GOSTcoin support for ccminer CUDA miner project, compatible with most nvidia cards
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

173 lines
4.8 KiB

/*
* qubit algorithm
*
*/
extern "C" {
#include "sph/sph_luffa.h"
#include "sph/sph_cubehash.h"
#include "sph/sph_shavite.h"
#include "sph/sph_simd.h"
#include "sph/sph_echo.h"
}
#include "miner.h"
#include "cuda_helper.h"
#include "x11/cuda_x11.h"
static uint32_t *d_hash[MAX_GPUS];
extern void qubit_luffa512_cpu_init(int thr_id, uint32_t threads);
extern void qubit_luffa512_cpu_setBlock_80(void *pdata);
extern void qubit_luffa512_cpu_hash_80(int thr_id, uint32_t threads, uint32_t startNounce, uint32_t *d_hash, int order);
extern "C" void qubithash(void *state, const void *input)
{
uint8_t _ALIGN(128) hash[64];
// luffa1-cubehash2-shavite3-simd4-echo5
sph_luffa512_context ctx_luffa;
sph_cubehash512_context ctx_cubehash;
sph_shavite512_context ctx_shavite;
sph_simd512_context ctx_simd;
sph_echo512_context ctx_echo;
sph_luffa512_init(&ctx_luffa);
sph_luffa512 (&ctx_luffa, input, 80);
sph_luffa512_close(&ctx_luffa, (void*) hash);
sph_cubehash512_init(&ctx_cubehash);
sph_cubehash512 (&ctx_cubehash, (const void*) hash, 64);
sph_cubehash512_close(&ctx_cubehash, (void*) hash);
sph_shavite512_init(&ctx_shavite);
sph_shavite512 (&ctx_shavite, (const void*) hash, 64);
sph_shavite512_close(&ctx_shavite, (void*) hash);
sph_simd512_init(&ctx_simd);
sph_simd512 (&ctx_simd, (const void*) hash, 64);
sph_simd512_close(&ctx_simd, (void*) hash);
sph_echo512_init(&ctx_echo);
sph_echo512 (&ctx_echo, (const void*) hash, 64);
sph_echo512_close(&ctx_echo, (void*) hash);
memcpy(state, hash, 32);
}
static bool init[MAX_GPUS] = { 0 };
extern "C" int scanhash_qubit(int thr_id, struct work* work, uint32_t max_nonce, unsigned long *hashes_done)
{
uint32_t _ALIGN(64) endiandata[20];
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
const uint32_t first_nonce = pdata[19];
uint32_t throughput = cuda_default_throughput(thr_id, 1U << 19); // 256*256*8
if (init[thr_id]) throughput = min(throughput, max_nonce - first_nonce);
if (opt_benchmark)
ptarget[7] = 0x007f;
if (!init[thr_id])
{
cudaSetDevice(device_map[thr_id]);
if (opt_cudaschedule == -1 && gpu_threads == 1) {
cudaDeviceReset();
// reduce cpu usage
cudaSetDeviceFlags(cudaDeviceScheduleBlockingSync);
CUDA_LOG_ERROR();
}
gpulog(LOG_INFO, thr_id, "Intensity set to %g, %u cuda threads", throughput2intensity(throughput), throughput);
qubit_luffa512_cpu_init(thr_id, throughput);
x11_cubehash512_cpu_init(thr_id, throughput);
x11_shavite512_cpu_init(thr_id, throughput);
x11_simd512_cpu_init(thr_id, throughput);
x11_echo512_cpu_init(thr_id, throughput);
CUDA_CALL_OR_RET_X(cudaMalloc(&d_hash[thr_id], (size_t) 64 * throughput), 0);
cuda_check_cpu_init(thr_id, throughput);
init[thr_id] = true;
}
for (int k=0; k < 20; k++)
be32enc(&endiandata[k], pdata[k]);
qubit_luffa512_cpu_setBlock_80((void*)endiandata);
cuda_check_cpu_setTarget(ptarget);
do {
int order = 0;
// Hash with CUDA
qubit_luffa512_cpu_hash_80(thr_id, throughput, pdata[19], d_hash[thr_id], order++);
x11_cubehash512_cpu_hash_64(thr_id, throughput, pdata[19], NULL, d_hash[thr_id], order++);
x11_shavite512_cpu_hash_64(thr_id, throughput, pdata[19], NULL, d_hash[thr_id], order++);
x11_simd512_cpu_hash_64(thr_id, throughput, pdata[19], NULL, d_hash[thr_id], order++);
x11_echo512_cpu_hash_64(thr_id, throughput, pdata[19], NULL, d_hash[thr_id], order++);
*hashes_done = pdata[19] - first_nonce + throughput;
uint32_t foundNonce = cuda_check_hash(thr_id, throughput, pdata[19], d_hash[thr_id]);
if (foundNonce != UINT32_MAX)
{
const uint32_t Htarg = ptarget[7];
uint32_t vhash64[8];
be32enc(&endiandata[19], foundNonce);
qubithash(vhash64, endiandata);
if (vhash64[7] <= Htarg && fulltest(vhash64, ptarget)) {
int res = 1;
uint32_t secNonce = cuda_check_hash_suppl(thr_id, throughput, pdata[19], d_hash[thr_id], 1);
work_set_target_ratio(work, vhash64);
if (secNonce != 0) {
be32enc(&endiandata[19], secNonce);
qubithash(vhash64, endiandata);
if (bn_hash_target_ratio(vhash64, ptarget) > work->shareratio[0])
work_set_target_ratio(work, vhash64);
pdata[21] = secNonce;
res++;
}
pdata[19] = foundNonce;
return res;
}
else {
applog(LOG_WARNING, "GPU #%d: result for nonce %08x does not validate on CPU!", device_map[thr_id], foundNonce);
}
}
if ((uint64_t) throughput + pdata[19] >= max_nonce) {
pdata[19] = max_nonce;
break;
}
pdata[19] += throughput;
} while (!work_restart[thr_id].restart);
*hashes_done = pdata[19] - first_nonce;
return 0;
}
// cleanup
extern "C" void free_qubit(int thr_id)
{
if (!init[thr_id])
return;
cudaThreadSynchronize();
cudaFree(d_hash[thr_id]);
x11_simd512_cpu_free(thr_id);
cuda_check_cpu_free(thr_id);
init[thr_id] = false;
cudaDeviceSynchronize();
}