portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-13 04:37:05 +01:00
Code Releases Activity

Sizable improvement in multigrid for unsquared.

Browse Source 
6000 matmuls CG unprec
2000 matmuls CG prec (4000 eo muls)
1050 matmuls PGCR on 16^3 x 32 x 8 m=.01

Substantial effort on timing and logging infrastructure
This commit is contained in:
Peter Boyle
2015-07-24 01:31:13 +09:00
parent 11c99d5e66
commit d1afebf71e
67 changed files with 945 additions and 753 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
benchmarks/Benchmark_comms.cc
View File

@ -11,15 +11,15 @@ int main (int argc, char ** argv)
std::vector<int> simd_layout = GridDefaultSimd(Nd,vComplexD::Nsimd());
std::vector<int> mpi_layout = GridDefaultMpi();
int threads = GridThread::GetThreads();
std::cout << "Grid is setup to use "<<threads<<" threads"<<std::endl;
std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
int Nloop=10;
int nmu=0;
for(int mu=0;mu<4;mu++) if (mpi_layout[mu]>1) nmu++;
std::cout << "===================================================================================================="<<std::endl;
std::cout << "= Benchmarking concurrent halo exchange in "<<nmu<<" dimensions"<<std::endl;
std::cout << "===================================================================================================="<<std::endl;
std::cout << " L "<<"\t\t"<<" Ls "<<"\t\t"<<"bytes"<<"\t\t"<<"MB/s uni"<<"\t\t"<<"MB/s bidi"<<std::endl;
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << "= Benchmarking concurrent halo exchange in "<<nmu<<" dimensions"<<std::endl;
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << " L "<<"\t\t"<<" Ls "<<"\t\t"<<"bytes"<<"\t\t"<<"MB/s uni"<<"\t\t"<<"MB/s bidi"<<std::endl;
@ -87,15 +87,15 @@ int main (int argc, char ** argv)
double time = stop-start; // microseconds
std::cout << lat<<"\t\t"<<Ls<<"\t\t"<<bytes<<"\t\t"<<xbytes/time<<"\t\t"<<bidibytes/time<<std::endl;
std::cout<<GridLogMessage << lat<<"\t\t"<<Ls<<"\t\t"<<bytes<<"\t\t"<<xbytes/time<<"\t\t"<<bidibytes/time<<std::endl;
}
}
std::cout << "===================================================================================================="<<std::endl;
std::cout << "= Benchmarking sequential halo exchange in "<<nmu<<" dimensions"<<std::endl;
std::cout << "===================================================================================================="<<std::endl;
std::cout << " L "<<"\t\t"<<" Ls "<<"\t\t"<<"bytes"<<"\t\t"<<"MB/s uni"<<"\t\t"<<"MB/s bidi"<<std::endl;
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << "= Benchmarking sequential halo exchange in "<<nmu<<" dimensions"<<std::endl;
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << " L "<<"\t\t"<<" Ls "<<"\t\t"<<"bytes"<<"\t\t"<<"MB/s uni"<<"\t\t"<<"MB/s bidi"<<std::endl;
for(int lat=4;lat<=32;lat+=2){
@ -163,7 +163,7 @@ int main (int argc, char ** argv)
double time = stop-start;
std::cout << lat<<"\t\t"<<Ls<<"\t\t"<<bytes<<"\t\t"<<xbytes/time<<"\t\t"<<bidibytes/time<<std::endl;
std::cout<<GridLogMessage << lat<<"\t\t"<<Ls<<"\t\t"<<bytes<<"\t\t"<<xbytes/time<<"\t\t"<<bidibytes/time<<std::endl;
}
}

40
benchmarks/Benchmark_dwf.cc
View File

@ -21,7 +21,7 @@ int main (int argc, char ** argv)
Grid_init(&argc,&argv);
int threads = GridThread::GetThreads();
std::cout << "Grid is setup to use "<<threads<<" threads"<<std::endl;
std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
std::vector<int> latt4 = GridDefaultLatt();
const int Ls=8;
@ -81,7 +81,7 @@ int main (int argc, char ** argv)
RealD M5 =1.8;
DomainWallFermion Dw(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
std::cout << "Calling Dw"<<std::endl;
std::cout<<GridLogMessage << "Calling Dw"<<std::endl;
int ncall=10;
double t0=usecond();
for(int i=0;i<ncall;i++){
@ -92,12 +92,12 @@ int main (int argc, char ** argv)
double volume=Ls; for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
double flops=1344*volume*ncall;
std::cout << "Called Dw"<<std::endl;
std::cout << "norm result "<< norm2(result)<<std::endl;
std::cout << "norm ref "<< norm2(ref)<<std::endl;
std::cout << "mflop/s = "<< flops/(t1-t0)<<std::endl;
std::cout<<GridLogMessage << "Called Dw"<<std::endl;
std::cout<<GridLogMessage << "norm result "<< norm2(result)<<std::endl;
std::cout<<GridLogMessage << "norm ref "<< norm2(ref)<<std::endl;
std::cout<<GridLogMessage << "mflop/s = "<< flops/(t1-t0)<<std::endl;
err = ref-result;
std::cout << "norm diff "<< norm2(err)<<std::endl;
std::cout<<GridLogMessage << "norm diff "<< norm2(err)<<std::endl;
if (1)
@ -120,11 +120,11 @@ int main (int argc, char ** argv)
ref = -0.5*ref;
}
Dw.Dhop(src,result,1);
std::cout << "Called DwDag"<<std::endl;
std::cout << "norm result "<< norm2(result)<<std::endl;
std::cout << "norm ref "<< norm2(ref)<<std::endl;
std::cout<<GridLogMessage << "Called DwDag"<<std::endl;
std::cout<<GridLogMessage << "norm result "<< norm2(result)<<std::endl;
std::cout<<GridLogMessage << "norm ref "<< norm2(ref)<<std::endl;
err = ref-result;
std::cout << "norm diff "<< norm2(err)<<std::endl;
std::cout<<GridLogMessage << "norm diff "<< norm2(err)<<std::endl;
LatticeFermion src_e (FrbGrid);
LatticeFermion src_o (FrbGrid);
@ -133,24 +133,32 @@ int main (int argc, char ** argv)
LatticeFermion r_eo (FGrid);
std::cout << "Calling Deo and Doe"<<std::endl;
std::cout<<GridLogMessage << "Calling Deo and Doe"<<std::endl;
pickCheckerboard(Even,src_e,src);
pickCheckerboard(Odd,src_o,src);
std::cout<<GridLogMessage << "src_e"<<norm2(src_e)<<std::endl;
std::cout<<GridLogMessage << "src_o"<<norm2(src_o)<<std::endl;
Dw.DhopEO(src_o,r_e,DaggerNo);
Dw.DhopOE(src_e,r_o,DaggerNo);
Dw.Dhop(src,result,DaggerNo);
Dw.Dhop (src ,result,DaggerNo);
std::cout<<GridLogMessage << "r_e"<<norm2(r_e)<<std::endl;
std::cout<<GridLogMessage << "r_o"<<norm2(r_o)<<std::endl;
std::cout<<GridLogMessage << "res"<<norm2(result)<<std::endl;
setCheckerboard(r_eo,r_o);
setCheckerboard(r_eo,r_e);
err = r_eo-result;
std::cout << "norm diff "<< norm2(err)<<std::endl;
std::cout<<GridLogMessage << "norm diff "<< norm2(err)<<std::endl;
pickCheckerboard(Even,src_e,err);
pickCheckerboard(Odd,src_o,err);
std::cout << "norm diff even "<< norm2(src_e)<<std::endl;
std::cout << "norm diff odd "<< norm2(src_o)<<std::endl;
std::cout<<GridLogMessage << "norm diff even "<< norm2(src_e)<<std::endl;
std::cout<<GridLogMessage << "norm diff odd "<< norm2(src_o)<<std::endl;
Grid_finalize();
}

48
benchmarks/Benchmark_memory_bandwidth.cc
View File

@ -17,13 +17,13 @@ int main (int argc, char ** argv)
std::vector<int> mpi_layout = GridDefaultMpi();
int threads = GridThread::GetThreads();
std::cout << "Grid is setup to use "<<threads<<" threads"<<std::endl;
std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
std::cout << "===================================================================================================="<<std::endl;
std::cout << "= Benchmarking fused AXPY bandwidth ; sizeof(Real) "<<sizeof(Real)<<std::endl;
std::cout << "===================================================================================================="<<std::endl;
std::cout << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s"<<"\t\t"<<"Gflop/s"<<std::endl;
std::cout << "----------------------------------------------------------"<<std::endl;
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << "= Benchmarking fused AXPY bandwidth ; sizeof(Real) "<<sizeof(Real)<<std::endl;
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s"<<"\t\t"<<"Gflop/s"<<std::endl;
std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
for(int lat=4;lat<=32;lat+=4){
@ -49,15 +49,15 @@ int main (int argc, char ** argv)
double flops=vol*Nvec*2;// mul,add
double bytes=3*vol*Nvec*sizeof(Real);
std::cout<<std::setprecision(3) << lat<<"\t\t"<<bytes<<" \t\t"<<bytes/time<<"\t\t"<<flops/time<<std::endl;
std::cout<<GridLogMessage<<std::setprecision(3) << lat<<"\t\t"<<bytes<<" \t\t"<<bytes/time<<"\t\t"<<flops/time<<std::endl;
}
std::cout << "===================================================================================================="<<std::endl;
std::cout << "= Benchmarking a*x + y bandwidth"<<std::endl;
std::cout << "===================================================================================================="<<std::endl;
std::cout << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s"<<"\t\t"<<"Gflop/s"<<std::endl;
std::cout << "----------------------------------------------------------"<<std::endl;
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << "= Benchmarking a*x + y bandwidth"<<std::endl;
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s"<<"\t\t"<<"Gflop/s"<<std::endl;
std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
for(int lat=4;lat<=32;lat+=4){
@ -81,14 +81,14 @@ int main (int argc, char ** argv)
double flops=vol*Nvec*2;// mul,add
double bytes=3*vol*Nvec*sizeof(Real);
std::cout<<std::setprecision(3) << lat<<"\t\t"<<bytes<<" \t\t"<<bytes/time<<"\t\t"<<flops/time<<std::endl;
std::cout<<GridLogMessage<<std::setprecision(3) << lat<<"\t\t"<<bytes<<" \t\t"<<bytes/time<<"\t\t"<<flops/time<<std::endl;
}
std::cout << "===================================================================================================="<<std::endl;
std::cout << "= Benchmarking SCALE bandwidth"<<std::endl;
std::cout << "===================================================================================================="<<std::endl;
std::cout << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s"<<"\t\t"<<"Gflop/s"<<std::endl;
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << "= Benchmarking SCALE bandwidth"<<std::endl;
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s"<<"\t\t"<<"Gflop/s"<<std::endl;
for(int lat=4;lat<=32;lat+=4){
@ -114,15 +114,15 @@ int main (int argc, char ** argv)
double bytes=2*vol*Nvec*sizeof(Real);
double flops=vol*Nvec*1;// mul
std::cout <<std::setprecision(3) << lat<<"\t\t"<<bytes<<" \t\t"<<bytes/time<<"\t\t"<<flops/time<<std::endl;
std::cout<<GridLogMessage <<std::setprecision(3) << lat<<"\t\t"<<bytes<<" \t\t"<<bytes/time<<"\t\t"<<flops/time<<std::endl;
}
std::cout << "===================================================================================================="<<std::endl;
std::cout << "= Benchmarking READ bandwidth"<<std::endl;
std::cout << "===================================================================================================="<<std::endl;
std::cout << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s"<<"\t\t"<<"Gflop/s"<<std::endl;
std::cout << "----------------------------------------------------------"<<std::endl;
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << "= Benchmarking READ bandwidth"<<std::endl;
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s"<<"\t\t"<<"Gflop/s"<<std::endl;
std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
for(int lat=4;lat<=32;lat+=4){
@ -147,7 +147,7 @@ int main (int argc, char ** argv)
double bytes=vol*Nvec*sizeof(Real);
double flops=vol*Nvec*2;// mul,add
std::cout<<std::setprecision(3) << lat<<"\t\t"<<bytes<<" \t\t"<<bytes/time<<"\t\t"<<flops/time<<std::endl;
std::cout<<GridLogMessage<<std::setprecision(3) << lat<<"\t\t"<<bytes<<" \t\t"<<bytes/time<<"\t\t"<<flops/time<<std::endl;
}

50
benchmarks/Benchmark_su3.cc
View File

@ -14,13 +14,13 @@ int main (int argc, char ** argv)
std::vector<int> mpi_layout = GridDefaultMpi();
int threads = GridThread::GetThreads();
std::cout << "Grid is setup to use "<<threads<<" threads"<<std::endl;
std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
std::cout << "===================================================================================================="<<std::endl;
std::cout << "= Benchmarking SU3xSU3 x= x*y"<<std::endl;
std::cout << "===================================================================================================="<<std::endl;
std::cout << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
std::cout << "----------------------------------------------------------"<<std::endl;
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << "= Benchmarking SU3xSU3 x= x*y"<<std::endl;
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
for(int lat=2;lat<=24;lat+=2){
@ -43,16 +43,16 @@ int main (int argc, char ** argv)
double bytes=3.0*vol*Nc*Nc*sizeof(Complex);
double footprint=2.0*vol*Nc*Nc*sizeof(Complex);
double flops=Nc*Nc*(6.0+8.0+8.0)*vol;
std::cout<<std::setprecision(3) << lat<<"\t\t"<<footprint<<" \t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
std::cout<<GridLogMessage<<std::setprecision(3) << lat<<"\t\t"<<footprint<<" \t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
}
std::cout << "===================================================================================================="<<std::endl;
std::cout << "= Benchmarking SU3xSU3 z= x*y"<<std::endl;
std::cout << "===================================================================================================="<<std::endl;
std::cout << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
std::cout << "----------------------------------------------------------"<<std::endl;
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << "= Benchmarking SU3xSU3 z= x*y"<<std::endl;
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
for(int lat=2;lat<=24;lat+=2){
@ -75,15 +75,15 @@ int main (int argc, char ** argv)
double bytes=3*vol*Nc*Nc*sizeof(Complex);
double flops=Nc*Nc*(6+8+8)*vol;
std::cout<<std::setprecision(3) << lat<<"\t\t"<<bytes<<" \t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
std::cout<<GridLogMessage<<std::setprecision(3) << lat<<"\t\t"<<bytes<<" \t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
}
std::cout << "===================================================================================================="<<std::endl;
std::cout << "= Benchmarking SU3xSU3 mult(z,x,y)"<<std::endl;
std::cout << "===================================================================================================="<<std::endl;
std::cout << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
std::cout << "----------------------------------------------------------"<<std::endl;
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << "= Benchmarking SU3xSU3 mult(z,x,y)"<<std::endl;
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
for(int lat=2;lat<=24;lat+=2){
@ -106,15 +106,15 @@ int main (int argc, char ** argv)
double bytes=3*vol*Nc*Nc*sizeof(Complex);
double flops=Nc*Nc*(6+8+8)*vol;
std::cout<<std::setprecision(3) << lat<<"\t\t"<<bytes<<" \t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
std::cout<<GridLogMessage<<std::setprecision(3) << lat<<"\t\t"<<bytes<<" \t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
}
std::cout << "===================================================================================================="<<std::endl;
std::cout << "= Benchmarking SU3xSU3 mac(z,x,y)"<<std::endl;
std::cout << "===================================================================================================="<<std::endl;
std::cout << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
std::cout << "----------------------------------------------------------"<<std::endl;
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << "= Benchmarking SU3xSU3 mac(z,x,y)"<<std::endl;
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
for(int lat=2;lat<=24;lat+=2){
@ -137,7 +137,7 @@ int main (int argc, char ** argv)
double bytes=3*vol*Nc*Nc*sizeof(Complex);
double flops=Nc*Nc*(8+8+8)*vol;
std::cout<<std::setprecision(3) << lat<<"\t\t"<<bytes<<" \t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
std::cout<<GridLogMessage<<std::setprecision(3) << lat<<"\t\t"<<bytes<<" \t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
}

34
benchmarks/Benchmark_wilson.cc
View File

@ -28,10 +28,10 @@ int main (int argc, char ** argv)
GridRedBlackCartesian RBGrid(latt_size,simd_layout,mpi_layout);
int threads = GridThread::GetThreads();
std::cout << "Grid is setup to use "<<threads<<" threads"<<std::endl;
std::cout << "Grid floating point word size is REALF"<< sizeof(RealF)<<std::endl;
std::cout << "Grid floating point word size is REALD"<< sizeof(RealD)<<std::endl;
std::cout << "Grid floating point word size is REAL"<< sizeof(Real)<<std::endl;
std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
std::cout<<GridLogMessage << "Grid floating point word size is REALF"<< sizeof(RealF)<<std::endl;
std::cout<<GridLogMessage << "Grid floating point word size is REALD"<< sizeof(RealD)<<std::endl;
std::cout<<GridLogMessage << "Grid floating point word size is REAL"<< sizeof(Real)<<std::endl;
std::vector<int> seeds({1,2,3,4});
GridParallelRNG pRNG(&Grid);
@ -58,8 +58,8 @@ int main (int argc, char ** argv)
for(int nn=0;nn<Nd;nn++){
random(pRNG,U[nn]);
if(0) {
if (nn==-1) { U[nn]=zero; std::cout << "zeroing gauge field in dir "<<nn<<std::endl; }
else { U[nn] = cone;std::cout << "unit gauge field in dir "<<nn<<std::endl; }
if (nn==-1) { U[nn]=zero; std::cout<<GridLogMessage << "zeroing gauge field in dir "<<nn<<std::endl; }
else { U[nn] = cone;std::cout<<GridLogMessage << "unit gauge field in dir "<<nn<<std::endl; }
}
pokeIndex<LorentzIndex>(Umu,U[nn],nn);
}
@ -89,7 +89,7 @@ int main (int argc, char ** argv)
RealD mass=0.1;
WilsonFermion Dw(Umu,Grid,RBGrid,mass);
std::cout << "Calling Dw"<<std::endl;
std::cout<<GridLogMessage << "Calling Dw"<<std::endl;
int ncall=10000;
double t0=usecond();
for(int i=0;i<ncall;i++){
@ -98,12 +98,12 @@ int main (int argc, char ** argv)
double t1=usecond();
double flops=1344*volume*ncall;
std::cout << "Called Dw"<<std::endl;
std::cout << "norm result "<< norm2(result)<<std::endl;
std::cout << "norm ref "<< norm2(ref)<<std::endl;
std::cout << "mflop/s = "<< flops/(t1-t0)<<std::endl;
std::cout<<GridLogMessage << "Called Dw"<<std::endl;
std::cout<<GridLogMessage << "norm result "<< norm2(result)<<std::endl;
std::cout<<GridLogMessage << "norm ref "<< norm2(ref)<<std::endl;
std::cout<<GridLogMessage << "mflop/s = "<< flops/(t1-t0)<<std::endl;
err = ref-result;
std::cout << "norm diff "<< norm2(err)<<std::endl;
std::cout<<GridLogMessage << "norm diff "<< norm2(err)<<std::endl;
// for(int ss=0;ss<10;ss++ ){
@ -112,7 +112,7 @@ int main (int argc, char ** argv)
for(int j=0;j<Nc;j++){
ComplexF * ref_p = (ComplexF *)&ref._odata[ss]()(i)(j);
ComplexF * res_p = (ComplexF *)&result._odata[ss]()(i)(j);
std::cout << ss<< " "<<i<<" "<<j<<" "<< (*ref_p)<<" " <<(*res_p)<<std::endl;
std::cout<<GridLogMessage << ss<< " "<<i<<" "<<j<<" "<< (*ref_p)<<" " <<(*res_p)<<std::endl;
}
}
}
@ -136,11 +136,11 @@ int main (int argc, char ** argv)
}
ref = -0.5*ref;
Dw.Dhop(src,result,1);
std::cout << "Called DwDag"<<std::endl;
std::cout << "norm result "<< norm2(result)<<std::endl;
std::cout << "norm ref "<< norm2(ref)<<std::endl;
std::cout<<GridLogMessage << "Called DwDag"<<std::endl;
std::cout<<GridLogMessage << "norm result "<< norm2(result)<<std::endl;
std::cout<<GridLogMessage << "norm ref "<< norm2(ref)<<std::endl;
err = ref-result;
std::cout << "norm diff "<< norm2(err)<<std::endl;
std::cout<<GridLogMessage << "norm diff "<< norm2(err)<<std::endl;
Grid_finalize();
}