2 * Author......: Jens Steube <jens.steube@gmail.com>
8 #include "include/constants.h"
9 #include "include/kernel_vendor.h"
16 #include "include/kernel_functions.c"
17 #include "types_ocl.c"
20 #define COMPARE_S "check_single_comp4.c"
21 #define COMPARE_M "check_multi_comp4.c"
23 __constant u32 padding[8] =
43 static void swap (__local RC4_KEY *rc4_key, const u8 i, const u8 j)
48 rc4_key->S[i] = rc4_key->S[j];
52 static void rc4_init_16 (__local RC4_KEY *rc4_key, const u32 data[4])
57 __local u32 *ptr = (__local u32 *) rc4_key->S;
60 for (u32 i = 0; i < 64; i++)
65 const u32 d0 = data[0] >> 0;
66 const u32 d1 = data[0] >> 8;
67 const u32 d2 = data[0] >> 16;
68 const u32 d3 = data[0] >> 24;
69 const u32 d4 = data[1] >> 0;
74 for (u32 i = 0; i < 255; i += 5)
76 j += rc4_key->S[i + 0] + d0; swap (rc4_key, i + 0, j);
77 j += rc4_key->S[i + 1] + d1; swap (rc4_key, i + 1, j);
78 j += rc4_key->S[i + 2] + d2; swap (rc4_key, i + 2, j);
79 j += rc4_key->S[i + 3] + d3; swap (rc4_key, i + 3, j);
80 j += rc4_key->S[i + 4] + d4; swap (rc4_key, i + 4, j);
83 j += rc4_key->S[255] + d0; swap (rc4_key, 255, j);
86 static u8 rc4_next_16 (__local RC4_KEY *rc4_key, u8 i, u8 j, __constant u32 in[4], u32 out[4])
88 for (u32 k = 0; k < 4; k++)
99 idx = rc4_key->S[i] + rc4_key->S[j];
101 xor4 |= rc4_key->S[idx] << 0;
106 swap (rc4_key, i, j);
108 idx = rc4_key->S[i] + rc4_key->S[j];
110 xor4 |= rc4_key->S[idx] << 8;
115 swap (rc4_key, i, j);
117 idx = rc4_key->S[i] + rc4_key->S[j];
119 xor4 |= rc4_key->S[idx] << 16;
124 swap (rc4_key, i, j);
126 idx = rc4_key->S[i] + rc4_key->S[j];
128 xor4 |= rc4_key->S[idx] << 24;
130 out[k] = in[k] ^ xor4;
136 __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m10410_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 pdf_t *pdf_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)
142 const u32 lid = get_local_id (0);
148 const u32 gid = get_global_id (0);
150 if (gid >= gid_max) return;
154 wordl0[0] = pws[gid].i[ 0];
155 wordl0[1] = pws[gid].i[ 1];
156 wordl0[2] = pws[gid].i[ 2];
157 wordl0[3] = pws[gid].i[ 3];
161 wordl1[0] = pws[gid].i[ 4];
162 wordl1[1] = pws[gid].i[ 5];
163 wordl1[2] = pws[gid].i[ 6];
164 wordl1[3] = pws[gid].i[ 7];
180 const u32 pw_l_len = pws[gid].pw_len;
182 if (combs_mode == COMBINATOR_MODE_BASE_RIGHT)
184 switch_buffer_by_offset (wordl0, wordl1, wordl2, wordl3, combs_buf[0].pw_len);
191 __local RC4_KEY rc4_keys[64];
192 __local RC4_KEY *rc4_key = &rc4_keys[lid];
198 for (u32 il_pos = 0; il_pos < combs_cnt; il_pos++)
202 wordr0[0] = combs_buf[il_pos].i[0];
203 wordr0[1] = combs_buf[il_pos].i[1];
204 wordr0[2] = combs_buf[il_pos].i[2];
205 wordr0[3] = combs_buf[il_pos].i[3];
209 wordr1[0] = combs_buf[il_pos].i[4];
210 wordr1[1] = combs_buf[il_pos].i[5];
211 wordr1[2] = combs_buf[il_pos].i[6];
212 wordr1[3] = combs_buf[il_pos].i[7];
228 if (combs_mode == COMBINATOR_MODE_BASE_LEFT)
230 switch_buffer_by_offset (wordr0, wordr1, wordr2, wordr3, pw_l_len);
235 w0[0] = wordl0[0] | wordr0[0];
236 w0[1] = wordl0[1] | wordr0[1];
247 rc4_init_16 (rc4_key, key);
251 rc4_next_16 (rc4_key, 0, 0, padding, out);
253 const u32 r0 = out[0];
254 const u32 r1 = out[1];
255 const u32 r2 = out[2];
256 const u32 r3 = out[3];
262 __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m10410_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 pdf_t *pdf_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)
266 __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m10410_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 pdf_t *pdf_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)
270 __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m10410_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 pdf_t *pdf_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)
276 const u32 lid = get_local_id (0);
282 const u32 gid = get_global_id (0);
284 if (gid >= gid_max) return;
288 wordl0[0] = pws[gid].i[ 0];
289 wordl0[1] = pws[gid].i[ 1];
290 wordl0[2] = pws[gid].i[ 2];
291 wordl0[3] = pws[gid].i[ 3];
295 wordl1[0] = pws[gid].i[ 4];
296 wordl1[1] = pws[gid].i[ 5];
297 wordl1[2] = pws[gid].i[ 6];
298 wordl1[3] = pws[gid].i[ 7];
314 const u32 pw_l_len = pws[gid].pw_len;
316 if (combs_mode == COMBINATOR_MODE_BASE_RIGHT)
318 switch_buffer_by_offset (wordl0, wordl1, wordl2, wordl3, combs_buf[0].pw_len);
325 const u32 search[4] =
327 digests_buf[digests_offset].digest_buf[DGST_R0],
328 digests_buf[digests_offset].digest_buf[DGST_R1],
329 digests_buf[digests_offset].digest_buf[DGST_R2],
330 digests_buf[digests_offset].digest_buf[DGST_R3]
337 __local RC4_KEY rc4_keys[64];
338 __local RC4_KEY *rc4_key = &rc4_keys[lid];
344 for (u32 il_pos = 0; il_pos < combs_cnt; il_pos++)
348 wordr0[0] = combs_buf[il_pos].i[0];
349 wordr0[1] = combs_buf[il_pos].i[1];
350 wordr0[2] = combs_buf[il_pos].i[2];
351 wordr0[3] = combs_buf[il_pos].i[3];
355 wordr1[0] = combs_buf[il_pos].i[4];
356 wordr1[1] = combs_buf[il_pos].i[5];
357 wordr1[2] = combs_buf[il_pos].i[6];
358 wordr1[3] = combs_buf[il_pos].i[7];
374 if (combs_mode == COMBINATOR_MODE_BASE_LEFT)
376 switch_buffer_by_offset (wordr0, wordr1, wordr2, wordr3, pw_l_len);
381 w0[0] = wordl0[0] | wordr0[0];
382 w0[1] = wordl0[1] | wordr0[1];
393 rc4_init_16 (rc4_key, key);
397 rc4_next_16 (rc4_key, 0, 0, padding, out);
399 const u32 r0 = out[0];
400 const u32 r1 = out[1];
401 const u32 r2 = out[2];
402 const u32 r3 = out[3];
408 __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m10410_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 pdf_t *pdf_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)
412 __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m10410_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 pdf_t *pdf_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)