portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-10 07:55:35 +00:00
Code Releases Activity

Added overlap comms compute mode

Browse Source
This commit is contained in:
paboyle 2016-01-03 01:38:11 +00:00
parent 4aac345bea
commit 331768dcff
13 changed files with 704 additions and 270 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
benchmarks/Benchmark_wilson.cc
View File

@ -85,11 +85,12 @@ int main (int argc, char ** argv)
Complex cone(1.0,0.0);
for(int nn=0;nn<Nd;nn++){
random(pRNG,U[nn]);
if(0) {
if (nn==-1) { U[nn]=zero; std::cout<<GridLogMessage << "zeroing gauge field in dir "<<nn<<std::endl; }
else { U[nn] = cone;std::cout<<GridLogMessage << "unit gauge field in dir "<<nn<<std::endl; }
if(1) {
if (nn!=2) { U[nn]=zero; std::cout<<GridLogMessage << "zeroing gauge field in dir "<<nn<<std::endl; }
// else { U[nn]= cone;std::cout<<GridLogMessage << "unit gauge field in dir "<<nn<<std::endl; }
else { std::cout<<GridLogMessage << "random gauge field in dir "<<nn<<std::endl; }
}
pokeIndex<LorentzIndex>(Umu,U[nn],nn);
PokeIndex<LorentzIndex>(Umu,U[nn],nn);
}
#endif

22
lib/Init.cc
View File

@ -238,15 +238,23 @@ void Grid_init(int *argc,char ***argv)
}
std::cout <<std::endl;
std::cout <<Logger::GREEN<< " GGG "<<Logger::RED<<" RRRR "<<Logger::BLUE<<" IIIII "<<Logger::PURPLE<<" DDDD "<<std::endl;
std::cout <<Logger::GREEN<< " G G "<<Logger::RED<<" R R "<<Logger::BLUE<<" I "<<Logger::PURPLE<<" D D "<<std::endl;
std::cout <<Logger::GREEN<< " G "<<Logger::RED<<" R R "<<Logger::BLUE<<" I "<<Logger::PURPLE<<" D D "<<std::endl;
std::cout <<Logger::GREEN<< " G GGG"<<Logger::RED<<" RRRR "<<Logger::BLUE<<" I "<<Logger::PURPLE<<" D D "<<std::endl;
std::cout <<Logger::GREEN<< " G G "<<Logger::RED<<" R R "<<Logger::BLUE<<" I "<<Logger::PURPLE<<" D D "<<std::endl;
std::cout <<Logger::GREEN<< " GGGG "<<Logger::RED<<" R R "<<Logger::BLUE<<" IIIII "<<Logger::PURPLE<<" DDDD "<<std::endl;
std::cout <<Logger::RED << "__|__|__|__|__"<< "|__|__|_"<< "_|__|__|"<< "__|__|__|__|__"<<std::endl;
std::cout <<Logger::RED << "__|__|__|__|__"<< "|__|__|_"<< "_|__|__|"<< "__|__|__|__|__"<<std::endl;
std::cout <<Logger::RED << "__|__|__| | "<< "| | | "<< " | | |"<< " | | | _|__"<<std::endl;
std::cout <<Logger::RED << "__|__ "<< " "<< " "<< " _|__"<<std::endl;
std::cout <<Logger::GREEN<< "__|_ GGGG "<<Logger::RED<<" RRRR "<<Logger::BLUE<<" III "<<Logger::PURPLE<<"DDDD _|__"<<std::endl;
std::cout <<Logger::GREEN<< "__|_ G "<<Logger::RED<<" R R "<<Logger::BLUE<<" I "<<Logger::PURPLE<<"D D _|__"<<std::endl;
std::cout <<Logger::GREEN<< "__|_ G "<<Logger::RED<<" R R "<<Logger::BLUE<<" I "<<Logger::PURPLE<<"D D _|__"<<std::endl;
std::cout <<Logger::GREEN<< "__|_ G GG "<<Logger::RED<<" RRRR "<<Logger::BLUE<<" I "<<Logger::PURPLE<<"D D _|__"<<std::endl;
std::cout <<Logger::GREEN<< "__|_ G G "<<Logger::RED<<" R R "<<Logger::BLUE<<" I "<<Logger::PURPLE<<"D D _|__"<<std::endl;
std::cout <<Logger::GREEN<< "__|_ GGGG "<<Logger::RED<<" R R "<<Logger::BLUE<<" III "<<Logger::PURPLE<<"DDDD _|__"<<std::endl;
std::cout <<Logger::BLUE << "__|__ "<< " "<< " "<< " _|__"<<std::endl;
std::cout <<Logger::BLUE << "__|__|__|__|__"<< "|__|____"<< "_|__|__|"<< "__|__|__|__|__"<<std::endl;
std::cout <<Logger::BLUE << "__|__|__|__|__"<< "|__|__|_"<< "_|__|__|"<< "__|__|__|__|__"<<std::endl;
std::cout <<Logger::BLUE << " | | | | "<< "| | | "<< " | | |"<< " | | | | "<<std::endl;
std::cout << std::endl;
std::cout << std::endl;
std::cout <<Logger::RED<< std::endl;
std::cout <<Logger::YELLOW<< std::endl;
std::cout << "Copyright (C) 2015 Peter Boyle, Azusa Yamaguchi, Guido Cossu, Antonin Portelli and other authors"<<std::endl;
std::cout << "Colours by Tadahito Boyle "<<std::endl;
std::cout << std::endl;

15
lib/Stencil.h
View File

@ -30,6 +30,8 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#ifndef GRID_STENCIL_H
#define GRID_STENCIL_H
#include <thread>
#include <stencil/Lebesgue.h> // subdir aggregate
//////////////////////////////////////////////////////////////////////////////////////////
@ -90,6 +92,7 @@ namespace Grid {
int _npoints; // Move to template param?
GridBase * _grid;
// npoints of these
std::vector<int> _directions;
std::vector<int> _distances;
@ -455,6 +458,18 @@ namespace Grid {
// Could allow a functional munging of the halo to another type during the comms.
// this could implement the 16bit/32bit/64bit compression.
void HaloExchange(const Lattice<vobj> &source,std::vector<cobj,alignedAllocator<cobj> > &u_comm_buf,compressor &compress)
{
std::thread thr = HaloExchangeBegin(source,u_comm_buf,compress);
thr.join();
}
std::thread HaloExchangeBegin(const Lattice<vobj> &source,std::vector<cobj,alignedAllocator<cobj> > & u_comm_buf,compressor &compress) {
return std::thread([&] { this->HaloExchangeBlocking(source,u_comm_buf,compress); });
// std::thread t(&HaloExchangeBlocking,this,source,u_comm_buf,compress);
// return t;
}
void HaloExchangeBlocking(const Lattice<vobj> &source,std::vector<cobj,alignedAllocator<cobj> > &u_comm_buf,compressor &compress)
{
// conformable(source._grid,_grid);
assert(source._grid==_grid);

8
lib/qcd/action/fermion/FermionOperatorImpl.h
View File

@ -110,10 +110,10 @@ namespace Grid {
// Single flavour four spinors with colour index
///////
template<class S,int Nrepresentation=Nc>
class WilsonImpl : public PeriodicGaugeImpl<S,Nrepresentation> {
class WilsonImpl : public PeriodicGaugeImpl< GaugeImplTypes< S,Nrepresentation> > {
public:
typedef PeriodicGaugeImpl<S,Nrepresentation> Gimpl;
typedef PeriodicGaugeImpl< GaugeImplTypes< S,Nrepresentation> > Gimpl;
INHERIT_GIMPL_TYPES(Gimpl);
@ -182,10 +182,10 @@ PARALLEL_FOR_LOOP
////////////////////////////////////////////////////////////////////////////////////////
template<class S,int Nrepresentation>
class GparityWilsonImpl : public ConjugateGaugeImpl<S,Nrepresentation> {
class GparityWilsonImpl : public ConjugateGaugeImpl< GaugeImplTypes<S,Nrepresentation> >{
public:
typedef ConjugateGaugeImpl<S,Nrepresentation> Gimpl;
typedef ConjugateGaugeImpl< GaugeImplTypes<S,Nrepresentation> > Gimpl;
INHERIT_GIMPL_TYPES(Gimpl);

78
lib/qcd/action/fermion/WilsonFermion.cc
View File

@ -227,7 +227,7 @@ PARALLEL_FOR_LOOP
out.checkerboard = in.checkerboard;
DhopInternal(Stencil,Umu,in,out,dag);
DhopInternalCommsCompute(Stencil,Umu,in,out,dag);
}
template<class Impl>
@ -238,7 +238,7 @@ PARALLEL_FOR_LOOP
assert(in.checkerboard==Even);
out.checkerboard = Odd;
DhopInternal(StencilEven,UmuOdd,in,out,dag);
DhopInternalCommsCompute(StencilEven,UmuOdd,in,out,dag);
}
template<class Impl>
@ -249,7 +249,7 @@ PARALLEL_FOR_LOOP
assert(in.checkerboard==Odd);
out.checkerboard = Even;
DhopInternal(StencilOdd,UmuEven,in,out,dag);
DhopInternalCommsCompute(StencilOdd,UmuEven,in,out,dag);
}
template<class Impl>
@ -320,6 +320,78 @@ PARALLEL_FOR_LOOP
}
};
template<class Impl>
void WilsonFermion<Impl>::DhopInternalCommsCompute(StencilImpl & st,DoubledGaugeField & U,
const FermionField &in, FermionField &out,int dag) {
assert((dag==DaggerNo) ||(dag==DaggerYes));
Compressor compressor(dag);
std::thread comms_thread = st.HaloExchangeBegin(in,comm_buf,compressor);
comms_thread.join();
bool local = true;
bool nonlocal = false;
if ( dag == DaggerYes ) {
if( HandOptDslash ) {
PARALLEL_FOR_LOOP
for(int sss=0;sss<in._grid->oSites();sss++){
Kernels::DiracOptHandDhopSiteDag(st,U,comm_buf,sss,sss,in,out,local,nonlocal);
}
} else {
PARALLEL_FOR_LOOP
for(int sss=0;sss<in._grid->oSites();sss++){
Kernels::DiracOptDhopSiteDag(st,U,comm_buf,sss,sss,in,out,local,nonlocal);
}
}
} else {
if( HandOptDslash ) {
PARALLEL_FOR_LOOP
for(int sss=0;sss<in._grid->oSites();sss++){
Kernels::DiracOptHandDhopSite(st,U,comm_buf,sss,sss,in,out,local,nonlocal);
}
} else {
PARALLEL_FOR_LOOP
for(int sss=0;sss<in._grid->oSites();sss++){
Kernels::DiracOptDhopSite(st,U,comm_buf,sss,sss,in,out,local,nonlocal);
}
}
}
local = false;
nonlocal = true;
if ( dag == DaggerYes ) {
if( HandOptDslash ) {
PARALLEL_FOR_LOOP
for(int sss=0;sss<in._grid->oSites();sss++){
Kernels::DiracOptHandDhopSiteDag(st,U,comm_buf,sss,sss,in,out,local,nonlocal);
}
} else {
PARALLEL_FOR_LOOP
for(int sss=0;sss<in._grid->oSites();sss++){
Kernels::DiracOptDhopSiteDag(st,U,comm_buf,sss,sss,in,out,local,nonlocal);
}
}
} else {
if( HandOptDslash ) {
PARALLEL_FOR_LOOP
for(int sss=0;sss<in._grid->oSites();sss++){
Kernels::DiracOptHandDhopSite(st,U,comm_buf,sss,sss,in,out,local,nonlocal);
}
} else {
PARALLEL_FOR_LOOP
for(int sss=0;sss<in._grid->oSites();sss++){
Kernels::DiracOptDhopSite(st,U,comm_buf,sss,sss,in,out,local,nonlocal);
}
}
}
};
FermOpTemplateInstantiate(WilsonFermion);

3
lib/qcd/action/fermion/WilsonFermion.h
View File

@ -114,6 +114,9 @@ namespace Grid {
void DhopInternal(StencilImpl & st,DoubledGaugeField & U,
const FermionField &in, FermionField &out,int dag) ;
void DhopInternalCommsCompute(StencilImpl & st,DoubledGaugeField & U,
const FermionField &in, FermionField &out,int dag) ;
// Constructor
WilsonFermion(GaugeField &_Umu,

137
lib/qcd/action/fermion/WilsonFermion5D.cc
View File

@ -396,6 +396,137 @@ PARALLEL_FOR_LOOP
}
dslashtime +=usecond();
}
template<class Impl>
void WilsonFermion5D<Impl>::DhopInternalCommsCompute(StencilImpl & st, LebesgueOrder &lo,
DoubledGaugeField & U,
const FermionField &in, FermionField &out,int dag)
{
// assert((dag==DaggerNo) ||(dag==DaggerYes));
Compressor compressor(dag);
// Assume balanced KMP_AFFINITY; this is forced in GridThread.h
int threads = GridThread::GetThreads();
int HT = GridThread::GetHyperThreads();
int cores = GridThread::GetCores();
int nwork = U._grid->oSites();
commtime -=usecond();
std::thread thr = st.HaloExchangeBegin(in,comm_buf,compressor);
commtime +=usecond();
// Dhop takes the 4d grid from U, and makes a 5d index for fermion
// Not loop ordering and data layout.
// Designed to create
// - per thread reuse in L1 cache for U
// - 8 linear access unit stride streams per thread for Fermion for hw prefetchable.
bool local = true;
bool nonlocal = false;
dslashtime -=usecond();
if ( dag == DaggerYes ) {
if( this->HandOptDslash ) {
PARALLEL_FOR_LOOP
for(int ss=0;ss<U._grid->oSites();ss++){
int sU=ss;
for(int s=0;s<Ls;s++){
int sF = s+Ls*sU;
Kernels::DiracOptHandDhopSiteDag(st,U,comm_buf,sF,sU,in,out,local,nonlocal);
}
}
} else {
PARALLEL_FOR_LOOP
for(int ss=0;ss<U._grid->oSites();ss++){
{
int sd;
for(sd=0;sd<Ls;sd++){
int sU=ss;
int sF = sd+Ls*sU;
Kernels::DiracOptDhopSiteDag(st,U,comm_buf,sF,sU,in,out,local,nonlocal);
}
}
}
}
} else {
if( this->HandOptDslash ) {
PARALLEL_FOR_LOOP
for(int ss=0;ss<U._grid->oSites();ss++){
int sU=ss;
for(int s=0;s<Ls;s++){
int sF = s+Ls*sU;
Kernels::DiracOptHandDhopSite(st,U,comm_buf,sF,sU,in,out,local,nonlocal);
}
}
} else {
PARALLEL_FOR_LOOP
for(int ss=0;ss<U._grid->oSites();ss++){
int sU=ss;
for(int s=0;s<Ls;s++){
int sF = s+Ls*sU;
Kernels::DiracOptDhopSite(st,U,comm_buf,sF,sU,in,out,local,nonlocal);
}
}
}
}
dslashtime +=usecond();
commtime -=usecond();
thr.join();
commtime +=usecond();
local = false;
nonlocal = true;
dslashtime -=usecond();
if ( dag == DaggerYes ) {
if( this->HandOptDslash ) {
PARALLEL_FOR_LOOP
for(int ss=0;ss<U._grid->oSites();ss++){
int sU=ss;
for(int s=0;s<Ls;s++){
int sF = s+Ls*sU;
Kernels::DiracOptHandDhopSiteDag(st,U,comm_buf,sF,sU,in,out,local,nonlocal);
}
}
} else {
PARALLEL_FOR_LOOP
for(int ss=0;ss<U._grid->oSites();ss++){
{
int sd;
for(sd=0;sd<Ls;sd++){
int sU=ss;
int sF = sd+Ls*sU;
Kernels::DiracOptDhopSiteDag(st,U,comm_buf,sF,sU,in,out,local,nonlocal);
}
}
}
}
} else {
if( this->HandOptDslash ) {
PARALLEL_FOR_LOOP
for(int ss=0;ss<U._grid->oSites();ss++){
int sU=ss;
for(int s=0;s<Ls;s++){
int sF = s+Ls*sU;
Kernels::DiracOptHandDhopSite(st,U,comm_buf,sF,sU,in,out,local,nonlocal);
}
}
} else {
PARALLEL_FOR_LOOP
for(int ss=0;ss<U._grid->oSites();ss++){
int sU=ss;
for(int s=0;s<Ls;s++){
int sF = s+Ls*sU;
Kernels::DiracOptDhopSite(st,U,comm_buf,sF,sU,in,out,local,nonlocal);
}
}
}
}
dslashtime +=usecond();
}
template<class Impl>
void WilsonFermion5D<Impl>::DhopOE(const FermionField &in, FermionField &out,int dag)
{
@ -405,7 +536,7 @@ void WilsonFermion5D<Impl>::DhopOE(const FermionField &in, FermionField &out,int
assert(in.checkerboard==Even);
out.checkerboard = Odd;
DhopInternal(StencilEven,LebesgueEvenOdd,UmuOdd,in,out,dag);
DhopInternalCommsCompute(StencilEven,LebesgueEvenOdd,UmuOdd,in,out,dag);
}
template<class Impl>
void WilsonFermion5D<Impl>::DhopEO(const FermionField &in, FermionField &out,int dag)
@ -416,7 +547,7 @@ void WilsonFermion5D<Impl>::DhopEO(const FermionField &in, FermionField &out,int
assert(in.checkerboard==Odd);
out.checkerboard = Even;
DhopInternal(StencilOdd,LebesgueEvenOdd,UmuEven,in,out,dag);
DhopInternalCommsCompute(StencilOdd,LebesgueEvenOdd,UmuEven,in,out,dag);
}
template<class Impl>
void WilsonFermion5D<Impl>::Dhop(const FermionField &in, FermionField &out,int dag)
@ -426,7 +557,7 @@ void WilsonFermion5D<Impl>::Dhop(const FermionField &in, FermionField &out,int d
out.checkerboard = in.checkerboard;
DhopInternal(Stencil,Lebesgue,Umu,in,out,dag);
DhopInternalCommsCompute(Stencil,Lebesgue,Umu,in,out,dag);
}
template<class Impl>
void WilsonFermion5D<Impl>::DW(const FermionField &in, FermionField &out,int dag)

7
lib/qcd/action/fermion/WilsonFermion5D.h
View File

@ -117,6 +117,13 @@ namespace Grid {
FermionField &out,
int dag);
void DhopInternalCommsCompute(StencilImpl & st,
LebesgueOrder &lo,
DoubledGaugeField &U,
const FermionField &in,
FermionField &out,
int dag);
// Constructors
WilsonFermion5D(GaugeField &_Umu,
GridCartesian &FiveDimGrid,

309
lib/qcd/action/fermion/WilsonKernels.cc
View File

@ -34,10 +34,11 @@ namespace QCD {
template<class Impl>
WilsonKernels<Impl>::WilsonKernels(const ImplParams &p): Base(p) {};
// Need controls to do interior, exterior, or both
template<class Impl>
void WilsonKernels<Impl>::DiracOptDhopSiteDag(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int sF,int sU,const FermionField &in, FermionField &out)
int sF,int sU,const FermionField &in, FermionField &out,bool local, bool nonlocal)
{
SiteHalfSpinor tmp;
SiteHalfSpinor chi;
@ -46,231 +47,425 @@ void WilsonKernels<Impl>::DiracOptDhopSiteDag(StencilImpl &st,DoubledGaugeField
StencilEntry *SE;
int ptype;
int num = 0;
result=zero;
///////////////////////////
// Xp
///////////////////////////
SE=st.GetEntry(ptype,Xp,sF);
if ( SE->_is_local && SE->_permute ) {
if (local && SE->_is_local ) {
if ( SE->_permute ) {
spProjXp(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
} else if ( SE->_is_local ) {
spProjXp(chi,in._odata[SE->_offset]);
} else {
spProjXp(chi,in._odata[SE->_offset]);
}
}
if ( nonlocal && (!SE->_is_local) ) {
chi=buf[SE->_offset];
}
Impl::multLink(Uchi,U._odata[sU],chi,Xp,SE,st);
spReconXp(result,Uchi);
if ( (local && SE->_is_local) || ( nonlocal && (!SE->_is_local)) ) {
Impl::multLink(Uchi,U._odata[sU],chi,Xp,SE,st);
accumReconXp(result,Uchi);
num++;
}
///////////////////////////
// Yp
///////////////////////////
SE=st.GetEntry(ptype,Yp,sF);
if ( SE->_is_local && SE->_permute ) {
if (local && SE->_is_local ) {
if ( SE->_permute ) {
spProjYp(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
} else if ( SE->_is_local ) {
spProjYp(chi,in._odata[SE->_offset]);
} else {
spProjYp(chi,in._odata[SE->_offset]);
}
}
if ( nonlocal && (!SE->_is_local) ) {
chi=buf[SE->_offset];
}
if ( (local && SE->_is_local) || ( nonlocal && (!SE->_is_local)) ) {
Impl::multLink(Uchi,U._odata[sU],chi,Yp,SE,st);
accumReconYp(result,Uchi);
num++;
}
///////////////////////////
// Zp
///////////////////////////
SE=st.GetEntry(ptype,Zp,sF);
if ( SE->_is_local && SE->_permute ) {
if (local && SE->_is_local ) {
if ( SE->_permute ) {
spProjZp(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
} else if ( SE->_is_local ) {
spProjZp(chi,in._odata[SE->_offset]);
} else {
spProjZp(chi,in._odata[SE->_offset]);
}
}
if ( nonlocal && (!SE->_is_local) ) {
chi=buf[SE->_offset];
}
if ( (local && SE->_is_local) || ( nonlocal && (!SE->_is_local)) ) {
Impl::multLink(Uchi,U._odata[sU],chi,Zp,SE,st);
accumReconZp(result,Uchi);
num++;
}
///////////////////////////
// Tp
///////////////////////////
SE=st.GetEntry(ptype,Tp,sF);
if ( SE->_is_local && SE->_permute ) {
if (local && SE->_is_local ) {
if ( SE->_permute ) {
spProjTp(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
} else if ( SE->_is_local ) {
spProjTp(chi,in._odata[SE->_offset]);
} else {
spProjTp(chi,in._odata[SE->_offset]);
}
}
if ( nonlocal && (!SE->_is_local) ) {
chi=buf[SE->_offset];
}
if ( (local && SE->_is_local) || ( nonlocal && (!SE->_is_local)) ) {
Impl::multLink(Uchi,U._odata[sU],chi,Tp,SE,st);
accumReconTp(result,Uchi);
num++;
}
///////////////////////////
// Xm
///////////////////////////
SE=st.GetEntry(ptype,Xm,sF);
if ( SE->_is_local && SE->_permute ) {
if (local && SE->_is_local ) {
if ( SE->_permute ) {
spProjXm(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
} else if ( SE->_is_local ) {
spProjXm(chi,in._odata[SE->_offset]);
} else {
spProjXm(chi,in._odata[SE->_offset]);
}
}
if ( nonlocal && (!SE->_is_local) ) {
chi=buf[SE->_offset];
}
if( (local && SE->_is_local) || ( nonlocal && (!SE->_is_local)) ) {
Impl::multLink(Uchi,U._odata[sU],chi,Xm,SE,st);
accumReconXm(result,Uchi);
num++;
}
///////////////////////////
// Ym
///////////////////////////
SE=st.GetEntry(ptype,Ym,sF);
if ( SE->_is_local && SE->_permute ) {
if (local && SE->_is_local ) {
if ( SE->_permute ) {
spProjYm(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
} else if ( SE->_is_local ) {
spProjYm(chi,in._odata[SE->_offset]);
} else {
spProjYm(chi,in._odata[SE->_offset]);
}
}
if ( nonlocal && (!SE->_is_local) ) {
chi=buf[SE->_offset];
}
if( (local && SE->_is_local) || ( nonlocal && (!SE->_is_local)) ) {
Impl::multLink(Uchi,U._odata[sU],chi,Ym,SE,st);
accumReconYm(result,Uchi);
num++;
}
///////////////////////////
// Zm
///////////////////////////
SE=st.GetEntry(ptype,Zm,sF);
if ( SE->_is_local && SE->_permute ) {
if (local && SE->_is_local ) {
if ( SE->_permute ) {
spProjZm(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
} else if ( SE->_is_local ) {
spProjZm(chi,in._odata[SE->_offset]);
} else {
spProjZm(chi,in._odata[SE->_offset]);
}
}
if ( nonlocal && (!SE->_is_local) ) {
chi=buf[SE->_offset];
}
if( (local && SE->_is_local) || ( nonlocal && (!SE->_is_local)) ) {
Impl::multLink(Uchi,U._odata[sU],chi,Zm,SE,st);
accumReconZm(result,Uchi);
num++;
}
///////////////////////////
// Tm
///////////////////////////
SE=st.GetEntry(ptype,Tm,sF);
if ( SE->_is_local && SE->_permute ) {
if (local && SE->_is_local ) {
if ( SE->_permute ) {
spProjTm(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
} else if ( SE->_is_local ) {
spProjTm(chi,in._odata[SE->_offset]);
} else {
spProjTm(chi,in._odata[SE->_offset]);
}
}
if ( nonlocal && (!SE->_is_local) ) {
chi=buf[SE->_offset];
}
if( (local && SE->_is_local) || ( nonlocal && (!SE->_is_local)) ) {
Impl::multLink(Uchi,U._odata[sU],chi,Tm,SE,st);
accumReconTm(result,Uchi);
num++;
}
if ( local ) {
vstream(out._odata[sF],result*(-0.5));
} else if ( num ) {
vstream(out._odata[sF],out._odata[sF]+result*(-0.5));
}
};
// Need controls to do interior, exterior, or both
template<class Impl>
void WilsonKernels<Impl>::DiracOptDhopSite(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int sF,int sU,const FermionField &in, FermionField &out)
int sF,int sU,const FermionField &in, FermionField &out,bool local, bool nonlocal)
{
SiteHalfSpinor tmp;
SiteHalfSpinor chi;
SiteSpinor result;
SiteHalfSpinor Uchi;
SiteSpinor result;
StencilEntry *SE;
int ptype;
int num = 0;
result=zero;
///////////////////////////
// Xp
///////////////////////////
SE=st.GetEntry(ptype,Xm,sF);
if ( SE->_is_local && SE->_permute ) {
if (local && SE->_is_local ) {
if ( SE->_permute ) {
spProjXp(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
} else if ( SE->_is_local ) {
spProjXp(chi,in._odata[SE->_offset]);
} else {
spProjXp(chi,in._odata[SE->_offset]);
}
}
if ( nonlocal && (!SE->_is_local) ) {
chi=buf[SE->_offset];
}
Impl::multLink(Uchi,U._odata[sU],chi,Xm,SE,st);
spReconXp(result,Uchi);
if ( (local && SE->_is_local) || ( nonlocal && (!SE->_is_local)) ) {
Impl::multLink(Uchi,U._odata[sU],chi,Xm,SE,st);
accumReconXp(result,Uchi);
num++;
}
///////////////////////////
// Yp
///////////////////////////
SE=st.GetEntry(ptype,Ym,sF);
if ( SE->_is_local && SE->_permute ) {
if (local && SE->_is_local ) {
if ( SE->_permute ) {
spProjYp(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
} else if ( SE->_is_local ) {
spProjYp(chi,in._odata[SE->_offset]);
} else {
spProjYp(chi,in._odata[SE->_offset]);
}
}
if ( nonlocal && (!SE->_is_local) ) {
chi=buf[SE->_offset];
}
if ( (local && SE->_is_local) || ( nonlocal && (!SE->_is_local)) ) {
Impl::multLink(Uchi,U._odata[sU],chi,Ym,SE,st);
accumReconYp(result,Uchi);
num++;
}
///////////////////////////
// Zp
///////////////////////////
SE=st.GetEntry(ptype,Zm,sF);
if ( SE->_is_local && SE->_permute ) {
if (local && SE->_is_local ) {
if ( SE->_permute ) {
spProjZp(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
} else if ( SE->_is_local ) {
spProjZp(chi,in._odata[SE->_offset]);
} else {
spProjZp(chi,in._odata[SE->_offset]);
}
}
if ( nonlocal && (!SE->_is_local) ) {
chi=buf[SE->_offset];
}
if ( (local && SE->_is_local) || ( nonlocal && (!SE->_is_local)) ) {
Impl::multLink(Uchi,U._odata[sU],chi,Zm,SE,st);
accumReconZp(result,Uchi);
num++;
}
///////////////////////////
// Tp
///////////////////////////
SE=st.GetEntry(ptype,Tm,sF);
if ( SE->_is_local && SE->_permute ) {
if (local && SE->_is_local ) {
if ( SE->_permute ) {
spProjTp(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
} else if ( SE->_is_local ) {
spProjTp(chi,in._odata[SE->_offset]);
} else {
spProjTp(chi,in._odata[SE->_offset]);
}
}
if ( nonlocal && (!SE->_is_local) ) {
chi=buf[SE->_offset];
}
if ( (local && SE->_is_local) || ( nonlocal && (!SE->_is_local)) ) {
Impl::multLink(Uchi,U._odata[sU],chi,Tm,SE,st);
accumReconTp(result,Uchi);
num++;
}
///////////////////////////
// Xm
///////////////////////////
SE=st.GetEntry(ptype,Xp,sF);
if ( SE->_is_local && SE->_permute ) {
if (local && SE->_is_local ) {
if ( SE->_permute ) {
spProjXm(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
} else if ( SE->_is_local ) {
spProjXm(chi,in._odata[SE->_offset]);
} else {
spProjXm(chi,in._odata[SE->_offset]);
}
}
if ( nonlocal && (!SE->_is_local) ) {
chi=buf[SE->_offset];
}
if( (local && SE->_is_local) || ( nonlocal && (!SE->_is_local)) ) {
Impl::multLink(Uchi,U._odata[sU],chi,Xp,SE,st);
accumReconXm(result,Uchi);
num++;
}
///////////////////////////
// Ym
///////////////////////////
SE=st.GetEntry(ptype,Yp,sF);
if ( SE->_is_local && SE->_permute ) {
if (local && SE->_is_local ) {
if ( SE->_permute ) {
spProjYm(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
} else if ( SE->_is_local ) {
spProjYm(chi,in._odata[SE->_offset]);
} else {
spProjYm(chi,in._odata[SE->_offset]);
}
}
if ( nonlocal && (!SE->_is_local) ) {
chi=buf[SE->_offset];
}
if( (local && SE->_is_local) || ( nonlocal && (!SE->_is_local)) ) {
Impl::multLink(Uchi,U._odata[sU],chi,Yp,SE,st);
accumReconYm(result,Uchi);
num++;
}
///////////////////////////
// Zm
///////////////////////////
SE=st.GetEntry(ptype,Zp,sF);
if ( SE->_is_local && SE->_permute ) {
if (local && SE->_is_local ) {
if ( SE->_permute ) {
spProjZm(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
} else if ( SE->_is_local ) {
spProjZm(chi,in._odata[SE->_offset]);
} else {
spProjZm(chi,in._odata[SE->_offset]);
}
}
if ( nonlocal && (!SE->_is_local) ) {
chi=buf[SE->_offset];
}
if( (local && SE->_is_local) || ( nonlocal && (!SE->_is_local)) ) {
Impl::multLink(Uchi,U._odata[sU],chi,Zp,SE,st);
accumReconZm(result,Uchi);
num++;
}
///////////////////////////
// Tm
///////////////////////////
SE=st.GetEntry(ptype,Tp,sF);
if ( SE->_is_local && SE->_permute ) {
if (local && SE->_is_local ) {
if ( SE->_permute ) {
spProjTm(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
} else if ( SE->_is_local ) {
spProjTm(chi,in._odata[SE->_offset]);
} else {
spProjTm(chi,in._odata[SE->_offset]);
}
}
if ( nonlocal && (!SE->_is_local) ) {
chi=buf[SE->_offset];
}
if( (local && SE->_is_local) || ( nonlocal && (!SE->_is_local)) ) {
Impl::multLink(Uchi,U._odata[sU],chi,Tp,SE,st);
accumReconTm(result,Uchi);
num++;
}
if ( local ) {
vstream(out._odata[sF],result*(-0.5));
}
} else if ( num ) {
vstream(out._odata[sF],out._odata[sF]+result*(-0.5));
}
};
template<class Impl>
void WilsonKernels<Impl>::DiracOptDhopDir(StencilImpl &st,DoubledGaugeField &U,
@ -405,7 +600,7 @@ void WilsonKernels<Impl>::DiracOptDhopDir(StencilImpl &st,DoubledGaugeField &U,
template<class Impl>
void WilsonKernels<Impl>::DiracOptAsmDhopSite(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int sF,int sU,const FermionField &in, FermionField &out,uint64_t *p)
int sF,int sU,const FermionField &in, FermionField &out,bool local, bool nonlocal)
{
DiracOptDhopSite(st,U,buf,sF,sU,in,out); // will template override for Wilson Nc=3
}

10
lib/qcd/action/fermion/WilsonKernels.h
View File

@ -48,11 +48,11 @@ namespace Grid {
public:
void DiracOptDhopSite(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int sF,int sU,const FermionField &in, FermionField &out);
int sF,int sU,const FermionField &in, FermionField &out,bool local= true, bool nonlocal=true);
void DiracOptDhopSiteDag(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int sF,int sU,const FermionField &in,FermionField &out);
int sF,int sU,const FermionField &in,FermionField &out,bool local= true, bool nonlocal=true);
void DiracOptDhopDir(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
@ -60,15 +60,15 @@ namespace Grid {
void DiracOptAsmDhopSite(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int sF,int sU,const FermionField &in, FermionField &out,uint64_t *);
int sF,int sU,const FermionField &in, FermionField &out,bool local= true, bool nonlocal=true);
void DiracOptHandDhopSite(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int sF,int sU,const FermionField &in, FermionField &out);
int sF,int sU,const FermionField &in, FermionField &out,bool local= true, bool nonlocal=true);
void DiracOptHandDhopSiteDag(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int sF,int sU,const FermionField &in, FermionField &out);
int sF,int sU,const FermionField &in, FermionField &out,bool local= true, bool nonlocal=true);
WilsonKernels(const ImplParams &p= ImplParams());

28
lib/qcd/action/fermion/WilsonKernelsHand.cc
View File

@ -312,7 +312,7 @@ namespace QCD {
template<class Impl>
void WilsonKernels<Impl >::DiracOptHandDhopSiteDag(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int sU,const FermionField &in, FermionField &out)
int ss,int sU,const FermionField &in, FermionField &out, bool Local, bool Nonlocal)
{
// std::cout << "Hand op Dhop "<<std::endl;
typedef typename Simd::scalar_type S;
@ -556,7 +556,7 @@ void WilsonKernels<Impl >::DiracOptHandDhopSiteDag(StencilImpl &st,DoubledGaugeF
template<class Impl>
void WilsonKernels<Impl >::DiracOptHandDhopSite(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int sU,const FermionField &in, FermionField &out)
int ss,int sU,const FermionField &in, FermionField &out, bool Local, bool Nonlocal)
{
typedef typename Simd::scalar_type S;
typedef typename Simd::vector_type V;
@ -802,7 +802,7 @@ void WilsonKernels<Impl >::DiracOptHandDhopSite(StencilImpl &st,DoubledGaugeFiel
template<>
void WilsonKernels<GparityWilsonImplF>::DiracOptHandDhopSite(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int sF,int sU,const FermionField &in, FermionField &out)
int sF,int sU,const FermionField &in, FermionField &out, bool Local, bool Nonlocal)
{
DiracOptDhopSite(st,U,buf,sF,sU,in,out); // will template override for Wilson Nc=3
}
@ -810,7 +810,7 @@ void WilsonKernels<GparityWilsonImplF>::DiracOptHandDhopSite(StencilImpl &st,Dou
template<>
void WilsonKernels<GparityWilsonImplF>::DiracOptHandDhopSiteDag(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int sF,int sU,const FermionField &in, FermionField &out)
int sF,int sU,const FermionField &in, FermionField &out, bool Local, bool Nonlocal)
{
DiracOptDhopSiteDag(st,U,buf,sF,sU,in,out); // will template override for Wilson Nc=3
}
@ -818,7 +818,7 @@ void WilsonKernels<GparityWilsonImplF>::DiracOptHandDhopSiteDag(StencilImpl &st,
template<>
void WilsonKernels<GparityWilsonImplD>::DiracOptHandDhopSite(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int sF,int sU,const FermionField &in, FermionField &out)
int sF,int sU,const FermionField &in, FermionField &out, bool Local, bool Nonlocal)
{
DiracOptDhopSite(st,U,buf,sF,sU,in,out); // will template override for Wilson Nc=3
}
@ -826,7 +826,7 @@ void WilsonKernels<GparityWilsonImplD>::DiracOptHandDhopSite(StencilImpl &st,Dou
template<>
void WilsonKernels<GparityWilsonImplD>::DiracOptHandDhopSiteDag(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int sF,int sU,const FermionField &in, FermionField &out)
int sF,int sU,const FermionField &in, FermionField &out, bool Local, bool Nonlocal)
{
DiracOptDhopSiteDag(st,U,buf,sF,sU,in,out); // will template override for Wilson Nc=3
}
@ -835,29 +835,29 @@ void WilsonKernels<GparityWilsonImplD>::DiracOptHandDhopSiteDag(StencilImpl &st,
template void WilsonKernels<WilsonImplF>::DiracOptHandDhopSite(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int sU,const FermionField &in, FermionField &out);
int ss,int sU,const FermionField &in, FermionField &out,bool l,bool n);
template void WilsonKernels<WilsonImplD>::DiracOptHandDhopSite(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int sU,const FermionField &in, FermionField &out);
int ss,int sU,const FermionField &in, FermionField &out, bool l, bool n);
template void WilsonKernels<WilsonImplF>::DiracOptHandDhopSiteDag(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int sU,const FermionField &in, FermionField &out);
int ss,int sU,const FermionField &in, FermionField &out, bool l, bool n);
template void WilsonKernels<WilsonImplD>::DiracOptHandDhopSiteDag(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int sU,const FermionField &in, FermionField &out);
int ss,int sU,const FermionField &in, FermionField &out, bool l, bool n);
template void WilsonKernels<GparityWilsonImplF>::DiracOptHandDhopSite(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int sU,const FermionField &in, FermionField &out);
int ss,int sU,const FermionField &in, FermionField &out, bool l, bool nl);
template void WilsonKernels<GparityWilsonImplD>::DiracOptHandDhopSite(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int sU,const FermionField &in, FermionField &out);
int ss,int sU,const FermionField &in, FermionField &out, bool l, bool nl);
template void WilsonKernels<GparityWilsonImplF>::DiracOptHandDhopSiteDag(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int sU,const FermionField &in, FermionField &out);
int ss,int sU,const FermionField &in, FermionField &out, bool l, bool nl);
template void WilsonKernels<GparityWilsonImplD>::DiracOptHandDhopSiteDag(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int sU,const FermionField &in, FermionField &out);
int ss,int sU,const FermionField &in, FermionField &out, bool l, bool nl);
}}

50
lib/qcd/action/gauge/GaugeImpl.h
View File

@ -45,10 +45,9 @@ template<class Gimpl> class WilsonLoops;
typedef typename GImpl::GaugeField GaugeField;
// Composition with smeared link, bc's etc.. probably need multiple inheritance
// Variable precision "S" and variable Nc
//
template<class S,int Nrepresentation=Nc>
class PeriodicGaugeImpl {
class GaugeImplTypes {
public:
typedef S Simd;
@ -62,6 +61,16 @@ template<class Gimpl> class WilsonLoops;
typedef Lattice<SiteGaugeLink> GaugeLinkField; // bit ugly naming; polarised gauge field, lorentz... all ugly
typedef Lattice<SiteGaugeField> GaugeField;
};
// Composition with smeared link, bc's etc.. probably need multiple inheritance
// Variable precision "S" and variable Nc
template<class GimplTypes>
class PeriodicGaugeImpl : public GimplTypes {
public:
INHERIT_GIMPL_TYPES(GimplTypes);
////////////////////////////////////////////////////////////////////////////////////////////////////////////
// Support needed for the assembly of loops including all boundary condition effects such as conjugate bcs
////////////////////////////////////////////////////////////////////////////////////////////////////////////
@ -92,38 +101,28 @@ template<class Gimpl> class WilsonLoops;
return true;
}
};
// Composition with smeared link, bc's etc.. probably need multiple inheritance
// Variable precision "S" and variable Nc
template<class S,int Nrepresentation=Nc>
class ConjugateGaugeImpl {
template<class GimplTypes>
class ConjugateGaugeImpl : public GimplTypes {
public:
typedef S Simd;
template<typename vtype> using iImplGaugeLink = iScalar<iScalar<iMatrix<vtype, Nrepresentation> > >;
template<typename vtype> using iImplGaugeField = iVector<iScalar<iMatrix<vtype, Nrepresentation> >, Nd >;
typedef iImplGaugeLink <Simd> SiteGaugeLink;
typedef iImplGaugeField <Simd> SiteGaugeField;
typedef Lattice<SiteGaugeLink> GaugeLinkField; // bit ugly naming; polarised gauge field, lorentz... all ugly
typedef Lattice<SiteGaugeField> GaugeField;
INHERIT_GIMPL_TYPES(GimplTypes);
////////////////////////////////////////////////////////////////////////////////////////////////////////////
// Support needed for the assembly of loops including all boundary condition effects such as Gparity.
////////////////////////////////////////////////////////////////////////////////////////////////////////////
template<class covariant> static
Lattice<covariant> CovShiftForward (const GaugeLinkField &Link, int mu, const Lattice<covariant> &field) {
return GparityBC::CovShiftForward(Link,mu,field);
return ConjugateBC::CovShiftForward(Link,mu,field);
}
template<class covariant> static
Lattice<covariant> CovShiftBackward(const GaugeLinkField &Link, int mu,const Lattice<covariant> &field) {
return GparityBC::CovShiftBackward(Link,mu,field);
return ConjugateBC::CovShiftBackward(Link,mu,field);
}
static inline
@ -162,14 +161,17 @@ template<class Gimpl> class WilsonLoops;
};
typedef GaugeImplTypes<vComplex,Nc> GimplTypesR;
typedef GaugeImplTypes<vComplexF,Nc> GimplTypesF;
typedef GaugeImplTypes<vComplexD,Nc> GimplTypesD;
typedef PeriodicGaugeImpl<vComplex ,Nc> PeriodicGimplR; // Real.. whichever prec
typedef PeriodicGaugeImpl<vComplexF,Nc> PeriodicGimplF; // Float
typedef PeriodicGaugeImpl<vComplexD,Nc> PeriodicGimplD; // Double
typedef PeriodicGaugeImpl<GimplTypesR> PeriodicGimplR; // Real.. whichever prec
typedef PeriodicGaugeImpl<GimplTypesF> PeriodicGimplF; // Float
typedef PeriodicGaugeImpl<GimplTypesD> PeriodicGimplD; // Double
typedef ConjugateGaugeImpl<vComplex ,Nc> ConjugateGimplR; // Real.. whichever prec
typedef ConjugateGaugeImpl<vComplexF,Nc> ConjugateGimplF; // Float
typedef ConjugateGaugeImpl<vComplexD,Nc> ConjugateGimplD; // Double
typedef ConjugateGaugeImpl<GimplTypesR> ConjugateGimplR; // Real.. whichever prec
typedef ConjugateGaugeImpl<GimplTypesF> ConjugateGimplF; // Float
typedef ConjugateGaugeImpl<GimplTypesD> ConjugateGimplD; // Double
}
}

2
lib/qcd/utils/CovariantCshift.h
View File

@ -55,7 +55,7 @@ namespace PeriodicBC {
}
namespace GparityBC {
namespace ConjugateBC {
// Must give right answers across boundary
// <----