add --noBatch benchmark option

This commit is contained in:
tevador 2020-07-04 14:57:56 +02:00
parent 9905ec9c5a
commit 5ce5f4906c

View File

@ -95,6 +95,7 @@ void printUsage(const char* executable) {
std::cout << " --ssse3 use optimized Argon2 for SSSE3 CPUs" << std::endl; std::cout << " --ssse3 use optimized Argon2 for SSSE3 CPUs" << std::endl;
std::cout << " --avx2 use optimized Argon2 for AVX2 CPUs" << std::endl; std::cout << " --avx2 use optimized Argon2 for AVX2 CPUs" << std::endl;
std::cout << " --auto select the best options for the current CPU" << std::endl; std::cout << " --auto select the best options for the current CPU" << std::endl;
std::cout << " --noBatch calculate hashes one by one (default: batch)" << std::endl;
} }
struct MemoryException : public std::exception { struct MemoryException : public std::exception {
@ -110,7 +111,10 @@ struct DatasetAllocException : public MemoryException {
} }
}; };
void mine(randomx_vm* vm, std::atomic<uint32_t>& atomicNonce, AtomicHash& result, uint32_t noncesCount, int thread, int cpuid=-1) { using MineFunc = void(randomx_vm * vm, std::atomic<uint32_t> & atomicNonce, AtomicHash & result, uint32_t noncesCount, int thread, int cpuid);
template<bool batch>
void mine(randomx_vm* vm, std::atomic<uint32_t>& atomicNonce, AtomicHash& result, uint32_t noncesCount, int thread, int cpuid = -1) {
if (cpuid >= 0) { if (cpuid >= 0) {
int rc = set_thread_affinity(cpuid); int rc = set_thread_affinity(cpuid);
if (rc) { if (rc) {
@ -123,19 +127,27 @@ void mine(randomx_vm* vm, std::atomic<uint32_t>& atomicNonce, AtomicHash& result
void* noncePtr = blockTemplate + 39; void* noncePtr = blockTemplate + 39;
auto nonce = atomicNonce.fetch_add(1); auto nonce = atomicNonce.fetch_add(1);
if (batch) {
store32(noncePtr, nonce); store32(noncePtr, nonce);
randomx_calculate_hash_first(vm, blockTemplate, sizeof(blockTemplate)); randomx_calculate_hash_first(vm, blockTemplate, sizeof(blockTemplate));
}
while (nonce < noncesCount) { while (nonce < noncesCount) {
if (batch) {
nonce = atomicNonce.fetch_add(1); nonce = atomicNonce.fetch_add(1);
}
store32(noncePtr, nonce); store32(noncePtr, nonce);
randomx_calculate_hash_next(vm, blockTemplate, sizeof(blockTemplate), &hash); (batch ? randomx_calculate_hash_next : randomx_calculate_hash)(vm, blockTemplate, sizeof(blockTemplate), &hash);
result.xorWith(hash); result.xorWith(hash);
if (!batch) {
nonce = atomicNonce.fetch_add(1);
}
} }
} }
int main(int argc, char** argv) { int main(int argc, char** argv) {
bool softAes, miningMode, verificationMode, help, largePages, jit, secure, ssse3, avx2, autoFlags; bool softAes, miningMode, verificationMode, help, largePages, jit, secure;
bool ssse3, avx2, autoFlags, noBatch;
int noncesCount, threadCount, initThreadCount; int noncesCount, threadCount, initThreadCount;
uint64_t threadAffinity; uint64_t threadAffinity;
int32_t seedValue; int32_t seedValue;
@ -159,10 +171,11 @@ int main(int argc, char** argv) {
readOption("--ssse3", argc, argv, ssse3); readOption("--ssse3", argc, argv, ssse3);
readOption("--avx2", argc, argv, avx2); readOption("--avx2", argc, argv, avx2);
readOption("--auto", argc, argv, autoFlags); readOption("--auto", argc, argv, autoFlags);
readOption("--noBatch", argc, argv, noBatch);
store32(&seed, seedValue); store32(&seed, seedValue);
std::cout << "RandomX benchmark v1.1.7" << std::endl; std::cout << "RandomX benchmark v1.1.8" << std::endl;
if (help) { if (help) {
printUsage(argv[0]); printUsage(argv[0]);
@ -264,6 +277,16 @@ int main(int argc, char** argv) {
std::cout << " - thread affinity (" << mask_to_string(threadAffinity) << ")" << std::endl; std::cout << " - thread affinity (" << mask_to_string(threadAffinity) << ")" << std::endl;
} }
MineFunc* func;
if (noBatch) {
func = &mine<false>;
}
else {
func = &mine<true>;
std::cout << " - batch mode" << std::endl;
}
std::cout << "Initializing"; std::cout << "Initializing";
if (miningMode) if (miningMode)
std::cout << " (" << initThreadCount << " thread" << (initThreadCount > 1 ? "s)" : ")"); std::cout << " (" << initThreadCount << " thread" << (initThreadCount > 1 ? "s)" : ")");
@ -334,14 +357,14 @@ int main(int argc, char** argv) {
int cpuid = -1; int cpuid = -1;
if (threadAffinity) if (threadAffinity)
cpuid = cpuid_from_mask(threadAffinity, i); cpuid = cpuid_from_mask(threadAffinity, i);
threads.push_back(std::thread(&mine, vms[i], std::ref(atomicNonce), std::ref(result), noncesCount, i, cpuid)); threads.push_back(std::thread(func, vms[i], std::ref(atomicNonce), std::ref(result), noncesCount, i, cpuid));
} }
for (unsigned i = 0; i < threads.size(); ++i) { for (unsigned i = 0; i < threads.size(); ++i) {
threads[i].join(); threads[i].join();
} }
} }
else { else {
mine(vms[0], std::ref(atomicNonce), std::ref(result), noncesCount, 0); func(vms[0], std::ref(atomicNonce), std::ref(result), noncesCount, 0, -1);
} }
double elapsed = sw.getElapsed(); double elapsed = sw.getElapsed();