portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-11-03 21:44:33 +00:00
Code Releases Activity

Some small steps towards a multigrid

Browse Source
This commit is contained in:
Peter Boyle
2015-06-22 12:49:44 +01:00
parent e415587e8f
commit dec68e5c0e
12 changed files with 243 additions and 278 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
lib/Make.inc
View File

@@ -1,4 +1,4 @@
HFILES=./algorithms/approx/bigfloat.h ./algorithms/approx/bigfloat_double.h ./algorithms/approx/Chebyshev.h ./algorithms/approx/MultiShiftFunction.h ./algorithms/approx/Remez.h ./algorithms/approx/Zolotarev.h ./algorithms/CoarsenedMatrix.h ./algorithms/iterative/AdefGeneric.h ./algorithms/iterative/ConjugateGradient.h ./algorithms/iterative/ConjugateGradientMultiShift.h ./algorithms/iterative/ConjugateResidual.h ./algorithms/iterative/NormalEquations.h ./algorithms/iterative/PrecGeneralisedConjugateResidual.h ./algorithms/iterative/SchurRedBlack.h ./algorithms/LinearOperator.h ./algorithms/SparseMatrix.h ./Algorithms.h ./AlignedAllocator.h ./cartesian/Cartesian_base.h ./cartesian/Cartesian_full.h ./cartesian/Cartesian_red_black.h ./Cartesian.h ./communicator/Communicator_base.h ./Communicator.h ./cshift/Cshift_common.h ./cshift/Cshift_mpi.h ./cshift/Cshift_none.h ./Cshift.h ./Grid.h ./GridConfig.h ./lattice/Lattice_arith.h ./lattice/Lattice_base.h ./lattice/Lattice_comparison.h ./lattice/Lattice_comparison_utils.h ./lattice/Lattice_conformable.h ./lattice/Lattice_coordinate.h ./lattice/Lattice_ET.h ./lattice/Lattice_local.h ./lattice/Lattice_overload.h ./lattice/Lattice_peekpoke.h ./lattice/Lattice_reality.h ./lattice/Lattice_reduction.h ./lattice/Lattice_rng.h ./lattice/Lattice_trace.h ./lattice/Lattice_transfer.h ./lattice/Lattice_transpose.h ./lattice/Lattice_unary.h ./lattice/Lattice_where.h ./Lattice.h ./parallelIO/NerscIO.h ./qcd/action/Actions.h ./qcd/action/fermion/CayleyFermion5D.h ./qcd/action/fermion/ContinuedFractionFermion5D.h ./qcd/action/fermion/DomainWallFermion.h ./qcd/action/fermion/FermionOperator.h ./qcd/action/fermion/g5HermitianLinop.h ./qcd/action/fermion/MobiusFermion.h ./qcd/action/fermion/MobiusZolotarevFermion.h ./qcd/action/fermion/OverlapWilsonCayleyTanhFermion.h ./qcd/action/fermion/OverlapWilsonCayleyZolotarevFermion.h ./qcd/action/fermion/OverlapWilsonContfracTanhFermion.h ./qcd/action/fermion/OverlapWilsonContfracZolotarevFermion.h ./qcd/action/fermion/OverlapWilsonPartialFractionTanhFermion.h ./qcd/action/fermion/OverlapWilsonPartialFractionZolotarevFermion.h ./qcd/action/fermion/PartialFractionFermion5D.h ./qcd/action/fermion/ScaledShamirFermion.h ./qcd/action/fermion/ShamirZolotarevFermion.h ./qcd/action/fermion/WilsonCompressor.h ./qcd/action/fermion/WilsonFermion.h ./qcd/action/fermion/WilsonFermion5D.h ./qcd/action/fermion/WilsonKernels.h ./qcd/action/gauge/GaugeActionBase.h ./qcd/action/gauge/WilsonGaugeAction.h ./qcd/QCD.h ./qcd/spin/Dirac.h ./qcd/spin/TwoSpinor.h ./qcd/utils/CovariantCshift.h ./qcd/utils/LinalgUtils.h ./qcd/utils/SpaceTimeGrid.h ./qcd/utils/SUn.h ./qcd/utils/WilsonLoops.h ./simd/Grid_avx.h ./simd/Grid_avx512.h ./simd/Grid_empty.h ./simd/Grid_neon.h ./simd/Grid_qpx.h ./simd/Grid_sse4.h ./simd/Grid_vector_types.h ./simd/Grid_vector_unops.h ./Simd.h ./stencil/Lebesgue.h ./Stencil.h ./tensors/Tensor_arith.h ./tensors/Tensor_arith_add.h ./tensors/Tensor_arith_mac.h ./tensors/Tensor_arith_mul.h ./tensors/Tensor_arith_scalar.h ./tensors/Tensor_arith_sub.h ./tensors/Tensor_class.h ./tensors/Tensor_determinant.h ./tensors/Tensor_exp.h ./tensors/Tensor_extract_merge.h ./tensors/Tensor_inner.h ./tensors/Tensor_logical.h ./tensors/Tensor_outer.h ./tensors/Tensor_peek.h ./tensors/Tensor_poke.h ./tensors/Tensor_reality.h ./tensors/Tensor_Ta.h ./tensors/Tensor_trace.h ./tensors/Tensor_traits.h ./tensors/Tensor_transpose.h ./tensors/Tensor_unary.h ./Tensors.h ./Threads.h
HFILES=./algorithms/approx/bigfloat.h ./algorithms/approx/bigfloat_double.h ./algorithms/approx/Chebyshev.h ./algorithms/approx/MultiShiftFunction.h ./algorithms/approx/Remez.h ./algorithms/approx/Zolotarev.h ./algorithms/CoarsenedMatrix.h ./algorithms/iterative/AdefGeneric.h ./algorithms/iterative/BfmHDCG.h ./algorithms/iterative/ConjugateGradient.h ./algorithms/iterative/ConjugateGradientMultiShift.h ./algorithms/iterative/ConjugateResidual.h ./algorithms/iterative/NormalEquations.h ./algorithms/iterative/PrecGeneralisedConjugateResidual.h ./algorithms/iterative/SchurRedBlack.h ./algorithms/LinearOperator.h ./algorithms/Preconditioner.h ./algorithms/SparseMatrix.h ./Algorithms.h ./AlignedAllocator.h ./cartesian/Cartesian_base.h ./cartesian/Cartesian_full.h ./cartesian/Cartesian_red_black.h ./Cartesian.h ./communicator/Communicator_base.h ./Communicator.h ./cshift/Cshift_common.h ./cshift/Cshift_mpi.h ./cshift/Cshift_none.h ./Cshift.h ./Grid.h ./GridConfig.h ./lattice/Lattice_arith.h ./lattice/Lattice_base.h ./lattice/Lattice_comparison.h ./lattice/Lattice_comparison_utils.h ./lattice/Lattice_conformable.h ./lattice/Lattice_coordinate.h ./lattice/Lattice_ET.h ./lattice/Lattice_local.h ./lattice/Lattice_overload.h ./lattice/Lattice_peekpoke.h ./lattice/Lattice_reality.h ./lattice/Lattice_reduction.h ./lattice/Lattice_rng.h ./lattice/Lattice_trace.h ./lattice/Lattice_transfer.h ./lattice/Lattice_transpose.h ./lattice/Lattice_unary.h ./lattice/Lattice_where.h ./Lattice.h ./parallelIO/NerscIO.h ./qcd/action/Actions.h ./qcd/action/DiffAction.h ./qcd/action/fermion/CayleyFermion5D.h ./qcd/action/fermion/ContinuedFractionFermion5D.h ./qcd/action/fermion/DomainWallFermion.h ./qcd/action/fermion/FermionOperator.h ./qcd/action/fermion/g5HermitianLinop.h ./qcd/action/fermion/MobiusFermion.h ./qcd/action/fermion/MobiusZolotarevFermion.h ./qcd/action/fermion/OverlapWilsonCayleyTanhFermion.h ./qcd/action/fermion/OverlapWilsonCayleyZolotarevFermion.h ./qcd/action/fermion/OverlapWilsonContfracTanhFermion.h ./qcd/action/fermion/OverlapWilsonContfracZolotarevFermion.h ./qcd/action/fermion/OverlapWilsonPartialFractionTanhFermion.h ./qcd/action/fermion/OverlapWilsonPartialFractionZolotarevFermion.h ./qcd/action/fermion/PartialFractionFermion5D.h ./qcd/action/fermion/ScaledShamirFermion.h ./qcd/action/fermion/ShamirZolotarevFermion.h ./qcd/action/fermion/WilsonCompressor.h ./qcd/action/fermion/WilsonFermion.h ./qcd/action/fermion/WilsonFermion5D.h ./qcd/action/fermion/WilsonKernels.h ./qcd/action/gauge/GaugeActionBase.h ./qcd/action/gauge/WilsonGaugeAction.h ./qcd/QCD.h ./qcd/spin/Dirac.h ./qcd/spin/TwoSpinor.h ./qcd/utils/CovariantCshift.h ./qcd/utils/LinalgUtils.h ./qcd/utils/SpaceTimeGrid.h ./qcd/utils/SUn.h ./qcd/utils/WilsonLoops.h ./simd/Grid_avx.h ./simd/Grid_avx512.h ./simd/Grid_empty.h ./simd/Grid_neon.h ./simd/Grid_qpx.h ./simd/Grid_sse4.h ./simd/Grid_vector_types.h ./simd/Grid_vector_unops.h ./Simd.h ./stencil/Lebesgue.h ./Stencil.h ./tensors/Tensor_arith.h ./tensors/Tensor_arith_add.h ./tensors/Tensor_arith_mac.h ./tensors/Tensor_arith_mul.h ./tensors/Tensor_arith_scalar.h ./tensors/Tensor_arith_sub.h ./tensors/Tensor_class.h ./tensors/Tensor_determinant.h ./tensors/Tensor_exp.h ./tensors/Tensor_extract_merge.h ./tensors/Tensor_inner.h ./tensors/Tensor_logical.h ./tensors/Tensor_outer.h ./tensors/Tensor_peek.h ./tensors/Tensor_poke.h ./tensors/Tensor_reality.h ./tensors/Tensor_Ta.h ./tensors/Tensor_trace.h ./tensors/Tensor_traits.h ./tensors/Tensor_transpose.h ./tensors/Tensor_unary.h ./Tensors.h ./Threads.h
CCFILES=./algorithms/approx/MultiShiftFunction.cc ./algorithms/approx/Remez.cc ./algorithms/approx/Zolotarev.cc ./GridInit.cc ./qcd/action/fermion/CayleyFermion5D.cc ./qcd/action/fermion/ContinuedFractionFermion5D.cc ./qcd/action/fermion/PartialFractionFermion5D.cc ./qcd/action/fermion/WilsonFermion.cc ./qcd/action/fermion/WilsonFermion5D.cc ./qcd/action/fermion/WilsonKernels.cc ./qcd/action/fermion/WilsonKernelsHand.cc ./qcd/spin/Dirac.cc ./qcd/utils/SpaceTimeGrid.cc ./stencil/Lebesgue.cc ./stencil/Stencil_common.cc

6
lib/algorithms/CoarsenedMatrix.h
View File

@@ -85,10 +85,6 @@ namespace Grid {
void Orthogonalise(void){
CoarseScalar InnerProd(CoarseGrid);
blockOrthogonalise(InnerProd,subspace);
#if 1
// CheckOrthogonal();
#endif
}
void CheckOrthogonal(void){
CoarseVector iProj(CoarseGrid);
@@ -125,7 +121,7 @@ namespace Grid {
RealD scale;
ConjugateGradient<FineField> CG(1.0e-4,10000);
ConjugateGradient<FineField> CG(1.0e-3,10000);
FineField noise(FineGrid);
FineField Mn(FineGrid);

2
lib/algorithms/iterative/ConjugateGradient.h
View File

@@ -15,7 +15,7 @@ public:
Integer MaxIterations;
int verbose;
ConjugateGradient(RealD tol,Integer maxit) : Tolerance(tol), MaxIterations(maxit) {
verbose=0;
verbose=1;
};

6
lib/algorithms/iterative/PrecGeneralisedConjugateResidual.h
View File

@@ -48,11 +48,11 @@ namespace Grid {
if(cp<rsq) {
Linop.HermOp(psi,r);
axpy(r,-1.0,src,r);
RealD true_resid = norm2(r);
RealD tr = norm2(r);
std::cout<<"PrecGeneralisedConjugateResidual: Converged on iteration " <<steps
<< " computed residual "<<sqrt(cp/ssq)
<< " true residual "<<true_resid
<< " target " <<Tolerance <<std::endl;
<< " true residual " <<sqrt(tr/ssq)
<< " target " <<Tolerance <<std::endl;
return;
}

39
lib/qcd/action/fermion/WilsonFermion5D.cc
View File

@@ -87,8 +87,8 @@ void WilsonFermion5D::DhopDir(const LatticeFermion &in, LatticeFermion &out,int
{
int dir = dir5-1; // Maps to the ordering above in "directions" that is passed to stencil
// we drop off the innermost fifth dimension
assert( (disp==1)||(disp==-1) );
assert( (dir>=0)&&(dir<4) ); //must do x,y,z or t;
// assert( (disp==1)||(disp==-1) );
// assert( (dir>=0)&&(dir<4) ); //must do x,y,z or t;
WilsonCompressor compressor(DaggerNo);
Stencil.HaloExchange<vSpinColourVector,vHalfSpinColourVector,WilsonCompressor>(in,comm_buf,compressor);
@@ -100,7 +100,7 @@ void WilsonFermion5D::DhopDir(const LatticeFermion &in, LatticeFermion &out,int
assert(dirdisp<=7);
assert(dirdisp>=0);
PARALLEL_FOR_LOOP
//PARALLEL_FOR_LOOP
for(int ss=0;ss<Umu._grid->oSites();ss++){
for(int s=0;s<Ls;s++){
int sU=ss;
@@ -114,7 +114,7 @@ void WilsonFermion5D::DhopInternal(CartesianStencil & st, LebesgueOrder &lo,
LatticeDoubledGaugeField & U,
const LatticeFermion &in, LatticeFermion &out,int dag)
{
assert((dag==DaggerNo) ||(dag==DaggerYes));
// assert((dag==DaggerNo) ||(dag==DaggerYes));
WilsonCompressor compressor(dag);
@@ -127,29 +127,32 @@ void WilsonFermion5D::DhopInternal(CartesianStencil & st, LebesgueOrder &lo,
// - 8 linear access unit stride streams per thread for Fermion for hw prefetchable.
if ( dag == DaggerYes ) {
if( HandOptDslash ) {
PARALLEL_FOR_LOOP
#pragma parallel for
for(int ss=0;ss<U._grid->oSites();ss++){
for(int s=0;s<Ls;s++){
//int sU=lo.Reorder(ss);
int sU=ss;
int sF = s+Ls*sU;
DiracOptHand::DhopSiteDag(st,U,comm_buf,sF,sU,in,out);
{
for(int s=0;s<Ls;s++){
int sU=ss;
int sF = s+Ls*sU;
DiracOptHand::DhopSiteDag(st,U,comm_buf,sF,sU,in,out);
}
}
}
} else {
PARALLEL_FOR_LOOP
#pragma parallel for
for(int ss=0;ss<U._grid->oSites();ss++){
for(int s=0;s<Ls;s++){
// int sU=lo.Reorder(ss);
int sU=ss;
int sF = s+Ls*sU;
DiracOpt::DhopSiteDag(st,U,comm_buf,sF,sU,in,out);
{
int sd;
for(sd=0;sd<Ls;sd++){
int sU=ss;
int sF = sd+Ls*sU;
DiracOpt::DhopSiteDag(st,U,comm_buf,sF,sU,in,out);
}
}
}
}
} else {
if( HandOptDslash ) {
PARALLEL_FOR_LOOP
#pragma parallel for
for(int ss=0;ss<U._grid->oSites();ss++){
for(int s=0;s<Ls;s++){
// int sU=lo.Reorder(ss);
@@ -160,7 +163,7 @@ PARALLEL_FOR_LOOP
}
} else {
PARALLEL_FOR_LOOP
#pragma parallel for
for(int ss=0;ss<U._grid->oSites();ss++){
for(int s=0;s<Ls;s++){
// int sU=lo.Reorder(ss);

4
lib/qcd/utils/SUn.h
View File

@@ -166,7 +166,7 @@ public:
su2SubGroupIndex(i0,i1,su2_index);
PARALLEL_FOR_LOOP
for(int ss=0;ss!=grid->oSites();ss++){
for(int ss=0;ss<grid->oSites();ss++){
subgroup._odata[ss]()()(0,0) = source._odata[ss]()()(i0,i0);
subgroup._odata[ss]()()(0,1) = source._odata[ss]()()(i0,i1);
subgroup._odata[ss]()()(1,0) = source._odata[ss]()()(i1,i0);
@@ -201,7 +201,7 @@ PARALLEL_FOR_LOOP
dest = 1.0; // start out with identity
PARALLEL_FOR_LOOP
for(int ss=0;ss!=grid->oSites();ss++){
for(int ss=0;ss<grid->oSites();ss++){
dest._odata[ss]()()(i0,i0) = subgroup._odata[ss]()()(0,0);
dest._odata[ss]()()(i0,i1) = subgroup._odata[ss]()()(0,1);
dest._odata[ss]()()(i1,i0) = subgroup._odata[ss]()()(1,0);