portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-09-20 01:05:38 +01:00
Code Releases Activity

Merge branch 'develop' into feature/hirep

Browse Source
This commit is contained in:
Guido Cossu 2016-09-13 10:01:51 +01:00
commit b9c80318a2
5 changed files with 291 additions and 304 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
bootstrap.sh
View File

@ -4,7 +4,7 @@ EIGEN_URL='http://bitbucket.org/eigen/eigen/get/3.2.9.tar.bz2'
FFTW_URL=http://www.fftw.org/fftw-3.3.4.tar.gz
echo "-- deploying Eigen source..."
wget ${EIGEN_URL}
wget ${EIGEN_URL} --no-check-certificate
./scripts/update_eigen.sh `basename ${EIGEN_URL}`
rm `basename ${EIGEN_URL}`

166
lib/parallelIO/BinaryIO.h
View File

@ -194,22 +194,22 @@ class BinaryIO {
std::vector<int> site({x,y,z,t});
if ( grid->IsBoss() ) {
fin.read((char *)&file_object,sizeof(file_object));
bytes += sizeof(file_object);
if(ieee32big) be32toh_v((void *)&file_object,sizeof(file_object));
if(ieee32) le32toh_v((void *)&file_object,sizeof(file_object));
if(ieee64big) be64toh_v((void *)&file_object,sizeof(file_object));
if(ieee64) le64toh_v((void *)&file_object,sizeof(file_object));
if (grid->IsBoss()) {
fin.read((char *)&file_object, sizeof(file_object));
bytes += sizeof(file_object);
if (ieee32big) be32toh_v((void *)&file_object, sizeof(file_object));
if (ieee32) le32toh_v((void *)&file_object, sizeof(file_object));
if (ieee64big) be64toh_v((void *)&file_object, sizeof(file_object));
if (ieee64) le64toh_v((void *)&file_object, sizeof(file_object));
munge(file_object,munged,csum);
munge(file_object, munged, csum);
}
// The boss who read the file has their value poked
pokeSite(munged,Umu,site);
}}}}
timer.Stop();
std::cout<<GridLogPerformance<<"readObjectSerial: read "<< bytes <<" bytes in "<<timer.Elapsed() <<" "
<< (double)bytes/ (double)timer.useconds() <<" MB/s " <<std::endl;
<< (double)bytes/ (double)timer.useconds() <<" MB/s " <<std::endl;
return csum;
}
@ -254,20 +254,20 @@ class BinaryIO {
if ( grid->IsBoss() ) {
if(ieee32big) htobe32_v((void *)&file_object,sizeof(file_object));
if(ieee32) htole32_v((void *)&file_object,sizeof(file_object));
if(ieee64big) htobe64_v((void *)&file_object,sizeof(file_object));
if(ieee64) htole64_v((void *)&file_object,sizeof(file_object));
if(ieee32big) htobe32_v((void *)&file_object,sizeof(file_object));
if(ieee32) htole32_v((void *)&file_object,sizeof(file_object));
if(ieee64big) htobe64_v((void *)&file_object,sizeof(file_object));
if(ieee64) htole64_v((void *)&file_object,sizeof(file_object));
// NB could gather an xstrip as an optimisation.
fout.write((char *)&file_object,sizeof(file_object));
bytes+=sizeof(file_object);
// NB could gather an xstrip as an optimisation.
fout.write((char *)&file_object,sizeof(file_object));
bytes+=sizeof(file_object);
}
}}}}
timer.Stop();
std::cout<<GridLogPerformance<<"writeObjectSerial: wrote "<< bytes <<" bytes in "<<timer.Elapsed() <<" "
<< (double)bytes/timer.useconds() <<" MB/s " <<std::endl;
<< (double)bytes/timer.useconds() <<" MB/s " <<std::endl;
return csum;
}
@ -305,15 +305,15 @@ class BinaryIO {
int l_idx=parallel.generator_idx(o_idx,i_idx);
if( rank == grid->ThisRank() ){
// std::cout << "rank" << rank<<" Getting state for index "<<l_idx<<std::endl;
parallel.GetState(saved,l_idx);
// std::cout << "rank" << rank<<" Getting state for index "<<l_idx<<std::endl;
parallel.GetState(saved,l_idx);
}
grid->Broadcast(rank,(void *)&saved[0],bytes);
if ( grid->IsBoss() ) {
Uint32Checksum((uint32_t *)&saved[0],bytes,csum);
fout.write((char *)&saved[0],bytes);
Uint32Checksum((uint32_t *)&saved[0],bytes,csum);
fout.write((char *)&saved[0],bytes);
}
}
@ -355,14 +355,14 @@ class BinaryIO {
int l_idx=parallel.generator_idx(o_idx,i_idx);
if ( grid->IsBoss() ) {
fin.read((char *)&saved[0],bytes);
Uint32Checksum((uint32_t *)&saved[0],bytes,csum);
fin.read((char *)&saved[0],bytes);
Uint32Checksum((uint32_t *)&saved[0],bytes,csum);
}
grid->Broadcast(0,(void *)&saved[0],bytes);
if( rank == grid->ThisRank() ){
parallel.SetState(saved,l_idx);
parallel.SetState(saved,l_idx);
}
}
@ -415,15 +415,15 @@ class BinaryIO {
if ( d == 0 ) parallel[d] = 0;
if (parallel[d]) {
range[d] = grid->_ldimensions[d];
start[d] = grid->_processor_coor[d]*range[d];
ioproc[d]= grid->_processor_coor[d];
range[d] = grid->_ldimensions[d];
start[d] = grid->_processor_coor[d]*range[d];
ioproc[d]= grid->_processor_coor[d];
} else {
range[d] = grid->_gdimensions[d];
start[d] = 0;
ioproc[d]= 0;
range[d] = grid->_gdimensions[d];
start[d] = 0;
ioproc[d]= 0;
if ( grid->_processor_coor[d] != 0 ) IOnode = 0;
if ( grid->_processor_coor[d] != 0 ) IOnode = 0;
}
slice_vol = slice_vol * range[d];
}
@ -434,9 +434,9 @@ class BinaryIO {
std::cout<< std::dec ;
std::cout<< GridLogMessage<< "Parallel read I/O to "<< file << " with " <<tmp<< " IOnodes for subslice ";
for(int d=0;d<grid->_ndimension;d++){
std::cout<< range[d];
if( d< grid->_ndimension-1 )
std::cout<< " x ";
std::cout<< range[d];
if( d< grid->_ndimension-1 )
std::cout<< " x ";
}
std::cout << std::endl;
}
@ -463,7 +463,7 @@ class BinaryIO {
// need to implement these loops in Nd independent way with a lexico conversion
for(int tlex=0;tlex<slice_vol;tlex++){
std::vector<int> tsite(nd); // temporary mixed up site
std::vector<int> gsite(nd);
std::vector<int> lsite(nd);
@ -472,8 +472,8 @@ class BinaryIO {
Lexicographic::CoorFromIndex(tsite,tlex,range);
for(int d=0;d<nd;d++){
lsite[d] = tsite[d]%grid->_ldimensions[d]; // local site
gsite[d] = tsite[d]+start[d]; // global site
lsite[d] = tsite[d]%grid->_ldimensions[d]; // local site
gsite[d] = tsite[d]+start[d]; // global site
}
/////////////////////////
@ -487,29 +487,29 @@ class BinaryIO {
// iorank reads from the seek
////////////////////////////////
if (myrank == iorank) {
fin.seekg(offset+g_idx*sizeof(fileObj));
fin.read((char *)&fileObj,sizeof(fileObj));
bytes+=sizeof(fileObj);
if(ieee32big) be32toh_v((void *)&fileObj,sizeof(fileObj));
if(ieee32) le32toh_v((void *)&fileObj,sizeof(fileObj));
if(ieee64big) be64toh_v((void *)&fileObj,sizeof(fileObj));
if(ieee64) le64toh_v((void *)&fileObj,sizeof(fileObj));
munge(fileObj,siteObj,csum);
fin.seekg(offset+g_idx*sizeof(fileObj));
fin.read((char *)&fileObj,sizeof(fileObj));
bytes+=sizeof(fileObj);
if(ieee32big) be32toh_v((void *)&fileObj,sizeof(fileObj));
if(ieee32) le32toh_v((void *)&fileObj,sizeof(fileObj));
if(ieee64big) be64toh_v((void *)&fileObj,sizeof(fileObj));
if(ieee64) le64toh_v((void *)&fileObj,sizeof(fileObj));
munge(fileObj,siteObj,csum);
}
}
// Possibly do transport through pt2pt
if ( rank != iorank ) {
if ( (myrank == rank) || (myrank==iorank) ) {
grid->SendRecvPacket((void *)&siteObj,(void *)&siteObj,iorank,rank,sizeof(siteObj));
}
if ( (myrank == rank) || (myrank==iorank) ) {
grid->SendRecvPacket((void *)&siteObj,(void *)&siteObj,iorank,rank,sizeof(siteObj));
}
}
// Poke at destination
if ( myrank == rank ) {
pokeLocalSite(siteObj,Umu,lsite);
pokeLocalSite(siteObj,Umu,lsite);
}
grid->Barrier(); // necessary?
}
@ -520,7 +520,7 @@ class BinaryIO {
timer.Stop();
std::cout<<GridLogPerformance<<"readObjectParallel: read "<< bytes <<" bytes in "<<timer.Elapsed() <<" "
<< (double)bytes/timer.useconds() <<" MB/s " <<std::endl;
<< (double)bytes/timer.useconds() <<" MB/s " <<std::endl;
return csum;
}
@ -558,15 +558,15 @@ class BinaryIO {
if ( d!= grid->_ndimension-1 ) parallel[d] = 0;
if (parallel[d]) {
range[d] = grid->_ldimensions[d];
start[d] = grid->_processor_coor[d]*range[d];
ioproc[d]= grid->_processor_coor[d];
range[d] = grid->_ldimensions[d];
start[d] = grid->_processor_coor[d]*range[d];
ioproc[d]= grid->_processor_coor[d];
} else {
range[d] = grid->_gdimensions[d];
start[d] = 0;
ioproc[d]= 0;
range[d] = grid->_gdimensions[d];
start[d] = 0;
ioproc[d]= 0;
if ( grid->_processor_coor[d] != 0 ) IOnode = 0;
if ( grid->_processor_coor[d] != 0 ) IOnode = 0;
}
slice_vol = slice_vol * range[d];
@ -577,9 +577,9 @@ class BinaryIO {
grid->GlobalSum(tmp);
std::cout<< GridLogMessage<< "Parallel write I/O from "<< file << " with " <<tmp<< " IOnodes for subslice ";
for(int d=0;d<grid->_ndimension;d++){
std::cout<< range[d];
if( d< grid->_ndimension-1 )
std::cout<< " x ";
std::cout<< range[d];
if( d< grid->_ndimension-1 )
std::cout<< " x ";
}
std::cout << std::endl;
}
@ -610,7 +610,7 @@ class BinaryIO {
// should aggregate a whole chunk and then write.
// need to implement these loops in Nd independent way with a lexico conversion
for(int tlex=0;tlex<slice_vol;tlex++){
std::vector<int> tsite(nd); // temporary mixed up site
std::vector<int> gsite(nd);
std::vector<int> lsite(nd);
@ -619,8 +619,8 @@ class BinaryIO {
Lexicographic::CoorFromIndex(tsite,tlex,range);
for(int d=0;d<nd;d++){
lsite[d] = tsite[d]%grid->_ldimensions[d]; // local site
gsite[d] = tsite[d]+start[d]; // global site
lsite[d] = tsite[d]%grid->_ldimensions[d]; // local site
gsite[d] = tsite[d]+start[d]; // global site
}
@ -640,26 +640,26 @@ class BinaryIO {
// Pair of nodes may need to do pt2pt send
if ( rank != iorank ) { // comms is necessary
if ( (myrank == rank) || (myrank==iorank) ) { // and we have to do it
// Send to IOrank
grid->SendRecvPacket((void *)&siteObj,(void *)&siteObj,rank,iorank,sizeof(siteObj));
}
if ( (myrank == rank) || (myrank==iorank) ) { // and we have to do it
// Send to IOrank
grid->SendRecvPacket((void *)&siteObj,(void *)&siteObj,rank,iorank,sizeof(siteObj));
}
}
grid->Barrier(); // necessary?
if (myrank == iorank) {
munge(siteObj,fileObj,csum);
munge(siteObj,fileObj,csum);
if(ieee32big) htobe32_v((void *)&fileObj,sizeof(fileObj));
if(ieee32) htole32_v((void *)&fileObj,sizeof(fileObj));
if(ieee64big) htobe64_v((void *)&fileObj,sizeof(fileObj));
if(ieee64) htole64_v((void *)&fileObj,sizeof(fileObj));
fout.seekp(offset+g_idx*sizeof(fileObj));
fout.write((char *)&fileObj,sizeof(fileObj));
bytes+=sizeof(fileObj);
if(ieee32big) htobe32_v((void *)&fileObj,sizeof(fileObj));
if(ieee32) htole32_v((void *)&fileObj,sizeof(fileObj));
if(ieee64big) htobe64_v((void *)&fileObj,sizeof(fileObj));
if(ieee64) htole64_v((void *)&fileObj,sizeof(fileObj));
fout.seekp(offset+g_idx*sizeof(fileObj));
fout.write((char *)&fileObj,sizeof(fileObj));
bytes+=sizeof(fileObj);
}
}
@ -668,7 +668,7 @@ class BinaryIO {
timer.Stop();
std::cout<<GridLogPerformance<<"writeObjectParallel: wrote "<< bytes <<" bytes in "<<timer.Elapsed() <<" "
<< (double)bytes/timer.useconds() <<" MB/s " <<std::endl;
<< (double)bytes/timer.useconds() <<" MB/s " <<std::endl;
return csum;
}

9
lib/qcd/QCD.h
View File

@ -55,11 +55,14 @@ namespace QCD {
//////////////////////////////////////////////////////////////////////////////
// QCD iMatrix types
// Index conventions: Lorentz x Spin x Colour
// note: static const int or constexpr will work for type deductions
// with the intel compiler (up to version 17)
//////////////////////////////////////////////////////////////////////////////
static const int ColourIndex = 2;
static const int SpinIndex = 1;
static const int LorentzIndex= 0;
#define ColourIndex 2
#define SpinIndex 1
#define LorentzIndex 0
// Also should make these a named enum type
static const int DaggerNo=0;
static const int DaggerYes=1;

289
lib/qcd/action/fermion/WilsonKernels.h
View File

@ -49,154 +49,171 @@ namespace Grid {
template<class Impl> class WilsonKernels : public FermionOperator<Impl> , public WilsonKernelsStatic {
public:
INHERIT_IMPL_TYPES(Impl);
typedef FermionOperator<Impl> Base;
INHERIT_IMPL_TYPES(Impl);
typedef FermionOperator<Impl> Base;
public:
template <bool EnableBool = true>
typename std::enable_if<Impl::Dimension == 3 && Nc == 3 &&EnableBool, void>::type
DiracOptDhopSite(
StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU, int Ls, int Ns, const FermionField &in,
FermionField &out) {
template <bool EnableBool = true>
typename std::enable_if<Impl::Dimension == 3 && Nc == 3 &&EnableBool, void>::type
DiracOptDhopSite(
StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU, int Ls, int Ns, const FermionField &in,
FermionField &out) {
#ifdef AVX512
if (AsmOpt) {
WilsonKernels<Impl>::DiracOptAsmDhopSite(st, lo, U, buf, sF, sU, Ls, Ns,
in, out);
if (AsmOpt) {
WilsonKernels<Impl>::DiracOptAsmDhopSite(st, lo, U, buf, sF, sU, Ls, Ns,
in, out);
} else {
} else {
#else
{
{
#endif
for (int site = 0; site < Ns; site++) {
for (int s = 0; s < Ls; s++) {
if (HandOpt)
WilsonKernels<Impl>::DiracOptHandDhopSite(st, lo, U, buf, sF, sU,
in, out);
else
WilsonKernels<Impl>::DiracOptGenericDhopSite(st, lo, U, buf, sF, sU,
in, out);
sF++;
}
sU++;
}
}
}
for (int site = 0; site < Ns; site++) {
for (int s = 0; s < Ls; s++) {
if (HandOpt)
WilsonKernels<Impl>::DiracOptHandDhopSite(st, lo, U, buf, sF, sU,
in, out);
else
WilsonKernels<Impl>::DiracOptGenericDhopSite(st, lo, U, buf, sF, sU,
in, out);
sF++;
}
sU++;
}
}
}
template <bool EnableBool = true>
typename std::enable_if<(Impl::Dimension != 3 || (Impl::Dimension == 3 && Nc != 3)) && EnableBool, void>::type
DiracOptDhopSite(
StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU, int Ls, int Ns, const FermionField &in,
FermionField &out) {
for (int site = 0; site < Ns; site++) {
for (int s = 0; s < Ls; s++) {
WilsonKernels<Impl>::DiracOptGenericDhopSite(st, lo, U, buf, sF, sU, in,
out);
sF++;
}
sU++;
}
}
template <bool EnableBool = true>
typename std::enable_if<(Impl::Dimension != 3 || (Impl::Dimension == 3 && Nc != 3)) && EnableBool, void>::type
DiracOptDhopSite(
StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU, int Ls, int Ns, const FermionField &in,
FermionField &out) {
for (int site = 0; site < Ns; site++) {
for (int s = 0; s < Ls; s++) {
WilsonKernels<Impl>::DiracOptGenericDhopSite(st, lo, U, buf, sF, sU, in,
out);
sF++;
}
sU++;
}
}
template <bool EnableBool = true>
typename std::enable_if<Impl::Dimension == 3 && Nc == 3 && EnableBool,
void>::type
DiracOptDhopSiteDag(
StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU, int Ls, int Ns, const FermionField &in,
FermionField &out) {
template <bool EnableBool = true>
typename std::enable_if<Impl::Dimension == 3 && Nc == 3 && EnableBool,
void>::type
DiracOptDhopSiteDag(
StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU, int Ls, int Ns, const FermionField &in,
FermionField &out) {
#ifdef AVX512
if (AsmOpt) {
WilsonKernels<Impl>::DiracOptAsmDhopSiteDag(st, lo, U, buf, sF, sU, Ls,
Ns, in, out);
} else {
if (AsmOpt) {
WilsonKernels<Impl>::DiracOptAsmDhopSiteDag(st, lo, U, buf, sF, sU, Ls,
Ns, in, out);
} else {
#else
{
{
#endif
for (int site = 0; site < Ns; site++) {
for (int s = 0; s < Ls; s++) {
if (HandOpt)
WilsonKernels<Impl>::DiracOptHandDhopSiteDag(st, lo, U, buf, sF, sU,
in, out);
else
WilsonKernels<Impl>::DiracOptGenericDhopSiteDag(st, lo, U, buf, sF,
sU, in, out);
sF++;
}
sU++;
for (int site = 0; site < Ns; site++) {
for (int s = 0; s < Ls; s++) {
if (HandOpt)
WilsonKernels<Impl>::DiracOptHandDhopSiteDag(st, lo, U, buf, sF, sU,
in, out);
else
WilsonKernels<Impl>::DiracOptGenericDhopSiteDag(st, lo, U, buf, sF,
sU, in, out);
sF++;
}
sU++;
}
}
}
template <bool EnableBool = true>
typename std::enable_if<
(Impl::Dimension != 3 || (Impl::Dimension == 3 && Nc != 3)) && EnableBool,
void>::type
DiracOptDhopSiteDag(
StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU, int Ls, int Ns, const FermionField &in,
FermionField &out) {
for (int site = 0; site < Ns; site++) {
for (int s = 0; s < Ls; s++) {
WilsonKernels<Impl>::DiracOptGenericDhopSiteDag(st, lo, U, buf, sF, sU,
in, out);
sF++;
}
sU++;
}
}
void DiracOptDhopDir(
StencilImpl &st, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU, const FermionField &in, FermionField &out, int dirdisp,
int gamma);
private:
// Specialised variants
void DiracOptGenericDhopSite(
StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU, const FermionField &in, FermionField &out);
void DiracOptGenericDhopSiteDag(
StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU, const FermionField &in, FermionField &out);
void DiracOptAsmDhopSite(
StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU, int Ls, int Ns, const FermionField &in,
FermionField &out);
void DiracOptAsmDhopSiteDag(
StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU, int Ls, int Ns, const FermionField &in,
FermionField &out);
void DiracOptHandDhopSite(
StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU, const FermionField &in, FermionField &out);
void DiracOptHandDhopSiteDag(
StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU, const FermionField &in, FermionField &out);
public:
WilsonKernels(const ImplParams &p = ImplParams());
};
///////////////////////////////////////////////////////////
// Default to no assembler implementation
///////////////////////////////////////////////////////////
template<class Impl>
void WilsonKernels<Impl >::DiracOptAsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
{
assert(0);
}
template<class Impl>
void WilsonKernels<Impl >::DiracOptAsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
{
assert(0);
}
}
}
}
template <bool EnableBool = true>
typename std::enable_if<
(Impl::Dimension != 3 || (Impl::Dimension == 3 && Nc != 3)) && EnableBool,
void>::type
DiracOptDhopSiteDag(
StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU, int Ls, int Ns, const FermionField &in,
FermionField &out) {
for (int site = 0; site < Ns; site++) {
for (int s = 0; s < Ls; s++) {
WilsonKernels<Impl>::DiracOptGenericDhopSiteDag(st, lo, U, buf, sF, sU,
in, out);
sF++;
}
sU++;
}
}
void DiracOptDhopDir(
StencilImpl &st, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU, const FermionField &in, FermionField &out, int dirdisp,
int gamma);
private:
// Specialised variants
void DiracOptGenericDhopSite(
StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU, const FermionField &in, FermionField &out);
void DiracOptGenericDhopSiteDag(
StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU, const FermionField &in, FermionField &out);
void DiracOptAsmDhopSite(
StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU, int Ls, int Ns, const FermionField &in,
FermionField &out);
void DiracOptAsmDhopSiteDag(
StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU, int Ls, int Ns, const FermionField &in,
FermionField &out);
void DiracOptHandDhopSite(
StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU, const FermionField &in, FermionField &out);
void DiracOptHandDhopSiteDag(
StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU, const FermionField &in, FermionField &out);
public:
WilsonKernels(const ImplParams &p = ImplParams());
};
}
}
#endif

129
lib/qcd/action/fermion/WilsonKernelsAsm.cc
View File

@ -1,4 +1,4 @@
/*************************************************************************************
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
@ -26,68 +26,56 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
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 */
*************************************************************************************/
/* END LEGAL */
#include <Grid.h>
namespace Grid {
namespace QCD {
///////////////////////////////////////////////////////////
// Default to no assembler implementation
///////////////////////////////////////////////////////////
template<class Impl>
void WilsonKernels<Impl>::DiracOptAsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
{
assert(0);
}
namespace QCD {
#if defined(AVX512)
///////////////////////////////////////////////////////////
// If we are AVX512 specialise the single precision routine
///////////////////////////////////////////////////////////
///////////////////////////////////////////////////////////
// If we are AVX512 specialise the single precision routine
///////////////////////////////////////////////////////////
#include <simd/Intel512wilson.h>
#include <simd/Intel512single.h>
static Vector<vComplexF> signs;
int setupSigns(void ){
Vector<vComplexF> bother(2);
signs = bother;
vrsign(signs[0]);
visign(signs[1]);
return 1;
}
static int signInit = setupSigns();
static Vector<vComplexF> signs;
int setupSigns(void ){
Vector<vComplexF> bother(2);
signs = bother;
vrsign(signs[0]);
visign(signs[1]);
return 1;
}
static int signInit = setupSigns();
#define label(A) ilabel(A)
#define ilabel(A) ".globl\n" #A ":\n"
#define MAYBEPERM(A,perm) if (perm) { A ; }
#define MULT_2SPIN(ptr,pf) MULT_ADDSUB_2SPIN(ptr,pf)
#define FX(A) WILSONASM_ ##A
#undef KERNEL_DAG
template<>
void WilsonKernels<WilsonImplF>::DiracOptAsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
template<>
void WilsonKernels<WilsonImplF>::DiracOptAsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
#define KERNEL_DAG
template<>
void WilsonKernels<WilsonImplF>::DiracOptAsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
template<>
void WilsonKernels<WilsonImplF>::DiracOptAsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
#undef VMOVIDUP
#undef VMOVRDUP
#undef MAYBEPERM
@ -98,43 +86,22 @@ void WilsonKernels<WilsonImplF>::DiracOptAsmDhopSiteDag(StencilImpl &st,Lebesgue
#define VMOVIDUP(A,B,C) VBCASTIDUPf(A,B,C)
#define VMOVRDUP(A,B,C) VBCASTRDUPf(A,B,C)
#define MULT_2SPIN(ptr,pf) MULT_ADDSUB_2SPIN_LS(ptr,pf)
#undef KERNEL_DAG
template<>
void WilsonKernels<DomainWallVec5dImplF>::DiracOptAsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
template<>
void WilsonKernels<DomainWallVec5dImplF>::DiracOptAsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
#define KERNEL_DAG
template<>
void WilsonKernels<DomainWallVec5dImplF>::DiracOptAsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
template<>
void WilsonKernels<DomainWallVec5dImplF>::DiracOptAsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
#endif
template void WilsonKernels<WilsonImplF>::DiracOptAsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out);
template void WilsonKernels<WilsonImplD>::DiracOptAsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out);
template void WilsonKernels<GparityWilsonImplF>::DiracOptAsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out);
template void WilsonKernels<GparityWilsonImplD>::DiracOptAsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out);
template void WilsonKernels<DomainWallVec5dImplF>::DiracOptAsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out);
template void WilsonKernels<DomainWallVec5dImplD>::DiracOptAsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out);
}}
}
}