mirror of https://github.com/hashcat/hashcat.git
The previous hash-mode 5000 covered Keccak-256 only. FIPS changed one padding byte while adopting Keccak as the SHA3 standard, which gives us different digests. Now we have separate kernels for SHA3 and Keccak. - Added hash-mode 17300 = SHA3-224 - Added hash-mode 17400 = SHA3-256 - Added hash-mode 17500 = SHA3-384 - Added hash-mode 17600 = SHA3-512 - Added hash-mode 17700 = Keccak-224 - Added hash-mode 17800 = Keccak-256 - Added hash-mode 17900 = Keccak-384 - Added hash-mode 18000 = Keccak-512 - Removed hash-mode 5000 = SHA-3 (Keccak)pull/1717/head
R. Yushaev
6 years ago
parent
9d7e39590b
commit
5c87720acc
@ -0,0 +1,479 @@
|
||||
/**
|
||||
* Author......: See docs/credits.txt
|
||||
* License.....: MIT
|
||||
*/
|
||||
|
||||
#define NEW_SIMD_CODE
|
||||
|
||||
#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_optimized.h"
|
||||
#include "inc_rp_optimized.cl"
|
||||
#include "inc_simd.cl"
|
||||
|
||||
__constant u64a keccakf_rndc[24] =
|
||||
{
|
||||
0x0000000000000001, 0x0000000000008082, 0x800000000000808a,
|
||||
0x8000000080008000, 0x000000000000808b, 0x0000000080000001,
|
||||
0x8000000080008081, 0x8000000000008009, 0x000000000000008a,
|
||||
0x0000000000000088, 0x0000000080008009, 0x000000008000000a,
|
||||
0x000000008000808b, 0x800000000000008b, 0x8000000000008089,
|
||||
0x8000000000008003, 0x8000000000008002, 0x8000000000000080,
|
||||
0x000000000000800a, 0x800000008000000a, 0x8000000080008081,
|
||||
0x8000000000008080, 0x0000000080000001, 0x8000000080008008
|
||||
};
|
||||
|
||||
#ifndef KECCAK_ROUNDS
|
||||
#define KECCAK_ROUNDS 24
|
||||
#endif
|
||||
|
||||
__kernel void m17300_m04 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 pw_buf0[4];
|
||||
u32 pw_buf1[4];
|
||||
|
||||
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];
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
u32x w0[4] = { 0 };
|
||||
u32x w1[4] = { 0 };
|
||||
u32x w2[4] = { 0 };
|
||||
u32x w3[4] = { 0 };
|
||||
|
||||
const u32x out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
|
||||
|
||||
append_0x06_2x4_VV (w0, w1, out_len);
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0[0]);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = 0;
|
||||
u64x a10 = 0;
|
||||
u64x a11 = 0;
|
||||
u64x a12 = 0;
|
||||
u64x a13 = 0;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0x8000000000000000;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = 0;
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_M_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
__kernel void m17300_m08 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17300_m16 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17300_s04 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 pw_buf0[4];
|
||||
u32 pw_buf1[4];
|
||||
|
||||
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];
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* 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]
|
||||
};
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
u32x w0[4] = { 0 };
|
||||
u32x w1[4] = { 0 };
|
||||
u32x w2[4] = { 0 };
|
||||
u32x w3[4] = { 0 };
|
||||
|
||||
const u32x out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
|
||||
|
||||
append_0x06_2x4_VV (w0, w1, out_len);
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0[0]);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = 0;
|
||||
u64x a10 = 0;
|
||||
u64x a11 = 0;
|
||||
u64x a12 = 0;
|
||||
u64x a13 = 0;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0x8000000000000000;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = 0;
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_S_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
__kernel void m17300_s08 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17300_s16 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
@ -0,0 +1,594 @@
|
||||
/**
|
||||
* Author......: See docs/credits.txt
|
||||
* License.....: MIT
|
||||
*/
|
||||
|
||||
#define NEW_SIMD_CODE
|
||||
|
||||
#include "inc_vendor.cl"
|
||||
#include "inc_hash_constants.h"
|
||||
#include "inc_hash_functions.cl"
|
||||
#include "inc_types.cl"
|
||||
#include "inc_common.cl"
|
||||
#include "inc_simd.cl"
|
||||
|
||||
__constant u64a keccakf_rndc[24] =
|
||||
{
|
||||
0x0000000000000001, 0x0000000000008082, 0x800000000000808a,
|
||||
0x8000000080008000, 0x000000000000808b, 0x0000000080000001,
|
||||
0x8000000080008081, 0x8000000000008009, 0x000000000000008a,
|
||||
0x0000000000000088, 0x0000000080008009, 0x000000008000000a,
|
||||
0x000000008000808b, 0x800000000000008b, 0x8000000000008089,
|
||||
0x8000000000008003, 0x8000000000008002, 0x8000000000000080,
|
||||
0x000000000000800a, 0x800000008000000a, 0x8000000080008081,
|
||||
0x8000000000008080, 0x0000000080000001, 0x8000000080008008
|
||||
};
|
||||
|
||||
#ifndef KECCAK_ROUNDS
|
||||
#define KECCAK_ROUNDS 24
|
||||
#endif
|
||||
|
||||
__kernel void m17300_m04 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 pw_buf0[4];
|
||||
u32 pw_buf1[4];
|
||||
|
||||
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];
|
||||
|
||||
const u32 pw_l_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
const u32x pw_r_len = pwlenx_create_combt (combs_buf, il_pos);
|
||||
|
||||
const u32x pw_len = pw_l_len + pw_r_len;
|
||||
|
||||
/**
|
||||
* concat password candidate
|
||||
*/
|
||||
|
||||
u32x wordl0[4] = { 0 };
|
||||
u32x wordl1[4] = { 0 };
|
||||
u32x wordl2[4] = { 0 };
|
||||
u32x wordl3[4] = { 0 };
|
||||
|
||||
wordl0[0] = pw_buf0[0];
|
||||
wordl0[1] = pw_buf0[1];
|
||||
wordl0[2] = pw_buf0[2];
|
||||
wordl0[3] = pw_buf0[3];
|
||||
wordl1[0] = pw_buf1[0];
|
||||
wordl1[1] = pw_buf1[1];
|
||||
wordl1[2] = pw_buf1[2];
|
||||
wordl1[3] = pw_buf1[3];
|
||||
|
||||
u32x wordr0[4] = { 0 };
|
||||
u32x wordr1[4] = { 0 };
|
||||
u32x wordr2[4] = { 0 };
|
||||
u32x wordr3[4] = { 0 };
|
||||
|
||||
wordr0[0] = ix_create_combt (combs_buf, il_pos, 0);
|
||||
wordr0[1] = ix_create_combt (combs_buf, il_pos, 1);
|
||||
wordr0[2] = ix_create_combt (combs_buf, il_pos, 2);
|
||||
wordr0[3] = ix_create_combt (combs_buf, il_pos, 3);
|
||||
wordr1[0] = ix_create_combt (combs_buf, il_pos, 4);
|
||||
wordr1[1] = ix_create_combt (combs_buf, il_pos, 5);
|
||||
wordr1[2] = ix_create_combt (combs_buf, il_pos, 6);
|
||||
wordr1[3] = ix_create_combt (combs_buf, il_pos, 7);
|
||||
|
||||
if (combs_mode == COMBINATOR_MODE_BASE_LEFT)
|
||||
{
|
||||
switch_buffer_by_offset_le_VV (wordr0, wordr1, wordr2, wordr3, pw_l_len);
|
||||
}
|
||||
else
|
||||
{
|
||||
switch_buffer_by_offset_le_VV (wordl0, wordl1, wordl2, wordl3, pw_r_len);
|
||||
}
|
||||
|
||||
u32x w0[4];
|
||||
u32x w1[4];
|
||||
u32x w2[4];
|
||||
u32x w3[4];
|
||||
|
||||
w0[0] = wordl0[0] | wordr0[0];
|
||||
w0[1] = wordl0[1] | wordr0[1];
|
||||
w0[2] = wordl0[2] | wordr0[2];
|
||||
w0[3] = wordl0[3] | wordr0[3];
|
||||
w1[0] = wordl1[0] | wordr1[0];
|
||||
w1[1] = wordl1[1] | wordr1[1];
|
||||
w1[2] = wordl1[2] | wordr1[2];
|
||||
w1[3] = wordl1[3] | wordr1[3];
|
||||
w2[0] = wordl2[0] | wordr2[0];
|
||||
w2[1] = wordl2[1] | wordr2[1];
|
||||
w2[2] = wordl2[2] | wordr2[2];
|
||||
w2[3] = wordl2[3] | wordr2[3];
|
||||
w3[0] = wordl3[0] | wordr3[0];
|
||||
w3[1] = wordl3[1] | wordr3[1];
|
||||
w3[2] = wordl3[2] | wordr3[2];
|
||||
w3[3] = wordl3[3] | wordr3[3];
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0[0]);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = hl32_to_64 (w2[1], w2[0]);
|
||||
u64x a10 = hl32_to_64 (w2[3], w2[2]);
|
||||
u64x a11 = hl32_to_64 (w3[1], w3[0]);
|
||||
u64x a12 = hl32_to_64 (w3[3], w3[2]);
|
||||
u64x a13 = 0;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0x8000000000000000;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = 0;
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_M_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
__kernel void m17300_m08 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17300_m16 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17300_s04 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 pw_buf0[4];
|
||||
u32 pw_buf1[4];
|
||||
|
||||
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];
|
||||
|
||||
const u32 pw_l_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* 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]
|
||||
};
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
const u32x pw_r_len = pwlenx_create_combt (combs_buf, il_pos);
|
||||
|
||||
const u32x pw_len = pw_l_len + pw_r_len;
|
||||
|
||||
/**
|
||||
* concat password candidate
|
||||
*/
|
||||
|
||||
u32x wordl0[4] = { 0 };
|
||||
u32x wordl1[4] = { 0 };
|
||||
u32x wordl2[4] = { 0 };
|
||||
u32x wordl3[4] = { 0 };
|
||||
|
||||
wordl0[0] = pw_buf0[0];
|
||||
wordl0[1] = pw_buf0[1];
|
||||
wordl0[2] = pw_buf0[2];
|
||||
wordl0[3] = pw_buf0[3];
|
||||
wordl1[0] = pw_buf1[0];
|
||||
wordl1[1] = pw_buf1[1];
|
||||
wordl1[2] = pw_buf1[2];
|
||||
wordl1[3] = pw_buf1[3];
|
||||
|
||||
u32x wordr0[4] = { 0 };
|
||||
u32x wordr1[4] = { 0 };
|
||||
u32x wordr2[4] = { 0 };
|
||||
u32x wordr3[4] = { 0 };
|
||||
|
||||
wordr0[0] = ix_create_combt (combs_buf, il_pos, 0);
|
||||
wordr0[1] = ix_create_combt (combs_buf, il_pos, 1);
|
||||
wordr0[2] = ix_create_combt (combs_buf, il_pos, 2);
|
||||
wordr0[3] = ix_create_combt (combs_buf, il_pos, 3);
|
||||
wordr1[0] = ix_create_combt (combs_buf, il_pos, 4);
|
||||
wordr1[1] = ix_create_combt (combs_buf, il_pos, 5);
|
||||
wordr1[2] = ix_create_combt (combs_buf, il_pos, 6);
|
||||
wordr1[3] = ix_create_combt (combs_buf, il_pos, 7);
|
||||
|
||||
if (combs_mode == COMBINATOR_MODE_BASE_LEFT)
|
||||
{
|
||||
switch_buffer_by_offset_le_VV (wordr0, wordr1, wordr2, wordr3, pw_l_len);
|
||||
}
|
||||
else
|
||||
{
|
||||
switch_buffer_by_offset_le_VV (wordl0, wordl1, wordl2, wordl3, pw_r_len);
|
||||
}
|
||||
|
||||
u32x w0[4];
|
||||
u32x w1[4];
|
||||
u32x w2[4];
|
||||
u32x w3[4];
|
||||
|
||||
w0[0] = wordl0[0] | wordr0[0];
|
||||
w0[1] = wordl0[1] | wordr0[1];
|
||||
w0[2] = wordl0[2] | wordr0[2];
|
||||
w0[3] = wordl0[3] | wordr0[3];
|
||||
w1[0] = wordl1[0] | wordr1[0];
|
||||
w1[1] = wordl1[1] | wordr1[1];
|
||||
w1[2] = wordl1[2] | wordr1[2];
|
||||
w1[3] = wordl1[3] | wordr1[3];
|
||||
w2[0] = wordl2[0] | wordr2[0];
|
||||
w2[1] = wordl2[1] | wordr2[1];
|
||||
w2[2] = wordl2[2] | wordr2[2];
|
||||
w2[3] = wordl2[3] | wordr2[3];
|
||||
w3[0] = wordl3[0] | wordr3[0];
|
||||
w3[1] = wordl3[1] | wordr3[1];
|
||||
w3[2] = wordl3[2] | wordr3[2];
|
||||
w3[3] = wordl3[3] | wordr3[3];
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0[0]);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = hl32_to_64 (w2[1], w2[0]);
|
||||
u64x a10 = hl32_to_64 (w2[3], w2[2]);
|
||||
u64x a11 = hl32_to_64 (w3[1], w3[0]);
|
||||
u64x a12 = hl32_to_64 (w3[3], w3[2]);
|
||||
u64x a13 = 0;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0x8000000000000000;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = 0;
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_S_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
__kernel void m17300_s08 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17300_s16 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
@ -0,0 +1,695 @@
|
||||
/**
|
||||
* Author......: See docs/credits.txt
|
||||
* License.....: MIT
|
||||
*/
|
||||
|
||||
#define NEW_SIMD_CODE
|
||||
|
||||
#include "inc_vendor.cl"
|
||||
#include "inc_hash_constants.h"
|
||||
#include "inc_hash_functions.cl"
|
||||
#include "inc_types.cl"
|
||||
#include "inc_common.cl"
|
||||
#include "inc_simd.cl"
|
||||
|
||||
__constant u64a keccakf_rndc[24] =
|
||||
{
|
||||
0x0000000000000001, 0x0000000000008082, 0x800000000000808a,
|
||||
0x8000000080008000, 0x000000000000808b, 0x0000000080000001,
|
||||
0x8000000080008081, 0x8000000000008009, 0x000000000000008a,
|
||||
0x0000000000000088, 0x0000000080008009, 0x000000008000000a,
|
||||
0x000000008000808b, 0x800000000000008b, 0x8000000000008089,
|
||||
0x8000000000008003, 0x8000000000008002, 0x8000000000000080,
|
||||
0x000000000000800a, 0x800000008000000a, 0x8000000080008081,
|
||||
0x8000000000008080, 0x0000000080000001, 0x8000000080008008
|
||||
};
|
||||
|
||||
#ifndef KECCAK_ROUNDS
|
||||
#define KECCAK_ROUNDS 24
|
||||
#endif
|
||||
|
||||
DECLSPEC void m17300m (u32 *w0, u32 *w1, u32 *w2, u32 *w3, const u32 pw_len, __global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
u32 w0l = w0[0];
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
const u32x w0r = ix_create_bft (bfs_buf, il_pos);
|
||||
|
||||
const u32x w0lr = w0l | w0r;
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0lr);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = hl32_to_64 (w2[1], w2[0]);
|
||||
u64x a10 = hl32_to_64 (w2[3], w2[2]);
|
||||
u64x a11 = hl32_to_64 (w3[1], w3[0]);
|
||||
u64x a12 = hl32_to_64 (w3[3], w3[2]);
|
||||
u64x a13 = 0;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0x8000000000000000;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = 0;
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_M_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
DECLSPEC void m17300s (u32 *w0, u32 *w1, u32 *w2, u32 *w3, const u32 pw_len, __global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
const u64 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]
|
||||
};
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
u32 w0l = w0[0];
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
const u32x w0r = ix_create_bft (bfs_buf, il_pos);
|
||||
|
||||
const u32x w0lr = w0l | w0r;
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0lr);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = hl32_to_64 (w2[1], w2[0]);
|
||||
u64x a10 = hl32_to_64 (w2[3], w2[2]);
|
||||
u64x a11 = hl32_to_64 (w3[1], w3[0]);
|
||||
u64x a12 = hl32_to_64 (w3[3], w3[2]);
|
||||
u64x a13 = 0;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0x8000000000000000;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = 0;
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_S_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
__kernel void m17300_m04 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
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;
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* main
|
||||
*/
|
||||
|
||||
m17300m (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m17300_m08 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
u32 w1[4];
|
||||
|
||||
w1[0] = pws[gid].i[ 4];
|
||||
w1[1] = pws[gid].i[ 5];
|
||||
w1[2] = pws[gid].i[ 6];
|
||||
w1[3] = pws[gid].i[ 7];
|
||||
|
||||
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;
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* main
|
||||
*/
|
||||
|
||||
m17300m (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m17300_m16 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
u32 w1[4];
|
||||
|
||||
w1[0] = pws[gid].i[ 4];
|
||||
w1[1] = pws[gid].i[ 5];
|
||||
w1[2] = pws[gid].i[ 6];
|
||||
w1[3] = pws[gid].i[ 7];
|
||||
|
||||
u32 w2[4];
|
||||
|
||||
w2[0] = pws[gid].i[ 8];
|
||||
w2[1] = pws[gid].i[ 9];
|
||||
w2[2] = pws[gid].i[10];
|
||||
w2[3] = pws[gid].i[11];
|
||||
|
||||
u32 w3[4];
|
||||
|
||||
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
|
||||
*/
|
||||
|
||||
m17300m (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m17300_s04 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
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;
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* main
|
||||
*/
|
||||
|
||||
m17300s (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m17300_s08 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
u32 w1[4];
|
||||
|
||||
w1[0] = pws[gid].i[ 4];
|
||||
w1[1] = pws[gid].i[ 5];
|
||||
w1[2] = pws[gid].i[ 6];
|
||||
w1[3] = pws[gid].i[ 7];
|
||||
|
||||
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;
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* main
|
||||
*/
|
||||
|
||||
m17300s (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m17300_s16 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
u32 w1[4];
|
||||
|
||||
w1[0] = pws[gid].i[ 4];
|
||||
w1[1] = pws[gid].i[ 5];
|
||||
w1[2] = pws[gid].i[ 6];
|
||||
w1[3] = pws[gid].i[ 7];
|
||||
|
||||
u32 w2[4];
|
||||
|
||||
w2[0] = pws[gid].i[ 8];
|
||||
w2[1] = pws[gid].i[ 9];
|
||||
w2[2] = pws[gid].i[10];
|
||||
w2[3] = pws[gid].i[11];
|
||||
|
||||
u32 w3[4];
|
||||
|
||||
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
|
||||
*/
|
||||
|
||||
m17300s (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
@ -0,0 +1,479 @@
|
||||
/**
|
||||
* Author......: See docs/credits.txt
|
||||
* License.....: MIT
|
||||
*/
|
||||
|
||||
#define NEW_SIMD_CODE
|
||||
|
||||
#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_optimized.h"
|
||||
#include "inc_rp_optimized.cl"
|
||||
#include "inc_simd.cl"
|
||||
|
||||
__constant u64a keccakf_rndc[24] =
|
||||
{
|
||||
0x0000000000000001, 0x0000000000008082, 0x800000000000808a,
|
||||
0x8000000080008000, 0x000000000000808b, 0x0000000080000001,
|
||||
0x8000000080008081, 0x8000000000008009, 0x000000000000008a,
|
||||
0x0000000000000088, 0x0000000080008009, 0x000000008000000a,
|
||||
0x000000008000808b, 0x800000000000008b, 0x8000000000008089,
|
||||
0x8000000000008003, 0x8000000000008002, 0x8000000000000080,
|
||||
0x000000000000800a, 0x800000008000000a, 0x8000000080008081,
|
||||
0x8000000000008080, 0x0000000080000001, 0x8000000080008008
|
||||
};
|
||||
|
||||
#ifndef KECCAK_ROUNDS
|
||||
#define KECCAK_ROUNDS 24
|
||||
#endif
|
||||
|
||||
__kernel void m17400_m04 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 pw_buf0[4];
|
||||
u32 pw_buf1[4];
|
||||
|
||||
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];
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
u32x w0[4] = { 0 };
|
||||
u32x w1[4] = { 0 };
|
||||
u32x w2[4] = { 0 };
|
||||
u32x w3[4] = { 0 };
|
||||
|
||||
const u32x out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
|
||||
|
||||
append_0x06_2x4_VV (w0, w1, out_len);
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0[0]);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = 0;
|
||||
u64x a10 = 0;
|
||||
u64x a11 = 0;
|
||||
u64x a12 = 0;
|
||||
u64x a13 = 0;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0x8000000000000000;
|
||||
u64x a32 = 0;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = h32_from_64 (a03);
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_M_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
__kernel void m17400_m08 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17400_m16 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17400_s04 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 pw_buf0[4];
|
||||
u32 pw_buf1[4];
|
||||
|
||||
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];
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* 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]
|
||||
};
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
u32x w0[4] = { 0 };
|
||||
u32x w1[4] = { 0 };
|
||||
u32x w2[4] = { 0 };
|
||||
u32x w3[4] = { 0 };
|
||||
|
||||
const u32x out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
|
||||
|
||||
append_0x06_2x4_VV (w0, w1, out_len);
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0[0]);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = 0;
|
||||
u64x a10 = 0;
|
||||
u64x a11 = 0;
|
||||
u64x a12 = 0;
|
||||
u64x a13 = 0;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0x8000000000000000;
|
||||
u64x a32 = 0;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = h32_from_64 (a03);
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_S_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
__kernel void m17400_s08 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17400_s16 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
@ -0,0 +1,696 @@
|
||||
/**
|
||||
* Author......: See docs/credits.txt
|
||||
* License.....: MIT
|
||||
*/
|
||||
|
||||
#define NEW_SIMD_CODE
|
||||
|
||||
#include "inc_vendor.cl"
|
||||
#include "inc_hash_constants.h"
|
||||
#include "inc_hash_functions.cl"
|
||||
#include "inc_types.cl"
|
||||
#include "inc_common.cl"
|
||||
#include "inc_simd.cl"
|
||||
|
||||
__constant u64a keccakf_rndc[24] =
|
||||
{
|
||||
0x0000000000000001, 0x0000000000008082, 0x800000000000808a,
|
||||
0x8000000080008000, 0x000000000000808b, 0x0000000080000001,
|
||||
0x8000000080008081, 0x8000000000008009, 0x000000000000008a,
|
||||
0x0000000000000088, 0x0000000080008009, 0x000000008000000a,
|
||||
0x000000008000808b, 0x800000000000008b, 0x8000000000008089,
|
||||
0x8000000000008003, 0x8000000000008002, 0x8000000000000080,
|
||||
0x000000000000800a, 0x800000008000000a, 0x8000000080008081,
|
||||
0x8000000000008080, 0x0000000080000001, 0x8000000080008008
|
||||
};
|
||||
|
||||
#ifndef KECCAK_ROUNDS
|
||||
#define KECCAK_ROUNDS 24
|
||||
#endif
|
||||
|
||||
DECLSPEC void m17400m (u32 *w0, u32 *w1, u32 *w2, u32 *w3, const u32 pw_len, __global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
u32 w0l = w0[0];
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
const u32x w0r = ix_create_bft (bfs_buf, il_pos);
|
||||
|
||||
const u32x w0lr = w0l | w0r;
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0lr);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = hl32_to_64 (w2[1], w2[0]);
|
||||
u64x a10 = hl32_to_64 (w2[3], w2[2]);
|
||||
u64x a11 = hl32_to_64 (w3[1], w3[0]);
|
||||
u64x a12 = hl32_to_64 (w3[3], w3[2]);
|
||||
u64x a13 = 0;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0x8000000000000000;
|
||||
u64x a32 = 0;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = h32_from_64 (a03);
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_M_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
DECLSPEC void m17400s (u32 *w0, u32 *w1, u32 *w2, u32 *w3, const u32 pw_len, __global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
const u64 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]
|
||||
};
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
u32 w0l = w0[0];
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
const u32x w0r = ix_create_bft (bfs_buf, il_pos);
|
||||
|
||||
const u32x w0lr = w0l | w0r;
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0lr);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = hl32_to_64 (w2[1], w2[0]);
|
||||
u64x a10 = hl32_to_64 (w2[3], w2[2]);
|
||||
u64x a11 = hl32_to_64 (w3[1], w3[0]);
|
||||
u64x a12 = hl32_to_64 (w3[3], w3[2]);
|
||||
u64x a13 = 0;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0x8000000000000000;
|
||||
u64x a32 = 0;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = h32_from_64 (a03);
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_S_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
__kernel void m17400_m04 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
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;
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* main
|
||||
*/
|
||||
|
||||
m17400m (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m17400_m08 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
u32 w1[4];
|
||||
|
||||
w1[0] = pws[gid].i[ 4];
|
||||
w1[1] = pws[gid].i[ 5];
|
||||
w1[2] = pws[gid].i[ 6];
|
||||
w1[3] = pws[gid].i[ 7];
|
||||
|
||||
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;
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* main
|
||||
*/
|
||||
|
||||
m17400m (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m17400_m16 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
u32 w1[4];
|
||||
|
||||
w1[0] = pws[gid].i[ 4];
|
||||
w1[1] = pws[gid].i[ 5];
|
||||
w1[2] = pws[gid].i[ 6];
|
||||
w1[3] = pws[gid].i[ 7];
|
||||
|
||||
u32 w2[4];
|
||||
|
||||
w2[0] = pws[gid].i[ 8];
|
||||
w2[1] = pws[gid].i[ 9];
|
||||
w2[2] = pws[gid].i[10];
|
||||
w2[3] = pws[gid].i[11];
|
||||
|
||||
u32 w3[4];
|
||||
|
||||
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
|
||||
*/
|
||||
|
||||
m17400m (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m17400_s04 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
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;
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* main
|
||||
*/
|
||||
|
||||
m17400s (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m17400_s08 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
u32 w1[4];
|
||||
|
||||
w1[0] = pws[gid].i[ 4];
|
||||
w1[1] = pws[gid].i[ 5];
|
||||
w1[2] = pws[gid].i[ 6];
|
||||
w1[3] = pws[gid].i[ 7];
|
||||
|
||||
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;
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* main
|
||||
*/
|
||||
|
||||
m17400s (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m17400_s16 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
u32 w1[4];
|
||||
|
||||
w1[0] = pws[gid].i[ 4];
|
||||
w1[1] = pws[gid].i[ 5];
|
||||
w1[2] = pws[gid].i[ 6];
|
||||
w1[3] = pws[gid].i[ 7];
|
||||
|
||||
u32 w2[4];
|
||||
|
||||
w2[0] = pws[gid].i[ 8];
|
||||
w2[1] = pws[gid].i[ 9];
|
||||
w2[2] = pws[gid].i[10];
|
||||
w2[3] = pws[gid].i[11];
|
||||
|
||||
u32 w3[4];
|
||||
|
||||
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
|
||||
*/
|
||||
|
||||
m17400s (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
@ -0,0 +1,479 @@
|
||||
/**
|
||||
* Author......: See docs/credits.txt
|
||||
* License.....: MIT
|
||||
*/
|
||||
|
||||
#define NEW_SIMD_CODE
|
||||
|
||||
#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_optimized.h"
|
||||
#include "inc_rp_optimized.cl"
|
||||
#include "inc_simd.cl"
|
||||
|
||||
__constant u64a keccakf_rndc[24] =
|
||||
{
|
||||
0x0000000000000001, 0x0000000000008082, 0x800000000000808a,
|
||||
0x8000000080008000, 0x000000000000808b, 0x0000000080000001,
|
||||
0x8000000080008081, 0x8000000000008009, 0x000000000000008a,
|
||||
0x0000000000000088, 0x0000000080008009, 0x000000008000000a,
|
||||
0x000000008000808b, 0x800000000000008b, 0x8000000000008089,
|
||||
0x8000000000008003, 0x8000000000008002, 0x8000000000000080,
|
||||
0x000000000000800a, 0x800000008000000a, 0x8000000080008081,
|
||||
0x8000000000008080, 0x0000000080000001, 0x8000000080008008
|
||||
};
|
||||
|
||||
#ifndef KECCAK_ROUNDS
|
||||
#define KECCAK_ROUNDS 24
|
||||
#endif
|
||||
|
||||
__kernel void m17500_m04 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 pw_buf0[4];
|
||||
u32 pw_buf1[4];
|
||||
|
||||
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];
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
u32x w0[4] = { 0 };
|
||||
u32x w1[4] = { 0 };
|
||||
u32x w2[4] = { 0 };
|
||||
u32x w3[4] = { 0 };
|
||||
|
||||
const u32x out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
|
||||
|
||||
append_0x06_2x4_VV (w0, w1, out_len);
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0[0]);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = 0;
|
||||
u64x a10 = 0;
|
||||
u64x a11 = 0;
|
||||
u64x a12 = 0;
|
||||
u64x a13 = 0;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0x8000000000000000;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = h32_from_64 (a03);
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_M_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
__kernel void m17500_m08 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17500_m16 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17500_s04 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 pw_buf0[4];
|
||||
u32 pw_buf1[4];
|
||||
|
||||
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];
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* 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]
|
||||
};
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
u32x w0[4] = { 0 };
|
||||
u32x w1[4] = { 0 };
|
||||
u32x w2[4] = { 0 };
|
||||
u32x w3[4] = { 0 };
|
||||
|
||||
const u32x out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
|
||||
|
||||
append_0x06_2x4_VV (w0, w1, out_len);
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0[0]);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = 0;
|
||||
u64x a10 = 0;
|
||||
u64x a11 = 0;
|
||||
u64x a12 = 0;
|
||||
u64x a13 = 0;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0x8000000000000000;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = h32_from_64 (a03);
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_S_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
__kernel void m17500_s08 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17500_s16 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
@ -0,0 +1,594 @@
|
||||
/**
|
||||
* Author......: See docs/credits.txt
|
||||
* License.....: MIT
|
||||
*/
|
||||
|
||||
#define NEW_SIMD_CODE
|
||||
|
||||
#include "inc_vendor.cl"
|
||||
#include "inc_hash_constants.h"
|
||||
#include "inc_hash_functions.cl"
|
||||
#include "inc_types.cl"
|
||||
#include "inc_common.cl"
|
||||
#include "inc_simd.cl"
|
||||
|
||||
__constant u64a keccakf_rndc[24] =
|
||||
{
|
||||
0x0000000000000001, 0x0000000000008082, 0x800000000000808a,
|
||||
0x8000000080008000, 0x000000000000808b, 0x0000000080000001,
|
||||
0x8000000080008081, 0x8000000000008009, 0x000000000000008a,
|
||||
0x0000000000000088, 0x0000000080008009, 0x000000008000000a,
|
||||
0x000000008000808b, 0x800000000000008b, 0x8000000000008089,
|
||||
0x8000000000008003, 0x8000000000008002, 0x8000000000000080,
|
||||
0x000000000000800a, 0x800000008000000a, 0x8000000080008081,
|
||||
0x8000000000008080, 0x0000000080000001, 0x8000000080008008
|
||||
};
|
||||
|
||||
#ifndef KECCAK_ROUNDS
|
||||
#define KECCAK_ROUNDS 24
|
||||
#endif
|
||||
|
||||
__kernel void m17500_m04 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 pw_buf0[4];
|
||||
u32 pw_buf1[4];
|
||||
|
||||
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];
|
||||
|
||||
const u32 pw_l_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
const u32x pw_r_len = pwlenx_create_combt (combs_buf, il_pos);
|
||||
|
||||
const u32x pw_len = pw_l_len + pw_r_len;
|
||||
|
||||
/**
|
||||
* concat password candidate
|
||||
*/
|
||||
|
||||
u32x wordl0[4] = { 0 };
|
||||
u32x wordl1[4] = { 0 };
|
||||
u32x wordl2[4] = { 0 };
|
||||
u32x wordl3[4] = { 0 };
|
||||
|
||||
wordl0[0] = pw_buf0[0];
|
||||
wordl0[1] = pw_buf0[1];
|
||||
wordl0[2] = pw_buf0[2];
|
||||
wordl0[3] = pw_buf0[3];
|
||||
wordl1[0] = pw_buf1[0];
|
||||
wordl1[1] = pw_buf1[1];
|
||||
wordl1[2] = pw_buf1[2];
|
||||
wordl1[3] = pw_buf1[3];
|
||||
|
||||
u32x wordr0[4] = { 0 };
|
||||
u32x wordr1[4] = { 0 };
|
||||
u32x wordr2[4] = { 0 };
|
||||
u32x wordr3[4] = { 0 };
|
||||
|
||||
wordr0[0] = ix_create_combt (combs_buf, il_pos, 0);
|
||||
wordr0[1] = ix_create_combt (combs_buf, il_pos, 1);
|
||||
wordr0[2] = ix_create_combt (combs_buf, il_pos, 2);
|
||||
wordr0[3] = ix_create_combt (combs_buf, il_pos, 3);
|
||||
wordr1[0] = ix_create_combt (combs_buf, il_pos, 4);
|
||||
wordr1[1] = ix_create_combt (combs_buf, il_pos, 5);
|
||||
wordr1[2] = ix_create_combt (combs_buf, il_pos, 6);
|
||||
wordr1[3] = ix_create_combt (combs_buf, il_pos, 7);
|
||||
|
||||
if (combs_mode == COMBINATOR_MODE_BASE_LEFT)
|
||||
{
|
||||
switch_buffer_by_offset_le_VV (wordr0, wordr1, wordr2, wordr3, pw_l_len);
|
||||
}
|
||||
else
|
||||
{
|
||||
switch_buffer_by_offset_le_VV (wordl0, wordl1, wordl2, wordl3, pw_r_len);
|
||||
}
|
||||
|
||||
u32x w0[4];
|
||||
u32x w1[4];
|
||||
u32x w2[4];
|
||||
u32x w3[4];
|
||||
|
||||
w0[0] = wordl0[0] | wordr0[0];
|
||||
w0[1] = wordl0[1] | wordr0[1];
|
||||
w0[2] = wordl0[2] | wordr0[2];
|
||||
w0[3] = wordl0[3] | wordr0[3];
|
||||
w1[0] = wordl1[0] | wordr1[0];
|
||||
w1[1] = wordl1[1] | wordr1[1];
|
||||
w1[2] = wordl1[2] | wordr1[2];
|
||||
w1[3] = wordl1[3] | wordr1[3];
|
||||
w2[0] = wordl2[0] | wordr2[0];
|
||||
w2[1] = wordl2[1] | wordr2[1];
|
||||
w2[2] = wordl2[2] | wordr2[2];
|
||||
w2[3] = wordl2[3] | wordr2[3];
|
||||
w3[0] = wordl3[0] | wordr3[0];
|
||||
w3[1] = wordl3[1] | wordr3[1];
|
||||
w3[2] = wordl3[2] | wordr3[2];
|
||||
w3[3] = wordl3[3] | wordr3[3];
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0[0]);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = hl32_to_64 (w2[1], w2[0]);
|
||||
u64x a10 = hl32_to_64 (w2[3], w2[2]);
|
||||
u64x a11 = hl32_to_64 (w3[1], w3[0]);
|
||||
u64x a12 = hl32_to_64 (w3[3], w3[2]);
|
||||
u64x a13 = 0;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0x8000000000000000;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = h32_from_64 (a03);
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_M_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
__kernel void m17500_m08 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17500_m16 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17500_s04 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 pw_buf0[4];
|
||||
u32 pw_buf1[4];
|
||||
|
||||
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];
|
||||
|
||||
const u32 pw_l_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* 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]
|
||||
};
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
const u32x pw_r_len = pwlenx_create_combt (combs_buf, il_pos);
|
||||
|
||||
const u32x pw_len = pw_l_len + pw_r_len;
|
||||
|
||||
/**
|
||||
* concat password candidate
|
||||
*/
|
||||
|
||||
u32x wordl0[4] = { 0 };
|
||||
u32x wordl1[4] = { 0 };
|
||||
u32x wordl2[4] = { 0 };
|
||||
u32x wordl3[4] = { 0 };
|
||||
|
||||
wordl0[0] = pw_buf0[0];
|
||||
wordl0[1] = pw_buf0[1];
|
||||
wordl0[2] = pw_buf0[2];
|
||||
wordl0[3] = pw_buf0[3];
|
||||
wordl1[0] = pw_buf1[0];
|
||||
wordl1[1] = pw_buf1[1];
|
||||
wordl1[2] = pw_buf1[2];
|
||||
wordl1[3] = pw_buf1[3];
|
||||
|
||||
u32x wordr0[4] = { 0 };
|
||||
u32x wordr1[4] = { 0 };
|
||||
u32x wordr2[4] = { 0 };
|
||||
u32x wordr3[4] = { 0 };
|
||||
|
||||
wordr0[0] = ix_create_combt (combs_buf, il_pos, 0);
|
||||
wordr0[1] = ix_create_combt (combs_buf, il_pos, 1);
|
||||
wordr0[2] = ix_create_combt (combs_buf, il_pos, 2);
|
||||
wordr0[3] = ix_create_combt (combs_buf, il_pos, 3);
|
||||
wordr1[0] = ix_create_combt (combs_buf, il_pos, 4);
|
||||
wordr1[1] = ix_create_combt (combs_buf, il_pos, 5);
|
||||
wordr1[2] = ix_create_combt (combs_buf, il_pos, 6);
|
||||
wordr1[3] = ix_create_combt (combs_buf, il_pos, 7);
|
||||
|
||||
if (combs_mode == COMBINATOR_MODE_BASE_LEFT)
|
||||
{
|
||||
switch_buffer_by_offset_le_VV (wordr0, wordr1, wordr2, wordr3, pw_l_len);
|
||||
}
|
||||
else
|
||||
{
|
||||
switch_buffer_by_offset_le_VV (wordl0, wordl1, wordl2, wordl3, pw_r_len);
|
||||
}
|
||||
|
||||
u32x w0[4];
|
||||
u32x w1[4];
|
||||
u32x w2[4];
|
||||
u32x w3[4];
|
||||
|
||||
w0[0] = wordl0[0] | wordr0[0];
|
||||
w0[1] = wordl0[1] | wordr0[1];
|
||||
w0[2] = wordl0[2] | wordr0[2];
|
||||
w0[3] = wordl0[3] | wordr0[3];
|
||||
w1[0] = wordl1[0] | wordr1[0];
|
||||
w1[1] = wordl1[1] | wordr1[1];
|
||||
w1[2] = wordl1[2] | wordr1[2];
|
||||
w1[3] = wordl1[3] | wordr1[3];
|
||||
w2[0] = wordl2[0] | wordr2[0];
|
||||
w2[1] = wordl2[1] | wordr2[1];
|
||||
w2[2] = wordl2[2] | wordr2[2];
|
||||
w2[3] = wordl2[3] | wordr2[3];
|
||||
w3[0] = wordl3[0] | wordr3[0];
|
||||
w3[1] = wordl3[1] | wordr3[1];
|
||||
w3[2] = wordl3[2] | wordr3[2];
|
||||
w3[3] = wordl3[3] | wordr3[3];
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0[0]);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = hl32_to_64 (w2[1], w2[0]);
|
||||
u64x a10 = hl32_to_64 (w2[3], w2[2]);
|
||||
u64x a11 = hl32_to_64 (w3[1], w3[0]);
|
||||
u64x a12 = hl32_to_64 (w3[3], w3[2]);
|
||||
u64x a13 = 0;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0x8000000000000000;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = h32_from_64 (a03);
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_S_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
__kernel void m17500_s08 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17500_s16 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
@ -0,0 +1,695 @@
|
||||
/**
|
||||
* Author......: See docs/credits.txt
|
||||
* License.....: MIT
|
||||
*/
|
||||
|
||||
#define NEW_SIMD_CODE
|
||||
|
||||
#include "inc_vendor.cl"
|
||||
#include "inc_hash_constants.h"
|
||||
#include "inc_hash_functions.cl"
|
||||
#include "inc_types.cl"
|
||||
#include "inc_common.cl"
|
||||
#include "inc_simd.cl"
|
||||
|
||||
__constant u64a keccakf_rndc[24] =
|
||||
{
|
||||
0x0000000000000001, 0x0000000000008082, 0x800000000000808a,
|
||||
0x8000000080008000, 0x000000000000808b, 0x0000000080000001,
|
||||
0x8000000080008081, 0x8000000000008009, 0x000000000000008a,
|
||||
0x0000000000000088, 0x0000000080008009, 0x000000008000000a,
|
||||
0x000000008000808b, 0x800000000000008b, 0x8000000000008089,
|
||||
0x8000000000008003, 0x8000000000008002, 0x8000000000000080,
|
||||
0x000000000000800a, 0x800000008000000a, 0x8000000080008081,
|
||||
0x8000000000008080, 0x0000000080000001, 0x8000000080008008
|
||||
};
|
||||
|
||||
#ifndef KECCAK_ROUNDS
|
||||
#define KECCAK_ROUNDS 24
|
||||
#endif
|
||||
|
||||
DECLSPEC void m17500m (u32 *w0, u32 *w1, u32 *w2, u32 *w3, const u32 pw_len, __global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
u32 w0l = w0[0];
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
const u32x w0r = ix_create_bft (bfs_buf, il_pos);
|
||||
|
||||
const u32x w0lr = w0l | w0r;
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0lr);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = hl32_to_64 (w2[1], w2[0]);
|
||||
u64x a10 = hl32_to_64 (w2[3], w2[2]);
|
||||
u64x a11 = hl32_to_64 (w3[1], w3[0]);
|
||||
u64x a12 = hl32_to_64 (w3[3], w3[2]);
|
||||
u64x a13 = 0;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0x8000000000000000;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = h32_from_64 (a03);
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_M_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
DECLSPEC void m17500s (u32 *w0, u32 *w1, u32 *w2, u32 *w3, const u32 pw_len, __global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
const u64 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]
|
||||
};
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
u32 w0l = w0[0];
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
const u32x w0r = ix_create_bft (bfs_buf, il_pos);
|
||||
|
||||
const u32x w0lr = w0l | w0r;
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0lr);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = hl32_to_64 (w2[1], w2[0]);
|
||||
u64x a10 = hl32_to_64 (w2[3], w2[2]);
|
||||
u64x a11 = hl32_to_64 (w3[1], w3[0]);
|
||||
u64x a12 = hl32_to_64 (w3[3], w3[2]);
|
||||
u64x a13 = 0;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0x8000000000000000;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = h32_from_64 (a03);
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_S_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
__kernel void m17500_m04 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
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;
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* main
|
||||
*/
|
||||
|
||||
m17500m (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m17500_m08 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
u32 w1[4];
|
||||
|
||||
w1[0] = pws[gid].i[ 4];
|
||||
w1[1] = pws[gid].i[ 5];
|
||||
w1[2] = pws[gid].i[ 6];
|
||||
w1[3] = pws[gid].i[ 7];
|
||||
|
||||
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;
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* main
|
||||
*/
|
||||
|
||||
m17500m (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m17500_m16 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
u32 w1[4];
|
||||
|
||||
w1[0] = pws[gid].i[ 4];
|
||||
w1[1] = pws[gid].i[ 5];
|
||||
w1[2] = pws[gid].i[ 6];
|
||||
w1[3] = pws[gid].i[ 7];
|
||||
|
||||
u32 w2[4];
|
||||
|
||||
w2[0] = pws[gid].i[ 8];
|
||||
w2[1] = pws[gid].i[ 9];
|
||||
w2[2] = pws[gid].i[10];
|
||||
w2[3] = pws[gid].i[11];
|
||||
|
||||
u32 w3[4];
|
||||
|
||||
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
|
||||
*/
|
||||
|
||||
m17500m (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m17500_s04 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
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;
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* main
|
||||
*/
|
||||
|
||||
m17500s (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m17500_s08 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
u32 w1[4];
|
||||
|
||||
w1[0] = pws[gid].i[ 4];
|
||||
w1[1] = pws[gid].i[ 5];
|
||||
w1[2] = pws[gid].i[ 6];
|
||||
w1[3] = pws[gid].i[ 7];
|
||||
|
||||
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;
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* main
|
||||
*/
|
||||
|
||||
m17500s (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m17500_s16 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
u32 w1[4];
|
||||
|
||||
w1[0] = pws[gid].i[ 4];
|
||||
w1[1] = pws[gid].i[ 5];
|
||||
w1[2] = pws[gid].i[ 6];
|
||||
w1[3] = pws[gid].i[ 7];
|
||||
|
||||
u32 w2[4];
|
||||
|
||||
w2[0] = pws[gid].i[ 8];
|
||||
w2[1] = pws[gid].i[ 9];
|
||||
w2[2] = pws[gid].i[10];
|
||||
w2[3] = pws[gid].i[11];
|
||||
|
||||
u32 w3[4];
|
||||
|
||||
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
|
||||
*/
|
||||
|
||||
m17500s (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
@ -0,0 +1,479 @@
|
||||
/**
|
||||
* Author......: See docs/credits.txt
|
||||
* License.....: MIT
|
||||
*/
|
||||
|
||||
#define NEW_SIMD_CODE
|
||||
|
||||
#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_optimized.h"
|
||||
#include "inc_rp_optimized.cl"
|
||||
#include "inc_simd.cl"
|
||||
|
||||
__constant u64a keccakf_rndc[24] =
|
||||
{
|
||||
0x0000000000000001, 0x0000000000008082, 0x800000000000808a,
|
||||
0x8000000080008000, 0x000000000000808b, 0x0000000080000001,
|
||||
0x8000000080008081, 0x8000000000008009, 0x000000000000008a,
|
||||
0x0000000000000088, 0x0000000080008009, 0x000000008000000a,
|
||||
0x000000008000808b, 0x800000000000008b, 0x8000000000008089,
|
||||
0x8000000000008003, 0x8000000000008002, 0x8000000000000080,
|
||||
0x000000000000800a, 0x800000008000000a, 0x8000000080008081,
|
||||
0x8000000000008080, 0x0000000080000001, 0x8000000080008008
|
||||
};
|
||||
|
||||
#ifndef KECCAK_ROUNDS
|
||||
#define KECCAK_ROUNDS 24
|
||||
#endif
|
||||
|
||||
__kernel void m17600_m04 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 pw_buf0[4];
|
||||
u32 pw_buf1[4];
|
||||
|
||||
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];
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
u32x w0[4] = { 0 };
|
||||
u32x w1[4] = { 0 };
|
||||
u32x w2[4] = { 0 };
|
||||
u32x w3[4] = { 0 };
|
||||
|
||||
const u32x out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
|
||||
|
||||
append_0x06_2x4_VV (w0, w1, out_len);
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0[0]);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = 0;
|
||||
u64x a10 = 0;
|
||||
u64x a11 = 0;
|
||||
u64x a12 = 0;
|
||||
u64x a13 = 0x8000000000000000;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = h32_from_64 (a03);
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_M_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
__kernel void m17600_m08 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17600_m16 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17600_s04 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 pw_buf0[4];
|
||||
u32 pw_buf1[4];
|
||||
|
||||
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];
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* 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]
|
||||
};
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
u32x w0[4] = { 0 };
|
||||
u32x w1[4] = { 0 };
|
||||
u32x w2[4] = { 0 };
|
||||
u32x w3[4] = { 0 };
|
||||
|
||||
const u32x out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
|
||||
|
||||
append_0x06_2x4_VV (w0, w1, out_len);
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0[0]);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = 0;
|
||||
u64x a10 = 0;
|
||||
u64x a11 = 0;
|
||||
u64x a12 = 0;
|
||||
u64x a13 = 0x8000000000000000;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = h32_from_64 (a03);
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_S_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
__kernel void m17600_s08 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17600_s16 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
@ -0,0 +1,594 @@
|
||||
/**
|
||||
* Author......: See docs/credits.txt
|
||||
* License.....: MIT
|
||||
*/
|
||||
|
||||
#define NEW_SIMD_CODE
|
||||
|
||||
#include "inc_vendor.cl"
|
||||
#include "inc_hash_constants.h"
|
||||
#include "inc_hash_functions.cl"
|
||||
#include "inc_types.cl"
|
||||
#include "inc_common.cl"
|
||||
#include "inc_simd.cl"
|
||||
|
||||
__constant u64a keccakf_rndc[24] =
|
||||
{
|
||||
0x0000000000000001, 0x0000000000008082, 0x800000000000808a,
|
||||
0x8000000080008000, 0x000000000000808b, 0x0000000080000001,
|
||||
0x8000000080008081, 0x8000000000008009, 0x000000000000008a,
|
||||
0x0000000000000088, 0x0000000080008009, 0x000000008000000a,
|
||||
0x000000008000808b, 0x800000000000008b, 0x8000000000008089,
|
||||
0x8000000000008003, 0x8000000000008002, 0x8000000000000080,
|
||||
0x000000000000800a, 0x800000008000000a, 0x8000000080008081,
|
||||
0x8000000000008080, 0x0000000080000001, 0x8000000080008008
|
||||
};
|
||||
|
||||
#ifndef KECCAK_ROUNDS
|
||||
#define KECCAK_ROUNDS 24
|
||||
#endif
|
||||
|
||||
__kernel void m17600_m04 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 pw_buf0[4];
|
||||
u32 pw_buf1[4];
|
||||
|
||||
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];
|
||||
|
||||
const u32 pw_l_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
const u32x pw_r_len = pwlenx_create_combt (combs_buf, il_pos);
|
||||
|
||||
const u32x pw_len = pw_l_len + pw_r_len;
|
||||
|
||||
/**
|
||||
* concat password candidate
|
||||
*/
|
||||
|
||||
u32x wordl0[4] = { 0 };
|
||||
u32x wordl1[4] = { 0 };
|
||||
u32x wordl2[4] = { 0 };
|
||||
u32x wordl3[4] = { 0 };
|
||||
|
||||
wordl0[0] = pw_buf0[0];
|
||||
wordl0[1] = pw_buf0[1];
|
||||
wordl0[2] = pw_buf0[2];
|
||||
wordl0[3] = pw_buf0[3];
|
||||
wordl1[0] = pw_buf1[0];
|
||||
wordl1[1] = pw_buf1[1];
|
||||
wordl1[2] = pw_buf1[2];
|
||||
wordl1[3] = pw_buf1[3];
|
||||
|
||||
u32x wordr0[4] = { 0 };
|
||||
u32x wordr1[4] = { 0 };
|
||||
u32x wordr2[4] = { 0 };
|
||||
u32x wordr3[4] = { 0 };
|
||||
|
||||
wordr0[0] = ix_create_combt (combs_buf, il_pos, 0);
|
||||
wordr0[1] = ix_create_combt (combs_buf, il_pos, 1);
|
||||
wordr0[2] = ix_create_combt (combs_buf, il_pos, 2);
|
||||
wordr0[3] = ix_create_combt (combs_buf, il_pos, 3);
|
||||
wordr1[0] = ix_create_combt (combs_buf, il_pos, 4);
|
||||
wordr1[1] = ix_create_combt (combs_buf, il_pos, 5);
|
||||
wordr1[2] = ix_create_combt (combs_buf, il_pos, 6);
|
||||
wordr1[3] = ix_create_combt (combs_buf, il_pos, 7);
|
||||
|
||||
if (combs_mode == COMBINATOR_MODE_BASE_LEFT)
|
||||
{
|
||||
switch_buffer_by_offset_le_VV (wordr0, wordr1, wordr2, wordr3, pw_l_len);
|
||||
}
|
||||
else
|
||||
{
|
||||
switch_buffer_by_offset_le_VV (wordl0, wordl1, wordl2, wordl3, pw_r_len);
|
||||
}
|
||||
|
||||
u32x w0[4];
|
||||
u32x w1[4];
|
||||
u32x w2[4];
|
||||
u32x w3[4];
|
||||
|
||||
w0[0] = wordl0[0] | wordr0[0];
|
||||
w0[1] = wordl0[1] | wordr0[1];
|
||||
w0[2] = wordl0[2] | wordr0[2];
|
||||
w0[3] = wordl0[3] | wordr0[3];
|
||||
w1[0] = wordl1[0] | wordr1[0];
|
||||
w1[1] = wordl1[1] | wordr1[1];
|
||||
w1[2] = wordl1[2] | wordr1[2];
|
||||
w1[3] = wordl1[3] | wordr1[3];
|
||||
w2[0] = wordl2[0] | wordr2[0];
|
||||
w2[1] = wordl2[1] | wordr2[1];
|
||||
w2[2] = wordl2[2] | wordr2[2];
|
||||
w2[3] = wordl2[3] | wordr2[3];
|
||||
w3[0] = wordl3[0] | wordr3[0];
|
||||
w3[1] = wordl3[1] | wordr3[1];
|
||||
w3[2] = wordl3[2] | wordr3[2];
|
||||
w3[3] = wordl3[3] | wordr3[3];
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0[0]);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = hl32_to_64 (w2[1], w2[0]);
|
||||
u64x a10 = hl32_to_64 (w2[3], w2[2]);
|
||||
u64x a11 = hl32_to_64 (w3[1], w3[0]);
|
||||
u64x a12 = hl32_to_64 (w3[3], w3[2]);
|
||||
u64x a13 = 0x8000000000000000;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = h32_from_64 (a03);
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_M_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
__kernel void m17600_m08 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17600_m16 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17600_s04 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 pw_buf0[4];
|
||||
u32 pw_buf1[4];
|
||||
|
||||
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];
|
||||
|
||||
const u32 pw_l_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* 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]
|
||||
};
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
const u32x pw_r_len = pwlenx_create_combt (combs_buf, il_pos);
|
||||
|
||||
const u32x pw_len = pw_l_len + pw_r_len;
|
||||
|
||||
/**
|
||||
* concat password candidate
|
||||
*/
|
||||
|
||||
u32x wordl0[4] = { 0 };
|
||||
u32x wordl1[4] = { 0 };
|
||||
u32x wordl2[4] = { 0 };
|
||||
u32x wordl3[4] = { 0 };
|
||||
|
||||
wordl0[0] = pw_buf0[0];
|
||||
wordl0[1] = pw_buf0[1];
|
||||
wordl0[2] = pw_buf0[2];
|
||||
wordl0[3] = pw_buf0[3];
|
||||
wordl1[0] = pw_buf1[0];
|
||||
wordl1[1] = pw_buf1[1];
|
||||
wordl1[2] = pw_buf1[2];
|
||||
wordl1[3] = pw_buf1[3];
|
||||
|
||||
u32x wordr0[4] = { 0 };
|
||||
u32x wordr1[4] = { 0 };
|
||||
u32x wordr2[4] = { 0 };
|
||||
u32x wordr3[4] = { 0 };
|
||||
|
||||
wordr0[0] = ix_create_combt (combs_buf, il_pos, 0);
|
||||
wordr0[1] = ix_create_combt (combs_buf, il_pos, 1);
|
||||
wordr0[2] = ix_create_combt (combs_buf, il_pos, 2);
|
||||
wordr0[3] = ix_create_combt (combs_buf, il_pos, 3);
|
||||
wordr1[0] = ix_create_combt (combs_buf, il_pos, 4);
|
||||
wordr1[1] = ix_create_combt (combs_buf, il_pos, 5);
|
||||
wordr1[2] = ix_create_combt (combs_buf, il_pos, 6);
|
||||
wordr1[3] = ix_create_combt (combs_buf, il_pos, 7);
|
||||
|
||||
if (combs_mode == COMBINATOR_MODE_BASE_LEFT)
|
||||
{
|
||||
switch_buffer_by_offset_le_VV (wordr0, wordr1, wordr2, wordr3, pw_l_len);
|
||||
}
|
||||
else
|
||||
{
|
||||
switch_buffer_by_offset_le_VV (wordl0, wordl1, wordl2, wordl3, pw_r_len);
|
||||
}
|
||||
|
||||
u32x w0[4];
|
||||
u32x w1[4];
|
||||
u32x w2[4];
|
||||
u32x w3[4];
|
||||
|
||||
w0[0] = wordl0[0] | wordr0[0];
|
||||
w0[1] = wordl0[1] | wordr0[1];
|
||||
w0[2] = wordl0[2] | wordr0[2];
|
||||
w0[3] = wordl0[3] | wordr0[3];
|
||||
w1[0] = wordl1[0] | wordr1[0];
|
||||
w1[1] = wordl1[1] | wordr1[1];
|
||||
w1[2] = wordl1[2] | wordr1[2];
|
||||
w1[3] = wordl1[3] | wordr1[3];
|
||||
w2[0] = wordl2[0] | wordr2[0];
|
||||
w2[1] = wordl2[1] | wordr2[1];
|
||||
w2[2] = wordl2[2] | wordr2[2];
|
||||
w2[3] = wordl2[3] | wordr2[3];
|
||||
w3[0] = wordl3[0] | wordr3[0];
|
||||
w3[1] = wordl3[1] | wordr3[1];
|
||||
w3[2] = wordl3[2] | wordr3[2];
|
||||
w3[3] = wordl3[3] | wordr3[3];
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0[0]);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = hl32_to_64 (w2[1], w2[0]);
|
||||
u64x a10 = hl32_to_64 (w2[3], w2[2]);
|
||||
u64x a11 = hl32_to_64 (w3[1], w3[0]);
|
||||
u64x a12 = hl32_to_64 (w3[3], w3[2]);
|
||||
u64x a13 = 0x8000000000000000;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = h32_from_64 (a03);
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_S_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
__kernel void m17600_s08 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17600_s16 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
@ -0,0 +1,696 @@
|
||||
/**
|
||||
* Author......: See docs/credits.txt
|
||||
* License.....: MIT
|
||||
*/
|
||||
|
||||
#define NEW_SIMD_CODE
|
||||
|
||||
#include "inc_vendor.cl"
|
||||
#include "inc_hash_constants.h"
|
||||
#include "inc_hash_functions.cl"
|
||||
#include "inc_types.cl"
|
||||
#include "inc_common.cl"
|
||||
#include "inc_simd.cl"
|
||||
|
||||
__constant u64a keccakf_rndc[24] =
|
||||
{
|
||||
0x0000000000000001, 0x0000000000008082, 0x800000000000808a,
|
||||
0x8000000080008000, 0x000000000000808b, 0x0000000080000001,
|
||||
0x8000000080008081, 0x8000000000008009, 0x000000000000008a,
|
||||
0x0000000000000088, 0x0000000080008009, 0x000000008000000a,
|
||||
0x000000008000808b, 0x800000000000008b, 0x8000000000008089,
|
||||
0x8000000000008003, 0x8000000000008002, 0x8000000000000080,
|
||||
0x000000000000800a, 0x800000008000000a, 0x8000000080008081,
|
||||
0x8000000000008080, 0x0000000080000001, 0x8000000080008008
|
||||
};
|
||||
|
||||
#ifndef KECCAK_ROUNDS
|
||||
#define KECCAK_ROUNDS 24
|
||||
#endif
|
||||
|
||||
DECLSPEC void m17600m (u32 *w0, u32 *w1, u32 *w2, u32 *w3, const u32 pw_len, __global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
u32 w0l = w0[0];
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
const u32x w0r = ix_create_bft (bfs_buf, il_pos);
|
||||
|
||||
const u32x w0lr = w0l | w0r;
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0lr);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = hl32_to_64 (w2[1], w2[0]);
|
||||
u64x a10 = hl32_to_64 (w2[3], w2[2]);
|
||||
u64x a11 = hl32_to_64 (w3[1], w3[0]);
|
||||
u64x a12 = hl32_to_64 (w3[3], w3[2]);
|
||||
u64x a13 = 0x8000000000000000;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = h32_from_64 (a03);
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_M_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
DECLSPEC void m17600s (u32 *w0, u32 *w1, u32 *w2, u32 *w3, const u32 pw_len, __global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
const u64 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]
|
||||
};
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
u32 w0l = w0[0];
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
const u32x w0r = ix_create_bft (bfs_buf, il_pos);
|
||||
|
||||
const u32x w0lr = w0l | w0r;
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0lr);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = hl32_to_64 (w2[1], w2[0]);
|
||||
u64x a10 = hl32_to_64 (w2[3], w2[2]);
|
||||
u64x a11 = hl32_to_64 (w3[1], w3[0]);
|
||||
u64x a12 = hl32_to_64 (w3[3], w3[2]);
|
||||
u64x a13 = 0x8000000000000000;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = h32_from_64 (a03);
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_S_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
__kernel void m17600_m04 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
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;
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* main
|
||||
*/
|
||||
|
||||
m17600m (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m17600_m08 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
u32 w1[4];
|
||||
|
||||
w1[0] = pws[gid].i[ 4];
|
||||
w1[1] = pws[gid].i[ 5];
|
||||
w1[2] = pws[gid].i[ 6];
|
||||
w1[3] = pws[gid].i[ 7];
|
||||
|
||||
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;
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* main
|
||||
*/
|
||||
|
||||
m17600m (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m17600_m16 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
u32 w1[4];
|
||||
|
||||
w1[0] = pws[gid].i[ 4];
|
||||
w1[1] = pws[gid].i[ 5];
|
||||
w1[2] = pws[gid].i[ 6];
|
||||
w1[3] = pws[gid].i[ 7];
|
||||
|
||||
u32 w2[4];
|
||||
|
||||
w2[0] = pws[gid].i[ 8];
|
||||
w2[1] = pws[gid].i[ 9];
|
||||
w2[2] = pws[gid].i[10];
|
||||
w2[3] = pws[gid].i[11];
|
||||
|
||||
u32 w3[4];
|
||||
|
||||
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
|
||||
*/
|
||||
|
||||
m17600m (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m17600_s04 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
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;
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* main
|
||||
*/
|
||||
|
||||
m17600s (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m17600_s08 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
u32 w1[4];
|
||||
|
||||
w1[0] = pws[gid].i[ 4];
|
||||
w1[1] = pws[gid].i[ 5];
|
||||
w1[2] = pws[gid].i[ 6];
|
||||
w1[3] = pws[gid].i[ 7];
|
||||
|
||||
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;
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* main
|
||||
*/
|
||||
|
||||
m17600s (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m17600_s16 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
u32 w1[4];
|
||||
|
||||
w1[0] = pws[gid].i[ 4];
|
||||
w1[1] = pws[gid].i[ 5];
|
||||
w1[2] = pws[gid].i[ 6];
|
||||
w1[3] = pws[gid].i[ 7];
|
||||
|
||||
u32 w2[4];
|
||||
|
||||
w2[0] = pws[gid].i[ 8];
|
||||
w2[1] = pws[gid].i[ 9];
|
||||
w2[2] = pws[gid].i[10];
|
||||
w2[3] = pws[gid].i[11];
|
||||
|
||||
u32 w3[4];
|
||||
|
||||
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
|
||||
*/
|
||||
|
||||
m17600s (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
@ -0,0 +1,479 @@
|
||||
/**
|
||||
* Author......: See docs/credits.txt
|
||||
* License.....: MIT
|
||||
*/
|
||||
|
||||
#define NEW_SIMD_CODE
|
||||
|
||||
#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_optimized.h"
|
||||
#include "inc_rp_optimized.cl"
|
||||
#include "inc_simd.cl"
|
||||
|
||||
__constant u64a keccakf_rndc[24] =
|
||||
{
|
||||
0x0000000000000001, 0x0000000000008082, 0x800000000000808a,
|
||||
0x8000000080008000, 0x000000000000808b, 0x0000000080000001,
|
||||
0x8000000080008081, 0x8000000000008009, 0x000000000000008a,
|
||||
0x0000000000000088, 0x0000000080008009, 0x000000008000000a,
|
||||
0x000000008000808b, 0x800000000000008b, 0x8000000000008089,
|
||||
0x8000000000008003, 0x8000000000008002, 0x8000000000000080,
|
||||
0x000000000000800a, 0x800000008000000a, 0x8000000080008081,
|
||||
0x8000000000008080, 0x0000000080000001, 0x8000000080008008
|
||||
};
|
||||
|
||||
#ifndef KECCAK_ROUNDS
|
||||
#define KECCAK_ROUNDS 24
|
||||
#endif
|
||||
|
||||
__kernel void m17700_m04 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 pw_buf0[4];
|
||||
u32 pw_buf1[4];
|
||||
|
||||
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];
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
u32x w0[4] = { 0 };
|
||||
u32x w1[4] = { 0 };
|
||||
u32x w2[4] = { 0 };
|
||||
u32x w3[4] = { 0 };
|
||||
|
||||
const u32x out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
|
||||
|
||||
append_0x01_2x4_VV (w0, w1, out_len);
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0[0]);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = 0;
|
||||
u64x a10 = 0;
|
||||
u64x a11 = 0;
|
||||
u64x a12 = 0;
|
||||
u64x a13 = 0;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0x8000000000000000;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = 0;
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_M_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
__kernel void m17700_m08 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17700_m16 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17700_s04 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 pw_buf0[4];
|
||||
u32 pw_buf1[4];
|
||||
|
||||
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];
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* 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]
|
||||
};
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
u32x w0[4] = { 0 };
|
||||
u32x w1[4] = { 0 };
|
||||
u32x w2[4] = { 0 };
|
||||
u32x w3[4] = { 0 };
|
||||
|
||||
const u32x out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
|
||||
|
||||
append_0x01_2x4_VV (w0, w1, out_len);
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0[0]);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = 0;
|
||||
u64x a10 = 0;
|
||||
u64x a11 = 0;
|
||||
u64x a12 = 0;
|
||||
u64x a13 = 0;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0x8000000000000000;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = 0;
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_S_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
__kernel void m17700_s08 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17700_s16 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
@ -0,0 +1,594 @@
|
||||
/**
|
||||
* Author......: See docs/credits.txt
|
||||
* License.....: MIT
|
||||
*/
|
||||
|
||||
#define NEW_SIMD_CODE
|
||||
|
||||
#include "inc_vendor.cl"
|
||||
#include "inc_hash_constants.h"
|
||||
#include "inc_hash_functions.cl"
|
||||
#include "inc_types.cl"
|
||||
#include "inc_common.cl"
|
||||
#include "inc_simd.cl"
|
||||
|
||||
__constant u64a keccakf_rndc[24] =
|
||||
{
|
||||
0x0000000000000001, 0x0000000000008082, 0x800000000000808a,
|
||||
0x8000000080008000, 0x000000000000808b, 0x0000000080000001,
|
||||
0x8000000080008081, 0x8000000000008009, 0x000000000000008a,
|
||||
0x0000000000000088, 0x0000000080008009, 0x000000008000000a,
|
||||
0x000000008000808b, 0x800000000000008b, 0x8000000000008089,
|
||||
0x8000000000008003, 0x8000000000008002, 0x8000000000000080,
|
||||
0x000000000000800a, 0x800000008000000a, 0x8000000080008081,
|
||||
0x8000000000008080, 0x0000000080000001, 0x8000000080008008
|
||||
};
|
||||
|
||||
#ifndef KECCAK_ROUNDS
|
||||
#define KECCAK_ROUNDS 24
|
||||
#endif
|
||||
|
||||
__kernel void m17700_m04 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 pw_buf0[4];
|
||||
u32 pw_buf1[4];
|
||||
|
||||
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];
|
||||
|
||||
const u32 pw_l_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
const u32x pw_r_len = pwlenx_create_combt (combs_buf, il_pos);
|
||||
|
||||
const u32x pw_len = pw_l_len + pw_r_len;
|
||||
|
||||
/**
|
||||
* concat password candidate
|
||||
*/
|
||||
|
||||
u32x wordl0[4] = { 0 };
|
||||
u32x wordl1[4] = { 0 };
|
||||
u32x wordl2[4] = { 0 };
|
||||
u32x wordl3[4] = { 0 };
|
||||
|
||||
wordl0[0] = pw_buf0[0];
|
||||
wordl0[1] = pw_buf0[1];
|
||||
wordl0[2] = pw_buf0[2];
|
||||
wordl0[3] = pw_buf0[3];
|
||||
wordl1[0] = pw_buf1[0];
|
||||
wordl1[1] = pw_buf1[1];
|
||||
wordl1[2] = pw_buf1[2];
|
||||
wordl1[3] = pw_buf1[3];
|
||||
|
||||
u32x wordr0[4] = { 0 };
|
||||
u32x wordr1[4] = { 0 };
|
||||
u32x wordr2[4] = { 0 };
|
||||
u32x wordr3[4] = { 0 };
|
||||
|
||||
wordr0[0] = ix_create_combt (combs_buf, il_pos, 0);
|
||||
wordr0[1] = ix_create_combt (combs_buf, il_pos, 1);
|
||||
wordr0[2] = ix_create_combt (combs_buf, il_pos, 2);
|
||||
wordr0[3] = ix_create_combt (combs_buf, il_pos, 3);
|
||||
wordr1[0] = ix_create_combt (combs_buf, il_pos, 4);
|
||||
wordr1[1] = ix_create_combt (combs_buf, il_pos, 5);
|
||||
wordr1[2] = ix_create_combt (combs_buf, il_pos, 6);
|
||||
wordr1[3] = ix_create_combt (combs_buf, il_pos, 7);
|
||||
|
||||
if (combs_mode == COMBINATOR_MODE_BASE_LEFT)
|
||||
{
|
||||
switch_buffer_by_offset_le_VV (wordr0, wordr1, wordr2, wordr3, pw_l_len);
|
||||
}
|
||||
else
|
||||
{
|
||||
switch_buffer_by_offset_le_VV (wordl0, wordl1, wordl2, wordl3, pw_r_len);
|
||||
}
|
||||
|
||||
u32x w0[4];
|
||||
u32x w1[4];
|
||||
u32x w2[4];
|
||||
u32x w3[4];
|
||||
|
||||
w0[0] = wordl0[0] | wordr0[0];
|
||||
w0[1] = wordl0[1] | wordr0[1];
|
||||
w0[2] = wordl0[2] | wordr0[2];
|
||||
w0[3] = wordl0[3] | wordr0[3];
|
||||
w1[0] = wordl1[0] | wordr1[0];
|
||||
w1[1] = wordl1[1] | wordr1[1];
|
||||
w1[2] = wordl1[2] | wordr1[2];
|
||||
w1[3] = wordl1[3] | wordr1[3];
|
||||
w2[0] = wordl2[0] | wordr2[0];
|
||||
w2[1] = wordl2[1] | wordr2[1];
|
||||
w2[2] = wordl2[2] | wordr2[2];
|
||||
w2[3] = wordl2[3] | wordr2[3];
|
||||
w3[0] = wordl3[0] | wordr3[0];
|
||||
w3[1] = wordl3[1] | wordr3[1];
|
||||
w3[2] = wordl3[2] | wordr3[2];
|
||||
w3[3] = wordl3[3] | wordr3[3];
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0[0]);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = hl32_to_64 (w2[1], w2[0]);
|
||||
u64x a10 = hl32_to_64 (w2[3], w2[2]);
|
||||
u64x a11 = hl32_to_64 (w3[1], w3[0]);
|
||||
u64x a12 = hl32_to_64 (w3[3], w3[2]);
|
||||
u64x a13 = 0;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0x8000000000000000;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = 0;
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_M_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
__kernel void m17700_m08 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17700_m16 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17700_s04 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 pw_buf0[4];
|
||||
u32 pw_buf1[4];
|
||||
|
||||
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];
|
||||
|
||||
const u32 pw_l_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* 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]
|
||||
};
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
const u32x pw_r_len = pwlenx_create_combt (combs_buf, il_pos);
|
||||
|
||||
const u32x pw_len = pw_l_len + pw_r_len;
|
||||
|
||||
/**
|
||||
* concat password candidate
|
||||
*/
|
||||
|
||||
u32x wordl0[4] = { 0 };
|
||||
u32x wordl1[4] = { 0 };
|
||||
u32x wordl2[4] = { 0 };
|
||||
u32x wordl3[4] = { 0 };
|
||||
|
||||
wordl0[0] = pw_buf0[0];
|
||||
wordl0[1] = pw_buf0[1];
|
||||
wordl0[2] = pw_buf0[2];
|
||||
wordl0[3] = pw_buf0[3];
|
||||
wordl1[0] = pw_buf1[0];
|
||||
wordl1[1] = pw_buf1[1];
|
||||
wordl1[2] = pw_buf1[2];
|
||||
wordl1[3] = pw_buf1[3];
|
||||
|
||||
u32x wordr0[4] = { 0 };
|
||||
u32x wordr1[4] = { 0 };
|
||||
u32x wordr2[4] = { 0 };
|
||||
u32x wordr3[4] = { 0 };
|
||||
|
||||
wordr0[0] = ix_create_combt (combs_buf, il_pos, 0);
|
||||
wordr0[1] = ix_create_combt (combs_buf, il_pos, 1);
|
||||
wordr0[2] = ix_create_combt (combs_buf, il_pos, 2);
|
||||
wordr0[3] = ix_create_combt (combs_buf, il_pos, 3);
|
||||
wordr1[0] = ix_create_combt (combs_buf, il_pos, 4);
|
||||
wordr1[1] = ix_create_combt (combs_buf, il_pos, 5);
|
||||
wordr1[2] = ix_create_combt (combs_buf, il_pos, 6);
|
||||
wordr1[3] = ix_create_combt (combs_buf, il_pos, 7);
|
||||
|
||||
if (combs_mode == COMBINATOR_MODE_BASE_LEFT)
|
||||
{
|
||||
switch_buffer_by_offset_le_VV (wordr0, wordr1, wordr2, wordr3, pw_l_len);
|
||||
}
|
||||
else
|
||||
{
|
||||
switch_buffer_by_offset_le_VV (wordl0, wordl1, wordl2, wordl3, pw_r_len);
|
||||
}
|
||||
|
||||
u32x w0[4];
|
||||
u32x w1[4];
|
||||
u32x w2[4];
|
||||
u32x w3[4];
|
||||
|
||||
w0[0] = wordl0[0] | wordr0[0];
|
||||
w0[1] = wordl0[1] | wordr0[1];
|
||||
w0[2] = wordl0[2] | wordr0[2];
|
||||
w0[3] = wordl0[3] | wordr0[3];
|
||||
w1[0] = wordl1[0] | wordr1[0];
|
||||
w1[1] = wordl1[1] | wordr1[1];
|
||||
w1[2] = wordl1[2] | wordr1[2];
|
||||
w1[3] = wordl1[3] | wordr1[3];
|
||||
w2[0] = wordl2[0] | wordr2[0];
|
||||
w2[1] = wordl2[1] | wordr2[1];
|
||||
w2[2] = wordl2[2] | wordr2[2];
|
||||
w2[3] = wordl2[3] | wordr2[3];
|
||||
w3[0] = wordl3[0] | wordr3[0];
|
||||
w3[1] = wordl3[1] | wordr3[1];
|
||||
w3[2] = wordl3[2] | wordr3[2];
|
||||
w3[3] = wordl3[3] | wordr3[3];
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0[0]);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = hl32_to_64 (w2[1], w2[0]);
|
||||
u64x a10 = hl32_to_64 (w2[3], w2[2]);
|
||||
u64x a11 = hl32_to_64 (w3[1], w3[0]);
|
||||
u64x a12 = hl32_to_64 (w3[3], w3[2]);
|
||||
u64x a13 = 0;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0x8000000000000000;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = 0;
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_S_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
__kernel void m17700_s08 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17700_s16 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
@ -0,0 +1,695 @@
|
||||
/**
|
||||
* Author......: See docs/credits.txt
|
||||
* License.....: MIT
|
||||
*/
|
||||
|
||||
#define NEW_SIMD_CODE
|
||||
|
||||
#include "inc_vendor.cl"
|
||||
#include "inc_hash_constants.h"
|
||||
#include "inc_hash_functions.cl"
|
||||
#include "inc_types.cl"
|
||||
#include "inc_common.cl"
|
||||
#include "inc_simd.cl"
|
||||
|
||||
__constant u64a keccakf_rndc[24] =
|
||||
{
|
||||
0x0000000000000001, 0x0000000000008082, 0x800000000000808a,
|
||||
0x8000000080008000, 0x000000000000808b, 0x0000000080000001,
|
||||
0x8000000080008081, 0x8000000000008009, 0x000000000000008a,
|
||||
0x0000000000000088, 0x0000000080008009, 0x000000008000000a,
|
||||
0x000000008000808b, 0x800000000000008b, 0x8000000000008089,
|
||||
0x8000000000008003, 0x8000000000008002, 0x8000000000000080,
|
||||
0x000000000000800a, 0x800000008000000a, 0x8000000080008081,
|
||||
0x8000000000008080, 0x0000000080000001, 0x8000000080008008
|
||||
};
|
||||
|
||||
#ifndef KECCAK_ROUNDS
|
||||
#define KECCAK_ROUNDS 24
|
||||
#endif
|
||||
|
||||
DECLSPEC void m17300m (u32 *w0, u32 *w1, u32 *w2, u32 *w3, const u32 pw_len, __global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
u32 w0l = w0[0];
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
const u32x w0r = ix_create_bft (bfs_buf, il_pos);
|
||||
|
||||
const u32x w0lr = w0l | w0r;
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0lr);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = hl32_to_64 (w2[1], w2[0]);
|
||||
u64x a10 = hl32_to_64 (w2[3], w2[2]);
|
||||
u64x a11 = hl32_to_64 (w3[1], w3[0]);
|
||||
u64x a12 = hl32_to_64 (w3[3], w3[2]);
|
||||
u64x a13 = 0;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0x8000000000000000;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = 0;
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_M_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
DECLSPEC void m17300s (u32 *w0, u32 *w1, u32 *w2, u32 *w3, const u32 pw_len, __global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
const u64 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]
|
||||
};
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
u32 w0l = w0[0];
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
const u32x w0r = ix_create_bft (bfs_buf, il_pos);
|
||||
|
||||
const u32x w0lr = w0l | w0r;
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0lr);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = hl32_to_64 (w2[1], w2[0]);
|
||||
u64x a10 = hl32_to_64 (w2[3], w2[2]);
|
||||
u64x a11 = hl32_to_64 (w3[1], w3[0]);
|
||||
u64x a12 = hl32_to_64 (w3[3], w3[2]);
|
||||
u64x a13 = 0;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0x8000000000000000;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = 0;
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_S_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
__kernel void m17700_m04 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
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;
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* main
|
||||
*/
|
||||
|
||||
m17300m (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m17700_m08 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
u32 w1[4];
|
||||
|
||||
w1[0] = pws[gid].i[ 4];
|
||||
w1[1] = pws[gid].i[ 5];
|
||||
w1[2] = pws[gid].i[ 6];
|
||||
w1[3] = pws[gid].i[ 7];
|
||||
|
||||
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;
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* main
|
||||
*/
|
||||
|
||||
m17300m (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m17700_m16 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
u32 w1[4];
|
||||
|
||||
w1[0] = pws[gid].i[ 4];
|
||||
w1[1] = pws[gid].i[ 5];
|
||||
w1[2] = pws[gid].i[ 6];
|
||||
w1[3] = pws[gid].i[ 7];
|
||||
|
||||
u32 w2[4];
|
||||
|
||||
w2[0] = pws[gid].i[ 8];
|
||||
w2[1] = pws[gid].i[ 9];
|
||||
w2[2] = pws[gid].i[10];
|
||||
w2[3] = pws[gid].i[11];
|
||||
|
||||
u32 w3[4];
|
||||
|
||||
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
|
||||
*/
|
||||
|
||||
m17300m (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m17700_s04 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
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;
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* main
|
||||
*/
|
||||
|
||||
m17300s (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m17700_s08 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
u32 w1[4];
|
||||
|
||||
w1[0] = pws[gid].i[ 4];
|
||||
w1[1] = pws[gid].i[ 5];
|
||||
w1[2] = pws[gid].i[ 6];
|
||||
w1[3] = pws[gid].i[ 7];
|
||||
|
||||
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;
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* main
|
||||
*/
|
||||
|
||||
m17300s (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m17700_s16 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
u32 w1[4];
|
||||
|
||||
w1[0] = pws[gid].i[ 4];
|
||||
w1[1] = pws[gid].i[ 5];
|
||||
w1[2] = pws[gid].i[ 6];
|
||||
w1[3] = pws[gid].i[ 7];
|
||||
|
||||
u32 w2[4];
|
||||
|
||||
w2[0] = pws[gid].i[ 8];
|
||||
w2[1] = pws[gid].i[ 9];
|
||||
w2[2] = pws[gid].i[10];
|
||||
w2[3] = pws[gid].i[11];
|
||||
|
||||
u32 w3[4];
|
||||
|
||||
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
|
||||
*/
|
||||
|
||||
m17300s (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
@ -0,0 +1,594 @@
|
||||
/**
|
||||
* Author......: See docs/credits.txt
|
||||
* License.....: MIT
|
||||
*/
|
||||
|
||||
#define NEW_SIMD_CODE
|
||||
|
||||
#include "inc_vendor.cl"
|
||||
#include "inc_hash_constants.h"
|
||||
#include "inc_hash_functions.cl"
|
||||
#include "inc_types.cl"
|
||||
#include "inc_common.cl"
|
||||
#include "inc_simd.cl"
|
||||
|
||||
__constant u64a keccakf_rndc[24] =
|
||||
{
|
||||
0x0000000000000001, 0x0000000000008082, 0x800000000000808a,
|
||||
0x8000000080008000, 0x000000000000808b, 0x0000000080000001,
|
||||
0x8000000080008081, 0x8000000000008009, 0x000000000000008a,
|
||||
0x0000000000000088, 0x0000000080008009, 0x000000008000000a,
|
||||
0x000000008000808b, 0x800000000000008b, 0x8000000000008089,
|
||||
0x8000000000008003, 0x8000000000008002, 0x8000000000000080,
|
||||
0x000000000000800a, 0x800000008000000a, 0x8000000080008081,
|
||||
0x8000000000008080, 0x0000000080000001, 0x8000000080008008
|
||||
};
|
||||
|
||||
#ifndef KECCAK_ROUNDS
|
||||
#define KECCAK_ROUNDS 24
|
||||
#endif
|
||||
|
||||
__kernel void m17800_m04 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 pw_buf0[4];
|
||||
u32 pw_buf1[4];
|
||||
|
||||
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];
|
||||
|
||||
const u32 pw_l_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
const u32x pw_r_len = pwlenx_create_combt (combs_buf, il_pos);
|
||||
|
||||
const u32x pw_len = pw_l_len + pw_r_len;
|
||||
|
||||
/**
|
||||
* concat password candidate
|
||||
*/
|
||||
|
||||
u32x wordl0[4] = { 0 };
|
||||
u32x wordl1[4] = { 0 };
|
||||
u32x wordl2[4] = { 0 };
|
||||
u32x wordl3[4] = { 0 };
|
||||
|
||||
wordl0[0] = pw_buf0[0];
|
||||
wordl0[1] = pw_buf0[1];
|
||||
wordl0[2] = pw_buf0[2];
|
||||
wordl0[3] = pw_buf0[3];
|
||||
wordl1[0] = pw_buf1[0];
|
||||
wordl1[1] = pw_buf1[1];
|
||||
wordl1[2] = pw_buf1[2];
|
||||
wordl1[3] = pw_buf1[3];
|
||||
|
||||
u32x wordr0[4] = { 0 };
|
||||
u32x wordr1[4] = { 0 };
|
||||
u32x wordr2[4] = { 0 };
|
||||
u32x wordr3[4] = { 0 };
|
||||
|
||||
wordr0[0] = ix_create_combt (combs_buf, il_pos, 0);
|
||||
wordr0[1] = ix_create_combt (combs_buf, il_pos, 1);
|
||||
wordr0[2] = ix_create_combt (combs_buf, il_pos, 2);
|
||||
wordr0[3] = ix_create_combt (combs_buf, il_pos, 3);
|
||||
wordr1[0] = ix_create_combt (combs_buf, il_pos, 4);
|
||||
wordr1[1] = ix_create_combt (combs_buf, il_pos, 5);
|
||||
wordr1[2] = ix_create_combt (combs_buf, il_pos, 6);
|
||||
wordr1[3] = ix_create_combt (combs_buf, il_pos, 7);
|
||||
|
||||
if (combs_mode == COMBINATOR_MODE_BASE_LEFT)
|
||||
{
|
||||
switch_buffer_by_offset_le_VV (wordr0, wordr1, wordr2, wordr3, pw_l_len);
|
||||
}
|
||||
else
|
||||
{
|
||||
switch_buffer_by_offset_le_VV (wordl0, wordl1, wordl2, wordl3, pw_r_len);
|
||||
}
|
||||
|
||||
u32x w0[4];
|
||||
u32x w1[4];
|
||||
u32x w2[4];
|
||||
u32x w3[4];
|
||||
|
||||
w0[0] = wordl0[0] | wordr0[0];
|
||||
w0[1] = wordl0[1] | wordr0[1];
|
||||
w0[2] = wordl0[2] | wordr0[2];
|
||||
w0[3] = wordl0[3] | wordr0[3];
|
||||
w1[0] = wordl1[0] | wordr1[0];
|
||||
w1[1] = wordl1[1] | wordr1[1];
|
||||
w1[2] = wordl1[2] | wordr1[2];
|
||||
w1[3] = wordl1[3] | wordr1[3];
|
||||
w2[0] = wordl2[0] | wordr2[0];
|
||||
w2[1] = wordl2[1] | wordr2[1];
|
||||
w2[2] = wordl2[2] | wordr2[2];
|
||||
w2[3] = wordl2[3] | wordr2[3];
|
||||
w3[0] = wordl3[0] | wordr3[0];
|
||||
w3[1] = wordl3[1] | wordr3[1];
|
||||
w3[2] = wordl3[2] | wordr3[2];
|
||||
w3[3] = wordl3[3] | wordr3[3];
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0[0]);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = hl32_to_64 (w2[1], w2[0]);
|
||||
u64x a10 = hl32_to_64 (w2[3], w2[2]);
|
||||
u64x a11 = hl32_to_64 (w3[1], w3[0]);
|
||||
u64x a12 = hl32_to_64 (w3[3], w3[2]);
|
||||
u64x a13 = 0;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0x8000000000000000;
|
||||
u64x a32 = 0;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = h32_from_64 (a03);
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_M_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
__kernel void m17800_m08 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17800_m16 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17800_s04 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 pw_buf0[4];
|
||||
u32 pw_buf1[4];
|
||||
|
||||
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];
|
||||
|
||||
const u32 pw_l_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* 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]
|
||||
};
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
const u32x pw_r_len = pwlenx_create_combt (combs_buf, il_pos);
|
||||
|
||||
const u32x pw_len = pw_l_len + pw_r_len;
|
||||
|
||||
/**
|
||||
* concat password candidate
|
||||
*/
|
||||
|
||||
u32x wordl0[4] = { 0 };
|
||||
u32x wordl1[4] = { 0 };
|
||||
u32x wordl2[4] = { 0 };
|
||||
u32x wordl3[4] = { 0 };
|
||||
|
||||
wordl0[0] = pw_buf0[0];
|
||||
wordl0[1] = pw_buf0[1];
|
||||
wordl0[2] = pw_buf0[2];
|
||||
wordl0[3] = pw_buf0[3];
|
||||
wordl1[0] = pw_buf1[0];
|
||||
wordl1[1] = pw_buf1[1];
|
||||
wordl1[2] = pw_buf1[2];
|
||||
wordl1[3] = pw_buf1[3];
|
||||
|
||||
u32x wordr0[4] = { 0 };
|
||||
u32x wordr1[4] = { 0 };
|
||||
u32x wordr2[4] = { 0 };
|
||||
u32x wordr3[4] = { 0 };
|
||||
|
||||
wordr0[0] = ix_create_combt (combs_buf, il_pos, 0);
|
||||
wordr0[1] = ix_create_combt (combs_buf, il_pos, 1);
|
||||
wordr0[2] = ix_create_combt (combs_buf, il_pos, 2);
|
||||
wordr0[3] = ix_create_combt (combs_buf, il_pos, 3);
|
||||
wordr1[0] = ix_create_combt (combs_buf, il_pos, 4);
|
||||
wordr1[1] = ix_create_combt (combs_buf, il_pos, 5);
|
||||
wordr1[2] = ix_create_combt (combs_buf, il_pos, 6);
|
||||
wordr1[3] = ix_create_combt (combs_buf, il_pos, 7);
|
||||
|
||||
if (combs_mode == COMBINATOR_MODE_BASE_LEFT)
|
||||
{
|
||||
switch_buffer_by_offset_le_VV (wordr0, wordr1, wordr2, wordr3, pw_l_len);
|
||||
}
|
||||
else
|
||||
{
|
||||
switch_buffer_by_offset_le_VV (wordl0, wordl1, wordl2, wordl3, pw_r_len);
|
||||
}
|
||||
|
||||
u32x w0[4];
|
||||
u32x w1[4];
|
||||
u32x w2[4];
|
||||
u32x w3[4];
|
||||
|
||||
w0[0] = wordl0[0] | wordr0[0];
|
||||
w0[1] = wordl0[1] | wordr0[1];
|
||||
w0[2] = wordl0[2] | wordr0[2];
|
||||
w0[3] = wordl0[3] | wordr0[3];
|
||||
w1[0] = wordl1[0] | wordr1[0];
|
||||
w1[1] = wordl1[1] | wordr1[1];
|
||||
w1[2] = wordl1[2] | wordr1[2];
|
||||
w1[3] = wordl1[3] | wordr1[3];
|
||||
w2[0] = wordl2[0] | wordr2[0];
|
||||
w2[1] = wordl2[1] | wordr2[1];
|
||||
w2[2] = wordl2[2] | wordr2[2];
|
||||
w2[3] = wordl2[3] | wordr2[3];
|
||||
w3[0] = wordl3[0] | wordr3[0];
|
||||
w3[1] = wordl3[1] | wordr3[1];
|
||||
w3[2] = wordl3[2] | wordr3[2];
|
||||
w3[3] = wordl3[3] | wordr3[3];
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0[0]);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = hl32_to_64 (w2[1], w2[0]);
|
||||
u64x a10 = hl32_to_64 (w2[3], w2[2]);
|
||||
u64x a11 = hl32_to_64 (w3[1], w3[0]);
|
||||
u64x a12 = hl32_to_64 (w3[3], w3[2]);
|
||||
u64x a13 = 0;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0x8000000000000000;
|
||||
u64x a32 = 0;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = h32_from_64 (a03);
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_S_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
__kernel void m17800_s08 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17800_s16 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
@ -0,0 +1,479 @@
|
||||
/**
|
||||
* Author......: See docs/credits.txt
|
||||
* License.....: MIT
|
||||
*/
|
||||
|
||||
#define NEW_SIMD_CODE
|
||||
|
||||
#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_optimized.h"
|
||||
#include "inc_rp_optimized.cl"
|
||||
#include "inc_simd.cl"
|
||||
|
||||
__constant u64a keccakf_rndc[24] =
|
||||
{
|
||||
0x0000000000000001, 0x0000000000008082, 0x800000000000808a,
|
||||
0x8000000080008000, 0x000000000000808b, 0x0000000080000001,
|
||||
0x8000000080008081, 0x8000000000008009, 0x000000000000008a,
|
||||
0x0000000000000088, 0x0000000080008009, 0x000000008000000a,
|
||||
0x000000008000808b, 0x800000000000008b, 0x8000000000008089,
|
||||
0x8000000000008003, 0x8000000000008002, 0x8000000000000080,
|
||||
0x000000000000800a, 0x800000008000000a, 0x8000000080008081,
|
||||
0x8000000000008080, 0x0000000080000001, 0x8000000080008008
|
||||
};
|
||||
|
||||
#ifndef KECCAK_ROUNDS
|
||||
#define KECCAK_ROUNDS 24
|
||||
#endif
|
||||
|
||||
__kernel void m17900_m04 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 pw_buf0[4];
|
||||
u32 pw_buf1[4];
|
||||
|
||||
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];
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
u32x w0[4] = { 0 };
|
||||
u32x w1[4] = { 0 };
|
||||
u32x w2[4] = { 0 };
|
||||
u32x w3[4] = { 0 };
|
||||
|
||||
const u32x out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
|
||||
|
||||
append_0x01_2x4_VV (w0, w1, out_len);
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0[0]);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = 0;
|
||||
u64x a10 = 0;
|
||||
u64x a11 = 0;
|
||||
u64x a12 = 0;
|
||||
u64x a13 = 0;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0x8000000000000000;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = h32_from_64 (a03);
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_M_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
__kernel void m17900_m08 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17900_m16 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17900_s04 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 pw_buf0[4];
|
||||
u32 pw_buf1[4];
|
||||
|
||||
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];
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* 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]
|
||||
};
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
u32x w0[4] = { 0 };
|
||||
u32x w1[4] = { 0 };
|
||||
u32x w2[4] = { 0 };
|
||||
u32x w3[4] = { 0 };
|
||||
|
||||
const u32x out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
|
||||
|
||||
append_0x01_2x4_VV (w0, w1, out_len);
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0[0]);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = 0;
|
||||
u64x a10 = 0;
|
||||
u64x a11 = 0;
|
||||
u64x a12 = 0;
|
||||
u64x a13 = 0;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0x8000000000000000;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = h32_from_64 (a03);
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_S_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
__kernel void m17900_s08 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17900_s16 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
@ -0,0 +1,594 @@
|
||||
/**
|
||||
* Author......: See docs/credits.txt
|
||||
* License.....: MIT
|
||||
*/
|
||||
|
||||
#define NEW_SIMD_CODE
|
||||
|
||||
#include "inc_vendor.cl"
|
||||
#include "inc_hash_constants.h"
|
||||
#include "inc_hash_functions.cl"
|
||||
#include "inc_types.cl"
|
||||
#include "inc_common.cl"
|
||||
#include "inc_simd.cl"
|
||||
|
||||
__constant u64a keccakf_rndc[24] =
|
||||
{
|
||||
0x0000000000000001, 0x0000000000008082, 0x800000000000808a,
|
||||
0x8000000080008000, 0x000000000000808b, 0x0000000080000001,
|
||||
0x8000000080008081, 0x8000000000008009, 0x000000000000008a,
|
||||
0x0000000000000088, 0x0000000080008009, 0x000000008000000a,
|
||||
0x000000008000808b, 0x800000000000008b, 0x8000000000008089,
|
||||
0x8000000000008003, 0x8000000000008002, 0x8000000000000080,
|
||||
0x000000000000800a, 0x800000008000000a, 0x8000000080008081,
|
||||
0x8000000000008080, 0x0000000080000001, 0x8000000080008008
|
||||
};
|
||||
|
||||
#ifndef KECCAK_ROUNDS
|
||||
#define KECCAK_ROUNDS 24
|
||||
#endif
|
||||
|
||||
__kernel void m17900_m04 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 pw_buf0[4];
|
||||
u32 pw_buf1[4];
|
||||
|
||||
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];
|
||||
|
||||
const u32 pw_l_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
const u32x pw_r_len = pwlenx_create_combt (combs_buf, il_pos);
|
||||
|
||||
const u32x pw_len = pw_l_len + pw_r_len;
|
||||
|
||||
/**
|
||||
* concat password candidate
|
||||
*/
|
||||
|
||||
u32x wordl0[4] = { 0 };
|
||||
u32x wordl1[4] = { 0 };
|
||||
u32x wordl2[4] = { 0 };
|
||||
u32x wordl3[4] = { 0 };
|
||||
|
||||
wordl0[0] = pw_buf0[0];
|
||||
wordl0[1] = pw_buf0[1];
|
||||
wordl0[2] = pw_buf0[2];
|
||||
wordl0[3] = pw_buf0[3];
|
||||
wordl1[0] = pw_buf1[0];
|
||||
wordl1[1] = pw_buf1[1];
|
||||
wordl1[2] = pw_buf1[2];
|
||||
wordl1[3] = pw_buf1[3];
|
||||
|
||||
u32x wordr0[4] = { 0 };
|
||||
u32x wordr1[4] = { 0 };
|
||||
u32x wordr2[4] = { 0 };
|
||||
u32x wordr3[4] = { 0 };
|
||||
|
||||
wordr0[0] = ix_create_combt (combs_buf, il_pos, 0);
|
||||
wordr0[1] = ix_create_combt (combs_buf, il_pos, 1);
|
||||
wordr0[2] = ix_create_combt (combs_buf, il_pos, 2);
|
||||
wordr0[3] = ix_create_combt (combs_buf, il_pos, 3);
|
||||
wordr1[0] = ix_create_combt (combs_buf, il_pos, 4);
|
||||
wordr1[1] = ix_create_combt (combs_buf, il_pos, 5);
|
||||
wordr1[2] = ix_create_combt (combs_buf, il_pos, 6);
|
||||
wordr1[3] = ix_create_combt (combs_buf, il_pos, 7);
|
||||
|
||||
if (combs_mode == COMBINATOR_MODE_BASE_LEFT)
|
||||
{
|
||||
switch_buffer_by_offset_le_VV (wordr0, wordr1, wordr2, wordr3, pw_l_len);
|
||||
}
|
||||
else
|
||||
{
|
||||
switch_buffer_by_offset_le_VV (wordl0, wordl1, wordl2, wordl3, pw_r_len);
|
||||
}
|
||||
|
||||
u32x w0[4];
|
||||
u32x w1[4];
|
||||
u32x w2[4];
|
||||
u32x w3[4];
|
||||
|
||||
w0[0] = wordl0[0] | wordr0[0];
|
||||
w0[1] = wordl0[1] | wordr0[1];
|
||||
w0[2] = wordl0[2] | wordr0[2];
|
||||
w0[3] = wordl0[3] | wordr0[3];
|
||||
w1[0] = wordl1[0] | wordr1[0];
|
||||
w1[1] = wordl1[1] | wordr1[1];
|
||||
w1[2] = wordl1[2] | wordr1[2];
|
||||
w1[3] = wordl1[3] | wordr1[3];
|
||||
w2[0] = wordl2[0] | wordr2[0];
|
||||
w2[1] = wordl2[1] | wordr2[1];
|
||||
w2[2] = wordl2[2] | wordr2[2];
|
||||
w2[3] = wordl2[3] | wordr2[3];
|
||||
w3[0] = wordl3[0] | wordr3[0];
|
||||
w3[1] = wordl3[1] | wordr3[1];
|
||||
w3[2] = wordl3[2] | wordr3[2];
|
||||
w3[3] = wordl3[3] | wordr3[3];
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0[0]);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = hl32_to_64 (w2[1], w2[0]);
|
||||
u64x a10 = hl32_to_64 (w2[3], w2[2]);
|
||||
u64x a11 = hl32_to_64 (w3[1], w3[0]);
|
||||
u64x a12 = hl32_to_64 (w3[3], w3[2]);
|
||||
u64x a13 = 0;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0x8000000000000000;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = h32_from_64 (a03);
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_M_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
__kernel void m17900_m08 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17900_m16 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17900_s04 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 pw_buf0[4];
|
||||
u32 pw_buf1[4];
|
||||
|
||||
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];
|
||||
|
||||
const u32 pw_l_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* 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]
|
||||
};
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
const u32x pw_r_len = pwlenx_create_combt (combs_buf, il_pos);
|
||||
|
||||
const u32x pw_len = pw_l_len + pw_r_len;
|
||||
|
||||
/**
|
||||
* concat password candidate
|
||||
*/
|
||||
|
||||
u32x wordl0[4] = { 0 };
|
||||
u32x wordl1[4] = { 0 };
|
||||
u32x wordl2[4] = { 0 };
|
||||
u32x wordl3[4] = { 0 };
|
||||
|
||||
wordl0[0] = pw_buf0[0];
|
||||
wordl0[1] = pw_buf0[1];
|
||||
wordl0[2] = pw_buf0[2];
|
||||
wordl0[3] = pw_buf0[3];
|
||||
wordl1[0] = pw_buf1[0];
|
||||
wordl1[1] = pw_buf1[1];
|
||||
wordl1[2] = pw_buf1[2];
|
||||
wordl1[3] = pw_buf1[3];
|
||||
|
||||
u32x wordr0[4] = { 0 };
|
||||
u32x wordr1[4] = { 0 };
|
||||
u32x wordr2[4] = { 0 };
|
||||
u32x wordr3[4] = { 0 };
|
||||
|
||||
wordr0[0] = ix_create_combt (combs_buf, il_pos, 0);
|
||||
wordr0[1] = ix_create_combt (combs_buf, il_pos, 1);
|
||||
wordr0[2] = ix_create_combt (combs_buf, il_pos, 2);
|
||||
wordr0[3] = ix_create_combt (combs_buf, il_pos, 3);
|
||||
wordr1[0] = ix_create_combt (combs_buf, il_pos, 4);
|
||||
wordr1[1] = ix_create_combt (combs_buf, il_pos, 5);
|
||||
wordr1[2] = ix_create_combt (combs_buf, il_pos, 6);
|
||||
wordr1[3] = ix_create_combt (combs_buf, il_pos, 7);
|
||||
|
||||
if (combs_mode == COMBINATOR_MODE_BASE_LEFT)
|
||||
{
|
||||
switch_buffer_by_offset_le_VV (wordr0, wordr1, wordr2, wordr3, pw_l_len);
|
||||
}
|
||||
else
|
||||
{
|
||||
switch_buffer_by_offset_le_VV (wordl0, wordl1, wordl2, wordl3, pw_r_len);
|
||||
}
|
||||
|
||||
u32x w0[4];
|
||||
u32x w1[4];
|
||||
u32x w2[4];
|
||||
u32x w3[4];
|
||||
|
||||
w0[0] = wordl0[0] | wordr0[0];
|
||||
w0[1] = wordl0[1] | wordr0[1];
|
||||
w0[2] = wordl0[2] | wordr0[2];
|
||||
w0[3] = wordl0[3] | wordr0[3];
|
||||
w1[0] = wordl1[0] | wordr1[0];
|
||||
w1[1] = wordl1[1] | wordr1[1];
|
||||
w1[2] = wordl1[2] | wordr1[2];
|
||||
w1[3] = wordl1[3] | wordr1[3];
|
||||
w2[0] = wordl2[0] | wordr2[0];
|
||||
w2[1] = wordl2[1] | wordr2[1];
|
||||
w2[2] = wordl2[2] | wordr2[2];
|
||||
w2[3] = wordl2[3] | wordr2[3];
|
||||
w3[0] = wordl3[0] | wordr3[0];
|
||||
w3[1] = wordl3[1] | wordr3[1];
|
||||
w3[2] = wordl3[2] | wordr3[2];
|
||||
w3[3] = wordl3[3] | wordr3[3];
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0[0]);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = hl32_to_64 (w2[1], w2[0]);
|
||||
u64x a10 = hl32_to_64 (w2[3], w2[2]);
|
||||
u64x a11 = hl32_to_64 (w3[1], w3[0]);
|
||||
u64x a12 = hl32_to_64 (w3[3], w3[2]);
|
||||
u64x a13 = 0;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0x8000000000000000;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = h32_from_64 (a03);
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_S_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
__kernel void m17900_s08 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m17900_s16 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
@ -0,0 +1,695 @@
|
||||
/**
|
||||
* Author......: See docs/credits.txt
|
||||
* License.....: MIT
|
||||
*/
|
||||
|
||||
#define NEW_SIMD_CODE
|
||||
|
||||
#include "inc_vendor.cl"
|
||||
#include "inc_hash_constants.h"
|
||||
#include "inc_hash_functions.cl"
|
||||
#include "inc_types.cl"
|
||||
#include "inc_common.cl"
|
||||
#include "inc_simd.cl"
|
||||
|
||||
__constant u64a keccakf_rndc[24] =
|
||||
{
|
||||
0x0000000000000001, 0x0000000000008082, 0x800000000000808a,
|
||||
0x8000000080008000, 0x000000000000808b, 0x0000000080000001,
|
||||
0x8000000080008081, 0x8000000000008009, 0x000000000000008a,
|
||||
0x0000000000000088, 0x0000000080008009, 0x000000008000000a,
|
||||
0x000000008000808b, 0x800000000000008b, 0x8000000000008089,
|
||||
0x8000000000008003, 0x8000000000008002, 0x8000000000000080,
|
||||
0x000000000000800a, 0x800000008000000a, 0x8000000080008081,
|
||||
0x8000000000008080, 0x0000000080000001, 0x8000000080008008
|
||||
};
|
||||
|
||||
#ifndef KECCAK_ROUNDS
|
||||
#define KECCAK_ROUNDS 24
|
||||
#endif
|
||||
|
||||
DECLSPEC void m17500m (u32 *w0, u32 *w1, u32 *w2, u32 *w3, const u32 pw_len, __global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
u32 w0l = w0[0];
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
const u32x w0r = ix_create_bft (bfs_buf, il_pos);
|
||||
|
||||
const u32x w0lr = w0l | w0r;
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0lr);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = hl32_to_64 (w2[1], w2[0]);
|
||||
u64x a10 = hl32_to_64 (w2[3], w2[2]);
|
||||
u64x a11 = hl32_to_64 (w3[1], w3[0]);
|
||||
u64x a12 = hl32_to_64 (w3[3], w3[2]);
|
||||
u64x a13 = 0;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0x8000000000000000;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = h32_from_64 (a03);
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_M_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
DECLSPEC void m17500s (u32 *w0, u32 *w1, u32 *w2, u32 *w3, const u32 pw_len, __global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
const u64 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]
|
||||
};
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
u32 w0l = w0[0];
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
const u32x w0r = ix_create_bft (bfs_buf, il_pos);
|
||||
|
||||
const u32x w0lr = w0l | w0r;
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0lr);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = hl32_to_64 (w2[1], w2[0]);
|
||||
u64x a10 = hl32_to_64 (w2[3], w2[2]);
|
||||
u64x a11 = hl32_to_64 (w3[1], w3[0]);
|
||||
u64x a12 = hl32_to_64 (w3[3], w3[2]);
|
||||
u64x a13 = 0;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0x8000000000000000;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = h32_from_64 (a03);
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_S_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
__kernel void m17900_m04 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
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;
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* main
|
||||
*/
|
||||
|
||||
m17500m (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m17900_m08 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
u32 w1[4];
|
||||
|
||||
w1[0] = pws[gid].i[ 4];
|
||||
w1[1] = pws[gid].i[ 5];
|
||||
w1[2] = pws[gid].i[ 6];
|
||||
w1[3] = pws[gid].i[ 7];
|
||||
|
||||
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;
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* main
|
||||
*/
|
||||
|
||||
m17500m (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m17900_m16 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
u32 w1[4];
|
||||
|
||||
w1[0] = pws[gid].i[ 4];
|
||||
w1[1] = pws[gid].i[ 5];
|
||||
w1[2] = pws[gid].i[ 6];
|
||||
w1[3] = pws[gid].i[ 7];
|
||||
|
||||
u32 w2[4];
|
||||
|
||||
w2[0] = pws[gid].i[ 8];
|
||||
w2[1] = pws[gid].i[ 9];
|
||||
w2[2] = pws[gid].i[10];
|
||||
w2[3] = pws[gid].i[11];
|
||||
|
||||
u32 w3[4];
|
||||
|
||||
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
|
||||
*/
|
||||
|
||||
m17500m (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m17900_s04 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
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;
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* main
|
||||
*/
|
||||
|
||||
m17500s (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m17900_s08 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
u32 w1[4];
|
||||
|
||||
w1[0] = pws[gid].i[ 4];
|
||||
w1[1] = pws[gid].i[ 5];
|
||||
w1[2] = pws[gid].i[ 6];
|
||||
w1[3] = pws[gid].i[ 7];
|
||||
|
||||
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;
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* main
|
||||
*/
|
||||
|
||||
m17500s (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m17900_s16 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
u32 w1[4];
|
||||
|
||||
w1[0] = pws[gid].i[ 4];
|
||||
w1[1] = pws[gid].i[ 5];
|
||||
w1[2] = pws[gid].i[ 6];
|
||||
w1[3] = pws[gid].i[ 7];
|
||||
|
||||
u32 w2[4];
|
||||
|
||||
w2[0] = pws[gid].i[ 8];
|
||||
w2[1] = pws[gid].i[ 9];
|
||||
w2[2] = pws[gid].i[10];
|
||||
w2[3] = pws[gid].i[11];
|
||||
|
||||
u32 w3[4];
|
||||
|
||||
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
|
||||
*/
|
||||
|
||||
m17500s (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
@ -0,0 +1,479 @@
|
||||
/**
|
||||
* Author......: See docs/credits.txt
|
||||
* License.....: MIT
|
||||
*/
|
||||
|
||||
#define NEW_SIMD_CODE
|
||||
|
||||
#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_optimized.h"
|
||||
#include "inc_rp_optimized.cl"
|
||||
#include "inc_simd.cl"
|
||||
|
||||
__constant u64a keccakf_rndc[24] =
|
||||
{
|
||||
0x0000000000000001, 0x0000000000008082, 0x800000000000808a,
|
||||
0x8000000080008000, 0x000000000000808b, 0x0000000080000001,
|
||||
0x8000000080008081, 0x8000000000008009, 0x000000000000008a,
|
||||
0x0000000000000088, 0x0000000080008009, 0x000000008000000a,
|
||||
0x000000008000808b, 0x800000000000008b, 0x8000000000008089,
|
||||
0x8000000000008003, 0x8000000000008002, 0x8000000000000080,
|
||||
0x000000000000800a, 0x800000008000000a, 0x8000000080008081,
|
||||
0x8000000000008080, 0x0000000080000001, 0x8000000080008008
|
||||
};
|
||||
|
||||
#ifndef KECCAK_ROUNDS
|
||||
#define KECCAK_ROUNDS 24
|
||||
#endif
|
||||
|
||||
__kernel void m18000_m04 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 pw_buf0[4];
|
||||
u32 pw_buf1[4];
|
||||
|
||||
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];
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
u32x w0[4] = { 0 };
|
||||
u32x w1[4] = { 0 };
|
||||
u32x w2[4] = { 0 };
|
||||
u32x w3[4] = { 0 };
|
||||
|
||||
const u32x out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
|
||||
|
||||
append_0x01_2x4_VV (w0, w1, out_len);
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0[0]);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = 0;
|
||||
u64x a10 = 0;
|
||||
u64x a11 = 0;
|
||||
u64x a12 = 0;
|
||||
u64x a13 = 0x8000000000000000;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = h32_from_64 (a03);
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_M_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
__kernel void m18000_m08 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m18000_m16 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m18000_s04 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 pw_buf0[4];
|
||||
u32 pw_buf1[4];
|
||||
|
||||
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];
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* 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]
|
||||
};
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
u32x w0[4] = { 0 };
|
||||
u32x w1[4] = { 0 };
|
||||
u32x w2[4] = { 0 };
|
||||
u32x w3[4] = { 0 };
|
||||
|
||||
const u32x out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
|
||||
|
||||
append_0x01_2x4_VV (w0, w1, out_len);
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0[0]);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = 0;
|
||||
u64x a10 = 0;
|
||||
u64x a11 = 0;
|
||||
u64x a12 = 0;
|
||||
u64x a13 = 0x8000000000000000;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = h32_from_64 (a03);
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_S_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
__kernel void m18000_s08 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m18000_s16 (__global pw_t *pws, __constant const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
@ -0,0 +1,594 @@
|
||||
/**
|
||||
* Author......: See docs/credits.txt
|
||||
* License.....: MIT
|
||||
*/
|
||||
|
||||
#define NEW_SIMD_CODE
|
||||
|
||||
#include "inc_vendor.cl"
|
||||
#include "inc_hash_constants.h"
|
||||
#include "inc_hash_functions.cl"
|
||||
#include "inc_types.cl"
|
||||
#include "inc_common.cl"
|
||||
#include "inc_simd.cl"
|
||||
|
||||
__constant u64a keccakf_rndc[24] =
|
||||
{
|
||||
0x0000000000000001, 0x0000000000008082, 0x800000000000808a,
|
||||
0x8000000080008000, 0x000000000000808b, 0x0000000080000001,
|
||||
0x8000000080008081, 0x8000000000008009, 0x000000000000008a,
|
||||
0x0000000000000088, 0x0000000080008009, 0x000000008000000a,
|
||||
0x000000008000808b, 0x800000000000008b, 0x8000000000008089,
|
||||
0x8000000000008003, 0x8000000000008002, 0x8000000000000080,
|
||||
0x000000000000800a, 0x800000008000000a, 0x8000000080008081,
|
||||
0x8000000000008080, 0x0000000080000001, 0x8000000080008008
|
||||
};
|
||||
|
||||
#ifndef KECCAK_ROUNDS
|
||||
#define KECCAK_ROUNDS 24
|
||||
#endif
|
||||
|
||||
__kernel void m18000_m04 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 pw_buf0[4];
|
||||
u32 pw_buf1[4];
|
||||
|
||||
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];
|
||||
|
||||
const u32 pw_l_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
const u32x pw_r_len = pwlenx_create_combt (combs_buf, il_pos);
|
||||
|
||||
const u32x pw_len = pw_l_len + pw_r_len;
|
||||
|
||||
/**
|
||||
* concat password candidate
|
||||
*/
|
||||
|
||||
u32x wordl0[4] = { 0 };
|
||||
u32x wordl1[4] = { 0 };
|
||||
u32x wordl2[4] = { 0 };
|
||||
u32x wordl3[4] = { 0 };
|
||||
|
||||
wordl0[0] = pw_buf0[0];
|
||||
wordl0[1] = pw_buf0[1];
|
||||
wordl0[2] = pw_buf0[2];
|
||||
wordl0[3] = pw_buf0[3];
|
||||
wordl1[0] = pw_buf1[0];
|
||||
wordl1[1] = pw_buf1[1];
|
||||
wordl1[2] = pw_buf1[2];
|
||||
wordl1[3] = pw_buf1[3];
|
||||
|
||||
u32x wordr0[4] = { 0 };
|
||||
u32x wordr1[4] = { 0 };
|
||||
u32x wordr2[4] = { 0 };
|
||||
u32x wordr3[4] = { 0 };
|
||||
|
||||
wordr0[0] = ix_create_combt (combs_buf, il_pos, 0);
|
||||
wordr0[1] = ix_create_combt (combs_buf, il_pos, 1);
|
||||
wordr0[2] = ix_create_combt (combs_buf, il_pos, 2);
|
||||
wordr0[3] = ix_create_combt (combs_buf, il_pos, 3);
|
||||
wordr1[0] = ix_create_combt (combs_buf, il_pos, 4);
|
||||
wordr1[1] = ix_create_combt (combs_buf, il_pos, 5);
|
||||
wordr1[2] = ix_create_combt (combs_buf, il_pos, 6);
|
||||
wordr1[3] = ix_create_combt (combs_buf, il_pos, 7);
|
||||
|
||||
if (combs_mode == COMBINATOR_MODE_BASE_LEFT)
|
||||
{
|
||||
switch_buffer_by_offset_le_VV (wordr0, wordr1, wordr2, wordr3, pw_l_len);
|
||||
}
|
||||
else
|
||||
{
|
||||
switch_buffer_by_offset_le_VV (wordl0, wordl1, wordl2, wordl3, pw_r_len);
|
||||
}
|
||||
|
||||
u32x w0[4];
|
||||
u32x w1[4];
|
||||
u32x w2[4];
|
||||
u32x w3[4];
|
||||
|
||||
w0[0] = wordl0[0] | wordr0[0];
|
||||
w0[1] = wordl0[1] | wordr0[1];
|
||||
w0[2] = wordl0[2] | wordr0[2];
|
||||
w0[3] = wordl0[3] | wordr0[3];
|
||||
w1[0] = wordl1[0] | wordr1[0];
|
||||
w1[1] = wordl1[1] | wordr1[1];
|
||||
w1[2] = wordl1[2] | wordr1[2];
|
||||
w1[3] = wordl1[3] | wordr1[3];
|
||||
w2[0] = wordl2[0] | wordr2[0];
|
||||
w2[1] = wordl2[1] | wordr2[1];
|
||||
w2[2] = wordl2[2] | wordr2[2];
|
||||
w2[3] = wordl2[3] | wordr2[3];
|
||||
w3[0] = wordl3[0] | wordr3[0];
|
||||
w3[1] = wordl3[1] | wordr3[1];
|
||||
w3[2] = wordl3[2] | wordr3[2];
|
||||
w3[3] = wordl3[3] | wordr3[3];
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0[0]);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = hl32_to_64 (w2[1], w2[0]);
|
||||
u64x a10 = hl32_to_64 (w2[3], w2[2]);
|
||||
u64x a11 = hl32_to_64 (w3[1], w3[0]);
|
||||
u64x a12 = hl32_to_64 (w3[3], w3[2]);
|
||||
u64x a13 = 0x8000000000000000;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = h32_from_64 (a03);
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_M_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
__kernel void m18000_m08 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m18000_m16 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m18000_s04 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 pw_buf0[4];
|
||||
u32 pw_buf1[4];
|
||||
|
||||
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];
|
||||
|
||||
const u32 pw_l_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* 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]
|
||||
};
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
const u32x pw_r_len = pwlenx_create_combt (combs_buf, il_pos);
|
||||
|
||||
const u32x pw_len = pw_l_len + pw_r_len;
|
||||
|
||||
/**
|
||||
* concat password candidate
|
||||
*/
|
||||
|
||||
u32x wordl0[4] = { 0 };
|
||||
u32x wordl1[4] = { 0 };
|
||||
u32x wordl2[4] = { 0 };
|
||||
u32x wordl3[4] = { 0 };
|
||||
|
||||
wordl0[0] = pw_buf0[0];
|
||||
wordl0[1] = pw_buf0[1];
|
||||
wordl0[2] = pw_buf0[2];
|
||||
wordl0[3] = pw_buf0[3];
|
||||
wordl1[0] = pw_buf1[0];
|
||||
wordl1[1] = pw_buf1[1];
|
||||
wordl1[2] = pw_buf1[2];
|
||||
wordl1[3] = pw_buf1[3];
|
||||
|
||||
u32x wordr0[4] = { 0 };
|
||||
u32x wordr1[4] = { 0 };
|
||||
u32x wordr2[4] = { 0 };
|
||||
u32x wordr3[4] = { 0 };
|
||||
|
||||
wordr0[0] = ix_create_combt (combs_buf, il_pos, 0);
|
||||
wordr0[1] = ix_create_combt (combs_buf, il_pos, 1);
|
||||
wordr0[2] = ix_create_combt (combs_buf, il_pos, 2);
|
||||
wordr0[3] = ix_create_combt (combs_buf, il_pos, 3);
|
||||
wordr1[0] = ix_create_combt (combs_buf, il_pos, 4);
|
||||
wordr1[1] = ix_create_combt (combs_buf, il_pos, 5);
|
||||
wordr1[2] = ix_create_combt (combs_buf, il_pos, 6);
|
||||
wordr1[3] = ix_create_combt (combs_buf, il_pos, 7);
|
||||
|
||||
if (combs_mode == COMBINATOR_MODE_BASE_LEFT)
|
||||
{
|
||||
switch_buffer_by_offset_le_VV (wordr0, wordr1, wordr2, wordr3, pw_l_len);
|
||||
}
|
||||
else
|
||||
{
|
||||
switch_buffer_by_offset_le_VV (wordl0, wordl1, wordl2, wordl3, pw_r_len);
|
||||
}
|
||||
|
||||
u32x w0[4];
|
||||
u32x w1[4];
|
||||
u32x w2[4];
|
||||
u32x w3[4];
|
||||
|
||||
w0[0] = wordl0[0] | wordr0[0];
|
||||
w0[1] = wordl0[1] | wordr0[1];
|
||||
w0[2] = wordl0[2] | wordr0[2];
|
||||
w0[3] = wordl0[3] | wordr0[3];
|
||||
w1[0] = wordl1[0] | wordr1[0];
|
||||
w1[1] = wordl1[1] | wordr1[1];
|
||||
w1[2] = wordl1[2] | wordr1[2];
|
||||
w1[3] = wordl1[3] | wordr1[3];
|
||||
w2[0] = wordl2[0] | wordr2[0];
|
||||
w2[1] = wordl2[1] | wordr2[1];
|
||||
w2[2] = wordl2[2] | wordr2[2];
|
||||
w2[3] = wordl2[3] | wordr2[3];
|
||||
w3[0] = wordl3[0] | wordr3[0];
|
||||
w3[1] = wordl3[1] | wordr3[1];
|
||||
w3[2] = wordl3[2] | wordr3[2];
|
||||
w3[3] = wordl3[3] | wordr3[3];
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0[0]);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = hl32_to_64 (w2[1], w2[0]);
|
||||
u64x a10 = hl32_to_64 (w2[3], w2[2]);
|
||||
u64x a11 = hl32_to_64 (w3[1], w3[0]);
|
||||
u64x a12 = hl32_to_64 (w3[3], w3[2]);
|
||||
u64x a13 = 0x8000000000000000;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = h32_from_64 (a03);
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_S_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
__kernel void m18000_s08 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
|
||||
__kernel void m18000_s16 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
}
|
||||
@ -0,0 +1,695 @@
|
||||
/**
|
||||
* Author......: See docs/credits.txt
|
||||
* License.....: MIT
|
||||
*/
|
||||
|
||||
#define NEW_SIMD_CODE
|
||||
|
||||
#include "inc_vendor.cl"
|
||||
#include "inc_hash_constants.h"
|
||||
#include "inc_hash_functions.cl"
|
||||
#include "inc_types.cl"
|
||||
#include "inc_common.cl"
|
||||
#include "inc_simd.cl"
|
||||
|
||||
__constant u64a keccakf_rndc[24] =
|
||||
{
|
||||
0x0000000000000001, 0x0000000000008082, 0x800000000000808a,
|
||||
0x8000000080008000, 0x000000000000808b, 0x0000000080000001,
|
||||
0x8000000080008081, 0x8000000000008009, 0x000000000000008a,
|
||||
0x0000000000000088, 0x0000000080008009, 0x000000008000000a,
|
||||
0x000000008000808b, 0x800000000000008b, 0x8000000000008089,
|
||||
0x8000000000008003, 0x8000000000008002, 0x8000000000000080,
|
||||
0x000000000000800a, 0x800000008000000a, 0x8000000080008081,
|
||||
0x8000000000008080, 0x0000000080000001, 0x8000000080008008
|
||||
};
|
||||
|
||||
#ifndef KECCAK_ROUNDS
|
||||
#define KECCAK_ROUNDS 24
|
||||
#endif
|
||||
|
||||
DECLSPEC void m17600m (u32 *w0, u32 *w1, u32 *w2, u32 *w3, const u32 pw_len, __global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
const u64 lid = get_local_id (0);
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
u32 w0l = w0[0];
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
const u32x w0r = ix_create_bft (bfs_buf, il_pos);
|
||||
|
||||
const u32x w0lr = w0l | w0r;
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0lr);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = hl32_to_64 (w2[1], w2[0]);
|
||||
u64x a10 = hl32_to_64 (w2[3], w2[2]);
|
||||
u64x a11 = hl32_to_64 (w3[1], w3[0]);
|
||||
u64x a12 = hl32_to_64 (w3[3], w3[2]);
|
||||
u64x a13 = 0x8000000000000000;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = h32_from_64 (a03);
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_M_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
DECLSPEC void m17600s (u32 *w0, u32 *w1, u32 *w2, u32 *w3, const u32 pw_len, __global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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)
|
||||
{
|
||||
/**
|
||||
* modifier
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
const u64 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]
|
||||
};
|
||||
|
||||
/**
|
||||
* loop
|
||||
*/
|
||||
|
||||
u32 w0l = w0[0];
|
||||
|
||||
for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
|
||||
{
|
||||
const u32x w0r = ix_create_bft (bfs_buf, il_pos);
|
||||
|
||||
const u32x w0lr = w0l | w0r;
|
||||
|
||||
/**
|
||||
* Keccak
|
||||
*/
|
||||
|
||||
u64x a00 = hl32_to_64 (w0[1], w0lr);
|
||||
u64x a01 = hl32_to_64 (w0[3], w0[2]);
|
||||
u64x a02 = hl32_to_64 (w1[1], w1[0]);
|
||||
u64x a03 = hl32_to_64 (w1[3], w1[2]);
|
||||
u64x a04 = hl32_to_64 (w2[1], w2[0]);
|
||||
u64x a10 = hl32_to_64 (w2[3], w2[2]);
|
||||
u64x a11 = hl32_to_64 (w3[1], w3[0]);
|
||||
u64x a12 = hl32_to_64 (w3[3], w3[2]);
|
||||
u64x a13 = 0x8000000000000000;
|
||||
u64x a14 = 0;
|
||||
u64x a20 = 0;
|
||||
u64x a21 = 0;
|
||||
u64x a22 = 0;
|
||||
u64x a23 = 0;
|
||||
u64x a24 = 0;
|
||||
u64x a30 = 0;
|
||||
u64x a31 = 0;
|
||||
u64x a32 = 0;
|
||||
u64x a33 = 0;
|
||||
u64x a34 = 0;
|
||||
u64x a40 = 0;
|
||||
u64x a41 = 0;
|
||||
u64x a42 = 0;
|
||||
u64x a43 = 0;
|
||||
u64x a44 = 0;
|
||||
|
||||
#define Rho_Pi(ad,r) \
|
||||
bc0 = ad; \
|
||||
ad = rotl64 (t, r); \
|
||||
t = bc0; \
|
||||
|
||||
#ifdef _unroll
|
||||
#pragma unroll
|
||||
#endif
|
||||
for (int round = 0; round < KECCAK_ROUNDS - 1; round++)
|
||||
{
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t; a40 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a01 ^= t; a11 ^= t; a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t; a42 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a14 ^= t; a24 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
Rho_Pi (a40, 62);
|
||||
Rho_Pi (a24, 18);
|
||||
Rho_Pi (a42, 39);
|
||||
Rho_Pi (a14, 61);
|
||||
Rho_Pi (a11, 20);
|
||||
Rho_Pi (a01, 44);
|
||||
|
||||
// Chi
|
||||
|
||||
bc0 = a00; bc1 = a01; bc2 = a02; bc3 = a03; bc4 = a04;
|
||||
a00 ^= ~bc1 & bc2; a01 ^= ~bc2 & bc3; a02 ^= ~bc3 & bc4; a03 ^= ~bc4 & bc0; a04 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a10; bc1 = a11; bc2 = a12; bc3 = a13; bc4 = a14;
|
||||
a10 ^= ~bc1 & bc2; a11 ^= ~bc2 & bc3; a12 ^= ~bc3 & bc4; a13 ^= ~bc4 & bc0; a14 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a20; bc1 = a21; bc2 = a22; bc3 = a23; bc4 = a24;
|
||||
a20 ^= ~bc1 & bc2; a21 ^= ~bc2 & bc3; a22 ^= ~bc3 & bc4; a23 ^= ~bc4 & bc0; a24 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a30; bc1 = a31; bc2 = a32; bc3 = a33; bc4 = a34;
|
||||
a30 ^= ~bc1 & bc2; a31 ^= ~bc2 & bc3; a32 ^= ~bc3 & bc4; a33 ^= ~bc4 & bc0; a34 ^= ~bc0 & bc1;
|
||||
|
||||
bc0 = a40; bc1 = a41; bc2 = a42; bc3 = a43; bc4 = a44;
|
||||
a40 ^= ~bc1 & bc2; a41 ^= ~bc2 & bc3; a42 ^= ~bc3 & bc4; a43 ^= ~bc4 & bc0; a44 ^= ~bc0 & bc1;
|
||||
|
||||
// Iota
|
||||
|
||||
a00 ^= keccakf_rndc[round];
|
||||
}
|
||||
|
||||
// Theta
|
||||
|
||||
u64x bc0 = a00 ^ a10 ^ a20 ^ a30 ^ a40;
|
||||
u64x bc1 = a01 ^ a11 ^ a21 ^ a31 ^ a41;
|
||||
u64x bc2 = a02 ^ a12 ^ a22 ^ a32 ^ a42;
|
||||
u64x bc3 = a03 ^ a13 ^ a23 ^ a33 ^ a43;
|
||||
u64x bc4 = a04 ^ a14 ^ a24 ^ a34 ^ a44;
|
||||
|
||||
u64x t;
|
||||
|
||||
t = bc4 ^ rotl64 (bc1, 1); a00 ^= t; a10 ^= t; a20 ^= t; a30 ^= t;
|
||||
t = bc0 ^ rotl64 (bc2, 1); a21 ^= t; a31 ^= t; a41 ^= t;
|
||||
t = bc1 ^ rotl64 (bc3, 1); a02 ^= t; a12 ^= t; a22 ^= t; a32 ^= t;
|
||||
t = bc2 ^ rotl64 (bc4, 1); a03 ^= t; a13 ^= t; a23 ^= t; a33 ^= t; a43 ^= t;
|
||||
t = bc3 ^ rotl64 (bc0, 1); a04 ^= t; a34 ^= t; a44 ^= t;
|
||||
|
||||
// Rho Pi
|
||||
|
||||
t = a01;
|
||||
|
||||
Rho_Pi (a20, 1);
|
||||
Rho_Pi (a12, 3);
|
||||
Rho_Pi (a21, 6);
|
||||
Rho_Pi (a32, 10);
|
||||
Rho_Pi (a33, 15);
|
||||
Rho_Pi (a03, 21);
|
||||
Rho_Pi (a10, 28);
|
||||
Rho_Pi (a31, 36);
|
||||
Rho_Pi (a13, 45);
|
||||
Rho_Pi (a41, 55);
|
||||
Rho_Pi (a44, 2);
|
||||
Rho_Pi (a04, 14);
|
||||
Rho_Pi (a30, 27);
|
||||
Rho_Pi (a43, 41);
|
||||
Rho_Pi (a34, 56);
|
||||
Rho_Pi (a23, 8);
|
||||
Rho_Pi (a22, 25);
|
||||
Rho_Pi (a02, 43);
|
||||
|
||||
#undef Rho_Pi
|
||||
|
||||
bc0 = a00;
|
||||
bc2 = a02;
|
||||
bc3 = a03;
|
||||
bc4 = a04;
|
||||
|
||||
a02 ^= ~bc3 & bc4;
|
||||
a03 ^= ~bc4 & bc0;
|
||||
|
||||
const u32x r0 = l32_from_64 (a03);
|
||||
const u32x r1 = h32_from_64 (a03);
|
||||
const u32x r2 = l32_from_64 (a02);
|
||||
const u32x r3 = h32_from_64 (a02);
|
||||
|
||||
COMPARE_S_SIMD (r0, r1, r2, r3);
|
||||
}
|
||||
}
|
||||
|
||||
__kernel void m18000_m04 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
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;
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* main
|
||||
*/
|
||||
|
||||
m17600m (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m18000_m08 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
u32 w1[4];
|
||||
|
||||
w1[0] = pws[gid].i[ 4];
|
||||
w1[1] = pws[gid].i[ 5];
|
||||
w1[2] = pws[gid].i[ 6];
|
||||
w1[3] = pws[gid].i[ 7];
|
||||
|
||||
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;
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* main
|
||||
*/
|
||||
|
||||
m17600m (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m18000_m16 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
u32 w1[4];
|
||||
|
||||
w1[0] = pws[gid].i[ 4];
|
||||
w1[1] = pws[gid].i[ 5];
|
||||
w1[2] = pws[gid].i[ 6];
|
||||
w1[3] = pws[gid].i[ 7];
|
||||
|
||||
u32 w2[4];
|
||||
|
||||
w2[0] = pws[gid].i[ 8];
|
||||
w2[1] = pws[gid].i[ 9];
|
||||
w2[2] = pws[gid].i[10];
|
||||
w2[3] = pws[gid].i[11];
|
||||
|
||||
u32 w3[4];
|
||||
|
||||
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
|
||||
*/
|
||||
|
||||
m17600m (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m18000_s04 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
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;
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* main
|
||||
*/
|
||||
|
||||
m17600s (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m18000_s08 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
u32 w1[4];
|
||||
|
||||
w1[0] = pws[gid].i[ 4];
|
||||
w1[1] = pws[gid].i[ 5];
|
||||
w1[2] = pws[gid].i[ 6];
|
||||
w1[3] = pws[gid].i[ 7];
|
||||
|
||||
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;
|
||||
|
||||
const u32 pw_len = pws[gid].pw_len;
|
||||
|
||||
/**
|
||||
* main
|
||||
*/
|
||||
|
||||
m17600s (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
|
||||
__kernel void m18000_s16 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const 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 u64 gid_max)
|
||||
{
|
||||
/**
|
||||
* base
|
||||
*/
|
||||
|
||||
const u64 gid = get_global_id (0);
|
||||
|
||||
if (gid >= gid_max) return;
|
||||
|
||||
u32 w0[4];
|
||||
|
||||
w0[0] = pws[gid].i[ 0];
|
||||
w0[1] = pws[gid].i[ 1];
|
||||
w0[2] = pws[gid].i[ 2];
|
||||
w0[3] = pws[gid].i[ 3];
|
||||
|
||||
u32 w1[4];
|
||||
|
||||
w1[0] = pws[gid].i[ 4];
|
||||
w1[1] = pws[gid].i[ 5];
|
||||
w1[2] = pws[gid].i[ 6];
|
||||
w1[3] = pws[gid].i[ 7];
|
||||
|
||||
u32 w2[4];
|
||||
|
||||
w2[0] = pws[gid].i[ 8];
|
||||
w2[1] = pws[gid].i[ 9];
|
||||
w2[2] = pws[gid].i[10];
|
||||
w2[3] = pws[gid].i[11];
|
||||
|
||||
u32 w3[4];
|
||||
|
||||
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
|
||||
*/
|
||||
|
||||
m17600s (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_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset);
|
||||
}
|
||||
Loading…
Reference in new issue