int scanhash_scrypt_jane(int thr_id, uint32_t *pdata,
const uint32_t *ptarget,
uint32_t max_nonce, unsigned long *hashes_done)
{
const uint32_t Htarg = ptarget[7];
int Nfactor = GetNfactor(bswap_32x4(pdata[17]));
if (Nfactor > scrypt_maxN) {
scrypt_fatal_error("scrypt: N out of range");
}
N = (1 << (Nfactor + 1));
parallel = 0;
int throughput = cuda_throughput(thr_id);
uint32_t *data[2] = { new uint32_t[20*throughput], new uint32_t[20*throughput] };
uint32_t *hash = new uint32_t[8*throughput];
uint32_t n = pdata[19] - 1;
// int i;
#if !defined(SCRYPT_TEST)
static int power_on_self_test = 0;
if (!power_on_self_test) {
power_on_self_test = 1;
if (!scrypt_power_on_self_test())
scrypt_fatal_error("scrypt: power on self test failed");
}
#endif
/* byte swap pdata into data[0]/[1] arrays */
for (int k=0; k<2; ++k) {
for(int z=0;z<20;z++) data[k][z] = bswap_32x4(pdata[z]);
for(int i=1;i<throughput;++i) memcpy(&data[k][20*i], &data[k][0], 20*sizeof(uint32_t));
}
scrypt_aligned_alloc Xbuf[2] = { scrypt_alloc(128 * throughput), scrypt_alloc(128 * throughput) };
scrypt_aligned_alloc Vbuf = scrypt_alloc((uint64_t)N * 128);
scrypt_aligned_alloc Ybuf = scrypt_alloc(128);
uint32_t nonce[2];
uint32_t* cuda_X[2] = { cuda_transferbuffer(thr_id,0), cuda_transferbuffer(thr_id,1) };
#if !defined(SCRYPT_CHOOSE_COMPILETIME)
scrypt_ROMixfn scrypt_ROMix = scrypt_getROMix();
#endif
int cur = 0, nxt = 1;
nonce[cur] = n+1;
for(int i=0;i<throughput;++i) {
uint32_t tmp_nonce = ++n;
data[cur][20*i + 19] = bswap_32x4(tmp_nonce);
}
/* 1: X = PBKDF2(password, salt) */
for(int i=0;i<throughput;++i)
scrypt_pbkdf2_1((unsigned char *)&data[cur][20*i], 80, (unsigned char *)&data[cur][20*i], 80, Xbuf[cur].ptr + 128 * i, 128);
/* 2: X = ROMix(X) in CUDA */
memcpy(cuda_X[cur], Xbuf[cur].ptr, 128 * throughput);
cuda_scrypt_HtoD(thr_id, cuda_X[cur], cur);
cuda_scrypt_serialize(thr_id, cur);
cuda_scrypt_core(thr_id, cur, N);
cuda_scrypt_done(thr_id, cur);
cuda_scrypt_DtoH(thr_id, cuda_X[cur], cur);
cuda_scrypt_flush(thr_id, cur);
do {
nonce[nxt] = n+1;
for(int i=0;i<throughput;++i) {
uint32_t tmp_nonce = ++n;
data[nxt][20*i + 19] = bswap_32x4(tmp_nonce);
}
/* 1: X = PBKDF2(password, salt) */
for(int i=0;i<throughput;++i)
scrypt_pbkdf2_1((unsigned char *)&data[nxt][20*i], 80, (unsigned char *)&data[nxt][20*i], 80, Xbuf[nxt].ptr + 128 * i, 128);
/* 2: X = ROMix(X) in CUDA */
memcpy(cuda_X[nxt], Xbuf[nxt].ptr, 128 * throughput);
cuda_scrypt_HtoD(thr_id, cuda_X[nxt], nxt);
cuda_scrypt_serialize(thr_id, nxt);
cuda_scrypt_core(thr_id, nxt, N);
cuda_scrypt_done(thr_id, nxt);
cuda_scrypt_DtoH(thr_id, cuda_X[nxt], nxt);
cuda_scrypt_flush(thr_id, nxt);
cuda_scrypt_sync(thr_id, cur);
#define VERIFY_ALL 0
#if VERIFY_ALL
{
/* 2: X = ROMix(X) */
for(int i=0;i<throughput;++i)
scrypt_ROMix_1((scrypt_mix_word_t *)(Xbuf[cur].ptr + 128 * i), (scrypt_mix_word_t *)Ybuf.ptr, (scrypt_mix_word_t *)Vbuf.ptr, N);
unsigned int err = 0;
for(int i=0;i<throughput;++i) {
unsigned char *ref = (Xbuf[cur].ptr + 128 * i);
unsigned char *dat = (unsigned char*)(cuda_X[cur] + 32 * i);
if (memcmp(ref, dat, 128) != 0)
{
err++;
#if 0
uint32_t *ref32 = (uint32_t*) ref;
uint32_t *dat32 = (uint32_t*) dat;
for (int j=0; j<32; ++j) {
if (ref32[j] != dat32[j])
fprintf(stderr, "ref32[i=%d][j=%d] = $%08x / $%08x\n", i, j, ref32[j], dat32[j]);
}
#endif
}
}
if (err > 0) fprintf(stderr, "%d out of %d hashes differ.\n", err, throughput);
}
#endif
memcpy(Xbuf[cur].ptr, cuda_X[cur], 128 * throughput);
/* 3: Out = PBKDF2(password, X) */
for(int i=0;i<throughput;++i)
scrypt_pbkdf2_1((unsigned char *)&data[cur][20*i], 80, Xbuf[cur].ptr + 128 * i, 128, (unsigned char *)&hash[8*i], 32);
for(int i=0;i<throughput;++i) {
volatile unsigned char *hashc = (unsigned char *)&hash[8*i];
if (hash[8*i+7] <= Htarg && fulltest(&hash[8*i], ptarget)) {
uint32_t tmp_nonce = nonce[cur]+i;
uint32_t thash[8], tdata[20];
for(int z=0;z<20;z++) tdata[z] = bswap_32x4(pdata[z]);
tdata[19] = bswap_32x4(tmp_nonce);
scrypt_pbkdf2_1((unsigned char *)tdata, 80, (unsigned char *)tdata, 80, Xbuf[cur].ptr + 128 * i, 128);
scrypt_ROMix_1((scrypt_mix_word_t *)(Xbuf[cur].ptr + 128 * i), (scrypt_mix_word_t *)(Ybuf.ptr), (scrypt_mix_word_t *)(Vbuf.ptr), N);
scrypt_pbkdf2_1((unsigned char *)tdata, 80, Xbuf[cur].ptr + 128 * i, 128, (unsigned char *)thash, 32);
if (memcmp(thash, &hash[8*i], 32) == 0)
{
*hashes_done = (n-throughput) - pdata[19] + 1;
pdata[19] = tmp_nonce;
scrypt_free(&Vbuf);
scrypt_free(&Ybuf);
scrypt_free(&Xbuf[0]); scrypt_free(&Xbuf[1]);
delete[] data[0]; delete[] data[1];
delete[] hash;
return 1;
}
else
{
applog(LOG_INFO, "GPU #%d: %s result does not validate on CPU (i=%d, s=%d)!", device_map[thr_id], device_name[thr_id], i, cur);
}
}
}
cur = (cur+1)&1; nxt = (nxt+1)&1;
} while ((n-throughput) < max_nonce && !work_restart[thr_id].restart);
scrypt_free(&Vbuf);
scrypt_free(&Ybuf);
scrypt_free(&Xbuf[0]); scrypt_free(&Xbuf[1]);
delete[] data[0]; delete[] data[1];
delete[] hash;
*hashes_done = (n-throughput) - pdata[19] + 1;
pdata[19] = (n-throughput);
return 0;
}
int scanhash_scrypt_jane(int thr_id, uint32_t *pdata,
const uint32_t *ptarget,
uint32_t max_nonce, struct timeval *tv_start, struct timeval *tv_end, unsigned long *hashes_done)
{
const uint32_t Htarg = ptarget[7];
static int s_Nfactor = 0;
if (s_Nfactor == 0 && strlen(jane_params) > 0)
applog(LOG_INFO, "Given scrypt-jane parameters: %s", jane_params);
int Nfactor = GetNfactor(bswap_32x4(pdata[17]));
if (Nfactor > scrypt_maxN) {
scrypt_fatal_error("scrypt: N out of range");
}
if (Nfactor != s_Nfactor)
{
// all of this isn't very thread-safe...
N = (1 << (Nfactor + 1));
applog(LOG_INFO, "Nfactor is %d (N=%d)!", Nfactor, N);
if (s_Nfactor != 0) {
// handle N-factor increase at runtime
// by adjusting the lookup_gap by factor 2
if (s_Nfactor == Nfactor-1)
for (int i=0; i < 8; ++i)
device_lookup_gap[i] *= 2;
}
s_Nfactor = Nfactor;
}
int throughput = cuda_throughput(thr_id);
if(throughput == 0)
return -1;
gettimeofday(tv_start, NULL);
uint32_t *data[2] = { new uint32_t[20*throughput], new uint32_t[20*throughput] };
uint32_t* hash[2] = { cuda_hashbuffer(thr_id,0), cuda_hashbuffer(thr_id,1) };
uint32_t n = pdata[19];
/* byte swap pdata into data[0]/[1] arrays */
for (int k=0; k<2; ++k) {
for(int z=0;z<20;z++) data[k][z] = bswap_32x4(pdata[z]);
for(int i=1;i<throughput;++i) memcpy(&data[k][20*i], &data[k][0], 20*sizeof(uint32_t));
}
if (parallel == 2) prepare_keccak512(thr_id, pdata);
scrypt_aligned_alloc Xbuf[2] = { scrypt_alloc(128 * throughput), scrypt_alloc(128 * throughput) };
scrypt_aligned_alloc Vbuf = scrypt_alloc((uint64_t)N * 128);
scrypt_aligned_alloc Ybuf = scrypt_alloc(128);
uint32_t nonce[2];
uint32_t* cuda_X[2] = { cuda_transferbuffer(thr_id,0), cuda_transferbuffer(thr_id,1) };
#if !defined(SCRYPT_CHOOSE_COMPILETIME)
scrypt_ROMixfn scrypt_ROMix = scrypt_getROMix();
#endif
int cur = 0, nxt = 1;
int iteration = 0;
do {
nonce[nxt] = n;
if (parallel < 2)
{
for(int i=0;i<throughput;++i) {
uint32_t tmp_nonce = n++;
data[nxt][20*i + 19] = bswap_32x4(tmp_nonce);
}
for(int i=0;i<throughput;++i)
scrypt_pbkdf2_1((unsigned char *)&data[nxt][20*i], 80, (unsigned char *)&data[nxt][20*i], 80, Xbuf[nxt].ptr + 128 * i, 128);
memcpy(cuda_X[nxt], Xbuf[nxt].ptr, 128 * throughput);
cuda_scrypt_serialize(thr_id, nxt);
cuda_scrypt_HtoD(thr_id, cuda_X[nxt], nxt);
cuda_scrypt_core(thr_id, nxt, N);
cuda_scrypt_done(thr_id, nxt);
cuda_scrypt_DtoH(thr_id, cuda_X[nxt], nxt, false);
cuda_scrypt_flush(thr_id, nxt);
if(!cuda_scrypt_sync(thr_id, cur))
{
return -1;
}
memcpy(Xbuf[cur].ptr, cuda_X[cur], 128 * throughput);
for(int i=0;i<throughput;++i)
scrypt_pbkdf2_1((unsigned char *)&data[cur][20*i], 80, Xbuf[cur].ptr + 128 * i, 128, (unsigned char *)(&hash[cur][8*i]), 32);
#define VERIFY_ALL 0
#if VERIFY_ALL
{
/* 2: X = ROMix(X) */
for(int i=0;i<throughput;++i)
scrypt_ROMix_1((scrypt_mix_word_t *)(Xbuf[cur].ptr + 128 * i), (scrypt_mix_word_t *)Ybuf.ptr, (scrypt_mix_word_t *)Vbuf.ptr, N);
unsigned int err = 0;
for(int i=0;i<throughput;++i) {
unsigned char *ref = (Xbuf[cur].ptr + 128 * i);
unsigned char *dat = (unsigned char*)(cuda_X[cur] + 32 * i);
if (memcmp(ref, dat, 128) != 0)
{
err++;
#if 0
uint32_t *ref32 = (uint32_t*) ref;
uint32_t *dat32 = (uint32_t*) dat;
for (int j=0; j<32; ++j) {
if (ref32[j] != dat32[j])
fprintf(stderr, "ref32[i=%d][j=%d] = $%08x / $%08x\n", i, j, ref32[j], dat32[j]);
}
#endif
}
}
if (err > 0) fprintf(stderr, "%d out of %d hashes differ.\n", err, throughput);
}
#endif
} else {
n += throughput;
cuda_scrypt_serialize(thr_id, nxt);
pre_keccak512(thr_id, nxt, nonce[nxt], throughput);
cuda_scrypt_core(thr_id, nxt, N);
cuda_scrypt_flush(thr_id, nxt);
post_keccak512(thr_id, nxt, nonce[nxt], throughput);
cuda_scrypt_done(thr_id, nxt);
cuda_scrypt_DtoH(thr_id, hash[nxt], nxt, true);
if(!cuda_scrypt_sync(thr_id, cur))
{
return -1;
}
}
if(iteration > 0)
{
for(int i=0;i<throughput;++i) {
volatile unsigned char *hashc = (unsigned char *)(&hash[cur][8*i]);
if (hash[cur][8*i+7] <= Htarg && fulltest(&hash[cur][8*i], ptarget)) {
uint32_t tmp_nonce = nonce[cur]+i;
uint32_t thash[8], tdata[20];
for(int z=0;z<20;z++) tdata[z] = bswap_32x4(pdata[z]);
tdata[19] = bswap_32x4(tmp_nonce);
scrypt_pbkdf2_1((unsigned char *)tdata, 80, (unsigned char *)tdata, 80, Xbuf[cur].ptr + 128 * i, 128);
scrypt_ROMix_1((scrypt_mix_word_t *)(Xbuf[cur].ptr + 128 * i), (scrypt_mix_word_t *)(Ybuf.ptr), (scrypt_mix_word_t *)(Vbuf.ptr), N);
scrypt_pbkdf2_1((unsigned char *)tdata, 80, Xbuf[cur].ptr + 128 * i, 128, (unsigned char *)thash, 32);
if (memcmp(thash, &hash[cur][8*i], 32) == 0)
{
//applog(LOG_INFO, "GPU #%d: %s result validates on CPU.", device_map[thr_id], device_name[thr_id]);
*hashes_done = n - pdata[19];
pdata[19] = tmp_nonce;
scrypt_free(&Vbuf);
scrypt_free(&Ybuf);
scrypt_free(&Xbuf[0]); scrypt_free(&Xbuf[1]);
delete[] data[0]; delete[] data[1];
gettimeofday(tv_end, NULL);
return 1;
}
else
{
applog(LOG_INFO, "GPU #%d: %s result does not validate on CPU (i=%d, s=%d)!", device_map[thr_id], device_name[thr_id], i, cur);
}
}
}
}
cur = (cur+1)&1;
nxt = (nxt+1)&1;
++iteration;
} while (n <= max_nonce && !work_restart[thr_id].restart);
scrypt_free(&Vbuf);
scrypt_free(&Ybuf);
scrypt_free(&Xbuf[0]); scrypt_free(&Xbuf[1]);
delete[] data[0]; delete[] data[1];
*hashes_done = n - pdata[19];
pdata[19] = n;
gettimeofday(tv_end, NULL);
return 0;
}
int scanhash_keccak(int thr_id, uint32_t *pdata, const uint32_t *ptarget,
uint32_t max_nonce, struct timeval *tv_start, struct timeval *tv_end, unsigned long *hashes_done)
{
int throughput = cuda_throughput(thr_id);
gettimeofday(tv_start, NULL);
uint32_t n = pdata[19] - 1;
// TESTING ONLY
// ((uint32_t*)ptarget)[7] = 0x0000000f;
const uint32_t Htarg = ptarget[7];
uint32_t endiandata[20];
for (int kk=0; kk < 20; kk++)
be32enc(&endiandata[kk], pdata[kk]);
cuda_prepare_keccak256(thr_id, endiandata, ptarget);
uint32_t *cuda_hash64[2] = { (uint32_t *)cuda_hashbuffer(thr_id, 0), (uint32_t *)cuda_hashbuffer(thr_id, 1) };
memset(cuda_hash64[0], 0xff, throughput * 8 * sizeof(uint32_t));
memset(cuda_hash64[1], 0xff, throughput * 8 * sizeof(uint32_t));
bool validate = false;
uint32_t nonce[2];
int cur = 0, nxt = 1;
// begin work on first CUDA stream
nonce[cur] = n+1; n += throughput;
cuda_do_keccak256(thr_id, 0, cuda_hash64[cur], nonce[cur], throughput, validate);
do {
nonce[nxt] = n+1; n += throughput;
if ((n-throughput) < max_nonce && !work_restart[thr_id].restart)
{
// begin work on next CUDA stream
cuda_do_keccak256(thr_id, 0, cuda_hash64[nxt], nonce[nxt], throughput, validate);
}
// synchronize current stream and get the "winning" nonce index, if any
cuda_scrypt_sync(thr_id, cur);
uint32_t result = *cuda_hash64[cur];
// optional full CPU based validation (see validate flag)
if (validate)
{
for (int i=0; i < throughput; ++i)
{
uint32_t hash64[8];
be32enc(&endiandata[19], nonce[cur]+i);
crypto_hash( (unsigned char*)hash64, (unsigned char*)&endiandata[0], 80 );
if (memcmp(hash64, &cuda_hash64[8*i], 32))
fprintf(stderr, "CPU and CUDA hashes (i=%d) differ!\n", i);
}
}
else if (result != 0xffffffff && result > pdata[19])
{
uint32_t hash64[8];
be32enc(&endiandata[19], result);
crypto_hash( (unsigned char*)hash64, (unsigned char*)&endiandata[0], 80 );
if (result >= nonce[cur] && result < nonce[cur]+throughput && hash64[7] <= Htarg && fulltest(hash64, ptarget)) {
*hashes_done = n-throughput - pdata[19] + 1;
pdata[19] = result;
gettimeofday(tv_end, NULL);
return true;
} else {
applog(LOG_INFO, "GPU #%d: %s result for nonce $%08x does not validate on CPU!", device_map[thr_id], device_name[thr_id], result);
}
}
cur = (cur + 1) % 2;
nxt = (nxt + 1) % 2;
} while ((n-throughput) < max_nonce && !work_restart[thr_id].restart);
*hashes_done = n-throughput - pdata[19] + 1;
if (n-throughput > pdata[19])
// CB: don't report values bigger max_nonce
pdata[19] = max_nonce > n-throughput ? n-throughput : max_nonce;
else
pdata[19] = 0xffffffffU; // CB: prevent nonce space overflow.
gettimeofday(tv_end, NULL);
return 0;
}