mirror of
https://github.com/paboyle/Grid.git
synced 2025-06-13 20:57:06 +01:00
Compare commits
1 Commits
feature/di
...
feature/ad
| Author | SHA1 | Date | |
|---|---|---|---|
| b284d50863 |
1
.gitignore
vendored
1
.gitignore
vendored
@ -88,7 +88,6 @@ Thumbs.db
|
||||
# build directory #
|
||||
###################
|
||||
build*/*
|
||||
Documentation/_build
|
||||
|
||||
# IDE related files #
|
||||
#####################
|
||||
|
||||
@ -44,22 +44,14 @@ directory
|
||||
#ifdef __NVCC__
|
||||
//disables nvcc specific warning in json.hpp
|
||||
#pragma clang diagnostic ignored "-Wdeprecated-register"
|
||||
|
||||
#if (__CUDACC_VER_MAJOR__ >= 11) && (__CUDACC_VER_MINOR__ >= 5)
|
||||
//disables nvcc specific warning in json.hpp
|
||||
#pragma nv_diag_suppress unsigned_compare_with_zero
|
||||
#pragma nv_diag_suppress cast_to_qualified_type
|
||||
//disables nvcc specific warning in many files
|
||||
#pragma nv_diag_suppress esa_on_defaulted_function_ignored
|
||||
#pragma nv_diag_suppress extra_semicolon
|
||||
#else
|
||||
//disables nvcc specific warning in json.hpp
|
||||
#pragma diag_suppress unsigned_compare_with_zero
|
||||
#pragma diag_suppress cast_to_qualified_type
|
||||
|
||||
//disables nvcc specific warning in many files
|
||||
#pragma diag_suppress esa_on_defaulted_function_ignored
|
||||
#pragma diag_suppress extra_semicolon
|
||||
#endif
|
||||
|
||||
//Eigen only
|
||||
#endif
|
||||
|
||||
// Disable vectorisation in Eigen on the Power8/9 and PowerPC
|
||||
|
||||
@ -36,7 +36,6 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
|
||||
#include <Grid/GridCore.h>
|
||||
#include <Grid/qcd/QCD.h>
|
||||
#include <Grid/qcd/spin/Spin.h>
|
||||
#include <Grid/qcd/gparity/Gparity.h>
|
||||
#include <Grid/qcd/utils/Utils.h>
|
||||
#include <Grid/qcd/representations/Representations.h>
|
||||
NAMESPACE_CHECK(GridQCDCore);
|
||||
|
||||
@ -16,7 +16,6 @@
|
||||
#include <functional>
|
||||
#include <stdio.h>
|
||||
#include <stdlib.h>
|
||||
#include <strings.h>
|
||||
#include <stdio.h>
|
||||
#include <signal.h>
|
||||
#include <ctime>
|
||||
|
||||
@ -14,11 +14,7 @@
|
||||
/* NVCC save and restore compile environment*/
|
||||
#ifdef __NVCC__
|
||||
#pragma push
|
||||
#if (__CUDACC_VER_MAJOR__ >= 11) && (__CUDACC_VER_MINOR__ >= 5)
|
||||
#pragma nv_diag_suppress code_is_unreachable
|
||||
#else
|
||||
#pragma diag_suppress code_is_unreachable
|
||||
#endif
|
||||
#pragma push_macro("__CUDA_ARCH__")
|
||||
#pragma push_macro("__NVCC__")
|
||||
#pragma push_macro("__CUDACC__")
|
||||
|
||||
@ -54,7 +54,6 @@ NAMESPACE_CHECK(BiCGSTAB);
|
||||
#include <Grid/algorithms/iterative/SchurRedBlack.h>
|
||||
#include <Grid/algorithms/iterative/ConjugateGradientMultiShift.h>
|
||||
#include <Grid/algorithms/iterative/ConjugateGradientMixedPrec.h>
|
||||
#include <Grid/algorithms/iterative/ConjugateGradientMultiShiftMixedPrec.h>
|
||||
#include <Grid/algorithms/iterative/BiCGSTABMixedPrec.h>
|
||||
#include <Grid/algorithms/iterative/BlockConjugateGradient.h>
|
||||
#include <Grid/algorithms/iterative/ConjugateGradientReliableUpdate.h>
|
||||
|
||||
@ -358,7 +358,7 @@ public:
|
||||
autoView( in_v , in, AcceleratorRead);
|
||||
autoView( out_v , out, AcceleratorWrite);
|
||||
autoView( Stencil_v , Stencil, AcceleratorRead);
|
||||
int npoint = geom.npoint;
|
||||
auto& geom_v = geom;
|
||||
typedef LatticeView<Cobj> Aview;
|
||||
|
||||
Vector<Aview> AcceleratorViewContainer;
|
||||
@ -380,7 +380,7 @@ public:
|
||||
int ptype;
|
||||
StencilEntry *SE;
|
||||
|
||||
for(int point=0;point<npoint;point++){
|
||||
for(int point=0;point<geom_v.npoint;point++){
|
||||
|
||||
SE=Stencil_v.GetEntry(ptype,point,ss);
|
||||
|
||||
@ -424,7 +424,7 @@ public:
|
||||
autoView( in_v , in, AcceleratorRead);
|
||||
autoView( out_v , out, AcceleratorWrite);
|
||||
autoView( Stencil_v , Stencil, AcceleratorRead);
|
||||
int npoint = geom.npoint;
|
||||
auto& geom_v = geom;
|
||||
typedef LatticeView<Cobj> Aview;
|
||||
|
||||
Vector<Aview> AcceleratorViewContainer;
|
||||
@ -442,8 +442,6 @@ public:
|
||||
for(int p=0; p<geom.npoint; p++)
|
||||
points[p] = geom.points_dagger[p];
|
||||
|
||||
auto points_p = &points[0];
|
||||
|
||||
RealD* dag_factor_p = &dag_factor[0];
|
||||
|
||||
accelerator_for(sss, Grid()->oSites()*nbasis, Nsimd, {
|
||||
@ -454,8 +452,8 @@ public:
|
||||
int ptype;
|
||||
StencilEntry *SE;
|
||||
|
||||
for(int p=0;p<npoint;p++){
|
||||
int point = points_p[p];
|
||||
for(int p=0;p<geom_v.npoint;p++){
|
||||
int point = points[p];
|
||||
|
||||
SE=Stencil_v.GetEntry(ptype,point,ss);
|
||||
|
||||
@ -710,8 +708,6 @@ public:
|
||||
for(int p=0; p<npoint; p++)
|
||||
points[p] = (dag && !hermitian) ? geom.points_dagger[p] : p;
|
||||
|
||||
auto points_p = &points[0];
|
||||
|
||||
Vector<Aview> AcceleratorViewContainer;
|
||||
for(int p=0;p<npoint;p++) AcceleratorViewContainer.push_back(a[p].View(AcceleratorRead));
|
||||
Aview *Aview_p = & AcceleratorViewContainer[0];
|
||||
@ -732,7 +728,7 @@ public:
|
||||
StencilEntry *SE;
|
||||
|
||||
for(int p=0;p<npoint;p++){
|
||||
int point = points_p[p];
|
||||
int point = points[p];
|
||||
SE=st_v.GetEntry(ptype,point,ss);
|
||||
|
||||
if(SE->_is_local) {
|
||||
@ -758,7 +754,7 @@ public:
|
||||
StencilEntry *SE;
|
||||
|
||||
for(int p=0;p<npoint;p++){
|
||||
int point = points_p[p];
|
||||
int point = points[p];
|
||||
SE=st_v.GetEntry(ptype,point,ss);
|
||||
|
||||
if(SE->_is_local) {
|
||||
|
||||
@ -136,7 +136,7 @@ public:
|
||||
flops=0;
|
||||
usec =0;
|
||||
Coordinate layout(Nd,1);
|
||||
sgrid = new GridCartesian(dimensions,layout,processors,*grid);
|
||||
sgrid = new GridCartesian(dimensions,layout,processors);
|
||||
};
|
||||
|
||||
~FFT ( void) {
|
||||
@ -182,7 +182,7 @@ public:
|
||||
pencil_gd[dim] = G*processors[dim];
|
||||
|
||||
// Pencil global vol LxLxGxLxL per node
|
||||
GridCartesian pencil_g(pencil_gd,layout,processors,*vgrid);
|
||||
GridCartesian pencil_g(pencil_gd,layout,processors);
|
||||
|
||||
// Construct pencils
|
||||
typedef typename vobj::scalar_object sobj;
|
||||
|
||||
@ -52,7 +52,6 @@ public:
|
||||
virtual void AdjOp (const Field &in, Field &out) = 0; // Abstract base
|
||||
virtual void HermOpAndNorm(const Field &in, Field &out,RealD &n1,RealD &n2)=0;
|
||||
virtual void HermOp(const Field &in, Field &out)=0;
|
||||
virtual ~LinearOperatorBase(){};
|
||||
};
|
||||
|
||||
|
||||
@ -508,7 +507,7 @@ class SchurStaggeredOperator : public SchurOperatorBase<Field> {
|
||||
virtual void MpcDag (const Field &in, Field &out){
|
||||
Mpc(in,out);
|
||||
}
|
||||
virtual void MpcDagMpc(const Field &in, Field &out) {
|
||||
virtual void MpcDagMpc(const Field &in, Field &out,RealD &ni,RealD &no) {
|
||||
assert(0);// Never need with staggered
|
||||
}
|
||||
};
|
||||
@ -531,16 +530,6 @@ public:
|
||||
template<class Field> class LinearFunction {
|
||||
public:
|
||||
virtual void operator() (const Field &in, Field &out) = 0;
|
||||
|
||||
virtual void operator() (const std::vector<Field> &in, std::vector<Field> &out)
|
||||
{
|
||||
assert(in.size() == out.size());
|
||||
|
||||
for (unsigned int i = 0; i < in.size(); ++i)
|
||||
{
|
||||
(*this)(in[i], out[i]);
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
template<class Field> class IdentityLinearFunction : public LinearFunction<Field> {
|
||||
@ -586,7 +575,6 @@ class HermOpOperatorFunction : public OperatorFunction<Field> {
|
||||
template<typename Field>
|
||||
class PlainHermOp : public LinearFunction<Field> {
|
||||
public:
|
||||
using LinearFunction<Field>::operator();
|
||||
LinearOperatorBase<Field> &_Linop;
|
||||
|
||||
PlainHermOp(LinearOperatorBase<Field>& linop) : _Linop(linop)
|
||||
@ -600,7 +588,6 @@ public:
|
||||
template<typename Field>
|
||||
class FunctionHermOp : public LinearFunction<Field> {
|
||||
public:
|
||||
using LinearFunction<Field>::operator();
|
||||
OperatorFunction<Field> & _poly;
|
||||
LinearOperatorBase<Field> &_Linop;
|
||||
|
||||
|
||||
@ -30,19 +30,13 @@ Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
|
||||
|
||||
NAMESPACE_BEGIN(Grid);
|
||||
|
||||
template<class Field> using Preconditioner = LinearFunction<Field> ;
|
||||
|
||||
/*
|
||||
template<class Field> class Preconditioner : public LinearFunction<Field> {
|
||||
using LinearFunction<Field>::operator();
|
||||
template<class Field> class Preconditioner : public LinearFunction<Field> {
|
||||
virtual void operator()(const Field &src, Field & psi)=0;
|
||||
};
|
||||
*/
|
||||
|
||||
template<class Field> class TrivialPrecon : public Preconditioner<Field> {
|
||||
public:
|
||||
using Preconditioner<Field>::operator();
|
||||
virtual void operator()(const Field &src, Field & psi){
|
||||
void operator()(const Field &src, Field & psi){
|
||||
psi = src;
|
||||
}
|
||||
TrivialPrecon(void){};
|
||||
|
||||
@ -48,7 +48,6 @@ public:
|
||||
virtual void Mdiag (const Field &in, Field &out)=0;
|
||||
virtual void Mdir (const Field &in, Field &out,int dir, int disp)=0;
|
||||
virtual void MdirAll (const Field &in, std::vector<Field> &out)=0;
|
||||
virtual ~SparseMatrixBase() {};
|
||||
};
|
||||
|
||||
/////////////////////////////////////////////////////////////////////////////////////////////
|
||||
@ -73,7 +72,7 @@ public:
|
||||
virtual void MeooeDag (const Field &in, Field &out)=0;
|
||||
virtual void MooeeDag (const Field &in, Field &out)=0;
|
||||
virtual void MooeeInvDag (const Field &in, Field &out)=0;
|
||||
virtual ~CheckerBoardedSparseMatrixBase() {};
|
||||
|
||||
};
|
||||
|
||||
NAMESPACE_END(Grid);
|
||||
|
||||
@ -36,8 +36,7 @@ NAMESPACE_BEGIN(Grid);
|
||||
template<class FieldD, class FieldF, typename std::enable_if< getPrecision<FieldD>::value == 2, int>::type = 0, typename std::enable_if< getPrecision<FieldF>::value == 1, int>::type = 0>
|
||||
class MixedPrecisionBiCGSTAB : public LinearFunction<FieldD>
|
||||
{
|
||||
public:
|
||||
using LinearFunction<FieldD>::operator();
|
||||
public:
|
||||
RealD Tolerance;
|
||||
RealD InnerTolerance; // Initial tolerance for inner CG. Defaults to Tolerance but can be changed
|
||||
Integer MaxInnerIterations;
|
||||
|
||||
@ -120,9 +120,6 @@ public:
|
||||
SolverTimer.Start();
|
||||
int k;
|
||||
for (k = 1; k <= MaxIterations; k++) {
|
||||
|
||||
GridStopWatch IterationTimer;
|
||||
IterationTimer.Start();
|
||||
c = cp;
|
||||
|
||||
MatrixTimer.Start();
|
||||
@ -155,14 +152,8 @@ public:
|
||||
LinearCombTimer.Stop();
|
||||
LinalgTimer.Stop();
|
||||
|
||||
IterationTimer.Stop();
|
||||
if ( (k % 500) == 0 ) {
|
||||
std::cout << GridLogMessage << "ConjugateGradient: Iteration " << k
|
||||
std::cout << GridLogIterative << "ConjugateGradient: Iteration " << k
|
||||
<< " residual " << sqrt(cp/ssq) << " target " << Tolerance << std::endl;
|
||||
} else {
|
||||
std::cout << GridLogIterative << "ConjugateGradient: Iteration " << k
|
||||
<< " residual " << sqrt(cp/ssq) << " target " << Tolerance << " took " << IterationTimer.Elapsed() << std::endl;
|
||||
}
|
||||
|
||||
// Stopping condition
|
||||
if (cp <= rsq) {
|
||||
@ -179,13 +170,13 @@ public:
|
||||
<< "\tTrue residual " << true_residual
|
||||
<< "\tTarget " << Tolerance << std::endl;
|
||||
|
||||
std::cout << GridLogMessage << "Time breakdown "<<std::endl;
|
||||
std::cout << GridLogMessage << "\tElapsed " << SolverTimer.Elapsed() <<std::endl;
|
||||
std::cout << GridLogMessage << "\tMatrix " << MatrixTimer.Elapsed() <<std::endl;
|
||||
std::cout << GridLogMessage << "\tLinalg " << LinalgTimer.Elapsed() <<std::endl;
|
||||
std::cout << GridLogMessage << "\tInner " << InnerTimer.Elapsed() <<std::endl;
|
||||
std::cout << GridLogMessage << "\tAxpyNorm " << AxpyNormTimer.Elapsed() <<std::endl;
|
||||
std::cout << GridLogMessage << "\tLinearComb " << LinearCombTimer.Elapsed() <<std::endl;
|
||||
std::cout << GridLogIterative << "Time breakdown "<<std::endl;
|
||||
std::cout << GridLogIterative << "\tElapsed " << SolverTimer.Elapsed() <<std::endl;
|
||||
std::cout << GridLogIterative << "\tMatrix " << MatrixTimer.Elapsed() <<std::endl;
|
||||
std::cout << GridLogIterative << "\tLinalg " << LinalgTimer.Elapsed() <<std::endl;
|
||||
std::cout << GridLogIterative << "\tInner " << InnerTimer.Elapsed() <<std::endl;
|
||||
std::cout << GridLogIterative << "\tAxpyNorm " << AxpyNormTimer.Elapsed() <<std::endl;
|
||||
std::cout << GridLogIterative << "\tLinearComb " << LinearCombTimer.Elapsed() <<std::endl;
|
||||
|
||||
if (ErrorOnNoConverge) assert(true_residual / Tolerance < 10000.0);
|
||||
|
||||
|
||||
@ -35,8 +35,7 @@ NAMESPACE_BEGIN(Grid);
|
||||
typename std::enable_if< getPrecision<FieldD>::value == 2, int>::type = 0,
|
||||
typename std::enable_if< getPrecision<FieldF>::value == 1, int>::type = 0>
|
||||
class MixedPrecisionConjugateGradient : public LinearFunction<FieldD> {
|
||||
public:
|
||||
using LinearFunction<FieldD>::operator();
|
||||
public:
|
||||
RealD Tolerance;
|
||||
RealD InnerTolerance; //Initial tolerance for inner CG. Defaults to Tolerance but can be changed
|
||||
Integer MaxInnerIterations;
|
||||
@ -49,7 +48,6 @@ NAMESPACE_BEGIN(Grid);
|
||||
Integer TotalInnerIterations; //Number of inner CG iterations
|
||||
Integer TotalOuterIterations; //Number of restarts
|
||||
Integer TotalFinalStepIterations; //Number of CG iterations in final patch-up step
|
||||
RealD TrueResidual;
|
||||
|
||||
//Option to speed up *inner single precision* solves using a LinearFunction that produces a guess
|
||||
LinearFunction<FieldF> *guesser;
|
||||
@ -69,7 +67,6 @@ NAMESPACE_BEGIN(Grid);
|
||||
}
|
||||
|
||||
void operator() (const FieldD &src_d_in, FieldD &sol_d){
|
||||
std::cout << GridLogMessage << "MixedPrecisionConjugateGradient: Starting mixed precision CG with outer tolerance " << Tolerance << " and inner tolerance " << InnerTolerance << std::endl;
|
||||
TotalInnerIterations = 0;
|
||||
|
||||
GridStopWatch TotalTimer;
|
||||
@ -99,7 +96,6 @@ NAMESPACE_BEGIN(Grid);
|
||||
FieldF sol_f(SinglePrecGrid);
|
||||
sol_f.Checkerboard() = cb;
|
||||
|
||||
std::cout<<GridLogMessage<<"MixedPrecisionConjugateGradient: Starting initial inner CG with tolerance " << inner_tol << std::endl;
|
||||
ConjugateGradient<FieldF> CG_f(inner_tol, MaxInnerIterations);
|
||||
CG_f.ErrorOnNoConverge = false;
|
||||
|
||||
@ -133,7 +129,6 @@ NAMESPACE_BEGIN(Grid);
|
||||
(*guesser)(src_f, sol_f);
|
||||
|
||||
//Inner CG
|
||||
std::cout<<GridLogMessage<<"MixedPrecisionConjugateGradient: Outer iteration " << outer_iter << " starting inner CG with tolerance " << inner_tol << std::endl;
|
||||
CG_f.Tolerance = inner_tol;
|
||||
InnerCGtimer.Start();
|
||||
CG_f(Linop_f, src_f, sol_f);
|
||||
@ -154,7 +149,6 @@ NAMESPACE_BEGIN(Grid);
|
||||
ConjugateGradient<FieldD> CG_d(Tolerance, MaxInnerIterations);
|
||||
CG_d(Linop_d, src_d_in, sol_d);
|
||||
TotalFinalStepIterations = CG_d.IterationsToComplete;
|
||||
TrueResidual = CG_d.TrueResidual;
|
||||
|
||||
TotalTimer.Stop();
|
||||
std::cout<<GridLogMessage<<"MixedPrecisionConjugateGradient: Inner CG iterations " << TotalInnerIterations << " Restarts " << TotalOuterIterations << " Final CG iterations " << TotalFinalStepIterations << std::endl;
|
||||
|
||||
@ -44,7 +44,7 @@ public:
|
||||
|
||||
using OperatorFunction<Field>::operator();
|
||||
|
||||
// RealD Tolerance;
|
||||
RealD Tolerance;
|
||||
Integer MaxIterations;
|
||||
Integer IterationsToComplete; //Number of iterations the CG took to finish. Filled in upon completion
|
||||
std::vector<int> IterationsToCompleteShift; // Iterations for this shift
|
||||
@ -52,7 +52,7 @@ public:
|
||||
MultiShiftFunction shifts;
|
||||
std::vector<RealD> TrueResidualShift;
|
||||
|
||||
ConjugateGradientMultiShift(Integer maxit, const MultiShiftFunction &_shifts) :
|
||||
ConjugateGradientMultiShift(Integer maxit,MultiShiftFunction &_shifts) :
|
||||
MaxIterations(maxit),
|
||||
shifts(_shifts)
|
||||
{
|
||||
@ -182,9 +182,6 @@ public:
|
||||
for(int s=0;s<nshift;s++) {
|
||||
axpby(psi[s],0.,-bs[s]*alpha[s],src,src);
|
||||
}
|
||||
|
||||
std::cout << GridLogIterative << "ConjugateGradientMultiShift: initial rn (|src|^2) =" << rn << " qq (|MdagM src|^2) =" << qq << " d ( dot(src, [MdagM + m_0]src) ) =" << d << " c=" << c << std::endl;
|
||||
|
||||
|
||||
///////////////////////////////////////
|
||||
// Timers
|
||||
@ -324,8 +321,8 @@ public:
|
||||
|
||||
std::cout << GridLogMessage << "Time Breakdown "<<std::endl;
|
||||
std::cout << GridLogMessage << "\tElapsed " << SolverTimer.Elapsed() <<std::endl;
|
||||
std::cout << GridLogMessage << "\tAXPY " << AXPYTimer.Elapsed() <<std::endl;
|
||||
std::cout << GridLogMessage << "\tMatrix " << MatrixTimer.Elapsed() <<std::endl;
|
||||
std::cout << GridLogMessage << "\tAXPY " << AXPYTimer.Elapsed() <<std::endl;
|
||||
std::cout << GridLogMessage << "\tMarix " << MatrixTimer.Elapsed() <<std::endl;
|
||||
std::cout << GridLogMessage << "\tShift " << ShiftTimer.Elapsed() <<std::endl;
|
||||
|
||||
IterationsToComplete = k;
|
||||
|
||||
@ -1,409 +0,0 @@
|
||||
/*************************************************************************************
|
||||
|
||||
Grid physics library, www.github.com/paboyle/Grid
|
||||
|
||||
Source file: ./lib/algorithms/iterative/ConjugateGradientMultiShift.h
|
||||
|
||||
Copyright (C) 2015
|
||||
|
||||
Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
|
||||
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
|
||||
Author: Christopher Kelly <ckelly@bnl.gov>
|
||||
|
||||
This program is free software; you can redistribute it and/or modify
|
||||
it under the terms of the GNU General Public License as published by
|
||||
the Free Software Foundation; either version 2 of the License, or
|
||||
(at your option) any later version.
|
||||
|
||||
This program is distributed in the hope that it will be useful,
|
||||
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||||
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
||||
GNU General Public License for more details.
|
||||
|
||||
You should have received a copy of the GNU General Public License along
|
||||
with this program; if not, write to the Free Software Foundation, Inc.,
|
||||
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
|
||||
|
||||
See the full license in the file "LICENSE" in the top level distribution directory
|
||||
*************************************************************************************/
|
||||
/* END LEGAL */
|
||||
#ifndef GRID_CONJUGATE_GRADIENT_MULTI_SHIFT_MIXEDPREC_H
|
||||
#define GRID_CONJUGATE_GRADIENT_MULTI_SHIFT_MIXEDPREC_H
|
||||
|
||||
NAMESPACE_BEGIN(Grid);
|
||||
|
||||
//CK 2020: A variant of the multi-shift conjugate gradient with the matrix multiplication in single precision.
|
||||
//The residual is stored in single precision, but the search directions and solution are stored in double precision.
|
||||
//Every update_freq iterations the residual is corrected in double precision.
|
||||
|
||||
//For safety the a final regular CG is applied to clean up if necessary
|
||||
|
||||
//Linop to add shift to input linop, used in cleanup CG
|
||||
namespace ConjugateGradientMultiShiftMixedPrecSupport{
|
||||
template<typename Field>
|
||||
class ShiftedLinop: public LinearOperatorBase<Field>{
|
||||
public:
|
||||
LinearOperatorBase<Field> &linop_base;
|
||||
RealD shift;
|
||||
|
||||
ShiftedLinop(LinearOperatorBase<Field> &_linop_base, RealD _shift): linop_base(_linop_base), shift(_shift){}
|
||||
|
||||
void OpDiag (const Field &in, Field &out){ assert(0); }
|
||||
void OpDir (const Field &in, Field &out,int dir,int disp){ assert(0); }
|
||||
void OpDirAll (const Field &in, std::vector<Field> &out){ assert(0); }
|
||||
|
||||
void Op (const Field &in, Field &out){ assert(0); }
|
||||
void AdjOp (const Field &in, Field &out){ assert(0); }
|
||||
|
||||
void HermOp(const Field &in, Field &out){
|
||||
linop_base.HermOp(in, out);
|
||||
axpy(out, shift, in, out);
|
||||
}
|
||||
|
||||
void HermOpAndNorm(const Field &in, Field &out,RealD &n1,RealD &n2){
|
||||
HermOp(in,out);
|
||||
ComplexD dot = innerProduct(in,out);
|
||||
n1=real(dot);
|
||||
n2=norm2(out);
|
||||
}
|
||||
};
|
||||
};
|
||||
|
||||
|
||||
template<class FieldD, class FieldF,
|
||||
typename std::enable_if< getPrecision<FieldD>::value == 2, int>::type = 0,
|
||||
typename std::enable_if< getPrecision<FieldF>::value == 1, int>::type = 0>
|
||||
class ConjugateGradientMultiShiftMixedPrec : public OperatorMultiFunction<FieldD>,
|
||||
public OperatorFunction<FieldD>
|
||||
{
|
||||
public:
|
||||
|
||||
using OperatorFunction<FieldD>::operator();
|
||||
|
||||
RealD Tolerance;
|
||||
Integer MaxIterations;
|
||||
Integer IterationsToComplete; //Number of iterations the CG took to finish. Filled in upon completion
|
||||
std::vector<int> IterationsToCompleteShift; // Iterations for this shift
|
||||
int verbose;
|
||||
MultiShiftFunction shifts;
|
||||
std::vector<RealD> TrueResidualShift;
|
||||
|
||||
int ReliableUpdateFreq; //number of iterations between reliable updates
|
||||
|
||||
GridBase* SinglePrecGrid; //Grid for single-precision fields
|
||||
LinearOperatorBase<FieldF> &Linop_f; //single precision
|
||||
|
||||
ConjugateGradientMultiShiftMixedPrec(Integer maxit, const MultiShiftFunction &_shifts,
|
||||
GridBase* _SinglePrecGrid, LinearOperatorBase<FieldF> &_Linop_f,
|
||||
int _ReliableUpdateFreq
|
||||
) :
|
||||
MaxIterations(maxit), shifts(_shifts), SinglePrecGrid(_SinglePrecGrid), Linop_f(_Linop_f), ReliableUpdateFreq(_ReliableUpdateFreq)
|
||||
{
|
||||
verbose=1;
|
||||
IterationsToCompleteShift.resize(_shifts.order);
|
||||
TrueResidualShift.resize(_shifts.order);
|
||||
}
|
||||
|
||||
void operator() (LinearOperatorBase<FieldD> &Linop, const FieldD &src, FieldD &psi)
|
||||
{
|
||||
GridBase *grid = src.Grid();
|
||||
int nshift = shifts.order;
|
||||
std::vector<FieldD> results(nshift,grid);
|
||||
(*this)(Linop,src,results,psi);
|
||||
}
|
||||
void operator() (LinearOperatorBase<FieldD> &Linop, const FieldD &src, std::vector<FieldD> &results, FieldD &psi)
|
||||
{
|
||||
int nshift = shifts.order;
|
||||
|
||||
(*this)(Linop,src,results);
|
||||
|
||||
psi = shifts.norm*src;
|
||||
for(int i=0;i<nshift;i++){
|
||||
psi = psi + shifts.residues[i]*results[i];
|
||||
}
|
||||
|
||||
return;
|
||||
}
|
||||
|
||||
void operator() (LinearOperatorBase<FieldD> &Linop_d, const FieldD &src_d, std::vector<FieldD> &psi_d)
|
||||
{
|
||||
GridBase *DoublePrecGrid = src_d.Grid();
|
||||
|
||||
////////////////////////////////////////////////////////////////////////
|
||||
// Convenience references to the info stored in "MultiShiftFunction"
|
||||
////////////////////////////////////////////////////////////////////////
|
||||
int nshift = shifts.order;
|
||||
|
||||
std::vector<RealD> &mass(shifts.poles); // Make references to array in "shifts"
|
||||
std::vector<RealD> &mresidual(shifts.tolerances);
|
||||
std::vector<RealD> alpha(nshift,1.0);
|
||||
|
||||
//Double precision search directions
|
||||
FieldD p_d(DoublePrecGrid);
|
||||
std::vector<FieldD> ps_d(nshift, DoublePrecGrid);// Search directions (double precision)
|
||||
|
||||
FieldD tmp_d(DoublePrecGrid);
|
||||
FieldD r_d(DoublePrecGrid);
|
||||
FieldD mmp_d(DoublePrecGrid);
|
||||
|
||||
assert(psi_d.size()==nshift);
|
||||
assert(mass.size()==nshift);
|
||||
assert(mresidual.size()==nshift);
|
||||
|
||||
// dynamic sized arrays on stack; 2d is a pain with vector
|
||||
RealD bs[nshift];
|
||||
RealD rsq[nshift];
|
||||
RealD z[nshift][2];
|
||||
int converged[nshift];
|
||||
|
||||
const int primary =0;
|
||||
|
||||
//Primary shift fields CG iteration
|
||||
RealD a,b,c,d;
|
||||
RealD cp,bp,qq; //prev
|
||||
|
||||
// Matrix mult fields
|
||||
FieldF r_f(SinglePrecGrid);
|
||||
FieldF p_f(SinglePrecGrid);
|
||||
FieldF tmp_f(SinglePrecGrid);
|
||||
FieldF mmp_f(SinglePrecGrid);
|
||||
FieldF src_f(SinglePrecGrid);
|
||||
precisionChange(src_f, src_d);
|
||||
|
||||
// Check lightest mass
|
||||
for(int s=0;s<nshift;s++){
|
||||
assert( mass[s]>= mass[primary] );
|
||||
converged[s]=0;
|
||||
}
|
||||
|
||||
// Wire guess to zero
|
||||
// Residuals "r" are src
|
||||
// First search direction "p" is also src
|
||||
cp = norm2(src_d);
|
||||
|
||||
// Handle trivial case of zero src.
|
||||
if( cp == 0. ){
|
||||
for(int s=0;s<nshift;s++){
|
||||
psi_d[s] = Zero();
|
||||
IterationsToCompleteShift[s] = 1;
|
||||
TrueResidualShift[s] = 0.;
|
||||
}
|
||||
return;
|
||||
}
|
||||
|
||||
for(int s=0;s<nshift;s++){
|
||||
rsq[s] = cp * mresidual[s] * mresidual[s];
|
||||
std::cout<<GridLogMessage<<"ConjugateGradientMultiShiftMixedPrec: shift "<< s <<" target resid "<<rsq[s]<<std::endl;
|
||||
ps_d[s] = src_d;
|
||||
}
|
||||
// r and p for primary
|
||||
r_f=src_f; //residual maintained in single
|
||||
p_f=src_f;
|
||||
p_d = src_d; //primary copy --- make this a reference to ps_d to save axpys
|
||||
|
||||
//MdagM+m[0]
|
||||
Linop_f.HermOpAndNorm(p_f,mmp_f,d,qq); // mmp = MdagM p d=real(dot(p, mmp)), qq=norm2(mmp)
|
||||
axpy(mmp_f,mass[0],p_f,mmp_f);
|
||||
RealD rn = norm2(p_f);
|
||||
d += rn*mass[0];
|
||||
|
||||
b = -cp /d;
|
||||
|
||||
// Set up the various shift variables
|
||||
int iz=0;
|
||||
z[0][1-iz] = 1.0;
|
||||
z[0][iz] = 1.0;
|
||||
bs[0] = b;
|
||||
for(int s=1;s<nshift;s++){
|
||||
z[s][1-iz] = 1.0;
|
||||
z[s][iz] = 1.0/( 1.0 - b*(mass[s]-mass[0]));
|
||||
bs[s] = b*z[s][iz];
|
||||
}
|
||||
|
||||
// r += b[0] A.p[0]
|
||||
// c= norm(r)
|
||||
c=axpy_norm(r_f,b,mmp_f,r_f);
|
||||
|
||||
for(int s=0;s<nshift;s++) {
|
||||
axpby(psi_d[s],0.,-bs[s]*alpha[s],src_d,src_d);
|
||||
}
|
||||
|
||||
///////////////////////////////////////
|
||||
// Timers
|
||||
///////////////////////////////////////
|
||||
GridStopWatch AXPYTimer, ShiftTimer, QRTimer, MatrixTimer, SolverTimer, PrecChangeTimer, CleanupTimer;
|
||||
|
||||
SolverTimer.Start();
|
||||
|
||||
// Iteration loop
|
||||
int k;
|
||||
|
||||
for (k=1;k<=MaxIterations;k++){
|
||||
a = c /cp;
|
||||
|
||||
//Update double precision search direction by residual
|
||||
PrecChangeTimer.Start();
|
||||
precisionChange(r_d, r_f);
|
||||
PrecChangeTimer.Stop();
|
||||
|
||||
AXPYTimer.Start();
|
||||
axpy(p_d,a,p_d,r_d);
|
||||
|
||||
for(int s=0;s<nshift;s++){
|
||||
if ( ! converged[s] ) {
|
||||
if (s==0){
|
||||
axpy(ps_d[s],a,ps_d[s],r_d);
|
||||
} else{
|
||||
RealD as =a *z[s][iz]*bs[s] /(z[s][1-iz]*b);
|
||||
axpby(ps_d[s],z[s][iz],as,r_d,ps_d[s]);
|
||||
}
|
||||
}
|
||||
}
|
||||
AXPYTimer.Stop();
|
||||
|
||||
PrecChangeTimer.Start();
|
||||
precisionChange(p_f, p_d); //get back single prec search direction for linop
|
||||
PrecChangeTimer.Stop();
|
||||
|
||||
cp=c;
|
||||
MatrixTimer.Start();
|
||||
Linop_f.HermOp(p_f,mmp_f);
|
||||
d=real(innerProduct(p_f,mmp_f));
|
||||
MatrixTimer.Stop();
|
||||
|
||||
AXPYTimer.Start();
|
||||
axpy(mmp_f,mass[0],p_f,mmp_f);
|
||||
AXPYTimer.Stop();
|
||||
RealD rn = norm2(p_f);
|
||||
d += rn*mass[0];
|
||||
|
||||
bp=b;
|
||||
b=-cp/d;
|
||||
|
||||
// Toggle the recurrence history
|
||||
bs[0] = b;
|
||||
iz = 1-iz;
|
||||
ShiftTimer.Start();
|
||||
for(int s=1;s<nshift;s++){
|
||||
if((!converged[s])){
|
||||
RealD z0 = z[s][1-iz];
|
||||
RealD z1 = z[s][iz];
|
||||
z[s][iz] = z0*z1*bp
|
||||
/ (b*a*(z1-z0) + z1*bp*(1- (mass[s]-mass[0])*b));
|
||||
bs[s] = b*z[s][iz]/z0; // NB sign rel to Mike
|
||||
}
|
||||
}
|
||||
ShiftTimer.Stop();
|
||||
|
||||
//Update double precision solutions
|
||||
AXPYTimer.Start();
|
||||
for(int s=0;s<nshift;s++){
|
||||
int ss = s;
|
||||
if( (!converged[s]) ) {
|
||||
axpy(psi_d[ss],-bs[s]*alpha[s],ps_d[s],psi_d[ss]);
|
||||
}
|
||||
}
|
||||
|
||||
//Perform reliable update if necessary; otherwise update residual from single-prec mmp
|
||||
RealD c_f = axpy_norm(r_f,b,mmp_f,r_f);
|
||||
AXPYTimer.Stop();
|
||||
|
||||
c = c_f;
|
||||
|
||||
if(k % ReliableUpdateFreq == 0){
|
||||
//Replace r with true residual
|
||||
MatrixTimer.Start();
|
||||
Linop_d.HermOp(psi_d[0],mmp_d);
|
||||
MatrixTimer.Stop();
|
||||
|
||||
AXPYTimer.Start();
|
||||
axpy(mmp_d,mass[0],psi_d[0],mmp_d);
|
||||
|
||||
RealD c_d = axpy_norm(r_d, -1.0, mmp_d, src_d);
|
||||
AXPYTimer.Stop();
|
||||
|
||||
std::cout<<GridLogMessage<<"ConjugateGradientMultiShiftMixedPrec k="<<k<< ", replaced |r|^2 = "<<c_f <<" with |r|^2 = "<<c_d<<std::endl;
|
||||
|
||||
PrecChangeTimer.Start();
|
||||
precisionChange(r_f, r_d);
|
||||
PrecChangeTimer.Stop();
|
||||
c = c_d;
|
||||
}
|
||||
|
||||
// Convergence checks
|
||||
int all_converged = 1;
|
||||
for(int s=0;s<nshift;s++){
|
||||
|
||||
if ( (!converged[s]) ){
|
||||
IterationsToCompleteShift[s] = k;
|
||||
|
||||
RealD css = c * z[s][iz]* z[s][iz];
|
||||
|
||||
if(css<rsq[s]){
|
||||
if ( ! converged[s] )
|
||||
std::cout<<GridLogMessage<<"ConjugateGradientMultiShiftMixedPrec k="<<k<<" Shift "<<s<<" has converged"<<std::endl;
|
||||
converged[s]=1;
|
||||
} else {
|
||||
all_converged=0;
|
||||
}
|
||||
|
||||
}
|
||||
}
|
||||
|
||||
if ( all_converged ){
|
||||
|
||||
SolverTimer.Stop();
|
||||
std::cout<<GridLogMessage<< "ConjugateGradientMultiShiftMixedPrec: All shifts have converged iteration "<<k<<std::endl;
|
||||
std::cout<<GridLogMessage<< "ConjugateGradientMultiShiftMixedPrec: Checking solutions"<<std::endl;
|
||||
|
||||
// Check answers
|
||||
for(int s=0; s < nshift; s++) {
|
||||
Linop_d.HermOpAndNorm(psi_d[s],mmp_d,d,qq);
|
||||
axpy(tmp_d,mass[s],psi_d[s],mmp_d);
|
||||
axpy(r_d,-alpha[s],src_d,tmp_d);
|
||||
RealD rn = norm2(r_d);
|
||||
RealD cn = norm2(src_d);
|
||||
TrueResidualShift[s] = std::sqrt(rn/cn);
|
||||
std::cout<<GridLogMessage<<"ConjugateGradientMultiShiftMixedPrec: shift["<<s<<"] true residual "<< TrueResidualShift[s] << " target " << mresidual[s] << std::endl;
|
||||
|
||||
//If we have not reached the desired tolerance, do a (mixed precision) CG cleanup
|
||||
if(rn >= rsq[s]){
|
||||
CleanupTimer.Start();
|
||||
std::cout<<GridLogMessage<<"ConjugateGradientMultiShiftMixedPrec: performing cleanup step for shift " << s << std::endl;
|
||||
|
||||
//Setup linear operators for final cleanup
|
||||
ConjugateGradientMultiShiftMixedPrecSupport::ShiftedLinop<FieldD> Linop_shift_d(Linop_d, mass[s]);
|
||||
ConjugateGradientMultiShiftMixedPrecSupport::ShiftedLinop<FieldF> Linop_shift_f(Linop_f, mass[s]);
|
||||
|
||||
MixedPrecisionConjugateGradient<FieldD,FieldF> cg(mresidual[s], MaxIterations, MaxIterations, SinglePrecGrid, Linop_shift_f, Linop_shift_d);
|
||||
cg(src_d, psi_d[s]);
|
||||
|
||||
TrueResidualShift[s] = cg.TrueResidual;
|
||||
CleanupTimer.Stop();
|
||||
}
|
||||
}
|
||||
|
||||
std::cout << GridLogMessage << "ConjugateGradientMultiShiftMixedPrec: Time Breakdown for body"<<std::endl;
|
||||
std::cout << GridLogMessage << "\tSolver " << SolverTimer.Elapsed() <<std::endl;
|
||||
std::cout << GridLogMessage << "\t\tAXPY " << AXPYTimer.Elapsed() <<std::endl;
|
||||
std::cout << GridLogMessage << "\t\tMatrix " << MatrixTimer.Elapsed() <<std::endl;
|
||||
std::cout << GridLogMessage << "\t\tShift " << ShiftTimer.Elapsed() <<std::endl;
|
||||
std::cout << GridLogMessage << "\t\tPrecision Change " << PrecChangeTimer.Elapsed() <<std::endl;
|
||||
std::cout << GridLogMessage << "\tFinal Cleanup " << CleanupTimer.Elapsed() <<std::endl;
|
||||
std::cout << GridLogMessage << "\tSolver+Cleanup " << SolverTimer.Elapsed() + CleanupTimer.Elapsed() << std::endl;
|
||||
|
||||
IterationsToComplete = k;
|
||||
|
||||
return;
|
||||
}
|
||||
|
||||
|
||||
}
|
||||
// ugly hack
|
||||
std::cout<<GridLogMessage<<"CG multi shift did not converge"<<std::endl;
|
||||
// assert(0);
|
||||
}
|
||||
|
||||
};
|
||||
NAMESPACE_END(Grid);
|
||||
#endif
|
||||
@ -33,19 +33,16 @@ namespace Grid {
|
||||
template<class Field>
|
||||
class ZeroGuesser: public LinearFunction<Field> {
|
||||
public:
|
||||
using LinearFunction<Field>::operator();
|
||||
virtual void operator()(const Field &src, Field &guess) { guess = Zero(); };
|
||||
};
|
||||
template<class Field>
|
||||
class DoNothingGuesser: public LinearFunction<Field> {
|
||||
public:
|
||||
using LinearFunction<Field>::operator();
|
||||
virtual void operator()(const Field &src, Field &guess) { };
|
||||
};
|
||||
template<class Field>
|
||||
class SourceGuesser: public LinearFunction<Field> {
|
||||
public:
|
||||
using LinearFunction<Field>::operator();
|
||||
virtual void operator()(const Field &src, Field &guess) { guess = src; };
|
||||
};
|
||||
|
||||
@ -57,24 +54,15 @@ class DeflatedGuesser: public LinearFunction<Field> {
|
||||
private:
|
||||
const std::vector<Field> &evec;
|
||||
const std::vector<RealD> &eval;
|
||||
const unsigned int N;
|
||||
|
||||
public:
|
||||
using LinearFunction<Field>::operator();
|
||||
|
||||
DeflatedGuesser(const std::vector<Field> & _evec,const std::vector<RealD> & _eval)
|
||||
: DeflatedGuesser(_evec, _eval, _evec.size())
|
||||
{}
|
||||
|
||||
DeflatedGuesser(const std::vector<Field> & _evec, const std::vector<RealD> & _eval, const unsigned int _N)
|
||||
: evec(_evec), eval(_eval), N(_N)
|
||||
{
|
||||
assert(evec.size()==eval.size());
|
||||
assert(N <= evec.size());
|
||||
}
|
||||
DeflatedGuesser(const std::vector<Field> & _evec,const std::vector<RealD> & _eval) : evec(_evec), eval(_eval) {};
|
||||
|
||||
virtual void operator()(const Field &src,Field &guess) {
|
||||
guess = Zero();
|
||||
assert(evec.size()==eval.size());
|
||||
auto N = evec.size();
|
||||
for (int i=0;i<N;i++) {
|
||||
const Field& tmp = evec[i];
|
||||
axpy(guess,TensorRemove(innerProduct(tmp,src)) / eval[i],tmp,guess);
|
||||
@ -91,7 +79,6 @@ private:
|
||||
const std::vector<RealD> &eval_coarse;
|
||||
public:
|
||||
|
||||
using LinearFunction<FineField>::operator();
|
||||
LocalCoherenceDeflatedGuesser(const std::vector<FineField> &_subspace,
|
||||
const std::vector<CoarseField> &_evec_coarse,
|
||||
const std::vector<RealD> &_eval_coarse)
|
||||
@ -113,43 +100,7 @@ public:
|
||||
blockPromote(guess_coarse,guess,subspace);
|
||||
guess.Checkerboard() = src.Checkerboard();
|
||||
};
|
||||
|
||||
void operator()(const std::vector<FineField> &src,std::vector<FineField> &guess) {
|
||||
int Nevec = (int)evec_coarse.size();
|
||||
int Nsrc = (int)src.size();
|
||||
// make temp variables
|
||||
std::vector<CoarseField> src_coarse(Nsrc,evec_coarse[0].Grid());
|
||||
std::vector<CoarseField> guess_coarse(Nsrc,evec_coarse[0].Grid());
|
||||
//Preporcessing
|
||||
std::cout << GridLogMessage << "Start BlockProject for loop" << std::endl;
|
||||
for (int j=0;j<Nsrc;j++)
|
||||
{
|
||||
guess_coarse[j] = Zero();
|
||||
std::cout << GridLogMessage << "BlockProject iter: " << j << std::endl;
|
||||
blockProject(src_coarse[j],src[j],subspace);
|
||||
}
|
||||
//deflation set up for eigen vector batchsize 1 and source batch size equal number of sources
|
||||
std::cout << GridLogMessage << "Start ProjectAccum for loop" << std::endl;
|
||||
for (int i=0;i<Nevec;i++)
|
||||
{
|
||||
std::cout << GridLogMessage << "ProjectAccum Nvec: " << i << std::endl;
|
||||
const CoarseField & tmp = evec_coarse[i];
|
||||
for (int j=0;j<Nsrc;j++)
|
||||
{
|
||||
axpy(guess_coarse[j],TensorRemove(innerProduct(tmp,src_coarse[j])) / eval_coarse[i],tmp,guess_coarse[j]);
|
||||
}
|
||||
}
|
||||
//postprocessing
|
||||
std::cout << GridLogMessage << "Start BlockPromote for loop" << std::endl;
|
||||
for (int j=0;j<Nsrc;j++)
|
||||
{
|
||||
std::cout << GridLogMessage << "BlockProject iter: " << j << std::endl;
|
||||
blockPromote(guess_coarse[j],guess[j],subspace);
|
||||
guess[j].Checkerboard() = src[j].Checkerboard();
|
||||
}
|
||||
};
|
||||
|
||||
};
|
||||
};
|
||||
|
||||
|
||||
|
||||
|
||||
@ -44,7 +44,6 @@ public:
|
||||
int, MinRes); // Must restart
|
||||
};
|
||||
|
||||
//This class is the input parameter class for some testing programs
|
||||
struct LocalCoherenceLanczosParams : Serializable {
|
||||
public:
|
||||
GRID_SERIALIZABLE_CLASS_MEMBERS(LocalCoherenceLanczosParams,
|
||||
@ -68,7 +67,6 @@ public:
|
||||
template<class Fobj,class CComplex,int nbasis>
|
||||
class ProjectedHermOp : public LinearFunction<Lattice<iVector<CComplex,nbasis > > > {
|
||||
public:
|
||||
using LinearFunction<Lattice<iVector<CComplex,nbasis > > >::operator();
|
||||
typedef iVector<CComplex,nbasis > CoarseSiteVector;
|
||||
typedef Lattice<CoarseSiteVector> CoarseField;
|
||||
typedef Lattice<CComplex> CoarseScalar; // used for inner products on fine field
|
||||
@ -99,7 +97,6 @@ public:
|
||||
template<class Fobj,class CComplex,int nbasis>
|
||||
class ProjectedFunctionHermOp : public LinearFunction<Lattice<iVector<CComplex,nbasis > > > {
|
||||
public:
|
||||
using LinearFunction<Lattice<iVector<CComplex,nbasis > > >::operator();
|
||||
typedef iVector<CComplex,nbasis > CoarseSiteVector;
|
||||
typedef Lattice<CoarseSiteVector> CoarseField;
|
||||
typedef Lattice<CComplex> CoarseScalar; // used for inner products on fine field
|
||||
@ -146,24 +143,16 @@ public:
|
||||
LinearOperatorBase<FineField> &_Linop;
|
||||
RealD _coarse_relax_tol;
|
||||
std::vector<FineField> &_subspace;
|
||||
|
||||
int _largestEvalIdxForReport; //The convergence of the LCL is based on the evals of the coarse grid operator, not those of the underlying fine grid operator
|
||||
//As a result we do not know what the eval range of the fine operator is until the very end, making tuning the Cheby bounds very difficult
|
||||
//To work around this issue, every restart we separately reconstruct the fine operator eval for the lowest and highest evec and print these
|
||||
//out alongside the evals of the coarse operator. To do so we need to know the index of the largest eval (i.e. Nstop-1)
|
||||
//NOTE: If largestEvalIdxForReport=-1 (default) then this is not performed
|
||||
|
||||
ImplicitlyRestartedLanczosSmoothedTester(LinearFunction<CoarseField> &Poly,
|
||||
OperatorFunction<FineField> &smoother,
|
||||
LinearOperatorBase<FineField> &Linop,
|
||||
std::vector<FineField> &subspace,
|
||||
RealD coarse_relax_tol=5.0e3,
|
||||
int largestEvalIdxForReport=-1)
|
||||
RealD coarse_relax_tol=5.0e3)
|
||||
: _smoother(smoother), _Linop(Linop), _Poly(Poly), _subspace(subspace),
|
||||
_coarse_relax_tol(coarse_relax_tol), _largestEvalIdxForReport(largestEvalIdxForReport)
|
||||
_coarse_relax_tol(coarse_relax_tol)
|
||||
{ };
|
||||
|
||||
//evalMaxApprox: approximation of largest eval of the fine Chebyshev operator (suitably wrapped by block projection)
|
||||
int TestConvergence(int j,RealD eresid,CoarseField &B, RealD &eval,RealD evalMaxApprox)
|
||||
{
|
||||
CoarseField v(B);
|
||||
@ -186,26 +175,12 @@ public:
|
||||
<<" |H B[i] - eval[i]B[i]|^2 / evalMaxApprox^2 " << std::setw(25) << vv
|
||||
<<std::endl;
|
||||
|
||||
if(_largestEvalIdxForReport != -1 && (j==0 || j==_largestEvalIdxForReport)){
|
||||
std::cout<<GridLogIRL << "Estimating true eval of fine grid operator for eval idx " << j << std::endl;
|
||||
RealD tmp_eval;
|
||||
ReconstructEval(j,eresid,B,tmp_eval,1.0); //don't use evalMaxApprox of coarse operator! (cf below)
|
||||
}
|
||||
|
||||
int conv=0;
|
||||
if( (vv<eresid*eresid) ) conv = 1;
|
||||
return conv;
|
||||
}
|
||||
|
||||
//This function is called at the end of the coarse grid Lanczos. It promotes the coarse eigenvector 'B' to the fine grid,
|
||||
//applies a smoother to the result then computes the computes the *fine grid* eigenvalue (output as 'eval').
|
||||
|
||||
//evalMaxApprox should be the approximation of the largest eval of the fine Hermop. However when this function is called by IRL it actually passes the largest eval of the *Chebyshev* operator (as this is the max approx used for the TestConvergence above)
|
||||
//As the largest eval of the Chebyshev is typically several orders of magnitude larger this makes the convergence test pass even when it should not.
|
||||
//We therefore ignore evalMaxApprox here and use a value of 1.0 (note this value is already used by TestCoarse)
|
||||
int ReconstructEval(int j,RealD eresid,CoarseField &B, RealD &eval,RealD evalMaxApprox)
|
||||
int ReconstructEval(int j,RealD eresid,CoarseField &B, RealD &eval,RealD evalMaxApprox)
|
||||
{
|
||||
evalMaxApprox = 1.0; //cf above
|
||||
GridBase *FineGrid = _subspace[0].Grid();
|
||||
int checkerboard = _subspace[0].Checkerboard();
|
||||
FineField fB(FineGrid);fB.Checkerboard() =checkerboard;
|
||||
@ -224,13 +199,13 @@ public:
|
||||
eval = vnum/vden;
|
||||
fv -= eval*fB;
|
||||
RealD vv = norm2(fv) / ::pow(evalMaxApprox,2.0);
|
||||
if ( j > nbasis ) eresid = eresid*_coarse_relax_tol;
|
||||
|
||||
|
||||
std::cout.precision(13);
|
||||
std::cout<<GridLogIRL << "[" << std::setw(3)<<j<<"] "
|
||||
<<"eval = "<<std::setw(25)<< eval << " (" << eval_poly << ")"
|
||||
<<" |H B[i] - eval[i]B[i]|^2 / evalMaxApprox^2 " << std::setw(25) << vv << " target " << eresid*eresid
|
||||
<<" |H B[i] - eval[i]B[i]|^2 / evalMaxApprox^2 " << std::setw(25) << vv
|
||||
<<std::endl;
|
||||
if ( j > nbasis ) eresid = eresid*_coarse_relax_tol;
|
||||
if( (vv<eresid*eresid) ) return 1;
|
||||
return 0;
|
||||
}
|
||||
@ -308,10 +283,6 @@ public:
|
||||
evals_coarse.resize(0);
|
||||
};
|
||||
|
||||
//The block inner product is the inner product on the fine grid locally summed over the blocks
|
||||
//to give a Lattice<Scalar> on the coarse grid. This function orthnormalizes the fine-grid subspace
|
||||
//vectors under the block inner product. This step must be performed after computing the fine grid
|
||||
//eigenvectors and before computing the coarse grid eigenvectors.
|
||||
void Orthogonalise(void ) {
|
||||
CoarseScalar InnerProd(_CoarseGrid);
|
||||
std::cout << GridLogMessage <<" Gramm-Schmidt pass 1"<<std::endl;
|
||||
@ -355,8 +326,6 @@ public:
|
||||
}
|
||||
}
|
||||
|
||||
//While this method serves to check the coarse eigenvectors, it also recomputes the eigenvalues from the smoothed reconstructed eigenvectors
|
||||
//hence the smoother can be tuned after running the coarse Lanczos by using a different smoother here
|
||||
void testCoarse(RealD resid,ChebyParams cheby_smooth,RealD relax)
|
||||
{
|
||||
assert(evals_fine.size() == nbasis);
|
||||
@ -405,31 +374,25 @@ public:
|
||||
evals_fine.resize(nbasis);
|
||||
subspace.resize(nbasis,_FineGrid);
|
||||
}
|
||||
|
||||
|
||||
//cheby_op: Parameters of the fine grid Chebyshev polynomial used for the Lanczos acceleration
|
||||
//cheby_smooth: Parameters of a separate Chebyshev polynomial used after the Lanczos has completed to smooth out high frequency noise in the reconstructed fine grid eigenvectors prior to computing the eigenvalue
|
||||
//relax: Reconstructed eigenvectors (post smoothing) are naturally not as precise as true eigenvectors. This factor acts as a multiplier on the stopping condition when determining whether the results satisfy the user provided stopping condition
|
||||
void calcCoarse(ChebyParams cheby_op,ChebyParams cheby_smooth,RealD relax,
|
||||
int Nstop, int Nk, int Nm,RealD resid,
|
||||
RealD MaxIt, RealD betastp, int MinRes)
|
||||
{
|
||||
Chebyshev<FineField> Cheby(cheby_op); //Chebyshev of fine operator on fine grid
|
||||
ProjectedHermOp<Fobj,CComplex,nbasis> Op(_FineOp,subspace); //Fine operator on coarse grid with intermediate fine grid conversion
|
||||
ProjectedFunctionHermOp<Fobj,CComplex,nbasis> ChebyOp (Cheby,_FineOp,subspace); //Chebyshev of fine operator on coarse grid with intermediate fine grid conversion
|
||||
Chebyshev<FineField> Cheby(cheby_op);
|
||||
ProjectedHermOp<Fobj,CComplex,nbasis> Op(_FineOp,subspace);
|
||||
ProjectedFunctionHermOp<Fobj,CComplex,nbasis> ChebyOp (Cheby,_FineOp,subspace);
|
||||
//////////////////////////////////////////////////////////////////////////////////////////////////
|
||||
// create a smoother and see if we can get a cheap convergence test and smooth inside the IRL
|
||||
//////////////////////////////////////////////////////////////////////////////////////////////////
|
||||
|
||||
Chebyshev<FineField> ChebySmooth(cheby_smooth); //lower order Chebyshev of fine operator on fine grid used to smooth regenerated eigenvectors
|
||||
ImplicitlyRestartedLanczosSmoothedTester<Fobj,CComplex,nbasis> ChebySmoothTester(ChebyOp,ChebySmooth,_FineOp,subspace,relax,Nstop-1);
|
||||
Chebyshev<FineField> ChebySmooth(cheby_smooth);
|
||||
ImplicitlyRestartedLanczosSmoothedTester<Fobj,CComplex,nbasis> ChebySmoothTester(ChebyOp,ChebySmooth,_FineOp,subspace,relax);
|
||||
|
||||
evals_coarse.resize(Nm);
|
||||
evec_coarse.resize(Nm,_CoarseGrid);
|
||||
|
||||
CoarseField src(_CoarseGrid); src=1.0;
|
||||
|
||||
//Note the "tester" here is also responsible for generating the fine grid eigenvalues which are output into the "evals_coarse" array
|
||||
ImplicitlyRestartedLanczos<CoarseField> IRL(ChebyOp,ChebyOp,ChebySmoothTester,Nstop,Nk,Nm,resid,MaxIt,betastp,MinRes);
|
||||
int Nconv=0;
|
||||
IRL.calc(evals_coarse,evec_coarse,src,Nconv,false);
|
||||
@ -440,14 +403,6 @@ public:
|
||||
std::cout << i << " Coarse eval = " << evals_coarse[i] << std::endl;
|
||||
}
|
||||
}
|
||||
|
||||
//Get the fine eigenvector 'i' by reconstruction
|
||||
void getFineEvecEval(FineField &evec, RealD &eval, const int i) const{
|
||||
blockPromote(evec_coarse[i],evec,subspace);
|
||||
eval = evals_coarse[i];
|
||||
}
|
||||
|
||||
|
||||
};
|
||||
|
||||
NAMESPACE_END(Grid);
|
||||
|
||||
@ -29,8 +29,6 @@ template<class Field> class PowerMethod
|
||||
RealD vnum = real(innerProduct(src_n,tmp)); // HermOp.
|
||||
RealD vden = norm2(src_n);
|
||||
RealD na = vnum/vden;
|
||||
|
||||
std::cout << GridLogIterative << "PowerMethod: Current approximation of largest eigenvalue " << na << std::endl;
|
||||
|
||||
if ( (fabs(evalMaxApprox/na - 1.0) < 0.001) || (i==_MAX_ITER_EST_-1) ) {
|
||||
evalMaxApprox = na;
|
||||
|
||||
@ -43,7 +43,7 @@ NAMESPACE_BEGIN(Grid);
|
||||
template<class Field>
|
||||
class PrecGeneralisedConjugateResidual : public LinearFunction<Field> {
|
||||
public:
|
||||
using LinearFunction<Field>::operator();
|
||||
|
||||
RealD Tolerance;
|
||||
Integer MaxIterations;
|
||||
int verbose;
|
||||
|
||||
@ -43,7 +43,7 @@ NAMESPACE_BEGIN(Grid);
|
||||
template<class Field>
|
||||
class PrecGeneralisedConjugateResidualNonHermitian : public LinearFunction<Field> {
|
||||
public:
|
||||
using LinearFunction<Field>::operator();
|
||||
|
||||
RealD Tolerance;
|
||||
Integer MaxIterations;
|
||||
int verbose;
|
||||
@ -119,8 +119,7 @@ public:
|
||||
RealD GCRnStep(const Field &src, Field &psi,RealD rsq){
|
||||
|
||||
RealD cp;
|
||||
ComplexD a, b;
|
||||
// ComplexD zAz;
|
||||
ComplexD a, b, zAz;
|
||||
RealD zAAz;
|
||||
ComplexD rq;
|
||||
|
||||
@ -147,7 +146,7 @@ public:
|
||||
//////////////////////////////////
|
||||
MatTimer.Start();
|
||||
Linop.Op(psi,Az);
|
||||
// zAz = innerProduct(Az,psi);
|
||||
zAz = innerProduct(Az,psi);
|
||||
zAAz= norm2(Az);
|
||||
MatTimer.Stop();
|
||||
|
||||
@ -171,7 +170,7 @@ public:
|
||||
|
||||
LinalgTimer.Start();
|
||||
|
||||
// zAz = innerProduct(Az,psi);
|
||||
zAz = innerProduct(Az,psi);
|
||||
zAAz= norm2(Az);
|
||||
|
||||
//p[0],q[0],qq[0]
|
||||
@ -213,7 +212,7 @@ public:
|
||||
MatTimer.Start();
|
||||
Linop.Op(z,Az);
|
||||
MatTimer.Stop();
|
||||
// zAz = innerProduct(Az,psi);
|
||||
zAz = innerProduct(Az,psi);
|
||||
zAAz= norm2(Az);
|
||||
|
||||
LinalgTimer.Start();
|
||||
|
||||
@ -132,31 +132,6 @@ namespace Grid {
|
||||
(*this)(_Matrix,in,out,guess);
|
||||
}
|
||||
|
||||
void RedBlackSource(Matrix &_Matrix, const std::vector<Field> &in, std::vector<Field> &src_o)
|
||||
{
|
||||
GridBase *grid = _Matrix.RedBlackGrid();
|
||||
Field tmp(grid);
|
||||
int nblock = in.size();
|
||||
for(int b=0;b<nblock;b++){
|
||||
RedBlackSource(_Matrix,in[b],tmp,src_o[b]);
|
||||
}
|
||||
}
|
||||
// James can write his own deflated guesser
|
||||
// with optimised code for the inner products
|
||||
// RedBlackSolveSplitGrid();
|
||||
// RedBlackSolve(_Matrix,src_o,sol_o);
|
||||
|
||||
void RedBlackSolution(Matrix &_Matrix, const std::vector<Field> &in, const std::vector<Field> &sol_o, std::vector<Field> &out)
|
||||
{
|
||||
GridBase *grid = _Matrix.RedBlackGrid();
|
||||
Field tmp(grid);
|
||||
int nblock = in.size();
|
||||
for(int b=0;b<nblock;b++) {
|
||||
pickCheckerboard(Even,tmp,in[b]);
|
||||
RedBlackSolution(_Matrix,sol_o[b],tmp,out[b]);
|
||||
}
|
||||
}
|
||||
|
||||
template<class Guesser>
|
||||
void operator()(Matrix &_Matrix, const std::vector<Field> &in, std::vector<Field> &out,Guesser &guess)
|
||||
{
|
||||
@ -175,29 +150,24 @@ namespace Grid {
|
||||
////////////////////////////////////////////////
|
||||
// Prepare RedBlack source
|
||||
////////////////////////////////////////////////
|
||||
RedBlackSource(_Matrix,in,src_o);
|
||||
// for(int b=0;b<nblock;b++){
|
||||
// RedBlackSource(_Matrix,in[b],tmp,src_o[b]);
|
||||
// }
|
||||
|
||||
for(int b=0;b<nblock;b++){
|
||||
RedBlackSource(_Matrix,in[b],tmp,src_o[b]);
|
||||
}
|
||||
////////////////////////////////////////////////
|
||||
// Make the guesses
|
||||
////////////////////////////////////////////////
|
||||
if ( subGuess ) guess_save.resize(nblock,grid);
|
||||
|
||||
|
||||
if(useSolnAsInitGuess) {
|
||||
for(int b=0;b<nblock;b++){
|
||||
for(int b=0;b<nblock;b++){
|
||||
if(useSolnAsInitGuess) {
|
||||
pickCheckerboard(Odd, sol_o[b], out[b]);
|
||||
} else {
|
||||
guess(src_o[b],sol_o[b]);
|
||||
}
|
||||
} else {
|
||||
guess(src_o, sol_o);
|
||||
}
|
||||
|
||||
if ( subGuess ) {
|
||||
for(int b=0;b<nblock;b++){
|
||||
guess_save[b] = sol_o[b];
|
||||
}
|
||||
if ( subGuess ) {
|
||||
guess_save[b] = sol_o[b];
|
||||
}
|
||||
}
|
||||
//////////////////////////////////////////////////////////////
|
||||
// Call the block solver
|
||||
|
||||
@ -9,30 +9,14 @@ NAMESPACE_BEGIN(Grid);
|
||||
#define AccSmall (3)
|
||||
#define Shared (4)
|
||||
#define SharedSmall (5)
|
||||
#undef GRID_MM_VERBOSE
|
||||
uint64_t total_shared;
|
||||
uint64_t total_device;
|
||||
uint64_t total_host;;
|
||||
void MemoryManager::PrintBytes(void)
|
||||
{
|
||||
std::cout << " MemoryManager : ------------------------------------ "<<std::endl;
|
||||
std::cout << " MemoryManager : PrintBytes "<<std::endl;
|
||||
std::cout << " MemoryManager : ------------------------------------ "<<std::endl;
|
||||
std::cout << " MemoryManager : "<<(total_shared>>20)<<" shared Mbytes "<<std::endl;
|
||||
std::cout << " MemoryManager : "<<(total_device>>20)<<" accelerator Mbytes "<<std::endl;
|
||||
std::cout << " MemoryManager : "<<(total_host>>20) <<" cpu Mbytes "<<std::endl;
|
||||
uint64_t cacheBytes;
|
||||
cacheBytes = CacheBytes[Cpu];
|
||||
std::cout << " MemoryManager : "<<(cacheBytes>>20) <<" cpu cache Mbytes "<<std::endl;
|
||||
cacheBytes = CacheBytes[Acc];
|
||||
std::cout << " MemoryManager : "<<(cacheBytes>>20) <<" acc cache Mbytes "<<std::endl;
|
||||
cacheBytes = CacheBytes[Shared];
|
||||
std::cout << " MemoryManager : "<<(cacheBytes>>20) <<" shared cache Mbytes "<<std::endl;
|
||||
|
||||
#ifdef GRID_CUDA
|
||||
cuda_mem();
|
||||
#endif
|
||||
|
||||
std::cout << " MemoryManager : "<<total_shared<<" shared bytes "<<std::endl;
|
||||
std::cout << " MemoryManager : "<<total_device<<" accelerator bytes "<<std::endl;
|
||||
std::cout << " MemoryManager : "<<total_host <<" cpu bytes "<<std::endl;
|
||||
}
|
||||
|
||||
//////////////////////////////////////////////////////////////////////
|
||||
@ -40,114 +24,86 @@ void MemoryManager::PrintBytes(void)
|
||||
//////////////////////////////////////////////////////////////////////
|
||||
MemoryManager::AllocationCacheEntry MemoryManager::Entries[MemoryManager::NallocType][MemoryManager::NallocCacheMax];
|
||||
int MemoryManager::Victim[MemoryManager::NallocType];
|
||||
int MemoryManager::Ncache[MemoryManager::NallocType] = { 2, 8, 2, 8, 2, 8 };
|
||||
uint64_t MemoryManager::CacheBytes[MemoryManager::NallocType];
|
||||
int MemoryManager::Ncache[MemoryManager::NallocType] = { 8, 32, 8, 32, 8, 32 };
|
||||
|
||||
//////////////////////////////////////////////////////////////////////
|
||||
// Actual allocation and deallocation utils
|
||||
//////////////////////////////////////////////////////////////////////
|
||||
void *MemoryManager::AcceleratorAllocate(size_t bytes)
|
||||
{
|
||||
total_device+=bytes;
|
||||
void *ptr = (void *) Lookup(bytes,Acc);
|
||||
if ( ptr == (void *) NULL ) {
|
||||
ptr = (void *) acceleratorAllocDevice(bytes);
|
||||
total_device+=bytes;
|
||||
}
|
||||
#ifdef GRID_MM_VERBOSE
|
||||
std::cout <<"AcceleratorAllocate "<<std::endl;
|
||||
PrintBytes();
|
||||
#endif
|
||||
return ptr;
|
||||
}
|
||||
void MemoryManager::AcceleratorFree (void *ptr,size_t bytes)
|
||||
{
|
||||
total_device-=bytes;
|
||||
void *__freeme = Insert(ptr,bytes,Acc);
|
||||
if ( __freeme ) {
|
||||
acceleratorFreeDevice(__freeme);
|
||||
total_device-=bytes;
|
||||
// PrintBytes();
|
||||
}
|
||||
#ifdef GRID_MM_VERBOSE
|
||||
std::cout <<"AcceleratorFree "<<std::endl;
|
||||
PrintBytes();
|
||||
#endif
|
||||
}
|
||||
void *MemoryManager::SharedAllocate(size_t bytes)
|
||||
{
|
||||
total_shared+=bytes;
|
||||
void *ptr = (void *) Lookup(bytes,Shared);
|
||||
if ( ptr == (void *) NULL ) {
|
||||
ptr = (void *) acceleratorAllocShared(bytes);
|
||||
total_shared+=bytes;
|
||||
// std::cout <<"AcceleratorAllocate: allocated Shared pointer "<<std::hex<<ptr<<std::dec<<std::endl;
|
||||
// PrintBytes();
|
||||
}
|
||||
#ifdef GRID_MM_VERBOSE
|
||||
std::cout <<"SharedAllocate "<<std::endl;
|
||||
PrintBytes();
|
||||
#endif
|
||||
return ptr;
|
||||
}
|
||||
void MemoryManager::SharedFree (void *ptr,size_t bytes)
|
||||
{
|
||||
total_shared-=bytes;
|
||||
void *__freeme = Insert(ptr,bytes,Shared);
|
||||
if ( __freeme ) {
|
||||
acceleratorFreeShared(__freeme);
|
||||
total_shared-=bytes;
|
||||
// PrintBytes();
|
||||
}
|
||||
#ifdef GRID_MM_VERBOSE
|
||||
std::cout <<"SharedFree "<<std::endl;
|
||||
PrintBytes();
|
||||
#endif
|
||||
}
|
||||
#ifdef GRID_UVM
|
||||
void *MemoryManager::CpuAllocate(size_t bytes)
|
||||
{
|
||||
total_host+=bytes;
|
||||
void *ptr = (void *) Lookup(bytes,Cpu);
|
||||
if ( ptr == (void *) NULL ) {
|
||||
ptr = (void *) acceleratorAllocShared(bytes);
|
||||
total_host+=bytes;
|
||||
}
|
||||
#ifdef GRID_MM_VERBOSE
|
||||
std::cout <<"CpuAllocate "<<std::endl;
|
||||
PrintBytes();
|
||||
#endif
|
||||
return ptr;
|
||||
}
|
||||
void MemoryManager::CpuFree (void *_ptr,size_t bytes)
|
||||
{
|
||||
total_host-=bytes;
|
||||
NotifyDeletion(_ptr);
|
||||
void *__freeme = Insert(_ptr,bytes,Cpu);
|
||||
if ( __freeme ) {
|
||||
acceleratorFreeShared(__freeme);
|
||||
total_host-=bytes;
|
||||
}
|
||||
#ifdef GRID_MM_VERBOSE
|
||||
std::cout <<"CpuFree "<<std::endl;
|
||||
PrintBytes();
|
||||
#endif
|
||||
}
|
||||
#else
|
||||
void *MemoryManager::CpuAllocate(size_t bytes)
|
||||
{
|
||||
total_host+=bytes;
|
||||
void *ptr = (void *) Lookup(bytes,Cpu);
|
||||
if ( ptr == (void *) NULL ) {
|
||||
ptr = (void *) acceleratorAllocCpu(bytes);
|
||||
total_host+=bytes;
|
||||
}
|
||||
#ifdef GRID_MM_VERBOSE
|
||||
std::cout <<"CpuAllocate "<<std::endl;
|
||||
PrintBytes();
|
||||
#endif
|
||||
return ptr;
|
||||
}
|
||||
void MemoryManager::CpuFree (void *_ptr,size_t bytes)
|
||||
{
|
||||
total_host-=bytes;
|
||||
NotifyDeletion(_ptr);
|
||||
void *__freeme = Insert(_ptr,bytes,Cpu);
|
||||
if ( __freeme ) {
|
||||
acceleratorFreeCpu(__freeme);
|
||||
total_host-=bytes;
|
||||
}
|
||||
#ifdef GRID_MM_VERBOSE
|
||||
std::cout <<"CpuFree "<<std::endl;
|
||||
PrintBytes();
|
||||
#endif
|
||||
}
|
||||
#endif
|
||||
|
||||
@ -159,6 +115,7 @@ void MemoryManager::Init(void)
|
||||
|
||||
char * str;
|
||||
int Nc;
|
||||
int NcS;
|
||||
|
||||
str= getenv("GRID_ALLOC_NCACHE_LARGE");
|
||||
if ( str ) {
|
||||
@ -224,13 +181,13 @@ void *MemoryManager::Insert(void *ptr,size_t bytes,int type)
|
||||
#ifdef ALLOCATION_CACHE
|
||||
bool small = (bytes < GRID_ALLOC_SMALL_LIMIT);
|
||||
int cache = type + small;
|
||||
return Insert(ptr,bytes,Entries[cache],Ncache[cache],Victim[cache],CacheBytes[cache]);
|
||||
return Insert(ptr,bytes,Entries[cache],Ncache[cache],Victim[cache]);
|
||||
#else
|
||||
return ptr;
|
||||
#endif
|
||||
}
|
||||
|
||||
void *MemoryManager::Insert(void *ptr,size_t bytes,AllocationCacheEntry *entries,int ncache,int &victim, uint64_t &cacheBytes)
|
||||
void *MemoryManager::Insert(void *ptr,size_t bytes,AllocationCacheEntry *entries,int ncache,int &victim)
|
||||
{
|
||||
assert(ncache>0);
|
||||
#ifdef GRID_OMP
|
||||
@ -254,7 +211,6 @@ void *MemoryManager::Insert(void *ptr,size_t bytes,AllocationCacheEntry *entries
|
||||
|
||||
if ( entries[v].valid ) {
|
||||
ret = entries[v].address;
|
||||
cacheBytes -= entries[v].bytes;
|
||||
entries[v].valid = 0;
|
||||
entries[v].address = NULL;
|
||||
entries[v].bytes = 0;
|
||||
@ -263,7 +219,6 @@ void *MemoryManager::Insert(void *ptr,size_t bytes,AllocationCacheEntry *entries
|
||||
entries[v].address=ptr;
|
||||
entries[v].bytes =bytes;
|
||||
entries[v].valid =1;
|
||||
cacheBytes += bytes;
|
||||
|
||||
return ret;
|
||||
}
|
||||
@ -273,13 +228,13 @@ void *MemoryManager::Lookup(size_t bytes,int type)
|
||||
#ifdef ALLOCATION_CACHE
|
||||
bool small = (bytes < GRID_ALLOC_SMALL_LIMIT);
|
||||
int cache = type+small;
|
||||
return Lookup(bytes,Entries[cache],Ncache[cache],CacheBytes[cache]);
|
||||
return Lookup(bytes,Entries[cache],Ncache[cache]);
|
||||
#else
|
||||
return NULL;
|
||||
#endif
|
||||
}
|
||||
|
||||
void *MemoryManager::Lookup(size_t bytes,AllocationCacheEntry *entries,int ncache,uint64_t & cacheBytes)
|
||||
void *MemoryManager::Lookup(size_t bytes,AllocationCacheEntry *entries,int ncache)
|
||||
{
|
||||
assert(ncache>0);
|
||||
#ifdef GRID_OMP
|
||||
@ -288,7 +243,6 @@ void *MemoryManager::Lookup(size_t bytes,AllocationCacheEntry *entries,int ncach
|
||||
for(int e=0;e<ncache;e++){
|
||||
if ( entries[e].valid && ( entries[e].bytes == bytes ) ) {
|
||||
entries[e].valid = 0;
|
||||
cacheBytes -= entries[e].bytes;
|
||||
return entries[e].address;
|
||||
}
|
||||
}
|
||||
|
||||
@ -82,15 +82,14 @@ private:
|
||||
static AllocationCacheEntry Entries[NallocType][NallocCacheMax];
|
||||
static int Victim[NallocType];
|
||||
static int Ncache[NallocType];
|
||||
static uint64_t CacheBytes[NallocType];
|
||||
|
||||
/////////////////////////////////////////////////
|
||||
// Free pool
|
||||
/////////////////////////////////////////////////
|
||||
static void *Insert(void *ptr,size_t bytes,int type) ;
|
||||
static void *Lookup(size_t bytes,int type) ;
|
||||
static void *Insert(void *ptr,size_t bytes,AllocationCacheEntry *entries,int ncache,int &victim,uint64_t &cbytes) ;
|
||||
static void *Lookup(size_t bytes,AllocationCacheEntry *entries,int ncache,uint64_t &cbytes) ;
|
||||
static void *Insert(void *ptr,size_t bytes,AllocationCacheEntry *entries,int ncache,int &victim) ;
|
||||
static void *Lookup(size_t bytes,AllocationCacheEntry *entries,int ncache) ;
|
||||
|
||||
static void PrintBytes(void);
|
||||
public:
|
||||
@ -170,7 +169,6 @@ private:
|
||||
|
||||
public:
|
||||
static void Print(void);
|
||||
static void PrintState( void* CpuPtr);
|
||||
static int isOpen (void* CpuPtr);
|
||||
static void ViewClose(void* CpuPtr,ViewMode mode);
|
||||
static void *ViewOpen (void* CpuPtr,size_t bytes,ViewMode mode,ViewAdvise hint);
|
||||
|
||||
@ -3,7 +3,7 @@
|
||||
|
||||
#warning "Using explicit device memory copies"
|
||||
NAMESPACE_BEGIN(Grid);
|
||||
//#define dprintf(...) printf ( __VA_ARGS__ ); fflush(stdout);
|
||||
//define dprintf(...) printf ( __VA_ARGS__ ); fflush(stdout);
|
||||
#define dprintf(...)
|
||||
|
||||
|
||||
@ -429,7 +429,6 @@ void MemoryManager::NotifyDeletion(void *_ptr)
|
||||
}
|
||||
void MemoryManager::Print(void)
|
||||
{
|
||||
PrintBytes();
|
||||
std::cout << GridLogDebug << "--------------------------------------------" << std::endl;
|
||||
std::cout << GridLogDebug << "Memory Manager " << std::endl;
|
||||
std::cout << GridLogDebug << "--------------------------------------------" << std::endl;
|
||||
@ -474,32 +473,6 @@ int MemoryManager::isOpen (void* _CpuPtr)
|
||||
}
|
||||
}
|
||||
|
||||
void MemoryManager::PrintState(void* _CpuPtr)
|
||||
{
|
||||
uint64_t CpuPtr = (uint64_t)_CpuPtr;
|
||||
|
||||
if ( EntryPresent(CpuPtr) ){
|
||||
auto AccCacheIterator = EntryLookup(CpuPtr);
|
||||
auto & AccCache = AccCacheIterator->second;
|
||||
std::string str;
|
||||
if ( AccCache.state==Empty ) str = std::string("Empty");
|
||||
if ( AccCache.state==CpuDirty ) str = std::string("CpuDirty");
|
||||
if ( AccCache.state==AccDirty ) str = std::string("AccDirty");
|
||||
if ( AccCache.state==Consistent)str = std::string("Consistent");
|
||||
if ( AccCache.state==EvictNext) str = std::string("EvictNext");
|
||||
|
||||
std::cout << GridLogMessage << "CpuAddr\t\tAccAddr\t\tState\t\tcpuLock\taccLock\tLRU_valid "<<std::endl;
|
||||
std::cout << GridLogMessage << "0x"<<std::hex<<AccCache.CpuPtr<<std::dec
|
||||
<< "\t0x"<<std::hex<<AccCache.AccPtr<<std::dec<<"\t" <<str
|
||||
<< "\t" << AccCache.cpuLock
|
||||
<< "\t" << AccCache.accLock
|
||||
<< "\t" << AccCache.LRU_valid<<std::endl;
|
||||
|
||||
} else {
|
||||
std::cout << GridLogMessage << "No Entry in AccCache table." << std::endl;
|
||||
}
|
||||
}
|
||||
|
||||
NAMESPACE_END(Grid);
|
||||
|
||||
#endif
|
||||
|
||||
@ -16,10 +16,6 @@ uint64_t MemoryManager::DeviceToHostXfer;
|
||||
void MemoryManager::ViewClose(void* AccPtr,ViewMode mode){};
|
||||
void *MemoryManager::ViewOpen(void* CpuPtr,size_t bytes,ViewMode mode,ViewAdvise hint){ return CpuPtr; };
|
||||
int MemoryManager::isOpen (void* CpuPtr) { return 0;}
|
||||
void MemoryManager::PrintState(void* CpuPtr)
|
||||
{
|
||||
std::cout << GridLogMessage << "Host<->Device memory movement not currently managed by Grid." << std::endl;
|
||||
};
|
||||
void MemoryManager::Print(void){};
|
||||
void MemoryManager::NotifyDeletion(void *ptr){};
|
||||
|
||||
|
||||
@ -33,8 +33,6 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
|
||||
|
||||
NAMESPACE_BEGIN(Grid);
|
||||
|
||||
bool Stencil_force_mpi = true;
|
||||
|
||||
///////////////////////////////////////////////////////////////
|
||||
// Info that is setup once and indept of cartesian layout
|
||||
///////////////////////////////////////////////////////////////
|
||||
|
||||
@ -35,8 +35,6 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
|
||||
|
||||
NAMESPACE_BEGIN(Grid);
|
||||
|
||||
extern bool Stencil_force_mpi ;
|
||||
|
||||
class CartesianCommunicator : public SharedMemory {
|
||||
|
||||
public:
|
||||
@ -53,11 +51,10 @@ public:
|
||||
// Communicator should know nothing of the physics grid, only processor grid.
|
||||
////////////////////////////////////////////
|
||||
int _Nprocessors; // How many in all
|
||||
int _processor; // linear processor rank
|
||||
unsigned long _ndimension;
|
||||
Coordinate _shm_processors; // Which dimensions get relayed out over processors lanes.
|
||||
Coordinate _processors; // Which dimensions get relayed out over processors lanes.
|
||||
int _processor; // linear processor rank
|
||||
Coordinate _processor_coor; // linear processor coordinate
|
||||
unsigned long _ndimension;
|
||||
static Grid_MPI_Comm communicator_world;
|
||||
Grid_MPI_Comm communicator;
|
||||
std::vector<Grid_MPI_Comm> communicator_halo;
|
||||
@ -98,9 +95,8 @@ public:
|
||||
int BossRank(void) ;
|
||||
int ThisRank(void) ;
|
||||
const Coordinate & ThisProcessorCoor(void) ;
|
||||
const Coordinate & ShmGrid(void) { return _shm_processors; } ;
|
||||
const Coordinate & ProcessorGrid(void) ;
|
||||
int ProcessorCount(void) ;
|
||||
int ProcessorCount(void) ;
|
||||
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
// very VERY rarely (Log, serial RNG) we need world without a grid
|
||||
@ -144,16 +140,16 @@ public:
|
||||
int bytes);
|
||||
|
||||
double StencilSendToRecvFrom(void *xmit,
|
||||
int xmit_to_rank,int do_xmit,
|
||||
int xmit_to_rank,
|
||||
void *recv,
|
||||
int recv_from_rank,int do_recv,
|
||||
int recv_from_rank,
|
||||
int bytes,int dir);
|
||||
|
||||
double StencilSendToRecvFromBegin(std::vector<CommsRequest_t> &list,
|
||||
void *xmit,
|
||||
int xmit_to_rank,int do_xmit,
|
||||
int xmit_to_rank,
|
||||
void *recv,
|
||||
int recv_from_rank,int do_recv,
|
||||
int recv_from_rank,
|
||||
int bytes,int dir);
|
||||
|
||||
|
||||
|
||||
@ -106,7 +106,7 @@ CartesianCommunicator::CartesianCommunicator(const Coordinate &processors)
|
||||
// Remap using the shared memory optimising routine
|
||||
// The remap creates a comm which must be freed
|
||||
////////////////////////////////////////////////////
|
||||
GlobalSharedMemory::OptimalCommunicator (processors,optimal_comm,_shm_processors);
|
||||
GlobalSharedMemory::OptimalCommunicator (processors,optimal_comm);
|
||||
InitFromMPICommunicator(processors,optimal_comm);
|
||||
SetCommunicator(optimal_comm);
|
||||
///////////////////////////////////////////////////
|
||||
@ -124,13 +124,12 @@ CartesianCommunicator::CartesianCommunicator(const Coordinate &processors,const
|
||||
int parent_ndimension = parent._ndimension; assert(_ndimension >= parent._ndimension);
|
||||
Coordinate parent_processor_coor(_ndimension,0);
|
||||
Coordinate parent_processors (_ndimension,1);
|
||||
Coordinate shm_processors (_ndimension,1);
|
||||
|
||||
// Can make 5d grid from 4d etc...
|
||||
int pad = _ndimension-parent_ndimension;
|
||||
for(int d=0;d<parent_ndimension;d++){
|
||||
parent_processor_coor[pad+d]=parent._processor_coor[d];
|
||||
parent_processors [pad+d]=parent._processors[d];
|
||||
shm_processors [pad+d]=parent._shm_processors[d];
|
||||
}
|
||||
|
||||
//////////////////////////////////////////////////////////////////////////////////////////////////////
|
||||
@ -155,7 +154,6 @@ CartesianCommunicator::CartesianCommunicator(const Coordinate &processors,const
|
||||
ccoor[d] = parent_processor_coor[d] % processors[d];
|
||||
scoor[d] = parent_processor_coor[d] / processors[d];
|
||||
ssize[d] = parent_processors[d] / processors[d];
|
||||
if ( processors[d] < shm_processors[d] ) shm_processors[d] = processors[d]; // subnode splitting.
|
||||
}
|
||||
|
||||
// rank within subcomm ; srank is rank of subcomm within blocks of subcomms
|
||||
@ -337,22 +335,22 @@ void CartesianCommunicator::SendToRecvFrom(void *xmit,
|
||||
}
|
||||
// Basic Halo comms primitive
|
||||
double CartesianCommunicator::StencilSendToRecvFrom( void *xmit,
|
||||
int dest, int dox,
|
||||
int dest,
|
||||
void *recv,
|
||||
int from, int dor,
|
||||
int from,
|
||||
int bytes,int dir)
|
||||
{
|
||||
std::vector<CommsRequest_t> list;
|
||||
double offbytes = StencilSendToRecvFromBegin(list,xmit,dest,dox,recv,from,dor,bytes,dir);
|
||||
double offbytes = StencilSendToRecvFromBegin(list,xmit,dest,recv,from,bytes,dir);
|
||||
StencilSendToRecvFromComplete(list,dir);
|
||||
return offbytes;
|
||||
}
|
||||
|
||||
double CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsRequest_t> &list,
|
||||
void *xmit,
|
||||
int dest,int dox,
|
||||
int dest,
|
||||
void *recv,
|
||||
int from,int dor,
|
||||
int from,
|
||||
int bytes,int dir)
|
||||
{
|
||||
int ncomm =communicator_halo.size();
|
||||
@ -372,42 +370,30 @@ double CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsReques
|
||||
double off_node_bytes=0.0;
|
||||
int tag;
|
||||
|
||||
if ( dor ) {
|
||||
if ( (gfrom ==MPI_UNDEFINED) || Stencil_force_mpi ) {
|
||||
tag= dir+from*32;
|
||||
ierr=MPI_Irecv(recv, bytes, MPI_CHAR,from,tag,communicator_halo[commdir],&rrq);
|
||||
assert(ierr==0);
|
||||
list.push_back(rrq);
|
||||
off_node_bytes+=bytes;
|
||||
}
|
||||
if ( gfrom ==MPI_UNDEFINED) {
|
||||
tag= dir+from*32;
|
||||
ierr=MPI_Irecv(recv, bytes, MPI_CHAR,from,tag,communicator_halo[commdir],&rrq);
|
||||
assert(ierr==0);
|
||||
list.push_back(rrq);
|
||||
off_node_bytes+=bytes;
|
||||
}
|
||||
|
||||
if (dox) {
|
||||
if ( (gdest == MPI_UNDEFINED) || Stencil_force_mpi ) {
|
||||
tag= dir+_processor*32;
|
||||
ierr =MPI_Isend(xmit, bytes, MPI_CHAR,dest,tag,communicator_halo[commdir],&xrq);
|
||||
assert(ierr==0);
|
||||
list.push_back(xrq);
|
||||
off_node_bytes+=bytes;
|
||||
} else {
|
||||
void *shm = (void *) this->ShmBufferTranslate(dest,recv);
|
||||
assert(shm!=NULL);
|
||||
acceleratorCopyDeviceToDeviceAsynch(xmit,shm,bytes);
|
||||
}
|
||||
|
||||
if ( gdest == MPI_UNDEFINED ) {
|
||||
tag= dir+_processor*32;
|
||||
ierr =MPI_Isend(xmit, bytes, MPI_CHAR,dest,tag,communicator_halo[commdir],&xrq);
|
||||
assert(ierr==0);
|
||||
list.push_back(xrq);
|
||||
off_node_bytes+=bytes;
|
||||
}
|
||||
|
||||
|
||||
if ( CommunicatorPolicy == CommunicatorPolicySequential ) {
|
||||
this->StencilSendToRecvFromComplete(list,dir);
|
||||
list.resize(0);
|
||||
}
|
||||
|
||||
return off_node_bytes;
|
||||
}
|
||||
void CartesianCommunicator::StencilSendToRecvFromComplete(std::vector<CommsRequest_t> &list,int dir)
|
||||
{
|
||||
// std::cout << "Copy Synchronised\n"<<std::endl;
|
||||
acceleratorCopySynchronise();
|
||||
|
||||
int nreq=list.size();
|
||||
|
||||
if (nreq==0) return;
|
||||
|
||||
@ -45,14 +45,12 @@ void CartesianCommunicator::Init(int *argc, char *** arv)
|
||||
CartesianCommunicator::CartesianCommunicator(const Coordinate &processors,const CartesianCommunicator &parent,int &srank)
|
||||
: CartesianCommunicator(processors)
|
||||
{
|
||||
_shm_processors = Coordinate(processors.size(),1);
|
||||
srank=0;
|
||||
SetCommunicator(communicator_world);
|
||||
}
|
||||
|
||||
CartesianCommunicator::CartesianCommunicator(const Coordinate &processors)
|
||||
{
|
||||
_shm_processors = Coordinate(processors.size(),1);
|
||||
_processors = processors;
|
||||
_ndimension = processors.size(); assert(_ndimension>=1);
|
||||
_processor_coor.resize(_ndimension);
|
||||
@ -113,18 +111,18 @@ void CartesianCommunicator::ShiftedRanks(int dim,int shift,int &source,int &dest
|
||||
}
|
||||
|
||||
double CartesianCommunicator::StencilSendToRecvFrom( void *xmit,
|
||||
int xmit_to_rank,int dox,
|
||||
int xmit_to_rank,
|
||||
void *recv,
|
||||
int recv_from_rank,int dor,
|
||||
int recv_from_rank,
|
||||
int bytes, int dir)
|
||||
{
|
||||
return 2.0*bytes;
|
||||
}
|
||||
double CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsRequest_t> &list,
|
||||
void *xmit,
|
||||
int xmit_to_rank,int dox,
|
||||
int xmit_to_rank,
|
||||
void *recv,
|
||||
int recv_from_rank,int dor,
|
||||
int recv_from_rank,
|
||||
int bytes, int dir)
|
||||
{
|
||||
return 2.0*bytes;
|
||||
|
||||
@ -93,10 +93,9 @@ public:
|
||||
// Create an optimal reordered communicator that makes MPI_Cart_create get it right
|
||||
//////////////////////////////////////////////////////////////////////////////////////
|
||||
static void Init(Grid_MPI_Comm comm); // Typically MPI_COMM_WORLD
|
||||
// Turns MPI_COMM_WORLD into right layout for Cartesian
|
||||
static void OptimalCommunicator (const Coordinate &processors,Grid_MPI_Comm & optimal_comm,Coordinate &ShmDims);
|
||||
static void OptimalCommunicatorHypercube (const Coordinate &processors,Grid_MPI_Comm & optimal_comm,Coordinate &ShmDims);
|
||||
static void OptimalCommunicatorSharedMemory(const Coordinate &processors,Grid_MPI_Comm & optimal_comm,Coordinate &ShmDims);
|
||||
static void OptimalCommunicator (const Coordinate &processors,Grid_MPI_Comm & optimal_comm); // Turns MPI_COMM_WORLD into right layout for Cartesian
|
||||
static void OptimalCommunicatorHypercube (const Coordinate &processors,Grid_MPI_Comm & optimal_comm); // Turns MPI_COMM_WORLD into right layout for Cartesian
|
||||
static void OptimalCommunicatorSharedMemory(const Coordinate &processors,Grid_MPI_Comm & optimal_comm); // Turns MPI_COMM_WORLD into right layout for Cartesian
|
||||
static void GetShmDims(const Coordinate &WorldDims,Coordinate &ShmDims);
|
||||
///////////////////////////////////////////////////
|
||||
// Provide shared memory facilities off comm world
|
||||
|
||||
@ -35,9 +35,6 @@ Author: Christoph Lehner <christoph@lhnr.de>
|
||||
#endif
|
||||
#ifdef GRID_HIP
|
||||
#include <hip/hip_runtime_api.h>
|
||||
#endif
|
||||
#ifdef GRID_SYCl
|
||||
|
||||
#endif
|
||||
|
||||
NAMESPACE_BEGIN(Grid);
|
||||
@ -73,7 +70,6 @@ void GlobalSharedMemory::Init(Grid_MPI_Comm comm)
|
||||
WorldNodes = WorldSize/WorldShmSize;
|
||||
assert( (WorldNodes * WorldShmSize) == WorldSize );
|
||||
|
||||
|
||||
// FIXME: Check all WorldShmSize are the same ?
|
||||
|
||||
/////////////////////////////////////////////////////////////////////
|
||||
@ -152,7 +148,7 @@ int Log2Size(int TwoToPower,int MAXLOG2)
|
||||
}
|
||||
return log2size;
|
||||
}
|
||||
void GlobalSharedMemory::OptimalCommunicator(const Coordinate &processors,Grid_MPI_Comm & optimal_comm,Coordinate &SHM)
|
||||
void GlobalSharedMemory::OptimalCommunicator(const Coordinate &processors,Grid_MPI_Comm & optimal_comm)
|
||||
{
|
||||
//////////////////////////////////////////////////////////////////////////////
|
||||
// Look and see if it looks like an HPE 8600 based on hostname conventions
|
||||
@ -165,8 +161,8 @@ void GlobalSharedMemory::OptimalCommunicator(const Coordinate &processors,Grid_M
|
||||
gethostname(name,namelen);
|
||||
int nscan = sscanf(name,"r%di%dn%d",&R,&I,&N) ;
|
||||
|
||||
if(nscan==3 && HPEhypercube ) OptimalCommunicatorHypercube(processors,optimal_comm,SHM);
|
||||
else OptimalCommunicatorSharedMemory(processors,optimal_comm,SHM);
|
||||
if(nscan==3 && HPEhypercube ) OptimalCommunicatorHypercube(processors,optimal_comm);
|
||||
else OptimalCommunicatorSharedMemory(processors,optimal_comm);
|
||||
}
|
||||
static inline int divides(int a,int b)
|
||||
{
|
||||
@ -221,7 +217,7 @@ void GlobalSharedMemory::GetShmDims(const Coordinate &WorldDims,Coordinate &ShmD
|
||||
dim=(dim+1) %ndimension;
|
||||
}
|
||||
}
|
||||
void GlobalSharedMemory::OptimalCommunicatorHypercube(const Coordinate &processors,Grid_MPI_Comm & optimal_comm,Coordinate &SHM)
|
||||
void GlobalSharedMemory::OptimalCommunicatorHypercube(const Coordinate &processors,Grid_MPI_Comm & optimal_comm)
|
||||
{
|
||||
////////////////////////////////////////////////////////////////
|
||||
// Assert power of two shm_size.
|
||||
@ -294,8 +290,7 @@ void GlobalSharedMemory::OptimalCommunicatorHypercube(const Coordinate &processo
|
||||
Coordinate HyperCoor(ndimension);
|
||||
|
||||
GetShmDims(WorldDims,ShmDims);
|
||||
SHM = ShmDims;
|
||||
|
||||
|
||||
////////////////////////////////////////////////////////////////
|
||||
// Establish torus of processes and nodes with sub-blockings
|
||||
////////////////////////////////////////////////////////////////
|
||||
@ -342,7 +337,7 @@ void GlobalSharedMemory::OptimalCommunicatorHypercube(const Coordinate &processo
|
||||
int ierr= MPI_Comm_split(WorldComm,0,rank,&optimal_comm);
|
||||
assert(ierr==0);
|
||||
}
|
||||
void GlobalSharedMemory::OptimalCommunicatorSharedMemory(const Coordinate &processors,Grid_MPI_Comm & optimal_comm,Coordinate &SHM)
|
||||
void GlobalSharedMemory::OptimalCommunicatorSharedMemory(const Coordinate &processors,Grid_MPI_Comm & optimal_comm)
|
||||
{
|
||||
////////////////////////////////////////////////////////////////
|
||||
// Identify subblock of ranks on node spreading across dims
|
||||
@ -354,8 +349,6 @@ void GlobalSharedMemory::OptimalCommunicatorSharedMemory(const Coordinate &proce
|
||||
Coordinate ShmCoor(ndimension); Coordinate NodeCoor(ndimension); Coordinate WorldCoor(ndimension);
|
||||
|
||||
GetShmDims(WorldDims,ShmDims);
|
||||
SHM=ShmDims;
|
||||
|
||||
////////////////////////////////////////////////////////////////
|
||||
// Establish torus of processes and nodes with sub-blockings
|
||||
////////////////////////////////////////////////////////////////
|
||||
@ -453,47 +446,7 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
|
||||
////////////////////////////////////////////////////////////////////////////////////////////
|
||||
// Hugetlbfs mapping intended
|
||||
////////////////////////////////////////////////////////////////////////////////////////////
|
||||
#if defined(GRID_CUDA) ||defined(GRID_HIP) || defined(GRID_SYCL)
|
||||
|
||||
//if defined(GRID_SYCL)
|
||||
#if 0
|
||||
void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
|
||||
{
|
||||
void * ShmCommBuf ;
|
||||
assert(_ShmSetup==1);
|
||||
assert(_ShmAlloc==0);
|
||||
|
||||
//////////////////////////////////////////////////////////////////////////////////////////////////////////
|
||||
// allocate the pointer array for shared windows for our group
|
||||
//////////////////////////////////////////////////////////////////////////////////////////////////////////
|
||||
MPI_Barrier(WorldShmComm);
|
||||
WorldShmCommBufs.resize(WorldShmSize);
|
||||
|
||||
///////////////////////////////////////////////////////////////////////////////////////////////////////////
|
||||
// Each MPI rank should allocate our own buffer
|
||||
///////////////////////////////////////////////////////////////////////////////////////////////////////////
|
||||
ShmCommBuf = acceleratorAllocDevice(bytes);
|
||||
|
||||
if (ShmCommBuf == (void *)NULL ) {
|
||||
std::cerr << " SharedMemoryMPI.cc acceleratorAllocDevice failed NULL pointer for " << bytes<<" bytes " << std::endl;
|
||||
exit(EXIT_FAILURE);
|
||||
}
|
||||
|
||||
std::cout << WorldRank << header " SharedMemoryMPI.cc acceleratorAllocDevice "<< bytes
|
||||
<< "bytes at "<< std::hex<< ShmCommBuf <<std::dec<<" for comms buffers " <<std::endl;
|
||||
|
||||
SharedMemoryZero(ShmCommBuf,bytes);
|
||||
|
||||
assert(WorldShmSize == 1);
|
||||
for(int r=0;r<WorldShmSize;r++){
|
||||
WorldShmCommBufs[r] = ShmCommBuf;
|
||||
}
|
||||
_ShmAllocBytes=bytes;
|
||||
_ShmAlloc=1;
|
||||
}
|
||||
#endif
|
||||
|
||||
#if defined(GRID_CUDA) ||defined(GRID_HIP) ||defined(GRID_SYCL)
|
||||
#if defined(GRID_CUDA) ||defined(GRID_HIP)
|
||||
void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
|
||||
{
|
||||
void * ShmCommBuf ;
|
||||
@ -517,16 +470,18 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
|
||||
// Each MPI rank should allocate our own buffer
|
||||
///////////////////////////////////////////////////////////////////////////////////////////////////////////
|
||||
ShmCommBuf = acceleratorAllocDevice(bytes);
|
||||
|
||||
if (ShmCommBuf == (void *)NULL ) {
|
||||
std::cerr << " SharedMemoryMPI.cc acceleratorAllocDevice failed NULL pointer for " << bytes<<" bytes " << std::endl;
|
||||
exit(EXIT_FAILURE);
|
||||
}
|
||||
if ( WorldRank == 0 ){
|
||||
// if ( WorldRank == 0 ){
|
||||
if ( 1 ){
|
||||
std::cout << WorldRank << header " SharedMemoryMPI.cc acceleratorAllocDevice "<< bytes
|
||||
<< "bytes at "<< std::hex<< ShmCommBuf <<std::dec<<" for comms buffers " <<std::endl;
|
||||
}
|
||||
SharedMemoryZero(ShmCommBuf,bytes);
|
||||
std::cout<< "Setting up IPC"<<std::endl;
|
||||
|
||||
///////////////////////////////////////////////////////////////////////////////////////////////////////////
|
||||
// Loop over ranks/gpu's on our node
|
||||
///////////////////////////////////////////////////////////////////////////////////////////////////////////
|
||||
@ -536,29 +491,6 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
|
||||
//////////////////////////////////////////////////
|
||||
// If it is me, pass around the IPC access key
|
||||
//////////////////////////////////////////////////
|
||||
void * thisBuf = ShmCommBuf;
|
||||
if(!Stencil_force_mpi) {
|
||||
#ifdef GRID_SYCL_LEVEL_ZERO_IPC
|
||||
typedef struct { int fd; pid_t pid ; } clone_mem_t;
|
||||
|
||||
auto zeDevice = cl::sycl::get_native<cl::sycl::backend::level_zero>(theGridAccelerator->get_device());
|
||||
auto zeContext = cl::sycl::get_native<cl::sycl::backend::level_zero>(theGridAccelerator->get_context());
|
||||
|
||||
ze_ipc_mem_handle_t ihandle;
|
||||
clone_mem_t handle;
|
||||
|
||||
if ( r==WorldShmRank ) {
|
||||
auto err = zeMemGetIpcHandle(zeContext,ShmCommBuf,&ihandle);
|
||||
if ( err != ZE_RESULT_SUCCESS ) {
|
||||
std::cout << "SharedMemoryMPI.cc zeMemGetIpcHandle failed for rank "<<r<<" "<<std::hex<<err<<std::dec<<std::endl;
|
||||
exit(EXIT_FAILURE);
|
||||
} else {
|
||||
std::cout << "SharedMemoryMPI.cc zeMemGetIpcHandle succeeded for rank "<<r<<" "<<std::hex<<err<<std::dec<<std::endl;
|
||||
}
|
||||
memcpy((void *)&handle.fd,(void *)&ihandle,sizeof(int));
|
||||
handle.pid = getpid();
|
||||
}
|
||||
#endif
|
||||
#ifdef GRID_CUDA
|
||||
cudaIpcMemHandle_t handle;
|
||||
if ( r==WorldShmRank ) {
|
||||
@ -579,7 +511,6 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
|
||||
}
|
||||
}
|
||||
#endif
|
||||
|
||||
//////////////////////////////////////////////////
|
||||
// Share this IPC handle across the Shm Comm
|
||||
//////////////////////////////////////////////////
|
||||
@ -595,35 +526,7 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
|
||||
///////////////////////////////////////////////////////////////
|
||||
// If I am not the source, overwrite thisBuf with remote buffer
|
||||
///////////////////////////////////////////////////////////////
|
||||
|
||||
#ifdef GRID_SYCL_LEVEL_ZERO_IPC
|
||||
if ( r!=WorldShmRank ) {
|
||||
thisBuf = nullptr;
|
||||
std::cout<<"mapping seeking remote pid/fd "
|
||||
<<handle.pid<<"/"
|
||||
<<handle.fd<<std::endl;
|
||||
|
||||
int pidfd = syscall(SYS_pidfd_open,handle.pid,0);
|
||||
std::cout<<"Using IpcHandle pidfd "<<pidfd<<"\n";
|
||||
// int myfd = syscall(SYS_pidfd_getfd,pidfd,handle.fd,0);
|
||||
int myfd = syscall(438,pidfd,handle.fd,0);
|
||||
|
||||
std::cout<<"Using IpcHandle myfd "<<myfd<<"\n";
|
||||
|
||||
memcpy((void *)&ihandle,(void *)&myfd,sizeof(int));
|
||||
|
||||
auto err = zeMemOpenIpcHandle(zeContext,zeDevice,ihandle,0,&thisBuf);
|
||||
if ( err != ZE_RESULT_SUCCESS ) {
|
||||
std::cout << "SharedMemoryMPI.cc "<<zeContext<<" "<<zeDevice<<std::endl;
|
||||
std::cout << "SharedMemoryMPI.cc zeMemOpenIpcHandle failed for rank "<<r<<" "<<std::hex<<err<<std::dec<<std::endl;
|
||||
exit(EXIT_FAILURE);
|
||||
} else {
|
||||
std::cout << "SharedMemoryMPI.cc zeMemOpenIpcHandle succeeded for rank "<<r<<std::endl;
|
||||
std::cout << "SharedMemoryMPI.cc zeMemOpenIpcHandle pointer is "<<std::hex<<thisBuf<<std::dec<<std::endl;
|
||||
}
|
||||
assert(thisBuf!=nullptr);
|
||||
}
|
||||
#endif
|
||||
void * thisBuf = ShmCommBuf;
|
||||
#ifdef GRID_CUDA
|
||||
if ( r!=WorldShmRank ) {
|
||||
auto err = cudaIpcOpenMemHandle(&thisBuf,handle,cudaIpcMemLazyEnablePeerAccess);
|
||||
@ -645,7 +548,6 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
|
||||
///////////////////////////////////////////////////////////////
|
||||
// Save a copy of the device buffers
|
||||
///////////////////////////////////////////////////////////////
|
||||
}
|
||||
WorldShmCommBufs[r] = thisBuf;
|
||||
#else
|
||||
WorldShmCommBufs[r] = ShmCommBuf;
|
||||
@ -655,8 +557,6 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
|
||||
_ShmAllocBytes=bytes;
|
||||
_ShmAlloc=1;
|
||||
}
|
||||
#endif
|
||||
|
||||
#else
|
||||
#ifdef GRID_MPI3_SHMMMAP
|
||||
void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
|
||||
@ -827,16 +727,16 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
|
||||
/////////////////////////////////////////////////////////////////////////
|
||||
void GlobalSharedMemory::SharedMemoryZero(void *dest,size_t bytes)
|
||||
{
|
||||
#if defined(GRID_CUDA) || defined(GRID_HIP) || defined(GRID_SYCL)
|
||||
acceleratorMemSet(dest,0,bytes);
|
||||
#ifdef GRID_CUDA
|
||||
cudaMemset(dest,0,bytes);
|
||||
#else
|
||||
bzero(dest,bytes);
|
||||
#endif
|
||||
}
|
||||
void GlobalSharedMemory::SharedMemoryCopy(void *dest,void *src,size_t bytes)
|
||||
{
|
||||
#if defined(GRID_CUDA) || defined(GRID_HIP) || defined(GRID_SYCL)
|
||||
acceleratorCopyToDevice(src,dest,bytes);
|
||||
#ifdef GRID_CUDA
|
||||
cudaMemcpy(dest,src,bytes,cudaMemcpyDefault);
|
||||
#else
|
||||
bcopy(src,dest,bytes);
|
||||
#endif
|
||||
@ -900,7 +800,7 @@ void SharedMemory::SetCommunicator(Grid_MPI_Comm comm)
|
||||
}
|
||||
#endif
|
||||
|
||||
//SharedMemoryTest();
|
||||
SharedMemoryTest();
|
||||
}
|
||||
//////////////////////////////////////////////////////////////////
|
||||
// On node barrier
|
||||
|
||||
@ -48,10 +48,9 @@ void GlobalSharedMemory::Init(Grid_MPI_Comm comm)
|
||||
_ShmSetup=1;
|
||||
}
|
||||
|
||||
void GlobalSharedMemory::OptimalCommunicator(const Coordinate &processors,Grid_MPI_Comm & optimal_comm,Coordinate &SHM)
|
||||
void GlobalSharedMemory::OptimalCommunicator(const Coordinate &processors,Grid_MPI_Comm & optimal_comm)
|
||||
{
|
||||
optimal_comm = WorldComm;
|
||||
SHM = Coordinate(processors.size(),1);
|
||||
}
|
||||
|
||||
////////////////////////////////////////////////////////////////////////////////////////////
|
||||
|
||||
@ -46,4 +46,3 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
|
||||
#include <Grid/lattice/Lattice_unary.h>
|
||||
#include <Grid/lattice/Lattice_transfer.h>
|
||||
#include <Grid/lattice/Lattice_basis.h>
|
||||
#include <Grid/lattice/Lattice_crc.h>
|
||||
|
||||
@ -225,7 +225,7 @@ void axpy(Lattice<vobj> &ret,sobj a,const Lattice<vobj> &x,const Lattice<vobj> &
|
||||
autoView( x_v , x, AcceleratorRead);
|
||||
autoView( y_v , y, AcceleratorRead);
|
||||
accelerator_for(ss,x_v.size(),vobj::Nsimd(),{
|
||||
auto tmp = a*coalescedRead(x_v[ss])+coalescedRead(y_v[ss]);
|
||||
auto tmp = a*x_v(ss)+y_v(ss);
|
||||
coalescedWrite(ret_v[ss],tmp);
|
||||
});
|
||||
}
|
||||
|
||||
@ -88,13 +88,6 @@ public:
|
||||
LatticeView<vobj> accessor(*( (LatticeAccelerator<vobj> *) this),mode);
|
||||
accessor.ViewClose();
|
||||
}
|
||||
|
||||
// Helper function to print the state of this object in the AccCache
|
||||
void PrintCacheState(void)
|
||||
{
|
||||
MemoryManager::PrintState(this->_odata);
|
||||
}
|
||||
|
||||
/////////////////////////////////////////////////////////////////////////////////
|
||||
// Return a view object that may be dereferenced in site loops.
|
||||
// The view is trivially copy constructible and may be copied to an accelerator device
|
||||
|
||||
@ -125,7 +125,7 @@ void basisRotate(VField &basis,Matrix& Qt,int j0, int j1, int k0,int k1,int Nm)
|
||||
|
||||
for(int k=k0; k<k1; ++k){
|
||||
auto tmp = coalescedRead(Bp[ss*nrot+j]);
|
||||
coalescedWrite(Bp[ss*nrot+j],tmp+ Qt_p[jj*Nm+k] * coalescedRead(basis_vp[k][sss]));
|
||||
coalescedWrite(Bp[ss*nrot+j],tmp+ Qt_p[jj*Nm+k] * coalescedRead(basis_v[k][sss]));
|
||||
}
|
||||
});
|
||||
|
||||
@ -134,7 +134,7 @@ void basisRotate(VField &basis,Matrix& Qt,int j0, int j1, int k0,int k1,int Nm)
|
||||
int jj =j0+j;
|
||||
int ss =sj/nrot;
|
||||
int sss=ss+s;
|
||||
coalescedWrite(basis_vp[jj][sss],coalescedRead(Bp[ss*nrot+j]));
|
||||
coalescedWrite(basis_v[jj][sss],coalescedRead(Bp[ss*nrot+j]));
|
||||
});
|
||||
}
|
||||
#endif
|
||||
|
||||
@ -1,55 +0,0 @@
|
||||
/*************************************************************************************
|
||||
|
||||
Grid physics library, www.github.com/paboyle/Grid
|
||||
|
||||
Source file: ./lib/lattice/Lattice_crc.h
|
||||
|
||||
Copyright (C) 2021
|
||||
|
||||
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
|
||||
|
||||
This program is free software; you can redistribute it and/or modify
|
||||
it under the terms of the GNU General Public License as published by
|
||||
the Free Software Foundation; either version 2 of the License, or
|
||||
(at your option) any later version.
|
||||
|
||||
This program is distributed in the hope that it will be useful,
|
||||
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||||
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
||||
GNU General Public License for more details.
|
||||
|
||||
You should have received a copy of the GNU General Public License along
|
||||
with this program; if not, write to the Free Software Foundation, Inc.,
|
||||
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
|
||||
|
||||
See the full license in the file "LICENSE" in the top level distribution directory
|
||||
*************************************************************************************/
|
||||
/* END LEGAL */
|
||||
#pragma once
|
||||
|
||||
NAMESPACE_BEGIN(Grid);
|
||||
|
||||
template<class vobj> void DumpSliceNorm(std::string s,Lattice<vobj> &f,int mu=-1)
|
||||
{
|
||||
auto ff = localNorm2(f);
|
||||
if ( mu==-1 ) mu = f.Grid()->Nd()-1;
|
||||
typedef typename vobj::tensor_reduced normtype;
|
||||
typedef typename normtype::scalar_object scalar;
|
||||
std::vector<scalar> sff;
|
||||
sliceSum(ff,sff,mu);
|
||||
for(int t=0;t<sff.size();t++){
|
||||
std::cout << s<<" "<<t<<" "<<sff[t]<<std::endl;
|
||||
}
|
||||
}
|
||||
|
||||
template<class vobj> uint32_t crc(Lattice<vobj> & buf)
|
||||
{
|
||||
autoView( buf_v , buf, CpuRead);
|
||||
return ::crc32(0L,(unsigned char *)&buf_v[0],(size_t)sizeof(vobj)*buf.oSites());
|
||||
}
|
||||
|
||||
#define CRC(U) std::cout << "FingerPrint "<<__FILE__ <<" "<< __LINE__ <<" "<< #U <<" "<<crc(U)<<std::endl;
|
||||
|
||||
NAMESPACE_END(Grid);
|
||||
|
||||
|
||||
@ -125,12 +125,6 @@ void pokeSite(const sobj &s,Lattice<vobj> &l,const Coordinate &site){
|
||||
//////////////////////////////////////////////////////////
|
||||
// Peek a scalar object from the SIMD array
|
||||
//////////////////////////////////////////////////////////
|
||||
template<class vobj>
|
||||
typename vobj::scalar_object peekSite(const Lattice<vobj> &l,const Coordinate &site){
|
||||
typename vobj::scalar_object s;
|
||||
peekSite(s,l,site);
|
||||
return s;
|
||||
}
|
||||
template<class vobj,class sobj>
|
||||
void peekSite(sobj &s,const Lattice<vobj> &l,const Coordinate &site){
|
||||
|
||||
|
||||
@ -142,15 +142,6 @@ inline typename vobj::scalar_objectD sumD(const vobj *arg, Integer osites)
|
||||
return sumD_cpu(arg,osites);
|
||||
#endif
|
||||
}
|
||||
template<class vobj>
|
||||
inline typename vobj::scalar_objectD sumD_large(const vobj *arg, Integer osites)
|
||||
{
|
||||
#if defined(GRID_CUDA)||defined(GRID_HIP)
|
||||
return sumD_gpu_large(arg,osites);
|
||||
#else
|
||||
return sumD_cpu(arg,osites);
|
||||
#endif
|
||||
}
|
||||
|
||||
template<class vobj>
|
||||
inline typename vobj::scalar_object sum(const Lattice<vobj> &arg)
|
||||
@ -168,22 +159,6 @@ inline typename vobj::scalar_object sum(const Lattice<vobj> &arg)
|
||||
return ssum;
|
||||
}
|
||||
|
||||
template<class vobj>
|
||||
inline typename vobj::scalar_object sum_large(const Lattice<vobj> &arg)
|
||||
{
|
||||
#if defined(GRID_CUDA)||defined(GRID_HIP)
|
||||
autoView( arg_v, arg, AcceleratorRead);
|
||||
Integer osites = arg.Grid()->oSites();
|
||||
auto ssum= sum_gpu_large(&arg_v[0],osites);
|
||||
#else
|
||||
autoView(arg_v, arg, CpuRead);
|
||||
Integer osites = arg.Grid()->oSites();
|
||||
auto ssum= sum_cpu(&arg_v[0],osites);
|
||||
#endif
|
||||
arg.Grid()->GlobalSum(ssum);
|
||||
return ssum;
|
||||
}
|
||||
|
||||
////////////////////////////////////////////////////////////////////////////////////////////////////
|
||||
// Deterministic Reduction operations
|
||||
////////////////////////////////////////////////////////////////////////////////////////////////////
|
||||
@ -232,7 +207,6 @@ inline ComplexD rankInnerProduct(const Lattice<vobj> &left,const Lattice<vobj> &
|
||||
const uint64_t sites = grid->oSites();
|
||||
|
||||
// Might make all code paths go this way.
|
||||
#if 0
|
||||
typedef decltype(innerProductD(vobj(),vobj())) inner_t;
|
||||
Vector<inner_t> inner_tmp(sites);
|
||||
auto inner_tmp_v = &inner_tmp[0];
|
||||
@ -242,31 +216,15 @@ inline ComplexD rankInnerProduct(const Lattice<vobj> &left,const Lattice<vobj> &
|
||||
autoView( right_v,right, AcceleratorRead);
|
||||
|
||||
// GPU - SIMT lane compliance...
|
||||
accelerator_for( ss, sites, nsimd,{
|
||||
auto x_l = left_v(ss);
|
||||
auto y_l = right_v(ss);
|
||||
coalescedWrite(inner_tmp_v[ss],innerProductD(x_l,y_l));
|
||||
accelerator_for( ss, sites, 1,{
|
||||
auto x_l = left_v[ss];
|
||||
auto y_l = right_v[ss];
|
||||
inner_tmp_v[ss]=innerProductD(x_l,y_l);
|
||||
});
|
||||
}
|
||||
#else
|
||||
typedef decltype(innerProduct(vobj(),vobj())) inner_t;
|
||||
Vector<inner_t> inner_tmp(sites);
|
||||
auto inner_tmp_v = &inner_tmp[0];
|
||||
|
||||
{
|
||||
autoView( left_v , left, AcceleratorRead);
|
||||
autoView( right_v,right, AcceleratorRead);
|
||||
|
||||
// GPU - SIMT lane compliance...
|
||||
accelerator_for( ss, sites, nsimd,{
|
||||
auto x_l = left_v(ss);
|
||||
auto y_l = right_v(ss);
|
||||
coalescedWrite(inner_tmp_v[ss],innerProduct(x_l,y_l));
|
||||
});
|
||||
}
|
||||
#endif
|
||||
// This is in single precision and fails some tests
|
||||
auto anrm = sumD(inner_tmp_v,sites);
|
||||
auto anrm = sum(inner_tmp_v,sites);
|
||||
nrm = anrm;
|
||||
return nrm;
|
||||
}
|
||||
@ -300,7 +258,7 @@ axpby_norm_fast(Lattice<vobj> &z,sobj a,sobj b,const Lattice<vobj> &x,const Latt
|
||||
conformable(x,y);
|
||||
|
||||
typedef typename vobj::scalar_type scalar_type;
|
||||
// typedef typename vobj::vector_typeD vector_type;
|
||||
typedef typename vobj::vector_typeD vector_type;
|
||||
RealD nrm;
|
||||
|
||||
GridBase *grid = x.Grid();
|
||||
@ -312,29 +270,17 @@ axpby_norm_fast(Lattice<vobj> &z,sobj a,sobj b,const Lattice<vobj> &x,const Latt
|
||||
autoView( x_v, x, AcceleratorRead);
|
||||
autoView( y_v, y, AcceleratorRead);
|
||||
autoView( z_v, z, AcceleratorWrite);
|
||||
#if 0
|
||||
|
||||
typedef decltype(innerProductD(x_v[0],y_v[0])) inner_t;
|
||||
Vector<inner_t> inner_tmp(sites);
|
||||
auto inner_tmp_v = &inner_tmp[0];
|
||||
|
||||
accelerator_for( ss, sites, nsimd,{
|
||||
auto tmp = a*x_v(ss)+b*y_v(ss);
|
||||
coalescedWrite(inner_tmp_v[ss],innerProductD(tmp,tmp));
|
||||
coalescedWrite(z_v[ss],tmp);
|
||||
accelerator_for( ss, sites, 1,{
|
||||
auto tmp = a*x_v[ss]+b*y_v[ss];
|
||||
inner_tmp_v[ss]=innerProductD(tmp,tmp);
|
||||
z_v[ss]=tmp;
|
||||
});
|
||||
nrm = real(TensorRemove(sum(inner_tmp_v,sites)));
|
||||
#else
|
||||
typedef decltype(innerProduct(x_v[0],y_v[0])) inner_t;
|
||||
Vector<inner_t> inner_tmp(sites);
|
||||
auto inner_tmp_v = &inner_tmp[0];
|
||||
|
||||
accelerator_for( ss, sites, nsimd,{
|
||||
auto tmp = a*x_v(ss)+b*y_v(ss);
|
||||
coalescedWrite(inner_tmp_v[ss],innerProduct(tmp,tmp));
|
||||
coalescedWrite(z_v[ss],tmp);
|
||||
});
|
||||
nrm = real(TensorRemove(sumD(inner_tmp_v,sites)));
|
||||
#endif
|
||||
grid->GlobalSum(nrm);
|
||||
return nrm;
|
||||
}
|
||||
@ -415,7 +361,6 @@ template<class vobj> inline void sliceSum(const Lattice<vobj> &Data,std::vector<
|
||||
// But easily avoided by using double precision fields
|
||||
///////////////////////////////////////////////////////
|
||||
typedef typename vobj::scalar_object sobj;
|
||||
typedef typename vobj::scalar_object::scalar_type scalar_type;
|
||||
GridBase *grid = Data.Grid();
|
||||
assert(grid!=NULL);
|
||||
|
||||
@ -474,19 +419,20 @@ template<class vobj> inline void sliceSum(const Lattice<vobj> &Data,std::vector<
|
||||
}
|
||||
|
||||
// sum over nodes.
|
||||
sobj gsum;
|
||||
for(int t=0;t<fd;t++){
|
||||
int pt = t/ld; // processor plane
|
||||
int lt = t%ld;
|
||||
if ( pt == grid->_processor_coor[orthogdim] ) {
|
||||
result[t]=lsSum[lt];
|
||||
gsum=lsSum[lt];
|
||||
} else {
|
||||
result[t]=Zero();
|
||||
gsum=Zero();
|
||||
}
|
||||
|
||||
grid->GlobalSum(gsum);
|
||||
|
||||
result[t]=gsum;
|
||||
}
|
||||
scalar_type * ptr = (scalar_type *) &result[0];
|
||||
int words = fd*sizeof(sobj)/sizeof(scalar_type);
|
||||
grid->GlobalSumVector(ptr, words);
|
||||
}
|
||||
|
||||
template<class vobj>
|
||||
|
||||
@ -23,7 +23,7 @@ unsigned int nextPow2(Iterator x) {
|
||||
}
|
||||
|
||||
template <class Iterator>
|
||||
int getNumBlocksAndThreads(const Iterator n, const size_t sizeofsobj, Iterator &threads, Iterator &blocks) {
|
||||
void getNumBlocksAndThreads(const Iterator n, const size_t sizeofsobj, Iterator &threads, Iterator &blocks) {
|
||||
|
||||
int device;
|
||||
#ifdef GRID_CUDA
|
||||
@ -37,13 +37,14 @@ int getNumBlocksAndThreads(const Iterator n, const size_t sizeofsobj, Iterator &
|
||||
Iterator sharedMemPerBlock = gpu_props[device].sharedMemPerBlock;
|
||||
Iterator maxThreadsPerBlock = gpu_props[device].maxThreadsPerBlock;
|
||||
Iterator multiProcessorCount = gpu_props[device].multiProcessorCount;
|
||||
/*
|
||||
|
||||
std::cout << GridLogDebug << "GPU has:" << std::endl;
|
||||
std::cout << GridLogDebug << "\twarpSize = " << warpSize << std::endl;
|
||||
std::cout << GridLogDebug << "\tsharedMemPerBlock = " << sharedMemPerBlock << std::endl;
|
||||
std::cout << GridLogDebug << "\tmaxThreadsPerBlock = " << maxThreadsPerBlock << std::endl;
|
||||
std::cout << GridLogDebug << "\tmaxThreadsPerBlock = " << warpSize << std::endl;
|
||||
std::cout << GridLogDebug << "\tmultiProcessorCount = " << multiProcessorCount << std::endl;
|
||||
*/
|
||||
|
||||
if (warpSize != WARP_SIZE) {
|
||||
std::cout << GridLogError << "The warp size of the GPU in use does not match the warp size set when compiling Grid." << std::endl;
|
||||
exit(EXIT_FAILURE);
|
||||
@ -51,14 +52,10 @@ int getNumBlocksAndThreads(const Iterator n, const size_t sizeofsobj, Iterator &
|
||||
|
||||
// let the number of threads in a block be a multiple of 2, starting from warpSize
|
||||
threads = warpSize;
|
||||
if ( threads*sizeofsobj > sharedMemPerBlock ) {
|
||||
std::cout << GridLogError << "The object is too large for the shared memory." << std::endl;
|
||||
return 0;
|
||||
}
|
||||
while( 2*threads*sizeofsobj < sharedMemPerBlock && 2*threads <= maxThreadsPerBlock ) threads *= 2;
|
||||
// keep all the streaming multiprocessors busy
|
||||
blocks = nextPow2(multiProcessorCount);
|
||||
return 1;
|
||||
|
||||
}
|
||||
|
||||
template <class sobj, class Iterator>
|
||||
@ -198,7 +195,7 @@ __global__ void reduceKernel(const vobj *lat, sobj *buffer, Iterator n) {
|
||||
// Possibly promote to double and sum
|
||||
/////////////////////////////////////////////////////////////////////////////////////////////////////////
|
||||
template <class vobj>
|
||||
inline typename vobj::scalar_objectD sumD_gpu_small(const vobj *lat, Integer osites)
|
||||
inline typename vobj::scalar_objectD sumD_gpu(const vobj *lat, Integer osites)
|
||||
{
|
||||
typedef typename vobj::scalar_objectD sobj;
|
||||
typedef decltype(lat) Iterator;
|
||||
@ -207,9 +204,7 @@ inline typename vobj::scalar_objectD sumD_gpu_small(const vobj *lat, Integer osi
|
||||
Integer size = osites*nsimd;
|
||||
|
||||
Integer numThreads, numBlocks;
|
||||
int ok = getNumBlocksAndThreads(size, sizeof(sobj), numThreads, numBlocks);
|
||||
assert(ok);
|
||||
|
||||
getNumBlocksAndThreads(size, sizeof(sobj), numThreads, numBlocks);
|
||||
Integer smemSize = numThreads * sizeof(sobj);
|
||||
|
||||
Vector<sobj> buffer(numBlocks);
|
||||
@ -220,54 +215,6 @@ inline typename vobj::scalar_objectD sumD_gpu_small(const vobj *lat, Integer osi
|
||||
auto result = buffer_v[0];
|
||||
return result;
|
||||
}
|
||||
|
||||
template <class vobj>
|
||||
inline typename vobj::scalar_objectD sumD_gpu_large(const vobj *lat, Integer osites)
|
||||
{
|
||||
typedef typename vobj::vector_type vector;
|
||||
typedef typename vobj::scalar_typeD scalarD;
|
||||
typedef typename vobj::scalar_objectD sobj;
|
||||
sobj ret;
|
||||
scalarD *ret_p = (scalarD *)&ret;
|
||||
|
||||
const int words = sizeof(vobj)/sizeof(vector);
|
||||
|
||||
Vector<vector> buffer(osites);
|
||||
vector *dat = (vector *)lat;
|
||||
vector *buf = &buffer[0];
|
||||
iScalar<vector> *tbuf =(iScalar<vector> *) &buffer[0];
|
||||
for(int w=0;w<words;w++) {
|
||||
|
||||
accelerator_for(ss,osites,1,{
|
||||
buf[ss] = dat[ss*words+w];
|
||||
});
|
||||
|
||||
ret_p[w] = sumD_gpu_small(tbuf,osites);
|
||||
}
|
||||
return ret;
|
||||
}
|
||||
|
||||
template <class vobj>
|
||||
inline typename vobj::scalar_objectD sumD_gpu(const vobj *lat, Integer osites)
|
||||
{
|
||||
typedef typename vobj::vector_type vector;
|
||||
typedef typename vobj::scalar_typeD scalarD;
|
||||
typedef typename vobj::scalar_objectD sobj;
|
||||
sobj ret;
|
||||
|
||||
Integer nsimd= vobj::Nsimd();
|
||||
Integer size = osites*nsimd;
|
||||
Integer numThreads, numBlocks;
|
||||
int ok = getNumBlocksAndThreads(size, sizeof(sobj), numThreads, numBlocks);
|
||||
|
||||
if ( ok ) {
|
||||
ret = sumD_gpu_small(lat,osites);
|
||||
} else {
|
||||
ret = sumD_gpu_large(lat,osites);
|
||||
}
|
||||
return ret;
|
||||
}
|
||||
|
||||
/////////////////////////////////////////////////////////////////////////////////////////////////////////
|
||||
// Return as same precision as input performing reduction in double precision though
|
||||
/////////////////////////////////////////////////////////////////////////////////////////////////////////
|
||||
@ -280,13 +227,6 @@ inline typename vobj::scalar_object sum_gpu(const vobj *lat, Integer osites)
|
||||
return result;
|
||||
}
|
||||
|
||||
template <class vobj>
|
||||
inline typename vobj::scalar_object sum_gpu_large(const vobj *lat, Integer osites)
|
||||
{
|
||||
typedef typename vobj::scalar_object sobj;
|
||||
sobj result;
|
||||
result = sumD_gpu_large(lat,osites);
|
||||
return result;
|
||||
}
|
||||
|
||||
|
||||
NAMESPACE_END(Grid);
|
||||
|
||||
@ -424,32 +424,9 @@ public:
|
||||
// MT implementation does not implement fast discard even though
|
||||
// in principle this is possible
|
||||
////////////////////////////////////////////////
|
||||
#if 1
|
||||
thread_for( lidx, _grid->lSites(), {
|
||||
|
||||
int gidx;
|
||||
int o_idx;
|
||||
int i_idx;
|
||||
int rank;
|
||||
Coordinate pcoor;
|
||||
Coordinate lcoor;
|
||||
Coordinate gcoor;
|
||||
_grid->LocalIndexToLocalCoor(lidx,lcoor);
|
||||
pcoor=_grid->ThisProcessorCoor();
|
||||
_grid->ProcessorCoorLocalCoorToGlobalCoor(pcoor,lcoor,gcoor);
|
||||
_grid->GlobalCoorToGlobalIndex(gcoor,gidx);
|
||||
|
||||
_grid->GlobalCoorToRankIndex(rank,o_idx,i_idx,gcoor);
|
||||
assert(rank == _grid->ThisRank() );
|
||||
|
||||
int l_idx=generator_idx(o_idx,i_idx);
|
||||
_generators[l_idx] = master_engine;
|
||||
Skip(_generators[l_idx],gidx); // Skip to next RNG sequence
|
||||
});
|
||||
#else
|
||||
// Everybody loops over global volume.
|
||||
thread_for( gidx, _grid->_gsites, {
|
||||
|
||||
// Where is it?
|
||||
int rank;
|
||||
int o_idx;
|
||||
@ -466,7 +443,6 @@ public:
|
||||
Skip(_generators[l_idx],gidx); // Skip to next RNG sequence
|
||||
}
|
||||
});
|
||||
#endif
|
||||
#else
|
||||
////////////////////////////////////////////////////////////////
|
||||
// Machine and thread decomposition dependent seeding is efficient
|
||||
|
||||
@ -85,76 +85,6 @@ template<class vobj> inline void setCheckerboard(Lattice<vobj> &full,const Latti
|
||||
});
|
||||
}
|
||||
|
||||
template<class vobj> inline void acceleratorPickCheckerboard(int cb,Lattice<vobj> &half,const Lattice<vobj> &full, int checker_dim_half=0)
|
||||
{
|
||||
half.Checkerboard() = cb;
|
||||
autoView(half_v, half, AcceleratorWrite);
|
||||
autoView(full_v, full, AcceleratorRead);
|
||||
Coordinate rdim_full = full.Grid()->_rdimensions;
|
||||
Coordinate rdim_half = half.Grid()->_rdimensions;
|
||||
unsigned long ndim_half = half.Grid()->_ndimension;
|
||||
Coordinate checker_dim_mask_half = half.Grid()->_checker_dim_mask;
|
||||
Coordinate ostride_half = half.Grid()->_ostride;
|
||||
accelerator_for(ss, full.Grid()->oSites(),full.Grid()->Nsimd(),{
|
||||
|
||||
Coordinate coor;
|
||||
int cbos;
|
||||
int linear=0;
|
||||
|
||||
Lexicographic::CoorFromIndex(coor,ss,rdim_full);
|
||||
assert(coor.size()==ndim_half);
|
||||
|
||||
for(int d=0;d<ndim_half;d++){
|
||||
if(checker_dim_mask_half[d]) linear += coor[d];
|
||||
}
|
||||
cbos = (linear&0x1);
|
||||
|
||||
if (cbos==cb) {
|
||||
int ssh=0;
|
||||
for(int d=0;d<ndim_half;d++) {
|
||||
if (d == checker_dim_half) ssh += ostride_half[d] * ((coor[d] / 2) % rdim_half[d]);
|
||||
else ssh += ostride_half[d] * (coor[d] % rdim_half[d]);
|
||||
}
|
||||
coalescedWrite(half_v[ssh],full_v(ss));
|
||||
}
|
||||
});
|
||||
}
|
||||
template<class vobj> inline void acceleratorSetCheckerboard(Lattice<vobj> &full,const Lattice<vobj> &half, int checker_dim_half=0)
|
||||
{
|
||||
int cb = half.Checkerboard();
|
||||
autoView(half_v , half, AcceleratorRead);
|
||||
autoView(full_v , full, AcceleratorWrite);
|
||||
Coordinate rdim_full = full.Grid()->_rdimensions;
|
||||
Coordinate rdim_half = half.Grid()->_rdimensions;
|
||||
unsigned long ndim_half = half.Grid()->_ndimension;
|
||||
Coordinate checker_dim_mask_half = half.Grid()->_checker_dim_mask;
|
||||
Coordinate ostride_half = half.Grid()->_ostride;
|
||||
accelerator_for(ss,full.Grid()->oSites(),full.Grid()->Nsimd(),{
|
||||
|
||||
Coordinate coor;
|
||||
int cbos;
|
||||
int linear=0;
|
||||
|
||||
Lexicographic::CoorFromIndex(coor,ss,rdim_full);
|
||||
assert(coor.size()==ndim_half);
|
||||
|
||||
for(int d=0;d<ndim_half;d++){
|
||||
if(checker_dim_mask_half[d]) linear += coor[d];
|
||||
}
|
||||
cbos = (linear&0x1);
|
||||
|
||||
if (cbos==cb) {
|
||||
int ssh=0;
|
||||
for(int d=0;d<ndim_half;d++){
|
||||
if (d == checker_dim_half) ssh += ostride_half[d] * ((coor[d] / 2) % rdim_half[d]);
|
||||
else ssh += ostride_half[d] * (coor[d] % rdim_half[d]);
|
||||
}
|
||||
coalescedWrite(full_v[ss],half_v(ssh));
|
||||
}
|
||||
|
||||
});
|
||||
}
|
||||
|
||||
////////////////////////////////////////////////////////////////////////////////////////////
|
||||
// Flexible Type Conversion for internal promotion to double as well as graceful
|
||||
// treatment of scalar-compatible types
|
||||
@ -434,22 +364,16 @@ inline void blockSum(Lattice<vobj> &coarseData,const Lattice<vobj> &fineData)
|
||||
autoView( coarseData_ , coarseData, AcceleratorWrite);
|
||||
autoView( fineData_ , fineData, AcceleratorRead);
|
||||
|
||||
auto coarseData_p = &coarseData_[0];
|
||||
auto fineData_p = &fineData_[0];
|
||||
|
||||
Coordinate fine_rdimensions = fine->_rdimensions;
|
||||
Coordinate coarse_rdimensions = coarse->_rdimensions;
|
||||
|
||||
vobj zz = Zero();
|
||||
|
||||
accelerator_for(sc,coarse->oSites(),1,{
|
||||
|
||||
// One thread per sub block
|
||||
Coordinate coor_c(_ndimension);
|
||||
Lexicographic::CoorFromIndex(coor_c,sc,coarse_rdimensions); // Block coordinate
|
||||
coarseData_[sc]=Zero();
|
||||
|
||||
vobj cd = zz;
|
||||
|
||||
for(int sb=0;sb<blockVol;sb++){
|
||||
|
||||
int sf;
|
||||
@ -459,11 +383,9 @@ inline void blockSum(Lattice<vobj> &coarseData,const Lattice<vobj> &fineData)
|
||||
for(int d=0;d<_ndimension;d++) coor_f[d]=coor_c[d]*block_r[d] + coor_b[d];
|
||||
Lexicographic::IndexFromCoor(coor_f,sf,fine_rdimensions);
|
||||
|
||||
cd=cd+fineData_p[sf];
|
||||
coarseData_[sc]=coarseData_[sc]+fineData_[sf];
|
||||
}
|
||||
|
||||
coarseData_p[sc] = cd;
|
||||
|
||||
});
|
||||
return;
|
||||
}
|
||||
@ -855,7 +777,7 @@ void ExtractSliceLocal(Lattice<vobj> &lowDim,const Lattice<vobj> & higherDim,int
|
||||
|
||||
|
||||
template<class vobj>
|
||||
void Replicate(const Lattice<vobj> &coarse,Lattice<vobj> & fine)
|
||||
void Replicate(Lattice<vobj> &coarse,Lattice<vobj> & fine)
|
||||
{
|
||||
typedef typename vobj::scalar_object sobj;
|
||||
|
||||
|
||||
@ -65,34 +65,29 @@ GridLogger GridLogSolver (1, "Solver", GridLogColours, "NORMAL");
|
||||
GridLogger GridLogError (1, "Error" , GridLogColours, "RED");
|
||||
GridLogger GridLogWarning(1, "Warning", GridLogColours, "YELLOW");
|
||||
GridLogger GridLogMessage(1, "Message", GridLogColours, "NORMAL");
|
||||
GridLogger GridLogMemory (1, "Memory", GridLogColours, "NORMAL");
|
||||
GridLogger GridLogDebug (1, "Debug", GridLogColours, "PURPLE");
|
||||
GridLogger GridLogPerformance(1, "Performance", GridLogColours, "GREEN");
|
||||
GridLogger GridLogIterative (1, "Iterative", GridLogColours, "BLUE");
|
||||
GridLogger GridLogIntegrator (1, "Integrator", GridLogColours, "BLUE");
|
||||
GridLogger GridLogHMC (1, "HMC", GridLogColours, "BLUE");
|
||||
|
||||
void GridLogConfigure(std::vector<std::string> &logstreams) {
|
||||
GridLogError.Active(1);
|
||||
GridLogError.Active(0);
|
||||
GridLogWarning.Active(0);
|
||||
GridLogMessage.Active(1); // at least the messages should be always on
|
||||
GridLogMemory.Active(0); // at least the messages should be always on
|
||||
GridLogIterative.Active(0);
|
||||
GridLogDebug.Active(0);
|
||||
GridLogPerformance.Active(0);
|
||||
GridLogIntegrator.Active(1);
|
||||
GridLogColours.Active(0);
|
||||
GridLogHMC.Active(1);
|
||||
|
||||
for (int i = 0; i < logstreams.size(); i++) {
|
||||
if (logstreams[i] == std::string("Memory")) GridLogMemory.Active(1);
|
||||
if (logstreams[i] == std::string("Error")) GridLogError.Active(1);
|
||||
if (logstreams[i] == std::string("Warning")) GridLogWarning.Active(1);
|
||||
if (logstreams[i] == std::string("NoMessage")) GridLogMessage.Active(0);
|
||||
if (logstreams[i] == std::string("Iterative")) GridLogIterative.Active(1);
|
||||
if (logstreams[i] == std::string("Debug")) GridLogDebug.Active(1);
|
||||
if (logstreams[i] == std::string("Performance")) GridLogPerformance.Active(1);
|
||||
if (logstreams[i] == std::string("NoIntegrator")) GridLogIntegrator.Active(0);
|
||||
if (logstreams[i] == std::string("NoHMC")) GridLogHMC.Active(0);
|
||||
if (logstreams[i] == std::string("Integrator")) GridLogIntegrator.Active(1);
|
||||
if (logstreams[i] == std::string("Colours")) GridLogColours.Active(1);
|
||||
}
|
||||
}
|
||||
|
||||
@ -182,8 +182,6 @@ extern GridLogger GridLogDebug ;
|
||||
extern GridLogger GridLogPerformance;
|
||||
extern GridLogger GridLogIterative ;
|
||||
extern GridLogger GridLogIntegrator ;
|
||||
extern GridLogger GridLogHMC;
|
||||
extern GridLogger GridLogMemory;
|
||||
extern Colours GridLogColours;
|
||||
|
||||
std::string demangle(const char* name) ;
|
||||
|
||||
@ -31,7 +31,6 @@ directory
|
||||
#include <fstream>
|
||||
#include <iomanip>
|
||||
#include <iostream>
|
||||
#include <string>
|
||||
#include <map>
|
||||
|
||||
#include <pwd.h>
|
||||
@ -577,8 +576,6 @@ class ScidacReader : public GridLimeReader {
|
||||
std::string rec_name(ILDG_BINARY_DATA);
|
||||
while ( limeReaderNextRecord(LimeR) == LIME_SUCCESS ) {
|
||||
if ( !strncmp(limeReaderType(LimeR), rec_name.c_str(),strlen(rec_name.c_str()) ) ) {
|
||||
// in principle should do the line below, but that breaks backard compatibility with old data
|
||||
// skipPastObjectRecord(std::string(GRID_FIELD_NORM));
|
||||
skipPastObjectRecord(std::string(SCIDAC_CHECKSUM));
|
||||
return;
|
||||
}
|
||||
@ -655,8 +652,7 @@ class IldgWriter : public ScidacWriter {
|
||||
// Fill ILDG header data struct
|
||||
//////////////////////////////////////////////////////
|
||||
ildgFormat ildgfmt ;
|
||||
const std::string stNC = std::to_string( Nc ) ;
|
||||
ildgfmt.field = std::string("su"+stNC+"gauge");
|
||||
ildgfmt.field = std::string("su3gauge");
|
||||
|
||||
if ( format == std::string("IEEE32BIG") ) {
|
||||
ildgfmt.precision = 32;
|
||||
@ -873,8 +869,7 @@ class IldgReader : public GridLimeReader {
|
||||
} else {
|
||||
|
||||
assert(found_ildgFormat);
|
||||
const std::string stNC = std::to_string( Nc ) ;
|
||||
assert ( ildgFormat_.field == std::string("su"+stNC+"gauge") );
|
||||
assert ( ildgFormat_.field == std::string("su3gauge") );
|
||||
|
||||
///////////////////////////////////////////////////////////////////////////////////////
|
||||
// Populate our Grid metadata as best we can
|
||||
@ -882,7 +877,7 @@ class IldgReader : public GridLimeReader {
|
||||
|
||||
std::ostringstream vers; vers << ildgFormat_.version;
|
||||
FieldMetaData_.hdr_version = vers.str();
|
||||
FieldMetaData_.data_type = std::string("4D_SU"+stNC+"_GAUGE_"+stNC+"x"+stNC);
|
||||
FieldMetaData_.data_type = std::string("4D_SU3_GAUGE_3X3");
|
||||
|
||||
FieldMetaData_.nd=4;
|
||||
FieldMetaData_.dimension.resize(4);
|
||||
|
||||
@ -6,8 +6,8 @@
|
||||
|
||||
Copyright (C) 2015
|
||||
|
||||
|
||||
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
|
||||
Author: Jamie Hudspith <renwick.james.hudspth@gmail.com>
|
||||
|
||||
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
|
||||
@ -182,8 +182,8 @@ class GaugeStatistics
|
||||
public:
|
||||
void operator()(Lattice<vLorentzColourMatrixD> & data,FieldMetaData &header)
|
||||
{
|
||||
header.link_trace = WilsonLoops<Impl>::linkTrace(data);
|
||||
header.plaquette = WilsonLoops<Impl>::avgPlaquette(data);
|
||||
header.link_trace=WilsonLoops<Impl>::linkTrace(data);
|
||||
header.plaquette =WilsonLoops<Impl>::avgPlaquette(data);
|
||||
}
|
||||
};
|
||||
typedef GaugeStatistics<PeriodicGimplD> PeriodicGaugeStatistics;
|
||||
@ -203,24 +203,20 @@ template<> inline void PrepareMetaData<vLorentzColourMatrixD>(Lattice<vLorentzCo
|
||||
//////////////////////////////////////////////////////////////////////
|
||||
inline void reconstruct3(LorentzColourMatrix & cm)
|
||||
{
|
||||
assert( Nc < 4 && Nc > 1 ) ;
|
||||
const int x=0;
|
||||
const int y=1;
|
||||
const int z=2;
|
||||
for(int mu=0;mu<Nd;mu++){
|
||||
#if Nc == 2
|
||||
cm(mu)()(1,0) = -adj(cm(mu)()(0,y)) ;
|
||||
cm(mu)()(1,1) = adj(cm(mu)()(0,x)) ;
|
||||
#else
|
||||
const int x=0 , y=1 , z=2 ; // a little disinenuous labelling
|
||||
cm(mu)()(2,x) = adj(cm(mu)()(0,y)*cm(mu)()(1,z)-cm(mu)()(0,z)*cm(mu)()(1,y)); //x= yz-zy
|
||||
cm(mu)()(2,y) = adj(cm(mu)()(0,z)*cm(mu)()(1,x)-cm(mu)()(0,x)*cm(mu)()(1,z)); //y= zx-xz
|
||||
cm(mu)()(2,z) = adj(cm(mu)()(0,x)*cm(mu)()(1,y)-cm(mu)()(0,y)*cm(mu)()(1,x)); //z= xy-yx
|
||||
#endif
|
||||
cm(mu)()(2,x) = adj(cm(mu)()(0,y)*cm(mu)()(1,z)-cm(mu)()(0,z)*cm(mu)()(1,y)); //x= yz-zy
|
||||
cm(mu)()(2,y) = adj(cm(mu)()(0,z)*cm(mu)()(1,x)-cm(mu)()(0,x)*cm(mu)()(1,z)); //y= zx-xz
|
||||
cm(mu)()(2,z) = adj(cm(mu)()(0,x)*cm(mu)()(1,y)-cm(mu)()(0,y)*cm(mu)()(1,x)); //z= xy-yx
|
||||
}
|
||||
}
|
||||
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
// Some data types for intermediate storage
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
template<typename vtype> using iLorentzColour2x3 = iVector<iVector<iVector<vtype, Nc>, Nc-1>, Nd >;
|
||||
template<typename vtype> using iLorentzColour2x3 = iVector<iVector<iVector<vtype, Nc>, 2>, Nd >;
|
||||
|
||||
typedef iLorentzColour2x3<Complex> LorentzColour2x3;
|
||||
typedef iLorentzColour2x3<ComplexF> LorentzColour2x3F;
|
||||
@ -282,6 +278,7 @@ struct GaugeSimpleMunger{
|
||||
|
||||
template <class fobj, class sobj>
|
||||
struct GaugeSimpleUnmunger {
|
||||
|
||||
void operator()(sobj &in, fobj &out) {
|
||||
for (int mu = 0; mu < Nd; mu++) {
|
||||
for (int i = 0; i < Nc; i++) {
|
||||
@ -320,8 +317,8 @@ template<class fobj,class sobj>
|
||||
struct Gauge3x2munger{
|
||||
void operator() (fobj &in,sobj &out){
|
||||
for(int mu=0;mu<Nd;mu++){
|
||||
for(int i=0;i<Nc-1;i++){
|
||||
for(int j=0;j<Nc;j++){
|
||||
for(int i=0;i<2;i++){
|
||||
for(int j=0;j<3;j++){
|
||||
out(mu)()(i,j) = in(mu)(i)(j);
|
||||
}}
|
||||
}
|
||||
@ -333,8 +330,8 @@ template<class fobj,class sobj>
|
||||
struct Gauge3x2unmunger{
|
||||
void operator() (sobj &in,fobj &out){
|
||||
for(int mu=0;mu<Nd;mu++){
|
||||
for(int i=0;i<Nc-1;i++){
|
||||
for(int j=0;j<Nc;j++){
|
||||
for(int i=0;i<2;i++){
|
||||
for(int j=0;j<3;j++){
|
||||
out(mu)(i)(j) = in(mu)()(i,j);
|
||||
}}
|
||||
}
|
||||
|
||||
@ -9,7 +9,6 @@
|
||||
Author: Matt Spraggs <matthew.spraggs@gmail.com>
|
||||
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
|
||||
Author: paboyle <paboyle@ph.ed.ac.uk>
|
||||
Author: Jamie Hudspith <renwick.james.hudspth@gmail.com>
|
||||
|
||||
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
|
||||
@ -31,8 +30,6 @@
|
||||
#ifndef GRID_NERSC_IO_H
|
||||
#define GRID_NERSC_IO_H
|
||||
|
||||
#include <string>
|
||||
|
||||
NAMESPACE_BEGIN(Grid);
|
||||
|
||||
using namespace Grid;
|
||||
@ -42,10 +39,8 @@ using namespace Grid;
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
class NerscIO : public BinaryIO {
|
||||
public:
|
||||
typedef Lattice<vLorentzColourMatrixD> GaugeField;
|
||||
|
||||
// Enable/disable exiting if the plaquette in the header does not match the value computed (default true)
|
||||
static bool & exitOnReadPlaquetteMismatch(){ static bool v=true; return v; }
|
||||
typedef Lattice<vLorentzColourMatrixD> GaugeField;
|
||||
|
||||
static inline void truncate(std::string file){
|
||||
std::ofstream fout(file,std::ios::out);
|
||||
@ -150,17 +145,15 @@ public:
|
||||
|
||||
std::string format(header.floating_point);
|
||||
|
||||
const int ieee32big = (format == std::string("IEEE32BIG"));
|
||||
const int ieee32 = (format == std::string("IEEE32"));
|
||||
const int ieee64big = (format == std::string("IEEE64BIG"));
|
||||
const int ieee64 = (format == std::string("IEEE64") || \
|
||||
format == std::string("IEEE64LITTLE"));
|
||||
int ieee32big = (format == std::string("IEEE32BIG"));
|
||||
int ieee32 = (format == std::string("IEEE32"));
|
||||
int ieee64big = (format == std::string("IEEE64BIG"));
|
||||
int ieee64 = (format == std::string("IEEE64") || format == std::string("IEEE64LITTLE"));
|
||||
|
||||
uint32_t nersc_csum,scidac_csuma,scidac_csumb;
|
||||
// depending on datatype, set up munger;
|
||||
// munger is a function of <floating point, Real, data_type>
|
||||
const std::string stNC = std::to_string( Nc ) ;
|
||||
if ( header.data_type == std::string("4D_SU"+stNC+"_GAUGE") ) {
|
||||
if ( header.data_type == std::string("4D_SU3_GAUGE") ) {
|
||||
if ( ieee32 || ieee32big ) {
|
||||
BinaryIO::readLatticeObject<vLorentzColourMatrixD, LorentzColour2x3F>
|
||||
(Umu,file,Gauge3x2munger<LorentzColour2x3F,LorentzColourMatrix>(), offset,format,
|
||||
@ -171,7 +164,7 @@ public:
|
||||
(Umu,file,Gauge3x2munger<LorentzColour2x3D,LorentzColourMatrix>(),offset,format,
|
||||
nersc_csum,scidac_csuma,scidac_csumb);
|
||||
}
|
||||
} else if ( header.data_type == std::string("4D_SU"+stNC+"_GAUGE_"+stNC+"x"+stNC) ) {
|
||||
} else if ( header.data_type == std::string("4D_SU3_GAUGE_3x3") ) {
|
||||
if ( ieee32 || ieee32big ) {
|
||||
BinaryIO::readLatticeObject<vLorentzColourMatrixD,LorentzColourMatrixF>
|
||||
(Umu,file,GaugeSimpleMunger<LorentzColourMatrixF,LorentzColourMatrix>(),offset,format,
|
||||
@ -205,7 +198,7 @@ public:
|
||||
std::cerr << " nersc_csum " <<std::hex<< nersc_csum << " " << header.checksum<< std::dec<< std::endl;
|
||||
exit(0);
|
||||
}
|
||||
if(exitOnReadPlaquetteMismatch()) assert(fabs(clone.plaquette -header.plaquette ) < 1.0e-5 );
|
||||
assert(fabs(clone.plaquette -header.plaquette ) < 1.0e-5 );
|
||||
assert(fabs(clone.link_trace-header.link_trace) < 1.0e-6 );
|
||||
assert(nersc_csum == header.checksum );
|
||||
|
||||
@ -216,29 +209,27 @@ public:
|
||||
template<class GaugeStats=PeriodicGaugeStatistics>
|
||||
static inline void writeConfiguration(Lattice<vLorentzColourMatrixD > &Umu,
|
||||
std::string file,
|
||||
std::string ens_label = std::string("DWF"),
|
||||
std::string ens_id = std::string("UKQCD"),
|
||||
unsigned int sequence_number = 1)
|
||||
std::string ens_label = std::string("DWF"))
|
||||
{
|
||||
writeConfiguration(Umu,file,0,1,ens_label,ens_id,sequence_number);
|
||||
writeConfiguration(Umu,file,0,1,ens_label);
|
||||
}
|
||||
template<class GaugeStats=PeriodicGaugeStatistics>
|
||||
static inline void writeConfiguration(Lattice<vLorentzColourMatrixD > &Umu,
|
||||
std::string file,
|
||||
int two_row,
|
||||
int bits32,
|
||||
std::string ens_label = std::string("DWF"),
|
||||
std::string ens_id = std::string("UKQCD"),
|
||||
unsigned int sequence_number = 1)
|
||||
std::string ens_label = std::string("DWF"))
|
||||
{
|
||||
typedef vLorentzColourMatrixD vobj;
|
||||
typedef typename vobj::scalar_object sobj;
|
||||
|
||||
FieldMetaData header;
|
||||
header.sequence_number = sequence_number;
|
||||
header.ensemble_id = ens_id;
|
||||
///////////////////////////////////////////
|
||||
// Following should become arguments
|
||||
///////////////////////////////////////////
|
||||
header.sequence_number = 1;
|
||||
header.ensemble_id = std::string("UKQCD");
|
||||
header.ensemble_label = ens_label;
|
||||
header.hdr_version = "1.0" ;
|
||||
|
||||
typedef LorentzColourMatrixD fobj3D;
|
||||
typedef LorentzColour2x3D fobj2D;
|
||||
@ -252,14 +243,10 @@ public:
|
||||
|
||||
uint64_t offset;
|
||||
|
||||
// Sod it -- always write NcxNc double
|
||||
header.floating_point = std::string("IEEE64BIG");
|
||||
const std::string stNC = std::to_string( Nc ) ;
|
||||
if( two_row ) {
|
||||
header.data_type = std::string("4D_SU" + stNC + "_GAUGE" );
|
||||
} else {
|
||||
header.data_type = std::string("4D_SU" + stNC + "_GAUGE_" + stNC + "x" + stNC );
|
||||
}
|
||||
// Sod it -- always write 3x3 double
|
||||
header.floating_point = std::string("IEEE64BIG");
|
||||
header.data_type = std::string("4D_SU3_GAUGE_3x3");
|
||||
GaugeSimpleUnmunger<fobj3D,sobj> munge;
|
||||
if ( grid->IsBoss() ) {
|
||||
truncate(file);
|
||||
offset = writeHeader(header,file);
|
||||
@ -267,15 +254,8 @@ public:
|
||||
grid->Broadcast(0,(void *)&offset,sizeof(offset));
|
||||
|
||||
uint32_t nersc_csum,scidac_csuma,scidac_csumb;
|
||||
if( two_row ) {
|
||||
Gauge3x2unmunger<fobj2D,sobj> munge;
|
||||
BinaryIO::writeLatticeObject<vobj,fobj2D>(Umu,file,munge,offset,header.floating_point,
|
||||
nersc_csum,scidac_csuma,scidac_csumb);
|
||||
} else {
|
||||
GaugeSimpleUnmunger<fobj3D,sobj> munge;
|
||||
BinaryIO::writeLatticeObject<vobj,fobj3D>(Umu,file,munge,offset,header.floating_point,
|
||||
nersc_csum,scidac_csuma,scidac_csumb);
|
||||
}
|
||||
BinaryIO::writeLatticeObject<vobj,fobj3D>(Umu,file,munge,offset,header.floating_point,
|
||||
nersc_csum,scidac_csuma,scidac_csumb);
|
||||
header.checksum = nersc_csum;
|
||||
if ( grid->IsBoss() ) {
|
||||
writeHeader(header,file);
|
||||
@ -307,7 +287,8 @@ public:
|
||||
header.plaquette=0.0;
|
||||
MachineCharacteristics(header);
|
||||
|
||||
uint64_t offset;
|
||||
uint64_t offset;
|
||||
|
||||
#ifdef RNG_RANLUX
|
||||
header.floating_point = std::string("UINT64");
|
||||
header.data_type = std::string("RANLUX48");
|
||||
@ -347,7 +328,7 @@ public:
|
||||
|
||||
GridBase *grid = parallel.Grid();
|
||||
|
||||
uint64_t offset = readHeader(file,grid,header);
|
||||
uint64_t offset = readHeader(file,grid,header);
|
||||
|
||||
FieldMetaData clone(header);
|
||||
|
||||
|
||||
@ -72,9 +72,17 @@ static long perf_event_open(struct perf_event_attr *hw_event, pid_t pid,
|
||||
inline uint64_t cyclecount(void){
|
||||
return 0;
|
||||
}
|
||||
#define __SSC_MARK(mark) __asm__ __volatile__ ("movl %0, %%ebx; .byte 0x64, 0x67, 0x90 " ::"i"(mark):"%ebx")
|
||||
#define __SSC_STOP __SSC_MARK(0x110)
|
||||
#define __SSC_START __SSC_MARK(0x111)
|
||||
|
||||
|
||||
#else
|
||||
|
||||
#define __SSC_MARK(mark)
|
||||
#define __SSC_STOP
|
||||
#define __SSC_START
|
||||
|
||||
/*
|
||||
* cycle counters arch dependent
|
||||
*/
|
||||
|
||||
@ -39,9 +39,9 @@ NAMESPACE_BEGIN(Grid)
|
||||
// C++11 time facilities better?
|
||||
inline double usecond(void) {
|
||||
struct timeval tv;
|
||||
tv.tv_sec = 0;
|
||||
tv.tv_usec = 0;
|
||||
#ifdef TIMERS_ON
|
||||
gettimeofday(&tv,NULL);
|
||||
#endif
|
||||
return 1.0*tv.tv_usec + 1.0e6*tv.tv_sec;
|
||||
}
|
||||
|
||||
|
||||
@ -16,12 +16,8 @@
|
||||
|
||||
#ifdef __NVCC__
|
||||
#pragma push
|
||||
#if (__CUDACC_VER_MAJOR__ >= 11) && (__CUDACC_VER_MINOR__ >= 5)
|
||||
#pragma nv_diag_suppress declared_but_not_referenced // suppress "function was declared but never referenced warning"
|
||||
#else
|
||||
#pragma diag_suppress declared_but_not_referenced // suppress "function was declared but never referenced warning"
|
||||
#endif
|
||||
#endif
|
||||
|
||||
#include "pugixml.h"
|
||||
|
||||
|
||||
@ -63,7 +63,6 @@ static constexpr int Ngp=2; // gparity index range
|
||||
#define ColourIndex (2)
|
||||
#define SpinIndex (1)
|
||||
#define LorentzIndex (0)
|
||||
#define GparityFlavourIndex (0)
|
||||
|
||||
// Also should make these a named enum type
|
||||
static constexpr int DaggerNo=0;
|
||||
@ -88,8 +87,6 @@ template<typename T> struct isCoarsened {
|
||||
template <typename T> using IfCoarsened = Invoke<std::enable_if< isCoarsened<T>::value,int> > ;
|
||||
template <typename T> using IfNotCoarsened = Invoke<std::enable_if<!isCoarsened<T>::value,int> > ;
|
||||
|
||||
const int GparityFlavourTensorIndex = 3; //TensorLevel counts from the bottom!
|
||||
|
||||
// ChrisK very keen to add extra space for Gparity doubling.
|
||||
//
|
||||
// Also add domain wall index, in a way where Wilson operator
|
||||
@ -113,10 +110,8 @@ template<typename vtype> using iHalfSpinColourVector = iScalar<iVector<iVec
|
||||
template<typename vtype> using iSpinColourSpinColourMatrix = iScalar<iMatrix<iMatrix<iMatrix<iMatrix<vtype, Nc>, Ns>, Nc>, Ns> >;
|
||||
|
||||
|
||||
template<typename vtype> using iGparityFlavourVector = iVector<iScalar<iScalar<vtype> >, Ngp>;
|
||||
template<typename vtype> using iGparitySpinColourVector = iVector<iVector<iVector<vtype, Nc>, Ns>, Ngp >;
|
||||
template<typename vtype> using iGparityHalfSpinColourVector = iVector<iVector<iVector<vtype, Nc>, Nhs>, Ngp >;
|
||||
template<typename vtype> using iGparityFlavourMatrix = iMatrix<iScalar<iScalar<vtype> >, Ngp>;
|
||||
|
||||
// Spin matrix
|
||||
typedef iSpinMatrix<Complex > SpinMatrix;
|
||||
@ -181,16 +176,6 @@ typedef iDoubleStoredColourMatrix<vComplex > vDoubleStoredColourMatrix;
|
||||
typedef iDoubleStoredColourMatrix<vComplexF> vDoubleStoredColourMatrixF;
|
||||
typedef iDoubleStoredColourMatrix<vComplexD> vDoubleStoredColourMatrixD;
|
||||
|
||||
//G-parity flavour matrix
|
||||
typedef iGparityFlavourMatrix<Complex> GparityFlavourMatrix;
|
||||
typedef iGparityFlavourMatrix<ComplexF> GparityFlavourMatrixF;
|
||||
typedef iGparityFlavourMatrix<ComplexD> GparityFlavourMatrixD;
|
||||
|
||||
typedef iGparityFlavourMatrix<vComplex> vGparityFlavourMatrix;
|
||||
typedef iGparityFlavourMatrix<vComplexF> vGparityFlavourMatrixF;
|
||||
typedef iGparityFlavourMatrix<vComplexD> vGparityFlavourMatrixD;
|
||||
|
||||
|
||||
// Spin vector
|
||||
typedef iSpinVector<Complex > SpinVector;
|
||||
typedef iSpinVector<ComplexF> SpinVectorF;
|
||||
@ -235,16 +220,6 @@ typedef iHalfSpinColourVector<ComplexD> HalfSpinColourVectorD;
|
||||
typedef iHalfSpinColourVector<vComplex > vHalfSpinColourVector;
|
||||
typedef iHalfSpinColourVector<vComplexF> vHalfSpinColourVectorF;
|
||||
typedef iHalfSpinColourVector<vComplexD> vHalfSpinColourVectorD;
|
||||
|
||||
//G-parity flavour vector
|
||||
typedef iGparityFlavourVector<Complex > GparityFlavourVector;
|
||||
typedef iGparityFlavourVector<ComplexF> GparityFlavourVectorF;
|
||||
typedef iGparityFlavourVector<ComplexD> GparityFlavourVectorD;
|
||||
|
||||
typedef iGparityFlavourVector<vComplex > vGparityFlavourVector;
|
||||
typedef iGparityFlavourVector<vComplexF> vGparityFlavourVectorF;
|
||||
typedef iGparityFlavourVector<vComplexD> vGparityFlavourVectorD;
|
||||
|
||||
|
||||
// singlets
|
||||
typedef iSinglet<Complex > TComplex; // FIXME This is painful. Tensor singlet complex type.
|
||||
|
||||
@ -40,29 +40,6 @@ class Action
|
||||
|
||||
public:
|
||||
bool is_smeared = false;
|
||||
RealD deriv_norm_sum;
|
||||
RealD deriv_max_sum;
|
||||
int deriv_num;
|
||||
RealD deriv_us;
|
||||
RealD S_us;
|
||||
RealD refresh_us;
|
||||
void reset_timer(void) {
|
||||
deriv_us = S_us = refresh_us = 0.0;
|
||||
deriv_num=0;
|
||||
deriv_norm_sum = deriv_max_sum=0.0;
|
||||
}
|
||||
void deriv_log(RealD nrm, RealD max) { deriv_max_sum+=max; deriv_norm_sum+=nrm; deriv_num++;}
|
||||
RealD deriv_max_average(void) { return deriv_max_sum/deriv_num; };
|
||||
RealD deriv_norm_average(void) { return deriv_norm_sum/deriv_num; };
|
||||
RealD deriv_timer(void) { return deriv_us; };
|
||||
RealD S_timer(void) { return deriv_us; };
|
||||
RealD refresh_timer(void) { return deriv_us; };
|
||||
void deriv_timer_start(void) { deriv_us-=usecond(); }
|
||||
void deriv_timer_stop(void) { deriv_us+=usecond(); }
|
||||
void refresh_timer_start(void) { refresh_us-=usecond(); }
|
||||
void refresh_timer_stop(void) { refresh_us+=usecond(); }
|
||||
void S_timer_start(void) { S_us-=usecond(); }
|
||||
void S_timer_stop(void) { S_us+=usecond(); }
|
||||
// Heatbath?
|
||||
virtual void refresh(const GaugeField& U, GridSerialRNG &sRNG, GridParallelRNG& pRNG) = 0; // refresh pseudofermions
|
||||
virtual RealD S(const GaugeField& U) = 0; // evaluate the action
|
||||
|
||||
@ -37,10 +37,6 @@ NAMESPACE_CHECK(ActionSet);
|
||||
#include <Grid/qcd/action/ActionParams.h>
|
||||
NAMESPACE_CHECK(ActionParams);
|
||||
|
||||
#include <Grid/qcd/action/filters/MomentumFilter.h>
|
||||
#include <Grid/qcd/action/filters/DirichletFilter.h>
|
||||
#include <Grid/qcd/action/filters/DDHMCFilter.h>
|
||||
|
||||
////////////////////////////////////////////
|
||||
// Gauge Actions
|
||||
////////////////////////////////////////////
|
||||
|
||||
@ -37,33 +37,24 @@ NAMESPACE_BEGIN(Grid);
|
||||
// These can move into a params header and be given MacroMagic serialisation
|
||||
struct GparityWilsonImplParams {
|
||||
Coordinate twists;
|
||||
//mu=Nd-1 is assumed to be the time direction and a twist value of 1 indicates antiperiodic BCs
|
||||
Coordinate dirichlet; // Blocksize of dirichlet BCs
|
||||
GparityWilsonImplParams() : twists(Nd, 0), dirichlet(Nd, 0) {};
|
||||
GparityWilsonImplParams() : twists(Nd, 0) {};
|
||||
};
|
||||
|
||||
struct WilsonImplParams {
|
||||
bool overlapCommsCompute;
|
||||
Coordinate dirichlet; // Blocksize of dirichlet BCs
|
||||
AcceleratorVector<Real,Nd> twist_n_2pi_L;
|
||||
AcceleratorVector<Complex,Nd> boundary_phases;
|
||||
WilsonImplParams() {
|
||||
dirichlet.resize(Nd,0);
|
||||
boundary_phases.resize(Nd, 1.0);
|
||||
twist_n_2pi_L.resize(Nd, 0.0);
|
||||
};
|
||||
WilsonImplParams(const AcceleratorVector<Complex,Nd> phi) : boundary_phases(phi), overlapCommsCompute(false) {
|
||||
twist_n_2pi_L.resize(Nd, 0.0);
|
||||
dirichlet.resize(Nd,0);
|
||||
}
|
||||
};
|
||||
|
||||
struct StaggeredImplParams {
|
||||
Coordinate dirichlet; // Blocksize of dirichlet BCs
|
||||
StaggeredImplParams()
|
||||
{
|
||||
dirichlet.resize(Nd,0);
|
||||
};
|
||||
StaggeredImplParams() {};
|
||||
};
|
||||
|
||||
struct OneFlavourRationalParams : Serializable {
|
||||
@ -72,11 +63,9 @@ struct StaggeredImplParams {
|
||||
RealD, hi,
|
||||
int, MaxIter,
|
||||
RealD, tolerance,
|
||||
RealD, mdtolerance,
|
||||
int, degree,
|
||||
int, precision,
|
||||
int, BoundsCheckFreq,
|
||||
RealD, BoundsCheckTol);
|
||||
int, BoundsCheckFreq);
|
||||
|
||||
// MaxIter and tolerance, vectors??
|
||||
|
||||
@ -87,62 +76,16 @@ struct StaggeredImplParams {
|
||||
RealD tol = 1.0e-8,
|
||||
int _degree = 10,
|
||||
int _precision = 64,
|
||||
int _BoundsCheckFreq=20,
|
||||
RealD mdtol = 1.0e-6,
|
||||
double _BoundsCheckTol=1e-6)
|
||||
int _BoundsCheckFreq=20)
|
||||
: lo(_lo),
|
||||
hi(_hi),
|
||||
MaxIter(_maxit),
|
||||
tolerance(tol),
|
||||
mdtolerance(mdtol),
|
||||
degree(_degree),
|
||||
precision(_precision),
|
||||
BoundsCheckFreq(_BoundsCheckFreq),
|
||||
BoundsCheckTol(_BoundsCheckTol){};
|
||||
BoundsCheckFreq(_BoundsCheckFreq){};
|
||||
};
|
||||
|
||||
/*Action parameters for the generalized rational action
|
||||
The approximation is for (M^dag M)^{1/inv_pow}
|
||||
where inv_pow is the denominator of the fractional power.
|
||||
Default inv_pow=2 for square root, making this equivalent to
|
||||
the OneFlavourRational action
|
||||
*/
|
||||
struct RationalActionParams : Serializable {
|
||||
GRID_SERIALIZABLE_CLASS_MEMBERS(RationalActionParams,
|
||||
int, inv_pow,
|
||||
RealD, lo, //low eigenvalue bound of rational approx
|
||||
RealD, hi, //high eigenvalue bound of rational approx
|
||||
int, MaxIter, //maximum iterations in msCG
|
||||
RealD, action_tolerance, //msCG tolerance in action evaluation
|
||||
int, action_degree, //rational approx tolerance in action evaluation
|
||||
RealD, md_tolerance, //msCG tolerance in MD integration
|
||||
int, md_degree, //rational approx tolerance in MD integration
|
||||
int, precision, //precision of floating point arithmetic
|
||||
int, BoundsCheckFreq); //frequency the approximation is tested (with Metropolis degree/tolerance); 0 disables the check
|
||||
// constructor
|
||||
RationalActionParams(int _inv_pow = 2,
|
||||
RealD _lo = 0.0,
|
||||
RealD _hi = 1.0,
|
||||
int _maxit = 1000,
|
||||
RealD _action_tolerance = 1.0e-8,
|
||||
int _action_degree = 10,
|
||||
RealD _md_tolerance = 1.0e-8,
|
||||
int _md_degree = 10,
|
||||
int _precision = 64,
|
||||
int _BoundsCheckFreq=20)
|
||||
: inv_pow(_inv_pow),
|
||||
lo(_lo),
|
||||
hi(_hi),
|
||||
MaxIter(_maxit),
|
||||
action_tolerance(_action_tolerance),
|
||||
action_degree(_action_degree),
|
||||
md_tolerance(_md_tolerance),
|
||||
md_degree(_md_degree),
|
||||
precision(_precision),
|
||||
BoundsCheckFreq(_BoundsCheckFreq){};
|
||||
};
|
||||
|
||||
|
||||
NAMESPACE_END(Grid);
|
||||
|
||||
#endif
|
||||
|
||||
@ -68,17 +68,9 @@ public:
|
||||
///////////////////////////////////////////////////////////////
|
||||
// Support for MADWF tricks
|
||||
///////////////////////////////////////////////////////////////
|
||||
RealD Mass(void) { return (mass_plus + mass_minus) / 2.0; };
|
||||
RealD MassPlus(void) { return mass_plus; };
|
||||
RealD MassMinus(void) { return mass_minus; };
|
||||
|
||||
RealD Mass(void) { return mass; };
|
||||
void SetMass(RealD _mass) {
|
||||
mass_plus=mass_minus=_mass;
|
||||
SetCoefficientsInternal(_zolo_hi,_gamma,_b,_c); // Reset coeffs
|
||||
} ;
|
||||
void SetMass(RealD _mass_plus, RealD _mass_minus) {
|
||||
mass_plus=_mass_plus;
|
||||
mass_minus=_mass_minus;
|
||||
mass=_mass;
|
||||
SetCoefficientsInternal(_zolo_hi,_gamma,_b,_c); // Reset coeffs
|
||||
} ;
|
||||
void P(const FermionField &psi, FermionField &chi);
|
||||
@ -116,7 +108,7 @@ public:
|
||||
void MeooeDag5D (const FermionField &in, FermionField &out);
|
||||
|
||||
// protected:
|
||||
RealD mass_plus, mass_minus;
|
||||
RealD mass;
|
||||
|
||||
// Save arguments to SetCoefficientsInternal
|
||||
Vector<Coeff_t> _gamma;
|
||||
|
||||
@ -1,435 +0,0 @@
|
||||
/*************************************************************************************
|
||||
|
||||
Grid physics library, www.github.com/paboyle/Grid
|
||||
|
||||
Source file: ./lib/qcd/action/fermion/WilsonCloverFermionImplementation.h
|
||||
|
||||
Copyright (C) 2017 - 2022
|
||||
|
||||
Author: paboyle <paboyle@ph.ed.ac.uk>
|
||||
Author: Daniel Richtmann <daniel.richtmann@gmail.com>
|
||||
Author: Mattia Bruno <mattia.bruno@cern.ch>
|
||||
|
||||
This program is free software; you can redistribute it and/or modify
|
||||
it under the terms of the GNU General Public License as published by
|
||||
the Free Software Foundation; either version 2 of the License, or
|
||||
(at your option) any later version.
|
||||
|
||||
This program is distributed in the hope that it will be useful,
|
||||
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||||
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
||||
GNU General Public License for more details.
|
||||
|
||||
You should have received a copy of the GNU General Public License along
|
||||
with this program; if not, write to the Free Software Foundation, Inc.,
|
||||
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
|
||||
|
||||
See the full license in the file "LICENSE" in the top level distribution directory
|
||||
*************************************************************************************/
|
||||
/* END LEGAL */
|
||||
|
||||
#pragma once
|
||||
|
||||
#include <Grid/Grid.h>
|
||||
#include <Grid/qcd/spin/Dirac.h>
|
||||
#include <Grid/qcd/action/fermion/WilsonCloverHelpers.h>
|
||||
|
||||
////////////////////////////////////////////
|
||||
// Standard Clover
|
||||
// (4+m0) + csw * clover_term
|
||||
// Exp Clover
|
||||
// (4+m0) * exp(csw/(4+m0) clover_term)
|
||||
// = (4+m0) + csw * clover_term + ...
|
||||
////////////////////////////////////////////
|
||||
|
||||
NAMESPACE_BEGIN(Grid);
|
||||
|
||||
|
||||
//////////////////////////////////
|
||||
// Generic Standard Clover
|
||||
//////////////////////////////////
|
||||
|
||||
template<class Impl>
|
||||
class CloverHelpers: public WilsonCloverHelpers<Impl> {
|
||||
public:
|
||||
|
||||
INHERIT_IMPL_TYPES(Impl);
|
||||
INHERIT_CLOVER_TYPES(Impl);
|
||||
|
||||
typedef WilsonCloverHelpers<Impl> Helpers;
|
||||
|
||||
static void Instantiate(CloverField& CloverTerm, CloverField& CloverTermInv, RealD csw_t, RealD diag_mass) {
|
||||
GridBase *grid = CloverTerm.Grid();
|
||||
CloverTerm += diag_mass;
|
||||
|
||||
int lvol = grid->lSites();
|
||||
int DimRep = Impl::Dimension;
|
||||
{
|
||||
autoView(CTv,CloverTerm,CpuRead);
|
||||
autoView(CTIv,CloverTermInv,CpuWrite);
|
||||
thread_for(site, lvol, {
|
||||
Coordinate lcoor;
|
||||
grid->LocalIndexToLocalCoor(site, lcoor);
|
||||
Eigen::MatrixXcd EigenCloverOp = Eigen::MatrixXcd::Zero(Ns * DimRep, Ns * DimRep);
|
||||
Eigen::MatrixXcd EigenInvCloverOp = Eigen::MatrixXcd::Zero(Ns * DimRep, Ns * DimRep);
|
||||
typename SiteClover::scalar_object Qx = Zero(), Qxinv = Zero();
|
||||
peekLocalSite(Qx, CTv, lcoor);
|
||||
|
||||
for (int j = 0; j < Ns; j++)
|
||||
for (int k = 0; k < Ns; k++)
|
||||
for (int a = 0; a < DimRep; a++)
|
||||
for (int b = 0; b < DimRep; b++){
|
||||
auto zz = Qx()(j, k)(a, b);
|
||||
EigenCloverOp(a + j * DimRep, b + k * DimRep) = std::complex<double>(zz);
|
||||
}
|
||||
|
||||
EigenInvCloverOp = EigenCloverOp.inverse();
|
||||
for (int j = 0; j < Ns; j++)
|
||||
for (int k = 0; k < Ns; k++)
|
||||
for (int a = 0; a < DimRep; a++)
|
||||
for (int b = 0; b < DimRep; b++)
|
||||
Qxinv()(j, k)(a, b) = EigenInvCloverOp(a + j * DimRep, b + k * DimRep);
|
||||
pokeLocalSite(Qxinv, CTIv, lcoor);
|
||||
});
|
||||
}
|
||||
}
|
||||
|
||||
static GaugeLinkField Cmunu(std::vector<GaugeLinkField> &U, GaugeLinkField &lambda, int mu, int nu) {
|
||||
return Helpers::Cmunu(U, lambda, mu, nu);
|
||||
}
|
||||
|
||||
};
|
||||
|
||||
|
||||
//////////////////////////////////
|
||||
// Generic Exp Clover
|
||||
//////////////////////////////////
|
||||
|
||||
template<class Impl>
|
||||
class ExpCloverHelpers: public WilsonCloverHelpers<Impl> {
|
||||
public:
|
||||
|
||||
INHERIT_IMPL_TYPES(Impl);
|
||||
INHERIT_CLOVER_TYPES(Impl);
|
||||
|
||||
template <typename vtype> using iImplClover = iScalar<iMatrix<iMatrix<vtype, Impl::Dimension>, Ns>>;
|
||||
typedef WilsonCloverHelpers<Impl> Helpers;
|
||||
|
||||
// Can this be avoided?
|
||||
static void IdentityTimesC(const CloverField& in, RealD c) {
|
||||
int DimRep = Impl::Dimension;
|
||||
|
||||
autoView(in_v, in, AcceleratorWrite);
|
||||
|
||||
accelerator_for(ss, in.Grid()->oSites(), 1, {
|
||||
for (int sa=0; sa<Ns; sa++)
|
||||
for (int ca=0; ca<DimRep; ca++)
|
||||
in_v[ss]()(sa,sa)(ca,ca) = c;
|
||||
});
|
||||
}
|
||||
|
||||
static int getNMAX(RealD prec, RealD R) {
|
||||
/* compute stop condition for exponential */
|
||||
int NMAX=1;
|
||||
RealD cond=R*R/2.;
|
||||
|
||||
while (cond*std::exp(R)>prec) {
|
||||
NMAX++;
|
||||
cond*=R/(double)(NMAX+1);
|
||||
}
|
||||
return NMAX;
|
||||
}
|
||||
|
||||
static int getNMAX(Lattice<iImplClover<vComplexD>> &t, RealD R) {return getNMAX(1e-12,R);}
|
||||
static int getNMAX(Lattice<iImplClover<vComplexF>> &t, RealD R) {return getNMAX(1e-6,R);}
|
||||
|
||||
static void Instantiate(CloverField& Clover, CloverField& CloverInv, RealD csw_t, RealD diag_mass) {
|
||||
GridBase* grid = Clover.Grid();
|
||||
CloverField ExpClover(grid);
|
||||
|
||||
int NMAX = getNMAX(Clover, 3.*csw_t/diag_mass);
|
||||
|
||||
Clover *= (1.0/diag_mass);
|
||||
|
||||
// Taylor expansion, slow but generic
|
||||
// Horner scheme: a0 + a1 x + a2 x^2 + .. = a0 + x (a1 + x(...))
|
||||
// qN = cN
|
||||
// qn = cn + qn+1 X
|
||||
std::vector<RealD> cn(NMAX+1);
|
||||
cn[0] = 1.0;
|
||||
for (int i=1; i<=NMAX; i++)
|
||||
cn[i] = cn[i-1] / RealD(i);
|
||||
|
||||
ExpClover = Zero();
|
||||
IdentityTimesC(ExpClover, cn[NMAX]);
|
||||
for (int i=NMAX-1; i>=0; i--)
|
||||
ExpClover = ExpClover * Clover + cn[i];
|
||||
|
||||
// prepare inverse
|
||||
CloverInv = (-1.0)*Clover;
|
||||
|
||||
Clover = ExpClover * diag_mass;
|
||||
|
||||
ExpClover = Zero();
|
||||
IdentityTimesC(ExpClover, cn[NMAX]);
|
||||
for (int i=NMAX-1; i>=0; i--)
|
||||
ExpClover = ExpClover * CloverInv + cn[i];
|
||||
|
||||
CloverInv = ExpClover * (1.0/diag_mass);
|
||||
|
||||
}
|
||||
|
||||
static GaugeLinkField Cmunu(std::vector<GaugeLinkField> &U, GaugeLinkField &lambda, int mu, int nu) {
|
||||
assert(0);
|
||||
return lambda;
|
||||
}
|
||||
|
||||
};
|
||||
|
||||
|
||||
//////////////////////////////////
|
||||
// Compact Standard Clover
|
||||
//////////////////////////////////
|
||||
|
||||
|
||||
template<class Impl>
|
||||
class CompactCloverHelpers: public CompactWilsonCloverHelpers<Impl>,
|
||||
public WilsonCloverHelpers<Impl> {
|
||||
public:
|
||||
|
||||
INHERIT_IMPL_TYPES(Impl);
|
||||
INHERIT_CLOVER_TYPES(Impl);
|
||||
INHERIT_COMPACT_CLOVER_TYPES(Impl);
|
||||
|
||||
typedef WilsonCloverHelpers<Impl> Helpers;
|
||||
typedef CompactWilsonCloverHelpers<Impl> CompactHelpers;
|
||||
|
||||
static void MassTerm(CloverField& Clover, RealD diag_mass) {
|
||||
Clover += diag_mass;
|
||||
}
|
||||
|
||||
static void Exponentiate_Clover(CloverDiagonalField& Diagonal,
|
||||
CloverTriangleField& Triangle,
|
||||
RealD csw_t, RealD diag_mass) {
|
||||
|
||||
// Do nothing
|
||||
}
|
||||
|
||||
// TODO: implement Cmunu for better performances with compact layout, but don't do it
|
||||
// here, but rather in WilsonCloverHelpers.h -> CompactWilsonCloverHelpers
|
||||
static GaugeLinkField Cmunu(std::vector<GaugeLinkField> &U, GaugeLinkField &lambda, int mu, int nu) {
|
||||
return Helpers::Cmunu(U, lambda, mu, nu);
|
||||
}
|
||||
};
|
||||
|
||||
//////////////////////////////////
|
||||
// Compact Exp Clover
|
||||
//////////////////////////////////
|
||||
|
||||
template<class Impl>
|
||||
class CompactExpCloverHelpers: public CompactWilsonCloverHelpers<Impl> {
|
||||
public:
|
||||
|
||||
INHERIT_IMPL_TYPES(Impl);
|
||||
INHERIT_CLOVER_TYPES(Impl);
|
||||
INHERIT_COMPACT_CLOVER_TYPES(Impl);
|
||||
|
||||
template <typename vtype> using iImplClover = iScalar<iMatrix<iMatrix<vtype, Impl::Dimension>, Ns>>;
|
||||
typedef CompactWilsonCloverHelpers<Impl> CompactHelpers;
|
||||
|
||||
static void MassTerm(CloverField& Clover, RealD diag_mass) {
|
||||
// do nothing!
|
||||
// mass term is multiplied to exp(Clover) below
|
||||
}
|
||||
|
||||
static int getNMAX(RealD prec, RealD R) {
|
||||
/* compute stop condition for exponential */
|
||||
int NMAX=1;
|
||||
RealD cond=R*R/2.;
|
||||
|
||||
while (cond*std::exp(R)>prec) {
|
||||
NMAX++;
|
||||
cond*=R/(double)(NMAX+1);
|
||||
}
|
||||
return NMAX;
|
||||
}
|
||||
|
||||
static int getNMAX(Lattice<iImplCloverDiagonal<vComplexD>> &t, RealD R) {return getNMAX(1e-12,R);}
|
||||
static int getNMAX(Lattice<iImplCloverDiagonal<vComplexF>> &t, RealD R) {return getNMAX(1e-6,R);}
|
||||
|
||||
static void ExponentiateHermitean6by6(const iMatrix<ComplexD,6> &arg, const RealD& alpha, const std::vector<RealD>& cN, const int Niter, iMatrix<ComplexD,6>& dest){
|
||||
|
||||
typedef iMatrix<ComplexD,6> mat;
|
||||
|
||||
RealD qn[6];
|
||||
RealD qnold[6];
|
||||
RealD p[5];
|
||||
RealD trA2, trA3, trA4;
|
||||
|
||||
mat A2, A3, A4, A5;
|
||||
A2 = alpha * alpha * arg * arg;
|
||||
A3 = alpha * arg * A2;
|
||||
A4 = A2 * A2;
|
||||
A5 = A2 * A3;
|
||||
|
||||
trA2 = toReal( trace(A2) );
|
||||
trA3 = toReal( trace(A3) );
|
||||
trA4 = toReal( trace(A4));
|
||||
|
||||
p[0] = toReal( trace(A3 * A3)) / 6.0 - 0.125 * trA4 * trA2 - trA3 * trA3 / 18.0 + trA2 * trA2 * trA2/ 48.0;
|
||||
p[1] = toReal( trace(A5)) / 5.0 - trA3 * trA2 / 6.0;
|
||||
p[2] = toReal( trace(A4)) / 4.0 - 0.125 * trA2 * trA2;
|
||||
p[3] = trA3 / 3.0;
|
||||
p[4] = 0.5 * trA2;
|
||||
|
||||
qnold[0] = cN[Niter];
|
||||
qnold[1] = 0.0;
|
||||
qnold[2] = 0.0;
|
||||
qnold[3] = 0.0;
|
||||
qnold[4] = 0.0;
|
||||
qnold[5] = 0.0;
|
||||
|
||||
for(int i = Niter-1; i >= 0; i--)
|
||||
{
|
||||
qn[0] = p[0] * qnold[5] + cN[i];
|
||||
qn[1] = p[1] * qnold[5] + qnold[0];
|
||||
qn[2] = p[2] * qnold[5] + qnold[1];
|
||||
qn[3] = p[3] * qnold[5] + qnold[2];
|
||||
qn[4] = p[4] * qnold[5] + qnold[3];
|
||||
qn[5] = qnold[4];
|
||||
|
||||
qnold[0] = qn[0];
|
||||
qnold[1] = qn[1];
|
||||
qnold[2] = qn[2];
|
||||
qnold[3] = qn[3];
|
||||
qnold[4] = qn[4];
|
||||
qnold[5] = qn[5];
|
||||
}
|
||||
|
||||
mat unit(1.0);
|
||||
|
||||
dest = (qn[0] * unit + qn[1] * alpha * arg + qn[2] * A2 + qn[3] * A3 + qn[4] * A4 + qn[5] * A5);
|
||||
|
||||
}
|
||||
|
||||
static void Exponentiate_Clover(CloverDiagonalField& Diagonal, CloverTriangleField& Triangle, RealD csw_t, RealD diag_mass) {
|
||||
|
||||
GridBase* grid = Diagonal.Grid();
|
||||
int NMAX = getNMAX(Diagonal, 3.*csw_t/diag_mass);
|
||||
|
||||
//
|
||||
// Implementation completely in Daniel's layout
|
||||
//
|
||||
|
||||
// Taylor expansion with Cayley-Hamilton recursion
|
||||
// underlying Horner scheme as above
|
||||
std::vector<RealD> cn(NMAX+1);
|
||||
cn[0] = 1.0;
|
||||
for (int i=1; i<=NMAX; i++){
|
||||
cn[i] = cn[i-1] / RealD(i);
|
||||
}
|
||||
|
||||
// Taken over from Daniel's implementation
|
||||
conformable(Diagonal, Triangle);
|
||||
|
||||
long lsites = grid->lSites();
|
||||
{
|
||||
typedef typename SiteCloverDiagonal::scalar_object scalar_object_diagonal;
|
||||
typedef typename SiteCloverTriangle::scalar_object scalar_object_triangle;
|
||||
typedef iMatrix<ComplexD,6> mat;
|
||||
|
||||
autoView(diagonal_v, Diagonal, CpuRead);
|
||||
autoView(triangle_v, Triangle, CpuRead);
|
||||
autoView(diagonalExp_v, Diagonal, CpuWrite);
|
||||
autoView(triangleExp_v, Triangle, CpuWrite);
|
||||
|
||||
thread_for(site, lsites, { // NOTE: Not on GPU because of (peek/poke)LocalSite
|
||||
|
||||
mat srcCloverOpUL(0.0); // upper left block
|
||||
mat srcCloverOpLR(0.0); // lower right block
|
||||
mat ExpCloverOp;
|
||||
|
||||
scalar_object_diagonal diagonal_tmp = Zero();
|
||||
scalar_object_diagonal diagonal_exp_tmp = Zero();
|
||||
scalar_object_triangle triangle_tmp = Zero();
|
||||
scalar_object_triangle triangle_exp_tmp = Zero();
|
||||
|
||||
Coordinate lcoor;
|
||||
grid->LocalIndexToLocalCoor(site, lcoor);
|
||||
|
||||
peekLocalSite(diagonal_tmp, diagonal_v, lcoor);
|
||||
peekLocalSite(triangle_tmp, triangle_v, lcoor);
|
||||
|
||||
int block;
|
||||
block = 0;
|
||||
for(int i = 0; i < 6; i++){
|
||||
for(int j = 0; j < 6; j++){
|
||||
if (i == j){
|
||||
srcCloverOpUL(i,j) = static_cast<ComplexD>(TensorRemove(diagonal_tmp()(block)(i)));
|
||||
}
|
||||
else{
|
||||
srcCloverOpUL(i,j) = static_cast<ComplexD>(TensorRemove(CompactHelpers::triangle_elem(triangle_tmp, block, i, j)));
|
||||
}
|
||||
}
|
||||
}
|
||||
block = 1;
|
||||
for(int i = 0; i < 6; i++){
|
||||
for(int j = 0; j < 6; j++){
|
||||
if (i == j){
|
||||
srcCloverOpLR(i,j) = static_cast<ComplexD>(TensorRemove(diagonal_tmp()(block)(i)));
|
||||
}
|
||||
else{
|
||||
srcCloverOpLR(i,j) = static_cast<ComplexD>(TensorRemove(CompactHelpers::triangle_elem(triangle_tmp, block, i, j)));
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// exp(Clover)
|
||||
|
||||
ExponentiateHermitean6by6(srcCloverOpUL,1.0/diag_mass,cn,NMAX,ExpCloverOp);
|
||||
|
||||
block = 0;
|
||||
for(int i = 0; i < 6; i++){
|
||||
for(int j = 0; j < 6; j++){
|
||||
if (i == j){
|
||||
diagonal_exp_tmp()(block)(i) = ExpCloverOp(i,j);
|
||||
}
|
||||
else if(i < j){
|
||||
triangle_exp_tmp()(block)(CompactHelpers::triangle_index(i, j)) = ExpCloverOp(i,j);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
ExponentiateHermitean6by6(srcCloverOpLR,1.0/diag_mass,cn,NMAX,ExpCloverOp);
|
||||
|
||||
block = 1;
|
||||
for(int i = 0; i < 6; i++){
|
||||
for(int j = 0; j < 6; j++){
|
||||
if (i == j){
|
||||
diagonal_exp_tmp()(block)(i) = ExpCloverOp(i,j);
|
||||
}
|
||||
else if(i < j){
|
||||
triangle_exp_tmp()(block)(CompactHelpers::triangle_index(i, j)) = ExpCloverOp(i,j);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
pokeLocalSite(diagonal_exp_tmp, diagonalExp_v, lcoor);
|
||||
pokeLocalSite(triangle_exp_tmp, triangleExp_v, lcoor);
|
||||
});
|
||||
}
|
||||
|
||||
Diagonal *= diag_mass;
|
||||
Triangle *= diag_mass;
|
||||
}
|
||||
|
||||
|
||||
static GaugeLinkField Cmunu(std::vector<GaugeLinkField> &U, GaugeLinkField &lambda, int mu, int nu) {
|
||||
assert(0);
|
||||
return lambda;
|
||||
}
|
||||
|
||||
};
|
||||
|
||||
|
||||
NAMESPACE_END(Grid);
|
||||
@ -1,241 +0,0 @@
|
||||
/*************************************************************************************
|
||||
|
||||
Grid physics library, www.github.com/paboyle/Grid
|
||||
|
||||
Source file: ./lib/qcd/action/fermion/CompactWilsonCloverFermion.h
|
||||
|
||||
Copyright (C) 2020 - 2022
|
||||
|
||||
Author: Daniel Richtmann <daniel.richtmann@gmail.com>
|
||||
Author: Nils Meyer <nils.meyer@ur.de>
|
||||
|
||||
This program is free software; you can redistribute it and/or modify
|
||||
it under the terms of the GNU General Public License as published by
|
||||
the Free Software Foundation; either version 2 of the License, or
|
||||
(at your option) any later version.
|
||||
|
||||
This program is distributed in the hope that it will be useful,
|
||||
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||||
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
||||
GNU General Public License for more details.
|
||||
|
||||
You should have received a copy of the GNU General Public License along
|
||||
with this program; if not, write to the Free Software Foundation, Inc.,
|
||||
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
|
||||
|
||||
See the full license in the file "LICENSE" in the top level distribution directory
|
||||
*************************************************************************************/
|
||||
/* END LEGAL */
|
||||
|
||||
#pragma once
|
||||
|
||||
#include <Grid/qcd/action/fermion/WilsonCloverTypes.h>
|
||||
#include <Grid/qcd/action/fermion/WilsonCloverHelpers.h>
|
||||
#include <Grid/qcd/action/fermion/CloverHelpers.h>
|
||||
|
||||
NAMESPACE_BEGIN(Grid);
|
||||
|
||||
// see Grid/qcd/action/fermion/WilsonCloverFermion.h for description
|
||||
//
|
||||
// Modifications done here:
|
||||
//
|
||||
// Original: clover term = 12x12 matrix per site
|
||||
//
|
||||
// But: Only two diagonal 6x6 hermitian blocks are non-zero (also true for original, verified by running)
|
||||
// Sufficient to store/transfer only the real parts of the diagonal and one triangular part
|
||||
// 2 * (6 + 15 * 2) = 72 real or 36 complex words to be stored/transfered
|
||||
//
|
||||
// Here: Above but diagonal as complex numbers, i.e., need to store/transfer
|
||||
// 2 * (6 * 2 + 15 * 2) = 84 real or 42 complex words
|
||||
//
|
||||
// Words per site and improvement compared to original (combined with the input and output spinors):
|
||||
//
|
||||
// - Original: 2*12 + 12*12 = 168 words -> 1.00 x less
|
||||
// - Minimal: 2*12 + 36 = 60 words -> 2.80 x less
|
||||
// - Here: 2*12 + 42 = 66 words -> 2.55 x less
|
||||
//
|
||||
// These improvements directly translate to wall-clock time
|
||||
//
|
||||
// Data layout:
|
||||
//
|
||||
// - diagonal and triangle part as separate lattice fields,
|
||||
// this was faster than as 1 combined field on all tested machines
|
||||
// - diagonal: as expected
|
||||
// - triangle: store upper right triangle in row major order
|
||||
// - graphical:
|
||||
// 0 1 2 3 4
|
||||
// 5 6 7 8
|
||||
// 9 10 11 = upper right triangle indices
|
||||
// 12 13
|
||||
// 14
|
||||
// 0
|
||||
// 1
|
||||
// 2
|
||||
// 3 = diagonal indices
|
||||
// 4
|
||||
// 5
|
||||
// 0
|
||||
// 1 5
|
||||
// 2 6 9 = lower left triangle indices
|
||||
// 3 7 10 12
|
||||
// 4 8 11 13 14
|
||||
//
|
||||
// Impact on total memory consumption:
|
||||
// - Original: (2 * 1 + 8 * 1/2) 12x12 matrices = 6 12x12 matrices = 864 complex words per site
|
||||
// - Here: (2 * 1 + 4 * 1/2) diagonal parts = 4 diagonal parts = 24 complex words per site
|
||||
// + (2 * 1 + 4 * 1/2) triangle parts = 4 triangle parts = 60 complex words per site
|
||||
// = 84 complex words per site
|
||||
|
||||
template<class Impl, class CloverHelpers>
|
||||
class CompactWilsonCloverFermion : public WilsonFermion<Impl>,
|
||||
public WilsonCloverHelpers<Impl>,
|
||||
public CompactWilsonCloverHelpers<Impl> {
|
||||
/////////////////////////////////////////////
|
||||
// Sizes
|
||||
/////////////////////////////////////////////
|
||||
|
||||
public:
|
||||
|
||||
INHERIT_COMPACT_CLOVER_SIZES(Impl);
|
||||
|
||||
/////////////////////////////////////////////
|
||||
// Type definitions
|
||||
/////////////////////////////////////////////
|
||||
|
||||
public:
|
||||
|
||||
INHERIT_IMPL_TYPES(Impl);
|
||||
INHERIT_CLOVER_TYPES(Impl);
|
||||
INHERIT_COMPACT_CLOVER_TYPES(Impl);
|
||||
|
||||
typedef WilsonFermion<Impl> WilsonBase;
|
||||
typedef WilsonCloverHelpers<Impl> Helpers;
|
||||
typedef CompactWilsonCloverHelpers<Impl> CompactHelpers;
|
||||
|
||||
/////////////////////////////////////////////
|
||||
// Constructors
|
||||
/////////////////////////////////////////////
|
||||
|
||||
public:
|
||||
|
||||
CompactWilsonCloverFermion(GaugeField& _Umu,
|
||||
GridCartesian& Fgrid,
|
||||
GridRedBlackCartesian& Hgrid,
|
||||
const RealD _mass,
|
||||
const RealD _csw_r = 0.0,
|
||||
const RealD _csw_t = 0.0,
|
||||
const RealD _cF = 1.0,
|
||||
const WilsonAnisotropyCoefficients& clover_anisotropy = WilsonAnisotropyCoefficients(),
|
||||
const ImplParams& impl_p = ImplParams());
|
||||
|
||||
/////////////////////////////////////////////
|
||||
// Member functions (implementing interface)
|
||||
/////////////////////////////////////////////
|
||||
|
||||
public:
|
||||
|
||||
virtual void Instantiatable() {};
|
||||
int ConstEE() override { return 0; };
|
||||
int isTrivialEE() override { return 0; };
|
||||
|
||||
void Dhop(const FermionField& in, FermionField& out, int dag) override;
|
||||
|
||||
void DhopOE(const FermionField& in, FermionField& out, int dag) override;
|
||||
|
||||
void DhopEO(const FermionField& in, FermionField& out, int dag) override;
|
||||
|
||||
void DhopDir(const FermionField& in, FermionField& out, int dir, int disp) override;
|
||||
|
||||
void DhopDirAll(const FermionField& in, std::vector<FermionField>& out) /* override */;
|
||||
|
||||
void M(const FermionField& in, FermionField& out) override;
|
||||
|
||||
void Mdag(const FermionField& in, FermionField& out) override;
|
||||
|
||||
void Meooe(const FermionField& in, FermionField& out) override;
|
||||
|
||||
void MeooeDag(const FermionField& in, FermionField& out) override;
|
||||
|
||||
void Mooee(const FermionField& in, FermionField& out) override;
|
||||
|
||||
void MooeeDag(const FermionField& in, FermionField& out) override;
|
||||
|
||||
void MooeeInv(const FermionField& in, FermionField& out) override;
|
||||
|
||||
void MooeeInvDag(const FermionField& in, FermionField& out) override;
|
||||
|
||||
void Mdir(const FermionField& in, FermionField& out, int dir, int disp) override;
|
||||
|
||||
void MdirAll(const FermionField& in, std::vector<FermionField>& out) override;
|
||||
|
||||
void MDeriv(GaugeField& force, const FermionField& X, const FermionField& Y, int dag) override;
|
||||
|
||||
void MooDeriv(GaugeField& mat, const FermionField& U, const FermionField& V, int dag) override;
|
||||
|
||||
void MeeDeriv(GaugeField& mat, const FermionField& U, const FermionField& V, int dag) override;
|
||||
|
||||
/////////////////////////////////////////////
|
||||
// Member functions (internals)
|
||||
/////////////////////////////////////////////
|
||||
|
||||
void MooeeInternal(const FermionField& in,
|
||||
FermionField& out,
|
||||
const CloverDiagonalField& diagonal,
|
||||
const CloverTriangleField& triangle);
|
||||
|
||||
/////////////////////////////////////////////
|
||||
// Helpers
|
||||
/////////////////////////////////////////////
|
||||
|
||||
void ImportGauge(const GaugeField& _Umu) override;
|
||||
|
||||
/////////////////////////////////////////////
|
||||
// Helpers
|
||||
/////////////////////////////////////////////
|
||||
|
||||
private:
|
||||
|
||||
template<class Field>
|
||||
const MaskField* getCorrectMaskField(const Field &in) const {
|
||||
if(in.Grid()->_isCheckerBoarded) {
|
||||
if(in.Checkerboard() == Odd) {
|
||||
return &this->BoundaryMaskOdd;
|
||||
} else {
|
||||
return &this->BoundaryMaskEven;
|
||||
}
|
||||
} else {
|
||||
return &this->BoundaryMask;
|
||||
}
|
||||
}
|
||||
|
||||
template<class Field>
|
||||
void ApplyBoundaryMask(Field& f) {
|
||||
const MaskField* m = getCorrectMaskField(f); assert(m != nullptr);
|
||||
assert(m != nullptr);
|
||||
CompactHelpers::ApplyBoundaryMask(f, *m);
|
||||
}
|
||||
|
||||
/////////////////////////////////////////////
|
||||
// Member Data
|
||||
/////////////////////////////////////////////
|
||||
|
||||
public:
|
||||
|
||||
RealD csw_r;
|
||||
RealD csw_t;
|
||||
RealD cF;
|
||||
|
||||
bool open_boundaries;
|
||||
|
||||
CloverDiagonalField Diagonal, DiagonalEven, DiagonalOdd;
|
||||
CloverDiagonalField DiagonalInv, DiagonalInvEven, DiagonalInvOdd;
|
||||
|
||||
CloverTriangleField Triangle, TriangleEven, TriangleOdd;
|
||||
CloverTriangleField TriangleInv, TriangleInvEven, TriangleInvOdd;
|
||||
|
||||
FermionField Tmp;
|
||||
|
||||
MaskField BoundaryMask, BoundaryMaskEven, BoundaryMaskOdd;
|
||||
};
|
||||
|
||||
NAMESPACE_END(Grid);
|
||||
@ -53,7 +53,6 @@ NAMESPACE_CHECK(Wilson);
|
||||
#include <Grid/qcd/action/fermion/WilsonTMFermion.h> // 4d wilson like
|
||||
NAMESPACE_CHECK(WilsonTM);
|
||||
#include <Grid/qcd/action/fermion/WilsonCloverFermion.h> // 4d wilson clover fermions
|
||||
#include <Grid/qcd/action/fermion/CompactWilsonCloverFermion.h> // 4d compact wilson clover fermions
|
||||
NAMESPACE_CHECK(WilsonClover);
|
||||
#include <Grid/qcd/action/fermion/WilsonFermion5D.h> // 5d base used by all 5d overlap types
|
||||
NAMESPACE_CHECK(Wilson5D);
|
||||
@ -116,9 +115,9 @@ typedef WilsonFermion<WilsonImplR> WilsonFermionR;
|
||||
typedef WilsonFermion<WilsonImplF> WilsonFermionF;
|
||||
typedef WilsonFermion<WilsonImplD> WilsonFermionD;
|
||||
|
||||
//typedef WilsonFermion<WilsonImplRL> WilsonFermionRL;
|
||||
//typedef WilsonFermion<WilsonImplFH> WilsonFermionFH;
|
||||
//typedef WilsonFermion<WilsonImplDF> WilsonFermionDF;
|
||||
typedef WilsonFermion<WilsonImplRL> WilsonFermionRL;
|
||||
typedef WilsonFermion<WilsonImplFH> WilsonFermionFH;
|
||||
typedef WilsonFermion<WilsonImplDF> WilsonFermionDF;
|
||||
|
||||
typedef WilsonFermion<WilsonAdjImplR> WilsonAdjFermionR;
|
||||
typedef WilsonFermion<WilsonAdjImplF> WilsonAdjFermionF;
|
||||
@ -138,93 +137,62 @@ typedef WilsonTMFermion<WilsonImplF> WilsonTMFermionF;
|
||||
typedef WilsonTMFermion<WilsonImplD> WilsonTMFermionD;
|
||||
|
||||
// Clover fermions
|
||||
template <typename WImpl> using WilsonClover = WilsonCloverFermion<WImpl, CloverHelpers<WImpl>>;
|
||||
template <typename WImpl> using WilsonExpClover = WilsonCloverFermion<WImpl, ExpCloverHelpers<WImpl>>;
|
||||
typedef WilsonCloverFermion<WilsonImplR> WilsonCloverFermionR;
|
||||
typedef WilsonCloverFermion<WilsonImplF> WilsonCloverFermionF;
|
||||
typedef WilsonCloverFermion<WilsonImplD> WilsonCloverFermionD;
|
||||
|
||||
typedef WilsonClover<WilsonImplR> WilsonCloverFermionR;
|
||||
typedef WilsonClover<WilsonImplF> WilsonCloverFermionF;
|
||||
typedef WilsonClover<WilsonImplD> WilsonCloverFermionD;
|
||||
typedef WilsonCloverFermion<WilsonAdjImplR> WilsonCloverAdjFermionR;
|
||||
typedef WilsonCloverFermion<WilsonAdjImplF> WilsonCloverAdjFermionF;
|
||||
typedef WilsonCloverFermion<WilsonAdjImplD> WilsonCloverAdjFermionD;
|
||||
|
||||
typedef WilsonExpClover<WilsonImplR> WilsonExpCloverFermionR;
|
||||
typedef WilsonExpClover<WilsonImplF> WilsonExpCloverFermionF;
|
||||
typedef WilsonExpClover<WilsonImplD> WilsonExpCloverFermionD;
|
||||
typedef WilsonCloverFermion<WilsonTwoIndexSymmetricImplR> WilsonCloverTwoIndexSymmetricFermionR;
|
||||
typedef WilsonCloverFermion<WilsonTwoIndexSymmetricImplF> WilsonCloverTwoIndexSymmetricFermionF;
|
||||
typedef WilsonCloverFermion<WilsonTwoIndexSymmetricImplD> WilsonCloverTwoIndexSymmetricFermionD;
|
||||
|
||||
typedef WilsonClover<WilsonAdjImplR> WilsonCloverAdjFermionR;
|
||||
typedef WilsonClover<WilsonAdjImplF> WilsonCloverAdjFermionF;
|
||||
typedef WilsonClover<WilsonAdjImplD> WilsonCloverAdjFermionD;
|
||||
|
||||
typedef WilsonClover<WilsonTwoIndexSymmetricImplR> WilsonCloverTwoIndexSymmetricFermionR;
|
||||
typedef WilsonClover<WilsonTwoIndexSymmetricImplF> WilsonCloverTwoIndexSymmetricFermionF;
|
||||
typedef WilsonClover<WilsonTwoIndexSymmetricImplD> WilsonCloverTwoIndexSymmetricFermionD;
|
||||
|
||||
typedef WilsonClover<WilsonTwoIndexAntiSymmetricImplR> WilsonCloverTwoIndexAntiSymmetricFermionR;
|
||||
typedef WilsonClover<WilsonTwoIndexAntiSymmetricImplF> WilsonCloverTwoIndexAntiSymmetricFermionF;
|
||||
typedef WilsonClover<WilsonTwoIndexAntiSymmetricImplD> WilsonCloverTwoIndexAntiSymmetricFermionD;
|
||||
|
||||
// Compact Clover fermions
|
||||
template <typename WImpl> using CompactWilsonClover = CompactWilsonCloverFermion<WImpl, CompactCloverHelpers<WImpl>>;
|
||||
template <typename WImpl> using CompactWilsonExpClover = CompactWilsonCloverFermion<WImpl, CompactExpCloverHelpers<WImpl>>;
|
||||
|
||||
typedef CompactWilsonClover<WilsonImplR> CompactWilsonCloverFermionR;
|
||||
typedef CompactWilsonClover<WilsonImplF> CompactWilsonCloverFermionF;
|
||||
typedef CompactWilsonClover<WilsonImplD> CompactWilsonCloverFermionD;
|
||||
|
||||
typedef CompactWilsonExpClover<WilsonImplR> CompactWilsonExpCloverFermionR;
|
||||
typedef CompactWilsonExpClover<WilsonImplF> CompactWilsonExpCloverFermionF;
|
||||
typedef CompactWilsonExpClover<WilsonImplD> CompactWilsonExpCloverFermionD;
|
||||
|
||||
typedef CompactWilsonClover<WilsonAdjImplR> CompactWilsonCloverAdjFermionR;
|
||||
typedef CompactWilsonClover<WilsonAdjImplF> CompactWilsonCloverAdjFermionF;
|
||||
typedef CompactWilsonClover<WilsonAdjImplD> CompactWilsonCloverAdjFermionD;
|
||||
|
||||
typedef CompactWilsonClover<WilsonTwoIndexSymmetricImplR> CompactWilsonCloverTwoIndexSymmetricFermionR;
|
||||
typedef CompactWilsonClover<WilsonTwoIndexSymmetricImplF> CompactWilsonCloverTwoIndexSymmetricFermionF;
|
||||
typedef CompactWilsonClover<WilsonTwoIndexSymmetricImplD> CompactWilsonCloverTwoIndexSymmetricFermionD;
|
||||
|
||||
typedef CompactWilsonClover<WilsonTwoIndexAntiSymmetricImplR> CompactWilsonCloverTwoIndexAntiSymmetricFermionR;
|
||||
typedef CompactWilsonClover<WilsonTwoIndexAntiSymmetricImplF> CompactWilsonCloverTwoIndexAntiSymmetricFermionF;
|
||||
typedef CompactWilsonClover<WilsonTwoIndexAntiSymmetricImplD> CompactWilsonCloverTwoIndexAntiSymmetricFermionD;
|
||||
typedef WilsonCloverFermion<WilsonTwoIndexAntiSymmetricImplR> WilsonCloverTwoIndexAntiSymmetricFermionR;
|
||||
typedef WilsonCloverFermion<WilsonTwoIndexAntiSymmetricImplF> WilsonCloverTwoIndexAntiSymmetricFermionF;
|
||||
typedef WilsonCloverFermion<WilsonTwoIndexAntiSymmetricImplD> WilsonCloverTwoIndexAntiSymmetricFermionD;
|
||||
|
||||
// Domain Wall fermions
|
||||
typedef DomainWallFermion<WilsonImplR> DomainWallFermionR;
|
||||
typedef DomainWallFermion<WilsonImplF> DomainWallFermionF;
|
||||
typedef DomainWallFermion<WilsonImplD> DomainWallFermionD;
|
||||
|
||||
//typedef DomainWallFermion<WilsonImplRL> DomainWallFermionRL;
|
||||
//typedef DomainWallFermion<WilsonImplFH> DomainWallFermionFH;
|
||||
//typedef DomainWallFermion<WilsonImplDF> DomainWallFermionDF;
|
||||
typedef DomainWallFermion<WilsonImplRL> DomainWallFermionRL;
|
||||
typedef DomainWallFermion<WilsonImplFH> DomainWallFermionFH;
|
||||
typedef DomainWallFermion<WilsonImplDF> DomainWallFermionDF;
|
||||
|
||||
typedef DomainWallEOFAFermion<WilsonImplR> DomainWallEOFAFermionR;
|
||||
typedef DomainWallEOFAFermion<WilsonImplF> DomainWallEOFAFermionF;
|
||||
typedef DomainWallEOFAFermion<WilsonImplD> DomainWallEOFAFermionD;
|
||||
|
||||
//typedef DomainWallEOFAFermion<WilsonImplRL> DomainWallEOFAFermionRL;
|
||||
//typedef DomainWallEOFAFermion<WilsonImplFH> DomainWallEOFAFermionFH;
|
||||
//typedef DomainWallEOFAFermion<WilsonImplDF> DomainWallEOFAFermionDF;
|
||||
typedef DomainWallEOFAFermion<WilsonImplRL> DomainWallEOFAFermionRL;
|
||||
typedef DomainWallEOFAFermion<WilsonImplFH> DomainWallEOFAFermionFH;
|
||||
typedef DomainWallEOFAFermion<WilsonImplDF> DomainWallEOFAFermionDF;
|
||||
|
||||
typedef MobiusFermion<WilsonImplR> MobiusFermionR;
|
||||
typedef MobiusFermion<WilsonImplF> MobiusFermionF;
|
||||
typedef MobiusFermion<WilsonImplD> MobiusFermionD;
|
||||
|
||||
//typedef MobiusFermion<WilsonImplRL> MobiusFermionRL;
|
||||
//typedef MobiusFermion<WilsonImplFH> MobiusFermionFH;
|
||||
//typedef MobiusFermion<WilsonImplDF> MobiusFermionDF;
|
||||
typedef MobiusFermion<WilsonImplRL> MobiusFermionRL;
|
||||
typedef MobiusFermion<WilsonImplFH> MobiusFermionFH;
|
||||
typedef MobiusFermion<WilsonImplDF> MobiusFermionDF;
|
||||
|
||||
typedef MobiusEOFAFermion<WilsonImplR> MobiusEOFAFermionR;
|
||||
typedef MobiusEOFAFermion<WilsonImplF> MobiusEOFAFermionF;
|
||||
typedef MobiusEOFAFermion<WilsonImplD> MobiusEOFAFermionD;
|
||||
|
||||
//typedef MobiusEOFAFermion<WilsonImplRL> MobiusEOFAFermionRL;
|
||||
//typedef MobiusEOFAFermion<WilsonImplFH> MobiusEOFAFermionFH;
|
||||
//typedef MobiusEOFAFermion<WilsonImplDF> MobiusEOFAFermionDF;
|
||||
typedef MobiusEOFAFermion<WilsonImplRL> MobiusEOFAFermionRL;
|
||||
typedef MobiusEOFAFermion<WilsonImplFH> MobiusEOFAFermionFH;
|
||||
typedef MobiusEOFAFermion<WilsonImplDF> MobiusEOFAFermionDF;
|
||||
|
||||
typedef ZMobiusFermion<ZWilsonImplR> ZMobiusFermionR;
|
||||
typedef ZMobiusFermion<ZWilsonImplF> ZMobiusFermionF;
|
||||
typedef ZMobiusFermion<ZWilsonImplD> ZMobiusFermionD;
|
||||
|
||||
//typedef ZMobiusFermion<ZWilsonImplRL> ZMobiusFermionRL;
|
||||
//typedef ZMobiusFermion<ZWilsonImplFH> ZMobiusFermionFH;
|
||||
//typedef ZMobiusFermion<ZWilsonImplDF> ZMobiusFermionDF;
|
||||
typedef ZMobiusFermion<ZWilsonImplRL> ZMobiusFermionRL;
|
||||
typedef ZMobiusFermion<ZWilsonImplFH> ZMobiusFermionFH;
|
||||
typedef ZMobiusFermion<ZWilsonImplDF> ZMobiusFermionDF;
|
||||
|
||||
// Ls vectorised
|
||||
typedef ScaledShamirFermion<WilsonImplR> ScaledShamirFermionR;
|
||||
@ -267,49 +235,49 @@ typedef WilsonFermion<GparityWilsonImplR> GparityWilsonFermionR;
|
||||
typedef WilsonFermion<GparityWilsonImplF> GparityWilsonFermionF;
|
||||
typedef WilsonFermion<GparityWilsonImplD> GparityWilsonFermionD;
|
||||
|
||||
//typedef WilsonFermion<GparityWilsonImplRL> GparityWilsonFermionRL;
|
||||
//typedef WilsonFermion<GparityWilsonImplFH> GparityWilsonFermionFH;
|
||||
//typedef WilsonFermion<GparityWilsonImplDF> GparityWilsonFermionDF;
|
||||
typedef WilsonFermion<GparityWilsonImplRL> GparityWilsonFermionRL;
|
||||
typedef WilsonFermion<GparityWilsonImplFH> GparityWilsonFermionFH;
|
||||
typedef WilsonFermion<GparityWilsonImplDF> GparityWilsonFermionDF;
|
||||
|
||||
typedef DomainWallFermion<GparityWilsonImplR> GparityDomainWallFermionR;
|
||||
typedef DomainWallFermion<GparityWilsonImplF> GparityDomainWallFermionF;
|
||||
typedef DomainWallFermion<GparityWilsonImplD> GparityDomainWallFermionD;
|
||||
|
||||
//typedef DomainWallFermion<GparityWilsonImplRL> GparityDomainWallFermionRL;
|
||||
//typedef DomainWallFermion<GparityWilsonImplFH> GparityDomainWallFermionFH;
|
||||
//typedef DomainWallFermion<GparityWilsonImplDF> GparityDomainWallFermionDF;
|
||||
typedef DomainWallFermion<GparityWilsonImplRL> GparityDomainWallFermionRL;
|
||||
typedef DomainWallFermion<GparityWilsonImplFH> GparityDomainWallFermionFH;
|
||||
typedef DomainWallFermion<GparityWilsonImplDF> GparityDomainWallFermionDF;
|
||||
|
||||
typedef DomainWallEOFAFermion<GparityWilsonImplR> GparityDomainWallEOFAFermionR;
|
||||
typedef DomainWallEOFAFermion<GparityWilsonImplF> GparityDomainWallEOFAFermionF;
|
||||
typedef DomainWallEOFAFermion<GparityWilsonImplD> GparityDomainWallEOFAFermionD;
|
||||
|
||||
//typedef DomainWallEOFAFermion<GparityWilsonImplRL> GparityDomainWallEOFAFermionRL;
|
||||
//typedef DomainWallEOFAFermion<GparityWilsonImplFH> GparityDomainWallEOFAFermionFH;
|
||||
//typedef DomainWallEOFAFermion<GparityWilsonImplDF> GparityDomainWallEOFAFermionDF;
|
||||
typedef DomainWallEOFAFermion<GparityWilsonImplRL> GparityDomainWallEOFAFermionRL;
|
||||
typedef DomainWallEOFAFermion<GparityWilsonImplFH> GparityDomainWallEOFAFermionFH;
|
||||
typedef DomainWallEOFAFermion<GparityWilsonImplDF> GparityDomainWallEOFAFermionDF;
|
||||
|
||||
typedef WilsonTMFermion<GparityWilsonImplR> GparityWilsonTMFermionR;
|
||||
typedef WilsonTMFermion<GparityWilsonImplF> GparityWilsonTMFermionF;
|
||||
typedef WilsonTMFermion<GparityWilsonImplD> GparityWilsonTMFermionD;
|
||||
|
||||
//typedef WilsonTMFermion<GparityWilsonImplRL> GparityWilsonTMFermionRL;
|
||||
//typedef WilsonTMFermion<GparityWilsonImplFH> GparityWilsonTMFermionFH;
|
||||
//typedef WilsonTMFermion<GparityWilsonImplDF> GparityWilsonTMFermionDF;
|
||||
typedef WilsonTMFermion<GparityWilsonImplRL> GparityWilsonTMFermionRL;
|
||||
typedef WilsonTMFermion<GparityWilsonImplFH> GparityWilsonTMFermionFH;
|
||||
typedef WilsonTMFermion<GparityWilsonImplDF> GparityWilsonTMFermionDF;
|
||||
|
||||
typedef MobiusFermion<GparityWilsonImplR> GparityMobiusFermionR;
|
||||
typedef MobiusFermion<GparityWilsonImplF> GparityMobiusFermionF;
|
||||
typedef MobiusFermion<GparityWilsonImplD> GparityMobiusFermionD;
|
||||
|
||||
//typedef MobiusFermion<GparityWilsonImplRL> GparityMobiusFermionRL;
|
||||
//typedef MobiusFermion<GparityWilsonImplFH> GparityMobiusFermionFH;
|
||||
//typedef MobiusFermion<GparityWilsonImplDF> GparityMobiusFermionDF;
|
||||
typedef MobiusFermion<GparityWilsonImplRL> GparityMobiusFermionRL;
|
||||
typedef MobiusFermion<GparityWilsonImplFH> GparityMobiusFermionFH;
|
||||
typedef MobiusFermion<GparityWilsonImplDF> GparityMobiusFermionDF;
|
||||
|
||||
typedef MobiusEOFAFermion<GparityWilsonImplR> GparityMobiusEOFAFermionR;
|
||||
typedef MobiusEOFAFermion<GparityWilsonImplF> GparityMobiusEOFAFermionF;
|
||||
typedef MobiusEOFAFermion<GparityWilsonImplD> GparityMobiusEOFAFermionD;
|
||||
|
||||
//typedef MobiusEOFAFermion<GparityWilsonImplRL> GparityMobiusEOFAFermionRL;
|
||||
//typedef MobiusEOFAFermion<GparityWilsonImplFH> GparityMobiusEOFAFermionFH;
|
||||
//typedef MobiusEOFAFermion<GparityWilsonImplDF> GparityMobiusEOFAFermionDF;
|
||||
typedef MobiusEOFAFermion<GparityWilsonImplRL> GparityMobiusEOFAFermionRL;
|
||||
typedef MobiusEOFAFermion<GparityWilsonImplFH> GparityMobiusEOFAFermionFH;
|
||||
typedef MobiusEOFAFermion<GparityWilsonImplDF> GparityMobiusEOFAFermionDF;
|
||||
|
||||
typedef ImprovedStaggeredFermion<StaggeredImplR> ImprovedStaggeredFermionR;
|
||||
typedef ImprovedStaggeredFermion<StaggeredImplF> ImprovedStaggeredFermionF;
|
||||
|
||||
@ -49,8 +49,6 @@ public:
|
||||
|
||||
virtual FermionField &tmp(void) = 0;
|
||||
|
||||
virtual void DirichletBlock(const Coordinate & _Block) { assert(0); };
|
||||
|
||||
GridBase * Grid(void) { return FermionGrid(); }; // this is all the linalg routines need to know
|
||||
GridBase * RedBlackGrid(void) { return FermionRedBlackGrid(); };
|
||||
|
||||
|
||||
@ -30,18 +30,6 @@ directory
|
||||
|
||||
NAMESPACE_BEGIN(Grid);
|
||||
|
||||
/*
|
||||
Policy implementation for G-parity boundary conditions
|
||||
|
||||
Rather than treating the gauge field as a flavored field, the Grid implementation of G-parity treats the gauge field as a regular
|
||||
field with complex conjugate boundary conditions. In order to ensure the second flavor interacts with the conjugate links and the first
|
||||
with the regular links we overload the functionality of doubleStore, whose purpose is to store the gauge field and the barrel-shifted gauge field
|
||||
to avoid communicating links when applying the Dirac operator, such that the double-stored field contains also a flavor index which maps to
|
||||
either the link or the conjugate link. This flavored field is then used by multLink to apply the correct link to a spinor.
|
||||
|
||||
Here the first Nd-1 directions are treated as "spatial", and a twist value of 1 indicates G-parity BCs in that direction.
|
||||
mu=Nd-1 is assumed to be the time direction and a twist value of 1 indicates antiperiodic BCs
|
||||
*/
|
||||
template <class S, class Representation = FundamentalRepresentation, class Options=CoeffReal>
|
||||
class GparityWilsonImpl : public ConjugateGaugeImpl<GaugeImplTypes<S, Representation::Dimension> > {
|
||||
public:
|
||||
@ -125,7 +113,7 @@ public:
|
||||
|| ((distance== 1)&&(icoor[direction]==1))
|
||||
|| ((distance==-1)&&(icoor[direction]==0));
|
||||
|
||||
permute_lane = permute_lane && SE->_around_the_world && St.parameters.twists[mmu] && mmu < Nd-1; //only if we are going around the world in a spatial direction
|
||||
permute_lane = permute_lane && SE->_around_the_world && St.parameters.twists[mmu]; //only if we are going around the world
|
||||
|
||||
//Apply the links
|
||||
int f_upper = permute_lane ? 1 : 0;
|
||||
@ -151,10 +139,10 @@ public:
|
||||
assert((distance == 1) || (distance == -1)); // nearest neighbour stencil hard code
|
||||
assert((sl == 1) || (sl == 2));
|
||||
|
||||
//If this site is an global boundary site, perform the G-parity flavor twist
|
||||
if ( mmu < Nd-1 && SE->_around_the_world && St.parameters.twists[mmu] ) {
|
||||
if ( SE->_around_the_world && St.parameters.twists[mmu] ) {
|
||||
|
||||
if ( sl == 2 ) {
|
||||
//Only do the twist for lanes on the edge of the physical node
|
||||
|
||||
ExtractBuffer<sobj> vals(Nsimd);
|
||||
|
||||
extract(chi,vals);
|
||||
@ -209,19 +197,6 @@ public:
|
||||
reg = memory;
|
||||
}
|
||||
|
||||
|
||||
//Poke 'poke_f0' onto flavor 0 and 'poke_f1' onto flavor 1 in direction mu of the doubled gauge field Uds
|
||||
inline void pokeGparityDoubledGaugeField(DoubledGaugeField &Uds, const GaugeLinkField &poke_f0, const GaugeLinkField &poke_f1, const int mu){
|
||||
autoView(poke_f0_v, poke_f0, CpuRead);
|
||||
autoView(poke_f1_v, poke_f1, CpuRead);
|
||||
autoView(Uds_v, Uds, CpuWrite);
|
||||
thread_foreach(ss,poke_f0_v,{
|
||||
Uds_v[ss](0)(mu) = poke_f0_v[ss]();
|
||||
Uds_v[ss](1)(mu) = poke_f1_v[ss]();
|
||||
});
|
||||
}
|
||||
|
||||
|
||||
inline void DoubleStore(GridBase *GaugeGrid,DoubledGaugeField &Uds,const GaugeField &Umu)
|
||||
{
|
||||
conformable(Uds.Grid(),GaugeGrid);
|
||||
@ -232,19 +207,14 @@ public:
|
||||
GaugeLinkField Uconj(GaugeGrid);
|
||||
|
||||
Lattice<iScalar<vInteger> > coor(GaugeGrid);
|
||||
|
||||
//Here the first Nd-1 directions are treated as "spatial", and a twist value of 1 indicates G-parity BCs in that direction.
|
||||
//mu=Nd-1 is assumed to be the time direction and a twist value of 1 indicates antiperiodic BCs
|
||||
for(int mu=0;mu<Nd-1;mu++){
|
||||
|
||||
if( Params.twists[mu] ){
|
||||
LatticeCoordinate(coor,mu);
|
||||
}
|
||||
|
||||
for(int mu=0;mu<Nd;mu++){
|
||||
|
||||
LatticeCoordinate(coor,mu);
|
||||
|
||||
U = PeekIndex<LorentzIndex>(Umu,mu);
|
||||
Uconj = conjugate(U);
|
||||
|
||||
// Implement the isospin rotation sign on the boundary between f=1 and f=0
|
||||
// This phase could come from a simple bc 1,1,-1,1 ..
|
||||
int neglink = GaugeGrid->GlobalDimensions()[mu]-1;
|
||||
if ( Params.twists[mu] ) {
|
||||
@ -259,7 +229,7 @@ public:
|
||||
thread_foreach(ss,U_v,{
|
||||
Uds_v[ss](0)(mu) = U_v[ss]();
|
||||
Uds_v[ss](1)(mu) = Uconj_v[ss]();
|
||||
});
|
||||
});
|
||||
}
|
||||
|
||||
U = adj(Cshift(U ,mu,-1)); // correct except for spanning the boundary
|
||||
@ -290,38 +260,6 @@ public:
|
||||
});
|
||||
}
|
||||
}
|
||||
|
||||
{ //periodic / antiperiodic temporal BCs
|
||||
int mu = Nd-1;
|
||||
int L = GaugeGrid->GlobalDimensions()[mu];
|
||||
int Lmu = L - 1;
|
||||
|
||||
LatticeCoordinate(coor, mu);
|
||||
|
||||
U = PeekIndex<LorentzIndex>(Umu, mu); //Get t-directed links
|
||||
|
||||
GaugeLinkField *Upoke = &U;
|
||||
|
||||
if(Params.twists[mu]){ //antiperiodic
|
||||
Utmp = where(coor == Lmu, -U, U);
|
||||
Upoke = &Utmp;
|
||||
}
|
||||
|
||||
Uconj = conjugate(*Upoke); //second flavor interacts with conjugate links
|
||||
pokeGparityDoubledGaugeField(Uds, *Upoke, Uconj, mu);
|
||||
|
||||
//Get the barrel-shifted field
|
||||
Utmp = adj(Cshift(U, mu, -1)); //is a forward shift!
|
||||
Upoke = &Utmp;
|
||||
|
||||
if(Params.twists[mu]){
|
||||
U = where(coor == 0, -Utmp, Utmp); //boundary phase
|
||||
Upoke = &U;
|
||||
}
|
||||
|
||||
Uconj = conjugate(*Upoke);
|
||||
pokeGparityDoubledGaugeField(Uds, *Upoke, Uconj, mu + 4);
|
||||
}
|
||||
}
|
||||
|
||||
inline void InsertForce4D(GaugeField &mat, FermionField &Btilde, FermionField &A, int mu) {
|
||||
@ -360,48 +298,28 @@ public:
|
||||
inline void extractLinkField(std::vector<GaugeLinkField> &mat, DoubledGaugeField &Uds){
|
||||
assert(0);
|
||||
}
|
||||
|
||||
inline void InsertForce5D(GaugeField &mat, FermionField &Btilde, FermionField Ã, int mu) {
|
||||
int Ls=Btilde.Grid()->_fdimensions[0];
|
||||
|
||||
{
|
||||
GridBase *GaugeGrid = mat.Grid();
|
||||
Lattice<iScalar<vInteger> > coor(GaugeGrid);
|
||||
|
||||
if( Params.twists[mu] ){
|
||||
LatticeCoordinate(coor,mu);
|
||||
}
|
||||
|
||||
autoView( mat_v , mat, AcceleratorWrite);
|
||||
autoView( Btilde_v , Btilde, AcceleratorRead);
|
||||
autoView( Atilde_v , Atilde, AcceleratorRead);
|
||||
accelerator_for(sss,mat.Grid()->oSites(), FermionField::vector_type::Nsimd(),{
|
||||
int sU=sss;
|
||||
typedef decltype(coalescedRead(mat_v[sU](mu)() )) ColorMatrixType;
|
||||
ColorMatrixType sum;
|
||||
zeroit(sum);
|
||||
for(int s=0;s<Ls;s++){
|
||||
int sF = s+Ls*sU;
|
||||
for(int spn=0;spn<Ns;spn++){ //sum over spin
|
||||
//Flavor 0
|
||||
auto bb = coalescedRead(Btilde_v[sF](0)(spn) ); //color vector
|
||||
auto aa = coalescedRead(Atilde_v[sF](0)(spn) );
|
||||
sum = sum + outerProduct(bb,aa);
|
||||
|
||||
//Flavor 1
|
||||
bb = coalescedRead(Btilde_v[sF](1)(spn) );
|
||||
aa = coalescedRead(Atilde_v[sF](1)(spn) );
|
||||
sum = sum + conjugate(outerProduct(bb,aa));
|
||||
}
|
||||
}
|
||||
coalescedWrite(mat_v[sU](mu)(), sum);
|
||||
});
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
|
||||
inline void InsertForce5D(GaugeField &mat, FermionField &Btilde, FermionField Ã, int mu) {
|
||||
|
||||
int Ls = Btilde.Grid()->_fdimensions[0];
|
||||
|
||||
GaugeLinkField tmp(mat.Grid());
|
||||
tmp = Zero();
|
||||
{
|
||||
autoView( tmp_v , tmp, CpuWrite);
|
||||
autoView( Atilde_v , Atilde, CpuRead);
|
||||
autoView( Btilde_v , Btilde, CpuRead);
|
||||
thread_for(ss,tmp.Grid()->oSites(),{
|
||||
for (int s = 0; s < Ls; s++) {
|
||||
int sF = s + Ls * ss;
|
||||
auto ttmp = traceIndex<SpinIndex>(outerProduct(Btilde_v[sF], Atilde_v[sF]));
|
||||
tmp_v[ss]() = tmp_v[ss]() + ttmp(0, 0) + conjugate(ttmp(1, 1));
|
||||
}
|
||||
});
|
||||
}
|
||||
PokeIndex<LorentzIndex>(mat, tmp, mu);
|
||||
return;
|
||||
}
|
||||
|
||||
};
|
||||
|
||||
@ -409,8 +327,8 @@ typedef GparityWilsonImpl<vComplex , FundamentalRepresentation,CoeffReal> Gparit
|
||||
typedef GparityWilsonImpl<vComplexF, FundamentalRepresentation,CoeffReal> GparityWilsonImplF; // Float
|
||||
typedef GparityWilsonImpl<vComplexD, FundamentalRepresentation,CoeffReal> GparityWilsonImplD; // Double
|
||||
|
||||
//typedef GparityWilsonImpl<vComplex , FundamentalRepresentation,CoeffRealHalfComms> GparityWilsonImplRL; // Real.. whichever prec
|
||||
//typedef GparityWilsonImpl<vComplexF, FundamentalRepresentation,CoeffRealHalfComms> GparityWilsonImplFH; // Float
|
||||
//typedef GparityWilsonImpl<vComplexD, FundamentalRepresentation,CoeffRealHalfComms> GparityWilsonImplDF; // Double
|
||||
typedef GparityWilsonImpl<vComplex , FundamentalRepresentation,CoeffRealHalfComms> GparityWilsonImplRL; // Real.. whichever prec
|
||||
typedef GparityWilsonImpl<vComplexF, FundamentalRepresentation,CoeffRealHalfComms> GparityWilsonImplFH; // Float
|
||||
typedef GparityWilsonImpl<vComplexD, FundamentalRepresentation,CoeffRealHalfComms> GparityWilsonImplDF; // Double
|
||||
|
||||
NAMESPACE_END(Grid);
|
||||
|
||||
@ -4,11 +4,10 @@
|
||||
|
||||
Source file: ./lib/qcd/action/fermion/WilsonCloverFermion.h
|
||||
|
||||
Copyright (C) 2017 - 2022
|
||||
Copyright (C) 2017
|
||||
|
||||
Author: Guido Cossu <guido.cossu@ed.ac.uk>
|
||||
Author: David Preti <>
|
||||
Author: Daniel Richtmann <daniel.richtmann@gmail.com>
|
||||
|
||||
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
|
||||
@ -30,9 +29,7 @@
|
||||
|
||||
#pragma once
|
||||
|
||||
#include <Grid/qcd/action/fermion/WilsonCloverTypes.h>
|
||||
#include <Grid/qcd/action/fermion/WilsonCloverHelpers.h>
|
||||
#include <Grid/qcd/action/fermion/CloverHelpers.h>
|
||||
#include <Grid/Grid.h>
|
||||
|
||||
NAMESPACE_BEGIN(Grid);
|
||||
|
||||
@ -52,16 +49,19 @@ NAMESPACE_BEGIN(Grid);
|
||||
// csw_r = csw_t to recover the isotropic version
|
||||
//////////////////////////////////////////////////////////////////
|
||||
|
||||
template<class Impl, class CloverHelpers>
|
||||
class WilsonCloverFermion : public WilsonFermion<Impl>,
|
||||
public WilsonCloverHelpers<Impl>
|
||||
template <class Impl>
|
||||
class WilsonCloverFermion : public WilsonFermion<Impl>
|
||||
{
|
||||
public:
|
||||
// Types definitions
|
||||
INHERIT_IMPL_TYPES(Impl);
|
||||
INHERIT_CLOVER_TYPES(Impl);
|
||||
template <typename vtype>
|
||||
using iImplClover = iScalar<iMatrix<iMatrix<vtype, Impl::Dimension>, Ns>>;
|
||||
typedef iImplClover<Simd> SiteCloverType;
|
||||
typedef Lattice<SiteCloverType> CloverFieldType;
|
||||
|
||||
typedef WilsonFermion<Impl> WilsonBase;
|
||||
typedef WilsonCloverHelpers<Impl> Helpers;
|
||||
public:
|
||||
typedef WilsonFermion<Impl> WilsonBase;
|
||||
|
||||
virtual int ConstEE(void) { return 0; };
|
||||
virtual void Instantiatable(void){};
|
||||
@ -72,7 +72,42 @@ public:
|
||||
const RealD _csw_r = 0.0,
|
||||
const RealD _csw_t = 0.0,
|
||||
const WilsonAnisotropyCoefficients &clover_anisotropy = WilsonAnisotropyCoefficients(),
|
||||
const ImplParams &impl_p = ImplParams());
|
||||
const ImplParams &impl_p = ImplParams()) : WilsonFermion<Impl>(_Umu,
|
||||
Fgrid,
|
||||
Hgrid,
|
||||
_mass, impl_p, clover_anisotropy),
|
||||
CloverTerm(&Fgrid),
|
||||
CloverTermInv(&Fgrid),
|
||||
CloverTermEven(&Hgrid),
|
||||
CloverTermOdd(&Hgrid),
|
||||
CloverTermInvEven(&Hgrid),
|
||||
CloverTermInvOdd(&Hgrid),
|
||||
CloverTermDagEven(&Hgrid),
|
||||
CloverTermDagOdd(&Hgrid),
|
||||
CloverTermInvDagEven(&Hgrid),
|
||||
CloverTermInvDagOdd(&Hgrid)
|
||||
{
|
||||
assert(Nd == 4); // require 4 dimensions
|
||||
|
||||
if (clover_anisotropy.isAnisotropic)
|
||||
{
|
||||
csw_r = _csw_r * 0.5 / clover_anisotropy.xi_0;
|
||||
diag_mass = _mass + 1.0 + (Nd - 1) * (clover_anisotropy.nu / clover_anisotropy.xi_0);
|
||||
}
|
||||
else
|
||||
{
|
||||
csw_r = _csw_r * 0.5;
|
||||
diag_mass = 4.0 + _mass;
|
||||
}
|
||||
csw_t = _csw_t * 0.5;
|
||||
|
||||
if (csw_r == 0)
|
||||
std::cout << GridLogWarning << "Initializing WilsonCloverFermion with csw_r = 0" << std::endl;
|
||||
if (csw_t == 0)
|
||||
std::cout << GridLogWarning << "Initializing WilsonCloverFermion with csw_t = 0" << std::endl;
|
||||
|
||||
ImportGauge(_Umu);
|
||||
}
|
||||
|
||||
virtual void M(const FermionField &in, FermionField &out);
|
||||
virtual void Mdag(const FermionField &in, FermionField &out);
|
||||
@ -89,21 +124,250 @@ public:
|
||||
void ImportGauge(const GaugeField &_Umu);
|
||||
|
||||
// Derivative parts unpreconditioned pseudofermions
|
||||
void MDeriv(GaugeField &force, const FermionField &X, const FermionField &Y, int dag);
|
||||
void MDeriv(GaugeField &force, const FermionField &X, const FermionField &Y, int dag)
|
||||
{
|
||||
conformable(X.Grid(), Y.Grid());
|
||||
conformable(X.Grid(), force.Grid());
|
||||
GaugeLinkField force_mu(force.Grid()), lambda(force.Grid());
|
||||
GaugeField clover_force(force.Grid());
|
||||
PropagatorField Lambda(force.Grid());
|
||||
|
||||
public:
|
||||
// Guido: Here we are hitting some performance issues:
|
||||
// need to extract the components of the DoubledGaugeField
|
||||
// for each call
|
||||
// Possible solution
|
||||
// Create a vector object to store them? (cons: wasting space)
|
||||
std::vector<GaugeLinkField> U(Nd, this->Umu.Grid());
|
||||
|
||||
Impl::extractLinkField(U, this->Umu);
|
||||
|
||||
force = Zero();
|
||||
// Derivative of the Wilson hopping term
|
||||
this->DhopDeriv(force, X, Y, dag);
|
||||
|
||||
///////////////////////////////////////////////////////////
|
||||
// Clover term derivative
|
||||
///////////////////////////////////////////////////////////
|
||||
Impl::outerProductImpl(Lambda, X, Y);
|
||||
//std::cout << "Lambda:" << Lambda << std::endl;
|
||||
|
||||
Gamma::Algebra sigma[] = {
|
||||
Gamma::Algebra::SigmaXY,
|
||||
Gamma::Algebra::SigmaXZ,
|
||||
Gamma::Algebra::SigmaXT,
|
||||
Gamma::Algebra::MinusSigmaXY,
|
||||
Gamma::Algebra::SigmaYZ,
|
||||
Gamma::Algebra::SigmaYT,
|
||||
Gamma::Algebra::MinusSigmaXZ,
|
||||
Gamma::Algebra::MinusSigmaYZ,
|
||||
Gamma::Algebra::SigmaZT,
|
||||
Gamma::Algebra::MinusSigmaXT,
|
||||
Gamma::Algebra::MinusSigmaYT,
|
||||
Gamma::Algebra::MinusSigmaZT};
|
||||
|
||||
/*
|
||||
sigma_{\mu \nu}=
|
||||
| 0 sigma[0] sigma[1] sigma[2] |
|
||||
| sigma[3] 0 sigma[4] sigma[5] |
|
||||
| sigma[6] sigma[7] 0 sigma[8] |
|
||||
| sigma[9] sigma[10] sigma[11] 0 |
|
||||
*/
|
||||
|
||||
int count = 0;
|
||||
clover_force = Zero();
|
||||
for (int mu = 0; mu < 4; mu++)
|
||||
{
|
||||
force_mu = Zero();
|
||||
for (int nu = 0; nu < 4; nu++)
|
||||
{
|
||||
if (mu == nu)
|
||||
continue;
|
||||
|
||||
RealD factor;
|
||||
if (nu == 4 || mu == 4)
|
||||
{
|
||||
factor = 2.0 * csw_t;
|
||||
}
|
||||
else
|
||||
{
|
||||
factor = 2.0 * csw_r;
|
||||
}
|
||||
PropagatorField Slambda = Gamma(sigma[count]) * Lambda; // sigma checked
|
||||
Impl::TraceSpinImpl(lambda, Slambda); // traceSpin ok
|
||||
force_mu -= factor*Cmunu(U, lambda, mu, nu); // checked
|
||||
count++;
|
||||
}
|
||||
|
||||
pokeLorentz(clover_force, U[mu] * force_mu, mu);
|
||||
}
|
||||
//clover_force *= csw;
|
||||
force += clover_force;
|
||||
}
|
||||
|
||||
// Computing C_{\mu \nu}(x) as in Eq.(B.39) in Zbigniew Sroczynski's PhD thesis
|
||||
GaugeLinkField Cmunu(std::vector<GaugeLinkField> &U, GaugeLinkField &lambda, int mu, int nu)
|
||||
{
|
||||
conformable(lambda.Grid(), U[0].Grid());
|
||||
GaugeLinkField out(lambda.Grid()), tmp(lambda.Grid());
|
||||
// insertion in upper staple
|
||||
// please check redundancy of shift operations
|
||||
|
||||
// C1+
|
||||
tmp = lambda * U[nu];
|
||||
out = Impl::ShiftStaple(Impl::CovShiftForward(tmp, nu, Impl::CovShiftBackward(U[mu], mu, Impl::CovShiftIdentityBackward(U[nu], nu))), mu);
|
||||
|
||||
// C2+
|
||||
tmp = U[mu] * Impl::ShiftStaple(adj(lambda), mu);
|
||||
out += Impl::ShiftStaple(Impl::CovShiftForward(U[nu], nu, Impl::CovShiftBackward(tmp, mu, Impl::CovShiftIdentityBackward(U[nu], nu))), mu);
|
||||
|
||||
// C3+
|
||||
tmp = U[nu] * Impl::ShiftStaple(adj(lambda), nu);
|
||||
out += Impl::ShiftStaple(Impl::CovShiftForward(U[nu], nu, Impl::CovShiftBackward(U[mu], mu, Impl::CovShiftIdentityBackward(tmp, nu))), mu);
|
||||
|
||||
// C4+
|
||||
out += Impl::ShiftStaple(Impl::CovShiftForward(U[nu], nu, Impl::CovShiftBackward(U[mu], mu, Impl::CovShiftIdentityBackward(U[nu], nu))), mu) * lambda;
|
||||
|
||||
// insertion in lower staple
|
||||
// C1-
|
||||
out -= Impl::ShiftStaple(lambda, mu) * Impl::ShiftStaple(Impl::CovShiftBackward(U[nu], nu, Impl::CovShiftBackward(U[mu], mu, U[nu])), mu);
|
||||
|
||||
// C2-
|
||||
tmp = adj(lambda) * U[nu];
|
||||
out -= Impl::ShiftStaple(Impl::CovShiftBackward(tmp, nu, Impl::CovShiftBackward(U[mu], mu, U[nu])), mu);
|
||||
|
||||
// C3-
|
||||
tmp = lambda * U[nu];
|
||||
out -= Impl::ShiftStaple(Impl::CovShiftBackward(U[nu], nu, Impl::CovShiftBackward(U[mu], mu, tmp)), mu);
|
||||
|
||||
// C4-
|
||||
out -= Impl::ShiftStaple(Impl::CovShiftBackward(U[nu], nu, Impl::CovShiftBackward(U[mu], mu, U[nu])), mu) * lambda;
|
||||
|
||||
return out;
|
||||
}
|
||||
|
||||
protected:
|
||||
// here fixing the 4 dimensions, make it more general?
|
||||
|
||||
RealD csw_r; // Clover coefficient - spatial
|
||||
RealD csw_t; // Clover coefficient - temporal
|
||||
RealD diag_mass; // Mass term
|
||||
CloverField CloverTerm, CloverTermInv; // Clover term
|
||||
CloverField CloverTermEven, CloverTermOdd; // Clover term EO
|
||||
CloverField CloverTermInvEven, CloverTermInvOdd; // Clover term Inv EO
|
||||
CloverField CloverTermDagEven, CloverTermDagOdd; // Clover term Dag EO
|
||||
CloverField CloverTermInvDagEven, CloverTermInvDagOdd; // Clover term Inv Dag EO
|
||||
};
|
||||
CloverFieldType CloverTerm, CloverTermInv; // Clover term
|
||||
CloverFieldType CloverTermEven, CloverTermOdd; // Clover term EO
|
||||
CloverFieldType CloverTermInvEven, CloverTermInvOdd; // Clover term Inv EO
|
||||
CloverFieldType CloverTermDagEven, CloverTermDagOdd; // Clover term Dag EO
|
||||
CloverFieldType CloverTermInvDagEven, CloverTermInvDagOdd; // Clover term Inv Dag EO
|
||||
|
||||
public:
|
||||
// eventually these can be compressed into 6x6 blocks instead of the 12x12
|
||||
// using the DeGrand-Rossi basis for the gamma matrices
|
||||
CloverFieldType fillCloverYZ(const GaugeLinkField &F)
|
||||
{
|
||||
CloverFieldType T(F.Grid());
|
||||
T = Zero();
|
||||
autoView(T_v,T,AcceleratorWrite);
|
||||
autoView(F_v,F,AcceleratorRead);
|
||||
accelerator_for(i, CloverTerm.Grid()->oSites(),1,
|
||||
{
|
||||
T_v[i]()(0, 1) = timesMinusI(F_v[i]()());
|
||||
T_v[i]()(1, 0) = timesMinusI(F_v[i]()());
|
||||
T_v[i]()(2, 3) = timesMinusI(F_v[i]()());
|
||||
T_v[i]()(3, 2) = timesMinusI(F_v[i]()());
|
||||
});
|
||||
|
||||
return T;
|
||||
}
|
||||
|
||||
CloverFieldType fillCloverXZ(const GaugeLinkField &F)
|
||||
{
|
||||
CloverFieldType T(F.Grid());
|
||||
T = Zero();
|
||||
|
||||
autoView(T_v, T,AcceleratorWrite);
|
||||
autoView(F_v, F,AcceleratorRead);
|
||||
accelerator_for(i, CloverTerm.Grid()->oSites(),1,
|
||||
{
|
||||
T_v[i]()(0, 1) = -F_v[i]()();
|
||||
T_v[i]()(1, 0) = F_v[i]()();
|
||||
T_v[i]()(2, 3) = -F_v[i]()();
|
||||
T_v[i]()(3, 2) = F_v[i]()();
|
||||
});
|
||||
|
||||
return T;
|
||||
}
|
||||
|
||||
CloverFieldType fillCloverXY(const GaugeLinkField &F)
|
||||
{
|
||||
CloverFieldType T(F.Grid());
|
||||
T = Zero();
|
||||
|
||||
autoView(T_v,T,AcceleratorWrite);
|
||||
autoView(F_v,F,AcceleratorRead);
|
||||
accelerator_for(i, CloverTerm.Grid()->oSites(),1,
|
||||
{
|
||||
T_v[i]()(0, 0) = timesMinusI(F_v[i]()());
|
||||
T_v[i]()(1, 1) = timesI(F_v[i]()());
|
||||
T_v[i]()(2, 2) = timesMinusI(F_v[i]()());
|
||||
T_v[i]()(3, 3) = timesI(F_v[i]()());
|
||||
});
|
||||
|
||||
return T;
|
||||
}
|
||||
|
||||
CloverFieldType fillCloverXT(const GaugeLinkField &F)
|
||||
{
|
||||
CloverFieldType T(F.Grid());
|
||||
T = Zero();
|
||||
|
||||
autoView( T_v , T, AcceleratorWrite);
|
||||
autoView( F_v , F, AcceleratorRead);
|
||||
accelerator_for(i, CloverTerm.Grid()->oSites(),1,
|
||||
{
|
||||
T_v[i]()(0, 1) = timesI(F_v[i]()());
|
||||
T_v[i]()(1, 0) = timesI(F_v[i]()());
|
||||
T_v[i]()(2, 3) = timesMinusI(F_v[i]()());
|
||||
T_v[i]()(3, 2) = timesMinusI(F_v[i]()());
|
||||
});
|
||||
|
||||
return T;
|
||||
}
|
||||
|
||||
CloverFieldType fillCloverYT(const GaugeLinkField &F)
|
||||
{
|
||||
CloverFieldType T(F.Grid());
|
||||
T = Zero();
|
||||
|
||||
autoView( T_v ,T,AcceleratorWrite);
|
||||
autoView( F_v ,F,AcceleratorRead);
|
||||
accelerator_for(i, CloverTerm.Grid()->oSites(),1,
|
||||
{
|
||||
T_v[i]()(0, 1) = -(F_v[i]()());
|
||||
T_v[i]()(1, 0) = (F_v[i]()());
|
||||
T_v[i]()(2, 3) = (F_v[i]()());
|
||||
T_v[i]()(3, 2) = -(F_v[i]()());
|
||||
});
|
||||
|
||||
return T;
|
||||
}
|
||||
|
||||
CloverFieldType fillCloverZT(const GaugeLinkField &F)
|
||||
{
|
||||
CloverFieldType T(F.Grid());
|
||||
|
||||
T = Zero();
|
||||
|
||||
autoView( T_v , T,AcceleratorWrite);
|
||||
autoView( F_v , F,AcceleratorRead);
|
||||
accelerator_for(i, CloverTerm.Grid()->oSites(),1,
|
||||
{
|
||||
T_v[i]()(0, 0) = timesI(F_v[i]()());
|
||||
T_v[i]()(1, 1) = timesMinusI(F_v[i]()());
|
||||
T_v[i]()(2, 2) = timesMinusI(F_v[i]()());
|
||||
T_v[i]()(3, 3) = timesI(F_v[i]()());
|
||||
});
|
||||
|
||||
return T;
|
||||
}
|
||||
};
|
||||
NAMESPACE_END(Grid);
|
||||
|
||||
|
||||
|
||||
@ -1,763 +0,0 @@
|
||||
/*************************************************************************************
|
||||
|
||||
Grid physics library, www.github.com/paboyle/Grid
|
||||
|
||||
Source file: ./lib/qcd/action/fermion/WilsonCloverHelpers.h
|
||||
|
||||
Copyright (C) 2021 - 2022
|
||||
|
||||
Author: Daniel Richtmann <daniel.richtmann@gmail.com>
|
||||
|
||||
This program is free software; you can redistribute it and/or modify
|
||||
it under the terms of the GNU General Public License as published by
|
||||
the Free Software Foundation; either version 2 of the License, or
|
||||
(at your option) any later version.
|
||||
|
||||
This program is distributed in the hope that it will be useful,
|
||||
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||||
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
||||
GNU General Public License for more details.
|
||||
|
||||
You should have received a copy of the GNU General Public License along
|
||||
with this program; if not, write to the Free Software Foundation, Inc.,
|
||||
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
|
||||
|
||||
See the full license in the file "LICENSE" in the top level distribution directory
|
||||
*************************************************************************************/
|
||||
/* END LEGAL */
|
||||
|
||||
#pragma once
|
||||
|
||||
// Helper routines that implement common clover functionality
|
||||
|
||||
NAMESPACE_BEGIN(Grid);
|
||||
|
||||
template<class Impl> class WilsonCloverHelpers {
|
||||
public:
|
||||
|
||||
INHERIT_IMPL_TYPES(Impl);
|
||||
INHERIT_CLOVER_TYPES(Impl);
|
||||
|
||||
// Computing C_{\mu \nu}(x) as in Eq.(B.39) in Zbigniew Sroczynski's PhD thesis
|
||||
static GaugeLinkField Cmunu(std::vector<GaugeLinkField> &U, GaugeLinkField &lambda, int mu, int nu)
|
||||
{
|
||||
conformable(lambda.Grid(), U[0].Grid());
|
||||
GaugeLinkField out(lambda.Grid()), tmp(lambda.Grid());
|
||||
// insertion in upper staple
|
||||
// please check redundancy of shift operations
|
||||
|
||||
// C1+
|
||||
tmp = lambda * U[nu];
|
||||
out = Impl::ShiftStaple(Impl::CovShiftForward(tmp, nu, Impl::CovShiftBackward(U[mu], mu, Impl::CovShiftIdentityBackward(U[nu], nu))), mu);
|
||||
|
||||
// C2+
|
||||
tmp = U[mu] * Impl::ShiftStaple(adj(lambda), mu);
|
||||
out += Impl::ShiftStaple(Impl::CovShiftForward(U[nu], nu, Impl::CovShiftBackward(tmp, mu, Impl::CovShiftIdentityBackward(U[nu], nu))), mu);
|
||||
|
||||
// C3+
|
||||
tmp = U[nu] * Impl::ShiftStaple(adj(lambda), nu);
|
||||
out += Impl::ShiftStaple(Impl::CovShiftForward(U[nu], nu, Impl::CovShiftBackward(U[mu], mu, Impl::CovShiftIdentityBackward(tmp, nu))), mu);
|
||||
|
||||
// C4+
|
||||
out += Impl::ShiftStaple(Impl::CovShiftForward(U[nu], nu, Impl::CovShiftBackward(U[mu], mu, Impl::CovShiftIdentityBackward(U[nu], nu))), mu) * lambda;
|
||||
|
||||
// insertion in lower staple
|
||||
// C1-
|
||||
out -= Impl::ShiftStaple(lambda, mu) * Impl::ShiftStaple(Impl::CovShiftBackward(U[nu], nu, Impl::CovShiftBackward(U[mu], mu, U[nu])), mu);
|
||||
|
||||
// C2-
|
||||
tmp = adj(lambda) * U[nu];
|
||||
out -= Impl::ShiftStaple(Impl::CovShiftBackward(tmp, nu, Impl::CovShiftBackward(U[mu], mu, U[nu])), mu);
|
||||
|
||||
// C3-
|
||||
tmp = lambda * U[nu];
|
||||
out -= Impl::ShiftStaple(Impl::CovShiftBackward(U[nu], nu, Impl::CovShiftBackward(U[mu], mu, tmp)), mu);
|
||||
|
||||
// C4-
|
||||
out -= Impl::ShiftStaple(Impl::CovShiftBackward(U[nu], nu, Impl::CovShiftBackward(U[mu], mu, U[nu])), mu) * lambda;
|
||||
|
||||
return out;
|
||||
}
|
||||
|
||||
static CloverField fillCloverYZ(const GaugeLinkField &F)
|
||||
{
|
||||
CloverField T(F.Grid());
|
||||
T = Zero();
|
||||
autoView(T_v,T,AcceleratorWrite);
|
||||
autoView(F_v,F,AcceleratorRead);
|
||||
accelerator_for(i, T.Grid()->oSites(),CloverField::vector_type::Nsimd(),
|
||||
{
|
||||
coalescedWrite(T_v[i]()(0, 1), coalescedRead(timesMinusI(F_v[i]()())));
|
||||
coalescedWrite(T_v[i]()(1, 0), coalescedRead(timesMinusI(F_v[i]()())));
|
||||
coalescedWrite(T_v[i]()(2, 3), coalescedRead(timesMinusI(F_v[i]()())));
|
||||
coalescedWrite(T_v[i]()(3, 2), coalescedRead(timesMinusI(F_v[i]()())));
|
||||
});
|
||||
|
||||
return T;
|
||||
}
|
||||
|
||||
static CloverField fillCloverXZ(const GaugeLinkField &F)
|
||||
{
|
||||
CloverField T(F.Grid());
|
||||
T = Zero();
|
||||
|
||||
autoView(T_v, T,AcceleratorWrite);
|
||||
autoView(F_v, F,AcceleratorRead);
|
||||
accelerator_for(i, T.Grid()->oSites(),CloverField::vector_type::Nsimd(),
|
||||
{
|
||||
coalescedWrite(T_v[i]()(0, 1), coalescedRead(-F_v[i]()()));
|
||||
coalescedWrite(T_v[i]()(1, 0), coalescedRead(F_v[i]()()));
|
||||
coalescedWrite(T_v[i]()(2, 3), coalescedRead(-F_v[i]()()));
|
||||
coalescedWrite(T_v[i]()(3, 2), coalescedRead(F_v[i]()()));
|
||||
});
|
||||
|
||||
return T;
|
||||
}
|
||||
|
||||
static CloverField fillCloverXY(const GaugeLinkField &F)
|
||||
{
|
||||
CloverField T(F.Grid());
|
||||
T = Zero();
|
||||
|
||||
autoView(T_v,T,AcceleratorWrite);
|
||||
autoView(F_v,F,AcceleratorRead);
|
||||
accelerator_for(i, T.Grid()->oSites(),CloverField::vector_type::Nsimd(),
|
||||
{
|
||||
coalescedWrite(T_v[i]()(0, 0), coalescedRead(timesMinusI(F_v[i]()())));
|
||||
coalescedWrite(T_v[i]()(1, 1), coalescedRead(timesI(F_v[i]()())));
|
||||
coalescedWrite(T_v[i]()(2, 2), coalescedRead(timesMinusI(F_v[i]()())));
|
||||
coalescedWrite(T_v[i]()(3, 3), coalescedRead(timesI(F_v[i]()())));
|
||||
});
|
||||
|
||||
return T;
|
||||
}
|
||||
|
||||
static CloverField fillCloverXT(const GaugeLinkField &F)
|
||||
{
|
||||
CloverField T(F.Grid());
|
||||
T = Zero();
|
||||
|
||||
autoView( T_v , T, AcceleratorWrite);
|
||||
autoView( F_v , F, AcceleratorRead);
|
||||
accelerator_for(i, T.Grid()->oSites(),CloverField::vector_type::Nsimd(),
|
||||
{
|
||||
coalescedWrite(T_v[i]()(0, 1), coalescedRead(timesI(F_v[i]()())));
|
||||
coalescedWrite(T_v[i]()(1, 0), coalescedRead(timesI(F_v[i]()())));
|
||||
coalescedWrite(T_v[i]()(2, 3), coalescedRead(timesMinusI(F_v[i]()())));
|
||||
coalescedWrite(T_v[i]()(3, 2), coalescedRead(timesMinusI(F_v[i]()())));
|
||||
});
|
||||
|
||||
return T;
|
||||
}
|
||||
|
||||
static CloverField fillCloverYT(const GaugeLinkField &F)
|
||||
{
|
||||
CloverField T(F.Grid());
|
||||
T = Zero();
|
||||
|
||||
autoView( T_v ,T,AcceleratorWrite);
|
||||
autoView( F_v ,F,AcceleratorRead);
|
||||
accelerator_for(i, T.Grid()->oSites(),CloverField::vector_type::Nsimd(),
|
||||
{
|
||||
coalescedWrite(T_v[i]()(0, 1), coalescedRead(-(F_v[i]()())));
|
||||
coalescedWrite(T_v[i]()(1, 0), coalescedRead((F_v[i]()())));
|
||||
coalescedWrite(T_v[i]()(2, 3), coalescedRead((F_v[i]()())));
|
||||
coalescedWrite(T_v[i]()(3, 2), coalescedRead(-(F_v[i]()())));
|
||||
});
|
||||
|
||||
return T;
|
||||
}
|
||||
|
||||
static CloverField fillCloverZT(const GaugeLinkField &F)
|
||||
{
|
||||
CloverField T(F.Grid());
|
||||
|
||||
T = Zero();
|
||||
|
||||
autoView( T_v , T,AcceleratorWrite);
|
||||
autoView( F_v , F,AcceleratorRead);
|
||||
accelerator_for(i, T.Grid()->oSites(),CloverField::vector_type::Nsimd(),
|
||||
{
|
||||
coalescedWrite(T_v[i]()(0, 0), coalescedRead(timesI(F_v[i]()())));
|
||||
coalescedWrite(T_v[i]()(1, 1), coalescedRead(timesMinusI(F_v[i]()())));
|
||||
coalescedWrite(T_v[i]()(2, 2), coalescedRead(timesMinusI(F_v[i]()())));
|
||||
coalescedWrite(T_v[i]()(3, 3), coalescedRead(timesI(F_v[i]()())));
|
||||
});
|
||||
|
||||
return T;
|
||||
}
|
||||
|
||||
template<class _Spinor>
|
||||
static accelerator_inline void multClover(_Spinor& phi, const SiteClover& C, const _Spinor& chi) {
|
||||
auto CC = coalescedRead(C);
|
||||
mult(&phi, &CC, &chi);
|
||||
}
|
||||
|
||||
template<class _SpinorField>
|
||||
inline void multCloverField(_SpinorField& out, const CloverField& C, const _SpinorField& phi) {
|
||||
const int Nsimd = SiteSpinor::Nsimd();
|
||||
autoView(out_v, out, AcceleratorWrite);
|
||||
autoView(phi_v, phi, AcceleratorRead);
|
||||
autoView(C_v, C, AcceleratorRead);
|
||||
typedef decltype(coalescedRead(out_v[0])) calcSpinor;
|
||||
accelerator_for(sss,out.Grid()->oSites(),Nsimd,{
|
||||
calcSpinor tmp;
|
||||
multClover(tmp,C_v[sss],phi_v(sss));
|
||||
coalescedWrite(out_v[sss],tmp);
|
||||
});
|
||||
}
|
||||
};
|
||||
|
||||
|
||||
////////////////////////////////////////////////////////
|
||||
|
||||
template<class Impl> class CompactWilsonCloverHelpers {
|
||||
public:
|
||||
|
||||
INHERIT_COMPACT_CLOVER_SIZES(Impl);
|
||||
|
||||
INHERIT_IMPL_TYPES(Impl);
|
||||
INHERIT_CLOVER_TYPES(Impl);
|
||||
INHERIT_COMPACT_CLOVER_TYPES(Impl);
|
||||
|
||||
#if 0
|
||||
static accelerator_inline typename SiteCloverTriangle::vector_type triangle_elem(const SiteCloverTriangle& triangle, int block, int i, int j) {
|
||||
assert(i != j);
|
||||
if(i < j) {
|
||||
return triangle()(block)(triangle_index(i, j));
|
||||
} else { // i > j
|
||||
return conjugate(triangle()(block)(triangle_index(i, j)));
|
||||
}
|
||||
}
|
||||
#else
|
||||
template<typename vobj>
|
||||
static accelerator_inline vobj triangle_elem(const iImplCloverTriangle<vobj>& triangle, int block, int i, int j) {
|
||||
assert(i != j);
|
||||
if(i < j) {
|
||||
return triangle()(block)(triangle_index(i, j));
|
||||
} else { // i > j
|
||||
return conjugate(triangle()(block)(triangle_index(i, j)));
|
||||
}
|
||||
}
|
||||
#endif
|
||||
|
||||
static accelerator_inline int triangle_index(int i, int j) {
|
||||
if(i == j)
|
||||
return 0;
|
||||
else if(i < j)
|
||||
return Nred * (Nred - 1) / 2 - (Nred - i) * (Nred - i - 1) / 2 + j - i - 1;
|
||||
else // i > j
|
||||
return Nred * (Nred - 1) / 2 - (Nred - j) * (Nred - j - 1) / 2 + i - j - 1;
|
||||
}
|
||||
|
||||
static void MooeeKernel_gpu(int Nsite,
|
||||
int Ls,
|
||||
const FermionField& in,
|
||||
FermionField& out,
|
||||
const CloverDiagonalField& diagonal,
|
||||
const CloverTriangleField& triangle) {
|
||||
autoView(diagonal_v, diagonal, AcceleratorRead);
|
||||
autoView(triangle_v, triangle, AcceleratorRead);
|
||||
autoView(in_v, in, AcceleratorRead);
|
||||
autoView(out_v, out, AcceleratorWrite);
|
||||
|
||||
typedef decltype(coalescedRead(out_v[0])) CalcSpinor;
|
||||
|
||||
const uint64_t NN = Nsite * Ls;
|
||||
|
||||
accelerator_for(ss, NN, Simd::Nsimd(), {
|
||||
int sF = ss;
|
||||
int sU = ss/Ls;
|
||||
CalcSpinor res;
|
||||
CalcSpinor in_t = in_v(sF);
|
||||
auto diagonal_t = diagonal_v(sU);
|
||||
auto triangle_t = triangle_v(sU);
|
||||
for(int block=0; block<Nhs; block++) {
|
||||
int s_start = block*Nhs;
|
||||
for(int i=0; i<Nred; i++) {
|
||||
int si = s_start + i/Nc, ci = i%Nc;
|
||||
res()(si)(ci) = diagonal_t()(block)(i) * in_t()(si)(ci);
|
||||
for(int j=0; j<Nred; j++) {
|
||||
if (j == i) continue;
|
||||
int sj = s_start + j/Nc, cj = j%Nc;
|
||||
res()(si)(ci) = res()(si)(ci) + triangle_elem(triangle_t, block, i, j) * in_t()(sj)(cj);
|
||||
};
|
||||
};
|
||||
};
|
||||
coalescedWrite(out_v[sF], res);
|
||||
});
|
||||
}
|
||||
|
||||
static void MooeeKernel_cpu(int Nsite,
|
||||
int Ls,
|
||||
const FermionField& in,
|
||||
FermionField& out,
|
||||
const CloverDiagonalField& diagonal,
|
||||
const CloverTriangleField& triangle) {
|
||||
autoView(diagonal_v, diagonal, CpuRead);
|
||||
autoView(triangle_v, triangle, CpuRead);
|
||||
autoView(in_v, in, CpuRead);
|
||||
autoView(out_v, out, CpuWrite);
|
||||
|
||||
typedef SiteSpinor CalcSpinor;
|
||||
|
||||
#if defined(A64FX) || defined(A64FXFIXEDSIZE)
|
||||
#define PREFETCH_CLOVER(BASE) { \
|
||||
uint64_t base; \
|
||||
int pf_dist_L1 = 1; \
|
||||
int pf_dist_L2 = -5; /* -> penalty -> disable */ \
|
||||
\
|
||||
if ((pf_dist_L1 >= 0) && (sU + pf_dist_L1 < Nsite)) { \
|
||||
base = (uint64_t)&diag_t()(pf_dist_L1+BASE)(0); \
|
||||
svprfd(svptrue_b64(), (int64_t*)(base + 0), SV_PLDL1STRM); \
|
||||
svprfd(svptrue_b64(), (int64_t*)(base + 256), SV_PLDL1STRM); \
|
||||
svprfd(svptrue_b64(), (int64_t*)(base + 512), SV_PLDL1STRM); \
|
||||
svprfd(svptrue_b64(), (int64_t*)(base + 768), SV_PLDL1STRM); \
|
||||
svprfd(svptrue_b64(), (int64_t*)(base + 1024), SV_PLDL1STRM); \
|
||||
svprfd(svptrue_b64(), (int64_t*)(base + 1280), SV_PLDL1STRM); \
|
||||
} \
|
||||
\
|
||||
if ((pf_dist_L2 >= 0) && (sU + pf_dist_L2 < Nsite)) { \
|
||||
base = (uint64_t)&diag_t()(pf_dist_L2+BASE)(0); \
|
||||
svprfd(svptrue_b64(), (int64_t*)(base + 0), SV_PLDL2STRM); \
|
||||
svprfd(svptrue_b64(), (int64_t*)(base + 256), SV_PLDL2STRM); \
|
||||
svprfd(svptrue_b64(), (int64_t*)(base + 512), SV_PLDL2STRM); \
|
||||
svprfd(svptrue_b64(), (int64_t*)(base + 768), SV_PLDL2STRM); \
|
||||
svprfd(svptrue_b64(), (int64_t*)(base + 1024), SV_PLDL2STRM); \
|
||||
svprfd(svptrue_b64(), (int64_t*)(base + 1280), SV_PLDL2STRM); \
|
||||
} \
|
||||
}
|
||||
// TODO: Implement/generalize this for other architectures
|
||||
// I played around a bit on KNL (see below) but didn't bring anything
|
||||
// #elif defined(AVX512)
|
||||
// #define PREFETCH_CLOVER(BASE) { \
|
||||
// uint64_t base; \
|
||||
// int pf_dist_L1 = 1; \
|
||||
// int pf_dist_L2 = +4; \
|
||||
// \
|
||||
// if ((pf_dist_L1 >= 0) && (sU + pf_dist_L1 < Nsite)) { \
|
||||
// base = (uint64_t)&diag_t()(pf_dist_L1+BASE)(0); \
|
||||
// _mm_prefetch((const char*)(base + 0), _MM_HINT_T0); \
|
||||
// _mm_prefetch((const char*)(base + 64), _MM_HINT_T0); \
|
||||
// _mm_prefetch((const char*)(base + 128), _MM_HINT_T0); \
|
||||
// _mm_prefetch((const char*)(base + 192), _MM_HINT_T0); \
|
||||
// _mm_prefetch((const char*)(base + 256), _MM_HINT_T0); \
|
||||
// _mm_prefetch((const char*)(base + 320), _MM_HINT_T0); \
|
||||
// } \
|
||||
// \
|
||||
// if ((pf_dist_L2 >= 0) && (sU + pf_dist_L2 < Nsite)) { \
|
||||
// base = (uint64_t)&diag_t()(pf_dist_L2+BASE)(0); \
|
||||
// _mm_prefetch((const char*)(base + 0), _MM_HINT_T1); \
|
||||
// _mm_prefetch((const char*)(base + 64), _MM_HINT_T1); \
|
||||
// _mm_prefetch((const char*)(base + 128), _MM_HINT_T1); \
|
||||
// _mm_prefetch((const char*)(base + 192), _MM_HINT_T1); \
|
||||
// _mm_prefetch((const char*)(base + 256), _MM_HINT_T1); \
|
||||
// _mm_prefetch((const char*)(base + 320), _MM_HINT_T1); \
|
||||
// } \
|
||||
// }
|
||||
#else
|
||||
#define PREFETCH_CLOVER(BASE)
|
||||
#endif
|
||||
|
||||
const uint64_t NN = Nsite * Ls;
|
||||
|
||||
thread_for(ss, NN, {
|
||||
int sF = ss;
|
||||
int sU = ss/Ls;
|
||||
CalcSpinor res;
|
||||
CalcSpinor in_t = in_v[sF];
|
||||
auto diag_t = diagonal_v[sU]; // "diag" instead of "diagonal" here to make code below easier to read
|
||||
auto triangle_t = triangle_v[sU];
|
||||
|
||||
// upper half
|
||||
PREFETCH_CLOVER(0);
|
||||
|
||||
auto in_cc_0_0 = conjugate(in_t()(0)(0)); // Nils: reduces number
|
||||
auto in_cc_0_1 = conjugate(in_t()(0)(1)); // of conjugates from
|
||||
auto in_cc_0_2 = conjugate(in_t()(0)(2)); // 30 to 20
|
||||
auto in_cc_1_0 = conjugate(in_t()(1)(0));
|
||||
auto in_cc_1_1 = conjugate(in_t()(1)(1));
|
||||
|
||||
res()(0)(0) = diag_t()(0)( 0) * in_t()(0)(0)
|
||||
+ triangle_t()(0)( 0) * in_t()(0)(1)
|
||||
+ triangle_t()(0)( 1) * in_t()(0)(2)
|
||||
+ triangle_t()(0)( 2) * in_t()(1)(0)
|
||||
+ triangle_t()(0)( 3) * in_t()(1)(1)
|
||||
+ triangle_t()(0)( 4) * in_t()(1)(2);
|
||||
|
||||
res()(0)(1) = triangle_t()(0)( 0) * in_cc_0_0;
|
||||
res()(0)(1) = diag_t()(0)( 1) * in_t()(0)(1)
|
||||
+ triangle_t()(0)( 5) * in_t()(0)(2)
|
||||
+ triangle_t()(0)( 6) * in_t()(1)(0)
|
||||
+ triangle_t()(0)( 7) * in_t()(1)(1)
|
||||
+ triangle_t()(0)( 8) * in_t()(1)(2)
|
||||
+ conjugate( res()(0)( 1));
|
||||
|
||||
res()(0)(2) = triangle_t()(0)( 1) * in_cc_0_0
|
||||
+ triangle_t()(0)( 5) * in_cc_0_1;
|
||||
res()(0)(2) = diag_t()(0)( 2) * in_t()(0)(2)
|
||||
+ triangle_t()(0)( 9) * in_t()(1)(0)
|
||||
+ triangle_t()(0)(10) * in_t()(1)(1)
|
||||
+ triangle_t()(0)(11) * in_t()(1)(2)
|
||||
+ conjugate( res()(0)( 2));
|
||||
|
||||
res()(1)(0) = triangle_t()(0)( 2) * in_cc_0_0
|
||||
+ triangle_t()(0)( 6) * in_cc_0_1
|
||||
+ triangle_t()(0)( 9) * in_cc_0_2;
|
||||
res()(1)(0) = diag_t()(0)( 3) * in_t()(1)(0)
|
||||
+ triangle_t()(0)(12) * in_t()(1)(1)
|
||||
+ triangle_t()(0)(13) * in_t()(1)(2)
|
||||
+ conjugate( res()(1)( 0));
|
||||
|
||||
res()(1)(1) = triangle_t()(0)( 3) * in_cc_0_0
|
||||
+ triangle_t()(0)( 7) * in_cc_0_1
|
||||
+ triangle_t()(0)(10) * in_cc_0_2
|
||||
+ triangle_t()(0)(12) * in_cc_1_0;
|
||||
res()(1)(1) = diag_t()(0)( 4) * in_t()(1)(1)
|
||||
+ triangle_t()(0)(14) * in_t()(1)(2)
|
||||
+ conjugate( res()(1)( 1));
|
||||
|
||||
res()(1)(2) = triangle_t()(0)( 4) * in_cc_0_0
|
||||
+ triangle_t()(0)( 8) * in_cc_0_1
|
||||
+ triangle_t()(0)(11) * in_cc_0_2
|
||||
+ triangle_t()(0)(13) * in_cc_1_0
|
||||
+ triangle_t()(0)(14) * in_cc_1_1;
|
||||
res()(1)(2) = diag_t()(0)( 5) * in_t()(1)(2)
|
||||
+ conjugate( res()(1)( 2));
|
||||
|
||||
vstream(out_v[sF]()(0)(0), res()(0)(0));
|
||||
vstream(out_v[sF]()(0)(1), res()(0)(1));
|
||||
vstream(out_v[sF]()(0)(2), res()(0)(2));
|
||||
vstream(out_v[sF]()(1)(0), res()(1)(0));
|
||||
vstream(out_v[sF]()(1)(1), res()(1)(1));
|
||||
vstream(out_v[sF]()(1)(2), res()(1)(2));
|
||||
|
||||
// lower half
|
||||
PREFETCH_CLOVER(1);
|
||||
|
||||
auto in_cc_2_0 = conjugate(in_t()(2)(0));
|
||||
auto in_cc_2_1 = conjugate(in_t()(2)(1));
|
||||
auto in_cc_2_2 = conjugate(in_t()(2)(2));
|
||||
auto in_cc_3_0 = conjugate(in_t()(3)(0));
|
||||
auto in_cc_3_1 = conjugate(in_t()(3)(1));
|
||||
|
||||
res()(2)(0) = diag_t()(1)( 0) * in_t()(2)(0)
|
||||
+ triangle_t()(1)( 0) * in_t()(2)(1)
|
||||
+ triangle_t()(1)( 1) * in_t()(2)(2)
|
||||
+ triangle_t()(1)( 2) * in_t()(3)(0)
|
||||
+ triangle_t()(1)( 3) * in_t()(3)(1)
|
||||
+ triangle_t()(1)( 4) * in_t()(3)(2);
|
||||
|
||||
res()(2)(1) = triangle_t()(1)( 0) * in_cc_2_0;
|
||||
res()(2)(1) = diag_t()(1)( 1) * in_t()(2)(1)
|
||||
+ triangle_t()(1)( 5) * in_t()(2)(2)
|
||||
+ triangle_t()(1)( 6) * in_t()(3)(0)
|
||||
+ triangle_t()(1)( 7) * in_t()(3)(1)
|
||||
+ triangle_t()(1)( 8) * in_t()(3)(2)
|
||||
+ conjugate( res()(2)( 1));
|
||||
|
||||
res()(2)(2) = triangle_t()(1)( 1) * in_cc_2_0
|
||||
+ triangle_t()(1)( 5) * in_cc_2_1;
|
||||
res()(2)(2) = diag_t()(1)( 2) * in_t()(2)(2)
|
||||
+ triangle_t()(1)( 9) * in_t()(3)(0)
|
||||
+ triangle_t()(1)(10) * in_t()(3)(1)
|
||||
+ triangle_t()(1)(11) * in_t()(3)(2)
|
||||
+ conjugate( res()(2)( 2));
|
||||
|
||||
res()(3)(0) = triangle_t()(1)( 2) * in_cc_2_0
|
||||
+ triangle_t()(1)( 6) * in_cc_2_1
|
||||
+ triangle_t()(1)( 9) * in_cc_2_2;
|
||||
res()(3)(0) = diag_t()(1)( 3) * in_t()(3)(0)
|
||||
+ triangle_t()(1)(12) * in_t()(3)(1)
|
||||
+ triangle_t()(1)(13) * in_t()(3)(2)
|
||||
+ conjugate( res()(3)( 0));
|
||||
|
||||
res()(3)(1) = triangle_t()(1)( 3) * in_cc_2_0
|
||||
+ triangle_t()(1)( 7) * in_cc_2_1
|
||||
+ triangle_t()(1)(10) * in_cc_2_2
|
||||
+ triangle_t()(1)(12) * in_cc_3_0;
|
||||
res()(3)(1) = diag_t()(1)( 4) * in_t()(3)(1)
|
||||
+ triangle_t()(1)(14) * in_t()(3)(2)
|
||||
+ conjugate( res()(3)( 1));
|
||||
|
||||
res()(3)(2) = triangle_t()(1)( 4) * in_cc_2_0
|
||||
+ triangle_t()(1)( 8) * in_cc_2_1
|
||||
+ triangle_t()(1)(11) * in_cc_2_2
|
||||
+ triangle_t()(1)(13) * in_cc_3_0
|
||||
+ triangle_t()(1)(14) * in_cc_3_1;
|
||||
res()(3)(2) = diag_t()(1)( 5) * in_t()(3)(2)
|
||||
+ conjugate( res()(3)( 2));
|
||||
|
||||
vstream(out_v[sF]()(2)(0), res()(2)(0));
|
||||
vstream(out_v[sF]()(2)(1), res()(2)(1));
|
||||
vstream(out_v[sF]()(2)(2), res()(2)(2));
|
||||
vstream(out_v[sF]()(3)(0), res()(3)(0));
|
||||
vstream(out_v[sF]()(3)(1), res()(3)(1));
|
||||
vstream(out_v[sF]()(3)(2), res()(3)(2));
|
||||
});
|
||||
}
|
||||
|
||||
static void MooeeKernel(int Nsite,
|
||||
int Ls,
|
||||
const FermionField& in,
|
||||
FermionField& out,
|
||||
const CloverDiagonalField& diagonal,
|
||||
const CloverTriangleField& triangle) {
|
||||
#if defined(GRID_CUDA) || defined(GRID_HIP)
|
||||
MooeeKernel_gpu(Nsite, Ls, in, out, diagonal, triangle);
|
||||
#else
|
||||
MooeeKernel_cpu(Nsite, Ls, in, out, diagonal, triangle);
|
||||
#endif
|
||||
}
|
||||
|
||||
static void Invert(const CloverDiagonalField& diagonal,
|
||||
const CloverTriangleField& triangle,
|
||||
CloverDiagonalField& diagonalInv,
|
||||
CloverTriangleField& triangleInv) {
|
||||
conformable(diagonal, diagonalInv);
|
||||
conformable(triangle, triangleInv);
|
||||
conformable(diagonal, triangle);
|
||||
|
||||
diagonalInv.Checkerboard() = diagonal.Checkerboard();
|
||||
triangleInv.Checkerboard() = triangle.Checkerboard();
|
||||
|
||||
GridBase* grid = diagonal.Grid();
|
||||
|
||||
long lsites = grid->lSites();
|
||||
|
||||
typedef typename SiteCloverDiagonal::scalar_object scalar_object_diagonal;
|
||||
typedef typename SiteCloverTriangle::scalar_object scalar_object_triangle;
|
||||
|
||||
autoView(diagonal_v, diagonal, CpuRead);
|
||||
autoView(triangle_v, triangle, CpuRead);
|
||||
autoView(diagonalInv_v, diagonalInv, CpuWrite);
|
||||
autoView(triangleInv_v, triangleInv, CpuWrite);
|
||||
|
||||
thread_for(site, lsites, { // NOTE: Not on GPU because of Eigen & (peek/poke)LocalSite
|
||||
Eigen::MatrixXcd clover_inv_eigen = Eigen::MatrixXcd::Zero(Ns*Nc, Ns*Nc);
|
||||
Eigen::MatrixXcd clover_eigen = Eigen::MatrixXcd::Zero(Ns*Nc, Ns*Nc);
|
||||
|
||||
scalar_object_diagonal diagonal_tmp = Zero();
|
||||
scalar_object_diagonal diagonal_inv_tmp = Zero();
|
||||
scalar_object_triangle triangle_tmp = Zero();
|
||||
scalar_object_triangle triangle_inv_tmp = Zero();
|
||||
|
||||
Coordinate lcoor;
|
||||
grid->LocalIndexToLocalCoor(site, lcoor);
|
||||
|
||||
peekLocalSite(diagonal_tmp, diagonal_v, lcoor);
|
||||
peekLocalSite(triangle_tmp, triangle_v, lcoor);
|
||||
|
||||
// TODO: can we save time here by inverting the two 6x6 hermitian matrices separately?
|
||||
for (long s_row=0;s_row<Ns;s_row++) {
|
||||
for (long s_col=0;s_col<Ns;s_col++) {
|
||||
if(abs(s_row - s_col) > 1 || s_row + s_col == 3) continue;
|
||||
int block = s_row / Nhs;
|
||||
int s_row_block = s_row % Nhs;
|
||||
int s_col_block = s_col % Nhs;
|
||||
for (long c_row=0;c_row<Nc;c_row++) {
|
||||
for (long c_col=0;c_col<Nc;c_col++) {
|
||||
int i = s_row_block * Nc + c_row;
|
||||
int j = s_col_block * Nc + c_col;
|
||||
if(i == j)
|
||||
clover_eigen(s_row*Nc+c_row, s_col*Nc+c_col) = static_cast<ComplexD>(TensorRemove(diagonal_tmp()(block)(i)));
|
||||
else
|
||||
clover_eigen(s_row*Nc+c_row, s_col*Nc+c_col) = static_cast<ComplexD>(TensorRemove(triangle_elem(triangle_tmp, block, i, j)));
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
clover_inv_eigen = clover_eigen.inverse();
|
||||
|
||||
for (long s_row=0;s_row<Ns;s_row++) {
|
||||
for (long s_col=0;s_col<Ns;s_col++) {
|
||||
if(abs(s_row - s_col) > 1 || s_row + s_col == 3) continue;
|
||||
int block = s_row / Nhs;
|
||||
int s_row_block = s_row % Nhs;
|
||||
int s_col_block = s_col % Nhs;
|
||||
for (long c_row=0;c_row<Nc;c_row++) {
|
||||
for (long c_col=0;c_col<Nc;c_col++) {
|
||||
int i = s_row_block * Nc + c_row;
|
||||
int j = s_col_block * Nc + c_col;
|
||||
if(i == j)
|
||||
diagonal_inv_tmp()(block)(i) = clover_inv_eigen(s_row*Nc+c_row, s_col*Nc+c_col);
|
||||
else if(i < j)
|
||||
triangle_inv_tmp()(block)(triangle_index(i, j)) = clover_inv_eigen(s_row*Nc+c_row, s_col*Nc+c_col);
|
||||
else
|
||||
continue;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
pokeLocalSite(diagonal_inv_tmp, diagonalInv_v, lcoor);
|
||||
pokeLocalSite(triangle_inv_tmp, triangleInv_v, lcoor);
|
||||
});
|
||||
}
|
||||
|
||||
static void ConvertLayout(const CloverField& full,
|
||||
CloverDiagonalField& diagonal,
|
||||
CloverTriangleField& triangle) {
|
||||
conformable(full, diagonal);
|
||||
conformable(full, triangle);
|
||||
|
||||
diagonal.Checkerboard() = full.Checkerboard();
|
||||
triangle.Checkerboard() = full.Checkerboard();
|
||||
|
||||
autoView(full_v, full, AcceleratorRead);
|
||||
autoView(diagonal_v, diagonal, AcceleratorWrite);
|
||||
autoView(triangle_v, triangle, AcceleratorWrite);
|
||||
|
||||
// NOTE: this function cannot be 'private' since nvcc forbids this for kernels
|
||||
accelerator_for(ss, full.Grid()->oSites(), 1, {
|
||||
for(int s_row = 0; s_row < Ns; s_row++) {
|
||||
for(int s_col = 0; s_col < Ns; s_col++) {
|
||||
if(abs(s_row - s_col) > 1 || s_row + s_col == 3) continue;
|
||||
int block = s_row / Nhs;
|
||||
int s_row_block = s_row % Nhs;
|
||||
int s_col_block = s_col % Nhs;
|
||||
for(int c_row = 0; c_row < Nc; c_row++) {
|
||||
for(int c_col = 0; c_col < Nc; c_col++) {
|
||||
int i = s_row_block * Nc + c_row;
|
||||
int j = s_col_block * Nc + c_col;
|
||||
if(i == j)
|
||||
diagonal_v[ss]()(block)(i) = full_v[ss]()(s_row, s_col)(c_row, c_col);
|
||||
else if(i < j)
|
||||
triangle_v[ss]()(block)(triangle_index(i, j)) = full_v[ss]()(s_row, s_col)(c_row, c_col);
|
||||
else
|
||||
continue;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
});
|
||||
}
|
||||
|
||||
|
||||
static void ConvertLayout(const CloverDiagonalField& diagonal,
|
||||
const CloverTriangleField& triangle,
|
||||
CloverField& full) {
|
||||
conformable(full, diagonal);
|
||||
conformable(full, triangle);
|
||||
|
||||
full.Checkerboard() = diagonal.Checkerboard();
|
||||
|
||||
full = Zero();
|
||||
|
||||
autoView(diagonal_v, diagonal, AcceleratorRead);
|
||||
autoView(triangle_v, triangle, AcceleratorRead);
|
||||
autoView(full_v, full, AcceleratorWrite);
|
||||
|
||||
// NOTE: this function cannot be 'private' since nvcc forbids this for kernels
|
||||
accelerator_for(ss, full.Grid()->oSites(), 1, {
|
||||
for(int s_row = 0; s_row < Ns; s_row++) {
|
||||
for(int s_col = 0; s_col < Ns; s_col++) {
|
||||
if(abs(s_row - s_col) > 1 || s_row + s_col == 3) continue;
|
||||
int block = s_row / Nhs;
|
||||
int s_row_block = s_row % Nhs;
|
||||
int s_col_block = s_col % Nhs;
|
||||
for(int c_row = 0; c_row < Nc; c_row++) {
|
||||
for(int c_col = 0; c_col < Nc; c_col++) {
|
||||
int i = s_row_block * Nc + c_row;
|
||||
int j = s_col_block * Nc + c_col;
|
||||
if(i == j)
|
||||
full_v[ss]()(s_row, s_col)(c_row, c_col) = diagonal_v[ss]()(block)(i);
|
||||
else
|
||||
full_v[ss]()(s_row, s_col)(c_row, c_col) = triangle_elem(triangle_v[ss], block, i, j);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
});
|
||||
}
|
||||
|
||||
static void ModifyBoundaries(CloverDiagonalField& diagonal, CloverTriangleField& triangle, RealD csw_t, RealD cF, RealD diag_mass) {
|
||||
// Checks/grid
|
||||
double t0 = usecond();
|
||||
conformable(diagonal, triangle);
|
||||
GridBase* grid = diagonal.Grid();
|
||||
|
||||
// Determine the boundary coordinates/sites
|
||||
double t1 = usecond();
|
||||
int t_dir = Nd - 1;
|
||||
Lattice<iScalar<vInteger>> t_coor(grid);
|
||||
LatticeCoordinate(t_coor, t_dir);
|
||||
int T = grid->GlobalDimensions()[t_dir];
|
||||
|
||||
// Set off-diagonal parts at boundary to zero -- OK
|
||||
double t2 = usecond();
|
||||
CloverTriangleField zeroTriangle(grid);
|
||||
zeroTriangle.Checkerboard() = triangle.Checkerboard();
|
||||
zeroTriangle = Zero();
|
||||
triangle = where(t_coor == 0, zeroTriangle, triangle);
|
||||
triangle = where(t_coor == T-1, zeroTriangle, triangle);
|
||||
|
||||
// Set diagonal to unity (scaled correctly) -- OK
|
||||
double t3 = usecond();
|
||||
CloverDiagonalField tmp(grid);
|
||||
tmp.Checkerboard() = diagonal.Checkerboard();
|
||||
tmp = -1.0 * csw_t + diag_mass;
|
||||
diagonal = where(t_coor == 0, tmp, diagonal);
|
||||
diagonal = where(t_coor == T-1, tmp, diagonal);
|
||||
|
||||
// Correct values next to boundary
|
||||
double t4 = usecond();
|
||||
if(cF != 1.0) {
|
||||
tmp = cF - 1.0;
|
||||
tmp += diagonal;
|
||||
diagonal = where(t_coor == 1, tmp, diagonal);
|
||||
diagonal = where(t_coor == T-2, tmp, diagonal);
|
||||
}
|
||||
|
||||
// Report timings
|
||||
double t5 = usecond();
|
||||
#if 0
|
||||
std::cout << GridLogMessage << "CompactWilsonCloverHelpers::ModifyBoundaries timings:"
|
||||
<< " checks = " << (t1 - t0) / 1e6
|
||||
<< ", coordinate = " << (t2 - t1) / 1e6
|
||||
<< ", off-diag zero = " << (t3 - t2) / 1e6
|
||||
<< ", diagonal unity = " << (t4 - t3) / 1e6
|
||||
<< ", near-boundary = " << (t5 - t4) / 1e6
|
||||
<< ", total = " << (t5 - t0) / 1e6
|
||||
<< std::endl;
|
||||
#endif
|
||||
}
|
||||
|
||||
template<class Field, class Mask>
|
||||
static strong_inline void ApplyBoundaryMask(Field& f, const Mask& m) {
|
||||
conformable(f, m);
|
||||
auto grid = f.Grid();
|
||||
const uint32_t Nsite = grid->oSites();
|
||||
const uint32_t Nsimd = grid->Nsimd();
|
||||
autoView(f_v, f, AcceleratorWrite);
|
||||
autoView(m_v, m, AcceleratorRead);
|
||||
// NOTE: this function cannot be 'private' since nvcc forbids this for kernels
|
||||
accelerator_for(ss, Nsite, Nsimd, {
|
||||
coalescedWrite(f_v[ss], m_v(ss) * f_v(ss));
|
||||
});
|
||||
}
|
||||
|
||||
template<class MaskField>
|
||||
static void SetupMasks(MaskField& full, MaskField& even, MaskField& odd) {
|
||||
assert(even.Grid()->_isCheckerBoarded && even.Checkerboard() == Even);
|
||||
assert(odd.Grid()->_isCheckerBoarded && odd.Checkerboard() == Odd);
|
||||
assert(!full.Grid()->_isCheckerBoarded);
|
||||
|
||||
GridBase* grid = full.Grid();
|
||||
int t_dir = Nd-1;
|
||||
Lattice<iScalar<vInteger>> t_coor(grid);
|
||||
LatticeCoordinate(t_coor, t_dir);
|
||||
int T = grid->GlobalDimensions()[t_dir];
|
||||
|
||||
MaskField zeroMask(grid); zeroMask = Zero();
|
||||
full = 1.0;
|
||||
full = where(t_coor == 0, zeroMask, full);
|
||||
full = where(t_coor == T-1, zeroMask, full);
|
||||
|
||||
pickCheckerboard(Even, even, full);
|
||||
pickCheckerboard(Odd, odd, full);
|
||||
}
|
||||
};
|
||||
|
||||
NAMESPACE_END(Grid);
|
||||
@ -1,90 +0,0 @@
|
||||
/*************************************************************************************
|
||||
|
||||
Grid physics library, www.github.com/paboyle/Grid
|
||||
|
||||
Source file: ./lib/qcd/action/fermion/WilsonCloverTypes.h
|
||||
|
||||
Copyright (C) 2021 - 2022
|
||||
|
||||
Author: Daniel Richtmann <daniel.richtmann@gmail.com>
|
||||
|
||||
This program is free software; you can redistribute it and/or modify
|
||||
it under the terms of the GNU General Public License as published by
|
||||
the Free Software Foundation; either version 2 of the License, or
|
||||
(at your option) any later version.
|
||||
|
||||
This program is distributed in the hope that it will be useful,
|
||||
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||||
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
||||
GNU General Public License for more details.
|
||||
|
||||
You should have received a copy of the GNU General Public License along
|
||||
with this program; if not, write to the Free Software Foundation, Inc.,
|
||||
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
|
||||
|
||||
See the full license in the file "LICENSE" in the top level distribution directory
|
||||
*************************************************************************************/
|
||||
/* END LEGAL */
|
||||
|
||||
#pragma once
|
||||
|
||||
NAMESPACE_BEGIN(Grid);
|
||||
|
||||
template<class Impl>
|
||||
class WilsonCloverTypes {
|
||||
public:
|
||||
INHERIT_IMPL_TYPES(Impl);
|
||||
|
||||
template <typename vtype> using iImplClover = iScalar<iMatrix<iMatrix<vtype, Impl::Dimension>, Ns>>;
|
||||
|
||||
typedef iImplClover<Simd> SiteClover;
|
||||
|
||||
typedef Lattice<SiteClover> CloverField;
|
||||
};
|
||||
|
||||
template<class Impl>
|
||||
class CompactWilsonCloverTypes {
|
||||
public:
|
||||
INHERIT_IMPL_TYPES(Impl);
|
||||
|
||||
static constexpr int Nred = Nc * Nhs; // 6
|
||||
static constexpr int Nblock = Nhs; // 2
|
||||
static constexpr int Ndiagonal = Nred; // 6
|
||||
static constexpr int Ntriangle = (Nred - 1) * Nc; // 15
|
||||
|
||||
template<typename vtype> using iImplCloverDiagonal = iScalar<iVector<iVector<vtype, Ndiagonal>, Nblock>>;
|
||||
template<typename vtype> using iImplCloverTriangle = iScalar<iVector<iVector<vtype, Ntriangle>, Nblock>>;
|
||||
|
||||
typedef iImplCloverDiagonal<Simd> SiteCloverDiagonal;
|
||||
typedef iImplCloverTriangle<Simd> SiteCloverTriangle;
|
||||
typedef iSinglet<Simd> SiteMask;
|
||||
|
||||
typedef Lattice<SiteCloverDiagonal> CloverDiagonalField;
|
||||
typedef Lattice<SiteCloverTriangle> CloverTriangleField;
|
||||
typedef Lattice<SiteMask> MaskField;
|
||||
};
|
||||
|
||||
#define INHERIT_CLOVER_TYPES(Impl) \
|
||||
typedef typename WilsonCloverTypes<Impl>::SiteClover SiteClover; \
|
||||
typedef typename WilsonCloverTypes<Impl>::CloverField CloverField;
|
||||
|
||||
#define INHERIT_COMPACT_CLOVER_TYPES(Impl) \
|
||||
typedef typename CompactWilsonCloverTypes<Impl>::SiteCloverDiagonal SiteCloverDiagonal; \
|
||||
typedef typename CompactWilsonCloverTypes<Impl>::SiteCloverTriangle SiteCloverTriangle; \
|
||||
typedef typename CompactWilsonCloverTypes<Impl>::SiteMask SiteMask; \
|
||||
typedef typename CompactWilsonCloverTypes<Impl>::CloverDiagonalField CloverDiagonalField; \
|
||||
typedef typename CompactWilsonCloverTypes<Impl>::CloverTriangleField CloverTriangleField; \
|
||||
typedef typename CompactWilsonCloverTypes<Impl>::MaskField MaskField; \
|
||||
/* ugly duplication but needed inside functionality classes */ \
|
||||
template<typename vtype> using iImplCloverDiagonal = \
|
||||
iScalar<iVector<iVector<vtype, CompactWilsonCloverTypes<Impl>::Ndiagonal>, CompactWilsonCloverTypes<Impl>::Nblock>>; \
|
||||
template<typename vtype> using iImplCloverTriangle = \
|
||||
iScalar<iVector<iVector<vtype, CompactWilsonCloverTypes<Impl>::Ntriangle>, CompactWilsonCloverTypes<Impl>::Nblock>>;
|
||||
|
||||
#define INHERIT_COMPACT_CLOVER_SIZES(Impl) \
|
||||
static constexpr int Nred = CompactWilsonCloverTypes<Impl>::Nred; \
|
||||
static constexpr int Nblock = CompactWilsonCloverTypes<Impl>::Nblock; \
|
||||
static constexpr int Ndiagonal = CompactWilsonCloverTypes<Impl>::Ndiagonal; \
|
||||
static constexpr int Ntriangle = CompactWilsonCloverTypes<Impl>::Ntriangle;
|
||||
|
||||
NAMESPACE_END(Grid);
|
||||
@ -68,12 +68,11 @@ public:
|
||||
/*****************************************************/
|
||||
/* Compress includes precision change if mpi data is not same */
|
||||
/*****************************************************/
|
||||
accelerator_inline void Compress(SiteHalfSpinor &buf,const SiteSpinor &in) const {
|
||||
typedef decltype(coalescedRead(buf)) sobj;
|
||||
sobj sp;
|
||||
auto sin = coalescedRead(in);
|
||||
projector::Proj(sp,sin,mu,dag);
|
||||
coalescedWrite(buf,sp);
|
||||
template<class _SiteHalfSpinor, class _SiteSpinor>
|
||||
accelerator_inline void Compress(_SiteHalfSpinor *buf,Integer o,const _SiteSpinor &in) const {
|
||||
_SiteHalfSpinor tmp;
|
||||
projector::Proj(tmp,in,mu,dag);
|
||||
vstream(buf[o],tmp);
|
||||
}
|
||||
|
||||
/*****************************************************/
|
||||
@ -83,18 +82,13 @@ public:
|
||||
const SiteHalfSpinor * __restrict__ vp0,
|
||||
const SiteHalfSpinor * __restrict__ vp1,
|
||||
Integer type,Integer o) const {
|
||||
#ifdef GRID_SIMT
|
||||
exchangeSIMT(mp[2*o],mp[2*o+1],vp0[o],vp1[o],type);
|
||||
#else
|
||||
SiteHalfSpinor tmp1;
|
||||
SiteHalfSpinor tmp2;
|
||||
exchange(tmp1,tmp2,vp0[o],vp1[o],type);
|
||||
vstream(mp[2*o ],tmp1);
|
||||
vstream(mp[2*o+1],tmp2);
|
||||
#endif
|
||||
}
|
||||
|
||||
|
||||
/*****************************************************/
|
||||
/* Have a decompression step if mpi data is not same */
|
||||
/*****************************************************/
|
||||
@ -111,28 +105,6 @@ public:
|
||||
const SiteSpinor * __restrict__ in,
|
||||
Integer j,Integer k, Integer m,Integer type) const
|
||||
{
|
||||
#ifdef GRID_SIMT
|
||||
typedef SiteSpinor vobj;
|
||||
typedef SiteHalfSpinor hvobj;
|
||||
typedef decltype(coalescedRead(*in)) sobj;
|
||||
typedef decltype(coalescedRead(*out0)) hsobj;
|
||||
|
||||
unsigned int Nsimd = vobj::Nsimd();
|
||||
unsigned int mask = Nsimd >> (type + 1);
|
||||
int lane = acceleratorSIMTlane(Nsimd);
|
||||
int j0 = lane &(~mask); // inner coor zero
|
||||
int j1 = lane |(mask) ; // inner coor one
|
||||
const vobj *vp0 = &in[k];
|
||||
const vobj *vp1 = &in[m];
|
||||
const vobj *vp = (lane&mask) ? vp1:vp0;
|
||||
auto sa = coalescedRead(*vp,j0);
|
||||
auto sb = coalescedRead(*vp,j1);
|
||||
hsobj psa, psb;
|
||||
projector::Proj(psa,sa,mu,dag);
|
||||
projector::Proj(psb,sb,mu,dag);
|
||||
coalescedWrite(out0[j],psa);
|
||||
coalescedWrite(out1[j],psb);
|
||||
#else
|
||||
SiteHalfSpinor temp1, temp2;
|
||||
SiteHalfSpinor temp3, temp4;
|
||||
projector::Proj(temp1,in[k],mu,dag);
|
||||
@ -140,7 +112,6 @@ public:
|
||||
exchange(temp3,temp4,temp1,temp2,type);
|
||||
vstream(out0[j],temp3);
|
||||
vstream(out1[j],temp4);
|
||||
#endif
|
||||
}
|
||||
|
||||
/*****************************************************/
|
||||
@ -150,7 +121,6 @@ public:
|
||||
|
||||
};
|
||||
|
||||
#if 0
|
||||
template<class _HCspinor,class _Hspinor,class _Spinor, class projector>
|
||||
class WilsonCompressorTemplate< _HCspinor, _Hspinor, _Spinor, projector,
|
||||
typename std::enable_if<!std::is_same<_HCspinor,_Hspinor>::value>::type >
|
||||
@ -179,23 +149,13 @@ public:
|
||||
/*****************************************************/
|
||||
/* Compress includes precision change if mpi data is not same */
|
||||
/*****************************************************/
|
||||
accelerator_inline void Compress(SiteHalfSpinor &buf,const SiteSpinor &in) const {
|
||||
SiteHalfSpinor hsp;
|
||||
template<class _SiteHalfSpinor, class _SiteSpinor>
|
||||
accelerator_inline void Compress(_SiteHalfSpinor *buf,Integer o,const _SiteSpinor &in) const {
|
||||
_SiteHalfSpinor hsp;
|
||||
SiteHalfCommSpinor *hbuf = (SiteHalfCommSpinor *)buf;
|
||||
projector::Proj(hsp,in,mu,dag);
|
||||
precisionChange((vComplexLow *)&hbuf[o],(vComplexHigh *)&hsp,Nw);
|
||||
}
|
||||
accelerator_inline void Compress(SiteHalfSpinor &buf,const SiteSpinor &in) const {
|
||||
#ifdef GRID_SIMT
|
||||
typedef decltype(coalescedRead(buf)) sobj;
|
||||
sobj sp;
|
||||
auto sin = coalescedRead(in);
|
||||
projector::Proj(sp,sin,mu,dag);
|
||||
coalescedWrite(buf,sp);
|
||||
#else
|
||||
projector::Proj(buf,in,mu,dag);
|
||||
#endif
|
||||
}
|
||||
|
||||
/*****************************************************/
|
||||
/* Exchange includes precision change if mpi data is not same */
|
||||
@ -243,7 +203,6 @@ public:
|
||||
accelerator_inline bool DecompressionStep(void) const { return true; }
|
||||
|
||||
};
|
||||
#endif
|
||||
|
||||
#define DECLARE_PROJ(Projector,Compressor,spProj) \
|
||||
class Projector { \
|
||||
@ -294,10 +253,35 @@ public:
|
||||
typedef typename Base::View_type View_type;
|
||||
typedef typename Base::StencilVector StencilVector;
|
||||
|
||||
void ZeroCountersi(void) { }
|
||||
void Reporti(int calls) { }
|
||||
double timer0;
|
||||
double timer1;
|
||||
double timer2;
|
||||
double timer3;
|
||||
double timer4;
|
||||
double timer5;
|
||||
double timer6;
|
||||
uint64_t callsi;
|
||||
void ZeroCountersi(void)
|
||||
{
|
||||
timer0=0;
|
||||
timer1=0;
|
||||
timer2=0;
|
||||
timer3=0;
|
||||
timer4=0;
|
||||
timer5=0;
|
||||
timer6=0;
|
||||
callsi=0;
|
||||
}
|
||||
void Reporti(int calls)
|
||||
{
|
||||
if ( timer0 ) std::cout << GridLogMessage << " timer0 (HaloGatherOpt) " <<timer0/calls <<std::endl;
|
||||
if ( timer1 ) std::cout << GridLogMessage << " timer1 (Communicate) " <<timer1/calls <<std::endl;
|
||||
if ( timer2 ) std::cout << GridLogMessage << " timer2 (CommsMerge ) " <<timer2/calls <<std::endl;
|
||||
if ( timer3 ) std::cout << GridLogMessage << " timer3 (commsMergeShm) " <<timer3/calls <<std::endl;
|
||||
if ( timer4 ) std::cout << GridLogMessage << " timer4 " <<timer4 <<std::endl;
|
||||
}
|
||||
|
||||
// Vector<int> surface_list;
|
||||
std::vector<int> surface_list;
|
||||
|
||||
WilsonStencil(GridBase *grid,
|
||||
int npoints,
|
||||
@ -307,11 +291,10 @@ public:
|
||||
: CartesianStencil<vobj,cobj,Parameters> (grid,npoints,checkerboard,directions,distances,p)
|
||||
{
|
||||
ZeroCountersi();
|
||||
// surface_list.resize(0);
|
||||
surface_list.resize(0);
|
||||
this->same_node.resize(npoints);
|
||||
};
|
||||
|
||||
/*
|
||||
void BuildSurfaceList(int Ls,int vol4){
|
||||
|
||||
// find same node for SHM
|
||||
@ -332,25 +315,32 @@ public:
|
||||
}
|
||||
}
|
||||
}
|
||||
*/
|
||||
|
||||
|
||||
template < class compressor>
|
||||
void HaloExchangeOpt(const Lattice<vobj> &source,compressor &compress)
|
||||
{
|
||||
std::vector<std::vector<CommsRequest_t> > reqs;
|
||||
this->HaloExchangeOptGather(source,compress);
|
||||
double t1=usecond();
|
||||
// Asynchronous MPI calls multidirectional, Isend etc...
|
||||
// Non-overlapped directions within a thread. Asynchronous calls except MPI3, threaded up to comm threads ways.
|
||||
this->Communicate();
|
||||
double t2=usecond(); timer1 += t2-t1;
|
||||
this->CommsMerge(compress);
|
||||
double t3=usecond(); timer2 += t3-t2;
|
||||
this->CommsMergeSHM(compress);
|
||||
double t4=usecond(); timer3 += t4-t3;
|
||||
}
|
||||
|
||||
template <class compressor>
|
||||
void HaloExchangeOptGather(const Lattice<vobj> &source,compressor &compress)
|
||||
{
|
||||
this->Prepare();
|
||||
double t0=usecond();
|
||||
this->HaloGatherOpt(source,compress);
|
||||
double t1=usecond();
|
||||
timer0 += t1-t0;
|
||||
callsi++;
|
||||
}
|
||||
|
||||
template <class compressor>
|
||||
@ -362,9 +352,12 @@ public:
|
||||
typedef typename compressor::SiteHalfSpinor SiteHalfSpinor;
|
||||
typedef typename compressor::SiteHalfCommSpinor SiteHalfCommSpinor;
|
||||
|
||||
this->mpi3synctime_g-=usecond();
|
||||
this->_grid->StencilBarrier();
|
||||
this->mpi3synctime_g+=usecond();
|
||||
|
||||
assert(source.Grid()==this->_grid);
|
||||
this->halogtime-=usecond();
|
||||
|
||||
this->u_comm_offset=0;
|
||||
|
||||
@ -400,6 +393,7 @@ public:
|
||||
}
|
||||
this->face_table_computed=1;
|
||||
assert(this->u_comm_offset==this->_unified_buffer_size);
|
||||
this->halogtime+=usecond();
|
||||
accelerator_barrier();
|
||||
}
|
||||
|
||||
|
||||
@ -75,10 +75,6 @@ public:
|
||||
FermionField _tmp;
|
||||
FermionField &tmp(void) { return _tmp; }
|
||||
|
||||
int Dirichlet;
|
||||
Coordinate Block;
|
||||
|
||||
/********** Deprecate timers **********/
|
||||
void Report(void);
|
||||
void ZeroCounters(void);
|
||||
double DhopCalls;
|
||||
@ -177,10 +173,7 @@ public:
|
||||
GridCartesian &FourDimGrid,
|
||||
GridRedBlackCartesian &FourDimRedBlackGrid,
|
||||
double _M5,const ImplParams &p= ImplParams());
|
||||
|
||||
virtual void DirichletBlock(const Coordinate & block)
|
||||
{
|
||||
}
|
||||
|
||||
// Constructors
|
||||
/*
|
||||
WilsonFermion5D(int simd,
|
||||
|
||||
@ -243,17 +243,17 @@ typedef WilsonImpl<vComplex, FundamentalRepresentation, CoeffReal > WilsonImplR
|
||||
typedef WilsonImpl<vComplexF, FundamentalRepresentation, CoeffReal > WilsonImplF; // Float
|
||||
typedef WilsonImpl<vComplexD, FundamentalRepresentation, CoeffReal > WilsonImplD; // Double
|
||||
|
||||
//typedef WilsonImpl<vComplex, FundamentalRepresentation, CoeffRealHalfComms > WilsonImplRL; // Real.. whichever prec
|
||||
//typedef WilsonImpl<vComplexF, FundamentalRepresentation, CoeffRealHalfComms > WilsonImplFH; // Float
|
||||
//typedef WilsonImpl<vComplexD, FundamentalRepresentation, CoeffRealHalfComms > WilsonImplDF; // Double
|
||||
typedef WilsonImpl<vComplex, FundamentalRepresentation, CoeffRealHalfComms > WilsonImplRL; // Real.. whichever prec
|
||||
typedef WilsonImpl<vComplexF, FundamentalRepresentation, CoeffRealHalfComms > WilsonImplFH; // Float
|
||||
typedef WilsonImpl<vComplexD, FundamentalRepresentation, CoeffRealHalfComms > WilsonImplDF; // Double
|
||||
|
||||
typedef WilsonImpl<vComplex, FundamentalRepresentation, CoeffComplex > ZWilsonImplR; // Real.. whichever prec
|
||||
typedef WilsonImpl<vComplexF, FundamentalRepresentation, CoeffComplex > ZWilsonImplF; // Float
|
||||
typedef WilsonImpl<vComplexD, FundamentalRepresentation, CoeffComplex > ZWilsonImplD; // Double
|
||||
|
||||
//typedef WilsonImpl<vComplex, FundamentalRepresentation, CoeffComplexHalfComms > ZWilsonImplRL; // Real.. whichever prec
|
||||
//typedef WilsonImpl<vComplexF, FundamentalRepresentation, CoeffComplexHalfComms > ZWilsonImplFH; // Float
|
||||
//typedef WilsonImpl<vComplexD, FundamentalRepresentation, CoeffComplexHalfComms > ZWilsonImplDF; // Double
|
||||
typedef WilsonImpl<vComplex, FundamentalRepresentation, CoeffComplexHalfComms > ZWilsonImplRL; // Real.. whichever prec
|
||||
typedef WilsonImpl<vComplexF, FundamentalRepresentation, CoeffComplexHalfComms > ZWilsonImplFH; // Float
|
||||
typedef WilsonImpl<vComplexD, FundamentalRepresentation, CoeffComplexHalfComms > ZWilsonImplDF; // Double
|
||||
|
||||
typedef WilsonImpl<vComplex, AdjointRepresentation, CoeffReal > WilsonAdjImplR; // Real.. whichever prec
|
||||
typedef WilsonImpl<vComplexF, AdjointRepresentation, CoeffReal > WilsonAdjImplF; // Float
|
||||
|
||||
@ -47,7 +47,7 @@ CayleyFermion5D<Impl>::CayleyFermion5D(GaugeField &_Umu,
|
||||
FiveDimRedBlackGrid,
|
||||
FourDimGrid,
|
||||
FourDimRedBlackGrid,_M5,p),
|
||||
mass_plus(_mass), mass_minus(_mass)
|
||||
mass(_mass)
|
||||
{
|
||||
}
|
||||
|
||||
@ -209,8 +209,8 @@ void CayleyFermion5D<Impl>::M5D (const FermionField &psi, FermionField &chi)
|
||||
{
|
||||
int Ls=this->Ls;
|
||||
Vector<Coeff_t> diag (Ls,1.0);
|
||||
Vector<Coeff_t> upper(Ls,-1.0); upper[Ls-1]=mass_minus;
|
||||
Vector<Coeff_t> lower(Ls,-1.0); lower[0] =mass_plus;
|
||||
Vector<Coeff_t> upper(Ls,-1.0); upper[Ls-1]=mass;
|
||||
Vector<Coeff_t> lower(Ls,-1.0); lower[0] =mass;
|
||||
M5D(psi,chi,chi,lower,diag,upper);
|
||||
}
|
||||
template<class Impl>
|
||||
@ -220,8 +220,8 @@ void CayleyFermion5D<Impl>::Meooe5D (const FermionField &psi, FermionField &D
|
||||
Vector<Coeff_t> diag = bs;
|
||||
Vector<Coeff_t> upper= cs;
|
||||
Vector<Coeff_t> lower= cs;
|
||||
upper[Ls-1]=-mass_minus*upper[Ls-1];
|
||||
lower[0] =-mass_plus*lower[0];
|
||||
upper[Ls-1]=-mass*upper[Ls-1];
|
||||
lower[0] =-mass*lower[0];
|
||||
M5D(psi,psi,Din,lower,diag,upper);
|
||||
}
|
||||
// FIXME Redunant with the above routine; check this and eliminate
|
||||
@ -235,8 +235,8 @@ template<class Impl> void CayleyFermion5D<Impl>::Meo5D (const FermionField &
|
||||
upper[i]=-ceo[i];
|
||||
lower[i]=-ceo[i];
|
||||
}
|
||||
upper[Ls-1]=-mass_minus*upper[Ls-1];
|
||||
lower[0] =-mass_plus*lower[0];
|
||||
upper[Ls-1]=-mass*upper[Ls-1];
|
||||
lower[0] =-mass*lower[0];
|
||||
M5D(psi,psi,chi,lower,diag,upper);
|
||||
}
|
||||
template<class Impl>
|
||||
@ -250,8 +250,8 @@ void CayleyFermion5D<Impl>::Mooee (const FermionField &psi, FermionField &
|
||||
upper[i]=-cee[i];
|
||||
lower[i]=-cee[i];
|
||||
}
|
||||
upper[Ls-1]=-mass_minus*upper[Ls-1];
|
||||
lower[0] =-mass_plus*lower[0];
|
||||
upper[Ls-1]=-mass*upper[Ls-1];
|
||||
lower[0] =-mass*lower[0];
|
||||
M5D(psi,psi,chi,lower,diag,upper);
|
||||
}
|
||||
template<class Impl>
|
||||
@ -266,9 +266,9 @@ void CayleyFermion5D<Impl>::MooeeDag (const FermionField &psi, FermionField &
|
||||
// Assemble the 5d matrix
|
||||
if ( s==0 ) {
|
||||
upper[s] = -cee[s+1] ;
|
||||
lower[s] = mass_minus*cee[Ls-1];
|
||||
lower[s] = mass*cee[Ls-1];
|
||||
} else if ( s==(Ls-1)) {
|
||||
upper[s] = mass_plus*cee[0];
|
||||
upper[s] = mass*cee[0];
|
||||
lower[s] = -cee[s-1];
|
||||
} else {
|
||||
upper[s]=-cee[s+1];
|
||||
@ -291,8 +291,8 @@ void CayleyFermion5D<Impl>::M5Ddag (const FermionField &psi, FermionField &chi)
|
||||
Vector<Coeff_t> diag(Ls,1.0);
|
||||
Vector<Coeff_t> upper(Ls,-1.0);
|
||||
Vector<Coeff_t> lower(Ls,-1.0);
|
||||
upper[Ls-1]=-mass_plus*upper[Ls-1];
|
||||
lower[0] =-mass_minus*lower[0];
|
||||
upper[Ls-1]=-mass*upper[Ls-1];
|
||||
lower[0] =-mass*lower[0];
|
||||
M5Ddag(psi,chi,chi,lower,diag,upper);
|
||||
}
|
||||
|
||||
@ -307,9 +307,9 @@ void CayleyFermion5D<Impl>::MeooeDag5D (const FermionField &psi, FermionField
|
||||
for (int s=0;s<Ls;s++){
|
||||
if ( s== 0 ) {
|
||||
upper[s] = cs[s+1];
|
||||
lower[s] =-mass_minus*cs[Ls-1];
|
||||
lower[s] =-mass*cs[Ls-1];
|
||||
} else if ( s==(Ls-1) ) {
|
||||
upper[s] =-mass_plus*cs[0];
|
||||
upper[s] =-mass*cs[0];
|
||||
lower[s] = cs[s-1];
|
||||
} else {
|
||||
upper[s] = cs[s+1];
|
||||
@ -552,7 +552,7 @@ void CayleyFermion5D<Impl>::SetCoefficientsInternal(RealD zolo_hi,Vector<Coeff_t
|
||||
|
||||
lee[i] =-cee[i+1]/bee[i]; // sub-diag entry on the ith column
|
||||
|
||||
leem[i]=mass_minus*cee[Ls-1]/bee[0];
|
||||
leem[i]=mass*cee[Ls-1]/bee[0];
|
||||
for(int j=0;j<i;j++) {
|
||||
assert(bee[j+1]!=Coeff_t(0.0));
|
||||
leem[i]*= aee[j]/bee[j+1];
|
||||
@ -560,7 +560,7 @@ void CayleyFermion5D<Impl>::SetCoefficientsInternal(RealD zolo_hi,Vector<Coeff_t
|
||||
|
||||
uee[i] =-aee[i]/bee[i]; // up-diag entry on the ith row
|
||||
|
||||
ueem[i]=mass_plus;
|
||||
ueem[i]=mass;
|
||||
for(int j=1;j<=i;j++) ueem[i]*= cee[j]/bee[j];
|
||||
ueem[i]*= aee[0]/bee[0];
|
||||
|
||||
@ -573,7 +573,7 @@ void CayleyFermion5D<Impl>::SetCoefficientsInternal(RealD zolo_hi,Vector<Coeff_t
|
||||
}
|
||||
|
||||
{
|
||||
Coeff_t delta_d=mass_minus*cee[Ls-1];
|
||||
Coeff_t delta_d=mass*cee[Ls-1];
|
||||
for(int j=0;j<Ls-1;j++) {
|
||||
assert(bee[j] != Coeff_t(0.0));
|
||||
delta_d *= cee[j]/bee[j];
|
||||
@ -642,10 +642,6 @@ void CayleyFermion5D<Impl>::ContractConservedCurrent( PropagatorField &q_in_1,
|
||||
Current curr_type,
|
||||
unsigned int mu)
|
||||
{
|
||||
|
||||
assert(mass_plus == mass_minus);
|
||||
RealD mass = mass_plus;
|
||||
|
||||
#if (!defined(GRID_HIP))
|
||||
Gamma::Algebra Gmu [] = {
|
||||
Gamma::Algebra::GammaX,
|
||||
@ -781,8 +777,6 @@ void CayleyFermion5D<Impl>::SeqConservedCurrent(PropagatorField &q_in,
|
||||
assert(mu>=0);
|
||||
assert(mu<Nd);
|
||||
|
||||
assert(mass_plus == mass_minus);
|
||||
RealD mass = mass_plus;
|
||||
|
||||
#if 0
|
||||
int tshift = (mu == Nd-1) ? 1 : 0;
|
||||
@ -834,7 +828,6 @@ void CayleyFermion5D<Impl>::SeqConservedCurrent(PropagatorField &q_in,
|
||||
|
||||
#if (!defined(GRID_HIP))
|
||||
int tshift = (mu == Nd-1) ? 1 : 0;
|
||||
unsigned int LLt = GridDefaultLatt()[Tp];
|
||||
////////////////////////////////////////////////
|
||||
// GENERAL CAYLEY CASE
|
||||
////////////////////////////////////////////////
|
||||
@ -887,7 +880,7 @@ void CayleyFermion5D<Impl>::SeqConservedCurrent(PropagatorField &q_in,
|
||||
}
|
||||
|
||||
std::vector<RealD> G_s(Ls,1.0);
|
||||
RealD sign = 1.0; // sign flip for vector/tadpole
|
||||
Integer sign = 1; // sign flip for vector/tadpole
|
||||
if ( curr_type == Current::Axial ) {
|
||||
for(int s=0;s<Ls/2;s++){
|
||||
G_s[s] = -1.0;
|
||||
@ -897,7 +890,7 @@ void CayleyFermion5D<Impl>::SeqConservedCurrent(PropagatorField &q_in,
|
||||
auto b=this->_b;
|
||||
auto c=this->_c;
|
||||
if ( b == 1 && c == 0 ) {
|
||||
sign = -1.0;
|
||||
sign = -1;
|
||||
}
|
||||
else {
|
||||
std::cerr << "Error: Tadpole implementation currently unavailable for non-Shamir actions." << std::endl;
|
||||
@ -908,8 +901,8 @@ void CayleyFermion5D<Impl>::SeqConservedCurrent(PropagatorField &q_in,
|
||||
for(int s=0;s<Ls;s++){
|
||||
|
||||
int sp = (s+1)%Ls;
|
||||
// int sr = Ls-1-s;
|
||||
// int srp= (sr+1)%Ls;
|
||||
int sr = Ls-1-s;
|
||||
int srp= (sr+1)%Ls;
|
||||
|
||||
// Mobius parameters
|
||||
auto b=this->bs[s];
|
||||
@ -941,13 +934,7 @@ void CayleyFermion5D<Impl>::SeqConservedCurrent(PropagatorField &q_in,
|
||||
tmp = Cshift(tmp,mu,-1);
|
||||
Impl::multLinkField(Utmp,this->Umu,tmp,mu+Nd); // Adjoint link
|
||||
tmp = -G_s[s]*( Utmp + gmu*Utmp );
|
||||
// Mask the time
|
||||
if (tmax == LLt - 1 && tshift == 1){ // quick fix to include timeslice 0 if tmax + tshift is over the last timeslice
|
||||
unsigned int t0 = 0;
|
||||
tmp = where(((lcoor==t0) || (lcoor>=tmin+tshift)),tmp,zz);
|
||||
} else {
|
||||
tmp = where((lcoor>=tmin+tshift),tmp,zz);
|
||||
}
|
||||
tmp = where((lcoor>=tmin+tshift),tmp,zz); // Mask the time
|
||||
L_Q += where((lcoor<=tmax+tshift),tmp,zz); // Position of current complicated
|
||||
|
||||
InsertSlice(L_Q, q_out, s , 0);
|
||||
|
||||
@ -66,17 +66,18 @@ CayleyFermion5D<Impl>::M5D(const FermionField &psi_i,
|
||||
M5Dcalls++;
|
||||
M5Dtime-=usecond();
|
||||
|
||||
uint64_t nloop = grid->oSites();
|
||||
uint64_t nloop = grid->oSites()/Ls;
|
||||
accelerator_for(sss,nloop,Simd::Nsimd(),{
|
||||
uint64_t s = sss%Ls;
|
||||
uint64_t ss= sss-s;
|
||||
uint64_t ss= sss*Ls;
|
||||
typedef decltype(coalescedRead(psi[0])) spinor;
|
||||
spinor tmp1, tmp2;
|
||||
uint64_t idx_u = ss+((s+1)%Ls);
|
||||
uint64_t idx_l = ss+((s+Ls-1)%Ls);
|
||||
spProj5m(tmp1,psi(idx_u));
|
||||
spProj5p(tmp2,psi(idx_l));
|
||||
coalescedWrite(chi[ss+s],pdiag[s]*phi(ss+s)+pupper[s]*tmp1+plower[s]*tmp2);
|
||||
for(int s=0;s<Ls;s++){
|
||||
uint64_t idx_u = ss+((s+1)%Ls);
|
||||
uint64_t idx_l = ss+((s+Ls-1)%Ls);
|
||||
spProj5m(tmp1,psi(idx_u));
|
||||
spProj5p(tmp2,psi(idx_l));
|
||||
coalescedWrite(chi[ss+s],pdiag[s]*phi(ss+s)+pupper[s]*tmp1+plower[s]*tmp2);
|
||||
}
|
||||
});
|
||||
M5Dtime+=usecond();
|
||||
}
|
||||
@ -107,17 +108,18 @@ CayleyFermion5D<Impl>::M5Ddag(const FermionField &psi_i,
|
||||
M5Dcalls++;
|
||||
M5Dtime-=usecond();
|
||||
|
||||
uint64_t nloop = grid->oSites();
|
||||
uint64_t nloop = grid->oSites()/Ls;
|
||||
accelerator_for(sss,nloop,Simd::Nsimd(),{
|
||||
uint64_t s = sss%Ls;
|
||||
uint64_t ss= sss-s;
|
||||
uint64_t ss=sss*Ls;
|
||||
typedef decltype(coalescedRead(psi[0])) spinor;
|
||||
spinor tmp1,tmp2;
|
||||
uint64_t idx_u = ss+((s+1)%Ls);
|
||||
uint64_t idx_l = ss+((s+Ls-1)%Ls);
|
||||
spProj5p(tmp1,psi(idx_u));
|
||||
spProj5m(tmp2,psi(idx_l));
|
||||
coalescedWrite(chi[ss+s],pdiag[s]*phi(ss+s)+pupper[s]*tmp1+plower[s]*tmp2);
|
||||
for(int s=0;s<Ls;s++){
|
||||
uint64_t idx_u = ss+((s+1)%Ls);
|
||||
uint64_t idx_l = ss+((s+Ls-1)%Ls);
|
||||
spProj5p(tmp1,psi(idx_u));
|
||||
spProj5m(tmp2,psi(idx_l));
|
||||
coalescedWrite(chi[ss+s],pdiag[s]*phi(ss+s)+pupper[s]*tmp1+plower[s]*tmp2);
|
||||
}
|
||||
});
|
||||
M5Dtime+=usecond();
|
||||
}
|
||||
|
||||
@ -1,373 +0,0 @@
|
||||
/*************************************************************************************
|
||||
|
||||
Grid physics library, www.github.com/paboyle/Grid
|
||||
|
||||
Source file: ./lib/qcd/action/fermion/CompactWilsonCloverFermionImplementation.h
|
||||
|
||||
Copyright (C) 2017 - 2022
|
||||
|
||||
Author: paboyle <paboyle@ph.ed.ac.uk>
|
||||
Author: Guido Cossu <guido.cossu@ed.ac.uk>
|
||||
Author: Daniel Richtmann <daniel.richtmann@gmail.com>
|
||||
|
||||
This program is free software; you can redistribute it and/or modify
|
||||
it under the terms of the GNU General Public License as published by
|
||||
the Free Software Foundation; either version 2 of the License, or
|
||||
(at your option) any later version.
|
||||
|
||||
This program is distributed in the hope that it will be useful,
|
||||
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||||
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
||||
GNU General Public License for more details.
|
||||
|
||||
You should have received a copy of the GNU General Public License along
|
||||
with this program; if not, write to the Free Software Foundation, Inc.,
|
||||
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
|
||||
|
||||
See the full license in the file "LICENSE" in the top level distribution directory
|
||||
*************************************************************************************/
|
||||
/* END LEGAL */
|
||||
|
||||
#include <Grid/Grid.h>
|
||||
#include <Grid/qcd/spin/Dirac.h>
|
||||
#include <Grid/qcd/action/fermion/CompactWilsonCloverFermion.h>
|
||||
|
||||
|
||||
NAMESPACE_BEGIN(Grid);
|
||||
template<class Impl, class CloverHelpers>
|
||||
CompactWilsonCloverFermion<Impl, CloverHelpers>::CompactWilsonCloverFermion(GaugeField& _Umu,
|
||||
GridCartesian& Fgrid,
|
||||
GridRedBlackCartesian& Hgrid,
|
||||
const RealD _mass,
|
||||
const RealD _csw_r,
|
||||
const RealD _csw_t,
|
||||
const RealD _cF,
|
||||
const WilsonAnisotropyCoefficients& clover_anisotropy,
|
||||
const ImplParams& impl_p)
|
||||
: WilsonBase(_Umu, Fgrid, Hgrid, _mass, impl_p, clover_anisotropy)
|
||||
, csw_r(_csw_r)
|
||||
, csw_t(_csw_t)
|
||||
, cF(_cF)
|
||||
, open_boundaries(impl_p.boundary_phases[Nd-1] == 0.0)
|
||||
, Diagonal(&Fgrid), Triangle(&Fgrid)
|
||||
, DiagonalEven(&Hgrid), TriangleEven(&Hgrid)
|
||||
, DiagonalOdd(&Hgrid), TriangleOdd(&Hgrid)
|
||||
, DiagonalInv(&Fgrid), TriangleInv(&Fgrid)
|
||||
, DiagonalInvEven(&Hgrid), TriangleInvEven(&Hgrid)
|
||||
, DiagonalInvOdd(&Hgrid), TriangleInvOdd(&Hgrid)
|
||||
, Tmp(&Fgrid)
|
||||
, BoundaryMask(&Fgrid)
|
||||
, BoundaryMaskEven(&Hgrid), BoundaryMaskOdd(&Hgrid)
|
||||
{
|
||||
assert(Nd == 4 && Nc == 3 && Ns == 4 && Impl::Dimension == 3);
|
||||
|
||||
csw_r *= 0.5;
|
||||
csw_t *= 0.5;
|
||||
if (clover_anisotropy.isAnisotropic)
|
||||
csw_r /= clover_anisotropy.xi_0;
|
||||
|
||||
ImportGauge(_Umu);
|
||||
if (open_boundaries) {
|
||||
this->BoundaryMaskEven.Checkerboard() = Even;
|
||||
this->BoundaryMaskOdd.Checkerboard() = Odd;
|
||||
CompactHelpers::SetupMasks(this->BoundaryMask, this->BoundaryMaskEven, this->BoundaryMaskOdd);
|
||||
}
|
||||
}
|
||||
|
||||
template<class Impl, class CloverHelpers>
|
||||
void CompactWilsonCloverFermion<Impl, CloverHelpers>::Dhop(const FermionField& in, FermionField& out, int dag) {
|
||||
WilsonBase::Dhop(in, out, dag);
|
||||
if(open_boundaries) ApplyBoundaryMask(out);
|
||||
}
|
||||
|
||||
template<class Impl, class CloverHelpers>
|
||||
void CompactWilsonCloverFermion<Impl, CloverHelpers>::DhopOE(const FermionField& in, FermionField& out, int dag) {
|
||||
WilsonBase::DhopOE(in, out, dag);
|
||||
if(open_boundaries) ApplyBoundaryMask(out);
|
||||
}
|
||||
|
||||
template<class Impl, class CloverHelpers>
|
||||
void CompactWilsonCloverFermion<Impl, CloverHelpers>::DhopEO(const FermionField& in, FermionField& out, int dag) {
|
||||
WilsonBase::DhopEO(in, out, dag);
|
||||
if(open_boundaries) ApplyBoundaryMask(out);
|
||||
}
|
||||
|
||||
template<class Impl, class CloverHelpers>
|
||||
void CompactWilsonCloverFermion<Impl, CloverHelpers>::DhopDir(const FermionField& in, FermionField& out, int dir, int disp) {
|
||||
WilsonBase::DhopDir(in, out, dir, disp);
|
||||
if(this->open_boundaries) ApplyBoundaryMask(out);
|
||||
}
|
||||
|
||||
template<class Impl, class CloverHelpers>
|
||||
void CompactWilsonCloverFermion<Impl, CloverHelpers>::DhopDirAll(const FermionField& in, std::vector<FermionField>& out) {
|
||||
WilsonBase::DhopDirAll(in, out);
|
||||
if(this->open_boundaries) {
|
||||
for(auto& o : out) ApplyBoundaryMask(o);
|
||||
}
|
||||
}
|
||||
|
||||
template<class Impl, class CloverHelpers>
|
||||
void CompactWilsonCloverFermion<Impl, CloverHelpers>::M(const FermionField& in, FermionField& out) {
|
||||
out.Checkerboard() = in.Checkerboard();
|
||||
WilsonBase::Dhop(in, out, DaggerNo); // call base to save applying bc
|
||||
Mooee(in, Tmp);
|
||||
axpy(out, 1.0, out, Tmp);
|
||||
if(open_boundaries) ApplyBoundaryMask(out);
|
||||
}
|
||||
|
||||
template<class Impl, class CloverHelpers>
|
||||
void CompactWilsonCloverFermion<Impl, CloverHelpers>::Mdag(const FermionField& in, FermionField& out) {
|
||||
out.Checkerboard() = in.Checkerboard();
|
||||
WilsonBase::Dhop(in, out, DaggerYes); // call base to save applying bc
|
||||
MooeeDag(in, Tmp);
|
||||
axpy(out, 1.0, out, Tmp);
|
||||
if(open_boundaries) ApplyBoundaryMask(out);
|
||||
}
|
||||
|
||||
template<class Impl, class CloverHelpers>
|
||||
void CompactWilsonCloverFermion<Impl, CloverHelpers>::Meooe(const FermionField& in, FermionField& out) {
|
||||
WilsonBase::Meooe(in, out);
|
||||
if(open_boundaries) ApplyBoundaryMask(out);
|
||||
}
|
||||
|
||||
template<class Impl, class CloverHelpers>
|
||||
void CompactWilsonCloverFermion<Impl, CloverHelpers>::MeooeDag(const FermionField& in, FermionField& out) {
|
||||
WilsonBase::MeooeDag(in, out);
|
||||
if(open_boundaries) ApplyBoundaryMask(out);
|
||||
}
|
||||
|
||||
template<class Impl, class CloverHelpers>
|
||||
void CompactWilsonCloverFermion<Impl, CloverHelpers>::Mooee(const FermionField& in, FermionField& out) {
|
||||
if(in.Grid()->_isCheckerBoarded) {
|
||||
if(in.Checkerboard() == Odd) {
|
||||
MooeeInternal(in, out, DiagonalOdd, TriangleOdd);
|
||||
} else {
|
||||
MooeeInternal(in, out, DiagonalEven, TriangleEven);
|
||||
}
|
||||
} else {
|
||||
MooeeInternal(in, out, Diagonal, Triangle);
|
||||
}
|
||||
if(open_boundaries) ApplyBoundaryMask(out);
|
||||
}
|
||||
|
||||
template<class Impl, class CloverHelpers>
|
||||
void CompactWilsonCloverFermion<Impl, CloverHelpers>::MooeeDag(const FermionField& in, FermionField& out) {
|
||||
Mooee(in, out); // blocks are hermitian
|
||||
}
|
||||
|
||||
template<class Impl, class CloverHelpers>
|
||||
void CompactWilsonCloverFermion<Impl, CloverHelpers>::MooeeInv(const FermionField& in, FermionField& out) {
|
||||
if(in.Grid()->_isCheckerBoarded) {
|
||||
if(in.Checkerboard() == Odd) {
|
||||
MooeeInternal(in, out, DiagonalInvOdd, TriangleInvOdd);
|
||||
} else {
|
||||
MooeeInternal(in, out, DiagonalInvEven, TriangleInvEven);
|
||||
}
|
||||
} else {
|
||||
MooeeInternal(in, out, DiagonalInv, TriangleInv);
|
||||
}
|
||||
if(open_boundaries) ApplyBoundaryMask(out);
|
||||
}
|
||||
|
||||
template<class Impl, class CloverHelpers>
|
||||
void CompactWilsonCloverFermion<Impl, CloverHelpers>::MooeeInvDag(const FermionField& in, FermionField& out) {
|
||||
MooeeInv(in, out); // blocks are hermitian
|
||||
}
|
||||
|
||||
template<class Impl, class CloverHelpers>
|
||||
void CompactWilsonCloverFermion<Impl, CloverHelpers>::Mdir(const FermionField& in, FermionField& out, int dir, int disp) {
|
||||
DhopDir(in, out, dir, disp);
|
||||
}
|
||||
|
||||
template<class Impl, class CloverHelpers>
|
||||
void CompactWilsonCloverFermion<Impl, CloverHelpers>::MdirAll(const FermionField& in, std::vector<FermionField>& out) {
|
||||
DhopDirAll(in, out);
|
||||
}
|
||||
|
||||
template<class Impl, class CloverHelpers>
|
||||
void CompactWilsonCloverFermion<Impl, CloverHelpers>::MDeriv(GaugeField& force, const FermionField& X, const FermionField& Y, int dag) {
|
||||
assert(!open_boundaries); // TODO check for changes required for open bc
|
||||
|
||||
// NOTE: code copied from original clover term
|
||||
conformable(X.Grid(), Y.Grid());
|
||||
conformable(X.Grid(), force.Grid());
|
||||
GaugeLinkField force_mu(force.Grid()), lambda(force.Grid());
|
||||
GaugeField clover_force(force.Grid());
|
||||
PropagatorField Lambda(force.Grid());
|
||||
|
||||
// Guido: Here we are hitting some performance issues:
|
||||
// need to extract the components of the DoubledGaugeField
|
||||
// for each call
|
||||
// Possible solution
|
||||
// Create a vector object to store them? (cons: wasting space)
|
||||
std::vector<GaugeLinkField> U(Nd, this->Umu.Grid());
|
||||
|
||||
Impl::extractLinkField(U, this->Umu);
|
||||
|
||||
force = Zero();
|
||||
// Derivative of the Wilson hopping term
|
||||
this->DhopDeriv(force, X, Y, dag);
|
||||
|
||||
///////////////////////////////////////////////////////////
|
||||
// Clover term derivative
|
||||
///////////////////////////////////////////////////////////
|
||||
Impl::outerProductImpl(Lambda, X, Y);
|
||||
//std::cout << "Lambda:" << Lambda << std::endl;
|
||||
|
||||
Gamma::Algebra sigma[] = {
|
||||
Gamma::Algebra::SigmaXY,
|
||||
Gamma::Algebra::SigmaXZ,
|
||||
Gamma::Algebra::SigmaXT,
|
||||
Gamma::Algebra::MinusSigmaXY,
|
||||
Gamma::Algebra::SigmaYZ,
|
||||
Gamma::Algebra::SigmaYT,
|
||||
Gamma::Algebra::MinusSigmaXZ,
|
||||
Gamma::Algebra::MinusSigmaYZ,
|
||||
Gamma::Algebra::SigmaZT,
|
||||
Gamma::Algebra::MinusSigmaXT,
|
||||
Gamma::Algebra::MinusSigmaYT,
|
||||
Gamma::Algebra::MinusSigmaZT};
|
||||
|
||||
/*
|
||||
sigma_{\mu \nu}=
|
||||
| 0 sigma[0] sigma[1] sigma[2] |
|
||||
| sigma[3] 0 sigma[4] sigma[5] |
|
||||
| sigma[6] sigma[7] 0 sigma[8] |
|
||||
| sigma[9] sigma[10] sigma[11] 0 |
|
||||
*/
|
||||
|
||||
int count = 0;
|
||||
clover_force = Zero();
|
||||
for (int mu = 0; mu < 4; mu++)
|
||||
{
|
||||
force_mu = Zero();
|
||||
for (int nu = 0; nu < 4; nu++)
|
||||
{
|
||||
if (mu == nu)
|
||||
continue;
|
||||
|
||||
RealD factor;
|
||||
if (nu == 4 || mu == 4)
|
||||
{
|
||||
factor = 2.0 * csw_t;
|
||||
}
|
||||
else
|
||||
{
|
||||
factor = 2.0 * csw_r;
|
||||
}
|
||||
PropagatorField Slambda = Gamma(sigma[count]) * Lambda; // sigma checked
|
||||
Impl::TraceSpinImpl(lambda, Slambda); // traceSpin ok
|
||||
force_mu -= factor*CloverHelpers::Cmunu(U, lambda, mu, nu); // checked
|
||||
count++;
|
||||
}
|
||||
|
||||
pokeLorentz(clover_force, U[mu] * force_mu, mu);
|
||||
}
|
||||
//clover_force *= csw;
|
||||
force += clover_force;
|
||||
}
|
||||
|
||||
template<class Impl, class CloverHelpers>
|
||||
void CompactWilsonCloverFermion<Impl, CloverHelpers>::MooDeriv(GaugeField& mat, const FermionField& U, const FermionField& V, int dag) {
|
||||
assert(0);
|
||||
}
|
||||
|
||||
template<class Impl, class CloverHelpers>
|
||||
void CompactWilsonCloverFermion<Impl, CloverHelpers>::MeeDeriv(GaugeField& mat, const FermionField& U, const FermionField& V, int dag) {
|
||||
assert(0);
|
||||
}
|
||||
|
||||
template<class Impl, class CloverHelpers>
|
||||
void CompactWilsonCloverFermion<Impl, CloverHelpers>::MooeeInternal(const FermionField& in,
|
||||
FermionField& out,
|
||||
const CloverDiagonalField& diagonal,
|
||||
const CloverTriangleField& triangle) {
|
||||
assert(in.Checkerboard() == Odd || in.Checkerboard() == Even);
|
||||
out.Checkerboard() = in.Checkerboard();
|
||||
conformable(in, out);
|
||||
conformable(in, diagonal);
|
||||
conformable(in, triangle);
|
||||
|
||||
CompactHelpers::MooeeKernel(diagonal.oSites(), 1, in, out, diagonal, triangle);
|
||||
}
|
||||
|
||||
template<class Impl, class CloverHelpers>
|
||||
void CompactWilsonCloverFermion<Impl, CloverHelpers>::ImportGauge(const GaugeField& _Umu) {
|
||||
// NOTE: parts copied from original implementation
|
||||
|
||||
// Import gauge into base class
|
||||
double t0 = usecond();
|
||||
WilsonBase::ImportGauge(_Umu); // NOTE: called here and in wilson constructor -> performed twice, but can't avoid that
|
||||
|
||||
// Initialize temporary variables
|
||||
double t1 = usecond();
|
||||
conformable(_Umu.Grid(), this->GaugeGrid());
|
||||
GridBase* grid = _Umu.Grid();
|
||||
typename Impl::GaugeLinkField Bx(grid), By(grid), Bz(grid), Ex(grid), Ey(grid), Ez(grid);
|
||||
CloverField TmpOriginal(grid);
|
||||
|
||||
// Compute the field strength terms mu>nu
|
||||
double t2 = usecond();
|
||||
WilsonLoops<Impl>::FieldStrength(Bx, _Umu, Zdir, Ydir);
|
||||
WilsonLoops<Impl>::FieldStrength(By, _Umu, Zdir, Xdir);
|
||||
WilsonLoops<Impl>::FieldStrength(Bz, _Umu, Ydir, Xdir);
|
||||
WilsonLoops<Impl>::FieldStrength(Ex, _Umu, Tdir, Xdir);
|
||||
WilsonLoops<Impl>::FieldStrength(Ey, _Umu, Tdir, Ydir);
|
||||
WilsonLoops<Impl>::FieldStrength(Ez, _Umu, Tdir, Zdir);
|
||||
|
||||
// Compute the Clover Operator acting on Colour and Spin
|
||||
// multiply here by the clover coefficients for the anisotropy
|
||||
double t3 = usecond();
|
||||
TmpOriginal = Helpers::fillCloverYZ(Bx) * csw_r;
|
||||
TmpOriginal += Helpers::fillCloverXZ(By) * csw_r;
|
||||
TmpOriginal += Helpers::fillCloverXY(Bz) * csw_r;
|
||||
TmpOriginal += Helpers::fillCloverXT(Ex) * csw_t;
|
||||
TmpOriginal += Helpers::fillCloverYT(Ey) * csw_t;
|
||||
TmpOriginal += Helpers::fillCloverZT(Ez) * csw_t;
|
||||
// Handle mass term based on clover policy
|
||||
CloverHelpers::MassTerm(TmpOriginal, this->diag_mass);
|
||||
|
||||
// Convert the data layout of the clover term
|
||||
double t4 = usecond();
|
||||
CompactHelpers::ConvertLayout(TmpOriginal, Diagonal, Triangle);
|
||||
|
||||
// Exponentiate the clover (nothing happens in case of the standard clover)
|
||||
double t5 = usecond();
|
||||
CloverHelpers::Exponentiate_Clover(Diagonal, Triangle, csw_t, this->diag_mass);
|
||||
|
||||
// Possible modify the boundary values
|
||||
double t6 = usecond();
|
||||
if(open_boundaries) CompactHelpers::ModifyBoundaries(Diagonal, Triangle, csw_t, cF, this->diag_mass);
|
||||
|
||||
// Invert the Clover term (explicit inversion needed for the improvement in case of open boundary conditions)
|
||||
double t7 = usecond();
|
||||
CompactHelpers::Invert(Diagonal, Triangle, DiagonalInv, TriangleInv);
|
||||
|
||||
// Fill the remaining clover fields
|
||||
double t8 = usecond();
|
||||
pickCheckerboard(Even, DiagonalEven, Diagonal);
|
||||
pickCheckerboard(Even, TriangleEven, Triangle);
|
||||
pickCheckerboard(Odd, DiagonalOdd, Diagonal);
|
||||
pickCheckerboard(Odd, TriangleOdd, Triangle);
|
||||
pickCheckerboard(Even, DiagonalInvEven, DiagonalInv);
|
||||
pickCheckerboard(Even, TriangleInvEven, TriangleInv);
|
||||
pickCheckerboard(Odd, DiagonalInvOdd, DiagonalInv);
|
||||
pickCheckerboard(Odd, TriangleInvOdd, TriangleInv);
|
||||
|
||||
// Report timings
|
||||
double t9 = usecond();
|
||||
|
||||
std::cout << GridLogDebug << "CompactWilsonCloverFermion::ImportGauge timings:" << std::endl;
|
||||
std::cout << GridLogDebug << "WilsonFermion::Importgauge = " << (t1 - t0) / 1e6 << std::endl;
|
||||
std::cout << GridLogDebug << "allocations = " << (t2 - t1) / 1e6 << std::endl;
|
||||
std::cout << GridLogDebug << "field strength = " << (t3 - t2) / 1e6 << std::endl;
|
||||
std::cout << GridLogDebug << "fill clover = " << (t4 - t3) / 1e6 << std::endl;
|
||||
std::cout << GridLogDebug << "convert = " << (t5 - t4) / 1e6 << std::endl;
|
||||
std::cout << GridLogDebug << "exponentiation = " << (t6 - t5) / 1e6 << std::endl;
|
||||
std::cout << GridLogDebug << "boundaries = " << (t7 - t6) / 1e6 << std::endl;
|
||||
std::cout << GridLogDebug << "inversions = " << (t8 - t7) / 1e6 << std::endl;
|
||||
std::cout << GridLogDebug << "pick cbs = " << (t9 - t8) / 1e6 << std::endl;
|
||||
std::cout << GridLogDebug << "total = " << (t9 - t0) / 1e6 << std::endl;
|
||||
}
|
||||
|
||||
NAMESPACE_END(Grid);
|
||||
@ -2,13 +2,12 @@
|
||||
|
||||
Grid physics library, www.github.com/paboyle/Grid
|
||||
|
||||
Source file: ./lib/qcd/action/fermion/WilsonCloverFermionImplementation.h
|
||||
Source file: ./lib/qcd/action/fermion/WilsonCloverFermion.cc
|
||||
|
||||
Copyright (C) 2017 - 2022
|
||||
Copyright (C) 2017
|
||||
|
||||
Author: paboyle <paboyle@ph.ed.ac.uk>
|
||||
Author: Guido Cossu <guido.cossu@ed.ac.uk>
|
||||
Author: Daniel Richtmann <daniel.richtmann@gmail.com>
|
||||
|
||||
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
|
||||
@ -34,48 +33,9 @@
|
||||
|
||||
NAMESPACE_BEGIN(Grid);
|
||||
|
||||
template<class Impl, class CloverHelpers>
|
||||
WilsonCloverFermion<Impl, CloverHelpers>::WilsonCloverFermion(GaugeField& _Umu,
|
||||
GridCartesian& Fgrid,
|
||||
GridRedBlackCartesian& Hgrid,
|
||||
const RealD _mass,
|
||||
const RealD _csw_r,
|
||||
const RealD _csw_t,
|
||||
const WilsonAnisotropyCoefficients& clover_anisotropy,
|
||||
const ImplParams& impl_p)
|
||||
: WilsonFermion<Impl>(_Umu, Fgrid, Hgrid, _mass, impl_p, clover_anisotropy)
|
||||
, CloverTerm(&Fgrid)
|
||||
, CloverTermInv(&Fgrid)
|
||||
, CloverTermEven(&Hgrid)
|
||||
, CloverTermOdd(&Hgrid)
|
||||
, CloverTermInvEven(&Hgrid)
|
||||
, CloverTermInvOdd(&Hgrid)
|
||||
, CloverTermDagEven(&Hgrid)
|
||||
, CloverTermDagOdd(&Hgrid)
|
||||
, CloverTermInvDagEven(&Hgrid)
|
||||
, CloverTermInvDagOdd(&Hgrid) {
|
||||
assert(Nd == 4); // require 4 dimensions
|
||||
|
||||
if(clover_anisotropy.isAnisotropic) {
|
||||
csw_r = _csw_r * 0.5 / clover_anisotropy.xi_0;
|
||||
diag_mass = _mass + 1.0 + (Nd - 1) * (clover_anisotropy.nu / clover_anisotropy.xi_0);
|
||||
} else {
|
||||
csw_r = _csw_r * 0.5;
|
||||
diag_mass = 4.0 + _mass;
|
||||
}
|
||||
csw_t = _csw_t * 0.5;
|
||||
|
||||
if(csw_r == 0)
|
||||
std::cout << GridLogWarning << "Initializing WilsonCloverFermion with csw_r = 0" << std::endl;
|
||||
if(csw_t == 0)
|
||||
std::cout << GridLogWarning << "Initializing WilsonCloverFermion with csw_t = 0" << std::endl;
|
||||
|
||||
ImportGauge(_Umu);
|
||||
}
|
||||
|
||||
// *NOT* EO
|
||||
template<class Impl, class CloverHelpers>
|
||||
void WilsonCloverFermion<Impl, CloverHelpers>::M(const FermionField &in, FermionField &out)
|
||||
template <class Impl>
|
||||
void WilsonCloverFermion<Impl>::M(const FermionField &in, FermionField &out)
|
||||
{
|
||||
FermionField temp(out.Grid());
|
||||
|
||||
@ -89,8 +49,8 @@ void WilsonCloverFermion<Impl, CloverHelpers>::M(const FermionField &in, Fermion
|
||||
out += temp;
|
||||
}
|
||||
|
||||
template<class Impl, class CloverHelpers>
|
||||
void WilsonCloverFermion<Impl, CloverHelpers>::Mdag(const FermionField &in, FermionField &out)
|
||||
template <class Impl>
|
||||
void WilsonCloverFermion<Impl>::Mdag(const FermionField &in, FermionField &out)
|
||||
{
|
||||
FermionField temp(out.Grid());
|
||||
|
||||
@ -104,16 +64,13 @@ void WilsonCloverFermion<Impl, CloverHelpers>::Mdag(const FermionField &in, Ferm
|
||||
out += temp;
|
||||
}
|
||||
|
||||
template<class Impl, class CloverHelpers>
|
||||
void WilsonCloverFermion<Impl, CloverHelpers>::ImportGauge(const GaugeField &_Umu)
|
||||
template <class Impl>
|
||||
void WilsonCloverFermion<Impl>::ImportGauge(const GaugeField &_Umu)
|
||||
{
|
||||
double t0 = usecond();
|
||||
WilsonFermion<Impl>::ImportGauge(_Umu);
|
||||
double t1 = usecond();
|
||||
GridBase *grid = _Umu.Grid();
|
||||
typename Impl::GaugeLinkField Bx(grid), By(grid), Bz(grid), Ex(grid), Ey(grid), Ez(grid);
|
||||
|
||||
double t2 = usecond();
|
||||
// Compute the field strength terms mu>nu
|
||||
WilsonLoops<Impl>::FieldStrength(Bx, _Umu, Zdir, Ydir);
|
||||
WilsonLoops<Impl>::FieldStrength(By, _Umu, Zdir, Xdir);
|
||||
@ -122,20 +79,52 @@ void WilsonCloverFermion<Impl, CloverHelpers>::ImportGauge(const GaugeField &_Um
|
||||
WilsonLoops<Impl>::FieldStrength(Ey, _Umu, Tdir, Ydir);
|
||||
WilsonLoops<Impl>::FieldStrength(Ez, _Umu, Tdir, Zdir);
|
||||
|
||||
double t3 = usecond();
|
||||
// Compute the Clover Operator acting on Colour and Spin
|
||||
// multiply here by the clover coefficients for the anisotropy
|
||||
CloverTerm = Helpers::fillCloverYZ(Bx) * csw_r;
|
||||
CloverTerm += Helpers::fillCloverXZ(By) * csw_r;
|
||||
CloverTerm += Helpers::fillCloverXY(Bz) * csw_r;
|
||||
CloverTerm += Helpers::fillCloverXT(Ex) * csw_t;
|
||||
CloverTerm += Helpers::fillCloverYT(Ey) * csw_t;
|
||||
CloverTerm += Helpers::fillCloverZT(Ez) * csw_t;
|
||||
|
||||
double t4 = usecond();
|
||||
CloverHelpers::Instantiate(CloverTerm, CloverTermInv, csw_t, this->diag_mass);
|
||||
CloverTerm = fillCloverYZ(Bx) * csw_r;
|
||||
CloverTerm += fillCloverXZ(By) * csw_r;
|
||||
CloverTerm += fillCloverXY(Bz) * csw_r;
|
||||
CloverTerm += fillCloverXT(Ex) * csw_t;
|
||||
CloverTerm += fillCloverYT(Ey) * csw_t;
|
||||
CloverTerm += fillCloverZT(Ez) * csw_t;
|
||||
CloverTerm += diag_mass;
|
||||
|
||||
int lvol = _Umu.Grid()->lSites();
|
||||
int DimRep = Impl::Dimension;
|
||||
|
||||
{
|
||||
autoView(CTv,CloverTerm,CpuRead);
|
||||
autoView(CTIv,CloverTermInv,CpuWrite);
|
||||
thread_for(site, lvol, {
|
||||
Coordinate lcoor;
|
||||
grid->LocalIndexToLocalCoor(site, lcoor);
|
||||
Eigen::MatrixXcd EigenCloverOp = Eigen::MatrixXcd::Zero(Ns * DimRep, Ns * DimRep);
|
||||
Eigen::MatrixXcd EigenInvCloverOp = Eigen::MatrixXcd::Zero(Ns * DimRep, Ns * DimRep);
|
||||
typename SiteCloverType::scalar_object Qx = Zero(), Qxinv = Zero();
|
||||
peekLocalSite(Qx, CTv, lcoor);
|
||||
//if (csw!=0){
|
||||
for (int j = 0; j < Ns; j++)
|
||||
for (int k = 0; k < Ns; k++)
|
||||
for (int a = 0; a < DimRep; a++)
|
||||
for (int b = 0; b < DimRep; b++){
|
||||
auto zz = Qx()(j, k)(a, b);
|
||||
EigenCloverOp(a + j * DimRep, b + k * DimRep) = std::complex<double>(zz);
|
||||
}
|
||||
// if (site==0) std::cout << "site =" << site << "\n" << EigenCloverOp << std::endl;
|
||||
|
||||
EigenInvCloverOp = EigenCloverOp.inverse();
|
||||
//std::cout << EigenInvCloverOp << std::endl;
|
||||
for (int j = 0; j < Ns; j++)
|
||||
for (int k = 0; k < Ns; k++)
|
||||
for (int a = 0; a < DimRep; a++)
|
||||
for (int b = 0; b < DimRep; b++)
|
||||
Qxinv()(j, k)(a, b) = EigenInvCloverOp(a + j * DimRep, b + k * DimRep);
|
||||
// if (site==0) std::cout << "site =" << site << "\n" << EigenInvCloverOp << std::endl;
|
||||
// }
|
||||
pokeLocalSite(Qxinv, CTIv, lcoor);
|
||||
});
|
||||
}
|
||||
|
||||
double t5 = usecond();
|
||||
// Separate the even and odd parts
|
||||
pickCheckerboard(Even, CloverTermEven, CloverTerm);
|
||||
pickCheckerboard(Odd, CloverTermOdd, CloverTerm);
|
||||
@ -148,47 +137,37 @@ void WilsonCloverFermion<Impl, CloverHelpers>::ImportGauge(const GaugeField &_Um
|
||||
|
||||
pickCheckerboard(Even, CloverTermInvDagEven, adj(CloverTermInv));
|
||||
pickCheckerboard(Odd, CloverTermInvDagOdd, adj(CloverTermInv));
|
||||
double t6 = usecond();
|
||||
|
||||
std::cout << GridLogDebug << "WilsonCloverFermion::ImportGauge timings:" << std::endl;
|
||||
std::cout << GridLogDebug << "WilsonFermion::Importgauge = " << (t1 - t0) / 1e6 << std::endl;
|
||||
std::cout << GridLogDebug << "allocations = " << (t2 - t1) / 1e6 << std::endl;
|
||||
std::cout << GridLogDebug << "field strength = " << (t3 - t2) / 1e6 << std::endl;
|
||||
std::cout << GridLogDebug << "fill clover = " << (t4 - t3) / 1e6 << std::endl;
|
||||
std::cout << GridLogDebug << "instantiation = " << (t5 - t4) / 1e6 << std::endl;
|
||||
std::cout << GridLogDebug << "pick cbs = " << (t6 - t5) / 1e6 << std::endl;
|
||||
std::cout << GridLogDebug << "total = " << (t6 - t0) / 1e6 << std::endl;
|
||||
}
|
||||
|
||||
template<class Impl, class CloverHelpers>
|
||||
void WilsonCloverFermion<Impl, CloverHelpers>::Mooee(const FermionField &in, FermionField &out)
|
||||
template <class Impl>
|
||||
void WilsonCloverFermion<Impl>::Mooee(const FermionField &in, FermionField &out)
|
||||
{
|
||||
this->MooeeInternal(in, out, DaggerNo, InverseNo);
|
||||
}
|
||||
|
||||
template<class Impl, class CloverHelpers>
|
||||
void WilsonCloverFermion<Impl, CloverHelpers>::MooeeDag(const FermionField &in, FermionField &out)
|
||||
template <class Impl>
|
||||
void WilsonCloverFermion<Impl>::MooeeDag(const FermionField &in, FermionField &out)
|
||||
{
|
||||
this->MooeeInternal(in, out, DaggerYes, InverseNo);
|
||||
}
|
||||
|
||||
template<class Impl, class CloverHelpers>
|
||||
void WilsonCloverFermion<Impl, CloverHelpers>::MooeeInv(const FermionField &in, FermionField &out)
|
||||
template <class Impl>
|
||||
void WilsonCloverFermion<Impl>::MooeeInv(const FermionField &in, FermionField &out)
|
||||
{
|
||||
this->MooeeInternal(in, out, DaggerNo, InverseYes);
|
||||
}
|
||||
|
||||
template<class Impl, class CloverHelpers>
|
||||
void WilsonCloverFermion<Impl, CloverHelpers>::MooeeInvDag(const FermionField &in, FermionField &out)
|
||||
template <class Impl>
|
||||
void WilsonCloverFermion<Impl>::MooeeInvDag(const FermionField &in, FermionField &out)
|
||||
{
|
||||
this->MooeeInternal(in, out, DaggerYes, InverseYes);
|
||||
}
|
||||
|
||||
template<class Impl, class CloverHelpers>
|
||||
void WilsonCloverFermion<Impl, CloverHelpers>::MooeeInternal(const FermionField &in, FermionField &out, int dag, int inv)
|
||||
template <class Impl>
|
||||
void WilsonCloverFermion<Impl>::MooeeInternal(const FermionField &in, FermionField &out, int dag, int inv)
|
||||
{
|
||||
out.Checkerboard() = in.Checkerboard();
|
||||
CloverField *Clover;
|
||||
CloverFieldType *Clover;
|
||||
assert(in.Checkerboard() == Odd || in.Checkerboard() == Even);
|
||||
|
||||
if (dag)
|
||||
@ -203,12 +182,12 @@ void WilsonCloverFermion<Impl, CloverHelpers>::MooeeInternal(const FermionField
|
||||
{
|
||||
Clover = (inv) ? &CloverTermInvDagEven : &CloverTermDagEven;
|
||||
}
|
||||
Helpers::multCloverField(out, *Clover, in);
|
||||
out = *Clover * in;
|
||||
}
|
||||
else
|
||||
{
|
||||
Clover = (inv) ? &CloverTermInv : &CloverTerm;
|
||||
Helpers::multCloverField(out, *Clover, in); // don't bother with adj, hermitian anyway
|
||||
out = adj(*Clover) * in;
|
||||
}
|
||||
}
|
||||
else
|
||||
@ -226,109 +205,29 @@ void WilsonCloverFermion<Impl, CloverHelpers>::MooeeInternal(const FermionField
|
||||
// std::cout << "Calling clover term Even" << std::endl;
|
||||
Clover = (inv) ? &CloverTermInvEven : &CloverTermEven;
|
||||
}
|
||||
Helpers::multCloverField(out, *Clover, in);
|
||||
out = *Clover * in;
|
||||
// std::cout << GridLogMessage << "*Clover.Checkerboard() " << (*Clover).Checkerboard() << std::endl;
|
||||
}
|
||||
else
|
||||
{
|
||||
Clover = (inv) ? &CloverTermInv : &CloverTerm;
|
||||
Helpers::multCloverField(out, *Clover, in);
|
||||
out = *Clover * in;
|
||||
}
|
||||
}
|
||||
|
||||
} // MooeeInternal
|
||||
|
||||
// Derivative parts unpreconditioned pseudofermions
|
||||
template<class Impl, class CloverHelpers>
|
||||
void WilsonCloverFermion<Impl, CloverHelpers>::MDeriv(GaugeField &force, const FermionField &X, const FermionField &Y, int dag)
|
||||
{
|
||||
conformable(X.Grid(), Y.Grid());
|
||||
conformable(X.Grid(), force.Grid());
|
||||
GaugeLinkField force_mu(force.Grid()), lambda(force.Grid());
|
||||
GaugeField clover_force(force.Grid());
|
||||
PropagatorField Lambda(force.Grid());
|
||||
|
||||
// Guido: Here we are hitting some performance issues:
|
||||
// need to extract the components of the DoubledGaugeField
|
||||
// for each call
|
||||
// Possible solution
|
||||
// Create a vector object to store them? (cons: wasting space)
|
||||
std::vector<GaugeLinkField> U(Nd, this->Umu.Grid());
|
||||
|
||||
Impl::extractLinkField(U, this->Umu);
|
||||
|
||||
force = Zero();
|
||||
// Derivative of the Wilson hopping term
|
||||
this->DhopDeriv(force, X, Y, dag);
|
||||
|
||||
///////////////////////////////////////////////////////////
|
||||
// Clover term derivative
|
||||
///////////////////////////////////////////////////////////
|
||||
Impl::outerProductImpl(Lambda, X, Y);
|
||||
//std::cout << "Lambda:" << Lambda << std::endl;
|
||||
|
||||
Gamma::Algebra sigma[] = {
|
||||
Gamma::Algebra::SigmaXY,
|
||||
Gamma::Algebra::SigmaXZ,
|
||||
Gamma::Algebra::SigmaXT,
|
||||
Gamma::Algebra::MinusSigmaXY,
|
||||
Gamma::Algebra::SigmaYZ,
|
||||
Gamma::Algebra::SigmaYT,
|
||||
Gamma::Algebra::MinusSigmaXZ,
|
||||
Gamma::Algebra::MinusSigmaYZ,
|
||||
Gamma::Algebra::SigmaZT,
|
||||
Gamma::Algebra::MinusSigmaXT,
|
||||
Gamma::Algebra::MinusSigmaYT,
|
||||
Gamma::Algebra::MinusSigmaZT};
|
||||
|
||||
/*
|
||||
sigma_{\mu \nu}=
|
||||
| 0 sigma[0] sigma[1] sigma[2] |
|
||||
| sigma[3] 0 sigma[4] sigma[5] |
|
||||
| sigma[6] sigma[7] 0 sigma[8] |
|
||||
| sigma[9] sigma[10] sigma[11] 0 |
|
||||
*/
|
||||
|
||||
int count = 0;
|
||||
clover_force = Zero();
|
||||
for (int mu = 0; mu < 4; mu++)
|
||||
{
|
||||
force_mu = Zero();
|
||||
for (int nu = 0; nu < 4; nu++)
|
||||
{
|
||||
if (mu == nu)
|
||||
continue;
|
||||
|
||||
RealD factor;
|
||||
if (nu == 4 || mu == 4)
|
||||
{
|
||||
factor = 2.0 * csw_t;
|
||||
}
|
||||
else
|
||||
{
|
||||
factor = 2.0 * csw_r;
|
||||
}
|
||||
PropagatorField Slambda = Gamma(sigma[count]) * Lambda; // sigma checked
|
||||
Impl::TraceSpinImpl(lambda, Slambda); // traceSpin ok
|
||||
force_mu -= factor*CloverHelpers::Cmunu(U, lambda, mu, nu); // checked
|
||||
count++;
|
||||
}
|
||||
|
||||
pokeLorentz(clover_force, U[mu] * force_mu, mu);
|
||||
}
|
||||
//clover_force *= csw;
|
||||
force += clover_force;
|
||||
}
|
||||
|
||||
// Derivative parts
|
||||
template<class Impl, class CloverHelpers>
|
||||
void WilsonCloverFermion<Impl, CloverHelpers>::MooDeriv(GaugeField &mat, const FermionField &X, const FermionField &Y, int dag)
|
||||
template <class Impl>
|
||||
void WilsonCloverFermion<Impl>::MooDeriv(GaugeField &mat, const FermionField &X, const FermionField &Y, int dag)
|
||||
{
|
||||
assert(0);
|
||||
}
|
||||
|
||||
// Derivative parts
|
||||
template<class Impl, class CloverHelpers>
|
||||
void WilsonCloverFermion<Impl, CloverHelpers>::MeeDeriv(GaugeField &mat, const FermionField &U, const FermionField &V, int dag)
|
||||
template <class Impl>
|
||||
void WilsonCloverFermion<Impl>::MeeDeriv(GaugeField &mat, const FermionField &U, const FermionField &V, int dag)
|
||||
{
|
||||
assert(0); // not implemented yet
|
||||
}
|
||||
|
||||
@ -60,8 +60,7 @@ WilsonFermion5D<Impl>::WilsonFermion5D(GaugeField &_Umu,
|
||||
UmuOdd (_FourDimRedBlackGrid),
|
||||
Lebesgue(_FourDimGrid),
|
||||
LebesgueEvenOdd(_FourDimRedBlackGrid),
|
||||
_tmp(&FiveDimRedBlackGrid),
|
||||
Dirichlet(0)
|
||||
_tmp(&FiveDimRedBlackGrid)
|
||||
{
|
||||
// some assertions
|
||||
assert(FiveDimGrid._ndimension==5);
|
||||
@ -92,19 +91,6 @@ WilsonFermion5D<Impl>::WilsonFermion5D(GaugeField &_Umu,
|
||||
assert(FourDimRedBlackGrid._simd_layout[d] ==FourDimGrid._simd_layout[d]);
|
||||
}
|
||||
|
||||
if ( p.dirichlet.size() == Nd+1) {
|
||||
Coordinate block = p.dirichlet;
|
||||
if ( block[0] || block[1] || block[2] || block[3] || block[4] ){
|
||||
Dirichlet = 1;
|
||||
Block = block;
|
||||
}
|
||||
} else {
|
||||
Coordinate block(Nd+1,0);
|
||||
Block = block;
|
||||
}
|
||||
|
||||
ZeroCounters();
|
||||
|
||||
if (Impl::LsVectorised) {
|
||||
|
||||
int nsimd = Simd::Nsimd();
|
||||
@ -232,14 +218,6 @@ void WilsonFermion5D<Impl>::ImportGauge(const GaugeField &_Umu)
|
||||
{
|
||||
GaugeField HUmu(_Umu.Grid());
|
||||
HUmu = _Umu*(-0.5);
|
||||
if ( Dirichlet ) {
|
||||
std::cout << GridLogMessage << " Dirichlet BCs 5d " <<Block<<std::endl;
|
||||
Coordinate GaugeBlock(Nd);
|
||||
for(int d=0;d<Nd;d++) GaugeBlock[d] = Block[d+1];
|
||||
std::cout << GridLogMessage << " Dirichlet BCs 4d " <<GaugeBlock<<std::endl;
|
||||
DirichletFilter<GaugeField> Filter(GaugeBlock);
|
||||
Filter.applyFilter(HUmu);
|
||||
}
|
||||
Impl::DoubleStore(GaugeGrid(),Umu,HUmu);
|
||||
pickCheckerboard(Even,UmuEven,Umu);
|
||||
pickCheckerboard(Odd ,UmuOdd,Umu);
|
||||
|
||||
@ -4,13 +4,12 @@ Grid physics library, www.github.com/paboyle/Grid
|
||||
|
||||
Source file: ./lib/qcd/action/fermion/WilsonFermion.cc
|
||||
|
||||
Copyright (C) 2022
|
||||
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>
|
||||
Author: Fabian Joswig <fabian.joswig@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
|
||||
@ -600,47 +599,11 @@ void WilsonFermion<Impl>::ContractConservedCurrent(PropagatorField &q_in_1,
|
||||
Current curr_type,
|
||||
unsigned int mu)
|
||||
{
|
||||
if(curr_type != Current::Vector)
|
||||
{
|
||||
std::cout << GridLogError << "Only the conserved vector current is implemented so far." << std::endl;
|
||||
exit(1);
|
||||
}
|
||||
|
||||
Gamma g5(Gamma::Algebra::Gamma5);
|
||||
conformable(_grid, q_in_1.Grid());
|
||||
conformable(_grid, q_in_2.Grid());
|
||||
conformable(_grid, q_out.Grid());
|
||||
auto UGrid= this->GaugeGrid();
|
||||
|
||||
PropagatorField tmp_shifted(UGrid);
|
||||
PropagatorField g5Lg5(UGrid);
|
||||
PropagatorField R(UGrid);
|
||||
PropagatorField gmuR(UGrid);
|
||||
|
||||
Gamma::Algebra Gmu [] = {
|
||||
Gamma::Algebra::GammaX,
|
||||
Gamma::Algebra::GammaY,
|
||||
Gamma::Algebra::GammaZ,
|
||||
Gamma::Algebra::GammaT,
|
||||
};
|
||||
Gamma gmu=Gamma(Gmu[mu]);
|
||||
|
||||
g5Lg5=g5*q_in_1*g5;
|
||||
tmp_shifted=Cshift(q_in_2,mu,1);
|
||||
Impl::multLinkField(R,this->Umu,tmp_shifted,mu);
|
||||
gmuR=gmu*R;
|
||||
|
||||
q_out=adj(g5Lg5)*R;
|
||||
q_out-=adj(g5Lg5)*gmuR;
|
||||
|
||||
tmp_shifted=Cshift(q_in_1,mu,1);
|
||||
Impl::multLinkField(g5Lg5,this->Umu,tmp_shifted,mu);
|
||||
g5Lg5=g5*g5Lg5*g5;
|
||||
R=q_in_2;
|
||||
gmuR=gmu*R;
|
||||
|
||||
q_out-=adj(g5Lg5)*R;
|
||||
q_out-=adj(g5Lg5)*gmuR;
|
||||
assert(0);
|
||||
}
|
||||
|
||||
|
||||
@ -654,51 +617,9 @@ void WilsonFermion<Impl>::SeqConservedCurrent(PropagatorField &q_in,
|
||||
unsigned int tmax,
|
||||
ComplexField &lattice_cmplx)
|
||||
{
|
||||
if(curr_type != Current::Vector)
|
||||
{
|
||||
std::cout << GridLogError << "Only the conserved vector current is implemented so far." << std::endl;
|
||||
exit(1);
|
||||
}
|
||||
|
||||
int tshift = (mu == Nd-1) ? 1 : 0;
|
||||
unsigned int LLt = GridDefaultLatt()[Tp];
|
||||
conformable(_grid, q_in.Grid());
|
||||
conformable(_grid, q_out.Grid());
|
||||
auto UGrid= this->GaugeGrid();
|
||||
|
||||
PropagatorField tmp(UGrid);
|
||||
PropagatorField Utmp(UGrid);
|
||||
PropagatorField L(UGrid);
|
||||
PropagatorField zz (UGrid);
|
||||
zz=Zero();
|
||||
LatticeInteger lcoor(UGrid); LatticeCoordinate(lcoor,Nd-1);
|
||||
|
||||
Gamma::Algebra Gmu [] = {
|
||||
Gamma::Algebra::GammaX,
|
||||
Gamma::Algebra::GammaY,
|
||||
Gamma::Algebra::GammaZ,
|
||||
Gamma::Algebra::GammaT,
|
||||
};
|
||||
Gamma gmu=Gamma(Gmu[mu]);
|
||||
|
||||
tmp = Cshift(q_in,mu,1);
|
||||
Impl::multLinkField(Utmp,this->Umu,tmp,mu);
|
||||
tmp = ( Utmp*lattice_cmplx - gmu*Utmp*lattice_cmplx ); // Forward hop
|
||||
tmp = where((lcoor>=tmin),tmp,zz); // Mask the time
|
||||
q_out = where((lcoor<=tmax),tmp,zz); // Position of current complicated
|
||||
|
||||
tmp = q_in *lattice_cmplx;
|
||||
tmp = Cshift(tmp,mu,-1);
|
||||
Impl::multLinkField(Utmp,this->Umu,tmp,mu+Nd); // Adjoint link
|
||||
tmp = -( Utmp + gmu*Utmp );
|
||||
// Mask the time
|
||||
if (tmax == LLt - 1 && tshift == 1){ // quick fix to include timeslice 0 if tmax + tshift is over the last timeslice
|
||||
unsigned int t0 = 0;
|
||||
tmp = where(((lcoor==t0) || (lcoor>=tmin+tshift)),tmp,zz);
|
||||
} else {
|
||||
tmp = where((lcoor>=tmin+tshift),tmp,zz);
|
||||
}
|
||||
q_out+= where((lcoor<=tmax+tshift),tmp,zz); // Position of current complicated
|
||||
assert(0);
|
||||
}
|
||||
|
||||
NAMESPACE_END(Grid);
|
||||
|
||||
@ -73,17 +73,17 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
//template<> void
|
||||
//WilsonKernels<WilsonImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
template<> void
|
||||
WilsonKernels<WilsonImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
//template<> void
|
||||
//WilsonKernels<ZWilsonImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
template<> void
|
||||
WilsonKernels<ZWilsonImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
|
||||
#undef INTERIOR_AND_EXTERIOR
|
||||
@ -102,17 +102,17 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldVi
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
//template<> void
|
||||
//WilsonKernels<WilsonImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
template<> void
|
||||
WilsonKernels<WilsonImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
//template<> void
|
||||
//WilsonKernels<ZWilsonImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
template<> void
|
||||
WilsonKernels<ZWilsonImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
|
||||
#undef INTERIOR_AND_EXTERIOR
|
||||
@ -131,17 +131,17 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldVi
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
//template<> void
|
||||
//WilsonKernels<WilsonImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
template<> void
|
||||
WilsonKernels<WilsonImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
//template<> void
|
||||
//WilsonKernels<ZWilsonImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
template<> void
|
||||
WilsonKernels<ZWilsonImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
|
||||
|
||||
@ -165,17 +165,17 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldVi
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
//template<> void
|
||||
//WilsonKernels<WilsonImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
template<> void
|
||||
WilsonKernels<WilsonImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
//template<> void
|
||||
//WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
template<> void
|
||||
WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
|
||||
#undef INTERIOR_AND_EXTERIOR
|
||||
@ -194,17 +194,17 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFiel
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
//template<> void
|
||||
//WilsonKernels<WilsonImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
template<> void
|
||||
WilsonKernels<WilsonImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
//template<> void
|
||||
//WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
template<> void
|
||||
WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
|
||||
#undef INTERIOR_AND_EXTERIOR
|
||||
@ -223,17 +223,17 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFiel
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
//template<> void
|
||||
//WilsonKernels<WilsonImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
template<> void
|
||||
WilsonKernels<WilsonImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
//template<> void
|
||||
//WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
template<> void
|
||||
WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
|
||||
|
||||
@ -280,17 +280,17 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
// template<> void
|
||||
// WilsonKernels<WilsonImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
template<> void
|
||||
WilsonKernels<WilsonImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
// template<> void
|
||||
// WilsonKernels<ZWilsonImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
template<> void
|
||||
WilsonKernels<ZWilsonImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
|
||||
#undef INTERIOR_AND_EXTERIOR
|
||||
@ -309,17 +309,17 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldVi
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
// template<> void
|
||||
// WilsonKernels<WilsonImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
template<> void
|
||||
WilsonKernels<WilsonImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
// template<> void
|
||||
// WilsonKernels<ZWilsonImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
template<> void
|
||||
WilsonKernels<ZWilsonImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
|
||||
#undef INTERIOR_AND_EXTERIOR
|
||||
@ -338,17 +338,17 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldVi
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
// template<> void
|
||||
// WilsonKernels<WilsonImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
template<> void
|
||||
WilsonKernels<WilsonImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
// template<> void
|
||||
// WilsonKernels<ZWilsonImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
template<> void
|
||||
WilsonKernels<ZWilsonImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
|
||||
/////////////////////////////////////////////////////////////////
|
||||
@ -371,17 +371,17 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldVi
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
// template<> void
|
||||
// WilsonKernels<WilsonImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
template<> void
|
||||
WilsonKernels<WilsonImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
// template<> void
|
||||
// WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
template<> void
|
||||
WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
|
||||
#undef INTERIOR_AND_EXTERIOR
|
||||
@ -400,17 +400,17 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFiel
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
// template<> void
|
||||
// WilsonKernels<WilsonImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
template<> void
|
||||
WilsonKernels<WilsonImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
// template<> void
|
||||
// WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
template<> void
|
||||
WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
|
||||
#undef INTERIOR_AND_EXTERIOR
|
||||
@ -429,17 +429,17 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFiel
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
// template<> void
|
||||
// WilsonKernels<WilsonImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
template<> void
|
||||
WilsonKernels<WilsonImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
// template<> void
|
||||
// WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
|
||||
template<> void
|
||||
WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
|
||||
|
||||
|
||||
|
||||
|
||||
@ -74,15 +74,15 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
//template<> void
|
||||
//WilsonKernels<WilsonImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
//
|
||||
//template<> void
|
||||
//WilsonKernels<ZWilsonImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<WilsonImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
template<> void
|
||||
WilsonKernels<ZWilsonImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
#undef INTERIOR_AND_EXTERIOR
|
||||
#define INTERIOR
|
||||
@ -97,15 +97,15 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldVi
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
//template<> void
|
||||
//WilsonKernels<WilsonImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
//
|
||||
//template<> void
|
||||
//WilsonKernels<ZWilsonImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<WilsonImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
template<> void
|
||||
WilsonKernels<ZWilsonImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
|
||||
#undef INTERIOR_AND_EXTERIOR
|
||||
@ -121,15 +121,15 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldVi
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
//template<> void
|
||||
//WilsonKernels<WilsonImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
//
|
||||
//template<> void
|
||||
//WilsonKernels<ZWilsonImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<WilsonImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
template<> void
|
||||
WilsonKernels<ZWilsonImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
/////////////////////////////////////////////////////////////////
|
||||
// XYZT vectorised, dag Kernel, single
|
||||
@ -148,15 +148,15 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldVi
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
//template<> void
|
||||
//WilsonKernels<WilsonImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
//
|
||||
//template<> void
|
||||
//WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<WilsonImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
template<> void
|
||||
WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
#undef INTERIOR_AND_EXTERIOR
|
||||
#define INTERIOR
|
||||
@ -171,15 +171,15 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFiel
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
//template<> void
|
||||
//WilsonKernels<WilsonImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
//
|
||||
//template<> void
|
||||
//WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<WilsonImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
template<> void
|
||||
WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
#undef INTERIOR_AND_EXTERIOR
|
||||
#undef INTERIOR
|
||||
@ -194,15 +194,15 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFiel
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
//template<> void
|
||||
//WilsonKernels<WilsonImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
//
|
||||
//template<> void
|
||||
//WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<WilsonImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
template<> void
|
||||
WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
#undef MAYBEPERM
|
||||
#undef MULT_2SPIN
|
||||
@ -228,14 +228,14 @@ WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSite(StencilView &st, DoubledGaugeF
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
//template<> void
|
||||
//WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
//template<> void
|
||||
//WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
#undef INTERIOR_AND_EXTERIOR
|
||||
#define INTERIOR
|
||||
@ -249,14 +249,14 @@ WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSiteInt(StencilView &st, DoubledGau
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
//template<> void
|
||||
//WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
//template<> void
|
||||
//WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
#undef INTERIOR_AND_EXTERIOR
|
||||
#undef INTERIOR
|
||||
@ -273,15 +273,15 @@ WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSiteExt(StencilView &st, DoubledGau
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
//template<> void
|
||||
//WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
//
|
||||
//template<> void
|
||||
//WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
template<> void
|
||||
WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
/////////////////////////////////////////////////////////////////
|
||||
// Ls vectorised, dag Kernel, single
|
||||
@ -299,14 +299,14 @@ WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSiteDag(StencilView &st, DoubledGau
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
//template<> void
|
||||
//WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
//template<> void
|
||||
//WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
#undef INTERIOR_AND_EXTERIOR
|
||||
#define INTERIOR
|
||||
@ -320,14 +320,14 @@ WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSiteDagInt(StencilView &st, Doubled
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
//template<> void
|
||||
//WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
//template<> void
|
||||
//WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
#undef INTERIOR_AND_EXTERIOR
|
||||
#undef INTERIOR
|
||||
@ -341,14 +341,14 @@ WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSiteDagExt(StencilView &st, Doubled
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
//template<> void
|
||||
//WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
//template<> void
|
||||
//WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
#endif // VEC 5D
|
||||
|
||||
@ -392,14 +392,14 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
//template<> void
|
||||
//WilsonKernels<WilsonImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
//template<> void
|
||||
//WilsonKernels<ZWilsonImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<WilsonImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<ZWilsonImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
#undef INTERIOR_AND_EXTERIOR
|
||||
#define INTERIOR
|
||||
@ -413,14 +413,14 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldVi
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
//template<> void
|
||||
//WilsonKernels<WilsonImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
//template<> void
|
||||
//WilsonKernels<ZWilsonImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<WilsonImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<ZWilsonImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
#undef INTERIOR_AND_EXTERIOR
|
||||
#undef INTERIOR
|
||||
@ -434,14 +434,14 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldVi
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
//template<> void
|
||||
//WilsonKernels<WilsonImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
//template<> void
|
||||
//WilsonKernels<ZWilsonImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<WilsonImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<ZWilsonImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
/////////////////////////////////////////////////////////////////
|
||||
// XYZT vectorised, dag Kernel, single
|
||||
@ -459,14 +459,14 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldVi
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
//template<> void
|
||||
//WilsonKernels<WilsonImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
//template<> void
|
||||
//WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<WilsonImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
#undef INTERIOR_AND_EXTERIOR
|
||||
#define INTERIOR
|
||||
@ -480,14 +480,14 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFiel
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
//template<> void
|
||||
//WilsonKernels<WilsonImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
//template<> void
|
||||
//WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<WilsonImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
#undef INTERIOR_AND_EXTERIOR
|
||||
#undef INTERIOR
|
||||
@ -501,14 +501,14 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFiel
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
//template<> void
|
||||
//WilsonKernels<WilsonImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
//template<> void
|
||||
//WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<WilsonImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
#undef MAYBEPERM
|
||||
#undef MULT_2SPIN
|
||||
@ -533,14 +533,14 @@ WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSite(StencilView &st, DoubledGaugeF
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
//template<> void
|
||||
//WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
//template<> void
|
||||
//WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
#undef INTERIOR_AND_EXTERIOR
|
||||
#define INTERIOR
|
||||
@ -554,14 +554,14 @@ WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSiteInt(StencilView &st, DoubledGau
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
//template<> void
|
||||
//WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
//template<> void
|
||||
//WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
#undef INTERIOR_AND_EXTERIOR
|
||||
#undef INTERIOR
|
||||
@ -577,14 +577,14 @@ WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSiteExt(StencilView &st, DoubledGau
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
//template<> void
|
||||
//WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
//template<> void
|
||||
//WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
/////////////////////////////////////////////////////////////////
|
||||
// Ls vectorised, dag Kernel, single
|
||||
@ -602,14 +602,14 @@ WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSiteDag(StencilView &st, DoubledGau
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
//template<> void
|
||||
//WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
//template<> void
|
||||
//WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
#undef INTERIOR_AND_EXTERIOR
|
||||
#define INTERIOR
|
||||
@ -623,14 +623,14 @@ WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSiteDagInt(StencilView &st, Doubled
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
//template<> void
|
||||
//WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
//template<> void
|
||||
//WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
#undef INTERIOR_AND_EXTERIOR
|
||||
#undef INTERIOR
|
||||
@ -645,14 +645,14 @@ WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSiteDagExt(StencilView &st, Doubled
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
//template<> void
|
||||
//WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
//template<> void
|
||||
//WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
|
||||
// int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
template<> void
|
||||
WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
|
||||
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
|
||||
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
|
||||
|
||||
#endif // VEC 5D
|
||||
|
||||
|
||||
@ -77,23 +77,23 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
|
||||
#define REGISTER
|
||||
|
||||
#ifdef GRID_SIMT
|
||||
#define LOAD_CHIMU(Ptype) \
|
||||
#define LOAD_CHIMU(ptype) \
|
||||
{const SiteSpinor & ref (in[offset]); \
|
||||
Chimu_00=coalescedReadPermute<Ptype>(ref()(0)(0),perm,lane); \
|
||||
Chimu_01=coalescedReadPermute<Ptype>(ref()(0)(1),perm,lane); \
|
||||
Chimu_02=coalescedReadPermute<Ptype>(ref()(0)(2),perm,lane); \
|
||||
Chimu_10=coalescedReadPermute<Ptype>(ref()(1)(0),perm,lane); \
|
||||
Chimu_11=coalescedReadPermute<Ptype>(ref()(1)(1),perm,lane); \
|
||||
Chimu_12=coalescedReadPermute<Ptype>(ref()(1)(2),perm,lane); \
|
||||
Chimu_20=coalescedReadPermute<Ptype>(ref()(2)(0),perm,lane); \
|
||||
Chimu_21=coalescedReadPermute<Ptype>(ref()(2)(1),perm,lane); \
|
||||
Chimu_22=coalescedReadPermute<Ptype>(ref()(2)(2),perm,lane); \
|
||||
Chimu_30=coalescedReadPermute<Ptype>(ref()(3)(0),perm,lane); \
|
||||
Chimu_31=coalescedReadPermute<Ptype>(ref()(3)(1),perm,lane); \
|
||||
Chimu_32=coalescedReadPermute<Ptype>(ref()(3)(2),perm,lane); }
|
||||
Chimu_00=coalescedReadPermute<ptype>(ref()(0)(0),perm,lane); \
|
||||
Chimu_01=coalescedReadPermute<ptype>(ref()(0)(1),perm,lane); \
|
||||
Chimu_02=coalescedReadPermute<ptype>(ref()(0)(2),perm,lane); \
|
||||
Chimu_10=coalescedReadPermute<ptype>(ref()(1)(0),perm,lane); \
|
||||
Chimu_11=coalescedReadPermute<ptype>(ref()(1)(1),perm,lane); \
|
||||
Chimu_12=coalescedReadPermute<ptype>(ref()(1)(2),perm,lane); \
|
||||
Chimu_20=coalescedReadPermute<ptype>(ref()(2)(0),perm,lane); \
|
||||
Chimu_21=coalescedReadPermute<ptype>(ref()(2)(1),perm,lane); \
|
||||
Chimu_22=coalescedReadPermute<ptype>(ref()(2)(2),perm,lane); \
|
||||
Chimu_30=coalescedReadPermute<ptype>(ref()(3)(0),perm,lane); \
|
||||
Chimu_31=coalescedReadPermute<ptype>(ref()(3)(1),perm,lane); \
|
||||
Chimu_32=coalescedReadPermute<ptype>(ref()(3)(2),perm,lane); }
|
||||
#define PERMUTE_DIR(dir) ;
|
||||
#else
|
||||
#define LOAD_CHIMU(Ptype) \
|
||||
#define LOAD_CHIMU(ptype) \
|
||||
{const SiteSpinor & ref (in[offset]); \
|
||||
Chimu_00=ref()(0)(0);\
|
||||
Chimu_01=ref()(0)(1);\
|
||||
@ -109,12 +109,12 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
|
||||
Chimu_32=ref()(3)(2);}
|
||||
|
||||
#define PERMUTE_DIR(dir) \
|
||||
permute##dir(Chi_00,Chi_00); \
|
||||
permute##dir(Chi_01,Chi_01); \
|
||||
permute##dir(Chi_02,Chi_02); \
|
||||
permute##dir(Chi_10,Chi_10); \
|
||||
permute##dir(Chi_11,Chi_11); \
|
||||
permute##dir(Chi_12,Chi_12);
|
||||
permute##dir(Chi_00,Chi_00); \
|
||||
permute##dir(Chi_01,Chi_01);\
|
||||
permute##dir(Chi_02,Chi_02);\
|
||||
permute##dir(Chi_10,Chi_10); \
|
||||
permute##dir(Chi_11,Chi_11);\
|
||||
permute##dir(Chi_12,Chi_12);
|
||||
|
||||
#endif
|
||||
|
||||
@ -371,91 +371,88 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
|
||||
result_32-= UChi_12;
|
||||
|
||||
#define HAND_STENCIL_LEGB(PROJ,PERM,DIR,RECON) \
|
||||
{int ptype; \
|
||||
SE=st.GetEntry(ptype,DIR,ss); \
|
||||
auto offset = SE->_offset; \
|
||||
auto local = SE->_is_local; \
|
||||
auto perm = SE->_permute; \
|
||||
if ( local ) { \
|
||||
LOAD_CHIMU(PERM); \
|
||||
PROJ; \
|
||||
if ( perm) { \
|
||||
PERMUTE_DIR(PERM); \
|
||||
} \
|
||||
} else { \
|
||||
LOAD_CHI; \
|
||||
} \
|
||||
acceleratorSynchronise(); \
|
||||
MULT_2SPIN(DIR); \
|
||||
RECON; }
|
||||
SE=st.GetEntry(ptype,DIR,ss); \
|
||||
offset = SE->_offset; \
|
||||
local = SE->_is_local; \
|
||||
perm = SE->_permute; \
|
||||
if ( local ) { \
|
||||
LOAD_CHIMU(PERM); \
|
||||
PROJ; \
|
||||
if ( perm) { \
|
||||
PERMUTE_DIR(PERM); \
|
||||
} \
|
||||
} else { \
|
||||
LOAD_CHI; \
|
||||
} \
|
||||
acceleratorSynchronise(); \
|
||||
MULT_2SPIN(DIR); \
|
||||
RECON;
|
||||
|
||||
#define HAND_STENCIL_LEG(PROJ,PERM,DIR,RECON) \
|
||||
{ SE=&st_p[DIR+8*ss]; \
|
||||
auto ptype=st_perm[DIR]; \
|
||||
auto offset = SE->_offset; \
|
||||
auto local = SE->_is_local; \
|
||||
auto perm = SE->_permute; \
|
||||
if ( local ) { \
|
||||
LOAD_CHIMU(PERM); \
|
||||
PROJ; \
|
||||
if ( perm) { \
|
||||
PERMUTE_DIR(PERM); \
|
||||
} \
|
||||
} else { \
|
||||
LOAD_CHI; \
|
||||
} \
|
||||
acceleratorSynchronise(); \
|
||||
MULT_2SPIN(DIR); \
|
||||
RECON; }
|
||||
#define HAND_STENCIL_LEG(PROJ,PERM,DIR,RECON) \
|
||||
SE=&st_p[DIR+8*ss]; \
|
||||
ptype=st_perm[DIR]; \
|
||||
offset = SE->_offset; \
|
||||
local = SE->_is_local; \
|
||||
perm = SE->_permute; \
|
||||
if ( local ) { \
|
||||
LOAD_CHIMU(PERM); \
|
||||
PROJ; \
|
||||
if ( perm) { \
|
||||
PERMUTE_DIR(PERM); \
|
||||
} \
|
||||
} else { \
|
||||
LOAD_CHI; \
|
||||
} \
|
||||
acceleratorSynchronise(); \
|
||||
MULT_2SPIN(DIR); \
|
||||
RECON;
|
||||
|
||||
#define HAND_STENCIL_LEGA(PROJ,PERM,DIR,RECON) \
|
||||
{ SE=&st_p[DIR+8*ss]; \
|
||||
auto ptype=st_perm[DIR]; \
|
||||
/*SE=st.GetEntry(ptype,DIR,ss);*/ \
|
||||
auto offset = SE->_offset; \
|
||||
auto perm = SE->_permute; \
|
||||
LOAD_CHIMU(PERM); \
|
||||
PROJ; \
|
||||
MULT_2SPIN(DIR); \
|
||||
RECON; }
|
||||
SE=&st_p[DIR+8*ss]; \
|
||||
ptype=st_perm[DIR]; \
|
||||
/*SE=st.GetEntry(ptype,DIR,ss);*/ \
|
||||
offset = SE->_offset; \
|
||||
perm = SE->_permute; \
|
||||
LOAD_CHIMU(PERM); \
|
||||
PROJ; \
|
||||
MULT_2SPIN(DIR); \
|
||||
RECON;
|
||||
|
||||
#define HAND_STENCIL_LEG_INT(PROJ,PERM,DIR,RECON) \
|
||||
{ int ptype; \
|
||||
SE=st.GetEntry(ptype,DIR,ss); \
|
||||
auto offset = SE->_offset; \
|
||||
auto local = SE->_is_local; \
|
||||
auto perm = SE->_permute; \
|
||||
if ( local ) { \
|
||||
LOAD_CHIMU(PERM); \
|
||||
PROJ; \
|
||||
if ( perm) { \
|
||||
PERMUTE_DIR(PERM); \
|
||||
} \
|
||||
} else if ( st.same_node[DIR] ) { \
|
||||
LOAD_CHI; \
|
||||
} \
|
||||
acceleratorSynchronise(); \
|
||||
if (local || st.same_node[DIR] ) { \
|
||||
MULT_2SPIN(DIR); \
|
||||
RECON; \
|
||||
} \
|
||||
acceleratorSynchronise(); }
|
||||
SE=st.GetEntry(ptype,DIR,ss); \
|
||||
offset = SE->_offset; \
|
||||
local = SE->_is_local; \
|
||||
perm = SE->_permute; \
|
||||
if ( local ) { \
|
||||
LOAD_CHIMU(PERM); \
|
||||
PROJ; \
|
||||
if ( perm) { \
|
||||
PERMUTE_DIR(PERM); \
|
||||
} \
|
||||
} else if ( st.same_node[DIR] ) { \
|
||||
LOAD_CHI; \
|
||||
} \
|
||||
acceleratorSynchronise(); \
|
||||
if (local || st.same_node[DIR] ) { \
|
||||
MULT_2SPIN(DIR); \
|
||||
RECON; \
|
||||
} \
|
||||
acceleratorSynchronise();
|
||||
|
||||
#define HAND_STENCIL_LEG_EXT(PROJ,PERM,DIR,RECON) \
|
||||
{ int ptype; \
|
||||
SE=st.GetEntry(ptype,DIR,ss); \
|
||||
auto offset = SE->_offset; \
|
||||
if((!SE->_is_local)&&(!st.same_node[DIR]) ) { \
|
||||
LOAD_CHI; \
|
||||
MULT_2SPIN(DIR); \
|
||||
RECON; \
|
||||
nmu++; \
|
||||
} \
|
||||
acceleratorSynchronise(); }
|
||||
SE=st.GetEntry(ptype,DIR,ss); \
|
||||
offset = SE->_offset; \
|
||||
if((!SE->_is_local)&&(!st.same_node[DIR]) ) { \
|
||||
LOAD_CHI; \
|
||||
MULT_2SPIN(DIR); \
|
||||
RECON; \
|
||||
nmu++; \
|
||||
} \
|
||||
acceleratorSynchronise();
|
||||
|
||||
#define HAND_RESULT(ss) \
|
||||
{ \
|
||||
SiteSpinor & ref (out[ss]); \
|
||||
#define HAND_RESULT(ss) \
|
||||
{ \
|
||||
SiteSpinor & ref (out[ss]); \
|
||||
coalescedWrite(ref()(0)(0),result_00,lane); \
|
||||
coalescedWrite(ref()(0)(1),result_01,lane); \
|
||||
coalescedWrite(ref()(0)(2),result_02,lane); \
|
||||
@ -566,6 +563,7 @@ WilsonKernels<Impl>::HandDhopSiteSycl(StencilVector st_perm,StencilEntry *st_p,
|
||||
|
||||
HAND_DECLARATIONS(Simt);
|
||||
|
||||
int offset,local,perm, ptype;
|
||||
StencilEntry *SE;
|
||||
HAND_STENCIL_LEG(XM_PROJ,3,Xp,XM_RECON);
|
||||
HAND_STENCIL_LEG(YM_PROJ,2,Yp,YM_RECON_ACCUM);
|
||||
@ -595,7 +593,9 @@ WilsonKernels<Impl>::HandDhopSite(StencilView &st, DoubledGaugeFieldView &U,Site
|
||||
|
||||
HAND_DECLARATIONS(Simt);
|
||||
|
||||
int offset,local,perm, ptype;
|
||||
StencilEntry *SE;
|
||||
|
||||
HAND_STENCIL_LEG(XM_PROJ,3,Xp,XM_RECON);
|
||||
HAND_STENCIL_LEG(YM_PROJ,2,Yp,YM_RECON_ACCUM);
|
||||
HAND_STENCIL_LEG(ZM_PROJ,1,Zp,ZM_RECON_ACCUM);
|
||||
@ -623,6 +623,8 @@ void WilsonKernels<Impl>::HandDhopSiteDag(StencilView &st,DoubledGaugeFieldView
|
||||
HAND_DECLARATIONS(Simt);
|
||||
|
||||
StencilEntry *SE;
|
||||
int offset,local,perm, ptype;
|
||||
|
||||
HAND_STENCIL_LEG(XP_PROJ,3,Xp,XP_RECON);
|
||||
HAND_STENCIL_LEG(YP_PROJ,2,Yp,YP_RECON_ACCUM);
|
||||
HAND_STENCIL_LEG(ZP_PROJ,1,Zp,ZP_RECON_ACCUM);
|
||||
@ -638,8 +640,8 @@ template<class Impl> accelerator_inline void
|
||||
WilsonKernels<Impl>::HandDhopSiteInt(StencilView &st,DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
|
||||
int ss,int sU,const FermionFieldView &in, FermionFieldView &out)
|
||||
{
|
||||
// auto st_p = st._entries_p;
|
||||
// auto st_perm = st._permute_type;
|
||||
auto st_p = st._entries_p;
|
||||
auto st_perm = st._permute_type;
|
||||
// T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
|
||||
typedef typename Simd::scalar_type S;
|
||||
typedef typename Simd::vector_type V;
|
||||
@ -650,6 +652,7 @@ WilsonKernels<Impl>::HandDhopSiteInt(StencilView &st,DoubledGaugeFieldView &U,Si
|
||||
|
||||
HAND_DECLARATIONS(Simt);
|
||||
|
||||
int offset,local,perm, ptype;
|
||||
StencilEntry *SE;
|
||||
ZERO_RESULT;
|
||||
HAND_STENCIL_LEG_INT(XM_PROJ,3,Xp,XM_RECON_ACCUM);
|
||||
@ -667,8 +670,8 @@ template<class Impl> accelerator_inline
|
||||
void WilsonKernels<Impl>::HandDhopSiteDagInt(StencilView &st,DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
|
||||
int ss,int sU,const FermionFieldView &in, FermionFieldView &out)
|
||||
{
|
||||
// auto st_p = st._entries_p;
|
||||
// auto st_perm = st._permute_type;
|
||||
auto st_p = st._entries_p;
|
||||
auto st_perm = st._permute_type;
|
||||
typedef typename Simd::scalar_type S;
|
||||
typedef typename Simd::vector_type V;
|
||||
typedef decltype( coalescedRead( in[0]()(0)(0) )) Simt;
|
||||
@ -679,6 +682,7 @@ void WilsonKernels<Impl>::HandDhopSiteDagInt(StencilView &st,DoubledGaugeFieldVi
|
||||
HAND_DECLARATIONS(Simt);
|
||||
|
||||
StencilEntry *SE;
|
||||
int offset,local,perm, ptype;
|
||||
ZERO_RESULT;
|
||||
HAND_STENCIL_LEG_INT(XP_PROJ,3,Xp,XP_RECON_ACCUM);
|
||||
HAND_STENCIL_LEG_INT(YP_PROJ,2,Yp,YP_RECON_ACCUM);
|
||||
@ -695,8 +699,8 @@ template<class Impl> accelerator_inline void
|
||||
WilsonKernels<Impl>::HandDhopSiteExt(StencilView &st,DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
|
||||
int ss,int sU,const FermionFieldView &in, FermionFieldView &out)
|
||||
{
|
||||
// auto st_p = st._entries_p;
|
||||
// auto st_perm = st._permute_type;
|
||||
auto st_p = st._entries_p;
|
||||
auto st_perm = st._permute_type;
|
||||
// T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
|
||||
typedef typename Simd::scalar_type S;
|
||||
typedef typename Simd::vector_type V;
|
||||
@ -707,7 +711,7 @@ WilsonKernels<Impl>::HandDhopSiteExt(StencilView &st,DoubledGaugeFieldView &U,Si
|
||||
|
||||
HAND_DECLARATIONS(Simt);
|
||||
|
||||
// int offset, ptype;
|
||||
int offset, ptype;
|
||||
StencilEntry *SE;
|
||||
int nmu=0;
|
||||
ZERO_RESULT;
|
||||
@ -726,8 +730,8 @@ template<class Impl> accelerator_inline
|
||||
void WilsonKernels<Impl>::HandDhopSiteDagExt(StencilView &st,DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
|
||||
int ss,int sU,const FermionFieldView &in, FermionFieldView &out)
|
||||
{
|
||||
// auto st_p = st._entries_p;
|
||||
// auto st_perm = st._permute_type;
|
||||
auto st_p = st._entries_p;
|
||||
auto st_perm = st._permute_type;
|
||||
typedef typename Simd::scalar_type S;
|
||||
typedef typename Simd::vector_type V;
|
||||
typedef decltype( coalescedRead( in[0]()(0)(0) )) Simt;
|
||||
@ -738,7 +742,7 @@ void WilsonKernels<Impl>::HandDhopSiteDagExt(StencilView &st,DoubledGaugeFieldVi
|
||||
HAND_DECLARATIONS(Simt);
|
||||
|
||||
StencilEntry *SE;
|
||||
// int offset, ptype;
|
||||
int offset, ptype;
|
||||
int nmu=0;
|
||||
ZERO_RESULT;
|
||||
HAND_STENCIL_LEG_EXT(XP_PROJ,3,Xp,XP_RECON_ACCUM);
|
||||
|
||||
@ -440,17 +440,6 @@ void WilsonKernels<Impl>::DhopDirKernel( StencilImpl &st, DoubledGaugeField &U,S
|
||||
|
||||
#define KERNEL_CALL(A) KERNEL_CALLNB(A); accelerator_barrier();
|
||||
|
||||
#define KERNEL_CALL_EXT(A) \
|
||||
const uint64_t NN = Nsite*Ls; \
|
||||
const uint64_t sz = st.surface_list.size(); \
|
||||
auto ptr = &st.surface_list[0]; \
|
||||
accelerator_forNB( ss, sz, Simd::Nsimd(), { \
|
||||
int sF = ptr[ss]; \
|
||||
int sU = ss/Ls; \
|
||||
WilsonKernels<Impl>::A(st_v,U_v,buf,sF,sU,in_v,out_v); \
|
||||
}); \
|
||||
accelerator_barrier();
|
||||
|
||||
#define ASM_CALL(A) \
|
||||
thread_for( ss, Nsite, { \
|
||||
int sU = ss; \
|
||||
|
||||
@ -1,44 +0,0 @@
|
||||
/*************************************************************************************
|
||||
|
||||
Grid physics library, www.github.com/paboyle/Grid
|
||||
|
||||
Source file: ./lib/ qcd/action/fermion/instantiation/CompactWilsonCloverFermionInstantiation.cc.master
|
||||
|
||||
Copyright (C) 2017 - 2022
|
||||
|
||||
Author: paboyle <paboyle@ph.ed.ac.uk>
|
||||
Author: Guido Cossu <guido.cossu@ed.ac.uk>
|
||||
Author: Daniel Richtmann <daniel.richtmann@gmail.com>
|
||||
Author: Mattia Bruno <mattia.bruno@cern.ch>
|
||||
|
||||
This program is free software; you can redistribute it and/or modify
|
||||
it under the terms of the GNU General Public License as published by
|
||||
the Free Software Foundation; either version 2 of the License, or
|
||||
(at your option) any later version.
|
||||
|
||||
This program is distributed in the hope that it will be useful,
|
||||
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||||
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
||||
GNU General Public License for more details.
|
||||
|
||||
You should have received a copy of the GNU General Public License along
|
||||
with this program; if not, write to the Free Software Foundation, Inc.,
|
||||
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
|
||||
|
||||
See the full license in the file "LICENSE" in the top level distribution directory
|
||||
*************************************************************************************/
|
||||
/* END LEGAL */
|
||||
|
||||
#include <Grid/Grid.h>
|
||||
#include <Grid/qcd/spin/Dirac.h>
|
||||
#include <Grid/qcd/action/fermion/CompactWilsonCloverFermion.h>
|
||||
#include <Grid/qcd/action/fermion/implementation/CompactWilsonCloverFermionImplementation.h>
|
||||
#include <Grid/qcd/action/fermion/CloverHelpers.h>
|
||||
|
||||
NAMESPACE_BEGIN(Grid);
|
||||
|
||||
#include "impl.h"
|
||||
template class CompactWilsonCloverFermion<IMPLEMENTATION, CompactCloverHelpers<IMPLEMENTATION>>;
|
||||
template class CompactWilsonCloverFermion<IMPLEMENTATION, CompactExpCloverHelpers<IMPLEMENTATION>>;
|
||||
|
||||
NAMESPACE_END(Grid);
|
||||
@ -0,0 +1 @@
|
||||
../CayleyFermion5DInstantiation.cc.master
|
||||
@ -0,0 +1 @@
|
||||
../ContinuedFractionFermion5DInstantiation.cc.master
|
||||
@ -0,0 +1 @@
|
||||
../DomainWallEOFAFermionInstantiation.cc.master
|
||||
@ -0,0 +1 @@
|
||||
../MobiusEOFAFermionInstantiation.cc.master
|
||||
@ -0,0 +1 @@
|
||||
../PartialFractionFermion5DInstantiation.cc.master
|
||||
@ -0,0 +1 @@
|
||||
../WilsonCloverFermionInstantiation.cc.master
|
||||
@ -0,0 +1 @@
|
||||
../WilsonFermion5DInstantiation.cc.master
|
||||
@ -0,0 +1 @@
|
||||
../WilsonFermionInstantiation.cc.master
|
||||
@ -0,0 +1 @@
|
||||
../WilsonKernelsInstantiationGparity.cc.master
|
||||
@ -0,0 +1 @@
|
||||
../WilsonTMFermionInstantiation.cc.master
|
||||
@ -0,0 +1 @@
|
||||
#define IMPLEMENTATION GparityWilsonImplDF
|
||||
@ -0,0 +1 @@
|
||||
../CayleyFermion5DInstantiation.cc.master
|
||||
@ -0,0 +1 @@
|
||||
../ContinuedFractionFermion5DInstantiation.cc.master
|
||||
@ -0,0 +1 @@
|
||||
../DomainWallEOFAFermionInstantiation.cc.master
|
||||
@ -0,0 +1 @@
|
||||
../MobiusEOFAFermionInstantiation.cc.master
|
||||
@ -0,0 +1 @@
|
||||
../PartialFractionFermion5DInstantiation.cc.master
|
||||
@ -0,0 +1 @@
|
||||
../WilsonCloverFermionInstantiation.cc.master
|
||||
@ -0,0 +1 @@
|
||||
../WilsonFermion5DInstantiation.cc.master
|
||||
@ -0,0 +1 @@
|
||||
../WilsonFermionInstantiation.cc.master
|
||||
@ -0,0 +1 @@
|
||||
../WilsonKernelsInstantiationGparity.cc.master
|
||||
@ -0,0 +1 @@
|
||||
../WilsonTMFermionInstantiation.cc.master
|
||||
@ -0,0 +1 @@
|
||||
#define IMPLEMENTATION GparityWilsonImplFH
|
||||
Some files were not shown because too many files have changed in this diff Show More
Reference in New Issue
Block a user