| @@ -3,60 +3,34 @@ | |||||
| #include <random> | #include <random> | ||||
| #include <future> | #include <future> | ||||
| #include <omp.h> | #include <omp.h> | ||||
| #include "/opt/OpenBLAS_zen_serial/include/cblas.h" | |||||
| #include "../cblas.h" | |||||
| #include "cpp_thread_safety_common.h" | |||||
| const blasint randomMatSize = 1024; //dimension of the random square matrices used | |||||
| const uint32_t numConcurrentThreads = 52; //number of concurrent calls of the functions being tested | |||||
| const uint32_t numTestRounds = 8; //number of testing rounds before success exit | |||||
| inline void pauser(){ | |||||
| /// a portable way to pause a program | |||||
| std::string dummy; | |||||
| std::cout << "Press enter to continue..."; | |||||
| std::getline(std::cin, dummy); | |||||
| } | |||||
| void launch_cblas_dgemm(double* A, double* B, double* C){ | |||||
| void launch_cblas_dgemm(double* A, double* B, double* C, const blasint randomMatSize){ | |||||
| cblas_dgemm(CblasColMajor, CblasNoTrans, CblasNoTrans, randomMatSize, randomMatSize, randomMatSize, 1.0, A, randomMatSize, B, randomMatSize, 0.1, C, randomMatSize); | cblas_dgemm(CblasColMajor, CblasNoTrans, CblasNoTrans, randomMatSize, randomMatSize, randomMatSize, 1.0, A, randomMatSize, B, randomMatSize, 0.1, C, randomMatSize); | ||||
| } | } | ||||
| void FillMatrices(std::vector<std::vector<double>>& matBlock, std::mt19937_64& PRNG, std::uniform_real_distribution<double>& rngdist){ | |||||
| for(uint32_t i=0; i<3; i++){ | |||||
| for(uint32_t j=0; j<(randomMatSize*randomMatSize); j++){ | |||||
| matBlock[i][j] = rngdist(PRNG); | |||||
| } | |||||
| } | |||||
| for(uint32_t i=3; i<(numConcurrentThreads*3); i+=3){ | |||||
| matBlock[i] = matBlock[0]; | |||||
| matBlock[i+1] = matBlock[1]; | |||||
| matBlock[i+2] = matBlock[2]; | |||||
| int main(int argc, char* argv[]){ | |||||
| blasint randomMatSize = 1024; //dimension of the random square matrices used | |||||
| uint32_t numConcurrentThreads = 52; //number of concurrent calls of the functions being tested | |||||
| uint32_t numTestRounds = 16; //number of testing rounds before success exit | |||||
| if (argc > 4){ | |||||
| std::cout<<"ERROR: too many arguments for thread safety tester"<<std::endl; | |||||
| abort(); | |||||
| } | } | ||||
| } | |||||
| std::mt19937_64 InitPRNG(){ | |||||
| std::random_device rd; | |||||
| std::mt19937_64 PRNG(rd()); //seed PRNG using /dev/urandom or similar OS provided RNG | |||||
| std::uniform_real_distribution<double> rngdist{-1.0, 1.0}; | |||||
| //make sure the internal state of the PRNG is properly mixed by generating 10M random numbers | |||||
| //PRNGs often have unreliable distribution uniformity and other statistical properties before their internal state is sufficiently mixed | |||||
| for (uint32_t i=0;i<10000000;i++) rngdist(PRNG); | |||||
| return PRNG; | |||||
| } | |||||
| void PrintMatrices(const std::vector<std::vector<double>>& matBlock){ | |||||
| for (uint32_t i=0;i<numConcurrentThreads*3;i++){ | |||||
| std::cout<<i<<std::endl; | |||||
| for (uint32_t j=0;j<randomMatSize;j++){ | |||||
| for (uint32_t k=0;k<randomMatSize;k++){ | |||||
| std::cout<<matBlock[i][j*randomMatSize + k]<<" "; | |||||
| } | |||||
| std::cout<<std::endl; | |||||
| if(argc == 4){ | |||||
| std::vector<std::string> cliArgs; | |||||
| for (int i = 1; i < argc; i++){ | |||||
| cliArgs.push_back(argv[i]); | |||||
| std::cout<<argv[i]<<std::endl; | |||||
| } | } | ||||
| std::cout<<std::endl; | |||||
| randomMatSize = std::stoul(cliArgs[0]); | |||||
| numConcurrentThreads = std::stoul(cliArgs[1]); | |||||
| numTestRounds = std::stoul(cliArgs[2]); | |||||
| } | } | ||||
| } | |||||
| int main(){ | |||||
| std::uniform_real_distribution<double> rngdist{-1.0, 1.0}; | std::uniform_real_distribution<double> rngdist{-1.0, 1.0}; | ||||
| std::vector<std::vector<double>> matBlock(numConcurrentThreads*3); | std::vector<std::vector<double>> matBlock(numConcurrentThreads*3); | ||||
| std::vector<std::future<void>> futureBlock(numConcurrentThreads); | std::vector<std::future<void>> futureBlock(numConcurrentThreads); | ||||
| @@ -81,17 +55,17 @@ int main(){ | |||||
| std::cout<<"done\n"; | std::cout<<"done\n"; | ||||
| //pauser(); | //pauser(); | ||||
| std::cout<<"Filling matrices with random numbers..."<<std::flush; | std::cout<<"Filling matrices with random numbers..."<<std::flush; | ||||
| FillMatrices(matBlock, PRNG, rngdist); | |||||
| //PrintMatrices(matBlock); | |||||
| FillMatrices(matBlock, PRNG, rngdist, randomMatSize, numConcurrentThreads); | |||||
| //PrintMatrices(matBlock, randomMatSize, numConcurrentThreads); | |||||
| std::cout<<"done\n"; | std::cout<<"done\n"; | ||||
| std::cout<<"Testing CBLAS DGEMM thread safety\n"; | std::cout<<"Testing CBLAS DGEMM thread safety\n"; | ||||
| omp_set_num_threads(numConcurrentThreads); | omp_set_num_threads(numConcurrentThreads); | ||||
| for(uint32_t R=0; R<numTestRounds; R++){ | for(uint32_t R=0; R<numTestRounds; R++){ | ||||
| std::cout<<"DGEMM round #"<<R<<std::endl; | std::cout<<"DGEMM round #"<<R<<std::endl; | ||||
| std::cout<<"Launching "<<numConcurrentThreads<<" threads..."<<std::flush; | |||||
| #pragma omp parallel for default(none) shared(futureBlock, matBlock) | |||||
| std::cout<<"Launching "<<numConcurrentThreads<<" threads simultaneously using OpenMP..."<<std::flush; | |||||
| #pragma omp parallel for default(none) shared(futureBlock, matBlock, randomMatSize, numConcurrentThreads) | |||||
| for(uint32_t i=0; i<numConcurrentThreads; i++){ | for(uint32_t i=0; i<numConcurrentThreads; i++){ | ||||
| futureBlock[i] = std::async(std::launch::async, launch_cblas_dgemm, &matBlock[i*3][0], &matBlock[i*3+1][0], &matBlock[i*3+2][0]); | |||||
| futureBlock[i] = std::async(std::launch::async, launch_cblas_dgemm, &matBlock[i*3][0], &matBlock[i*3+1][0], &matBlock[i*3+2][0], randomMatSize); | |||||
| //launch_cblas_dgemm( &matBlock[i][0], &matBlock[i+1][0], &matBlock[i+2][0]); | //launch_cblas_dgemm( &matBlock[i][0], &matBlock[i+1][0], &matBlock[i+2][0]); | ||||
| } | } | ||||
| std::cout<<"done\n"; | std::cout<<"done\n"; | ||||
| @@ -100,7 +74,7 @@ int main(){ | |||||
| futureBlock[i].get(); | futureBlock[i].get(); | ||||
| } | } | ||||
| std::cout<<"done\n"; | std::cout<<"done\n"; | ||||
| //PrintMatrices(matBlock); | |||||
| //PrintMatrices(matBlock, randomMatSize, numConcurrentThreads); | |||||
| std::cout<<"Comparing results from different threads..."<<std::flush; | std::cout<<"Comparing results from different threads..."<<std::flush; | ||||
| for(uint32_t i=3; i<(numConcurrentThreads*3); i+=3){ | for(uint32_t i=3; i<(numConcurrentThreads*3); i+=3){ | ||||
| for(uint32_t j=0; j<(randomMatSize*randomMatSize); j++){ | for(uint32_t j=0; j<(randomMatSize*randomMatSize); j++){ | ||||