portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-21 01:02:02 +01:00
Code Releases Activity

Merge branch 'feature/dirichlet' of https://github.com/paboyle/Grid into feature/dirichlet

Browse Source
This commit is contained in:
Peter Boyle
2023-03-21 16:04:27 -04:00
parent 8b43be39c0 e1c326558a
commit 7db8dd7a95
19 changed files with 377 additions and 93 deletions
Download Patch File Download Diff File Expand all files Collapse all files

34
Grid/qcd/action/fermion/WilsonCompressor.h
View File

@ -484,24 +484,26 @@ public:
int dag = compress.dag;
int face_idx=0;
#define vet_same_node(a,b) \
{ auto tmp = b; }
if ( dag ) {
assert(this->same_node[Xp]==this->HaloGatherDir(source,XpCompress,Xp,face_idx));
assert(this->same_node[Yp]==this->HaloGatherDir(source,YpCompress,Yp,face_idx));
assert(this->same_node[Zp]==this->HaloGatherDir(source,ZpCompress,Zp,face_idx));
assert(this->same_node[Tp]==this->HaloGatherDir(source,TpCompress,Tp,face_idx));
assert(this->same_node[Xm]==this->HaloGatherDir(source,XmCompress,Xm,face_idx));
assert(this->same_node[Ym]==this->HaloGatherDir(source,YmCompress,Ym,face_idx));
assert(this->same_node[Zm]==this->HaloGatherDir(source,ZmCompress,Zm,face_idx));
assert(this->same_node[Tm]==this->HaloGatherDir(source,TmCompress,Tm,face_idx));
vet_same_node(this->same_node[Xp],this->HaloGatherDir(source,XpCompress,Xp,face_idx));
vet_same_node(this->same_node[Yp],this->HaloGatherDir(source,YpCompress,Yp,face_idx));
vet_same_node(this->same_node[Zp],this->HaloGatherDir(source,ZpCompress,Zp,face_idx));
vet_same_node(this->same_node[Tp],this->HaloGatherDir(source,TpCompress,Tp,face_idx));
vet_same_node(this->same_node[Xm],this->HaloGatherDir(source,XmCompress,Xm,face_idx));
vet_same_node(this->same_node[Ym],this->HaloGatherDir(source,YmCompress,Ym,face_idx));
vet_same_node(this->same_node[Zm],this->HaloGatherDir(source,ZmCompress,Zm,face_idx));
vet_same_node(this->same_node[Tm],this->HaloGatherDir(source,TmCompress,Tm,face_idx));
} else {
assert(this->same_node[Xp]==this->HaloGatherDir(source,XmCompress,Xp,face_idx));
assert(this->same_node[Yp]==this->HaloGatherDir(source,YmCompress,Yp,face_idx));
assert(this->same_node[Zp]==this->HaloGatherDir(source,ZmCompress,Zp,face_idx));
assert(this->same_node[Tp]==this->HaloGatherDir(source,TmCompress,Tp,face_idx));
assert(this->same_node[Xm]==this->HaloGatherDir(source,XpCompress,Xm,face_idx));
assert(this->same_node[Ym]==this->HaloGatherDir(source,YpCompress,Ym,face_idx));
assert(this->same_node[Zm]==this->HaloGatherDir(source,ZpCompress,Zm,face_idx));
assert(this->same_node[Tm]==this->HaloGatherDir(source,TpCompress,Tm,face_idx));
vet_same_node(this->same_node[Xp],this->HaloGatherDir(source,XmCompress,Xp,face_idx));
vet_same_node(this->same_node[Yp],this->HaloGatherDir(source,YmCompress,Yp,face_idx));
vet_same_node(this->same_node[Zp],this->HaloGatherDir(source,ZmCompress,Zp,face_idx));
vet_same_node(this->same_node[Tp],this->HaloGatherDir(source,TmCompress,Tp,face_idx));
vet_same_node(this->same_node[Xm],this->HaloGatherDir(source,XpCompress,Xm,face_idx));
vet_same_node(this->same_node[Ym],this->HaloGatherDir(source,YpCompress,Ym,face_idx));
vet_same_node(this->same_node[Zm],this->HaloGatherDir(source,ZpCompress,Zm,face_idx));
vet_same_node(this->same_node[Tm],this->HaloGatherDir(source,TpCompress,Tm,face_idx));
}
this->face_table_computed=1;
assert(this->u_comm_offset==this->_unified_buffer_size);

7
Grid/qcd/action/fermion/WilsonKernels.h
View File

@ -52,13 +52,6 @@ public:
typedef AcceleratorVector<int,STENCIL_MAX> StencilVector;
public:
#ifdef GRID_SYCL
#define SYCL_HACK
#endif
#ifdef SYCL_HACK
static void HandDhopSiteSycl(StencilVector st_perm,StencilEntry *st_p, SiteDoubledGaugeField *U,SiteHalfSpinor *buf,
int ss,int sU,const SiteSpinor *in, SiteSpinor *out);
#endif
static void DhopKernel(int Opt,StencilImpl &st, DoubledGaugeField &U, SiteHalfSpinor * buf,
int Ls, int Nsite, const FermionField &in, FermionField &out,

4
Grid/qcd/action/fermion/implementation/WilsonFermion5DImplementation.h
View File

@ -63,6 +63,10 @@ WilsonFermion5D<Impl>::WilsonFermion5D(GaugeField &_Umu,
_tmp(&FiveDimRedBlackGrid),
Dirichlet(0)
{
Stencil.lo = &Lebesgue;
StencilEven.lo = &LebesgueEvenOdd;
StencilOdd.lo = &LebesgueEvenOdd;
// some assertions
assert(FiveDimGrid._ndimension==5);
assert(FourDimGrid._ndimension==4);

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

@ -60,6 +60,9 @@ WilsonFermion<Impl>::WilsonFermion(GaugeField &_Umu, GridCartesian &Fgrid,
_tmp(&Hgrid),
anisotropyCoeff(anis)
{
Stencil.lo = &Lebesgue;
StencilEven.lo = &LebesgueEvenOdd;
StencilOdd.lo = &LebesgueEvenOdd;
// Allocate the required comms buffer
ImportGauge(_Umu);
if (anisotropyCoeff.isAnisotropic){

14
Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h
View File

@ -433,11 +433,23 @@ void WilsonKernels<Impl>::DhopDirKernel( StencilImpl &st, DoubledGaugeField &U,S
});
#define ASM_CALL(A) \
thread_for( ss, Nsite, { \
thread_for( sss, Nsite, { \
int ss = st.lo->Reorder(sss); \
int sU = ss; \
int sF = ss*Ls; \
WilsonKernels<Impl>::A(st_v,U_v,buf,sF,sU,Ls,1,in_v,out_v); \
});
#define ASM_CALL_SLICE(A) \
auto grid = in.Grid() ; \
int nt = grid->LocalDimensions()[4]; \
int nxyz = Nsite/nt ; \
for(int t=0;t<nt;t++){ \
thread_for( sss, nxyz, { \
int ss = t*nxyz+sss; \
int sU = ss; \
int sF = ss*Ls; \
WilsonKernels<Impl>::A(st_v,U_v,buf,sF,sU,Ls,1,in_v,out_v); \
});}
template <class Impl>
void WilsonKernels<Impl>::DhopKernel(int Opt,StencilImpl &st, DoubledGaugeField &U, SiteHalfSpinor * buf,

2
Grid/qcd/action/pseudofermion/GeneralEvenOddRationalRatio.h
View File

@ -127,6 +127,8 @@ NAMESPACE_BEGIN(Grid);
ApproxNegPowerAction.tolerances[i] = action_tolerance[i];
ApproxHalfPowerAction.tolerances[i] = action_tolerance[i];
ApproxNegHalfPowerAction.tolerances[i]= action_tolerance[i];
}
for(int i=0;i<ApproxPowerMD.tolerances.size();i++){
ApproxPowerMD.tolerances[i] = md_tolerance[i];
ApproxNegPowerMD.tolerances[i] = md_tolerance[i];
ApproxHalfPowerMD.tolerances[i] = md_tolerance[i];

11
Grid/qcd/action/pseudofermion/GeneralEvenOddRationalRatioMixedPrec.h
View File

@ -29,6 +29,8 @@
#ifndef QCD_PSEUDOFERMION_GENERAL_EVEN_ODD_RATIONAL_RATIO_MIXED_PREC_H
#define QCD_PSEUDOFERMION_GENERAL_EVEN_ODD_RATIONAL_RATIO_MIXED_PREC_H
#include <Grid/algorithms/iterative/ConjugateGradientMultiShiftCleanup.h>
NAMESPACE_BEGIN(Grid);
/////////////////////////////////////////////////////////////////////////////////////////////////////////////
@ -58,7 +60,7 @@ NAMESPACE_BEGIN(Grid);
//Allow derived classes to override the multishift CG
virtual void multiShiftInverse(bool numerator, const MultiShiftFunction &approx, const Integer MaxIter, const FermionFieldD &in, FermionFieldD &out){
#if 0
SchurDifferentiableOperator<ImplD> schurOp(numerator ? NumOp : DenOp);
SchurDifferentiableOperator<ImplD> schurOp(numerator ? NumOpD : DenOpD);
ConjugateGradientMultiShift<FermionFieldD> msCG(MaxIter, approx);
msCG(schurOp,in, out);
#else
@ -66,7 +68,8 @@ NAMESPACE_BEGIN(Grid);
SchurDifferentiableOperator<ImplF> schurOpF(numerator ? NumOpF : DenOpF);
FermionFieldD2 inD2(NumOpD2.FermionRedBlackGrid());
FermionFieldD2 outD2(NumOpD2.FermionRedBlackGrid());
// Action better with higher precision?
ConjugateGradientMultiShiftMixedPrec<FermionFieldD2, FermionFieldF> msCG(MaxIter, approx, NumOpF.FermionRedBlackGrid(), schurOpF, ReliableUpdateFreq);
precisionChange(inD2,in);
std::cout << "msCG single solve "<<norm2(inD2)<<" " <<norm2(in)<<std::endl;
@ -76,12 +79,12 @@ NAMESPACE_BEGIN(Grid);
}
virtual void multiShiftInverse(bool numerator, const MultiShiftFunction &approx, const Integer MaxIter, const FermionFieldD &in, std::vector<FermionFieldD> &out_elems, FermionFieldD &out){
SchurDifferentiableOperator<ImplD2> schurOpD2(numerator ? NumOpD2 : DenOpD2);
SchurDifferentiableOperator<ImplF> schurOpF(numerator ? NumOpF : DenOpF);
SchurDifferentiableOperator<ImplF> schurOpF (numerator ? NumOpF : DenOpF);
FermionFieldD2 inD2(NumOpD2.FermionRedBlackGrid());
FermionFieldD2 outD2(NumOpD2.FermionRedBlackGrid());
std::vector<FermionFieldD2> out_elemsD2(out_elems.size(),NumOpD2.FermionRedBlackGrid());
ConjugateGradientMultiShiftMixedPrec<FermionFieldD2, FermionFieldF> msCG(MaxIter, approx, NumOpF.FermionRedBlackGrid(), schurOpF, ReliableUpdateFreq);
ConjugateGradientMultiShiftMixedPrecCleanup<FermionFieldD2, FermionFieldF> msCG(MaxIter, approx, NumOpF.FermionRedBlackGrid(), schurOpF, ReliableUpdateFreq);
precisionChange(inD2,in);
std::cout << "msCG in "<<norm2(inD2)<<" " <<norm2(in)<<std::endl;
msCG(schurOpD2, inD2, out_elemsD2, outD2);