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

Merge pull request #23 from paboyle/develop

Browse Source 
Sync
This commit is contained in:
Christoph Lehner 2021-03-01 12:33:51 +01:00 committed by GitHub
commit 9c9566b9c9
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
51 changed files with 6081 additions and 1091 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

60
Grid/algorithms/CoarsenedMatrix.h
View File

@ -775,7 +775,26 @@ public:
for(int p=0;p<npoint;p++) AcceleratorViewContainer[p].ViewClose();
}
CoarsenedMatrix(GridCartesian &CoarseGrid, GridRedBlackCartesian &CoarseRBGrid, int hermitian_=0) :
CoarsenedMatrix(GridCartesian &CoarseGrid, int hermitian_=0) :
_grid(&CoarseGrid),
_cbgrid(new GridRedBlackCartesian(&CoarseGrid)),
geom(CoarseGrid._ndimension),
hermitian(hermitian_),
Stencil(&CoarseGrid,geom.npoint,Even,geom.directions,geom.displacements,0),
StencilEven(_cbgrid,geom.npoint,Even,geom.directions,geom.displacements,0),
StencilOdd(_cbgrid,geom.npoint,Odd,geom.directions,geom.displacements,0),
A(geom.npoint,&CoarseGrid),
Aeven(geom.npoint,_cbgrid),
Aodd(geom.npoint,_cbgrid),
AselfInv(&CoarseGrid),
AselfInvEven(_cbgrid),
AselfInvOdd(_cbgrid),
dag_factor(nbasis*nbasis)
{
fillFactor();
};
CoarsenedMatrix(GridCartesian &CoarseGrid, GridRedBlackCartesian &CoarseRBGrid, int hermitian_=0) :
_grid(&CoarseGrid),
_cbgrid(&CoarseRBGrid),
@ -817,6 +836,8 @@ public:
typedef Lattice<typename Fobj::tensor_reduced> FineComplexField;
typedef typename Fobj::scalar_type scalar_type;
std::cout << GridLogMessage<< "CoarsenMatrix "<< std::endl;
FineComplexField one(FineGrid); one=scalar_type(1.0,0.0);
FineComplexField zero(FineGrid); zero=scalar_type(0.0,0.0);
@ -847,11 +868,13 @@ public:
CoarseScalar InnerProd(Grid());
std::cout << GridLogMessage<< "CoarsenMatrix Orthog "<< std::endl;
// Orthogonalise the subblocks over the basis
blockOrthogonalise(InnerProd,Subspace.subspace);
// Compute the matrix elements of linop between this orthonormal
// set of vectors.
std::cout << GridLogMessage<< "CoarsenMatrix masks "<< std::endl;
int self_stencil=-1;
for(int p=0;p<geom.npoint;p++)
{
@ -890,7 +913,7 @@ public:
phi=Subspace.subspace[i];
// std::cout << GridLogMessage<< "CoarsenMatrix vector "<<i << std::endl;
std::cout << GridLogMessage<< "CoarsenMatrix vector "<<i << std::endl;
linop.OpDirAll(phi,Mphi_p);
linop.OpDiag (phi,Mphi_p[geom.npoint-1]);
@ -919,6 +942,18 @@ public:
autoView( A_self , A[self_stencil], AcceleratorWrite);
accelerator_for(ss, Grid()->oSites(), Fobj::Nsimd(),{ coalescedWrite(A_p[ss](j,i),oZProj_v(ss)); });
if ( hermitian && (disp==-1) ) {
for(int pp=0;pp<geom.npoint;pp++){// Find the opposite link and set <j|A|i> = <i|A|j>*
int dirp = geom.directions[pp];
int dispp = geom.displacements[pp];
if ( (dirp==dir) && (dispp==1) ){
auto sft = conjugate(Cshift(oZProj,dir,1));
autoView( sft_v , sft , AcceleratorWrite);
autoView( A_pp , A[pp], AcceleratorWrite);
accelerator_for(ss, Grid()->oSites(), Fobj::Nsimd(),{ coalescedWrite(A_pp[ss](i,j),sft_v(ss)); });
}
}
}
}
}
@ -957,33 +992,12 @@ public:
}
if(hermitian) {
std::cout << GridLogMessage << " ForceHermitian, new code "<<std::endl;
ForceHermitian();
}
InvertSelfStencilLink(); std::cout << GridLogMessage << "Coarse self link inverted" << std::endl;
FillHalfCbs(); std::cout << GridLogMessage << "Coarse half checkerboards filled" << std::endl;
}
void ForceHermitian(void) {
CoarseMatrix Diff (Grid());
for(int p=0;p<geom.npoint;p++){
int dir = geom.directions[p];
int disp = geom.displacements[p];
if(disp==-1) {
// Find the opposite link
for(int pp=0;pp<geom.npoint;pp++){
int dirp = geom.directions[pp];
int dispp = geom.displacements[pp];
if ( (dirp==dir) && (dispp==1) ){
// Diff = adj(Cshift(A[p],dir,1)) - A[pp];
// std::cout << GridLogMessage<<" Replacing stencil leg "<<pp<<" with leg "<<p<< " diff "<<norm2(Diff) <<std::endl;
A[pp] = adj(Cshift(A[p],dir,1));
}
}
}
}
}
void InvertSelfStencilLink() {
std::cout << GridLogDebug << "CoarsenedMatrix::InvertSelfStencilLink" << std::endl;
int localVolume = Grid()->lSites();

3
Grid/communicator/Communicator_base.h
View File

@ -1,4 +1,3 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
@ -108,6 +107,8 @@ public:
////////////////////////////////////////////////////////////
// Reduction
////////////////////////////////////////////////////////////
void GlobalMax(RealD &);
void GlobalMax(RealF &);
void GlobalSum(RealF &);
void GlobalSumVector(RealF *,int N);
void GlobalSum(RealD &);

10
Grid/communicator/Communicator_mpi3.cc
View File

@ -275,6 +275,16 @@ void CartesianCommunicator::GlobalXOR(uint64_t &u){
int ierr=MPI_Allreduce(MPI_IN_PLACE,&u,1,MPI_UINT64_T,MPI_BXOR,communicator);
assert(ierr==0);
}
void CartesianCommunicator::GlobalMax(float &f)
{
int ierr=MPI_Allreduce(MPI_IN_PLACE,&f,1,MPI_FLOAT,MPI_MAX,communicator);
assert(ierr==0);
}
void CartesianCommunicator::GlobalMax(double &d)
{
int ierr = MPI_Allreduce(MPI_IN_PLACE,&d,1,MPI_DOUBLE,MPI_MAX,communicator);
assert(ierr==0);
}
void CartesianCommunicator::GlobalSum(float &f){
int ierr=MPI_Allreduce(MPI_IN_PLACE,&f,1,MPI_FLOAT,MPI_SUM,communicator);
assert(ierr==0);

2
Grid/communicator/Communicator_none.cc
View File

@ -67,6 +67,8 @@ CartesianCommunicator::CartesianCommunicator(const Coordinate &processors)
CartesianCommunicator::~CartesianCommunicator(){}
void CartesianCommunicator::GlobalMax(float &){}
void CartesianCommunicator::GlobalMax(double &){}
void CartesianCommunicator::GlobalSum(float &){}
void CartesianCommunicator::GlobalSumVector(float *,int N){}
void CartesianCommunicator::GlobalSum(double &){}

54
Grid/lattice/Lattice_reduction.h
View File

@ -96,8 +96,34 @@ inline typename vobj::scalar_objectD sumD_cpu(const vobj *arg, Integer osites)
ssobj ret = ssum;
return ret;
}
/*
Threaded max, don't use for now
template<class Double>
inline Double max(const Double *arg, Integer osites)
{
// const int Nsimd = vobj::Nsimd();
const int nthread = GridThread::GetThreads();
std::vector<Double> maxarray(nthread);
thread_for(thr,nthread, {
int nwork, mywork, myoff;
nwork = osites;
GridThread::GetWork(nwork,thr,mywork,myoff);
Double max=arg[0];
for(int ss=myoff;ss<mywork+myoff; ss++){
if( arg[ss] > max ) max = arg[ss];
}
maxarray[thr]=max;
});
Double tmax=maxarray[0];
for(int i=0;i<nthread;i++){
if (maxarray[i]>tmax) tmax = maxarray[i];
}
return tmax;
}
*/
template<class vobj>
inline typename vobj::scalar_object sum(const vobj *arg, Integer osites)
{
@ -141,6 +167,32 @@ template<class vobj> inline RealD norm2(const Lattice<vobj> &arg){
return real(nrm);
}
//The global maximum of the site norm2
template<class vobj> inline RealD maxLocalNorm2(const Lattice<vobj> &arg)
{
typedef typename vobj::tensor_reduced vscalar; //iScalar<iScalar<.... <vPODtype> > >
typedef typename vscalar::scalar_object scalar; //iScalar<iScalar<.... <PODtype> > >
Lattice<vscalar> inner = localNorm2(arg);
auto grid = arg.Grid();
RealD max;
for(int l=0;l<grid->lSites();l++){
Coordinate coor;
scalar val;
RealD r;
grid->LocalIndexToLocalCoor(l,coor);
peekLocalSite(val,inner,coor);
r=real(TensorRemove(val));
if( (l==0) || (r>max)){
max=r;
}
}
grid->GlobalMax(max);
return max;
}
// Double inner product
template<class vobj>
inline ComplexD rankInnerProduct(const Lattice<vobj> &left,const Lattice<vobj> &right)

7
Grid/lattice/Lattice_view.h
View File

@ -67,9 +67,14 @@ public:
accelerator_inline const vobj & operator()(size_t i) const { return this->_odata[i]; }
#endif
#if 1
// accelerator_inline const vobj & operator[](size_t i) const { return this->_odata[i]; };
accelerator_inline vobj & operator[](size_t i) const { return this->_odata[i]; };
#else
accelerator_inline const vobj & operator[](size_t i) const { return this->_odata[i]; };
accelerator_inline vobj & operator[](size_t i) { return this->_odata[i]; };
#endif
accelerator_inline uint64_t begin(void) const { return 0;};
accelerator_inline uint64_t end(void) const { return this->_odata_size; };
accelerator_inline uint64_t size(void) const { return this->_odata_size; };

22
Grid/parallelIO/IldgIO.h
View File

@ -123,7 +123,7 @@ assert(GRID_FIELD_NORM_CALC(FieldNormMetaData_, n2ck) < 1.0e-5);
////////////////////////////////////////////////////////////
// Helper to fill out metadata
////////////////////////////////////////////////////////////
template<class vobj> void ScidacMetaData(Lattice<vobj> & field,
template<class vobj> void ScidacMetaData(Lattice<vobj> & field,
FieldMetaData &header,
scidacRecord & _scidacRecord,
scidacFile & _scidacFile)
@ -619,12 +619,12 @@ class IldgWriter : public ScidacWriter {
// Don't require scidac records EXCEPT checksum
// Use Grid MetaData object if present.
////////////////////////////////////////////////////////////////
template <class vsimd>
void writeConfiguration(Lattice<iLorentzColourMatrix<vsimd> > &Umu,int sequence,std::string LFN,std::string description)
template <class stats = PeriodicGaugeStatistics>
void writeConfiguration(Lattice<vLorentzColourMatrixD > &Umu,int sequence,std::string LFN,std::string description)
{
GridBase * grid = Umu.Grid();
typedef Lattice<iLorentzColourMatrix<vsimd> > GaugeField;
typedef iLorentzColourMatrix<vsimd> vobj;
typedef Lattice<vLorentzColourMatrixD> GaugeField;
typedef vLorentzColourMatrixD vobj;
typedef typename vobj::scalar_object sobj;
////////////////////////////////////////
@ -636,6 +636,9 @@ class IldgWriter : public ScidacWriter {
ScidacMetaData(Umu,header,_scidacRecord,_scidacFile);
stats Stats;
Stats(Umu,header);
std::string format = header.floating_point;
header.ensemble_id = description;
header.ensemble_label = description;
@ -705,10 +708,10 @@ class IldgReader : public GridLimeReader {
// Else use ILDG MetaData object if present.
// Else use SciDAC MetaData object if present.
////////////////////////////////////////////////////////////////
template <class vsimd>
void readConfiguration(Lattice<iLorentzColourMatrix<vsimd> > &Umu, FieldMetaData &FieldMetaData_) {
template <class stats = PeriodicGaugeStatistics>
void readConfiguration(Lattice<vLorentzColourMatrixD> &Umu, FieldMetaData &FieldMetaData_) {
typedef Lattice<iLorentzColourMatrix<vsimd> > GaugeField;
typedef Lattice<vLorentzColourMatrixD > GaugeField;
typedef typename GaugeField::vector_object vobj;
typedef typename vobj::scalar_object sobj;
@ -921,7 +924,8 @@ class IldgReader : public GridLimeReader {
if ( found_FieldMetaData || found_usqcdInfo ) {
FieldMetaData checker;
GaugeStatistics(Umu,checker);
stats Stats;
Stats(Umu,checker);
assert(fabs(checker.plaquette - FieldMetaData_.plaquette )<1.0e-5);
assert(fabs(checker.link_trace - FieldMetaData_.link_trace)<1.0e-5);
std::cout << GridLogMessage<<"Plaquette and link trace match " << std::endl;

34
Grid/parallelIO/MetaData.h
View File

@ -176,29 +176,18 @@ template<class vobj> inline void PrepareMetaData(Lattice<vobj> & field, FieldMet
GridMetaData(grid,header);
MachineCharacteristics(header);
}
inline void GaugeStatistics(Lattice<vLorentzColourMatrixF> & data,FieldMetaData &header)
template<class Impl>
class GaugeStatistics
{
// How to convert data precision etc...
header.link_trace=WilsonLoops<PeriodicGimplF>::linkTrace(data);
header.plaquette =WilsonLoops<PeriodicGimplF>::avgPlaquette(data);
}
inline void GaugeStatistics(Lattice<vLorentzColourMatrixD> & data,FieldMetaData &header)
{
// How to convert data precision etc...
header.link_trace=WilsonLoops<PeriodicGimplD>::linkTrace(data);
header.plaquette =WilsonLoops<PeriodicGimplD>::avgPlaquette(data);
}
template<> inline void PrepareMetaData<vLorentzColourMatrixF>(Lattice<vLorentzColourMatrixF> & field, FieldMetaData &header)
{
GridBase *grid = field.Grid();
std::string format = getFormatString<vLorentzColourMatrixF>();
header.floating_point = format;
header.checksum = 0x0; // Nersc checksum unused in ILDG, Scidac
GridMetaData(grid,header);
GaugeStatistics(field,header);
MachineCharacteristics(header);
}
public:
void operator()(Lattice<vLorentzColourMatrixD> & data,FieldMetaData &header)
{
header.link_trace=WilsonLoops<Impl>::linkTrace(data);
header.plaquette =WilsonLoops<Impl>::avgPlaquette(data);
}
};
typedef GaugeStatistics<PeriodicGimplD> PeriodicGaugeStatistics;
typedef GaugeStatistics<ConjugateGimplD> ConjugateGaugeStatistics;
template<> inline void PrepareMetaData<vLorentzColourMatrixD>(Lattice<vLorentzColourMatrixD> & field, FieldMetaData &header)
{
GridBase *grid = field.Grid();
@ -206,7 +195,6 @@ template<> inline void PrepareMetaData<vLorentzColourMatrixD>(Lattice<vLorentzCo
header.floating_point = format;
header.checksum = 0x0; // Nersc checksum unused in ILDG, Scidac
GridMetaData(grid,header);
GaugeStatistics(field,header);
MachineCharacteristics(header);
}

46
Grid/parallelIO/NerscIO.h
View File

@ -40,6 +40,8 @@ using namespace Grid;
class NerscIO : public BinaryIO {
public:
typedef Lattice<vLorentzColourMatrixD> GaugeField;
static inline void truncate(std::string file){
std::ofstream fout(file,std::ios::out);
}
@ -129,12 +131,12 @@ public:
// Now the meat: the object readers
/////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
template<class vsimd>
static inline void readConfiguration(Lattice<iLorentzColourMatrix<vsimd> > &Umu,
template<class GaugeStats=PeriodicGaugeStatistics>
static inline void readConfiguration(GaugeField &Umu,
FieldMetaData& header,
std::string file)
std::string file,
GaugeStats GaugeStatisticsCalculator=GaugeStats())
{
typedef Lattice<iLorentzColourMatrix<vsimd> > GaugeField;
GridBase *grid = Umu.Grid();
uint64_t offset = readHeader(file,Umu.Grid(),header);
@ -153,23 +155,23 @@ public:
// munger is a function of <floating point, Real, data_type>
if ( header.data_type == std::string("4D_SU3_GAUGE") ) {
if ( ieee32 || ieee32big ) {
BinaryIO::readLatticeObject<iLorentzColourMatrix<vsimd>, LorentzColour2x3F>
BinaryIO::readLatticeObject<vLorentzColourMatrixD, LorentzColour2x3F>
(Umu,file,Gauge3x2munger<LorentzColour2x3F,LorentzColourMatrix>(), offset,format,
nersc_csum,scidac_csuma,scidac_csumb);
}
if ( ieee64 || ieee64big ) {
BinaryIO::readLatticeObject<iLorentzColourMatrix<vsimd>, LorentzColour2x3D>
BinaryIO::readLatticeObject<vLorentzColourMatrixD, LorentzColour2x3D>
(Umu,file,Gauge3x2munger<LorentzColour2x3D,LorentzColourMatrix>(),offset,format,
nersc_csum,scidac_csuma,scidac_csumb);
}
} else if ( header.data_type == std::string("4D_SU3_GAUGE_3x3") ) {
if ( ieee32 || ieee32big ) {
BinaryIO::readLatticeObject<iLorentzColourMatrix<vsimd>,LorentzColourMatrixF>
BinaryIO::readLatticeObject<vLorentzColourMatrixD,LorentzColourMatrixF>
(Umu,file,GaugeSimpleMunger<LorentzColourMatrixF,LorentzColourMatrix>(),offset,format,
nersc_csum,scidac_csuma,scidac_csumb);
}
if ( ieee64 || ieee64big ) {
BinaryIO::readLatticeObject<iLorentzColourMatrix<vsimd>,LorentzColourMatrixD>
BinaryIO::readLatticeObject<vLorentzColourMatrixD,LorentzColourMatrixD>
(Umu,file,GaugeSimpleMunger<LorentzColourMatrixD,LorentzColourMatrix>(),offset,format,
nersc_csum,scidac_csuma,scidac_csumb);
}
@ -177,7 +179,7 @@ public:
assert(0);
}
GaugeStatistics(Umu,clone);
GaugeStats Stats; Stats(Umu,clone);
std::cout<<GridLogMessage <<"NERSC Configuration "<<file<<" checksum "<<std::hex<<nersc_csum<< std::dec
<<" header "<<std::hex<<header.checksum<<std::dec <<std::endl;
@ -203,15 +205,13 @@ public:
std::cout<<GridLogMessage <<"NERSC Configuration "<<file<< " and plaquette, link trace, and checksum agree"<<std::endl;
}
template<class vsimd>
static inline void writeConfiguration(Lattice<iLorentzColourMatrix<vsimd> > &Umu,
template<class GaugeStats=PeriodicGaugeStatistics>
static inline void writeConfiguration(Lattice<vLorentzColourMatrixD > &Umu,
std::string file,
int two_row,
int bits32)
{
typedef Lattice<iLorentzColourMatrix<vsimd> > GaugeField;
typedef iLorentzColourMatrix<vsimd> vobj;
typedef vLorentzColourMatrixD vobj;
typedef typename vobj::scalar_object sobj;
FieldMetaData header;
@ -229,7 +229,7 @@ public:
GridMetaData(grid,header);
assert(header.nd==4);
GaugeStatistics(Umu,header);
GaugeStats Stats; Stats(Umu,header);
MachineCharacteristics(header);
uint64_t offset;
@ -238,19 +238,19 @@ public:
header.floating_point = std::string("IEEE64BIG");
header.data_type = std::string("4D_SU3_GAUGE_3x3");
GaugeSimpleUnmunger<fobj3D,sobj> munge;
if ( grid->IsBoss() ) {
truncate(file);
offset = writeHeader(header,file);
}
grid->Broadcast(0,(void *)&offset,sizeof(offset));
if ( grid->IsBoss() ) {
truncate(file);
offset = writeHeader(header,file);
}
grid->Broadcast(0,(void *)&offset,sizeof(offset));
uint32_t nersc_csum,scidac_csuma,scidac_csumb;
BinaryIO::writeLatticeObject<vobj,fobj3D>(Umu,file,munge,offset,header.floating_point,
nersc_csum,scidac_csuma,scidac_csumb);
header.checksum = nersc_csum;
if ( grid->IsBoss() ) {
writeHeader(header,file);
}
if ( grid->IsBoss() ) {
writeHeader(header,file);
}
std::cout<<GridLogMessage <<"Written NERSC Configuration on "<< file << " checksum "
<<std::hex<<header.checksum

2
Grid/parallelIO/OpenQcdIO.h
View File

@ -154,7 +154,7 @@ public:
grid->Barrier(); timer.Stop();
std::cout << Grid::GridLogMessage << "OpenQcdIO::readConfiguration: redistribute overhead " << timer.Elapsed() << std::endl;
GaugeStatistics(Umu, clone);
PeriodicGaugeStatistics Stats; Stats(Umu, clone);
RealD plaq_diff = fabs(clone.plaquette - header.plaquette);

2
Grid/parallelIO/OpenQcdIOChromaReference.h
View File

@ -208,7 +208,7 @@ public:
FieldMetaData clone(header);
GaugeStatistics(Umu, clone);
PeriodicGaugeStatistics Stats; Stats(Umu, clone);
RealD plaq_diff = fabs(clone.plaquette - header.plaquette);

7
Grid/qcd/QCD.h
View File

@ -80,6 +80,13 @@ template<typename T> struct isSpinor {
template <typename T> using IfSpinor = Invoke<std::enable_if< isSpinor<T>::value,int> > ;
template <typename T> using IfNotSpinor = Invoke<std::enable_if<!isSpinor<T>::value,int> > ;
const int CoarseIndex = 4;
template<typename T> struct isCoarsened {
static constexpr bool value = (CoarseIndex<=T::TensorLevel);
};
template <typename T> using IfCoarsened = Invoke<std::enable_if< isCoarsened<T>::value,int> > ;
template <typename T> using IfNotCoarsened = Invoke<std::enable_if<!isCoarsened<T>::value,int> > ;
// ChrisK very keen to add extra space for Gparity doubling.
//
// Also add domain wall index, in a way where Wilson operator

14
Grid/qcd/action/fermion/MADWF.h
View File

@ -85,7 +85,7 @@ class MADWF
maxiter =_maxiter;
};
void operator() (const FermionFieldo &src4,FermionFieldo &sol5)
void operator() (const FermionFieldo &src,FermionFieldo &sol5)
{
std::cout << GridLogMessage<< " ************************************************" << std::endl;
std::cout << GridLogMessage<< " MADWF-like algorithm " << std::endl;
@ -114,8 +114,16 @@ class MADWF
///////////////////////////////////////
//Import source, include Dminus factors
///////////////////////////////////////
Mato.ImportPhysicalFermionSource(src4,b);
std::cout << GridLogMessage << " src4 " <<norm2(src4)<<std::endl;
GridBase *src_grid = src.Grid();
assert( (src_grid == Mato.GaugeGrid()) || (src_grid == Mato.FermionGrid()));
if ( src_grid == Mato.GaugeGrid() ) {
Mato.ImportPhysicalFermionSource(src,b);
} else {
b=src;
}
std::cout << GridLogMessage << " src " <<norm2(src)<<std::endl;
std::cout << GridLogMessage << " b " <<norm2(b)<<std::endl;
defect = b;

8
Grid/qcd/action/fermion/WilsonImpl.h
View File

@ -106,11 +106,15 @@ public:
const _SpinorField & phi,
int mu)
{
const int Nsimd = SiteHalfSpinor::Nsimd();
autoView( out_v, out, AcceleratorWrite);
autoView( phi_v, phi, AcceleratorRead);
autoView( Umu_v, Umu, AcceleratorRead);
accelerator_for(sss,out.Grid()->oSites(),1,{
multLink(out_v[sss],Umu_v[sss],phi_v[sss],mu);
typedef decltype(coalescedRead(out_v[0])) calcSpinor;
accelerator_for(sss,out.Grid()->oSites(),Nsimd,{
calcSpinor tmp;
multLink(tmp,Umu_v[sss],phi_v(sss),mu);
coalescedWrite(out_v[sss],tmp);
});
}

3
Grid/qcd/action/fermion/implementation/WilsonFermionImplementation.h
View File

@ -397,6 +397,7 @@ void WilsonFermion<Impl>::DhopDerivEO(GaugeField &mat, const FermionField &U, co
template <class Impl>
void WilsonFermion<Impl>::Dhop(const FermionField &in, FermionField &out, int dag)
{
DhopCalls+=2;
conformable(in.Grid(), _grid); // verifies full grid
conformable(in.Grid(), out.Grid());
@ -408,6 +409,7 @@ void WilsonFermion<Impl>::Dhop(const FermionField &in, FermionField &out, int da
template <class Impl>
void WilsonFermion<Impl>::DhopOE(const FermionField &in, FermionField &out, int dag)
{
DhopCalls++;
conformable(in.Grid(), _cbgrid); // verifies half grid
conformable(in.Grid(), out.Grid()); // drops the cb check
@ -420,6 +422,7 @@ void WilsonFermion<Impl>::DhopOE(const FermionField &in, FermionField &out, int
template <class Impl>
void WilsonFermion<Impl>::DhopEO(const FermionField &in, FermionField &out,int dag)
{
DhopCalls++;
conformable(in.Grid(), _cbgrid); // verifies half grid
conformable(in.Grid(), out.Grid()); // drops the cb check

38
Grid/qcd/action/gauge/Gauge.cc Normal file
View File

@ -0,0 +1,38 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/action/gauge/Gauge.cc
Copyright (C) 2020
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
Author: paboyle <paboyle@ph.ed.ac.uk>
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution
directory
*************************************************************************************/
/* END LEGAL */
#include <Grid/qcd/action/fermion/FermionCore.h>
NAMESPACE_BEGIN(Grid);
std::vector<int> ConjugateGaugeImplBase::_conjDirs;
NAMESPACE_END(Grid);

79
Grid/qcd/action/gauge/GaugeImplementations.h
View File

@ -59,14 +59,14 @@ public:
}
static inline GaugeLinkField
CovShiftIdentityBackward(const GaugeLinkField &Link, int mu) {
return Cshift(adj(Link), mu, -1);
return PeriodicBC::CovShiftIdentityBackward(Link, mu);
}
static inline GaugeLinkField
CovShiftIdentityForward(const GaugeLinkField &Link, int mu) {
return Link;
return PeriodicBC::CovShiftIdentityForward(Link,mu);
}
static inline GaugeLinkField ShiftStaple(const GaugeLinkField &Link, int mu) {
return Cshift(Link, mu, 1);
return PeriodicBC::ShiftStaple(Link,mu);
}
static inline bool isPeriodicGaugeField(void) { return true; }
@ -74,7 +74,13 @@ public:
// Composition with smeared link, bc's etc.. probably need multiple inheritance
// Variable precision "S" and variable Nc
template <class GimplTypes> class ConjugateGaugeImpl : public GimplTypes {
class ConjugateGaugeImplBase {
protected:
static std::vector<int> _conjDirs;
};
template <class GimplTypes> class ConjugateGaugeImpl : public GimplTypes, ConjugateGaugeImplBase {
private:
public:
INHERIT_GIMPL_TYPES(GimplTypes);
@ -84,47 +90,56 @@ public:
////////////////////////////////////////////////////////////////////////////////////////////////////////////
template <class covariant>
static Lattice<covariant> CovShiftForward(const GaugeLinkField &Link, int mu,
const Lattice<covariant> &field) {
return ConjugateBC::CovShiftForward(Link, mu, field);
const Lattice<covariant> &field)
{
assert(_conjDirs.size() == Nd);
if(_conjDirs[mu])
return ConjugateBC::CovShiftForward(Link, mu, field);
else
return PeriodicBC::CovShiftForward(Link, mu, field);
}
template <class covariant>
static Lattice<covariant> CovShiftBackward(const GaugeLinkField &Link, int mu,
const Lattice<covariant> &field) {
return ConjugateBC::CovShiftBackward(Link, mu, field);
const Lattice<covariant> &field)
{
assert(_conjDirs.size() == Nd);
if(_conjDirs[mu])
return ConjugateBC::CovShiftBackward(Link, mu, field);
else
return PeriodicBC::CovShiftBackward(Link, mu, field);
}
static inline GaugeLinkField
CovShiftIdentityBackward(const GaugeLinkField &Link, int mu) {
GridBase *grid = Link.Grid();
int Lmu = grid->GlobalDimensions()[mu] - 1;
Lattice<iScalar<vInteger>> coor(grid);
LatticeCoordinate(coor, mu);
GaugeLinkField tmp(grid);
tmp = adj(Link);
tmp = where(coor == Lmu, conjugate(tmp), tmp);
return Cshift(tmp, mu, -1); // moves towards positive mu
CovShiftIdentityBackward(const GaugeLinkField &Link, int mu)
{
assert(_conjDirs.size() == Nd);
if(_conjDirs[mu])
return ConjugateBC::CovShiftIdentityBackward(Link, mu);
else
return PeriodicBC::CovShiftIdentityBackward(Link, mu);
}
static inline GaugeLinkField
CovShiftIdentityForward(const GaugeLinkField &Link, int mu) {
return Link;
CovShiftIdentityForward(const GaugeLinkField &Link, int mu)
{
assert(_conjDirs.size() == Nd);
if(_conjDirs[mu])
return ConjugateBC::CovShiftIdentityForward(Link,mu);
else
return PeriodicBC::CovShiftIdentityForward(Link,mu);
}
static inline GaugeLinkField ShiftStaple(const GaugeLinkField &Link, int mu) {
GridBase *grid = Link.Grid();
int Lmu = grid->GlobalDimensions()[mu] - 1;
Lattice<iScalar<vInteger>> coor(grid);
LatticeCoordinate(coor, mu);
GaugeLinkField tmp(grid);
tmp = Cshift(Link, mu, 1);
tmp = where(coor == Lmu, conjugate(tmp), tmp);
return tmp;
static inline GaugeLinkField ShiftStaple(const GaugeLinkField &Link, int mu)
{
assert(_conjDirs.size() == Nd);
if(_conjDirs[mu])
return ConjugateBC::ShiftStaple(Link,mu);
else
return PeriodicBC::ShiftStaple(Link,mu);
}
static inline void setDirections(std::vector<int> &conjDirs) { _conjDirs=conjDirs; }
static inline std::vector<int> getDirections(void) { return _conjDirs; }
static inline bool isPeriodicGaugeField(void) { return false; }
};

3
Grid/qcd/hmc/checkpointers/BaseCheckpointer.h
View File

@ -74,7 +74,7 @@ public:
conf_file = os.str();
}
}
virtual ~BaseHmcCheckpointer(){};
void check_filename(const std::string &filename){
std::ifstream f(filename.c_str());
if(!f.good()){
@ -82,7 +82,6 @@ public:
abort();
};
}
virtual void initialize(const CheckpointerParameters &Params) = 0;
virtual void CheckpointRestore(int traj, typename Impl::Field &U,

5
Grid/qcd/hmc/checkpointers/ILDGCheckpointer.h
View File

@ -45,6 +45,7 @@ private:
public:
INHERIT_GIMPL_TYPES(Implementation);
typedef GaugeStatistics<Implementation> GaugeStats;
ILDGHmcCheckpointer(const CheckpointerParameters &Params_) { initialize(Params_); }
@ -78,7 +79,7 @@ public:
BinaryIO::writeRNG(sRNG, pRNG, rng, 0,nersc_csum,scidac_csuma,scidac_csumb);
IldgWriter _IldgWriter(grid->IsBoss());
_IldgWriter.open(config);
_IldgWriter.writeConfiguration(U, traj, config, config);
_IldgWriter.writeConfiguration<GaugeStats>(U, traj, config, config);
_IldgWriter.close();
std::cout << GridLogMessage << "Written ILDG Configuration on " << config
@ -105,7 +106,7 @@ public:
FieldMetaData header;
IldgReader _IldgReader;
_IldgReader.open(config);
_IldgReader.readConfiguration(U,header); // format from the header
_IldgReader.readConfiguration<GaugeStats>(U,header); // format from the header
_IldgReader.close();
std::cout << GridLogMessage << "Read ILDG Configuration from " << config

7
Grid/qcd/hmc/checkpointers/NerscCheckpointer.h
View File

@ -43,7 +43,8 @@ private:
public:
INHERIT_GIMPL_TYPES(Gimpl); // only for gauge configurations
typedef GaugeStatistics<Gimpl> GaugeStats;
NerscHmcCheckpointer(const CheckpointerParameters &Params_) { initialize(Params_); }
void initialize(const CheckpointerParameters &Params_) {
@ -60,7 +61,7 @@ public:
int precision32 = 1;
int tworow = 0;
NerscIO::writeRNGState(sRNG, pRNG, rng);
NerscIO::writeConfiguration(U, config, tworow, precision32);
NerscIO::writeConfiguration<GaugeStats>(U, config, tworow, precision32);
}
};
@ -74,7 +75,7 @@ public:
FieldMetaData header;
NerscIO::readRNGState(sRNG, pRNG, header, rng);
NerscIO::readConfiguration(U, header, config);
NerscIO::readConfiguration<GaugeStats>(U, header, config);
};
};

28
Grid/qcd/hmc/integrators/Integrator.h
View File

@ -33,6 +33,7 @@ directory
#define INTEGRATOR_INCLUDED
#include <memory>
#include "MomentumFilter.h"
NAMESPACE_BEGIN(Grid);
@ -78,8 +79,19 @@ protected:
RepresentationPolicy Representations;
IntegratorParameters Params;
//Filters allow the user to manipulate the conjugate momentum, for example to freeze links in DDHMC
//It is applied whenever the momentum is updated / refreshed
//The default filter does nothing
MomentumFilterBase<MomentaField> const* MomFilter;
const ActionSet<Field, RepresentationPolicy> as;
//Get a pointer to a shared static instance of the "do-nothing" momentum filter to serve as a default
static MomentumFilterBase<MomentaField> const* getDefaultMomFilter(){
static MomentumFilterNone<MomentaField> filter;
return &filter;
}
void update_P(Field& U, int level, double ep)
{
t_P[level] += ep;
@ -135,6 +147,8 @@ protected:
// Force from the other representations
as[level].apply(update_P_hireps, Representations, Mom, U, ep);
MomFilter->applyFilter(Mom);
}
void update_U(Field& U, double ep)
@ -174,11 +188,23 @@ public:
t_P.resize(levels, 0.0);
t_U = 0.0;
// initialization of smearer delegated outside of Integrator
//Default the momentum filter to "do-nothing"
MomFilter = getDefaultMomFilter();
};
virtual ~Integrator() {}
virtual std::string integrator_name() = 0;
//Set the momentum filter allowing for manipulation of the conjugate momentum
void setMomentumFilter(const MomentumFilterBase<MomentaField> &filter){
MomFilter = &filter;
}
//Access the conjugate momentum
const MomentaField & getMomentum() const{ return P; }
void print_parameters()
{
@ -249,6 +275,8 @@ public:
// Refresh the higher representation actions
as[level].apply(refresh_hireps, Representations, pRNG);
}
MomFilter->applyFilter(P);
}
// to be used by the actionlevel class to iterate

94
Grid/qcd/hmc/integrators/MomentumFilter.h Normal file
View File

@ -0,0 +1,94 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/hmc/integrators/MomentumFilter.h
Copyright (C) 2015
Author: Christopher Kelly <ckelly@bnl.gov>
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution
directory
*************************************************************************************/
/* END LEGAL */
//--------------------------------------------------------------------
#ifndef MOMENTUM_FILTER
#define MOMENTUM_FILTER
NAMESPACE_BEGIN(Grid);
//These filter objects allow the user to manipulate the conjugate momentum as part of the update / refresh
template<typename MomentaField>
struct MomentumFilterBase{
virtual void applyFilter(MomentaField &P) const;
};
//Do nothing
template<typename MomentaField>
struct MomentumFilterNone: public MomentumFilterBase<MomentaField>{
void applyFilter(MomentaField &P) const override{}
};
//Multiply each site/direction by a Lorentz vector complex number field
//Can be used to implement a mask, zeroing out sites
template<typename MomentaField>
struct MomentumFilterApplyPhase: public MomentumFilterBase<MomentaField>{
typedef typename MomentaField::vector_type vector_type; //SIMD-vectorized complex type
typedef typename MomentaField::scalar_type scalar_type; //scalar complex type
typedef iVector<iScalar<iScalar<vector_type> >, Nd > LorentzScalarType; //complex phase for each site/direction
typedef Lattice<LorentzScalarType> LatticeLorentzScalarType;
LatticeLorentzScalarType phase;
MomentumFilterApplyPhase(const LatticeLorentzScalarType _phase): phase(_phase){}
//Default to uniform field of (1,0)
MomentumFilterApplyPhase(GridBase* _grid): phase(_grid){
LorentzScalarType one;
for(int mu=0;mu<Nd;mu++)
one(mu)()() = scalar_type(1.);
phase = one;
}
void applyFilter(MomentaField &P) const override{
conformable(P,phase);
autoView( P_v , P, AcceleratorWrite);
autoView( phase_v , phase, AcceleratorRead);
accelerator_for(ss,P_v.size(),MomentaField::vector_type::Nsimd(),{
auto site_mom = P_v(ss);
auto site_phase = phase_v(ss);
for(int mu=0;mu<Nd;mu++)
site_mom(mu) = site_mom(mu) * site_phase(mu);
coalescedWrite(P_v[ss], site_mom);
});
}
};
NAMESPACE_END(Grid);
#endif

2
Grid/qcd/modules/Modules.h
View File

@ -99,7 +99,7 @@ public:
virtual Prod* getPtr() = 0;
// add a getReference?
virtual ~HMCModuleBase(){};
virtual void print_parameters(){}; // default to nothing
};

179
Grid/qcd/spin/TwoSpinor.h
View File

@ -128,7 +128,6 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void s
}
template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void spProjTm (iVector<vtype,Nhs> &hspin,const iVector<vtype,Ns> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
hspin(0)=fspin(0)-fspin(2);
hspin(1)=fspin(1)-fspin(3);
}
@ -138,40 +137,50 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void s
* 0 0 -1 0
* 0 0 0 -1
*/
template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void spProj5p (iVector<vtype,Nhs> &hspin,const iVector<vtype,Ns> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
hspin(0)=fspin(0);
hspin(1)=fspin(1);
}
template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void spProj5m (iVector<vtype,Nhs> &hspin,const iVector<vtype,Ns> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
hspin(0)=fspin(2);
hspin(1)=fspin(3);
}
// template<class vtype> accelerator_inline void fspProj5p (iVector<vtype,Ns> &rfspin,const iVector<vtype,Ns> &fspin)
template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void spProj5p (iVector<vtype,Ns> &rfspin,const iVector<vtype,Ns> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
rfspin(0)=fspin(0);
rfspin(1)=fspin(1);
rfspin(2)=Zero();
rfspin(3)=Zero();
}
// template<class vtype> accelerator_inline void fspProj5m (iVector<vtype,Ns> &rfspin,const iVector<vtype,Ns> &fspin)
template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void spProj5m (iVector<vtype,Ns> &rfspin,const iVector<vtype,Ns> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
rfspin(0)=Zero();
rfspin(1)=Zero();
rfspin(2)=fspin(2);
rfspin(3)=fspin(3);
}
template<class vtype,int N,IfCoarsened<iVector<vtype,N> > = 0> accelerator_inline void spProj5p (iVector<vtype,N> &rfspin,const iVector<vtype,N> &fspin)
{
const int hN = N>>1;
for(int s=0;s<hN;s++){
rfspin(s)=fspin(s);
rfspin(s+hN)=Zero();
}
}
template<class vtype,int N,IfCoarsened<iVector<vtype,N> > = 0> accelerator_inline void spProj5m (iVector<vtype,N> &rfspin,const iVector<vtype,N> &fspin)
{
const int hN = N>>1;
for(int s=0;s<hN;s++){
rfspin(s)=Zero();
rfspin(s+hN)=fspin(s+hN);
}
}
////////////////////////////////////////////////////////////////////////////////////////////////////////////////
// Reconstruction routines to move back again to four spin
////////////////////////////////////////////////////////////////////////////////////////////////////////////////
@ -183,7 +192,6 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void s
*/
template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void spReconXp (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
fspin(0)=hspin(0);
fspin(1)=hspin(1);
fspin(2)=timesMinusI(hspin(1));
@ -191,7 +199,6 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void s
}
template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void spReconXm (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
fspin(0)=hspin(0);
fspin(1)=hspin(1);
fspin(2)=timesI(hspin(1));
@ -199,7 +206,6 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void s
}
template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void accumReconXp (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
fspin(0)+=hspin(0);
fspin(1)+=hspin(1);
fspin(2)-=timesI(hspin(1));
@ -207,7 +213,6 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void a
}
template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void accumReconXm (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
fspin(0)+=hspin(0);
fspin(1)+=hspin(1);
fspin(2)+=timesI(hspin(1));
@ -221,7 +226,6 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void a
template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void spReconYp (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
fspin(0)=hspin(0);
fspin(1)=hspin(1);
fspin(2)= hspin(1);
@ -229,7 +233,6 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void s
}
template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void spReconYm (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
fspin(0)=hspin(0);
fspin(1)=hspin(1);
fspin(2)=-hspin(1);
@ -237,7 +240,6 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void s
}
template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void accumReconYp (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
fspin(0)+=hspin(0);
fspin(1)+=hspin(1);
fspin(2)+=hspin(1);
@ -245,7 +247,6 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void a
}
template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void accumReconYm (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
fspin(0)+=hspin(0);
fspin(1)+=hspin(1);
fspin(2)-=hspin(1);
@ -260,7 +261,6 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void a
*/
template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void spReconZp (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
fspin(0)=hspin(0);
fspin(1)=hspin(1);
fspin(2)=timesMinusI(hspin(0));
@ -268,7 +268,6 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void s
}
template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void spReconZm (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
fspin(0)=hspin(0);
fspin(1)=hspin(1);
fspin(2)= timesI(hspin(0));
@ -276,7 +275,6 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void s
}
template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void accumReconZp (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
fspin(0)+=hspin(0);
fspin(1)+=hspin(1);
fspin(2)-=timesI(hspin(0));
@ -284,7 +282,6 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void a
}
template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void accumReconZm (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
fspin(0)+=hspin(0);
fspin(1)+=hspin(1);
fspin(2)+=timesI(hspin(0));
@ -298,7 +295,6 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void a
*/
template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void spReconTp (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
fspin(0)=hspin(0);
fspin(1)=hspin(1);
fspin(2)=hspin(0);
@ -306,7 +302,6 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void s
}
template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void spReconTm (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
fspin(0)=hspin(0);
fspin(1)=hspin(1);
fspin(2)=-hspin(0);
@ -314,7 +309,6 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void s
}
template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void accumReconTp (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
fspin(0)+=hspin(0);
fspin(1)+=hspin(1);
fspin(2)+=hspin(0);
@ -322,7 +316,6 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void a
}
template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void accumReconTm (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
fspin(0)+=hspin(0);
fspin(1)+=hspin(1);
fspin(2)-=hspin(0);
@ -336,7 +329,6 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void a
*/
template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void spRecon5p (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
fspin(0)=hspin(0)+hspin(0); // add is lower latency than mul
fspin(1)=hspin(1)+hspin(1); // probably no measurable diffence though
fspin(2)=Zero();
@ -344,7 +336,6 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void s
}
template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void spRecon5m (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
fspin(0)=Zero();
fspin(1)=Zero();
fspin(2)=hspin(0)+hspin(0);
@ -352,7 +343,6 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void s
}
template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void accumRecon5p (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
fspin(0)+=hspin(0)+hspin(0);
fspin(1)+=hspin(1)+hspin(1);
}
@ -372,7 +362,6 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void a
//////////
template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spProjXp (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
for(int i=0;i<N;i++) {
spProjXp(hspin._internal[i],fspin._internal[i]);
}
@ -426,26 +415,21 @@ template<class rtype,class vtype,int N> accelerator_inline void accumReconXp (iM
}}
}
////////
// Xm
////////
template<class rtype,class vtype> accelerator_inline void spProjXm (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
spProjXm(hspin._internal,fspin._internal);
}
template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spProjXm (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
for(int i=0;i<N;i++) {
spProjXm(hspin._internal[i],fspin._internal[i]);
}
}
template<class rtype,class vtype,int N> accelerator_inline void spProjXm (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
spProjXm(hspin._internal[i][j],fspin._internal[i][j]);
@ -455,19 +439,16 @@ template<class rtype,class vtype,int N> accelerator_inline void spProjXm (iMatri
template<class rtype,class vtype> accelerator_inline void spReconXm (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
spReconXm(hspin._internal,fspin._internal);
}
template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spReconXm (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
for(int i=0;i<N;i++) {
spReconXm(hspin._internal[i],fspin._internal[i]);
}
}
template<class rtype,class vtype,int N> accelerator_inline void spReconXm (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
spReconXm(hspin._internal[i][j],fspin._internal[i][j]);
@ -476,45 +457,37 @@ template<class rtype,class vtype,int N> accelerator_inline void spReconXm (iMatr
template<class rtype,class vtype> accelerator_inline void accumReconXm (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
accumReconXm(hspin._internal,fspin._internal);
}
template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void accumReconXm (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
for(int i=0;i<N;i++) {
accumReconXm(hspin._internal[i],fspin._internal[i]);
}
}
template<class rtype,class vtype,int N> accelerator_inline void accumReconXm (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
accumReconXm(hspin._internal[i][j],fspin._internal[i][j]);
}}
}
////////
// Yp
////////
template<class rtype,class vtype> accelerator_inline void spProjYp (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
spProjYp(hspin._internal,fspin._internal);
}
template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spProjYp (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
for(int i=0;i<N;i++) {
spProjYp(hspin._internal[i],fspin._internal[i]);
}
}
template<class rtype,class vtype,int N> accelerator_inline void spProjYp (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
spProjYp(hspin._internal[i][j],fspin._internal[i][j]);
@ -524,19 +497,16 @@ template<class rtype,class vtype,int N> accelerator_inline void spProjYp (iMatri
template<class rtype,class vtype> accelerator_inline void spReconYp (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
spReconYp(hspin._internal,fspin._internal);
}
template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spReconYp (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
for(int i=0;i<N;i++) {
spReconYp(hspin._internal[i],fspin._internal[i]);
}
}
template<class rtype,class vtype,int N> accelerator_inline void spReconYp (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
spReconYp(hspin._internal[i][j],fspin._internal[i][j]);
@ -545,66 +515,55 @@ template<class rtype,class vtype,int N> accelerator_inline void spReconYp (iMatr
template<class rtype,class vtype> accelerator_inline void accumReconYp (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
accumReconYp(hspin._internal,fspin._internal);
}
template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void accumReconYp (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
for(int i=0;i<N;i++) {
accumReconYp(hspin._internal[i],fspin._internal[i]);
}
}
template<class rtype,class vtype,int N> accelerator_inline void accumReconYp (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
accumReconYp(hspin._internal[i][j],fspin._internal[i][j]);
}}
}
////////
// Ym
////////
template<class rtype,class vtype> accelerator_inline void spProjYm (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
spProjYm(hspin._internal,fspin._internal);
}
template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spProjYm (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
for(int i=0;i<N;i++) {
spProjYm(hspin._internal[i],fspin._internal[i]);
}
}
template<class rtype,class vtype,int N> accelerator_inline void spProjYm (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
spProjYm(hspin._internal[i][j],fspin._internal[i][j]);
}}
}
template<class rtype,class vtype> accelerator_inline void spReconYm (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
spReconYm(hspin._internal,fspin._internal);
}
template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spReconYm (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,const iVector<vtype,N> >::type *temp;
for(int i=0;i<N;i++) {
spReconYm(hspin._internal[i],fspin._internal[i]);
}
}
template<class rtype,class vtype,int N> accelerator_inline void spReconYm (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
spReconYm(hspin._internal[i][j],fspin._internal[i][j]);
@ -613,19 +572,16 @@ template<class rtype,class vtype,int N> accelerator_inline void spReconYm (iMatr
template<class rtype,class vtype> accelerator_inline void accumReconYm (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
accumReconYm(hspin._internal,fspin._internal);
}
template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void accumReconYm (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
for(int i=0;i<N;i++) {
accumReconYm(hspin._internal[i],fspin._internal[i]);
}
}
template<class rtype,class vtype,int N> accelerator_inline void accumReconYm (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
accumReconYm(hspin._internal[i][j],fspin._internal[i][j]);
@ -638,66 +594,57 @@ template<class rtype,class vtype,int N> accelerator_inline void accumReconYm (iM
////////
template<class rtype,class vtype> accelerator_inline void spProjZp (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
spProjZp(hspin._internal,fspin._internal);
}
template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spProjZp (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
for(int i=0;i<N;i++) {
spProjZp(hspin._internal[i],fspin._internal[i]);
}
}
template<class rtype,class vtype,int N> accelerator_inline void spProjZp (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
spProjZp(hspin._internal[i][j],fspin._internal[i][j]);
}}
}}
}
template<class rtype,class vtype> accelerator_inline void spReconZp (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
spReconZp(hspin._internal,fspin._internal);
}
template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spReconZp (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
for(int i=0;i<N;i++) {
spReconZp(hspin._internal[i],fspin._internal[i]);
}
}
template<class rtype,class vtype,int N> accelerator_inline void spReconZp (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
spReconZp(hspin._internal[i][j],fspin._internal[i][j]);
}}
}}
}
template<class rtype,class vtype> accelerator_inline void accumReconZp (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
accumReconZp(hspin._internal,fspin._internal);
}
template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void accumReconZp (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
for(int i=0;i<N;i++) {
accumReconZp(hspin._internal[i],fspin._internal[i]);
}
}
template<class rtype,class vtype,int N> accelerator_inline void accumReconZp (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
accumReconZp(hspin._internal[i][j],fspin._internal[i][j]);
}}
}}
}
@ -706,62 +653,53 @@ template<class rtype,class vtype,int N> accelerator_inline void accumReconZp (iM
////////
template<class rtype,class vtype> accelerator_inline void spProjZm (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
spProjZm(hspin._internal,fspin._internal);
}
template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spProjZm (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
for(int i=0;i<N;i++) {
spProjZm(hspin._internal[i],fspin._internal[i]);
}
}
template<class rtype,class vtype,int N> accelerator_inline void spProjZm (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
spProjZm(hspin._internal[i][j],fspin._internal[i][j]);
}}
}}
}
template<class rtype,class vtype> accelerator_inline void spReconZm (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
spReconZm(hspin._internal,fspin._internal);
}
template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spReconZm (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
for(int i=0;i<N;i++) {
spReconZm(hspin._internal[i],fspin._internal[i]);
}
}
template<class rtype,class vtype,int N> accelerator_inline void spReconZm (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
spReconZm(hspin._internal[i][j],fspin._internal[i][j]);
}}
}}
}
template<class rtype,class vtype> accelerator_inline void accumReconZm (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
accumReconZm(hspin._internal,fspin._internal);
}
template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void accumReconZm (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
for(int i=0;i<N;i++) {
accumReconZm(hspin._internal[i],fspin._internal[i]);
}
}
template<class rtype,class vtype,int N> accelerator_inline void accumReconZm (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
accumReconZm(hspin._internal[i][j],fspin._internal[i][j]);
@ -774,41 +712,35 @@ template<class rtype,class vtype,int N> accelerator_inline void accumReconZm (iM
////////
template<class rtype,class vtype> accelerator_inline void spProjTp (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
spProjTp(hspin._internal,fspin._internal);
}
template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spProjTp (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
for(int i=0;i<N;i++) {
spProjTp(hspin._internal[i],fspin._internal[i]);
}
}
template<class rtype,class vtype,int N> accelerator_inline void spProjTp (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
spProjTp(hspin._internal[i][j],fspin._internal[i][j]);
}}
}}
}
template<class rtype,class vtype> accelerator_inline void spReconTp (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
spReconTp(hspin._internal,fspin._internal);
}
template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spReconTp (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
for(int i=0;i<N;i++) {
spReconTp(hspin._internal[i],fspin._internal[i]);
}
}
template<class rtype,class vtype,int N> accelerator_inline void spReconTp (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
spReconTp(hspin._internal[i][j],fspin._internal[i][j]);
@ -817,44 +749,37 @@ template<class rtype,class vtype,int N> accelerator_inline void spReconTp (iMatr
template<class rtype,class vtype> accelerator_inline void accumReconTp (iScalar<rtype> &hspin, iScalar<vtype> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
accumReconTp(hspin._internal,fspin._internal);
}
template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void accumReconTp (iVector<rtype,N> &hspin, const iVector<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
for(int i=0;i<N;i++) {
accumReconTp(hspin._internal[i],fspin._internal[i]);
}
}
template<class rtype,class vtype,int N> accelerator_inline void accumReconTp (iMatrix<rtype,N> &hspin, const iMatrix<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
accumReconTp(hspin._internal[i][j],fspin._internal[i][j]);
}}
}
////////
// Tm
////////
template<class rtype,class vtype> accelerator_inline void spProjTm (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
spProjTm(hspin._internal,fspin._internal);
}
template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spProjTm (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
for(int i=0;i<N;i++) {
spProjTm(hspin._internal[i],fspin._internal[i]);
}
}
template<class rtype,class vtype,int N> accelerator_inline void spProjTm (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
spProjTm(hspin._internal[i][j],fspin._internal[i][j]);
@ -864,19 +789,16 @@ template<class rtype,class vtype,int N> accelerator_inline void spProjTm (iMatri
template<class rtype,class vtype> accelerator_inline void spReconTm (iScalar<rtype> &hspin, const iScalar<vtype> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
spReconTm(hspin._internal,fspin._internal);
}
template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spReconTm (iVector<rtype,N> &hspin, const iVector<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
for(int i=0;i<N;i++) {
spReconTm(hspin._internal[i],fspin._internal[i]);
}
}
template<class rtype,class vtype,int N> accelerator_inline void spReconTm (iMatrix<rtype,N> &hspin, const iMatrix<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
spReconTm(hspin._internal[i][j],fspin._internal[i][j]);
@ -885,44 +807,37 @@ template<class rtype,class vtype,int N> accelerator_inline void spReconTm (iMatr
template<class rtype,class vtype> accelerator_inline void accumReconTm (iScalar<rtype> &hspin, const iScalar<vtype> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
accumReconTm(hspin._internal,fspin._internal);
}
template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void accumReconTm (iVector<rtype,N> &hspin, const iVector<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
for(int i=0;i<N;i++) {
accumReconTm(hspin._internal[i],fspin._internal[i]);
}
}
template<class rtype,class vtype,int N> accelerator_inline void accumReconTm (iMatrix<rtype,N> &hspin, const iMatrix<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
accumReconTm(hspin._internal[i][j],fspin._internal[i][j]);
}}
}
////////
// 5p
////////
template<class rtype,class vtype> accelerator_inline void spProj5p (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
template<class rtype,class vtype,IfNotCoarsened<iScalar<vtype> > = 0> accelerator_inline void spProj5p (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
spProj5p(hspin._internal,fspin._internal);
}
template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spProj5p (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
for(int i=0;i<N;i++) {
spProj5p(hspin._internal[i],fspin._internal[i]);
}
}
template<class rtype,class vtype,int N> accelerator_inline void spProj5p (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
template<class rtype,class vtype,int N,IfNotCoarsened<iScalar<vtype> > = 0> accelerator_inline void spProj5p (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
spProj5p(hspin._internal[i][j],fspin._internal[i][j]);
@ -931,19 +846,16 @@ template<class rtype,class vtype,int N> accelerator_inline void spProj5p (iMatri
template<class rtype,class vtype> accelerator_inline void spRecon5p (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
spRecon5p(hspin._internal,fspin._internal);
}
template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spRecon5p (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
for(int i=0;i<N;i++) {
spRecon5p(hspin._internal[i],fspin._internal[i]);
}
}
template<class rtype,class vtype,int N> accelerator_inline void spRecon5p (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
spRecon5p(hspin._internal[i][j],fspin._internal[i][j]);
@ -952,19 +864,16 @@ template<class rtype,class vtype,int N> accelerator_inline void spRecon5p (iMatr
template<class rtype,class vtype> accelerator_inline void accumRecon5p (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
accumRecon5p(hspin._internal,fspin._internal);
}
template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void accumRecon5p (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
for(int i=0;i<N;i++) {
accumRecon5p(hspin._internal[i],fspin._internal[i]);
}
}
template<class rtype,class vtype,int N> accelerator_inline void accumRecon5p (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
accumRecon5p(hspin._internal[i][j],fspin._internal[i][j]);
@ -972,24 +881,18 @@ template<class rtype,class vtype,int N> accelerator_inline void accumRecon5p (iM
}
// four spinor projectors for chiral proj
// template<class vtype> accelerator_inline void fspProj5p (iScalar<vtype> &hspin,const iScalar<vtype> &fspin)
template<class vtype> accelerator_inline void spProj5p (iScalar<vtype> &hspin,const iScalar<vtype> &fspin)
template<class vtype,IfNotCoarsened<iScalar<vtype> > = 0> accelerator_inline void spProj5p (iScalar<vtype> &hspin,const iScalar<vtype> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
spProj5p(hspin._internal,fspin._internal);
}
// template<class vtype,int N> accelerator_inline void fspProj5p (iVector<vtype,N> &hspin,iVector<vtype,N> &fspin)
template<class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spProj5p (iVector<vtype,N> &hspin,const iVector<vtype,N> &fspin)
template<class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0,IfNotCoarsened<iScalar<vtype> > = 0> accelerator_inline void spProj5p (iVector<vtype,N> &hspin,const iVector<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
for(int i=0;i<N;i++) {
spProj5p(hspin._internal[i],fspin._internal[i]);
}
}
// template<class vtype,int N> accelerator_inline void fspProj5p (iMatrix<vtype,N> &hspin,iMatrix<vtype,N> &fspin)
template<class vtype,int N> accelerator_inline void spProj5p (iMatrix<vtype,N> &hspin,const iMatrix<vtype,N> &fspin)
template<class vtype,int N,IfNotCoarsened<iScalar<vtype> > = 0> accelerator_inline void spProj5p (iMatrix<vtype,N> &hspin,const iMatrix<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
spProj5p(hspin._internal[i][j],fspin._internal[i][j]);
@ -1001,17 +904,17 @@ template<class vtype,int N> accelerator_inline void spProj5p (iMatrix<vtype,N> &
// 5m
////////
template<class rtype,class vtype> accelerator_inline void spProj5m (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
template<class rtype,class vtype,IfNotCoarsened<iScalar<vtype> > = 0> accelerator_inline void spProj5m (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
{
spProj5m(hspin._internal,fspin._internal);
}
template<class rtype,class vtype,int N,IfNotSpinor<iVector<rtype,N> > = 0> accelerator_inline void spProj5m (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
template<class rtype,class vtype,int N,IfNotSpinor<iVector<rtype,N> > = 0,IfNotCoarsened<iScalar<vtype> > = 0> accelerator_inline void spProj5m (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
{
for(int i=0;i<N;i++) {
spProj5m(hspin._internal[i],fspin._internal[i]);
}
}
template<class rtype,class vtype,int N> accelerator_inline void spProj5m (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
template<class rtype,class vtype,int N,IfNotCoarsened<iScalar<vtype> > = 0> accelerator_inline void spProj5m (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
{
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
@ -1021,40 +924,34 @@ template<class rtype,class vtype,int N> accelerator_inline void spProj5m (iMatri
template<class rtype,class vtype> accelerator_inline void spRecon5m (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
spRecon5m(hspin._internal,fspin._internal);
}
template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spRecon5m (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
for(int i=0;i<N;i++) {
spRecon5m(hspin._internal[i],fspin._internal[i]);
}
}
template<class rtype,class vtype,int N> accelerator_inline void spRecon5m (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
spRecon5m(hspin._internal[i][j],fspin._internal[i][j]);
}}
}}
}
template<class rtype,class vtype> accelerator_inline void accumRecon5m (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
accumRecon5m(hspin._internal,fspin._internal);
}
template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void accumRecon5m (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
for(int i=0;i<N;i++) {
accumRecon5m(hspin._internal[i],fspin._internal[i]);
}
}
template<class rtype,class vtype,int N> accelerator_inline void accumRecon5m (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
accumRecon5m(hspin._internal[i][j],fspin._internal[i][j]);
@ -1063,24 +960,18 @@ template<class rtype,class vtype,int N> accelerator_inline void accumRecon5m (iM
// four spinor projectors for chiral proj
// template<class vtype> accelerator_inline void fspProj5m (iScalar<vtype> &hspin,const iScalar<vtype> &fspin)
template<class vtype> accelerator_inline void spProj5m (iScalar<vtype> &hspin,const iScalar<vtype> &fspin)
template<class vtype,IfNotCoarsened<iScalar<vtype> > = 0> accelerator_inline void spProj5m (iScalar<vtype> &hspin,const iScalar<vtype> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
spProj5m(hspin._internal,fspin._internal);
}
// template<class vtype,int N> accelerator_inline void fspProj5m (iVector<vtype,N> &hspin,iVector<vtype,N> &fspin)
template<class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spProj5m (iVector<vtype,N> &hspin,const iVector<vtype,N> &fspin)
template<class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0,IfNotCoarsened<iScalar<vtype> > = 0> accelerator_inline void spProj5m (iVector<vtype,N> &hspin,const iVector<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
for(int i=0;i<N;i++) {
spProj5m(hspin._internal[i],fspin._internal[i]);
}
}
// template<class vtype,int N> accelerator_inline void fspProj5m (iMatrix<vtype,N> &hspin,iMatrix<vtype,N> &fspin)
template<class vtype,int N> accelerator_inline void spProj5m (iMatrix<vtype,N> &hspin,const iMatrix<vtype,N> &fspin)
template<class vtype,int N,IfNotCoarsened<iScalar<vtype> > = 0> accelerator_inline void spProj5m (iMatrix<vtype,N> &hspin,const iMatrix<vtype,N> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
spProj5m(hspin._internal[i][j],fspin._internal[i][j]);

1620
Grid/qcd/utils/BaryonUtils.h
View File

File diff suppressed because it is too large Load Diff

51
Grid/qcd/utils/CovariantCshift.h
View File

@ -53,6 +53,24 @@ namespace PeriodicBC {
return Cshift(tmp,mu,-1);// moves towards positive mu
}
template<class gauge> Lattice<gauge>
CovShiftIdentityBackward(const Lattice<gauge> &Link, int mu)
{
return Cshift(adj(Link), mu, -1);
}
template<class gauge> Lattice<gauge>
CovShiftIdentityForward(const Lattice<gauge> &Link, int mu)
{
return Link;
}
template<class gauge> Lattice<gauge>
ShiftStaple(const Lattice<gauge> &Link, int mu)
{
return Cshift(Link, mu, 1);
}
template<class gauge,class Expr,typename std::enable_if<is_lattice_expr<Expr>::value,void>::type * = nullptr>
auto CovShiftForward(const Lattice<gauge> &Link,
int mu,
@ -70,6 +88,7 @@ namespace PeriodicBC {
return CovShiftBackward(Link,mu,arg);
}
}
@ -139,6 +158,38 @@ namespace ConjugateBC {
// std::cout<<"Gparity::CovCshiftBackward mu="<<mu<<std::endl;
return Cshift(tmp,mu,-1);// moves towards positive mu
}
template<class gauge> Lattice<gauge>
CovShiftIdentityBackward(const Lattice<gauge> &Link, int mu) {
GridBase *grid = Link.Grid();
int Lmu = grid->GlobalDimensions()[mu] - 1;
Lattice<iScalar<vInteger>> coor(grid);
LatticeCoordinate(coor, mu);
Lattice<gauge> tmp(grid);
tmp = adj(Link);
tmp = where(coor == Lmu, conjugate(tmp), tmp);
return Cshift(tmp, mu, -1); // moves towards positive mu
}
template<class gauge> Lattice<gauge>
CovShiftIdentityForward(const Lattice<gauge> &Link, int mu) {
return Link;
}
template<class gauge> Lattice<gauge>
ShiftStaple(const Lattice<gauge> &Link, int mu)
{
GridBase *grid = Link.Grid();
int Lmu = grid->GlobalDimensions()[mu] - 1;
Lattice<iScalar<vInteger>> coor(grid);
LatticeCoordinate(coor, mu);
Lattice<gauge> tmp(grid);
tmp = Cshift(Link, mu, 1);
tmp = where(coor == Lmu, conjugate(tmp), tmp);
return tmp;
}
template<class gauge,class Expr,typename std::enable_if<is_lattice_expr<Expr>::value,void>::type * = nullptr>
auto CovShiftForward(const Lattice<gauge> &Link,

28
Grid/qcd/utils/LinalgUtils.h
View File

@ -154,8 +154,8 @@ void axpby_ssp_pminus(Lattice<vobj> &z,Coeff a,const Lattice<vobj> &x,Coeff b,co
accelerator_for(sss,nloop,vobj::Nsimd(),{
uint64_t ss = sss*Ls;
decltype(coalescedRead(y_v[ss+sp])) tmp;
spProj5m(tmp,y_v(ss+sp));
tmp = a*x_v(ss+s)+b*tmp;
spProj5m(tmp,y_v(ss+sp));
tmp = a*x_v(ss+s)+b*tmp;
coalescedWrite(z_v[ss+s],tmp);
});
}
@ -188,7 +188,6 @@ void G5R5(Lattice<vobj> &z,const Lattice<vobj> &x)
z.Checkerboard() = x.Checkerboard();
conformable(x,z);
int Ls = grid->_rdimensions[0];
Gamma G5(Gamma::Algebra::Gamma5);
autoView( x_v, x, AcceleratorRead);
autoView( z_v, z, AcceleratorWrite);
uint64_t nloop = grid->oSites()/Ls;
@ -196,7 +195,13 @@ void G5R5(Lattice<vobj> &z,const Lattice<vobj> &x)
uint64_t ss = sss*Ls;
for(int s=0;s<Ls;s++){
int sp = Ls-1-s;
coalescedWrite(z_v[ss+sp],G5*x_v(ss+s));
auto tmp = x_v(ss+s);
decltype(tmp) tmp_p;
decltype(tmp) tmp_m;
spProj5p(tmp_p,tmp);
spProj5m(tmp_m,tmp);
// Use of spProj5m, 5p captures the coarse space too
coalescedWrite(z_v[ss+sp],tmp_p - tmp_m);
}
});
}
@ -208,10 +213,20 @@ void G5C(Lattice<vobj> &z, const Lattice<vobj> &x)
z.Checkerboard() = x.Checkerboard();
conformable(x, z);
Gamma G5(Gamma::Algebra::Gamma5);
z = G5 * x;
autoView( x_v, x, AcceleratorRead);
autoView( z_v, z, AcceleratorWrite);
uint64_t nloop = grid->oSites();
accelerator_for(ss,nloop,vobj::Nsimd(),{
auto tmp = x_v(ss);
decltype(tmp) tmp_p;
decltype(tmp) tmp_m;
spProj5p(tmp_p,tmp);
spProj5m(tmp_m,tmp);
coalescedWrite(z_v[ss],tmp_p - tmp_m);
});
}
/*
template<class CComplex, int nbasis>
void G5C(Lattice<iVector<CComplex, nbasis>> &z, const Lattice<iVector<CComplex, nbasis>> &x)
{
@ -234,6 +249,7 @@ void G5C(Lattice<iVector<CComplex, nbasis>> &z, const Lattice<iVector<CComplex,
}
});
}
*/
NAMESPACE_END(Grid);

23
Grid/simd/Grid_gpu_vec.h
View File

@ -60,11 +60,25 @@ template<class pair>
class GpuComplex {
public:
pair z;
typedef decltype(z.x) real;
typedef decltype(z.x) Real;
public:
accelerator_inline GpuComplex() = default;
accelerator_inline GpuComplex(real re,real im) { z.x=re; z.y=im; };
accelerator_inline GpuComplex(Real re,Real im) { z.x=re; z.y=im; };
accelerator_inline GpuComplex(const GpuComplex &zz) { z = zz.z;};
accelerator_inline Real real(void) const { return z.x; };
accelerator_inline Real imag(void) const { return z.y; };
accelerator_inline GpuComplex &operator*=(const GpuComplex &r) {
*this = (*this) * r;
return *this;
}
accelerator_inline GpuComplex &operator+=(const GpuComplex &r) {
*this = (*this) + r;
return *this;
}
accelerator_inline GpuComplex &operator-=(const GpuComplex &r) {
*this = (*this) - r;
return *this;
}
friend accelerator_inline GpuComplex operator+(const GpuComplex &lhs,const GpuComplex &rhs) {
GpuComplex r ;
r.z.x = lhs.z.x + rhs.z.x;
@ -157,6 +171,11 @@ typedef GpuVector<NSIMD_RealD, double > GpuVectorRD;
typedef GpuVector<NSIMD_ComplexD, GpuComplexD > GpuVectorCD;
typedef GpuVector<NSIMD_Integer, Integer > GpuVectorI;
accelerator_inline GpuComplexF timesI(const GpuComplexF &r) { return(GpuComplexF(-r.imag(),r.real()));}
accelerator_inline GpuComplexD timesI(const GpuComplexD &r) { return(GpuComplexD(-r.imag(),r.real()));}
accelerator_inline GpuComplexF timesMinusI(const GpuComplexF &r){ return(GpuComplexF(r.imag(),-r.real()));}
accelerator_inline GpuComplexD timesMinusI(const GpuComplexD &r){ return(GpuComplexD(r.imag(),-r.real()));}
accelerator_inline float half2float(half h)
{
float f;

12
Grid/simd/Simd.h
View File

@ -148,10 +148,14 @@ accelerator_inline void sub (ComplexF * __restrict__ y,const ComplexF * __restri
accelerator_inline void add (ComplexF * __restrict__ y,const ComplexF * __restrict__ l,const ComplexF *__restrict__ r){ *y = (*l) + (*r); }
//conjugate already supported for complex
accelerator_inline ComplexF timesI(const ComplexF &r) { return(r*ComplexF(0.0,1.0));}
accelerator_inline ComplexD timesI(const ComplexD &r) { return(r*ComplexD(0.0,1.0));}
accelerator_inline ComplexF timesMinusI(const ComplexF &r){ return(r*ComplexF(0.0,-1.0));}
accelerator_inline ComplexD timesMinusI(const ComplexD &r){ return(r*ComplexD(0.0,-1.0));}
accelerator_inline ComplexF timesI(const ComplexF &r) { return(ComplexF(-r.imag(),r.real()));}
accelerator_inline ComplexD timesI(const ComplexD &r) { return(ComplexD(-r.imag(),r.real()));}
accelerator_inline ComplexF timesMinusI(const ComplexF &r){ return(ComplexF(r.imag(),-r.real()));}
accelerator_inline ComplexD timesMinusI(const ComplexD &r){ return(ComplexD(r.imag(),-r.real()));}
//accelerator_inline ComplexF timesI(const ComplexF &r) { return(r*ComplexF(0.0,1.0));}
//accelerator_inline ComplexD timesI(const ComplexD &r) { return(r*ComplexD(0.0,1.0));}
//accelerator_inline ComplexF timesMinusI(const ComplexF &r){ return(r*ComplexF(0.0,-1.0));}
//accelerator_inline ComplexD timesMinusI(const ComplexD &r){ return(r*ComplexD(0.0,-1.0));}
// define projections to real and imaginay parts
accelerator_inline ComplexF projReal(const ComplexF &r){return( ComplexF(r.real(), 0.0));}

62
Grid/tensors/Tensor_SIMT.h
View File

@ -64,6 +64,68 @@ void coalescedWriteNonTemporal(vobj & __restrict__ vec,const vobj & __restrict__
}
#else
#ifndef GRID_SYCL
// Use the scalar as our own complex on GPU
template<class vsimd,IfSimd<vsimd> = 0> accelerator_inline
typename vsimd::scalar_type
coalescedRead(const vsimd & __restrict__ vec,int lane=acceleratorSIMTlane(vsimd::Nsimd()))
{
typedef typename vsimd::scalar_type S;
S * __restrict__ p=(S *)&vec;
return p[lane];
}
template<int ptype,class vsimd,IfSimd<vsimd> = 0> accelerator_inline
typename vsimd::scalar_type
coalescedReadPermute(const vsimd & __restrict__ vec,int doperm,int lane=acceleratorSIMTlane(vsimd::Nsimd()))
{
typedef typename vsimd::scalar_type S;
S * __restrict__ p=(S *)&vec;
int mask = vsimd::Nsimd() >> (ptype + 1);
int plane= doperm ? lane ^ mask : lane;
return p[plane];
}
template<class vsimd,IfSimd<vsimd> = 0> accelerator_inline
void coalescedWrite(vsimd & __restrict__ vec,
const typename vsimd::scalar_type & __restrict__ extracted,
int lane=acceleratorSIMTlane(vsimd::Nsimd()))
{
typedef typename vsimd::scalar_type S;
S * __restrict__ p=(S *)&vec;
p[lane]=extracted;
}
#else
template<class vsimd,IfSimd<vsimd> = 0> accelerator_inline
typename vsimd::vector_type::datum
coalescedRead(const vsimd & __restrict__ vec,int lane=acceleratorSIMTlane(vsimd::Nsimd()))
{
typedef typename vsimd::vector_type::datum S;
S * __restrict__ p=(S *)&vec;
return p[lane];
}
template<int ptype,class vsimd,IfSimd<vsimd> = 0> accelerator_inline
typename vsimd::vector_type::datum
coalescedReadPermute(const vsimd & __restrict__ vec,int doperm,int lane=acceleratorSIMTlane(vsimd::Nsimd()))
{
typedef typename vsimd::vector_type::datum S;
S * __restrict__ p=(S *)&vec;
int mask = vsimd::Nsimd() >> (ptype + 1);
int plane= doperm ? lane ^ mask : lane;
return p[plane];
}
template<class vsimd,IfSimd<vsimd> = 0> accelerator_inline
void coalescedWrite(vsimd & __restrict__ vec,
const typename vsimd::vector_type::datum & __restrict__ extracted,
int lane=acceleratorSIMTlane(vsimd::Nsimd()))
{
typedef typename vsimd::vector_type::datum S;
S * __restrict__ p=(S *)&vec;
p[lane]=extracted;
}
#endif
//////////////////////////////////////////
// Extract and insert slices on the GPU
//////////////////////////////////////////

14
Grid/tensors/Tensor_Ta.h
View File

@ -117,7 +117,19 @@ accelerator_inline iMatrix<vtype,N> ProjectOnGroup(const iMatrix<vtype,N> &arg)
ret._internal[b][c] -= pr * ret._internal[c1][c];
}
}
}
// Normalise last row
{
int c1 = N-1;
zeroit(inner);
for(int c2=0;c2<N;c2++)
inner += innerProduct(ret._internal[c1][c2],ret._internal[c1][c2]);
nrm = sqrt(inner);
nrm = 1.0/nrm;
for(int c2=0;c2<N;c2++)
ret._internal[c1][c2]*= nrm;
}
// assuming the determinant is ok
return ret;

1
Grid/threads/Accelerator.cc
View File

@ -1,6 +1,7 @@
#include <Grid/GridCore.h>
NAMESPACE_BEGIN(Grid);
int acceleratorAbortOnGpuError=1;
uint32_t accelerator_threads=2;
uint32_t acceleratorThreads(void) {return accelerator_threads;};
void acceleratorThreads(uint32_t t) {accelerator_threads = t;};

56
Grid/threads/Accelerator.h
View File

@ -100,9 +100,11 @@ void acceleratorInit(void);
#define accelerator __host__ __device__
#define accelerator_inline __host__ __device__ inline
extern int acceleratorAbortOnGpuError;
accelerator_inline int acceleratorSIMTlane(int Nsimd) {
#ifdef GRID_SIMT
return threadIdx.z;
return threadIdx.x;
#else
return 0;
#endif
@ -110,28 +112,67 @@ accelerator_inline int acceleratorSIMTlane(int Nsimd) {
#define accelerator_for2dNB( iter1, num1, iter2, num2, nsimd, ... ) \
{ \
int nt=acceleratorThreads(); \
typedef uint64_t Iterator; \
auto lambda = [=] accelerator \
(Iterator iter1,Iterator iter2,Iterator lane) mutable { \
__VA_ARGS__; \
}; \
int nt=acceleratorThreads(); \
dim3 cu_threads(acceleratorThreads(),1,nsimd); \
dim3 cu_threads(nsimd,acceleratorThreads(),1); \
dim3 cu_blocks ((num1+nt-1)/nt,num2,1); \
LambdaApply<<<cu_blocks,cu_threads>>>(num1,num2,nsimd,lambda); \
}
#define accelerator_for6dNB(iter1, num1, \
iter2, num2, \
iter3, num3, \
iter4, num4, \
iter5, num5, \
iter6, num6, ... ) \
{ \
typedef uint64_t Iterator; \
auto lambda = [=] accelerator \
(Iterator iter1,Iterator iter2, \
Iterator iter3,Iterator iter4, \
Iterator iter5,Iterator iter6) mutable { \
__VA_ARGS__; \
}; \
dim3 cu_blocks (num1,num2,num3); \
dim3 cu_threads(num4,num5,num6); \
Lambda6Apply<<<cu_blocks,cu_threads>>>(num1,num2,num3,num4,num5,num6,lambda); \
}
template<typename lambda> __global__
void LambdaApply(uint64_t num1, uint64_t num2, uint64_t num3, lambda Lambda)
{
uint64_t x = threadIdx.x + blockDim.x*blockIdx.x;
uint64_t y = threadIdx.y + blockDim.y*blockIdx.y;
uint64_t z = threadIdx.z;
// Weird permute is to make lane coalesce for large blocks
uint64_t x = threadIdx.y + blockDim.y*blockIdx.x;
uint64_t y = threadIdx.z + blockDim.z*blockIdx.y;
uint64_t z = threadIdx.x;
if ( (x < num1) && (y<num2) && (z<num3) ) {
Lambda(x,y,z);
}
}
template<typename lambda> __global__
void Lambda6Apply(uint64_t num1, uint64_t num2, uint64_t num3,
uint64_t num4, uint64_t num5, uint64_t num6,
lambda Lambda)
{
uint64_t iter1 = blockIdx.x;
uint64_t iter2 = blockIdx.y;
uint64_t iter3 = blockIdx.z;
uint64_t iter4 = threadIdx.x;
uint64_t iter5 = threadIdx.y;
uint64_t iter6 = threadIdx.z;
if ( (iter1 < num1) && (iter2<num2) && (iter3<num3)
&& (iter4 < num4) && (iter5<num5) && (iter6<num6) )
{
Lambda(iter1,iter2,iter3,iter4,iter5,iter6);
}
}
#define accelerator_barrier(dummy) \
{ \
cudaDeviceSynchronize(); \
@ -140,6 +181,7 @@ void LambdaApply(uint64_t num1, uint64_t num2, uint64_t num3, lambda Lambda)
printf("Cuda error %s \n", cudaGetErrorString( err )); \
puts(__FILE__); \
printf("Line %d\n",__LINE__); \
if (acceleratorAbortOnGpuError) assert(err==cudaSuccess); \
} \
}
@ -218,7 +260,7 @@ accelerator_inline int acceleratorSIMTlane(int Nsimd) {
cl::sycl::range<3> global{unum1,unum2,nsimd}; \
cgh.parallel_for<class dslash>( \
cl::sycl::nd_range<3>(global,local), \
[=] (cl::sycl::nd_item<3> item) mutable { \
[=] (cl::sycl::nd_item<3> item) /*mutable*/ { \
auto iter1 = item.get_global_id(0); \
auto iter2 = item.get_global_id(1); \
auto lane = item.get_global_id(2); \

9
configure.ac
View File

@ -7,7 +7,12 @@ AM_INIT_AUTOMAKE([subdir-objects 1.13])
AM_EXTRA_RECURSIVE_TARGETS([tests bench])
AC_CONFIG_MACRO_DIR([m4])
AC_CONFIG_SRCDIR([Grid/Grid.h])
AC_CONFIG_HEADERS([Grid/Config.h],[sed -i 's|PACKAGE_|GRID_|' Grid/Config.h])
AC_CONFIG_HEADERS([Grid/Config.h],[[$SED_INPLACE -e 's|PACKAGE_|GRID_|' -e 's|[[:space:]]PACKAGE[[:space:]]| GRID_PACKAGE |' -e 's|[[:space:]]VERSION[[:space:]]| GRID_PACKAGE_VERSION |' Grid/Config.h]],
[if test x"$host_os" == x"${host_os#darwin}" ; then]
[SED_INPLACE="sed -i"]
[else]
[SED_INPLACE="sed -i .bak"]
[fi])
m4_ifdef([AM_SILENT_RULES], [AM_SILENT_RULES([yes])])
################ Get git info
@ -125,7 +130,7 @@ esac
############### fermions
AC_ARG_ENABLE([fermion-reps],
[AC_HELP_STRING([--fermion-reps=yes|no], [enable extra fermion representation support])],
[AC_HELP_STRING([--enable-fermion-reps=yes|no], [enable extra fermion representation support])],
[ac_FERMION_REPS=${enable_fermion_reps}], [ac_FERMION_REPS=yes])
AM_CONDITIONAL(BUILD_FERMION_REPS, [ test "${ac_FERMION_REPS}X" == "yesX" ])

17
tests/core/Test_main.cc
View File

@ -231,6 +231,20 @@ int main(int argc, char **argv) {
scalar = localInnerProduct(cVec, cVec);
scalar = localNorm2(cVec);
std::cout << "Testing maxLocalNorm2" <<std::endl;
LatticeComplex rand_scalar(&Fine);
random(FineRNG, rand_scalar); //uniform [0,1]
for(Integer gsite=0;gsite<Fine.gSites();gsite++){ //check on every site independently
scalar = rand_scalar;
TComplex big(10.0);
Coordinate coor;
Fine.GlobalIndexToGlobalCoor(gsite,coor);
pokeSite(big,scalar,coor);
RealD Linfty = maxLocalNorm2(scalar);
assert(Linfty == 100.0);
}
// -=,+=,*=,()
// add,+,sub,-,mult,mac,*
// adj,conjugate
@ -549,7 +563,8 @@ int main(int argc, char **argv) {
std::vector<int> shiftcoor = coor;
shiftcoor[dir] = (shiftcoor[dir] + shift + latt_size[dir]) %
(latt_size[dir] / mpi_layout[dir]);
(latt_size[dir]);
// (latt_size[dir] / mpi_layout[dir]);
std::vector<int> rl(4);
for (int dd = 0; dd < 4; dd++) {

5
tests/core/Test_reunitarise.cc
View File

@ -103,7 +103,7 @@ int main (int argc, char ** argv)
detU= Determinant(U) ;
detU=detU-1.0;
std::cout << "Determinant before screw up " << norm2(detU)<<std::endl;
std::cout << "Determinant defect before screw up " << norm2(detU)<<std::endl;
std::cout << " Screwing up determinant " << std::endl;
@ -113,7 +113,8 @@ int main (int argc, char ** argv)
auto element = PeekIndex<ColourIndex>(U,Nc-1,i);
element = element * phase;
PokeIndex<ColourIndex>(U,element,Nc-1,i);
}
}
U=U*0.1;
UU=U;
detU= Determinant(U) ;

4
tests/debug/Test_cayley_mres.cc
View File

@ -117,8 +117,8 @@ int main (int argc, char ** argv)
else
{
std::cout<<GridLogMessage <<"Using cold configuration"<<std::endl;
SU<Nc>::ColdConfiguration(Umu);
// SU<Nc>::HotConfiguration(RNG4,Umu);
//SU<Nc>::ColdConfiguration(Umu);
SU<Nc>::HotConfiguration(RNG4,Umu);
}
RealD mass=0.3;

154
tests/forces/Test_momentum_filter.cc Normal file
View File

@ -0,0 +1,154 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./tests/Test_wilson_force.cc
Copyright (C) 2015
Author: Christopher Kelly <ckelly@bnl.gov>
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#include <Grid/Grid.h>
using namespace std;
using namespace Grid;
//Get the mu-directected links on the upper boundary and the bulk remainder
template<typename Field>
void getLinksBoundaryBulk(Field &bound, Field &bulk, Field &from, const Coordinate &latt_size){
bound = Zero(); bulk = Zero();
for(int mu=0;mu<Nd;mu++){
LatticeInteger mucoor(bound.Grid());
LatticeCoordinate(mucoor, mu);
bound = where( mucoor == (Integer)(latt_size[mu] - 1), from, bound );
bulk = where( mucoor != (Integer)(latt_size[mu] - 1), from, bulk );
}
}
int main (int argc, char ** argv)
{
Grid_init(&argc,&argv);
Coordinate latt_size = GridDefaultLatt();
Coordinate simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd());
Coordinate mpi_layout = GridDefaultMpi();
GridCartesian Grid(latt_size,simd_layout,mpi_layout);
GridRedBlackCartesian RBGrid(&Grid);
int threads = GridThread::GetThreads();
std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
std::vector<int> seeds({1,2,3,4});
GridParallelRNG pRNG(&Grid);
pRNG.SeedFixedIntegers(seeds);
typedef PeriodicGimplR Gimpl;
typedef WilsonGaugeAction<Gimpl> GaugeAction;
typedef NoHirep Representation; //fundamental
typedef NoSmearing<Gimpl> Smearing;
typedef MinimumNorm2<Gimpl, Smearing> Omelyan;
typedef Gimpl::Field Field;
typedef MomentumFilterApplyPhase<Field> Filter;
Filter filter(&Grid);
//Setup a filter that disables link update on links passing through the global lattice boundary
typedef Filter::LatticeLorentzScalarType MaskType;
typedef Filter::LorentzScalarType MaskSiteType;
MaskSiteType zero, one;
for(int mu=0;mu<Nd;mu++){
zero(mu)()() = 0.;
one(mu)()() = 1.;
}
MaskType zeroField(&Grid), oneField(&Grid);
zeroField = zero;
oneField = one;
filter.phase = oneField; //make every site 1.0
//Zero mu-directed links at upper boundary
for(int mu=0;mu<Nd;mu++){
LatticeInteger mucoor(&Grid);
LatticeCoordinate(mucoor, mu);
filter.phase = where( mucoor == (Integer)(latt_size[mu] - 1) , zeroField, filter.phase );
}
//Start with a random gauge field
Field U(&Grid);
SU<Nc>::HotConfiguration(pRNG,U);
//Get the original links on the bulk and boundary for later use
Field Ubnd_orig(&Grid), Ubulk_orig(&Grid);
getLinksBoundaryBulk(Ubnd_orig, Ubulk_orig, U, latt_size);
ActionSet<Field,Representation> actions(1);
double beta=6;
GaugeAction gauge_action(beta);
actions[0].push_back(&gauge_action);
Smearing smear;
IntegratorParameters params(1,1.); //1 MD step
Omelyan integrator(&Grid, params, actions, smear);
integrator.setMomentumFilter(filter);
integrator.refresh(U, pRNG); //doesn't actually change the gauge field
//Check the momentum is zero on the boundary
const auto &P = integrator.getMomentum();
Field Pbnd(&Grid), Pbulk(&Grid);
getLinksBoundaryBulk(Pbnd, Pbulk, const_cast<Field&>(P), latt_size);
RealD Pbnd_nrm = norm2(Pbnd); //expect zero
std::cout << GridLogMessage << "After refresh, norm2 of mu-directed conjugate momentum on boundary is: " << Pbnd_nrm << " (expect 0)" << std::endl;
RealD Pbulk_nrm = norm2(Pbulk); //expect non-zero
std::cout << GridLogMessage << "After refresh, norm2 of bulk conjugate momentum is: " << Pbulk_nrm << " (expect non-zero)" << std::endl;
//Evolve the gauge field
integrator.integrate(U);
//Check momentum is still zero on boundary
getLinksBoundaryBulk(Pbnd, Pbulk, const_cast<Field&>(P), latt_size);
Pbnd_nrm = norm2(Pbnd); //expect zero
std::cout << GridLogMessage << "After integrate, norm2 of mu-directed conjugate momentum on boundary is: " << Pbnd_nrm << " (expect 0)" << std::endl;
Pbulk_nrm = norm2(Pbulk); //expect non-zero
std::cout << GridLogMessage << "After integrate, norm2 of bulk conjugate momentum is: " << Pbulk_nrm << " (expect non-zero)" << std::endl;
//Get the new bulk and bound links
Field Ubnd_new(&Grid), Ubulk_new(&Grid);
getLinksBoundaryBulk(Ubnd_new, Ubulk_new, U, latt_size);
Field Ubnd_diff = Ubnd_new - Ubnd_orig;
Field Ubulk_diff = Ubulk_new - Ubulk_orig;
RealD Ubnd_change = norm2( Ubnd_diff );
RealD Ubulk_change = norm2( Ubulk_diff );
std::cout << GridLogMessage << "After integrate, norm2 of change in mu-directed boundary links is : " << Ubnd_change << " (expect 0)" << std::endl;
std::cout << GridLogMessage << "After integrate, norm2 of change in bulk links is : " << Ubulk_change << " (expect non-zero)" << std::endl;
Grid_finalize();
}

7
tests/hmc/Test_hmc_EODWFRatio_Gparity.cc
View File

@ -81,6 +81,10 @@ int main(int argc, char **argv) {
// that have a complex construction
// standard
RealD beta = 5.6 ;
const int nu = 3;
std::vector<int> twists(Nd,0);
twists[nu] = 1;
ConjugateGimplD::setDirections(twists);
ConjugateWilsonGaugeActionR Waction(beta);
const int Ls = 8;
@ -93,9 +97,6 @@ int main(int argc, char **argv) {
// temporarily need a gauge field
LatticeGaugeField U(GridPtr);
const int nu = 3;
std::vector<int> twists(Nd,0);
twists[nu] = 1;
FermionAction::ImplParams params;
params.twists = twists;
Real mass=0.04;

4
tests/hmc/Test_hmc_GparityIwasakiGauge.cc
View File

@ -79,6 +79,10 @@ int main(int argc, char **argv) {
// that have a complex construction
// standard
RealD beta = 2.6 ;
const int nu = 3;
std::vector<int> twists(Nd,0);
twists[nu] = 1;
ConjugateGimplD::setDirections(twists);
ConjugateIwasakiGaugeActionR Waction(beta);

3
tests/hmc/Test_hmc_GparityWilsonGauge.cc
View File

@ -80,6 +80,9 @@ int main(int argc, char **argv) {
// that have a complex construction
// standard
RealD beta = 5.6 ;
std::vector<int> twists(Nd,0);
twists[3] = 1;
ConjugateGimplD::setDirections(twists);
ConjugateWilsonGaugeActionR Waction(beta);

15
tests/solver/Test_dwf_hdcr.cc
View File

@ -222,9 +222,16 @@ int main (int argc, char ** argv)
GridCartesian *Coarse4d = SpaceTimeGrid::makeFourDimGrid(clatt, GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());;
GridCartesian *Coarse5d = SpaceTimeGrid::makeFiveDimGrid(1,Coarse4d);
GridCartesian *CoarseCoarse4d = SpaceTimeGrid::makeFourDimGrid(cclatt, GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());;
GridCartesian *CoarseCoarse4d = SpaceTimeGrid::makeFourDimGrid(cclatt, GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());
GridCartesian *CoarseCoarse5d = SpaceTimeGrid::makeFiveDimGrid(1,CoarseCoarse4d);
GridRedBlackCartesian * Coarse4dRB = SpaceTimeGrid::makeFourDimRedBlackGrid(Coarse4d);
GridRedBlackCartesian * Coarse5dRB = SpaceTimeGrid::makeFiveDimRedBlackGrid(1,Coarse4d);
GridRedBlackCartesian *CoarseCoarse4dRB = SpaceTimeGrid::makeFourDimRedBlackGrid(CoarseCoarse4d);
GridRedBlackCartesian *CoarseCoarse5dRB = SpaceTimeGrid::makeFiveDimRedBlackGrid(1,CoarseCoarse4d);
std::vector<int> seeds4({1,2,3,4});
std::vector<int> seeds5({5,6,7,8});
std::vector<int> cseeds({5,6,7,8});
@ -282,8 +289,7 @@ int main (int argc, char ** argv)
Gamma5R5HermitianLinearOperator<DomainWallFermionR,LatticeFermion> HermIndefOp(Ddwf);
Level1Op LDOp(*Coarse5d,1); LDOp.CoarsenOperator(FGrid,HermIndefOp,Aggregates);
Level1Op LDOp(*Coarse5d,*Coarse5dRB,1); LDOp.CoarsenOperator(FGrid,HermIndefOp,Aggregates);
//////////////////////////////////////////////////
// Deflate the course space. Recursive multigrid?
@ -311,12 +317,11 @@ int main (int argc, char ** argv)
}
}
Level2Op L2Op(*CoarseCoarse5d,1); // Hermitian matrix
Level2Op L2Op(*CoarseCoarse5d,*CoarseCoarse5dRB,1); // Hermitian matrix
typedef Level2Op::CoarseVector CoarseCoarseVector;
HermitianLinearOperator<Level1Op,CoarseVector> L1LinOp(LDOp);
L2Op.CoarsenOperator(Coarse5d,L1LinOp,CoarseAggregates);
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << " Running CoarseCoarse grid Lanczos "<< std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;

397
tests/solver/Test_dwf_hdcr_16_rb.cc Normal file
View File

@ -0,0 +1,397 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./tests/Test_dwf_hdcr.cc
Copyright (C) 2015
Author: Antonin Portelli <antonin.portelli@me.com>
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: paboyle <paboyle@ph.ed.ac.uk>
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#include <Grid/Grid.h>
#include <Grid/algorithms/iterative/PrecGeneralisedConjugateResidual.h>
using namespace std;
using namespace Grid;
/* Params
* Grid:
* block1(4)
* block2(4)
*
* Subspace
* * Fine : Subspace(nbasis,hi,lo,order,first,step) -- 32, 60,0.02,500,100,100
* * Coarse: Subspace(nbasis,hi,lo,order,first,step) -- 32, 18,0.02,500,100,100
* Smoother:
* * Fine: Cheby(hi, lo, order) -- 60,0.5,10
* * Coarse: Cheby(hi, lo, order) -- 12,0.1,4
* Lanczos:
* CoarseCoarse IRL( Nk, Nm, Nstop, poly(lo,hi,order)) 24,36,24,0.002,4.0,61
*/
RealD InverseApproximation(RealD x){
return 1.0/x;
}
template<class Field> class SolverWrapper : public LinearFunction<Field> {
private:
CheckerBoardedSparseMatrixBase<Field> & _Matrix;
SchurRedBlackBase<Field> & _Solver;
public:
/////////////////////////////////////////////////////
// Wrap the usual normal equations trick
/////////////////////////////////////////////////////
SolverWrapper(CheckerBoardedSparseMatrixBase<Field> &Matrix,
SchurRedBlackBase<Field> &Solver)
: _Matrix(Matrix), _Solver(Solver) {};
void operator() (const Field &in, Field &out){
_Solver(_Matrix,in,out); // Mdag M out = Mdag in
}
};
template<class Field,class Matrix> class ChebyshevSmoother : public LinearFunction<Field>
{
public:
typedef LinearOperatorBase<Field> FineOperator;
Matrix & _SmootherMatrix;
FineOperator & _SmootherOperator;
Chebyshev<Field> Cheby;
ChebyshevSmoother(RealD _lo,RealD _hi,int _ord, FineOperator &SmootherOperator,Matrix &SmootherMatrix) :
_SmootherOperator(SmootherOperator),
_SmootherMatrix(SmootherMatrix),
Cheby(_lo,_hi,_ord,InverseApproximation)
{};
void operator() (const Field &in, Field &out)
{
Field tmp(in.Grid());
MdagMLinearOperator<Matrix,Field> MdagMOp(_SmootherMatrix);
_SmootherOperator.AdjOp(in,tmp);
Cheby(MdagMOp,tmp,out);
}
};
template<class Field,class Matrix> class MirsSmoother : public LinearFunction<Field>
{
public:
typedef LinearOperatorBase<Field> FineOperator;
Matrix & SmootherMatrix;
FineOperator & SmootherOperator;
RealD tol;
RealD shift;
int maxit;
MirsSmoother(RealD _shift,RealD _tol,int _maxit,FineOperator &_SmootherOperator,Matrix &_SmootherMatrix) :
shift(_shift),tol(_tol),maxit(_maxit),
SmootherOperator(_SmootherOperator),
SmootherMatrix(_SmootherMatrix)
{};
void operator() (const Field &in, Field &out)
{
ZeroGuesser<Field> Guess;
ConjugateGradient<Field> CG(tol,maxit,false);
Field src(in.Grid());
ShiftedMdagMLinearOperator<SparseMatrixBase<Field>,Field> MdagMOp(SmootherMatrix,shift);
SmootherOperator.AdjOp(in,src);
Guess(src,out);
CG(MdagMOp,src,out);
}
};
template<class Fobj,class CComplex,int nbasis, class Matrix, class Guesser, class CoarseSolver>
class MultiGridPreconditioner : public LinearFunction< Lattice<Fobj> > {
public:
typedef Aggregation<Fobj,CComplex,nbasis> Aggregates;
typedef CoarsenedMatrix<Fobj,CComplex,nbasis> CoarseOperator;
typedef typename Aggregation<Fobj,CComplex,nbasis>::CoarseVector CoarseVector;
typedef typename Aggregation<Fobj,CComplex,nbasis>::CoarseMatrix CoarseMatrix;
typedef typename Aggregation<Fobj,CComplex,nbasis>::FineField FineField;
typedef LinearOperatorBase<FineField> FineOperator;
typedef LinearFunction <FineField> FineSmoother;
Aggregates & _Aggregates;
CoarseOperator & _CoarseOperator;
Matrix & _FineMatrix;
FineOperator & _FineOperator;
Guesser & _Guess;
FineSmoother & _Smoother;
CoarseSolver & _CoarseSolve;
int level; void Level(int lv) {level = lv; };
#define GridLogLevel std::cout << GridLogMessage <<std::string(level,'\t')<< " Level "<<level <<" "
MultiGridPreconditioner(Aggregates &Agg, CoarseOperator &Coarse,
FineOperator &Fine,Matrix &FineMatrix,
FineSmoother &Smoother,
Guesser &Guess_,
CoarseSolver &CoarseSolve_)
: _Aggregates(Agg),
_CoarseOperator(Coarse),
_FineOperator(Fine),
_FineMatrix(FineMatrix),
_Smoother(Smoother),
_Guess(Guess_),
_CoarseSolve(CoarseSolve_),
level(1) { }
virtual void operator()(const FineField &in, FineField & out)
{
CoarseVector Csrc(_CoarseOperator.Grid());
CoarseVector Csol(_CoarseOperator.Grid());
FineField vec1(in.Grid());
FineField vec2(in.Grid());
double t;
// Fine Smoother
t=-usecond();
_Smoother(in,out);
t+=usecond();
GridLogLevel << "Smoother took "<< t/1000.0<< "ms" <<std::endl;
// Update the residual
_FineOperator.Op(out,vec1); sub(vec1, in ,vec1);
// Fine to Coarse
t=-usecond();
_Aggregates.ProjectToSubspace (Csrc,vec1);
t+=usecond();
GridLogLevel << "Project to coarse took "<< t/1000.0<< "ms" <<std::endl;
// Coarse correction
t=-usecond();
_CoarseSolve(Csrc,Csol);
t+=usecond();
GridLogLevel << "Coarse solve took "<< t/1000.0<< "ms" <<std::endl;
// Coarse to Fine
t=-usecond();
_Aggregates.PromoteFromSubspace(Csol,vec1);
add(out,out,vec1);
t+=usecond();
GridLogLevel << "Promote to this level took "<< t/1000.0<< "ms" <<std::endl;
// Residual
_FineOperator.Op(out,vec1); sub(vec1 ,in , vec1);
// Fine Smoother
t=-usecond();
_Smoother(vec1,vec2);
t+=usecond();
GridLogLevel << "Smoother took "<< t/1000.0<< "ms" <<std::endl;
add( out,out,vec2);
}
};
int main (int argc, char ** argv)
{
Grid_init(&argc,&argv);
const int Ls=16;
GridCartesian * UGrid = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());
GridRedBlackCartesian * UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
GridCartesian * FGrid = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid);
///////////////////////////////////////////////////
// Construct a coarsened grid; utility for this?
///////////////////////////////////////////////////
std::vector<int> block ({2,2,2,2});
std::vector<int> blockc ({2,2,2,2});
const int nbasis= 32;
const int nbasisc= 32;
auto clatt = GridDefaultLatt();
for(int d=0;d<clatt.size();d++){
clatt[d] = clatt[d]/block[d];
}
auto cclatt = clatt;
for(int d=0;d<clatt.size();d++){
cclatt[d] = clatt[d]/blockc[d];
}
GridCartesian *Coarse4d = SpaceTimeGrid::makeFourDimGrid(clatt, GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());;
GridCartesian *Coarse5d = SpaceTimeGrid::makeFiveDimGrid(1,Coarse4d);
// GridCartesian *CoarseCoarse4d = SpaceTimeGrid::makeFourDimGrid(cclatt, GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());;
// GridCartesian *CoarseCoarse5d = SpaceTimeGrid::makeFiveDimGrid(1,CoarseCoarse4d);
std::vector<int> seeds4({1,2,3,4});
std::vector<int> seeds5({5,6,7,8});
std::vector<int> cseeds({5,6,7,8});
GridParallelRNG RNG5(FGrid); RNG5.SeedFixedIntegers(seeds5);
GridParallelRNG RNG4(UGrid); RNG4.SeedFixedIntegers(seeds4);
GridParallelRNG CRNG(Coarse5d);CRNG.SeedFixedIntegers(cseeds);
LatticeFermion src(FGrid); gaussian(RNG5,src);// src=src+g5*src;
LatticeFermion result(FGrid);
LatticeGaugeField Umu(UGrid);
FieldMetaData header;
std::string file("./ckpoint_lat.4000");
//std::string file("./ckpoint_lat.1000");
NerscIO::readConfiguration(Umu,header,file);
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Building g5R5 hermitian DWF operator" <<std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
RealD mass=0.001;
RealD M5=1.8;
DomainWallFermionR Ddwf(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
typedef Aggregation<vSpinColourVector,vTComplex,nbasis> Subspace;
typedef CoarsenedMatrix<vSpinColourVector,vTComplex,nbasis> CoarseOperator;
typedef CoarseOperator::CoarseVector CoarseVector;
typedef CoarseOperator::siteVector siteVector;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Calling Aggregation class to build subspace" <<std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
MdagMLinearOperator<DomainWallFermionR,LatticeFermion> HermDefOp(Ddwf);
Subspace Aggregates(Coarse5d,FGrid,0);
assert ( (nbasis & 0x1)==0);
{
int nb=nbasis/2;
LatticeFermion A(FGrid);
LatticeFermion B(FGrid);
// Aggregates.CreateSubspaceChebyshev(RNG5,HermDefOp,nb,60.0,0.002,1000,800,100,0.0);
// Aggregates.CreateSubspaceChebyshev(RNG5,HermDefOp,nb,60.0,0.02,1000,800,100,0.0);
Aggregates.CreateSubspaceChebyshev(RNG5,HermDefOp,nb,60.0,0.05,500,200,150,0.0);//
// Aggregates.CreateSubspaceChebyshev(RNG5,HermDefOp,nb,60.0,0.01,1000,100,100,0.0); // Slightly faster
for(int n=0;n<nb;n++){
std::cout << GridLogMessage << " G5R5 "<<n<<std::endl;
G5R5(Aggregates.subspace[n+nb],Aggregates.subspace[n]);
std::cout << GridLogMessage << " Projection "<<n<<std::endl;
A = Aggregates.subspace[n];
B = Aggregates.subspace[n+nb];
std::cout << GridLogMessage << " Copy "<<n<<std::endl;
Aggregates.subspace[n] = A+B; // 1+G5 // eigen value of G5R5 is +1
std::cout << GridLogMessage << " P+ "<<n<<std::endl;
Aggregates.subspace[n+nb]= A-B; // 1-G5 // eigen value of G5R5 is -1
std::cout << GridLogMessage << " P- "<<n<<std::endl;
}
}
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Building coarse representation of Indef operator" <<std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
typedef CoarsenedMatrix<vSpinColourVector,vTComplex,nbasis> Level1Op;
typedef CoarsenedMatrix<siteVector,iScalar<vTComplex>,nbasisc> Level2Op;
Gamma5R5HermitianLinearOperator<DomainWallFermionR,LatticeFermion> HermIndefOp(Ddwf);
GridRedBlackCartesian * Coarse4dRB = SpaceTimeGrid::makeFourDimRedBlackGrid(Coarse4d);
std::cout << " Making 5D coarse RB grid " <<std::endl;
GridRedBlackCartesian * Coarse5dRB = SpaceTimeGrid::makeFiveDimRedBlackGrid(1,Coarse4d);
std::cout << " Made 5D coarse RB grid " <<std::endl;
Level1Op LDOp(*Coarse5d,*Coarse5dRB,1); LDOp.CoarsenOperator(FGrid,HermIndefOp,Aggregates);
//////////////////////////////////////////////////
// Deflate the course space. Recursive multigrid?
//////////////////////////////////////////////////
typedef Aggregation<siteVector,iScalar<vTComplex>,nbasisc> CoarseSubspace;
// CoarseSubspace CoarseAggregates(CoarseCoarse5d,Coarse5d,0);
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Build deflation space in coarse operator "<< std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
MdagMLinearOperator<CoarseOperator,CoarseVector> PosdefLdop(LDOp);
typedef Level2Op::CoarseVector CoarseCoarseVector;
CoarseVector c_src(Coarse5d); c_src=1.0;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Building 3 level Multigrid "<< std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
typedef MultiGridPreconditioner<vSpinColourVector, vTComplex,nbasis, DomainWallFermionR,ZeroGuesser<CoarseVector> , SolverWrapper<CoarseVector> > TwoLevelMG;
typedef MultiGridPreconditioner<siteVector,iScalar<vTComplex>,nbasisc,Level1Op, DeflatedGuesser<CoarseCoarseVector>, NormalEquations<CoarseCoarseVector> > CoarseMG;
typedef MultiGridPreconditioner<vSpinColourVector, vTComplex,nbasis, DomainWallFermionR,ZeroGuesser<CoarseVector>, LinearFunction<CoarseVector> > ThreeLevelMG;
ChebyshevSmoother<LatticeFermion,DomainWallFermionR> FineSmoother(0.5,60.0,12,HermIndefOp,Ddwf);
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Calling 2 level Multigrid "<< std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
result=Zero();
ZeroGuesser<CoarseVector> CoarseZeroGuesser;
ConjugateGradient<CoarseVector> CoarseCG(0.005,1000);
// SchurDiagMooeeOperator<CoarseOperator,CoarseVector> CoarseMpcDagMpc(LDOp);
SchurRedBlackDiagMooeeSolve<CoarseVector> CoarseRBCG(CoarseCG);
SolverWrapper<CoarseVector> CoarseSolver(LDOp,CoarseRBCG);
// NormalEquations<CoarseVector> CoarseCGNE(LDOp,CoarseCG,CoarseZeroGuesser);
TwoLevelMG TwoLevelPrecon(Aggregates, LDOp,
HermIndefOp,Ddwf,
FineSmoother,
CoarseZeroGuesser,
CoarseSolver);
TwoLevelPrecon.Level(1);
PrecGeneralisedConjugateResidual<LatticeFermion> l1PGCR(1.0e-8,20,HermIndefOp,TwoLevelPrecon,16,16);
l1PGCR.Level(1);
l1PGCR(src,result);
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Calling CG "<< std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
ConjugateGradient<LatticeFermion> pCG(1.0e-8,60000);
result=Zero();
// pCG(HermDefOp,src,result);
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Calling red black CG "<< std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
result=Zero();
LatticeFermion src_o(FrbGrid);
LatticeFermion result_o(FrbGrid);
pickCheckerboard(Odd,src_o,src);
result_o=Zero();
SchurDiagMooeeOperator<DomainWallFermionR,LatticeFermion> HermOpEO(Ddwf);
// pCG(HermOpEO,src_o,result_o);
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << " Fine PowerMethod "<< std::endl;
PowerMethod<LatticeFermion> PM; PM(HermDefOp,src);
std::cout<<GridLogMessage << " Coarse PowerMethod "<< std::endl;
PowerMethod<CoarseVector> cPM; cPM(PosdefLdop,c_src);
// std::cout<<GridLogMessage << " CoarseCoarse PowerMethod "<< std::endl;
// PowerMethod<CoarseCoarseVector> ccPM; ccPM(IRLHermOpL2,cc_src);
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Done "<< std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
Grid_finalize();
}

477
tests/solver/Test_dwf_hdcr_24_regression.cc Normal file
View File

@ -0,0 +1,477 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./tests/Test_dwf_hdcr.cc
Copyright (C) 2015
Author: Antonin Portelli <antonin.portelli@me.com>
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: paboyle <paboyle@ph.ed.ac.uk>
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#include <Grid/Grid.h>
#include <Grid/algorithms/iterative/PrecGeneralisedConjugateResidual.h>
using namespace std;
using namespace Grid;
/* Params
* Grid:
* block1(4)
* block2(4)
*
* Subspace
* * Fine : Subspace(nbasis,hi,lo,order,first,step) -- 32, 60,0.02,500,100,100
* * Coarse: Subspace(nbasis,hi,lo,order,first,step) -- 32, 18,0.02,500,100,100
* Smoother:
* * Fine: Cheby(hi, lo, order) -- 60,0.5,10
* * Coarse: Cheby(hi, lo, order) -- 12,0.1,4
* Lanczos:
* CoarseCoarse IRL( Nk, Nm, Nstop, poly(lo,hi,order)) 24,36,24,0.002,4.0,61
*/
RealD InverseApproximation(RealD x){
return 1.0/x;
}
template<class Field,class Matrix> class ChebyshevSmoother : public LinearFunction<Field>
{
public:
typedef LinearOperatorBase<Field> FineOperator;
Matrix & _SmootherMatrix;
FineOperator & _SmootherOperator;
Chebyshev<Field> Cheby;
ChebyshevSmoother(RealD _lo,RealD _hi,int _ord, FineOperator &SmootherOperator,Matrix &SmootherMatrix) :
_SmootherOperator(SmootherOperator),
_SmootherMatrix(SmootherMatrix),
Cheby(_lo,_hi,_ord,InverseApproximation)
{};
void operator() (const Field &in, Field &out)
{
Field tmp(in.Grid());
MdagMLinearOperator<Matrix,Field> MdagMOp(_SmootherMatrix);
_SmootherOperator.AdjOp(in,tmp);
Cheby(MdagMOp,tmp,out);
}
};
template<class Field,class Matrix> class MirsSmoother : public LinearFunction<Field>
{
public:
typedef LinearOperatorBase<Field> FineOperator;
Matrix & SmootherMatrix;
FineOperator & SmootherOperator;
RealD tol;
RealD shift;
int maxit;
MirsSmoother(RealD _shift,RealD _tol,int _maxit,FineOperator &_SmootherOperator,Matrix &_SmootherMatrix) :
shift(_shift),tol(_tol),maxit(_maxit),
SmootherOperator(_SmootherOperator),
SmootherMatrix(_SmootherMatrix)
{};
void operator() (const Field &in, Field &out)
{
ZeroGuesser<Field> Guess;
ConjugateGradient<Field> CG(tol,maxit,false);
Field src(in.Grid());
ShiftedMdagMLinearOperator<SparseMatrixBase<Field>,Field> MdagMOp(SmootherMatrix,shift);
SmootherOperator.AdjOp(in,src);
Guess(src,out);
CG(MdagMOp,src,out);
}
};
template<class Fobj,class CComplex,int nbasis, class Matrix, class Guesser, class CoarseSolver>
class MultiGridPreconditioner : public LinearFunction< Lattice<Fobj> > {
public:
typedef Aggregation<Fobj,CComplex,nbasis> Aggregates;
typedef CoarsenedMatrix<Fobj,CComplex,nbasis> CoarseOperator;
typedef typename Aggregation<Fobj,CComplex,nbasis>::CoarseVector CoarseVector;
typedef typename Aggregation<Fobj,CComplex,nbasis>::CoarseMatrix CoarseMatrix;
typedef typename Aggregation<Fobj,CComplex,nbasis>::FineField FineField;
typedef LinearOperatorBase<FineField> FineOperator;
typedef LinearFunction <FineField> FineSmoother;
Aggregates & _Aggregates;
CoarseOperator & _CoarseOperator;
Matrix & _FineMatrix;
FineOperator & _FineOperator;
Guesser & _Guess;
FineSmoother & _Smoother;
CoarseSolver & _CoarseSolve;
int level; void Level(int lv) {level = lv; };
#define GridLogLevel std::cout << GridLogMessage <<std::string(level,'\t')<< " Level "<<level <<" "
MultiGridPreconditioner(Aggregates &Agg, CoarseOperator &Coarse,
FineOperator &Fine,Matrix &FineMatrix,
FineSmoother &Smoother,
Guesser &Guess_,
CoarseSolver &CoarseSolve_)
: _Aggregates(Agg),
_CoarseOperator(Coarse),
_FineOperator(Fine),
_FineMatrix(FineMatrix),
_Smoother(Smoother),
_Guess(Guess_),
_CoarseSolve(CoarseSolve_),
level(1) { }
virtual void operator()(const FineField &in, FineField & out)
{
CoarseVector Csrc(_CoarseOperator.Grid());
CoarseVector Csol(_CoarseOperator.Grid());
FineField vec1(in.Grid());
FineField vec2(in.Grid());
double t;
// Fine Smoother
t=-usecond();
_Smoother(in,out);
t+=usecond();
GridLogLevel << "Smoother took "<< t/1000.0<< "ms" <<std::endl;
// Update the residual
_FineOperator.Op(out,vec1); sub(vec1, in ,vec1);
// Fine to Coarse
t=-usecond();
_Aggregates.ProjectToSubspace (Csrc,vec1);
t+=usecond();
GridLogLevel << "Project to coarse took "<< t/1000.0<< "ms" <<std::endl;
// Coarse correction
t=-usecond();
_CoarseSolve(Csrc,Csol);
t+=usecond();
GridLogLevel << "Coarse solve took "<< t/1000.0<< "ms" <<std::endl;
// Coarse to Fine
t=-usecond();
_Aggregates.PromoteFromSubspace(Csol,vec1);
add(out,out,vec1);
t+=usecond();
GridLogLevel << "Promote to this level took "<< t/1000.0<< "ms" <<std::endl;
// Residual
_FineOperator.Op(out,vec1); sub(vec1 ,in , vec1);
// Fine Smoother
t=-usecond();
_Smoother(vec1,vec2);
t+=usecond();
GridLogLevel << "Smoother took "<< t/1000.0<< "ms" <<std::endl;
add( out,out,vec2);
}
};
int main (int argc, char ** argv)
{
Grid_init(&argc,&argv);
const int Ls=24;
GridCartesian * UGrid = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());
GridRedBlackCartesian * UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
GridCartesian * FGrid = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid);
///////////////////////////////////////////////////
// Construct a coarsened grid; utility for this?
///////////////////////////////////////////////////
std::vector<int> block ({2,2,2,2});
std::vector<int> blockc ({2,2,2,2});
const int nbasis= 40;
const int nbasisc= 40;
auto clatt = GridDefaultLatt();
for(int d=0;d<clatt.size();d++){
clatt[d] = clatt[d]/block[d];
}
auto cclatt = clatt;
for(int d=0;d<clatt.size();d++){
cclatt[d] = clatt[d]/blockc[d];
}
GridCartesian *Coarse4d = SpaceTimeGrid::makeFourDimGrid(clatt, GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());;
GridCartesian *Coarse5d = SpaceTimeGrid::makeFiveDimGrid(1,Coarse4d);
// GridCartesian *CoarseCoarse4d = SpaceTimeGrid::makeFourDimGrid(cclatt, GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());;
// GridCartesian *CoarseCoarse5d = SpaceTimeGrid::makeFiveDimGrid(1,CoarseCoarse4d);
std::vector<int> seeds4({1,2,3,4});
std::vector<int> seeds5({5,6,7,8});
std::vector<int> cseeds({5,6,7,8});
GridParallelRNG RNG5(FGrid); RNG5.SeedFixedIntegers(seeds5);
GridParallelRNG RNG4(UGrid); RNG4.SeedFixedIntegers(seeds4);
GridParallelRNG CRNG(Coarse5d);CRNG.SeedFixedIntegers(cseeds);
LatticeFermion src(FGrid); gaussian(RNG5,src);// src=src+g5*src;
LatticeFermion result(FGrid);
LatticeGaugeField Umu(UGrid);
FieldMetaData header;
// std::string file("./ckpoint_lat.4000");
// std::string file("./ckpoint_lat.1000");
// NerscIO::readConfiguration(Umu,header,file);
SU<Nc>::HotConfiguration(RNG4,Umu);
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Building g5R5 hermitian DWF operator" <<std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
RealD mass=0.00078;
RealD M5=1.8;
DomainWallFermionR Ddwf(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
typedef Aggregation<vSpinColourVector,vTComplex,nbasis> Subspace;
typedef CoarsenedMatrix<vSpinColourVector,vTComplex,nbasis> CoarseOperator;
typedef CoarseOperator::CoarseVector CoarseVector;
typedef CoarseOperator::siteVector siteVector;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Calling Aggregation class to build subspace" <<std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
MdagMLinearOperator<DomainWallFermionR,LatticeFermion> HermDefOp(Ddwf);
Subspace Aggregates(Coarse5d,FGrid,0);
assert ( (nbasis & 0x1)==0);
{
int nb=nbasis/2;
LatticeFermion A(FGrid);
LatticeFermion B(FGrid);
// Aggregates.CreateSubspaceChebyshev(RNG5,HermDefOp,nb,60.0,0.002,1000,800,100,0.0);
// Aggregates.CreateSubspaceChebyshev(RNG5,HermDefOp,nb,60.0,0.02,1000,800,100,0.0);
Aggregates.CreateSubspaceChebyshev(RNG5,HermDefOp,nb,60.0,0.01,400,50,50,0.0); // Slightly faster
for(int n=0;n<nb;n++){
std::cout << GridLogMessage << " G5R5 "<<n<<std::endl;
G5R5(Aggregates.subspace[n+nb],Aggregates.subspace[n]);
std::cout << GridLogMessage << " Projection "<<n<<std::endl;
A = Aggregates.subspace[n];
B = Aggregates.subspace[n+nb];
std::cout << GridLogMessage << " Copy "<<n<<std::endl;
Aggregates.subspace[n] = A+B; // 1+G5 // eigen value of G5R5 is +1
std::cout << GridLogMessage << " P+ "<<n<<std::endl;
Aggregates.subspace[n+nb]= A-B; // 1-G5 // eigen value of G5R5 is -1
std::cout << GridLogMessage << " P- "<<n<<std::endl;
}
}
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Building coarse representation of Indef operator" <<std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
typedef CoarsenedMatrix<vSpinColourVector,vTComplex,nbasis> Level1Op;
typedef CoarsenedMatrix<siteVector,iScalar<vTComplex>,nbasisc> Level2Op;
Gamma5R5HermitianLinearOperator<DomainWallFermionR,LatticeFermion> HermIndefOp(Ddwf);
GridRedBlackCartesian * Coarse4dRB = SpaceTimeGrid::makeFourDimRedBlackGrid(Coarse4d);
std::cout << " Making 5D coarse RB grid " <<std::endl;
GridRedBlackCartesian * Coarse5dRB = SpaceTimeGrid::makeFiveDimRedBlackGrid(1,Coarse4d);
std::cout << " Made 5D coarse RB grid " <<std::endl;
Level1Op LDOp(*Coarse5d,*Coarse5dRB,1);
std::cout << " LDOp.CoarsenOperator " <<std::endl;
LDOp.CoarsenOperator(FGrid,HermIndefOp,Aggregates);
std::cout << " Coarsened Operator " <<std::endl;
//////////////////////////////////////////////////
// Deflate the course space. Recursive multigrid?
//////////////////////////////////////////////////
typedef Aggregation<siteVector,iScalar<vTComplex>,nbasisc> CoarseSubspace;
// CoarseSubspace CoarseAggregates(CoarseCoarse5d,Coarse5d,0);
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Build deflation space in coarse operator "<< std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
MdagMLinearOperator<CoarseOperator,CoarseVector> PosdefLdop(LDOp);
/*
{
int nb=nbasisc/2;
CoarseAggregates.CreateSubspaceChebyshev(CRNG,PosdefLdop,nb,15.0,0.02,1000,800,100,0.0);
for(int n=0;n<nb;n++){
autoView( subspace , CoarseAggregates.subspace[n],CpuWrite);
autoView( subspace_g5, CoarseAggregates.subspace[n+nb],CpuWrite);
for(int nn=0;nn<nb;nn++){
for(int site=0;site<Coarse5d->oSites();site++){
subspace_g5[site](nn) = subspace[site](nn);
subspace_g5[site](nn+nb)=-subspace[site](nn+nb);
}
}
}
}
*/
typedef Level2Op::CoarseVector CoarseCoarseVector;
/*
Level2Op L2Op(*CoarseCoarse5d,1); // Hermitian matrix
HermitianLinearOperator<Level1Op,CoarseVector> L1LinOp(LDOp);
L2Op.CoarsenOperator(Coarse5d,L1LinOp,CoarseAggregates);
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << " Running CoarseCoarse grid Lanczos "<< std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
MdagMLinearOperator<Level2Op,CoarseCoarseVector> IRLHermOpL2(L2Op);
CoarseCoarseVector cc_src(CoarseCoarse5d); cc_src=1.0;
*/
/*
Chebyshev<CoarseCoarseVector> IRLChebyL2(0.001,15.0,301);
FunctionHermOp<CoarseCoarseVector> IRLOpChebyL2(IRLChebyL2,IRLHermOpL2);
PlainHermOp<CoarseCoarseVector> IRLOpL2 (IRLHermOpL2);
int cNk=24;
int cNm=36;
int cNstop=24;
ImplicitlyRestartedLanczos<CoarseCoarseVector> IRLL2(IRLOpChebyL2,IRLOpL2,cNstop,cNk,cNm,1.0e-3,20);
int cNconv;
std::vector<RealD> eval2(cNm);
std::vector<CoarseCoarseVector> evec2(cNm,CoarseCoarse5d);
IRLL2.calc(eval2,evec2,cc_src,cNconv);
ConjugateGradient<CoarseCoarseVector> CoarseCoarseCG(0.1,1000);
DeflatedGuesser<CoarseCoarseVector> DeflCoarseCoarseGuesser(evec2,eval2);
NormalEquations<CoarseCoarseVector> DeflCoarseCoarseCGNE(L2Op,CoarseCoarseCG,DeflCoarseCoarseGuesser);
*/
/*
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << " Running Coarse grid Lanczos "<< std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
MdagMLinearOperator<Level1Op,CoarseVector> IRLHermOp(LDOp);
// Chebyshev<CoarseVector> IRLCheby(0.001,15.0,301);
Chebyshev<CoarseVector> IRLCheby(0.03,12.0,101);
FunctionHermOp<CoarseVector> IRLOpCheby(IRLCheby,IRLHermOp);
PlainHermOp<CoarseVector> IRLOp (IRLHermOp);
int Nk=64;
int Nm=128;
int Nstop=Nk;
ImplicitlyRestartedLanczos<CoarseVector> IRL(IRLOpCheby,IRLOp,Nstop,Nk,Nm,1.0e-3,20);
int Nconv;
std::vector<RealD> eval(Nm);
std::vector<CoarseVector> evec(Nm,Coarse5d);
IRL.calc(eval,evec,c_src,Nconv);
*/
CoarseVector c_src(Coarse5d); c_src=1.0;
// DeflatedGuesser<CoarseVector> DeflCoarseGuesser(evec,eval);
// NormalEquations<CoarseVector> DeflCoarseCGNE(LDOp,CoarseCG,DeflCoarseGuesser);
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Building 3 level Multigrid "<< std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
// typedef MultiGridPreconditioner<vSpinColourVector, vTComplex,nbasis, DomainWallFermionR,DeflatedGuesser<CoarseVector> , NormalEquations<CoarseVector> > TwoLevelMG;
typedef MultiGridPreconditioner<vSpinColourVector, vTComplex,nbasis, DomainWallFermionR,ZeroGuesser<CoarseVector> , NormalEquations<CoarseVector> > TwoLevelMG;
typedef MultiGridPreconditioner<siteVector,iScalar<vTComplex>,nbasisc,Level1Op, DeflatedGuesser<CoarseCoarseVector>, NormalEquations<CoarseCoarseVector> > CoarseMG;
typedef MultiGridPreconditioner<vSpinColourVector, vTComplex,nbasis, DomainWallFermionR,ZeroGuesser<CoarseVector>, LinearFunction<CoarseVector> > ThreeLevelMG;
ChebyshevSmoother<LatticeFermion,DomainWallFermionR> FineSmoother(0.25,60.0,12,HermIndefOp,Ddwf);
/*
// MultiGrid preconditioner acting on the coarse space <-> coarsecoarse space
ChebyshevSmoother<CoarseVector, Level1Op > CoarseSmoother(0.1,15.0,3,L1LinOp,LDOp);
// MirsSmoother<CoarseVector, Level1Op > CoarseCGSmoother(0.1,0.1,4,L1LinOp,LDOp);
// MirsSmoother<LatticeFermion,DomainWallFermionR> FineCGSmoother(0.0,0.01,8,HermIndefOp,Ddwf);
CoarseMG Level2Precon (CoarseAggregates, L2Op,
L1LinOp,LDOp,
CoarseSmoother,
DeflCoarseCoarseGuesser,
DeflCoarseCoarseCGNE);
Level2Precon.Level(2);
// PGCR Applying this solver to solve the coarse space problem
PrecGeneralisedConjugateResidual<CoarseVector> l2PGCR(0.1, 100, L1LinOp,Level2Precon,16,16);
l2PGCR.Level(2);
// Wrap the 2nd level solver in a MultiGrid preconditioner acting on the fine space
ZeroGuesser<CoarseVector> CoarseZeroGuesser;
ThreeLevelMG ThreeLevelPrecon(Aggregates, LDOp,
HermIndefOp,Ddwf,
FineSmoother,
CoarseZeroGuesser,
l2PGCR);
ThreeLevelPrecon.Level(1);
// Apply the fine-coarse-coarsecoarse 2 deep MG preconditioner in an outer PGCR on the fine fgrid
PrecGeneralisedConjugateResidual<LatticeFermion> l1PGCR(1.0e-8,1000,HermIndefOp,ThreeLevelPrecon,16,16);
l1PGCR.Level(1);
*/
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Calling 2 level Multigrid "<< std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
result=Zero();
ZeroGuesser<CoarseVector> CoarseZeroGuesser;
ConjugateGradient<CoarseVector> CoarseCG(0.01,1000);
NormalEquations<CoarseVector> CoarseCGNE(LDOp,CoarseCG,CoarseZeroGuesser);
TwoLevelMG TwoLevelPrecon(Aggregates, LDOp,
HermIndefOp,Ddwf,
FineSmoother,
CoarseZeroGuesser,
CoarseCGNE);
TwoLevelPrecon.Level(1);
PrecGeneralisedConjugateResidual<LatticeFermion> l1PGCR(1.0e-8,20,HermIndefOp,TwoLevelPrecon,16,16);
l1PGCR.Level(1);
l1PGCR(src,result);
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Calling CG "<< std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
ConjugateGradient<LatticeFermion> pCG(1.0e-8,60000);
result=Zero();
// pCG(HermDefOp,src,result);
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Calling red black CG "<< std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
result=Zero();
LatticeFermion src_o(FrbGrid);
LatticeFermion result_o(FrbGrid);
pickCheckerboard(Odd,src_o,src);
result_o=Zero();
SchurDiagMooeeOperator<DomainWallFermionR,LatticeFermion> HermOpEO(Ddwf);
// pCG(HermOpEO,src_o,result_o);
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << " Fine PowerMethod "<< std::endl;
PowerMethod<LatticeFermion> PM; PM(HermDefOp,src);
std::cout<<GridLogMessage << " Coarse PowerMethod "<< std::endl;
PowerMethod<CoarseVector> cPM; cPM(PosdefLdop,c_src);
// std::cout<<GridLogMessage << " CoarseCoarse PowerMethod "<< std::endl;
// PowerMethod<CoarseCoarseVector> ccPM; ccPM(IRLHermOpL2,cc_src);
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Done "<< std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
Grid_finalize();
}

8
tests/solver/Test_dwf_hdcr_2level.cc
View File

@ -262,6 +262,8 @@ int main (int argc, char ** argv)
GridCartesian *Coarse4d = SpaceTimeGrid::makeFourDimGrid(clatt, GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());;
GridCartesian *Coarse5d = SpaceTimeGrid::makeFiveDimGrid(1,Coarse4d);
GridRedBlackCartesian * Coarse4dRB = SpaceTimeGrid::makeFourDimRedBlackGrid(Coarse4d);
GridRedBlackCartesian * Coarse5dRB = SpaceTimeGrid::makeFiveDimRedBlackGrid(1,Coarse4d);
std::vector<int> seeds4({1,2,3,4});
std::vector<int> seeds5({5,6,7,8});
@ -328,7 +330,7 @@ int main (int argc, char ** argv)
Gamma5R5HermitianLinearOperator<DomainWallFermionR,LatticeFermion> HermIndefOp(Ddwf);
Level1Op LDOp(*Coarse5d,1); LDOp.CoarsenOperator(FGrid,HermIndefOp,Aggregates);
Level1Op LDOp(*Coarse5d,*Coarse5dRB,1); LDOp.CoarsenOperator(FGrid,HermIndefOp,Aggregates);
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << " Running Coarse grid Lanczos "<< std::endl;
@ -352,7 +354,9 @@ int main (int argc, char ** argv)
// ConjugateGradient<CoarseVector> CoarseCG(0.01,1000);
ConjugateGradient<CoarseVector> CoarseCG(0.02,1000);// 14.7s
ConjugateGradient<CoarseVector> CoarseCG(0.01,2000);// 14.7s
eval.resize(0);
evec.resize(0,Coarse5d);
DeflatedGuesser<CoarseVector> DeflCoarseGuesser(evec,eval);
NormalEquations<CoarseVector> DeflCoarseCGNE(LDOp,CoarseCG,DeflCoarseGuesser);

397
tests/solver/Test_dwf_hdcr_48_rb.cc Normal file
View File

@ -0,0 +1,397 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./tests/Test_dwf_hdcr.cc
Copyright (C) 2015
Author: Antonin Portelli <antonin.portelli@me.com>
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: paboyle <paboyle@ph.ed.ac.uk>
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#include <Grid/Grid.h>
#include <Grid/algorithms/iterative/PrecGeneralisedConjugateResidual.h>
using namespace std;
using namespace Grid;
/* Params
* Grid:
* block1(4)
* block2(4)
*
* Subspace
* * Fine : Subspace(nbasis,hi,lo,order,first,step) -- 32, 60,0.02,500,100,100
* * Coarse: Subspace(nbasis,hi,lo,order,first,step) -- 32, 18,0.02,500,100,100
* Smoother:
* * Fine: Cheby(hi, lo, order) -- 60,0.5,10
* * Coarse: Cheby(hi, lo, order) -- 12,0.1,4
* Lanczos:
* CoarseCoarse IRL( Nk, Nm, Nstop, poly(lo,hi,order)) 24,36,24,0.002,4.0,61
*/
RealD InverseApproximation(RealD x){
return 1.0/x;
}
template<class Field> class SolverWrapper : public LinearFunction<Field> {
private:
CheckerBoardedSparseMatrixBase<Field> & _Matrix;
SchurRedBlackBase<Field> & _Solver;
public:
/////////////////////////////////////////////////////
// Wrap the usual normal equations trick
/////////////////////////////////////////////////////
SolverWrapper(CheckerBoardedSparseMatrixBase<Field> &Matrix,
SchurRedBlackBase<Field> &Solver)
: _Matrix(Matrix), _Solver(Solver) {};
void operator() (const Field &in, Field &out){
_Solver(_Matrix,in,out); // Mdag M out = Mdag in
}
};
template<class Field,class Matrix> class ChebyshevSmoother : public LinearFunction<Field>
{
public:
typedef LinearOperatorBase<Field> FineOperator;
Matrix & _SmootherMatrix;
FineOperator & _SmootherOperator;
Chebyshev<Field> Cheby;
ChebyshevSmoother(RealD _lo,RealD _hi,int _ord, FineOperator &SmootherOperator,Matrix &SmootherMatrix) :
_SmootherOperator(SmootherOperator),
_SmootherMatrix(SmootherMatrix),
Cheby(_lo,_hi,_ord,InverseApproximation)
{};
void operator() (const Field &in, Field &out)
{
Field tmp(in.Grid());
MdagMLinearOperator<Matrix,Field> MdagMOp(_SmootherMatrix);
_SmootherOperator.AdjOp(in,tmp);
Cheby(MdagMOp,tmp,out);
}
};
template<class Field,class Matrix> class MirsSmoother : public LinearFunction<Field>
{
public:
typedef LinearOperatorBase<Field> FineOperator;
Matrix & SmootherMatrix;
FineOperator & SmootherOperator;
RealD tol;
RealD shift;
int maxit;
MirsSmoother(RealD _shift,RealD _tol,int _maxit,FineOperator &_SmootherOperator,Matrix &_SmootherMatrix) :
shift(_shift),tol(_tol),maxit(_maxit),
SmootherOperator(_SmootherOperator),
SmootherMatrix(_SmootherMatrix)
{};
void operator() (const Field &in, Field &out)
{
ZeroGuesser<Field> Guess;
ConjugateGradient<Field> CG(tol,maxit,false);
Field src(in.Grid());
ShiftedMdagMLinearOperator<SparseMatrixBase<Field>,Field> MdagMOp(SmootherMatrix,shift);
SmootherOperator.AdjOp(in,src);
Guess(src,out);
CG(MdagMOp,src,out);
}
};
template<class Fobj,class CComplex,int nbasis, class Matrix, class Guesser, class CoarseSolver>
class MultiGridPreconditioner : public LinearFunction< Lattice<Fobj> > {
public:
typedef Aggregation<Fobj,CComplex,nbasis> Aggregates;
typedef CoarsenedMatrix<Fobj,CComplex,nbasis> CoarseOperator;
typedef typename Aggregation<Fobj,CComplex,nbasis>::CoarseVector CoarseVector;
typedef typename Aggregation<Fobj,CComplex,nbasis>::CoarseMatrix CoarseMatrix;
typedef typename Aggregation<Fobj,CComplex,nbasis>::FineField FineField;
typedef LinearOperatorBase<FineField> FineOperator;
typedef LinearFunction <FineField> FineSmoother;
Aggregates & _Aggregates;
CoarseOperator & _CoarseOperator;
Matrix & _FineMatrix;
FineOperator & _FineOperator;
Guesser & _Guess;
FineSmoother & _Smoother;
CoarseSolver & _CoarseSolve;
int level; void Level(int lv) {level = lv; };
#define GridLogLevel std::cout << GridLogMessage <<std::string(level,'\t')<< " Level "<<level <<" "
MultiGridPreconditioner(Aggregates &Agg, CoarseOperator &Coarse,
FineOperator &Fine,Matrix &FineMatrix,
FineSmoother &Smoother,
Guesser &Guess_,
CoarseSolver &CoarseSolve_)
: _Aggregates(Agg),
_CoarseOperator(Coarse),
_FineOperator(Fine),
_FineMatrix(FineMatrix),
_Smoother(Smoother),
_Guess(Guess_),
_CoarseSolve(CoarseSolve_),
level(1) { }
virtual void operator()(const FineField &in, FineField & out)
{
CoarseVector Csrc(_CoarseOperator.Grid());
CoarseVector Csol(_CoarseOperator.Grid());
FineField vec1(in.Grid());
FineField vec2(in.Grid());
double t;
// Fine Smoother
t=-usecond();
_Smoother(in,out);
t+=usecond();
GridLogLevel << "Smoother took "<< t/1000.0<< "ms" <<std::endl;
// Update the residual
_FineOperator.Op(out,vec1); sub(vec1, in ,vec1);
// Fine to Coarse
t=-usecond();
_Aggregates.ProjectToSubspace (Csrc,vec1);
t+=usecond();
GridLogLevel << "Project to coarse took "<< t/1000.0<< "ms" <<std::endl;
// Coarse correction
t=-usecond();
_CoarseSolve(Csrc,Csol);
t+=usecond();
GridLogLevel << "Coarse solve took "<< t/1000.0<< "ms" <<std::endl;
// Coarse to Fine
t=-usecond();
_Aggregates.PromoteFromSubspace(Csol,vec1);
add(out,out,vec1);
t+=usecond();
GridLogLevel << "Promote to this level took "<< t/1000.0<< "ms" <<std::endl;
// Residual
_FineOperator.Op(out,vec1); sub(vec1 ,in , vec1);
// Fine Smoother
t=-usecond();
_Smoother(vec1,vec2);
t+=usecond();
GridLogLevel << "Smoother took "<< t/1000.0<< "ms" <<std::endl;
add( out,out,vec2);
}
};
int main (int argc, char ** argv)
{
Grid_init(&argc,&argv);
const int Ls=24;
GridCartesian * UGrid = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());
GridRedBlackCartesian * UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
GridCartesian * FGrid = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid);
///////////////////////////////////////////////////
// Construct a coarsened grid; utility for this?
///////////////////////////////////////////////////
std::vector<int> block ({2,2,2,2});
//std::vector<int> block ({2,2,2,2});
const int nbasis= 40;
const int nbasisc= 40;
auto clatt = GridDefaultLatt();
for(int d=0;d<clatt.size();d++){
clatt[d] = clatt[d]/block[d];
}
GridCartesian *Coarse4d = SpaceTimeGrid::makeFourDimGrid(clatt, GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());;
GridCartesian *Coarse5d = SpaceTimeGrid::makeFiveDimGrid(1,Coarse4d);
// GridCartesian *CoarseCoarse4d = SpaceTimeGrid::makeFourDimGrid(cclatt, GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());;
// GridCartesian *CoarseCoarse5d = SpaceTimeGrid::makeFiveDimGrid(1,CoarseCoarse4d);
std::vector<int> seeds4({1,2,3,4});
std::vector<int> seeds5({5,6,7,8});
std::vector<int> cseeds({5,6,7,8});
GridParallelRNG RNG5(FGrid); RNG5.SeedFixedIntegers(seeds5);
GridParallelRNG RNG4(UGrid); RNG4.SeedFixedIntegers(seeds4);
GridParallelRNG CRNG(Coarse5d);CRNG.SeedFixedIntegers(cseeds);
LatticeFermion src(FGrid); gaussian(RNG5,src);// src=src+g5*src;
LatticeFermion result(FGrid);
LatticeGaugeField Umu(UGrid);
FieldMetaData header;
//std::string file("./ckpoint_lat.4000");
std::string file("./ckpoint_lat.1000");
NerscIO::readConfiguration(Umu,header,file);
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Building g5R5 hermitian DWF operator" <<std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
RealD mass=0.00078;
RealD M5=1.8;
DomainWallFermionR Ddwf(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
typedef Aggregation<vSpinColourVector,vTComplex,nbasis> Subspace;
typedef CoarsenedMatrix<vSpinColourVector,vTComplex,nbasis> CoarseOperator;
typedef CoarseOperator::CoarseVector CoarseVector;
typedef CoarseOperator::siteVector siteVector;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Calling Aggregation class to build subspace" <<std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
MdagMLinearOperator<DomainWallFermionR,LatticeFermion> HermDefOp(Ddwf);
Subspace Aggregates(Coarse5d,FGrid,0);
assert ( (nbasis & 0x1)==0);
{
int nb=nbasis/2;
LatticeFermion A(FGrid);
LatticeFermion B(FGrid);
// Aggregates.CreateSubspaceChebyshev(RNG5,HermDefOp,nb,60.0,0.002,1000,800,100,0.0);
// Aggregates.CreateSubspaceChebyshev(RNG5,HermDefOp,nb,60.0,0.02,1000,800,100,0.0);
Aggregates.CreateSubspaceChebyshev(RNG5,HermDefOp,nb,60.0,0.01,1000,100,100,0.0); // Slightly faster
for(int n=0;n<nb;n++){
std::cout << GridLogMessage << " G5R5 "<<n<<std::endl;
G5R5(Aggregates.subspace[n+nb],Aggregates.subspace[n]);
std::cout << GridLogMessage << " Projection "<<n<<std::endl;
A = Aggregates.subspace[n];
B = Aggregates.subspace[n+nb];
std::cout << GridLogMessage << " Copy "<<n<<std::endl;
Aggregates.subspace[n] = A+B; // 1+G5 // eigen value of G5R5 is +1
std::cout << GridLogMessage << " P+ "<<n<<std::endl;
Aggregates.subspace[n+nb]= A-B; // 1-G5 // eigen value of G5R5 is -1
std::cout << GridLogMessage << " P- "<<n<<std::endl;
}
}
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Building coarse representation of Indef operator" <<std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
typedef CoarsenedMatrix<vSpinColourVector,vTComplex,nbasis> Level1Op;
typedef CoarsenedMatrix<siteVector,iScalar<vTComplex>,nbasisc> Level2Op;
Gamma5R5HermitianLinearOperator<DomainWallFermionR,LatticeFermion> HermIndefOp(Ddwf);
GridRedBlackCartesian * Coarse4dRB = SpaceTimeGrid::makeFourDimRedBlackGrid(Coarse4d);
GridRedBlackCartesian * Coarse5dRB = SpaceTimeGrid::makeFiveDimRedBlackGrid(1,Coarse4d);
Level1Op LDOp(*Coarse5d,*Coarse5dRB,1); LDOp.CoarsenOperator(FGrid,HermIndefOp,Aggregates);
//////////////////////////////////////////////////
// Deflate the course space. Recursive multigrid?
//////////////////////////////////////////////////
typedef Aggregation<siteVector,iScalar<vTComplex>,nbasisc> CoarseSubspace;
// CoarseSubspace CoarseAggregates(CoarseCoarse5d,Coarse5d,0);
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Build deflation space in coarse operator "<< std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
MdagMLinearOperator<CoarseOperator,CoarseVector> PosdefLdop(LDOp);
typedef Level2Op::CoarseVector CoarseCoarseVector;
CoarseVector c_src(Coarse5d); c_src=1.0;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Building 3 level Multigrid "<< std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
typedef MultiGridPreconditioner<vSpinColourVector, vTComplex,nbasis, DomainWallFermionR,ZeroGuesser<CoarseVector> , SolverWrapper<CoarseVector> > TwoLevelMG;
typedef MultiGridPreconditioner<siteVector,iScalar<vTComplex>,nbasisc,Level1Op, DeflatedGuesser<CoarseCoarseVector>, NormalEquations<CoarseCoarseVector> > CoarseMG;
typedef MultiGridPreconditioner<vSpinColourVector, vTComplex,nbasis, DomainWallFermionR,ZeroGuesser<CoarseVector>, LinearFunction<CoarseVector> > ThreeLevelMG;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Calling 2 level Multigrid "<< std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::vector<RealD> tols({0.015});
std::vector<int> ords({12});
std::vector<RealD> los({0.8});
for(int l=0;l<los.size();l++){
for(int o=0;o<ords.size();o++){
for(int t=0;t<tols.size();t++){
result=Zero();
std::cout << GridLogMessage <<" tol " << tols[t] << " cheby order " <<ords[o]<< " lo "<<los[l] <<std::endl;
ChebyshevSmoother<LatticeFermion,DomainWallFermionR> FineSmoother(los[l],60.0,ords[o],HermIndefOp,Ddwf);
ZeroGuesser<CoarseVector> CoarseZeroGuesser;
ConjugateGradient<CoarseVector> CoarseCG(tols[t],10000);
SchurRedBlackDiagMooeeSolve<CoarseVector> CoarseRBCG(CoarseCG);
SolverWrapper<CoarseVector> CoarseSolver(LDOp,CoarseRBCG);
TwoLevelMG TwoLevelPrecon(Aggregates, LDOp,
HermIndefOp,Ddwf,
FineSmoother,
CoarseZeroGuesser,
CoarseSolver);
TwoLevelPrecon.Level(1);
PrecGeneralisedConjugateResidual<LatticeFermion> l1PGCR(1.0e-8,20,HermIndefOp,TwoLevelPrecon,16,16);
l1PGCR.Level(1);
l1PGCR(src,result);
}}}
ConjugateGradient<LatticeFermion> pCG(1.0e-8,60000);
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Calling red black CG "<< std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
result=Zero();
LatticeFermion src_o(FrbGrid);
LatticeFermion result_o(FrbGrid);
pickCheckerboard(Odd,src_o,src);
result_o=Zero();
SchurDiagMooeeOperator<DomainWallFermionR,LatticeFermion> HermOpEO(Ddwf);
pCG(HermOpEO,src_o,result_o);
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Calling CG "<< std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
result=Zero();
pCG(HermDefOp,src,result);
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << " Fine PowerMethod "<< std::endl;
PowerMethod<LatticeFermion> PM; PM(HermDefOp,src);
std::cout<<GridLogMessage << " Coarse PowerMethod "<< std::endl;
PowerMethod<CoarseVector> cPM; cPM(PosdefLdop,c_src);
// std::cout<<GridLogMessage << " CoarseCoarse PowerMethod "<< std::endl;
// PowerMethod<CoarseCoarseVector> ccPM; ccPM(IRLHermOpL2,cc_src);
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Done "<< std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
Grid_finalize();
}

473
tests/solver/Test_dwf_hdcr_48_regression.cc Normal file
View File

@ -0,0 +1,473 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./tests/Test_dwf_hdcr.cc
Copyright (C) 2015
Author: Antonin Portelli <antonin.portelli@me.com>
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: paboyle <paboyle@ph.ed.ac.uk>
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#include <Grid/Grid.h>
#include <Grid/algorithms/iterative/PrecGeneralisedConjugateResidual.h>
using namespace std;
using namespace Grid;
/* Params
* Grid:
* block1(4)
* block2(4)
*
* Subspace
* * Fine : Subspace(nbasis,hi,lo,order,first,step) -- 32, 60,0.02,500,100,100
* * Coarse: Subspace(nbasis,hi,lo,order,first,step) -- 32, 18,0.02,500,100,100
* Smoother:
* * Fine: Cheby(hi, lo, order) -- 60,0.5,10
* * Coarse: Cheby(hi, lo, order) -- 12,0.1,4
* Lanczos:
* CoarseCoarse IRL( Nk, Nm, Nstop, poly(lo,hi,order)) 24,36,24,0.002,4.0,61
*/
RealD InverseApproximation(RealD x){
return 1.0/x;
}
template<class Field,class Matrix> class ChebyshevSmoother : public LinearFunction<Field>
{
public:
typedef LinearOperatorBase<Field> FineOperator;
Matrix & _SmootherMatrix;
FineOperator & _SmootherOperator;
Chebyshev<Field> Cheby;
ChebyshevSmoother(RealD _lo,RealD _hi,int _ord, FineOperator &SmootherOperator,Matrix &SmootherMatrix) :
_SmootherOperator(SmootherOperator),
_SmootherMatrix(SmootherMatrix),
Cheby(_lo,_hi,_ord,InverseApproximation)
{};
void operator() (const Field &in, Field &out)
{
Field tmp(in.Grid());
MdagMLinearOperator<Matrix,Field> MdagMOp(_SmootherMatrix);
_SmootherOperator.AdjOp(in,tmp);
Cheby(MdagMOp,tmp,out);
}
};
template<class Field,class Matrix> class MirsSmoother : public LinearFunction<Field>
{
public:
typedef LinearOperatorBase<Field> FineOperator;
Matrix & SmootherMatrix;
FineOperator & SmootherOperator;
RealD tol;
RealD shift;
int maxit;
MirsSmoother(RealD _shift,RealD _tol,int _maxit,FineOperator &_SmootherOperator,Matrix &_SmootherMatrix) :
shift(_shift),tol(_tol),maxit(_maxit),
SmootherOperator(_SmootherOperator),
SmootherMatrix(_SmootherMatrix)
{};
void operator() (const Field &in, Field &out)
{
ZeroGuesser<Field> Guess;
ConjugateGradient<Field> CG(tol,maxit,false);
Field src(in.Grid());
ShiftedMdagMLinearOperator<SparseMatrixBase<Field>,Field> MdagMOp(SmootherMatrix,shift);
SmootherOperator.AdjOp(in,src);
Guess(src,out);
CG(MdagMOp,src,out);
}
};
template<class Fobj,class CComplex,int nbasis, class Matrix, class Guesser, class CoarseSolver>
class MultiGridPreconditioner : public LinearFunction< Lattice<Fobj> > {
public:
typedef Aggregation<Fobj,CComplex,nbasis> Aggregates;
typedef CoarsenedMatrix<Fobj,CComplex,nbasis> CoarseOperator;
typedef typename Aggregation<Fobj,CComplex,nbasis>::CoarseVector CoarseVector;
typedef typename Aggregation<Fobj,CComplex,nbasis>::CoarseMatrix CoarseMatrix;
typedef typename Aggregation<Fobj,CComplex,nbasis>::FineField FineField;
typedef LinearOperatorBase<FineField> FineOperator;
typedef LinearFunction <FineField> FineSmoother;
Aggregates & _Aggregates;
CoarseOperator & _CoarseOperator;
Matrix & _FineMatrix;
FineOperator & _FineOperator;
Guesser & _Guess;
FineSmoother & _Smoother;
CoarseSolver & _CoarseSolve;
int level; void Level(int lv) {level = lv; };
#define GridLogLevel std::cout << GridLogMessage <<std::string(level,'\t')<< " Level "<<level <<" "
MultiGridPreconditioner(Aggregates &Agg, CoarseOperator &Coarse,
FineOperator &Fine,Matrix &FineMatrix,
FineSmoother &Smoother,
Guesser &Guess_,
CoarseSolver &CoarseSolve_)
: _Aggregates(Agg),
_CoarseOperator(Coarse),
_FineOperator(Fine),
_FineMatrix(FineMatrix),
_Smoother(Smoother),
_Guess(Guess_),
_CoarseSolve(CoarseSolve_),
level(1) { }
virtual void operator()(const FineField &in, FineField & out)
{
CoarseVector Csrc(_CoarseOperator.Grid());
CoarseVector Csol(_CoarseOperator.Grid());
FineField vec1(in.Grid());
FineField vec2(in.Grid());
double t;
// Fine Smoother
t=-usecond();
_Smoother(in,out);
t+=usecond();
GridLogLevel << "Smoother took "<< t/1000.0<< "ms" <<std::endl;
// Update the residual
_FineOperator.Op(out,vec1); sub(vec1, in ,vec1);
// Fine to Coarse
t=-usecond();
_Aggregates.ProjectToSubspace (Csrc,vec1);
t+=usecond();
GridLogLevel << "Project to coarse took "<< t/1000.0<< "ms" <<std::endl;
// Coarse correction
t=-usecond();
_CoarseSolve(Csrc,Csol);
t+=usecond();
GridLogLevel << "Coarse solve took "<< t/1000.0<< "ms" <<std::endl;
// Coarse to Fine
t=-usecond();
_Aggregates.PromoteFromSubspace(Csol,vec1);
add(out,out,vec1);
t+=usecond();
GridLogLevel << "Promote to this level took "<< t/1000.0<< "ms" <<std::endl;
// Residual
_FineOperator.Op(out,vec1); sub(vec1 ,in , vec1);
// Fine Smoother
t=-usecond();
_Smoother(vec1,vec2);
t+=usecond();
GridLogLevel << "Smoother took "<< t/1000.0<< "ms" <<std::endl;
add( out,out,vec2);
}
};
int main (int argc, char ** argv)
{
Grid_init(&argc,&argv);
const int Ls=24;
GridCartesian * UGrid = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());
GridRedBlackCartesian * UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
GridCartesian * FGrid = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid);
///////////////////////////////////////////////////
// Construct a coarsened grid; utility for this?
///////////////////////////////////////////////////
std::vector<int> block ({2,2,2,2});
std::vector<int> blockc ({2,2,2,2});
const int nbasis= 40;
const int nbasisc= 40;
auto clatt = GridDefaultLatt();
for(int d=0;d<clatt.size();d++){
clatt[d] = clatt[d]/block[d];
}
auto cclatt = clatt;
for(int d=0;d<clatt.size();d++){
cclatt[d] = clatt[d]/blockc[d];
}
GridCartesian *Coarse4d = SpaceTimeGrid::makeFourDimGrid(clatt, GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());;
GridCartesian *Coarse5d = SpaceTimeGrid::makeFiveDimGrid(1,Coarse4d);
// GridCartesian *CoarseCoarse4d = SpaceTimeGrid::makeFourDimGrid(cclatt, GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());;
// GridCartesian *CoarseCoarse5d = SpaceTimeGrid::makeFiveDimGrid(1,CoarseCoarse4d);
std::vector<int> seeds4({1,2,3,4});
std::vector<int> seeds5({5,6,7,8});
std::vector<int> cseeds({5,6,7,8});
GridParallelRNG RNG5(FGrid); RNG5.SeedFixedIntegers(seeds5);
GridParallelRNG RNG4(UGrid); RNG4.SeedFixedIntegers(seeds4);
GridParallelRNG CRNG(Coarse5d);CRNG.SeedFixedIntegers(cseeds);
LatticeFermion src(FGrid); gaussian(RNG5,src);// src=src+g5*src;
LatticeFermion result(FGrid);
LatticeGaugeField Umu(UGrid);
FieldMetaData header;
// std::string file("./ckpoint_lat.4000");
std::string file("./ckpoint_lat.1000");
NerscIO::readConfiguration(Umu,header,file);
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Building g5R5 hermitian DWF operator" <<std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
RealD mass=0.00078;
RealD M5=1.8;
DomainWallFermionR Ddwf(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
typedef Aggregation<vSpinColourVector,vTComplex,nbasis> Subspace;
typedef CoarsenedMatrix<vSpinColourVector,vTComplex,nbasis> CoarseOperator;
typedef CoarseOperator::CoarseVector CoarseVector;
typedef CoarseOperator::siteVector siteVector;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Calling Aggregation class to build subspace" <<std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
MdagMLinearOperator<DomainWallFermionR,LatticeFermion> HermDefOp(Ddwf);
Subspace Aggregates(Coarse5d,FGrid,0);
assert ( (nbasis & 0x1)==0);
{
int nb=nbasis/2;
LatticeFermion A(FGrid);
LatticeFermion B(FGrid);
// Aggregates.CreateSubspaceChebyshev(RNG5,HermDefOp,nb,60.0,0.002,1000,800,100,0.0);
// Aggregates.CreateSubspaceChebyshev(RNG5,HermDefOp,nb,60.0,0.02,1000,800,100,0.0);
Aggregates.CreateSubspaceChebyshev(RNG5,HermDefOp,nb,60.0,0.01,1000,100,100,0.0); // Slightly faster
for(int n=0;n<nb;n++){
std::cout << GridLogMessage << " G5R5 "<<n<<std::endl;
G5R5(Aggregates.subspace[n+nb],Aggregates.subspace[n]);
std::cout << GridLogMessage << " Projection "<<n<<std::endl;
A = Aggregates.subspace[n];
B = Aggregates.subspace[n+nb];
std::cout << GridLogMessage << " Copy "<<n<<std::endl;
Aggregates.subspace[n] = A+B; // 1+G5 // eigen value of G5R5 is +1
std::cout << GridLogMessage << " P+ "<<n<<std::endl;
Aggregates.subspace[n+nb]= A-B; // 1-G5 // eigen value of G5R5 is -1
std::cout << GridLogMessage << " P- "<<n<<std::endl;
}
}
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Building coarse representation of Indef operator" <<std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
typedef CoarsenedMatrix<vSpinColourVector,vTComplex,nbasis> Level1Op;
typedef CoarsenedMatrix<siteVector,iScalar<vTComplex>,nbasisc> Level2Op;
Gamma5R5HermitianLinearOperator<DomainWallFermionR,LatticeFermion> HermIndefOp(Ddwf);
GridRedBlackCartesian * Coarse4dRB = SpaceTimeGrid::makeFourDimRedBlackGrid(Coarse4d);
std::cout << " Making 5D coarse RB grid " <<std::endl;
GridRedBlackCartesian * Coarse5dRB = SpaceTimeGrid::makeFiveDimRedBlackGrid(1,Coarse4d);
std::cout << " Made 5D coarse RB grid " <<std::endl;
Level1Op LDOp(*Coarse5d,*Coarse5dRB,1); LDOp.CoarsenOperator(FGrid,HermIndefOp,Aggregates);
//////////////////////////////////////////////////
// Deflate the course space. Recursive multigrid?
//////////////////////////////////////////////////
typedef Aggregation<siteVector,iScalar<vTComplex>,nbasisc> CoarseSubspace;
// CoarseSubspace CoarseAggregates(CoarseCoarse5d,Coarse5d,0);
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Build deflation space in coarse operator "<< std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
MdagMLinearOperator<CoarseOperator,CoarseVector> PosdefLdop(LDOp);
/*
{
int nb=nbasisc/2;
CoarseAggregates.CreateSubspaceChebyshev(CRNG,PosdefLdop,nb,15.0,0.02,1000,800,100,0.0);
for(int n=0;n<nb;n++){
autoView( subspace , CoarseAggregates.subspace[n],CpuWrite);
autoView( subspace_g5, CoarseAggregates.subspace[n+nb],CpuWrite);
for(int nn=0;nn<nb;nn++){
for(int site=0;site<Coarse5d->oSites();site++){
subspace_g5[site](nn) = subspace[site](nn);
subspace_g5[site](nn+nb)=-subspace[site](nn+nb);
}
}
}
}
*/
typedef Level2Op::CoarseVector CoarseCoarseVector;
/*
Level2Op L2Op(*CoarseCoarse5d,1); // Hermitian matrix
HermitianLinearOperator<Level1Op,CoarseVector> L1LinOp(LDOp);
L2Op.CoarsenOperator(Coarse5d,L1LinOp,CoarseAggregates);
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << " Running CoarseCoarse grid Lanczos "<< std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
MdagMLinearOperator<Level2Op,CoarseCoarseVector> IRLHermOpL2(L2Op);
CoarseCoarseVector cc_src(CoarseCoarse5d); cc_src=1.0;
*/
/*
Chebyshev<CoarseCoarseVector> IRLChebyL2(0.001,15.0,301);
FunctionHermOp<CoarseCoarseVector> IRLOpChebyL2(IRLChebyL2,IRLHermOpL2);
PlainHermOp<CoarseCoarseVector> IRLOpL2 (IRLHermOpL2);
int cNk=24;
int cNm=36;
int cNstop=24;
ImplicitlyRestartedLanczos<CoarseCoarseVector> IRLL2(IRLOpChebyL2,IRLOpL2,cNstop,cNk,cNm,1.0e-3,20);
int cNconv;
std::vector<RealD> eval2(cNm);
std::vector<CoarseCoarseVector> evec2(cNm,CoarseCoarse5d);
IRLL2.calc(eval2,evec2,cc_src,cNconv);
ConjugateGradient<CoarseCoarseVector> CoarseCoarseCG(0.1,1000);
DeflatedGuesser<CoarseCoarseVector> DeflCoarseCoarseGuesser(evec2,eval2);
NormalEquations<CoarseCoarseVector> DeflCoarseCoarseCGNE(L2Op,CoarseCoarseCG,DeflCoarseCoarseGuesser);
*/
/*
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << " Running Coarse grid Lanczos "<< std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
MdagMLinearOperator<Level1Op,CoarseVector> IRLHermOp(LDOp);
// Chebyshev<CoarseVector> IRLCheby(0.001,15.0,301);
Chebyshev<CoarseVector> IRLCheby(0.03,12.0,101);
FunctionHermOp<CoarseVector> IRLOpCheby(IRLCheby,IRLHermOp);
PlainHermOp<CoarseVector> IRLOp (IRLHermOp);
int Nk=64;
int Nm=128;
int Nstop=Nk;
ImplicitlyRestartedLanczos<CoarseVector> IRL(IRLOpCheby,IRLOp,Nstop,Nk,Nm,1.0e-3,20);
int Nconv;
std::vector<RealD> eval(Nm);
std::vector<CoarseVector> evec(Nm,Coarse5d);
IRL.calc(eval,evec,c_src,Nconv);
*/
CoarseVector c_src(Coarse5d); c_src=1.0;
// DeflatedGuesser<CoarseVector> DeflCoarseGuesser(evec,eval);
// NormalEquations<CoarseVector> DeflCoarseCGNE(LDOp,CoarseCG,DeflCoarseGuesser);
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Building 3 level Multigrid "<< std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
// typedef MultiGridPreconditioner<vSpinColourVector, vTComplex,nbasis, DomainWallFermionR,DeflatedGuesser<CoarseVector> , NormalEquations<CoarseVector> > TwoLevelMG;
typedef MultiGridPreconditioner<vSpinColourVector, vTComplex,nbasis, DomainWallFermionR,ZeroGuesser<CoarseVector> , NormalEquations<CoarseVector> > TwoLevelMG;
typedef MultiGridPreconditioner<siteVector,iScalar<vTComplex>,nbasisc,Level1Op, DeflatedGuesser<CoarseCoarseVector>, NormalEquations<CoarseCoarseVector> > CoarseMG;
typedef MultiGridPreconditioner<vSpinColourVector, vTComplex,nbasis, DomainWallFermionR,ZeroGuesser<CoarseVector>, LinearFunction<CoarseVector> > ThreeLevelMG;
ChebyshevSmoother<LatticeFermion,DomainWallFermionR> FineSmoother(0.25,60.0,12,HermIndefOp,Ddwf);
/*
// MultiGrid preconditioner acting on the coarse space <-> coarsecoarse space
ChebyshevSmoother<CoarseVector, Level1Op > CoarseSmoother(0.1,15.0,3,L1LinOp,LDOp);
// MirsSmoother<CoarseVector, Level1Op > CoarseCGSmoother(0.1,0.1,4,L1LinOp,LDOp);
// MirsSmoother<LatticeFermion,DomainWallFermionR> FineCGSmoother(0.0,0.01,8,HermIndefOp,Ddwf);
CoarseMG Level2Precon (CoarseAggregates, L2Op,
L1LinOp,LDOp,
CoarseSmoother,
DeflCoarseCoarseGuesser,
DeflCoarseCoarseCGNE);
Level2Precon.Level(2);
// PGCR Applying this solver to solve the coarse space problem
PrecGeneralisedConjugateResidual<CoarseVector> l2PGCR(0.1, 100, L1LinOp,Level2Precon,16,16);
l2PGCR.Level(2);
// Wrap the 2nd level solver in a MultiGrid preconditioner acting on the fine space
ZeroGuesser<CoarseVector> CoarseZeroGuesser;
ThreeLevelMG ThreeLevelPrecon(Aggregates, LDOp,
HermIndefOp,Ddwf,
FineSmoother,
CoarseZeroGuesser,
l2PGCR);
ThreeLevelPrecon.Level(1);
// Apply the fine-coarse-coarsecoarse 2 deep MG preconditioner in an outer PGCR on the fine fgrid
PrecGeneralisedConjugateResidual<LatticeFermion> l1PGCR(1.0e-8,1000,HermIndefOp,ThreeLevelPrecon,16,16);
l1PGCR.Level(1);
*/
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Calling 2 level Multigrid "<< std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
result=Zero();
ZeroGuesser<CoarseVector> CoarseZeroGuesser;
ConjugateGradient<CoarseVector> CoarseCG(0.01,1000);
NormalEquations<CoarseVector> CoarseCGNE(LDOp,CoarseCG,CoarseZeroGuesser);
TwoLevelMG TwoLevelPrecon(Aggregates, LDOp,
HermIndefOp,Ddwf,
FineSmoother,
CoarseZeroGuesser,
CoarseCGNE);
TwoLevelPrecon.Level(1);
PrecGeneralisedConjugateResidual<LatticeFermion> l1PGCR(1.0e-8,20,HermIndefOp,TwoLevelPrecon,16,16);
l1PGCR.Level(1);
l1PGCR(src,result);
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Calling CG "<< std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
ConjugateGradient<LatticeFermion> pCG(1.0e-8,60000);
result=Zero();
// pCG(HermDefOp,src,result);
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Calling red black CG "<< std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
result=Zero();
LatticeFermion src_o(FrbGrid);
LatticeFermion result_o(FrbGrid);
pickCheckerboard(Odd,src_o,src);
result_o=Zero();
SchurDiagMooeeOperator<DomainWallFermionR,LatticeFermion> HermOpEO(Ddwf);
pCG(HermOpEO,src_o,result_o);
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << " Fine PowerMethod "<< std::endl;
PowerMethod<LatticeFermion> PM; PM(HermDefOp,src);
std::cout<<GridLogMessage << " Coarse PowerMethod "<< std::endl;
PowerMethod<CoarseVector> cPM; cPM(PosdefLdop,c_src);
// std::cout<<GridLogMessage << " CoarseCoarse PowerMethod "<< std::endl;
// PowerMethod<CoarseCoarseVector> ccPM; ccPM(IRLHermOpL2,cc_src);
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
std::cout<<GridLogMessage << "Done "<< std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
Grid_finalize();
}

9
tests/solver/Test_dwf_multigrid.cc
View File

@ -370,6 +370,11 @@ int main (int argc, char ** argv)
GridCartesian *CoarseCoarse4d = SpaceTimeGrid::makeFourDimGrid(cclatt, GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());;
GridCartesian *CoarseCoarse5d = SpaceTimeGrid::makeFiveDimGrid(1,CoarseCoarse4d);
GridRedBlackCartesian * Coarse4dRB = SpaceTimeGrid::makeFourDimRedBlackGrid(Coarse4d);
GridRedBlackCartesian * Coarse5dRB = SpaceTimeGrid::makeFiveDimRedBlackGrid(1,Coarse4d);
GridRedBlackCartesian *CoarseCoarse4dRB = SpaceTimeGrid::makeFourDimRedBlackGrid(CoarseCoarse4d);
GridRedBlackCartesian *CoarseCoarse5dRB = SpaceTimeGrid::makeFiveDimRedBlackGrid(1,CoarseCoarse4d);
std::vector<int> seeds4({1,2,3,4});
std::vector<int> seeds5({5,6,7,8});
std::vector<int> cseeds({5,6,7,8});
@ -434,8 +439,8 @@ int main (int argc, char ** argv)
std::cout<<GridLogMessage << "Building coarse representation of Indef operator" <<std::endl;
std::cout<<GridLogMessage << "**************************************************"<< std::endl;
Level1Op LDOp(*Coarse5d,1); LDOp.CoarsenOperator(FGrid,HermIndefOp,Aggregates);
Level1Op LDOpPV(*Coarse5d,1); LDOpPV.CoarsenOperator(FGrid,HermIndefOpPV,Aggregates);
Level1Op LDOp(*Coarse5d,*Coarse5dRB,1); LDOp.CoarsenOperator(FGrid,HermIndefOp,Aggregates);
Level1Op LDOpPV(*Coarse5d,*Coarse5dRB,1); LDOpPV.CoarsenOperator(FGrid,HermIndefOpPV,Aggregates);
std::cout<<GridLogMessage << "**************************************************"<< std::endl;

4
tests/solver/Test_hw_multigrid.cc
View File

@ -274,6 +274,8 @@ int main (int argc, char ** argv)
GridCartesian *Coarse4d = SpaceTimeGrid::makeFourDimGrid(clatt, GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());;
GridCartesian *Coarse5d = SpaceTimeGrid::makeFiveDimGrid(Ls,Coarse4d);
GridRedBlackCartesian * Coarse4dRB = SpaceTimeGrid::makeFourDimRedBlackGrid(Coarse4d);
GridRedBlackCartesian * Coarse5dRB = SpaceTimeGrid::makeFiveDimRedBlackGrid(1,Coarse4d);
std::vector<int> seeds({1,2,3,4});
GridParallelRNG RNG5(FGrid); RNG5.SeedFixedIntegers(seeds);
@ -335,7 +337,7 @@ int main (int argc, char ** argv)
NonHermitianLinearOperator<DomainWallFermionR,LatticeFermion> LinOpDwf(Ddwf);
Level1Op LDOp (*Coarse5d,0);
Level1Op LDOp (*Coarse5d,*Coarse5dRB,0);
std::cout<<GridLogMessage << " Callinig Coarsen the operator " <<std::endl;
LDOp.CoarsenOperator(FGrid,LinOpDwf,Aggregates5D);

1287
tests/solver/Test_hw_multigrid_mixed_48.cc Normal file
View File

File diff suppressed because it is too large Load Diff

1326
tests/solver/Test_hw_multigrid_mixed_48_rb.cc Normal file
View File

File diff suppressed because it is too large Load Diff