arno/hashcat
1
0
Fork 0
mirror of https://github.com/hashcat/hashcat.git synced 2025-07-31 19:08:43 +00:00
Code Issues Releases Wiki Activity

Fix hc_bytealign_be() declaration for CUDA: use u32x instead of u32

Browse Source 
Add hc_uint4_t to SCRYPT to work-around Intel OpenCL alignment bug.
Align large buffers (V1-V4) manually to 1k-byte boundaries.
Replace uint4 xor operator with xor_uint4() function.
This commit is contained in:
Jens Steube 2025-07-16 12:43:11 +02:00
parent dd02267bf2
commit 77b5c65e52
3 changed files with 45 additions and 30 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
OpenCL/inc_common.cl
View File

@ -1887,7 +1887,7 @@ DECLSPEC u32 hc_bytealign_S (const u32 a, const u32 b, const int c)
return r; return r;
} }
#else #else
DECLSPEC u32 hc_bytealign_be (const u32 a, const u32 b, const int c) DECLSPEC u32 hc_bytealign_be (const u32x a, const u32x b, const int c)
{ {
const int c_mod_4 = c & 3; const int c_mod_4 = c & 3;

50
OpenCL/inc_hash_scrypt.cl
View File

@ -9,6 +9,18 @@
#include "inc_common.h" #include "inc_common.h"
#include "inc_hash_scrypt.h" #include "inc_hash_scrypt.h"
DECLSPEC hc_uint4_t xor_uint4 (const hc_uint4_t a, const hc_uint4_t b)
{
hc_uint4_t r;
r.x = a.x ^ b.x;
r.y = a.y ^ b.y;
r.z = a.z ^ b.z;
r.w = a.w ^ b.w;
return r;
}
#if SCRYPT_R > 1 #if SCRYPT_R > 1
DECLSPEC void scrypt_shuffle (PRIVATE_AS u32 *TI) DECLSPEC void scrypt_shuffle (PRIVATE_AS u32 *TI)
{ {
@ -144,27 +156,27 @@ DECLSPEC void scrypt_smix_init (GLOBAL_AS u32 *P, PRIVATE_AS u32 *X, GLOBAL_AS v
const u32 xd4 = bid / 4; const u32 xd4 = bid / 4;
const u32 xm4 = bid & 3; const u32 xm4 = bid & 3;
PRIVATE_AS uint4 *X4 = (PRIVATE_AS uint4 *) X; PRIVATE_AS hc_uint4_t *X4 = (PRIVATE_AS hc_uint4_t *) X;
GLOBAL_AS uint4 *V; GLOBAL_AS hc_uint4_t *V;
switch (xm4) switch (xm4)
{ {
case 0: V = (GLOBAL_AS uint4 *) V0; break; case 0: V = (GLOBAL_AS hc_uint4_t *) ALIGN_PTR_1k (V0); break;
case 1: V = (GLOBAL_AS uint4 *) V1; break; case 1: V = (GLOBAL_AS hc_uint4_t *) ALIGN_PTR_1k (V1); break;
case 2: V = (GLOBAL_AS uint4 *) V2; break; case 2: V = (GLOBAL_AS hc_uint4_t *) ALIGN_PTR_1k (V2); break;
case 3: V = (GLOBAL_AS uint4 *) V3; break; case 3: V = (GLOBAL_AS hc_uint4_t *) ALIGN_PTR_1k (V3); break;
} }
GLOBAL_AS uint4 *Vx = V + (xd4 * lsz * ySIZE * zSIZE) + (lid * ySIZE * zSIZE); GLOBAL_AS hc_uint4_t *Vx = V + (xd4 * lsz * ySIZE * zSIZE) + (lid * ySIZE * zSIZE);
for (u32 i = 0; i < STATE_CNT4; i++) X[i] = P[i]; for (u32 i = 0; i < STATE_CNT4; i++) X[i] = P[i];
for (u32 y = 0; y < ySIZE; y++) for (u32 y = 0; y < ySIZE; y++)
{ {
GLOBAL_AS uint4 *Vxx = Vx + (y * zSIZE); GLOBAL_AS hc_uint4_t *Vxx = Vx + (y * zSIZE);
for (u32 z = 0; z < zSIZE; z++) *Vxx++ = X4[z]; for (u32 z = 0; z < zSIZE; z++) Vxx[z] = X4[z];
for (u32 i = 0; i < (1 << SCRYPT_TMTO); i++) for (u32 i = 0; i < (1 << SCRYPT_TMTO); i++)
{ {
@ -187,20 +199,20 @@ DECLSPEC void scrypt_smix_loop (GLOBAL_AS u32 *P, PRIVATE_AS u32 *X, PRIVATE_AS
const u32 xd4 = bid / 4; const u32 xd4 = bid / 4;
const u32 xm4 = bid & 3; const u32 xm4 = bid & 3;
PRIVATE_AS uint4 *X4 = (PRIVATE_AS uint4 *) X; PRIVATE_AS hc_uint4_t *X4 = (PRIVATE_AS hc_uint4_t *) X;
PRIVATE_AS uint4 *T4 = (PRIVATE_AS uint4 *) T; PRIVATE_AS hc_uint4_t *T4 = (PRIVATE_AS hc_uint4_t *) T;
GLOBAL_AS uint4 *V; GLOBAL_AS hc_uint4_t *V;
switch (xm4) switch (xm4)
{ {
case 0: V = (GLOBAL_AS uint4 *) V0; break; case 0: V = (GLOBAL_AS hc_uint4_t *) ALIGN_PTR_1k (V0); break;
case 1: V = (GLOBAL_AS uint4 *) V1; break; case 1: V = (GLOBAL_AS hc_uint4_t *) ALIGN_PTR_1k (V1); break;
case 2: V = (GLOBAL_AS uint4 *) V2; break; case 2: V = (GLOBAL_AS hc_uint4_t *) ALIGN_PTR_1k (V2); break;
case 3: V = (GLOBAL_AS uint4 *) V3; break; case 3: V = (GLOBAL_AS hc_uint4_t *) ALIGN_PTR_1k (V3); break;
} }
GLOBAL_AS uint4 *Vx = V + (xd4 * lsz * ySIZE * zSIZE) + (lid * ySIZE * zSIZE); GLOBAL_AS hc_uint4_t *Vx = V + (xd4 * lsz * ySIZE * zSIZE) + (lid * ySIZE * zSIZE);
for (u32 i = 0; i < STATE_CNT4; i++) X[i] = P[i]; for (u32 i = 0; i < STATE_CNT4; i++) X[i] = P[i];
@ -216,7 +228,7 @@ DECLSPEC void scrypt_smix_loop (GLOBAL_AS u32 *P, PRIVATE_AS u32 *X, PRIVATE_AS
const u32 km = k - (y << SCRYPT_TMTO); const u32 km = k - (y << SCRYPT_TMTO);
GLOBAL_AS uint4 *Vxx = Vx + (y * zSIZE); GLOBAL_AS hc_uint4_t *Vxx = Vx + (y * zSIZE);
for (u32 z = 0; z < zSIZE; z++) T4[z] = *Vxx++; for (u32 z = 0; z < zSIZE; z++) T4[z] = *Vxx++;
@ -229,7 +241,7 @@ DECLSPEC void scrypt_smix_loop (GLOBAL_AS u32 *P, PRIVATE_AS u32 *X, PRIVATE_AS
#endif #endif
} }
for (u32 z = 0; z < zSIZE; z++) X4[z] = X4[z] ^ T4[z]; for (u32 z = 0; z < zSIZE; z++) X4[z] = xor_uint4 (X4[z], T4[z]);
salsa_r (X); salsa_r (X);

23
OpenCL/inc_hash_scrypt.h
View File

@ -25,24 +25,27 @@
#define SALSA_CNT4 (SALSA_SZ / 4) #define SALSA_CNT4 (SALSA_SZ / 4)
#define SALSA_CNT44 ((SALSA_SZ / 4) / 4) #define SALSA_CNT44 ((SALSA_SZ / 4) / 4)
//#define VIDX(bid4,lsz,lid,ySIZE,zSIZE,y,z) (((bid4) * (lsz) * (ySIZE) * (zSIZE)) + ((lid) * (ySIZE) * (zSIZE)) + ((y) * (zSIZE)) + (z)) // should be safe, because in backend.c we use:
// u64 size_extra_buffer1 = 4096;
// size_extra_buffer1 += base_chunk_size;
#define ALIGN_PTR_1k(p) ((GLOBAL_AS hc_uint4_t *) (((u64) (p) + 1023) & ~1023UL))
#if defined IS_CUDA #if defined IS_INTEL_SDK
DECLSPEC uint4 operator ^ (const uint4 a, const uint4 b) typedef struct
{ {
uint4 r; u32 x, y, z, w;
r.x = a.x ^ b.x; } hc_uint4_t;
r.y = a.y ^ b.y;
r.z = a.z ^ b.z;
r.w = a.w ^ b.w;
return r; #else
}
typedef uint4 hc_uint4_t;
#endif #endif
DECLSPEC hc_uint4_t xor_uint4 (const hc_uint4_t a, const hc_uint4_t b);
typedef struct typedef struct
{ {
#ifndef SCRYPT_TMP_ELEM #ifndef SCRYPT_TMP_ELEM