1
0
mirror of https://github.com/paboyle/Grid.git synced 2024-09-20 09:15:38 +01:00

PartialFraction Hw with Zolo and Tanh approx converged under CG and passed EO breakdown

and hermiticity tests.
This commit is contained in:
Peter Boyle 2015-06-04 13:28:37 +01:00
parent 5b1ba66604
commit b9e9777912
21 changed files with 501 additions and 56 deletions

View File

@ -1 +1 @@
/usr/share/automake-1.14/INSTALL
/opt/local/share/automake-1.15/INSTALL

30
configure vendored
View File

@ -2574,7 +2574,7 @@ test -n "$target_alias" &&
NONENONEs,x,x, &&
program_prefix=${target_alias}-
am__api_version='1.14'
am__api_version='1.15'
# Find a good install program. We prefer a C program (faster),
# so one script is as good as another. But avoid the broken or
@ -2746,8 +2746,8 @@ test "$program_suffix" != NONE &&
ac_script='s/[\\$]/&&/g;s/;s,x,x,$//'
program_transform_name=`$as_echo "$program_transform_name" | sed "$ac_script"`
# expand $ac_aux_dir to an absolute path
am_aux_dir=`cd $ac_aux_dir && pwd`
# Expand $ac_aux_dir to an absolute path.
am_aux_dir=`cd "$ac_aux_dir" && pwd`
if test x"${MISSING+set}" != xset; then
case $am_aux_dir in
@ -2766,7 +2766,7 @@ else
$as_echo "$as_me: WARNING: 'missing' script is too old or missing" >&2;}
fi
if test x"${install_sh}" != xset; then
if test x"${install_sh+set}" != xset; then
case $am_aux_dir in
*\ * | *\ *)
install_sh="\${SHELL} '$am_aux_dir/install-sh'" ;;
@ -3094,8 +3094,8 @@ MAKEINFO=${MAKEINFO-"${am_missing_run}makeinfo"}
# <http://lists.gnu.org/archive/html/automake/2012-07/msg00014.html>
mkdir_p='$(MKDIR_P)'
# We need awk for the "check" target. The system "awk" is bad on
# some platforms.
# We need awk for the "check" target (and possibly the TAP driver). The
# system "awk" is bad on some platforms.
# Always define AMTAR for backward compatibility. Yes, it's still used
# in the wild :-( We should find a proper way to deprecate it ...
AMTAR='$${TAR-tar}'
@ -3154,6 +3154,7 @@ END
fi
ac_config_headers="$ac_config_headers lib/GridConfig.h"
# Check whether --enable-silent-rules was given.
@ -6656,9 +6657,6 @@ fi
# Check whether --enable-simd was given.
if test "${enable_simd+set}" = set; then :
enableval=$enable_simd; \
@ -6673,21 +6671,21 @@ supported=no
case ${ac_SIMD} in
SSE4)
echo Configuring for SSE4
if test x"$ax_cv_support_ssse3_ext" = x"yes"; then
$as_echo "#define SSE4 1" >>confdefs.h
supported=yes
if test x"$ax_cv_support_ssse3_ext" = x"yes"; then supported=yes
else
{ $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: Your processor does not support SSE4 instructions" >&5
{ $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: Your processor does not support SSE4 instructions" >&5
$as_echo "$as_me: WARNING: Your processor does not support SSE4 instructions" >&2;}
fi
;;
AVX)
echo Configuring for AVX
if test x"$ax_cv_support_avx_ext" = x"yes"; then
$as_echo "#define AVX1 1" >>confdefs.h
supported=yes
if test x"$ax_cv_support_avx_ext" = x"yes"; then supported=yes
else
{ $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: Your processor does not support AVX instructions" >&5
$as_echo "$as_me: WARNING: Your processor does not support AVX instructions" >&2;}
@ -6695,10 +6693,10 @@ $as_echo "$as_me: WARNING: Your processor does not support AVX instructions" >&2
;;
AVX2)
echo Configuring for AVX2
if test x"$ax_cv_support_avx2_ext" = x"yes"; then
$as_echo "#define AVX2 1" >>confdefs.h
supported=yes
if test x"$ax_cv_support_avx2_ext" = x"yes"; then supported=yes
else
{ $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: Your processor does not support AVX2 instructions" >&5
$as_echo "$as_me: WARNING: Your processor does not support AVX2 instructions" >&2;}

View File

@ -66,9 +66,6 @@ Please install or provide the correct path to your installation
Info at: http://www.mpfr.org/)])
AC_ARG_ENABLE([simd],[AC_HELP_STRING([--enable-simd=SSE4|AVX|AVX2|AVX512|MIC],\
[Select instructions to be SSE4.0, AVX 1.0, AVX 2.0+FMA, AVX 512, MIC])],\
[ac_SIMD=${enable_simd}],[ac_SIMD=AVX2])
@ -78,17 +75,17 @@ supported=no
case ${ac_SIMD} in
SSE4)
echo Configuring for SSE4
if test x"$ax_cv_support_ssse3_ext" = x"yes"; then dnl minimal support for SSE4
AC_DEFINE([SSE4],[1],[SSE4] )
supported=yes
if test x"$ax_cv_support_ssse3_ext" = x"yes"; then dnl minimal support for SSE4
supported=yes
else
AC_MSG_WARN([Your processor does not support SSE4 instructions])
AC_MSG_WARN([Your processor does not support SSE4 instructions])
fi
;;
AVX)
echo Configuring for AVX
if test x"$ax_cv_support_avx_ext" = x"yes"; then dnl minimal support for AVX
AC_DEFINE([AVX1],[1],[AVX] )
if test x"$ax_cv_support_avx_ext" = x"yes"; then dnl minimal support for AVX
supported=yes
else
AC_MSG_WARN([Your processor does not support AVX instructions])
@ -96,8 +93,8 @@ case ${ac_SIMD} in
;;
AVX2)
echo Configuring for AVX2
if test x"$ax_cv_support_avx2_ext" = x"yes"; then dnl minimal support for AVX2
AC_DEFINE([AVX2],[1],[AVX2] )
if test x"$ax_cv_support_avx2_ext" = x"yes"; then dnl minimal support for AVX2
supported=yes
else
AC_MSG_WARN([Your processor does not support AVX2 instructions])

View File

@ -1,4 +1,4 @@
HFILES=./Cshift.h ./simd/Grid_avx.h ./simd/Grid_vector_types.h ./simd/Grid_sse4.h ./simd/Grid_avx512.h ./simd/Old/Grid_vRealD.h ./simd/Old/Grid_vComplexD.h ./simd/Old/Grid_vInteger.h ./simd/Old/Grid_vComplexF.h ./simd/Old/Grid_vRealF.h ./simd/Grid_qpx.h ./Tensors.h ./Algorithms.h ./communicator/Communicator_base.h ./lattice/Lattice_rng.h ./lattice/Lattice_reduction.h ./lattice/Lattice_transfer.h ./lattice/Lattice_peekpoke.h ./lattice/Lattice_coordinate.h ./lattice/Lattice_comparison.h ./lattice/Lattice_overload.h ./lattice/Lattice_reality.h ./lattice/Lattice_local.h ./lattice/Lattice_conformable.h ./lattice/Lattice_where.h ./lattice/Lattice_arith.h ./lattice/Lattice_base.h ./lattice/Lattice_ET.h ./lattice/Lattice_transpose.h ./lattice/Lattice_trace.h ./Stencil.h ./tensors/Tensor_arith_sub.h ./tensors/Tensor_poke.h ./tensors/Tensor_arith_mul.h ./tensors/Tensor_class.h ./tensors/Tensor_transpose.h ./tensors/Tensor_arith_mac.h ./tensors/Tensor_arith_scalar.h ./tensors/Tensor_reality.h ./tensors/Tensor_trace.h ./tensors/Tensor_arith_add.h ./tensors/Tensor_outer.h ./tensors/Tensor_inner.h ./tensors/Tensor_traits.h ./tensors/Tensor_Ta.h ./tensors/Tensor_peek.h ./tensors/Tensor_arith.h ./tensors/Tensor_extract_merge.h ./Communicator.h ./Cartesian.h ./parallelIO/NerscIO.h ./qcd/QCD.h ./qcd/SpaceTimeGrid.h ./qcd/LinalgUtils.h ./qcd/TwoSpinor.h ./qcd/action/Actions.h ./qcd/action/fermion/CayleyFermion5D.h ./qcd/action/fermion/ScaledShamirFermion.h ./qcd/action/fermion/MobiusFermion.h ./qcd/action/fermion/OverlapWilsonContfracTanhFermion.h ./qcd/action/fermion/PartialFractionFermion5D.h ./qcd/action/fermion/ShamirZolotarevFermion.h ./qcd/action/fermion/FermionOperator.h ./qcd/action/fermion/WilsonFermion5D.h ./qcd/action/fermion/WilsonCompressor.h ./qcd/action/fermion/WilsonKernels.h ./qcd/action/fermion/DomainWallFermion.h ./qcd/action/fermion/OverlapWilsonContfracZolotarevFermion.h ./qcd/action/fermion/MobiusZolotarevFermion.h ./qcd/action/fermion/OverlapWilsonCayleyTanhFermion.h ./qcd/action/fermion/WilsonFermion.h ./qcd/action/fermion/ContinuedFractionFermion5D.h ./qcd/action/fermion/OverlapWilsonCayleyZolotarevFermion.h ./qcd/Dirac.h ./cshift/Cshift_common.h ./cshift/Cshift_none.h ./cshift/Cshift_mpi.h ./Simd.h ./GridConfig.h ./cartesian/Cartesian_base.h ./cartesian/Cartesian_red_black.h ./cartesian/Cartesian_full.h ./AlignedAllocator.h ./Lattice.h ./Threads.h ./Comparison.h ./Grid.h ./algorithms/iterative/SchurRedBlack.h ./algorithms/iterative/NormalEquations.h ./algorithms/iterative/ConjugateGradient.h ./algorithms/approx/Chebyshev.h ./algorithms/approx/Zolotarev.h ./algorithms/approx/bigfloat.h ./algorithms/approx/bigfloat_double.h ./algorithms/approx/Remez.h ./algorithms/LinearOperator.h ./algorithms/SparseMatrix.h ./stencil/Lebesgue.h
HFILES=./algorithms/approx/bigfloat.h ./algorithms/approx/bigfloat_double.h ./algorithms/approx/Chebyshev.h ./algorithms/approx/Remez.h ./algorithms/approx/Zolotarev.h ./algorithms/iterative/ConjugateGradient.h ./algorithms/iterative/NormalEquations.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 ./Comparison.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_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_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/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/Dirac.h ./qcd/LinalgUtils.h ./qcd/QCD.h ./qcd/SpaceTimeGrid.h ./qcd/TwoSpinor.h ./simd/Grid_avx.h ./simd/Grid_avx512.h ./simd/Grid_qpx.h ./simd/Grid_sse4.h ./simd/Grid_vector_types.h ./simd/Old/Grid_vComplexD.h ./simd/Old/Grid_vComplexF.h ./simd/Old/Grid_vInteger.h ./simd/Old/Grid_vRealD.h ./simd/Old/Grid_vRealF.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_extract_merge.h ./tensors/Tensor_inner.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.h ./Threads.h
CCFILES=./qcd/SpaceTimeGrid.cc ./qcd/action/fermion/WilsonKernels.cc ./qcd/action/fermion/PartialFractionFermion5D.cc ./qcd/action/fermion/CayleyFermion5D.cc ./qcd/action/fermion/WilsonKernelsHand.cc ./qcd/action/fermion/WilsonFermion.cc ./qcd/action/fermion/ContinuedFractionFermion5D.cc ./qcd/action/fermion/WilsonFermion5D.cc ./qcd/Dirac.cc ./GridInit.cc ./algorithms/approx/Remez.cc ./algorithms/approx/Zolotarev.cc ./stencil/Lebesgue.cc ./stencil/Stencil_common.cc
CCFILES=./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/Dirac.cc ./qcd/SpaceTimeGrid.cc ./stencil/Lebesgue.cc ./stencil/Stencil_common.cc

View File

@ -293,7 +293,7 @@ static void sncndnFK(INTERNAL_PRECISION u, INTERNAL_PRECISION k,
* Set type = 0 for the Zolotarev approximation, which is zero at x = 0, and
* type = 1 for the approximation which is infinite at x = 0. */
zolotarev_data* grid_zolotarev(PRECISION epsilon, int n, int type) {
zolotarev_data* zolotarev(PRECISION epsilon, int n, int type) {
INTERNAL_PRECISION A, c, cp, kp, ksq, sn, cn, dn, Kp, Kj, z, z0, t, M, F,
l, invlambda, xi, xisq, *tv, s, opl;
int m, czero, ts;
@ -414,7 +414,19 @@ zolotarev_data* grid_zolotarev(PRECISION epsilon, int n, int type) {
return zd;
}
zolotarev_data* grid_higham(PRECISION epsilon, int n) {
void zolotarev_free(zolotarev_data *zdata)
{
free(zdata -> a);
free(zdata -> ap);
free(zdata -> alpha);
free(zdata -> beta);
free(zdata -> gamma);
free(zdata);
}
zolotarev_data* higham(PRECISION epsilon, int n) {
INTERNAL_PRECISION A, M, c, cp, z, z0, t, epssq;
int m, czero;
zolotarev_data *zd;

View File

@ -77,8 +77,9 @@ typedef struct {
* zolotarev_data structure. The arguments must satisfy the constraints that
* epsilon > 0, n > 0, and type = 0 or 1. */
ZOLOTAREV_DATA* grid_higham(PRECISION epsilon, int n) ;
ZOLOTAREV_DATA* grid_zolotarev(PRECISION epsilon, int n, int type);
ZOLOTAREV_DATA* higham(PRECISION epsilon, int n) ;
ZOLOTAREV_DATA* zolotarev(PRECISION epsilon, int n, int type);
void zolotarev_free(zolotarev_data *zdata);
#endif
#ifdef __cplusplus

View File

@ -62,6 +62,8 @@
// Partial fraction
//////////////////////
#include <qcd/action/fermion/PartialFractionFermion5D.h>
#include <qcd/action/fermion/OverlapWilsonPartialFractionTanhFermion.h>
#include <qcd/action/fermion/OverlapWilsonPartialFractionZolotarevFermion.h>
// Chroma interface defining FermionAction

View File

@ -9,6 +9,16 @@ namespace Grid {
}
void ContinuedFractionFermion5D::SetCoefficientsZolotarev(RealD zolo_hi,Approx::zolotarev_data *zdata)
{
// How to check Ls matches??
// std::cout << Ls << " Ls"<<std::endl;
// std::cout << zdata->n << " - n"<<std::endl;
// std::cout << zdata->da << " -da "<<std::endl;
// std::cout << zdata->db << " -db"<<std::endl;
// std::cout << zdata->dn << " -dn"<<std::endl;
// std::cout << zdata->dd << " -dd"<<std::endl;
assert(zdata->db==Ls);// Beta has Ls coeffs
R=(1+this->mass)/(1-this->mass);
Beta.resize(Ls);
@ -29,7 +39,7 @@ namespace Grid {
ZoloHiInv =1.0/zolo_hi;
double dw_diag = (4.0-M5)*ZoloHiInv;
dw_diag = (4.0-M5)*ZoloHiInv;
See.resize(Ls);
Aee.resize(Ls);
@ -105,8 +115,6 @@ namespace Grid {
}
void ContinuedFractionFermion5D::Mooee (const LatticeFermion &psi, LatticeFermion &chi)
{
double dw_diag = (4.0-M5)*ZoloHiInv;
int sign=1;
for(int s=0;s<Ls;s++){
if ( s==0 ) {

View File

@ -37,9 +37,8 @@ namespace Grid {
void SetCoefficientsTanh(Approx::zolotarev_data *zdata,RealD scale);
void SetCoefficientsZolotarev(RealD zolo_hi,Approx::zolotarev_data *zdata);;
Approx::zolotarev_data *zdata;
// Cont frac
RealD dw_diag;
RealD mass;
RealD R;
RealD ZoloHiInv;

View File

@ -29,13 +29,14 @@ namespace Grid {
{
RealD eps = 1.0;
Approx::zolotarev_data *zdata = Approx::grid_higham(eps,this->Ls);// eps is ignored for higham
Approx::zolotarev_data *zdata = Approx::higham(eps,this->Ls);// eps is ignored for higham
assert(zdata->n==this->Ls);
std::cout << "DomainWallFermion with Ls="<<Ls<<std::endl;
// Call base setter
this->CayleyFermion5D::SetCoefficientsTanh(zdata,1.0,0.0);
Approx::zolotarev_free(zdata);
}
};

View File

@ -31,11 +31,13 @@ namespace Grid {
RealD eps = 1.0;
std::cout << "MobiusFermion (b="<<b<<",c="<<c<<") with Ls= "<<Ls<<" Tanh approx"<<std::endl;
Approx::zolotarev_data *zdata = Approx::grid_higham(eps,this->Ls);// eps is ignored for higham
Approx::zolotarev_data *zdata = Approx::higham(eps,this->Ls);// eps is ignored for higham
assert(zdata->n==this->Ls);
// Call base setter
this->CayleyFermion5D::SetCoefficientsTanh(zdata,b,c);
Approx::zolotarev_free(zdata);
}

View File

@ -31,7 +31,7 @@ namespace Grid {
{
RealD eps = lo/hi;
Approx::zolotarev_data *zdata = Approx::grid_zolotarev(eps,this->Ls,0);// eps is ignored for higham
Approx::zolotarev_data *zdata = Approx::zolotarev(eps,this->Ls,0);
assert(zdata->n==this->Ls);
std::cout << "MobiusZolotarevFermion (b="<<b<<",c="<<c<<") with Ls= "<<Ls<<" Zolotarev range ["<<lo<<","<<hi<<"]"<<std::endl;
@ -39,6 +39,7 @@ namespace Grid {
// Call base setter
this->CayleyFermion5D::SetCoefficientsZolotarev(hi,zdata,b,c);
Approx::zolotarev_free(zdata);
}
};

View File

@ -30,8 +30,9 @@ namespace Grid {
{
assert((Ls&0x1)==1); // Odd Ls required
int nrational=Ls-1;// Even rational order
zdata = Approx::grid_higham(1.0,nrational);// eps is ignored for higham
Approx::zolotarev_data *zdata = Approx::higham(1.0,nrational);// eps is ignored for higham
SetCoefficientsTanh(zdata,scale);
Approx::zolotarev_free(zdata);
}
};
}

View File

@ -30,12 +30,12 @@ namespace Grid {
{
assert((Ls&0x1)==1); // Odd Ls required
int nrational=Ls-1;// Even rational order
int nrational=Ls;// Odd rational order
RealD eps = lo/hi;
Approx::zolotarev_data *zdata = Approx::grid_zolotarev(eps,nrational,0);
Approx::zolotarev_data *zdata = Approx::zolotarev(eps,nrational,0);
SetCoefficientsZolotarev(hi,zdata);
Approx::zolotarev_free(zdata);
}
};

View File

@ -0,0 +1,40 @@
#ifndef OVERLAP_WILSON_PARTFRAC_TANH_FERMION_H
#define OVERLAP_WILSON_PARTFRAC_TANH_FERMION_H
#include <Grid.h>
namespace Grid {
namespace QCD {
class OverlapWilsonPartialFractionTanhFermion : public PartialFractionFermion5D
{
public:
virtual void Instantiatable(void){};
// Constructors
OverlapWilsonPartialFractionTanhFermion(LatticeGaugeField &_Umu,
GridCartesian &FiveDimGrid,
GridRedBlackCartesian &FiveDimRedBlackGrid,
GridCartesian &FourDimGrid,
GridRedBlackCartesian &FourDimRedBlackGrid,
RealD _mass,RealD _M5,
RealD scale) :
// b+c=scale, b-c = 0 <=> b =c = scale/2
PartialFractionFermion5D(_Umu,
FiveDimGrid,
FiveDimRedBlackGrid,
FourDimGrid,
FourDimRedBlackGrid,_mass,_M5)
{
assert((Ls&0x1)==1); // Odd Ls required
int nrational=Ls-1;// Even rational order
Approx::zolotarev_data *zdata = Approx::higham(1.0,nrational);// eps is ignored for higham
SetCoefficientsTanh(zdata,scale);
Approx::zolotarev_free(zdata);
}
};
}
}
#endif

View File

@ -0,0 +1,44 @@
#ifndef OVERLAP_WILSON_PARTFRAC_ZOLOTAREV_FERMION_H
#define OVERLAP_WILSON_PARTFRAC_ZOLOTAREV_FERMION_H
#include <Grid.h>
namespace Grid {
namespace QCD {
class OverlapWilsonPartialFractionZolotarevFermion : public PartialFractionFermion5D
{
public:
virtual void Instantiatable(void){};
// Constructors
OverlapWilsonPartialFractionZolotarevFermion(LatticeGaugeField &_Umu,
GridCartesian &FiveDimGrid,
GridRedBlackCartesian &FiveDimRedBlackGrid,
GridCartesian &FourDimGrid,
GridRedBlackCartesian &FourDimRedBlackGrid,
RealD _mass,RealD _M5,
RealD lo,RealD hi):
// b+c=scale, b-c = 0 <=> b =c = scale/2
PartialFractionFermion5D(_Umu,
FiveDimGrid,
FiveDimRedBlackGrid,
FourDimGrid,
FourDimRedBlackGrid,_mass,_M5)
{
assert((Ls&0x1)==1); // Odd Ls required
int nrational=Ls;// Odd rational order
RealD eps = lo/hi;
Approx::zolotarev_data *zdata = Approx::zolotarev(eps,nrational,0);
SetCoefficientsZolotarev(hi,zdata);
Approx::zolotarev_free(zdata);
}
};
}
}
#endif

View File

@ -1 +1,310 @@
#include <Grid.h>
namespace Grid {
namespace QCD {
void PartialFractionFermion5D::Meooe_internal(const LatticeFermion &psi, LatticeFermion &chi,int dag)
{
// this does both dag and undag but is trivial; make a common helper routing
int sign = dag ? (-1) : 1;
if ( psi.checkerboard == Odd ) {
DhopEO(psi,chi,DaggerNo);
} else {
DhopOE(psi,chi,DaggerNo);
}
int nblock=(Ls-1)/2;
for(int b=0;b<nblock;b++){
int s = 2*b;
ag5xpby_ssp(chi,-scale,chi,0.0,chi,s,s);
ag5xpby_ssp(chi, scale,chi,0.0,chi,s+1,s+1);
}
ag5xpby_ssp(chi,p[nblock]*scale/amax,chi,0.0,chi,Ls-1,Ls-1);
}
void PartialFractionFermion5D::Mooee_internal(const LatticeFermion &psi, LatticeFermion &chi,int dag)
{
// again dag and undag are trivially related
int sign = dag ? (-1) : 1;
int nblock=(Ls-1)/2;
for(int b=0;b<nblock;b++){
int s = 2*b;
RealD pp = p[nblock-1-b];
RealD qq = q[nblock-1-b];
// Do each 2x2 block aligned at s and multiplies Dw site diagonal by G5 so Hw
ag5xpby_ssp(chi,-dw_diag*scale,psi,amax*sqrt(qq)*scale,psi, s ,s+1);
ag5xpby_ssp(chi, dw_diag*scale,psi,amax*sqrt(qq)*scale,psi, s+1,s);
axpby_ssp (chi, 1.0, chi,sqrt(amax*pp)*scale*sign,psi,s+1,Ls-1);
}
{
RealD R=(1+mass)/(1-mass);
//R g5 psi[Ls-1] + p[0] H
ag5xpbg5y_ssp(chi,R*scale,psi,p[nblock]*scale*dw_diag/amax,psi,Ls-1,Ls-1);
for(int b=0;b<nblock;b++){
int s = 2*b+1;
RealD pp = p[nblock-1-b];
axpby_ssp(chi,1.0,chi,-sqrt(amax*pp)*scale*sign,psi,Ls-1,s);
}
}
}
void PartialFractionFermion5D::MooeeInv_internal(const LatticeFermion &psi, LatticeFermion &chi,int dag)
{
int sign = dag ? (-1) : 1;
LatticeFermion tmp(psi._grid);
///////////////////////////////////////////////////////////////////////////////////////
//Linv
///////////////////////////////////////////////////////////////////////////////////////
int nblock=(Ls-1)/2;
axpy(chi,0.0,psi,psi); // Identity piece
for(int b=0;b<nblock;b++){
int s = 2*b;
RealD pp = p[nblock-1-b];
RealD qq = q[nblock-1-b];
RealD coeff1=sign*sqrt(amax*amax*amax*pp*qq) / ( dw_diag*dw_diag + amax*amax* qq);
RealD coeff2=sign*sqrt(amax*pp)*dw_diag / ( dw_diag*dw_diag + amax*amax* qq); // Implicit g5 here
axpby_ssp (chi,1.0,chi,coeff1,psi,Ls-1,s);
axpbg5y_ssp(chi,1.0,chi,coeff2,psi,Ls-1,s+1);
}
///////////////////////////////////////////////////////////////////////////////////////
//Dinv (note D isn't really diagonal -- just diagonal enough that we can still invert)
// Compute Seeinv (coeff of gamma5)
///////////////////////////////////////////////////////////////////////////////////////
RealD R=(1+mass)/(1-mass);
RealD Seeinv = R + p[nblock]*dw_diag/amax;
for(int b=0;b<nblock;b++){
Seeinv += p[nblock-1-b]*dw_diag/amax / ( dw_diag*dw_diag/amax/amax + q[nblock-1-b]);
}
Seeinv = 1.0/Seeinv;
for(int b=0;b<nblock;b++){
int s = 2*b;
RealD pp = p[nblock-1-b];
RealD qq = q[nblock-1-b];
RealD coeff1=dw_diag / ( dw_diag*dw_diag + amax*amax* qq); // Implicit g5 here
RealD coeff2=amax*sqrt(qq) / ( dw_diag*dw_diag + amax*amax* qq);
ag5xpby_ssp (tmp,-coeff1,chi,coeff2,chi,s,s+1);
ag5xpby_ssp (tmp, coeff1,chi,coeff2,chi,s+1,s);
}
ag5xpby_ssp (tmp, Seeinv,chi,0.0,chi,Ls-1,Ls-1);
///////////////////////////////////////////////////////////////////////////////////////
// Uinv
///////////////////////////////////////////////////////////////////////////////////////
for(int b=0;b<nblock;b++){
int s = 2*b;
RealD pp = p[nblock-1-b];
RealD qq = q[nblock-1-b];
RealD coeff1=-sign*sqrt(amax*amax*amax*pp*qq) / ( dw_diag*dw_diag + amax*amax* qq);
RealD coeff2=-sign*sqrt(amax*pp)*dw_diag / ( dw_diag*dw_diag + amax*amax* qq); // Implicit g5 here
axpby_ssp (chi,1.0/scale,tmp,coeff1/scale,tmp,s,Ls-1);
axpbg5y_ssp(chi,1.0/scale,tmp,coeff2/scale,tmp,s+1,Ls-1);
}
axpby_ssp (chi, 1.0/scale,tmp,0.0,tmp,Ls-1,Ls-1);
}
void PartialFractionFermion5D::M_internal(const LatticeFermion &psi, LatticeFermion &chi,int dag)
{
LatticeFermion D(psi._grid);
int sign = dag ? (-1) : 1;
// For partial frac Hw case (b5=c5=1) chroma quirkily computes
//
// Conventions for partfrac appear to be a mess.
// Tony's Nara lectures have
//
// BlockDiag( H/p_i 1 | 1 )
// ( 1 p_i H / q_i^2 | 0 )
// ---------------------------------
// ( -1 0 | R +p0 H )
//
//Chroma ( -2H 2sqrt(q_i) | 0 )
// (2 sqrt(q_i) 2H | 2 sqrt(p_i) )
// ---------------------------------
// ( 0 -2 sqrt(p_i) | 2 R gamma_5 + p0 2H
//
// Edwards/Joo/Kennedy/Wenger
//
// Here, the "beta's" selected by chroma to scale the unphysical bulk constraint fields
// incorporate the approx scale factor. This is obtained by propagating the
// scale on "H" out to the off diagonal elements as follows:
//
// BlockDiag( H/p_i 1 | 1 )
// ( 1 p_i H / q_i^2 | 0 )
// ---------------------------------
// ( -1 0 | R + p_0 H )
//
// becomes:
// BlockDiag( H/ sp_i 1 | 1 )
// ( 1 sp_i H / s^2q_i^2 | 0 )
// ---------------------------------
// ( -1 0 | R + p_0/s H )
//
//
// This is implemented in Chroma by
// p0' = p0/approxMax
// p_i' = p_i*approxMax
// q_i' = q_i*approxMax*approxMax
//
// After the equivalence transform is applied the matrix becomes
//
//Chroma ( -2H sqrt(q'_i) | 0 )
// (sqrt(q'_i) 2H | sqrt(p'_i) )
// ---------------------------------
// ( 0 -sqrt(p'_i) | 2 R gamma_5 + p'0 2H
//
// = ( -2H sqrt(q_i)amax | 0 )
// (sqrt(q_i)amax 2H | sqrt(p_i*amax) )
// ---------------------------------
// ( 0 -sqrt(p_i)*amax | 2 R gamma_5 + p0/amax 2H
//
DW(psi,D,DaggerNo);
int nblock=(Ls-1)/2;
for(int b=0;b<nblock;b++){
int s = 2*b;
double pp = p[nblock-1-b];
double qq = q[nblock-1-b];
// Do each 2x2 block aligned at s and
ag5xpby_ssp(chi,-1.0*scale,D,amax*sqrt(qq)*scale,psi, s ,s+1); // Multiplies Dw by G5 so Hw
ag5xpby_ssp(chi, 1.0*scale,D,amax*sqrt(qq)*scale,psi, s+1,s);
// Pick up last column
axpby_ssp (chi, 1.0, chi,sqrt(amax*pp)*scale*sign,psi,s+1,Ls-1);
}
{
double R=(1+this->mass)/(1-this->mass);
//R g5 psi[Ls] + p[0] H
ag5xpbg5y_ssp(chi,R*scale,psi,p[nblock]*scale/amax,D,Ls-1,Ls-1);
for(int b=0;b<nblock;b++){
int s = 2*b+1;
double pp = p[nblock-1-b];
axpby_ssp(chi,1.0,chi,-sqrt(amax*pp)*scale*sign,psi,Ls-1,s);
}
}
}
RealD PartialFractionFermion5D::M (const LatticeFermion &in, LatticeFermion &out)
{
M_internal(in,out,DaggerNo);
return norm2(out);
}
RealD PartialFractionFermion5D::Mdag (const LatticeFermion &in, LatticeFermion &out)
{
M_internal(in,out,DaggerYes);
return norm2(out);
}
void PartialFractionFermion5D::Meooe (const LatticeFermion &in, LatticeFermion &out)
{
Meooe_internal(in,out,DaggerNo);
}
void PartialFractionFermion5D::MeooeDag (const LatticeFermion &in, LatticeFermion &out)
{
Meooe_internal(in,out,DaggerYes);
}
void PartialFractionFermion5D::Mooee (const LatticeFermion &in, LatticeFermion &out)
{
Mooee_internal(in,out,DaggerNo);
}
void PartialFractionFermion5D::MooeeDag (const LatticeFermion &in, LatticeFermion &out)
{
Mooee_internal(in,out,DaggerYes);
}
void PartialFractionFermion5D::MooeeInv (const LatticeFermion &in, LatticeFermion &out)
{
MooeeInv_internal(in,out,DaggerNo);
}
void PartialFractionFermion5D::MooeeInvDag (const LatticeFermion &in, LatticeFermion &out)
{
MooeeInv_internal(in,out,DaggerYes);
}
void PartialFractionFermion5D::SetCoefficientsTanh(Approx::zolotarev_data *zdata,RealD scale){
SetCoefficientsZolotarev(1.0/scale,zdata);
}
void PartialFractionFermion5D::SetCoefficientsZolotarev(RealD zolo_hi,Approx::zolotarev_data *zdata){
// check on degree matching
// std::cout << Ls << " Ls"<<std::endl;
// std::cout << zdata->n << " - n"<<std::endl;
// std::cout << zdata->da << " -da "<<std::endl;
// std::cout << zdata->db << " -db"<<std::endl;
// std::cout << zdata->dn << " -dn"<<std::endl;
// std::cout << zdata->dd << " -dd"<<std::endl;
assert(Ls == (2*zdata->da -1) );
// Part frac
// RealD R;
R=(1+mass)/(1-mass);
dw_diag = (4.0-M5);
// std::vector<RealD> p;
// std::vector<RealD> q;
p.resize(zdata->da);
q.resize(zdata->dd);
for(int n=0;n<zdata->da;n++){
p[n] = zdata -> alpha[n];
}
for(int n=0;n<zdata->dd;n++){
q[n] = -zdata -> ap[n];
}
scale= part_frac_chroma_convention ? 2.0 : 1.0; // Chroma conventions annoy me
amax=zolo_hi;
}
// Constructors
PartialFractionFermion5D::PartialFractionFermion5D(LatticeGaugeField &_Umu,
GridCartesian &FiveDimGrid,
GridRedBlackCartesian &FiveDimRedBlackGrid,
GridCartesian &FourDimGrid,
GridRedBlackCartesian &FourDimRedBlackGrid,
RealD _mass,RealD M5) :
WilsonFermion5D(_Umu,
FiveDimGrid, FiveDimRedBlackGrid,
FourDimGrid, FourDimRedBlackGrid,M5),
mass(_mass)
{
assert((Ls&0x1)==1); // Odd Ls required
int nrational=Ls-1;
Approx::zolotarev_data *zdata = Approx::higham(1.0,nrational);
// NB: chroma uses a cast to "float" for the zolotarev range(!?).
// this creates a real difference in the operator which I do not like but we can replicate here
// to demonstrate compatibility
// RealD eps = (zolo_lo / zolo_hi);
// zdata = bfm_zolotarev(eps,nrational,0);
SetCoefficientsTanh(zdata,1.0);
Approx::zolotarev_free(zdata);
}
}
}

View File

@ -9,6 +9,13 @@ namespace Grid {
{
public:
const int part_frac_chroma_convention=1;
void Meooe_internal(const LatticeFermion &in, LatticeFermion &out,int dag);
void Mooee_internal(const LatticeFermion &in, LatticeFermion &out,int dag);
void MooeeInv_internal(const LatticeFermion &in, LatticeFermion &out,int dag);
void M_internal(const LatticeFermion &in, LatticeFermion &out,int dag);
// override multiply
virtual RealD M (const LatticeFermion &in, LatticeFermion &out);
virtual RealD Mdag (const LatticeFermion &in, LatticeFermion &out);
@ -21,16 +28,7 @@ namespace Grid {
virtual void MooeeInv (const LatticeFermion &in, LatticeFermion &out);
virtual void MooeeInvDag (const LatticeFermion &in, LatticeFermion &out);
private:
virtual void PartialFractionCoefficients(void);
Approx::zolotarev_data *zdata;
// Part frac
double R;
std::vector<double> p;
std::vector<double> q;
virtual void Instantiatable(void) =0; // ensure no make-eee
// Constructors
PartialFractionFermion5D(LatticeGaugeField &_Umu,
@ -40,6 +38,20 @@ namespace Grid {
GridRedBlackCartesian &FourDimRedBlackGrid,
RealD _mass,RealD M5);
protected:
virtual void SetCoefficientsTanh(Approx::zolotarev_data *zdata,RealD scale);
virtual void SetCoefficientsZolotarev(RealD zolo_hi,Approx::zolotarev_data *zdata);
// Part frac
RealD mass;
RealD dw_diag;
RealD R;
RealD amax;
RealD scale;
std::vector<double> p;
std::vector<double> q;
};

View File

@ -1,5 +1,5 @@
#ifndef GRID_QCD_DWF_H
#define GRID_QCD_DWF_H
#ifndef GRID_QCD_WILSON_FERMION_5D_H
#define GRID_QCD_WILSON_FERMION_5D_H
namespace Grid {

View File

@ -82,6 +82,16 @@ int main (int argc, char ** argv)
OverlapWilsonContFracZolotarevFermion Dcfz(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,0.1,6.0);
TestCGinversions<OverlapWilsonContFracZolotarevFermion>(Dcfz,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,&RNG4,&RNG5);
std::cout <<"OverlapWilsonPartialFractionTanhFermion test"<<std::endl;
OverlapWilsonPartialFractionTanhFermion Dpf(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,1.0);
TestCGinversions<OverlapWilsonPartialFractionTanhFermion>(Dpf,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,&RNG4,&RNG5);
std::cout <<"OverlapWilsonPartialFractionZolotarevFermion test"<<std::endl;
OverlapWilsonPartialFractionZolotarevFermion Dpfz(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,0.1,6.0);
TestCGinversions<OverlapWilsonPartialFractionZolotarevFermion>(Dpfz,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,&RNG4,&RNG5);
Grid_finalize();
}
template<class What>

View File

@ -57,6 +57,14 @@ int main (int argc, char ** argv)
OverlapWilsonContFracZolotarevFermion Dcfz(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,0.1,6.0);
TestWhat<OverlapWilsonContFracZolotarevFermion>(Dcfz,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,&RNG4,&RNG5);
std::cout <<"OverlapWilsonPartialFractionTanhFermion test"<<std::endl;
OverlapWilsonPartialFractionTanhFermion Dpf(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,1.0);
TestWhat<OverlapWilsonPartialFractionTanhFermion>(Dpf,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,&RNG4,&RNG5);
std::cout <<"OverlapWilsonPartialFractionZolotarevFermion test"<<std::endl;
OverlapWilsonPartialFractionZolotarevFermion Dpfz(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,0.1,6.0);
TestWhat<OverlapWilsonPartialFractionZolotarevFermion>(Dpfz,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,&RNG4,&RNG5);
Grid_finalize();
}