portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-14 13:57:07 +01:00
Code Releases Activity

Merge branch 'develop' into feature/gpu-port

Browse Source
This commit is contained in:
Peter Boyle
2018-12-13 05:11:34 +00:00
parent adbdc4e65b c509bd3fe2
commit b57a4d32aa
647 changed files with 49155 additions and 11160 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
tests/solver/Test_dwf_mrhs_cg.cc
View File

@ -114,7 +114,7 @@ int main (int argc, char ** argv)
{
FGrid->Barrier();
ScidacWriter _ScidacWriter;
ScidacWriter _ScidacWriter(FGrid->IsBoss());
_ScidacWriter.open(file);
std::cout << GridLogMessage << "****************************************************************** "<<std::endl;
std::cout << GridLogMessage << " Writing out gauge field "<<std::endl;
@ -144,7 +144,7 @@ int main (int argc, char ** argv)
std::cout << GridLogMessage << "****************************************************************** "<<std::endl;
std::stringstream filefn; filefn << filef << "."<< n;
ScidacWriter _ScidacWriter;
ScidacWriter _ScidacWriter(FGrid->IsBoss());
_ScidacWriter.open(filefn.str());
_ScidacWriter.writeScidacFieldRecord(src[n],record);
_ScidacWriter.close();

9
tests/solver/Test_dwf_mrhs_cg_mpi.cc
View File

@ -38,6 +38,7 @@ int main (int argc, char ** argv)
typedef typename DomainWallFermionR::ComplexField ComplexField;
typename DomainWallFermionR::ImplParams params;
double stp=1.0e-5;
const int Ls=4;
Grid_init(&argc,&argv);
@ -197,7 +198,7 @@ int main (int argc, char ** argv)
MdagMLinearOperator<DomainWallFermionR,FermionField> HermOp(Ddwf);
MdagMLinearOperator<DomainWallFermionR,FermionField> HermOpCk(Dchk);
ConjugateGradient<FermionField> CG((1.0e-2),10000);
ConjugateGradient<FermionField> CG((stp),10000);
s_res = Zero();
CG(HermOp,s_src,s_res);
@ -227,5 +228,11 @@ int main (int argc, char ** argv)
std::cout << GridLogMessage<<" resid["<<n<<"] "<< norm2(tmp)/norm2(src[n])<<std::endl;
}
for(int s=0;s<nrhs;s++) result[s]=zero;
int blockDim = 0;//not used for BlockCGVec
BlockConjugateGradient<FermionField> BCGV (BlockCGVec,blockDim,stp,10000);
BCGV.PrintInterval=10;
BCGV(HermOpCk,src,result);
Grid_finalize();
}

220
tests/solver/Test_mobius_bcg.cc Normal file
View File

@ -0,0 +1,220 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./tests/Test_dwf_mrhs_cg.cc
Copyright (C) 2015
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#include <Grid/Grid.h>
#include <Grid/algorithms/iterative/BlockConjugateGradient.h>
using namespace std;
using namespace Grid;
using namespace Grid::QCD;
int main (int argc, char ** argv)
{
typedef typename MobiusFermionR::FermionField FermionField;
typedef typename MobiusFermionR::ComplexField ComplexField;
typename MobiusFermionR::ImplParams params;
const int Ls=12;
Grid_init(&argc,&argv);
std::vector<int> latt_size = GridDefaultLatt();
std::vector<int> simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd());
std::vector<int> mpi_layout = GridDefaultMpi();
std::vector<int> mpi_split (mpi_layout.size(),1);
std::vector<int> split_coor (mpi_layout.size(),1);
std::vector<int> split_dim (mpi_layout.size(),1);
std::vector<ComplexD> boundary_phases(Nd,1.);
boundary_phases[Nd-1]=-1.;
params.boundary_phases = boundary_phases;
GridCartesian * UGrid = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(),
GridDefaultSimd(Nd,vComplex::Nsimd()),
GridDefaultMpi());
GridCartesian * FGrid = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
GridRedBlackCartesian * rbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid);
/////////////////////////////////////////////
// Split into 1^4 mpi communicators
/////////////////////////////////////////////
for(int i=0;i<argc;i++){
if(std::string(argv[i]) == "--split"){
for(int k=0;k<mpi_layout.size();k++){
std::stringstream ss;
ss << argv[i+1+k];
ss >> mpi_split[k];
}
break;
}
}
double stp = 1.e-8;
int nrhs = 1;
int me;
for(int i=0;i<mpi_layout.size();i++){
// split_dim[i] = (mpi_layout[i]/mpi_split[i]);
nrhs *= (mpi_layout[i]/mpi_split[i]);
// split_coor[i] = FGrid._processor_coor[i]/mpi_split[i];
}
std::cout << GridLogMessage << "Creating split grids " <<std::endl;
GridCartesian * SGrid = new GridCartesian(GridDefaultLatt(),
GridDefaultSimd(Nd,vComplex::Nsimd()),
mpi_split,
*UGrid,me);
std::cout << GridLogMessage <<"Creating split ferm grids " <<std::endl;
GridCartesian * SFGrid = SpaceTimeGrid::makeFiveDimGrid(Ls,SGrid);
std::cout << GridLogMessage <<"Creating split rb grids " <<std::endl;
GridRedBlackCartesian * SrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(SGrid);
std::cout << GridLogMessage <<"Creating split ferm rb grids " <<std::endl;
GridRedBlackCartesian * SFrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,SGrid);
std::cout << GridLogMessage << "Made the grids"<<std::endl;
///////////////////////////////////////////////
// Set up the problem as a 4d spreadout job
///////////////////////////////////////////////
std::vector<int> seeds({1,2,3,4});
std::vector<FermionField> src(nrhs,FGrid);
std::vector<FermionField> src_chk(nrhs,FGrid);
std::vector<FermionField> result(nrhs,FGrid);
FermionField tmp(FGrid);
std::cout << GridLogMessage << "Made the Fermion Fields"<<std::endl;
for(int s=0;s<nrhs;s++) result[s]=zero;
GridParallelRNG pRNG5(FGrid); pRNG5.SeedFixedIntegers(seeds);
for(int s=0;s<nrhs;s++) {
random(pRNG5,src[s]);
std::cout << GridLogMessage << " src ["<<s<<"] "<<norm2(src[s])<<std::endl;
}
std::cout << GridLogMessage << "Intialised the Fermion Fields"<<std::endl;
LatticeGaugeField Umu(UGrid);
if(0) {
FieldMetaData header;
std::string file("./lat.in");
NerscIO::readConfiguration(Umu,header,file);
std::cout << GridLogMessage << " "<<file<<" successfully read" <<std::endl;
} else {
GridParallelRNG pRNG(UGrid );
std::cout << GridLogMessage << "Intialising 4D RNG "<<std::endl;
pRNG.SeedFixedIntegers(seeds);
std::cout << GridLogMessage << "Intialised 4D RNG "<<std::endl;
SU3::HotConfiguration(pRNG,Umu);
std::cout << GridLogMessage << "Intialised the HOT Gauge Field"<<std::endl;
std::cout << " Site zero "<< Umu._odata[0] <<std::endl;
}
/////////////////
// MPI only sends
/////////////////
LatticeGaugeField s_Umu(SGrid);
FermionField s_src(SFGrid);
FermionField s_tmp(SFGrid);
FermionField s_res(SFGrid);
std::cout << GridLogMessage << "Made the split grid fields"<<std::endl;
///////////////////////////////////////////////////////////////
// split the source out using MPI instead of I/O
///////////////////////////////////////////////////////////////
Grid_split (Umu,s_Umu);
Grid_split (src,s_src);
std::cout << GridLogMessage << " split rank " <<me << " s_src "<<norm2(s_src)<<std::endl;
///////////////////////////////////////////////////////////////
// Set up N-solvers as trivially parallel
///////////////////////////////////////////////////////////////
std::cout << GridLogMessage << " Building the solvers"<<std::endl;
// RealD mass=0.00107;
RealD mass=0.1;
RealD M5=1.8;
RealD mobius_factor=32./12.;
RealD mobius_b=0.5*(mobius_factor+1.);
RealD mobius_c=0.5*(mobius_factor-1.);
MobiusFermionR Dchk(Umu,*FGrid,*FrbGrid,*UGrid,*rbGrid,mass,M5,mobius_b,mobius_c,params);
MobiusFermionR Ddwf(s_Umu,*SFGrid,*SFrbGrid,*SGrid,*SrbGrid,mass,M5,mobius_b,mobius_c,params);
std::cout << GridLogMessage << "****************************************************************** "<<std::endl;
std::cout << GridLogMessage << " Calling DWF CG "<<std::endl;
std::cout << GridLogMessage << "****************************************************************** "<<std::endl;
MdagMLinearOperator<MobiusFermionR,FermionField> HermOp(Ddwf);
MdagMLinearOperator<MobiusFermionR,FermionField> HermOpCk(Dchk);
ConjugateGradient<FermionField> CG((stp),100000);
s_res = zero;
CG(HermOp,s_src,s_res);
std::cout << GridLogMessage << " split residual norm "<<norm2(s_res)<<std::endl;
/////////////////////////////////////////////////////////////
// Report how long they all took
/////////////////////////////////////////////////////////////
std::vector<uint32_t> iterations(nrhs,0);
iterations[me] = CG.IterationsToComplete;
for(int n=0;n<nrhs;n++){
UGrid->GlobalSum(iterations[n]);
std::cout << GridLogMessage<<" Rank "<<n<<" "<< iterations[n]<<" CG iterations"<<std::endl;
}
/////////////////////////////////////////////////////////////
// Gather and residual check on the results
/////////////////////////////////////////////////////////////
std::cout << GridLogMessage<< "Unsplitting the result"<<std::endl;
Grid_unsplit(result,s_res);
std::cout << GridLogMessage<< "Checking the residuals"<<std::endl;
for(int n=0;n<nrhs;n++){
std::cout << GridLogMessage<< " res["<<n<<"] norm "<<norm2(result[n])<<std::endl;
HermOpCk.HermOp(result[n],tmp); tmp = tmp - src[n];
std::cout << GridLogMessage<<" resid["<<n<<"] "<< std::sqrt(norm2(tmp)/norm2(src[n]))<<std::endl;
}
for(int s=0;s<nrhs;s++){
result[s]=zero;
}
/////////////////////////////////////////////////////////////
// Try block CG
/////////////////////////////////////////////////////////////
int blockDim = 0;//not used for BlockCGVec
BlockConjugateGradient<FermionField> BCGV (BlockCGrQVec,blockDim,stp,100000);
{
BCGV(HermOpCk,src,result);
}
Grid_finalize();
}

144
tests/solver/Test_mobius_bcg_nosplit.cc Normal file
View File

@ -0,0 +1,144 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./tests/Test_dwf_mrhs_cg.cc
Copyright (C) 2015
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#include <Grid/Grid.h>
#include <Grid/algorithms/iterative/BlockConjugateGradient.h>
using namespace std;
using namespace Grid;
using namespace Grid::QCD;
int main (int argc, char ** argv)
{
typedef typename DomainWallFermionR::FermionField FermionField;
typedef typename DomainWallFermionR::ComplexField ComplexField;
typename DomainWallFermionR::ImplParams params;
const int Ls=16;
Grid_init(&argc,&argv);
std::vector<int> latt_size = GridDefaultLatt();
std::vector<int> simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd());
std::vector<int> mpi_layout = GridDefaultMpi();
std::vector<ComplexD> boundary_phases(Nd,1.);
boundary_phases[Nd-1]=-1.;
params.boundary_phases = boundary_phases;
GridCartesian * UGrid = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(),
GridDefaultSimd(Nd,vComplex::Nsimd()),
GridDefaultMpi());
GridCartesian * FGrid = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
GridRedBlackCartesian * rbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid);
double stp = 1.e-8;
int nrhs = 2;
///////////////////////////////////////////////
// Set up the problem as a 4d spreadout job
///////////////////////////////////////////////
std::vector<int> seeds({1,2,3,4});
std::vector<FermionField> src(nrhs,FGrid);
std::vector<FermionField> src_chk(nrhs,FGrid);
std::vector<FermionField> result(nrhs,FGrid);
FermionField tmp(FGrid);
std::cout << GridLogMessage << "Made the Fermion Fields"<<std::endl;
for(int s=0;s<nrhs;s++) result[s]=zero;
GridParallelRNG pRNG5(FGrid); pRNG5.SeedFixedIntegers(seeds);
for(int s=0;s<nrhs;s++) {
random(pRNG5,src[s]);
std::cout << GridLogMessage << " src ["<<s<<"] "<<norm2(src[s])<<std::endl;
}
std::cout << GridLogMessage << "Intialised the Fermion Fields"<<std::endl;
LatticeGaugeField Umu(UGrid);
int conf = 0;
if(conf==0) {
FieldMetaData header;
std::string file("./lat.in");
NerscIO::readConfiguration(Umu,header,file);
std::cout << GridLogMessage << " Config "<<file<<" successfully read" <<std::endl;
} else if (conf==1){
GridParallelRNG pRNG(UGrid );
pRNG.SeedFixedIntegers(seeds);
SU3::HotConfiguration(pRNG,Umu);
std::cout << GridLogMessage << "Intialised the HOT Gauge Field"<<std::endl;
} else {
SU3::ColdConfiguration(Umu);
std::cout << GridLogMessage << "Intialised the COLD Gauge Field"<<std::endl;
}
///////////////////////////////////////////////////////////////
// Set up N-solvers as trivially parallel
///////////////////////////////////////////////////////////////
std::cout << GridLogMessage << " Building the solvers"<<std::endl;
RealD mass=0.01;
RealD M5=1.8;
DomainWallFermionR Ddwf(Umu,*FGrid,*FrbGrid,*UGrid,*rbGrid,mass,M5,params);
std::cout << GridLogMessage << "****************************************************************** "<<std::endl;
std::cout << GridLogMessage << " Calling DWF CG "<<std::endl;
std::cout << GridLogMessage << "****************************************************************** "<<std::endl;
MdagMLinearOperator<DomainWallFermionR,FermionField> HermOp(Ddwf);
ConjugateGradient<FermionField> CG((stp),100000);
for(int rhs=0;rhs<1;rhs++){
result[rhs] = zero;
CG(HermOp,src[rhs],result[rhs]);
}
for(int rhs=0;rhs<1;rhs++){
std::cout << " Result["<<rhs<<"] norm = "<<norm2(result[rhs])<<std::endl;
}
/////////////////////////////////////////////////////////////
// Try block CG
/////////////////////////////////////////////////////////////
int blockDim = 0;//not used for BlockCGVec
for(int s=0;s<nrhs;s++){
result[s]=zero;
}
BlockConjugateGradient<FermionField> BCGV (BlockCGrQVec,blockDim,stp,100000);
{
BCGV(HermOp,src,result);
}
for(int rhs=0;rhs<nrhs;rhs++){
std::cout << " Result["<<rhs<<"] norm = "<<norm2(result[rhs])<<std::endl;
}
Grid_finalize();
}

148
tests/solver/Test_mobius_bcg_phys_nosplit.cc Normal file
View File

@ -0,0 +1,148 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./tests/Test_dwf_mrhs_cg.cc
Copyright (C) 2015
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#include <Grid/Grid.h>
#include <Grid/algorithms/iterative/BlockConjugateGradient.h>
using namespace std;
using namespace Grid;
using namespace Grid::QCD;
int main (int argc, char ** argv)
{
typedef typename DomainWallFermionR::FermionField FermionField;
typedef typename DomainWallFermionR::ComplexField ComplexField;
typename DomainWallFermionR::ImplParams params;
const int Ls=16;
Grid_init(&argc,&argv);
std::vector<int> latt_size = GridDefaultLatt();
std::vector<int> simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd());
std::vector<int> mpi_layout = GridDefaultMpi();
std::vector<ComplexD> boundary_phases(Nd,1.);
boundary_phases[Nd-1]=-1.;
params.boundary_phases = boundary_phases;
GridCartesian * UGrid = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(),
GridDefaultSimd(Nd,vComplex::Nsimd()),
GridDefaultMpi());
GridCartesian * FGrid = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
GridRedBlackCartesian * rbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid);
double stp = 1.e-8;
int nrhs = 2;
///////////////////////////////////////////////
// Set up the problem as a 4d spreadout job
///////////////////////////////////////////////
std::vector<int> seeds({1,2,3,4});
std::vector<FermionField> src4(nrhs,UGrid);
std::vector<FermionField> src(nrhs,FGrid);
std::vector<FermionField> src_chk(nrhs,FGrid);
std::vector<FermionField> result(nrhs,FGrid);
FermionField tmp(FGrid);
std::cout << GridLogMessage << "Made the Fermion Fields"<<std::endl;
for(int s=0;s<nrhs;s++) result[s]=zero;
GridParallelRNG pRNG4(UGrid); pRNG4.SeedFixedIntegers(seeds);
for(int s=0;s<nrhs;s++) {
random(pRNG4,src4[s]);
std::cout << GridLogMessage << " src ["<<s<<"] "<<norm2(src[s])<<std::endl;
}
std::cout << GridLogMessage << "Intialised the Fermion Fields"<<std::endl;
LatticeGaugeField Umu(UGrid);
int conf = 0;
if(conf==0) {
FieldMetaData header;
std::string file("./lat.in");
NerscIO::readConfiguration(Umu,header,file);
std::cout << GridLogMessage << " Config "<<file<<" successfully read" <<std::endl;
} else if (conf==1){
GridParallelRNG pRNG(UGrid );
pRNG.SeedFixedIntegers(seeds);
SU3::HotConfiguration(pRNG,Umu);
std::cout << GridLogMessage << "Intialised the HOT Gauge Field"<<std::endl;
} else {
SU3::ColdConfiguration(Umu);
std::cout << GridLogMessage << "Intialised the COLD Gauge Field"<<std::endl;
}
///////////////////////////////////////////////////////////////
// Set up N-solvers as trivially parallel
///////////////////////////////////////////////////////////////
std::cout << GridLogMessage << " Building the solvers"<<std::endl;
RealD mass=0.01;
RealD M5=1.8;
DomainWallFermionR Ddwf(Umu,*FGrid,*FrbGrid,*UGrid,*rbGrid,mass,M5,params);
for(int s=0;s<nrhs;s++) {
Ddwf.ImportPhysicalFermionSource(src4[s],src[s]);
}
std::cout << GridLogMessage << "****************************************************************** "<<std::endl;
std::cout << GridLogMessage << " Calling DWF CG "<<std::endl;
std::cout << GridLogMessage << "****************************************************************** "<<std::endl;
MdagMLinearOperator<DomainWallFermionR,FermionField> HermOp(Ddwf);
ConjugateGradient<FermionField> CG((stp),100000);
for(int rhs=0;rhs<1;rhs++){
result[rhs] = zero;
// CG(HermOp,src[rhs],result[rhs]);
}
for(int rhs=0;rhs<1;rhs++){
std::cout << " Result["<<rhs<<"] norm = "<<norm2(result[rhs])<<std::endl;
}
/////////////////////////////////////////////////////////////
// Try block CG
/////////////////////////////////////////////////////////////
int blockDim = 0;//not used for BlockCGVec
for(int s=0;s<nrhs;s++){
result[s]=zero;
}
BlockConjugateGradient<FermionField> BCGV (BlockCGrQVec,blockDim,stp,100000);
{
BCGV(HermOp,src,result);
}
for(int rhs=0;rhs<nrhs;rhs++){
std::cout << " Result["<<rhs<<"] norm = "<<norm2(result[rhs])<<std::endl;
}
Grid_finalize();
}

147
tests/solver/Test_mobius_bcg_prec_nosplit.cc Normal file
View File

@ -0,0 +1,147 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./tests/Test_dwf_mrhs_cg.cc
Copyright (C) 2015
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#include <Grid/Grid.h>
#include <Grid/algorithms/iterative/BlockConjugateGradient.h>
using namespace std;
using namespace Grid;
using namespace Grid::QCD;
int main (int argc, char ** argv)
{
typedef typename DomainWallFermionR::FermionField FermionField;
typedef typename DomainWallFermionR::ComplexField ComplexField;
typename DomainWallFermionR::ImplParams params;
const int Ls=16;
Grid_init(&argc,&argv);
std::vector<int> latt_size = GridDefaultLatt();
std::vector<int> simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd());
std::vector<int> mpi_layout = GridDefaultMpi();
std::vector<ComplexD> boundary_phases(Nd,1.);
boundary_phases[Nd-1]=-1.;
params.boundary_phases = boundary_phases;
GridCartesian * UGrid = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(),
GridDefaultSimd(Nd,vComplex::Nsimd()),
GridDefaultMpi());
GridCartesian * FGrid = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
GridRedBlackCartesian * rbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid);
double stp = 1.e-8;
int nrhs = 4;
///////////////////////////////////////////////
// Set up the problem as a 4d spreadout job
///////////////////////////////////////////////
std::vector<int> seeds({1,2,3,4});
std::vector<FermionField> src(nrhs,FGrid);
std::vector<FermionField> src_chk(nrhs,FGrid);
std::vector<FermionField> result(nrhs,FGrid);
FermionField tmp(FGrid);
std::cout << GridLogMessage << "Made the Fermion Fields"<<std::endl;
for(int s=0;s<nrhs;s++) result[s]=zero;
GridParallelRNG pRNG5(FGrid); pRNG5.SeedFixedIntegers(seeds);
for(int s=0;s<nrhs;s++) {
random(pRNG5,src[s]);
std::cout << GridLogMessage << " src ["<<s<<"] "<<norm2(src[s])<<std::endl;
}
std::cout << GridLogMessage << "Intialised the Fermion Fields"<<std::endl;
LatticeGaugeField Umu(UGrid);
int conf = 2;
if(conf==0) {
FieldMetaData header;
std::string file("./lat.in");
NerscIO::readConfiguration(Umu,header,file);
std::cout << GridLogMessage << " Config "<<file<<" successfully read" <<std::endl;
} else if (conf==1){
GridParallelRNG pRNG(UGrid );
pRNG.SeedFixedIntegers(seeds);
SU3::HotConfiguration(pRNG,Umu);
std::cout << GridLogMessage << "Intialised the HOT Gauge Field"<<std::endl;
} else {
SU3::ColdConfiguration(Umu);
std::cout << GridLogMessage << "Intialised the COLD Gauge Field"<<std::endl;
}
///////////////////////////////////////////////////////////////
// Set up N-solvers as trivially parallel
///////////////////////////////////////////////////////////////
std::cout << GridLogMessage << " Building the solvers"<<std::endl;
RealD mass=0.01;
RealD M5=1.8;
DomainWallFermionR Ddwf(Umu,*FGrid,*FrbGrid,*UGrid,*rbGrid,mass,M5,params);
std::cout << GridLogMessage << "****************************************************************** "<<std::endl;
std::cout << GridLogMessage << " Calling DWF CG "<<std::endl;
std::cout << GridLogMessage << "****************************************************************** "<<std::endl;
MdagMLinearOperator<DomainWallFermionR,FermionField> HermOp(Ddwf);
ConjugateGradient<FermionField> CG((stp),100000);
for(int rhs=0;rhs<1;rhs++){
result[rhs] = zero;
CG(HermOp,src[rhs],result[rhs]);
}
for(int rhs=0;rhs<1;rhs++){
std::cout << " Result["<<rhs<<"] norm = "<<norm2(result[rhs])<<std::endl;
}
/////////////////////////////////////////////////////////////
// Try block CG
/////////////////////////////////////////////////////////////
int blockDim = 0;//not used for BlockCGVec
for(int s=0;s<nrhs;s++){
result[s]=zero;
}
{
BlockConjugateGradient<FermionField> BCGV (BlockCGrQVec,blockDim,stp,100000);
SchurRedBlackDiagTwoSolve<FermionField> SchurSolver(BCGV);
SchurSolver(Ddwf,src,result);
}
for(int rhs=0;rhs<nrhs;rhs++){
std::cout << " Result["<<rhs<<"] norm = "<<norm2(result[rhs])<<std::endl;
}
Grid_finalize();
}

128
tests/solver/Test_staggered_block_cg_prec.cc
View File

@ -67,34 +67,70 @@ int main (int argc, char ** argv)
GridParallelRNG pRNG(UGrid ); pRNG.SeedFixedIntegers(seeds);
GridParallelRNG pRNG5(FGrid); pRNG5.SeedFixedIntegers(seeds);
FermionField src(FGrid); random(pRNG5,src);
FermionField src(FGrid);
FermionField tt(FGrid);
#if 1
random(pRNG5,src);
#else
src=zero;
ComplexField coor(FGrid);
LatticeCoordinate(coor,0);
for(int ss=0;ss<FGrid->oSites();ss++){
src._odata[ss]()()(0)=coor._odata[ss]()()();
}
LatticeCoordinate(coor,1);
for(int ss=0;ss<FGrid->oSites();ss++){
src._odata[ss]()()(0)+=coor._odata[ss]()()();
}
#endif
FermionField src_o(FrbGrid); pickCheckerboard(Odd,src_o,src);
FermionField result_o(FrbGrid); result_o=Zero();
RealD nrm = norm2(src);
LatticeGaugeField Umu(UGrid); SU3::HotConfiguration(pRNG,Umu);
double volume=1;
for(int mu=0;mu<Nd;mu++){
volume=volume*latt_size[mu];
}
RealD mass=0.003;
ImprovedStaggeredFermion5DR Ds(Umu,Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass);
RealD c1=9.0/8.0;
RealD c2=-1.0/24.0;
RealD u0=1.0;
ImprovedStaggeredFermion5DR Ds(Umu,Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,c1,c2,u0);
SchurStaggeredOperator<ImprovedStaggeredFermion5DR,FermionField> HermOp(Ds);
ConjugateGradient<FermionField> CG(1.0e-8,10000);
int blockDim = 0;
BlockConjugateGradient<FermionField> BCGrQ(BlockCGrQ,blockDim,1.0e-8,10000);
BlockConjugateGradient<FermionField> BCG (BlockCG,blockDim,1.0e-8,10000);
BlockConjugateGradient<FermionField> BCG (BlockCGrQ,blockDim,1.0e-8,10000);
BlockConjugateGradient<FermionField> BCGv (BlockCGrQVec,blockDim,1.0e-8,10000);
BlockConjugateGradient<FermionField> mCG (CGmultiRHS,blockDim,1.0e-8,10000);
std::cout << GridLogMessage << "****************************************************************** "<<std::endl;
std::cout << GridLogMessage << " Calling 4d CG "<<std::endl;
std::cout << GridLogMessage << "****************************************************************** "<<std::endl;
ImprovedStaggeredFermionR Ds4d(Umu,Umu,*UGrid,*UrbGrid,mass);
ImprovedStaggeredFermionR Ds4d(Umu,Umu,*UGrid,*UrbGrid,mass,c1,c2,u0);
SchurStaggeredOperator<ImprovedStaggeredFermionR,FermionField> HermOp4d(Ds4d);
FermionField src4d(UGrid); random(pRNG,src4d);
FermionField src4d_o(UrbGrid); pickCheckerboard(Odd,src4d_o,src4d);
FermionField result4d_o(UrbGrid);
result4d_o=Zero();
CG(HermOp4d,src4d_o,result4d_o);
double deodoe_flops=(16*(3*(6+8+8)) + 15*3*2)*volume; // == 66*16 + == 1146
{
double t1=usecond();
CG(HermOp4d,src4d_o,result4d_o);
double t2=usecond();
double ncall=CG.IterationsToComplete;
double flops = deodoe_flops * ncall;
std::cout<<GridLogMessage << "usec = "<< (t2-t1)<<std::endl;
std::cout<<GridLogMessage << "flops = "<< flops<<std::endl;
std::cout<<GridLogMessage << "mflop/s = "<< flops/(t2-t1)<<std::endl;
HermOp4d.Report();
}
Ds4d.Report();
std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
@ -103,7 +139,17 @@ int main (int argc, char ** argv)
std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
Ds.ZeroCounters();
result_o=Zero();
CG(HermOp,src_o,result_o);
{
double t1=usecond();
CG(HermOp,src_o,result_o);
double t2=usecond();
double ncall=CG.IterationsToComplete*Ls;
double flops = deodoe_flops * ncall;
std::cout<<GridLogMessage << "usec = "<< (t2-t1)<<std::endl;
std::cout<<GridLogMessage << "flops = "<< flops<<std::endl;
std::cout<<GridLogMessage << "mflop/s = "<< flops/(t2-t1)<<std::endl;
HermOp.Report();
}
Ds.Report();
std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
@ -112,7 +158,37 @@ int main (int argc, char ** argv)
std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
Ds.ZeroCounters();
result_o=Zero();
mCG(HermOp,src_o,result_o);
{
double t1=usecond();
mCG(HermOp,src_o,result_o);
double t2=usecond();
double ncall=mCG.IterationsToComplete*Ls;
double flops = deodoe_flops * ncall;
std::cout<<GridLogMessage << "usec = "<< (t2-t1)<<std::endl;
std::cout<<GridLogMessage << "flops = "<< flops<<std::endl;
std::cout<<GridLogMessage << "mflop/s = "<< flops/(t2-t1)<<std::endl;
HermOp.Report();
}
Ds.Report();
std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
std::cout << GridLogMessage << " Calling Block CGrQ for "<<Ls <<" right hand sides" <<std::endl;
std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
Ds.ZeroCounters();
result_o=Zero();
{
double t1=usecond();
BCGrQ(HermOp,src_o,result_o);
double t2=usecond();
double ncall=BCGrQ.IterationsToComplete*Ls;
double flops = deodoe_flops * ncall;
std::cout<<GridLogMessage << "usec = "<< (t2-t1)<<std::endl;
std::cout<<GridLogMessage << "flops = "<< flops<<std::endl;
std::cout<<GridLogMessage << "mflop/s = "<< flops/(t2-t1)<<std::endl;
HermOp.Report();
}
Ds.Report();
std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
@ -120,11 +196,45 @@ int main (int argc, char ** argv)
std::cout << GridLogMessage << " Calling Block CG for "<<Ls <<" right hand sides" <<std::endl;
std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
Ds.ZeroCounters();
result_o=Zero();
BCGrQ(HermOp,src_o,result_o);
result_o=zero;
{
double t1=usecond();
BCG(HermOp,src_o,result_o);
double t2=usecond();
double ncall=BCGrQ.IterationsToComplete*Ls;
double flops = deodoe_flops * ncall;
std::cout<<GridLogMessage << "usec = "<< (t2-t1)<<std::endl;
std::cout<<GridLogMessage << "flops = "<< flops<<std::endl;
std::cout<<GridLogMessage << "mflop/s = "<< flops/(t2-t1)<<std::endl;
HermOp.Report();
}
Ds.Report();
std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
std::cout << GridLogMessage << "****************************************************************** "<<std::endl;
std::cout << GridLogMessage << " Calling BCGvec "<<std::endl;
std::cout << GridLogMessage << "****************************************************************** "<<std::endl;
std::vector<FermionField> src_v (Ls,UrbGrid);
std::vector<FermionField> result_v(Ls,UrbGrid);
for(int s=0;s<Ls;s++) result_v[s] = zero;
for(int s=0;s<Ls;s++) {
FermionField src4(UGrid);
ExtractSlice(src4,src,s,0);
pickCheckerboard(Odd,src_v[s],src4);
}
{
double t1=usecond();
BCGv(HermOp4d,src_v,result_v);
double t2=usecond();
double ncall=BCGv.IterationsToComplete*Ls;
double flops = deodoe_flops * ncall;
std::cout<<GridLogMessage << "usec = "<< (t2-t1)<<std::endl;
std::cout<<GridLogMessage << "flops = "<< flops<<std::endl;
std::cout<<GridLogMessage << "mflop/s = "<< flops/(t2-t1)<<std::endl;
// HermOp4d.Report();
}
Grid_finalize();
}

54
tests/solver/Test_staggered_block_cg_unprec.cc
View File

@ -74,7 +74,16 @@ int main (int argc, char ** argv)
LatticeGaugeField Umu(UGrid); SU3::HotConfiguration(pRNG,Umu);
RealD mass=0.003;
ImprovedStaggeredFermion5DR Ds(Umu,Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass);
RealD c1=9.0/8.0;
RealD c2=-1.0/24.0;
RealD u0=1.0;
double volume=1;
for(int mu=0;mu<Nd;mu++){
volume=volume*latt_size[mu];
}
ImprovedStaggeredFermion5DR Ds(Umu,Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,c1,c2,u0);
MdagMLinearOperator<ImprovedStaggeredFermion5DR,FermionField> HermOp(Ds);
ConjugateGradient<FermionField> CG(1.0e-8,10000);
@ -86,11 +95,23 @@ int main (int argc, char ** argv)
std::cout << GridLogMessage << "****************************************************************** "<<std::endl;
std::cout << GridLogMessage << " Calling 4d CG "<<std::endl;
std::cout << GridLogMessage << "****************************************************************** "<<std::endl;
ImprovedStaggeredFermionR Ds4d(Umu,Umu,*UGrid,*UrbGrid,mass);
ImprovedStaggeredFermionR Ds4d(Umu,Umu,*UGrid,*UrbGrid,mass,c1,c2,u0);
MdagMLinearOperator<ImprovedStaggeredFermionR,FermionField> HermOp4d(Ds4d);
FermionField src4d(UGrid); random(pRNG,src4d);
FermionField result4d(UGrid); result4d=Zero();
CG(HermOp4d,src4d,result4d);
double deodoe_flops=(16*(3*(6+8+8)) + 15*3*2)*volume; // == 66*16 + == 1146
{
double t1=usecond();
CG(HermOp4d,src4d,result4d);
double t2=usecond();
double ncall=CG.IterationsToComplete;
double flops = deodoe_flops * ncall;
std::cout<<GridLogMessage << "usec = "<< (t2-t1)<<std::endl;
std::cout<<GridLogMessage << "flops = "<< flops<<std::endl;
std::cout<<GridLogMessage << "mflop/s = "<< flops/(t2-t1)<<std::endl;
}
std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
@ -98,9 +119,18 @@ int main (int argc, char ** argv)
std::cout << GridLogMessage << " Calling 5d CG for "<<Ls <<" right hand sides" <<std::endl;
std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
result=Zero();
{
Ds.ZeroCounters();
double t1=usecond();
CG(HermOp,src,result);
double t2=usecond();
double ncall=CG.IterationsToComplete;
double flops = deodoe_flops * ncall;
std::cout<<GridLogMessage << "usec = "<< (t2-t1)<<std::endl;
std::cout<<GridLogMessage << "flops = "<< flops<<std::endl;
std::cout<<GridLogMessage << "mflop/s = "<< flops/(t2-t1)<<std::endl;
Ds.Report();
}
std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
@ -108,7 +138,16 @@ int main (int argc, char ** argv)
std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
result=Zero();
Ds.ZeroCounters();
{
double t1=usecond();
mCG(HermOp,src,result);
double t2=usecond();
double ncall=CG.IterationsToComplete;
double flops = deodoe_flops * ncall;
std::cout<<GridLogMessage << "usec = "<< (t2-t1)<<std::endl;
std::cout<<GridLogMessage << "flops = "<< flops<<std::endl;
std::cout<<GridLogMessage << "mflop/s = "<< flops/(t2-t1)<<std::endl;
}
Ds.Report();
std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
@ -117,7 +156,16 @@ int main (int argc, char ** argv)
std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
result=Zero();
Ds.ZeroCounters();
{
double t1=usecond();
BCGrQ(HermOp,src,result);
double t2=usecond();
double ncall=CG.IterationsToComplete;
double flops = deodoe_flops * ncall;
std::cout<<GridLogMessage << "usec = "<< (t2-t1)<<std::endl;
std::cout<<GridLogMessage << "flops = "<< flops<<std::endl;
std::cout<<GridLogMessage << "mflop/s = "<< flops/(t2-t1)<<std::endl;
}
Ds.Report();
std::cout << GridLogMessage << "************************************************************************ "<<std::endl;

17
tests/solver/Test_staggered_cg_prec.cc
View File

@ -71,7 +71,10 @@ int main (int argc, char ** argv)
}
RealD mass=0.003;
ImprovedStaggeredFermionR Ds(Umu,Umu,Grid,RBGrid,mass);
RealD c1=9.0/8.0;
RealD c2=-1.0/24.0;
RealD u0=1.0;
ImprovedStaggeredFermionR Ds(Umu,Umu,Grid,RBGrid,mass,c1,c2,u0);
FermionField res_o(&RBGrid);
FermionField src_o(&RBGrid);
@ -80,7 +83,19 @@ int main (int argc, char ** argv)
SchurStaggeredOperator<ImprovedStaggeredFermionR,FermionField> HermOpEO(Ds);
ConjugateGradient<FermionField> CG(1.0e-8,10000);
double t1=usecond();
CG(HermOpEO,src_o,res_o);
double t2=usecond();
// Schur solver: uses DeoDoe => volume * 1146
double ncall=CG.IterationsToComplete;
double flops=(16*(3*(6+8+8)) + 15*3*2)*volume*ncall; // == 66*16 + == 1146
std::cout<<GridLogMessage << "usec = "<< (t2-t1)<<std::endl;
std::cout<<GridLogMessage << "flops = "<< flops<<std::endl;
std::cout<<GridLogMessage << "mflop/s = "<< flops/(t2-t1)<<std::endl;
FermionField tmp(&RBGrid);

5
tests/solver/Test_staggered_cg_schur.cc
View File

@ -65,7 +65,10 @@ int main (int argc, char ** argv)
FermionField resid(&Grid);
RealD mass=0.1;
ImprovedStaggeredFermionR Ds(Umu,Umu,Grid,RBGrid,mass);
RealD c1=9.0/8.0;
RealD c2=-1.0/24.0;
RealD u0=1.0;
ImprovedStaggeredFermionR Ds(Umu,Umu,Grid,RBGrid,mass,c1,c2,u0);
ConjugateGradient<FermionField> CG(1.0e-8,10000);
SchurRedBlackStaggeredSolve<FermionField> SchurSolver(CG);

5
tests/solver/Test_staggered_cg_unprec.cc
View File

@ -73,7 +73,10 @@ int main (int argc, char ** argv)
}
RealD mass=0.1;
ImprovedStaggeredFermionR Ds(Umu,Umu,Grid,RBGrid,mass);
RealD c1=9.0/8.0;
RealD c2=-1.0/24.0;
RealD u0=1.0;
ImprovedStaggeredFermionR Ds(Umu,Umu,Grid,RBGrid,mass,c1,c2,u0);
MdagMLinearOperator<ImprovedStaggeredFermionR,FermionField> HermOp(Ds);
ConjugateGradient<FermionField> CG(1.0e-6,10000);

121
tests/solver/Test_staggered_multishift.cc Normal file
View File

@ -0,0 +1,121 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./tests/Test_wilson_cg_unprec.cc
Copyright (C) 2015
Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#include <Grid/Grid.h>
#include <Grid/algorithms/iterative/BlockConjugateGradient.h>
using namespace std;
using namespace Grid;
using namespace Grid::QCD;
template<class d>
struct scal {
d internal;
};
Gamma::Algebra Gmu [] = {
Gamma::Algebra::GammaX,
Gamma::Algebra::GammaY,
Gamma::Algebra::GammaZ,
Gamma::Algebra::GammaT
};
int main (int argc, char ** argv)
{
typedef typename ImprovedStaggeredFermionR::FermionField FermionField;
typename ImprovedStaggeredFermionR::ImplParams params;
Grid_init(&argc,&argv);
std::vector<int> latt_size = GridDefaultLatt();
std::vector<int> simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd());
std::vector<int> mpi_layout = GridDefaultMpi();
GridCartesian Grid(latt_size,simd_layout,mpi_layout);
GridRedBlackCartesian RBGrid(&Grid);
std::vector<int> seeds({1,2,3,4});
GridParallelRNG pRNG(&Grid); pRNG.SeedFixedIntegers(seeds);
LatticeGaugeField Umu(&Grid); SU3::HotConfiguration(pRNG,Umu);
double volume=1;
for(int mu=0;mu<Nd;mu++){
volume=volume*latt_size[mu];
}
////////////////////////////////////////
// sqrt
////////////////////////////////////////
double lo=0.001;
double hi=1.0;
int precision=64;
int degree=10;
AlgRemez remez(lo,hi,precision);
remez.generateApprox(degree,1,2);
MultiShiftFunction Sqrt(remez,1.0e-6,false);
std::cout<<GridLogMessage << "Generating degree "<<degree<<" for x^(1/2)"<<std::endl;
////////////////////////////////////////////
// Setup staggered
////////////////////////////////////////////
RealD mass=0.003;
RealD c1=9.0/8.0;
RealD c2=-1.0/24.0;
RealD u0=1.0;
ImprovedStaggeredFermionR Ds(Umu,Umu,Grid,RBGrid,mass,c1,c2,u0);
SchurStaggeredOperator<ImprovedStaggeredFermionR,FermionField> HermOpEO(Ds);
FermionField src(&Grid); random(pRNG,src);
FermionField src_o(&RBGrid);
pickCheckerboard(Odd,src_o,src);
/////////////////////////////////
//Multishift CG
/////////////////////////////////
std::vector<FermionField> result(degree,&RBGrid);
ConjugateGradientMultiShift<FermionField> MSCG(10000,Sqrt);
double deodoe_flops=(1205+15*degree)*volume; // == 66*16 + == 1146
double t1=usecond();
MSCG(HermOpEO,src_o,result);
double t2=usecond();
double ncall=MSCG.IterationsToComplete;
double flops = deodoe_flops * ncall;
std::cout<<GridLogMessage << "usec = "<< (t2-t1)<<std::endl;
std::cout<<GridLogMessage << "flops = "<< flops<<std::endl;
std::cout<<GridLogMessage << "mflop/s = "<< flops/(t2-t1)<<std::endl;
// HermOpEO.Report();
Grid_finalize();
}