portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-11-04 05:54:32 +00:00
Code Releases Activity

Merge pull request #130 from giltirn/gparity-handunroll

Browse Source 
Gparity handunroll
This commit is contained in:
Peter Boyle
2017-09-21 10:11:00 +01:00
committed by GitHub
parent 18c335198a 59bd1fe21b
commit bfb68e6f02
6 changed files with 722 additions and 204 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
benchmarks/Benchmark_dwf.cc
View File

@@ -51,7 +51,13 @@ int main (int argc, char ** argv)
std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
std::vector<int> latt4 = GridDefaultLatt();
const int Ls=16;
int Ls=16;
for(int i=0;i<argc;i++)
if(std::string(argv[i]) == "-Ls"){
std::stringstream ss(argv[i+1]); ss >> Ls;
}
GridCartesian * UGrid = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());
GridRedBlackCartesian * UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
GridCartesian * FGrid = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);

190
benchmarks/Benchmark_gparity.cc Normal file
View File

@@ -0,0 +1,190 @@
#include <Grid/Grid.h>
#include <sstream>
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
};
typedef typename GparityDomainWallFermionF::FermionField GparityLatticeFermionF;
typedef typename GparityDomainWallFermionD::FermionField GparityLatticeFermionD;
int main (int argc, char ** argv)
{
Grid_init(&argc,&argv);
int Ls=16;
for(int i=0;i<argc;i++)
if(std::string(argv[i]) == "-Ls"){
std::stringstream ss(argv[i+1]); ss >> Ls;
}
int threads = GridThread::GetThreads();
std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
std::cout<<GridLogMessage << "Ls = " << Ls << std::endl;
std::vector<int> latt4 = GridDefaultLatt();
GridCartesian * UGrid = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplexF::Nsimd()),GridDefaultMpi());
GridRedBlackCartesian * UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
GridCartesian * FGrid = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid);
std::vector<int> seeds4({1,2,3,4});
std::vector<int> seeds5({5,6,7,8});
std::cout << GridLogMessage << "Initialising 4d RNG" << std::endl;
GridParallelRNG RNG4(UGrid); RNG4.SeedFixedIntegers(seeds4);
std::cout << GridLogMessage << "Initialising 5d RNG" << std::endl;
GridParallelRNG RNG5(FGrid); RNG5.SeedFixedIntegers(seeds5);
std::cout << GridLogMessage << "Initialised RNGs" << std::endl;
GparityLatticeFermionF src (FGrid); random(RNG5,src);
RealD N2 = 1.0/::sqrt(norm2(src));
src = src*N2;
GparityLatticeFermionF result(FGrid); result=zero;
GparityLatticeFermionF ref(FGrid); ref=zero;
GparityLatticeFermionF tmp(FGrid);
GparityLatticeFermionF err(FGrid);
std::cout << GridLogMessage << "Drawing gauge field" << std::endl;
LatticeGaugeFieldF Umu(UGrid);
SU3::HotConfiguration(RNG4,Umu);
std::cout << GridLogMessage << "Random gauge initialised " << std::endl;
RealD mass=0.1;
RealD M5 =1.8;
RealD NP = UGrid->_Nprocessors;
RealD NN = UGrid->NodeCount();
std::cout << GridLogMessage<< "*****************************************************************" <<std::endl;
std::cout << GridLogMessage<< "* Kernel options --dslash-generic, --dslash-unroll, --dslash-asm" <<std::endl;
std::cout << GridLogMessage<< "*****************************************************************" <<std::endl;
std::cout << GridLogMessage<< "*****************************************************************" <<std::endl;
std::cout << GridLogMessage<< "* Benchmarking DomainWallFermion::Dhop "<<std::endl;
std::cout << GridLogMessage<< "* Vectorising space-time by "<<vComplexF::Nsimd()<<std::endl;
#ifdef GRID_OMP
if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsAndCompute ) std::cout << GridLogMessage<< "* Using Overlapped Comms/Compute" <<std::endl;
if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsThenCompute) std::cout << GridLogMessage<< "* Using sequential comms compute" <<std::endl;
#endif
if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptGeneric ) std::cout << GridLogMessage<< "* Using GENERIC Nc WilsonKernels" <<std::endl;
if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptHandUnroll) std::cout << GridLogMessage<< "* Using 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<< "* SINGLE/SINGLE"<<std::endl;
GparityDomainWallFermionF Dw(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
int ncall =1000;
if (1) {
FGrid->Barrier();
Dw.ZeroCounters();
Dw.Dhop(src,result,0);
std::cout<<GridLogMessage<<"Called warmup"<<std::endl;
double t0=usecond();
for(int i=0;i<ncall;i++){
__SSC_START;
Dw.Dhop(src,result,0);
__SSC_STOP;
}
double t1=usecond();
FGrid->Barrier();
double volume=Ls; for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
double flops=2*1344*volume*ncall;
std::cout<<GridLogMessage << "Called Dw "<<ncall<<" times in "<<t1-t0<<" us"<<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;
std::cout<<GridLogMessage << "mflop/s per rank = "<< flops/(t1-t0)/NP<<std::endl;
std::cout<<GridLogMessage << "mflop/s per node = "<< flops/(t1-t0)/NN<<std::endl;
Dw.Report();
}
std::cout << GridLogMessage<< "* SINGLE/HALF"<<std::endl;
GparityDomainWallFermionFH DwH(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
if (1) {
FGrid->Barrier();
DwH.ZeroCounters();
DwH.Dhop(src,result,0);
double t0=usecond();
for(int i=0;i<ncall;i++){
__SSC_START;
DwH.Dhop(src,result,0);
__SSC_STOP;
}
double t1=usecond();
FGrid->Barrier();
double volume=Ls; for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
double flops=2*1344*volume*ncall;
std::cout<<GridLogMessage << "Called half prec comms Dw "<<ncall<<" times in "<<t1-t0<<" us"<<std::endl;
std::cout<<GridLogMessage << "mflop/s = "<< flops/(t1-t0)<<std::endl;
std::cout<<GridLogMessage << "mflop/s per rank = "<< flops/(t1-t0)/NP<<std::endl;
std::cout<<GridLogMessage << "mflop/s per node = "<< flops/(t1-t0)/NN<<std::endl;
DwH.Report();
}
GridCartesian * UGrid_d = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplexD::Nsimd()),GridDefaultMpi());
GridRedBlackCartesian * UrbGrid_d = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid_d);
GridCartesian * FGrid_d = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid_d);
GridRedBlackCartesian * FrbGrid_d = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid_d);
std::cout << GridLogMessage<< "* DOUBLE/DOUBLE"<<std::endl;
GparityLatticeFermionD src_d(FGrid_d);
precisionChange(src_d,src);
LatticeGaugeFieldD Umu_d(UGrid_d);
precisionChange(Umu_d,Umu);
GparityLatticeFermionD result_d(FGrid_d);
GparityDomainWallFermionD DwD(Umu_d,*FGrid_d,*FrbGrid_d,*UGrid_d,*UrbGrid_d,mass,M5);
if (1) {
FGrid_d->Barrier();
DwD.ZeroCounters();
DwD.Dhop(src_d,result_d,0);
std::cout<<GridLogMessage<<"Called warmup"<<std::endl;
double t0=usecond();
for(int i=0;i<ncall;i++){
__SSC_START;
DwD.Dhop(src_d,result_d,0);
__SSC_STOP;
}
double t1=usecond();
FGrid_d->Barrier();
double volume=Ls; for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
double flops=2*1344*volume*ncall;
std::cout<<GridLogMessage << "Called Dw "<<ncall<<" times in "<<t1-t0<<" us"<<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;
std::cout<<GridLogMessage << "mflop/s per rank = "<< flops/(t1-t0)/NP<<std::endl;
std::cout<<GridLogMessage << "mflop/s per node = "<< flops/(t1-t0)/NN<<std::endl;
DwD.Report();
}
Grid_finalize();
}

6
lib/qcd/action/fermion/FermionOperatorImpl.h
View File

@@ -538,6 +538,12 @@ class GparityWilsonImpl : public ConjugateGaugeImpl<GaugeImplTypes<S, Nrepresent
}
template <class ref>
inline void loadLinkElement(Simd &reg, ref &memory) {
reg = memory;
}
inline void DoubleStore(GridBase *GaugeGrid,DoubledGaugeField &Uds,const GaugeField &Umu)
{
conformable(Uds._grid,GaugeGrid);

541
lib/qcd/action/fermion/WilsonKernelsHand.cc
View File

@@ -30,60 +30,181 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#define REGISTER
#define LOAD_CHIMU \
{const SiteSpinor & ref (in._odata[offset]); \
Chimu_00=ref()(0)(0);\
Chimu_01=ref()(0)(1);\
Chimu_02=ref()(0)(2);\
Chimu_10=ref()(1)(0);\
Chimu_11=ref()(1)(1);\
Chimu_12=ref()(1)(2);\
Chimu_20=ref()(2)(0);\
Chimu_21=ref()(2)(1);\
Chimu_22=ref()(2)(2);\
Chimu_30=ref()(3)(0);\
Chimu_31=ref()(3)(1);\
Chimu_32=ref()(3)(2);}
#define LOAD_CHIMU_BODY(F) \
Chimu_00=ref(F)(0)(0); \
Chimu_01=ref(F)(0)(1); \
Chimu_02=ref(F)(0)(2); \
Chimu_10=ref(F)(1)(0); \
Chimu_11=ref(F)(1)(1); \
Chimu_12=ref(F)(1)(2); \
Chimu_20=ref(F)(2)(0); \
Chimu_21=ref(F)(2)(1); \
Chimu_22=ref(F)(2)(2); \
Chimu_30=ref(F)(3)(0); \
Chimu_31=ref(F)(3)(1); \
Chimu_32=ref(F)(3)(2)
#define LOAD_CHI\
{const SiteHalfSpinor &ref(buf[offset]); \
Chi_00 = ref()(0)(0);\
Chi_01 = ref()(0)(1);\
Chi_02 = ref()(0)(2);\
Chi_10 = ref()(1)(0);\
Chi_11 = ref()(1)(1);\
Chi_12 = ref()(1)(2);}
#define LOAD_CHIMU(DIR,F,PERM) \
{ const SiteSpinor & ref (in._odata[offset]); LOAD_CHIMU_BODY(F); }
#define LOAD_CHI_BODY(F) \
Chi_00 = ref(F)(0)(0);\
Chi_01 = ref(F)(0)(1);\
Chi_02 = ref(F)(0)(2);\
Chi_10 = ref(F)(1)(0);\
Chi_11 = ref(F)(1)(1);\
Chi_12 = ref(F)(1)(2)
#define LOAD_CHI(DIR,F,PERM) \
{const SiteHalfSpinor &ref(buf[offset]); LOAD_CHI_BODY(F); }
//G-parity implementations using in-place intrinsic ops
//1l 1h -> 1h 1l
//0l 0h , 1h 1l -> 0l 1h 0h,1l
//0h,1l -> 1l,0h
//if( (distance == 1 && !perm_will_occur) || (distance == -1 && perm_will_occur) )
//Pulled fermion through forwards face, GPBC on upper component
//Need 0= 0l 1h 1= 1l 0h
//else if( (distance == -1 && !perm) || (distance == 1 && perm) )
//Pulled fermion through backwards face, GPBC on lower component
//Need 0= 1l 0h 1= 0l 1h
//1l 1h -> 1h 1l
//0l 0h , 1h 1l -> 0l 1h 0h,1l
#define DO_TWIST_0L_1H(INTO,S,C,F, PERM, tmp1, tmp2, tmp3) \
permute##PERM(tmp1, ref(1)(S)(C)); \
exchange##PERM(tmp2,tmp3, ref(0)(S)(C), tmp1); \
INTO = tmp2;
//0l 0h -> 0h 0l
//1l 1h, 0h 0l -> 1l 0h, 1h 0l
#define DO_TWIST_1L_0H(INTO,S,C,F, PERM, tmp1, tmp2, tmp3) \
permute##PERM(tmp1, ref(0)(S)(C)); \
exchange##PERM(tmp2,tmp3, ref(1)(S)(C), tmp1); \
INTO = tmp2;
#define LOAD_CHI_SETUP(DIR,F) \
g = F; \
direction = st._directions[DIR]; \
distance = st._distances[DIR]; \
sl = st._grid->_simd_layout[direction]; \
inplace_twist = 0; \
if(SE->_around_the_world && this->Params.twists[DIR % 4]){ \
if(sl == 1){ \
g = (F+1) % 2; \
}else{ \
inplace_twist = 1; \
} \
}
#define LOAD_CHIMU_GPARITY_INPLACE_TWIST(DIR,F,PERM) \
{ const SiteSpinor &ref(in._odata[offset]); \
LOAD_CHI_SETUP(DIR,F); \
if(!inplace_twist){ \
LOAD_CHIMU_BODY(g); \
}else{ \
if( ( F==0 && ((distance == 1 && !perm) || (distance == -1 && perm)) ) || \
( F==1 && ((distance == -1 && !perm) || (distance == 1 && perm)) ) ){ \
DO_TWIST_0L_1H(Chimu_00,0,0,F,PERM, U_00,U_01,U_10); \
DO_TWIST_0L_1H(Chimu_01,0,1,F,PERM, U_11,U_20,U_21); \
DO_TWIST_0L_1H(Chimu_02,0,2,F,PERM, U_00,U_01,U_10); \
DO_TWIST_0L_1H(Chimu_10,1,0,F,PERM, U_11,U_20,U_21); \
DO_TWIST_0L_1H(Chimu_11,1,1,F,PERM, U_00,U_01,U_10); \
DO_TWIST_0L_1H(Chimu_12,1,2,F,PERM, U_11,U_20,U_21); \
DO_TWIST_0L_1H(Chimu_20,2,0,F,PERM, U_00,U_01,U_10); \
DO_TWIST_0L_1H(Chimu_21,2,1,F,PERM, U_11,U_20,U_21); \
DO_TWIST_0L_1H(Chimu_22,2,2,F,PERM, U_00,U_01,U_10); \
DO_TWIST_0L_1H(Chimu_30,3,0,F,PERM, U_11,U_20,U_21); \
DO_TWIST_0L_1H(Chimu_31,3,1,F,PERM, U_00,U_01,U_10); \
DO_TWIST_0L_1H(Chimu_32,3,2,F,PERM, U_11,U_20,U_21); \
}else{ \
DO_TWIST_1L_0H(Chimu_00,0,0,F,PERM, U_00,U_01,U_10); \
DO_TWIST_1L_0H(Chimu_01,0,1,F,PERM, U_11,U_20,U_21); \
DO_TWIST_1L_0H(Chimu_02,0,2,F,PERM, U_00,U_01,U_10); \
DO_TWIST_1L_0H(Chimu_10,1,0,F,PERM, U_11,U_20,U_21); \
DO_TWIST_1L_0H(Chimu_11,1,1,F,PERM, U_00,U_01,U_10); \
DO_TWIST_1L_0H(Chimu_12,1,2,F,PERM, U_11,U_20,U_21); \
DO_TWIST_1L_0H(Chimu_20,2,0,F,PERM, U_00,U_01,U_10); \
DO_TWIST_1L_0H(Chimu_21,2,1,F,PERM, U_11,U_20,U_21); \
DO_TWIST_1L_0H(Chimu_22,2,2,F,PERM, U_00,U_01,U_10); \
DO_TWIST_1L_0H(Chimu_30,3,0,F,PERM, U_11,U_20,U_21); \
DO_TWIST_1L_0H(Chimu_31,3,1,F,PERM, U_00,U_01,U_10); \
DO_TWIST_1L_0H(Chimu_32,3,2,F,PERM, U_11,U_20,U_21); \
} \
} \
}
#define LOAD_CHI_GPARITY_INPLACE_TWIST(DIR,F,PERM) \
{ const SiteHalfSpinor &ref(buf[offset]); \
LOAD_CHI_SETUP(DIR,F); \
if(!inplace_twist){ \
LOAD_CHI_BODY(g); \
}else{ \
if( ( F==0 && ((distance == 1 && !perm) || (distance == -1 && perm)) ) || \
( F==1 && ((distance == -1 && !perm) || (distance == 1 && perm)) ) ){ \
DO_TWIST_0L_1H(Chi_00,0,0,F,PERM, U_00,U_01,U_10); \
DO_TWIST_0L_1H(Chi_01,0,1,F,PERM, U_11,U_20,U_21); \
DO_TWIST_0L_1H(Chi_02,0,2,F,PERM, UChi_00,UChi_01,UChi_02); \
DO_TWIST_0L_1H(Chi_10,1,0,F,PERM, UChi_10,UChi_11,UChi_12); \
DO_TWIST_0L_1H(Chi_11,1,1,F,PERM, U_00,U_01,U_10); \
DO_TWIST_0L_1H(Chi_12,1,2,F,PERM, U_11,U_20,U_21); \
}else{ \
DO_TWIST_1L_0H(Chi_00,0,0,F,PERM, U_00,U_01,U_10); \
DO_TWIST_1L_0H(Chi_01,0,1,F,PERM, U_11,U_20,U_21); \
DO_TWIST_1L_0H(Chi_02,0,2,F,PERM, UChi_00,UChi_01,UChi_02); \
DO_TWIST_1L_0H(Chi_10,1,0,F,PERM, UChi_10,UChi_11,UChi_12); \
DO_TWIST_1L_0H(Chi_11,1,1,F,PERM, U_00,U_01,U_10); \
DO_TWIST_1L_0H(Chi_12,1,2,F,PERM, U_11,U_20,U_21); \
} \
} \
}
#define LOAD_CHI_GPARITY(DIR,F,PERM) LOAD_CHI_GPARITY_INPLACE_TWIST(DIR,F,PERM)
#define LOAD_CHIMU_GPARITY(DIR,F,PERM) LOAD_CHIMU_GPARITY_INPLACE_TWIST(DIR,F,PERM)
// To splat or not to splat depends on the implementation
#define MULT_2SPIN(A)\
{auto & ref(U._odata[sU](A)); \
#define MULT_2SPIN_BODY \
Impl::loadLinkElement(U_00,ref()(0,0)); \
Impl::loadLinkElement(U_10,ref()(1,0)); \
Impl::loadLinkElement(U_20,ref()(2,0)); \
Impl::loadLinkElement(U_01,ref()(0,1)); \
Impl::loadLinkElement(U_11,ref()(1,1)); \
Impl::loadLinkElement(U_21,ref()(2,1)); \
UChi_00 = U_00*Chi_00;\
UChi_10 = U_00*Chi_10;\
UChi_01 = U_10*Chi_00;\
UChi_11 = U_10*Chi_10;\
UChi_02 = U_20*Chi_00;\
UChi_12 = U_20*Chi_10;\
UChi_00+= U_01*Chi_01;\
UChi_10+= U_01*Chi_11;\
UChi_01+= U_11*Chi_01;\
UChi_11+= U_11*Chi_11;\
UChi_02+= U_21*Chi_01;\
UChi_12+= U_21*Chi_11;\
UChi_00 = U_00*Chi_00; \
UChi_10 = U_00*Chi_10; \
UChi_01 = U_10*Chi_00; \
UChi_11 = U_10*Chi_10; \
UChi_02 = U_20*Chi_00; \
UChi_12 = U_20*Chi_10; \
UChi_00+= U_01*Chi_01; \
UChi_10+= U_01*Chi_11; \
UChi_01+= U_11*Chi_01; \
UChi_11+= U_11*Chi_11; \
UChi_02+= U_21*Chi_01; \
UChi_12+= U_21*Chi_11; \
Impl::loadLinkElement(U_00,ref()(0,2)); \
Impl::loadLinkElement(U_10,ref()(1,2)); \
Impl::loadLinkElement(U_20,ref()(2,2)); \
UChi_00+= U_00*Chi_02;\
UChi_10+= U_00*Chi_12;\
UChi_01+= U_10*Chi_02;\
UChi_11+= U_10*Chi_12;\
UChi_02+= U_20*Chi_02;\
UChi_12+= U_20*Chi_12;}
UChi_00+= U_00*Chi_02; \
UChi_10+= U_00*Chi_12; \
UChi_01+= U_10*Chi_02; \
UChi_11+= U_10*Chi_12; \
UChi_02+= U_20*Chi_02; \
UChi_12+= U_20*Chi_12
#define MULT_2SPIN(A,F) \
{auto & ref(U._odata[sU](A)); MULT_2SPIN_BODY; }
#define MULT_2SPIN_GPARITY(A,F) \
{auto & ref(U._odata[sU](F)(A)); MULT_2SPIN_BODY; }
#define PERMUTE_DIR(dir) \
@@ -307,84 +428,87 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
result_31-= UChi_11; \
result_32-= UChi_12;
#define HAND_STENCIL_LEG(PROJ,PERM,DIR,RECON) \
#define HAND_STENCIL_LEG(PROJ,PERM,DIR,RECON,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \
SE=st.GetEntry(ptype,DIR,ss); \
offset = SE->_offset; \
local = SE->_is_local; \
perm = SE->_permute; \
if ( local ) { \
LOAD_CHIMU; \
LOAD_CHIMU_IMPL(DIR,F,PERM); \
PROJ; \
if ( perm) { \
PERMUTE_DIR(PERM); \
} \
} else { \
LOAD_CHI; \
LOAD_CHI_IMPL(DIR,F,PERM); \
} \
MULT_2SPIN(DIR); \
MULT_2SPIN_IMPL(DIR,F); \
RECON;
#define HAND_STENCIL_LEG_INT(PROJ,PERM,DIR,RECON) \
#define HAND_STENCIL_LEG_INT(PROJ,PERM,DIR,RECON,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \
SE=st.GetEntry(ptype,DIR,ss); \
offset = SE->_offset; \
local = SE->_is_local; \
perm = SE->_permute; \
if ( local ) { \
LOAD_CHIMU; \
LOAD_CHIMU_IMPL(DIR,F,PERM); \
PROJ; \
if ( perm) { \
PERMUTE_DIR(PERM); \
} \
} else if ( st.same_node[DIR] ) { \
LOAD_CHI; \
LOAD_CHI_IMPL(DIR,F,PERM); \
} \
if (local || st.same_node[DIR] ) { \
MULT_2SPIN(DIR); \
MULT_2SPIN_IMPL(DIR,F); \
RECON; \
}
#define HAND_STENCIL_LEG_EXT(PROJ,PERM,DIR,RECON) \
#define HAND_STENCIL_LEG_EXT(PROJ,PERM,DIR,RECON,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \
SE=st.GetEntry(ptype,DIR,ss); \
offset = SE->_offset; \
local = SE->_is_local; \
perm = SE->_permute; \
if((!SE->_is_local)&&(!st.same_node[DIR]) ) { \
LOAD_CHI; \
MULT_2SPIN(DIR); \
LOAD_CHI_IMPL(DIR,F,PERM); \
MULT_2SPIN_IMPL(DIR,F); \
RECON; \
nmu++; \
}
#define HAND_RESULT(ss) \
#define HAND_RESULT(ss,F) \
{ \
SiteSpinor & ref (out._odata[ss]); \
vstream(ref()(0)(0),result_00); \
vstream(ref()(0)(1),result_01); \
vstream(ref()(0)(2),result_02); \
vstream(ref()(1)(0),result_10); \
vstream(ref()(1)(1),result_11); \
vstream(ref()(1)(2),result_12); \
vstream(ref()(2)(0),result_20); \
vstream(ref()(2)(1),result_21); \
vstream(ref()(2)(2),result_22); \
vstream(ref()(3)(0),result_30); \
vstream(ref()(3)(1),result_31); \
vstream(ref()(3)(2),result_32); \
vstream(ref(F)(0)(0),result_00); \
vstream(ref(F)(0)(1),result_01); \
vstream(ref(F)(0)(2),result_02); \
vstream(ref(F)(1)(0),result_10); \
vstream(ref(F)(1)(1),result_11); \
vstream(ref(F)(1)(2),result_12); \
vstream(ref(F)(2)(0),result_20); \
vstream(ref(F)(2)(1),result_21); \
vstream(ref(F)(2)(2),result_22); \
vstream(ref(F)(3)(0),result_30); \
vstream(ref(F)(3)(1),result_31); \
vstream(ref(F)(3)(2),result_32); \
}
#define HAND_RESULT_EXT(ss) \
#define HAND_RESULT_EXT(ss,F) \
if (nmu){ \
SiteSpinor & ref (out._odata[ss]); \
ref()(0)(0)+=result_00; \
ref()(0)(1)+=result_01; \
ref()(0)(2)+=result_02; \
ref()(1)(0)+=result_10; \
ref()(1)(1)+=result_11; \
ref()(1)(2)+=result_12; \
ref()(2)(0)+=result_20; \
ref()(2)(1)+=result_21; \
ref()(2)(2)+=result_22; \
ref()(3)(0)+=result_30; \
ref()(3)(1)+=result_31; \
ref()(3)(2)+=result_32; \
ref(F)(0)(0)+=result_00; \
ref(F)(0)(1)+=result_01; \
ref(F)(0)(2)+=result_02; \
ref(F)(1)(0)+=result_10; \
ref(F)(1)(1)+=result_11; \
ref(F)(1)(2)+=result_12; \
ref(F)(2)(0)+=result_20; \
ref(F)(2)(1)+=result_21; \
ref(F)(2)(2)+=result_22; \
ref(F)(3)(0)+=result_30; \
ref(F)(3)(1)+=result_31; \
ref(F)(3)(2)+=result_32; \
}
@@ -463,15 +587,18 @@ WilsonKernels<Impl>::HandDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGauge
int offset,local,perm, ptype;
StencilEntry *SE;
HAND_STENCIL_LEG(XM_PROJ,3,Xp,XM_RECON);
HAND_STENCIL_LEG(YM_PROJ,2,Yp,YM_RECON_ACCUM);
HAND_STENCIL_LEG(ZM_PROJ,1,Zp,ZM_RECON_ACCUM);
HAND_STENCIL_LEG(TM_PROJ,0,Tp,TM_RECON_ACCUM);
HAND_STENCIL_LEG(XP_PROJ,3,Xm,XP_RECON_ACCUM);
HAND_STENCIL_LEG(YP_PROJ,2,Ym,YP_RECON_ACCUM);
HAND_STENCIL_LEG(ZP_PROJ,1,Zm,ZP_RECON_ACCUM);
HAND_STENCIL_LEG(TP_PROJ,0,Tm,TP_RECON_ACCUM);
HAND_RESULT(ss);
#define HAND_DOP_SITE(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \
HAND_STENCIL_LEG(XM_PROJ,3,Xp,XM_RECON,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG(YM_PROJ,2,Yp,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG(ZM_PROJ,1,Zp,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG(TM_PROJ,0,Tp,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG(XP_PROJ,3,Xm,XP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG(YP_PROJ,2,Ym,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG(ZP_PROJ,1,Zm,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG(TP_PROJ,0,Tm,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_RESULT(ss,F)
HAND_DOP_SITE(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN);
}
template<class Impl>
@@ -486,15 +613,18 @@ void WilsonKernels<Impl>::HandDhopSiteDag(StencilImpl &st,LebesgueOrder &lo,Doub
StencilEntry *SE;
int offset,local,perm, ptype;
HAND_STENCIL_LEG(XP_PROJ,3,Xp,XP_RECON);
HAND_STENCIL_LEG(YP_PROJ,2,Yp,YP_RECON_ACCUM);
HAND_STENCIL_LEG(ZP_PROJ,1,Zp,ZP_RECON_ACCUM);
HAND_STENCIL_LEG(TP_PROJ,0,Tp,TP_RECON_ACCUM);
HAND_STENCIL_LEG(XM_PROJ,3,Xm,XM_RECON_ACCUM);
HAND_STENCIL_LEG(YM_PROJ,2,Ym,YM_RECON_ACCUM);
HAND_STENCIL_LEG(ZM_PROJ,1,Zm,ZM_RECON_ACCUM);
HAND_STENCIL_LEG(TM_PROJ,0,Tm,TM_RECON_ACCUM);
HAND_RESULT(ss);
#define HAND_DOP_SITE_DAG(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \
HAND_STENCIL_LEG(XP_PROJ,3,Xp,XP_RECON,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG(YP_PROJ,2,Yp,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG(ZP_PROJ,1,Zp,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG(TP_PROJ,0,Tp,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG(XM_PROJ,3,Xm,XM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG(YM_PROJ,2,Ym,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG(ZM_PROJ,1,Zm,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG(TM_PROJ,0,Tm,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_RESULT(ss,F)
HAND_DOP_SITE_DAG(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN);
}
template<class Impl> void
@@ -509,16 +639,20 @@ WilsonKernels<Impl>::HandDhopSiteInt(StencilImpl &st,LebesgueOrder &lo,DoubledGa
int offset,local,perm, ptype;
StencilEntry *SE;
ZERO_RESULT;
HAND_STENCIL_LEG_INT(XM_PROJ,3,Xp,XM_RECON_ACCUM);
HAND_STENCIL_LEG_INT(YM_PROJ,2,Yp,YM_RECON_ACCUM);
HAND_STENCIL_LEG_INT(ZM_PROJ,1,Zp,ZM_RECON_ACCUM);
HAND_STENCIL_LEG_INT(TM_PROJ,0,Tp,TM_RECON_ACCUM);
HAND_STENCIL_LEG_INT(XP_PROJ,3,Xm,XP_RECON_ACCUM);
HAND_STENCIL_LEG_INT(YP_PROJ,2,Ym,YP_RECON_ACCUM);
HAND_STENCIL_LEG_INT(ZP_PROJ,1,Zm,ZP_RECON_ACCUM);
HAND_STENCIL_LEG_INT(TP_PROJ,0,Tm,TP_RECON_ACCUM);
HAND_RESULT(ss);
#define HAND_DOP_SITE_INT(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \
ZERO_RESULT; \
HAND_STENCIL_LEG_INT(XM_PROJ,3,Xp,XM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_INT(YM_PROJ,2,Yp,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_INT(ZM_PROJ,1,Zp,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_INT(TM_PROJ,0,Tp,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_INT(XP_PROJ,3,Xm,XP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_INT(YP_PROJ,2,Ym,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_INT(ZP_PROJ,1,Zm,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_INT(TP_PROJ,0,Tm,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_RESULT(ss,F)
HAND_DOP_SITE_INT(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN);
}
template<class Impl>
@@ -532,16 +666,20 @@ void WilsonKernels<Impl>::HandDhopSiteDagInt(StencilImpl &st,LebesgueOrder &lo,D
StencilEntry *SE;
int offset,local,perm, ptype;
ZERO_RESULT;
HAND_STENCIL_LEG_INT(XP_PROJ,3,Xp,XP_RECON_ACCUM);
HAND_STENCIL_LEG_INT(YP_PROJ,2,Yp,YP_RECON_ACCUM);
HAND_STENCIL_LEG_INT(ZP_PROJ,1,Zp,ZP_RECON_ACCUM);
HAND_STENCIL_LEG_INT(TP_PROJ,0,Tp,TP_RECON_ACCUM);
HAND_STENCIL_LEG_INT(XM_PROJ,3,Xm,XM_RECON_ACCUM);
HAND_STENCIL_LEG_INT(YM_PROJ,2,Ym,YM_RECON_ACCUM);
HAND_STENCIL_LEG_INT(ZM_PROJ,1,Zm,ZM_RECON_ACCUM);
HAND_STENCIL_LEG_INT(TM_PROJ,0,Tm,TM_RECON_ACCUM);
HAND_RESULT(ss);
#define HAND_DOP_SITE_DAG_INT(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \
ZERO_RESULT; \
HAND_STENCIL_LEG_INT(XP_PROJ,3,Xp,XP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_INT(YP_PROJ,2,Yp,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_INT(ZP_PROJ,1,Zp,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_INT(TP_PROJ,0,Tp,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_INT(XM_PROJ,3,Xm,XM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_INT(YM_PROJ,2,Ym,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_INT(ZM_PROJ,1,Zm,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_INT(TM_PROJ,0,Tm,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_RESULT(ss,F)
HAND_DOP_SITE_DAG_INT(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN);
}
template<class Impl> void
@@ -557,16 +695,20 @@ WilsonKernels<Impl>::HandDhopSiteExt(StencilImpl &st,LebesgueOrder &lo,DoubledGa
int offset,local,perm, ptype;
StencilEntry *SE;
int nmu=0;
ZERO_RESULT;
HAND_STENCIL_LEG_EXT(XM_PROJ,3,Xp,XM_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(YM_PROJ,2,Yp,YM_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(ZM_PROJ,1,Zp,ZM_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(TM_PROJ,0,Tp,TM_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(XP_PROJ,3,Xm,XP_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(YP_PROJ,2,Ym,YP_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(ZP_PROJ,1,Zm,ZP_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(TP_PROJ,0,Tm,TP_RECON_ACCUM);
HAND_RESULT_EXT(ss);
#define HAND_DOP_SITE_EXT(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \
ZERO_RESULT; \
HAND_STENCIL_LEG_EXT(XM_PROJ,3,Xp,XM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_EXT(YM_PROJ,2,Yp,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_EXT(ZM_PROJ,1,Zp,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_EXT(TM_PROJ,0,Tp,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_EXT(XP_PROJ,3,Xm,XP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_EXT(YP_PROJ,2,Ym,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_EXT(ZP_PROJ,1,Zm,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_EXT(TP_PROJ,0,Tm,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_RESULT_EXT(ss,F)
HAND_DOP_SITE_EXT(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN);
}
template<class Impl>
@@ -581,16 +723,20 @@ void WilsonKernels<Impl>::HandDhopSiteDagExt(StencilImpl &st,LebesgueOrder &lo,D
StencilEntry *SE;
int offset,local,perm, ptype;
int nmu=0;
ZERO_RESULT;
HAND_STENCIL_LEG_EXT(XP_PROJ,3,Xp,XP_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(YP_PROJ,2,Yp,YP_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(ZP_PROJ,1,Zp,ZP_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(TP_PROJ,0,Tp,TP_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(XM_PROJ,3,Xm,XM_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(YM_PROJ,2,Ym,YM_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(ZM_PROJ,1,Zm,ZM_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(TM_PROJ,0,Tm,TM_RECON_ACCUM);
HAND_RESULT_EXT(ss);
#define HAND_DOP_SITE_DAG_EXT(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \
ZERO_RESULT; \
HAND_STENCIL_LEG_EXT(XP_PROJ,3,Xp,XP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_EXT(YP_PROJ,2,Yp,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_EXT(ZP_PROJ,1,Zp,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_EXT(TP_PROJ,0,Tp,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_EXT(XM_PROJ,3,Xm,XM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_EXT(YM_PROJ,2,Ym,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_EXT(ZM_PROJ,1,Zm,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_EXT(TM_PROJ,0,Tm,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_RESULT_EXT(ss,F)
HAND_DOP_SITE_DAG_EXT(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN);
}
////////////////////////////////////////////////
@@ -646,10 +792,123 @@ void WilsonKernels<Impl>::HandDhopSiteDagExt(StencilImpl &st,LebesgueOrder &lo,D
const FermionField &in, \
FermionField &out){ assert(0); } \
HAND_SPECIALISE_EMPTY(GparityWilsonImplF);
HAND_SPECIALISE_EMPTY(GparityWilsonImplD);
HAND_SPECIALISE_EMPTY(GparityWilsonImplFH);
HAND_SPECIALISE_EMPTY(GparityWilsonImplDF);
#define HAND_SPECIALISE_GPARITY(IMPL) \
template<> void \
WilsonKernels<IMPL>::HandDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
int ss,int sU,const FermionField &in, FermionField &out) \
{ \
typedef IMPL Impl; \
typedef typename Simd::scalar_type S; \
typedef typename Simd::vector_type V; \
\
HAND_DECLARATIONS(ignore); \
\
int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist; \
StencilEntry *SE; \
HAND_DOP_SITE(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
HAND_DOP_SITE(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
} \
\
template<> \
void WilsonKernels<IMPL>::HandDhopSiteDag(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
int ss,int sU,const FermionField &in, FermionField &out) \
{ \
typedef IMPL Impl; \
typedef typename Simd::scalar_type S; \
typedef typename Simd::vector_type V; \
\
HAND_DECLARATIONS(ignore); \
\
StencilEntry *SE; \
int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist; \
HAND_DOP_SITE_DAG(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
HAND_DOP_SITE_DAG(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
} \
\
template<> void \
WilsonKernels<IMPL>::HandDhopSiteInt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
int ss,int sU,const FermionField &in, FermionField &out) \
{ \
typedef IMPL Impl; \
typedef typename Simd::scalar_type S; \
typedef typename Simd::vector_type V; \
\
HAND_DECLARATIONS(ignore); \
\
int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist; \
StencilEntry *SE; \
HAND_DOP_SITE_INT(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
HAND_DOP_SITE_INT(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
} \
\
template<> \
void WilsonKernels<IMPL>::HandDhopSiteDagInt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
int ss,int sU,const FermionField &in, FermionField &out) \
{ \
typedef IMPL Impl; \
typedef typename Simd::scalar_type S; \
typedef typename Simd::vector_type V; \
\
HAND_DECLARATIONS(ignore); \
\
StencilEntry *SE; \
int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist; \
HAND_DOP_SITE_DAG_INT(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
HAND_DOP_SITE_DAG_INT(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
} \
\
template<> void \
WilsonKernels<IMPL>::HandDhopSiteExt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
int ss,int sU,const FermionField &in, FermionField &out) \
{ \
typedef IMPL Impl; \
typedef typename Simd::scalar_type S; \
typedef typename Simd::vector_type V; \
\
HAND_DECLARATIONS(ignore); \
\
int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist; \
StencilEntry *SE; \
int nmu=0; \
HAND_DOP_SITE_EXT(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
nmu = 0; \
HAND_DOP_SITE_EXT(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
} \
template<> \
void WilsonKernels<IMPL>::HandDhopSiteDagExt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
int ss,int sU,const FermionField &in, FermionField &out) \
{ \
typedef IMPL Impl; \
typedef typename Simd::scalar_type S; \
typedef typename Simd::vector_type V; \
\
HAND_DECLARATIONS(ignore); \
\
StencilEntry *SE; \
int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist; \
int nmu=0; \
HAND_DOP_SITE_DAG_EXT(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
nmu = 0; \
HAND_DOP_SITE_DAG_EXT(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
}
HAND_SPECIALISE_GPARITY(GparityWilsonImplF);
HAND_SPECIALISE_GPARITY(GparityWilsonImplD);
HAND_SPECIALISE_GPARITY(GparityWilsonImplFH);
HAND_SPECIALISE_GPARITY(GparityWilsonImplDF);
////////////// Wilson ; uses this implementation /////////////////////

13
lib/simd/Grid_vector_types.h
View File

@@ -376,7 +376,18 @@ class Grid_simd {
Optimization::Exchange::Exchange0(out1.v,out2.v,in1.v,in2.v);
}
}
friend inline void exchange0(Grid_simd &out1,Grid_simd &out2,Grid_simd in1,Grid_simd in2){
Optimization::Exchange::Exchange0(out1.v,out2.v,in1.v,in2.v);
}
friend inline void exchange1(Grid_simd &out1,Grid_simd &out2,Grid_simd in1,Grid_simd in2){
Optimization::Exchange::Exchange1(out1.v,out2.v,in1.v,in2.v);
}
friend inline void exchange2(Grid_simd &out1,Grid_simd &out2,Grid_simd in1,Grid_simd in2){
Optimization::Exchange::Exchange2(out1.v,out2.v,in1.v,in2.v);
}
friend inline void exchange3(Grid_simd &out1,Grid_simd &out2,Grid_simd in1,Grid_simd in2){
Optimization::Exchange::Exchange3(out1.v,out2.v,in1.v,in2.v);
}
////////////////////////////////////////////////////////////////////
// General permute; assumes vector length is same across
// all subtypes; may not be a good assumption, but could

148
tests/core/Test_gparity.cc
View File

@@ -33,22 +33,68 @@ using namespace std;
using namespace Grid;
using namespace Grid::QCD;
typedef typename GparityDomainWallFermionR::FermionField FermionField;
//typedef GparityDomainWallFermionD GparityDiracOp;
//typedef DomainWallFermionD StandardDiracOp;
//#define DOP_PARAMS
typedef GparityMobiusFermionD GparityDiracOp;
typedef MobiusFermionD StandardDiracOp;
#define DOP_PARAMS ,1.5, 0.5
typedef typename GparityDiracOp::FermionField GparityFermionField;
typedef typename GparityDiracOp::GaugeField GparityGaugeField;
typedef typename GparityFermionField::vector_type vComplexType;
typedef typename StandardDiracOp::FermionField StandardFermionField;
typedef typename StandardDiracOp::GaugeField StandardGaugeField;
enum{ same_vComplex = std::is_same<vComplexType, typename StandardFermionField::vector_type>::value };
static_assert(same_vComplex == 1, "Dirac Operators must have same underlying SIMD complex type");
int main (int argc, char ** argv)
{
const int nu = 3;
int nu = 0;
Grid_init(&argc,&argv);
for(int i=1;i<argc;i++){
if(std::string(argv[i]) == "--Gparity-dir"){
std::stringstream ss; ss << argv[i+1]; ss >> nu;
std::cout << GridLogMessage << "Set Gparity direction to " << nu << std::endl;
}
}
std::cout << GridLogMessage<< "*****************************************************************" <<std::endl;
std::cout << GridLogMessage<< "* Kernel options --dslash-generic, --dslash-unroll, --dslash-asm" <<std::endl;
std::cout << GridLogMessage<< "*****************************************************************" <<std::endl;
std::cout << GridLogMessage<< "*****************************************************************" <<std::endl;
std::cout << GridLogMessage<< "* Testing Gparity Dirac operator "<<std::endl;
std::cout << GridLogMessage<< "* Vectorising space-time by "<<vComplexType::Nsimd()<<std::endl;
#ifdef GRID_OMP
if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsAndCompute ) std::cout << GridLogMessage<< "* Using Overlapped Comms/Compute" <<std::endl;
if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsThenCompute) std::cout << GridLogMessage<< "* Using sequential comms compute" <<std::endl;
#endif
if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptGeneric ) std::cout << GridLogMessage<< "* Using GENERIC Nc WilsonKernels" <<std::endl;
if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptHandUnroll) std::cout << GridLogMessage<< "* Using UNROLLED 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;
const int Ls=4;
const int L =4;
std::vector<int> latt_2f(Nd,L);
std::vector<int> latt_1f(Nd,L); latt_1f[nu] = 2*L;
//const int L =4;
//std::vector<int> latt_2f(Nd,L);
std::vector<int> latt_2f = GridDefaultLatt();
std::vector<int> latt_1f(latt_2f); latt_1f[nu] = 2*latt_2f[nu];
int L = latt_2f[nu];
std::vector<int> simd_layout = GridDefaultSimd(Nd,vComplexType::Nsimd());
std::cout << GridLogMessage << "SIMD layout: ";
for(int i=0;i<simd_layout.size();i++) std::cout << simd_layout[i] << " ";
std::cout << std::endl;
std::vector<int> simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd());
std::vector<int> mpi_layout = GridDefaultMpi(); //node layout
GridCartesian * UGrid_1f = SpaceTimeGrid::makeFourDimGrid(latt_1f, simd_layout, mpi_layout);
@@ -67,13 +113,13 @@ int main (int argc, char ** argv)
GridParallelRNG RNG5_2f(FGrid_2f); RNG5_2f.SeedFixedIntegers(seeds5);
GridParallelRNG RNG4_2f(UGrid_2f); RNG4_2f.SeedFixedIntegers(seeds4);
LatticeGaugeField Umu_2f(UGrid_2f);
GparityGaugeField Umu_2f(UGrid_2f);
SU3::HotConfiguration(RNG4_2f,Umu_2f);
LatticeFermion src (FGrid_2f);
LatticeFermion tmpsrc(FGrid_2f);
FermionField src_2f(FGrid_2f);
LatticeFermion src_1f(FGrid_1f);
StandardFermionField src (FGrid_2f);
StandardFermionField tmpsrc(FGrid_2f);
GparityFermionField src_2f(FGrid_2f);
StandardFermionField src_1f(FGrid_1f);
// Replicate fermion source
random(RNG5_2f,src);
@@ -81,8 +127,8 @@ int main (int argc, char ** argv)
tmpsrc=src*2.0;
PokeIndex<0>(src_2f,tmpsrc,1);
LatticeFermion result_1f(FGrid_1f); result_1f=zero;
LatticeGaugeField Umu_1f(UGrid_1f);
StandardFermionField result_1f(FGrid_1f); result_1f=zero;
StandardGaugeField Umu_1f(UGrid_1f);
Replicate(Umu_2f,Umu_1f);
//Coordinate grid for reference
@@ -92,7 +138,7 @@ int main (int argc, char ** argv)
//Copy-conjugate the gauge field
//First C-shift the lattice by Lx/2
{
LatticeGaugeField Umu_shift = conjugate( Cshift(Umu_1f,nu,L) );
StandardGaugeField Umu_shift = conjugate( Cshift(Umu_1f,nu,L) );
Umu_1f = where( xcoor_1f >= Integer(L), Umu_shift, Umu_1f );
// hack test to check the same
@@ -101,7 +147,7 @@ int main (int argc, char ** argv)
cout << GridLogMessage << "Umu diff " << norm2(Umu_shift)<<std::endl;
//Make the gauge field antiperiodic in nu-direction
LatticeColourMatrix Unu(UGrid_1f);
decltype(PeekIndex<LorentzIndex>(Umu_1f,nu)) Unu(UGrid_1f);
Unu = PeekIndex<LorentzIndex>(Umu_1f,nu);
Unu = where(xcoor_1f == Integer(2*L-1), -Unu, Unu);
PokeIndex<LorentzIndex>(Umu_1f,Unu,nu);
@@ -115,33 +161,33 @@ int main (int argc, char ** argv)
RealD mass=0.0;
RealD M5=1.8;
DomainWallFermionR Ddwf(Umu_1f,*FGrid_1f,*FrbGrid_1f,*UGrid_1f,*UrbGrid_1f,mass,M5);
StandardDiracOp Ddwf(Umu_1f,*FGrid_1f,*FrbGrid_1f,*UGrid_1f,*UrbGrid_1f,mass,M5 DOP_PARAMS);
LatticeFermion src_o_1f(FrbGrid_1f);
LatticeFermion result_o_1f(FrbGrid_1f);
StandardFermionField src_o_1f(FrbGrid_1f);
StandardFermionField result_o_1f(FrbGrid_1f);
pickCheckerboard(Odd,src_o_1f,src_1f);
result_o_1f=zero;
SchurDiagMooeeOperator<DomainWallFermionR,LatticeFermion> HermOpEO(Ddwf);
ConjugateGradient<LatticeFermion> CG(1.0e-8,10000);
SchurDiagMooeeOperator<StandardDiracOp,StandardFermionField> HermOpEO(Ddwf);
ConjugateGradient<StandardFermionField> CG(1.0e-8,10000);
CG(HermOpEO,src_o_1f,result_o_1f);
// const int nu = 3;
std::vector<int> twists(Nd,0);
twists[nu] = 1;
GparityDomainWallFermionR::ImplParams params;
GparityDiracOp::ImplParams params;
params.twists = twists;
GparityDomainWallFermionR GPDdwf(Umu_2f,*FGrid_2f,*FrbGrid_2f,*UGrid_2f,*UrbGrid_2f,mass,M5,params);
GparityDiracOp GPDdwf(Umu_2f,*FGrid_2f,*FrbGrid_2f,*UGrid_2f,*UrbGrid_2f,mass,M5 DOP_PARAMS,params);
for(int disp=-1;disp<=1;disp+=2)
for(int mu=0;mu<5;mu++)
{
FermionField Dsrc_2f(FGrid_2f);
GparityFermionField Dsrc_2f(FGrid_2f);
LatticeFermion Dsrc_1f(FGrid_1f);
LatticeFermion Dsrc_2freplica(FGrid_1f);
LatticeFermion Dsrc_2freplica0(FGrid_1f);
LatticeFermion Dsrc_2freplica1(FGrid_1f);
StandardFermionField Dsrc_1f(FGrid_1f);
StandardFermionField Dsrc_2freplica(FGrid_1f);
StandardFermionField Dsrc_2freplica0(FGrid_1f);
StandardFermionField Dsrc_2freplica1(FGrid_1f);
if ( mu ==0 ) {
std::cout << GridLogMessage<< " Cross checking entire hopping term"<<std::endl;
@@ -156,8 +202,8 @@ int main (int argc, char ** argv)
std::cout << GridLogMessage << "S norms "<< norm2(src_2f) << " " << norm2(src_1f) <<std::endl;
std::cout << GridLogMessage << "D norms "<< norm2(Dsrc_2f)<< " " << norm2(Dsrc_1f) <<std::endl;
LatticeFermion Dsrc_2f0(FGrid_2f); Dsrc_2f0 = PeekIndex<0>(Dsrc_2f,0);
LatticeFermion Dsrc_2f1(FGrid_2f); Dsrc_2f1 = PeekIndex<0>(Dsrc_2f,1);
StandardFermionField Dsrc_2f0(FGrid_2f); Dsrc_2f0 = PeekIndex<0>(Dsrc_2f,0);
StandardFermionField Dsrc_2f1(FGrid_2f); Dsrc_2f1 = PeekIndex<0>(Dsrc_2f,1);
// Dsrc_2f1 = Dsrc_2f1 - Dsrc_2f0;
// std::cout << GridLogMessage << " Cross check two halves " <<norm2(Dsrc_2f1)<<std::endl;
@@ -174,20 +220,20 @@ int main (int argc, char ** argv)
}
{
FermionField chi (FGrid_2f); gaussian(RNG5_2f,chi);
FermionField phi (FGrid_2f); gaussian(RNG5_2f,phi);
GparityFermionField chi (FGrid_2f); gaussian(RNG5_2f,chi);
GparityFermionField phi (FGrid_2f); gaussian(RNG5_2f,phi);
FermionField chi_e (FrbGrid_2f);
FermionField chi_o (FrbGrid_2f);
GparityFermionField chi_e (FrbGrid_2f);
GparityFermionField chi_o (FrbGrid_2f);
FermionField dchi_e (FrbGrid_2f);
FermionField dchi_o (FrbGrid_2f);
GparityFermionField dchi_e (FrbGrid_2f);
GparityFermionField dchi_o (FrbGrid_2f);
FermionField phi_e (FrbGrid_2f);
FermionField phi_o (FrbGrid_2f);
GparityFermionField phi_e (FrbGrid_2f);
GparityFermionField phi_o (FrbGrid_2f);
FermionField dphi_e (FrbGrid_2f);
FermionField dphi_o (FrbGrid_2f);
GparityFermionField dphi_e (FrbGrid_2f);
GparityFermionField dphi_o (FrbGrid_2f);
pickCheckerboard(Even,chi_e,chi);
pickCheckerboard(Odd ,chi_o,chi);
@@ -212,14 +258,14 @@ int main (int argc, char ** argv)
}
FermionField result_2f(FGrid_2f); result_2f=zero;
FermionField src_o_2f(FrbGrid_2f);
FermionField result_o_2f(FrbGrid_2f);
GparityFermionField result_2f(FGrid_2f); result_2f=zero;
GparityFermionField src_o_2f(FrbGrid_2f);
GparityFermionField result_o_2f(FrbGrid_2f);
pickCheckerboard(Odd,src_o_2f,src_2f);
result_o_2f=zero;
ConjugateGradient<FermionField> CG2f(1.0e-8,10000);
SchurDiagMooeeOperator<GparityDomainWallFermionR,FermionField> HermOpEO2f(GPDdwf);
ConjugateGradient<GparityFermionField> CG2f(1.0e-8,10000);
SchurDiagMooeeOperator<GparityDiracOp,GparityFermionField> HermOpEO2f(GPDdwf);
CG2f(HermOpEO2f,src_o_2f,result_o_2f);
std::cout << "2f cb "<<result_o_2f.checkerboard<<std::endl;
@@ -227,10 +273,10 @@ int main (int argc, char ** argv)
std::cout << " result norms " <<norm2(result_o_2f)<<" " <<norm2(result_o_1f)<<std::endl;
LatticeFermion res0o (FrbGrid_2f);
LatticeFermion res1o (FrbGrid_2f);
LatticeFermion res0 (FGrid_2f);
LatticeFermion res1 (FGrid_2f);
StandardFermionField res0o (FrbGrid_2f);
StandardFermionField res1o (FrbGrid_2f);
StandardFermionField res0 (FGrid_2f);
StandardFermionField res1 (FGrid_2f);
res0=zero;
res1=zero;
@@ -244,9 +290,9 @@ int main (int argc, char ** argv)
setCheckerboard(res0,res0o);
setCheckerboard(res1,res1o);
LatticeFermion replica (FGrid_1f);
LatticeFermion replica0(FGrid_1f);
LatticeFermion replica1(FGrid_1f);
StandardFermionField replica (FGrid_1f);
StandardFermionField replica0(FGrid_1f);
StandardFermionField replica1(FGrid_1f);
Replicate(res0,replica0);
Replicate(res1,replica1);