portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-13 12:47:05 +01:00
Code Releases Activity

Adding utilities for perf profiling

Browse Source
This commit is contained in:
Guido Cossu
2018-01-29 11:11:45 +01:00
parent 655a69259a
commit fb24e3a7d2
2 changed files with 95 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
benchmarks/Benchmark_wilson.cc
View File

@ -4,7 +4,7 @@
Source file: ./benchmarks/Benchmark_wilson.cc
Copyright (C) 2015
Copyright (C) 2018
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: paboyle <paboyle@ph.ed.ac.uk>
@ -32,6 +32,9 @@ using namespace std;
using namespace Grid;
using namespace Grid::QCD;
#include "Grid/util/Profiling.h"
template<class d>
struct scal {
d internal;
@ -45,6 +48,7 @@ struct scal {
};
bool overlapComms = false;
bool perfProfiling = false;
int main (int argc, char ** argv)
{
@ -53,6 +57,9 @@ int main (int argc, char ** argv)
if( GridCmdOptionExists(argv,argv+argc,"--asynch") ){
overlapComms = true;
}
if( GridCmdOptionExists(argv,argv+argc,"--perf") ){
perfProfiling = true;
}
long unsigned int single_site_flops = 8*QCD::Nc*(7+16*QCD::Nc);
@ -144,6 +151,21 @@ int main (int argc, char ** argv)
double t1=usecond();
double flops=single_site_flops*volume*ncall;
if (perfProfiling){
std::cout<<GridLogMessage << "Profiling Dw with perf"<<std::endl;
System::profile("kernel", [&]() {
for(int i=0;i<ncall;i++){
Dw.Dhop(src,result,0);
}
});
std::cout<<GridLogMessage << "Generated kernel.data"<<std::endl;
std::cout<<GridLogMessage << "Use with: perf report -i kernel.data"<<std::endl;
}
std::cout<<GridLogMessage << "Called Dw"<<std::endl;
std::cout<<GridLogMessage << "flops per site " << single_site_flops << std::endl;
std::cout<<GridLogMessage << "norm result "<< norm2(result)<<std::endl;