Merge branch 'master' into chulwoo-dec12-2015

.gitignore

@@ -5,11 +5,11 @@
 *.o
 *.obj
 
+
 # Editor files #
 ################
 *~
 *#
-*.sublime-*
 
 # Precompiled Headers #
 #######################
@@ -48,9 +48,6 @@ Config.h.in
 config.log
 config.status
 .deps
-Make.inc
-eigen.inc
-Eigen.inc
 
 # http://www.gnu.org/software/autoconf #
 ########################################
@@ -65,8 +62,19 @@ stamp-h1
 config.sub
 config.guess
 INSTALL
-.dirstamp
-ltmain.sh
+
+# Packages #
+############
+# it's better to unpack these files and commit the raw source
+# git has its own built in compression methods
+*.7z
+*.dmg
+*.gz
+*.iso
+*.jar
+*.rar
+*.tar
+*.zip
  
 # Logs and databases #
 ######################
@@ -83,7 +91,6 @@ ltmain.sh
 .Trashes
 ehthumbs.db
 Thumbs.db
-.dirstamp
 
 # build directory #
 ###################
@@ -93,25 +100,3 @@ build*/*
 #####################
 *.xcodeproj/*
 build.sh
-.vscode
-*.code-workspace
-
-# Eigen source #
-################
-Grid/Eigen
-Eigen/*
-
-# libtool macros #
-##################
-m4/lt*
-m4/libtool.m4
-
-# github pages #
-################
-gh-pages/
-
-# generated sources #
-#####################
-Grid/qcd/spin/gamma-gen/*.h
-Grid/qcd/spin/gamma-gen/*.cc
-Grid/util/Version.h

.travis.yml

@@ -1,61 +1,79 @@
 language: cpp
 
-cache:
-  directories:
-    - clang
-
 matrix:
   include:
     - os:        osx
-      osx_image: xcode8.3
+      osx_image: xcode7.2
       compiler: clang
-      env: PREC=single
     - os:        osx
-      osx_image: xcode8.3
-      compiler: clang
-      env: PREC=double
+      osx_image: xcode7.2
+      compiler: gcc
+      env: VERSION=-5
+    - compiler: gcc
+      addons:
+        apt:
+          sources:
+            - ubuntu-toolchain-r-test
+          packages:
+            - g++-4.9
+            - libmpfr-dev
+            - libgmp-dev
+            - libmpc-dev
+            - binutils-dev
+      env: VERSION=-4.9
+    - compiler: gcc
+      addons:
+        apt:
+          sources:
+            - ubuntu-toolchain-r-test
+          packages:
+            - g++-5
+            - libmpfr-dev
+            - libgmp-dev
+            - libmpc-dev
+            - binutils-dev
+      env: VERSION=-5
+    - compiler: clang
+      addons:
+        apt:
+          sources:
+            - ubuntu-toolchain-r-test
+            - llvm-toolchain-precise-3.7
+          packages:
+            - clang-3.7
+            - libmpfr-dev
+            - libgmp-dev
+            - libmpc-dev
+            - binutils-dev
+      env: VERSION=-3.7
+    - compiler: clang
+      addons:
+        apt:
+          sources:
+            - ubuntu-toolchain-r-test
+            - llvm-toolchain-precise-3.8
+          packages:
+            - clang-3.8
+            - libmpfr-dev
+            - libgmp-dev
+            - libmpc-dev
+            - binutils-dev
+      env: VERSION=-3.8
       
 before_install:
-    - export GRIDDIR=`pwd`
-    - if [[ "$TRAVIS_OS_NAME" == "linux" ]] && [[ "$CC" == "clang" ]] && [ ! -e clang/bin ]; then wget $CLANG_LINK; tar -xf `basename $CLANG_LINK`; mkdir clang; mv clang+*/* clang/; fi
-    - if [[ "$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
+    - if [[ "$TRAVIS_OS_NAME" == "osx" ]] && [[ "$CC" == "gcc" ]]; then brew install gcc5; fi
     
 install:
-    - export CWD=`pwd`
-    - echo $CWD
     - export CC=$CC$VERSION
     - export CXX=$CXX$VERSION
-    - echo $PATH
-    - which autoconf
-    - autoconf  --version
-    - which automake
-    - automake  --version
-    - which $CC
-    - $CC  --version
-    - 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
+    - ./scripts/reconfigure_script
     - 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
+    - ../configure CXXFLAGS="-msse4.2 -O3 -std=c++11" LIBS="-lmpfr -lgmp" --enable-precision=single --enable-simd=SSE4 --enable-comms=none
     - make -j4
-    - make install
-    - cd $CWD/build
-    - ../configure --enable-precision=$PREC --enable-simd=SSE4 --enable-comms=none --with-lime=$CWD/build/lime/install ${EXTRACONF}
-    - make -j4 
-    - ./benchmarks/Benchmark_dwf --threads 1 --debug-signals
-    - make check
+    - ./benchmarks/Benchmark_dwf --threads 1

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

@@ -1,37 +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
-
- //disables and intel compiler specific warning (in json.hpp)
-#pragma warning disable 488  
-
-
-#endif

Grid/GridStd.h

@@ -1,29 +0,0 @@
-#ifndef GRID_STD_H
-#define GRID_STD_H
-
-///////////////////
-// Std C++ dependencies
-///////////////////
-#include <cassert>
-#include <complex>
-#include <vector>
-#include <string>
-#include <iostream>
-#include <iomanip>
-#include <random>
-#include <functional>
-#include <stdio.h>
-#include <stdlib.h>
-#include <stdio.h>
-#include <signal.h>
-#include <ctime>
-#include <sys/time.h>
-#include <chrono>
-#include <zlib.h>
-
-///////////////////
-// Grid config
-///////////////////
-#include "Config.h"
-
-#endif /* GRID_STD_H */

Grid/Grid_Eigen_Dense.h

@@ -1,14 +0,0 @@
-#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
-#include <Grid/Eigen/Dense>
-#if defined __GNUC__
-#pragma GCC diagnostic pop
-#endif

Grid/Makefile.am

@@ -1,63 +0,0 @@
-extra_sources=
-extra_headers=
-
-if BUILD_COMMS_MPI3
-  extra_sources+=communicator/Communicator_mpi3.cc
-  extra_sources+=communicator/Communicator_base.cc
-  extra_sources+=communicator/SharedMemoryMPI.cc
-  extra_sources+=communicator/SharedMemory.cc
-endif
-
-if BUILD_COMMS_NONE
-  extra_sources+=communicator/Communicator_none.cc
-  extra_sources+=communicator/Communicator_base.cc
-  extra_sources+=communicator/SharedMemoryNone.cc
-  extra_sources+=communicator/SharedMemory.cc
-endif
-
-if BUILD_HDF5
-  extra_sources+=serialisation/Hdf5IO.cc 
-  extra_headers+=serialisation/Hdf5IO.h
-  extra_headers+=serialisation/Hdf5Type.h
-endif
-
-all: version-cache
-
-version-cache:
-	@if [ `git status --porcelain | grep -v '??' | wc -l` -gt 0 ]; then\
-		a="uncommited changes";\
-	else\
-		a="clean";\
-	fi;\
-	echo "`git log -n 1 --format=format:"#define GITHASH \\"%H:%d $$a\\"%n" HEAD`" > vertmp;\
-	if [ -e version-cache ]; then\
-		d=`diff vertmp version-cache`;\
-		if [ "$${d}" != "" ]; then\
-			mv vertmp version-cache;\
-			rm -f Version.h;\
-		fi;\
-	else\
-		mv vertmp version-cache;\
-		rm -f Version.h;\
-	fi;\
-	rm -f vertmp
-
-Version.h:
-	cp version-cache Version.h
-
-.PHONY: version-cache
-
-#
-# Libraries
-#
-include Make.inc
-include Eigen.inc
-
-lib_LIBRARIES = libGrid.a
-
-CCFILES += $(extra_sources)
-HFILES  += $(extra_headers) Config.h Version.h
-
-libGrid_a_SOURCES              = $(CCFILES)
-libGrid_adir                   = $(includedir)/Grid
-nobase_dist_pkginclude_HEADERS = $(HFILES) $(eigen_files) $(eigen_unsupp_files)

Grid/algorithms/Algorithms.h

@@ -1,63 +0,0 @@
-    /*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: ./lib/Algorithms.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_ALGORITHMS_H
-#define GRID_ALGORITHMS_H
-
-#include <Grid/algorithms/SparseMatrix.h>
-#include <Grid/algorithms/LinearOperator.h>
-#include <Grid/algorithms/Preconditioner.h>
-
-#include <Grid/algorithms/approx/Zolotarev.h>
-#include <Grid/algorithms/approx/Chebyshev.h>
-#include <Grid/algorithms/approx/Remez.h>
-#include <Grid/algorithms/approx/MultiShiftFunction.h>
-#include <Grid/algorithms/approx/Forecast.h>
-
-#include <Grid/algorithms/iterative/Deflation.h>
-#include <Grid/algorithms/iterative/ConjugateGradient.h>
-#include <Grid/algorithms/iterative/ConjugateResidual.h>
-#include <Grid/algorithms/iterative/NormalEquations.h>
-#include <Grid/algorithms/iterative/SchurRedBlack.h>
-#include <Grid/algorithms/iterative/ConjugateGradientMultiShift.h>
-#include <Grid/algorithms/iterative/ConjugateGradientMixedPrec.h>
-#include <Grid/algorithms/iterative/BlockConjugateGradient.h>
-#include <Grid/algorithms/iterative/ConjugateGradientReliableUpdate.h>
-#include <Grid/algorithms/iterative/MinimalResidual.h>
-#include <Grid/algorithms/iterative/GeneralisedMinimalResidual.h>
-#include <Grid/algorithms/iterative/CommunicationAvoidingGeneralisedMinimalResidual.h>
-#include <Grid/algorithms/iterative/FlexibleGeneralisedMinimalResidual.h>
-#include <Grid/algorithms/iterative/FlexibleCommunicationAvoidingGeneralisedMinimalResidual.h>
-#include <Grid/algorithms/iterative/MixedPrecisionFlexibleGeneralisedMinimalResidual.h>
-#include <Grid/algorithms/iterative/ImplicitlyRestartedLanczos.h>
-#include <Grid/algorithms/iterative/PowerMethod.h>
-
-#include <Grid/algorithms/CoarsenedMatrix.h>
-#include <Grid/algorithms/FFT.h>
-
-#endif

Grid/algorithms/FFT.h

@@ -1,306 +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 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;
-    
-    std::vector<int> dimensions;
-    std::vector<int> processors;
-    std::vector<int> 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;
-      std::vector<int> 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,std::vector<int> 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){
-      std::vector<int> 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];
-      
-      std::vector<int> layout(Nd,1);
-      std::vector<int> 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);
-      
-
-      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._odata[0];
-        FFTW_scalar *out= (FFTW_scalar *)&pgbuf._odata[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
-      std::vector<int> lcoor(Nd), gcoor(Nd);
-      result = source;
-      int pc = processor_coor[dim];
-      for(int p=0;p<processors[dim];p++) {
-        PARALLEL_REGION
-        {
-          std::vector<int> cbuf(Nd);
-          sobj s;
-          
-          PARALLEL_FOR_LOOP_INTERN
-          for(int idx=0;idx<sgrid->lSites();idx++) {
-            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();
-      PARALLEL_REGION
-      {
-        std::vector<int> cbuf(Nd);
-        
-        PARALLEL_FOR_LOOP_INTERN
-        for(int idx=0;idx<NN;idx++) {
-          pencil_g.LocalIndexToLocalCoor(idx, cbuf);
-          if ( cbuf[dim] == 0 ) {  // restricts loop to plane at lcoor[dim]==0
-            FFTW_scalar *in = (FFTW_scalar *)&pgbuf._odata[idx];
-            FFTW_scalar *out= (FFTW_scalar *)&pgbuf._odata[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
-      PARALLEL_REGION
-      {
-        std::vector<int> clbuf(Nd), cgbuf(Nd);
-        sobj s;
-        
-        PARALLEL_FOR_LOOP_INTERN
-        for(int idx=0;idx<sgrid->lSites();idx++) {
-          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
-    }
-  };
-}
-
-#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 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] = std::conj(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(std::abs(G[j][j]) > std::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 = std::conj(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;
-      };
-  };
-
-}
-
-#endif

Grid/algorithms/iterative/BlockConjugateGradient.h

@@ -1,698 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/algorithms/iterative/BlockConjugateGradient.h
-
-Copyright (C) 2017
-
-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_BLOCK_CONJUGATE_GRADIENT_H
-#define GRID_BLOCK_CONJUGATE_GRADIENT_H
-
-
-namespace Grid {
-
-enum BlockCGtype { BlockCG, BlockCGrQ, CGmultiRHS, BlockCGVec, BlockCGrQVec };
-
-//////////////////////////////////////////////////////////////////////////
-// Block conjugate gradient. Dimension zero should be the block direction
-//////////////////////////////////////////////////////////////////////////
-template <class Field>
-class BlockConjugateGradient : public OperatorFunction<Field> {
- public:
-
-  typedef typename Field::scalar_type scomplex;
-
-  int blockDim ;
-  int Nblock;
-
-  BlockCGtype CGtype;
-  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 PrintInterval; //GridLogMessages or Iterative
-  
-  BlockConjugateGradient(BlockCGtype cgtype,int _Orthog,RealD tol, Integer maxit, bool err_on_no_conv = true)
-    : Tolerance(tol), CGtype(cgtype),   blockDim(_Orthog),  MaxIterations(maxit), ErrorOnNoConverge(err_on_no_conv),PrintInterval(100)
-  {};
-
-////////////////////////////////////////////////////////////////////////////////////////////////////
-// Thin QR factorisation (google it)
-////////////////////////////////////////////////////////////////////////////////////////////////////
-  ////////////////////////////////////////////////////////////////////////////////////////////////////
-  //Dimensions
-  // R_{ferm x Nblock} =  Q_{ferm x Nblock} x  C_{Nblock x Nblock} -> ferm x Nblock
-  //
-  // Rdag R = m_rr = Herm = L L^dag        <-- Cholesky decomposition (LLT routine in Eigen)
-  //
-  //   Q  C = R => Q = R C^{-1}
-  //
-  // Want  Ident = Q^dag Q = C^{-dag} R^dag R C^{-1} = C^{-dag} L L^dag C^{-1} = 1_{Nblock x Nblock} 
-  //
-  // Set C = L^{dag}, and then Q^dag Q = ident 
-  //
-  // Checks:
-  // Cdag C = Rdag R ; passes.
-  // QdagQ  = 1      ; passes
-  ////////////////////////////////////////////////////////////////////////////////////////////////////
-void ThinQRfact (Eigen::MatrixXcd &m_rr,
-		 Eigen::MatrixXcd &C,
-		 Eigen::MatrixXcd &Cinv,
-		 Field & Q,
-		 const Field & R)
-{
-  int Orthog = blockDim; // First dimension is block dim; this is an assumption
-  sliceInnerProductMatrix(m_rr,R,R,Orthog);
-
-  // Force manifest hermitian to avoid rounding related
-  m_rr = 0.5*(m_rr+m_rr.adjoint());
-
-  Eigen::MatrixXcd L    = m_rr.llt().matrixL(); 
-
-  C    = L.adjoint();
-  Cinv = C.inverse();
-  ////////////////////////////////////////////////////////////////////////////////////////////////////
-  // Q = R C^{-1}
-  //
-  // Q_j  = R_i Cinv(i,j) 
-  //
-  // NB maddMatrix conventions are Right multiplication X[j] a[j,i] already
-  ////////////////////////////////////////////////////////////////////////////////////////////////////
-  sliceMulMatrix(Q,Cinv,R,Orthog);
-}
-// see comments above
-void ThinQRfact (Eigen::MatrixXcd &m_rr,
-		 Eigen::MatrixXcd &C,
-		 Eigen::MatrixXcd &Cinv,
-		 std::vector<Field> & Q,
-		 const std::vector<Field> & R)
-{
-  InnerProductMatrix(m_rr,R,R);
-
-  m_rr = 0.5*(m_rr+m_rr.adjoint());
-
-  Eigen::MatrixXcd L    = m_rr.llt().matrixL(); 
-
-  C    = L.adjoint();
-  Cinv = C.inverse();
-
-  MulMatrix(Q,Cinv,R);
-}
-
-////////////////////////////////////////////////////////////////////////////////////////////////////
-// Call one of several implementations
-////////////////////////////////////////////////////////////////////////////////////////////////////
-void operator()(LinearOperatorBase<Field> &Linop, const Field &Src, Field &Psi) 
-{
-  if ( CGtype == BlockCGrQ ) {
-    BlockCGrQsolve(Linop,Src,Psi);
-  } else if (CGtype == CGmultiRHS ) {
-    CGmultiRHSsolve(Linop,Src,Psi);
-  } else {
-    assert(0);
-  }
-}
-virtual void operator()(LinearOperatorBase<Field> &Linop, const std::vector<Field> &Src, std::vector<Field> &Psi) 
-{
-  if ( CGtype == BlockCGrQVec ) {
-    BlockCGrQsolveVec(Linop,Src,Psi);
-  } else {
-    assert(0);
-  }
-}
-
-////////////////////////////////////////////////////////////////////////////
-// BlockCGrQ implementation:
-//--------------------------
-// X is guess/Solution
-// B is RHS
-// Solve A X_i = B_i    ;        i refers to Nblock index
-////////////////////////////////////////////////////////////////////////////
-void BlockCGrQsolve(LinearOperatorBase<Field> &Linop, const Field &B, Field &X) 
-{
-  int Orthog = blockDim; // First dimension is block dim; this is an assumption
-  Nblock = B._grid->_fdimensions[Orthog];
-/* FAKE */
-  Nblock=8;
-  std::cout<<GridLogMessage<<" Block Conjugate Gradient : Orthog "<<Orthog<<" Nblock "<<Nblock<<std::endl;
-
-  X.checkerboard = B.checkerboard;
-  conformable(X, B);
-
-  Field tmp(B);
-  Field Q(B);
-  Field D(B);
-  Field Z(B);
-  Field AD(B);
-
-  Eigen::MatrixXcd m_DZ     = Eigen::MatrixXcd::Identity(Nblock,Nblock);
-  Eigen::MatrixXcd m_M      = Eigen::MatrixXcd::Identity(Nblock,Nblock);
-  Eigen::MatrixXcd m_rr     = Eigen::MatrixXcd::Zero(Nblock,Nblock);
-
-  Eigen::MatrixXcd m_C      = Eigen::MatrixXcd::Zero(Nblock,Nblock);
-  Eigen::MatrixXcd m_Cinv   = Eigen::MatrixXcd::Zero(Nblock,Nblock);
-  Eigen::MatrixXcd m_S      = Eigen::MatrixXcd::Zero(Nblock,Nblock);
-  Eigen::MatrixXcd m_Sinv   = Eigen::MatrixXcd::Zero(Nblock,Nblock);
-
-  Eigen::MatrixXcd m_tmp    = Eigen::MatrixXcd::Identity(Nblock,Nblock);
-  Eigen::MatrixXcd m_tmp1   = Eigen::MatrixXcd::Identity(Nblock,Nblock);
-
-  // Initial residual computation & set up
-  std::vector<RealD> residuals(Nblock);
-  std::vector<RealD> ssq(Nblock);
-
-  sliceNorm(ssq,B,Orthog);
-  RealD sssum=0;
-  for(int b=0;b<Nblock;b++) sssum+=ssq[b];
-
-  sliceNorm(residuals,B,Orthog);
-  for(int b=0;b<Nblock;b++){ assert(std::isnan(residuals[b])==0); }
-
-  sliceNorm(residuals,X,Orthog);
-  for(int b=0;b<Nblock;b++){ assert(std::isnan(residuals[b])==0); }
-
-  /************************************************************************
-   * Block conjugate gradient rQ (Sebastien Birk Thesis, after Dubrulle 2001)
-   ************************************************************************
-   * Dimensions:
-   *
-   *   X,B==(Nferm x Nblock)
-   *   A==(Nferm x Nferm)
-   *  
-   * Nferm = Nspin x Ncolour x Ncomplex x Nlattice_site
-   * 
-   * QC = R = B-AX, D = Q     ; QC => Thin QR factorisation (google it)
-   * for k: 
-   *   Z  = AD
-   *   M  = [D^dag Z]^{-1}
-   *   X  = X + D MC
-   *   QS = Q - ZM
-   *   D  = Q + D S^dag
-   *   C  = S C
-   */
-  ///////////////////////////////////////
-  // Initial block: initial search dir is guess
-  ///////////////////////////////////////
-  std::cout << GridLogMessage<<"BlockCGrQ algorithm initialisation " <<std::endl;
-
-  //1.  QC = R = B-AX, D = Q     ; QC => Thin QR factorisation (google it)
-  Linop.HermOp(X, AD);
-  tmp = B - AD;  
-
-  ThinQRfact (m_rr, m_C, m_Cinv, Q, tmp);
-  D=Q;
-
-  std::cout << GridLogMessage<<"BlockCGrQ computed initial residual and QR fact " <<std::endl;
-
-  ///////////////////////////////////////
-  // Timers
-  ///////////////////////////////////////
-  GridStopWatch sliceInnerTimer;
-  GridStopWatch sliceMaddTimer;
-  GridStopWatch QRTimer;
-  GridStopWatch MatrixTimer;
-  GridStopWatch SolverTimer;
-  SolverTimer.Start();
-
-  int k;
-  for (k = 1; k <= MaxIterations; k++) {
-
-    //3. Z  = AD
-    MatrixTimer.Start();
-    Linop.HermOp(D, Z);      
-    MatrixTimer.Stop();
-
-    //4. M  = [D^dag Z]^{-1}
-    sliceInnerTimer.Start();
-    sliceInnerProductMatrix(m_DZ,D,Z,Orthog);
-    sliceInnerTimer.Stop();
-    m_M       = m_DZ.inverse();
-    
-    //5. X  = X + D MC
-    m_tmp     = m_M * m_C;
-    sliceMaddTimer.Start();
-    sliceMaddMatrix(X,m_tmp, D,X,Orthog);     
-    sliceMaddTimer.Stop();
-
-    //6. QS = Q - ZM
-    sliceMaddTimer.Start();
-    sliceMaddMatrix(tmp,m_M,Z,Q,Orthog,-1.0);
-    sliceMaddTimer.Stop();
-    QRTimer.Start();
-    ThinQRfact (m_rr, m_S, m_Sinv, Q, tmp);
-    QRTimer.Stop();
-    
-    //7. D  = Q + D S^dag
-    m_tmp = m_S.adjoint();
-
-    sliceMaddTimer.Start();
-    sliceMaddMatrix(D,m_tmp,D,Q,Orthog);
-    sliceMaddTimer.Stop();
-
-    //8. C  = S C
-    m_C = m_S*m_C;
-    
-    /*********************
-     * convergence monitor
-     *********************
-     */
-    m_rr = m_C.adjoint() * m_C;
-
-    RealD max_resid=0;
-    RealD rrsum=0;
-    RealD rr;
-
-    for(int b=0;b<Nblock;b++) {
-      rrsum+=real(m_rr(b,b));
-      rr = real(m_rr(b,b))/ssq[b];
-      if ( rr > max_resid ) max_resid = rr;
-    }
-
-    std::cout << GridLogIterative << "\titeration "<<k<<" rr_sum "<<rrsum<<" ssq_sum "<< sssum
-	      <<" ave "<<std::sqrt(rrsum/sssum) << " max "<< max_resid <<std::endl;
-
-    if ( max_resid < Tolerance*Tolerance ) { 
-
-      SolverTimer.Stop();
-
-      std::cout << GridLogMessage<<"BlockCGrQ converged in "<<k<<" iterations"<<std::endl;
-
-      for(int b=0;b<Nblock;b++){
-	std::cout << GridLogMessage<< "\t\tblock "<<b<<" computed resid "
-		  << std::sqrt(real(m_rr(b,b))/ssq[b])<<std::endl;
-      }
-      std::cout << GridLogMessage<<"\tMax residual is "<<std::sqrt(max_resid)<<std::endl;
-
-      Linop.HermOp(X, AD);
-      AD = AD-B;
-      std::cout << GridLogMessage <<"\t True residual is " << std::sqrt(norm2(AD)/norm2(B)) <<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 << "\tInnerProd  " << sliceInnerTimer.Elapsed() <<std::endl;
-      std::cout << GridLogMessage << "\tMaddMatrix " << sliceMaddTimer.Elapsed()  <<std::endl;
-      std::cout << GridLogMessage << "\tThinQRfact " << QRTimer.Elapsed()  <<std::endl;
-	    
-      IterationsToComplete = k;
-      return;
-    }
-
-  }
-  std::cout << GridLogMessage << "BlockConjugateGradient(rQ) did NOT converge" << std::endl;
-
-  if (ErrorOnNoConverge) assert(0);
-  IterationsToComplete = k;
-}
-//////////////////////////////////////////////////////////////////////////
-// multiRHS conjugate gradient. Dimension zero should be the block direction
-// Use this for spread out across nodes
-//////////////////////////////////////////////////////////////////////////
-void CGmultiRHSsolve(LinearOperatorBase<Field> &Linop, const Field &Src, Field &Psi) 
-{
-  int Orthog = blockDim; // First dimension is block dim
-  Nblock = Src._grid->_fdimensions[Orthog];
-
-  std::cout<<GridLogMessage<<"MultiRHS Conjugate Gradient : Orthog "<<Orthog<<" Nblock "<<Nblock<<std::endl;
-
-  Psi.checkerboard = Src.checkerboard;
-  conformable(Psi, Src);
-
-  Field P(Src);
-  Field AP(Src);
-  Field R(Src);
-  
-  std::vector<ComplexD> v_pAp(Nblock);
-  std::vector<RealD> v_rr (Nblock);
-  std::vector<RealD> v_rr_inv(Nblock);
-  std::vector<RealD> v_alpha(Nblock);
-  std::vector<RealD> v_beta(Nblock);
-
-  // Initial residual computation & set up
-  std::vector<RealD> residuals(Nblock);
-  std::vector<RealD> ssq(Nblock);
-
-  sliceNorm(ssq,Src,Orthog);
-  RealD sssum=0;
-  for(int b=0;b<Nblock;b++) sssum+=ssq[b];
-
-  sliceNorm(residuals,Src,Orthog);
-  for(int b=0;b<Nblock;b++){ assert(std::isnan(residuals[b])==0); }
-
-  sliceNorm(residuals,Psi,Orthog);
-  for(int b=0;b<Nblock;b++){ assert(std::isnan(residuals[b])==0); }
-
-  // Initial search dir is guess
-  Linop.HermOp(Psi, AP);
-
-  R = Src - AP;  
-  P = R;
-  sliceNorm(v_rr,R,Orthog);
-
-  GridStopWatch sliceInnerTimer;
-  GridStopWatch sliceMaddTimer;
-  GridStopWatch sliceNormTimer;
-  GridStopWatch MatrixTimer;
-  GridStopWatch SolverTimer;
-
-  SolverTimer.Start();
-  int k;
-  for (k = 1; k <= MaxIterations; k++) {
-
-    RealD rrsum=0;
-    for(int b=0;b<Nblock;b++) rrsum+=real(v_rr[b]);
-
-    std::cout << GridLogIterative << "\titeration "<<k<<" rr_sum "<<rrsum<<" ssq_sum "<< sssum
-	      <<" / "<<std::sqrt(rrsum/sssum) <<std::endl;
-
-    MatrixTimer.Start();
-    Linop.HermOp(P, AP);
-    MatrixTimer.Stop();
-
-    // Alpha
-    sliceInnerTimer.Start();
-    sliceInnerProductVector(v_pAp,P,AP,Orthog);
-    sliceInnerTimer.Stop();
-    for(int b=0;b<Nblock;b++){
-      v_alpha[b] = v_rr[b]/real(v_pAp[b]);
-    }
-
-    // Psi, R update
-    sliceMaddTimer.Start();
-    sliceMaddVector(Psi,v_alpha, P,Psi,Orthog);     // add alpha *  P to psi
-    sliceMaddVector(R  ,v_alpha,AP,  R,Orthog,-1.0);// sub alpha * AP to resid
-    sliceMaddTimer.Stop();
-
-    // Beta
-    for(int b=0;b<Nblock;b++){
-      v_rr_inv[b] = 1.0/v_rr[b];
-    }
-    sliceNormTimer.Start();
-    sliceNorm(v_rr,R,Orthog);
-    sliceNormTimer.Stop();
-    for(int b=0;b<Nblock;b++){
-      v_beta[b] = v_rr_inv[b] *v_rr[b];
-    }
-
-    // Search update
-    sliceMaddTimer.Start();
-    sliceMaddVector(P,v_beta,P,R,Orthog);
-    sliceMaddTimer.Stop();
-
-    /*********************
-     * convergence monitor
-     *********************
-     */
-    RealD max_resid=0;
-    for(int b=0;b<Nblock;b++){
-      RealD rr = v_rr[b]/ssq[b];
-      if ( rr > max_resid ) max_resid = rr;
-    }
-    
-    if ( max_resid < Tolerance*Tolerance ) { 
-
-      SolverTimer.Stop();
-
-      std::cout << GridLogMessage<<"MultiRHS solver converged in " <<k<<" iterations"<<std::endl;
-      for(int b=0;b<Nblock;b++){
-	std::cout << GridLogMessage<< "\t\tBlock "<<b<<" computed resid "<< std::sqrt(v_rr[b]/ssq[b])<<std::endl;
-      }
-      std::cout << GridLogMessage<<"\tMax residual is "<<std::sqrt(max_resid)<<std::endl;
-
-      Linop.HermOp(Psi, AP);
-      AP = AP-Src;
-      std::cout <<GridLogMessage << "\tTrue residual is " << std::sqrt(norm2(AP)/norm2(Src)) <<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 << "\tInnerProd  " << sliceInnerTimer.Elapsed() <<std::endl;
-      std::cout << GridLogMessage << "\tNorm       " << sliceNormTimer.Elapsed() <<std::endl;
-      std::cout << GridLogMessage << "\tMaddMatrix " << sliceMaddTimer.Elapsed()  <<std::endl;
-
-
-      IterationsToComplete = k;
-      return;
-    }
-
-  }
-  std::cout << GridLogMessage << "MultiRHSConjugateGradient did NOT converge" << std::endl;
-
-  if (ErrorOnNoConverge) assert(0);
-  IterationsToComplete = k;
-}
-
-void InnerProductMatrix(Eigen::MatrixXcd &m , const std::vector<Field> &X, const std::vector<Field> &Y){
-  for(int b=0;b<Nblock;b++){
-  for(int bp=0;bp<Nblock;bp++) {
-    m(b,bp) = innerProduct(X[b],Y[bp]);  
-  }}
-}
-void MaddMatrix(std::vector<Field> &AP, Eigen::MatrixXcd &m , const std::vector<Field> &X,const std::vector<Field> &Y,RealD scale=1.0){
-  // Should make this cache friendly with site outermost, parallel_for
-  // Deal with case AP aliases with either Y or X
-  std::vector<Field> tmp(Nblock,X[0]);
-  for(int b=0;b<Nblock;b++){
-    tmp[b]   = Y[b];
-    for(int bp=0;bp<Nblock;bp++) {
-      tmp[b] = tmp[b] + (scale*m(bp,b))*X[bp]; 
-    }
-  }
-  for(int b=0;b<Nblock;b++){
-    AP[b] = tmp[b];
-  }
-}
-void MulMatrix(std::vector<Field> &AP, Eigen::MatrixXcd &m , const std::vector<Field> &X){
-  // Should make this cache friendly with site outermost, parallel_for
-  for(int b=0;b<Nblock;b++){
-    AP[b] = zero;
-    for(int bp=0;bp<Nblock;bp++) {
-      AP[b] += (m(bp,b))*X[bp]; 
-    }
-  }
-}
-double normv(const std::vector<Field> &P){
-  double nn = 0.0;
-  for(int b=0;b<Nblock;b++) {
-    nn+=norm2(P[b]);
-  }
-  return nn;
-}
-
-////////////////////////////////////////////////////////////////////////////
-// BlockCGrQvec implementation:
-//--------------------------
-// X is guess/Solution
-// B is RHS
-// Solve A X_i = B_i    ;        i refers to Nblock index
-////////////////////////////////////////////////////////////////////////////
-void BlockCGrQsolveVec(LinearOperatorBase<Field> &Linop, const std::vector<Field> &B, std::vector<Field> &X) 
-{
-  Nblock = B.size();
-  assert(Nblock == X.size());
-
-  std::cout<<GridLogMessage<<" Block Conjugate Gradient Vec rQ : Nblock "<<Nblock<<std::endl;
-
-  for(int b=0;b<Nblock;b++){ 
-    X[b].checkerboard = B[b].checkerboard;
-    conformable(X[b], B[b]);
-    conformable(X[b], X[0]); 
-  }
-
-  Field Fake(B[0]);
-
-  std::vector<Field> tmp(Nblock,Fake);
-  std::vector<Field>   Q(Nblock,Fake);
-  std::vector<Field>   D(Nblock,Fake);
-  std::vector<Field>   Z(Nblock,Fake);
-  std::vector<Field>  AD(Nblock,Fake);
-
-  Eigen::MatrixXcd m_DZ     = Eigen::MatrixXcd::Identity(Nblock,Nblock);
-  Eigen::MatrixXcd m_M      = Eigen::MatrixXcd::Identity(Nblock,Nblock);
-  Eigen::MatrixXcd m_rr     = Eigen::MatrixXcd::Zero(Nblock,Nblock);
-
-  Eigen::MatrixXcd m_C      = Eigen::MatrixXcd::Zero(Nblock,Nblock);
-  Eigen::MatrixXcd m_Cinv   = Eigen::MatrixXcd::Zero(Nblock,Nblock);
-  Eigen::MatrixXcd m_S      = Eigen::MatrixXcd::Zero(Nblock,Nblock);
-  Eigen::MatrixXcd m_Sinv   = Eigen::MatrixXcd::Zero(Nblock,Nblock);
-
-  Eigen::MatrixXcd m_tmp    = Eigen::MatrixXcd::Identity(Nblock,Nblock);
-  Eigen::MatrixXcd m_tmp1   = Eigen::MatrixXcd::Identity(Nblock,Nblock);
-
-  // Initial residual computation & set up
-  std::vector<RealD> residuals(Nblock);
-  std::vector<RealD> ssq(Nblock);
-
-  RealD sssum=0;
-  for(int b=0;b<Nblock;b++){ ssq[b] = norm2(B[b]);}
-  for(int b=0;b<Nblock;b++) sssum+=ssq[b];
-
-  for(int b=0;b<Nblock;b++){ residuals[b] = norm2(B[b]);}
-  for(int b=0;b<Nblock;b++){ assert(std::isnan(residuals[b])==0); }
-
-  for(int b=0;b<Nblock;b++){ residuals[b] = norm2(X[b]);}
-  for(int b=0;b<Nblock;b++){ assert(std::isnan(residuals[b])==0); }
-
-  /************************************************************************
-   * Block conjugate gradient rQ (Sebastien Birk Thesis, after Dubrulle 2001)
-   ************************************************************************
-   * Dimensions:
-   *
-   *   X,B==(Nferm x Nblock)
-   *   A==(Nferm x Nferm)
-   *  
-   * Nferm = Nspin x Ncolour x Ncomplex x Nlattice_site
-   * 
-   * QC = R = B-AX, D = Q     ; QC => Thin QR factorisation (google it)
-   * for k: 
-   *   Z  = AD
-   *   M  = [D^dag Z]^{-1}
-   *   X  = X + D MC
-   *   QS = Q - ZM
-   *   D  = Q + D S^dag
-   *   C  = S C
-   */
-  ///////////////////////////////////////
-  // Initial block: initial search dir is guess
-  ///////////////////////////////////////
-  std::cout << GridLogMessage<<"BlockCGrQvec algorithm initialisation " <<std::endl;
-
-  //1.  QC = R = B-AX, D = Q     ; QC => Thin QR factorisation (google it)
-  for(int b=0;b<Nblock;b++) {
-    Linop.HermOp(X[b], AD[b]);
-    tmp[b] = B[b] - AD[b];  
-  }
-
-  ThinQRfact (m_rr, m_C, m_Cinv, Q, tmp);
-
-  for(int b=0;b<Nblock;b++) D[b]=Q[b];
-
-  std::cout << GridLogMessage<<"BlockCGrQ vec computed initial residual and QR fact " <<std::endl;
-
-  ///////////////////////////////////////
-  // Timers
-  ///////////////////////////////////////
-  GridStopWatch sliceInnerTimer;
-  GridStopWatch sliceMaddTimer;
-  GridStopWatch QRTimer;
-  GridStopWatch MatrixTimer;
-  GridStopWatch SolverTimer;
-  SolverTimer.Start();
-
-  int k;
-  for (k = 1; k <= MaxIterations; k++) {
-
-    //3. Z  = AD
-    MatrixTimer.Start();
-    for(int b=0;b<Nblock;b++) Linop.HermOp(D[b], Z[b]);      
-    MatrixTimer.Stop();
-
-    //4. M  = [D^dag Z]^{-1}
-    sliceInnerTimer.Start();
-    InnerProductMatrix(m_DZ,D,Z);
-    sliceInnerTimer.Stop();
-    m_M       = m_DZ.inverse();
-    
-    //5. X  = X + D MC
-    m_tmp     = m_M * m_C;
-    sliceMaddTimer.Start();
-    MaddMatrix(X,m_tmp, D,X);     
-    sliceMaddTimer.Stop();
-
-    //6. QS = Q - ZM
-    sliceMaddTimer.Start();
-    MaddMatrix(tmp,m_M,Z,Q,-1.0);
-    sliceMaddTimer.Stop();
-    QRTimer.Start();
-    ThinQRfact (m_rr, m_S, m_Sinv, Q, tmp);
-    QRTimer.Stop();
-    
-    //7. D  = Q + D S^dag
-    m_tmp = m_S.adjoint();
-    sliceMaddTimer.Start();
-    MaddMatrix(D,m_tmp,D,Q);
-    sliceMaddTimer.Stop();
-
-    //8. C  = S C
-    m_C = m_S*m_C;
-    
-    /*********************
-     * convergence monitor
-     *********************
-     */
-    m_rr = m_C.adjoint() * m_C;
-
-    RealD max_resid=0;
-    RealD rrsum=0;
-    RealD rr;
-
-    for(int b=0;b<Nblock;b++) {
-      rrsum+=real(m_rr(b,b));
-      rr = real(m_rr(b,b))/ssq[b];
-      if ( rr > max_resid ) max_resid = rr;
-    }
-
-    std::cout << GridLogIterative << "\t Block Iteration "<<k<<" ave resid "<< sqrt(rrsum/sssum) << " max "<< sqrt(max_resid) <<std::endl;
-
-    if ( max_resid < Tolerance*Tolerance ) { 
-
-      SolverTimer.Stop();
-
-      std::cout << GridLogMessage<<"BlockCGrQ converged in "<<k<<" iterations"<<std::endl;
-
-      for(int b=0;b<Nblock;b++){
-	std::cout << GridLogMessage<< "\t\tblock "<<b<<" computed resid "<< std::sqrt(real(m_rr(b,b))/ssq[b])<<std::endl;
-      }
-      std::cout << GridLogMessage<<"\tMax residual is "<<std::sqrt(max_resid)<<std::endl;
-
-      for(int b=0;b<Nblock;b++) Linop.HermOp(X[b], AD[b]);
-      for(int b=0;b<Nblock;b++) AD[b] = AD[b]-B[b];
-      std::cout << GridLogMessage <<"\t True residual is " << std::sqrt(normv(AD)/normv(B)) <<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 << "\tInnerProd  " << sliceInnerTimer.Elapsed() <<std::endl;
-      std::cout << GridLogMessage << "\tMaddMatrix " << sliceMaddTimer.Elapsed()  <<std::endl;
-      std::cout << GridLogMessage << "\tThinQRfact " << QRTimer.Elapsed()  <<std::endl;
-	    
-      IterationsToComplete = k;
-      return;
-    }
-
-  }
-  std::cout << GridLogMessage << "BlockConjugateGradient(rQ) did NOT converge" << std::endl;
-
-  if (ErrorOnNoConverge) assert(0);
-  IterationsToComplete = k;
-}
-
-
-
-};
-
-}
-#endif

Grid/algorithms/iterative/CommunicationAvoidingGeneralisedMinimalResidual.h

@@ -1,244 +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:
-  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<std::complex<double>> y;
-  std::vector<std::complex<double>> gamma;
-  std::vector<std::complex<double>> c;
-  std::vector<std::complex<double>> 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));
-
-    v[0] = (1. / gamma[0]) * r;
-    LinalgTimer.Stop();
-
-    for (int i=0; i<RestartLength; i++) {
-
-      IterationCount++;
-
-      arnoldiStep(LinOp, v, w, i);
-
-      qrUpdate(i);
-
-      cp = std::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 - H(iter, i) * v[i];
-    }
-
-    H(iter, iter + 1) = sqrt(norm2(w));
-    v[iter + 1] = (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] * H(iter, i) + c[i] * H(iter, i + 1);
-      H(iter, i)     = std::conj(c[i]) * H(iter, i) + std::conj(s[i]) * H(iter, i + 1);
-      H(iter, i + 1) = tmp;
-    }
-
-    // Compute new Givens Rotation
-    ComplexD 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]     = std::conj(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] - H(k, i) * y[k];
-      y[i] = y[i] / H(i, i);
-    }
-
-    for (int i = 0; i <= iter; i++)
-      psi = psi + v[i] * y[i];
-    CompSolutionTimer.Stop();
-  }
-};
-}
-#endif

Grid/algorithms/iterative/ConjugateGradient.h

@@ -1,178 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/algorithms/iterative/ConjugateGradient.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 */
-#ifndef GRID_CONJUGATE_GRADIENT_H
-#define GRID_CONJUGATE_GRADIENT_H
-
-namespace Grid {
-
-/////////////////////////////////////////////////////////////
-// Base classes for iterative processes based on operators
-// single input vec, single output vec.
-/////////////////////////////////////////////////////////////
-
-template <class Field>
-class ConjugateGradient : public OperatorFunction<Field> {
- 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
-  
-  ConjugateGradient(RealD tol, Integer maxit, bool err_on_no_conv = true)
-      : Tolerance(tol),
-        MaxIterations(maxit),
-        ErrorOnNoConverge(err_on_no_conv){};
-
-  void operator()(LinearOperatorBase<Field> &Linop, const Field &src, Field &psi) {
-
-
-    psi.checkerboard = src.checkerboard;
-    conformable(psi, src);
-
-    RealD cp, c, a, d, b, ssq, qq, b_pred;
-
-    Field p(src);
-    Field mmp(src);
-    Field r(src);
-
-    // Initial residual computation & set up
-    RealD guess = norm2(psi);
-    assert(std::isnan(guess) == 0);
-
-    
-    Linop.HermOpAndNorm(psi, mmp, d, b);
-
-    r = src - mmp;
-    p = r;
-
-    a = norm2(p);
-    cp = a;
-    ssq = norm2(src);
-
-    std::cout << GridLogIterative << std::setprecision(8) << "ConjugateGradient: guess " << guess << std::endl;
-    std::cout << GridLogIterative << std::setprecision(8) << "ConjugateGradient:   src " << ssq << std::endl;
-    std::cout << GridLogIterative << std::setprecision(8) << "ConjugateGradient:    mp " << d << std::endl;
-    std::cout << GridLogIterative << std::setprecision(8) << "ConjugateGradient:   mmp " << b << std::endl;
-    std::cout << GridLogIterative << std::setprecision(8) << "ConjugateGradient:  cp,r " << cp << std::endl;
-    std::cout << GridLogIterative << std::setprecision(8) << "ConjugateGradient:     p " << a << std::endl;
-
-    RealD rsq = Tolerance * Tolerance * ssq;
-
-    // Check if guess is really REALLY good :)
-    if (cp <= rsq) {
-      std::cout << GridLogMessage << "ConjugateGradient guess is converged already " << std::endl;
-      IterationsToComplete = 0;	
-      return;
-    }
-
-    std::cout << GridLogIterative << std::setprecision(8)
-              << "ConjugateGradient: k=0 residual " << cp << " target " << rsq << std::endl;
-
-    GridStopWatch LinalgTimer;
-    GridStopWatch InnerTimer;
-    GridStopWatch AxpyNormTimer;
-    GridStopWatch LinearCombTimer;
-    GridStopWatch MatrixTimer;
-    GridStopWatch SolverTimer;
-
-    SolverTimer.Start();
-    int k;
-    for (k = 1; k <= MaxIterations; k++) {
-      c = cp;
-
-      MatrixTimer.Start();
-      Linop.HermOp(p, mmp);
-      MatrixTimer.Stop();
-
-      LinalgTimer.Start();
-
-      InnerTimer.Start();
-      ComplexD dc  = innerProduct(p,mmp);
-      InnerTimer.Stop();
-      d = dc.real();
-      a = c / d;
-
-      AxpyNormTimer.Start();
-      cp = axpy_norm(r, -a, mmp, r);
-      AxpyNormTimer.Stop();
-      b = cp / c;
-
-      LinearCombTimer.Start();
-      parallel_for(int ss=0;ss<src._grid->oSites();ss++){
-	vstream(psi[ss], a      *  p[ss] + psi[ss]);
-	vstream(p  [ss], b      *  p[ss] + r[ss]);
-      }
-      LinearCombTimer.Stop();
-      LinalgTimer.Stop();
-
-      std::cout << GridLogIterative << "ConjugateGradient: Iteration " << k
-                << " residual^2 " << sqrt(cp/ssq) << " target " << Tolerance << std::endl;
-
-      // Stopping condition
-      if (cp <= rsq) {
-        SolverTimer.Stop();
-        Linop.HermOpAndNorm(psi, mmp, d, qq);
-        p = mmp - src;
-
-        RealD srcnorm = sqrt(norm2(src));
-        RealD resnorm = sqrt(norm2(p));
-        RealD true_residual = resnorm / srcnorm;
-
-        std::cout << GridLogMessage << "ConjugateGradient Converged on iteration " << k << std::endl;
-        std::cout << GridLogMessage << "\tComputed residual " << 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;
-	std::cout << GridLogMessage << "\tInner      " << InnerTimer.Elapsed() <<std::endl;
-	std::cout << GridLogMessage << "\tAxpyNorm   " << AxpyNormTimer.Elapsed() <<std::endl;
-	std::cout << GridLogMessage << "\tLinearComb " << LinearCombTimer.Elapsed() <<std::endl;
-
-        if (ErrorOnNoConverge) assert(true_residual / Tolerance < 10000.0);
-
-	IterationsToComplete = k;	
-
-        return;
-      }
-    }
-    std::cout << GridLogMessage << "ConjugateGradient did NOT converge "<<k<<" / "<< MaxIterations<< std::endl;
-
-    if (ErrorOnNoConverge) assert(0);
-    IterationsToComplete = k;
-
-  }
-};
-}
-#endif

Grid/algorithms/iterative/ConjugateGradientMixedPrec.h

@@ -1,164 +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 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();
-      
-	zeroit(sol_f);
-
-	//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;
-    }
-  };
-
-
-
-}
-
-#endif

Grid/algorithms/iterative/ConjugateGradientReliableUpdate.h

@@ -1,256 +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 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 " << 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 = sqrt(norm2(src));
-	  RealD resnorm = 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 " << 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;      
-    }    
-  };
-
-
-};
-
-
-
-#endif

Grid/algorithms/iterative/Deflation.h

@@ -1,108 +0,0 @@
-    /*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/algorithms/iterative/ImplicitlyRestartedLanczos.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_DEFLATION_H
-#define GRID_DEFLATION_H
-
-namespace Grid { 
-
-template<class Field>
-class ZeroGuesser: public LinearFunction<Field> {
-public:
-  virtual void operator()(const Field &src, Field &guess) { guess = zero; };
-};
-template<class Field>
-class DoNothingGuesser: public LinearFunction<Field> {
-public:
-  virtual void operator()(const Field &src, Field &guess) {  };
-};
-template<class Field>
-class SourceGuesser: public LinearFunction<Field> {
-public:
-  virtual void operator()(const Field &src, Field &guess) { guess = src; };
-};
-
-////////////////////////////////
-// Fine grid deflation
-////////////////////////////////
-template<class Field>
-class DeflatedGuesser: public LinearFunction<Field> {
-private:
-  const std::vector<Field> &evec;
-  const std::vector<RealD> &eval;
-
-public:
-
-  DeflatedGuesser(const std::vector<Field> & _evec,const std::vector<RealD> & _eval) : evec(_evec), eval(_eval) {};
-
-  virtual void operator()(const Field &src,Field &guess) {
-    guess = zero;
-    assert(evec.size()==eval.size());
-    auto N = evec.size();
-    for (int i=0;i<N;i++) {
-      const Field& tmp = evec[i];
-      axpy(guess,TensorRemove(innerProduct(tmp,src)) / eval[i],tmp,guess);
-    }
-    guess.checkerboard = src.checkerboard;
-  }
-};
-
-template<class FineField, class CoarseField>
-class LocalCoherenceDeflatedGuesser: public LinearFunction<FineField> {
-private:
-  const std::vector<FineField>   &subspace;
-  const std::vector<CoarseField> &evec_coarse;
-  const std::vector<RealD>       &eval_coarse;
-public:
-  
-  LocalCoherenceDeflatedGuesser(const std::vector<FineField>   &_subspace,
-				const std::vector<CoarseField> &_evec_coarse,
-				const std::vector<RealD>       &_eval_coarse)
-    : subspace(_subspace), 
-      evec_coarse(_evec_coarse), 
-      eval_coarse(_eval_coarse)  
-  {
-  }
-  
-  void operator()(const FineField &src,FineField &guess) { 
-    int N = (int)evec_coarse.size();
-    CoarseField src_coarse(evec_coarse[0]._grid);
-    CoarseField guess_coarse(evec_coarse[0]._grid);    guess_coarse = zero;
-    blockProject(src_coarse,src,subspace);    
-    for (int i=0;i<N;i++) {
-      const CoarseField & tmp = evec_coarse[i];
-      axpy(guess_coarse,TensorRemove(innerProduct(tmp,src_coarse)) / eval_coarse[i],tmp,guess_coarse);
-    }
-    blockPromote(guess_coarse,guess,subspace);
-    guess.checkerboard = src.checkerboard;
-  };
-};
-
-
-
-}
-#endif

Grid/algorithms/iterative/FlexibleCommunicationAvoidingGeneralisedMinimalResidual.h

@@ -1,256 +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:
-  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<std::complex<double>> y;
-  std::vector<std::complex<double>> gamma;
-  std::vector<std::complex<double>> c;
-  std::vector<std::complex<double>> 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 = std::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 - H(iter, i) * v[i];
-    }
-
-    H(iter, iter + 1) = sqrt(norm2(w));
-    v[iter + 1] = (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] * H(iter, i) + c[i] * H(iter, i + 1);
-      H(iter, i)     = std::conj(c[i]) * H(iter, i) + std::conj(s[i]) * H(iter, i + 1);
-      H(iter, i + 1) = tmp;
-    }
-
-    // Compute new Givens Rotation
-    ComplexD 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]     = std::conj(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] - H(k, i) * y[k];
-      y[i] = y[i] / H(i, i);
-    }
-
-    for (int i = 0; i <= iter; i++)
-      psi = psi + z[i] * y[i];
-    CompSolutionTimer.Stop();
-  }
-};
-}
-#endif

Grid/algorithms/iterative/FlexibleGeneralisedMinimalResidual.h

@@ -1,254 +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:
-  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<std::complex<double>> y;
-  std::vector<std::complex<double>> gamma;
-  std::vector<std::complex<double>> c;
-  std::vector<std::complex<double>> 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 = std::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 - H(iter, i) * v[i];
-    }
-
-    H(iter, iter + 1) = sqrt(norm2(w));
-    v[iter + 1] = (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] * H(iter, i) + c[i] * H(iter, i + 1);
-      H(iter, i)     = std::conj(c[i]) * H(iter, i) + std::conj(s[i]) * H(iter, i + 1);
-      H(iter, i + 1) = tmp;
-    }
-
-    // Compute new Givens Rotation
-    ComplexD 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]     = std::conj(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] - H(k, i) * y[k];
-      y[i] = y[i] / H(i, i);
-    }
-
-    for (int i = 0; i <= iter; i++)
-      psi = psi + z[i] * y[i];
-    CompSolutionTimer.Stop();
-  }
-};
-}
-#endif

Grid/algorithms/iterative/GeneralisedMinimalResidual.h

@@ -1,242 +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:
-  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<std::complex<double>> y;
-  std::vector<std::complex<double>> gamma;
-  std::vector<std::complex<double>> c;
-  std::vector<std::complex<double>> 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 = std::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 - H(iter, i) * v[i];
-    }
-
-    H(iter, iter + 1) = sqrt(norm2(w));
-    v[iter + 1] = (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] * H(iter, i) + c[i] * H(iter, i + 1);
-      H(iter, i)     = std::conj(c[i]) * H(iter, i) + std::conj(s[i]) * H(iter, i + 1);
-      H(iter, i + 1) = tmp;
-    }
-
-    // Compute new Givens Rotation
-    ComplexD 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]     = std::conj(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] - H(k, i) * y[k];
-      y[i] = y[i] / H(i, i);
-    }
-
-    for (int i = 0; i <= iter; i++)
-      psi = psi + v[i] * y[i];
-    CompSolutionTimer.Stop();
-  }
-};
-}
-#endif

Grid/algorithms/iterative/ImplicitlyRestartedLanczos.h

@@ -1,842 +0,0 @@
-    /*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/algorithms/iterative/ImplicitlyRestartedLanczos.h
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Chulwoo Jung <chulwoo@bnl.gov>
-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_BIRL_H
-#define GRID_BIRL_H
-
-#include <string.h> //memset
-//#include <zlib.h>
-#include <sys/stat.h>
-
-namespace Grid { 
-
-  ////////////////////////////////////////////////////////
-  // Move following 100 LOC to lattice/Lattice_basis.h
-  ////////////////////////////////////////////////////////
-template<class Field>
-void basisOrthogonalize(std::vector<Field> &basis,Field &w,int k) 
-{
-  for(int j=0; j<k; ++j){
-    auto ip = innerProduct(basis[j],w);
-    w = w - ip*basis[j];
-  }
-}
-
-template<class Field>
-void basisRotate(std::vector<Field> &basis,Eigen::MatrixXd& Qt,int j0, int j1, int k0,int k1,int Nm) 
-{
-  typedef typename Field::vector_object vobj;
-  GridBase* grid = basis[0]._grid;
-      
-  parallel_region
-  {
-
-    std::vector < vobj , commAllocator<vobj> > B(Nm); // Thread private
-       
-    parallel_for_internal(int ss=0;ss < grid->oSites();ss++){
-      for(int j=j0; j<j1; ++j) B[j]=0.;
-      
-      for(int j=j0; j<j1; ++j){
-	for(int k=k0; k<k1; ++k){
-	  B[j] +=Qt(j,k) * basis[k]._odata[ss];
-	}
-      }
-      for(int j=j0; j<j1; ++j){
-	  basis[j]._odata[ss] = B[j];
-      }
-    }
-  }
-}
-
-// Extract a single rotated vector
-template<class Field>
-void basisRotateJ(Field &result,std::vector<Field> &basis,Eigen::MatrixXd& Qt,int j, int k0,int k1,int Nm) 
-{
-  typedef typename Field::vector_object vobj;
-  GridBase* grid = basis[0]._grid;
-
-  result.checkerboard = basis[0].checkerboard;
-  parallel_for(int ss=0;ss < grid->oSites();ss++){
-    vobj B = zero;
-    for(int k=k0; k<k1; ++k){
-      B +=Qt(j,k) * basis[k]._odata[ss];
-    }
-    result._odata[ss] = B;
-  }
-}
-
-template<class Field>
-void basisReorderInPlace(std::vector<Field> &_v,std::vector<RealD>& sort_vals, std::vector<int>& idx) 
-{
-  int vlen = idx.size();
-
-  assert(vlen>=1);
-  assert(vlen<=sort_vals.size());
-  assert(vlen<=_v.size());
-
-  for (size_t i=0;i<vlen;i++) {
-
-    if (idx[i] != i) {
-
-      //////////////////////////////////////
-      // idx[i] is a table of desired sources giving a permutation.
-      // Swap v[i] with v[idx[i]].
-      // Find  j>i for which _vnew[j] = _vold[i],
-      // track the move idx[j] => idx[i]
-      // track the move idx[i] => i
-      //////////////////////////////////////
-      size_t j;
-      for (j=i;j<idx.size();j++)
-	if (idx[j]==i)
-	  break;
-
-      assert(idx[i] > i);     assert(j!=idx.size());      assert(idx[j]==i);
-
-      std::swap(_v[i]._odata,_v[idx[i]]._odata); // should use vector move constructor, no data copy
-      std::swap(sort_vals[i],sort_vals[idx[i]]);
-
-      idx[j] = idx[i];
-      idx[i] = i;
-    }
-  }
-}
-
-inline std::vector<int> basisSortGetIndex(std::vector<RealD>& sort_vals) 
-{
-  std::vector<int> idx(sort_vals.size());
-  std::iota(idx.begin(), idx.end(), 0);
-
-  // sort indexes based on comparing values in v
-  std::sort(idx.begin(), idx.end(), [&sort_vals](int i1, int i2) {
-    return ::fabs(sort_vals[i1]) < ::fabs(sort_vals[i2]);
-  });
-  return idx;
-}
-
-template<class Field>
-void basisSortInPlace(std::vector<Field> & _v,std::vector<RealD>& sort_vals, bool reverse) 
-{
-  std::vector<int> idx = basisSortGetIndex(sort_vals);
-  if (reverse)
-    std::reverse(idx.begin(), idx.end());
-  
-  basisReorderInPlace(_v,sort_vals,idx);
-}
-
-/////////////////////////////////////////////////////////////
-// Implicitly restarted lanczos
-/////////////////////////////////////////////////////////////
-template<class Field> class ImplicitlyRestartedLanczosTester 
-{
- public:
-  virtual int TestConvergence(int j,RealD resid,Field &evec, RealD &eval,RealD evalMaxApprox)=0;
-  virtual int ReconstructEval(int j,RealD resid,Field &evec, RealD &eval,RealD evalMaxApprox)=0;
-};
-
-enum IRLdiagonalisation { 
-  IRLdiagonaliseWithDSTEGR,
-  IRLdiagonaliseWithQR,
-  IRLdiagonaliseWithEigen
-};
-
-template<class Field> class ImplicitlyRestartedLanczosHermOpTester  : public ImplicitlyRestartedLanczosTester<Field>
-{
- public:
-
-  LinearFunction<Field>       &_HermOp;
-  ImplicitlyRestartedLanczosHermOpTester(LinearFunction<Field> &HermOp) : _HermOp(HermOp)  {  };
-  int ReconstructEval(int j,RealD resid,Field &B, RealD &eval,RealD evalMaxApprox)
-  {
-    return TestConvergence(j,resid,B,eval,evalMaxApprox);
-  }
-  int TestConvergence(int j,RealD eresid,Field &B, RealD &eval,RealD evalMaxApprox)
-  {
-    Field v(B);
-    RealD eval_poly = eval;
-    // Apply operator
-    _HermOp(B,v);
-
-    RealD vnum = real(innerProduct(B,v)); // HermOp.
-    RealD vden = norm2(B);
-    RealD vv0  = norm2(v);
-    eval   = vnum/vden;
-    v -= eval*B;
-
-    RealD vv = norm2(v) / ::pow(evalMaxApprox,2.0);
-
-    std::cout.precision(13);
-    std::cout<<GridLogIRL  << "[" << std::setw(3)<<j<<"] "
-	     <<"eval = "<<std::setw(25)<< eval << " (" << eval_poly << ")"
-	     <<" |H B[i] - eval[i]B[i]|^2 / evalMaxApprox^2 " << std::setw(25) << vv
-	     <<std::endl;
-
-    int conv=0;
-    if( (vv<eresid*eresid) ) conv = 1;
-
-    return conv;
-  }
-};
-
-template<class Field> 
-class ImplicitlyRestartedLanczos {
- private:
-  const RealD small = 1.0e-8;
-  int MaxIter;
-  int MinRestart; // Minimum number of restarts; only check for convergence after
-  int Nstop;   // Number of evecs checked for convergence
-  int Nk;      // Number of converged sought
-  //  int Np;      // Np -- Number of spare vecs in krylov space //  == Nm - Nk
-  int Nm;      // Nm -- total number of vectors
-  IRLdiagonalisation diagonalisation;
-  int orth_period;
-    
-  RealD OrthoTime;
-  RealD eresid, betastp;
-  ////////////////////////////////
-  // Embedded objects
-  ////////////////////////////////
-  LinearFunction<Field>       &_PolyOp;
-  LinearFunction<Field>       &_HermOp;
-  ImplicitlyRestartedLanczosTester<Field> &_Tester;
-  // Default tester provided (we need a ref to something in default case)
-  ImplicitlyRestartedLanczosHermOpTester<Field> SimpleTester;
-  /////////////////////////
-  // Constructor
-  /////////////////////////
-  
-public:       
-
-  //////////////////////////////////////////////////////////////////
-  // PAB:
-  //////////////////////////////////////////////////////////////////
-  // Too many options  & knobs. 
-  // Eliminate:
-  //   orth_period
-  //   betastp
-  //   MinRestart
-  //
-  // Do we really need orth_period
-  // What is the theoretical basis & guarantees of betastp ?
-  // Nstop=Nk viable?
-  // MinRestart avoidable with new convergence test?
-  // Could cut to PolyOp, HermOp, Tester, Nk, Nm, resid, maxiter (+diagonalisation)
-  // HermOp could be eliminated if we dropped the Power method for max eval.
-  // -- also: The eval, eval2, eval2_copy stuff is still unnecessarily unclear
-  //////////////////////////////////////////////////////////////////
- ImplicitlyRestartedLanczos(LinearFunction<Field> & PolyOp,
-			    LinearFunction<Field> & HermOp,
-			    ImplicitlyRestartedLanczosTester<Field> & Tester,
-			    int _Nstop, // sought vecs
-			    int _Nk, // sought vecs
-			    int _Nm, // spare vecs
-			    RealD _eresid, // resid in lmdue deficit 
-			    int _MaxIter, // Max iterations
-			    RealD _betastp=0.0, // if beta(k) < betastp: converged
-			    int _MinRestart=1, int _orth_period = 1,
-			    IRLdiagonalisation _diagonalisation= IRLdiagonaliseWithEigen) :
-    SimpleTester(HermOp), _PolyOp(PolyOp),      _HermOp(HermOp), _Tester(Tester),
-    Nstop(_Nstop)  ,      Nk(_Nk),      Nm(_Nm),
-    eresid(_eresid),      betastp(_betastp),
-    MaxIter(_MaxIter)  ,      MinRestart(_MinRestart),
-    orth_period(_orth_period), diagonalisation(_diagonalisation)  { };
-
-    ImplicitlyRestartedLanczos(LinearFunction<Field> & PolyOp,
-			       LinearFunction<Field> & HermOp,
-			       int _Nstop, // sought vecs
-			       int _Nk, // sought vecs
-			       int _Nm, // spare vecs
-			       RealD _eresid, // resid in lmdue deficit 
-			       int _MaxIter, // Max iterations
-			       RealD _betastp=0.0, // if beta(k) < betastp: converged
-			       int _MinRestart=1, int _orth_period = 1,
-			       IRLdiagonalisation _diagonalisation= IRLdiagonaliseWithEigen) :
-    SimpleTester(HermOp),  _PolyOp(PolyOp),      _HermOp(HermOp), _Tester(SimpleTester),
-    Nstop(_Nstop)  ,      Nk(_Nk),      Nm(_Nm),
-    eresid(_eresid),      betastp(_betastp),
-    MaxIter(_MaxIter)  ,      MinRestart(_MinRestart),
-    orth_period(_orth_period), diagonalisation(_diagonalisation)  { };
-
-  ////////////////////////////////
-  // Helpers
-  ////////////////////////////////
-  template<typename T>  static RealD normalise(T& v) 
-  {
-    RealD nn = norm2(v);
-    nn = sqrt(nn);
-    v = v * (1.0/nn);
-    return nn;
-  }
-
-  void orthogonalize(Field& w, std::vector<Field>& evec,int k)
-  {
-    OrthoTime-=usecond()/1e6;
-    basisOrthogonalize(evec,w,k);
-    normalise(w);
-    OrthoTime+=usecond()/1e6;
-  }
-
-/* Rudy Arthur's thesis pp.137
-------------------------
-Require: M > K P = M − K †
-Compute the factorization AVM = VM HM + fM eM 
-repeat
-  Q=I
-  for i = 1,...,P do
-    QiRi =HM −θiI Q = QQi
-    H M = Q †i H M Q i
-  end for
-  βK =HM(K+1,K) σK =Q(M,K)
-  r=vK+1βK +rσK
-  VK =VM(1:M)Q(1:M,1:K)
-  HK =HM(1:K,1:K)
-  →AVK =VKHK +fKe†K † Extend to an M = K + P step factorization AVM = VMHM + fMeM
-until convergence
-*/
-  void calc(std::vector<RealD>& eval, std::vector<Field>& evec,  const Field& src, int& Nconv, bool reverse=false)
-  {
-    GridBase *grid = src._grid;
-    assert(grid == evec[0]._grid);
-    
-    GridLogIRL.TimingMode(1);
-    std::cout << GridLogIRL <<"**************************************************************************"<< std::endl;
-    std::cout << GridLogIRL <<" ImplicitlyRestartedLanczos::calc() starting iteration 0 /  "<< MaxIter<< std::endl;
-    std::cout << GridLogIRL <<"**************************************************************************"<< std::endl;
-    std::cout << GridLogIRL <<" -- seek   Nk    = " << Nk    <<" vectors"<< std::endl;
-    std::cout << GridLogIRL <<" -- accept Nstop = " << Nstop <<" vectors"<< std::endl;
-    std::cout << GridLogIRL <<" -- total  Nm    = " << Nm    <<" vectors"<< std::endl;
-    std::cout << GridLogIRL <<" -- size of eval = " << eval.size() << std::endl;
-    std::cout << GridLogIRL <<" -- size of evec = " << evec.size() << std::endl;
-    if ( diagonalisation == IRLdiagonaliseWithDSTEGR ) {
-      std::cout << GridLogIRL << "Diagonalisation is DSTEGR "<<std::endl;
-    } else if ( diagonalisation == IRLdiagonaliseWithQR ) { 
-      std::cout << GridLogIRL << "Diagonalisation is QR "<<std::endl;
-    }  else if ( diagonalisation == IRLdiagonaliseWithEigen ) { 
-      std::cout << GridLogIRL << "Diagonalisation is Eigen "<<std::endl;
-    }
-    std::cout << GridLogIRL <<"**************************************************************************"<< std::endl;
-	
-    assert(Nm <= evec.size() && Nm <= eval.size());
-    
-    // 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_IRL_MEVAPP_ = 50;
-      for (int i=0;i<_MAX_ITER_IRL_MEVAPP_;i++) {
-	normalise(src_n);
-	_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) < 0.05)
-	  i=_MAX_ITER_IRL_MEVAPP_;
-	evalMaxApprox = na;
-	std::cout << GridLogIRL << " Approximation of largest eigenvalue: " << evalMaxApprox << std::endl;
-	src_n = tmp;
-      }
-    }
-	
-    std::vector<RealD> lme(Nm);  
-    std::vector<RealD> lme2(Nm);
-    std::vector<RealD> eval2(Nm);
-    std::vector<RealD> eval2_copy(Nm);
-    Eigen::MatrixXd Qt = Eigen::MatrixXd::Zero(Nm,Nm);
-
-    Field f(grid);
-    Field v(grid);
-    int k1 = 1;
-    int k2 = Nk;
-    RealD beta_k;
-
-    Nconv = 0;
-  
-    // Set initial vector
-    evec[0] = src;
-    normalise(evec[0]);
-	
-    // Initial Nk steps
-    OrthoTime=0.;
-    for(int k=0; k<Nk; ++k) step(eval,lme,evec,f,Nm,k);
-    std::cout<<GridLogIRL <<"Initial "<< Nk <<"steps done "<<std::endl;
-    std::cout<<GridLogIRL <<"Initial steps:OrthoTime "<<OrthoTime<< "seconds"<<std::endl;
-
-    //////////////////////////////////
-    // Restarting loop begins
-    //////////////////////////////////
-    int iter;
-    for(iter = 0; iter<MaxIter; ++iter){
-      
-      OrthoTime=0.;
-
-      std::cout<< GridLogMessage <<" **********************"<< std::endl;
-      std::cout<< GridLogMessage <<" Restart iteration = "<< iter << std::endl;
-      std::cout<< GridLogMessage <<" **********************"<< std::endl;
-
-      std::cout<<GridLogIRL <<" running "<<Nm-Nk <<" steps: "<<std::endl;
-      for(int k=Nk; k<Nm; ++k) step(eval,lme,evec,f,Nm,k);
-      f *= lme[Nm-1];
-
-      std::cout<<GridLogIRL <<" "<<Nm-Nk <<" steps done "<<std::endl;
-      std::cout<<GridLogIRL <<"Initial steps:OrthoTime "<<OrthoTime<< "seconds"<<std::endl;
-	  
-      //////////////////////////////////
-      // getting eigenvalues
-      //////////////////////////////////
-      for(int k=0; k<Nm; ++k){
-	eval2[k] = eval[k+k1-1];
-	lme2[k] = lme[k+k1-1];
-      }
-      Qt = Eigen::MatrixXd::Identity(Nm,Nm);
-      diagonalize(eval2,lme2,Nm,Nm,Qt,grid);
-      std::cout<<GridLogIRL <<" diagonalized "<<std::endl;
-
-      //////////////////////////////////
-      // sorting
-      //////////////////////////////////
-      eval2_copy = eval2;
-      std::partial_sort(eval2.begin(),eval2.begin()+Nm,eval2.end(),std::greater<RealD>());
-      std::cout<<GridLogIRL <<" evals sorted "<<std::endl;
-      const int chunk=8;
-      for(int io=0; io<k2;io+=chunk){
-	std::cout<<GridLogIRL << "eval "<< std::setw(3) << io ;
-	for(int ii=0;ii<chunk;ii++){
-	  if ( (io+ii)<k2 )
-	    std::cout<< " "<< std::setw(12)<< eval2[io+ii];
-	}
-	std::cout << std::endl;
-      }
-
-      //////////////////////////////////
-      // Implicitly shifted QR transformations
-      //////////////////////////////////
-      Qt = Eigen::MatrixXd::Identity(Nm,Nm);
-      for(int ip=k2; ip<Nm; ++ip){ 
-	QR_decomp(eval,lme,Nm,Nm,Qt,eval2[ip],k1,Nm);
-      }
-      std::cout<<GridLogIRL <<"QR decomposed "<<std::endl;
-
-      assert(k2<Nm);      assert(k2<Nm);      assert(k1>0);
-
-      basisRotate(evec,Qt,k1-1,k2+1,0,Nm,Nm); /// big constraint on the basis
-      std::cout<<GridLogIRL <<"basisRotated  by Qt"<<std::endl;
-      
-      ////////////////////////////////////////////////////
-      // Compressed vector f and beta(k2)
-      ////////////////////////////////////////////////////
-      f *= Qt(k2-1,Nm-1);
-      f += lme[k2-1] * evec[k2];
-      beta_k = norm2(f);
-      beta_k = sqrt(beta_k);
-      std::cout<<GridLogIRL<<" beta(k) = "<<beta_k<<std::endl;
-	  
-      RealD betar = 1.0/beta_k;
-      evec[k2] = betar * f;
-      lme[k2-1] = beta_k;
-	  
-      ////////////////////////////////////////////////////
-      // Convergence test
-      ////////////////////////////////////////////////////
-      for(int k=0; k<Nm; ++k){    
-	eval2[k] = eval[k];
-	lme2[k] = lme[k];
-      }
-      Qt = Eigen::MatrixXd::Identity(Nm,Nm);
-      diagonalize(eval2,lme2,Nk,Nm,Qt,grid);
-      std::cout<<GridLogIRL <<" Diagonalized "<<std::endl;
-	  
-      Nconv = 0;
-      if (iter >= MinRestart) {
-
-	std::cout << GridLogIRL << "Test convergence: rotate subset of vectors to test convergence " << std::endl;
-
-	Field B(grid); B.checkerboard = evec[0].checkerboard;
-
-	//  power of two search pattern;  not every evalue in eval2 is assessed.
-	int allconv =1;
-	for(int jj = 1; jj<=Nstop; jj*=2){
-	  int j = Nstop-jj;
-	  RealD e = eval2_copy[j]; // Discard the evalue
-	  basisRotateJ(B,evec,Qt,j,0,Nk,Nm);	    
-	  if( !_Tester.TestConvergence(j,eresid,B,e,evalMaxApprox) ) {
-	    allconv=0;
-	  }
-	}
-	// Do evec[0] for good measure
-	{ 
-	  int j=0;
-	  RealD e = eval2_copy[0]; 
-	  basisRotateJ(B,evec,Qt,j,0,Nk,Nm);	    
-	  if( !_Tester.TestConvergence(j,eresid,B,e,evalMaxApprox) ) allconv=0;
-	}
-	if ( allconv ) Nconv = Nstop;
-
-	// test if we converged, if so, terminate
-	std::cout<<GridLogIRL<<" #modes converged: >= "<<Nconv<<"/"<<Nstop<<std::endl;
-	//	if( Nconv>=Nstop || beta_k < betastp){
-	if( Nconv>=Nstop){
-	  goto converged;
-	}
-	  
-      } else {
-	std::cout << GridLogIRL << "iter < MinRestart: do not yet test for convergence\n";
-      } // end of iter loop
-    }
-
-    std::cout<<GridLogError<<"\n NOT converged.\n";
-    abort();
-	
-  converged:
-    {
-      Field B(grid); B.checkerboard = evec[0].checkerboard;
-      basisRotate(evec,Qt,0,Nk,0,Nk,Nm);	    
-      std::cout << GridLogIRL << " Rotated basis"<<std::endl;
-      Nconv=0;
-      //////////////////////////////////////////////////////////////////////
-      // Full final convergence test; unconditionally applied
-      //////////////////////////////////////////////////////////////////////
-      for(int j = 0; j<=Nk; j++){
-	B=evec[j];
-	if( _Tester.ReconstructEval(j,eresid,B,eval2[j],evalMaxApprox) ) {
-	  Nconv++;
-	}
-      }
-
-      if ( Nconv < Nstop )
-	std::cout << GridLogIRL << "Nconv ("<<Nconv<<") < Nstop ("<<Nstop<<")"<<std::endl;
-
-      eval=eval2;
-      
-      //Keep only converged
-      eval.resize(Nconv);// Nstop?
-      evec.resize(Nconv,grid);// Nstop?
-      basisSortInPlace(evec,eval,reverse);
-      
-    }
-       
-    std::cout << GridLogIRL <<"**************************************************************************"<< std::endl;
-    std::cout << GridLogIRL << "ImplicitlyRestartedLanczos CONVERGED ; Summary :\n";
-    std::cout << GridLogIRL <<"**************************************************************************"<< std::endl;
-    std::cout << GridLogIRL << " -- Iterations  = "<< iter   << "\n";
-    std::cout << GridLogIRL << " -- beta(k)     = "<< beta_k << "\n";
-    std::cout << GridLogIRL << " -- Nconv       = "<< Nconv  << "\n";
-    std::cout << GridLogIRL <<"**************************************************************************"<< std::endl;
-  }
-
- private:
-/* Saad PP. 195
-1. Choose an initial vector v1 of 2-norm unity. Set β1 ≡ 0, v0 ≡ 0
-2. For k = 1,2,...,m Do:
-3. wk:=Avk−βkv_{k−1}      
-4. αk:=(wk,vk)       // 
-5. wk:=wk−αkvk       // wk orthog vk 
-6. βk+1 := ∥wk∥2. If βk+1 = 0 then Stop
-7. vk+1 := wk/βk+1
-8. EndDo
- */
-  void step(std::vector<RealD>& lmd,
-	    std::vector<RealD>& lme, 
-	    std::vector<Field>& evec,
-	    Field& w,int Nm,int k)
-  {
-    const RealD tiny = 1.0e-20;
-    assert( k< Nm );
-
-    GridStopWatch gsw_op,gsw_o;
-
-    Field& evec_k = evec[k];
-
-    _PolyOp(evec_k,w);    std::cout<<GridLogIRL << "PolyOp" <<std::endl;
-
-    if(k>0) w -= lme[k-1] * evec[k-1];
-
-    ComplexD zalph = innerProduct(evec_k,w); // 4. αk:=(wk,vk)
-    RealD     alph = real(zalph);
-
-    w = w - alph * evec_k;// 5. wk:=wk−αkvk
-
-    RealD beta = normalise(w); // 6. βk+1 := ∥wk∥2. If βk+1 = 0 then Stop
-    // 7. vk+1 := wk/βk+1
-
-    lmd[k] = alph;
-    lme[k] = beta;
-
-    if (k>0 && k % orth_period == 0) {
-      orthogonalize(w,evec,k); // orthonormalise
-      std::cout<<GridLogIRL << "Orthogonalised " <<std::endl;
-    }
-
-    if(k < Nm-1) evec[k+1] = w;
-
-    std::cout<<GridLogIRL << "alpha[" << k << "] = " << zalph << " beta[" << k << "] = "<<beta<<std::endl;
-    if ( beta < tiny ) 
-      std::cout<<GridLogIRL << " beta is tiny "<<beta<<std::endl;
-  }
-
-  void diagonalize_Eigen(std::vector<RealD>& lmd, std::vector<RealD>& lme, 
-			 int Nk, int Nm,  
-			 Eigen::MatrixXd & Qt, // Nm x Nm
-			 GridBase *grid)
-  {
-    Eigen::MatrixXd TriDiag = Eigen::MatrixXd::Zero(Nk,Nk);
-
-    for(int i=0;i<Nk;i++)   TriDiag(i,i)   = lmd[i];
-    for(int i=0;i<Nk-1;i++) TriDiag(i,i+1) = lme[i];
-    for(int i=0;i<Nk-1;i++) TriDiag(i+1,i) = lme[i];
-    
-    Eigen::SelfAdjointEigenSolver<Eigen::MatrixXd> eigensolver(TriDiag);
-
-    for (int i = 0; i < Nk; i++) {
-      lmd[Nk-1-i] = eigensolver.eigenvalues()(i);
-    }
-    for (int i = 0; i < Nk; i++) {
-      for (int j = 0; j < Nk; j++) {
-	Qt(Nk-1-i,j) = eigensolver.eigenvectors()(j,i);
-      }
-    }
-  }
-
-  ///////////////////////////////////////////////////////////////////////////
-  // File could end here if settle on Eigen ??? !!!
-  ///////////////////////////////////////////////////////////////////////////
-  void QR_decomp(std::vector<RealD>& lmd,   // Nm 
-		 std::vector<RealD>& lme,   // Nm 
-		 int Nk, int Nm,            // Nk, Nm
-		 Eigen::MatrixXd& Qt,       // Nm x Nm matrix
-		 RealD Dsh, int kmin, int kmax)
-  {
-    int k = kmin-1;
-    RealD x;
-    
-    RealD Fden = 1.0/hypot(lmd[k]-Dsh,lme[k]);
-    RealD c = ( lmd[k] -Dsh) *Fden;
-    RealD s = -lme[k] *Fden;
-      
-    RealD tmpa1 = lmd[k];
-    RealD tmpa2 = lmd[k+1];
-    RealD tmpb  = lme[k];
-
-    lmd[k]   = c*c*tmpa1 +s*s*tmpa2 -2.0*c*s*tmpb;
-    lmd[k+1] = s*s*tmpa1 +c*c*tmpa2 +2.0*c*s*tmpb;
-    lme[k]   = c*s*(tmpa1-tmpa2) +(c*c-s*s)*tmpb;
-    x        =-s*lme[k+1];
-    lme[k+1] = c*lme[k+1];
-      
-    for(int i=0; i<Nk; ++i){
-      RealD Qtmp1 = Qt(k,i);
-      RealD Qtmp2 = Qt(k+1,i);
-      Qt(k,i)  = c*Qtmp1 - s*Qtmp2;
-      Qt(k+1,i)= s*Qtmp1 + c*Qtmp2; 
-    }
-
-    // Givens transformations
-    for(int k = kmin; k < kmax-1; ++k){
-      
-      RealD Fden = 1.0/hypot(x,lme[k-1]);
-      RealD c = lme[k-1]*Fden;
-      RealD s = - x*Fden;
-	
-      RealD tmpa1 = lmd[k];
-      RealD tmpa2 = lmd[k+1];
-      RealD tmpb  = lme[k];
-
-      lmd[k]   = c*c*tmpa1 +s*s*tmpa2 -2.0*c*s*tmpb;
-      lmd[k+1] = s*s*tmpa1 +c*c*tmpa2 +2.0*c*s*tmpb;
-      lme[k]   = c*s*(tmpa1-tmpa2) +(c*c-s*s)*tmpb;
-      lme[k-1] = c*lme[k-1] -s*x;
-
-      if(k != kmax-2){
-	x = -s*lme[k+1];
-	lme[k+1] = c*lme[k+1];
-      }
-
-      for(int i=0; i<Nk; ++i){
-	RealD Qtmp1 = Qt(k,i);
-	RealD Qtmp2 = Qt(k+1,i);
-	Qt(k,i)     = c*Qtmp1 -s*Qtmp2;
-	Qt(k+1,i)   = s*Qtmp1 +c*Qtmp2;
-      }
-    }
-  }
-
-  void diagonalize(std::vector<RealD>& lmd, std::vector<RealD>& lme, 
-		   int Nk, int Nm,   
-		   Eigen::MatrixXd & Qt,
-		   GridBase *grid)
-  {
-    Qt = Eigen::MatrixXd::Identity(Nm,Nm);
-    if ( diagonalisation == IRLdiagonaliseWithDSTEGR ) {
-      diagonalize_lapack(lmd,lme,Nk,Nm,Qt,grid);
-    } else if ( diagonalisation == IRLdiagonaliseWithQR ) { 
-      diagonalize_QR(lmd,lme,Nk,Nm,Qt,grid);
-    }  else if ( diagonalisation == IRLdiagonaliseWithEigen ) { 
-      diagonalize_Eigen(lmd,lme,Nk,Nm,Qt,grid);
-    } else { 
-      assert(0);
-    }
-  }
-
-#ifdef USE_LAPACK
-void LAPACK_dstegr(char *jobz, char *range, int *n, double *d, double *e,
-                   double *vl, double *vu, int *il, int *iu, double *abstol,
-                   int *m, double *w, double *z, int *ldz, int *isuppz,
-                   double *work, int *lwork, int *iwork, int *liwork,
-                   int *info);
-#endif
-
-void diagonalize_lapack(std::vector<RealD>& lmd,
-			std::vector<RealD>& lme, 
-			int Nk, int Nm,  
-			Eigen::MatrixXd& Qt,
-			GridBase *grid)
-{
-#ifdef USE_LAPACK
-  const int size = Nm;
-  int NN = Nk;
-  double evals_tmp[NN];
-  double evec_tmp[NN][NN];
-  memset(evec_tmp[0],0,sizeof(double)*NN*NN);
-  double DD[NN];
-  double EE[NN];
-  for (int i = 0; i< NN; i++) {
-    for (int j = i - 1; j <= i + 1; j++) {
-      if ( j < NN && j >= 0 ) {
-	if (i==j) DD[i] = lmd[i];
-	if (i==j) evals_tmp[i] = lmd[i];
-	if (j==(i-1)) EE[j] = lme[j];
-      }
-    }
-  }
-  int evals_found;
-  int lwork = ( (18*NN) > (1+4*NN+NN*NN)? (18*NN):(1+4*NN+NN*NN)) ;
-  int liwork =  3+NN*10 ;
-  int iwork[liwork];
-  double work[lwork];
-  int isuppz[2*NN];
-  char jobz = 'V'; // calculate evals & evecs
-  char range = 'I'; // calculate all evals
-  //    char range = 'A'; // calculate all evals
-  char uplo = 'U'; // refer to upper half of original matrix
-  char compz = 'I'; // Compute eigenvectors of tridiagonal matrix
-  int ifail[NN];
-  int info;
-  int total = grid->_Nprocessors;
-  int node  = grid->_processor;
-  int interval = (NN/total)+1;
-  double vl = 0.0, vu = 0.0;
-  int il = interval*node+1 , iu = interval*(node+1);
-  if (iu > NN)  iu=NN;
-  double tol = 0.0;
-  if (1) {
-    memset(evals_tmp,0,sizeof(double)*NN);
-    if ( il <= NN){
-      LAPACK_dstegr(&jobz, &range, &NN,
-		    (double*)DD, (double*)EE,
-		    &vl, &vu, &il, &iu, // these four are ignored if second parameteris 'A'
-		    &tol, // tolerance
-		    &evals_found, evals_tmp, (double*)evec_tmp, &NN,
-		    isuppz,
-		    work, &lwork, iwork, &liwork,
-		    &info);
-      for (int i = iu-1; i>= il-1; i--){
-	evals_tmp[i] = evals_tmp[i - (il-1)];
-	if (il>1) evals_tmp[i-(il-1)]=0.;
-	for (int j = 0; j< NN; j++){
-	  evec_tmp[i][j] = evec_tmp[i - (il-1)][j];
-	  if (il>1) evec_tmp[i-(il-1)][j]=0.;
-	}
-      }
-    }
-    {
-      grid->GlobalSumVector(evals_tmp,NN);
-      grid->GlobalSumVector((double*)evec_tmp,NN*NN);
-    }
-  } 
-  // Safer to sort instead of just reversing it, 
-  // but the document of the routine says evals are sorted in increasing order. 
-  // qr gives evals in decreasing order.
-  for(int i=0;i<NN;i++){
-    lmd [NN-1-i]=evals_tmp[i];
-    for(int j=0;j<NN;j++){
-      Qt((NN-1-i),j)=evec_tmp[i][j];
-    }
-  }
-#else 
-  assert(0);
-#endif
-}
-
-void diagonalize_QR(std::vector<RealD>& lmd, std::vector<RealD>& lme, 
-		    int Nk, int Nm,   
-		    Eigen::MatrixXd & Qt,
-		    GridBase *grid)
-{
-  int QRiter = 100*Nm;
-  int kmin = 1;
-  int kmax = Nk;
-  
-  // (this should be more sophisticated)
-  for(int iter=0; iter<QRiter; ++iter){
-    
-    // determination of 2x2 leading submatrix
-    RealD dsub = lmd[kmax-1]-lmd[kmax-2];
-    RealD dd = sqrt(dsub*dsub + 4.0*lme[kmax-2]*lme[kmax-2]);
-    RealD Dsh = 0.5*(lmd[kmax-2]+lmd[kmax-1] +dd*(dsub/fabs(dsub)));
-    // (Dsh: shift)
-    
-    // transformation
-    QR_decomp(lmd,lme,Nk,Nm,Qt,Dsh,kmin,kmax); // Nk, Nm
-    
-    // Convergence criterion (redef of kmin and kamx)
-    for(int j=kmax-1; j>= kmin; --j){
-      RealD dds = fabs(lmd[j-1])+fabs(lmd[j]);
-      if(fabs(lme[j-1])+dds > dds){
-	kmax = j+1;
-	goto continued;
-      }
-    }
-    QRiter = iter;
-    return;
-    
-  continued:
-    for(int j=0; j<kmax-1; ++j){
-      RealD dds = fabs(lmd[j])+fabs(lmd[j+1]);
-      if(fabs(lme[j])+dds > dds){
-	kmin = j+1;
-	break;
-      }
-    }
-  }
-  std::cout << GridLogError << "[QL method] Error - Too many iteration: "<<QRiter<<"\n";
-  abort();
-}
-};
-}
-#endif

Grid/algorithms/iterative/LocalCoherenceLanczos.h

@@ -1,406 +0,0 @@
-    /*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/algorithms/iterative/LocalCoherenceLanczos.h
-
-    Copyright (C) 2015
-
-Author: Christoph Lehner <clehner@bnl.gov>
-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_LOCAL_COHERENCE_IRL_H
-#define GRID_LOCAL_COHERENCE_IRL_H
-
-namespace Grid { 
-
-
-struct LanczosParams : Serializable {
- public:
-  GRID_SERIALIZABLE_CLASS_MEMBERS(LanczosParams,
-				  ChebyParams, Cheby,/*Chebyshev*/
-				  int, Nstop,    /*Vecs in Lanczos must converge Nstop < Nk < Nm*/
-				  int, Nk,       /*Vecs in Lanczos seek converge*/
-				  int, Nm,       /*Total vecs in Lanczos include restart*/
-				  RealD, resid,  /*residual*/
- 				  int, MaxIt, 
-				  RealD, betastp,  /* ? */
-				  int, MinRes);    // Must restart
-};
-
-struct LocalCoherenceLanczosParams : Serializable {
- public:
-  GRID_SERIALIZABLE_CLASS_MEMBERS(LocalCoherenceLanczosParams,
-				  bool, saveEvecs,
-				  bool, doFine,
-				  bool, doFineRead,
-				  bool, doCoarse,
-	       			  bool, doCoarseRead,
-				  LanczosParams, FineParams,
-				  LanczosParams, CoarseParams,
-				  ChebyParams,   Smoother,
-				  RealD        , coarse_relax_tol,
-				  std::vector<int>, blockSize,
-				  std::string, config,
-				  std::vector < std::complex<double>  >, omega,
-				  RealD, mass,
-				  RealD, M5);
-};
-
-// Duplicate functionality; ProjectedFunctionHermOp could be used with the trivial function
-template<class Fobj,class CComplex,int nbasis>
-class ProjectedHermOp : public LinearFunction<Lattice<iVector<CComplex,nbasis > > > {
-public:
-  typedef iVector<CComplex,nbasis >           CoarseSiteVector;
-  typedef Lattice<CoarseSiteVector>           CoarseField;
-  typedef Lattice<CComplex>   CoarseScalar; // used for inner products on fine field
-  typedef Lattice<Fobj>          FineField;
-
-  LinearOperatorBase<FineField> &_Linop;
-  std::vector<FineField>        &subspace;
-
-  ProjectedHermOp(LinearOperatorBase<FineField>& linop, std::vector<FineField> & _subspace) : 
-    _Linop(linop), subspace(_subspace)
-  {  
-    assert(subspace.size() >0);
-  };
-
-  void operator()(const CoarseField& in, CoarseField& out) {
-    GridBase *FineGrid = subspace[0]._grid;    
-    int   checkerboard = subspace[0].checkerboard;
-      
-    FineField fin (FineGrid);     fin.checkerboard= checkerboard;
-    FineField fout(FineGrid);   fout.checkerboard = checkerboard;
-
-    blockPromote(in,fin,subspace);       std::cout<<GridLogIRL<<"ProjectedHermop : Promote to fine"<<std::endl;
-    _Linop.HermOp(fin,fout);             std::cout<<GridLogIRL<<"ProjectedHermop : HermOp (fine) "<<std::endl;
-    blockProject(out,fout,subspace);     std::cout<<GridLogIRL<<"ProjectedHermop : Project to coarse "<<std::endl;
-  }
-};
-
-template<class Fobj,class CComplex,int nbasis>
-class ProjectedFunctionHermOp : public LinearFunction<Lattice<iVector<CComplex,nbasis > > > {
-public:
-  typedef iVector<CComplex,nbasis >           CoarseSiteVector;
-  typedef Lattice<CoarseSiteVector>           CoarseField;
-  typedef Lattice<CComplex>   CoarseScalar; // used for inner products on fine field
-  typedef Lattice<Fobj>          FineField;
-
-
-  OperatorFunction<FineField>   & _poly;
-  LinearOperatorBase<FineField> &_Linop;
-  std::vector<FineField>        &subspace;
-
-  ProjectedFunctionHermOp(OperatorFunction<FineField> & poly,
-			  LinearOperatorBase<FineField>& linop, 
-			  std::vector<FineField> & _subspace) :
-    _poly(poly),
-    _Linop(linop),
-    subspace(_subspace)
-  {  };
-
-  void operator()(const CoarseField& in, CoarseField& out) {
-    
-    GridBase *FineGrid = subspace[0]._grid;    
-    int   checkerboard = subspace[0].checkerboard;
-
-    FineField fin (FineGrid); fin.checkerboard =checkerboard;
-    FineField fout(FineGrid);fout.checkerboard =checkerboard;
-    
-    blockPromote(in,fin,subspace);             std::cout<<GridLogIRL<<"ProjectedFunctionHermop : Promote to fine"<<std::endl;
-    _poly(_Linop,fin,fout);                    std::cout<<GridLogIRL<<"ProjectedFunctionHermop : Poly "<<std::endl;
-    blockProject(out,fout,subspace);           std::cout<<GridLogIRL<<"ProjectedFunctionHermop : Project to coarse "<<std::endl;
-  }
-};
-
-template<class Fobj,class CComplex,int nbasis>
-class ImplicitlyRestartedLanczosSmoothedTester  : public ImplicitlyRestartedLanczosTester<Lattice<iVector<CComplex,nbasis > > >
-{
- public:
-  typedef iVector<CComplex,nbasis >           CoarseSiteVector;
-  typedef Lattice<CoarseSiteVector>           CoarseField;
-  typedef Lattice<CComplex>   CoarseScalar; // used for inner products on fine field
-  typedef Lattice<Fobj>          FineField;
-
-  LinearFunction<CoarseField> & _Poly;
-  OperatorFunction<FineField>   & _smoother;
-  LinearOperatorBase<FineField> &_Linop;
-  RealD                          _coarse_relax_tol;
-  std::vector<FineField>        &_subspace;
-  
-  ImplicitlyRestartedLanczosSmoothedTester(LinearFunction<CoarseField>   &Poly,
-					   OperatorFunction<FineField>   &smoother,
-					   LinearOperatorBase<FineField> &Linop,
-					   std::vector<FineField>        &subspace,
-					   RealD coarse_relax_tol=5.0e3) 
-    : _smoother(smoother), _Linop(Linop), _Poly(Poly), _subspace(subspace),
-      _coarse_relax_tol(coarse_relax_tol)  
-  {    };
-
-  int TestConvergence(int j,RealD eresid,CoarseField &B, RealD &eval,RealD evalMaxApprox)
-  {
-    CoarseField v(B);
-    RealD eval_poly = eval;
-
-    // Apply operator
-    _Poly(B,v);
-
-    RealD vnum = real(innerProduct(B,v)); // HermOp.
-    RealD vden = norm2(B);
-    RealD vv0  = norm2(v);
-    eval   = vnum/vden;
-    v -= eval*B;
-
-    RealD vv = norm2(v) / ::pow(evalMaxApprox,2.0);
-
-    std::cout.precision(13);
-    std::cout<<GridLogIRL  << "[" << std::setw(3)<<j<<"] "
-	     <<"eval = "<<std::setw(25)<< eval << " (" << eval_poly << ")"
-	     <<" |H B[i] - eval[i]B[i]|^2 / evalMaxApprox^2 " << std::setw(25) << vv
-	     <<std::endl;
-
-    int conv=0;
-    if( (vv<eresid*eresid) ) conv = 1;
-    return conv;
-  }
-  int ReconstructEval(int j,RealD eresid,CoarseField &B, RealD &eval,RealD evalMaxApprox)
-  {
-    GridBase *FineGrid = _subspace[0]._grid;    
-    int checkerboard   = _subspace[0].checkerboard;
-    FineField fB(FineGrid);fB.checkerboard =checkerboard;
-    FineField fv(FineGrid);fv.checkerboard =checkerboard;
-
-    blockPromote(B,fv,_subspace);  
-    
-    _smoother(_Linop,fv,fB); 
-
-    RealD eval_poly = eval;
-    _Linop.HermOp(fB,fv);
-
-    RealD vnum = real(innerProduct(fB,fv)); // HermOp.
-    RealD vden = norm2(fB);
-    RealD vv0  = norm2(fv);
-    eval   = vnum/vden;
-    fv -= eval*fB;
-    RealD vv = norm2(fv) / ::pow(evalMaxApprox,2.0);
-
-    std::cout.precision(13);
-    std::cout<<GridLogIRL  << "[" << std::setw(3)<<j<<"] "
-	     <<"eval = "<<std::setw(25)<< eval << " (" << eval_poly << ")"
-	     <<" |H B[i] - eval[i]B[i]|^2 / evalMaxApprox^2 " << std::setw(25) << vv
-	     <<std::endl;
-    if ( j > nbasis ) eresid = eresid*_coarse_relax_tol;
-    if( (vv<eresid*eresid) ) return 1;
-    return 0;
-  }
-};
-
-////////////////////////////////////////////
-// Make serializable Lanczos params
-////////////////////////////////////////////
-template<class Fobj,class CComplex,int nbasis>
-class LocalCoherenceLanczos 
-{
-public:
-  typedef iVector<CComplex,nbasis >           CoarseSiteVector;
-  typedef Lattice<CComplex>                   CoarseScalar; // used for inner products on fine field
-  typedef Lattice<CoarseSiteVector>           CoarseField;
-  typedef Lattice<Fobj>                       FineField;
-
-protected:
-  GridBase *_CoarseGrid;
-  GridBase *_FineGrid;
-  int _checkerboard;
-  LinearOperatorBase<FineField>                 & _FineOp;
-  
-  std::vector<RealD>                              &evals_fine;
-  std::vector<RealD>                              &evals_coarse; 
-  std::vector<FineField>                          &subspace;
-  std::vector<CoarseField>                        &evec_coarse;
-
-private:
-  std::vector<RealD>                              _evals_fine;
-  std::vector<RealD>                              _evals_coarse; 
-  std::vector<FineField>                          _subspace;
-  std::vector<CoarseField>                        _evec_coarse;
-
-public:
-
-  LocalCoherenceLanczos(GridBase *FineGrid,
-			GridBase *CoarseGrid,
-			LinearOperatorBase<FineField> &FineOp,
-			int checkerboard) :
-    _CoarseGrid(CoarseGrid),
-    _FineGrid(FineGrid),
-    _FineOp(FineOp),
-    _checkerboard(checkerboard),
-    evals_fine  (_evals_fine),
-    evals_coarse(_evals_coarse),
-    subspace    (_subspace),
-    evec_coarse(_evec_coarse)
-  {
-    evals_fine.resize(0);
-    evals_coarse.resize(0);
-  };
-  //////////////////////////////////////////////////////////////////////////
-  // Alternate constructore, external storage for use by Hadrons module
-  //////////////////////////////////////////////////////////////////////////
-  LocalCoherenceLanczos(GridBase *FineGrid,
-			GridBase *CoarseGrid,
-			LinearOperatorBase<FineField> &FineOp,
-			int checkerboard,
-			std::vector<FineField>   &ext_subspace,
-			std::vector<CoarseField> &ext_coarse,
-			std::vector<RealD>       &ext_eval_fine,
-			std::vector<RealD>       &ext_eval_coarse
-			) :
-    _CoarseGrid(CoarseGrid),
-    _FineGrid(FineGrid),
-    _FineOp(FineOp),
-    _checkerboard(checkerboard),
-    evals_fine  (ext_eval_fine), 
-    evals_coarse(ext_eval_coarse),
-    subspace    (ext_subspace),
-    evec_coarse (ext_coarse)
-  {
-    evals_fine.resize(0);
-    evals_coarse.resize(0);
-  };
-
-  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);
-  };
-
-  template<typename T>  static RealD normalise(T& v) 
-  {
-    RealD nn = norm2(v);
-    nn = ::sqrt(nn);
-    v = v * (1.0/nn);
-    return nn;
-  }
-  /*
-  void fakeFine(void)
-  {
-    int Nk = nbasis;
-    subspace.resize(Nk,_FineGrid);
-    subspace[0]=1.0;
-    subspace[0].checkerboard=_checkerboard;
-    normalise(subspace[0]);
-    PlainHermOp<FineField>    Op(_FineOp);
-    for(int k=1;k<Nk;k++){
-      subspace[k].checkerboard=_checkerboard;
-      Op(subspace[k-1],subspace[k]);
-      normalise(subspace[k]);
-    }
-  }
-  */
-
-  void testFine(RealD resid) 
-  {
-    assert(evals_fine.size() == nbasis);
-    assert(subspace.size() == nbasis);
-    PlainHermOp<FineField>    Op(_FineOp);
-    ImplicitlyRestartedLanczosHermOpTester<FineField> SimpleTester(Op);
-    for(int k=0;k<nbasis;k++){
-      assert(SimpleTester.ReconstructEval(k,resid,subspace[k],evals_fine[k],1.0)==1);
-    }
-  }
-
-  void testCoarse(RealD resid,ChebyParams cheby_smooth,RealD relax) 
-  {
-    assert(evals_fine.size() == nbasis);
-    assert(subspace.size() == nbasis);
-    //////////////////////////////////////////////////////////////////////////////////////////////////
-    // create a smoother and see if we can get a cheap convergence test and smooth inside the IRL
-    //////////////////////////////////////////////////////////////////////////////////////////////////
-    Chebyshev<FineField>                          ChebySmooth(cheby_smooth);
-    ProjectedFunctionHermOp<Fobj,CComplex,nbasis> ChebyOp (ChebySmooth,_FineOp,subspace);
-    ImplicitlyRestartedLanczosSmoothedTester<Fobj,CComplex,nbasis> ChebySmoothTester(ChebyOp,ChebySmooth,_FineOp,subspace,relax);
-
-    for(int k=0;k<evec_coarse.size();k++){
-      if ( k < nbasis ) { 
-	assert(ChebySmoothTester.ReconstructEval(k,resid,evec_coarse[k],evals_coarse[k],1.0)==1);
-      } else { 
-	assert(ChebySmoothTester.ReconstructEval(k,resid*relax,evec_coarse[k],evals_coarse[k],1.0)==1);
-      }
-    }
-  }
-
-  void calcFine(ChebyParams cheby_parms,int Nstop,int Nk,int Nm,RealD resid, 
-		RealD MaxIt, RealD betastp, int MinRes)
-  {
-    assert(nbasis<=Nm);
-    Chebyshev<FineField>      Cheby(cheby_parms);
-    FunctionHermOp<FineField> ChebyOp(Cheby,_FineOp);
-    PlainHermOp<FineField>    Op(_FineOp);
-
-    evals_fine.resize(Nm);
-    subspace.resize(Nm,_FineGrid);
-
-    ImplicitlyRestartedLanczos<FineField> IRL(ChebyOp,Op,Nstop,Nk,Nm,resid,MaxIt,betastp,MinRes);
-
-    FineField src(_FineGrid); src=1.0; src.checkerboard = _checkerboard;
-
-    int Nconv;
-    IRL.calc(evals_fine,subspace,src,Nconv,false);
-    
-    // Shrink down to number saved
-    assert(Nstop>=nbasis);
-    assert(Nconv>=nbasis);
-    evals_fine.resize(nbasis);
-    subspace.resize(nbasis,_FineGrid);
-  }
-  void calcCoarse(ChebyParams cheby_op,ChebyParams cheby_smooth,RealD relax,
-		  int Nstop, int Nk, int Nm,RealD resid, 
-		  RealD MaxIt, RealD betastp, int MinRes)
-  {
-    Chebyshev<FineField>                          Cheby(cheby_op);
-    ProjectedHermOp<Fobj,CComplex,nbasis>         Op(_FineOp,subspace);
-    ProjectedFunctionHermOp<Fobj,CComplex,nbasis> ChebyOp (Cheby,_FineOp,subspace);
-    //////////////////////////////////////////////////////////////////////////////////////////////////
-    // create a smoother and see if we can get a cheap convergence test and smooth inside the IRL
-    //////////////////////////////////////////////////////////////////////////////////////////////////
-
-    Chebyshev<FineField>                                           ChebySmooth(cheby_smooth);
-    ImplicitlyRestartedLanczosSmoothedTester<Fobj,CComplex,nbasis> ChebySmoothTester(ChebyOp,ChebySmooth,_FineOp,subspace,relax);
-
-    evals_coarse.resize(Nm);
-    evec_coarse.resize(Nm,_CoarseGrid);
-
-    CoarseField src(_CoarseGrid);     src=1.0; 
-
-    ImplicitlyRestartedLanczos<CoarseField> IRL(ChebyOp,ChebyOp,ChebySmoothTester,Nstop,Nk,Nm,resid,MaxIt,betastp,MinRes);
-    int Nconv=0;
-    IRL.calc(evals_coarse,evec_coarse,src,Nconv,false);
-    assert(Nconv>=Nstop);
-    evals_coarse.resize(Nstop);
-    evec_coarse.resize (Nstop,_CoarseGrid);
-    for (int i=0;i<Nstop;i++){
-      std::cout << i << " Coarse eval = " << evals_coarse[i]  << std::endl;
-    }
-  }
-};
-
-}
-#endif

Grid/algorithms/iterative/MinimalResidual.h

@@ -1,156 +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:
-  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);
-
-    Complex a, c;
-    Real    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:    mp " << d << 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

@@ -1,273 +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:
-  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<std::complex<double>> y;
-  std::vector<std::complex<double>> gamma;
-  std::vector<std::complex<double>> c;
-  std::vector<std::complex<double>> 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 = std::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 - H(iter, i) * v[i];
-    }
-
-    H(iter, iter + 1) = sqrt(norm2(w));
-    v[iter + 1] = (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] * H(iter, i) + c[i] * H(iter, i + 1);
-      H(iter, i)     = std::conj(c[i]) * H(iter, i) + std::conj(s[i]) * H(iter, i + 1);
-      H(iter, i + 1) = tmp;
-    }
-
-    // Compute new Givens Rotation
-    ComplexD 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]     = std::conj(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] - H(k, i) * y[k];
-      y[i] = y[i] / 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/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.cc

@@ -1,125 +0,0 @@
-#include <Grid/GridCore.h>
-#include <fcntl.h>
-
-namespace Grid {
-
-MemoryStats *MemoryProfiler::stats = nullptr;
-bool         MemoryProfiler::debug = false;
-
-int PointerCache::victim;
-
-PointerCache::PointerCacheEntry PointerCache::Entries[PointerCache::Ncache];
-
-void *PointerCache::Insert(void *ptr,size_t bytes) {
-
-  if (bytes < 4096 ) return ptr;
-
-#ifdef GRID_OMP
-  assert(omp_in_parallel()==0);
-#endif 
-
-  void * ret = NULL;
-  int v = -1;
-
-  for(int e=0;e<Ncache;e++) {
-    if ( Entries[e].valid==0 ) {
-      v=e; 
-      break;
-    }
-  }
-
-  if ( v==-1 ) {
-    v=victim;
-    victim = (victim+1)%Ncache;
-  }
-
-  if ( Entries[v].valid ) {
-    ret = Entries[v].address;
-    Entries[v].valid = 0;
-    Entries[v].address = NULL;
-    Entries[v].bytes = 0;
-  }
-
-  Entries[v].address=ptr;
-  Entries[v].bytes  =bytes;
-  Entries[v].valid  =1;
-
-  return ret;
-}
-
-void *PointerCache::Lookup(size_t bytes) {
-
- if (bytes < 4096 ) return NULL;
-
-#ifdef _OPENMP
-  assert(omp_in_parallel()==0);
-#endif 
-
-  for(int e=0;e<Ncache;e++){
-    if ( Entries[e].valid && ( Entries[e].bytes == bytes ) ) {
-      Entries[e].valid = 0;
-      return Entries[e].address;
-    }
-  }
-  return NULL;
-}
-
-
-void check_huge_pages(void *Buf,uint64_t BYTES)
-{
-#ifdef __linux__
-  int fd = open("/proc/self/pagemap", O_RDONLY);
-  assert(fd >= 0);
-  const int page_size = 4096;
-  uint64_t virt_pfn = (uint64_t)Buf / page_size;
-  off_t offset = sizeof(uint64_t) * virt_pfn;
-  uint64_t npages = (BYTES + page_size-1) / page_size;
-  uint64_t pagedata[npages];
-  uint64_t ret = lseek(fd, offset, SEEK_SET);
-  assert(ret == offset);
-  ret = ::read(fd, pagedata, sizeof(uint64_t)*npages);
-  assert(ret == sizeof(uint64_t) * npages);
-  int nhugepages = npages / 512;
-  int n4ktotal, nnothuge;
-  n4ktotal = 0;
-  nnothuge = 0;
-  for (int i = 0; i < nhugepages; ++i) {
-    uint64_t baseaddr = (pagedata[i*512] & 0x7fffffffffffffULL) * page_size;
-    for (int j = 0; j < 512; ++j) {
-      uint64_t pageaddr = (pagedata[i*512+j] & 0x7fffffffffffffULL) * page_size;
-      ++n4ktotal;
-      if (pageaddr != baseaddr + j * page_size)
-	++nnothuge;
-      }
-  }
-  int rank = CartesianCommunicator::RankWorld();
-  printf("rank %d Allocated %d 4k pages, %d not in huge pages\n", rank, n4ktotal, nnothuge);
-#endif
-}
-
-std::string sizeString(const size_t bytes)
-{
-  constexpr unsigned int bufSize = 256;
-  const char             *suffixes[7] = {"", "K", "M", "G", "T", "P", "E"};
-  char                   buf[256];
-  size_t                 s     = 0;
-  double                 count = bytes;
-  
-  while (count >= 1024 && s < 7)
-  {
-      s++;
-      count /= 1024;
-  }
-  if (count - floor(count) == 0.0)
-  {
-      snprintf(buf, bufSize, "%d %sB", (int)count, suffixes[s]);
-  }
-  else
-  {
-      snprintf(buf, bufSize, "%.1f %sB", count, suffixes[s]);
-  }
-  
-  return std::string(buf);
-}
-
-}

Grid/allocator/AlignedAllocator.h

@@ -1,315 +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
-
-namespace Grid {
-
-  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) ;
-
-  };
-  
-  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);
-
-    _Tp *ptr = (_Tp *) PointerCache::Lookup(bytes);
-    //    if ( ptr != NULL ) 
-    //      std::cout << "alignedAllocator "<<__n << " cache hit "<< std::hex << ptr <<std::dec <<std::endl;
-
-    //////////////////
-    // Hack 2MB align; could make option probably doesn't need configurability
-    //////////////////
-//define GRID_ALLOC_ALIGN (128)
-#define GRID_ALLOC_ALIGN (2*1024*1024)
-#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
-    //    std::cout << "alignedAllocator " << std::hex << ptr <<std::dec <<std::endl;
-    // First touch optimise in threaded loop
-    uint8_t *cp = (uint8_t *)ptr;
-#ifdef GRID_OMP
-#pragma omp parallel for
-#endif
-    for(size_type n=0;n<bytes;n+=4096){
-      cp[n]=0;
-    }
-    return ptr;
-  }
-
-  void deallocate(pointer __p, size_type __n) { 
-    size_type bytes = __n * sizeof(_Tp);
-
-    profilerFree(bytes);
-
-    pointer __freeme = (pointer)PointerCache::Insert((void *)__p,bytes);
-
-#ifdef HAVE_MM_MALLOC_H
-    if ( __freeme ) _mm_free((void *)__freeme); 
-#else
-    if ( __freeme ) free((void *)__freeme);
-#endif
-  }
-  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; }
-
-//////////////////////////////////////////////////////////////////////////////////////////
-// MPI3 : comms must use shm region
-// SHMEM: comms must use symmetric heap
-//////////////////////////////////////////////////////////////////////////////////////////
-#ifdef GRID_COMMS_SHMEM
-extern "C" { 
-#include <mpp/shmem.h>
-extern void * shmem_align(size_t, size_t);
-extern void  shmem_free(void *);
-}
-#define PARANOID_SYMMETRIC_HEAP
-#endif
-
-template<typename _Tp>
-class commAllocator {
-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 commAllocator<_Tp1> other; };
-  commAllocator() throw() { }
-  commAllocator(const commAllocator&) throw() { }
-  template<typename _Tp1> commAllocator(const commAllocator<_Tp1>&) throw() { }
-  ~commAllocator() throw() { }
-  pointer       address(reference __x)       const { return &__x; }
-  size_type  max_size() const throw() { return size_t(-1) / sizeof(_Tp); }
-
-#ifdef GRID_COMMS_SHMEM
-  pointer allocate(size_type __n, const void* _p= 0)
-  {
-    size_type bytes = __n*sizeof(_Tp);
-
-    profilerAllocate(bytes);
-#ifdef CRAY
-    _Tp *ptr = (_Tp *) shmem_align(bytes,64);
-#else
-    _Tp *ptr = (_Tp *) shmem_align(64,bytes);
-#endif
-#ifdef PARANOID_SYMMETRIC_HEAP
-    static void * bcast;
-    static long  psync[_SHMEM_REDUCE_SYNC_SIZE];
-
-    bcast = (void *) ptr;
-    shmem_broadcast32((void *)&bcast,(void *)&bcast,sizeof(void *)/4,0,0,0,shmem_n_pes(),psync);
-
-    if ( bcast != ptr ) {
-      std::printf("inconsistent alloc pe %d %lx %lx \n",shmem_my_pe(),bcast,ptr);std::fflush(stdout);
-      //      BACKTRACEFILE();
-      exit(0);
-    }
-    assert( bcast == (void *) ptr);
-#endif 
-    return ptr;
-  }
-  void deallocate(pointer __p, size_type __n) { 
-    size_type bytes = __n*sizeof(_Tp);
-
-    profilerFree(bytes);
-    shmem_free((void *)__p);
-  }
-#else
-  pointer allocate(size_type __n, const void* _p= 0) 
-  {
-    size_type bytes = __n*sizeof(_Tp);
-    
-    profilerAllocate(bytes);
-#ifdef HAVE_MM_MALLOC_H
-    _Tp * ptr = (_Tp *) _mm_malloc(bytes, GRID_ALLOC_ALIGN);
-#else
-    _Tp * ptr = (_Tp *) memalign(GRID_ALLOC_ALIGN, bytes);
-#endif
-    uint8_t *cp = (uint8_t *)ptr;
-    if ( ptr ) { 
-    // One touch per 4k page, static OMP loop to catch same loop order
-#ifdef GRID_OMP
-#pragma omp parallel for schedule(static)
-#endif
-      for(size_type n=0;n<bytes;n+=4096){
-	cp[n]=0;
-      }
-    }
-    return ptr;
-  }
-  void deallocate(pointer __p, size_type __n) {
-    size_type bytes = __n*sizeof(_Tp);
-
-    profilerFree(bytes);
-#ifdef HAVE_MM_MALLOC_H
-    _mm_free((void *)__p); 
-#else
-    free((void *)__p);
-#endif
-  }
-#endif
-  void construct(pointer __p, const _Tp& __val) { };
-  void construct(pointer __p) { };
-  void destroy(pointer __p) { };
-};
-template<typename _Tp>  inline bool operator==(const commAllocator<_Tp>&, const commAllocator<_Tp>&){ return true; }
-template<typename _Tp>  inline bool operator!=(const commAllocator<_Tp>&, const commAllocator<_Tp>&){ return false; }
-
-////////////////////////////////////////////////////////////////////////////////
-// Template typedefs
-////////////////////////////////////////////////////////////////////////////////
-template<class T> using Vector     = std::vector<T,alignedAllocator<T> >;           
-template<class T> using commVector = std::vector<T,commAllocator<T> >;              
-template<class T> using Matrix     = std::vector<std::vector<T,alignedAllocator<T> > >;
-    
-}; // namespace 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 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 std::vector<int> &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 std::vector<int> &dimensions,
-		  const std::vector<int> &simd_layout,
-		  const std::vector<int> &processor_grid,
-		  const GridCartesian &parent) : GridBase(processor_grid,parent,dummy)
-    {
-      Init(dimensions,simd_layout,processor_grid);
-    }
-    GridCartesian(const std::vector<int> &dimensions,
-		  const std::vector<int> &simd_layout,
-		  const std::vector<int> &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 std::vector<int> &dimensions,
-		  const std::vector<int> &simd_layout,
-		  const std::vector<int> &processor_grid) : GridBase(processor_grid)
-    {
-      Init(dimensions,simd_layout,processor_grid);
-    }
-
-    virtual ~GridCartesian() = default;
-
-    void Init(const std::vector<int> &dimensions,
-	      const std::vector<int> &simd_layout,
-	      const std::vector<int> &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];
-      }
-    };
-
-};
-}
-#endif

Grid/cartesian/Cartesian_red_black.h

@@ -1,320 +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 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:
-    std::vector<int> _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 std::vector<int> &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)
-    {
-      std::vector<int> 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;
-      std::vector<int> 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 std::vector<int> &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;
-#if 0
-    ////////////////////////////////////////////////////////////
-    // Create redblack grid ;; deprecate these. Should not
-    // need direct creation of redblack without a full grid to base on
-    ////////////////////////////////////////////////////////////
-    GridRedBlackCartesian(const GridBase *base,
-			  const std::vector<int> &dimensions,
-			  const std::vector<int> &simd_layout,
-			  const std::vector<int> &processor_grid,
-			  const std::vector<int> &checker_dim_mask,
-			  int checker_dim
-			  ) :  GridBase(processor_grid,*base) 
-    {
-      Init(dimensions,simd_layout,processor_grid,checker_dim_mask,checker_dim);
-    }
-
-    ////////////////////////////////////////////////////////////
-    // Create redblack grid
-    ////////////////////////////////////////////////////////////
-    GridRedBlackCartesian(const GridBase *base,
-			  const std::vector<int> &dimensions,
-			  const std::vector<int> &simd_layout,
-			  const std::vector<int> &processor_grid) : GridBase(processor_grid,*base) 
-    {
-      std::vector<int> checker_dim_mask(dimensions.size(),1);
-      int checker_dim = 0;
-      Init(dimensions,simd_layout,processor_grid,checker_dim_mask,checker_dim);
-    }
-#endif
-
-    void Init(const std::vector<int> &dimensions,
-              const std::vector<int> &simd_layout,
-              const std::vector<int> &processor_grid,
-              const std::vector<int> &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(std::vector<int> &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(std::vector<int> &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;
-    }
-};
-}
-#endif

Grid/communicator/Communicator_base.cc

@@ -1,76 +0,0 @@
-    /*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/communicator/Communicator_none.cc
-
-    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 <Grid/GridCore.h>
-#include <fcntl.h>
-#include <unistd.h>
-#include <limits.h>
-#include <sys/mman.h>
-
-namespace Grid {
-
-///////////////////////////////////////////////////////////////
-// Info that is setup once and indept of cartesian layout
-///////////////////////////////////////////////////////////////
-CartesianCommunicator::CommunicatorPolicy_t  
-CartesianCommunicator::CommunicatorPolicy= CartesianCommunicator::CommunicatorPolicyConcurrent;
-int CartesianCommunicator::nCommThreads = -1;
-
-/////////////////////////////////
-// Grid information queries
-/////////////////////////////////
-int                      CartesianCommunicator::Dimensions(void)        { return _ndimension; };
-int                      CartesianCommunicator::IsBoss(void)            { return _processor==0; };
-int                      CartesianCommunicator::BossRank(void)          { return 0; };
-int                      CartesianCommunicator::ThisRank(void)          { return _processor; };
-const std::vector<int> & CartesianCommunicator::ThisProcessorCoor(void) { return _processor_coor; };
-const std::vector<int> & CartesianCommunicator::ProcessorGrid(void)     { return _processors; };
-int                      CartesianCommunicator::ProcessorCount(void)    { return _Nprocessors; };
-
-////////////////////////////////////////////////////////////////////////////////
-// very VERY rarely (Log, serial RNG) we need world without a grid
-////////////////////////////////////////////////////////////////////////////////
-
-void CartesianCommunicator::GlobalSum(ComplexF &c)
-{
-  GlobalSumVector((float *)&c,2);
-}
-void CartesianCommunicator::GlobalSumVector(ComplexF *c,int N)
-{
-  GlobalSumVector((float *)c,2*N);
-}
-void CartesianCommunicator::GlobalSum(ComplexD &c)
-{
-  GlobalSumVector((double *)&c,2);
-}
-void CartesianCommunicator::GlobalSumVector(ComplexD *c,int N)
-{
-  GlobalSumVector((double *)c,2*N);
-}
-  
-}
-

Grid/communicator/Communicator_base.h

@@ -1,207 +0,0 @@
-
-    /*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/communicator/Communicator_base.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_COMMUNICATOR_BASE_H
-#define GRID_COMMUNICATOR_BASE_H
-
-///////////////////////////////////
-// Processor layout information
-///////////////////////////////////
-#include <Grid/communicator/SharedMemory.h>
-
-namespace Grid {
-
-class CartesianCommunicator : public SharedMemory {
-
-public:    
-
-  ////////////////////////////////////////////
-  // Policies
-  ////////////////////////////////////////////
-  enum CommunicatorPolicy_t { CommunicatorPolicyConcurrent, CommunicatorPolicySequential };
-  static CommunicatorPolicy_t CommunicatorPolicy;
-  static void SetCommunicatorPolicy(CommunicatorPolicy_t policy ) { CommunicatorPolicy = policy; }
-  static int       nCommThreads;
-
-  ////////////////////////////////////////////
-  // Communicator should know nothing of the physics grid, only processor grid.
-  ////////////////////////////////////////////
-  int              _Nprocessors;     // How many in all
-  std::vector<int> _processors;      // Which dimensions get relayed out over processors lanes.
-  int              _processor;       // linear processor rank
-  std::vector<int> _processor_coor;  // linear processor coordinate
-  unsigned long    _ndimension;
-  static Grid_MPI_Comm      communicator_world;
-  Grid_MPI_Comm             communicator;
-  std::vector<Grid_MPI_Comm> communicator_halo;
-  
-  ////////////////////////////////////////////////
-  // Must call in Grid startup
-  ////////////////////////////////////////////////
-  static void Init(int *argc, char ***argv);
-
-  ////////////////////////////////////////////////
-  // Constructors to sub-divide a parent communicator
-  // and default to comm world
-  ////////////////////////////////////////////////
-  CartesianCommunicator(const std::vector<int> &processors,const CartesianCommunicator &parent,int &srank);
-  CartesianCommunicator(const std::vector<int> &pdimensions_in);
-  virtual ~CartesianCommunicator();
-
- private:
-
-  ////////////////////////////////////////////////
-  // Private initialise from an MPI communicator
-  // Can use after an MPI_Comm_split, but hidden from user so private
-  ////////////////////////////////////////////////
-  void InitFromMPICommunicator(const std::vector<int> &processors, Grid_MPI_Comm communicator_base);
-
- public:
-
-  
-  ////////////////////////////////////////////////////////////////////////////////////////
-  // Wraps MPI_Cart routines, or implements equivalent on other impls
-  ////////////////////////////////////////////////////////////////////////////////////////
-  void ShiftedRanks(int dim,int shift,int & source, int & dest);
-  int  RankFromProcessorCoor(std::vector<int> &coor);
-  void ProcessorCoorFromRank(int rank,std::vector<int> &coor);
-  
-  int                      Dimensions(void)        ;
-  int                      IsBoss(void)            ;
-  int                      BossRank(void)          ;
-  int                      ThisRank(void)          ;
-  const std::vector<int> & ThisProcessorCoor(void) ;
-  const std::vector<int> & ProcessorGrid(void)     ;
-  int                      ProcessorCount(void)    ;
-
-  ////////////////////////////////////////////////////////////////////////////////
-  // very VERY rarely (Log, serial RNG) we need world without a grid
-  ////////////////////////////////////////////////////////////////////////////////
-  static int  RankWorld(void) ;
-  static void BroadcastWorld(int root,void* data, int bytes);
-  
-  ////////////////////////////////////////////////////////////
-  // Reduction
-  ////////////////////////////////////////////////////////////
-  void GlobalSum(RealF &);
-  void GlobalSumVector(RealF *,int N);
-  void GlobalSum(RealD &);
-  void GlobalSumVector(RealD *,int N);
-  void GlobalSum(uint32_t &);
-  void GlobalSum(uint64_t &);
-  void GlobalSum(ComplexF &c);
-  void GlobalSumVector(ComplexF *c,int N);
-  void GlobalSum(ComplexD &c);
-  void GlobalSumVector(ComplexD *c,int N);
-  void GlobalXOR(uint32_t &);
-  void GlobalXOR(uint64_t &);
-  
-  template<class obj> void GlobalSum(obj &o){
-    typedef typename obj::scalar_type scalar_type;
-    int words = sizeof(obj)/sizeof(scalar_type);
-    scalar_type * ptr = (scalar_type *)& o;
-    GlobalSumVector(ptr,words);
-  }
-  
-  ////////////////////////////////////////////////////////////
-  // Face exchange, buffer swap in translational invariant way
-  ////////////////////////////////////////////////////////////
-  void SendToRecvFrom(void *xmit,
-		      int xmit_to_rank,
-		      void *recv,
-		      int recv_from_rank,
-		      int bytes);
-  
-  void SendRecvPacket(void *xmit,
-		      void *recv,
-		      int xmit_to_rank,
-		      int recv_from_rank,
-		      int bytes);
-  
-  void SendToRecvFromBegin(std::vector<CommsRequest_t> &list,
-			   void *xmit,
-			   int xmit_to_rank,
-			   void *recv,
-			   int recv_from_rank,
-			   int bytes);
-  
-  void SendToRecvFromComplete(std::vector<CommsRequest_t> &waitall);
-
-  double StencilSendToRecvFrom(void *xmit,
-			       int xmit_to_rank,
-			       void *recv,
-			       int recv_from_rank,
-			       int bytes,int dir);
-
-  double StencilSendToRecvFromBegin(std::vector<CommsRequest_t> &list,
-				    void *xmit,
-				    int xmit_to_rank,
-				    void *recv,
-				    int recv_from_rank,
-				    int bytes,int dir);
-  
-  
-  void StencilSendToRecvFromComplete(std::vector<CommsRequest_t> &waitall,int i);
-  void StencilBarrier(void);
-
-  ////////////////////////////////////////////////////////////
-  // Barrier
-  ////////////////////////////////////////////////////////////
-  void Barrier(void);
-  
-  ////////////////////////////////////////////////////////////
-  // Broadcast a buffer and composite larger
-  ////////////////////////////////////////////////////////////
-  void Broadcast(int root,void* data, int bytes);
-
-  ////////////////////////////////////////////////////////////
-  // All2All down one dimension
-  ////////////////////////////////////////////////////////////
-  template<class T> void AllToAll(int dim,std::vector<T> &in, std::vector<T> &out){
-    assert(dim>=0);
-    assert(dim<_ndimension);
-    assert(in.size()==out.size());
-    int numnode = _processors[dim];
-    uint64_t bytes=sizeof(T);
-    uint64_t words=in.size()/numnode;
-    assert(numnode * words == in.size());
-    assert(words < (1ULL<<31));
-    AllToAll(dim,(void *)&in[0],(void *)&out[0],words,bytes);
-  }
-  void AllToAll(int dim  ,void *in,void *out,uint64_t words,uint64_t bytes);
-  void AllToAll(void  *in,void *out,uint64_t words         ,uint64_t bytes);
-  
-  template<class obj> void Broadcast(int root,obj &data)
-    {
-      Broadcast(root,(void *)&data,sizeof(data));
-    };
-
-}; 
-}
-
-#endif

Grid/communicator/Communicator_mpi3.cc

@@ -1,509 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/communicator/Communicator_mpi.cc
-
-    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 <Grid/GridCore.h>
-#include <Grid/communicator/SharedMemory.h>
-
-namespace Grid {
-
-Grid_MPI_Comm       CartesianCommunicator::communicator_world;
-
-////////////////////////////////////////////
-// First initialise of comms system
-////////////////////////////////////////////
-void CartesianCommunicator::Init(int *argc, char ***argv) 
-{
-
-  int flag;
-  int provided;
-
-  MPI_Initialized(&flag); // needed to coexist with other libs apparently
-  if ( !flag ) {
-    MPI_Init_thread(argc,argv,MPI_THREAD_MULTIPLE,&provided);
-    //If only 1 comms thread we require any threading mode other than SINGLE, but for multiple comms threads we need MULTIPLE
-    if( (nCommThreads == 1 && provided == MPI_THREAD_SINGLE) ||
-        (nCommThreads > 1 && provided != MPI_THREAD_MULTIPLE) )
-      assert(0);
-  }
-
-  // Never clean up as done once.
-  MPI_Comm_dup (MPI_COMM_WORLD,&communicator_world);
-
-  Grid_quiesce_nodes();
-  GlobalSharedMemory::Init(communicator_world);
-  GlobalSharedMemory::SharedMemoryAllocate(
-		   GlobalSharedMemory::MAX_MPI_SHM_BYTES,
-		   GlobalSharedMemory::Hugepages);
-  Grid_unquiesce_nodes();
-}
-
-///////////////////////////////////////////////////////////////////////////
-// Use cartesian communicators now even in MPI3
-///////////////////////////////////////////////////////////////////////////
-void CartesianCommunicator::ShiftedRanks(int dim,int shift,int &source,int &dest)
-{
-  int ierr=MPI_Cart_shift(communicator,dim,shift,&source,&dest);
-  assert(ierr==0);
-}
-int CartesianCommunicator::RankFromProcessorCoor(std::vector<int> &coor)
-{
-  int rank;
-  int ierr=MPI_Cart_rank  (communicator, &coor[0], &rank);
-  assert(ierr==0);
-  return rank;
-}
-void  CartesianCommunicator::ProcessorCoorFromRank(int rank, std::vector<int> &coor)
-{
-  coor.resize(_ndimension);
-  int ierr=MPI_Cart_coords  (communicator, rank, _ndimension,&coor[0]);
-  assert(ierr==0);
-}
-
-////////////////////////////////////////////////////////////////////////////////////////////////////////
-// Initialises from communicator_world
-////////////////////////////////////////////////////////////////////////////////////////////////////////
-CartesianCommunicator::CartesianCommunicator(const std::vector<int> &processors) 
-{
-  MPI_Comm optimal_comm;
-  ////////////////////////////////////////////////////
-  // Remap using the shared memory optimising routine
-  // The remap creates a comm which must be freed
-  ////////////////////////////////////////////////////
-  GlobalSharedMemory::OptimalCommunicator    (processors,optimal_comm);
-  InitFromMPICommunicator(processors,optimal_comm);
-  SetCommunicator(optimal_comm);
-  ///////////////////////////////////////////////////
-  // Free the temp communicator
-  ///////////////////////////////////////////////////
-  MPI_Comm_free(&optimal_comm);
-}
-
-//////////////////////////////////
-// Try to subdivide communicator
-//////////////////////////////////
-CartesianCommunicator::CartesianCommunicator(const std::vector<int> &processors,const CartesianCommunicator &parent,int &srank)    
-{
-  _ndimension = processors.size();  assert(_ndimension>=1);
-  int parent_ndimension = parent._ndimension; assert(_ndimension >= parent._ndimension);
-  std::vector<int> parent_processor_coor(_ndimension,0);
-  std::vector<int> parent_processors    (_ndimension,1);
-
-  // Can make 5d grid from 4d etc...
-  int pad = _ndimension-parent_ndimension;
-  for(int d=0;d<parent_ndimension;d++){
-    parent_processor_coor[pad+d]=parent._processor_coor[d];
-    parent_processors    [pad+d]=parent._processors[d];
-  }
-
-  //////////////////////////////////////////////////////////////////////////////////////////////////////
-  // split the communicator
-  //////////////////////////////////////////////////////////////////////////////////////////////////////
-  //  int Nparent = parent._processors ; 
-  int Nparent;
-  MPI_Comm_size(parent.communicator,&Nparent);
-
-  int childsize=1;
-  for(int d=0;d<processors.size();d++) {
-    childsize *= processors[d];
-  }
-  int Nchild = Nparent/childsize;
-  assert (childsize * Nchild == Nparent);
-
-  std::vector<int> ccoor(_ndimension); // coor within subcommunicator
-  std::vector<int> scoor(_ndimension); // coor of split within parent
-  std::vector<int> ssize(_ndimension); // coor of split within parent
-
-  for(int d=0;d<_ndimension;d++){
-    ccoor[d] = parent_processor_coor[d] % processors[d];
-    scoor[d] = parent_processor_coor[d] / processors[d];
-    ssize[d] = parent_processors[d]     / processors[d];
-  }
-
-  // rank within subcomm ; srank is rank of subcomm within blocks of subcomms
-  int crank;  
-  // Mpi uses the reverse Lexico convention to us; so reversed routines called
-  Lexicographic::IndexFromCoorReversed(ccoor,crank,processors); // processors is the split grid dimensions
-  Lexicographic::IndexFromCoorReversed(scoor,srank,ssize);      // ssize is the number of split grids
-
-  MPI_Comm comm_split;
-  if ( Nchild > 1 ) { 
-
-    if(0){
-      std::cout << GridLogMessage<<"Child communicator of "<< std::hex << parent.communicator << std::dec<<std::endl;
-      std::cout << GridLogMessage<<" parent grid["<< parent._ndimension<<"]    ";
-      for(int d=0;d<parent._ndimension;d++)  std::cout << parent._processors[d] << " ";
-      std::cout<<std::endl;
-      
-      std::cout << GridLogMessage<<" child grid["<< _ndimension <<"]    ";
-      for(int d=0;d<processors.size();d++)  std::cout << processors[d] << " ";
-      std::cout<<std::endl;
-      
-      std::cout << GridLogMessage<<" old rank "<< parent._processor<<" coor ["<< parent._ndimension <<"]    ";
-      for(int d=0;d<parent._ndimension;d++)  std::cout << parent._processor_coor[d] << " ";
-      std::cout<<std::endl;
-      
-      std::cout << GridLogMessage<<" new split "<< srank<<" scoor ["<< _ndimension <<"]    ";
-      for(int d=0;d<processors.size();d++)  std::cout << scoor[d] << " ";
-      std::cout<<std::endl;
-      
-      std::cout << GridLogMessage<<" new rank "<< crank<<" coor ["<< _ndimension <<"]    ";
-      for(int d=0;d<processors.size();d++)  std::cout << ccoor[d] << " ";
-      std::cout<<std::endl;
-
-      //////////////////////////////////////////////////////////////////////////////////////////////////////
-      // Declare victory
-      //////////////////////////////////////////////////////////////////////////////////////////////////////
-      std::cout << GridLogMessage<<"Divided communicator "<< parent._Nprocessors<<" into "
-		<< Nchild <<" communicators with " << childsize << " ranks"<<std::endl;
-      std::cout << " Split communicator " <<comm_split <<std::endl;
-    }
-
-    ////////////////////////////////////////////////////////////////
-    // Split the communicator
-    ////////////////////////////////////////////////////////////////
-    int ierr= MPI_Comm_split(parent.communicator,srank,crank,&comm_split);
-    assert(ierr==0);
-
-  } else {
-    srank = 0;
-    int ierr = MPI_Comm_dup (parent.communicator,&comm_split);
-    assert(ierr==0);
-  }
-
-  //////////////////////////////////////////////////////////////////////////////////////////////////////
-  // Set up from the new split communicator
-  //////////////////////////////////////////////////////////////////////////////////////////////////////
-  InitFromMPICommunicator(processors,comm_split);
-
-  //////////////////////////////////////////////////////////////////////////////////////////////////////
-  // Take the right SHM buffers
-  //////////////////////////////////////////////////////////////////////////////////////////////////////
-  SetCommunicator(comm_split);
-  
-  ///////////////////////////////////////////////
-  // Free the temp communicator 
-  ///////////////////////////////////////////////
-  MPI_Comm_free(&comm_split);
-
-  if(0){ 
-    std::cout << " ndim " <<_ndimension<<" " << parent._ndimension << std::endl;
-    for(int d=0;d<processors.size();d++){
-      std::cout << d<< " " << _processor_coor[d] <<" " <<  ccoor[d]<<std::endl;
-    }
-  }
-  for(int d=0;d<processors.size();d++){
-    assert(_processor_coor[d] == ccoor[d] );
-  }
-}
-
-void CartesianCommunicator::InitFromMPICommunicator(const std::vector<int> &processors, MPI_Comm communicator_base)
-{
-  ////////////////////////////////////////////////////
-  // Creates communicator, and the communicator_halo
-  ////////////////////////////////////////////////////
-  _ndimension = processors.size();
-  _processor_coor.resize(_ndimension);
-
-  /////////////////////////////////
-  // Count the requested nodes
-  /////////////////////////////////
-  _Nprocessors=1;
-  _processors = processors;
-  for(int i=0;i<_ndimension;i++){
-    _Nprocessors*=_processors[i];
-  }
-
-  std::vector<int> periodic(_ndimension,1);
-  MPI_Cart_create(communicator_base, _ndimension,&_processors[0],&periodic[0],0,&communicator);
-  MPI_Comm_rank(communicator,&_processor);
-  MPI_Cart_coords(communicator,_processor,_ndimension,&_processor_coor[0]);
-
-  if ( 0 && (communicator_base != communicator_world) ) {
-    std::cout << "InitFromMPICommunicator Cartesian communicator created with a non-world communicator"<<std::endl;
-    std::cout << " new communicator rank "<<_processor<< " coor ["<<_ndimension<<"] ";
-    for(int d=0;d<_processors.size();d++){
-      std::cout << _processor_coor[d]<<" ";
-    }
-    std::cout << std::endl;
-  }
-
-  int Size;
-  MPI_Comm_size(communicator,&Size);
-
-  communicator_halo.resize (2*_ndimension);
-  for(int i=0;i<_ndimension*2;i++){
-    MPI_Comm_dup(communicator,&communicator_halo[i]);
-  }
-  assert(Size==_Nprocessors);
-}
-
-CartesianCommunicator::~CartesianCommunicator()
-{
-  int MPI_is_finalised;
-  MPI_Finalized(&MPI_is_finalised);
-  if (communicator && !MPI_is_finalised) {
-    MPI_Comm_free(&communicator);
-    for(int i=0;i<communicator_halo.size();i++){
-      MPI_Comm_free(&communicator_halo[i]);
-    }
-  }  
-}
-void CartesianCommunicator::GlobalSum(uint32_t &u){
-  int ierr=MPI_Allreduce(MPI_IN_PLACE,&u,1,MPI_UINT32_T,MPI_SUM,communicator);
-  assert(ierr==0);
-}
-void CartesianCommunicator::GlobalSum(uint64_t &u){
-  int ierr=MPI_Allreduce(MPI_IN_PLACE,&u,1,MPI_UINT64_T,MPI_SUM,communicator);
-  assert(ierr==0);
-}
-void CartesianCommunicator::GlobalXOR(uint32_t &u){
-  int ierr=MPI_Allreduce(MPI_IN_PLACE,&u,1,MPI_UINT32_T,MPI_BXOR,communicator);
-  assert(ierr==0);
-}
-void CartesianCommunicator::GlobalXOR(uint64_t &u){
-  int ierr=MPI_Allreduce(MPI_IN_PLACE,&u,1,MPI_UINT64_T,MPI_BXOR,communicator);
-  assert(ierr==0);
-}
-void CartesianCommunicator::GlobalSum(float &f){
-  int ierr=MPI_Allreduce(MPI_IN_PLACE,&f,1,MPI_FLOAT,MPI_SUM,communicator);
-  assert(ierr==0);
-}
-void CartesianCommunicator::GlobalSumVector(float *f,int N)
-{
-  int ierr=MPI_Allreduce(MPI_IN_PLACE,f,N,MPI_FLOAT,MPI_SUM,communicator);
-  assert(ierr==0);
-}
-void CartesianCommunicator::GlobalSum(double &d)
-{
-  int ierr = MPI_Allreduce(MPI_IN_PLACE,&d,1,MPI_DOUBLE,MPI_SUM,communicator);
-  assert(ierr==0);
-}
-void CartesianCommunicator::GlobalSumVector(double *d,int N)
-{
-  int ierr = MPI_Allreduce(MPI_IN_PLACE,d,N,MPI_DOUBLE,MPI_SUM,communicator);
-  assert(ierr==0);
-}
-// Basic Halo comms primitive
-void CartesianCommunicator::SendToRecvFrom(void *xmit,
-					   int dest,
-					   void *recv,
-					   int from,
-					   int bytes)
-{
-  std::vector<CommsRequest_t> reqs(0);
-  //    unsigned long  xcrc = crc32(0L, Z_NULL, 0);
-  //    unsigned long  rcrc = crc32(0L, Z_NULL, 0);
-  //    xcrc = crc32(xcrc,(unsigned char *)xmit,bytes);
-  SendToRecvFromBegin(reqs,xmit,dest,recv,from,bytes);
-  SendToRecvFromComplete(reqs);
-  //    rcrc = crc32(rcrc,(unsigned char *)recv,bytes);
-  //    printf("proc %d SendToRecvFrom %d bytes %lx %lx\n",_processor,bytes,xcrc,rcrc);
-}
-void CartesianCommunicator::SendRecvPacket(void *xmit,
-					   void *recv,
-					   int sender,
-					   int receiver,
-					   int bytes)
-{
-  MPI_Status stat;
-  assert(sender != receiver);
-  int tag = sender;
-  if ( _processor == sender ) {
-    MPI_Send(xmit, bytes, MPI_CHAR,receiver,tag,communicator);
-  }
-  if ( _processor == receiver ) { 
-    MPI_Recv(recv, bytes, MPI_CHAR,sender,tag,communicator,&stat);
-  }
-}
-// Basic Halo comms primitive
-void CartesianCommunicator::SendToRecvFromBegin(std::vector<CommsRequest_t> &list,
-						void *xmit,
-						int dest,
-						void *recv,
-						int from,
-						int bytes)
-{
-  int myrank = _processor;
-  int ierr;
-
-  if ( CommunicatorPolicy == CommunicatorPolicyConcurrent ) { 
-    MPI_Request xrq;
-    MPI_Request rrq;
-
-    ierr =MPI_Irecv(recv, bytes, MPI_CHAR,from,from,communicator,&rrq);
-    ierr|=MPI_Isend(xmit, bytes, MPI_CHAR,dest,_processor,communicator,&xrq);
-    
-    assert(ierr==0);
-    list.push_back(xrq);
-    list.push_back(rrq);
-  } else { 
-    // Give the CPU to MPI immediately; can use threads to overlap optionally
-    ierr=MPI_Sendrecv(xmit,bytes,MPI_CHAR,dest,myrank,
-		      recv,bytes,MPI_CHAR,from, from,
-		      communicator,MPI_STATUS_IGNORE);
-    assert(ierr==0);
-  }
-}
-
-double CartesianCommunicator::StencilSendToRecvFrom( void *xmit,
-						     int dest,
-						     void *recv,
-						     int from,
-						     int bytes,int dir)
-{
-  std::vector<CommsRequest_t> list;
-  double offbytes = StencilSendToRecvFromBegin(list,xmit,dest,recv,from,bytes,dir);
-  StencilSendToRecvFromComplete(list,dir);
-  return offbytes;
-}
-
-double CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsRequest_t> &list,
-							 void *xmit,
-							 int dest,
-							 void *recv,
-							 int from,
-							 int bytes,int dir)
-{
-  int ncomm  =communicator_halo.size(); 
-  int commdir=dir%ncomm;
-
-  MPI_Request xrq;
-  MPI_Request rrq;
-
-  int ierr;
-  int gdest = ShmRanks[dest];
-  int gfrom = ShmRanks[from];
-  int gme   = ShmRanks[_processor];
-
-  assert(dest != _processor);
-  assert(from != _processor);
-  assert(gme  == ShmRank);
-  double off_node_bytes=0.0;
-
-  if ( gfrom ==MPI_UNDEFINED) {
-    ierr=MPI_Irecv(recv, bytes, MPI_CHAR,from,from,communicator_halo[commdir],&rrq);
-    assert(ierr==0);
-    list.push_back(rrq);
-    off_node_bytes+=bytes;
-  }
-
-  if ( gdest == MPI_UNDEFINED ) {
-    ierr =MPI_Isend(xmit, bytes, MPI_CHAR,dest,_processor,communicator_halo[commdir],&xrq);
-    assert(ierr==0);
-    list.push_back(xrq);
-    off_node_bytes+=bytes;
-  }
-
-  if ( CommunicatorPolicy == CommunicatorPolicySequential ) { 
-    this->StencilSendToRecvFromComplete(list,dir);
-  }
-
-  return off_node_bytes;
-}
-void CartesianCommunicator::StencilSendToRecvFromComplete(std::vector<CommsRequest_t> &waitall,int dir)
-{
-  SendToRecvFromComplete(waitall);
-}
-void CartesianCommunicator::StencilBarrier(void)
-{
-  MPI_Barrier  (ShmComm);
-}
-void CartesianCommunicator::SendToRecvFromComplete(std::vector<CommsRequest_t> &list)
-{
-  int nreq=list.size();
-
-  if (nreq==0) return;
-
-  std::vector<MPI_Status> status(nreq);
-  int ierr = MPI_Waitall(nreq,&list[0],&status[0]);
-  assert(ierr==0);
-  list.resize(0);
-}
-void CartesianCommunicator::Barrier(void)
-{
-  int ierr = MPI_Barrier(communicator);
-  assert(ierr==0);
-}
-void CartesianCommunicator::Broadcast(int root,void* data, int bytes)
-{
-  int ierr=MPI_Bcast(data,
-		     bytes,
-		     MPI_BYTE,
-		     root,
-		     communicator);
-  assert(ierr==0);
-}
-int CartesianCommunicator::RankWorld(void){ 
-  int r; 
-  MPI_Comm_rank(communicator_world,&r);
-  return r;
-}
-void CartesianCommunicator::BroadcastWorld(int root,void* data, int bytes)
-{
-  int ierr= MPI_Bcast(data,
-		      bytes,
-		      MPI_BYTE,
-		      root,
-		      communicator_world);
-  assert(ierr==0);
-}
-
-void CartesianCommunicator::AllToAll(int dim,void  *in,void *out,uint64_t words,uint64_t bytes)
-{
-  std::vector<int> row(_ndimension,1);
-  assert(dim>=0 && dim<_ndimension);
-
-  //  Split the communicator
-  row[dim] = _processors[dim];
-
-  int me;
-  CartesianCommunicator Comm(row,*this,me);
-  Comm.AllToAll(in,out,words,bytes);
-}
-void CartesianCommunicator::AllToAll(void  *in,void *out,uint64_t words,uint64_t bytes)
-{
-  // MPI is a pain and uses "int" arguments
-  // 64*64*64*128*16 == 500Million elements of data.
-  // When 24*4 bytes multiples get 50x 10^9 >>> 2x10^9 Y2K bug.
-  // (Turns up on 32^3 x 64 Gparity too)
-  MPI_Datatype object;
-  int iwords; 
-  int ibytes;
-  iwords = words;
-  ibytes = bytes;
-  assert(words == iwords); // safe to cast to int ?
-  assert(bytes == ibytes); // safe to cast to int ?
-  MPI_Type_contiguous(ibytes,MPI_BYTE,&object);
-  MPI_Type_commit(&object);
-  MPI_Alltoall(in,iwords,object,out,iwords,object,communicator);
-  MPI_Type_free(&object);
-}
-
-
-
-}
-

Grid/communicator/SharedMemory.cc

@@ -1,92 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/communicator/SharedMemory.cc
-
-    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 <Grid/GridCore.h>
-
-namespace Grid { 
-
-// static data
-
-uint64_t            GlobalSharedMemory::MAX_MPI_SHM_BYTES   = 1024LL*1024LL*1024LL; 
-int                 GlobalSharedMemory::Hugepages = 0;
-int                 GlobalSharedMemory::_ShmSetup;
-int                 GlobalSharedMemory::_ShmAlloc;
-uint64_t            GlobalSharedMemory::_ShmAllocBytes;
-
-std::vector<void *> GlobalSharedMemory::WorldShmCommBufs;
-
-Grid_MPI_Comm       GlobalSharedMemory::WorldShmComm;
-int                 GlobalSharedMemory::WorldShmRank;
-int                 GlobalSharedMemory::WorldShmSize;
-std::vector<int>    GlobalSharedMemory::WorldShmRanks;
-
-Grid_MPI_Comm       GlobalSharedMemory::WorldComm;
-int                 GlobalSharedMemory::WorldSize;
-int                 GlobalSharedMemory::WorldRank;
-
-int                 GlobalSharedMemory::WorldNodes;
-int                 GlobalSharedMemory::WorldNode;
-
-void GlobalSharedMemory::SharedMemoryFree(void)
-{
-  assert(_ShmAlloc);
-  assert(_ShmAllocBytes>0);
-  for(int r=0;r<WorldShmSize;r++){
-    munmap(WorldShmCommBufs[r],_ShmAllocBytes);
-  }
-  _ShmAlloc = 0;
-  _ShmAllocBytes = 0;
-}
-/////////////////////////////////
-// Alloc, free shmem region
-/////////////////////////////////
-void *SharedMemory::ShmBufferMalloc(size_t bytes){
-  //  bytes = (bytes+sizeof(vRealD))&(~(sizeof(vRealD)-1));// align up bytes
-  void *ptr = (void *)heap_top;
-  heap_top  += bytes;
-  heap_bytes+= bytes;
-  if (heap_bytes >= heap_size) {
-    std::cout<< " ShmBufferMalloc exceeded shared heap size -- try increasing with --shm <MB> flag" <<std::endl;
-    std::cout<< " Parameter specified in units of MB (megabytes) " <<std::endl;
-    std::cout<< " Current value is " << (heap_size/(1024*1024)) <<std::endl;
-    assert(heap_bytes<heap_size);
-  }
-  return ptr;
-}
-void SharedMemory::ShmBufferFreeAll(void) { 
-  heap_top  =(size_t)ShmBufferSelf();
-  heap_bytes=0;
-}
-void *SharedMemory::ShmBufferSelf(void)
-{
-  return ShmCommBufs[ShmRank];
-}
-
-
-
-}

Grid/communicator/SharedMemory.h

@@ -1,167 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/communicator/SharedMemory.cc
-
-    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 */
-
-
-// TODO
-// 1) move includes into SharedMemory.cc
-//
-// 2) split shared memory into a) optimal communicator creation from comm world
-// 
-//                             b) shared memory buffers container
-//                                -- static globally shared; init once
-//                                -- per instance set of buffers.
-//                                   
-
-#pragma once 
-
-#include <Grid/GridCore.h>
-
-#if defined (GRID_COMMS_MPI3) 
-#include <mpi.h>
-#endif 
-#include <semaphore.h>
-#include <fcntl.h>
-#include <unistd.h>
-#include <limits.h>
-#include <sys/types.h>
-#include <sys/ipc.h>
-#include <sys/shm.h>
-#include <sys/mman.h>
-#include <zlib.h>
-#ifdef HAVE_NUMAIF_H
-#include <numaif.h>
-#endif
-
-namespace Grid {
-
-#if defined (GRID_COMMS_MPI3) 
-  typedef MPI_Comm    Grid_MPI_Comm;
-  typedef MPI_Request CommsRequest_t;
-#else 
-  typedef int CommsRequest_t;
-  typedef int Grid_MPI_Comm;
-#endif
-
-class GlobalSharedMemory {
- private:
-  static const int     MAXLOG2RANKSPERNODE = 16;            
-
-  // Init once lock on the buffer allocation
-  static int      _ShmSetup;
-  static int      _ShmAlloc;
-  static uint64_t _ShmAllocBytes;
-
- public:
-  static int      ShmSetup(void)      { return _ShmSetup; }
-  static int      ShmAlloc(void)      { return _ShmAlloc; }
-  static uint64_t ShmAllocBytes(void) { return _ShmAllocBytes; }
-  static uint64_t      MAX_MPI_SHM_BYTES;
-  static int           Hugepages;
-
-  static std::vector<void *> WorldShmCommBufs;
-
-  static Grid_MPI_Comm WorldComm;
-  static int           WorldRank;
-  static int           WorldSize;
-
-  static Grid_MPI_Comm WorldShmComm;
-  static int           WorldShmRank;
-  static int           WorldShmSize;
-
-  static int           WorldNodes;
-  static int           WorldNode;
-
-  static std::vector<int>  WorldShmRanks;
-
-  //////////////////////////////////////////////////////////////////////////////////////
-  // Create an optimal reordered communicator that makes MPI_Cart_create get it right
-  //////////////////////////////////////////////////////////////////////////////////////
-  static void Init(Grid_MPI_Comm comm); // Typically MPI_COMM_WORLD
-  static void OptimalCommunicator(const std::vector<int> &processors,Grid_MPI_Comm & optimal_comm);  // Turns MPI_COMM_WORLD into right layout for Cartesian
-  static void OptimalCommunicatorHypercube(const std::vector<int> &processors,Grid_MPI_Comm & optimal_comm);  // Turns MPI_COMM_WORLD into right layout for Cartesian
-  static void OptimalCommunicatorSharedMemory(const std::vector<int> &processors,Grid_MPI_Comm & optimal_comm);  // Turns MPI_COMM_WORLD into right layout for Cartesian
-  ///////////////////////////////////////////////////
-  // Provide shared memory facilities off comm world
-  ///////////////////////////////////////////////////
-  static void SharedMemoryAllocate(uint64_t bytes, int flags);
-  static void SharedMemoryFree(void);
-
-};
-
-//////////////////////////////
-// one per communicator
-//////////////////////////////
-class SharedMemory 
-{
- private:
-  static const int     MAXLOG2RANKSPERNODE = 16;            
-
-  size_t heap_top;
-  size_t heap_bytes;
-  size_t heap_size;
-
- protected:
-
-  Grid_MPI_Comm    ShmComm; // for barriers
-  int    ShmRank; 
-  int    ShmSize;
-  std::vector<void *> ShmCommBufs;
-  std::vector<int>    ShmRanks;// Mapping comm ranks to Shm ranks
-
- public:
-  SharedMemory() {};
-  ~SharedMemory();
-  ///////////////////////////////////////////////////////////////////////////////////////
-  // set the buffers & sizes
-  ///////////////////////////////////////////////////////////////////////////////////////
-  void SetCommunicator(Grid_MPI_Comm comm);
-
-  ////////////////////////////////////////////////////////////////////////
-  // For this instance ; disjoint buffer sets between splits if split grid
-  ////////////////////////////////////////////////////////////////////////
-  void ShmBarrier(void); 
-
-  ///////////////////////////////////////////////////
-  // Call on any instance
-  ///////////////////////////////////////////////////
-  void SharedMemoryTest(void);
-  void *ShmBufferSelf(void);
-  void *ShmBuffer    (int rank);
-  void *ShmBufferTranslate(int rank,void * local_p);
-  void *ShmBufferMalloc(size_t bytes);
-  void  ShmBufferFreeAll(void) ;
-  
-  //////////////////////////////////////////////////////////////////////////
-  // Make info on Nodes & ranks and Shared memory available
-  //////////////////////////////////////////////////////////////////////////
-  int NodeCount(void) { return GlobalSharedMemory::WorldNodes;};
-  int RankCount(void) { return GlobalSharedMemory::WorldSize;};
-
-};
-
-}

Grid/communicator/SharedMemoryMPI.cc

@@ -1,667 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/communicator/SharedMemory.cc
-
-    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 <Grid/GridCore.h>
-#include <pwd.h>
-
-namespace Grid { 
-
-/*Construct from an MPI communicator*/
-void GlobalSharedMemory::Init(Grid_MPI_Comm comm)
-{
-  assert(_ShmSetup==0);
-  WorldComm = comm;
-  MPI_Comm_rank(WorldComm,&WorldRank);
-  MPI_Comm_size(WorldComm,&WorldSize);
-  // WorldComm, WorldSize, WorldRank
-
-  /////////////////////////////////////////////////////////////////////
-  // Split into groups that can share memory
-  /////////////////////////////////////////////////////////////////////
-  MPI_Comm_split_type(comm, MPI_COMM_TYPE_SHARED, 0, MPI_INFO_NULL,&WorldShmComm);
-  MPI_Comm_rank(WorldShmComm     ,&WorldShmRank);
-  MPI_Comm_size(WorldShmComm     ,&WorldShmSize);
-  // WorldShmComm, WorldShmSize, WorldShmRank
-
-  // WorldNodes
-  WorldNodes = WorldSize/WorldShmSize;
-  assert( (WorldNodes * WorldShmSize) == WorldSize );
-
-  // FIXME: Check all WorldShmSize are the same ?
-
-  /////////////////////////////////////////////////////////////////////
-  // find world ranks in our SHM group (i.e. which ranks are on our node)
-  /////////////////////////////////////////////////////////////////////
-  MPI_Group WorldGroup, ShmGroup;
-  MPI_Comm_group (WorldComm, &WorldGroup); 
-  MPI_Comm_group (WorldShmComm, &ShmGroup);
-
-  std::vector<int> world_ranks(WorldSize);   for(int r=0;r<WorldSize;r++) world_ranks[r]=r;
-
-  WorldShmRanks.resize(WorldSize); 
-  MPI_Group_translate_ranks (WorldGroup,WorldSize,&world_ranks[0],ShmGroup, &WorldShmRanks[0]); 
-
-  ///////////////////////////////////////////////////////////////////
-  // Identify who is in my group and nominate the leader
-  ///////////////////////////////////////////////////////////////////
-  int g=0;
-  std::vector<int> MyGroup;
-  MyGroup.resize(WorldShmSize);
-  for(int rank=0;rank<WorldSize;rank++){
-    if(WorldShmRanks[rank]!=MPI_UNDEFINED){
-      assert(g<WorldShmSize);
-      MyGroup[g++] = rank;
-    }
-  }
-  
-  std::sort(MyGroup.begin(),MyGroup.end(),std::less<int>());
-  int myleader = MyGroup[0];
-  
-  std::vector<int> leaders_1hot(WorldSize,0);
-  std::vector<int> leaders_group(WorldNodes,0);
-  leaders_1hot [ myleader ] = 1;
-    
-  ///////////////////////////////////////////////////////////////////
-  // global sum leaders over comm world
-  ///////////////////////////////////////////////////////////////////
-  int ierr=MPI_Allreduce(MPI_IN_PLACE,&leaders_1hot[0],WorldSize,MPI_INT,MPI_SUM,WorldComm);
-  assert(ierr==0);
-
-  ///////////////////////////////////////////////////////////////////
-  // find the group leaders world rank
-  ///////////////////////////////////////////////////////////////////
-  int group=0;
-  for(int l=0;l<WorldSize;l++){
-    if(leaders_1hot[l]){
-      leaders_group[group++] = l;
-    }
-  }
-
-  ///////////////////////////////////////////////////////////////////
-  // Identify the node of the group in which I (and my leader) live
-  ///////////////////////////////////////////////////////////////////
-  WorldNode=-1;
-  for(int g=0;g<WorldNodes;g++){
-    if (myleader == leaders_group[g]){
-      WorldNode=g;
-    }
-  }
-  assert(WorldNode!=-1);
-  _ShmSetup=1;
-}
-// Gray encode support 
-int BinaryToGray (int  binary) {
-  int gray = (binary>>1)^binary;
-  return gray;
-}
-int Log2Size(int TwoToPower,int MAXLOG2)
-{
-  int log2size = -1;
-  for(int i=0;i<=MAXLOG2;i++){
-    if ( (0x1<<i) == TwoToPower ) {
-      log2size = i;
-      break;
-    }
-  }
-  return log2size;
-}
-void GlobalSharedMemory::OptimalCommunicator(const std::vector<int> &processors,Grid_MPI_Comm & optimal_comm)
-{
-  //////////////////////////////////////////////////////////////////////////////
-  // Look and see if it looks like an HPE 8600 based on hostname conventions
-  //////////////////////////////////////////////////////////////////////////////
-  const int namelen = _POSIX_HOST_NAME_MAX;
-  char name[namelen];
-  int R;
-  int I;
-  int N;
-  gethostname(name,namelen);
-  int nscan = sscanf(name,"r%di%dn%d",&R,&I,&N) ;
-
-  if(nscan==3) OptimalCommunicatorHypercube(processors,optimal_comm);
-  else         OptimalCommunicatorSharedMemory(processors,optimal_comm);
-}
-void GlobalSharedMemory::OptimalCommunicatorHypercube(const std::vector<int> &processors,Grid_MPI_Comm & optimal_comm)
-{
-  ////////////////////////////////////////////////////////////////
-  // Assert power of two shm_size.
-  ////////////////////////////////////////////////////////////////
-  int log2size = Log2Size(WorldShmSize,MAXLOG2RANKSPERNODE);
-  assert(log2size != -1);
-
-  ////////////////////////////////////////////////////////////////
-  // Identify the hypercube coordinate of this node using hostname
-  ////////////////////////////////////////////////////////////////
-  // n runs 0...7 9...16 18...25 27...34     (8*4)  5 bits
-  // i runs 0..7                                    3 bits
-  // r runs 0..3                                    2 bits
-  // 2^10 = 1024 nodes
-  const int maxhdim = 10; 
-  std::vector<int> HyperCubeCoords(maxhdim,0);
-  std::vector<int> RootHyperCubeCoords(maxhdim,0);
-  int R;
-  int I;
-  int N;
-  const int namelen = _POSIX_HOST_NAME_MAX;
-  char name[namelen];
-
-  // Parse ICE-XA hostname to get hypercube location
-  gethostname(name,namelen);
-  int nscan = sscanf(name,"r%di%dn%d",&R,&I,&N) ;
-  assert(nscan==3);
-
-  int nlo = N%9;
-  int nhi = N/9;
-  uint32_t hypercoor = (R<<8)|(I<<5)|(nhi<<3)|nlo ;
-  uint32_t rootcoor  = hypercoor;
-
-  //////////////////////////////////////////////////////////////////
-  // Print debug info
-  //////////////////////////////////////////////////////////////////
-  for(int d=0;d<maxhdim;d++){
-    HyperCubeCoords[d] = (hypercoor>>d)&0x1;
-  }
-
-  std::string hname(name);
-  std::cout << "hostname "<<hname<<std::endl;
-  std::cout << "R " << R << " I " << I << " N "<< N
-            << " hypercoor 0x"<<std::hex<<hypercoor<<std::dec<<std::endl;
-
-  //////////////////////////////////////////////////////////////////
-  // broadcast node 0's base coordinate for this partition.
-  //////////////////////////////////////////////////////////////////
-  MPI_Bcast(&rootcoor, sizeof(rootcoor), MPI_BYTE, 0, WorldComm); 
-  hypercoor=hypercoor-rootcoor;
-  assert(hypercoor<WorldSize);
-  assert(hypercoor>=0);
-
-  //////////////////////////////////////
-  // Printing
-  //////////////////////////////////////
-  for(int d=0;d<maxhdim;d++){
-    HyperCubeCoords[d] = (hypercoor>>d)&0x1;
-  }
-
-  ////////////////////////////////////////////////////////////////
-  // Identify subblock of ranks on node spreading across dims
-  // in a maximally symmetrical way
-  ////////////////////////////////////////////////////////////////
-  int ndimension              = processors.size();
-  std::vector<int> processor_coor(ndimension);
-  std::vector<int> WorldDims = processors;   std::vector<int> ShmDims  (ndimension,1);  std::vector<int> NodeDims (ndimension);
-  std::vector<int> ShmCoor  (ndimension);    std::vector<int> NodeCoor (ndimension);    std::vector<int> WorldCoor(ndimension);
-  std::vector<int> HyperCoor(ndimension);
-  int dim = 0;
-  for(int l2=0;l2<log2size;l2++){
-    while ( (WorldDims[dim] / ShmDims[dim]) <= 1 ) dim=(dim+1)%ndimension;
-    ShmDims[dim]*=2;
-    dim=(dim+1)%ndimension;
-  }
-
-  ////////////////////////////////////////////////////////////////
-  // Establish torus of processes and nodes with sub-blockings
-  ////////////////////////////////////////////////////////////////
-  for(int d=0;d<ndimension;d++){
-    NodeDims[d] = WorldDims[d]/ShmDims[d];
-  }
-  ////////////////////////////////////////////////////////////////
-  // Map Hcube according to physical lattice 
-  // must partition. Loop over dims and find out who would join.
-  ////////////////////////////////////////////////////////////////
-  int hcoor = hypercoor;
-  for(int d=0;d<ndimension;d++){
-     int bits = Log2Size(NodeDims[d],MAXLOG2RANKSPERNODE);
-     int msk  = (0x1<<bits)-1;
-     HyperCoor[d]=hcoor & msk;  
-     HyperCoor[d]=BinaryToGray(HyperCoor[d]); // Space filling curve magic
-     hcoor = hcoor >> bits;
-  } 
-  ////////////////////////////////////////////////////////////////
-  // Check processor counts match
-  ////////////////////////////////////////////////////////////////
-  int Nprocessors=1;
-  for(int i=0;i<ndimension;i++){
-    Nprocessors*=processors[i];
-  }
-  assert(WorldSize==Nprocessors);
-
-  ////////////////////////////////////////////////////////////////
-  // Establish mapping between lexico physics coord and WorldRank
-  ////////////////////////////////////////////////////////////////
-  int rank;
-
-  Lexicographic::CoorFromIndexReversed(NodeCoor,WorldNode   ,NodeDims);
-
-  for(int d=0;d<ndimension;d++) NodeCoor[d]=HyperCoor[d];
-
-  Lexicographic::CoorFromIndexReversed(ShmCoor ,WorldShmRank,ShmDims);
-  for(int d=0;d<ndimension;d++) WorldCoor[d] = NodeCoor[d]*ShmDims[d]+ShmCoor[d];
-  Lexicographic::IndexFromCoorReversed(WorldCoor,rank,WorldDims);
-
-  /////////////////////////////////////////////////////////////////
-  // Build the new communicator
-  /////////////////////////////////////////////////////////////////
-  int ierr= MPI_Comm_split(WorldComm,0,rank,&optimal_comm);
-  assert(ierr==0);
-}
-void GlobalSharedMemory::OptimalCommunicatorSharedMemory(const std::vector<int> &processors,Grid_MPI_Comm & optimal_comm)
-{
-  ////////////////////////////////////////////////////////////////
-  // Assert power of two shm_size.
-  ////////////////////////////////////////////////////////////////
-  int log2size = Log2Size(WorldShmSize,MAXLOG2RANKSPERNODE);
-  assert(log2size != -1);
-
-  ////////////////////////////////////////////////////////////////
-  // Identify subblock of ranks on node spreading across dims
-  // in a maximally symmetrical way
-  ////////////////////////////////////////////////////////////////
-  int ndimension              = processors.size();
-  std::vector<int> processor_coor(ndimension);
-  std::vector<int> WorldDims = processors;   std::vector<int> ShmDims  (ndimension,1);  std::vector<int> NodeDims (ndimension);
-  std::vector<int> ShmCoor  (ndimension);    std::vector<int> NodeCoor (ndimension);    std::vector<int> WorldCoor(ndimension);
-  int dim = 0;
-  for(int l2=0;l2<log2size;l2++){
-    while ( (WorldDims[dim] / ShmDims[dim]) <= 1 ) dim=(dim+1)%ndimension;
-    ShmDims[dim]*=2;
-    dim=(dim+1)%ndimension;
-  }
-
-  ////////////////////////////////////////////////////////////////
-  // Establish torus of processes and nodes with sub-blockings
-  ////////////////////////////////////////////////////////////////
-  for(int d=0;d<ndimension;d++){
-    NodeDims[d] = WorldDims[d]/ShmDims[d];
-  }
-
-  ////////////////////////////////////////////////////////////////
-  // Check processor counts match
-  ////////////////////////////////////////////////////////////////
-  int Nprocessors=1;
-  for(int i=0;i<ndimension;i++){
-    Nprocessors*=processors[i];
-  }
-  assert(WorldSize==Nprocessors);
-
-  ////////////////////////////////////////////////////////////////
-  // Establish mapping between lexico physics coord and WorldRank
-  ////////////////////////////////////////////////////////////////
-  int rank;
-
-  Lexicographic::CoorFromIndexReversed(NodeCoor,WorldNode   ,NodeDims);
-  Lexicographic::CoorFromIndexReversed(ShmCoor ,WorldShmRank,ShmDims);
-  for(int d=0;d<ndimension;d++) WorldCoor[d] = NodeCoor[d]*ShmDims[d]+ShmCoor[d];
-  Lexicographic::IndexFromCoorReversed(WorldCoor,rank,WorldDims);
-
-  /////////////////////////////////////////////////////////////////
-  // Build the new communicator
-  /////////////////////////////////////////////////////////////////
-  int ierr= MPI_Comm_split(WorldComm,0,rank,&optimal_comm);
-  assert(ierr==0);
-}
-////////////////////////////////////////////////////////////////////////////////////////////
-// SHMGET
-////////////////////////////////////////////////////////////////////////////////////////////
-#ifdef GRID_MPI3_SHMGET
-void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
-{
-  std::cout << "SharedMemoryAllocate "<< bytes<< " shmget implementation "<<std::endl;
-  assert(_ShmSetup==1);
-  assert(_ShmAlloc==0);
-
-  //////////////////////////////////////////////////////////////////////////////////////////////////////////
-  // allocate the shared windows for our group
-  //////////////////////////////////////////////////////////////////////////////////////////////////////////
-  MPI_Barrier(WorldShmComm);
-  WorldShmCommBufs.resize(WorldShmSize);
-  std::vector<int> shmids(WorldShmSize);
-
-  if ( WorldShmRank == 0 ) {
-    for(int r=0;r<WorldShmSize;r++){
-      size_t size = bytes;
-      key_t key   = IPC_PRIVATE;
-      int flags = IPC_CREAT | SHM_R | SHM_W;
-#ifdef SHM_HUGETLB
-      if (Hugepages) flags|=SHM_HUGETLB;
-#endif
-      if ((shmids[r]= shmget(key,size, flags)) ==-1) {
-        int errsv = errno;
-        printf("Errno %d\n",errsv);
-        printf("key   %d\n",key);
-        printf("size  %ld\n",size);
-        printf("flags %d\n",flags);
-        perror("shmget");
-        exit(1);
-      }
-    }
-  }
-  MPI_Barrier(WorldShmComm);
-  MPI_Bcast(&shmids[0],WorldShmSize*sizeof(int),MPI_BYTE,0,WorldShmComm);
-  MPI_Barrier(WorldShmComm);
-
-  for(int r=0;r<WorldShmSize;r++){
-    WorldShmCommBufs[r] = (uint64_t *)shmat(shmids[r], NULL,0);
-    if (WorldShmCommBufs[r] == (uint64_t *)-1) {
-      perror("Shared memory attach failure");
-      shmctl(shmids[r], IPC_RMID, NULL);
-      exit(2);
-    }
-  }
-  MPI_Barrier(WorldShmComm);
-  ///////////////////////////////////
-  // Mark for clean up
-  ///////////////////////////////////
-  for(int r=0;r<WorldShmSize;r++){
-    shmctl(shmids[r], IPC_RMID,(struct shmid_ds *)NULL);
-  }
-  MPI_Barrier(WorldShmComm);
-
-  _ShmAlloc=1;
-  _ShmAllocBytes  = bytes;
-}
-#endif
- 
-////////////////////////////////////////////////////////////////////////////////////////////
-// Hugetlbfs mapping intended
-////////////////////////////////////////////////////////////////////////////////////////////
-#ifdef GRID_MPI3_SHMMMAP
-void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
-{
-  std::cout << "SharedMemoryAllocate "<< bytes<< " MMAP implementation "<< GRID_SHM_PATH <<std::endl;
-  assert(_ShmSetup==1);
-  assert(_ShmAlloc==0);
-  //////////////////////////////////////////////////////////////////////////////////////////////////////////
-  // allocate the shared windows for our group
-  //////////////////////////////////////////////////////////////////////////////////////////////////////////
-  MPI_Barrier(WorldShmComm);
-  WorldShmCommBufs.resize(WorldShmSize);
-  
-  ////////////////////////////////////////////////////////////////////////////////////////////
-  // Hugetlbfs and others map filesystems as mappable huge pages
-  ////////////////////////////////////////////////////////////////////////////////////////////
-  char shm_name [NAME_MAX];
-  for(int r=0;r<WorldShmSize;r++){
-    
-    sprintf(shm_name,GRID_SHM_PATH "/Grid_mpi3_shm_%d_%d",WorldNode,r);
-    int fd=open(shm_name,O_RDWR|O_CREAT,0666);
-    if ( fd == -1) { 
-      printf("open %s failed\n",shm_name);
-      perror("open hugetlbfs");
-      exit(0);
-    }
-    int mmap_flag = MAP_SHARED ;
-#ifdef MAP_POPULATE    
-    mmap_flag|=MAP_POPULATE;
-#endif
-#ifdef MAP_HUGETLB
-    if ( flags ) mmap_flag |= MAP_HUGETLB;
-#endif
-    void *ptr = (void *) mmap(NULL, bytes, PROT_READ | PROT_WRITE, mmap_flag,fd, 0); 
-    if ( ptr == (void *)MAP_FAILED ) {    
-      printf("mmap %s failed\n",shm_name);
-      perror("failed mmap");      assert(0);    
-    }
-    assert(((uint64_t)ptr&0x3F)==0);
-    close(fd);
-    WorldShmCommBufs[r] =ptr;
-    //    std::cout << "Set WorldShmCommBufs["<<r<<"]="<<ptr<< "("<< bytes<< "bytes)"<<std::endl;
-  }
-  _ShmAlloc=1;
-  _ShmAllocBytes  = bytes;
-};
-#endif // MMAP
-
-#ifdef GRID_MPI3_SHM_NONE
-void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
-{
-  std::cout << "SharedMemoryAllocate "<< bytes<< " MMAP anonymous implementation "<<std::endl;
-  assert(_ShmSetup==1);
-  assert(_ShmAlloc==0);
-  //////////////////////////////////////////////////////////////////////////////////////////////////////////
-  // allocate the shared windows for our group
-  //////////////////////////////////////////////////////////////////////////////////////////////////////////
-  MPI_Barrier(WorldShmComm);
-  WorldShmCommBufs.resize(WorldShmSize);
-  
-  ////////////////////////////////////////////////////////////////////////////////////////////
-  // Hugetlbf and others map filesystems as mappable huge pages
-  ////////////////////////////////////////////////////////////////////////////////////////////
-  char shm_name [NAME_MAX];
-  assert(WorldShmSize == 1);
-  for(int r=0;r<WorldShmSize;r++){
-    
-    int fd=-1;
-    int mmap_flag = MAP_SHARED |MAP_ANONYMOUS ;
-#ifdef MAP_POPULATE    
-    mmap_flag|=MAP_POPULATE;
-#endif
-#ifdef MAP_HUGETLB
-    if ( flags ) mmap_flag |= MAP_HUGETLB;
-#endif
-    void *ptr = (void *) mmap(NULL, bytes, PROT_READ | PROT_WRITE, mmap_flag,fd, 0); 
-    if ( ptr == (void *)MAP_FAILED ) {    
-      printf("mmap %s failed\n",shm_name);
-      perror("failed mmap");      assert(0);    
-    }
-    assert(((uint64_t)ptr&0x3F)==0);
-    close(fd);
-    WorldShmCommBufs[r] =ptr;
-    //    std::cout << "Set WorldShmCommBufs["<<r<<"]="<<ptr<< "("<< bytes<< "bytes)"<<std::endl;
-  }
-  _ShmAlloc=1;
-  _ShmAllocBytes  = bytes;
-};
-#endif // MMAP
-
-#ifdef GRID_MPI3_SHMOPEN
-////////////////////////////////////////////////////////////////////////////////////////////
-// POSIX SHMOPEN ; as far as I know Linux does not allow EXPLICIT HugePages with this case
-// tmpfs (Larry Meadows says) does not support explicit huge page, and this is used for 
-// the posix shm virtual file system
-////////////////////////////////////////////////////////////////////////////////////////////
-void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
-{ 
-  std::cout << "SharedMemoryAllocate "<< bytes<< " SHMOPEN implementation "<<std::endl;
-  assert(_ShmSetup==1);
-  assert(_ShmAlloc==0); 
-  MPI_Barrier(WorldShmComm);
-  WorldShmCommBufs.resize(WorldShmSize);
-
-  char shm_name [NAME_MAX];
-  if ( WorldShmRank == 0 ) {
-    for(int r=0;r<WorldShmSize;r++){
-	
-      size_t size = bytes;
-      
-      struct passwd *pw = getpwuid (getuid());
-      sprintf(shm_name,"/Grid_%s_mpi3_shm_%d_%d",pw->pw_name,WorldNode,r);
-      
-      shm_unlink(shm_name);
-      int fd=shm_open(shm_name,O_RDWR|O_CREAT,0666);
-      if ( fd < 0 ) {	perror("failed shm_open");	assert(0);      }
-      ftruncate(fd, size);
-	
-      int mmap_flag = MAP_SHARED;
-#ifdef MAP_POPULATE 
-      mmap_flag |= MAP_POPULATE;
-#endif
-#ifdef MAP_HUGETLB
-      if (flags) mmap_flag |= MAP_HUGETLB;
-#endif
-      void * ptr =  mmap(NULL,size, PROT_READ | PROT_WRITE, mmap_flag, fd, 0);
-      
-      //      std::cout << "Set WorldShmCommBufs["<<r<<"]="<<ptr<< "("<< size<< "bytes)"<<std::endl;
-      if ( ptr == (void * )MAP_FAILED ) {       
-	perror("failed mmap");     
-	assert(0);    
-      }
-      assert(((uint64_t)ptr&0x3F)==0);
-      
-      WorldShmCommBufs[r] =ptr;
-      close(fd);
-    }
-  }
-
-  MPI_Barrier(WorldShmComm);
-  
-  if ( WorldShmRank != 0 ) { 
-    for(int r=0;r<WorldShmSize;r++){
-
-      size_t size = bytes ;
-      
-      struct passwd *pw = getpwuid (getuid());
-      sprintf(shm_name,"/Grid_%s_mpi3_shm_%d_%d",pw->pw_name,WorldNode,r);
-      
-      int fd=shm_open(shm_name,O_RDWR,0666);
-      if ( fd<0 ) {	perror("failed shm_open");	assert(0);      }
-      
-      void * ptr =  mmap(NULL,size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
-      if ( ptr == MAP_FAILED ) {       perror("failed mmap");      assert(0);    }
-      assert(((uint64_t)ptr&0x3F)==0);
-      WorldShmCommBufs[r] =ptr;
-
-      close(fd);
-    }
-  }
-  _ShmAlloc=1;
-  _ShmAllocBytes = bytes;
-}
-#endif
-
-
-
-
-  ////////////////////////////////////////////////////////
-  // Global shared functionality finished
-  // Now move to per communicator functionality
-  ////////////////////////////////////////////////////////
-void SharedMemory::SetCommunicator(Grid_MPI_Comm comm)
-{
-  int rank, size;
-  MPI_Comm_rank(comm,&rank);
-  MPI_Comm_size(comm,&size);
-  ShmRanks.resize(size);
-
-  /////////////////////////////////////////////////////////////////////
-  // Split into groups that can share memory
-  /////////////////////////////////////////////////////////////////////
-  MPI_Comm_split_type(comm, MPI_COMM_TYPE_SHARED, 0, MPI_INFO_NULL,&ShmComm);
-  MPI_Comm_rank(ShmComm     ,&ShmRank);
-  MPI_Comm_size(ShmComm     ,&ShmSize);
-  ShmCommBufs.resize(ShmSize);
-
-  //////////////////////////////////////////////////////////////////////
-  // Map ShmRank to WorldShmRank and use the right buffer
-  //////////////////////////////////////////////////////////////////////
-  assert (GlobalSharedMemory::ShmAlloc()==1);
-  heap_size = GlobalSharedMemory::ShmAllocBytes();
-  for(int r=0;r<ShmSize;r++){
-
-    uint32_t wsr = (r==ShmRank) ? GlobalSharedMemory::WorldShmRank : 0 ;
-
-    MPI_Allreduce(MPI_IN_PLACE,&wsr,1,MPI_UINT32_T,MPI_SUM,ShmComm);
-
-    ShmCommBufs[r] = GlobalSharedMemory::WorldShmCommBufs[wsr];
-    //    std::cout << "SetCommunicator ShmCommBufs ["<< r<< "] = "<< ShmCommBufs[r]<< "  wsr = "<<wsr<<std::endl;
-  }
-  ShmBufferFreeAll();
-
-  /////////////////////////////////////////////////////////////////////
-  // find comm ranks in our SHM group (i.e. which ranks are on our node)
-  /////////////////////////////////////////////////////////////////////
-  MPI_Group FullGroup, ShmGroup;
-  MPI_Comm_group (comm   , &FullGroup); 
-  MPI_Comm_group (ShmComm, &ShmGroup);
-
-  std::vector<int> ranks(size);   for(int r=0;r<size;r++) ranks[r]=r;
-  MPI_Group_translate_ranks (FullGroup,size,&ranks[0],ShmGroup, &ShmRanks[0]); 
-}
-//////////////////////////////////////////////////////////////////
-// On node barrier
-//////////////////////////////////////////////////////////////////
-void SharedMemory::ShmBarrier(void)
-{
-  MPI_Barrier  (ShmComm);
-}
-//////////////////////////////////////////////////////////////////////////////////////////////////////////
-// Test the shared memory is working
-//////////////////////////////////////////////////////////////////////////////////////////////////////////
-void SharedMemory::SharedMemoryTest(void)
-{
-  ShmBarrier();
-  if ( ShmRank == 0 ) {
-    for(int r=0;r<ShmSize;r++){
-      uint64_t * check = (uint64_t *) ShmCommBufs[r];
-      check[0] = GlobalSharedMemory::WorldNode;
-      check[1] = r;
-      check[2] = 0x5A5A5A;
-    }
-  }
-  ShmBarrier();
-  for(int r=0;r<ShmSize;r++){
-    uint64_t * check = (uint64_t *) ShmCommBufs[r];
-    
-    assert(check[0]==GlobalSharedMemory::WorldNode);
-    assert(check[1]==r);
-    assert(check[2]==0x5A5A5A);
-    
-  }
-  ShmBarrier();
-}
-
-void *SharedMemory::ShmBuffer(int rank)
-{
-  int gpeer = ShmRanks[rank];
-  if (gpeer == MPI_UNDEFINED){
-    return NULL;
-  } else { 
-    return ShmCommBufs[gpeer];
-  }
-}
-void *SharedMemory::ShmBufferTranslate(int rank,void * local_p)
-{
-  static int count =0;
-  int gpeer = ShmRanks[rank];
-  assert(gpeer!=ShmRank); // never send to self
-  if (gpeer == MPI_UNDEFINED){
-    return NULL;
-  } else { 
-    uint64_t offset = (uint64_t)local_p - (uint64_t)ShmCommBufs[ShmRank];
-    uint64_t remote = (uint64_t)ShmCommBufs[gpeer]+offset;
-    return (void *) remote;
-  }
-}
-SharedMemory::~SharedMemory()
-{
-  int MPI_is_finalised;  MPI_Finalized(&MPI_is_finalised);
-  if ( !MPI_is_finalised ) { 
-    MPI_Comm_free(&ShmComm);
-  }
-};
-
-}

Grid/communicator/SharedMemoryNone.cc

@@ -1,128 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/communicator/SharedMemory.cc
-
-    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 <Grid/GridCore.h>
-
-namespace Grid { 
-
-/*Construct from an MPI communicator*/
-void GlobalSharedMemory::Init(Grid_MPI_Comm comm)
-{
-  assert(_ShmSetup==0);
-  WorldComm = 0;
-  WorldRank = 0;
-  WorldSize = 1;
-  WorldShmComm = 0 ;
-  WorldShmRank = 0 ;
-  WorldShmSize = 1 ;
-  WorldNodes   = 1 ;
-  WorldNode    = 0 ;
-  WorldShmRanks.resize(WorldSize); WorldShmRanks[0] = 0;
-  WorldShmCommBufs.resize(1);
-  _ShmSetup=1;
-}
-
-void GlobalSharedMemory::OptimalCommunicator(const std::vector<int> &processors,Grid_MPI_Comm & optimal_comm)
-{
-  optimal_comm = WorldComm;
-}
-
-////////////////////////////////////////////////////////////////////////////////////////////
-// Hugetlbfs mapping intended, use anonymous mmap
-////////////////////////////////////////////////////////////////////////////////////////////
-void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
-{
-  void * ShmCommBuf ; 
-  assert(_ShmSetup==1);
-  assert(_ShmAlloc==0);
-  int mmap_flag =0;
-#ifdef MAP_ANONYMOUS
-  mmap_flag = mmap_flag| MAP_SHARED | MAP_ANONYMOUS;
-#endif
-#ifdef MAP_ANON
-  mmap_flag = mmap_flag| MAP_SHARED | MAP_ANON;
-#endif
-#ifdef MAP_HUGETLB
-  if ( flags ) mmap_flag |= MAP_HUGETLB;
-#endif
-  ShmCommBuf =(void *) mmap(NULL, bytes, PROT_READ | PROT_WRITE, mmap_flag, -1, 0); 
-  if (ShmCommBuf == (void *)MAP_FAILED) {
-    perror("mmap failed ");
-    exit(EXIT_FAILURE);  
-  }
-#ifdef MADV_HUGEPAGE
-  if (!Hugepages ) madvise(ShmCommBuf,bytes,MADV_HUGEPAGE);
-#endif
-  bzero(ShmCommBuf,bytes);
-  WorldShmCommBufs[0] = ShmCommBuf;
-  _ShmAllocBytes=bytes;
-  _ShmAlloc=1;
-};
-
-  ////////////////////////////////////////////////////////
-  // Global shared functionality finished
-  // Now move to per communicator functionality
-  ////////////////////////////////////////////////////////
-void SharedMemory::SetCommunicator(Grid_MPI_Comm comm)
-{
-  assert(GlobalSharedMemory::ShmAlloc()==1);
-  ShmRanks.resize(1);
-  ShmCommBufs.resize(1);
-  ShmRanks[0] = 0;
-  ShmRank     = 0;
-  ShmSize     = 1;
-  //////////////////////////////////////////////////////////////////////
-  // Map ShmRank to WorldShmRank and use the right buffer
-  //////////////////////////////////////////////////////////////////////
-  ShmCommBufs[0] = GlobalSharedMemory::WorldShmCommBufs[0];
-  heap_size      = GlobalSharedMemory::ShmAllocBytes();
-  ShmBufferFreeAll();
-  return;
-}
-//////////////////////////////////////////////////////////////////
-// On node barrier
-//////////////////////////////////////////////////////////////////
-void SharedMemory::ShmBarrier(void){ return ; }
-
-//////////////////////////////////////////////////////////////////////////////////////////////////////////
-// Test the shared memory is working
-//////////////////////////////////////////////////////////////////////////////////////////////////////////
-void SharedMemory::SharedMemoryTest(void) { return; }
-
-void *SharedMemory::ShmBuffer(int rank)
-{
-  return NULL;
-}
-void *SharedMemory::ShmBufferTranslate(int rank,void * local_p)
-{
-  return NULL;
-}
-SharedMemory::~SharedMemory()
-{};
-
-}

Grid/lattice/Lattice_ET.h

@@ -1,466 +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 Grid {
-
-////////////////////////////////////////////////////
-// Predicated where support
-////////////////////////////////////////////////////
-template <class iobj, class vobj, class robj>
-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();
-  const int words = sizeof(vobj) / sizeof(vector_type);
-
-  std::vector<Integer> mask(Nsimd);
-  std::vector<scalar_object> truevals(Nsimd);
-  std::vector<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;
-}
-
-////////////////////////////////////////////
-// recursive evaluation of expressions; Could
-// switch to generic approach with variadics, a la
-// Antonin's Lat Sim but the repack to variadic with popped
-// from tuple is hideous; C++14 introduces std::make_index_sequence for this
-////////////////////////////////////////////
-
-// leaf eval of lattice ; should enable if protect using traits
-
-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 <typename T> using is_lattice_expr = std::is_base_of<LatticeExpressionBase,T >;
-
-//Specialization of getVectorType for lattices
-template<typename T>
-struct getVectorType<Lattice<T> >{
-  typedef typename Lattice<T>::vector_object type;
-};
- 
-template<class sobj>
-inline sobj eval(const unsigned int ss, const sobj &arg)
-{
-  return arg;
-}
-template <class lobj>
-inline const lobj &eval(const unsigned int ss, const Lattice<lobj> &arg) {
-  return arg._odata[ss];
-}
-
-// handle nodes in syntax tree
-template <typename Op, typename T1>
-auto inline eval(
-    const unsigned int ss,
-    const LatticeUnaryExpression<Op, T1> &expr)  // eval one operand
-    -> decltype(expr.first.func(eval(ss, std::get<0>(expr.second)))) {
-  return expr.first.func(eval(ss, std::get<0>(expr.second)));
-}
-
-template <typename Op, typename T1, typename T2>
-auto inline eval(
-    const unsigned int ss,
-    const LatticeBinaryExpression<Op, T1, T2> &expr)  // eval two operands
-    -> decltype(expr.first.func(eval(ss, std::get<0>(expr.second)),
-                                eval(ss, std::get<1>(expr.second)))) {
-  return expr.first.func(eval(ss, std::get<0>(expr.second)),
-                         eval(ss, std::get<1>(expr.second)));
-}
-
-template <typename Op, typename T1, typename T2, typename T3>
-auto inline eval(const unsigned int ss,
-                 const LatticeTrinaryExpression<Op, T1, T2, T3>
-                     &expr)  // eval three operands
-    -> decltype(expr.first.func(eval(ss, std::get<0>(expr.second)),
-                                eval(ss, std::get<1>(expr.second)),
-                                eval(ss, std::get<2>(expr.second)))) {
-  return expr.first.func(eval(ss, std::get<0>(expr.second)),
-                         eval(ss, std::get<1>(expr.second)),
-                         eval(ss, std::get<2>(expr.second)));
-}
-
-//////////////////////////////////////////////////////////////////////////
-// Obtain the grid 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 GridFromExpression(GridBase *&grid, const T1 &lat)  // Lattice leaf
-{
-  if (grid) {
-    conformable(grid, lat._grid);
-  }
-  grid = lat._grid;
-}
-template <class T1,
-          typename std::enable_if<!is_lattice<T1>::value, T1>::type * = nullptr>
-inline void GridFromExpression(GridBase *&grid,
-                               const T1 &notlat)  // non-lattice leaf
-{}
-template <typename Op, typename T1>
-inline void GridFromExpression(GridBase *&grid,
-                               const LatticeUnaryExpression<Op, T1> &expr) {
-  GridFromExpression(grid, std::get<0>(expr.second));  // recurse
-}
-
-template <typename Op, typename T1, typename T2>
-inline void GridFromExpression(
-    GridBase *&grid, const LatticeBinaryExpression<Op, T1, T2> &expr) {
-  GridFromExpression(grid, std::get<0>(expr.second));  // recurse
-  GridFromExpression(grid, std::get<1>(expr.second));
-}
-template <typename Op, typename T1, typename T2, typename T3>
-inline void GridFromExpression(
-    GridBase *&grid, const LatticeTrinaryExpression<Op, T1, T2, T3> &expr) {
-  GridFromExpression(grid, std::get<0>(expr.second));  // recurse
-  GridFromExpression(grid, std::get<1>(expr.second));
-  GridFromExpression(grid, std::get<2>(expr.second));
-}
-
-//////////////////////////////////////////////////////////////////////////
-// 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;
-  //  std::cout<<GridLogMessage<<"Lattice leaf cb "<<cb<<std::endl;
-}
-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
-{
-  //  std::cout<<GridLogMessage<<"Non lattice leaf cb"<<cb<<std::endl;
-}
-template <typename Op, typename T1>
-inline void CBFromExpression(int &cb,
-                             const LatticeUnaryExpression<Op, T1> &expr) {
-  CBFromExpression(cb, std::get<0>(expr.second));  // recurse
-  //  std::cout<<GridLogMessage<<"Unary node cb "<<cb<<std::endl;
-}
-
-template <typename Op, typename T1, typename T2>
-inline void CBFromExpression(int &cb,
-                             const LatticeBinaryExpression<Op, T1, T2> &expr) {
-  CBFromExpression(cb, std::get<0>(expr.second));  // recurse
-  CBFromExpression(cb, std::get<1>(expr.second));
-  //  std::cout<<GridLogMessage<<"Binary node cb "<<cb<<std::endl;
-}
-template <typename Op, typename T1, typename T2, typename T3>
-inline void CBFromExpression(
-    int &cb, const LatticeTrinaryExpression<Op, T1, T2, T3> &expr) {
-  CBFromExpression(cb, std::get<0>(expr.second));  // recurse
-  CBFromExpression(cb, std::get<1>(expr.second));
-  CBFromExpression(cb, std::get<2>(expr.second));
-  //  std::cout<<GridLogMessage<<"Trinary node cb "<<cb<<std::endl;
-}
-
-////////////////////////////////////////////
-// Unary operators and funcs
-////////////////////////////////////////////
-#define GridUnopClass(name, ret)                                          \
-  template <class arg>                                                    \
-  struct name {                                                           \
-    static auto 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 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 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), const T1 &>(       \
-          std::make_pair(GRID_UNOP(name)(), std::forward_as_tuple(arg)))) { \
-    return LatticeUnaryExpression<GRID_UNOP(name), const T1 &>(             \
-        std::make_pair(GRID_UNOP(name)(), std::forward_as_tuple(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), const T1 &, const T2 &>(  \
-              std::make_pair(GRID_BINOP(name)(),                              \
-                             std::forward_as_tuple(lhs, rhs)))) {             \
-    return LatticeBinaryExpression<GRID_BINOP(name), const T1 &, const T2 &>( \
-        std::make_pair(GRID_BINOP(name)(), std::forward_as_tuple(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), const T1 &, const T2 &>(  \
-              std::make_pair(GRID_BINOP(name)(),                              \
-                             std::forward_as_tuple(lhs, rhs)))) {             \
-    return LatticeBinaryExpression<GRID_BINOP(name), const T1 &, const T2 &>( \
-        std::make_pair(GRID_BINOP(name)(), std::forward_as_tuple(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), const T1 &, const T2 &,  \
-                                   const T3 &>(std::make_pair(                 \
-              GRID_TRINOP(name)(), std::forward_as_tuple(pred, lhs, rhs)))) {  \
-    return LatticeTrinaryExpression<GRID_TRINOP(name), const T1 &, const T2 &, \
-                                    const T3 &>(std::make_pair(                \
-        GRID_TRINOP(name)(), std::forward_as_tuple(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.first.func(eval(0, std::get<0>(expr.second))))> {
-  Lattice<decltype(expr.first.func(eval(0, std::get<0>(expr.second))))> ret(
-      expr);
-  return ret;
-}
-template <class Op, class T1, class T2>
-auto closure(const LatticeBinaryExpression<Op, T1, T2> &expr)
-    -> Lattice<decltype(expr.first.func(eval(0, std::get<0>(expr.second)),
-                                        eval(0, std::get<1>(expr.second))))> {
-  Lattice<decltype(expr.first.func(eval(0, std::get<0>(expr.second)),
-                                   eval(0, std::get<1>(expr.second))))>
-      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.first.func(eval(0, std::get<0>(expr.second)),
-                                        eval(0, std::get<1>(expr.second)),
-                                        eval(0, std::get<2>(expr.second))))> {
-  Lattice<decltype(expr.first.func(eval(0, std::get<0>(expr.second)),
-                                   eval(0, std::get<1>(expr.second)),
-                                   eval(0, std::get<2>(expr.second))))>
-      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
-}
-
-#if 0
-using namespace Grid;
-        
- int main(int argc,char **argv){
-   
-   Lattice<double> v1(16);
-   Lattice<double> v2(16);
-   Lattice<double> v3(16);
-
-   BinaryAdd<double,double> tmp;
-   LatticeBinaryExpression<BinaryAdd<double,double>,Lattice<double> &,Lattice<double> &> 
-     expr(std::make_pair(tmp,
-    std::forward_as_tuple(v1,v2)));
-   tmp.func(eval(0,v1),eval(0,v2));
-
-   auto var = v1+v2;
-   std::cout<<GridLogMessage<<typeid(var).name()<<std::endl;
-
-   v3=v1+v2;
-   v3=v1+v2+v1*v2;
- };
-
-void testit(Lattice<double> &v1,Lattice<double> &v2,Lattice<double> &v3)
-{
-   v3=v1+v2+v1*v2;
-}
-#endif
-
-#endif

Grid/lattice/Lattice_reduction.h

@@ -1,733 +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 */
-#ifndef GRID_LATTICE_REDUCTION_H
-#define GRID_LATTICE_REDUCTION_H
-
-#include <Grid/Grid_Eigen_Dense.h>
-
-namespace Grid {
-#ifdef GRID_WARN_SUBOPTIMAL
-#warning "Optimisation alert all these reduction loops are NOT threaded "
-#endif     
-
-  ////////////////////////////////////////////////////////////////////////////////////////////////////
-  // Deterministic Reduction operations
-  ////////////////////////////////////////////////////////////////////////////////////////////////////
-template<class vobj> inline RealD norm2(const Lattice<vobj> &arg){
-  auto nrm = innerProduct(arg,arg);
-  return std::real(nrm); 
-}
-
-// Double inner product
-template<class vobj>
-inline ComplexD innerProduct(const Lattice<vobj> &left,const Lattice<vobj> &right)
-{
-  typedef typename vobj::scalar_type scalar_type;
-  typedef typename vobj::vector_typeD vector_type;
-  GridBase *grid = left._grid;
-  const int pad = 8;
-
-  ComplexD  inner;
-  Vector<ComplexD> sumarray(grid->SumArraySize()*pad);
-
-  parallel_for(int thr=0;thr<grid->SumArraySize();thr++){
-    int nwork, mywork, myoff;
-    GridThread::GetWork(left._grid->oSites(),thr,mywork,myoff);
-    
-    decltype(innerProductD(left._odata[0],right._odata[0])) vinner=zero; // private to thread; sub summation
-    for(int ss=myoff;ss<mywork+myoff; ss++){
-      vinner = vinner + innerProductD(left._odata[ss],right._odata[ss]);
-    }
-    // All threads sum across SIMD; reduce serial work at end
-    // one write per cacheline with streaming store
-    ComplexD tmp = Reduce(TensorRemove(vinner)) ;
-    vstream(sumarray[thr*pad],tmp);
-  }
-  
-  inner=0.0;
-  for(int i=0;i<grid->SumArraySize();i++){
-    inner = inner+sumarray[i*pad];
-  } 
-  right._grid->GlobalSum(inner);
-  return inner;
-}
-
-/////////////////////////
-// Fast axpby_norm
-// z = a x + b y
-// return norm z
-/////////////////////////
-template<class sobj,class vobj> strong_inline RealD 
-axpy_norm_fast(Lattice<vobj> &z,sobj a,const Lattice<vobj> &x,const Lattice<vobj> &y) 
-{
-  sobj one(1.0);
-  return axpby_norm_fast(z,a,one,x,y);
-}
-
-template<class sobj,class vobj> strong_inline RealD 
-axpby_norm_fast(Lattice<vobj> &z,sobj a,sobj b,const Lattice<vobj> &x,const Lattice<vobj> &y) 
-{
-  const int pad = 8;
-  z.checkerboard = x.checkerboard;
-  conformable(z,x);
-  conformable(x,y);
-
-  typedef typename vobj::scalar_type scalar_type;
-  typedef typename vobj::vector_typeD vector_type;
-  RealD  nrm;
-  
-  GridBase *grid = x._grid;
-  
-  Vector<RealD> sumarray(grid->SumArraySize()*pad);
-  
-  parallel_for(int thr=0;thr<grid->SumArraySize();thr++){
-    int nwork, mywork, myoff;
-    GridThread::GetWork(x._grid->oSites(),thr,mywork,myoff);
-    
-    // private to thread; sub summation
-    decltype(innerProductD(z._odata[0],z._odata[0])) vnrm=zero; 
-    for(int ss=myoff;ss<mywork+myoff; ss++){
-      vobj tmp = a*x._odata[ss]+b*y._odata[ss];
-      vnrm = vnrm + innerProductD(tmp,tmp);
-      vstream(z._odata[ss],tmp);
-    }
-    vstream(sumarray[thr*pad],real(Reduce(TensorRemove(vnrm)))) ;
-  }
-  
-  nrm = 0.0; // sum across threads; linear in thread count but fast
-  for(int i=0;i<grid->SumArraySize();i++){
-    nrm = nrm+sumarray[i*pad];
-  } 
-  z._grid->GlobalSum(nrm);
-  return nrm; 
-}
-
- 
-template<class Op,class T1>
-inline auto sum(const LatticeUnaryExpression<Op,T1> & expr)
-  ->typename decltype(expr.first.func(eval(0,std::get<0>(expr.second))))::scalar_object
-{
-  return sum(closure(expr));
-}
-
-template<class Op,class T1,class T2>
-inline auto sum(const LatticeBinaryExpression<Op,T1,T2> & expr)
-      ->typename decltype(expr.first.func(eval(0,std::get<0>(expr.second)),eval(0,std::get<1>(expr.second))))::scalar_object
-{
-  return sum(closure(expr));
-}
-
-
-template<class Op,class T1,class T2,class T3>
-inline auto sum(const LatticeTrinaryExpression<Op,T1,T2,T3> & expr)
-  ->typename decltype(expr.first.func(eval(0,std::get<0>(expr.second)),
-				      eval(0,std::get<1>(expr.second)),
-				      eval(0,std::get<2>(expr.second))
-				      ))::scalar_object
-{
-  return sum(closure(expr));
-}
-
-template<class vobj>
-inline typename vobj::scalar_object sum(const Lattice<vobj> &arg)
-{
-  GridBase *grid=arg._grid;
-  int Nsimd = grid->Nsimd();
-  
-  std::vector<vobj,alignedAllocator<vobj> > sumarray(grid->SumArraySize());
-  for(int i=0;i<grid->SumArraySize();i++){
-    sumarray[i]=zero;
-  }
-  
-  parallel_for(int thr=0;thr<grid->SumArraySize();thr++){
-    int nwork, mywork, myoff;
-    GridThread::GetWork(grid->oSites(),thr,mywork,myoff);
-    
-    vobj vvsum=zero;
-    for(int ss=myoff;ss<mywork+myoff; ss++){
-      vvsum = vvsum + arg._odata[ss];
-    }
-    sumarray[thr]=vvsum;
-  }
-  
-  vobj vsum=zero;  // sum across threads
-  for(int i=0;i<grid->SumArraySize();i++){
-    vsum = vsum+sumarray[i];
-  } 
-  
-  typedef typename vobj::scalar_object sobj;
-  sobj ssum=zero;
-  
-  std::vector<sobj>               buf(Nsimd);
-  extract(vsum,buf);
-  
-  for(int i=0;i<Nsimd;i++) ssum = ssum + buf[i];
-  arg._grid->GlobalSum(ssum);
-  
-  return ssum;
-}
-
-
-//////////////////////////////////////////////////////////////////////////////////////////////////////////////
-// sliceSum, sliceInnerProduct, sliceAxpy, sliceNorm etc...
-//////////////////////////////////////////////////////////////////////////////////////////////////////////////
-
-template<class vobj> inline void sliceSum(const Lattice<vobj> &Data,std::vector<typename vobj::scalar_object> &result,int orthogdim)
-{
-  ///////////////////////////////////////////////////////
-  // FIXME precision promoted summation
-  // may be important for correlation functions
-  // But easily avoided by using double precision fields
-  ///////////////////////////////////////////////////////
-  typedef typename vobj::scalar_object sobj;
-  GridBase  *grid = Data._grid;
-  assert(grid!=NULL);
-
-  const int    Nd = grid->_ndimension;
-  const int Nsimd = grid->Nsimd();
-
-  assert(orthogdim >= 0);
-  assert(orthogdim < Nd);
-
-  int fd=grid->_fdimensions[orthogdim];
-  int ld=grid->_ldimensions[orthogdim];
-  int rd=grid->_rdimensions[orthogdim];
-
-  std::vector<vobj,alignedAllocator<vobj> > lvSum(rd); // will locally sum vectors first
-  std::vector<sobj> lsSum(ld,zero);                    // sum across these down to scalars
-  std::vector<sobj> extracted(Nsimd);                  // splitting the SIMD
-
-  result.resize(fd); // And then global sum to return the same vector to every node 
-  for(int r=0;r<rd;r++){
-    lvSum[r]=zero;
-  }
-
-  int e1=    grid->_slice_nblock[orthogdim];
-  int e2=    grid->_slice_block [orthogdim];
-  int stride=grid->_slice_stride[orthogdim];
-
-  // sum over reduced dimension planes, breaking out orthog dir
-  // Parallel over orthog direction
-  parallel_for(int r=0;r<rd;r++){
-
-    int so=r*grid->_ostride[orthogdim]; // base offset for start of plane 
-
-    for(int n=0;n<e1;n++){
-      for(int b=0;b<e2;b++){
-	int ss= so+n*stride+b;
-	lvSum[r]=lvSum[r]+Data._odata[ss];
-      }
-    }
-  }
-
-  // Sum across simd lanes in the plane, breaking out orthog dir.
-  std::vector<int> icoor(Nd);
-
-  for(int rt=0;rt<rd;rt++){
-
-    extract(lvSum[rt],extracted);
-
-    for(int idx=0;idx<Nsimd;idx++){
-
-      grid->iCoorFromIindex(icoor,idx);
-
-      int ldx =rt+icoor[orthogdim]*rd;
-
-      lsSum[ldx]=lsSum[ldx]+extracted[idx];
-
-    }
-  }
-  
-  // sum over nodes.
-  sobj gsum;
-  for(int t=0;t<fd;t++){
-    int pt = t/ld; // processor plane
-    int lt = t%ld;
-    if ( pt == grid->_processor_coor[orthogdim] ) {
-      gsum=lsSum[lt];
-    } else {
-      gsum=zero;
-    }
-
-    grid->GlobalSum(gsum);
-
-    result[t]=gsum;
-  }
-}
-
-template<class vobj>
-static void mySliceInnerProductVector( std::vector<ComplexD> & result, const Lattice<vobj> &lhs,const Lattice<vobj> &rhs,int orthogdim)
-{
-  // std::cout << GridLogMessage << "Start mySliceInnerProductVector" << std::endl;
-
-  typedef typename vobj::scalar_type scalar_type;
-  std::vector<scalar_type> lsSum;
-  localSliceInnerProductVector(result, lhs, rhs, lsSum, orthogdim);
-  globalSliceInnerProductVector(result, lhs, lsSum, orthogdim);
-  // std::cout << GridLogMessage << "End mySliceInnerProductVector" << std::endl;
-}
-
-template <class vobj>
-static void localSliceInnerProductVector(std::vector<ComplexD> &result, const Lattice<vobj> &lhs, const Lattice<vobj> &rhs, std::vector<typename vobj::scalar_type> &lsSum, int orthogdim)
-{
-  // std::cout << GridLogMessage << "Start prep" << std::endl;
-  typedef typename vobj::vector_type   vector_type;
-  typedef typename vobj::scalar_type   scalar_type;
-  GridBase  *grid = lhs._grid;
-  assert(grid!=NULL);
-  conformable(grid,rhs._grid);
-
-  const int    Nd = grid->_ndimension;
-  const int Nsimd = grid->Nsimd();
-
-  assert(orthogdim >= 0);
-  assert(orthogdim < Nd);
-
-  int fd=grid->_fdimensions[orthogdim];
-  int ld=grid->_ldimensions[orthogdim];
-  int rd=grid->_rdimensions[orthogdim];
-  // std::cout << GridLogMessage << "Start alloc" << std::endl;
-
-  std::vector<vector_type,alignedAllocator<vector_type> > lvSum(rd); // will locally sum vectors first
-  lsSum.resize(ld,scalar_type(0.0));                    // sum across these down to scalars
-  std::vector<iScalar<scalar_type>> extracted(Nsimd);   // splitting the SIMD  
-  // std::cout << GridLogMessage << "End alloc" << std::endl;
-
-  result.resize(fd); // And then global sum to return the same vector to every node for IO to file
-  for(int r=0;r<rd;r++){
-    lvSum[r]=zero;
-  }
-
-  int e1=    grid->_slice_nblock[orthogdim];
-  int e2=    grid->_slice_block [orthogdim];
-  int stride=grid->_slice_stride[orthogdim];
-  // std::cout << GridLogMessage << "End prep" << std::endl;
-  // std::cout << GridLogMessage << "Start parallel inner product, _rd = " << rd << std::endl;
-  vector_type vv;
-  parallel_for(int r=0;r<rd;r++)
-  {
-
-    int so=r*grid->_ostride[orthogdim]; // base offset for start of plane 
-
-    for(int n=0;n<e1;n++){
-      for(int b=0;b<e2;b++){
-        int ss = so + n * stride + b;
-        vv = TensorRemove(innerProduct(lhs._odata[ss], rhs._odata[ss]));
-        lvSum[r] = lvSum[r] + vv;
-      }
-    }
-  }
-  // std::cout << GridLogMessage << "End parallel inner product" << std::endl;
-
-  // Sum across simd lanes in the plane, breaking out orthog dir.
-  std::vector<int> icoor(Nd);
-  for(int rt=0;rt<rd;rt++){
-
-    iScalar<vector_type> temp; 
-    temp._internal = lvSum[rt];
-    extract(temp,extracted);
-
-    for(int idx=0;idx<Nsimd;idx++){
-
-      grid->iCoorFromIindex(icoor,idx);
-
-      int ldx =rt+icoor[orthogdim]*rd;
-
-      lsSum[ldx]=lsSum[ldx]+extracted[idx]._internal;
-
-    }
-  }
-  // std::cout << GridLogMessage << "End sum over simd lanes" << std::endl;
-}
-template <class vobj>
-static void globalSliceInnerProductVector(std::vector<ComplexD> &result, const Lattice<vobj> &lhs, std::vector<typename vobj::scalar_type> &lsSum, int orthogdim)
-{
-  typedef typename vobj::scalar_type scalar_type;
-  GridBase *grid = lhs._grid;
-  int fd = result.size();
-  int ld = lsSum.size();
-  // sum over nodes.
-  std::vector<scalar_type> gsum;
-  gsum.resize(fd, scalar_type(0.0));
-  // std::cout << GridLogMessage << "Start of gsum[t] creation:" << std::endl;
-  for(int t=0;t<fd;t++){
-    int pt = t/ld; // processor plane
-    int lt = t%ld;
-    if ( pt == grid->_processor_coor[orthogdim] ) {
-      gsum[t]=lsSum[lt];
-    }
-  }
-  // std::cout << GridLogMessage << "End of gsum[t] creation:" << std::endl;
-  // std::cout << GridLogMessage << "Start of GlobalSumVector:" << std::endl;
-  grid->GlobalSumVector(&gsum[0], fd);
-  // std::cout << GridLogMessage << "End of GlobalSumVector:" << std::endl;
-
-  result = gsum;
-}
-template<class vobj>
-static void sliceInnerProductVector( std::vector<ComplexD> & result, const Lattice<vobj> &lhs,const Lattice<vobj> &rhs,int orthogdim) 
-{
-  typedef typename vobj::vector_type   vector_type;
-  typedef typename vobj::scalar_type   scalar_type;
-  GridBase  *grid = lhs._grid;
-  assert(grid!=NULL);
-  conformable(grid,rhs._grid);
-
-  const int    Nd = grid->_ndimension;
-  const int Nsimd = grid->Nsimd();
-
-  assert(orthogdim >= 0);
-  assert(orthogdim < Nd);
-
-  int fd=grid->_fdimensions[orthogdim];
-  int ld=grid->_ldimensions[orthogdim];
-  int rd=grid->_rdimensions[orthogdim];
-
-  std::vector<vector_type,alignedAllocator<vector_type> > lvSum(rd); // will locally sum vectors first
-  std::vector<scalar_type > lsSum(ld,scalar_type(0.0));                    // sum across these down to scalars
-  std::vector<iScalar<scalar_type> > extracted(Nsimd);                  // splitting the SIMD
-
-  result.resize(fd); // And then global sum to return the same vector to every node for IO to file
-  for(int r=0;r<rd;r++){
-    lvSum[r]=zero;
-  }
-
-  int e1=    grid->_slice_nblock[orthogdim];
-  int e2=    grid->_slice_block [orthogdim];
-  int stride=grid->_slice_stride[orthogdim];
-
-  parallel_for(int r=0;r<rd;r++){
-
-    int so=r*grid->_ostride[orthogdim]; // base offset for start of plane 
-
-    for(int n=0;n<e1;n++){
-      for(int b=0;b<e2;b++){
-	int ss= so+n*stride+b;
-	vector_type vv = TensorRemove(innerProduct(lhs._odata[ss],rhs._odata[ss]));
-	lvSum[r]=lvSum[r]+vv;
-      }
-    }
-  }
-
-  // Sum across simd lanes in the plane, breaking out orthog dir.
-  std::vector<int> icoor(Nd);
-  for(int rt=0;rt<rd;rt++){
-
-    iScalar<vector_type> temp; 
-    temp._internal = lvSum[rt];
-    extract(temp,extracted);
-
-    for(int idx=0;idx<Nsimd;idx++){
-
-      grid->iCoorFromIindex(icoor,idx);
-
-      int ldx =rt+icoor[orthogdim]*rd;
-
-      lsSum[ldx]=lsSum[ldx]+extracted[idx]._internal;
-
-    }
-  }
-  
-  // sum over nodes.
-  scalar_type gsum;
-  for(int t=0;t<fd;t++){
-    int pt = t/ld; // processor plane
-    int lt = t%ld;
-    if ( pt == grid->_processor_coor[orthogdim] ) {
-      gsum=lsSum[lt];
-    } else {
-      gsum=scalar_type(0.0);
-    }
-
-    grid->GlobalSum(gsum);
-
-    result[t]=gsum;
-  }
-}
-template<class vobj>
-static void sliceNorm (std::vector<RealD> &sn,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;
-  
-  int Nblock = rhs._grid->GlobalDimensions()[Orthog];
-  std::vector<ComplexD> ip(Nblock);
-  sn.resize(Nblock);
-  
-  sliceInnerProductVector(ip,rhs,rhs,Orthog);
-  for(int ss=0;ss<Nblock;ss++){
-    sn[ss] = real(ip[ss]);
-  }
-};
-
-
-template<class vobj>
-static void sliceMaddVector(Lattice<vobj> &R,std::vector<RealD> &a,const Lattice<vobj> &X,const Lattice<vobj> &Y,
-			    int orthogdim,RealD scale=1.0) 
-{    
-  typedef typename vobj::scalar_object sobj;
-  typedef typename vobj::scalar_type scalar_type;
-  typedef typename vobj::vector_type vector_type;
-  typedef typename vobj::tensor_reduced tensor_reduced;
-  
-  scalar_type zscale(scale);
-
-  GridBase *grid  = X._grid;
-
-  int Nsimd  =grid->Nsimd();
-  int Nblock =grid->GlobalDimensions()[orthogdim];
-
-  int fd     =grid->_fdimensions[orthogdim];
-  int ld     =grid->_ldimensions[orthogdim];
-  int rd     =grid->_rdimensions[orthogdim];
-
-  int e1     =grid->_slice_nblock[orthogdim];
-  int e2     =grid->_slice_block [orthogdim];
-  int stride =grid->_slice_stride[orthogdim];
-
-  std::vector<int> icoor;
-
-  for(int r=0;r<rd;r++){
-
-    int so=r*grid->_ostride[orthogdim]; // base offset for start of plane 
-
-    vector_type    av;
-
-    for(int l=0;l<Nsimd;l++){
-      grid->iCoorFromIindex(icoor,l);
-      int ldx =r+icoor[orthogdim]*rd;
-      scalar_type *as =(scalar_type *)&av;
-      as[l] = scalar_type(a[ldx])*zscale;
-    }
-
-    tensor_reduced at; at=av;
-
-    parallel_for_nest2(int n=0;n<e1;n++){
-      for(int b=0;b<e2;b++){
-	int ss= so+n*stride+b;
-	R._odata[ss] = at*X._odata[ss]+Y._odata[ss];
-      }
-    }
-  }
-};
-
-/*
-inline GridBase         *makeSubSliceGrid(const GridBase *BlockSolverGrid,int Orthog)
-{
-  int NN    = BlockSolverGrid->_ndimension;
-  int nsimd = BlockSolverGrid->Nsimd();
-  
-  std::vector<int> latt_phys(0);
-  std::vector<int> simd_phys(0);
-  std::vector<int>  mpi_phys(0);
-  
-  for(int d=0;d<NN;d++){
-    if( d!=Orthog ) { 
-      latt_phys.push_back(BlockSolverGrid->_fdimensions[d]);
-      simd_phys.push_back(BlockSolverGrid->_simd_layout[d]);
-      mpi_phys.push_back(BlockSolverGrid->_processors[d]);
-    }
-  }
-  return (GridBase *)new GridCartesian(latt_phys,simd_phys,mpi_phys); 
-}
-*/
-
-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);
-  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];
-#pragma omp parallel 
-  {
-    std::vector<vobj> s_x(Nblock);
-
-#pragma omp for collapse(2)
-    for(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[o+i*ostride];
-      }
-
-      vobj dot;
-      for(int i=0;i<Nblock;i++){
-	dot = Y[o+i*ostride];
-	for(int j=0;j<Nblock;j++){
-	  dot = dot + s_x[j]*(scale*aa(j,i));
-	}
-	R[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;
-  //  GridBase *SliceGrid = makeSubSliceGrid(FullGrid,Orthog);
-  //  Lattice<vobj> Xslice(SliceGrid);
-  //  Lattice<vobj> Rslice(SliceGrid);
-
-  assert( FullGrid->_simd_layout[Orthog]==1);
-  int nh =  FullGrid->_ndimension;
-  //  int nl = SliceGrid->_ndimension;
-  int nl=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];
-#pragma omp parallel 
-  {
-    std::vector<vobj> s_x(Nblock);
-
-#pragma omp for collapse(2)
-    for(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[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[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;
-
-#pragma omp parallel 
-  {
-    std::vector<vobj> Left(Nblock);
-    std::vector<vobj> Right(Nblock);
-    Eigen::MatrixXcd  mat_thread = Eigen::MatrixXcd::Zero(Nblock,Nblock);
-
-#pragma omp for collapse(2)
-    for(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[o+i*ostride];
-	Right[i] = rhs[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);
-	mat_thread(i,j) += Reduce(rtmp);
-      }}
-    }}
-#pragma omp 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;
-}
-
-} /*END NAMESPACE GRID*/
-#endif
-
-
-

Grid/lattice/Lattice_rng.h

@@ -1,516 +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 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));
-  }
-  template<class distribution,class generator> 
-  void fillScalar(ComplexD &s,distribution &dist,generator &gen)
-  {
-    s=ComplexD(dist(gen),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;
-
-      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 {
-
-    double _time_counter;
-
-  public:
-    GridBase *_grid;
-    unsigned int _vol;
-
-    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);
-
-      parallel_for(int ss=0;ss<osites;ss++){
-        std::vector<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._odata[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.
-      parallel_for(int gidx=0;gidx<_grid->_gsites;gidx++){
-
-	// Where is it?
-	int rank,o_idx,i_idx;
-	std::vector<int> 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);
-	}
-
-	parallel_for(int t=0;t<Nthread;t++) {
-	  // 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
-      std::vector<int> gcoor;
-      int rank,o_idx,i_idx;
-      _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); }
-
-}
-#endif

Grid/log/Log.cc

@@ -1,116 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/Log.cc
-
-Copyright (C) 2015
-
-Author: Antonin Portelli <antonin.portelli@me.com>
-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 */
-#include <Grid/GridCore.h>
-#include <Grid/util/CompilerCompatible.h>
-
-#include <cxxabi.h>
-#include <memory>
-
-namespace Grid {
-
-  std::string demangle(const char* name) {
-    
-    int status = -4; // some arbitrary value to eliminate the compiler warning
-    
-    // enable c++11 by passing the flag -std=c++11 to g++
-    std::unique_ptr<char, void(*)(void*)> res {
-      abi::__cxa_demangle(name, NULL, NULL, &status),
-	std::free
-	};
-    
-    return (status==0) ? res.get() : name ;
-  }
-  
-GridStopWatch Logger::GlobalStopWatch;
-int Logger::timestamp;
-std::ostream Logger::devnull(0);
-
-void GridLogTimestamp(int on){
-  Logger::Timestamp(on);
-}
-
-Colours GridLogColours(0);
-GridLogger GridLogMG     (1, "MG"    , GridLogColours, "NORMAL");
-GridLogger GridLogIRL    (1, "IRL"   , GridLogColours, "NORMAL");
-GridLogger GridLogSolver (1, "Solver", GridLogColours, "NORMAL");
-GridLogger GridLogError  (1, "Error" , GridLogColours, "RED");
-GridLogger GridLogWarning(1, "Warning", GridLogColours, "YELLOW");
-GridLogger GridLogMessage(1, "Message", GridLogColours, "NORMAL");
-GridLogger GridLogDebug  (1, "Debug", GridLogColours, "PURPLE");
-GridLogger GridLogPerformance(1, "Performance", GridLogColours, "GREEN");
-GridLogger GridLogIterative  (1, "Iterative", GridLogColours, "BLUE");
-GridLogger GridLogIntegrator (1, "Integrator", GridLogColours, "BLUE");
-
-void GridLogConfigure(std::vector<std::string> &logstreams) {
-  GridLogError.Active(0);
-  GridLogWarning.Active(0);
-  GridLogMessage.Active(1); // at least the messages should be always on
-  GridLogIterative.Active(0);
-  GridLogDebug.Active(0);
-  GridLogPerformance.Active(0);
-  GridLogIntegrator.Active(1);
-  GridLogColours.Active(0);
-
-  for (int i = 0; i < logstreams.size(); i++) {
-    if (logstreams[i] == std::string("Error"))       GridLogError.Active(1);
-    if (logstreams[i] == std::string("Warning"))     GridLogWarning.Active(1);
-    if (logstreams[i] == std::string("NoMessage"))   GridLogMessage.Active(0);
-    if (logstreams[i] == std::string("Iterative"))   GridLogIterative.Active(1);
-    if (logstreams[i] == std::string("Debug"))       GridLogDebug.Active(1);
-    if (logstreams[i] == std::string("Performance")) GridLogPerformance.Active(1);
-    if (logstreams[i] == std::string("Integrator"))  GridLogIntegrator.Active(1);
-    if (logstreams[i] == std::string("Colours"))     GridLogColours.Active(1);
-  }
-}
-
-////////////////////////////////////////////////////////////
-// Verbose limiter on MPI tasks
-////////////////////////////////////////////////////////////
-void Grid_quiesce_nodes(void) {
-  int me = 0;
-#if defined(GRID_COMMS_MPI) || defined(GRID_COMMS_MPI3) || defined(GRID_COMMS_MPIT)
-  MPI_Comm_rank(MPI_COMM_WORLD, &me);
-#endif
-#ifdef GRID_COMMS_SHMEM
-  me = shmem_my_pe();
-#endif
-  if (me) {
-    std::cout.setstate(std::ios::badbit);
-  }
-}
-
-void Grid_unquiesce_nodes(void) {
-#ifdef GRID_COMMS_MPI
-  std::cout.clear();
-#endif
-}
-}

Grid/log/Log.h

@@ -1,219 +0,0 @@
-    /*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/Log.h
-
-    Copyright (C) 2015
-
-    Author: Antonin Portelli <antonin.portelli@me.com>
-    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 */
-
-#include <map>
-
-#ifndef GRID_LOG_H
-#define GRID_LOG_H
-
-#ifdef HAVE_EXECINFO_H
-#include <execinfo.h>
-#endif
-
-namespace Grid {
-
-//////////////////////////////////////////////////////////////////////////////////////////////////
-// Dress the output; use std::chrono for time stamping via the StopWatch class
-//////////////////////////////////////////////////////////////////////////////////////////////////
-
-
-class Colours{
-protected:
-  bool is_active;
-public:
-  std::map<std::string, std::string> colour;
-
-  Colours(bool activate=false){
-    Active(activate);
-  };
-
-  void Active(bool activate){
-    is_active=activate;
-    if (is_active){
-     colour["BLACK"]  ="\033[30m";
-     colour["RED"]    ="\033[31m";
-     colour["GREEN"]  ="\033[32m";
-     colour["YELLOW"] ="\033[33m";
-     colour["BLUE"]   ="\033[34m";
-     colour["PURPLE"] ="\033[35m";
-     colour["CYAN"]   ="\033[36m";
-     colour["WHITE"]  ="\033[37m";
-     colour["NORMAL"] ="\033[0;39m";
-    } else {
-      colour["BLACK"] ="";
-      colour["RED"]   ="";
-      colour["GREEN"] ="";
-      colour["YELLOW"]="";
-      colour["BLUE"]  ="";
-      colour["PURPLE"]="";
-      colour["CYAN"]  ="";
-      colour["WHITE"] ="";
-      colour["NORMAL"]="";
-    }
-  };
-};
-
-
-class Logger {
-protected:
-  Colours &Painter;
-  int active;
-  int timing_mode;
-  int topWidth{-1}, chanWidth{-1};
-  static int timestamp;
-  std::string name, topName;
-  std::string COLOUR;
-
-public:
-  static GridStopWatch GlobalStopWatch;
-  GridStopWatch         LocalStopWatch;
-  GridStopWatch *StopWatch;
-  static std::ostream devnull;
-
-  std::string background() {return Painter.colour["NORMAL"];}
-  std::string evidence() {return Painter.colour["YELLOW"];}
-  std::string colour() {return Painter.colour[COLOUR];}
-
-  Logger(std::string topNm, int on, std::string nm, Colours& col_class, std::string col)  : active(on),
-    name(nm),
-    topName(topNm),
-    Painter(col_class),
-    timing_mode(0),
-    COLOUR(col) 
-    {
-      StopWatch = & GlobalStopWatch;
-    };
-  
-  void Active(int on) {active = on;};
-  int  isActive(void) {return active;};
-  static void Timestamp(int on) {timestamp = on;};
-  void Reset(void) { 
-    StopWatch->Reset(); 
-    StopWatch->Start(); 
-  }
-  void TimingMode(int on) { 
-    timing_mode = on; 
-    if(on) { 
-      StopWatch = &LocalStopWatch;
-      Reset(); 
-    }
-  }
-  void setTopWidth(const int w) {topWidth = w;}
-  void setChanWidth(const int w) {chanWidth = w;}
-
-  friend std::ostream& operator<< (std::ostream& stream, Logger& log){
-
-    if ( log.active ) {
-      stream << log.background()<<  std::left;
-      if (log.topWidth > 0)
-      {
-        stream << std::setw(log.topWidth);
-      }
-      stream << log.topName << log.background()<< " : ";
-      stream << log.colour() <<  std::left;
-      if (log.chanWidth > 0)
-      {
-        stream << std::setw(log.chanWidth);
-      }
-      stream << log.name << log.background() << " : ";
-      if ( log.timestamp ) {
-	log.StopWatch->Stop();
-	GridTime now = log.StopWatch->Elapsed();
-	
-	if ( log.timing_mode==1 ) log.StopWatch->Reset();
-	log.StopWatch->Start();
-	stream << log.evidence()
-	       << now	       << log.background() << " : " ;
-      }
-      stream << log.colour();
-      return stream;
-    } else { 
-      return devnull;
-    }
-  }
-
-};
-
-class GridLogger: public Logger {
-public:
-  GridLogger(int on, std::string nm, Colours&col_class, std::string col_key = "NORMAL"):
-  Logger("Grid", on, nm, col_class, col_key){};
-};
-
-void GridLogConfigure(std::vector<std::string> &logstreams);
-
-extern GridLogger GridLogMG;
-extern GridLogger GridLogIRL;
-extern GridLogger GridLogSolver;
-extern GridLogger GridLogError;
-extern GridLogger GridLogWarning;
-extern GridLogger GridLogMessage;
-extern GridLogger GridLogDebug  ;
-extern GridLogger GridLogPerformance;
-extern GridLogger GridLogIterative  ;
-extern GridLogger GridLogIntegrator  ;
-extern Colours    GridLogColours;
-
- std::string demangle(const char* name) ;
-
-#define _NBACKTRACE (256)
-extern void * Grid_backtrace_buffer[_NBACKTRACE];
-
-#define BACKTRACEFILE() {\
-char string[20];					\
-std::sprintf(string,"backtrace.%d",CartesianCommunicator::RankWorld()); \
-std::FILE * fp = std::fopen(string,"w");				\
-BACKTRACEFP(fp)\
-std::fclose(fp);	    \
-}
-
-
-#ifdef HAVE_EXECINFO_H
-#define BACKTRACEFP(fp) { \
-int symbols    = backtrace        (Grid_backtrace_buffer,_NBACKTRACE);\
-char **strings = backtrace_symbols(Grid_backtrace_buffer,symbols);\
-for (int i = 0; i < symbols; i++){\
-  std::fprintf (fp,"BackTrace Strings: %d %s\n",i, demangle(strings[i]).c_str()); std::fflush(fp); \
-}\
-}
-#else 
-#define BACKTRACEFP(fp) { \
-std::fprintf (fp,"BT %d %lx\n",0, __builtin_return_address(0)); std::fflush(fp); \
-std::fprintf (fp,"BT %d %lx\n",1, __builtin_return_address(1)); std::fflush(fp); \
-std::fprintf (fp,"BT %d %lx\n",2, __builtin_return_address(2)); std::fflush(fp); \
-std::fprintf (fp,"BT %d %lx\n",3, __builtin_return_address(3)); std::fflush(fp); \
-}
-#endif
-
-#define BACKTRACE() BACKTRACEFP(stdout) 
-
-
-}
-#endif

Grid/parallelIO/BinaryIO.cc

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

Grid/parallelIO/BinaryIO.h

@@ -1,760 +0,0 @@
-    /*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/parallelIO/BinaryIO.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_BINARY_IO_H
-#define GRID_BINARY_IO_H
-
-#if defined(GRID_COMMS_MPI) || defined(GRID_COMMS_MPI3) || defined(GRID_COMMS_MPIT) 
-#define USE_MPI_IO
-#else
-#undef  USE_MPI_IO
-#endif
-
-#ifdef HAVE_ENDIAN_H
-#include <endian.h>
-#endif
-
-#include <arpa/inet.h>
-#include <algorithm>
-
-namespace Grid { 
-
-
-/////////////////////////////////////////////////////////////////////////////////
-// Byte reversal garbage
-/////////////////////////////////////////////////////////////////////////////////
-inline uint32_t byte_reverse32(uint32_t f) { 
-      f = ((f&0xFF)<<24) | ((f&0xFF00)<<8) | ((f&0xFF0000)>>8) | ((f&0xFF000000UL)>>24) ; 
-      return f;
-}
-inline uint64_t byte_reverse64(uint64_t f) { 
-  uint64_t g;
-  g = ((f&0xFF)<<24) | ((f&0xFF00)<<8) | ((f&0xFF0000)>>8) | ((f&0xFF000000UL)>>24) ; 
-  g = g << 32;
-  f = f >> 32;
-  g|= ((f&0xFF)<<24) | ((f&0xFF00)<<8) | ((f&0xFF0000)>>8) | ((f&0xFF000000UL)>>24) ; 
-  return g;
-}
-
-#if BYTE_ORDER == BIG_ENDIAN 
-inline uint64_t Grid_ntohll(uint64_t A) { return A; }
-#else
-inline uint64_t Grid_ntohll(uint64_t A) { 
-  return byte_reverse64(A);
-}
-#endif
-
-// A little helper
-inline void removeWhitespace(std::string &key)
-{
-  key.erase(std::remove_if(key.begin(), key.end(), ::isspace),key.end());
-}
-
-///////////////////////////////////////////////////////////////////////////////////////////////////
-// Static class holding the parallel IO code
-// Could just use a namespace
-///////////////////////////////////////////////////////////////////////////////////////////////////
-class BinaryIO {
- public:
-  static int latticeWriteMaxRetry;
-
-  /////////////////////////////////////////////////////////////////////////////
-  // more byte manipulation helpers
-  /////////////////////////////////////////////////////////////////////////////
-
-  template<class vobj> static inline void Uint32Checksum(Lattice<vobj> &lat,uint32_t &nersc_csum)
-  {
-    typedef typename vobj::scalar_object sobj;
-
-    GridBase *grid = lat._grid;
-    uint64_t lsites = grid->lSites();
-
-    std::vector<sobj> scalardata(lsites); 
-    unvectorizeToLexOrdArray(scalardata,lat);    
-
-    NerscChecksum(grid,scalardata,nersc_csum);
-  }
-
-  template <class fobj>
-  static inline void NerscChecksum(GridBase *grid, std::vector<fobj> &fbuf, uint32_t &nersc_csum)
-  {
-    const uint64_t size32 = sizeof(fobj) / sizeof(uint32_t);
-
-    uint64_t lsites = grid->lSites();
-    if (fbuf.size() == 1)
-    {
-      lsites = 1;
-    }
-
-PARALLEL_REGION
-    {
-      uint32_t nersc_csum_thr = 0;
-
-PARALLEL_FOR_LOOP_INTERN
-      for (uint64_t local_site = 0; local_site < lsites; local_site++)
-      {
-        uint32_t *site_buf = (uint32_t *)&fbuf[local_site];
-        for (uint64_t j = 0; j < size32; j++)
-        {
-          nersc_csum_thr = nersc_csum_thr + site_buf[j];
-        }
-      }
-
-PARALLEL_CRITICAL
-      {
-        nersc_csum += nersc_csum_thr;
-      }
-    }
-  }
-
-  template<class fobj> static inline void ScidacChecksum(GridBase *grid,std::vector<fobj> &fbuf,uint32_t &scidac_csuma,uint32_t &scidac_csumb)
-  {
-    const uint64_t size32 = sizeof(fobj)/sizeof(uint32_t);
-
-
-    int nd = grid->_ndimension;
-
-    uint64_t lsites              =grid->lSites();
-    if (fbuf.size()==1) {
-      lsites=1;
-    }
-    std::vector<int> local_vol   =grid->LocalDimensions();
-    std::vector<int> local_start =grid->LocalStarts();
-    std::vector<int> global_vol  =grid->FullDimensions();
-
-PARALLEL_REGION
-    { 
-      std::vector<int> coor(nd);
-      uint32_t scidac_csuma_thr=0;
-      uint32_t scidac_csumb_thr=0;
-      uint32_t site_crc=0;
-
-PARALLEL_FOR_LOOP_INTERN
-      for(uint64_t local_site=0;local_site<lsites;local_site++){
-
-	uint32_t * site_buf = (uint32_t *)&fbuf[local_site];
-
-	/* 
-	 * Scidac csum  is rather more heavyweight
-	 * FIXME -- 128^3 x 256 x 16 will overflow.
-	 */
-	
-	int global_site;
-
-	Lexicographic::CoorFromIndex(coor,local_site,local_vol);
-
-	for(int d=0;d<nd;d++) {
-	  coor[d] = coor[d]+local_start[d];
-	}
-
-	Lexicographic::IndexFromCoor(coor,global_site,global_vol);
-
-	uint32_t gsite29   = global_site%29;
-	uint32_t gsite31   = global_site%31;
-	
-	site_crc = crc32(0,(unsigned char *)site_buf,sizeof(fobj));
-	//	std::cout << "Site "<<local_site << " crc "<<std::hex<<site_crc<<std::dec<<std::endl;
-	//	std::cout << "Site "<<local_site << std::hex<<site_buf[0] <<site_buf[1]<<std::dec <<std::endl;
-	scidac_csuma_thr ^= site_crc<<gsite29 | site_crc>>(32-gsite29);
-	scidac_csumb_thr ^= site_crc<<gsite31 | site_crc>>(32-gsite31);
-      }
-
-PARALLEL_CRITICAL
-      {
-	scidac_csuma^= scidac_csuma_thr;
-	scidac_csumb^= scidac_csumb_thr;
-      }
-    }
-  }
-
-  // Network is big endian
-  static inline void htobe32_v(void *file_object,uint32_t bytes){ be32toh_v(file_object,bytes);} 
-  static inline void htobe64_v(void *file_object,uint32_t bytes){ be64toh_v(file_object,bytes);} 
-  static inline void htole32_v(void *file_object,uint32_t bytes){ le32toh_v(file_object,bytes);} 
-  static inline void htole64_v(void *file_object,uint32_t bytes){ le64toh_v(file_object,bytes);} 
-
-  static inline void be32toh_v(void *file_object,uint64_t bytes)
-  {
-    uint32_t * f = (uint32_t *)file_object;
-    uint64_t count = bytes/sizeof(uint32_t);
-    parallel_for(uint64_t i=0;i<count;i++){  
-      f[i] = ntohl(f[i]);
-    }
-  }
-  // LE must Swap and switch to host
-  static inline void le32toh_v(void *file_object,uint64_t bytes)
-  {
-    uint32_t *fp = (uint32_t *)file_object;
-
-    uint64_t count = bytes/sizeof(uint32_t);
-    parallel_for(uint64_t i=0;i<count;i++){  
-      uint32_t f;
-      f = fp[i];
-      // got network order and the network to host
-      f = ((f&0xFF)<<24) | ((f&0xFF00)<<8) | ((f&0xFF0000)>>8) | ((f&0xFF000000UL)>>24) ; 
-      fp[i] = ntohl(f);
-    }
-  }
-
-  // BE is same as network
-  static inline void be64toh_v(void *file_object,uint64_t bytes)
-  {
-    uint64_t * f = (uint64_t *)file_object;
-    uint64_t count = bytes/sizeof(uint64_t);
-    parallel_for(uint64_t i=0;i<count;i++){  
-      f[i] = Grid_ntohll(f[i]);
-    }
-  }
-  
-  // LE must swap and switch;
-  static inline void le64toh_v(void *file_object,uint64_t bytes)
-  {
-    uint64_t *fp = (uint64_t *)file_object;
-    uint64_t count = bytes/sizeof(uint64_t);
-    parallel_for(uint64_t i=0;i<count;i++){  
-      uint64_t f,g;
-      f = fp[i];
-      // got network order and the network to host
-      g = ((f&0xFF)<<24) | ((f&0xFF00)<<8) | ((f&0xFF0000)>>8) | ((f&0xFF000000UL)>>24) ; 
-      g = g << 32;
-      f = f >> 32;
-      g|= ((f&0xFF)<<24) | ((f&0xFF00)<<8) | ((f&0xFF0000)>>8) | ((f&0xFF000000UL)>>24) ; 
-      fp[i] = Grid_ntohll(g);
-    }
-  }
-  /////////////////////////////////////////////////////////////////////////////
-  // Real action:
-  // Read or Write distributed lexico array of ANY object to a specific location in file 
-  //////////////////////////////////////////////////////////////////////////////////////
-
-  static const int BINARYIO_MASTER_APPEND = 0x10;
-  static const int BINARYIO_UNORDERED     = 0x08;
-  static const int BINARYIO_LEXICOGRAPHIC = 0x04;
-  static const int BINARYIO_READ          = 0x02;
-  static const int BINARYIO_WRITE         = 0x01;
-
-  template<class word,class fobj>
-  static inline void IOobject(word w,
-			      GridBase *grid,
-			      std::vector<fobj> &iodata,
-			      std::string file,
-			      uint64_t& offset,
-			      const std::string &format, int control,
-			      uint32_t &nersc_csum,
-			      uint32_t &scidac_csuma,
-			      uint32_t &scidac_csumb)
-  {
-    grid->Barrier();
-    GridStopWatch timer; 
-    GridStopWatch bstimer;
-    
-    nersc_csum=0;
-    scidac_csuma=0;
-    scidac_csumb=0;
-
-    int ndim                 = grid->Dimensions();
-    int nrank                = grid->ProcessorCount();
-    int myrank               = grid->ThisRank();
-
-    std::vector<int>  psizes = grid->ProcessorGrid(); 
-    std::vector<int>  pcoor  = grid->ThisProcessorCoor();
-    std::vector<int> gLattice= grid->GlobalDimensions();
-    std::vector<int> lLattice= grid->LocalDimensions();
-
-    std::vector<int> lStart(ndim);
-    std::vector<int> gStart(ndim);
-
-    // Flatten the file
-    uint64_t lsites = grid->lSites();
-    if ( control & BINARYIO_MASTER_APPEND )  {
-      assert(iodata.size()==1);
-    } else {
-      assert(lsites==iodata.size());
-    }
-    for(int d=0;d<ndim;d++){
-      gStart[d] = lLattice[d]*pcoor[d];
-      lStart[d] = 0;
-    }
-
-#ifdef USE_MPI_IO
-    std::vector<int> distribs(ndim,MPI_DISTRIBUTE_BLOCK);
-    std::vector<int> dargs   (ndim,MPI_DISTRIBUTE_DFLT_DARG);
-    MPI_Datatype mpiObject;
-    MPI_Datatype fileArray;
-    MPI_Datatype localArray;
-    MPI_Datatype mpiword;
-    MPI_Offset disp = offset;
-    MPI_File fh ;
-    MPI_Status status;
-    int numword;
-
-    if ( sizeof( word ) == sizeof(float ) ) {
-      numword = sizeof(fobj)/sizeof(float);
-      mpiword = MPI_FLOAT;
-    } else {
-      numword = sizeof(fobj)/sizeof(double);
-      mpiword = MPI_DOUBLE;
-    }
-
-    //////////////////////////////////////////////////////////////////////////////
-    // Sobj in MPI phrasing
-    //////////////////////////////////////////////////////////////////////////////
-    int ierr;
-    ierr = MPI_Type_contiguous(numword,mpiword,&mpiObject);    assert(ierr==0);
-    ierr = MPI_Type_commit(&mpiObject);
-
-    //////////////////////////////////////////////////////////////////////////////
-    // File global array data type
-    //////////////////////////////////////////////////////////////////////////////
-    ierr=MPI_Type_create_subarray(ndim,&gLattice[0],&lLattice[0],&gStart[0],MPI_ORDER_FORTRAN, mpiObject,&fileArray);    assert(ierr==0);
-    ierr=MPI_Type_commit(&fileArray);    assert(ierr==0);
-
-    //////////////////////////////////////////////////////////////////////////////
-    // local lattice array
-    //////////////////////////////////////////////////////////////////////////////
-    ierr=MPI_Type_create_subarray(ndim,&lLattice[0],&lLattice[0],&lStart[0],MPI_ORDER_FORTRAN, mpiObject,&localArray);    assert(ierr==0);
-    ierr=MPI_Type_commit(&localArray);    assert(ierr==0);
-#endif
-
-    //////////////////////////////////////////////////////////////////////////////
-    // Byte order
-    //////////////////////////////////////////////////////////////////////////////
-    int ieee32big = (format == std::string("IEEE32BIG"));
-    int ieee32    = (format == std::string("IEEE32"));
-    int ieee64big = (format == std::string("IEEE64BIG"));
-    int ieee64    = (format == std::string("IEEE64"));
-    assert(ieee64||ieee32|ieee64big||ieee32big);
-    assert((ieee64+ieee32+ieee64big+ieee32big)==1);
-    //////////////////////////////////////////////////////////////////////////////
-    // Do the I/O
-    //////////////////////////////////////////////////////////////////////////////
-    if ( control & BINARYIO_READ ) { 
-
-      timer.Start();
-
-      if ( (control & BINARYIO_LEXICOGRAPHIC) && (nrank > 1) ) {
-#ifdef USE_MPI_IO
-	std::cout<< GridLogMessage<<"IOobject: MPI read I/O "<< file<< std::endl;
-	ierr=MPI_File_open(grid->communicator,(char *) file.c_str(), MPI_MODE_RDONLY, MPI_INFO_NULL, &fh);    assert(ierr==0);
-	ierr=MPI_File_set_view(fh, disp, mpiObject, fileArray, "native", MPI_INFO_NULL);    assert(ierr==0);
-	ierr=MPI_File_read_all(fh, &iodata[0], 1, localArray, &status);    assert(ierr==0);
-	MPI_File_close(&fh);
-	MPI_Type_free(&fileArray);
-	MPI_Type_free(&localArray);
-#else 
-	assert(0);
-#endif
-      } else {
-	std::cout << GridLogMessage <<"IOobject: C++ read I/O " << file << " : "
-                  << iodata.size() * sizeof(fobj) << " bytes and offset " << offset << std::endl;
-        std::ifstream fin;
-	fin.open(file, std::ios::binary | std::ios::in);
-        if (control & BINARYIO_MASTER_APPEND)
-        {
-          fin.seekg(-sizeof(fobj), fin.end);
-        }
-        else
-        {
-          fin.seekg(offset + myrank * lsites * sizeof(fobj));
-        }
-        fin.read((char *)&iodata[0], iodata.size() * sizeof(fobj));
-        assert(fin.fail() == 0);
-        fin.close();
-      }
-      timer.Stop();
-
-      grid->Barrier();
-
-      bstimer.Start();
-      ScidacChecksum(grid,iodata,scidac_csuma,scidac_csumb);
-      if (ieee32big) be32toh_v((void *)&iodata[0], sizeof(fobj)*iodata.size());
-      if (ieee32)    le32toh_v((void *)&iodata[0], sizeof(fobj)*iodata.size());
-      if (ieee64big) be64toh_v((void *)&iodata[0], sizeof(fobj)*iodata.size());
-      if (ieee64)    le64toh_v((void *)&iodata[0], sizeof(fobj)*iodata.size());
-      NerscChecksum(grid,iodata,nersc_csum);
-      bstimer.Stop();
-    }
-    
-    if ( control & BINARYIO_WRITE ) { 
-
-      bstimer.Start();
-      NerscChecksum(grid,iodata,nersc_csum);
-      if (ieee32big) htobe32_v((void *)&iodata[0], sizeof(fobj)*iodata.size());
-      if (ieee32)    htole32_v((void *)&iodata[0], sizeof(fobj)*iodata.size());
-      if (ieee64big) htobe64_v((void *)&iodata[0], sizeof(fobj)*iodata.size());
-      if (ieee64)    htole64_v((void *)&iodata[0], sizeof(fobj)*iodata.size());
-      ScidacChecksum(grid,iodata,scidac_csuma,scidac_csumb);
-      bstimer.Stop();
-
-      grid->Barrier();
-
-      timer.Start();
-      if ( (control & BINARYIO_LEXICOGRAPHIC) && (nrank > 1) ) {
-#ifdef USE_MPI_IO
-        std::cout << GridLogMessage <<"IOobject: MPI write I/O " << file << std::endl;
-        ierr = MPI_File_open(grid->communicator, (char *)file.c_str(), MPI_MODE_RDWR | MPI_MODE_CREATE, MPI_INFO_NULL, &fh);
-	//        std::cout << GridLogMessage << "Checking for errors" << std::endl;
-        if (ierr != MPI_SUCCESS)
-        {
-          char error_string[BUFSIZ];
-          int length_of_error_string, error_class;
-
-          MPI_Error_class(ierr, &error_class);
-          MPI_Error_string(error_class, error_string, &length_of_error_string);
-          fprintf(stderr, "%3d: %s\n", myrank, error_string);
-          MPI_Error_string(ierr, error_string, &length_of_error_string);
-          fprintf(stderr, "%3d: %s\n", myrank, error_string);
-          MPI_Abort(MPI_COMM_WORLD, 1); //assert(ierr == 0);
-        }
-
-        std::cout << GridLogDebug << "MPI write I/O set view " << file << std::endl;
-        ierr = MPI_File_set_view(fh, disp, mpiObject, fileArray, "native", MPI_INFO_NULL);
-        assert(ierr == 0);
-
-        std::cout << GridLogDebug << "MPI write I/O write all " << file << std::endl;
-        ierr = MPI_File_write_all(fh, &iodata[0], 1, localArray, &status);
-        assert(ierr == 0);
-
-        MPI_Offset os;
-        MPI_File_get_position(fh, &os);
-        MPI_File_get_byte_offset(fh, os, &disp);
-        offset = disp;
-
-
-        MPI_File_close(&fh);
-        MPI_Type_free(&fileArray);
-        MPI_Type_free(&localArray);
-#else 
-	assert(0);
-#endif
-      } else { 
-
-        std::cout << GridLogMessage << "IOobject: C++ write I/O " << file << " : "
-                  << iodata.size() * sizeof(fobj) << " bytes and offset " << offset << std::endl;
-        
-	std::ofstream fout; 
-	fout.exceptions ( std::fstream::failbit | std::fstream::badbit );
-	try {
-	  if (offset) { // Must already exist and contain data
-	    fout.open(file,std::ios::binary|std::ios::out|std::ios::in);
-	  } else {     // Allow create
-	    fout.open(file,std::ios::binary|std::ios::out);
-	  }
-	} catch (const std::fstream::failure& exc) {
-	  std::cout << GridLogError << "Error in opening the file " << file << " for output" <<std::endl;
-	  std::cout << GridLogError << "Exception description: " << exc.what() << std::endl;
-	  //	  std::cout << GridLogError << "Probable cause: wrong path, inaccessible location "<< std::endl;
-#ifdef USE_MPI_IO
-	  MPI_Abort(MPI_COMM_WORLD,1);
-#else
-	  exit(1);
-#endif
-	}
-	
-	if ( control & BINARYIO_MASTER_APPEND )  {
-	  try {
-	    fout.seekp(0,fout.end);
-	  } catch (const std::fstream::failure& exc) {
-	    std::cout << "Exception in seeking file end " << file << std::endl;
-	  }
-	} else {
-	  try { 
-	    fout.seekp(offset+myrank*lsites*sizeof(fobj));
-	  } catch (const std::fstream::failure& exc) {
-	    std::cout << "Exception in seeking file " << file <<" offset "<< offset << std::endl;
-	  }
-	}
-
-	try {
-	  fout.write((char *)&iodata[0],iodata.size()*sizeof(fobj));//assert( fout.fail()==0);
-	}
-	catch (const std::fstream::failure& exc) {
-	  std::cout << "Exception in writing file " << file << std::endl;
-	  std::cout << GridLogError << "Exception description: "<< exc.what() << std::endl;
-#ifdef USE_MPI_IO
-	  MPI_Abort(MPI_COMM_WORLD,1);
-#else
-	  exit(1);
-#endif
-	}
-  offset  = fout.tellp();
-	fout.close();
-      }
-      timer.Stop();
-    }
-    
-    std::cout<<GridLogMessage<<"IOobject: ";
-    if ( control & BINARYIO_READ) std::cout << " read  ";
-    else                          std::cout << " write ";
-    uint64_t bytes = sizeof(fobj)*iodata.size()*nrank;
-    std::cout<< bytes <<" bytes in "<<timer.Elapsed() <<" "
-	     << (double)bytes/ (double)timer.useconds() <<" MB/s "<<std::endl;
-
-    std::cout<<GridLogMessage<<"IOobject: endian and checksum overhead "<<bstimer.Elapsed()  <<std::endl;
-
-    //////////////////////////////////////////////////////////////////////////////
-    // Safety check
-    //////////////////////////////////////////////////////////////////////////////
-    // if the data size is 1 we do not want to sum over the MPI ranks
-    if (iodata.size() != 1){
-      grid->Barrier();
-      grid->GlobalSum(nersc_csum);
-      grid->GlobalXOR(scidac_csuma);
-      grid->GlobalXOR(scidac_csumb);
-      grid->Barrier();
-    }
-  }
-
-  /////////////////////////////////////////////////////////////////////////////
-  // Read a Lattice of object
-  //////////////////////////////////////////////////////////////////////////////////////
-  template<class vobj,class fobj,class munger>
-  static inline void readLatticeObject(Lattice<vobj> &Umu,
-				       std::string file,
-				       munger munge,
-				       uint64_t offset,
-				       const std::string &format,
-				       uint32_t &nersc_csum,
-				       uint32_t &scidac_csuma,
-				       uint32_t &scidac_csumb)
-  {
-    typedef typename vobj::scalar_object sobj;
-    typedef typename vobj::Realified::scalar_type word;    word w=0;
-
-    GridBase *grid = Umu._grid;
-    uint64_t lsites = grid->lSites();
-
-    std::vector<sobj> scalardata(lsites); 
-    std::vector<fobj>     iodata(lsites); // Munge, checksum, byte order in here
-    
-    IOobject(w,grid,iodata,file,offset,format,BINARYIO_READ|BINARYIO_LEXICOGRAPHIC,
-	     nersc_csum,scidac_csuma,scidac_csumb);
-
-    GridStopWatch timer; 
-    timer.Start();
-
-    parallel_for(uint64_t x=0;x<lsites;x++) munge(iodata[x], scalardata[x]);
-
-    vectorizeFromLexOrdArray(scalardata,Umu);    
-    grid->Barrier();
-
-    timer.Stop();
-    std::cout<<GridLogMessage<<"readLatticeObject: vectorize overhead "<<timer.Elapsed()  <<std::endl;
-  }
-
-  /////////////////////////////////////////////////////////////////////////////
-  // Write a Lattice of object
-  //////////////////////////////////////////////////////////////////////////////////////
-  template<class vobj,class fobj,class munger>
-    static inline void writeLatticeObject(Lattice<vobj> &Umu,
-					  std::string file,
-					  munger munge,
-					  uint64_t offset,
-					  const std::string &format,
-					  uint32_t &nersc_csum,
-					  uint32_t &scidac_csuma,
-					  uint32_t &scidac_csumb)
-  {
-    typedef typename vobj::scalar_object sobj;
-    typedef typename vobj::Realified::scalar_type word;    word w=0;
-    GridBase *grid = Umu._grid;
-    uint64_t lsites = grid->lSites(), offsetCopy = offset;
-    int attemptsLeft = std::max(0, BinaryIO::latticeWriteMaxRetry);
-    bool checkWrite = (BinaryIO::latticeWriteMaxRetry >= 0);
-
-    std::vector<sobj> scalardata(lsites); 
-    std::vector<fobj>     iodata(lsites); // Munge, checksum, byte order in here
-
-    //////////////////////////////////////////////////////////////////////////////
-    // Munge [ .e.g 3rd row recon ]
-    //////////////////////////////////////////////////////////////////////////////
-    GridStopWatch timer; timer.Start();
-    unvectorizeToLexOrdArray(scalardata,Umu);    
-
-    parallel_for(uint64_t x=0;x<lsites;x++) munge(scalardata[x],iodata[x]);
-
-    grid->Barrier();
-    timer.Stop();
-    while (attemptsLeft >= 0)
-    {
-      grid->Barrier();
-      IOobject(w,grid,iodata,file,offset,format,BINARYIO_WRITE|BINARYIO_LEXICOGRAPHIC,
-	             nersc_csum,scidac_csuma,scidac_csumb);
-      if (checkWrite)
-      {
-        std::vector<fobj> ckiodata(lsites);
-        uint32_t          cknersc_csum, ckscidac_csuma, ckscidac_csumb;
-        uint64_t          ckoffset = offsetCopy;
-
-        std::cout << GridLogMessage << "writeLatticeObject: read back object" << std::endl;
-        grid->Barrier();
-        IOobject(w,grid,ckiodata,file,ckoffset,format,BINARYIO_READ|BINARYIO_LEXICOGRAPHIC,
-	               cknersc_csum,ckscidac_csuma,ckscidac_csumb);
-        if ((cknersc_csum != nersc_csum) or (ckscidac_csuma != scidac_csuma) or (ckscidac_csumb != scidac_csumb))
-        {
-          std::cout << GridLogMessage << "writeLatticeObject: read test checksum failure, re-writing (" << attemptsLeft << " attempt(s) remaining)" << std::endl;
-          offset = offsetCopy;
-          parallel_for(uint64_t x=0;x<lsites;x++) munge(scalardata[x],iodata[x]);
-        }
-        else
-        {
-          std::cout << GridLogMessage << "writeLatticeObject: read test checksum correct" << std::endl;
-          break;
-        }
-      }
-      attemptsLeft--;
-    }
-    
-
-    std::cout<<GridLogMessage<<"writeLatticeObject: unvectorize overhead "<<timer.Elapsed()  <<std::endl;
-  }
-  
-  /////////////////////////////////////////////////////////////////////////////
-  // Read a RNG;  use IOobject and lexico map to an array of state 
-  //////////////////////////////////////////////////////////////////////////////////////
-  static inline void readRNG(GridSerialRNG &serial,
-			     GridParallelRNG &parallel,
-			     std::string file,
-			     uint64_t offset,
-			     uint32_t &nersc_csum,
-			     uint32_t &scidac_csuma,
-			     uint32_t &scidac_csumb)
-  {
-    typedef typename GridSerialRNG::RngStateType RngStateType;
-    const int RngStateCount = GridSerialRNG::RngStateCount;
-    typedef std::array<RngStateType,RngStateCount> RNGstate;
-    typedef RngStateType word;    word w=0;
-
-    std::string format = "IEEE32BIG";
-
-    GridBase *grid = parallel._grid;
-    uint64_t gsites = grid->gSites();
-    uint64_t lsites = grid->lSites();
-
-    uint32_t nersc_csum_tmp   = 0;
-    uint32_t scidac_csuma_tmp = 0;
-    uint32_t scidac_csumb_tmp = 0;
-
-    GridStopWatch timer;
-
-    std::cout << GridLogMessage << "RNG read I/O on file " << file << std::endl;
-
-    std::vector<RNGstate> iodata(lsites);
-    IOobject(w,grid,iodata,file,offset,format,BINARYIO_READ|BINARYIO_LEXICOGRAPHIC,
-	     nersc_csum,scidac_csuma,scidac_csumb);
-
-    timer.Start();
-    parallel_for(uint64_t lidx=0;lidx<lsites;lidx++){
-      std::vector<RngStateType> tmp(RngStateCount);
-      std::copy(iodata[lidx].begin(),iodata[lidx].end(),tmp.begin());
-      parallel.SetState(tmp,lidx);
-    }
-    timer.Stop();
-
-    iodata.resize(1);
-    IOobject(w,grid,iodata,file,offset,format,BINARYIO_READ|BINARYIO_MASTER_APPEND,
-	     nersc_csum_tmp,scidac_csuma_tmp,scidac_csumb_tmp);
-
-    {
-      std::vector<RngStateType> tmp(RngStateCount);
-      std::copy(iodata[0].begin(),iodata[0].end(),tmp.begin());
-      serial.SetState(tmp,0);
-    }
-
-    nersc_csum   = nersc_csum   + nersc_csum_tmp;
-    scidac_csuma = scidac_csuma ^ scidac_csuma_tmp;
-    scidac_csumb = scidac_csumb ^ scidac_csumb_tmp;
-
-    std::cout << GridLogMessage << "RNG file nersc_checksum   " << std::hex << nersc_csum << std::dec << std::endl;
-    std::cout << GridLogMessage << "RNG file scidac_checksuma " << std::hex << scidac_csuma << std::dec << std::endl;
-    std::cout << GridLogMessage << "RNG file scidac_checksumb " << std::hex << scidac_csumb << std::dec << std::endl;
-
-    std::cout << GridLogMessage << "RNG state overhead " << timer.Elapsed() << std::endl;
-  }
-  /////////////////////////////////////////////////////////////////////////////
-  // Write a RNG; lexico map to an array of state and use IOobject
-  //////////////////////////////////////////////////////////////////////////////////////
-  static inline void writeRNG(GridSerialRNG &serial,
-			      GridParallelRNG &parallel,
-			      std::string file,
-			      uint64_t offset,
-			      uint32_t &nersc_csum,
-			      uint32_t &scidac_csuma,
-			      uint32_t &scidac_csumb)
-  {
-    typedef typename GridSerialRNG::RngStateType RngStateType;
-    typedef RngStateType word; word w=0;
-    const int RngStateCount = GridSerialRNG::RngStateCount;
-    typedef std::array<RngStateType,RngStateCount> RNGstate;
-
-    GridBase *grid = parallel._grid;
-    uint64_t gsites = grid->gSites();
-    uint64_t lsites = grid->lSites();
-
-    uint32_t nersc_csum_tmp;
-    uint32_t scidac_csuma_tmp;
-    uint32_t scidac_csumb_tmp;
-
-    GridStopWatch timer;
-    std::string format = "IEEE32BIG";
-
-    std::cout << GridLogMessage << "RNG write I/O on file " << file << std::endl;
-
-    timer.Start();
-    std::vector<RNGstate> iodata(lsites);
-    parallel_for(uint64_t lidx=0;lidx<lsites;lidx++){
-      std::vector<RngStateType> tmp(RngStateCount);
-      parallel.GetState(tmp,lidx);
-      std::copy(tmp.begin(),tmp.end(),iodata[lidx].begin());
-    }
-    timer.Stop();
-
-    IOobject(w,grid,iodata,file,offset,format,BINARYIO_WRITE|BINARYIO_LEXICOGRAPHIC,
-	     nersc_csum,scidac_csuma,scidac_csumb);
-    iodata.resize(1);
-    {
-      std::vector<RngStateType> tmp(RngStateCount);
-      serial.GetState(tmp,0);
-      std::copy(tmp.begin(),tmp.end(),iodata[0].begin());
-    }
-    IOobject(w,grid,iodata,file,offset,format,BINARYIO_WRITE|BINARYIO_MASTER_APPEND,
-	     nersc_csum_tmp,scidac_csuma_tmp,scidac_csumb_tmp);
-
-    nersc_csum   = nersc_csum   + nersc_csum_tmp;
-    scidac_csuma = scidac_csuma ^ scidac_csuma_tmp;
-    scidac_csumb = scidac_csumb ^ scidac_csumb_tmp;
-    
-    std::cout << GridLogMessage << "RNG file checksum " << std::hex << nersc_csum    << std::dec << std::endl;
-    std::cout << GridLogMessage << "RNG file checksuma " << std::hex << scidac_csuma << std::dec << std::endl;
-    std::cout << GridLogMessage << "RNG file checksumb " << std::hex << scidac_csumb << std::dec << std::endl;
-    std::cout << GridLogMessage << "RNG state overhead " << timer.Elapsed() << std::endl;
-  }
-};
-
-}
-#endif

Grid/parallelIO/IldgIO.h

@@ -1,938 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/parallelIO/IldgIO.h
-
-Copyright (C) 2015
-
-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_ILDG_IO_H
-#define GRID_ILDG_IO_H
-
-#ifdef HAVE_LIME
-#include <algorithm>
-#include <fstream>
-#include <iomanip>
-#include <iostream>
-#include <map>
-
-#include <pwd.h>
-#include <sys/utsname.h>
-#include <unistd.h>
-
-//C-Lime is a must have for this functionality
-extern "C" {  
-#include "lime.h"
-}
-
-namespace Grid {
-namespace QCD {
-
-#define GRID_FIELD_NORM "FieldNormMetaData"
-#define GRID_FIELD_NORM_CALC(FieldNormMetaData_, n2ck) \
-0.5*fabs(FieldNormMetaData_.norm2 - n2ck)/(FieldNormMetaData_.norm2 + n2ck)
-#define GRID_FIELD_NORM_CHECK(FieldNormMetaData_, n2ck) \
-assert(GRID_FIELD_NORM_CALC(FieldNormMetaData_, n2ck) < 1.0e-5);
-
-  /////////////////////////////////
-  // Encode word types as strings
-  /////////////////////////////////
- template<class word> inline std::string ScidacWordMnemonic(void){ return std::string("unknown"); }
- template<> inline std::string ScidacWordMnemonic<double>  (void){ return std::string("D"); }
- template<> inline std::string ScidacWordMnemonic<float>   (void){ return std::string("F"); }
- template<> inline std::string ScidacWordMnemonic< int32_t>(void){ return std::string("I32_t"); }
- template<> inline std::string ScidacWordMnemonic<uint32_t>(void){ return std::string("U32_t"); }
- template<> inline std::string ScidacWordMnemonic< int64_t>(void){ return std::string("I64_t"); }
- template<> inline std::string ScidacWordMnemonic<uint64_t>(void){ return std::string("U64_t"); }
-
-  /////////////////////////////////////////
-  // Encode a generic tensor as a string
-  /////////////////////////////////////////
- template<class vobj> std::string ScidacRecordTypeString(int &colors, int &spins, int & typesize,int &datacount) { 
-
-   typedef typename getPrecision<vobj>::real_scalar_type stype;
-
-   int _ColourN       = indexRank<ColourIndex,vobj>();
-   int _ColourScalar  =  isScalar<ColourIndex,vobj>();
-   int _ColourVector  =  isVector<ColourIndex,vobj>();
-   int _ColourMatrix  =  isMatrix<ColourIndex,vobj>();
-
-   int _SpinN       = indexRank<SpinIndex,vobj>();
-   int _SpinScalar  =  isScalar<SpinIndex,vobj>();
-   int _SpinVector  =  isVector<SpinIndex,vobj>();
-   int _SpinMatrix  =  isMatrix<SpinIndex,vobj>();
-
-   int _LorentzN       = indexRank<LorentzIndex,vobj>();
-   int _LorentzScalar  =  isScalar<LorentzIndex,vobj>();
-   int _LorentzVector  =  isVector<LorentzIndex,vobj>();
-   int _LorentzMatrix  =  isMatrix<LorentzIndex,vobj>();
-
-   std::stringstream stream;
-
-   stream << "GRID_";
-   stream << ScidacWordMnemonic<stype>();
-
-   if ( _LorentzVector )   stream << "_LorentzVector"<<_LorentzN;
-   if ( _LorentzMatrix )   stream << "_LorentzMatrix"<<_LorentzN;
-
-   if ( _SpinVector )   stream << "_SpinVector"<<_SpinN;
-   if ( _SpinMatrix )   stream << "_SpinMatrix"<<_SpinN;
-
-   if ( _ColourVector )   stream << "_ColourVector"<<_ColourN;
-   if ( _ColourMatrix )   stream << "_ColourMatrix"<<_ColourN;
-
-   if ( _ColourScalar && _LorentzScalar && _SpinScalar )   stream << "_Complex";
-
-
-   typesize = sizeof(typename vobj::scalar_type);
-
-   if ( _ColourMatrix ) typesize*= _ColourN*_ColourN;
-   else                 typesize*= _ColourN;
-
-   if ( _SpinMatrix )   typesize*= _SpinN*_SpinN;
-   else                 typesize*= _SpinN;
-
-   colors    = _ColourN;
-   spins     = _SpinN;
-   datacount = _LorentzN;
-
-   return stream.str();
- }
- 
- template<class vobj> std::string ScidacRecordTypeString(Lattice<vobj> & lat,int &colors, int &spins, int & typesize,int &datacount) { 
-   return ScidacRecordTypeString<vobj>(colors,spins,typesize,datacount);
- };
-
-
- ////////////////////////////////////////////////////////////
- // Helper to fill out metadata
- ////////////////////////////////////////////////////////////
- template<class vobj> void ScidacMetaData(Lattice<vobj> & field,
-					  FieldMetaData &header,
-					  scidacRecord & _scidacRecord,
-					  scidacFile   & _scidacFile) 
- {
-   typedef typename getPrecision<vobj>::real_scalar_type stype;
-
-   /////////////////////////////////////
-   // Pull Grid's metadata
-   /////////////////////////////////////
-   PrepareMetaData(field,header);
-
-   /////////////////////////////////////
-   // Scidac Private File structure
-   /////////////////////////////////////
-   _scidacFile              = scidacFile(field._grid);
-
-   /////////////////////////////////////
-   // Scidac Private Record structure
-   /////////////////////////////////////
-   scidacRecord sr;
-   sr.datatype   = ScidacRecordTypeString(field,sr.colors,sr.spins,sr.typesize,sr.datacount);
-   sr.date       = header.creation_date;
-   sr.precision  = ScidacWordMnemonic<stype>();
-   sr.recordtype = GRID_IO_FIELD;
-
-   _scidacRecord = sr;
-
-   //   std::cout << GridLogMessage << "Build SciDAC datatype " <<sr.datatype<<std::endl;
- }
- 
- ///////////////////////////////////////////////////////
- // Scidac checksum
- ///////////////////////////////////////////////////////
- static int scidacChecksumVerify(scidacChecksum &scidacChecksum_,uint32_t scidac_csuma,uint32_t scidac_csumb)
- {
-   uint32_t scidac_checksuma = stoull(scidacChecksum_.suma,0,16);
-   uint32_t scidac_checksumb = stoull(scidacChecksum_.sumb,0,16);
-   if ( scidac_csuma !=scidac_checksuma) return 0;
-   if ( scidac_csumb !=scidac_checksumb) return 0;
-   return 1;
- }
-
-////////////////////////////////////////////////////////////////////////////////////
-// Lime, ILDG and Scidac I/O classes
-////////////////////////////////////////////////////////////////////////////////////
-class GridLimeReader : public BinaryIO {
- public:
-   ///////////////////////////////////////////////////
-   // FIXME: format for RNG? Now just binary out instead
-   ///////////////////////////////////////////////////
-
-   FILE       *File;
-   LimeReader *LimeR;
-   std::string filename;
-
-   /////////////////////////////////////////////
-   // Open the file
-   /////////////////////////////////////////////
-   void open(const std::string &_filename) 
-   {
-     filename= _filename;
-     File = fopen(filename.c_str(), "r");
-     if (File == nullptr)
-     {
-       std::cerr << "cannot open file '" << filename << "'" << std::endl;
-       abort();
-     }
-     LimeR = limeCreateReader(File);
-   }
-   /////////////////////////////////////////////
-   // Close the file
-   /////////////////////////////////////////////
-   void close(void){
-     fclose(File);
-     //     limeDestroyReader(LimeR);
-   }
-
-  ////////////////////////////////////////////
-  // Read a generic lattice field and verify checksum
-  ////////////////////////////////////////////
-  template<class vobj>
-  void readLimeLatticeBinaryObject(Lattice<vobj> &field,std::string record_name)
-  {
-    typedef typename vobj::scalar_object sobj;
-    scidacChecksum scidacChecksum_;
-    FieldNormMetaData  FieldNormMetaData_;
-    uint32_t nersc_csum,scidac_csuma,scidac_csumb;
-
-    std::string format = getFormatString<vobj>();
-
-    while ( limeReaderNextRecord(LimeR) == LIME_SUCCESS ) { 
-
-      uint64_t file_bytes =limeReaderBytes(LimeR);
-
-      //      std::cout << GridLogMessage << limeReaderType(LimeR) << " "<< file_bytes <<" bytes "<<std::endl;
-      //      std::cout << GridLogMessage<< " readLimeObject seeking "<<  record_name <<" found record :" <<limeReaderType(LimeR) <<std::endl;
-
-      if ( !strncmp(limeReaderType(LimeR), record_name.c_str(),strlen(record_name.c_str()) )  ) {
-
-	//	std::cout << GridLogMessage<< " readLimeLatticeBinaryObject matches ! " <<std::endl;
-
-	uint64_t PayloadSize = sizeof(sobj) * field._grid->_gsites;
-
-	//	std::cout << "R sizeof(sobj)= " <<sizeof(sobj)<<std::endl;
-	//	std::cout << "R Gsites " <<field._grid->_gsites<<std::endl;
-	//	std::cout << "R Payload expected " <<PayloadSize<<std::endl;
-	//	std::cout << "R file size " <<file_bytes <<std::endl;
-
-	assert(PayloadSize == file_bytes);// Must match or user error
-
-	uint64_t offset= ftello(File);
-	//	std::cout << " ReadLatticeObject from offset "<<offset << std::endl;
-	BinarySimpleMunger<sobj,sobj> munge;
-	BinaryIO::readLatticeObject< vobj, sobj >(field, filename, munge, offset, format,nersc_csum,scidac_csuma,scidac_csumb);
-	std::cout << GridLogMessage << "SciDAC checksum A " << std::hex << scidac_csuma << std::dec << std::endl;
-	std::cout << GridLogMessage << "SciDAC checksum B " << std::hex << scidac_csumb << std::dec << std::endl;
-	/////////////////////////////////////////////
-	// Insist checksum is next record
-	/////////////////////////////////////////////
-	readScidacChecksum(scidacChecksum_,FieldNormMetaData_);
-	/////////////////////////////////////////////
-	// Verify checksums
-	/////////////////////////////////////////////
-	if(FieldNormMetaData_.norm2 != 0.0){ 
-	  RealD n2ck = norm2(field);
-	  std::cout << GridLogMessage << "Field norm: metadata= " << FieldNormMetaData_.norm2 
-              << " / field= " << n2ck << " / rdiff= " << GRID_FIELD_NORM_CALC(FieldNormMetaData_,n2ck) << std::endl;
-	  GRID_FIELD_NORM_CHECK(FieldNormMetaData_,n2ck);
-	}
-	assert(scidacChecksumVerify(scidacChecksum_,scidac_csuma,scidac_csumb)==1);
-
-	// find out if next field is a GridFieldNorm
-	return;
-      }
-    }
-  }
-  void readScidacChecksum(scidacChecksum     &scidacChecksum_,
-			  FieldNormMetaData  &FieldNormMetaData_)
-  {
-    FieldNormMetaData_.norm2 =0.0;
-    std::string scidac_str(SCIDAC_CHECKSUM);
-    std::string field_norm_str(GRID_FIELD_NORM);
-    while ( limeReaderNextRecord(LimeR) == LIME_SUCCESS ) { 
-      uint64_t nbytes = limeReaderBytes(LimeR);//size of this record (configuration)
-      std::vector<char> xmlc(nbytes+1,'\0');
-      limeReaderReadData((void *)&xmlc[0], &nbytes, LimeR);    
-      std::string xmlstring = std::string(&xmlc[0]);
-      XmlReader RD(xmlstring, true, "");
-      if ( !strncmp(limeReaderType(LimeR), field_norm_str.c_str(),strlen(field_norm_str.c_str()) )  ) {
-	//	std::cout << "FieldNormMetaData "<<xmlstring<<std::endl;
-	read(RD,field_norm_str,FieldNormMetaData_);
-      }
-      if ( !strncmp(limeReaderType(LimeR), scidac_str.c_str(),strlen(scidac_str.c_str()) )  ) {
-	//	std::cout << SCIDAC_CHECKSUM << " " <<xmlstring<<std::endl;
-	read(RD,std::string("scidacChecksum"),scidacChecksum_);
-	return;
-      }      
-    }
-    assert(0);
-  }
-  ////////////////////////////////////////////
-  // Read a generic serialisable object
-  ////////////////////////////////////////////
-  void readLimeObject(std::string &xmlstring,std::string record_name)
-  {
-    // should this be a do while; can we miss a first record??
-    while ( limeReaderNextRecord(LimeR) == LIME_SUCCESS ) { 
-
-      //      std::cout << GridLogMessage<< " readLimeObject seeking "<< record_name <<" found record :" <<limeReaderType(LimeR) <<std::endl;
-      uint64_t nbytes = limeReaderBytes(LimeR);//size of this record (configuration)
-
-      if ( !strncmp(limeReaderType(LimeR), record_name.c_str(),strlen(record_name.c_str()) )  ) {
-
-	//	std::cout << GridLogMessage<< " readLimeObject matches ! " << record_name <<std::endl;
-	std::vector<char> xmlc(nbytes+1,'\0');
-	limeReaderReadData((void *)&xmlc[0], &nbytes, LimeR);    
-	//	std::cout << GridLogMessage<< " readLimeObject matches XML " << &xmlc[0] <<std::endl;
-
-	xmlstring = std::string(&xmlc[0]);
-	return;
-      }
-
-    }  
-    assert(0);
-  }
-
-  template<class serialisable_object>
-  void readLimeObject(serialisable_object &object,std::string object_name,std::string record_name)
-  {
-    std::string xmlstring;
-
-    readLimeObject(xmlstring, record_name);
-    XmlReader RD(xmlstring, true, "");
-    read(RD,object_name,object);
-  }
-};
-
-class GridLimeWriter : public BinaryIO 
-{
- public:
-
-   ///////////////////////////////////////////////////
-   // FIXME: format for RNG? Now just binary out instead
-   // FIXME: collective calls or not ?
-   //      : must know if I am the I/O boss
-   ///////////////////////////////////////////////////
-   FILE       *File;
-   LimeWriter *LimeW;
-   std::string filename;
-   bool        boss_node;
-   GridLimeWriter( bool isboss = true) {
-     boss_node = isboss;
-   }
-   void open(const std::string &_filename) { 
-     filename= _filename;
-     if ( boss_node ) {
-       File = fopen(filename.c_str(), "w");
-       LimeW = limeCreateWriter(File); assert(LimeW != NULL );
-     }
-   }
-   /////////////////////////////////////////////
-   // Close the file
-   /////////////////////////////////////////////
-   void close(void) {
-     if ( boss_node ) {
-       fclose(File);
-     }
-     //  limeDestroyWriter(LimeW);
-   }
-  ///////////////////////////////////////////////////////
-  // Lime utility functions
-  ///////////////////////////////////////////////////////
-  int createLimeRecordHeader(std::string message, int MB, int ME, size_t PayloadSize)
-  {
-    if ( boss_node ) {
-      LimeRecordHeader *h;
-      h = limeCreateHeader(MB, ME, const_cast<char *>(message.c_str()), PayloadSize);
-      assert(limeWriteRecordHeader(h, LimeW) >= 0);
-      limeDestroyHeader(h);
-    }
-    return LIME_SUCCESS;
-  }
-  ////////////////////////////////////////////
-  // Write a generic serialisable object
-  ////////////////////////////////////////////
-  void writeLimeObject(int MB,int ME,XmlWriter &writer,std::string object_name,std::string record_name)
-  {
-    if ( boss_node ) {
-      std::string xmlstring = writer.docString();
-
-      //    std::cout << "WriteLimeObject" << record_name <<std::endl;
-      uint64_t nbytes = xmlstring.size();
-      //    std::cout << " xmlstring "<< nbytes<< " " << xmlstring <<std::endl;
-      int err;
-      LimeRecordHeader *h = limeCreateHeader(MB, ME,const_cast<char *>(record_name.c_str()), nbytes); 
-      assert(h!= NULL);
-      
-      err=limeWriteRecordHeader(h, LimeW);                    assert(err>=0);
-      err=limeWriteRecordData(&xmlstring[0], &nbytes, LimeW); assert(err>=0);
-      err=limeWriterCloseRecord(LimeW);                       assert(err>=0);
-      limeDestroyHeader(h);
-    }
-  }
-
-  template<class serialisable_object>
-  void writeLimeObject(int MB,int ME,serialisable_object &object,std::string object_name,std::string record_name, const unsigned int scientificPrec = 0)
-  {
-    XmlWriter WR("","");
-
-    if (scientificPrec)
-    {
-      WR.scientificFormat(true);
-      WR.setPrecision(scientificPrec);
-    }
-    write(WR,object_name,object);
-    writeLimeObject(MB, ME, WR, object_name, record_name);
-  }
-  ////////////////////////////////////////////////////
-  // Write a generic lattice field and csum
-  // This routine is Collectively called by all nodes
-  // in communicator used by the field._grid
-  ////////////////////////////////////////////////////
-  template<class vobj>
-  void writeLimeLatticeBinaryObject(Lattice<vobj> &field,std::string record_name)
-  {
-    ////////////////////////////////////////////////////////////////////
-    // NB: FILE and iostream are jointly writing disjoint sequences in the
-    // the same file through different file handles (integer units).
-    // 
-    // These are both buffered, so why I think this code is right is as follows.
-    //
-    // i)  write record header to FILE *File, telegraphing the size; flush
-    // ii) ftello reads the offset from FILE *File . 
-    // iii) iostream / MPI Open independently seek this offset. Write sequence direct to disk.
-    //      Closes iostream and flushes.
-    // iv) fseek on FILE * to end of this disjoint section.
-    //  v) Continue writing scidac record.
-    ////////////////////////////////////////////////////////////////////
-    
-    GridBase *grid = field._grid;
-    assert(boss_node == field._grid->IsBoss() );
-
-    FieldNormMetaData FNMD; FNMD.norm2 = norm2(field);
-
-    ////////////////////////////////////////////
-    // Create record header
-    ////////////////////////////////////////////
-    typedef typename vobj::scalar_object sobj;
-    int err;
-    uint32_t nersc_csum,scidac_csuma,scidac_csumb;
-    uint64_t PayloadSize = sizeof(sobj) * grid->_gsites;
-    if ( boss_node ) {
-      createLimeRecordHeader(record_name, 0, 0, PayloadSize);
-      fflush(File);
-    }
-    
-    //    std::cout << "W sizeof(sobj)"      <<sizeof(sobj)<<std::endl;
-    //    std::cout << "W Gsites "           <<field._grid->_gsites<<std::endl;
-    //    std::cout << "W Payload expected " <<PayloadSize<<std::endl;
-
-    ////////////////////////////////////////////////
-    // Check all nodes agree on file position
-    ////////////////////////////////////////////////
-    uint64_t offset1;
-    if ( boss_node ) {
-      offset1 = ftello(File);    
-    }
-    grid->Broadcast(0,(void *)&offset1,sizeof(offset1));
-
-    ///////////////////////////////////////////
-    // The above is collective. Write by other means into the binary record
-    ///////////////////////////////////////////
-    std::string format = getFormatString<vobj>();
-    BinarySimpleMunger<sobj,sobj> munge;
-    BinaryIO::writeLatticeObject<vobj,sobj>(field, filename, munge, offset1, format,nersc_csum,scidac_csuma,scidac_csumb);
-
-    ///////////////////////////////////////////
-    // Wind forward and close the record
-    ///////////////////////////////////////////
-    if ( boss_node ) {
-      fseek(File,0,SEEK_END);             
-      uint64_t offset2 = ftello(File);     //    std::cout << " now at offset "<<offset2 << std::endl;
-      assert( (offset2-offset1) == PayloadSize);
-    }
-
-    /////////////////////////////////////////////////////////////
-    // Check MPI-2 I/O did what we expect to file
-    /////////////////////////////////////////////////////////////
-
-    if ( boss_node ) { 
-      err=limeWriterCloseRecord(LimeW);  assert(err>=0);
-    }
-    ////////////////////////////////////////
-    // Write checksum element, propagaing forward from the BinaryIO
-    // Always pair a checksum with a binary object, and close message
-    ////////////////////////////////////////
-    scidacChecksum checksum;
-    std::stringstream streama; streama << std::hex << scidac_csuma;
-    std::stringstream streamb; streamb << std::hex << scidac_csumb;
-    checksum.suma= streama.str();
-    checksum.sumb= streamb.str();
-    if ( boss_node ) { 
-      writeLimeObject(0,0,FNMD,std::string(GRID_FIELD_NORM),std::string(GRID_FIELD_NORM));
-      writeLimeObject(0,1,checksum,std::string("scidacChecksum"),std::string(SCIDAC_CHECKSUM));
-    }
-  }
-};
-
-class ScidacWriter : public GridLimeWriter {
- public:
-
-  ScidacWriter(bool isboss =true ) : GridLimeWriter(isboss)  { };
-
-  template<class SerialisableUserFile>
-  void writeScidacFileRecord(GridBase *grid,SerialisableUserFile &_userFile)
-  {
-    scidacFile    _scidacFile(grid);
-    if ( this->boss_node ) {
-      writeLimeObject(1,0,_scidacFile,_scidacFile.SerialisableClassName(),std::string(SCIDAC_PRIVATE_FILE_XML));
-      writeLimeObject(0,1,_userFile,_userFile.SerialisableClassName(),std::string(SCIDAC_FILE_XML));
-    }
-  }
-  ////////////////////////////////////////////////
-  // Write generic lattice field in scidac format
-  ////////////////////////////////////////////////
-  template <class vobj, class userRecord>
-  void writeScidacFieldRecord(Lattice<vobj> &field,userRecord _userRecord,
-                              const unsigned int recordScientificPrec = 0) 
-  {
-    GridBase * grid = field._grid;
-
-    ////////////////////////////////////////
-    // fill the Grid header
-    ////////////////////////////////////////
-    FieldMetaData header;
-    scidacRecord  _scidacRecord;
-    scidacFile    _scidacFile;
-
-    ScidacMetaData(field,header,_scidacRecord,_scidacFile);
-
-    //////////////////////////////////////////////
-    // Fill the Lime file record by record
-    //////////////////////////////////////////////
-    if ( this->boss_node ) {
-      writeLimeObject(1,0,header ,std::string("FieldMetaData"),std::string(GRID_FORMAT)); // Open message 
-      writeLimeObject(0,0,_userRecord,_userRecord.SerialisableClassName(),std::string(SCIDAC_RECORD_XML), recordScientificPrec);
-      writeLimeObject(0,0,_scidacRecord,_scidacRecord.SerialisableClassName(),std::string(SCIDAC_PRIVATE_RECORD_XML));
-    }
-    // Collective call
-    writeLimeLatticeBinaryObject(field,std::string(ILDG_BINARY_DATA));      // Closes message with checksum
-  }
-};
-
-
-class ScidacReader : public GridLimeReader {
- public:
-
-   template<class SerialisableUserFile>
-   void readScidacFileRecord(GridBase *grid,SerialisableUserFile &_userFile)
-   {
-     scidacFile    _scidacFile(grid);
-     readLimeObject(_scidacFile,_scidacFile.SerialisableClassName(),std::string(SCIDAC_PRIVATE_FILE_XML));
-     readLimeObject(_userFile,_userFile.SerialisableClassName(),std::string(SCIDAC_FILE_XML));
-   }
-  ////////////////////////////////////////////////
-  // Write generic lattice field in scidac format
-  ////////////////////////////////////////////////
-  template <class vobj, class userRecord>
-  void readScidacFieldRecord(Lattice<vobj> &field,userRecord &_userRecord) 
-  {
-    typedef typename vobj::scalar_object sobj;
-    GridBase * grid = field._grid;
-
-    ////////////////////////////////////////
-    // fill the Grid header
-    ////////////////////////////////////////
-    FieldMetaData header;
-    scidacRecord  _scidacRecord;
-    scidacFile    _scidacFile;
-
-    //////////////////////////////////////////////
-    // Fill the Lime file record by record
-    //////////////////////////////////////////////
-    readLimeObject(header ,std::string("FieldMetaData"),std::string(GRID_FORMAT)); // Open message 
-    readLimeObject(_userRecord,_userRecord.SerialisableClassName(),std::string(SCIDAC_RECORD_XML));
-    readLimeObject(_scidacRecord,_scidacRecord.SerialisableClassName(),std::string(SCIDAC_PRIVATE_RECORD_XML));
-    readLimeLatticeBinaryObject(field,std::string(ILDG_BINARY_DATA));
-  }
-  void skipPastBinaryRecord(void) {
-    std::string rec_name(ILDG_BINARY_DATA);
-    while ( limeReaderNextRecord(LimeR) == LIME_SUCCESS ) { 
-      if ( !strncmp(limeReaderType(LimeR), rec_name.c_str(),strlen(rec_name.c_str()) )  ) {
-	skipPastObjectRecord(std::string(SCIDAC_CHECKSUM));
-	return;
-      }
-    }    
-  }
-  void skipPastObjectRecord(std::string rec_name) {
-    while ( limeReaderNextRecord(LimeR) == LIME_SUCCESS ) { 
-      if ( !strncmp(limeReaderType(LimeR), rec_name.c_str(),strlen(rec_name.c_str()) )  ) {
-	return;
-      }
-    }
-  }
-  void skipScidacFieldRecord() {
-    skipPastObjectRecord(std::string(GRID_FORMAT));
-    skipPastObjectRecord(std::string(SCIDAC_RECORD_XML));
-    skipPastObjectRecord(std::string(SCIDAC_PRIVATE_RECORD_XML));
-    skipPastBinaryRecord();
-  }
-};
-
-
-class IldgWriter : public ScidacWriter {
- public:
-  
-  IldgWriter(bool isboss) : ScidacWriter(isboss) {};
-
-  ///////////////////////////////////
-  // A little helper
-  ///////////////////////////////////
-  void writeLimeIldgLFN(std::string &LFN)
-  {
-    uint64_t PayloadSize = LFN.size();
-    int err;
-    createLimeRecordHeader(ILDG_DATA_LFN, 0 , 0, PayloadSize);
-    err=limeWriteRecordData(const_cast<char*>(LFN.c_str()), &PayloadSize,LimeW); assert(err>=0);
-    err=limeWriterCloseRecord(LimeW); assert(err>=0);
-  }
-
-  ////////////////////////////////////////////////////////////////
-  // Special ILDG operations ; gauge configs only.
-  // Don't require scidac records EXCEPT checksum
-  // Use Grid MetaData object if present.
-  ////////////////////////////////////////////////////////////////
-  template <class vsimd>
-  void writeConfiguration(Lattice<iLorentzColourMatrix<vsimd> > &Umu,int sequence,std::string LFN,std::string description) 
-  {
-    GridBase * grid = Umu._grid;
-    typedef Lattice<iLorentzColourMatrix<vsimd> > GaugeField;
-    typedef iLorentzColourMatrix<vsimd> vobj;
-    typedef typename vobj::scalar_object sobj;
-
-    ////////////////////////////////////////
-    // fill the Grid header
-    ////////////////////////////////////////
-    FieldMetaData header;
-    scidacRecord  _scidacRecord;
-    scidacFile    _scidacFile;
-
-    ScidacMetaData(Umu,header,_scidacRecord,_scidacFile);
-
-    std::string format = header.floating_point;
-    header.ensemble_id    = description;
-    header.ensemble_label = description;
-    header.sequence_number = sequence;
-    header.ildg_lfn = LFN;
-
-    assert ( (format == std::string("IEEE32BIG"))  
-           ||(format == std::string("IEEE64BIG")) );
-
-    //////////////////////////////////////////////////////
-    // Fill ILDG header data struct
-    //////////////////////////////////////////////////////
-    ildgFormat ildgfmt ;
-    ildgfmt.field     = std::string("su3gauge");
-
-    if ( format == std::string("IEEE32BIG") ) { 
-      ildgfmt.precision = 32;
-    } else { 
-      ildgfmt.precision = 64;
-    }
-    ildgfmt.version = 1.0;
-    ildgfmt.lx = header.dimension[0];
-    ildgfmt.ly = header.dimension[1];
-    ildgfmt.lz = header.dimension[2];
-    ildgfmt.lt = header.dimension[3];
-    assert(header.nd==4);
-    assert(header.nd==header.dimension.size());
-
-    //////////////////////////////////////////////////////////////////////////////
-    // Field norm tests
-    //////////////////////////////////////////////////////////////////////////////
-    FieldNormMetaData FieldNormMetaData_;
-    FieldNormMetaData_.norm2 = norm2(Umu);
-
-    //////////////////////////////////////////////////////////////////////////////
-    // Fill the USQCD info field
-    //////////////////////////////////////////////////////////////////////////////
-    usqcdInfo info;
-    info.version=1.0;
-    info.plaq   = header.plaquette;
-    info.linktr = header.link_trace;
-
-    //    std::cout << GridLogMessage << " Writing config; IldgIO n2 "<< FieldNormMetaData_.norm2<<std::endl;
-    //////////////////////////////////////////////
-    // Fill the Lime file record by record
-    //////////////////////////////////////////////
-    writeLimeObject(1,0,header ,std::string("FieldMetaData"),std::string(GRID_FORMAT)); // Open message 
-    writeLimeObject(0,0,FieldNormMetaData_,FieldNormMetaData_.SerialisableClassName(),std::string(GRID_FIELD_NORM));
-    writeLimeObject(0,0,_scidacFile,_scidacFile.SerialisableClassName(),std::string(SCIDAC_PRIVATE_FILE_XML));
-    writeLimeObject(0,1,info,info.SerialisableClassName(),std::string(SCIDAC_FILE_XML));
-    writeLimeObject(1,0,_scidacRecord,_scidacRecord.SerialisableClassName(),std::string(SCIDAC_PRIVATE_RECORD_XML));
-    writeLimeObject(0,0,info,info.SerialisableClassName(),std::string(SCIDAC_RECORD_XML));
-    writeLimeObject(0,0,ildgfmt,std::string("ildgFormat")   ,std::string(ILDG_FORMAT)); // rec
-    writeLimeIldgLFN(header.ildg_lfn);                                                 // rec
-    writeLimeLatticeBinaryObject(Umu,std::string(ILDG_BINARY_DATA));      // Closes message with checksum
-    //    limeDestroyWriter(LimeW);
-  }
-};
-
-class IldgReader : public GridLimeReader {
- public:
-
-  ////////////////////////////////////////////////////////////////
-  // Read either Grid/SciDAC/ILDG configuration
-  // Don't require scidac records EXCEPT checksum
-  // Use Grid MetaData object if present.
-  // Else use ILDG MetaData object if present.
-  // Else use SciDAC MetaData object if present.
-  ////////////////////////////////////////////////////////////////
-  template <class vsimd>
-  void readConfiguration(Lattice<iLorentzColourMatrix<vsimd> > &Umu, FieldMetaData &FieldMetaData_) {
-
-    typedef Lattice<iLorentzColourMatrix<vsimd> > GaugeField;
-    typedef typename GaugeField::vector_object  vobj;
-    typedef typename vobj::scalar_object sobj;
-
-    typedef LorentzColourMatrixF fobj;
-    typedef LorentzColourMatrixD dobj;
-
-    GridBase *grid = Umu._grid;
-
-    std::vector<int> dims = Umu._grid->FullDimensions();
-
-    assert(dims.size()==4);
-
-    // Metadata holders
-    ildgFormat     ildgFormat_    ;
-    std::string    ildgLFN_       ;
-    scidacChecksum scidacChecksum_; 
-    usqcdInfo      usqcdInfo_     ;
-    FieldNormMetaData FieldNormMetaData_;
-
-    // track what we read from file
-    int found_ildgFormat    =0;
-    int found_ildgLFN       =0;
-    int found_scidacChecksum=0;
-    int found_usqcdInfo     =0;
-    int found_ildgBinary =0;
-    int found_FieldMetaData =0;
-    int found_FieldNormMetaData =0;
-    uint32_t nersc_csum;
-    uint32_t scidac_csuma;
-    uint32_t scidac_csumb;
-
-    // Binary format
-    std::string format;
-
-    //////////////////////////////////////////////////////////////////////////
-    // Loop over all records
-    // -- Order is poorly guaranteed except ILDG header preceeds binary section.
-    // -- Run like an event loop.
-    // -- Impose trust hierarchy. Grid takes precedence & look for ILDG, and failing
-    //    that Scidac. 
-    // -- Insist on Scidac checksum record.
-    //////////////////////////////////////////////////////////////////////////
-
-    while ( limeReaderNextRecord(LimeR) == LIME_SUCCESS ) { 
-
-      uint64_t nbytes = limeReaderBytes(LimeR);//size of this record (configuration)
-      
-      //////////////////////////////////////////////////////////////////
-      // If not BINARY_DATA read a string and parse
-      //////////////////////////////////////////////////////////////////
-      if ( strncmp(limeReaderType(LimeR), ILDG_BINARY_DATA,strlen(ILDG_BINARY_DATA) )  ) {
-	
-	// Copy out the string
-	std::vector<char> xmlc(nbytes+1,'\0');
-	limeReaderReadData((void *)&xmlc[0], &nbytes, LimeR);    
-	//	std::cout << GridLogMessage<< "Non binary record :" <<limeReaderType(LimeR) <<std::endl; //<<"\n"<<(&xmlc[0])<<std::endl;
-
-	//////////////////////////////////
-	// ILDG format record
-
-	std::string xmlstring(&xmlc[0]);
-	if ( !strncmp(limeReaderType(LimeR), ILDG_FORMAT,strlen(ILDG_FORMAT)) ) { 
-
-	  XmlReader RD(xmlstring, true, "");
-	  read(RD,"ildgFormat",ildgFormat_);
-
-	  if ( ildgFormat_.precision == 64 ) format = std::string("IEEE64BIG");
-	  if ( ildgFormat_.precision == 32 ) format = std::string("IEEE32BIG");
-
-	  assert( ildgFormat_.lx == dims[0]);
-	  assert( ildgFormat_.ly == dims[1]);
-	  assert( ildgFormat_.lz == dims[2]);
-	  assert( ildgFormat_.lt == dims[3]);
-
-	  found_ildgFormat = 1;
-	}
-
-	if ( !strncmp(limeReaderType(LimeR), ILDG_DATA_LFN,strlen(ILDG_DATA_LFN)) ) {
-	  FieldMetaData_.ildg_lfn = xmlstring;
-	  found_ildgLFN = 1;
-	}
-
-	if ( !strncmp(limeReaderType(LimeR), GRID_FORMAT,strlen(ILDG_FORMAT)) ) { 
-
-	  XmlReader RD(xmlstring, true, "");
-	  read(RD,"FieldMetaData",FieldMetaData_);
-
-	  format = FieldMetaData_.floating_point;
-
-	  assert(FieldMetaData_.dimension[0] == dims[0]);
-	  assert(FieldMetaData_.dimension[1] == dims[1]);
-	  assert(FieldMetaData_.dimension[2] == dims[2]);
-	  assert(FieldMetaData_.dimension[3] == dims[3]);
-
-	  found_FieldMetaData = 1;
-	}
-
-	if ( !strncmp(limeReaderType(LimeR), SCIDAC_RECORD_XML,strlen(SCIDAC_RECORD_XML)) ) { 
-	  // is it a USQCD info field
-	  if ( xmlstring.find(std::string("usqcdInfo")) != std::string::npos ) { 
-	    //	    std::cout << GridLogMessage<<"...found a usqcdInfo field"<<std::endl;
-	    XmlReader RD(xmlstring, true, "");
-	    read(RD,"usqcdInfo",usqcdInfo_);
-	    found_usqcdInfo = 1;
-	  }
-	}
-
-	if ( !strncmp(limeReaderType(LimeR), SCIDAC_CHECKSUM,strlen(SCIDAC_CHECKSUM)) ) { 
-	  XmlReader RD(xmlstring, true, "");
-	  read(RD,"scidacChecksum",scidacChecksum_);
-	  found_scidacChecksum = 1;
-	}
-
-	if ( !strncmp(limeReaderType(LimeR), GRID_FIELD_NORM,strlen(GRID_FIELD_NORM)) ) { 
-	  XmlReader RD(xmlstring, true, "");
-	  read(RD,GRID_FIELD_NORM,FieldNormMetaData_);
-	  found_FieldNormMetaData = 1;
-	}
-
-      } else {  
-	/////////////////////////////////
-	// Binary data
-	/////////////////////////////////
-	//	std::cout << GridLogMessage << "ILDG Binary record found : "  ILDG_BINARY_DATA << std::endl;
-	uint64_t offset= ftello(File);
-	if ( format == std::string("IEEE64BIG") ) {
-	  GaugeSimpleMunger<dobj, sobj> munge;
-	  BinaryIO::readLatticeObject< vobj, dobj >(Umu, filename, munge, offset, format,nersc_csum,scidac_csuma,scidac_csumb);
-	} else { 
-	  GaugeSimpleMunger<fobj, sobj> munge;
-	  BinaryIO::readLatticeObject< vobj, fobj >(Umu, filename, munge, offset, format,nersc_csum,scidac_csuma,scidac_csumb);
-	}
-
-	found_ildgBinary = 1;
-      }
-
-    }
-
-    //////////////////////////////////////////////////////
-    // Minimally must find binary segment and checksum
-    // Since this is an ILDG reader require ILDG format
-    //////////////////////////////////////////////////////
-    assert(found_ildgBinary);
-    assert(found_ildgFormat);
-    assert(found_scidacChecksum);
-
-    // Must find something with the lattice dimensions
-    assert(found_FieldMetaData||found_ildgFormat);
-
-    if ( found_FieldMetaData ) {
-
-      std::cout << GridLogMessage<<"Grid MetaData was record found: configuration was probably written by Grid ! Yay ! "<<std::endl;
-
-    } else { 
-
-      assert(found_ildgFormat);
-      assert ( ildgFormat_.field == std::string("su3gauge") );
-
-      ///////////////////////////////////////////////////////////////////////////////////////
-      // Populate our Grid metadata as best we can
-      ///////////////////////////////////////////////////////////////////////////////////////
-
-      std::ostringstream vers; vers << ildgFormat_.version;
-      FieldMetaData_.hdr_version = vers.str();
-      FieldMetaData_.data_type = std::string("4D_SU3_GAUGE_3X3");
-
-      FieldMetaData_.nd=4;
-      FieldMetaData_.dimension.resize(4);
-
-      FieldMetaData_.dimension[0] = ildgFormat_.lx ;
-      FieldMetaData_.dimension[1] = ildgFormat_.ly ;
-      FieldMetaData_.dimension[2] = ildgFormat_.lz ;
-      FieldMetaData_.dimension[3] = ildgFormat_.lt ;
-
-      if ( found_usqcdInfo ) { 
-	FieldMetaData_.plaquette = usqcdInfo_.plaq;
-	FieldMetaData_.link_trace= usqcdInfo_.linktr;
-	std::cout << GridLogMessage <<"This configuration was probably written by USQCD "<<std::endl;
-	std::cout << GridLogMessage <<"USQCD xml record Plaquette : "<<FieldMetaData_.plaquette<<std::endl;
-	std::cout << GridLogMessage <<"USQCD xml record LinkTrace : "<<FieldMetaData_.link_trace<<std::endl;
-      } else { 
-	FieldMetaData_.plaquette = 0.0;
-	FieldMetaData_.link_trace= 0.0;
-	std::cout << GridLogWarning << "This configuration is unsafe with no plaquette records that can verify it !!! "<<std::endl;
-      }
-    }
-
-    ////////////////////////////////////////////////////////////
-    // Really really want to mandate a scidac checksum
-    ////////////////////////////////////////////////////////////
-    if ( found_FieldNormMetaData ) { 
-      RealD nn = norm2(Umu);
-      GRID_FIELD_NORM_CHECK(FieldNormMetaData_,nn);
-      std::cout << GridLogMessage<<"FieldNormMetaData matches " << std::endl;
-    }  else { 
-      std::cout << GridLogWarning<<"FieldNormMetaData not found. " << std::endl;
-    }
-    if ( found_scidacChecksum ) {
-      FieldMetaData_.scidac_checksuma = stoull(scidacChecksum_.suma,0,16);
-      FieldMetaData_.scidac_checksumb = stoull(scidacChecksum_.sumb,0,16);
-      scidacChecksumVerify(scidacChecksum_,scidac_csuma,scidac_csumb);
-      assert( scidac_csuma ==FieldMetaData_.scidac_checksuma);
-      assert( scidac_csumb ==FieldMetaData_.scidac_checksumb);
-      std::cout << GridLogMessage<<"SciDAC checksums match " << std::endl;
-    } else { 
-      std::cout << GridLogWarning<<"SciDAC checksums not found. This is unsafe. " << std::endl;
-      assert(0); // Can I insist always checksum ?
-    }
-
-    if ( found_FieldMetaData || found_usqcdInfo ) {
-      FieldMetaData checker;
-      GaugeStatistics(Umu,checker);
-      assert(fabs(checker.plaquette  - FieldMetaData_.plaquette )<1.0e-5);
-      assert(fabs(checker.link_trace - FieldMetaData_.link_trace)<1.0e-5);
-      std::cout << GridLogMessage<<"Plaquette and link trace match " << std::endl;
-    }
-  }
- };
-
-}}
-
-//HAVE_LIME
-#endif
-
-#endif

Grid/parallelIO/IldgIOtypes.h

@@ -1,237 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/parallelIO/IldgIO.h
-
-Copyright (C) 2015
-
-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_ILDGTYPES_IO_H
-#define GRID_ILDGTYPES_IO_H
-
-#ifdef HAVE_LIME
-extern "C" { // for linkage
-#include "lime.h"
-}
-
-namespace Grid {
-
-/////////////////////////////////////////////////////////////////////////////////
-// Data representation of records that enter ILDG and SciDac formats
-/////////////////////////////////////////////////////////////////////////////////
-
-#define GRID_FORMAT      "grid-format"
-#define ILDG_FORMAT      "ildg-format"
-#define ILDG_BINARY_DATA "ildg-binary-data"
-#define ILDG_DATA_LFN    "ildg-data-lfn"
-#define SCIDAC_CHECKSUM           "scidac-checksum"
-#define SCIDAC_PRIVATE_FILE_XML   "scidac-private-file-xml"
-#define SCIDAC_FILE_XML           "scidac-file-xml"
-#define SCIDAC_PRIVATE_RECORD_XML "scidac-private-record-xml"
-#define SCIDAC_RECORD_XML         "scidac-record-xml"
-#define SCIDAC_BINARY_DATA        "scidac-binary-data"
-// Unused SCIDAC records names; could move to support this functionality
-#define SCIDAC_SITELIST           "scidac-sitelist"
-
-  ////////////////////////////////////////////////////////////
-  const int GRID_IO_SINGLEFILE = 0; // hardcode lift from QIO compat
-  const int GRID_IO_MULTIFILE  = 1; // hardcode lift from QIO compat
-  const int GRID_IO_FIELD      = 0; // hardcode lift from QIO compat
-  const int GRID_IO_GLOBAL     = 1; // hardcode lift from QIO compat
-  ////////////////////////////////////////////////////////////
-
-/////////////////////////////////////////////////////////////////////////////////
-// QIO uses mandatory "private" records fixed format
-// Private is in principle "opaque" however it can't be changed now because that would break existing 
-// file compatability, so should be correct to assume the undocumented but defacto file structure.
-/////////////////////////////////////////////////////////////////////////////////
-
-struct emptyUserRecord : Serializable { 
-  GRID_SERIALIZABLE_CLASS_MEMBERS(emptyUserRecord,int,dummy);
-  emptyUserRecord() { dummy=0; };
-};
-
-////////////////////////
-// Scidac private file xml
-// <?xml version="1.0" encoding="UTF-8"?><scidacFile><version>1.1</version><spacetime>4</spacetime><dims>16 16 16 32 </dims><volfmt>0</volfmt></scidacFile>
-////////////////////////
-struct scidacFile : Serializable {
- public:
-  GRID_SERIALIZABLE_CLASS_MEMBERS(scidacFile,
-                                  double, version,
-                                  int, spacetime,
-				  std::string, dims, // must convert to int
-                                  int, volfmt);
-
-  std::vector<int> getDimensions(void) { 
-    std::stringstream stream(dims);
-    std::vector<int> dimensions;
-    int n;
-    while(stream >> n){
-      dimensions.push_back(n);
-    }
-    return dimensions;
-  }
-
-  void setDimensions(std::vector<int> dimensions) { 
-    char delimiter = ' ';
-    std::stringstream stream;
-    for(int i=0;i<dimensions.size();i++){ 
-      stream << dimensions[i];
-      if ( i != dimensions.size()-1) { 
-	stream << delimiter <<std::endl;
-      }
-    }
-    dims = stream.str();
-  }
-
-  // Constructor provides Grid
-  scidacFile() =default; // default constructor
-  scidacFile(GridBase * grid){
-    version      = 1.0;
-    spacetime    = grid->_ndimension;
-    setDimensions(grid->FullDimensions()); 
-    volfmt       = GRID_IO_SINGLEFILE;
-  }
-
-};
-
-///////////////////////////////////////////////////////////////////////
-// scidac-private-record-xml : example
-// <scidacRecord>
-// <version>1.1</version><date>Tue Jul 26 21:14:44 2011 UTC</date><recordtype>0</recordtype>
-// <datatype>QDP_D3_ColorMatrix</datatype><precision>D</precision><colors>3</colors><spins>4</spins>
-// <typesize>144</typesize><datacount>4</datacount>
-// </scidacRecord>
-///////////////////////////////////////////////////////////////////////
-
-struct scidacRecord : Serializable {
- public:
-  GRID_SERIALIZABLE_CLASS_MEMBERS(scidacRecord,
-                                  double, version,
-                                  std::string, date,
-				  int, recordtype,
-				  std::string, datatype,
-				  std::string, precision,
-				  int, colors,
-				  int, spins,
-				  int, typesize,
-				  int, datacount);
-
-  scidacRecord()
-  : version(1.0), recordtype(0), colors(0), spins(0), typesize(0), datacount(0)
-  {}
-};
-
-////////////////////////
-// ILDG format
-////////////////////////
-struct ildgFormat : Serializable {
-public:
-  GRID_SERIALIZABLE_CLASS_MEMBERS(ildgFormat,
-				  double, version,
-				  std::string, field,
-				  int, precision,
-				  int, lx,
-				  int, ly,
-				  int, lz,
-				  int, lt);
-  ildgFormat() { version=1.0; };
-};
-////////////////////////
-// USQCD info
-////////////////////////
-struct usqcdInfo : Serializable { 
- public:
-  GRID_SERIALIZABLE_CLASS_MEMBERS(usqcdInfo,
-				  double, version,
-				  double, plaq,
-				  double, linktr,
-				  std::string, info);
-  usqcdInfo() { 
-    version=1.0; 
-  };
-};
-////////////////////////
-// Scidac Checksum
-////////////////////////
-struct scidacChecksum : Serializable { 
- public:
-  GRID_SERIALIZABLE_CLASS_MEMBERS(scidacChecksum,
-				  double, version,
-				  std::string, suma,
-				  std::string, sumb);
-  scidacChecksum() { 
-    version=1.0; 
-  };
-};
-////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-// Type:           scidac-file-xml         <title>MILC ILDG archival gauge configuration</title>
-////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-
-////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-// Type:           
-////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-
-////////////////////////
-// Scidac private file xml 
-// <?xml version="1.0" encoding="UTF-8"?><scidacFile><version>1.1</version><spacetime>4</spacetime><dims>16 16 16 32 </dims><volfmt>0</volfmt></scidacFile> 
-////////////////////////                                                                                                                                                                              
-
-#if 0
-////////////////////////////////////////////////////////////////////////////////////////
-// From http://www.physics.utah.edu/~detar/scidac/qio_2p3.pdf
-////////////////////////////////////////////////////////////////////////////////////////
-struct usqcdPropFile : Serializable { 
- public:
-  GRID_SERIALIZABLE_CLASS_MEMBERS(usqcdPropFile,
-				  double, version,
-				  std::string, type,
-				  std::string, info);
-  usqcdPropFile() { 
-    version=1.0; 
-  };
-};
-struct usqcdSourceInfo : Serializable { 
- public:
-  GRID_SERIALIZABLE_CLASS_MEMBERS(usqcdSourceInfo,
-				  double, version,
-				  std::string, info);
-  usqcdSourceInfo() { 
-    version=1.0; 
-  };
-};
-struct usqcdPropInfo : Serializable { 
- public:
-  GRID_SERIALIZABLE_CLASS_MEMBERS(usqcdPropInfo,
-				  double, version,
-				  int, spin,
-				  int, color,
-				  std::string, info);
-  usqcdPropInfo() { 
-    version=1.0; 
-  };
-};
-#endif
-
-}
-#endif
-#endif

Grid/parallelIO/MetaData.h

@@ -1,331 +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 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)
-      {}
-    };
-
-  namespace QCD {
-
-    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=Grid::QCD::WilsonLoops<PeriodicGimplF>::linkTrace(data);
-   header.plaquette =Grid::QCD::WilsonLoops<PeriodicGimplF>::avgPlaquette(data);
- }
- inline void GaugeStatistics(Lattice<vLorentzColourMatrixD> & data,FieldMetaData &header)
- {
-   // How to convert data precision etc...
-   header.link_trace=Grid::QCD::WilsonLoops<PeriodicGimplD>::linkTrace(data);
-   header.plaquette =Grid::QCD::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);
-	  }}
-	}
-      }
-    };
-  }
-
-
-}

Grid/parallelIO/NerscIO.h

@@ -1,363 +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 Grid {
-  namespace QCD {
-
-    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)
-      {
-      uint64_t offset=0;
-      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;
-	std::cout << Umu[0]<<std::endl;
-	std::cout << Umu[1]<<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;
-      }
-
-    };
-
-  }}
-#endif

Grid/perfmon/Stat.cc

@@ -1,245 +0,0 @@
-#include <Grid/GridCore.h>
-#include <Grid/perfmon/PerfCount.h>
-#include <Grid/perfmon/Stat.h>
-
-namespace Grid { 
-
-bool PmuStat::pmu_initialized=false;
-
-
-void PmuStat::init(const char *regname)
-{
-#ifdef __x86_64__
-  name = regname;
-  if (!pmu_initialized)
-    {
-      std::cout<<"initialising pmu"<<std::endl;
-      pmu_initialized = true;
-      pmu_init();
-    }
-  clear();
-#endif
-}
-void PmuStat::clear(void)
-{
-#ifdef __x86_64__
-  count = 0;
-  tregion = 0;
-  pmc0 = 0;
-  pmc1 = 0;
-  inst = 0;
-  cyc = 0;
-  ref = 0;
-  tcycles = 0;
-  reads = 0;
-  writes = 0;
-#endif
-}
-void PmuStat::print(void)
-{
-#ifdef __x86_64__
-  std::cout <<"Reg "<<std::string(name)<<":\n";
-  std::cout <<"  region "<<tregion<<std::endl;
-  std::cout <<"  cycles "<<tcycles<<std::endl;
-  std::cout <<"  inst   "<<inst   <<std::endl;
-  std::cout <<"  cyc    "<<cyc    <<std::endl;
-  std::cout <<"  ref    "<<ref    <<std::endl;
-  std::cout <<"  pmc0   "<<pmc0   <<std::endl;
-  std::cout <<"  pmc1   "<<pmc1   <<std::endl;
-  std::cout <<"  count  "<<count  <<std::endl;
-  std::cout <<"  reads  "<<reads  <<std::endl;
-  std::cout <<"  writes "<<writes <<std::endl;
-#endif
-}
-void PmuStat::start(void)
-{
-#ifdef __x86_64__
-  pmu_start();
-  ++count;
-  xmemctrs(&mrstart, &mwstart);
-  tstart = __rdtsc();
-#endif
-}
-void PmuStat::enter(int t)
-{
-#ifdef __x86_64__
-  counters[0][t] = __rdpmc(0);
-  counters[1][t] = __rdpmc(1);
-  counters[2][t] = __rdpmc((1<<30)|0);
-  counters[3][t] = __rdpmc((1<<30)|1);
-  counters[4][t] = __rdpmc((1<<30)|2);
-  counters[5][t] = __rdtsc();
-#endif
-}
-void PmuStat::exit(int t)
-{
-#ifdef __x86_64__
-  counters[0][t] = __rdpmc(0) - counters[0][t];
-  counters[1][t] = __rdpmc(1) - counters[1][t];
-  counters[2][t] = __rdpmc((1<<30)|0) - counters[2][t];
-  counters[3][t] = __rdpmc((1<<30)|1) - counters[3][t];
-  counters[4][t] = __rdpmc((1<<30)|2) - counters[4][t];
-  counters[5][t] = __rdtsc() - counters[5][t];
-#endif
-}
-void PmuStat::accum(int nthreads)
-{
-#ifdef __x86_64__
-  tend = __rdtsc();
-  xmemctrs(&mrend, &mwend);
-  pmu_stop();
-  for (int t = 0; t < nthreads; ++t) {
-    pmc0 += counters[0][t];
-    pmc1 += counters[1][t];
-    inst += counters[2][t];
-    cyc += counters[3][t];
-    ref += counters[4][t];
-    tcycles += counters[5][t];
-  }
-  uint64_t region = tend - tstart;
-  tregion += region;
-  uint64_t mreads = mrend - mrstart;
-  reads += mreads;
-  uint64_t mwrites = mwend - mwstart;
-  writes += mwrites;
-#endif
-}
-
-
-void PmuStat::pmu_fini(void) {}
-void PmuStat::pmu_start(void) {};
-void PmuStat::pmu_stop(void) {};
-void PmuStat::pmu_init(void)
-{
-#ifdef _KNIGHTS_LANDING_
-  KNLsetup();
-#endif
-}
-void PmuStat::xmemctrs(uint64_t *mr, uint64_t *mw)
-{
-#ifdef _KNIGHTS_LANDING_
-  ctrs c;
-  KNLreadctrs(c);
-  uint64_t emr = 0, emw = 0;
-  for (int i = 0; i < NEDC; ++i)
-    {
-      emr += c.edcrd[i];
-      emw += c.edcwr[i];
-    }
-  *mr = emr;
-  *mw = emw;
-#else
-  *mr = *mw = 0;
-#endif
-}
-
-#ifdef _KNIGHTS_LANDING_
-
-struct knl_gbl_ PmuStat::gbl;
-
-#define PMU_MEM
-
-void PmuStat::KNLevsetup(const char *ename, int &fd, int event, int umask)
-{
-  char fname[1024];
-  snprintf(fname, sizeof(fname), "%s/type", ename);
-  FILE *fp = fopen(fname, "r");
-  if (fp == 0) {
-    ::printf("open %s", fname);
-    ::exit(0);
-  }
-  int type;
-  int ret = fscanf(fp, "%d", &type);
-  assert(ret == 1);
-  fclose(fp);
-  //  std::cout << "Using PMU type "<<type<<" from " << std::string(ename) <<std::endl;
-
-  struct perf_event_attr hw = {};
-  hw.size = sizeof(hw);
-  hw.type = type;
-  // see /sys/devices/uncore_*/format/*
-  // All of the events we are interested in are configured the same way, but
-  // that isn't always true. Proper code would parse the format files
-  hw.config = event | (umask << 8);
-  //hw.read_format = PERF_FORMAT_GROUP;
-  // unfortunately the above only works within a single PMU; might
-  // as well just read them one at a time
-  int cpu = 0;
-  fd = perf_event_open(&hw, -1, cpu, -1, 0);
-  if (fd == -1) {
-    ::printf("CPU %d, box %s, event 0x%lx", cpu, ename, hw.config);
-    ::exit(0);
-  } else { 
-    //    std::cout << "event "<<std::string(ename)<<" set up for fd "<<fd<<" hw.config "<<hw.config <<std::endl;
-  }
-}
-
-
- void PmuStat::KNLsetup(void){
-
-   int ret;
-   char fname[1024];
-
-   // MC RPQ inserts and WPQ inserts (reads & writes)
-   for (int mc = 0; mc < NMC; ++mc)
-     {
-       ::snprintf(fname, sizeof(fname), "/sys/devices/uncore_imc_%d",mc);
-       // RPQ Inserts
-       KNLevsetup(fname, gbl.mc_rd[mc], 0x1, 0x1);
-       // WPQ Inserts
-       KNLevsetup(fname, gbl.mc_wr[mc], 0x2, 0x1);
-     }
-   // EDC RPQ inserts and WPQ inserts
-   for (int edc=0; edc < NEDC; ++edc)
-     {
-       ::snprintf(fname, sizeof(fname), "/sys/devices/uncore_edc_eclk_%d",edc);
-       // RPQ inserts
-       KNLevsetup(fname, gbl.edc_rd[edc], 0x1, 0x1);
-       // WPQ inserts
-       KNLevsetup(fname, gbl.edc_wr[edc], 0x2, 0x1);
-     }
-   // EDC HitE, HitM, MissE, MissM
-   for (int edc=0; edc < NEDC; ++edc)
-     {
-       ::snprintf(fname, sizeof(fname), "/sys/devices/uncore_edc_uclk_%d", edc);
-       KNLevsetup(fname, gbl.edc_hite[edc], 0x2, 0x1);
-       KNLevsetup(fname, gbl.edc_hitm[edc], 0x2, 0x2);
-       KNLevsetup(fname, gbl.edc_misse[edc], 0x2, 0x4);
-       KNLevsetup(fname, gbl.edc_missm[edc], 0x2, 0x8);
-     }
- }
-
-uint64_t PmuStat::KNLreadctr(int fd)
-{
-  uint64_t data;
-  size_t s = ::read(fd, &data, sizeof(data));
-  if (s != sizeof(uint64_t)){
-    ::printf("read counter %lu", s);
-    ::exit(0);
-  }
-  return data;
-}
-
-void PmuStat::KNLreadctrs(ctrs &c)
-{
-  for (int i = 0; i < NMC; ++i)
-    {
-      c.mcrd[i] = KNLreadctr(gbl.mc_rd[i]);
-      c.mcwr[i] = KNLreadctr(gbl.mc_wr[i]);
-    }
-  for (int i = 0; i < NEDC; ++i)
-    {
-      c.edcrd[i] = KNLreadctr(gbl.edc_rd[i]);
-      c.edcwr[i] = KNLreadctr(gbl.edc_wr[i]);
-    }
-  for (int i = 0; i < NEDC; ++i)
-    {
-      c.edchite[i] = KNLreadctr(gbl.edc_hite[i]);
-      c.edchitm[i] = KNLreadctr(gbl.edc_hitm[i]);
-      c.edcmisse[i] = KNLreadctr(gbl.edc_misse[i]);
-      c.edcmissm[i] = KNLreadctr(gbl.edc_missm[i]);
-    }
-}
-
-#endif
-}

Grid/perfmon/Stat.h

@@ -1,104 +0,0 @@
-#ifndef _GRID_STAT_H
-#define _GRID_STAT_H
-
-#ifdef AVX512
-#define _KNIGHTS_LANDING_ROOTONLY
-#endif
-
-namespace 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
-    
-  };
-
-}
-#endif
-
-

Grid/qcd/LatticeTheories.h

@@ -1,124 +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 */
-#ifndef GRID_LT_H
-#define GRID_LT_H
-namespace Grid{
-
-// First steps in the complete generalization of the Physics part
-// Design not final
-namespace LatticeTheories {
-
-template <int Dimensions>
-struct LatticeTheory {
-  static const int Nd = Dimensions;
-  static const int Nds = Dimensions * 2;  // double stored field
-  template <typename vtype>
-  using iSinglet = iScalar<iScalar<iScalar<vtype> > >;
-};
-
-template <int Dimensions, int Colours>
-struct LatticeGaugeTheory : public LatticeTheory<Dimensions> {
-  static const int Nds = Dimensions * 2;
-  static const int Nd = Dimensions;
-  static const int Nc = Colours;
-
-  template <typename vtype> 
-  using iColourMatrix = iScalar<iScalar<iMatrix<vtype, Nc> > >;
-  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 iColourVector = iScalar<iScalar<iVector<vtype, Nc> > >;
-};
-
-template <int Dimensions, int Colours, int Spin>
-struct FermionicLatticeGaugeTheory
-    : public LatticeGaugeTheory<Dimensions, Colours> {
-  static const int Nd = Dimensions;
-  static const int Nds = Dimensions * 2;
-  static const int Nc = Colours;
-  static const int Ns = Spin;
-
-  template <typename vtype>
-  using iSpinMatrix = iScalar<iMatrix<iScalar<vtype>, Ns> >;
-  template <typename vtype>
-  using iSpinColourMatrix = iScalar<iMatrix<iMatrix<vtype, Nc>, Ns> >;
-  template <typename vtype>
-  using iSpinVector = iScalar<iVector<iScalar<vtype>, Ns> >;
-  template <typename vtype>
-  using iSpinColourVector = iScalar<iVector<iVector<vtype, Nc>, Ns> >;
-  // These 2 only if Spin is a multiple of 2
-  static const int Nhs = Spin / 2;
-  template <typename vtype>
-  using iHalfSpinVector = iScalar<iVector<iScalar<vtype>, Nhs> >;
-  template <typename vtype>
-  using iHalfSpinColourVector = iScalar<iVector<iVector<vtype, Nc>, Nhs> >;
-
-  //tests
-  typedef iColourMatrix<Complex> ColourMatrix;
-  typedef iColourMatrix<ComplexF> ColourMatrixF;
-  typedef iColourMatrix<ComplexD> ColourMatrixD;
-
-
-};
-
-// Examples, not complete now.
-struct QCD : public FermionicLatticeGaugeTheory<4, 3, 4> {
-    static const int Xp = 0;
-    static const int Yp = 1;
-    static const int Zp = 2;
-    static const int Tp = 3;
-    static const int Xm = 4;
-    static const int Ym = 5;
-    static const int Zm = 6;
-    static const int Tm = 7;
-
-    typedef FermionicLatticeGaugeTheory FLGT;
-
-    typedef FLGT::iSpinMatrix<Complex  >          SpinMatrix;
-    typedef FLGT::iSpinMatrix<ComplexF >          SpinMatrixF;
-    typedef FLGT::iSpinMatrix<ComplexD >          SpinMatrixD;
-
-};
-struct QED : public FermionicLatticeGaugeTheory<4, 1, 4> {//fill
-};
-
-template <int Dimensions>
-struct Scalar : public LatticeTheory<Dimensions> {};
-
-};  // LatticeTheories
-
-} // Grid
-
-#endif

Grid/qcd/action/ActionBase.h

@@ -1,56 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/ActionBase.h
-
-Copyright (C) 2015-2016
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: neo <cossu@post.kek.jp>
-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 ACTION_BASE_H
-#define ACTION_BASE_H
-
-namespace Grid {
-namespace QCD {
-
-template <class GaugeField >
-class Action 
-{
-
- public:
-  bool is_smeared = false;
-  // Heatbath?
-  virtual void refresh(const GaugeField& U, GridParallelRNG& pRNG) = 0; // refresh pseudofermions
-  virtual RealD S(const GaugeField& U) = 0;                             // evaluate the action
-  virtual void deriv(const GaugeField& U, GaugeField& dSdU) = 0;        // evaluate the action derivative
-  virtual std::string action_name()    = 0;                             // return the action name
-  virtual std::string LogParameters()  = 0;                             // prints action parameters
-  virtual ~Action(){}
-};
-
-}
-}
-
-#endif // ACTION_BASE_H

Grid/qcd/action/ActionCore.h

@@ -1,61 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/ActionCore.h
-
-Copyright (C) 2015
-
-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 QCD_ACTION_CORE
-#define QCD_ACTION_CORE
-
-#include <Grid/qcd/action/ActionBase.h>
-#include <Grid/qcd/action/ActionSet.h>
-#include <Grid/qcd/action/ActionParams.h>
-
-////////////////////////////////////////////
-// Gauge Actions
-////////////////////////////////////////////
-#include <Grid/qcd/action/gauge/Gauge.h>
-
-////////////////////////////////////////////
-// Fermion prereqs
-////////////////////////////////////////////
-#include <Grid/qcd/action/fermion/FermionCore.h>
-
-////////////////////////////////////////////
-// Scalar Actions
-////////////////////////////////////////////
-#include <Grid/qcd/action/scalar/Scalar.h>
-
-////////////////////////////////////////////
-// Utility functions
-////////////////////////////////////////////
-#include <Grid/qcd/utils/Metric.h>
-#include <Grid/qcd/utils/CovariantLaplacian.h>
-
-
-
-
-#endif

Grid/qcd/action/ActionParams.h

@@ -1,98 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/ActionParams.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_QCD_ACTION_PARAMS_H
-#define GRID_QCD_ACTION_PARAMS_H
-
-namespace Grid {
-namespace QCD {
-
-  // These can move into a params header and be given MacroMagic serialisation
-  struct GparityWilsonImplParams {
-    bool overlapCommsCompute;
-    std::vector<int> twists;
-    GparityWilsonImplParams() : twists(Nd, 0), overlapCommsCompute(false){};
-  };
-  
-  struct WilsonImplParams {
-    bool overlapCommsCompute;
-    std::vector<Real> twist_n_2pi_L;
-    std::vector<Complex> boundary_phases;
-    WilsonImplParams() : overlapCommsCompute(false) {
-      boundary_phases.resize(Nd, 1.0);
-      twist_n_2pi_L.resize(Nd, 0.0);
-    };
-    WilsonImplParams(const std::vector<Complex> phi) : boundary_phases(phi), overlapCommsCompute(false) {
-      twist_n_2pi_L.resize(Nd, 0.0);
-    }
-  };
-
-  struct StaggeredImplParams {
-    StaggeredImplParams()  {};
-  };
-  
-  struct OneFlavourRationalParams : Serializable {
-    GRID_SERIALIZABLE_CLASS_MEMBERS(OneFlavourRationalParams, 
-				    RealD, lo, 
-				    RealD, hi, 
-				    int,   MaxIter, 
-				    RealD, tolerance, 
-				    int,   degree, 
-				    int,   precision,
-				    int,   BoundsCheckFreq);
-    
-    // MaxIter and tolerance, vectors??
-    
-    // constructor 
-    OneFlavourRationalParams(	RealD _lo      = 0.0, 
-				RealD _hi      = 1.0, 
-				int _maxit     = 1000,
-				RealD tol      = 1.0e-8, 
-                           	int _degree    = 10,
-				int _precision = 64,
-				int _BoundsCheckFreq=20)
-      : lo(_lo),
-	hi(_hi),
-	MaxIter(_maxit),
-	tolerance(tol),
-	degree(_degree),
-        precision(_precision),
-        BoundsCheckFreq(_BoundsCheckFreq){};
-  };
-  
-  
-}
-}
-
-
-
-
-#endif

Grid/qcd/action/ActionSet.h

@@ -1,116 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/ActionSet.h
-
-Copyright (C) 2015
-
-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 ACTION_SET_H
-#define ACTION_SET_H
-
-namespace Grid {
-
-// Should drop this namespace here
-namespace QCD {
-
-//////////////////////////////////
-// Indexing of tuple types
-//////////////////////////////////
-
-template <class T, class Tuple>
-struct Index;
-
-template <class T, class... Types>
-struct Index<T, std::tuple<T, Types...>> {
-  static const std::size_t value = 0;
-};
-
-template <class T, class U, class... Types>
-struct Index<T, std::tuple<U, Types...>> {
-  static const std::size_t value = 1 + Index<T, std::tuple<Types...>>::value;
-};
-
-
-////////////////////////////////////////////
-// Action Level
-// Action collection 
-// in a integration level
-// (for multilevel integration schemes)
-////////////////////////////////////////////
-
-template <class Field, class Repr = NoHirep >
-struct ActionLevel {
- public:
-  unsigned int multiplier;
-
-  // Fundamental repr actions separated because of the smearing
-  typedef Action<Field>* ActPtr;
-
-  // construct a tuple of vectors of the actions for the corresponding higher
-  // representation fields
-  typedef typename AccessTypes<Action, Repr>::VectorCollection action_collection;
-  typedef typename  AccessTypes<Action, Repr>::FieldTypeCollection action_hirep_types;
-
-  action_collection actions_hirep;
-  std::vector<ActPtr>& actions;
-
-  explicit ActionLevel(unsigned int mul = 1) : 
-  actions(std::get<0>(actions_hirep)), multiplier(mul) {
-    // initialize the hirep vectors to zero.
-    // apply(this->resize, actions_hirep, 0); //need a working resize
-    assert(mul >= 1);
-  }
-
-  template < class GenField >
-  void push_back(Action<GenField>* ptr) {
-    // insert only in the correct vector
-    std::get< Index < GenField, action_hirep_types>::value >(actions_hirep).push_back(ptr);
-  };
-
-  template <class ActPtr>
-  static void resize(ActPtr ap, unsigned int n) {
-    ap->resize(n);
-  }
-
-  // Loop on tuple for a callable function
-  template <std::size_t I = 1, typename Callable, typename ...Args>
-  inline typename std::enable_if<I == std::tuple_size<action_collection>::value, void>::type apply(Callable, Repr& R,Args&...) const {}
-
-  template <std::size_t I = 1, typename Callable, typename ...Args>
-  inline typename std::enable_if<I < std::tuple_size<action_collection>::value, void>::type apply(Callable fn, Repr& R, Args&... arguments) const {
-    fn(std::get<I>(actions_hirep), std::get<I>(R.rep), arguments...);
-    apply<I + 1>(fn, R, arguments...);
-  }  
-
-};
-
-// Define the ActionSet
-template <class GaugeField, class R>
-using ActionSet = std::vector<ActionLevel<GaugeField, R> >;
-
-} // QCD
-} // Grid
-
-#endif  // ACTION_SET_H

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 Grid {
-namespace QCD {
-
-  // 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) );
-      };
-  };
-}}
-
-#endif

Grid/qcd/action/fermion/CayleyFermion5D.cc

@@ -1,662 +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/Grid_Eigen_Dense.h>
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/CayleyFermion5D.h>
-
-namespace Grid {
-namespace QCD {
-
- template<class Impl>
- CayleyFermion5D<Impl>::CayleyFermion5D(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)
- { 
- }
-
-///////////////////////////////////////////////////////////////
-// Physical surface field utilities
-///////////////////////////////////////////////////////////////
-template<class Impl>  
-void CayleyFermion5D<Impl>::ExportPhysicalFermionSolution(const FermionField &solution5d,FermionField &exported4d)
-{
-  int Ls = this->Ls;
-  FermionField tmp(this->FermionGrid());
-  tmp = solution5d;
-  conformable(solution5d._grid,this->FermionGrid());
-  conformable(exported4d._grid,this->GaugeGrid());
-  axpby_ssp_pminus(tmp, 0., solution5d, 1., solution5d, 0, 0);
-  axpby_ssp_pplus (tmp, 1., tmp       , 1., solution5d, 0, Ls-1);
-  ExtractSlice(exported4d, tmp, 0, 0);
-}
-template<class Impl>  
-void CayleyFermion5D<Impl>::P(const FermionField &psi, FermionField &chi)
-{
-  int Ls= this->Ls;
-  chi=zero;
-  for(int s=0;s<Ls;s++){
-    axpby_ssp_pminus(chi,1.0,chi,1.0,psi,s,s);
-    axpby_ssp_pplus (chi,1.0,chi,1.0,psi,s,(s+1)%Ls);
-  }
-}
-template<class Impl>  
-void CayleyFermion5D<Impl>::Pdag(const FermionField &psi, FermionField &chi)
-{
-  int Ls= this->Ls;
-  chi=zero;
-  for(int s=0;s<Ls;s++){
-    axpby_ssp_pminus(chi,1.0,chi,1.0,psi,s,s);
-    axpby_ssp_pplus (chi,1.0,chi,1.0,psi,s,(s-1+Ls)%Ls);
-  }
-}
-template<class Impl>  
-void CayleyFermion5D<Impl>::ExportPhysicalFermionSource(const FermionField &solution5d,FermionField &exported4d)
-{
-  int Ls = this->Ls;
-  FermionField tmp(this->FermionGrid());
-  tmp = solution5d;
-  conformable(solution5d._grid,this->FermionGrid());
-  conformable(exported4d._grid,this->GaugeGrid());
-  axpby_ssp_pplus (tmp, 0., solution5d, 1., solution5d, 0, 0);
-  axpby_ssp_pminus(tmp, 1., tmp       , 1., solution5d, 0, Ls-1);
-  ExtractSlice(exported4d, tmp, 0, 0);
-}
-template<class Impl>
-void CayleyFermion5D<Impl>::ImportUnphysicalFermion(const FermionField &input4d,FermionField &imported5d)
-{
-  int Ls = this->Ls;
-  FermionField tmp(this->FermionGrid());
-  conformable(imported5d._grid,this->FermionGrid());
-  conformable(input4d._grid   ,this->GaugeGrid());
-  tmp = zero;
-  InsertSlice(input4d, tmp, 0   , 0);
-  InsertSlice(input4d, tmp, Ls-1, 0);
-  axpby_ssp_pplus (tmp, 0., tmp, 1., tmp, 0, 0);
-  axpby_ssp_pminus(tmp, 0., tmp, 1., tmp, Ls-1, Ls-1);
-  imported5d=tmp;
-}
-
-template<class Impl>  
-void CayleyFermion5D<Impl>::ImportPhysicalFermionSource(const FermionField &input4d,FermionField &imported5d)
-{
-  int Ls = this->Ls;
-  FermionField tmp(this->FermionGrid());
-  conformable(imported5d._grid,this->FermionGrid());
-  conformable(input4d._grid   ,this->GaugeGrid());
-  tmp = zero;
-  InsertSlice(input4d, tmp, 0   , 0);
-  InsertSlice(input4d, tmp, Ls-1, 0);
-  axpby_ssp_pplus (tmp, 0., tmp, 1., tmp, 0, 0);
-  axpby_ssp_pminus(tmp, 0., tmp, 1., tmp, Ls-1, Ls-1);
-  Dminus(tmp,imported5d);
-}
-template<class Impl>  
-void CayleyFermion5D<Impl>::Dminus(const FermionField &psi, FermionField &chi)
-{
-  int Ls=this->Ls;
-
-  FermionField tmp_f(this->FermionGrid());
-  this->DW(psi,tmp_f,DaggerNo);
-
-  for(int s=0;s<Ls;s++){
-    axpby_ssp(chi,Coeff_t(1.0),psi,-cs[s],tmp_f,s,s);// chi = (1-c[s] D_W) psi
-  }
-}
-template<class Impl>  
-void CayleyFermion5D<Impl>::DminusDag(const FermionField &psi, FermionField &chi)
-{
-  int Ls=this->Ls;
-
-  FermionField tmp_f(this->FermionGrid());
-  this->DW(psi,tmp_f,DaggerYes);
-
-  for(int s=0;s<Ls;s++){
-    axpby_ssp(chi,Coeff_t(1.0),psi,conjugate(-cs[s]),tmp_f,s,s);// chi = (1-c[s] D_W) psi
-  }
-}
-
-template<class Impl> void CayleyFermion5D<Impl>::CayleyReport(void)
-{
-  this->Report();
-  std::vector<int> latt = GridDefaultLatt();          
-  RealD volume = this->Ls;  for(int mu=0;mu<Nd;mu++) volume=volume*latt[mu];
-  RealD NP     = this->_FourDimGrid->_Nprocessors;
-  if ( M5Dcalls > 0 ) {
-    std::cout << GridLogMessage << "#### M5D calls report " << std::endl;
-    std::cout << GridLogMessage << "CayleyFermion5D Number of M5D Calls     : " << M5Dcalls   << std::endl;
-    std::cout << GridLogMessage << "CayleyFermion5D ComputeTime/Calls       : " << M5Dtime / M5Dcalls << " us" << std::endl;
-
-    // Flops = 6.0*(Nc*Ns) *Ls*vol
-    RealD mflops = 6.0*12*volume*M5Dcalls/M5Dtime/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;
-  }
-
-  if ( MooeeInvCalls > 0 ) {
-
-    std::cout << GridLogMessage << "#### MooeeInv calls report " << std::endl;
-    std::cout << GridLogMessage << "CayleyFermion5D Number of MooeeInv Calls     : " << MooeeInvCalls   << std::endl;
-    std::cout << GridLogMessage << "CayleyFermion5D ComputeTime/Calls            : " << MooeeInvTime / MooeeInvCalls << " us" << std::endl;
-
-    // Flops = MADD * Ls *Ls *4dvol * spin/colour/complex
-    RealD mflops = 2.0*24*this->Ls*volume*MooeeInvCalls/MooeeInvTime/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;
-  }
-
-}
-template<class Impl> void CayleyFermion5D<Impl>::CayleyZeroCounters(void)
-{
-  this->ZeroCounters();
-  M5Dflops=0;
-  M5Dcalls=0;
-  M5Dtime=0;
-  MooeeInvFlops=0;
-  MooeeInvCalls=0;
-  MooeeInvTime=0;
-}
-
-template<class Impl>  
-void CayleyFermion5D<Impl>::M5D   (const FermionField &psi, FermionField &chi)
-{
-  int Ls=this->Ls;
-  std::vector<Coeff_t> diag (Ls,1.0);
-  std::vector<Coeff_t> upper(Ls,-1.0); upper[Ls-1]=mass;
-  std::vector<Coeff_t> lower(Ls,-1.0); lower[0]   =mass;
-  M5D(psi,chi,chi,lower,diag,upper);
-}
-template<class Impl>
-void CayleyFermion5D<Impl>::Meooe5D    (const FermionField &psi, FermionField &Din)
-{
-  int Ls=this->Ls;
-  std::vector<Coeff_t> diag = bs;
-  std::vector<Coeff_t> upper= cs;
-  std::vector<Coeff_t> lower= cs; 
-  upper[Ls-1]=-mass*upper[Ls-1];
-  lower[0]   =-mass*lower[0];
-  M5D(psi,psi,Din,lower,diag,upper);
-}
-// FIXME Redunant with the above routine; check this and eliminate
-template<class Impl> void CayleyFermion5D<Impl>::Meo5D     (const FermionField &psi, FermionField &chi)
-{
-  int Ls=this->Ls;
-  std::vector<Coeff_t> diag = beo;
-  std::vector<Coeff_t> upper(Ls);
-  std::vector<Coeff_t> lower(Ls);
-  for(int i=0;i<Ls;i++) {
-    upper[i]=-ceo[i];
-    lower[i]=-ceo[i];
-  }
-  upper[Ls-1]=-mass*upper[Ls-1];
-  lower[0]   =-mass*lower[0];
-  M5D(psi,psi,chi,lower,diag,upper);
-}
-template<class Impl>
-void CayleyFermion5D<Impl>::Mooee       (const FermionField &psi, FermionField &chi)
-{
-  int Ls=this->Ls;
-  std::vector<Coeff_t> diag = bee;
-  std::vector<Coeff_t> upper(Ls);
-  std::vector<Coeff_t> lower(Ls);
-  for(int i=0;i<Ls;i++) {
-    upper[i]=-cee[i];
-    lower[i]=-cee[i];
-  }
-  upper[Ls-1]=-mass*upper[Ls-1];
-  lower[0]   =-mass*lower[0];
-  M5D(psi,psi,chi,lower,diag,upper);
-}
-template<class Impl>
-void CayleyFermion5D<Impl>::MooeeDag    (const FermionField &psi, FermionField &chi)
-{
-  int Ls=this->Ls;
-  std::vector<Coeff_t> diag = bee;
-  std::vector<Coeff_t> upper(Ls);
-  std::vector<Coeff_t> lower(Ls);
-
-  for (int s=0;s<Ls;s++){
-    // Assemble the 5d matrix
-    if ( s==0 ) {
-      upper[s] = -cee[s+1] ;
-      lower[s] = mass*cee[Ls-1];
-    } else if ( s==(Ls-1)) { 
-      upper[s] = mass*cee[0];
-      lower[s] = -cee[s-1];
-    } else {
-      upper[s]=-cee[s+1];
-      lower[s]=-cee[s-1];
-    }
-  }
-  // Conjugate the terms 
-  for (int s=0;s<Ls;s++){
-    diag[s] =conjugate(diag[s]);
-    upper[s]=conjugate(upper[s]);
-    lower[s]=conjugate(lower[s]);
-  }
-  M5Ddag(psi,psi,chi,lower,diag,upper);
-}
-
-template<class Impl>
-void CayleyFermion5D<Impl>::M5Ddag (const FermionField &psi, FermionField &chi)
-{
-  int Ls=this->Ls;
-  std::vector<Coeff_t> diag(Ls,1.0);
-  std::vector<Coeff_t> upper(Ls,-1.0);
-  std::vector<Coeff_t> lower(Ls,-1.0);
-  upper[Ls-1]=-mass*upper[Ls-1];
-  lower[0]   =-mass*lower[0];
-  M5Ddag(psi,chi,chi,lower,diag,upper);
-}
-
-template<class Impl>
-void CayleyFermion5D<Impl>::MeooeDag5D    (const FermionField &psi, FermionField &Din)
-{
-  int Ls=this->Ls;
-  std::vector<Coeff_t> diag =bs;
-  std::vector<Coeff_t> upper=cs;
-  std::vector<Coeff_t> lower=cs; 
-
-  for (int s=0;s<Ls;s++){
-    if ( s== 0 ) {
-      upper[s] = cs[s+1];
-      lower[s] =-mass*cs[Ls-1];
-    } else if ( s==(Ls-1) ) { 
-      upper[s] =-mass*cs[0];
-      lower[s] = cs[s-1];
-    } else { 
-      upper[s] = cs[s+1];
-      lower[s] = cs[s-1];
-    }
-    upper[s] = conjugate(upper[s]);
-    lower[s] = conjugate(lower[s]);
-    diag[s]  = conjugate(diag[s]);
-  }
-  M5Ddag(psi,psi,Din,lower,diag,upper);
-}
-
-template<class Impl>
-RealD CayleyFermion5D<Impl>::M    (const FermionField &psi, FermionField &chi)
-{
-  int Ls=this->Ls;
-  
-  FermionField Din(psi._grid);
-  
-  // Assemble Din
-  Meooe5D(psi,Din);
-  
-  this->DW(Din,chi,DaggerNo);
-  // ((b D_W + D_w hop terms +1) on s-diag
-  axpby(chi,1.0,1.0,chi,psi); 
-  
-  M5D(psi,chi);
-  return(norm2(chi));
-}
-
-template<class Impl>
-RealD CayleyFermion5D<Impl>::Mdag (const FermionField &psi, FermionField &chi)
-{
-  // Under adjoint
-  //D1+        D1- P-    ->   D1+^dag   P+ D2-^dag
-  //D2- P+     D2+            P-D1-^dag D2+dag
-  
-  FermionField Din(psi._grid);
-  // Apply Dw
-  this->DW(psi,Din,DaggerYes); 
-  
-  MeooeDag5D(Din,chi);
-  
-  M5Ddag(psi,chi);
-  // ((b D_W + D_w hop terms +1) on s-diag
-  axpby (chi,1.0,1.0,chi,psi); 
-  return norm2(chi);
-}
-
-// half checkerboard operations
-template<class Impl>
-void CayleyFermion5D<Impl>::Meooe       (const FermionField &psi, FermionField &chi)
-{
-  int Ls=this->Ls;
-
-  Meooe5D(psi,this->tmp()); 
-
-  if ( psi.checkerboard == Odd ) {
-    this->DhopEO(this->tmp(),chi,DaggerNo);
-  } else {
-    this->DhopOE(this->tmp(),chi,DaggerNo);
-  }
-}
-
-template<class Impl>
-void CayleyFermion5D<Impl>::MeooeDag    (const FermionField &psi, FermionField &chi)
-{
-  // Apply 4d dslash
-  if ( psi.checkerboard == Odd ) {
-    this->DhopEO(psi,this->tmp(),DaggerYes);
-  } else {
-    this->DhopOE(psi,this->tmp(),DaggerYes);
-  }
-  MeooeDag5D(this->tmp(),chi); 
-}
-
-template<class Impl>
-void  CayleyFermion5D<Impl>::Mdir (const FermionField &psi, FermionField &chi,int dir,int disp){
-  Meo5D(psi,this->tmp());
-  // Apply 4d dslash fragment
-  this->DhopDir(this->tmp(),chi,dir,disp);
-}
-// force terms; five routines; default to Dhop on diagonal
-template<class Impl>
-void CayleyFermion5D<Impl>::MDeriv  (GaugeField &mat,const FermionField &U,const FermionField &V,int dag)
-{
-  FermionField Din(V._grid);
-  
-  if ( dag == DaggerNo ) {
-    //      U d/du [D_w D5] V = U d/du DW D5 V
-    Meooe5D(V,Din);
-    this->DhopDeriv(mat,U,Din,dag);
-  } else {
-    //      U d/du [D_w D5]^dag V = U D5^dag d/du DW^dag Y // implicit adj on U in call
-    Meooe5D(U,Din);
-    this->DhopDeriv(mat,Din,V,dag);
-  }
-};
-template<class Impl>
-void CayleyFermion5D<Impl>::MoeDeriv(GaugeField &mat,const FermionField &U,const FermionField &V,int dag)
-{
-  FermionField Din(V._grid);
-  
-  if ( dag == DaggerNo ) {
-    //      U d/du [D_w D5] V = U d/du DW D5 V
-    Meooe5D(V,Din);
-    this->DhopDerivOE(mat,U,Din,dag);
-  } else {
-    //      U d/du [D_w D5]^dag V = U D5^dag d/du DW^dag Y // implicit adj on U in call
-    Meooe5D(U,Din);
-    this->DhopDerivOE(mat,Din,V,dag);
-  }
-};
-template<class Impl>
-void CayleyFermion5D<Impl>::MeoDeriv(GaugeField &mat,const FermionField &U,const FermionField &V,int dag)
-{
-  FermionField Din(V._grid);
-  
-  if ( dag == DaggerNo ) {
-    //      U d/du [D_w D5] V = U d/du DW D5 V
-    Meooe5D(V,Din);
-    this->DhopDerivEO(mat,U,Din,dag);
-  } else {
-    //      U d/du [D_w D5]^dag V = U D5^dag d/du DW^dag Y // implicit adj on U in call
-    Meooe5D(U,Din);
-    this->DhopDerivEO(mat,Din,V,dag);
-  }
-};
-  
-// Tanh
-template<class Impl>
-void CayleyFermion5D<Impl>::SetCoefficientsTanh(Approx::zolotarev_data *zdata,RealD b,RealD c)
-{
-  std::vector<Coeff_t> gamma(this->Ls);
-  for(int s=0;s<this->Ls;s++) gamma[s] = zdata->gamma[s];
-  SetCoefficientsInternal(1.0,gamma,b,c);
-}
-//Zolo
-template<class Impl>
-void CayleyFermion5D<Impl>::SetCoefficientsZolotarev(RealD zolo_hi,Approx::zolotarev_data *zdata,RealD b,RealD c)
-{
-  std::vector<Coeff_t> gamma(this->Ls);
-  for(int s=0;s<this->Ls;s++) gamma[s] = zdata->gamma[s];
-  SetCoefficientsInternal(zolo_hi,gamma,b,c);
-}
-//Zolo
-template<class Impl>
-void CayleyFermion5D<Impl>::SetCoefficientsInternal(RealD zolo_hi,std::vector<Coeff_t> & gamma,RealD b,RealD c)
-{
-  int Ls=this->Ls;
-
-  ///////////////////////////////////////////////////////////
-  // The Cayley coeffs (unprec)
-  ///////////////////////////////////////////////////////////
-  assert(gamma.size()==Ls);
-
-  omega.resize(Ls);
-  bs.resize(Ls);
-  cs.resize(Ls);
-  as.resize(Ls);
-  
-  // 
-  // Ts = (    [bs+cs]Dw        )^-1 (    (bs+cs) Dw         )
-  //     -(g5  -------       -1 )    ( g5 ---------     + 1  )
-  //      (   {2+(bs-cs)Dw}     )    (    2+(bs-cs) Dw       )
-  //
-  //  bs = 1/2( (1/omega_s + 1)*b + (1/omega - 1)*c ) = 1/2(  1/omega(b+c) + (b-c) )
-  //  cs = 1/2( (1/omega_s - 1)*b + (1/omega + 1)*c ) = 1/2(  1/omega(b+c) - (b-c) )
-  //
-  // bs+cs = 0.5*( 1/omega(b+c) + (b-c) + 1/omega(b+c) - (b-c) ) = 1/omega(b+c)
-  // bs-cs = 0.5*( 1/omega(b+c) + (b-c) - 1/omega(b+c) + (b-c) ) = b-c
-  //
-  // So 
-  //
-  // Ts = (    [b+c]Dw/omega_s    )^-1 (    (b+c) Dw /omega_s        )
-  //     -(g5  -------         -1 )    ( g5 ---------           + 1  )
-  //      (   {2+(b-c)Dw}         )    (    2+(b-c) Dw               )
-  //
-  // Ts = (    [b+c]Dw            )^-1 (    (b+c) Dw                 )
-  //     -(g5  -------    -omega_s)    ( g5 ---------      + omega_s )
-  //      (   {2+(b-c)Dw}         )    (    2+(b-c) Dw               )
-  // 
-    
-  double bpc = b+c;
-  double bmc = b-c;
-  _b = b;
-  _c = c;
-  _gamma  = gamma; // Save the parameters so we can change mass later.
-  _zolo_hi= zolo_hi;
-  for(int i=0; i < Ls; i++){
-    as[i] = 1.0;
-    omega[i] = _gamma[i]*_zolo_hi; //NB reciprocal relative to Chroma NEF code
-    assert(omega[i]!=Coeff_t(0.0));
-    bs[i] = 0.5*(bpc/omega[i] + bmc);
-    cs[i] = 0.5*(bpc/omega[i] - bmc);
-  }
-
-  ////////////////////////////////////////////////////////
-  // Constants for the preconditioned matrix Cayley form
-  ////////////////////////////////////////////////////////
-  bee.resize(Ls);
-  cee.resize(Ls);
-  beo.resize(Ls);
-  ceo.resize(Ls);
-  
-  for(int i=0;i<Ls;i++){
-    bee[i]=as[i]*(bs[i]*(4.0-this->M5) +1.0);     
-    assert(bee[i]!=Coeff_t(0.0));
-    cee[i]=as[i]*(1.0-cs[i]*(4.0-this->M5));
-    beo[i]=as[i]*bs[i];
-    ceo[i]=-as[i]*cs[i];
-  }
-  aee.resize(Ls);
-  aeo.resize(Ls);
-  for(int i=0;i<Ls;i++){
-    aee[i]=cee[i];
-    aeo[i]=ceo[i];
-  }
-  
-  //////////////////////////////////////////
-  // LDU decomposition of eeoo
-  //////////////////////////////////////////
-  dee.resize(Ls);
-  lee.resize(Ls);
-  leem.resize(Ls);
-  uee.resize(Ls);
-  ueem.resize(Ls);
-  
-  for(int i=0;i<Ls;i++){
-    
-    dee[i] = bee[i];
-    
-    if ( i < Ls-1 ) {
-
-      assert(bee[i]!=Coeff_t(0.0));
-      assert(bee[0]!=Coeff_t(0.0));
-      
-      lee[i] =-cee[i+1]/bee[i]; // sub-diag entry on the ith column
-      
-      leem[i]=mass*cee[Ls-1]/bee[0];
-      for(int j=0;j<i;j++) {
-	assert(bee[j+1]!=Coeff_t(0.0));
-	leem[i]*= aee[j]/bee[j+1];
-      }
-      
-      uee[i] =-aee[i]/bee[i];   // up-diag entry on the ith row
-      
-      ueem[i]=mass;
-      for(int j=1;j<=i;j++) ueem[i]*= cee[j]/bee[j];
-      ueem[i]*= aee[0]/bee[0];
-      
-    } else { 
-      lee[i] =0.0;
-      leem[i]=0.0;
-      uee[i] =0.0;
-      ueem[i]=0.0;
-    }
-  }
-	
-  { 
-    Coeff_t delta_d=mass*cee[Ls-1];
-    for(int j=0;j<Ls-1;j++) {
-      assert(bee[j] != Coeff_t(0.0));
-      delta_d *= cee[j]/bee[j];
-    }
-    dee[Ls-1] += delta_d;
-  }  
-
-  int inv=1;
-  this->MooeeInternalCompute(0,inv,MatpInv,MatmInv);
-  this->MooeeInternalCompute(1,inv,MatpInvDag,MatmInvDag);
-}
-
-
-template<class Impl>
-void CayleyFermion5D<Impl>::MooeeInternalCompute(int dag, int inv,
-						 Vector<iSinglet<Simd> > & Matp,
-						 Vector<iSinglet<Simd> > & Matm)
-{
-  int Ls=this->Ls;
-
-  GridBase *grid = this->FermionRedBlackGrid();
-  int LLs = grid->_rdimensions[0];
-
-  if ( LLs == Ls ) {
-    return; // Not vectorised in 5th direction
-  }
-
-  Eigen::MatrixXcd Pplus  = Eigen::MatrixXcd::Zero(Ls,Ls);
-  Eigen::MatrixXcd Pminus = Eigen::MatrixXcd::Zero(Ls,Ls);
-  
-  for(int s=0;s<Ls;s++){
-    Pplus(s,s) = bee[s];
-    Pminus(s,s)= bee[s];
-  }
-  
-  for(int s=0;s<Ls-1;s++){
-    Pminus(s,s+1) = -cee[s];
-  }
-  
-  for(int s=0;s<Ls-1;s++){
-    Pplus(s+1,s) = -cee[s+1];
-  }
-  Pplus (0,Ls-1) = mass*cee[0];
-  Pminus(Ls-1,0) = mass*cee[Ls-1];
-  
-  Eigen::MatrixXcd PplusMat ;
-  Eigen::MatrixXcd PminusMat;
-  
-  if ( inv ) {
-    PplusMat =Pplus.inverse();
-    PminusMat=Pminus.inverse();
-  } else { 
-    PplusMat =Pplus;
-    PminusMat=Pminus;
-  }
-  
-  if(dag){
-    PplusMat.adjointInPlace();
-    PminusMat.adjointInPlace();
-  }
-  
-  typedef typename SiteHalfSpinor::scalar_type scalar_type;
-  const int Nsimd=Simd::Nsimd();
-  Matp.resize(Ls*LLs);
-  Matm.resize(Ls*LLs);
-
-  for(int s2=0;s2<Ls;s2++){
-  for(int s1=0;s1<LLs;s1++){
-    int istride = LLs;
-    int ostride = 1;
-    Simd Vp;
-    Simd Vm;
-    scalar_type *sp = (scalar_type *)&Vp;
-    scalar_type *sm = (scalar_type *)&Vm;
-    for(int l=0;l<Nsimd;l++){
-      if ( switcheroo<Coeff_t>::iscomplex() ) {
-	sp[l] = PplusMat (l*istride+s1*ostride,s2);
-	sm[l] = PminusMat(l*istride+s1*ostride,s2);
-      } else { 
-      // if real
-	scalar_type tmp;
-	tmp = PplusMat (l*istride+s1*ostride,s2);
-	sp[l] = scalar_type(tmp.real(),tmp.real());
-	tmp = PminusMat(l*istride+s1*ostride,s2);
-	sm[l] = scalar_type(tmp.real(),tmp.real());
-      }
-    }
-    Matp[LLs*s2+s1] = Vp;
-    Matm[LLs*s2+s1] = Vm;
-  }}
-}
-
-
-  FermOpTemplateInstantiate(CayleyFermion5D);
-  GparityFermOpTemplateInstantiate(CayleyFermion5D);
-
-}}
-
-

Grid/qcd/action/fermion/CayleyFermion5D.h

@@ -1,226 +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 */
-#ifndef  GRID_QCD_CAYLEY_FERMION_H
-#define  GRID_QCD_CAYLEY_FERMION_H
-
-#include <Grid/qcd/action/fermion/WilsonFermion5D.h>
-
-namespace Grid {
-
-  namespace QCD {
-
-     template<typename T> struct switcheroo   {
-       static inline int iscomplex()  { return 0; }
-
-       template<class vec>
-       static inline vec mult(vec a, vec b) {
-	 return real_mult(a,b);
-       }
-     };
-     template<> struct switcheroo<ComplexD> {
-       static inline int iscomplex()  { return 1; }
-
-       template<class vec>
-       static inline vec mult(vec a, vec b) {
-	 return a*b;
-       }
-     };
-     template<> struct switcheroo<ComplexF> {
-       static inline int iscomplex()  { return 1; }
-       template<class vec>
-       static inline vec mult(vec a, vec b) {
-	 return a*b;
-       }
-     };
-
-
-    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,
-	       std::vector<Coeff_t> &lower,
-	       std::vector<Coeff_t> &diag,
-	       std::vector<Coeff_t> &upper);
-
-      void M5Ddag(const FermionField &psi,
-		  const FermionField &phi,
-		  FermionField &chi,
-		  std::vector<Coeff_t> &lower,
-		  std::vector<Coeff_t> &diag,
-		  std::vector<Coeff_t> &upper);
-
-      void MooeeInternal(const FermionField &in, FermionField &out,int dag,int inv);
-      void MooeeInternalCompute(int dag, int inv, Vector<iSinglet<Simd> > & Matp, Vector<iSinglet<Simd> > & Matm);
-
-      void MooeeInternalAsm(const FermionField &in, FermionField &out,
-			    int LLs, int site,
-			    Vector<iSinglet<Simd> > &Matp,
-			    Vector<iSinglet<Simd> > &Matm);
-      void MooeeInternalZAsm(const FermionField &in, FermionField &out,
-			    int LLs, int site,
-			    Vector<iSinglet<Simd> > &Matp,
-			    Vector<iSinglet<Simd> > &Matm);
-
-
-      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
-      std::vector<Coeff_t> _gamma;
-      RealD                _zolo_hi;
-      RealD                _b;
-      RealD                _c;
-
-      // Cayley form Moebius (tanh and zolotarev)
-      std::vector<Coeff_t> omega;
-      std::vector<Coeff_t> bs;    // S dependent coeffs
-      std::vector<Coeff_t> cs;
-      std::vector<Coeff_t> as;
-      // For preconditioning Cayley form
-      std::vector<Coeff_t> bee;
-      std::vector<Coeff_t> cee;
-      std::vector<Coeff_t> aee;
-      std::vector<Coeff_t> beo;
-      std::vector<Coeff_t> ceo;
-      std::vector<Coeff_t> aeo;
-      // LDU factorisation of the eeoo matrix
-      std::vector<Coeff_t> lee;
-      std::vector<Coeff_t> leem;
-      std::vector<Coeff_t> uee;
-      std::vector<Coeff_t> ueem;
-      std::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,std::vector<Coeff_t> & gamma,RealD b,RealD c);
-    };
-
-  }
-}
-#define INSTANTIATE_DPERP(A)\
-template void CayleyFermion5D< A >::M5D(const FermionField &psi,const FermionField &phi,FermionField &chi,\
-					std::vector<Coeff_t> &lower,std::vector<Coeff_t> &diag,std::vector<Coeff_t> &upper); \
-template void CayleyFermion5D< A >::M5Ddag(const FermionField &psi,const FermionField &phi,FermionField &chi,\
-					   std::vector<Coeff_t> &lower,std::vector<Coeff_t> &diag,std::vector<Coeff_t> &upper); \
-template void CayleyFermion5D< A >::MooeeInv    (const FermionField &psi, FermionField &chi); \
-template void CayleyFermion5D< A >::MooeeInvDag (const FermionField &psi, FermionField &chi);
-
-#undef  CAYLEY_DPERP_DENSE
-#define  CAYLEY_DPERP_CACHE
-#undef  CAYLEY_DPERP_LINALG
-#define CAYLEY_DPERP_VEC
-
-#endif

Grid/qcd/action/fermion/CayleyFermion5Dcache.cc

@@ -1,249 +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 Grid {
-namespace QCD {
-
-  // FIXME -- make a version of these routines with site loop outermost for cache reuse.
-
-  // Pminus fowards
-  // Pplus  backwards..
-template<class Impl>  
-void CayleyFermion5D<Impl>::M5D(const FermionField &psi,
-				const FermionField &phi, 
-				FermionField &chi,
-				std::vector<Coeff_t> &lower,
-				std::vector<Coeff_t> &diag,
-				std::vector<Coeff_t> &upper)
-{
-  int Ls =this->Ls;
-  GridBase *grid=psi._grid;
-  assert(phi.checkerboard == psi.checkerboard);
-  chi.checkerboard=psi.checkerboard;
-  // Flops = 6.0*(Nc*Ns) *Ls*vol
-  M5Dcalls++;
-  M5Dtime-=usecond();
-
-  parallel_for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
-    for(int s=0;s<Ls;s++){
-      auto tmp = psi._odata[0];
-      if ( s==0 ) {
- 	                            spProj5m(tmp,psi._odata[ss+s+1]);
-	chi[ss+s]=diag[s]*phi[ss+s]+upper[s]*tmp;
-
-	                    spProj5p(tmp,psi._odata[ss+Ls-1]);
-	chi[ss+s]=chi[ss+s]+lower[s]*tmp;
-      } else if ( s==(Ls-1)) {
-	                            spProj5m(tmp,psi._odata[ss+0]);
-	chi[ss+s]=diag[s]*phi[ss+s]+upper[s]*tmp;
-
- 	                    spProj5p(tmp,psi._odata[ss+s-1]);
-	chi[ss+s]=chi[ss+s]+lower[s]*tmp;
-      } else { 
-	                            spProj5m(tmp,psi._odata[ss+s+1]);
-	chi[ss+s]=diag[s]*phi[ss+s]+upper[s]*tmp;
-
-	                    spProj5p(tmp,psi._odata[ss+s-1]);
-	chi[ss+s]=chi[ss+s]+lower[s]*tmp;
-      }
-    }
-  }
-  M5Dtime+=usecond();
-}
-
-template<class Impl>  
-void CayleyFermion5D<Impl>::M5Ddag(const FermionField &psi,
-				   const FermionField &phi, 
-				   FermionField &chi,
-				   std::vector<Coeff_t> &lower,
-				   std::vector<Coeff_t> &diag,
-				   std::vector<Coeff_t> &upper)
-{
-  int Ls =this->Ls;
-  GridBase *grid=psi._grid;
-  assert(phi.checkerboard == psi.checkerboard);
-  chi.checkerboard=psi.checkerboard;
-
-  // Flops = 6.0*(Nc*Ns) *Ls*vol
-  M5Dcalls++;
-  M5Dtime-=usecond();
-
-  parallel_for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
-    auto tmp = psi._odata[0];
-    for(int s=0;s<Ls;s++){
-      if ( s==0 ) {
-	spProj5p(tmp,psi._odata[ss+s+1]);
-	chi[ss+s]=diag[s]*phi[ss+s]+upper[s]*tmp;
-
-	spProj5m(tmp,psi._odata[ss+Ls-1]);
-	chi[ss+s]=chi[ss+s]+lower[s]*tmp;
-      } else if ( s==(Ls-1)) {
-	spProj5p(tmp,psi._odata[ss+0]);
-	chi[ss+s]=diag[s]*phi[ss+s]+upper[s]*tmp;
-
-	spProj5m(tmp,psi._odata[ss+s-1]);
-	chi[ss+s]=chi[ss+s]+lower[s]*tmp;
-      } else { 
-	spProj5p(tmp,psi._odata[ss+s+1]);
-	chi[ss+s]=diag[s]*phi[ss+s]+upper[s]*tmp;
-
-	spProj5m(tmp,psi._odata[ss+s-1]);
-	chi[ss+s]=chi[ss+s]+lower[s]*tmp;
-      }
-    }
-  }
-  M5Dtime+=usecond();
-}
-
-template<class Impl>
-void CayleyFermion5D<Impl>::MooeeInv    (const FermionField &psi, FermionField &chi)
-{
-  GridBase *grid=psi._grid;
-  int Ls=this->Ls;
-
-  chi.checkerboard=psi.checkerboard;
-
-  MooeeInvCalls++;
-  MooeeInvTime-=usecond();
-
-  parallel_for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
-    auto tmp = psi._odata[0];
-
-    // flops = 12*2*Ls + 12*2*Ls + 3*12*Ls + 12*2*Ls  = 12*Ls * (9) = 108*Ls flops
-    // Apply (L^{\prime})^{-1}
-    chi[ss]=psi[ss]; // chi[0]=psi[0]
-    for(int s=1;s<Ls;s++){
-                            spProj5p(tmp,chi[ss+s-1]);  
-      chi[ss+s] = psi[ss+s]-lee[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]);    
-      chi[ss+Ls-1] = chi[ss+Ls-1] - leem[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]); 
-      chi[ss+s] = (1.0/dee[s])*chi[ss+s]-(ueem[s]/dee[Ls-1])*tmp;
-    }	
-    chi[ss+Ls-1]= (1.0/dee[Ls-1])*chi[ss+Ls-1];
-      
-    // Apply U^{-1}
-    for (int s=Ls-2;s>=0;s--){
-                            spProj5m(tmp,chi[ss+s+1]);  
-      chi[ss+s] = chi[ss+s] - uee[s]*tmp;
-    }
-  }
-
-  MooeeInvTime+=usecond();
-
-}
-
-template<class Impl>
-void CayleyFermion5D<Impl>::MooeeInvDag (const FermionField &psi, FermionField &chi)
-{
-  GridBase *grid=psi._grid;
-  int Ls=this->Ls;
-
-  assert(psi.checkerboard == psi.checkerboard);
-  chi.checkerboard=psi.checkerboard;
-
-  std::vector<Coeff_t> ueec(Ls);
-  std::vector<Coeff_t> deec(Ls);
-  std::vector<Coeff_t> leec(Ls);
-  std::vector<Coeff_t> ueemc(Ls);
-  std::vector<Coeff_t> leemc(Ls);
-  for(int s=0;s<ueec.size();s++){
-    ueec[s] = conjugate(uee[s]);
-    deec[s] = conjugate(dee[s]);
-    leec[s] = conjugate(lee[s]);
-    ueemc[s]= conjugate(ueem[s]);
-    leemc[s]= conjugate(leem[s]);
-  }
-  MooeeInvCalls++;
-  MooeeInvTime-=usecond();
-
-  parallel_for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
-
-    auto tmp = psi._odata[0];
-
-    // Apply (U^{\prime})^{-dagger}
-    chi[ss]=psi[ss];
-    for (int s=1;s<Ls;s++){
-                            spProj5m(tmp,chi[ss+s-1]);
-      chi[ss+s] = psi[ss+s]-ueec[s-1]*tmp;
-    }
-    // U_m^{-\dagger} 
-    for (int s=0;s<Ls-1;s++){
-                                   spProj5p(tmp,chi[ss+s]);
-      chi[ss+Ls-1] = chi[ss+Ls-1] - ueemc[s]*tmp;
-    }
-
-    // L_m^{-\dagger} D^{-dagger}
-    for (int s=0;s<Ls-1;s++){
-      spProj5m(tmp,chi[ss+Ls-1]);
-      chi[ss+s] = (1.0/deec[s])*chi[ss+s]-(leemc[s]/deec[Ls-1])*tmp;
-    }	
-    chi[ss+Ls-1]= (1.0/deec[Ls-1])*chi[ss+Ls-1];
-  
-    // Apply L^{-dagger}
-    for (int s=Ls-2;s>=0;s--){
-      spProj5p(tmp,chi[ss+s+1]);
-      chi[ss+s] = chi[ss+s] - leec[s]*tmp;
-    }
-  }
-
-  MooeeInvTime+=usecond();
-
-}
-
-#ifdef CAYLEY_DPERP_CACHE
-  INSTANTIATE_DPERP(WilsonImplF);
-  INSTANTIATE_DPERP(WilsonImplD);
-  INSTANTIATE_DPERP(GparityWilsonImplF);
-  INSTANTIATE_DPERP(GparityWilsonImplD);
-  INSTANTIATE_DPERP(ZWilsonImplF);
-  INSTANTIATE_DPERP(ZWilsonImplD);
-
-  INSTANTIATE_DPERP(WilsonImplFH);
-  INSTANTIATE_DPERP(WilsonImplDF);
-  INSTANTIATE_DPERP(GparityWilsonImplFH);
-  INSTANTIATE_DPERP(GparityWilsonImplDF);
-  INSTANTIATE_DPERP(ZWilsonImplFH);
-  INSTANTIATE_DPERP(ZWilsonImplDF);
-#endif
-
-}}

Grid/qcd/action/fermion/CayleyFermion5Ddense.cc

@@ -1,156 +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/Grid_Eigen_Dense.h>
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/CayleyFermion5D.h>
-
-
-namespace Grid {
-namespace QCD {
-  /*
-   * Dense matrix versions of routines
-   */
-template<class Impl>
-void CayleyFermion5D<Impl>::MooeeInvDag (const FermionField &psi, FermionField &chi)
-{
-  this->MooeeInternal(psi,chi,DaggerYes,InverseYes);
-}
-template<class Impl>
-void CayleyFermion5D<Impl>::MooeeInv(const FermionField &psi, FermionField &chi)
-{
-  this->MooeeInternal(psi,chi,DaggerNo,InverseYes);
-}
-
-template<class Impl>
-void CayleyFermion5D<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) = bee[s];
-    Pminus(s,s)= bee[s];
-  }
-  
-  for(int s=0;s<Ls-1;s++){
-    Pminus(s,s+1) = -cee[s];
-  }
-  
-  for(int s=0;s<Ls-1;s++){
-    Pplus(s+1,s) = -cee[s+1];
-  }
-  Pplus (0,Ls-1) = mass*cee[0];
-  Pminus(Ls-1,0) = mass*cee[Ls-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 CAYLEY_DPERP_DENSE
-INSTANTIATE_DPERP(GparityWilsonImplF);
-INSTANTIATE_DPERP(GparityWilsonImplD);
-INSTANTIATE_DPERP(WilsonImplF);
-INSTANTIATE_DPERP(WilsonImplD);
-INSTANTIATE_DPERP(ZWilsonImplF);
-INSTANTIATE_DPERP(ZWilsonImplD);
-
-template void CayleyFermion5D<GparityWilsonImplF>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
-template void CayleyFermion5D<GparityWilsonImplD>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
-template void CayleyFermion5D<WilsonImplF>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
-template void CayleyFermion5D<WilsonImplD>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
-template void CayleyFermion5D<ZWilsonImplF>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
-template void CayleyFermion5D<ZWilsonImplD>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
-
-INSTANTIATE_DPERP(GparityWilsonImplFH);
-INSTANTIATE_DPERP(GparityWilsonImplDF);
-INSTANTIATE_DPERP(WilsonImplFH);
-INSTANTIATE_DPERP(WilsonImplDF);
-INSTANTIATE_DPERP(ZWilsonImplFH);
-INSTANTIATE_DPERP(ZWilsonImplDF);
-
-template void CayleyFermion5D<GparityWilsonImplFH>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
-template void CayleyFermion5D<GparityWilsonImplDF>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
-template void CayleyFermion5D<WilsonImplFH>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
-template void CayleyFermion5D<WilsonImplDF>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
-template void CayleyFermion5D<ZWilsonImplFH>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
-template void CayleyFermion5D<ZWilsonImplDF>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
-#endif
-
-}}

Grid/qcd/action/fermion/CayleyFermion5Dssp.cc

@@ -1,164 +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 Grid {
-namespace QCD {
-
-  // FIXME -- make a version of these routines with site loop outermost for cache reuse.
-  // Pminus fowards
-  // Pplus  backwards
-template<class Impl>  
-void CayleyFermion5D<Impl>::M5D(const FermionField &psi,
-				const FermionField &phi, 
-				FermionField &chi,
-				std::vector<Coeff_t> &lower,
-				std::vector<Coeff_t> &diag,
-				std::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 CayleyFermion5D<Impl>::M5Ddag(const FermionField &psi,
-				   const FermionField &phi, 
-				   FermionField &chi,
-				   std::vector<Coeff_t> &lower,
-				   std::vector<Coeff_t> &diag,
-				   std::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 CayleyFermion5D<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;
-  // 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,-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,-leem[s],chi,Ls-1,s);
-  }
-  // U_m^{-1} D^{-1}
-  for (int s=0;s<Ls-1;s++){
-    // Chi[s] + 1/d chi[s] 
-    axpby_ssp_pplus(chi,one/dee[s],chi,-ueem[s]/dee[Ls-1],chi,s,Ls-1);
-  }	
-  axpby_ssp(chi,one/dee[Ls-1],chi,czero,chi,Ls-1,Ls-1); // Modest avoidable 
-  
-  // Apply U^{-1}
-  for (int s=Ls-2;s>=0;s--){
-    axpby_ssp_pminus (chi,one,chi,-uee[s],chi,s,s+1);  // chi[Ls]
-  }
-}
-
-template<class Impl>
-void CayleyFermion5D<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;
-  // 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(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(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(dee[s]),chi,-conjugate(leem[s]/dee[Ls-1]),chi,s,Ls-1);
-  }	
-  axpby_ssp(chi,one/conjugate(dee[Ls-1]),chi,czero,chi,Ls-1,Ls-1); // Modest avoidable 
-  
-  // Apply L^{-dagger}
-  for (int s=Ls-2;s>=0;s--){
-    axpby_ssp_pplus (chi,one,chi,-conjugate(lee[s]),chi,s,s+1);  // chi[Ls]
-  }
-}
-
-
-#ifdef CAYLEY_DPERP_LINALG
-  INSTANTIATE_DPERP(WilsonImplF);
-  INSTANTIATE_DPERP(WilsonImplD);
-  INSTANTIATE_DPERP(GparityWilsonImplF);
-  INSTANTIATE_DPERP(GparityWilsonImplD);
-  INSTANTIATE_DPERP(ZWilsonImplF);
-  INSTANTIATE_DPERP(ZWilsonImplD);
-
-  INSTANTIATE_DPERP(WilsonImplFH);
-  INSTANTIATE_DPERP(WilsonImplDF);
-  INSTANTIATE_DPERP(GparityWilsonImplFH);
-  INSTANTIATE_DPERP(GparityWilsonImplDF);
-  INSTANTIATE_DPERP(ZWilsonImplFH);
-  INSTANTIATE_DPERP(ZWilsonImplDF);
-#endif
-
-}
-}

Grid/qcd/action/fermion/CayleyFermion5Dvec.cc

@@ -1,828 +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 Grid {
-namespace QCD {  
-  /*
-   * Dense matrix versions of routines
-   */
-template<class Impl>
-void CayleyFermion5D<Impl>::MooeeInvDag (const FermionField &psi, FermionField &chi)
-{
-  this->MooeeInternal(psi,chi,DaggerYes,InverseYes);
-}
-  
-template<class Impl>
-void CayleyFermion5D<Impl>::MooeeInv(const FermionField &psi, FermionField &chi)
-{
-  this->MooeeInternal(psi,chi,DaggerNo,InverseYes);
-}
-template<class Impl>  
-void CayleyFermion5D<Impl>::M5D(const FermionField &psi,
-				const FermionField &phi, 
-				FermionField &chi,
-				std::vector<Coeff_t> &lower,
-				std::vector<Coeff_t> &diag,
-				std::vector<Coeff_t> &upper)
-{
-  GridBase *grid=psi._grid;
-  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);
-
-  chi.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);
-
-  parallel_for(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,
-				   const FermionField &phi, 
-				   FermionField &chi,
-				   std::vector<Coeff_t> &lower,
-				   std::vector<Coeff_t> &diag,
-				   std::vector<Coeff_t> &upper)
-{
-  GridBase *grid=psi._grid;
-  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);
-
-  chi.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();
-  parallel_for(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, FermionField &chi,
-					     int LLs, int site,
-					     Vector<iSinglet<Simd> > &Matp,
-					     Vector<iSinglet<Simd> > &Matm)
-{
-#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, FermionField &chi,
-					     int LLs, int site, Vector<iSinglet<Simd> > &Matp, Vector<iSinglet<Simd> > &Matm)
-{
-#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)=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
-  {
-  // 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)
-{
-  int Ls=this->Ls;
-  int LLs = psi._grid->_rdimensions[0];
-  int vol = psi._grid->oSites()/LLs;
-
-  chi.checkerboard=psi.checkerboard;
-  
-  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() ) {
-    parallel_for(auto site=0;site<vol;site++){
-      MooeeInternalZAsm(psi,chi,LLs,site,*_Matp,*_Matm);
-    }
-  } else { 
-    parallel_for(auto site=0;site<vol;site++){
-      MooeeInternalAsm(psi,chi,LLs,site,*_Matp,*_Matm);
-    }
-  }
-  MooeeInvTime+=usecond();
-}
-
-INSTANTIATE_DPERP(DomainWallVec5dImplD);
-INSTANTIATE_DPERP(DomainWallVec5dImplF);
-INSTANTIATE_DPERP(ZDomainWallVec5dImplD);
-INSTANTIATE_DPERP(ZDomainWallVec5dImplF);
-
-INSTANTIATE_DPERP(DomainWallVec5dImplDF);
-INSTANTIATE_DPERP(DomainWallVec5dImplFH);
-INSTANTIATE_DPERP(ZDomainWallVec5dImplDF);
-INSTANTIATE_DPERP(ZDomainWallVec5dImplFH);
-
-template void CayleyFermion5D<DomainWallVec5dImplF>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
-template void CayleyFermion5D<DomainWallVec5dImplD>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
-template void CayleyFermion5D<ZDomainWallVec5dImplF>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
-template void CayleyFermion5D<ZDomainWallVec5dImplD>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
-
-template void CayleyFermion5D<DomainWallVec5dImplFH>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
-template void CayleyFermion5D<DomainWallVec5dImplDF>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
-template void CayleyFermion5D<ZDomainWallVec5dImplFH>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
-template void CayleyFermion5D<ZDomainWallVec5dImplDF>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
-
-
-
-}}

Grid/qcd/action/fermion/DomainWallEOFAFermion.cc

@@ -1,438 +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 */
-
-#include <Grid/Grid_Eigen_Dense.h>
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/DomainWallEOFAFermion.h>
-
-namespace Grid {
-namespace QCD {
-
-    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)
-    {
-        int Ls = this->Ls;
-
-        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)
-    {
-        int Ls = this->Ls;
-
-        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); }
-        }
-
-        std::vector<Coeff_t> diag(Ls,1.0);
-        std::vector<Coeff_t> upper(Ls,-1.0); upper[Ls-1] = mq1 + shiftm;
-        std::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); }
-        }
-
-        std::vector<Coeff_t> diag(Ls,1.0);
-        std::vector<Coeff_t> upper(Ls,-1.0); upper[Ls-1] = mq1 + shiftp;
-        std::vector<Coeff_t> lower(Ls,-1.0); lower[0]    = mq1 + shiftm;
-
-        #if(0)
-            std::cout << GridLogMessage << "DomainWallEOFAFermion::M5Ddag(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->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;
-
-        std::vector<Coeff_t> diag = this->bee;
-        std::vector<Coeff_t> upper(Ls);
-        std::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;
-
-        std::vector<Coeff_t> diag = this->bee;
-        std::vector<Coeff_t> upper(Ls);
-        std::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, std::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;
-        }
-
-        int inv = 1;
-        this->MooeeInternalCompute(0, inv, this->MatpInv, this->MatmInv);
-        this->MooeeInternalCompute(1, inv, this->MatpInvDag, this->MatmInvDag);
-    }
-
-    // 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);
-    }
-
-    template<class Impl>
-    void DomainWallEOFAFermion<Impl>::MooeeInternalCompute(int dag, int inv,
-        Vector<iSinglet<Simd> >& Matp, Vector<iSinglet<Simd> >& Matm)
-    {
-        int Ls = this->Ls;
-
-        GridBase* grid = this->FermionRedBlackGrid();
-        int LLs = grid->_rdimensions[0];
-
-        if(LLs == Ls){ return; } // Not vectorised in 5th direction
-
-        Eigen::MatrixXcd Pplus  = Eigen::MatrixXcd::Zero(Ls,Ls);
-        Eigen::MatrixXcd Pminus = Eigen::MatrixXcd::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::MatrixXcd PplusMat ;
-        Eigen::MatrixXcd PminusMat;
-
-        #if(0)
-            std::cout << GridLogMessage << "Pplus:" << std::endl;
-            for(int s=0; s<Ls; ++s){
-                for(int ss=0; ss<Ls; ++ss){
-                    std::cout << Pplus(s,ss) << "\t";
-                }
-                std::cout << std::endl;
-            }
-            std::cout << GridLogMessage << "Pminus:" << std::endl;
-            for(int s=0; s<Ls; ++s){
-                for(int ss=0; ss<Ls; ++ss){
-                    std::cout << Pminus(s,ss) << "\t";
-                }
-                std::cout << std::endl;
-            }
-        #endif
-
-        if(inv) {
-            PplusMat  = Pplus.inverse();
-            PminusMat = Pminus.inverse();
-        } else {
-            PplusMat  = Pplus;
-            PminusMat = Pminus;
-        }
-
-        if(dag){
-            PplusMat.adjointInPlace();
-            PminusMat.adjointInPlace();
-        }
-
-        typedef typename SiteHalfSpinor::scalar_type scalar_type;
-        const int Nsimd = Simd::Nsimd();
-        Matp.resize(Ls*LLs);
-        Matm.resize(Ls*LLs);
-
-        for(int s2=0; s2<Ls; s2++){
-        for(int s1=0; s1<LLs; s1++){
-            int istride = LLs;
-            int ostride = 1;
-            Simd Vp;
-            Simd Vm;
-            scalar_type *sp = (scalar_type*) &Vp;
-            scalar_type *sm = (scalar_type*) &Vm;
-            for(int l=0; l<Nsimd; l++){
-                if(switcheroo<Coeff_t>::iscomplex()) {
-                    sp[l] = PplusMat (l*istride+s1*ostride,s2);
-                    sm[l] = PminusMat(l*istride+s1*ostride,s2);
-                } else {
-                    // if real
-                    scalar_type tmp;
-                    tmp = PplusMat (l*istride+s1*ostride,s2);
-                    sp[l] = scalar_type(tmp.real(),tmp.real());
-                    tmp = PminusMat(l*istride+s1*ostride,s2);
-                    sm[l] = scalar_type(tmp.real(),tmp.real());
-                }
-            }
-            Matp[LLs*s2+s1] = Vp;
-            Matm[LLs*s2+s1] = Vm;
-        }}
-    }
-
-    FermOpTemplateInstantiate(DomainWallEOFAFermion);
-    GparityFermOpTemplateInstantiate(DomainWallEOFAFermion);
-
-}}

Grid/qcd/action/fermion/DomainWallEOFAFermion.h

@@ -1,115 +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 */
-#ifndef  GRID_QCD_DOMAIN_WALL_EOFA_FERMION_H
-#define  GRID_QCD_DOMAIN_WALL_EOFA_FERMION_H
-
-#include <Grid/qcd/action/fermion/AbstractEOFAFermion.h>
-
-namespace Grid {
-namespace QCD {
-
-  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,
-        std::vector<Coeff_t>& lower, std::vector<Coeff_t>& diag, std::vector<Coeff_t>& upper);
-
-      void M5Ddag(const FermionField& psi, const FermionField& phi, FermionField& chi,
-        std::vector<Coeff_t>& lower, std::vector<Coeff_t>& diag, std::vector<Coeff_t>& upper);
-
-      void MooeeInternal(const FermionField& in, FermionField& out, int dag, int inv);
-
-      void MooeeInternalCompute(int dag, int inv, Vector<iSinglet<Simd>>& Matp, Vector<iSinglet<Simd>>& Matm);
-
-      void MooeeInternalAsm(const FermionField& in, FermionField& out, int LLs, int site,
-        Vector<iSinglet<Simd>>& Matp, Vector<iSinglet<Simd>>& Matm);
-
-      void MooeeInternalZAsm(const FermionField& in, FermionField& out, int LLs, int site,
-        Vector<iSinglet<Simd>>& Matp, Vector<iSinglet<Simd>>& Matm);
-
-      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, std::vector<Coeff_t>& gamma, RealD b, RealD c);
-  };
-}}
-
-#define INSTANTIATE_DPERP_DWF_EOFA(A)\
-template void DomainWallEOFAFermion<A>::M5D(const FermionField& psi, const FermionField& phi, FermionField& chi, \
-  std::vector<Coeff_t>& lower, std::vector<Coeff_t>& diag, std::vector<Coeff_t>& upper); \
-template void DomainWallEOFAFermion<A>::M5Ddag(const FermionField& psi, const FermionField& phi, FermionField& chi, \
-  std::vector<Coeff_t>& lower, std::vector<Coeff_t>& diag, std::vector<Coeff_t>& upper); \
-template void DomainWallEOFAFermion<A>::MooeeInv(const FermionField& psi, FermionField& chi); \
-template void DomainWallEOFAFermion<A>::MooeeInvDag(const FermionField& psi, FermionField& chi);
-
-#undef  DOMAIN_WALL_EOFA_DPERP_DENSE
-#define DOMAIN_WALL_EOFA_DPERP_CACHE
-#undef  DOMAIN_WALL_EOFA_DPERP_LINALG
-#define DOMAIN_WALL_EOFA_DPERP_VEC
-
-#endif

Grid/qcd/action/fermion/DomainWallEOFAFermioncache.cc

@@ -1,248 +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 Grid {
-namespace QCD {
-
-    // 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, std::vector<Coeff_t>& lower, std::vector<Coeff_t>& diag, std::vector<Coeff_t>& upper)
-    {
-        int Ls = this->Ls;
-        GridBase* grid = psi._grid;
-
-        assert(phi.checkerboard == psi.checkerboard);
-        chi.checkerboard = psi.checkerboard;
-        // Flops = 6.0*(Nc*Ns) *Ls*vol
-        this->M5Dcalls++;
-        this->M5Dtime -= usecond();
-
-        parallel_for(int ss=0; ss<grid->oSites(); ss+=Ls){ // adds Ls
-            for(int s=0; s<Ls; s++){
-                auto tmp = psi._odata[0];
-                if(s==0) {
-                    spProj5m(tmp, psi._odata[ss+s+1]);
-                    chi[ss+s] = diag[s]*phi[ss+s] + upper[s]*tmp;
-                    spProj5p(tmp, psi._odata[ss+Ls-1]);
-                    chi[ss+s] = chi[ss+s] + lower[s]*tmp;
-                } else if(s==(Ls-1)) {
-                    spProj5m(tmp, psi._odata[ss+0]);
-                    chi[ss+s] = diag[s]*phi[ss+s] + upper[s]*tmp;
-                    spProj5p(tmp, psi._odata[ss+s-1]);
-                    chi[ss+s] = chi[ss+s] + lower[s]*tmp;
-                } else {
-                    spProj5m(tmp, psi._odata[ss+s+1]);
-                    chi[ss+s] = diag[s]*phi[ss+s] + upper[s]*tmp;
-                    spProj5p(tmp, psi._odata[ss+s-1]);
-                    chi[ss+s] = chi[ss+s] + lower[s]*tmp;
-                }
-            }
-        }
-
-        this->M5Dtime += usecond();
-    }
-
-    template<class Impl>
-    void DomainWallEOFAFermion<Impl>::M5Ddag(const FermionField& psi, const FermionField& phi,
-        FermionField& chi, std::vector<Coeff_t>& lower, std::vector<Coeff_t>& diag, std::vector<Coeff_t>& upper)
-    {
-        int Ls = this->Ls;
-        GridBase* grid = psi._grid;
-        assert(phi.checkerboard == psi.checkerboard);
-        chi.checkerboard=psi.checkerboard;
-
-        // Flops = 6.0*(Nc*Ns) *Ls*vol
-        this->M5Dcalls++;
-        this->M5Dtime -= usecond();
-
-        parallel_for(int ss=0; ss<grid->oSites(); ss+=Ls){ // adds Ls
-            auto tmp = psi._odata[0];
-            for(int s=0; s<Ls; s++){
-                if(s==0) {
-                    spProj5p(tmp, psi._odata[ss+s+1]);
-                    chi[ss+s] = diag[s]*phi[ss+s] + upper[s]*tmp;
-                    spProj5m(tmp, psi._odata[ss+Ls-1]);
-                    chi[ss+s] = chi[ss+s] + lower[s]*tmp;
-                } else if(s==(Ls-1)) {
-                    spProj5p(tmp, psi._odata[ss+0]);
-                    chi[ss+s] = diag[s]*phi[ss+s] + upper[s]*tmp;
-                    spProj5m(tmp, psi._odata[ss+s-1]);
-                    chi[ss+s] = chi[ss+s] + lower[s]*tmp;
-                } else {
-                    spProj5p(tmp, psi._odata[ss+s+1]);
-                    chi[ss+s] = diag[s]*phi[ss+s] + upper[s]*tmp;
-                    spProj5m(tmp, psi._odata[ss+s-1]);
-                    chi[ss+s] = chi[ss+s] + lower[s]*tmp;
-                }
-            }
-        }
-
-        this->M5Dtime += usecond();
-    }
-
-    template<class Impl>
-    void DomainWallEOFAFermion<Impl>::MooeeInv(const FermionField& psi, FermionField& chi)
-    {
-        GridBase* grid = psi._grid;
-        int Ls = this->Ls;
-
-        chi.checkerboard = psi.checkerboard;
-
-        this->MooeeInvCalls++;
-        this->MooeeInvTime -= usecond();
-
-        parallel_for(int ss=0; ss<grid->oSites(); ss+=Ls){ // adds Ls
-
-            auto tmp1 = psi._odata[0];
-            auto tmp2 = psi._odata[0];
-
-            // flops = 12*2*Ls + 12*2*Ls + 3*12*Ls + 12*2*Ls  = 12*Ls * (9) = 108*Ls flops
-            // Apply (L^{\prime})^{-1}
-            chi[ss] = psi[ss]; // chi[0]=psi[0]
-            for(int s=1; s<Ls; s++){
-                spProj5p(tmp1, chi[ss+s-1]);
-                chi[ss+s] = psi[ss+s] - this->lee[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]);
-                chi[ss+Ls-1] = chi[ss+Ls-1] - this->leem[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]);
-                chi[ss+s] = (1.0/this->dee[s])*chi[ss+s] - (this->ueem[s]/this->dee[Ls])*tmp1;
-            }
-            spProj5m(tmp2, chi[ss+Ls-1]);
-            chi[ss+Ls-1] = (1.0/this->dee[Ls])*tmp1 + (1.0/this->dee[Ls-1])*tmp2;
-
-            // Apply U^{-1}
-            for(int s=Ls-2; s>=0; s--){
-                spProj5m(tmp1, chi[ss+s+1]);
-                chi[ss+s] = chi[ss+s] - this->uee[s]*tmp1;
-            }
-        }
-
-        this->MooeeInvTime += usecond();
-    }
-
-    template<class Impl>
-    void DomainWallEOFAFermion<Impl>::MooeeInvDag(const FermionField& psi, FermionField& chi)
-    {
-        GridBase* grid = psi._grid;
-        int Ls = this->Ls;
-
-        assert(psi.checkerboard == psi.checkerboard);
-        chi.checkerboard = psi.checkerboard;
-
-        std::vector<Coeff_t> ueec(Ls);
-        std::vector<Coeff_t> deec(Ls+1);
-        std::vector<Coeff_t> leec(Ls);
-        std::vector<Coeff_t> ueemc(Ls);
-        std::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();
-
-        parallel_for(int ss=0; ss<grid->oSites(); ss+=Ls){ // adds Ls
-
-            auto tmp1 = psi._odata[0];
-            auto tmp2 = psi._odata[0];
-
-            // Apply (U^{\prime})^{-dagger}
-            chi[ss] = psi[ss];
-            for(int s=1; s<Ls; s++){
-                spProj5m(tmp1, chi[ss+s-1]);
-                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]);
-                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]);
-                chi[ss+s] = (1.0/deec[s])*chi[ss+s] - (leemc[s]/deec[Ls-1])*tmp1;
-            }
-            spProj5p(tmp2, chi[ss+Ls-1]);
-            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]);
-                chi[ss+s] = chi[ss+s] - leec[s]*tmp1;
-            }
-        }
-
-        this->MooeeInvTime += usecond();
-    }
-
-    #ifdef DOMAIN_WALL_EOFA_DPERP_CACHE
-
-        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
-
-}}

Grid/qcd/action/fermion/DomainWallEOFAFermiondense.cc

@@ -1,159 +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 Grid {
-namespace QCD {
-
-    /*
-    * 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
-
-}}

Grid/qcd/action/fermion/DomainWallEOFAFermionssp.cc

@@ -1,168 +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 Grid {
-namespace QCD {
-
-    // 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, std::vector<Coeff_t>& lower, std::vector<Coeff_t>& diag, std::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, std::vector<Coeff_t>& lower, std::vector<Coeff_t>& diag, std::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
-
-}}

Grid/qcd/action/fermion/DomainWallEOFAFermionvec.cc

@@ -1,605 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/DomainWallEOFAFermionvec.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 Grid {
-namespace QCD {
-
-    /*
-    * 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>::M5D(const FermionField& psi, const FermionField& phi,
-        FermionField& chi, std::vector<Coeff_t>& lower, std::vector<Coeff_t>& diag, std::vector<Coeff_t>& upper)
-    {
-        GridBase* grid = psi._grid;
-        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);
-
-        chi.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];
-        }}
-
-        this->M5Dcalls++;
-        this->M5Dtime -= usecond();
-
-        assert(Nc == 3);
-
-        parallel_for(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
-        }
-
-        this->M5Dtime += usecond();
-    }
-
-    template<class Impl>
-    void DomainWallEOFAFermion<Impl>::M5Ddag(const FermionField& psi, const FermionField& phi,
-        FermionField& chi, std::vector<Coeff_t>& lower, std::vector<Coeff_t>& diag, std::vector<Coeff_t>& upper)
-    {
-        GridBase* grid = psi._grid;
-        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);
-
-        chi.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];
-        }}
-
-        this->M5Dcalls++;
-        this->M5Dtime -= usecond();
-
-        parallel_for(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
-
-        }
-
-        this->M5Dtime += usecond();
-    }
-
-    #ifdef AVX512
-        #include<simd/Intel512common.h>
-        #include<simd/Intel512avx.h>
-        #include<simd/Intel512single.h>
-    #endif
-
-    template<class Impl>
-    void DomainWallEOFAFermion<Impl>::MooeeInternalAsm(const FermionField& psi, FermionField& chi,
-        int LLs, int site, Vector<iSinglet<Simd> >& Matp, Vector<iSinglet<Simd> >& Matm)
-    {
-        #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 DomainWallEOFAFermion<Impl>::MooeeInternalZAsm(const FermionField& psi, FermionField& chi,
-        int LLs, int site, Vector<iSinglet<Simd> >& Matp, Vector<iSinglet<Simd> >& Matm)
-    {
-        std::cout << "Error: zMobius not implemented for EOFA" << std::endl;
-        exit(-1);
-    };
-
-    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;
-
-        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 = &this->MatpInvDag;
-            _Matm = &this->MatmInvDag;
-        }
-
-        if(inv && (!dag)){
-            _Matp = &this->MatpInv;
-            _Matm = &this->MatmInv;
-        }
-
-        if(!inv){
-            MooeeInternalCompute(dag, inv, Matp, Matm);
-            _Matp = &Matp;
-            _Matm = &Matm;
-        }
-
-        assert(_Matp->size() == Ls*LLs);
-
-        this->MooeeInvCalls++;
-        this->MooeeInvTime -= usecond();
-
-        if(switcheroo<Coeff_t>::iscomplex()){
-            parallel_for(auto site=0; site<vol; site++){
-                MooeeInternalZAsm(psi, chi, LLs, site, *_Matp, *_Matm);
-            }
-        } else {
-            parallel_for(auto site=0; site<vol; site++){
-                MooeeInternalAsm(psi, chi, LLs, site, *_Matp, *_Matm);
-            }
-        }
-
-        this->MooeeInvTime += usecond();
-    }
-
-    #ifdef DOMAIN_WALL_EOFA_DPERP_VEC
-
-        INSTANTIATE_DPERP_DWF_EOFA(DomainWallVec5dImplD);
-        INSTANTIATE_DPERP_DWF_EOFA(DomainWallVec5dImplF);
-        INSTANTIATE_DPERP_DWF_EOFA(ZDomainWallVec5dImplD);
-        INSTANTIATE_DPERP_DWF_EOFA(ZDomainWallVec5dImplF);
-
-        INSTANTIATE_DPERP_DWF_EOFA(DomainWallVec5dImplDF);
-        INSTANTIATE_DPERP_DWF_EOFA(DomainWallVec5dImplFH);
-        INSTANTIATE_DPERP_DWF_EOFA(ZDomainWallVec5dImplDF);
-        INSTANTIATE_DPERP_DWF_EOFA(ZDomainWallVec5dImplFH);
-
-        template void DomainWallEOFAFermion<DomainWallVec5dImplF>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-        template void DomainWallEOFAFermion<DomainWallVec5dImplD>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-        template void DomainWallEOFAFermion<ZDomainWallVec5dImplF>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-        template void DomainWallEOFAFermion<ZDomainWallVec5dImplD>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-
-        template void DomainWallEOFAFermion<DomainWallVec5dImplFH>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-        template void DomainWallEOFAFermion<DomainWallVec5dImplDF>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-        template void DomainWallEOFAFermion<ZDomainWallVec5dImplFH>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-        template void DomainWallEOFAFermion<ZDomainWallVec5dImplDF>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-
-    #endif
-
-}}

Grid/qcd/action/fermion/DomainWallFermion.h

@@ -1,142 +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 Grid {
-
-  namespace QCD {
-
-    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;
-	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(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);
-      }
-
-    };
-
-  }
-}
-
-#endif

Grid/qcd/action/fermion/Fermion.h

@@ -1,335 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: ./lib/qcd/action/fermion/Fermion_base_aggregate.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 */
-#ifndef  GRID_QCD_FERMION_H
-#define  GRID_QCD_FERMION_H
-
-////////////////////////////////////////////////////////////////////////////////////////////////////
-// Explicit explicit template instantiation is still required in the .cc files
-//
-// - CayleyFermion5D.cc
-// - PartialFractionFermion5D.cc
-// - WilsonFermion5D.cc
-// - WilsonKernelsHand.cc
-// - ContinuedFractionFermion5D.cc
-// - WilsonFermion.cc
-// - WilsonKernels.cc
-// - DomainWallEOFAFermion.cc
-// - MobiusEOFAFermion.cc
-//
-// The explicit instantiation is only avoidable if we move this source to headers and end up with include/parse/recompile
-// for EVERY .cc file. This define centralises the list and restores global push of impl cases
-////////////////////////////////////////////////////////////////////////////////////////////////////
-
-////////////////////////////////////////////
-// Fermion operators / actions
-////////////////////////////////////////////
-
-#include <Grid/qcd/action/fermion/WilsonFermion.h>       // 4d wilson like
-#include <Grid/qcd/action/fermion/WilsonTMFermion.h>     // 4d wilson like
-#include <Grid/qcd/action/fermion/WilsonCloverFermion.h> // 4d wilson clover fermions
-#include <Grid/qcd/action/fermion/WilsonFermion5D.h>     // 5d base used by all 5d overlap types
-
-#include <Grid/qcd/action/fermion/ImprovedStaggeredFermion.h>
-#include <Grid/qcd/action/fermion/ImprovedStaggeredFermion5D.h>
-
-#include <Grid/qcd/action/fermion/CayleyFermion5D.h>     // Cayley types
-#include <Grid/qcd/action/fermion/DomainWallFermion.h>
-#include <Grid/qcd/action/fermion/DomainWallEOFAFermion.h>
-#include <Grid/qcd/action/fermion/MobiusFermion.h>
-#include <Grid/qcd/action/fermion/MobiusEOFAFermion.h>
-#include <Grid/qcd/action/fermion/ZMobiusFermion.h>
-#include <Grid/qcd/action/fermion/SchurDiagTwoKappa.h>
-#include <Grid/qcd/action/fermion/ScaledShamirFermion.h>
-#include <Grid/qcd/action/fermion/MobiusZolotarevFermion.h>
-#include <Grid/qcd/action/fermion/ShamirZolotarevFermion.h>
-#include <Grid/qcd/action/fermion/OverlapWilsonCayleyTanhFermion.h>
-#include <Grid/qcd/action/fermion/OverlapWilsonCayleyZolotarevFermion.h>
-#include <Grid/qcd/action/fermion/ContinuedFractionFermion5D.h>               // Continued fraction
-#include <Grid/qcd/action/fermion/OverlapWilsonContfracTanhFermion.h>
-#include <Grid/qcd/action/fermion/OverlapWilsonContfracZolotarevFermion.h>
-#include <Grid/qcd/action/fermion/PartialFractionFermion5D.h>                 // Partial fraction
-#include <Grid/qcd/action/fermion/OverlapWilsonPartialFractionTanhFermion.h>
-#include <Grid/qcd/action/fermion/OverlapWilsonPartialFractionZolotarevFermion.h>
-///////////////////////////////////////////////////////////////////////////////
-// G5 herm -- this has to live in QCD since dirac matrix is not in the broader sector of code
-///////////////////////////////////////////////////////////////////////////////
-#include <Grid/qcd/action/fermion/g5HermitianLinop.h>
-
-///////////////////////////////////////////////////////////////////////////////
-// Fourier accelerated Pauli Villars inverse support
-///////////////////////////////////////////////////////////////////////////////
-#include <Grid/qcd/action/fermion/WilsonTMFermion5D.h>   
-
-////////////////////////////////////////////////////////////////////////////////
-// Move this group to a DWF specific tools/algorithms subdir? 
-////////////////////////////////////////////////////////////////////////////////
-#include <Grid/qcd/action/fermion/FourierAcceleratedPV.h>
-#include <Grid/qcd/action/fermion/PauliVillarsInverters.h>
-#include <Grid/qcd/action/fermion/Reconstruct5Dprop.h>
-#include <Grid/qcd/action/fermion/MADWF.h>
-
-////////////////////////////////////////////////////////////////////////////////////////////////////
-// More maintainable to maintain the following typedef list centrally, as more "impl" targets
-// are added, (e.g. extension for gparity, half precision project in comms etc..)
-////////////////////////////////////////////////////////////////////////////////////////////////////
-
-// Cayley 5d
-namespace Grid {
-  namespace QCD {
-
-typedef WilsonFermion<WilsonImplR> WilsonFermionR;
-typedef WilsonFermion<WilsonImplF> WilsonFermionF;
-typedef WilsonFermion<WilsonImplD> WilsonFermionD;
-
-typedef WilsonFermion<WilsonImplRL> WilsonFermionRL;
-typedef WilsonFermion<WilsonImplFH> WilsonFermionFH;
-typedef WilsonFermion<WilsonImplDF> WilsonFermionDF;
-
-typedef WilsonFermion<WilsonAdjImplR> WilsonAdjFermionR;
-typedef WilsonFermion<WilsonAdjImplF> WilsonAdjFermionF;
-typedef WilsonFermion<WilsonAdjImplD> WilsonAdjFermionD;
-
-typedef WilsonFermion<WilsonTwoIndexSymmetricImplR> WilsonTwoIndexSymmetricFermionR;
-typedef WilsonFermion<WilsonTwoIndexSymmetricImplF> WilsonTwoIndexSymmetricFermionF;
-typedef WilsonFermion<WilsonTwoIndexSymmetricImplD> WilsonTwoIndexSymmetricFermionD;
-
-typedef WilsonFermion<WilsonTwoIndexAntiSymmetricImplR> WilsonTwoIndexAntiSymmetricFermionR;
-typedef WilsonFermion<WilsonTwoIndexAntiSymmetricImplF> WilsonTwoIndexAntiSymmetricFermionF;
-typedef WilsonFermion<WilsonTwoIndexAntiSymmetricImplD> WilsonTwoIndexAntiSymmetricFermionD;
-
-// Twisted mass fermion
-typedef WilsonTMFermion<WilsonImplR> WilsonTMFermionR;
-typedef WilsonTMFermion<WilsonImplF> WilsonTMFermionF;
-typedef WilsonTMFermion<WilsonImplD> WilsonTMFermionD;
-
-// Clover fermions
-typedef WilsonCloverFermion<WilsonImplR> WilsonCloverFermionR;
-typedef WilsonCloverFermion<WilsonImplF> WilsonCloverFermionF;
-typedef WilsonCloverFermion<WilsonImplD> WilsonCloverFermionD;
-
-typedef WilsonCloverFermion<WilsonAdjImplR> WilsonCloverAdjFermionR;
-typedef WilsonCloverFermion<WilsonAdjImplF> WilsonCloverAdjFermionF;
-typedef WilsonCloverFermion<WilsonAdjImplD> WilsonCloverAdjFermionD;
-
-typedef WilsonCloverFermion<WilsonTwoIndexSymmetricImplR> WilsonCloverTwoIndexSymmetricFermionR;
-typedef WilsonCloverFermion<WilsonTwoIndexSymmetricImplF> WilsonCloverTwoIndexSymmetricFermionF;
-typedef WilsonCloverFermion<WilsonTwoIndexSymmetricImplD> WilsonCloverTwoIndexSymmetricFermionD;
-
-typedef WilsonCloverFermion<WilsonTwoIndexAntiSymmetricImplR> WilsonCloverTwoIndexAntiSymmetricFermionR;
-typedef WilsonCloverFermion<WilsonTwoIndexAntiSymmetricImplF> WilsonCloverTwoIndexAntiSymmetricFermionF;
-typedef WilsonCloverFermion<WilsonTwoIndexAntiSymmetricImplD> WilsonCloverTwoIndexAntiSymmetricFermionD;
-
-// Domain Wall fermions
-typedef DomainWallFermion<WilsonImplR> DomainWallFermionR;
-typedef DomainWallFermion<WilsonImplF> DomainWallFermionF;
-typedef DomainWallFermion<WilsonImplD> DomainWallFermionD;
-
-typedef DomainWallFermion<WilsonImplRL> DomainWallFermionRL;
-typedef DomainWallFermion<WilsonImplFH> DomainWallFermionFH;
-typedef DomainWallFermion<WilsonImplDF> DomainWallFermionDF;
-
-typedef DomainWallEOFAFermion<WilsonImplR> DomainWallEOFAFermionR;
-typedef DomainWallEOFAFermion<WilsonImplF> DomainWallEOFAFermionF;
-typedef DomainWallEOFAFermion<WilsonImplD> DomainWallEOFAFermionD;
-
-typedef DomainWallEOFAFermion<WilsonImplRL> DomainWallEOFAFermionRL;
-typedef DomainWallEOFAFermion<WilsonImplFH> DomainWallEOFAFermionFH;
-typedef DomainWallEOFAFermion<WilsonImplDF> DomainWallEOFAFermionDF;
-
-typedef MobiusFermion<WilsonImplR> MobiusFermionR;
-typedef MobiusFermion<WilsonImplF> MobiusFermionF;
-typedef MobiusFermion<WilsonImplD> MobiusFermionD;
-
-typedef MobiusFermion<WilsonImplRL> MobiusFermionRL;
-typedef MobiusFermion<WilsonImplFH> MobiusFermionFH;
-typedef MobiusFermion<WilsonImplDF> MobiusFermionDF;
-
-typedef MobiusEOFAFermion<WilsonImplR> MobiusEOFAFermionR;
-typedef MobiusEOFAFermion<WilsonImplF> MobiusEOFAFermionF;
-typedef MobiusEOFAFermion<WilsonImplD> MobiusEOFAFermionD;
-
-typedef MobiusEOFAFermion<WilsonImplRL> MobiusEOFAFermionRL;
-typedef MobiusEOFAFermion<WilsonImplFH> MobiusEOFAFermionFH;
-typedef MobiusEOFAFermion<WilsonImplDF> MobiusEOFAFermionDF;
-
-typedef ZMobiusFermion<ZWilsonImplR> ZMobiusFermionR;
-typedef ZMobiusFermion<ZWilsonImplF> ZMobiusFermionF;
-typedef ZMobiusFermion<ZWilsonImplD> ZMobiusFermionD;
-
-typedef ZMobiusFermion<ZWilsonImplRL> ZMobiusFermionRL;
-typedef ZMobiusFermion<ZWilsonImplFH> ZMobiusFermionFH;
-typedef ZMobiusFermion<ZWilsonImplDF> ZMobiusFermionDF;
-
-// Ls vectorised
-typedef DomainWallFermion<DomainWallVec5dImplR> DomainWallFermionVec5dR;
-typedef DomainWallFermion<DomainWallVec5dImplF> DomainWallFermionVec5dF;
-typedef DomainWallFermion<DomainWallVec5dImplD> DomainWallFermionVec5dD;
-
-typedef DomainWallFermion<DomainWallVec5dImplRL> DomainWallFermionVec5dRL;
-typedef DomainWallFermion<DomainWallVec5dImplFH> DomainWallFermionVec5dFH;
-typedef DomainWallFermion<DomainWallVec5dImplDF> DomainWallFermionVec5dDF;
-
-typedef DomainWallEOFAFermion<DomainWallVec5dImplR> DomainWallEOFAFermionVec5dR;
-typedef DomainWallEOFAFermion<DomainWallVec5dImplF> DomainWallEOFAFermionVec5dF;
-typedef DomainWallEOFAFermion<DomainWallVec5dImplD> DomainWallEOFAFermionVec5dD;
-
-typedef DomainWallEOFAFermion<DomainWallVec5dImplRL> DomainWallEOFAFermionVec5dRL;
-typedef DomainWallEOFAFermion<DomainWallVec5dImplFH> DomainWallEOFAFermionVec5dFH;
-typedef DomainWallEOFAFermion<DomainWallVec5dImplDF> DomainWallEOFAFermionVec5dDF;
-
-typedef MobiusFermion<DomainWallVec5dImplR> MobiusFermionVec5dR;
-typedef MobiusFermion<DomainWallVec5dImplF> MobiusFermionVec5dF;
-typedef MobiusFermion<DomainWallVec5dImplD> MobiusFermionVec5dD;
-
-typedef MobiusFermion<DomainWallVec5dImplRL> MobiusFermionVec5dRL;
-typedef MobiusFermion<DomainWallVec5dImplFH> MobiusFermionVec5dFH;
-typedef MobiusFermion<DomainWallVec5dImplDF> MobiusFermionVec5dDF;
-
-typedef MobiusEOFAFermion<DomainWallVec5dImplR> MobiusEOFAFermionVec5dR;
-typedef MobiusEOFAFermion<DomainWallVec5dImplF> MobiusEOFAFermionVec5dF;
-typedef MobiusEOFAFermion<DomainWallVec5dImplD> MobiusEOFAFermionVec5dD;
-
-typedef MobiusEOFAFermion<DomainWallVec5dImplRL> MobiusEOFAFermionVec5dRL;
-typedef MobiusEOFAFermion<DomainWallVec5dImplFH> MobiusEOFAFermionVec5dFH;
-typedef MobiusEOFAFermion<DomainWallVec5dImplDF> MobiusEOFAFermionVec5dDF;
-
-typedef ZMobiusFermion<ZDomainWallVec5dImplR> ZMobiusFermionVec5dR;
-typedef ZMobiusFermion<ZDomainWallVec5dImplF> ZMobiusFermionVec5dF;
-typedef ZMobiusFermion<ZDomainWallVec5dImplD> ZMobiusFermionVec5dD;
-
-typedef ZMobiusFermion<ZDomainWallVec5dImplRL> ZMobiusFermionVec5dRL;
-typedef ZMobiusFermion<ZDomainWallVec5dImplFH> ZMobiusFermionVec5dFH;
-typedef ZMobiusFermion<ZDomainWallVec5dImplDF> ZMobiusFermionVec5dDF;
-
-typedef ScaledShamirFermion<WilsonImplR> ScaledShamirFermionR;
-typedef ScaledShamirFermion<WilsonImplF> ScaledShamirFermionF;
-typedef ScaledShamirFermion<WilsonImplD> ScaledShamirFermionD;
-
-typedef MobiusZolotarevFermion<WilsonImplR> MobiusZolotarevFermionR;
-typedef MobiusZolotarevFermion<WilsonImplF> MobiusZolotarevFermionF;
-typedef MobiusZolotarevFermion<WilsonImplD> MobiusZolotarevFermionD;
-typedef ShamirZolotarevFermion<WilsonImplR> ShamirZolotarevFermionR;
-typedef ShamirZolotarevFermion<WilsonImplF> ShamirZolotarevFermionF;
-typedef ShamirZolotarevFermion<WilsonImplD> ShamirZolotarevFermionD;
-
-typedef OverlapWilsonCayleyTanhFermion<WilsonImplR> OverlapWilsonCayleyTanhFermionR;
-typedef OverlapWilsonCayleyTanhFermion<WilsonImplF> OverlapWilsonCayleyTanhFermionF;
-typedef OverlapWilsonCayleyTanhFermion<WilsonImplD> OverlapWilsonCayleyTanhFermionD;
-typedef OverlapWilsonCayleyZolotarevFermion<WilsonImplR> OverlapWilsonCayleyZolotarevFermionR;
-typedef OverlapWilsonCayleyZolotarevFermion<WilsonImplF> OverlapWilsonCayleyZolotarevFermionF;
-typedef OverlapWilsonCayleyZolotarevFermion<WilsonImplD> OverlapWilsonCayleyZolotarevFermionD;
-
-// Continued fraction
-typedef OverlapWilsonContFracTanhFermion<WilsonImplR> OverlapWilsonContFracTanhFermionR;
-typedef OverlapWilsonContFracTanhFermion<WilsonImplF> OverlapWilsonContFracTanhFermionF;
-typedef OverlapWilsonContFracTanhFermion<WilsonImplD> OverlapWilsonContFracTanhFermionD;
-typedef OverlapWilsonContFracZolotarevFermion<WilsonImplR> OverlapWilsonContFracZolotarevFermionR;
-typedef OverlapWilsonContFracZolotarevFermion<WilsonImplF> OverlapWilsonContFracZolotarevFermionF;
-typedef OverlapWilsonContFracZolotarevFermion<WilsonImplD> OverlapWilsonContFracZolotarevFermionD;
-
-// Partial fraction
-typedef OverlapWilsonPartialFractionTanhFermion<WilsonImplR> OverlapWilsonPartialFractionTanhFermionR;
-typedef OverlapWilsonPartialFractionTanhFermion<WilsonImplF> OverlapWilsonPartialFractionTanhFermionF;
-typedef OverlapWilsonPartialFractionTanhFermion<WilsonImplD> OverlapWilsonPartialFractionTanhFermionD;
-
-typedef OverlapWilsonPartialFractionZolotarevFermion<WilsonImplR> OverlapWilsonPartialFractionZolotarevFermionR;
-typedef OverlapWilsonPartialFractionZolotarevFermion<WilsonImplF> OverlapWilsonPartialFractionZolotarevFermionF;
-typedef OverlapWilsonPartialFractionZolotarevFermion<WilsonImplD> OverlapWilsonPartialFractionZolotarevFermionD;
-
-// Gparity cases; partial list until tested
-typedef WilsonFermion<GparityWilsonImplR>     GparityWilsonFermionR;
-typedef WilsonFermion<GparityWilsonImplF>     GparityWilsonFermionF;
-typedef WilsonFermion<GparityWilsonImplD>     GparityWilsonFermionD;
-
-typedef WilsonFermion<GparityWilsonImplRL>     GparityWilsonFermionRL;
-typedef WilsonFermion<GparityWilsonImplFH>     GparityWilsonFermionFH;
-typedef WilsonFermion<GparityWilsonImplDF>     GparityWilsonFermionDF;
-
-typedef DomainWallFermion<GparityWilsonImplR> GparityDomainWallFermionR;
-typedef DomainWallFermion<GparityWilsonImplF> GparityDomainWallFermionF;
-typedef DomainWallFermion<GparityWilsonImplD> GparityDomainWallFermionD;
-
-typedef DomainWallFermion<GparityWilsonImplRL> GparityDomainWallFermionRL;
-typedef DomainWallFermion<GparityWilsonImplFH> GparityDomainWallFermionFH;
-typedef DomainWallFermion<GparityWilsonImplDF> GparityDomainWallFermionDF;
-
-typedef DomainWallEOFAFermion<GparityWilsonImplR> GparityDomainWallEOFAFermionR;
-typedef DomainWallEOFAFermion<GparityWilsonImplF> GparityDomainWallEOFAFermionF;
-typedef DomainWallEOFAFermion<GparityWilsonImplD> GparityDomainWallEOFAFermionD;
-
-typedef DomainWallEOFAFermion<GparityWilsonImplRL> GparityDomainWallEOFAFermionRL;
-typedef DomainWallEOFAFermion<GparityWilsonImplFH> GparityDomainWallEOFAFermionFH;
-typedef DomainWallEOFAFermion<GparityWilsonImplDF> GparityDomainWallEOFAFermionDF;
-
-typedef WilsonTMFermion<GparityWilsonImplR> GparityWilsonTMFermionR;
-typedef WilsonTMFermion<GparityWilsonImplF> GparityWilsonTMFermionF;
-typedef WilsonTMFermion<GparityWilsonImplD> GparityWilsonTMFermionD;
-
-typedef WilsonTMFermion<GparityWilsonImplRL> GparityWilsonTMFermionRL;
-typedef WilsonTMFermion<GparityWilsonImplFH> GparityWilsonTMFermionFH;
-typedef WilsonTMFermion<GparityWilsonImplDF> GparityWilsonTMFermionDF;
-
-typedef MobiusFermion<GparityWilsonImplR> GparityMobiusFermionR;
-typedef MobiusFermion<GparityWilsonImplF> GparityMobiusFermionF;
-typedef MobiusFermion<GparityWilsonImplD> GparityMobiusFermionD;
-
-typedef MobiusFermion<GparityWilsonImplRL> GparityMobiusFermionRL;
-typedef MobiusFermion<GparityWilsonImplFH> GparityMobiusFermionFH;
-typedef MobiusFermion<GparityWilsonImplDF> GparityMobiusFermionDF;
-
-typedef MobiusEOFAFermion<GparityWilsonImplR> GparityMobiusEOFAFermionR;
-typedef MobiusEOFAFermion<GparityWilsonImplF> GparityMobiusEOFAFermionF;
-typedef MobiusEOFAFermion<GparityWilsonImplD> GparityMobiusEOFAFermionD;
-
-typedef MobiusEOFAFermion<GparityWilsonImplRL> GparityMobiusEOFAFermionRL;
-typedef MobiusEOFAFermion<GparityWilsonImplFH> GparityMobiusEOFAFermionFH;
-typedef MobiusEOFAFermion<GparityWilsonImplDF> GparityMobiusEOFAFermionDF;
-
-typedef ImprovedStaggeredFermion<StaggeredImplR> ImprovedStaggeredFermionR;
-typedef ImprovedStaggeredFermion<StaggeredImplF> ImprovedStaggeredFermionF;
-typedef ImprovedStaggeredFermion<StaggeredImplD> ImprovedStaggeredFermionD;
-
-typedef ImprovedStaggeredFermion5D<StaggeredImplR> ImprovedStaggeredFermion5DR;
-typedef ImprovedStaggeredFermion5D<StaggeredImplF> ImprovedStaggeredFermion5DF;
-typedef ImprovedStaggeredFermion5D<StaggeredImplD> ImprovedStaggeredFermion5DD;
-
-typedef ImprovedStaggeredFermion5D<StaggeredVec5dImplR> ImprovedStaggeredFermionVec5dR;
-typedef ImprovedStaggeredFermion5D<StaggeredVec5dImplF> ImprovedStaggeredFermionVec5dF;
-typedef ImprovedStaggeredFermion5D<StaggeredVec5dImplD> ImprovedStaggeredFermionVec5dD;
-
-
-  }}
-
-////////////////////
-// Scalar QED actions
-// TODO: this needs to move to another header after rename to Fermion.h
-////////////////////
-#include <Grid/qcd/action/scalar/Scalar.h>
-#include <Grid/qcd/action/gauge/Photon.h>
-
-#endif

Grid/qcd/action/fermion/FermionCore.h

@@ -1,93 +0,0 @@
-    /*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/Fermion_base_aggregate.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 */
-#ifndef  GRID_QCD_FERMION_CORE_H
-#define  GRID_QCD_FERMION_CORE_H
-
-#include <Grid/GridCore.h>
-#include <Grid/GridQCDcore.h>
-#include <Grid/qcd/action/ActionCore.h>
-
-////////////////////////////////////////////
-// Fermion prereqs
-////////////////////////////////////////////
-#include <Grid/qcd/action/fermion/WilsonCompressor.h>     //used by all wilson type fermions
-#include <Grid/qcd/action/fermion/FermionOperatorImpl.h>
-#include <Grid/qcd/action/fermion/FermionOperator.h>
-#include <Grid/qcd/action/fermion/WilsonKernels.h>        //used by all wilson type fermions
-#include <Grid/qcd/action/fermion/StaggeredKernels.h>        //used by all wilson type fermions
-
-#define FermOpStaggeredTemplateInstantiate(A) \
-  template class A<StaggeredImplF>; \
-  template class A<StaggeredImplD>; 
-
-#define FermOpStaggeredVec5dTemplateInstantiate(A) \
-  template class A<StaggeredVec5dImplF>; \
-  template class A<StaggeredVec5dImplD>; 
-
-#define FermOp4dVecTemplateInstantiate(A) \
-  template class A<WilsonImplF>;		\
-  template class A<WilsonImplD>;		\
-  template class A<ZWilsonImplF>;		\
-  template class A<ZWilsonImplD>;		\
-  template class A<GparityWilsonImplF>;		\
-  template class A<GparityWilsonImplD>;		\
-  template class A<WilsonImplFH>;		\
-  template class A<WilsonImplDF>;		\
-  template class A<ZWilsonImplFH>;		\
-  template class A<ZWilsonImplDF>;		\
-  template class A<GparityWilsonImplFH>;		\
-  template class A<GparityWilsonImplDF>;		
-
-
-#define AdjointFermOpTemplateInstantiate(A) \
-  template class A<WilsonAdjImplF>; \
-  template class A<WilsonAdjImplD>; 
-
-#define TwoIndexFermOpTemplateInstantiate(A) \
-  template class A<WilsonTwoIndexSymmetricImplF>; \
-  template class A<WilsonTwoIndexSymmetricImplD>; \
-  template class A<WilsonTwoIndexAntiSymmetricImplF>; \
-  template class A<WilsonTwoIndexAntiSymmetricImplD>;
-
-#define FermOp5dVecTemplateInstantiate(A) \
-  template class A<DomainWallVec5dImplF>;	\
-  template class A<DomainWallVec5dImplD>;	\
-  template class A<ZDomainWallVec5dImplF>;	\
-  template class A<ZDomainWallVec5dImplD>;	\
-  template class A<DomainWallVec5dImplFH>;	\
-  template class A<DomainWallVec5dImplDF>;	\
-  template class A<ZDomainWallVec5dImplFH>;	\
-  template class A<ZDomainWallVec5dImplDF>;	
-
-#define FermOpTemplateInstantiate(A) \
- FermOp4dVecTemplateInstantiate(A) \
- FermOp5dVecTemplateInstantiate(A) 
-
-#define GparityFermOpTemplateInstantiate(A) 
-
-#endif

Grid/qcd/action/fermion/FourierAcceleratedPV.h

@@ -1,237 +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 Grid {
-namespace QCD {
-
-  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 = QCD::SpaceTimeGrid::makeFiveDimGrid(2*group_in_s, (GridCartesian*)Umu._grid);
-    gridRB5D = QCD::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)*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;
-  }
-
-};
-}}

Grid/qcd/action/fermion/ImprovedStaggeredFermion.cc

@@ -1,604 +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>
-
-namespace Grid {
-namespace QCD {
-
-const std::vector<int> 
-ImprovedStaggeredFermionStatic::directions({0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3});
-const std::vector<int> 
-ImprovedStaggeredFermionStatic::displacements({1, 1, 1, 1, -1, -1, -1, -1, 3, 3, 3, 3, -3, -3, -3, -3});
-
-/////////////////////////////////
-// 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),
-      StencilEven(&Hgrid, npoint, Even, directions, displacements),  // source is Even
-      StencilOdd(&Hgrid, npoint, Odd, directions, displacements),  // 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
-    ////////////////////////
-    PARALLEL_FOR_LOOP
-    for (int sss = 0; sss < B._grid->oSites(); sss++) {
-      Kernels::DhopDir(st, U, UUU, st.CommBuf(), sss, sss, B, Btilde, 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);
-
-  PARALLEL_FOR_LOOP
-  for (int sss = 0; sss < in._grid->oSites(); sss++) {
-    Kernels::DhopDir(Stencil, Umu, UUUmu, Stencil.CommBuf(), sss, sss, in, out, 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
-      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,UUU,st.CommBuf(),1,sU,in,out,1,0); 
-        }
-      } else {
-        for (int ss = myblock; ss < myblock+myn; ++ss) {
-	  // Interior = 1; Exterior = 0;
-          int sU = ss;
-          Kernels::DhopSite(st,lo,U,UUU,st.CommBuf(),1,sU,in,out,1,0);
-        }
-      }
-    } else {
-      st.CommunicateThreaded();
-    }
-  }
-  DhopComputeTime    += usecond();
-
-  // First to enter, last to leave timing
-  DhopFaceTime    -= usecond();
-  st.CommsMerge(compressor);
-  DhopFaceTime    -= usecond();
-
-  DhopComputeTime2    -= usecond();
-  if (dag == DaggerYes) {
-    int sz=st.surface_list.size();
-    parallel_for (int ss = 0; ss < sz; ss++) {
-      int sU = st.surface_list[ss];
-      Kernels::DhopSiteDag(st,lo,U,UUU,st.CommBuf(),1,sU,in,out,0,1);
-    }
-  } else {
-    int sz=st.surface_list.size();
-    parallel_for (int ss = 0; ss < sz; ss++) {
-      int sU = st.surface_list[ss];
-      Kernels::DhopSite(st,lo,U,UUU,st.CommBuf(),1,sU,in,out,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();
-
-  DhopComputeTime -= usecond();
-  if (dag == DaggerYes) {
-    parallel_for (int sss = 0; sss < in._grid->oSites(); sss++) {
-      Kernels::DhopSiteDag(st, lo, U, UUU, st.CommBuf(), 1, sss, in, out);
-    }
-  } else {
-    parallel_for (int sss = 0; sss < in._grid->oSites(); sss++) {
-      Kernels::DhopSite(st, lo, U, UUU, st.CommBuf(), 1, sss, in, out);
-    }
-  }
-  DhopComputeTime += usecond();
-  DhopTotalTime   += usecond();
-};
-
-  ////////////////////////////////////////////////////////////////
-  // Reporting
-  ////////////////////////////////////////////////////////////////
-template<class Impl>
-void ImprovedStaggeredFermion<Impl>::Report(void) 
-{
-  std::vector<int> latt = GridDefaultLatt();          
-  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);
-
-}
-
-
-FermOpStaggeredTemplateInstantiate(ImprovedStaggeredFermion);
-
-  //AdjointFermOpTemplateInstantiate(ImprovedStaggeredFermion);
-  //TwoIndexFermOpTemplateInstantiate(ImprovedStaggeredFermion);
-
-}}

Grid/qcd/action/fermion/ImprovedStaggeredFermion.h

@@ -1,205 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/ImprovedStaggered.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 */
-#ifndef GRID_QCD_IMPR_STAG_FERMION_H
-#define GRID_QCD_IMPR_STAG_FERMION_H
-
-namespace Grid {
-
-namespace QCD {
-
-class ImprovedStaggeredFermionStatic {
- public:
-  static const std::vector<int> directions;
-  static const std::vector<int> displacements;
-  static const int npoint = 16;
-};
-
-template <class Impl>
-class ImprovedStaggeredFermion : public StaggeredKernels<Impl>, public ImprovedStaggeredFermionStatic {
- 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 _grid; }
-  GridBase *GaugeRedBlackGrid(void) { return _cbgrid; }
-  GridBase *FermionGrid(void) { return _grid; }
-  GridBase *FermionRedBlackGrid(void) { return _cbgrid; }
-
-  //////////////////////////////////////////////////////////////////
-  // override multiply; cut number routines if pass dagger argument
-  // and also make interface more uniformly consistent
-  //////////////////////////////////////////////////////////////////
-  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 MeooeDag(const FermionField &in, FermionField &out);
-  void Mooee(const FermionField &in, FermionField &out);
-  void MooeeDag(const FermionField &in, FermionField &out);
-  void MooeeInv(const FermionField &in, FermionField &out);
-  void MooeeInvDag(const FermionField &in, FermionField &out);
-
-  ////////////////////////
-  // Derivative interface
-  ////////////////////////
-  // Interface calls an internal routine
-  void DhopDeriv  (GaugeField &mat, const FermionField &U, const FermionField &V, int dag);
-  void DhopDerivOE(GaugeField &mat, const FermionField &U, const FermionField &V, int dag);
-  void DhopDerivEO(GaugeField &mat, const FermionField &U, const FermionField &V, int dag);
-
-  ///////////////////////////////////////////////////////////////
-  // 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);
-
-  ///////////////////////////////////////////////////////////////
-  // Multigrid assistance; force term uses too
-  ///////////////////////////////////////////////////////////////
-  void Mdir(const FermionField &in, FermionField &out, int dir, int disp);
-  void DhopDir(const FermionField &in, FermionField &out, int dir, int disp);
-
-  ///////////////////////////////////////////////////////////////
-  // Extra methods added by derived
-  ///////////////////////////////////////////////////////////////
-  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 DhopInternalSerialComms(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);
-
-  //////////////////////////////////////////////////////////////////////////
-  // Grid own interface Constructor
-  //////////////////////////////////////////////////////////////////////////
-  ImprovedStaggeredFermion(GaugeField &_Uthin, GaugeField &_Ufat, GridCartesian &Fgrid,
-			   GridRedBlackCartesian &Hgrid, RealD _mass,
-			   RealD _c1, RealD _c2,RealD _u0,
-			   const ImplParams &p = ImplParams());
-
-  //////////////////////////////////////////////////////////////////////////
-  // MILC constructor no gauge fields
-  //////////////////////////////////////////////////////////////////////////
-  ImprovedStaggeredFermion(GridCartesian &Fgrid, GridRedBlackCartesian &Hgrid, RealD _mass,
-			   RealD _c1=1.0, RealD _c2=1.0,RealD _u0=1.0,
-			   const ImplParams &p = ImplParams());
-
-  // DoubleStore impl dependent
-  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);
-  DoubledGaugeField &GetU(void)   { return Umu ; } ;
-  DoubledGaugeField &GetUUU(void) { return UUUmu; };
-  void CopyGaugeCheckerboards(void);
-
-  ///////////////////////////////////////////////////////////////
-  // Data members require to support the functionality
-  ///////////////////////////////////////////////////////////////
-
-  //    protected:
- public:
-  // any other parameters of action ???
-  virtual int   isTrivialEE(void) { return 1; };
-  virtual RealD Mass(void) { return mass; }
-  RealD mass;
-  RealD u0;
-  RealD c1;
-  RealD c2;
-
-  GridBase *_grid;
-  GridBase *_cbgrid;
-
-  // 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;
-  
-  ///////////////////////////////////////////////////////////////
-  // 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);
-};
-
-typedef ImprovedStaggeredFermion<StaggeredImplF> ImprovedStaggeredFermionF;
-typedef ImprovedStaggeredFermion<StaggeredImplD> ImprovedStaggeredFermionD;
-
-}
-}
-#endif

Grid/qcd/action/fermion/ImprovedStaggeredFermion5D.cc

@@ -1,654 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/ImprovedStaggeredFermion5D.cc
-
-    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 */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/ImprovedStaggeredFermion5D.h>
-#include <Grid/perfmon/PerfCount.h>
-
-namespace Grid {
-namespace QCD {
-  
-// S-direction is INNERMOST and takes no part in the parity.
-const std::vector<int> 
-ImprovedStaggeredFermion5DStatic::directions({1,2,3,4,1,2,3,4,1,2,3,4,1,2,3,4});
-const std::vector<int> 
-ImprovedStaggeredFermion5DStatic::displacements({1, 1, 1, 1, -1, -1, -1, -1, 3, 3, 3, 3, -3, -3, -3, -3});
-
-  // 5d lattice for DWF.
-template<class Impl>
-ImprovedStaggeredFermion5D<Impl>::ImprovedStaggeredFermion5D(GridCartesian         &FiveDimGrid,
-							     GridRedBlackCartesian &FiveDimRedBlackGrid,
-							     GridCartesian         &FourDimGrid,
-							     GridRedBlackCartesian &FourDimRedBlackGrid,
-							     RealD _mass,
-							     RealD _c1,RealD _c2, RealD _u0,
-							     const ImplParams &p) :
-  Kernels(p),
-  _FiveDimGrid        (&FiveDimGrid),
-  _FiveDimRedBlackGrid(&FiveDimRedBlackGrid),
-  _FourDimGrid        (&FourDimGrid),
-  _FourDimRedBlackGrid(&FourDimRedBlackGrid),
-  Stencil    (&FiveDimGrid,npoint,Even,directions,displacements),
-  StencilEven(&FiveDimRedBlackGrid,npoint,Even,directions,displacements), // source is Even
-  StencilOdd (&FiveDimRedBlackGrid,npoint,Odd ,directions,displacements), // source is Odd
-  mass(_mass),
-  c1(_c1),
-  c2(_c2),
-  u0(_u0),
-  Umu(&FourDimGrid),
-  UmuEven(&FourDimRedBlackGrid),
-  UmuOdd (&FourDimRedBlackGrid),
-  UUUmu(&FourDimGrid),
-  UUUmuEven(&FourDimRedBlackGrid),
-  UUUmuOdd(&FourDimRedBlackGrid),
-  Lebesgue(&FourDimGrid),
-  LebesgueEvenOdd(&FourDimRedBlackGrid),
-  _tmp(&FiveDimRedBlackGrid)
-{
-
-  // some assertions
-  assert(FiveDimGrid._ndimension==5);
-  assert(FourDimGrid._ndimension==4);
-  assert(FourDimRedBlackGrid._ndimension==4);
-  assert(FiveDimRedBlackGrid._ndimension==5);
-  assert(FiveDimRedBlackGrid._checker_dim==1); // Don't checker the s direction
-
-  // extent of fifth dim and not spread out
-  Ls=FiveDimGrid._fdimensions[0];
-  assert(FiveDimRedBlackGrid._fdimensions[0]==Ls);
-  assert(FiveDimGrid._processors[0]         ==1);
-  assert(FiveDimRedBlackGrid._processors[0] ==1);
-
-  // Other dimensions must match the decomposition of the four-D fields 
-  for(int d=0;d<4;d++){
-    assert(FiveDimGrid._processors[d+1]         ==FourDimGrid._processors[d]);
-    assert(FiveDimRedBlackGrid._processors[d+1] ==FourDimGrid._processors[d]);
-    assert(FourDimRedBlackGrid._processors[d]   ==FourDimGrid._processors[d]);
-
-    assert(FiveDimGrid._fdimensions[d+1]        ==FourDimGrid._fdimensions[d]);
-    assert(FiveDimRedBlackGrid._fdimensions[d+1]==FourDimGrid._fdimensions[d]);
-    assert(FourDimRedBlackGrid._fdimensions[d]  ==FourDimGrid._fdimensions[d]);
-
-    assert(FiveDimGrid._simd_layout[d+1]        ==FourDimGrid._simd_layout[d]);
-    assert(FiveDimRedBlackGrid._simd_layout[d+1]==FourDimGrid._simd_layout[d]);
-    assert(FourDimRedBlackGrid._simd_layout[d]  ==FourDimGrid._simd_layout[d]);
-  }
-
-  if (Impl::LsVectorised) { 
-
-    int nsimd = Simd::Nsimd();
-    
-    // Dimension zero of the five-d is the Ls direction
-    assert(FiveDimGrid._simd_layout[0]        ==nsimd);
-    assert(FiveDimRedBlackGrid._simd_layout[0]==nsimd);
-
-    for(int d=0;d<4;d++){
-      assert(FourDimGrid._simd_layout[d]=1);
-      assert(FourDimRedBlackGrid._simd_layout[d]=1);
-      assert(FiveDimRedBlackGrid._simd_layout[d+1]==1);
-    }
-
-  } else {
-    
-    // Dimension zero of the five-d is the Ls direction
-    assert(FiveDimRedBlackGrid._simd_layout[0]==1);
-    assert(FiveDimGrid._simd_layout[0]        ==1);
-
-  }
-  int LLs = FiveDimGrid._rdimensions[0];
-  int vol4= FourDimGrid.oSites();
-  Stencil.BuildSurfaceList(LLs,vol4);
-
-  vol4=FourDimRedBlackGrid.oSites();
-  StencilEven.BuildSurfaceList(LLs,vol4);
-  StencilOdd.BuildSurfaceList(LLs,vol4);
-}
-template <class Impl>
-void ImprovedStaggeredFermion5D<Impl>::CopyGaugeCheckerboards(void)
-{
-  pickCheckerboard(Even, UmuEven,  Umu);
-  pickCheckerboard(Odd,  UmuOdd ,  Umu);
-  pickCheckerboard(Even, UUUmuEven,UUUmu);
-  pickCheckerboard(Odd,  UUUmuOdd, UUUmu);
-}
-template<class Impl>
-ImprovedStaggeredFermion5D<Impl>::ImprovedStaggeredFermion5D(GaugeField &_Uthin,GaugeField &_Ufat,
-							     GridCartesian         &FiveDimGrid,
-							     GridRedBlackCartesian &FiveDimRedBlackGrid,
-							     GridCartesian         &FourDimGrid,
-							     GridRedBlackCartesian &FourDimRedBlackGrid,
-							     RealD _mass,
-							     RealD _c1,RealD _c2, RealD _u0,
-							     const ImplParams &p) :
-  ImprovedStaggeredFermion5D(FiveDimGrid,FiveDimRedBlackGrid,
-			     FourDimGrid,FourDimRedBlackGrid,
-			     _mass,_c1,_c2,_u0,p)
-{
-  ImportGauge(_Uthin,_Ufat);
-}
-
-///////////////////////////////////////////////////
-// For MILC use; pass three link U's and 1 link U
-///////////////////////////////////////////////////
-template <class Impl>
-void ImprovedStaggeredFermion5D<Impl>::ImportGaugeSimple(const GaugeField &_Utriple,const GaugeField &_Ufat) 
-{
-  /////////////////////////////////////////////////////////////////
-  // Trivial import; phases and fattening and such like preapplied
-  /////////////////////////////////////////////////////////////////
-  for (int mu = 0; mu < Nd; mu++) {
-
-    auto U = PeekIndex<LorentzIndex>(_Utriple, mu);
-    Impl::InsertGaugeField(UUUmu,U,mu);
-
-    U = adj( Cshift(U, mu, -3));
-    Impl::InsertGaugeField(UUUmu,-U,mu+4);
-
-    U = PeekIndex<LorentzIndex>(_Ufat, mu);
-    Impl::InsertGaugeField(Umu,U,mu);
-
-    U = adj( Cshift(U, mu, -1));
-    Impl::InsertGaugeField(Umu,-U,mu+4);
-
-  }
-  CopyGaugeCheckerboards();
-}
-template <class Impl>
-void ImprovedStaggeredFermion5D<Impl>::ImportGaugeSimple(const DoubledGaugeField &_UUU,const DoubledGaugeField &_U) 
-{
-  /////////////////////////////////////////////////////////////////
-  // Trivial import; phases and fattening and such like preapplied
-  /////////////////////////////////////////////////////////////////
-  Umu   = _U;
-  UUUmu = _UUU;
-  CopyGaugeCheckerboards();
-}
-template<class Impl>
-void ImprovedStaggeredFermion5D<Impl>::ImportGauge(const GaugeField &_Uthin,const GaugeField &_Ufat)
-{
-  ////////////////////////////////////////////////////////
-  // 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++) {
-
-    auto 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();
-}
-template<class Impl>
-void ImprovedStaggeredFermion5D<Impl>::DhopDir(const FermionField &in, FermionField &out,int dir5,int disp)
-{
-  int dir = dir5-1; // Maps to the ordering above in "directions" that is passed to stencil
-                    // we drop off the innermost fifth dimension
-
-  Compressor compressor;
-  Stencil.HaloExchange(in,compressor);
-
-  parallel_for(int ss=0;ss<Umu._grid->oSites();ss++){
-    for(int s=0;s<Ls;s++){
-      int sU=ss;
-      int sF = s+Ls*sU; 
-      Kernels::DhopDir(Stencil, Umu, UUUmu, Stencil.CommBuf(), sF, sU, in, out, dir, disp);
-    }
-  }
-};
-
-template<class Impl>
-void ImprovedStaggeredFermion5D<Impl>::DerivInternal(StencilImpl & st,
-            DoubledGaugeField & U,
-            DoubledGaugeField & UUU,
-            GaugeField &mat,
-            const FermionField &A,
-            const FermionField &B,
-            int dag)
-{
-  // No force terms in multi-rhs solver staggered
-  assert(0);
-}
-
-template<class Impl>
-void ImprovedStaggeredFermion5D<Impl>::DhopDeriv(GaugeField &mat,
-				      const FermionField &A,
-				      const FermionField &B,
-				      int dag)
-{
-  assert(0);
-}
-
-template<class Impl>
-void ImprovedStaggeredFermion5D<Impl>::DhopDerivEO(GaugeField &mat,
-					const FermionField &A,
-					const FermionField &B,
-					int dag)
-{
-  assert(0);
-}
-
-
-template<class Impl>
-void ImprovedStaggeredFermion5D<Impl>::DhopDerivOE(GaugeField &mat,
-					const FermionField &A,
-					const FermionField &B,
-					int dag)
-{
-  assert(0);
-}
-
-/*CHANGE */
-template<class Impl>
-void ImprovedStaggeredFermion5D<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 ImprovedStaggeredFermion5D<Impl>::DhopInternalOverlappedComms(StencilImpl & st, LebesgueOrder &lo,
-								   DoubledGaugeField & U,DoubledGaugeField & UUU,
-								   const FermionField &in, FermionField &out,int dag)
-{
-#ifdef GRID_OMP
-  //  assert((dag==DaggerNo) ||(dag==DaggerYes));
-
-  Compressor compressor; 
-
-  int LLs = in._grid->_rdimensions[0];
-  int len =  U._grid->oSites();
-
-  DhopFaceTime-=usecond();
-  st.Prepare();
-  st.HaloGather(in,compressor);
-  //  st.HaloExchangeOptGather(in,compressor); // Wilson compressor
-  st.CommsMergeSHM(compressor);// Could do this inside parallel region overlapped with comms
-  DhopFaceTime+=usecond();
-
-  double ctime=0;
-  double ptime=0;
-
-  //////////////////////////////////////////////////////////////////////////////////////////////////////
-  // Ugly explicit thread mapping introduced for OPA reasons.
-  //////////////////////////////////////////////////////////////////////////////////////////////////////
-#pragma omp parallel reduction(max:ctime) reduction(max:ptime)
-  {
-    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) {
-      double start = usecond();
-      nthreads -= ncomms;
-      int ttid  = tid - ncomms;
-      int n     = U._grid->oSites(); // 4d vol
-      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
-      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,UUU,st.CommBuf(),LLs,sU,in,out,1,0); //<---------
-        }
-      } else {
-        for (int ss = myblock; ss < myblock+myn; ++ss) {
-	  // Interior = 1; Exterior = 0;
-          int sU = ss;
-          Kernels::DhopSite(st,lo,U,UUU,st.CommBuf(),LLs,sU,in,out,1,0); //<------------
-        }
-      }
-        ptime = usecond() - start;
-    } else {
-      double start = usecond();
-      st.CommunicateThreaded();
-      ctime = usecond() - start;
-    }
-  }
-  DhopCommTime += ctime;
-  DhopComputeTime+=ptime;
-
-  // First to enter, last to leave timing
-  st.CollateThreads();
-
-  DhopFaceTime-=usecond();
-  st.CommsMerge(compressor);
-  DhopFaceTime+=usecond();
-
-  DhopComputeTime2-=usecond();
-  if (dag == DaggerYes) {
-    int sz=st.surface_list.size();
-    parallel_for (int ss = 0; ss < sz; ss++) {
-      int sU = st.surface_list[ss];
-      Kernels::DhopSiteDag(st,lo,U,UUU,st.CommBuf(),LLs,sU,in,out,0,1); //<----------
-    }
-  } else {
-    int sz=st.surface_list.size();
-    parallel_for (int ss = 0; ss < sz; ss++) {
-      int sU = st.surface_list[ss];
-      Kernels::DhopSite(st,lo,U,UUU,st.CommBuf(),LLs,sU,in,out,0,1);//<----------
-    }
-  }
-  DhopComputeTime2+=usecond();
-#else
-  assert(0);
-#endif
-
-}
-
-template<class Impl>
-void ImprovedStaggeredFermion5D<Impl>::DhopInternalSerialComms(StencilImpl & st, LebesgueOrder &lo,
-						    DoubledGaugeField & U,DoubledGaugeField & UUU,
-						    const FermionField &in, FermionField &out,int dag)
-{
-  Compressor compressor;
-  int LLs = in._grid->_rdimensions[0];
-
-
-
- //double t1=usecond();
-  DhopTotalTime -= usecond();
-  DhopCommTime -= usecond();
-  st.HaloExchange(in,compressor);
-  DhopCommTime += usecond();
-  
-  DhopComputeTime -= usecond();
-  // Dhop takes the 4d grid from U, and makes a 5d index for fermion
-  if (dag == DaggerYes) {
-    parallel_for (int ss = 0; ss < U._grid->oSites(); ss++) {
-      int sU=ss;
-      Kernels::DhopSiteDag(st, lo, U, UUU, st.CommBuf(), LLs, sU,in, out);
-    }
-  } else {
-    parallel_for (int ss = 0; ss < U._grid->oSites(); ss++) {
-      int sU=ss;
-      Kernels::DhopSite(st,lo,U,UUU,st.CommBuf(),LLs,sU,in,out);
-    }
-  }
-  DhopComputeTime += usecond();
-  DhopTotalTime   += usecond();
- //double t2=usecond();
- //std::cout << __FILE__ << " " << __func__  << " Total Time " << DhopTotalTime << std::endl;
- //std::cout << __FILE__ << " " << __func__  << " Total Time Org " << t2-t1 << std::endl;
- //std::cout << __FILE__ << " " << __func__  << " Comml Time " << DhopCommTime << std::endl;
- //std::cout << __FILE__ << " " << __func__  << " Compute Time " << DhopComputeTime << std::endl;
-
-}
-/*CHANGE END*/
-
-/* ORG
-template<class Impl>
-void ImprovedStaggeredFermion5D<Impl>::DhopInternal(StencilImpl & st, LebesgueOrder &lo,
-						    DoubledGaugeField & U,DoubledGaugeField & UUU,
-						    const FermionField &in, FermionField &out,int dag)
-{
-  Compressor compressor;
-  int LLs = in._grid->_rdimensions[0];
-
-
-
-  DhopTotalTime -= usecond();
-  DhopCommTime -= usecond();
-  st.HaloExchange(in,compressor);
-  DhopCommTime += usecond();
-  
-  DhopComputeTime -= usecond();
-  // Dhop takes the 4d grid from U, and makes a 5d index for fermion
-  if (dag == DaggerYes) {
-    parallel_for (int ss = 0; ss < U._grid->oSites(); ss++) {
-      int sU=ss;
-      Kernels::DhopSiteDag(st, lo, U, UUU, st.CommBuf(), LLs, sU,in, out);
-    }
-  } else {
-    parallel_for (int ss = 0; ss < U._grid->oSites(); ss++) {
-      int sU=ss;
-	Kernels::DhopSite(st,lo,U,UUU,st.CommBuf(),LLs,sU,in,out);
-    }
-  }
-  DhopComputeTime += usecond();
-  DhopTotalTime   += usecond();
-}
-*/
-
-
-template<class Impl>
-void ImprovedStaggeredFermion5D<Impl>::DhopOE(const FermionField &in, FermionField &out,int dag)
-{
-  DhopCalls+=1;
-  conformable(in._grid,FermionRedBlackGrid());    // 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 ImprovedStaggeredFermion5D<Impl>::DhopEO(const FermionField &in, FermionField &out,int dag)
-{
-  DhopCalls+=1;
-  conformable(in._grid,FermionRedBlackGrid());    // 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 ImprovedStaggeredFermion5D<Impl>::Dhop(const FermionField &in, FermionField &out,int dag)
-{
-  DhopCalls+=2;
-  conformable(in._grid,FermionGrid()); // verifies full grid
-  conformable(in._grid,out._grid);
-
-  out.checkerboard = in.checkerboard;
-
-  DhopInternal(Stencil,Lebesgue,Umu,UUUmu,in,out,dag);
-}
-
-template<class Impl>
-void ImprovedStaggeredFermion5D<Impl>::Report(void) 
-{
-  std::vector<int> latt = GridDefaultLatt();          
-  RealD volume = Ls;  for(int mu=0;mu<Nd;mu++) volume=volume*latt[mu];
-  RealD NP = _FourDimGrid->_Nprocessors;
-  RealD NN = _FourDimGrid->NodeCount();
-
-  std::cout << GridLogMessage << "#### Dhop calls report " << std::endl;
-
-  std::cout << GridLogMessage << "ImprovedStaggeredFermion5D Number of DhopEO Calls   : " 
-	    << DhopCalls   << std::endl;
-  std::cout << GridLogMessage << "ImprovedStaggeredFermion5D TotalTime   /Calls       : " 
-	    << DhopTotalTime   / DhopCalls << " us" << std::endl;
-  std::cout << GridLogMessage << "ImprovedStaggeredFermion5D CommTime    /Calls       : " 
-	    << DhopCommTime    / DhopCalls << " us" << std::endl;
-  std::cout << GridLogMessage << "ImprovedStaggeredFermion5D ComputeTime/Calls        : " 
-	    << DhopComputeTime / DhopCalls << " us" << std::endl;
-
-  // Average the compute time
-  _FourDimGrid->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 << "ImprovedStaggeredFermion5D Stencil"    <<std::endl;  Stencil.Report();
-  std::cout << GridLogMessage << "ImprovedStaggeredFermion5D StencilEven"<<std::endl;  StencilEven.Report();
-  std::cout << GridLogMessage << "ImprovedStaggeredFermion5D StencilOdd" <<std::endl;  StencilOdd.Report();
-}
-template<class Impl>
-void ImprovedStaggeredFermion5D<Impl>::ZeroCounters(void) 
-{
-  DhopCalls       = 0;
-  DhopTotalTime    = 0;
-  DhopCommTime    = 0;
-  DhopComputeTime = 0;
-  DhopFaceTime    = 0;
-
-
-  Stencil.ZeroCounters();
-  StencilEven.ZeroCounters();
-  StencilOdd.ZeroCounters();
-}
-
-/////////////////////////////////////////////////////////////////////////
-// Implement the general interface. Here we use SAME mass on all slices
-/////////////////////////////////////////////////////////////////////////
-template <class Impl>
-void ImprovedStaggeredFermion5D<Impl>::Mdir(const FermionField &in, FermionField &out, int dir, int disp) {
-  DhopDir(in, out, dir, disp);
-}
-template <class Impl>
-RealD ImprovedStaggeredFermion5D<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 ImprovedStaggeredFermion5D<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 ImprovedStaggeredFermion5D<Impl>::Meooe(const FermionField &in, FermionField &out) {
-  if (in.checkerboard == Odd) {
-    DhopEO(in, out, DaggerNo);
-  } else {
-    DhopOE(in, out, DaggerNo);
-  }
-}
-template <class Impl>
-void ImprovedStaggeredFermion5D<Impl>::MeooeDag(const FermionField &in, FermionField &out) {
-  if (in.checkerboard == Odd) {
-    DhopEO(in, out, DaggerYes);
-  } else {
-    DhopOE(in, out, DaggerYes);
-  }
-}
-
-template <class Impl>
-void ImprovedStaggeredFermion5D<Impl>::Mooee(const FermionField &in, FermionField &out) {
-  out.checkerboard = in.checkerboard;
-  typename FermionField::scalar_type scal(mass);
-  out = scal * in;
-}
-
-template <class Impl>
-void ImprovedStaggeredFermion5D<Impl>::MooeeDag(const FermionField &in, FermionField &out) {
-  out.checkerboard = in.checkerboard;
-  Mooee(in, out);
-}
-
-template <class Impl>
-void ImprovedStaggeredFermion5D<Impl>::MooeeInv(const FermionField &in, FermionField &out) {
-  out.checkerboard = in.checkerboard;
-  out = (1.0 / (mass)) * in;
-}
-
-template <class Impl>
-void ImprovedStaggeredFermion5D<Impl>::MooeeInvDag(const FermionField &in,
-                                      FermionField &out) {
-  out.checkerboard = in.checkerboard;
-  MooeeInv(in, out);
-}
-
-//////////////////////////////////////////////////////// 
-// Conserved current - not yet implemented.
-////////////////////////////////////////////////////////
-template <class Impl>
-void ImprovedStaggeredFermion5D<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 ImprovedStaggeredFermion5D<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);
-
-}
-
-FermOpStaggeredTemplateInstantiate(ImprovedStaggeredFermion5D);
-FermOpStaggeredVec5dTemplateInstantiate(ImprovedStaggeredFermion5D);
-  
-}}
-
-
-

Grid/qcd/action/fermion/ImprovedStaggeredFermion5D.h

@@ -1,234 +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 */
-#ifndef  GRID_QCD_IMPROVED_STAGGERED_FERMION_5D_H
-#define  GRID_QCD_IMPROVED_STAGGERED_FERMION_5D_H
-
-namespace Grid {
-namespace QCD {
-
-  ////////////////////////////////////////////////////////////////////////////////
-  // 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);
-  };
-
-}}
-
-#endif

Grid/qcd/action/fermion/MADWF.h

@@ -1,193 +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 Grid {
-namespace QCD {
-
-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);
-
-  }
-
-};
-
-}}

Grid/qcd/action/fermion/MobiusEOFAFermion.cc

@@ -1,502 +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 */
-
-#include <Grid/Grid_Eigen_Dense.h>
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/MobiusEOFAFermion.h>
-
-namespace Grid {
-namespace QCD {
-
-  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)
-    {
-      int Ls = this->Ls;
-
-      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)
-    {
-      int Ls = this->Ls;
-
-      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;
-
-      std::vector<Coeff_t> diag(Ls,1.0);
-      std::vector<Coeff_t> upper(Ls,-1.0);  upper[Ls-1] = this->mq1;
-      std::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;
-
-      std::vector<Coeff_t> diag(Ls,1.0);
-      std::vector<Coeff_t> upper(Ls,-1.0);  upper[Ls-1] = this->mq1;
-      std::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
-      std::vector<Coeff_t> diag = this->bee;
-      std::vector<Coeff_t> upper(Ls);
-      std::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
-      std::vector<Coeff_t> diag = this->bee;
-      std::vector<Coeff_t> upper(Ls);
-      std::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);
-        std::vector<Coeff_t> d = Mooee_shift;
-        std::vector<Coeff_t> u(Ls,0.0);
-        std::vector<Coeff_t> y(Ls,0.0);
-        std::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] = std::pow(this->bee[s],s) * std::pow(this->cee[s],Ls-1-s); }
-          else{ MooeeInv_shift_lc[s] = std::pow(this->bee[s],Ls-1-s) * std::pow(this->cee[s],s); }
-
-          // MooeeInv_shift_norm
-          MooeeInv_shift_norm[s] = -MooeeInvDag_shift_lc[s] /
-            ( std::pow(this->bee[s],Ls) + mq1*std::pow(this->cee[s],Ls) ) / N;
-
-          // MooeeInvDag_shift_norm
-          if(pm == 1){ MooeeInvDag_shift_norm[s] = -std::pow(this->bee[s],s) * std::pow(this->cee[s],Ls-1-s) /
-            ( std::pow(this->bee[s],Ls) + mq1*std::pow(this->cee[s],Ls) ) / N; }
-          else{ MooeeInvDag_shift_norm[s] = -std::pow(this->bee[s],Ls-1-s) * std::pow(this->cee[s],s) /
-            ( std::pow(this->bee[s],Ls) + mq1*std::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);
-      }
-    }
-
-    template<class Impl>
-    void MobiusEOFAFermion<Impl>::MooeeInternalCompute(int dag, int inv,
-      Vector<iSinglet<Simd> >& Matp, Vector<iSinglet<Simd> >& Matm)
-    {
-      int Ls = this->Ls;
-
-      GridBase* grid = this->FermionRedBlackGrid();
-      int LLs = grid->_rdimensions[0];
-
-      if(LLs == Ls){ return; } // Not vectorised in 5th direction
-
-      Eigen::MatrixXcd Pplus  = Eigen::MatrixXcd::Zero(Ls,Ls);
-      Eigen::MatrixXcd Pminus = Eigen::MatrixXcd::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];
-        Pplus(s+1,s) = -this->cee[s+1];
-      }
-
-      Pplus (0,Ls-1) = this->mq1*this->cee[0];
-      Pminus(Ls-1,0) = this->mq1*this->cee[Ls-1];
-
-      if(this->shift != 0.0){
-        RealD c = 0.5 * this->alpha;
-        RealD d = 0.5;
-        RealD N = this->shift * this->k * ( std::pow(c+d,Ls) + this->mq1*std::pow(c-d,Ls) );
-        if(this->pm == 1) {
-          for(int s=0; s<Ls; ++s){
-            Pplus(s,Ls-1) += N * std::pow(-1.0,s) * std::pow(c-d,s) / std::pow(c+d,Ls+s+1);
-          }
-        } else {
-          for(int s=0; s<Ls; ++s){
-            Pminus(s,0) += N * std::pow(-1.0,s+1) * std::pow(c-d,Ls-1-s) / std::pow(c+d,2*Ls-s);
-          }
-        }
-      }
-
-      Eigen::MatrixXcd PplusMat ;
-      Eigen::MatrixXcd PminusMat;
-
-      if(inv) {
-        PplusMat  = Pplus.inverse();
-        PminusMat = Pminus.inverse();
-      } else {
-        PplusMat  = Pplus;
-        PminusMat = Pminus;
-      }
-
-      if(dag){
-        PplusMat.adjointInPlace();
-        PminusMat.adjointInPlace();
-      }
-
-      typedef typename SiteHalfSpinor::scalar_type scalar_type;
-      const int Nsimd = Simd::Nsimd();
-      Matp.resize(Ls*LLs);
-      Matm.resize(Ls*LLs);
-
-      for(int s2=0; s2<Ls; s2++){
-      for(int s1=0; s1<LLs; s1++){
-        int istride = LLs;
-        int ostride = 1;
-        Simd Vp;
-        Simd Vm;
-        scalar_type *sp = (scalar_type*) &Vp;
-        scalar_type *sm = (scalar_type*) &Vm;
-        for(int l=0; l<Nsimd; l++){
-          if(switcheroo<Coeff_t>::iscomplex()) {
-            sp[l] = PplusMat (l*istride+s1*ostride,s2);
-            sm[l] = PminusMat(l*istride+s1*ostride,s2);
-          } else {
-            // if real
-            scalar_type tmp;
-            tmp = PplusMat (l*istride+s1*ostride,s2);
-            sp[l] = scalar_type(tmp.real(),tmp.real());
-            tmp = PminusMat(l*istride+s1*ostride,s2);
-            sm[l] = scalar_type(tmp.real(),tmp.real());
-          }
-        }
-        Matp[LLs*s2+s1] = Vp;
-        Matm[LLs*s2+s1] = Vm;
-      }}
-  }
-
-  FermOpTemplateInstantiate(MobiusEOFAFermion);
-  GparityFermOpTemplateInstantiate(MobiusEOFAFermion);
-
-}}

Grid/qcd/action/fermion/MobiusEOFAFermion.h

@@ -1,133 +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 Grid {
-namespace QCD {
-
-  template<class Impl>
-  class MobiusEOFAFermion : public AbstractEOFAFermion<Impl>
-  {
-    public:
-      INHERIT_IMPL_TYPES(Impl);
-
-    public:
-      // Shift operator coefficients for red-black preconditioned Mobius EOFA
-      std::vector<Coeff_t> Mooee_shift;
-      std::vector<Coeff_t> MooeeInv_shift_lc;
-      std::vector<Coeff_t> MooeeInv_shift_norm;
-      std::vector<Coeff_t> MooeeInvDag_shift_lc;
-      std::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,
-        std::vector<Coeff_t>& lower, std::vector<Coeff_t>& diag, std::vector<Coeff_t>& upper);
-
-      void M5D_shift(const FermionField& psi, const FermionField& phi, FermionField& chi,
-        std::vector<Coeff_t>& lower, std::vector<Coeff_t>& diag, std::vector<Coeff_t>& upper,
-        std::vector<Coeff_t>& shift_coeffs);
-
-      void M5Ddag(const FermionField& psi, const FermionField& phi, FermionField& chi,
-        std::vector<Coeff_t>& lower, std::vector<Coeff_t>& diag, std::vector<Coeff_t>& upper);
-
-      void M5Ddag_shift(const FermionField& psi, const FermionField& phi, FermionField& chi,
-        std::vector<Coeff_t>& lower, std::vector<Coeff_t>& diag, std::vector<Coeff_t>& upper,
-        std::vector<Coeff_t>& shift_coeffs);
-
-      void MooeeInternal(const FermionField& in, FermionField& out, int dag, int inv);
-
-      void MooeeInternalCompute(int dag, int inv, Vector<iSinglet<Simd>>& Matp, Vector<iSinglet<Simd>>& Matm);
-
-      void MooeeInternalAsm(const FermionField& in, FermionField& out, int LLs, int site,
-        Vector<iSinglet<Simd>>& Matp, Vector<iSinglet<Simd>>& Matm);
-
-      void MooeeInternalZAsm(const FermionField& in, FermionField& out, int LLs, int site,
-        Vector<iSinglet<Simd>>& Matp, Vector<iSinglet<Simd>>& Matm);
-
-      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);
-  };
-}}
-
-#define INSTANTIATE_DPERP_MOBIUS_EOFA(A)\
-template void MobiusEOFAFermion<A>::M5D(const FermionField& psi, const FermionField& phi, FermionField& chi, \
-  std::vector<Coeff_t>& lower, std::vector<Coeff_t>& diag, std::vector<Coeff_t>& upper); \
-template void MobiusEOFAFermion<A>::M5D_shift(const FermionField& psi, const FermionField& phi, FermionField& chi, \
-  std::vector<Coeff_t>& lower, std::vector<Coeff_t>& diag, std::vector<Coeff_t>& upper, std::vector<Coeff_t>& shift_coeffs); \
-template void MobiusEOFAFermion<A>::M5Ddag(const FermionField& psi, const FermionField& phi, FermionField& chi, \
-  std::vector<Coeff_t>& lower, std::vector<Coeff_t>& diag, std::vector<Coeff_t>& upper); \
-template void MobiusEOFAFermion<A>::M5Ddag_shift(const FermionField& psi, const FermionField& phi, FermionField& chi, \
-  std::vector<Coeff_t>& lower, std::vector<Coeff_t>& diag, std::vector<Coeff_t>& upper, std::vector<Coeff_t>& shift_coeffs); \
-template void MobiusEOFAFermion<A>::MooeeInv(const FermionField& psi, FermionField& chi); \
-template void MobiusEOFAFermion<A>::MooeeInv_shift(const FermionField& psi, FermionField& chi); \
-template void MobiusEOFAFermion<A>::MooeeInvDag(const FermionField& psi, FermionField& chi); \
-template void MobiusEOFAFermion<A>::MooeeInvDag_shift(const FermionField& psi, FermionField& chi);
-
-#undef  MOBIUS_EOFA_DPERP_DENSE
-#define MOBIUS_EOFA_DPERP_CACHE
-#undef  MOBIUS_EOFA_DPERP_LINALG
-#define MOBIUS_EOFA_DPERP_VEC
-
-#endif

Grid/qcd/action/fermion/MobiusEOFAFermioncache.cc

@@ -1,429 +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 Grid {
-namespace QCD {
-
-  // FIXME -- make a version of these routines with site loop outermost for cache reuse.
-
-  template<class Impl>
-  void MobiusEOFAFermion<Impl>::M5D(const FermionField &psi, const FermionField &phi, FermionField &chi,
-    std::vector<Coeff_t> &lower, std::vector<Coeff_t> &diag, std::vector<Coeff_t> &upper)
-  {
-    int Ls = this->Ls;
-    GridBase *grid = psi._grid;
-
-    assert(phi.checkerboard == psi.checkerboard);
-    chi.checkerboard = psi.checkerboard;
-
-    // Flops = 6.0*(Nc*Ns) *Ls*vol
-    this->M5Dcalls++;
-    this->M5Dtime -= usecond();
-
-    parallel_for(int ss=0; ss<grid->oSites(); ss+=Ls){
-      for(int s=0; s<Ls; s++){
-        auto tmp = psi._odata[0];
-        if(s==0){
-          spProj5m(tmp, psi._odata[ss+s+1]);
-          chi[ss+s] = diag[s]*phi[ss+s] + upper[s]*tmp;
-          spProj5p(tmp, psi._odata[ss+Ls-1]);
-          chi[ss+s] = chi[ss+s] + lower[s]*tmp;
-        } else if(s==(Ls-1)) {
-          spProj5m(tmp, psi._odata[ss+0]);
-          chi[ss+s] = diag[s]*phi[ss+s] + upper[s]*tmp;
-          spProj5p(tmp, psi._odata[ss+s-1]);
-          chi[ss+s] = chi[ss+s] + lower[s]*tmp;
-        } else {
-          spProj5m(tmp, psi._odata[ss+s+1]);
-          chi[ss+s] = diag[s]*phi[ss+s] + upper[s]*tmp;
-          spProj5p(tmp, psi._odata[ss+s-1]);
-          chi[ss+s] = chi[ss+s] + lower[s]*tmp;
-        }
-      }
-    }
-
-    this->M5Dtime += usecond();
-  }
-
-  template<class Impl>
-  void MobiusEOFAFermion<Impl>::M5D_shift(const FermionField &psi, const FermionField &phi, FermionField &chi,
-    std::vector<Coeff_t> &lower, std::vector<Coeff_t> &diag, std::vector<Coeff_t> &upper,
-    std::vector<Coeff_t> &shift_coeffs)
-  {
-    int Ls = this->Ls;
-    int shift_s = (this->pm == 1) ? (Ls-1) : 0; // s-component modified by shift operator
-    GridBase *grid = psi._grid;
-
-    assert(phi.checkerboard == psi.checkerboard);
-    chi.checkerboard = psi.checkerboard;
-
-    // Flops = 6.0*(Nc*Ns) *Ls*vol
-    this->M5Dcalls++;
-    this->M5Dtime -= usecond();
-
-    parallel_for(int ss=0; ss<grid->oSites(); ss+=Ls){
-      for(int s=0; s<Ls; s++){
-        auto tmp = psi._odata[0];
-        if(s==0){
-          spProj5m(tmp, psi._odata[ss+s+1]);
-          chi[ss+s] = diag[s]*phi[ss+s] + upper[s]*tmp;
-          spProj5p(tmp, psi._odata[ss+Ls-1]);
-          chi[ss+s] = chi[ss+s] + lower[s]*tmp;
-        } else if(s==(Ls-1)) {
-          spProj5m(tmp, psi._odata[ss+0]);
-          chi[ss+s] = diag[s]*phi[ss+s] + upper[s]*tmp;
-          spProj5p(tmp, psi._odata[ss+s-1]);
-          chi[ss+s] = chi[ss+s] + lower[s]*tmp;
-        } else {
-          spProj5m(tmp, psi._odata[ss+s+1]);
-          chi[ss+s] = diag[s]*phi[ss+s] + upper[s]*tmp;
-          spProj5p(tmp, psi._odata[ss+s-1]);
-          chi[ss+s] = chi[ss+s] + lower[s]*tmp;
-        }
-        if(this->pm == 1){ spProj5p(tmp, psi._odata[ss+shift_s]); }
-        else{ spProj5m(tmp, psi._odata[ss+shift_s]); }
-        chi[ss+s] = chi[ss+s] + shift_coeffs[s]*tmp;
-      }
-    }
-
-    this->M5Dtime += usecond();
-  }
-
-  template<class Impl>
-  void MobiusEOFAFermion<Impl>::M5Ddag(const FermionField &psi, const FermionField &phi, FermionField &chi,
-    std::vector<Coeff_t> &lower, std::vector<Coeff_t> &diag, std::vector<Coeff_t> &upper)
-  {
-    int Ls = this->Ls;
-    GridBase *grid = psi._grid;
-
-    assert(phi.checkerboard == psi.checkerboard);
-    chi.checkerboard = psi.checkerboard;
-
-    // Flops = 6.0*(Nc*Ns) *Ls*vol
-    this->M5Dcalls++;
-    this->M5Dtime -= usecond();
-
-    parallel_for(int ss=0; ss<grid->oSites(); ss+=Ls){
-      auto tmp = psi._odata[0];
-      for(int s=0; s<Ls; s++){
-        if(s==0) {
-          spProj5p(tmp, psi._odata[ss+s+1]);
-          chi[ss+s] = diag[s]*phi[ss+s] + upper[s]*tmp;
-          spProj5m(tmp, psi._odata[ss+Ls-1]);
-          chi[ss+s] = chi[ss+s] + lower[s]*tmp;
-        } else if(s==(Ls-1)) {
-          spProj5p(tmp, psi._odata[ss+0]);
-          chi[ss+s] = diag[s]*phi[ss+s] + upper[s]*tmp;
-          spProj5m(tmp, psi._odata[ss+s-1]);
-          chi[ss+s] = chi[ss+s] + lower[s]*tmp;
-        } else {
-          spProj5p(tmp, psi._odata[ss+s+1]);
-          chi[ss+s] = diag[s]*phi[ss+s] + upper[s]*tmp;
-          spProj5m(tmp, psi._odata[ss+s-1]);
-          chi[ss+s] = chi[ss+s] + lower[s]*tmp;
-        }
-      }
-    }
-
-    this->M5Dtime += usecond();
-  }
-
-  template<class Impl>
-  void MobiusEOFAFermion<Impl>::M5Ddag_shift(const FermionField &psi, const FermionField &phi, FermionField &chi,
-    std::vector<Coeff_t> &lower, std::vector<Coeff_t> &diag, std::vector<Coeff_t> &upper,
-    std::vector<Coeff_t> &shift_coeffs)
-  {
-    int Ls = this->Ls;
-    int shift_s = (this->pm == 1) ? (Ls-1) : 0; // s-component modified by shift operator
-    GridBase *grid = psi._grid;
-
-    assert(phi.checkerboard == psi.checkerboard);
-    chi.checkerboard = psi.checkerboard;
-
-    // Flops = 6.0*(Nc*Ns) *Ls*vol
-    this->M5Dcalls++;
-    this->M5Dtime -= usecond();
-
-    parallel_for(int ss=0; ss<grid->oSites(); ss+=Ls){
-      chi[ss+Ls-1] = zero;
-      auto tmp = psi._odata[0];
-      for(int s=0; s<Ls; s++){
-        if(s==0) {
-          spProj5p(tmp, psi._odata[ss+s+1]);
-          chi[ss+s] = diag[s]*phi[ss+s] + upper[s]*tmp;
-          spProj5m(tmp, psi._odata[ss+Ls-1]);
-          chi[ss+s] = chi[ss+s] + lower[s]*tmp;
-        } else if(s==(Ls-1)) {
-          spProj5p(tmp, psi._odata[ss+0]);
-          chi[ss+s] = chi[ss+s] + diag[s]*phi[ss+s] + upper[s]*tmp;
-          spProj5m(tmp, psi._odata[ss+s-1]);
-          chi[ss+s] = chi[ss+s] + lower[s]*tmp;
-        } else {
-          spProj5p(tmp, psi._odata[ss+s+1]);
-          chi[ss+s] = diag[s]*phi[ss+s] + upper[s]*tmp;
-          spProj5m(tmp, psi._odata[ss+s-1]);
-          chi[ss+s] = chi[ss+s] + lower[s]*tmp;
-        }
-        if(this->pm == 1){ spProj5p(tmp, psi._odata[ss+s]); }
-        else{ spProj5m(tmp, psi._odata[ss+s]); }
-        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, FermionField &chi)
-  {
-    if(this->shift != 0.0){ MooeeInv_shift(psi,chi); return; }
-
-    GridBase *grid = psi._grid;
-    int Ls = this->Ls;
-
-    chi.checkerboard = psi.checkerboard;
-
-    this->MooeeInvCalls++;
-    this->MooeeInvTime -= usecond();
-
-    parallel_for(int ss=0; ss<grid->oSites(); ss+=Ls){
-
-      auto tmp = psi._odata[0];
-
-      // Apply (L^{\prime})^{-1}
-      chi[ss] = psi[ss]; // chi[0]=psi[0]
-      for(int s=1; s<Ls; s++){
-        spProj5p(tmp, chi[ss+s-1]);
-        chi[ss+s] = psi[ss+s] - this->lee[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]);
-        chi[ss+Ls-1] = chi[ss+Ls-1] - this->leem[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]);
-        chi[ss+s] = (1.0/this->dee[s])*chi[ss+s] - (this->ueem[s]/this->dee[Ls-1])*tmp;
-      }
-      chi[ss+Ls-1] = (1.0/this->dee[Ls-1])*chi[ss+Ls-1];
-
-      // Apply U^{-1}
-      for(int s=Ls-2; s>=0; s--){
-        spProj5m(tmp, chi[ss+s+1]);
-        chi[ss+s] = chi[ss+s] - this->uee[s]*tmp;
-      }
-    }
-
-    this->MooeeInvTime += usecond();
-  }
-
-  template<class Impl>
-  void MobiusEOFAFermion<Impl>::MooeeInv_shift(const FermionField &psi, FermionField &chi)
-  {
-    GridBase *grid = psi._grid;
-    int Ls = this->Ls;
-
-    chi.checkerboard = psi.checkerboard;
-
-    this->MooeeInvCalls++;
-    this->MooeeInvTime -= usecond();
-
-    parallel_for(int ss=0; ss<grid->oSites(); ss+=Ls){
-
-      auto tmp1        = psi._odata[0];
-      auto tmp2        = psi._odata[0];
-      auto tmp2_spProj = psi._odata[0];
-
-      // Apply (L^{\prime})^{-1} and accumulate MooeeInv_shift_lc[j]*psi[j] in tmp2
-      chi[ss] = psi[ss]; // chi[0]=psi[0]
-      tmp2 = MooeeInv_shift_lc[0]*psi[ss];
-      for(int s=1; s<Ls; s++){
-        spProj5p(tmp1, chi[ss+s-1]);
-        chi[ss+s] = psi[ss+s] - this->lee[s-1]*tmp1;
-        tmp2 = tmp2 + MooeeInv_shift_lc[s]*psi[ss+s];
-      }
-      if(this->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]);
-        chi[ss+Ls-1] = chi[ss+Ls-1] - this->leem[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]);
-        chi[ss+s] = (1.0/this->dee[s])*chi[ss+s] - (this->ueem[s]/this->dee[Ls-1])*tmp1;
-      }
-      // chi[ss+Ls-1] = (1.0/this->dee[Ls-1])*chi[ss+Ls-1] + MooeeInv_shift_norm[Ls-1]*tmp2_spProj;
-      chi[ss+Ls-1] = (1.0/this->dee[Ls-1])*chi[ss+Ls-1];
-      spProj5m(tmp1, chi[ss+Ls-1]);
-      chi[ss+Ls-1] = chi[ss+Ls-1] + MooeeInv_shift_norm[Ls-1]*tmp2_spProj;
-
-      // Apply U^{-1} and add shift term
-      for(int s=Ls-2; s>=0; s--){
-        chi[ss+s] = chi[ss+s] - this->uee[s]*tmp1;
-        spProj5m(tmp1, chi[ss+s]);
-        chi[ss+s] = chi[ss+s] + MooeeInv_shift_norm[s]*tmp2_spProj;
-      }
-    }
-
-    this->MooeeInvTime += usecond();
-  }
-
-  template<class Impl>
-  void MobiusEOFAFermion<Impl>::MooeeInvDag(const FermionField &psi, FermionField &chi)
-  {
-    if(this->shift != 0.0){ MooeeInvDag_shift(psi,chi); return; }
-
-    GridBase *grid = psi._grid;
-    int Ls = this->Ls;
-
-    chi.checkerboard = psi.checkerboard;
-
-    this->MooeeInvCalls++;
-    this->MooeeInvTime -= usecond();
-
-    parallel_for(int ss=0; ss<grid->oSites(); ss+=Ls){
-
-      auto tmp = psi._odata[0];
-
-      // Apply (U^{\prime})^{-dag}
-      chi[ss] = psi[ss];
-      for(int s=1; s<Ls; s++){
-        spProj5m(tmp, chi[ss+s-1]);
-        chi[ss+s] = psi[ss+s] - this->uee[s-1]*tmp;
-      }
-
-      // U_m^{-\dag}
-      for(int s=0; s<Ls-1; s++){
-        spProj5p(tmp, chi[ss+s]);
-        chi[ss+Ls-1] = chi[ss+Ls-1] - this->ueem[s]*tmp;
-      }
-
-      // L_m^{-\dag} D^{-dag}
-      for(int s=0; s<Ls-1; s++){
-        spProj5m(tmp, chi[ss+Ls-1]);
-        chi[ss+s] = (1.0/this->dee[s])*chi[ss+s] - (this->leem[s]/this->dee[Ls-1])*tmp;
-      }
-      chi[ss+Ls-1] = (1.0/this->dee[Ls-1])*chi[ss+Ls-1];
-
-      // Apply L^{-dag}
-      for(int s=Ls-2; s>=0; s--){
-        spProj5p(tmp, chi[ss+s+1]);
-        chi[ss+s] = chi[ss+s] - this->lee[s]*tmp;
-      }
-    }
-
-    this->MooeeInvTime += usecond();
-  }
-
-  template<class Impl>
-  void MobiusEOFAFermion<Impl>::MooeeInvDag_shift(const FermionField &psi, FermionField &chi)
-  {
-    GridBase *grid = psi._grid;
-    int Ls = this->Ls;
-
-    chi.checkerboard = psi.checkerboard;
-
-    this->MooeeInvCalls++;
-    this->MooeeInvTime -= usecond();
-
-    parallel_for(int ss=0; ss<grid->oSites(); ss+=Ls){
-
-      auto tmp1        = psi._odata[0];
-      auto tmp2        = psi._odata[0];
-      auto tmp2_spProj = psi._odata[0];
-
-      // Apply (U^{\prime})^{-dag} and accumulate MooeeInvDag_shift_lc[j]*psi[j] in tmp2
-      chi[ss] = psi[ss];
-      tmp2 = MooeeInvDag_shift_lc[0]*psi[ss];
-      for(int s=1; s<Ls; s++){
-        spProj5m(tmp1, chi[ss+s-1]);
-        chi[ss+s] = psi[ss+s] - this->uee[s-1]*tmp1;
-        tmp2 = tmp2 + MooeeInvDag_shift_lc[s]*psi[ss+s];
-      }
-      if(this->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]);
-        chi[ss+Ls-1] = chi[ss+Ls-1] - this->ueem[s]*tmp1;
-      }
-
-      // L_m^{-\dag} D^{-dag}
-      for(int s=0; s<Ls-1; s++){
-        spProj5m(tmp1, chi[ss+Ls-1]);
-        chi[ss+s] = (1.0/this->dee[s])*chi[ss+s] - (this->leem[s]/this->dee[Ls-1])*tmp1;
-      }
-      chi[ss+Ls-1] = (1.0/this->dee[Ls-1])*chi[ss+Ls-1];
-      spProj5p(tmp1, chi[ss+Ls-1]);
-      chi[ss+Ls-1] = chi[ss+Ls-1] + MooeeInvDag_shift_norm[Ls-1]*tmp2_spProj;
-
-      // Apply L^{-dag}
-      for(int s=Ls-2; s>=0; s--){
-        chi[ss+s] = chi[ss+s] - this->lee[s]*tmp1;
-        spProj5p(tmp1, chi[ss+s]);
-        chi[ss+s] = chi[ss+s] + MooeeInvDag_shift_norm[s]*tmp2_spProj;
-      }
-    }
-
-    this->MooeeInvTime += usecond();
-  }
-
-  #ifdef MOBIUS_EOFA_DPERP_CACHE
-
-    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
-
-}}

Grid/qcd/action/fermion/MobiusEOFAFermiondense.cc

@@ -1,184 +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 Grid {
-namespace QCD {
-
-  /*
-  * 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
-
-}}

Grid/qcd/action/fermion/MobiusEOFAFermionssp.cc

@@ -1,290 +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 Grid {
-namespace QCD {
-
-  // 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, std::vector<Coeff_t>& lower, std::vector<Coeff_t>& diag, std::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, std::vector<Coeff_t>& lower, std::vector<Coeff_t>& diag, std::vector<Coeff_t>& upper,
-    std::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, std::vector<Coeff_t>& lower, std::vector<Coeff_t>& diag, std::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, std::vector<Coeff_t>& lower, std::vector<Coeff_t>& diag, std::vector<Coeff_t>& upper,
-    std::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
-
-}}

Grid/qcd/action/fermion/MobiusEOFAFermionvec.cc

@@ -1,983 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/MobiusEOFAFermionvec.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 Grid {
-namespace QCD {
-
-  /*
-  * 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>::M5D(const FermionField& psi, const FermionField& phi,
-    FermionField& chi, std::vector<Coeff_t>& lower, std::vector<Coeff_t>& diag, std::vector<Coeff_t>& upper)
-  {
-    GridBase* grid  = psi._grid;
-    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);
-
-    chi.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];
-    }}
-
-    this->M5Dcalls++;
-    this->M5Dtime -= usecond();
-
-    assert(Nc == 3);
-
-    parallel_for(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
-    }
-
-    this->M5Dtime += usecond();
-  }
-
-  template<class Impl>
-  void MobiusEOFAFermion<Impl>::M5D_shift(const FermionField& psi, const FermionField& phi,
-    FermionField& chi, std::vector<Coeff_t>& lower, std::vector<Coeff_t>& diag, std::vector<Coeff_t>& upper,
-    std::vector<Coeff_t>& shift_coeffs)
-  {
-    #if 0
-
-      this->M5D(psi, phi, chi, lower, diag, upper);
-
-      // FIXME: possible gain from vectorizing shift operation as well?
-      Coeff_t one(1.0);
-      int Ls = this->Ls;
-      for(int s=0; s<Ls; s++){
-        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); }
-      }
-
-    #else
-
-      GridBase* grid  = psi._grid;
-      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);
-      Vector<iSinglet<Simd>> s(LLs);
-
-      assert(Ls/LLs == nsimd);
-      assert(phi.checkerboard == psi.checkerboard);
-
-      chi.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];
-      scalar_type* s_p = (scalar_type*) &s[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];
-        s_p[ss] = shift_coeffs[s];
-      }}
-
-      this->M5Dcalls++;
-      this->M5Dtime -= usecond();
-
-      assert(Nc == 3);
-
-      parallel_for(int ss=0; ss<grid->oSites(); ss+=LLs){ // adds LLs
-
-        int vs     = (this->pm == 1) ? LLs-1 : 0;
-        Simd hs_00 = (this->pm == 1) ? psi[ss+vs]()(2)(0) : psi[ss+vs]()(0)(0);
-        Simd hs_01 = (this->pm == 1) ? psi[ss+vs]()(2)(1) : psi[ss+vs]()(0)(1);
-        Simd hs_02 = (this->pm == 1) ? psi[ss+vs]()(2)(2) : psi[ss+vs]()(0)(2);
-        Simd hs_10 = (this->pm == 1) ? psi[ss+vs]()(3)(0) : psi[ss+vs]()(1)(0);
-        Simd hs_11 = (this->pm == 1) ? psi[ss+vs]()(3)(1) : psi[ss+vs]()(1)(1);
-        Simd hs_12 = (this->pm == 1) ? psi[ss+vs]()(3)(2) : psi[ss+vs]()(1)(2);
-
-        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(this->pm == 1 && vs <= v){
-            hs_00.v = Optimization::Rotate::tRotate<2>(hs_00.v);
-            hs_01.v = Optimization::Rotate::tRotate<2>(hs_01.v);
-            hs_02.v = Optimization::Rotate::tRotate<2>(hs_02.v);
-            hs_10.v = Optimization::Rotate::tRotate<2>(hs_10.v);
-            hs_11.v = Optimization::Rotate::tRotate<2>(hs_11.v);
-            hs_12.v = Optimization::Rotate::tRotate<2>(hs_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);
-          }
-
-          if(this->pm == -1 && vs >= v){
-            hs_00.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hs_00.v);
-            hs_01.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hs_01.v);
-            hs_02.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hs_02.v);
-            hs_10.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hs_10.v);
-            hs_11.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hs_11.v);
-            hs_12.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hs_12.v);
-          }
-
-          // Can force these to real arithmetic and save 2x.
-          Simd p_00 = (this->pm == 1) ? switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(0)(0)) + switcheroo<Coeff_t>::mult(l[v]()()(), hm_00)
-                                      : switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(0)(0)) + switcheroo<Coeff_t>::mult(l[v]()()(), hm_00)
-                                                                                                 + switcheroo<Coeff_t>::mult(s[v]()()(), hs_00);
-          Simd p_01 = (this->pm == 1) ? switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(0)(1)) + switcheroo<Coeff_t>::mult(l[v]()()(), hm_01)
-                                      : switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(0)(1)) + switcheroo<Coeff_t>::mult(l[v]()()(), hm_01)
-                                                                                                 + switcheroo<Coeff_t>::mult(s[v]()()(), hs_01);
-          Simd p_02 = (this->pm == 1) ? switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(0)(2)) + switcheroo<Coeff_t>::mult(l[v]()()(), hm_02)
-                                      : switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(0)(2)) + switcheroo<Coeff_t>::mult(l[v]()()(), hm_02)
-                                                                                                 + switcheroo<Coeff_t>::mult(s[v]()()(), hs_02);
-          Simd p_10 = (this->pm == 1) ? switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(1)(0)) + switcheroo<Coeff_t>::mult(l[v]()()(), hm_10)
-                                      : switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(1)(0)) + switcheroo<Coeff_t>::mult(l[v]()()(), hm_10)
-                                                                                                 + switcheroo<Coeff_t>::mult(s[v]()()(), hs_10);
-          Simd p_11 = (this->pm == 1) ? switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(1)(1)) + switcheroo<Coeff_t>::mult(l[v]()()(), hm_11)
-                                      : switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(1)(1)) + switcheroo<Coeff_t>::mult(l[v]()()(), hm_11)
-                                                                                                 + switcheroo<Coeff_t>::mult(s[v]()()(), hs_11);
-          Simd p_12 = (this->pm == 1) ? switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(1)(2)) + switcheroo<Coeff_t>::mult(l[v]()()(), hm_12)
-                                      : switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(1)(2)) + switcheroo<Coeff_t>::mult(l[v]()()(), hm_12)
-                                                                                                 + switcheroo<Coeff_t>::mult(s[v]()()(), hs_12);
-          Simd p_20 = (this->pm == 1) ? switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(2)(0)) + switcheroo<Coeff_t>::mult(u[v]()()(), hp_00)
-                                                                                                 + switcheroo<Coeff_t>::mult(s[v]()()(), hs_00)
-                                      : switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(2)(0)) + switcheroo<Coeff_t>::mult(u[v]()()(), hp_00);
-          Simd p_21 = (this->pm == 1) ? switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(2)(1)) + switcheroo<Coeff_t>::mult(u[v]()()(), hp_01)
-                                                                                                 + switcheroo<Coeff_t>::mult(s[v]()()(), hs_01)
-                                      : switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(2)(1)) + switcheroo<Coeff_t>::mult(u[v]()()(), hp_01);
-          Simd p_22 = (this->pm == 1) ? switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(2)(2)) + switcheroo<Coeff_t>::mult(u[v]()()(), hp_02)
-                                                                                                 + switcheroo<Coeff_t>::mult(s[v]()()(), hs_02)
-                                      : switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(2)(2)) + switcheroo<Coeff_t>::mult(u[v]()()(), hp_02);
-          Simd p_30 = (this->pm == 1) ? switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(3)(0)) + switcheroo<Coeff_t>::mult(u[v]()()(), hp_10)
-                                                                                                 + switcheroo<Coeff_t>::mult(s[v]()()(), hs_10)
-                                      : switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(3)(0)) + switcheroo<Coeff_t>::mult(u[v]()()(), hp_10);
-          Simd p_31 = (this->pm == 1) ? switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(3)(1)) + switcheroo<Coeff_t>::mult(u[v]()()(), hp_11)
-                                                                                                 + switcheroo<Coeff_t>::mult(s[v]()()(), hs_11)
-                                      : switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(3)(1)) + switcheroo<Coeff_t>::mult(u[v]()()(), hp_11);
-          Simd p_32 = (this->pm == 1) ? switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(3)(2)) + switcheroo<Coeff_t>::mult(u[v]()()(), hp_12)
-                                                                                                 + switcheroo<Coeff_t>::mult(s[v]()()(), hs_12)
-                                      : 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);
-        }
-      }
-
-      this->M5Dtime += usecond();
-
-    #endif
-  }
-
-  template<class Impl>
-  void MobiusEOFAFermion<Impl>::M5Ddag(const FermionField& psi, const FermionField& phi,
-    FermionField& chi, std::vector<Coeff_t>& lower, std::vector<Coeff_t>& diag, std::vector<Coeff_t>& upper)
-  {
-    GridBase* grid = psi._grid;
-    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);
-
-    chi.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];
-    }}
-
-    this->M5Dcalls++;
-    this->M5Dtime -= usecond();
-
-    parallel_for(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
-
-    }
-
-    this->M5Dtime += usecond();
-  }
-
-  template<class Impl>
-  void MobiusEOFAFermion<Impl>::M5Ddag_shift(const FermionField& psi, const FermionField& phi,
-    FermionField& chi, std::vector<Coeff_t>& lower, std::vector<Coeff_t>& diag, std::vector<Coeff_t>& upper,
-    std::vector<Coeff_t>& shift_coeffs)
-  {
-    #if 0
-
-      this->M5Ddag(psi, phi, chi, lower, diag, upper);
-
-      // FIXME: possible gain from vectorizing shift operation as well?
-      Coeff_t one(1.0);
-      int Ls = this->Ls;
-      for(int s=0; s<Ls; s++){
-        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); }
-      }
-
-    #else
-
-      GridBase* grid = psi._grid;
-      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);
-      Vector<iSinglet<Simd>> s(LLs);
-
-      assert(Ls/LLs == nsimd);
-      assert(phi.checkerboard == psi.checkerboard);
-
-      chi.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];
-      scalar_type* s_p = (scalar_type*) &s[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];
-        s_p[ss] = shift_coeffs[s];
-      }}
-
-      this->M5Dcalls++;
-      this->M5Dtime -= usecond();
-
-      parallel_for(int ss=0; ss<grid->oSites(); ss+=LLs){ // adds LLs
-
-        int vs     = (this->pm == 1) ? LLs-1 : 0;
-        Simd hs_00 = (this->pm == 1) ? psi[ss+vs]()(0)(0) : psi[ss+vs]()(2)(0);
-        Simd hs_01 = (this->pm == 1) ? psi[ss+vs]()(0)(1) : psi[ss+vs]()(2)(1);
-        Simd hs_02 = (this->pm == 1) ? psi[ss+vs]()(0)(2) : psi[ss+vs]()(2)(2);
-        Simd hs_10 = (this->pm == 1) ? psi[ss+vs]()(1)(0) : psi[ss+vs]()(3)(0);
-        Simd hs_11 = (this->pm == 1) ? psi[ss+vs]()(1)(1) : psi[ss+vs]()(3)(1);
-        Simd hs_12 = (this->pm == 1) ? psi[ss+vs]()(1)(2) : psi[ss+vs]()(3)(2);
-
-        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(this->pm == 1 && vs <= v){
-            hs_00.v = Optimization::Rotate::tRotate<2>(hs_00.v);
-            hs_01.v = Optimization::Rotate::tRotate<2>(hs_01.v);
-            hs_02.v = Optimization::Rotate::tRotate<2>(hs_02.v);
-            hs_10.v = Optimization::Rotate::tRotate<2>(hs_10.v);
-            hs_11.v = Optimization::Rotate::tRotate<2>(hs_11.v);
-            hs_12.v = Optimization::Rotate::tRotate<2>(hs_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);
-          }
-
-          if(this->pm == -1 && vs >= v){
-            hs_00.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hs_00.v);
-            hs_01.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hs_01.v);
-            hs_02.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hs_02.v);
-            hs_10.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hs_10.v);
-            hs_11.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hs_11.v);
-            hs_12.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hs_12.v);
-          }
-
-          Simd p_00 = (this->pm == 1) ? switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(0)(0)) + switcheroo<Coeff_t>::mult(u[v]()()(), hp_00)
-                                                                                                 + switcheroo<Coeff_t>::mult(s[v]()()(), hs_00)
-                                      : switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(0)(0)) + switcheroo<Coeff_t>::mult(u[v]()()(), hp_00);
-          Simd p_01 = (this->pm == 1) ? switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(0)(1)) + switcheroo<Coeff_t>::mult(u[v]()()(), hp_01)
-                                                                                                 + switcheroo<Coeff_t>::mult(s[v]()()(), hs_01)
-                                      : switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(0)(1)) + switcheroo<Coeff_t>::mult(u[v]()()(), hp_01);
-          Simd p_02 = (this->pm == 1) ? switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(0)(2)) + switcheroo<Coeff_t>::mult(u[v]()()(), hp_02)
-                                                                                                 + switcheroo<Coeff_t>::mult(s[v]()()(), hs_02)
-                                      : switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(0)(2)) + switcheroo<Coeff_t>::mult(u[v]()()(), hp_02);
-          Simd p_10 = (this->pm == 1) ? switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(1)(0)) + switcheroo<Coeff_t>::mult(u[v]()()(), hp_10)
-                                                                                                 + switcheroo<Coeff_t>::mult(s[v]()()(), hs_10)
-                                      : switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(1)(0)) + switcheroo<Coeff_t>::mult(u[v]()()(), hp_10);
-          Simd p_11 = (this->pm == 1) ? switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(1)(1)) + switcheroo<Coeff_t>::mult(u[v]()()(), hp_11)
-                                                                                                 + switcheroo<Coeff_t>::mult(s[v]()()(), hs_11)
-                                      : switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(1)(1)) + switcheroo<Coeff_t>::mult(u[v]()()(), hp_11);
-          Simd p_12 = (this->pm == 1) ? switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(1)(2)) + switcheroo<Coeff_t>::mult(u[v]()()(), hp_12)
-                                                                                                 + switcheroo<Coeff_t>::mult(s[v]()()(), hs_12)
-                                      : switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(1)(2)) + switcheroo<Coeff_t>::mult(u[v]()()(), hp_12);
-          Simd p_20 = (this->pm == 1) ? switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(2)(0)) + switcheroo<Coeff_t>::mult(l[v]()()(), hm_00)
-                                      : switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(2)(0)) + switcheroo<Coeff_t>::mult(l[v]()()(), hm_00)
-                                                                                                 + switcheroo<Coeff_t>::mult(s[v]()()(), hs_00);
-          Simd p_21 = (this->pm == 1) ? switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(2)(1)) + switcheroo<Coeff_t>::mult(l[v]()()(), hm_01)
-                                      : switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(2)(1)) + switcheroo<Coeff_t>::mult(l[v]()()(), hm_01)
-                                                                                                 + switcheroo<Coeff_t>::mult(s[v]()()(), hs_01);
-          Simd p_22 = (this->pm == 1) ? switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(2)(2)) + switcheroo<Coeff_t>::mult(l[v]()()(), hm_02)
-                                      : switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(2)(2)) + switcheroo<Coeff_t>::mult(l[v]()()(), hm_02)
-                                                                                                 + switcheroo<Coeff_t>::mult(s[v]()()(), hs_02);
-          Simd p_30 = (this->pm == 1) ? switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(3)(0)) + switcheroo<Coeff_t>::mult(l[v]()()(), hm_10)
-                                      : switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(3)(0)) + switcheroo<Coeff_t>::mult(l[v]()()(), hm_10)
-                                                                                                 + switcheroo<Coeff_t>::mult(s[v]()()(), hs_10);
-          Simd p_31 = (this->pm == 1) ? switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(3)(1)) + switcheroo<Coeff_t>::mult(l[v]()()(), hm_11)
-                                      : switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(3)(1)) + switcheroo<Coeff_t>::mult(l[v]()()(), hm_11)
-                                                                                                 + switcheroo<Coeff_t>::mult(s[v]()()(), hs_11);
-          Simd p_32 = (this->pm == 1) ? switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(3)(2)) + switcheroo<Coeff_t>::mult(l[v]()()(), hm_12)
-                                      : switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(3)(2)) + switcheroo<Coeff_t>::mult(l[v]()()(), hm_12)
-                                                                                                 + switcheroo<Coeff_t>::mult(s[v]()()(), hs_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);
-
-        }
-
-      }
-
-      this->M5Dtime += usecond();
-
-    #endif
-  }
-
-  #ifdef AVX512
-    #include<simd/Intel512common.h>
-    #include<simd/Intel512avx.h>
-    #include<simd/Intel512single.h>
-  #endif
-
-  template<class Impl>
-  void MobiusEOFAFermion<Impl>::MooeeInternalAsm(const FermionField& psi, FermionField& chi,
-    int LLs, int site, Vector<iSinglet<Simd> >& Matp, Vector<iSinglet<Simd> >& Matm)
-  {
-    #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 MobiusEOFAFermion<Impl>::MooeeInternalZAsm(const FermionField& psi, FermionField& chi,
-    int LLs, int site, Vector<iSinglet<Simd> >& Matp, Vector<iSinglet<Simd> >& Matm)
-  {
-    std::cout << "Error: zMobius not implemented for EOFA" << std::endl;
-    exit(-1);
-  };
-
-  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;
-
-    chi.checkerboard = psi.checkerboard;
-
-    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 = &this->MatpInvDag;
-      _Matm = &this->MatmInvDag;
-    }
-
-    if(inv && (!dag)){
-      _Matp = &this->MatpInv;
-      _Matm = &this->MatmInv;
-    }
-
-    if(!inv){
-      MooeeInternalCompute(dag, inv, Matp, Matm);
-      _Matp = &Matp;
-      _Matm = &Matm;
-    }
-
-    assert(_Matp->size() == Ls*LLs);
-
-    this->MooeeInvCalls++;
-    this->MooeeInvTime -= usecond();
-
-    if(switcheroo<Coeff_t>::iscomplex()){
-      parallel_for(auto site=0; site<vol; site++){
-        MooeeInternalZAsm(psi, chi, LLs, site, *_Matp, *_Matm);
-      }
-    } else {
-      parallel_for(auto site=0; site<vol; site++){
-        MooeeInternalAsm(psi, chi, LLs, site, *_Matp, *_Matm);
-      }
-    }
-
-    this->MooeeInvTime += usecond();
-  }
-
-  #ifdef MOBIUS_EOFA_DPERP_VEC
-
-    INSTANTIATE_DPERP_MOBIUS_EOFA(DomainWallVec5dImplD);
-    INSTANTIATE_DPERP_MOBIUS_EOFA(DomainWallVec5dImplF);
-    INSTANTIATE_DPERP_MOBIUS_EOFA(ZDomainWallVec5dImplD);
-    INSTANTIATE_DPERP_MOBIUS_EOFA(ZDomainWallVec5dImplF);
-
-    INSTANTIATE_DPERP_MOBIUS_EOFA(DomainWallVec5dImplDF);
-    INSTANTIATE_DPERP_MOBIUS_EOFA(DomainWallVec5dImplFH);
-    INSTANTIATE_DPERP_MOBIUS_EOFA(ZDomainWallVec5dImplDF);
-    INSTANTIATE_DPERP_MOBIUS_EOFA(ZDomainWallVec5dImplFH);
-
-    template void MobiusEOFAFermion<DomainWallVec5dImplF>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-    template void MobiusEOFAFermion<DomainWallVec5dImplD>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-    template void MobiusEOFAFermion<ZDomainWallVec5dImplF>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-    template void MobiusEOFAFermion<ZDomainWallVec5dImplD>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-
-    template void MobiusEOFAFermion<DomainWallVec5dImplFH>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-    template void MobiusEOFAFermion<DomainWallVec5dImplDF>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-    template void MobiusEOFAFermion<ZDomainWallVec5dImplFH>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-    template void MobiusEOFAFermion<ZDomainWallVec5dImplDF>::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv);
-
-  #endif
-
-}}

Grid/qcd/action/fermion/PauliVillarsInverters.h

@@ -1,95 +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 Grid {
-namespace QCD {
-
-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);
-  };
-};
-
-
-}
-}

Grid/qcd/action/fermion/Reconstruct5Dprop.h

@@ -1,135 +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 Grid {
-namespace QCD {
-
-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);
-    
-  }
-};
-
-}
-}

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 */
-#ifndef  _SCHUR_DIAG_TWO_KAPPA_H
-#define  _SCHUR_DIAG_TWO_KAPPA_H
-
-namespace Grid {
-
-  // This is specific to (Z)mobius fermions
-  template<class Matrix, class Field>
-    class KappaSimilarityTransform {
-  public:
-    INHERIT_IMPL_TYPES(Matrix);
-    std::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];
-    parallel_for(int ss=0;ss<grid->oSites();ss++){
-      vobj tmp = s[ss % Ls]*in._odata[ss];
-      vstream(out._odata[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);
-    }
-  };
-
-}
-
-#endif

Grid/qcd/action/fermion/StaggeredKernels.cc

@@ -1,294 +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>
-
-namespace Grid {
-namespace QCD {
-
-int StaggeredKernelsStatic::Opt= StaggeredKernelsStatic::OptGeneric;
-int StaggeredKernelsStatic::Comms = StaggeredKernelsStatic::CommsAndCompute;
-
-#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._odata[SE->_offset], ptype);		\
-    } else {							\
-      chi_p = &in._odata[SE->_offset];				\
-    }								\
-  } else {							\
-    chi_p = &buf[SE->_offset];					\
-  }								\
-  multLink(Uchi, U._odata[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._odata[SE->_offset], ptype);		\
-    } else {							\
-      chi_p = &in._odata[SE->_offset];				\
-    }								\
-  } else if ( st.same_node[Dir] ) {				\
-    chi_p = &buf[SE->_offset];					\
-  }								\
-  if (SE->_is_local || st.same_node[Dir] ) {			\
-    multLink(Uchi, U._odata[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._odata[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, 
-					     DoubledGaugeField &U, DoubledGaugeField &UUU,
-					     SiteSpinor *buf, int LLs, int sU, 
-					     const FermionField &in, FermionField &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._odata[sF], Uchi);
-  }
-};
-
-  ///////////////////////////////////////////////////
-  // Only contributions from interior of our node
-  ///////////////////////////////////////////////////
-template <class Impl>
-void StaggeredKernels<Impl>::DhopSiteGenericInt(StencilImpl &st, LebesgueOrder &lo, 
-						DoubledGaugeField &U, DoubledGaugeField &UUU,
-						SiteSpinor *buf, int LLs, int sU, 
-						const FermionField &in, FermionField &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._odata[sF], Uchi);
-  }
-};
-
-
-  ///////////////////////////////////////////////////
-  // Only contributions from exterior of our node
-  ///////////////////////////////////////////////////
-template <class Impl>
-void StaggeredKernels<Impl>::DhopSiteGenericExt(StencilImpl &st, LebesgueOrder &lo, 
-						DoubledGaugeField &U, DoubledGaugeField &UUU,
-						SiteSpinor *buf, int LLs, int sU,
-						const FermionField &in, FermionField &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._odata[sF] = out._odata[sF] - Uchi;
-      } else { 
-	out._odata[sF] = out._odata[sF] + Uchi;
-      }
-    }
-  }
-};
-
-////////////////////////////////////////////////////////////////////////////////////
-// Driving / wrapping routine to select right kernel
-////////////////////////////////////////////////////////////////////////////////////
-
-template <class Impl>
-void StaggeredKernels<Impl>::DhopSiteDag(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, DoubledGaugeField &UUU,
-					 SiteSpinor *buf, int LLs, int sU,
-					 const FermionField &in, FermionField &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, DoubledGaugeField &U, DoubledGaugeField &UUU,
-				      SiteSpinor *buf, int LLs, int sU,
-				      const FermionField &in, FermionField &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, DoubledGaugeField &U, DoubledGaugeField &UUU,
-				      SiteSpinor *buf, int LLs,
-				      int sU, const FermionField &in, FermionField &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>::DhopDir( StencilImpl &st, DoubledGaugeField &U,  DoubledGaugeField &UUU, SiteSpinor *buf, int sF,
-				      int sU, const FermionField &in, FermionField &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);
-}
-
-FermOpStaggeredTemplateInstantiate(StaggeredKernels);
-FermOpStaggeredVec5dTemplateInstantiate(StaggeredKernels);
-
-}}
-

Grid/qcd/action/fermion/StaggeredKernels.h

@@ -1,122 +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 */
-#ifndef GRID_QCD_STAGGERED_KERNELS_H
-#define GRID_QCD_STAGGERED_KERNELS_H
-
-namespace Grid {
-namespace QCD {
-
-  ////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-  // 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 DhopDir(StencilImpl &st, DoubledGaugeField &U, DoubledGaugeField &UUU, SiteSpinor * buf,
-		      int sF, int sU, const FermionField &in, FermionField &out, int dir,int disp);
-
-   ///////////////////////////////////////////////////////////////////////////////////////
-   // Generic Nc kernels
-   ///////////////////////////////////////////////////////////////////////////////////////
-   void DhopSiteGeneric(StencilImpl &st, LebesgueOrder &lo, 
-			DoubledGaugeField &U, DoubledGaugeField &UUU, 
-			SiteSpinor * buf, int LLs, int sU, 
-			const FermionField &in, FermionField &out,int dag);
-   void DhopSiteGenericInt(StencilImpl &st, LebesgueOrder &lo, 
-			   DoubledGaugeField &U, DoubledGaugeField &UUU, 
-			   SiteSpinor * buf, int LLs, int sU, 
-			   const FermionField &in, FermionField &out,int dag);
-   void DhopSiteGenericExt(StencilImpl &st, LebesgueOrder &lo, 
-			   DoubledGaugeField &U, DoubledGaugeField &UUU,
-			   SiteSpinor * buf, int LLs, int sU, 
-			   const FermionField &in, FermionField &out,int dag);
-
-   ///////////////////////////////////////////////////////////////////////////////////////
-   // Nc=3 specific kernels
-   ///////////////////////////////////////////////////////////////////////////////////////
-   void DhopSiteHand(StencilImpl &st, LebesgueOrder &lo, 
-		     DoubledGaugeField &U,DoubledGaugeField &UUU, 
-		     SiteSpinor * buf, int LLs, int sU, 
-		     const FermionField &in, FermionField &out,int dag);
-   void DhopSiteHandInt(StencilImpl &st, LebesgueOrder &lo, 
-			DoubledGaugeField &U,DoubledGaugeField &UUU, 
-			SiteSpinor * buf, int LLs, int sU, 
-			const FermionField &in, FermionField &out,int dag);
-   void DhopSiteHandExt(StencilImpl &st, LebesgueOrder &lo, 
-			DoubledGaugeField &U,DoubledGaugeField &UUU, 
-			SiteSpinor * buf, int LLs, int sU, 
-			const FermionField &in, FermionField &out,int dag);
-
-   ///////////////////////////////////////////////////////////////////////////////////////
-   // Asm Nc=3 specific kernels
-   ///////////////////////////////////////////////////////////////////////////////////////
-   void DhopSiteAsm(StencilImpl &st, LebesgueOrder &lo, 
-		    DoubledGaugeField &U,DoubledGaugeField &UUU, 
-		    SiteSpinor * buf, int LLs, int sU, 
-		    const FermionField &in, FermionField &out,int dag);
-   ///////////////////////////////////////////////////////////////////////////////////////////////////
-   // Generic interface; fan out to right routine
-   ///////////////////////////////////////////////////////////////////////////////////////////////////
-   void DhopSite(StencilImpl &st, LebesgueOrder &lo, 
-		 DoubledGaugeField &U, DoubledGaugeField &UUU, 
-		 SiteSpinor * buf, int LLs, int sU,
-		 const FermionField &in, FermionField &out, int interior=1,int exterior=1);
-
-   void DhopSiteDag(StencilImpl &st, LebesgueOrder &lo, 
-		    DoubledGaugeField &U, DoubledGaugeField &UUU, 
-		    SiteSpinor * buf, int LLs, int sU,
-		    const FermionField &in, FermionField &out, int interior=1,int exterior=1);
-
-   void DhopSite(StencilImpl &st, LebesgueOrder &lo, 
-		 DoubledGaugeField &U, DoubledGaugeField &UUU, 
-		 SiteSpinor * buf, int LLs, int sU,
-		 const FermionField &in, FermionField &out, int dag, int interior,int exterior);
-  
-public:
-
-  StaggeredKernels(const ImplParams &p = ImplParams());
-
-};
-    
-}}
-
-#endif

Grid/qcd/action/fermion/StaggeredKernelsAsm.cc

@@ -1,968 +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>
-
-#ifdef AVX512
-#include <Grid/simd/Intel512common.h>
-#include <Grid/simd/Intel512avx.h>
-#endif
-
-// Interleave operations from two directions
-// This looks just like a 2 spin multiply and reuse same sequence from the Wilson
-// Kernel. But the spin index becomes a mu index instead.
-#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 UChi_00 %zmm12
-#define UChi_01 %zmm13
-#define UChi_02 %zmm14
-#define UChi_10 %zmm15
-#define UChi_11 %zmm16
-#define UChi_12 %zmm17
-#define UChi_20 %zmm18
-#define UChi_21 %zmm19
-#define UChi_22 %zmm20
-#define UChi_30 %zmm21
-#define UChi_31 %zmm22
-#define UChi_32 %zmm23
-
-#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 pUChi_00 %%zmm12
-#define pUChi_01 %%zmm13
-#define pUChi_02 %%zmm14
-#define pUChi_10 %%zmm15
-#define pUChi_11 %%zmm16
-#define pUChi_12 %%zmm17
-#define pUChi_20 %%zmm18
-#define pUChi_21 %%zmm19
-#define pUChi_22 %%zmm20
-#define pUChi_30 %%zmm21
-#define pUChi_31 %%zmm22
-#define pUChi_32 %%zmm23
-
-#define T0 %zmm24
-#define T1 %zmm25
-#define T2 %zmm26
-#define T3 %zmm27
-
-#define Z00 %zmm26
-#define Z10 %zmm27
-#define Z0 Z00
-#define Z1 %zmm28
-#define Z2 %zmm29
-
-#define Z3 %zmm30
-#define Z4 %zmm31
-#define Z5 Chi_31
-#define Z6 Chi_32
-
-#define MULT_ADD_LS(g0,g1,g2,g3)					\
-  asm ( "movq %0, %%r8 \n\t"					\
-	"movq %1, %%r9 \n\t"						\
-        "movq %2, %%r10 \n\t"						\
-        "movq %3, %%r11 \n\t" :  : "r"(g0), "r"(g1), "r"(g2), "r"(g3) : "%r8","%r9","%r10","%r11" );\
-  asm (									\
-  VSHUF(Chi_00,T0)      VSHUF(Chi_10,T1)				\
-  VSHUF(Chi_20,T2)      VSHUF(Chi_30,T3)				\
-  VMADDSUBIDUP(0,%r8,T0,UChi_00) VMADDSUBIDUP(0,%r9,T1,UChi_10)		\
-  VMADDSUBIDUP(3,%r8,T0,UChi_01) VMADDSUBIDUP(3,%r9,T1,UChi_11)		\
-  VMADDSUBIDUP(6,%r8,T0,UChi_02) VMADDSUBIDUP(6,%r9,T1,UChi_12)		\
-  VMADDSUBIDUP(0,%r10,T2,UChi_20) VMADDSUBIDUP(0,%r11,T3,UChi_30)		\
-  VMADDSUBIDUP(3,%r10,T2,UChi_21) VMADDSUBIDUP(3,%r11,T3,UChi_31)		\
-  VMADDSUBIDUP(6,%r10,T2,UChi_22) VMADDSUBIDUP(6,%r11,T3,UChi_32)		\
-  VMADDSUBRDUP(0,%r8,Chi_00,UChi_00) VMADDSUBRDUP(0,%r9,Chi_10,UChi_10) \
-  VMADDSUBRDUP(3,%r8,Chi_00,UChi_01) VMADDSUBRDUP(3,%r9,Chi_10,UChi_11) \
-  VMADDSUBRDUP(6,%r8,Chi_00,UChi_02) VMADDSUBRDUP(6,%r9,Chi_10,UChi_12) \
-  VMADDSUBRDUP(0,%r10,Chi_20,UChi_20) VMADDSUBRDUP(0,%r11,Chi_30,UChi_30) \
-  VMADDSUBRDUP(3,%r10,Chi_20,UChi_21) VMADDSUBRDUP(3,%r11,Chi_30,UChi_31) \
-  VMADDSUBRDUP(6,%r10,Chi_20,UChi_22) VMADDSUBRDUP(6,%r11,Chi_30,UChi_32) \
-  VSHUF(Chi_01,T0)	  VSHUF(Chi_11,T1)				\
-  VSHUF(Chi_21,T2)	  VSHUF(Chi_31,T3)				\
-  VMADDSUBIDUP(1,%r8,T0,UChi_00)     VMADDSUBIDUP(1,%r9,T1,UChi_10)	\
-  VMADDSUBIDUP(4,%r8,T0,UChi_01)     VMADDSUBIDUP(4,%r9,T1,UChi_11)	\
-  VMADDSUBIDUP(7,%r8,T0,UChi_02)     VMADDSUBIDUP(7,%r9,T1,UChi_12)	\
-  VMADDSUBIDUP(1,%r10,T2,UChi_20)     VMADDSUBIDUP(1,%r11,T3,UChi_30)	\
-  VMADDSUBIDUP(4,%r10,T2,UChi_21)     VMADDSUBIDUP(4,%r11,T3,UChi_31)	\
-  VMADDSUBIDUP(7,%r10,T2,UChi_22)     VMADDSUBIDUP(7,%r11,T3,UChi_32)	\
-  VMADDSUBRDUP(1,%r8,Chi_01,UChi_00) VMADDSUBRDUP(1,%r9,Chi_11,UChi_10) \
-  VMADDSUBRDUP(4,%r8,Chi_01,UChi_01) VMADDSUBRDUP(4,%r9,Chi_11,UChi_11) \
-  VMADDSUBRDUP(7,%r8,Chi_01,UChi_02) VMADDSUBRDUP(7,%r9,Chi_11,UChi_12) \
-  VMADDSUBRDUP(1,%r10,Chi_21,UChi_20) VMADDSUBRDUP(1,%r11,Chi_31,UChi_30) \
-  VMADDSUBRDUP(4,%r10,Chi_21,UChi_21) VMADDSUBRDUP(4,%r11,Chi_31,UChi_31) \
-  VMADDSUBRDUP(7,%r10,Chi_21,UChi_22) VMADDSUBRDUP(7,%r11,Chi_31,UChi_32) \
-  VSHUF(Chi_02,T0)    VSHUF(Chi_12,T1)					\
-  VSHUF(Chi_22,T2)    VSHUF(Chi_32,T3)					\
-  VMADDSUBIDUP(2,%r8,T0,UChi_00)     VMADDSUBIDUP(2,%r9,T1,UChi_10)     \
-  VMADDSUBIDUP(5,%r8,T0,UChi_01)     VMADDSUBIDUP(5,%r9,T1,UChi_11)     \
-  VMADDSUBIDUP(8,%r8,T0,UChi_02)     VMADDSUBIDUP(8,%r9,T1,UChi_12)     \
-  VMADDSUBIDUP(2,%r10,T2,UChi_20)     VMADDSUBIDUP(2,%r11,T3,UChi_30)     \
-  VMADDSUBIDUP(5,%r10,T2,UChi_21)     VMADDSUBIDUP(5,%r11,T3,UChi_31)     \
-  VMADDSUBIDUP(8,%r10,T2,UChi_22)     VMADDSUBIDUP(8,%r11,T3,UChi_32)     \
-  VMADDSUBRDUP(2,%r8,Chi_02,UChi_00) VMADDSUBRDUP(2,%r9,Chi_12,UChi_10) \
-  VMADDSUBRDUP(5,%r8,Chi_02,UChi_01) VMADDSUBRDUP(5,%r9,Chi_12,UChi_11) \
-  VMADDSUBRDUP(8,%r8,Chi_02,UChi_02) VMADDSUBRDUP(8,%r9,Chi_12,UChi_12) \
-  VMADDSUBRDUP(2,%r10,Chi_22,UChi_20) VMADDSUBRDUP(2,%r11,Chi_32,UChi_30) \
-  VMADDSUBRDUP(5,%r10,Chi_22,UChi_21) VMADDSUBRDUP(5,%r11,Chi_32,UChi_31) \
-  VMADDSUBRDUP(8,%r10,Chi_22,UChi_22) VMADDSUBRDUP(8,%r11,Chi_32,UChi_32) );
-
-#define MULT_LS(g0,g1,g2,g3)					\
-  asm ( "movq %0, %%r8 \n\t"					\
-	"movq %1, %%r9 \n\t"						\
-        "movq %2, %%r10 \n\t"						\
-        "movq %3, %%r11 \n\t" :  : "r"(g0), "r"(g1), "r"(g2), "r"(g3) : "%r8","%r9","%r10","%r11" );\
-  asm (									\
-  VSHUF(Chi_00,T0)      VSHUF(Chi_10,T1)				\
-  VSHUF(Chi_20,T2)      VSHUF(Chi_30,T3)				\
-  VMULIDUP(0,%r8,T0,UChi_00) VMULIDUP(0,%r9,T1,UChi_10)		\
-  VMULIDUP(3,%r8,T0,UChi_01) VMULIDUP(3,%r9,T1,UChi_11)		\
-  VMULIDUP(6,%r8,T0,UChi_02) VMULIDUP(6,%r9,T1,UChi_12)		\
-  VMULIDUP(0,%r10,T2,UChi_20) VMULIDUP(0,%r11,T3,UChi_30)		\
-  VMULIDUP(3,%r10,T2,UChi_21) VMULIDUP(3,%r11,T3,UChi_31)		\
-  VMULIDUP(6,%r10,T2,UChi_22) VMULIDUP(6,%r11,T3,UChi_32)		\
-  VMADDSUBRDUP(0,%r8,Chi_00,UChi_00) VMADDSUBRDUP(0,%r9,Chi_10,UChi_10) \
-  VMADDSUBRDUP(3,%r8,Chi_00,UChi_01) VMADDSUBRDUP(3,%r9,Chi_10,UChi_11) \
-  VMADDSUBRDUP(6,%r8,Chi_00,UChi_02) VMADDSUBRDUP(6,%r9,Chi_10,UChi_12) \
-  VMADDSUBRDUP(0,%r10,Chi_20,UChi_20) VMADDSUBRDUP(0,%r11,Chi_30,UChi_30) \
-  VMADDSUBRDUP(3,%r10,Chi_20,UChi_21) VMADDSUBRDUP(3,%r11,Chi_30,UChi_31) \
-  VMADDSUBRDUP(6,%r10,Chi_20,UChi_22) VMADDSUBRDUP(6,%r11,Chi_30,UChi_32) \
-  VSHUF(Chi_01,T0)	  VSHUF(Chi_11,T1)				\
-  VSHUF(Chi_21,T2)	  VSHUF(Chi_31,T3)				\
-  VMADDSUBIDUP(1,%r8,T0,UChi_00)     VMADDSUBIDUP(1,%r9,T1,UChi_10)	\
-  VMADDSUBIDUP(4,%r8,T0,UChi_01)     VMADDSUBIDUP(4,%r9,T1,UChi_11)	\
-  VMADDSUBIDUP(7,%r8,T0,UChi_02)     VMADDSUBIDUP(7,%r9,T1,UChi_12)	\
-  VMADDSUBIDUP(1,%r10,T2,UChi_20)     VMADDSUBIDUP(1,%r11,T3,UChi_30)	\
-  VMADDSUBIDUP(4,%r10,T2,UChi_21)     VMADDSUBIDUP(4,%r11,T3,UChi_31)	\
-  VMADDSUBIDUP(7,%r10,T2,UChi_22)     VMADDSUBIDUP(7,%r11,T3,UChi_32)	\
-  VMADDSUBRDUP(1,%r8,Chi_01,UChi_00) VMADDSUBRDUP(1,%r9,Chi_11,UChi_10) \
-  VMADDSUBRDUP(4,%r8,Chi_01,UChi_01) VMADDSUBRDUP(4,%r9,Chi_11,UChi_11) \
-  VMADDSUBRDUP(7,%r8,Chi_01,UChi_02) VMADDSUBRDUP(7,%r9,Chi_11,UChi_12) \
-  VMADDSUBRDUP(1,%r10,Chi_21,UChi_20) VMADDSUBRDUP(1,%r11,Chi_31,UChi_30) \
-  VMADDSUBRDUP(4,%r10,Chi_21,UChi_21) VMADDSUBRDUP(4,%r11,Chi_31,UChi_31) \
-  VMADDSUBRDUP(7,%r10,Chi_21,UChi_22) VMADDSUBRDUP(7,%r11,Chi_31,UChi_32) \
-  VSHUF(Chi_02,T0)    VSHUF(Chi_12,T1)					\
-  VSHUF(Chi_22,T2)    VSHUF(Chi_32,T3)					\
-  VMADDSUBIDUP(2,%r8,T0,UChi_00)     VMADDSUBIDUP(2,%r9,T1,UChi_10)     \
-  VMADDSUBIDUP(5,%r8,T0,UChi_01)     VMADDSUBIDUP(5,%r9,T1,UChi_11)     \
-  VMADDSUBIDUP(8,%r8,T0,UChi_02)     VMADDSUBIDUP(8,%r9,T1,UChi_12)     \
-  VMADDSUBIDUP(2,%r10,T2,UChi_20)     VMADDSUBIDUP(2,%r11,T3,UChi_30)     \
-  VMADDSUBIDUP(5,%r10,T2,UChi_21)     VMADDSUBIDUP(5,%r11,T3,UChi_31)     \
-  VMADDSUBIDUP(8,%r10,T2,UChi_22)     VMADDSUBIDUP(8,%r11,T3,UChi_32)     \
-  VMADDSUBRDUP(2,%r8,Chi_02,UChi_00) VMADDSUBRDUP(2,%r9,Chi_12,UChi_10) \
-  VMADDSUBRDUP(5,%r8,Chi_02,UChi_01) VMADDSUBRDUP(5,%r9,Chi_12,UChi_11) \
-  VMADDSUBRDUP(8,%r8,Chi_02,UChi_02) VMADDSUBRDUP(8,%r9,Chi_12,UChi_12) \
-  VMADDSUBRDUP(2,%r10,Chi_22,UChi_20) VMADDSUBRDUP(2,%r11,Chi_32,UChi_30) \
-  VMADDSUBRDUP(5,%r10,Chi_22,UChi_21) VMADDSUBRDUP(5,%r11,Chi_32,UChi_31) \
-  VMADDSUBRDUP(8,%r10,Chi_22,UChi_22) VMADDSUBRDUP(8,%r11,Chi_32,UChi_32) );
-
-#define MULT_ADD_XYZTa(g0,g1)					\
-  asm ( "movq %0, %%r8 \n\t"					\
-	"movq %1, %%r9 \n\t"	 :  : "r"(g0), "r"(g1) : "%r8","%r9");\
-	   __asm__ (						\
-	   VSHUF(Chi_00,T0)				\
-	   VSHUF(Chi_10,T1)						\
-	   VMOVIDUP(0,%r8,Z0 )						\
-           VMOVIDUP(3,%r8,Z1 )						\
-           VMOVIDUP(6,%r8,Z2 )						\
-           VMADDSUB(Z0,T0,UChi_00)					\
-	   VMADDSUB(Z1,T0,UChi_01)					\
-	   VMADDSUB(Z2,T0,UChi_02)					\
-									\
-	   VMOVIDUP(0,%r9,Z0 )						\
-           VMOVIDUP(3,%r9,Z1 )						\
-           VMOVIDUP(6,%r9,Z2 )						\
-           VMADDSUB(Z0,T1,UChi_10)					\
-           VMADDSUB(Z1,T1,UChi_11)            \
-           VMADDSUB(Z2,T1,UChi_12)            \
-	   							\
-								\
-	   VMOVRDUP(0,%r8,Z3 )					\
-	   VMOVRDUP(3,%r8,Z4 )					\
-	   VMOVRDUP(6,%r8,Z5 )					\
-           VMADDSUB(Z3,Chi_00,UChi_00)/*rr * ir = ri rr*/	\
-           VMADDSUB(Z4,Chi_00,UChi_01)				\
-           VMADDSUB(Z5,Chi_00,UChi_02)				\
-								\
-	   VMOVRDUP(0,%r9,Z3 )					\
-	   VMOVRDUP(3,%r9,Z4 )					\
-	   VMOVRDUP(6,%r9,Z5 )					\
-           VMADDSUB(Z3,Chi_10,UChi_10)				\
-           VMADDSUB(Z4,Chi_10,UChi_11)\
-           VMADDSUB(Z5,Chi_10,UChi_12)				\
-	   							\
-								\
-	   VMOVIDUP(1,%r8,Z0 )					\
-	   VMOVIDUP(4,%r8,Z1 )					\
-	   VMOVIDUP(7,%r8,Z2 )					\
-	   VSHUF(Chi_01,T0)					\
-           VMADDSUB(Z0,T0,UChi_00)				\
-           VMADDSUB(Z1,T0,UChi_01)				\
-           VMADDSUB(Z2,T0,UChi_02)				\
-								\
-	   VMOVIDUP(1,%r9,Z0 )					\
-	   VMOVIDUP(4,%r9,Z1 )					\
-	   VMOVIDUP(7,%r9,Z2 )					\
-	   VSHUF(Chi_11,T1)					\
-           VMADDSUB(Z0,T1,UChi_10)				\
-           VMADDSUB(Z1,T1,UChi_11)				\
-           VMADDSUB(Z2,T1,UChi_12)				\
-								\
-	   VMOVRDUP(1,%r8,Z3 )					\
-	   VMOVRDUP(4,%r8,Z4 )					\
-	   VMOVRDUP(7,%r8,Z5 )					\
-           VMADDSUB(Z3,Chi_01,UChi_00)				\
-           VMADDSUB(Z4,Chi_01,UChi_01)				\
-           VMADDSUB(Z5,Chi_01,UChi_02)				\
-								\
-	   VMOVRDUP(1,%r9,Z3 )					\
-	   VMOVRDUP(4,%r9,Z4 )					\
-	   VMOVRDUP(7,%r9,Z5 )					\
-           VMADDSUB(Z3,Chi_11,UChi_10)				\
-           VMADDSUB(Z4,Chi_11,UChi_11)				\
-           VMADDSUB(Z5,Chi_11,UChi_12)				\
-	   							\
-	   VSHUF(Chi_02,T0)					\
-	   VSHUF(Chi_12,T1)					\
-	   VMOVIDUP(2,%r8,Z0 )					\
-	   VMOVIDUP(5,%r8,Z1 )					\
-	   VMOVIDUP(8,%r8,Z2 )					\
-           VMADDSUB(Z0,T0,UChi_00)				\
-           VMADDSUB(Z1,T0,UChi_01)			      \
-           VMADDSUB(Z2,T0,UChi_02)			      \
-	   VMOVIDUP(2,%r9,Z0 )					\
-	   VMOVIDUP(5,%r9,Z1 )					\
-	   VMOVIDUP(8,%r9,Z2 )					\
-           VMADDSUB(Z0,T1,UChi_10)			      \
-           VMADDSUB(Z1,T1,UChi_11)			      \
-           VMADDSUB(Z2,T1,UChi_12)			      \
-	   /*55*/					      \
-	   VMOVRDUP(2,%r8,Z3 )		  \
-	   VMOVRDUP(5,%r8,Z4 )					\
-	   VMOVRDUP(8,%r8,Z5 )				      \
-           VMADDSUB(Z3,Chi_02,UChi_00)			      \
-           VMADDSUB(Z4,Chi_02,UChi_01)			      \
-           VMADDSUB(Z5,Chi_02,UChi_02)			      \
-	   VMOVRDUP(2,%r9,Z3 )		  \
-	   VMOVRDUP(5,%r9,Z4 )					\
-	   VMOVRDUP(8,%r9,Z5 )				      \
-           VMADDSUB(Z3,Chi_12,UChi_10)			      \
-           VMADDSUB(Z4,Chi_12,UChi_11)			      \
-           VMADDSUB(Z5,Chi_12,UChi_12)			      \
-	   /*61 insns*/							);
-
-#define MULT_ADD_XYZT(g0,g1)					\
-  asm ( "movq %0, %%r8 \n\t"					\
-	"movq %1, %%r9 \n\t"	 :  : "r"(g0), "r"(g1) : "%r8","%r9");\
-  __asm__ (							  \
-  VSHUFMEM(0,%r8,Z00)		   VSHUFMEM(0,%r9,Z10)			\
-  VRDUP(Chi_00,T0)           VIDUP(Chi_00,Chi_00)	          \
-   VRDUP(Chi_10,T1)           VIDUP(Chi_10,Chi_10)		  \
-   VMUL(Z00,Chi_00,Z1)        VMUL(Z10,Chi_10,Z2)		  \
-   VSHUFMEM(3,%r8,Z00)	      VSHUFMEM(3,%r9,Z10)		  \
-   VMUL(Z00,Chi_00,Z3)        VMUL(Z10,Chi_10,Z4)		  \
-   VSHUFMEM(6,%r8,Z00)	      VSHUFMEM(6,%r9,Z10)		  \
-   VMUL(Z00,Chi_00,Z5)        VMUL(Z10,Chi_10,Z6)		  \
-   VMADDMEM(0,%r8,T0,UChi_00)  VMADDMEM(0,%r9,T1,UChi_10)		  \
-   VMADDMEM(3,%r8,T0,UChi_01)  VMADDMEM(3,%r9,T1,UChi_11)		  \
-   VMADDMEM(6,%r8,T0,UChi_02)  VMADDMEM(6,%r9,T1,UChi_12)		  \
-   VSHUFMEM(1,%r8,Z00)	      VSHUFMEM(1,%r9,Z10)		  \
-   VRDUP(Chi_01,T0)           VIDUP(Chi_01,Chi_01)		  \
-   VRDUP(Chi_11,T1)           VIDUP(Chi_11,Chi_11)		  \
-   VMADD(Z00,Chi_01,Z1)       VMADD(Z10,Chi_11,Z2)		  \
-   VSHUFMEM(4,%r8,Z00)	      VSHUFMEM(4,%r9,Z10)		  \
-   VMADD(Z00,Chi_01,Z3)       VMADD(Z10,Chi_11,Z4)		  \
-   VSHUFMEM(7,%r8,Z00)	      VSHUFMEM(7,%r9,Z10)		  \
-   VMADD(Z00,Chi_01,Z5)       VMADD(Z10,Chi_11,Z6)		  \
-   VMADDMEM(1,%r8,T0,UChi_00) VMADDMEM(1,%r9,T1,UChi_10)	  \
-   VMADDMEM(4,%r8,T0,UChi_01) VMADDMEM(4,%r9,T1,UChi_11)	  \
-   VMADDMEM(7,%r8,T0,UChi_02) VMADDMEM(7,%r9,T1,UChi_12)	  \
-   VSHUFMEM(2,%r8,Z00)	      VSHUFMEM(2,%r9,Z10)			\
-   VRDUP(Chi_02,T0)           VIDUP(Chi_02,Chi_02)			\
-   VRDUP(Chi_12,T1)           VIDUP(Chi_12,Chi_12)			\
-   VMADD(Z00,Chi_02,Z1)       VMADD(Z10,Chi_12,Z2)		  \
-   VSHUFMEM(5,%r8,Z00)	      VSHUFMEM(5,%r9,Z10)		  \
-   VMADD(Z00,Chi_02,Z3)       VMADD(Z10,Chi_12,Z4)		  \
-   VSHUFMEM(8,%r8,Z00)	      VSHUFMEM(8,%r9,Z10)		  \
-   VMADD(Z00,Chi_02,Z5)       VMADD(Z10,Chi_12,Z6)		  \
-   VMADDSUBMEM(2,%r8,T0,Z1)   VMADDSUBMEM(2,%r9,T1,Z2)		  \
-   VMADDSUBMEM(5,%r8,T0,Z3)   VMADDSUBMEM(5,%r9,T1,Z4)	          \
-   VMADDSUBMEM(8,%r8,T0,Z5)   VMADDSUBMEM(8,%r9,T1,Z6)	       \
-   VADD(Z1,UChi_00,UChi_00)   VADD(Z2,UChi_10,UChi_10)	       \
-   VADD(Z3,UChi_01,UChi_01)   VADD(Z4,UChi_11,UChi_11)	       \
-   VADD(Z5,UChi_02,UChi_02)   VADD(Z6,UChi_12,UChi_12) );
-
-#define MULT_XYZT(g0,g1)					\
-    asm ( "movq %0, %%r8 \n\t"						\
-	"movq %1, %%r9 \n\t" :  : "r"(g0), "r"(g1) : "%r8","%r9" ); \
-	   __asm__ (						\
-	   VSHUF(Chi_00,T0)				\
-	   VSHUF(Chi_10,T1)						\
-	   VMOVIDUP(0,%r8,Z0 )						\
-           VMOVIDUP(3,%r8,Z1 )						\
-           VMOVIDUP(6,%r8,Z2 )						\
-	   /*6*/							\
-           VMUL(Z0,T0,UChi_00)            \
-           VMUL(Z1,T0,UChi_01)            \
-           VMUL(Z2,T0,UChi_02)            \
-	   VMOVIDUP(0,%r9,Z0 )						\
-           VMOVIDUP(3,%r9,Z1 )						\
-           VMOVIDUP(6,%r9,Z2 )						\
-           VMUL(Z0,T1,UChi_10)            \
-           VMUL(Z1,T1,UChi_11)            \
-           VMUL(Z2,T1,UChi_12)            \
-	   VMOVRDUP(0,%r8,Z3 )					\
-	   VMOVRDUP(3,%r8,Z4 )					\
-	   VMOVRDUP(6,%r8,Z5 )					\
-	   /*18*/						\
-           VMADDSUB(Z3,Chi_00,UChi_00)				\
-           VMADDSUB(Z4,Chi_00,UChi_01)\
-           VMADDSUB(Z5,Chi_00,UChi_02) \
-	   VMOVRDUP(0,%r9,Z3 )					\
-	   VMOVRDUP(3,%r9,Z4 )					\
-	   VMOVRDUP(6,%r9,Z5 )					\
-           VMADDSUB(Z3,Chi_10,UChi_10)				\
-           VMADDSUB(Z4,Chi_10,UChi_11)\
-           VMADDSUB(Z5,Chi_10,UChi_12)				\
-	   VMOVIDUP(1,%r8,Z0 )					\
-	   VMOVIDUP(4,%r8,Z1 )					\
-	   VMOVIDUP(7,%r8,Z2 )					\
-	   /*28*/						\
-	   VSHUF(Chi_01,T0)					\
-           VMADDSUB(Z0,T0,UChi_00)      \
-           VMADDSUB(Z1,T0,UChi_01)       \
-           VMADDSUB(Z2,T0,UChi_02)        \
-	   VMOVIDUP(1,%r9,Z0 )					\
-	   VMOVIDUP(4,%r9,Z1 )					\
-	   VMOVIDUP(7,%r9,Z2 )					\
-	   VSHUF(Chi_11,T1)					\
-           VMADDSUB(Z0,T1,UChi_10)				\
-           VMADDSUB(Z1,T1,UChi_11)				\
-           VMADDSUB(Z2,T1,UChi_12)        \
-	   VMOVRDUP(1,%r8,Z3 )					\
-	   VMOVRDUP(4,%r8,Z4 )					\
-	   VMOVRDUP(7,%r8,Z5 )					\
-           /*38*/						\
-           VMADDSUB(Z3,Chi_01,UChi_00)    \
-           VMADDSUB(Z4,Chi_01,UChi_01)    \
-           VMADDSUB(Z5,Chi_01,UChi_02)    \
-	   VMOVRDUP(1,%r9,Z3 )					\
-	   VMOVRDUP(4,%r9,Z4 )					\
-	   VMOVRDUP(7,%r9,Z5 )					\
-           VMADDSUB(Z3,Chi_11,UChi_10)				\
-           VMADDSUB(Z4,Chi_11,UChi_11)    \
-           VMADDSUB(Z5,Chi_11,UChi_12)				\
-	   /*48*/						\
-	   VSHUF(Chi_02,T0)					\
-	   VSHUF(Chi_12,T1)					\
-	   VMOVIDUP(2,%r8,Z0 )					\
-	   VMOVIDUP(5,%r8,Z1 )					\
-	   VMOVIDUP(8,%r8,Z2 )					\
-           VMADDSUB(Z0,T0,UChi_00)				\
-           VMADDSUB(Z1,T0,UChi_01)			      \
-           VMADDSUB(Z2,T0,UChi_02)			      \
-	   VMOVIDUP(2,%r9,Z0 )					\
-	   VMOVIDUP(5,%r9,Z1 )					\
-	   VMOVIDUP(8,%r9,Z2 )					\
-           VMADDSUB(Z0,T1,UChi_10)			      \
-           VMADDSUB(Z1,T1,UChi_11)			      \
-           VMADDSUB(Z2,T1,UChi_12)			      \
-	   /*55*/					      \
-	   VMOVRDUP(2,%r8,Z3 )		  \
-	   VMOVRDUP(5,%r8,Z4 )					\
-	   VMOVRDUP(8,%r8,Z5 )				      \
-           VMADDSUB(Z3,Chi_02,UChi_00)			      \
-           VMADDSUB(Z4,Chi_02,UChi_01)			      \
-           VMADDSUB(Z5,Chi_02,UChi_02)			      \
-	   VMOVRDUP(2,%r9,Z3 )		  \
-	   VMOVRDUP(5,%r9,Z4 )					\
-	   VMOVRDUP(8,%r9,Z5 )				      \
-           VMADDSUB(Z3,Chi_12,UChi_10)			      \
-           VMADDSUB(Z4,Chi_12,UChi_11)			      \
-           VMADDSUB(Z5,Chi_12,UChi_12)			      \
-	   /*61 insns*/							);
-
-#define MULT_XYZTa(g0,g1)					\
-  asm ( "movq %0, %%r8 \n\t"					\
-	"movq %1, %%r9 \n\t" :  : "r"(g0), "r"(g1) : "%r8","%r9" ); \
-  __asm__ (							  \
-   VSHUFMEM(0,%r8,Z00)		   VSHUFMEM(0,%r9,Z10)	  \
-   VRDUP(Chi_00,T0)           VIDUP(Chi_00,Chi_00)	          \
-   VRDUP(Chi_10,T1)           VIDUP(Chi_10,Chi_10)		  \
-   VMUL(Z00,Chi_00,Z1)        VMUL(Z10,Chi_10,Z2)		  \
-   VSHUFMEM(3,%r8,Z00)	      VSHUFMEM(3,%r9,Z10)		  \
-   VMUL(Z00,Chi_00,Z3)        VMUL(Z10,Chi_10,Z4)		  \
-   VSHUFMEM(6,%r8,Z00)	      VSHUFMEM(6,%r9,Z10)		  \
-   VMUL(Z00,Chi_00,Z5)        VMUL(Z10,Chi_10,Z6)		  \
-   VMULMEM(0,%r8,T0,UChi_00)  VMULMEM(0,%r9,T1,UChi_10)		  \
-   VMULMEM(3,%r8,T0,UChi_01)  VMULMEM(3,%r9,T1,UChi_11)		  \
-   VMULMEM(6,%r8,T0,UChi_02)  VMULMEM(6,%r9,T1,UChi_12)		  \
-   VSHUFMEM(1,%r8,Z00)	      VSHUFMEM(1,%r9,Z10)		  \
-   VRDUP(Chi_01,T0)           VIDUP(Chi_01,Chi_01)		  \
-   VRDUP(Chi_11,T1)           VIDUP(Chi_11,Chi_11)		  \
-   VMADD(Z00,Chi_01,Z1)       VMADD(Z10,Chi_11,Z2)		  \
-   VSHUFMEM(4,%r8,Z00)	      VSHUFMEM(4,%r9,Z10)		  \
-   VMADD(Z00,Chi_01,Z3)       VMADD(Z10,Chi_11,Z4)		  \
-   VSHUFMEM(7,%r8,Z00)	      VSHUFMEM(7,%r9,Z10)		  \
-   VMADD(Z00,Chi_01,Z5)       VMADD(Z10,Chi_11,Z6)		  \
-   VMADDMEM(1,%r8,T0,UChi_00) VMADDMEM(1,%r9,T1,UChi_10)	  \
-   VMADDMEM(4,%r8,T0,UChi_01) VMADDMEM(4,%r9,T1,UChi_11)	  \
-   VMADDMEM(7,%r8,T0,UChi_02) VMADDMEM(7,%r9,T1,UChi_12)	  \
-   VSHUFMEM(2,%r8,Z00)	      VSHUFMEM(2,%r9,Z10)			\
-   VRDUP(Chi_02,T0)           VIDUP(Chi_02,Chi_02)			\
-   VRDUP(Chi_12,T1)           VIDUP(Chi_12,Chi_12)			\
-   VMADD(Z00,Chi_02,Z1)       VMADD(Z10,Chi_12,Z2)		  \
-   VSHUFMEM(5,%r8,Z00)	      VSHUFMEM(5,%r9,Z10)		  \
-   VMADD(Z00,Chi_02,Z3)       VMADD(Z10,Chi_12,Z4)		  \
-   VSHUFMEM(8,%r8,Z00)	      VSHUFMEM(8,%r9,Z10)		  \
-   VMADD(Z00,Chi_02,Z5)       VMADD(Z10,Chi_12,Z6)		  \
-   VMADDSUBMEM(2,%r8,T0,Z1)   VMADDSUBMEM(2,%r9,T1,Z2)		  \
-   VMADDSUBMEM(5,%r8,T0,Z3)   VMADDSUBMEM(5,%r9,T1,Z4)	          \
-   VMADDSUBMEM(8,%r8,T0,Z5)   VMADDSUBMEM(8,%r9,T1,Z6)	       \
-   VADD(Z1,UChi_00,UChi_00)   VADD(Z2,UChi_10,UChi_10)	       \
-   VADD(Z3,UChi_01,UChi_01)   VADD(Z4,UChi_11,UChi_11)	       \
-   VADD(Z5,UChi_02,UChi_02)   VADD(Z6,UChi_12,UChi_12) );
-
-
-#define LOAD_CHI(a0,a1,a2,a3)						\
-  asm (									\
-       "movq %0, %%r8 \n\t"						\
-       VLOAD(0,%%r8,pChi_00)						\
-       VLOAD(1,%%r8,pChi_01)						\
-       VLOAD(2,%%r8,pChi_02)						\
-       : : "r" (a0) : "%r8" );						\
-  asm (									\
-       "movq %0, %%r8 \n\t"						\
-       VLOAD(0,%%r8,pChi_10)						\
-       VLOAD(1,%%r8,pChi_11)						\
-       VLOAD(2,%%r8,pChi_12)						\
-       : : "r" (a1) : "%r8" );						\
-  asm (									\
-       "movq %0, %%r8 \n\t"						\
-       VLOAD(0,%%r8,pChi_20)						\
-       VLOAD(1,%%r8,pChi_21)						\
-       VLOAD(2,%%r8,pChi_22)						\
-       : : "r" (a2) : "%r8" );						\
-  asm (									\
-       "movq %0, %%r8 \n\t"						\
-       VLOAD(0,%%r8,pChi_30)						\
-       VLOAD(1,%%r8,pChi_31)						\
-       VLOAD(2,%%r8,pChi_32)						\
-       : : "r" (a3) : "%r8" );						
-
-#define LOAD_CHIa(a0,a1)						\
-  asm (									\
-       "movq %0, %%r8 \n\t"						\
-       VLOAD(0,%%r8,pChi_00)						\
-       VLOAD(1,%%r8,pChi_01)						\
-       VLOAD(2,%%r8,pChi_02)						\
-       : : "r" (a0) : "%r8" );						\
-  asm (									\
-       "movq %0, %%r8 \n\t"						\
-       VLOAD(0,%%r8,pChi_10)						\
-       VLOAD(1,%%r8,pChi_11)						\
-       VLOAD(2,%%r8,pChi_12)						\
-       : : "r" (a1) : "%r8" );						
-
-#define PF_CHI(a0)							
-#define PF_CHIa(a0)							\
-  asm (									\
-       "movq %0, %%r8 \n\t"						\
-       VPREFETCH1(0,%%r8)						\
-       VPREFETCH1(1,%%r8)						\
-       VPREFETCH1(2,%%r8)						\
-       : : "r" (a0) : "%r8" );						\
-
-#define PF_GAUGE_XYZT(a0)							
-#define PF_GAUGE_XYZTa(a0)						\
-  asm (									\
-       "movq %0, %%r8 \n\t"						\
-       VPREFETCH1(0,%%r8)						\
-       VPREFETCH1(1,%%r8)						\
-       VPREFETCH1(2,%%r8)						\
-       VPREFETCH1(3,%%r8)						\
-       VPREFETCH1(4,%%r8)						\
-       VPREFETCH1(5,%%r8)						\
-       VPREFETCH1(6,%%r8)						\
-       VPREFETCH1(7,%%r8)						\
-       VPREFETCH1(8,%%r8)						\
-       : : "r" (a0) : "%r8" );						\
-
-#define PF_GAUGE_LS(a0)							
-#define PF_GAUGE_LSa(a0)							\
-  asm (									\
-       "movq %0, %%r8 \n\t"						\
-       VPREFETCH1(0,%%r8)						\
-       VPREFETCH1(1,%%r8)						\
-       : : "r" (a0) : "%r8" );						\
-  
-
-#define REDUCE(out)					\
-  asm (							\
-  VADD(UChi_00,UChi_10,UChi_00)				\
-  VADD(UChi_01,UChi_11,UChi_01)				\
-  VADD(UChi_02,UChi_12,UChi_02)				\
-  VADD(UChi_30,UChi_20,UChi_30)				\
-  VADD(UChi_31,UChi_21,UChi_31)				\
-  VADD(UChi_32,UChi_22,UChi_32)				\
-  VADD(UChi_00,UChi_30,UChi_00)				\
-  VADD(UChi_01,UChi_31,UChi_01)				\
-  VADD(UChi_02,UChi_32,UChi_02)				);	\
-  asm (								\
-       VSTORE(0,%0,pUChi_00)					\
-       VSTORE(1,%0,pUChi_01)					\
-       VSTORE(2,%0,pUChi_02)					\
-       : : "r" (out) : "memory" );
-
-#define nREDUCE(out)							\
-  asm (									\
-       VADD(UChi_00,UChi_10,UChi_00)					\
-       VADD(UChi_01,UChi_11,UChi_01)					\
-       VADD(UChi_02,UChi_12,UChi_02)					\
-       VADD(UChi_30,UChi_20,UChi_30)					\
-       VADD(UChi_31,UChi_21,UChi_31)					\
-       VADD(UChi_32,UChi_22,UChi_32)					\
-       VADD(UChi_00,UChi_30,UChi_00)					\
-       VADD(UChi_01,UChi_31,UChi_01)					\
-       VADD(UChi_02,UChi_32,UChi_02)				);	\
-  asm (VZERO(Chi_00)							\
-       VSUB(UChi_00,Chi_00,UChi_00)					\
-       VSUB(UChi_01,Chi_00,UChi_01)					\
-       VSUB(UChi_02,Chi_00,UChi_02)				);	\
-  asm (								\
-       VSTORE(0,%0,pUChi_00)					\
-       VSTORE(1,%0,pUChi_01)					\
-       VSTORE(2,%0,pUChi_02)					\
-       : : "r" (out) : "memory" );
-
-#define REDUCEa(out)					\
-  asm (							\
-  VADD(UChi_00,UChi_10,UChi_00)				\
-  VADD(UChi_01,UChi_11,UChi_01)				\
-  VADD(UChi_02,UChi_12,UChi_02)	);			\
-  asm (									\
-       VSTORE(0,%0,pUChi_00)						\
-       VSTORE(1,%0,pUChi_01)						\
-       VSTORE(2,%0,pUChi_02)						\
-       : : "r" (out) : "memory" );
-
-// FIXME is sign right in the VSUB ?
-#define nREDUCEa(out)					\
-  asm (							\
-  VADD(UChi_00,UChi_10,UChi_00)				\
-  VADD(UChi_01,UChi_11,UChi_01)				\
-  VADD(UChi_02,UChi_12,UChi_02)	);			\
-  asm (VZERO(Chi_00)							\
-       VSUB(UChi_00,Chi_00,UChi_00)					\
-       VSUB(UChi_01,Chi_00,UChi_01)					\
-       VSUB(UChi_02,Chi_00,UChi_02)				);	\
-  asm (									\
-       VSTORE(0,%0,pUChi_00)				\
-       VSTORE(1,%0,pUChi_01)				\
-       VSTORE(2,%0,pUChi_02)				\
-       : : "r" (out) : "memory" );
-
-#define PERMUTE_DIR(dir)			\
-      permute##dir(Chi_0,Chi_0);\
-      permute##dir(Chi_1,Chi_1);\
-      permute##dir(Chi_2,Chi_2);
-
-namespace Grid {
-namespace QCD {
-
-template <class Impl>
-void StaggeredKernels<Impl>::DhopSiteAsm(StencilImpl &st, LebesgueOrder &lo, 
-					 DoubledGaugeField &U, DoubledGaugeField &UUU,
-					 SiteSpinor *buf, int LLs, int sU, 
-					 const FermionField &in, FermionField &out,int dag) 
-{
-  assert(0);
-};
-
-
-//#define CONDITIONAL_MOVE(l,o,out) if ( l ) { out = (uint64_t) &in._odata[o] ; } else { out =(uint64_t) &buf[o]; }
-
-#define CONDITIONAL_MOVE(l,o,out) { const SiteSpinor *ptr = l? in_p : buf; out = (uint64_t) &ptr[o]; }
-
-#define PREPARE_XYZT(X,Y,Z,T,skew,UU)			\
-  PREPARE(X,Y,Z,T,skew,UU);				\
-  PF_GAUGE_XYZT(gauge0);					\
-  PF_GAUGE_XYZT(gauge1);					\
-  PF_GAUGE_XYZT(gauge2);					\
-  PF_GAUGE_XYZT(gauge3);					
-
-#define PREPARE_LS(X,Y,Z,T,skew,UU)			\
-  PREPARE(X,Y,Z,T,skew,UU);				\
-  PF_GAUGE_LS(gauge0);					\
-  PF_GAUGE_LS(gauge1);					\
-  PF_GAUGE_LS(gauge2);					\
-  PF_GAUGE_LS(gauge3);					
-
-#define PREPARE(X,Y,Z,T,skew,UU)					\
-  SE0=st.GetEntry(ptype,X+skew,sF);					\
-  o0 = SE0->_offset;							\
-  l0 = SE0->_is_local;							\
-  p0 = SE0->_permute;							\
-  CONDITIONAL_MOVE(l0,o0,addr0);					\
-  PF_CHI(addr0);							\
-  									\
-  SE1=st.GetEntry(ptype,Y+skew,sF);					\
-  o1 = SE1->_offset;							\
-  l1 = SE1->_is_local;							\
-  p1 = SE1->_permute;							\
-  CONDITIONAL_MOVE(l1,o1,addr1);					\
-  PF_CHI(addr1);							\
-  									\
-  SE2=st.GetEntry(ptype,Z+skew,sF);					\
-  o2 = SE2->_offset;							\
-  l2 = SE2->_is_local;							\
-  p2 = SE2->_permute;							\
-  CONDITIONAL_MOVE(l2,o2,addr2);					\
-  PF_CHI(addr2);							\
-  									\
-  SE3=st.GetEntry(ptype,T+skew,sF);					\
-  o3 = SE3->_offset;							\
-  l3 = SE3->_is_local;							\
-  p3 = SE3->_permute;							\
-  CONDITIONAL_MOVE(l3,o3,addr3);					\
-  PF_CHI(addr3);							\
-  									\
-  gauge0 =(uint64_t)&UU._odata[sU]( X );				\
-  gauge1 =(uint64_t)&UU._odata[sU]( Y );				\
-  gauge2 =(uint64_t)&UU._odata[sU]( Z );				\
-  gauge3 =(uint64_t)&UU._odata[sU]( T ); 
-  
-  // This is the single precision 5th direction vectorised kernel
-#include <Grid/simd/Intel512single.h>
-template <> void StaggeredKernels<StaggeredVec5dImplF>::DhopSiteAsm(StencilImpl &st, LebesgueOrder &lo, 
-								    DoubledGaugeField &U, DoubledGaugeField &UUU,
-								    SiteSpinor *buf, int LLs, int sU, 
-								    const FermionField &in, FermionField &out,int dag) 
-{
-#ifdef AVX512
-  uint64_t gauge0,gauge1,gauge2,gauge3;
-  uint64_t addr0,addr1,addr2,addr3;
-  const SiteSpinor *in_p; in_p = &in._odata[0];
-
-  int o0,o1,o2,o3; // offsets
-  int l0,l1,l2,l3; // local 
-  int p0,p1,p2,p3; // perm
-  int ptype;
-  StencilEntry *SE0;
-  StencilEntry *SE1;
-  StencilEntry *SE2;
-  StencilEntry *SE3;
-
-   for(int s=0;s<LLs;s++){
-
-    int sF=s+LLs*sU;
-    // Xp, Yp, Zp, Tp
-    PREPARE(Xp,Yp,Zp,Tp,0,U);
-    LOAD_CHI(addr0,addr1,addr2,addr3);
-    MULT_LS(gauge0,gauge1,gauge2,gauge3);  
-
-    PREPARE(Xm,Ym,Zm,Tm,0,U);
-    LOAD_CHI(addr0,addr1,addr2,addr3);
-    MULT_ADD_LS(gauge0,gauge1,gauge2,gauge3);  
-
-    PREPARE(Xp,Yp,Zp,Tp,8,UUU);
-    LOAD_CHI(addr0,addr1,addr2,addr3);
-    MULT_ADD_LS(gauge0,gauge1,gauge2,gauge3);
-
-    PREPARE(Xm,Ym,Zm,Tm,8,UUU);
-    LOAD_CHI(addr0,addr1,addr2,addr3);
-    MULT_ADD_LS(gauge0,gauge1,gauge2,gauge3);
-
-    addr0 = (uint64_t) &out._odata[sF];
-    if ( dag ) {
-      nREDUCE(addr0);
-    } else { 
-      REDUCE(addr0);
-    }
-   }
-#else 
-    assert(0);
-#endif
-   
-}
-
-#include <Grid/simd/Intel512double.h>
-template <> void StaggeredKernels<StaggeredVec5dImplD>::DhopSiteAsm(StencilImpl &st, LebesgueOrder &lo, 
-								    DoubledGaugeField &U, DoubledGaugeField &UUU,
-								    SiteSpinor *buf, int LLs, int sU, 
-								    const FermionField &in, FermionField &out,int dag) 
-{
-#ifdef AVX512
-  uint64_t gauge0,gauge1,gauge2,gauge3;
-  uint64_t addr0,addr1,addr2,addr3;
-  const SiteSpinor *in_p; in_p = &in._odata[0];
-
-  int o0,o1,o2,o3; // offsets
-  int l0,l1,l2,l3; // local 
-  int p0,p1,p2,p3; // perm
-  int ptype;
-  StencilEntry *SE0;
-  StencilEntry *SE1;
-  StencilEntry *SE2;
-  StencilEntry *SE3;
-
-  for(int s=0;s<LLs;s++){
-    int sF=s+LLs*sU;
-    // Xp, Yp, Zp, Tp
-    PREPARE(Xp,Yp,Zp,Tp,0,U);
-    LOAD_CHI(addr0,addr1,addr2,addr3);
-    MULT_LS(gauge0,gauge1,gauge2,gauge3);  
-
-    PREPARE(Xm,Ym,Zm,Tm,0,U);
-    LOAD_CHI(addr0,addr1,addr2,addr3);
-    MULT_ADD_LS(gauge0,gauge1,gauge2,gauge3);  
-
-    PREPARE(Xp,Yp,Zp,Tp,8,UUU);
-    LOAD_CHI(addr0,addr1,addr2,addr3);
-    MULT_ADD_LS(gauge0,gauge1,gauge2,gauge3);
-
-    PREPARE(Xm,Ym,Zm,Tm,8,UUU);
-    LOAD_CHI(addr0,addr1,addr2,addr3);
-    MULT_ADD_LS(gauge0,gauge1,gauge2,gauge3);
-
-    addr0 = (uint64_t) &out._odata[sF];
-    if ( dag ) {
-      nREDUCE(addr0);
-    } else { 
-      REDUCE(addr0);
-    }
-  }
-#else 
-  assert(0);
-#endif
-}
-   
-   
-
-
-#define PERMUTE_DIR3 __asm__ (	\
-  VPERM3(Chi_00,Chi_00)	\
-  VPERM3(Chi_01,Chi_01)	\
-  VPERM3(Chi_02,Chi_02)	);
-
-#define PERMUTE_DIR2 __asm__ (	\
-  VPERM2(Chi_10,Chi_10)	\
-  VPERM2(Chi_11,Chi_11)	\
-  VPERM2(Chi_12,Chi_12) );
-
-#define PERMUTE_DIR1 __asm__ (	\
-  VPERM1(Chi_00,Chi_00)	\
-  VPERM1(Chi_01,Chi_01)	\
-  VPERM1(Chi_02,Chi_02)	);
-
-#define PERMUTE_DIR0 __asm__ (			\
-  VPERM0(Chi_10,Chi_10)	\
-  VPERM0(Chi_11,Chi_11)	\
-  VPERM0(Chi_12,Chi_12) );
-
-#define PERMUTE01 \
-  if ( p0 ) { PERMUTE_DIR3; }\
-  if ( p1 ) { PERMUTE_DIR2; }
-
-#define PERMUTE23 \
-  if ( p2 ) { PERMUTE_DIR1; }\
-  if ( p3 ) { PERMUTE_DIR0; }
-
-  // This is the single precision 5th direction vectorised kernel
-
-#include <Grid/simd/Intel512single.h>
-template <> void StaggeredKernels<StaggeredImplF>::DhopSiteAsm(StencilImpl &st, LebesgueOrder &lo, 
-							       DoubledGaugeField &U, DoubledGaugeField &UUU,
-							       SiteSpinor *buf, int LLs, int sU, 
-							       const FermionField &in, FermionField &out,int dag) 
-{
-#ifdef AVX512
-  uint64_t gauge0,gauge1,gauge2,gauge3;
-  uint64_t addr0,addr1,addr2,addr3;
-  const SiteSpinor *in_p; in_p = &in._odata[0];
-
-  int o0,o1,o2,o3; // offsets
-  int l0,l1,l2,l3; // local 
-  int p0,p1,p2,p3; // perm
-  int ptype;
-  StencilEntry *SE0;
-  StencilEntry *SE1;
-  StencilEntry *SE2;
-  StencilEntry *SE3;
-
-  for(int s=0;s<LLs;s++){
-    
-    int sF=s+LLs*sU;
-    // Xp, Yp, Zp, Tp
-    PREPARE(Xp,Yp,Zp,Tp,0,U);
-    LOAD_CHIa(addr0,addr1);
-    PERMUTE01;
-    MULT_XYZT(gauge0,gauge1);
-    LOAD_CHIa(addr2,addr3);
-    PERMUTE23;
-    MULT_ADD_XYZT(gauge2,gauge3);  
-
-    PREPARE(Xm,Ym,Zm,Tm,0,U);
-    LOAD_CHIa(addr0,addr1);
-    PERMUTE01;
-    MULT_ADD_XYZT(gauge0,gauge1);
-    LOAD_CHIa(addr2,addr3);
-    PERMUTE23;
-    MULT_ADD_XYZT(gauge2,gauge3);  
-
-    PREPARE(Xp,Yp,Zp,Tp,8,UUU);
-    LOAD_CHIa(addr0,addr1);
-    PERMUTE01;
-    MULT_ADD_XYZT(gauge0,gauge1);
-    LOAD_CHIa(addr2,addr3);
-    PERMUTE23;
-    MULT_ADD_XYZT(gauge2,gauge3);  
-    
-    PREPARE(Xm,Ym,Zm,Tm,8,UUU);
-    LOAD_CHIa(addr0,addr1);
-    PERMUTE01;
-    MULT_ADD_XYZT(gauge0,gauge1);
-    LOAD_CHIa(addr2,addr3);
-    PERMUTE23;
-    MULT_ADD_XYZT(gauge2,gauge3);  
-
-    addr0 = (uint64_t) &out._odata[sF];
-    if ( dag ) { 
-      nREDUCEa(addr0);
-    } else { 
-      REDUCEa(addr0);
-    }
-  }
-#else 
-  assert(0);
-#endif
-}
-
-#include <Grid/simd/Intel512double.h>
-template <> void StaggeredKernels<StaggeredImplD>::DhopSiteAsm(StencilImpl &st, LebesgueOrder &lo, 
-							       DoubledGaugeField &U, DoubledGaugeField &UUU,
-							       SiteSpinor *buf, int LLs, int sU, 
-							       const FermionField &in, FermionField &out,int dag) 
-{
-#ifdef AVX512
-  uint64_t gauge0,gauge1,gauge2,gauge3;
-  uint64_t addr0,addr1,addr2,addr3;
-  const SiteSpinor *in_p; in_p = &in._odata[0];
-
-  int o0,o1,o2,o3; // offsets
-  int l0,l1,l2,l3; // local 
-  int p0,p1,p2,p3; // perm
-  int ptype;
-  StencilEntry *SE0;
-  StencilEntry *SE1;
-  StencilEntry *SE2;
-  StencilEntry *SE3;
-
-  for(int s=0;s<LLs;s++){
-    
-    int sF=s+LLs*sU;
-    // Xp, Yp, Zp, Tp
-    PREPARE(Xp,Yp,Zp,Tp,0,U);
-    LOAD_CHIa(addr0,addr1);
-    PERMUTE01;
-    MULT_XYZT(gauge0,gauge1);
-    LOAD_CHIa(addr2,addr3);
-    PERMUTE23;
-    MULT_ADD_XYZT(gauge2,gauge3);  
-    
-    PREPARE(Xm,Ym,Zm,Tm,0,U);
-    LOAD_CHIa(addr0,addr1);
-    PERMUTE01;
-    MULT_ADD_XYZT(gauge0,gauge1);
-    LOAD_CHIa(addr2,addr3);
-    PERMUTE23;
-    MULT_ADD_XYZT(gauge2,gauge3);  
-    
-    PREPARE(Xp,Yp,Zp,Tp,8,UUU);
-    LOAD_CHIa(addr0,addr1);
-    PERMUTE01;
-    MULT_ADD_XYZT(gauge0,gauge1);
-    LOAD_CHIa(addr2,addr3);
-    PERMUTE23;
-    MULT_ADD_XYZT(gauge2,gauge3);  
-    
-    PREPARE(Xm,Ym,Zm,Tm,8,UUU);
-    LOAD_CHIa(addr0,addr1);
-    PERMUTE01;
-    MULT_ADD_XYZT(gauge0,gauge1);
-    LOAD_CHIa(addr2,addr3);
-    PERMUTE23;
-    MULT_ADD_XYZT(gauge2,gauge3);  
-    
-    addr0 = (uint64_t) &out._odata[sF];
-    if ( dag ) {
-      nREDUCEa(addr0);
-    } else { 
-      REDUCEa(addr0);
-    }
-  }
-#else 
-  assert(0);
-#endif
-}
-
-#define KERNEL_INSTANTIATE(CLASS,FUNC,IMPL)			    \
-  template void CLASS<IMPL>::FUNC(StencilImpl &st, LebesgueOrder &lo,	\
-				  DoubledGaugeField &U,			\
-				  DoubledGaugeField &UUU,		\
-				  SiteSpinor *buf, int LLs,		\
-				  int sU, const FermionField &in, FermionField &out,int dag);
-
-KERNEL_INSTANTIATE(StaggeredKernels,DhopSiteAsm,StaggeredImplD);
-KERNEL_INSTANTIATE(StaggeredKernels,DhopSiteAsm,StaggeredImplF);
-KERNEL_INSTANTIATE(StaggeredKernels,DhopSiteAsm,StaggeredVec5dImplD);
-KERNEL_INSTANTIATE(StaggeredKernels,DhopSiteAsm,StaggeredVec5dImplF);
-
-}}
-

Grid/qcd/action/fermion/StaggeredKernelsHand.cc

@@ -1,399 +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>
-
-
-#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._odata[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._odata[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._odata);					\
-    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._odata);				\
-    if ( perm) {					\
-      PERMUTE_DIR(Perm);				\
-    }							\
-  } else if ( st.same_node[Dir] ) {			\
-    LOAD_CHI(buf);					\
-  }							\
-  if (SE->_is_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;				\
-  perm   = SE->_permute;				\
-  if ((!SE->_is_local) && (!st.same_node[Dir]) ) {		\
-    nmu++;							\
-    { LOAD_CHI(buf);	  }					\
-    { MULT_ADD(U,Dir,even); }					\
-  }								
-
-namespace Grid {
-namespace QCD {
-
-
-template <class Impl>
-void StaggeredKernels<Impl>::DhopSiteHand(StencilImpl &st, LebesgueOrder &lo, 
-					  DoubledGaugeField &U,DoubledGaugeField &UUU,
-					  SiteSpinor *buf, int LLs, int sU, 
-					  const FermionField &in, FermionField &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._odata[sF],result);
-  }
-}
-
-
-template <class Impl>
-void StaggeredKernels<Impl>::DhopSiteHandInt(StencilImpl &st, LebesgueOrder &lo, 
-					     DoubledGaugeField &U, DoubledGaugeField &UUU,
-					     SiteSpinor *buf, int LLs, int sU, 
-					     const FermionField &in, FermionField &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;
-
-    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._odata[sF],result);
-  }
-}
-
-
-template <class Impl>
-void StaggeredKernels<Impl>::DhopSiteHandExt(StencilImpl &st, LebesgueOrder &lo, 
-					     DoubledGaugeField &U, DoubledGaugeField &UUU,
-					     SiteSpinor *buf, int LLs, int sU, 
-					     const FermionField &in, FermionField &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;
-
-    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._odata[sF] = out._odata[sF] + result;
-    }
-  }
-}
-
-
-#define DHOP_SITE_HAND_INSTANTIATE(IMPL)				\
-  template void StaggeredKernels<IMPL>::DhopSiteHand(StencilImpl &st, LebesgueOrder &lo, \
-						     DoubledGaugeField &U,DoubledGaugeField &UUU, \
-						     SiteSpinor *buf, int LLs, int sU, \
-						     const FermionField &in, FermionField &out, int dag); \
-									\
-  template void StaggeredKernels<IMPL>::DhopSiteHandInt(StencilImpl &st, LebesgueOrder &lo, \
-						     DoubledGaugeField &U,DoubledGaugeField &UUU, \
-						     SiteSpinor *buf, int LLs, int sU, \
-						     const FermionField &in, FermionField &out, int dag); \
-									\
-  template void StaggeredKernels<IMPL>::DhopSiteHandExt(StencilImpl &st, LebesgueOrder &lo, \
-						     DoubledGaugeField &U,DoubledGaugeField &UUU, \
-						     SiteSpinor *buf, int LLs, int sU, \
-						     const FermionField &in, FermionField &out, int dag); \
-
-DHOP_SITE_HAND_INSTANTIATE(StaggeredImplD);
-DHOP_SITE_HAND_INSTANTIATE(StaggeredImplF);
-DHOP_SITE_HAND_INSTANTIATE(StaggeredVec5dImplD);
-DHOP_SITE_HAND_INSTANTIATE(StaggeredVec5dImplF);
-
-
-}
-}
-

Grid/qcd/action/fermion/WilsonCloverFermion.cc

@@ -1,243 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: ./lib/qcd/action/fermion/WilsonCloverFermion.cc
-
-    Copyright (C) 2017
-
-    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 */
-#include <Grid/Grid.h>
-//#include <Grid/Eigen/Dense>
-#include <Grid/qcd/spin/Dirac.h>
-
-namespace Grid
-{
-namespace QCD
-{
-
-// *NOT* EO
-template <class Impl>
-RealD WilsonCloverFermion<Impl>::M(const FermionField &in, FermionField &out)
-{
-  FermionField temp(out._grid);
-
-  // Wilson term
-  out.checkerboard = in.checkerboard;
-  this->Dhop(in, out, DaggerNo);
-
-  // Clover term
-  Mooee(in, temp);
-
-  out += temp;
-  return norm2(out);
-}
-
-template <class Impl>
-RealD WilsonCloverFermion<Impl>::Mdag(const FermionField &in, FermionField &out)
-{
-  FermionField temp(out._grid);
-
-  // Wilson term
-  out.checkerboard = in.checkerboard;
-  this->Dhop(in, out, DaggerYes);
-
-  // Clover term
-  MooeeDag(in, temp);
-
-  out += temp;
-  return norm2(out);
-}
-
-template <class Impl>
-void WilsonCloverFermion<Impl>::ImportGauge(const GaugeField &_Umu)
-{
-  WilsonFermion<Impl>::ImportGauge(_Umu);
-  GridBase *grid = _Umu._grid;
-  typename Impl::GaugeLinkField Bx(grid), By(grid), Bz(grid), Ex(grid), Ey(grid), Ez(grid);
-
-  // Compute the field strength terms mu>nu
-  WilsonLoops<Impl>::FieldStrength(Bx, _Umu, Zdir, Ydir);
-  WilsonLoops<Impl>::FieldStrength(By, _Umu, Zdir, Xdir);
-  WilsonLoops<Impl>::FieldStrength(Bz, _Umu, Ydir, Xdir);
-  WilsonLoops<Impl>::FieldStrength(Ex, _Umu, Tdir, Xdir);
-  WilsonLoops<Impl>::FieldStrength(Ey, _Umu, Tdir, Ydir);
-  WilsonLoops<Impl>::FieldStrength(Ez, _Umu, Tdir, Zdir);
-
-  // Compute the Clover Operator acting on Colour and Spin
-  // multiply here by the clover coefficients for the anisotropy
-  CloverTerm  = fillCloverYZ(Bx) * csw_r;
-  CloverTerm += fillCloverXZ(By) * csw_r;
-  CloverTerm += fillCloverXY(Bz) * csw_r;
-  CloverTerm += fillCloverXT(Ex) * csw_t;
-  CloverTerm += fillCloverYT(Ey) * csw_t;
-  CloverTerm += fillCloverZT(Ez) * csw_t;
-  CloverTerm += diag_mass;
-
-  int lvol = _Umu._grid->lSites();
-  int DimRep = Impl::Dimension;
-
-  Eigen::MatrixXcd EigenCloverOp = Eigen::MatrixXcd::Zero(Ns * DimRep, Ns * DimRep);
-  Eigen::MatrixXcd EigenInvCloverOp = Eigen::MatrixXcd::Zero(Ns * DimRep, Ns * DimRep);
-
-  std::vector<int> lcoor;
-  typename SiteCloverType::scalar_object Qx = zero, Qxinv = zero;
-
-  for (int site = 0; site < lvol; site++)
-  {
-    grid->LocalIndexToLocalCoor(site, lcoor);
-    EigenCloverOp = Eigen::MatrixXcd::Zero(Ns * DimRep, Ns * DimRep);
-    peekLocalSite(Qx, CloverTerm, lcoor);
-    Qxinv = zero;
-    //if (csw!=0){
-    for (int j = 0; j < Ns; j++)
-      for (int k = 0; k < Ns; k++)
-        for (int a = 0; a < DimRep; a++)
-          for (int b = 0; b < DimRep; b++)
-            EigenCloverOp(a + j * DimRep, b + k * DimRep) = Qx()(j, k)(a, b);
-    //   if (site==0) std::cout << "site =" << site << "\n" << EigenCloverOp << std::endl;
-
-    EigenInvCloverOp = EigenCloverOp.inverse();
-    //std::cout << EigenInvCloverOp << std::endl;
-    for (int j = 0; j < Ns; j++)
-      for (int k = 0; k < Ns; k++)
-        for (int a = 0; a < DimRep; a++)
-          for (int b = 0; b < DimRep; b++)
-            Qxinv()(j, k)(a, b) = EigenInvCloverOp(a + j * DimRep, b + k * DimRep);
-    //    if (site==0) std::cout << "site =" << site << "\n" << EigenInvCloverOp << std::endl;
-    //  }
-    pokeLocalSite(Qxinv, CloverTermInv, lcoor);
-  }
-
-  // Separate the even and odd parts
-  pickCheckerboard(Even, CloverTermEven, CloverTerm);
-  pickCheckerboard(Odd, CloverTermOdd, CloverTerm);
-
-  pickCheckerboard(Even, CloverTermDagEven, adj(CloverTerm));
-  pickCheckerboard(Odd, CloverTermDagOdd, adj(CloverTerm));
-
-  pickCheckerboard(Even, CloverTermInvEven, CloverTermInv);
-  pickCheckerboard(Odd, CloverTermInvOdd, CloverTermInv);
-
-  pickCheckerboard(Even, CloverTermInvDagEven, adj(CloverTermInv));
-  pickCheckerboard(Odd, CloverTermInvDagOdd, adj(CloverTermInv));
-}
-
-template <class Impl>
-void WilsonCloverFermion<Impl>::Mooee(const FermionField &in, FermionField &out)
-{
-  this->MooeeInternal(in, out, DaggerNo, InverseNo);
-}
-
-template <class Impl>
-void WilsonCloverFermion<Impl>::MooeeDag(const FermionField &in, FermionField &out)
-{
-  this->MooeeInternal(in, out, DaggerYes, InverseNo);
-}
-
-template <class Impl>
-void WilsonCloverFermion<Impl>::MooeeInv(const FermionField &in, FermionField &out)
-{
-  this->MooeeInternal(in, out, DaggerNo, InverseYes);
-}
-
-template <class Impl>
-void WilsonCloverFermion<Impl>::MooeeInvDag(const FermionField &in, FermionField &out)
-{
-  this->MooeeInternal(in, out, DaggerYes, InverseYes);
-}
-
-template <class Impl>
-void WilsonCloverFermion<Impl>::MooeeInternal(const FermionField &in, FermionField &out, int dag, int inv)
-{
-  out.checkerboard = in.checkerboard;
-  CloverFieldType *Clover;
-  assert(in.checkerboard == Odd || in.checkerboard == Even);
-
-  if (dag)
-  {
-    if (in._grid->_isCheckerBoarded)
-    {
-      if (in.checkerboard == Odd)
-      {
-        Clover = (inv) ? &CloverTermInvDagOdd : &CloverTermDagOdd;
-      }
-      else
-      {
-        Clover = (inv) ? &CloverTermInvDagEven : &CloverTermDagEven;
-      }
-      out = *Clover * in;
-    }
-    else
-    {
-      Clover = (inv) ? &CloverTermInv : &CloverTerm;
-      out = adj(*Clover) * in;
-    }
-  }
-  else
-  {
-    if (in._grid->_isCheckerBoarded)
-    {
-
-      if (in.checkerboard == Odd)
-      {
-        //  std::cout << "Calling clover term Odd" << std::endl;
-        Clover = (inv) ? &CloverTermInvOdd : &CloverTermOdd;
-      }
-      else
-      {
-        //  std::cout << "Calling clover term Even" << std::endl;
-        Clover = (inv) ? &CloverTermInvEven : &CloverTermEven;
-      }
-      out = *Clover * in;
-      //  std::cout << GridLogMessage << "*Clover.checkerboard "  << (*Clover).checkerboard << std::endl;
-    }
-    else
-    {
-      Clover = (inv) ? &CloverTermInv : &CloverTerm;
-      out = *Clover * in;
-    }
-  }
-
-} // MooeeInternal
-
-
-// Derivative parts
-template <class Impl>
-void WilsonCloverFermion<Impl>::MooDeriv(GaugeField &mat, const FermionField &X, const FermionField &Y, int dag)
-{
-  assert(0);
-}
-
-// Derivative parts
-template <class Impl>
-void WilsonCloverFermion<Impl>::MeeDeriv(GaugeField &mat, const FermionField &U, const FermionField &V, int dag)
-{
-  assert(0); // not implemented yet
-}
-
-FermOpTemplateInstantiate(WilsonCloverFermion);
-AdjointFermOpTemplateInstantiate(WilsonCloverFermion);
-TwoIndexFermOpTemplateInstantiate(WilsonCloverFermion);
-//GparityFermOpTemplateInstantiate(WilsonCloverFermion);
-}
-}

Grid/qcd/action/fermion/WilsonCloverFermion.h

@@ -1,367 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: ./lib/qcd/action/fermion/WilsonCloverFermion.h
-
-    Copyright (C) 2017
-
-    Author: Guido Cossu <guido.cossu@ed.ac.uk>
-    Author: David Preti <>
-
-    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_CLOVER_FERMION_H
-#define GRID_QCD_WILSON_CLOVER_FERMION_H
-
-#include <Grid/Grid.h>
-
-namespace Grid
-{
-namespace QCD
-{
-
-///////////////////////////////////////////////////////////////////
-// Wilson Clover
-//
-// Operator ( with anisotropy coefficients):
-//
-// Q =   1 + (Nd-1)/xi_0 + m
-//     + W_t + (nu/xi_0) * W_s
-//     - 1/2*[ csw_t * sum_s (sigma_ts F_ts) + (csw_s/xi_0) * sum_ss (sigma_ss F_ss)  ]
-//
-// s spatial, t temporal directions.
-// where W_t and W_s are the temporal and spatial components of the
-// Wilson Dirac operator
-//
-// csw_r = csw_t to recover the isotropic version
-//////////////////////////////////////////////////////////////////
-
-template <class Impl>
-class WilsonCloverFermion : public WilsonFermion<Impl>
-{
-public:
-  // Types definitions
-  INHERIT_IMPL_TYPES(Impl);
-  template <typename vtype>
-  using iImplClover = iScalar<iMatrix<iMatrix<vtype, Impl::Dimension>, Ns>>;
-  typedef iImplClover<Simd> SiteCloverType;
-  typedef Lattice<SiteCloverType> CloverFieldType;
-
-public:
-  typedef WilsonFermion<Impl> WilsonBase;
-
-  virtual int    ConstEE(void)     { return 0; };
-  virtual void Instantiatable(void){};
-  // Constructors
-  WilsonCloverFermion(GaugeField &_Umu, GridCartesian &Fgrid,
-                      GridRedBlackCartesian &Hgrid,
-                      const RealD _mass,
-                      const RealD _csw_r = 0.0,
-                      const RealD _csw_t = 0.0,
-                      const WilsonAnisotropyCoefficients &clover_anisotropy = WilsonAnisotropyCoefficients(),
-                      const ImplParams &impl_p = ImplParams()) : WilsonFermion<Impl>(_Umu,
-                                                                                     Fgrid,
-                                                                                     Hgrid,
-                                                                                     _mass, impl_p, clover_anisotropy),
-                                                                 CloverTerm(&Fgrid),
-                                                                 CloverTermInv(&Fgrid),
-                                                                 CloverTermEven(&Hgrid),
-                                                                 CloverTermOdd(&Hgrid),
-                                                                 CloverTermInvEven(&Hgrid),
-                                                                 CloverTermInvOdd(&Hgrid),
-                                                                 CloverTermDagEven(&Hgrid),
-                                                                 CloverTermDagOdd(&Hgrid),
-                                                                 CloverTermInvDagEven(&Hgrid),
-                                                                 CloverTermInvDagOdd(&Hgrid)
-  {
-    assert(Nd == 4); // require 4 dimensions
-
-    if (clover_anisotropy.isAnisotropic)
-    {
-      csw_r = _csw_r * 0.5 / clover_anisotropy.xi_0;
-      diag_mass = _mass + 1.0 + (Nd - 1) * (clover_anisotropy.nu / clover_anisotropy.xi_0);
-    }
-    else
-    {
-      csw_r = _csw_r * 0.5;
-      diag_mass = 4.0 + _mass;
-    }
-    csw_t = _csw_t * 0.5;
-
-    if (csw_r == 0)
-      std::cout << GridLogWarning << "Initializing WilsonCloverFermion with csw_r = 0" << std::endl;
-    if (csw_t == 0)
-      std::cout << GridLogWarning << "Initializing WilsonCloverFermion with csw_t = 0" << std::endl;
-
-    ImportGauge(_Umu);
-  }
-
-  virtual RealD M(const FermionField &in, FermionField &out);
-  virtual RealD Mdag(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 MooeeInternal(const FermionField &in, FermionField &out, int dag, int inv);
-
-  //virtual void MDeriv(GaugeField &mat, const FermionField &U, const FermionField &V, int dag);
-  virtual void MooDeriv(GaugeField &mat, const FermionField &U, const FermionField &V, int dag);
-  virtual void MeeDeriv(GaugeField &mat, const FermionField &U, const FermionField &V, int dag);
-
-  void ImportGauge(const GaugeField &_Umu);
-
-  // Derivative parts unpreconditioned pseudofermions
-  void MDeriv(GaugeField &force, const FermionField &X, const FermionField &Y, int dag)
-  {
-    conformable(X._grid, Y._grid);
-    conformable(X._grid, force._grid);
-    GaugeLinkField force_mu(force._grid), lambda(force._grid);
-    GaugeField clover_force(force._grid);
-    PropagatorField Lambda(force._grid);
-
-    // Guido: Here we are hitting some performance issues:
-    // need to extract the components of the DoubledGaugeField
-    // for each call
-    // Possible solution
-    // Create a vector object to store them? (cons: wasting space)
-    std::vector<GaugeLinkField> U(Nd, this->Umu._grid);
-
-    Impl::extractLinkField(U, this->Umu);
-
-    force = zero;
-    // Derivative of the Wilson hopping term
-    this->DhopDeriv(force, X, Y, dag);
-
-    ///////////////////////////////////////////////////////////
-    // Clover term derivative
-    ///////////////////////////////////////////////////////////
-    Impl::outerProductImpl(Lambda, X, Y);
-    //std::cout << "Lambda:" << Lambda << std::endl;
-
-    Gamma::Algebra sigma[] = {
-        Gamma::Algebra::SigmaXY,
-        Gamma::Algebra::SigmaXZ,
-        Gamma::Algebra::SigmaXT,
-        Gamma::Algebra::MinusSigmaXY,
-        Gamma::Algebra::SigmaYZ,
-        Gamma::Algebra::SigmaYT,
-        Gamma::Algebra::MinusSigmaXZ,
-        Gamma::Algebra::MinusSigmaYZ,
-        Gamma::Algebra::SigmaZT,
-        Gamma::Algebra::MinusSigmaXT,
-        Gamma::Algebra::MinusSigmaYT,
-        Gamma::Algebra::MinusSigmaZT};
-
-    /*
-      sigma_{\mu \nu}=
-      | 0         sigma[0]  sigma[1]  sigma[2] |
-      | sigma[3]    0       sigma[4]  sigma[5] |
-      | sigma[6]  sigma[7]     0      sigma[8] |
-      | sigma[9]  sigma[10] sigma[11]   0      |
-    */
-
-    int count = 0;
-    clover_force = zero;
-    for (int mu = 0; mu < 4; mu++)
-    {
-      force_mu = zero;
-      for (int nu = 0; nu < 4; nu++)
-      {
-        if (mu == nu)
-        continue;
-        
-        RealD factor;
-        if (nu == 4 || mu == 4)
-        {
-          factor = 2.0 * csw_t;
-        }
-        else
-        {
-          factor = 2.0 * csw_r;
-        }
-        PropagatorField Slambda = Gamma(sigma[count]) * Lambda; // sigma checked
-        Impl::TraceSpinImpl(lambda, Slambda);                   // traceSpin ok
-        force_mu -= factor*Cmunu(U, lambda, mu, nu);                   // checked
-        count++;
-      }
-
-      pokeLorentz(clover_force, U[mu] * force_mu, mu);
-    }
-    //clover_force *= csw;
-    force += clover_force;
-  }
-
-  // Computing C_{\mu \nu}(x) as in Eq.(B.39) in Zbigniew Sroczynski's PhD thesis
-  GaugeLinkField Cmunu(std::vector<GaugeLinkField> &U, GaugeLinkField &lambda, int mu, int nu)
-  {
-    conformable(lambda._grid, U[0]._grid);
-    GaugeLinkField out(lambda._grid), tmp(lambda._grid);
-    // insertion in upper staple
-    // please check redundancy of shift operations
-
-    // C1+
-    tmp = lambda * U[nu];
-    out = Impl::ShiftStaple(Impl::CovShiftForward(tmp, nu, Impl::CovShiftBackward(U[mu], mu, Impl::CovShiftIdentityBackward(U[nu], nu))), mu);
-
-    // C2+
-    tmp = U[mu] * Impl::ShiftStaple(adj(lambda), mu);
-    out += Impl::ShiftStaple(Impl::CovShiftForward(U[nu], nu, Impl::CovShiftBackward(tmp, mu, Impl::CovShiftIdentityBackward(U[nu], nu))), mu);
-
-    // C3+
-    tmp = U[nu] * Impl::ShiftStaple(adj(lambda), nu);
-    out += Impl::ShiftStaple(Impl::CovShiftForward(U[nu], nu, Impl::CovShiftBackward(U[mu], mu, Impl::CovShiftIdentityBackward(tmp, nu))), mu);
-
-    // C4+
-    out += Impl::ShiftStaple(Impl::CovShiftForward(U[nu], nu, Impl::CovShiftBackward(U[mu], mu, Impl::CovShiftIdentityBackward(U[nu], nu))), mu) * lambda;
-
-    // insertion in lower staple
-    // C1-
-    out -= Impl::ShiftStaple(lambda, mu) * Impl::ShiftStaple(Impl::CovShiftBackward(U[nu], nu, Impl::CovShiftBackward(U[mu], mu, U[nu])), mu);
-
-    // C2-
-    tmp = adj(lambda) * U[nu];
-    out -= Impl::ShiftStaple(Impl::CovShiftBackward(tmp, nu, Impl::CovShiftBackward(U[mu], mu, U[nu])), mu);
-
-    // C3-
-    tmp = lambda * U[nu];
-    out -= Impl::ShiftStaple(Impl::CovShiftBackward(U[nu], nu, Impl::CovShiftBackward(U[mu], mu, tmp)), mu);
-
-    // C4-
-    out -= Impl::ShiftStaple(Impl::CovShiftBackward(U[nu], nu, Impl::CovShiftBackward(U[mu], mu, U[nu])), mu) * lambda;
-
-    return out;
-  }
-
-private:
-  // here fixing the 4 dimensions, make it more general?
-
-  RealD csw_r;                                               // Clover coefficient - spatial
-  RealD csw_t;                                               // Clover coefficient - temporal
-  RealD diag_mass;                                           // Mass term
-  CloverFieldType CloverTerm, CloverTermInv;                 // Clover term
-  CloverFieldType CloverTermEven, CloverTermOdd;             // Clover term EO
-  CloverFieldType CloverTermInvEven, CloverTermInvOdd;       // Clover term Inv EO
-  CloverFieldType CloverTermDagEven, CloverTermDagOdd;       // Clover term Dag EO
-  CloverFieldType CloverTermInvDagEven, CloverTermInvDagOdd; // Clover term Inv Dag EO
-
-  // eventually these can be compressed into 6x6 blocks instead of the 12x12
-  // using the DeGrand-Rossi basis for the gamma matrices
-  CloverFieldType fillCloverYZ(const GaugeLinkField &F)
-  {
-    CloverFieldType T(F._grid);
-    T = zero;
-    PARALLEL_FOR_LOOP
-    for (int i = 0; i < CloverTerm._grid->oSites(); i++)
-    {
-      T._odata[i]()(0, 1) = timesMinusI(F._odata[i]()());
-      T._odata[i]()(1, 0) = timesMinusI(F._odata[i]()());
-      T._odata[i]()(2, 3) = timesMinusI(F._odata[i]()());
-      T._odata[i]()(3, 2) = timesMinusI(F._odata[i]()());
-    }
-
-    return T;
-  }
-
-  CloverFieldType fillCloverXZ(const GaugeLinkField &F)
-  {
-    CloverFieldType T(F._grid);
-    T = zero;
-    PARALLEL_FOR_LOOP
-    for (int i = 0; i < CloverTerm._grid->oSites(); i++)
-    {
-      T._odata[i]()(0, 1) = -F._odata[i]()();
-      T._odata[i]()(1, 0) = F._odata[i]()();
-      T._odata[i]()(2, 3) = -F._odata[i]()();
-      T._odata[i]()(3, 2) = F._odata[i]()();
-    }
-
-    return T;
-  }
-
-  CloverFieldType fillCloverXY(const GaugeLinkField &F)
-  {
-    CloverFieldType T(F._grid);
-    T = zero;
-    PARALLEL_FOR_LOOP
-    for (int i = 0; i < CloverTerm._grid->oSites(); i++)
-    {
-
-      T._odata[i]()(0, 0) = timesMinusI(F._odata[i]()());
-      T._odata[i]()(1, 1) = timesI(F._odata[i]()());
-      T._odata[i]()(2, 2) = timesMinusI(F._odata[i]()());
-      T._odata[i]()(3, 3) = timesI(F._odata[i]()());
-    }
-
-    return T;
-  }
-
-  CloverFieldType fillCloverXT(const GaugeLinkField &F)
-  {
-    CloverFieldType T(F._grid);
-    T = zero;
-    PARALLEL_FOR_LOOP
-    for (int i = 0; i < CloverTerm._grid->oSites(); i++)
-    {
-      T._odata[i]()(0, 1) = timesI(F._odata[i]()());
-      T._odata[i]()(1, 0) = timesI(F._odata[i]()());
-      T._odata[i]()(2, 3) = timesMinusI(F._odata[i]()());
-      T._odata[i]()(3, 2) = timesMinusI(F._odata[i]()());
-    }
-
-    return T;
-  }
-
-  CloverFieldType fillCloverYT(const GaugeLinkField &F)
-  {
-    CloverFieldType T(F._grid);
-    T = zero;
-    PARALLEL_FOR_LOOP
-    for (int i = 0; i < CloverTerm._grid->oSites(); i++)
-    {
-      T._odata[i]()(0, 1) = -(F._odata[i]()());
-      T._odata[i]()(1, 0) = (F._odata[i]()());
-      T._odata[i]()(2, 3) = (F._odata[i]()());
-      T._odata[i]()(3, 2) = -(F._odata[i]()());
-    }
-
-    return T;
-  }
-
-  CloverFieldType fillCloverZT(const GaugeLinkField &F)
-  {
-    CloverFieldType T(F._grid);
-    T = zero;
-    PARALLEL_FOR_LOOP
-    for (int i = 0; i < CloverTerm._grid->oSites(); i++)
-    {
-      T._odata[i]()(0, 0) = timesI(F._odata[i]()());
-      T._odata[i]()(1, 1) = timesMinusI(F._odata[i]()());
-      T._odata[i]()(2, 2) = timesMinusI(F._odata[i]()());
-      T._odata[i]()(3, 3) = timesI(F._odata[i]()());
-    }
-
-    return T;
-  }
-};
-}
-}
-
-#endif // GRID_QCD_WILSON_CLOVER_FERMION_H

Grid/qcd/action/fermion/WilsonCompressor.h

@@ -1,373 +0,0 @@
-    /*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/WilsonCompressor.h
-
-    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 */
-#ifndef  GRID_QCD_WILSON_COMPRESSOR_H
-#define  GRID_QCD_WILSON_COMPRESSOR_H
-
-namespace Grid {
-namespace QCD {
-
-/////////////////////////////////////////////////////////////////////////////////////////////
-// optimised versions supporting half precision too
-/////////////////////////////////////////////////////////////////////////////////////////////
-
-template<class _HCspinor,class _Hspinor,class _Spinor, class projector,typename SFINAE = void >
-class WilsonCompressorTemplate;
-
-
-template<class _HCspinor,class _Hspinor,class _Spinor, class projector>
-class WilsonCompressorTemplate< _HCspinor, _Hspinor, _Spinor, projector,
-  typename std::enable_if<std::is_same<_HCspinor,_Hspinor>::value>::type >
-{
- public:
-  
-  int mu,dag;  
-
-  void Point(int p) { mu=p; };
-
-  WilsonCompressorTemplate(int _dag=0){
-    dag = _dag;
-  }
-
-  typedef _Spinor         SiteSpinor;
-  typedef _Hspinor     SiteHalfSpinor;
-  typedef _HCspinor SiteHalfCommSpinor;
-  typedef typename SiteHalfCommSpinor::vector_type vComplexLow;
-  typedef typename SiteHalfSpinor::vector_type     vComplexHigh;
-  constexpr static int Nw=sizeof(SiteHalfSpinor)/sizeof(vComplexHigh);
-
-  inline int CommDatumSize(void) {
-    return sizeof(SiteHalfCommSpinor);
-  }
-
-  /*****************************************************/
-  /* Compress includes precision change if mpi data is not same */
-  /*****************************************************/
-  inline void Compress(SiteHalfSpinor * __restrict__ buf,Integer o,const SiteSpinor &in) {
-    SiteHalfSpinor tmp;
-    projector::Proj(tmp,in,mu,dag);
-    vstream(buf[o],tmp);
-  }
-
-  /*****************************************************/
-  /* Exchange includes precision change if mpi data is not same */
-  /*****************************************************/
-  inline void Exchange(SiteHalfSpinor * __restrict__ mp,
-                       const SiteHalfSpinor * __restrict__ vp0,
-                       const SiteHalfSpinor * __restrict__ vp1,
-		       Integer type,Integer o){
-    SiteHalfSpinor tmp1;
-    SiteHalfSpinor tmp2;
-    exchange(tmp1,tmp2,vp0[o],vp1[o],type);
-    vstream(mp[2*o  ],tmp1);
-    vstream(mp[2*o+1],tmp2);
-  }
-
-  /*****************************************************/
-  /* Have a decompression step if mpi data is not same */
-  /*****************************************************/
-  inline void Decompress(SiteHalfSpinor * __restrict__ out,
-			 SiteHalfSpinor * __restrict__ in, Integer o) {    
-    assert(0);
-  }
-
-  /*****************************************************/
-  /* Compress Exchange                                 */
-  /*****************************************************/
-  inline void CompressExchange(SiteHalfSpinor * __restrict__ out0,
-			       SiteHalfSpinor * __restrict__ out1,
-			       const SiteSpinor * __restrict__ in,
-			       Integer j,Integer k, Integer m,Integer type){
-    SiteHalfSpinor temp1, temp2,temp3,temp4;
-    projector::Proj(temp1,in[k],mu,dag);
-    projector::Proj(temp2,in[m],mu,dag);
-    exchange(temp3,temp4,temp1,temp2,type);
-    vstream(out0[j],temp3);
-    vstream(out1[j],temp4);
-  }
-
-  /*****************************************************/
-  /* Pass the info to the stencil */
-  /*****************************************************/
-  inline bool DecompressionStep(void) { return false; }
-
-};
-
-template<class _HCspinor,class _Hspinor,class _Spinor, class projector>
-class WilsonCompressorTemplate< _HCspinor, _Hspinor, _Spinor, projector,
-  typename std::enable_if<!std::is_same<_HCspinor,_Hspinor>::value>::type >
-{
- public:
-  
-  int mu,dag;  
-
-  void Point(int p) { mu=p; };
-
-  WilsonCompressorTemplate(int _dag=0){
-    dag = _dag;
-  }
-
-  typedef _Spinor         SiteSpinor;
-  typedef _Hspinor     SiteHalfSpinor;
-  typedef _HCspinor SiteHalfCommSpinor;
-  typedef typename SiteHalfCommSpinor::vector_type vComplexLow;
-  typedef typename SiteHalfSpinor::vector_type     vComplexHigh;
-  constexpr static int Nw=sizeof(SiteHalfSpinor)/sizeof(vComplexHigh);
-
-  inline int CommDatumSize(void) {
-    return sizeof(SiteHalfCommSpinor);
-  }
-
-  /*****************************************************/
-  /* Compress includes precision change if mpi data is not same */
-  /*****************************************************/
-  inline void Compress(SiteHalfSpinor *buf,Integer o,const SiteSpinor &in) {
-    SiteHalfSpinor hsp;
-    SiteHalfCommSpinor *hbuf = (SiteHalfCommSpinor *)buf;
-    projector::Proj(hsp,in,mu,dag);
-    precisionChange((vComplexLow *)&hbuf[o],(vComplexHigh *)&hsp,Nw);
-  }
-
-  /*****************************************************/
-  /* Exchange includes precision change if mpi data is not same */
-  /*****************************************************/
-  inline void Exchange(SiteHalfSpinor *mp,
-                       SiteHalfSpinor *vp0,
-                       SiteHalfSpinor *vp1,
-		       Integer type,Integer o){
-    SiteHalfSpinor vt0,vt1;
-    SiteHalfCommSpinor *vpp0 = (SiteHalfCommSpinor *)vp0;
-    SiteHalfCommSpinor *vpp1 = (SiteHalfCommSpinor *)vp1;
-    precisionChange((vComplexHigh *)&vt0,(vComplexLow *)&vpp0[o],Nw);
-    precisionChange((vComplexHigh *)&vt1,(vComplexLow *)&vpp1[o],Nw);
-    exchange(mp[2*o],mp[2*o+1],vt0,vt1,type);
-  }
-
-  /*****************************************************/
-  /* Have a decompression step if mpi data is not same */
-  /*****************************************************/
-  inline void Decompress(SiteHalfSpinor *out,
-			 SiteHalfSpinor *in, Integer o){
-    SiteHalfCommSpinor *hin=(SiteHalfCommSpinor *)in;
-    precisionChange((vComplexHigh *)&out[o],(vComplexLow *)&hin[o],Nw);
-  }
-
-  /*****************************************************/
-  /* Compress Exchange                                 */
-  /*****************************************************/
-  inline void CompressExchange(SiteHalfSpinor *out0,
-			       SiteHalfSpinor *out1,
-			       const SiteSpinor *in,
-			       Integer j,Integer k, Integer m,Integer type){
-    SiteHalfSpinor temp1, temp2,temp3,temp4;
-    SiteHalfCommSpinor *hout0 = (SiteHalfCommSpinor *)out0;
-    SiteHalfCommSpinor *hout1 = (SiteHalfCommSpinor *)out1;
-    projector::Proj(temp1,in[k],mu,dag);
-    projector::Proj(temp2,in[m],mu,dag);
-    exchange(temp3,temp4,temp1,temp2,type);
-    precisionChange((vComplexLow *)&hout0[j],(vComplexHigh *)&temp3,Nw);
-    precisionChange((vComplexLow *)&hout1[j],(vComplexHigh *)&temp4,Nw);
-  }
-
-  /*****************************************************/
-  /* Pass the info to the stencil */
-  /*****************************************************/
-  inline bool DecompressionStep(void) { return true; }
-
-};
-
-#define DECLARE_PROJ(Projector,Compressor,spProj)			\
-  class Projector {							\
-  public:								\
-    template<class hsp,class fsp>					\
-    static void Proj(hsp &result,const fsp &in,int mu,int dag){			\
-      spProj(result,in);						\
-    }									\
-  };									\
-template<typename HCS,typename HS,typename S> using Compressor = WilsonCompressorTemplate<HCS,HS,S,Projector>;
-
-DECLARE_PROJ(WilsonXpProjector,WilsonXpCompressor,spProjXp);
-DECLARE_PROJ(WilsonYpProjector,WilsonYpCompressor,spProjYp);
-DECLARE_PROJ(WilsonZpProjector,WilsonZpCompressor,spProjZp);
-DECLARE_PROJ(WilsonTpProjector,WilsonTpCompressor,spProjTp);
-DECLARE_PROJ(WilsonXmProjector,WilsonXmCompressor,spProjXm);
-DECLARE_PROJ(WilsonYmProjector,WilsonYmCompressor,spProjYm);
-DECLARE_PROJ(WilsonZmProjector,WilsonZmCompressor,spProjZm);
-DECLARE_PROJ(WilsonTmProjector,WilsonTmCompressor,spProjTm);
-
-class WilsonProjector {
- public:
-  template<class hsp,class fsp>
-  static void Proj(hsp &result,const fsp &in,int mu,int dag){
-    int mudag=dag? mu : (mu+Nd)%(2*Nd);
-    switch(mudag) {
-    case Xp:	spProjXp(result,in);	break;
-    case Yp:	spProjYp(result,in);	break;
-    case Zp:	spProjZp(result,in);	break;
-    case Tp:	spProjTp(result,in);	break;
-    case Xm:	spProjXm(result,in);	break;
-    case Ym:	spProjYm(result,in);	break;
-    case Zm:	spProjZm(result,in);	break;
-    case Tm:	spProjTm(result,in);	break;
-    default: 	assert(0);	        break;
-    }
-  }
-};
-template<typename HCS,typename HS,typename S> using WilsonCompressor = WilsonCompressorTemplate<HCS,HS,S,WilsonProjector>;
-
-// Fast comms buffer manipulation which should inline right through (avoid direction
-// dependent logic that prevents inlining
-template<class vobj,class cobj>
-class WilsonStencil : public CartesianStencil<vobj,cobj> {
-public:
-  double timer0;
-  double timer1;
-  double timer2;
-  double timer3;
-  double timer4;
-  double timer5;
-  double timer6;
-  uint64_t callsi;
-  void ZeroCountersi(void)
-  {
-    timer0=0;
-    timer1=0;
-    timer2=0;
-    timer3=0;
-    timer4=0;
-    timer5=0;
-    timer6=0;
-    callsi=0;
-  }
-  void Reporti(int calls)
-  {
-    if ( timer0 ) std::cout << GridLogMessage << " timer0 (HaloGatherOpt) " <<timer0/calls <<std::endl;
-    if ( timer1 ) std::cout << GridLogMessage << " timer1 (Communicate)   " <<timer1/calls <<std::endl;
-    if ( timer2 ) std::cout << GridLogMessage << " timer2 (CommsMerge )   " <<timer2/calls <<std::endl;
-    if ( timer3 ) std::cout << GridLogMessage << " timer3 (commsMergeShm) " <<timer3/calls <<std::endl;
-    if ( timer4 ) std::cout << GridLogMessage << " timer4 " <<timer4 <<std::endl;
-  }
-
-  WilsonStencil(GridBase *grid,
-		int npoints,
-		int checkerboard,
-		const std::vector<int> &directions,
-		const std::vector<int> &distances)  
-    : CartesianStencil<vobj,cobj> (grid,npoints,checkerboard,directions,distances) 
-  { 
-    ZeroCountersi();
-  };
-
-
-  template < class compressor>
-  void HaloExchangeOpt(const Lattice<vobj> &source,compressor &compress) 
-  {
-    std::vector<std::vector<CommsRequest_t> > reqs;
-    this->HaloExchangeOptGather(source,compress);
-    double t1=usecond();
-    // Asynchronous MPI calls multidirectional, Isend etc...
-    //    this->CommunicateBegin(reqs);
-    //    this->CommunicateComplete(reqs);
-    // Non-overlapped directions within a thread. Asynchronous calls except MPI3, threaded up to comm threads ways.
-    this->Communicate();
-    double t2=usecond(); timer1 += t2-t1;
-    this->CommsMerge(compress);
-    double t3=usecond(); timer2 += t3-t2;
-    this->CommsMergeSHM(compress);
-    double t4=usecond(); timer3 += t4-t3;
-  }
-  
-  template <class compressor>
-  void HaloExchangeOptGather(const Lattice<vobj> &source,compressor &compress) 
-  {
-    this->Prepare();
-    double t0=usecond();
-    this->HaloGatherOpt(source,compress);
-    double t1=usecond();
-    timer0 += t1-t0;
-    callsi++;
-  }
-
-  template <class compressor>
-  void HaloGatherOpt(const Lattice<vobj> &source,compressor &compress)
-  {
-    // Strategy. Inherit types from Compressor.
-    // Use types to select the write direction by directon compressor
-    typedef typename compressor::SiteSpinor         SiteSpinor;
-    typedef typename compressor::SiteHalfSpinor     SiteHalfSpinor;
-    typedef typename compressor::SiteHalfCommSpinor SiteHalfCommSpinor;
-
-    this->mpi3synctime_g-=usecond();
-    this->_grid->StencilBarrier();
-    this->mpi3synctime_g+=usecond();
-
-    assert(source._grid==this->_grid);
-    this->halogtime-=usecond();
-    
-    this->u_comm_offset=0;
-      
-    WilsonXpCompressor<SiteHalfCommSpinor,SiteHalfSpinor,SiteSpinor> XpCompress; 
-    WilsonYpCompressor<SiteHalfCommSpinor,SiteHalfSpinor,SiteSpinor> YpCompress; 
-    WilsonZpCompressor<SiteHalfCommSpinor,SiteHalfSpinor,SiteSpinor> ZpCompress; 
-    WilsonTpCompressor<SiteHalfCommSpinor,SiteHalfSpinor,SiteSpinor> TpCompress;
-    WilsonXmCompressor<SiteHalfCommSpinor,SiteHalfSpinor,SiteSpinor> XmCompress; 
-    WilsonYmCompressor<SiteHalfCommSpinor,SiteHalfSpinor,SiteSpinor> YmCompress; 
-    WilsonZmCompressor<SiteHalfCommSpinor,SiteHalfSpinor,SiteSpinor> ZmCompress; 
-    WilsonTmCompressor<SiteHalfCommSpinor,SiteHalfSpinor,SiteSpinor> TmCompress;
-
-    int dag = compress.dag;
-    int face_idx=0;
-    if ( dag ) { 
-      assert(this->same_node[Xp]==this->HaloGatherDir(source,XpCompress,Xp,face_idx));
-      assert(this->same_node[Yp]==this->HaloGatherDir(source,YpCompress,Yp,face_idx));
-      assert(this->same_node[Zp]==this->HaloGatherDir(source,ZpCompress,Zp,face_idx));
-      assert(this->same_node[Tp]==this->HaloGatherDir(source,TpCompress,Tp,face_idx));
-      assert(this->same_node[Xm]==this->HaloGatherDir(source,XmCompress,Xm,face_idx));
-      assert(this->same_node[Ym]==this->HaloGatherDir(source,YmCompress,Ym,face_idx));
-      assert(this->same_node[Zm]==this->HaloGatherDir(source,ZmCompress,Zm,face_idx));
-      assert(this->same_node[Tm]==this->HaloGatherDir(source,TmCompress,Tm,face_idx));
-    } else {
-      assert(this->same_node[Xp]==this->HaloGatherDir(source,XmCompress,Xp,face_idx));
-      assert(this->same_node[Yp]==this->HaloGatherDir(source,YmCompress,Yp,face_idx));
-      assert(this->same_node[Zp]==this->HaloGatherDir(source,ZmCompress,Zp,face_idx));
-      assert(this->same_node[Tp]==this->HaloGatherDir(source,TmCompress,Tp,face_idx));
-      assert(this->same_node[Xm]==this->HaloGatherDir(source,XpCompress,Xm,face_idx));
-      assert(this->same_node[Ym]==this->HaloGatherDir(source,YpCompress,Ym,face_idx));
-      assert(this->same_node[Zm]==this->HaloGatherDir(source,ZpCompress,Zm,face_idx));
-      assert(this->same_node[Tm]==this->HaloGatherDir(source,TpCompress,Tm,face_idx));
-    }
-    this->face_table_computed=1;
-    assert(this->u_comm_offset==this->_unified_buffer_size);
-    this->halogtime+=usecond();
-  }
-
- };
-
-}} // namespace close
-#endif

Grid/qcd/action/fermion/WilsonFermion.cc

@@ -1,562 +0,0 @@
-
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonFermion.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/WilsonFermion.h>
-
-namespace Grid {
-namespace QCD {
-
-const std::vector<int> WilsonFermionStatic::directions({0, 1, 2, 3, 0, 1, 2, 3});
-const std::vector<int> WilsonFermionStatic::displacements({1, 1, 1, 1, -1, -1, -1, -1});
-int WilsonFermionStatic::HandOptDslash;
-
-/////////////////////////////////
-// Constructor and gauge import
-/////////////////////////////////
-
-template <class Impl>
-WilsonFermion<Impl>::WilsonFermion(GaugeField &_Umu, GridCartesian &Fgrid,
-                                   GridRedBlackCartesian &Hgrid, RealD _mass,
-                                   const ImplParams &p,
-                                   const WilsonAnisotropyCoefficients &anis)
-    : Kernels(p),
-      _grid(&Fgrid),
-      _cbgrid(&Hgrid),
-      Stencil(&Fgrid, npoint, Even, directions, displacements),
-      StencilEven(&Hgrid, npoint, Even, directions,displacements),  // source is Even
-      StencilOdd(&Hgrid, npoint, Odd, directions,displacements),  // source is Odd
-      mass(_mass),
-      Lebesgue(_grid),
-      LebesgueEvenOdd(_cbgrid),
-      Umu(&Fgrid),
-      UmuEven(&Hgrid),
-      UmuOdd(&Hgrid),
-      _tmp(&Hgrid),
-      anisotropyCoeff(anis)
-{
-  // Allocate the required comms buffer
-  ImportGauge(_Umu);
-  if  (anisotropyCoeff.isAnisotropic){
-    diag_mass = mass + 1.0 + (Nd-1)*(anisotropyCoeff.nu / anisotropyCoeff.xi_0);
-  } else {
-    diag_mass = 4.0 + mass;
-  }
-
-
-}
-
-template <class Impl>
-void WilsonFermion<Impl>::ImportGauge(const GaugeField &_Umu) {
-  GaugeField HUmu(_Umu._grid);
-
-  //Here multiply the anisotropy coefficients
-  if (anisotropyCoeff.isAnisotropic)
-  {
-
-    for (int mu = 0; mu < Nd; mu++)
-    {
-      GaugeLinkField U_dir = (-0.5)*PeekIndex<LorentzIndex>(_Umu, mu);
-      if (mu != anisotropyCoeff.t_direction)
-        U_dir *= (anisotropyCoeff.nu / anisotropyCoeff.xi_0);
-
-      PokeIndex<LorentzIndex>(HUmu, U_dir, mu);
-    }
-  }
-  else
-  {
-    HUmu = _Umu * (-0.5);
-  }
-  Impl::DoubleStore(GaugeGrid(), Umu, HUmu);
-  pickCheckerboard(Even, UmuEven, Umu);
-  pickCheckerboard(Odd, UmuOdd, Umu);
-}
-
-/////////////////////////////
-// Implement the interface
-/////////////////////////////
-
-template <class Impl>
-RealD WilsonFermion<Impl>::M(const FermionField &in, FermionField &out) {
-  out.checkerboard = in.checkerboard;
-  Dhop(in, out, DaggerNo);
-  return axpy_norm(out, diag_mass, in, out);
-}
-
-template <class Impl>
-RealD WilsonFermion<Impl>::Mdag(const FermionField &in, FermionField &out) {
-  out.checkerboard = in.checkerboard;
-  Dhop(in, out, DaggerYes);
-  return axpy_norm(out, diag_mass, in, out);
-}
-
-template <class Impl>
-void WilsonFermion<Impl>::Meooe(const FermionField &in, FermionField &out) {
-  if (in.checkerboard == Odd) {
-    DhopEO(in, out, DaggerNo);
-  } else {
-    DhopOE(in, out, DaggerNo);
-  }
-}
-
-template <class Impl>
-void WilsonFermion<Impl>::MeooeDag(const FermionField &in, FermionField &out) {
-  if (in.checkerboard == Odd) {
-    DhopEO(in, out, DaggerYes);
-  } else {
-    DhopOE(in, out, DaggerYes);
-  }
-}
-  
-template <class Impl>
-void WilsonFermion<Impl>::Mooee(const FermionField &in, FermionField &out) {
-  out.checkerboard = in.checkerboard;
-  typename FermionField::scalar_type scal(diag_mass);
-  out = scal * in;
-}
-
-template <class Impl>
-void WilsonFermion<Impl>::MooeeDag(const FermionField &in, FermionField &out) {
-  out.checkerboard = in.checkerboard;
-  Mooee(in, out);
-}
-
-template<class Impl>
-void WilsonFermion<Impl>::MooeeInv(const FermionField &in, FermionField &out) {
-  out.checkerboard = in.checkerboard;
-  out = (1.0/(diag_mass))*in;
-}
-  
-template<class Impl>
-void WilsonFermion<Impl>::MooeeInvDag(const FermionField &in, FermionField &out) {
-  out.checkerboard = in.checkerboard;
-  MooeeInv(in,out);
-}
-template<class Impl>
-void WilsonFermion<Impl>::MomentumSpacePropagator(FermionField &out, const FermionField &in,RealD _m,std::vector<double> twist)
-{  
-  typedef typename FermionField::vector_type vector_type;
-  typedef typename FermionField::scalar_type ScalComplex;
-  typedef Lattice<iSinglet<vector_type> > LatComplex;
-  
-  // what type LatticeComplex 
-  conformable(_grid,out._grid);
-  
-  Gamma::Algebra Gmu [] = {
-    Gamma::Algebra::GammaX,
-    Gamma::Algebra::GammaY,
-    Gamma::Algebra::GammaZ,
-    Gamma::Algebra::GammaT
-  };
-  
-  std::vector<int> latt_size   = _grid->_fdimensions;
-  
-  FermionField   num  (_grid); num  = zero;
-  LatComplex    wilson(_grid); wilson= zero;
-  LatComplex     one  (_grid); one = ScalComplex(1.0,0.0);
-  
-  LatComplex denom(_grid); denom= zero;
-  LatComplex kmu(_grid); 
-  ScalComplex ci(0.0,1.0);
-  // momphase = n * 2pi / L
-  for(int mu=0;mu<Nd;mu++) {
-    
-    LatticeCoordinate(kmu,mu);
-    
-    RealD TwoPiL =  M_PI * 2.0/ latt_size[mu];
-    
-    kmu = TwoPiL * kmu;
-    kmu = kmu + TwoPiL * one * twist[mu];//momentum for twisted boundary conditions
-    
-    wilson = wilson + 2.0*sin(kmu*0.5)*sin(kmu*0.5); // Wilson term
-    
-    num = num - sin(kmu)*ci*(Gamma(Gmu[mu])*in);    // derivative term
-    
-    denom=denom + sin(kmu)*sin(kmu);
-  }
-  
-  wilson = wilson + _m;     // 2 sin^2 k/2 + m
-  
-  num   = num + wilson*in;     // -i gmu sin k + 2 sin^2 k/2 + m
-  
-  denom= denom+wilson*wilson; // sin^2 k + (2 sin^2 k/2 + m)^2
-  
-  denom= one/denom;
-  
-  out = num*denom; // [ -i gmu sin k + 2 sin^2 k/2 + m] / [ sin^2 k + (2 sin^2 k/2 + m)^2 ]
-  
-}
-  
-
-///////////////////////////////////
-// Internal
-///////////////////////////////////
-
-template <class Impl>
-void WilsonFermion<Impl>::DerivInternal(StencilImpl &st, DoubledGaugeField &U,
-                                        GaugeField &mat, const FermionField &A,
-                                        const FermionField &B, int dag) {
-  assert((dag == DaggerNo) || (dag == DaggerYes));
-
-  Compressor compressor(dag);
-
-  FermionField Btilde(B._grid);
-  FermionField Atilde(B._grid);
-  Atilde = A;//redundant
-
-  st.HaloExchange(B, compressor);
-
-  for (int mu = 0; mu < Nd; mu++) {
-    ////////////////////////////////////////////////////////////////////////
-    // Flip gamma (1+g)<->(1-g) if dag
-    ////////////////////////////////////////////////////////////////////////
-    int gamma = mu;
-    if (!dag) gamma += Nd;
-
-    ////////////////////////
-    // Call the single hop
-    ////////////////////////
-    parallel_for (int sss = 0; sss < B._grid->oSites(); sss++) {
-      Kernels::DhopDir(st, U, st.CommBuf(), sss, sss, B, Btilde, mu, gamma);
-    }
-
-    //////////////////////////////////////////////////
-    // spin trace outer product
-    //////////////////////////////////////////////////
-    Impl::InsertForce4D(mat, Btilde, Atilde, mu);
-  }
-}
-
-template <class Impl>
-void WilsonFermion<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, mat, U, V, dag);
-}
-
-template <class Impl>
-void WilsonFermion<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); not general, leaving as a comment (Guido)
-  // Motivation: look at the SchurDiff operator
-  
-  assert(V.checkerboard == Even);
-  assert(U.checkerboard == Odd);
-  mat.checkerboard = Odd;
-
-  DerivInternal(StencilEven, UmuOdd, mat, U, V, dag);
-}
-
-template <class Impl>
-void WilsonFermion<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, mat, U, V, dag);
-}
-
-template <class Impl>
-void WilsonFermion<Impl>::Dhop(const FermionField &in, FermionField &out, int dag) {
-  conformable(in._grid, _grid);  // verifies full grid
-  conformable(in._grid, out._grid);
-
-  out.checkerboard = in.checkerboard;
-
-  DhopInternal(Stencil, Lebesgue, Umu, in, out, dag);
-}
-
-template <class Impl>
-void WilsonFermion<Impl>::DhopOE(const FermionField &in, FermionField &out, int dag) {
-  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, in, out, dag);
-}
-
-template <class Impl>
-void WilsonFermion<Impl>::DhopEO(const FermionField &in, FermionField &out,int dag) {
-  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, in, out, dag);
-}
-
-template <class Impl>
-void WilsonFermion<Impl>::Mdir(const FermionField &in, FermionField &out, int dir, int disp) {
-  DhopDir(in, out, dir, disp);
-}
-
-template <class Impl>
-void WilsonFermion<Impl>::DhopDir(const FermionField &in, FermionField &out, int dir, int disp) {
-  int skip = (disp == 1) ? 0 : 1;
-  int dirdisp = dir + skip * 4;
-  int gamma = dir + (1 - skip) * 4;
-
-  DhopDirDisp(in, out, dirdisp, gamma, DaggerNo);
-};
-
-template <class Impl>
-void WilsonFermion<Impl>::DhopDirDisp(const FermionField &in, FermionField &out,int dirdisp, int gamma, int dag) {
-  Compressor compressor(dag);
-
-  Stencil.HaloExchange(in, compressor);
-
-  parallel_for (int sss = 0; sss < in._grid->oSites(); sss++) {
-    Kernels::DhopDir(Stencil, Umu, Stencil.CommBuf(), sss, sss, in, out, dirdisp, gamma);
-  }
-} 
-/*Change starts*/
-template <class Impl>
-void WilsonFermion<Impl>::DhopInternal(StencilImpl &st, LebesgueOrder &lo,
-                                       DoubledGaugeField &U,
-                                       const FermionField &in,
-                                       FermionField &out, int dag) {
-#ifdef GRID_OMP
-  if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsAndCompute )
-    DhopInternalOverlappedComms(st,lo,U,in,out,dag);
-  else
-#endif 
-    DhopInternalSerial(st,lo,U,in,out,dag);
-
-}
-
-template <class Impl>
-void WilsonFermion<Impl>::DhopInternalOverlappedComms(StencilImpl &st, LebesgueOrder &lo,
-                                       DoubledGaugeField &U,
-                                       const FermionField &in,
-                                       FermionField &out, int dag) {
-  assert((dag == DaggerNo) || (dag == DaggerYes));
-#ifdef GRID_OMP
-  Compressor compressor;
-  int len =  U._grid->oSites();
-  const int LLs =  1;
-
-  st.Prepare();
-  st.HaloGather(in,compressor);
-  st.CommsMergeSHM(compressor);
-#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
-     if (dag == DaggerYes) {
-
-        for (int sss = myblock; sss < myblock+myn; ++sss) {
-         Kernels::DhopSiteDag(st, lo, U, st.CommBuf(), sss, sss, 1, 1, in, out);
-       }
-     } else {
-        for (int sss = myblock; sss < myblock+myn; ++sss) {
-         Kernels::DhopSite(st, lo, U, st.CommBuf(), sss, sss, 1, 1, in, out);
-       }
-    } //else
-
-    } else {
-      st.CommunicateThreaded();
-    }
-
-  Compressor compressor(dag);
-
-  if (dag == DaggerYes) {
-    parallel_for (int sss = 0; sss < in._grid->oSites(); sss++) {
-      Kernels::DhopSiteDag(st, lo, U, st.CommBuf(), sss, sss, 1, 1, in, out);
-    }
-  } else {
-    parallel_for (int sss = 0; sss < in._grid->oSites(); sss++) {
-      Kernels::DhopSite(st, lo, U, st.CommBuf(), sss, sss, 1, 1, in, out);
-    }
-  }
-
-  }  //pragma
-#else
-  assert(0);
-#endif
-};
-
-
-template <class Impl>
-void WilsonFermion<Impl>::DhopInternalSerial(StencilImpl &st, LebesgueOrder &lo,
-                                       DoubledGaugeField &U,
-                                       const FermionField &in,
-                                       FermionField &out, int dag) {
-  assert((dag == DaggerNo) || (dag == DaggerYes));
-  Compressor compressor(dag);
-  st.HaloExchange(in, compressor);
-
-  if (dag == DaggerYes) {
-    parallel_for (int sss = 0; sss < in._grid->oSites(); sss++) {
-      Kernels::DhopSiteDag(st, lo, U, st.CommBuf(), sss, sss, 1, 1, in, out);
-    }
-  } else {
-    parallel_for (int sss = 0; sss < in._grid->oSites(); sss++) {
-      Kernels::DhopSite(st, lo, U, st.CommBuf(), sss, sss, 1, 1, in, out);
-    }
-  }
-};
-/*Change ends */
-
-/*******************************************************************************
- * Conserved current utilities for Wilson fermions, for contracting propagators
- * to make a conserved current sink or inserting the conserved current 
- * sequentially.
- ******************************************************************************/
-template <class Impl>
-void WilsonFermion<Impl>::ContractConservedCurrent(PropagatorField &q_in_1,
-                                                   PropagatorField &q_in_2,
-                                                   PropagatorField &q_out,
-                                                   Current curr_type,
-                                                   unsigned int mu)
-{
-    Gamma g5(Gamma::Algebra::Gamma5);
-    conformable(_grid, q_in_1._grid);
-    conformable(_grid, q_in_2._grid);
-    conformable(_grid, q_out._grid);
-    PropagatorField tmp1(_grid), tmp2(_grid);
-    q_out = zero;
-
-    // Forward, need q1(x + mu), q2(x). Backward, need q1(x), q2(x + mu).
-    // Inefficient comms method but not performance critical.
-    tmp1 = Cshift(q_in_1, mu, 1);
-    tmp2 = Cshift(q_in_2, mu, 1);
-    parallel_for (unsigned int sU = 0; sU < Umu._grid->oSites(); ++sU)
-    {
-        Kernels::ContractConservedCurrentSiteFwd(tmp1._odata[sU],
-                                                 q_in_2._odata[sU],
-                                                 q_out._odata[sU],
-                                                 Umu, sU, mu);
-        Kernels::ContractConservedCurrentSiteBwd(q_in_1._odata[sU],
-                                                 tmp2._odata[sU],
-                                                 q_out._odata[sU],
-                                                 Umu, sU, mu);
-    }
-}
-
-
-template <class Impl>
-void WilsonFermion<Impl>::SeqConservedCurrent(PropagatorField &q_in, 
-                                              PropagatorField &q_out,
-                                              Current curr_type,
-                                              unsigned int mu,
-                                              unsigned int tmin, 
-                                              unsigned int tmax,
-					      ComplexField &lattice_cmplx)
-{
-    conformable(_grid, q_in._grid);
-    conformable(_grid, q_out._grid);
-    PropagatorField tmpFwd(_grid), tmpBwd(_grid), tmp(_grid);
-    unsigned int tshift = (mu == Tp) ? 1 : 0;
-    unsigned int LLt    = GridDefaultLatt()[Tp];
-
-    q_out = zero;
-    LatticeInteger coords(_grid);
-    LatticeCoordinate(coords, Tp);
-
-    // Need q(x + mu) and q(x - mu).
-    tmp = Cshift(q_in, mu, 1);
-    tmpFwd = tmp*lattice_cmplx;
-    tmp = lattice_cmplx*q_in;
-    tmpBwd = Cshift(tmp, mu, -1);
-
-    parallel_for (unsigned int sU = 0; sU < Umu._grid->oSites(); ++sU)
-    {
-        // Compute the sequential conserved current insertion only if our simd
-        // object contains a timeslice we need.
-        vInteger t_mask   = ((coords._odata[sU] >= tmin) &&
-                             (coords._odata[sU] <= tmax));
-        Integer timeSlices = Reduce(t_mask);
-
-        if (timeSlices > 0)
-        {
-            Kernels::SeqConservedCurrentSiteFwd(tmpFwd._odata[sU], 
-                                                q_out._odata[sU], 
-                                                Umu, sU, mu, t_mask);
-        }
-
-        // Repeat for backward direction.
-        t_mask     = ((coords._odata[sU] >= (tmin + tshift)) && 
-                      (coords._odata[sU] <= (tmax + tshift)));
-
-	//if tmax = LLt-1 (last timeslice) include timeslice 0 if the time is shifted (mu=3)	
-	unsigned int t0 = 0;
-	if((tmax==LLt-1) && (tshift==1)) t_mask = (t_mask || (coords._odata[sU] == t0 ));
-
-        timeSlices = Reduce(t_mask);
-
-        if (timeSlices > 0)
-        {
-            Kernels::SeqConservedCurrentSiteBwd(tmpBwd._odata[sU], 
-                                                q_out._odata[sU], 
-                                                Umu, sU, mu, t_mask);
-        }
-    }
-
-
-}
-
-FermOpTemplateInstantiate(WilsonFermion);
-AdjointFermOpTemplateInstantiate(WilsonFermion);
-TwoIndexFermOpTemplateInstantiate(WilsonFermion);
-GparityFermOpTemplateInstantiate(WilsonFermion);
-}
-}

Grid/qcd/action/fermion/WilsonFermion.h

@@ -1,200 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonFermion.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_H
-#define GRID_QCD_WILSON_FERMION_H
-
-namespace Grid {
-
-namespace QCD {
-
-class WilsonFermionStatic {
- public:
-  static int HandOptDslash;  // these are a temporary hack
-  static int MortonOrder;
-  static const std::vector<int> directions;
-  static const std::vector<int> displacements;
-  static const int npoint = 8;
-};
-
- struct WilsonAnisotropyCoefficients: Serializable
- {
-  GRID_SERIALIZABLE_CLASS_MEMBERS(WilsonAnisotropyCoefficients,
-  bool, isAnisotropic,
-  int, t_direction,
-  double, xi_0,
-  double, nu);
-
-  WilsonAnisotropyCoefficients():
-    isAnisotropic(false), 
-    t_direction(Nd-1), 
-    xi_0(1.0), 
-    nu(1.0){}
-};
-
-template <class Impl>
-class WilsonFermion : public WilsonKernels<Impl>, public WilsonFermionStatic {
- public:
-  INHERIT_IMPL_TYPES(Impl);
-  typedef WilsonKernels<Impl> Kernels;
-
-  ///////////////////////////////////////////////////////////////
-  // Implement the abstract base
-  ///////////////////////////////////////////////////////////////
-  GridBase *GaugeGrid(void) { return _grid; }
-  GridBase *GaugeRedBlackGrid(void) { return _cbgrid; }
-  GridBase *FermionGrid(void) { return _grid; }
-  GridBase *FermionRedBlackGrid(void) { return _cbgrid; }
-
-  FermionField _tmp;
-  FermionField &tmp(void) { return _tmp; }
-
-  //////////////////////////////////////////////////////////////////
-  // override multiply; cut number routines if pass dagger argument
-  // and also make interface more uniformly consistent
-  //////////////////////////////////////////////////////////////////
-  virtual RealD M(const FermionField &in, FermionField &out);
-  virtual RealD Mdag(const FermionField &in, FermionField &out);
-
-  /////////////////////////////////////////////////////////
-  // half checkerboard operations
-  // could remain virtual so we  can derive Clover from Wilson base
-  /////////////////////////////////////////////////////////
-  void Meooe(const FermionField &in, FermionField &out);
-  void MeooeDag(const FermionField &in, FermionField &out);
-
-  // 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);
-
-  virtual void  MomentumSpacePropagator(FermionField &out,const FermionField &in,RealD _mass,std::vector<double> twist) ;
-
-  ////////////////////////
-  // Derivative interface
-  ////////////////////////
-  // Interface calls an internal routine
-  void DhopDeriv(GaugeField &mat,const FermionField &U,const FermionField &V,int dag);
-  void DhopDerivOE(GaugeField &mat,const FermionField &U,const FermionField &V,int dag);
-  void DhopDerivEO(GaugeField &mat,const FermionField &U,const FermionField &V,int dag);
-
-  ///////////////////////////////////////////////////////////////
-  // 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);
-
-  ///////////////////////////////////////////////////////////////
-  // Multigrid assistance; force term uses too
-  ///////////////////////////////////////////////////////////////
-  void Mdir(const FermionField &in, FermionField &out, int dir, int disp);
-  void DhopDir(const FermionField &in, FermionField &out, int dir, int disp);
-  void DhopDirDisp(const FermionField &in, FermionField &out, int dirdisp,
-                   int gamma, int dag);
-
-  ///////////////////////////////////////////////////////////////
-  // Extra methods added by derived
-  ///////////////////////////////////////////////////////////////
-  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 DhopInternalSerial(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);
-
-  // Constructor
-  WilsonFermion(GaugeField &_Umu, GridCartesian &Fgrid,
-                GridRedBlackCartesian &Hgrid, RealD _mass, 
-                const ImplParams &p = ImplParams(), 
-                const WilsonAnisotropyCoefficients &anis = WilsonAnisotropyCoefficients() );
-  
-  // DoubleStore impl dependent
-  void ImportGauge(const GaugeField &_Umu);
-
-  ///////////////////////////////////////////////////////////////
-  // Data members require to support the functionality
-  ///////////////////////////////////////////////////////////////
-
-  //    protected:
- public:
-  virtual RealD Mass(void) { return mass; }
-  virtual int   isTrivialEE(void) { return 1; };
-  RealD mass;
-  RealD diag_mass;
-
-  GridBase *_grid;
-  GridBase *_cbgrid;
-
-  // 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;
-
-  WilsonAnisotropyCoefficients anisotropyCoeff;
-  
-  ///////////////////////////////////////////////////////////////
-  // 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);
-};
-
-typedef WilsonFermion<WilsonImplF> WilsonFermionF;
-typedef WilsonFermion<WilsonImplD> WilsonFermionD;
-
-
-}
-}
-#endif

Grid/qcd/action/fermion/WilsonKernels.cc

@@ -1,455 +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 */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-
-namespace Grid {
-namespace QCD {
-
-int WilsonKernelsStatic::Opt   = WilsonKernelsStatic::OptGeneric;
-int WilsonKernelsStatic::Comms = WilsonKernelsStatic::CommsAndCompute;
-
-template <class Impl>
-WilsonKernels<Impl>::WilsonKernels(const ImplParams &p) : Base(p){};
-
-////////////////////////////////////////////
-// Generic implementation; move to different file?
-////////////////////////////////////////////
-  
-#define GENERIC_STENCIL_LEG(Dir,spProj,Recon)			\
-  SE = st.GetEntry(ptype, Dir, sF);				\
-  if (SE->_is_local) {						\
-    chi_p = &chi;						\
-    if (SE->_permute) {						\
-      spProj(tmp, in._odata[SE->_offset]);			\
-      permute(chi, tmp, ptype);					\
-    } else {							\
-      spProj(chi, in._odata[SE->_offset]);			\
-    }								\
-  } else {							\
-    chi_p = &buf[SE->_offset];					\
-  }								\
-  Impl::multLink(Uchi, U._odata[sU], *chi_p, Dir, SE, st);	\
-  Recon(result, Uchi);
-  
-#define GENERIC_STENCIL_LEG_INT(Dir,spProj,Recon)		\
-  SE = st.GetEntry(ptype, Dir, sF);				\
-  if (SE->_is_local) {						\
-    chi_p = &chi;						\
-    if (SE->_permute) {						\
-      spProj(tmp, in._odata[SE->_offset]);			\
-      permute(chi, tmp, ptype);					\
-    } else {							\
-      spProj(chi, in._odata[SE->_offset]);			\
-    }								\
-  } else if ( st.same_node[Dir] ) {				\
-      chi_p = &buf[SE->_offset];				\
-  }								\
-  if (SE->_is_local || st.same_node[Dir] ) {			\
-    Impl::multLink(Uchi, U._odata[sU], *chi_p, Dir, SE, st);	\
-    Recon(result, Uchi);					\
-  }
-
-#define GENERIC_STENCIL_LEG_EXT(Dir,spProj,Recon)		\
-  SE = st.GetEntry(ptype, Dir, sF);				\
-  if ((!SE->_is_local) && (!st.same_node[Dir]) ) {		\
-    chi_p = &buf[SE->_offset];					\
-    Impl::multLink(Uchi, U._odata[sU], *chi_p, Dir, SE, st);	\
-    Recon(result, Uchi);					\
-    nmu++;							\
-  }
-
-#define GENERIC_DHOPDIR_LEG(Dir,spProj,Recon)			\
-  if (gamma == Dir) {						\
-    if (SE->_is_local && SE->_permute) {			\
-      spProj(tmp, in._odata[SE->_offset]);			\
-      permute(chi, tmp, ptype);					\
-    } else if (SE->_is_local) {					\
-      spProj(chi, in._odata[SE->_offset]);			\
-    } else {							\
-      chi = buf[SE->_offset];					\
-    }								\
-    Impl::multLink(Uchi, U._odata[sU], chi, dir, SE, st);	\
-    Recon(result, Uchi);					\
-  }
-
-  ////////////////////////////////////////////////////////////////////
-  // All legs kernels ; comms then compute
-  ////////////////////////////////////////////////////////////////////
-template <class Impl>
-void WilsonKernels<Impl>::GenericDhopSiteDag(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
-					     SiteHalfSpinor *buf, int sF,
-					     int sU, const FermionField &in, FermionField &out)
-{
-  SiteHalfSpinor tmp;
-  SiteHalfSpinor chi;
-  SiteHalfSpinor *chi_p;
-  SiteHalfSpinor Uchi;
-  SiteSpinor result;
-  StencilEntry *SE;
-  int ptype;
-
-  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);
-  vstream(out._odata[sF], result);
-};
-
-template <class Impl>
-void WilsonKernels<Impl>::GenericDhopSite(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
-					  SiteHalfSpinor *buf, int sF,
-					  int sU, const FermionField &in, FermionField &out) 
-{
-  SiteHalfSpinor tmp;
-  SiteHalfSpinor chi;
-  SiteHalfSpinor *chi_p;
-  SiteHalfSpinor Uchi;
-  SiteSpinor result;
-  StencilEntry *SE;
-  int ptype;
-
-  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);
-  vstream(out._odata[sF], result);
-};
-  ////////////////////////////////////////////////////////////////////
-  // Interior kernels
-  ////////////////////////////////////////////////////////////////////
-template <class Impl>
-void WilsonKernels<Impl>::GenericDhopSiteDagInt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
-						SiteHalfSpinor *buf, int sF,
-						int sU, const FermionField &in, FermionField &out)
-{
-  SiteHalfSpinor tmp;
-  SiteHalfSpinor chi;
-  SiteHalfSpinor *chi_p;
-  SiteHalfSpinor Uchi;
-  SiteSpinor result;
-  StencilEntry *SE;
-  int ptype;
-
-  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);
-  vstream(out._odata[sF], result);
-};
-
-template <class Impl>
-void WilsonKernels<Impl>::GenericDhopSiteInt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
-					     SiteHalfSpinor *buf, int sF,
-					     int sU, const FermionField &in, FermionField &out) 
-{
-  SiteHalfSpinor tmp;
-  SiteHalfSpinor chi;
-  SiteHalfSpinor *chi_p;
-  SiteHalfSpinor Uchi;
-  SiteSpinor 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);
-  vstream(out._odata[sF], result);
-};
-////////////////////////////////////////////////////////////////////
-// Exterior kernels
-////////////////////////////////////////////////////////////////////
-template <class Impl>
-void WilsonKernels<Impl>::GenericDhopSiteDagExt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
-						SiteHalfSpinor *buf, int sF,
-						int sU, const FermionField &in, FermionField &out)
-{
-  SiteHalfSpinor tmp;
-  SiteHalfSpinor chi;
-  SiteHalfSpinor *chi_p;
-  SiteHalfSpinor Uchi;
-  SiteSpinor result;
-  StencilEntry *SE;
-  int ptype;
-  int nmu=0;
-  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 ) { 
-    out._odata[sF] = out._odata[sF] + result; 
-  }
-};
-
-template <class Impl>
-void WilsonKernels<Impl>::GenericDhopSiteExt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
-					     SiteHalfSpinor *buf, int sF,
-					     int sU, const FermionField &in, FermionField &out) 
-{
-  SiteHalfSpinor tmp;
-  SiteHalfSpinor chi;
-  SiteHalfSpinor *chi_p;
-  SiteHalfSpinor Uchi;
-  SiteSpinor result;
-  StencilEntry *SE;
-  int ptype;
-  int nmu=0;
-  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 ) { 
-    out._odata[sF] = out._odata[sF] + result; 
-  }
-};
-
-template <class Impl>
-void WilsonKernels<Impl>::DhopDir( StencilImpl &st, DoubledGaugeField &U,SiteHalfSpinor *buf, int sF,
-					   int sU, const FermionField &in, FermionField &out, int dir, int gamma) {
-
-  SiteHalfSpinor tmp;
-  SiteHalfSpinor chi;
-  SiteSpinor result;
-  SiteHalfSpinor Uchi;
-  StencilEntry *SE;
-  int ptype;
-
-  SE = st.GetEntry(ptype, dir, sF);
-  GENERIC_DHOPDIR_LEG(Xp,spProjXp,spReconXp);
-  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);
-  vstream(out._odata[sF], result);
-}
-
-/*******************************************************************************
- * 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,
-                                                  DoubledGaugeField &U,
-                                                  unsigned int sU,
-                                                  unsigned int mu,
-                                                  bool switch_sign)
-{
-    SitePropagator result, tmp;
-    Gamma g5(Gamma::Algebra::Gamma5);
-    Impl::multLinkProp(tmp, U._odata[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,
-                                                  DoubledGaugeField &U,
-                                                  unsigned int sU,
-                                                  unsigned int mu,
-                                                  bool switch_sign)
-{
-    SitePropagator result, tmp;
-    Gamma g5(Gamma::Algebra::Gamma5);
-    Impl::multLinkProp(tmp, U._odata[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;
-    }
-}
-
-// G-parity requires more specialised implementation.
-#define NO_CURR_SITE(Impl) \
-template <> \
-void WilsonKernels<Impl>::ContractConservedCurrentSiteFwd( \
-                                                  const SitePropagator &q_in_1, \
-                                                  const SitePropagator &q_in_2, \
-                                                  SitePropagator &q_out,        \
-                                                  DoubledGaugeField &U,         \
-                                                  unsigned int sU,              \
-                                                  unsigned int mu,              \
-                                                  bool switch_sign)             \
-{ \
-    assert(0); \
-} \
-template <> \
-void WilsonKernels<Impl>::ContractConservedCurrentSiteBwd( \
-                                                  const SitePropagator &q_in_1, \
-                                                  const SitePropagator &q_in_2, \
-                                                  SitePropagator &q_out,        \
-                                                  DoubledGaugeField &U,         \
-                                                  unsigned int mu,              \
-                                                  unsigned int sU,              \
-                                                  bool switch_sign)             \
-{ \
-    assert(0); \
-}
-
-NO_CURR_SITE(GparityWilsonImplF);
-NO_CURR_SITE(GparityWilsonImplD);
-NO_CURR_SITE(GparityWilsonImplFH);
-NO_CURR_SITE(GparityWilsonImplDF);
-
-
-/*******************************************************************************
- * 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,
-                                                     DoubledGaugeField &U,
-                                                     unsigned int sU,
-                                                     unsigned int mu,
-                                                     vInteger t_mask,
-                                                     bool switch_sign)
-{
-    SitePropagator result;
-    Impl::multLinkProp(result, U._odata[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,
-                                                     DoubledGaugeField &U,
-                                                     unsigned int sU,
-                                                     unsigned int mu,
-                                                     vInteger t_mask,
-                                                     bool switch_sign)
-{
-    SitePropagator result;
-    Impl::multLinkProp(result, U._odata[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;
-    }
-}
-
-FermOpTemplateInstantiate(WilsonKernels);
-AdjointFermOpTemplateInstantiate(WilsonKernels);
-TwoIndexFermOpTemplateInstantiate(WilsonKernels);
-
-}}
-

Grid/qcd/action/fermion/WilsonKernels.h

@@ -1,281 +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 */
-#ifndef GRID_QCD_DHOP_H
-#define GRID_QCD_DHOP_H
-
-namespace Grid {
-namespace QCD {
-
-  ////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-  // 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:
-
-  template <bool EnableBool = true>
-  typename std::enable_if<Impl::isFundamental==true && Nc == 3 &&EnableBool, void>::type
-  DhopSite(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
-		   int sF, int sU, int Ls, int Ns, const FermionField &in, FermionField &out,int interior=1,int exterior=1) 
-  {
-    bgq_l1p_optimisation(1);
-    switch(Opt) {
-#if defined(AVX512) || defined (QPX)
-    case OptInlineAsm:
-      if(interior&&exterior) WilsonKernels<Impl>::AsmDhopSite   (st,lo,U,buf,sF,sU,Ls,Ns,in,out);
-      else if (interior)     WilsonKernels<Impl>::AsmDhopSiteInt(st,lo,U,buf,sF,sU,Ls,Ns,in,out);
-      else if (exterior)     WilsonKernels<Impl>::AsmDhopSiteExt(st,lo,U,buf,sF,sU,Ls,Ns,in,out);
-      else assert(0);
-      break;
-#endif
-    case OptHandUnroll:
-         for (int site = 0; site < Ns; site++) {
-	   for (int s = 0; s < Ls; s++) {
-	     if(interior&&exterior) WilsonKernels<Impl>::HandDhopSite(st,lo,U,buf,sF,sU,in,out);
-	     else if (interior)     WilsonKernels<Impl>::HandDhopSiteInt(st,lo,U,buf,sF,sU,in,out);
-	     else if (exterior)     WilsonKernels<Impl>::HandDhopSiteExt(st,lo,U,buf,sF,sU,in,out);
-	     sF++;
-	   }
-	   sU++;
-         }
-      break;
-    case OptGeneric:
-         for (int site = 0; site < Ns; site++) {
-	   for (int s = 0; s < Ls; s++) {
-	     if(interior&&exterior) WilsonKernels<Impl>::GenericDhopSite(st,lo,U,buf,sF,sU,in,out);
-	     else if (interior)     WilsonKernels<Impl>::GenericDhopSiteInt(st,lo,U,buf,sF,sU,in,out);
-	     else if (exterior)     WilsonKernels<Impl>::GenericDhopSiteExt(st,lo,U,buf,sF,sU,in,out);
-	     else assert(0);
-	     sF++;
-	   }
-	   sU++;
-       } 
-      break;
-    default:
-      assert(0);
-    }
-    bgq_l1p_optimisation(0);
-  }
-     
-  template <bool EnableBool = true>
-  typename std::enable_if<(Impl::isFundamental==false || (Impl::isFundamental==true && Nc != 3)) && EnableBool, void>::type
-  DhopSite(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
-	   int sF, int sU, int Ls, int Ns, const FermionField &in, FermionField &out,int interior=1,int exterior=1 ) {
-    // no kernel choice  
-    for (int site = 0; site < Ns; site++) {
-      for (int s = 0; s < Ls; s++) {
-	if(interior&&exterior) WilsonKernels<Impl>::GenericDhopSite(st,lo,U,buf,sF,sU,in,out);
-	else if (interior)     WilsonKernels<Impl>::GenericDhopSiteInt(st,lo,U,buf,sF,sU,in,out);
-	else if (exterior)     WilsonKernels<Impl>::GenericDhopSiteExt(st,lo,U,buf,sF,sU,in,out);
-	else assert(0);
-	sF++;
-      }
-      sU++;
-    }
-  }
-     
-  template <bool EnableBool = true>
-  typename std::enable_if<Impl::isFundamental==true && Nc == 3 && EnableBool,void>::type
-  DhopSiteDag(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
-	      int sF, int sU, int Ls, int Ns, const FermionField &in, FermionField &out,int interior=1,int exterior=1) 
-{
-    bgq_l1p_optimisation(1);
-    switch(Opt) {
-#if defined(AVX512) || defined (QPX)
-    case OptInlineAsm:
-      if(interior&&exterior) WilsonKernels<Impl>::AsmDhopSiteDag   (st,lo,U,buf,sF,sU,Ls,Ns,in,out);
-      else if (interior)     WilsonKernels<Impl>::AsmDhopSiteDagInt(st,lo,U,buf,sF,sU,Ls,Ns,in,out);
-      else if (exterior)     WilsonKernels<Impl>::AsmDhopSiteDagExt(st,lo,U,buf,sF,sU,Ls,Ns,in,out);
-      else assert(0);
-      break;
-#endif
-    case OptHandUnroll:
-      for (int site = 0; site < Ns; site++) {
-	for (int s = 0; s < Ls; s++) {
-	  if(interior&&exterior) WilsonKernels<Impl>::HandDhopSiteDag(st,lo,U,buf,sF,sU,in,out);
-	  else if (interior)     WilsonKernels<Impl>::HandDhopSiteDagInt(st,lo,U,buf,sF,sU,in,out);
-	  else if (exterior)     WilsonKernels<Impl>::HandDhopSiteDagExt(st,lo,U,buf,sF,sU,in,out);
-	  else assert(0);
-	  sF++;
-	}
-	sU++;
-      }
-      break;
-    case OptGeneric:
-      for (int site = 0; site < Ns; site++) {
-	for (int s = 0; s < Ls; s++) {
-	  if(interior&&exterior) WilsonKernels<Impl>::GenericDhopSiteDag(st,lo,U,buf,sF,sU,in,out);
-	  else if (interior)     WilsonKernels<Impl>::GenericDhopSiteDagInt(st,lo,U,buf,sF,sU,in,out);
-	  else if (exterior)     WilsonKernels<Impl>::GenericDhopSiteDagExt(st,lo,U,buf,sF,sU,in,out);
-	  else assert(0);
-	  sF++;
-	}
-	sU++;
-      }
-      break;
-    default:
-      assert(0);
-    }
-    bgq_l1p_optimisation(0);
-  }
-
-  template <bool EnableBool = true>
-  typename std::enable_if<(Impl::isFundamental==false || (Impl::isFundamental==true && Nc != 3)) && EnableBool,void>::type
-  DhopSiteDag(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,SiteHalfSpinor * buf,
-		      int sF, int sU, int Ls, int Ns, const FermionField &in, FermionField &out,int interior=1,int exterior=1) {
-
-    for (int site = 0; site < Ns; site++) {
-      for (int s = 0; s < Ls; s++) {
-	if(interior&&exterior) WilsonKernels<Impl>::GenericDhopSiteDag(st,lo,U,buf,sF,sU,in,out);
-	else if (interior)     WilsonKernels<Impl>::GenericDhopSiteDagInt(st,lo,U,buf,sF,sU,in,out);
-	else if (exterior)     WilsonKernels<Impl>::GenericDhopSiteDagExt(st,lo,U,buf,sF,sU,in,out);
-	else assert(0);
-	sF++;
-      }
-      sU++;
-    }
-  }
-
-  void DhopDir(StencilImpl &st, DoubledGaugeField &U,SiteHalfSpinor * buf,
-		       int sF, int sU, const FermionField &in, FermionField &out, int dirdisp, int gamma);
-      
-  //////////////////////////////////////////////////////////////////////////////
-  // Utilities for inserting Wilson conserved current.
-  //////////////////////////////////////////////////////////////////////////////
-  void ContractConservedCurrentSiteFwd(const SitePropagator &q_in_1,
-                                       const SitePropagator &q_in_2,
-                                       SitePropagator &q_out,
-                                       DoubledGaugeField &U,
-                                       unsigned int sU,
-                                       unsigned int mu,
-                                       bool switch_sign = false);
-  void ContractConservedCurrentSiteBwd(const SitePropagator &q_in_1,
-                                       const SitePropagator &q_in_2,
-                                       SitePropagator &q_out,
-                                       DoubledGaugeField &U,
-                                       unsigned int sU,
-                                       unsigned int mu,
-                                       bool switch_sign = false);
-  void SeqConservedCurrentSiteFwd(const SitePropagator &q_in, 
-                                  SitePropagator &q_out,
-                                  DoubledGaugeField &U,
-                                  unsigned int sU,
-                                  unsigned int mu,
-                                  vInteger t_mask,
-                                  bool switch_sign = false);
-  void SeqConservedCurrentSiteBwd(const SitePropagator &q_in,
-                                  SitePropagator &q_out,
-                                  DoubledGaugeField &U,
-                                  unsigned int sU,
-                                  unsigned int mu,
-                                  vInteger t_mask,
-                                  bool switch_sign = false);
-
-private:
-     // Specialised variants
-  void GenericDhopSite(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
-		       int sF, int sU, const FermionField &in, FermionField &out);
-      
-  void GenericDhopSiteDag(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
-			  int sF, int sU, const FermionField &in, FermionField &out);
-
-  void GenericDhopSiteInt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
-			  int sF, int sU, const FermionField &in, FermionField &out);
-      
-  void GenericDhopSiteDagInt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
-			     int sF, int sU, const FermionField &in, FermionField &out);
-
-  void GenericDhopSiteExt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
-			  int sF, int sU, const FermionField &in, FermionField &out);
-      
-  void GenericDhopSiteDagExt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
-			     int sF, int sU, const FermionField &in, FermionField &out);
-
-
-  void AsmDhopSite(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
-		   int sF, int sU, int Ls, int Ns, const FermionField &in,FermionField &out);
-
-  void AsmDhopSiteDag(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
-		      int sF, int sU, int Ls, int Ns, const FermionField &in, FermionField &out);
-
-  void AsmDhopSiteInt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
-		      int sF, int sU, int Ls, int Ns, const FermionField &in,FermionField &out);
-
-  void AsmDhopSiteDagInt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
-			 int sF, int sU, int Ls, int Ns, const FermionField &in, FermionField &out);
-
-  void AsmDhopSiteExt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
-		      int sF, int sU, int Ls, int Ns, const FermionField &in,FermionField &out);
-
-  void AsmDhopSiteDagExt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
-			 int sF, int sU, int Ls, int Ns, const FermionField &in, FermionField &out);
-
-
-  void HandDhopSite(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
-		    int sF, int sU, const FermionField &in, FermionField &out);
-
-  void HandDhopSiteDag(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
-		       int sF, int sU, const FermionField &in, FermionField &out);
-      
-  void HandDhopSiteInt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
-		       int sF, int sU, const FermionField &in, FermionField &out);
-  
-  void HandDhopSiteDagInt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
-			  int sF, int sU, const FermionField &in, FermionField &out);
-  
-  void HandDhopSiteExt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
-		       int sF, int sU, const FermionField &in, FermionField &out);
-  
-  void HandDhopSiteDagExt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
-			  int sF, int sU, const FermionField &in, FermionField &out);
-  
-public:
-
-  WilsonKernels(const ImplParams &p = ImplParams());
-
-};
-    
-}}
-
-#endif