Debugging offsets in BinaryIO

.gitignore

@@ -83,7 +83,6 @@ ltmain.sh
 .Trashes
 ehthumbs.db
 Thumbs.db
-.dirstamp
 
 # build directory #
 ###################
@@ -98,8 +97,11 @@ build.sh
 
 # Eigen source #
 ################
-Grid/Eigen
-Eigen/*
+lib/Eigen/*
+
+# FFTW source #
+################
+lib/fftw/*
 
 # libtool macros #
 ##################
@@ -110,8 +112,21 @@ m4/libtool.m4
 ################
 gh-pages/
 
+# Buck files #
+##############
+.buck*
+buck-out
+BUCK
+make-bin-BUCK.sh
+
 # generated sources #
 #####################
-Grid/qcd/spin/gamma-gen/*.h
-Grid/qcd/spin/gamma-gen/*.cc
-Grid/util/Version.h
+lib/qcd/spin/gamma-gen/*.h
+lib/qcd/spin/gamma-gen/*.cc
+lib/version.h
+
+# vs code editor files #
+########################
+.vscode/
+.vscode/settings.json
+settings.json

.travis.yml

@@ -9,11 +9,6 @@ matrix:
     - os:        osx
       osx_image: xcode8.3
       compiler: clang
-      env: PREC=single
-    - os:        osx
-      osx_image: xcode8.3
-      compiler: clang
-      env: PREC=double
       
 before_install:
     - export GRIDDIR=`pwd`
@@ -21,11 +16,9 @@ before_install:
     - if [[ "$TRAVIS_OS_NAME" == "linux" ]] && [[ "$CC" == "clang" ]]; then export PATH="${GRIDDIR}/clang/bin:${PATH}"; fi
     - if [[ "$TRAVIS_OS_NAME" == "linux" ]] && [[ "$CC" == "clang" ]]; then export LD_LIBRARY_PATH="${GRIDDIR}/clang/lib:${LD_LIBRARY_PATH}"; fi
     - if [[ "$TRAVIS_OS_NAME" == "osx" ]]; then brew update; fi
-    - if [[ "$TRAVIS_OS_NAME" == "osx" ]]; then brew install libmpc openssl; fi
+    - if [[ "$TRAVIS_OS_NAME" == "osx" ]]; then brew install libmpc; fi
     
 install:
-    - export CWD=`pwd`
-    - echo $CWD
     - export CC=$CC$VERSION
     - export CXX=$CXX$VERSION
     - echo $PATH
@@ -38,24 +31,17 @@ install:
     - which $CXX
     - $CXX --version
     - if [[ "$TRAVIS_OS_NAME" == "osx" ]]; then export LDFLAGS='-L/usr/local/lib'; fi
-    - if [[ "$TRAVIS_OS_NAME" == "osx" ]]; then export EXTRACONF='--with-openssl=/usr/local/opt/openssl'; fi
     
 script:
     - ./bootstrap.sh
     - mkdir build
     - cd build
-    - mkdir lime
-    - cd lime
-    - mkdir build
-    - cd build
-    - wget http://usqcd-software.github.io/downloads/c-lime/lime-1.3.2.tar.gz
-    - tar xf lime-1.3.2.tar.gz
-    - cd lime-1.3.2
-    - ./configure --prefix=$CWD/build/lime/install
-    - make -j4
-    - make install
-    - cd $CWD/build
-    - ../configure --enable-precision=$PREC --enable-simd=SSE4 --enable-comms=none --with-lime=$CWD/build/lime/install ${EXTRACONF}
+    - ../configure --enable-precision=single --enable-simd=SSE4 --enable-comms=none
     - make -j4 
     - ./benchmarks/Benchmark_dwf --threads 1 --debug-signals
+    - echo make clean
+    - ../configure --enable-precision=double --enable-simd=SSE4 --enable-comms=none
+    - make -j4
+    - ./benchmarks/Benchmark_dwf --threads 1 --debug-signals
     - make check
+

ChangeLog

@@ -1,5 +0,0 @@
-Version : 0.8.0
-
-- Clang 3.5 and above, ICPC v16 and above, GCC 6.3 and above recommended
-- MPI and MPI3 comms optimisations for KNL and OPA finished
-- Half precision comms

Grid/DisableWarnings.h

@@ -1,63 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/DisableWarnings.h
-
-Copyright (C) 2016
-
-Author: Guido Cossu <guido.cossu@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 */
-
-#ifndef DISABLE_WARNINGS_H
-#define DISABLE_WARNINGS_H
-
-
-
-#if defined __GNUC__ && __GNUC__>=6
-#pragma GCC diagnostic ignored "-Wignored-attributes"
-#endif
-
- //disables and intel compiler specific warning (in json.hpp)
-#pragma warning disable 488  
-
-#ifdef __NVCC__
- //disables nvcc specific warning in json.hpp
-#pragma clang diagnostic ignored "-Wdeprecated-register"
-#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
-
-//Eigen only
-#endif
-
-// Disable vectorisation in Eigen on the Power8/9 and PowerPC
-#ifdef  __ALTIVEC__
-#define  EIGEN_DONT_VECTORIZE
-#endif
-#ifdef  __VSX__
-#define  EIGEN_DONT_VECTORIZE
-#endif
-
-#endif

Grid/Grid_Eigen_Dense.h

@@ -1,41 +0,0 @@
-#include <Grid/GridCore.h>
-#pragma once
-// Force Eigen to use MKL if Grid has been configured with --enable-mkl
-#ifdef USE_MKL
-#define EIGEN_USE_MKL_ALL
-#endif
-
-
-#if defined __GNUC__
-#pragma GCC diagnostic push
-#pragma GCC diagnostic ignored "-Wdeprecated-declarations"
-#endif
-
-/* NVCC save and restore compile environment*/
-#ifdef __NVCC__
-#pragma push
-#pragma diag_suppress code_is_unreachable
-#pragma push_macro("__CUDA_ARCH__")
-#pragma push_macro("__NVCC__")
-#pragma push_macro("__CUDACC__")
-#undef __NVCC__
-#undef __CUDACC__
-#undef __CUDA_ARCH__
-#define __NVCC__REDEFINE__
-#endif
-
-#include <Grid/Eigen/Dense>
-#include <Grid/Eigen/unsupported/CXX11/Tensor>
-
-/* NVCC restore */
-#ifdef __NVCC__REDEFINE__
-#pragma pop_macro("__CUDACC__")
-#pragma pop_macro("__NVCC__")
-#pragma pop_macro("__CUDA_ARCH__")
-#pragma pop
-#endif
-
-#if defined __GNUC__
-#pragma GCC diagnostic pop
-#endif
-

Grid/Grid_Eigen_Tensor.h

@@ -1 +0,0 @@
-#include <Grid/Grid_Eigen_Dense.h>

Grid/Namespace.h

@@ -1,38 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/Namespace.h
-
-Copyright (C) 2016
-
-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
-
-#include <type_traits>
-#include <cassert>
-
-#define NAMESPACE_BEGIN(A) namespace A {
-#define NAMESPACE_END(A)   }
-#define GRID_NAMESPACE_BEGIN NAMESPACE_BEGIN(Grid)
-#define GRID_NAMESPACE_END   NAMESPACE_END(Grid)
-#define NAMESPACE_CHECK(x) struct namespaceTEST##x {};  static_assert(std::is_same<namespaceTEST##x, ::namespaceTEST##x>::value,"Not in :: at"  ); 

Grid/algorithms/CoarsenedMatrix.h

@@ -1,512 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/algorithms/CoarsenedMatrix.h
-
-    Copyright (C) 2015
-
-Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is 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_ALGORITHM_COARSENED_MATRIX_H
-#define  GRID_ALGORITHM_COARSENED_MATRIX_H
-
-
-NAMESPACE_BEGIN(Grid);
-
-class Geometry {
-  //    int dimension;
-public:
-  int npoint;
-  std::vector<int> directions   ;
-  std::vector<int> displacements;
-
-  Geometry(int _d)  {
-  
-    int base = (_d==5) ? 1:0;
-
-    // make coarse grid stencil for 4d , not 5d
-    if ( _d==5 ) _d=4;
-
-    npoint = 2*_d+1;
-    directions.resize(npoint);
-    displacements.resize(npoint);
-    for(int d=0;d<_d;d++){
-      directions[2*d  ] = d+base;
-      directions[2*d+1] = d+base;
-      displacements[2*d  ] = +1;
-      displacements[2*d+1] = -1;
-    }
-    directions   [2*_d]=0;
-    displacements[2*_d]=0;
-      
-    //// report back
-    std::cout<<GridLogMessage<<"directions    :";
-    for(int d=0;d<npoint;d++) std::cout<< directions[d]<< " ";
-    std::cout <<std::endl;
-    std::cout<<GridLogMessage<<"displacements :";
-    for(int d=0;d<npoint;d++) std::cout<< displacements[d]<< " ";
-    std::cout<<std::endl;
-  }
-  
-  /*
-  // Original cleaner code
-  Geometry(int _d) : dimension(_d), npoint(2*_d+1), directions(npoint), displacements(npoint) {
-  for(int d=0;d<dimension;d++){
-  directions[2*d  ] = d;
-  directions[2*d+1] = d;
-  displacements[2*d  ] = +1;
-  displacements[2*d+1] = -1;
-  }
-  directions   [2*dimension]=0;
-  displacements[2*dimension]=0;
-  }
-  std::vector<int> GetDelta(int point) {
-  std::vector<int> delta(dimension,0);
-  delta[directions[point]] = displacements[point];
-  return delta;
-  };
-  */    
-
-};
-  
-template<class Fobj,class CComplex,int nbasis>
-class Aggregation   {
-public:
-  typedef iVector<CComplex,nbasis >             siteVector;
-  typedef Lattice<siteVector>                 CoarseVector;
-  typedef Lattice<iMatrix<CComplex,nbasis > > CoarseMatrix;
-
-  typedef Lattice< CComplex >   CoarseScalar; // used for inner products on fine field
-  typedef Lattice<Fobj >        FineField;
-
-  GridBase *CoarseGrid;
-  GridBase *FineGrid;
-  std::vector<Lattice<Fobj> > subspace;
-  int checkerboard;
-  int Checkerboard(void){return checkerboard;}
-  Aggregation(GridBase *_CoarseGrid,GridBase *_FineGrid,int _checkerboard) : 
-    CoarseGrid(_CoarseGrid),
-    FineGrid(_FineGrid),
-    subspace(nbasis,_FineGrid),
-    checkerboard(_checkerboard)
-  {
-  };
-  
-  void Orthogonalise(void){
-    CoarseScalar InnerProd(CoarseGrid); 
-    std::cout << GridLogMessage <<" Gramm-Schmidt pass 1"<<std::endl;
-    blockOrthogonalise(InnerProd,subspace);
-    std::cout << GridLogMessage <<" Gramm-Schmidt pass 2"<<std::endl;
-    blockOrthogonalise(InnerProd,subspace);
-    //      std::cout << GridLogMessage <<" Gramm-Schmidt checking orthogonality"<<std::endl;
-    //      CheckOrthogonal();
-  } 
-  void CheckOrthogonal(void){
-    CoarseVector iProj(CoarseGrid); 
-    CoarseVector eProj(CoarseGrid); 
-    for(int i=0;i<nbasis;i++){
-      blockProject(iProj,subspace[i],subspace);
-      eProj=Zero(); 
-      thread_for(ss, CoarseGrid->oSites(),{
-	eProj[ss](i)=CComplex(1.0);
-      });
-      eProj=eProj - iProj;
-      std::cout<<GridLogMessage<<"Orthog check error "<<i<<" " << norm2(eProj)<<std::endl;
-    }
-    std::cout<<GridLogMessage <<"CheckOrthog done"<<std::endl;
-  }
-  void ProjectToSubspace(CoarseVector &CoarseVec,const FineField &FineVec){
-    blockProject(CoarseVec,FineVec,subspace);
-  }
-  void PromoteFromSubspace(const CoarseVector &CoarseVec,FineField &FineVec){
-    FineVec.Checkerboard() = subspace[0].Checkerboard();
-    blockPromote(CoarseVec,FineVec,subspace);
-  }
-  void CreateSubspaceRandom(GridParallelRNG &RNG){
-    for(int i=0;i<nbasis;i++){
-      random(RNG,subspace[i]);
-      std::cout<<GridLogMessage<<" norm subspace["<<i<<"] "<<norm2(subspace[i])<<std::endl;
-    }
-    Orthogonalise();
-  }
-
-  /*
-    virtual void CreateSubspaceLanczos(GridParallelRNG  &RNG,LinearOperatorBase<FineField> &hermop,int nn=nbasis) 
-    {
-    // Run a Lanczos with sloppy convergence
-    const int Nstop = nn;
-    const int Nk = nn+20;
-    const int Np = nn+20;
-    const int Nm = Nk+Np;
-    const int MaxIt= 10000;
-    RealD resid = 1.0e-3;
-
-    Chebyshev<FineField> Cheb(0.5,64.0,21);
-    ImplicitlyRestartedLanczos<FineField> IRL(hermop,Cheb,Nstop,Nk,Nm,resid,MaxIt);
-    //	IRL.lock = 1;
-
-    FineField noise(FineGrid); gaussian(RNG,noise);
-    FineField tmp(FineGrid); 
-    std::vector<RealD>     eval(Nm);
-    std::vector<FineField> evec(Nm,FineGrid);
-
-    int Nconv;
-    IRL.calc(eval,evec,
-    noise,
-    Nconv);
-
-    // pull back nn vectors
-    for(int b=0;b<nn;b++){
-
-    subspace[b]   = evec[b];
-
-    std::cout << GridLogMessage <<"subspace["<<b<<"] = "<<norm2(subspace[b])<<std::endl;
-
-    hermop.Op(subspace[b],tmp); 
-    std::cout<<GridLogMessage << "filtered["<<b<<"] <f|MdagM|f> "<<norm2(tmp)<<std::endl;
-
-    noise = tmp -  sqrt(eval[b])*subspace[b] ;
-
-    std::cout<<GridLogMessage << " lambda_"<<b<<" = "<< eval[b] <<"  ;  [ M - Lambda ]_"<<b<<" vec_"<<b<<"  = " <<norm2(noise)<<std::endl;
-
-    noise = tmp +  eval[b]*subspace[b] ;
-
-    std::cout<<GridLogMessage << " lambda_"<<b<<" = "<< eval[b] <<"  ;  [ M - Lambda ]_"<<b<<" vec_"<<b<<"  = " <<norm2(noise)<<std::endl;
-
-    }
-    Orthogonalise();
-    for(int b=0;b<nn;b++){
-    std::cout << GridLogMessage <<"subspace["<<b<<"] = "<<norm2(subspace[b])<<std::endl;
-    }
-    }
-  */
-  virtual void CreateSubspace(GridParallelRNG  &RNG,LinearOperatorBase<FineField> &hermop,int nn=nbasis) {
-
-    RealD scale;
-
-    ConjugateGradient<FineField> CG(1.0e-2,10000);
-    FineField noise(FineGrid);
-    FineField Mn(FineGrid);
-
-    for(int b=0;b<nn;b++){
-	
-      subspace[b] = Zero();
-      gaussian(RNG,noise);
-      scale = std::pow(norm2(noise),-0.5); 
-      noise=noise*scale;
-	
-      hermop.Op(noise,Mn); std::cout<<GridLogMessage << "noise   ["<<b<<"] <n|MdagM|n> "<<norm2(Mn)<<std::endl;
-
-      for(int i=0;i<1;i++){
-
-	CG(hermop,noise,subspace[b]);
-
-	noise = subspace[b];
-	scale = std::pow(norm2(noise),-0.5); 
-	noise=noise*scale;
-
-      }
-
-      hermop.Op(noise,Mn); std::cout<<GridLogMessage << "filtered["<<b<<"] <f|MdagM|f> "<<norm2(Mn)<<std::endl;
-      subspace[b]   = noise;
-
-    }
-
-    Orthogonalise();
-
-  }
-};
-// Fine Object == (per site) type of fine field
-// nbasis      == number of deflation vectors
-template<class Fobj,class CComplex,int nbasis>
-class CoarsenedMatrix : public SparseMatrixBase<Lattice<iVector<CComplex,nbasis > > >  {
-public:
-    
-  typedef iVector<CComplex,nbasis >             siteVector;
-  typedef Lattice<siteVector>                 CoarseVector;
-  typedef Lattice<iMatrix<CComplex,nbasis > > CoarseMatrix;
-
-  typedef Lattice< CComplex >   CoarseScalar; // used for inner products on fine field
-  typedef Lattice<Fobj >        FineField;
-
-  ////////////////////
-  // Data members
-  ////////////////////
-  Geometry         geom;
-  GridBase *       _grid; 
-
-  CartesianStencil<siteVector,siteVector,int> Stencil; 
-
-  std::vector<CoarseMatrix> A;
-
-      
-  ///////////////////////
-  // Interface
-  ///////////////////////
-  GridBase * Grid(void)         { return _grid; };   // this is all the linalg routines need to know
-
-  RealD M (const CoarseVector &in, CoarseVector &out){
-
-    conformable(_grid,in.Grid());
-    conformable(in.Grid(),out.Grid());
-
-    SimpleCompressor<siteVector> compressor;
-    Stencil.HaloExchange(in,compressor);
-    auto in_v = in.View();
-    auto out_v = in.View();
-    thread_for(ss,Grid()->oSites(),{
-      siteVector res = Zero();
-      siteVector nbr;
-      int ptype;
-      StencilEntry *SE;
-      for(int point=0;point<geom.npoint;point++){
-
-	SE=Stencil.GetEntry(ptype,point,ss);
-	  
-	if(SE->_is_local&&SE->_permute) { 
-	  permute(nbr,in_v[SE->_offset],ptype);
-	} else if(SE->_is_local) { 
-	  nbr = in_v[SE->_offset];
-	} else {
-	  nbr = Stencil.CommBuf()[SE->_offset];
-	}
-	auto A_point = A[point].View();
-	res = res + A_point[ss]*nbr;
-      }
-      vstream(out_v[ss],res);
-    });
-    return norm2(out);
-  };
-
-  RealD Mdag (const CoarseVector &in, CoarseVector &out){
-    // // corresponds to Petrov-Galerkin coarsening
-    // return M(in,out);
-    
-    // corresponds to Galerkin coarsening
-    CoarseVector tmp(Grid());
-    G5C(tmp, in);
-    M(tmp, out);
-    G5C(out, out);
-    return norm2(out);
-  };
-
-  void Mdir(const CoarseVector &in, CoarseVector &out, int dir, int disp){
-    
-    conformable(_grid,in.Grid());
-    conformable(in.Grid(),out.Grid());
-    
-    SimpleCompressor<siteVector> compressor;
-    Stencil.HaloExchange(in,compressor);
-    
-    auto point = [dir, disp](){
-      if(dir == 0 and disp == 0)
-	return 8;
-      else
-	return (4 * dir + 1 - disp) / 2;
-    }();
-
-    auto out_v = out.View();
-    auto in_v  = in.View();
-    thread_for(ss,Grid()->oSites(),{
-      siteVector res = Zero();
-      siteVector nbr;
-      int ptype;
-      StencilEntry *SE;
-      
-      SE=Stencil.GetEntry(ptype,point,ss);
-      
-      if(SE->_is_local&&SE->_permute) {
-	permute(nbr,in_v[SE->_offset],ptype);
-      } else if(SE->_is_local) {
-	nbr = in_v[SE->_offset];
-      } else {
-	nbr = Stencil.CommBuf()[SE->_offset];
-      }
-
-      auto A_point = A[point].View();
-      res = res + A_point[ss]*nbr;
-      
-      vstream(out_v[ss],res);
-    });
-  };
-
-  void Mdiag(const CoarseVector &in, CoarseVector &out){
-    Mdir(in, out, 0, 0); // use the self coupling (= last) point of the stencil
-  };
-
-  
- CoarsenedMatrix(GridCartesian &CoarseGrid) 	: 
-
-    _grid(&CoarseGrid),
-    geom(CoarseGrid._ndimension),
-    Stencil(&CoarseGrid,geom.npoint,Even,geom.directions,geom.displacements,0),
-    A(geom.npoint,&CoarseGrid)
-  {
-  };
-
-  void CoarsenOperator(GridBase *FineGrid,LinearOperatorBase<Lattice<Fobj> > &linop,
-		       Aggregation<Fobj,CComplex,nbasis> & Subspace){
-
-    FineField iblock(FineGrid); // contributions from within this block
-    FineField oblock(FineGrid); // contributions from outwith this block
-
-    FineField     phi(FineGrid);
-    FineField     tmp(FineGrid);
-    FineField     zz(FineGrid); zz=Zero();
-    FineField    Mphi(FineGrid);
-
-    Lattice<iScalar<vInteger> > coor(FineGrid);
-
-    CoarseVector iProj(Grid()); 
-    CoarseVector oProj(Grid()); 
-    CoarseScalar InnerProd(Grid()); 
-
-    // Orthogonalise the subblocks over the basis
-    blockOrthogonalise(InnerProd,Subspace.subspace);
-
-    // Compute the matrix elements of linop between this orthonormal
-    // set of vectors.
-    int self_stencil=-1;
-    for(int p=0;p<geom.npoint;p++){ 
-      A[p]=Zero();
-      if( geom.displacements[p]==0){
-	self_stencil=p;
-      }
-    }
-    assert(self_stencil!=-1);
-
-    for(int i=0;i<nbasis;i++){
-      phi=Subspace.subspace[i];
-	
-      std::cout<<GridLogMessage<<"("<<i<<").."<<std::endl;
-
-      for(int p=0;p<geom.npoint;p++){ 
-
-	int dir   = geom.directions[p];
-	int disp  = geom.displacements[p];
-
-	Integer block=(FineGrid->_rdimensions[dir])/(Grid()->_rdimensions[dir]);
-
-	LatticeCoordinate(coor,dir);
-
-	if ( disp==0 ){
-	  linop.OpDiag(phi,Mphi);
-	}
-	else  {
-	  linop.OpDir(phi,Mphi,dir,disp); 
-	}
-
-	////////////////////////////////////////////////////////////////////////
-	// Pick out contributions coming from this cell and neighbour cell
-	////////////////////////////////////////////////////////////////////////
-	if ( disp==0 ) {
-	  iblock = Mphi;
-	  oblock = Zero();
-	} else if ( disp==1 ) {
-	  oblock = where(mod(coor,block)==(block-1),Mphi,zz);
-	  iblock = where(mod(coor,block)!=(block-1),Mphi,zz);
-	} else if ( disp==-1 ) {
-	  oblock = where(mod(coor,block)==(Integer)0,Mphi,zz);
-	  iblock = where(mod(coor,block)!=(Integer)0,Mphi,zz);
-	} else {
-	  assert(0);
-	}
-
-	Subspace.ProjectToSubspace(iProj,iblock);
-	Subspace.ProjectToSubspace(oProj,oblock);
-	//	  blockProject(iProj,iblock,Subspace.subspace);
-	//	  blockProject(oProj,oblock,Subspace.subspace);
-	auto iProj_v = iProj.View() ;
-	auto oProj_v = oProj.View() ;
-	auto A_p     =  A[p].View();
-	auto A_self  = A[self_stencil].View();
-	thread_for(ss, Grid()->oSites(),{
-	  for(int j=0;j<nbasis;j++){
-	    if( disp!= 0 ) {
-	      A_p[ss](j,i) = oProj_v[ss](j);
-	    }
-	    A_self[ss](j,i) =	A_self[ss](j,i) + iProj_v[ss](j);
-	  }
-	});
-      }
-    }
-
-#if 0
-    ///////////////////////////
-    // test code worth preserving in if block
-    ///////////////////////////
-    std::cout<<GridLogMessage<< " Computed matrix elements "<< self_stencil <<std::endl;
-    for(int p=0;p<geom.npoint;p++){
-      std::cout<<GridLogMessage<< "A["<<p<<"]" << std::endl;
-      std::cout<<GridLogMessage<< A[p] << std::endl;
-    }
-    std::cout<<GridLogMessage<< " picking by block0 "<< self_stencil <<std::endl;
-
-    phi=Subspace.subspace[0];
-    std::vector<int> bc(FineGrid->_ndimension,0);
-
-    blockPick(Grid(),phi,tmp,bc);      // Pick out a block
-    linop.Op(tmp,Mphi);                // Apply big dop
-    blockProject(iProj,Mphi,Subspace.subspace); // project it and print it
-    std::cout<<GridLogMessage<< " Computed matrix elements from block zero only "<<std::endl;
-    std::cout<<GridLogMessage<< iProj <<std::endl;
-    std::cout<<GridLogMessage<<"Computed Coarse Operator"<<std::endl;
-#endif
-      //      ForceHermitian();
-      // AssertHermitian();
-      // ForceDiagonal();
-  }
-
-  void ForceHermitian(void) {
-    for(int d=0;d<4;d++){
-      int dd=d+1;
-      A[2*d] = adj(Cshift(A[2*d+1],dd,1));
-    }
-    //      A[8] = 0.5*(A[8] + adj(A[8]));
-  }
-  void AssertHermitian(void) {
-    CoarseMatrix AA    (Grid());
-    CoarseMatrix AAc   (Grid());
-    CoarseMatrix Diff  (Grid());
-    for(int d=0;d<4;d++){
-	
-      int dd=d+1;
-      AAc = Cshift(A[2*d+1],dd,1);
-      AA  = A[2*d];
-	
-      Diff = AA - adj(AAc);
-
-      std::cout<<GridLogMessage<<"Norm diff dim "<<d<<" "<< norm2(Diff)<<std::endl;
-      std::cout<<GridLogMessage<<"Norm dim "<<d<<" "<< norm2(AA)<<std::endl;
-	  
-    }
-    Diff = A[8] - adj(A[8]);
-    std::cout<<GridLogMessage<<"Norm diff local "<< norm2(Diff)<<std::endl;
-    std::cout<<GridLogMessage<<"Norm local "<< norm2(A[8])<<std::endl;
-  }
-    
-};
-
-NAMESPACE_END(Grid);
-#endif

Grid/algorithms/FFT.h

@@ -1,291 +0,0 @@
-
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/Cshift.h
-
-    Copyright (C) 2015
-
-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 */
-#ifndef _GRID_FFT_H_
-#define _GRID_FFT_H_
-
-#ifdef HAVE_FFTW
-#ifdef USE_MKL
-#include <fftw/fftw3.h>
-#else
-#include <fftw3.h>
-#endif
-#endif
-
-
-NAMESPACE_BEGIN(Grid);
-
-template<class scalar> struct FFTW { };
-
-#ifdef HAVE_FFTW	
-template<> struct FFTW<ComplexD> {
-public:
-
-  typedef fftw_complex FFTW_scalar;
-  typedef fftw_plan    FFTW_plan;
-
-  static FFTW_plan fftw_plan_many_dft(int rank, const int *n,int howmany,
-				      FFTW_scalar *in, const int *inembed,		
-				      int istride, int idist,		
-				      FFTW_scalar *out, const int *onembed,		
-				      int ostride, int odist,		
-				      int sign, unsigned flags) {
-    return ::fftw_plan_many_dft(rank,n,howmany,in,inembed,istride,idist,out,onembed,ostride,odist,sign,flags);
-  }	  
-    
-  static void fftw_flops(const FFTW_plan p,double *add, double *mul, double *fmas){
-    ::fftw_flops(p,add,mul,fmas);
-  }
-
-  inline static void fftw_execute_dft(const FFTW_plan p,FFTW_scalar *in,FFTW_scalar *out) {
-    ::fftw_execute_dft(p,in,out);
-  }
-  inline static void fftw_destroy_plan(const FFTW_plan p) {
-    ::fftw_destroy_plan(p);
-  }
-};
-
-template<> struct FFTW<ComplexF> {
-public:
-
-  typedef fftwf_complex FFTW_scalar;
-  typedef fftwf_plan    FFTW_plan;
-
-  static FFTW_plan fftw_plan_many_dft(int rank, const int *n,int howmany,
-				      FFTW_scalar *in, const int *inembed,		
-				      int istride, int idist,		
-				      FFTW_scalar *out, const int *onembed,		
-				      int ostride, int odist,		
-				      int sign, unsigned flags) {
-    return ::fftwf_plan_many_dft(rank,n,howmany,in,inembed,istride,idist,out,onembed,ostride,odist,sign,flags);
-  }	  
-    
-  static void fftw_flops(const FFTW_plan p,double *add, double *mul, double *fmas){
-    ::fftwf_flops(p,add,mul,fmas);
-  }
-
-  inline static void fftw_execute_dft(const FFTW_plan p,FFTW_scalar *in,FFTW_scalar *out) {
-    ::fftwf_execute_dft(p,in,out);
-  }
-  inline static void fftw_destroy_plan(const FFTW_plan p) {
-    ::fftwf_destroy_plan(p);
-  }
-};
-
-#endif
-
-#ifndef FFTW_FORWARD
-#define FFTW_FORWARD (-1)
-#define FFTW_BACKWARD (+1)
-#endif
-
-class FFT {
-private:
-    
-  GridCartesian *vgrid;
-  GridCartesian *sgrid;
-    
-  int Nd;
-  double flops;
-  double flops_call;
-  uint64_t usec;
-    
-  Coordinate dimensions;
-  Coordinate processors;
-  Coordinate processor_coor;
-    
-public:
-    
-  static const int forward=FFTW_FORWARD;
-  static const int backward=FFTW_BACKWARD;
-    
-  double Flops(void) {return flops;}
-  double MFlops(void) {return flops/usec;}
-  double USec(void)   {return (double)usec;}    
-
-  FFT ( GridCartesian * grid ) :
-    vgrid(grid),
-    Nd(grid->_ndimension),
-    dimensions(grid->_fdimensions),
-    processors(grid->_processors),
-    processor_coor(grid->_processor_coor)
-  {
-    flops=0;
-    usec =0;
-    Coordinate layout(Nd,1);
-    sgrid = new GridCartesian(dimensions,layout,processors);
-  };
-    
-  ~FFT ( void)  {
-    delete sgrid;
-  }
-    
-  template<class vobj>
-  void FFT_dim_mask(Lattice<vobj> &result,const Lattice<vobj> &source,Coordinate mask,int sign){
-
-    conformable(result.Grid(),vgrid);
-    conformable(source.Grid(),vgrid);
-    Lattice<vobj> tmp(vgrid);
-    tmp = source;
-    for(int d=0;d<Nd;d++){
-      if( mask[d] ) {
-	FFT_dim(result,tmp,d,sign);
-	tmp=result;
-      }
-    }
-  }
-
-  template<class vobj>
-  void FFT_all_dim(Lattice<vobj> &result,const Lattice<vobj> &source,int sign){
-    Coordinate mask(Nd,1);
-    FFT_dim_mask(result,source,mask,sign);
-  }
-
-
-  template<class vobj>
-  void FFT_dim(Lattice<vobj> &result,const Lattice<vobj> &source,int dim, int sign){
-#ifndef HAVE_FFTW
-    assert(0);
-#else
-    conformable(result.Grid(),vgrid);
-    conformable(source.Grid(),vgrid);
-
-    int L = vgrid->_ldimensions[dim];
-    int G = vgrid->_fdimensions[dim];
-      
-    Coordinate layout(Nd,1);
-    Coordinate pencil_gd(vgrid->_fdimensions);
-      
-    pencil_gd[dim] = G*processors[dim];
-      
-    // Pencil global vol LxLxGxLxL per node
-    GridCartesian pencil_g(pencil_gd,layout,processors);
-      
-    // Construct pencils
-    typedef typename vobj::scalar_object sobj;
-    typedef typename sobj::scalar_type   scalar;
-      
-    Lattice<sobj> pgbuf(&pencil_g);
-    auto pgbuf_v = pgbuf.View();
-
-    typedef typename FFTW<scalar>::FFTW_scalar FFTW_scalar;
-    typedef typename FFTW<scalar>::FFTW_plan   FFTW_plan;
-      
-    int Ncomp = sizeof(sobj)/sizeof(scalar);
-    int Nlow  = 1;
-    for(int d=0;d<dim;d++){
-      Nlow*=vgrid->_ldimensions[d];
-    }
-      
-    int rank = 1;  /* 1d transforms */
-    int n[] = {G}; /* 1d transforms of length G */
-    int howmany = Ncomp;
-    int odist,idist,istride,ostride;
-    idist   = odist   = 1;          /* Distance between consecutive FT's */
-    istride = ostride = Ncomp*Nlow; /* distance between two elements in the same FT */
-    int *inembed = n, *onembed = n;
-      
-    scalar div;
-    if ( sign == backward ) div = 1.0/G;
-    else if ( sign == forward ) div = 1.0;
-    else assert(0);
-      
-    FFTW_plan p;
-    {
-      FFTW_scalar *in = (FFTW_scalar *)&pgbuf_v[0];
-      FFTW_scalar *out= (FFTW_scalar *)&pgbuf_v[0];
-      p = FFTW<scalar>::fftw_plan_many_dft(rank,n,howmany,
-					   in,inembed,
-					   istride,idist,
-					   out,onembed,
-					   ostride, odist,
-					   sign,FFTW_ESTIMATE);
-    }
-      
-    // Barrel shift and collect global pencil
-    Coordinate lcoor(Nd), gcoor(Nd);
-    result = source;
-    int pc = processor_coor[dim];
-    for(int p=0;p<processors[dim];p++) {
-      thread_for(idx, sgrid->lSites(),{
-          Coordinate cbuf(Nd);
-          sobj s;
-	  sgrid->LocalIndexToLocalCoor(idx,cbuf);
-	  peekLocalSite(s,result,cbuf);
-	  cbuf[dim]+=((pc+p) % processors[dim])*L;
-	  //            cbuf[dim]+=p*L;
-	  pokeLocalSite(s,pgbuf,cbuf);
-      });
-      if (p != processors[dim] - 1) {
-	result = Cshift(result,dim,L);
-      }
-    }
-      
-    // Loop over orthog coords
-    int NN=pencil_g.lSites();
-    GridStopWatch timer;
-    timer.Start();
-    thread_for( idx,NN,{
-        Coordinate cbuf(Nd);
-	pencil_g.LocalIndexToLocalCoor(idx, cbuf);
-	if ( cbuf[dim] == 0 ) {  // restricts loop to plane at lcoor[dim]==0
-	  FFTW_scalar *in = (FFTW_scalar *)&pgbuf_v[idx];
-	  FFTW_scalar *out= (FFTW_scalar *)&pgbuf_v[idx];
-	  FFTW<scalar>::fftw_execute_dft(p,in,out);
-	}
-    });
-    timer.Stop();
-      
-    // performance counting
-    double add,mul,fma;
-    FFTW<scalar>::fftw_flops(p,&add,&mul,&fma);
-    flops_call = add+mul+2.0*fma;
-    usec += timer.useconds();
-    flops+= flops_call*NN;
-      
-    // writing out result
-    thread_for(idx,sgrid->lSites(),{
-	Coordinate clbuf(Nd), cgbuf(Nd);
-	sobj s;
-	sgrid->LocalIndexToLocalCoor(idx,clbuf);
-	cgbuf = clbuf;
-	cgbuf[dim] = clbuf[dim]+L*pc;
-	peekLocalSite(s,pgbuf,cgbuf);
-	pokeLocalSite(s,result,clbuf);
-    });
-    result = result*div;
-      
-    // destroying plan
-    FFTW<scalar>::fftw_destroy_plan(p);
-#endif
-  }
-};
-
-NAMESPACE_END(Grid);
-
-#endif

Grid/algorithms/LinearOperator.h

@@ -1,495 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/algorithms/LinearOperator.h
-
-    Copyright (C) 2015
-
-Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
-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);
-
-/////////////////////////////////////////////////////////////////////////////////////////////
-// LinearOperators Take a something and return a something.
-/////////////////////////////////////////////////////////////////////////////////////////////
-//
-// Hopefully linearity is satisfied and the AdjOp is indeed the Hermitian Conjugateugate (transpose if real):
-//SBase
-//   i)  F(a x + b y) = aF(x) + b F(y).
-//  ii)  <x|Op|y> = <y|AdjOp|x>^\ast
-//
-// Would be fun to have a test linearity & Herm Conj function!
-/////////////////////////////////////////////////////////////////////////////////////////////
-template<class Field> class LinearOperatorBase {
-public:
-
-  // Support for coarsening to a multigrid
-  virtual void OpDiag (const Field &in, Field &out) = 0; // Abstract base
-  virtual void OpDir  (const Field &in, Field &out,int dir,int disp) = 0; // Abstract base
-
-  virtual void Op     (const Field &in, Field &out) = 0; // Abstract base
-  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;
-};
-
-
-/////////////////////////////////////////////////////////////////////////////////////////////
-// By sharing the class for Sparse Matrix across multiple operator wrappers, we can share code
-// between RB and non-RB variants. Sparse matrix is like the fermion action def, and then
-// the wrappers implement the specialisation of "Op" and "AdjOp" to the cases minimising
-// replication of code.
-//
-// I'm not entirely happy with implementation; to share the Schur code between herm and non-herm
-// while still having a "OpAndNorm" in the abstract base I had to implement it in both cases
-// with an assert trap in the non-herm. This isn't right; there must be a better C++ way to
-// do it, but I fear it required multiple inheritance and mixed in abstract base classes
-/////////////////////////////////////////////////////////////////////////////////////////////
-
-////////////////////////////////////////////////////////////////////
-// Construct herm op from non-herm matrix
-////////////////////////////////////////////////////////////////////
-template<class Matrix,class Field>
-class MdagMLinearOperator : public LinearOperatorBase<Field> {
-  Matrix &_Mat;
-public:
-  MdagMLinearOperator(Matrix &Mat): _Mat(Mat){};
-
-  // Support for coarsening to a multigrid
-  void OpDiag (const Field &in, Field &out) {
-    _Mat.Mdiag(in,out);
-  }
-  void OpDir  (const Field &in, Field &out,int dir,int disp) {
-    _Mat.Mdir(in,out,dir,disp);
-  }
-  void Op     (const Field &in, Field &out){
-    _Mat.M(in,out);
-  }
-  void AdjOp     (const Field &in, Field &out){
-    _Mat.Mdag(in,out);
-  }
-  void HermOpAndNorm(const Field &in, Field &out,RealD &n1,RealD &n2){
-    _Mat.MdagM(in,out,n1,n2);
-  }
-  void HermOp(const Field &in, Field &out){
-    RealD n1,n2;
-    HermOpAndNorm(in,out,n1,n2);
-  }
-};
-
-////////////////////////////////////////////////////////////////////
-// Construct herm op and shift it for mgrid smoother
-////////////////////////////////////////////////////////////////////
-template<class Matrix,class Field>
-class ShiftedMdagMLinearOperator : public LinearOperatorBase<Field> {
-  Matrix &_Mat;
-  RealD _shift;
-public:
-  ShiftedMdagMLinearOperator(Matrix &Mat,RealD shift): _Mat(Mat), _shift(shift){};
-  // Support for coarsening to a multigrid
-  void OpDiag (const Field &in, Field &out) {
-    _Mat.Mdiag(in,out);
-    assert(0);
-  }
-  void OpDir  (const Field &in, Field &out,int dir,int disp) {
-    _Mat.Mdir(in,out,dir,disp);
-    assert(0);
-  }
-  void Op     (const Field &in, Field &out){
-    _Mat.M(in,out);
-    assert(0);
-  }
-  void AdjOp     (const Field &in, Field &out){
-    _Mat.Mdag(in,out);
-    assert(0);
-  }
-  void HermOpAndNorm(const Field &in, Field &out,RealD &n1,RealD &n2){
-    _Mat.MdagM(in,out,n1,n2);
-    out = out + _shift*in;
-
-    ComplexD dot;	
-    dot= innerProduct(in,out);
-    n1=real(dot);
-    n2=norm2(out);
-  }
-  void HermOp(const Field &in, Field &out){
-    RealD n1,n2;
-    HermOpAndNorm(in,out,n1,n2);
-  }
-};
-
-////////////////////////////////////////////////////////////////////
-// Wrap an already herm matrix
-////////////////////////////////////////////////////////////////////
-template<class Matrix,class Field>
-class HermitianLinearOperator : public LinearOperatorBase<Field> {
-  Matrix &_Mat;
-public:
-  HermitianLinearOperator(Matrix &Mat): _Mat(Mat){};
-  // Support for coarsening to a multigrid
-  void OpDiag (const Field &in, Field &out) {
-    _Mat.Mdiag(in,out);
-  }
-  void OpDir  (const Field &in, Field &out,int dir,int disp) {
-    _Mat.Mdir(in,out,dir,disp);
-  }
-  void Op     (const Field &in, Field &out){
-    _Mat.M(in,out);
-  }
-  void AdjOp     (const Field &in, Field &out){
-    _Mat.M(in,out);
-  }
-  void HermOpAndNorm(const Field &in, Field &out,RealD &n1,RealD &n2){
-    _Mat.M(in,out);
-	
-    ComplexD dot= innerProduct(in,out); n1=real(dot);
-    n2=norm2(out);
-  }
-  void HermOp(const Field &in, Field &out){
-    _Mat.M(in,out);
-  }
-};
-
-    //////////////////////////////////////////////////////////
-    // Even Odd Schur decomp operators; there are several
-    // ways to introduce the even odd checkerboarding
-    //////////////////////////////////////////////////////////
-
-    template<class Field>
-    class SchurOperatorBase :  public LinearOperatorBase<Field> {
-    public:
-      virtual  RealD Mpc      (const Field &in, Field &out) =0;
-      virtual  RealD MpcDag   (const Field &in, Field &out) =0;
-      virtual void MpcDagMpc(const Field &in, Field &out,RealD &ni,RealD &no) {
-      Field tmp(in.Grid());
-      tmp.Checkerboard() = in.Checkerboard();
-	ni=Mpc(in,tmp);
-	no=MpcDag(tmp,out);
-      }
-      virtual void HermOpAndNorm(const Field &in, Field &out,RealD &n1,RealD &n2){
-      out.Checkerboard() = in.Checkerboard();
-	MpcDagMpc(in,out,n1,n2);
-      }
-      virtual void HermOp(const Field &in, Field &out){
-	RealD n1,n2;
-	HermOpAndNorm(in,out,n1,n2);
-      }
-      void Op     (const Field &in, Field &out){
-	Mpc(in,out);
-      }
-      void AdjOp     (const Field &in, Field &out){ 
-	MpcDag(in,out);
-      }
-      // Support for coarsening to a multigrid
-      void OpDiag (const Field &in, Field &out) {
-	assert(0); // must coarsen the unpreconditioned system
-      }
-      void OpDir  (const Field &in, Field &out,int dir,int disp) {
-	assert(0);
-      }
-    };
-    template<class Matrix,class Field>
-    class SchurDiagMooeeOperator :  public SchurOperatorBase<Field> {
-    public:
-      Matrix &_Mat;
-      SchurDiagMooeeOperator (Matrix &Mat): _Mat(Mat){};
-      virtual  RealD Mpc      (const Field &in, Field &out) {
-      Field tmp(in.Grid());
-      tmp.Checkerboard() = !in.Checkerboard();
-	//std::cout <<"grid pointers: in._grid="<< in._grid << " out._grid=" << out._grid << "  _Mat.Grid=" << _Mat.Grid() << " _Mat.RedBlackGrid=" << _Mat.RedBlackGrid() << std::endl;
-
-	_Mat.Meooe(in,tmp);
-	_Mat.MooeeInv(tmp,out);
-	_Mat.Meooe(out,tmp);
-
-      //std::cout << "cb in " << in.Checkerboard() << "  cb out " << out.Checkerboard() << std::endl;
-	_Mat.Mooee(in,out);
-	return axpy_norm(out,-1.0,tmp,out);
-      }
-      virtual  RealD MpcDag   (const Field &in, Field &out){
-	Field tmp(in.Grid());
-
-	_Mat.MeooeDag(in,tmp);
-        _Mat.MooeeInvDag(tmp,out);
-	_Mat.MeooeDag(out,tmp);
-
-	_Mat.MooeeDag(in,out);
-	return axpy_norm(out,-1.0,tmp,out);
-      }
-    };
-    template<class Matrix,class Field>
-      class SchurDiagOneOperator :  public SchurOperatorBase<Field> {
-    protected:
-      Matrix &_Mat;
-    public:
-      SchurDiagOneOperator (Matrix &Mat): _Mat(Mat){};
-
-      virtual  RealD Mpc      (const Field &in, Field &out) {
-	Field tmp(in.Grid());
-
-	_Mat.Meooe(in,out);
-	_Mat.MooeeInv(out,tmp);
-	_Mat.Meooe(tmp,out);
-	_Mat.MooeeInv(out,tmp);
-
-	return axpy_norm(out,-1.0,tmp,in);
-      }
-      virtual  RealD MpcDag   (const Field &in, Field &out){
-	Field tmp(in.Grid());
-
-	_Mat.MooeeInvDag(in,out);
-	_Mat.MeooeDag(out,tmp);
-	_Mat.MooeeInvDag(tmp,out);
-	_Mat.MeooeDag(out,tmp);
-
-	return axpy_norm(out,-1.0,tmp,in);
-      }
-    };
-    template<class Matrix,class Field>
-      class SchurDiagTwoOperator :  public SchurOperatorBase<Field> {
-    protected:
-      Matrix &_Mat;
-    public:
-      SchurDiagTwoOperator (Matrix &Mat): _Mat(Mat){};
-
-      virtual  RealD Mpc      (const Field &in, Field &out) {
-	Field tmp(in.Grid());
-
-	_Mat.MooeeInv(in,out);
-	_Mat.Meooe(out,tmp);
-	_Mat.MooeeInv(tmp,out);
-	_Mat.Meooe(out,tmp);
-
-	return axpy_norm(out,-1.0,tmp,in);
-      }
-      virtual  RealD MpcDag   (const Field &in, Field &out){
-	Field tmp(in.Grid());
-
-	_Mat.MeooeDag(in,out);
-	_Mat.MooeeInvDag(out,tmp);
-	_Mat.MeooeDag(tmp,out);
-	_Mat.MooeeInvDag(out,tmp);
-
-	return axpy_norm(out,-1.0,tmp,in);
-      }
-    };
-    ///////////////////////////////////////////////////////////////////////////////////////////////////
-    // Left  handed Moo^-1 ; (Moo - Moe Mee^-1 Meo) psi = eta  -->  ( 1 - Moo^-1 Moe Mee^-1 Meo ) psi = Moo^-1 eta
-    // Right handed Moo^-1 ; (Moo - Moe Mee^-1 Meo) Moo^-1 Moo psi = eta  -->  ( 1 - Moe Mee^-1 Meo ) Moo^-1 phi=eta ; psi = Moo^-1 phi
-    ///////////////////////////////////////////////////////////////////////////////////////////////////
-    template<class Matrix,class Field> using SchurDiagOneRH = SchurDiagTwoOperator<Matrix,Field> ;
-    template<class Matrix,class Field> using SchurDiagOneLH = SchurDiagOneOperator<Matrix,Field> ;
-    ///////////////////////////////////////////////////////////////////////////////////////////////////
-    //  Staggered use
-    ///////////////////////////////////////////////////////////////////////////////////////////////////
-    template<class Matrix,class Field>
-      class SchurStaggeredOperator :  public SchurOperatorBase<Field> {
-    protected:
-      Matrix &_Mat;
-      Field tmp;
-      RealD mass;
-      double tMpc;
-      double tIP;
-      double tMeo;
-      double taxpby_norm;
-      uint64_t ncall;
-public:
-      void Report(void)
-      {
-	std::cout << GridLogMessage << " HermOpAndNorm.Mpc "<< tMpc/ncall<<" usec "<<std::endl;
-	std::cout << GridLogMessage << " HermOpAndNorm.IP  "<< tIP /ncall<<" usec "<<std::endl;
-	std::cout << GridLogMessage << " Mpc.MeoMoe        "<< tMeo/ncall<<" usec "<<std::endl;
-	std::cout << GridLogMessage << " Mpc.axpby_norm    "<< taxpby_norm/ncall<<" usec "<<std::endl;
-      }
-      SchurStaggeredOperator (Matrix &Mat): _Mat(Mat), tmp(_Mat.RedBlackGrid()) 
-      { 
-	assert( _Mat.isTrivialEE() );
-	mass = _Mat.Mass();
-	tMpc=0;
-	tIP =0;
-        tMeo=0;
-        taxpby_norm=0;
-	ncall=0;
-      }
-  virtual void HermOpAndNorm(const Field &in, Field &out,RealD &n1,RealD &n2){
-	ncall++;
-	tMpc-=usecond();
-    n2 = Mpc(in,out);
-	tMpc+=usecond();
-	tIP-=usecond();
-    ComplexD dot= innerProduct(in,out);
-	tIP+=usecond();
-    n1 = real(dot);
-  }
-  virtual void HermOp(const Field &in, Field &out){
-	ncall++;
-	tMpc-=usecond();
-	_Mat.Meooe(in,out);
-	_Mat.Meooe(out,tmp);
-	tMpc+=usecond();
-	taxpby_norm-=usecond();
-	axpby(out,-1.0,mass*mass,tmp,in);
-	taxpby_norm+=usecond();
-  }
-  virtual  RealD Mpc      (const Field &in, Field &out) 
-  {
-
-    Field tmp(in.Grid());
-    Field tmp2(in.Grid());
-
-    //    std::cout << GridLogIterative << " HermOp.Mpc "<<std::endl;
-    _Mat.Mooee(in,out);
-    _Mat.Mooee(out,tmp);
-    //    std::cout << GridLogIterative << " HermOp.MooeeMooee "<<std::endl;
-
-    tMeo-=usecond();
-    _Mat.Meooe(in,out);
-    _Mat.Meooe(out,tmp);
-    tMeo+=usecond();
-    taxpby_norm-=usecond();
-    RealD nn=axpby_norm(out,-1.0,mass*mass,tmp,in);
-    taxpby_norm+=usecond();
-    return nn;
-  }
-  virtual  RealD MpcDag   (const Field &in, Field &out){
-    return Mpc(in,out);
-  }
-  virtual void MpcDagMpc(const Field &in, Field &out,RealD &ni,RealD &no) {
-    assert(0);// Never need with staggered
-  }
-};
-template<class Matrix,class Field> using SchurStagOperator = SchurStaggeredOperator<Matrix,Field>;
-
-
-/////////////////////////////////////////////////////////////
-// Base classes for functions of operators
-/////////////////////////////////////////////////////////////
-template<class Field> class OperatorFunction {
-public:
-  virtual void operator() (LinearOperatorBase<Field> &Linop, const Field &in, Field &out) = 0;
-  virtual void operator() (LinearOperatorBase<Field> &Linop, const std::vector<Field> &in,std::vector<Field> &out) {
-    assert(in.size()==out.size());
-    for(int k=0;k<in.size();k++){
-      (*this)(Linop,in[k],out[k]);
-    }
-  };
-};
-
-template<class Field> class LinearFunction {
-public:
-  virtual void operator() (const Field &in, Field &out) = 0;
-};
-
-template<class Field> class IdentityLinearFunction : public LinearFunction<Field> {
-public:
-  void operator() (const Field &in, Field &out){
-    out = in;
-  };
-};
-
-
-/////////////////////////////////////////////////////////////
-// Base classes for Multishift solvers for operators
-/////////////////////////////////////////////////////////////
-template<class Field> class OperatorMultiFunction {
-public:
-  virtual void operator() (LinearOperatorBase<Field> &Linop, const Field &in, std::vector<Field> &out) = 0;
-};
-
-// FIXME : To think about
-
-// Chroma functionality list defining LinearOperator
-/*
-  virtual void operator() (T& chi, const T& psi, enum PlusMinus isign) const = 0;
-  virtual void operator() (T& chi, const T& psi, enum PlusMinus isign, Real epsilon) const
-  virtual const Subset& subset() const = 0;
-  virtual unsigned long nFlops() const { return 0; }
-  virtual void deriv(P& ds_u, const T& chi, const T& psi, enum PlusMinus isign) const
-  class UnprecLinearOperator : public DiffLinearOperator<T,P,Q>
-  const Subset& subset() const {return all;}
-  };
-*/
-
-////////////////////////////////////////////////////////////////////////////////////////////
-// Hermitian operator Linear function and operator function
-////////////////////////////////////////////////////////////////////////////////////////////
-template<class Field>
-class HermOpOperatorFunction : public OperatorFunction<Field> {
-  void operator() (LinearOperatorBase<Field> &Linop, const Field &in, Field &out) {
-    Linop.HermOp(in,out);
-  };
-};
-
-template<typename Field>
-class PlainHermOp : public LinearFunction<Field> {
-public:
-  LinearOperatorBase<Field> &_Linop;
-      
-  PlainHermOp(LinearOperatorBase<Field>& linop) : _Linop(linop) 
-  {}
-      
-  void operator()(const Field& in, Field& out) {
-    _Linop.HermOp(in,out);
-  }
-};
-
-template<typename Field>
-class FunctionHermOp : public LinearFunction<Field> {
-public:
-  OperatorFunction<Field>   & _poly;
-  LinearOperatorBase<Field> &_Linop;
-      
-  FunctionHermOp(OperatorFunction<Field> & poly,LinearOperatorBase<Field>& linop) 
-    : _poly(poly), _Linop(linop) {};
-      
-  void operator()(const Field& in, Field& out) {
-    _poly(_Linop,in,out);
-  }
-};
-
-template<class Field>
-class Polynomial : public OperatorFunction<Field> {
-private:
-  std::vector<RealD> Coeffs;
-public:
-  using OperatorFunction<Field>::operator();
-
-  Polynomial(std::vector<RealD> &_Coeffs) : Coeffs(_Coeffs) { };
-
-  // Implement the required interface
-  void operator() (LinearOperatorBase<Field> &Linop, const Field &in, Field &out) {
-
-    Field AtoN(in.Grid());
-    Field Mtmp(in.Grid());
-    AtoN = in;
-    out = AtoN*Coeffs[0];
-    for(int n=1;n<Coeffs.size();n++){
-      Mtmp = AtoN;
-      Linop.HermOp(Mtmp,AtoN);
-      out=out+AtoN*Coeffs[n];
-    }
-  };
-};
-
-NAMESPACE_END(Grid);

Grid/algorithms/SparseMatrix.h

@@ -1,79 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/algorithms/SparseMatrix.h
-
-    Copyright (C) 2015
-
-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 */
-#ifndef  GRID_ALGORITHM_SPARSE_MATRIX_H
-#define  GRID_ALGORITHM_SPARSE_MATRIX_H
-
-
-NAMESPACE_BEGIN(Grid);
-
-/////////////////////////////////////////////////////////////////////////////////////////////
-// Interface defining what I expect of a general sparse matrix, such as a Fermion action
-/////////////////////////////////////////////////////////////////////////////////////////////
-template<class Field> class SparseMatrixBase {
-public:
-  virtual GridBase *Grid(void) =0;
-  // Full checkerboar operations
-  virtual RealD M    (const Field &in, Field &out)=0;
-  virtual RealD Mdag (const Field &in, Field &out)=0;
-  virtual void  MdagM(const Field &in, Field &out,RealD &ni,RealD &no) {
-    Field tmp (in.Grid());
-    ni=M(in,tmp);
-    no=Mdag(tmp,out);
-  }
-  virtual  void Mdiag    (const Field &in, Field &out)=0;
-  virtual  void Mdir     (const Field &in, Field &out,int dir, int disp)=0;
-};
-
-/////////////////////////////////////////////////////////////////////////////////////////////
-// Interface augmented by a red black sparse matrix, such as a Fermion action
-/////////////////////////////////////////////////////////////////////////////////////////////
-template<class Field> class CheckerBoardedSparseMatrixBase : public SparseMatrixBase<Field> {
-public:
-  virtual GridBase *RedBlackGrid(void)=0;
-
-      //////////////////////////////////////////////////////////////////////
-      // Query the even even properties to make algorithmic decisions
-      //////////////////////////////////////////////////////////////////////
-      virtual RealD  Mass(void)        { return 0.0; };
-      virtual int    ConstEE(void)     { return 1; }; // Disable assumptions unless overridden
-      virtual int    isTrivialEE(void) { return 0; }; // by a derived class that knows better
-
-  // half checkerboard operaions
-  virtual  void Meooe    (const Field &in, Field &out)=0;
-  virtual  void Mooee    (const Field &in, Field &out)=0;
-  virtual  void MooeeInv (const Field &in, Field &out)=0;
-
-  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;
-
-};
-
-NAMESPACE_END(Grid);
-
-#endif

Grid/algorithms/approx/Chebyshev.h

@@ -1,379 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/algorithms/approx/Chebyshev.h
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Christoph Lehner <clehner@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_CHEBYSHEV_H
-#define GRID_CHEBYSHEV_H
-
-#include <Grid/algorithms/LinearOperator.h>
-
-NAMESPACE_BEGIN(Grid);
-
-struct ChebyParams : Serializable {
-  GRID_SERIALIZABLE_CLASS_MEMBERS(ChebyParams,
-				  RealD, alpha,  
-				  RealD, beta,   
-				  int, Npoly);
-};
-
-////////////////////////////////////////////////////////////////////////////////////////////
-// Generic Chebyshev approximations
-////////////////////////////////////////////////////////////////////////////////////////////
-template<class Field>
-class Chebyshev : public OperatorFunction<Field> {
-private:
-  using OperatorFunction<Field>::operator();
-
-  std::vector<RealD> Coeffs;
-  int order;
-  RealD hi;
-  RealD lo;
-
-public:
-  void csv(std::ostream &out){
-    RealD diff = hi-lo;
-    RealD delta = diff*1.0e-9;
-    for (RealD x=lo; x<hi; x+=delta) {
-      delta*=1.1;
-      RealD f = approx(x);
-      out<< x<<" "<<f<<std::endl;
-    }
-    return;
-  }
-
-  // Convenience for plotting the approximation
-  void   PlotApprox(std::ostream &out) {
-    out<<"Polynomial approx ["<<lo<<","<<hi<<"]"<<std::endl;
-    for(RealD x=lo;x<hi;x+=(hi-lo)/50.0){
-      out <<x<<"\t"<<approx(x)<<std::endl;
-    }
-  };
-
-  Chebyshev(){};
-  Chebyshev(ChebyParams p){ Init(p.alpha,p.beta,p.Npoly);};
-  Chebyshev(RealD _lo,RealD _hi,int _order, RealD (* func)(RealD) ) {Init(_lo,_hi,_order,func);};
-  Chebyshev(RealD _lo,RealD _hi,int _order) {Init(_lo,_hi,_order);};
-
-  ////////////////////////////////////////////////////////////////////////////////////////////////////
-  // c.f. numerical recipes "chebft"/"chebev". This is sec 5.8 "Chebyshev approximation".
-  ////////////////////////////////////////////////////////////////////////////////////////////////////
-  // CJ: the one we need for Lanczos
-  void Init(RealD _lo,RealD _hi,int _order)
-  {
-    lo=_lo;
-    hi=_hi;
-    order=_order;
-      
-    if(order < 2) exit(-1);
-    Coeffs.resize(order);
-    Coeffs.assign(0.,order);
-    Coeffs[order-1] = 1.;
-  };
-
-  void Init(RealD _lo,RealD _hi,int _order, RealD (* func)(RealD))
-  {
-    lo=_lo;
-    hi=_hi;
-    order=_order;
-      
-    if(order < 2) exit(-1);
-    Coeffs.resize(order);
-    for(int j=0;j<order;j++){
-      RealD s=0;
-      for(int k=0;k<order;k++){
-	RealD y=std::cos(M_PI*(k+0.5)/order);
-	RealD x=0.5*(y*(hi-lo)+(hi+lo));
-	RealD f=func(x);
-	s=s+f*std::cos( j*M_PI*(k+0.5)/order );
-      }
-      Coeffs[j] = s * 2.0/order;
-    }
-  };
-
-    
-  void JacksonSmooth(void){
-    RealD M=order;
-    RealD alpha = M_PI/(M+2);
-    RealD lmax = std::cos(alpha);
-    RealD sumUsq =0;
-    std::vector<RealD> U(M);
-    std::vector<RealD> a(M);
-    std::vector<RealD> g(M);
-    for(int n=0;n<=M;n++){
-      U[n] = std::sin((n+1)*std::acos(lmax))/std::sin(std::acos(lmax));
-      sumUsq += U[n]*U[n];
-    }      
-    sumUsq = std::sqrt(sumUsq);
-
-    for(int i=1;i<=M;i++){
-      a[i] = U[i]/sumUsq;
-    }
-    g[0] = 1.0;
-    for(int m=1;m<=M;m++){
-      g[m] = 0;
-      for(int i=0;i<=M-m;i++){
-	g[m]+= a[i]*a[m+i];
-      }
-    }
-    for(int m=1;m<=M;m++){
-      Coeffs[m]*=g[m];
-    }
-  }
-  RealD approx(RealD x) // Convenience for plotting the approximation
-  {
-    RealD Tn;
-    RealD Tnm;
-    RealD Tnp;
-      
-    RealD y=( x-0.5*(hi+lo))/(0.5*(hi-lo));
-      
-    RealD T0=1;
-    RealD T1=y;
-      
-    RealD sum;
-    sum = 0.5*Coeffs[0]*T0;
-    sum+= Coeffs[1]*T1;
-      
-    Tn =T1;
-    Tnm=T0;
-    for(int i=2;i<order;i++){
-      Tnp=2*y*Tn-Tnm;
-      Tnm=Tn;
-      Tn =Tnp;
-      sum+= Tn*Coeffs[i];
-    }
-    return sum;
-  };
-
-  RealD approxD(RealD x)
-  {
-    RealD Un;
-    RealD Unm;
-    RealD Unp;
-      
-    RealD y=( x-0.5*(hi+lo))/(0.5*(hi-lo));
-      
-    RealD U0=1;
-    RealD U1=2*y;
-      
-    RealD sum;
-    sum = Coeffs[1]*U0;
-    sum+= Coeffs[2]*U1*2.0;
-      
-    Un =U1;
-    Unm=U0;
-    for(int i=2;i<order-1;i++){
-      Unp=2*y*Un-Unm;
-      Unm=Un;
-      Un =Unp;
-      sum+= Un*Coeffs[i+1]*(i+1.0);
-    }
-    return sum/(0.5*(hi-lo));
-  };
-    
-  RealD approxInv(RealD z, RealD x0, int maxiter, RealD resid) {
-    RealD x = x0;
-    RealD eps;
-      
-    int i;
-    for (i=0;i<maxiter;i++) {
-      eps = approx(x) - z;
-      if (fabs(eps / z) < resid)
-	return x;
-      x = x - eps / approxD(x);
-    }
-      
-    return std::numeric_limits<double>::quiet_NaN();
-  }
-    
-  // Implement the required interface
-  void operator() (LinearOperatorBase<Field> &Linop, const Field &in, Field &out) {
-
-    GridBase *grid=in.Grid();
-
-    // std::cout << "Chevyshef(): in.Grid()="<<in.Grid()<<std::endl;
-    //std::cout <<" Linop.Grid()="<<Linop.Grid()<<"Linop.RedBlackGrid()="<<Linop.RedBlackGrid()<<std::endl;
-
-    int vol=grid->gSites();
-
-    Field T0(grid); T0 = in;  
-    Field T1(grid); 
-    Field T2(grid);
-    Field y(grid);
-      
-    Field *Tnm = &T0;
-    Field *Tn  = &T1;
-    Field *Tnp = &T2;
-
-    // Tn=T1 = (xscale M + mscale)in
-    RealD xscale = 2.0/(hi-lo);
-    RealD mscale = -(hi+lo)/(hi-lo);
-    Linop.HermOp(T0,y);
-    T1=y*xscale+in*mscale;
-
-    // sum = .5 c[0] T0 + c[1] T1
-    out = (0.5*Coeffs[0])*T0 + Coeffs[1]*T1;
-    for(int n=2;n<order;n++){
-	
-      Linop.HermOp(*Tn,y);
-
-      y=xscale*y+mscale*(*Tn);
-
-      *Tnp=2.0*y-(*Tnm);
-
-      out=out+Coeffs[n]* (*Tnp);
-
-      // Cycle pointers to avoid copies
-      Field *swizzle = Tnm;
-      Tnm    =Tn;
-      Tn     =Tnp;
-      Tnp    =swizzle;
-	  
-    }
-  }
-};
-
-
-template<class Field>
-class ChebyshevLanczos : public Chebyshev<Field> {
-private:
-  std::vector<RealD> Coeffs;
-  int order;
-  RealD alpha;
-  RealD beta;
-  RealD mu;
-
-public:
-  ChebyshevLanczos(RealD _alpha,RealD _beta,RealD _mu,int _order) :
-    alpha(_alpha),
-    beta(_beta),
-    mu(_mu)
-  {
-    order=_order;
-    Coeffs.resize(order);
-    for(int i=0;i<_order;i++){
-      Coeffs[i] = 0.0;
-    }
-    Coeffs[order-1]=1.0;
-  };
-
-  void csv(std::ostream &out){
-    for (RealD x=-1.2*alpha; x<1.2*alpha; x+=(2.0*alpha)/10000) {
-      RealD f = approx(x);
-      out<< x<<" "<<f<<std::endl;
-    }
-    return;
-  }
-
-  RealD approx(RealD xx) // Convenience for plotting the approximation
-  {
-    RealD Tn;
-    RealD Tnm;
-    RealD Tnp;
-    Real aa = alpha * alpha;
-    Real bb = beta  *  beta;
-      
-    RealD x = ( 2.0 * (xx-mu)*(xx-mu) - (aa+bb) ) / (aa-bb);
-
-    RealD y= x;
-      
-    RealD T0=1;
-    RealD T1=y;
-      
-    RealD sum;
-    sum = 0.5*Coeffs[0]*T0;
-    sum+= Coeffs[1]*T1;
-      
-    Tn =T1;
-    Tnm=T0;
-    for(int i=2;i<order;i++){
-      Tnp=2*y*Tn-Tnm;
-      Tnm=Tn;
-      Tn =Tnp;
-      sum+= Tn*Coeffs[i];
-    }
-    return sum;
-  };
-
-  // shift_Multiply in Rudy's code
-  void AminusMuSq(LinearOperatorBase<Field> &Linop, const Field &in, Field &out) 
-  {
-    GridBase *grid=in.Grid();
-    Field tmp(grid);
-
-    RealD aa= alpha*alpha;
-    RealD bb= beta * beta;
-
-    Linop.HermOp(in,out);
-    out = out - mu*in;
-
-    Linop.HermOp(out,tmp);
-    tmp = tmp - mu * out;
-
-    out = (2.0/ (aa-bb) ) * tmp -  ((aa+bb)/(aa-bb))*in;
-  };
-  // Implement the required interface
-  void operator() (LinearOperatorBase<Field> &Linop, const Field &in, Field &out) {
-
-    GridBase *grid=in.Grid();
-
-    int vol=grid->gSites();
-
-    Field T0(grid); T0 = in;  
-    Field T1(grid); 
-    Field T2(grid);
-    Field  y(grid);
-      
-    Field *Tnm = &T0;
-    Field *Tn  = &T1;
-    Field *Tnp = &T2;
-
-    // Tn=T1 = (xscale M )*in
-    AminusMuSq(Linop,T0,T1);
-
-    // sum = .5 c[0] T0 + c[1] T1
-    out = (0.5*Coeffs[0])*T0 + Coeffs[1]*T1;
-    for(int n=2;n<order;n++){
-	
-      AminusMuSq(Linop,*Tn,y);
-
-      *Tnp=2.0*y-(*Tnm);
-
-      out=out+Coeffs[n]* (*Tnp);
-
-      // Cycle pointers to avoid copies
-      Field *swizzle = Tnm;
-      Tnm    =Tn;
-      Tn     =Tnp;
-      Tnp    =swizzle;
-	  
-    }
-  }
-};
-NAMESPACE_END(Grid);
-#endif

Grid/algorithms/approx/Forecast.h

@@ -1,152 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/algorithms/approx/Forecast.h
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: David Murphy <dmurphy@phys.columbia.edu>
-
-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 INCLUDED_FORECAST_H
-#define INCLUDED_FORECAST_H
-
-NAMESPACE_BEGIN(Grid);
-
-// Abstract base class.
-// Takes a matrix (Mat), a source (phi), and a vector of Fields (chi)
-// and returns a forecasted solution to the system D*psi = phi (psi).
-template<class Matrix, class Field>
-class Forecast
-{
-public:
-  virtual Field operator()(Matrix &Mat, const Field& phi, const std::vector<Field>& chi) = 0;
-};
-
-// Implementation of Brower et al.'s chronological inverter (arXiv:hep-lat/9509012),
-// used to forecast solutions across poles of the EOFA heatbath.
-//
-// Modified from CPS (cps_pp/src/util/dirac_op/d_op_base/comsrc/minresext.C)
-template<class Matrix, class Field>
-class ChronoForecast : public Forecast<Matrix,Field>
-{
-public:
-  Field operator()(Matrix &Mat, const Field& phi, const std::vector<Field>& prev_solns)
-  {
-    int degree = prev_solns.size();
-    Field chi(phi); // forecasted solution
-
-    // Trivial cases
-    if(degree == 0){ chi = Zero(); return chi; }
-    else if(degree == 1){ return prev_solns[0]; }
-
-    //    RealD dot;
-    ComplexD xp;
-    Field r(phi); // residual
-    Field Mv(phi);
-    std::vector<Field> v(prev_solns); // orthonormalized previous solutions
-    std::vector<Field> MdagMv(degree,phi);
-
-    // Array to hold the matrix elements
-    std::vector<std::vector<ComplexD>> G(degree, std::vector<ComplexD>(degree));
-
-    // Solution and source vectors
-    std::vector<ComplexD> a(degree);
-    std::vector<ComplexD> b(degree);
-
-    // Orthonormalize the vector basis
-    for(int i=0; i<degree; i++){
-      v[i] *= 1.0/std::sqrt(norm2(v[i]));
-      for(int j=i+1; j<degree; j++){ v[j] -= innerProduct(v[i],v[j]) * v[i]; }
-    }
-
-    // Perform sparse matrix multiplication and construct rhs
-    for(int i=0; i<degree; i++){
-      b[i] = innerProduct(v[i],phi);
-      Mat.M(v[i],Mv);
-      Mat.Mdag(Mv,MdagMv[i]);
-      G[i][i] = innerProduct(v[i],MdagMv[i]);
-    }
-
-    // Construct the matrix
-    for(int j=0; j<degree; j++){
-      for(int k=j+1; k<degree; k++){
-	G[j][k] = innerProduct(v[j],MdagMv[k]);
-	G[k][j] = conjugate(G[j][k]);
-      }}
-
-    // Gauss-Jordan elimination with partial pivoting
-    for(int i=0; i<degree; i++){
-
-      // Perform partial pivoting
-      int k = i;
-      for(int j=i+1; j<degree; j++){ if(abs(G[j][j]) > abs(G[k][k])){ k = j; } }
-      if(k != i){
-	xp = b[k];
-	b[k] = b[i];
-	b[i] = xp;
-	for(int j=0; j<degree; j++){
-	  xp = G[k][j];
-	  G[k][j] = G[i][j];
-	  G[i][j] = xp;
-	}
-      }
-
-      // Convert matrix to upper triangular form
-      for(int j=i+1; j<degree; j++){
-	xp = G[j][i]/G[i][i];
-	b[j] -= xp * b[i];
-	for(int k=0; k<degree; k++){ G[j][k] -= xp*G[i][k]; }
-      }
-    }
-
-    // Use Gaussian elimination to solve equations and calculate initial guess
-    chi = Zero();
-    r = phi;
-    for(int i=degree-1; i>=0; i--){
-      a[i] = 0.0;
-      for(int j=i+1; j<degree; j++){ a[i] += G[i][j] * a[j]; }
-      a[i] = (b[i]-a[i])/G[i][i];
-      chi += a[i]*v[i];
-      r -= a[i]*MdagMv[i];
-    }
-
-    RealD true_r(0.0);
-    ComplexD tmp;
-    for(int i=0; i<degree; i++){
-      tmp = -b[i];
-      for(int j=0; j<degree; j++){ tmp += G[i][j]*a[j]; }
-      tmp = conjugate(tmp)*tmp;
-      true_r += std::sqrt(tmp.real());
-    }
-
-    RealD error = std::sqrt(norm2(r)/norm2(phi));
-    std::cout << GridLogMessage << "ChronoForecast: |res|/|src| = " << error << std::endl;
-
-    return chi;
-  };
-};
-
-NAMESPACE_END(Grid);
-
-#endif

Grid/algorithms/iterative/CommunicationAvoidingGeneralisedMinimalResidual.h

@@ -1,248 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/algorithms/iterative/CommunicationAvoidingGeneralisedMinimalResidual.h
-
-Copyright (C) 2015
-
-Author: Daniel Richtmann <daniel.richtmann@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 */
-#ifndef GRID_COMMUNICATION_AVOIDING_GENERALISED_MINIMAL_RESIDUAL_H
-#define GRID_COMMUNICATION_AVOIDING_GENERALISED_MINIMAL_RESIDUAL_H
-
-namespace Grid {
-
-template<class Field>
-class CommunicationAvoidingGeneralisedMinimalResidual : public OperatorFunction<Field> {
- public:
-  using OperatorFunction<Field>::operator();
-
-  bool ErrorOnNoConverge; // Throw an assert when CAGMRES fails to converge,
-                          // defaults to true
-
-  RealD   Tolerance;
-
-  Integer MaxIterations;
-  Integer RestartLength;
-  Integer MaxNumberOfRestarts;
-  Integer IterationCount; // Number of iterations the CAGMRES took to finish,
-                          // filled in upon completion
-
-  GridStopWatch MatrixTimer;
-  GridStopWatch LinalgTimer;
-  GridStopWatch QrTimer;
-  GridStopWatch CompSolutionTimer;
-
-  Eigen::MatrixXcd H;
-
-  std::vector<ComplexD> y;
-  std::vector<ComplexD> gamma;
-  std::vector<ComplexD> c;
-  std::vector<ComplexD> s;
-
-  CommunicationAvoidingGeneralisedMinimalResidual(RealD   tol,
-                                                  Integer maxit,
-                                                  Integer restart_length,
-                                                  bool    err_on_no_conv = true)
-      : Tolerance(tol)
-      , MaxIterations(maxit)
-      , RestartLength(restart_length)
-      , MaxNumberOfRestarts(MaxIterations/RestartLength + ((MaxIterations%RestartLength == 0) ? 0 : 1))
-      , ErrorOnNoConverge(err_on_no_conv)
-      , H(Eigen::MatrixXcd::Zero(RestartLength, RestartLength + 1)) // sizes taken from DD-αAMG code base
-      , y(RestartLength + 1, 0.)
-      , gamma(RestartLength + 1, 0.)
-      , c(RestartLength + 1, 0.)
-      , s(RestartLength + 1, 0.) {};
-
-  void operator()(LinearOperatorBase<Field> &LinOp, const Field &src, Field &psi) {
-
-    std::cout << GridLogWarning << "This algorithm currently doesn't differ from regular GMRES" << std::endl;
-
-    psi.Checkerboard() = src.Checkerboard();
-    conformable(psi, src);
-
-    RealD guess = norm2(psi);
-    assert(std::isnan(guess) == 0);
-
-    RealD cp;
-    RealD ssq = norm2(src);
-    RealD rsq = Tolerance * Tolerance * ssq;
-
-    Field r(src.Grid());
-
-    std::cout << std::setprecision(4) << std::scientific;
-    std::cout << GridLogIterative << "CommunicationAvoidingGeneralisedMinimalResidual: guess " << guess << std::endl;
-    std::cout << GridLogIterative << "CommunicationAvoidingGeneralisedMinimalResidual:   src " << ssq   << std::endl;
-
-    MatrixTimer.Reset();
-    LinalgTimer.Reset();
-    QrTimer.Reset();
-    CompSolutionTimer.Reset();
-
-    GridStopWatch SolverTimer;
-    SolverTimer.Start();
-
-    IterationCount = 0;
-
-    for (int k=0; k<MaxNumberOfRestarts; k++) {
-
-      cp = outerLoopBody(LinOp, src, psi, rsq);
-
-      // Stopping condition
-      if (cp <= rsq) {
-
-        SolverTimer.Stop();
-
-        LinOp.Op(psi,r);
-        axpy(r,-1.0,src,r);
-
-        RealD srcnorm       = sqrt(ssq);
-        RealD resnorm       = sqrt(norm2(r));
-        RealD true_residual = resnorm / srcnorm;
-
-        std::cout << GridLogMessage        << "CommunicationAvoidingGeneralisedMinimalResidual: Converged on iteration " << IterationCount
-                  << " computed residual " << sqrt(cp / ssq)
-                  << " true residual "     << true_residual
-                  << " target "            << Tolerance << std::endl;
-
-        std::cout << GridLogMessage << "CAGMRES Time elapsed: Total   " <<       SolverTimer.Elapsed() << std::endl;
-        std::cout << GridLogMessage << "CAGMRES Time elapsed: Matrix  " <<       MatrixTimer.Elapsed() << std::endl;
-        std::cout << GridLogMessage << "CAGMRES Time elapsed: Linalg  " <<       LinalgTimer.Elapsed() << std::endl;
-        std::cout << GridLogMessage << "CAGMRES Time elapsed: QR      " <<           QrTimer.Elapsed() << std::endl;
-        std::cout << GridLogMessage << "CAGMRES Time elapsed: CompSol " << CompSolutionTimer.Elapsed() << std::endl;
-        return;
-      }
-    }
-
-    std::cout << GridLogMessage << "CommunicationAvoidingGeneralisedMinimalResidual did NOT converge" << std::endl;
-
-    if (ErrorOnNoConverge)
-      assert(0);
-  }
-
-  RealD outerLoopBody(LinearOperatorBase<Field> &LinOp, const Field &src, Field &psi, RealD rsq) {
-
-    RealD cp = 0;
-
-    Field w(src.Grid());
-    Field r(src.Grid());
-
-    // this should probably be made a class member so that it is only allocated once, not in every restart
-    std::vector<Field> v(RestartLength + 1, src.Grid()); for (auto &elem : v) elem = Zero();
-
-    MatrixTimer.Start();
-    LinOp.Op(psi, w);
-    MatrixTimer.Stop();
-
-    LinalgTimer.Start();
-    r = src - w;
-
-    gamma[0] = sqrt(norm2(r));
-
-    ComplexD scale = 1.0/gamma[0];
-    v[0] = scale * r;
-
-    LinalgTimer.Stop();
-
-    for (int i=0; i<RestartLength; i++) {
-
-      IterationCount++;
-
-      arnoldiStep(LinOp, v, w, i);
-
-      qrUpdate(i);
-
-      cp = norm(gamma[i+1]);
-
-      std::cout << GridLogIterative << "CommunicationAvoidingGeneralisedMinimalResidual: Iteration " << IterationCount
-                << " residual " << cp << " target " << rsq << std::endl;
-
-      if ((i == RestartLength - 1) || (IterationCount == MaxIterations) || (cp <= rsq)) {
-
-        computeSolution(v, psi, i);
-
-        return cp;
-      }
-    }
-
-    assert(0); // Never reached
-    return cp;
-  }
-
-  void arnoldiStep(LinearOperatorBase<Field> &LinOp, std::vector<Field> &v, Field &w, int iter) {
-
-    MatrixTimer.Start();
-    LinOp.Op(v[iter], w);
-    MatrixTimer.Stop();
-
-    LinalgTimer.Start();
-    for (int i = 0; i <= iter; ++i) {
-      H(iter, i) = innerProduct(v[i], w);
-      w = w - ComplexD(H(iter, i)) * v[i];
-    }
-
-    H(iter, iter + 1) = sqrt(norm2(w));
-    v[iter + 1] = ComplexD(1. / H(iter, iter + 1)) * w;
-    LinalgTimer.Stop();
-  }
-
-  void qrUpdate(int iter) {
-
-    QrTimer.Start();
-    for (int i = 0; i < iter ; ++i) {
-      auto tmp       = -s[i] * ComplexD(H(iter, i)) + c[i] * ComplexD(H(iter, i + 1));
-      H(iter, i)     = conjugate(c[i]) * ComplexD(H(iter, i)) + conjugate(s[i]) * ComplexD(H(iter, i + 1));
-      H(iter, i + 1) = tmp;
-    }
-
-    // Compute new Givens Rotation
-    auto nu     = sqrt(std::norm(H(iter, iter)) + std::norm(H(iter, iter + 1)));
-    c[iter]     = H(iter, iter) / nu;
-    s[iter]     = H(iter, iter + 1) / nu;
-
-    // Apply new Givens rotation
-    H(iter, iter)     = nu;
-    H(iter, iter + 1) = 0.;
-
-    gamma[iter + 1] = -s[iter] * gamma[iter];
-    gamma[iter]     = conjugate(c[iter]) * gamma[iter];
-    QrTimer.Stop();
-  }
-
-  void computeSolution(std::vector<Field> const &v, Field &psi, int iter) {
-
-    CompSolutionTimer.Start();
-    for (int i = iter; i >= 0; i--) {
-      y[i] = gamma[i];
-      for (int k = i + 1; k <= iter; k++)
-        y[i] = y[i] - ComplexD(H(k, i)) * y[k];
-      y[i] = y[i] / ComplexD(H(i, i));
-    }
-
-    for (int i = 0; i <= iter; i++)
-      psi = psi + v[i] * y[i];
-    CompSolutionTimer.Stop();
-  }
-};
-}
-#endif

Grid/algorithms/iterative/ConjugateGradientMixedPrec.h

@@ -1,161 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/algorithms/iterative/ConjugateGradientMixedPrec.h
-
-    Copyright (C) 2015
-
-Author: Christopher Kelly <ckelly@phys.columbia.edu>
-
-    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_MIXED_PREC_H
-#define GRID_CONJUGATE_GRADIENT_MIXED_PREC_H
-
-NAMESPACE_BEGIN(Grid);
-
-  //Mixed precision restarted defect correction CG
-  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 MixedPrecisionConjugateGradient : public LinearFunction<FieldD> {
-  public:                                                
-    RealD   Tolerance;
-    RealD   InnerTolerance; //Initial tolerance for inner CG. Defaults to Tolerance but can be changed
-    Integer MaxInnerIterations;
-    Integer MaxOuterIterations;
-    GridBase* SinglePrecGrid; //Grid for single-precision fields
-    RealD OuterLoopNormMult; //Stop the outer loop and move to a final double prec solve when the residual is OuterLoopNormMult * Tolerance
-    LinearOperatorBase<FieldF> &Linop_f;
-    LinearOperatorBase<FieldD> &Linop_d;
-
-    Integer TotalInnerIterations; //Number of inner CG iterations
-    Integer TotalOuterIterations; //Number of restarts
-    Integer TotalFinalStepIterations; //Number of CG iterations in final patch-up step
-
-    //Option to speed up *inner single precision* solves using a LinearFunction that produces a guess
-    LinearFunction<FieldF> *guesser;
-    
-    MixedPrecisionConjugateGradient(RealD tol, 
-				    Integer maxinnerit, 
-				    Integer maxouterit, 
-				    GridBase* _sp_grid, 
-				    LinearOperatorBase<FieldF> &_Linop_f, 
-				    LinearOperatorBase<FieldD> &_Linop_d) :
-      Linop_f(_Linop_f), Linop_d(_Linop_d),
-      Tolerance(tol), InnerTolerance(tol), MaxInnerIterations(maxinnerit), MaxOuterIterations(maxouterit), SinglePrecGrid(_sp_grid),
-      OuterLoopNormMult(100.), guesser(NULL){ };
-
-    void useGuesser(LinearFunction<FieldF> &g){
-      guesser = &g;
-    }
-  
-  void operator() (const FieldD &src_d_in, FieldD &sol_d){
-    TotalInnerIterations = 0;
-	
-    GridStopWatch TotalTimer;
-    TotalTimer.Start();
-    
-    int cb = src_d_in.Checkerboard();
-    sol_d.Checkerboard() = cb;
-    
-    RealD src_norm = norm2(src_d_in);
-    RealD stop = src_norm * Tolerance*Tolerance;
-
-    GridBase* DoublePrecGrid = src_d_in.Grid();
-    FieldD tmp_d(DoublePrecGrid);
-    tmp_d.Checkerboard() = cb;
-    
-    FieldD tmp2_d(DoublePrecGrid);
-    tmp2_d.Checkerboard() = cb;
-    
-    FieldD src_d(DoublePrecGrid);
-    src_d = src_d_in; //source for next inner iteration, computed from residual during operation
-    
-    RealD inner_tol = InnerTolerance;
-    
-    FieldF src_f(SinglePrecGrid);
-    src_f.Checkerboard() = cb;
-    
-    FieldF sol_f(SinglePrecGrid);
-    sol_f.Checkerboard() = cb;
-    
-    ConjugateGradient<FieldF> CG_f(inner_tol, MaxInnerIterations);
-    CG_f.ErrorOnNoConverge = false;
-
-    GridStopWatch InnerCGtimer;
-
-    GridStopWatch PrecChangeTimer;
-    
-    Integer &outer_iter = TotalOuterIterations; //so it will be equal to the final iteration count
-      
-    for(outer_iter = 0; outer_iter < MaxOuterIterations; outer_iter++){
-      //Compute double precision rsd and also new RHS vector.
-      Linop_d.HermOp(sol_d, tmp_d);
-      RealD norm = axpy_norm(src_d, -1., tmp_d, src_d_in); //src_d is residual vector
-      
-      std::cout<<GridLogMessage<<"MixedPrecisionConjugateGradient: Outer iteration " <<outer_iter<<" residual "<< norm<< " target "<< stop<<std::endl;
-
-      if(norm < OuterLoopNormMult * stop){
-	std::cout<<GridLogMessage<<"MixedPrecisionConjugateGradient: Outer iteration converged on iteration " <<outer_iter <<std::endl;
-	break;
-      }
-      while(norm * inner_tol * inner_tol < stop) inner_tol *= 2;  // inner_tol = sqrt(stop/norm) ??
-
-      PrecChangeTimer.Start();
-      precisionChange(src_f, src_d);
-      PrecChangeTimer.Stop();
-      
-      sol_f = Zero();
-
-      //Optionally improve inner solver guess (eg using known eigenvectors)
-      if(guesser != NULL)
-	(*guesser)(src_f, sol_f);
-
-      //Inner CG
-      CG_f.Tolerance = inner_tol;
-      InnerCGtimer.Start();
-      CG_f(Linop_f, src_f, sol_f);
-      InnerCGtimer.Stop();
-      TotalInnerIterations += CG_f.IterationsToComplete;
-      
-      //Convert sol back to double and add to double prec solution
-      PrecChangeTimer.Start();
-      precisionChange(tmp_d, sol_f);
-      PrecChangeTimer.Stop();
-      
-      axpy(sol_d, 1.0, tmp_d, sol_d);
-    }
-    
-    //Final trial CG
-    std::cout<<GridLogMessage<<"MixedPrecisionConjugateGradient: Starting final patch-up double-precision solve"<<std::endl;
-    
-    ConjugateGradient<FieldD> CG_d(Tolerance, MaxInnerIterations);
-    CG_d(Linop_d, src_d_in, sol_d);
-    TotalFinalStepIterations = CG_d.IterationsToComplete;
-
-    TotalTimer.Stop();
-    std::cout<<GridLogMessage<<"MixedPrecisionConjugateGradient: Inner CG iterations " << TotalInnerIterations << " Restarts " << TotalOuterIterations << " Final CG iterations " << TotalFinalStepIterations << std::endl;
-    std::cout<<GridLogMessage<<"MixedPrecisionConjugateGradient: Total time " << TotalTimer.Elapsed() << " Precision change " << PrecChangeTimer.Elapsed() << " Inner CG total " << InnerCGtimer.Elapsed() << std::endl;
-  }
-};
-
-NAMESPACE_END(Grid);
-
-#endif

Grid/algorithms/iterative/ConjugateGradientMultiShift.h

@@ -1,325 +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>
-
-    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_MULTI_SHIFT_GRADIENT_H
-#define GRID_CONJUGATE_MULTI_SHIFT_GRADIENT_H
-
-NAMESPACE_BEGIN(Grid);
-
-/////////////////////////////////////////////////////////////
-// Base classes for iterative processes based on operators
-// single input vec, single output vec.
-/////////////////////////////////////////////////////////////
-
-template<class Field> 
-class ConjugateGradientMultiShift : public OperatorMultiFunction<Field>,
-				    public OperatorFunction<Field>
-{
-public:                                                
-
-  using OperatorFunction<Field>::operator();
-
-  RealD   Tolerance;
-  Integer MaxIterations;
-    Integer IterationsToComplete; //Number of iterations the CG took to finish. Filled in upon completion
-  int verbose;
-  MultiShiftFunction shifts;
-
-  ConjugateGradientMultiShift(Integer maxit,MultiShiftFunction &_shifts) : 
-    MaxIterations(maxit),
-    shifts(_shifts)
-  { 
-    verbose=1;
-  }
-
-  void operator() (LinearOperatorBase<Field> &Linop, const Field &src, Field &psi)
-  {
-    GridBase *grid = src.Grid();
-    int nshift = shifts.order;
-    std::vector<Field> results(nshift,grid);
-    (*this)(Linop,src,results,psi);
-  }
-  void operator() (LinearOperatorBase<Field> &Linop, const Field &src, std::vector<Field> &results, Field &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<Field> &Linop, const Field &src, std::vector<Field> &psi)
-  {
-  
-    GridBase *grid = src.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);
-    std::vector<Field>   ps(nshift,grid);// Search directions
-
-    assert(psi.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
-    Field r(grid);
-    Field p(grid);
-    Field tmp(grid);
-    Field mmp(grid);
-  
-    // 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);
-    for(int s=0;s<nshift;s++){
-      rsq[s] = cp * mresidual[s] * mresidual[s];
-      std::cout<<GridLogMessage<<"ConjugateGradientMultiShift: shift "<<s
-	       <<" target resid "<<rsq[s]<<std::endl;
-      ps[s] = src;
-    }
-    // r and p for primary
-    r=src;
-    p=src;
-  
-    //MdagM+m[0]
-    Linop.HermOpAndNorm(p,mmp,d,qq);
-    axpy(mmp,mass[0],p,mmp);
-    RealD rn = norm2(p);
-    d += rn*mass[0];
-  
-    // have verified that inner product of 
-    // p and mmp is equal to d after this since
-    // the d computation is tricky
-    //  qq = real(innerProduct(p,mmp));
-    //  std::cout<<GridLogMessage << "debug equal ?  qq "<<qq<<" d "<< d<<std::endl;
-  
-    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,b,mmp,r);
-  
-    for(int s=0;s<nshift;s++) {
-      axpby(psi[s],0.,-bs[s]*alpha[s],src,src);
-    }
-  
-  ///////////////////////////////////////
-  // Timers
-  ///////////////////////////////////////
-  GridStopWatch AXPYTimer;
-  GridStopWatch ShiftTimer;
-  GridStopWatch QRTimer;
-  GridStopWatch MatrixTimer;
-  GridStopWatch SolverTimer;
-  SolverTimer.Start();
-  
-    // Iteration loop
-    int k;
-  
-    for (k=1;k<=MaxIterations;k++){
-    
-      a = c /cp;
-    AXPYTimer.Start();
-      axpy(p,a,p,r);
-    AXPYTimer.Stop();
-    
-      // Note to self - direction ps is iterated seperately
-      // for each shift. Does not appear to have any scope
-      // for avoiding linear algebra in "single" case.
-      // 
-      // However SAME r is used. Could load "r" and update
-      // ALL ps[s]. 2/3 Bandwidth saving
-      // New Kernel: Load r, vector of coeffs, vector of pointers ps
-    AXPYTimer.Start();
-      for(int s=0;s<nshift;s++){
-	if ( ! converged[s] ) { 
-	  if (s==0){
-	    axpy(ps[s],a,ps[s],r);
-	  } else{
-	    RealD as =a *z[s][iz]*bs[s] /(z[s][1-iz]*b);
-	    axpby(ps[s],z[s][iz],as,r,ps[s]);
-	  }
-	}
-      }
-    AXPYTimer.Stop();
-    
-      cp=c;
-    MatrixTimer.Start();  
-    //Linop.HermOpAndNorm(p,mmp,d,qq); // d is used
-    // The below is faster on KNL
-    Linop.HermOp(p,mmp); 
-    d=real(innerProduct(p,mmp));
-    
-    MatrixTimer.Stop();  
-
-    AXPYTimer.Start();
-      axpy(mmp,mass[0],p,mmp);
-    AXPYTimer.Stop();
-      RealD rn = norm2(p);
-      d += rn*mass[0];
-    
-      bp=b;
-      b=-cp/d;
-    
-    AXPYTimer.Start();
-      c=axpy_norm(r,b,mmp,r);
-    AXPYTimer.Stop();
-
-      // 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();
-    
-      for(int s=0;s<nshift;s++){
-	int ss = s;
-	// Scope for optimisation here in case of "single".
-	// Could load psi[0] and pull all ps[s] in.
-	//      if ( single ) ss=primary;
-	// Bandwith saving in single case is Ls * 3 -> 2+Ls, so ~ 3x saving
-	// Pipelined CG gain:
-	//
-	// New Kernel: Load r, vector of coeffs, vector of pointers ps
-	// New Kernel: Load psi[0], vector of coeffs, vector of pointers ps
-	// If can predict the coefficient bs then we can fuse these and avoid write reread cyce
-	//  on ps[s].
-	// Before:  3 x npole  + 3 x npole
-	// After :  2 x npole (ps[s])        => 3x speed up of multishift CG.
-      
-	if( (!converged[s]) ) { 
-	  axpy(psi[ss],-bs[s]*alpha[s],ps[s],psi[ss]);
-	}
-      }
-    
-      // Convergence checks
-      int all_converged = 1;
-      for(int s=0;s<nshift;s++){
-      
-	if ( (!converged[s]) ){
-	
-	  RealD css  = c * z[s][iz]* z[s][iz];
-	
-	  if(css<rsq[s]){
-	    if ( ! converged[s] )
-	      std::cout<<GridLogMessage<<"ConjugateGradientMultiShift k="<<k<<" Shift "<<s<<" has converged"<<std::endl;
-	    converged[s]=1;
-	  } else {
-	    all_converged=0;
-	  }
-
-	}
-      }
-    
-      if ( all_converged ){
-
-    SolverTimer.Stop();
-
-
-	std::cout<<GridLogMessage<< "CGMultiShift: All shifts have converged iteration "<<k<<std::endl;
-	std::cout<<GridLogMessage<< "CGMultiShift: Checking solutions"<<std::endl;
-      
-	// Check answers 
-	for(int s=0; s < nshift; s++) { 
-	  Linop.HermOpAndNorm(psi[s],mmp,d,qq);
-	  axpy(tmp,mass[s],psi[s],mmp);
-	  axpy(r,-alpha[s],src,tmp);
-	  RealD rn = norm2(r);
-	  RealD cn = norm2(src);
-	  std::cout<<GridLogMessage<<"CGMultiShift: shift["<<s<<"] true residual "<<std::sqrt(rn/cn)<<std::endl;
-	}
-
-      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 << "\tMarix    " << MatrixTimer.Elapsed()     <<std::endl;
-      std::cout << GridLogMessage << "\tShift    " << ShiftTimer.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

Grid/algorithms/iterative/ConjugateGradientReliableUpdate.h

@@ -1,258 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/algorithms/iterative/ConjugateGradientReliableUpdate.h
-
-    Copyright (C) 2015
-
-Author: Christopher Kelly <ckelly@phys.columbia.edu>
-
-    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_RELIABLE_UPDATE_H
-#define GRID_CONJUGATE_GRADIENT_RELIABLE_UPDATE_H
-
-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 ConjugateGradientReliableUpdate : public LinearFunction<FieldD> {
-public:
-  bool ErrorOnNoConverge;  // throw an assert when the CG fails to converge.
-  // Defaults true.
-  RealD Tolerance;
-  Integer MaxIterations;
-  Integer IterationsToComplete; //Number of iterations the CG took to finish. Filled in upon completion
-  Integer ReliableUpdatesPerformed;
-
-  bool DoFinalCleanup; //Final DP cleanup, defaults to true
-  Integer IterationsToCleanup; //Final DP cleanup step iterations
-    
-  LinearOperatorBase<FieldF> &Linop_f;
-  LinearOperatorBase<FieldD> &Linop_d;
-  GridBase* SinglePrecGrid;
-  RealD Delta; //reliable update parameter
-
-  //Optional ability to switch to a different linear operator once the tolerance reaches a certain point. Useful for single/half -> single/single
-  LinearOperatorBase<FieldF> *Linop_fallback;
-  RealD fallback_transition_tol;
-
-    
-  ConjugateGradientReliableUpdate(RealD tol, Integer maxit, RealD _delta, GridBase* _sp_grid, LinearOperatorBase<FieldF> &_Linop_f, LinearOperatorBase<FieldD> &_Linop_d, bool err_on_no_conv = true)
-    : Tolerance(tol),
-      MaxIterations(maxit),
-      Delta(_delta),
-      Linop_f(_Linop_f),
-      Linop_d(_Linop_d),
-      SinglePrecGrid(_sp_grid),
-      ErrorOnNoConverge(err_on_no_conv),
-      DoFinalCleanup(true),
-      Linop_fallback(NULL)
-  {};
-
-  void setFallbackLinop(LinearOperatorBase<FieldF> &_Linop_fallback, const RealD _fallback_transition_tol){
-    Linop_fallback = &_Linop_fallback;
-    fallback_transition_tol = _fallback_transition_tol;      
-  }
-    
-  void operator()(const FieldD &src, FieldD &psi) {
-    LinearOperatorBase<FieldF> *Linop_f_use = &Linop_f;
-    bool using_fallback = false;
-      
-    psi.Checkerboard() = src.Checkerboard();
-    conformable(psi, src);
-
-    RealD cp, c, a, d, b, ssq, qq, b_pred;
-
-    FieldD p(src);
-    FieldD mmp(src);
-    FieldD r(src);
-
-    // Initial residual computation & set up
-    RealD guess = norm2(psi);
-    assert(std::isnan(guess) == 0);
-    
-    Linop_d.HermOpAndNorm(psi, mmp, d, b);
-    
-    r = src - mmp;
-    p = r;
-
-    a = norm2(p);
-    cp = a;
-    ssq = norm2(src);
-
-    std::cout << GridLogIterative << std::setprecision(4) << "ConjugateGradientReliableUpdate: guess " << guess << std::endl;
-    std::cout << GridLogIterative << std::setprecision(4) << "ConjugateGradientReliableUpdate:   src " << ssq << std::endl;
-    std::cout << GridLogIterative << std::setprecision(4) << "ConjugateGradientReliableUpdate:    mp " << d << std::endl;
-    std::cout << GridLogIterative << std::setprecision(4) << "ConjugateGradientReliableUpdate:   mmp " << b << std::endl;
-    std::cout << GridLogIterative << std::setprecision(4) << "ConjugateGradientReliableUpdate:  cp,r " << cp << std::endl;
-    std::cout << GridLogIterative << std::setprecision(4) << "ConjugateGradientReliableUpdate:     p " << a << std::endl;
-
-    RealD rsq = Tolerance * Tolerance * ssq;
-
-    // Check if guess is really REALLY good :)
-    if (cp <= rsq) {
-      std::cout << GridLogMessage << "ConjugateGradientReliableUpdate guess was REALLY good\n";
-      std::cout << GridLogMessage << "\tComputed residual " << std::sqrt(cp / ssq)<<std::endl;
-      return;
-    }
-
-    //Single prec initialization
-    FieldF r_f(SinglePrecGrid);
-    r_f.Checkerboard() = r.Checkerboard();
-    precisionChange(r_f, r);
-
-    FieldF psi_f(r_f);
-    psi_f = Zero();
-
-    FieldF p_f(r_f);
-    FieldF mmp_f(r_f);
-
-    RealD MaxResidSinceLastRelUp = cp; //initial residual    
-    
-    std::cout << GridLogIterative << std::setprecision(4)
-	      << "ConjugateGradient: k=0 residual " << cp << " target " << rsq << std::endl;
-
-    GridStopWatch LinalgTimer;
-    GridStopWatch MatrixTimer;
-    GridStopWatch SolverTimer;
-
-    SolverTimer.Start();
-    int k = 0;
-    int l = 0;
-    
-    for (k = 1; k <= MaxIterations; k++) {
-      c = cp;
-
-      MatrixTimer.Start();
-      Linop_f_use->HermOpAndNorm(p_f, mmp_f, d, qq);
-      MatrixTimer.Stop();
-
-      LinalgTimer.Start();
-
-      a = c / d;
-      b_pred = a * (a * qq - d) / c;
-
-      cp = axpy_norm(r_f, -a, mmp_f, r_f);
-      b = cp / c;
-
-      // Fuse these loops ; should be really easy
-      psi_f = a * p_f + psi_f;
-      //p_f = p_f * b + r_f;
-
-      LinalgTimer.Stop();
-
-      std::cout << GridLogIterative << "ConjugateGradientReliableUpdate: Iteration " << k
-		<< " residual " << cp << " target " << rsq << std::endl;
-      std::cout << GridLogDebug << "a = "<< a << " b_pred = "<< b_pred << "  b = "<< b << std::endl;
-      std::cout << GridLogDebug << "qq = "<< qq << " d = "<< d << "  c = "<< c << std::endl;
-
-      if(cp > MaxResidSinceLastRelUp){
-	std::cout << GridLogIterative << "ConjugateGradientReliableUpdate: updating MaxResidSinceLastRelUp : " << MaxResidSinceLastRelUp << " -> " << cp << std::endl;
-	MaxResidSinceLastRelUp = cp;
-      }
-	  
-      // Stopping condition
-      if (cp <= rsq) {
-	//Although not written in the paper, I assume that I have to add on the final solution
-	precisionChange(mmp, psi_f);
-	psi = psi + mmp;
-	
-	
-	SolverTimer.Stop();
-	Linop_d.HermOpAndNorm(psi, mmp, d, qq);
-	p = mmp - src;
-
-	RealD srcnorm = std::sqrt(norm2(src));
-	RealD resnorm = std::sqrt(norm2(p));
-	RealD true_residual = resnorm / srcnorm;
-
-	std::cout << GridLogMessage << "ConjugateGradientReliableUpdate Converged on iteration " << k << " after " << l << " reliable updates" << std::endl;
-	std::cout << GridLogMessage << "\tComputed residual " << std::sqrt(cp / ssq)<<std::endl;
-	std::cout << GridLogMessage << "\tTrue residual " << true_residual<<std::endl;
-	std::cout << GridLogMessage << "\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;
-
-	IterationsToComplete = k;	
-	ReliableUpdatesPerformed = l;
-	  
-	if(DoFinalCleanup){
-	  //Do a final CG to cleanup
-	  std::cout << GridLogMessage << "ConjugateGradientReliableUpdate performing final cleanup.\n";
-	  ConjugateGradient<FieldD> CG(Tolerance,MaxIterations);
-	  CG.ErrorOnNoConverge = ErrorOnNoConverge;
-	  CG(Linop_d,src,psi);
-	  IterationsToCleanup = CG.IterationsToComplete;
-	}
-	else if (ErrorOnNoConverge) assert(true_residual / Tolerance < 10000.0);
-
-	std::cout << GridLogMessage << "ConjugateGradientReliableUpdate complete.\n";
-	return;
-      }
-      else if(cp < Delta * MaxResidSinceLastRelUp) { //reliable update
-	std::cout << GridLogMessage << "ConjugateGradientReliableUpdate "
-		  << cp << "(residual) < " << Delta << "(Delta) * " << MaxResidSinceLastRelUp << "(MaxResidSinceLastRelUp) on iteration " << k << " : performing reliable update\n";
-	precisionChange(mmp, psi_f);
-	psi = psi + mmp;
-
-	Linop_d.HermOpAndNorm(psi, mmp, d, qq);
-	r = src - mmp;
-
-	psi_f = Zero();
-	precisionChange(r_f, r);
-	cp = norm2(r);
-	MaxResidSinceLastRelUp = cp;
-
-	b = cp/c;
-	  
-	std::cout << GridLogMessage << "ConjugateGradientReliableUpdate new residual " << cp << std::endl;
-	  
-	l = l+1;
-      }
-
-      p_f = p_f * b + r_f; //update search vector after reliable update appears to help convergence
-
-      if(!using_fallback && Linop_fallback != NULL && cp < fallback_transition_tol){
-	std::cout << GridLogMessage << "ConjugateGradientReliableUpdate switching to fallback linear operator on iteration " << k << " at residual " << cp << std::endl;
-	Linop_f_use = Linop_fallback;
-	using_fallback = true;
-      }
-
-	
-    }
-    std::cout << GridLogMessage << "ConjugateGradientReliableUpdate did NOT converge"
-	      << std::endl;
-      
-    if (ErrorOnNoConverge) assert(0);
-    IterationsToComplete = k;
-    ReliableUpdatesPerformed = l;      
-  }    
-};
-
-
-NAMESPACE_END(Grid);
-
-
-
-#endif

Grid/algorithms/iterative/ConjugateResidual.h

@@ -1,113 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/algorithms/iterative/ConjugateResidual.h
-
-    Copyright (C) 2015
-
-Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
-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 */
-#ifndef GRID_CONJUGATE_RESIDUAL_H
-#define GRID_CONJUGATE_RESIDUAL_H
-
-NAMESPACE_BEGIN(Grid);
-
-/////////////////////////////////////////////////////////////
-// Base classes for iterative processes based on operators
-// single input vec, single output vec.
-/////////////////////////////////////////////////////////////
-
-template<class Field> 
-class ConjugateResidual : public OperatorFunction<Field> {
-public:                                                
-  using OperatorFunction<Field>::operator();
-
-  RealD   Tolerance;
-  Integer MaxIterations;
-  int verbose;
-
-  ConjugateResidual(RealD tol,Integer maxit) : Tolerance(tol), MaxIterations(maxit) { 
-    verbose=0;
-  };
-
-  void operator() (LinearOperatorBase<Field> &Linop,const Field &src, Field &psi){
-
-    RealD a, b; // c, d;
-    RealD cp, ssq,rsq;
-      
-    RealD rAr, rAAr, rArp;
-    RealD pAp, pAAp;
-
-    GridBase *grid = src.Grid();
-    psi=Zero();
-    Field r(grid),  p(grid), Ap(grid), Ar(grid);
-      
-    r=src;
-    p=src;
-
-    Linop.HermOpAndNorm(p,Ap,pAp,pAAp);
-    Linop.HermOpAndNorm(r,Ar,rAr,rAAr);
-
-    cp =norm2(r);
-    ssq=norm2(src);
-    rsq=Tolerance*Tolerance*ssq;
-
-    if (verbose) std::cout<<GridLogMessage<<"ConjugateResidual: iteration " <<0<<" residual "<<cp<< " target"<< rsq<<std::endl;
-
-    for(int k=1;k<MaxIterations;k++){
-
-      a = rAr/pAAp;
-
-      axpy(psi,a,p,psi);
-
-      cp = axpy_norm(r,-a,Ap,r);
-
-      rArp=rAr;
-
-      Linop.HermOpAndNorm(r,Ar,rAr,rAAr);
-
-      b   =rAr/rArp;
- 
-      axpy(p,b,p,r);
-      pAAp=axpy_norm(Ap,b,Ap,Ar);
-	
-      if(verbose) std::cout<<GridLogMessage<<"ConjugateResidual: iteration " <<k<<" residual "<<cp<< " target"<< rsq<<std::endl;
-
-      if(cp<rsq) {
-	Linop.HermOp(psi,Ap);
-	axpy(r,-1.0,src,Ap);
-	RealD true_resid = norm2(r)/ssq;
-	std::cout<<GridLogMessage<<"ConjugateResidual: Converged on iteration " <<k
-		 << " computed residual "<<std::sqrt(cp/ssq)
-		 << " true residual "<<std::sqrt(true_resid)
-		 << " target "       <<Tolerance <<std::endl;
-	return;
-      }
-
-    }
-
-    std::cout<<GridLogMessage<<"ConjugateResidual did NOT converge"<<std::endl;
-    assert(0);
-  }
-};
-NAMESPACE_END(Grid);
-#endif

Grid/algorithms/iterative/FlexibleCommunicationAvoidingGeneralisedMinimalResidual.h

@@ -1,258 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/algorithms/iterative/FlexibleCommunicationAvoidingGeneralisedMinimalResidual.h
-
-Copyright (C) 2015
-
-Author: Daniel Richtmann <daniel.richtmann@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 */
-#ifndef GRID_FLEXIBLE_COMMUNICATION_AVOIDING_GENERALISED_MINIMAL_RESIDUAL_H
-#define GRID_FLEXIBLE_COMMUNICATION_AVOIDING_GENERALISED_MINIMAL_RESIDUAL_H
-
-namespace Grid {
-
-template<class Field>
-class FlexibleCommunicationAvoidingGeneralisedMinimalResidual : public OperatorFunction<Field> {
- public:
-  using OperatorFunction<Field>::operator();
-
-  bool ErrorOnNoConverge; // Throw an assert when FCAGMRES fails to converge,
-                          // defaults to true
-
-  RealD   Tolerance;
-
-  Integer MaxIterations;
-  Integer RestartLength;
-  Integer MaxNumberOfRestarts;
-  Integer IterationCount; // Number of iterations the FCAGMRES took to finish,
-                          // filled in upon completion
-
-  GridStopWatch MatrixTimer;
-  GridStopWatch PrecTimer;
-  GridStopWatch LinalgTimer;
-  GridStopWatch QrTimer;
-  GridStopWatch CompSolutionTimer;
-
-  Eigen::MatrixXcd H;
-
-  std::vector<ComplexD> y;
-  std::vector<ComplexD> gamma;
-  std::vector<ComplexD> c;
-  std::vector<ComplexD> s;
-
-  LinearFunction<Field> &Preconditioner;
-
-  FlexibleCommunicationAvoidingGeneralisedMinimalResidual(RealD   tol,
-                                                          Integer maxit,
-                                                          LinearFunction<Field> &Prec,
-                                                          Integer restart_length,
-                                                          bool    err_on_no_conv = true)
-      : Tolerance(tol)
-      , MaxIterations(maxit)
-      , RestartLength(restart_length)
-      , MaxNumberOfRestarts(MaxIterations/RestartLength + ((MaxIterations%RestartLength == 0) ? 0 : 1))
-      , ErrorOnNoConverge(err_on_no_conv)
-      , H(Eigen::MatrixXcd::Zero(RestartLength, RestartLength + 1)) // sizes taken from DD-αAMG code base
-      , y(RestartLength + 1, 0.)
-      , gamma(RestartLength + 1, 0.)
-      , c(RestartLength + 1, 0.)
-      , s(RestartLength + 1, 0.)
-      , Preconditioner(Prec) {};
-
-  void operator()(LinearOperatorBase<Field> &LinOp, const Field &src, Field &psi) {
-
-    std::cout << GridLogWarning << "This algorithm currently doesn't differ from regular FGMRES" << std::endl;
-
-    psi.Checkerboard() = src.Checkerboard();
-    conformable(psi, src);
-
-    RealD guess = norm2(psi);
-    assert(std::isnan(guess) == 0);
-
-    RealD cp;
-    RealD ssq = norm2(src);
-    RealD rsq = Tolerance * Tolerance * ssq;
-
-    Field r(src.Grid());
-
-    std::cout << std::setprecision(4) << std::scientific;
-    std::cout << GridLogIterative << "FlexibleCommunicationAvoidingGeneralisedMinimalResidual: guess " << guess << std::endl;
-    std::cout << GridLogIterative << "FlexibleCommunicationAvoidingGeneralisedMinimalResidual:   src " << ssq   << std::endl;
-
-    PrecTimer.Reset();
-    MatrixTimer.Reset();
-    LinalgTimer.Reset();
-    QrTimer.Reset();
-    CompSolutionTimer.Reset();
-
-    GridStopWatch SolverTimer;
-    SolverTimer.Start();
-
-    IterationCount = 0;
-
-    for (int k=0; k<MaxNumberOfRestarts; k++) {
-
-      cp = outerLoopBody(LinOp, src, psi, rsq);
-
-      // Stopping condition
-      if (cp <= rsq) {
-
-        SolverTimer.Stop();
-
-        LinOp.Op(psi,r);
-        axpy(r,-1.0,src,r);
-
-        RealD srcnorm       = sqrt(ssq);
-        RealD resnorm       = sqrt(norm2(r));
-        RealD true_residual = resnorm / srcnorm;
-
-        std::cout << GridLogMessage        << "FlexibleCommunicationAvoidingGeneralisedMinimalResidual: Converged on iteration " << IterationCount
-                  << " computed residual " << sqrt(cp / ssq)
-                  << " true residual "     << true_residual
-                  << " target "            << Tolerance << std::endl;
-
-        std::cout << GridLogMessage << "FCAGMRES Time elapsed: Total   " <<       SolverTimer.Elapsed() << std::endl;
-        std::cout << GridLogMessage << "FCAGMRES Time elapsed: Precon  " <<         PrecTimer.Elapsed() << std::endl;
-        std::cout << GridLogMessage << "FCAGMRES Time elapsed: Matrix  " <<       MatrixTimer.Elapsed() << std::endl;
-        std::cout << GridLogMessage << "FCAGMRES Time elapsed: Linalg  " <<       LinalgTimer.Elapsed() << std::endl;
-        std::cout << GridLogMessage << "FCAGMRES Time elapsed: QR      " <<           QrTimer.Elapsed() << std::endl;
-        std::cout << GridLogMessage << "FCAGMRES Time elapsed: CompSol " << CompSolutionTimer.Elapsed() << std::endl;
-        return;
-      }
-    }
-
-    std::cout << GridLogMessage << "FlexibleCommunicationAvoidingGeneralisedMinimalResidual did NOT converge" << std::endl;
-
-    if (ErrorOnNoConverge)
-      assert(0);
-  }
-
-  RealD outerLoopBody(LinearOperatorBase<Field> &LinOp, const Field &src, Field &psi, RealD rsq) {
-
-    RealD cp = 0;
-
-    Field w(src.Grid());
-    Field r(src.Grid());
-
-    // these should probably be made class members so that they are only allocated once, not in every restart
-    std::vector<Field> v(RestartLength + 1, src.Grid()); for (auto &elem : v) elem = Zero();
-    std::vector<Field> z(RestartLength + 1, src.Grid()); for (auto &elem : z) elem = Zero();
-
-    MatrixTimer.Start();
-    LinOp.Op(psi, w);
-    MatrixTimer.Stop();
-
-    LinalgTimer.Start();
-    r = src - w;
-
-    gamma[0] = sqrt(norm2(r));
-
-    v[0] = (1. / gamma[0]) * r;
-    LinalgTimer.Stop();
-
-    for (int i=0; i<RestartLength; i++) {
-
-      IterationCount++;
-
-      arnoldiStep(LinOp, v, z, w, i);
-
-      qrUpdate(i);
-
-      cp = norm(gamma[i+1]);
-
-      std::cout << GridLogIterative << "FlexibleCommunicationAvoidingGeneralisedMinimalResidual: Iteration " << IterationCount
-                << " residual " << cp << " target " << rsq << std::endl;
-
-      if ((i == RestartLength - 1) || (IterationCount == MaxIterations) || (cp <= rsq)) {
-
-        computeSolution(z, psi, i);
-
-        return cp;
-      }
-    }
-
-    assert(0); // Never reached
-    return cp;
-  }
-
-  void arnoldiStep(LinearOperatorBase<Field> &LinOp, std::vector<Field> &v, std::vector<Field> &z, Field &w, int iter) {
-
-    PrecTimer.Start();
-    Preconditioner(v[iter], z[iter]);
-    PrecTimer.Stop();
-
-    MatrixTimer.Start();
-    LinOp.Op(z[iter], w);
-    MatrixTimer.Stop();
-
-    LinalgTimer.Start();
-    for (int i = 0; i <= iter; ++i) {
-      H(iter, i) = innerProduct(v[i], w);
-      w = w - ComplexD(H(iter, i)) * v[i];
-    }
-
-    H(iter, iter + 1) = sqrt(norm2(w));
-    v[iter + 1] = ComplexD(1. / H(iter, iter + 1)) * w;
-    LinalgTimer.Stop();
-  }
-
-  void qrUpdate(int iter) {
-
-    QrTimer.Start();
-    for (int i = 0; i < iter ; ++i) {
-      auto tmp       = -s[i] * ComplexD(H(iter, i)) + c[i] * ComplexD(H(iter, i + 1));
-      H(iter, i)     = conjugate(c[i]) * ComplexD(H(iter, i)) + conjugate(s[i]) * ComplexD(H(iter, i + 1));
-      H(iter, i + 1) = tmp;
-    }
-
-    // Compute new Givens Rotation
-    auto nu = sqrt(std::norm(H(iter, iter)) + std::norm(H(iter, iter + 1)));
-    c[iter]     = H(iter, iter) / nu;
-    s[iter]     = H(iter, iter + 1) / nu;
-
-    // Apply new Givens rotation
-    H(iter, iter)     = nu;
-    H(iter, iter + 1) = 0.;
-
-    gamma[iter + 1] = -s[iter] * gamma[iter];
-    gamma[iter]     = conjugate(c[iter]) * gamma[iter];
-    QrTimer.Stop();
-  }
-
-  void computeSolution(std::vector<Field> const &z, Field &psi, int iter) {
-
-    CompSolutionTimer.Start();
-    for (int i = iter; i >= 0; i--) {
-      y[i] = gamma[i];
-      for (int k = i + 1; k <= iter; k++)
-        y[i] = y[i] - ComplexD(H(k, i)) * y[k];
-      y[i] = y[i] / ComplexD(H(i, i));
-    }
-
-    for (int i = 0; i <= iter; i++)
-      psi = psi + z[i] * y[i];
-    CompSolutionTimer.Stop();
-  }
-};
-}
-#endif

Grid/algorithms/iterative/FlexibleGeneralisedMinimalResidual.h

@@ -1,256 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/algorithms/iterative/FlexibleGeneralisedMinimalResidual.h
-
-Copyright (C) 2015
-
-Author: Daniel Richtmann <daniel.richtmann@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 */
-#ifndef GRID_FLEXIBLE_GENERALISED_MINIMAL_RESIDUAL_H
-#define GRID_FLEXIBLE_GENERALISED_MINIMAL_RESIDUAL_H
-
-namespace Grid {
-
-template<class Field>
-class FlexibleGeneralisedMinimalResidual : public OperatorFunction<Field> {
- public:
-  using OperatorFunction<Field>::operator();
-
-  bool ErrorOnNoConverge; // Throw an assert when FGMRES fails to converge,
-                          // defaults to true
-
-  RealD   Tolerance;
-
-  Integer MaxIterations;
-  Integer RestartLength;
-  Integer MaxNumberOfRestarts;
-  Integer IterationCount; // Number of iterations the FGMRES took to finish,
-                          // filled in upon completion
-
-  GridStopWatch MatrixTimer;
-  GridStopWatch PrecTimer;
-  GridStopWatch LinalgTimer;
-  GridStopWatch QrTimer;
-  GridStopWatch CompSolutionTimer;
-
-  Eigen::MatrixXcd H;
-
-  std::vector<ComplexD> y;
-  std::vector<ComplexD> gamma;
-  std::vector<ComplexD> c;
-  std::vector<ComplexD> s;
-
-  LinearFunction<Field> &Preconditioner;
-
-  FlexibleGeneralisedMinimalResidual(RealD   tol,
-                                     Integer maxit,
-                                     LinearFunction<Field> &Prec,
-                                     Integer restart_length,
-                                     bool    err_on_no_conv = true)
-      : Tolerance(tol)
-      , MaxIterations(maxit)
-      , RestartLength(restart_length)
-      , MaxNumberOfRestarts(MaxIterations/RestartLength + ((MaxIterations%RestartLength == 0) ? 0 : 1))
-      , ErrorOnNoConverge(err_on_no_conv)
-      , H(Eigen::MatrixXcd::Zero(RestartLength, RestartLength + 1)) // sizes taken from DD-αAMG code base
-      , y(RestartLength + 1, 0.)
-      , gamma(RestartLength + 1, 0.)
-      , c(RestartLength + 1, 0.)
-      , s(RestartLength + 1, 0.)
-      , Preconditioner(Prec) {};
-
-  void operator()(LinearOperatorBase<Field> &LinOp, const Field &src, Field &psi) {
-
-    psi.Checkerboard() = src.Checkerboard();
-    conformable(psi, src);
-
-    RealD guess = norm2(psi);
-    assert(std::isnan(guess) == 0);
-
-    RealD cp;
-    RealD ssq = norm2(src);
-    RealD rsq = Tolerance * Tolerance * ssq;
-
-    Field r(src.Grid());
-
-    std::cout << std::setprecision(4) << std::scientific;
-    std::cout << GridLogIterative << "FlexibleGeneralisedMinimalResidual: guess " << guess << std::endl;
-    std::cout << GridLogIterative << "FlexibleGeneralisedMinimalResidual:   src " << ssq   << std::endl;
-
-    PrecTimer.Reset();
-    MatrixTimer.Reset();
-    LinalgTimer.Reset();
-    QrTimer.Reset();
-    CompSolutionTimer.Reset();
-
-    GridStopWatch SolverTimer;
-    SolverTimer.Start();
-
-    IterationCount = 0;
-
-    for (int k=0; k<MaxNumberOfRestarts; k++) {
-
-      cp = outerLoopBody(LinOp, src, psi, rsq);
-
-      // Stopping condition
-      if (cp <= rsq) {
-
-        SolverTimer.Stop();
-
-        LinOp.Op(psi,r);
-        axpy(r,-1.0,src,r);
-
-        RealD srcnorm       = sqrt(ssq);
-        RealD resnorm       = sqrt(norm2(r));
-        RealD true_residual = resnorm / srcnorm;
-
-        std::cout << GridLogMessage        << "FlexibleGeneralisedMinimalResidual: Converged on iteration " << IterationCount
-                  << " computed residual " << sqrt(cp / ssq)
-                  << " true residual "     << true_residual
-                  << " target "            << Tolerance << std::endl;
-
-        std::cout << GridLogMessage << "FGMRES Time elapsed: Total   " <<       SolverTimer.Elapsed() << std::endl;
-        std::cout << GridLogMessage << "FGMRES Time elapsed: Precon  " <<         PrecTimer.Elapsed() << std::endl;
-        std::cout << GridLogMessage << "FGMRES Time elapsed: Matrix  " <<       MatrixTimer.Elapsed() << std::endl;
-        std::cout << GridLogMessage << "FGMRES Time elapsed: Linalg  " <<       LinalgTimer.Elapsed() << std::endl;
-        std::cout << GridLogMessage << "FGMRES Time elapsed: QR      " <<           QrTimer.Elapsed() << std::endl;
-        std::cout << GridLogMessage << "FGMRES Time elapsed: CompSol " << CompSolutionTimer.Elapsed() << std::endl;
-        return;
-      }
-    }
-
-    std::cout << GridLogMessage << "FlexibleGeneralisedMinimalResidual did NOT converge" << std::endl;
-
-    if (ErrorOnNoConverge)
-      assert(0);
-  }
-
-  RealD outerLoopBody(LinearOperatorBase<Field> &LinOp, const Field &src, Field &psi, RealD rsq) {
-
-    RealD cp = 0;
-
-    Field w(src.Grid());
-    Field r(src.Grid());
-
-    // these should probably be made class members so that they are only allocated once, not in every restart
-    std::vector<Field> v(RestartLength + 1, src.Grid()); for (auto &elem : v) elem = Zero();
-    std::vector<Field> z(RestartLength + 1, src.Grid()); for (auto &elem : z) elem = Zero();
-
-    MatrixTimer.Start();
-    LinOp.Op(psi, w);
-    MatrixTimer.Stop();
-
-    LinalgTimer.Start();
-    r = src - w;
-
-    gamma[0] = sqrt(norm2(r));
-
-    v[0] = (1. / gamma[0]) * r;
-    LinalgTimer.Stop();
-
-    for (int i=0; i<RestartLength; i++) {
-
-      IterationCount++;
-
-      arnoldiStep(LinOp, v, z, w, i);
-
-      qrUpdate(i);
-
-      cp = norm(gamma[i+1]);
-
-      std::cout << GridLogIterative << "FlexibleGeneralisedMinimalResidual: Iteration " << IterationCount
-                << " residual " << cp << " target " << rsq << std::endl;
-
-      if ((i == RestartLength - 1) || (IterationCount == MaxIterations) || (cp <= rsq)) {
-
-        computeSolution(z, psi, i);
-
-        return cp;
-      }
-    }
-
-    assert(0); // Never reached
-    return cp;
-  }
-
-  void arnoldiStep(LinearOperatorBase<Field> &LinOp, std::vector<Field> &v, std::vector<Field> &z, Field &w, int iter) {
-
-    PrecTimer.Start();
-    Preconditioner(v[iter], z[iter]);
-    PrecTimer.Stop();
-
-    MatrixTimer.Start();
-    LinOp.Op(z[iter], w);
-    MatrixTimer.Stop();
-
-    LinalgTimer.Start();
-    for (int i = 0; i <= iter; ++i) {
-      H(iter, i) = innerProduct(v[i], w);
-      w = w - ComplexD(H(iter, i)) * v[i];
-    }
-
-    H(iter, iter + 1) = sqrt(norm2(w));
-    v[iter + 1] = ComplexD(1. / H(iter, iter + 1)) * w;
-    LinalgTimer.Stop();
-  }
-
-  void qrUpdate(int iter) {
-
-    QrTimer.Start();
-    for (int i = 0; i < iter ; ++i) {
-      auto tmp       = -s[i] * ComplexD(H(iter, i)) + c[i] * ComplexD(H(iter, i + 1));
-      H(iter, i)     = conjugate(c[i]) * ComplexD(H(iter, i)) + conjugate(s[i]) * ComplexD(H(iter, i + 1));
-      H(iter, i + 1) = tmp;
-    }
-
-    // Compute new Givens Rotation
-    auto nu = sqrt(std::norm(H(iter, iter)) + std::norm(H(iter, iter + 1)));
-    c[iter]     = H(iter, iter) / nu;
-    s[iter]     = H(iter, iter + 1) / nu;
-
-    // Apply new Givens rotation
-    H(iter, iter)     = nu;
-    H(iter, iter + 1) = 0.;
-
-    gamma[iter + 1] = -s[iter] * gamma[iter];
-    gamma[iter]     = conjugate(c[iter]) * gamma[iter];
-    QrTimer.Stop();
-  }
-
-  void computeSolution(std::vector<Field> const &z, Field &psi, int iter) {
-
-    CompSolutionTimer.Start();
-    for (int i = iter; i >= 0; i--) {
-      y[i] = gamma[i];
-      for (int k = i + 1; k <= iter; k++)
-        y[i] = y[i] - ComplexD(H(k, i)) * y[k];
-      y[i] = y[i] / ComplexD(H(i, i));
-    }
-
-    for (int i = 0; i <= iter; i++)
-      psi = psi + z[i] * y[i];
-    CompSolutionTimer.Stop();
-  }
-};
-}
-#endif

Grid/algorithms/iterative/GeneralisedMinimalResidual.h

@@ -1,244 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/algorithms/iterative/GeneralisedMinimalResidual.h
-
-Copyright (C) 2015
-
-Author: Daniel Richtmann <daniel.richtmann@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 */
-#ifndef GRID_GENERALISED_MINIMAL_RESIDUAL_H
-#define GRID_GENERALISED_MINIMAL_RESIDUAL_H
-
-namespace Grid {
-
-template<class Field>
-class GeneralisedMinimalResidual : public OperatorFunction<Field> {
- public:
-  using OperatorFunction<Field>::operator();
-
-  bool ErrorOnNoConverge; // Throw an assert when GMRES fails to converge,
-                          // defaults to true
-
-  RealD   Tolerance;
-
-  Integer MaxIterations;
-  Integer RestartLength;
-  Integer MaxNumberOfRestarts;
-  Integer IterationCount; // Number of iterations the GMRES took to finish,
-                          // filled in upon completion
-
-  GridStopWatch MatrixTimer;
-  GridStopWatch LinalgTimer;
-  GridStopWatch QrTimer;
-  GridStopWatch CompSolutionTimer;
-
-  Eigen::MatrixXcd H;
-
-  std::vector<ComplexD> y;
-  std::vector<ComplexD> gamma;
-  std::vector<ComplexD> c;
-  std::vector<ComplexD> s;
-
-  GeneralisedMinimalResidual(RealD   tol,
-                             Integer maxit,
-                             Integer restart_length,
-                             bool    err_on_no_conv = true)
-      : Tolerance(tol)
-      , MaxIterations(maxit)
-      , RestartLength(restart_length)
-      , MaxNumberOfRestarts(MaxIterations/RestartLength + ((MaxIterations%RestartLength == 0) ? 0 : 1))
-      , ErrorOnNoConverge(err_on_no_conv)
-      , H(Eigen::MatrixXcd::Zero(RestartLength, RestartLength + 1)) // sizes taken from DD-αAMG code base
-      , y(RestartLength + 1, 0.)
-      , gamma(RestartLength + 1, 0.)
-      , c(RestartLength + 1, 0.)
-      , s(RestartLength + 1, 0.) {};
-
-  void operator()(LinearOperatorBase<Field> &LinOp, const Field &src, Field &psi) {
-
-    psi.Checkerboard() = src.Checkerboard();
-    conformable(psi, src);
-
-    RealD guess = norm2(psi);
-    assert(std::isnan(guess) == 0);
-
-    RealD cp;
-    RealD ssq = norm2(src);
-    RealD rsq = Tolerance * Tolerance * ssq;
-
-    Field r(src.Grid());
-
-    std::cout << std::setprecision(4) << std::scientific;
-    std::cout << GridLogIterative << "GeneralisedMinimalResidual: guess " << guess << std::endl;
-    std::cout << GridLogIterative << "GeneralisedMinimalResidual:   src " << ssq   << std::endl;
-
-    MatrixTimer.Reset();
-    LinalgTimer.Reset();
-    QrTimer.Reset();
-    CompSolutionTimer.Reset();
-
-    GridStopWatch SolverTimer;
-    SolverTimer.Start();
-
-    IterationCount = 0;
-
-    for (int k=0; k<MaxNumberOfRestarts; k++) {
-
-      cp = outerLoopBody(LinOp, src, psi, rsq);
-
-      // Stopping condition
-      if (cp <= rsq) {
-
-        SolverTimer.Stop();
-
-        LinOp.Op(psi,r);
-        axpy(r,-1.0,src,r);
-
-        RealD srcnorm       = sqrt(ssq);
-        RealD resnorm       = sqrt(norm2(r));
-        RealD true_residual = resnorm / srcnorm;
-
-        std::cout << GridLogMessage        << "GeneralisedMinimalResidual: Converged on iteration " << IterationCount
-                  << " computed residual " << sqrt(cp / ssq)
-                  << " true residual "     << true_residual
-                  << " target "            << Tolerance << std::endl;
-
-        std::cout << GridLogMessage << "GMRES Time elapsed: Total   " <<       SolverTimer.Elapsed() << std::endl;
-        std::cout << GridLogMessage << "GMRES Time elapsed: Matrix  " <<       MatrixTimer.Elapsed() << std::endl;
-        std::cout << GridLogMessage << "GMRES Time elapsed: Linalg  " <<       LinalgTimer.Elapsed() << std::endl;
-        std::cout << GridLogMessage << "GMRES Time elapsed: QR      " <<           QrTimer.Elapsed() << std::endl;
-        std::cout << GridLogMessage << "GMRES Time elapsed: CompSol " << CompSolutionTimer.Elapsed() << std::endl;
-        return;
-      }
-    }
-
-    std::cout << GridLogMessage << "GeneralisedMinimalResidual did NOT converge" << std::endl;
-
-    if (ErrorOnNoConverge)
-      assert(0);
-  }
-
-  RealD outerLoopBody(LinearOperatorBase<Field> &LinOp, const Field &src, Field &psi, RealD rsq) {
-
-    RealD cp = 0;
-
-    Field w(src.Grid());
-    Field r(src.Grid());
-
-    // this should probably be made a class member so that it is only allocated once, not in every restart
-    std::vector<Field> v(RestartLength + 1, src.Grid()); for (auto &elem : v) elem = Zero();
-
-    MatrixTimer.Start();
-    LinOp.Op(psi, w);
-    MatrixTimer.Stop();
-
-    LinalgTimer.Start();
-    r = src - w;
-
-    gamma[0] = sqrt(norm2(r));
-
-    v[0] = (1. / gamma[0]) * r;
-    LinalgTimer.Stop();
-
-    for (int i=0; i<RestartLength; i++) {
-
-      IterationCount++;
-
-      arnoldiStep(LinOp, v, w, i);
-
-      qrUpdate(i);
-
-      cp = norm(gamma[i+1]);
-
-      std::cout << GridLogIterative << "GeneralisedMinimalResidual: Iteration " << IterationCount
-                << " residual " << cp << " target " << rsq << std::endl;
-
-      if ((i == RestartLength - 1) || (IterationCount == MaxIterations) || (cp <= rsq)) {
-
-        computeSolution(v, psi, i);
-
-        return cp;
-      }
-    }
-
-    assert(0); // Never reached
-    return cp;
-  }
-
-  void arnoldiStep(LinearOperatorBase<Field> &LinOp, std::vector<Field> &v, Field &w, int iter) {
-
-    MatrixTimer.Start();
-    LinOp.Op(v[iter], w);
-    MatrixTimer.Stop();
-
-    LinalgTimer.Start();
-    for (int i = 0; i <= iter; ++i) {
-      H(iter, i) = innerProduct(v[i], w);
-      w = w - ComplexD(H(iter, i)) * v[i];
-    }
-
-    H(iter, iter + 1) = sqrt(norm2(w));
-    v[iter + 1] = ComplexD(1. / H(iter, iter + 1)) * w;
-    LinalgTimer.Stop();
-  }
-
-  void qrUpdate(int iter) {
-
-    QrTimer.Start();
-    for (int i = 0; i < iter ; ++i) {
-      auto tmp       = -s[i] * ComplexD(H(iter, i)) + c[i] * ComplexD(H(iter, i + 1));
-      H(iter, i)     = conjugate(c[i]) * ComplexD(H(iter, i)) + conjugate(s[i]) * ComplexD(H(iter, i + 1));
-      H(iter, i + 1) = tmp;
-    }
-
-    // Compute new Givens Rotation
-    auto nu = sqrt(std::norm(H(iter, iter)) + std::norm(H(iter, iter + 1)));
-    c[iter]     = H(iter, iter) / nu;
-    s[iter]     = H(iter, iter + 1) / nu;
-
-    // Apply new Givens rotation
-    H(iter, iter)     = nu;
-    H(iter, iter + 1) = 0.;
-
-    gamma[iter + 1] = -s[iter] * gamma[iter];
-    gamma[iter]     = conjugate(c[iter]) * gamma[iter];
-    QrTimer.Stop();
-  }
-
-  void computeSolution(std::vector<Field> const &v, Field &psi, int iter) {
-
-    CompSolutionTimer.Start();
-    for (int i = iter; i >= 0; i--) {
-      y[i] = gamma[i];
-      for (int k = i + 1; k <= iter; k++)
-        y[i] = y[i] - ComplexD(H(k, i)) * y[k];
-      y[i] = y[i] / ComplexD(H(i, i));
-    }
-
-    for (int i = 0; i <= iter; i++)
-      psi = psi + v[i] * y[i];
-    CompSolutionTimer.Stop();
-  }
-};
-}
-#endif

Grid/algorithms/iterative/MinimalResidual.h

@@ -1,157 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/algorithms/iterative/MinimalResidual.h
-
-Copyright (C) 2015
-
-Author: Daniel Richtmann <daniel.richtmann@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 */
-#ifndef GRID_MINIMAL_RESIDUAL_H
-#define GRID_MINIMAL_RESIDUAL_H
-
-namespace Grid {
-
-template<class Field> class MinimalResidual : public OperatorFunction<Field> {
- public:
-  using OperatorFunction<Field>::operator();
-
-  bool ErrorOnNoConverge; // throw an assert when the MR fails to converge.
-                          // Defaults true.
-  RealD   Tolerance;
-  Integer MaxIterations;
-  RealD   overRelaxParam;
-  Integer IterationsToComplete; // Number of iterations the MR took to finish.
-                                // Filled in upon completion
-
-  MinimalResidual(RealD tol, Integer maxit, Real ovrelparam = 1.0, bool err_on_no_conv = true)
-    : Tolerance(tol), MaxIterations(maxit), overRelaxParam(ovrelparam), ErrorOnNoConverge(err_on_no_conv){};
-
-  void operator()(LinearOperatorBase<Field> &Linop, const Field &src, Field &psi) {
-
-    psi.Checkerboard() = src.Checkerboard();
-    conformable(psi, src);
-
-    ComplexD a, c;
-    RealD    d;
-
-    Field Mr(src);
-    Field r(src);
-
-    // Initial residual computation & set up
-    RealD guess = norm2(psi);
-    assert(std::isnan(guess) == 0);
-
-    RealD ssq = norm2(src);
-    RealD rsq = Tolerance * Tolerance * ssq;
-
-    Linop.Op(psi, Mr);
-
-    r = src - Mr;
-
-    RealD cp = norm2(r);
-
-    std::cout << std::setprecision(4) << std::scientific;
-    std::cout << GridLogIterative << "MinimalResidual: guess " << guess << std::endl;
-    std::cout << GridLogIterative << "MinimalResidual:   src " << ssq << std::endl;
-    std::cout << GridLogIterative << "MinimalResidual:  cp,r " << cp << std::endl;
-
-    if (cp <= rsq) {
-      return;
-    }
-
-    std::cout << GridLogIterative << "MinimalResidual: k=0 residual " << cp << " target " << rsq << std::endl;
-
-    GridStopWatch LinalgTimer;
-    GridStopWatch MatrixTimer;
-    GridStopWatch SolverTimer;
-
-    SolverTimer.Start();
-    int k;
-    for (k = 1; k <= MaxIterations; k++) {
-
-      MatrixTimer.Start();
-      Linop.Op(r, Mr);
-      MatrixTimer.Stop();
-
-      LinalgTimer.Start();
-
-      c = innerProduct(Mr, r);
-
-      d = norm2(Mr);
-
-      a = c / d;
-
-      a = a * overRelaxParam;
-
-      psi = psi + r * a;
-
-      r = r - Mr * a;
-
-      cp = norm2(r);
-
-      LinalgTimer.Stop();
-
-      std::cout << GridLogIterative << "MinimalResidual: Iteration " << k
-                << " residual " << cp << " target " << rsq << std::endl;
-      std::cout << GridLogDebug << "a = " << a << " c = " << c << " d = " << d << std::endl;
-
-      // Stopping condition
-      if (cp <= rsq) {
-        SolverTimer.Stop();
-
-        Linop.Op(psi, Mr);
-        r = src - Mr;
-
-        RealD srcnorm       = sqrt(ssq);
-        RealD resnorm       = sqrt(norm2(r));
-        RealD true_residual = resnorm / srcnorm;
-
-        std::cout << GridLogMessage        << "MinimalResidual Converged on iteration " << k
-                  << " computed residual " << sqrt(cp / ssq)
-                  << " true residual "     << true_residual
-                  << " target "            << Tolerance << std::endl;
-
-        std::cout << GridLogMessage << "MR Time elapsed: Total   " << SolverTimer.Elapsed() << std::endl;
-        std::cout << GridLogMessage << "MR Time elapsed: Matrix  " << MatrixTimer.Elapsed() << std::endl;
-        std::cout << GridLogMessage << "MR Time elapsed: Linalg  " << LinalgTimer.Elapsed() << std::endl;
-
-        if (ErrorOnNoConverge)
-          assert(true_residual / Tolerance < 10000.0);
-
-        IterationsToComplete = k;
-
-        return;
-      }
-    }
-
-    std::cout << GridLogMessage << "MinimalResidual did NOT converge"
-              << std::endl;
-
-    if (ErrorOnNoConverge)
-      assert(0);
-
-    IterationsToComplete = k;
-  }
-};
-} // namespace Grid
-#endif

Grid/algorithms/iterative/MixedPrecisionFlexibleGeneralisedMinimalResidual.h

@@ -1,276 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/algorithms/iterative/MixedPrecisionFlexibleGeneralisedMinimalResidual.h
-
-Copyright (C) 2015
-
-Author: Daniel Richtmann <daniel.richtmann@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 */
-#ifndef GRID_MIXED_PRECISION_FLEXIBLE_GENERALISED_MINIMAL_RESIDUAL_H
-#define GRID_MIXED_PRECISION_FLEXIBLE_GENERALISED_MINIMAL_RESIDUAL_H
-
-namespace 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 MixedPrecisionFlexibleGeneralisedMinimalResidual : public OperatorFunction<FieldD> {
- public:
-
-  using OperatorFunction<FieldD>::operator();
-
-  bool ErrorOnNoConverge; // Throw an assert when MPFGMRES fails to converge,
-                          // defaults to true
-
-  RealD   Tolerance;
-
-  Integer MaxIterations;
-  Integer RestartLength;
-  Integer MaxNumberOfRestarts;
-  Integer IterationCount; // Number of iterations the MPFGMRES took to finish,
-                          // filled in upon completion
-
-  GridStopWatch MatrixTimer;
-  GridStopWatch PrecTimer;
-  GridStopWatch LinalgTimer;
-  GridStopWatch QrTimer;
-  GridStopWatch CompSolutionTimer;
-  GridStopWatch ChangePrecTimer;
-
-  Eigen::MatrixXcd H;
-
-  std::vector<ComplexD> y;
-  std::vector<ComplexD> gamma;
-  std::vector<ComplexD> c;
-  std::vector<ComplexD> s;
-
-  GridBase* SinglePrecGrid;
-
-  LinearFunction<FieldF> &Preconditioner;
-
-  MixedPrecisionFlexibleGeneralisedMinimalResidual(RealD   tol,
-                                                   Integer maxit,
-                                                   GridBase * sp_grid,
-                                                   LinearFunction<FieldF> &Prec,
-                                                   Integer restart_length,
-                                                   bool    err_on_no_conv = true)
-      : Tolerance(tol)
-      , MaxIterations(maxit)
-      , RestartLength(restart_length)
-      , MaxNumberOfRestarts(MaxIterations/RestartLength + ((MaxIterations%RestartLength == 0) ? 0 : 1))
-      , ErrorOnNoConverge(err_on_no_conv)
-      , H(Eigen::MatrixXcd::Zero(RestartLength, RestartLength + 1)) // sizes taken from DD-αAMG code base
-      , y(RestartLength + 1, 0.)
-      , gamma(RestartLength + 1, 0.)
-      , c(RestartLength + 1, 0.)
-      , s(RestartLength + 1, 0.)
-      , SinglePrecGrid(sp_grid)
-      , Preconditioner(Prec) {};
-
-  void operator()(LinearOperatorBase<FieldD> &LinOp, const FieldD &src, FieldD &psi) {
-
-    psi.Checkerboard() = src.Checkerboard();
-    conformable(psi, src);
-
-    RealD guess = norm2(psi);
-    assert(std::isnan(guess) == 0);
-
-    RealD cp;
-    RealD ssq = norm2(src);
-    RealD rsq = Tolerance * Tolerance * ssq;
-
-    FieldD r(src.Grid());
-
-    std::cout << std::setprecision(4) << std::scientific;
-    std::cout << GridLogIterative << "MPFGMRES: guess " << guess << std::endl;
-    std::cout << GridLogIterative << "MPFGMRES:   src " << ssq   << std::endl;
-
-    PrecTimer.Reset();
-    MatrixTimer.Reset();
-    LinalgTimer.Reset();
-    QrTimer.Reset();
-    CompSolutionTimer.Reset();
-    ChangePrecTimer.Reset();
-
-    GridStopWatch SolverTimer;
-    SolverTimer.Start();
-
-    IterationCount = 0;
-
-    for (int k=0; k<MaxNumberOfRestarts; k++) {
-
-      cp = outerLoopBody(LinOp, src, psi, rsq);
-
-      // Stopping condition
-      if (cp <= rsq) {
-
-        SolverTimer.Stop();
-
-        LinOp.Op(psi,r);
-        axpy(r,-1.0,src,r);
-
-        RealD srcnorm       = sqrt(ssq);
-        RealD resnorm       = sqrt(norm2(r));
-        RealD true_residual = resnorm / srcnorm;
-
-        std::cout << GridLogMessage        << "MPFGMRES: Converged on iteration " << IterationCount
-                  << " computed residual " << sqrt(cp / ssq)
-                  << " true residual "     << true_residual
-                  << " target "            << Tolerance << std::endl;
-
-        std::cout << GridLogMessage << "MPFGMRES Time elapsed: Total      " <<       SolverTimer.Elapsed() << std::endl;
-        std::cout << GridLogMessage << "MPFGMRES Time elapsed: Precon     " <<         PrecTimer.Elapsed() << std::endl;
-        std::cout << GridLogMessage << "MPFGMRES Time elapsed: Matrix     " <<       MatrixTimer.Elapsed() << std::endl;
-        std::cout << GridLogMessage << "MPFGMRES Time elapsed: Linalg     " <<       LinalgTimer.Elapsed() << std::endl;
-        std::cout << GridLogMessage << "MPFGMRES Time elapsed: QR         " <<           QrTimer.Elapsed() << std::endl;
-        std::cout << GridLogMessage << "MPFGMRES Time elapsed: CompSol    " << CompSolutionTimer.Elapsed() << std::endl;
-        std::cout << GridLogMessage << "MPFGMRES Time elapsed: PrecChange " <<   ChangePrecTimer.Elapsed() << std::endl;
-        return;
-      }
-    }
-
-    std::cout << GridLogMessage << "MPFGMRES did NOT converge" << std::endl;
-
-    if (ErrorOnNoConverge)
-      assert(0);
-  }
-
-  RealD outerLoopBody(LinearOperatorBase<FieldD> &LinOp, const FieldD &src, FieldD &psi, RealD rsq) {
-
-    RealD cp = 0;
-
-    FieldD w(src.Grid());
-    FieldD r(src.Grid());
-
-    // these should probably be made class members so that they are only allocated once, not in every restart
-    std::vector<FieldD> v(RestartLength + 1, src.Grid()); for (auto &elem : v) elem = Zero();
-    std::vector<FieldD> z(RestartLength + 1, src.Grid()); for (auto &elem : z) elem = Zero();
-
-    MatrixTimer.Start();
-    LinOp.Op(psi, w);
-    MatrixTimer.Stop();
-
-    LinalgTimer.Start();
-    r = src - w;
-
-    gamma[0] = sqrt(norm2(r));
-
-    v[0] = (1. / gamma[0]) * r;
-    LinalgTimer.Stop();
-
-    for (int i=0; i<RestartLength; i++) {
-
-      IterationCount++;
-
-      arnoldiStep(LinOp, v, z, w, i);
-
-      qrUpdate(i);
-
-      cp = norm(gamma[i+1]);
-
-      std::cout << GridLogIterative << "MPFGMRES: Iteration " << IterationCount
-                << " residual " << cp << " target " << rsq << std::endl;
-
-      if ((i == RestartLength - 1) || (IterationCount == MaxIterations) || (cp <= rsq)) {
-
-        computeSolution(z, psi, i);
-
-        return cp;
-      }
-    }
-
-    assert(0); // Never reached
-    return cp;
-  }
-
-  void arnoldiStep(LinearOperatorBase<FieldD> &LinOp, std::vector<FieldD> &v, std::vector<FieldD> &z, FieldD &w, int iter) {
-
-    FieldF v_f(SinglePrecGrid);
-    FieldF z_f(SinglePrecGrid);
-
-    ChangePrecTimer.Start();
-    precisionChange(v_f, v[iter]);
-    precisionChange(z_f, z[iter]);
-    ChangePrecTimer.Stop();
-
-    PrecTimer.Start();
-    Preconditioner(v_f, z_f);
-    PrecTimer.Stop();
-
-    ChangePrecTimer.Start();
-    precisionChange(z[iter], z_f);
-    ChangePrecTimer.Stop();
-
-    MatrixTimer.Start();
-    LinOp.Op(z[iter], w);
-    MatrixTimer.Stop();
-
-    LinalgTimer.Start();
-    for (int i = 0; i <= iter; ++i) {
-      H(iter, i) = innerProduct(v[i], w);
-      w = w - ComplexD(H(iter, i)) * v[i];
-    }
-
-    H(iter, iter + 1) = sqrt(norm2(w));
-    v[iter + 1] = ComplexD(1. / H(iter, iter + 1)) * w;
-    LinalgTimer.Stop();
-  }
-
-  void qrUpdate(int iter) {
-
-    QrTimer.Start();
-    for (int i = 0; i < iter ; ++i) {
-      auto tmp       = -s[i] * ComplexD(H(iter, i)) + c[i] * ComplexD(H(iter, i + 1));
-      H(iter, i)     = conjugate(c[i]) * ComplexD(H(iter, i)) + conjugate(s[i]) * ComplexD(H(iter, i + 1));
-      H(iter, i + 1) = tmp;
-    }
-
-    // Compute new Givens Rotation
-    auto nu = sqrt(std::norm(H(iter, iter)) + std::norm(H(iter, iter + 1)));
-    c[iter]     = H(iter, iter) / nu;
-    s[iter]     = H(iter, iter + 1) / nu;
-
-    // Apply new Givens rotation
-    H(iter, iter)     = nu;
-    H(iter, iter + 1) = 0.;
-
-    gamma[iter + 1] = -s[iter] * gamma[iter];
-    gamma[iter]     = conjugate(c[iter]) * gamma[iter];
-    QrTimer.Stop();
-  }
-
-  void computeSolution(std::vector<FieldD> const &z, FieldD &psi, int iter) {
-
-    CompSolutionTimer.Start();
-    for (int i = iter; i >= 0; i--) {
-      y[i] = gamma[i];
-      for (int k = i + 1; k <= iter; k++)
-        y[i] = y[i] - ComplexD(H(k, i)) * y[k];
-      y[i] = y[i] / ComplexD(H(i, i));
-    }
-
-    for (int i = 0; i <= iter; i++)
-      psi = psi + z[i] * y[i];
-    CompSolutionTimer.Stop();
-  }
-};
-}
-#endif

Grid/algorithms/iterative/PowerMethod.h

@@ -1,45 +0,0 @@
-#pragma once
-namespace Grid {
-template<class Field> class PowerMethod  
-{ 
- public: 
-
-  template<typename T>  static RealD normalise(T& v) 
-  {
-    RealD nn = norm2(v);
-    nn = sqrt(nn);
-    v = v * (1.0/nn);
-    return nn;
-  }
-
-  RealD operator()(LinearOperatorBase<Field> &HermOp, const Field &src) 
-  { 
-    GridBase *grid = src.Grid(); 
-    
-    // quickly get an idea of the largest eigenvalue to more properly normalize the residuum 
-    RealD evalMaxApprox = 0.0; 
-    auto src_n = src; 
-    auto tmp = src; 
-    const int _MAX_ITER_EST_ = 50; 
-
-    for (int i=0;i<_MAX_ITER_EST_;i++) { 
-      
-      normalise(src_n); 
-      HermOp.HermOp(src_n,tmp); 
-      RealD vnum = real(innerProduct(src_n,tmp)); // HermOp. 
-      RealD vden = norm2(src_n); 
-      RealD na = vnum/vden; 
-      
-      if ( (fabs(evalMaxApprox/na - 1.0) < 0.01) || (i==_MAX_ITER_EST_-1) ) { 
- 	evalMaxApprox = na; 
- 	return evalMaxApprox; 
-      } 
-      evalMaxApprox = na; 
-      std::cout << GridLogMessage << " Approximation of largest eigenvalue: " << evalMaxApprox << std::endl;
-      src_n = tmp;
-    }
-    assert(0);
-    return 0;
-  }
-};
-}

Grid/algorithms/iterative/PrecConjugateResidual.h

@@ -1,119 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/algorithms/iterative/PrecConjugateResidual.h
-
-    Copyright (C) 2015
-
-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 */
-#ifndef GRID_PREC_CONJUGATE_RESIDUAL_H
-#define GRID_PREC_CONJUGATE_RESIDUAL_H
-
-NAMESPACE_BEGIN(Grid);
-
-/////////////////////////////////////////////////////////////
-// Base classes for iterative processes based on operators
-// single input vec, single output vec.
-/////////////////////////////////////////////////////////////
-
-template<class Field> 
-class PrecConjugateResidual : public OperatorFunction<Field> {
-public:                                                
-  RealD   Tolerance;
-  Integer MaxIterations;
-  int verbose;
-  LinearFunction<Field> &Preconditioner;
-
-  PrecConjugateResidual(RealD tol,Integer maxit,LinearFunction<Field> &Prec) : Tolerance(tol), MaxIterations(maxit),      Preconditioner(Prec)
-  { 
-    verbose=1;
-  };
-
-  void operator() (LinearOperatorBase<Field> &Linop,const Field &src, Field &psi){
-
-    RealD a, b, c, d;
-    RealD cp, ssq,rsq;
-      
-    RealD rAr, rAAr, rArp;
-    RealD pAp, pAAp;
-
-    GridBase *grid = src.Grid();
-    Field r(grid),  p(grid), Ap(grid), Ar(grid), z(grid);
-      
-    psi=zero;
-    r  = src;
-    Preconditioner(r,p);
-
-      
-
-    Linop.HermOpAndNorm(p,Ap,pAp,pAAp);
-    Ar=Ap;
-    rAr=pAp;
-    rAAr=pAAp;
-
-    cp =norm2(r);
-    ssq=norm2(src);
-    rsq=Tolerance*Tolerance*ssq;
-
-    if (verbose) std::cout<<GridLogMessage<<"PrecConjugateResidual: iteration " <<0<<" residual "<<cp<< " target"<< rsq<<std::endl;
-
-    for(int k=0;k<MaxIterations;k++){
-
-
-      Preconditioner(Ap,z);
-      RealD rq= real(innerProduct(Ap,z)); 
-
-      a = rAr/rq;
-
-      axpy(psi,a,p,psi);
-      cp = axpy_norm(r,-a,z,r);
-
-      rArp=rAr;
-
-      Linop.HermOpAndNorm(r,Ar,rAr,rAAr);
-
-      b   =rAr/rArp;
- 
-      axpy(p,b,p,r);
-      pAAp=axpy_norm(Ap,b,Ap,Ar);
-	
-      if(verbose) std::cout<<GridLogMessage<<"PrecConjugateResidual: iteration " <<k<<" residual "<<cp<< " target"<< rsq<<std::endl;
-
-      if(cp<rsq) {
-	Linop.HermOp(psi,Ap);
-	axpy(r,-1.0,src,Ap);
-	RealD true_resid = norm2(r)/ssq;
-	std::cout<<GridLogMessage<<"PrecConjugateResidual: Converged on iteration " <<k
-		 << " computed residual "<<sqrt(cp/ssq)
-		 << " true residual "<<sqrt(true_resid)
-		 << " target "       <<Tolerance <<std::endl;
-	return;
-      }
-
-    }
-
-    std::cout<<GridLogMessage<<"PrecConjugateResidual did NOT converge"<<std::endl;
-    assert(0);
-  }
-};
-NAMESPACE_END(Grid);
-#endif

Grid/algorithms/iterative/PrecGeneralisedConjugateResidual.h

@@ -1,243 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/algorithms/iterative/PrecGeneralisedConjugateResidual.h
-
-    Copyright (C) 2015
-
-Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
-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 */
-#ifndef GRID_PREC_GCR_H
-#define GRID_PREC_GCR_H
-
-///////////////////////////////////////////////////////////////////////////////////////////////////////
-//VPGCR Abe and Zhang, 2005.
-//INTERNATIONAL JOURNAL OF NUMERICAL ANALYSIS AND MODELING
-//Computing and Information Volume 2, Number 2, Pages 147-161
-//NB. Likely not original reference since they are focussing on a preconditioner variant.
-//    but VPGCR was nicely written up in their paper
-///////////////////////////////////////////////////////////////////////////////////////////////////////
-NAMESPACE_BEGIN(Grid);
-
-template<class Field>
-class PrecGeneralisedConjugateResidual : public OperatorFunction<Field> {
-public:                                                
-  using OperatorFunction<Field>::operator();
-
-  RealD   Tolerance;
-  Integer MaxIterations;
-  int verbose;
-  int mmax;
-  int nstep;
-  int steps;
-  GridStopWatch PrecTimer;
-  GridStopWatch MatTimer;
-  GridStopWatch LinalgTimer;
-
-  LinearFunction<Field> &Preconditioner;
-
-  PrecGeneralisedConjugateResidual(RealD tol,Integer maxit,LinearFunction<Field> &Prec,int _mmax,int _nstep) : 
-    Tolerance(tol), 
-    MaxIterations(maxit),
-    Preconditioner(Prec),
-    mmax(_mmax),
-    nstep(_nstep)
-  { 
-    verbose=1;
-  };
-
-  void operator() (LinearOperatorBase<Field> &Linop,const Field &src, Field &psi){
-
-    psi=Zero();
-    RealD cp, ssq,rsq;
-    ssq=norm2(src);
-    rsq=Tolerance*Tolerance*ssq;
-      
-    Field r(src.Grid());
-
-    PrecTimer.Reset();
-    MatTimer.Reset();
-    LinalgTimer.Reset();
-
-    GridStopWatch SolverTimer;
-    SolverTimer.Start();
-
-    steps=0;
-    for(int k=0;k<MaxIterations;k++){
-
-      cp=GCRnStep(Linop,src,psi,rsq);
-
-      std::cout<<GridLogMessage<<"VPGCR("<<mmax<<","<<nstep<<") "<< steps <<" steps cp = "<<cp<<std::endl;
-
-      if(cp<rsq) {
-
-	SolverTimer.Stop();
-
-	Linop.HermOp(psi,r);
-	axpy(r,-1.0,src,r);
-	RealD tr = norm2(r);
-	std::cout<<GridLogMessage<<"PrecGeneralisedConjugateResidual: Converged on iteration " <<steps
-		 << " computed residual "<<sqrt(cp/ssq)
-		 << " true residual "    <<sqrt(tr/ssq)
-		 << " target "           <<Tolerance <<std::endl;
-
-	std::cout<<GridLogMessage<<"VPGCR Time elapsed: Total  "<< SolverTimer.Elapsed() <<std::endl;
-	std::cout<<GridLogMessage<<"VPGCR Time elapsed: Precon "<<   PrecTimer.Elapsed() <<std::endl;
-	std::cout<<GridLogMessage<<"VPGCR Time elapsed: Matrix "<<    MatTimer.Elapsed() <<std::endl;
-	std::cout<<GridLogMessage<<"VPGCR Time elapsed: Linalg "<< LinalgTimer.Elapsed() <<std::endl;
-	return;
-      }
-
-    }
-    std::cout<<GridLogMessage<<"Variable Preconditioned GCR did not converge"<<std::endl;
-    assert(0);
-  }
-
-  RealD GCRnStep(LinearOperatorBase<Field> &Linop,const Field &src, Field &psi,RealD rsq){
-
-    RealD cp;
-    RealD a, b;
-    RealD zAz, zAAz;
-    RealD rq;
-
-    GridBase *grid = src.Grid();
-
-    Field r(grid);
-    Field z(grid);
-    Field tmp(grid);
-    Field ttmp(grid);
-    Field Az(grid);
-
-    ////////////////////////////////
-    // history for flexible orthog
-    ////////////////////////////////
-    std::vector<Field> q(mmax,grid);
-    std::vector<Field> p(mmax,grid);
-    std::vector<RealD> qq(mmax);
-      
-
-    //////////////////////////////////
-    // initial guess x0 is taken as nonzero.
-    // r0=src-A x0 = src
-    //////////////////////////////////
-    MatTimer.Start();
-    Linop.HermOpAndNorm(psi,Az,zAz,zAAz); 
-    MatTimer.Stop();
-    
-    LinalgTimer.Start();
-    r=src-Az;
-    LinalgTimer.Stop();
-    
-    /////////////////////
-    // p = Prec(r)
-    /////////////////////
-    PrecTimer.Start();
-    Preconditioner(r,z);
-    PrecTimer.Stop();
-
-    MatTimer.Start();
-    Linop.HermOp(z,tmp); 
-    MatTimer.Stop();
-
-    LinalgTimer.Start();
-    ttmp=tmp;
-    tmp=tmp-r;
-    LinalgTimer.Stop();
-
-    /*
-      std::cout<<GridLogMessage<<r<<std::endl;
-      std::cout<<GridLogMessage<<z<<std::endl;
-      std::cout<<GridLogMessage<<ttmp<<std::endl;
-      std::cout<<GridLogMessage<<tmp<<std::endl;
-    */
-
-    MatTimer.Start();
-    Linop.HermOpAndNorm(z,Az,zAz,zAAz); 
-    MatTimer.Stop();
-
-    LinalgTimer.Start();
-    //p[0],q[0],qq[0] 
-    p[0]= z;
-    q[0]= Az;
-    qq[0]= zAAz;
-    
-    cp =norm2(r);
-    LinalgTimer.Stop();
-
-    for(int k=0;k<nstep;k++){
-
-      steps++;
-
-      int kp     = k+1;
-      int peri_k = k %mmax;
-      int peri_kp= kp%mmax;
-
-      LinalgTimer.Start();
-      rq= real(innerProduct(r,q[peri_k])); // what if rAr not real?
-      a = rq/qq[peri_k];
-
-      axpy(psi,a,p[peri_k],psi);         
-
-      cp = axpy_norm(r,-a,q[peri_k],r);
-      LinalgTimer.Stop();
-
-      if((k==nstep-1)||(cp<rsq)){
-	return cp;
-      }
-
-      std::cout<<GridLogMessage<< " VPGCR_step["<<steps<<"]  resid " <<sqrt(cp/rsq)<<std::endl; 
-
-      PrecTimer.Start();
-      Preconditioner(r,z);// solve Az = r
-      PrecTimer.Stop();
-
-      MatTimer.Start();
-      Linop.HermOpAndNorm(z,Az,zAz,zAAz);
-      Linop.HermOp(z,tmp);
-      MatTimer.Stop();
-
-      LinalgTimer.Start();
-      tmp=tmp-r;
-      std::cout<<GridLogMessage<< " Preconditioner resid " <<sqrt(norm2(tmp)/norm2(r))<<std::endl; 
-
-      q[peri_kp]=Az;
-      p[peri_kp]=z;
-
-      int northog = ((kp)>(mmax-1))?(mmax-1):(kp);  // if more than mmax done, we orthog all mmax history.
-      for(int back=0;back<northog;back++){
-
-	int peri_back=(k-back)%mmax;   	  assert((k-back)>=0);
-
-	b=-real(innerProduct(q[peri_back],Az))/qq[peri_back];
-	p[peri_kp]=p[peri_kp]+b*p[peri_back];
-	q[peri_kp]=q[peri_kp]+b*q[peri_back];
-
-      }
-      qq[peri_kp]=norm2(q[peri_kp]); // could use axpy_norm
-      LinalgTimer.Stop();
-    }
-    assert(0); // never reached
-    return cp;
-  }
-};
-NAMESPACE_END(Grid);
-#endif

Grid/algorithms/iterative/SchurRedBlack.h

@@ -1,486 +0,0 @@
-    /*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/algorithms/iterative/SchurRedBlack.h
-
-    Copyright (C) 2015
-
-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 */
-#ifndef GRID_SCHUR_RED_BLACK_H
-#define GRID_SCHUR_RED_BLACK_H
-
-
-  /*
-   * Red black Schur decomposition
-   *
-   *  M = (Mee Meo) =  (1             0 )   (Mee   0               )  (1 Mee^{-1} Meo)
-   *      (Moe Moo)    (Moe Mee^-1    1 )   (0   Moo-Moe Mee^-1 Meo)  (0   1         )
-   *                =         L                     D                     U
-   *
-   * L^-1 = (1              0 )
-   *        (-MoeMee^{-1}   1 )   
-   * L^{dag} = ( 1       Mee^{-dag} Moe^{dag} )
-   *           ( 0       1                    )
-   * L^{-d}  = ( 1      -Mee^{-dag} Moe^{dag} )
-   *           ( 0       1                    )
-   *
-   * U^-1 = (1   -Mee^{-1} Meo)
-   *        (0    1           )
-   * U^{dag} = ( 1                 0)
-   *           (Meo^dag Mee^{-dag} 1)
-   * U^{-dag} = (  1                 0)
-   *            (-Meo^dag Mee^{-dag} 1)
-   ***********************
-   *     M psi = eta
-   ***********************
-   *Odd
-   * i)                 D_oo psi_o =  L^{-1}  eta_o
-   *                        eta_o' = (D_oo)^dag (eta_o - Moe Mee^{-1} eta_e)
-   *
-   * Wilson:
-   *      (D_oo)^{\dag} D_oo psi_o = (D_oo)^dag L^{-1}  eta_o
-   * Stag:
-   *      D_oo psi_o = L^{-1}  eta =    (eta_o - Moe Mee^{-1} eta_e)
-   *
-   * L^-1 eta_o= (1              0 ) (e
-   *             (-MoeMee^{-1}   1 )   
-   *
-   *Even
-   * ii)  Mee psi_e + Meo psi_o = src_e
-   *
-   *   => sol_e = M_ee^-1 * ( src_e - Meo sol_o )...
-   *
-   * 
-   * TODO: Other options:
-   * 
-   * a) change checkerboards for Schur e<->o
-   *
-   * Left precon by Moo^-1
-   * b) Doo^{dag} M_oo^-dag Moo^-1 Doo psi_0 =  (D_oo)^dag M_oo^-dag Moo^-1 L^{-1}  eta_o
-   *                              eta_o'     = (D_oo)^dag  M_oo^-dag Moo^-1 (eta_o - Moe Mee^{-1} eta_e)
-   *
-   * Right precon by Moo^-1
-   * c) M_oo^-dag Doo^{dag} Doo Moo^-1 phi_0 = M_oo^-dag (D_oo)^dag L^{-1}  eta_o
-   *                              eta_o'     = M_oo^-dag (D_oo)^dag (eta_o - Moe Mee^{-1} eta_e)
-   *                              psi_o = M_oo^-1 phi_o
-   * TODO: Deflation 
-   */
-namespace Grid {
-
-  ///////////////////////////////////////////////////////////////////////////////////////////////////////
-  // Use base class to share code
-  ///////////////////////////////////////////////////////////////////////////////////////////////////////
-  ///////////////////////////////////////////////////////////////////////////////////////////////////////
-  // Take a matrix and form a Red Black solver calling a Herm solver
-  // Use of RB info prevents making SchurRedBlackSolve conform to standard interface
-  ///////////////////////////////////////////////////////////////////////////////////////////////////////
-  template<class Field> class SchurRedBlackBase {
-  protected:
-    typedef CheckerBoardedSparseMatrixBase<Field> Matrix;
-    OperatorFunction<Field> & _HermitianRBSolver;
-    int CBfactorise;
-    bool subGuess;
-    bool useSolnAsInitGuess; // if true user-supplied solution vector is used as initial guess for solver
-  public:
-
-    SchurRedBlackBase(OperatorFunction<Field> &HermitianRBSolver, const bool initSubGuess = false,
-        const bool _solnAsInitGuess = false)  :
-    _HermitianRBSolver(HermitianRBSolver),
-    useSolnAsInitGuess(_solnAsInitGuess)
-    { 
-      CBfactorise = 0;
-      subtractGuess(initSubGuess);
-    };
-    void subtractGuess(const bool initSubGuess)
-    {
-      subGuess = initSubGuess;
-    }
-    bool isSubtractGuess(void)
-    {
-      return subGuess;
-    }
-
-    /////////////////////////////////////////////////////////////
-    // Shared code
-    /////////////////////////////////////////////////////////////
-    void operator() (Matrix & _Matrix,const Field &in, Field &out){
-      ZeroGuesser<Field> guess;
-      (*this)(_Matrix,in,out,guess);
-    }
-    void operator()(Matrix &_Matrix, const std::vector<Field> &in, std::vector<Field> &out) 
-    {
-      ZeroGuesser<Field> guess;
-      (*this)(_Matrix,in,out,guess);
-    }
-
-    template<class Guesser>
-    void operator()(Matrix &_Matrix, const std::vector<Field> &in, std::vector<Field> &out,Guesser &guess) 
-    {
-      GridBase *grid = _Matrix.RedBlackGrid();
-      GridBase *fgrid= _Matrix.Grid();
-      int nblock = in.size();
-
-      std::vector<Field> src_o(nblock,grid);
-      std::vector<Field> sol_o(nblock,grid);
-      
-      std::vector<Field> guess_save;
-
-      Field resid(fgrid);
-      Field tmp(grid);
-
-      ////////////////////////////////////////////////
-      // Prepare RedBlack source
-      ////////////////////////////////////////////////
-      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);
-
-      for(int b=0;b<nblock;b++){
-        if(useSolnAsInitGuess) {
-          pickCheckerboard(Odd, sol_o[b], out[b]);
-        } else {
-          guess(src_o[b],sol_o[b]); 
-        }
-
-	if ( subGuess ) { 
-	  guess_save[b] = sol_o[b];
-	}
-      }
-      //////////////////////////////////////////////////////////////
-      // Call the block solver
-      //////////////////////////////////////////////////////////////
-      std::cout<<GridLogMessage << "SchurRedBlackBase calling the solver for "<<nblock<<" RHS" <<std::endl;
-      RedBlackSolve(_Matrix,src_o,sol_o);
-
-      ////////////////////////////////////////////////
-      // A2A boolean behavioural control & reconstruct other checkerboard
-      ////////////////////////////////////////////////
-      for(int b=0;b<nblock;b++) {
-
-	if (subGuess)   sol_o[b] = sol_o[b] - guess_save[b];
-
-	///////// Needs even source //////////////
-	pickCheckerboard(Even,tmp,in[b]);
-	RedBlackSolution(_Matrix,sol_o[b],tmp,out[b]);
-
-	/////////////////////////////////////////////////
-	// Check unprec residual if possible
-	/////////////////////////////////////////////////
-	if ( ! subGuess ) {
-	  _Matrix.M(out[b],resid); 
-	  resid = resid-in[b];
-	  RealD ns = norm2(in[b]);
-	  RealD nr = norm2(resid);
-	
-	  std::cout<<GridLogMessage<< "SchurRedBlackBase solver true unprec resid["<<b<<"] "<<std::sqrt(nr/ns) << std::endl;
-	} else {
-	  std::cout<<GridLogMessage<< "SchurRedBlackBase Guess subtracted after solve["<<b<<"] " << std::endl;
-	}
-
-      }
-    }
-    template<class Guesser>
-    void operator() (Matrix & _Matrix,const Field &in, Field &out,Guesser &guess){
-
-      // FIXME CGdiagonalMee not implemented virtual function
-      // FIXME use CBfactorise to control schur decomp
-      GridBase *grid = _Matrix.RedBlackGrid();
-      GridBase *fgrid= _Matrix.Grid();
-
-      Field resid(fgrid);
-      Field src_o(grid);
-      Field src_e(grid);
-      Field sol_o(grid);
-
-      ////////////////////////////////////////////////
-      // RedBlack source
-      ////////////////////////////////////////////////
-      RedBlackSource(_Matrix,in,src_e,src_o);
-
-      ////////////////////////////////
-      // Construct the guess
-      ////////////////////////////////
-      if(useSolnAsInitGuess) {
-        pickCheckerboard(Odd, sol_o, out);
-      } else {
-        guess(src_o,sol_o);
-      }
-
-      Field  guess_save(grid);
-      guess_save = sol_o;
-
-      //////////////////////////////////////////////////////////////
-      // Call the red-black solver
-      //////////////////////////////////////////////////////////////
-      RedBlackSolve(_Matrix,src_o,sol_o);
-
-      ////////////////////////////////////////////////
-      // Fionn A2A boolean behavioural control
-      ////////////////////////////////////////////////
-      if (subGuess)      sol_o= sol_o-guess_save;
-
-      ///////////////////////////////////////////////////
-      // RedBlack solution needs the even source
-      ///////////////////////////////////////////////////
-      RedBlackSolution(_Matrix,sol_o,src_e,out);
-
-      // Verify the unprec residual
-      if ( ! subGuess ) {
-        _Matrix.M(out,resid); 
-        resid = resid-in;
-        RealD ns = norm2(in);
-        RealD nr = norm2(resid);
-
-        std::cout<<GridLogMessage << "SchurRedBlackBase solver true unprec resid "<< std::sqrt(nr/ns) << std::endl;
-      } else {
-        std::cout << GridLogMessage << "SchurRedBlackBase Guess subtracted after solve." << std::endl;
-      }
-    }     
-    
-    /////////////////////////////////////////////////////////////
-    // Override in derived. 
-    /////////////////////////////////////////////////////////////
-    virtual void RedBlackSource  (Matrix & _Matrix,const Field &src, Field &src_e,Field &src_o)                =0;
-    virtual void RedBlackSolution(Matrix & _Matrix,const Field &sol_o, const Field &src_e,Field &sol)          =0;
-    virtual void RedBlackSolve   (Matrix & _Matrix,const Field &src_o, Field &sol_o)                           =0;
-    virtual void RedBlackSolve   (Matrix & _Matrix,const std::vector<Field> &src_o,  std::vector<Field> &sol_o)=0;
-
-  };
-
-  template<class Field> class SchurRedBlackStaggeredSolve : public SchurRedBlackBase<Field> {
-  public:
-    typedef CheckerBoardedSparseMatrixBase<Field> Matrix;
-
-    SchurRedBlackStaggeredSolve(OperatorFunction<Field> &HermitianRBSolver, const bool initSubGuess = false,
-        const bool _solnAsInitGuess = false) 
-      :    SchurRedBlackBase<Field> (HermitianRBSolver,initSubGuess,_solnAsInitGuess) 
-    {
-    }
-
-    //////////////////////////////////////////////////////
-    // Override RedBlack specialisation
-    //////////////////////////////////////////////////////
-    virtual void RedBlackSource(Matrix & _Matrix,const Field &src, Field &src_e,Field &src_o)
-    {
-      GridBase *grid = _Matrix.RedBlackGrid();
-      GridBase *fgrid= _Matrix.Grid();
-
-      Field   tmp(grid);
-      Field  Mtmp(grid);
-
-      pickCheckerboard(Even,src_e,src);
-      pickCheckerboard(Odd ,src_o,src);
-
-      /////////////////////////////////////////////////////
-      // src_o = (source_o - Moe MeeInv source_e)
-      /////////////////////////////////////////////////////
-      _Matrix.MooeeInv(src_e,tmp);     assert(  tmp.Checkerboard() ==Even);
-      _Matrix.Meooe   (tmp,Mtmp);      assert( Mtmp.Checkerboard() ==Odd);     
-      tmp=src_o-Mtmp;                  assert(  tmp.Checkerboard() ==Odd);     
-
-      _Matrix.Mooee(tmp,src_o); // Extra factor of "m" in source from dumb choice of matrix norm.
-    }
-    virtual void RedBlackSolution(Matrix & _Matrix,const Field &sol_o, const Field &src_e_c,Field &sol)
-    {
-      GridBase *grid = _Matrix.RedBlackGrid();
-      GridBase *fgrid= _Matrix.Grid();
-
-      Field   tmp(grid);
-      Field   sol_e(grid);
-      Field   src_e(grid);
-
-      src_e = src_e_c; // Const correctness
-
-      ///////////////////////////////////////////////////
-      // sol_e = M_ee^-1 * ( src_e - Meo sol_o )...
-      ///////////////////////////////////////////////////
-      _Matrix.Meooe(sol_o,tmp);        assert(  tmp.Checkerboard()   ==Even);
-      src_e = src_e-tmp;               assert(  src_e.Checkerboard() ==Even);
-      _Matrix.MooeeInv(src_e,sol_e);   assert(  sol_e.Checkerboard() ==Even);
-     
-      setCheckerboard(sol,sol_e); assert(  sol_e.Checkerboard() ==Even);
-      setCheckerboard(sol,sol_o); assert(  sol_o.Checkerboard() ==Odd );
-    }
-    virtual void RedBlackSolve   (Matrix & _Matrix,const Field &src_o, Field &sol_o)
-    {
-      SchurStaggeredOperator<Matrix,Field> _HermOpEO(_Matrix);
-      this->_HermitianRBSolver(_HermOpEO,src_o,sol_o);  assert(sol_o.Checkerboard()==Odd);
-    };
-    virtual void RedBlackSolve   (Matrix & _Matrix,const std::vector<Field> &src_o,  std::vector<Field> &sol_o)
-    {
-      SchurStaggeredOperator<Matrix,Field> _HermOpEO(_Matrix);
-      this->_HermitianRBSolver(_HermOpEO,src_o,sol_o); 
-    }
-  };
-  template<class Field> using SchurRedBlackStagSolve = SchurRedBlackStaggeredSolve<Field>;
-
-  ///////////////////////////////////////////////////////////////////////////////////////////////////////
-  // Site diagonal has Mooee on it.
-  ///////////////////////////////////////////////////////////////////////////////////////////////////////
-  template<class Field> class SchurRedBlackDiagMooeeSolve : public SchurRedBlackBase<Field> {
-  public:
-    typedef CheckerBoardedSparseMatrixBase<Field> Matrix;
-
-    SchurRedBlackDiagMooeeSolve(OperatorFunction<Field> &HermitianRBSolver, const bool initSubGuess = false,
-        const bool _solnAsInitGuess = false)  
-      : SchurRedBlackBase<Field> (HermitianRBSolver,initSubGuess,_solnAsInitGuess) {};
-
-
-    //////////////////////////////////////////////////////
-    // Override RedBlack specialisation
-    //////////////////////////////////////////////////////
-    virtual void RedBlackSource(Matrix & _Matrix,const Field &src, Field &src_e,Field &src_o)
-    {
-      GridBase *grid = _Matrix.RedBlackGrid();
-      GridBase *fgrid= _Matrix.Grid();
-
-      Field   tmp(grid);
-      Field  Mtmp(grid);
-
-      pickCheckerboard(Even,src_e,src);
-      pickCheckerboard(Odd ,src_o,src);
-
-      /////////////////////////////////////////////////////
-      // src_o = Mdag * (source_o - Moe MeeInv source_e)
-      /////////////////////////////////////////////////////
-      _Matrix.MooeeInv(src_e,tmp);     assert(  tmp.Checkerboard() ==Even);
-      _Matrix.Meooe   (tmp,Mtmp);      assert( Mtmp.Checkerboard() ==Odd);     
-      tmp=src_o-Mtmp;                  assert(  tmp.Checkerboard() ==Odd);     
-
-      // get the right MpcDag
-      SchurDiagMooeeOperator<Matrix,Field> _HermOpEO(_Matrix);
-      _HermOpEO.MpcDag(tmp,src_o);     assert(src_o.Checkerboard() ==Odd);       
-
-    }
-    virtual void RedBlackSolution(Matrix & _Matrix,const Field &sol_o, const Field &src_e,Field &sol)
-    {
-      GridBase *grid = _Matrix.RedBlackGrid();
-      GridBase *fgrid= _Matrix.Grid();
-
-      Field   tmp(grid);
-      Field  sol_e(grid);
-      Field  src_e_i(grid);
-      ///////////////////////////////////////////////////
-      // sol_e = M_ee^-1 * ( src_e - Meo sol_o )...
-      ///////////////////////////////////////////////////
-      _Matrix.Meooe(sol_o,tmp);          assert(  tmp.Checkerboard()   ==Even);
-      src_e_i = src_e-tmp;               assert(  src_e_i.Checkerboard() ==Even);
-      _Matrix.MooeeInv(src_e_i,sol_e);   assert(  sol_e.Checkerboard() ==Even);
-     
-      setCheckerboard(sol,sol_e); assert(  sol_e.Checkerboard() ==Even);
-      setCheckerboard(sol,sol_o); assert(  sol_o.Checkerboard() ==Odd );
-    }
-    virtual void RedBlackSolve   (Matrix & _Matrix,const Field &src_o, Field &sol_o)
-    {
-      SchurDiagMooeeOperator<Matrix,Field> _HermOpEO(_Matrix);
-      this->_HermitianRBSolver(_HermOpEO,src_o,sol_o);  assert(sol_o.Checkerboard()==Odd);
-    };
-    virtual void RedBlackSolve   (Matrix & _Matrix,const std::vector<Field> &src_o,  std::vector<Field> &sol_o)
-    {
-      SchurDiagMooeeOperator<Matrix,Field> _HermOpEO(_Matrix);
-      this->_HermitianRBSolver(_HermOpEO,src_o,sol_o); 
-    }
-  };
-
-  ///////////////////////////////////////////////////////////////////////////////////////////////////////
-  // Site diagonal is identity, right preconditioned by Mee^inv
-  // ( 1 - Meo Moo^inv Moe Mee^inv  ) phi =( 1 - Meo Moo^inv Moe Mee^inv  ) Mee psi =  = eta  = eta
-  //=> psi = MeeInv phi
-  ///////////////////////////////////////////////////////////////////////////////////////////////////////
-  template<class Field> class SchurRedBlackDiagTwoSolve : public SchurRedBlackBase<Field> {
-  public:
-    typedef CheckerBoardedSparseMatrixBase<Field> Matrix;
-
-    /////////////////////////////////////////////////////
-    // Wrap the usual normal equations Schur trick
-    /////////////////////////////////////////////////////
-  SchurRedBlackDiagTwoSolve(OperatorFunction<Field> &HermitianRBSolver, const bool initSubGuess = false,
-      const bool _solnAsInitGuess = false)  
-    : SchurRedBlackBase<Field>(HermitianRBSolver,initSubGuess,_solnAsInitGuess) {};
-
-    virtual void RedBlackSource(Matrix & _Matrix,const Field &src, Field &src_e,Field &src_o)
-    {
-      GridBase *grid = _Matrix.RedBlackGrid();
-      GridBase *fgrid= _Matrix.Grid();
-
-      SchurDiagTwoOperator<Matrix,Field> _HermOpEO(_Matrix);
-      
-      Field   tmp(grid);
-      Field  Mtmp(grid);
-
-      pickCheckerboard(Even,src_e,src);
-      pickCheckerboard(Odd ,src_o,src);
-    
-      /////////////////////////////////////////////////////
-      // src_o = Mdag * (source_o - Moe MeeInv source_e)
-      /////////////////////////////////////////////////////
-      _Matrix.MooeeInv(src_e,tmp);     assert(  tmp.Checkerboard() ==Even);
-      _Matrix.Meooe   (tmp,Mtmp);      assert( Mtmp.Checkerboard() ==Odd);     
-      tmp=src_o-Mtmp;                  assert(  tmp.Checkerboard() ==Odd);     
-
-      // get the right MpcDag
-      _HermOpEO.MpcDag(tmp,src_o);     assert(src_o.Checkerboard() ==Odd);       
-    }
-
-    virtual void RedBlackSolution(Matrix & _Matrix,const Field &sol_o, const Field &src_e,Field &sol)
-    {
-      GridBase *grid = _Matrix.RedBlackGrid();
-      GridBase *fgrid= _Matrix.Grid();
-
-      Field   sol_o_i(grid);
-      Field   tmp(grid);
-      Field   sol_e(grid);
-
-      ////////////////////////////////////////////////
-      // MooeeInv due to pecond
-      ////////////////////////////////////////////////
-      _Matrix.MooeeInv(sol_o,tmp);
-      sol_o_i = tmp;
-
-      ///////////////////////////////////////////////////
-      // sol_e = M_ee^-1 * ( src_e - Meo sol_o )...
-      ///////////////////////////////////////////////////
-      _Matrix.Meooe(sol_o_i,tmp);    assert(  tmp.Checkerboard()   ==Even);
-      tmp = src_e-tmp;               assert(  src_e.Checkerboard() ==Even);
-      _Matrix.MooeeInv(tmp,sol_e);   assert(  sol_e.Checkerboard() ==Even);
-     
-      setCheckerboard(sol,sol_e);    assert(  sol_e.Checkerboard() ==Even);
-      setCheckerboard(sol,sol_o_i);  assert(  sol_o_i.Checkerboard() ==Odd );
-    };
-
-    virtual void RedBlackSolve   (Matrix & _Matrix,const Field &src_o, Field &sol_o)
-    {
-      SchurDiagTwoOperator<Matrix,Field> _HermOpEO(_Matrix);
-      this->_HermitianRBSolver(_HermOpEO,src_o,sol_o);
-    };
-    virtual void RedBlackSolve   (Matrix & _Matrix,const std::vector<Field> &src_o,  std::vector<Field> &sol_o)
-    {
-      SchurDiagTwoOperator<Matrix,Field> _HermOpEO(_Matrix);
-      this->_HermitianRBSolver(_HermOpEO,src_o,sol_o); 
-    }
-  };
-}
-#endif

Grid/allocator/AlignedAllocator.h

@@ -1,244 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/AlignedAllocator.h
-
-    Copyright (C) 2015
-
-Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
-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 */
-#ifndef GRID_ALIGNED_ALLOCATOR_H
-#define GRID_ALIGNED_ALLOCATOR_H
-
-#ifdef HAVE_MALLOC_MALLOC_H
-#include <malloc/malloc.h>
-#endif
-#ifdef HAVE_MALLOC_H
-#include <malloc.h>
-#endif
-
-#ifdef HAVE_MM_MALLOC_H
-#include <mm_malloc.h>
-#endif
-
-#define POINTER_CACHE
-#define GRID_ALLOC_ALIGN (2*1024*1024)
-
-NAMESPACE_BEGIN(Grid);
-
-// Move control to configure.ac and Config.h?
-#ifdef POINTER_CACHE
-class PointerCache {
-private:
-
-  static const int Ncache=8;
-  static int victim;
-
-  typedef struct { 
-    void *address;
-    size_t bytes;
-    int valid;
-  } PointerCacheEntry;
-    
-  static PointerCacheEntry Entries[Ncache];
-
-public:
-
-
-  static void *Insert(void *ptr,size_t bytes) ;
-  static void *Lookup(size_t bytes) ;
-
-};
-#endif  
-
-std::string sizeString(size_t bytes);
-
-struct MemoryStats
-{
-  size_t totalAllocated{0}, maxAllocated{0}, 
-    currentlyAllocated{0}, totalFreed{0};
-};
-    
-class MemoryProfiler
-{
-public:
-  static MemoryStats *stats;
-  static bool        debug;
-};
-
-#define memString(bytes) std::to_string(bytes) + " (" + sizeString(bytes) + ")"
-#define profilerDebugPrint						\
-  if (MemoryProfiler::stats)						\
-    {									\
-      auto s = MemoryProfiler::stats;					\
-      std::cout << GridLogDebug << "[Memory debug] Stats " << MemoryProfiler::stats << std::endl; \
-      std::cout << GridLogDebug << "[Memory debug] total  : " << memString(s->totalAllocated) \
-		<< std::endl;						\
-      std::cout << GridLogDebug << "[Memory debug] max    : " << memString(s->maxAllocated) \
-		<< std::endl;						\
-      std::cout << GridLogDebug << "[Memory debug] current: " << memString(s->currentlyAllocated) \
-		<< std::endl;						\
-      std::cout << GridLogDebug << "[Memory debug] freed  : " << memString(s->totalFreed) \
-		<< std::endl;						\
-    }
-
-#define profilerAllocate(bytes)						\
-  if (MemoryProfiler::stats)						\
-    {									\
-      auto s = MemoryProfiler::stats;					\
-      s->totalAllocated     += (bytes);					\
-      s->currentlyAllocated += (bytes);					\
-      s->maxAllocated        = std::max(s->maxAllocated, s->currentlyAllocated); \
-    }									\
-  if (MemoryProfiler::debug)						\
-    {									\
-      std::cout << GridLogDebug << "[Memory debug] allocating " << memString(bytes) << std::endl; \
-      profilerDebugPrint;						\
-    }
-
-#define profilerFree(bytes)						\
-  if (MemoryProfiler::stats)						\
-    {									\
-      auto s = MemoryProfiler::stats;					\
-      s->totalFreed         += (bytes);					\
-      s->currentlyAllocated -= (bytes);					\
-    }									\
-  if (MemoryProfiler::debug)						\
-    {									\
-      std::cout << GridLogDebug << "[Memory debug] freeing " << memString(bytes) << std::endl; \
-      profilerDebugPrint;						\
-    }
-
-void check_huge_pages(void *Buf,uint64_t BYTES);
-
-////////////////////////////////////////////////////////////////////
-// A lattice of something, but assume the something is SIMDized.
-////////////////////////////////////////////////////////////////////
-
-template<typename _Tp>
-class alignedAllocator {
-public: 
-  typedef std::size_t     size_type;
-  typedef std::ptrdiff_t  difference_type;
-  typedef _Tp*       pointer;
-  typedef const _Tp* const_pointer;
-  typedef _Tp&       reference;
-  typedef const _Tp& const_reference;
-  typedef _Tp        value_type;
-
-  template<typename _Tp1>  struct rebind { typedef alignedAllocator<_Tp1> other; };
-  alignedAllocator() throw() { }
-  alignedAllocator(const alignedAllocator&) throw() { }
-  template<typename _Tp1> alignedAllocator(const alignedAllocator<_Tp1>&) throw() { }
-  ~alignedAllocator() throw() { }
-  pointer       address(reference __x)       const { return &__x; }
-  size_type  max_size() const throw() { return size_t(-1) / sizeof(_Tp); }
-
-  pointer allocate(size_type __n, const void* _p= 0)
-  { 
-    size_type bytes = __n*sizeof(_Tp);
-    profilerAllocate(bytes);
-
-
-#ifdef POINTER_CACHE
-    _Tp *ptr = (_Tp *) PointerCache::Lookup(bytes);
-#else
-    pointer ptr = nullptr;
-#endif
-
-#ifdef GRID_NVCC
-    ////////////////////////////////////
-    // Unified (managed) memory
-    ////////////////////////////////////
-    if ( ptr == (_Tp *) NULL ) {
-      auto err = cudaMallocManaged((void **)&ptr,bytes);
-      if( err != cudaSuccess ) {
-	ptr = (_Tp *) NULL;
-	std::cerr << " cudaMallocManaged failed for " << bytes<<" bytes " <<cudaGetErrorString(err)<< std::endl;
-	assert(0);
-      }
-    }
-    assert( ptr != (_Tp *)NULL);
-#else 
-    //////////////////////////////////////////////////////////////////////////////////////////
-    // 2MB align; could make option probably doesn't need configurability
-    //////////////////////////////////////////////////////////////////////////////////////////
-  #ifdef HAVE_MM_MALLOC_H
-    if ( ptr == (_Tp *) NULL ) ptr = (_Tp *) _mm_malloc(bytes,GRID_ALLOC_ALIGN);
-  #else
-    if ( ptr == (_Tp *) NULL ) ptr = (_Tp *) memalign(GRID_ALLOC_ALIGN,bytes);
-  #endif
-    assert( ptr != (_Tp *)NULL);
-
-    //////////////////////////////////////////////////
-    // First touch optimise in threaded loop 
-    //////////////////////////////////////////////////
-    uint64_t *cp = (uint64_t *)ptr;
-    thread_for(n,bytes/sizeof(uint64_t), { // need only one touch per page
-      cp[n]=0;
-    });
-#endif
-    return ptr;
-  }
-
-  void deallocate(pointer __p, size_type __n) { 
-    size_type bytes = __n * sizeof(_Tp);
-
-    profilerFree(bytes);
-
-#ifdef POINTER_CACHE
-    pointer __freeme = (pointer)PointerCache::Insert((void *)__p,bytes);
-#else 
-    pointer __freeme = __p;
-#endif
-
-#ifdef GRID_NVCC
-    if ( __freeme ) cudaFree((void *)__freeme);
-#else 
-  #ifdef HAVE_MM_MALLOC_H
-    if ( __freeme ) _mm_free((void *)__freeme); 
-  #else
-    if ( __freeme ) free((void *)__freeme);
-  #endif
-#endif
-  }
-
-  // FIXME: hack for the copy constructor, eventually it must be avoided
-  void construct(pointer __p, const _Tp& __val) { new((void *)__p) _Tp(__val); };
-  //void construct(pointer __p, const _Tp& __val) { };
-  void construct(pointer __p) { };
-  void destroy(pointer __p) { };
-};
-template<typename _Tp>  inline bool operator==(const alignedAllocator<_Tp>&, const alignedAllocator<_Tp>&){ return true; }
-template<typename _Tp>  inline bool operator!=(const alignedAllocator<_Tp>&, const alignedAllocator<_Tp>&){ return false; }
-
-////////////////////////////////////////////////////////////////////////////////
-// Template typedefs
-////////////////////////////////////////////////////////////////////////////////
-template<class T> using commAllocator = alignedAllocator<T>;
-template<class T> using Vector     = std::vector<T,alignedAllocator<T> >;           
-template<class T> using commVector = std::vector<T,alignedAllocator<T> >;
-template<class T> using Matrix     = std::vector<std::vector<T,alignedAllocator<T> > >;
-
-NAMESPACE_END(Grid);
-
-#endif

Grid/cartesian/Cartesian_base.h

@@ -1,290 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/cartesian/Cartesian_base.h
-
-    Copyright (C) 2015
-
-    Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-    Author: paboyle <paboyle@ph.ed.ac.uk>
-    Author: Guido Cossu <guido.cossu@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 */
-#ifndef GRID_CARTESIAN_BASE_H
-#define GRID_CARTESIAN_BASE_H
-
-NAMESPACE_BEGIN(Grid);
-
-//////////////////////////////////////////////////////////////////////
-// Commicator provides information on the processor grid
-//////////////////////////////////////////////////////////////////////
-//    unsigned long _ndimension;
-//    Coordinate _processors; // processor grid
-//    int              _processor;  // linear processor rank
-//    Coordinate _processor_coor;  // linear processor rank
-//////////////////////////////////////////////////////////////////////
-class GridBase : public CartesianCommunicator , public GridThread {
-
-public:
-  int dummy;
-  // Give Lattice access
-  template<class object> friend class Lattice;
-
-  GridBase(const Coordinate & processor_grid) : CartesianCommunicator(processor_grid) {}; 
-
-  GridBase(const Coordinate & processor_grid,
-	   const CartesianCommunicator &parent,
-	   int &split_rank) 
-    : CartesianCommunicator(processor_grid,parent,split_rank) {};
-
-  GridBase(const Coordinate & processor_grid,
-	   const CartesianCommunicator &parent) 
-    : CartesianCommunicator(processor_grid,parent,dummy) {};
-
-  virtual ~GridBase() = default;
-
-
-  // Physics Grid information.
-  Coordinate _simd_layout;// Which dimensions get relayed out over simd lanes.
-  Coordinate _fdimensions;// (full) Global dimensions of array prior to cb removal
-  Coordinate _gdimensions;// Global dimensions of array after cb removal
-  Coordinate _ldimensions;// local dimensions of array with processor images removed
-  Coordinate _rdimensions;// Reduced local dimensions with simd lane images and processor images removed 
-  Coordinate _ostride;    // Outer stride for each dimension
-  Coordinate _istride;    // Inner stride i.e. within simd lane
-  int _osites;                  // _isites*_osites = product(dimensions).
-  int _isites;
-  int _fsites;                  // _isites*_osites = product(dimensions).
-  int _gsites;
-  Coordinate _slice_block;// subslice information
-  Coordinate _slice_stride;
-  Coordinate _slice_nblock;
-
-  Coordinate _lstart;     // local start of array in gcoors _processor_coor[d]*_ldimensions[d]
-  Coordinate _lend  ;     // local end of array in gcoors   _processor_coor[d]*_ldimensions[d]+_ldimensions_[d]-1
-
-    bool _isCheckerBoarded; 
-
-public:
-
-  ////////////////////////////////////////////////////////////////
-  // Checkerboarding interface is virtual and overridden by 
-  // GridCartesian / GridRedBlackCartesian
-  ////////////////////////////////////////////////////////////////
-  virtual int CheckerBoarded(int dim)=0;
-  virtual int CheckerBoard(const Coordinate &site)=0;
-  virtual int CheckerBoardDestination(int source_cb,int shift,int dim)=0;
-  virtual int CheckerBoardShift(int source_cb,int dim,int shift,int osite)=0;
-  virtual int CheckerBoardShiftForCB(int source_cb,int dim,int shift,int cb)=0;
-  virtual int CheckerBoardFromOindex (int Oindex)=0;
-  virtual int CheckerBoardFromOindexTable (int Oindex)=0;
-
-  //////////////////////////////////////////////////////////////////////////////////////////////
-  // Local layout calculations
-  //////////////////////////////////////////////////////////////////////////////////////////////
-  // These routines are key. Subdivide the linearised cartesian index into
-  //      "inner" index identifying which simd lane of object<vFcomplex> is associated with coord
-  //      "outer" index identifying which element of _odata in class "Lattice" is associated with coord.
-  //
-  // Compared to, say, Blitz++ we simply need to store BOTH an inner stride and an outer
-  // stride per dimension. The cost of evaluating the indexing information is doubled for an n-dimensional
-  // coordinate. Note, however, for data parallel operations the "inner" indexing cost is not paid and all
-  // lanes are operated upon simultaneously.
-  
-  virtual int oIndex(Coordinate &coor)
-  {
-    int idx=0;
-    // Works with either global or local coordinates
-    for(int d=0;d<_ndimension;d++) idx+=_ostride[d]*(coor[d]%_rdimensions[d]);
-    return idx;
-  }
-  virtual int iIndex(Coordinate &lcoor)
-  {
-    int idx=0;
-    for(int d=0;d<_ndimension;d++) idx+=_istride[d]*(lcoor[d]/_rdimensions[d]);
-    return idx;
-  }
-  inline int oIndexReduced(Coordinate &ocoor)
-  {
-    int idx=0; 
-    // ocoor is already reduced so can eliminate the modulo operation
-    // for fast indexing and inline the routine
-    for(int d=0;d<_ndimension;d++) idx+=_ostride[d]*ocoor[d];
-    return idx;
-  }
-  inline void oCoorFromOindex (Coordinate& coor,int Oindex){
-    Lexicographic::CoorFromIndex(coor,Oindex,_rdimensions);
-  }
-
-  inline void InOutCoorToLocalCoor (Coordinate &ocoor, Coordinate &icoor, Coordinate &lcoor) {
-    lcoor.resize(_ndimension);
-    for (int d = 0; d < _ndimension; d++)
-      lcoor[d] = ocoor[d] + _rdimensions[d] * icoor[d];
-  }
-
-  //////////////////////////////////////////////////////////
-  // SIMD lane addressing
-  //////////////////////////////////////////////////////////
-  inline void iCoorFromIindex(Coordinate &coor,int lane)
-  {
-    Lexicographic::CoorFromIndex(coor,lane,_simd_layout);
-  }
-
-  inline int PermuteDim(int dimension){
-    return _simd_layout[dimension]>1;
-  }
-  inline int PermuteType(int dimension){
-    int permute_type=0;
-    //
-    // Best way to encode this would be to present a mask 
-    // for which simd dimensions are rotated, and the rotation
-    // size. If there is only one simd dimension rotated, this is just 
-    // a permute. 
-    //
-    // Cases: PermuteType == 1,2,4,8
-    // Distance should be either 0,1,2..
-    //
-    if ( _simd_layout[dimension] > 2 ) { 
-      for(int d=0;d<_ndimension;d++){
-	if ( d != dimension ) assert ( (_simd_layout[d]==1)  );
-      }
-      permute_type = RotateBit; // How to specify distance; this is not just direction.
-      return permute_type;
-    }
-
-    for(int d=_ndimension-1;d>dimension;d--){
-      if (_simd_layout[d]>1 ) permute_type++;
-    }
-    return permute_type;
-  }
-  ////////////////////////////////////////////////////////////////
-  // Array sizing queries
-  ////////////////////////////////////////////////////////////////
-
-  inline int iSites(void) const { return _isites; };
-  inline int Nsimd(void)  const { return _isites; };// Synonymous with iSites
-  inline int oSites(void) const { return _osites; };
-  inline int lSites(void) const { return _isites*_osites; }; 
-  inline int gSites(void) const { return _isites*_osites*_Nprocessors; }; 
-  inline int Nd    (void) const { return _ndimension;};
-
-  inline const Coordinate LocalStarts(void)             { return _lstart;    };
-  inline const Coordinate &FullDimensions(void)         { return _fdimensions;};
-  inline const Coordinate &GlobalDimensions(void)       { return _gdimensions;};
-  inline const Coordinate &LocalDimensions(void)        { return _ldimensions;};
-  inline const Coordinate &VirtualLocalDimensions(void) { return _ldimensions;};
-
-  ////////////////////////////////////////////////////////////////
-  // Utility to print the full decomposition details 
-  ////////////////////////////////////////////////////////////////
-
-  void show_decomposition(){
-    std::cout << GridLogMessage << "\tFull Dimensions    : " << _fdimensions << std::endl;
-    std::cout << GridLogMessage << "\tSIMD layout        : " << _simd_layout << std::endl;
-    std::cout << GridLogMessage << "\tGlobal Dimensions  : " << _gdimensions << std::endl;
-    std::cout << GridLogMessage << "\tLocal Dimensions   : " << _ldimensions << std::endl;
-    std::cout << GridLogMessage << "\tReduced Dimensions : " << _rdimensions << std::endl;
-    std::cout << GridLogMessage << "\tOuter strides      : " << _ostride << std::endl;
-    std::cout << GridLogMessage << "\tInner strides      : " << _istride << std::endl;
-    std::cout << GridLogMessage << "\tiSites             : " << _isites << std::endl;
-    std::cout << GridLogMessage << "\toSites             : " << _osites << std::endl;
-    std::cout << GridLogMessage << "\tlSites             : " << lSites() << std::endl;        
-    std::cout << GridLogMessage << "\tgSites             : " << gSites() << std::endl;
-    std::cout << GridLogMessage << "\tNd                 : " << _ndimension << std::endl;             
-  } 
-
-  ////////////////////////////////////////////////////////////////
-  // Global addressing
-  ////////////////////////////////////////////////////////////////
-  void GlobalIndexToGlobalCoor(int gidx,Coordinate &gcoor){
-    assert(gidx< gSites());
-    Lexicographic::CoorFromIndex(gcoor,gidx,_gdimensions);
-  }
-  void LocalIndexToLocalCoor(int lidx,Coordinate &lcoor){
-    assert(lidx<lSites());
-    Lexicographic::CoorFromIndex(lcoor,lidx,_ldimensions);
-  }
-  void GlobalCoorToGlobalIndex(const Coordinate & gcoor,int & gidx){
-    gidx=0;
-    int mult=1;
-    for(int mu=0;mu<_ndimension;mu++) {
-      gidx+=mult*gcoor[mu];
-      mult*=_gdimensions[mu];
-    }
-  }
-  void GlobalCoorToProcessorCoorLocalCoor(Coordinate &pcoor,Coordinate &lcoor,const Coordinate &gcoor)
-  {
-    pcoor.resize(_ndimension);
-    lcoor.resize(_ndimension);
-    for(int mu=0;mu<_ndimension;mu++){
-      int _fld  = _fdimensions[mu]/_processors[mu];
-      pcoor[mu] = gcoor[mu]/_fld;
-      lcoor[mu] = gcoor[mu]%_fld;
-    }
-  }
-  void GlobalCoorToRankIndex(int &rank, int &o_idx, int &i_idx ,const Coordinate &gcoor)
-  {
-    Coordinate pcoor;
-    Coordinate lcoor;
-    GlobalCoorToProcessorCoorLocalCoor(pcoor,lcoor,gcoor);
-    rank = RankFromProcessorCoor(pcoor);
-    /*
-      Coordinate cblcoor(lcoor);
-      for(int d=0;d<cblcoor.size();d++){
-      if( this->CheckerBoarded(d) ) {
-      cblcoor[d] = lcoor[d]/2;
-      }
-      }
-    */
-    i_idx= iIndex(lcoor);
-    o_idx= oIndex(lcoor);
-  }
-
-  void RankIndexToGlobalCoor(int rank, int o_idx, int i_idx , Coordinate &gcoor)
-  {
-    gcoor.resize(_ndimension);
-    Coordinate coor(_ndimension);
-
-    ProcessorCoorFromRank(rank,coor);
-    for(int mu=0;mu<_ndimension;mu++) gcoor[mu] = _ldimensions[mu]*coor[mu];
-
-    iCoorFromIindex(coor,i_idx);
-    for(int mu=0;mu<_ndimension;mu++) gcoor[mu] += _rdimensions[mu]*coor[mu];
-
-    oCoorFromOindex (coor,o_idx);
-    for(int mu=0;mu<_ndimension;mu++) gcoor[mu] += coor[mu];
-      
-  }
-  void RankIndexCbToFullGlobalCoor(int rank, int o_idx, int i_idx, int cb,Coordinate &fcoor)
-  {
-    RankIndexToGlobalCoor(rank,o_idx,i_idx ,fcoor);
-    if(CheckerBoarded(0)){
-      fcoor[0] = fcoor[0]*2+cb;
-    }
-  }
-  void ProcessorCoorLocalCoorToGlobalCoor(Coordinate &Pcoor,Coordinate &Lcoor,Coordinate &gcoor)
-  {
-    gcoor.resize(_ndimension);
-    for(int mu=0;mu<_ndimension;mu++) gcoor[mu] = Pcoor[mu]*_ldimensions[mu]+Lcoor[mu];
-  }
-};
-
-NAMESPACE_END(Grid);
-#endif

Grid/cartesian/Cartesian_full.h

@@ -1,174 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/cartesian/Cartesian_full.h
-
-    Copyright (C) 2015
-
-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 */
-#ifndef GRID_CARTESIAN_FULL_H
-#define GRID_CARTESIAN_FULL_H
-
-NAMESPACE_BEGIN(Grid);
-    
-/////////////////////////////////////////////////////////////////////////////////////////
-// Grid Support.
-/////////////////////////////////////////////////////////////////////////////////////////
-
-class GridCartesian: public GridBase {
-
-public:
-  int dummy;
-  virtual int  CheckerBoardFromOindexTable (int Oindex) {
-    return 0;
-  }
-  virtual int  CheckerBoardFromOindex (int Oindex)
-  {
-    return 0;
-  }
-  virtual int CheckerBoarded(int dim){
-    return 0;
-  }
-  virtual int CheckerBoard(const Coordinate &site){
-    return 0;
-  }
-  virtual int CheckerBoardDestination(int cb,int shift,int dim){
-    return 0;
-  }
-  virtual int CheckerBoardShiftForCB(int source_cb,int dim,int shift, int ocb){
-    return shift;
-  }
-  virtual int CheckerBoardShift(int source_cb,int dim,int shift, int osite){
-    return shift;
-  }
-  /////////////////////////////////////////////////////////////////////////
-  // Constructor takes a parent grid and possibly subdivides communicator.
-  /////////////////////////////////////////////////////////////////////////
-  GridCartesian(const Coordinate &dimensions,
-		const Coordinate &simd_layout,
-		const Coordinate &processor_grid,
-		const GridCartesian &parent) : GridBase(processor_grid,parent,dummy)
-  {
-    Init(dimensions,simd_layout,processor_grid);
-  }
-  GridCartesian(const Coordinate &dimensions,
-		const Coordinate &simd_layout,
-		const Coordinate &processor_grid,
-		const GridCartesian &parent,int &split_rank) : GridBase(processor_grid,parent,split_rank)
-  {
-    Init(dimensions,simd_layout,processor_grid);
-  }
-  /////////////////////////////////////////////////////////////////////////
-  // Construct from comm world
-  /////////////////////////////////////////////////////////////////////////
-  GridCartesian(const Coordinate &dimensions,
-		const Coordinate &simd_layout,
-		const Coordinate &processor_grid) : GridBase(processor_grid)
-  {
-    Init(dimensions,simd_layout,processor_grid);
-  }
-
-  virtual ~GridCartesian() = default;
-
-  void Init(const Coordinate &dimensions,
-	    const Coordinate &simd_layout,
-	    const Coordinate &processor_grid)
-  {
-    ///////////////////////
-    // Grid information
-    ///////////////////////
-      _isCheckerBoarded = false;
-    _ndimension = dimensions.size();
-
-    _fdimensions.resize(_ndimension);
-    _gdimensions.resize(_ndimension);
-    _ldimensions.resize(_ndimension);
-    _rdimensions.resize(_ndimension);
-    _simd_layout.resize(_ndimension);
-    _lstart.resize(_ndimension);
-    _lend.resize(_ndimension);
-
-    _ostride.resize(_ndimension);
-    _istride.resize(_ndimension);
-
-    _fsites = _gsites = _osites = _isites = 1;
-
-    for (int d = 0; d < _ndimension; d++)
-      {
-        _fdimensions[d] = dimensions[d];   // Global dimensions
-        _gdimensions[d] = _fdimensions[d]; // Global dimensions
-        _simd_layout[d] = simd_layout[d];
-        _fsites = _fsites * _fdimensions[d];
-        _gsites = _gsites * _gdimensions[d];
-
-        // Use a reduced simd grid
-        _ldimensions[d] = _gdimensions[d] / _processors[d]; //local dimensions
-        //std::cout << _ldimensions[d] << "  " << _gdimensions[d] << "  " << _processors[d] << std::endl;
-        assert(_ldimensions[d] * _processors[d] == _gdimensions[d]);
-
-        _rdimensions[d] = _ldimensions[d] / _simd_layout[d]; //overdecomposition
-        assert(_rdimensions[d] * _simd_layout[d] == _ldimensions[d]);
-
-        _lstart[d] = _processor_coor[d] * _ldimensions[d];
-        _lend[d] = _processor_coor[d] * _ldimensions[d] + _ldimensions[d] - 1;
-        _osites *= _rdimensions[d];
-        _isites *= _simd_layout[d];
-
-        // Addressing support
-        if (d == 0)
-	  {
-	    _ostride[d] = 1;
-	    _istride[d] = 1;
-	  }
-        else
-	  {
-	    _ostride[d] = _ostride[d - 1] * _rdimensions[d - 1];
-	    _istride[d] = _istride[d - 1] * _simd_layout[d - 1];
-	  }
-      }
-
-    ///////////////////////
-    // subplane information
-    ///////////////////////
-    _slice_block.resize(_ndimension);
-    _slice_stride.resize(_ndimension);
-    _slice_nblock.resize(_ndimension);
-
-    int block = 1;
-    int nblock = 1;
-    for (int d = 0; d < _ndimension; d++)
-      nblock *= _rdimensions[d];
-
-    for (int d = 0; d < _ndimension; d++)
-      {
-        nblock /= _rdimensions[d];
-        _slice_block[d] = block;
-        _slice_stride[d] = _ostride[d] * _rdimensions[d];
-        _slice_nblock[d] = nblock;
-        block = block * _rdimensions[d];
-      }
-  };
-
-};
-
-NAMESPACE_END(Grid);
-#endif

Grid/cartesian/Cartesian_red_black.h

@@ -1,289 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/cartesian/Cartesian_red_black.h
-
-    Copyright (C) 2015
-
-Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
-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 */
-#ifndef GRID_CARTESIAN_RED_BLACK_H
-#define GRID_CARTESIAN_RED_BLACK_H
-
-NAMESPACE_BEGIN(Grid);
-
-static const int CbRed  =0;
-static const int CbBlack=1;
-static const int Even   =CbRed;
-static const int Odd    =CbBlack;
-    
-// Specialise this for red black grids storing half the data like a chess board.
-class GridRedBlackCartesian : public GridBase
-{
-public:
-  Coordinate _checker_dim_mask;
-  int              _checker_dim;
-  std::vector<int> _checker_board;
-
-  virtual int CheckerBoarded(int dim){
-    if( dim==_checker_dim) return 1;
-    else return 0;
-  }
-  virtual int CheckerBoard(const Coordinate &site){
-    int linear=0;
-    assert(site.size()==_ndimension);
-    for(int d=0;d<_ndimension;d++){ 
-      if(_checker_dim_mask[d])
-	linear=linear+site[d];
-    }
-    return (linear&0x1);
-  }
-
-  // Depending on the cb of site, we toggle source cb.
-  // for block #b, element #e = (b, e)
-  // we need 
-  virtual int CheckerBoardShiftForCB(int source_cb,int dim,int shift,int ocb){
-    if(dim != _checker_dim) return shift;
-
-    int fulldim =_fdimensions[dim];
-    shift = (shift+fulldim)%fulldim;
-
-    // Probably faster with table lookup;
-    // or by looping over x,y,z and multiply rather than computing checkerboard.
-	  
-    if ( (source_cb+ocb)&1 ) {
-      return (shift)/2;
-    } else {
-      return (shift+1)/2;
-    }
-  }
-  virtual int  CheckerBoardFromOindexTable (int Oindex) {
-    return _checker_board[Oindex];
-  }
-  virtual int  CheckerBoardFromOindex (int Oindex)
-  {
-    Coordinate ocoor;
-    oCoorFromOindex(ocoor,Oindex);
-    return CheckerBoard(ocoor);
-  }
-  virtual int CheckerBoardShift(int source_cb,int dim,int shift,int osite){
-
-    if(dim != _checker_dim) return shift;
-
-    int ocb=CheckerBoardFromOindex(osite);
-      
-    return CheckerBoardShiftForCB(source_cb,dim,shift,ocb);
-  }
-    
-  virtual int CheckerBoardDestination(int source_cb,int shift,int dim){
-    if ( _checker_dim_mask[dim]  ) {
-      // If _fdimensions[checker_dim] is odd, then shifting by 1 in other dims
-      // does NOT cause a parity hop.
-      int add=(dim==_checker_dim) ? 0 : _fdimensions[_checker_dim];
-      if ( (shift+add) &0x1) {
-	return 1-source_cb;
-      } else {
-	return source_cb;
-      }
-    } else {
-      return source_cb;
-
-    }
-  };
-
-  ////////////////////////////////////////////////////////////
-  // Create Redblack from original grid; require full grid pointer ?
-  ////////////////////////////////////////////////////////////
-  GridRedBlackCartesian(const GridBase *base) : GridBase(base->_processors,*base)
-  {
-    int dims = base->_ndimension;
-    Coordinate checker_dim_mask(dims,1);
-    int checker_dim = 0;
-    Init(base->_fdimensions,base->_simd_layout,base->_processors,checker_dim_mask,checker_dim);
-  };
-
-  ////////////////////////////////////////////////////////////
-  // Create redblack from original grid, with non-trivial checker dim mask
-  ////////////////////////////////////////////////////////////
-  GridRedBlackCartesian(const GridBase *base,
-			const Coordinate &checker_dim_mask,
-			int checker_dim
-			) :  GridBase(base->_processors,*base) 
-  {
-    Init(base->_fdimensions,base->_simd_layout,base->_processors,checker_dim_mask,checker_dim)  ;
-  }
-
-  virtual ~GridRedBlackCartesian() = default;
-
-  void Init(const Coordinate &dimensions,
-	    const Coordinate &simd_layout,
-	    const Coordinate &processor_grid,
-	    const Coordinate &checker_dim_mask,
-	    int checker_dim)
-  {
-
-      _isCheckerBoarded = true;
-    _checker_dim = checker_dim;
-    assert(checker_dim_mask[checker_dim] == 1);
-    _ndimension = dimensions.size();
-    assert(checker_dim_mask.size() == _ndimension);
-    assert(processor_grid.size() == _ndimension);
-    assert(simd_layout.size() == _ndimension);
-
-    _fdimensions.resize(_ndimension);
-    _gdimensions.resize(_ndimension);
-    _ldimensions.resize(_ndimension);
-    _rdimensions.resize(_ndimension);
-    _simd_layout.resize(_ndimension);
-    _lstart.resize(_ndimension);
-    _lend.resize(_ndimension);
-
-    _ostride.resize(_ndimension);
-    _istride.resize(_ndimension);
-
-    _fsites = _gsites = _osites = _isites = 1;
-
-    _checker_dim_mask = checker_dim_mask;
-
-    for (int d = 0; d < _ndimension; d++)
-      {
-        _fdimensions[d] = dimensions[d];
-        _gdimensions[d] = _fdimensions[d];
-        _fsites = _fsites * _fdimensions[d];
-        _gsites = _gsites * _gdimensions[d];
-
-        if (d == _checker_dim)
-	  {
-	    assert((_gdimensions[d] & 0x1) == 0);
-	    _gdimensions[d] = _gdimensions[d] / 2; // Remove a checkerboard
-	    _gsites /= 2;
-	  }
-        _ldimensions[d] = _gdimensions[d] / _processors[d];
-        assert(_ldimensions[d] * _processors[d] == _gdimensions[d]);
-        _lstart[d] = _processor_coor[d] * _ldimensions[d];
-        _lend[d] = _processor_coor[d] * _ldimensions[d] + _ldimensions[d] - 1;
-
-        // Use a reduced simd grid
-        _simd_layout[d] = simd_layout[d];
-        _rdimensions[d] = _ldimensions[d] / _simd_layout[d]; // this is not checking if this is integer
-        assert(_rdimensions[d] * _simd_layout[d] == _ldimensions[d]);
-        assert(_rdimensions[d] > 0);
-
-        // all elements of a simd vector must have same checkerboard.
-        // If Ls vectorised, this must still be the case; e.g. dwf rb5d
-        if (_simd_layout[d] > 1)
-	  {
-	    if (checker_dim_mask[d])
-	      {
-		assert((_rdimensions[d] & 0x1) == 0);
-	      }
-	  }
-
-        _osites *= _rdimensions[d];
-        _isites *= _simd_layout[d];
-
-        // Addressing support
-        if (d == 0)
-	  {
-	    _ostride[d] = 1;
-	    _istride[d] = 1;
-	  }
-        else
-	  {
-	    _ostride[d] = _ostride[d - 1] * _rdimensions[d - 1];
-	    _istride[d] = _istride[d - 1] * _simd_layout[d - 1];
-	  }
-      }
-
-    ////////////////////////////////////////////////////////////////////////////////////////////
-    // subplane information
-    ////////////////////////////////////////////////////////////////////////////////////////////
-    _slice_block.resize(_ndimension);
-    _slice_stride.resize(_ndimension);
-    _slice_nblock.resize(_ndimension);
-
-    int block = 1;
-    int nblock = 1;
-    for (int d = 0; d < _ndimension; d++)
-      nblock *= _rdimensions[d];
-
-    for (int d = 0; d < _ndimension; d++)
-      {
-        nblock /= _rdimensions[d];
-        _slice_block[d] = block;
-        _slice_stride[d] = _ostride[d] * _rdimensions[d];
-        _slice_nblock[d] = nblock;
-        block = block * _rdimensions[d];
-      }
-
-    ////////////////////////////////////////////////
-    // Create a checkerboard lookup table
-    ////////////////////////////////////////////////
-    int rvol = 1;
-    for (int d = 0; d < _ndimension; d++)
-      {
-        rvol = rvol * _rdimensions[d];
-      }
-    _checker_board.resize(rvol);
-    for (int osite = 0; osite < _osites; osite++)
-      {
-        _checker_board[osite] = CheckerBoardFromOindex(osite);
-      }
-  };
-
-protected:
-  virtual int oIndex(Coordinate &coor)
-  {
-    int idx = 0;
-    for (int d = 0; d < _ndimension; d++)
-      {
-        if (d == _checker_dim)
-	  {
-	    idx += _ostride[d] * ((coor[d] / 2) % _rdimensions[d]);
-	  }
-        else
-	  {
-	    idx += _ostride[d] * (coor[d] % _rdimensions[d]);
-	  }
-      }
-    return idx;
-  };
-
-  virtual int iIndex(Coordinate &lcoor)
-  {
-    int idx = 0;
-    for (int d = 0; d < _ndimension; d++)
-      {
-        if (d == _checker_dim)
-	  {
-	    idx += _istride[d] * (lcoor[d] / (2 * _rdimensions[d]));
-	  }
-        else
-	  {
-	    idx += _istride[d] * (lcoor[d] / _rdimensions[d]);
-	  }
-      }
-    return idx;
-  }
-};
-NAMESPACE_END(Grid);
-#endif

Grid/lattice/Lattice_ET.h

@@ -1,407 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/lattice/Lattice_ET.h
-
-Copyright (C) 2015
-
-Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: neo <cossu@post.kek.jp>
-
-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_LATTICE_ET_H
-#define GRID_LATTICE_ET_H
-
-#include <iostream>
-#include <tuple>
-#include <typeinfo>
-#include <vector>
-
-NAMESPACE_BEGIN(Grid);
-
-////////////////////////////////////////////////////
-// Predicated where support
-////////////////////////////////////////////////////
-template <class iobj, class vobj, class robj>
-accelerator_inline vobj predicatedWhere(const iobj &predicate, const vobj &iftrue,
-                            const robj &iffalse) {
-  typename std::remove_const<vobj>::type ret;
-
-  typedef typename vobj::scalar_object scalar_object;
-  typedef typename vobj::scalar_type scalar_type;
-  typedef typename vobj::vector_type vector_type;
-
-  const int Nsimd = vobj::vector_type::Nsimd();
-
-  ExtractBuffer<Integer> mask(Nsimd);
-  ExtractBuffer<scalar_object> truevals(Nsimd);
-  ExtractBuffer<scalar_object> falsevals(Nsimd);
-
-  extract(iftrue, truevals);
-  extract(iffalse, falsevals);
-  extract<vInteger, Integer>(TensorRemove(predicate), mask);
-
-  for (int s = 0; s < Nsimd; s++) {
-    if (mask[s]) falsevals[s] = truevals[s];
-  }
-
-  merge(ret, falsevals);
-  return ret;
-}
-
-/////////////////////////////////////////////////////
-//Specialization of getVectorType for lattices
-/////////////////////////////////////////////////////
-template<typename T>
-struct getVectorType<Lattice<T> >{
-  typedef typename Lattice<T>::vector_object type;
-};
-
-////////////////////////////////////////////
-//--  recursive evaluation of expressions; --
-// handle leaves of syntax tree
-///////////////////////////////////////////////////
-template<class sobj> accelerator_inline 
-sobj eval(const uint64_t ss, const sobj &arg)
-{
-  return arg;
-}
-
-template <class lobj> accelerator_inline 
-const lobj & eval(const uint64_t ss, const LatticeView<lobj> &arg) 
-{
-  return arg[ss];
-}
-template <class lobj> accelerator_inline 
-const lobj & eval(const uint64_t ss, const Lattice<lobj> &arg) 
-{
-  auto view = arg.View();
-  return view[ss];
-}
-
-///////////////////////////////////////////////////
-// handle nodes in syntax tree- eval one operand
-///////////////////////////////////////////////////
-template <typename Op, typename T1> accelerator_inline 
-auto eval(const uint64_t ss, const LatticeUnaryExpression<Op, T1> &expr)  
-  -> decltype(expr.op.func( eval(ss, expr.arg1)))
-{
-  return expr.op.func( eval(ss, expr.arg1) );
-}
-///////////////////////
-// eval two operands
-///////////////////////
-template <typename Op, typename T1, typename T2> accelerator_inline
-auto eval(const uint64_t ss, const LatticeBinaryExpression<Op, T1, T2> &expr)  
-  -> decltype(expr.op.func( eval(ss,expr.arg1),eval(ss,expr.arg2)))
-{
-  return expr.op.func( eval(ss,expr.arg1), eval(ss,expr.arg2) );
-}
-///////////////////////
-// eval three operands
-///////////////////////
-template <typename Op, typename T1, typename T2, typename T3> accelerator_inline
-auto eval(const uint64_t ss, const LatticeTrinaryExpression<Op, T1, T2, T3> &expr)  
-  -> decltype(expr.op.func(eval(ss, expr.arg1), eval(ss, expr.arg2), eval(ss, expr.arg3)))
-{
-  return expr.op.func(eval(ss, expr.arg1), eval(ss, expr.arg2), eval(ss, expr.arg3));
-}
-
-//////////////////////////////////////////////////////////////////////////
-// Obtain the grid from an expression, ensuring conformable. This must follow a
-// tree recursion; must retain grid pointer in the LatticeView class which sucks
-// Use a different method, and make it void *.
-// Perhaps a conformable method.
-//////////////////////////////////////////////////////////////////////////
-template <class T1,typename std::enable_if<is_lattice<T1>::value, T1>::type * = nullptr>
-accelerator_inline void GridFromExpression(GridBase *&grid, const T1 &lat)  // Lattice leaf
-{
-  lat.Conformable(grid);
-}
-
-template <class T1,typename std::enable_if<!is_lattice<T1>::value, T1>::type * = nullptr>
-accelerator_inline 
-void GridFromExpression(GridBase *&grid,const T1 &notlat)  // non-lattice leaf
-{}
-
-template <typename Op, typename T1>
-accelerator_inline 
-void GridFromExpression(GridBase *&grid,const LatticeUnaryExpression<Op, T1> &expr) 
-{
-  GridFromExpression(grid, expr.arg1);  // recurse
-}
-
-template <typename Op, typename T1, typename T2>
-accelerator_inline 
-void GridFromExpression(GridBase *&grid, const LatticeBinaryExpression<Op, T1, T2> &expr) 
-{
-  GridFromExpression(grid, expr.arg1);  // recurse
-  GridFromExpression(grid, expr.arg2);
-}
-template <typename Op, typename T1, typename T2, typename T3>
-accelerator_inline 
-void GridFromExpression(GridBase *&grid, const LatticeTrinaryExpression<Op, T1, T2, T3> &expr) 
-{
-  GridFromExpression(grid, expr.arg1);  // recurse
-  GridFromExpression(grid, expr.arg2);  // recurse
-  GridFromExpression(grid, expr.arg3);  // recurse
-}
-
-//////////////////////////////////////////////////////////////////////////
-// Obtain the CB from an expression, ensuring conformable. This must follow a
-// tree recursion
-//////////////////////////////////////////////////////////////////////////
-template <class T1,typename std::enable_if<is_lattice<T1>::value, T1>::type * = nullptr>
-inline void CBFromExpression(int &cb, const T1 &lat)  // Lattice leaf
-{
-  if ((cb == Odd) || (cb == Even)) {
-    assert(cb == lat.Checkerboard());
-  }
-  cb = lat.Checkerboard();
-}
-template <class T1,typename std::enable_if<!is_lattice<T1>::value, T1>::type * = nullptr>
-inline void CBFromExpression(int &cb, const T1 &notlat)  // non-lattice leaf
-{
-}
-
-template <typename Op, typename T1> inline 
-void CBFromExpression(int &cb,const LatticeUnaryExpression<Op, T1> &expr) 
-{
-  CBFromExpression(cb, expr.arg1);  // recurse AST
-}
-
-template <typename Op, typename T1, typename T2> inline 
-void CBFromExpression(int &cb,const LatticeBinaryExpression<Op, T1, T2> &expr) 
-{
-  CBFromExpression(cb, expr.arg1);  // recurse AST
-  CBFromExpression(cb, expr.arg2);  // recurse AST
-}
-template <typename Op, typename T1, typename T2, typename T3>
-inline void CBFromExpression(int &cb, const LatticeTrinaryExpression<Op, T1, T2, T3> &expr) 
-{
-  CBFromExpression(cb, expr.arg1);  // recurse AST
-  CBFromExpression(cb, expr.arg2);  // recurse AST
-  CBFromExpression(cb, expr.arg3);  // recurse AST
-}
-
-////////////////////////////////////////////
-// Unary operators and funcs
-////////////////////////////////////////////
-#define GridUnopClass(name, ret)					\
-  template <class arg>							\
-  struct name {								\
-    static auto accelerator_inline func(const arg a) -> decltype(ret) { return ret; } \
-  };
-
-GridUnopClass(UnarySub, -a);
-GridUnopClass(UnaryNot, Not(a));
-GridUnopClass(UnaryAdj, adj(a));
-GridUnopClass(UnaryConj, conjugate(a));
-GridUnopClass(UnaryTrace, trace(a));
-GridUnopClass(UnaryTranspose, transpose(a));
-GridUnopClass(UnaryTa, Ta(a));
-GridUnopClass(UnaryProjectOnGroup, ProjectOnGroup(a));
-GridUnopClass(UnaryReal, real(a));
-GridUnopClass(UnaryImag, imag(a));
-GridUnopClass(UnaryToReal, toReal(a));
-GridUnopClass(UnaryToComplex, toComplex(a));
-GridUnopClass(UnaryTimesI, timesI(a));
-GridUnopClass(UnaryTimesMinusI, timesMinusI(a));
-GridUnopClass(UnaryAbs, abs(a));
-GridUnopClass(UnarySqrt, sqrt(a));
-GridUnopClass(UnaryRsqrt, rsqrt(a));
-GridUnopClass(UnarySin, sin(a));
-GridUnopClass(UnaryCos, cos(a));
-GridUnopClass(UnaryAsin, asin(a));
-GridUnopClass(UnaryAcos, acos(a));
-GridUnopClass(UnaryLog, log(a));
-GridUnopClass(UnaryExp, exp(a));
-
-////////////////////////////////////////////
-// Binary operators
-////////////////////////////////////////////
-#define GridBinOpClass(name, combination)			\
-  template <class left, class right>				\
-  struct name {							\
-    static auto accelerator_inline				\
-    func(const left &lhs, const right &rhs)			\
-      -> decltype(combination) const				\
-    {								\
-      return combination;					\
-    }								\
-  };
-
-GridBinOpClass(BinaryAdd, lhs + rhs);
-GridBinOpClass(BinarySub, lhs - rhs);
-GridBinOpClass(BinaryMul, lhs *rhs);
-GridBinOpClass(BinaryDiv, lhs /rhs);
-GridBinOpClass(BinaryAnd, lhs &rhs);
-GridBinOpClass(BinaryOr, lhs | rhs);
-GridBinOpClass(BinaryAndAnd, lhs &&rhs);
-GridBinOpClass(BinaryOrOr, lhs || rhs);
-
-////////////////////////////////////////////////////
-// Trinary conditional op
-////////////////////////////////////////////////////
-#define GridTrinOpClass(name, combination)				\
-  template <class predicate, class left, class right>			\
-  struct name {								\
-    static auto accelerator_inline					\
-    func(const predicate &pred, const left &lhs, const right &rhs)	\
-      -> decltype(combination) const					\
-    {									\
-      return combination;						\
-    }									\
-  };
-
-GridTrinOpClass(TrinaryWhere,
-		(predicatedWhere<predicate, 
-		 typename std::remove_reference<left>::type,
-		 typename std::remove_reference<right>::type>(pred, lhs,rhs)));
-
-////////////////////////////////////////////
-// Operator syntactical glue
-////////////////////////////////////////////
-
-#define GRID_UNOP(name)   name<decltype(eval(0, arg))>
-#define GRID_BINOP(name)  name<decltype(eval(0, lhs)), decltype(eval(0, rhs))>
-#define GRID_TRINOP(name) name<decltype(eval(0, pred)), decltype(eval(0, lhs)), decltype(eval(0, rhs))>
-
-#define GRID_DEF_UNOP(op, name)						\
-  template <typename T1, typename std::enable_if<is_lattice<T1>::value||is_lattice_expr<T1>::value,T1>::type * = nullptr> \
-  inline auto op(const T1 &arg) ->decltype(LatticeUnaryExpression<GRID_UNOP(name),T1>(GRID_UNOP(name)(), arg)) \
-  {									\
-    return     LatticeUnaryExpression<GRID_UNOP(name),T1>(GRID_UNOP(name)(), arg); \
-  }
-
-#define GRID_BINOP_LEFT(op, name)					\
-  template <typename T1, typename T2,					\
-            typename std::enable_if<is_lattice<T1>::value||is_lattice_expr<T1>::value,T1>::type * = nullptr> \
-  inline auto op(const T1 &lhs, const T2 &rhs)				\
-    ->decltype(LatticeBinaryExpression<GRID_BINOP(name),T1,T2>(GRID_BINOP(name)(),lhs,rhs)) \
-  {									\
-    return     LatticeBinaryExpression<GRID_BINOP(name),T1,T2>(GRID_BINOP(name)(),lhs,rhs);\
-  }
-
-#define GRID_BINOP_RIGHT(op, name)					\
-  template <typename T1, typename T2,					\
-            typename std::enable_if<!is_lattice<T1>::value&&!is_lattice_expr<T1>::value,T1>::type * = nullptr, \
-            typename std::enable_if< is_lattice<T2>::value|| is_lattice_expr<T2>::value,T2>::type * = nullptr> \
-  inline auto op(const T1 &lhs, const T2 &rhs)				\
-    ->decltype(LatticeBinaryExpression<GRID_BINOP(name),T1,T2>(GRID_BINOP(name)(),lhs, rhs)) \
-  {									\
-    return     LatticeBinaryExpression<GRID_BINOP(name),T1,T2>(GRID_BINOP(name)(),lhs, rhs); \
-  }
-
-#define GRID_DEF_BINOP(op, name)		\
-  GRID_BINOP_LEFT(op, name);			\
-  GRID_BINOP_RIGHT(op, name);
-
-#define GRID_DEF_TRINOP(op, name)					\
-  template <typename T1, typename T2, typename T3>			\
-  inline auto op(const T1 &pred, const T2 &lhs, const T3 &rhs)		\
-    ->decltype(LatticeTrinaryExpression<GRID_TRINOP(name),T1,T2,T3>(GRID_TRINOP(name)(),pred, lhs, rhs)) \
-  {									\
-    return LatticeTrinaryExpression<GRID_TRINOP(name),T1,T2,T3>(GRID_TRINOP(name)(),pred, lhs, rhs); \
-  }
-
-////////////////////////
-// Operator definitions
-////////////////////////
-GRID_DEF_UNOP(operator-, UnarySub);
-GRID_DEF_UNOP(Not, UnaryNot);
-GRID_DEF_UNOP(operator!, UnaryNot);
-GRID_DEF_UNOP(adj, UnaryAdj);
-GRID_DEF_UNOP(conjugate, UnaryConj);
-GRID_DEF_UNOP(trace, UnaryTrace);
-GRID_DEF_UNOP(transpose, UnaryTranspose);
-GRID_DEF_UNOP(Ta, UnaryTa);
-GRID_DEF_UNOP(ProjectOnGroup, UnaryProjectOnGroup);
-GRID_DEF_UNOP(real, UnaryReal);
-GRID_DEF_UNOP(imag, UnaryImag);
-GRID_DEF_UNOP(toReal, UnaryToReal);
-GRID_DEF_UNOP(toComplex, UnaryToComplex);
-GRID_DEF_UNOP(timesI, UnaryTimesI);
-GRID_DEF_UNOP(timesMinusI, UnaryTimesMinusI);
-GRID_DEF_UNOP(abs, UnaryAbs);  // abs overloaded in cmath C++98; DON'T do the
-                               // abs-fabs-dabs-labs thing
-GRID_DEF_UNOP(sqrt, UnarySqrt);
-GRID_DEF_UNOP(rsqrt, UnaryRsqrt);
-GRID_DEF_UNOP(sin, UnarySin);
-GRID_DEF_UNOP(cos, UnaryCos);
-GRID_DEF_UNOP(asin, UnaryAsin);
-GRID_DEF_UNOP(acos, UnaryAcos);
-GRID_DEF_UNOP(log, UnaryLog);
-GRID_DEF_UNOP(exp, UnaryExp);
-
-GRID_DEF_BINOP(operator+, BinaryAdd);
-GRID_DEF_BINOP(operator-, BinarySub);
-GRID_DEF_BINOP(operator*, BinaryMul);
-GRID_DEF_BINOP(operator/, BinaryDiv);
-
-GRID_DEF_BINOP(operator&, BinaryAnd);
-GRID_DEF_BINOP(operator|, BinaryOr);
-GRID_DEF_BINOP(operator&&, BinaryAndAnd);
-GRID_DEF_BINOP(operator||, BinaryOrOr);
-
-GRID_DEF_TRINOP(where, TrinaryWhere);
-
-/////////////////////////////////////////////////////////////
-// Closure convenience to force expression to evaluate
-/////////////////////////////////////////////////////////////
-template <class Op, class T1>
-auto closure(const LatticeUnaryExpression<Op, T1> &expr)
-  -> Lattice<decltype(expr.op.func(eval(0, expr.arg1)))> 
-{
-  Lattice<decltype(expr.op.func(eval(0, expr.arg1)))> ret(expr);
-  return ret;
-}
-template <class Op, class T1, class T2>
-auto closure(const LatticeBinaryExpression<Op, T1, T2> &expr)
-  -> Lattice<decltype(expr.op.func(eval(0, expr.arg1),eval(0, expr.arg2)))> 
-{
-  Lattice<decltype(expr.op.func(eval(0, expr.arg1),eval(0, expr.arg2)))> ret(expr);
-  return ret;
-}
-template <class Op, class T1, class T2, class T3>
-auto closure(const LatticeTrinaryExpression<Op, T1, T2, T3> &expr)
-  -> Lattice<decltype(expr.op.func(eval(0, expr.arg1),
-				   eval(0, expr.arg2),
-				   eval(0, expr.arg3)))> 
-{
-  Lattice<decltype(expr.op.func(eval(0, expr.arg1),
-				eval(0, expr.arg2),
-				eval(0, expr.arg3)))>  ret(expr);
-  return ret;
-}
-
-#undef GRID_UNOP
-#undef GRID_BINOP
-#undef GRID_TRINOP
-
-#undef GRID_DEF_UNOP
-#undef GRID_DEF_BINOP
-#undef GRID_DEF_TRINOP
-
-NAMESPACE_END(Grid);
-
-#endif

Grid/lattice/Lattice_arith.h

@@ -1,257 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/lattice/Lattice_arith.h
-
-    Copyright (C) 2015
-
-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 */
-#ifndef GRID_LATTICE_ARITH_H
-#define GRID_LATTICE_ARITH_H
-
-NAMESPACE_BEGIN(Grid);
-
-//////////////////////////////////////////////////////////////////////////////////////////////////////
-//  avoid copy back routines for mult, mac, sub, add
-//////////////////////////////////////////////////////////////////////////////////////////////////////
-template<class obj1,class obj2,class obj3> inline
-void mult(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const Lattice<obj3> &rhs){
-  ret.Checkerboard() = lhs.Checkerboard();
-  auto ret_v = ret.View();
-  auto lhs_v = lhs.View();
-  auto rhs_v = rhs.View();
-  conformable(ret,rhs);
-  conformable(lhs,rhs);
-  accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
-    decltype(coalescedRead(obj1())) tmp;
-    auto lhs_t = lhs_v(ss);
-    auto rhs_t = rhs_v(ss);
-    mult(&tmp,&lhs_t,&rhs_t);
-    coalescedWrite(ret_v[ss],tmp);
-  });
-}
-  
-template<class obj1,class obj2,class obj3> inline
-void mac(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const Lattice<obj3> &rhs){
-  ret.Checkerboard() = lhs.Checkerboard();
-  conformable(ret,rhs);
-  conformable(lhs,rhs);
-  auto ret_v = ret.View();
-  auto lhs_v = lhs.View();
-  auto rhs_v = rhs.View();
-  accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
-    decltype(coalescedRead(obj1())) tmp;
-    auto lhs_t=lhs_v(ss);
-    auto rhs_t=rhs_v(ss);
-    mac(&tmp,&lhs_t,&rhs_t);
-    coalescedWrite(ret_v[ss],tmp);
-  });
-}
-  
-template<class obj1,class obj2,class obj3> inline
-void sub(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const Lattice<obj3> &rhs){
-  ret.Checkerboard() = lhs.Checkerboard();
-  conformable(ret,rhs);
-  conformable(lhs,rhs);
-  auto ret_v = ret.View();
-  auto lhs_v = lhs.View();
-  auto rhs_v = rhs.View();
-  accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
-    decltype(coalescedRead(obj1())) tmp;
-    auto lhs_t=lhs_v(ss);
-    auto rhs_t=rhs_v(ss);
-    sub(&tmp,&lhs_t,&rhs_t);
-    coalescedWrite(ret_v[ss],tmp);
-  });
-}
-template<class obj1,class obj2,class obj3> inline
-void add(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const Lattice<obj3> &rhs){
-  ret.Checkerboard() = lhs.Checkerboard();
-  conformable(ret,rhs);
-  conformable(lhs,rhs);
-  auto ret_v = ret.View();
-  auto lhs_v = lhs.View();
-  auto rhs_v = rhs.View();
-  accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
-    decltype(coalescedRead(obj1())) tmp;
-    auto lhs_t=lhs_v(ss);
-    auto rhs_t=rhs_v(ss);
-    add(&tmp,&lhs_t,&rhs_t);
-    coalescedWrite(ret_v[ss],tmp);
-  });
-}
-  
-//////////////////////////////////////////////////////////////////////////////////////////////////////
-//  avoid copy back routines for mult, mac, sub, add
-//////////////////////////////////////////////////////////////////////////////////////////////////////
-template<class obj1,class obj2,class obj3> inline
-void mult(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
-  ret.Checkerboard() = lhs.Checkerboard();
-  conformable(lhs,ret);
-  auto ret_v = ret.View();
-  auto lhs_v = lhs.View();
-  accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
-    decltype(coalescedRead(obj1())) tmp;
-    mult(&tmp,&lhs_v(ss),&rhs);
-    coalescedWrite(ret_v[ss],tmp);
-  });
-}
-  
-template<class obj1,class obj2,class obj3> inline
-void mac(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
-  ret.Checkerboard() = lhs.Checkerboard();
-  conformable(ret,lhs);
-  auto ret_v = ret.View();
-  auto lhs_v = lhs.View();
-  accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
-    decltype(coalescedRead(obj1())) tmp;
-    auto lhs_t=lhs_v(ss);
-    mac(&tmp,&lhs_t,&rhs);
-    coalescedWrite(ret_v[ss],tmp);
-  });
-}
-  
-template<class obj1,class obj2,class obj3> inline
-void sub(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
-  ret.Checkerboard() = lhs.Checkerboard();
-  conformable(ret,lhs);
-  auto ret_v = ret.View();
-  auto lhs_v = lhs.View();
-  accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
-    decltype(coalescedRead(obj1())) tmp;
-    auto lhs_t=lhs_v(ss);
-    sub(&tmp,&lhs_t,&rhs);
-    coalescedWrite(ret_v[ss],tmp);
-  });
-}
-template<class obj1,class obj2,class obj3> inline
-void add(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
-  ret.Checkerboard() = lhs.Checkerboard();
-  conformable(lhs,ret);
-  auto ret_v = ret.View();
-  auto lhs_v = lhs.View();
-  accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
-    decltype(coalescedRead(obj1())) tmp;
-    auto lhs_t=lhs_v(ss);
-    add(&tmp,&lhs_t,&rhs);
-    coalescedWrite(ret_v[ss],tmp);
-  });
-}
-
-//////////////////////////////////////////////////////////////////////////////////////////////////////
-//  avoid copy back routines for mult, mac, sub, add
-//////////////////////////////////////////////////////////////////////////////////////////////////////
-template<class obj1,class obj2,class obj3> inline
-void mult(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
-  ret.Checkerboard() = rhs.Checkerboard();
-  conformable(ret,rhs);
-  auto ret_v = ret.View();
-  auto rhs_v = lhs.View();
-  accelerator_for(ss,rhs_v.size(),obj1::Nsimd(),{
-    decltype(coalescedRead(obj1())) tmp;
-    auto rhs_t=rhs_v(ss);
-    mult(&tmp,&lhs,&rhs_t);
-    coalescedWrite(ret_v[ss],tmp);
-  });
-}
-  
-template<class obj1,class obj2,class obj3> inline
-void mac(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
-  ret.Checkerboard() = rhs.Checkerboard();
-  conformable(ret,rhs);
-  auto ret_v = ret.View();
-  auto rhs_v = lhs.View();
-  accelerator_for(ss,rhs_v.size(),obj1::Nsimd(),{
-    decltype(coalescedRead(obj1())) tmp;
-    auto rhs_t=rhs_v(ss);
-    mac(&tmp,&lhs,&rhs_t);
-    coalescedWrite(ret_v[ss],tmp);
-  });
-}
-  
-template<class obj1,class obj2,class obj3> inline
-void sub(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
-  ret.Checkerboard() = rhs.Checkerboard();
-  conformable(ret,rhs);
-  auto ret_v = ret.View();
-  auto rhs_v = lhs.View();
-  accelerator_for(ss,rhs_v.size(),obj1::Nsimd(),{
-    decltype(coalescedRead(obj1())) tmp;
-    auto rhs_t=rhs_v(ss);
-    sub(&tmp,&lhs,&rhs_t);
-    coalescedWrite(ret_v[ss],tmp);
-  });
-}
-template<class obj1,class obj2,class obj3> inline
-void add(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
-  ret.Checkerboard() = rhs.Checkerboard();
-  conformable(ret,rhs);
-  auto ret_v = ret.View();
-  auto rhs_v = lhs.View();
-  accelerator_for(ss,rhs_v.size(),obj1::Nsimd(),{
-    decltype(coalescedRead(obj1())) tmp;
-    auto rhs_t=rhs_v(ss);
-    add(&tmp,&lhs,&rhs_t);
-    coalescedWrite(ret_v[ss],tmp);
-  });
-}
-  
-template<class sobj,class vobj> inline
-void axpy(Lattice<vobj> &ret,sobj a,const Lattice<vobj> &x,const Lattice<vobj> &y){
-  ret.Checkerboard() = x.Checkerboard();
-  conformable(ret,x);
-  conformable(x,y);
-  auto ret_v = ret.View();
-  auto x_v = x.View();
-  auto y_v = y.View();
-  accelerator_for(ss,x_v.size(),vobj::Nsimd(),{
-    auto tmp = a*x_v(ss)+y_v(ss);
-    coalescedWrite(ret_v[ss],tmp);
-  });
-}
-template<class sobj,class vobj> inline
-void axpby(Lattice<vobj> &ret,sobj a,sobj b,const Lattice<vobj> &x,const Lattice<vobj> &y){
-  ret.Checkerboard() = x.Checkerboard();
-  conformable(ret,x);
-  conformable(x,y);
-  auto ret_v = ret.View();
-  auto x_v = x.View();
-  auto y_v = y.View();
-  accelerator_for(ss,x_v.size(),vobj::Nsimd(),{
-    auto tmp = a*x_v(ss)+b*y_v(ss);
-    coalescedWrite(ret_v[ss],tmp);
-  });
-}
-
-template<class sobj,class vobj> inline
-RealD axpy_norm(Lattice<vobj> &ret,sobj a,const Lattice<vobj> &x,const Lattice<vobj> &y)
-{
-    return axpy_norm_fast(ret,a,x,y);
-}
-template<class sobj,class vobj> inline
-RealD axpby_norm(Lattice<vobj> &ret,sobj a,sobj b,const Lattice<vobj> &x,const Lattice<vobj> &y)
-{
-    return axpby_norm_fast(ret,a,b,x,y);
-}
-
-NAMESPACE_END(Grid);
-#endif

Grid/lattice/Lattice_base.h

@@ -1,466 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/lattice/Lattice_base.h
-
-Copyright (C) 2015
-
-Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is 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 
-
-#define STREAMING_STORES
-
-NAMESPACE_BEGIN(Grid);
-
-extern int GridCshiftPermuteMap[4][16];
-
-///////////////////////////////////////////////////////////////////
-// Base class which can be used by traits to pick up behaviour
-///////////////////////////////////////////////////////////////////
-class LatticeBase {};
-
-/////////////////////////////////////////////////////////////////////////////////////////
-// Conformable checks; same instance of Grid required
-/////////////////////////////////////////////////////////////////////////////////////////
-void accelerator_inline conformable(GridBase *lhs,GridBase *rhs)
-{
-  assert(lhs == rhs);
-}
-
-////////////////////////////////////////////////////////////////////////////
-// Minimal base class containing only data valid to access from accelerator
-// _odata will be a managed pointer in CUDA
-////////////////////////////////////////////////////////////////////////////
-// Force access to lattice through a view object.
-// prevents writing of code that will not offload to GPU, but perhaps annoyingly
-// strict since host could could in principle direct access through the lattice object
-// Need to decide programming model.
-#define LATTICE_VIEW_STRICT
-template<class vobj> class LatticeAccelerator : public LatticeBase
-{
-protected:
-  GridBase *_grid;
-  int checkerboard;
-  vobj     *_odata;    // A managed pointer
-  uint64_t _odata_size;    
-public:
-  accelerator_inline LatticeAccelerator() : checkerboard(0), _odata(nullptr), _odata_size(0), _grid(nullptr) { }; 
-  accelerator_inline uint64_t oSites(void) const { return _odata_size; };
-  accelerator_inline int  Checkerboard(void) const { return checkerboard; };
-  accelerator_inline int &Checkerboard(void) { return this->checkerboard; }; // can assign checkerboard on a container, not a view
-  accelerator_inline void Conformable(GridBase * &grid) const
-  { 
-    if (grid) conformable(grid, _grid);
-    else      grid = _grid;
-  };
-};
-
-/////////////////////////////////////////////////////////////////////////////////////////
-// A View class which provides accessor to the data.
-// This will be safe to call from accelerator_for and is trivially copy constructible
-// The copy constructor for this will need to be used by device lambda functions
-/////////////////////////////////////////////////////////////////////////////////////////
-template<class vobj> 
-class LatticeView : public LatticeAccelerator<vobj>
-{
-public:
-
-
-  // Rvalue
-#ifdef __CUDA_ARCH__
-  accelerator_inline const typename vobj::scalar_object operator()(size_t i) const { return coalescedRead(this->_odata[i]); }
-#else 
-  accelerator_inline const vobj & operator()(size_t i) const { return this->_odata[i]; }
-#endif
-
-  accelerator_inline const vobj & operator[](size_t i) const { return this->_odata[i]; };
-  accelerator_inline vobj       & operator[](size_t i)       { return this->_odata[i]; };
-
-  accelerator_inline uint64_t begin(void) const { return 0;};
-  accelerator_inline uint64_t end(void)   const { return this->_odata_size; };
-  accelerator_inline uint64_t size(void)  const { return this->_odata_size; };
-
-  LatticeView(const LatticeAccelerator<vobj> &refer_to_me) : LatticeAccelerator<vobj> (refer_to_me)
-  {
-  }
-};
-
-/////////////////////////////////////////////////////////////////////////////////////////
-// Lattice expression types used by ET to assemble the AST
-// 
-// Need to be able to detect code paths according to the whether a lattice object or not
-// so introduce some trait type things
-/////////////////////////////////////////////////////////////////////////////////////////
-
-class LatticeExpressionBase {};
-
-template <typename T> using is_lattice = std::is_base_of<LatticeBase, T>;
-template <typename T> using is_lattice_expr = std::is_base_of<LatticeExpressionBase,T >;
-
-template<class T, bool isLattice> struct ViewMapBase { typedef T Type; };
-template<class T>                 struct ViewMapBase<T,true> { typedef LatticeView<typename T::vector_object> Type; };
-template<class T> using ViewMap = ViewMapBase<T,std::is_base_of<LatticeBase, T>::value >;
-
-template <typename Op, typename _T1>                           
-class LatticeUnaryExpression : public  LatticeExpressionBase 
-{
-public:
-  typedef typename ViewMap<_T1>::Type T1;
-  Op op;
-  T1 arg1;
-  LatticeUnaryExpression(Op _op,const _T1 &_arg1) : op(_op), arg1(_arg1) {};
-};
-
-template <typename Op, typename _T1, typename _T2>              
-class LatticeBinaryExpression : public LatticeExpressionBase 
-{
-public:
-  typedef typename ViewMap<_T1>::Type T1;
-  typedef typename ViewMap<_T2>::Type T2;
-  Op op;
-  T1 arg1;
-  T2 arg2;
-  LatticeBinaryExpression(Op _op,const _T1 &_arg1,const _T2 &_arg2) : op(_op), arg1(_arg1), arg2(_arg2) {};
-};
-
-template <typename Op, typename _T1, typename _T2, typename _T3> 
-class LatticeTrinaryExpression : public LatticeExpressionBase 
-{
-public:
-  typedef typename ViewMap<_T1>::Type T1;
-  typedef typename ViewMap<_T2>::Type T2;
-  typedef typename ViewMap<_T3>::Type T3;
-  Op op;
-  T1 arg1;
-  T2 arg2;
-  T3 arg3;
-  LatticeTrinaryExpression(Op _op,const _T1 &_arg1,const _T2 &_arg2,const _T3 &_arg3) : op(_op), arg1(_arg1), arg2(_arg2), arg3(_arg3) {};
-};
-
-/////////////////////////////////////////////////////////////////////////////////////////
-// The real lattice class, with normal copy and assignment semantics.
-// This contains extra (host resident) grid pointer data that may be accessed by host code
-/////////////////////////////////////////////////////////////////////////////////////////
-template<class vobj>
-class Lattice : public LatticeAccelerator<vobj>
-{
-public:
-  GridBase *Grid(void) const { return this->_grid; }
-  ///////////////////////////////////////////////////
-  // Member types
-  ///////////////////////////////////////////////////
-  typedef typename vobj::scalar_type scalar_type;
-  typedef typename vobj::vector_type vector_type;
-  typedef vobj vector_object;
-
-private:
-  void dealloc(void)
-  {
-    alignedAllocator<vobj> alloc;
-    if( this->_odata_size ) {
-      alloc.deallocate(this->_odata,this->_odata_size);
-      this->_odata=nullptr;
-      this->_odata_size=0;
-    }
-  }
-  void resize(uint64_t size)
-  {
-    alignedAllocator<vobj> alloc;
-    if ( this->_odata_size != size ) {
-      dealloc();
-    }
-    this->_odata_size = size;
-    if ( size ) 
-      this->_odata      = alloc.allocate(this->_odata_size);
-    else 
-      this->_odata      = nullptr;
-  }
-public:
-  /////////////////////////////////////////////////////////////////////////////////
-  // 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
-  // in device lambdas
-  /////////////////////////////////////////////////////////////////////////////////
-  LatticeView<vobj> View (void) const 
-  {
-    LatticeView<vobj> accessor(*( (LatticeAccelerator<vobj> *) this));
-    return accessor;
-  }
-  
-  ~Lattice() { 
-    if ( this->_odata_size ) {
-      dealloc();
-    }
-   }
-  ////////////////////////////////////////////////////////////////////////////////
-  // Expression Template closure support
-  ////////////////////////////////////////////////////////////////////////////////
-  template <typename Op, typename T1> inline Lattice<vobj> & operator=(const LatticeUnaryExpression<Op,T1> &expr)
-  {
-    GridBase *egrid(nullptr);
-    GridFromExpression(egrid,expr);
-    assert(egrid!=nullptr);
-    conformable(this->_grid,egrid);
-
-    int cb=-1;
-    CBFromExpression(cb,expr);
-    assert( (cb==Odd) || (cb==Even));
-    this->checkerboard=cb;
-
-    auto me  = View();
-    accelerator_for(ss,me.size(),1,{
-      auto tmp = eval(ss,expr);
-      vstream(me[ss],tmp);
-    });
-    return *this;
-  }
-  template <typename Op, typename T1,typename T2> inline Lattice<vobj> & operator=(const LatticeBinaryExpression<Op,T1,T2> &expr)
-  {
-    GridBase *egrid(nullptr);
-    GridFromExpression(egrid,expr);
-    assert(egrid!=nullptr);
-    conformable(this->_grid,egrid);
-
-    int cb=-1;
-    CBFromExpression(cb,expr);
-    assert( (cb==Odd) || (cb==Even));
-    this->checkerboard=cb;
-
-    auto me  = View();
-    accelerator_for(ss,me.size(),1,{
-      auto tmp = eval(ss,expr);
-      vstream(me[ss],tmp);
-    });
-    return *this;
-  }
-  template <typename Op, typename T1,typename T2,typename T3> inline Lattice<vobj> & operator=(const LatticeTrinaryExpression<Op,T1,T2,T3> &expr)
-  {
-    GridBase *egrid(nullptr);
-    GridFromExpression(egrid,expr);
-    assert(egrid!=nullptr);
-    conformable(this->_grid,egrid);
-
-    int cb=-1;
-    CBFromExpression(cb,expr);
-    assert( (cb==Odd) || (cb==Even));
-    this->checkerboard=cb;
-    auto me  = View();
-    accelerator_for(ss,me.size(),1,{
-      auto tmp = eval(ss,expr);
-      vstream(me[ss],tmp);
-    });
-    return *this;
-  }
-  //GridFromExpression is tricky to do
-  template<class Op,class T1>
-  Lattice(const LatticeUnaryExpression<Op,T1> & expr) {
-    this->_grid = nullptr;
-    GridFromExpression(this->_grid,expr);
-    assert(this->_grid!=nullptr);
-
-    int cb=-1;
-    CBFromExpression(cb,expr);
-    assert( (cb==Odd) || (cb==Even));
-    this->checkerboard=cb;
-
-    resize(this->_grid->oSites());
-
-    *this = expr;
-  }
-  template<class Op,class T1, class T2>
-  Lattice(const LatticeBinaryExpression<Op,T1,T2> & expr) {
-    this->_grid = nullptr;
-    GridFromExpression(this->_grid,expr);
-    assert(this->_grid!=nullptr);
-
-    int cb=-1;
-    CBFromExpression(cb,expr);
-    assert( (cb==Odd) || (cb==Even));
-    this->checkerboard=cb;
-
-    resize(this->_grid->oSites());
-
-    *this = expr;
-  }
-  template<class Op,class T1, class T2, class T3>
-  Lattice(const LatticeTrinaryExpression<Op,T1,T2,T3> & expr) {
-    this->_grid = nullptr;
-    GridFromExpression(this->_grid,expr);
-    assert(this->_grid!=nullptr);
-
-    int cb=-1;
-    CBFromExpression(cb,expr);
-    assert( (cb==Odd) || (cb==Even));
-    this->checkerboard=cb;
-
-    resize(this->_grid->oSites());
-
-    *this = expr;
-  }
-
-  template<class sobj> inline Lattice<vobj> & operator = (const sobj & r){
-    auto me  = View();
-    thread_for(ss,me.size(),{
-      me[ss] = r;
-    });
-    return *this;
-  }
-
-  //////////////////////////////////////////////////////////////////
-  // Follow rule of five, with Constructor requires "grid" passed
-  // to user defined constructor
-  ///////////////////////////////////////////
-  // user defined constructor
-  ///////////////////////////////////////////
-  Lattice(GridBase *grid) { 
-    this->_grid = grid;
-    resize(this->_grid->oSites());
-    assert((((uint64_t)&this->_odata[0])&0xF) ==0);
-    this->checkerboard=0;
-  }
-  
-  //  virtual ~Lattice(void) = default;
-    
-  void reset(GridBase* grid) {
-    if (this->_grid != grid) {
-      this->_grid = grid;
-      this->_odata.resize(grid->oSites());
-      this->checkerboard = 0;
-    }
-  }
-  ///////////////////////////////////////////
-  // copy constructor
-  ///////////////////////////////////////////
-  Lattice(const Lattice& r){ 
-    //    std::cout << "Lattice constructor(const Lattice &) "<<this<<std::endl; 
-    this->_grid = r.Grid();
-    resize(this->_grid->oSites());
-    *this = r;
-  }
-  ///////////////////////////////////////////
-  // move constructor
-  ///////////////////////////////////////////
-  Lattice(Lattice && r){ 
-    this->_grid = r.Grid();
-    this->_odata      = r._odata;
-    this->_odata_size = r._odata_size;
-    this->checkerboard= r.Checkerboard();
-    r._odata      = nullptr;
-    r._odata_size = 0;
-  }
-  ///////////////////////////////////////////
-  // assignment template
-  ///////////////////////////////////////////
-  template<class robj> inline Lattice<vobj> & operator = (const Lattice<robj> & r){
-    typename std::enable_if<!std::is_same<robj,vobj>::value,int>::type i=0;
-    conformable(*this,r);
-    this->checkerboard = r.Checkerboard();
-    auto me =   View();
-    auto him= r.View();
-    accelerator_for(ss,me.size(),vobj::Nsimd(),{
-      coalescedWrite(me[ss],him(ss));
-    });
-    return *this;
-  }
-
-  ///////////////////////////////////////////
-  // Copy assignment 
-  ///////////////////////////////////////////
-  inline Lattice<vobj> & operator = (const Lattice<vobj> & r){
-    this->checkerboard = r.Checkerboard();
-    conformable(*this,r);
-    auto me =   View();
-    auto him= r.View();
-    accelerator_for(ss,me.size(),vobj::Nsimd(),{
-      coalescedWrite(me[ss],him(ss));
-    });
-    return *this;
-  }
-  ///////////////////////////////////////////
-  // Move assignment possible if same type
-  ///////////////////////////////////////////
-  inline Lattice<vobj> & operator = (Lattice<vobj> && r){
-
-    resize(0); // deletes if appropriate
-    this->_grid       = r.Grid();
-    this->_odata      = r._odata;
-    this->_odata_size = r._odata_size;
-    this->checkerboard= r.Checkerboard();
-
-    r._odata      = nullptr;
-    r._odata_size = 0;
-    
-    return *this;
-  }
-
-  /////////////////////////////////////////////////////////////////////////////
-  // *=,+=,-= operators inherit behvour from correspond */+/- operation
-  /////////////////////////////////////////////////////////////////////////////
-  template<class T> inline Lattice<vobj> &operator *=(const T &r) {
-    *this = (*this)*r;
-    return *this;
-  }
-  
-  template<class T> inline Lattice<vobj> &operator -=(const T &r) {
-    *this = (*this)-r;
-    return *this;
-  }
-  template<class T> inline Lattice<vobj> &operator +=(const T &r) {
-    *this = (*this)+r;
-    return *this;
-  }
-
-  friend inline void swap(Lattice &l, Lattice &r) { 
-    conformable(l,r);
-    LatticeAccelerator<vobj> tmp;
-    LatticeAccelerator<vobj> *lp = (LatticeAccelerator<vobj> *)&l;
-    LatticeAccelerator<vobj> *rp = (LatticeAccelerator<vobj> *)&r;
-    tmp = *lp;    *lp=*rp;    *rp=tmp;
-  }
-
-}; // class Lattice
-
-template<class vobj> std::ostream& operator<< (std::ostream& stream, const Lattice<vobj> &o){
-  typedef typename vobj::scalar_object sobj;
-  for(int g=0;g<o.Grid()->_gsites;g++){
-
-    Coordinate gcoor;
-    o.Grid()->GlobalIndexToGlobalCoor(g,gcoor);
-
-    sobj ss;
-    peekSite(ss,o,gcoor);
-    stream<<"[";
-    for(int d=0;d<gcoor.size();d++){
-      stream<<gcoor[d];
-      if(d!=gcoor.size()-1) stream<<",";
-    }
-    stream<<"]\t";
-    stream<<ss<<std::endl;
-  }
-  return stream;
-}
-  
-NAMESPACE_END(Grid);
-

Grid/lattice/Lattice_comparison.h

@@ -1,207 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/lattice/Lattice_comparison.h
-
-    Copyright (C) 2015
-
-Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
-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 */
-#ifndef GRID_LATTICE_COMPARISON_H
-#define GRID_LATTICE_COMPARISON_H
-
-NAMESPACE_BEGIN(Grid);
-
-//////////////////////////////////////////////////////////////////////////
-// relational operators
-// 
-// Support <,>,<=,>=,==,!=
-//
-//Query supporting bitwise &, |, ^, !
-//Query supporting logical &&, ||, 
-//////////////////////////////////////////////////////////////////////////
-
-typedef iScalar<vInteger> vPredicate ;
-
-/*
-template <class iobj, class vobj, class robj> accelerator_inline 
-vobj predicatedWhere(const iobj &predicate, const vobj &iftrue, const robj &iffalse) 
-{
-  typename std::remove_const<vobj>::type ret;
-
-  typedef typename vobj::scalar_object scalar_object;
-  typedef typename vobj::scalar_type scalar_type;
-  typedef typename vobj::vector_type vector_type;
-
-  const int Nsimd = vobj::vector_type::Nsimd();
-
-  ExtractBuffer<Integer> mask(Nsimd);
-  ExtractBuffer<scalar_object> truevals(Nsimd);
-  ExtractBuffer<scalar_object> falsevals(Nsimd);
-
-  extract(iftrue, truevals);
-  extract(iffalse, falsevals);
-  extract<vInteger, Integer>(TensorRemove(predicate), mask);
-
-  for (int s = 0; s < Nsimd; s++) {
-    if (mask[s]) falsevals[s] = truevals[s];
-  }
-
-  merge(ret, falsevals);
-  return ret;
-}
-*/
-//////////////////////////////////////////////////////////////////////////
-// compare lattice to lattice
-//////////////////////////////////////////////////////////////////////////
-
-template<class vfunctor,class lobj,class robj>  
-inline Lattice<vPredicate> LLComparison(vfunctor op,const Lattice<lobj> &lhs,const Lattice<robj> &rhs)
-{
-  Lattice<vPredicate> ret(rhs.Grid());
-  auto lhs_v = lhs.View();
-  auto rhs_v = rhs.View();
-  auto ret_v = ret.View();
-  thread_for( ss, rhs_v.size(), {
-      ret_v[ss]=op(lhs_v[ss],rhs_v[ss]);
-  });
-  return ret;
-}
-//////////////////////////////////////////////////////////////////////////
-// compare lattice to scalar
-//////////////////////////////////////////////////////////////////////////
-template<class vfunctor,class lobj,class robj> 
-inline Lattice<vPredicate> LSComparison(vfunctor op,const Lattice<lobj> &lhs,const robj &rhs)
-{
-  Lattice<vPredicate> ret(lhs.Grid());
-  auto lhs_v = lhs.View();
-  auto ret_v = ret.View();
-  thread_for( ss, lhs_v.size(), {
-    ret_v[ss]=op(lhs_v[ss],rhs);
-  });
-  return ret;
-}
-//////////////////////////////////////////////////////////////////////////
-// compare scalar to lattice
-//////////////////////////////////////////////////////////////////////////
-template<class vfunctor,class lobj,class robj> 
-inline Lattice<vPredicate> SLComparison(vfunctor op,const lobj &lhs,const Lattice<robj> &rhs)
-{
-  Lattice<vPredicate> ret(rhs.Grid());
-  auto rhs_v = rhs.View();
-  auto ret_v = ret.View();
-  thread_for( ss, rhs_v.size(), {
-    ret_v[ss]=op(lhs,rhs_v[ss]);
-  });
-  return ret;
-}
-  
-//////////////////////////////////////////////////////////////////////////
-// Map to functors
-//////////////////////////////////////////////////////////////////////////
-// Less than
-template<class lobj,class robj>
-inline Lattice<vPredicate> operator < (const Lattice<lobj> & lhs, const Lattice<robj> & rhs) {
-  return LLComparison(vlt<lobj,robj>(),lhs,rhs);
-}
-template<class lobj,class robj>
-inline Lattice<vPredicate> operator < (const Lattice<lobj> & lhs, const robj & rhs) {
-  return LSComparison(vlt<lobj,robj>(),lhs,rhs);
-}
-template<class lobj,class robj>
-inline Lattice<vPredicate> operator < (const lobj & lhs, const Lattice<robj> & rhs) {
-  return SLComparison(vlt<lobj,robj>(),lhs,rhs);
-}
-  
-// Less than equal
-template<class lobj,class robj>
-inline Lattice<vPredicate> operator <= (const Lattice<lobj> & lhs, const Lattice<robj> & rhs) {
-  return LLComparison(vle<lobj,robj>(),lhs,rhs);
-}
-template<class lobj,class robj>
-inline Lattice<vPredicate> operator <= (const Lattice<lobj> & lhs, const robj & rhs) {
-  return LSComparison(vle<lobj,robj>(),lhs,rhs);
-}
-template<class lobj,class robj>
-inline Lattice<vPredicate> operator <= (const lobj & lhs, const Lattice<robj> & rhs) {
-  return SLComparison(vle<lobj,robj>(),lhs,rhs);
-}
-  
-// Greater than 
-template<class lobj,class robj>
-inline Lattice<vPredicate> operator > (const Lattice<lobj> & lhs, const Lattice<robj> & rhs) {
-  return LLComparison(vgt<lobj,robj>(),lhs,rhs);
-}
-template<class lobj,class robj>
-inline Lattice<vPredicate> operator > (const Lattice<lobj> & lhs, const robj & rhs) {
-  return LSComparison(vgt<lobj,robj>(),lhs,rhs);
-}
-template<class lobj,class robj>
-inline Lattice<vPredicate> operator > (const lobj & lhs, const Lattice<robj> & rhs) {
-  return SLComparison(vgt<lobj,robj>(),lhs,rhs);
-}
-  
-  
-// Greater than equal
-template<class lobj,class robj>
-inline Lattice<vPredicate> operator >= (const Lattice<lobj> & lhs, const Lattice<robj> & rhs) {
-  return LLComparison(vge<lobj,robj>(),lhs,rhs);
-}
-template<class lobj,class robj>
-inline Lattice<vPredicate> operator >= (const Lattice<lobj> & lhs, const robj & rhs) {
-  return LSComparison(vge<lobj,robj>(),lhs,rhs);
-}
-template<class lobj,class robj>
-inline Lattice<vPredicate> operator >= (const lobj & lhs, const Lattice<robj> & rhs) {
-  return SLComparison(vge<lobj,robj>(),lhs,rhs);
-}
-   
-// equal
-template<class lobj,class robj>
-inline Lattice<vPredicate> operator == (const Lattice<lobj> & lhs, const Lattice<robj> & rhs) {
-  return LLComparison(veq<lobj,robj>(),lhs,rhs);
-}
-template<class lobj,class robj>
-inline Lattice<vPredicate> operator == (const Lattice<lobj> & lhs, const robj & rhs) {
-  return LSComparison(veq<lobj,robj>(),lhs,rhs);
-}
-template<class lobj,class robj>
-inline Lattice<vPredicate> operator == (const lobj & lhs, const Lattice<robj> & rhs) {
-  return SLComparison(veq<lobj,robj>(),lhs,rhs);
-}
-   
-   
-// not equal
-template<class lobj,class robj>
-inline Lattice<vPredicate> operator != (const Lattice<lobj> & lhs, const Lattice<robj> & rhs) {
-  return LLComparison(vne<lobj,robj>(),lhs,rhs);
-}
-template<class lobj,class robj>
-inline Lattice<vPredicate> operator != (const Lattice<lobj> & lhs, const robj & rhs) {
-  return LSComparison(vne<lobj,robj>(),lhs,rhs);
-}
-template<class lobj,class robj>
-inline Lattice<vPredicate> operator != (const lobj & lhs, const Lattice<robj> & rhs) {
-  return SLComparison(vne<lobj,robj>(),lhs,rhs);
-}
-NAMESPACE_END(Grid);
-#endif

Grid/lattice/Lattice_coordinate.h

@@ -1,74 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/lattice/Lattice_coordinate.h
-
-    Copyright (C) 2015
-
-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 iobj> inline void LatticeCoordinate(Lattice<iobj> &l,int mu)
-{
-  typedef typename iobj::scalar_type scalar_type;
-  typedef typename iobj::vector_type vector_type;
-
-  GridBase *grid = l.Grid();
-  int Nsimd = grid->iSites();
-
-  Coordinate gcoor;
-  ExtractBuffer<scalar_type> mergebuf(Nsimd);
-
-  vector_type vI;
-  auto l_v = l.View();
-  for(int o=0;o<grid->oSites();o++){
-    for(int i=0;i<grid->iSites();i++){
-      grid->RankIndexToGlobalCoor(grid->ThisRank(),o,i,gcoor);
-      mergebuf[i]=(Integer)gcoor[mu];
-    }
-    merge<vector_type,scalar_type>(vI,mergebuf);
-    l_v[o]=vI;
-  }
-};
-
-// LatticeCoordinate();
-// FIXME for debug; deprecate this; made obscelete by 
-template<class vobj> void lex_sites(Lattice<vobj> &l){
-  auto l_v = l.View();
-  Real *v_ptr = (Real *)&l_v[0];
-  size_t o_len = l.Grid()->oSites();
-  size_t v_len = sizeof(vobj)/sizeof(vRealF);
-  size_t vec_len = vRealF::Nsimd();
-
-  for(int i=0;i<o_len;i++){
-    for(int j=0;j<v_len;j++){
-      for(int vv=0;vv<vec_len;vv+=2){
-	v_ptr[i*v_len*vec_len+j*vec_len+vv  ]= i+vv*500;
-	v_ptr[i*v_len*vec_len+j*vec_len+vv+1]= i+vv*500;
-      }
-    }}
-}
-
-NAMESPACE_END(Grid);
-

Grid/lattice/Lattice_local.h

@@ -1,87 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/lattice/Lattice_local.h
-
-    Copyright (C) 2015
-
-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 */
-#ifndef GRID_LATTICE_LOCALREDUCTION_H
-#define GRID_LATTICE_LOCALREDUCTION_H
-
-///////////////////////////////////////////////
-// localInner, localNorm, outerProduct
-///////////////////////////////////////////////
-
-NAMESPACE_BEGIN(Grid);
-
-/////////////////////////////////////////////////////
-// Non site, reduced locally reduced routines
-/////////////////////////////////////////////////////
-  
-// localNorm2,
-template<class vobj>
-inline auto localNorm2 (const Lattice<vobj> &rhs)-> Lattice<typename vobj::tensor_reduced>
-{
-  Lattice<typename vobj::tensor_reduced> ret(rhs.Grid());
-  auto rhs_v = rhs.View();
-  auto ret_v = ret.View();
-  accelerator_for(ss,rhs_v.size(),vobj::Nsimd(),{
-    coalescedWrite(ret_v[ss],innerProduct(rhs_v(ss),rhs_v(ss)));
-  });
-  return ret;
-}
-  
-// localInnerProduct
-template<class vobj>
-inline auto localInnerProduct (const Lattice<vobj> &lhs,const Lattice<vobj> &rhs) -> Lattice<typename vobj::tensor_reduced>
-{
-  Lattice<typename vobj::tensor_reduced> ret(rhs.Grid());
-  auto lhs_v = lhs.View();
-  auto rhs_v = rhs.View();
-  auto ret_v = ret.View();
-  accelerator_for(ss,rhs_v.size(),vobj::Nsimd(),{
-    coalescedWrite(ret_v[ss],innerProduct(lhs_v(ss),rhs_v(ss)));
-  });
-  return ret;
-}
-  
-// outerProduct Scalar x Scalar -> Scalar
-//              Vector x Vector -> Matrix
-template<class ll,class rr>
-inline auto outerProduct (const Lattice<ll> &lhs,const Lattice<rr> &rhs) -> Lattice<decltype(outerProduct(ll(),rr()))>
-{
-  typedef decltype(coalescedRead(ll())) sll;
-  typedef decltype(coalescedRead(rr())) srr;
-  Lattice<decltype(outerProduct(ll(),rr()))> ret(rhs.Grid());
-  auto lhs_v = lhs.View();
-  auto rhs_v = rhs.View();
-  auto ret_v = ret.View();
-  accelerator_for(ss,rhs_v.size(),1,{
-    // FIXME had issues with scalar version of outer 
-    // Use vector [] operator and don't read coalesce this loop
-    ret_v[ss]=outerProduct(lhs_v[ss],rhs_v[ss]);
-  });
-  return ret;
-}
-NAMESPACE_END(Grid);
-#endif

Grid/lattice/Lattice_matrix_reduction.h

@@ -1,202 +0,0 @@
-/*************************************************************************************
-    Grid physics library, www.github.com/paboyle/Grid 
-    Source file: ./lib/lattice/Lattice_reduction.h
-    Copyright (C) 2015
-Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-    This program is 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_Eigen_Dense.h>
-
-#ifdef GRID_WARN_SUBOPTIMAL
-#warning "Optimisation alert all these reduction loops are NOT threaded "
-#endif     
-
-NAMESPACE_BEGIN(Grid);
-
-template<class vobj>
-static void sliceMaddMatrix (Lattice<vobj> &R,Eigen::MatrixXcd &aa,const Lattice<vobj> &X,const Lattice<vobj> &Y,int Orthog,RealD scale=1.0) 
-{    
-  typedef typename vobj::scalar_object sobj;
-  typedef typename vobj::scalar_type scalar_type;
-  typedef typename vobj::vector_type vector_type;
-
-  int Nblock = X.Grid()->GlobalDimensions()[Orthog];
-
-  GridBase *FullGrid  = X.Grid();
-  //  GridBase *SliceGrid = makeSubSliceGrid(FullGrid,Orthog);
-
-  //  Lattice<vobj> Xslice(SliceGrid);
-  //  Lattice<vobj> Rslice(SliceGrid);
-
-  assert( FullGrid->_simd_layout[Orthog]==1);
-
-  //FIXME package in a convenient iterator
-  //Should loop over a plane orthogonal to direction "Orthog"
-  int stride=FullGrid->_slice_stride[Orthog];
-  int block =FullGrid->_slice_block [Orthog];
-  int nblock=FullGrid->_slice_nblock[Orthog];
-  int ostride=FullGrid->_ostride[Orthog];
-  auto X_v = X.View();
-  auto Y_v = Y.View();
-  auto R_v = R.View();
-  thread_region
-  {
-    std::vector<vobj> s_x(Nblock);
-
-    thread_loop_collapse2( (int n=0;n<nblock;n++),{
-      for(int b=0;b<block;b++){
-	int o  = n*stride + b;
-
-	for(int i=0;i<Nblock;i++){
-	  s_x[i] = X_v[o+i*ostride];
-	}
-
-	vobj dot;
-	for(int i=0;i<Nblock;i++){
-	  dot = Y_v[o+i*ostride];
-	  for(int j=0;j<Nblock;j++){
-	    dot = dot + s_x[j]*(scale*aa(j,i));
-	  }
-	  R_v[o+i*ostride]=dot;
-	}
-      }});
-  }
-};
-
-template<class vobj>
-static void sliceMulMatrix (Lattice<vobj> &R,Eigen::MatrixXcd &aa,const Lattice<vobj> &X,int Orthog,RealD scale=1.0) 
-{    
-  typedef typename vobj::scalar_object sobj;
-  typedef typename vobj::scalar_type scalar_type;
-  typedef typename vobj::vector_type vector_type;
-
-  int Nblock = X.Grid()->GlobalDimensions()[Orthog];
-
-  GridBase *FullGrid  = X.Grid();
-  assert( FullGrid->_simd_layout[Orthog]==1);
-
-  //FIXME package in a convenient iterator
-  //Should loop over a plane orthogonal to direction "Orthog"
-  int stride=FullGrid->_slice_stride[Orthog];
-  int block =FullGrid->_slice_block [Orthog];
-  int nblock=FullGrid->_slice_nblock[Orthog];
-  int ostride=FullGrid->_ostride[Orthog];
-
-  auto X_v = X.View();
-  auto R_v = R.View();
-
-  thread_region
-  {
-    std::vector<vobj> s_x(Nblock);
-    
-    thread_loop_collapse2( (int n=0;n<nblock;n++),{
-      for(int b=0;b<block;b++){
-	int o  = n*stride + b;
-
-	for(int i=0;i<Nblock;i++){
-	  s_x[i] = X_v[o+i*ostride];
-	}
-
-	vobj dot;
-	for(int i=0;i<Nblock;i++){
-	  dot = s_x[0]*(scale*aa(0,i));
-	  for(int j=1;j<Nblock;j++){
-	    dot = dot + s_x[j]*(scale*aa(j,i));
-	  }
-	  R_v[o+i*ostride]=dot;
-	}
-    }});
-  }
-
-};
-
-
-template<class vobj>
-static void sliceInnerProductMatrix(  Eigen::MatrixXcd &mat, const Lattice<vobj> &lhs,const Lattice<vobj> &rhs,int Orthog) 
-{
-  typedef typename vobj::scalar_object sobj;
-  typedef typename vobj::scalar_type scalar_type;
-  typedef typename vobj::vector_type vector_type;
-  
-  GridBase *FullGrid  = lhs.Grid();
-  //  GridBase *SliceGrid = makeSubSliceGrid(FullGrid,Orthog);
-  
-  int Nblock = FullGrid->GlobalDimensions()[Orthog];
-  
-  //  Lattice<vobj> Lslice(SliceGrid);
-  //  Lattice<vobj> Rslice(SliceGrid);
-  
-  mat = Eigen::MatrixXcd::Zero(Nblock,Nblock);
-
-  assert( FullGrid->_simd_layout[Orthog]==1);
-  //  int nh =  FullGrid->_ndimension;
-  //  int nl = SliceGrid->_ndimension;
-  //  int nl = nh-1;
-
-  //FIXME package in a convenient iterator
-  //Should loop over a plane orthogonal to direction "Orthog"
-  int stride=FullGrid->_slice_stride[Orthog];
-  int block =FullGrid->_slice_block [Orthog];
-  int nblock=FullGrid->_slice_nblock[Orthog];
-  int ostride=FullGrid->_ostride[Orthog];
-
-  typedef typename vobj::vector_typeD vector_typeD;
-  auto lhs_v = lhs.View();
-  auto rhs_v = rhs.View();
-  thread_region {
-    std::vector<vobj> Left(Nblock);
-    std::vector<vobj> Right(Nblock);
-    Eigen::MatrixXcd  mat_thread = Eigen::MatrixXcd::Zero(Nblock,Nblock);
-
-    thread_loop_collapse2((int n=0;n<nblock;n++),{
-      for(int b=0;b<block;b++){
-
-	int o  = n*stride + b;
-
-	for(int i=0;i<Nblock;i++){
-	  Left [i] = lhs_v[o+i*ostride];
-	  Right[i] = rhs_v[o+i*ostride];
-	}
-
-	for(int i=0;i<Nblock;i++){
-	  for(int j=0;j<Nblock;j++){
-	    auto tmp = innerProduct(Left[i],Right[j]);
-	    auto rtmp = TensorRemove(tmp);
-	    ComplexD z = Reduce(rtmp);
-	    mat_thread(i,j) += std::complex<double>(real(z),imag(z));
-	  }}
-    }});
-    thread_critical {
-      mat += mat_thread;
-    }  
-  }
-
-  for(int i=0;i<Nblock;i++){
-    for(int j=0;j<Nblock;j++){
-      ComplexD sum = mat(i,j);
-      FullGrid->GlobalSum(sum);
-      mat(i,j)=sum;
-    }}
-
-  return;
-}
-
-NAMESPACE_END(Grid);
-
-
-

Grid/lattice/Lattice_peekpoke.h

@@ -1,217 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/lattice/Lattice_peekpoke.h
-
-    Copyright (C) 2015
-
-Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-
-    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_LATTICE_PEEK_H
-#define GRID_LATTICE_PEEK_H
-
-///////////////////////////////////////////////
-// Peeking and poking around
-///////////////////////////////////////////////
-
-NAMESPACE_BEGIN(Grid);
-
-
-// FIXME accelerator_loop and accelerator_inline these
-////////////////////////////////////////////////////////////////////////////////////////////////////
-// Peek internal indices of a Lattice object
-////////////////////////////////////////////////////////////////////////////////////////////////////
-template<int Index,class vobj> 
-auto PeekIndex(const Lattice<vobj> &lhs,int i) -> Lattice<decltype(peekIndex<Index>(vobj(),i))>
-{
-  Lattice<decltype(peekIndex<Index>(vobj(),i))> ret(lhs.Grid());
-  ret.Checkerboard()=lhs.Checkerboard();
-  auto ret_v = ret.View();
-  auto lhs_v = lhs.View();
-  thread_for( ss, lhs_v.size(), {
-    ret_v[ss] = peekIndex<Index>(lhs_v[ss],i);
-  });
-  return ret;
-};
-template<int Index,class vobj> 
-auto PeekIndex(const Lattice<vobj> &lhs,int i,int j) -> Lattice<decltype(peekIndex<Index>(vobj(),i,j))>
-{
-  Lattice<decltype(peekIndex<Index>(vobj(),i,j))> ret(lhs.Grid());
-  ret.Checkerboard()=lhs.Checkerboard();
-  auto ret_v = ret.View();
-  auto lhs_v = lhs.View();
-  thread_for( ss, lhs_v.size(), {
-    ret_v[ss] = peekIndex<Index>(lhs_v[ss],i,j);
-  });
-  return ret;
-};
-
-////////////////////////////////////////////////////////////////////////////////////////////////////
-// Poke internal indices of a Lattice object
-////////////////////////////////////////////////////////////////////////////////////////////////////
-template<int Index,class vobj>  
-void PokeIndex(Lattice<vobj> &lhs,const Lattice<decltype(peekIndex<Index>(vobj(),0))> & rhs,int i)
-{
-  auto rhs_v = rhs.View();
-  auto lhs_v = lhs.View();
-  thread_for( ss, lhs_v.size(), {
-    pokeIndex<Index>(lhs_v[ss],rhs_v[ss],i);
-  });
-}
-template<int Index,class vobj> 
-void PokeIndex(Lattice<vobj> &lhs,const Lattice<decltype(peekIndex<Index>(vobj(),0,0))> & rhs,int i,int j)
-{
-  auto rhs_v = rhs.View();
-  auto lhs_v = lhs.View();
-  thread_for( ss, lhs_v.size(), {
-    pokeIndex<Index>(lhs_v[ss],rhs_v[ss],i,j);
-  });
-}
-
-//////////////////////////////////////////////////////
-// Poke a scalar object into the SIMD array
-//////////////////////////////////////////////////////
-template<class vobj,class sobj> 
-void pokeSite(const sobj &s,Lattice<vobj> &l,const Coordinate &site){
-
-  GridBase *grid=l.Grid();
-
-  typedef typename vobj::scalar_type scalar_type;
-  typedef typename vobj::vector_type vector_type;
-
-  int Nsimd = grid->Nsimd();
-
-  assert( l.Checkerboard()== l.Grid()->CheckerBoard(site));
-  assert( sizeof(sobj)*Nsimd == sizeof(vobj));
-
-  int rank,odx,idx;
-  // Optional to broadcast from node 0.
-  grid->GlobalCoorToRankIndex(rank,odx,idx,site);
-  grid->Broadcast(grid->BossRank(),s);
-
-  // extract-modify-merge cycle is easiest way and this is not perf critical
-  ExtractBuffer<sobj> buf(Nsimd);
-  auto l_v = l.View();
-  if ( rank == grid->ThisRank() ) {
-    extract(l_v[odx],buf);
-    buf[idx] = s;
-    merge(l_v[odx],buf);
-  }
-
-  return;
-};
-
-
-//////////////////////////////////////////////////////////
-// Peek a scalar object from the SIMD array
-//////////////////////////////////////////////////////////
-template<class vobj,class sobj>
-void peekSite(sobj &s,const Lattice<vobj> &l,const Coordinate &site){
-        
-  GridBase *grid=l.Grid();
-
-  typedef typename vobj::scalar_type scalar_type;
-  typedef typename vobj::vector_type vector_type;
-
-  int Nsimd = grid->Nsimd();
-
-  assert( l.Checkerboard() == l.Grid()->CheckerBoard(site));
-
-  int rank,odx,idx;
-  grid->GlobalCoorToRankIndex(rank,odx,idx,site);
-
-  ExtractBuffer<sobj> buf(Nsimd);
-  auto l_v = l.View();
-  extract(l_v[odx],buf);
-
-  s = buf[idx];
-
-  grid->Broadcast(rank,s);
-
-  return;
-};
-
-
-//////////////////////////////////////////////////////////
-// Peek a scalar object from the SIMD array
-//////////////////////////////////////////////////////////
-template<class vobj,class sobj>
-void peekLocalSite(sobj &s,const Lattice<vobj> &l,Coordinate &site){
-        
-  GridBase *grid = l.Grid();
-
-  typedef typename vobj::scalar_type scalar_type;
-  typedef typename vobj::vector_type vector_type;
-
-  int Nsimd = grid->Nsimd();
-
-  assert( l.Checkerboard()== l.Grid()->CheckerBoard(site));
-  assert( sizeof(sobj)*Nsimd == sizeof(vobj));
-
-  static const int words=sizeof(vobj)/sizeof(vector_type);
-  int odx,idx;
-  idx= grid->iIndex(site);
-  odx= grid->oIndex(site);
-  
-  auto l_v = l.View();
-  scalar_type * vp = (scalar_type *)&l_v[odx];
-  scalar_type * pt = (scalar_type *)&s;
-      
-  for(int w=0;w<words;w++){
-    pt[w] = vp[idx+w*Nsimd];
-  }
-      
-  return;
-};
-
-template<class vobj,class sobj>
-void pokeLocalSite(const sobj &s,Lattice<vobj> &l,Coordinate &site){
-
-  GridBase *grid=l.Grid();
-
-  typedef typename vobj::scalar_type scalar_type;
-  typedef typename vobj::vector_type vector_type;
-
-  int Nsimd = grid->Nsimd();
-
-  assert( l.Checkerboard()== l.Grid()->CheckerBoard(site));
-  assert( sizeof(sobj)*Nsimd == sizeof(vobj));
-
-  static const int words=sizeof(vobj)/sizeof(vector_type);
-  int odx,idx;
-  idx= grid->iIndex(site);
-  odx= grid->oIndex(site);
-
-  auto l_v = l.View();
-  scalar_type * vp = (scalar_type *)&l_v[odx];
-  scalar_type * pt = (scalar_type *)&s;
-  for(int w=0;w<words;w++){
-    vp[idx+w*Nsimd] = pt[w];
-  }
-
-  return;
-};
-
-NAMESPACE_END(Grid);
-#endif
-

Grid/lattice/Lattice_reduction_gpu.h

@@ -1,226 +0,0 @@
-NAMESPACE_BEGIN(Grid);
-
-#define WARP_SIZE 32
-extern cudaDeviceProp *gpu_props;
-__device__ unsigned int retirementCount = 0;
-
-template <class Iterator>
-unsigned int nextPow2(Iterator x) {
-  --x;
-  x |= x >> 1;
-  x |= x >> 2;
-  x |= x >> 4;
-  x |= x >> 8;
-  x |= x >> 16;
-  return ++x;
-}
-
-template <class Iterator>
-void getNumBlocksAndThreads(const Iterator n, const size_t sizeofsobj, Iterator &threads, Iterator &blocks) {
-  
-  int device;
-  cudaGetDevice(&device);
-  
-  Iterator warpSize            = gpu_props[device].warpSize;
-  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);
-  }
-  
-  // let the number of threads in a block be a multiple of 2, starting from warpSize
-  threads = warpSize;
-  while( 2*threads*sizeofsobj < sharedMemPerBlock && 2*threads <= maxThreadsPerBlock ) threads *= 2;
-  // keep all the streaming multiprocessors busy
-  blocks = nextPow2(multiProcessorCount);
-  
-}
-
-template <class sobj, class Iterator>
-__device__ void reduceBlock(volatile sobj *sdata, sobj mySum, const Iterator tid) {
-  
-  Iterator blockSize = blockDim.x;
-  
-  // cannot use overloaded operators for sobj as they are not volatile-qualified
-  memcpy((void *)&sdata[tid], (void *)&mySum, sizeof(sobj));
-  __syncwarp();
-  
-  const Iterator VEC = WARP_SIZE;
-  const Iterator vid = tid & (VEC-1);
-  
-  sobj beta, temp;
-  memcpy((void *)&beta, (void *)&mySum, sizeof(sobj));
-  
-  for (int i = VEC/2; i > 0; i>>=1) {
-    if (vid < i) {
-      memcpy((void *)&temp, (void *)&sdata[tid+i], sizeof(sobj));
-      beta += temp;
-      memcpy((void *)&sdata[tid], (void *)&beta, sizeof(sobj));
-    }
-    __syncwarp();
-  }
-  __syncthreads();
-  
-  if (threadIdx.x == 0) {
-    beta  = Zero();
-    for (Iterator i = 0; i < blockSize; i += VEC) {
-      memcpy((void *)&temp, (void *)&sdata[i], sizeof(sobj));
-      beta  += temp;
-    }
-    memcpy((void *)&sdata[0], (void *)&beta, sizeof(sobj));
-  }
-  __syncthreads();
-}
-
-
-template <class vobj, class sobj, class Iterator>
-__device__ void reduceBlocks(const vobj *g_idata, sobj *g_odata, Iterator n) 
-{
-  constexpr Iterator nsimd = vobj::Nsimd();
-  
-  Iterator blockSize = blockDim.x;
-  
-  // force shared memory alignment
-  extern __shared__ __align__(COALESCE_GRANULARITY) unsigned char shmem_pointer[];
-  // it's not possible to have two extern __shared__ arrays with same name
-  // but different types in different scopes -- need to cast each time
-  sobj *sdata = (sobj *)shmem_pointer;
-  
-  // first level of reduction,
-  // each thread writes result in mySum
-  Iterator tid = threadIdx.x;
-  Iterator i = blockIdx.x*(blockSize*2) + threadIdx.x;
-  Iterator gridSize = blockSize*2*gridDim.x;
-  sobj mySum = Zero();
-  
-  while (i < n) {
-    Iterator lane = i % nsimd;
-    Iterator ss   = i / nsimd;
-    auto tmp = extractLane(lane,g_idata[ss]);
-    sobj tmpD;
-    tmpD=tmp;
-    mySum   +=tmpD;
-    
-    if (i + blockSize < n) {
-      lane = (i+blockSize) % nsimd;
-      ss   = (i+blockSize) / nsimd;
-      tmp = extractLane(lane,g_idata[ss]);
-      tmpD = tmp;
-      mySum += tmpD;
-    }
-    i += gridSize;
-  }
-  
-  // copy mySum to shared memory and perform
-  // reduction for all threads in this block
-  reduceBlock(sdata, mySum, tid);
-  if (tid == 0) g_odata[blockIdx.x] = sdata[0];
-}
-
-template <class vobj, class sobj,class Iterator>
-__global__ void reduceKernel(const vobj *lat, sobj *buffer, Iterator n) {
-  
-  Iterator blockSize = blockDim.x;
-  
-  // perform reduction for this block and
-  // write result to global memory buffer
-  reduceBlocks(lat, buffer, n);
-  
-  if (gridDim.x > 1) {
-    
-    const Iterator tid = threadIdx.x;
-    __shared__ bool amLast;
-    // force shared memory alignment
-    extern __shared__ __align__(COALESCE_GRANULARITY) unsigned char shmem_pointer[];
-    // it's not possible to have two extern __shared__ arrays with same name
-    // but different types in different scopes -- need to cast each time
-    sobj *smem = (sobj *)shmem_pointer;
-    
-    // wait until all outstanding memory instructions in this thread are finished
-    __threadfence();
-    
-    if (tid==0) {
-      unsigned int ticket = atomicInc(&retirementCount, gridDim.x);
-      // true if this block is the last block to be done
-      amLast = (ticket == gridDim.x-1);
-    }
-    
-    // each thread must read the correct value of amLast
-    __syncthreads();
-    
-    if (amLast) {
-      // reduce buffer[0], ..., buffer[gridDim.x-1]
-      Iterator i = tid;
-      sobj mySum = Zero();
-      
-      while (i < gridDim.x) {
-        mySum += buffer[i];
-        i += blockSize;
-      }
-      
-      reduceBlock(smem, mySum, tid);
-      
-      if (tid==0) {
-        buffer[0] = smem[0];
-        // reset count variable
-        retirementCount = 0;
-      }
-    }
-  }
-}
-
-/////////////////////////////////////////////////////////////////////////////////////////////////////////
-// Possibly promote to double and sum
-/////////////////////////////////////////////////////////////////////////////////////////////////////////
-template <class vobj>
-inline typename vobj::scalar_objectD sumD_gpu(const vobj *lat, Integer osites) 
-{
-  typedef typename vobj::scalar_objectD sobj;
-  typedef decltype(lat) Iterator;
-  
-  Integer nsimd= vobj::Nsimd();
-  Integer size = osites*nsimd;
-
-  Integer numThreads, numBlocks;
-  getNumBlocksAndThreads(size, sizeof(sobj), numThreads, numBlocks);
-  Integer smemSize = numThreads * sizeof(sobj);
-
-  Vector<sobj> buffer(numBlocks);
-  sobj *buffer_v = &buffer[0];
-  
-  reduceKernel<<< numBlocks, numThreads, smemSize >>>(lat, buffer_v, size);
-  cudaDeviceSynchronize();
-  
-  cudaError err = cudaGetLastError();
-  if ( cudaSuccess != err ) {
-    printf("Cuda error %s\n",cudaGetErrorString( err ));
-    exit(0);
-  }
-  auto result = buffer_v[0];
-  return result;
-}
-/////////////////////////////////////////////////////////////////////////////////////////////////////////
-// Return as same precision as input performing reduction in double precision though
-/////////////////////////////////////////////////////////////////////////////////////////////////////////
-template <class vobj>
-inline typename vobj::scalar_object sum_gpu(const vobj *lat, Integer osites) 
-{
-  typedef typename vobj::scalar_object sobj;
-  sobj result;
-  result = sumD_gpu(lat,osites);
-  return result;
-}
-
-
-
-NAMESPACE_END(Grid);

Grid/lattice/Lattice_rng.h

@@ -1,525 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/lattice/Lattice_rng.h
-
-    Copyright (C) 2015
-
-    Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-    Author: Guido Cossu <guido.cossu@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 */
-#ifndef GRID_LATTICE_RNG_H
-#define GRID_LATTICE_RNG_H
-
-#include <random>
-
-#ifdef RNG_SITMO
-#include <Grid/sitmo_rng/sitmo_prng_engine.hpp>
-#endif 
-
-#if defined(RNG_SITMO)
-#define RNG_FAST_DISCARD
-#else 
-#undef  RNG_FAST_DISCARD
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-//////////////////////////////////////////////////////////////
-// Allow the RNG state to be less dense than the fine grid
-//////////////////////////////////////////////////////////////
-inline int RNGfillable(GridBase *coarse,GridBase *fine)
-{
-
-  int rngdims = coarse->_ndimension;
-
-  // trivially extended in higher dims, with locality guaranteeing RNG state is local to node
-  int lowerdims   = fine->_ndimension - coarse->_ndimension;
-  assert(lowerdims >= 0);
-  for(int d=0;d<lowerdims;d++){
-    assert(fine->_simd_layout[d]==1);
-    assert(fine->_processors[d]==1);
-  }
-
-  int multiplicity=1;
-  for(int d=0;d<lowerdims;d++){
-    multiplicity=multiplicity*fine->_rdimensions[d];
-  }
-  // local and global volumes subdivide cleanly after SIMDization
-  for(int d=0;d<rngdims;d++){
-    int fd= d+lowerdims;
-    assert(coarse->_processors[d]  == fine->_processors[fd]);
-    assert(coarse->_simd_layout[d] == fine->_simd_layout[fd]);
-    assert(((fine->_rdimensions[fd] / coarse->_rdimensions[d])* coarse->_rdimensions[d])==fine->_rdimensions[fd]); 
-
-    multiplicity = multiplicity *fine->_rdimensions[fd] / coarse->_rdimensions[d]; 
-  }
-  return multiplicity;
-}
-
-  
-// merge of April 11 2017
-// this function is necessary for the LS vectorised field
-inline int RNGfillable_general(GridBase *coarse,GridBase *fine)
-{
-  int rngdims = coarse->_ndimension;
-    
-  // trivially extended in higher dims, with locality guaranteeing RNG state is local to node
-  int lowerdims   = fine->_ndimension - coarse->_ndimension;  assert(lowerdims >= 0);
-  // assumes that the higher dimensions are not using more processors
-  // all further divisions are local
-  for(int d=0;d<lowerdims;d++) assert(fine->_processors[d]==1);
-  for(int d=0;d<rngdims;d++) assert(coarse->_processors[d] == fine->_processors[d+lowerdims]);
-
-  // then divide the number of local sites
-  // check that the total number of sims agree, meanse the iSites are the same
-  assert(fine->Nsimd() == coarse->Nsimd());
-
-  // check that the two grids divide cleanly
-  assert( (fine->lSites() / coarse->lSites() ) * coarse->lSites() == fine->lSites() );
-
-  return fine->lSites() / coarse->lSites();
-}
-  
-// real scalars are one component
-template<class scalar,class distribution,class generator> 
-void fillScalar(scalar &s,distribution &dist,generator & gen)
-{
-  s=dist(gen);
-}
-template<class distribution,class generator> 
-void fillScalar(ComplexF &s,distribution &dist, generator &gen)
-{
-  //  s=ComplexF(dist(gen),dist(gen));
-  s.real(dist(gen));
-  s.imag(dist(gen));
-}
-template<class distribution,class generator> 
-void fillScalar(ComplexD &s,distribution &dist,generator &gen)
-{
-  //  s=ComplexD(dist(gen),dist(gen));
-  s.real(dist(gen));
-  s.imag(dist(gen));
-}
-  
-class GridRNGbase {
-public:
-  // One generator per site.
-  // Uniform and Gaussian distributions from these generators.
-#ifdef RNG_RANLUX
-  typedef std::ranlux48 RngEngine;
-  typedef uint64_t      RngStateType;
-  static const int RngStateCount = 15;
-#endif 
-#ifdef RNG_MT19937 
-  typedef std::mt19937 RngEngine;
-  typedef uint32_t     RngStateType;
-  static const int     RngStateCount = std::mt19937::state_size;
-#endif
-#ifdef RNG_SITMO
-  typedef sitmo::prng_engine 	RngEngine;
-  typedef uint64_t    	RngStateType;
-  static const int    	RngStateCount = 13;
-#endif
-
-  std::vector<RngEngine>                             _generators;
-  std::vector<std::uniform_real_distribution<RealD> > _uniform;
-  std::vector<std::normal_distribution<RealD> >       _gaussian;
-  std::vector<std::discrete_distribution<int32_t> >   _bernoulli;
-  std::vector<std::uniform_int_distribution<uint32_t> > _uid;
-
-  ///////////////////////
-  // support for parallel init
-  ///////////////////////
-#ifdef RNG_FAST_DISCARD
-  static void Skip(RngEngine &eng,uint64_t site)
-  {
-    /////////////////////////////////////////////////////////////////////////////////////
-    // Skip by 2^40 elements between successive lattice sites
-    // This goes by 10^12.
-    // Consider quenched updating; likely never exceeding rate of 1000 sweeps
-    // per second on any machine. This gives us of order 10^9 seconds, or 100 years
-    // skip ahead.
-    // For HMC unlikely to go at faster than a solve per second, and 
-    // tens of seconds per trajectory so this is clean in all reasonable cases,
-    // and margin of safety is orders of magnitude.
-    // We could hack Sitmo to skip in the higher order words of state if necessary
-      //
-      // Replace with 2^30 ; avoid problem on large volumes
-      //
-    /////////////////////////////////////////////////////////////////////////////////////
-    //      uint64_t skip = site+1;  //   Old init Skipped then drew.  Checked compat with faster init
-    const int shift = 30;
-
-    ////////////////////////////////////////////////////////////////////
-    // Weird compiler bug in Intel 2018.1 under O3 was generating 32bit and not 64 bit left shift.
-    ////////////////////////////////////////////////////////////////////
-    volatile uint64_t skip = site;
-
-    skip = skip<<shift;
-
-    assert((skip >> shift)==site); // check for overflow
-
-    eng.discard(skip);
-    //      std::cout << " Engine  " <<site << " state " <<eng<<std::endl;
-  } 
-#endif
-  static RngEngine Reseed(RngEngine &eng)
-  {
-    std::vector<uint32_t> newseed;
-    std::uniform_int_distribution<uint32_t> uid;
-    return Reseed(eng,newseed,uid);
-  }
-  static RngEngine Reseed(RngEngine &eng,std::vector<uint32_t> & newseed,
-			  std::uniform_int_distribution<uint32_t> &uid)
-  {
-    const int reseeds=4;
-      
-    newseed.resize(reseeds);
-    for(int i=0;i<reseeds;i++){
-      newseed[i] = uid(eng);
-    }
-    std::seed_seq sseq(newseed.begin(),newseed.end());
-    return RngEngine(sseq);
-  }    
-
-  void GetState(std::vector<RngStateType> & saved,RngEngine &eng) {
-    saved.resize(RngStateCount);
-    std::stringstream ss;
-    ss<<eng;
-    ss.seekg(0,ss.beg);
-    for(int i=0;i<RngStateCount;i++){
-      ss>>saved[i];
-    }
-  }
-  void GetState(std::vector<RngStateType> & saved,int gen) {
-    GetState(saved,_generators[gen]);
-  }
-  void SetState(std::vector<RngStateType> & saved,RngEngine &eng){
-    assert(saved.size()==RngStateCount);
-    std::stringstream ss;
-    for(int i=0;i<RngStateCount;i++){
-      ss<< saved[i]<<" ";
-    }
-    ss.seekg(0,ss.beg);
-    ss>>eng;
-  }
-  void SetState(std::vector<RngStateType> & saved,int gen){
-    SetState(saved,_generators[gen]);
-  }
-  void SetEngine(RngEngine &Eng, int gen){
-    _generators[gen]=Eng;
-  }
-  void GetEngine(RngEngine &Eng, int gen){
-    Eng=_generators[gen];
-  }
-  template<class source> void Seed(source &src, int gen)
-  {
-    _generators[gen] = RngEngine(src);
-  }    
-};
-
-class GridSerialRNG : public GridRNGbase {
-public:
-
-  GridSerialRNG() : GridRNGbase() {
-    _generators.resize(1);
-    _uniform.resize(1,std::uniform_real_distribution<RealD>{0,1});
-    _gaussian.resize(1,std::normal_distribution<RealD>(0.0,1.0) );
-    _bernoulli.resize(1,std::discrete_distribution<int32_t>{1,1});
-    _uid.resize(1,std::uniform_int_distribution<uint32_t>() );
-  }
-
-  template <class sobj,class distribution> inline void fill(sobj &l,std::vector<distribution> &dist){
-
-    typedef typename sobj::scalar_type scalar_type;
- 
-    int words = sizeof(sobj)/sizeof(scalar_type);
-
-    scalar_type *buf = (scalar_type *) & l;
-
-    dist[0].reset();
-    for(int idx=0;idx<words;idx++){
-      fillScalar(buf[idx],dist[0],_generators[0]);
-    }
-
-    CartesianCommunicator::BroadcastWorld(0,(void *)&l,sizeof(l));
-
-  }
-
-  template <class distribution>  inline void fill(ComplexF &l,std::vector<distribution> &dist){
-    dist[0].reset();
-    fillScalar(l,dist[0],_generators[0]);
-    CartesianCommunicator::BroadcastWorld(0,(void *)&l,sizeof(l));
-  }
-  template <class distribution>  inline void fill(ComplexD &l,std::vector<distribution> &dist){
-    dist[0].reset();
-    fillScalar(l,dist[0],_generators[0]);
-    CartesianCommunicator::BroadcastWorld(0,(void *)&l,sizeof(l));
-  }
-  template <class distribution>  inline void fill(RealF &l,std::vector<distribution> &dist){
-    dist[0].reset();
-    fillScalar(l,dist[0],_generators[0]);
-    CartesianCommunicator::BroadcastWorld(0,(void *)&l,sizeof(l));
-  }
-  template <class distribution>  inline void fill(RealD &l,std::vector<distribution> &dist){
-    dist[0].reset();
-    fillScalar(l,dist[0],_generators[0]);
-    CartesianCommunicator::BroadcastWorld(0,(void *)&l,sizeof(l));
-  }
-  // vector fill
-  template <class distribution>  inline void fill(vComplexF &l,std::vector<distribution> &dist){
-    RealF *pointer=(RealF *)&l;
-    dist[0].reset();
-    for(int i=0;i<2*vComplexF::Nsimd();i++){
-      fillScalar(pointer[i],dist[0],_generators[0]);
-    }
-    CartesianCommunicator::BroadcastWorld(0,(void *)&l,sizeof(l));
-  }
-  template <class distribution>  inline void fill(vComplexD &l,std::vector<distribution> &dist){
-    RealD *pointer=(RealD *)&l;
-    dist[0].reset();
-    for(int i=0;i<2*vComplexD::Nsimd();i++){
-      fillScalar(pointer[i],dist[0],_generators[0]);
-    }
-    CartesianCommunicator::BroadcastWorld(0,(void *)&l,sizeof(l));
-  }
-  template <class distribution>  inline void fill(vRealF &l,std::vector<distribution> &dist){
-    RealF *pointer=(RealF *)&l;
-    dist[0].reset();
-    for(int i=0;i<vRealF::Nsimd();i++){
-      fillScalar(pointer[i],dist[0],_generators[0]);
-    }
-    CartesianCommunicator::BroadcastWorld(0,(void *)&l,sizeof(l));
-  }
-  template <class distribution>  inline void fill(vRealD &l,std::vector<distribution> &dist){
-    RealD *pointer=(RealD *)&l;
-    dist[0].reset();
-    for(int i=0;i<vRealD::Nsimd();i++){
-      fillScalar(pointer[i],dist[0],_generators[0]);
-    }
-    CartesianCommunicator::BroadcastWorld(0,(void *)&l,sizeof(l));
-  }
-    
-  void SeedFixedIntegers(const std::vector<int> &seeds){
-    CartesianCommunicator::BroadcastWorld(0,(void *)&seeds[0],sizeof(int)*seeds.size());
-    std::seed_seq src(seeds.begin(),seeds.end());
-    Seed(src,0);
-  }
-
-    void SeedUniqueString(const std::string &s){
-      std::vector<int> seeds;
-      std::stringstream sha;
-      seeds = GridChecksum::sha256_seeds(s);
-      for(int i=0;i<seeds.size();i++) { 
-        sha << std::hex << seeds[i];
-      }
-      std::cout << GridLogMessage << "Intialising serial RNG with unique string '" 
-                << s << "'" << std::endl;
-      std::cout << GridLogMessage << "Seed SHA256: " << sha.str() << std::endl;
-      SeedFixedIntegers(seeds);
-    }
-};
-
-class GridParallelRNG : public GridRNGbase {
-private:
-  double _time_counter;
-  GridBase *_grid;
-  unsigned int _vol;
-
-public:
-  GridBase *Grid(void) const { return _grid; }
-  int generator_idx(int os,int is) {
-    return is*_grid->oSites()+os;
-  }
-
-  GridParallelRNG(GridBase *grid) : GridRNGbase() {
-    _grid = grid;
-    _vol  =_grid->iSites()*_grid->oSites();
-
-    _generators.resize(_vol);
-    _uniform.resize(_vol,std::uniform_real_distribution<RealD>{0,1});
-    _gaussian.resize(_vol,std::normal_distribution<RealD>(0.0,1.0) );
-    _bernoulli.resize(_vol,std::discrete_distribution<int32_t>{1,1});
-    _uid.resize(_vol,std::uniform_int_distribution<uint32_t>() );
-  }
-
-  template <class vobj,class distribution> inline void fill(Lattice<vobj> &l,std::vector<distribution> &dist){
-
-    typedef typename vobj::scalar_object scalar_object;
-    typedef typename vobj::scalar_type scalar_type;
-    typedef typename vobj::vector_type vector_type;
-
-    double inner_time_counter = usecond();
-
-    int multiplicity = RNGfillable_general(_grid, l.Grid()); // l has finer or same grid
-    int Nsimd  = _grid->Nsimd();  // guaranteed to be the same for l.Grid() too
-    int osites = _grid->oSites();  // guaranteed to be <= l.Grid()->oSites() by a factor multiplicity
-    int words  = sizeof(scalar_object) / sizeof(scalar_type);
-
-    auto l_v = l.View();
-    thread_for( ss, osites, {
-      ExtractBuffer<scalar_object> buf(Nsimd);
-      for (int m = 0; m < multiplicity; m++) {  // Draw from same generator multiplicity times
-
-	int sm = multiplicity * ss + m;  // Maps the generator site to the fine site
-
-	for (int si = 0; si < Nsimd; si++) {
-            
-	  int gdx = generator_idx(ss, si);  // index of generator state
-	  scalar_type *pointer = (scalar_type *)&buf[si];
-	  dist[gdx].reset();
-	  for (int idx = 0; idx < words; idx++) 
-	    fillScalar(pointer[idx], dist[gdx], _generators[gdx]);
-	}
-	// merge into SIMD lanes, FIXME suboptimal implementation
-	merge(l_v[sm], buf);
-      }
-      });
-    //    });
-
-    _time_counter += usecond()- inner_time_counter;
-  }
-
-    void SeedUniqueString(const std::string &s){
-      std::vector<int> seeds;
-      seeds = GridChecksum::sha256_seeds(s);
-      std::cout << GridLogMessage << "Intialising parallel RNG with unique string '" 
-                << s << "'" << std::endl;
-      std::cout << GridLogMessage << "Seed SHA256: " << GridChecksum::sha256_string(seeds) << std::endl;
-      SeedFixedIntegers(seeds);
-    }
-  void SeedFixedIntegers(const std::vector<int> &seeds){
-
-    // Everyone generates the same seed_seq based on input seeds
-    CartesianCommunicator::BroadcastWorld(0,(void *)&seeds[0],sizeof(int)*seeds.size());
-
-    std::seed_seq source(seeds.begin(),seeds.end());
-
-    RngEngine master_engine(source);
-
-#ifdef RNG_FAST_DISCARD
-    ////////////////////////////////////////////////
-    // Skip ahead through a single stream.
-    // Applicable to SITMO and other has based/crypto RNGs
-    // Should be applicable to Mersenne Twister, but the C++11
-    // MT implementation does not implement fast discard even though
-    // in principle this is possible
-    ////////////////////////////////////////////////
-
-    // Everybody loops over global volume.
-    thread_for( gidx, _grid->_gsites, {
-	// Where is it?
-	int rank;
-	int o_idx;
-	int i_idx;
-
-	Coordinate gcoor;
-	_grid->GlobalIndexToGlobalCoor(gidx,gcoor);
-	_grid->GlobalCoorToRankIndex(rank,o_idx,i_idx,gcoor);
-	
-	// If this is one of mine we take it
-	if( 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 
-    ////////////////////////////////////////////////////////////////
-    // Machine and thread decomposition dependent seeding is efficient
-    // and maximally parallel; but NOT reproducible from machine to machine. 
-    // Not ideal, but fastest way to reseed all nodes.
-    ////////////////////////////////////////////////////////////////
-    {
-      // Obtain one Reseed per processor
-      int Nproc = _grid->ProcessorCount();
-      std::vector<RngEngine> seeders(Nproc);
-      int me= _grid->ThisRank();
-      for(int p=0;p<Nproc;p++){
-	seeders[p] = Reseed(master_engine);
-      }
-      master_engine = seeders[me];
-    }
-
-    {
-      // Obtain one reseeded generator per thread
-      int Nthread = GridThread::GetThreads();
-      std::vector<RngEngine> seeders(Nthread);
-      for(int t=0;t<Nthread;t++){
-	seeders[t] = Reseed(master_engine);
-      }
-
-      thread_for( t, Nthread, {
-	// set up one per local site in threaded fashion
-	std::vector<uint32_t> newseeds;
-	std::uniform_int_distribution<uint32_t> uid;	
-	for(int l=0;l<_grid->lSites();l++) {
-	  if ( (l%Nthread)==t ) {
-	    _generators[l] = Reseed(seeders[t],newseeds,uid);
-	  }
-	}
-      });
-    }
-#endif
-  }
-
-  void Report(){
-    std::cout << GridLogMessage << "Time spent in the fill() routine by GridParallelRNG: "<< _time_counter/1e3 << " ms" << std::endl;
-  }
-
-
-  ////////////////////////////////////////////////////////////////////////
-  // Support for rigorous test of RNG's
-  // Return uniform random uint32_t from requested site generator
-  ////////////////////////////////////////////////////////////////////////
-  uint32_t GlobalU01(int gsite){
-
-    uint32_t the_number;
-    // who
-    int rank,o_idx,i_idx;
-    Coordinate gcoor;
-    _grid->GlobalIndexToGlobalCoor(gsite,gcoor);
-    _grid->GlobalCoorToRankIndex(rank,o_idx,i_idx,gcoor);
-
-    // draw
-    int l_idx=generator_idx(o_idx,i_idx);
-    if( rank == _grid->ThisRank() ){
-      the_number = _uid[l_idx](_generators[l_idx]);
-    }
-      
-    // share & return
-    _grid->Broadcast(rank,(void *)&the_number,sizeof(the_number));
-    return the_number;
-  }
-
-};
-
-template <class vobj> inline void random(GridParallelRNG &rng,Lattice<vobj> &l)   { rng.fill(l,rng._uniform);  }
-template <class vobj> inline void gaussian(GridParallelRNG &rng,Lattice<vobj> &l) { rng.fill(l,rng._gaussian); }
-template <class vobj> inline void bernoulli(GridParallelRNG &rng,Lattice<vobj> &l){ rng.fill(l,rng._bernoulli);}
-
-template <class sobj> inline void random(GridSerialRNG &rng,sobj &l)   { rng.fill(l,rng._uniform  ); }
-template <class sobj> inline void gaussian(GridSerialRNG &rng,sobj &l) { rng.fill(l,rng._gaussian ); }
-template <class sobj> inline void bernoulli(GridSerialRNG &rng,sobj &l){ rng.fill(l,rng._bernoulli); }
-
-NAMESPACE_END(Grid);
-#endif

Grid/lattice/Lattice_trace.h

@@ -1,69 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/lattice/Lattice_trace.h
-
-    Copyright (C) 2015
-
-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 */
-#ifndef GRID_LATTICE_TRACE_H
-#define GRID_LATTICE_TRACE_H
-
-///////////////////////////////////////////////
-// Tracing, transposing, peeking, poking
-///////////////////////////////////////////////
-
-NAMESPACE_BEGIN(Grid);
-
-////////////////////////////////////////////////////////////////////////////////////////////////////
-// Trace
-////////////////////////////////////////////////////////////////////////////////////////////////////
-template<class vobj>
-inline auto trace(const Lattice<vobj> &lhs)  -> Lattice<decltype(trace(vobj()))>
-{
-  Lattice<decltype(trace(vobj()))> ret(lhs.Grid());
-  auto ret_v = ret.View();
-  auto lhs_v = lhs.View();
-  accelerator_for( ss, lhs_v.size(), vobj::Nsimd(), {
-    coalescedWrite(ret_v[ss], trace(lhs_v(ss)));
-  });
-  return ret;
-};
-    
-////////////////////////////////////////////////////////////////////////////////////////////////////
-// Trace Index level dependent operation
-////////////////////////////////////////////////////////////////////////////////////////////////////
-template<int Index,class vobj>
-inline auto TraceIndex(const Lattice<vobj> &lhs) -> Lattice<decltype(traceIndex<Index>(vobj()))>
-{
-  Lattice<decltype(traceIndex<Index>(vobj()))> ret(lhs.Grid());
-  auto ret_v = ret.View();
-  auto lhs_v = lhs.View();
-  accelerator_for( ss, lhs_v.size(), vobj::Nsimd(), {
-    coalescedWrite(ret_v[ss], traceIndex<Index>(lhs_v(ss)));
-  });
-  return ret;
-};
-
-NAMESPACE_END(Grid);
-#endif
-

Grid/lattice/Lattice_transpose.h

@@ -1,68 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/lattice/Lattice_transpose.h
-
-    Copyright (C) 2015
-
-Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
-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 */
-#ifndef GRID_LATTICE_TRANSPOSE_H
-#define GRID_LATTICE_TRANSPOSE_H
-
-///////////////////////////////////////////////
-// Transpose
-///////////////////////////////////////////////
-
-NAMESPACE_BEGIN(Grid);
-
-////////////////////////////////////////////////////////////////////////////////////////////////////
-// Transpose
-////////////////////////////////////////////////////////////////////////////////////////////////////
-template<class vobj>
-inline Lattice<vobj> transpose(const Lattice<vobj> &lhs){
-  Lattice<vobj> ret(lhs.Grid());
-  auto ret_v = ret.View();
-  auto lhs_v = lhs.View();
-  accelerator_for(ss,lhs_v.size(),vobj::Nsimd(),{
-    coalescedWrite(ret_v[ss], transpose(lhs_v(ss)));
-  });
-  return ret;
-};
-    
-////////////////////////////////////////////////////////////////////////////////////////////////////
-// Index level dependent transpose
-////////////////////////////////////////////////////////////////////////////////////////////////////
-template<int Index,class vobj>
-inline auto TransposeIndex(const Lattice<vobj> &lhs) -> Lattice<decltype(transposeIndex<Index>(vobj()))>
-{
-  Lattice<decltype(transposeIndex<Index>(vobj()))> ret(lhs.Grid());
-  auto ret_v = ret.View();
-  auto lhs_v = lhs.View();
-  accelerator_for(ss,lhs_v.size(),vobj::Nsimd(),{
-    coalescedWrite(ret_v[ss] , transposeIndex<Index>(lhs_v(ss)));
-  });
-  return ret;
-};
-
-NAMESPACE_END(Grid);
-#endif

Grid/lattice/Lattice_unary.h

@@ -1,80 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/lattice/Lattice_unary.h
-
-    Copyright (C) 2015
-
-Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: neo <cossu@post.kek.jp>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is 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_LATTICE_UNARY_H
-#define GRID_LATTICE_UNARY_H
-
-NAMESPACE_BEGIN(Grid);
-
-template<class obj> Lattice<obj> pow(const Lattice<obj> &rhs_i,RealD y){
-  Lattice<obj> ret_i(rhs_i.Grid());
-  auto rhs = rhs_i.View();
-  auto ret = ret_i.View();
-  ret.Checkerboard() = rhs.Checkerboard();
-  accelerator_for(ss,rhs.size(),1,{
-      ret[ss]=pow(rhs[ss],y);
-  });
-  return ret_i;
-}
-template<class obj> Lattice<obj> mod(const Lattice<obj> &rhs_i,Integer y){
-  Lattice<obj> ret_i(rhs_i.Grid());
-  auto rhs = rhs_i.View();
-  auto ret = ret_i.View();
-  ret.Checkerboard() = rhs.Checkerboard();
-  accelerator_for(ss,rhs.size(),obj::Nsimd(),{
-    coalescedWrite(ret[ss],mod(rhs(ss),y));
-  });
-  return ret_i;
-}
-
-template<class obj> Lattice<obj> div(const Lattice<obj> &rhs_i,Integer y){
-  Lattice<obj> ret_i(rhs_i.Grid());
-  auto ret = ret_i.View();
-  auto rhs = rhs_i.View();
-  ret.Checkerboard() = rhs_i.Checkerboard();
-  accelerator_for(ss,rhs.size(),obj::Nsimd(),{
-    coalescedWrite(ret[ss],div(rhs(ss),y));
-  });
-  return ret_i;
-}
-
-template<class obj> Lattice<obj> expMat(const Lattice<obj> &rhs_i, RealD alpha, Integer Nexp = DEFAULT_MAT_EXP){
-  Lattice<obj> ret_i(rhs_i.Grid());
-  auto rhs = rhs_i.View();
-  auto ret = ret_i.View();
-  ret.Checkerboard() = rhs.Checkerboard();
-  accelerator_for(ss,rhs.size(),obj::Nsimd(),{
-    coalescedWrite(ret[ss],Exponentiate(rhs(ss),alpha, Nexp));
-  });
-  return ret_i;
-}
-
-NAMESPACE_END(Grid);
-#endif

Grid/parallelIO/BinaryIO.cc

@@ -1,3 +0,0 @@
-#include <Grid/GridCore.h>
-
-int Grid::BinaryIO::latticeWriteMaxRetry = -1;

Grid/parallelIO/MetaData.h

@@ -1,330 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/parallelIO/NerscIO.h
-
-    Copyright (C) 2015
-
-
-    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 */
-
-#include <algorithm>
-#include <iostream>
-#include <iomanip>
-#include <fstream>
-#include <map>
-#include <unistd.h>
-#include <sys/utsname.h>
-#include <pwd.h>
-
-NAMESPACE_BEGIN(Grid);
-
-///////////////////////////////////////////////////////
-// Precision mapping
-///////////////////////////////////////////////////////
-template<class vobj> static std::string getFormatString (void)
-{
-  std::string format;
-  typedef typename getPrecision<vobj>::real_scalar_type stype;
-  if ( sizeof(stype) == sizeof(float) ) {
-    format = std::string("IEEE32BIG");
-  }
-  if ( sizeof(stype) == sizeof(double) ) {
-    format = std::string("IEEE64BIG");
-  }
-  return format;
-};
-
-  ////////////////////////////////////////////////////////////////////////////////
-  // header specification/interpretation
-  ////////////////////////////////////////////////////////////////////////////////
-    class FieldNormMetaData : Serializable {
-    public:
-      GRID_SERIALIZABLE_CLASS_MEMBERS(FieldNormMetaData, double, norm2);
-    };
-    class FieldMetaData : Serializable {
-    public:
-
-      GRID_SERIALIZABLE_CLASS_MEMBERS(FieldMetaData,
-				      int, nd,
-				      std::vector<int>, dimension,
-				      std::vector<std::string>, boundary,
-				      int, data_start,
-				      std::string, hdr_version,
-				      std::string, storage_format,
-				      double, link_trace,
-				      double, plaquette,
-				      uint32_t, checksum,
-				      uint32_t, scidac_checksuma,
-				      uint32_t, scidac_checksumb,
-				      unsigned int, sequence_number,
-				      std::string, data_type,
-				      std::string, ensemble_id,
-				      std::string, ensemble_label,
-				      std::string, ildg_lfn,
-				      std::string, creator,
-				      std::string, creator_hardware,
-				      std::string, creation_date,
-				      std::string, archive_date,
-				      std::string, floating_point);
-      // WARNING: non-initialised values might lead to twisted parallel IO
-      // issues, std::string are fine because they initliase to size 0
-      // as per C++ standard.
-      FieldMetaData(void) 
-      : nd(4), dimension(4,0), boundary(4, ""), data_start(0),
-      link_trace(0.), plaquette(0.), checksum(0),
-      scidac_checksuma(0), scidac_checksumb(0), sequence_number(0)
-      {}
-  };
-
-// PB disable using namespace - this is a header and forces namesapce visibility for all 
-// including files
-//using namespace Grid;
-
-//////////////////////////////////////////////////////////////////////
-// Bit and Physical Checksumming and QA of data
-//////////////////////////////////////////////////////////////////////
-inline void GridMetaData(GridBase *grid,FieldMetaData &header)
-{
-  int nd = grid->_ndimension;
-  header.nd = nd;
-  header.dimension.resize(nd);
-  header.boundary.resize(nd);
-  header.data_start = 0;
-  for(int d=0;d<nd;d++) {
-    header.dimension[d] = grid->_fdimensions[d];
-  }
-  for(int d=0;d<nd;d++) {
-    header.boundary[d] = std::string("PERIODIC");
-  }
-}
-
-inline void MachineCharacteristics(FieldMetaData &header)
-{
-  // Who
-  struct passwd *pw = getpwuid (getuid());
-  if (pw) header.creator = std::string(pw->pw_name); 
-
-  // When
-  std::time_t t = std::time(nullptr);
-  std::tm tm_ = *std::localtime(&t);
-  std::ostringstream oss; 
-  //      oss << std::put_time(&tm_, "%c %Z");
-  header.creation_date = oss.str();
-  header.archive_date  = header.creation_date;
-
-  // What
-  struct utsname name;  uname(&name);
-  header.creator_hardware = std::string(name.nodename)+"-";
-  header.creator_hardware+= std::string(name.machine)+"-";
-  header.creator_hardware+= std::string(name.sysname)+"-";
-  header.creator_hardware+= std::string(name.release);
-}
-
-#define dump_meta_data(field, s)					\
-  s << "BEGIN_HEADER"      << std::endl;				\
-  s << "HDR_VERSION = "    << field.hdr_version    << std::endl;	\
-  s << "DATATYPE = "       << field.data_type      << std::endl;	\
-  s << "STORAGE_FORMAT = " << field.storage_format << std::endl;	\
-  for(int i=0;i<4;i++){							\
-    s << "DIMENSION_" << i+1 << " = " << field.dimension[i] << std::endl ; \
-  }									\
-  s << "LINK_TRACE = " << std::setprecision(10) << field.link_trace << std::endl; \
-  s << "PLAQUETTE  = " << std::setprecision(10) << field.plaquette  << std::endl; \
-  for(int i=0;i<4;i++){							\
-    s << "BOUNDARY_"<<i+1<<" = " << field.boundary[i] << std::endl;	\
-  }									\
-									\
-  s << "CHECKSUM = "<< std::hex << std::setw(10) << field.checksum << std::dec<<std::endl; \
-  s << "SCIDAC_CHECKSUMA = "<< std::hex << std::setw(10) << field.scidac_checksuma << std::dec<<std::endl; \
-  s << "SCIDAC_CHECKSUMB = "<< std::hex << std::setw(10) << field.scidac_checksumb << std::dec<<std::endl; \
-  s << "ENSEMBLE_ID = "     << field.ensemble_id      << std::endl;	\
-  s << "ENSEMBLE_LABEL = "  << field.ensemble_label   << std::endl;	\
-  s << "SEQUENCE_NUMBER = " << field.sequence_number  << std::endl;	\
-  s << "CREATOR = "         << field.creator          << std::endl;	\
-  s << "CREATOR_HARDWARE = "<< field.creator_hardware << std::endl;	\
-  s << "CREATION_DATE = "   << field.creation_date    << std::endl;	\
-  s << "ARCHIVE_DATE = "    << field.archive_date     << std::endl;	\
-  s << "FLOATING_POINT = "  << field.floating_point   << std::endl;	\
-  s << "END_HEADER"         << std::endl;
-
-template<class vobj> inline void PrepareMetaData(Lattice<vobj> & field, FieldMetaData &header)
-{
-  GridBase *grid = field.Grid();
-  std::string format = getFormatString<vobj>();
-  header.floating_point = format;
-  header.checksum = 0x0; // Nersc checksum unused in ILDG, Scidac
-  GridMetaData(grid,header); 
-  MachineCharacteristics(header);
-}
-inline void GaugeStatistics(Lattice<vLorentzColourMatrixF> & data,FieldMetaData &header)
-{
-  // How to convert data precision etc...
-  header.link_trace=WilsonLoops<PeriodicGimplF>::linkTrace(data);
-  header.plaquette =WilsonLoops<PeriodicGimplF>::avgPlaquette(data);
-}
-inline void GaugeStatistics(Lattice<vLorentzColourMatrixD> & data,FieldMetaData &header)
-{
-  // How to convert data precision etc...
-  header.link_trace=WilsonLoops<PeriodicGimplD>::linkTrace(data);
-  header.plaquette =WilsonLoops<PeriodicGimplD>::avgPlaquette(data);
-}
-template<> inline void PrepareMetaData<vLorentzColourMatrixF>(Lattice<vLorentzColourMatrixF> & field, FieldMetaData &header)
-{
-   
-  GridBase *grid = field.Grid();
-  std::string format = getFormatString<vLorentzColourMatrixF>();
-  header.floating_point = format;
-  header.checksum = 0x0; // Nersc checksum unused in ILDG, Scidac
-  GridMetaData(grid,header); 
-  GaugeStatistics(field,header);
-  MachineCharacteristics(header);
-}
-template<> inline void PrepareMetaData<vLorentzColourMatrixD>(Lattice<vLorentzColourMatrixD> & field, FieldMetaData &header)
-{
-  GridBase *grid = field.Grid();
-  std::string format = getFormatString<vLorentzColourMatrixD>();
-  header.floating_point = format;
-  header.checksum = 0x0; // Nersc checksum unused in ILDG, Scidac
-  GridMetaData(grid,header); 
-  GaugeStatistics(field,header);
-  MachineCharacteristics(header);
-}
-
-//////////////////////////////////////////////////////////////////////
-// Utilities ; these are QCD aware
-//////////////////////////////////////////////////////////////////////
-inline void reconstruct3(LorentzColourMatrix & cm)
-{
-  const int x=0;
-  const int y=1;
-  const int z=2;
-  for(int mu=0;mu<Nd;mu++){
-    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>, 2>, Nd >;
-
-typedef iLorentzColour2x3<Complex>  LorentzColour2x3;
-typedef iLorentzColour2x3<ComplexF> LorentzColour2x3F;
-typedef iLorentzColour2x3<ComplexD> LorentzColour2x3D;
-
-/////////////////////////////////////////////////////////////////////////////////
-// Simple classes for precision conversion
-/////////////////////////////////////////////////////////////////////////////////
-template <class fobj, class sobj>
-struct BinarySimpleUnmunger {
-  typedef typename getPrecision<fobj>::real_scalar_type fobj_stype;
-  typedef typename getPrecision<sobj>::real_scalar_type sobj_stype;
-  
-  void operator()(sobj &in, fobj &out) {
-    // take word by word and transform accoding to the status
-    fobj_stype *out_buffer = (fobj_stype *)&out;
-    sobj_stype *in_buffer = (sobj_stype *)&in;
-    size_t fobj_words = sizeof(out) / sizeof(fobj_stype);
-    size_t sobj_words = sizeof(in) / sizeof(sobj_stype);
-    assert(fobj_words == sobj_words);
-    
-    for (unsigned int word = 0; word < sobj_words; word++)
-      out_buffer[word] = in_buffer[word];  // type conversion on the fly
-    
-  }
-};
-
-template <class fobj, class sobj>
-struct BinarySimpleMunger {
-  typedef typename getPrecision<fobj>::real_scalar_type fobj_stype;
-  typedef typename getPrecision<sobj>::real_scalar_type sobj_stype;
-
-  void operator()(fobj &in, sobj &out) {
-    // take word by word and transform accoding to the status
-    fobj_stype *in_buffer = (fobj_stype *)&in;
-    sobj_stype *out_buffer = (sobj_stype *)&out;
-    size_t fobj_words = sizeof(in) / sizeof(fobj_stype);
-    size_t sobj_words = sizeof(out) / sizeof(sobj_stype);
-    assert(fobj_words == sobj_words);
-    
-    for (unsigned int word = 0; word < sobj_words; word++)
-      out_buffer[word] = in_buffer[word];  // type conversion on the fly
-    
-  }
-};
-
-
-template<class fobj,class sobj>
-struct GaugeSimpleMunger{
-  void operator()(fobj &in, sobj &out) {
-    for (int mu = 0; mu < Nd; mu++) {
-      for (int i = 0; i < Nc; i++) {
-	for (int j = 0; j < Nc; j++) {
-	  out(mu)()(i, j) = in(mu)()(i, j);
-	}}
-    }
-  };
-};
-
-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++) {
-	for (int j = 0; j < Nc; j++) {
-	  out(mu)()(i, j) = in(mu)()(i, j);
-	}}
-    }
-  };
-};
-
-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<2;i++){
-	for(int j=0;j<3;j++){
-	  out(mu)()(i,j) = in(mu)(i)(j);
-	}}
-    }
-    reconstruct3(out);
-  }
-};
-
-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<2;i++){
-	for(int j=0;j<3;j++){
-	  out(mu)(i)(j) = in(mu)()(i,j);
-	}}
-    }
-  }
-};
-
-NAMESPACE_END(Grid);
-

Grid/parallelIO/NerscIO.h

@@ -1,359 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/parallelIO/NerscIO.h
-
-    Copyright (C) 2015
-
-    Author: Matt Spraggs <matthew.spraggs@gmail.com>
-    Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-    Author: paboyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is 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_NERSC_IO_H
-#define GRID_NERSC_IO_H
-
-NAMESPACE_BEGIN(Grid);
-
-using namespace Grid;
-
-////////////////////////////////////////////////////////////////////////////////
-// Write and read from fstream; comput header offset for payload
-////////////////////////////////////////////////////////////////////////////////
-class NerscIO : public BinaryIO { 
-public:
-
-  static inline void truncate(std::string file){
-    std::ofstream fout(file,std::ios::out);
-  }
-  
-  static inline unsigned int writeHeader(FieldMetaData &field,std::string file)
-  {
-    std::ofstream fout(file,std::ios::out|std::ios::in);
-    fout.seekp(0,std::ios::beg);
-    dump_meta_data(field, fout);
-    field.data_start = fout.tellp();
-    return field.data_start;
-  }
-
-  // for the header-reader
-  static inline int readHeader(std::string file,GridBase *grid,  FieldMetaData &field)
-  {
-    std::map<std::string,std::string> header;
-    std::string line;
-
-    //////////////////////////////////////////////////
-    // read the header
-    //////////////////////////////////////////////////
-    std::ifstream fin(file);
-
-    getline(fin,line); // read one line and insist is 
-
-    removeWhitespace(line);
-    std::cout << GridLogMessage << "* " << line << std::endl;
-
-    assert(line==std::string("BEGIN_HEADER"));
-
-    do {
-      getline(fin,line); // read one line
-      std::cout << GridLogMessage << "* "<<line<< std::endl;
-      int eq = line.find("=");
-      if(eq >0) {
-	std::string key=line.substr(0,eq);
-	std::string val=line.substr(eq+1);
-	removeWhitespace(key);
-	removeWhitespace(val);
-      
-	header[key] = val;
-      }
-    } while( line.find("END_HEADER") == std::string::npos );
-
-    field.data_start = fin.tellg();
-
-    //////////////////////////////////////////////////
-    // chomp the values
-    //////////////////////////////////////////////////
-    field.hdr_version    = header["HDR_VERSION"];
-    field.data_type      = header["DATATYPE"];
-    field.storage_format = header["STORAGE_FORMAT"];
-  
-    field.dimension[0] = std::stol(header["DIMENSION_1"]);
-    field.dimension[1] = std::stol(header["DIMENSION_2"]);
-    field.dimension[2] = std::stol(header["DIMENSION_3"]);
-    field.dimension[3] = std::stol(header["DIMENSION_4"]);
-
-    assert(grid->_ndimension == 4);
-    for(int d=0;d<4;d++){
-      assert(grid->_fdimensions[d]==field.dimension[d]);
-    }
-
-    field.link_trace = std::stod(header["LINK_TRACE"]);
-    field.plaquette  = std::stod(header["PLAQUETTE"]);
-
-    field.boundary[0] = header["BOUNDARY_1"];
-    field.boundary[1] = header["BOUNDARY_2"];
-    field.boundary[2] = header["BOUNDARY_3"];
-    field.boundary[3] = header["BOUNDARY_4"];
-
-    field.checksum = std::stoul(header["CHECKSUM"],0,16);
-    field.ensemble_id      = header["ENSEMBLE_ID"];
-    field.ensemble_label   = header["ENSEMBLE_LABEL"];
-    field.sequence_number  = std::stol(header["SEQUENCE_NUMBER"]);
-    field.creator          = header["CREATOR"];
-    field.creator_hardware = header["CREATOR_HARDWARE"];
-    field.creation_date    = header["CREATION_DATE"];
-    field.archive_date     = header["ARCHIVE_DATE"];
-    field.floating_point   = header["FLOATING_POINT"];
-
-    return field.data_start;
-  }
-
-  /////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-  // Now the meat: the object readers
-  /////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-
-  template<class vsimd>
-  static inline void readConfiguration(Lattice<iLorentzColourMatrix<vsimd> > &Umu,
-				       FieldMetaData& header,
-				       std::string file)
-  {
-    typedef Lattice<iLorentzColourMatrix<vsimd> > GaugeField;
-
-    GridBase *grid = Umu.Grid();
-    uint64_t offset = readHeader(file,Umu.Grid(),header);
-
-    FieldMetaData clone(header);
-
-    std::string format(header.floating_point);
-
-    int ieee32big = (format == std::string("IEEE32BIG"));
-    int ieee32    = (format == std::string("IEEE32"));
-    int ieee64big = (format == std::string("IEEE64BIG"));
-    int ieee64    = (format == std::string("IEEE64"));
-
-    uint32_t nersc_csum,scidac_csuma,scidac_csumb;
-    // depending on datatype, set up munger;
-    // munger is a function of <floating point, Real, data_type>
-    if ( header.data_type == std::string("4D_SU3_GAUGE") ) {
-      if ( ieee32 || ieee32big ) {
-	BinaryIO::readLatticeObject<iLorentzColourMatrix<vsimd>, LorentzColour2x3F> 
-	  (Umu,file,Gauge3x2munger<LorentzColour2x3F,LorentzColourMatrix>(), offset,format,
-	   nersc_csum,scidac_csuma,scidac_csumb);
-      }
-      if ( ieee64 || ieee64big ) {
-	BinaryIO::readLatticeObject<iLorentzColourMatrix<vsimd>, LorentzColour2x3D> 
-	  (Umu,file,Gauge3x2munger<LorentzColour2x3D,LorentzColourMatrix>(),offset,format,
-	   nersc_csum,scidac_csuma,scidac_csumb);
-      }
-    } else if ( header.data_type == std::string("4D_SU3_GAUGE_3x3") ) {
-      if ( ieee32 || ieee32big ) {
-	BinaryIO::readLatticeObject<iLorentzColourMatrix<vsimd>,LorentzColourMatrixF>
-	  (Umu,file,GaugeSimpleMunger<LorentzColourMatrixF,LorentzColourMatrix>(),offset,format,
-	   nersc_csum,scidac_csuma,scidac_csumb);
-      }
-      if ( ieee64 || ieee64big ) {
-	BinaryIO::readLatticeObject<iLorentzColourMatrix<vsimd>,LorentzColourMatrixD>
-	  (Umu,file,GaugeSimpleMunger<LorentzColourMatrixD,LorentzColourMatrix>(),offset,format,
-	   nersc_csum,scidac_csuma,scidac_csumb);
-      }
-    } else {
-      assert(0);
-    }
-
-    GaugeStatistics(Umu,clone);
-
-    std::cout<<GridLogMessage <<"NERSC Configuration "<<file<<" checksum "<<std::hex<<nersc_csum<< std::dec
-	     <<" header   "<<std::hex<<header.checksum<<std::dec <<std::endl;
-    std::cout<<GridLogMessage <<"NERSC Configuration "<<file<<" plaquette "<<clone.plaquette
-	     <<" header    "<<header.plaquette<<std::endl;
-    std::cout<<GridLogMessage <<"NERSC Configuration "<<file<<" link_trace "<<clone.link_trace
-	     <<" header    "<<header.link_trace<<std::endl;
-
-    if ( fabs(clone.plaquette -header.plaquette ) >=  1.0e-5 ) { 
-      std::cout << " Plaquette mismatch "<<std::endl;
-    }
-    if ( nersc_csum != header.checksum ) { 
-      std::cerr << " checksum mismatch " << std::endl;
-      std::cerr << " plaqs " << clone.plaquette << " " << header.plaquette << std::endl;
-      std::cerr << " trace " << clone.link_trace<< " " << header.link_trace<< std::endl;
-      std::cerr << " nersc_csum  " <<std::hex<< nersc_csum << " " << header.checksum<< std::dec<< std::endl;
-      exit(0);
-    }
-    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 );
-      
-    std::cout<<GridLogMessage <<"NERSC Configuration "<<file<< " and plaquette, link trace, and checksum agree"<<std::endl;
-  }
-
-  template<class vsimd>
-  static inline void writeConfiguration(Lattice<iLorentzColourMatrix<vsimd> > &Umu,
-					std::string file, 
-					int two_row,
-					int bits32)
-  {
-    typedef Lattice<iLorentzColourMatrix<vsimd> > GaugeField;
-
-    typedef iLorentzColourMatrix<vsimd> vobj;
-    typedef typename vobj::scalar_object sobj;
-
-    FieldMetaData header;
-    ///////////////////////////////////////////
-    // Following should become arguments
-    ///////////////////////////////////////////
-    header.sequence_number = 1;
-    header.ensemble_id     = "UKQCD";
-    header.ensemble_label  = "DWF";
-
-    typedef LorentzColourMatrixD fobj3D;
-    typedef LorentzColour2x3D    fobj2D;
-  
-    GridBase *grid = Umu.Grid();
-
-    GridMetaData(grid,header);
-    assert(header.nd==4);
-    GaugeStatistics(Umu,header);
-    MachineCharacteristics(header);
-
-	uint64_t offset;
-
-    // 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);
-	}
-	grid->Broadcast(0,(void *)&offset,sizeof(offset));
-
-    uint32_t 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);
-	}
-
-    std::cout<<GridLogMessage <<"Written NERSC Configuration on "<< file << " checksum "
-	     <<std::hex<<header.checksum
-	     <<std::dec<<" plaq "<< header.plaquette <<std::endl;
-
-  }
-  ///////////////////////////////
-  // RNG state
-  ///////////////////////////////
-  static inline void writeRNGState(GridSerialRNG &serial,GridParallelRNG &parallel,std::string file)
-  {
-    typedef typename GridParallelRNG::RngStateType RngStateType;
-
-    // Following should become arguments
-    FieldMetaData header;
-    header.sequence_number = 1;
-    header.ensemble_id     = "UKQCD";
-    header.ensemble_label  = "DWF";
-
-    GridBase *grid = parallel.Grid();
-
-    GridMetaData(grid,header);
-    assert(header.nd==4);
-    header.link_trace=0.0;
-    header.plaquette=0.0;
-    MachineCharacteristics(header);
-
-	uint64_t offset;
-  
-#ifdef RNG_RANLUX
-    header.floating_point = std::string("UINT64");
-    header.data_type      = std::string("RANLUX48");
-#endif
-#ifdef RNG_MT19937
-    header.floating_point = std::string("UINT32");
-    header.data_type      = std::string("MT19937");
-#endif
-#ifdef RNG_SITMO
-    header.floating_point = std::string("UINT64");
-    header.data_type      = std::string("SITMO");
-#endif
-
-	if ( grid->IsBoss() ) { 
-    truncate(file);
-    offset = writeHeader(header,file);
-	}
-	grid->Broadcast(0,(void *)&offset,sizeof(offset));
-	
-    uint32_t nersc_csum,scidac_csuma,scidac_csumb;
-    BinaryIO::writeRNG(serial,parallel,file,offset,nersc_csum,scidac_csuma,scidac_csumb);
-    header.checksum = nersc_csum;
-	if ( grid->IsBoss() ) { 
-    offset = writeHeader(header,file);
-	}
-
-    std::cout<<GridLogMessage 
-	     <<"Written NERSC RNG STATE "<<file<< " checksum "
-	     <<std::hex<<header.checksum
-	     <<std::dec<<std::endl;
-
-  }
-    
-  static inline void readRNGState(GridSerialRNG &serial,GridParallelRNG & parallel,FieldMetaData& header,std::string file)
-  {
-    typedef typename GridParallelRNG::RngStateType RngStateType;
-
-    GridBase *grid = parallel.Grid();
-
-	uint64_t offset = readHeader(file,grid,header);
-
-    FieldMetaData clone(header);
-
-    std::string format(header.floating_point);
-    std::string data_type(header.data_type);
-
-#ifdef RNG_RANLUX
-    assert(format == std::string("UINT64"));
-    assert(data_type == std::string("RANLUX48"));
-#endif
-#ifdef RNG_MT19937
-    assert(format == std::string("UINT32"));
-    assert(data_type == std::string("MT19937"));
-#endif
-#ifdef RNG_SITMO
-    assert(format == std::string("UINT64"));
-    assert(data_type == std::string("SITMO"));
-#endif
-
-    // depending on datatype, set up munger;
-    // munger is a function of <floating point, Real, data_type>
-    uint32_t nersc_csum,scidac_csuma,scidac_csumb;
-    BinaryIO::readRNG(serial,parallel,file,offset,nersc_csum,scidac_csuma,scidac_csumb);
-
-    if ( nersc_csum != header.checksum ) { 
-      std::cerr << "checksum mismatch "<<std::hex<< nersc_csum <<" "<<header.checksum<<std::dec<<std::endl;
-      exit(0);
-    }
-    assert(nersc_csum == header.checksum );
-
-    std::cout<<GridLogMessage <<"Read NERSC RNG file "<<file<< " format "<< data_type <<std::endl;
-  }
-};
-
-NAMESPACE_END(QCD);
-
-#endif

Grid/perfmon/Stat.h

@@ -1,105 +0,0 @@
-#ifndef _GRID_STAT_H
-#define _GRID_STAT_H
-
-#ifdef AVX512
-#define _KNIGHTS_LANDING_ROOTONLY
-#endif
-
-NAMESPACE_BEGIN(Grid); 
-
-///////////////////////////////////////////////////////////////////////////////
-// Extra KNL counters from MCDRAM
-///////////////////////////////////////////////////////////////////////////////
-#ifdef _KNIGHTS_LANDING_
-#define NMC 6
-#define NEDC 8
-struct ctrs
-{
-  uint64_t mcrd[NMC];
-  uint64_t mcwr[NMC];
-  uint64_t edcrd[NEDC]; 
-  uint64_t edcwr[NEDC];
-  uint64_t edchite[NEDC];
-  uint64_t edchitm[NEDC];
-  uint64_t edcmisse[NEDC];
-  uint64_t edcmissm[NEDC];
-};
-// Peter/Azusa:
-// Our modification of a code provided by Larry Meadows from Intel
-// Verified by email exchange non-NDA, ok for github. Should be as uses /sys/devices/ FS
-// so is already public and in the linux kernel for KNL.
-struct knl_gbl_
-{
-  int mc_rd[NMC];
-  int mc_wr[NMC];
-  int edc_rd[NEDC];
-  int edc_wr[NEDC];
-  int edc_hite[NEDC];
-  int edc_hitm[NEDC];
-  int edc_misse[NEDC];
-  int edc_missm[NEDC];
-};
-#endif
-///////////////////////////////////////////////////////////////////////////////
-
-class PmuStat
-{
-  uint64_t counters[8][256];
-#ifdef _KNIGHTS_LANDING_
-  static struct knl_gbl_ gbl;
-#endif
-  const char *name;
-
-  uint64_t reads;     // memory reads
-  uint64_t writes;    // memory writes
-  uint64_t mrstart;   // memory read counter at start of parallel region
-  uint64_t mrend;     // memory read counter at end of parallel region
-  uint64_t mwstart;   // memory write counter at start of parallel region
-  uint64_t mwend;     // memory write counter at end of parallel region
-
-  // cumulative counters
-  uint64_t count;     // number of invocations
-  uint64_t tregion;   // total time in parallel region (from thread 0)
-  uint64_t tcycles;   // total cycles inside parallel region
-  uint64_t inst, ref, cyc;   // fixed counters
-  uint64_t pmc0, pmc1;// pmu
-  // add memory counters here
-  // temp variables
-  uint64_t tstart;    // tsc at start of parallel region
-  uint64_t tend;      // tsc at end of parallel region
-  // map for ctrs values
-  // 0 pmc0 start
-  // 1 pmc0 end
-  // 2 pmc1 start
-  // 3 pmc1 end
-  // 4 tsc start
-  // 5 tsc end
-  static bool pmu_initialized;
-public:
-  static bool is_init(void){ return pmu_initialized;}
-  static void pmu_init(void);
-  static void pmu_fini(void);
-  static void pmu_start(void);
-  static void pmu_stop(void);
-  void accum(int nthreads);
-  static void xmemctrs(uint64_t *mr, uint64_t *mw);
-  void start(void);
-  void enter(int t);
-  void exit(int t);
-  void print(void);
-  void init(const char *regname);
-  void clear(void);
-#ifdef _KNIGHTS_LANDING_
-  static void     KNLsetup(void);
-  static uint64_t KNLreadctr(int fd);
-  static void     KNLreadctrs(ctrs &c);
-  static void     KNLevsetup(const char *ename, int &fd, int event, int umask);
-#endif
-    
-};
-
-NAMESPACE_END(Grid); 
-
-#endif
-
-

Grid/qcd/QCD.h

@@ -1,529 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/QCD.h
-
-    Copyright (C) 2015
-
-Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: neo <cossu@post.kek.jp>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is 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);
-
-static constexpr int Xdir = 0;
-static constexpr int Ydir = 1;
-static constexpr int Zdir = 2;
-static constexpr int Tdir = 3;
-
-static constexpr int Xp = 0;
-static constexpr int Yp = 1;
-static constexpr int Zp = 2;
-static constexpr int Tp = 3;
-static constexpr int Xm = 4;
-static constexpr int Ym = 5;
-static constexpr int Zm = 6;
-static constexpr int Tm = 7;
-
-static constexpr int Nc=3;
-static constexpr int Ns=4;
-static constexpr int Nd=4;
-static constexpr int Nhs=2; // half spinor
-static constexpr int Nds=8; // double stored gauge field
-static constexpr int Ngp=2; // gparity index range
-
-//////////////////////////////////////////////////////////////////////////////
-// QCD iMatrix types
-// Index conventions:                            Lorentz x Spin x Colour
-// note: static constexpr int or constexpr will work for type deductions
-//       with the intel compiler (up to version 17)
-//////////////////////////////////////////////////////////////////////////////
-#define ColourIndex  (2)
-#define SpinIndex    (1)
-#define LorentzIndex (0)
-
-// Also should make these a named enum type
-static constexpr int DaggerNo=0;
-static constexpr int DaggerYes=1;
-static constexpr int InverseNo=0;
-static constexpr int InverseYes=1;
-
-// Useful traits is this a spin index
-//typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
-
-const int SpinorIndex = 2;
-template<typename T> struct isSpinor {
-  static constexpr bool value = (SpinorIndex==T::TensorLevel);
-};
-template <typename T> using IfSpinor    = Invoke<std::enable_if< isSpinor<T>::value,int> > ;
-template <typename T> using IfNotSpinor = Invoke<std::enable_if<!isSpinor<T>::value,int> > ;
-
-// ChrisK very keen to add extra space for Gparity doubling.
-//
-// Also add domain wall index, in a way where Wilson operator 
-// naturally distributes across the 5th dimensions.
-//
-// That probably makes for GridRedBlack4dCartesian grid.
-
-// s,sp,c,spc,lc
-
-template<typename vtype> using iSinglet                   = iScalar<iScalar<iScalar<vtype> > >;
-template<typename vtype> using iSpinMatrix                = iScalar<iMatrix<iScalar<vtype>, Ns> >;
-template<typename vtype> using iColourMatrix              = iScalar<iScalar<iMatrix<vtype, Nc> > > ;
-template<typename vtype> using iSpinColourMatrix          = iScalar<iMatrix<iMatrix<vtype, Nc>, Ns> >;
-template<typename vtype> using iLorentzColourMatrix       = iVector<iScalar<iMatrix<vtype, Nc> >, Nd > ;
-template<typename vtype> using iDoubleStoredColourMatrix  = iVector<iScalar<iMatrix<vtype, Nc> >, Nds > ;
-template<typename vtype> using iSpinVector                = iScalar<iVector<iScalar<vtype>, Ns> >;
-template<typename vtype> using iColourVector              = iScalar<iScalar<iVector<vtype, Nc> > >;
-template<typename vtype> using iSpinColourVector          = iScalar<iVector<iVector<vtype, Nc>, Ns> >;
-template<typename vtype> using iHalfSpinVector            = iScalar<iVector<iScalar<vtype>, Nhs> >;
-template<typename vtype> using iHalfSpinColourVector      = iScalar<iVector<iVector<vtype, Nc>, Nhs> >;
-    template<typename vtype> using iSpinColourSpinColourMatrix  = iScalar<iMatrix<iMatrix<iMatrix<iMatrix<vtype, Nc>, Ns>, Nc>, Ns> >;
-
-
-template<typename vtype> using iGparitySpinColourVector       = iVector<iVector<iVector<vtype, Nc>, Ns>, Ngp >;
-template<typename vtype> using iGparityHalfSpinColourVector   = iVector<iVector<iVector<vtype, Nc>, Nhs>, Ngp >;
-
-// Spin matrix
-typedef iSpinMatrix<Complex  >          SpinMatrix;
-typedef iSpinMatrix<ComplexF >          SpinMatrixF;
-typedef iSpinMatrix<ComplexD >          SpinMatrixD;
-
-typedef iSpinMatrix<vComplex >          vSpinMatrix;
-typedef iSpinMatrix<vComplexF>          vSpinMatrixF;
-typedef iSpinMatrix<vComplexD>          vSpinMatrixD;
-
-// Colour Matrix
-typedef iColourMatrix<Complex  >        ColourMatrix;
-typedef iColourMatrix<ComplexF >        ColourMatrixF;
-typedef iColourMatrix<ComplexD >        ColourMatrixD;
-
-typedef iColourMatrix<vComplex >        vColourMatrix;
-typedef iColourMatrix<vComplexF>        vColourMatrixF;
-typedef iColourMatrix<vComplexD>        vColourMatrixD;
-
-// SpinColour matrix
-typedef iSpinColourMatrix<Complex  >    SpinColourMatrix;
-typedef iSpinColourMatrix<ComplexF >    SpinColourMatrixF;
-typedef iSpinColourMatrix<ComplexD >    SpinColourMatrixD;
-
-typedef iSpinColourMatrix<vComplex >    vSpinColourMatrix;
-typedef iSpinColourMatrix<vComplexF>    vSpinColourMatrixF;
-typedef iSpinColourMatrix<vComplexD>    vSpinColourMatrixD;
-
-    // SpinColourSpinColour matrix
-    typedef iSpinColourSpinColourMatrix<Complex  >    SpinColourSpinColourMatrix;
-    typedef iSpinColourSpinColourMatrix<ComplexF >    SpinColourSpinColourMatrixF;
-    typedef iSpinColourSpinColourMatrix<ComplexD >    SpinColourSpinColourMatrixD;
-
-    typedef iSpinColourSpinColourMatrix<vComplex >    vSpinColourSpinColourMatrix;
-    typedef iSpinColourSpinColourMatrix<vComplexF>    vSpinColourSpinColourMatrixF;
-    typedef iSpinColourSpinColourMatrix<vComplexD>    vSpinColourSpinColourMatrixD;
-
-    // SpinColourSpinColour matrix
-    typedef iSpinColourSpinColourMatrix<Complex  >    SpinColourSpinColourMatrix;
-    typedef iSpinColourSpinColourMatrix<ComplexF >    SpinColourSpinColourMatrixF;
-    typedef iSpinColourSpinColourMatrix<ComplexD >    SpinColourSpinColourMatrixD;
-
-    typedef iSpinColourSpinColourMatrix<vComplex >    vSpinColourSpinColourMatrix;
-    typedef iSpinColourSpinColourMatrix<vComplexF>    vSpinColourSpinColourMatrixF;
-    typedef iSpinColourSpinColourMatrix<vComplexD>    vSpinColourSpinColourMatrixD;
-
-// LorentzColour
-typedef iLorentzColourMatrix<Complex  > LorentzColourMatrix;
-typedef iLorentzColourMatrix<ComplexF > LorentzColourMatrixF;
-typedef iLorentzColourMatrix<ComplexD > LorentzColourMatrixD;
-
-typedef iLorentzColourMatrix<vComplex > vLorentzColourMatrix;
-typedef iLorentzColourMatrix<vComplexF> vLorentzColourMatrixF;
-typedef iLorentzColourMatrix<vComplexD> vLorentzColourMatrixD;
-
-// DoubleStored gauge field
-typedef iDoubleStoredColourMatrix<Complex  > DoubleStoredColourMatrix;
-typedef iDoubleStoredColourMatrix<ComplexF > DoubleStoredColourMatrixF;
-typedef iDoubleStoredColourMatrix<ComplexD > DoubleStoredColourMatrixD;
-
-typedef iDoubleStoredColourMatrix<vComplex > vDoubleStoredColourMatrix;
-typedef iDoubleStoredColourMatrix<vComplexF> vDoubleStoredColourMatrixF;
-typedef iDoubleStoredColourMatrix<vComplexD> vDoubleStoredColourMatrixD;
-
-// Spin vector
-typedef iSpinVector<Complex >           SpinVector;
-typedef iSpinVector<ComplexF>           SpinVectorF;
-typedef iSpinVector<ComplexD>           SpinVectorD;
-
-typedef iSpinVector<vComplex >           vSpinVector;
-typedef iSpinVector<vComplexF>           vSpinVectorF;
-typedef iSpinVector<vComplexD>           vSpinVectorD;
-
-// Colour vector
-typedef iColourVector<Complex >         ColourVector;
-typedef iColourVector<ComplexF>         ColourVectorF;
-typedef iColourVector<ComplexD>         ColourVectorD;
-
-typedef iColourVector<vComplex >         vColourVector;
-typedef iColourVector<vComplexF>         vColourVectorF;
-typedef iColourVector<vComplexD>         vColourVectorD;
-
-// SpinColourVector
-typedef iSpinColourVector<Complex >     SpinColourVector;
-typedef iSpinColourVector<ComplexF>     SpinColourVectorF;
-typedef iSpinColourVector<ComplexD>     SpinColourVectorD;
-
-typedef iSpinColourVector<vComplex >     vSpinColourVector;
-typedef iSpinColourVector<vComplexF>     vSpinColourVectorF;
-typedef iSpinColourVector<vComplexD>     vSpinColourVectorD;
-
-// HalfSpin vector
-typedef iHalfSpinVector<Complex >       HalfSpinVector;
-typedef iHalfSpinVector<ComplexF>       HalfSpinVectorF;
-typedef iHalfSpinVector<ComplexD>       HalfSpinVectorD;
-
-typedef iHalfSpinVector<vComplex >       vHalfSpinVector;
-typedef iHalfSpinVector<vComplexF>       vHalfSpinVectorF;
-typedef iHalfSpinVector<vComplexD>       vHalfSpinVectorD;
-
-// HalfSpinColour vector
-typedef iHalfSpinColourVector<Complex > HalfSpinColourVector;
-typedef iHalfSpinColourVector<ComplexF> HalfSpinColourVectorF;
-typedef iHalfSpinColourVector<ComplexD> HalfSpinColourVectorD;
-    
-typedef iHalfSpinColourVector<vComplex > vHalfSpinColourVector;
-typedef iHalfSpinColourVector<vComplexF> vHalfSpinColourVectorF;
-typedef iHalfSpinColourVector<vComplexD> vHalfSpinColourVectorD;
-    
-// singlets
-typedef iSinglet<Complex >         TComplex;     // FIXME This is painful. Tensor singlet complex type.
-typedef iSinglet<ComplexF>         TComplexF;    // FIXME This is painful. Tensor singlet complex type.
-typedef iSinglet<ComplexD>         TComplexD;    // FIXME This is painful. Tensor singlet complex type.
-
-typedef iSinglet<vComplex >        vTComplex ;   // what if we don't know the tensor structure
-typedef iSinglet<vComplexF>        vTComplexF;   // what if we don't know the tensor structure
-typedef iSinglet<vComplexD>        vTComplexD;   // what if we don't know the tensor structure
-
-typedef iSinglet<Real >            TReal;        // Shouldn't need these; can I make it work without?
-typedef iSinglet<RealF>            TRealF;       // Shouldn't need these; can I make it work without?
-typedef iSinglet<RealD>            TRealD;       // Shouldn't need these; can I make it work without?
-
-typedef iSinglet<vReal >           vTReal;      
-typedef iSinglet<vRealF>           vTRealF;      
-typedef iSinglet<vRealD>           vTRealD;      
-
-typedef iSinglet<vInteger>         vTInteger;
-typedef iSinglet<Integer >         TInteger;
-
-
-// Lattices of these
-typedef Lattice<vColourMatrix>          LatticeColourMatrix;
-typedef Lattice<vColourMatrixF>         LatticeColourMatrixF;
-typedef Lattice<vColourMatrixD>         LatticeColourMatrixD;
-
-typedef Lattice<vSpinMatrix>            LatticeSpinMatrix;
-typedef Lattice<vSpinMatrixF>           LatticeSpinMatrixF;
-typedef Lattice<vSpinMatrixD>           LatticeSpinMatrixD;
-
-typedef Lattice<vSpinColourMatrix>      LatticeSpinColourMatrix;
-typedef Lattice<vSpinColourMatrixF>     LatticeSpinColourMatrixF;
-typedef Lattice<vSpinColourMatrixD>     LatticeSpinColourMatrixD;
-
-typedef Lattice<vSpinColourSpinColourMatrix>      LatticeSpinColourSpinColourMatrix;
-typedef Lattice<vSpinColourSpinColourMatrixF>     LatticeSpinColourSpinColourMatrixF;
-typedef Lattice<vSpinColourSpinColourMatrixD>     LatticeSpinColourSpinColourMatrixD;
-
-typedef Lattice<vLorentzColourMatrix>  LatticeLorentzColourMatrix;
-typedef Lattice<vLorentzColourMatrixF> LatticeLorentzColourMatrixF;
-typedef Lattice<vLorentzColourMatrixD> LatticeLorentzColourMatrixD;
-
-// DoubleStored gauge field
-typedef Lattice<vDoubleStoredColourMatrix>  LatticeDoubleStoredColourMatrix;
-typedef Lattice<vDoubleStoredColourMatrixF> LatticeDoubleStoredColourMatrixF;
-typedef Lattice<vDoubleStoredColourMatrixD> LatticeDoubleStoredColourMatrixD;
-
-typedef Lattice<vSpinVector>            LatticeSpinVector;
-typedef Lattice<vSpinVectorF>           LatticeSpinVectorF;
-typedef Lattice<vSpinVectorD>           LatticeSpinVectorD;
-
-typedef Lattice<vColourVector>          LatticeColourVector;
-typedef Lattice<vColourVectorF>         LatticeColourVectorF;
-typedef Lattice<vColourVectorD>         LatticeColourVectorD;
-
-typedef Lattice<vSpinColourVector>      LatticeSpinColourVector;
-typedef Lattice<vSpinColourVectorF>     LatticeSpinColourVectorF;
-typedef Lattice<vSpinColourVectorD>     LatticeSpinColourVectorD;
-
-typedef Lattice<vHalfSpinVector>        LatticeHalfSpinVector;
-typedef Lattice<vHalfSpinVectorF>       LatticeHalfSpinVectorF;
-typedef Lattice<vHalfSpinVectorD>       LatticeHalfSpinVectorD;
-
-typedef Lattice<vHalfSpinColourVector>  LatticeHalfSpinColourVector;
-typedef Lattice<vHalfSpinColourVectorF> LatticeHalfSpinColourVectorF;
-typedef Lattice<vHalfSpinColourVectorD> LatticeHalfSpinColourVectorD;
-
-typedef Lattice<vTReal>            LatticeReal;
-typedef Lattice<vTRealF>           LatticeRealF;
-typedef Lattice<vTRealD>           LatticeRealD;
-
-typedef Lattice<vTComplex>         LatticeComplex;
-typedef Lattice<vTComplexF>        LatticeComplexF;
-typedef Lattice<vTComplexD>        LatticeComplexD;
-
-typedef Lattice<vTInteger>         LatticeInteger; // Predicates for "where"
-
-
-///////////////////////////////////////////
-// Physical names for things
-///////////////////////////////////////////
-typedef LatticeHalfSpinColourVector  LatticeHalfFermion;
-typedef LatticeHalfSpinColourVectorF LatticeHalfFermionF;
-typedef LatticeHalfSpinColourVectorF LatticeHalfFermionD;
-
-typedef LatticeSpinColourVector      LatticeFermion;
-typedef LatticeSpinColourVectorF     LatticeFermionF;
-typedef LatticeSpinColourVectorD     LatticeFermionD;
-
-typedef LatticeSpinColourMatrix                LatticePropagator;
-typedef LatticeSpinColourMatrixF               LatticePropagatorF;
-typedef LatticeSpinColourMatrixD               LatticePropagatorD;
-
-typedef LatticeLorentzColourMatrix             LatticeGaugeField;
-typedef LatticeLorentzColourMatrixF            LatticeGaugeFieldF;
-typedef LatticeLorentzColourMatrixD            LatticeGaugeFieldD;
-
-typedef LatticeDoubleStoredColourMatrix        LatticeDoubledGaugeField;
-typedef LatticeDoubleStoredColourMatrixF       LatticeDoubledGaugeFieldF;
-typedef LatticeDoubleStoredColourMatrixD       LatticeDoubledGaugeFieldD;
-
-template<class GF> using LorentzScalar = Lattice<iScalar<typename GF::vector_object::element> >;
-
-// Uhgg... typing this hurt  ;)
-// (my keyboard got burning hot when I typed this, must be the anti-Fermion)
-typedef Lattice<vColourVector>          LatticeStaggeredFermion;    
-typedef Lattice<vColourVectorF>         LatticeStaggeredFermionF;    
-typedef Lattice<vColourVectorD>         LatticeStaggeredFermionD;    
-
-typedef Lattice<vColourMatrix>          LatticeStaggeredPropagator; 
-typedef Lattice<vColourMatrixF>         LatticeStaggeredPropagatorF; 
-typedef Lattice<vColourMatrixD>         LatticeStaggeredPropagatorD; 
-
-//////////////////////////////////////////////////////////////////////////////
-// Peek and Poke named after physics attributes
-//////////////////////////////////////////////////////////////////////////////
-
-//spin
-template<class vobj> auto peekSpin(const vobj &rhs,int i) -> decltype(PeekIndex<SpinIndex>(rhs,0))
-{
-  return PeekIndex<SpinIndex>(rhs,i);
-}
-template<class vobj> auto peekSpin(const vobj &rhs,int i,int j) -> decltype(PeekIndex<SpinIndex>(rhs,0,0))
-{
-  return PeekIndex<SpinIndex>(rhs,i,j);
-}
-template<class vobj> auto peekSpin(const Lattice<vobj> &rhs,int i) -> decltype(PeekIndex<SpinIndex>(rhs,0))
-{
-  return PeekIndex<SpinIndex>(rhs,i);
-}
-template<class vobj> auto peekSpin(const Lattice<vobj> &rhs,int i,int j) -> decltype(PeekIndex<SpinIndex>(rhs,0,0))
-{
-  return PeekIndex<SpinIndex>(rhs,i,j);
-}
-//colour
-template<class vobj> auto peekColour(const vobj &rhs,int i) -> decltype(PeekIndex<ColourIndex>(rhs,0))
-{
-  return PeekIndex<ColourIndex>(rhs,i);
-}
-template<class vobj> auto peekColour(const vobj &rhs,int i,int j) -> decltype(PeekIndex<ColourIndex>(rhs,0,0))
-{
-  return PeekIndex<ColourIndex>(rhs,i,j);
-}
-template<class vobj> auto peekColour(const Lattice<vobj> &rhs,int i) -> decltype(PeekIndex<ColourIndex>(rhs,0))
-{
-  return PeekIndex<ColourIndex>(rhs,i);
-}
-template<class vobj> auto peekColour(const Lattice<vobj> &rhs,int i,int j) -> decltype(PeekIndex<ColourIndex>(rhs,0,0))
-{
-  return PeekIndex<ColourIndex>(rhs,i,j);
-}
-//lorentz
-template<class vobj> auto peekLorentz(const vobj &rhs,int i) -> decltype(PeekIndex<LorentzIndex>(rhs,0))
-{
-  return PeekIndex<LorentzIndex>(rhs,i);
-}
-template<class vobj> auto peekLorentz(const Lattice<vobj> &rhs,int i) -> decltype(PeekIndex<LorentzIndex>(rhs,0))
-{
-  return PeekIndex<LorentzIndex>(rhs,i);
-}
-
-//////////////////////////////////////////////
-// Poke lattice
-//////////////////////////////////////////////
-template<class vobj> 
-void pokeColour(Lattice<vobj> &lhs,
-		const Lattice<decltype(peekIndex<ColourIndex>(vobj(),0))> & rhs,
-		int i)
-{
-  PokeIndex<ColourIndex>(lhs,rhs,i);
-}
-template<class vobj> 
-void pokeColour(Lattice<vobj> &lhs,
-		const Lattice<decltype(peekIndex<ColourIndex>(vobj(),0,0))> & rhs,
-		int i,int j)
-{
-  PokeIndex<ColourIndex>(lhs,rhs,i,j);
-}
-template<class vobj> 
-void pokeSpin(Lattice<vobj> &lhs,
-              const Lattice<decltype(peekIndex<SpinIndex>(vobj(),0))> & rhs,
-              int i)
-{
-  PokeIndex<SpinIndex>(lhs,rhs,i);
-}
-template<class vobj> 
-void pokeSpin(Lattice<vobj> &lhs,
-              const Lattice<decltype(peekIndex<SpinIndex>(vobj(),0,0))> & rhs,
-              int i,int j)
-{
-  PokeIndex<SpinIndex>(lhs,rhs,i,j);
-}
-template<class vobj> 
-void pokeLorentz(Lattice<vobj> &lhs,
-		 const Lattice<decltype(peekIndex<LorentzIndex>(vobj(),0))> & rhs,
-		 int i)
-{
-  PokeIndex<LorentzIndex>(lhs,rhs,i);
-}
-
-//////////////////////////////////////////////
-// Poke scalars
-//////////////////////////////////////////////
-template<class vobj> void pokeSpin(vobj &lhs,const decltype(peekIndex<SpinIndex>(lhs,0)) & rhs,int i)
-{
-  pokeIndex<SpinIndex>(lhs,rhs,i);
-}
-template<class vobj> void pokeSpin(vobj &lhs,const decltype(peekIndex<SpinIndex>(lhs,0,0)) & rhs,int i,int j)
-{
-  pokeIndex<SpinIndex>(lhs,rhs,i,j);
-}
-
-template<class vobj> void pokeColour(vobj &lhs,const decltype(peekIndex<ColourIndex>(lhs,0)) & rhs,int i)
-{
-  pokeIndex<ColourIndex>(lhs,rhs,i);
-}
-template<class vobj> void pokeColour(vobj &lhs,const decltype(peekIndex<ColourIndex>(lhs,0,0)) & rhs,int i,int j)
-{
-  pokeIndex<ColourIndex>(lhs,rhs,i,j);
-}
-
-template<class vobj> void pokeLorentz(vobj &lhs,const decltype(peekIndex<LorentzIndex>(lhs,0)) & rhs,int i)
-{
-  pokeIndex<LorentzIndex>(lhs,rhs,i);
-}
-
-//////////////////////////////////////////////
-// Fermion <-> propagator assignements
-//////////////////////////////////////////////
-    //template <class Prop, class Ferm>
-    template <class Fimpl>
-      void FermToProp(typename Fimpl::PropagatorField &p, const typename Fimpl::FermionField &f, const int s, const int c)
-{
-  for(int j = 0; j < Ns; ++j)
-    {
-      auto pjs = peekSpin(p, j, s);
-      auto fj  = peekSpin(f, j);
-            
-            for(int i = 0; i < Fimpl::Dimension; ++i)
-	{
-	  pokeColour(pjs, peekColour(fj, i), i, c);
-	}
-      pokeSpin(p, pjs, j, s);
-    }
-}
-    
-    //template <class Prop, class Ferm>
-    template <class Fimpl>
-      void PropToFerm(typename Fimpl::FermionField &f, const typename Fimpl::PropagatorField &p, const int s, const int c)
-{
-  for(int j = 0; j < Ns; ++j)
-    {
-      auto pjs = peekSpin(p, j, s);
-      auto fj  = peekSpin(f, j);
-            
-            for(int i = 0; i < Fimpl::Dimension; ++i)
-	{
-	  pokeColour(fj, peekColour(pjs, i, c), i);
-	}
-      pokeSpin(f, fj, j);
-    }
-}
-    
-//////////////////////////////////////////////
-// transpose array and scalar
-//////////////////////////////////////////////
-template<int Index,class vobj> inline Lattice<vobj> transposeSpin(const Lattice<vobj> &lhs){
-  return transposeIndex<SpinIndex>(lhs);
-}
-template<int Index,class vobj> inline Lattice<vobj> transposeColour(const Lattice<vobj> &lhs){
-  return transposeIndex<ColourIndex>(lhs);
-}
-template<int Index,class vobj> inline vobj transposeSpin(const vobj &lhs){
-  return transposeIndex<SpinIndex>(lhs);
-}
-template<int Index,class vobj> inline vobj transposeColour(const vobj &lhs){
-  return transposeIndex<ColourIndex>(lhs);
-}
-
-//////////////////////////////////////////
-// Trace lattice and non-lattice
-//////////////////////////////////////////
-template<int Index,class vobj>
-inline auto traceSpin(const Lattice<vobj> &lhs) -> Lattice<decltype(traceIndex<SpinIndex>(vobj()))>
-{
-  return traceIndex<SpinIndex>(lhs);
-}
-template<int Index,class vobj>
-inline auto traceColour(const Lattice<vobj> &lhs) -> Lattice<decltype(traceIndex<ColourIndex>(vobj()))>
-{
-  return traceIndex<ColourIndex>(lhs);
-}
-template<int Index,class vobj>
-inline auto traceSpin(const vobj &lhs) -> Lattice<decltype(traceIndex<SpinIndex>(lhs))>
-{
-  return traceIndex<SpinIndex>(lhs);
-}
-template<int Index,class vobj>
-inline auto traceColour(const vobj &lhs) -> Lattice<decltype(traceIndex<ColourIndex>(lhs))>
-{
-  return traceIndex<ColourIndex>(lhs);
-}
-
-//////////////////////////////////////////
-// Current types
-//////////////////////////////////////////
-GRID_SERIALIZABLE_ENUM(Current, undef,
-		       Vector,  0,
-		       Axial,   1,
-		       Tadpole, 2);
-
-NAMESPACE_END(Grid);
-

Grid/qcd/action/fermion/AbstractEOFAFermion.h

@@ -1,100 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/AbstractEOFAFermion.h
-
-Copyright (C) 2017
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: David Murphy <dmurphy@phys.columbia.edu>
-
-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_QCD_ABSTRACT_EOFA_FERMION_H
-#define  GRID_QCD_ABSTRACT_EOFA_FERMION_H
-
-#include <Grid/qcd/action/fermion/CayleyFermion5D.h>
-
-NAMESPACE_BEGIN(Grid);
-
-// DJM: Abstract base class for EOFA fermion types.
-// Defines layout of additional EOFA-specific parameters and operators.
-// Use to construct EOFA pseudofermion actions that are agnostic to
-// Shamir / Mobius / etc., and ensure that no one can construct EOFA
-// pseudofermion action with non-EOFA fermion type.
-template<class Impl>
-class AbstractEOFAFermion : public CayleyFermion5D<Impl> {
-public:
-  INHERIT_IMPL_TYPES(Impl);
-
-public:
-  // Fermion operator: D(mq1) + shift*\gamma_{5}*R_{5}*\Delta_{\pm}(mq2,mq3)*P_{\pm}
-  RealD mq1;
-  RealD mq2;
-  RealD mq3;
-  RealD shift;
-  int pm;
-
-  RealD alpha; // Mobius scale
-  RealD k;     // EOFA normalization constant
-
-  virtual void Instantiatable(void) = 0;
-
-  // EOFA-specific operations
-  // Force user to implement in derived classes
-  virtual void  Omega    (const FermionField& in, FermionField& out, int sign, int dag) = 0;
-  virtual void  Dtilde   (const FermionField& in, FermionField& out) = 0;
-  virtual void  DtildeInv(const FermionField& in, FermionField& out) = 0;
-
-  // Implement derivatives in base class:
-  // for EOFA both DWF and Mobius just need d(Dw)/dU
-  virtual void MDeriv(GaugeField& mat, const FermionField& U, const FermionField& V, int dag){
-    this->DhopDeriv(mat, U, V, dag);
-  };
-  virtual void MoeDeriv(GaugeField& mat, const FermionField& U, const FermionField& V, int dag){
-    this->DhopDerivOE(mat, U, V, dag);
-  };
-  virtual void MeoDeriv(GaugeField& mat, const FermionField& U, const FermionField& V, int dag){
-    this->DhopDerivEO(mat, U, V, dag);
-  };
-
-  // Recompute 5D coefficients for different value of shift constant
-  // (needed for heatbath loop over poles)
-  virtual void RefreshShiftCoefficients(RealD new_shift) = 0;
-
-  // Constructors
-  AbstractEOFAFermion(GaugeField& _Umu, GridCartesian& FiveDimGrid, GridRedBlackCartesian& FiveDimRedBlackGrid,
-		      GridCartesian& FourDimGrid, GridRedBlackCartesian& FourDimRedBlackGrid,
-		      RealD _mq1, RealD _mq2, RealD _mq3, RealD _shift, int _pm,
-		      RealD _M5, RealD _b, RealD _c, const ImplParams& p=ImplParams())
-    : CayleyFermion5D<Impl>(_Umu, FiveDimGrid, FiveDimRedBlackGrid, FourDimGrid, FourDimRedBlackGrid,
-			    _mq1, _M5, p), mq1(_mq1), mq2(_mq2), mq3(_mq3), shift(_shift), pm(_pm)
-  {
-    int Ls = this->Ls;
-    this->alpha = _b + _c;
-    this->k = this->alpha * (_mq3-_mq2) * std::pow(this->alpha+1.0,2*Ls) /
-      ( std::pow(this->alpha+1.0,Ls) + _mq2*std::pow(this->alpha-1.0,Ls) ) /
-      ( std::pow(this->alpha+1.0,Ls) + _mq3*std::pow(this->alpha-1.0,Ls) );
-  };
-};
-
-NAMESPACE_END(Grid);
-
-#endif

Grid/qcd/action/fermion/CayleyFermion5D.h

@@ -1,169 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: ./lib/qcd/action/fermion/CayleyFermion5D.h
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-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
-
-#include <Grid/qcd/action/fermion/WilsonFermion5D.h>
-
-NAMESPACE_BEGIN(Grid);
-
-template<class Impl>
-class CayleyFermion5D : public WilsonFermion5D<Impl>
-{
-public:
-  INHERIT_IMPL_TYPES(Impl);
-public:
-
-  // override multiply
-  virtual RealD  M    (const FermionField &in, FermionField &out);
-  virtual RealD  Mdag (const FermionField &in, FermionField &out);
-
-  // half checkerboard operations
-  virtual void   Meooe       (const FermionField &in, FermionField &out);
-  virtual void   MeooeDag    (const FermionField &in, FermionField &out);
-  virtual void   Mooee       (const FermionField &in, FermionField &out);
-  virtual void   MooeeDag    (const FermionField &in, FermionField &out);
-  virtual void   MooeeInv    (const FermionField &in, FermionField &out);
-  virtual void   MooeeInvDag (const FermionField &in, FermionField &out);
-  virtual void   Meo5D (const FermionField &psi, FermionField &chi);
-
-  virtual void   M5D   (const FermionField &psi, FermionField &chi);
-  virtual void   M5Ddag(const FermionField &psi, FermionField &chi);
-
-  ///////////////////////////////////////////////////////////////
-  // Physical surface field utilities
-  ///////////////////////////////////////////////////////////////
-  virtual void Dminus(const FermionField &psi, FermionField &chi);
-  virtual void DminusDag(const FermionField &psi, FermionField &chi);
-  virtual void ExportPhysicalFermionSolution(const FermionField &solution5d,FermionField &exported4d);
-  virtual void ExportPhysicalFermionSource(const FermionField &solution5d, FermionField &exported4d);
-  virtual void ImportPhysicalFermionSource(const FermionField &input4d,FermionField &imported5d);
-  virtual void ImportUnphysicalFermion(const FermionField &solution5d, FermionField &exported4d);
-
-  ///////////////////////////////////////////////////////////////
-  // Support for MADWF tricks
-  ///////////////////////////////////////////////////////////////
-  RealD Mass(void) { return mass; };
-  void  SetMass(RealD _mass) { 
-    mass=_mass; 
-    SetCoefficientsInternal(_zolo_hi,_gamma,_b,_c);  // Reset coeffs
-  } ;
-  void  P(const FermionField &psi, FermionField &chi);
-  void  Pdag(const FermionField &psi, FermionField &chi);
-  
-  /////////////////////////////////////////////////////
-  // Instantiate different versions depending on Impl
-  /////////////////////////////////////////////////////
-  void M5D(const FermionField &psi,
-	   const FermionField &phi,
-	   FermionField &chi,
-	   Vector<Coeff_t> &lower,
-	   Vector<Coeff_t> &diag,
-	   Vector<Coeff_t> &upper);
-
-  void M5Ddag(const FermionField &psi,
-	      const FermionField &phi,
-	      FermionField &chi,
-	      Vector<Coeff_t> &lower,
-	      Vector<Coeff_t> &diag,
-	      Vector<Coeff_t> &upper);
-
-  virtual void   Instantiatable(void)=0;
-
-  // force terms; five routines; default to Dhop on diagonal
-  virtual void MDeriv  (GaugeField &mat,const FermionField &U,const FermionField &V,int dag);
-  virtual void MoeDeriv(GaugeField &mat,const FermionField &U,const FermionField &V,int dag);
-  virtual void MeoDeriv(GaugeField &mat,const FermionField &U,const FermionField &V,int dag);
-
-  // Efficient support for multigrid coarsening
-  virtual void  Mdir (const FermionField &in, FermionField &out,int dir,int disp);
-
-  void   Meooe5D       (const FermionField &in, FermionField &out);
-  void   MeooeDag5D    (const FermionField &in, FermionField &out);
-
-  //    protected:
-  RealD mass;
-
-  // Save arguments to SetCoefficientsInternal
-  Vector<Coeff_t> _gamma;
-  RealD                _zolo_hi;
-  RealD                _b;
-  RealD                _c;
-
-  // Cayley form Moebius (tanh and zolotarev)
-  Vector<Coeff_t> omega;
-  Vector<Coeff_t> bs;    // S dependent coeffs
-  Vector<Coeff_t> cs;
-  Vector<Coeff_t> as;
-  // For preconditioning Cayley form
-  Vector<Coeff_t> bee;
-  Vector<Coeff_t> cee;
-  Vector<Coeff_t> aee;
-  Vector<Coeff_t> beo;
-  Vector<Coeff_t> ceo;
-  Vector<Coeff_t> aeo;
-  // LDU factorisation of the eeoo matrix
-  Vector<Coeff_t> lee;
-  Vector<Coeff_t> leem;
-  Vector<Coeff_t> uee;
-  Vector<Coeff_t> ueem;
-  Vector<Coeff_t> dee;
-
-  // Matrices of 5d ee inverse params
-  Vector<iSinglet<Simd> >  MatpInv;
-  Vector<iSinglet<Simd> >  MatmInv;
-  Vector<iSinglet<Simd> >  MatpInvDag;
-  Vector<iSinglet<Simd> >  MatmInvDag;
-
-  // Constructors
-  CayleyFermion5D(GaugeField &_Umu,
-		  GridCartesian         &FiveDimGrid,
-		  GridRedBlackCartesian &FiveDimRedBlackGrid,
-		  GridCartesian         &FourDimGrid,
-		  GridRedBlackCartesian &FourDimRedBlackGrid,
-		  RealD _mass,RealD _M5,const ImplParams &p= ImplParams());
-
-  void CayleyReport(void);
-  void CayleyZeroCounters(void);
-
-  double M5Dflops;
-  double M5Dcalls;
-  double M5Dtime;
-
-  double MooeeInvFlops;
-  double MooeeInvCalls;
-  double MooeeInvTime;
-
-protected:
-  virtual void SetCoefficientsZolotarev(RealD zolohi,Approx::zolotarev_data *zdata,RealD b,RealD c);
-  virtual void SetCoefficientsTanh(Approx::zolotarev_data *zdata,RealD b,RealD c);
-  virtual void SetCoefficientsInternal(RealD zolo_hi,Vector<Coeff_t> & gamma,RealD b,RealD c);
-};
-
-NAMESPACE_END(Grid);
-

Grid/qcd/action/fermion/ContinuedFractionFermion5D.h

@@ -1,104 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/ContinuedFractionFermion5D.h
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-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 */
-#ifndef  GRID_QCD_CONTINUED_FRACTION_H
-#define  GRID_QCD_CONTINUED_FRACTION_H
-
-#include <Grid/qcd/action/fermion/WilsonFermion5D.h>
-
-NAMESPACE_BEGIN(Grid);
-
-template<class Impl>
-class ContinuedFractionFermion5D : public WilsonFermion5D<Impl>
-{
-public:
-  INHERIT_IMPL_TYPES(Impl);
-public:
-
-  // override multiply
-  virtual RealD  M    (const FermionField &in, FermionField &out);
-  virtual RealD  Mdag (const FermionField &in, FermionField &out);
-
-  // half checkerboard operaions
-  virtual void   Meooe       (const FermionField &in, FermionField &out);
-  virtual void   MeooeDag    (const FermionField &in, FermionField &out);
-  virtual void   Mooee       (const FermionField &in, FermionField &out);
-  virtual void   MooeeDag    (const FermionField &in, FermionField &out);
-  virtual void   MooeeInv    (const FermionField &in, FermionField &out);
-  virtual void   MooeeInvDag (const FermionField &in, FermionField &out);
-
-  // force terms; five routines; default to Dhop on diagonal
-  virtual void MDeriv  (GaugeField &mat,const FermionField &U,const FermionField &V,int dag);
-  virtual void MoeDeriv(GaugeField &mat,const FermionField &U,const FermionField &V,int dag);
-  virtual void MeoDeriv(GaugeField &mat,const FermionField &U,const FermionField &V,int dag);
-
-  //      virtual void   Instantiatable(void)=0;
-  virtual void   Instantiatable(void) =0;
-
-  // Efficient support for multigrid coarsening
-  virtual void  Mdir (const FermionField &in, FermionField &out,int dir,int disp);
-
-      ///////////////////////////////////////////////////////////////
-      // Physical surface field utilities
-      ///////////////////////////////////////////////////////////////
-      //      virtual void Dminus(const FermionField &psi, FermionField &chi);     // Inherit trivial case
-      //      virtual void DminusDag(const FermionField &psi, FermionField &chi);  // Inherit trivial case
-      virtual void ExportPhysicalFermionSolution(const FermionField &solution5d,FermionField &exported4d);
-      virtual void ImportPhysicalFermionSource  (const FermionField &input4d,FermionField &imported5d);
-
-  // Constructors
-  ContinuedFractionFermion5D(GaugeField &_Umu,
-			     GridCartesian         &FiveDimGrid,
-			     GridRedBlackCartesian &FiveDimRedBlackGrid,
-			     GridCartesian         &FourDimGrid,
-			     GridRedBlackCartesian &FourDimRedBlackGrid,
-			     RealD _mass,RealD M5,const ImplParams &p= ImplParams());
-
-protected:
-
-  void SetCoefficientsTanh(Approx::zolotarev_data *zdata,RealD scale);
-  void SetCoefficientsZolotarev(RealD zolo_hi,Approx::zolotarev_data *zdata);;
-
-  // Cont frac
-  RealD dw_diag;
-  RealD mass;
-  RealD R;
-  RealD ZoloHiInv;
-  Vector<double> Beta;
-  Vector<double> cc;;
-  Vector<double> cc_d;;
-  Vector<double> sqrt_cc;
-  Vector<double> See;
-  Vector<double> Aee;
-
-};
-
-
-NAMESPACE_END(Grid);
-
-#endif

Grid/qcd/action/fermion/DomainWallEOFAFermion.h

@@ -1,90 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/DomainWallEOFAFermion.h
-
-Copyright (C) 2017
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: David Murphy <dmurphy@phys.columbia.edu>
-
-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/AbstractEOFAFermion.h>
-
-NAMESPACE_BEGIN(Grid);
-
-template<class Impl>
-class DomainWallEOFAFermion : public AbstractEOFAFermion<Impl>
-{
-public:
-  INHERIT_IMPL_TYPES(Impl);
-
-public:
-  // Modified (0,Ls-1) and (Ls-1,0) elements of Mooee
-  // for red-black preconditioned Shamir EOFA
-  Coeff_t dm;
-  Coeff_t dp;
-
-  virtual void Instantiatable(void) {};
-
-  // EOFA-specific operations
-  virtual void  Omega      (const FermionField& in, FermionField& out, int sign, int dag);
-  virtual void  Dtilde     (const FermionField& in, FermionField& out);
-  virtual void  DtildeInv  (const FermionField& in, FermionField& out);
-
-  // override multiply
-  virtual RealD M          (const FermionField& in, FermionField& out);
-  virtual RealD Mdag       (const FermionField& in, FermionField& out);
-
-  // half checkerboard operations
-  virtual void  Mooee      (const FermionField& in, FermionField& out);
-  virtual void  MooeeDag   (const FermionField& in, FermionField& out);
-  virtual void  MooeeInv   (const FermionField& in, FermionField& out);
-  virtual void  MooeeInvDag(const FermionField& in, FermionField& out);
-
-  virtual void   M5D       (const FermionField& psi, FermionField& chi);
-  virtual void   M5Ddag    (const FermionField& psi, FermionField& chi);
-
-  /////////////////////////////////////////////////////
-  // Instantiate different versions depending on Impl
-  /////////////////////////////////////////////////////
-  void M5D(const FermionField& psi, const FermionField& phi, FermionField& chi,
-	   Vector<Coeff_t>& lower, Vector<Coeff_t>& diag, Vector<Coeff_t>& upper);
-
-  void M5Ddag(const FermionField& psi, const FermionField& phi, FermionField& chi,
-	      Vector<Coeff_t>& lower, Vector<Coeff_t>& diag, Vector<Coeff_t>& upper);
-
-  virtual void RefreshShiftCoefficients(RealD new_shift);
-
-  // Constructors
-  DomainWallEOFAFermion(GaugeField& _Umu, GridCartesian& FiveDimGrid, GridRedBlackCartesian& FiveDimRedBlackGrid,
-			GridCartesian& FourDimGrid, GridRedBlackCartesian& FourDimRedBlackGrid,
-			RealD _mq1, RealD _mq2, RealD _mq3, RealD _shift, int pm,
-			RealD _M5, const ImplParams& p=ImplParams());
-
-protected:
-  void SetCoefficientsInternal(RealD zolo_hi, Vector<Coeff_t>& gamma, RealD b, RealD c);
-};
-
-NAMESPACE_END(Grid);
-

Grid/qcd/action/fermion/DomainWallFermion.h

@@ -1,139 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/DomainWallFermion.h
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Vera Guelpers <V.M.Guelpers@soton.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 */
-#ifndef  GRID_QCD_DOMAIN_WALL_FERMION_H
-#define  GRID_QCD_DOMAIN_WALL_FERMION_H
-
-#include <Grid/qcd/action/fermion/FermionCore.h>
-
-NAMESPACE_BEGIN(Grid);
-
-template<class Impl>
-class DomainWallFermion : public CayleyFermion5D<Impl>
-{
-public:
-  INHERIT_IMPL_TYPES(Impl);
-public:
-
-  void FreePropagator(const FermionField &in,FermionField &out,RealD mass,std::vector<Complex> boundary, std::vector<double> twist, bool fiveD) {
-	FermionField in_k(in.Grid());
-	FermionField prop_k(in.Grid());
-
-	FFT theFFT((GridCartesian *) in.Grid());
-
-	//phase for boundary condition
-	ComplexField coor(in.Grid());
-	ComplexField ph(in.Grid());  ph = Zero();
-	FermionField in_buf(in.Grid()); in_buf = Zero();
-	typedef typename Simd::scalar_type Scalar;
-	Scalar ci(0.0,1.0);
-	assert(twist.size() == Nd);//check that twist is Nd
-	assert(boundary.size() == Nd);//check that boundary conditions is Nd
-	int shift = 0;
-	if(fiveD) shift = 1;
-	for(unsigned int nu = 0; nu < Nd; nu++)
-	{
-	  // Shift coordinate lattice index by 1 to account for 5th dimension.
-          LatticeCoordinate(coor, nu + shift);
-	  double boundary_phase = ::acos(real(boundary[nu]));
-	  ph = ph + boundary_phase*coor*((1./(in.Grid()->_fdimensions[nu+shift])));
-	  //momenta for propagator shifted by twist+boundary
-	  twist[nu] = twist[nu] + boundary_phase/((2.0*M_PI));
-	}
-	in_buf = exp(ci*ph*(-1.0))*in;
-
-	if(fiveD){//FFT only on temporal and spatial dimensions
-          std::vector<int> mask(Nd+1,1); mask[0] = 0;
-	  theFFT.FFT_dim_mask(in_k,in_buf,mask,FFT::forward);
-          this->MomentumSpacePropagatorHt_5d(prop_k,in_k,mass,twist);
-          theFFT.FFT_dim_mask(out,prop_k,mask,FFT::backward);
-        }
-	else{
-	  theFFT.FFT_all_dim(in_k,in,FFT::forward);
-          this->MomentumSpacePropagatorHt(prop_k,in_k,mass,twist);
-	  theFFT.FFT_all_dim(out,prop_k,FFT::backward);
-        }
-	//phase for boundary condition
-	out = out * exp(Scalar(2.0*M_PI)*ci*ph);
-      };
-
-      virtual void FreePropagator(const FermionField &in,FermionField &out,RealD mass,std::vector<Complex> boundary,std::vector<double> twist) {
-        bool fiveD = true; //5d propagator by default
-	FreePropagator(in,out,mass,boundary,twist,fiveD);
-      };
-
-      virtual void FreePropagator(const FermionField &in,FermionField &out,RealD mass, bool fiveD) {
-	std::vector<double> twist(Nd,0.0); //default: periodic boundarys in all directions
-	std::vector<Complex> boundary;
-	for(int i=0;i<Nd;i++) boundary.push_back(1);//default: periodic boundary conditions
-	FreePropagator(in,out,mass,boundary,twist,fiveD);
-      };
-
-      virtual void FreePropagator(const FermionField &in,FermionField &out,RealD mass) {
-        bool fiveD = true; //5d propagator by default
-	std::vector<double> twist(Nd,0.0); //default: twist angle 0
-	std::vector<Complex> boundary;
-	for(int i=0;i<Nd;i++) boundary.push_back(1); //default: periodic boundary conditions
-	FreePropagator(in,out,mass,boundary,twist,fiveD);
-      };
-
-  virtual void   Instantiatable(void) {};
-  // Constructors
-  DomainWallFermion(GaugeField &_Umu,
-		    GridCartesian         &FiveDimGrid,
-		    GridRedBlackCartesian &FiveDimRedBlackGrid,
-		    GridCartesian         &FourDimGrid,
-		    GridRedBlackCartesian &FourDimRedBlackGrid,
-		    RealD _mass,RealD _M5,const ImplParams &p= ImplParams()) : 
-
-
-    CayleyFermion5D<Impl>(_Umu,
-			  FiveDimGrid,
-			  FiveDimRedBlackGrid,
-			  FourDimGrid,
-			  FourDimRedBlackGrid,_mass,_M5,p)
-
-  {
-    RealD eps = 1.0;
-
-    Approx::zolotarev_data *zdata = Approx::higham(eps,this->Ls);// eps is ignored for higham
-    assert(zdata->n==this->Ls);
-	
-    //    std::cout<<GridLogMessage << "DomainWallFermion with Ls="<<this->Ls<<std::endl;
-    // Call base setter
-    this->SetCoefficientsTanh(zdata,1.0,0.0);
-
-    Approx::zolotarev_free(zdata);
-  }
-
-};
-
-NAMESPACE_END(Grid);
-
-#endif

Grid/qcd/action/fermion/DomainWallVec5dImpl.h

@@ -1,213 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/FermionOperatorImpl.h
-
-Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@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 S,class Representation = FundamentalRepresentation, class Options=CoeffReal>
-class DomainWallVec5dImpl :  public PeriodicGaugeImpl< GaugeImplTypes< S,Representation::Dimension> > { 
-public:
-
-  typedef PeriodicGaugeImpl<GaugeImplTypes<S, Representation::Dimension> > Gimpl;
-  INHERIT_GIMPL_TYPES(Gimpl);
-
-  static const int Dimension = Representation::Dimension;
-  static const bool isFundamental = Representation::isFundamental;
-  static const bool LsVectorised=true;
-  static const int Nhcs = Options::Nhcs;
-      
-  typedef typename Options::_Coeff_t Coeff_t;      
-  typedef typename Options::template PrecisionMapper<Simd>::LowerPrecVector SimdL;
-  
-  template <typename vtype> using iImplSpinor            = iScalar<iVector<iVector<vtype, Dimension>, Ns> >;
-  template <typename vtype> using iImplPropagator        = iScalar<iMatrix<iMatrix<vtype, Dimension>, Ns> >;
-  template <typename vtype> using iImplHalfSpinor        = iScalar<iVector<iVector<vtype, Dimension>, Nhs> >;
-  template <typename vtype> using iImplHalfCommSpinor    = iScalar<iVector<iVector<vtype, Dimension>, Nhcs> >;
-  template <typename vtype> using iImplDoubledGaugeField = iVector<iScalar<iMatrix<vtype, Dimension> >, Nds>;
-  template <typename vtype> using iImplGaugeField        = iVector<iScalar<iMatrix<vtype, Dimension> >, Nd>;
-  template <typename vtype> using iImplGaugeLink         = iScalar<iScalar<iMatrix<vtype, Dimension> > >;
-  
-  typedef iImplSpinor<Simd>            SiteSpinor;
-  typedef iImplPropagator<Simd>        SitePropagator;
-  typedef iImplHalfSpinor<Simd>        SiteHalfSpinor;
-  typedef iImplHalfCommSpinor<SimdL>   SiteHalfCommSpinor;
-  typedef Lattice<SiteSpinor>          FermionField;
-  typedef Lattice<SitePropagator>      PropagatorField;
-
-  /////////////////////////////////////////////////
-  // Make the doubled gauge field a *scalar*
-  /////////////////////////////////////////////////
-  typedef iImplDoubledGaugeField<typename Simd::scalar_type>  SiteDoubledGaugeField;  // This is a scalar
-  typedef iImplGaugeField<typename Simd::scalar_type>         SiteScalarGaugeField;  // scalar
-  typedef iImplGaugeLink<typename Simd::scalar_type>          SiteScalarGaugeLink;  // scalar
-  typedef Lattice<SiteDoubledGaugeField>                      DoubledGaugeField;
-      
-  typedef WilsonCompressor<SiteHalfCommSpinor,SiteHalfSpinor, SiteSpinor> Compressor;
-  typedef WilsonImplParams ImplParams;
-  typedef WilsonStencil<SiteSpinor, SiteHalfSpinor,ImplParams> StencilImpl;
-  typedef typename StencilImpl::View_type StencilView;
-  
-  ImplParams Params;
-
-  DomainWallVec5dImpl(const ImplParams &p = ImplParams()) : Params(p){};
-      
-  template <class ref>
-  static accelerator_inline void loadLinkElement(Simd &reg, ref &memory) 
-  {
-    vsplat(reg, memory);
-  }
-
-  template<class _Spinor>
-  static accelerator_inline void multLink(_Spinor &phi, const SiteDoubledGaugeField &U,
-					  const _Spinor &chi, int mu, StencilEntry *SE,
-					  StencilView &St) 
-  {
-#ifdef GPU_VEC
-    // Gauge link is scalarised
-    mult(&phi(), &U(mu), &chi());
-#else
-    SiteGaugeLink UU;
-    for (int i = 0; i < Dimension; i++) {
-      for (int j = 0; j < Dimension; j++) {
-        vsplat(UU()()(i, j), U(mu)()(i, j));
-      }
-    }
-    mult(&phi(), &UU(), &chi());
-#endif
-  }
-
-  inline void DoubleStore(GridBase *GaugeGrid, DoubledGaugeField &Uds,const GaugeField &Umu) 
-  {
-    SiteScalarGaugeField  ScalarUmu;
-    SiteDoubledGaugeField ScalarUds;
-    
-    GaugeLinkField U(Umu.Grid());
-    GaugeField  Uadj(Umu.Grid());
-    for (int mu = 0; mu < Nd; mu++) {
-      U = PeekIndex<LorentzIndex>(Umu, mu);
-      U = adj(Cshift(U, mu, -1));
-      PokeIndex<LorentzIndex>(Uadj, U, mu);
-    }
-    
-    for (int lidx = 0; lidx < GaugeGrid->lSites(); lidx++) {
-      Coordinate lcoor;
-      GaugeGrid->LocalIndexToLocalCoor(lidx, lcoor);
-      
-      peekLocalSite(ScalarUmu, Umu, lcoor);
-      for (int mu = 0; mu < 4; mu++) ScalarUds(mu) = ScalarUmu(mu);
-      
-      peekLocalSite(ScalarUmu, Uadj, lcoor);
-      for (int mu = 0; mu < 4; mu++) ScalarUds(mu + 4) = ScalarUmu(mu);
-      
-      pokeLocalSite(ScalarUds, Uds, lcoor);
-    }
-  }
-      
-  inline void InsertForce4D(GaugeField &mat, FermionField &Btilde,FermionField &A, int mu) 
-  {
-    assert(0);
-  }
-
-  inline void outerProductImpl(PropagatorField &mat, const FermionField &Btilde, const FermionField &A){
-    assert(0);
-  } 
-
-  inline void TraceSpinImpl(GaugeLinkField &mat, PropagatorField&P) {
-    assert(0);
-  }
-
-  inline void extractLinkField(std::vector<GaugeLinkField> &mat, DoubledGaugeField &Uds){
-    assert(0);
-  }
-
-
-  inline void InsertForce5D(GaugeField &mat, FermionField &Btilde, FermionField &Atilde, int mu) {
-
-    assert(0);
-    // Following lines to be revised after Peter's addition of half prec
-    // missing put lane...
-    /*
-      typedef decltype(traceIndex<SpinIndex>(outerProduct(Btilde[0], Atilde[0]))) result_type;
-      unsigned int LLs = Btilde.Grid()->_rdimensions[0];
-      conformable(Atilde.Grid(),Btilde.Grid());
-      GridBase* grid = mat.Grid();
-      GridBase* Bgrid = Btilde.Grid();
-      unsigned int dimU = grid->Nd();
-      unsigned int dimF = Bgrid->Nd();
-      GaugeLinkField tmp(grid); 
-      tmp = Zero();
-    
-      // FIXME 
-      // Current implementation works, thread safe, probably suboptimal
-      // Passing through the local coordinate for grid transformation
-      // the force grid is in general very different from the Ls vectorized grid
-
-      for (int so = 0; so < grid->oSites(); so++) {
-      std::vector<typename result_type::scalar_object> vres(Bgrid->Nsimd());
-      std::vector<int> ocoor;  grid->oCoorFromOindex(ocoor,so); 
-      for (int si = 0; si < tmp.Grid()->iSites(); si++){
-      typename result_type::scalar_object scalar_object; scalar_object = Zero();
-      std::vector<int> local_coor;      
-      std::vector<int> icoor; grid->iCoorFromIindex(icoor,si);
-      grid->InOutCoorToLocalCoor(ocoor, icoor, local_coor);
-      for (int s = 0; s < LLs; s++) {
-      std::vector<int> slocal_coor(dimF);
-      slocal_coor[0] = s;
-      for (int s4d = 1; s4d< dimF; s4d++) slocal_coor[s4d] = local_coor[s4d-1];
-      int sF = Bgrid->oIndexReduced(slocal_coor);  
-      assert(sF < Bgrid->oSites());
-
-      extract(traceIndex<SpinIndex>(outerProduct(Btilde[sF], Atilde[sF])), vres); 
-      // sum across the 5d dimension
-      for (auto v : vres) scalar_object += v;  
-      }
-      tmp[so].putlane(scalar_object, si);
-      }
-      }
-      PokeIndex<LorentzIndex>(mat, tmp, mu);
-    */
-  }
-};
-typedef DomainWallVec5dImpl<vComplex ,FundamentalRepresentation, CoeffReal> DomainWallVec5dImplR; // Real.. whichever prec
-typedef DomainWallVec5dImpl<vComplexF,FundamentalRepresentation, CoeffReal> DomainWallVec5dImplF; // Float
-typedef DomainWallVec5dImpl<vComplexD,FundamentalRepresentation, CoeffReal> DomainWallVec5dImplD; // Double
- 
-typedef DomainWallVec5dImpl<vComplex ,FundamentalRepresentation, CoeffRealHalfComms> DomainWallVec5dImplRL; // Real.. whichever prec
-typedef DomainWallVec5dImpl<vComplexF,FundamentalRepresentation, CoeffRealHalfComms> DomainWallVec5dImplFH; // Float
-typedef DomainWallVec5dImpl<vComplexD,FundamentalRepresentation, CoeffRealHalfComms> DomainWallVec5dImplDF; // Double
- 
-typedef DomainWallVec5dImpl<vComplex ,FundamentalRepresentation,CoeffComplex> ZDomainWallVec5dImplR; // Real.. whichever prec
-typedef DomainWallVec5dImpl<vComplexF,FundamentalRepresentation,CoeffComplex> ZDomainWallVec5dImplF; // Float
-typedef DomainWallVec5dImpl<vComplexD,FundamentalRepresentation,CoeffComplex> ZDomainWallVec5dImplD; // Double
- 
-typedef DomainWallVec5dImpl<vComplex ,FundamentalRepresentation,CoeffComplexHalfComms> ZDomainWallVec5dImplRL; // Real.. whichever prec
-typedef DomainWallVec5dImpl<vComplexF,FundamentalRepresentation,CoeffComplexHalfComms> ZDomainWallVec5dImplFH; // Float
-typedef DomainWallVec5dImpl<vComplexD,FundamentalRepresentation,CoeffComplexHalfComms> ZDomainWallVec5dImplDF; // Double
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/FermionOperator.h

@@ -1,189 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/FermionOperator.h
-
-    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: Vera Guelpers <V.M.Guelpers@soton.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);
-
-////////////////////////////////////////////////////////////////
-// Allow to select  between gauge representation rank bc's, flavours etc.
-// and single/double precision.
-////////////////////////////////////////////////////////////////
-    
-template<class Impl>
-class FermionOperator : public CheckerBoardedSparseMatrixBase<typename Impl::FermionField>, public Impl
-{
-public:
-
-  INHERIT_IMPL_TYPES(Impl);
-
-  FermionOperator(const ImplParams &p= ImplParams()) : Impl(p) {};
-  virtual ~FermionOperator(void) = default;
-
-  virtual FermionField &tmp(void) = 0;
-
-  GridBase * Grid(void)   { return FermionGrid(); };   // this is all the linalg routines need to know
-  GridBase * RedBlackGrid(void) { return FermionRedBlackGrid(); };
-
-  virtual GridBase *FermionGrid(void)         =0;
-  virtual GridBase *FermionRedBlackGrid(void) =0;
-  virtual GridBase *GaugeGrid(void)           =0;
-  virtual GridBase *GaugeRedBlackGrid(void)   =0;
-
-  // override multiply
-  virtual RealD  M    (const FermionField &in, FermionField &out)=0;
-  virtual RealD  Mdag (const FermionField &in, FermionField &out)=0;
-
-  // half checkerboard operaions
-  virtual void   Meooe       (const FermionField &in, FermionField &out)=0;
-  virtual void   MeooeDag    (const FermionField &in, FermionField &out)=0;
-  virtual void   Mooee       (const FermionField &in, FermionField &out)=0;
-  virtual void   MooeeDag    (const FermionField &in, FermionField &out)=0;
-  virtual void   MooeeInv    (const FermionField &in, FermionField &out)=0;
-  virtual void   MooeeInvDag (const FermionField &in, FermionField &out)=0;
-
-  // non-hermitian hopping term; half cb or both
-  virtual void Dhop  (const FermionField &in, FermionField &out,int dag)=0;
-  virtual void DhopOE(const FermionField &in, FermionField &out,int dag)=0;
-  virtual void DhopEO(const FermionField &in, FermionField &out,int dag)=0;
-  virtual void DhopDir(const FermionField &in, FermionField &out,int dir,int disp)=0; // implemented by WilsonFermion and WilsonFermion5D
-
-  // force terms; five routines; default to Dhop on diagonal
-  virtual void MDeriv  (GaugeField &mat,const FermionField &U,const FermionField &V,int dag){DhopDeriv(mat,U,V,dag);};
-  virtual void MoeDeriv(GaugeField &mat,const FermionField &U,const FermionField &V,int dag){DhopDerivOE(mat,U,V,dag);};
-  virtual void MeoDeriv(GaugeField &mat,const FermionField &U,const FermionField &V,int dag){DhopDerivEO(mat,U,V,dag);};
-  virtual void MooDeriv(GaugeField &mat,const FermionField &U,const FermionField &V,int dag){mat=Zero();}; // Clover can override these
-  virtual void MeeDeriv(GaugeField &mat,const FermionField &U,const FermionField &V,int dag){mat=Zero();};
-
-  virtual void DhopDeriv  (GaugeField &mat,const FermionField &U,const FermionField &V,int dag)=0;
-  virtual void DhopDerivEO(GaugeField &mat,const FermionField &U,const FermionField &V,int dag)=0;
-  virtual void DhopDerivOE(GaugeField &mat,const FermionField &U,const FermionField &V,int dag)=0;
-
-
-  virtual void  Mdiag  (const FermionField &in, FermionField &out) { Mooee(in,out);};   // Same as Mooee applied to both CB's
-  virtual void  Mdir   (const FermionField &in, FermionField &out,int dir,int disp)=0;   // case by case Wilson, Clover, Cayley, ContFrac, PartFrac
-
-
-      virtual void  MomentumSpacePropagator(FermionField &out,const FermionField &in,RealD _m,std::vector<double> twist) { assert(0);};
-
-      virtual void  FreePropagator(const FermionField &in,FermionField &out,RealD mass,std::vector<Complex> boundary,std::vector<double> twist) 
-      {
-	FFT theFFT((GridCartesian *) in.Grid());
-
-	typedef typename Simd::scalar_type Scalar;
-
-	FermionField in_k(in.Grid());
-	FermionField prop_k(in.Grid());
-
-	//phase for boundary condition
-	ComplexField coor(in.Grid());
-	ComplexField ph(in.Grid());  ph = Zero();
-	FermionField in_buf(in.Grid()); in_buf = Zero();
-
-	Scalar ci(0.0,1.0);
-	assert(twist.size() == Nd);//check that twist is Nd
-	assert(boundary.size() == Nd);//check that boundary conditions is Nd
-	for(unsigned int nu = 0; nu < Nd; nu++)
-	{
-          LatticeCoordinate(coor, nu);
-	  double boundary_phase = ::acos(real(boundary[nu]));
-	  ph = ph + boundary_phase*coor*((1./(in.Grid()->_fdimensions[nu])));
-	  //momenta for propagator shifted by twist+boundary
-	  twist[nu] = twist[nu] + boundary_phase/((2.0*M_PI));
-	}
-	in_buf = exp(ci*ph*(-1.0))*in;
-
-	theFFT.FFT_all_dim(in_k,in_buf,FFT::forward);
-        this->MomentumSpacePropagator(prop_k,in_k,mass,twist);
-	theFFT.FFT_all_dim(out,prop_k,FFT::backward);
-
-	//phase for boundary condition
-        out = out * exp(Scalar(2.0*M_PI)*ci*ph);
-
-      };
-
-      virtual void FreePropagator(const FermionField &in,FermionField &out,RealD mass) {
-	std::vector<Complex> boundary;
-	for(int i=0;i<Nd;i++) boundary.push_back(1);//default: periodic boundary conditions
-	std::vector<double> twist(Nd,0.0); //default: periodic boundarys in all directions
-	FreePropagator(in,out,mass,boundary,twist);
-      };
-
-  ///////////////////////////////////////////////
-  // Updates gauge field during HMC
-  ///////////////////////////////////////////////
-  virtual void ImportGauge(const GaugeField & _U)=0;
-
-  //////////////////////////////////////////////////////////////////////
-  // Conserved currents, either contract at sink or insert sequentially.
-  //////////////////////////////////////////////////////////////////////
-  virtual void ContractConservedCurrent(PropagatorField &q_in_1,
-					PropagatorField &q_in_2,
-					PropagatorField &q_out,
-					Current curr_type,
-					unsigned int mu)=0;
-  virtual void SeqConservedCurrent(PropagatorField &q_in, 
-				   PropagatorField &q_out,
-				   Current curr_type,
-				   unsigned int mu,
-				   unsigned int tmin, 
-				   unsigned int tmax,
-				   ComplexField &lattice_cmplx)=0;
-
-      // Only reimplemented in Wilson5D 
-      // Default to just a zero correlation function
-  virtual void ContractJ5q(FermionField &q_in   ,ComplexField &J5q) { J5q=Zero(); };
-  virtual void ContractJ5q(PropagatorField &q_in,ComplexField &J5q) { J5q=Zero(); };
-
-      ///////////////////////////////////////////////
-      // Physical field import/export
-      ///////////////////////////////////////////////
-      virtual void Dminus(const FermionField &psi, FermionField &chi)    { chi=psi; }
-      virtual void DminusDag(const FermionField &psi, FermionField &chi) { chi=psi; }
-      virtual void ImportPhysicalFermionSource(const FermionField &input,FermionField &imported)
-      {
-	imported = input;
-      };
-      virtual void ImportUnphysicalFermion(const FermionField &input,FermionField &imported)
-      {
-	imported=input;
-      };
-      virtual void ExportPhysicalFermionSolution(const FermionField &solution,FermionField &exported)
-      {
-	exported=solution;
-      };
-      virtual void ExportPhysicalFermionSource(const FermionField &solution,FermionField &exported)
-      {
-	exported=solution;
-      };
-};
-
-NAMESPACE_END(Grid);
-

Grid/qcd/action/fermion/FermionOperatorImpl.h

@@ -1,189 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/FermionOperatorImpl.h
-
-Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is 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 parameter class constructs to package
-// externally control Fermion implementations
-// in orthogonal directions
-//
-// Ultimately need Impl to always define types where XXX is opaque
-//
-//    typedef typename XXX               Simd;
-//    typedef typename XXX     GaugeLinkField;        
-//    typedef typename XXX         GaugeField;
-//    typedef typename XXX      GaugeActField;
-//    typedef typename XXX       FermionField;
-//    typedef typename XXX    PropagatorField;
-//    typedef typename XXX  DoubledGaugeField;
-//    typedef typename XXX         SiteSpinor;
-//    typedef typename XXX     SitePropagator;
-//    typedef typename XXX     SiteHalfSpinor;	
-//    typedef typename XXX         Compressor;	
-//
-// and Methods:
-//    void ImportGauge(GridBase *GaugeGrid,DoubledGaugeField &Uds,const GaugeField &Umu)
-//    void DoubleStore(GridBase *GaugeGrid,DoubledGaugeField &Uds,const GaugeField &Umu)
-//    void multLink(SiteHalfSpinor &phi,const SiteDoubledGaugeField &U,const SiteHalfSpinor &chi,int mu,StencilEntry *SE,StencilImpl::View_type &St)
-//    void InsertForce4D(GaugeField &mat,const FermionField &Btilde,const FermionField &A,int mu)
-//    void InsertForce5D(GaugeField &mat,const FermionField &Btilde,const FermionField &A,int mu)
-//
-//
-// To acquire the typedefs from "Base" (either a base class or template param) use:
-//
-// INHERIT_GIMPL_TYPES(Base)
-// INHERIT_FIMPL_TYPES(Base)
-// INHERIT_IMPL_TYPES(Base)
-//
-// The Fermion operators will do the following:
-//
-// struct MyOpParams { 
-//   RealD mass;
-// };
-//
-//
-// template<class Impl>
-// class MyOp : public<Impl> { 
-// public:
-//
-//    INHERIT_ALL_IMPL_TYPES(Impl);
-//
-//    MyOp(MyOpParams Myparm, ImplParams &ImplParam) :  Impl(ImplParam)
-//    {
-//
-//    };
-//    
-//  }
-//////////////////////////////////////////////
-
-template <class T> struct SamePrecisionMapper {
-  typedef T HigherPrecVector ;
-  typedef T LowerPrecVector ;
-};
-template <class T> struct LowerPrecisionMapper {  };
-template <> struct LowerPrecisionMapper<vRealF> {
-  typedef vRealF HigherPrecVector ;
-  typedef vRealH LowerPrecVector ;
-};
-template <> struct LowerPrecisionMapper<vRealD> {
-  typedef vRealD HigherPrecVector ;
-  typedef vRealF LowerPrecVector ;
-};
-template <> struct LowerPrecisionMapper<vComplexF> {
-  typedef vComplexF HigherPrecVector ;
-  typedef vComplexH LowerPrecVector ;
-};
-template <> struct LowerPrecisionMapper<vComplexD> {
-  typedef vComplexD HigherPrecVector ;
-  typedef vComplexF LowerPrecVector ;
-};
-
-struct CoeffReal {
-public:
-  typedef RealD _Coeff_t;
-  static const int Nhcs = 2;
-  template<class Simd> using PrecisionMapper = SamePrecisionMapper<Simd>;
-};
-struct CoeffRealHalfComms {
-public:
-  typedef RealD _Coeff_t;
-  static const int Nhcs = 1;
-  template<class Simd> using PrecisionMapper = LowerPrecisionMapper<Simd>;
-};
-struct CoeffComplex {
-public:
-  typedef ComplexD _Coeff_t;
-  static const int Nhcs = 2;
-  template<class Simd> using PrecisionMapper = SamePrecisionMapper<Simd>;
-};
-struct CoeffComplexHalfComms {
-public:
-  typedef ComplexD _Coeff_t;
-  static const int Nhcs = 1;
-  template<class Simd> using PrecisionMapper = LowerPrecisionMapper<Simd>;
-};
-
-////////////////////////////////////////////////////////////////////////
-// Implementation dependent fermion types
-////////////////////////////////////////////////////////////////////////
-  
-#define INHERIT_FIMPL_TYPES(Impl)\
-  typedef typename Impl::Coeff_t                     Coeff_t;           \
-  typedef Impl Impl_t;							\
-  typedef typename Impl::FermionField           FermionField;		\
-  typedef typename Impl::PropagatorField     PropagatorField;		\
-  typedef typename Impl::DoubledGaugeField DoubledGaugeField;		\
-  typedef typename Impl::SiteDoubledGaugeField SiteDoubledGaugeField;	\
-  typedef typename Impl::SiteSpinor               SiteSpinor;		\
-  typedef typename Impl::SitePropagator       SitePropagator;		\
-  typedef typename Impl::SiteHalfSpinor       SiteHalfSpinor;		\
-  typedef typename Impl::Compressor               Compressor;		\
-  typedef typename Impl::StencilImpl             StencilImpl;		\
-  typedef typename Impl::ImplParams               ImplParams;	        \
-  typedef typename Impl::StencilImpl::View_type  StencilView;		\
-  typedef typename ViewMap<FermionField>::Type      FermionFieldView;	\
-  typedef typename ViewMap<DoubledGaugeField>::Type DoubledGaugeFieldView;
-
-#define INHERIT_IMPL_TYPES(Base)		\
-  INHERIT_GIMPL_TYPES(Base)			\
-  INHERIT_FIMPL_TYPES(Base)
-
-NAMESPACE_END(Grid);
-NAMESPACE_CHECK(ImplBase);  
-/////////////////////////////////////////////////////////////////////////////
-// Single flavour four spinors with colour index
-/////////////////////////////////////////////////////////////////////////////
-#include <Grid/qcd/action/fermion/WilsonImpl.h> 
-NAMESPACE_CHECK(ImplWilson);  
-   
-////////////////////////////////////////////////////////////////////////////////////////
-// Flavour doubled spinors; is Gparity the only? what about C*?
-////////////////////////////////////////////////////////////////////////////////////////
-#include <Grid/qcd/action/fermion/GparityWilsonImpl.h> 
-NAMESPACE_CHECK(ImplGparityWilson);  
-
-/////////////////////////////////////////////////////////////////////////////
-// Single flavour one component spinors with colour index
-/////////////////////////////////////////////////////////////////////////////
-#include <Grid/qcd/action/fermion/StaggeredImpl.h> 
-NAMESPACE_CHECK(ImplStaggered);  
-
-/////////////////////////////////////////////////////////////////////////////
-// Single flavour one component spinors with colour index. 5d vec
-/////////////////////////////////////////////////////////////////////////////
-#include <Grid/qcd/action/fermion/StaggeredVec5dImpl.h> 
-NAMESPACE_CHECK(ImplStaggered5dVec);  
-
-

Grid/qcd/action/fermion/FourierAcceleratedPV.h

@@ -1,238 +0,0 @@
-
-    /*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/FourierAcceleratedPV.h
-
-    Copyright (C) 2015
-
-Author: Christoph Lehner (lifted with permission by Peter Boyle, brought back to Grid)
-Author: Peter Boyle <pabobyle@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<typename M>
-    void get_real_const_bc(M& m, RealD& _b, RealD& _c) {
-    ComplexD b,c;
-    b=m.bs[0];
-    c=m.cs[0];
-    std::cout << GridLogMessage << "b=" << b << ", c=" << c << std::endl;
-    for (size_t i=1;i<m.bs.size();i++) {
-      assert(m.bs[i] == b);
-      assert(m.cs[i] == c);
-    }
-    assert(b.imag() == 0.0);
-    assert(c.imag() == 0.0);
-    _b = b.real();
-    _c = c.real();
-  }
-
-
-template<typename Vi, typename M, typename G>
-class FourierAcceleratedPV {
- public:
-
-  ConjugateGradient<Vi> &cg;
-  M& dwfPV;
-  G& Umu;
-  GridCartesian* grid5D;
-  GridRedBlackCartesian* gridRB5D;
-  int group_in_s;
-
-  FourierAcceleratedPV(M& _dwfPV, G& _Umu, ConjugateGradient<Vi> &_cg, int _group_in_s = 2) 
-   : dwfPV(_dwfPV), Umu(_Umu), cg(_cg), group_in_s(_group_in_s) 
-  {
-    assert( dwfPV.FermionGrid()->_fdimensions[0] % (2*group_in_s) == 0);
-    grid5D   = SpaceTimeGrid::makeFiveDimGrid(2*group_in_s, (GridCartesian*)Umu.Grid());
-    gridRB5D = SpaceTimeGrid::makeFiveDimRedBlackGrid(2*group_in_s, (GridCartesian*)Umu.Grid());
-  }
-
-  void rotatePV(const Vi& _src, Vi& dst, bool forward) const {
-
-    GridStopWatch gsw1, gsw2;
-
-    typedef typename Vi::scalar_type Coeff_t;
-    int Ls = dst.Grid()->_fdimensions[0];
-
-    Vi _tmp(dst.Grid());
-    double phase = M_PI / (double)Ls;
-    Coeff_t bzero(0.0,0.0);
-
-    FFT theFFT((GridCartesian*)dst.Grid());
-
-    if (!forward) {
-      gsw1.Start();
-      for (int s=0;s<Ls;s++) {
-	Coeff_t a(::cos(phase*s),-::sin(phase*s));
-	axpby_ssp(_tmp,a,_src,bzero,_src,s,s);
-      }
-      gsw1.Stop();
-
-      gsw2.Start();
-      theFFT.FFT_dim(dst,_tmp,0,FFT::forward);
-      gsw2.Stop();
-
-    } else {
-
-      gsw2.Start();
-      theFFT.FFT_dim(_tmp,_src,0,FFT::backward);
-      gsw2.Stop();
-
-      gsw1.Start();
-      for (int s=0;s<Ls;s++) {
-	Coeff_t a(::cos(phase*s),::sin(phase*s));
-	axpby_ssp(dst,a,_tmp,bzero,_tmp,s,s);
-      }
-      gsw1.Stop();
-    }
-
-    std::cout << GridLogMessage << "Timing rotatePV: " << gsw1.Elapsed() << ", " << gsw2.Elapsed() << std::endl;
-
-  }
-
-  void pvInv(const Vi& _src, Vi& _dst) const {
-
-    std::cout << GridLogMessage << "Fourier-Accelerated Outer Pauli Villars"<<std::endl;
-
-    typedef typename Vi::scalar_type Coeff_t;
-    int Ls = _dst.Grid()->_fdimensions[0];
-
-    GridStopWatch gswT;
-    gswT.Start();
-
-    RealD b,c;
-    get_real_const_bc(dwfPV,b,c);
-    RealD M5 = dwfPV.M5;
-    
-    // U(true) Rightinv TMinv U(false) = Minv
-
-    Vi _src_diag(_dst.Grid());
-    Vi _src_diag_slice(dwfPV.GaugeGrid());
-    Vi _dst_diag_slice(dwfPV.GaugeGrid());
-    Vi _src_diag_slices(grid5D);
-    Vi _dst_diag_slices(grid5D);
-    Vi _dst_diag(_dst.Grid());
-
-    rotatePV(_src,_src_diag,false);
-
-    // now do TM solves
-    Gamma G5(Gamma::Algebra::Gamma5);
-
-    GridStopWatch gswA, gswB;
-
-    gswA.Start();
-
-    typedef typename M::Impl_t Impl;
-    //WilsonTMFermion<Impl> tm(x.Umu,*x.UGridF,*x.UrbGridF,0.0,0.0,solver_outer.parent.par.wparams_f);
-    std::vector<RealD> vmass(grid5D->_fdimensions[0],0.0);
-    std::vector<RealD> vmu(grid5D->_fdimensions[0],0.0);
-
-    WilsonTMFermion5D<Impl> tm(Umu,*grid5D,*gridRB5D,
-			   *(GridCartesian*)dwfPV.GaugeGrid(),
-			   *(GridRedBlackCartesian*)dwfPV.GaugeRedBlackGrid(),
-			   vmass,vmu);
-    
-    //SchurRedBlackDiagTwoSolve<Vi> sol(cg);
-    SchurRedBlackDiagMooeeSolve<Vi> sol(cg); // same performance as DiagTwo
-    gswA.Stop();
-
-    gswB.Start();
-
-    for (int sgroup=0;sgroup<Ls/2/group_in_s;sgroup++) {
-
-      for (int sidx=0;sidx<group_in_s;sidx++) {
-
-	int s = sgroup*group_in_s + sidx;
-	//	int sprime = Ls-s-1;
-
-	RealD phase = M_PI / (RealD)Ls * (2.0 * s + 1.0);
-	RealD cosp = ::cos(phase);
-	RealD sinp = ::sin(phase);
-	RealD denom = b*b + c*c + 2.0*b*c*cosp;
-	RealD mass = -(b*b*M5 + c*(1.0 - cosp + c*M5) + b*(-1.0 + cosp + 2.0*c*cosp*M5))/denom;
-	RealD mu = (b+c)*sinp/denom;
-
-	vmass[2*sidx + 0] = mass;
-	vmass[2*sidx + 1] = mass;
-	vmu[2*sidx + 0] = mu;
-	vmu[2*sidx + 1] = -mu;
-
-      }
-
-      tm.update(vmass,vmu);
-
-      for (int sidx=0;sidx<group_in_s;sidx++) {
-
-	int s = sgroup*group_in_s + sidx;
-	int sprime = Ls-s-1;
-
-	ExtractSlice(_src_diag_slice,_src_diag,s,0);
-	InsertSlice(_src_diag_slice,_src_diag_slices,2*sidx + 0,0);
-
-	ExtractSlice(_src_diag_slice,_src_diag,sprime,0);
-	InsertSlice(_src_diag_slice,_src_diag_slices,2*sidx + 1,0);
-
-      }
-
-      GridStopWatch gsw;
-      gsw.Start();
-      _dst_diag_slices = Zero(); // zero guess
-      sol(tm,_src_diag_slices,_dst_diag_slices);
-      gsw.Stop();
-      std::cout << GridLogMessage << "Solve[sgroup=" << sgroup << "] completed in " << gsw.Elapsed() << ", " << gswA.Elapsed() << std::endl;
-
-      for (int sidx=0;sidx<group_in_s;sidx++) {
-
-	int s = sgroup*group_in_s + sidx;
-	int sprime = Ls-s-1;
-
-	RealD phase = M_PI / (RealD)Ls * (2.0 * s + 1.0);
-	RealD cosp = ::cos(phase);
-	RealD sinp = ::sin(phase);
-
-	// now rotate with inverse of
-	Coeff_t pA = b + c*cosp;
-	Coeff_t pB = - Coeff_t(0.0,1.0)*Coeff_t(c*sinp);
-	Coeff_t pABden = pA*pA - pB*pB;
-	// (pA + pB * G5) * (pA - pB*G5) = (pA^2 - pB^2)
-      
-	ExtractSlice(_dst_diag_slice,_dst_diag_slices,2*sidx + 0,0);
-	_dst_diag_slice = (pA/pABden) * _dst_diag_slice - (pB/pABden) * (G5 * _dst_diag_slice);
-	InsertSlice(_dst_diag_slice,_dst_diag,s,0);
-	
-	ExtractSlice(_dst_diag_slice,_dst_diag_slices,2*sidx + 1,0);
-	_dst_diag_slice = (pA/pABden) * _dst_diag_slice + (pB/pABden) * (G5 * _dst_diag_slice);
-	InsertSlice(_dst_diag_slice,_dst_diag,sprime,0);
-      }
-    }
-    gswB.Stop();
-
-    rotatePV(_dst_diag,_dst,true);
-
-    gswT.Stop();
-    std::cout << GridLogMessage << "PV completed in " << gswT.Elapsed() << " (Setup: " << gswA.Elapsed() << ", s-loop: " << gswB.Elapsed() << ")" << std::endl;
-  }
-
-};
-NAMESPACE_END(Grid);
-

Grid/qcd/action/fermion/GparityWilsonImpl.h

@@ -1,321 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/FermionOperatorImpl.h
-
-Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@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 S, class Representation = FundamentalRepresentation, class Options=CoeffReal>
-class GparityWilsonImpl : public ConjugateGaugeImpl<GaugeImplTypes<S, Representation::Dimension> > {
-public:
-
- static const int Dimension = Representation::Dimension;
- static const bool isFundamental = Representation::isFundamental;
- static const int Nhcs = Options::Nhcs;
- static const bool LsVectorised=false;
-
- typedef ConjugateGaugeImpl< GaugeImplTypes<S,Dimension> > Gimpl;
- INHERIT_GIMPL_TYPES(Gimpl);
- 
- typedef typename Options::_Coeff_t Coeff_t;
- typedef typename Options::template PrecisionMapper<Simd>::LowerPrecVector SimdL;
-      
- template <typename vtype> using iImplSpinor            = iVector<iVector<iVector<vtype, Dimension>, Ns>,   Ngp>;
- template <typename vtype> using iImplPropagator        = iVector<iMatrix<iMatrix<vtype, Dimension>, Ns>,   Ngp>;
- template <typename vtype> using iImplHalfSpinor        = iVector<iVector<iVector<vtype, Dimension>, Nhs>,  Ngp>;
- template <typename vtype> using iImplHalfCommSpinor    = iVector<iVector<iVector<vtype, Dimension>, Nhcs>, Ngp>;
- template <typename vtype> using iImplDoubledGaugeField = iVector<iVector<iScalar<iMatrix<vtype, Dimension> >, Nds>, Ngp>;
-
-  typedef iImplSpinor<Simd>            SiteSpinor;
-  typedef iImplPropagator<Simd>        SitePropagator;
-  typedef iImplHalfSpinor<Simd>        SiteHalfSpinor;
-  typedef iImplHalfCommSpinor<SimdL>   SiteHalfCommSpinor;
-  typedef iImplDoubledGaugeField<Simd> SiteDoubledGaugeField;
-
-  typedef Lattice<SiteSpinor> FermionField;
-  typedef Lattice<SitePropagator> PropagatorField;
-  typedef Lattice<SiteDoubledGaugeField> DoubledGaugeField;
- 
-  typedef GparityWilsonImplParams ImplParams;
-  typedef WilsonCompressor<SiteHalfCommSpinor,SiteHalfSpinor, SiteSpinor> Compressor;
-  typedef WilsonStencil<SiteSpinor, SiteHalfSpinor, ImplParams> StencilImpl;
-  typedef typename StencilImpl::View_type StencilView;
-      
-  ImplParams Params;
-
-  GparityWilsonImpl(const ImplParams &p = ImplParams()) : Params(p){};
-
-  // provide the multiply by link that is differentiated between Gparity (with
-  // flavour index) and non-Gparity
-  template<class _Spinor>
-  static accelerator_inline void multLink(_Spinor &phi, 
-					  const SiteDoubledGaugeField &U,
-					  const _Spinor &chi, 
-					  int mu) 
-  {
-    assert(0);
-  } 
-  template<class _Spinor>
-  static accelerator_inline void multLink(_Spinor &phi, 
-					  const SiteDoubledGaugeField &U,
-					  const _Spinor &chi, 
-					  int mu, 
-					  StencilEntry *SE,
-					  StencilView &St) 
-  {
-    int direction = St._directions[mu];
-    int distance  = St._distances[mu];
-    int ptype     = St._permute_type[mu];
-    int sl        = St._simd_layout[direction];
-    Coordinate icoor;
-
-#ifdef __CUDA_ARCH__
-    _Spinor tmp;
-
-    const int Nsimd =SiteDoubledGaugeField::Nsimd();
-    int s = SIMTlane(Nsimd);
-    St.iCoorFromIindex(icoor,s);
-
-    int mmu = mu % Nd;
-    if ( SE->_around_the_world && St.parameters.twists[mmu] ) {
-      
-      int permute_lane = (sl==1) 
-    	|| ((distance== 1)&&(icoor[direction]==1))
-	|| ((distance==-1)&&(icoor[direction]==0));
-
-      if ( permute_lane ) { 
-	tmp(0) = chi(1);
-	tmp(1) = chi(0);
-      } else {
-	tmp(0) = chi(0);
-	tmp(1) = chi(1);
-      }
-
-      auto UU0=coalescedRead(U(0)(mu));
-      auto UU1=coalescedRead(U(1)(mu));
-
-      mult(&phi(0),&UU0,&tmp(0));
-      mult(&phi(1),&UU1,&tmp(1));
-
-    } else {
-
-      auto UU0=coalescedRead(U(0)(mu));
-      auto UU1=coalescedRead(U(1)(mu));
-
-      mult(&phi(0),&UU0,&chi(0));
-      mult(&phi(1),&UU1,&chi(1));
-
-    }
-
-#else
-    typedef _Spinor vobj;
-    typedef typename SiteHalfSpinor::scalar_object sobj;
-    typedef typename SiteHalfSpinor::vector_type   vector_type;
-	
-    vobj vtmp;
-    sobj stmp;
-        
-    const int Nsimd =vector_type::Nsimd();
-    
-    // Fixme X.Y.Z.T hardcode in stencil
-    int mmu = mu % Nd;
-        
-    // assert our assumptions
-    assert((distance == 1) || (distance == -1));  // nearest neighbour stencil hard code
-    assert((sl == 1) || (sl == 2));
-
-    if ( SE->_around_the_world && St.parameters.twists[mmu] ) {
-
-      if ( sl == 2 ) {
-       
-	ExtractBuffer<sobj> vals(Nsimd);
-
-	extract(chi,vals);
-	for(int s=0;s<Nsimd;s++){
-
-	  St.iCoorFromIindex(icoor,s);
-              
-	  assert((icoor[direction]==0)||(icoor[direction]==1));
-              
-	  int permute_lane;
-	  if ( distance == 1) {
-	    permute_lane = icoor[direction]?1:0;
-	  } else {
-	    permute_lane = icoor[direction]?0:1;
-	  }
-              
-	  if ( permute_lane ) { 
-	    stmp(0) = vals[s](1);
-	    stmp(1) = vals[s](0);
-	    vals[s] = stmp;
-	  }
-	}
-	merge(vtmp,vals);
-            
-      } else { 
-	vtmp(0) = chi(1);
-	vtmp(1) = chi(0);
-      }
-      mult(&phi(0),&U(0)(mu),&vtmp(0));
-      mult(&phi(1),&U(1)(mu),&vtmp(1));
-     
-    } else { 
-      mult(&phi(0),&U(0)(mu),&chi(0));
-      mult(&phi(1),&U(1)(mu),&chi(1));
-    }
-#endif   
-  }
-
-  template <class ref>
-  static accelerator_inline void loadLinkElement(Simd &reg, ref &memory) 
-  {
-    reg = memory;
-  }
-
-  inline void DoubleStore(GridBase *GaugeGrid,DoubledGaugeField &Uds,const GaugeField &Umu)
-  {
-    conformable(Uds.Grid(),GaugeGrid);
-    conformable(Umu.Grid(),GaugeGrid);
-   
-    GaugeLinkField Utmp (GaugeGrid);
-    GaugeLinkField U    (GaugeGrid);
-    GaugeLinkField Uconj(GaugeGrid);
-   
-    Lattice<iScalar<vInteger> > coor(GaugeGrid);
-        
-    for(int mu=0;mu<Nd;mu++){
-          
-      LatticeCoordinate(coor,mu);
-          
-      U     = PeekIndex<LorentzIndex>(Umu,mu);
-      Uconj = conjugate(U);
-     
-      // This phase could come from a simple bc 1,1,-1,1 ..
-      int neglink = GaugeGrid->GlobalDimensions()[mu]-1;
-      if ( Params.twists[mu] ) { 
-	Uconj = where(coor==neglink,-Uconj,Uconj);
-      }
-	  
-      auto U_v = U.View();
-      auto Uds_v = Uds.View();
-      auto Uconj_v = Uconj.View();
-      auto Utmp_v= Utmp.View();
-      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
-      Uconj = adj(Cshift(Uconj,mu,-1));
- 
-      Utmp = U;
-      if ( Params.twists[mu] ) { 
-	Utmp = where(coor==0,Uconj,Utmp);
-      }
-
-      thread_foreach(ss,Utmp_v,{
-	Uds_v[ss](0)(mu+4) = Utmp_v[ss]();
-      });
-          
-      Utmp = Uconj;
-      if ( Params.twists[mu] ) { 
-	Utmp = where(coor==0,U,Utmp);
-      }
-	  
-      thread_foreach(ss,Utmp_v,{
-        Uds_v[ss](1)(mu+4) = Utmp_v[ss]();
-      });
-          
-    }
-  }
-      
-  inline void InsertForce4D(GaugeField &mat, FermionField &Btilde, FermionField &A, int mu) {
-
-    // DhopDir provides U or Uconj depending on coor/flavour.
-    GaugeLinkField link(mat.Grid());
-    // use lorentz for flavour as hack.
-    auto tmp = TraceIndex<SpinIndex>(outerProduct(Btilde, A));
-    auto link_v = link.View();
-    auto tmp_v = tmp.View();
-    thread_foreach(ss,tmp_v,{
-      link_v[ss]() = tmp_v[ss](0, 0) + conjugate(tmp_v[ss](1, 1));
-    });
-    PokeIndex<LorentzIndex>(mat, link, mu);
-    return;
-  }
-      
- inline void outerProductImpl(PropagatorField &mat, const FermionField &Btilde, const FermionField &A){
-   //mat = outerProduct(Btilde, A);
-   assert(0);
-  }
-
-  inline void TraceSpinImpl(GaugeLinkField &mat, PropagatorField&P) {
-    assert(0);
-    /*
-    auto tmp = TraceIndex<SpinIndex>(P);
-    parallel_for(auto ss = tmp.begin(); ss < tmp.end(); ss++) {
-      mat[ss]() = tmp[ss](0, 0) + conjugate(tmp[ss](1, 1));
-    }
-    */
-  }
-
-  inline void extractLinkField(std::vector<GaugeLinkField> &mat, DoubledGaugeField &Uds){
-    assert(0);
-  }
-  
-  inline void InsertForce5D(GaugeField &mat, FermionField &Btilde, FermionField &Atilde, int mu) {
-
-    int Ls = Btilde.Grid()->_fdimensions[0];
-        
-    GaugeLinkField tmp(mat.Grid());
-    tmp = Zero();
-    auto tmp_v = tmp.View();
-    auto Atilde_v = Atilde.View();
-    auto Btilde_v = Btilde.View();
-    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;
-  }
-  
-};
-
-typedef GparityWilsonImpl<vComplex , FundamentalRepresentation,CoeffReal> GparityWilsonImplR;  // Real.. whichever prec
-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
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/ImprovedStaggeredFermion5D.h

@@ -1,231 +0,0 @@
-
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/ImprovedStaggeredFermion5D.h
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: AzusaYamaguchi <ayamaguc@staffmail.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);
-
-////////////////////////////////////////////////////////////////////////////////
-// This is the 4d red black case appropriate to support
-////////////////////////////////////////////////////////////////////////////////
-
-class ImprovedStaggeredFermion5DStatic { 
-public:
-  // S-direction is INNERMOST and takes no part in the parity.
-  static const std::vector<int> directions;
-  static const std::vector<int> displacements;
-  const int npoint = 16;
-};
-
-template<class Impl>
-class ImprovedStaggeredFermion5D :  public StaggeredKernels<Impl>, public ImprovedStaggeredFermion5DStatic 
-{
-public:
-  INHERIT_IMPL_TYPES(Impl);
-  typedef StaggeredKernels<Impl> Kernels;
-
-  FermionField _tmp;
-  FermionField &tmp(void) { return _tmp; }
-
-  ////////////////////////////////////////
-  // Performance monitoring
-  ////////////////////////////////////////
-  void Report(void);
-  void ZeroCounters(void);
-  double DhopTotalTime;
-  double DhopCalls;
-  double DhopCommTime;
-  double DhopComputeTime;
-      double DhopComputeTime2;
-      double DhopFaceTime;
-
-  ///////////////////////////////////////////////////////////////
-  // Implement the abstract base
-  ///////////////////////////////////////////////////////////////
-  GridBase *GaugeGrid(void)              { return _FourDimGrid ;}
-  GridBase *GaugeRedBlackGrid(void)      { return _FourDimRedBlackGrid ;}
-  GridBase *FermionGrid(void)            { return _FiveDimGrid;}
-  GridBase *FermionRedBlackGrid(void)    { return _FiveDimRedBlackGrid;}
-
-  // full checkerboard operations; leave unimplemented as abstract for now
-  RealD  M    (const FermionField &in, FermionField &out);
-  RealD  Mdag (const FermionField &in, FermionField &out);
-
-  // half checkerboard operations
-  void   Meooe       (const FermionField &in, FermionField &out);
-  void   Mooee       (const FermionField &in, FermionField &out);
-  void   MooeeInv    (const FermionField &in, FermionField &out);
-
-  void   MeooeDag    (const FermionField &in, FermionField &out);
-  void   MooeeDag    (const FermionField &in, FermionField &out);
-  void   MooeeInvDag (const FermionField &in, FermionField &out);
-
-  void   Mdir   (const FermionField &in, FermionField &out,int dir,int disp);
-  void DhopDir(const FermionField &in, FermionField &out,int dir,int disp);
-
-  // These can be overridden by fancy 5d chiral action
-  void DhopDeriv  (GaugeField &mat,const FermionField &U,const FermionField &V,int dag);
-  void DhopDerivEO(GaugeField &mat,const FermionField &U,const FermionField &V,int dag);
-  void DhopDerivOE(GaugeField &mat,const FermionField &U,const FermionField &V,int dag);
-
-  // Implement hopping term non-hermitian hopping term; half cb or both
-  void Dhop  (const FermionField &in, FermionField &out,int dag);
-  void DhopOE(const FermionField &in, FermionField &out,int dag);
-  void DhopEO(const FermionField &in, FermionField &out,int dag);
-
-    
-  ///////////////////////////////////////////////////////////////
-  // New methods added 
-  ///////////////////////////////////////////////////////////////
-  void DerivInternal(StencilImpl & st,
-		     DoubledGaugeField & U,
-		     DoubledGaugeField & UUU,
-		     GaugeField &mat,
-		     const FermionField &A,
-		     const FermionField &B,
-		     int dag);
-    
-  void DhopInternal(StencilImpl & st,
-		    LebesgueOrder &lo,
-		    DoubledGaugeField &U,
-		    DoubledGaugeField &UUU,
-		    const FermionField &in, 
-		    FermionField &out,
-		    int dag);
-    
-    void DhopInternalOverlappedComms(StencilImpl & st,
-		      LebesgueOrder &lo,
-		      DoubledGaugeField &U,
-		      DoubledGaugeField &UUU,
-		      const FermionField &in, 
-		      FermionField &out,
-		      int dag);
-
-    void DhopInternalSerialComms(StencilImpl & st,
-		      LebesgueOrder &lo,
-		      DoubledGaugeField &U,
-		      DoubledGaugeField &UUU,
-		      const FermionField &in, 
-		      FermionField &out,
-		      int dag);
-    
-    
-  // Constructors
-    ////////////////////////////////////////////////////////////////////////////////////////////////
-    // Grid internal interface -- Thin link and fat link, with coefficients
-    ////////////////////////////////////////////////////////////////////////////////////////////////
-  ImprovedStaggeredFermion5D(GaugeField &_Uthin,
-			     GaugeField &_Ufat,
-			     GridCartesian         &FiveDimGrid,
-			     GridRedBlackCartesian &FiveDimRedBlackGrid,
-			     GridCartesian         &FourDimGrid,
-			     GridRedBlackCartesian &FourDimRedBlackGrid,
-			     double _mass,
-			       RealD _c1, RealD _c2,RealD _u0,
-			       const ImplParams &p= ImplParams());
-    ////////////////////////////////////////////////////////////////////////////////////////////////
-    // MILC constructor ; triple links, no rescale factors; must be externally pre multiplied
-    ////////////////////////////////////////////////////////////////////////////////////////////////
-    ImprovedStaggeredFermion5D(GridCartesian         &FiveDimGrid,
-			       GridRedBlackCartesian &FiveDimRedBlackGrid,
-			       GridCartesian         &FourDimGrid,
-			       GridRedBlackCartesian &FourDimRedBlackGrid,
-			       double _mass,
-			       RealD _c1=1.0, RealD _c2=1.0,RealD _u0=1.0,
-			     const ImplParams &p= ImplParams());
-    
-    // DoubleStore gauge field in operator
-    void ImportGauge      (const GaugeField &_Uthin ) { assert(0); }
-  void ImportGauge(const GaugeField &_Uthin,const GaugeField &_Ufat);
-    void ImportGaugeSimple(const GaugeField &_UUU,const GaugeField &_U);
-    void ImportGaugeSimple(const DoubledGaugeField &_UUU,const DoubledGaugeField &_U);
-    // Give a reference; can be used to do an assignment or copy back out after import
-    // if Carleton wants to cache them and not use the ImportSimple
-    DoubledGaugeField &GetU(void)   { return Umu ; } ;
-    DoubledGaugeField &GetUUU(void) { return UUUmu; };
-    void CopyGaugeCheckerboards(void);
-    
-  ///////////////////////////////////////////////////////////////
-  // Data members require to support the functionality
-  ///////////////////////////////////////////////////////////////
-public:
-    
-    virtual int   isTrivialEE(void) { return 1; };
-    virtual RealD Mass(void) { return mass; }
-    
-  GridBase *_FourDimGrid;
-  GridBase *_FourDimRedBlackGrid;
-  GridBase *_FiveDimGrid;
-  GridBase *_FiveDimRedBlackGrid;
-    
-  RealD mass;
-  RealD c1;
-  RealD c2;
-  RealD u0;
-  int Ls;
-    
-  //Defines the stencils for even and odd
-  StencilImpl Stencil; 
-  StencilImpl StencilEven; 
-  StencilImpl StencilOdd; 
-    
-  // Copy of the gauge field , with even and odd subsets
-  DoubledGaugeField Umu;
-  DoubledGaugeField UmuEven;
-  DoubledGaugeField UmuOdd;
-
-  DoubledGaugeField UUUmu;
-  DoubledGaugeField UUUmuEven;
-  DoubledGaugeField UUUmuOdd;
-    
-  LebesgueOrder Lebesgue;
-  LebesgueOrder LebesgueEvenOdd;
-    
-  // Comms buffer
-  std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> >  comm_buf;
-    
-  ///////////////////////////////////////////////////////////////
-  // Conserved current utilities
-  ///////////////////////////////////////////////////////////////
-  void ContractConservedCurrent(PropagatorField &q_in_1,
-				PropagatorField &q_in_2,
-				PropagatorField &q_out,
-				Current curr_type,
-				unsigned int mu);
-  void SeqConservedCurrent(PropagatorField &q_in,
-			   PropagatorField &q_out,
-			   Current curr_type,
-			   unsigned int mu, 
-			   unsigned int tmin,
-                             unsigned int tmax,
-                 	     ComplexField &lattice_cmplx);
-};
-
-NAMESPACE_END(Grid);
-

Grid/qcd/action/fermion/MADWF.h

@@ -1,192 +0,0 @@
-    /*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/algorithms/iterative/MADWF.h
-
-    Copyright (C) 2015
-
-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 Fieldi, class Fieldo,IfNotSame<Fieldi,Fieldo> X=0>
-inline void convert(const Fieldi &from,Fieldo &to) 
-{
-  precisionChange(to,from);
-}
-template <class Fieldi, class Fieldo,IfSame<Fieldi,Fieldo> X=0>
-inline void convert(const Fieldi &from,Fieldo &to) 
-{
-  to=from;
-}
-
-template<class Matrixo,class Matrixi,class PVinverter,class SchurSolver, class Guesser> 
-class MADWF 
-{
- private:
-  typedef typename Matrixo::FermionField FermionFieldo;
-  typedef typename Matrixi::FermionField FermionFieldi;
-
-  PVinverter  & PauliVillarsSolvero;// For the outer field
-  SchurSolver & SchurSolveri;       // For the inner approx field
-  Guesser     & Guesseri;           // To deflate the inner approx solves
-
-  Matrixo & Mato;                   // Action object for outer
-  Matrixi & Mati;                   // Action object for inner
-
-  RealD target_resid;
-  int   maxiter;
- public:
-
-  MADWF(Matrixo &_Mato,
-	Matrixi &_Mati, 
-	PVinverter &_PauliVillarsSolvero, 
-	SchurSolver &_SchurSolveri,
-	Guesser & _Guesseri,
-	RealD resid,
-	int _maxiter) :
-
-  Mato(_Mato),Mati(_Mati),
-    SchurSolveri(_SchurSolveri),
-    PauliVillarsSolvero(_PauliVillarsSolvero),Guesseri(_Guesseri)
-  {   
-    target_resid=resid;
-    maxiter     =_maxiter; 
-  };
-
-  void operator() (const FermionFieldo &src4,FermionFieldo &sol5)
-  {
-    std::cout << GridLogMessage<< " ************************************************" << std::endl;
-    std::cout << GridLogMessage<< "  MADWF-like algorithm                           " << std::endl;
-    std::cout << GridLogMessage<< " ************************************************" << std::endl;
-
-    FermionFieldi    c0i(Mati.GaugeGrid()); // 4d 
-    FermionFieldi    y0i(Mati.GaugeGrid()); // 4d
-    FermionFieldo    c0 (Mato.GaugeGrid()); // 4d 
-    FermionFieldo    y0 (Mato.GaugeGrid()); // 4d
-
-    FermionFieldo    A(Mato.FermionGrid()); // Temporary outer
-    FermionFieldo    B(Mato.FermionGrid()); // Temporary outer
-    FermionFieldo    b(Mato.FermionGrid()); // 5d source
-
-    FermionFieldo    c(Mato.FermionGrid()); // PVinv source; reused so store
-    FermionFieldo    defect(Mato.FermionGrid()); // 5d source
-
-    FermionFieldi   ci(Mati.FermionGrid()); 
-    FermionFieldi   yi(Mati.FermionGrid()); 
-    FermionFieldi   xi(Mati.FermionGrid()); 
-    FermionFieldi srci(Mati.FermionGrid()); 
-    FermionFieldi   Ai(Mati.FermionGrid()); 
-
-    RealD m=Mati.Mass();
-
-    ///////////////////////////////////////
-    //Import source, include Dminus factors
-    ///////////////////////////////////////
-    Mato.ImportPhysicalFermionSource(src4,b); 
-    std::cout << GridLogMessage << " src4 " <<norm2(src4)<<std::endl;
-    std::cout << GridLogMessage << " b    " <<norm2(b)<<std::endl;
-
-    defect = b;
-    sol5=Zero();
-    for (int i=0;i<maxiter;i++) {
-
-      ///////////////////////////////////////
-      // Set up c0 from current defect
-      ///////////////////////////////////////
-      PauliVillarsSolvero(Mato,defect,A);
-      Mato.Pdag(A,c);
-      ExtractSlice(c0, c, 0 , 0);
-
-      ////////////////////////////////////////////////
-      // Solve the inner system with surface term c0
-      ////////////////////////////////////////////////
-      ci = Zero();  
-      convert(c0,c0i); // Possible precison change
-      InsertSlice(c0i,ci,0, 0);
-
-      // Dwm P y = Dwm x = D(1) P (c0,0,0,0)^T
-      Mati.P(ci,Ai);
-      Mati.SetMass(1.0);      Mati.M(Ai,srci);      Mati.SetMass(m);
-      SchurSolveri(Mati,srci,xi,Guesseri); 
-      Mati.Pdag(xi,yi);
-      ExtractSlice(y0i, yi, 0 , 0);
-      convert(y0i,y0); // Possible precision change
-
-      //////////////////////////////////////
-      // Propagate solution back to outer system
-      // Build Pdag PV^-1 Dm P [-sol4,c2,c3... cL]
-      //////////////////////////////////////
-      c0 = - y0;
-      InsertSlice(c0, c, 0   , 0);
-
-      /////////////////////////////
-      // Reconstruct the bulk solution Pdag PV^-1 Dm P 
-      /////////////////////////////
-      Mato.P(c,B);
-      Mato.M(B,A);
-      PauliVillarsSolvero(Mato,A,B);
-      Mato.Pdag(B,A);
-
-      //////////////////////////////
-      // Reinsert surface prop
-      //////////////////////////////
-      InsertSlice(y0,A,0,0);
-
-      //////////////////////////////
-      // Convert from y back to x 
-      //////////////////////////////
-      Mato.P(A,B);
-
-      //         sol5' = sol5 + M^-1 defect
-      //               = sol5 + M^-1 src - M^-1 M sol5  ...
-      sol5 = sol5 + B;
-      std::cout << GridLogMessage << "***************************************" <<std::endl;
-      std::cout << GridLogMessage << " Sol5 update "<<std::endl;
-      std::cout << GridLogMessage << "***************************************" <<std::endl;
-      std::cout << GridLogMessage << " Sol5 now "<<norm2(sol5)<<std::endl;
-      std::cout << GridLogMessage << " delta    "<<norm2(B)<<std::endl;
-
-       // New defect  = b - M sol5
-       Mato.M(sol5,A);
-       defect = b - A;
-
-       std::cout << GridLogMessage << " defect   "<<norm2(defect)<<std::endl;
-
-       double resid = ::sqrt(norm2(defect) / norm2(b));
-       std::cout << GridLogMessage << "Residual " << i << ": " << resid  << std::endl;
-       std::cout << GridLogMessage << "***************************************" <<std::endl;
-
-       if (resid < target_resid) {
-	 return;
-       }
-    }
-
-    std::cout << GridLogMessage << "MADWF : Exceeded maxiter "<<std::endl;
-    assert(0);
-
-  }
-
-};
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/MobiusEOFAFermion.h

@@ -1,104 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/MobiusEOFAFermion.h
-
-Copyright (C) 2017
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: David Murphy <dmurphy@phys.columbia.edu>
-
-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_QCD_MOBIUS_EOFA_FERMION_H
-#define  GRID_QCD_MOBIUS_EOFA_FERMION_H
-
-#include <Grid/qcd/action/fermion/AbstractEOFAFermion.h>
-
-NAMESPACE_BEGIN(Grid);
-
-template<class Impl>
-class MobiusEOFAFermion : public AbstractEOFAFermion<Impl>
-{
-public:
-  INHERIT_IMPL_TYPES(Impl);
-
-public:
-  // Shift operator coefficients for red-black preconditioned Mobius EOFA
-  Vector<Coeff_t> Mooee_shift;
-  Vector<Coeff_t> MooeeInv_shift_lc;
-  Vector<Coeff_t> MooeeInv_shift_norm;
-  Vector<Coeff_t> MooeeInvDag_shift_lc;
-  Vector<Coeff_t> MooeeInvDag_shift_norm;
-
-  virtual void Instantiatable(void) {};
-
-  // EOFA-specific operations
-  virtual void  Omega            (const FermionField& in, FermionField& out, int sign, int dag);
-  virtual void  Dtilde           (const FermionField& in, FermionField& out);
-  virtual void  DtildeInv        (const FermionField& in, FermionField& out);
-
-  // override multiply
-  virtual RealD M                (const FermionField& in, FermionField& out);
-  virtual RealD Mdag             (const FermionField& in, FermionField& out);
-
-  // half checkerboard operations
-  virtual void  Mooee            (const FermionField& in, FermionField& out);
-  virtual void  MooeeDag         (const FermionField& in, FermionField& out);
-  virtual void  MooeeInv         (const FermionField& in, FermionField& out);
-  virtual void  MooeeInv_shift   (const FermionField& in, FermionField& out);
-  virtual void  MooeeInvDag      (const FermionField& in, FermionField& out);
-  virtual void  MooeeInvDag_shift(const FermionField& in, FermionField& out);
-
-  virtual void   M5D             (const FermionField& psi, FermionField& chi);
-  virtual void   M5Ddag          (const FermionField& psi, FermionField& chi);
-
-  /////////////////////////////////////////////////////
-  // Instantiate different versions depending on Impl
-  /////////////////////////////////////////////////////
-  void M5D(const FermionField& psi, const FermionField& phi, FermionField& chi,
-	   Vector<Coeff_t>& lower, Vector<Coeff_t>& diag, Vector<Coeff_t>& upper);
-
-  void M5D_shift(const FermionField& psi, const FermionField& phi, FermionField& chi,
-		 Vector<Coeff_t>& lower, Vector<Coeff_t>& diag, Vector<Coeff_t>& upper,
-		 Vector<Coeff_t>& shift_coeffs);
-
-  void M5Ddag(const FermionField& psi, const FermionField& phi, FermionField& chi,
-	      Vector<Coeff_t>& lower, Vector<Coeff_t>& diag, Vector<Coeff_t>& upper);
-
-  void M5Ddag_shift(const FermionField& psi, const FermionField& phi, FermionField& chi,
-		    Vector<Coeff_t>& lower, Vector<Coeff_t>& diag, Vector<Coeff_t>& upper,
-		    Vector<Coeff_t>& shift_coeffs);
-
-  virtual void RefreshShiftCoefficients(RealD new_shift);
-
-  // Constructors
-  MobiusEOFAFermion(GaugeField& _Umu, GridCartesian& FiveDimGrid, GridRedBlackCartesian& FiveDimRedBlackGrid,
-		    GridCartesian& FourDimGrid, GridRedBlackCartesian& FourDimRedBlackGrid,
-		    RealD _mq1, RealD _mq2, RealD _mq3, RealD _shift, int pm,
-		    RealD _M5, RealD _b, RealD _c, const ImplParams& p=ImplParams());
-
-protected:
-  void SetCoefficientsPrecondShiftOps(void);
-};
-
-NAMESPACE_END(Grid);
-
-#endif

Grid/qcd/action/fermion/MobiusFermion.h

@@ -1,77 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/MobiusFermion.h
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-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 */
-#ifndef  GRID_QCD_MOBIUS_FERMION_H
-#define  GRID_QCD_MOBIUS_FERMION_H
-
-#include <Grid/qcd/action/fermion/FermionCore.h>
-
-NAMESPACE_BEGIN(Grid);
-
-template<class Impl>
-class MobiusFermion : public CayleyFermion5D<Impl>
-{
-public:
-  INHERIT_IMPL_TYPES(Impl);
-public:
-
-  virtual void   Instantiatable(void) {};
-  // Constructors
-  MobiusFermion(GaugeField &_Umu,
-		GridCartesian         &FiveDimGrid,
-		GridRedBlackCartesian &FiveDimRedBlackGrid,
-		GridCartesian         &FourDimGrid,
-		GridRedBlackCartesian &FourDimRedBlackGrid,
-		RealD _mass,RealD _M5,
-		RealD b, RealD c,const ImplParams &p= ImplParams()) : 
-      
-    CayleyFermion5D<Impl>(_Umu,
-			  FiveDimGrid,
-			  FiveDimRedBlackGrid,
-			  FourDimGrid,
-			  FourDimRedBlackGrid,_mass,_M5,p)
-
-  {
-    RealD eps = 1.0;
-
-    std::cout<<GridLogMessage << "MobiusFermion (b="<<b<<",c="<<c<<") with Ls= "<<this->Ls<<" Tanh approx"<<std::endl;
-    Approx::zolotarev_data *zdata = Approx::higham(eps,this->Ls);// eps is ignored for higham
-    assert(zdata->n==this->Ls);
-	
-    // Call base setter
-    this->SetCoefficientsTanh(zdata,b,c);
-
-    Approx::zolotarev_free(zdata);
- 
-  }
-
-};
-
-NAMESPACE_END(Grid);
-
-#endif

Grid/qcd/action/fermion/MobiusZolotarevFermion.h

@@ -1,78 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/MobiusZolotarevFermion.h
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-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 */
-#ifndef  GRID_QCD_MOBIUS_ZOLOTAREV_FERMION_H
-#define  GRID_QCD_MOBIUS_ZOLOTAREV_FERMION_H
-
-#include <Grid/qcd/action/fermion/FermionCore.h>
-
-NAMESPACE_BEGIN(Grid);
-
-template<class Impl>
-class MobiusZolotarevFermion : public CayleyFermion5D<Impl>
-{
-public:
-  INHERIT_IMPL_TYPES(Impl);
-public:
-
-  virtual void   Instantiatable(void) {};
-  // Constructors
-  MobiusZolotarevFermion(GaugeField &_Umu,
-			 GridCartesian         &FiveDimGrid,
-			 GridRedBlackCartesian &FiveDimRedBlackGrid,
-			 GridCartesian         &FourDimGrid,
-			 GridRedBlackCartesian &FourDimRedBlackGrid,
-			 RealD _mass,RealD _M5,
-			 RealD b, RealD c,
-			 RealD lo, RealD hi,const ImplParams &p= ImplParams()) : 
-      
-    CayleyFermion5D<Impl>(_Umu,
-			  FiveDimGrid,
-			  FiveDimRedBlackGrid,
-			  FourDimGrid,
-			  FourDimRedBlackGrid,_mass,_M5,p)
-
-  {
-    RealD eps = lo/hi;
-
-    Approx::zolotarev_data *zdata = Approx::zolotarev(eps,this->Ls,0);
-    assert(zdata->n==this->Ls);
-
-    std::cout<<GridLogMessage << "MobiusZolotarevFermion (b="<<b<<",c="<<c<<") with Ls= "<<this->Ls<<" Zolotarev range ["<<lo<<","<<hi<<"]"<<std::endl;
-	
-    // Call base setter
-    this->SetCoefficientsZolotarev(hi,zdata,b,c);
- 
-    Approx::zolotarev_free(zdata);
-  }
-
-};
-
-NAMESPACE_END(Grid);
-
-#endif

Grid/qcd/action/fermion/OverlapWilsonContfracTanhFermion.h

@@ -1,70 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/OverlapWilsonContfracTanhFermion.h
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-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 */
-#ifndef OVERLAP_WILSON_CONTFRAC_TANH_FERMION_H
-#define OVERLAP_WILSON_CONTFRAC_TANH_FERMION_H
-
-#include <Grid/qcd/action/fermion/FermionCore.h>
-
-NAMESPACE_BEGIN(Grid);
-
-template<class Impl>
-class OverlapWilsonContFracTanhFermion : public ContinuedFractionFermion5D<Impl>
-{
-public:
-  INHERIT_IMPL_TYPES(Impl);
-public:
-
-  virtual void   Instantiatable(void){};
-  // Constructors
-  OverlapWilsonContFracTanhFermion(GaugeField &_Umu,
-				   GridCartesian         &FiveDimGrid,
-				   GridRedBlackCartesian &FiveDimRedBlackGrid,
-				   GridCartesian         &FourDimGrid,
-				   GridRedBlackCartesian &FourDimRedBlackGrid,
-				   RealD _mass,RealD _M5,
-				   RealD scale,const ImplParams &p= ImplParams()) :
-      
-    // b+c=scale, b-c = 0 <=> b =c = scale/2
-    ContinuedFractionFermion5D<Impl>(_Umu,
-				     FiveDimGrid,
-				     FiveDimRedBlackGrid,
-				     FourDimGrid,
-				     FourDimRedBlackGrid,_mass,_M5,p)
-  {
-    assert((this->Ls&0x1)==1); // Odd Ls required
-    int nrational=this->Ls-1;// Even rational order
-    Approx::zolotarev_data *zdata = Approx::higham(1.0,nrational);// eps is ignored for higham
-    this->SetCoefficientsTanh(zdata,scale);
-    Approx::zolotarev_free(zdata);
-  }
-};
-
-NAMESPACE_END(Grid);
-
-#endif

Grid/qcd/action/fermion/OverlapWilsonContfracZolotarevFermion.h

@@ -1,72 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/OverlapWilsonContfracZolotarevFermion.h
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-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 */
-#ifndef OVERLAP_WILSON_CONTFRAC_ZOLOTAREV_FERMION_H
-#define OVERLAP_WILSON_CONTFRAC_ZOLOTAREV_FERMION_H
-
-#include <Grid/qcd/action/fermion/FermionCore.h>
-
-NAMESPACE_BEGIN(Grid);
-
-template<class Impl>
-class OverlapWilsonContFracZolotarevFermion : public ContinuedFractionFermion5D<Impl>
-{
-public:
-  INHERIT_IMPL_TYPES(Impl);
-
-  virtual void   Instantiatable(void){};
-  // Constructors
-  OverlapWilsonContFracZolotarevFermion(GaugeField &_Umu,
-					GridCartesian         &FiveDimGrid,
-					GridRedBlackCartesian &FiveDimRedBlackGrid,
-					GridCartesian         &FourDimGrid,
-					GridRedBlackCartesian &FourDimRedBlackGrid,
-					RealD _mass,RealD _M5,
-					RealD lo,RealD hi,const ImplParams &p= ImplParams()):
-      
-    // b+c=scale, b-c = 0 <=> b =c = scale/2
-    ContinuedFractionFermion5D<Impl>(_Umu,
-				     FiveDimGrid,
-				     FiveDimRedBlackGrid,
-				     FourDimGrid,
-				     FourDimRedBlackGrid,_mass,_M5,p)
-  {
-    assert((this->Ls&0x1)==1); // Odd Ls required
-
-    int nrational=this->Ls;// Odd rational order
-    RealD eps = lo/hi;
-
-    Approx::zolotarev_data *zdata = Approx::zolotarev(eps,nrational,0);
-    this->SetCoefficientsZolotarev(hi,zdata);
-    Approx::zolotarev_free(zdata);
-
-  }
-};
-
-NAMESPACE_END(Grid);
-#endif

Grid/qcd/action/fermion/OverlapWilsonPartialFractionTanhFermion.h

@@ -1,69 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/OverlapWilsonPartialFractionTanhFermion.h
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-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 */
-#ifndef OVERLAP_WILSON_PARTFRAC_TANH_FERMION_H
-#define OVERLAP_WILSON_PARTFRAC_TANH_FERMION_H
-
-#include <Grid/qcd/action/fermion/FermionCore.h>
-
-NAMESPACE_BEGIN(Grid);
-
-template<class Impl>
-class OverlapWilsonPartialFractionTanhFermion : public PartialFractionFermion5D<Impl>
-{
-public:
-  INHERIT_IMPL_TYPES(Impl);
-public:
-
-  virtual void   Instantiatable(void){};
-  // Constructors
-  OverlapWilsonPartialFractionTanhFermion(GaugeField &_Umu,
-					  GridCartesian         &FiveDimGrid,
-					  GridRedBlackCartesian &FiveDimRedBlackGrid,
-					  GridCartesian         &FourDimGrid,
-					  GridRedBlackCartesian &FourDimRedBlackGrid,
-					  RealD _mass,RealD _M5,
-					  RealD scale,const ImplParams &p= ImplParams()) :
-      
-    // b+c=scale, b-c = 0 <=> b =c = scale/2
-    PartialFractionFermion5D<Impl>(_Umu,
-				   FiveDimGrid,
-				   FiveDimRedBlackGrid,
-				   FourDimGrid,
-				   FourDimRedBlackGrid,_mass,_M5,p)
-  {
-    assert((this->Ls&0x1)==1); // Odd Ls required
-    int nrational=this->Ls-1;// Even rational order
-    Approx::zolotarev_data *zdata = Approx::higham(1.0,nrational);// eps is ignored for higham
-    this->SetCoefficientsTanh(zdata,scale);
-    Approx::zolotarev_free(zdata);
-  }
-};
-
-NAMESPACE_END(Grid);
-#endif

Grid/qcd/action/fermion/OverlapWilsonPartialFractionZolotarevFermion.h

@@ -1,73 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/OverlapWilsonPartialFractionZolotarevFermion.h
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-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 */
-#ifndef OVERLAP_WILSON_PARTFRAC_ZOLOTAREV_FERMION_H
-#define OVERLAP_WILSON_PARTFRAC_ZOLOTAREV_FERMION_H
-
-#include <Grid/qcd/action/fermion/FermionCore.h>
-
-NAMESPACE_BEGIN(Grid);
-
-template<class Impl>
-class OverlapWilsonPartialFractionZolotarevFermion : public PartialFractionFermion5D<Impl>
-{
-public:
-  INHERIT_IMPL_TYPES(Impl);
-
-  virtual void   Instantiatable(void){};
-  // Constructors
-  OverlapWilsonPartialFractionZolotarevFermion(GaugeField &_Umu,
-					       GridCartesian         &FiveDimGrid,
-					       GridRedBlackCartesian &FiveDimRedBlackGrid,
-					       GridCartesian         &FourDimGrid,
-					       GridRedBlackCartesian &FourDimRedBlackGrid,
-					       RealD _mass,RealD _M5,
-					       RealD lo,RealD hi,const ImplParams &p= ImplParams()):
-      
-    // b+c=scale, b-c = 0 <=> b =c = scale/2
-    PartialFractionFermion5D<Impl>(_Umu,
-				   FiveDimGrid,
-				   FiveDimRedBlackGrid,
-				   FourDimGrid,
-				   FourDimRedBlackGrid,_mass,_M5,p)
-  {
-    assert((this->Ls&0x1)==1); // Odd Ls required
-
-    int nrational=this->Ls;// Odd rational order
-    RealD eps = lo/hi;
-
-    Approx::zolotarev_data *zdata = Approx::zolotarev(eps,nrational,0);
-    this->SetCoefficientsZolotarev(hi,zdata);
-    Approx::zolotarev_free(zdata);
-
-  }
-};
-
-NAMESPACE_END(Grid);
-
-#endif

Grid/qcd/action/fermion/PartialFractionFermion5D.h

@@ -1,103 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/PartialFractionFermion5D.h
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-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 */
-#ifndef  GRID_QCD_PARTIAL_FRACTION_H
-#define  GRID_QCD_PARTIAL_FRACTION_H
-
-#include <Grid/qcd/action/fermion/WilsonFermion5D.h>
-
-NAMESPACE_BEGIN(Grid);
-
-template<class Impl>
-class PartialFractionFermion5D : public WilsonFermion5D<Impl>
-{
-public:
-  INHERIT_IMPL_TYPES(Impl);
-
-  const int part_frac_chroma_convention=1;
-
-  void   Meooe_internal(const FermionField &in, FermionField &out,int dag);
-  void   Mooee_internal(const FermionField &in, FermionField &out,int dag);
-  void   MooeeInv_internal(const FermionField &in, FermionField &out,int dag);
-  void   M_internal(const FermionField &in, FermionField &out,int dag);
-
-  // override multiply
-  virtual RealD  M    (const FermionField &in, FermionField &out);
-  virtual RealD  Mdag (const FermionField &in, FermionField &out);
-
-  // half checkerboard operaions
-  virtual void   Meooe       (const FermionField &in, FermionField &out);
-  virtual void   MeooeDag    (const FermionField &in, FermionField &out);
-  virtual void   Mooee       (const FermionField &in, FermionField &out);
-  virtual void   MooeeDag    (const FermionField &in, FermionField &out);
-  virtual void   MooeeInv    (const FermionField &in, FermionField &out);
-  virtual void   MooeeInvDag (const FermionField &in, FermionField &out);
-
-  // force terms; five routines; default to Dhop on diagonal
-  virtual void MDeriv  (GaugeField &mat,const FermionField &U,const FermionField &V,int dag);
-  virtual void MoeDeriv(GaugeField &mat,const FermionField &U,const FermionField &V,int dag);
-  virtual void MeoDeriv(GaugeField &mat,const FermionField &U,const FermionField &V,int dag);
-
-  virtual void   Instantiatable(void) =0; // ensure no make-eee
-
-  // Efficient support for multigrid coarsening
-  virtual void  Mdir (const FermionField &in, FermionField &out,int dir,int disp);
-
-      ///////////////////////////////////////////////////////////////
-      // Physical surface field utilities
-      ///////////////////////////////////////////////////////////////
-      virtual void ExportPhysicalFermionSolution(const FermionField &solution5d,FermionField &exported4d);
-      virtual void ImportPhysicalFermionSource  (const FermionField &input4d,FermionField &imported5d);
-
-  // Constructors
-  PartialFractionFermion5D(GaugeField &_Umu,
-			   GridCartesian         &FiveDimGrid,
-			   GridRedBlackCartesian &FiveDimRedBlackGrid,
-			   GridCartesian         &FourDimGrid,
-			   GridRedBlackCartesian &FourDimRedBlackGrid,
-			   RealD _mass,RealD M5,const ImplParams &p= ImplParams());
-
-protected:
-
-  virtual void SetCoefficientsTanh(Approx::zolotarev_data *zdata,RealD scale);
-  virtual void SetCoefficientsZolotarev(RealD zolo_hi,Approx::zolotarev_data *zdata);
-
-  // Part frac
-  RealD mass;
-  RealD dw_diag;
-  RealD R;
-  RealD amax;
-  RealD scale;
-  Vector<double> p; 
-  Vector<double> q;
-
-};
-
-NAMESPACE_END(Grid);
-
-#endif

Grid/qcd/action/fermion/PauliVillarsInverters.h

@@ -1,92 +0,0 @@
-    /*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/algorithms/iterative/SchurRedBlack.h
-
-    Copyright (C) 2015
-
-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 Field>
-class PauliVillarsSolverUnprec
-{
- public:
-  ConjugateGradient<Field> & CG;
-  PauliVillarsSolverUnprec(  ConjugateGradient<Field> &_CG) : CG(_CG){};
-
-  template<class Matrix>
-  void operator() (Matrix &_Matrix,const Field &src,Field &sol)
-  {
-    RealD m = _Matrix.Mass();
-    Field A  (_Matrix.FermionGrid());
-
-    MdagMLinearOperator<Matrix,Field> HermOp(_Matrix);
-
-    _Matrix.SetMass(1.0);
-    _Matrix.Mdag(src,A);
-    CG(HermOp,A,sol);
-    _Matrix.SetMass(m);
-  };
-};
-
-template<class Field,class SchurSolverType>
-class PauliVillarsSolverRBprec
-{
- public:
-  SchurSolverType & SchurSolver;
-  PauliVillarsSolverRBprec( SchurSolverType &_SchurSolver) : SchurSolver(_SchurSolver){};
-
-  template<class Matrix>
-  void operator() (Matrix &_Matrix,const Field &src,Field &sol)
-  {
-    RealD m = _Matrix.Mass();
-    Field A  (_Matrix.FermionGrid());
-
-    _Matrix.SetMass(1.0);
-    SchurSolver(_Matrix,src,sol);
-    _Matrix.SetMass(m);
-  };
-};
-
-template<class Field,class GaugeField>
-class PauliVillarsSolverFourierAccel
-{
- public:
-  GaugeField      & Umu;
-  ConjugateGradient<Field> & CG;
-
-  PauliVillarsSolverFourierAccel(GaugeField &_Umu,ConjugateGradient<Field> &_CG) :  Umu(_Umu), CG(_CG)
-  {
-  };
-
-  template<class Matrix>
-  void operator() (Matrix &_Matrix,const Field &src,Field &sol)
-  {
-    FourierAcceleratedPV<Field, Matrix, typename Matrix::GaugeField > faPV(_Matrix,Umu,CG) ;
-    faPV.pvInv(src,sol);
-  };
-};
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/Reconstruct5Dprop.h

@@ -1,134 +0,0 @@
-    /*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/algorithms/iterative/SchurRedBlack.h
-
-    Copyright (C) 2015
-
-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 Field,class PVinverter> class Reconstruct5DfromPhysical {
- private:
-  PVinverter & PauliVillarsSolver;
- public:
-
- /////////////////////////////////////////////////////
- // First cut works, 10 Oct 2018.
- //
- // Must form a plan to get this into production for Zmobius acceleration
- // of the Mobius exact AMA corrections.
- //
- // TODO : understand absence of contact term in eqns in Hantao's thesis
- //        sol4 is contact term subtracted, but thesis & Brower's paper suggests not.
- //
- // Step 1: Localise PV inverse in a routine. [DONE]
- // Step 2: Schur based PV inverse            [DONE]
- // Step 3: Fourier accelerated PV inverse    [DONE]
- //
- /////////////////////////////////////////////////////
- 
-  Reconstruct5DfromPhysical(PVinverter &_PauliVillarsSolver) 
-    : PauliVillarsSolver(_PauliVillarsSolver) 
-  { 
-  };
-
-
-   template<class Matrix>
-   void PV(Matrix &_Matrix,const Field &src,Field &sol)
-   {
-     RealD m = _Matrix.Mass();
-     _Matrix.SetMass(1.0);
-     _Matrix.M(src,sol);
-     _Matrix.SetMass(m);
-   }
-   template<class Matrix>
-   void PVdag(Matrix &_Matrix,const Field &src,Field &sol)
-   {
-     RealD m = _Matrix.Mass();
-     _Matrix.SetMass(1.0);
-     _Matrix.Mdag(src,sol);
-     _Matrix.SetMass(m);
-   }
-  template<class Matrix>
-  void operator() (Matrix & _Matrix,const Field &sol4,const Field &src4, Field &sol5){
-
-    int Ls =  _Matrix.Ls;
-
-    Field psi4(_Matrix.GaugeGrid());
-    Field psi(_Matrix.FermionGrid());
-    Field A  (_Matrix.FermionGrid());
-    Field B  (_Matrix.FermionGrid());
-    Field c  (_Matrix.FermionGrid());
-
-    typedef typename Matrix::Coeff_t Coeff_t;
-
-    std::cout << GridLogMessage<< " ************************************************" << std::endl;
-    std::cout << GridLogMessage<< " Reconstruct5Dprop: c.f. MADWF algorithm         " << std::endl;
-    std::cout << GridLogMessage<< " ************************************************" << std::endl;
-
-    ///////////////////////////////////////
-    //Import source, include Dminus factors
-    ///////////////////////////////////////
-    _Matrix.ImportPhysicalFermionSource(src4,B); 
-
-    ///////////////////////////////////////
-    // Set up c from src4
-    ///////////////////////////////////////
-    PauliVillarsSolver(_Matrix,B,A);
-    _Matrix.Pdag(A,c);
-
-    //////////////////////////////////////
-    // Build Pdag PV^-1 Dm P [-sol4,c2,c3... cL]
-    //////////////////////////////////////
-    psi4 = - sol4;
-    InsertSlice(psi4, psi, 0   , 0);
-    for (int s=1;s<Ls;s++) {
-      ExtractSlice(psi4,c,s,0);
-       InsertSlice(psi4,psi,s,0);
-    }
-
-    /////////////////////////////
-    // Pdag PV^-1 Dm P 
-    /////////////////////////////
-    _Matrix.P(psi,B);
-    _Matrix.M(B,A);
-    PauliVillarsSolver(_Matrix,A,B);
-    _Matrix.Pdag(B,A);
-
-    //////////////////////////////
-    // Reinsert surface prop
-    //////////////////////////////
-    InsertSlice(sol4,A,0,0);
-
-    //////////////////////////////
-    // Convert from y back to x 
-    //////////////////////////////
-    _Matrix.P(A,sol5);
-    
-  }
-};
-
-NAMESPACE_END(Grid);
-

Grid/qcd/action/fermion/SchurDiagTwoKappa.h

@@ -1,102 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: SchurDiagTwoKappa.h
-
-    Copyright (C) 2017
-
-Author: Christoph Lehner
-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);
-
-// This is specific to (Z)mobius fermions
-template<class Matrix, class Field>
-class KappaSimilarityTransform {
-public:
-  INHERIT_IMPL_TYPES(Matrix);
-  Vector<Coeff_t> kappa, kappaDag, kappaInv, kappaInvDag;
-
-  KappaSimilarityTransform (Matrix &zmob) {
-    for (int i=0;i<(int)zmob.bs.size();i++) {
-      Coeff_t k = 1.0 / ( 2.0 * (zmob.bs[i] *(4 - zmob.M5) + 1.0) );
-      kappa.push_back( k );
-      kappaDag.push_back( conj(k) );
-      kappaInv.push_back( 1.0 / k );
-      kappaInvDag.push_back( 1.0 / conj(k) );
-    }
-  }
-
-  template<typename vobj>
-  void sscale(const Lattice<vobj>& in, Lattice<vobj>& out, Coeff_t* s) {
-    GridBase *grid=out.Grid();
-    out.Checkerboard() = in.Checkerboard();
-    assert(grid->_simd_layout[0] == 1); // should be fine for ZMobius for now
-    int Ls = grid->_rdimensions[0];
-    thread_for(ss, grid->oSites(),
-    {
-      vobj tmp = s[ss % Ls]*in[ss];
-      vstream(out[ss],tmp);
-    });
-  }
-
-  RealD sscale_norm(const Field& in, Field& out, Coeff_t* s) {
-    sscale(in,out,s);
-    return norm2(out);
-  }
-
-  virtual RealD M       (const Field& in, Field& out) { return sscale_norm(in,out,&kappa[0]);   }
-  virtual RealD MDag    (const Field& in, Field& out) { return sscale_norm(in,out,&kappaDag[0]);}
-  virtual RealD MInv    (const Field& in, Field& out) { return sscale_norm(in,out,&kappaInv[0]);}
-  virtual RealD MInvDag (const Field& in, Field& out) { return sscale_norm(in,out,&kappaInvDag[0]);}
-
-};
-
-template<class Matrix,class Field>
-class SchurDiagTwoKappaOperator :  public SchurOperatorBase<Field> {
-public:
-  KappaSimilarityTransform<Matrix, Field> _S;
-  SchurDiagTwoOperator<Matrix, Field> _Mat;
-
-  SchurDiagTwoKappaOperator (Matrix &Mat): _S(Mat), _Mat(Mat) {};
-
-  virtual  RealD Mpc      (const Field &in, Field &out) {
-    Field tmp(in.Grid());
-
-    _S.MInv(in,out);
-    _Mat.Mpc(out,tmp);
-    return _S.M(tmp,out);
-
-  }
-  virtual  RealD MpcDag   (const Field &in, Field &out){
-    Field tmp(in.Grid());
-
-    _S.MDag(in,out);
-    _Mat.MpcDag(out,tmp);
-    return _S.MInvDag(tmp,out);
-  }
-};
-
-NAMESPACE_END(Grid);
-
-

Grid/qcd/action/fermion/StaggeredImpl.h

@@ -1,175 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/FermionOperatorImpl.h
-
-Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@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 S, class Representation = FundamentalRepresentation >
-class StaggeredImpl : public PeriodicGaugeImpl<GaugeImplTypes<S, Representation::Dimension > > 
-{
-
-public:
-
-  typedef RealD  _Coeff_t ;
-  static const int Dimension = Representation::Dimension;
-  static const bool isFundamental = Representation::isFundamental;
-  static const bool LsVectorised=false;
-  typedef PeriodicGaugeImpl<GaugeImplTypes<S, Dimension > > Gimpl;
-      
-  //Necessary?
-  constexpr bool is_fundamental() const{return Dimension == Nc ? 1 : 0;}
-    
-  typedef _Coeff_t Coeff_t;
-
-  INHERIT_GIMPL_TYPES(Gimpl);
-      
-  template <typename vtype> using iImplSpinor            = iScalar<iScalar<iVector<vtype, Dimension> > >;
-  template <typename vtype> using iImplHalfSpinor        = iScalar<iScalar<iVector<vtype, Dimension> > >;
-  template <typename vtype> using iImplDoubledGaugeField = iVector<iScalar<iMatrix<vtype, Dimension> >, Nds>;
-  template <typename vtype> using iImplPropagator        = iScalar<iScalar<iMatrix<vtype, Dimension> > >;
-    
-  typedef iImplSpinor<Simd>            SiteSpinor;
-  typedef iImplHalfSpinor<Simd>        SiteHalfSpinor;
-  typedef iImplDoubledGaugeField<Simd> SiteDoubledGaugeField;
-  typedef iImplPropagator<Simd>        SitePropagator;
-    
-  typedef Lattice<SiteSpinor>            FermionField;
-  typedef Lattice<SiteDoubledGaugeField> DoubledGaugeField;
-  typedef Lattice<SitePropagator> PropagatorField;
-    
-  typedef StaggeredImplParams ImplParams;
-  typedef SimpleCompressor<SiteSpinor> Compressor;
-  typedef CartesianStencil<SiteSpinor, SiteSpinor, ImplParams> StencilImpl;
-  typedef typename StencilImpl::View_type StencilView;
-
-  ImplParams Params;
-    
-  StaggeredImpl(const ImplParams &p = ImplParams()) : Params(p){};
-      
-  static accelerator_inline void multLink(SiteSpinor &phi,
-		       const SiteDoubledGaugeField &U,
-		       const SiteSpinor &chi,
-		       int mu)
-  {
-    mult(&phi(), &U(mu), &chi());
-  }
-  static accelerator_inline void multLinkAdd(SiteSpinor &phi,
-			  const SiteDoubledGaugeField &U,
-			  const SiteSpinor &chi,
-			  int mu)
-  {
-    mac(&phi(), &U(mu), &chi());
-  }
-      
-  template <class ref>
-  static accelerator_inline void loadLinkElement(Simd &reg, ref &memory) 
-  {
-    reg = memory;
-  }
-      
-    inline void InsertGaugeField(DoubledGaugeField &U_ds,
-				 const GaugeLinkField &U,int mu)
-    {
-      PokeIndex<LorentzIndex>(U_ds, U, mu);
-    }
-  inline void DoubleStore(GridBase *GaugeGrid,
-			  DoubledGaugeField &UUUds, // for Naik term
-			  DoubledGaugeField &Uds,
-			  const GaugeField &Uthin,
-			  const GaugeField &Ufat) {
-    conformable(Uds.Grid(), GaugeGrid);
-    conformable(Uthin.Grid(), GaugeGrid);
-    conformable(Ufat.Grid(), GaugeGrid);
-    GaugeLinkField U(GaugeGrid);
-    GaugeLinkField UU(GaugeGrid);
-    GaugeLinkField UUU(GaugeGrid);
-    GaugeLinkField Udag(GaugeGrid);
-    GaugeLinkField UUUdag(GaugeGrid);
-    for (int mu = 0; mu < Nd; mu++) {
-
-      // Staggered Phase.
-      Lattice<iScalar<vInteger> > coor(GaugeGrid);
-      Lattice<iScalar<vInteger> > x(GaugeGrid); LatticeCoordinate(x,0);
-      Lattice<iScalar<vInteger> > y(GaugeGrid); LatticeCoordinate(y,1);
-      Lattice<iScalar<vInteger> > z(GaugeGrid); LatticeCoordinate(z,2);
-      Lattice<iScalar<vInteger> > t(GaugeGrid); LatticeCoordinate(t,3);
-
-      Lattice<iScalar<vInteger> > lin_z(GaugeGrid); lin_z=x+y;
-      Lattice<iScalar<vInteger> > lin_t(GaugeGrid); lin_t=x+y+z;
-
-      ComplexField phases(GaugeGrid);	phases=1.0;
-
-      if ( mu == 1 ) phases = where( mod(x    ,2)==(Integer)0, phases,-phases);
-      if ( mu == 2 ) phases = where( mod(lin_z,2)==(Integer)0, phases,-phases);
-      if ( mu == 3 ) phases = where( mod(lin_t,2)==(Integer)0, phases,-phases);
-
-      // 1 hop based on fat links
-      U      = PeekIndex<LorentzIndex>(Ufat, mu);
-      Udag   = adj( Cshift(U, mu, -1));
-
-      U    = U    *phases;
-      Udag = Udag *phases;
-
-	InsertGaugeField(Uds,U,mu);
-	InsertGaugeField(Uds,Udag,mu+4);
-	//	PokeIndex<LorentzIndex>(Uds, U, mu);
-	//	PokeIndex<LorentzIndex>(Uds, Udag, mu + 4);
-
-      // 3 hop based on thin links. Crazy huh ?
-      U  = PeekIndex<LorentzIndex>(Uthin, mu);
-      UU = Gimpl::CovShiftForward(U,mu,U);
-      UUU= Gimpl::CovShiftForward(U,mu,UU);
-	
-      UUUdag = adj( Cshift(UUU, mu, -3));
-
-      UUU    = UUU    *phases;
-      UUUdag = UUUdag *phases;
-
-	InsertGaugeField(UUUds,UUU,mu);
-	InsertGaugeField(UUUds,UUUdag,mu+4);
-
-    }
-  }
-
-  inline void InsertForce4D(GaugeField &mat, FermionField &Btilde, FermionField &A,int mu){
-    GaugeLinkField link(mat.Grid());
-    link = TraceIndex<SpinIndex>(outerProduct(Btilde,A)); 
-    PokeIndex<LorentzIndex>(mat,link,mu);
-  }   
-      
-  inline void InsertForce5D(GaugeField &mat, FermionField &Btilde, FermionField &Atilde,int mu){
-    assert (0); 
-    // Must never hit
-  }
-};
-typedef StaggeredImpl<vComplex,  FundamentalRepresentation > StaggeredImplR;   // Real.. whichever prec
-typedef StaggeredImpl<vComplexF, FundamentalRepresentation > StaggeredImplF;  // Float
-typedef StaggeredImpl<vComplexD, FundamentalRepresentation > StaggeredImplD;  // Double
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/StaggeredKernels.h

@@ -1,117 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/StaggeredKernels.h
-
-Copyright (C) 2015
-
-Author: Azusa Yamaguchi, Peter Boyle
-
-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)
-
-  ////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-  // Helper routines that implement Staggered stencil for a single site.
-  ////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-class StaggeredKernelsStatic { 
- public:
-  enum { OptGeneric, OptHandUnroll, OptInlineAsm };
-  enum { CommsAndCompute, CommsThenCompute };
-  static int Opt;
-  static int Comms;
-};
- 
-template<class Impl> class StaggeredKernels : public FermionOperator<Impl> , public StaggeredKernelsStatic { 
- public:
-   
-  INHERIT_IMPL_TYPES(Impl);
-  typedef FermionOperator<Impl> Base;
-   
-public:
-    
-   void DhopDirKernel(StencilImpl &st, DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU, SiteSpinor * buf,
-		      int sF, int sU, const FermionFieldView &in, FermionFieldView &out, int dir,int disp);
-
-   ///////////////////////////////////////////////////////////////////////////////////////
-   // Generic Nc kernels
-   ///////////////////////////////////////////////////////////////////////////////////////
-   void DhopSiteGeneric(StencilImpl &st, LebesgueOrder &lo, 
-			DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU, 
-			SiteSpinor * buf, int LLs, int sU, 
-			const FermionFieldView &in, FermionFieldView &out,int dag);
-   void DhopSiteGenericInt(StencilImpl &st, LebesgueOrder &lo, 
-			   DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU, 
-			   SiteSpinor * buf, int LLs, int sU, 
-			   const FermionFieldView &in, FermionFieldView &out,int dag);
-   void DhopSiteGenericExt(StencilImpl &st, LebesgueOrder &lo, 
-			   DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU,
-			   SiteSpinor * buf, int LLs, int sU, 
-			   const FermionFieldView &in, FermionFieldView &out,int dag);
-
-   ///////////////////////////////////////////////////////////////////////////////////////
-   // Nc=3 specific kernels
-   ///////////////////////////////////////////////////////////////////////////////////////
-   void DhopSiteHand(StencilImpl &st, LebesgueOrder &lo, 
-		     DoubledGaugeFieldView &U,DoubledGaugeFieldView &UUU, 
-		     SiteSpinor * buf, int LLs, int sU, 
-		     const FermionFieldView &in, FermionFieldView &out,int dag);
-   void DhopSiteHandInt(StencilImpl &st, LebesgueOrder &lo, 
-			DoubledGaugeFieldView &U,DoubledGaugeFieldView &UUU, 
-			SiteSpinor * buf, int LLs, int sU, 
-			const FermionFieldView &in, FermionFieldView &out,int dag);
-   void DhopSiteHandExt(StencilImpl &st, LebesgueOrder &lo, 
-			DoubledGaugeFieldView &U,DoubledGaugeFieldView &UUU, 
-			SiteSpinor * buf, int LLs, int sU, 
-			const FermionFieldView &in, FermionFieldView &out,int dag);
-
-   ///////////////////////////////////////////////////////////////////////////////////////
-   // Asm Nc=3 specific kernels
-   ///////////////////////////////////////////////////////////////////////////////////////
-   void DhopSiteAsm(StencilImpl &st, LebesgueOrder &lo, 
-		    DoubledGaugeFieldView &U,DoubledGaugeFieldView &UUU, 
-		    SiteSpinor * buf, int LLs, int sU, 
-		    const FermionFieldView &in, FermionFieldView &out,int dag);
-   ///////////////////////////////////////////////////////////////////////////////////////////////////
-   // Generic interface; fan out to right routine
-   ///////////////////////////////////////////////////////////////////////////////////////////////////
-   void DhopSite(StencilImpl &st, LebesgueOrder &lo, 
-		 DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU, 
-		 SiteSpinor * buf, int LLs, int sU,
-		 const FermionFieldView &in, FermionFieldView &out, int interior=1,int exterior=1);
-
-   void DhopSiteDag(StencilImpl &st, LebesgueOrder &lo, 
-		    DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU, 
-		    SiteSpinor * buf, int LLs, int sU,
-		    const FermionFieldView &in, FermionFieldView &out, int interior=1,int exterior=1);
-
-   void DhopSite(StencilImpl &st, LebesgueOrder &lo, 
-		 DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU, 
-		 SiteSpinor * buf, int LLs, int sU,
-		 const FermionFieldView &in, FermionFieldView &out, int dag, int interior,int exterior);
-  
-public:
-
-  StaggeredKernels(const ImplParams &p = ImplParams());
-
-};
-NAMESPACE_END(Grid);    

Grid/qcd/action/fermion/StaggeredVec5dImpl.h

@@ -1,203 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/FermionOperatorImpl.h
-
-Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@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 S, class Representation = FundamentalRepresentation >
-class StaggeredVec5dImpl : public PeriodicGaugeImpl<GaugeImplTypes<S, Representation::Dimension > > {
-
-public:
-
-  static const int Dimension = Representation::Dimension;
-    static const bool isFundamental = Representation::isFundamental;
-  static const bool LsVectorised=true;
-  typedef RealD   Coeff_t ;
-  typedef PeriodicGaugeImpl<GaugeImplTypes<S, Dimension > > Gimpl;
-      
-  //Necessary?
-  constexpr bool is_fundamental() const{return Dimension == Nc ? 1 : 0;}
-
-
-  INHERIT_GIMPL_TYPES(Gimpl);
-
-  template <typename vtype> using iImplSpinor            = iScalar<iScalar<iVector<vtype, Dimension> > >;
-  template <typename vtype> using iImplHalfSpinor        = iScalar<iScalar<iVector<vtype, Dimension> > >;
-  template <typename vtype> using iImplDoubledGaugeField = iVector<iScalar<iMatrix<vtype, Dimension> >, Nds>;
-  template <typename vtype> using iImplGaugeField        = iVector<iScalar<iMatrix<vtype, Dimension> >, Nd>;
-  template <typename vtype> using iImplGaugeLink         = iScalar<iScalar<iMatrix<vtype, Dimension> > >;
-  template <typename vtype> using iImplPropagator        = iScalar<iScalar<iMatrix<vtype, Dimension> > >;
-
-  // Make the doubled gauge field a *scalar*
-  typedef iImplDoubledGaugeField<typename Simd::scalar_type>  SiteDoubledGaugeField;  // This is a scalar
-  typedef iImplGaugeField<typename Simd::scalar_type>         SiteScalarGaugeField;  // scalar
-  typedef iImplGaugeLink<typename Simd::scalar_type>          SiteScalarGaugeLink;  // scalar
-  typedef iImplPropagator<Simd>        SitePropagator;
-
-  typedef Lattice<SiteDoubledGaugeField> DoubledGaugeField;
-  typedef Lattice<SitePropagator> PropagatorField;
-    
-  typedef iImplSpinor<Simd>            SiteSpinor;
-  typedef iImplHalfSpinor<Simd>        SiteHalfSpinor;
-
-    
-  typedef Lattice<SiteSpinor>            FermionField;
-    
-  typedef StaggeredImplParams ImplParams;
-  typedef SimpleCompressor<SiteSpinor> Compressor;
-  typedef CartesianStencil<SiteSpinor, SiteSpinor, ImplParams> StencilImpl;
-  typedef typename StencilImpl::View_type StencilView;
-    
-  ImplParams Params;
-    
-  StaggeredVec5dImpl(const ImplParams &p = ImplParams()) : Params(p){};
-
-  template <class ref>
-  static accelerator_inline void loadLinkElement(Simd &reg, ref &memory) 
-  {
-    vsplat(reg, memory);
-  }
-
-  static accelerator_inline void multLink(SiteHalfSpinor &phi, 
-					  const SiteDoubledGaugeField &U,
-					  const SiteHalfSpinor &chi, 
-					  int mu) 
-  {
-    SiteGaugeLink UU;
-    for (int i = 0; i < Dimension; i++) {
-      for (int j = 0; j < Dimension; j++) {
-	vsplat(UU()()(i, j), U(mu)()(i, j));
-      }
-    }
-    mult(&phi(), &UU(), &chi());
-  }
-  static accelerator_inline void multLinkAdd(SiteHalfSpinor &phi, 
-					     const SiteDoubledGaugeField &U,
-					     const SiteHalfSpinor &chi, 
-					     int mu) 
-  {
-    SiteGaugeLink UU;
-    for (int i = 0; i < Dimension; i++) {
-      for (int j = 0; j < Dimension; j++) {
-	vsplat(UU()()(i, j), U(mu)()(i, j));
-      }
-    }
-    mac(&phi(), &UU(), &chi());
-  }
-      
-  inline void InsertGaugeField(DoubledGaugeField &U_ds,const GaugeLinkField &U,int mu)
-  {
-    GridBase *GaugeGrid = U_ds.Grid();
-    thread_for(lidx, GaugeGrid->lSites(),{
-
-	SiteScalarGaugeLink   ScalarU;
-	SiteDoubledGaugeField ScalarUds;
-	
-	Coordinate lcoor;
-	GaugeGrid->LocalIndexToLocalCoor(lidx, lcoor);
-	peekLocalSite(ScalarUds, U_ds, lcoor);
-	
-	peekLocalSite(ScalarU, U, lcoor);
-	ScalarUds(mu) = ScalarU();
-	
-    });
-  }
-  inline void DoubleStore(GridBase *GaugeGrid,
-			  DoubledGaugeField &UUUds, // for Naik term
-			  DoubledGaugeField &Uds,
-			  const GaugeField &Uthin,
-			  const GaugeField &Ufat) 
-  {
-
-    GridBase * InputGrid = Uthin.Grid();
-    conformable(InputGrid,Ufat.Grid());
-
-    GaugeLinkField U(InputGrid);
-    GaugeLinkField UU(InputGrid);
-    GaugeLinkField UUU(InputGrid);
-    GaugeLinkField Udag(InputGrid);
-    GaugeLinkField UUUdag(InputGrid);
-
-    for (int mu = 0; mu < Nd; mu++) {
-
-      // Staggered Phase.
-      Lattice<iScalar<vInteger> > coor(InputGrid);
-      Lattice<iScalar<vInteger> > x(InputGrid); LatticeCoordinate(x,0);
-      Lattice<iScalar<vInteger> > y(InputGrid); LatticeCoordinate(y,1);
-      Lattice<iScalar<vInteger> > z(InputGrid); LatticeCoordinate(z,2);
-      Lattice<iScalar<vInteger> > t(InputGrid); LatticeCoordinate(t,3);
-
-      Lattice<iScalar<vInteger> > lin_z(InputGrid); lin_z=x+y;
-      Lattice<iScalar<vInteger> > lin_t(InputGrid); lin_t=x+y+z;
-
-      ComplexField phases(InputGrid);	phases=1.0;
-
-      if ( mu == 1 ) phases = where( mod(x    ,2)==(Integer)0, phases,-phases);
-      if ( mu == 2 ) phases = where( mod(lin_z,2)==(Integer)0, phases,-phases);
-      if ( mu == 3 ) phases = where( mod(lin_t,2)==(Integer)0, phases,-phases);
-
-      // 1 hop based on fat links
-      U      = PeekIndex<LorentzIndex>(Ufat, mu);
-      Udag   = adj( Cshift(U, mu, -1));
-
-      U    = U    *phases;
-      Udag = Udag *phases;
-
-      InsertGaugeField(Uds,U,mu);
-      InsertGaugeField(Uds,Udag,mu+4);
-
-      // 3 hop based on thin links. Crazy huh ?
-      U  = PeekIndex<LorentzIndex>(Uthin, mu);
-      UU = Gimpl::CovShiftForward(U,mu,U);
-      UUU= Gimpl::CovShiftForward(U,mu,UU);
-	
-      UUUdag = adj( Cshift(UUU, mu, -3));
-
-      UUU    = UUU    *phases;
-      UUUdag = UUUdag *phases;
-
-      InsertGaugeField(UUUds,UUU,mu);
-      InsertGaugeField(UUUds,UUUdag,mu+4);
-
-    }
-  }
-
-  inline void InsertForce4D(GaugeField &mat, FermionField &Btilde, FermionField &A,int mu){
-    assert(0);
-  }   
-      
-  inline void InsertForce5D(GaugeField &mat, FermionField &Btilde, FermionField &Atilde,int mu){
-    assert (0); 
-  }
-};
-typedef StaggeredVec5dImpl<vComplex,  FundamentalRepresentation > StaggeredVec5dImplR;   // Real.. whichever prec
-typedef StaggeredVec5dImpl<vComplexF, FundamentalRepresentation > StaggeredVec5dImplF;  // Float
-typedef StaggeredVec5dImpl<vComplexD, FundamentalRepresentation > StaggeredVec5dImplD;  // Double
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/WilsonFermion5D.h

@@ -1,240 +0,0 @@
-
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/WilsonFermion5D.h
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is 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_QCD_WILSON_FERMION_5D_H
-#define  GRID_QCD_WILSON_FERMION_5D_H
-
-#include <Grid/perfmon/Stat.h>
-
-NAMESPACE_BEGIN(Grid);
-
-////////////////////////////////////////////////////////////////////////////////
-// This is the 4d red black case appropriate to support
-//
-// parity = (x+y+z+t)|2;
-// generalised five dim fermions like mobius, zolotarev etc..	
-//
-// i.e. even even contains fifth dim hopping term.
-//
-// [DIFFERS from original CPS red black implementation parity = (x+y+z+t+s)|2 ]
-////////////////////////////////////////////////////////////////////////////////
-
-////////////////////////////////////////////////////////////////////////////////
-// This is the 4d red black case appropriate to support
-//
-// parity = (x+y+z+t)|2;
-// generalised five dim fermions like mobius, zolotarev etc..	
-//
-// i.e. even even contains fifth dim hopping term.
-//
-// [DIFFERS from original CPS red black implementation parity = (x+y+z+t+s)|2 ]
-////////////////////////////////////////////////////////////////////////////////
-
-class WilsonFermion5DStatic { 
-public:
-  // S-direction is INNERMOST and takes no part in the parity.
-  static const std::vector<int> directions;
-  static const std::vector<int> displacements;
-  static constexpr int npoint = 8;
-};
-
-template<class Impl>
-class WilsonFermion5D : public WilsonKernels<Impl>, public WilsonFermion5DStatic
-{
-public:
-  INHERIT_IMPL_TYPES(Impl);
-  typedef WilsonKernels<Impl> Kernels;
-  PmuStat stat;
-
-  FermionField _tmp;
-  FermionField &tmp(void) { return _tmp; }
-
-  void Report(void);
-  void ZeroCounters(void);
-  double DhopCalls;
-  double DhopCommTime;
-  double DhopComputeTime;
-  double DhopComputeTime2;
-  double DhopFaceTime;
-  double DhopTotalTime;
-
-  double DerivCalls;
-  double DerivCommTime;
-  double DerivComputeTime;
-  double DerivDhopComputeTime;
-
-  ///////////////////////////////////////////////////////////////
-  // Implement the abstract base
-  ///////////////////////////////////////////////////////////////
-  GridBase *GaugeGrid(void)              { return _FourDimGrid ;}
-  GridBase *GaugeRedBlackGrid(void)      { return _FourDimRedBlackGrid ;}
-  GridBase *FermionGrid(void)            { return _FiveDimGrid;}
-  GridBase *FermionRedBlackGrid(void)    { return _FiveDimRedBlackGrid;}
-
-  // full checkerboard operations; leave unimplemented as abstract for now
-  virtual RealD  M    (const FermionField &in, FermionField &out){assert(0); return 0.0;};
-  virtual RealD  Mdag (const FermionField &in, FermionField &out){assert(0); return 0.0;};
-
-  // half checkerboard operations; leave unimplemented as abstract for now
-  virtual void   Meooe       (const FermionField &in, FermionField &out){assert(0);};
-  virtual void   Mooee       (const FermionField &in, FermionField &out){assert(0);};
-  virtual void   MooeeInv    (const FermionField &in, FermionField &out){assert(0);};
-
-  virtual void   MeooeDag    (const FermionField &in, FermionField &out){assert(0);};
-  virtual void   MooeeDag    (const FermionField &in, FermionField &out){assert(0);};
-  virtual void   MooeeInvDag (const FermionField &in, FermionField &out){assert(0);};
-  virtual void   Mdir   (const FermionField &in, FermionField &out,int dir,int disp){assert(0);};   // case by case Wilson, Clover, Cayley, ContFrac, PartFrac
-
-  // These can be overridden by fancy 5d chiral action
-  virtual void DhopDeriv  (GaugeField &mat,const FermionField &U,const FermionField &V,int dag);
-  virtual void DhopDerivEO(GaugeField &mat,const FermionField &U,const FermionField &V,int dag);
-  virtual void DhopDerivOE(GaugeField &mat,const FermionField &U,const FermionField &V,int dag);
-
-      void MomentumSpacePropagatorHt_5d(FermionField &out,const FermionField &in,RealD mass,std::vector<double> twist) ;
-      void MomentumSpacePropagatorHt(FermionField &out,const FermionField &in,RealD mass,std::vector<double> twist) ;
-      void MomentumSpacePropagatorHw(FermionField &out,const FermionField &in,RealD mass,std::vector<double> twist) ;
-
-  // Implement hopping term non-hermitian hopping term; half cb or both
-  // Implement s-diagonal DW
-  void DW    (const FermionField &in, FermionField &out,int dag);
-  void Dhop  (const FermionField &in, FermionField &out,int dag);
-  void DhopOE(const FermionField &in, FermionField &out,int dag);
-  void DhopEO(const FermionField &in, FermionField &out,int dag);
-
-  // add a DhopComm
-  // -- suboptimal interface will presently trigger multiple comms.
-  void DhopDir(const FermionField &in, FermionField &out,int dir,int disp);
-    
-  ///////////////////////////////////////////////////////////////
-  // New methods added 
-  ///////////////////////////////////////////////////////////////
-  void DerivInternal(StencilImpl & st,
-		     DoubledGaugeField & U,
-		     GaugeField &mat,
-		     const FermionField &A,
-		     const FermionField &B,
-		     int dag);
-    
-  void DhopInternal(StencilImpl & st,
-		    LebesgueOrder &lo,
-		    DoubledGaugeField &U,
-		    const FermionField &in, 
-		    FermionField &out,
-		    int dag);
-
-  void DhopInternalOverlappedComms(StencilImpl & st,
-				   LebesgueOrder &lo,
-				   DoubledGaugeField &U,
-				   const FermionField &in, 
-				   FermionField &out,
-				   int dag);
-
-  void DhopInternalSerialComms(StencilImpl & st,
-			       LebesgueOrder &lo,
-			       DoubledGaugeField &U,
-			       const FermionField &in, 
-			       FermionField &out,
-			       int dag);
-    
-  // Constructors
-  WilsonFermion5D(GaugeField &_Umu,
-		  GridCartesian         &FiveDimGrid,
-		  GridRedBlackCartesian &FiveDimRedBlackGrid,
-		  GridCartesian         &FourDimGrid,
-		  GridRedBlackCartesian &FourDimRedBlackGrid,
-		  double _M5,const ImplParams &p= ImplParams());
-    
-  // Constructors
-  /*
-    WilsonFermion5D(int simd, 
-    GaugeField &_Umu,
-    GridCartesian         &FiveDimGrid,
-    GridRedBlackCartesian &FiveDimRedBlackGrid,
-    GridCartesian         &FourDimGrid,
-    double _M5,const ImplParams &p= ImplParams());
-  */
-    
-  // DoubleStore
-  void ImportGauge(const GaugeField &_Umu);
-    
-  ///////////////////////////////////////////////////////////////
-  // Data members require to support the functionality
-  ///////////////////////////////////////////////////////////////
-public:
-    
-  // Add these to the support from Wilson
-  GridBase *_FourDimGrid;
-  GridBase *_FourDimRedBlackGrid;
-  GridBase *_FiveDimGrid;
-  GridBase *_FiveDimRedBlackGrid;
-    
-  double                        M5;
-  int Ls;
-    
-  //Defines the stencils for even and odd
-  StencilImpl Stencil; 
-  StencilImpl StencilEven; 
-  StencilImpl StencilOdd; 
-    
-  // Copy of the gauge field , with even and odd subsets
-  DoubledGaugeField Umu;
-  DoubledGaugeField UmuEven;
-  DoubledGaugeField UmuOdd;
-    
-  LebesgueOrder Lebesgue;
-  LebesgueOrder LebesgueEvenOdd;
-    
-  // Comms buffer
-  std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> >  comm_buf;
-    
-  ///////////////////////////////////////////////////////////////
-  // Conserved current utilities
-  ///////////////////////////////////////////////////////////////
-  void ContractConservedCurrent(PropagatorField &q_in_1,
-				PropagatorField &q_in_2,
-				PropagatorField &q_out,
-				Current curr_type, 
-				unsigned int mu);
-  void SeqConservedCurrent(PropagatorField &q_in,
-			   PropagatorField &q_out,
-			   Current curr_type,
-			   unsigned int mu,
-			   unsigned int tmin,
-			   unsigned int tmax,
-			   ComplexField &lattice_cmplx);
-
-  void ContractJ5q(PropagatorField &q_in,ComplexField &J5q);
-  void ContractJ5q(FermionField &q_in,ComplexField &J5q);
-
-};
-
-NAMESPACE_END(Grid);
-
-#endif

Grid/qcd/action/fermion/WilsonImpl.h

@@ -1,226 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/FermionOperatorImpl.h
-
-Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@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);
-
-  
-/////////////////////////////////////////////////////////////////////////////
-// Single flavour four spinors with colour index
-/////////////////////////////////////////////////////////////////////////////
-template <class S, class Representation = FundamentalRepresentation,class Options = CoeffReal >
-class WilsonImpl : public PeriodicGaugeImpl<GaugeImplTypes<S, Representation::Dimension > > {
-public:
-
-  static const int Dimension = Representation::Dimension;
-  static const bool isFundamental = Representation::isFundamental;
-  static const bool LsVectorised=false;
-  static const int Nhcs = Options::Nhcs;
-
-  typedef PeriodicGaugeImpl<GaugeImplTypes<S, Dimension > > Gimpl;
-  INHERIT_GIMPL_TYPES(Gimpl);
-      
-  //Necessary?
-  constexpr bool is_fundamental() const{return Dimension == Nc ? 1 : 0;}
-    
-  typedef typename Options::_Coeff_t Coeff_t;
-  typedef typename Options::template PrecisionMapper<Simd>::LowerPrecVector SimdL;
-      
-  template <typename vtype> using iImplSpinor            = iScalar<iVector<iVector<vtype, Dimension>, Ns> >;
-  template <typename vtype> using iImplPropagator        = iScalar<iMatrix<iMatrix<vtype, Dimension>, Ns> >;
-  template <typename vtype> using iImplHalfSpinor        = iScalar<iVector<iVector<vtype, Dimension>, Nhs> >;
-  template <typename vtype> using iImplHalfCommSpinor    = iScalar<iVector<iVector<vtype, Dimension>, Nhcs> >;
-  template <typename vtype> using iImplDoubledGaugeField = iVector<iScalar<iMatrix<vtype, Dimension> >, Nds>;
-    
-  typedef iImplSpinor<Simd>            SiteSpinor;
-  typedef iImplPropagator<Simd>        SitePropagator;
-  typedef iImplHalfSpinor<Simd>        SiteHalfSpinor;
-  typedef iImplHalfCommSpinor<SimdL>   SiteHalfCommSpinor;
-  typedef iImplDoubledGaugeField<Simd> SiteDoubledGaugeField;
-    
-  typedef Lattice<SiteSpinor>            FermionField;
-  typedef Lattice<SitePropagator>        PropagatorField;
-  typedef Lattice<SiteDoubledGaugeField> DoubledGaugeField;
-    
-  typedef WilsonCompressor<SiteHalfCommSpinor,SiteHalfSpinor, SiteSpinor> Compressor;
-  typedef WilsonImplParams ImplParams;
-  typedef WilsonStencil<SiteSpinor, SiteHalfSpinor,ImplParams> StencilImpl;
-  typedef typename StencilImpl::View_type StencilView;
-    
-  ImplParams Params;
-
-  WilsonImpl(const ImplParams &p = ImplParams()) : Params(p){
-    assert(Params.boundary_phases.size() == Nd);
-  };
-
-  template<class _Spinor>
-  static accelerator_inline void multLink(_Spinor &phi,
-					  const SiteDoubledGaugeField &U,
-					  const _Spinor &chi,
-					  int mu) 
-  {
-    auto UU = coalescedRead(U(mu));
-    mult(&phi(), &UU, &chi());
-  }
-  template<class _Spinor>
-  static accelerator_inline void multLink(_Spinor &phi,
-					  const SiteDoubledGaugeField &U,
-					  const _Spinor &chi,
-					  int mu,
-					  StencilEntry *SE,
-					  StencilView &St) 
-  {
-    multLink(phi,U,chi,mu);
-  }
-    
-      
-  template <class ref>
-  static accelerator_inline void loadLinkElement(Simd &reg, ref &memory) 
-  {
-    reg = memory;
-  }
-      
-  inline void DoubleStore(GridBase *GaugeGrid,
-			  DoubledGaugeField &Uds,
-			  const GaugeField &Umu) 
-  {
-    typedef typename Simd::scalar_type scalar_type;
-
-    conformable(Uds.Grid(), GaugeGrid);
-    conformable(Umu.Grid(), GaugeGrid);
-
-    GaugeLinkField U(GaugeGrid);
-    GaugeLinkField tmp(GaugeGrid);
-
-    Lattice<iScalar<vInteger> > coor(GaugeGrid);
-      ////////////////////////////////////////////////////
-      // apply any boundary phase or twists
-      ////////////////////////////////////////////////////
-    for (int mu = 0; mu < Nd; mu++) {
-
-	////////// boundary phase /////////////
-      auto pha = Params.boundary_phases[mu];
-      scalar_type phase( real(pha),imag(pha) );
-
-	int L   = GaugeGrid->GlobalDimensions()[mu];
-        int Lmu = L - 1;
-
-      LatticeCoordinate(coor, mu);
-
-      U = PeekIndex<LorentzIndex>(Umu, mu);
-
-	// apply any twists
-	RealD theta = Params.twist_n_2pi_L[mu] * 2*M_PI / L;
-	if ( theta != 0.0) { 
-	  scalar_type twphase(::cos(theta),::sin(theta));
-	  U = twphase*U;
-	  std::cout << GridLogMessage << " Twist ["<<mu<<"] "<< Params.twist_n_2pi_L[mu]<< " phase"<<phase <<std::endl;
-	}
-
-      tmp = where(coor == Lmu, phase * U, U);
-      PokeIndex<LorentzIndex>(Uds, tmp, mu);
-
-      U = adj(Cshift(U, mu, -1));
-      U = where(coor == 0, conjugate(phase) * U, U); 
-      PokeIndex<LorentzIndex>(Uds, U, mu + 4);
-    }
-  }
-
-  inline void InsertForce4D(GaugeField &mat, FermionField &Btilde, FermionField &A,int mu){
-    GaugeLinkField link(mat.Grid());
-    link = TraceIndex<SpinIndex>(outerProduct(Btilde,A)); 
-    PokeIndex<LorentzIndex>(mat,link,mu);
-  }   
-      
-    inline void outerProductImpl(PropagatorField &mat, const FermionField &B, const FermionField &A){
-      mat = outerProduct(B,A); 
-    }  
-
-    inline void TraceSpinImpl(GaugeLinkField &mat, PropagatorField&P) {
-      mat = TraceIndex<SpinIndex>(P); 
-    }
-      
-    inline void extractLinkField(std::vector<GaugeLinkField> &mat, DoubledGaugeField &Uds){
-      for (int mu = 0; mu < Nd; mu++)
-      mat[mu] = PeekIndex<LorentzIndex>(Uds, mu);
-    }
-
-
-  inline void InsertForce5D(GaugeField &mat, FermionField &Btilde, FermionField &Atilde,int mu){
-      
-    int Ls=Btilde.Grid()->_fdimensions[0];
-    GaugeLinkField tmp(mat.Grid());
-    tmp = Zero();
-    auto tmp_v = tmp.View();
-    auto Btilde_v = Btilde.View();
-    auto Atilde_v = Atilde.View();
-    thread_for(sss,tmp.Grid()->oSites(),{
-      int sU=sss;
-      for(int s=0;s<Ls;s++){
-	int sF = s+Ls*sU;
-	tmp_v[sU] = tmp_v[sU]+ traceIndex<SpinIndex>(outerProduct(Btilde_v[sF],Atilde_v[sF])); // ordering here
-      }
-    });
-    PokeIndex<LorentzIndex>(mat,tmp,mu);
-      
-  }
-};
-
-
-typedef WilsonImpl<vComplex,  FundamentalRepresentation, CoeffReal > WilsonImplR;  // Real.. whichever prec
-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, 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,  AdjointRepresentation, CoeffReal > WilsonAdjImplR;   // Real.. whichever prec
-typedef WilsonImpl<vComplexF, AdjointRepresentation, CoeffReal > WilsonAdjImplF;  // Float
-typedef WilsonImpl<vComplexD, AdjointRepresentation, CoeffReal > WilsonAdjImplD;  // Double
- 
-typedef WilsonImpl<vComplex,  TwoIndexSymmetricRepresentation, CoeffReal > WilsonTwoIndexSymmetricImplR;   // Real.. whichever prec
-typedef WilsonImpl<vComplexF, TwoIndexSymmetricRepresentation, CoeffReal > WilsonTwoIndexSymmetricImplF;  // Float
-typedef WilsonImpl<vComplexD, TwoIndexSymmetricRepresentation, CoeffReal > WilsonTwoIndexSymmetricImplD;  // Double
- 
-typedef WilsonImpl<vComplex,  TwoIndexAntiSymmetricRepresentation, CoeffReal > WilsonTwoIndexAntiSymmetricImplR;   // Real.. whichever prec
-typedef WilsonImpl<vComplexF, TwoIndexAntiSymmetricRepresentation, CoeffReal > WilsonTwoIndexAntiSymmetricImplF;  // Float
-typedef WilsonImpl<vComplexD, TwoIndexAntiSymmetricRepresentation, CoeffReal > WilsonTwoIndexAntiSymmetricImplD;  // Double
-
-
-NAMESPACE_END(Grid);
-

Grid/qcd/action/fermion/WilsonKernels.h

@@ -1,167 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.h
-
-Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is 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);
-
-////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-// Helper routines that implement Wilson stencil for a single site.
-// Common to both the WilsonFermion and WilsonFermion5D
-////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-class WilsonKernelsStatic { 
-public:
-  enum { OptGeneric, OptHandUnroll, OptInlineAsm };
-  enum { CommsAndCompute, CommsThenCompute };
-  static int Opt;  
-  static int Comms;
-};
- 
-template<class Impl> class WilsonKernels : public FermionOperator<Impl> , public WilsonKernelsStatic { 
-public:
-
-  INHERIT_IMPL_TYPES(Impl);
-  typedef FermionOperator<Impl> Base;
-   
-public:
-
-  static void DhopKernel(int Opt,StencilImpl &st,  DoubledGaugeField &U, SiteHalfSpinor * buf,
-			 int Ls, int Nsite, const FermionField &in, FermionField &out,
-			 int interior=1,int exterior=1) ;
-
-  static void DhopDagKernel(int Opt,StencilImpl &st,  DoubledGaugeField &U, SiteHalfSpinor * buf,
-			    int Ls, int Nsite, const FermionField &in, FermionField &out,
-			    int interior=1,int exterior=1) ;
-
-  static void DhopDirKernel(StencilImpl &st, DoubledGaugeField &U,SiteHalfSpinor * buf,
-			    int Ls, int Nsite, const FermionField &in, FermionField &out, int dirdisp, int gamma);
-
-  //////////////////////////////////////////////////////////////////////////////
-  // Utilities for inserting Wilson conserved current.
-  //////////////////////////////////////////////////////////////////////////////
-  static void ContractConservedCurrentSiteFwd(const SitePropagator &q_in_1,
-                                       const SitePropagator &q_in_2,
-                                       SitePropagator &q_out,
-                                       DoubledGaugeFieldView &U,
-                                       unsigned int sU,
-                                       unsigned int mu,
-                                       bool switch_sign = false);
-
-  static void ContractConservedCurrentSiteBwd(const SitePropagator &q_in_1,
-                                       const SitePropagator &q_in_2,
-                                       SitePropagator &q_out,
-                                       DoubledGaugeFieldView &U,
-                                       unsigned int sU,
-                                       unsigned int mu,
-                                       bool switch_sign = false);
-
-  static void SeqConservedCurrentSiteFwd(const SitePropagator &q_in, 
-                                  SitePropagator &q_out,
-                                  DoubledGaugeFieldView &U,
-                                  unsigned int sU,
-                                  unsigned int mu,
-                                  vPredicate t_mask,
-                                  bool switch_sign = false);
-
-  static void SeqConservedCurrentSiteBwd(const SitePropagator &q_in,
-                                  SitePropagator &q_out,
-                                  DoubledGaugeFieldView &U,
-                                  unsigned int sU,
-                                  unsigned int mu,
-                                  vPredicate t_mask,
-                                  bool switch_sign = false);
-
-private:
-
-  static accelerator void DhopDirK(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor * buf,
-				   int sF, int sU, const FermionFieldView &in, FermionFieldView &out, int dirdisp, int gamma);
-      
-  // Specialised variants
-  static accelerator void GenericDhopSite(StencilView &st,  DoubledGaugeFieldView &U, SiteHalfSpinor * buf,
-					  int sF, int sU, const FermionFieldView &in, FermionFieldView &out);
-      
-  static accelerator void GenericDhopSiteDag(StencilView &st,  DoubledGaugeFieldView &U, SiteHalfSpinor * buf,
-						    int sF, int sU, const FermionFieldView &in, FermionFieldView &out);
-  
-  static accelerator void GenericDhopSiteInt(StencilView &st,  DoubledGaugeFieldView &U, SiteHalfSpinor * buf,
-						    int sF, int sU, const FermionFieldView &in, FermionFieldView &out);
-      
-  static accelerator void GenericDhopSiteDagInt(StencilView &st,  DoubledGaugeFieldView &U, SiteHalfSpinor * buf,
-						int sF, int sU, const FermionFieldView &in, FermionFieldView &out);
-  
-  static accelerator void GenericDhopSiteExt(StencilView &st,  DoubledGaugeFieldView &U, SiteHalfSpinor * buf,
-					     int sF, int sU, const FermionFieldView &in, FermionFieldView &out);
-      
-  static accelerator void GenericDhopSiteDagExt(StencilView &st,  DoubledGaugeFieldView &U, SiteHalfSpinor * buf,
-						       int sF, int sU, const FermionFieldView &in, FermionFieldView &out);
-
-  static void AsmDhopSite(StencilView &st,  DoubledGaugeFieldView &U, SiteHalfSpinor * buf,
-			  int sF, int sU, int Ls, int Nsite, const FermionFieldView &in,FermionFieldView &out);
-  
-  static void AsmDhopSiteDag(StencilView &st,  DoubledGaugeFieldView &U, SiteHalfSpinor * buf,
-			     int sF, int sU, int Ls, int Nsite, const FermionFieldView &in, FermionFieldView &out);
-  
-  static void AsmDhopSiteInt(StencilView &st,  DoubledGaugeFieldView &U, SiteHalfSpinor * buf,
-			     int sF, int sU, int Ls, int Nsite, const FermionFieldView &in,FermionFieldView &out);
-  
-  static void AsmDhopSiteDagInt(StencilView &st,  DoubledGaugeFieldView &U, SiteHalfSpinor * buf,
-				int sF, int sU, int Ls, int Nsite, const FermionFieldView &in, FermionFieldView &out);
-  
-  static void AsmDhopSiteExt(StencilView &st,  DoubledGaugeFieldView &U, SiteHalfSpinor * buf,
-			     int sF, int sU, int Ls, int Nsite, const FermionFieldView &in,FermionFieldView &out);
-  
-  static void AsmDhopSiteDagExt(StencilView &st,  DoubledGaugeFieldView &U, SiteHalfSpinor * buf,
-				int sF, int sU, int Ls, int Nsite, const FermionFieldView &in, FermionFieldView &out);
-
-// Keep Hand unrolled temporarily  
-  static accelerator void HandDhopSite(StencilView &st,  DoubledGaugeFieldView &U, SiteHalfSpinor * buf,
-				       int sF, int sU, const FermionFieldView &in, FermionFieldView &out);
-  
-  static accelerator void HandDhopSiteDag(StencilView &st,  DoubledGaugeFieldView &U, SiteHalfSpinor * buf,
-					  int sF, int sU, const FermionFieldView &in, FermionFieldView &out);
-  
-  static accelerator void HandDhopSiteInt(StencilView &st,  DoubledGaugeFieldView &U, SiteHalfSpinor * buf,
-					  int sF, int sU, const FermionFieldView &in, FermionFieldView &out);
-  
-  static accelerator void HandDhopSiteDagInt(StencilView &st,  DoubledGaugeFieldView &U, SiteHalfSpinor * buf,
-					     int sF, int sU, const FermionFieldView &in, FermionFieldView &out);
-  
-  static accelerator void HandDhopSiteExt(StencilView &st,  DoubledGaugeFieldView &U, SiteHalfSpinor * buf,
-					  int sF, int sU, const FermionFieldView &in, FermionFieldView &out);
-  
-  static accelerator void HandDhopSiteDagExt(StencilView &st,  DoubledGaugeFieldView &U, SiteHalfSpinor * buf,
-					     int sF, int sU, const FermionFieldView &in, FermionFieldView &out);
- public:
- WilsonKernels(const ImplParams &p = ImplParams()) : Base(p){};
-};
-    
-NAMESPACE_END(Grid);
-
-

Grid/qcd/action/fermion/WilsonTMFermion.h

@@ -1,74 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/WilsonTMFermion.h
-
-    Copyright (C) 2015
-
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is 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/FermionCore.h>
-#include <Grid/qcd/action/fermion/WilsonFermion.h>
-
-NAMESPACE_BEGIN(Grid);
-
-template<class Impl>
-class WilsonTMFermion : public WilsonFermion<Impl>
-{
-public:
-  INHERIT_IMPL_TYPES(Impl);
-public:
-
-  virtual void   Instantiatable(void) {};
-  // Constructors
-  WilsonTMFermion(GaugeField &_Umu,
-		  GridCartesian         &Fgrid,
-		  GridRedBlackCartesian &Hgrid, 
-		  RealD _mass,
-		  RealD _mu,
-		  const ImplParams &p= ImplParams()
-		  ) :
-    WilsonFermion<Impl>(_Umu,
-			Fgrid,
-			Hgrid,
-			_mass,p)
-
-  {
-    mu = _mu;
-  }
-
-
-  // allow override for twisted mass and clover
-  virtual void Mooee(const FermionField &in, FermionField &out) ;
-  virtual void MooeeDag(const FermionField &in, FermionField &out) ;
-  virtual void MooeeInv(const FermionField &in, FermionField &out) ;
-  virtual void MooeeInvDag(const FermionField &in, FermionField &out) ;
-
-private:
-  RealD mu; // TwistedMass parameter
-
-};
-
-NAMESPACE_END(Grid);
-
-

Grid/qcd/action/fermion/WilsonTMFermion5D.h

@@ -1,152 +0,0 @@
-    /*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/WilsonTMFermion5D.h
-
-    Copyright (C) 2015
-
-Author: paboyle <paboyle@ph.ed.ac.uk> ; NB Christoph did similar in GPT
-
-    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/FermionCore.h>
-#include <Grid/qcd/action/fermion/WilsonFermion.h>
-
-NAMESPACE_BEGIN(Grid);
-    
-template<class Impl>
-class WilsonTMFermion5D : public WilsonFermion5D<Impl>
-{
- public:
-  INHERIT_IMPL_TYPES(Impl);
- public:
-  
-  virtual void   Instantiatable(void) {};
-  
-  // Constructors
- WilsonTMFermion5D(GaugeField &_Umu,
-		   GridCartesian         &Fgrid,
-		   GridRedBlackCartesian &Frbgrid, 
-		   GridCartesian         &Ugrid,
-		   GridRedBlackCartesian &Urbgrid, 
-		   const std::vector<RealD> _mass,
-		   const std::vector<RealD> _mu,
-		   const ImplParams &p= ImplParams()
-		   ) :
-  WilsonFermion5D<Impl>(_Umu,
-			Fgrid,
-			Frbgrid,
-			Ugrid,
-			Urbgrid,
-			4.0,p)
-   
-    {
-      update(_mass,_mu);
-    }
-  
-  virtual void Meooe(const FermionField &in, FermionField &out) {
-    if (in.Checkerboard() == Odd) {
-      this->DhopEO(in, out, DaggerNo);
-    } else {
-      this->DhopOE(in, out, DaggerNo);
-    }
-  }
-  
-  virtual void MeooeDag(const FermionField &in, FermionField &out) {
-    if (in.Checkerboard() == Odd) {
-      this->DhopEO(in, out, DaggerYes);
-    } else {
-      this->DhopOE(in, out, DaggerYes);
-    }
-  }	
-  
-  // allow override for twisted mass and clover
-  virtual void Mooee(const FermionField &in, FermionField &out) {
-    out.Checkerboard() = in.Checkerboard();
-    //axpibg5x(out,in,a,b); // out = a*in + b*i*G5*in
-    for (int s=0;s<(int)this->mass.size();s++) {
-      ComplexD a = 4.0+this->mass[s];
-      ComplexD b(0.0,this->mu[s]);
-      axpbg5y_ssp(out,a,in,b,in,s,s);
-    }
-  }
-  
-  virtual void MooeeDag(const FermionField &in, FermionField &out) {
-    out.Checkerboard() = in.Checkerboard();
-    for (int s=0;s<(int)this->mass.size();s++) {
-      ComplexD a = 4.0+this->mass[s];
-      ComplexD b(0.0,-this->mu[s]);
-      axpbg5y_ssp(out,a,in,b,in,s,s);
-    }
-  }
-  virtual void MooeeInv(const FermionField &in, FermionField &out) {
-    for (int s=0;s<(int)this->mass.size();s++) {
-      RealD m    = this->mass[s];
-      RealD tm   = this->mu[s];
-      RealD mtil = 4.0+this->mass[s];
-      RealD sq   = mtil*mtil+tm*tm;
-      ComplexD a    = mtil/sq;
-      ComplexD b(0.0, -tm /sq);
-      axpbg5y_ssp(out,a,in,b,in,s,s);
-    }
-  }
-  virtual void MooeeInvDag(const FermionField &in, FermionField &out) {
-    for (int s=0;s<(int)this->mass.size();s++) {
-      RealD m    = this->mass[s];
-      RealD tm   = this->mu[s];
-      RealD mtil = 4.0+this->mass[s];
-      RealD sq   = mtil*mtil+tm*tm;
-      ComplexD a    = mtil/sq;
-      ComplexD b(0.0,tm /sq);
-      axpbg5y_ssp(out,a,in,b,in,s,s);
-    }
-  }
-  
-  virtual RealD M(const FermionField &in, FermionField &out) {
-    out.Checkerboard() = in.Checkerboard();
-    this->Dhop(in, out, DaggerNo);
-    FermionField tmp(out.Grid());
-    for (int s=0;s<(int)this->mass.size();s++) {
-      ComplexD a = 4.0+this->mass[s];
-      ComplexD b(0.0,this->mu[s]);
-      axpbg5y_ssp(tmp,a,in,b,in,s,s);
-    }
-    return axpy_norm(out, 1.0, tmp, out);
-  }
-  
-  // needed for fast PV
-  void update(const std::vector<RealD>& _mass, const std::vector<RealD>& _mu) {
-    assert(_mass.size() == _mu.size());
-    assert(_mass.size() == this->FermionGrid()->_fdimensions[0]);
-    this->mass = _mass;
-    this->mu = _mu;
-  }
-  
- private:
-  std::vector<RealD> mu;
-  std::vector<RealD> mass;
-  
-};
-   
-typedef WilsonTMFermion5D<WilsonImplF> WilsonTMFermion5DF; 
-typedef WilsonTMFermion5D<WilsonImplD> WilsonTMFermion5DD; 
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/ZMobiusFermion.h

@@ -1,73 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/MobiusFermion.h
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-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
-
-#include <Grid/qcd/action/fermion/FermionCore.h>
-
-NAMESPACE_BEGIN(Grid);
-
-template<class Impl>
-class ZMobiusFermion : public CayleyFermion5D<Impl>
-{
-public:
-  INHERIT_IMPL_TYPES(Impl);
-public:
-
-  virtual void   Instantiatable(void) {};
-  // Constructors
-  ZMobiusFermion(GaugeField &_Umu,
-		 GridCartesian         &FiveDimGrid,
-		 GridRedBlackCartesian &FiveDimRedBlackGrid,
-		 GridCartesian         &FourDimGrid,
-		 GridRedBlackCartesian &FourDimRedBlackGrid,
-		 RealD _mass,RealD _M5,
-		 std::vector<ComplexD> &gamma, RealD b,RealD c,const ImplParams &p= ImplParams()) : 
-      
-    CayleyFermion5D<Impl>(_Umu,
-			  FiveDimGrid,
-			  FiveDimRedBlackGrid,
-			  FourDimGrid,
-			  FourDimRedBlackGrid,_mass,_M5,p)
-
-  {
-    //    RealD eps = 1.0;
-    std::cout<<GridLogMessage << "ZMobiusFermion (b="<<b<<",c="<<c<<") with Ls= "<<this->Ls<<" gamma passed in"<<std::endl;
-    Vector<Coeff_t> zgamma(this->Ls);
-    for(int s=0;s<this->Ls;s++){
-      zgamma[s] = gamma[s];
-    }
-
-    // Call base setter
-    this->SetCoefficientsInternal(1.0,zgamma,b,c);
-  }
-
-};
-
-NAMESPACE_END(Grid);
-

Grid/qcd/action/fermion/deprecated/DomainWallEOFAFermiondense.h

@@ -1,158 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/DomainWallEOFAFermiondense.cc
-
-Copyright (C) 2017
-
-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: David Murphy <dmurphy@phys.columbia.edu>
-
-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_Eigen_Dense.h>
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/DomainWallEOFAFermion.h>
-
-NAMESPACE_BEGIN(Grid);
-
-/*
- * Dense matrix versions of routines
- */
-template<class Impl>
-void DomainWallEOFAFermion<Impl>::MooeeInvDag(const FermionField& psi, FermionField& chi)
-{
-  this->MooeeInternal(psi, chi, DaggerYes, InverseYes);
-}
-
-template<class Impl>
-void DomainWallEOFAFermion<Impl>::MooeeInv(const FermionField& psi, FermionField& chi)
-{
-  this->MooeeInternal(psi, chi, DaggerNo, InverseYes);
-}
-
-template<class Impl>
-void DomainWallEOFAFermion<Impl>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv)
-{
-  int Ls = this->Ls;
-  int LLs = psi.Grid()->_rdimensions[0];
-  int vol = psi.Grid()->oSites()/LLs;
-
-  chi.Checkerboard() = psi.Checkerboard();
-
-  assert(Ls==LLs);
-
-  Eigen::MatrixXd Pplus  = Eigen::MatrixXd::Zero(Ls,Ls);
-  Eigen::MatrixXd Pminus = Eigen::MatrixXd::Zero(Ls,Ls);
-
-  for(int s=0;s<Ls;s++){
-    Pplus(s,s)  = this->bee[s];
-    Pminus(s,s) = this->bee[s];
-  }
-
-  for(int s=0; s<Ls-1; s++){
-    Pminus(s,s+1) = -this->cee[s];
-  }
-
-  for(int s=0; s<Ls-1; s++){
-    Pplus(s+1,s) = -this->cee[s+1];
-  }
-
-  Pplus (0,Ls-1) = this->dp;
-  Pminus(Ls-1,0) = this->dm;
-
-  Eigen::MatrixXd PplusMat ;
-  Eigen::MatrixXd PminusMat;
-
-  if(inv) {
-    PplusMat  = Pplus.inverse();
-    PminusMat = Pminus.inverse();
-  } else {
-    PplusMat  = Pplus;
-    PminusMat = Pminus;
-  }
-
-  if(dag){
-    PplusMat.adjointInPlace();
-    PminusMat.adjointInPlace();
-  }
-
-  // For the non-vectorised s-direction this is simple
-
-  for(auto site=0; site<vol; site++){
-
-    SiteSpinor     SiteChi;
-    SiteHalfSpinor SitePplus;
-    SiteHalfSpinor SitePminus;
-
-    for(int s1=0; s1<Ls; s1++){
-      SiteChi = Zero();
-      for(int s2=0; s2<Ls; s2++){
-	int lex2 = s2 + Ls*site;
-	if(PplusMat(s1,s2) != 0.0){
-	  spProj5p(SitePplus,psi[lex2]);
-	  accumRecon5p(SiteChi, PplusMat(s1,s2)*SitePplus);
-	}
-	if(PminusMat(s1,s2) != 0.0){
-	  spProj5m(SitePminus, psi[lex2]);
-	  accumRecon5m(SiteChi, PminusMat(s1,s2)*SitePminus);
-	}
-      }
-      chi[s1+Ls*site] = SiteChi*0.5;
-    }
-  }
-}
-
-#ifdef DOMAIN_WALL_EOFA_DPERP_DENSE
-
-INSTANTIATE_DPERP_DWF_EOFA(GparityWilsonImplF);
-INSTANTIATE_DPERP_DWF_EOFA(GparityWilsonImplD);
-INSTANTIATE_DPERP_DWF_EOFA(WilsonImplF);
-INSTANTIATE_DPERP_DWF_EOFA(WilsonImplD);
-INSTANTIATE_DPERP_DWF_EOFA(ZWilsonImplF);
-INSTANTIATE_DPERP_DWF_EOFA(ZWilsonImplD);
-
-template void DomainWallEOFAFermion<GparityWilsonImplF>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-template void DomainWallEOFAFermion<GparityWilsonImplD>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-template void DomainWallEOFAFermion<WilsonImplF>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-template void DomainWallEOFAFermion<WilsonImplD>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-template void DomainWallEOFAFermion<ZWilsonImplF>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-template void DomainWallEOFAFermion<ZWilsonImplD>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-
-INSTANTIATE_DPERP_DWF_EOFA(GparityWilsonImplFH);
-INSTANTIATE_DPERP_DWF_EOFA(GparityWilsonImplDF);
-INSTANTIATE_DPERP_DWF_EOFA(WilsonImplFH);
-INSTANTIATE_DPERP_DWF_EOFA(WilsonImplDF);
-INSTANTIATE_DPERP_DWF_EOFA(ZWilsonImplFH);
-INSTANTIATE_DPERP_DWF_EOFA(ZWilsonImplDF);
-
-template void DomainWallEOFAFermion<GparityWilsonImplFH>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-template void DomainWallEOFAFermion<GparityWilsonImplDF>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-template void DomainWallEOFAFermion<WilsonImplFH>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-template void DomainWallEOFAFermion<WilsonImplDF>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-template void DomainWallEOFAFermion<ZWilsonImplFH>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-template void DomainWallEOFAFermion<ZWilsonImplDF>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-
-#endif
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/deprecated/DomainWallEOFAFermionssp.h

@@ -1,167 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/DomainWallEOFAFermionssp.cc
-
-Copyright (C) 2017
-
-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: David Murphy <dmurphy@phys.columbia.edu>
-
-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/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/DomainWallEOFAFermion.h>
-
-NAMESPACE_BEGIN(Grid);
-
-// FIXME -- make a version of these routines with site loop outermost for cache reuse.
-// Pminus fowards
-// Pplus  backwards
-template<class Impl>
-void DomainWallEOFAFermion<Impl>::M5D(const FermionField& psi, const FermionField& phi,
-				      FermionField& chi, Vector<Coeff_t>& lower, Vector<Coeff_t>& diag, Vector<Coeff_t>& upper)
-{
-  Coeff_t one(1.0);
-  int Ls = this->Ls;
-  for(int s=0; s<Ls; s++){
-    if(s==0) {
-      axpby_ssp_pminus(chi, diag[s], phi, upper[s], psi, s, s+1);
-      axpby_ssp_pplus (chi, one, chi, lower[s], psi, s, Ls-1);
-    } else if (s==(Ls-1)) {
-      axpby_ssp_pminus(chi, diag[s], phi, upper[s], psi, s, 0);
-      axpby_ssp_pplus (chi, one, chi, lower[s], psi, s, s-1);
-    } else {
-      axpby_ssp_pminus(chi, diag[s], phi, upper[s], psi, s, s+1);
-      axpby_ssp_pplus(chi, one, chi, lower[s], psi, s, s-1);
-    }
-  }
-}
-
-template<class Impl>
-void DomainWallEOFAFermion<Impl>::M5Ddag(const FermionField& psi, const FermionField& phi,
-					 FermionField& chi, Vector<Coeff_t>& lower, Vector<Coeff_t>& diag, Vector<Coeff_t>& upper)
-{
-  Coeff_t one(1.0);
-  int Ls = this->Ls;
-  for(int s=0; s<Ls; s++){
-    if(s==0) {
-      axpby_ssp_pplus (chi, diag[s], phi, upper[s], psi, s, s+1);
-      axpby_ssp_pminus(chi, one, chi, lower[s], psi, s, Ls-1);
-    } else if (s==(Ls-1)) {
-      axpby_ssp_pplus (chi, diag[s], phi, upper[s], psi, s, 0);
-      axpby_ssp_pminus(chi, one, chi, lower[s], psi, s, s-1);
-    } else {
-      axpby_ssp_pplus (chi, diag[s], phi, upper[s], psi, s, s+1);
-      axpby_ssp_pminus(chi, one, chi, lower[s], psi, s, s-1);
-    }
-  }
-}
-
-template<class Impl>
-void DomainWallEOFAFermion<Impl>::MooeeInv(const FermionField& psi, FermionField& chi)
-{
-  Coeff_t one(1.0);
-  Coeff_t czero(0.0);
-  chi.Checkerboard() = psi.Checkerboard();
-  int Ls = this->Ls;
-
-  FermionField tmp(psi.Grid());
-
-  // Apply (L^{\prime})^{-1}
-  axpby_ssp(chi, one, psi, czero, psi, 0, 0);      // chi[0]=psi[0]
-  for(int s=1; s<Ls; s++){
-    axpby_ssp_pplus(chi, one, psi, -this->lee[s-1], chi, s, s-1);// recursion Psi[s] -lee P_+ chi[s-1]
-  }
-
-  // L_m^{-1}
-  for(int s=0; s<Ls-1; s++){ // Chi[ee] = 1 - sum[s<Ls-1] -leem[s]P_- chi
-    axpby_ssp_pminus(chi, one, chi, -this->leem[s], chi, Ls-1, s);
-  }
-
-  // U_m^{-1} D^{-1}
-  for(int s=0; s<Ls-1; s++){
-    axpby_ssp_pplus(chi, one/this->dee[s], chi, -this->ueem[s]/this->dee[Ls], chi, s, Ls-1);
-  }
-  axpby_ssp_pminus(tmp, czero, chi, one/this->dee[Ls-1], chi, Ls-1, Ls-1);
-  axpby_ssp_pplus(chi, one, tmp, one/this->dee[Ls], chi, Ls-1, Ls-1);
-
-  // Apply U^{-1}
-  for(int s=Ls-2; s>=0; s--){
-    axpby_ssp_pminus(chi, one, chi, -this->uee[s], chi, s, s+1);  // chi[Ls]
-  }
-}
-
-template<class Impl>
-void DomainWallEOFAFermion<Impl>::MooeeInvDag(const FermionField& psi, FermionField& chi)
-{
-  Coeff_t one(1.0);
-  Coeff_t czero(0.0);
-  chi.Checkerboard() = psi.Checkerboard();
-  int Ls = this->Ls;
-
-  FermionField tmp(psi.Grid());
-
-  // Apply (U^{\prime})^{-dagger}
-  axpby_ssp(chi, one, psi, czero, psi, 0, 0);      // chi[0]=psi[0]
-  for(int s=1; s<Ls; s++){
-    axpby_ssp_pminus(chi, one, psi, -conjugate(this->uee[s-1]), chi, s, s-1);
-  }
-
-  // U_m^{-\dagger}
-  for(int s=0; s<Ls-1; s++){
-    axpby_ssp_pplus(chi, one, chi, -conjugate(this->ueem[s]), chi, Ls-1, s);
-  }
-
-  // L_m^{-\dagger} D^{-dagger}
-  for(int s=0; s<Ls-1; s++){
-    axpby_ssp_pminus(chi, one/conjugate(this->dee[s]), chi, -conjugate(this->leem[s]/this->dee[Ls-1]), chi, s, Ls-1);
-  }
-  axpby_ssp_pminus(tmp, czero, chi, one/conjugate(this->dee[Ls-1]), chi, Ls-1, Ls-1);
-  axpby_ssp_pplus(chi, one, tmp, one/conjugate(this->dee[Ls]), chi, Ls-1, Ls-1);
-
-  // Apply L^{-dagger}
-  for(int s=Ls-2; s>=0; s--){
-    axpby_ssp_pplus(chi, one, chi, -conjugate(this->lee[s]), chi, s, s+1);  // chi[Ls]
-  }
-}
-
-#ifdef DOMAIN_WALL_EOFA_DPERP_LINALG
-
-INSTANTIATE_DPERP_DWF_EOFA(WilsonImplF);
-INSTANTIATE_DPERP_DWF_EOFA(WilsonImplD);
-INSTANTIATE_DPERP_DWF_EOFA(GparityWilsonImplF);
-INSTANTIATE_DPERP_DWF_EOFA(GparityWilsonImplD);
-INSTANTIATE_DPERP_DWF_EOFA(ZWilsonImplF);
-INSTANTIATE_DPERP_DWF_EOFA(ZWilsonImplD);
-
-INSTANTIATE_DPERP_DWF_EOFA(WilsonImplFH);
-INSTANTIATE_DPERP_DWF_EOFA(WilsonImplDF);
-INSTANTIATE_DPERP_DWF_EOFA(GparityWilsonImplFH);
-INSTANTIATE_DPERP_DWF_EOFA(GparityWilsonImplDF);
-INSTANTIATE_DPERP_DWF_EOFA(ZWilsonImplFH);
-INSTANTIATE_DPERP_DWF_EOFA(ZWilsonImplDF);
-
-#endif
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/deprecated/MobiusEOFAFermiondense.h

@@ -1,183 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/MobiusEOFAFermiondense.cc
-
-Copyright (C) 2017
-
-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: David Murphy <dmurphy@phys.columbia.edu>
-
-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_Eigen_Dense.h>
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/MobiusEOFAFermion.h>
-
-NAMESPACE_BEGIN(Grid);
-
-/*
- * Dense matrix versions of routines
- */
-template<class Impl>
-void MobiusEOFAFermion<Impl>::MooeeInv(const FermionField& psi, FermionField& chi)
-{
-  this->MooeeInternal(psi, chi, DaggerNo, InverseYes);
-}
-
-template<class Impl>
-void MobiusEOFAFermion<Impl>::MooeeInv_shift(const FermionField& psi, FermionField& chi)
-{
-  this->MooeeInternal(psi, chi, DaggerNo, InverseYes);
-}
-
-template<class Impl>
-void MobiusEOFAFermion<Impl>::MooeeInvDag(const FermionField& psi, FermionField& chi)
-{
-  this->MooeeInternal(psi, chi, DaggerYes, InverseYes);
-}
-
-template<class Impl>
-void MobiusEOFAFermion<Impl>::MooeeInvDag_shift(const FermionField& psi, FermionField& chi)
-{
-  this->MooeeInternal(psi, chi, DaggerYes, InverseYes);
-}
-
-template<class Impl>
-void MobiusEOFAFermion<Impl>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv)
-{
-  int Ls = this->Ls;
-  int LLs = psi.Grid()->_rdimensions[0];
-  int vol = psi.Grid()->oSites()/LLs;
-
-  int pm      = this->pm;
-  RealD shift = this->shift;
-  RealD alpha = this->alpha;
-  RealD k     = this->k;
-  RealD mq1   = this->mq1;
-
-  chi.Checkerboard() = psi.Checkerboard();
-
-  assert(Ls==LLs);
-
-  Eigen::MatrixXd Pplus  = Eigen::MatrixXd::Zero(Ls,Ls);
-  Eigen::MatrixXd Pminus = Eigen::MatrixXd::Zero(Ls,Ls);
-
-  for(int s=0;s<Ls;s++){
-    Pplus(s,s)  = this->bee[s];
-    Pminus(s,s) = this->bee[s];
-  }
-
-  for(int s=0; s<Ls-1; s++){
-    Pminus(s,s+1) = -this->cee[s];
-  }
-
-  for(int s=0; s<Ls-1; s++){
-    Pplus(s+1,s) = -this->cee[s+1];
-  }
-  Pplus (0,Ls-1) = mq1*this->cee[0];
-  Pminus(Ls-1,0) = mq1*this->cee[Ls-1];
-
-  if(shift != 0.0){
-    Coeff_t N = 2.0 * ( std::pow(alpha+1.0,Ls) + mq1*std::pow(alpha-1.0,Ls) );
-    for(int s=0; s<Ls; ++s){
-      if(pm == 1){ Pplus(s,Ls-1) += shift * k * N * std::pow(-1.0,s) * std::pow(alpha-1.0,s) / std::pow(alpha+1.0,Ls+s+1); }
-      else{ Pminus(Ls-1-s,Ls-1) -= shift * k * N * std::pow(-1.0,s) * std::pow(alpha-1.0,s) / std::pow(alpha+1.0,Ls+s+1); }
-    }
-  }
-
-  Eigen::MatrixXd PplusMat ;
-  Eigen::MatrixXd PminusMat;
-
-  if(inv){
-    PplusMat  = Pplus.inverse();
-    PminusMat = Pminus.inverse();
-  } else {
-    PplusMat  = Pplus;
-    PminusMat = Pminus;
-  }
-
-  if(dag){
-    PplusMat.adjointInPlace();
-    PminusMat.adjointInPlace();
-  }
-
-  // For the non-vectorised s-direction this is simple
-
-  for(auto site=0; site<vol; site++){
-
-    SiteSpinor     SiteChi;
-    SiteHalfSpinor SitePplus;
-    SiteHalfSpinor SitePminus;
-
-    for(int s1=0; s1<Ls; s1++){
-      SiteChi = Zero();
-      for(int s2=0; s2<Ls; s2++){
-	int lex2 = s2 + Ls*site;
-	if(PplusMat(s1,s2) != 0.0){
-	  spProj5p(SitePplus,psi[lex2]);
-	  accumRecon5p(SiteChi, PplusMat(s1,s2)*SitePplus);
-	}
-	if(PminusMat(s1,s2) != 0.0){
-	  spProj5m(SitePminus, psi[lex2]);
-	  accumRecon5m(SiteChi, PminusMat(s1,s2)*SitePminus);
-	}
-      }
-      chi[s1+Ls*site] = SiteChi*0.5;
-    }
-  }
-}
-
-#ifdef MOBIUS_EOFA_DPERP_DENSE
-
-INSTANTIATE_DPERP_MOBIUS_EOFA(GparityWilsonImplF);
-INSTANTIATE_DPERP_MOBIUS_EOFA(GparityWilsonImplD);
-INSTANTIATE_DPERP_MOBIUS_EOFA(WilsonImplF);
-INSTANTIATE_DPERP_MOBIUS_EOFA(WilsonImplD);
-INSTANTIATE_DPERP_MOBIUS_EOFA(ZWilsonImplF);
-INSTANTIATE_DPERP_MOBIUS_EOFA(ZWilsonImplD);
-
-template void MobiusEOFAFermion<GparityWilsonImplF>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-template void MobiusEOFAFermion<GparityWilsonImplD>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-template void MobiusEOFAFermion<WilsonImplF>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-template void MobiusEOFAFermion<WilsonImplD>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-template void MobiusEOFAFermion<ZWilsonImplF>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-template void MobiusEOFAFermion<ZWilsonImplD>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-
-INSTANTIATE_DPERP_MOBIUS_EOFA(GparityWilsonImplFH);
-INSTANTIATE_DPERP_MOBIUS_EOFA(GparityWilsonImplDF);
-INSTANTIATE_DPERP_MOBIUS_EOFA(WilsonImplFH);
-INSTANTIATE_DPERP_MOBIUS_EOFA(WilsonImplDF);
-INSTANTIATE_DPERP_MOBIUS_EOFA(ZWilsonImplFH);
-INSTANTIATE_DPERP_MOBIUS_EOFA(ZWilsonImplDF);
-
-template void MobiusEOFAFermion<GparityWilsonImplFH>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-template void MobiusEOFAFermion<GparityWilsonImplDF>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-template void MobiusEOFAFermion<WilsonImplFH>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-template void MobiusEOFAFermion<WilsonImplDF>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-template void MobiusEOFAFermion<ZWilsonImplFH>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-template void MobiusEOFAFermion<ZWilsonImplDF>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-
-#endif
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/deprecated/MobiusEOFAFermionssp.h

@@ -1,289 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/MobiusEOFAFermionssp.cc
-
-Copyright (C) 2017
-
-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: David Murphy <dmurphy@phys.columbia.edu>
-
-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/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/MobiusEOFAFermion.h>
-
-NAMESPACE_BEGIN(Grid);
-
-// FIXME -- make a version of these routines with site loop outermost for cache reuse.
-// Pminus fowards
-// Pplus  backwards
-template<class Impl>
-void MobiusEOFAFermion<Impl>::M5D(const FermionField& psi, const FermionField& phi,
-				  FermionField& chi, Vector<Coeff_t>& lower, Vector<Coeff_t>& diag, Vector<Coeff_t>& upper)
-{
-  Coeff_t one(1.0);
-  int Ls = this->Ls;
-  for(int s=0; s<Ls; s++){
-    if(s==0) {
-      axpby_ssp_pminus(chi, diag[s], phi, upper[s], psi, s, s+1);
-      axpby_ssp_pplus (chi, one, chi, lower[s], psi, s, Ls-1);
-    } else if (s==(Ls-1)) {
-      axpby_ssp_pminus(chi, diag[s], phi, upper[s], psi, s, 0);
-      axpby_ssp_pplus (chi, one, chi, lower[s], psi, s, s-1);
-    } else {
-      axpby_ssp_pminus(chi, diag[s], phi, upper[s], psi, s, s+1);
-      axpby_ssp_pplus(chi, one, chi, lower[s], psi, s, s-1);
-    }
-  }
-}
-
-template<class Impl>
-void MobiusEOFAFermion<Impl>::M5D_shift(const FermionField& psi, const FermionField& phi,
-					FermionField& chi, Vector<Coeff_t>& lower, Vector<Coeff_t>& diag, Vector<Coeff_t>& upper,
-					Vector<Coeff_t>& shift_coeffs)
-{
-  Coeff_t one(1.0);
-  int Ls = this->Ls;
-  for(int s=0; s<Ls; s++){
-    if(s==0) {
-      axpby_ssp_pminus(chi, diag[s], phi, upper[s], psi, s, s+1);
-      axpby_ssp_pplus (chi, one, chi, lower[s], psi, s, Ls-1);
-    } else if (s==(Ls-1)) {
-      axpby_ssp_pminus(chi, diag[s], phi, upper[s], psi, s, 0);
-      axpby_ssp_pplus (chi, one, chi, lower[s], psi, s, s-1);
-    } else {
-      axpby_ssp_pminus(chi, diag[s], phi, upper[s], psi, s, s+1);
-      axpby_ssp_pplus(chi, one, chi, lower[s], psi, s, s-1);
-    }
-    if(this->pm == 1){ axpby_ssp_pplus(chi, one, chi, shift_coeffs[s], psi, s, Ls-1); }
-    else{ axpby_ssp_pminus(chi, one, chi, shift_coeffs[s], psi, s, 0); }
-  }
-}
-
-template<class Impl>
-void MobiusEOFAFermion<Impl>::M5Ddag(const FermionField& psi, const FermionField& phi,
-				     FermionField& chi, Vector<Coeff_t>& lower, Vector<Coeff_t>& diag, Vector<Coeff_t>& upper)
-{
-  Coeff_t one(1.0);
-  int Ls = this->Ls;
-  for(int s=0; s<Ls; s++){
-    if(s==0) {
-      axpby_ssp_pplus (chi, diag[s], phi, upper[s], psi, s, s+1);
-      axpby_ssp_pminus(chi, one, chi, lower[s], psi, s, Ls-1);
-    } else if (s==(Ls-1)) {
-      axpby_ssp_pplus (chi, diag[s], phi, upper[s], psi, s, 0);
-      axpby_ssp_pminus(chi, one, chi, lower[s], psi, s, s-1);
-    } else {
-      axpby_ssp_pplus (chi, diag[s], phi, upper[s], psi, s, s+1);
-      axpby_ssp_pminus(chi, one, chi, lower[s], psi, s, s-1);
-    }
-  }
-}
-
-template<class Impl>
-void MobiusEOFAFermion<Impl>::M5Ddag_shift(const FermionField& psi, const FermionField& phi,
-					   FermionField& chi, Vector<Coeff_t>& lower, Vector<Coeff_t>& diag, Vector<Coeff_t>& upper,
-					   Vector<Coeff_t>& shift_coeffs)
-{
-  Coeff_t one(1.0);
-  int Ls = this->Ls;
-  for(int s=0; s<Ls; s++){
-    if(s==0) {
-      axpby_ssp_pplus (chi, diag[s], phi, upper[s], psi, s, s+1);
-      axpby_ssp_pminus(chi, one, chi, lower[s], psi, s, Ls-1);
-    } else if (s==(Ls-1)) {
-      axpby_ssp_pplus (chi, diag[s], phi, upper[s], psi, s, 0);
-      axpby_ssp_pminus(chi, one, chi, lower[s], psi, s, s-1);
-    } else {
-      axpby_ssp_pplus (chi, diag[s], phi, upper[s], psi, s, s+1);
-      axpby_ssp_pminus(chi, one, chi, lower[s], psi, s, s-1);
-    }
-    if(this->pm == 1){ axpby_ssp_pplus(chi, one, chi, shift_coeffs[s], psi, Ls-1, s); }
-    else{ axpby_ssp_pminus(chi, one, chi, shift_coeffs[s], psi, 0, s); }
-  }
-}
-
-template<class Impl>
-void MobiusEOFAFermion<Impl>::MooeeInv(const FermionField& psi, FermionField& chi)
-{
-  if(this->shift != 0.0){ MooeeInv_shift(psi,chi); return; }
-
-  Coeff_t one(1.0);
-  Coeff_t czero(0.0);
-  chi.Checkerboard() = psi.Checkerboard();
-  int Ls = this->Ls;
-
-  // Apply (L^{\prime})^{-1}
-  axpby_ssp(chi, one, psi, czero, psi, 0, 0);      // chi[0]=psi[0]
-  for(int s=1; s<Ls; s++){
-    axpby_ssp_pplus(chi, one, psi, -this->lee[s-1], chi, s, s-1);// recursion Psi[s] -lee P_+ chi[s-1]
-  }
-
-  // L_m^{-1}
-  for(int s=0; s<Ls-1; s++){ // Chi[ee] = 1 - sum[s<Ls-1] -leem[s]P_- chi
-    axpby_ssp_pminus(chi, one, chi, -this->leem[s], chi, Ls-1, s);
-  }
-
-  // U_m^{-1} D^{-1}
-  for(int s=0; s<Ls-1; s++){
-    axpby_ssp_pplus(chi, one/this->dee[s], chi, -this->ueem[s]/this->dee[Ls-1], chi, s, Ls-1);
-  }
-  axpby_ssp(chi, one/this->dee[Ls-1], chi, czero, chi, Ls-1, Ls-1);
-
-  // Apply U^{-1}
-  for(int s=Ls-2; s>=0; s--){
-    axpby_ssp_pminus(chi, one, chi, -this->uee[s], chi, s, s+1);  // chi[Ls]
-  }
-}
-
-template<class Impl>
-void MobiusEOFAFermion<Impl>::MooeeInv_shift(const FermionField& psi, FermionField& chi)
-{
-  Coeff_t one(1.0);
-  Coeff_t czero(0.0);
-  chi.Checkerboard() = psi.Checkerboard();
-  int Ls = this->Ls;
-
-  FermionField tmp(psi.Grid());
-
-  // Apply (L^{\prime})^{-1}
-  axpby_ssp(chi, one, psi, czero, psi, 0, 0);      // chi[0]=psi[0]
-  axpby_ssp(tmp, czero, tmp, this->MooeeInv_shift_lc[0], psi, 0, 0);
-  for(int s=1; s<Ls; s++){
-    axpby_ssp_pplus(chi, one, psi, -this->lee[s-1], chi, s, s-1);// recursion Psi[s] -lee P_+ chi[s-1]
-    axpby_ssp(tmp, one, tmp, this->MooeeInv_shift_lc[s], psi, 0, s);
-  }
-
-  // L_m^{-1}
-  for(int s=0; s<Ls-1; s++){ // Chi[ee] = 1 - sum[s<Ls-1] -leem[s]P_- chi
-    axpby_ssp_pminus(chi, one, chi, -this->leem[s], chi, Ls-1, s);
-  }
-
-  // U_m^{-1} D^{-1}
-  for(int s=0; s<Ls-1; s++){
-    axpby_ssp_pplus(chi, one/this->dee[s], chi, -this->ueem[s]/this->dee[Ls-1], chi, s, Ls-1);
-  }
-  axpby_ssp(chi, one/this->dee[Ls-1], chi, czero, chi, Ls-1, Ls-1);
-
-  // Apply U^{-1} and add shift term
-  if(this->pm == 1){ axpby_ssp_pplus(chi, one, chi, this->MooeeInv_shift_norm[Ls-1], tmp, Ls-1, 0); }
-  else{ axpby_ssp_pminus(chi, one, chi, this->MooeeInv_shift_norm[Ls-1], tmp, Ls-1, 0); }
-  for(int s=Ls-2; s>=0; s--){
-    axpby_ssp_pminus(chi, one, chi, -this->uee[s], chi, s, s+1);  // chi[Ls]
-    if(this->pm == 1){ axpby_ssp_pplus(chi, one, chi, this->MooeeInv_shift_norm[s], tmp, s, 0); }
-    else{ axpby_ssp_pminus(chi, one, chi, this->MooeeInv_shift_norm[s], tmp, s, 0); }
-  }
-}
-
-template<class Impl>
-void MobiusEOFAFermion<Impl>::MooeeInvDag(const FermionField& psi, FermionField& chi)
-{
-  if(this->shift != 0.0){ MooeeInvDag_shift(psi,chi); return; }
-
-  Coeff_t one(1.0);
-  Coeff_t czero(0.0);
-  chi.Checkerboard() = psi.Checkerboard();
-  int Ls = this->Ls;
-
-  // Apply (U^{\prime})^{-dagger}
-  axpby_ssp(chi, one, psi, czero, psi, 0, 0);      // chi[0]=psi[0]
-  for(int s=1; s<Ls; s++){
-    axpby_ssp_pminus(chi, one, psi, -conjugate(this->uee[s-1]), chi, s, s-1);
-  }
-
-  // U_m^{-\dagger}
-  for(int s=0; s<Ls-1; s++){
-    axpby_ssp_pplus(chi, one, chi, -conjugate(this->ueem[s]), chi, Ls-1, s);
-  }
-
-  // L_m^{-\dagger} D^{-dagger}
-  for(int s=0; s<Ls-1; s++){
-    axpby_ssp_pminus(chi, one/conjugate(this->dee[s]), chi, -conjugate(this->leem[s]/this->dee[Ls-1]), chi, s, Ls-1);
-  }
-  axpby_ssp(chi, one/conjugate(this->dee[Ls-1]), chi, czero, chi, Ls-1, Ls-1);
-
-  // Apply L^{-dagger}
-  for(int s=Ls-2; s>=0; s--){
-    axpby_ssp_pplus(chi, one, chi, -conjugate(this->lee[s]), chi, s, s+1);  // chi[Ls]
-  }
-}
-
-template<class Impl>
-void MobiusEOFAFermion<Impl>::MooeeInvDag_shift(const FermionField& psi, FermionField& chi)
-{
-  Coeff_t one(1.0);
-  Coeff_t czero(0.0);
-  chi.Checkerboard() = psi.Checkerboard();
-  int Ls = this->Ls;
-
-  FermionField tmp(psi.Grid());
-
-  // Apply (U^{\prime})^{-dagger} and accumulate (MooeeInvDag_shift_lc)_{j} \psi_{j} in tmp[0]
-  axpby_ssp(chi, one, psi, czero, psi, 0, 0);      // chi[0]=psi[0]
-  axpby_ssp(tmp, czero, tmp, this->MooeeInvDag_shift_lc[0], psi, 0, 0);
-  for(int s=1; s<Ls; s++){
-    axpby_ssp_pminus(chi, one, psi, -conjugate(this->uee[s-1]), chi, s, s-1);
-    axpby_ssp(tmp, one, tmp, this->MooeeInvDag_shift_lc[s], psi, 0, s);
-  }
-
-  // U_m^{-\dagger}
-  for(int s=0; s<Ls-1; s++){
-    axpby_ssp_pplus(chi, one, chi, -conjugate(this->ueem[s]), chi, Ls-1, s);
-  }
-
-  // L_m^{-\dagger} D^{-dagger}
-  for(int s=0; s<Ls-1; s++){
-    axpby_ssp_pminus(chi, one/conjugate(this->dee[s]), chi, -conjugate(this->leem[s]/this->dee[Ls-1]), chi, s, Ls-1);
-  }
-  axpby_ssp(chi, one/conjugate(this->dee[Ls-1]), chi, czero, chi, Ls-1, Ls-1);
-
-  // Apply L^{-dagger} and add shift
-  if(this->pm == 1){ axpby_ssp_pplus(chi, one, chi, this->MooeeInvDag_shift_norm[Ls-1], tmp, Ls-1, 0); }
-  else{ axpby_ssp_pminus(chi, one, chi, this->MooeeInvDag_shift_norm[Ls-1], tmp, Ls-1, 0); }
-  for(int s=Ls-2; s>=0; s--){
-    axpby_ssp_pplus(chi, one, chi, -conjugate(this->lee[s]), chi, s, s+1);  // chi[Ls]
-    if(this->pm == 1){ axpby_ssp_pplus(chi, one, chi, this->MooeeInvDag_shift_norm[s], tmp, s, 0); }
-    else{ axpby_ssp_pminus(chi, one, chi, this->MooeeInvDag_shift_norm[s], tmp, s, 0); }
-  }
-}
-
-#ifdef MOBIUS_EOFA_DPERP_LINALG
-
-INSTANTIATE_DPERP_MOBIUS_EOFA(WilsonImplF);
-INSTANTIATE_DPERP_MOBIUS_EOFA(WilsonImplD);
-INSTANTIATE_DPERP_MOBIUS_EOFA(GparityWilsonImplF);
-INSTANTIATE_DPERP_MOBIUS_EOFA(GparityWilsonImplD);
-INSTANTIATE_DPERP_MOBIUS_EOFA(ZWilsonImplF);
-INSTANTIATE_DPERP_MOBIUS_EOFA(ZWilsonImplD);
-
-INSTANTIATE_DPERP_MOBIUS_EOFA(WilsonImplFH);
-INSTANTIATE_DPERP_MOBIUS_EOFA(WilsonImplDF);
-INSTANTIATE_DPERP_MOBIUS_EOFA(GparityWilsonImplFH);
-INSTANTIATE_DPERP_MOBIUS_EOFA(GparityWilsonImplDF);
-INSTANTIATE_DPERP_MOBIUS_EOFA(ZWilsonImplFH);
-INSTANTIATE_DPERP_MOBIUS_EOFA(ZWilsonImplDF);
-
-#endif
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/implementation/CayleyFermion5Dcache.h

@@ -1,235 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/CayleyFermion5D.cc
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is 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/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/CayleyFermion5D.h>
-
-
-NAMESPACE_BEGIN(Grid);
-
-// Pminus fowards
-// Pplus  backwards..
-template<class Impl>  
-void
-CayleyFermion5D<Impl>::M5D(const FermionField &psi_i,
-			       const FermionField &phi_i, 
-			       FermionField &chi_i,
-			       Vector<Coeff_t> &lower,
-			       Vector<Coeff_t> &diag,
-			       Vector<Coeff_t> &upper)
-{
-  
-  chi_i.Checkerboard()=psi_i.Checkerboard();
-  GridBase *grid=psi_i.Grid();
-  auto psi = psi_i.View();
-  auto phi = phi_i.View();
-  auto chi = chi_i.View();
-  assert(phi.Checkerboard() == psi.Checkerboard());
-
-  int Ls =this->Ls;
-
-  // 10 = 3 complex mult + 2 complex add
-  // Flops = 10.0*(Nc*Ns) *Ls*vol (/2 for red black counting)
-  M5Dcalls++;
-  M5Dtime-=usecond();
-
-  uint64_t nloop = grid->oSites()/Ls;
-  accelerator_for(sss,nloop,Simd::Nsimd(),{
-    uint64_t ss= sss*Ls;
-    typedef decltype(coalescedRead(psi[0])) spinor;
-    spinor tmp1, 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],diag[s]*phi(ss+s)+upper[s]*tmp1+lower[s]*tmp2);
-    }
-  });
-  M5Dtime+=usecond();
-}
-
-template<class Impl>  
-void
-CayleyFermion5D<Impl>::M5Ddag(const FermionField &psi_i,
-			      const FermionField &phi_i, 
-			      FermionField &chi_i,
-			      Vector<Coeff_t> &lower,
-			      Vector<Coeff_t> &diag,
-			      Vector<Coeff_t> &upper)
-{
-  chi_i.Checkerboard()=psi_i.Checkerboard();
-  GridBase *grid=psi_i.Grid();
-  auto psi = psi_i.View();
-  auto phi = phi_i.View();
-  auto chi = chi_i.View();
-  assert(phi.Checkerboard() == psi.Checkerboard());
-
-  int Ls=this->Ls;
-
-  // Flops = 6.0*(Nc*Ns) *Ls*vol
-  M5Dcalls++;
-  M5Dtime-=usecond();
-
-  uint64_t nloop = grid->oSites()/Ls;
-  accelerator_for(sss,nloop,Simd::Nsimd(),{
-    uint64_t ss=sss*Ls;
-    typedef decltype(coalescedRead(psi[0])) spinor;
-    spinor tmp1,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],diag[s]*phi(ss+s)+upper[s]*tmp1+lower[s]*tmp2);
-    }
-  });
-  M5Dtime+=usecond();
-}
-
-template<class Impl>
-void
-CayleyFermion5D<Impl>::MooeeInv    (const FermionField &psi_i, FermionField &chi_i)
-{
-  chi_i.Checkerboard()=psi_i.Checkerboard();
-  GridBase *grid=psi_i.Grid();
-
-  auto psi = psi_i.View();
-  auto chi = chi_i.View();
-
-  int Ls=this->Ls;
-
-  auto plee  = & lee [0];
-  auto pdee  = & dee [0];
-  auto puee  = & uee [0];
-  auto pleem = & leem[0];
-  auto pueem = & ueem[0];
-
-  MooeeInvCalls++;
-  MooeeInvTime-=usecond();
-  uint64_t nloop = grid->oSites()/Ls;
-  accelerator_for(sss,nloop,Simd::Nsimd(),{
-    uint64_t ss=sss*Ls;
-    typedef decltype(coalescedRead(psi[0])) spinor;
-    spinor tmp;
-
-    // flops = 12*2*Ls + 12*2*Ls + 3*12*Ls + 12*2*Ls  = 12*Ls * (9) = 108*Ls flops
-    // Apply (L^{\prime})^{-1}
-    coalescedWrite(chi[ss],psi(ss)); // chi[0]=psi[0]
-    for(int s=1;s<Ls;s++){
-      spProj5p(tmp,chi(ss+s-1));  
-      coalescedWrite(chi[ss+s] , psi(ss+s)-plee[s-1]*tmp);
-    }
-
-    // L_m^{-1} 
-    for (int s=0;s<Ls-1;s++){ // Chi[ee] = 1 - sum[s<Ls-1] -pleem[s]P_- chi
-      spProj5m(tmp,chi(ss+s));    
-      coalescedWrite(chi[ss+Ls-1], chi(ss+Ls-1) - pleem[s]*tmp);
-    }
-
-    // U_m^{-1} D^{-1}
-    for (int s=0;s<Ls-1;s++){
-      // Chi[s] + 1/d chi[s] 
-      spProj5p(tmp,chi(ss+Ls-1)); 
-      coalescedWrite(chi[ss+s], (1.0/pdee[s])*chi(ss+s)-(pueem[s]/pdee[Ls-1])*tmp);
-    }	
-    coalescedWrite(chi[ss+Ls-1], (1.0/pdee[Ls-1])*chi(ss+Ls-1));
-      
-    // Apply U^{-1}
-    for (int s=Ls-2;s>=0;s--){
-      spProj5m(tmp,chi(ss+s+1));  
-      coalescedWrite(chi[ss+s], chi(ss+s) - puee[s]*tmp);
-    }
-  });
-
-  MooeeInvTime+=usecond();
-
-}
-
-template<class Impl>
-void
-CayleyFermion5D<Impl>::MooeeInvDag (const FermionField &psi_i, FermionField &chi_i)
-{
-  chi_i.Checkerboard()=psi_i.Checkerboard();
-  GridBase *grid=psi_i.Grid();
-  int Ls=this->Ls;
-
-  auto psi = psi_i.View();
-  auto chi = chi_i.View();
-
-  auto plee  = & lee [0];
-  auto pdee  = & dee [0];
-  auto puee  = & uee [0];
-  auto pleem = & leem[0];
-  auto pueem = & ueem[0];
-
-  assert(psi.Checkerboard() == psi.Checkerboard());
-
-  MooeeInvCalls++;
-  MooeeInvTime-=usecond();
-
-
-  uint64_t nloop = grid->oSites()/Ls;
-  accelerator_for(sss,nloop,Simd::Nsimd(),{
-    uint64_t ss=sss*Ls;
-    typedef decltype(coalescedRead(psi[0])) spinor;
-    spinor tmp;
-
-    // Apply (U^{\prime})^{-dagger}
-    coalescedWrite(chi[ss],psi(ss));
-    for (int s=1;s<Ls;s++){
-      spProj5m(tmp,chi(ss+s-1));
-      coalescedWrite(chi[ss+s], psi(ss+s)-conjugate(puee[s-1])*tmp);
-    }
-    // U_m^{-\dagger} 
-    for (int s=0;s<Ls-1;s++){
-      spProj5p(tmp,chi(ss+s));
-      coalescedWrite(chi[ss+Ls-1], chi(ss+Ls-1) - conjugate(pueem[s])*tmp);
-    }
-
-    // L_m^{-\dagger} D^{-dagger}
-    for (int s=0;s<Ls-1;s++){
-      spProj5m(tmp,chi(ss+Ls-1));
-      coalescedWrite(chi[ss+s], conjugate(1.0/pdee[s])*chi(ss+s)-conjugate(pleem[s]/pdee[Ls-1])*tmp);
-    }	
-    coalescedWrite(chi[ss+Ls-1], conjugate(1.0/pdee[Ls-1])*chi(ss+Ls-1));
-  
-    // Apply L^{-dagger}
-    for (int s=Ls-2;s>=0;s--){
-      spProj5p(tmp,chi(ss+s+1));
-      coalescedWrite(chi[ss+s], chi(ss+s) - conjugate(plee[s])*tmp);
-    }
-  });
-  MooeeInvTime+=usecond();
-
-}
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/implementation/CayleyFermion5Dvec.h

@@ -1,831 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/CayleyFermion5D.cc
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is 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/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/CayleyFermion5D.h>
-
-NAMESPACE_BEGIN(Grid);
-
-/*
- * Dense matrix versions of routines
- */
-template<class Impl>
-void
-CayleyFermion5D<Impl>::MooeeInvDag(const FermionField &psi, FermionField &chi)
-{
-  EnableIf<Impl::LsVectorised&&EnableBool,int> sfinae=0;
-  this->MooeeInternal(psi,chi,DaggerYes,InverseYes);
-}
-  
-template<class Impl>
-void
-CayleyFermion5D<Impl>::MooeeInv(const FermionField &psi, FermionField &chi)
-{
-  EnableIf<Impl::LsVectorised&&EnableBool,int> sfinae=0;
-  this->MooeeInternal(psi,chi,DaggerNo,InverseYes);
-}
-template<class Impl>  
-void
-CayleyFermion5D<Impl>::M5D(const FermionField &psi_i,
-			   const FermionField &phi_i, 
-			   FermionField &chi_i,
-			   Vector<Coeff_t> &lower,
-			   Vector<Coeff_t> &diag,
-			   Vector<Coeff_t> &upper)
-{
-  EnableIf<Impl::LsVectorised&&EnableBool,int> sfinae=0;
-  chi_i.Checkerboard()=psi_i.Checkerboard();
-  GridBase *grid=psi_i.Grid();
-  auto psi = psi_i.View();
-  auto phi = phi_i.View();
-  auto chi = chi_i.View();
-  int Ls   = this->Ls;
-  int LLs  = grid->_rdimensions[0];
-  const int nsimd= Simd::Nsimd();
-
-  Vector<iSinglet<Simd> > u(LLs);
-  Vector<iSinglet<Simd> > l(LLs);
-  Vector<iSinglet<Simd> > d(LLs);
-
-  assert(Ls/LLs==nsimd);
-  assert(phi.Checkerboard() == psi.Checkerboard());
-
-  // just directly address via type pun
-  typedef typename Simd::scalar_type scalar_type;
-  scalar_type * u_p = (scalar_type *)&u[0];
-  scalar_type * l_p = (scalar_type *)&l[0];
-  scalar_type * d_p = (scalar_type *)&d[0];
-
-  for(int o=0;o<LLs;o++){ // outer
-    for(int i=0;i<nsimd;i++){ //inner
-      int s  = o+i*LLs;
-      int ss = o*nsimd+i;
-      u_p[ss] = upper[s];
-      l_p[ss] = lower[s];
-      d_p[ss] = diag[s];
-    }}
-
-
-  M5Dcalls++;
-  M5Dtime-=usecond();
-
-  assert(Nc==3);
-
-  thread_loop( (int ss=0;ss<grid->oSites();ss+=LLs),{ // adds LLs
-#if 0
-    alignas(64) SiteHalfSpinor hp;
-    alignas(64) SiteHalfSpinor hm;
-    alignas(64) SiteSpinor fp;
-    alignas(64) SiteSpinor fm;
-
-    for(int v=0;v<LLs;v++){
-
-      int vp=(v+1)%LLs;
-      int vm=(v+LLs-1)%LLs;
-
-      spProj5m(hp,psi[ss+vp]);
-      spProj5p(hm,psi[ss+vm]);
-
-      if ( vp<=v ) rotate(hp,hp,1);
-      if ( vm>=v ) rotate(hm,hm,nsimd-1);
-	
-      hp=0.5*hp;
-      hm=0.5*hm;
-
-      spRecon5m(fp,hp);
-      spRecon5p(fm,hm);
-
-      chi[ss+v] = d[v]*phi[ss+v];
-      chi[ss+v] = chi[ss+v]     +u[v]*fp;
-      chi[ss+v] = chi[ss+v]     +l[v]*fm;
-
-    }
-#else
-    for(int v=0;v<LLs;v++){
-      
-      vprefetch(psi[ss+v+LLs]);
-
-      int vp= (v==LLs-1) ? 0     : v+1;
-      int vm= (v==0    ) ? LLs-1 : v-1;
-	
-      Simd hp_00 = psi[ss+vp]()(2)(0); 
-      Simd hp_01 = psi[ss+vp]()(2)(1); 
-      Simd hp_02 = psi[ss+vp]()(2)(2); 
-      Simd hp_10 = psi[ss+vp]()(3)(0); 
-      Simd hp_11 = psi[ss+vp]()(3)(1); 
-      Simd hp_12 = psi[ss+vp]()(3)(2); 
-	
-      Simd hm_00 = psi[ss+vm]()(0)(0); 
-      Simd hm_01 = psi[ss+vm]()(0)(1); 
-      Simd hm_02 = psi[ss+vm]()(0)(2); 
-      Simd hm_10 = psi[ss+vm]()(1)(0); 
-      Simd hm_11 = psi[ss+vm]()(1)(1); 
-      Simd hm_12 = psi[ss+vm]()(1)(2); 
-
-      if ( vp<=v ) {
-	hp_00.v = Optimization::Rotate::tRotate<2>(hp_00.v);
-	hp_01.v = Optimization::Rotate::tRotate<2>(hp_01.v);
-	hp_02.v = Optimization::Rotate::tRotate<2>(hp_02.v);
-	hp_10.v = Optimization::Rotate::tRotate<2>(hp_10.v);
-	hp_11.v = Optimization::Rotate::tRotate<2>(hp_11.v);
-	hp_12.v = Optimization::Rotate::tRotate<2>(hp_12.v);
-      }
-      if ( vm>=v ) {
-	hm_00.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_00.v);
-	hm_01.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_01.v);
-	hm_02.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_02.v);
-	hm_10.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_10.v);
-	hm_11.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_11.v);
-	hm_12.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_12.v);
-      }
-
-      // Can force these to real arithmetic and save 2x.
-      Simd p_00  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(0)(0))  + switcheroo<Coeff_t>::mult(l[v]()()(),hm_00); 
-      Simd p_01  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(0)(1))  + switcheroo<Coeff_t>::mult(l[v]()()(),hm_01); 
-      Simd p_02  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(0)(2))  + switcheroo<Coeff_t>::mult(l[v]()()(),hm_02); 
-      Simd p_10  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(1)(0))  + switcheroo<Coeff_t>::mult(l[v]()()(),hm_10); 
-      Simd p_11  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(1)(1))  + switcheroo<Coeff_t>::mult(l[v]()()(),hm_11); 
-      Simd p_12  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(1)(2))  + switcheroo<Coeff_t>::mult(l[v]()()(),hm_12); 
-      Simd p_20  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(2)(0))  + switcheroo<Coeff_t>::mult(u[v]()()(),hp_00); 
-      Simd p_21  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(2)(1))  + switcheroo<Coeff_t>::mult(u[v]()()(),hp_01); 
-      Simd p_22  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(2)(2))  + switcheroo<Coeff_t>::mult(u[v]()()(),hp_02);  
-      Simd p_30  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(3)(0))  + switcheroo<Coeff_t>::mult(u[v]()()(),hp_10); 
-      Simd p_31  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(3)(1))  + switcheroo<Coeff_t>::mult(u[v]()()(),hp_11); 
-      Simd p_32  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(3)(2))  + switcheroo<Coeff_t>::mult(u[v]()()(),hp_12); 
-
-      vstream(chi[ss+v]()(0)(0),p_00);
-      vstream(chi[ss+v]()(0)(1),p_01);
-      vstream(chi[ss+v]()(0)(2),p_02);
-      vstream(chi[ss+v]()(1)(0),p_10);
-      vstream(chi[ss+v]()(1)(1),p_11);
-      vstream(chi[ss+v]()(1)(2),p_12);
-      vstream(chi[ss+v]()(2)(0),p_20);
-      vstream(chi[ss+v]()(2)(1),p_21);
-      vstream(chi[ss+v]()(2)(2),p_22);
-      vstream(chi[ss+v]()(3)(0),p_30);
-      vstream(chi[ss+v]()(3)(1),p_31);
-      vstream(chi[ss+v]()(3)(2),p_32);
-
-    }
-#endif
-  });
-  M5Dtime+=usecond();
-}
-
-template<class Impl>  
-void
-CayleyFermion5D<Impl>::M5Ddag(const FermionField &psi_i,
-			      const FermionField &phi_i, 
-			      FermionField &chi_i,
-			      Vector<Coeff_t> &lower,
-			      Vector<Coeff_t> &diag,
-			      Vector<Coeff_t> &upper)
-{
-  EnableIf<Impl::LsVectorised&&EnableBool,int> sfinae=0;
-  chi_i.Checkerboard()=psi_i.Checkerboard();
-  GridBase *grid=psi_i.Grid();
-  auto psi=psi_i.View();
-  auto phi=phi_i.View();
-  auto chi=chi_i.View();
-  int Ls   = this->Ls;
-  int LLs  = grid->_rdimensions[0];
-  int nsimd= Simd::Nsimd();
-
-  Vector<iSinglet<Simd> > u(LLs);
-  Vector<iSinglet<Simd> > l(LLs);
-  Vector<iSinglet<Simd> > d(LLs);
-
-  assert(Ls/LLs==nsimd);
-  assert(phi.Checkerboard() == psi.Checkerboard());
-
-  // just directly address via type pun
-  typedef typename Simd::scalar_type scalar_type;
-  scalar_type * u_p = (scalar_type *)&u[0];
-  scalar_type * l_p = (scalar_type *)&l[0];
-  scalar_type * d_p = (scalar_type *)&d[0];
-
-  for(int o=0;o<LLs;o++){ // outer
-    for(int i=0;i<nsimd;i++){ //inner
-      int s  = o+i*LLs;
-      int ss = o*nsimd+i;
-      u_p[ss] = upper[s];
-      l_p[ss] = lower[s];
-      d_p[ss] = diag[s];
-    }}
-
-  M5Dcalls++;
-  M5Dtime-=usecond();
-  thread_loop( (int ss=0;ss<grid->oSites();ss+=LLs),{ // adds LLs
-#if 0
-    alignas(64) SiteHalfSpinor hp;
-    alignas(64) SiteHalfSpinor hm;
-    alignas(64) SiteSpinor fp;
-    alignas(64) SiteSpinor fm;
-
-    for(int v=0;v<LLs;v++){
-
-      int vp=(v+1)%LLs;
-      int vm=(v+LLs-1)%LLs;
-
-      spProj5p(hp,psi[ss+vp]);
-      spProj5m(hm,psi[ss+vm]);
-
-      if ( vp<=v ) rotate(hp,hp,1);
-      if ( vm>=v ) rotate(hm,hm,nsimd-1);
-      
-      hp=hp*0.5;
-      hm=hm*0.5;
-      spRecon5p(fp,hp);
-      spRecon5m(fm,hm);
-
-      chi[ss+v] = d[v]*phi[ss+v]+u[v]*fp;
-      chi[ss+v] = chi[ss+v]     +l[v]*fm;
-
-    }
-#else
-    for(int v=0;v<LLs;v++){
-
-      vprefetch(psi[ss+v+LLs]);
-
-      int vp= (v==LLs-1) ? 0     : v+1;
-      int vm= (v==0    ) ? LLs-1 : v-1;
-	
-      Simd hp_00 = psi[ss+vp]()(0)(0); 
-      Simd hp_01 = psi[ss+vp]()(0)(1); 
-      Simd hp_02 = psi[ss+vp]()(0)(2); 
-      Simd hp_10 = psi[ss+vp]()(1)(0); 
-      Simd hp_11 = psi[ss+vp]()(1)(1); 
-      Simd hp_12 = psi[ss+vp]()(1)(2); 
-	
-      Simd hm_00 = psi[ss+vm]()(2)(0); 
-      Simd hm_01 = psi[ss+vm]()(2)(1); 
-      Simd hm_02 = psi[ss+vm]()(2)(2); 
-      Simd hm_10 = psi[ss+vm]()(3)(0); 
-      Simd hm_11 = psi[ss+vm]()(3)(1); 
-      Simd hm_12 = psi[ss+vm]()(3)(2); 
-
-      if ( vp<=v ) {
-	hp_00.v = Optimization::Rotate::tRotate<2>(hp_00.v);
-	hp_01.v = Optimization::Rotate::tRotate<2>(hp_01.v);
-	hp_02.v = Optimization::Rotate::tRotate<2>(hp_02.v);
-	hp_10.v = Optimization::Rotate::tRotate<2>(hp_10.v);
-	hp_11.v = Optimization::Rotate::tRotate<2>(hp_11.v);
-	hp_12.v = Optimization::Rotate::tRotate<2>(hp_12.v);
-      }
-      if ( vm>=v ) {
-	hm_00.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_00.v);
-	hm_01.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_01.v);
-	hm_02.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_02.v);
-	hm_10.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_10.v);
-	hm_11.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_11.v);
-	hm_12.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_12.v);
-      }
-
-      Simd p_00  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(0)(0))  + switcheroo<Coeff_t>::mult(u[v]()()(),hp_00); 
-      Simd p_01  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(0)(1))  + switcheroo<Coeff_t>::mult(u[v]()()(),hp_01); 
-      Simd p_02  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(0)(2))  + switcheroo<Coeff_t>::mult(u[v]()()(),hp_02); 
-      Simd p_10  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(1)(0))  + switcheroo<Coeff_t>::mult(u[v]()()(),hp_10); 
-      Simd p_11  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(1)(1))  + switcheroo<Coeff_t>::mult(u[v]()()(),hp_11); 
-      Simd p_12  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(1)(2))  + switcheroo<Coeff_t>::mult(u[v]()()(),hp_12); 
-
-      Simd p_20  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(2)(0))  + switcheroo<Coeff_t>::mult(l[v]()()(),hm_00); 
-      Simd p_21  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(2)(1))  + switcheroo<Coeff_t>::mult(l[v]()()(),hm_01); 
-      Simd p_22  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(2)(2))  + switcheroo<Coeff_t>::mult(l[v]()()(),hm_02);  
-      Simd p_30  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(3)(0))  + switcheroo<Coeff_t>::mult(l[v]()()(),hm_10); 
-      Simd p_31  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(3)(1))  + switcheroo<Coeff_t>::mult(l[v]()()(),hm_11); 
-      Simd p_32  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(3)(2))  + switcheroo<Coeff_t>::mult(l[v]()()(),hm_12); 
-
-      vstream(chi[ss+v]()(0)(0),p_00);
-      vstream(chi[ss+v]()(0)(1),p_01);
-      vstream(chi[ss+v]()(0)(2),p_02);
-      vstream(chi[ss+v]()(1)(0),p_10);
-      vstream(chi[ss+v]()(1)(1),p_11);
-      vstream(chi[ss+v]()(1)(2),p_12);
-      vstream(chi[ss+v]()(2)(0),p_20);
-      vstream(chi[ss+v]()(2)(1),p_21);
-      vstream(chi[ss+v]()(2)(2),p_22);
-      vstream(chi[ss+v]()(3)(0),p_30);
-      vstream(chi[ss+v]()(3)(1),p_31);
-      vstream(chi[ss+v]()(3)(2),p_32);
-    }
-#endif
-  });
-  M5Dtime+=usecond();
-}
-
-
-#ifdef AVX512 
-#include <simd/Intel512common.h>
-#include <simd/Intel512avx.h>
-#include <simd/Intel512single.h>
-#endif 
-
-template<class Impl>
-void
-CayleyFermion5D<Impl>::MooeeInternalAsm(const FermionField &psi_i, FermionField &chi_i,
-					int LLs, int site,
-					Vector<iSinglet<Simd> > &Matp,
-					Vector<iSinglet<Simd> > &Matm)
-{
-  EnableIf<Impl::LsVectorised&&EnableBool,int> sfinae=0;
-  auto psi = psi_i.View();
-  auto chi = chi_i.View();
-#ifndef AVX512
-  {
-    SiteHalfSpinor BcastP;
-    SiteHalfSpinor BcastM;
-    SiteHalfSpinor SiteChiP;
-    SiteHalfSpinor SiteChiM;
-
-    // Ls*Ls * 2 * 12 * vol flops
-    for(int s1=0;s1<LLs;s1++){ 
-      for(int s2=0;s2<LLs;s2++){ 
-	for(int  l=0; l<Simd::Nsimd();l++){ // simd lane
-
-	  int s=s2+l*LLs;
-	  int lex=s2+LLs*site;
-	
-	  if ( s2==0 && l==0) {
-	    SiteChiP=Zero();
-	    SiteChiM=Zero();
-	  }
-	
-	  for(int sp=0;sp<2;sp++){
-	    for(int co=0;co<Nc;co++){
-	      vbroadcast(BcastP()(sp  )(co),psi[lex]()(sp)(co),l);
-	    }}
-	  for(int sp=0;sp<2;sp++){
-	    for(int co=0;co<Nc;co++){
-	      vbroadcast(BcastM()(sp  )(co),psi[lex]()(sp+2)(co),l);
-	    }}
-
-	  for(int sp=0;sp<2;sp++){
-	    for(int co=0;co<Nc;co++){
-	      SiteChiP()(sp)(co)=real_madd(Matp[LLs*s+s1]()()(),BcastP()(sp)(co),SiteChiP()(sp)(co)); // 1100 us.
-	      SiteChiM()(sp)(co)=real_madd(Matm[LLs*s+s1]()()(),BcastM()(sp)(co),SiteChiM()(sp)(co)); // each found by commenting out
-	    }}
-
-	}}
-      {
-	int lex = s1+LLs*site;
-	for(int sp=0;sp<2;sp++){
-	  for(int co=0;co<Nc;co++){
-	    vstream(chi[lex]()(sp)(co), SiteChiP()(sp)(co));
-	    vstream(chi[lex]()(sp+2)(co), SiteChiM()(sp)(co));
-	  }}
-      }
-    }
-
-  }
-#else
-  {
-    // pointers
-    //  MASK_REGS;
-#define Chi_00 %%zmm1
-#define Chi_01 %%zmm2
-#define Chi_02 %%zmm3
-#define Chi_10 %%zmm4
-#define Chi_11 %%zmm5
-#define Chi_12 %%zmm6
-#define Chi_20 %%zmm7
-#define Chi_21 %%zmm8
-#define Chi_22 %%zmm9
-#define Chi_30 %%zmm10
-#define Chi_31 %%zmm11
-#define Chi_32 %%zmm12
-
-#define BCAST0   %%zmm13
-#define BCAST1   %%zmm14
-#define BCAST2   %%zmm15
-#define BCAST3   %%zmm16
-#define BCAST4   %%zmm17
-#define BCAST5   %%zmm18
-#define BCAST6   %%zmm19
-#define BCAST7   %%zmm20
-#define BCAST8   %%zmm21
-#define BCAST9   %%zmm22
-#define BCAST10  %%zmm23
-#define BCAST11  %%zmm24
-
-    int incr=LLs*LLs*sizeof(iSinglet<Simd>);
-    for(int s1=0;s1<LLs;s1++){ 
-      for(int s2=0;s2<LLs;s2++){ 
-	int lex=s2+LLs*site;
-	uint64_t a0 = (uint64_t)&Matp[LLs*s2+s1]; // should be cacheable
-	uint64_t a1 = (uint64_t)&Matm[LLs*s2+s1];
-	uint64_t a2 = (uint64_t)&psi[lex];
-	for(int  l=0; l<Simd::Nsimd();l++){ // simd lane
-	  if ( (s2+l)==0 ) {
-	    asm (
-		 VPREFETCH1(0,%2)  	     VPREFETCH1(0,%1)
-		 VPREFETCH1(12,%2)  	     VPREFETCH1(13,%2)
-		 VPREFETCH1(14,%2)  	     VPREFETCH1(15,%2)         
-		 VBCASTCDUP(0,%2,BCAST0)   
-		 VBCASTCDUP(1,%2,BCAST1)   
-		 VBCASTCDUP(2,%2,BCAST2)   
-		 VBCASTCDUP(3,%2,BCAST3)   
-		 VBCASTCDUP(4,%2,BCAST4)     VMULMEM (0,%0,BCAST0,Chi_00)
-		 VBCASTCDUP(5,%2,BCAST5)     VMULMEM (0,%0,BCAST1,Chi_01)
-		 VBCASTCDUP(6,%2,BCAST6)     VMULMEM (0,%0,BCAST2,Chi_02)
-		 VBCASTCDUP(7,%2,BCAST7)     VMULMEM (0,%0,BCAST3,Chi_10)
-		 VBCASTCDUP(8,%2,BCAST8)     VMULMEM (0,%0,BCAST4,Chi_11)
-		 VBCASTCDUP(9,%2,BCAST9)     VMULMEM (0,%0,BCAST5,Chi_12)
-		 VBCASTCDUP(10,%2,BCAST10)   VMULMEM (0,%1,BCAST6,Chi_20)
-		 VBCASTCDUP(11,%2,BCAST11)   VMULMEM (0,%1,BCAST7,Chi_21)
-		 VMULMEM (0,%1,BCAST8,Chi_22)         
-		 VMULMEM (0,%1,BCAST9,Chi_30)
-		 VMULMEM (0,%1,BCAST10,Chi_31)       
-		 VMULMEM (0,%1,BCAST11,Chi_32)
-		 : : "r" (a0), "r" (a1), "r" (a2)  );
-	  } else { 
-	    asm (
-		 VBCASTCDUP(0,%2,BCAST0)   VMADDMEM (0,%0,BCAST0,Chi_00)
-		 VBCASTCDUP(1,%2,BCAST1)   VMADDMEM (0,%0,BCAST1,Chi_01)
-		 VBCASTCDUP(2,%2,BCAST2)   VMADDMEM (0,%0,BCAST2,Chi_02)
-		 VBCASTCDUP(3,%2,BCAST3)   VMADDMEM (0,%0,BCAST3,Chi_10)
-		 VBCASTCDUP(4,%2,BCAST4)   VMADDMEM (0,%0,BCAST4,Chi_11)
-		 VBCASTCDUP(5,%2,BCAST5)   VMADDMEM (0,%0,BCAST5,Chi_12)
-		 VBCASTCDUP(6,%2,BCAST6)   VMADDMEM (0,%1,BCAST6,Chi_20)
-		 VBCASTCDUP(7,%2,BCAST7)   VMADDMEM (0,%1,BCAST7,Chi_21)
-		 VBCASTCDUP(8,%2,BCAST8)   VMADDMEM (0,%1,BCAST8,Chi_22)
-		 VBCASTCDUP(9,%2,BCAST9)   VMADDMEM (0,%1,BCAST9,Chi_30)
-		 VBCASTCDUP(10,%2,BCAST10)  VMADDMEM (0,%1,BCAST10,Chi_31)
-		 VBCASTCDUP(11,%2,BCAST11)  VMADDMEM (0,%1,BCAST11,Chi_32) 
-		 : : "r" (a0), "r" (a1), "r" (a2)  );
-	  }
-	  a0 = a0+incr;
-	  a1 = a1+incr;
-	a2 = a2+sizeof(typename Simd::scalar_type);
-	}}
-      {
-	int lexa = s1+LLs*site;
-	asm (
-	     VSTORE(0,%0,Chi_00) VSTORE(1 ,%0,Chi_01)  VSTORE(2 ,%0,Chi_02)		
-	     VSTORE(3,%0,Chi_10) VSTORE(4 ,%0,Chi_11)  VSTORE(5 ,%0,Chi_12)		
-	     VSTORE(6,%0,Chi_20) VSTORE(7 ,%0,Chi_21)  VSTORE(8 ,%0,Chi_22)		
-	     VSTORE(9,%0,Chi_30) VSTORE(10,%0,Chi_31)  VSTORE(11,%0,Chi_32)		
-	     : : "r" ((uint64_t)&chi[lexa]) : "memory" );
-
-      }
-    }
-  }
-#undef Chi_00
-#undef Chi_01
-#undef Chi_02
-#undef Chi_10
-#undef Chi_11
-#undef Chi_12
-#undef Chi_20
-#undef Chi_21
-#undef Chi_22
-#undef Chi_30
-#undef Chi_31
-#undef Chi_32
-
-#undef BCAST0
-#undef BCAST1
-#undef BCAST2
-#undef BCAST3
-#undef BCAST4
-#undef BCAST5
-#undef BCAST6
-#undef BCAST7
-#undef BCAST8
-#undef BCAST9
-#undef BCAST10
-#undef BCAST11
-#endif
-};
-
-// Z-mobius version
-template<class Impl>
-void
-CayleyFermion5D<Impl>::MooeeInternalZAsm(const FermionField &psi_i, FermionField &chi_i,
-					 int LLs, int site, Vector<iSinglet<Simd> > &Matp, Vector<iSinglet<Simd> > &Matm)
-{
-  EnableIf<Impl::LsVectorised,int> sfinae=0;
-#ifndef AVX512
-  {
-    auto psi = psi_i.View();
-    auto chi = chi_i.View();
-
-    SiteHalfSpinor BcastP;
-    SiteHalfSpinor BcastM;
-    SiteHalfSpinor SiteChiP;
-    SiteHalfSpinor SiteChiM;
-
-    // Ls*Ls * 2 * 12 * vol flops
-    for(int s1=0;s1<LLs;s1++){ 
-      for(int s2=0;s2<LLs;s2++){ 
-	for(int  l=0; l<Simd::Nsimd();l++){ // simd lane
-
-	  int s=s2+l*LLs;
-	  int lex=s2+LLs*site;
-	
-	  if ( s2==0 && l==0) {
-	    SiteChiP=Zero();
-	    SiteChiM=Zero();
-	  }
-	
-	  for(int sp=0;sp<2;sp++){
-	    for(int co=0;co<Nc;co++){
-	      vbroadcast(BcastP()(sp  )(co),psi[lex]()(sp)(co),l);
-	    }}
-	  for(int sp=0;sp<2;sp++){
-	    for(int co=0;co<Nc;co++){
-	      vbroadcast(BcastM()(sp  )(co),psi[lex]()(sp+2)(co),l);
-	    }}
-
-	  for(int sp=0;sp<2;sp++){
-	    for(int co=0;co<Nc;co++){
-	      SiteChiP()(sp)(co)=SiteChiP()(sp)(co)+ Matp[LLs*s+s1]()()()*BcastP()(sp)(co); 
-	      SiteChiM()(sp)(co)=SiteChiM()(sp)(co)+ Matm[LLs*s+s1]()()()*BcastM()(sp)(co); 
-	    }}
-
-
-	}}
-      {
-	int lex = s1+LLs*site;
-	for(int sp=0;sp<2;sp++){
-	  for(int co=0;co<Nc;co++){
-	    vstream(chi[lex]()(sp)(co), SiteChiP()(sp)(co));
-	    vstream(chi[lex]()(sp+2)(co), SiteChiM()(sp)(co));
-	  }}
-      }
-    }
-
-  }
-#else
-  {
-    auto psi = psi_i.View();
-    auto chi = chi_i.View();
-    // pointers
-    //  MASK_REGS;
-#define Chi_00 %zmm0
-#define Chi_01 %zmm1
-#define Chi_02 %zmm2
-#define Chi_10 %zmm3
-#define Chi_11 %zmm4
-#define Chi_12 %zmm5
-#define Chi_20 %zmm6
-#define Chi_21 %zmm7
-#define Chi_22 %zmm8
-#define Chi_30 %zmm9
-#define Chi_31 %zmm10
-#define Chi_32 %zmm11
-#define pChi_00 %%zmm0
-#define pChi_01 %%zmm1
-#define pChi_02 %%zmm2
-#define pChi_10 %%zmm3
-#define pChi_11 %%zmm4
-#define pChi_12 %%zmm5
-#define pChi_20 %%zmm6
-#define pChi_21 %%zmm7
-#define pChi_22 %%zmm8
-#define pChi_30 %%zmm9
-#define pChi_31 %%zmm10
-#define pChi_32 %%zmm11
-
-#define BCAST_00   %zmm12
-#define  SHUF_00   %zmm13
-#define BCAST_01   %zmm14
-#define  SHUF_01   %zmm15
-#define BCAST_02   %zmm16
-#define  SHUF_02   %zmm17
-#define BCAST_10   %zmm18
-#define  SHUF_10   %zmm19
-#define BCAST_11   %zmm20
-#define  SHUF_11   %zmm21
-#define BCAST_12   %zmm22
-#define  SHUF_12   %zmm23
-
-#define Mp  %zmm24
-#define Mps %zmm25
-#define Mm  %zmm26
-#define Mms %zmm27
-#define N 8
-    int incr=LLs*LLs*sizeof(iSinglet<Simd>);
-    for(int s1=0;s1<LLs;s1++){ 
-      for(int s2=0;s2<LLs;s2++){ 
-	int lex=s2+LLs*site;
-	uint64_t a0 = (uint64_t)&Matp[LLs*s2+s1]; // should be cacheable
-	uint64_t a1 = (uint64_t)&Matm[LLs*s2+s1];
-	uint64_t a2 = (uint64_t)&psi[lex];
-	for(int  l=0; l<Simd::Nsimd();l++){ // simd lane
-	  if ( (s2+l)==0 ) {
-	    LOAD64(%r8,a0);
-	    LOAD64(%r9,a1);
-	    LOAD64(%r10,a2);
-	    asm (
-		 VLOAD(0,%r8,Mp)// i r
-		 VLOAD(0,%r9,Mm)
-		 VSHUF(Mp,Mps)  // r i 
-		 VSHUF(Mm,Mms)
-		 VPREFETCH1(12,%r10)  	     VPREFETCH1(13,%r10)
-		 VPREFETCH1(14,%r10)  	     VPREFETCH1(15,%r10)         
-
-		 VMULIDUP(0*N,%r10,Mps,Chi_00)
-		 VMULIDUP(1*N,%r10,Mps,Chi_01)
-		 VMULIDUP(2*N,%r10,Mps,Chi_02)
-		 VMULIDUP(3*N,%r10,Mps,Chi_10)
-		 VMULIDUP(4*N,%r10,Mps,Chi_11)
-		 VMULIDUP(5*N,%r10,Mps,Chi_12)
-
-		 VMULIDUP(6*N ,%r10,Mms,Chi_20)
-		 VMULIDUP(7*N ,%r10,Mms,Chi_21)
-		 VMULIDUP(8*N ,%r10,Mms,Chi_22)
-		 VMULIDUP(9*N ,%r10,Mms,Chi_30)
-		 VMULIDUP(10*N,%r10,Mms,Chi_31)
-		 VMULIDUP(11*N,%r10,Mms,Chi_32)
-
-		 VMADDSUBRDUP(0*N,%r10,Mp,Chi_00)
-		 VMADDSUBRDUP(1*N,%r10,Mp,Chi_01)
-		 VMADDSUBRDUP(2*N,%r10,Mp,Chi_02)
-		 VMADDSUBRDUP(3*N,%r10,Mp,Chi_10)
-		 VMADDSUBRDUP(4*N,%r10,Mp,Chi_11)
-		 VMADDSUBRDUP(5*N,%r10,Mp,Chi_12)
-
-		 VMADDSUBRDUP(6*N ,%r10,Mm,Chi_20)
-		 VMADDSUBRDUP(7*N ,%r10,Mm,Chi_21)
-		 VMADDSUBRDUP(8*N ,%r10,Mm,Chi_22)
-		 VMADDSUBRDUP(9*N ,%r10,Mm,Chi_30)
-		 VMADDSUBRDUP(10*N,%r10,Mm,Chi_31)
-		 VMADDSUBRDUP(11*N,%r10,Mm,Chi_32)
-		 );
-	  } else { 
-	    LOAD64(%r8,a0);
-	    LOAD64(%r9,a1);
-	    LOAD64(%r10,a2);
-	    asm (
-		 VLOAD(0,%r8,Mp)
-		 VSHUF(Mp,Mps)
-
-		 VLOAD(0,%r9,Mm)
-		 VSHUF(Mm,Mms)
-
-		 VMADDSUBIDUP(0*N,%r10,Mps,Chi_00) //  Mri * Pii +- Cir
-		 VMADDSUBIDUP(1*N,%r10,Mps,Chi_01)
-		 VMADDSUBIDUP(2*N,%r10,Mps,Chi_02)
-		 VMADDSUBIDUP(3*N,%r10,Mps,Chi_10)
-		 VMADDSUBIDUP(4*N,%r10,Mps,Chi_11)
-		 VMADDSUBIDUP(5*N,%r10,Mps,Chi_12)
-
-		 VMADDSUBIDUP(6 *N,%r10,Mms,Chi_20)
-		 VMADDSUBIDUP(7 *N,%r10,Mms,Chi_21)
-		 VMADDSUBIDUP(8 *N,%r10,Mms,Chi_22)
-		 VMADDSUBIDUP(9 *N,%r10,Mms,Chi_30)
-		 VMADDSUBIDUP(10*N,%r10,Mms,Chi_31)
-		 VMADDSUBIDUP(11*N,%r10,Mms,Chi_32)
-
-		 VMADDSUBRDUP(0*N,%r10,Mp,Chi_00) //  Cir = Mir * Prr +- ( Mri * Pii +- Cir) 
-		 VMADDSUBRDUP(1*N,%r10,Mp,Chi_01) //  Ci = MiPr + Ci + MrPi ;    Cr = MrPr - ( MiPi - Cr)
-		 VMADDSUBRDUP(2*N,%r10,Mp,Chi_02)
-		 VMADDSUBRDUP(3*N,%r10,Mp,Chi_10)
-		 VMADDSUBRDUP(4*N,%r10,Mp,Chi_11)
-		 VMADDSUBRDUP(5*N,%r10,Mp,Chi_12)
-
-		 VMADDSUBRDUP(6 *N,%r10,Mm,Chi_20)
-		 VMADDSUBRDUP(7 *N,%r10,Mm,Chi_21)
-		 VMADDSUBRDUP(8 *N,%r10,Mm,Chi_22)
-		 VMADDSUBRDUP(9 *N,%r10,Mm,Chi_30)
-		 VMADDSUBRDUP(10*N,%r10,Mm,Chi_31)
-		 VMADDSUBRDUP(11*N,%r10,Mm,Chi_32)
-		 );
-	  }
-	  a0 = a0+incr;
-	  a1 = a1+incr;
-	a2 = a2+sizeof(typename Simd::scalar_type);
-	}}
-      {
-	int lexa = s1+LLs*site;
-	/*
-	  SiteSpinor tmp;
-	  asm (
-	  VSTORE(0,%0,pChi_00) VSTORE(1 ,%0,pChi_01)  VSTORE(2 ,%0,pChi_02)		
-	  VSTORE(3,%0,pChi_10) VSTORE(4 ,%0,pChi_11)  VSTORE(5 ,%0,pChi_12)		
-	  VSTORE(6,%0,pChi_20) VSTORE(7 ,%0,pChi_21)  VSTORE(8 ,%0,pChi_22)		
-	  VSTORE(9,%0,pChi_30) VSTORE(10,%0,pChi_31)  VSTORE(11,%0,pChi_32)		
-	  : : "r" ((uint64_t)&tmp) : "memory" );
-	*/
-
-	asm (
-	     VSTORE(0,%0,pChi_00) VSTORE(1 ,%0,pChi_01)  VSTORE(2 ,%0,pChi_02)		
-	     VSTORE(3,%0,pChi_10) VSTORE(4 ,%0,pChi_11)  VSTORE(5 ,%0,pChi_12)		
-	     VSTORE(6,%0,pChi_20) VSTORE(7 ,%0,pChi_21)  VSTORE(8 ,%0,pChi_22)		
-	     VSTORE(9,%0,pChi_30) VSTORE(10,%0,pChi_31)  VSTORE(11,%0,pChi_32)		
-	     : : "r" ((uint64_t)&chi[lexa]) : "memory" );
-
-	//      if ( 1 || (site==0) ) { 
-	//	std::cout<<site << " s1 "<<s1<<"\n\t"<<tmp << "\n't" << chi[lexa] <<"\n\t"<<tmp-chi[lexa]<<std::endl;
-	//      }
-      }
-    }
-  }
-#undef Chi_00
-#undef Chi_01
-#undef Chi_02
-#undef Chi_10
-#undef Chi_11
-#undef Chi_12
-#undef Chi_20
-#undef Chi_21
-#undef Chi_22
-#undef Chi_30
-#undef Chi_31
-#undef Chi_32
-
-#undef BCAST0
-#undef BCAST1
-#undef BCAST2
-#undef BCAST3
-#undef BCAST4
-#undef BCAST5
-#undef BCAST6
-#undef BCAST7
-#undef BCAST8
-#undef BCAST9
-#undef BCAST10
-#undef BCAST11
-
-#endif
-};
-
-
-template<class Impl>
-void
-CayleyFermion5D<Impl>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv)
-{
-  EnableIf<Impl::LsVectorised,int> sfinae=0;
-  chi.Checkerboard()=psi.Checkerboard();
-
-  int Ls=this->Ls;
-  int LLs = psi.Grid()->_rdimensions[0];
-  int vol = psi.Grid()->oSites()/LLs;
-
-  
-  Vector<iSinglet<Simd> >  Matp;
-  Vector<iSinglet<Simd> >  Matm;
-  Vector<iSinglet<Simd> >  *_Matp;
-  Vector<iSinglet<Simd> >  *_Matm;
-  
-  //  MooeeInternalCompute(dag,inv,Matp,Matm);
-  if ( inv && dag ) { 
-    _Matp = &MatpInvDag;
-    _Matm = &MatmInvDag;
-  }
-  if ( inv && (!dag) ) { 
-    _Matp = &MatpInv;
-    _Matm = &MatmInv;
-  } 
-  if ( !inv ) {
-    MooeeInternalCompute(dag,inv,Matp,Matm);
-    _Matp = &Matp;
-    _Matm = &Matm;
-  }
-  assert(_Matp->size()==Ls*LLs);
-
-  MooeeInvCalls++;
-  MooeeInvTime-=usecond();
-
-  if ( switcheroo<Coeff_t>::iscomplex() ) {
-    thread_loop( (auto site=0;site<vol;site++),{
-      MooeeInternalZAsm(psi,chi,LLs,site,*_Matp,*_Matm);
-    });
-  } else { 
-    thread_loop( (auto site=0;site<vol;site++),{
-      MooeeInternalAsm(psi,chi,LLs,site,*_Matp,*_Matm);
-    });
-  }
-  MooeeInvTime+=usecond();
-}
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/implementation/ContinuedFractionFermion5DImplementation.h

@@ -1,321 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/ContinuedFractionFermion5D.cc
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-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 */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/ContinuedFractionFermion5D.h>
-
-#pragma once
-
-NAMESPACE_BEGIN(Grid);
-
-template<class Impl>
-void ContinuedFractionFermion5D<Impl>::SetCoefficientsTanh(Approx::zolotarev_data *zdata,RealD scale)
-{
-  SetCoefficientsZolotarev(1.0/scale,zdata);
-}
-template<class Impl>
-void ContinuedFractionFermion5D<Impl>::SetCoefficientsZolotarev(RealD zolo_hi,Approx::zolotarev_data *zdata)
-{
-  // How to check Ls matches??
-  //      std::cout<<GridLogMessage << Ls << " Ls"<<std::endl;
-  //      std::cout<<GridLogMessage << zdata->n  << " - n"<<std::endl;
-  //      std::cout<<GridLogMessage << zdata->da << " -da "<<std::endl;
-  //      std::cout<<GridLogMessage << zdata->db << " -db"<<std::endl;
-  //      std::cout<<GridLogMessage << zdata->dn << " -dn"<<std::endl;
-  //      std::cout<<GridLogMessage << zdata->dd << " -dd"<<std::endl;
-  int Ls = this->Ls;
-  assert(zdata->db==Ls);// Beta has Ls coeffs
-
-  R=(1+this->mass)/(1-this->mass);
-
-  Beta.resize(Ls);
-  cc.resize(Ls);
-  cc_d.resize(Ls);
-  sqrt_cc.resize(Ls);
-  for(int i=0; i < Ls ; i++){
-    Beta[i] = zdata -> beta[i];
-    cc[i] = 1.0/Beta[i];
-    cc_d[i]=std::sqrt(cc[i]);
-  }
-    
-  cc_d[Ls-1]=1.0;
-  for(int i=0; i < Ls-1 ; i++){
-    sqrt_cc[i]= std::sqrt(cc[i]*cc[i+1]);
-  }    
-  sqrt_cc[Ls-2]=std::sqrt(cc[Ls-2]);
-
-
-  ZoloHiInv =1.0/zolo_hi;
-  dw_diag = (4.0-this->M5)*ZoloHiInv;
-    
-  See.resize(Ls);
-  Aee.resize(Ls);
-  int sign=1;
-  for(int s=0;s<Ls;s++){
-    Aee[s] = sign * Beta[s] * dw_diag;
-    sign   = - sign;
-  }
-  Aee[Ls-1] += R;
-    
-  See[0] = Aee[0];
-  for(int s=1;s<Ls;s++){
-    See[s] = Aee[s] - 1.0/See[s-1];
-  }
-  for(int s=0;s<Ls;s++){
-    std::cout<<GridLogMessage <<"s = "<<s<<" Beta "<<Beta[s]<<" Aee "<<Aee[s] <<" See "<<See[s] <<std::endl;
-  }
-}
-
-
-
-template<class Impl>
-RealD  ContinuedFractionFermion5D<Impl>::M           (const FermionField &psi, FermionField &chi)
-{
-  int Ls = this->Ls;
-
-  FermionField D(psi.Grid());
-
-  this->DW(psi,D,DaggerNo); 
-
-  int sign=1;
-  for(int s=0;s<Ls;s++){
-    if ( s==0 ) {
-      ag5xpby_ssp(chi,cc[0]*Beta[0]*sign*ZoloHiInv,D,sqrt_cc[0],psi,s,s+1); // Multiplies Dw by G5 so Hw
-    } else if ( s==(Ls-1) ){
-      RealD R=(1.0+mass)/(1.0-mass);
-      ag5xpby_ssp(chi,Beta[s]*ZoloHiInv,D,sqrt_cc[s-1],psi,s,s-1);
-      ag5xpby_ssp(chi,R,psi,1.0,chi,s,s);
-    } else {
-      ag5xpby_ssp(chi,cc[s]*Beta[s]*sign*ZoloHiInv,D,sqrt_cc[s],psi,s,s+1);
-      axpby_ssp(chi,1.0,chi,sqrt_cc[s-1],psi,s,s-1);
-    }
-    sign=-sign; 
-  }
-  return norm2(chi);
-}
-template<class Impl>
-RealD  ContinuedFractionFermion5D<Impl>::Mdag        (const FermionField &psi, FermionField &chi)
-{
-  // This matrix is already hermitian. (g5 Dw) = Dw dag g5 = (g5 Dw)dag
-  // The rest of matrix is symmetric.
-  // Can ignore "dag"
-  return M(psi,chi);
-}
-template<class Impl>
-void  ContinuedFractionFermion5D<Impl>::Mdir (const FermionField &psi, FermionField &chi,int dir,int disp){
-  int Ls = this->Ls;
-
-  this->DhopDir(psi,chi,dir,disp); // Dslash on diagonal. g5 Dslash is hermitian
-
-  int sign=1;
-  for(int s=0;s<Ls;s++){
-    if ( s==(Ls-1) ){
-      ag5xpby_ssp(chi,Beta[s]*ZoloHiInv,chi,0.0,chi,s,s);
-    } else {
-      ag5xpby_ssp(chi,cc[s]*Beta[s]*sign*ZoloHiInv,chi,0.0,chi,s,s);
-    }
-    sign=-sign; 
-  }
-}
-template<class Impl>
-void   ContinuedFractionFermion5D<Impl>::Meooe       (const FermionField &psi, FermionField &chi)
-{
-  int Ls = this->Ls;
-
-  // Apply 4d dslash
-  if ( psi.Checkerboard() == Odd ) {
-    this->DhopEO(psi,chi,DaggerNo); // Dslash on diagonal. g5 Dslash is hermitian
-  } else {
-    this->DhopOE(psi,chi,DaggerNo); // Dslash on diagonal. g5 Dslash is hermitian
-  }
-      
-  int sign=1;
-  for(int s=0;s<Ls;s++){
-    if ( s==(Ls-1) ){
-      ag5xpby_ssp(chi,Beta[s]*ZoloHiInv,chi,0.0,chi,s,s);
-    } else {
-      ag5xpby_ssp(chi,cc[s]*Beta[s]*sign*ZoloHiInv,chi,0.0,chi,s,s);
-    }
-    sign=-sign; 
-  }
-}
-template<class Impl>
-void   ContinuedFractionFermion5D<Impl>::MeooeDag    (const FermionField &psi, FermionField &chi)
-{
-  this->Meooe(psi,chi);
-}
-template<class Impl>
-void   ContinuedFractionFermion5D<Impl>::Mooee       (const FermionField &psi, FermionField &chi)
-{
-  int Ls = this->Ls;
-
-  int sign=1;
-  for(int s=0;s<Ls;s++){
-    if ( s==0 ) {
-      ag5xpby_ssp(chi,cc[0]*Beta[0]*sign*dw_diag,psi,sqrt_cc[0],psi,s,s+1); // Multiplies Dw by G5 so Hw
-    } else if ( s==(Ls-1) ){
-      // Drop the CC here.
-      double R=(1+mass)/(1-mass);
-      ag5xpby_ssp(chi,Beta[s]*dw_diag,psi,sqrt_cc[s-1],psi,s,s-1);
-      ag5xpby_ssp(chi,R,psi,1.0,chi,s,s);
-    } else {
-      ag5xpby_ssp(chi,cc[s]*Beta[s]*sign*dw_diag,psi,sqrt_cc[s],psi,s,s+1);
-      axpby_ssp(chi,1.0,chi,sqrt_cc[s-1],psi,s,s-1);
-    }
-    sign=-sign; 
-  }
-}
-
-template<class Impl>
-void   ContinuedFractionFermion5D<Impl>::MooeeDag    (const FermionField &psi, FermionField &chi)
-{
-  this->Mooee(psi,chi);
-}
-template<class Impl>
-void   ContinuedFractionFermion5D<Impl>::MooeeInv    (const FermionField &psi, FermionField &chi)
-{
-  int Ls = this->Ls;
-
-  // Apply Linv
-  axpby_ssp(chi,1.0/cc_d[0],psi,0.0,psi,0,0); 
-  for(int s=1;s<Ls;s++){
-    axpbg5y_ssp(chi,1.0/cc_d[s],psi,-1.0/See[s-1],chi,s,s-1);
-  }
-  // Apply Dinv
-  for(int s=0;s<Ls;s++){
-    ag5xpby_ssp(chi,1.0/See[s],chi,0.0,chi,s,s); //only appearance of See[0]
-  }
-  // Apply Uinv = (Linv)^T
-  axpby_ssp(chi,1.0/cc_d[Ls-1],chi,0.0,chi,Ls-1,Ls-1);
-  for(int s=Ls-2;s>=0;s--){
-    axpbg5y_ssp(chi,1.0/cc_d[s],chi,-1.0*cc_d[s+1]/See[s]/cc_d[s],chi,s,s+1);
-  }
-}
-template<class Impl>
-void   ContinuedFractionFermion5D<Impl>::MooeeInvDag (const FermionField &psi, FermionField &chi)
-{
-  this->MooeeInv(psi,chi);
-}
-
-// force terms; five routines; default to Dhop on diagonal
-template<class Impl>
-void ContinuedFractionFermion5D<Impl>::MDeriv  (GaugeField &mat,const FermionField &U,const FermionField &V,int dag)
-{
-  int Ls = this->Ls;
-
-  FermionField D(V.Grid());
-
-  int sign=1;
-  for(int s=0;s<Ls;s++){
-    if ( s==(Ls-1) ){
-      ag5xpby_ssp(D,Beta[s]*ZoloHiInv,U,0.0,U,s,s);
-    } else {
-      ag5xpby_ssp(D,cc[s]*Beta[s]*sign*ZoloHiInv,U,0.0,U,s,s);
-    }
-    sign=-sign; 
-  }
-  this->DhopDeriv(mat,D,V,DaggerNo); 
-};
-template<class Impl>
-void ContinuedFractionFermion5D<Impl>::MoeDeriv(GaugeField &mat,const FermionField &U,const FermionField &V,int dag)
-{
-  int Ls = this->Ls;
-
-  FermionField D(V.Grid());
-
-  int sign=1;
-  for(int s=0;s<Ls;s++){
-    if ( s==(Ls-1) ){
-      ag5xpby_ssp(D,Beta[s]*ZoloHiInv,U,0.0,U,s,s);
-    } else {
-      ag5xpby_ssp(D,cc[s]*Beta[s]*sign*ZoloHiInv,U,0.0,U,s,s);
-    }
-    sign=-sign; 
-  }
-  this->DhopDerivOE(mat,D,V,DaggerNo); 
-};
-template<class Impl>
-void ContinuedFractionFermion5D<Impl>::MeoDeriv(GaugeField &mat,const FermionField &U,const FermionField &V,int dag)
-{
-  int Ls = this->Ls;
-
-  FermionField D(V.Grid());
-
-  int sign=1;
-  for(int s=0;s<Ls;s++){
-    if ( s==(Ls-1) ){
-      ag5xpby_ssp(D,Beta[s]*ZoloHiInv,U,0.0,U,s,s);
-    } else {
-      ag5xpby_ssp(D,cc[s]*Beta[s]*sign*ZoloHiInv,U,0.0,U,s,s);
-    }
-    sign=-sign; 
-  }
-  this->DhopDerivEO(mat,D,V,DaggerNo); 
-};
-    
-// Constructors
-template<class Impl>
-ContinuedFractionFermion5D<Impl>::ContinuedFractionFermion5D(
-							     GaugeField &_Umu,
-							     GridCartesian         &FiveDimGrid,
-							     GridRedBlackCartesian &FiveDimRedBlackGrid,
-							     GridCartesian         &FourDimGrid,
-							     GridRedBlackCartesian &FourDimRedBlackGrid,
-							     RealD _mass,RealD M5,const ImplParams &p) :
-  WilsonFermion5D<Impl>(_Umu,
-			FiveDimGrid, FiveDimRedBlackGrid,
-			FourDimGrid, FourDimRedBlackGrid,M5,p),
-  mass(_mass)
-{
-  int Ls = this->Ls;
-  assert((Ls&0x1)==1); // Odd Ls required
-}
-
-    template<class Impl>
-    void ContinuedFractionFermion5D<Impl>::ExportPhysicalFermionSolution(const FermionField &solution5d,FermionField &exported4d)
-    {
-      int Ls = this->Ls;
-      conformable(solution5d.Grid(),this->FermionGrid());
-      conformable(exported4d.Grid(),this->GaugeGrid());
-      ExtractSlice(exported4d, solution5d, Ls-1, Ls-1);
-    }
-    template<class Impl>
-    void ContinuedFractionFermion5D<Impl>::ImportPhysicalFermionSource(const FermionField &input4d,FermionField &imported5d)
-    {
-      int Ls = this->Ls;
-      conformable(imported5d.Grid(),this->FermionGrid());
-      conformable(input4d.Grid()   ,this->GaugeGrid());
-      FermionField tmp(this->FermionGrid());
-      tmp=Zero();
-      InsertSlice(input4d, tmp, Ls-1, Ls-1);
-      tmp=Gamma(Gamma::Algebra::Gamma5)*tmp;
-      this->Dminus(tmp,imported5d);
-    }
-
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionCache.h

@@ -1,227 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/DomainWallEOFAFermioncache.cc
-
-Copyright (C) 2017
-
-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: David Murphy <dmurphy@phys.columbia.edu>
-
-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/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/DomainWallEOFAFermion.h>
-
-NAMESPACE_BEGIN(Grid);
-
-// FIXME -- make a version of these routines with site loop outermost for cache reuse.
-// Pminus fowards
-// Pplus  backwards..
-template<class Impl>
-void DomainWallEOFAFermion<Impl>::M5D(const FermionField& psi_i, const FermionField& phi_i,FermionField& chi_i, 
-				      Vector<Coeff_t>& lower, Vector<Coeff_t>& diag, Vector<Coeff_t>& upper)
-{
-  chi_i.Checkerboard() = psi_i.Checkerboard();
-  int Ls = this->Ls;
-  GridBase* grid = psi_i.Grid();
-  auto phi = phi_i.View();
-  auto psi = psi_i.View();
-  auto chi = chi_i.View();
-  assert(phi.Checkerboard() == psi.Checkerboard());
-  // Flops = 6.0*(Nc*Ns) *Ls*vol
-  this->M5Dcalls++;
-  this->M5Dtime -= usecond();
-  
-  auto nloop=grid->oSites()/Ls;
-  accelerator_for(sss,nloop,Simd::Nsimd(),{
-    auto ss=sss*Ls;
-    typedef decltype(coalescedRead(psi[0])) spinor;
-    for(int s=0; s<Ls; s++){
-      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], diag[s]*phi(ss+s) + upper[s]*tmp1 + lower[s]*tmp2);
-    }
-  });
-
-  this->M5Dtime += usecond();
-}
-
-template<class Impl>
-void DomainWallEOFAFermion<Impl>::M5Ddag(const FermionField& psi_i, const FermionField& phi_i, FermionField& chi_i, 
-					 Vector<Coeff_t>& lower, Vector<Coeff_t>& diag, Vector<Coeff_t>& upper)
-{
-  chi_i.Checkerboard() = psi_i.Checkerboard();
-  GridBase* grid = psi_i.Grid();
-  int Ls = this->Ls;
-
-  auto psi = psi_i.View();
-  auto phi = phi_i.View();
-  auto chi = chi_i.View();
-  assert(phi.Checkerboard() == psi.Checkerboard());
-
-  // Flops = 6.0*(Nc*Ns) *Ls*vol
-  this->M5Dcalls++;
-  this->M5Dtime -= usecond();
-
-  auto nloop=grid->oSites()/Ls;
-  accelerator_for(sss,nloop,Simd::Nsimd(),{
-    typedef decltype(coalescedRead(psi[0])) spinor;
-    auto ss=sss*Ls;
-    for(int s=0; s<Ls; s++){
-      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], diag[s]*phi(ss+s) + upper[s]*tmp1 + lower[s]*tmp2);
-    }
-  });
-
-  this->M5Dtime += usecond();
-}
-
-template<class Impl>
-void DomainWallEOFAFermion<Impl>::MooeeInv(const FermionField& psi_i, FermionField& chi_i)
-{
-  chi_i.Checkerboard() = psi_i.Checkerboard();
-  GridBase* grid = psi_i.Grid();
-  auto psi=psi_i.View();
-  auto chi=chi_i.View();
-  int Ls = this->Ls;
-
-  auto plee  = & this->lee[0];
-  auto pdee  = & this->dee[0];
-  auto puee  = & this->uee[0];
-
-  auto pleem = & this->leem[0];
-  auto pueem = & this->ueem[0];
-
-  this->MooeeInvCalls++;
-  this->MooeeInvTime -= usecond();
-  uint64_t nloop=grid->oSites()/Ls;
-  accelerator_for(sss,nloop,Simd::Nsimd(),{
-    auto ss=sss*Ls;
-    typedef decltype(coalescedRead(psi[0])) spinor;
-    spinor tmp1,tmp2;
-
-    // flops = 12*2*Ls + 12*2*Ls + 3*12*Ls + 12*2*Ls  = 12*Ls * (9) = 108*Ls flops
-    // Apply (L^{\prime})^{-1}
-    coalescedWrite(chi[ss],psi(ss)); // chi[0]=psi[0]
-    for(int s=1; s<Ls; s++){
-      spProj5p(tmp1, chi(ss+s-1));
-      coalescedWrite(chi[ss+s], psi(ss+s) - plee[s-1]*tmp1);
-    }
-
-    // L_m^{-1}
-    for(int s=0; s<Ls-1; s++){ // Chi[ee] = 1 - sum[s<Ls-1] -leem[s]P_- chi
-      spProj5m(tmp1, chi(ss+s));
-      coalescedWrite(chi[ss+Ls-1], chi(ss+Ls-1) - pleem[s]*tmp1);
-    }
-
-    // U_m^{-1} D^{-1}
-    for(int s=0; s<Ls-1; s++){ // Chi[s] + 1/d chi[s]
-      spProj5p(tmp1, chi(ss+Ls-1));
-      coalescedWrite(chi[ss+s], (1.0/pdee[s])*chi(ss+s) - (pueem[s]/pdee[Ls])*tmp1);
-    }
-    spProj5m(tmp2, chi(ss+Ls-1));
-    coalescedWrite(chi[ss+Ls-1],(1.0/pdee[Ls])*tmp1 + (1.0/pdee[Ls-1])*tmp2);
-
-    // Apply U^{-1}
-    for(int s=Ls-2; s>=0; s--){
-      spProj5m(tmp1, chi(ss+s+1));
-      coalescedWrite(chi[ss+s], chi(ss+s) - puee[s]*tmp1);
-    }
-  });
-  this->MooeeInvTime += usecond();
-}
-
-template<class Impl>
-void DomainWallEOFAFermion<Impl>::MooeeInvDag(const FermionField& psi_i, FermionField& chi_i)
-{
-  chi_i.Checkerboard() = psi_i.Checkerboard();
-  GridBase* grid = psi_i.Grid();
-  auto psi = psi_i.View();
-  auto chi = chi_i.View();
-  int Ls = this->Ls;
-
-  assert(psi.Checkerboard() == psi.Checkerboard());
-
-  Vector<Coeff_t> ueec(Ls);
-  Vector<Coeff_t> deec(Ls+1);
-  Vector<Coeff_t> leec(Ls);
-  Vector<Coeff_t> ueemc(Ls);
-  Vector<Coeff_t> leemc(Ls);
-
-  for(int s=0; s<ueec.size(); s++){
-    ueec[s]  = conjugate(this->uee[s]);
-    deec[s]  = conjugate(this->dee[s]);
-    leec[s]  = conjugate(this->lee[s]);
-    ueemc[s] = conjugate(this->ueem[s]);
-    leemc[s] = conjugate(this->leem[s]);
-  }
-  deec[Ls] = conjugate(this->dee[Ls]);
-
-  this->MooeeInvCalls++;
-  this->MooeeInvTime -= usecond();
-  auto nloop = grid->oSites()/Ls;
-  accelerator_for(sss,nloop,Simd::Nsimd(),{
-    typedef decltype(coalescedRead(psi[0])) spinor;
-    spinor tmp1,tmp2;
-    auto ss=sss*Ls;
-
-    // Apply (U^{\prime})^{-dagger}
-    coalescedWrite(chi[ss], psi(ss));
-    for(int s=1; s<Ls; s++){
-      spProj5m(tmp1, chi(ss+s-1));
-      coalescedWrite(chi[ss+s], psi(ss+s) - ueec[s-1]*tmp1);
-    }
-
-    // U_m^{-\dagger}
-    for(int s=0; s<Ls-1; s++){
-      spProj5p(tmp1, chi(ss+s));
-      coalescedWrite(chi[ss+Ls-1], chi(ss+Ls-1) - ueemc[s]*tmp1);
-    }
-
-    // L_m^{-\dagger} D^{-dagger}
-    for(int s=0; s<Ls-1; s++){
-      spProj5m(tmp1, chi(ss+Ls-1));
-      coalescedWrite(chi[ss+s] ,(1.0/deec[s])*chi(ss+s) - (leemc[s]/deec[Ls-1])*tmp1);
-    }
-    spProj5p(tmp2, chi(ss+Ls-1));
-    coalescedWrite(chi[ss+Ls-1], (1.0/deec[Ls-1])*tmp1 + (1.0/deec[Ls])*tmp2);
-
-    // Apply L^{-dagger}
-    for(int s=Ls-2; s>=0; s--){
-      spProj5p(tmp1, chi(ss+s+1));
-      coalescedWrite(chi[ss+s],chi(ss+s) - leec[s]*tmp1);
-    }
-  });
-
-  this->MooeeInvTime += usecond();
-}
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionImplementation.h

@@ -1,321 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/DomainWallEOFAFermion.cc
-
-Copyright (C) 2017
-
-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: David Murphy <dmurphy@phys.columbia.edu>
-
-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_Eigen_Dense.h>
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/DomainWallEOFAFermion.h>
-
-NAMESPACE_BEGIN(Grid);
-
-template<class Impl>
-DomainWallEOFAFermion<Impl>::DomainWallEOFAFermion(
-						   GaugeField            &_Umu,
-						   GridCartesian         &FiveDimGrid,
-						   GridRedBlackCartesian &FiveDimRedBlackGrid,
-						   GridCartesian         &FourDimGrid,
-						   GridRedBlackCartesian &FourDimRedBlackGrid,
-						   RealD _mq1, RealD _mq2, RealD _mq3,
-						   RealD _shift, int _pm, RealD _M5, const ImplParams &p) :
-  AbstractEOFAFermion<Impl>(_Umu, FiveDimGrid, FiveDimRedBlackGrid,
-			    FourDimGrid, FourDimRedBlackGrid, _mq1, _mq2, _mq3,
-			    _shift, _pm, _M5, 1.0, 0.0, p)
-{
-  RealD eps = 1.0;
-  Approx::zolotarev_data *zdata = Approx::higham(eps,this->Ls);
-  assert(zdata->n == this->Ls);
-
-  std::cout << GridLogMessage << "DomainWallEOFAFermion with Ls=" << this->Ls << std::endl;
-  this->SetCoefficientsTanh(zdata, 1.0, 0.0);
-
-  Approx::zolotarev_free(zdata);
-}
-
-/***************************************************************
- * Additional EOFA operators only called outside the inverter.
- * Since speed is not essential, simple axpby-style
- * implementations should be fine.
- ***************************************************************/
-template<class Impl>
-void DomainWallEOFAFermion<Impl>::Omega(const FermionField& psi, FermionField& Din, int sign, int dag)
-{
-  int Ls = this->Ls;
-
-  Din = Zero();
-  if((sign == 1) && (dag == 0)){ axpby_ssp(Din, 0.0, psi, 1.0, psi, Ls-1, 0); }
-  else if((sign == -1) && (dag == 0)){ axpby_ssp(Din, 0.0, psi, 1.0, psi, 0, 0); }
-  else if((sign == 1 ) && (dag == 1)){ axpby_ssp(Din, 0.0, psi, 1.0, psi, 0, Ls-1); }
-  else if((sign == -1) && (dag == 1)){ axpby_ssp(Din, 0.0, psi, 1.0, psi, 0, 0); }
-}
-
-// This is just the identity for DWF
-template<class Impl>
-void DomainWallEOFAFermion<Impl>::Dtilde(const FermionField& psi, FermionField& chi){ chi = psi; }
-
-// This is just the identity for DWF
-template<class Impl>
-void DomainWallEOFAFermion<Impl>::DtildeInv(const FermionField& psi, FermionField& chi){ chi = psi; }
-
-/*****************************************************************************************************/
-
-template<class Impl>
-RealD DomainWallEOFAFermion<Impl>::M(const FermionField& psi, FermionField& chi)
-{
-  FermionField Din(psi.Grid());
-
-  this->Meooe5D(psi, Din);
-  this->DW(Din, chi, DaggerNo);
-  axpby(chi, 1.0, 1.0, chi, psi);
-  this->M5D(psi, chi);
-  return(norm2(chi));
-}
-
-template<class Impl>
-RealD DomainWallEOFAFermion<Impl>::Mdag(const FermionField& psi, FermionField& chi)
-{
-  FermionField Din(psi.Grid());
-
-  this->DW(psi, Din, DaggerYes);
-  this->MeooeDag5D(Din, chi);
-  this->M5Ddag(psi, chi);
-  axpby(chi, 1.0, 1.0, chi, psi);
-  return(norm2(chi));
-}
-
-/********************************************************************
- * Performance critical fermion operators called inside the inverter
- ********************************************************************/
-
-template<class Impl>
-void DomainWallEOFAFermion<Impl>::M5D(const FermionField& psi, FermionField& chi)
-{
-  int   Ls    = this->Ls;
-  int   pm    = this->pm;
-  RealD shift = this->shift;
-  RealD mq1   = this->mq1;
-  RealD mq2   = this->mq2;
-  RealD mq3   = this->mq3;
-
-  // coefficients for shift operator ( = shift*\gamma_{5}*R_{5}*\Delta_{\pm}(mq2,mq3)*P_{\pm} )
-  Coeff_t shiftp(0.0), shiftm(0.0);
-  if(shift != 0.0){
-    if(pm == 1){ shiftp = shift*(mq3-mq2); }
-    else{ shiftm = -shift*(mq3-mq2); }
-  }
-
-  Vector<Coeff_t> diag(Ls,1.0);
-  Vector<Coeff_t> upper(Ls,-1.0); upper[Ls-1] = mq1 + shiftm;
-  Vector<Coeff_t> lower(Ls,-1.0); lower[0]    = mq1 + shiftp;
-
-#if(0)
-  std::cout << GridLogMessage << "DomainWallEOFAFermion::M5D(FF&,FF&):" << std::endl;
-  for(int i=0; i<diag.size(); ++i){
-    std::cout << GridLogMessage << "diag[" << i << "] =" << diag[i] << std::endl;
-  }
-  for(int i=0; i<upper.size(); ++i){
-    std::cout << GridLogMessage << "upper[" << i << "] =" << upper[i] << std::endl;
-  }
-  for(int i=0; i<lower.size(); ++i){
-    std::cout << GridLogMessage << "lower[" << i << "] =" << lower[i] << std::endl;
-  }
-#endif
-
-  this->M5D(psi, chi, chi, lower, diag, upper);
-}
-
-template<class Impl>
-void DomainWallEOFAFermion<Impl>::M5Ddag(const FermionField& psi, FermionField& chi)
-{
-  int   Ls    = this->Ls;
-  int   pm    = this->pm;
-  RealD shift = this->shift;
-  RealD mq1   = this->mq1;
-  RealD mq2   = this->mq2;
-  RealD mq3   = this->mq3;
-
-  // coefficients for shift operator ( = shift*\gamma_{5}*R_{5}*\Delta_{\pm}(mq2,mq3)*P_{\pm} )
-  Coeff_t shiftp(0.0), shiftm(0.0);
-  if(shift != 0.0){
-    if(pm == 1){ shiftp = shift*(mq3-mq2); }
-    else{ shiftm = -shift*(mq3-mq2); }
-  }
-
-  Vector<Coeff_t> diag(Ls,1.0);
-  Vector<Coeff_t> upper(Ls,-1.0); upper[Ls-1] = mq1 + shiftp;
-  Vector<Coeff_t> lower(Ls,-1.0); lower[0]    = mq1 + shiftm;
-
-  this->M5Ddag(psi, chi, chi, lower, diag, upper);
-}
-
-// half checkerboard operations
-template<class Impl>
-void DomainWallEOFAFermion<Impl>::Mooee(const FermionField& psi, FermionField& chi)
-{
-  int Ls = this->Ls;
-
-  Vector<Coeff_t> diag = this->bee;
-  Vector<Coeff_t> upper(Ls);
-  Vector<Coeff_t> lower(Ls);
-
-  for(int s=0; s<Ls; s++){
-    upper[s] = -this->cee[s];
-    lower[s] = -this->cee[s];
-  }
-  upper[Ls-1] = this->dm;
-  lower[0]    = this->dp;
-
-  this->M5D(psi, psi, chi, lower, diag, upper);
-}
-
-template<class Impl>
-void DomainWallEOFAFermion<Impl>::MooeeDag(const FermionField& psi, FermionField& chi)
-{
-  int Ls = this->Ls;
-
-  Vector<Coeff_t> diag = this->bee;
-  Vector<Coeff_t> upper(Ls);
-  Vector<Coeff_t> lower(Ls);
-
-  for(int s=0; s<Ls; s++){
-    upper[s] = -this->cee[s];
-    lower[s] = -this->cee[s];
-  }
-  upper[Ls-1] = this->dp;
-  lower[0]    = this->dm;
-
-  this->M5Ddag(psi, psi, chi, lower, diag, upper);
-}
-
-/****************************************************************************************/
-
-//Zolo
-template<class Impl>
-void DomainWallEOFAFermion<Impl>::SetCoefficientsInternal(RealD zolo_hi, Vector<Coeff_t>& gamma, RealD b, RealD c)
-{
-  int   Ls    = this->Ls;
-  int   pm    = this->pm;
-  RealD mq1   = this->mq1;
-  RealD mq2   = this->mq2;
-  RealD mq3   = this->mq3;
-  RealD shift = this->shift;
-
-  ////////////////////////////////////////////////////////
-  // Constants for the preconditioned matrix Cayley form
-  ////////////////////////////////////////////////////////
-  this->bs.resize(Ls);
-  this->cs.resize(Ls);
-  this->aee.resize(Ls);
-  this->aeo.resize(Ls);
-  this->bee.resize(Ls);
-  this->beo.resize(Ls);
-  this->cee.resize(Ls);
-  this->ceo.resize(Ls);
-
-  for(int i=0; i<Ls; ++i){
-    this->bee[i] = 4.0 - this->M5 + 1.0;
-    this->cee[i] = 1.0;
-  }
-
-  for(int i=0; i<Ls; ++i){
-    this->aee[i] = this->cee[i];
-    this->bs[i] = this->beo[i] = 1.0;
-    this->cs[i] = this->ceo[i] = 0.0;
-  }
-
-  //////////////////////////////////////////
-  // EOFA shift terms
-  //////////////////////////////////////////
-  if(pm == 1){
-    this->dp = mq1*this->cee[0] + shift*(mq3-mq2);
-    this->dm = mq1*this->cee[Ls-1];
-  } else if(this->pm == -1) {
-    this->dp = mq1*this->cee[0];
-    this->dm = mq1*this->cee[Ls-1] - shift*(mq3-mq2);
-  } else {
-    this->dp = mq1*this->cee[0];
-    this->dm = mq1*this->cee[Ls-1];
-  }
-
-  //////////////////////////////////////////
-  // LDU decomposition of eeoo
-  //////////////////////////////////////////
-  this->dee.resize(Ls+1);
-  this->lee.resize(Ls);
-  this->leem.resize(Ls);
-  this->uee.resize(Ls);
-  this->ueem.resize(Ls);
-
-  for(int i=0; i<Ls; ++i){
-
-    if(i < Ls-1){
-
-      this->lee[i] = -this->cee[i+1]/this->bee[i]; // sub-diag entry on the ith column
-
-      this->leem[i] = this->dm/this->bee[i];
-      for(int j=0; j<i; j++){ this->leem[i] *= this->aee[j]/this->bee[j]; }
-
-      this->dee[i] = this->bee[i];
-
-      this->uee[i] = -this->aee[i]/this->bee[i];   // up-diag entry on the ith row
-
-      this->ueem[i] = this->dp / this->bee[0];
-      for(int j=1; j<=i; j++){ this->ueem[i] *= this->cee[j]/this->bee[j]; }
-
-    } else {
-
-      this->lee[i]  = 0.0;
-      this->leem[i] = 0.0;
-      this->uee[i]  = 0.0;
-      this->ueem[i] = 0.0;
-
-    }
-  }
-
-  {
-    Coeff_t delta_d = 1.0 / this->bee[0];
-    for(int j=1; j<Ls-1; j++){ delta_d *= this->cee[j] / this->bee[j]; }
-    this->dee[Ls-1] = this->bee[Ls-1] + this->cee[0] * this->dm * delta_d;
-    this->dee[Ls] = this->bee[Ls-1] + this->cee[Ls-1] * this->dp * delta_d;
-  }
-}
-
-// Recompute Cayley-form coefficients for different shift
-template<class Impl>
-void DomainWallEOFAFermion<Impl>::RefreshShiftCoefficients(RealD new_shift)
-{
-  this->shift = new_shift;
-  Approx::zolotarev_data *zdata = Approx::higham(1.0, this->Ls);
-  this->SetCoefficientsTanh(zdata, 1.0, 0.0);
-}
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/implementation/ImprovedStaggeredFermionImplementation.h

@@ -1,615 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/ImprovedStaggeredFermion.cc
-
-Copyright (C) 2015
-
-Author: Azusa Yamaguchi, Peter Boyle
-
-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>
-
-#pragma once 
-
-NAMESPACE_BEGIN(Grid);
-
-/////////////////////////////////
-// Constructor and gauge import
-/////////////////////////////////
-
-template <class Impl>
-ImprovedStaggeredFermion<Impl>::ImprovedStaggeredFermion(GridCartesian &Fgrid, GridRedBlackCartesian &Hgrid, 
-							 RealD _mass,
-							 RealD _c1, RealD _c2,RealD _u0,
-							 const ImplParams &p)
-  : Kernels(p),
-    _grid(&Fgrid),
-    _cbgrid(&Hgrid),
-    Stencil(&Fgrid, npoint, Even, directions, displacements,p),
-    StencilEven(&Hgrid, npoint, Even, directions, displacements,p),  // source is Even
-    StencilOdd(&Hgrid, npoint, Odd, directions, displacements,p),  // source is Odd
-    mass(_mass),
-    Lebesgue(_grid),
-    LebesgueEvenOdd(_cbgrid),
-    Umu(&Fgrid),
-    UmuEven(&Hgrid),
-    UmuOdd(&Hgrid),
-    UUUmu(&Fgrid),
-    UUUmuEven(&Hgrid),
-    UUUmuOdd(&Hgrid) ,
-    _tmp(&Hgrid)
-{
-  int vol4;
-  int LLs=1;
-  c1=_c1;
-  c2=_c2;
-  u0=_u0;
-  vol4= _grid->oSites();
-  Stencil.BuildSurfaceList(LLs,vol4);
-  vol4= _cbgrid->oSites();
-  StencilEven.BuildSurfaceList(LLs,vol4);
-  StencilOdd.BuildSurfaceList(LLs,vol4);
-}
-
-template <class Impl>
-ImprovedStaggeredFermion<Impl>::ImprovedStaggeredFermion(GaugeField &_Uthin, GaugeField &_Ufat, GridCartesian &Fgrid,
-							 GridRedBlackCartesian &Hgrid, RealD _mass,
-							 RealD _c1, RealD _c2,RealD _u0,
-							 const ImplParams &p)
-  : ImprovedStaggeredFermion(Fgrid,Hgrid,_mass,_c1,_c2,_u0,p)
-{
-  ImportGauge(_Uthin,_Ufat);
-}
-
-////////////////////////////////////////////////////////////
-// Momentum space propagator should be 
-// https://arxiv.org/pdf/hep-lat/9712010.pdf
-//
-// mom space action.
-//   gamma_mu i ( c1 sin pmu + c2 sin 3 pmu ) + m
-//
-// must track through staggered flavour/spin reduction in literature to 
-// turn to free propagator for the one component chi field, a la page 4/5
-// of above link to implmement fourier based solver.
-////////////////////////////////////////////////////////////
-template <class Impl>
-void ImprovedStaggeredFermion<Impl>::ImportGaugeSimple(const GaugeField &_Utriple,const GaugeField &_Ufat) 
-{
-  /////////////////////////////////////////////////////////////////
-  // Trivial import; phases and fattening and such like preapplied
-  /////////////////////////////////////////////////////////////////
-  GaugeLinkField U(GaugeGrid());
-
-  for (int mu = 0; mu < Nd; mu++) {
-
-    U = PeekIndex<LorentzIndex>(_Utriple, mu);
-    PokeIndex<LorentzIndex>(UUUmu, U, mu );
-
-    U = adj( Cshift(U, mu, -3));
-    PokeIndex<LorentzIndex>(UUUmu, -U, mu+4 );
-
-    U = PeekIndex<LorentzIndex>(_Ufat, mu);
-    PokeIndex<LorentzIndex>(Umu, U, mu);
-
-    U = adj( Cshift(U, mu, -1));
-    PokeIndex<LorentzIndex>(Umu, -U, mu+4);
-
-  }
-  CopyGaugeCheckerboards();
-}
-template <class Impl>
-void ImprovedStaggeredFermion<Impl>::ImportGaugeSimple(const DoubledGaugeField &_UUU,const DoubledGaugeField &_U) 
-{
-
-  Umu   = _U;
-  UUUmu = _UUU;
-  CopyGaugeCheckerboards();
-}
-
-template <class Impl>
-void ImprovedStaggeredFermion<Impl>::CopyGaugeCheckerboards(void)
-{
-  pickCheckerboard(Even, UmuEven,  Umu);
-  pickCheckerboard(Odd,  UmuOdd ,  Umu);
-  pickCheckerboard(Even, UUUmuEven,UUUmu);
-  pickCheckerboard(Odd,  UUUmuOdd, UUUmu);
-}
-template <class Impl>
-void ImprovedStaggeredFermion<Impl>::ImportGauge(const GaugeField &_Uthin,const GaugeField &_Ufat) 
-{
-  GaugeLinkField U(GaugeGrid());
-
-  ////////////////////////////////////////////////////////
-  // Double Store should take two fields for Naik and one hop separately.
-  ////////////////////////////////////////////////////////
-  Impl::DoubleStore(GaugeGrid(), UUUmu, Umu, _Uthin, _Ufat );
-
-  ////////////////////////////////////////////////////////
-  // Apply scale factors to get the right fermion Kinetic term
-  // Could pass coeffs into the double store to save work.
-  // 0.5 ( U p(x+mu) - Udag(x-mu) p(x-mu) ) 
-  ////////////////////////////////////////////////////////
-  for (int mu = 0; mu < Nd; mu++) {
-
-    U = PeekIndex<LorentzIndex>(Umu, mu);
-    PokeIndex<LorentzIndex>(Umu, U*( 0.5*c1/u0), mu );
-    
-    U = PeekIndex<LorentzIndex>(Umu, mu+4);
-    PokeIndex<LorentzIndex>(Umu, U*(-0.5*c1/u0), mu+4);
-
-    U = PeekIndex<LorentzIndex>(UUUmu, mu);
-    PokeIndex<LorentzIndex>(UUUmu, U*( 0.5*c2/u0/u0/u0), mu );
-    
-    U = PeekIndex<LorentzIndex>(UUUmu, mu+4);
-    PokeIndex<LorentzIndex>(UUUmu, U*(-0.5*c2/u0/u0/u0), mu+4);
-  }
-
-  CopyGaugeCheckerboards();
-}
-
-/////////////////////////////
-// Implement the interface
-/////////////////////////////
-
-template <class Impl>
-RealD ImprovedStaggeredFermion<Impl>::M(const FermionField &in, FermionField &out) {
-  out.Checkerboard() = in.Checkerboard();
-  Dhop(in, out, DaggerNo);
-  return axpy_norm(out, mass, in, out);
-}
-
-template <class Impl>
-RealD ImprovedStaggeredFermion<Impl>::Mdag(const FermionField &in, FermionField &out) {
-  out.Checkerboard() = in.Checkerboard();
-  Dhop(in, out, DaggerYes);
-  return axpy_norm(out, mass, in, out);
-}
-
-template <class Impl>
-void ImprovedStaggeredFermion<Impl>::Meooe(const FermionField &in, FermionField &out) {
-  if (in.Checkerboard() == Odd) {
-    DhopEO(in, out, DaggerNo);
-  } else {
-    DhopOE(in, out, DaggerNo);
-  }
-}
-template <class Impl>
-void ImprovedStaggeredFermion<Impl>::MeooeDag(const FermionField &in, FermionField &out) {
-  if (in.Checkerboard() == Odd) {
-    DhopEO(in, out, DaggerYes);
-  } else {
-    DhopOE(in, out, DaggerYes);
-  }
-}
-
-template <class Impl>
-void ImprovedStaggeredFermion<Impl>::Mooee(const FermionField &in, FermionField &out) {
-  out.Checkerboard() = in.Checkerboard();
-  typename FermionField::scalar_type scal(mass);
-  out = scal * in;
-}
-
-template <class Impl>
-void ImprovedStaggeredFermion<Impl>::MooeeDag(const FermionField &in, FermionField &out) {
-  out.Checkerboard() = in.Checkerboard();
-  Mooee(in, out);
-}
-
-template <class Impl>
-void ImprovedStaggeredFermion<Impl>::MooeeInv(const FermionField &in, FermionField &out) {
-  out.Checkerboard() = in.Checkerboard();
-  out = (1.0 / (mass)) * in;
-}
-
-template <class Impl>
-void ImprovedStaggeredFermion<Impl>::MooeeInvDag(const FermionField &in,
-						 FermionField &out) {
-  out.Checkerboard() = in.Checkerboard();
-  MooeeInv(in, out);
-}
-
-///////////////////////////////////
-// Internal
-///////////////////////////////////
-
-template <class Impl>
-void ImprovedStaggeredFermion<Impl>::DerivInternal(StencilImpl &st, DoubledGaugeField &U, DoubledGaugeField &UUU, 
-						   GaugeField & mat,
-						   const FermionField &A, const FermionField &B, int dag) {
-  assert((dag == DaggerNo) || (dag == DaggerYes));
-
-  Compressor compressor;
-
-  FermionField Btilde(B.Grid());
-  FermionField Atilde(B.Grid());
-  Atilde = A;
-
-  st.HaloExchange(B, compressor);
-
-  for (int mu = 0; mu < Nd; mu++) {
-
-    ////////////////////////
-    // Call the single hop
-    ////////////////////////
-    auto U_v   = U.View();
-    auto UUU_v = UUU.View();
-    auto B_v   = B.View();
-    auto Btilde_v   = Btilde.View();
-    thread_for(sss,B.Grid()->oSites(),{
-      Kernels::DhopDirKernel(st, U_v, UUU_v, st.CommBuf(), sss, sss, B_v, Btilde_v, mu,1);
-    });
-
-    // Force in three link terms
-    //
-    //    Impl::InsertForce4D(mat, Btilde, Atilde, mu);
-    //
-    // dU_ac(x)/dt = i p_ab U_bc(x)
-    //
-    // => dS_f/dt = dS_f/dU_ac(x) . dU_ac(x)/dt =  i p_ab U_bc(x) dS_f/dU_ac(x) 
-    //
-    // One link: form fragments S_f = A U B 
-    //
-    //         write Btilde = U(x) B(x+mu)
-    //
-    // mat+= TraceIndex<SpinIndex>(outerProduct(Btilde,A)); 
-    // 
-    // Three link: form fragments S_f = A UUU B 
-    //
-    // mat+= outer ( A, UUUB) <-- Best take DhopDeriv with one linke or identity matrix
-    // mat+= outer ( AU, UUB) <-- and then use covariant cshift?
-    // mat+= outer ( AUU, UB) <-- Returned from call to DhopDir
-
-    assert(0);// need to figure out the force interface with a blasted three link term.
-    
-  }
-}
-
-template <class Impl>
-void ImprovedStaggeredFermion<Impl>::DhopDeriv(GaugeField &mat, const FermionField &U, const FermionField &V, int dag) {
-
-  conformable(U.Grid(), _grid);
-  conformable(U.Grid(), V.Grid());
-  conformable(U.Grid(), mat.Grid());
-
-  mat.Checkerboard() = U.Checkerboard();
-
-  DerivInternal(Stencil, Umu, UUUmu, mat, U, V, dag);
-}
-
-template <class Impl>
-void ImprovedStaggeredFermion<Impl>::DhopDerivOE(GaugeField &mat, const FermionField &U, const FermionField &V, int dag) {
-
-  conformable(U.Grid(), _cbgrid);
-  conformable(U.Grid(), V.Grid());
-  conformable(U.Grid(), mat.Grid());
-
-  assert(V.Checkerboard() == Even);
-  assert(U.Checkerboard() == Odd);
-  mat.Checkerboard() = Odd;
-
-  DerivInternal(StencilEven, UmuOdd, UUUmuOdd, mat, U, V, dag);
-}
-
-template <class Impl>
-void ImprovedStaggeredFermion<Impl>::DhopDerivEO(GaugeField &mat, const FermionField &U, const FermionField &V, int dag) {
-
-  conformable(U.Grid(), _cbgrid);
-  conformable(U.Grid(), V.Grid());
-  conformable(U.Grid(), mat.Grid());
-
-  assert(V.Checkerboard() == Odd);
-  assert(U.Checkerboard() == Even);
-  mat.Checkerboard() = Even;
-
-  DerivInternal(StencilOdd, UmuEven, UUUmuEven, mat, U, V, dag);
-}
-
-template <class Impl>
-void ImprovedStaggeredFermion<Impl>::Dhop(const FermionField &in, FermionField &out, int dag) 
-{
-  DhopCalls+=2;
-  conformable(in.Grid(), _grid);  // verifies full grid
-  conformable(in.Grid(), out.Grid());
-
-  out.Checkerboard() = in.Checkerboard();
-
-  DhopInternal(Stencil, Lebesgue, Umu, UUUmu, in, out, dag);
-}
-
-template <class Impl>
-void ImprovedStaggeredFermion<Impl>::DhopOE(const FermionField &in, FermionField &out, int dag) 
-{
-  DhopCalls+=1;
-  conformable(in.Grid(), _cbgrid);    // verifies half grid
-  conformable(in.Grid(), out.Grid());  // drops the cb check
-
-  assert(in.Checkerboard() == Even);
-  out.Checkerboard() = Odd;
-
-  DhopInternal(StencilEven, LebesgueEvenOdd, UmuOdd, UUUmuOdd, in, out, dag);
-}
-
-template <class Impl>
-void ImprovedStaggeredFermion<Impl>::DhopEO(const FermionField &in, FermionField &out, int dag) 
-{
-  DhopCalls+=1;
-  conformable(in.Grid(), _cbgrid);    // verifies half grid
-  conformable(in.Grid(), out.Grid());  // drops the cb check
-
-  assert(in.Checkerboard() == Odd);
-  out.Checkerboard() = Even;
-
-  DhopInternal(StencilOdd, LebesgueEvenOdd, UmuEven, UUUmuEven, in, out, dag);
-}
-
-template <class Impl>
-void ImprovedStaggeredFermion<Impl>::Mdir(const FermionField &in, FermionField &out, int dir, int disp) {
-  DhopDir(in, out, dir, disp);
-}
-
-template <class Impl>
-void ImprovedStaggeredFermion<Impl>::DhopDir(const FermionField &in, FermionField &out, int dir, int disp) {
-
-  Compressor compressor;
-  Stencil.HaloExchange(in, compressor);
-  auto Umu_v   =   Umu.View();
-  auto UUUmu_v = UUUmu.View();
-  auto in_v    =  in.View();
-  auto out_v   = out.View();
-  thread_for( sss, in.Grid()->oSites(),{
-    Kernels::DhopDirKernel(Stencil, Umu_v, UUUmu_v, Stencil.CommBuf(), sss, sss, in_v, out_v, dir, disp);
-  });
-};
-
-template <class Impl>
-void ImprovedStaggeredFermion<Impl>::DhopInternal(StencilImpl &st, LebesgueOrder &lo,
-						  DoubledGaugeField &U,
-						  DoubledGaugeField &UUU,
-						  const FermionField &in,
-						  FermionField &out, int dag) 
-{
-#ifdef GRID_OMP
-  if ( StaggeredKernelsStatic::Comms == StaggeredKernelsStatic::CommsAndCompute )
-    DhopInternalOverlappedComms(st,lo,U,UUU,in,out,dag);
-  else
-#endif
-    DhopInternalSerialComms(st,lo,U,UUU,in,out,dag);
-}
-template <class Impl>
-void ImprovedStaggeredFermion<Impl>::DhopInternalOverlappedComms(StencilImpl &st, LebesgueOrder &lo,
-								 DoubledGaugeField &U,
-								 DoubledGaugeField &UUU,
-								 const FermionField &in,
-								 FermionField &out, int dag) 
-{
-#ifdef GRID_OMP
-  Compressor compressor; 
-  int len =  U.Grid()->oSites();
-  const int LLs =  1;
-
-  DhopTotalTime   -= usecond();
-
-  DhopFaceTime    -= usecond();
-  st.Prepare();
-  st.HaloGather(in,compressor);
-  st.CommsMergeSHM(compressor);
-  DhopFaceTime    += usecond();
-
-  //////////////////////////////////////////////////////////////////////////////////////////////////////
-  // Ugly explicit thread mapping introduced for OPA reasons.
-  //////////////////////////////////////////////////////////////////////////////////////////////////////
-  DhopComputeTime    -= usecond();
-#pragma omp parallel 
-  {
-    int tid = omp_get_thread_num();
-    int nthreads = omp_get_num_threads();
-    int ncomms = CartesianCommunicator::nCommThreads;
-    if (ncomms == -1) ncomms = 1;
-    assert(nthreads > ncomms);
-
-    if (tid >= ncomms) {
-      nthreads -= ncomms;
-      int ttid  = tid - ncomms;
-      int n     = len;
-      int chunk = n / nthreads;
-      int rem   = n % nthreads;
-      int myblock, myn;
-      if (ttid < rem) {
-        myblock = ttid * chunk + ttid;
-        myn = chunk+1;
-      } else {
-        myblock = ttid*chunk + rem;
-        myn = chunk;
-      }
-
-      // do the compute
-      auto U_v   = U.View();
-      auto UUU_v = UUU.View();
-      auto in_v  = in.View();
-      auto out_v = out.View();
-      if (dag == DaggerYes) {
-        for (int ss = myblock; ss < myblock+myn; ++ss) {
-          int sU = ss;
-	  // Interior = 1; Exterior = 0; must implement for staggered
-          Kernels::DhopSiteDag(st,lo,U_v,UUU_v,st.CommBuf(),1,sU,in_v,out_v,1,0); 
-        }
-      } else {
-        for (int ss = myblock; ss < myblock+myn; ++ss) {
-	  // Interior = 1; Exterior = 0;
-          int sU = ss;
-          Kernels::DhopSite(st,lo,U_v,UUU_v,st.CommBuf(),1,sU,in_v,out_v,1,0);
-        }
-      }
-    } else {
-      st.CommunicateThreaded();
-    }
-  }
-  DhopComputeTime    += usecond();
-
-  // First to enter, last to leave timing
-  DhopFaceTime    -= usecond();
-  st.CommsMerge(compressor);
-  DhopFaceTime    -= usecond();
-
-  DhopComputeTime2    -= usecond();
-  {
-    auto U_v   = U.View();
-    auto UUU_v = UUU.View();
-    auto in_v  = in.View();
-    auto out_v = out.View();
-    if (dag == DaggerYes) {
-      int sz=st.surface_list.size();
-      thread_for(ss,sz,{
-	int sU = st.surface_list[ss];
-	Kernels::DhopSiteDag(st,lo,U_v,UUU_v,st.CommBuf(),1,sU,in_v,out_v,0,1);
-      });
-    } else {
-      int sz=st.surface_list.size();
-      thread_for(ss,sz,{
-	int sU = st.surface_list[ss];
-	Kernels::DhopSite(st,lo,U_v,UUU_v,st.CommBuf(),1,sU,in_v,out_v,0,1);
-      });
-    }
-  }
-  DhopComputeTime2    += usecond();
-#else
-  assert(0);
-#endif
-}
-
-
-template <class Impl>
-void ImprovedStaggeredFermion<Impl>::DhopInternalSerialComms(StencilImpl &st, LebesgueOrder &lo,
-							     DoubledGaugeField &U,
-							     DoubledGaugeField &UUU,
-							     const FermionField &in,
-							     FermionField &out, int dag) 
-{
-  assert((dag == DaggerNo) || (dag == DaggerYes));
-
-  DhopTotalTime   -= usecond();
-
-  DhopCommTime    -= usecond();
-  Compressor compressor;
-  st.HaloExchange(in, compressor);
-  DhopCommTime    += usecond();
-
-  auto U_v   =   U.View();
-  auto UUU_v = UUU.View();
-  auto in_v  =  in.View();
-  auto out_v = out.View();
-  DhopComputeTime -= usecond();
-  if (dag == DaggerYes) {
-    thread_for(sss, in.Grid()->oSites(),{
-      Kernels::DhopSiteDag(st, lo, U_v, UUU_v, st.CommBuf(), 1, sss, in_v, out_v);
-    });
-  } else {
-    thread_for(sss, in.Grid()->oSites(),{
-      Kernels::DhopSite(st, lo, U_v, UUU_v, st.CommBuf(), 1, sss, in_v, out_v);
-    });
-  }
-  DhopComputeTime += usecond();
-  DhopTotalTime   += usecond();
-};
-
-  ////////////////////////////////////////////////////////////////
-  // Reporting
-  ////////////////////////////////////////////////////////////////
-template<class Impl>
-void ImprovedStaggeredFermion<Impl>::Report(void) 
-{
-  Coordinate latt = _grid->GlobalDimensions();
-  RealD volume = 1;  for(int mu=0;mu<Nd;mu++) volume=volume*latt[mu];
-  RealD NP = _grid->_Nprocessors;
-  RealD NN = _grid->NodeCount();
-
-  std::cout << GridLogMessage << "#### Dhop calls report " << std::endl;
-
-  std::cout << GridLogMessage << "ImprovedStaggeredFermion Number of DhopEO Calls   : " 
-	    << DhopCalls   << std::endl;
-  std::cout << GridLogMessage << "ImprovedStaggeredFermion TotalTime   /Calls       : " 
-	    << DhopTotalTime   / DhopCalls << " us" << std::endl;
-  std::cout << GridLogMessage << "ImprovedStaggeredFermion CommTime    /Calls       : " 
-	    << DhopCommTime    / DhopCalls << " us" << std::endl;
-  std::cout << GridLogMessage << "ImprovedStaggeredFermion ComputeTime/Calls        : " 
-	    << DhopComputeTime / DhopCalls << " us" << std::endl;
-
-  // Average the compute time
-  _grid->GlobalSum(DhopComputeTime);
-  DhopComputeTime/=NP;
-
-  RealD mflops = 1154*volume*DhopCalls/DhopComputeTime/2; // 2 for red black counting
-  std::cout << GridLogMessage << "Average mflops/s per call                : " << mflops << std::endl;
-  std::cout << GridLogMessage << "Average mflops/s per call per rank       : " << mflops/NP << std::endl;
-  std::cout << GridLogMessage << "Average mflops/s per call per node       : " << mflops/NN << std::endl;
-  
-  RealD Fullmflops = 1154*volume*DhopCalls/(DhopTotalTime)/2; // 2 for red black counting
-  std::cout << GridLogMessage << "Average mflops/s per call (full)         : " << Fullmflops << std::endl;
-  std::cout << GridLogMessage << "Average mflops/s per call per rank (full): " << Fullmflops/NP << std::endl;
-  std::cout << GridLogMessage << "Average mflops/s per call per node (full): " << Fullmflops/NN << std::endl;
-
-  std::cout << GridLogMessage << "ImprovedStaggeredFermion Stencil"    <<std::endl;  Stencil.Report();
-  std::cout << GridLogMessage << "ImprovedStaggeredFermion StencilEven"<<std::endl;  StencilEven.Report();
-  std::cout << GridLogMessage << "ImprovedStaggeredFermion StencilOdd" <<std::endl;  StencilOdd.Report();
-}
-template<class Impl>
-void ImprovedStaggeredFermion<Impl>::ZeroCounters(void) 
-{
-  DhopCalls       = 0;
-  DhopTotalTime   = 0;
-  DhopCommTime    = 0;
-  DhopComputeTime = 0;
-  DhopFaceTime    = 0;
-
-  Stencil.ZeroCounters();
-  StencilEven.ZeroCounters();
-  StencilOdd.ZeroCounters();
-}
-
-
-//////////////////////////////////////////////////////// 
-// Conserved current - not yet implemented.
-////////////////////////////////////////////////////////
-template <class Impl>
-void ImprovedStaggeredFermion<Impl>::ContractConservedCurrent(PropagatorField &q_in_1,
-							      PropagatorField &q_in_2,
-							      PropagatorField &q_out,
-							      Current curr_type,
-							      unsigned int mu)
-{
-  assert(0);
-}
-
-template <class Impl>
-void ImprovedStaggeredFermion<Impl>::SeqConservedCurrent(PropagatorField &q_in,
-                                                         PropagatorField &q_out,
-                                                         Current curr_type,
-                                                         unsigned int mu, 
-                                                         unsigned int tmin,
-                                              unsigned int tmax,
-					      ComplexField &lattice_cmplx)
-{
-  assert(0);
-
-}
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/implementation/MobiusEOFAFermionCache.h

@@ -1,453 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/MobiusEOFAFermioncache.cc
-
-Copyright (C) 2017
-
-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: David Murphy <dmurphy@phys.columbia.edu>
-
-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/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/MobiusEOFAFermion.h>
-
-NAMESPACE_BEGIN(Grid);
-
- 
-template<class Impl>
-void MobiusEOFAFermion<Impl>::M5D(const FermionField &psi_i, const FermionField &phi_i, FermionField &chi_i,
-				  Vector<Coeff_t> &lower, Vector<Coeff_t> &diag, Vector<Coeff_t> &upper)
-{
-  chi_i.Checkerboard() = psi_i.Checkerboard();
-  GridBase *grid = psi_i.Grid();
-  int Ls = this->Ls;
-  auto psi = psi_i.View();
-  auto phi = phi_i.View();
-  auto chi = chi_i.View();
-
-  assert(phi.Checkerboard() == psi.Checkerboard());
-
-  // Flops = 6.0*(Nc*Ns) *Ls*vol
-  this->M5Dcalls++;
-  this->M5Dtime -= usecond();
-
-  int nloop = grid->oSites()/Ls;
-  accelerator_for(sss,nloop,Simd::Nsimd(),{
-    uint64_t ss = sss*Ls;
-    typedef decltype(coalescedRead(psi[0])) spinor;
-    spinor tmp1;
-    spinor 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], diag[s]*phi(ss+s) + upper[s]*tmp1 + lower[s]*tmp2);
-    }
-  });
-
-  this->M5Dtime += usecond();
-}
-
-template<class Impl>
-void MobiusEOFAFermion<Impl>::M5D_shift(const FermionField &psi_i, const FermionField &phi_i, FermionField &chi_i,
-					Vector<Coeff_t> &lower, Vector<Coeff_t> &diag, Vector<Coeff_t> &upper,
-					Vector<Coeff_t> &shift_coeffs)
-{
-  chi_i.Checkerboard() = psi_i.Checkerboard();
-  GridBase *grid = psi_i.Grid();
-  int Ls = this->Ls;
-  auto psi = psi_i.View();
-  auto phi = phi_i.View();
-  auto chi = chi_i.View();
-
-  auto pm  = this->pm;
-  int shift_s = (pm == 1) ? (Ls-1) : 0; // s-component modified by shift operator
-
-  assert(phi.Checkerboard() == psi.Checkerboard());
-
-  // Flops = 6.0*(Nc*Ns) *Ls*vol
-  this->M5Dcalls++;
-  this->M5Dtime -= usecond();
-
-  int nloop = grid->oSites()/Ls;
-  accelerator_for(sss,nloop,Simd::Nsimd(),{
-    uint64_t ss = sss*Ls;
-    typedef decltype(coalescedRead(psi[0])) spinor;
-    spinor tmp1;
-    spinor tmp2;
-    spinor tmp;
-    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));
-
-      if(pm == 1){ spProj5p(tmp, psi(ss+shift_s)); }
-      else       { spProj5m(tmp, psi(ss+shift_s)); }
-
-      coalescedWrite(chi[ss+s], diag[s]*phi(ss+s) + upper[s]*tmp1 +lower[s]*tmp2 + shift_coeffs[s]*tmp);
-    }
-  });
-
-  this->M5Dtime += usecond();
-}
-
-template<class Impl>
-void MobiusEOFAFermion<Impl>::M5Ddag(const FermionField &psi_i, const FermionField &phi_i, FermionField &chi_i,
-				     Vector<Coeff_t> &lower, Vector<Coeff_t> &diag, Vector<Coeff_t> &upper)
-{
-  chi_i.Checkerboard() = psi_i.Checkerboard();
-  GridBase *grid = psi_i.Grid();
-  int Ls = this->Ls;
-  auto psi = psi_i.View();
-  auto phi = phi_i.View();
-  auto chi = chi_i.View();
-
-  assert(phi.Checkerboard() == psi.Checkerboard());
-
-  // Flops = 6.0*(Nc*Ns) *Ls*vol
-  this->M5Dcalls++;
-  this->M5Dtime -= usecond();
-
-  int nloop = grid->oSites()/Ls;
-  accelerator_for(sss,nloop,Simd::Nsimd(), {
-    uint64_t ss = sss*Ls;
-
-    typedef decltype(coalescedRead(psi[0])) spinor;
-    spinor tmp1, 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], diag[s]*phi(ss+s) + upper[s]*tmp1 + lower[s]*tmp2);
-    }
-  });
-
-  this->M5Dtime += usecond();
-}
-
-template<class Impl>
-void MobiusEOFAFermion<Impl>::M5Ddag_shift(const FermionField &psi_i, const FermionField &phi_i, FermionField &chi_i,
-					   Vector<Coeff_t> &lower, Vector<Coeff_t> &diag, Vector<Coeff_t> &upper,
-					   Vector<Coeff_t> &shift_coeffs)
-{
-  chi_i.Checkerboard() = psi_i.Checkerboard();
-  GridBase *grid = psi_i.Grid();
-  int Ls = this->Ls;
-  int shift_s = (this->pm == 1) ? (Ls-1) : 0; // s-component modified by shift operator
-  auto psi = psi_i.View();
-  auto phi = phi_i.View();
-  auto chi = chi_i.View();
-
-  assert(phi.Checkerboard() == psi.Checkerboard());
-
-  // Flops = 6.0*(Nc*Ns) *Ls*vol
-  this->M5Dcalls++;
-  this->M5Dtime -= usecond();
-
-  auto pm = this->pm;
-
-  int nloop = grid->oSites()/Ls;
-  accelerator_for(sss,nloop,Simd::Nsimd(),{
-    uint64_t ss = sss*Ls;
-
-    typedef decltype(coalescedRead(psi[0])) spinor;
-    spinor tmp1, tmp2, tmp;
-    tmp1=Zero();
-    coalescedWrite(chi[ss+Ls-1],tmp1);
-
-    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));
-
-      if(s==(Ls-1)) coalescedWrite(chi[ss+s], chi(ss+s)+ diag[s]*phi(ss+s) + upper[s]*tmp1 + lower[s]*tmp2);
-      else          coalescedWrite(chi[ss+s], diag[s]*phi(ss+s) + upper[s]*tmp1 + lower[s]*tmp2);
-      if(pm == 1){ spProj5p(tmp, psi(ss+s)); }
-      else       { spProj5m(tmp, psi(ss+s)); }
-
-      coalescedWrite(chi[ss+shift_s],chi(ss+shift_s)+shift_coeffs[s]*tmp);
-    }
-  });
-
-  this->M5Dtime += usecond();
-}
-
-template<class Impl>
-void MobiusEOFAFermion<Impl>::MooeeInv(const FermionField &psi_i, FermionField &chi_i)
-{
-  chi_i.Checkerboard() = psi_i.Checkerboard();
-  GridBase *grid = psi_i.Grid();
-  int Ls = this->Ls;
-  auto psi = psi_i.View();
-  auto chi = chi_i.View();
-
-  auto plee = & this->lee [0];
-  auto pdee = & this->dee [0];
-  auto puee = & this->uee [0];
-  auto pleem= & this->leem[0];
-  auto pueem= & this->ueem[0];
-
-  if(this->shift != 0.0){ MooeeInv_shift(psi_i,chi_i); return; }
-
-  this->MooeeInvCalls++;
-  this->MooeeInvTime -= usecond();
-
-  int nloop = grid->oSites()/Ls;
-  accelerator_for(sss,nloop,Simd::Nsimd(),{
-
-    uint64_t ss = sss*Ls;
-
-    typedef decltype(coalescedRead(psi[0])) spinor;
-    spinor tmp;
-
-    // Apply (L^{\prime})^{-1}
-    coalescedWrite(chi[ss], psi(ss)); // chi[0]=psi[0]
-    for(int s=1; s<Ls; s++){
-      spProj5p(tmp, chi(ss+s-1));
-      coalescedWrite(chi[ss+s], psi(ss+s) - plee[s-1]*tmp);
-    }
-
-    // L_m^{-1}
-    for(int s=0; s<Ls-1; s++){ // Chi[ee] = 1 - sum[s<Ls-1] -leem[s]P_- chi
-      spProj5m(tmp, chi(ss+s));
-      coalescedWrite(chi[ss+Ls-1], chi(ss+Ls-1) - pleem[s]*tmp);
-    }
-
-    // U_m^{-1} D^{-1}
-    for(int s=0; s<Ls-1; s++){ // Chi[s] + 1/d chi[s]
-      spProj5p(tmp, chi(ss+Ls-1));
-      coalescedWrite(chi[ss+s], (1.0/pdee[s])*chi(ss+s) - (pueem[s]/pdee[Ls-1])*tmp);
-    }
-    coalescedWrite(chi[ss+Ls-1], (1.0/pdee[Ls-1])*chi(ss+Ls-1));
-
-    // Apply U^{-1}
-    for(int s=Ls-2; s>=0; s--){
-      spProj5m(tmp, chi(ss+s+1));
-      coalescedWrite(chi[ss+s], chi(ss+s) - puee[s]*tmp);
-    }
-  });
-   
-  this->MooeeInvTime += usecond();
-}
-
-template<class Impl>
-void MobiusEOFAFermion<Impl>::MooeeInv_shift(const FermionField &psi_i, FermionField &chi_i)
-{
-  chi_i.Checkerboard() = psi_i.Checkerboard();
-  GridBase *grid = psi_i.Grid();
-  int Ls = this->Ls;
-  auto psi = psi_i.View();
-  auto chi = chi_i.View();
-
-  auto pm = this->pm;
-  auto plee = & this->lee [0];
-  auto pdee = & this->dee [0];
-  auto puee = & this->uee [0];
-  auto pleem= & this->leem[0];
-  auto pueem= & this->ueem[0];
-  auto pMooeeInv_shift_lc   = &MooeeInv_shift_lc[0];
-  auto pMooeeInv_shift_norm = &MooeeInv_shift_norm[0];
-  this->MooeeInvCalls++;
-  this->MooeeInvTime -= usecond();
-
-  int nloop = grid->oSites()/Ls;
-  accelerator_for(sss,nloop,Simd::Nsimd(),{
-
-    uint64_t ss = sss*Ls;
-
-    typedef decltype(coalescedRead(psi[0])) spinor;
-    spinor tmp1,tmp2,tmp2_spProj;
-
-    // Apply (L^{\prime})^{-1} and accumulate MooeeInv_shift_lc[j]*psi[j] in tmp2
-    coalescedWrite(chi[ss], psi(ss)); // chi[0]=psi[0]
-    tmp2 = pMooeeInv_shift_lc[0]*psi(ss);
-    for(int s=1; s<Ls; s++){
-      spProj5p(tmp1, chi(ss+s-1));
-      coalescedWrite(chi[ss+s], psi(ss+s) - plee[s-1]*tmp1);
-      tmp2 = tmp2 + pMooeeInv_shift_lc[s]*psi(ss+s);
-    }
-    if(pm == 1){ spProj5p(tmp2_spProj, tmp2);}
-    else       { spProj5m(tmp2_spProj, tmp2); }
-
-    // L_m^{-1}
-    for(int s=0; s<Ls-1; s++){ // Chi[ee] = 1 - sum[s<Ls-1] -leem[s]P_- chi
-      spProj5m(tmp1, chi(ss+s));
-      coalescedWrite(chi[ss+Ls-1], chi(ss+Ls-1) - pleem[s]*tmp1);
-    }
-
-    // U_m^{-1} D^{-1}
-    for(int s=0; s<Ls-1; s++){ // Chi[s] + 1/d chi[s]
-      spProj5p(tmp1, chi(ss+Ls-1));
-      coalescedWrite(chi[ss+s], (1.0/pdee[s])*chi(ss+s) - (pueem[s]/pdee[Ls-1])*tmp1);
-    }
-    // chi[ss+Ls-1] = (1.0/pdee[Ls-1])*chi[ss+Ls-1] + MooeeInv_shift_norm[Ls-1]*tmp2_spProj;
-    coalescedWrite(chi[ss+Ls-1], (1.0/pdee[Ls-1])*chi(ss+Ls-1));
-    spProj5m(tmp1, chi(ss+Ls-1));
-    coalescedWrite(chi[ss+Ls-1], chi(ss+Ls-1) + pMooeeInv_shift_norm[Ls-1]*tmp2_spProj);
-
-    // Apply U^{-1} and add shift term
-    for(int s=Ls-2; s>=0; s--){
-      coalescedWrite(chi[ss+s] , chi(ss+s) - puee[s]*tmp1);
-      spProj5m(tmp1, chi(ss+s));
-      coalescedWrite(chi[ss+s], chi(ss+s) + pMooeeInv_shift_norm[s]*tmp2_spProj);
-    }
-  });
-
-  this->MooeeInvTime += usecond();
-}
-
-template<class Impl>
-void MobiusEOFAFermion<Impl>::MooeeInvDag(const FermionField &psi_i, FermionField &chi_i)
-{
-  if(this->shift != 0.0){ MooeeInvDag_shift(psi_i,chi_i); return; }
-
-  chi_i.Checkerboard() = psi_i.Checkerboard();
-  GridBase *grid = psi_i.Grid();
-  int Ls = this->Ls;
-  auto psi = psi_i.View();
-  auto chi = chi_i.View();
-
-  auto plee = & this->lee [0];
-  auto pdee = & this->dee [0];
-  auto puee = & this->uee [0];
-  auto pleem= & this->leem[0];
-  auto pueem= & this->ueem[0];
-
-  this->MooeeInvCalls++;
-  this->MooeeInvTime -= usecond();
-
-  int nloop = grid->oSites()/Ls;
-  accelerator_for(sss,nloop,Simd::Nsimd(),{
-
-    uint64_t ss = sss*Ls;
-
-    typedef decltype(coalescedRead(psi[0])) spinor;
-    spinor tmp;
-
-    // Apply (U^{\prime})^{-dag}
-    coalescedWrite(chi[ss], psi(ss));
-    for(int s=1; s<Ls; s++){
-      spProj5m(tmp, chi(ss+s-1));
-      coalescedWrite(chi[ss+s], psi(ss+s) - puee[s-1]*tmp);
-    }
-    
-    // U_m^{-\dag}
-    for(int s=0; s<Ls-1; s++){
-      spProj5p(tmp, chi(ss+s));
-      coalescedWrite(chi[ss+Ls-1], chi(ss+Ls-1) - pueem[s]*tmp);
-    }
-
-    // L_m^{-\dag} D^{-dag}
-    for(int s=0; s<Ls-1; s++){
-      spProj5m(tmp, chi(ss+Ls-1));
-      coalescedWrite(chi[ss+s], (1.0/pdee[s])*chi(ss+s) - (pleem[s]/pdee[Ls-1])*tmp);
-    }
-    coalescedWrite(chi[ss+Ls-1], (1.0/pdee[Ls-1])*chi(ss+Ls-1));
-
-    // Apply L^{-dag}
-    for(int s=Ls-2; s>=0; s--){
-      spProj5p(tmp, chi(ss+s+1));
-      coalescedWrite(chi[ss+s], chi(ss+s) - plee[s]*tmp);
-    }
-  });
-
-  this->MooeeInvTime += usecond();
-}
-
-template<class Impl>
-void MobiusEOFAFermion<Impl>::MooeeInvDag_shift(const FermionField &psi_i, FermionField &chi_i)
-{
-  chi_i.Checkerboard() = psi_i.Checkerboard();
-  GridBase *grid = psi_i.Grid();
-  auto psi = psi_i.View();
-  auto chi = chi_i.View();
-  int Ls = this->Ls;
-
-  auto pm = this->pm;
-  auto plee = & this->lee [0];
-  auto pdee = & this->dee [0];
-  auto puee = & this->uee [0];
-  auto pleem= & this->leem[0];
-  auto pueem= & this->ueem[0];
-  auto pMooeeInvDag_shift_lc   = &MooeeInvDag_shift_lc[0];
-  auto pMooeeInvDag_shift_norm = &MooeeInvDag_shift_norm[0];
-
-  this->MooeeInvCalls++;
-  this->MooeeInvTime -= usecond();
-
-  int nloop = grid->oSites()/Ls;
-  accelerator_for(sss,nloop,Simd::Nsimd(),{
-
-    uint64_t ss = sss*Ls;
-
-    typedef decltype(coalescedRead(psi[0])) spinor;
-    spinor tmp1,tmp2,tmp2_spProj;
-
-    // Apply (U^{\prime})^{-dag} and accumulate MooeeInvDag_shift_lc[j]*psi[j] in tmp2
-    coalescedWrite(chi[ss], psi(ss));
-    tmp2 = pMooeeInvDag_shift_lc[0]*psi(ss);
-    for(int s=1; s<Ls; s++){
-      spProj5m(tmp1, chi(ss+s-1));
-      coalescedWrite(chi[ss+s],psi(ss+s) - puee[s-1]*tmp1);
-      tmp2 = tmp2 + pMooeeInvDag_shift_lc[s]*psi(ss+s);
-    }
-
-    if(pm == 1){ spProj5p(tmp2_spProj, tmp2);}
-    else       { spProj5m(tmp2_spProj, tmp2);}
-
-    // U_m^{-\dag}
-    for(int s=0; s<Ls-1; s++){
-      spProj5p(tmp1, chi(ss+s));
-      coalescedWrite(chi[ss+Ls-1], chi(ss+Ls-1) - pueem[s]*tmp1);
-    }
-
-    // L_m^{-\dag} D^{-dag}
-    for(int s=0; s<Ls-1; s++){
-      spProj5m(tmp1, chi(ss+Ls-1));
-      coalescedWrite(chi[ss+s], (1.0/pdee[s])*chi(ss+s) - (pleem[s]/pdee[Ls-1])*tmp1);
-    }
-    coalescedWrite(chi[ss+Ls-1], (1.0/pdee[Ls-1])*chi(ss+Ls-1));
-    spProj5p(tmp1, chi(ss+Ls-1));
-    coalescedWrite(chi[ss+Ls-1], chi(ss+Ls-1) + pMooeeInvDag_shift_norm[Ls-1]*tmp2_spProj);
-
-    // Apply L^{-dag}
-    for(int s=Ls-2; s>=0; s--){
-      coalescedWrite(chi[ss+s], chi(ss+s) - plee[s]*tmp1);
-      spProj5p(tmp1, chi(ss+s));
-      coalescedWrite(chi[ss+s], chi(ss+s) + pMooeeInvDag_shift_norm[s]*tmp2_spProj);
-    }
-  });
-
-  this->MooeeInvTime += usecond();
-}
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/implementation/MobiusEOFAFermionImplementation.h

@@ -1,407 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/MobiusEOFAFermion.cc
-
-Copyright (C) 2017
-
-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: David Murphy <dmurphy@phys.columbia.edu>
-
-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/FermionCore.h>
-#include <Grid/qcd/action/fermion/MobiusEOFAFermion.h>
-
-NAMESPACE_BEGIN(Grid);
-
-template<class Impl>
-MobiusEOFAFermion<Impl>::MobiusEOFAFermion(
-					   GaugeField            &_Umu,
-					   GridCartesian         &FiveDimGrid,
-					   GridRedBlackCartesian &FiveDimRedBlackGrid,
-					   GridCartesian         &FourDimGrid,
-					   GridRedBlackCartesian &FourDimRedBlackGrid,
-					   RealD _mq1, RealD _mq2, RealD _mq3,
-					   RealD _shift, int _pm, RealD _M5,
-					   RealD _b, RealD _c, const ImplParams &p) :
-  AbstractEOFAFermion<Impl>(_Umu, FiveDimGrid, FiveDimRedBlackGrid,
-			    FourDimGrid, FourDimRedBlackGrid, _mq1, _mq2, _mq3,
-			    _shift, _pm, _M5, _b, _c, p)
-{
-  int Ls = this->Ls;
-
-  RealD eps = 1.0;
-  Approx::zolotarev_data *zdata = Approx::higham(eps, this->Ls);
-  assert(zdata->n == this->Ls);
-
-  std::cout << GridLogMessage << "MobiusEOFAFermion (b=" << _b <<
-    ",c=" << _c << ") with Ls=" << Ls << std::endl;
-  this->SetCoefficientsTanh(zdata, _b, _c);
-  std::cout << GridLogMessage << "EOFA parameters: (mq1=" << _mq1 <<
-    ",mq2=" << _mq2 << ",mq3=" << _mq3 << ",shift=" << _shift <<
-    ",pm=" << _pm << ")" << std::endl;
-
-  Approx::zolotarev_free(zdata);
-
-  if(_shift != 0.0){
-    SetCoefficientsPrecondShiftOps();
-  } else {
-    Mooee_shift.resize(Ls, 0.0);
-    MooeeInv_shift_lc.resize(Ls, 0.0);
-    MooeeInv_shift_norm.resize(Ls, 0.0);
-    MooeeInvDag_shift_lc.resize(Ls, 0.0);
-    MooeeInvDag_shift_norm.resize(Ls, 0.0);
-  }
-}
-
-/****************************************************************
- * Additional EOFA operators only called outside the inverter.  
- * Since speed is not essential, simple axpby-style
- * implementations should be fine.
- ***************************************************************/
-template<class Impl>
-void MobiusEOFAFermion<Impl>::Omega(const FermionField& psi, FermionField& Din, int sign, int dag)
-{
-  int Ls = this->Ls;
-  RealD alpha = this->alpha;
-
-  Din = Zero();
-  if((sign == 1) && (dag == 0)) { // \Omega_{+}
-    for(int s=0; s<Ls; ++s){
-      axpby_ssp(Din, 0.0, psi, 2.0*std::pow(1.0-alpha,Ls-s-1)/std::pow(1.0+alpha,Ls-s), psi, s, 0);
-    }
-  } else if((sign == -1) && (dag == 0)) { // \Omega_{-}
-    for(int s=0; s<Ls; ++s){
-      axpby_ssp(Din, 0.0, psi, 2.0*std::pow(1.0-alpha,s)/std::pow(1.0+alpha,s+1), psi, s, 0);
-    }
-  } else if((sign == 1 ) && (dag == 1)) { // \Omega_{+}^{\dagger}
-    for(int sp=0; sp<Ls; ++sp){
-      axpby_ssp(Din, 1.0, Din, 2.0*std::pow(1.0-alpha,Ls-sp-1)/std::pow(1.0+alpha,Ls-sp), psi, 0, sp);
-    }
-  } else if((sign == -1) && (dag == 1)) { // \Omega_{-}^{\dagger}
-    for(int sp=0; sp<Ls; ++sp){
-      axpby_ssp(Din, 1.0, Din, 2.0*std::pow(1.0-alpha,sp)/std::pow(1.0+alpha,sp+1), psi, 0, sp);
-    }
-  }
-}
-
-// This is the operator relating the usual Ddwf to TWQCD's EOFA Dirac operator (arXiv:1706.05843, Eqn. 6).
-// It also relates the preconditioned and unpreconditioned systems described in Appendix B.2.
-template<class Impl>
-void MobiusEOFAFermion<Impl>::Dtilde(const FermionField& psi, FermionField& chi)
-{
-  int Ls    = this->Ls;
-  RealD b   = 0.5 * ( 1.0 + this->alpha );
-  RealD c   = 0.5 * ( 1.0 - this->alpha );
-  RealD mq1 = this->mq1;
-
-  for(int s=0; s<Ls; ++s){
-    if(s == 0) {
-      axpby_ssp_pminus(chi, b, psi, -c, psi, s, s+1);
-      axpby_ssp_pplus (chi, 1.0, chi, mq1*c, psi, s, Ls-1);
-    } else if(s == (Ls-1)) {
-      axpby_ssp_pminus(chi, b, psi, mq1*c, psi, s, 0);
-      axpby_ssp_pplus (chi, 1.0, chi, -c, psi, s, s-1);
-    } else {
-      axpby_ssp_pminus(chi, b, psi, -c, psi, s, s+1);
-      axpby_ssp_pplus (chi, 1.0, chi, -c, psi, s, s-1);
-    }
-  }
-}
-
-template<class Impl>
-void MobiusEOFAFermion<Impl>::DtildeInv(const FermionField& psi, FermionField& chi)
-{
-  int Ls = this->Ls;
-  RealD m = this->mq1;
-  RealD c = 0.5 * this->alpha;
-  RealD d = 0.5;
-
-  RealD DtInv_p(0.0), DtInv_m(0.0);
-  RealD N = std::pow(c+d,Ls) + m*std::pow(c-d,Ls);
-  FermionField tmp(this->FermionGrid());
-
-  for(int s=0; s<Ls; ++s){
-    for(int sp=0; sp<Ls; ++sp){
-
-      DtInv_p = m * std::pow(-1.0,s-sp+1) * std::pow(c-d,Ls+s-sp) / std::pow(c+d,s-sp+1) / N;
-      DtInv_p += (s < sp) ? 0.0 : std::pow(-1.0,s-sp) * std::pow(c-d,s-sp) / std::pow(c+d,s-sp+1);
-      DtInv_m = m * std::pow(-1.0,sp-s+1) * std::pow(c-d,Ls+sp-s) / std::pow(c+d,sp-s+1) / N;
-      DtInv_m += (s > sp) ? 0.0 : std::pow(-1.0,sp-s) * std::pow(c-d,sp-s) / std::pow(c+d,sp-s+1);
-
-      if(sp == 0){
-	axpby_ssp_pplus (tmp, 0.0, tmp, DtInv_p, psi, s, sp);
-	axpby_ssp_pminus(tmp, 0.0, tmp, DtInv_m, psi, s, sp);
-      } else {
-	axpby_ssp_pplus (tmp, 1.0, tmp, DtInv_p, psi, s, sp);
-	axpby_ssp_pminus(tmp, 1.0, tmp, DtInv_m, psi, s, sp);
-      }
-
-    }}
-}
-
-/*****************************************************************************************************/
-
-template<class Impl>
-RealD MobiusEOFAFermion<Impl>::M(const FermionField& psi, FermionField& chi)
-{
-  FermionField Din(psi.Grid());
-
-  this->Meooe5D(psi, Din);
-  this->DW(Din, chi, DaggerNo);
-  axpby(chi, 1.0, 1.0, chi, psi);
-  this->M5D(psi, chi);
-  return(norm2(chi));
-}
-
-template<class Impl>
-RealD MobiusEOFAFermion<Impl>::Mdag(const FermionField& psi, FermionField& chi)
-{
-  FermionField Din(psi.Grid());
-
-  this->DW(psi, Din, DaggerYes);
-  this->MeooeDag5D(Din, chi);
-  this->M5Ddag(psi, chi);
-  axpby(chi, 1.0, 1.0, chi, psi);
-  return(norm2(chi));
-}
-
-/********************************************************************
- * Performance critical fermion operators called inside the inverter
- ********************************************************************/
-
-template<class Impl>
-void MobiusEOFAFermion<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] = this->mq1;
-  Vector<Coeff_t> lower(Ls,-1.0);  lower[0]    = this->mq1;
-
-  // no shift term
-  if(this->shift == 0.0){ this->M5D(psi, chi, chi, lower, diag, upper); }
-
-  // fused M + shift operation
-  else{ this->M5D_shift(psi, chi, chi, lower, diag, upper, Mooee_shift); }
-}
-
-template<class Impl>
-void MobiusEOFAFermion<Impl>::M5Ddag(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] = this->mq1;
-  Vector<Coeff_t> lower(Ls,-1.0);  lower[0]    = this->mq1;
-
-  // no shift term
-  if(this->shift == 0.0){ this->M5Ddag(psi, chi, chi, lower, diag, upper); }
-
-  // fused M + shift operation
-  else{ this->M5Ddag_shift(psi, chi, chi, lower, diag, upper, Mooee_shift); }
-}
-
-// half checkerboard operations
-template<class Impl>
-void MobiusEOFAFermion<Impl>::Mooee(const FermionField& psi, FermionField& chi)
-{
-  int Ls = this->Ls;
-
-  // coefficients of Mooee
-  Vector<Coeff_t> diag = this->bee;
-  Vector<Coeff_t> upper(Ls);
-  Vector<Coeff_t> lower(Ls);
-  for(int s=0; s<Ls; s++){
-    upper[s] = -this->cee[s];
-    lower[s] = -this->cee[s];
-  }
-  upper[Ls-1] *= -this->mq1;
-  lower[0]    *= -this->mq1;
-
-  // no shift term
-  if(this->shift == 0.0){ this->M5D(psi, psi, chi, lower, diag, upper); }
-
-  // fused M + shift operation
-  else { this->M5D_shift(psi, psi, chi, lower, diag, upper, Mooee_shift); }
-}
-
-template<class Impl>
-void MobiusEOFAFermion<Impl>::MooeeDag(const FermionField& psi, FermionField& chi)
-{
-  int Ls = this->Ls;
-
-  // coefficients of MooeeDag
-  Vector<Coeff_t> diag = this->bee;
-  Vector<Coeff_t> upper(Ls);
-  Vector<Coeff_t> lower(Ls);
-  for(int s=0; s<Ls; s++){
-    if(s==0) {
-      upper[s] = -this->cee[s+1];
-      lower[s] = this->mq1*this->cee[Ls-1];
-    } else if(s==(Ls-1)) {
-      upper[s] = this->mq1*this->cee[0];
-      lower[s] = -this->cee[s-1];
-    } else {
-      upper[s] = -this->cee[s+1];
-      lower[s] = -this->cee[s-1];
-    }
-  }
-
-  // no shift term
-  if(this->shift == 0.0){ this->M5Ddag(psi, psi, chi, lower, diag, upper); }
-
-  // fused M + shift operation
-  else{ this->M5Ddag_shift(psi, psi, chi, lower, diag, upper, Mooee_shift); }
-}
-
-/****************************************************************************************/
-
-// Computes coefficients for applying Cayley preconditioned shift operators
-//  (Mooee + \Delta) --> Mooee_shift
-//  (Mooee + \Delta)^{-1} --> MooeeInv_shift_lc, MooeeInv_shift_norm
-//  (Mooee + \Delta)^{-dag} --> MooeeInvDag_shift_lc, MooeeInvDag_shift_norm
-// For the latter two cases, the operation takes the form
-//  [ (Mooee + \Delta)^{-1} \psi ]_{i} = Mooee_{ij} \psi_{j} +
-//      ( MooeeInv_shift_norm )_{i} ( \sum_{j} [ MooeeInv_shift_lc ]_{j} P_{pm} \psi_{j} )
-template<class Impl>
-void MobiusEOFAFermion<Impl>::SetCoefficientsPrecondShiftOps()
-{
-  int   Ls    = this->Ls;
-  int   pm    = this->pm;
-  RealD alpha = this->alpha;
-  RealD k     = this->k;
-  RealD mq1   = this->mq1;
-  RealD shift = this->shift;
-
-  // Initialize
-  Mooee_shift.resize(Ls);
-  MooeeInv_shift_lc.resize(Ls);
-  MooeeInv_shift_norm.resize(Ls);
-  MooeeInvDag_shift_lc.resize(Ls);
-  MooeeInvDag_shift_norm.resize(Ls);
-
-  // Construct Mooee_shift
-  int idx(0);
-  Coeff_t N = ( (pm == 1) ? 1.0 : -1.0 ) * (2.0*shift*k) *
-    ( std::pow(alpha+1.0,Ls) + mq1*std::pow(alpha-1.0,Ls) );
-  for(int s=0; s<Ls; ++s){
-    idx = (pm == 1) ? (s) : (Ls-1-s);
-    Mooee_shift[idx] = N * std::pow(-1.0,s) * std::pow(alpha-1.0,s) / std::pow(alpha+1.0,Ls+s+1);
-  }
-
-  // Tridiagonal solve for MooeeInvDag_shift_lc
-  {
-    Coeff_t m(0.0);
-    Vector<Coeff_t> d = Mooee_shift;
-    Vector<Coeff_t> u(Ls,0.0);
-    Vector<Coeff_t> y(Ls,0.0);
-    Vector<Coeff_t> q(Ls,0.0);
-    if(pm == 1){ u[0] = 1.0; }
-    else{ u[Ls-1] = 1.0; }
-
-    // Tridiagonal matrix algorithm + Sherman-Morrison formula
-    //
-    // We solve
-    //  ( Mooee' + u \otimes v ) MooeeInvDag_shift_lc = Mooee_shift
-    // where Mooee' is the tridiagonal part of Mooee_{+}, and
-    // u = (1,0,...,0) and v = (0,...,0,mq1*cee[0]) are chosen
-    // so that the outer-product u \otimes v gives the (0,Ls-1)
-    // entry of Mooee_{+}.
-    //
-    // We do this as two solves: Mooee'*y = d and Mooee'*q = u,
-    // and then construct the solution to the original system
-    //  MooeeInvDag_shift_lc = y - <v,y> / ( 1 + <v,q> ) q
-    if(pm == 1){
-      for(int s=1; s<Ls; ++s){
-	m = -this->cee[s] / this->bee[s-1];
-	d[s] -= m*d[s-1];
-	u[s] -= m*u[s-1];
-      }
-    }
-    y[Ls-1] = d[Ls-1] / this->bee[Ls-1];
-    q[Ls-1] = u[Ls-1] / this->bee[Ls-1];
-    for(int s=Ls-2; s>=0; --s){
-      if(pm == 1){
-	y[s] = d[s] / this->bee[s];
-	q[s] = u[s] / this->bee[s];
-      } else {
-	y[s] = ( d[s] + this->cee[s]*y[s+1] ) / this->bee[s];
-	q[s] = ( u[s] + this->cee[s]*q[s+1] ) / this->bee[s];
-      }
-    }
-
-    // Construct MooeeInvDag_shift_lc
-    for(int s=0; s<Ls; ++s){
-      if(pm == 1){
-	MooeeInvDag_shift_lc[s] = y[s] - mq1*this->cee[0]*y[Ls-1] /
-	  (1.0+mq1*this->cee[0]*q[Ls-1]) * q[s];
-      } else {
-	MooeeInvDag_shift_lc[s] = y[s] - mq1*this->cee[Ls-1]*y[0] /
-	  (1.0+mq1*this->cee[Ls-1]*q[0]) * q[s];
-      }
-    }
-
-    // Compute remaining coefficients
-    N = (pm == 1) ? (1.0 + MooeeInvDag_shift_lc[Ls-1]) : (1.0 + MooeeInvDag_shift_lc[0]);
-    for(int s=0; s<Ls; ++s){
-
-      // MooeeInv_shift_lc
-      if(pm == 1){ MooeeInv_shift_lc[s] = pow(this->bee[s],s)      * pow(this->cee[s],Ls-1-s); }
-      else       { MooeeInv_shift_lc[s] = pow(this->bee[s],Ls-1-s) * pow(this->cee[s],s); }
-
-      // MooeeInv_shift_norm
-      MooeeInv_shift_norm[s] = -MooeeInvDag_shift_lc[s] /
-	( pow(this->bee[s],Ls) + mq1*pow(this->cee[s],Ls) ) / N;
-
-      // MooeeInvDag_shift_norm
-      if(pm == 1){ MooeeInvDag_shift_norm[s] = -pow(this->bee[s],s) * pow(this->cee[s],(Ls-1-s)) /
-     	  ( pow(this->bee[s],Ls) + mq1*pow(this->cee[s],Ls) ) / N; }
-      else{ MooeeInvDag_shift_norm[s] = -pow(this->bee[s],(Ls-1-s)) * pow(this->cee[s],s) /
-	  ( pow(this->bee[s],Ls) + mq1*pow(this->cee[s],Ls) ) / N; }
-    }
-  }
-}
-
-// Recompute coefficients for a different value of shift constant
-template<class Impl>
-void MobiusEOFAFermion<Impl>::RefreshShiftCoefficients(RealD new_shift)
-{
-  this->shift = new_shift;
-  if(new_shift != 0.0){
-    SetCoefficientsPrecondShiftOps();
-  } else {
-    int Ls = this->Ls;
-    Mooee_shift.resize(Ls,0.0);
-    MooeeInv_shift_lc.resize(Ls,0.0);
-    MooeeInv_shift_norm.resize(Ls,0.0);
-    MooeeInvDag_shift_lc.resize(Ls,0.0);
-    MooeeInvDag_shift_norm.resize(Ls,0.0);
-  }
-}
-
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/implementation/PartialFractionFermion5DImplementation.h

@@ -1,450 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/PartialFractionFermion5D.cc
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-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 */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/PartialFractionFermion5D.h>
-
-NAMESPACE_BEGIN(Grid);
-
-template<class Impl>
-void  PartialFractionFermion5D<Impl>::Mdir (const FermionField &psi, FermionField &chi,int dir,int disp){
-  // this does both dag and undag but is trivial; make a common helper routing
-  int Ls = this->Ls;
-
-  this->DhopDir(psi,chi,dir,disp);
-
-  int nblock=(Ls-1)/2;
-  for(int b=0;b<nblock;b++){
-    int s = 2*b;
-    ag5xpby_ssp(chi,-scale,chi,0.0,chi,s,s); 
-    ag5xpby_ssp(chi, scale,chi,0.0,chi,s+1,s+1); 
-  }
-  ag5xpby_ssp(chi,p[nblock]*scale/amax,chi,0.0,chi,Ls-1,Ls-1);
-
-}
-template<class Impl>
-void   PartialFractionFermion5D<Impl>::Meooe_internal(const FermionField &psi, FermionField &chi,int dag)
-{
-  int Ls = this->Ls;
-  if ( psi.Checkerboard() == Odd ) {
-    this->DhopEO(psi,chi,DaggerNo);
-  } else {
-    this->DhopOE(psi,chi,DaggerNo);
-  }
-
-  int nblock=(Ls-1)/2;
-  for(int b=0;b<nblock;b++){
-    int s = 2*b;
-    ag5xpby_ssp(chi,-scale,chi,0.0,chi,s,s); 
-    ag5xpby_ssp(chi, scale,chi,0.0,chi,s+1,s+1); 
-  }
-  ag5xpby_ssp(chi,p[nblock]*scale/amax,chi,0.0,chi,Ls-1,Ls-1);
-}
-
-template<class Impl>
-void   PartialFractionFermion5D<Impl>::Mooee_internal(const FermionField &psi, FermionField &chi,int dag)
-{
-  // again dag and undag are trivially related
-  int sign = dag ? (-1) : 1;
-  int Ls = this->Ls;
-      
-  int nblock=(Ls-1)/2;
-  for(int b=0;b<nblock;b++){
-	
-    int s = 2*b;
-    RealD pp = p[nblock-1-b];
-    RealD qq = q[nblock-1-b];
-	
-    // Do each 2x2 block aligned at s and multiplies Dw site diagonal by G5 so Hw
-    ag5xpby_ssp(chi,-dw_diag*scale,psi,amax*sqrt(qq)*scale,psi, s  ,s+1); 
-    ag5xpby_ssp(chi, dw_diag*scale,psi,amax*sqrt(qq)*scale,psi, s+1,s);
-    axpby_ssp  (chi, 1.0, chi,sqrt(amax*pp)*scale*sign,psi,s+1,Ls-1);
-  }
-      
-  {
-    RealD R=(1+mass)/(1-mass);
-    //R g5 psi[Ls-1] + p[0] H
-    ag5xpbg5y_ssp(chi,R*scale,psi,p[nblock]*scale*dw_diag/amax,psi,Ls-1,Ls-1);
-	
-    for(int b=0;b<nblock;b++){
-      int s = 2*b+1;
-      RealD pp = p[nblock-1-b];
-      axpby_ssp(chi,1.0,chi,-sqrt(amax*pp)*scale*sign,psi,Ls-1,s);
-    }
-  }
-}
-
-template<class Impl>
-void   PartialFractionFermion5D<Impl>::MooeeInv_internal(const FermionField &psi, FermionField &chi,int dag)
-{
-  int sign = dag ? (-1) : 1;
-  int Ls = this->Ls;
-
-  FermionField tmp(psi.Grid());
-      
-  ///////////////////////////////////////////////////////////////////////////////////////
-  //Linv
-  ///////////////////////////////////////////////////////////////////////////////////////
-  int nblock=(Ls-1)/2;
-
-  axpy(chi,0.0,psi,psi); // Identity piece
-      
-  for(int b=0;b<nblock;b++){
-    int s = 2*b;
-    RealD pp = p[nblock-1-b];
-    RealD qq = q[nblock-1-b];
-    RealD coeff1=sign*sqrt(amax*amax*amax*pp*qq) / ( dw_diag*dw_diag + amax*amax* qq);
-    RealD coeff2=sign*sqrt(amax*pp)*dw_diag / ( dw_diag*dw_diag + amax*amax* qq); // Implicit g5 here
-    axpby_ssp  (chi,1.0,chi,coeff1,psi,Ls-1,s);
-    axpbg5y_ssp(chi,1.0,chi,coeff2,psi,Ls-1,s+1);
-  }
-      
-  ///////////////////////////////////////////////////////////////////////////////////////
-  //Dinv (note D isn't really diagonal -- just diagonal enough that we can still invert)
-  // Compute Seeinv (coeff of gamma5)
-  ///////////////////////////////////////////////////////////////////////////////////////
-  RealD R=(1+mass)/(1-mass);
-  RealD Seeinv = R + p[nblock]*dw_diag/amax;
-  for(int b=0;b<nblock;b++){
-    Seeinv += p[nblock-1-b]*dw_diag/amax / ( dw_diag*dw_diag/amax/amax + q[nblock-1-b]);
-  }    
-  Seeinv = 1.0/Seeinv;
-      
-  for(int b=0;b<nblock;b++){
-    int s = 2*b;
-    RealD pp = p[nblock-1-b];
-    RealD qq = q[nblock-1-b];
-    RealD coeff1=dw_diag / ( dw_diag*dw_diag + amax*amax* qq); // Implicit g5 here
-    RealD coeff2=amax*sqrt(qq) / ( dw_diag*dw_diag + amax*amax* qq);
-    ag5xpby_ssp  (tmp,-coeff1,chi,coeff2,chi,s,s+1);
-    ag5xpby_ssp  (tmp, coeff1,chi,coeff2,chi,s+1,s);
-  }
-  ag5xpby_ssp  (tmp, Seeinv,chi,0.0,chi,Ls-1,Ls-1);
-      
-  ///////////////////////////////////////////////////////////////////////////////////////
-  // Uinv
-  ///////////////////////////////////////////////////////////////////////////////////////
-  for(int b=0;b<nblock;b++){
-    int s = 2*b;
-    RealD pp = p[nblock-1-b];
-    RealD qq = q[nblock-1-b];
-    RealD coeff1=-sign*sqrt(amax*amax*amax*pp*qq) / ( dw_diag*dw_diag + amax*amax* qq);
-    RealD coeff2=-sign*sqrt(amax*pp)*dw_diag / ( dw_diag*dw_diag + amax*amax* qq); // Implicit g5 here
-    axpby_ssp  (chi,1.0/scale,tmp,coeff1/scale,tmp,s,Ls-1);
-    axpbg5y_ssp(chi,1.0/scale,tmp,coeff2/scale,tmp,s+1,Ls-1);
-  }
-  axpby_ssp  (chi, 1.0/scale,tmp,0.0,tmp,Ls-1,Ls-1);
-}
-
-template<class Impl>
-void   PartialFractionFermion5D<Impl>::M_internal(const FermionField &psi, FermionField &chi,int dag)
-{
-  FermionField D(psi.Grid());
-  
-  int Ls = this->Ls;
-  int sign = dag ? (-1) : 1;
-
-  // For partial frac Hw case (b5=c5=1) chroma quirkily computes
-  //
-  // Conventions for partfrac appear to be a mess.
-  // Tony's Nara lectures have
-  //
-  // BlockDiag(  H/p_i  1             | 1       )    
-  //          (  1      p_i H / q_i^2 | 0       )  
-  //           ---------------------------------
-  //           ( -1      0                | R  +p0 H  )
-  //
-  //Chroma     ( -2H    2sqrt(q_i)    |   0         )
-  //           (2 sqrt(q_i)   2H      |  2 sqrt(p_i) )
-  //           ---------------------------------
-  //           ( 0     -2 sqrt(p_i)   |  2 R gamma_5 + p0 2H
-  //
-  // Edwards/Joo/Kennedy/Wenger
-  //
-  // Here, the "beta's" selected by chroma to scale the unphysical bulk constraint fields
-  // incorporate the approx scale factor. This is obtained by propagating the
-  // scale on "H" out to the off diagonal elements as follows:
-  //
-  // BlockDiag(  H/p_i  1             | 1       ) 
-  //          (  1      p_i H / q_i^2 | 0       )  
-  //           ---------------------------------
-  //          ( -1      0                | R  + p_0 H  )
-  //
-  // becomes:
-  // BlockDiag(  H/ sp_i  1               | 1             ) 
-  //          (  1      sp_i H / s^2q_i^2 | 0             )  
-  //           ---------------------------------
-  //           ( -1      0                | R + p_0/s H   )
-  //
-  //
-  // This is implemented in Chroma by
-  //           p0' = p0/approxMax
-  //           p_i' = p_i*approxMax
-  //           q_i' = q_i*approxMax*approxMax
-  //
-  // After the equivalence transform is applied the matrix becomes
-  // 
-  //Chroma     ( -2H    sqrt(q'_i)    |   0         )
-  //           (sqrt(q'_i)   2H       |   sqrt(p'_i) )
-  //           ---------------------------------
-  //           ( 0     -sqrt(p'_i)    |  2 R gamma_5 + p'0 2H
-  //
-  //     =     ( -2H    sqrt(q_i)amax    |   0              )
-  //           (sqrt(q_i)amax   2H       |   sqrt(p_i*amax) )
-  //           ---------------------------------
-  //           ( 0     -sqrt(p_i)*amax   |  2 R gamma_5 + p0/amax 2H
-  //
-
-  this->DW(psi,D,DaggerNo); 
-
-  int nblock=(Ls-1)/2;
-  for(int b=0;b<nblock;b++){
-	
-    int s = 2*b;
-    double pp = p[nblock-1-b];
-    double qq = q[nblock-1-b];
-	
-    // Do each 2x2 block aligned at s and
-    ag5xpby_ssp(chi,-1.0*scale,D,amax*sqrt(qq)*scale,psi, s  ,s+1); // Multiplies Dw by G5 so Hw
-    ag5xpby_ssp(chi, 1.0*scale,D,amax*sqrt(qq)*scale,psi, s+1,s);
-	
-    // Pick up last column
-    axpby_ssp  (chi, 1.0, chi,sqrt(amax*pp)*scale*sign,psi,s+1,Ls-1);
-  }
-	
-  {
-    double R=(1+this->mass)/(1-this->mass);
-    //R g5 psi[Ls] + p[0] H
-    ag5xpbg5y_ssp(chi,R*scale,psi,p[nblock]*scale/amax,D,Ls-1,Ls-1);
-	
-    for(int b=0;b<nblock;b++){
-      int s = 2*b+1;
-      double pp = p[nblock-1-b];
-      axpby_ssp(chi,1.0,chi,-sqrt(amax*pp)*scale*sign,psi,Ls-1,s);
-    }
-  }
-
-}
-
-template<class Impl>
-RealD  PartialFractionFermion5D<Impl>::M    (const FermionField &in, FermionField &out)
-{
-  M_internal(in,out,DaggerNo);
-  return norm2(out);
-}
-template<class Impl>
-RealD  PartialFractionFermion5D<Impl>::Mdag (const FermionField &in, FermionField &out)
-{
-  M_internal(in,out,DaggerYes);
-  return norm2(out);
-}
-
-template<class Impl>
-void PartialFractionFermion5D<Impl>::Meooe       (const FermionField &in, FermionField &out)
-{
-  Meooe_internal(in,out,DaggerNo);
-}
-template<class Impl>
-void PartialFractionFermion5D<Impl>::MeooeDag    (const FermionField &in, FermionField &out)
-{
-  Meooe_internal(in,out,DaggerYes);
-}
-template<class Impl>
-void PartialFractionFermion5D<Impl>::Mooee       (const FermionField &in, FermionField &out)
-{
-  Mooee_internal(in,out,DaggerNo);
-}
-template<class Impl>
-void PartialFractionFermion5D<Impl>::MooeeDag    (const FermionField &in, FermionField &out)
-{
-  Mooee_internal(in,out,DaggerYes);
-}
-
-template<class Impl>
-void PartialFractionFermion5D<Impl>::MooeeInv    (const FermionField &in, FermionField &out)
-{
-  MooeeInv_internal(in,out,DaggerNo);
-}
-template<class Impl>
-void PartialFractionFermion5D<Impl>::MooeeInvDag (const FermionField &in, FermionField &out)
-{
-  MooeeInv_internal(in,out,DaggerYes);
-}
-
-
-// force terms; five routines; default to Dhop on diagonal
-template<class Impl>
-void PartialFractionFermion5D<Impl>::MDeriv  (GaugeField &mat,const FermionField &U,const FermionField &V,int dag)
-{
-  int Ls = this->Ls;
-
-  FermionField D(V.Grid());
-
-  int nblock=(Ls-1)/2;
-  for(int b=0;b<nblock;b++){
-    int s = 2*b;
-    ag5xpby_ssp(D,-scale,U,0.0,U,s,s); 
-    ag5xpby_ssp(D, scale,U,0.0,U,s+1,s+1); 
-  }
-  ag5xpby_ssp(D,p[nblock]*scale/amax,U,0.0,U,Ls-1,Ls-1);
-
-  this->DhopDeriv(mat,D,V,DaggerNo); 
-};
-template<class Impl>
-void PartialFractionFermion5D<Impl>::MoeDeriv(GaugeField &mat,const FermionField &U,const FermionField &V,int dag)
-{
-  int Ls = this->Ls;
-
-  FermionField D(V.Grid());
-
-  int nblock=(Ls-1)/2;
-  for(int b=0;b<nblock;b++){
-    int s = 2*b;
-    ag5xpby_ssp(D,-scale,U,0.0,U,s,s); 
-    ag5xpby_ssp(D, scale,U,0.0,U,s+1,s+1); 
-  }
-  ag5xpby_ssp(D,p[nblock]*scale/amax,U,0.0,U,Ls-1,Ls-1);
-
-  this->DhopDerivOE(mat,D,V,DaggerNo); 
-};
-template<class Impl>
-void PartialFractionFermion5D<Impl>::MeoDeriv(GaugeField &mat,const FermionField &U,const FermionField &V,int dag)
-{
-  int Ls = this->Ls;
-
-  FermionField D(V.Grid());
-
-  int nblock=(Ls-1)/2;
-  for(int b=0;b<nblock;b++){
-    int s = 2*b;
-    ag5xpby_ssp(D,-scale,U,0.0,U,s,s); 
-    ag5xpby_ssp(D, scale,U,0.0,U,s+1,s+1); 
-  }
-  ag5xpby_ssp(D,p[nblock]*scale/amax,U,0.0,U,Ls-1,Ls-1);
-
-  this->DhopDerivEO(mat,D,V,DaggerNo); 
-};
-
-template<class Impl>
-void  PartialFractionFermion5D<Impl>::SetCoefficientsTanh(Approx::zolotarev_data *zdata,RealD scale){
-  SetCoefficientsZolotarev(1.0/scale,zdata);
-}
-template<class Impl>
-void  PartialFractionFermion5D<Impl>::SetCoefficientsZolotarev(RealD zolo_hi,Approx::zolotarev_data *zdata){
-
-  // check on degree matching
-  //      std::cout<<GridLogMessage << Ls << " Ls"<<std::endl;
-  //      std::cout<<GridLogMessage << zdata->n  << " - n"<<std::endl;
-  //      std::cout<<GridLogMessage << zdata->da << " -da "<<std::endl;
-  //      std::cout<<GridLogMessage << zdata->db << " -db"<<std::endl;
-  //      std::cout<<GridLogMessage << zdata->dn << " -dn"<<std::endl;
-  //      std::cout<<GridLogMessage << zdata->dd << " -dd"<<std::endl;
-  int Ls = this->Ls;
-
-  assert(Ls == (2*zdata->da -1) );
-
-  // Part frac
-  //      RealD R;
-  R=(1+mass)/(1-mass);
-  dw_diag = (4.0-this->M5);
-
-  //      std::vector<RealD> p; 
-  //      std::vector<RealD> q;
-  p.resize(zdata->da);
-  q.resize(zdata->dd);
-	
-  for(int n=0;n<zdata->da;n++){
-    p[n] = zdata -> alpha[n];
-  }
-  for(int n=0;n<zdata->dd;n++){
-    q[n] = -zdata -> ap[n];
-  }
-      
-  scale= part_frac_chroma_convention ? 2.0 : 1.0; // Chroma conventions annoy me
-
-  amax=zolo_hi;
-}
-
-    template<class Impl>
-    void PartialFractionFermion5D<Impl>::ExportPhysicalFermionSolution(const FermionField &solution5d,FermionField &exported4d)
-    {
-      int Ls = this->Ls;
-      conformable(solution5d.Grid(),this->FermionGrid());
-      conformable(exported4d.Grid(),this->GaugeGrid());
-      ExtractSlice(exported4d, solution5d, Ls-1, Ls-1);
-    }
-    template<class Impl>
-    void PartialFractionFermion5D<Impl>::ImportPhysicalFermionSource(const FermionField &input4d,FermionField &imported5d)
-    {
-      int Ls = this->Ls;
-      conformable(imported5d.Grid(),this->FermionGrid());
-      conformable(input4d.Grid()   ,this->GaugeGrid());
-      FermionField tmp(this->FermionGrid());
-      tmp=Zero();
-      InsertSlice(input4d, tmp, Ls-1, Ls-1);
-      tmp=Gamma(Gamma::Algebra::Gamma5)*tmp;
-      this->Dminus(tmp,imported5d);
-    }
-
-// Constructors
-template<class Impl>
-PartialFractionFermion5D<Impl>::PartialFractionFermion5D(GaugeField &_Umu,
-							 GridCartesian         &FiveDimGrid,
-							 GridRedBlackCartesian &FiveDimRedBlackGrid,
-							 GridCartesian         &FourDimGrid,
-							 GridRedBlackCartesian &FourDimRedBlackGrid,
-							 RealD _mass,RealD M5,
-							 const ImplParams &p) :
-  WilsonFermion5D<Impl>(_Umu,
-			FiveDimGrid, FiveDimRedBlackGrid,
-			FourDimGrid, FourDimRedBlackGrid,M5,p),
-  mass(_mass)
-
-{
-  int Ls = this->Ls;
-
-  assert((Ls&0x1)==1); // Odd Ls required
-  int nrational=Ls-1;
-
-
-  Approx::zolotarev_data *zdata = Approx::higham(1.0,nrational);
-
-  // NB: chroma uses a cast to "float" for the zolotarev range(!?).
-  // this creates a real difference in the operator which I do not like but we can replicate here
-  // to demonstrate compatibility
-  //      RealD eps = (zolo_lo / zolo_hi);
-  //      zdata = bfm_zolotarev(eps,nrational,0);
-      
-  SetCoefficientsTanh(zdata,1.0);
-
-  Approx::zolotarev_free(zdata);
-
-}
-
-NAMESPACE_END(Grid);
-

Grid/qcd/action/fermion/implementation/StaggeredKernelsHand.h

@@ -1,393 +0,0 @@
-    /*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/StaggerdKernelsHand.cc
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is 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>
-
-#pragma once
-
-NAMESPACE_BEGIN(Grid);
-
-#define LOAD_CHI(b)		\
-  const SiteSpinor & ref (b[offset]);	\
-    Chi_0=ref()()(0);\
-    Chi_1=ref()()(1);\
-    Chi_2=ref()()(2);
-
-
-// To splat or not to splat depends on the implementation
-#define MULT(A,UChi)				\
-  auto & ref(U[sU](A));			\
-   Impl::loadLinkElement(U_00,ref()(0,0));      \
-   Impl::loadLinkElement(U_10,ref()(1,0));      \
-   Impl::loadLinkElement(U_20,ref()(2,0));      \
-   Impl::loadLinkElement(U_01,ref()(0,1));      \
-   Impl::loadLinkElement(U_11,ref()(1,1));      \
-   Impl::loadLinkElement(U_21,ref()(2,1));      \
-   Impl::loadLinkElement(U_02,ref()(0,2));     \
-   Impl::loadLinkElement(U_12,ref()(1,2));     \
-   Impl::loadLinkElement(U_22,ref()(2,2));     \
-    UChi ## _0  = U_00*Chi_0;	       \
-    UChi ## _1  = U_10*Chi_0;\
-    UChi ## _2  = U_20*Chi_0;\
-    UChi ## _0 += U_01*Chi_1;\
-    UChi ## _1 += U_11*Chi_1;\
-    UChi ## _2 += U_21*Chi_1;\
-    UChi ## _0 += U_02*Chi_2;\
-    UChi ## _1 += U_12*Chi_2;\
-    UChi ## _2 += U_22*Chi_2;
-
-#define MULT_ADD(U,A,UChi)			\
-  auto & ref(U[sU](A));			\
-   Impl::loadLinkElement(U_00,ref()(0,0));      \
-   Impl::loadLinkElement(U_10,ref()(1,0));      \
-   Impl::loadLinkElement(U_20,ref()(2,0));      \
-   Impl::loadLinkElement(U_01,ref()(0,1));      \
-   Impl::loadLinkElement(U_11,ref()(1,1));      \
-   Impl::loadLinkElement(U_21,ref()(2,1));      \
-   Impl::loadLinkElement(U_02,ref()(0,2));     \
-   Impl::loadLinkElement(U_12,ref()(1,2));     \
-   Impl::loadLinkElement(U_22,ref()(2,2));     \
-    UChi ## _0 += U_00*Chi_0;	       \
-    UChi ## _1 += U_10*Chi_0;\
-    UChi ## _2 += U_20*Chi_0;\
-    UChi ## _0 += U_01*Chi_1;\
-    UChi ## _1 += U_11*Chi_1;\
-    UChi ## _2 += U_21*Chi_1;\
-    UChi ## _0 += U_02*Chi_2;\
-    UChi ## _1 += U_12*Chi_2;\
-    UChi ## _2 += U_22*Chi_2;
-
-
-#define PERMUTE_DIR(dir)			\
-  permute##dir(Chi_0,Chi_0);			\
-  permute##dir(Chi_1,Chi_1);			\
-  permute##dir(Chi_2,Chi_2);
-
-
-#define HAND_STENCIL_LEG_BASE(Dir,Perm,skew)	\
-  SE=st.GetEntry(ptype,Dir+skew,sF);	\
-  offset = SE->_offset;			\
-  local  = SE->_is_local;		\
-  perm   = SE->_permute;		\
-  if ( local ) {						\
-    LOAD_CHI(in);					\
-    if ( perm) {						\
-      PERMUTE_DIR(Perm);					\
-    }								\
-  } else {							\
-    LOAD_CHI(buf);						\
-  }								
-
-#define HAND_STENCIL_LEG_BEGIN(Dir,Perm,skew,even)		\
-  HAND_STENCIL_LEG_BASE(Dir,Perm,skew)				\
-  {								\
-    MULT(Dir,even);						\
-  }
-
-#define HAND_STENCIL_LEG(U,Dir,Perm,skew,even)			\
-  HAND_STENCIL_LEG_BASE(Dir,Perm,skew)				\
-  {								\
-    MULT_ADD(U,Dir,even);					\
-  }
-
-
-
-#define HAND_STENCIL_LEG_INT(U,Dir,Perm,skew,even)	\
-  SE=st.GetEntry(ptype,Dir+skew,sF);			\
-  offset = SE->_offset;					\
-  local  = SE->_is_local;				\
-  perm   = SE->_permute;				\
-  if ( local ) {					\
-    LOAD_CHI(in);				\
-    if ( perm) {					\
-      PERMUTE_DIR(Perm);				\
-    }							\
-  } else if ( st.same_node[Dir] ) {			\
-    LOAD_CHI(buf);					\
-  }							\
-  if (local || st.same_node[Dir] ) {		\
-    MULT_ADD(U,Dir,even);				\
-  }
-
-#define HAND_STENCIL_LEG_EXT(U,Dir,Perm,skew,even)	\
-  SE=st.GetEntry(ptype,Dir+skew,sF);			\
-  offset = SE->_offset;					\
-  local  = SE->_is_local;				\
-  if ((!local) && (!st.same_node[Dir]) ) {		\
-    nmu++;							\
-    { LOAD_CHI(buf);	  }					\
-    { MULT_ADD(U,Dir,even); }					\
-  }								
-
-
-template <class Impl>
-void StaggeredKernels<Impl>::DhopSiteHand(StencilImpl &st, LebesgueOrder &lo, 
-					  DoubledGaugeFieldView &U,DoubledGaugeFieldView &UUU,
-					  SiteSpinor *buf, int LLs, int sU, 
-					  const FermionFieldView &in, FermionFieldView &out,int dag) 
-{
-  typedef typename Simd::scalar_type S;
-  typedef typename Simd::vector_type V;
-
-  Simd even_0; // 12 regs on knc
-  Simd even_1;
-  Simd even_2;
-  Simd odd_0; // 12 regs on knc
-  Simd odd_1;
-  Simd odd_2;
-
-  Simd Chi_0;    // two spinor; 6 regs
-  Simd Chi_1;
-  Simd Chi_2;
-  
-  Simd U_00;  // two rows of U matrix
-  Simd U_10;
-  Simd U_20;  
-  Simd U_01;
-  Simd U_11;
-  Simd U_21;  // 2 reg left.
-  Simd U_02;
-  Simd U_12;
-  Simd U_22; 
-
-  SiteSpinor result;
-  int offset,local,perm, ptype;
-
-  StencilEntry *SE;
-  int skew;
-
-  for(int s=0;s<LLs;s++){
-    int sF=s+LLs*sU;
-
-    skew = 0;
-    HAND_STENCIL_LEG_BEGIN(Xp,3,skew,even);  
-    HAND_STENCIL_LEG_BEGIN(Yp,2,skew,odd);   
-    HAND_STENCIL_LEG      (U,Zp,1,skew,even);  
-    HAND_STENCIL_LEG      (U,Tp,0,skew,odd);  
-    HAND_STENCIL_LEG      (U,Xm,3,skew,even);  
-    HAND_STENCIL_LEG      (U,Ym,2,skew,odd);   
-    HAND_STENCIL_LEG      (U,Zm,1,skew,even);  
-    HAND_STENCIL_LEG      (U,Tm,0,skew,odd);  
-    skew = 8;
-    HAND_STENCIL_LEG(UUU,Xp,3,skew,even);  
-    HAND_STENCIL_LEG(UUU,Yp,2,skew,odd);   
-    HAND_STENCIL_LEG(UUU,Zp,1,skew,even);  
-    HAND_STENCIL_LEG(UUU,Tp,0,skew,odd);  
-    HAND_STENCIL_LEG(UUU,Xm,3,skew,even);  
-    HAND_STENCIL_LEG(UUU,Ym,2,skew,odd);   
-    HAND_STENCIL_LEG(UUU,Zm,1,skew,even);  
-    HAND_STENCIL_LEG(UUU,Tm,0,skew,odd);  
-    
-    if ( dag ) {
-      result()()(0) = - even_0 - odd_0;
-      result()()(1) = - even_1 - odd_1;
-      result()()(2) = - even_2 - odd_2;
-    } else { 
-      result()()(0) = even_0 + odd_0;
-      result()()(1) = even_1 + odd_1;
-      result()()(2) = even_2 + odd_2;
-    }
-    vstream(out[sF],result);
-  }
-}
-
-
-template <class Impl>
-void StaggeredKernels<Impl>::DhopSiteHandInt(StencilImpl &st, LebesgueOrder &lo, 
-					     DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU,
-					     SiteSpinor *buf, int LLs, int sU, 
-					     const FermionFieldView &in, FermionFieldView &out,int dag) 
-{
-  typedef typename Simd::scalar_type S;
-  typedef typename Simd::vector_type V;
-
-  Simd even_0; // 12 regs on knc
-  Simd even_1;
-  Simd even_2;
-  Simd odd_0; // 12 regs on knc
-  Simd odd_1;
-  Simd odd_2;
-
-  Simd Chi_0;    // two spinor; 6 regs
-  Simd Chi_1;
-  Simd Chi_2;
-  
-  Simd U_00;  // two rows of U matrix
-  Simd U_10;
-  Simd U_20;  
-  Simd U_01;
-  Simd U_11;
-  Simd U_21;  // 2 reg left.
-  Simd U_02;
-  Simd U_12;
-  Simd U_22; 
-
-  SiteSpinor result;
-  int offset, ptype, local, perm;
-
-  StencilEntry *SE;
-  int skew;
-
-  for(int s=0;s<LLs;s++){
-    int sF=s+LLs*sU;
-
-    even_0 = Zero();    even_1 = Zero();    even_2 = Zero();
-     odd_0 = Zero();     odd_1 = Zero();     odd_2 = Zero();
-
-    skew = 0;
-    HAND_STENCIL_LEG_INT(U,Xp,3,skew,even);  
-    HAND_STENCIL_LEG_INT(U,Yp,2,skew,odd);   
-    HAND_STENCIL_LEG_INT(U,Zp,1,skew,even);  
-    HAND_STENCIL_LEG_INT(U,Tp,0,skew,odd);  
-    HAND_STENCIL_LEG_INT(U,Xm,3,skew,even);  
-    HAND_STENCIL_LEG_INT(U,Ym,2,skew,odd);   
-    HAND_STENCIL_LEG_INT(U,Zm,1,skew,even);  
-    HAND_STENCIL_LEG_INT(U,Tm,0,skew,odd);  
-    skew = 8;
-    HAND_STENCIL_LEG_INT(UUU,Xp,3,skew,even);  
-    HAND_STENCIL_LEG_INT(UUU,Yp,2,skew,odd);   
-    HAND_STENCIL_LEG_INT(UUU,Zp,1,skew,even);  
-    HAND_STENCIL_LEG_INT(UUU,Tp,0,skew,odd);  
-    HAND_STENCIL_LEG_INT(UUU,Xm,3,skew,even);  
-    HAND_STENCIL_LEG_INT(UUU,Ym,2,skew,odd);   
-    HAND_STENCIL_LEG_INT(UUU,Zm,1,skew,even);  
-    HAND_STENCIL_LEG_INT(UUU,Tm,0,skew,odd);  
-
-    // Assume every site must be connected to at least one interior point. No 1^4 subvols.
-    if ( dag ) {
-      result()()(0) = - even_0 - odd_0;
-      result()()(1) = - even_1 - odd_1;
-      result()()(2) = - even_2 - odd_2;
-    } else { 
-      result()()(0) = even_0 + odd_0;
-      result()()(1) = even_1 + odd_1;
-      result()()(2) = even_2 + odd_2;
-    }
-    vstream(out[sF],result);
-  }
-}
-
-
-template <class Impl>
-void StaggeredKernels<Impl>::DhopSiteHandExt(StencilImpl &st, LebesgueOrder &lo, 
-					     DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU,
-					     SiteSpinor *buf, int LLs, int sU, 
-					     const FermionFieldView &in, FermionFieldView &out,int dag) 
-{
-  typedef typename Simd::scalar_type S;
-  typedef typename Simd::vector_type V;
-
-  Simd even_0; // 12 regs on knc
-  Simd even_1;
-  Simd even_2;
-  Simd odd_0; // 12 regs on knc
-  Simd odd_1;
-  Simd odd_2;
-
-  Simd Chi_0;    // two spinor; 6 regs
-  Simd Chi_1;
-  Simd Chi_2;
-  
-  Simd U_00;  // two rows of U matrix
-  Simd U_10;
-  Simd U_20;  
-  Simd U_01;
-  Simd U_11;
-  Simd U_21;  // 2 reg left.
-  Simd U_02;
-  Simd U_12;
-  Simd U_22; 
-
-  SiteSpinor result;
-  int offset, ptype, local;
-
-  StencilEntry *SE;
-  int skew;
-
-  for(int s=0;s<LLs;s++){
-    int sF=s+LLs*sU;
-
-    even_0 = Zero();    even_1 = Zero();    even_2 = Zero();
-     odd_0 = Zero();     odd_1 = Zero();     odd_2 = Zero();
-    int nmu=0;
-    skew = 0;
-    HAND_STENCIL_LEG_EXT(U,Xp,3,skew,even);  
-    HAND_STENCIL_LEG_EXT(U,Yp,2,skew,odd);   
-    HAND_STENCIL_LEG_EXT(U,Zp,1,skew,even);  
-    HAND_STENCIL_LEG_EXT(U,Tp,0,skew,odd);  
-    HAND_STENCIL_LEG_EXT(U,Xm,3,skew,even);  
-    HAND_STENCIL_LEG_EXT(U,Ym,2,skew,odd);   
-    HAND_STENCIL_LEG_EXT(U,Zm,1,skew,even);  
-    HAND_STENCIL_LEG_EXT(U,Tm,0,skew,odd);  
-    skew = 8;
-    HAND_STENCIL_LEG_EXT(UUU,Xp,3,skew,even);  
-    HAND_STENCIL_LEG_EXT(UUU,Yp,2,skew,odd);   
-    HAND_STENCIL_LEG_EXT(UUU,Zp,1,skew,even);  
-    HAND_STENCIL_LEG_EXT(UUU,Tp,0,skew,odd);  
-    HAND_STENCIL_LEG_EXT(UUU,Xm,3,skew,even);  
-    HAND_STENCIL_LEG_EXT(UUU,Ym,2,skew,odd);   
-    HAND_STENCIL_LEG_EXT(UUU,Zm,1,skew,even);  
-    HAND_STENCIL_LEG_EXT(UUU,Tm,0,skew,odd);  
-
-    // Add sum of all exterior connected stencil legs
-    if ( nmu ) { 
-      if ( dag ) {
-	result()()(0) = - even_0 - odd_0;
-	result()()(1) = - even_1 - odd_1;
-	result()()(2) = - even_2 - odd_2;
-      } else { 
-	result()()(0) = even_0 + odd_0;
-	result()()(1) = even_1 + odd_1;
-	result()()(2) = even_2 + odd_2;
-      }
-      out[sF] = out[sF] + result;
-    }
-  }
-}
-
-#define DHOP_SITE_HAND_INSTANTIATE(IMPL)				\
-  template void StaggeredKernels<IMPL>::DhopSiteHand(StencilImpl &st, LebesgueOrder &lo, \
-						     DoubledGaugeFieldView &U,DoubledGaugeFieldView &UUU, \
-						     SiteSpinor *buf, int LLs, int sU, \
-						     const FermionFieldView &in, FermionFieldView &out, int dag); \
-									\
-  template void StaggeredKernels<IMPL>::DhopSiteHandInt(StencilImpl &st, LebesgueOrder &lo, \
-						     DoubledGaugeFieldView &U,DoubledGaugeFieldView &UUU, \
-						     SiteSpinor *buf, int LLs, int sU, \
-						     const FermionFieldView &in, FermionFieldView &out, int dag); \
-									\
-  template void StaggeredKernels<IMPL>::DhopSiteHandExt(StencilImpl &st, LebesgueOrder &lo, \
-						     DoubledGaugeFieldView &U,DoubledGaugeFieldView &UUU, \
-						     SiteSpinor *buf, int LLs, int sU, \
-						     const FermionFieldView &in, FermionFieldView &out, int dag); \
-
-#undef LOAD_CHI
-
-NAMESPACE_END(Grid);
-
-

Grid/qcd/action/fermion/implementation/StaggeredKernelsImplementation.h

@@ -1,290 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Azusa Yamaguchi, Peter Boyle
-
-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/qcd/action/fermion/FermionCore.h>
-
-#pragma once
-
-NAMESPACE_BEGIN(Grid);
-
-#define GENERIC_STENCIL_LEG(U,Dir,skew,multLink)		\
-  SE = st.GetEntry(ptype, Dir+skew, sF);			\
-  if (SE->_is_local ) {						\
-    if (SE->_permute) {						\
-      chi_p = &chi;						\
-      permute(chi,  in[SE->_offset], ptype);		\
-    } else {							\
-      chi_p = &in[SE->_offset];				\
-    }								\
-  } else {							\
-    chi_p = &buf[SE->_offset];					\
-  }								\
-  multLink(Uchi, U[sU], *chi_p, Dir);			
-
-#define GENERIC_STENCIL_LEG_INT(U,Dir,skew,multLink)		\
-  SE = st.GetEntry(ptype, Dir+skew, sF);			\
-  if (SE->_is_local ) {						\
-    if (SE->_permute) {						\
-      chi_p = &chi;						\
-      permute(chi,  in[SE->_offset], ptype);		\
-    } else {							\
-      chi_p = &in[SE->_offset];				\
-    }								\
-  } else if ( st.same_node[Dir] ) {				\
-    chi_p = &buf[SE->_offset];					\
-  }								\
-  if (SE->_is_local || st.same_node[Dir] ) {			\
-    multLink(Uchi, U[sU], *chi_p, Dir);			\
-  }
-
-#define GENERIC_STENCIL_LEG_EXT(U,Dir,skew,multLink)		\
-  SE = st.GetEntry(ptype, Dir+skew, sF);			\
-  if ((!SE->_is_local) && (!st.same_node[Dir]) ) {		\
-    nmu++;							\
-    chi_p = &buf[SE->_offset];					\
-    multLink(Uchi, U[sU], *chi_p, Dir);			\
-  }
-
-template <class Impl>
-StaggeredKernels<Impl>::StaggeredKernels(const ImplParams &p) : Base(p){};
-
-////////////////////////////////////////////////////////////////////////////////////
-// Generic implementation; move to different file?
-// Int, Ext, Int+Ext cases for comms overlap
-////////////////////////////////////////////////////////////////////////////////////
-template <class Impl>
-void StaggeredKernels<Impl>::DhopSiteGeneric(StencilImpl &st, LebesgueOrder &lo, 
-					     DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU,
-					     SiteSpinor *buf, int LLs, int sU, 
-					     const FermionFieldView &in, FermionFieldView &out, int dag) {
-  const SiteSpinor *chi_p;
-  SiteSpinor chi;
-  SiteSpinor Uchi;
-  StencilEntry *SE;
-  int ptype;
-  int skew;
-
-  for(int s=0;s<LLs;s++){
-    int sF=LLs*sU+s;
-    skew = 0;
-    GENERIC_STENCIL_LEG(U,Xp,skew,Impl::multLink);
-    GENERIC_STENCIL_LEG(U,Yp,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG(U,Zp,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG(U,Tp,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG(U,Xm,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG(U,Ym,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG(U,Zm,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG(U,Tm,skew,Impl::multLinkAdd);
-    skew=8;
-    GENERIC_STENCIL_LEG(UUU,Xp,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG(UUU,Yp,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG(UUU,Zp,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG(UUU,Tp,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG(UUU,Xm,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG(UUU,Ym,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG(UUU,Zm,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG(UUU,Tm,skew,Impl::multLinkAdd);
-    if ( dag ) { 
-      Uchi = - Uchi;
-    } 
-    vstream(out[sF], Uchi);
-  }
-};
-
-  ///////////////////////////////////////////////////
-  // Only contributions from interior of our node
-  ///////////////////////////////////////////////////
-template <class Impl>
-void StaggeredKernels<Impl>::DhopSiteGenericInt(StencilImpl &st, LebesgueOrder &lo, 
-						DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU,
-						SiteSpinor *buf, int LLs, int sU, 
-						const FermionFieldView &in, FermionFieldView &out,int dag) {
-  const SiteSpinor *chi_p;
-  SiteSpinor chi;
-  SiteSpinor Uchi;
-  StencilEntry *SE;
-  int ptype;
-  int skew ;
-
-  for(int s=0;s<LLs;s++){
-    int sF=LLs*sU+s;
-    skew = 0;
-    Uchi=Zero();
-    GENERIC_STENCIL_LEG_INT(U,Xp,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG_INT(U,Yp,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG_INT(U,Zp,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG_INT(U,Tp,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG_INT(U,Xm,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG_INT(U,Ym,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG_INT(U,Zm,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG_INT(U,Tm,skew,Impl::multLinkAdd);
-    skew=8;
-    GENERIC_STENCIL_LEG_INT(UUU,Xp,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG_INT(UUU,Yp,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG_INT(UUU,Zp,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG_INT(UUU,Tp,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG_INT(UUU,Xm,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG_INT(UUU,Ym,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG_INT(UUU,Zm,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG_INT(UUU,Tm,skew,Impl::multLinkAdd);
-    if ( dag ) {
-      Uchi = - Uchi;
-    }
-    vstream(out[sF], Uchi);
-  }
-};
-
-
-  ///////////////////////////////////////////////////
-  // Only contributions from exterior of our node
-  ///////////////////////////////////////////////////
-template <class Impl>
-void StaggeredKernels<Impl>::DhopSiteGenericExt(StencilImpl &st, LebesgueOrder &lo, 
-						DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU,
-						SiteSpinor *buf, int LLs, int sU,
-						const FermionFieldView &in, FermionFieldView &out,int dag) {
-  const SiteSpinor *chi_p;
-  //  SiteSpinor chi;
-  SiteSpinor Uchi;
-  StencilEntry *SE;
-  int ptype;
-  int nmu=0;
-  int skew ;
-
-  for(int s=0;s<LLs;s++){
-    int sF=LLs*sU+s;
-    skew = 0;
-    Uchi=Zero();
-    GENERIC_STENCIL_LEG_EXT(U,Xp,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG_EXT(U,Yp,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG_EXT(U,Zp,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG_EXT(U,Tp,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG_EXT(U,Xm,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG_EXT(U,Ym,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG_EXT(U,Zm,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG_EXT(U,Tm,skew,Impl::multLinkAdd);
-    skew=8;
-    GENERIC_STENCIL_LEG_EXT(UUU,Xp,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG_EXT(UUU,Yp,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG_EXT(UUU,Zp,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG_EXT(UUU,Tp,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG_EXT(UUU,Xm,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG_EXT(UUU,Ym,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG_EXT(UUU,Zm,skew,Impl::multLinkAdd);
-    GENERIC_STENCIL_LEG_EXT(UUU,Tm,skew,Impl::multLinkAdd);
-
-    if ( nmu ) { 
-      if ( dag ) { 
-	out[sF] = out[sF] - Uchi;
-      } else { 
-	out[sF] = out[sF] + Uchi;
-      }
-    }
-  }
-};
-
-////////////////////////////////////////////////////////////////////////////////////
-// Driving / wrapping routine to select right kernel
-////////////////////////////////////////////////////////////////////////////////////
-
-template <class Impl>
-void StaggeredKernels<Impl>::DhopSiteDag(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU,
-					 SiteSpinor *buf, int LLs, int sU,
-					 const FermionFieldView &in, FermionFieldView &out,
-					 int interior,int exterior)
-{
-  int dag=1;
-  DhopSite(st,lo,U,UUU,buf,LLs,sU,in,out,dag,interior,exterior);
-};
-
-template <class Impl>
-void StaggeredKernels<Impl>::DhopSite(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU,
-				      SiteSpinor *buf, int LLs, int sU,
-				      const FermionFieldView &in, FermionFieldView &out,
-				      int interior,int exterior)
-{
-  int dag=0;
-  DhopSite(st,lo,U,UUU,buf,LLs,sU,in,out,dag,interior,exterior);
-};
-
-template <class Impl>
-void StaggeredKernels<Impl>::DhopSite(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU,
-				      SiteSpinor *buf, int LLs,
-				      int sU, const FermionFieldView &in, FermionFieldView &out,
-				      int dag,int interior,int exterior) 
-{
-  switch(Opt) {
-#ifdef AVX512
-  case OptInlineAsm:
-    if ( interior && exterior ) {
-      DhopSiteAsm(st,lo,U,UUU,buf,LLs,sU,in,out,dag);
-    } else { 
-      std::cout << GridLogError << "Cannot overlap comms and compute with Staggered assembly"<<std::endl;
-      assert(0);
-    }
-    break;
-#endif
-  case OptHandUnroll:
-    if ( interior && exterior ) {
-      DhopSiteHand   (st,lo,U,UUU,buf,LLs,sU,in,out,dag);
-    } else if ( interior ) {
-      DhopSiteHandInt(st,lo,U,UUU,buf,LLs,sU,in,out,dag);
-    } else if ( exterior ) {
-      DhopSiteHandExt(st,lo,U,UUU,buf,LLs,sU,in,out,dag);
-    }
-    break;
-  case OptGeneric:
-    if ( interior && exterior ) {
-      DhopSiteGeneric   (st,lo,U,UUU,buf,LLs,sU,in,out,dag);
-    } else if ( interior ) {
-      DhopSiteGenericInt(st,lo,U,UUU,buf,LLs,sU,in,out,dag);
-    } else if ( exterior ) {
-      DhopSiteGenericExt(st,lo,U,UUU,buf,LLs,sU,in,out,dag);
-    }
-    break;
-  default:
-    std::cout<<"Oops Opt = "<<Opt<<std::endl;
-    assert(0);
-    break;
-  }
-};
-
-template <class Impl>
-void StaggeredKernels<Impl>::DhopDirKernel( StencilImpl &st, DoubledGaugeFieldView &U,  DoubledGaugeFieldView &UUU, SiteSpinor *buf, int sF,
-					    int sU, const FermionFieldView &in, FermionFieldView &out, int dir, int disp) 
-{
-  // Disp should be either +1,-1,+3,-3
-  // What about "dag" ?
-  // Because we work out pU . dS/dU 
-  // U
-  assert(0);
-}
-
-NAMESPACE_END(Grid);
-
-

Grid/qcd/action/fermion/implementation/WilsonKernelsAsmAvx512.h

@@ -1,716 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-
-
-    Source file: ./lib/qcd/action/fermion/WilsonKernelsAsmAvx512.h
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is 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
-
-#if defined(AVX512) 
-    ///////////////////////////////////////////////////////////
-    // If we are AVX512 specialise the single precision routine
-    ///////////////////////////////////////////////////////////
-#include <simd/Intel512wilson.h>
-#include <simd/Intel512single.h>
-
-/// Switch off the 5d vectorised code optimisations
-#undef DWFVEC5D
-
-static Vector<vComplexF> signsF;
-
-  template<typename vtype>    
-  int setupSigns(Vector<vtype>& signs ){
-    Vector<vtype> bother(2);
-    signs = bother;
-    vrsign(signs[0]);
-    visign(signs[1]);
-    return 1;
-  }
-
-  static int signInitF = setupSigns(signsF);
-
-#define MAYBEPERM(A,perm) if (perm) { A ; }
-#define MULT_2SPIN(ptr,pf) MULT_ADDSUB_2SPIN(ptr,pf)
-#define COMPLEX_SIGNS(isigns) vComplexF *isigns = &signsF[0];  
-  
-/////////////////////////////////////////////////////////////////
-// XYZT vectorised, undag Kernel, single
-/////////////////////////////////////////////////////////////////
-#undef KERNEL_DAG
-#define INTERIOR_AND_EXTERIOR
-#undef INTERIOR
-#undef EXTERIOR
-template<> void 
-WilsonKernels<WilsonImplF>::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<ZWilsonImplF>::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
-#undef EXTERIOR
-template<> void 
-WilsonKernels<WilsonImplF>::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<ZWilsonImplF>::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
-#undef INTERIOR
-#define EXTERIOR
-template<> void 
-WilsonKernels<WilsonImplF>::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<ZWilsonImplF>::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
-/////////////////////////////////////////////////////////////////
-#define KERNEL_DAG
-#define INTERIOR_AND_EXTERIOR
-#undef INTERIOR
-#undef EXTERIOR
-template<> void 
-WilsonKernels<WilsonImplF>::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<ZWilsonImplF>::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
-#undef EXTERIOR
-template<> void 
-WilsonKernels<WilsonImplF>::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<ZWilsonImplF>::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
-#define EXTERIOR
-template<> void 
-WilsonKernels<WilsonImplF>::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<ZWilsonImplF>::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
-#define MAYBEPERM(A,B) 
-#define MULT_2SPIN(ptr,pf) MULT_ADDSUB_2SPIN_LS(ptr,pf)
-				    
-/////////////////////////////////////////////////////////////////
-// Ls vectorised, undag Kernel, single
-/////////////////////////////////////////////////////////////////
-
-#ifdef DWFVEC5D
-
-#undef KERNEL_DAG
-#define INTERIOR_AND_EXTERIOR
-#undef INTERIOR
-#undef EXTERIOR
-template<> void 
-WilsonKernels<DomainWallVec5dImplF>::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<ZDomainWallVec5dImplF>::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
-#undef EXTERIOR
-template<> void 
-WilsonKernels<DomainWallVec5dImplF>::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<ZDomainWallVec5dImplF>::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
-#define EXTERIOR
-#undef  MULT_2SPIN
-#define MULT_2SPIN(ptr,pf) MULT_ADDSUB_2SPIN_LSNOPF(ptr,pf)
-template<> void 
-WilsonKernels<DomainWallVec5dImplF>::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<ZDomainWallVec5dImplF>::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
-/////////////////////////////////////////////////////////////////
-#define KERNEL_DAG
-#define INTERIOR_AND_EXTERIOR
-#undef INTERIOR
-#undef EXTERIOR
-template<> void 
-WilsonKernels<DomainWallVec5dImplF>::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<ZDomainWallVec5dImplF>::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
-#undef EXTERIOR
-template<> void 
-WilsonKernels<DomainWallVec5dImplF>::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<ZDomainWallVec5dImplF>::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
-#define EXTERIOR
-template<> void 
-WilsonKernels<DomainWallVec5dImplF>::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<ZDomainWallVec5dImplF>::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
-
-#undef COMPLEX_SIGNS
-#undef MAYBEPERM
-#undef MULT_2SPIN
-	
-
-
-///////////////////////////////////////////////////////////
-// If we are AVX512 specialise the double precision routine
-///////////////////////////////////////////////////////////
-
-#include <simd/Intel512double.h>
-    
-static Vector<vComplexD> signsD;
-static int signInitD = setupSigns(signsD);
-    
-#define MAYBEPERM(A,perm) if (perm) { A ; }
-#define MULT_2SPIN(ptr,pf) MULT_ADDSUB_2SPIN(ptr,pf)
-#define COMPLEX_SIGNS(isigns) vComplexD *isigns = &signsD[0];  
-
-
-#define INTERIOR_AND_EXTERIOR    
-#undef  INTERIOR
-#undef  EXTERIOR
-  
-/////////////////////////////////////////////////////////////////
-// XYZT vectorised, undag Kernel, single
-/////////////////////////////////////////////////////////////////
-#undef KERNEL_DAG
-#define INTERIOR_AND_EXTERIOR
-#undef INTERIOR
-#undef EXTERIOR
-template<> void 
-WilsonKernels<WilsonImplD>::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<ZWilsonImplD>::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
-#undef EXTERIOR
-template<> void 
-WilsonKernels<WilsonImplD>::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<ZWilsonImplD>::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
-#define EXTERIOR
-template<> void 
-WilsonKernels<WilsonImplD>::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<ZWilsonImplD>::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
-/////////////////////////////////////////////////////////////////
-#define KERNEL_DAG
-#define INTERIOR_AND_EXTERIOR
-#undef INTERIOR
-#undef EXTERIOR
-template<> void 
-WilsonKernels<WilsonImplD>::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<ZWilsonImplD>::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
-#undef EXTERIOR
-template<> void 
-WilsonKernels<WilsonImplD>::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<ZWilsonImplD>::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
-#define EXTERIOR
-template<> void 
-WilsonKernels<WilsonImplD>::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<ZWilsonImplD>::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
-#define MAYBEPERM(A,B) 
-#define MULT_2SPIN(ptr,pf) MULT_ADDSUB_2SPIN_LS(ptr,pf)
-				    
-/////////////////////////////////////////////////////////////////
-// Ls vectorised, undag Kernel, single
-/////////////////////////////////////////////////////////////////
-#ifdef DWFVEC5D
-
-#undef KERNEL_DAG
-#define INTERIOR_AND_EXTERIOR
-#undef INTERIOR
-#undef EXTERIOR
-template<> void 
-WilsonKernels<DomainWallVec5dImplD>::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<ZDomainWallVec5dImplD>::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
-#undef EXTERIOR
-template<> void 
-WilsonKernels<DomainWallVec5dImplD>::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<ZDomainWallVec5dImplD>::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
-#define EXTERIOR
-#undef  MULT_2SPIN
-#define MULT_2SPIN(ptr,pf) MULT_ADDSUB_2SPIN_LSNOPF(ptr,pf)
-template<> void 
-WilsonKernels<DomainWallVec5dImplD>::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<ZDomainWallVec5dImplD>::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
-/////////////////////////////////////////////////////////////////
-#define KERNEL_DAG
-#define INTERIOR_AND_EXTERIOR
-#undef INTERIOR
-#undef EXTERIOR
-template<> void 
-WilsonKernels<DomainWallVec5dImplD>::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<ZDomainWallVec5dImplD>::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
-#undef EXTERIOR
-template<> void 
-WilsonKernels<DomainWallVec5dImplD>::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<ZDomainWallVec5dImplD>::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
-#define EXTERIOR
-
-template<> void 
-WilsonKernels<DomainWallVec5dImplD>::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<ZDomainWallVec5dImplD>::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
-
-#undef COMPLEX_SIGNS
-#undef MAYBEPERM
-#undef MULT_2SPIN
-
-#undef Chi_00
-#undef Chi_01
-#undef Chi_02
-#undef Chi_10
-#undef Chi_11
-#undef Chi_12
-#undef Chi_20
-#undef Chi_21
-#undef Chi_22
-#undef Chi_30
-#undef Chi_31
-#undef Chi_32
-
-#undef UChi_00
-#undef UChi_01
-#undef UChi_02
-#undef UChi_10
-#undef UChi_11
-#undef UChi_12
-#undef UChi_20
-#undef UChi_21
-#undef UChi_22
-#undef UChi_30
-#undef UChi_31
-#undef UChi_32
-
-#undef Psi_00
-#undef Psi_01
-#undef Psi_02
-#undef Psi_10
-#undef Psi_11
-#undef Psi_12
-#undef Psi_20
-#undef Psi_21
-#undef Psi_22
-#undef Psi_30
-#undef Psi_31
-#undef Psi_32
-
-#undef Phi_00
-#undef Phi_01
-#undef Phi_02
-#undef Phi_10
-#undef Phi_11
-#undef Phi_12
-#undef Phi_20
-#undef Phi_21
-#undef Phi_22
-#undef Phi_30
-#undef Phi_31
-#undef Phi_32
-
-
-#endif //AVX512

Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h

@@ -1,86 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-
-
-    Source file: ./lib/qcd/action/fermion/WilsonKernelsAsm.cc
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Guido Cossu <guido.cossu@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
-
-#include <Grid/qcd/action/fermion/FermionCore.h>
-
-NAMESPACE_BEGIN(Grid);
-
-///////////////////////////////////////////////////////////
-// Default to no assembler implementation
-// Will specialise to 
-///////////////////////////////////////////////////////////
-template<class Impl> void 
-WilsonKernels<Impl >::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
-				  int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-{
-  assert(0);
-}
-
-template<class Impl> void 
-WilsonKernels<Impl >::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
-				     int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-{
-  assert(0);
-}
-
-template<class Impl> void 
-WilsonKernels<Impl >::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
-				     int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-{
-  assert(0);
-}
-
-template<class Impl> void 
-WilsonKernels<Impl >::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
-					int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-{
-  assert(0);
-}
-
-template<class Impl> void 
-WilsonKernels<Impl >::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
-				     int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-{
-  assert(0);
-}
-
-template<class Impl> void 
-WilsonKernels<Impl >::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
-					int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-{
-  assert(0);
-}
-
-NAMESPACE_END(Grid);
-

Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h

@@ -1,551 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is 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/FermionCore.h>
-
-NAMESPACE_BEGIN(Grid);
-
-
-////////////////////////////////////////////
-// Generic implementation; move to different file?
-////////////////////////////////////////////
-
-accelerator_inline void get_stencil(StencilEntry * mem, StencilEntry &chip)
-{
-#ifdef __CUDA_ARCH__
-  static_assert(sizeof(StencilEntry)==sizeof(uint4),"Unexpected Stencil Entry Size"); 
-  uint4 * mem_pun  = (uint4 *)mem; // force 128 bit loads
-  uint4 * chip_pun = (uint4 *)&chip;
-  * chip_pun = * mem_pun;
-#else 
-  chip = *mem;
-#endif
-  return;
-}
-  
-#define GENERIC_STENCIL_LEG(Dir,spProj,Recon)			\
-  SE = st.GetEntry(ptype, Dir, sF);				\
-  if (SE->_is_local) {						\
-    int perm= SE->_permute;					\
-    auto tmp = coalescedReadPermute(in[SE->_offset],ptype,perm,lane);	\
-    spProj(chi,tmp);						\
-  } else {							\
-    chi = coalescedRead(buf[SE->_offset],lane);			\
-  }								\
-  synchronise();						\
-  Impl::multLink(Uchi, U[sU], chi, Dir, SE, st);		\
-  Recon(result, Uchi);
-  
-#define GENERIC_STENCIL_LEG_INT(Dir,spProj,Recon)		\
-  SE = st.GetEntry(ptype, Dir, sF);				\
-  if (SE->_is_local) {						\
-    int perm= SE->_permute;					\
-    auto tmp = coalescedReadPermute(in[SE->_offset],ptype,perm,lane);	\
-    spProj(chi,tmp);						\
-  } else if ( st.same_node[Dir] ) {				\
-    chi = coalescedRead(buf[SE->_offset],lane);			\
-  }								\
-  synchronise();						\
-  if (SE->_is_local || st.same_node[Dir] ) {			\
-    Impl::multLink(Uchi, U[sU], chi, Dir, SE, st);		\
-    Recon(result, Uchi);					\
-  }								\
-  synchronise();						
-
-#define GENERIC_STENCIL_LEG_EXT(Dir,spProj,Recon)		\
-  SE = st.GetEntry(ptype, Dir, sF);				\
-  if ((!SE->_is_local) && (!st.same_node[Dir]) ) {		\
-    auto chi = coalescedRead(buf[SE->_offset],lane);		\
-    Impl::multLink(Uchi, U[sU], chi, Dir, SE, st);		\
-    Recon(result, Uchi);					\
-    nmu++;							\
-  }								\
-  synchronise();						
-
-#define GENERIC_DHOPDIR_LEG(Dir,spProj,Recon)			\
-  if (gamma == Dir) {						\
-    if (SE->_is_local ) {					\
-      int perm= SE->_permute;					\
-      auto tmp = coalescedReadPermute(in[SE->_offset],ptype,perm,lane);	\
-      spProj(chi,tmp);						\
-    } else {							\
-      chi = coalescedRead(buf[SE->_offset],lane);		\
-    }								\
-    synchronise();						\
-    Impl::multLink(Uchi, U[sU], chi, dir, SE, st);		\
-    Recon(result, Uchi);					\
-    synchronise();						\
-  }
-
-  ////////////////////////////////////////////////////////////////////
-  // All legs kernels ; comms then compute
-  ////////////////////////////////////////////////////////////////////
-template <class Impl>
-void WilsonKernels<Impl>::GenericDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,
-					     SiteHalfSpinor *buf, int sF,
-					     int sU, const FermionFieldView &in, FermionFieldView &out)
-{
-  typedef decltype(coalescedRead(buf[0]))   calcHalfSpinor;
-  typedef decltype(coalescedRead(in[0])) calcSpinor;
-  calcHalfSpinor chi;
-  //  calcHalfSpinor *chi_p;
-  calcHalfSpinor Uchi;
-  calcSpinor result;
-  StencilEntry *SE;
-  int ptype;
-  const int Nsimd = SiteHalfSpinor::Nsimd();
-  const int lane=SIMTlane(Nsimd);
-  GENERIC_STENCIL_LEG(Xp,spProjXp,spReconXp);
-  GENERIC_STENCIL_LEG(Yp,spProjYp,accumReconYp);
-  GENERIC_STENCIL_LEG(Zp,spProjZp,accumReconZp);
-  GENERIC_STENCIL_LEG(Tp,spProjTp,accumReconTp);
-  GENERIC_STENCIL_LEG(Xm,spProjXm,accumReconXm);
-  GENERIC_STENCIL_LEG(Ym,spProjYm,accumReconYm);
-  GENERIC_STENCIL_LEG(Zm,spProjZm,accumReconZm);
-  GENERIC_STENCIL_LEG(Tm,spProjTm,accumReconTm);
-  coalescedWrite(out[sF],result,lane);
-};
-
-template <class Impl>
-void WilsonKernels<Impl>::GenericDhopSite(StencilView &st, DoubledGaugeFieldView &U,
-					  SiteHalfSpinor *buf, int sF,
-					  int sU, const FermionFieldView &in, FermionFieldView &out) 
-{
-  typedef decltype(coalescedRead(buf[0])) calcHalfSpinor;
-  typedef decltype(coalescedRead(in[0]))  calcSpinor;
-  calcHalfSpinor chi;
-  //  calcHalfSpinor *chi_p;
-  calcHalfSpinor Uchi;
-  calcSpinor result;
-  StencilEntry *SE;
-  int ptype;
-
-  const int Nsimd = SiteHalfSpinor::Nsimd();
-  const int lane=SIMTlane(Nsimd);
-  GENERIC_STENCIL_LEG(Xm,spProjXp,spReconXp);
-  GENERIC_STENCIL_LEG(Ym,spProjYp,accumReconYp);
-  GENERIC_STENCIL_LEG(Zm,spProjZp,accumReconZp);
-  GENERIC_STENCIL_LEG(Tm,spProjTp,accumReconTp);
-  GENERIC_STENCIL_LEG(Xp,spProjXm,accumReconXm);
-  GENERIC_STENCIL_LEG(Yp,spProjYm,accumReconYm);
-  GENERIC_STENCIL_LEG(Zp,spProjZm,accumReconZm);
-  GENERIC_STENCIL_LEG(Tp,spProjTm,accumReconTm);
-  coalescedWrite(out[sF], result,lane);
-};
-  ////////////////////////////////////////////////////////////////////
-  // Interior kernels
-  ////////////////////////////////////////////////////////////////////
-template <class Impl>
-void WilsonKernels<Impl>::GenericDhopSiteDagInt(StencilView &st,  DoubledGaugeFieldView &U,
-						SiteHalfSpinor *buf, int sF,
-						int sU, const FermionFieldView &in, FermionFieldView &out)
-{
-  typedef decltype(coalescedRead(buf[0])) calcHalfSpinor;
-  typedef decltype(coalescedRead(in[0]))  calcSpinor;
-  calcHalfSpinor chi;
-  //  calcHalfSpinor *chi_p;
-  calcHalfSpinor Uchi;
-  calcSpinor result;
-  StencilEntry *SE;
-  int ptype;
-  const int Nsimd = SiteHalfSpinor::Nsimd();
-  const int lane=SIMTlane(Nsimd);
-
-  result=Zero();
-  GENERIC_STENCIL_LEG_INT(Xp,spProjXp,accumReconXp);
-  GENERIC_STENCIL_LEG_INT(Yp,spProjYp,accumReconYp);
-  GENERIC_STENCIL_LEG_INT(Zp,spProjZp,accumReconZp);
-  GENERIC_STENCIL_LEG_INT(Tp,spProjTp,accumReconTp);
-  GENERIC_STENCIL_LEG_INT(Xm,spProjXm,accumReconXm);
-  GENERIC_STENCIL_LEG_INT(Ym,spProjYm,accumReconYm);
-  GENERIC_STENCIL_LEG_INT(Zm,spProjZm,accumReconZm);
-  GENERIC_STENCIL_LEG_INT(Tm,spProjTm,accumReconTm);
-  coalescedWrite(out[sF], result,lane);
-};
-
-template <class Impl>
-void WilsonKernels<Impl>::GenericDhopSiteInt(StencilView &st,  DoubledGaugeFieldView &U,
-							 SiteHalfSpinor *buf, int sF,
-							 int sU, const FermionFieldView &in, FermionFieldView &out) 
-{
-  typedef decltype(coalescedRead(buf[0])) calcHalfSpinor;
-  typedef decltype(coalescedRead(in[0]))  calcSpinor;
-  const int Nsimd = SiteHalfSpinor::Nsimd();
-  const int lane=SIMTlane(Nsimd);
-
-  calcHalfSpinor chi;
-  //  calcHalfSpinor *chi_p;
-  calcHalfSpinor Uchi;
-  calcSpinor result;
-  StencilEntry *SE;
-  int ptype;
-  result=Zero();
-  GENERIC_STENCIL_LEG_INT(Xm,spProjXp,accumReconXp);
-  GENERIC_STENCIL_LEG_INT(Ym,spProjYp,accumReconYp);
-  GENERIC_STENCIL_LEG_INT(Zm,spProjZp,accumReconZp);
-  GENERIC_STENCIL_LEG_INT(Tm,spProjTp,accumReconTp);
-  GENERIC_STENCIL_LEG_INT(Xp,spProjXm,accumReconXm);
-  GENERIC_STENCIL_LEG_INT(Yp,spProjYm,accumReconYm);
-  GENERIC_STENCIL_LEG_INT(Zp,spProjZm,accumReconZm);
-  GENERIC_STENCIL_LEG_INT(Tp,spProjTm,accumReconTm);
-  coalescedWrite(out[sF], result,lane);
-};
-////////////////////////////////////////////////////////////////////
-// Exterior kernels
-////////////////////////////////////////////////////////////////////
-template <class Impl>
-void WilsonKernels<Impl>::GenericDhopSiteDagExt(StencilView &st,  DoubledGaugeFieldView &U,
-						SiteHalfSpinor *buf, int sF,
-						int sU, const FermionFieldView &in, FermionFieldView &out)
-{
-  typedef decltype(coalescedRead(buf[0])) calcHalfSpinor;
-  typedef decltype(coalescedRead(in[0]))  calcSpinor;
-  //  calcHalfSpinor *chi_p;
-  calcHalfSpinor Uchi;
-  calcSpinor result;
-  StencilEntry *SE;
-  int ptype;
-  int nmu=0;
-  const int Nsimd = SiteHalfSpinor::Nsimd();
-  const int lane=SIMTlane(Nsimd);
-  result=Zero();
-  GENERIC_STENCIL_LEG_EXT(Xp,spProjXp,accumReconXp);
-  GENERIC_STENCIL_LEG_EXT(Yp,spProjYp,accumReconYp);
-  GENERIC_STENCIL_LEG_EXT(Zp,spProjZp,accumReconZp);
-  GENERIC_STENCIL_LEG_EXT(Tp,spProjTp,accumReconTp);
-  GENERIC_STENCIL_LEG_EXT(Xm,spProjXm,accumReconXm);
-  GENERIC_STENCIL_LEG_EXT(Ym,spProjYm,accumReconYm);
-  GENERIC_STENCIL_LEG_EXT(Zm,spProjZm,accumReconZm);
-  GENERIC_STENCIL_LEG_EXT(Tm,spProjTm,accumReconTm);
-  if ( nmu ) { 
-    auto out_t = coalescedRead(out[sF],lane);
-    out_t = out_t + result;
-    coalescedWrite(out[sF],out_t,lane);
-  }
-};
-
-template <class Impl>
-void WilsonKernels<Impl>::GenericDhopSiteExt(StencilView &st,  DoubledGaugeFieldView &U,
-					     SiteHalfSpinor *buf, int sF,
-					     int sU, const FermionFieldView &in, FermionFieldView &out) 
-{
-  typedef decltype(coalescedRead(buf[0])) calcHalfSpinor;
-  typedef decltype(coalescedRead(in[0]))  calcSpinor;
-  //  calcHalfSpinor *chi_p;
-  calcHalfSpinor Uchi;
-  calcSpinor result;
-  StencilEntry *SE;
-  int ptype;
-  int nmu=0;
-  const int Nsimd = SiteHalfSpinor::Nsimd();
-  const int lane=SIMTlane(Nsimd);
-  result=Zero();
-  GENERIC_STENCIL_LEG_EXT(Xm,spProjXp,accumReconXp);
-  GENERIC_STENCIL_LEG_EXT(Ym,spProjYp,accumReconYp);
-  GENERIC_STENCIL_LEG_EXT(Zm,spProjZp,accumReconZp);
-  GENERIC_STENCIL_LEG_EXT(Tm,spProjTp,accumReconTp);
-  GENERIC_STENCIL_LEG_EXT(Xp,spProjXm,accumReconXm);
-  GENERIC_STENCIL_LEG_EXT(Yp,spProjYm,accumReconYm);
-  GENERIC_STENCIL_LEG_EXT(Zp,spProjZm,accumReconZm);
-  GENERIC_STENCIL_LEG_EXT(Tp,spProjTm,accumReconTm);
-  if ( nmu ) { 
-    auto out_t = coalescedRead(out[sF],lane);
-    out_t = out_t + result;
-    coalescedWrite(out[sF],out_t,lane);
-  }
-};
-
-template <class Impl>
-void WilsonKernels<Impl>::DhopDirK( StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf, int sF,
-				    int sU, const FermionFieldView &in, FermionFieldView &out, int dir, int gamma) 
-{
-  typedef decltype(coalescedRead(buf[0])) calcHalfSpinor;
-  typedef decltype(coalescedRead(in[0]))  calcSpinor;
-  calcHalfSpinor chi;
-  calcSpinor result;
-  calcHalfSpinor Uchi;
-  StencilEntry *SE;
-  int ptype;
-  const int Nsimd = SiteHalfSpinor::Nsimd();
-  const int lane=SIMTlane(Nsimd);
-
-  SE = st.GetEntry(ptype, dir, sF);
-  if (gamma == Xp) {						
-    if (SE->_is_local ) {					
-      int perm= SE->_permute;					
-      auto tmp = coalescedReadPermute(in[SE->_offset],ptype,perm,lane);	
-      spProjXp(chi,tmp);						
-    } else {							
-      chi = coalescedRead(buf[SE->_offset],lane);			
-    }								
-    Impl::multLink(Uchi, U[sU], chi, dir, SE, st);		
-    spReconXp(result, Uchi);					
-  }
-
-  GENERIC_DHOPDIR_LEG(Yp,spProjYp,spReconYp);
-  GENERIC_DHOPDIR_LEG(Zp,spProjZp,spReconZp);
-  GENERIC_DHOPDIR_LEG(Tp,spProjTp,spReconTp);
-  GENERIC_DHOPDIR_LEG(Xm,spProjXm,spReconXm);
-  GENERIC_DHOPDIR_LEG(Ym,spProjYm,spReconYm);
-  GENERIC_DHOPDIR_LEG(Zm,spProjZm,spReconZm);
-  GENERIC_DHOPDIR_LEG(Tm,spProjTm,spReconTm);
-  coalescedWrite(out[sF], result,lane);
-}
-
-template <class Impl>
-void WilsonKernels<Impl>::DhopDirKernel( StencilImpl &st, DoubledGaugeField &U,SiteHalfSpinor *buf, int Ls,
-					 int Nsite, const FermionField &in, FermionField &out, int dirdisp, int gamma) 
-{
-  assert(dirdisp<=7);
-  assert(dirdisp>=0);
-
-   auto U_v   = U.View();
-   auto in_v  = in.View();
-   auto out_v = out.View();
-   auto st_v  = st.View();
-   accelerator_for(ss,Nsite,Simd::Nsimd(),{
-    for(int s=0;s<Ls;s++){
-      int sU=ss;
-      int sF = s+Ls*sU; 
-      DhopDirK(st_v,U_v,st.CommBuf(),sF,sU,in_v,out_v,dirdisp,gamma);
-    }
-  });
-} 
-
-#define KERNEL_CALLNB(A) \
-  const uint64_t    NN = Nsite*Ls;					\
-  accelerator_forNB( ss, NN, Simd::Nsimd(), {				\
-      int sF = ss;							\
-      int sU = ss/Ls;							\
-      WilsonKernels<Impl>::A(st_v,U_v,buf,sF,sU,in_v,out_v);		\
-  });
-
-#define KERNEL_CALL(A) KERNEL_CALLNB(A); accelerator_barrier(); 
-
-#define ASM_CALL(A)							\
-  thread_for( ss, Nsite, {						\
-    int sU = ss;							\
-    int sF = ss*Ls;							\
-    WilsonKernels<Impl>::A(st_v,U_v,buf,sF,sU,Ls,1,in_v,out_v);		\
-  });
-
-template <class Impl>
-void WilsonKernels<Impl>::DhopKernel(int Opt,StencilImpl &st,  DoubledGaugeField &U, SiteHalfSpinor * buf,
-				     int Ls, int Nsite, const FermionField &in, FermionField &out,
-				     int interior,int exterior) 
-{
-    auto U_v   =   U.View();
-    auto in_v  =  in.View();
-    auto out_v = out.View();
-    auto st_v  =  st.View();
-
-   if( interior && exterior ) { 
-     if (Opt == WilsonKernelsStatic::OptGeneric    ) { KERNEL_CALL(GenericDhopSite); return;}
-#ifndef GRID_NVCC
-     if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSite);    return;}
-     if (Opt == WilsonKernelsStatic::OptInlineAsm  ) {  ASM_CALL(AsmDhopSite); printf(".");    return;}
-#endif
-   } else if( interior ) {
-     if (Opt == WilsonKernelsStatic::OptGeneric    ) { KERNEL_CALLNB(GenericDhopSiteInt); return;}
-#ifndef GRID_NVCC
-     if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALLNB(HandDhopSiteInt);    return;}
-     if (Opt == WilsonKernelsStatic::OptInlineAsm  ) {  ASM_CALL(AsmDhopSiteInt); printf("-");    return;}
-#endif
-   } else if( exterior ) { 
-     if (Opt == WilsonKernelsStatic::OptGeneric    ) { KERNEL_CALL(GenericDhopSiteExt); return;}
-#ifndef GRID_NVCC
-     if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSiteExt);    return;}
-     if (Opt == WilsonKernelsStatic::OptInlineAsm  ) {  ASM_CALL(AsmDhopSiteExt); printf("+");    return;}
-#endif
-   }
-   assert(0 && " Kernel optimisation case not covered ");
-  }
-  template <class Impl>
-  void WilsonKernels<Impl>::DhopDagKernel(int Opt,StencilImpl &st,  DoubledGaugeField &U, SiteHalfSpinor * buf,
-					  int Ls, int Nsite, const FermionField &in, FermionField &out,
-					  int interior,int exterior) 
-  {
-    auto U_v   = U.View();
-    auto in_v  = in.View();
-    auto out_v = out.View();
-    auto st_v  = st.View();
-
-   if( interior && exterior ) { 
-     if (Opt == WilsonKernelsStatic::OptGeneric    ) { KERNEL_CALL(GenericDhopSiteDag); return;}
-#ifndef GRID_NVCC
-     if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSiteDag);    return;}
-     if (Opt == WilsonKernelsStatic::OptInlineAsm  ) {  ASM_CALL(AsmDhopSiteDag);     return;}
-#endif
-   } else if( interior ) {
-     if (Opt == WilsonKernelsStatic::OptGeneric    ) { KERNEL_CALL(GenericDhopSiteDagInt); return;}
-#ifndef GRID_NVCC
-     if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSiteDagInt);    return;}
-     if (Opt == WilsonKernelsStatic::OptInlineAsm  ) {  ASM_CALL(AsmDhopSiteDagInt);     return;}
-#endif
-   } else if( exterior ) { 
-     if (Opt == WilsonKernelsStatic::OptGeneric    ) { KERNEL_CALL(GenericDhopSiteDagExt); return;}
-#ifndef GRID_NVCC
-     if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSiteDagExt);    return;}
-     if (Opt == WilsonKernelsStatic::OptInlineAsm  ) {  ASM_CALL(AsmDhopSiteDagExt);     return;}
-#endif
-   }
-   assert(0 && " Kernel optimisation case not covered ");
-  }
-
-/*******************************************************************************
- * Conserved current utilities for Wilson fermions, for contracting propagators
- * to make a conserved current sink or inserting the conserved current 
- * sequentially. Common to both 4D and 5D.
- ******************************************************************************/
-// N.B. Functions below assume a -1/2 factor within U.
-#define WilsonCurrentFwd(expr, mu) ((expr - Gamma::gmu[mu]*expr))
-#define WilsonCurrentBwd(expr, mu) ((expr + Gamma::gmu[mu]*expr))
-
-/*******************************************************************************
- * Name: ContractConservedCurrentSiteFwd
- * Operation: (1/2) * q2[x] * U(x) * (g[mu] - 1) * q1[x + mu]
- * Notes: - DoubledGaugeField U assumed to contain -1/2 factor.
- *        - Pass in q_in_1 shifted in +ve mu direction.
- ******************************************************************************/
-template<class Impl>
-void WilsonKernels<Impl>::ContractConservedCurrentSiteFwd(const SitePropagator &q_in_1,
-							  const SitePropagator &q_in_2,
-							  SitePropagator &q_out,
-							  DoubledGaugeFieldView &U,
-							  unsigned int sU,
-							  unsigned int mu,
-							  bool switch_sign)
-{
-  SitePropagator result, tmp;
-  Gamma g5(Gamma::Algebra::Gamma5);
-
-  Impl::multLink(tmp, U[sU], q_in_1, mu);
-
-  result = g5 * adj(q_in_2) * g5 * WilsonCurrentFwd(tmp, mu);
-
-  if (switch_sign) {
-    q_out -= result;
-  } else {
-    q_out += result;
-  }
-}
-
-/*******************************************************************************
- * Name: ContractConservedCurrentSiteBwd
- * Operation: (1/2) * q2[x + mu] * U^dag(x) * (g[mu] + 1) * q1[x]
- * Notes: - DoubledGaugeField U assumed to contain -1/2 factor.
- *        - Pass in q_in_2 shifted in +ve mu direction.
- ******************************************************************************/
-template<class Impl>
-void WilsonKernels<Impl>::ContractConservedCurrentSiteBwd(const SitePropagator &q_in_1,
-							  const SitePropagator &q_in_2,
-							  SitePropagator &q_out,
-							  DoubledGaugeFieldView &U,
-							  unsigned int sU,
-							  unsigned int mu,
-							  bool switch_sign)
-{
-  SitePropagator result, tmp;
-  Gamma g5(Gamma::Algebra::Gamma5);
-
-  Impl::multLink(tmp, U[sU], q_in_1, mu + Nd);
-
-  result = g5 * adj(q_in_2) * g5 * WilsonCurrentBwd(tmp, mu);
-  if (switch_sign) {
-    q_out += result;
-  } else {
-    q_out -= result;
-  }
-}
-
-/*******************************************************************************
- * Name: SeqConservedCurrentSiteFwd
- * Operation: (1/2) * U(x) * (g[mu] - 1) * q[x + mu]
- * Notes: - DoubledGaugeField U assumed to contain -1/2 factor.
- *        - Pass in q_in shifted in +ve mu direction.
- ******************************************************************************/
-template<class Impl>
-void WilsonKernels<Impl>::SeqConservedCurrentSiteFwd(const SitePropagator &q_in,
-                                                     SitePropagator &q_out,
-                                                     DoubledGaugeFieldView &U,
-                                                     unsigned int sU,
-                                                     unsigned int mu,
-                                                     vPredicate t_mask,
-                                                     bool switch_sign)
-{
-  SitePropagator result;
-  
-  Impl::multLink(result, U[sU], q_in, mu);
-  result = WilsonCurrentFwd(result, mu);
-
-  // Zero any unwanted timeslice entries.
-  result = predicatedWhere(t_mask, result, 0.*result);
-  
-  if (switch_sign) {
-    q_out -= result;
-  } else {
-    q_out += result;
-  }
-}
-
-/*******************************************************************************
- * Name: SeqConservedCurrentSiteFwd
- * Operation: (1/2) * U^dag(x) * (g[mu] + 1) * q[x - mu]
- * Notes: - DoubledGaugeField U assumed to contain -1/2 factor.
- *        - Pass in q_in shifted in -ve mu direction.
- ******************************************************************************/
-template<class Impl>
-void WilsonKernels<Impl>::SeqConservedCurrentSiteBwd(const SitePropagator &q_in, 
-                                                     SitePropagator &q_out,
-                                                     DoubledGaugeFieldView &U,
-                                                     unsigned int sU,
-                                                     unsigned int mu,
-                                                     vPredicate t_mask,
-                                                     bool switch_sign)
-{
-  SitePropagator result;
-  Impl::multLink(result, U[sU], q_in, mu + Nd);
-  result = WilsonCurrentBwd(result, mu);
-
-  // Zero any unwanted timeslice entries.
-  result = predicatedWhere(t_mask, result, 0.*result);
-  
-  if (switch_sign) {
-    q_out += result;
-  } else {
-    q_out -= result;
-  }
-}
-
-
-NAMESPACE_END(Grid);
-

Grid/qcd/action/fermion/implementation/WilsonTMFermionImplementation.h

@@ -1,97 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/WilsonTMFermion.cc
-
-    Copyright (C) 2015
-
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is 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/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/WilsonTMFermion.h>
-
-#pragma once
-
-NAMESPACE_BEGIN(Grid);
-
-/*
- * BF sequence
- *
- void bfmbase<Float>::MooeeInv(Fermion_t psi, 
- Fermion_t chi, 
- int dag, int cb)
-
- double m    = this->mass;
- double tm   = this->twistedmass;
- double mtil = 4.0+this->mass;
-
- double sq = mtil*mtil + tm*tm;
-
- double a = mtil/sq;
- double b = -tm /sq;
- if(dag) b=-b;
- axpibg5x(chi,psi,a,b);
-
- void bfmbase<Float>::Mooee(Fermion_t psi, 
- Fermion_t chi, 
- int dag,int cb)
- double a = 4.0+this->mass;
- double b = this->twistedmass;
- if(dag) b=-b;
- axpibg5x(chi,psi,a,b);
-*/
-
-template<class Impl>
-void WilsonTMFermion<Impl>::Mooee(const FermionField &in, FermionField &out) {
-  RealD a = 4.0+this->mass;
-  RealD b = this->mu;
-  out.Checkerboard() = in.Checkerboard();
-  axpibg5x(out,in,a,b);
-}
-template<class Impl>
-void WilsonTMFermion<Impl>::MooeeDag(const FermionField &in, FermionField &out) {
-  RealD a = 4.0+this->mass;
-  RealD b = -this->mu;
-  out.Checkerboard() = in.Checkerboard();
-  axpibg5x(out,in,a,b);
-}
-template<class Impl>
-void WilsonTMFermion<Impl>::MooeeInv(const FermionField &in, FermionField &out) {
-  RealD m    = this->mass;
-  RealD tm   = this->mu;
-  RealD mtil = 4.0+m;
-  RealD sq   = mtil*mtil+tm*tm;
-  RealD a    = mtil/sq;
-  RealD b    = -tm /sq;
-  axpibg5x(out,in,a,b);
-}
-template<class Impl>
-void WilsonTMFermion<Impl>::MooeeInvDag(const FermionField &in, FermionField &out) {
-  RealD m    = this->mass;
-  RealD tm   = this->mu;
-  RealD mtil = 4.0+m;
-  RealD sq   = mtil*mtil+tm*tm;
-  RealD a    = mtil/sq;
-  RealD b    = tm /sq;
-  axpibg5x(out,in,a,b);
-}
-
-NAMESPACE_END(Grid);