portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-07-30 11:17:07 +01:00
Code Releases Activity

Merge branch 'develop' into feature/feynman-rules

Browse Source 
# Conflicts:
#	lib/Threads.h
#	lib/qcd/action/fermion/WilsonFermion.cc
#	lib/qcd/action/fermion/WilsonFermion.h
#	lib/qcd/utils/SUn.h
#	lib/simd/Grid_avx.h
#	lib/simd/Intel512common.h
This commit is contained in:
Antonin Portelli
2016-10-19 18:35:18 +01:00
parent a123dcd7e9 7af9b87318
commit 997fd882ff
84 changed files with 6162 additions and 2851 deletions
Download Patch File Download Diff File Expand all files Collapse all files

510
lib/qcd/action/fermion/FermionOperatorImpl.h
View File

@@ -1,35 +1,36 @@
/*************************************************************************************
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/action/fermion/FermionOperatorImpl.h
Source file: ./lib/qcd/action/fermion/FermionOperatorImpl.h
Copyright (C) 2015
Copyright (C) 2015
Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
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 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.
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.
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 GRID_QCD_FERMION_OPERATOR_IMPL_H
#define GRID_QCD_FERMION_OPERATOR_IMPL_H
See the full license in the file "LICENSE" in the top level distribution
directory
*************************************************************************************/
/* END LEGAL */
#ifndef GRID_QCD_FERMION_OPERATOR_IMPL_H
#define GRID_QCD_FERMION_OPERATOR_IMPL_H
namespace Grid {
@@ -99,258 +100,281 @@ namespace Grid {
typedef typename Impl::SiteSpinor SiteSpinor; \
typedef typename Impl::SiteHalfSpinor SiteHalfSpinor; \
typedef typename Impl::Compressor Compressor; \
typedef typename Impl::StencilImpl StencilImpl; \
typedef typename Impl::ImplParams ImplParams; \
typedef typename Impl::StencilImpl StencilImpl; \
typedef typename Impl::ImplParams ImplParams; \
typedef typename Impl::Coeff_t Coeff_t;
#define INHERIT_IMPL_TYPES(Base) \
INHERIT_GIMPL_TYPES(Base)\
INHERIT_GIMPL_TYPES(Base) \
INHERIT_FIMPL_TYPES(Base)
///////
// Single flavour four spinors with colour index
///////
template<class S,int Nrepresentation=Nc,class _Coeff_t = RealD>
class WilsonImpl : public PeriodicGaugeImpl< GaugeImplTypes< S, Nrepresentation> > {
template <class S, class Representation = FundamentalRepresentation,class _Coeff_t = RealD >
class WilsonImpl
: public PeriodicGaugeImpl<GaugeImplTypes<S, Representation::Dimension > > {
public:
static const int Dimension = Representation::Dimension;
typedef PeriodicGaugeImpl<GaugeImplTypes<S, Dimension > > Gimpl;
//Necessary?
constexpr bool is_fundamental() const{return Dimension == Nc ? 1 : 0;}
const bool LsVectorised=false;
typedef _Coeff_t Coeff_t;
typedef PeriodicGaugeImpl< GaugeImplTypes< S,Nrepresentation> > Gimpl;
INHERIT_GIMPL_TYPES(Gimpl);
template<typename vtype> using iImplSpinor = iScalar<iVector<iVector<vtype, Nrepresentation>, Ns> >;
template<typename vtype> using iImplHalfSpinor = iScalar<iVector<iVector<vtype, Nrepresentation>, Nhs> >;
template<typename vtype> using iImplDoubledGaugeField = iVector<iScalar<iMatrix<vtype, Nrepresentation> >, Nds >;
typedef iImplSpinor <Simd> SiteSpinor;
typedef iImplHalfSpinor<Simd> SiteHalfSpinor;
typedef iImplDoubledGaugeField<Simd> SiteDoubledGaugeField;
typedef Lattice<SiteSpinor> FermionField;
template <typename vtype> using iImplSpinor = iScalar<iVector<iVector<vtype, Dimension>, Ns> >;
template <typename vtype> using iImplHalfSpinor = iScalar<iVector<iVector<vtype, Dimension>, Nhs> >;
template <typename vtype> using iImplDoubledGaugeField = iVector<iScalar<iMatrix<vtype, Dimension> >, Nds>;
typedef iImplSpinor<Simd> SiteSpinor;
typedef iImplHalfSpinor<Simd> SiteHalfSpinor;
typedef iImplDoubledGaugeField<Simd> SiteDoubledGaugeField;
typedef Lattice<SiteSpinor> FermionField;
typedef Lattice<SiteDoubledGaugeField> DoubledGaugeField;
typedef WilsonCompressor<SiteHalfSpinor,SiteSpinor> Compressor;
typedef WilsonCompressor<SiteHalfSpinor, SiteSpinor> Compressor;
typedef WilsonImplParams ImplParams;
typedef WilsonStencil<SiteSpinor,SiteHalfSpinor> StencilImpl;
typedef WilsonStencil<SiteSpinor, SiteHalfSpinor> StencilImpl;
ImplParams Params;
WilsonImpl(const ImplParams &p= ImplParams()) : Params(p) {};
WilsonImpl(const ImplParams &p = ImplParams()) : Params(p){};
bool overlapCommsCompute(void) { return Params.overlapCommsCompute; };
inline void multLink(SiteHalfSpinor &phi,const SiteDoubledGaugeField &U,const SiteHalfSpinor &chi,int mu,StencilEntry *SE,StencilImpl &St){
mult(&phi(),&U(mu),&chi());
}
template<class ref>
inline void loadLinkElement(Simd & reg,ref &memory){
reg = memory;
}
inline void DoubleStore(GridBase *GaugeGrid,DoubledGaugeField &Uds,const GaugeField &Umu)
{
conformable(Uds._grid,GaugeGrid);
conformable(Umu._grid,GaugeGrid);
GaugeLinkField U(GaugeGrid);
for(int mu=0;mu<Nd;mu++){
U = PeekIndex<LorentzIndex>(Umu,mu);
PokeIndex<LorentzIndex>(Uds,U,mu);
U = adj(Cshift(U,mu,-1));
PokeIndex<LorentzIndex>(Uds,U,mu+4);
}
inline void multLink(SiteHalfSpinor &phi,
const SiteDoubledGaugeField &U,
const SiteHalfSpinor &chi,
int mu,
StencilEntry *SE,
StencilImpl &St) {
mult(&phi(), &U(mu), &chi());
}
template <class ref>
inline void loadLinkElement(Simd &reg,
ref &memory) {
reg = memory;
}
inline void DoubleStore(GridBase *GaugeGrid,
DoubledGaugeField &Uds,
const GaugeField &Umu) {
conformable(Uds._grid, GaugeGrid);
conformable(Umu._grid, GaugeGrid);
GaugeLinkField U(GaugeGrid);
for (int mu = 0; mu < Nd; mu++) {
U = PeekIndex<LorentzIndex>(Umu, mu);
PokeIndex<LorentzIndex>(Uds, U, mu);
U = adj(Cshift(U, mu, -1));
PokeIndex<LorentzIndex>(Uds, U, mu + 4);
}
}
inline void InsertForce4D(GaugeField &mat, FermionField &Btilde, FermionField &A,int mu){
GaugeLinkField link(mat._grid);
link = TraceIndex<SpinIndex>(outerProduct(Btilde,A));
PokeIndex<LorentzIndex>(mat,link,mu);
}
inline void InsertForce5D(GaugeField &mat, FermionField &Btilde, FermionField &Atilde,int mu){
int Ls=Btilde._grid->_fdimensions[0];
GaugeLinkField tmp(mat._grid);
tmp = zero;
PARALLEL_FOR_LOOP
for(int sss=0;sss<tmp._grid->oSites();sss++){
int sU=sss;
for(int s=0;s<Ls;s++){
int sF = s+Ls*sU;
tmp[sU] = tmp[sU]+ traceIndex<SpinIndex>(outerProduct(Btilde[sF],Atilde[sF])); // ordering here
PARALLEL_FOR_LOOP
for(int sss=0;sss<tmp._grid->oSites();sss++){
int sU=sss;
for(int s=0;s<Ls;s++){
int sF = s+Ls*sU;
tmp[sU] = tmp[sU]+ traceIndex<SpinIndex>(outerProduct(Btilde[sF],Atilde[sF])); // ordering here
}
}
}
PokeIndex<LorentzIndex>(mat,tmp,mu);
}
};
///////
// Single flavour four spinors with colour index, 5d redblack
///////
template<class S,int Nrepresentation=Nc,class _Coeff_t = RealD>
class DomainWallVec5dImpl : public PeriodicGaugeImpl< GaugeImplTypes< S,Nrepresentation> > {
public:
static const int Dimension = Nrepresentation;
const bool LsVectorised=true;
typedef _Coeff_t Coeff_t;
typedef PeriodicGaugeImpl< GaugeImplTypes< S,Nrepresentation> > Gimpl;
typedef _Coeff_t Coeff_t;
typedef PeriodicGaugeImpl<GaugeImplTypes<S, Nrepresentation> > Gimpl;
INHERIT_GIMPL_TYPES(Gimpl);
template<typename vtype> using iImplSpinor = iScalar<iVector<iVector<vtype, Nrepresentation>, Ns> >;
template<typename vtype> using iImplHalfSpinor = iScalar<iVector<iVector<vtype, Nrepresentation>, Nhs> >;
template<typename vtype> using iImplDoubledGaugeField = iVector<iScalar<iMatrix<vtype, Nrepresentation> >, Nds >;
template<typename vtype> using iImplGaugeField = iVector<iScalar<iMatrix<vtype, Nrepresentation> >, Nd >;
template<typename vtype> using iImplGaugeLink = iScalar<iScalar<iMatrix<vtype, Nrepresentation> > >;
typedef iImplSpinor <Simd> SiteSpinor;
typedef iImplHalfSpinor<Simd> SiteHalfSpinor;
typedef Lattice<SiteSpinor> FermionField;
template <typename vtype> using iImplSpinor = iScalar<iVector<iVector<vtype, Nrepresentation>, Ns> >;
template <typename vtype> using iImplHalfSpinor = iScalar<iVector<iVector<vtype, Nrepresentation>, Nhs> >;
template <typename vtype> using iImplDoubledGaugeField = iVector<iScalar<iMatrix<vtype, Nrepresentation> >, Nds>;
template <typename vtype> using iImplGaugeField = iVector<iScalar<iMatrix<vtype, Nrepresentation> >, Nd>;
template <typename vtype> using iImplGaugeLink = iScalar<iScalar<iMatrix<vtype, Nrepresentation> > >;
typedef iImplSpinor<Simd> SiteSpinor;
typedef iImplHalfSpinor<Simd> SiteHalfSpinor;
typedef Lattice<SiteSpinor> FermionField;
// Make the doubled gauge field a *scalar*
typedef iImplDoubledGaugeField<typename Simd::scalar_type> SiteDoubledGaugeField; // This is a scalar
typedef iImplGaugeField<typename Simd::scalar_type> SiteScalarGaugeField; // scalar
typedef iImplGaugeLink <typename Simd::scalar_type> SiteScalarGaugeLink; // scalar
typedef Lattice<SiteDoubledGaugeField> DoubledGaugeField;
typedef WilsonCompressor<SiteHalfSpinor,SiteSpinor> Compressor;
typedef iImplDoubledGaugeField<typename Simd::scalar_type>
SiteDoubledGaugeField; // This is a scalar
typedef iImplGaugeField<typename Simd::scalar_type>
SiteScalarGaugeField; // scalar
typedef iImplGaugeLink<typename Simd::scalar_type>
SiteScalarGaugeLink; // scalar
typedef Lattice<SiteDoubledGaugeField> DoubledGaugeField;
typedef WilsonCompressor<SiteHalfSpinor, SiteSpinor> Compressor;
typedef WilsonImplParams ImplParams;
typedef WilsonStencil<SiteSpinor,SiteHalfSpinor> StencilImpl;
typedef WilsonStencil<SiteSpinor, SiteHalfSpinor> StencilImpl;
ImplParams Params;
DomainWallVec5dImpl(const ImplParams &p= ImplParams()) : Params(p) {};
DomainWallVec5dImpl(const ImplParams &p = ImplParams()) : Params(p){};
bool overlapCommsCompute(void) { return false; };
template<class ref>
inline void loadLinkElement(Simd & reg,ref &memory){
vsplat(reg,memory);
template <class ref>
inline void loadLinkElement(Simd &reg, ref &memory) {
vsplat(reg, memory);
}
inline void multLink(SiteHalfSpinor &phi,const SiteDoubledGaugeField &U,const SiteHalfSpinor &chi,int mu,StencilEntry *SE,StencilImpl &St)
{
inline void multLink(SiteHalfSpinor &phi, const SiteDoubledGaugeField &U,
const SiteHalfSpinor &chi, int mu, StencilEntry *SE,
StencilImpl &St) {
SiteGaugeLink UU;
for(int i=0;i<Nrepresentation;i++){
for(int j=0;j<Nrepresentation;j++){
vsplat(UU()()(i,j),U(mu)()(i,j));
for (int i = 0; i < Nrepresentation; i++) {
for (int j = 0; j < Nrepresentation; j++) {
vsplat(UU()()(i, j), U(mu)()(i, j));
}
}
mult(&phi(),&UU(),&chi());
mult(&phi(), &UU(), &chi());
}
inline void DoubleStore(GridBase *GaugeGrid,DoubledGaugeField &Uds,const GaugeField &Umu)
{
SiteScalarGaugeField ScalarUmu;
inline void DoubleStore(GridBase *GaugeGrid, DoubledGaugeField &Uds,
const GaugeField &Umu) {
SiteScalarGaugeField ScalarUmu;
SiteDoubledGaugeField ScalarUds;
GaugeLinkField U (Umu._grid);
GaugeField Uadj(Umu._grid);
for(int mu=0;mu<Nd;mu++){
U = PeekIndex<LorentzIndex>(Umu,mu);
U = adj(Cshift(U,mu,-1));
PokeIndex<LorentzIndex>(Uadj,U,mu);
}
for(int lidx=0;lidx<GaugeGrid->lSites();lidx++){
std::vector<int> lcoor;
GaugeGrid->LocalIndexToLocalCoor(lidx,lcoor);
peekLocalSite(ScalarUmu,Umu,lcoor);
for(int mu=0;mu<4;mu++) ScalarUds(mu) = ScalarUmu(mu);
peekLocalSite(ScalarUmu,Uadj,lcoor);
for(int mu=0;mu<4;mu++) ScalarUds(mu+4) = ScalarUmu(mu);
pokeLocalSite(ScalarUds,Uds,lcoor);
}
}
inline void InsertForce4D(GaugeField &mat, FermionField &Btilde, FermionField &A,int mu){
assert(0);
}
inline void InsertForce5D(GaugeField &mat, FermionField &Btilde, FermionField &Atilde,int mu){
GaugeLinkField U(Umu._grid);
GaugeField Uadj(Umu._grid);
for (int mu = 0; mu < Nd; mu++) {
U = PeekIndex<LorentzIndex>(Umu, mu);
U = adj(Cshift(U, mu, -1));
PokeIndex<LorentzIndex>(Uadj, U, mu);
}
for (int lidx = 0; lidx < GaugeGrid->lSites(); lidx++) {
std::vector<int> lcoor;
GaugeGrid->LocalIndexToLocalCoor(lidx, lcoor);
peekLocalSite(ScalarUmu, Umu, lcoor);
for (int mu = 0; mu < 4; mu++) ScalarUds(mu) = ScalarUmu(mu);
peekLocalSite(ScalarUmu, Uadj, lcoor);
for (int mu = 0; mu < 4; mu++) ScalarUds(mu + 4) = ScalarUmu(mu);
pokeLocalSite(ScalarUds, Uds, lcoor);
}
}
inline void InsertForce4D(GaugeField &mat, FermionField &Btilde,
FermionField &A, int mu) {
assert(0);
}
inline void InsertForce5D(GaugeField &mat, FermionField &Btilde,
FermionField &Atilde, int mu) {
assert(0);
}
};
////////////////////////////////////////////////////////////////////////////////////////
// Flavour doubled spinors; is Gparity the only? what about C*?
////////////////////////////////////////////////////////////////////////////////////////
template<class S,int Nrepresentation,class _Coeff_t = RealD>
class GparityWilsonImpl : public ConjugateGaugeImpl< GaugeImplTypes<S,Nrepresentation> >{
template <class S, int Nrepresentation,class _Coeff_t = RealD>
class GparityWilsonImpl
: public ConjugateGaugeImpl<GaugeImplTypes<S, Nrepresentation> > {
public:
static const int Dimension = Nrepresentation;
const bool LsVectorised=false;
typedef _Coeff_t Coeff_t;
typedef ConjugateGaugeImpl< GaugeImplTypes<S,Nrepresentation> > Gimpl;
INHERIT_GIMPL_TYPES(Gimpl);
template<typename vtype> using iImplSpinor = iVector<iVector<iVector<vtype, Nrepresentation>, Ns>, Ngp >;
template<typename vtype> using iImplHalfSpinor = iVector<iVector<iVector<vtype, Nrepresentation>, Nhs>, Ngp >;
template<typename vtype> using iImplDoubledGaugeField = iVector<iVector<iScalar<iMatrix<vtype, Nrepresentation> >, Nds >, Ngp >;
typedef iImplSpinor <Simd> SiteSpinor;
typedef iImplHalfSpinor<Simd> SiteHalfSpinor;
typedef iImplDoubledGaugeField<Simd> SiteDoubledGaugeField;
typedef Lattice<SiteSpinor> FermionField;
template <typename vtype>
using iImplSpinor =
iVector<iVector<iVector<vtype, Nrepresentation>, Ns>, Ngp>;
template <typename vtype>
using iImplHalfSpinor =
iVector<iVector<iVector<vtype, Nrepresentation>, Nhs>, Ngp>;
template <typename vtype>
using iImplDoubledGaugeField =
iVector<iVector<iScalar<iMatrix<vtype, Nrepresentation> >, Nds>, Ngp>;
typedef iImplSpinor<Simd> SiteSpinor;
typedef iImplHalfSpinor<Simd> SiteHalfSpinor;
typedef iImplDoubledGaugeField<Simd> SiteDoubledGaugeField;
typedef Lattice<SiteSpinor> FermionField;
typedef Lattice<SiteDoubledGaugeField> DoubledGaugeField;
typedef WilsonCompressor<SiteHalfSpinor,SiteSpinor> Compressor;
typedef WilsonStencil<SiteSpinor,SiteHalfSpinor> StencilImpl;
typedef WilsonCompressor<SiteHalfSpinor, SiteSpinor> Compressor;
typedef WilsonStencil<SiteSpinor, SiteHalfSpinor> StencilImpl;
typedef GparityWilsonImplParams ImplParams;
ImplParams Params;
GparityWilsonImpl(const ImplParams &p= ImplParams()) : Params(p) {};
GparityWilsonImpl(const ImplParams &p = ImplParams()) : Params(p){};
bool overlapCommsCompute(void) { return Params.overlapCommsCompute; };
// provide the multiply by link that is differentiated between Gparity (with flavour index) and non-Gparity
inline void multLink(SiteHalfSpinor &phi,const SiteDoubledGaugeField &U,const SiteHalfSpinor &chi,int mu,StencilEntry *SE,StencilImpl &St){
// provide the multiply by link that is differentiated between Gparity (with
// flavour index) and non-Gparity
inline void multLink(SiteHalfSpinor &phi, const SiteDoubledGaugeField &U,
const SiteHalfSpinor &chi, int mu, StencilEntry *SE,
StencilImpl &St) {
typedef SiteHalfSpinor vobj;
typedef typename SiteHalfSpinor::scalar_object sobj;
vobj vtmp;
sobj stmp;
GridBase *grid = St._grid;
const int Nsimd = grid->Nsimd();
int direction = St._directions[mu];
int distance = St._distances[mu];
int ptype = St._permute_type[mu];
int sl = St._grid->_simd_layout[direction];
int direction = St._directions[mu];
int distance = St._distances[mu];
int ptype = St._permute_type[mu];
int sl = St._grid->_simd_layout[direction];
// Fixme X.Y.Z.T hardcode in stencil
int mmu = mu % Nd;
int mmu = mu % Nd;
// assert our assumptions
assert((distance==1)||(distance==-1)); // nearest neighbour stencil hard code
assert((sl==1)||(sl==2));
assert((distance == 1) || (distance == -1)); // nearest neighbour stencil hard code
assert((sl == 1) || (sl == 2));
std::vector<int> icoor;
if ( SE->_around_the_world && Params.twists[mmu] ) {
if ( sl == 2 ) {
@@ -391,7 +415,7 @@ PARALLEL_FOR_LOOP
mult(&phi(1),&U(1)(mu),&chi(1));
}
}
}
inline void DoubleStore(GridBase *GaugeGrid,DoubledGaugeField &Uds,const GaugeField &Umu)
{
@@ -404,7 +428,7 @@ PARALLEL_FOR_LOOP
GaugeLinkField Uconj(GaugeGrid);
Lattice<iScalar<vInteger> > coor(GaugeGrid);
for(int mu=0;mu<Nd;mu++){
@@ -412,19 +436,19 @@ PARALLEL_FOR_LOOP
U = PeekIndex<LorentzIndex>(Umu,mu);
Uconj = conjugate(U);
// This phase could come from a simple bc 1,1,-1,1 ..
int neglink = GaugeGrid->GlobalDimensions()[mu]-1;
if ( Params.twists[mu] ) {
Uconj = where(coor==neglink,-Uconj,Uconj);
}
PARALLEL_FOR_LOOP
for(auto ss=U.begin();ss<U.end();ss++){
Uds[ss](0)(mu) = U[ss]();
Uds[ss](1)(mu) = Uconj[ss]();
}
PARALLEL_FOR_LOOP
for(auto ss=U.begin();ss<U.end();ss++){
Uds[ss](0)(mu) = U[ss]();
Uds[ss](1)(mu) = Uconj[ss]();
}
U = adj(Cshift(U ,mu,-1)); // correct except for spanning the boundary
Uconj = adj(Cshift(Uconj,mu,-1));
@@ -434,80 +458,86 @@ PARALLEL_FOR_LOOP
Utmp = where(coor==0,Uconj,Utmp);
}
PARALLEL_FOR_LOOP
for(auto ss=U.begin();ss<U.end();ss++){
Uds[ss](0)(mu+4) = Utmp[ss]();
}
PARALLEL_FOR_LOOP
for(auto ss=U.begin();ss<U.end();ss++){
Uds[ss](0)(mu+4) = Utmp[ss]();
}
Utmp = Uconj;
if ( Params.twists[mu] ) {
Utmp = where(coor==0,U,Utmp);
}
PARALLEL_FOR_LOOP
for(auto ss=U.begin();ss<U.end();ss++){
Uds[ss](1)(mu+4) = Utmp[ss]();
}
PARALLEL_FOR_LOOP
for(auto ss=U.begin();ss<U.end();ss++){
Uds[ss](1)(mu+4) = Utmp[ss]();
}
}
}
inline void InsertForce4D(GaugeField &mat, FermionField &Btilde, FermionField &A,int mu){
inline void InsertForce4D(GaugeField &mat, FermionField &Btilde,
FermionField &A, int mu) {
// DhopDir provides U or Uconj depending on coor/flavour.
GaugeLinkField link(mat._grid);
// use lorentz for flavour as hack.
PARALLEL_FOR_LOOP
for(auto ss=link.begin();ss<link.end();ss++){
auto ttmp = traceIndex<SpinIndex>(outerProduct(Btilde[ss],A[ss]));
link[ss]() = ttmp(0,0) + conjugate(ttmp(1,1)) ;
}
PokeIndex<LorentzIndex>(mat,link,mu);
auto tmp = TraceIndex<SpinIndex>(outerProduct(Btilde, A));
PARALLEL_FOR_LOOP
for (auto ss = tmp.begin(); ss < tmp.end(); ss++) {
link[ss]() = tmp[ss](0, 0) - conjugate(tmp[ss](1, 1));
}
PokeIndex<LorentzIndex>(mat, link, mu);
return;
}
inline void InsertForce5D(GaugeField &mat, FermionField &Btilde, FermionField &Atilde,int mu){
int Ls=Btilde._grid->_fdimensions[0];
inline void InsertForce5D(GaugeField &mat, FermionField &Btilde,
FermionField &Atilde, int mu) {
int Ls = Btilde._grid->_fdimensions[0];
GaugeLinkField tmp(mat._grid);
tmp = zero;
PARALLEL_FOR_LOOP
for(int ss=0;ss<tmp._grid->oSites();ss++){
for(int s=0;s<Ls;s++){
int sF = s+Ls*ss;
auto ttmp = traceIndex<SpinIndex>(outerProduct(Btilde[sF],Atilde[sF]));
tmp[ss]() = tmp[ss]()+ ttmp(0,0) + conjugate(ttmp(1,1));
PARALLEL_FOR_LOOP
for (int ss = 0; ss < tmp._grid->oSites(); ss++) {
for (int s = 0; s < Ls; s++) {
int sF = s + Ls * ss;
auto ttmp = traceIndex<SpinIndex>(outerProduct(Btilde[sF], Atilde[sF]));
tmp[ss]() = tmp[ss]() + ttmp(0, 0) + conjugate(ttmp(1, 1));
}
}
}
PokeIndex<LorentzIndex>(mat,tmp,mu);
PokeIndex<LorentzIndex>(mat, tmp, mu);
return;
}
};
typedef WilsonImpl<vComplex ,Nc> WilsonImplR; // Real.. whichever prec
typedef WilsonImpl<vComplexF,Nc> WilsonImplF; // Float
typedef WilsonImpl<vComplexD,Nc> WilsonImplD; // Double
typedef WilsonImpl<vComplex, FundamentalRepresentation > WilsonImplR; // Real.. whichever prec
typedef WilsonImpl<vComplexF, FundamentalRepresentation > WilsonImplF; // Float
typedef WilsonImpl<vComplexD, FundamentalRepresentation > WilsonImplD; // Double
typedef WilsonImpl<vComplex ,Nc,ComplexD> ZWilsonImplR; // Real.. whichever prec
typedef WilsonImpl<vComplexF,Nc,ComplexD> ZWilsonImplF; // Float
typedef WilsonImpl<vComplexD,Nc,ComplexD> ZWilsonImplD; // Double
typedef WilsonImpl<vComplex, FundamentalRepresentation, ComplexD > ZWilsonImplR; // Real.. whichever prec
typedef WilsonImpl<vComplexF, FundamentalRepresentation, ComplexD > ZWilsonImplF; // Float
typedef WilsonImpl<vComplexD, FundamentalRepresentation, ComplexD > ZWilsonImplD; // Double
typedef WilsonImpl<vComplex, AdjointRepresentation > WilsonAdjImplR; // Real.. whichever prec
typedef WilsonImpl<vComplexF, AdjointRepresentation > WilsonAdjImplF; // Float
typedef WilsonImpl<vComplexD, AdjointRepresentation > WilsonAdjImplD; // Double
typedef WilsonImpl<vComplex, TwoIndexSymmetricRepresentation > WilsonTwoIndexSymmetricImplR; // Real.. whichever prec
typedef WilsonImpl<vComplexF, TwoIndexSymmetricRepresentation > WilsonTwoIndexSymmetricImplF; // Float
typedef WilsonImpl<vComplexD, TwoIndexSymmetricRepresentation > WilsonTwoIndexSymmetricImplD; // Double
typedef DomainWallVec5dImpl<vComplex ,Nc> DomainWallVec5dImplR; // Real.. whichever prec
typedef DomainWallVec5dImpl<vComplexF,Nc> DomainWallVec5dImplF; // Float
typedef DomainWallVec5dImpl<vComplexD,Nc> DomainWallVec5dImplD; // Double
typedef DomainWallVec5dImpl<vComplex ,Nc,ComplexD> ZDomainWallVec5dImplR; // Real.. whichever prec
typedef DomainWallVec5dImpl<vComplexF,Nc,ComplexD> ZDomainWallVec5dImplF; // Float
typedef DomainWallVec5dImpl<vComplexD,Nc,ComplexD> ZDomainWallVec5dImplD; // Double
typedef DomainWallVec5dImpl<vComplex ,Nc> DomainWallVec5dImplR; // Real.. whichever prec
typedef DomainWallVec5dImpl<vComplexF,Nc> DomainWallVec5dImplF; // Float
typedef DomainWallVec5dImpl<vComplexD,Nc> DomainWallVec5dImplD; // Double
typedef GparityWilsonImpl<vComplex ,Nc> GparityWilsonImplR; // Real.. whichever prec
typedef GparityWilsonImpl<vComplexF,Nc> GparityWilsonImplF; // Float
typedef GparityWilsonImpl<vComplexD,Nc> GparityWilsonImplD; // Double
}
typedef GparityWilsonImpl<vComplex, Nc> GparityWilsonImplR; // Real.. whichever prec
typedef GparityWilsonImpl<vComplexF, Nc> GparityWilsonImplF; // Float
typedef GparityWilsonImpl<vComplexD, Nc> GparityWilsonImplD; // Double
}
}
#endif

546
lib/qcd/action/fermion/WilsonFermion.cc
View File

@@ -1,130 +1,129 @@
/*************************************************************************************
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/action/fermion/WilsonFermion.cc
Source file: ./lib/qcd/action/fermion/WilsonFermion.cc
Copyright (C) 2015
Copyright (C) 2015
Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
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 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.
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.
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 */
See the full license in the file "LICENSE" in the top level distribution
directory
*************************************************************************************/
/* END LEGAL */
#include <Grid.h>
namespace Grid {
namespace QCD {
const std::vector<int> WilsonFermionStatic::directions ({0,1,2,3, 0, 1, 2, 3});
const std::vector<int> WilsonFermionStatic::displacements({1,1,1,1,-1,-1,-1,-1});
int WilsonFermionStatic::HandOptDslash;
const std::vector<int> WilsonFermionStatic::directions({0, 1, 2, 3, 0, 1, 2,
3});
const std::vector<int> WilsonFermionStatic::displacements({1, 1, 1, 1, -1, -1,
-1, -1});
int WilsonFermionStatic::HandOptDslash;
/////////////////////////////////
// Constructor and gauge import
/////////////////////////////////
/////////////////////////////////
// Constructor and gauge import
/////////////////////////////////
template<class Impl>
WilsonFermion<Impl>::WilsonFermion(GaugeField &_Umu,
GridCartesian &Fgrid,
GridRedBlackCartesian &Hgrid,
RealD _mass,const ImplParams &p) :
Kernels(p),
_grid(&Fgrid),
_cbgrid(&Hgrid),
Stencil (&Fgrid,npoint,Even,directions,displacements),
StencilEven(&Hgrid,npoint,Even,directions,displacements), // source is Even
StencilOdd (&Hgrid,npoint,Odd ,directions,displacements), // source is Odd
mass(_mass),
Lebesgue(_grid),
LebesgueEvenOdd(_cbgrid),
Umu(&Fgrid),
UmuEven(&Hgrid),
UmuOdd (&Hgrid)
{
// Allocate the required comms buffer
ImportGauge(_Umu);
template <class Impl>
WilsonFermion<Impl>::WilsonFermion(GaugeField &_Umu, GridCartesian &Fgrid,
GridRedBlackCartesian &Hgrid, RealD _mass,
const ImplParams &p)
: Kernels(p),
_grid(&Fgrid),
_cbgrid(&Hgrid),
Stencil(&Fgrid, npoint, Even, directions, displacements),
StencilEven(&Hgrid, npoint, Even, directions,
displacements), // source is Even
StencilOdd(&Hgrid, npoint, Odd, directions,
displacements), // source is Odd
mass(_mass),
Lebesgue(_grid),
LebesgueEvenOdd(_cbgrid),
Umu(&Fgrid),
UmuEven(&Hgrid),
UmuOdd(&Hgrid) {
// Allocate the required comms buffer
ImportGauge(_Umu);
}
template <class Impl>
void WilsonFermion<Impl>::ImportGauge(const GaugeField &_Umu) {
GaugeField HUmu(_Umu._grid);
HUmu = _Umu * (-0.5);
Impl::DoubleStore(GaugeGrid(), Umu, HUmu);
pickCheckerboard(Even, UmuEven, Umu);
pickCheckerboard(Odd, UmuOdd, Umu);
}
/////////////////////////////
// Implement the interface
/////////////////////////////
template <class Impl>
RealD WilsonFermion<Impl>::M(const FermionField &in, FermionField &out) {
out.checkerboard = in.checkerboard;
Dhop(in, out, DaggerNo);
return axpy_norm(out, 4 + mass, in, out);
}
template <class Impl>
RealD WilsonFermion<Impl>::Mdag(const FermionField &in, FermionField &out) {
out.checkerboard = in.checkerboard;
Dhop(in, out, DaggerYes);
return axpy_norm(out, 4 + mass, in, out);
}
template <class Impl>
void WilsonFermion<Impl>::Meooe(const FermionField &in, FermionField &out) {
if (in.checkerboard == Odd) {
DhopEO(in, out, DaggerNo);
} else {
DhopOE(in, out, DaggerNo);
}
}
template<class Impl>
void WilsonFermion<Impl>::ImportGauge(const GaugeField &_Umu)
{
GaugeField HUmu(_Umu._grid);
HUmu = _Umu*(-0.5);
Impl::DoubleStore(GaugeGrid(),Umu,HUmu);
pickCheckerboard(Even,UmuEven,Umu);
pickCheckerboard(Odd ,UmuOdd,Umu);
template <class Impl>
void WilsonFermion<Impl>::MeooeDag(const FermionField &in, FermionField &out) {
if (in.checkerboard == Odd) {
DhopEO(in, out, DaggerYes);
} else {
DhopOE(in, out, DaggerYes);
}
}
/////////////////////////////
// Implement the interface
/////////////////////////////
template<class Impl>
RealD WilsonFermion<Impl>::M(const FermionField &in, FermionField &out)
{
out.checkerboard=in.checkerboard;
Dhop(in,out,DaggerNo);
return axpy_norm(out,4+mass,in,out);
}
template<class Impl>
RealD WilsonFermion<Impl>::Mdag(const FermionField &in, FermionField &out)
{
out.checkerboard=in.checkerboard;
Dhop(in,out,DaggerYes);
return axpy_norm(out,4+mass,in,out);
}
template<class Impl>
void WilsonFermion<Impl>::Meooe(const FermionField &in, FermionField &out)
{
if ( in.checkerboard == Odd ) {
DhopEO(in,out,DaggerNo);
} else {
DhopOE(in,out,DaggerNo);
}
}
template<class Impl>
void WilsonFermion<Impl>::MeooeDag(const FermionField &in, FermionField &out)
{
if ( in.checkerboard == Odd ) {
DhopEO(in,out,DaggerYes);
} else {
DhopOE(in,out,DaggerYes);
}
}
template<class Impl>
template <class Impl>
void WilsonFermion<Impl>::Mooee(const FermionField &in, FermionField &out) {
out.checkerboard = in.checkerboard;
typename FermionField::scalar_type scal(4.0+mass);
out = scal*in;
typename FermionField::scalar_type scal(4.0 + mass);
out = scal * in;
}
template<class Impl>
template <class Impl>
void WilsonFermion<Impl>::MooeeDag(const FermionField &in, FermionField &out) {
out.checkerboard = in.checkerboard;
Mooee(in,out);
Mooee(in, out);
}
template<class Impl>
void WilsonFermion<Impl>::MooeeInv(const FermionField &in, FermionField &out) {
out.checkerboard = in.checkerboard;
@@ -136,8 +135,9 @@ namespace QCD {
out.checkerboard = in.checkerboard;
MooeeInv(in,out);
}
template<class Impl>
void WilsonFermion<Impl>:: MomentumSpacePropagator(FermionField &out, const FermionField &in,RealD _m) {
void WilsonFermion<Impl>::MomentumSpacePropagator(FermionField &out, const FermionField &in,RealD _m) {
// what type LatticeComplex
conformable(_grid,out._grid);
@@ -190,184 +190,182 @@ namespace QCD {
out = num*denom; // [ -i gmu sin k + 2 sin^2 k/2 + m] / [ sin^2 k + (2 sin^2 k/2 + m)^2 ]
}
///////////////////////////////////
// Internal
///////////////////////////////////
template<class Impl>
void WilsonFermion<Impl>::DerivInternal(StencilImpl & st,
DoubledGaugeField & U,
GaugeField &mat,
const FermionField &A,
const FermionField &B,int dag) {
assert((dag==DaggerNo) ||(dag==DaggerYes));
Compressor compressor(dag);
FermionField Btilde(B._grid);
FermionField Atilde(B._grid);
Atilde = A;
st.HaloExchange(B,compressor);
for(int mu=0;mu<Nd;mu++){
////////////////////////////////////////////////////////////////////////
// Flip gamma (1+g)<->(1-g) if dag
////////////////////////////////////////////////////////////////////////
int gamma = mu;
if ( !dag ) gamma+= Nd;
////////////////////////
// Call the single hop
////////////////////////
PARALLEL_FOR_LOOP
for(int sss=0;sss<B._grid->oSites();sss++){
Kernels::DiracOptDhopDir(st,U,st.comm_buf,sss,sss,B,Btilde,mu,gamma);
}
//////////////////////////////////////////////////
// spin trace outer product
//////////////////////////////////////////////////
Impl::InsertForce4D(mat,Btilde,Atilde,mu);
}
}
template<class Impl>
void WilsonFermion<Impl>::DhopDeriv(GaugeField &mat,const FermionField &U,const FermionField &V,int dag)
{
conformable(U._grid,_grid);
conformable(U._grid,V._grid);
conformable(U._grid,mat._grid);
mat.checkerboard = U.checkerboard;
DerivInternal(Stencil,Umu,mat,U,V,dag);
}
template<class Impl>
void WilsonFermion<Impl>::DhopDerivOE(GaugeField &mat,const FermionField &U,const FermionField &V,int dag)
{
conformable(U._grid,_cbgrid);
conformable(U._grid,V._grid);
conformable(U._grid,mat._grid);
assert(V.checkerboard==Even);
assert(U.checkerboard==Odd);
mat.checkerboard = Odd;
DerivInternal(StencilEven,UmuOdd,mat,U,V,dag);
}
template<class Impl>
void WilsonFermion<Impl>::DhopDerivEO(GaugeField &mat,const FermionField &U,const FermionField &V,int dag)
{
conformable(U._grid,_cbgrid);
conformable(U._grid,V._grid);
conformable(U._grid,mat._grid);
assert(V.checkerboard==Odd);
assert(U.checkerboard==Even);
mat.checkerboard = Even;
DerivInternal(StencilOdd,UmuEven,mat,U,V,dag);
}
template<class Impl>
void WilsonFermion<Impl>::Dhop(const FermionField &in, FermionField &out,int dag) {
conformable(in._grid,_grid); // verifies full grid
conformable(in._grid,out._grid);
out.checkerboard = in.checkerboard;
DhopInternal(Stencil,Lebesgue,Umu,in,out,dag);
}
template<class Impl>
void WilsonFermion<Impl>::DhopOE(const FermionField &in, FermionField &out,int dag) {
conformable(in._grid,_cbgrid); // verifies half grid
conformable(in._grid,out._grid); // drops the cb check
assert(in.checkerboard==Even);
out.checkerboard = Odd;
DhopInternal(StencilEven,LebesgueEvenOdd,UmuOdd,in,out,dag);
}
template<class Impl>
void WilsonFermion<Impl>::DhopEO(const FermionField &in, FermionField &out,int dag) {
conformable(in._grid,_cbgrid); // verifies half grid
conformable(in._grid,out._grid); // drops the cb check
assert(in.checkerboard==Odd);
out.checkerboard = Even;
DhopInternal(StencilOdd,LebesgueEvenOdd,UmuEven,in,out,dag);
}
template<class Impl>
void WilsonFermion<Impl>::Mdir (const FermionField &in, FermionField &out,int dir,int disp) {
DhopDir(in,out,dir,disp);
}
template<class Impl>
void WilsonFermion<Impl>::DhopDir(const FermionField &in, FermionField &out,int dir,int disp){
int skip = (disp==1) ? 0 : 1;
int dirdisp = dir+skip*4;
int gamma = dir+(1-skip)*4;
DhopDirDisp(in,out,dirdisp,gamma,DaggerNo);
};
template<class Impl>
void WilsonFermion<Impl>::DhopDirDisp(const FermionField &in, FermionField &out,int dirdisp,int gamma,int dag) {
Compressor compressor(dag);
Stencil.HaloExchange(in,compressor);
PARALLEL_FOR_LOOP
for(int sss=0;sss<in._grid->oSites();sss++){
Kernels::DiracOptDhopDir(Stencil,Umu,Stencil.comm_buf,sss,sss,in,out,dirdisp,gamma);
}
};
template<class Impl>
void WilsonFermion<Impl>::DhopInternal(StencilImpl & st,LebesgueOrder& lo,DoubledGaugeField & U,
const FermionField &in, FermionField &out,int dag)
{
assert((dag==DaggerNo) ||(dag==DaggerYes));
Compressor compressor(dag);
st.HaloExchange(in,compressor);
if ( dag == DaggerYes ) {
PARALLEL_FOR_LOOP
for(int sss=0;sss<in._grid->oSites();sss++){
Kernels::DiracOptDhopSiteDag(st,lo,U,st.comm_buf,sss,sss,1,1,in,out);
}
} else {
PARALLEL_FOR_LOOP
for(int sss=0;sss<in._grid->oSites();sss++){
Kernels::DiracOptDhopSite(st,lo,U,st.comm_buf,sss,sss,1,1,in,out);
}
}
};
FermOpTemplateInstantiate(WilsonFermion);
GparityFermOpTemplateInstantiate(WilsonFermion);
///////////////////////////////////
// Internal
///////////////////////////////////
}}
template <class Impl>
void WilsonFermion<Impl>::DerivInternal(StencilImpl &st, DoubledGaugeField &U,
GaugeField &mat, const FermionField &A,
const FermionField &B, int dag) {
assert((dag == DaggerNo) || (dag == DaggerYes));
Compressor compressor(dag);
FermionField Btilde(B._grid);
FermionField Atilde(B._grid);
Atilde = A;
st.HaloExchange(B, compressor);
for (int mu = 0; mu < Nd; mu++) {
////////////////////////////////////////////////////////////////////////
// Flip gamma (1+g)<->(1-g) if dag
////////////////////////////////////////////////////////////////////////
int gamma = mu;
if (!dag) gamma += Nd;
////////////////////////
// Call the single hop
////////////////////////
PARALLEL_FOR_LOOP
for (int sss = 0; sss < B._grid->oSites(); sss++) {
Kernels::DiracOptDhopDir(st, U, st.comm_buf, sss, sss, B, Btilde, mu,
gamma);
}
//////////////////////////////////////////////////
// spin trace outer product
//////////////////////////////////////////////////
Impl::InsertForce4D(mat, Btilde, Atilde, mu);
}
}
template <class Impl>
void WilsonFermion<Impl>::DhopDeriv(GaugeField &mat, const FermionField &U,
const FermionField &V, int dag) {
conformable(U._grid, _grid);
conformable(U._grid, V._grid);
conformable(U._grid, mat._grid);
mat.checkerboard = U.checkerboard;
DerivInternal(Stencil, Umu, mat, U, V, dag);
}
template <class Impl>
void WilsonFermion<Impl>::DhopDerivOE(GaugeField &mat, const FermionField &U,
const FermionField &V, int dag) {
conformable(U._grid, _cbgrid);
conformable(U._grid, V._grid);
conformable(U._grid, mat._grid);
assert(V.checkerboard == Even);
assert(U.checkerboard == Odd);
mat.checkerboard = Odd;
DerivInternal(StencilEven, UmuOdd, mat, U, V, dag);
}
template <class Impl>
void WilsonFermion<Impl>::DhopDerivEO(GaugeField &mat, const FermionField &U,
const FermionField &V, int dag) {
conformable(U._grid, _cbgrid);
conformable(U._grid, V._grid);
conformable(U._grid, mat._grid);
assert(V.checkerboard == Odd);
assert(U.checkerboard == Even);
mat.checkerboard = Even;
DerivInternal(StencilOdd, UmuEven, mat, U, V, dag);
}
template <class Impl>
void WilsonFermion<Impl>::Dhop(const FermionField &in, FermionField &out,
int dag) {
conformable(in._grid, _grid); // verifies full grid
conformable(in._grid, out._grid);
out.checkerboard = in.checkerboard;
DhopInternal(Stencil, Lebesgue, Umu, in, out, dag);
}
template <class Impl>
void WilsonFermion<Impl>::DhopOE(const FermionField &in, FermionField &out,
int dag) {
conformable(in._grid, _cbgrid); // verifies half grid
conformable(in._grid, out._grid); // drops the cb check
assert(in.checkerboard == Even);
out.checkerboard = Odd;
DhopInternal(StencilEven, LebesgueEvenOdd, UmuOdd, in, out, dag);
}
template <class Impl>
void WilsonFermion<Impl>::DhopEO(const FermionField &in, FermionField &out,
int dag) {
conformable(in._grid, _cbgrid); // verifies half grid
conformable(in._grid, out._grid); // drops the cb check
assert(in.checkerboard == Odd);
out.checkerboard = Even;
DhopInternal(StencilOdd, LebesgueEvenOdd, UmuEven, in, out, dag);
}
template <class Impl>
void WilsonFermion<Impl>::Mdir(const FermionField &in, FermionField &out,
int dir, int disp) {
DhopDir(in, out, dir, disp);
}
template <class Impl>
void WilsonFermion<Impl>::DhopDir(const FermionField &in, FermionField &out,
int dir, int disp) {
int skip = (disp == 1) ? 0 : 1;
int dirdisp = dir + skip * 4;
int gamma = dir + (1 - skip) * 4;
DhopDirDisp(in, out, dirdisp, gamma, DaggerNo);
};
template <class Impl>
void WilsonFermion<Impl>::DhopDirDisp(const FermionField &in, FermionField &out,
int dirdisp, int gamma, int dag) {
Compressor compressor(dag);
Stencil.HaloExchange(in, compressor);
PARALLEL_FOR_LOOP
for (int sss = 0; sss < in._grid->oSites(); sss++) {
Kernels::DiracOptDhopDir(Stencil, Umu, Stencil.comm_buf, sss, sss, in, out,
dirdisp, gamma);
}
};
template <class Impl>
void WilsonFermion<Impl>::DhopInternal(StencilImpl &st, LebesgueOrder &lo,
DoubledGaugeField &U,
const FermionField &in,
FermionField &out, int dag) {
assert((dag == DaggerNo) || (dag == DaggerYes));
Compressor compressor(dag);
st.HaloExchange(in, compressor);
if (dag == DaggerYes) {
PARALLEL_FOR_LOOP
for (int sss = 0; sss < in._grid->oSites(); sss++) {
Kernels::DiracOptDhopSiteDag(st, lo, U, st.comm_buf, sss, sss, 1, 1, in,
out);
}
} else {
PARALLEL_FOR_LOOP
for (int sss = 0; sss < in._grid->oSites(); sss++) {
Kernels::DiracOptDhopSite(st, lo, U, st.comm_buf, sss, sss, 1, 1, in,
out);
}
}
};
FermOpTemplateInstantiate(WilsonFermion);
AdjointFermOpTemplateInstantiate(WilsonFermion);
TwoIndexFermOpTemplateInstantiate(WilsonFermion);
GparityFermOpTemplateInstantiate(WilsonFermion);
}
}

265
lib/qcd/action/fermion/WilsonFermion.h
View File

@@ -1,163 +1,154 @@
/*************************************************************************************
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/action/fermion/WilsonFermion.h
Source file: ./lib/qcd/action/fermion/WilsonFermion.h
Copyright (C) 2015
Copyright (C) 2015
Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
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 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.
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.
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 GRID_QCD_WILSON_FERMION_H
#define GRID_QCD_WILSON_FERMION_H
See the full license in the file "LICENSE" in the top level distribution
directory
*************************************************************************************/
/* END LEGAL */
#ifndef GRID_QCD_WILSON_FERMION_H
#define GRID_QCD_WILSON_FERMION_H
namespace Grid {
namespace QCD {
namespace QCD {
class WilsonFermionStatic {
public:
static int HandOptDslash; // these are a temporary hack
static int MortonOrder;
static const std::vector<int> directions ;
static const std::vector<int> displacements;
static const int npoint=8;
};
class WilsonFermionStatic {
public:
static int HandOptDslash; // these are a temporary hack
static int MortonOrder;
static const std::vector<int> directions;
static const std::vector<int> displacements;
static const int npoint = 8;
};
template<class Impl>
class WilsonFermion : public WilsonKernels<Impl>, public WilsonFermionStatic
{
public:
INHERIT_IMPL_TYPES(Impl);
typedef WilsonKernels<Impl> Kernels;
template <class Impl>
class WilsonFermion : public WilsonKernels<Impl>, public WilsonFermionStatic {
public:
INHERIT_IMPL_TYPES(Impl);
typedef WilsonKernels<Impl> Kernels;
///////////////////////////////////////////////////////////////
// Implement the abstract base
///////////////////////////////////////////////////////////////
GridBase *GaugeGrid(void) { return _grid ;}
GridBase *GaugeRedBlackGrid(void) { return _cbgrid ;}
GridBase *FermionGrid(void) { return _grid;}
GridBase *FermionRedBlackGrid(void) { return _cbgrid;}
///////////////////////////////////////////////////////////////
// Implement the abstract base
///////////////////////////////////////////////////////////////
GridBase *GaugeGrid(void) { return _grid; }
GridBase *GaugeRedBlackGrid(void) { return _cbgrid; }
GridBase *FermionGrid(void) { return _grid; }
GridBase *FermionRedBlackGrid(void) { return _cbgrid; }
//////////////////////////////////////////////////////////////////
// override multiply; cut number routines if pass dagger argument
// and also make interface more uniformly consistent
//////////////////////////////////////////////////////////////////
RealD M(const FermionField &in, FermionField &out);
RealD Mdag(const FermionField &in, FermionField &out);
//////////////////////////////////////////////////////////////////
// override multiply; cut number routines if pass dagger argument
// and also make interface more uniformly consistent
//////////////////////////////////////////////////////////////////
RealD M(const FermionField &in, FermionField &out);
RealD Mdag(const FermionField &in, FermionField &out);
/////////////////////////////////////////////////////////
// half checkerboard operations
// could remain virtual so we can derive Clover from Wilson base
/////////////////////////////////////////////////////////
void Meooe(const FermionField &in, FermionField &out) ;
void MeooeDag(const FermionField &in, FermionField &out) ;
/////////////////////////////////////////////////////////
// half checkerboard operations
// could remain virtual so we can derive Clover from Wilson base
/////////////////////////////////////////////////////////
void Meooe(const FermionField &in, FermionField &out);
void MeooeDag(const FermionField &in, FermionField &out);
// allow override for twisted mass and clover
virtual void Mooee(const FermionField &in, FermionField &out) ;
virtual void MooeeDag(const FermionField &in, FermionField &out) ;
virtual void MooeeInv(const FermionField &in, FermionField &out) ;
virtual void MooeeInvDag(const FermionField &in, FermionField &out) ;
// allow override for twisted mass and clover
virtual void Mooee(const FermionField &in, FermionField &out);
virtual void MooeeDag(const FermionField &in, FermionField &out);
virtual void MooeeInv(const FermionField &in, FermionField &out);
virtual void MooeeInvDag(const FermionField &in, FermionField &out);
virtual void MomentumSpacePropagator(FermionField &out,const FermionField &in,RealD _mass) ;
virtual void MomentumSpacePropagator(FermionField &out,const FermionField &in,RealD _mass) ;
////////////////////////
// Derivative interface
////////////////////////
// Interface calls an internal routine
void DhopDeriv(GaugeField &mat,const FermionField &U,const FermionField &V,int dag);
void DhopDerivOE(GaugeField &mat,const FermionField &U,const FermionField &V,int dag);
void DhopDerivEO(GaugeField &mat,const FermionField &U,const FermionField &V,int dag);
////////////////////////
// Derivative interface
////////////////////////
// Interface calls an internal routine
void DhopDeriv(GaugeField &mat,const FermionField &U,const FermionField &V,int dag);
void DhopDerivOE(GaugeField &mat,const FermionField &U,const FermionField &V,int dag);
void DhopDerivEO(GaugeField &mat,const FermionField &U,const FermionField &V,int dag);
///////////////////////////////////////////////////////////////
// non-hermitian hopping term; half cb or both
///////////////////////////////////////////////////////////////
void Dhop(const FermionField &in, FermionField &out, int dag);
void DhopOE(const FermionField &in, FermionField &out, int dag);
void DhopEO(const FermionField &in, FermionField &out, int dag);
///////////////////////////////////////////////////////////////
// Multigrid assistance; force term uses too
///////////////////////////////////////////////////////////////
void Mdir(const FermionField &in, FermionField &out, int dir, int disp);
void DhopDir(const FermionField &in, FermionField &out, int dir, int disp);
void DhopDirDisp(const FermionField &in, FermionField &out, int dirdisp,
int gamma, int dag);
///////////////////////////////////////////////////////////////
// Extra methods added by derived
///////////////////////////////////////////////////////////////
void DerivInternal(StencilImpl &st, DoubledGaugeField &U, GaugeField &mat,
const FermionField &A, const FermionField &B, int dag);
void DhopInternal(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
const FermionField &in, FermionField &out, int dag);
// Constructor
WilsonFermion(GaugeField &_Umu, GridCartesian &Fgrid,
GridRedBlackCartesian &Hgrid, RealD _mass,
const ImplParams &p = ImplParams());
// DoubleStore impl dependent
void ImportGauge(const GaugeField &_Umu);
///////////////////////////////////////////////////////////////
// Data members require to support the functionality
///////////////////////////////////////////////////////////////
// protected:
public:
RealD mass;
GridBase *_grid;
GridBase *_cbgrid;
// Defines the stencils for even and odd
StencilImpl Stencil;
StencilImpl StencilEven;
StencilImpl StencilOdd;
// Copy of the gauge field , with even and odd subsets
DoubledGaugeField Umu;
DoubledGaugeField UmuEven;
DoubledGaugeField UmuOdd;
LebesgueOrder Lebesgue;
LebesgueOrder LebesgueEvenOdd;
};
typedef WilsonFermion<WilsonImplF> WilsonFermionF;
typedef WilsonFermion<WilsonImplD> WilsonFermionD;
///////////////////////////////////////////////////////////////
// non-hermitian hopping term; half cb or both
///////////////////////////////////////////////////////////////
void Dhop(const FermionField &in, FermionField &out,int dag) ;
void DhopOE(const FermionField &in, FermionField &out,int dag) ;
void DhopEO(const FermionField &in, FermionField &out,int dag) ;
///////////////////////////////////////////////////////////////
// Multigrid assistance; force term uses too
///////////////////////////////////////////////////////////////
void Mdir (const FermionField &in, FermionField &out,int dir,int disp) ;
void DhopDir(const FermionField &in, FermionField &out,int dir,int disp);
void DhopDirDisp(const FermionField &in, FermionField &out,int dirdisp,int gamma,int dag) ;
///////////////////////////////////////////////////////////////
// Extra methods added by derived
///////////////////////////////////////////////////////////////
void DerivInternal(StencilImpl & st,
DoubledGaugeField & U,
GaugeField &mat,
const FermionField &A,
const FermionField &B,
int dag);
void DhopInternal(StencilImpl & st,LebesgueOrder & lo,DoubledGaugeField & U,
const FermionField &in, FermionField &out,int dag) ;
// Constructor
WilsonFermion(GaugeField &_Umu,
GridCartesian &Fgrid,
GridRedBlackCartesian &Hgrid,
RealD _mass,
const ImplParams &p= ImplParams()
) ;
// DoubleStore impl dependent
void ImportGauge(const GaugeField &_Umu);
///////////////////////////////////////////////////////////////
// Data members require to support the functionality
///////////////////////////////////////////////////////////////
// protected:
public:
RealD mass;
GridBase * _grid;
GridBase * _cbgrid;
//Defines the stencils for even and odd
StencilImpl Stencil;
StencilImpl StencilEven;
StencilImpl StencilOdd;
// Copy of the gauge field , with even and odd subsets
DoubledGaugeField Umu;
DoubledGaugeField UmuEven;
DoubledGaugeField UmuOdd;
LebesgueOrder Lebesgue;
LebesgueOrder LebesgueEvenOdd;
};
typedef WilsonFermion<WilsonImplF> WilsonFermionF;
typedef WilsonFermion<WilsonImplD> WilsonFermionD;
}
}
}
#endif

209
lib/qcd/action/fermion/WilsonFermion5D.cc
View File

@@ -42,11 +42,11 @@ const std::vector<int> WilsonFermion5DStatic::displacements({1,1,1,1,-1,-1,-1,-1
// 5d lattice for DWF.
template<class Impl>
WilsonFermion5D<Impl>::WilsonFermion5D(GaugeField &_Umu,
GridCartesian &FiveDimGrid,
GridRedBlackCartesian &FiveDimRedBlackGrid,
GridCartesian &FourDimGrid,
GridRedBlackCartesian &FourDimRedBlackGrid,
RealD _M5,const ImplParams &p) :
GridCartesian &FiveDimGrid,
GridRedBlackCartesian &FiveDimRedBlackGrid,
GridCartesian &FourDimGrid,
GridRedBlackCartesian &FourDimRedBlackGrid,
RealD _M5,const ImplParams &p) :
Kernels(p),
_FiveDimGrid (&FiveDimGrid),
_FiveDimRedBlackGrid(&FiveDimRedBlackGrid),
@@ -135,10 +135,10 @@ WilsonFermion5D<Impl>::WilsonFermion5D(GaugeField &_Umu,
/*
template<class Impl>
WilsonFermion5D<Impl>::WilsonFermion5D(int simd,GaugeField &_Umu,
GridCartesian &FiveDimGrid,
GridRedBlackCartesian &FiveDimRedBlackGrid,
GridCartesian &FourDimGrid,
RealD _M5,const ImplParams &p) :
GridCartesian &FiveDimGrid,
GridRedBlackCartesian &FiveDimRedBlackGrid,
GridCartesian &FourDimGrid,
RealD _M5,const ImplParams &p) :
{
int nsimd = Simd::Nsimd();
@@ -175,6 +175,73 @@ WilsonFermion5D<Impl>::WilsonFermion5D(int simd,GaugeField &_Umu,
}
*/
template<class Impl>
void WilsonFermion5D<Impl>::Report(void)
{
std::vector<int> latt = GridDefaultLatt();
RealD volume = Ls; for(int mu=0;mu<Nd;mu++) volume=volume*latt[mu];
RealD NP = _FourDimGrid->_Nprocessors;
if ( DhopCalls > 0 ) {
std::cout << GridLogMessage << "#### Dhop calls report " << std::endl;
std::cout << GridLogMessage << "WilsonFermion5D Number of Dhop Calls : " << DhopCalls << std::endl;
std::cout << GridLogMessage << "WilsonFermion5D Total Communication time : " << DhopCommTime
<< " us" << std::endl;
std::cout << GridLogMessage << "WilsonFermion5D CommTime/Calls : "
<< DhopCommTime / DhopCalls << " us" << std::endl;
std::cout << GridLogMessage << "WilsonFermion5D Total Compute time : "
<< DhopComputeTime << " us" << std::endl;
std::cout << GridLogMessage << "WilsonFermion5D ComputeTime/Calls : "
<< DhopComputeTime / DhopCalls << " us" << std::endl;
RealD mflops = 1344*volume*DhopCalls/DhopComputeTime;
std::cout << GridLogMessage << "Average mflops/s per call : " << mflops << std::endl;
std::cout << GridLogMessage << "Average mflops/s per call per node : " << mflops/NP << std::endl;
}
if ( DerivCalls > 0 ) {
std::cout << GridLogMessage << "#### Deriv calls report "<< std::endl;
std::cout << GridLogMessage << "WilsonFermion5D Number of Deriv Calls : " <<DerivCalls <<std::endl;
std::cout << GridLogMessage << "WilsonFermion5D Total Communication time : " <<DerivCommTime <<" us"<<std::endl;
std::cout << GridLogMessage << "WilsonFermion5D CommTime/Calls : " <<DerivCommTime/DerivCalls<<" us" <<std::endl;
std::cout << GridLogMessage << "WilsonFermion5D Total Compute time : " <<DerivComputeTime <<" us"<<std::endl;
std::cout << GridLogMessage << "WilsonFermion5D ComputeTime/Calls : " <<DerivComputeTime/DerivCalls<<" us" <<std::endl;
std::cout << GridLogMessage << "WilsonFermion5D Total Dhop Compute time : " <<DerivDhopComputeTime <<" us"<<std::endl;
std::cout << GridLogMessage << "WilsonFermion5D Dhop ComputeTime/Calls : " <<DerivDhopComputeTime/DerivCalls<<" us" <<std::endl;
RealD mflops = 144*volume*DerivCalls/DerivDhopComputeTime;
std::cout << GridLogMessage << "Average mflops/s per call : " << mflops << std::endl;
std::cout << GridLogMessage << "Average mflops/s per call per node : " << mflops/NP << std::endl;
}
if (DerivCalls > 0 || DhopCalls > 0){
std::cout << GridLogMessage << "WilsonFermion5D Stencil"<<std::endl; Stencil.Report();
std::cout << GridLogMessage << "WilsonFermion5D StencilEven"<<std::endl; StencilEven.Report();
std::cout << GridLogMessage << "WilsonFermion5D StencilOdd"<<std::endl; StencilOdd.Report();
}
}
template<class Impl>
void WilsonFermion5D<Impl>::ZeroCounters(void) {
DhopCalls = 0;
DhopCommTime = 0;
DhopComputeTime = 0;
DerivCalls = 0;
DerivCommTime = 0;
DerivComputeTime = 0;
DerivDhopComputeTime = 0;
Stencil.ZeroCounters();
StencilEven.ZeroCounters();
StencilOdd.ZeroCounters();
}
template<class Impl>
void WilsonFermion5D<Impl>::ImportGauge(const GaugeField &_Umu)
{
@@ -215,12 +282,13 @@ PARALLEL_FOR_LOOP
template<class Impl>
void WilsonFermion5D<Impl>::DerivInternal(StencilImpl & st,
DoubledGaugeField & U,
GaugeField &mat,
const FermionField &A,
const FermionField &B,
int dag)
DoubledGaugeField & U,
GaugeField &mat,
const FermionField &A,
const FermionField &B,
int dag)
{
DerivCalls++;
assert((dag==DaggerNo) ||(dag==DaggerYes));
conformable(st._grid,A._grid);
@@ -231,51 +299,53 @@ void WilsonFermion5D<Impl>::DerivInternal(StencilImpl & st,
FermionField Btilde(B._grid);
FermionField Atilde(B._grid);
DerivCommTime-=usecond();
st.HaloExchange(B,compressor);
DerivCommTime+=usecond();
Atilde=A;
for(int mu=0;mu<Nd;mu++){
DerivComputeTime-=usecond();
for (int mu = 0; mu < Nd; mu++) {
////////////////////////////////////////////////////////////////////////
// Flip gamma if dag
////////////////////////////////////////////////////////////////////////
int gamma = mu;
if ( !dag ) gamma+= Nd;
if (!dag) gamma += Nd;
////////////////////////
// Call the single hop
////////////////////////
PARALLEL_FOR_LOOP
for(int sss=0;sss<U._grid->oSites();sss++){
for(int s=0;s<Ls;s++){
int sU=sss;
int sF = s+Ls*sU;
DerivDhopComputeTime -= usecond();
PARALLEL_FOR_LOOP
for (int sss = 0; sss < U._grid->oSites(); sss++) {
for (int s = 0; s < Ls; s++) {
int sU = sss;
int sF = s + Ls * sU;
assert ( sF< B._grid->oSites());
assert ( sU< U._grid->oSites());
assert(sF < B._grid->oSites());
assert(sU < U._grid->oSites());
Kernels::DiracOptDhopDir(st,U,st.comm_buf,sF,sU,B,Btilde,mu,gamma);
////////////////////////////
// spin trace outer product
////////////////////////////
Kernels::DiracOptDhopDir(st, U, st.comm_buf, sF, sU, B, Btilde, mu,
gamma);
////////////////////////////
// spin trace outer product
////////////////////////////
}
}
Impl::InsertForce5D(mat,Btilde,Atilde,mu);
DerivDhopComputeTime += usecond();
Impl::InsertForce5D(mat, Btilde, Atilde, mu);
}
DerivComputeTime += usecond();
}
template<class Impl>
void WilsonFermion5D<Impl>::DhopDeriv( GaugeField &mat,
const FermionField &A,
const FermionField &B,
int dag)
const FermionField &A,
const FermionField &B,
int dag)
{
conformable(A._grid,FermionGrid());
conformable(A._grid,B._grid);
@@ -288,9 +358,9 @@ void WilsonFermion5D<Impl>::DhopDeriv( GaugeField &mat,
template<class Impl>
void WilsonFermion5D<Impl>::DhopDerivEO(GaugeField &mat,
const FermionField &A,
const FermionField &B,
int dag)
const FermionField &A,
const FermionField &B,
int dag)
{
conformable(A._grid,FermionRedBlackGrid());
conformable(GaugeRedBlackGrid(),mat._grid);
@@ -306,9 +376,9 @@ void WilsonFermion5D<Impl>::DhopDerivEO(GaugeField &mat,
template<class Impl>
void WilsonFermion5D<Impl>::DhopDerivOE(GaugeField &mat,
const FermionField &A,
const FermionField &B,
int dag)
const FermionField &A,
const FermionField &B,
int dag)
{
conformable(A._grid,FermionRedBlackGrid());
conformable(GaugeRedBlackGrid(),mat._grid);
@@ -323,32 +393,61 @@ void WilsonFermion5D<Impl>::DhopDerivOE(GaugeField &mat,
template<class Impl>
void WilsonFermion5D<Impl>::DhopInternal(StencilImpl & st, LebesgueOrder &lo,
DoubledGaugeField & U,
const FermionField &in, FermionField &out,int dag)
DoubledGaugeField & U,
const FermionField &in, FermionField &out,int dag)
{
DhopCalls++;
// assert((dag==DaggerNo) ||(dag==DaggerYes));
Compressor compressor(dag);
int LLs = in._grid->_rdimensions[0];
DhopCommTime-=usecond();
st.HaloExchange(in,compressor);
DhopCommTime+=usecond();
DhopComputeTime-=usecond();
// Dhop takes the 4d grid from U, and makes a 5d index for fermion
if ( dag == DaggerYes ) {
PARALLEL_FOR_LOOP
for(int ss=0;ss<U._grid->oSites();ss++){
int sU=ss;
int sF=LLs*sU;
Kernels::DiracOptDhopSiteDag(st,lo,U,st.comm_buf,sF,sU,LLs,1,in,out);
if (dag == DaggerYes) {
PARALLEL_FOR_LOOP
for (int ss = 0; ss < U._grid->oSites(); ss++) {
int sU = ss;
int sF = LLs * sU;
Kernels::DiracOptDhopSiteDag(st, lo, U, st.comm_buf, sF, sU, LLs, 1, in,
out);
}
#ifdef AVX512
} else if (stat.is_init() ) {
int nthreads;
stat.start();
#pragma omp parallel
{
#pragma omp master
nthreads = omp_get_num_threads();
int mythread = omp_get_thread_num();
stat.enter(mythread);
#pragma omp for nowait
for(int ss=0;ss<U._grid->oSites();ss++)
{
int sU=ss;
int sF=LLs*sU;
Kernels::DiracOptDhopSite(st,lo,U,st.comm_buf,sF,sU,LLs,1,in,out);
}
stat.exit(mythread);
}
stat.accum(nthreads);
#endif
} else {
PARALLEL_FOR_LOOP
for(int ss=0;ss<U._grid->oSites();ss++){
int sU=ss;
int sF=LLs*sU;
Kernels::DiracOptDhopSite(st,lo,U,st.comm_buf,sF,sU,LLs,1,in,out);
PARALLEL_FOR_LOOP
for (int ss = 0; ss < U._grid->oSites(); ss++) {
int sU = ss;
int sF = LLs * sU;
Kernels::DiracOptDhopSite(st, lo, U, st.comm_buf, sF, sU, LLs, 1, in,
out);
}
}
DhopComputeTime+=usecond();
}

14
lib/qcd/action/fermion/WilsonFermion5D.h
View File

@@ -31,6 +31,8 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#ifndef GRID_QCD_WILSON_FERMION_5D_H
#define GRID_QCD_WILSON_FERMION_5D_H
#include <Grid/Stat.h>
namespace Grid {
namespace QCD {
@@ -60,6 +62,18 @@ namespace Grid {
public:
INHERIT_IMPL_TYPES(Impl);
typedef WilsonKernels<Impl> Kernels;
PmuStat stat;
void Report(void);
void ZeroCounters(void);
double DhopCalls;
double DhopCommTime;
double DhopComputeTime;
double DerivCalls;
double DerivCommTime;
double DerivComputeTime;
double DerivDhopComputeTime;
///////////////////////////////////////////////////////////////
// Implement the abstract base

654
lib/qcd/action/fermion/WilsonKernels.cc
View File

@@ -1,103 +1,54 @@
/*************************************************************************************
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
Copyright (C) 2015
Copyright (C) 2015
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 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.
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.
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 */
See the full license in the file "LICENSE" in the top level distribution
directory
*************************************************************************************/
/* END LEGAL */
#include <Grid.h>
namespace Grid {
namespace QCD {
int WilsonKernelsStatic::HandOpt;
int WilsonKernelsStatic::AsmOpt;
int WilsonKernelsStatic::HandOpt;
int WilsonKernelsStatic::AsmOpt;
template<class Impl>
WilsonKernels<Impl>::WilsonKernels(const ImplParams &p): Base(p) {};
template <class Impl>
WilsonKernels<Impl>::WilsonKernels(const ImplParams &p) : Base(p){};
template<class Impl>
void WilsonKernels<Impl>::DiracOptDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int sF,int sU,int Ls, int Ns, const FermionField &in, FermionField &out)
{
#ifdef AVX512
if ( AsmOpt ) {
////////////////////////////////////////////
// Generic implementation; move to different file?
////////////////////////////////////////////
WilsonKernels<Impl>::DiracOptAsmDhopSite(st,lo,U,buf,sF,sU,Ls,Ns,in,out);
} else {
#else
{
#endif
for(int site=0;site<Ns;site++) {
for(int s=0;s<Ls;s++) {
if (HandOpt) WilsonKernels<Impl>::DiracOptHandDhopSite(st,lo,U,buf,sF,sU,in,out);
else WilsonKernels<Impl>::DiracOptGenericDhopSite(st,lo,U,buf,sF,sU,in,out);
sF++;
}
sU++;
}
}
}
template<class Impl>
void WilsonKernels<Impl>::DiracOptDhopSiteDag(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int sF,int sU,int Ls, int Ns, const FermionField &in, FermionField &out)
{
#ifdef AVX512
if ( AsmOpt ) {
WilsonKernels<Impl>::DiracOptAsmDhopSiteDag(st,lo,U,buf,sF,sU,Ls,Ns,in,out);
} else {
#else
{
#endif
for(int site=0;site<Ns;site++) {
for(int s=0;s<Ls;s++) {
if (HandOpt) WilsonKernels<Impl>::DiracOptHandDhopSiteDag(st,lo,U,buf,sF,sU,in,out);
else WilsonKernels<Impl>::DiracOptGenericDhopSiteDag(st,lo,U,buf,sF,sU,in,out);
sF++;
}
sU++;
}
}
}
////////////////////////////////////////////
// Generic implementation; move to different file?
////////////////////////////////////////////
template<class Impl>
void WilsonKernels<Impl>::DiracOptGenericDhopSiteDag(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int sF,int sU,const FermionField &in, FermionField &out)
{
SiteHalfSpinor tmp;
SiteHalfSpinor chi;
template <class Impl>
void WilsonKernels<Impl>::DiracOptGenericDhopSiteDag(
StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf, int sF,
int sU, const FermionField &in, FermionField &out) {
SiteHalfSpinor tmp;
SiteHalfSpinor chi;
SiteHalfSpinor *chi_p;
SiteHalfSpinor Uchi;
SiteSpinor result;
@@ -107,176 +58,175 @@ void WilsonKernels<Impl>::DiracOptGenericDhopSiteDag(StencilImpl &st,LebesgueOrd
///////////////////////////
// Xp
///////////////////////////
SE=st.GetEntry(ptype,Xp,sF);
SE = st.GetEntry(ptype, Xp, sF);
if (SE->_is_local ) {
if (SE->_is_local) {
chi_p = &chi;
if ( SE->_permute ) {
spProjXp(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
if (SE->_permute) {
spProjXp(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else {
spProjXp(chi,in._odata[SE->_offset]);
spProjXp(chi, in._odata[SE->_offset]);
}
} else {
chi_p=&buf[SE->_offset];
} else {
chi_p = &buf[SE->_offset];
}
Impl::multLink(Uchi,U._odata[sU],*chi_p,Xp,SE,st);
spReconXp(result,Uchi);
Impl::multLink(Uchi, U._odata[sU], *chi_p, Xp, SE, st);
spReconXp(result, Uchi);
///////////////////////////
// Yp
///////////////////////////
SE=st.GetEntry(ptype,Yp,sF);
SE = st.GetEntry(ptype, Yp, sF);
if ( SE->_is_local ) {
if (SE->_is_local) {
chi_p = &chi;
if ( SE->_permute ) {
spProjYp(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
if (SE->_permute) {
spProjYp(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else {
spProjYp(chi,in._odata[SE->_offset]);
spProjYp(chi, in._odata[SE->_offset]);
}
} else {
chi_p=&buf[SE->_offset];
} else {
chi_p = &buf[SE->_offset];
}
Impl::multLink(Uchi,U._odata[sU],*chi_p,Yp,SE,st);
accumReconYp(result,Uchi);
Impl::multLink(Uchi, U._odata[sU], *chi_p, Yp, SE, st);
accumReconYp(result, Uchi);
///////////////////////////
// Zp
///////////////////////////
SE=st.GetEntry(ptype,Zp,sF);
SE = st.GetEntry(ptype, Zp, sF);
if ( SE->_is_local ) {
if (SE->_is_local) {
chi_p = &chi;
if ( SE->_permute ) {
spProjZp(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
if (SE->_permute) {
spProjZp(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else {
spProjZp(chi,in._odata[SE->_offset]);
spProjZp(chi, in._odata[SE->_offset]);
}
} else {
chi_p=&buf[SE->_offset];
} else {
chi_p = &buf[SE->_offset];
}
Impl::multLink(Uchi,U._odata[sU],*chi_p,Zp,SE,st);
accumReconZp(result,Uchi);
Impl::multLink(Uchi, U._odata[sU], *chi_p, Zp, SE, st);
accumReconZp(result, Uchi);
///////////////////////////
// Tp
///////////////////////////
SE=st.GetEntry(ptype,Tp,sF);
SE = st.GetEntry(ptype, Tp, sF);
if ( SE->_is_local ) {
if (SE->_is_local) {
chi_p = &chi;
if ( SE->_permute ) {
spProjTp(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
if (SE->_permute) {
spProjTp(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else {
spProjTp(chi,in._odata[SE->_offset]);
spProjTp(chi, in._odata[SE->_offset]);
}
} else {
chi_p=&buf[SE->_offset];
chi_p = &buf[SE->_offset];
}
Impl::multLink(Uchi,U._odata[sU],*chi_p,Tp,SE,st);
accumReconTp(result,Uchi);
Impl::multLink(Uchi, U._odata[sU], *chi_p, Tp, SE, st);
accumReconTp(result, Uchi);
///////////////////////////
// Xm
///////////////////////////
SE=st.GetEntry(ptype,Xm,sF);
SE = st.GetEntry(ptype, Xm, sF);
if ( SE->_is_local ) {
if (SE->_is_local) {
chi_p = &chi;
if ( SE->_permute ) {
spProjXm(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
if (SE->_permute) {
spProjXm(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else {
spProjXm(chi,in._odata[SE->_offset]);
spProjXm(chi, in._odata[SE->_offset]);
}
} else {
chi_p=&buf[SE->_offset];
chi_p = &buf[SE->_offset];
}
Impl::multLink(Uchi,U._odata[sU],*chi_p,Xm,SE,st);
accumReconXm(result,Uchi);
Impl::multLink(Uchi, U._odata[sU], *chi_p, Xm, SE, st);
accumReconXm(result, Uchi);
///////////////////////////
// Ym
///////////////////////////
SE=st.GetEntry(ptype,Ym,sF);
SE = st.GetEntry(ptype, Ym, sF);
if ( SE->_is_local ) {
if (SE->_is_local) {
chi_p = &chi;
if ( SE->_permute ) {
spProjYm(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
if (SE->_permute) {
spProjYm(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else {
spProjYm(chi,in._odata[SE->_offset]);
spProjYm(chi, in._odata[SE->_offset]);
}
} else {
chi_p=&buf[SE->_offset];
chi_p = &buf[SE->_offset];
}
Impl::multLink(Uchi,U._odata[sU],*chi_p,Ym,SE,st);
accumReconYm(result,Uchi);
Impl::multLink(Uchi, U._odata[sU], *chi_p, Ym, SE, st);
accumReconYm(result, Uchi);
///////////////////////////
// Zm
///////////////////////////
SE=st.GetEntry(ptype,Zm,sF);
SE = st.GetEntry(ptype, Zm, sF);
if ( SE->_is_local ) {
if (SE->_is_local) {
chi_p = &chi;
if ( SE->_permute ) {
spProjZm(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
if (SE->_permute) {
spProjZm(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else {
spProjZm(chi,in._odata[SE->_offset]);
spProjZm(chi, in._odata[SE->_offset]);
}
} else {
chi_p=&buf[SE->_offset];
chi_p = &buf[SE->_offset];
}
Impl::multLink(Uchi,U._odata[sU],*chi_p,Zm,SE,st);
accumReconZm(result,Uchi);
Impl::multLink(Uchi, U._odata[sU], *chi_p, Zm, SE, st);
accumReconZm(result, Uchi);
///////////////////////////
// Tm
///////////////////////////
SE=st.GetEntry(ptype,Tm,sF);
SE = st.GetEntry(ptype, Tm, sF);
if ( SE->_is_local ) {
if (SE->_is_local) {
chi_p = &chi;
if ( SE->_permute ) {
spProjTm(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
} else {
spProjTm(chi,in._odata[SE->_offset]);
if (SE->_permute) {
spProjTm(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else {
spProjTm(chi, in._odata[SE->_offset]);
}
} else {
chi_p=&buf[SE->_offset];
chi_p = &buf[SE->_offset];
}
Impl::multLink(Uchi,U._odata[sU],*chi_p,Tm,SE,st);
accumReconTm(result,Uchi);
Impl::multLink(Uchi, U._odata[sU], *chi_p, Tm, SE, st);
accumReconTm(result, Uchi);
vstream(out._odata[sF],result);
vstream(out._odata[sF], result);
};
// Need controls to do interior, exterior, or both
template<class Impl>
void WilsonKernels<Impl>::DiracOptGenericDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int sF,int sU,const FermionField &in, FermionField &out)
{
SiteHalfSpinor tmp;
SiteHalfSpinor chi;
SiteHalfSpinor *chi_p;
// Need controls to do interior, exterior, or both
template <class Impl>
void WilsonKernels<Impl>::DiracOptGenericDhopSite(
StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf, int sF,
int sU, const FermionField &in, FermionField &out) {
SiteHalfSpinor tmp;
SiteHalfSpinor chi;
SiteHalfSpinor *chi_p;
SiteHalfSpinor Uchi;
SiteSpinor result;
StencilEntry *SE;
@@ -285,296 +235,298 @@ void WilsonKernels<Impl>::DiracOptGenericDhopSite(StencilImpl &st,LebesgueOrder
///////////////////////////
// Xp
///////////////////////////
SE=st.GetEntry(ptype,Xm,sF);
SE = st.GetEntry(ptype, Xm, sF);
if ( SE->_is_local ) {
if (SE->_is_local) {
chi_p = &chi;
if ( SE->_permute ) {
spProjXp(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
if (SE->_permute) {
spProjXp(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else {
spProjXp(chi,in._odata[SE->_offset]);
spProjXp(chi, in._odata[SE->_offset]);
}
} else {
chi_p=&buf[SE->_offset];
} else {
chi_p = &buf[SE->_offset];
}
Impl::multLink(Uchi,U._odata[sU],*chi_p,Xm,SE,st);
spReconXp(result,Uchi);
Impl::multLink(Uchi, U._odata[sU], *chi_p, Xm, SE, st);
spReconXp(result, Uchi);
///////////////////////////
// Yp
///////////////////////////
SE=st.GetEntry(ptype,Ym,sF);
SE = st.GetEntry(ptype, Ym, sF);
if ( SE->_is_local ) {
if (SE->_is_local) {
chi_p = &chi;
if ( SE->_permute ) {
spProjYp(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
if (SE->_permute) {
spProjYp(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else {
spProjYp(chi,in._odata[SE->_offset]);
spProjYp(chi, in._odata[SE->_offset]);
}
} else {
chi_p=&buf[SE->_offset];
} else {
chi_p = &buf[SE->_offset];
}
Impl::multLink(Uchi,U._odata[sU],*chi_p,Ym,SE,st);
accumReconYp(result,Uchi);
Impl::multLink(Uchi, U._odata[sU], *chi_p, Ym, SE, st);
accumReconYp(result, Uchi);
///////////////////////////
// Zp
///////////////////////////
SE=st.GetEntry(ptype,Zm,sF);
SE = st.GetEntry(ptype, Zm, sF);
if ( SE->_is_local ) {
if (SE->_is_local) {
chi_p = &chi;
if ( SE->_permute ) {
spProjZp(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
if (SE->_permute) {
spProjZp(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else {
spProjZp(chi,in._odata[SE->_offset]);
spProjZp(chi, in._odata[SE->_offset]);
}
} else {
chi_p=&buf[SE->_offset];
} else {
chi_p = &buf[SE->_offset];
}
Impl::multLink(Uchi,U._odata[sU],*chi_p,Zm,SE,st);
accumReconZp(result,Uchi);
Impl::multLink(Uchi, U._odata[sU], *chi_p, Zm, SE, st);
accumReconZp(result, Uchi);
///////////////////////////
// Tp
///////////////////////////
SE=st.GetEntry(ptype,Tm,sF);
SE = st.GetEntry(ptype, Tm, sF);
if ( SE->_is_local ) {
if (SE->_is_local) {
chi_p = &chi;
if ( SE->_permute ) {
spProjTp(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
if (SE->_permute) {
spProjTp(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else {
spProjTp(chi,in._odata[SE->_offset]);
spProjTp(chi, in._odata[SE->_offset]);
}
} else {
chi_p=&buf[SE->_offset];
chi_p = &buf[SE->_offset];
}
Impl::multLink(Uchi,U._odata[sU],*chi_p,Tm,SE,st);
accumReconTp(result,Uchi);
Impl::multLink(Uchi, U._odata[sU], *chi_p, Tm, SE, st);
accumReconTp(result, Uchi);
///////////////////////////
// Xm
///////////////////////////
SE=st.GetEntry(ptype,Xp,sF);
SE = st.GetEntry(ptype, Xp, sF);
if ( SE->_is_local ) {
if (SE->_is_local) {
chi_p = &chi;
if ( SE->_permute ) {
spProjXm(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
if (SE->_permute) {
spProjXm(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else {
spProjXm(chi,in._odata[SE->_offset]);
spProjXm(chi, in._odata[SE->_offset]);
}
} else {
chi_p=&buf[SE->_offset];
chi_p = &buf[SE->_offset];
}
Impl::multLink(Uchi,U._odata[sU],*chi_p,Xp,SE,st);
accumReconXm(result,Uchi);
Impl::multLink(Uchi, U._odata[sU], *chi_p, Xp, SE, st);
accumReconXm(result, Uchi);
///////////////////////////
// Ym
///////////////////////////
SE=st.GetEntry(ptype,Yp,sF);
SE = st.GetEntry(ptype, Yp, sF);
if ( SE->_is_local ) {
if (SE->_is_local) {
chi_p = &chi;
if ( SE->_permute ) {
spProjYm(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
if (SE->_permute) {
spProjYm(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else {
spProjYm(chi,in._odata[SE->_offset]);
spProjYm(chi, in._odata[SE->_offset]);
}
} else {
chi_p=&buf[SE->_offset];
chi_p = &buf[SE->_offset];
}
Impl::multLink(Uchi,U._odata[sU],*chi_p,Yp,SE,st);
accumReconYm(result,Uchi);
Impl::multLink(Uchi, U._odata[sU], *chi_p, Yp, SE, st);
accumReconYm(result, Uchi);
///////////////////////////
// Zm
///////////////////////////
SE=st.GetEntry(ptype,Zp,sF);
SE = st.GetEntry(ptype, Zp, sF);
if ( SE->_is_local ) {
if (SE->_is_local) {
chi_p = &chi;
if ( SE->_permute ) {
spProjZm(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
if (SE->_permute) {
spProjZm(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else {
spProjZm(chi,in._odata[SE->_offset]);
spProjZm(chi, in._odata[SE->_offset]);
}
} else {
chi_p=&buf[SE->_offset];
chi_p = &buf[SE->_offset];
}
Impl::multLink(Uchi,U._odata[sU],*chi_p,Zp,SE,st);
accumReconZm(result,Uchi);
Impl::multLink(Uchi, U._odata[sU], *chi_p, Zp, SE, st);
accumReconZm(result, Uchi);
///////////////////////////
// Tm
///////////////////////////
SE=st.GetEntry(ptype,Tp,sF);
SE = st.GetEntry(ptype, Tp, sF);
if ( SE->_is_local ) {
if (SE->_is_local) {
chi_p = &chi;
if ( SE->_permute ) {
spProjTm(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
} else {
spProjTm(chi,in._odata[SE->_offset]);
if (SE->_permute) {
spProjTm(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else {
spProjTm(chi, in._odata[SE->_offset]);
}
} else {
chi_p=&buf[SE->_offset];
chi_p = &buf[SE->_offset];
}
Impl::multLink(Uchi,U._odata[sU],*chi_p,Tp,SE,st);
accumReconTm(result,Uchi);
Impl::multLink(Uchi, U._odata[sU], *chi_p, Tp, SE, st);
accumReconTm(result, Uchi);
vstream(out._odata[sF],result);
vstream(out._odata[sF], result);
};
template<class Impl>
void WilsonKernels<Impl>::DiracOptDhopDir(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int sF,int sU,const FermionField &in, FermionField &out,int dir,int gamma)
{
SiteHalfSpinor tmp;
SiteHalfSpinor chi;
SiteSpinor result;
template <class Impl>
void WilsonKernels<Impl>::DiracOptDhopDir(
StencilImpl &st, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf, int sF,
int sU, const FermionField &in, FermionField &out, int dir, int gamma) {
SiteHalfSpinor tmp;
SiteHalfSpinor chi;
SiteSpinor result;
SiteHalfSpinor Uchi;
StencilEntry *SE;
int ptype;
SE=st.GetEntry(ptype,dir,sF);
SE = st.GetEntry(ptype, dir, sF);
// Xp
if(gamma==Xp){
if ( SE->_is_local && SE->_permute ) {
spProjXp(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
} else if ( SE->_is_local ) {
spProjXp(chi,in._odata[SE->_offset]);
} else {
chi=buf[SE->_offset];
if (gamma == Xp) {
if (SE->_is_local && SE->_permute) {
spProjXp(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else if (SE->_is_local) {
spProjXp(chi, in._odata[SE->_offset]);
} else {
chi = buf[SE->_offset];
}
Impl::multLink(Uchi,U._odata[sU],chi,dir,SE,st);
spReconXp(result,Uchi);
Impl::multLink(Uchi, U._odata[sU], chi, dir, SE, st);
spReconXp(result, Uchi);
}
// Yp
if ( gamma==Yp ){
if ( SE->_is_local && SE->_permute ) {
spProjYp(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
} else if ( SE->_is_local ) {
spProjYp(chi,in._odata[SE->_offset]);
} else {
chi=buf[SE->_offset];
if (gamma == Yp) {
if (SE->_is_local && SE->_permute) {
spProjYp(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else if (SE->_is_local) {
spProjYp(chi, in._odata[SE->_offset]);
} else {
chi = buf[SE->_offset];
}
Impl::multLink(Uchi,U._odata[sU],chi,dir,SE,st);
spReconYp(result,Uchi);
Impl::multLink(Uchi, U._odata[sU], chi, dir, SE, st);
spReconYp(result, Uchi);
}
// Zp
if ( gamma ==Zp ){
if ( SE->_is_local && SE->_permute ) {
spProjZp(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
} else if ( SE->_is_local ) {
spProjZp(chi,in._odata[SE->_offset]);
} else {
chi=buf[SE->_offset];
if (gamma == Zp) {
if (SE->_is_local && SE->_permute) {
spProjZp(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else if (SE->_is_local) {
spProjZp(chi, in._odata[SE->_offset]);
} else {
chi = buf[SE->_offset];
}
Impl::multLink(Uchi,U._odata[sU],chi,dir,SE,st);
spReconZp(result,Uchi);
Impl::multLink(Uchi, U._odata[sU], chi, dir, SE, st);
spReconZp(result, Uchi);
}
// Tp
if ( gamma ==Tp ){
if ( SE->_is_local && SE->_permute ) {
spProjTp(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
} else if ( SE->_is_local ) {
spProjTp(chi,in._odata[SE->_offset]);
} else {
chi=buf[SE->_offset];
if (gamma == Tp) {
if (SE->_is_local && SE->_permute) {
spProjTp(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else if (SE->_is_local) {
spProjTp(chi, in._odata[SE->_offset]);
} else {
chi = buf[SE->_offset];
}
Impl::multLink(Uchi,U._odata[sU],chi,dir,SE,st);
spReconTp(result,Uchi);
Impl::multLink(Uchi, U._odata[sU], chi, dir, SE, st);
spReconTp(result, Uchi);
}
// Xm
if ( gamma==Xm ){
if ( SE->_is_local && SE->_permute ) {
spProjXm(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
} else if ( SE->_is_local ) {
spProjXm(chi,in._odata[SE->_offset]);
} else {
chi=buf[SE->_offset];
if (gamma == Xm) {
if (SE->_is_local && SE->_permute) {
spProjXm(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else if (SE->_is_local) {
spProjXm(chi, in._odata[SE->_offset]);
} else {
chi = buf[SE->_offset];
}
Impl::multLink(Uchi,U._odata[sU],chi,dir,SE,st);
spReconXm(result,Uchi);
Impl::multLink(Uchi, U._odata[sU], chi, dir, SE, st);
spReconXm(result, Uchi);
}
// Ym
if ( gamma == Ym ){
if ( SE->_is_local && SE->_permute ) {
spProjYm(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
} else if ( SE->_is_local ) {
spProjYm(chi,in._odata[SE->_offset]);
} else {
chi=buf[SE->_offset];
if (gamma == Ym) {
if (SE->_is_local && SE->_permute) {
spProjYm(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else if (SE->_is_local) {
spProjYm(chi, in._odata[SE->_offset]);
} else {
chi = buf[SE->_offset];
}
Impl::multLink(Uchi,U._odata[sU],chi,dir,SE,st);
spReconYm(result,Uchi);
Impl::multLink(Uchi, U._odata[sU], chi, dir, SE, st);
spReconYm(result, Uchi);
}
// Zm
if ( gamma == Zm ){
if ( SE->_is_local && SE->_permute ) {
spProjZm(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
} else if ( SE->_is_local ) {
spProjZm(chi,in._odata[SE->_offset]);
} else {
chi=buf[SE->_offset];
if (gamma == Zm) {
if (SE->_is_local && SE->_permute) {
spProjZm(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else if (SE->_is_local) {
spProjZm(chi, in._odata[SE->_offset]);
} else {
chi = buf[SE->_offset];
}
Impl::multLink(Uchi,U._odata[sU],chi,dir,SE,st);
spReconZm(result,Uchi);
}
// Tm
if ( gamma==Tm ) {
if ( SE->_is_local && SE->_permute ) {
spProjTm(tmp,in._odata[SE->_offset]);
permute(chi,tmp,ptype);
} else if ( SE->_is_local ) {
spProjTm(chi,in._odata[SE->_offset]);
} else {
chi=buf[SE->_offset];
}
Impl::multLink(Uchi,U._odata[sU],chi,dir,SE,st);
spReconTm(result,Uchi);
Impl::multLink(Uchi, U._odata[sU], chi, dir, SE, st);
spReconZm(result, Uchi);
}
vstream(out._odata[sF],result);
// Tm
if (gamma == Tm) {
if (SE->_is_local && SE->_permute) {
spProjTm(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else if (SE->_is_local) {
spProjTm(chi, in._odata[SE->_offset]);
} else {
chi = buf[SE->_offset];
}
Impl::multLink(Uchi, U._odata[sU], chi, dir, SE, st);
spReconTm(result, Uchi);
}
vstream(out._odata[sF], result);
}
FermOpTemplateInstantiate(WilsonKernels);
FermOpTemplateInstantiate(WilsonKernels);
AdjointFermOpTemplateInstantiate(WilsonKernels);
TwoIndexFermOpTemplateInstantiate(WilsonKernels);
}}

238
lib/qcd/action/fermion/WilsonKernels.h
View File

@@ -1,34 +1,35 @@
/*************************************************************************************
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/action/fermion/WilsonKernels.h
Source file: ./lib/qcd/action/fermion/WilsonKernels.h
Copyright (C) 2015
Copyright (C) 2015
Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
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 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.
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.
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 GRID_QCD_DHOP_H
#define GRID_QCD_DHOP_H
See the full license in the file "LICENSE" in the top level distribution
directory
*************************************************************************************/
/* END LEGAL */
#ifndef GRID_QCD_DHOP_H
#define GRID_QCD_DHOP_H
namespace Grid {
@@ -48,55 +49,158 @@ namespace Grid {
template<class Impl> class WilsonKernels : public FermionOperator<Impl> , public WilsonKernelsStatic {
public:
INHERIT_IMPL_TYPES(Impl);
typedef FermionOperator<Impl> Base;
INHERIT_IMPL_TYPES(Impl);
typedef FermionOperator<Impl> Base;
public:
void DiracOptDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU,int Ls, int Ns, const FermionField &in, FermionField &out);
void DiracOptDhopSiteDag(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int sF,int sU,int Ls, int Ns, const FermionField &in,FermionField &out);
template <bool EnableBool = true>
typename std::enable_if<Impl::Dimension == 3 && Nc == 3 &&EnableBool, void>::type
DiracOptDhopSite(
StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU, int Ls, int Ns, const FermionField &in,
FermionField &out) {
#ifdef AVX512
if (AsmOpt) {
WilsonKernels<Impl>::DiracOptAsmDhopSite(st, lo, U, buf, sF, sU, Ls, Ns,
in, out);
} else {
#else
{
#endif
for (int site = 0; site < Ns; site++) {
for (int s = 0; s < Ls; s++) {
if (HandOpt)
WilsonKernels<Impl>::DiracOptHandDhopSite(st, lo, U, buf, sF, sU,
in, out);
else
WilsonKernels<Impl>::DiracOptGenericDhopSite(st, lo, U, buf, sF, sU,
in, out);
sF++;
}
sU++;
}
}
}
template <bool EnableBool = true>
typename std::enable_if<(Impl::Dimension != 3 || (Impl::Dimension == 3 && Nc != 3)) && EnableBool, void>::type
DiracOptDhopSite(
StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU, int Ls, int Ns, const FermionField &in,
FermionField &out) {
for (int site = 0; site < Ns; site++) {
for (int s = 0; s < Ls; s++) {
WilsonKernels<Impl>::DiracOptGenericDhopSite(st, lo, U, buf, sF, sU, in,
out);
sF++;
}
sU++;
}
}
template <bool EnableBool = true>
typename std::enable_if<Impl::Dimension == 3 && Nc == 3 && EnableBool,
void>::type
DiracOptDhopSiteDag(
StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU, int Ls, int Ns, const FermionField &in,
FermionField &out) {
#ifdef AVX512
if (AsmOpt) {
WilsonKernels<Impl>::DiracOptAsmDhopSiteDag(st, lo, U, buf, sF, sU, Ls,
Ns, in, out);
} else {
#else
{
#endif
for (int site = 0; site < Ns; site++) {
for (int s = 0; s < Ls; s++) {
if (HandOpt)
WilsonKernels<Impl>::DiracOptHandDhopSiteDag(st, lo, U, buf, sF, sU,
in, out);
else
WilsonKernels<Impl>::DiracOptGenericDhopSiteDag(st, lo, U, buf, sF,
sU, in, out);
sF++;
}
sU++;
}
}
}
template <bool EnableBool = true>
typename std::enable_if<
(Impl::Dimension != 3 || (Impl::Dimension == 3 && Nc != 3)) && EnableBool,
void>::type
DiracOptDhopSiteDag(
StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU, int Ls, int Ns, const FermionField &in,
FermionField &out) {
for (int site = 0; site < Ns; site++) {
for (int s = 0; s < Ls; s++) {
WilsonKernels<Impl>::DiracOptGenericDhopSiteDag(st, lo, U, buf, sF, sU,
in, out);
sF++;
}
sU++;
}
}
void DiracOptDhopDir(
StencilImpl &st, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU, const FermionField &in, FermionField &out, int dirdisp,
int gamma);
private:
// Specialised variants
void DiracOptGenericDhopSite(
StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU, const FermionField &in, FermionField &out);
void DiracOptGenericDhopSiteDag(
StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU, const FermionField &in, FermionField &out);
void DiracOptAsmDhopSite(
StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU, int Ls, int Ns, const FermionField &in,
FermionField &out);
void DiracOptAsmDhopSiteDag(
StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU, int Ls, int Ns, const FermionField &in,
FermionField &out);
void DiracOptHandDhopSite(
StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU, const FermionField &in, FermionField &out);
void DiracOptHandDhopSiteDag(
StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
std::vector<SiteHalfSpinor, alignedAllocator<SiteHalfSpinor> > &buf,
int sF, int sU, const FermionField &in, FermionField &out);
public:
WilsonKernels(const ImplParams &p = ImplParams());
};
}
}
void DiracOptDhopDir(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int sF,int sU,const FermionField &in, FermionField &out,int dirdisp,int gamma);
private:
// Specialised variants
void DiracOptGenericDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int sF,int sU, const FermionField &in, FermionField &out);
void DiracOptGenericDhopSiteDag(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int sF,int sU,const FermionField &in,FermionField &out);
void DiracOptAsmDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int sF,int sU,int Ls, int Ns, const FermionField &in, FermionField &out);
void DiracOptAsmDhopSiteDag(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int sF,int sU,int Ls, int Ns, const FermionField &in, FermionField &out);
void DiracOptHandDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int sF,int sU,const FermionField &in, FermionField &out);
void DiracOptHandDhopSiteDag(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int sF,int sU,const FermionField &in, FermionField &out);
public:
WilsonKernels(const ImplParams &p= ImplParams());
};
}
}
#endif

157
lib/qcd/action/fermion/WilsonKernelsAsm.cc
View File

@@ -1,4 +1,4 @@
/*************************************************************************************
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
@@ -26,68 +26,77 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
*************************************************************************************/
/* END LEGAL */
#include <Grid.h>
namespace Grid {
namespace QCD {
namespace QCD {
///////////////////////////////////////////////////////////
// Default to no assembler implementation
///////////////////////////////////////////////////////////
template<class Impl>
void WilsonKernels<Impl >::DiracOptAsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
{
assert(0);
}
template<class Impl>
void WilsonKernels<Impl >::DiracOptAsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
{
assert(0);
}
///////////////////////////////////////////////////////////
// Default to no assembler implementation
///////////////////////////////////////////////////////////
template<class Impl>
void WilsonKernels<Impl >::DiracOptAsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
{
assert(0);
}
#if defined(AVX512)
///////////////////////////////////////////////////////////
// If we are AVX512 specialise the single precision routine
///////////////////////////////////////////////////////////
///////////////////////////////////////////////////////////
// If we are AVX512 specialise the single precision routine
///////////////////////////////////////////////////////////
#include <simd/Intel512wilson.h>
#include <simd/Intel512single.h>
static Vector<vComplexF> signs;
int setupSigns(void ){
Vector<vComplexF> bother(2);
signs = bother;
vrsign(signs[0]);
visign(signs[1]);
return 1;
}
static int signInit = setupSigns();
static Vector<vComplexF> signs;
int setupSigns(void ){
Vector<vComplexF> bother(2);
signs = bother;
vrsign(signs[0]);
visign(signs[1]);
return 1;
}
static int signInit = setupSigns();
#define label(A) ilabel(A)
#define ilabel(A) ".globl\n" #A ":\n"
#define MAYBEPERM(A,perm) if (perm) { A ; }
#define MULT_2SPIN(ptr,pf) MULT_ADDSUB_2SPIN(ptr,pf)
#define FX(A) WILSONASM_ ##A
#undef KERNEL_DAG
template<>
void WilsonKernels<WilsonImplF>::DiracOptAsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
template<>
void WilsonKernels<WilsonImplF>::DiracOptAsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
#define KERNEL_DAG
template<>
void WilsonKernels<WilsonImplF>::DiracOptAsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
template<>
void WilsonKernels<WilsonImplF>::DiracOptAsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
#undef VMOVIDUP
#undef VMOVRDUP
#undef MAYBEPERM
@@ -98,43 +107,43 @@ void WilsonKernels<WilsonImplF>::DiracOptAsmDhopSiteDag(StencilImpl &st,Lebesgue
#define VMOVIDUP(A,B,C) VBCASTIDUPf(A,B,C)
#define VMOVRDUP(A,B,C) VBCASTRDUPf(A,B,C)
#define MULT_2SPIN(ptr,pf) MULT_ADDSUB_2SPIN_LS(ptr,pf)
#undef KERNEL_DAG
template<>
void WilsonKernels<DomainWallVec5dImplF>::DiracOptAsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
template<>
void WilsonKernels<DomainWallVec5dImplF>::DiracOptAsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
#define KERNEL_DAG
template<>
void WilsonKernels<DomainWallVec5dImplF>::DiracOptAsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
template<>
void WilsonKernels<DomainWallVec5dImplF>::DiracOptAsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
#endif
#define INSTANTIATE_ASM(A)\
template void WilsonKernels<A>::DiracOptAsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,\
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,\
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out);\
template void WilsonKernels<A>::DiracOptAsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,\
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,\
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out);\
template void WilsonKernels<WilsonImplF>::DiracOptAsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out);
template void WilsonKernels<WilsonImplD>::DiracOptAsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out);
template void WilsonKernels<GparityWilsonImplF>::DiracOptAsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out);
template void WilsonKernels<GparityWilsonImplD>::DiracOptAsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out);
template void WilsonKernels<DomainWallVec5dImplF>::DiracOptAsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out);
template void WilsonKernels<DomainWallVec5dImplD>::DiracOptAsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out);
}}
INSTANTIATE_ASM(WilsonImplF);
INSTANTIATE_ASM(WilsonImplD);
INSTANTIATE_ASM(ZWilsonImplF);
INSTANTIATE_ASM(ZWilsonImplD);
INSTANTIATE_ASM(GparityWilsonImplF);
INSTANTIATE_ASM(GparityWilsonImplD);
INSTANTIATE_ASM(DomainWallVec5dImplF);
INSTANTIATE_ASM(DomainWallVec5dImplD);
INSTANTIATE_ASM(ZDomainWallVec5dImplF);
INSTANTIATE_ASM(ZDomainWallVec5dImplD);
}
}

4
lib/qcd/action/fermion/WilsonKernelsAsmBody.h
View File

@@ -134,7 +134,9 @@
////////////////////////////////
// Xm
////////////////////////////////
#ifndef STREAM_STORE
basep= (uint64_t) &out._odata[ss];
#endif
// basep= st.GetPFInfo(nent,plocal); nent++;
if ( local ) {
LOAD64(%r10,isigns); // times i => shuffle and xor the real part sign bit
@@ -229,7 +231,9 @@
LOAD_CHI(base);
}
base= (uint64_t) &out._odata[ss];
#ifndef STREAM_STORE
PREFETCH_CHIMU(base);
#endif
{
MULT_2SPIN_DIR_PFTM(Tm,basep);
}

8
lib/qcd/action/fermion/WilsonKernelsHand.cc
View File

@@ -311,8 +311,8 @@ namespace Grid {
namespace QCD {
template<class Impl>
void WilsonKernels<Impl >::DiracOptHandDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,
template<class Impl>
void WilsonKernels<Impl>::DiracOptHandDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int sU,const FermionField &in, FermionField &out)
{
@@ -554,8 +554,8 @@ void WilsonKernels<Impl >::DiracOptHandDhopSite(StencilImpl &st,LebesgueOrder &l
}
}
template<class Impl>
void WilsonKernels<Impl >::DiracOptHandDhopSiteDag(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,
template<class Impl>
void WilsonKernels<Impl>::DiracOptHandDhopSiteDag(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int sU,const FermionField &in, FermionField &out)
{