#define _SHA1_
+#define NEW_SIMD_CODE
+
#include "include/constants.h"
#include "include/kernel_vendor.h"
#include "include/kernel_functions.c"
#include "OpenCL/types_ocl.c"
#include "OpenCL/common.c"
-
-#define COMPARE_S "OpenCL/check_single_comp4.c"
-#define COMPARE_M "OpenCL/check_multi_comp4.c"
+#include "OpenCL/simd.c"
static void m08100m (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], const u32 pw_len, __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_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 bfs_cnt, const u32 digests_cnt, const u32 digests_offset)
{
u32 w0l = w0[0];
- for (u32 il_pos = 0; il_pos < bfs_cnt; il_pos++)
+ for (u32 il_pos = 0; il_pos < bfs_cnt; il_pos += VECT_SIZE)
{
- const u32 w0r = bfs_buf[il_pos].i;
+ const u32x w0r = w0r_create_bft (bfs_buf, il_pos);
- w0[0] = w0l | w0r;
+ const u32x w0lr = w0l | w0r;
/**
* prepend salt
*/
- u32 w0_t[4];
- u32 w1_t[4];
- u32 w2_t[4];
- u32 w3_t[4];
+ u32x w0_t[4];
+ u32x w1_t[4];
+ u32x w2_t[4];
+ u32x w3_t[4];
w0_t[0] = salt_buf0[0];
w0_t[1] = salt_buf0[1];
- w0_t[2] = w0[0];
+ w0_t[2] = w0lr;
w0_t[3] = w0[1];
w1_t[0] = w0[2];
w1_t[1] = w0[3];
* sha1
*/
- u32 a = SHA1M_A;
- u32 b = SHA1M_B;
- u32 c = SHA1M_C;
- u32 d = SHA1M_D;
- u32 e = SHA1M_E;
+ u32x a = SHA1M_A;
+ u32x b = SHA1M_B;
+ u32x c = SHA1M_C;
+ u32x d = SHA1M_D;
+ u32x e = SHA1M_E;
#undef K
#define K SHA1C00
w3_t[2] = rotl32 ((w2_t[3] ^ w1_t[2] ^ w0_t[0] ^ w3_t[2]), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w3_t[2]);
w3_t[3] = rotl32 ((w3_t[0] ^ w1_t[3] ^ w0_t[1] ^ w3_t[3]), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w3_t[3]);
- const u32 r0 = d;
- const u32 r1 = e;
- const u32 r2 = c;
- const u32 r3 = b;
-
- #include COMPARE_M
+ COMPARE_M_SIMD (d, e, c, b);
}
}
* reverse
*/
- const u32 e_rev = rotl32 (search[1], 2u);
+ const u32 e_rev = rotl32_S (search[1], 2u);
/**
* salt
u32 w0l = w0[0];
- for (u32 il_pos = 0; il_pos < bfs_cnt; il_pos++)
+ for (u32 il_pos = 0; il_pos < bfs_cnt; il_pos += VECT_SIZE)
{
- const u32 w0r = bfs_buf[il_pos].i;
+ const u32x w0r = w0r_create_bft (bfs_buf, il_pos);
- w0[0] = w0l | w0r;
+ const u32x w0lr = w0l | w0r;
/**
* prepend salt
*/
- u32 w0_t[4];
- u32 w1_t[4];
- u32 w2_t[4];
- u32 w3_t[4];
+ u32x w0_t[4];
+ u32x w1_t[4];
+ u32x w2_t[4];
+ u32x w3_t[4];
w0_t[0] = salt_buf0[0];
w0_t[1] = salt_buf0[1];
- w0_t[2] = w0[0];
+ w0_t[2] = w0lr;
w0_t[3] = w0[1];
w1_t[0] = w0[2];
w1_t[1] = w0[3];
* sha1
*/
- u32 a = SHA1M_A;
- u32 b = SHA1M_B;
- u32 c = SHA1M_C;
- u32 d = SHA1M_D;
- u32 e = SHA1M_E;
+ u32x a = SHA1M_A;
+ u32x b = SHA1M_B;
+ u32x c = SHA1M_C;
+ u32x d = SHA1M_D;
+ u32x e = SHA1M_E;
#undef K
#define K SHA1C00
w2_t[3] = rotl32 ((w2_t[0] ^ w0_t[3] ^ w3_t[1] ^ w2_t[3]), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w2_t[3]);
w3_t[0] = rotl32 ((w2_t[1] ^ w1_t[0] ^ w3_t[2] ^ w3_t[0]), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w3_t[0]);
- if (allx (e != e_rev)) continue;
+ if (MATCHES_NONE_VS (e, e_rev)) continue;
w3_t[1] = rotl32 ((w2_t[2] ^ w1_t[1] ^ w3_t[3] ^ w3_t[1]), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w3_t[1]);
w3_t[2] = rotl32 ((w2_t[3] ^ w1_t[2] ^ w0_t[0] ^ w3_t[2]), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w3_t[2]);
w3_t[3] = rotl32 ((w3_t[0] ^ w1_t[3] ^ w0_t[1] ^ w3_t[3]), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w3_t[3]);
- const u32 r0 = d;
- const u32 r1 = e;
- const u32 r2 = c;
- const u32 r3 = b;
-
- #include COMPARE_S
+ COMPARE_S_SIMD (d, e, c, b);
}
}
* base
*/
- w0[0] = swap32 (w0[0]);
- w0[1] = swap32 (w0[1]);
- w0[2] = swap32 (w0[2]);
- w0[3] = swap32 (w0[3]);
+ w0[0] = swap32_S (w0[0]);
+ w0[1] = swap32_S (w0[1]);
+ w0[2] = swap32_S (w0[2]);
+ w0[3] = swap32_S (w0[3]);
- append_0x80_2x4 (w0, w1, pw_len + 1);
+ append_0x80_2x4_S (w0, w1, pw_len + 1);
- w0[0] = swap32 (w0[0]);
- w0[1] = swap32 (w0[1]);
- w0[2] = swap32 (w0[2]);
- w0[3] = swap32 (w0[3]);
- w1[0] = swap32 (w1[0]);
+ w0[0] = swap32_S (w0[0]);
+ w0[1] = swap32_S (w0[1]);
+ w0[2] = swap32_S (w0[2]);
+ w0[3] = swap32_S (w0[3]);
+ w1[0] = swap32_S (w1[0]);
/**
* main
* base
*/
- w0[0] = swap32 (w0[0]);
- w0[1] = swap32 (w0[1]);
- w0[2] = swap32 (w0[2]);
- w0[3] = swap32 (w0[3]);
- w1[0] = swap32 (w1[0]);
- w1[1] = swap32 (w1[1]);
- w1[2] = swap32 (w1[2]);
- w1[3] = swap32 (w1[3]);
-
- append_0x80_3x4 (w0, w1, w2, pw_len + 1);
-
- w0[0] = swap32 (w0[0]);
- w0[1] = swap32 (w0[1]);
- w0[2] = swap32 (w0[2]);
- w0[3] = swap32 (w0[3]);
- w1[0] = swap32 (w1[0]);
- w1[1] = swap32 (w1[1]);
- w1[2] = swap32 (w1[2]);
- w1[3] = swap32 (w1[3]);
- w2[0] = swap32 (w2[0]);
+ w0[0] = swap32_S (w0[0]);
+ w0[1] = swap32_S (w0[1]);
+ w0[2] = swap32_S (w0[2]);
+ w0[3] = swap32_S (w0[3]);
+ w1[0] = swap32_S (w1[0]);
+ w1[1] = swap32_S (w1[1]);
+ w1[2] = swap32_S (w1[2]);
+ w1[3] = swap32_S (w1[3]);
+
+ append_0x80_3x4_S (w0, w1, w2, pw_len + 1);
+
+ w0[0] = swap32_S (w0[0]);
+ w0[1] = swap32_S (w0[1]);
+ w0[2] = swap32_S (w0[2]);
+ w0[3] = swap32_S (w0[3]);
+ w1[0] = swap32_S (w1[0]);
+ w1[1] = swap32_S (w1[1]);
+ w1[2] = swap32_S (w1[2]);
+ w1[3] = swap32_S (w1[3]);
+ w2[0] = swap32_S (w2[0]);
/**
* main
* base
*/
- w0[0] = swap32 (w0[0]);
- w0[1] = swap32 (w0[1]);
- w0[2] = swap32 (w0[2]);
- w0[3] = swap32 (w0[3]);
- w1[0] = swap32 (w1[0]);
- w1[1] = swap32 (w1[1]);
- w1[2] = swap32 (w1[2]);
- w1[3] = swap32 (w1[3]);
- w2[0] = swap32 (w2[0]);
- w2[1] = swap32 (w2[1]);
- w2[2] = swap32 (w2[2]);
- w2[3] = swap32 (w2[3]);
- w3[0] = swap32 (w3[0]);
- w3[1] = swap32 (w3[1]);
+ w0[0] = swap32_S (w0[0]);
+ w0[1] = swap32_S (w0[1]);
+ w0[2] = swap32_S (w0[2]);
+ w0[3] = swap32_S (w0[3]);
+ w1[0] = swap32_S (w1[0]);
+ w1[1] = swap32_S (w1[1]);
+ w1[2] = swap32_S (w1[2]);
+ w1[3] = swap32_S (w1[3]);
+ w2[0] = swap32_S (w2[0]);
+ w2[1] = swap32_S (w2[1]);
+ w2[2] = swap32_S (w2[2]);
+ w2[3] = swap32_S (w2[3]);
+ w3[0] = swap32_S (w3[0]);
+ w3[1] = swap32_S (w3[1]);
w3[2] = 0;
w3[3] = 0;
- append_0x80_4x4 (w0, w1, w2, w3, pw_len + 1);
-
- w0[0] = swap32 (w0[0]);
- w0[1] = swap32 (w0[1]);
- w0[2] = swap32 (w0[2]);
- w0[3] = swap32 (w0[3]);
- w1[0] = swap32 (w1[0]);
- w1[1] = swap32 (w1[1]);
- w1[2] = swap32 (w1[2]);
- w1[3] = swap32 (w1[3]);
- w2[0] = swap32 (w2[0]);
- w2[1] = swap32 (w2[1]);
- w2[2] = swap32 (w2[2]);
- w2[3] = swap32 (w2[3]);
- w3[0] = swap32 (w3[0]);
- w3[1] = swap32 (w3[1]);
+ append_0x80_4x4_S (w0, w1, w2, w3, pw_len + 1);
+
+ w0[0] = swap32_S (w0[0]);
+ w0[1] = swap32_S (w0[1]);
+ w0[2] = swap32_S (w0[2]);
+ w0[3] = swap32_S (w0[3]);
+ w1[0] = swap32_S (w1[0]);
+ w1[1] = swap32_S (w1[1]);
+ w1[2] = swap32_S (w1[2]);
+ w1[3] = swap32_S (w1[3]);
+ w2[0] = swap32_S (w2[0]);
+ w2[1] = swap32_S (w2[1]);
+ w2[2] = swap32_S (w2[2]);
+ w2[3] = swap32_S (w2[3]);
+ w3[0] = swap32_S (w3[0]);
+ w3[1] = swap32_S (w3[1]);
w3[2] = 0;
w3[3] = 0;
* base
*/
- w0[0] = swap32 (w0[0]);
- w0[1] = swap32 (w0[1]);
- w0[2] = swap32 (w0[2]);
- w0[3] = swap32 (w0[3]);
+ w0[0] = swap32_S (w0[0]);
+ w0[1] = swap32_S (w0[1]);
+ w0[2] = swap32_S (w0[2]);
+ w0[3] = swap32_S (w0[3]);
- append_0x80_2x4 (w0, w1, pw_len + 1);
+ append_0x80_2x4_S (w0, w1, pw_len + 1);
- w0[0] = swap32 (w0[0]);
- w0[1] = swap32 (w0[1]);
- w0[2] = swap32 (w0[2]);
- w0[3] = swap32 (w0[3]);
- w1[0] = swap32 (w1[0]);
+ w0[0] = swap32_S (w0[0]);
+ w0[1] = swap32_S (w0[1]);
+ w0[2] = swap32_S (w0[2]);
+ w0[3] = swap32_S (w0[3]);
+ w1[0] = swap32_S (w1[0]);
/**
* main
* base
*/
- w0[0] = swap32 (w0[0]);
- w0[1] = swap32 (w0[1]);
- w0[2] = swap32 (w0[2]);
- w0[3] = swap32 (w0[3]);
- w1[0] = swap32 (w1[0]);
- w1[1] = swap32 (w1[1]);
- w1[2] = swap32 (w1[2]);
- w1[3] = swap32 (w1[3]);
-
- append_0x80_3x4 (w0, w1, w2, pw_len + 1);
-
- w0[0] = swap32 (w0[0]);
- w0[1] = swap32 (w0[1]);
- w0[2] = swap32 (w0[2]);
- w0[3] = swap32 (w0[3]);
- w1[0] = swap32 (w1[0]);
- w1[1] = swap32 (w1[1]);
- w1[2] = swap32 (w1[2]);
- w1[3] = swap32 (w1[3]);
- w2[0] = swap32 (w2[0]);
+ w0[0] = swap32_S (w0[0]);
+ w0[1] = swap32_S (w0[1]);
+ w0[2] = swap32_S (w0[2]);
+ w0[3] = swap32_S (w0[3]);
+ w1[0] = swap32_S (w1[0]);
+ w1[1] = swap32_S (w1[1]);
+ w1[2] = swap32_S (w1[2]);
+ w1[3] = swap32_S (w1[3]);
+
+ append_0x80_3x4_S (w0, w1, w2, pw_len + 1);
+
+ w0[0] = swap32_S (w0[0]);
+ w0[1] = swap32_S (w0[1]);
+ w0[2] = swap32_S (w0[2]);
+ w0[3] = swap32_S (w0[3]);
+ w1[0] = swap32_S (w1[0]);
+ w1[1] = swap32_S (w1[1]);
+ w1[2] = swap32_S (w1[2]);
+ w1[3] = swap32_S (w1[3]);
+ w2[0] = swap32_S (w2[0]);
/**
* main
* base
*/
- w0[0] = swap32 (w0[0]);
- w0[1] = swap32 (w0[1]);
- w0[2] = swap32 (w0[2]);
- w0[3] = swap32 (w0[3]);
- w1[0] = swap32 (w1[0]);
- w1[1] = swap32 (w1[1]);
- w1[2] = swap32 (w1[2]);
- w1[3] = swap32 (w1[3]);
- w2[0] = swap32 (w2[0]);
- w2[1] = swap32 (w2[1]);
- w2[2] = swap32 (w2[2]);
- w2[3] = swap32 (w2[3]);
- w3[0] = swap32 (w3[0]);
- w3[1] = swap32 (w3[1]);
+ w0[0] = swap32_S (w0[0]);
+ w0[1] = swap32_S (w0[1]);
+ w0[2] = swap32_S (w0[2]);
+ w0[3] = swap32_S (w0[3]);
+ w1[0] = swap32_S (w1[0]);
+ w1[1] = swap32_S (w1[1]);
+ w1[2] = swap32_S (w1[2]);
+ w1[3] = swap32_S (w1[3]);
+ w2[0] = swap32_S (w2[0]);
+ w2[1] = swap32_S (w2[1]);
+ w2[2] = swap32_S (w2[2]);
+ w2[3] = swap32_S (w2[3]);
+ w3[0] = swap32_S (w3[0]);
+ w3[1] = swap32_S (w3[1]);
w3[2] = 0;
w3[3] = 0;
- append_0x80_4x4 (w0, w1, w2, w3, pw_len + 1);
-
- w0[0] = swap32 (w0[0]);
- w0[1] = swap32 (w0[1]);
- w0[2] = swap32 (w0[2]);
- w0[3] = swap32 (w0[3]);
- w1[0] = swap32 (w1[0]);
- w1[1] = swap32 (w1[1]);
- w1[2] = swap32 (w1[2]);
- w1[3] = swap32 (w1[3]);
- w2[0] = swap32 (w2[0]);
- w2[1] = swap32 (w2[1]);
- w2[2] = swap32 (w2[2]);
- w2[3] = swap32 (w2[3]);
- w3[0] = swap32 (w3[0]);
- w3[1] = swap32 (w3[1]);
+ append_0x80_4x4_S (w0, w1, w2, w3, pw_len + 1);
+
+ w0[0] = swap32_S (w0[0]);
+ w0[1] = swap32_S (w0[1]);
+ w0[2] = swap32_S (w0[2]);
+ w0[3] = swap32_S (w0[3]);
+ w1[0] = swap32_S (w1[0]);
+ w1[1] = swap32_S (w1[1]);
+ w1[2] = swap32_S (w1[2]);
+ w1[3] = swap32_S (w1[3]);
+ w2[0] = swap32_S (w2[0]);
+ w2[1] = swap32_S (w2[1]);
+ w2[2] = swap32_S (w2[2]);
+ w2[3] = swap32_S (w2[3]);
+ w3[0] = swap32_S (w3[0]);
+ w3[1] = swap32_S (w3[1]);
w3[2] = 0;
w3[3] = 0;
projects / hashcat.git / blobdiff