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

Added a test an fixed it for red black precon Ls innermost vectorised DWF

Browse Source
This commit is contained in:
paboyle
2016-06-07 13:16:56 -07:00
parent e503ef5590
commit 8ac021de73
6 changed files with 71 additions and 29 deletions
Download Patch File Download Diff File Expand all files Collapse all files

54
benchmarks/Benchmark_dwf.cc
View File

@ -71,7 +71,6 @@ int main (int argc, char ** argv)
std::cout << GridLogMessage << "Making s innermost grids"<<std::endl;
GridCartesian * sUGrid = SpaceTimeGrid::makeFourDimDWFGrid(GridDefaultLatt(),GridDefaultMpi());
GridRedBlackCartesian * sUrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(sUGrid);
GridCartesian * sFGrid = SpaceTimeGrid::makeFiveDimDWFGrid(Ls,UGrid);
std::cout << GridLogMessage << "Making s innermost rb grids"<<std::endl;
GridRedBlackCartesian * sFrbGrid = SpaceTimeGrid::makeFiveDimDWFRedBlackGrid(Ls,UGrid);
@ -162,13 +161,13 @@ int main (int argc, char ** argv)
// Dw.Report();
}
if (0)
if (1)
{
typedef WilsonFermion5D<DomainWallRedBlack5dImplF> WilsonFermion5DF;
LatticeFermionF ssrc(sFGrid);
LatticeFermionF sref(sFGrid);
LatticeFermionF sresult(sFGrid);
WilsonFermion5DF sDw(1,Umu,*sFGrid,*sFrbGrid,*sUGrid,*sUrbGrid,M5,params);
WilsonFermion5DF sDw(1,Umu,*sFGrid,*sFrbGrid,*sUGrid,M5,params);
for(int x=0;x<latt4[0];x++){
for(int y=0;y<latt4[1];y++){
@ -196,7 +195,7 @@ int main (int argc, char ** argv)
std::cout<<GridLogMessage << "mflop/s per node = "<< flops/(t1-t0)/NP<<std::endl;
// sDw.Report();
if(1){
if(0){
for(int i=0;i< PerformanceCounter::NumTypes(); i++ ){
sDw.Dhop(ssrc,sresult,0);
PerformanceCounter Counter(i);
@ -226,6 +225,53 @@ int main (int argc, char ** argv)
}}}}}
std::cout<<" difference between normal and simd is "<<sum<<std::endl;
if (1) {
LatticeFermionF sr_eo(sFGrid);
LatticeFermionF serr(sFGrid);
LatticeFermion ssrc_e (sFrbGrid);
LatticeFermion ssrc_o (sFrbGrid);
LatticeFermion sr_e (sFrbGrid);
LatticeFermion sr_o (sFrbGrid);
pickCheckerboard(Even,ssrc_e,ssrc);
pickCheckerboard(Odd,ssrc_o,ssrc);
setCheckerboard(sr_eo,ssrc_o);
setCheckerboard(sr_eo,ssrc_e);
serr = sr_eo-ssrc;
std::cout<<GridLogMessage << "EO src norm diff "<< norm2(serr)<<std::endl;
sr_e = zero;
sr_o = zero;
double t0=usecond();
for(int i=0;i<ncall;i++){
sDw.DhopEO(ssrc_o,sr_e,DaggerNo);
}
double t1=usecond();
double volume=Ls; for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
double flops=(1344.0*volume*ncall)/2;
std::cout<<GridLogMessage << "sDeo mflop/s = "<< flops/(t1-t0)<<std::endl;
std::cout<<GridLogMessage << "sDeo mflop/s per node "<< flops/(t1-t0)/NP<<std::endl;
sDw.DhopEO(ssrc_o,sr_e,DaggerNo);
sDw.DhopOE(ssrc_e,sr_o,DaggerNo);
sDw.Dhop (ssrc ,sresult,DaggerNo);
pickCheckerboard(Even,ssrc_e,sresult);
pickCheckerboard(Odd ,ssrc_o,sresult);
ssrc_e = ssrc_e - sr_e;
std::cout<<GridLogMessage << "sE norm diff "<< norm2(ssrc_e)<<std::endl;
ssrc_o = ssrc_o - sr_o;
std::cout<<GridLogMessage << "sO norm diff "<< norm2(ssrc_o)<<std::endl;
}
}
if (1)

6
benchmarks/Make.inc
View File

@ -1,5 +1,5 @@
bin_PROGRAMS = Benchmark_comms Benchmark_dwf Benchmark_memory_asynch Benchmark_memory_bandwidth Benchmark_su3 Benchmark_wilson Benchmark_zmm
bin_PROGRAMS = Benchmark_comms Benchmark_dwf Benchmark_dwf_ntpf Benchmark_memory_asynch Benchmark_memory_bandwidth Benchmark_su3 Benchmark_wilson Benchmark_zmm
Benchmark_comms_SOURCES=Benchmark_comms.cc
@ -10,6 +10,10 @@ Benchmark_dwf_SOURCES=Benchmark_dwf.cc
Benchmark_dwf_LDADD=-lGrid
Benchmark_dwf_ntpf_SOURCES=Benchmark_dwf_ntpf.cc
Benchmark_dwf_ntpf_LDADD=-lGrid
Benchmark_memory_asynch_SOURCES=Benchmark_memory_asynch.cc
Benchmark_memory_asynch_LDADD=-lGrid