arno/hashcat
1
0
Fork 0
mirror of https://github.com/hashcat/hashcat.git synced 2024-11-13 19:28:56 +00:00
Code Issues Releases Wiki Activity
12fa3d6bfc
hashcat/OpenCL/m00120_a3.cl

869 lines
36 KiB
Common Lisp
Raw Normal View History

Initial commit
2015-12-04 14:47:52 +00:00
/**
* Author......: Jens Steube <jens.steube@gmail.com>
* License.....: MIT
*/
#define _SHA1_
Converted _a3 kernels, use SIMD for CPU and GPU
2016-01-23 14:32:31 +00:00
#define NEW_SIMD_CODE
Initial commit
2015-12-04 14:47:52 +00:00
#include "include/constants.h"
#include "include/kernel_vendor.h"
#define DGST_R0 3
#define DGST_R1 4
#define DGST_R2 2
#define DGST_R3 1
#include "include/kernel_functions.c"
Fix path to includes
2016-01-03 00:48:05 +00:00
#include "OpenCL/types_ocl.c"
#include "OpenCL/common.c"
Converted _a3 kernels, use SIMD for CPU and GPU
2016-01-23 14:32:31 +00:00
#include "OpenCL/simd.c"
Initial commit
2015-12-04 14:47:52 +00:00
Replace the substring GPU to a more appropriate "device" or "kernel" substring depending on the context
2016-01-05 07:26:44 +00:00
static void m00120m (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)
Initial commit
2015-12-04 14:47:52 +00:00
{
/**
* modifier
*/
const u32 gid = get_global_id (0);
const u32 lid = get_local_id (0);
/**
* salt
*/
u32 salt_buf0[4];
salt_buf0[0] = salt_bufs[salt_pos].salt_buf[ 0];
salt_buf0[1] = salt_bufs[salt_pos].salt_buf[ 1];
salt_buf0[2] = salt_bufs[salt_pos].salt_buf[ 2];
salt_buf0[3] = salt_bufs[salt_pos].salt_buf[ 3];
u32 salt_buf1[4];
salt_buf1[0] = salt_bufs[salt_pos].salt_buf[ 4];
salt_buf1[1] = salt_bufs[salt_pos].salt_buf[ 5];
salt_buf1[2] = salt_bufs[salt_pos].salt_buf[ 6];
salt_buf1[3] = salt_bufs[salt_pos].salt_buf[ 7];
u32 salt_buf2[4];
salt_buf2[0] = 0;
salt_buf2[1] = 0;
salt_buf2[2] = 0;
salt_buf2[3] = 0;
u32 salt_buf3[4];
salt_buf3[0] = 0;
salt_buf3[1] = 0;
salt_buf3[2] = 0;
salt_buf3[3] = 0;
const u32 salt_len = salt_bufs[salt_pos].salt_len;
const u32 pw_salt_len = pw_len + salt_len;
/**
* prepend salt
*/
- Dropped all vector code since new GPU's are all scalar, makes the code much easier - Some performance on low-end GPU may drop because of that, but only for a few hash-modes - Dropped scalar code (aka warp) since we do not have any vector datatypes anymore - Renamed C++ overloading functions memcat32_9 -> memcat_c32_w4x4_a3x4 - Still need to fix kernels to new function names, needs to be done manually - Temperature Management needs to be rewritten partially because of conflicting datatypes names - Added code to create different codepaths for NV on AMD in runtime in host (see data.vendor_id) - Added code to create different codepaths for NV on AMD in runtime in kernels (see IS_NV and IS_AMD) - First tests working for -m 0, for example - Great performance increases in general for NV so far - Tested amp_* and markov_* kernel - Migrated special NV optimizations for rule processor
2015-12-15 11:04:22 +00:00
u32 w0_t[4];
u32 w1_t[4];
u32 w2_t[4];
u32 w3_t[4];
Initial commit
2015-12-04 14:47:52 +00:00
Converted _a3 kernels, use SIMD for CPU and GPU
2016-01-23 14:32:31 +00:00
w0_t[0] = swap32_S (w0[0]);
w0_t[1] = swap32_S (w0[1]);
w0_t[2] = swap32_S (w0[2]);
w0_t[3] = swap32_S (w0[3]);
w1_t[0] = swap32_S (w1[0]);
w1_t[1] = swap32_S (w1[1]);
w1_t[2] = swap32_S (w1[2]);
w1_t[3] = swap32_S (w1[3]);
w2_t[0] = swap32_S (w2[0]);
w2_t[1] = swap32_S (w2[1]);
w2_t[2] = swap32_S (w2[2]);
w2_t[3] = swap32_S (w2[3]);
w3_t[0] = swap32_S (w3[0]);
w3_t[1] = swap32_S (w3[1]);
w3_t[2] = swap32_S (w3[2]);
w3_t[3] = swap32_S (w3[3]);
switch_buffer_by_offset_le_S (w0_t, w1_t, w2_t, w3_t, salt_len);
Initial commit
2015-12-04 14:47:52 +00:00
w0_t[0] |= salt_buf0[0];
w0_t[1] |= salt_buf0[1];
w0_t[2] |= salt_buf0[2];
w0_t[3] |= salt_buf0[3];
w1_t[0] |= salt_buf1[0];
w1_t[1] |= salt_buf1[1];
w1_t[2] |= salt_buf1[2];
w1_t[3] |= salt_buf1[3];
w2_t[0] |= salt_buf2[0];
w2_t[1] |= salt_buf2[1];
w2_t[2] |= salt_buf2[2];
w2_t[3] |= salt_buf2[3];
w3_t[0] |= salt_buf3[0];
w3_t[1] |= salt_buf3[1];
w3_t[2] |= salt_buf3[2];
w3_t[3] |= salt_buf3[3];
Converted _a3 kernels, use SIMD for CPU and GPU
2016-01-23 14:32:31 +00:00
w0_t[0] = swap32_S (w0_t[0]);
w0_t[1] = swap32_S (w0_t[1]);
w0_t[2] = swap32_S (w0_t[2]);
w0_t[3] = swap32_S (w0_t[3]);
w1_t[0] = swap32_S (w1_t[0]);
w1_t[1] = swap32_S (w1_t[1]);
w1_t[2] = swap32_S (w1_t[2]);
w1_t[3] = swap32_S (w1_t[3]);
w2_t[0] = swap32_S (w2_t[0]);
w2_t[1] = swap32_S (w2_t[1]);
w2_t[2] = swap32_S (w2_t[2]);
w2_t[3] = swap32_S (w2_t[3]);
w3_t[0] = swap32_S (w3_t[0]);
w3_t[1] = swap32_S (w3_t[1]);
w3_t[2] = swap32_S (w3_t[2]);
w3_t[3] = swap32_S (w3_t[3]);
Initial commit
2015-12-04 14:47:52 +00:00
/**
* loop
*/
- Dropped all vector code since new GPU's are all scalar, makes the code much easier - Some performance on low-end GPU may drop because of that, but only for a few hash-modes - Dropped scalar code (aka warp) since we do not have any vector datatypes anymore - Renamed C++ overloading functions memcat32_9 -> memcat_c32_w4x4_a3x4 - Still need to fix kernels to new function names, needs to be done manually - Temperature Management needs to be rewritten partially because of conflicting datatypes names - Added code to create different codepaths for NV on AMD in runtime in host (see data.vendor_id) - Added code to create different codepaths for NV on AMD in runtime in kernels (see IS_NV and IS_AMD) - First tests working for -m 0, for example - Great performance increases in general for NV so far - Tested amp_* and markov_* kernel - Migrated special NV optimizations for rule processor
2015-12-15 11:04:22 +00:00
u32 w0l = w0[0];
Initial commit
2015-12-04 14:47:52 +00:00
Converted _a3 kernels, use SIMD for CPU and GPU
2016-01-23 14:32:31 +00:00
for (u32 il_pos = 0; il_pos < bfs_cnt; il_pos += VECT_SIZE)
Initial commit
2015-12-04 14:47:52 +00:00
{
Prepare NEW_SIMD_MODE for -a 1 kernels
2016-02-20 15:13:06 +00:00
const u32x w0r = ix_create_bft (bfs_buf, il_pos);
Initial commit
2015-12-04 14:47:52 +00:00
Converted _a3 kernels, use SIMD for CPU and GPU
2016-01-23 14:32:31 +00:00
const u32x w0lr = w0l | w0r;
Initial commit
2015-12-04 14:47:52 +00:00
Converted _a3 kernels, use SIMD for CPU and GPU
2016-01-23 14:32:31 +00:00
u32x wx[16];
Initial commit
2015-12-04 14:47:52 +00:00
wx[ 0] = w0_t[0];
wx[ 1] = w0_t[1];
wx[ 2] = w0_t[2];
wx[ 3] = w0_t[3];
wx[ 4] = w1_t[0];
wx[ 5] = w1_t[1];
wx[ 6] = w1_t[2];
wx[ 7] = w1_t[3];
wx[ 8] = w2_t[0];
wx[ 9] = w2_t[1];
wx[10] = w2_t[2];
wx[11] = w2_t[3];
wx[12] = w3_t[0];
wx[13] = w3_t[1];
wx[14] = w3_t[2];
wx[15] = w3_t[3];
Converted _a3 kernels, use SIMD for CPU and GPU
2016-01-23 14:32:31 +00:00
overwrite_at_be (wx, w0lr, salt_len);
Initial commit
2015-12-04 14:47:52 +00:00
Converted _a3 kernels, use SIMD for CPU and GPU
2016-01-23 14:32:31 +00:00
u32x w0_t[4];
u32x w1_t[4];
u32x w2_t[4];
u32x w3_t[4];
Initial commit
2015-12-04 14:47:52 +00:00
w0_t[0] = wx[ 0];
w0_t[1] = wx[ 1];
w0_t[2] = wx[ 2];
w0_t[3] = wx[ 3];
w1_t[0] = wx[ 4];
w1_t[1] = wx[ 5];
w1_t[2] = wx[ 6];
w1_t[3] = wx[ 7];
w2_t[0] = wx[ 8];
w2_t[1] = wx[ 9];
w2_t[2] = wx[10];
w2_t[3] = wx[11];
w3_t[0] = wx[12];
w3_t[1] = wx[13];
w3_t[2] = 0;
w3_t[3] = pw_salt_len * 8;
/**
* sha1
*/
Converted _a3 kernels, use SIMD for CPU and GPU
2016-01-23 14:32:31 +00:00
u32x a = SHA1M_A;
u32x b = SHA1M_B;
u32x c = SHA1M_C;
u32x d = SHA1M_D;
u32x e = SHA1M_E;
Initial commit
2015-12-04 14:47:52 +00:00
#undef K
#define K SHA1C00
SHA1_STEP (SHA1_F0o, a, b, c, d, e, w0_t[0]);
SHA1_STEP (SHA1_F0o, e, a, b, c, d, w0_t[1]);
SHA1_STEP (SHA1_F0o, d, e, a, b, c, w0_t[2]);
SHA1_STEP (SHA1_F0o, c, d, e, a, b, w0_t[3]);
SHA1_STEP (SHA1_F0o, b, c, d, e, a, w1_t[0]);
SHA1_STEP (SHA1_F0o, a, b, c, d, e, w1_t[1]);
SHA1_STEP (SHA1_F0o, e, a, b, c, d, w1_t[2]);
SHA1_STEP (SHA1_F0o, d, e, a, b, c, w1_t[3]);
SHA1_STEP (SHA1_F0o, c, d, e, a, b, w2_t[0]);
SHA1_STEP (SHA1_F0o, b, c, d, e, a, w2_t[1]);
SHA1_STEP (SHA1_F0o, a, b, c, d, e, w2_t[2]);
SHA1_STEP (SHA1_F0o, e, a, b, c, d, w2_t[3]);
SHA1_STEP (SHA1_F0o, d, e, a, b, c, w3_t[0]);
SHA1_STEP (SHA1_F0o, c, d, e, a, b, w3_t[1]);
SHA1_STEP (SHA1_F0o, b, c, d, e, a, w3_t[2]);
SHA1_STEP (SHA1_F0o, a, b, c, d, e, w3_t[3]);
w0_t[0] = rotl32 ((w3_t[1] ^ w2_t[0] ^ w0_t[2] ^ w0_t[0]), 1u); SHA1_STEP (SHA1_F0o, e, a, b, c, d, w0_t[0]);
w0_t[1] = rotl32 ((w3_t[2] ^ w2_t[1] ^ w0_t[3] ^ w0_t[1]), 1u); SHA1_STEP (SHA1_F0o, d, e, a, b, c, w0_t[1]);
w0_t[2] = rotl32 ((w3_t[3] ^ w2_t[2] ^ w1_t[0] ^ w0_t[2]), 1u); SHA1_STEP (SHA1_F0o, c, d, e, a, b, w0_t[2]);
w0_t[3] = rotl32 ((w0_t[0] ^ w2_t[3] ^ w1_t[1] ^ w0_t[3]), 1u); SHA1_STEP (SHA1_F0o, b, c, d, e, a, w0_t[3]);
#undef K
#define K SHA1C01
w1_t[0] = rotl32 ((w0_t[1] ^ w3_t[0] ^ w1_t[2] ^ w1_t[0]), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w1_t[0]);
w1_t[1] = rotl32 ((w0_t[2] ^ w3_t[1] ^ w1_t[3] ^ w1_t[1]), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w1_t[1]);
w1_t[2] = rotl32 ((w0_t[3] ^ w3_t[2] ^ w2_t[0] ^ w1_t[2]), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w1_t[2]);
w1_t[3] = rotl32 ((w1_t[0] ^ w3_t[3] ^ w2_t[1] ^ w1_t[3]), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w1_t[3]);
w2_t[0] = rotl32 ((w1_t[1] ^ w0_t[0] ^ w2_t[2] ^ w2_t[0]), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w2_t[0]);
w2_t[1] = rotl32 ((w1_t[2] ^ w0_t[1] ^ w2_t[3] ^ w2_t[1]), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w2_t[1]);
w2_t[2] = rotl32 ((w1_t[3] ^ w0_t[2] ^ w3_t[0] ^ w2_t[2]), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w2_t[2]);
w2_t[3] = rotl32 ((w2_t[0] ^ w0_t[3] ^ w3_t[1] ^ w2_t[3]), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w2_t[3]);
w3_t[0] = rotl32 ((w2_t[1] ^ w1_t[0] ^ w3_t[2] ^ w3_t[0]), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w3_t[0]);
w3_t[1] = rotl32 ((w2_t[2] ^ w1_t[1] ^ w3_t[3] ^ w3_t[1]), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w3_t[1]);
w3_t[2] = rotl32 ((w2_t[3] ^ w1_t[2] ^ w0_t[0] ^ w3_t[2]), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w3_t[2]);
w3_t[3] = rotl32 ((w3_t[0] ^ w1_t[3] ^ w0_t[1] ^ w3_t[3]), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w3_t[3]);
w0_t[0] = rotl32 ((w3_t[1] ^ w2_t[0] ^ w0_t[2] ^ w0_t[0]), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w0_t[0]);
w0_t[1] = rotl32 ((w3_t[2] ^ w2_t[1] ^ w0_t[3] ^ w0_t[1]), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w0_t[1]);
w0_t[2] = rotl32 ((w3_t[3] ^ w2_t[2] ^ w1_t[0] ^ w0_t[2]), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w0_t[2]);
w0_t[3] = rotl32 ((w0_t[0] ^ w2_t[3] ^ w1_t[1] ^ w0_t[3]), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w0_t[3]);
w1_t[0] = rotl32 ((w0_t[1] ^ w3_t[0] ^ w1_t[2] ^ w1_t[0]), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w1_t[0]);
w1_t[1] = rotl32 ((w0_t[2] ^ w3_t[1] ^ w1_t[3] ^ w1_t[1]), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w1_t[1]);
w1_t[2] = rotl32 ((w0_t[3] ^ w3_t[2] ^ w2_t[0] ^ w1_t[2]), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w1_t[2]);
w1_t[3] = rotl32 ((w1_t[0] ^ w3_t[3] ^ w2_t[1] ^ w1_t[3]), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w1_t[3]);
#undef K
#define K SHA1C02
w2_t[0] = rotl32 ((w1_t[1] ^ w0_t[0] ^ w2_t[2] ^ w2_t[0]), 1u); SHA1_STEP (SHA1_F2o, a, b, c, d, e, w2_t[0]);
w2_t[1] = rotl32 ((w1_t[2] ^ w0_t[1] ^ w2_t[3] ^ w2_t[1]), 1u); SHA1_STEP (SHA1_F2o, e, a, b, c, d, w2_t[1]);
w2_t[2] = rotl32 ((w1_t[3] ^ w0_t[2] ^ w3_t[0] ^ w2_t[2]), 1u); SHA1_STEP (SHA1_F2o, d, e, a, b, c, w2_t[2]);
w2_t[3] = rotl32 ((w2_t[0] ^ w0_t[3] ^ w3_t[1] ^ w2_t[3]), 1u); SHA1_STEP (SHA1_F2o, c, d, e, a, b, w2_t[3]);
w3_t[0] = rotl32 ((w2_t[1] ^ w1_t[0] ^ w3_t[2] ^ w3_t[0]), 1u); SHA1_STEP (SHA1_F2o, b, c, d, e, a, w3_t[0]);
w3_t[1] = rotl32 ((w2_t[2] ^ w1_t[1] ^ w3_t[3] ^ w3_t[1]), 1u); SHA1_STEP (SHA1_F2o, a, b, c, d, e, w3_t[1]);
w3_t[2] = rotl32 ((w2_t[3] ^ w1_t[2] ^ w0_t[0] ^ w3_t[2]), 1u); SHA1_STEP (SHA1_F2o, e, a, b, c, d, w3_t[2]);
w3_t[3] = rotl32 ((w3_t[0] ^ w1_t[3] ^ w0_t[1] ^ w3_t[3]), 1u); SHA1_STEP (SHA1_F2o, d, e, a, b, c, w3_t[3]);
w0_t[0] = rotl32 ((w3_t[1] ^ w2_t[0] ^ w0_t[2] ^ w0_t[0]), 1u); SHA1_STEP (SHA1_F2o, c, d, e, a, b, w0_t[0]);
w0_t[1] = rotl32 ((w3_t[2] ^ w2_t[1] ^ w0_t[3] ^ w0_t[1]), 1u); SHA1_STEP (SHA1_F2o, b, c, d, e, a, w0_t[1]);
w0_t[2] = rotl32 ((w3_t[3] ^ w2_t[2] ^ w1_t[0] ^ w0_t[2]), 1u); SHA1_STEP (SHA1_F2o, a, b, c, d, e, w0_t[2]);
w0_t[3] = rotl32 ((w0_t[0] ^ w2_t[3] ^ w1_t[1] ^ w0_t[3]), 1u); SHA1_STEP (SHA1_F2o, e, a, b, c, d, w0_t[3]);
w1_t[0] = rotl32 ((w0_t[1] ^ w3_t[0] ^ w1_t[2] ^ w1_t[0]), 1u); SHA1_STEP (SHA1_F2o, d, e, a, b, c, w1_t[0]);
w1_t[1] = rotl32 ((w0_t[2] ^ w3_t[1] ^ w1_t[3] ^ w1_t[1]), 1u); SHA1_STEP (SHA1_F2o, c, d, e, a, b, w1_t[1]);
w1_t[2] = rotl32 ((w0_t[3] ^ w3_t[2] ^ w2_t[0] ^ w1_t[2]), 1u); SHA1_STEP (SHA1_F2o, b, c, d, e, a, w1_t[2]);
w1_t[3] = rotl32 ((w1_t[0] ^ w3_t[3] ^ w2_t[1] ^ w1_t[3]), 1u); SHA1_STEP (SHA1_F2o, a, b, c, d, e, w1_t[3]);
w2_t[0] = rotl32 ((w1_t[1] ^ w0_t[0] ^ w2_t[2] ^ w2_t[0]), 1u); SHA1_STEP (SHA1_F2o, e, a, b, c, d, w2_t[0]);
w2_t[1] = rotl32 ((w1_t[2] ^ w0_t[1] ^ w2_t[3] ^ w2_t[1]), 1u); SHA1_STEP (SHA1_F2o, d, e, a, b, c, w2_t[1]);
w2_t[2] = rotl32 ((w1_t[3] ^ w0_t[2] ^ w3_t[0] ^ w2_t[2]), 1u); SHA1_STEP (SHA1_F2o, c, d, e, a, b, w2_t[2]);
w2_t[3] = rotl32 ((w2_t[0] ^ w0_t[3] ^ w3_t[1] ^ w2_t[3]), 1u); SHA1_STEP (SHA1_F2o, b, c, d, e, a, w2_t[3]);
#undef K
#define K SHA1C03
w3_t[0] = rotl32 ((w2_t[1] ^ w1_t[0] ^ w3_t[2] ^ w3_t[0]), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w3_t[0]);
w3_t[1] = rotl32 ((w2_t[2] ^ w1_t[1] ^ w3_t[3] ^ w3_t[1]), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w3_t[1]);
w3_t[2] = rotl32 ((w2_t[3] ^ w1_t[2] ^ w0_t[0] ^ w3_t[2]), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w3_t[2]);
w3_t[3] = rotl32 ((w3_t[0] ^ w1_t[3] ^ w0_t[1] ^ w3_t[3]), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w3_t[3]);
w0_t[0] = rotl32 ((w3_t[1] ^ w2_t[0] ^ w0_t[2] ^ w0_t[0]), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w0_t[0]);
w0_t[1] = rotl32 ((w3_t[2] ^ w2_t[1] ^ w0_t[3] ^ w0_t[1]), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w0_t[1]);
w0_t[2] = rotl32 ((w3_t[3] ^ w2_t[2] ^ w1_t[0] ^ w0_t[2]), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w0_t[2]);
w0_t[3] = rotl32 ((w0_t[0] ^ w2_t[3] ^ w1_t[1] ^ w0_t[3]), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w0_t[3]);
w1_t[0] = rotl32 ((w0_t[1] ^ w3_t[0] ^ w1_t[2] ^ w1_t[0]), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w1_t[0]);
w1_t[1] = rotl32 ((w0_t[2] ^ w3_t[1] ^ w1_t[3] ^ w1_t[1]), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w1_t[1]);
w1_t[2] = rotl32 ((w0_t[3] ^ w3_t[2] ^ w2_t[0] ^ w1_t[2]), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w1_t[2]);
w1_t[3] = rotl32 ((w1_t[0] ^ w3_t[3] ^ w2_t[1] ^ w1_t[3]), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w1_t[3]);
w2_t[0] = rotl32 ((w1_t[1] ^ w0_t[0] ^ w2_t[2] ^ w2_t[0]), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w2_t[0]);
w2_t[1] = rotl32 ((w1_t[2] ^ w0_t[1] ^ w2_t[3] ^ w2_t[1]), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w2_t[1]);
w2_t[2] = rotl32 ((w1_t[3] ^ w0_t[2] ^ w3_t[0] ^ w2_t[2]), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w2_t[2]);
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]);
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]);
Converted _a3 kernels, use SIMD for CPU and GPU
2016-01-23 14:32:31 +00:00
COMPARE_M_SIMD (d, e, c, b);
Initial commit
2015-12-04 14:47:52 +00:00
}
}
Replace the substring GPU to a more appropriate "device" or "kernel" substring depending on the context
2016-01-05 07:26:44 +00:00
static void m00120s (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)
Initial commit
2015-12-04 14:47:52 +00:00
{
/**
* modifier
*/
const u32 gid = get_global_id (0);
const u32 lid = get_local_id (0);
/**
* digest
*/
const u32 search[4] =
{
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]
};
/**
* reverse
*/
Converted _a3 kernels, use SIMD for CPU and GPU
2016-01-23 14:32:31 +00:00
const u32 e_rev = rotl32_S (search[1], 2u);
Initial commit
2015-12-04 14:47:52 +00:00
/**
* salt
*/
u32 salt_buf0[4];
salt_buf0[0] = salt_bufs[salt_pos].salt_buf[ 0];
salt_buf0[1] = salt_bufs[salt_pos].salt_buf[ 1];
salt_buf0[2] = salt_bufs[salt_pos].salt_buf[ 2];
salt_buf0[3] = salt_bufs[salt_pos].salt_buf[ 3];
u32 salt_buf1[4];
salt_buf1[0] = salt_bufs[salt_pos].salt_buf[ 4];
salt_buf1[1] = salt_bufs[salt_pos].salt_buf[ 5];
salt_buf1[2] = salt_bufs[salt_pos].salt_buf[ 6];
salt_buf1[3] = salt_bufs[salt_pos].salt_buf[ 7];
u32 salt_buf2[4];
salt_buf2[0] = 0;
salt_buf2[1] = 0;
salt_buf2[2] = 0;
salt_buf2[3] = 0;
u32 salt_buf3[4];
salt_buf3[0] = 0;
salt_buf3[1] = 0;
salt_buf3[2] = 0;
salt_buf3[3] = 0;
const u32 salt_len = salt_bufs[salt_pos].salt_len;
const u32 pw_salt_len = pw_len + salt_len;
/**
* prepend salt
*/
- Dropped all vector code since new GPU's are all scalar, makes the code much easier - Some performance on low-end GPU may drop because of that, but only for a few hash-modes - Dropped scalar code (aka warp) since we do not have any vector datatypes anymore - Renamed C++ overloading functions memcat32_9 -> memcat_c32_w4x4_a3x4 - Still need to fix kernels to new function names, needs to be done manually - Temperature Management needs to be rewritten partially because of conflicting datatypes names - Added code to create different codepaths for NV on AMD in runtime in host (see data.vendor_id) - Added code to create different codepaths for NV on AMD in runtime in kernels (see IS_NV and IS_AMD) - First tests working for -m 0, for example - Great performance increases in general for NV so far - Tested amp_* and markov_* kernel - Migrated special NV optimizations for rule processor
2015-12-15 11:04:22 +00:00
u32 w0_t[4];
u32 w1_t[4];
u32 w2_t[4];
u32 w3_t[4];
Initial commit
2015-12-04 14:47:52 +00:00
Converted _a3 kernels, use SIMD for CPU and GPU
2016-01-23 14:32:31 +00:00
w0_t[0] = swap32_S (w0[0]);
w0_t[1] = swap32_S (w0[1]);
w0_t[2] = swap32_S (w0[2]);
w0_t[3] = swap32_S (w0[3]);
w1_t[0] = swap32_S (w1[0]);
w1_t[1] = swap32_S (w1[1]);
w1_t[2] = swap32_S (w1[2]);
w1_t[3] = swap32_S (w1[3]);
w2_t[0] = swap32_S (w2[0]);
w2_t[1] = swap32_S (w2[1]);
w2_t[2] = swap32_S (w2[2]);
w2_t[3] = swap32_S (w2[3]);
w3_t[0] = swap32_S (w3[0]);
w3_t[1] = swap32_S (w3[1]);
w3_t[2] = swap32_S (w3[2]);
w3_t[3] = swap32_S (w3[3]);
switch_buffer_by_offset_le_S (w0_t, w1_t, w2_t, w3_t, salt_len);
Initial commit
2015-12-04 14:47:52 +00:00
w0_t[0] |= salt_buf0[0];
w0_t[1] |= salt_buf0[1];
w0_t[2] |= salt_buf0[2];
w0_t[3] |= salt_buf0[3];
w1_t[0] |= salt_buf1[0];
w1_t[1] |= salt_buf1[1];
w1_t[2] |= salt_buf1[2];
w1_t[3] |= salt_buf1[3];
w2_t[0] |= salt_buf2[0];
w2_t[1] |= salt_buf2[1];
w2_t[2] |= salt_buf2[2];
w2_t[3] |= salt_buf2[3];
w3_t[0] |= salt_buf3[0];
w3_t[1] |= salt_buf3[1];
w3_t[2] |= salt_buf3[2];
w3_t[3] |= salt_buf3[3];
Converted _a3 kernels, use SIMD for CPU and GPU
2016-01-23 14:32:31 +00:00
w0_t[0] = swap32_S (w0_t[0]);
w0_t[1] = swap32_S (w0_t[1]);
w0_t[2] = swap32_S (w0_t[2]);
w0_t[3] = swap32_S (w0_t[3]);
w1_t[0] = swap32_S (w1_t[0]);
w1_t[1] = swap32_S (w1_t[1]);
w1_t[2] = swap32_S (w1_t[2]);
w1_t[3] = swap32_S (w1_t[3]);
w2_t[0] = swap32_S (w2_t[0]);
w2_t[1] = swap32_S (w2_t[1]);
w2_t[2] = swap32_S (w2_t[2]);
w2_t[3] = swap32_S (w2_t[3]);
w3_t[0] = swap32_S (w3_t[0]);
w3_t[1] = swap32_S (w3_t[1]);
w3_t[2] = swap32_S (w3_t[2]);
w3_t[3] = swap32_S (w3_t[3]);
Initial commit
2015-12-04 14:47:52 +00:00
/**
* loop
*/
- Dropped all vector code since new GPU's are all scalar, makes the code much easier - Some performance on low-end GPU may drop because of that, but only for a few hash-modes - Dropped scalar code (aka warp) since we do not have any vector datatypes anymore - Renamed C++ overloading functions memcat32_9 -> memcat_c32_w4x4_a3x4 - Still need to fix kernels to new function names, needs to be done manually - Temperature Management needs to be rewritten partially because of conflicting datatypes names - Added code to create different codepaths for NV on AMD in runtime in host (see data.vendor_id) - Added code to create different codepaths for NV on AMD in runtime in kernels (see IS_NV and IS_AMD) - First tests working for -m 0, for example - Great performance increases in general for NV so far - Tested amp_* and markov_* kernel - Migrated special NV optimizations for rule processor
2015-12-15 11:04:22 +00:00
u32 w0l = w0[0];
Initial commit
2015-12-04 14:47:52 +00:00
Converted _a3 kernels, use SIMD for CPU and GPU
2016-01-23 14:32:31 +00:00
for (u32 il_pos = 0; il_pos < bfs_cnt; il_pos += VECT_SIZE)
Initial commit
2015-12-04 14:47:52 +00:00
{
Prepare NEW_SIMD_MODE for -a 1 kernels
2016-02-20 15:13:06 +00:00
const u32x w0r = ix_create_bft (bfs_buf, il_pos);
Initial commit
2015-12-04 14:47:52 +00:00
Converted _a3 kernels, use SIMD for CPU and GPU
2016-01-23 14:32:31 +00:00
const u32x w0lr = w0l | w0r;
Initial commit
2015-12-04 14:47:52 +00:00
Converted _a3 kernels, use SIMD for CPU and GPU
2016-01-23 14:32:31 +00:00
u32x wx[16];
Initial commit
2015-12-04 14:47:52 +00:00
wx[ 0] = w0_t[0];
wx[ 1] = w0_t[1];
wx[ 2] = w0_t[2];
wx[ 3] = w0_t[3];
wx[ 4] = w1_t[0];
wx[ 5] = w1_t[1];
wx[ 6] = w1_t[2];
wx[ 7] = w1_t[3];
wx[ 8] = w2_t[0];
wx[ 9] = w2_t[1];
wx[10] = w2_t[2];
wx[11] = w2_t[3];
wx[12] = w3_t[0];
wx[13] = w3_t[1];
wx[14] = w3_t[2];
wx[15] = w3_t[3];
Converted _a3 kernels, use SIMD for CPU and GPU
2016-01-23 14:32:31 +00:00
overwrite_at_be (wx, w0lr, salt_len);
Initial commit
2015-12-04 14:47:52 +00:00
Converted _a3 kernels, use SIMD for CPU and GPU
2016-01-23 14:32:31 +00:00
u32x w0_t[4];
u32x w1_t[4];
u32x w2_t[4];
u32x w3_t[4];
Initial commit
2015-12-04 14:47:52 +00:00
w0_t[0] = wx[ 0];
w0_t[1] = wx[ 1];
w0_t[2] = wx[ 2];
w0_t[3] = wx[ 3];
w1_t[0] = wx[ 4];
w1_t[1] = wx[ 5];
w1_t[2] = wx[ 6];
w1_t[3] = wx[ 7];
w2_t[0] = wx[ 8];
w2_t[1] = wx[ 9];
w2_t[2] = wx[10];
w2_t[3] = wx[11];
w3_t[0] = wx[12];
w3_t[1] = wx[13];
w3_t[2] = 0;
w3_t[3] = pw_salt_len * 8;
/**
* sha1
*/
Converted _a3 kernels, use SIMD for CPU and GPU
2016-01-23 14:32:31 +00:00
u32x a = SHA1M_A;
u32x b = SHA1M_B;
u32x c = SHA1M_C;
u32x d = SHA1M_D;
u32x e = SHA1M_E;
Initial commit
2015-12-04 14:47:52 +00:00
#undef K
#define K SHA1C00
SHA1_STEP (SHA1_F0o, a, b, c, d, e, w0_t[0]);
SHA1_STEP (SHA1_F0o, e, a, b, c, d, w0_t[1]);
SHA1_STEP (SHA1_F0o, d, e, a, b, c, w0_t[2]);
SHA1_STEP (SHA1_F0o, c, d, e, a, b, w0_t[3]);
SHA1_STEP (SHA1_F0o, b, c, d, e, a, w1_t[0]);
SHA1_STEP (SHA1_F0o, a, b, c, d, e, w1_t[1]);
SHA1_STEP (SHA1_F0o, e, a, b, c, d, w1_t[2]);
SHA1_STEP (SHA1_F0o, d, e, a, b, c, w1_t[3]);
SHA1_STEP (SHA1_F0o, c, d, e, a, b, w2_t[0]);
SHA1_STEP (SHA1_F0o, b, c, d, e, a, w2_t[1]);
SHA1_STEP (SHA1_F0o, a, b, c, d, e, w2_t[2]);
SHA1_STEP (SHA1_F0o, e, a, b, c, d, w2_t[3]);
SHA1_STEP (SHA1_F0o, d, e, a, b, c, w3_t[0]);
SHA1_STEP (SHA1_F0o, c, d, e, a, b, w3_t[1]);
SHA1_STEP (SHA1_F0o, b, c, d, e, a, w3_t[2]);
SHA1_STEP (SHA1_F0o, a, b, c, d, e, w3_t[3]);
w0_t[0] = rotl32 ((w3_t[1] ^ w2_t[0] ^ w0_t[2] ^ w0_t[0]), 1u); SHA1_STEP (SHA1_F0o, e, a, b, c, d, w0_t[0]);
w0_t[1] = rotl32 ((w3_t[2] ^ w2_t[1] ^ w0_t[3] ^ w0_t[1]), 1u); SHA1_STEP (SHA1_F0o, d, e, a, b, c, w0_t[1]);
w0_t[2] = rotl32 ((w3_t[3] ^ w2_t[2] ^ w1_t[0] ^ w0_t[2]), 1u); SHA1_STEP (SHA1_F0o, c, d, e, a, b, w0_t[2]);
w0_t[3] = rotl32 ((w0_t[0] ^ w2_t[3] ^ w1_t[1] ^ w0_t[3]), 1u); SHA1_STEP (SHA1_F0o, b, c, d, e, a, w0_t[3]);
#undef K
#define K SHA1C01
w1_t[0] = rotl32 ((w0_t[1] ^ w3_t[0] ^ w1_t[2] ^ w1_t[0]), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w1_t[0]);
w1_t[1] = rotl32 ((w0_t[2] ^ w3_t[1] ^ w1_t[3] ^ w1_t[1]), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w1_t[1]);
w1_t[2] = rotl32 ((w0_t[3] ^ w3_t[2] ^ w2_t[0] ^ w1_t[2]), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w1_t[2]);
w1_t[3] = rotl32 ((w1_t[0] ^ w3_t[3] ^ w2_t[1] ^ w1_t[3]), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w1_t[3]);
w2_t[0] = rotl32 ((w1_t[1] ^ w0_t[0] ^ w2_t[2] ^ w2_t[0]), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w2_t[0]);
w2_t[1] = rotl32 ((w1_t[2] ^ w0_t[1] ^ w2_t[3] ^ w2_t[1]), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w2_t[1]);
w2_t[2] = rotl32 ((w1_t[3] ^ w0_t[2] ^ w3_t[0] ^ w2_t[2]), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w2_t[2]);
w2_t[3] = rotl32 ((w2_t[0] ^ w0_t[3] ^ w3_t[1] ^ w2_t[3]), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w2_t[3]);
w3_t[0] = rotl32 ((w2_t[1] ^ w1_t[0] ^ w3_t[2] ^ w3_t[0]), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w3_t[0]);
w3_t[1] = rotl32 ((w2_t[2] ^ w1_t[1] ^ w3_t[3] ^ w3_t[1]), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w3_t[1]);
w3_t[2] = rotl32 ((w2_t[3] ^ w1_t[2] ^ w0_t[0] ^ w3_t[2]), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w3_t[2]);
w3_t[3] = rotl32 ((w3_t[0] ^ w1_t[3] ^ w0_t[1] ^ w3_t[3]), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w3_t[3]);
w0_t[0] = rotl32 ((w3_t[1] ^ w2_t[0] ^ w0_t[2] ^ w0_t[0]), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w0_t[0]);
w0_t[1] = rotl32 ((w3_t[2] ^ w2_t[1] ^ w0_t[3] ^ w0_t[1]), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w0_t[1]);
w0_t[2] = rotl32 ((w3_t[3] ^ w2_t[2] ^ w1_t[0] ^ w0_t[2]), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w0_t[2]);
w0_t[3] = rotl32 ((w0_t[0] ^ w2_t[3] ^ w1_t[1] ^ w0_t[3]), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w0_t[3]);
w1_t[0] = rotl32 ((w0_t[1] ^ w3_t[0] ^ w1_t[2] ^ w1_t[0]), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w1_t[0]);
w1_t[1] = rotl32 ((w0_t[2] ^ w3_t[1] ^ w1_t[3] ^ w1_t[1]), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w1_t[1]);
w1_t[2] = rotl32 ((w0_t[3] ^ w3_t[2] ^ w2_t[0] ^ w1_t[2]), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w1_t[2]);
w1_t[3] = rotl32 ((w1_t[0] ^ w3_t[3] ^ w2_t[1] ^ w1_t[3]), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w1_t[3]);
#undef K
#define K SHA1C02
w2_t[0] = rotl32 ((w1_t[1] ^ w0_t[0] ^ w2_t[2] ^ w2_t[0]), 1u); SHA1_STEP (SHA1_F2o, a, b, c, d, e, w2_t[0]);
w2_t[1] = rotl32 ((w1_t[2] ^ w0_t[1] ^ w2_t[3] ^ w2_t[1]), 1u); SHA1_STEP (SHA1_F2o, e, a, b, c, d, w2_t[1]);
w2_t[2] = rotl32 ((w1_t[3] ^ w0_t[2] ^ w3_t[0] ^ w2_t[2]), 1u); SHA1_STEP (SHA1_F2o, d, e, a, b, c, w2_t[2]);
w2_t[3] = rotl32 ((w2_t[0] ^ w0_t[3] ^ w3_t[1] ^ w2_t[3]), 1u); SHA1_STEP (SHA1_F2o, c, d, e, a, b, w2_t[3]);
w3_t[0] = rotl32 ((w2_t[1] ^ w1_t[0] ^ w3_t[2] ^ w3_t[0]), 1u); SHA1_STEP (SHA1_F2o, b, c, d, e, a, w3_t[0]);
w3_t[1] = rotl32 ((w2_t[2] ^ w1_t[1] ^ w3_t[3] ^ w3_t[1]), 1u); SHA1_STEP (SHA1_F2o, a, b, c, d, e, w3_t[1]);
w3_t[2] = rotl32 ((w2_t[3] ^ w1_t[2] ^ w0_t[0] ^ w3_t[2]), 1u); SHA1_STEP (SHA1_F2o, e, a, b, c, d, w3_t[2]);
w3_t[3] = rotl32 ((w3_t[0] ^ w1_t[3] ^ w0_t[1] ^ w3_t[3]), 1u); SHA1_STEP (SHA1_F2o, d, e, a, b, c, w3_t[3]);
w0_t[0] = rotl32 ((w3_t[1] ^ w2_t[0] ^ w0_t[2] ^ w0_t[0]), 1u); SHA1_STEP (SHA1_F2o, c, d, e, a, b, w0_t[0]);
w0_t[1] = rotl32 ((w3_t[2] ^ w2_t[1] ^ w0_t[3] ^ w0_t[1]), 1u); SHA1_STEP (SHA1_F2o, b, c, d, e, a, w0_t[1]);
w0_t[2] = rotl32 ((w3_t[3] ^ w2_t[2] ^ w1_t[0] ^ w0_t[2]), 1u); SHA1_STEP (SHA1_F2o, a, b, c, d, e, w0_t[2]);
w0_t[3] = rotl32 ((w0_t[0] ^ w2_t[3] ^ w1_t[1] ^ w0_t[3]), 1u); SHA1_STEP (SHA1_F2o, e, a, b, c, d, w0_t[3]);
w1_t[0] = rotl32 ((w0_t[1] ^ w3_t[0] ^ w1_t[2] ^ w1_t[0]), 1u); SHA1_STEP (SHA1_F2o, d, e, a, b, c, w1_t[0]);
w1_t[1] = rotl32 ((w0_t[2] ^ w3_t[1] ^ w1_t[3] ^ w1_t[1]), 1u); SHA1_STEP (SHA1_F2o, c, d, e, a, b, w1_t[1]);
w1_t[2] = rotl32 ((w0_t[3] ^ w3_t[2] ^ w2_t[0] ^ w1_t[2]), 1u); SHA1_STEP (SHA1_F2o, b, c, d, e, a, w1_t[2]);
w1_t[3] = rotl32 ((w1_t[0] ^ w3_t[3] ^ w2_t[1] ^ w1_t[3]), 1u); SHA1_STEP (SHA1_F2o, a, b, c, d, e, w1_t[3]);
w2_t[0] = rotl32 ((w1_t[1] ^ w0_t[0] ^ w2_t[2] ^ w2_t[0]), 1u); SHA1_STEP (SHA1_F2o, e, a, b, c, d, w2_t[0]);
w2_t[1] = rotl32 ((w1_t[2] ^ w0_t[1] ^ w2_t[3] ^ w2_t[1]), 1u); SHA1_STEP (SHA1_F2o, d, e, a, b, c, w2_t[1]);
w2_t[2] = rotl32 ((w1_t[3] ^ w0_t[2] ^ w3_t[0] ^ w2_t[2]), 1u); SHA1_STEP (SHA1_F2o, c, d, e, a, b, w2_t[2]);
w2_t[3] = rotl32 ((w2_t[0] ^ w0_t[3] ^ w3_t[1] ^ w2_t[3]), 1u); SHA1_STEP (SHA1_F2o, b, c, d, e, a, w2_t[3]);
#undef K
#define K SHA1C03
w3_t[0] = rotl32 ((w2_t[1] ^ w1_t[0] ^ w3_t[2] ^ w3_t[0]), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w3_t[0]);
w3_t[1] = rotl32 ((w2_t[2] ^ w1_t[1] ^ w3_t[3] ^ w3_t[1]), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w3_t[1]);
w3_t[2] = rotl32 ((w2_t[3] ^ w1_t[2] ^ w0_t[0] ^ w3_t[2]), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w3_t[2]);
w3_t[3] = rotl32 ((w3_t[0] ^ w1_t[3] ^ w0_t[1] ^ w3_t[3]), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w3_t[3]);
w0_t[0] = rotl32 ((w3_t[1] ^ w2_t[0] ^ w0_t[2] ^ w0_t[0]), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w0_t[0]);
w0_t[1] = rotl32 ((w3_t[2] ^ w2_t[1] ^ w0_t[3] ^ w0_t[1]), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w0_t[1]);
w0_t[2] = rotl32 ((w3_t[3] ^ w2_t[2] ^ w1_t[0] ^ w0_t[2]), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w0_t[2]);
w0_t[3] = rotl32 ((w0_t[0] ^ w2_t[3] ^ w1_t[1] ^ w0_t[3]), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w0_t[3]);
w1_t[0] = rotl32 ((w0_t[1] ^ w3_t[0] ^ w1_t[2] ^ w1_t[0]), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w1_t[0]);
w1_t[1] = rotl32 ((w0_t[2] ^ w3_t[1] ^ w1_t[3] ^ w1_t[1]), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w1_t[1]);
w1_t[2] = rotl32 ((w0_t[3] ^ w3_t[2] ^ w2_t[0] ^ w1_t[2]), 1u); SHA1_STEP (SHA1_F1, a, b, c, d, e, w1_t[2]);
w1_t[3] = rotl32 ((w1_t[0] ^ w3_t[3] ^ w2_t[1] ^ w1_t[3]), 1u); SHA1_STEP (SHA1_F1, e, a, b, c, d, w1_t[3]);
w2_t[0] = rotl32 ((w1_t[1] ^ w0_t[0] ^ w2_t[2] ^ w2_t[0]), 1u); SHA1_STEP (SHA1_F1, d, e, a, b, c, w2_t[0]);
w2_t[1] = rotl32 ((w1_t[2] ^ w0_t[1] ^ w2_t[3] ^ w2_t[1]), 1u); SHA1_STEP (SHA1_F1, c, d, e, a, b, w2_t[1]);
w2_t[2] = rotl32 ((w1_t[3] ^ w0_t[2] ^ w3_t[0] ^ w2_t[2]), 1u); SHA1_STEP (SHA1_F1, b, c, d, e, a, w2_t[2]);
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]);
Converted _a3 kernels, use SIMD for CPU and GPU
2016-01-23 14:32:31 +00:00
if (MATCHES_NONE_VS (e, e_rev)) continue;
Initial commit
2015-12-04 14:47:52 +00:00
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]);
Converted _a3 kernels, use SIMD for CPU and GPU
2016-01-23 14:32:31 +00:00
COMPARE_S_SIMD (d, e, c, b);
Initial commit
2015-12-04 14:47:52 +00:00
}
}
Upgrade kernel to support dynamic local work sizes
2016-01-19 15:06:03 +00:00
__kernel void m00120_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 bfs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
Initial commit
2015-12-04 14:47:52 +00:00
{
/**
* base
*/
const u32 gid = get_global_id (0);
if (gid >= gid_max) return;
- Dropped all vector code since new GPU's are all scalar, makes the code much easier - Some performance on low-end GPU may drop because of that, but only for a few hash-modes - Dropped scalar code (aka warp) since we do not have any vector datatypes anymore - Renamed C++ overloading functions memcat32_9 -> memcat_c32_w4x4_a3x4 - Still need to fix kernels to new function names, needs to be done manually - Temperature Management needs to be rewritten partially because of conflicting datatypes names - Added code to create different codepaths for NV on AMD in runtime in host (see data.vendor_id) - Added code to create different codepaths for NV on AMD in runtime in kernels (see IS_NV and IS_AMD) - First tests working for -m 0, for example - Great performance increases in general for NV so far - Tested amp_* and markov_* kernel - Migrated special NV optimizations for rule processor
2015-12-15 11:04:22 +00:00
u32 w0[4];
Initial commit
2015-12-04 14:47:52 +00:00
w0[0] = pws[gid].i[ 0];
w0[1] = pws[gid].i[ 1];
w0[2] = pws[gid].i[ 2];
w0[3] = pws[gid].i[ 3];
- Dropped all vector code since new GPU's are all scalar, makes the code much easier - Some performance on low-end GPU may drop because of that, but only for a few hash-modes - Dropped scalar code (aka warp) since we do not have any vector datatypes anymore - Renamed C++ overloading functions memcat32_9 -> memcat_c32_w4x4_a3x4 - Still need to fix kernels to new function names, needs to be done manually - Temperature Management needs to be rewritten partially because of conflicting datatypes names - Added code to create different codepaths for NV on AMD in runtime in host (see data.vendor_id) - Added code to create different codepaths for NV on AMD in runtime in kernels (see IS_NV and IS_AMD) - First tests working for -m 0, for example - Great performance increases in general for NV so far - Tested amp_* and markov_* kernel - Migrated special NV optimizations for rule processor
2015-12-15 11:04:22 +00:00
u32 w1[4];
Initial commit
2015-12-04 14:47:52 +00:00
w1[0] = 0;
w1[1] = 0;
w1[2] = 0;
w1[3] = 0;
- Dropped all vector code since new GPU's are all scalar, makes the code much easier - Some performance on low-end GPU may drop because of that, but only for a few hash-modes - Dropped scalar code (aka warp) since we do not have any vector datatypes anymore - Renamed C++ overloading functions memcat32_9 -> memcat_c32_w4x4_a3x4 - Still need to fix kernels to new function names, needs to be done manually - Temperature Management needs to be rewritten partially because of conflicting datatypes names - Added code to create different codepaths for NV on AMD in runtime in host (see data.vendor_id) - Added code to create different codepaths for NV on AMD in runtime in kernels (see IS_NV and IS_AMD) - First tests working for -m 0, for example - Great performance increases in general for NV so far - Tested amp_* and markov_* kernel - Migrated special NV optimizations for rule processor
2015-12-15 11:04:22 +00:00
u32 w2[4];
Initial commit
2015-12-04 14:47:52 +00:00
w2[0] = 0;
w2[1] = 0;
w2[2] = 0;
w2[3] = 0;
- Dropped all vector code since new GPU's are all scalar, makes the code much easier - Some performance on low-end GPU may drop because of that, but only for a few hash-modes - Dropped scalar code (aka warp) since we do not have any vector datatypes anymore - Renamed C++ overloading functions memcat32_9 -> memcat_c32_w4x4_a3x4 - Still need to fix kernels to new function names, needs to be done manually - Temperature Management needs to be rewritten partially because of conflicting datatypes names - Added code to create different codepaths for NV on AMD in runtime in host (see data.vendor_id) - Added code to create different codepaths for NV on AMD in runtime in kernels (see IS_NV and IS_AMD) - First tests working for -m 0, for example - Great performance increases in general for NV so far - Tested amp_* and markov_* kernel - Migrated special NV optimizations for rule processor
2015-12-15 11:04:22 +00:00
u32 w3[4];
Initial commit
2015-12-04 14:47:52 +00:00
w3[0] = 0;
w3[1] = 0;
w3[2] = 0;
w3[3] = 0;
const u32 pw_len = pws[gid].pw_len;
/**
* main
*/
m00120m (w0, w1, w2, w3, pw_len, pws, rules_buf, combs_buf, bfs_buf, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_scryptV_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, bfs_cnt, digests_cnt, digests_offset);
}
Upgrade kernel to support dynamic local work sizes
2016-01-19 15:06:03 +00:00
__kernel void m00120_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)
Initial commit
2015-12-04 14:47:52 +00:00
{
/**
* base
*/
const u32 gid = get_global_id (0);
if (gid >= gid_max) return;
- Dropped all vector code since new GPU's are all scalar, makes the code much easier - Some performance on low-end GPU may drop because of that, but only for a few hash-modes - Dropped scalar code (aka warp) since we do not have any vector datatypes anymore - Renamed C++ overloading functions memcat32_9 -> memcat_c32_w4x4_a3x4 - Still need to fix kernels to new function names, needs to be done manually - Temperature Management needs to be rewritten partially because of conflicting datatypes names - Added code to create different codepaths for NV on AMD in runtime in host (see data.vendor_id) - Added code to create different codepaths for NV on AMD in runtime in kernels (see IS_NV and IS_AMD) - First tests working for -m 0, for example - Great performance increases in general for NV so far - Tested amp_* and markov_* kernel - Migrated special NV optimizations for rule processor
2015-12-15 11:04:22 +00:00
u32 w0[4];
Initial commit
2015-12-04 14:47:52 +00:00
w0[0] = pws[gid].i[ 0];
w0[1] = pws[gid].i[ 1];
w0[2] = pws[gid].i[ 2];
w0[3] = pws[gid].i[ 3];
- Dropped all vector code since new GPU's are all scalar, makes the code much easier - Some performance on low-end GPU may drop because of that, but only for a few hash-modes - Dropped scalar code (aka warp) since we do not have any vector datatypes anymore - Renamed C++ overloading functions memcat32_9 -> memcat_c32_w4x4_a3x4 - Still need to fix kernels to new function names, needs to be done manually - Temperature Management needs to be rewritten partially because of conflicting datatypes names - Added code to create different codepaths for NV on AMD in runtime in host (see data.vendor_id) - Added code to create different codepaths for NV on AMD in runtime in kernels (see IS_NV and IS_AMD) - First tests working for -m 0, for example - Great performance increases in general for NV so far - Tested amp_* and markov_* kernel - Migrated special NV optimizations for rule processor
2015-12-15 11:04:22 +00:00
u32 w1[4];
Initial commit
2015-12-04 14:47:52 +00:00
w1[0] = pws[gid].i[ 4];
w1[1] = pws[gid].i[ 5];
w1[2] = pws[gid].i[ 6];
w1[3] = pws[gid].i[ 7];
- Dropped all vector code since new GPU's are all scalar, makes the code much easier - Some performance on low-end GPU may drop because of that, but only for a few hash-modes - Dropped scalar code (aka warp) since we do not have any vector datatypes anymore - Renamed C++ overloading functions memcat32_9 -> memcat_c32_w4x4_a3x4 - Still need to fix kernels to new function names, needs to be done manually - Temperature Management needs to be rewritten partially because of conflicting datatypes names - Added code to create different codepaths for NV on AMD in runtime in host (see data.vendor_id) - Added code to create different codepaths for NV on AMD in runtime in kernels (see IS_NV and IS_AMD) - First tests working for -m 0, for example - Great performance increases in general for NV so far - Tested amp_* and markov_* kernel - Migrated special NV optimizations for rule processor
2015-12-15 11:04:22 +00:00
u32 w2[4];
Initial commit
2015-12-04 14:47:52 +00:00
w2[0] = 0;
w2[1] = 0;
w2[2] = 0;
w2[3] = 0;
- Dropped all vector code since new GPU's are all scalar, makes the code much easier - Some performance on low-end GPU may drop because of that, but only for a few hash-modes - Dropped scalar code (aka warp) since we do not have any vector datatypes anymore - Renamed C++ overloading functions memcat32_9 -> memcat_c32_w4x4_a3x4 - Still need to fix kernels to new function names, needs to be done manually - Temperature Management needs to be rewritten partially because of conflicting datatypes names - Added code to create different codepaths for NV on AMD in runtime in host (see data.vendor_id) - Added code to create different codepaths for NV on AMD in runtime in kernels (see IS_NV and IS_AMD) - First tests working for -m 0, for example - Great performance increases in general for NV so far - Tested amp_* and markov_* kernel - Migrated special NV optimizations for rule processor
2015-12-15 11:04:22 +00:00
u32 w3[4];
Initial commit
2015-12-04 14:47:52 +00:00
w3[0] = 0;
w3[1] = 0;
w3[2] = 0;
w3[3] = 0;
const u32 pw_len = pws[gid].pw_len;
/**
* main
*/
m00120m (w0, w1, w2, w3, pw_len, pws, rules_buf, combs_buf, bfs_buf, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_scryptV_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, bfs_cnt, digests_cnt, digests_offset);
}
Upgrade kernel to support dynamic local work sizes
2016-01-19 15:06:03 +00:00
__kernel void m00120_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)
Initial commit
2015-12-04 14:47:52 +00:00
{
/**
* base
*/
const u32 gid = get_global_id (0);
if (gid >= gid_max) return;
- Dropped all vector code since new GPU's are all scalar, makes the code much easier - Some performance on low-end GPU may drop because of that, but only for a few hash-modes - Dropped scalar code (aka warp) since we do not have any vector datatypes anymore - Renamed C++ overloading functions memcat32_9 -> memcat_c32_w4x4_a3x4 - Still need to fix kernels to new function names, needs to be done manually - Temperature Management needs to be rewritten partially because of conflicting datatypes names - Added code to create different codepaths for NV on AMD in runtime in host (see data.vendor_id) - Added code to create different codepaths for NV on AMD in runtime in kernels (see IS_NV and IS_AMD) - First tests working for -m 0, for example - Great performance increases in general for NV so far - Tested amp_* and markov_* kernel - Migrated special NV optimizations for rule processor
2015-12-15 11:04:22 +00:00
u32 w0[4];
Initial commit
2015-12-04 14:47:52 +00:00
w0[0] = pws[gid].i[ 0];
w0[1] = pws[gid].i[ 1];
w0[2] = pws[gid].i[ 2];
w0[3] = pws[gid].i[ 3];
- Dropped all vector code since new GPU's are all scalar, makes the code much easier - Some performance on low-end GPU may drop because of that, but only for a few hash-modes - Dropped scalar code (aka warp) since we do not have any vector datatypes anymore - Renamed C++ overloading functions memcat32_9 -> memcat_c32_w4x4_a3x4 - Still need to fix kernels to new function names, needs to be done manually - Temperature Management needs to be rewritten partially because of conflicting datatypes names - Added code to create different codepaths for NV on AMD in runtime in host (see data.vendor_id) - Added code to create different codepaths for NV on AMD in runtime in kernels (see IS_NV and IS_AMD) - First tests working for -m 0, for example - Great performance increases in general for NV so far - Tested amp_* and markov_* kernel - Migrated special NV optimizations for rule processor
2015-12-15 11:04:22 +00:00
u32 w1[4];
Initial commit
2015-12-04 14:47:52 +00:00
w1[0] = pws[gid].i[ 4];
w1[1] = pws[gid].i[ 5];
w1[2] = pws[gid].i[ 6];
w1[3] = pws[gid].i[ 7];
- Dropped all vector code since new GPU's are all scalar, makes the code much easier - Some performance on low-end GPU may drop because of that, but only for a few hash-modes - Dropped scalar code (aka warp) since we do not have any vector datatypes anymore - Renamed C++ overloading functions memcat32_9 -> memcat_c32_w4x4_a3x4 - Still need to fix kernels to new function names, needs to be done manually - Temperature Management needs to be rewritten partially because of conflicting datatypes names - Added code to create different codepaths for NV on AMD in runtime in host (see data.vendor_id) - Added code to create different codepaths for NV on AMD in runtime in kernels (see IS_NV and IS_AMD) - First tests working for -m 0, for example - Great performance increases in general for NV so far - Tested amp_* and markov_* kernel - Migrated special NV optimizations for rule processor
2015-12-15 11:04:22 +00:00
u32 w2[4];
Initial commit
2015-12-04 14:47:52 +00:00
w2[0] = pws[gid].i[ 8];
w2[1] = pws[gid].i[ 9];
w2[2] = pws[gid].i[10];
w2[3] = pws[gid].i[11];
- Dropped all vector code since new GPU's are all scalar, makes the code much easier - Some performance on low-end GPU may drop because of that, but only for a few hash-modes - Dropped scalar code (aka warp) since we do not have any vector datatypes anymore - Renamed C++ overloading functions memcat32_9 -> memcat_c32_w4x4_a3x4 - Still need to fix kernels to new function names, needs to be done manually - Temperature Management needs to be rewritten partially because of conflicting datatypes names - Added code to create different codepaths for NV on AMD in runtime in host (see data.vendor_id) - Added code to create different codepaths for NV on AMD in runtime in kernels (see IS_NV and IS_AMD) - First tests working for -m 0, for example - Great performance increases in general for NV so far - Tested amp_* and markov_* kernel - Migrated special NV optimizations for rule processor
2015-12-15 11:04:22 +00:00
u32 w3[4];
Initial commit
2015-12-04 14:47:52 +00:00
w3[0] = pws[gid].i[12];
w3[1] = pws[gid].i[13];
w3[2] = 0;
w3[3] = 0;
const u32 pw_len = pws[gid].pw_len;
/**
* main
*/
m00120m (w0, w1, w2, w3, pw_len, pws, rules_buf, combs_buf, bfs_buf, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_scryptV_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, bfs_cnt, digests_cnt, digests_offset);
}
Upgrade kernel to support dynamic local work sizes
2016-01-19 15:06:03 +00:00
__kernel void m00120_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 bfs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
Initial commit
2015-12-04 14:47:52 +00:00
{
/**
* base
*/
const u32 gid = get_global_id (0);
if (gid >= gid_max) return;
- Dropped all vector code since new GPU's are all scalar, makes the code much easier - Some performance on low-end GPU may drop because of that, but only for a few hash-modes - Dropped scalar code (aka warp) since we do not have any vector datatypes anymore - Renamed C++ overloading functions memcat32_9 -> memcat_c32_w4x4_a3x4 - Still need to fix kernels to new function names, needs to be done manually - Temperature Management needs to be rewritten partially because of conflicting datatypes names - Added code to create different codepaths for NV on AMD in runtime in host (see data.vendor_id) - Added code to create different codepaths for NV on AMD in runtime in kernels (see IS_NV and IS_AMD) - First tests working for -m 0, for example - Great performance increases in general for NV so far - Tested amp_* and markov_* kernel - Migrated special NV optimizations for rule processor
2015-12-15 11:04:22 +00:00
u32 w0[4];
Initial commit
2015-12-04 14:47:52 +00:00
w0[0] = pws[gid].i[ 0];
w0[1] = pws[gid].i[ 1];
w0[2] = pws[gid].i[ 2];
w0[3] = pws[gid].i[ 3];
- Dropped all vector code since new GPU's are all scalar, makes the code much easier - Some performance on low-end GPU may drop because of that, but only for a few hash-modes - Dropped scalar code (aka warp) since we do not have any vector datatypes anymore - Renamed C++ overloading functions memcat32_9 -> memcat_c32_w4x4_a3x4 - Still need to fix kernels to new function names, needs to be done manually - Temperature Management needs to be rewritten partially because of conflicting datatypes names - Added code to create different codepaths for NV on AMD in runtime in host (see data.vendor_id) - Added code to create different codepaths for NV on AMD in runtime in kernels (see IS_NV and IS_AMD) - First tests working for -m 0, for example - Great performance increases in general for NV so far - Tested amp_* and markov_* kernel - Migrated special NV optimizations for rule processor
2015-12-15 11:04:22 +00:00
u32 w1[4];
Initial commit
2015-12-04 14:47:52 +00:00
w1[0] = 0;
w1[1] = 0;
w1[2] = 0;
w1[3] = 0;
- Dropped all vector code since new GPU's are all scalar, makes the code much easier - Some performance on low-end GPU may drop because of that, but only for a few hash-modes - Dropped scalar code (aka warp) since we do not have any vector datatypes anymore - Renamed C++ overloading functions memcat32_9 -> memcat_c32_w4x4_a3x4 - Still need to fix kernels to new function names, needs to be done manually - Temperature Management needs to be rewritten partially because of conflicting datatypes names - Added code to create different codepaths for NV on AMD in runtime in host (see data.vendor_id) - Added code to create different codepaths for NV on AMD in runtime in kernels (see IS_NV and IS_AMD) - First tests working for -m 0, for example - Great performance increases in general for NV so far - Tested amp_* and markov_* kernel - Migrated special NV optimizations for rule processor
2015-12-15 11:04:22 +00:00
u32 w2[4];
Initial commit
2015-12-04 14:47:52 +00:00
w2[0] = 0;
w2[1] = 0;
w2[2] = 0;
w2[3] = 0;
- Dropped all vector code since new GPU's are all scalar, makes the code much easier - Some performance on low-end GPU may drop because of that, but only for a few hash-modes - Dropped scalar code (aka warp) since we do not have any vector datatypes anymore - Renamed C++ overloading functions memcat32_9 -> memcat_c32_w4x4_a3x4 - Still need to fix kernels to new function names, needs to be done manually - Temperature Management needs to be rewritten partially because of conflicting datatypes names - Added code to create different codepaths for NV on AMD in runtime in host (see data.vendor_id) - Added code to create different codepaths for NV on AMD in runtime in kernels (see IS_NV and IS_AMD) - First tests working for -m 0, for example - Great performance increases in general for NV so far - Tested amp_* and markov_* kernel - Migrated special NV optimizations for rule processor
2015-12-15 11:04:22 +00:00
u32 w3[4];
Initial commit
2015-12-04 14:47:52 +00:00
w3[0] = 0;
w3[1] = 0;
w3[2] = 0;
w3[3] = 0;
const u32 pw_len = pws[gid].pw_len;
/**
* main
*/
m00120s (w0, w1, w2, w3, pw_len, pws, rules_buf, combs_buf, bfs_buf, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_scryptV_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, bfs_cnt, digests_cnt, digests_offset);
}
Upgrade kernel to support dynamic local work sizes
2016-01-19 15:06:03 +00:00
__kernel void m00120_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)
Initial commit
2015-12-04 14:47:52 +00:00
{
/**
* base
*/
const u32 gid = get_global_id (0);
if (gid >= gid_max) return;
- Dropped all vector code since new GPU's are all scalar, makes the code much easier - Some performance on low-end GPU may drop because of that, but only for a few hash-modes - Dropped scalar code (aka warp) since we do not have any vector datatypes anymore - Renamed C++ overloading functions memcat32_9 -> memcat_c32_w4x4_a3x4 - Still need to fix kernels to new function names, needs to be done manually - Temperature Management needs to be rewritten partially because of conflicting datatypes names - Added code to create different codepaths for NV on AMD in runtime in host (see data.vendor_id) - Added code to create different codepaths for NV on AMD in runtime in kernels (see IS_NV and IS_AMD) - First tests working for -m 0, for example - Great performance increases in general for NV so far - Tested amp_* and markov_* kernel - Migrated special NV optimizations for rule processor
2015-12-15 11:04:22 +00:00
u32 w0[4];
Initial commit
2015-12-04 14:47:52 +00:00
w0[0] = pws[gid].i[ 0];
w0[1] = pws[gid].i[ 1];
w0[2] = pws[gid].i[ 2];
w0[3] = pws[gid].i[ 3];
- Dropped all vector code since new GPU's are all scalar, makes the code much easier - Some performance on low-end GPU may drop because of that, but only for a few hash-modes - Dropped scalar code (aka warp) since we do not have any vector datatypes anymore - Renamed C++ overloading functions memcat32_9 -> memcat_c32_w4x4_a3x4 - Still need to fix kernels to new function names, needs to be done manually - Temperature Management needs to be rewritten partially because of conflicting datatypes names - Added code to create different codepaths for NV on AMD in runtime in host (see data.vendor_id) - Added code to create different codepaths for NV on AMD in runtime in kernels (see IS_NV and IS_AMD) - First tests working for -m 0, for example - Great performance increases in general for NV so far - Tested amp_* and markov_* kernel - Migrated special NV optimizations for rule processor
2015-12-15 11:04:22 +00:00
u32 w1[4];
Initial commit
2015-12-04 14:47:52 +00:00
w1[0] = pws[gid].i[ 4];
w1[1] = pws[gid].i[ 5];
w1[2] = pws[gid].i[ 6];
w1[3] = pws[gid].i[ 7];
- Dropped all vector code since new GPU's are all scalar, makes the code much easier - Some performance on low-end GPU may drop because of that, but only for a few hash-modes - Dropped scalar code (aka warp) since we do not have any vector datatypes anymore - Renamed C++ overloading functions memcat32_9 -> memcat_c32_w4x4_a3x4 - Still need to fix kernels to new function names, needs to be done manually - Temperature Management needs to be rewritten partially because of conflicting datatypes names - Added code to create different codepaths for NV on AMD in runtime in host (see data.vendor_id) - Added code to create different codepaths for NV on AMD in runtime in kernels (see IS_NV and IS_AMD) - First tests working for -m 0, for example - Great performance increases in general for NV so far - Tested amp_* and markov_* kernel - Migrated special NV optimizations for rule processor
2015-12-15 11:04:22 +00:00
u32 w2[4];
Initial commit
2015-12-04 14:47:52 +00:00
w2[0] = 0;
w2[1] = 0;
w2[2] = 0;
w2[3] = 0;
- Dropped all vector code since new GPU's are all scalar, makes the code much easier - Some performance on low-end GPU may drop because of that, but only for a few hash-modes - Dropped scalar code (aka warp) since we do not have any vector datatypes anymore - Renamed C++ overloading functions memcat32_9 -> memcat_c32_w4x4_a3x4 - Still need to fix kernels to new function names, needs to be done manually - Temperature Management needs to be rewritten partially because of conflicting datatypes names - Added code to create different codepaths for NV on AMD in runtime in host (see data.vendor_id) - Added code to create different codepaths for NV on AMD in runtime in kernels (see IS_NV and IS_AMD) - First tests working for -m 0, for example - Great performance increases in general for NV so far - Tested amp_* and markov_* kernel - Migrated special NV optimizations for rule processor
2015-12-15 11:04:22 +00:00
u32 w3[4];
Initial commit
2015-12-04 14:47:52 +00:00
w3[0] = 0;
w3[1] = 0;
w3[2] = 0;
w3[3] = 0;
const u32 pw_len = pws[gid].pw_len;
/**
* main
*/
m00120s (w0, w1, w2, w3, pw_len, pws, rules_buf, combs_buf, bfs_buf, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_scryptV_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, bfs_cnt, digests_cnt, digests_offset);
}
Upgrade kernel to support dynamic local work sizes
2016-01-19 15:06:03 +00:00
__kernel void m00120_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)
Initial commit
2015-12-04 14:47:52 +00:00
{
/**
* base
*/
const u32 gid = get_global_id (0);
if (gid >= gid_max) return;
- Dropped all vector code since new GPU's are all scalar, makes the code much easier - Some performance on low-end GPU may drop because of that, but only for a few hash-modes - Dropped scalar code (aka warp) since we do not have any vector datatypes anymore - Renamed C++ overloading functions memcat32_9 -> memcat_c32_w4x4_a3x4 - Still need to fix kernels to new function names, needs to be done manually - Temperature Management needs to be rewritten partially because of conflicting datatypes names - Added code to create different codepaths for NV on AMD in runtime in host (see data.vendor_id) - Added code to create different codepaths for NV on AMD in runtime in kernels (see IS_NV and IS_AMD) - First tests working for -m 0, for example - Great performance increases in general for NV so far - Tested amp_* and markov_* kernel - Migrated special NV optimizations for rule processor
2015-12-15 11:04:22 +00:00
u32 w0[4];
Initial commit
2015-12-04 14:47:52 +00:00
w0[0] = pws[gid].i[ 0];
w0[1] = pws[gid].i[ 1];
w0[2] = pws[gid].i[ 2];
w0[3] = pws[gid].i[ 3];
- Dropped all vector code since new GPU's are all scalar, makes the code much easier - Some performance on low-end GPU may drop because of that, but only for a few hash-modes - Dropped scalar code (aka warp) since we do not have any vector datatypes anymore - Renamed C++ overloading functions memcat32_9 -> memcat_c32_w4x4_a3x4 - Still need to fix kernels to new function names, needs to be done manually - Temperature Management needs to be rewritten partially because of conflicting datatypes names - Added code to create different codepaths for NV on AMD in runtime in host (see data.vendor_id) - Added code to create different codepaths for NV on AMD in runtime in kernels (see IS_NV and IS_AMD) - First tests working for -m 0, for example - Great performance increases in general for NV so far - Tested amp_* and markov_* kernel - Migrated special NV optimizations for rule processor
2015-12-15 11:04:22 +00:00
u32 w1[4];
Initial commit
2015-12-04 14:47:52 +00:00
w1[0] = pws[gid].i[ 4];
w1[1] = pws[gid].i[ 5];
w1[2] = pws[gid].i[ 6];
w1[3] = pws[gid].i[ 7];
- Dropped all vector code since new GPU's are all scalar, makes the code much easier - Some performance on low-end GPU may drop because of that, but only for a few hash-modes - Dropped scalar code (aka warp) since we do not have any vector datatypes anymore - Renamed C++ overloading functions memcat32_9 -> memcat_c32_w4x4_a3x4 - Still need to fix kernels to new function names, needs to be done manually - Temperature Management needs to be rewritten partially because of conflicting datatypes names - Added code to create different codepaths for NV on AMD in runtime in host (see data.vendor_id) - Added code to create different codepaths for NV on AMD in runtime in kernels (see IS_NV and IS_AMD) - First tests working for -m 0, for example - Great performance increases in general for NV so far - Tested amp_* and markov_* kernel - Migrated special NV optimizations for rule processor
2015-12-15 11:04:22 +00:00
u32 w2[4];
Initial commit
2015-12-04 14:47:52 +00:00
w2[0] = pws[gid].i[ 8];
w2[1] = pws[gid].i[ 9];
w2[2] = pws[gid].i[10];
w2[3] = pws[gid].i[11];
- Dropped all vector code since new GPU's are all scalar, makes the code much easier - Some performance on low-end GPU may drop because of that, but only for a few hash-modes - Dropped scalar code (aka warp) since we do not have any vector datatypes anymore - Renamed C++ overloading functions memcat32_9 -> memcat_c32_w4x4_a3x4 - Still need to fix kernels to new function names, needs to be done manually - Temperature Management needs to be rewritten partially because of conflicting datatypes names - Added code to create different codepaths for NV on AMD in runtime in host (see data.vendor_id) - Added code to create different codepaths for NV on AMD in runtime in kernels (see IS_NV and IS_AMD) - First tests working for -m 0, for example - Great performance increases in general for NV so far - Tested amp_* and markov_* kernel - Migrated special NV optimizations for rule processor
2015-12-15 11:04:22 +00:00
u32 w3[4];
Initial commit
2015-12-04 14:47:52 +00:00
w3[0] = pws[gid].i[12];
w3[1] = pws[gid].i[13];
w3[2] = 0;
w3[3] = 0;
const u32 pw_len = pws[gid].pw_len;
/**
* main
*/
m00120s (w0, w1, w2, w3, pw_len, pws, rules_buf, combs_buf, bfs_buf, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_scryptV_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, bfs_cnt, digests_cnt, digests_offset);
}
Reference in New Issue Copy Permalink