4];
314 pw_buf1[1] = pws[gid].i[ 5];
315 pw_buf1[2] = pws[gid].i[ 6];
316 pw_buf1[3] = pws[gid].i[ 7];
318 const u32 pw_len = pws[gid].pw_len;
326 o_buf[0] = pdf_bufs[salt_pos].o_buf[0];
327 o_buf[1] = pdf_bufs[salt_pos].o_buf[1];
328 o_buf[2] = pdf_bufs[salt_pos].o_buf[2];
329 o_buf[3] = pdf_bufs[salt_pos].o_buf[3];
330 o_buf[4] = pdf_bufs[salt_pos].o_buf[4];
331 o_buf[5] = pdf_bufs[salt_pos].o_buf[5];
332 o_buf[6] = pdf_bufs[salt_pos].o_buf[6];
333 o_buf[7] = pdf_bufs[salt_pos].o_buf[7];
335 u32 P = pdf_bufs[salt_pos].P;
339 id_buf[0] = pdf_bufs[salt_pos].id_buf[0];
340 id_buf[1] = pdf_bufs[salt_pos].id_buf[1];
341 id_buf[2] = pdf_bufs[salt_pos].id_buf[2];
342 id_buf[3] = pdf_bufs[salt_pos].id_buf[3];
348 const u32 search[4] =
350 digests_buf[digests_offset].digest_buf[DGST_R0],
351 digests_buf[digests_offset].digest_buf[DGST_R1],
352 digests_buf[digests_offset].digest_buf[DGST_R2],
353 digests_buf[digests_offset].digest_buf[DGST_R3]
360 for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
367 const u32x out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
395 switch_buffer_by_offset_le (p0, p1, p2, p3, out_len);
398 // truncate at 32 is wanted, not a bug!
425 md5_transform (w0, w1, w2, w3, digest);
444 md5_transform (w0, w1, w2, w3, digest);
447 u32x b = digest[1] & 0xff;
451 COMPARE_S_SIMD (a, b, c, d);
455 __kernel void m10420_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 pdf_t *pdf_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)
459 __kernel void m10420_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 pdf_t *pdf_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)