#define _MD5_
+#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"
-
-#define uint_to_hex_lower8(i) l_bin2asc[(i)]
+#include "OpenCL/simd.c"
+
+#if VECT_SIZE == 1
+#define uint_to_hex_lower8(i) (u32x) (l_bin2asc[(i)])
+#elif VECT_SIZE == 2
+#define uint_to_hex_lower8(i) (u32x) (l_bin2asc[(i).s0], l_bin2asc[(i).s1])
+#elif VECT_SIZE == 4
+#define uint_to_hex_lower8(i) (u32x) (l_bin2asc[(i).s0], l_bin2asc[(i).s1], l_bin2asc[(i).s2], l_bin2asc[(i).s3])
+#elif VECT_SIZE == 8
+#define uint_to_hex_lower8(i) (u32x) (l_bin2asc[(i).s0], l_bin2asc[(i).s1], l_bin2asc[(i).s2], l_bin2asc[(i).s3], l_bin2asc[(i).s4], l_bin2asc[(i).s5], l_bin2asc[(i).s6], l_bin2asc[(i).s7])
+#endif
static void m11100m (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, __local u32 l_bin2asc[256])
{
const u32 salt_len = salt_bufs[salt_pos].salt_len - 4;
- switch_buffer_by_offset (salt_buf0, salt_buf1, salt_buf2, salt_buf3, pw_len);
+ switch_buffer_by_offset_le_S (salt_buf0, salt_buf1, salt_buf2, salt_buf3, pw_len);
const u32 pw_salt_len = pw_len + salt_len;
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;
- u32 w0_t[4];
+ u32x w0_t[4];
- w0_t[0] = w0[0] | salt_buf0[0];
+ w0_t[0] = w0lr | salt_buf0[0];
w0_t[1] = w0[1] | salt_buf0[1];
w0_t[2] = w0[2] | salt_buf0[2];
w0_t[3] = w0[3] | salt_buf0[3];
- u32 w1_t[4];
+ u32x w1_t[4];
w1_t[0] = w1[0] | salt_buf1[0];
w1_t[1] = w1[1] | salt_buf1[1];
w1_t[2] = w1[2] | salt_buf1[2];
w1_t[3] = w1[3] | salt_buf1[3];
- u32 w2_t[4];
+ u32x w2_t[4];
w2_t[0] = w2[0] | salt_buf2[0];
w2_t[1] = w2[1] | salt_buf2[1];
w2_t[2] = w2[2] | salt_buf2[2];
w2_t[3] = w2[3] | salt_buf2[3];
- u32 w3_t[4];
+ u32x w3_t[4];
w3_t[0] = w3[0] | salt_buf3[0];
w3_t[1] = w3[1] | salt_buf3[1];
* md5 ($pass.$salt)
*/
- u32 a = MD5M_A;
- u32 b = MD5M_B;
- u32 c = MD5M_C;
- u32 d = MD5M_D;
+ u32x a = MD5M_A;
+ u32x b = MD5M_B;
+ u32x c = MD5M_C;
+ u32x d = MD5M_D;
MD5_STEP (MD5_Fo, a, b, c, d, w0_t[0], MD5C00, MD5S00);
MD5_STEP (MD5_Fo, d, a, b, c, w0_t[1], MD5C01, MD5S01);
MD5_STEP (MD5_I , c, d, a, b, w0_t[2], MD5C3e, MD5S32);
MD5_STEP (MD5_I , b, c, d, a, w2_t[1], MD5C3f, MD5S33);
- const u32 r0 = a;
- const u32 r1 = d;
- const u32 r2 = c;
- const u32 r3 = b;
-
- #include COMPARE_M
+ COMPARE_M_SIMD (a, d, c, b);
}
}
const u32 salt_len = salt_bufs[salt_pos].salt_len - 4;
- switch_buffer_by_offset (salt_buf0, salt_buf1, salt_buf2, salt_buf3, pw_len);
+ switch_buffer_by_offset_le_S (salt_buf0, salt_buf1, salt_buf2, salt_buf3, pw_len);
const u32 pw_salt_len = pw_len + salt_len;
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;
- u32 w0_t[4];
+ u32x w0_t[4];
- w0_t[0] = w0[0] | salt_buf0[0];
+ w0_t[0] = w0lr | salt_buf0[0];
w0_t[1] = w0[1] | salt_buf0[1];
w0_t[2] = w0[2] | salt_buf0[2];
w0_t[3] = w0[3] | salt_buf0[3];
- u32 w1_t[4];
+ u32x w1_t[4];
w1_t[0] = w1[0] | salt_buf1[0];
w1_t[1] = w1[1] | salt_buf1[1];
w1_t[2] = w1[2] | salt_buf1[2];
w1_t[3] = w1[3] | salt_buf1[3];
- u32 w2_t[4];
+ u32x w2_t[4];
w2_t[0] = w2[0] | salt_buf2[0];
w2_t[1] = w2[1] | salt_buf2[1];
w2_t[2] = w2[2] | salt_buf2[2];
w2_t[3] = w2[3] | salt_buf2[3];
- u32 w3_t[4];
+ u32x w3_t[4];
w3_t[0] = w3[0] | salt_buf3[0];
w3_t[1] = w3[1] | salt_buf3[1];
* md5 ($pass.$salt)
*/
- u32 a = MD5M_A;
- u32 b = MD5M_B;
- u32 c = MD5M_C;
- u32 d = MD5M_D;
+ u32x a = MD5M_A;
+ u32x b = MD5M_B;
+ u32x c = MD5M_C;
+ u32x d = MD5M_D;
MD5_STEP (MD5_Fo, a, b, c, d, w0_t[0], MD5C00, MD5S00);
MD5_STEP (MD5_Fo, d, a, b, c, w0_t[1], MD5C01, MD5S01);
MD5_STEP (MD5_I , c, d, a, b, w0_t[2], MD5C3e, MD5S32);
MD5_STEP (MD5_I , b, c, d, a, w2_t[1], MD5C3f, MD5S33);
- const u32 r0 = a;
- const u32 r1 = d;
- const u32 r2 = c;
- const u32 r3 = b;
-
- #include COMPARE_S
+ COMPARE_S_SIMD (a, d, c, b);
}
}