portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-14 09:45:36 +00:00
Code Releases Activity
685eea3d0f
Grid/tests/Test_cayley_even_odd_vec.cc

289 lines
12 KiB
C++
Raw Normal View History

Adding files for multiple implementations (cache opt) and Ls vectorisation of the 5D cayley form chiral fermions for the 5d matrix. With Ls entirely in the vector direction, s-hopping terms involve rotations. The serial dependence of the LDU inversion for Mobius and 4d even odd checkerboarding is removed by simply applying Ls^2 operations (vectorised many ways) as a dense matrix operation. This should give similar throughput but high flops (non-compulsory flops) but enable use of the KNL cache friendly kernels throughout the code. Ls is still constrained to be a multiple of Nsimd, which is as much as 8 for AVX512 with single precision.
2016-07-14 22:59:21 +01:00
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./tests/Test_cayley_even_odd.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>
using namespace std;
using namespace Grid;
GLobal edit for QCD namespace removal & NAMESPACE macros
2018-01-15 09:37:58 +00:00
;
Adding files for multiple implementations (cache opt) and Ls vectorisation of the 5D cayley form chiral fermions for the 5d matrix. With Ls entirely in the vector direction, s-hopping terms involve rotations. The serial dependence of the LDU inversion for Mobius and 4d even odd checkerboarding is removed by simply applying Ls^2 operations (vectorised many ways) as a dense matrix operation. This should give similar throughput but high flops (non-compulsory flops) but enable use of the KNL cache friendly kernels throughout the code. Ls is still constrained to be a multiple of Nsimd, which is as much as 8 for AVX512 with single precision.
2016-07-14 22:59:21 +01:00
template<class d>
struct scal {
d internal;
};
New automatic implementation of gamma matrices, Meson and SeqGamma are broken
2017-01-24 03:13:43 +00:00
Gamma::Algebra Gmu [] = {
Gamma::Algebra::GammaX,
Gamma::Algebra::GammaY,
Gamma::Algebra::GammaZ,
Gamma::Algebra::GammaT
};
Adding files for multiple implementations (cache opt) and Ls vectorisation of the 5D cayley form chiral fermions for the 5d matrix. With Ls entirely in the vector direction, s-hopping terms involve rotations. The serial dependence of the LDU inversion for Mobius and 4d even odd checkerboarding is removed by simply applying Ls^2 operations (vectorised many ways) as a dense matrix operation. This should give similar throughput but high flops (non-compulsory flops) but enable use of the KNL cache friendly kernels throughout the code. Ls is still constrained to be a multiple of Nsimd, which is as much as 8 for AVX512 with single precision.
2016-07-14 22:59:21 +01:00
template<class What>
void TestWhat(What & Ddwf,
GridCartesian * FGrid, GridRedBlackCartesian * FrbGrid,
GridCartesian * UGrid,
RealD mass, RealD M5,
GridParallelRNG *RNG4, GridParallelRNG *RNG5);
int main (int argc, char ** argv)
{
Grid_init(&argc,&argv);
int threads = GridThread::GetThreads();
std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
const int Ls=16;
Coordinate handling GPU friendly. Avoid std::vector
2018-02-24 22:19:28 +00:00
auto latt4 =GridDefaultLatt();
Adding files for multiple implementations (cache opt) and Ls vectorisation of the 5D cayley form chiral fermions for the 5d matrix. With Ls entirely in the vector direction, s-hopping terms involve rotations. The serial dependence of the LDU inversion for Mobius and 4d even odd checkerboarding is removed by simply applying Ls^2 operations (vectorised many ways) as a dense matrix operation. This should give similar throughput but high flops (non-compulsory flops) but enable use of the KNL cache friendly kernels throughout the code. Ls is still constrained to be a multiple of Nsimd, which is as much as 8 for AVX512 with single precision.
2016-07-14 22:59:21 +01:00
GridCartesian * UGrid = SpaceTimeGrid::makeFourDimGrid(latt4, GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());
GridRedBlackCartesian * UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
GridCartesian * FGrid = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid);
GridCartesian * sUGrid = SpaceTimeGrid::makeFourDimDWFGrid(latt4,GridDefaultMpi());
GridRedBlackCartesian * sUrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(sUGrid);
GridCartesian * sFGrid = SpaceTimeGrid::makeFiveDimDWFGrid(Ls,UGrid);
GridRedBlackCartesian * sFrbGrid = SpaceTimeGrid::makeFiveDimDWFRedBlackGrid(Ls,UGrid);
std::vector<int> seeds4({1,2,3,4});
std::vector<int> seeds5({5,6,7,8});
GridParallelRNG RNG4(UGrid); RNG4.SeedFixedIntegers(seeds4);
GridParallelRNG RNG5(FGrid); RNG5.SeedFixedIntegers(seeds5);
GridParallelRNG sRNG4(sUGrid); sRNG4.SeedFixedIntegers(seeds4);
GridParallelRNG sRNG5(sFGrid); sRNG5.SeedFixedIntegers(seeds5);
Hot start SU3
2016-12-22 17:51:45 +00:00
LatticeGaugeField Umu(UGrid);
SU3::HotConfiguration(RNG4,Umu);
Adding files for multiple implementations (cache opt) and Ls vectorisation of the 5D cayley form chiral fermions for the 5d matrix. With Ls entirely in the vector direction, s-hopping terms involve rotations. The serial dependence of the LDU inversion for Mobius and 4d even odd checkerboarding is removed by simply applying Ls^2 operations (vectorised many ways) as a dense matrix operation. This should give similar throughput but high flops (non-compulsory flops) but enable use of the KNL cache friendly kernels throughout the code. Ls is still constrained to be a multiple of Nsimd, which is as much as 8 for AVX512 with single precision.
2016-07-14 22:59:21 +01:00
RealD mass=0.1;
RealD M5 =1.8;
std::cout<<GridLogMessage<<"**************************************************************"<<std::endl;
std::cout<<GridLogMessage <<"DomainWallFermion vectorised test"<<std::endl;
std::cout<<GridLogMessage<<"**************************************************************"<<std::endl;
DomainWallFermionR Ddwf(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
TestWhat<DomainWallFermionR>(Ddwf,FGrid,FrbGrid,UGrid,mass,M5,&RNG4,&RNG5);
RealD b=1.5;// Scale factor b+c=2, b-c=1
RealD c=0.5;
std::cout<<GridLogMessage<<"**************************************************************"<<std::endl;
std::cout<<GridLogMessage <<"MobiusFermion test"<<std::endl;
std::cout<<GridLogMessage<<"**************************************************************"<<std::endl;
MobiusFermionR Dmob(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,b,c);
TestWhat<MobiusFermionR>(Dmob,FGrid,FrbGrid,UGrid,mass,M5,&RNG4,&RNG5);
Feature for z-Mobius prep
2016-08-15 22:31:29 +01:00
std::cout<<GridLogMessage<<"**************************************************************"<<std::endl;
std::cout<<GridLogMessage <<"Z-MobiusFermion test"<<std::endl;
std::cout<<GridLogMessage<<"**************************************************************"<<std::endl;
std::vector<ComplexD> gamma(Ls,std::complex<double>(1.0,0.0));
ZMobiusFermionR zDmob(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,gamma,b,c);
TestWhat<ZMobiusFermionR>(zDmob,FGrid,FrbGrid,UGrid,mass,M5,&RNG4,&RNG5);
Adding files for multiple implementations (cache opt) and Ls vectorisation of the 5D cayley form chiral fermions for the 5d matrix. With Ls entirely in the vector direction, s-hopping terms involve rotations. The serial dependence of the LDU inversion for Mobius and 4d even odd checkerboarding is removed by simply applying Ls^2 operations (vectorised many ways) as a dense matrix operation. This should give similar throughput but high flops (non-compulsory flops) but enable use of the KNL cache friendly kernels throughout the code. Ls is still constrained to be a multiple of Nsimd, which is as much as 8 for AVX512 with single precision.
2016-07-14 22:59:21 +01:00
std::cout<<GridLogMessage<<"**************************************************************"<<std::endl;
std::cout<<GridLogMessage <<"MobiusZolotarevFermion test"<<std::endl;
std::cout<<GridLogMessage<<"**************************************************************"<<std::endl;
MobiusZolotarevFermionR Dzolo(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,b,c,0.1,2.0);
TestWhat<MobiusZolotarevFermionR>(Dzolo,FGrid,FrbGrid,UGrid,mass,M5,&RNG4,&RNG5);
std::cout<<GridLogMessage<<"**************************************************************"<<std::endl;
std::cout<<GridLogMessage <<"ScaledShamirFermion test"<<std::endl;
std::cout<<GridLogMessage<<"**************************************************************"<<std::endl;
ScaledShamirFermionR Dsham(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,2.0);
TestWhat<ScaledShamirFermionR>(Dsham,FGrid,FrbGrid,UGrid,mass,M5,&RNG4,&RNG5);
std::cout<<GridLogMessage<<"**************************************************************"<<std::endl;
std::cout<<GridLogMessage <<"ShamirZolotarevFermion test"<<std::endl;
std::cout<<GridLogMessage<<"**************************************************************"<<std::endl;
ShamirZolotarevFermionR Dshamz(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,0.1,2.0);
TestWhat<ShamirZolotarevFermionR>(Dshamz,FGrid,FrbGrid,UGrid,mass,M5,&RNG4,&RNG5);
std::cout<<GridLogMessage<<"**************************************************************"<<std::endl;
std::cout<<GridLogMessage <<"OverlapWilsonCayleyTanhFermion test"<<std::endl;
std::cout<<GridLogMessage<<"**************************************************************"<<std::endl;
OverlapWilsonCayleyTanhFermionR Dov(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,1.0);
TestWhat<OverlapWilsonCayleyTanhFermionR>(Dov,FGrid,FrbGrid,UGrid,mass,M5,&RNG4,&RNG5);
std::cout<<GridLogMessage<<"**************************************************************"<<std::endl;
std::cout<<GridLogMessage <<"OverlapWilsonCayleyZolotarevFermion test"<<std::endl;
std::cout<<GridLogMessage<<"**************************************************************"<<std::endl;
OverlapWilsonCayleyZolotarevFermionR Dovz(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,0.1,2.0);
TestWhat<OverlapWilsonCayleyZolotarevFermionR>(Dovz,FGrid,FrbGrid,UGrid,mass,M5,&RNG4,&RNG5);
std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
Grid_finalize();
}
template<class What>
void TestWhat(What & Ddwf,
GridCartesian * FGrid, GridRedBlackCartesian * FrbGrid,
GridCartesian * UGrid,
RealD mass, RealD M5,
GridParallelRNG *RNG4,
GridParallelRNG *RNG5)
{
LatticeFermion src (FGrid); random(*RNG5,src);
LatticeFermion phi (FGrid); random(*RNG5,phi);
LatticeFermion chi (FGrid); random(*RNG5,chi);
Zero changes. (I mean literally)
2018-01-27 23:46:02 +00:00
LatticeFermion result(FGrid); result=Zero();
LatticeFermion ref(FGrid); ref=Zero();
LatticeFermion tmp(FGrid); tmp=Zero();
LatticeFermion err(FGrid); tmp=Zero();
Adding files for multiple implementations (cache opt) and Ls vectorisation of the 5D cayley form chiral fermions for the 5d matrix. With Ls entirely in the vector direction, s-hopping terms involve rotations. The serial dependence of the LDU inversion for Mobius and 4d even odd checkerboarding is removed by simply applying Ls^2 operations (vectorised many ways) as a dense matrix operation. This should give similar throughput but high flops (non-compulsory flops) but enable use of the KNL cache friendly kernels throughout the code. Ls is still constrained to be a multiple of Nsimd, which is as much as 8 for AVX512 with single precision.
2016-07-14 22:59:21 +01:00
LatticeFermion src_e (FrbGrid);
LatticeFermion src_o (FrbGrid);
LatticeFermion r_e (FrbGrid);
LatticeFermion r_o (FrbGrid);
LatticeFermion r_eo (FGrid);
LatticeFermion r_eeoo(FGrid);
std::cout<<GridLogMessage<<"=========================================================="<<std::endl;
std::cout<<GridLogMessage<<"= Testing that Meo + Moe + Moo + Mee = Munprec "<<std::endl;
std::cout<<GridLogMessage<<"=========================================================="<<std::endl;
pickCheckerboard(Even,src_e,src);
pickCheckerboard(Odd,src_o,src);
Ddwf.Meooe(src_e,r_o); std::cout<<GridLogMessage<<"Applied Meo"<<std::endl;
Ddwf.Meooe(src_o,r_e); std::cout<<GridLogMessage<<"Applied Moe"<<std::endl;
setCheckerboard(r_eo,r_o);
setCheckerboard(r_eo,r_e);
Ddwf.Mooee(src_e,r_e); std::cout<<GridLogMessage<<"Applied Mee"<<std::endl;
Ddwf.Mooee(src_o,r_o); std::cout<<GridLogMessage<<"Applied Moo"<<std::endl;
setCheckerboard(r_eeoo,r_e);
setCheckerboard(r_eeoo,r_o);
r_eo=r_eo+r_eeoo;
Ddwf.M(src,ref);
// std::cout<<GridLogMessage << r_eo<<std::endl;
// std::cout<<GridLogMessage << ref <<std::endl;
err= ref - r_eo;
std::cout<<GridLogMessage << "EO norm diff "<< norm2(err)<< " "<<norm2(ref)<< " " << norm2(r_eo) <<std::endl;
LatticeComplex cerr(FGrid);
cerr = localInnerProduct(err,err);
// std::cout<<GridLogMessage << cerr<<std::endl;
std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
std::cout<<GridLogMessage<<"= Test Ddagger is the dagger of D by requiring "<<std::endl;
std::cout<<GridLogMessage<<"= < phi | Deo | chi > * = < chi | Deo^dag| phi> "<<std::endl;
std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
LatticeFermion chi_e (FrbGrid);
LatticeFermion chi_o (FrbGrid);
LatticeFermion dchi_e (FrbGrid);
LatticeFermion dchi_o (FrbGrid);
LatticeFermion phi_e (FrbGrid);
LatticeFermion phi_o (FrbGrid);
LatticeFermion dphi_e (FrbGrid);
LatticeFermion dphi_o (FrbGrid);
pickCheckerboard(Even,chi_e,chi);
pickCheckerboard(Odd ,chi_o,chi);
pickCheckerboard(Even,phi_e,phi);
pickCheckerboard(Odd ,phi_o,phi);
Ddwf.Meooe(chi_e,dchi_o);
Ddwf.Meooe(chi_o,dchi_e);
Ddwf.MeooeDag(phi_e,dphi_o);
Ddwf.MeooeDag(phi_o,dphi_e);
ComplexD pDce = innerProduct(phi_e,dchi_e);
ComplexD pDco = innerProduct(phi_o,dchi_o);
ComplexD cDpe = innerProduct(chi_e,dphi_e);
ComplexD cDpo = innerProduct(chi_o,dphi_o);
std::cout<<GridLogMessage <<"e "<<pDce<<" "<<cDpe <<std::endl;
std::cout<<GridLogMessage <<"o "<<pDco<<" "<<cDpo <<std::endl;
std::cout<<GridLogMessage <<"pDce - conj(cDpo) "<< pDce-conj(cDpo) <<std::endl;
std::cout<<GridLogMessage <<"pDco - conj(cDpe) "<< pDco-conj(cDpe) <<std::endl;
std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
std::cout<<GridLogMessage<<"= Test MeeInv Mee = 1 "<<std::endl;
std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
pickCheckerboard(Even,chi_e,chi);
pickCheckerboard(Odd ,chi_o,chi);
Ddwf.Mooee(chi_e,src_e);
Ddwf.MooeeInv(src_e,phi_e);
Ddwf.Mooee(chi_o,src_o);
Ddwf.MooeeInv(src_o,phi_o);
setCheckerboard(phi,phi_e);
setCheckerboard(phi,phi_o);
err = phi-chi;
std::cout<<GridLogMessage << "norm diff "<< norm2(err)<< std::endl;
std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
std::cout<<GridLogMessage<<"= Test MeeInvDag MeeDag = 1 "<<std::endl;
std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
pickCheckerboard(Even,chi_e,chi);
pickCheckerboard(Odd ,chi_o,chi);
Ddwf.MooeeDag(chi_e,src_e);
Ddwf.MooeeInvDag(src_e,phi_e);
Ddwf.MooeeDag(chi_o,src_o);
Ddwf.MooeeInvDag(src_o,phi_o);
setCheckerboard(phi,phi_e);
setCheckerboard(phi,phi_o);
err = phi-chi;
std::cout<<GridLogMessage << "norm diff "<< norm2(err)<< std::endl;
}
Copy Permalink