2 * Author......: Jens Steube <jens.steube@gmail.com>
8 #include "include/constants.h"
9 #include "include/kernel_vendor.h"
28 #include "include/kernel_functions.c"
29 #include "types_amd.c"
30 #include "common_amd.c"
31 #include "include/rp_gpu.h"
35 #define VECT_COMPARE_S "check_single_vect1_comp4.c"
36 #define VECT_COMPARE_M "check_multi_vect1_comp4.c"
40 #define VECT_COMPARE_S "check_single_vect4_comp4.c"
41 #define VECT_COMPARE_M "check_multi_vect4_comp4.c"
44 __constant u32 padding[8] =
64 static void swap (__local RC4_KEY *rc4_key, const u8 i, const u8 j)
69 rc4_key->S[i] = rc4_key->S[j];
73 static void rc4_init_16 (__local RC4_KEY *rc4_key, const u32 data[4])
78 __local u32 *ptr = (__local u32 *) rc4_key->S;
81 for (u32 i = 0; i < 64; i++)
86 const u32 d0 = data[0] >> 0;
87 const u32 d1 = data[0] >> 8;
88 const u32 d2 = data[0] >> 16;
89 const u32 d3 = data[0] >> 24;
90 const u32 d4 = data[1] >> 0;
95 for (u32 i = 0; i < 255; i += 5)
97 j += rc4_key->S[i + 0] + d0; swap (rc4_key, i + 0, j);
98 j += rc4_key->S[i + 1] + d1; swap (rc4_key, i + 1, j);
99 j += rc4_key->S[i + 2] + d2; swap (rc4_key, i + 2, j);
100 j += rc4_key->S[i + 3] + d3; swap (rc4_key, i + 3, j);
101 j += rc4_key->S[i + 4] + d4; swap (rc4_key, i + 4, j);
104 j += rc4_key->S[255] + d0; swap (rc4_key, 255, j);
107 static u8 rc4_next_16 (__local RC4_KEY *rc4_key, u8 i, u8 j, __constant u32 in[4], u32 out[4])
109 for (u32 k = 0; k < 4; k++)
118 swap (rc4_key, i, j);
120 idx = rc4_key->S[i] + rc4_key->S[j];
122 xor4 |= rc4_key->S[idx] << 0;
127 swap (rc4_key, i, j);
129 idx = rc4_key->S[i] + rc4_key->S[j];
131 xor4 |= rc4_key->S[idx] << 8;
136 swap (rc4_key, i, j);
138 idx = rc4_key->S[i] + rc4_key->S[j];
140 xor4 |= rc4_key->S[idx] << 16;
145 swap (rc4_key, i, j);
147 idx = rc4_key->S[i] + rc4_key->S[j];
149 xor4 |= rc4_key->S[idx] << 24;
151 out[k] = in[k] ^ xor4;
157 __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 rules_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
163 const u32 lid = get_local_id (0);
169 const u32 gid = get_global_id (0);
171 if (gid >= gid_max) return;
175 pw_buf0[0] = pws[gid].i[ 0];
176 pw_buf0[1] = pws[gid].i[ 1];
177 pw_buf0[2] = pws[gid].i[ 2];
178 pw_buf0[3] = pws[gid].i[ 3];
182 pw_buf1[0] = pws[gid].i[ 4];
183 pw_buf1[1] = pws[gid].i[ 5];
184 pw_buf1[2] = pws[gid].i[ 6];
185 pw_buf1[3] = pws[gid].i[ 7];
187 const u32 pw_len = pws[gid].pw_len;
193 __local RC4_KEY rc4_keys[64];
194 __local RC4_KEY *rc4_key = &rc4_keys[lid];
200 for (u32 il_pos = 0; il_pos < rules_cnt; il_pos++)
230 const u32 out_len = apply_rules (rules_buf[il_pos].cmds, w0, w1, pw_len);
241 rc4_init_16 (rc4_key, key);
245 rc4_next_16 (rc4_key, 0, 0, padding, out);
247 const u32x r0 = out[0];
248 const u32x r1 = out[1];
249 const u32x r2 = out[2];
250 const u32x r3 = out[3];
252 #include VECT_COMPARE_M
256 __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 rules_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
260 __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 rules_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
264 __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 rules_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
270 const u32 lid = get_local_id (0);
276 const u32 gid = get_global_id (0);
278 if (gid >= gid_max) return;
282 pw_buf0[0] = pws[gid].i[ 0];
283 pw_buf0[1] = pws[gid].i[ 1];
284 pw_buf0[2] = pws[gid].i[ 2];
285 pw_buf0[3] = pws[gid].i[ 3];
289 pw_buf1[0] = pws[gid].i[ 4];
290 pw_buf1[1] = pws[gid].i[ 5];
291 pw_buf1[2] = pws[gid].i[ 6];
292 pw_buf1[3] = pws[gid].i[ 7];
294 const u32 pw_len = pws[gid].pw_len;
300 const u32 search[4] =
302 digests_buf[digests_offset].digest_buf[DGST_R0],
303 digests_buf[digests_offset].digest_buf[DGST_R1],
304 digests_buf[digests_offset].digest_buf[DGST_R2],
305 digests_buf[digests_offset].digest_buf[DGST_R3]
312 __local RC4_KEY rc4_keys[64];
313 __local RC4_KEY *rc4_key = &rc4_keys[lid];
319 for (u32 il_pos = 0; il_pos < rules_cnt; il_pos++)
349 const u32 out_len = apply_rules (rules_buf[il_pos].cmds, w0, w1, pw_len);
360 rc4_init_16 (rc4_key, key);
364 rc4_next_16 (rc4_key, 0, 0, padding, out);
366 const u32x r0 = out[0];
367 const u32x r1 = out[1];
368 const u32x r2 = out[2];
369 const u32x r3 = out[3];
371 #include VECT_COMPARE_S
375 __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 rules_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
379 __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 rules_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)