/**
- * Author......: Jens Steube <jens.steube@gmail.com>
+ * Authors.....: Jens Steube <jens.steube@gmail.com>
+ * Gabriele Gristina <matrix@hashcat.net>
+ * magnum <john.magnum@hushmail.com>
+ *
* License.....: MIT
*/
#define _DES_
-#include "include/constants.h"
-#include "include/kernel_vendor.h"
+#define NEW_SIMD_CODE
-#define DGST_R0 0
-#define DGST_R1 1
-#define DGST_R2 2
-#define DGST_R3 3
-
-#include "include/kernel_functions.c"
-#include "OpenCL/types_ocl.c"
-#include "OpenCL/common.c"
-#include "include/rp_kernel.h"
-#include "OpenCL/rp.c"
-
-#define COMPARE_S "OpenCL/check_single_comp4.c"
-#define COMPARE_M "OpenCL/check_multi_comp4.c"
+#include "inc_vendor.cl"
+#include "inc_hash_constants.h"
+#include "inc_hash_functions.cl"
+#include "inc_types.cl"
+#include "inc_common.cl"
+#include "inc_rp.h"
+#include "inc_rp.cl"
+#include "inc_simd.cl"
#define PERM_OP(a,b,tt,n,m) \
{ \
},
};
+#if VECT_SIZE == 1
#define BOX(i,n,S) (S)[(n)][(i)]
-
-static void _des_crypt_keysetup (u32 c, u32 d, u32 Kc[16], u32 Kd[16], __local u32 s_skb[8][64])
+#elif VECT_SIZE == 2
+#define BOX(i,n,S) (u32x) ((S)[(n)][(i).s0], (S)[(n)][(i).s1])
+#elif VECT_SIZE == 4
+#define BOX(i,n,S) (u32x) ((S)[(n)][(i).s0], (S)[(n)][(i).s1], (S)[(n)][(i).s2], (S)[(n)][(i).s3])
+#elif VECT_SIZE == 8
+#define BOX(i,n,S) (u32x) ((S)[(n)][(i).s0], (S)[(n)][(i).s1], (S)[(n)][(i).s2], (S)[(n)][(i).s3], (S)[(n)][(i).s4], (S)[(n)][(i).s5], (S)[(n)][(i).s6], (S)[(n)][(i).s7])
+#elif VECT_SIZE == 16
+#define BOX(i,n,S) (u32x) ((S)[(n)][(i).s0], (S)[(n)][(i).s1], (S)[(n)][(i).s2], (S)[(n)][(i).s3], (S)[(n)][(i).s4], (S)[(n)][(i).s5], (S)[(n)][(i).s6], (S)[(n)][(i).s7], (S)[(n)][(i).s8], (S)[(n)][(i).s9], (S)[(n)][(i).sa], (S)[(n)][(i).sb], (S)[(n)][(i).sc], (S)[(n)][(i).sd], (S)[(n)][(i).se], (S)[(n)][(i).sf])
+#endif
+
+void _des_crypt_keysetup (u32x c, u32x d, u32x Kc[16], u32x Kd[16], __local u32 (*s_skb)[64])
{
- u32 tt;
+ u32x tt;
PERM_OP (d, c, tt, 4, 0x0f0f0f0f);
HPERM_OP (c, tt, 2, 0xcccc0000);
c = c & 0x0fffffff;
+ #ifdef _unroll
#pragma unroll
+ #endif
for (u32 i = 0; i < 16; i++)
{
if ((i < 2) || (i == 8) || (i == 15))
c = c & 0x0fffffff;
d = d & 0x0fffffff;
- const u32 c00 = (c >> 0) & 0x0000003f;
- const u32 c06 = (c >> 6) & 0x00383003;
- const u32 c07 = (c >> 7) & 0x0000003c;
- const u32 c13 = (c >> 13) & 0x0000060f;
- const u32 c20 = (c >> 20) & 0x00000001;
-
- u32 s = BOX (((c00 >> 0) & 0xff), 0, s_skb)
- | BOX (((c06 >> 0) & 0xff)
- |((c07 >> 0) & 0xff), 1, s_skb)
- | BOX (((c13 >> 0) & 0xff)
- |((c06 >> 8) & 0xff), 2, s_skb)
- | BOX (((c20 >> 0) & 0xff)
- |((c13 >> 8) & 0xff)
- |((c06 >> 16) & 0xff), 3, s_skb);
-
- const u32 d00 = (d >> 0) & 0x00003c3f;
- const u32 d07 = (d >> 7) & 0x00003f03;
- const u32 d21 = (d >> 21) & 0x0000000f;
- const u32 d22 = (d >> 22) & 0x00000030;
-
- u32 t = BOX (((d00 >> 0) & 0xff), 4, s_skb)
- | BOX (((d07 >> 0) & 0xff)
- |((d00 >> 8) & 0xff), 5, s_skb)
- | BOX (((d07 >> 8) & 0xff), 6, s_skb)
- | BOX (((d21 >> 0) & 0xff)
- |((d22 >> 0) & 0xff), 7, s_skb);
+ const u32x c00 = (c >> 0) & 0x0000003f;
+ const u32x c06 = (c >> 6) & 0x00383003;
+ const u32x c07 = (c >> 7) & 0x0000003c;
+ const u32x c13 = (c >> 13) & 0x0000060f;
+ const u32x c20 = (c >> 20) & 0x00000001;
+
+ u32x s = BOX (((c00 >> 0) & 0xff), 0, s_skb)
+ | BOX (((c06 >> 0) & 0xff)
+ |((c07 >> 0) & 0xff), 1, s_skb)
+ | BOX (((c13 >> 0) & 0xff)
+ |((c06 >> 8) & 0xff), 2, s_skb)
+ | BOX (((c20 >> 0) & 0xff)
+ |((c13 >> 8) & 0xff)
+ |((c06 >> 16) & 0xff), 3, s_skb);
+
+ const u32x d00 = (d >> 0) & 0x00003c3f;
+ const u32x d07 = (d >> 7) & 0x00003f03;
+ const u32x d21 = (d >> 21) & 0x0000000f;
+ const u32x d22 = (d >> 22) & 0x00000030;
+
+ u32x t = BOX (((d00 >> 0) & 0xff), 4, s_skb)
+ | BOX (((d07 >> 0) & 0xff)
+ |((d00 >> 8) & 0xff), 5, s_skb)
+ | BOX (((d07 >> 8) & 0xff), 6, s_skb)
+ | BOX (((d21 >> 0) & 0xff)
+ |((d22 >> 0) & 0xff), 7, s_skb);
Kc[i] = ((t << 16) | (s & 0x0000ffff));
Kd[i] = ((s >> 16) | (t & 0xffff0000));
}
}
-static void _des_crypt_encrypt (u32 iv[2], u32 mask, u32 Kc[16], u32 Kd[16], __local u32 s_SPtrans[8][64])
+void _des_crypt_encrypt (u32x iv[2], u32 mask, u32x Kc[16], u32x Kd[16], __local u32 (*s_SPtrans)[64])
{
const u32 E1 = (mask >> 2) & 0x3f0;
const u32 E0 = mask & 0x3f;
- u32 r = 0;
- u32 l = 0;
+ u32x r = 0;
+ u32x l = 0;
for (u32 i = 0; i < 25; i++)
{
+ #ifdef _unroll
+ #pragma unroll
+ #endif
for (u32 j = 0; j < 16; j += 2)
{
- u32 t;
- u32 u;
+ u32x t;
+ u32x u;
t = r ^ (r >> 16);
u = t & E0;
| BOX (((t >> 24) & 0x3f), 7, s_SPtrans);
}
- u32 tt;
+ u32x tt;
tt = l;
l = r;
iv[1] = rotl32 (l, 31);
}
-__kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m01500_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 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 rules_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void m01500_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 void *esalt_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)
{
/**
* modifier
*/
+ const u32 gid = get_global_id (0);
const u32 lid = get_local_id (0);
+ const u32 lsz = get_local_size (0);
/**
- * base
+ * sbox, kbox
*/
- const u32 gid = get_global_id (0);
+ __local u32 s_SPtrans[8][64];
+ __local u32 s_skb[8][64];
- u32 pw_buf0[4];
+ for (u32 i = lid; i < 64; i += lsz)
+ {
+ s_SPtrans[0][i] = c_SPtrans[0][i];
+ s_SPtrans[1][i] = c_SPtrans[1][i];
+ s_SPtrans[2][i] = c_SPtrans[2][i];
+ s_SPtrans[3][i] = c_SPtrans[3][i];
+ s_SPtrans[4][i] = c_SPtrans[4][i];
+ s_SPtrans[5][i] = c_SPtrans[5][i];
+ s_SPtrans[6][i] = c_SPtrans[6][i];
+ s_SPtrans[7][i] = c_SPtrans[7][i];
+
+ s_skb[0][i] = c_skb[0][i];
+ s_skb[1][i] = c_skb[1][i];
+ s_skb[2][i] = c_skb[2][i];
+ s_skb[3][i] = c_skb[3][i];
+ s_skb[4][i] = c_skb[4][i];
+ s_skb[5][i] = c_skb[5][i];
+ s_skb[6][i] = c_skb[6][i];
+ s_skb[7][i] = c_skb[7][i];
+ }
- pw_buf0[0] = pws[gid].i[ 0];
- pw_buf0[1] = pws[gid].i[ 1];
- pw_buf0[2] = 0;
- pw_buf0[3] = 0;
+ barrier (CLK_LOCAL_MEM_FENCE);
- const u32 pw_len = pws[gid].pw_len;
+ if (gid >= gid_max) return;
/**
- * sbox, kbox
+ * base
*/
- __local u32 s_SPtrans[8][64];
- __local u32 s_skb[8][64];
-
- s_SPtrans[0][lid] = c_SPtrans[0][lid];
- s_SPtrans[1][lid] = c_SPtrans[1][lid];
- s_SPtrans[2][lid] = c_SPtrans[2][lid];
- s_SPtrans[3][lid] = c_SPtrans[3][lid];
- s_SPtrans[4][lid] = c_SPtrans[4][lid];
- s_SPtrans[5][lid] = c_SPtrans[5][lid];
- s_SPtrans[6][lid] = c_SPtrans[6][lid];
- s_SPtrans[7][lid] = c_SPtrans[7][lid];
-
- s_skb[0][lid] = c_skb[0][lid];
- s_skb[1][lid] = c_skb[1][lid];
- s_skb[2][lid] = c_skb[2][lid];
- s_skb[3][lid] = c_skb[3][lid];
- s_skb[4][lid] = c_skb[4][lid];
- s_skb[5][lid] = c_skb[5][lid];
- s_skb[6][lid] = c_skb[6][lid];
- s_skb[7][lid] = c_skb[7][lid];
+ u32 pw_buf0[4];
+ u32 pw_buf1[4];
- barrier (CLK_LOCAL_MEM_FENCE);
+ pw_buf0[0] = pws[gid].i[0];
+ pw_buf0[1] = pws[gid].i[1];
+ pw_buf0[2] = pws[gid].i[2];
+ pw_buf0[3] = pws[gid].i[3];
+ pw_buf1[0] = pws[gid].i[4];
+ pw_buf1[1] = pws[gid].i[5];
+ pw_buf1[2] = pws[gid].i[6];
+ pw_buf1[3] = pws[gid].i[7];
- if (gid >= gid_max) return;
+ const u32 pw_len = pws[gid].pw_len;
/**
* salt
* main
*/
- for (u32 il_pos = 0; il_pos < rules_cnt; il_pos++)
+ for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
{
- u32 w0[4];
-
- w0[0] = pw_buf0[0];
- w0[1] = pw_buf0[1];
- w0[2] = 0;
- w0[3] = 0;
-
- u32 w1[4];
-
- w1[0] = 0;
- w1[1] = 0;
- w1[2] = 0;
- w1[3] = 0;
-
- u32 w2[4];
-
- w2[0] = 0;
- w2[1] = 0;
- w2[2] = 0;
- w2[3] = 0;
-
- u32 w3[4];
-
- w3[0] = 0;
- w3[1] = 0;
- w3[2] = 0;
- w3[3] = 0;
+ u32x w0[4] = { 0 };
+ u32x w1[4] = { 0 };
+ u32x w2[4] = { 0 };
+ u32x w3[4] = { 0 };
- u32 out_len = apply_rules (rules_buf[il_pos].cmds, w0, w1, pw_len);
+ const u32x out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
- out_len = (out_len >= 8) ? 8 : out_len;
-
- u32 data[2];
+ u32x data[2];
data[0] = (w0[0] << 1) & 0xfefefefe;
data[1] = (w0[1] << 1) & 0xfefefefe;
- u32 Kc[16];
- u32 Kd[16];
+ u32x Kc[16];
+ u32x Kd[16];
_des_crypt_keysetup (data[0], data[1], Kc, Kd, s_skb);
- u32 iv[2];
+ u32x iv[2];
_des_crypt_encrypt (iv, mask, Kc, Kd, s_SPtrans);
- const u32 r0 = iv[0];
- const u32 r1 = iv[1];
- const u32 r2 = 0;
- const u32 r3 = 0;
+ u32x z = 0;
- #include COMPARE_M
+ COMPARE_M_SIMD (iv[0], iv[1], z, z);
}
}
-__kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m01500_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 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, const u32 combs_mode, const u32 gid_max)
+__kernel void m01500_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 void *esalt_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)
{
}
-__kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m01500_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 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, const u32 combs_mode, const u32 gid_max)
+__kernel void m01500_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 void *esalt_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)
{
}
-__kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m01500_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 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 rules_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void m01500_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 void *esalt_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)
{
/**
* modifier
*/
+ const u32 gid = get_global_id (0);
const u32 lid = get_local_id (0);
+ const u32 lsz = get_local_size (0);
/**
- * base
+ * sbox, kbox
*/
- const u32 gid = get_global_id (0);
+ __local u32 s_SPtrans[8][64];
+ __local u32 s_skb[8][64];
- u32 pw_buf0[4];
+ for (u32 i = lid; i < 64; i += lsz)
+ {
+ s_SPtrans[0][i] = c_SPtrans[0][i];
+ s_SPtrans[1][i] = c_SPtrans[1][i];
+ s_SPtrans[2][i] = c_SPtrans[2][i];
+ s_SPtrans[3][i] = c_SPtrans[3][i];
+ s_SPtrans[4][i] = c_SPtrans[4][i];
+ s_SPtrans[5][i] = c_SPtrans[5][i];
+ s_SPtrans[6][i] = c_SPtrans[6][i];
+ s_SPtrans[7][i] = c_SPtrans[7][i];
+
+ s_skb[0][i] = c_skb[0][i];
+ s_skb[1][i] = c_skb[1][i];
+ s_skb[2][i] = c_skb[2][i];
+ s_skb[3][i] = c_skb[3][i];
+ s_skb[4][i] = c_skb[4][i];
+ s_skb[5][i] = c_skb[5][i];
+ s_skb[6][i] = c_skb[6][i];
+ s_skb[7][i] = c_skb[7][i];
+ }
- pw_buf0[0] = pws[gid].i[ 0];
- pw_buf0[1] = pws[gid].i[ 1];
- pw_buf0[2] = 0;
- pw_buf0[3] = 0;
+ barrier (CLK_LOCAL_MEM_FENCE);
- const u32 pw_len = pws[gid].pw_len;
+ if (gid >= gid_max) return;
/**
- * sbox, kbox
+ * base
*/
- __local u32 s_SPtrans[8][64];
- __local u32 s_skb[8][64];
-
- s_SPtrans[0][lid] = c_SPtrans[0][lid];
- s_SPtrans[1][lid] = c_SPtrans[1][lid];
- s_SPtrans[2][lid] = c_SPtrans[2][lid];
- s_SPtrans[3][lid] = c_SPtrans[3][lid];
- s_SPtrans[4][lid] = c_SPtrans[4][lid];
- s_SPtrans[5][lid] = c_SPtrans[5][lid];
- s_SPtrans[6][lid] = c_SPtrans[6][lid];
- s_SPtrans[7][lid] = c_SPtrans[7][lid];
-
- s_skb[0][lid] = c_skb[0][lid];
- s_skb[1][lid] = c_skb[1][lid];
- s_skb[2][lid] = c_skb[2][lid];
- s_skb[3][lid] = c_skb[3][lid];
- s_skb[4][lid] = c_skb[4][lid];
- s_skb[5][lid] = c_skb[5][lid];
- s_skb[6][lid] = c_skb[6][lid];
- s_skb[7][lid] = c_skb[7][lid];
+ u32 pw_buf0[4];
+ u32 pw_buf1[4];
- barrier (CLK_LOCAL_MEM_FENCE);
+ pw_buf0[0] = pws[gid].i[0];
+ pw_buf0[1] = pws[gid].i[1];
+ pw_buf0[2] = pws[gid].i[2];
+ pw_buf0[3] = pws[gid].i[3];
+ pw_buf1[0] = pws[gid].i[4];
+ pw_buf1[1] = pws[gid].i[5];
+ pw_buf1[2] = pws[gid].i[6];
+ pw_buf1[3] = pws[gid].i[7];
- if (gid >= gid_max) return;
+ const u32 pw_len = pws[gid].pw_len;
/**
* salt
{
digests_buf[digests_offset].digest_buf[DGST_R0],
digests_buf[digests_offset].digest_buf[DGST_R1],
- digests_buf[digests_offset].digest_buf[DGST_R2],
- digests_buf[digests_offset].digest_buf[DGST_R3]
+ 0,
+ 0
};
/**
* main
*/
- for (u32 il_pos = 0; il_pos < rules_cnt; il_pos++)
+ for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
{
- u32 w0[4];
-
- w0[0] = pw_buf0[0];
- w0[1] = pw_buf0[1];
- w0[2] = 0;
- w0[3] = 0;
-
- u32 w1[4];
-
- w1[0] = 0;
- w1[1] = 0;
- w1[2] = 0;
- w1[3] = 0;
-
- u32 w2[4];
-
- w2[0] = 0;
- w2[1] = 0;
- w2[2] = 0;
- w2[3] = 0;
-
- u32 w3[4];
-
- w3[0] = 0;
- w3[1] = 0;
- w3[2] = 0;
- w3[3] = 0;
-
- u32 out_len = apply_rules (rules_buf[il_pos].cmds, w0, w1, pw_len);
+ u32x w0[4] = { 0 };
+ u32x w1[4] = { 0 };
+ u32x w2[4] = { 0 };
+ u32x w3[4] = { 0 };
- out_len = (out_len >= 8) ? 8 : out_len;
+ const u32x out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
- u32 data[2];
+ u32x data[2];
data[0] = (w0[0] << 1) & 0xfefefefe;
data[1] = (w0[1] << 1) & 0xfefefefe;
- u32 Kc[16];
- u32 Kd[16];
+ u32x Kc[16];
+ u32x Kd[16];
_des_crypt_keysetup (data[0], data[1], Kc, Kd, s_skb);
- u32 iv[2];
+ u32x iv[2];
_des_crypt_encrypt (iv, mask, Kc, Kd, s_SPtrans);
- const u32 r0 = iv[0];
- const u32 r1 = iv[1];
- const u32 r2 = 0;
- const u32 r3 = 0;
+ u32x z = 0;
- #include COMPARE_S
+ COMPARE_S_SIMD (iv[0], iv[1], z, z);
}
}
-__kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m01500_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 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, const u32 combs_mode, const u32 gid_max)
+__kernel void m01500_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 void *esalt_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)
{
}
-__kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m01500_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 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, const u32 combs_mode, const u32 gid_max)
+__kernel void m01500_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 void *esalt_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)
{
}
projects / hashcat.git / blobdiff