portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-10-25 02:04:48 +01:00
Code Releases Activity

Perf count control

Browse Source
This commit is contained in:
Peter Boyle
2018-05-12 17:57:32 -04:00
parent 87c5c0271b
commit eb921041d0
1 changed files with 22 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

31
benchmarks/Benchmark_dwf_sweep.cc
View File

@@ -30,7 +30,6 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
using namespace std; using namespace std;
using namespace Grid; using namespace Grid;
;
template<class d> template<class d>
struct scal { struct scal {
@@ -60,7 +59,7 @@ int main (int argc, char ** argv)
if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptInlineAsm ) std::cout << GridLogMessage<< "* Using Asm Nc=3 WilsonKernels" <<std::endl; if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptInlineAsm ) std::cout << GridLogMessage<< "* Using Asm Nc=3 WilsonKernels" <<std::endl;
std::cout << GridLogMessage<< "*****************************************************************" <<std::endl; std::cout << GridLogMessage<< "*****************************************************************" <<std::endl;
const int Ls=8; const int Ls=12;
int threads = GridThread::GetThreads(); int threads = GridThread::GetThreads();
std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl; std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
std::cout<<GridLogMessage << "=========================================================================="<<std::endl; std::cout<<GridLogMessage << "=========================================================================="<<std::endl;
@@ -88,6 +87,7 @@ int main (int argc, char ** argv)
} }
} }
std::cout<<GridLogMessage << "=========================================================================="<<std::endl; std::cout<<GridLogMessage << "=========================================================================="<<std::endl;
#ifdef PERFCOUNT
{ {
std::vector<int> latt4(4,16); std::vector<int> latt4(4,16);
std::cout<<GridLogMessage << "16^4 Dw miss rate"<<std::endl; std::cout<<GridLogMessage << "16^4 Dw miss rate"<<std::endl;
@@ -95,7 +95,7 @@ int main (int argc, char ** argv)
std::cout<<GridLogMessage << "16^4 sDw miss rate"<<std::endl; std::cout<<GridLogMessage << "16^4 sDw miss rate"<<std::endl;
benchsDw(latt4,Ls,threads,1); benchsDw(latt4,Ls,threads,1);
} }
#endif
Grid_finalize(); Grid_finalize();
} }
@@ -183,19 +183,21 @@ void benchDw(std::vector<int> & latt4, int Ls, int threads,int report )
if (ncall < 5 ) exit(0); if (ncall < 5 ) exit(0);
Dw.Dhop(src,result,0); Dw.Dhop(src,result,0);
#ifdef PERFCOUNT
PerformanceCounter Counter(8); PerformanceCounter Counter(8);
Counter.Start(); Counter.Start();
#endif
t0=usecond(); t0=usecond();
for(int i=0;i<ncall;i++){ for(int i=0;i<ncall;i++){
Dw.Dhop(src,result,0); Dw.Dhop(src,result,0);
} }
t1=usecond(); t1=usecond();
#ifdef PERFCOUNT
Counter.Stop(); Counter.Stop();
if ( report ) { if ( report ) {
Counter.Report(); Counter.Report();
} }
#endif
if ( ! report ) { if ( ! report ) {
double volume=Ls; for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu]; double volume=Ls; for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
double flops=1344*volume*ncall; double flops=1344*volume*ncall;
@@ -312,18 +314,24 @@ void benchsDw(std::vector<int> & latt4, int Ls, int threads, int report )
int ncall =1+(int) ((5.0*1000*1000)/(t1-t0)); int ncall =1+(int) ((5.0*1000*1000)/(t1-t0));
#endif #endif
#ifdef PERFCOUNT
PerformanceCounter Counter(8); PerformanceCounter Counter(8);
Counter.Start(); Counter.Start();
#endif
t0=usecond(); t0=usecond();
for(int i=0;i<ncall;i++){ for(int i=0;i<ncall;i++){
sDw.Dhop(ssrc,sresult,0); sDw.Dhop(ssrc,sresult,0);
} }
t1=usecond(); t1=usecond();
Counter.Stop();
#ifdef PERFCOUNT
Counter.Stop();
if ( report ) { if ( report ) {
Counter.Report(); Counter.Report();
} else { }
#endif
if ( !report){
double volume=Ls; for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu]; double volume=Ls; for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
double flops=1344*volume*ncall; double flops=1344*volume*ncall;
std::cout<<"\t"<< flops/(t1-t0); std::cout<<"\t"<< flops/(t1-t0);
@@ -347,8 +355,10 @@ void benchsDw(std::vector<int> & latt4, int Ls, int threads, int report )
sr_o = Zero(); sr_o = Zero();
sDw.DhopEO(ssrc_o,sr_e,DaggerNo); sDw.DhopEO(ssrc_o,sr_e,DaggerNo);
#ifdef PERFCOUNT
PerformanceCounter CounterSdw(8); PerformanceCounter CounterSdw(8);
CounterSdw.Start(); CounterSdw.Start();
#endif
t0=usecond(); t0=usecond();
for(int i=0;i<ncall;i++){ for(int i=0;i<ncall;i++){
__SSC_START; __SSC_START;
@@ -356,11 +366,14 @@ void benchsDw(std::vector<int> & latt4, int Ls, int threads, int report )
__SSC_STOP; __SSC_STOP;
} }
t1=usecond(); t1=usecond();
#ifdef PERFCOUNT
CounterSdw.Stop(); CounterSdw.Stop();
if ( report ) { if ( report ) {
CounterSdw.Report(); CounterSdw.Report();
} else { }
#endif
if ( ! report ) {
double volume=Ls; for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu]; double volume=Ls; for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
double flops=(1344.0*volume*ncall)/2; double flops=(1344.0*volume*ncall)/2;
std::cout<<"\t"<< flops/(t1-t0); std::cout<<"\t"<< flops/(t1-t0);