floraachy
/
OpenBLAS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

more output, use OMP for simultaneous thread launch 

Launching threads sequentially has a significant chance of giving a false negative. If an OMP parallel for construct is used to launch the concurrent calculations, then all threads start almost exactly at the same time, thus they have the highest chance of clashing.
This commit is contained in:
TiborGY 2019-03-15 13:26:36 +01:00 committed by GitHub
parent 1b88a2b1ed
commit 2e8cd9f3ad
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 17 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
cpp_thread_test/dgemm_thread_safety.cpp
View File

@ -2,11 +2,12 @@
#include <vector>
#include <random>
#include <future>
#include <omp.h>
#include "/opt/OpenBLAS_zen_serial/include/cblas.h"
const blasint randomMatSize = 1024; //dimension of the random square matrices used
const uint32_t numConcurrentThreads = 64; //number of concurrent calls of the functions being tested
const uint32_t numTestRounds = 32; //number of testing rounds before success exit
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
@ -60,6 +61,14 @@ int main(){
std::vector<std::vector<double>> matBlock(numConcurrentThreads*3);
std::vector<std::future<void>> futureBlock(numConcurrentThreads);
std::cout<<"*----------------------------*\n";
std::cout<<"| DGEMM thread safety tester |\n";
std::cout<<"*----------------------------*\n";
std::cout<<"Size of random matrices(N=M=K): "<<randomMatSize<<'\n';
std::cout<<"Number of concurrent calls into OpenBLAS : "<<numConcurrentThreads<<'\n';
std::cout<<"Number of testing rounds : "<<numTestRounds<<'\n';
std::cout<<"This test will need "<<(static_cast<uint64_t>(randomMatSize*randomMatSize)*numConcurrentThreads*3*8)/static_cast<double>(1024*1024)<<" MiB of RAM\n"<<std::endl;
std::cout<<"Initializing random number generator..."<<std::flush;
std::mt19937_64 PRNG = InitPRNG();
std::cout<<"done\n";
@ -70,14 +79,17 @@ int main(){
matBlock[i].resize(randomMatSize*randomMatSize);
}
std::cout<<"done\n";
//pauser();
std::cout<<"Filling matrices with random numbers..."<<std::flush;
FillMatrices(matBlock, PRNG, rngdist);
//PrintMatrices(matBlock);
std::cout<<"done\n";
std::cout<<"Testing CBLAS DGEMM thread safety\n";
omp_set_num_threads(numConcurrentThreads);
for(uint32_t R=0; R<numTestRounds; R++){
std::cout<<"DGEMM round #"<<R<<std::endl;
std::cout<<"Launching "<<numConcurrentThreads<<" threads..."<<std::flush;
#pragma omp parallel for default(none) shared(futureBlock, matBlock)
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]);
//launch_cblas_dgemm( &matBlock[i][0], &matBlock[i+1][0], &matBlock[i+2][0]);
@ -94,11 +106,13 @@ int main(){
for(uint32_t j=0; j<(randomMatSize*randomMatSize); j++){
if (std::abs(matBlock[i+2][j] - matBlock[2][j]) > 1.0E-13){
std::cout<<"ERROR: one of the threads returned a different result!"<<i+2<<std::endl;
std::cout<<"CBLAS DGEMM thread safety test FAILED!"<<std::endl;
return -1;
}
}
}
std::cout<<"OK!"<<std::endl;
}
std::cout<<"CBLAS DGEMM thread safety test PASSED!"<<std::endl;
return 0;
}