GOSTSec
/
ccminer-gostd-lite
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity
Browse Source

whirlpool: restore old source code for SM 3.0 

SM 3.0 implementation need a manual define in whirlpool.cu...

alexis variant is 2x slower on SM3.0 (GT 740)
master 2.0-rc2
Tanguy Pruvot 8 years ago
parent
e388c11c02
commit
73f6720121
5 changed files with 2409 additions and 169 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Unified View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 2
      Makefile.am
  2. 2
      quark/cuda_skein512.cu
  3. 137
      x15/cuda_x15_whirlpool.cu
  4. 2365
      x15/cuda_x15_whirlpool_sm3.cu
  5. 72
      x15/whirlpool.cu

2
Makefile.am
Unescape View File

@ -67,7 +67,7 @@ ccminer_SOURCES = elist.h miner.h compat.h \
x11/cuda_x11_luffa512_Cubehash.cu x11/x11evo.cu x11/timetravel.cu \ x11/cuda_x11_luffa512_Cubehash.cu x11/x11evo.cu x11/timetravel.cu \
x13/x13.cu x13/cuda_x13_hamsi512.cu x13/cuda_x13_fugue512.cu \ x13/x13.cu x13/cuda_x13_hamsi512.cu x13/cuda_x13_fugue512.cu \
x15/x14.cu x15/x15.cu x15/cuda_x14_shabal512.cu x15/cuda_x15_whirlpool.cu \ x15/x14.cu x15/x15.cu x15/cuda_x14_shabal512.cu x15/cuda_x15_whirlpool.cu \
x15/whirlpool.cu \ x15/whirlpool.cu x15/cuda_x15_whirlpool_sm3.cu \
x17/x17.cu x17/hmq17.cu x17/cuda_x17_haval256.cu x17/cuda_x17_sha512.cu \ x17/x17.cu x17/hmq17.cu x17/cuda_x17_haval256.cu x17/cuda_x17_sha512.cu \
x11/c11.cu x11/s3.cu x11/sib.cu x11/veltor.cu x11/cuda_streebog.cu x11/c11.cu x11/s3.cu x11/sib.cu x11/veltor.cu x11/cuda_streebog.cu

2
quark/cuda_skein512.cu
Unescape View File

@ -1,4 +1,4 @@
/* SKEIN 64 and 80 based on Alexis Provost version */ /* SKEIN 64 and 80 based on Alexis Provos version */
#define TPB52 512 #define TPB52 512
#define TPB50 256 #define TPB50 256

137
x15/cuda_x15_whirlpool.cu
Unescape View File

@ -743,140 +743,3 @@ void x15_whirlpool_cpu_hash_64(int thr_id, uint32_t threads, uint32_t startNounc
x15_whirlpool_cpu_hash_64(thr_id, threads, d_hash); x15_whirlpool_cpu_hash_64(thr_id, threads, d_hash);
} }
#if 0
__global__ __launch_bounds__(TPB64,2)
void x15_whirlpool_gpu_hash_64_final(uint32_t threads,const uint64_t* __restrict__ g_hash, uint32_t* resNonce, const uint64_t target)
{
__shared__ uint2 sharedMemory[7][256];
if (threadIdx.x < 256) {
const uint2 tmp = __ldg((uint2*)&b0[threadIdx.x]);
sharedMemory[0][threadIdx.x] = tmp;
sharedMemory[1][threadIdx.x] = ROL8(tmp);
sharedMemory[2][threadIdx.x] = ROL16(tmp);
sharedMemory[3][threadIdx.x] = ROL24(tmp);
sharedMemory[4][threadIdx.x] = SWAPUINT2(tmp);
sharedMemory[5][threadIdx.x] = ROR24(tmp);
sharedMemory[6][threadIdx.x] = ROR16(tmp);
}
const uint32_t thread = (blockDim.x * blockIdx.x + threadIdx.x);
if (thread < threads){
uint2 hash[8], n[8], h[ 8], backup;
uint2 tmp[8] = {
{0xC0EE0B30,0x672990AF},{0x28282828,0x28282828},{0x28282828,0x28282828},{0x28282828,0x28282828},
{0x28282828,0x28282828},{0x28282828,0x28282828},{0x28282828,0x28282828},{0x28282828,0x28282828}
};
*(uint2x4*)&hash[ 0] = __ldg4((uint2x4*)&g_hash[(thread<<3) + 0]);
*(uint2x4*)&hash[ 4] = __ldg4((uint2x4*)&g_hash[(thread<<3) + 4]);
__syncthreads();
#pragma unroll 8
for(int i=0;i<8;i++)
n[i]=hash[i];
// __syncthreads();
tmp[ 0]^= d_ROUND_ELT(sharedMemory,n, 0, 7, 6, 5, 4, 3, 2, 1);
tmp[ 1]^= d_ROUND_ELT_LDG(sharedMemory,n, 1, 0, 7, 6, 5, 4, 3, 2);
tmp[ 2]^= d_ROUND_ELT(sharedMemory,n, 2, 1, 0, 7, 6, 5, 4, 3);
tmp[ 3]^= d_ROUND_ELT_LDG(sharedMemory,n, 3, 2, 1, 0, 7, 6, 5, 4);
tmp[ 4]^= d_ROUND_ELT(sharedMemory,n, 4, 3, 2, 1, 0, 7, 6, 5);
tmp[ 5]^= d_ROUND_ELT_LDG(sharedMemory,n, 5, 4, 3, 2, 1, 0, 7, 6);
tmp[ 6]^= d_ROUND_ELT(sharedMemory,n, 6, 5, 4, 3, 2, 1, 0, 7);
tmp[ 7]^= d_ROUND_ELT_LDG(sharedMemory,n, 7, 6, 5, 4, 3, 2, 1, 0);
for (int i=1; i <10; i++){
TRANSFER(n, tmp);
tmp[ 0] = d_ROUND_ELT1_LDG(sharedMemory,n, 0, 7, 6, 5, 4, 3, 2, 1, precomputed_round_key_64[(i-1)*8+0]);
tmp[ 1] = d_ROUND_ELT1( sharedMemory,n, 1, 0, 7, 6, 5, 4, 3, 2, precomputed_round_key_64[(i-1)*8+1]);
tmp[ 2] = d_ROUND_ELT1( sharedMemory,n, 2, 1, 0, 7, 6, 5, 4, 3, precomputed_round_key_64[(i-1)*8+2]);
tmp[ 3] = d_ROUND_ELT1_LDG(sharedMemory,n, 3, 2, 1, 0, 7, 6, 5, 4, precomputed_round_key_64[(i-1)*8+3]);
tmp[ 4] = d_ROUND_ELT1( sharedMemory,n, 4, 3, 2, 1, 0, 7, 6, 5, precomputed_round_key_64[(i-1)*8+4]);
tmp[ 5] = d_ROUND_ELT1( sharedMemory,n, 5, 4, 3, 2, 1, 0, 7, 6, precomputed_round_key_64[(i-1)*8+5]);
tmp[ 6] = d_ROUND_ELT1( sharedMemory,n, 6, 5, 4, 3, 2, 1, 0, 7, precomputed_round_key_64[(i-1)*8+6]);
tmp[ 7] = d_ROUND_ELT1_LDG(sharedMemory,n, 7, 6, 5, 4, 3, 2, 1, 0, precomputed_round_key_64[(i-1)*8+7]);
}
TRANSFER(h, tmp);
#pragma unroll 8
for (int i=0; i<8; i++)
h[i] = h[i] ^ hash[i];
#pragma unroll 6
for (int i=1; i<7; i++)
n[i]=vectorize(0);
n[0] = vectorize(0x80);
n[7] = vectorize(0x2000000000000);
#pragma unroll 8
for (int i=0; i < 8; i++) {
n[i] = n[i] ^ h[i];
}
backup = h[ 3];
// #pragma unroll 8
for (int i=0; i < 8; i++) {
tmp[ 0] = d_ROUND_ELT1(sharedMemory, h, 0, 7, 6, 5, 4, 3, 2, 1, InitVector_RC[i]);
tmp[ 1] = d_ROUND_ELT(sharedMemory, h, 1, 0, 7, 6, 5, 4, 3, 2);
tmp[ 2] = d_ROUND_ELT_LDG(sharedMemory, h, 2, 1, 0, 7, 6, 5, 4, 3);
tmp[ 3] = d_ROUND_ELT(sharedMemory, h, 3, 2, 1, 0, 7, 6, 5, 4);
tmp[ 4] = d_ROUND_ELT_LDG(sharedMemory, h, 4, 3, 2, 1, 0, 7, 6, 5);
tmp[ 5] = d_ROUND_ELT(sharedMemory, h, 5, 4, 3, 2, 1, 0, 7, 6);
tmp[ 6] = d_ROUND_ELT_LDG(sharedMemory, h, 6, 5, 4, 3, 2, 1, 0, 7);
tmp[ 7] = d_ROUND_ELT(sharedMemory, h, 7, 6, 5, 4, 3, 2, 1, 0);
TRANSFER(h, tmp);
tmp[ 0] = d_ROUND_ELT1(sharedMemory,n, 0, 7, 6, 5, 4, 3, 2, 1, tmp[0]);
tmp[ 1] = d_ROUND_ELT1(sharedMemory,n, 1, 0, 7, 6, 5, 4, 3, 2, tmp[1]);
tmp[ 2] = d_ROUND_ELT1_LDG(sharedMemory,n, 2, 1, 0, 7, 6, 5, 4, 3, tmp[2]);
tmp[ 3] = d_ROUND_ELT1(sharedMemory,n, 3, 2, 1, 0, 7, 6, 5, 4, tmp[3]);
tmp[ 4] = d_ROUND_ELT1(sharedMemory,n, 4, 3, 2, 1, 0, 7, 6, 5, tmp[4]);
tmp[ 5] = d_ROUND_ELT1(sharedMemory,n, 5, 4, 3, 2, 1, 0, 7, 6, tmp[5]);
tmp[ 6] = d_ROUND_ELT1(sharedMemory,n, 6, 5, 4, 3, 2, 1, 0, 7, tmp[6]);
tmp[ 7] = d_ROUND_ELT1_LDG(sharedMemory,n, 7, 6, 5, 4, 3, 2, 1, 0, tmp[7]);
TRANSFER(n, tmp);
}
tmp[ 0] = d_ROUND_ELT1(sharedMemory, h, 0, 7, 6, 5, 4, 3, 2, 1, InitVector_RC[8]);
tmp[ 1] = d_ROUND_ELT(sharedMemory, h, 1, 0, 7, 6, 5, 4, 3, 2);
tmp[ 2] = d_ROUND_ELT_LDG(sharedMemory, h, 2, 1, 0, 7, 6, 5, 4, 3);
tmp[ 3] = d_ROUND_ELT(sharedMemory, h, 3, 2, 1, 0, 7, 6, 5, 4);
tmp[ 4] = d_ROUND_ELT_LDG(sharedMemory, h, 4, 3, 2, 1, 0, 7, 6, 5);
tmp[ 5] = d_ROUND_ELT(sharedMemory, h, 5, 4, 3, 2, 1, 0, 7, 6);
tmp[ 6] = d_ROUND_ELT(sharedMemory, h, 6, 5, 4, 3, 2, 1, 0, 7);
tmp[ 7] = d_ROUND_ELT(sharedMemory, h, 7, 6, 5, 4, 3, 2, 1, 0);
TRANSFER(h, tmp);
tmp[ 0] = d_ROUND_ELT1(sharedMemory,n, 0, 7, 6, 5, 4, 3, 2, 1, tmp[0]);
tmp[ 1] = d_ROUND_ELT1(sharedMemory,n, 1, 0, 7, 6, 5, 4, 3, 2, tmp[1]);
tmp[ 2] = d_ROUND_ELT1(sharedMemory,n, 2, 1, 0, 7, 6, 5, 4, 3, tmp[2]);
tmp[ 3] = d_ROUND_ELT1(sharedMemory,n, 3, 2, 1, 0, 7, 6, 5, 4, tmp[3]);
tmp[ 4] = d_ROUND_ELT1(sharedMemory,n, 4, 3, 2, 1, 0, 7, 6, 5, tmp[4]);
tmp[ 5] = d_ROUND_ELT1(sharedMemory,n, 5, 4, 3, 2, 1, 0, 7, 6, tmp[5]);
tmp[ 6] = d_ROUND_ELT1_LDG(sharedMemory,n, 6, 5, 4, 3, 2, 1, 0, 7, tmp[6]);
tmp[ 7] = d_ROUND_ELT1(sharedMemory,n, 7, 6, 5, 4, 3, 2, 1, 0, tmp[7]);
n[ 3] = backup ^ d_ROUND_ELT(sharedMemory, h, 3, 2, 1, 0, 7, 6, 5, 4)
^ d_ROUND_ELT(sharedMemory,tmp, 3, 2, 1, 0, 7, 6, 5, 4);
if(devectorize(n[3]) <= target) {
uint32_t tmp = atomicExch(&resNonce[0], thread);
if (tmp != UINT32_MAX)
resNonce[1] = tmp;
}
}
}
extern void x15_whirlpool_cpu_hash_64_final(int thr_id, uint32_t threads, uint32_t *d_hash, uint32_t *d_resNonce, const uint64_t target)
{
dim3 grid((threads + TPB64-1) / TPB64);
dim3 block(TPB64);
x15_whirlpool_gpu_hash_64_final <<<grid, block>>> (threads, (uint64_t*)d_hash,d_resNonce,target);
}
#endif

2365
x15/cuda_x15_whirlpool_sm3.cu
View File

File diff suppressed because it is too large Load Diff

72
x15/whirlpool.cu
Unescape View File

@ -1,30 +1,33 @@
/* /*
* whirlpool routine (djm) * whirlpool routine
*/ */
extern "C" extern "C" {
{ #include <sph/sph_whirlpool.h>
#include "sph/sph_whirlpool.h" #include <miner.h>
#include "miner.h"
} }
#include "cuda_helper.h" #include <cuda_helper.h>
//static uint32_t *d_hash[MAX_GPUS]; //#define SM3_VARIANT
#ifdef SM3_VARIANT
static uint32_t *d_hash[MAX_GPUS];
extern void whirlpool512_init_sm3(int thr_id, uint32_t threads, int mode);
extern void whirlpool512_free_sm3(int thr_id);
extern void whirlpool512_setBlock_80_sm3(void *pdata, const void *ptarget);
extern void whirlpool512_hash_64_sm3(int thr_id, uint32_t threads, uint32_t startNounce, uint32_t *d_nonceVector, uint32_t *d_hash, int order);
extern void whirlpool512_hash_80_sm3(int thr_id, uint32_t threads, uint32_t startNounce, uint32_t *d_hash);
extern uint32_t whirlpool512_finalhash_64(int thr_id, uint32_t threads, uint32_t startNounce, uint32_t *d_nonceVector, uint32_t *d_hash, int order);
//#define _DEBUG
#define _DEBUG_PREFIX "whirl"
#include <cuda_debug.cuh>
#else
extern void x15_whirlpool_cpu_init(int thr_id, uint32_t threads, int mode); extern void x15_whirlpool_cpu_init(int thr_id, uint32_t threads, int mode);
extern void x15_whirlpool_cpu_free(int thr_id); extern void x15_whirlpool_cpu_free(int thr_id);
extern void whirlpool512_setBlock_80(void *pdata, const void *ptarget); extern void whirlpool512_setBlock_80(void *pdata, const void *ptarget);
//extern void x15_whirlpool_cpu_hash_64(int thr_id, uint32_t threads, uint32_t startNounce, uint32_t *d_nonceVector, uint32_t *d_hash, int order);
//extern void whirlpool512_cpu_hash_80(int thr_id, uint32_t threads, uint32_t startNounce, uint32_t *d_hash, int order);
//extern uint32_t whirlpool512_cpu_finalhash_64(int thr_id, uint32_t threads, uint32_t startNounce, uint32_t *d_nonceVector, uint32_t *d_hash, int order);
extern void whirlpool512_cpu_hash_80(int thr_id, uint32_t threads, uint32_t startNounce, uint32_t *resNonces, const uint64_t target); extern void whirlpool512_cpu_hash_80(int thr_id, uint32_t threads, uint32_t startNounce, uint32_t *resNonces, const uint64_t target);
#endif
//#define _DEBUG
#define _DEBUG_PREFIX "whirl"
#include "cuda_debug.cuh"
// CPU Hash function // CPU Hash function
extern "C" void wcoinhash(void *state, const void *input) extern "C" void wcoinhash(void *state, const void *input)
@ -90,10 +93,12 @@ extern "C" int scanhash_whirl(int thr_id, struct work* work, uint32_t max_nonce,
CUDA_LOG_ERROR(); CUDA_LOG_ERROR();
} }
gpulog(LOG_INFO, thr_id, "Intensity set to %g, %u cuda threads", throughput2intensity(throughput), throughput); gpulog(LOG_INFO, thr_id, "Intensity set to %g, %u cuda threads", throughput2intensity(throughput), throughput);
#ifdef SM3_VARIANT
//CUDA_SAFE_CALL(cudaMalloc(&d_hash[thr_id], (size_t) 64 * throughput)); CUDA_SAFE_CALL(cudaMalloc(&d_hash[thr_id], (size_t) 64 * throughput));
whirlpool512_init_sm3(thr_id, throughput, 1 /* old whirlpool */);
#else
x15_whirlpool_cpu_init(thr_id, throughput, 1 /* old whirlpool */); x15_whirlpool_cpu_init(thr_id, throughput, 1 /* old whirlpool */);
#endif
init[thr_id] = true; init[thr_id] = true;
} }
@ -101,22 +106,26 @@ extern "C" int scanhash_whirl(int thr_id, struct work* work, uint32_t max_nonce,
be32enc(&endiandata[k], pdata[k]); be32enc(&endiandata[k], pdata[k]);
} }
#ifdef SM3_VARIANT
whirlpool512_setBlock_80_sm3((void*)endiandata, ptarget);
#else
whirlpool512_setBlock_80((void*)endiandata, ptarget); whirlpool512_setBlock_80((void*)endiandata, ptarget);
#endif
do { do {
/* #ifdef SM3_VARIANT
int order = 0; int order = 1;
whirlpool512_cpu_hash_80(thr_id, throughput, pdata[19], d_hash[thr_id], order++); whirlpool512_hash_80_sm3(thr_id, throughput, pdata[19], d_hash[thr_id]);
TRACE64(" 80 :", d_hash); TRACE64(" 80 :", d_hash);
x15_whirlpool_cpu_hash_64(thr_id, throughput, pdata[19], NULL, d_hash[thr_id], order++); whirlpool512_hash_64_sm3(thr_id, throughput, pdata[19], NULL, d_hash[thr_id], order++);
TRACE64(" 64 :", d_hash); TRACE64(" 64 :", d_hash);
x15_whirlpool_cpu_hash_64(thr_id, throughput, pdata[19], NULL, d_hash[thr_id], order++); whirlpool512_hash_64_sm3(thr_id, throughput, pdata[19], NULL, d_hash[thr_id], order++);
TRACE64(" 64 :", d_hash); TRACE64(" 64 :", d_hash);
work->nonces[0] = whirlpool512_cpu_finalhash_64(thr_id, throughput, pdata[19], NULL, d_hash[thr_id], order++); work->nonces[0] = whirlpool512_finalhash_64(thr_id, throughput, pdata[19], NULL, d_hash[thr_id], order++);
*/ #else
*hashes_done = pdata[19] - first_nonce + throughput;
whirlpool512_cpu_hash_80(thr_id, throughput, pdata[19], work->nonces, *(uint64_t*)&ptarget[6]); whirlpool512_cpu_hash_80(thr_id, throughput, pdata[19], work->nonces, *(uint64_t*)&ptarget[6]);
#endif
*hashes_done = pdata[19] - first_nonce + throughput;
if (work->nonces[0] != UINT32_MAX && bench_algo < 0) if (work->nonces[0] != UINT32_MAX && bench_algo < 0)
{ {
@ -159,9 +168,12 @@ extern "C" void free_whirl(int thr_id)
cudaThreadSynchronize(); cudaThreadSynchronize();
//cudaFree(d_hash[thr_id]); #ifdef SM3_VARIANT
cudaFree(d_hash[thr_id]);
whirlpool512_free_sm3(thr_id);
#else
x15_whirlpool_cpu_free(thr_id); x15_whirlpool_cpu_free(thr_id);
#endif
init[thr_id] = false; init[thr_id] = false;
cudaDeviceSynchronize(); cudaDeviceSynchronize();

Write Preview
Loading…
Cancel
Save