Hack - may revert this in future

.gitignore

@@ -88,7 +88,6 @@ Thumbs.db
 # build directory #
 ###################
 build*/*
-Documentation/_build
 
 # IDE related files #
 #####################

.travis.yml

@@ -0,0 +1,56 @@
+language: cpp
+
+cache:
+  directories:
+    - clang
+
+matrix:
+  include:
+    - os:        osx
+      osx_image: xcode8.3
+      compiler: clang
+      
+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
+    
+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 [[ "$TRAVIS_OS_NAME" == "osx" ]]; then export EXTRACONF='--with-openssl=/usr/local/opt/openssl'; fi
+    
+script:
+    - ./bootstrap.sh
+    - mkdir build
+    - cd build
+    - mkdir lime
+    - cd lime
+    - mkdir build
+    - cd build
+    - wget http://usqcd-software.github.io/downloads/c-lime/lime-1.3.2.tar.gz
+    - tar xf lime-1.3.2.tar.gz
+    - cd lime-1.3.2
+    - ./configure --prefix=$CWD/build/lime/install
+    - make -j4
+    - make install
+    - cd $CWD/build
+    - ../configure --enable-simd=SSE4 --enable-comms=none --with-lime=$CWD/build/lime/install ${EXTRACONF}
+    - make -j4 
+    - ./benchmarks/Benchmark_dwf --threads 1 --debug-signals
+    - make check

Grid/DisableWarnings.h

@@ -37,9 +37,7 @@ directory
 #endif
 
  //disables and intel compiler specific warning (in json.hpp)
-#ifdef __ICC
 #pragma warning disable 488  
-#endif
 
 #ifdef __NVCC__
  //disables nvcc specific warning in json.hpp

Grid/Makefile.am

@@ -54,11 +54,9 @@ Version.h: version-cache
 include Make.inc
 include Eigen.inc
 
+extra_sources+=$(ZWILS_FERMION_FILES)
 extra_sources+=$(WILS_FERMION_FILES)
 extra_sources+=$(STAG_FERMION_FILES)
-if BUILD_ZMOBIUS
-  extra_sources+=$(ZWILS_FERMION_FILES)
-endif
 if BUILD_GPARITY
   extra_sources+=$(GP_FERMION_FILES)
 endif

Grid/algorithms/CoarsenedMatrix.h

@@ -31,7 +31,6 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 #ifndef  GRID_ALGORITHM_COARSENED_MATRIX_H
 #define  GRID_ALGORITHM_COARSENED_MATRIX_H
 
-#include <Grid/qcd/QCD.h> // needed for Dagger(Yes|No), Inverse(Yes|No)
 
 NAMESPACE_BEGIN(Grid);
 
@@ -60,14 +59,12 @@ inline void blockMaskedInnerProduct(Lattice<CComplex> &CoarseInner,
 class Geometry {
 public:
   int npoint;
-  int base;
   std::vector<int> directions   ;
   std::vector<int> displacements;
-  std::vector<int> points_dagger;
 
   Geometry(int _d)  {
     
-    base = (_d==5) ? 1:0;
+    int base = (_d==5) ? 1:0;
 
     // make coarse grid stencil for 4d , not 5d
     if ( _d==5 ) _d=4;
@@ -75,51 +72,16 @@ public:
     npoint = 2*_d+1;
     directions.resize(npoint);
     displacements.resize(npoint);
-    points_dagger.resize(npoint);
     for(int d=0;d<_d;d++){
       directions[d   ] = d+base;
       directions[d+_d] = d+base;
       displacements[d  ] = +1;
       displacements[d+_d]= -1;
-      points_dagger[d   ] = d+_d;
-      points_dagger[d+_d] = d;
     }
     directions   [2*_d]=0;
     displacements[2*_d]=0;
-    points_dagger[2*_d]=2*_d;
   }
 
-  int point(int dir, int disp) {
-    assert(disp == -1 || disp == 0 || disp == 1);
-    assert(base+0 <= dir && dir < base+4);
-
-    // directions faster index = new indexing
-    // 4d (base = 0):
-    // point 0  1  2  3  4  5  6  7  8
-    // dir   0  1  2  3  0  1  2  3  0
-    // disp +1 +1 +1 +1 -1 -1 -1 -1  0
-    // 5d (base = 1):
-    // point 0  1  2  3  4  5  6  7  8
-    // dir   1  2  3  4  1  2  3  4  0
-    // disp +1 +1 +1 +1 -1 -1 -1 -1  0
-
-    // displacements faster index = old indexing
-    // 4d (base = 0):
-    // point 0  1  2  3  4  5  6  7  8
-    // dir   0  0  1  1  2  2  3  3  0
-    // disp +1 -1 +1 -1 +1 -1 +1 -1  0
-    // 5d (base = 1):
-    // point 0  1  2  3  4  5  6  7  8
-    // dir   1  1  2  2  3  3  4  4  0
-    // disp +1 -1 +1 -1 +1 -1 +1 -1  0
-
-    if(dir == 0 and disp == 0)
-      return 8;
-    else // New indexing
-      return (1 - disp) / 2 * 4 + dir - base;
-    // else // Old indexing
-    //   return (4 * (dir - base) + 1 - disp) / 2;
-  }
 };
   
 template<class Fobj,class CComplex,int nbasis>
@@ -296,7 +258,7 @@ public:
 // Fine Object == (per site) type of fine field
 // nbasis      == number of deflation vectors
 template<class Fobj,class CComplex,int nbasis>
-class CoarsenedMatrix : public CheckerBoardedSparseMatrixBase<Lattice<iVector<CComplex,nbasis > > >  {
+class CoarsenedMatrix : public SparseMatrixBase<Lattice<iVector<CComplex,nbasis > > >  {
 public:
     
   typedef iVector<CComplex,nbasis >           siteVector;
@@ -306,59 +268,33 @@ public:
   typedef iMatrix<CComplex,nbasis >  Cobj;
   typedef Lattice< CComplex >   CoarseScalar; // used for inner products on fine field
   typedef Lattice<Fobj >        FineField;
-  typedef CoarseVector FermionField;
-
-  // enrich interface, use default implementation as in FermionOperator ///////
-  void Dminus(CoarseVector const& in, CoarseVector& out) { out = in; }
-  void DminusDag(CoarseVector const& in, CoarseVector& out) { out = in; }
-  void ImportPhysicalFermionSource(CoarseVector const& input, CoarseVector& imported) { imported = input; }
-  void ImportUnphysicalFermion(CoarseVector const& input, CoarseVector& imported) { imported = input; }
-  void ExportPhysicalFermionSolution(CoarseVector const& solution, CoarseVector& exported) { exported = solution; };
-  void ExportPhysicalFermionSource(CoarseVector const& solution, CoarseVector& exported) { exported = solution; };
 
   ////////////////////
   // Data members
   ////////////////////
   Geometry         geom;
   GridBase *       _grid; 
-  GridBase*        _cbgrid;
   int hermitian;
 
   CartesianStencil<siteVector,siteVector,int> Stencil; 
-  CartesianStencil<siteVector,siteVector,int> StencilEven;
-  CartesianStencil<siteVector,siteVector,int> StencilOdd;
 
   std::vector<CoarseMatrix> A;
-  std::vector<CoarseMatrix> Aeven;
-  std::vector<CoarseMatrix> Aodd;
-
-  CoarseMatrix AselfInv;
-  CoarseMatrix AselfInvEven;
-  CoarseMatrix AselfInvOdd;
-
-  Vector<RealD> dag_factor;
-
+    
   ///////////////////////
   // Interface
   ///////////////////////
   GridBase * Grid(void)         { return _grid; };   // this is all the linalg routines need to know
-  GridBase * RedBlackGrid()     { return _cbgrid; };
-
-  int ConstEE() { return 0; }
 
   void M (const CoarseVector &in, CoarseVector &out)
   {
     conformable(_grid,in.Grid());
     conformable(in.Grid(),out.Grid());
-    out.Checkerboard() = in.Checkerboard();
 
     SimpleCompressor<siteVector> compressor;
 
     Stencil.HaloExchange(in,compressor);
     autoView( in_v , in, AcceleratorRead);
     autoView( out_v , out, AcceleratorWrite);
-    autoView( Stencil_v  , Stencil, AcceleratorRead);
-    auto& geom_v = geom;
     typedef LatticeView<Cobj> Aview;
       
     Vector<Aview> AcceleratorViewContainer;
@@ -380,14 +316,14 @@ public:
       int ptype;
       StencilEntry *SE;
 
-      for(int point=0;point<geom_v.npoint;point++){
+      for(int point=0;point<geom.npoint;point++){
 
-	SE=Stencil_v.GetEntry(ptype,point,ss);
+	SE=Stencil.GetEntry(ptype,point,ss);
 	  
 	if(SE->_is_local) { 
 	  nbr = coalescedReadPermute(in_v[SE->_offset],ptype,SE->_permute);
 	} else {
-	  nbr = coalescedRead(Stencil_v.CommBuf()[SE->_offset]);
+	  nbr = coalescedRead(Stencil.CommBuf()[SE->_offset]);
 	}
 	acceleratorSynchronise();
 
@@ -408,74 +344,12 @@ public:
       return M(in,out);
     } else {
       // corresponds to Galerkin coarsening
-      return MdagNonHermitian(in, out);
+      CoarseVector tmp(Grid());
+      G5C(tmp, in); 
+      M(tmp, out);
+      G5C(out, out);
     }
   };
-
-  void MdagNonHermitian(const CoarseVector &in, CoarseVector &out)
-  {
-    conformable(_grid,in.Grid());
-    conformable(in.Grid(),out.Grid());
-    out.Checkerboard() = in.Checkerboard();
-
-    SimpleCompressor<siteVector> compressor;
-
-    Stencil.HaloExchange(in,compressor);
-    autoView( in_v , in, AcceleratorRead);
-    autoView( out_v , out, AcceleratorWrite);
-    autoView( Stencil_v  , Stencil, AcceleratorRead);
-    auto& geom_v = geom;
-    typedef LatticeView<Cobj> Aview;
-
-    Vector<Aview> AcceleratorViewContainer;
-
-    for(int p=0;p<geom.npoint;p++) AcceleratorViewContainer.push_back(A[p].View(AcceleratorRead));
-    Aview *Aview_p = & AcceleratorViewContainer[0];
-
-    const int Nsimd = CComplex::Nsimd();
-    typedef decltype(coalescedRead(in_v[0])) calcVector;
-    typedef decltype(coalescedRead(in_v[0](0))) calcComplex;
-
-    int osites=Grid()->oSites();
-
-    Vector<int> points(geom.npoint, 0);
-    for(int p=0; p<geom.npoint; p++)
-      points[p] = geom.points_dagger[p];
-
-    auto points_p = &points[0];
-
-    RealD* dag_factor_p = &dag_factor[0];
-
-    accelerator_for(sss, Grid()->oSites()*nbasis, Nsimd, {
-      int ss = sss/nbasis;
-      int b  = sss%nbasis;
-      calcComplex res = Zero();
-      calcVector nbr;
-      int ptype;
-      StencilEntry *SE;
-
-      for(int p=0;p<geom_v.npoint;p++){
-        int point = points_p[p];
-
-	SE=Stencil_v.GetEntry(ptype,point,ss);
-
-	if(SE->_is_local) {
-	  nbr = coalescedReadPermute(in_v[SE->_offset],ptype,SE->_permute);
-	} else {
-	  nbr = coalescedRead(Stencil_v.CommBuf()[SE->_offset]);
-	}
-	acceleratorSynchronise();
-
-	for(int bb=0;bb<nbasis;bb++) {
-	  res = res + dag_factor_p[b*nbasis+bb]*coalescedRead(Aview_p[point][ss](b,bb))*nbr(bb);
-	}
-      }
-      coalescedWrite(out_v[ss](b),res);
-      });
-
-    for(int p=0;p<geom.npoint;p++) AcceleratorViewContainer[p].ViewClose();
-  }
-
   void MdirComms(const CoarseVector &in)
   {
     SimpleCompressor<siteVector> compressor;
@@ -485,7 +359,6 @@ public:
   {
     conformable(_grid,in.Grid());
     conformable(_grid,out.Grid());
-    out.Checkerboard() = in.Checkerboard();
 
     typedef LatticeView<Cobj> Aview;
     Vector<Aview> AcceleratorViewContainer;
@@ -494,7 +367,6 @@ public:
 
     autoView( out_v , out, AcceleratorWrite);
     autoView( in_v  , in, AcceleratorRead);
-    autoView( Stencil_v  , Stencil, AcceleratorRead);
 
     const int Nsimd = CComplex::Nsimd();
     typedef decltype(coalescedRead(in_v[0])) calcVector;
@@ -508,12 +380,12 @@ public:
       int ptype;
       StencilEntry *SE;
 
-      SE=Stencil_v.GetEntry(ptype,point,ss);
+      SE=Stencil.GetEntry(ptype,point,ss);
 	  
       if(SE->_is_local) { 
 	nbr = coalescedReadPermute(in_v[SE->_offset],ptype,SE->_permute);
       } else {
-	nbr = coalescedRead(Stencil_v.CommBuf()[SE->_offset]);
+	nbr = coalescedRead(Stencil.CommBuf()[SE->_offset]);
       }
       acceleratorSynchronise();
 
@@ -541,7 +413,34 @@ public:
 
     this->MdirComms(in);
 
-    MdirCalc(in,out,geom.point(dir,disp));
+    int ndim = in.Grid()->Nd();
+
+    //////////////
+    // 4D action like wilson
+    // 0+ => 0 
+    // 0- => 1
+    // 1+ => 2 
+    // 1- => 3
+    // etc..
+    //////////////
+    // 5D action like DWF
+    // 1+ => 0 
+    // 1- => 1
+    // 2+ => 2 
+    // 2- => 3
+    // etc..
+    auto point = [dir, disp, ndim](){
+      if(dir == 0 and disp == 0)
+	return 8;
+      else if ( ndim==4 ) { 
+	return (4 * dir + 1 - disp) / 2;
+      } else { 
+	return (4 * (dir-1) + 1 - disp) / 2;
+      }
+    }();
+
+    MdirCalc(in,out,point);
+
   };
 
   void Mdiag(const CoarseVector &in, CoarseVector &out)
@@ -550,298 +449,23 @@ public:
     MdirCalc(in, out, point); // No comms
   };
 
-  void Mooee(const CoarseVector &in, CoarseVector &out) {
-    MooeeInternal(in, out, DaggerNo, InverseNo);
-  }
-
-  void MooeeInv(const CoarseVector &in, CoarseVector &out) {
-    MooeeInternal(in, out, DaggerNo, InverseYes);
-  }
-
-  void MooeeDag(const CoarseVector &in, CoarseVector &out) {
-    MooeeInternal(in, out, DaggerYes, InverseNo);
-  }
-
-  void MooeeInvDag(const CoarseVector &in, CoarseVector &out) {
-    MooeeInternal(in, out, DaggerYes, InverseYes);
-  }
-
-  void Meooe(const CoarseVector &in, CoarseVector &out) {
-    if(in.Checkerboard() == Odd) {
-      DhopEO(in, out, DaggerNo);
-    } else {
-      DhopOE(in, out, DaggerNo);
-    }
-  }
-
-  void MeooeDag(const CoarseVector &in, CoarseVector &out) {
-    if(in.Checkerboard() == Odd) {
-      DhopEO(in, out, DaggerYes);
-    } else {
-      DhopOE(in, out, DaggerYes);
-    }
-  }
-
-  void Dhop(const CoarseVector &in, CoarseVector &out, int dag) {
-    conformable(in.Grid(), _grid); // verifies full grid
-    conformable(in.Grid(), out.Grid());
-
-    out.Checkerboard() = in.Checkerboard();
-
-    DhopInternal(Stencil, A, in, out, dag);
-  }
-
-  void DhopOE(const CoarseVector &in, CoarseVector &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, Aodd, in, out, dag);
-  }
-
-  void DhopEO(const CoarseVector &in, CoarseVector &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, Aeven, in, out, dag);
-  }
-
-  void MooeeInternal(const CoarseVector &in, CoarseVector &out, int dag, int inv) {
-    out.Checkerboard() = in.Checkerboard();
-    assert(in.Checkerboard() == Odd || in.Checkerboard() == Even);
-
-    CoarseMatrix *Aself = nullptr;
-    if(in.Grid()->_isCheckerBoarded) {
-      if(in.Checkerboard() == Odd) {
-        Aself = (inv) ? &AselfInvOdd : &Aodd[geom.npoint-1];
-        DselfInternal(StencilOdd, *Aself, in, out, dag);
-      } else {
-        Aself = (inv) ? &AselfInvEven : &Aeven[geom.npoint-1];
-        DselfInternal(StencilEven, *Aself, in, out, dag);
-      }
-    } else {
-      Aself = (inv) ? &AselfInv : &A[geom.npoint-1];
-      DselfInternal(Stencil, *Aself, in, out, dag);
-    }
-    assert(Aself != nullptr);
-  }
-
-  void DselfInternal(CartesianStencil<siteVector,siteVector,int> &st, CoarseMatrix &a,
-                       const CoarseVector &in, CoarseVector &out, int dag) {
-    int point = geom.npoint-1;
-    autoView( out_v, out, AcceleratorWrite);
-    autoView( in_v,  in,  AcceleratorRead);
-    autoView( st_v,  st,  AcceleratorRead);
-    autoView( a_v,   a,   AcceleratorRead);
-
-    const int Nsimd = CComplex::Nsimd();
-    typedef decltype(coalescedRead(in_v[0])) calcVector;
-    typedef decltype(coalescedRead(in_v[0](0))) calcComplex;
-
-    RealD* dag_factor_p = &dag_factor[0];
-
-    if(dag) {
-      accelerator_for(sss, in.Grid()->oSites()*nbasis, Nsimd, {
-        int ss = sss/nbasis;
-        int b  = sss%nbasis;
-        calcComplex res = Zero();
-        calcVector nbr;
-        int ptype;
-        StencilEntry *SE;
-
-        SE=st_v.GetEntry(ptype,point,ss);
-
-        if(SE->_is_local) {
-          nbr = coalescedReadPermute(in_v[SE->_offset],ptype,SE->_permute);
-        } else {
-          nbr = coalescedRead(st_v.CommBuf()[SE->_offset]);
-        }
-        acceleratorSynchronise();
-
-        for(int bb=0;bb<nbasis;bb++) {
-          res = res + dag_factor_p[b*nbasis+bb]*coalescedRead(a_v[ss](b,bb))*nbr(bb);
-        }
-        coalescedWrite(out_v[ss](b),res);
-      });
-    } else {
-      accelerator_for(sss, in.Grid()->oSites()*nbasis, Nsimd, {
-        int ss = sss/nbasis;
-        int b  = sss%nbasis;
-        calcComplex res = Zero();
-        calcVector nbr;
-        int ptype;
-        StencilEntry *SE;
-
-        SE=st_v.GetEntry(ptype,point,ss);
-
-        if(SE->_is_local) {
-          nbr = coalescedReadPermute(in_v[SE->_offset],ptype,SE->_permute);
-        } else {
-          nbr = coalescedRead(st_v.CommBuf()[SE->_offset]);
-        }
-        acceleratorSynchronise();
-
-        for(int bb=0;bb<nbasis;bb++) {
-          res = res + coalescedRead(a_v[ss](b,bb))*nbr(bb);
-        }
-        coalescedWrite(out_v[ss](b),res);
-      });
-    }
-  }
-
-  void DhopInternal(CartesianStencil<siteVector,siteVector,int> &st, std::vector<CoarseMatrix> &a,
-                    const CoarseVector &in, CoarseVector &out, int dag) {
-    SimpleCompressor<siteVector> compressor;
-
-    st.HaloExchange(in,compressor);
-    autoView( in_v,  in,  AcceleratorRead);
-    autoView( out_v, out, AcceleratorWrite);
-    autoView( st_v , st,  AcceleratorRead);
-    typedef LatticeView<Cobj> Aview;
-
-    // determine in what order we need the points
-    int npoint = geom.npoint-1;
-    Vector<int> points(npoint, 0);
-    for(int p=0; p<npoint; p++)
-      points[p] = (dag && !hermitian) ? geom.points_dagger[p] : p;
-
-    auto points_p = &points[0];
-
-    Vector<Aview> AcceleratorViewContainer;
-    for(int p=0;p<npoint;p++) AcceleratorViewContainer.push_back(a[p].View(AcceleratorRead));
-    Aview *Aview_p = & AcceleratorViewContainer[0];
-
-    const int Nsimd = CComplex::Nsimd();
-    typedef decltype(coalescedRead(in_v[0])) calcVector;
-    typedef decltype(coalescedRead(in_v[0](0))) calcComplex;
-
-    RealD* dag_factor_p = &dag_factor[0];
-
-    if(dag) {
-      accelerator_for(sss, in.Grid()->oSites()*nbasis, Nsimd, {
-        int ss = sss/nbasis;
-        int b  = sss%nbasis;
-        calcComplex res = Zero();
-        calcVector nbr;
-        int ptype;
-        StencilEntry *SE;
-
-        for(int p=0;p<npoint;p++){
-          int point = points_p[p];
-          SE=st_v.GetEntry(ptype,point,ss);
-
-          if(SE->_is_local) {
-            nbr = coalescedReadPermute(in_v[SE->_offset],ptype,SE->_permute);
-          } else {
-            nbr = coalescedRead(st_v.CommBuf()[SE->_offset]);
-          }
-          acceleratorSynchronise();
-
-          for(int bb=0;bb<nbasis;bb++) {
-            res = res + dag_factor_p[b*nbasis+bb]*coalescedRead(Aview_p[point][ss](b,bb))*nbr(bb);
-          }
-        }
-        coalescedWrite(out_v[ss](b),res);
-      });
-    } else {
-      accelerator_for(sss, in.Grid()->oSites()*nbasis, Nsimd, {
-        int ss = sss/nbasis;
-        int b  = sss%nbasis;
-        calcComplex res = Zero();
-        calcVector nbr;
-        int ptype;
-        StencilEntry *SE;
-
-        for(int p=0;p<npoint;p++){
-          int point = points_p[p];
-          SE=st_v.GetEntry(ptype,point,ss);
-
-          if(SE->_is_local) {
-            nbr = coalescedReadPermute(in_v[SE->_offset],ptype,SE->_permute);
-          } else {
-            nbr = coalescedRead(st_v.CommBuf()[SE->_offset]);
-          }
-          acceleratorSynchronise();
-
-          for(int bb=0;bb<nbasis;bb++) {
-            res = res + coalescedRead(Aview_p[point][ss](b,bb))*nbr(bb);
-          }
-        }
-        coalescedWrite(out_v[ss](b),res);
-      });
-    }
-
-    for(int p=0;p<npoint;p++) AcceleratorViewContainer[p].ViewClose();
-  }
   
-  CoarsenedMatrix(GridCartesian &CoarseGrid, int hermitian_=0) 	:
+ CoarsenedMatrix(GridCartesian &CoarseGrid, int hermitian_=0) 	: 
+
     _grid(&CoarseGrid),
-    _cbgrid(new GridRedBlackCartesian(&CoarseGrid)),
     geom(CoarseGrid._ndimension),
     hermitian(hermitian_),
     Stencil(&CoarseGrid,geom.npoint,Even,geom.directions,geom.displacements,0),
-    StencilEven(_cbgrid,geom.npoint,Even,geom.directions,geom.displacements,0),
-    StencilOdd(_cbgrid,geom.npoint,Odd,geom.directions,geom.displacements,0),
-    A(geom.npoint,&CoarseGrid),
-    Aeven(geom.npoint,_cbgrid),
-    Aodd(geom.npoint,_cbgrid),
-    AselfInv(&CoarseGrid),
-    AselfInvEven(_cbgrid),
-    AselfInvOdd(_cbgrid),
-    dag_factor(nbasis*nbasis)
+      A(geom.npoint,&CoarseGrid)
   {
-    fillFactor();
   };
 
-  CoarsenedMatrix(GridCartesian &CoarseGrid, GridRedBlackCartesian &CoarseRBGrid, int hermitian_=0) 	:
-
-    _grid(&CoarseGrid),
-    _cbgrid(&CoarseRBGrid),
-    geom(CoarseGrid._ndimension),
-    hermitian(hermitian_),
-    Stencil(&CoarseGrid,geom.npoint,Even,geom.directions,geom.displacements,0),
-    StencilEven(&CoarseRBGrid,geom.npoint,Even,geom.directions,geom.displacements,0),
-    StencilOdd(&CoarseRBGrid,geom.npoint,Odd,geom.directions,geom.displacements,0),
-    A(geom.npoint,&CoarseGrid),
-    Aeven(geom.npoint,&CoarseRBGrid),
-    Aodd(geom.npoint,&CoarseRBGrid),
-    AselfInv(&CoarseGrid),
-    AselfInvEven(&CoarseRBGrid),
-    AselfInvOdd(&CoarseRBGrid),
-    dag_factor(nbasis*nbasis)
-  {
-    fillFactor();
-  };
-
-  void fillFactor() {
-    Eigen::MatrixXd dag_factor_eigen = Eigen::MatrixXd::Ones(nbasis, nbasis);
-    if(!hermitian) {
-      const int nb = nbasis/2;
-      dag_factor_eigen.block(0,nb,nb,nb) *= -1.0;
-      dag_factor_eigen.block(nb,0,nb,nb) *= -1.0;
-    }
-
-    // GPU readable prefactor
-    thread_for(i, nbasis*nbasis, {
-      int j = i/nbasis;
-      int k = i%nbasis;
-      dag_factor[i] = dag_factor_eigen(j, k);
-    });
-  }
-
   void CoarsenOperator(GridBase *FineGrid,LinearOperatorBase<Lattice<Fobj> > &linop,
 		       Aggregation<Fobj,CComplex,nbasis> & Subspace)
   {
     typedef Lattice<typename Fobj::tensor_reduced> FineComplexField;
     typedef typename Fobj::scalar_type scalar_type;
 
-    std::cout << GridLogMessage<< "CoarsenMatrix "<< std::endl;
-
     FineComplexField one(FineGrid); one=scalar_type(1.0,0.0);
     FineComplexField zero(FineGrid); zero=scalar_type(0.0,0.0);
 
@@ -872,13 +496,11 @@ public:
 
     CoarseScalar InnerProd(Grid()); 
 
-    std::cout << GridLogMessage<< "CoarsenMatrix Orthog "<< std::endl;
     // Orthogonalise the subblocks over the basis
     blockOrthogonalise(InnerProd,Subspace.subspace);
 
     // Compute the matrix elements of linop between this orthonormal
     // set of vectors.
-    std::cout << GridLogMessage<< "CoarsenMatrix masks "<< std::endl;
     int self_stencil=-1;
     for(int p=0;p<geom.npoint;p++)
     { 
@@ -917,7 +539,7 @@ public:
 
       phi=Subspace.subspace[i];
 
-      std::cout << GridLogMessage<< "CoarsenMatrix vector "<<i << std::endl;
+      //      std::cout << GridLogMessage<< "CoarsenMatrix vector "<<i << std::endl;
       linop.OpDirAll(phi,Mphi_p);
       linop.OpDiag  (phi,Mphi_p[geom.npoint-1]);
 
@@ -946,18 +568,6 @@ public:
 	    autoView( A_self  , A[self_stencil], AcceleratorWrite);
 
 	    accelerator_for(ss, Grid()->oSites(), Fobj::Nsimd(),{ coalescedWrite(A_p[ss](j,i),oZProj_v(ss)); });
-	    if ( hermitian && (disp==-1) ) {
-	      for(int pp=0;pp<geom.npoint;pp++){// Find the opposite link and set <j|A|i> = <i|A|j>*
-		int dirp   = geom.directions[pp];
-		int dispp  = geom.displacements[pp];
-		if ( (dirp==dir) && (dispp==1) ){
-		  auto sft = conjugate(Cshift(oZProj,dir,1));
-		  autoView( sft_v    ,  sft  , AcceleratorWrite);
-		  autoView( A_pp     ,  A[pp], AcceleratorWrite);
-		  accelerator_for(ss, Grid()->oSites(), Fobj::Nsimd(),{ coalescedWrite(A_pp[ss](i,j),sft_v(ss)); });
-		}
-	      }
-	    }
 
 	  }
 	}
@@ -996,54 +606,28 @@ public:
     }
     if(hermitian) {
       std::cout << GridLogMessage << " ForceHermitian, new code "<<std::endl;
+      ForceHermitian();
     }
-
-    InvertSelfStencilLink(); std::cout << GridLogMessage << "Coarse self link inverted" << std::endl;
-    FillHalfCbs(); std::cout << GridLogMessage << "Coarse half checkerboards filled" << std::endl;
   }
 
-  void InvertSelfStencilLink() {
-    std::cout << GridLogDebug << "CoarsenedMatrix::InvertSelfStencilLink" << std::endl;
-    int localVolume = Grid()->lSites();
-
-    typedef typename Cobj::scalar_object scalar_object;
-
-    autoView(Aself_v,    A[geom.npoint-1], CpuRead);
-    autoView(AselfInv_v, AselfInv,         CpuWrite);
-    thread_for(site, localVolume, { // NOTE: Not able to bring this to GPU because of Eigen + peek/poke
-      Eigen::MatrixXcd selfLinkEigen    = Eigen::MatrixXcd::Zero(nbasis, nbasis);
-      Eigen::MatrixXcd selfLinkInvEigen = Eigen::MatrixXcd::Zero(nbasis, nbasis);
-
-      scalar_object selfLink    = Zero();
-      scalar_object selfLinkInv = Zero();
-
-      Coordinate lcoor;
-
-      Grid()->LocalIndexToLocalCoor(site, lcoor);
-      peekLocalSite(selfLink, Aself_v, lcoor);
-
-      for (int i = 0; i < nbasis; ++i)
-        for (int j = 0; j < nbasis; ++j)
-          selfLinkEigen(i, j) = static_cast<ComplexD>(TensorRemove(selfLink(i, j)));
-
-      selfLinkInvEigen = selfLinkEigen.inverse();
-
-      for(int i = 0; i < nbasis; ++i)
-        for(int j = 0; j < nbasis; ++j)
-          selfLinkInv(i, j) = selfLinkInvEigen(i, j);
-
-      pokeLocalSite(selfLinkInv, AselfInv_v, lcoor);
-    });
-  }
-
-  void FillHalfCbs() {
-    std::cout << GridLogDebug << "CoarsenedMatrix::FillHalfCbs" << std::endl;
-    for(int p = 0; p < geom.npoint; ++p) {
-      pickCheckerboard(Even, Aeven[p], A[p]);
-      pickCheckerboard(Odd, Aodd[p], A[p]);
+  void ForceHermitian(void) {
+    CoarseMatrix Diff  (Grid());
+    for(int p=0;p<geom.npoint;p++){
+      int dir   = geom.directions[p];
+      int disp  = geom.displacements[p];
+      if(disp==-1) {
+	// Find the opposite link
+	for(int pp=0;pp<geom.npoint;pp++){
+	  int dirp   = geom.directions[pp];
+	  int dispp  = geom.displacements[pp];
+	  if ( (dirp==dir) && (dispp==1) ){
+	    //	    Diff = adj(Cshift(A[p],dir,1)) - A[pp]; 
+	    //	    std::cout << GridLogMessage<<" Replacing stencil leg "<<pp<<" with leg "<<p<< " diff "<<norm2(Diff) <<std::endl;
+	    A[pp] = adj(Cshift(A[p],dir,1));
+	  }
+	}
+      }
     }
-    pickCheckerboard(Even, AselfInvEven, AselfInv);
-    pickCheckerboard(Odd, AselfInvOdd, AselfInv);
   }
 };
 

Grid/algorithms/FFT.h

@@ -136,7 +136,7 @@ public:
     flops=0;
     usec =0;
     Coordinate layout(Nd,1);
-    sgrid = new GridCartesian(dimensions,layout,processors,*grid);
+    sgrid = new GridCartesian(dimensions,layout,processors);
   };
     
   ~FFT ( void)  {
@@ -182,7 +182,7 @@ public:
     pencil_gd[dim] = G*processors[dim];
       
     // Pencil global vol LxLxGxLxL per node
-    GridCartesian pencil_g(pencil_gd,layout,processors,*vgrid);
+    GridCartesian pencil_g(pencil_gd,layout,processors);
       
     // Construct pencils
     typedef typename vobj::scalar_object sobj;

Grid/algorithms/LinearOperator.h

@@ -530,16 +530,6 @@ public:
 template<class Field> class LinearFunction {
 public:
   virtual void operator() (const Field &in, Field &out) = 0;
-
-  virtual void operator() (const std::vector<Field> &in, std::vector<Field> &out)
-  {
-    assert(in.size() == out.size());
-
-    for (unsigned int i = 0; i < in.size(); ++i)
-    {
-      (*this)(in[i], out[i]);
-    }
-  }
 };
 
 template<class Field> class IdentityLinearFunction : public LinearFunction<Field> {

Grid/algorithms/iterative/Deflation.h

@@ -54,23 +54,15 @@ class DeflatedGuesser: public LinearFunction<Field> {
 private:
   const std::vector<Field> &evec;
   const std::vector<RealD> &eval;
-  const unsigned int       N;
 
 public:
 
-  DeflatedGuesser(const std::vector<Field> & _evec,const std::vector<RealD> & _eval)
-  : DeflatedGuesser(_evec, _eval, _evec.size())
-  {}
-
-  DeflatedGuesser(const std::vector<Field> & _evec, const std::vector<RealD> & _eval, const unsigned int _N)
-  : evec(_evec), eval(_eval), N(_N)
-  {
-    assert(evec.size()==eval.size());
-    assert(N <= evec.size());
-  } 
+  DeflatedGuesser(const std::vector<Field> & _evec,const std::vector<RealD> & _eval) : evec(_evec), eval(_eval) {};
 
   virtual void operator()(const Field &src,Field &guess) {
     guess = Zero();
+    assert(evec.size()==eval.size());
+    auto N = evec.size();
     for (int i=0;i<N;i++) {
       const Field& tmp = evec[i];
       axpy(guess,TensorRemove(innerProduct(tmp,src)) / eval[i],tmp,guess);

Grid/algorithms/iterative/SchurRedBlack.h

@@ -132,31 +132,6 @@ namespace Grid {
       (*this)(_Matrix,in,out,guess);
     }
 
-    void RedBlackSource(Matrix &_Matrix, const std::vector<Field> &in, std::vector<Field> &src_o) 
-    {
-      GridBase *grid = _Matrix.RedBlackGrid();
-      Field tmp(grid);
-      int nblock = in.size();
-      for(int b=0;b<nblock;b++){
-	RedBlackSource(_Matrix,in[b],tmp,src_o[b]);
-      }
-    }
-    // James can write his own deflated guesser
-    // with optimised code for the inner products
-    //    RedBlackSolveSplitGrid();
-    //    RedBlackSolve(_Matrix,src_o,sol_o); 
-
-    void RedBlackSolution(Matrix &_Matrix, const std::vector<Field> &in, const std::vector<Field> &sol_o, std::vector<Field> &out)
-    {
-      GridBase *grid = _Matrix.RedBlackGrid();
-      Field tmp(grid);
-      int nblock = in.size();
-      for(int b=0;b<nblock;b++) {
-	pickCheckerboard(Even,tmp,in[b]);
-	RedBlackSolution(_Matrix,sol_o[b],tmp,out[b]);
-      }
-    }
-
     template<class Guesser>
     void operator()(Matrix &_Matrix, const std::vector<Field> &in, std::vector<Field> &out,Guesser &guess) 
     {
@@ -175,29 +150,24 @@ namespace Grid {
       ////////////////////////////////////////////////
       // Prepare RedBlack source
       ////////////////////////////////////////////////
-      RedBlackSource(_Matrix,in,src_o);
-	//      for(int b=0;b<nblock;b++){
-	//	RedBlackSource(_Matrix,in[b],tmp,src_o[b]);
-	//      }
-      
+      for(int b=0;b<nblock;b++){
+	RedBlackSource(_Matrix,in[b],tmp,src_o[b]);
+      }
       ////////////////////////////////////////////////
       // Make the guesses
       ////////////////////////////////////////////////
       if ( subGuess ) guess_save.resize(nblock,grid);
 
-      
-      if(useSolnAsInitGuess) {
-        for(int b=0;b<nblock;b++){
+      for(int b=0;b<nblock;b++){
+        if(useSolnAsInitGuess) {
           pickCheckerboard(Odd, sol_o[b], out[b]);
+        } else {
+          guess(src_o[b],sol_o[b]); 
         }
-      } else {
-        guess(src_o, sol_o); 
-      }
 
-	    if ( subGuess ) { 
-        for(int b=0;b<nblock;b++){
-          guess_save[b] = sol_o[b];
-        }
+	if ( subGuess ) { 
+	  guess_save[b] = sol_o[b];
+	}
       }
       //////////////////////////////////////////////////////////////
       // Call the block solver

Grid/allocator/AlignedAllocator.h

@@ -173,8 +173,7 @@ template<class T> using cshiftAllocator = devAllocator<T>;
 template<class T> using cshiftAllocator = std::allocator<T>;
 #endif
 
-template<class T> using Vector        = std::vector<T,uvmAllocator<T> >;           
-template<class T> using stencilVector = std::vector<T,alignedAllocator<T> >;           
+template<class T> using Vector     = std::vector<T,uvmAllocator<T> >;           
 template<class T> using commVector = std::vector<T,devAllocator<T> >;
 template<class T> using cshiftVector = std::vector<T,cshiftAllocator<T> >;
 

Grid/allocator/MemoryManager.h

@@ -34,6 +34,8 @@ NAMESPACE_BEGIN(Grid);
 
 // Move control to configure.ac and Config.h?
 
+#define ALLOCATION_CACHE
+#define GRID_ALLOC_ALIGN (2*1024*1024)
 #define GRID_ALLOC_SMALL_LIMIT (4096)
 
 /*Pinning pages is costly*/

Grid/allocator/MemoryManagerCache.cc

@@ -1,12 +1,11 @@
 #include <Grid/GridCore.h>
+
 #ifndef GRID_UVM
 
 #warning "Using explicit device memory copies"
 NAMESPACE_BEGIN(Grid);
-//define dprintf(...) printf ( __VA_ARGS__ ); fflush(stdout);
 #define dprintf(...)
 
-
 ////////////////////////////////////////////////////////////
 // For caching copies of data on device
 ////////////////////////////////////////////////////////////
@@ -104,7 +103,7 @@ void MemoryManager::AccDiscard(AcceleratorViewEntry &AccCache)
   ///////////////////////////////////////////////////////////
   assert(AccCache.state!=Empty);
   
-   dprintf("MemoryManager: Discard(%llx) %llx\n",(uint64_t)AccCache.CpuPtr,(uint64_t)AccCache.AccPtr); 
+  //  dprintf("MemoryManager: Discard(%llx) %llx\n",(uint64_t)AccCache.CpuPtr,(uint64_t)AccCache.AccPtr); 
   assert(AccCache.accLock==0);
   assert(AccCache.cpuLock==0);
   assert(AccCache.CpuPtr!=(uint64_t)NULL);
@@ -112,7 +111,7 @@ void MemoryManager::AccDiscard(AcceleratorViewEntry &AccCache)
     AcceleratorFree((void *)AccCache.AccPtr,AccCache.bytes);
     DeviceBytes   -=AccCache.bytes;
     LRUremove(AccCache);
-    dprintf("MemoryManager: Free(%llx) LRU %lld Total %lld\n",(uint64_t)AccCache.AccPtr,DeviceLRUBytes,DeviceBytes);  
+    //    dprintf("MemoryManager: Free(%llx) LRU %lld Total %lld\n",(uint64_t)AccCache.AccPtr,DeviceLRUBytes,DeviceBytes);  
   }
   uint64_t CpuPtr = AccCache.CpuPtr;
   EntryErase(CpuPtr);
@@ -126,7 +125,7 @@ void MemoryManager::Evict(AcceleratorViewEntry &AccCache)
   ///////////////////////////////////////////////////////////////////////////
   assert(AccCache.state!=Empty);
   
-  dprintf("MemoryManager: Evict(%llx) %llx\n",(uint64_t)AccCache.CpuPtr,(uint64_t)AccCache.AccPtr); 
+  //  dprintf("MemoryManager: Evict(%llx) %llx\n",(uint64_t)AccCache.CpuPtr,(uint64_t)AccCache.AccPtr); 
   assert(AccCache.accLock==0);
   assert(AccCache.cpuLock==0);
   if(AccCache.state==AccDirty) {
@@ -137,7 +136,7 @@ void MemoryManager::Evict(AcceleratorViewEntry &AccCache)
     AcceleratorFree((void *)AccCache.AccPtr,AccCache.bytes);
     DeviceBytes   -=AccCache.bytes;
     LRUremove(AccCache);
-    dprintf("MemoryManager: Free(%llx) footprint now %lld \n",(uint64_t)AccCache.AccPtr,DeviceBytes);  
+    //    dprintf("MemoryManager: Free(%llx) footprint now %lld \n",(uint64_t)AccCache.AccPtr,DeviceBytes);  
   }
   uint64_t CpuPtr = AccCache.CpuPtr;
   EntryErase(CpuPtr);
@@ -150,7 +149,7 @@ void MemoryManager::Flush(AcceleratorViewEntry &AccCache)
   assert(AccCache.AccPtr!=(uint64_t)NULL);
   assert(AccCache.CpuPtr!=(uint64_t)NULL);
   acceleratorCopyFromDevice((void *)AccCache.AccPtr,(void *)AccCache.CpuPtr,AccCache.bytes);
-  dprintf("MemoryManager: Flush  %llx -> %llx\n",(uint64_t)AccCache.AccPtr,(uint64_t)AccCache.CpuPtr); fflush(stdout);
+  //  dprintf("MemoryManager: Flush  %llx -> %llx\n",(uint64_t)AccCache.AccPtr,(uint64_t)AccCache.CpuPtr); fflush(stdout);
   DeviceToHostBytes+=AccCache.bytes;
   DeviceToHostXfer++;
   AccCache.state=Consistent;
@@ -165,7 +164,7 @@ void MemoryManager::Clone(AcceleratorViewEntry &AccCache)
     AccCache.AccPtr=(uint64_t)AcceleratorAllocate(AccCache.bytes);
     DeviceBytes+=AccCache.bytes;
   }
-  dprintf("MemoryManager: Clone %llx <- %llx\n",(uint64_t)AccCache.AccPtr,(uint64_t)AccCache.CpuPtr); fflush(stdout);
+  //  dprintf("MemoryManager: Clone %llx <- %llx\n",(uint64_t)AccCache.AccPtr,(uint64_t)AccCache.CpuPtr); fflush(stdout);
   acceleratorCopyToDevice((void *)AccCache.CpuPtr,(void *)AccCache.AccPtr,AccCache.bytes);
   HostToDeviceBytes+=AccCache.bytes;
   HostToDeviceXfer++;
@@ -228,24 +227,18 @@ uint64_t MemoryManager::AcceleratorViewOpen(uint64_t CpuPtr,size_t bytes,ViewMod
   // Find if present, otherwise get or force an empty
   ////////////////////////////////////////////////////////////////////////////
   if ( EntryPresent(CpuPtr)==0 ){
+    EvictVictims(bytes);
     EntryCreate(CpuPtr,bytes,mode,hint);
   }
 
   auto AccCacheIterator = EntryLookup(CpuPtr);
   auto & AccCache = AccCacheIterator->second;
-  if (!AccCache.AccPtr) {
-    EvictVictims(bytes); 
-  } 
+  
   assert((mode==AcceleratorRead)||(mode==AcceleratorWrite)||(mode==AcceleratorWriteDiscard));
 
   assert(AccCache.cpuLock==0);  // Programming error
 
   if(AccCache.state!=Empty) {
-    dprintf("ViewOpen found entry %llx %llx : %lld %lld\n",
-		    (uint64_t)AccCache.CpuPtr,
-		    (uint64_t)CpuPtr,
-		    (uint64_t)AccCache.bytes,
-		    (uint64_t)bytes);
     assert(AccCache.CpuPtr == CpuPtr);
     assert(AccCache.bytes  ==bytes);
   }
@@ -292,21 +285,21 @@ uint64_t MemoryManager::AcceleratorViewOpen(uint64_t CpuPtr,size_t bytes,ViewMod
       AccCache.state  = Consistent; // CpuDirty + AccRead => Consistent
     }
     AccCache.accLock++;
-    dprintf("Copied CpuDirty entry into device accLock %d\n",AccCache.accLock);
+    //    printf("Copied CpuDirty entry into device accLock %d\n",AccCache.accLock);
   } else if(AccCache.state==Consistent) {
     if((mode==AcceleratorWrite)||(mode==AcceleratorWriteDiscard))
       AccCache.state  = AccDirty;   // Consistent + AcceleratorWrite=> AccDirty
     else
       AccCache.state  = Consistent; // Consistent + AccRead => Consistent
     AccCache.accLock++;
-    dprintf("Consistent entry into device accLock %d\n",AccCache.accLock);
+    //    printf("Consistent entry into device accLock %d\n",AccCache.accLock);
   } else if(AccCache.state==AccDirty) {
     if((mode==AcceleratorWrite)||(mode==AcceleratorWriteDiscard))
       AccCache.state  = AccDirty; // AccDirty + AcceleratorWrite=> AccDirty
     else
       AccCache.state  = AccDirty; // AccDirty + AccRead => AccDirty
     AccCache.accLock++;
-    dprintf("AccDirty entry into device accLock %d\n",AccCache.accLock);
+    //    printf("AccDirty entry into device accLock %d\n",AccCache.accLock);
   } else {
     assert(0);
   }
@@ -368,16 +361,13 @@ uint64_t MemoryManager::CpuViewOpen(uint64_t CpuPtr,size_t bytes,ViewMode mode,V
   // Find if present, otherwise get or force an empty
   ////////////////////////////////////////////////////////////////////////////
   if ( EntryPresent(CpuPtr)==0 ){
+    EvictVictims(bytes);
     EntryCreate(CpuPtr,bytes,mode,transient);
   }
 
   auto AccCacheIterator = EntryLookup(CpuPtr);
   auto & AccCache = AccCacheIterator->second;
-
-  if (!AccCache.AccPtr) {
-     EvictVictims(bytes);
-  }
-
+  
   assert((mode==CpuRead)||(mode==CpuWrite));
   assert(AccCache.accLock==0);  // Programming error
 

Grid/allocator/MemoryManagerShared.cc

@@ -1,6 +1,7 @@
 #include <Grid/GridCore.h>
 #ifdef GRID_UVM
 
+#warning "Grid is assuming unified virtual memory address space"
 NAMESPACE_BEGIN(Grid);
 /////////////////////////////////////////////////////////////////////////////////
 // View management is 1:1 address space mapping

Grid/cartesian/Cartesian_red_black.h

@@ -36,7 +36,7 @@ static const int CbBlack=1;
 static const int Even   =CbRed;
 static const int Odd    =CbBlack;
 
-accelerator_inline int RedBlackCheckerBoardFromOindex (int oindex,const Coordinate &rdim,const Coordinate &chk_dim_msk)
+accelerator_inline int RedBlackCheckerBoardFromOindex (int oindex, Coordinate &rdim, Coordinate &chk_dim_msk)
 {
   int nd=rdim.size();
   Coordinate coor(nd);

Grid/communicator/Communicator_base.cc

@@ -33,8 +33,6 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 
 NAMESPACE_BEGIN(Grid);
 
-bool Stencil_force_mpi = true;
-
 ///////////////////////////////////////////////////////////////
 // Info that is setup once and indept of cartesian layout
 ///////////////////////////////////////////////////////////////

Grid/communicator/Communicator_base.h

@@ -1,3 +1,4 @@
+
 /*************************************************************************************
 
     Grid physics library, www.github.com/paboyle/Grid 
@@ -35,8 +36,6 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 
 NAMESPACE_BEGIN(Grid);
 
-extern bool Stencil_force_mpi ;
-
 class CartesianCommunicator : public SharedMemory {
 
 public:    
@@ -109,8 +108,6 @@ public:
   ////////////////////////////////////////////////////////////
   // Reduction
   ////////////////////////////////////////////////////////////
-  void GlobalMax(RealD &);
-  void GlobalMax(RealF &);
   void GlobalSum(RealF &);
   void GlobalSumVector(RealF *,int N);
   void GlobalSum(RealD &);

Grid/communicator/Communicator_mpi3.cc

@@ -275,16 +275,6 @@ 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::GlobalMax(float &f)
-{
-  int ierr=MPI_Allreduce(MPI_IN_PLACE,&f,1,MPI_FLOAT,MPI_MAX,communicator);
-  assert(ierr==0);
-}
-void CartesianCommunicator::GlobalMax(double &d)
-{
-  int ierr = MPI_Allreduce(MPI_IN_PLACE,&d,1,MPI_DOUBLE,MPI_MAX,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);
@@ -370,7 +360,7 @@ double CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsReques
   double off_node_bytes=0.0;
   int tag;
 
-  if ( (gfrom ==MPI_UNDEFINED) || Stencil_force_mpi ) {
+  if ( gfrom ==MPI_UNDEFINED) {
     tag= dir+from*32;
     ierr=MPI_Irecv(recv, bytes, MPI_CHAR,from,tag,communicator_halo[commdir],&rrq);
     assert(ierr==0);
@@ -378,18 +368,12 @@ double CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsReques
     off_node_bytes+=bytes;
   }
 
-  if ( (gdest == MPI_UNDEFINED) || Stencil_force_mpi ) {
+  if ( gdest == MPI_UNDEFINED ) {
     tag= dir+_processor*32;
     ierr =MPI_Isend(xmit, bytes, MPI_CHAR,dest,tag,communicator_halo[commdir],&xrq);
     assert(ierr==0);
     list.push_back(xrq);
     off_node_bytes+=bytes;
-  } else {
-    // TODO : make a OMP loop on CPU, call threaded bcopy
-    void *shm = (void *) this->ShmBufferTranslate(dest,recv);
-    assert(shm!=NULL);
-    acceleratorCopyDeviceToDeviceAsynch(xmit,shm,bytes);
-    acceleratorCopySynchronise(); // MPI prob slower
   }
 
   if ( CommunicatorPolicy == CommunicatorPolicySequential ) {

Grid/communicator/Communicator_none.cc

@@ -67,8 +67,6 @@ CartesianCommunicator::CartesianCommunicator(const Coordinate &processors)
 
 CartesianCommunicator::~CartesianCommunicator(){}
 
-void CartesianCommunicator::GlobalMax(float &){}
-void CartesianCommunicator::GlobalMax(double &){}
 void CartesianCommunicator::GlobalSum(float &){}
 void CartesianCommunicator::GlobalSumVector(float *,int N){}
 void CartesianCommunicator::GlobalSum(double &){}

Grid/communicator/SharedMemory.h

@@ -102,7 +102,7 @@ public:
   ///////////////////////////////////////////////////
   static void SharedMemoryAllocate(uint64_t bytes, int flags);
   static void SharedMemoryFree(void);
-  static void SharedMemoryCopy(void *dest,void *src,size_t bytes);
+  static void SharedMemoryCopy(void *dest,const void *src,size_t bytes);
   static void SharedMemoryZero(void *dest,size_t bytes);
 
 };

Grid/communicator/SharedMemoryMPI.cc

@@ -7,7 +7,6 @@
     Copyright (C) 2015
 
 Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Christoph Lehner <christoph@lhnr.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
@@ -35,9 +34,6 @@ Author: Christoph Lehner <christoph@lhnr.de>
 #endif
 #ifdef GRID_HIP
 #include <hip/hip_runtime_api.h>
-#endif
-#ifdef GRID_SYCl
-
 #endif
 
 NAMESPACE_BEGIN(Grid); 
@@ -73,7 +69,6 @@ void GlobalSharedMemory::Init(Grid_MPI_Comm comm)
   WorldNodes = WorldSize/WorldShmSize;
   assert( (WorldNodes * WorldShmSize) == WorldSize );
 
-
   // FIXME: Check all WorldShmSize are the same ?
 
   /////////////////////////////////////////////////////////////////////
@@ -174,23 +169,6 @@ static inline int divides(int a,int b)
 }
 void GlobalSharedMemory::GetShmDims(const Coordinate &WorldDims,Coordinate &ShmDims)
 {
-  ////////////////////////////////////////////////////////////////
-  // Allow user to configure through environment variable
-  ////////////////////////////////////////////////////////////////
-  char* str = getenv(("GRID_SHM_DIMS_" + std::to_string(ShmDims.size())).c_str());
-  if ( str ) {
-    std::vector<int> IntShmDims;
-    GridCmdOptionIntVector(std::string(str),IntShmDims);
-    assert(IntShmDims.size() == WorldDims.size());
-    long ShmSize = 1;
-    for (int dim=0;dim<WorldDims.size();dim++) {
-      ShmSize *= (ShmDims[dim] = IntShmDims[dim]);
-      assert(divides(ShmDims[dim],WorldDims[dim]));
-    }
-    assert(ShmSize == WorldShmSize);
-    return;
-  }
-  
   ////////////////////////////////////////////////////////////////
   // Powers of 2,3,5 only in prime decomposition for now
   ////////////////////////////////////////////////////////////////
@@ -450,47 +428,7 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
 ////////////////////////////////////////////////////////////////////////////////////////////
 // Hugetlbfs mapping intended
 ////////////////////////////////////////////////////////////////////////////////////////////
-#if defined(GRID_CUDA) ||defined(GRID_HIP)  || defined(GRID_SYCL)
-
-//if defined(GRID_SYCL)
-#if 0
-void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
-{
-  void * ShmCommBuf ; 
-  assert(_ShmSetup==1);
-  assert(_ShmAlloc==0);
-
-  //////////////////////////////////////////////////////////////////////////////////////////////////////////
-  // allocate the pointer array for shared windows for our group
-  //////////////////////////////////////////////////////////////////////////////////////////////////////////
-  MPI_Barrier(WorldShmComm);
-  WorldShmCommBufs.resize(WorldShmSize);
-
-  ///////////////////////////////////////////////////////////////////////////////////////////////////////////
-  // Each MPI rank should allocate our own buffer
-  ///////////////////////////////////////////////////////////////////////////////////////////////////////////
-  ShmCommBuf = acceleratorAllocDevice(bytes);
-
-  if (ShmCommBuf == (void *)NULL ) {
-    std::cerr << " SharedMemoryMPI.cc acceleratorAllocDevice failed NULL pointer for " << bytes<<" bytes " << std::endl;
-    exit(EXIT_FAILURE);  
-  }
-
-  std::cout << WorldRank << header " SharedMemoryMPI.cc acceleratorAllocDevice "<< bytes 
-	    << "bytes at "<< std::hex<< ShmCommBuf <<std::dec<<" for comms buffers " <<std::endl;
-
-  SharedMemoryZero(ShmCommBuf,bytes);
-
-  assert(WorldShmSize == 1);
-  for(int r=0;r<WorldShmSize;r++){
-    WorldShmCommBufs[r] = ShmCommBuf;
-  }
-  _ShmAllocBytes=bytes;
-  _ShmAlloc=1;
-}
-#endif
-
-#if defined(GRID_CUDA) ||defined(GRID_HIP) ||defined(GRID_SYCL)  
+#if defined(GRID_CUDA) ||defined(GRID_HIP)
 void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
 {
   void * ShmCommBuf ; 
@@ -514,16 +452,18 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
   // Each MPI rank should allocate our own buffer
   ///////////////////////////////////////////////////////////////////////////////////////////////////////////
   ShmCommBuf = acceleratorAllocDevice(bytes);
+
   if (ShmCommBuf == (void *)NULL ) {
     std::cerr << " SharedMemoryMPI.cc acceleratorAllocDevice failed NULL pointer for " << bytes<<" bytes " << std::endl;
     exit(EXIT_FAILURE);  
   }
-  if ( WorldRank == 0 ){
+  //  if ( WorldRank == 0 ){
+  if ( 1 ){
     std::cout << WorldRank << header " SharedMemoryMPI.cc acceleratorAllocDevice "<< bytes 
 	      << "bytes at "<< std::hex<< ShmCommBuf <<std::dec<<" for comms buffers " <<std::endl;
   }
   SharedMemoryZero(ShmCommBuf,bytes);
-  std::cout<< "Setting up IPC"<<std::endl;
+
   ///////////////////////////////////////////////////////////////////////////////////////////////////////////
   // Loop over ranks/gpu's on our node
   ///////////////////////////////////////////////////////////////////////////////////////////////////////////
@@ -533,29 +473,6 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
     //////////////////////////////////////////////////
     // If it is me, pass around the IPC access key
     //////////////////////////////////////////////////
-    void * thisBuf = ShmCommBuf;
-    if(!Stencil_force_mpi) {
-#ifdef GRID_SYCL_LEVEL_ZERO_IPC
-    typedef struct { int fd; pid_t pid ; } clone_mem_t;
-
-    auto zeDevice    = cl::sycl::get_native<cl::sycl::backend::level_zero>(theGridAccelerator->get_device());
-    auto zeContext   = cl::sycl::get_native<cl::sycl::backend::level_zero>(theGridAccelerator->get_context());
-      
-    ze_ipc_mem_handle_t ihandle;
-    clone_mem_t handle;
-
-    if ( r==WorldShmRank ) { 
-      auto err = zeMemGetIpcHandle(zeContext,ShmCommBuf,&ihandle);
-      if ( err != ZE_RESULT_SUCCESS ) {
-	std::cout << "SharedMemoryMPI.cc zeMemGetIpcHandle failed for rank "<<r<<" "<<std::hex<<err<<std::dec<<std::endl;
-	exit(EXIT_FAILURE);
-      } else {
-	std::cout << "SharedMemoryMPI.cc zeMemGetIpcHandle succeeded for rank "<<r<<" "<<std::hex<<err<<std::dec<<std::endl;
-      }
-      memcpy((void *)&handle.fd,(void *)&ihandle,sizeof(int));
-      handle.pid = getpid();
-    }
-#endif
 #ifdef GRID_CUDA
     cudaIpcMemHandle_t handle;
     if ( r==WorldShmRank ) { 
@@ -576,7 +493,6 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
       }
     }
 #endif
-
     //////////////////////////////////////////////////
     // Share this IPC handle across the Shm Comm
     //////////////////////////////////////////////////
@@ -592,35 +508,7 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
     ///////////////////////////////////////////////////////////////
     // If I am not the source, overwrite thisBuf with remote buffer
     ///////////////////////////////////////////////////////////////
-
-#ifdef GRID_SYCL_LEVEL_ZERO_IPC
-    if ( r!=WorldShmRank ) {
-      thisBuf = nullptr;
-      std::cout<<"mapping seeking remote pid/fd "
-	       <<handle.pid<<"/"
-	       <<handle.fd<<std::endl;
-
-      int pidfd = syscall(SYS_pidfd_open,handle.pid,0);
-      std::cout<<"Using IpcHandle pidfd "<<pidfd<<"\n";
-      //      int myfd  = syscall(SYS_pidfd_getfd,pidfd,handle.fd,0);
-      int myfd  = syscall(438,pidfd,handle.fd,0);
-
-      std::cout<<"Using IpcHandle myfd "<<myfd<<"\n";
-      
-      memcpy((void *)&ihandle,(void *)&myfd,sizeof(int));
-
-      auto err = zeMemOpenIpcHandle(zeContext,zeDevice,ihandle,0,&thisBuf);
-      if ( err != ZE_RESULT_SUCCESS ) {
-	std::cout << "SharedMemoryMPI.cc "<<zeContext<<" "<<zeDevice<<std::endl;
-	std::cout << "SharedMemoryMPI.cc zeMemOpenIpcHandle failed for rank "<<r<<" "<<std::hex<<err<<std::dec<<std::endl; 
-	exit(EXIT_FAILURE);
-      } else {
-	std::cout << "SharedMemoryMPI.cc zeMemOpenIpcHandle succeeded for rank "<<r<<std::endl;
-	std::cout << "SharedMemoryMPI.cc zeMemOpenIpcHandle pointer is "<<std::hex<<thisBuf<<std::dec<<std::endl;
-      }
-      assert(thisBuf!=nullptr);
-    }
-#endif
+    void * thisBuf = ShmCommBuf;
 #ifdef GRID_CUDA
     if ( r!=WorldShmRank ) { 
       auto err = cudaIpcOpenMemHandle(&thisBuf,handle,cudaIpcMemLazyEnablePeerAccess);
@@ -642,7 +530,6 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
     ///////////////////////////////////////////////////////////////
     // Save a copy of the device buffers
     ///////////////////////////////////////////////////////////////
-    }
     WorldShmCommBufs[r] = thisBuf;
 #else
     WorldShmCommBufs[r] = ShmCommBuf;
@@ -652,8 +539,6 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
   _ShmAllocBytes=bytes;
   _ShmAlloc=1;
 }
-#endif
-
 #else 
 #ifdef GRID_MPI3_SHMMMAP
 void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
@@ -824,16 +709,16 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
 /////////////////////////////////////////////////////////////////////////
 void GlobalSharedMemory::SharedMemoryZero(void *dest,size_t bytes)
 {
-#if defined(GRID_CUDA) || defined(GRID_HIP) || defined(GRID_SYCL)
-  acceleratorMemSet(dest,0,bytes);
+#ifdef GRID_CUDA
+  cudaMemset(dest,0,bytes);
 #else
   bzero(dest,bytes);
 #endif
 }
-void GlobalSharedMemory::SharedMemoryCopy(void *dest,void *src,size_t bytes)
+void GlobalSharedMemory::SharedMemoryCopy(void *dest,const void *src,size_t bytes)
 {
-#if defined(GRID_CUDA) || defined(GRID_HIP) || defined(GRID_SYCL)
-  acceleratorCopyToDevice(src,dest,bytes);
+#ifdef GRID_CUDA
+  cudaMemcpy(dest,src,bytes,cudaMemcpyDefault);
 #else   
   bcopy(src,dest,bytes);
 #endif
@@ -897,7 +782,7 @@ void SharedMemory::SetCommunicator(Grid_MPI_Comm comm)
   }
 #endif
 
-  //SharedMemoryTest();
+  SharedMemoryTest();
 }
 //////////////////////////////////////////////////////////////////
 // On node barrier

Grid/communicator/SharedMemoryNone.cc

@@ -29,7 +29,6 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 #include <Grid/GridCore.h>
 
 NAMESPACE_BEGIN(Grid); 
-#define header "SharedMemoryNone: "
 
 /*Construct from an MPI communicator*/
 void GlobalSharedMemory::Init(Grid_MPI_Comm comm)
@@ -56,38 +55,6 @@ void GlobalSharedMemory::OptimalCommunicator(const Coordinate &processors,Grid_M
 ////////////////////////////////////////////////////////////////////////////////////////////
 // Hugetlbfs mapping intended, use anonymous mmap
 ////////////////////////////////////////////////////////////////////////////////////////////
-#if 1
-void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
-{
-  std::cout << header "SharedMemoryAllocate "<< bytes<< " GPU implementation "<<std::endl;
-  void * ShmCommBuf ; 
-  assert(_ShmSetup==1);
-  assert(_ShmAlloc==0);
-
-  ///////////////////////////////////////////////////////////////////////////////////////////////////////////
-  // Each MPI rank should allocate our own buffer
-  ///////////////////////////////////////////////////////////////////////////////////////////////////////////
-  ShmCommBuf = acceleratorAllocDevice(bytes);
-
-  if (ShmCommBuf == (void *)NULL ) {
-    std::cerr << " SharedMemoryNone.cc acceleratorAllocDevice failed NULL pointer for " << bytes<<" bytes " << std::endl;
-    exit(EXIT_FAILURE);  
-  }
-  if ( WorldRank == 0 ){
-    std::cout << WorldRank << header " SharedMemoryNone.cc acceleratorAllocDevice "<< bytes 
-	      << "bytes at "<< std::hex<< ShmCommBuf <<std::dec<<" for comms buffers " <<std::endl;
-  }
-  SharedMemoryZero(ShmCommBuf,bytes);
-
-  ///////////////////////////////////////////////////////////////////////////////////////////////////////////
-  // Loop over ranks/gpu's on our node
-  ///////////////////////////////////////////////////////////////////////////////////////////////////////////
-  WorldShmCommBufs[0] = ShmCommBuf;
-
-  _ShmAllocBytes=bytes;
-  _ShmAlloc=1;
-}
-#else
 void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
 {
   void * ShmCommBuf ; 
@@ -116,15 +83,7 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
   _ShmAllocBytes=bytes;
   _ShmAlloc=1;
 };
-#endif
-void GlobalSharedMemory::SharedMemoryZero(void *dest,size_t bytes)
-{
-  acceleratorMemSet(dest,0,bytes);
-}
-void GlobalSharedMemory::SharedMemoryCopy(void *dest,void *src,size_t bytes)
-{
-  acceleratorCopyToDevice(src,dest,bytes);
-}
+
 ////////////////////////////////////////////////////////
 // Global shared functionality finished
 // Now move to per communicator functionality

Grid/cshift/Cshift_common.h

@@ -110,11 +110,9 @@ Gather_plane_extract(const Lattice<vobj> &rhs,
   int n1=rhs.Grid()->_slice_stride[dimension];
 
   if ( cbmask ==0x3){
-#ifdef ACCELERATOR_CSHIFT
+#ifdef ACCELERATOR_CSHIFT    
     autoView(rhs_v , rhs, AcceleratorRead);
-    accelerator_for(nn,e1*e2,1,{
-	int n = nn%e1;
-	int b = nn/e1;
+    accelerator_for2d(n,e1,b,e2,1,{
 	int o      =   n*n1;
 	int offset = b+n*e2;
 	
@@ -137,9 +135,7 @@ Gather_plane_extract(const Lattice<vobj> &rhs,
     std::cout << " Dense packed buffer WARNING " <<std::endl; // Does this get called twice once for each cb?
 #ifdef ACCELERATOR_CSHIFT    
     autoView(rhs_v , rhs, AcceleratorRead);
-    accelerator_for(nn,e1*e2,1,{
-	int n = nn%e1;
-	int b = nn/e1;
+    accelerator_for2d(n,e1,b,e2,1,{
 
 	Coordinate coor;
 
@@ -261,9 +257,7 @@ template<class vobj> void Scatter_plane_merge(Lattice<vobj> &rhs,ExtractPointerA
     int _slice_block = rhs.Grid()->_slice_block[dimension];
 #ifdef ACCELERATOR_CSHIFT    
     autoView( rhs_v , rhs, AcceleratorWrite);
-    accelerator_for(nn,e1*e2,1,{
-	int n = nn%e1;
-	int b = nn/e1;
+    accelerator_for2d(n,e1,b,e2,1,{
 	int o      = n*_slice_stride;
 	int offset = b+n*_slice_block;
 	merge(rhs_v[so+o+b],pointers,offset);
@@ -280,7 +274,7 @@ template<class vobj> void Scatter_plane_merge(Lattice<vobj> &rhs,ExtractPointerA
 
     // Case of SIMD split AND checker dim cannot currently be hit, except in 
     // Test_cshift_red_black code.
-    std::cout << "Scatter_plane merge assert(0); think this is buggy FIXME "<< std::endl;// think this is buggy FIXME
+    //    std::cout << "Scatter_plane merge assert(0); think this is buggy FIXME "<< std::endl;// think this is buggy FIXME
     std::cout<<" Unthreaded warning -- buffer is not densely packed ??"<<std::endl;
     assert(0); // This will fail if hit on GPU
     autoView( rhs_v, rhs, CpuWrite);

Grid/cshift/Cshift_mpi.h

@@ -122,8 +122,8 @@ template<class vobj> void Cshift_comms(Lattice<vobj> &ret,const Lattice<vobj> &r
   assert(shift<fd);
   
   int buffer_size = rhs.Grid()->_slice_nblock[dimension]*rhs.Grid()->_slice_block[dimension];
-  static cshiftVector<vobj> send_buf; send_buf.resize(buffer_size);
-  static cshiftVector<vobj> recv_buf; recv_buf.resize(buffer_size);
+  cshiftVector<vobj> send_buf(buffer_size);
+  cshiftVector<vobj> recv_buf(buffer_size);
     
   int cb= (cbmask==0x2)? Odd : Even;
   int sshift= rhs.Grid()->CheckerBoardShiftForCB(rhs.Checkerboard(),dimension,shift,cb);
@@ -198,8 +198,8 @@ template<class vobj> void  Cshift_comms_simd(Lattice<vobj> &ret,const Lattice<vo
   int buffer_size = grid->_slice_nblock[dimension]*grid->_slice_block[dimension];
   //  int words = sizeof(vobj)/sizeof(vector_type);
 
-  static std::vector<cshiftVector<scalar_object> >  send_buf_extract; send_buf_extract.resize(Nsimd);
-  static std::vector<cshiftVector<scalar_object> >  recv_buf_extract; recv_buf_extract.resize(Nsimd);
+  std::vector<cshiftVector<scalar_object> >  send_buf_extract(Nsimd);
+  std::vector<cshiftVector<scalar_object> >  recv_buf_extract(Nsimd);
   scalar_object *  recv_buf_extract_mpi;
   scalar_object *  send_buf_extract_mpi;
  
@@ -294,8 +294,8 @@ template<class vobj> void Cshift_comms(Lattice<vobj> &ret,const Lattice<vobj> &r
   assert(shift<fd);
   
   int buffer_size = rhs.Grid()->_slice_nblock[dimension]*rhs.Grid()->_slice_block[dimension];
-  static cshiftVector<vobj> send_buf_v; send_buf_v.resize(buffer_size);
-  static cshiftVector<vobj> recv_buf_v; recv_buf_v.resize(buffer_size);
+  cshiftVector<vobj> send_buf_v(buffer_size);
+  cshiftVector<vobj> recv_buf_v(buffer_size);
   vobj *send_buf;
   vobj *recv_buf;
   {
@@ -381,8 +381,8 @@ template<class vobj> void  Cshift_comms_simd(Lattice<vobj> &ret,const Lattice<vo
   int buffer_size = grid->_slice_nblock[dimension]*grid->_slice_block[dimension];
   //  int words = sizeof(vobj)/sizeof(vector_type);
 
-  static std::vector<cshiftVector<scalar_object> >  send_buf_extract; send_buf_extract.resize(Nsimd);
-  static std::vector<cshiftVector<scalar_object> >  recv_buf_extract; recv_buf_extract.resize(Nsimd);
+  std::vector<cshiftVector<scalar_object> >  send_buf_extract(Nsimd);
+  std::vector<cshiftVector<scalar_object> >  recv_buf_extract(Nsimd);
   scalar_object *  recv_buf_extract_mpi;
   scalar_object *  send_buf_extract_mpi;
   {

Grid/lattice/Lattice_arith.h

@@ -225,7 +225,7 @@ void axpy(Lattice<vobj> &ret,sobj a,const Lattice<vobj> &x,const Lattice<vobj> &
   autoView( x_v , x, AcceleratorRead);
   autoView( y_v , y, AcceleratorRead);
   accelerator_for(ss,x_v.size(),vobj::Nsimd(),{
-    auto tmp = a*coalescedRead(x_v[ss])+coalescedRead(y_v[ss]);
+    auto tmp = a*x_v(ss)+y_v(ss);
     coalescedWrite(ret_v[ss],tmp);
   });
 }

Grid/lattice/Lattice_basis.h

@@ -62,7 +62,7 @@ void basisRotate(VField &basis,Matrix& Qt,int j0, int j1, int k0,int k1,int Nm)
     basis_v.push_back(basis[k].View(AcceleratorWrite));
   }
 
-#if ( (!defined(GRID_CUDA)) )
+#if ( (!defined(GRID_SYCL)) && (!defined(GRID_CUDA)) )
   int max_threads = thread_max();
   Vector < vobj > Bt(Nm * max_threads);
   thread_region
@@ -125,7 +125,7 @@ void basisRotate(VField &basis,Matrix& Qt,int j0, int j1, int k0,int k1,int Nm)
 
 	for(int k=k0; k<k1; ++k){
 	  auto tmp = coalescedRead(Bp[ss*nrot+j]);
-	  coalescedWrite(Bp[ss*nrot+j],tmp+ Qt_p[jj*Nm+k] * coalescedRead(basis_vp[k][sss]));
+	  coalescedWrite(Bp[ss*nrot+j],tmp+ Qt_p[jj*Nm+k] * coalescedRead(basis_v[k][sss]));
 	}
       });
 
@@ -134,7 +134,7 @@ void basisRotate(VField &basis,Matrix& Qt,int j0, int j1, int k0,int k1,int Nm)
 	int jj  =j0+j;
 	int ss =sj/nrot;
 	int sss=ss+s;
-	coalescedWrite(basis_vp[jj][sss],coalescedRead(Bp[ss*nrot+j]));
+	coalescedWrite(basis_v[jj][sss],coalescedRead(Bp[ss*nrot+j]));
       });
   }
 #endif
@@ -164,8 +164,7 @@ void basisRotateJ(Field &result,std::vector<Field> &basis,Eigen::MatrixXd& Qt,in
   auto basis_vp=& basis_v[0];
   autoView(result_v,result,AcceleratorWrite);
   accelerator_for(ss, grid->oSites(),vobj::Nsimd(),{
-    vobj zzz=Zero();
-    auto B=coalescedRead(zzz);
+    auto B=coalescedRead(zz);
     for(int k=k0; k<k1; ++k){
       B +=Qt_j[k] * coalescedRead(basis_vp[k][ss]);
     }

Grid/lattice/Lattice_reduction.h

@@ -96,34 +96,8 @@ inline typename vobj::scalar_objectD sumD_cpu(const vobj *arg, Integer osites)
   ssobj ret = ssum;
   return ret;
 }
-/*
-Threaded max, don't use for now
-template<class Double>
-inline Double max(const Double *arg, Integer osites)
-{
-  //  const int Nsimd = vobj::Nsimd();
-  const int nthread = GridThread::GetThreads();
 
-  std::vector<Double> maxarray(nthread);
-  
-  thread_for(thr,nthread, {
-    int nwork, mywork, myoff;
-    nwork = osites;
-    GridThread::GetWork(nwork,thr,mywork,myoff);
-    Double max=arg[0];
-    for(int ss=myoff;ss<mywork+myoff; ss++){
-      if( arg[ss] > max ) max = arg[ss];
-    }
-    maxarray[thr]=max;
-  });
-  
-  Double tmax=maxarray[0];
-  for(int i=0;i<nthread;i++){
-    if (maxarray[i]>tmax) tmax = maxarray[i];
-  } 
-  return tmax;
-}
-*/
+
 template<class vobj>
 inline typename vobj::scalar_object sum(const vobj *arg, Integer osites)
 {
@@ -167,32 +141,6 @@ template<class vobj> inline RealD norm2(const Lattice<vobj> &arg){
   return real(nrm); 
 }
 
-//The global maximum of the site norm2
-template<class vobj> inline RealD maxLocalNorm2(const Lattice<vobj> &arg)
-{
-  typedef typename vobj::tensor_reduced vscalar;  //iScalar<iScalar<.... <vPODtype> > >
-  typedef typename vscalar::scalar_object  scalar;   //iScalar<iScalar<.... <PODtype> > >
-
-  Lattice<vscalar> inner = localNorm2(arg);
-
-  auto grid = arg.Grid();
-
-  RealD max;
-  for(int l=0;l<grid->lSites();l++){
-    Coordinate coor;
-    scalar val;
-    RealD r;
-    grid->LocalIndexToLocalCoor(l,coor);
-    peekLocalSite(val,inner,coor);
-    r=real(TensorRemove(val));
-    if( (l==0) || (r>max)){
-      max=r;
-    }
-  }
-  grid->GlobalMax(max);
-  return max;
-}
-
 // Double inner product
 template<class vobj>
 inline ComplexD rankInnerProduct(const Lattice<vobj> &left,const Lattice<vobj> &right)
@@ -361,7 +309,6 @@ template<class vobj> inline void sliceSum(const Lattice<vobj> &Data,std::vector<
   // But easily avoided by using double precision fields
   ///////////////////////////////////////////////////////
   typedef typename vobj::scalar_object sobj;
-  typedef typename vobj::scalar_object::scalar_type scalar_type;
   GridBase  *grid = Data.Grid();
   assert(grid!=NULL);
 
@@ -420,19 +367,20 @@ template<class vobj> inline void sliceSum(const Lattice<vobj> &Data,std::vector<
   }
   
   // sum over nodes.
+  sobj gsum;
   for(int t=0;t<fd;t++){
     int pt = t/ld; // processor plane
     int lt = t%ld;
     if ( pt == grid->_processor_coor[orthogdim] ) {
-      result[t]=lsSum[lt];
+      gsum=lsSum[lt];
     } else {
-      result[t]=Zero();
+      gsum=Zero();
     }
 
+    grid->GlobalSum(gsum);
+
+    result[t]=gsum;
   }
-  scalar_type * ptr = (scalar_type *) &result[0];
-  int words = fd*sizeof(sobj)/sizeof(scalar_type);
-  grid->GlobalSumVector(ptr, words);
 }
 
 template<class vobj>

Grid/lattice/Lattice_transfer.h

@@ -97,20 +97,6 @@ accelerator_inline void convertType(ComplexF & out, const std::complex<float> &
   out = in;
 }
 
-template<typename T>
-accelerator_inline EnableIf<isGridFundamental<T>> convertType(T & out, const T & in) {
-  out = in;
-}
-
-// This would allow for conversions between GridFundamental types, but is not strictly needed as yet
-/*template<typename T1, typename T2>
-accelerator_inline typename std::enable_if<isGridFundamental<T1>::value && isGridFundamental<T2>::value>::type
-// Or to make this very broad, conversions between anything that's not a GridTensor could be allowed
-//accelerator_inline typename std::enable_if<!isGridTensor<T1>::value && !isGridTensor<T2>::value>::type
-convertType(T1 & out, const T2 & in) {
-  out = in;
-}*/
-
 #ifdef GRID_SIMT
 accelerator_inline void convertType(vComplexF & out, const ComplexF & in) {
   ((ComplexF*)&out)[acceleratorSIMTlane(vComplexF::Nsimd())] = in;
@@ -131,18 +117,18 @@ accelerator_inline void convertType(vComplexD2 & out, const vComplexF & in) {
   Optimization::PrecisionChange::StoD(in.v,out._internal[0].v,out._internal[1].v);
 }
 
-template<typename T1,typename T2>
-accelerator_inline void convertType(iScalar<T1> & out, const iScalar<T2> & in) {
-  convertType(out._internal,in._internal);
-}
+template<typename T1,typename T2,int N>
+  accelerator_inline void convertType(iMatrix<T1,N> & out, const iMatrix<T2,N> & in);
+template<typename T1,typename T2,int N>
+  accelerator_inline void convertType(iVector<T1,N> & out, const iVector<T2,N> & in);
 
-template<typename T1,typename T2>
-accelerator_inline NotEnableIf<isGridScalar<T1>> convertType(T1 & out, const iScalar<T2> & in) {
+template<typename T1,typename T2, typename std::enable_if<!isGridScalar<T1>::value, T1>::type* = nullptr>
+accelerator_inline void convertType(T1 & out, const iScalar<T2> & in) {
   convertType(out,in._internal);
 }
 
 template<typename T1,typename T2>
-accelerator_inline NotEnableIf<isGridScalar<T2>> convertType(iScalar<T1> & out, const T2 & in) {
+accelerator_inline void convertType(iScalar<T1> & out, const T2 & in) {
   convertType(out._internal,in);
 }
 
@@ -159,6 +145,11 @@ accelerator_inline void convertType(iVector<T1,N> & out, const iVector<T2,N> & i
     convertType(out._internal[i],in._internal[i]);
 }
 
+template<typename T, typename std::enable_if<isGridFundamental<T>::value, T>::type* = nullptr>
+accelerator_inline void convertType(T & out, const T & in) {
+  out = in;
+}
+
 template<typename T1,typename T2>
 accelerator_inline void convertType(Lattice<T1> & out, const Lattice<T2> & in) {
   autoView( out_v , out,AcceleratorWrite);
@@ -364,22 +355,16 @@ inline void blockSum(Lattice<vobj> &coarseData,const Lattice<vobj> &fineData)
   autoView( coarseData_ , coarseData, AcceleratorWrite);
   autoView( fineData_   , fineData, AcceleratorRead);
 
-  auto coarseData_p = &coarseData_[0];
-  auto fineData_p = &fineData_[0];
-  
   Coordinate fine_rdimensions = fine->_rdimensions;
   Coordinate coarse_rdimensions = coarse->_rdimensions;
-
-  vobj zz = Zero();
   
   accelerator_for(sc,coarse->oSites(),1,{
 
       // One thread per sub block
       Coordinate coor_c(_ndimension);
       Lexicographic::CoorFromIndex(coor_c,sc,coarse_rdimensions);  // Block coordinate
+      coarseData_[sc]=Zero();
 
-      vobj cd = zz;
-      
       for(int sb=0;sb<blockVol;sb++){
 
 	int sf;
@@ -389,11 +374,9 @@ inline void blockSum(Lattice<vobj> &coarseData,const Lattice<vobj> &fineData)
 	for(int d=0;d<_ndimension;d++) coor_f[d]=coor_c[d]*block_r[d] + coor_b[d];
 	Lexicographic::IndexFromCoor(coor_f,sf,fine_rdimensions);
 
-	cd=cd+fineData_p[sf];
+	coarseData_[sc]=coarseData_[sc]+fineData_[sf];
       }
 
-      coarseData_p[sc] = cd;
-
     });
   return;
 }

Grid/lattice/Lattice_view.h

@@ -67,14 +67,9 @@ public:
   accelerator_inline const vobj & operator()(size_t i) const { return this->_odata[i]; }
 #endif
 
-#if 1
-  //  accelerator_inline const vobj & operator[](size_t i) const { return this->_odata[i]; };
-  accelerator_inline vobj       & operator[](size_t i) const { return this->_odata[i]; };
-#else
   accelerator_inline const vobj & operator[](size_t i) const { return this->_odata[i]; };
   accelerator_inline vobj       & operator[](size_t i)       { return this->_odata[i]; };
-#endif
-  
+
   accelerator_inline uint64_t begin(void) const { return 0;};
   accelerator_inline uint64_t end(void)   const { return this->_odata_size; };
   accelerator_inline uint64_t size(void)  const { return this->_odata_size; };

Grid/lattice/Lattice_where.h

@@ -43,7 +43,7 @@ inline void whereWolf(Lattice<vobj> &ret,const Lattice<iobj> &predicate,Lattice<
   conformable(iftrue,predicate);
   conformable(iftrue,ret);
 
-  GridBase *grid=iftrue.Grid();
+  GridBase *grid=iftrue._grid;
 
   typedef typename vobj::scalar_object scalar_object;
   typedef typename vobj::scalar_type scalar_type;
@@ -52,23 +52,22 @@ inline void whereWolf(Lattice<vobj> &ret,const Lattice<iobj> &predicate,Lattice<
 
   const int Nsimd = grid->Nsimd();
 
-  autoView(iftrue_v,iftrue,CpuRead);
-  autoView(iffalse_v,iffalse,CpuRead);
-  autoView(predicate_v,predicate,CpuRead);
-  autoView(ret_v,ret,CpuWrite);
-  Integer NN= grid->oSites();
-  thread_for(ss,NN,{
-    Integer mask;
-    scalar_object trueval;
-    scalar_object falseval;
-    for(int l=0;l<Nsimd;l++){
-      trueval =extractLane(l,iftrue_v[ss]);
-      falseval=extractLane(l,iffalse_v[ss]);
-      mask    =extractLane(l,predicate_v[ss]);
-      if (mask) falseval=trueval;
-      insertLane(l,ret_v[ss],falseval);
+  std::vector<Integer> mask(Nsimd);
+  std::vector<scalar_object> truevals (Nsimd);
+  std::vector<scalar_object> falsevals(Nsimd);
+
+  parallel_for(int ss=0;ss<iftrue._grid->oSites(); ss++){
+
+    extract(iftrue._odata[ss]   ,truevals);
+    extract(iffalse._odata[ss]  ,falsevals);
+    extract<vInteger,Integer>(TensorRemove(predicate._odata[ss]),mask);
+
+    for(int s=0;s<Nsimd;s++){
+      if (mask[s]) falsevals[s]=truevals[s];
     }
-  });
+
+    merge(ret._odata[ss],falsevals);
+  }
 }
 
 template<class vobj,class iobj>
@@ -77,9 +76,9 @@ inline Lattice<vobj> whereWolf(const Lattice<iobj> &predicate,Lattice<vobj> &ift
   conformable(iftrue,iffalse);
   conformable(iftrue,predicate);
 
-  Lattice<vobj> ret(iftrue.Grid());
+  Lattice<vobj> ret(iftrue._grid);
 
-  whereWolf(ret,predicate,iftrue,iffalse);
+  where(ret,predicate,iftrue,iffalse);
 
   return ret;
 }

Grid/parallelIO/IldgIO.h

@@ -123,7 +123,7 @@ assert(GRID_FIELD_NORM_CALC(FieldNormMetaData_, n2ck) < 1.0e-5);
  ////////////////////////////////////////////////////////////
  // Helper to fill out metadata
  ////////////////////////////////////////////////////////////
-template<class vobj> void ScidacMetaData(Lattice<vobj> & field,
+ template<class vobj> void ScidacMetaData(Lattice<vobj> & field,
 					  FieldMetaData &header,
 					  scidacRecord & _scidacRecord,
 					  scidacFile   & _scidacFile) 
@@ -619,12 +619,12 @@ class IldgWriter : public ScidacWriter {
   // Don't require scidac records EXCEPT checksum
   // Use Grid MetaData object if present.
   ////////////////////////////////////////////////////////////////
-  template <class stats = PeriodicGaugeStatistics>
-  void writeConfiguration(Lattice<vLorentzColourMatrixD > &Umu,int sequence,std::string LFN,std::string description) 
+  template <class vsimd>
+  void writeConfiguration(Lattice<iLorentzColourMatrix<vsimd> > &Umu,int sequence,std::string LFN,std::string description) 
   {
     GridBase * grid = Umu.Grid();
-    typedef Lattice<vLorentzColourMatrixD> GaugeField;
-    typedef vLorentzColourMatrixD vobj;
+    typedef Lattice<iLorentzColourMatrix<vsimd> > GaugeField;
+    typedef iLorentzColourMatrix<vsimd> vobj;
     typedef typename vobj::scalar_object sobj;
 
     ////////////////////////////////////////
@@ -636,9 +636,6 @@ class IldgWriter : public ScidacWriter {
 
     ScidacMetaData(Umu,header,_scidacRecord,_scidacFile);
 
-    stats Stats;
-    Stats(Umu,header);
-    
     std::string format = header.floating_point;
     header.ensemble_id    = description;
     header.ensemble_label = description;
@@ -708,10 +705,10 @@ class IldgReader : public GridLimeReader {
   // Else use ILDG MetaData object if present.
   // Else use SciDAC MetaData object if present.
   ////////////////////////////////////////////////////////////////
-  template <class stats = PeriodicGaugeStatistics>
-  void readConfiguration(Lattice<vLorentzColourMatrixD> &Umu, FieldMetaData &FieldMetaData_) {
+  template <class vsimd>
+  void readConfiguration(Lattice<iLorentzColourMatrix<vsimd> > &Umu, FieldMetaData &FieldMetaData_) {
 
-    typedef Lattice<vLorentzColourMatrixD > GaugeField;
+    typedef Lattice<iLorentzColourMatrix<vsimd> > GaugeField;
     typedef typename GaugeField::vector_object  vobj;
     typedef typename vobj::scalar_object sobj;
 
@@ -924,8 +921,7 @@ class IldgReader : public GridLimeReader {
 
     if ( found_FieldMetaData || found_usqcdInfo ) {
       FieldMetaData checker;
-      stats Stats;
-      Stats(Umu,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;

Grid/parallelIO/MetaData.h

@@ -128,7 +128,7 @@ inline void MachineCharacteristics(FieldMetaData &header)
   std::time_t t = std::time(nullptr);
   std::tm tm_ = *std::localtime(&t);
   std::ostringstream oss; 
-  oss << std::put_time(&tm_, "%c %Z");
+  //      oss << std::put_time(&tm_, "%c %Z");
   header.creation_date = oss.str();
   header.archive_date  = header.creation_date;
 
@@ -176,18 +176,29 @@ template<class vobj> inline void PrepareMetaData(Lattice<vobj> & field, FieldMet
   GridMetaData(grid,header); 
   MachineCharacteristics(header);
 }
-template<class Impl>
-class GaugeStatistics
+inline void GaugeStatistics(Lattice<vLorentzColourMatrixF> & data,FieldMetaData &header)
 {
-public:
-  void operator()(Lattice<vLorentzColourMatrixD> & data,FieldMetaData &header)
-  {
-    header.link_trace=WilsonLoops<Impl>::linkTrace(data);
-    header.plaquette =WilsonLoops<Impl>::avgPlaquette(data);
-  }
-};
-typedef GaugeStatistics<PeriodicGimplD> PeriodicGaugeStatistics;
-typedef GaugeStatistics<ConjugateGimplD> ConjugateGaugeStatistics;
+  // How to convert data precision etc...
+  header.link_trace=WilsonLoops<PeriodicGimplF>::linkTrace(data);
+  header.plaquette =WilsonLoops<PeriodicGimplF>::avgPlaquette(data);
+}
+inline void GaugeStatistics(Lattice<vLorentzColourMatrixD> & data,FieldMetaData &header)
+{
+  // How to convert data precision etc...
+  header.link_trace=WilsonLoops<PeriodicGimplD>::linkTrace(data);
+  header.plaquette =WilsonLoops<PeriodicGimplD>::avgPlaquette(data);
+}
+template<> inline void PrepareMetaData<vLorentzColourMatrixF>(Lattice<vLorentzColourMatrixF> & field, FieldMetaData &header)
+{
+   
+  GridBase *grid = field.Grid();
+  std::string format = getFormatString<vLorentzColourMatrixF>();
+  header.floating_point = format;
+  header.checksum = 0x0; // Nersc checksum unused in ILDG, Scidac
+  GridMetaData(grid,header); 
+  GaugeStatistics(field,header);
+  MachineCharacteristics(header);
+}
 template<> inline void PrepareMetaData<vLorentzColourMatrixD>(Lattice<vLorentzColourMatrixD> & field, FieldMetaData &header)
 {
   GridBase *grid = field.Grid();
@@ -195,6 +206,7 @@ template<> inline void PrepareMetaData<vLorentzColourMatrixD>(Lattice<vLorentzCo
   header.floating_point = format;
   header.checksum = 0x0; // Nersc checksum unused in ILDG, Scidac
   GridMetaData(grid,header); 
+  GaugeStatistics(field,header);
   MachineCharacteristics(header);
 }
 

Grid/parallelIO/NerscIO.h

@@ -40,8 +40,6 @@ using namespace Grid;
 class NerscIO : public BinaryIO { 
 public:
 
-  typedef Lattice<vLorentzColourMatrixD> GaugeField;
-
   static inline void truncate(std::string file){
     std::ofstream fout(file,std::ios::out);
   }
@@ -131,12 +129,12 @@ public:
   // Now the meat: the object readers
   /////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 
-  template<class GaugeStats=PeriodicGaugeStatistics>
-  static inline void readConfiguration(GaugeField &Umu,
+  template<class vsimd>
+  static inline void readConfiguration(Lattice<iLorentzColourMatrix<vsimd> > &Umu,
 				       FieldMetaData& header,
-				       std::string file,
-				       GaugeStats GaugeStatisticsCalculator=GaugeStats())
+				       std::string file)
   {
+    typedef Lattice<iLorentzColourMatrix<vsimd> > GaugeField;
 
     GridBase *grid = Umu.Grid();
     uint64_t offset = readHeader(file,Umu.Grid(),header);
@@ -155,23 +153,23 @@ public:
     // munger is a function of <floating point, Real, data_type>
     if ( header.data_type == std::string("4D_SU3_GAUGE") ) {
       if ( ieee32 || ieee32big ) {
-	BinaryIO::readLatticeObject<vLorentzColourMatrixD, LorentzColour2x3F> 
+	BinaryIO::readLatticeObject<iLorentzColourMatrix<vsimd>, LorentzColour2x3F> 
 	  (Umu,file,Gauge3x2munger<LorentzColour2x3F,LorentzColourMatrix>(), offset,format,
 	   nersc_csum,scidac_csuma,scidac_csumb);
       }
       if ( ieee64 || ieee64big ) {
-	BinaryIO::readLatticeObject<vLorentzColourMatrixD, LorentzColour2x3D> 
+	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<vLorentzColourMatrixD,LorentzColourMatrixF>
+	BinaryIO::readLatticeObject<iLorentzColourMatrix<vsimd>,LorentzColourMatrixF>
 	  (Umu,file,GaugeSimpleMunger<LorentzColourMatrixF,LorentzColourMatrix>(),offset,format,
 	   nersc_csum,scidac_csuma,scidac_csumb);
       }
       if ( ieee64 || ieee64big ) {
-	BinaryIO::readLatticeObject<vLorentzColourMatrixD,LorentzColourMatrixD>
+	BinaryIO::readLatticeObject<iLorentzColourMatrix<vsimd>,LorentzColourMatrixD>
 	  (Umu,file,GaugeSimpleMunger<LorentzColourMatrixD,LorentzColourMatrix>(),offset,format,
 	   nersc_csum,scidac_csuma,scidac_csumb);
       }
@@ -179,7 +177,7 @@ public:
       assert(0);
     }
 
-    GaugeStats Stats; Stats(Umu,clone);
+    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;
@@ -205,22 +203,15 @@ public:
     std::cout<<GridLogMessage <<"NERSC Configuration "<<file<< " and plaquette, link trace, and checksum agree"<<std::endl;
   }
 
-  // Preferred interface
-  template<class GaugeStats=PeriodicGaugeStatistics>
-  static inline void writeConfiguration(Lattice<vLorentzColourMatrixD > &Umu,
-					std::string file, 
-					std::string ens_label = std::string("DWF"))
-  {
-    writeConfiguration(Umu,file,0,1,ens_label);
-  }
-  template<class GaugeStats=PeriodicGaugeStatistics>
-  static inline void writeConfiguration(Lattice<vLorentzColourMatrixD > &Umu,
+  template<class vsimd>
+  static inline void writeConfiguration(Lattice<iLorentzColourMatrix<vsimd> > &Umu,
 					std::string file, 
 					int two_row,
-					int bits32,
-					std::string ens_label = std::string("DWF"))
+					int bits32)
   {
-    typedef vLorentzColourMatrixD vobj;
+    typedef Lattice<iLorentzColourMatrix<vsimd> > GaugeField;
+
+    typedef iLorentzColourMatrix<vsimd> vobj;
     typedef typename vobj::scalar_object sobj;
 
     FieldMetaData header;
@@ -228,8 +219,8 @@ public:
     // Following should become arguments
     ///////////////////////////////////////////
     header.sequence_number = 1;
-    header.ensemble_id     = std::string("UKQCD");
-    header.ensemble_label  = ens_label;
+    header.ensemble_id     = "UKQCD";
+    header.ensemble_label  = "DWF";
 
     typedef LorentzColourMatrixD fobj3D;
     typedef LorentzColour2x3D    fobj2D;
@@ -238,28 +229,28 @@ public:
 
     GridMetaData(grid,header);
     assert(header.nd==4);
-    GaugeStats Stats; Stats(Umu,header);
+    GaugeStatistics(Umu,header);
     MachineCharacteristics(header);
 
-    uint64_t offset;
+	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));
+	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);
-    }
+	if ( grid->IsBoss() ) { 
+    writeHeader(header,file);
+	}
 
     std::cout<<GridLogMessage <<"Written NERSC Configuration on "<< file << " checksum "
 	     <<std::hex<<header.checksum

Grid/parallelIO/OpenQcdIO.h

@@ -154,7 +154,7 @@ public:
     grid->Barrier(); timer.Stop();
     std::cout << Grid::GridLogMessage << "OpenQcdIO::readConfiguration: redistribute overhead " << timer.Elapsed() << std::endl;
 
-    PeriodicGaugeStatistics Stats; Stats(Umu, clone);
+    GaugeStatistics(Umu, clone);
 
     RealD plaq_diff = fabs(clone.plaquette - header.plaquette);
 

Grid/parallelIO/OpenQcdIOChromaReference.h

@@ -208,7 +208,7 @@ public:
 
     FieldMetaData clone(header);
 
-    PeriodicGaugeStatistics Stats; Stats(Umu, clone);
+    GaugeStatistics(Umu, clone);
 
     RealD plaq_diff = fabs(clone.plaquette - header.plaquette);
 

Grid/qcd/QCD.h

@@ -80,13 +80,6 @@ template<typename T> struct isSpinor {
 template <typename T> using IfSpinor    = Invoke<std::enable_if< isSpinor<T>::value,int> > ;
 template <typename T> using IfNotSpinor = Invoke<std::enable_if<!isSpinor<T>::value,int> > ;
 
-const int CoarseIndex = 4;
-template<typename T> struct isCoarsened {
-   static constexpr bool value = (CoarseIndex<=T::TensorLevel);
-};
-template <typename T> using IfCoarsened    = Invoke<std::enable_if< isCoarsened<T>::value,int> > ;
-template <typename T> using IfNotCoarsened = Invoke<std::enable_if<!isCoarsened<T>::value,int> > ;
-
 // ChrisK very keen to add extra space for Gparity doubling.
 //
 // Also add domain wall index, in a way where Wilson operator 

Grid/qcd/action/ActionBase.h

@@ -41,7 +41,7 @@ class Action
 public:
   bool is_smeared = false;
   // Heatbath?
-  virtual void refresh(const GaugeField& U, GridSerialRNG &sRNG, GridParallelRNG& pRNG) = 0; // refresh pseudofermions
+  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

Grid/qcd/action/fermion/Fermion.h

@@ -115,9 +115,9 @@ typedef WilsonFermion<WilsonImplR> WilsonFermionR;
 typedef WilsonFermion<WilsonImplF> WilsonFermionF;
 typedef WilsonFermion<WilsonImplD> WilsonFermionD;
 
-//typedef WilsonFermion<WilsonImplRL> WilsonFermionRL;
-//typedef WilsonFermion<WilsonImplFH> WilsonFermionFH;
-//typedef WilsonFermion<WilsonImplDF> WilsonFermionDF;
+typedef WilsonFermion<WilsonImplRL> WilsonFermionRL;
+typedef WilsonFermion<WilsonImplFH> WilsonFermionFH;
+typedef WilsonFermion<WilsonImplDF> WilsonFermionDF;
 
 typedef WilsonFermion<WilsonAdjImplR> WilsonAdjFermionR;
 typedef WilsonFermion<WilsonAdjImplF> WilsonAdjFermionF;
@@ -158,41 +158,41 @@ typedef DomainWallFermion<WilsonImplR> DomainWallFermionR;
 typedef DomainWallFermion<WilsonImplF> DomainWallFermionF;
 typedef DomainWallFermion<WilsonImplD> DomainWallFermionD;
 
-//typedef DomainWallFermion<WilsonImplRL> DomainWallFermionRL;
-//typedef DomainWallFermion<WilsonImplFH> DomainWallFermionFH;
-//typedef DomainWallFermion<WilsonImplDF> DomainWallFermionDF;
+typedef DomainWallFermion<WilsonImplRL> DomainWallFermionRL;
+typedef DomainWallFermion<WilsonImplFH> DomainWallFermionFH;
+typedef DomainWallFermion<WilsonImplDF> DomainWallFermionDF;
 
 typedef DomainWallEOFAFermion<WilsonImplR> DomainWallEOFAFermionR;
 typedef DomainWallEOFAFermion<WilsonImplF> DomainWallEOFAFermionF;
 typedef DomainWallEOFAFermion<WilsonImplD> DomainWallEOFAFermionD;
 
-//typedef DomainWallEOFAFermion<WilsonImplRL> DomainWallEOFAFermionRL;
-//typedef DomainWallEOFAFermion<WilsonImplFH> DomainWallEOFAFermionFH;
-//typedef DomainWallEOFAFermion<WilsonImplDF> DomainWallEOFAFermionDF;
+typedef DomainWallEOFAFermion<WilsonImplRL> DomainWallEOFAFermionRL;
+typedef DomainWallEOFAFermion<WilsonImplFH> DomainWallEOFAFermionFH;
+typedef DomainWallEOFAFermion<WilsonImplDF> DomainWallEOFAFermionDF;
 
 typedef MobiusFermion<WilsonImplR> MobiusFermionR;
 typedef MobiusFermion<WilsonImplF> MobiusFermionF;
 typedef MobiusFermion<WilsonImplD> MobiusFermionD;
 
-//typedef MobiusFermion<WilsonImplRL> MobiusFermionRL;
-//typedef MobiusFermion<WilsonImplFH> MobiusFermionFH;
-//typedef MobiusFermion<WilsonImplDF> MobiusFermionDF;
+typedef MobiusFermion<WilsonImplRL> MobiusFermionRL;
+typedef MobiusFermion<WilsonImplFH> MobiusFermionFH;
+typedef MobiusFermion<WilsonImplDF> MobiusFermionDF;
 
 typedef MobiusEOFAFermion<WilsonImplR> MobiusEOFAFermionR;
 typedef MobiusEOFAFermion<WilsonImplF> MobiusEOFAFermionF;
 typedef MobiusEOFAFermion<WilsonImplD> MobiusEOFAFermionD;
 
-//typedef MobiusEOFAFermion<WilsonImplRL> MobiusEOFAFermionRL;
-//typedef MobiusEOFAFermion<WilsonImplFH> MobiusEOFAFermionFH;
-//typedef MobiusEOFAFermion<WilsonImplDF> MobiusEOFAFermionDF;
+typedef MobiusEOFAFermion<WilsonImplRL> MobiusEOFAFermionRL;
+typedef MobiusEOFAFermion<WilsonImplFH> MobiusEOFAFermionFH;
+typedef MobiusEOFAFermion<WilsonImplDF> MobiusEOFAFermionDF;
 
 typedef ZMobiusFermion<ZWilsonImplR> ZMobiusFermionR;
 typedef ZMobiusFermion<ZWilsonImplF> ZMobiusFermionF;
 typedef ZMobiusFermion<ZWilsonImplD> ZMobiusFermionD;
 
-//typedef ZMobiusFermion<ZWilsonImplRL> ZMobiusFermionRL;
-//typedef ZMobiusFermion<ZWilsonImplFH> ZMobiusFermionFH;
-//typedef ZMobiusFermion<ZWilsonImplDF> ZMobiusFermionDF;
+typedef ZMobiusFermion<ZWilsonImplRL> ZMobiusFermionRL;
+typedef ZMobiusFermion<ZWilsonImplFH> ZMobiusFermionFH;
+typedef ZMobiusFermion<ZWilsonImplDF> ZMobiusFermionDF;
 
 // Ls vectorised
 typedef ScaledShamirFermion<WilsonImplR> ScaledShamirFermionR;
@@ -235,49 +235,49 @@ typedef WilsonFermion<GparityWilsonImplR>     GparityWilsonFermionR;
 typedef WilsonFermion<GparityWilsonImplF>     GparityWilsonFermionF;
 typedef WilsonFermion<GparityWilsonImplD>     GparityWilsonFermionD;
 
-//typedef WilsonFermion<GparityWilsonImplRL>     GparityWilsonFermionRL;
-//typedef WilsonFermion<GparityWilsonImplFH>     GparityWilsonFermionFH;
-//typedef WilsonFermion<GparityWilsonImplDF>     GparityWilsonFermionDF;
+typedef WilsonFermion<GparityWilsonImplRL>     GparityWilsonFermionRL;
+typedef WilsonFermion<GparityWilsonImplFH>     GparityWilsonFermionFH;
+typedef WilsonFermion<GparityWilsonImplDF>     GparityWilsonFermionDF;
 
 typedef DomainWallFermion<GparityWilsonImplR> GparityDomainWallFermionR;
 typedef DomainWallFermion<GparityWilsonImplF> GparityDomainWallFermionF;
 typedef DomainWallFermion<GparityWilsonImplD> GparityDomainWallFermionD;
 
-//typedef DomainWallFermion<GparityWilsonImplRL> GparityDomainWallFermionRL;
-//typedef DomainWallFermion<GparityWilsonImplFH> GparityDomainWallFermionFH;
-//typedef DomainWallFermion<GparityWilsonImplDF> GparityDomainWallFermionDF;
+typedef DomainWallFermion<GparityWilsonImplRL> GparityDomainWallFermionRL;
+typedef DomainWallFermion<GparityWilsonImplFH> GparityDomainWallFermionFH;
+typedef DomainWallFermion<GparityWilsonImplDF> GparityDomainWallFermionDF;
 
 typedef DomainWallEOFAFermion<GparityWilsonImplR> GparityDomainWallEOFAFermionR;
 typedef DomainWallEOFAFermion<GparityWilsonImplF> GparityDomainWallEOFAFermionF;
 typedef DomainWallEOFAFermion<GparityWilsonImplD> GparityDomainWallEOFAFermionD;
 
-//typedef DomainWallEOFAFermion<GparityWilsonImplRL> GparityDomainWallEOFAFermionRL;
-//typedef DomainWallEOFAFermion<GparityWilsonImplFH> GparityDomainWallEOFAFermionFH;
-//typedef DomainWallEOFAFermion<GparityWilsonImplDF> GparityDomainWallEOFAFermionDF;
+typedef DomainWallEOFAFermion<GparityWilsonImplRL> GparityDomainWallEOFAFermionRL;
+typedef DomainWallEOFAFermion<GparityWilsonImplFH> GparityDomainWallEOFAFermionFH;
+typedef DomainWallEOFAFermion<GparityWilsonImplDF> GparityDomainWallEOFAFermionDF;
 
 typedef WilsonTMFermion<GparityWilsonImplR> GparityWilsonTMFermionR;
 typedef WilsonTMFermion<GparityWilsonImplF> GparityWilsonTMFermionF;
 typedef WilsonTMFermion<GparityWilsonImplD> GparityWilsonTMFermionD;
 
-//typedef WilsonTMFermion<GparityWilsonImplRL> GparityWilsonTMFermionRL;
-//typedef WilsonTMFermion<GparityWilsonImplFH> GparityWilsonTMFermionFH;
-//typedef WilsonTMFermion<GparityWilsonImplDF> GparityWilsonTMFermionDF;
+typedef WilsonTMFermion<GparityWilsonImplRL> GparityWilsonTMFermionRL;
+typedef WilsonTMFermion<GparityWilsonImplFH> GparityWilsonTMFermionFH;
+typedef WilsonTMFermion<GparityWilsonImplDF> GparityWilsonTMFermionDF;
 
 typedef MobiusFermion<GparityWilsonImplR> GparityMobiusFermionR;
 typedef MobiusFermion<GparityWilsonImplF> GparityMobiusFermionF;
 typedef MobiusFermion<GparityWilsonImplD> GparityMobiusFermionD;
 
-//typedef MobiusFermion<GparityWilsonImplRL> GparityMobiusFermionRL;
-//typedef MobiusFermion<GparityWilsonImplFH> GparityMobiusFermionFH;
-//typedef MobiusFermion<GparityWilsonImplDF> GparityMobiusFermionDF;
+typedef MobiusFermion<GparityWilsonImplRL> GparityMobiusFermionRL;
+typedef MobiusFermion<GparityWilsonImplFH> GparityMobiusFermionFH;
+typedef MobiusFermion<GparityWilsonImplDF> GparityMobiusFermionDF;
 
 typedef MobiusEOFAFermion<GparityWilsonImplR> GparityMobiusEOFAFermionR;
 typedef MobiusEOFAFermion<GparityWilsonImplF> GparityMobiusEOFAFermionF;
 typedef MobiusEOFAFermion<GparityWilsonImplD> GparityMobiusEOFAFermionD;
 
-//typedef MobiusEOFAFermion<GparityWilsonImplRL> GparityMobiusEOFAFermionRL;
-//typedef MobiusEOFAFermion<GparityWilsonImplFH> GparityMobiusEOFAFermionFH;
-//typedef MobiusEOFAFermion<GparityWilsonImplDF> GparityMobiusEOFAFermionDF;
+typedef MobiusEOFAFermion<GparityWilsonImplRL> GparityMobiusEOFAFermionRL;
+typedef MobiusEOFAFermion<GparityWilsonImplFH> GparityMobiusEOFAFermionFH;
+typedef MobiusEOFAFermion<GparityWilsonImplDF> GparityMobiusEOFAFermionDF;
 
 typedef ImprovedStaggeredFermion<StaggeredImplR> ImprovedStaggeredFermionR;
 typedef ImprovedStaggeredFermion<StaggeredImplF> ImprovedStaggeredFermionF;
@@ -291,6 +291,12 @@ typedef ImprovedStaggeredFermion5D<StaggeredImplR> ImprovedStaggeredFermion5DR;
 typedef ImprovedStaggeredFermion5D<StaggeredImplF> ImprovedStaggeredFermion5DF;
 typedef ImprovedStaggeredFermion5D<StaggeredImplD> ImprovedStaggeredFermion5DD;
 
+#ifndef GRID_CUDA
+typedef ImprovedStaggeredFermion5D<StaggeredVec5dImplR> ImprovedStaggeredFermionVec5dR;
+typedef ImprovedStaggeredFermion5D<StaggeredVec5dImplF> ImprovedStaggeredFermionVec5dF;
+typedef ImprovedStaggeredFermion5D<StaggeredVec5dImplD> ImprovedStaggeredFermionVec5dD;
+#endif
+
 NAMESPACE_END(Grid);
 
 ////////////////////

Grid/qcd/action/fermion/FermionOperatorImpl.h

@@ -153,8 +153,8 @@ public:
   typedef typename Impl::StencilImpl             StencilImpl;		\
   typedef typename Impl::ImplParams               ImplParams;	        \
   typedef typename Impl::StencilImpl::View_type  StencilView;		\
-  typedef const typename ViewMap<FermionField>::Type      FermionFieldView;	\
-  typedef const typename ViewMap<DoubledGaugeField>::Type DoubledGaugeFieldView;
+  typedef typename ViewMap<FermionField>::Type      FermionFieldView;	\
+  typedef typename ViewMap<DoubledGaugeField>::Type DoubledGaugeFieldView;
 
 #define INHERIT_IMPL_TYPES(Base)		\
   INHERIT_GIMPL_TYPES(Base)			\
@@ -183,8 +183,7 @@ NAMESPACE_CHECK(ImplStaggered);
 /////////////////////////////////////////////////////////////////////////////
 // Single flavour one component spinors with colour index. 5d vec
 /////////////////////////////////////////////////////////////////////////////
-// Deprecate Vec5d
-//#include <Grid/qcd/action/fermion/StaggeredVec5dImpl.h> 
-//NAMESPACE_CHECK(ImplStaggered5dVec);  
+#include <Grid/qcd/action/fermion/StaggeredVec5dImpl.h> 
+NAMESPACE_CHECK(ImplStaggered5dVec);  
 
 

Grid/qcd/action/fermion/GparityWilsonImpl.h

@@ -97,30 +97,42 @@ public:
     Coordinate icoor;
 
 #ifdef GRID_SIMT
+    _Spinor tmp;
+
     const int Nsimd =SiteDoubledGaugeField::Nsimd();
     int s = acceleratorSIMTlane(Nsimd);
     St.iCoorFromIindex(icoor,s);
 
     int mmu = mu % Nd;
+    if ( SE->_around_the_world && St.parameters.twists[mmu] ) {
+      
+      int permute_lane = (sl==1) 
+    	|| ((distance== 1)&&(icoor[direction]==1))
+	|| ((distance==-1)&&(icoor[direction]==0));
 
-    auto UU0=coalescedRead(U(0)(mu));
-    auto UU1=coalescedRead(U(1)(mu));
-    
-    //Decide whether we do a G-parity flavor twist
-    //Note: this assumes (but does not check) that sl==1 || sl==2 i.e. max 2 SIMD lanes in G-parity dir
-    //It also assumes (but does not check) that abs(distance) == 1
-    int permute_lane = (sl==1) 
-    || ((distance== 1)&&(icoor[direction]==1))
-    || ((distance==-1)&&(icoor[direction]==0));
+      if ( permute_lane ) { 
+	tmp(0) = chi(1);
+	tmp(1) = chi(0);
+      } else {
+	tmp(0) = chi(0);
+	tmp(1) = chi(1);
+      }
 
-    permute_lane = permute_lane && SE->_around_the_world && St.parameters.twists[mmu]; //only if we are going around the world
+      auto UU0=coalescedRead(U(0)(mu));
+      auto UU1=coalescedRead(U(1)(mu));
 
-    //Apply the links
-    int f_upper = permute_lane ? 1 : 0;
-    int f_lower = !f_upper;
+      mult(&phi(0),&UU0,&tmp(0));
+      mult(&phi(1),&UU1,&tmp(1));
 
-    mult(&phi(0),&UU0,&chi(f_upper));
-    mult(&phi(1),&UU1,&chi(f_lower));
+    } else {
+
+      auto UU0=coalescedRead(U(0)(mu));
+      auto UU1=coalescedRead(U(1)(mu));
+
+      mult(&phi(0),&UU0,&chi(0));
+      mult(&phi(1),&UU1,&chi(1));
+
+    }
 
 #else
     typedef _Spinor vobj;
@@ -327,8 +339,8 @@ typedef GparityWilsonImpl<vComplex , FundamentalRepresentation,CoeffReal> Gparit
 typedef GparityWilsonImpl<vComplexF, FundamentalRepresentation,CoeffReal> GparityWilsonImplF;  // Float
 typedef GparityWilsonImpl<vComplexD, FundamentalRepresentation,CoeffReal> GparityWilsonImplD;  // Double
  
-//typedef GparityWilsonImpl<vComplex , FundamentalRepresentation,CoeffRealHalfComms> GparityWilsonImplRL;  // Real.. whichever prec
-//typedef GparityWilsonImpl<vComplexF, FundamentalRepresentation,CoeffRealHalfComms> GparityWilsonImplFH;  // Float
-//typedef GparityWilsonImpl<vComplexD, FundamentalRepresentation,CoeffRealHalfComms> GparityWilsonImplDF;  // Double
+typedef GparityWilsonImpl<vComplex , FundamentalRepresentation,CoeffRealHalfComms> GparityWilsonImplRL;  // Real.. whichever prec
+typedef GparityWilsonImpl<vComplexF, FundamentalRepresentation,CoeffRealHalfComms> GparityWilsonImplFH;  // Float
+typedef GparityWilsonImpl<vComplexD, FundamentalRepresentation,CoeffRealHalfComms> GparityWilsonImplDF;  // Double
 
 NAMESPACE_END(Grid);

Grid/qcd/action/fermion/MADWF.h

@@ -85,7 +85,7 @@ class MADWF
       maxiter     =_maxiter;
     };
    
-  void operator() (const FermionFieldo &src,FermionFieldo &sol5)
+  void operator() (const FermionFieldo &src4,FermionFieldo &sol5)
   {
     std::cout << GridLogMessage<< " ************************************************" << std::endl;
     std::cout << GridLogMessage<< "  MADWF-like algorithm                           " << std::endl;
@@ -114,16 +114,8 @@ class MADWF
     ///////////////////////////////////////
     //Import source, include Dminus factors
     ///////////////////////////////////////
-    GridBase *src_grid = src.Grid();
-
-    assert( (src_grid == Mato.GaugeGrid()) || (src_grid == Mato.FermionGrid()));
-
-    if ( src_grid == Mato.GaugeGrid() ) {
-      Mato.ImportPhysicalFermionSource(src,b);
-    } else {
-      b=src;
-    }
-    std::cout << GridLogMessage << " src " <<norm2(src)<<std::endl;
+    Mato.ImportPhysicalFermionSource(src4,b); 
+    std::cout << GridLogMessage << " src4 " <<norm2(src4)<<std::endl;
     std::cout << GridLogMessage << " b    " <<norm2(b)<<std::endl;
 
     defect = b;

Grid/qcd/action/fermion/StaggeredImpl.h

@@ -72,23 +72,19 @@ public:
     
   StaggeredImpl(const ImplParams &p = ImplParams()) : Params(p){};
       
-  template<class _Spinor>
-  static accelerator_inline void multLink(_Spinor &phi,
+  static accelerator_inline void multLink(SiteSpinor &phi,
 		       const SiteDoubledGaugeField &U,
-		       const _Spinor &chi,
+		       const SiteSpinor &chi,
 		       int mu)
   {
-    auto UU = coalescedRead(U(mu));
-    mult(&phi(), &UU, &chi());
+    mult(&phi(), &U(mu), &chi());
   }
-  template<class _Spinor>
-  static accelerator_inline void multLinkAdd(_Spinor &phi,
+  static accelerator_inline void multLinkAdd(SiteSpinor &phi,
 			  const SiteDoubledGaugeField &U,
-			  const _Spinor &chi,
+			  const SiteSpinor &chi,
 			  int mu)
   {
-    auto UU = coalescedRead(U(mu));
-    mac(&phi(), &UU, &chi());
+    mac(&phi(), &U(mu), &chi());
   }
       
   template <class ref>

Grid/qcd/action/fermion/WilsonCloverFermion.h

@@ -245,7 +245,7 @@ public:
     return out;
   }
 
-protected:
+private:
   // here fixing the 4 dimensions, make it more general?
 
   RealD csw_r;                                               // Clover coefficient - spatial

Grid/qcd/action/fermion/WilsonCompressor.h

@@ -61,19 +61,18 @@ public:
   typedef typename SiteHalfSpinor::vector_type     vComplexHigh;
   constexpr static int Nw=sizeof(SiteHalfSpinor)/sizeof(vComplexHigh);
 
-  accelerator_inline int CommDatumSize(void) const {
+  accelerator_inline int CommDatumSize(void) {
     return sizeof(SiteHalfCommSpinor);
   }
 
   /*****************************************************/
   /* Compress includes precision change if mpi data is not same */
   /*****************************************************/
-  accelerator_inline void Compress(SiteHalfSpinor &buf,const SiteSpinor &in) const {
-    typedef decltype(coalescedRead(buf)) sobj;
-    sobj sp;
-    auto sin = coalescedRead(in);
-    projector::Proj(sp,sin,mu,dag);
-    coalescedWrite(buf,sp);
+  template<class _SiteHalfSpinor, class _SiteSpinor>
+  accelerator_inline void Compress(_SiteHalfSpinor *buf,Integer o,const _SiteSpinor &in) {
+    _SiteHalfSpinor tmp;
+    projector::Proj(tmp,in,mu,dag);
+    vstream(buf[o],tmp);
   }
 
   /*****************************************************/
@@ -82,24 +81,19 @@ public:
   accelerator_inline void Exchange(SiteHalfSpinor *mp,
 				   const SiteHalfSpinor * __restrict__ vp0,
 				   const SiteHalfSpinor * __restrict__ vp1,
-				   Integer type,Integer o) const {
-#ifdef GRID_SIMT
-    exchangeSIMT(mp[2*o],mp[2*o+1],vp0[o],vp1[o],type);
-#else
+				   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);
-#endif
   }
 
-
   /*****************************************************/
   /* Have a decompression step if mpi data is not same */
   /*****************************************************/
   accelerator_inline void Decompress(SiteHalfSpinor * __restrict__ out,
-				     SiteHalfSpinor * __restrict__ in, Integer o) const {    
+				     SiteHalfSpinor * __restrict__ in, Integer o) {    
     assert(0);
   }
 
@@ -109,30 +103,8 @@ public:
   accelerator_inline void CompressExchange(SiteHalfSpinor * __restrict__ out0,
 					   SiteHalfSpinor * __restrict__ out1,
 					   const SiteSpinor * __restrict__ in,
-					   Integer j,Integer k, Integer m,Integer type) const
+					   Integer j,Integer k, Integer m,Integer type)
   {
-#ifdef GRID_SIMT
-    typedef SiteSpinor vobj;
-    typedef SiteHalfSpinor hvobj;
-    typedef decltype(coalescedRead(*in))    sobj;
-    typedef decltype(coalescedRead(*out0)) hsobj;
-
-    unsigned int Nsimd = vobj::Nsimd();
-    unsigned int mask = Nsimd >> (type + 1);
-    int lane = acceleratorSIMTlane(Nsimd);
-    int j0 = lane &(~mask); // inner coor zero
-    int j1 = lane |(mask) ; // inner coor one
-    const vobj *vp0 = &in[k];
-    const vobj *vp1 = &in[m];
-    const vobj *vp = (lane&mask) ? vp1:vp0;
-    auto sa = coalescedRead(*vp,j0);
-    auto sb = coalescedRead(*vp,j1);
-    hsobj psa, psb;
-    projector::Proj(psa,sa,mu,dag);
-    projector::Proj(psb,sb,mu,dag);
-    coalescedWrite(out0[j],psa);
-    coalescedWrite(out1[j],psb);
-#else
     SiteHalfSpinor temp1, temp2;
     SiteHalfSpinor temp3, temp4;
     projector::Proj(temp1,in[k],mu,dag);
@@ -140,17 +112,15 @@ public:
     exchange(temp3,temp4,temp1,temp2,type);
     vstream(out0[j],temp3);
     vstream(out1[j],temp4);
-#endif
   }
 
   /*****************************************************/
   /* Pass the info to the stencil */
   /*****************************************************/
-  accelerator_inline bool DecompressionStep(void) const { return false; }
+  accelerator_inline bool DecompressionStep(void) { return false; }
 
 };
 
-#if 0
 template<class _HCspinor,class _Hspinor,class _Spinor, class projector>
 class WilsonCompressorTemplate< _HCspinor, _Hspinor, _Spinor, projector,
 				typename std::enable_if<!std::is_same<_HCspinor,_Hspinor>::value>::type >
@@ -172,30 +142,20 @@ public:
   typedef typename SiteHalfSpinor::vector_type     vComplexHigh;
   constexpr static int Nw=sizeof(SiteHalfSpinor)/sizeof(vComplexHigh);
 
-  accelerator_inline int CommDatumSize(void) const {
+  accelerator_inline int CommDatumSize(void) {
     return sizeof(SiteHalfCommSpinor);
   }
 
   /*****************************************************/
   /* Compress includes precision change if mpi data is not same */
   /*****************************************************/
-  accelerator_inline void Compress(SiteHalfSpinor &buf,const SiteSpinor &in) const {
-    SiteHalfSpinor hsp;
+  template<class _SiteHalfSpinor, class _SiteSpinor>
+  accelerator_inline void Compress(_SiteHalfSpinor *buf,Integer o,const _SiteSpinor &in) {
+    _SiteHalfSpinor hsp;
     SiteHalfCommSpinor *hbuf = (SiteHalfCommSpinor *)buf;
     projector::Proj(hsp,in,mu,dag);
     precisionChange((vComplexLow *)&hbuf[o],(vComplexHigh *)&hsp,Nw);
   }
-  accelerator_inline void Compress(SiteHalfSpinor &buf,const SiteSpinor &in) const {
-#ifdef GRID_SIMT
-    typedef decltype(coalescedRead(buf)) sobj;
-    sobj sp;
-    auto sin = coalescedRead(in);
-    projector::Proj(sp,sin,mu,dag);
-    coalescedWrite(buf,sp);
-#else
-    projector::Proj(buf,in,mu,dag);
-#endif
-  }
 
   /*****************************************************/
   /* Exchange includes precision change if mpi data is not same */
@@ -203,7 +163,7 @@ public:
   accelerator_inline void Exchange(SiteHalfSpinor *mp,
                        SiteHalfSpinor *vp0,
                        SiteHalfSpinor *vp1,
-		       Integer type,Integer o) const {
+		       Integer type,Integer o){
     SiteHalfSpinor vt0,vt1;
     SiteHalfCommSpinor *vpp0 = (SiteHalfCommSpinor *)vp0;
     SiteHalfCommSpinor *vpp1 = (SiteHalfCommSpinor *)vp1;
@@ -215,7 +175,7 @@ public:
   /*****************************************************/
   /* Have a decompression step if mpi data is not same */
   /*****************************************************/
-  accelerator_inline void Decompress(SiteHalfSpinor *out, SiteHalfSpinor *in, Integer o) const {
+  accelerator_inline void Decompress(SiteHalfSpinor *out, SiteHalfSpinor *in, Integer o){
     SiteHalfCommSpinor *hin=(SiteHalfCommSpinor *)in;
     precisionChange((vComplexHigh *)&out[o],(vComplexLow *)&hin[o],Nw);
   }
@@ -226,7 +186,7 @@ public:
   accelerator_inline void CompressExchange(SiteHalfSpinor *out0,
 			       SiteHalfSpinor *out1,
 			       const SiteSpinor *in,
-			       Integer j,Integer k, Integer m,Integer type) const {
+			       Integer j,Integer k, Integer m,Integer type){
     SiteHalfSpinor temp1, temp2,temp3,temp4;
     SiteHalfCommSpinor *hout0 = (SiteHalfCommSpinor *)out0;
     SiteHalfCommSpinor *hout1 = (SiteHalfCommSpinor *)out1;
@@ -240,10 +200,9 @@ public:
   /*****************************************************/
   /* Pass the info to the stencil */
   /*****************************************************/
-  accelerator_inline bool DecompressionStep(void) const { return true; }
+  accelerator_inline bool DecompressionStep(void) { return true; }
 
 };
-#endif
 
 #define DECLARE_PROJ(Projector,Compressor,spProj)			\
   class Projector {							\
@@ -294,8 +253,33 @@ public:
   typedef typename Base::View_type View_type;
   typedef typename Base::StencilVector StencilVector;
 
-  void ZeroCountersi(void)  {  }
-  void Reporti(int calls)  {  }
+  double timer0;
+  double timer1;
+  double timer2;
+  double timer3;
+  double timer4;
+  double timer5;
+  double timer6;
+  uint64_t callsi;
+  void ZeroCountersi(void)
+  {
+    timer0=0;
+    timer1=0;
+    timer2=0;
+    timer3=0;
+    timer4=0;
+    timer5=0;
+    timer6=0;
+    callsi=0;
+  }
+  void Reporti(int calls)
+  {
+    if ( timer0 ) std::cout << GridLogMessage << " timer0 (HaloGatherOpt) " <<timer0/calls <<std::endl;
+    if ( timer1 ) std::cout << GridLogMessage << " timer1 (Communicate)   " <<timer1/calls <<std::endl;
+    if ( timer2 ) std::cout << GridLogMessage << " timer2 (CommsMerge )   " <<timer2/calls <<std::endl;
+    if ( timer3 ) std::cout << GridLogMessage << " timer3 (commsMergeShm) " <<timer3/calls <<std::endl;
+    if ( timer4 ) std::cout << GridLogMessage << " timer4 " <<timer4 <<std::endl;
+  }
 
   std::vector<int> surface_list;
 
@@ -337,18 +321,26 @@ public:
   {
     std::vector<std::vector<CommsRequest_t> > reqs;
     this->HaloExchangeOptGather(source,compress);
+    double t1=usecond();
     // Asynchronous MPI calls multidirectional, Isend etc...
     // Non-overlapped directions within a thread. Asynchronous calls except MPI3, threaded up to comm threads ways.
     this->Communicate();
+    double t2=usecond(); timer1 += t2-t1;
     this->CommsMerge(compress);
+    double t3=usecond(); timer2 += t3-t2;
     this->CommsMergeSHM(compress);
+    double t4=usecond(); timer3 += t4-t3;
   }
   
   template <class compressor>
   void HaloExchangeOptGather(const Lattice<vobj> &source,compressor &compress) 
   {
     this->Prepare();
+    double t0=usecond();
     this->HaloGatherOpt(source,compress);
+    double t1=usecond();
+    timer0 += t1-t0;
+    callsi++;
   }
 
   template <class compressor>
@@ -360,9 +352,12 @@ public:
     typedef typename compressor::SiteHalfSpinor     SiteHalfSpinor;
     typedef typename compressor::SiteHalfCommSpinor SiteHalfCommSpinor;
 
+    this->mpi3synctime_g-=usecond();
     this->_grid->StencilBarrier();
+    this->mpi3synctime_g+=usecond();
 
     assert(source.Grid()==this->_grid);
+    this->halogtime-=usecond();
     
     this->u_comm_offset=0;
       
@@ -398,6 +393,7 @@ public:
     }
     this->face_table_computed=1;
     assert(this->u_comm_offset==this->_unified_buffer_size);
+    this->halogtime+=usecond();
     accelerator_barrier();
   }
 

Grid/qcd/action/fermion/WilsonImpl.h

@@ -72,7 +72,7 @@ public:
   typedef WilsonCompressor<SiteHalfCommSpinor,SiteHalfSpinor, SiteSpinor> Compressor;
   typedef WilsonImplParams ImplParams;
   typedef WilsonStencil<SiteSpinor, SiteHalfSpinor,ImplParams> StencilImpl;
-  typedef const typename StencilImpl::View_type StencilView;
+  typedef typename StencilImpl::View_type StencilView;
     
   ImplParams Params;
 
@@ -106,15 +106,11 @@ public:
 			    const _SpinorField & phi,
 			    int mu)
   {
-    const int Nsimd = SiteHalfSpinor::Nsimd();
     autoView( out_v, out, AcceleratorWrite);
     autoView( phi_v, phi, AcceleratorRead);
     autoView( Umu_v, Umu, AcceleratorRead);
-    typedef decltype(coalescedRead(out_v[0]))   calcSpinor;
-    accelerator_for(sss,out.Grid()->oSites(),Nsimd,{
-	calcSpinor tmp;
-	multLink(tmp,Umu_v[sss],phi_v(sss),mu);
-	coalescedWrite(out_v[sss],tmp);
+    accelerator_for(sss,out.Grid()->oSites(),1,{
+	multLink(out_v[sss],Umu_v[sss],phi_v[sss],mu);
     });
   }
 					   
@@ -184,22 +180,18 @@ public:
       mat = TraceIndex<SpinIndex>(P); 
     }
       
-    inline void extractLinkField(std::vector<GaugeLinkField> &mat, DoubledGaugeField &Uds)
-    {
+    inline void extractLinkField(std::vector<GaugeLinkField> &mat, DoubledGaugeField &Uds){
       for (int mu = 0; mu < Nd; mu++)
       mat[mu] = PeekIndex<LorentzIndex>(Uds, mu);
     }
 
-  inline void InsertForce5D(GaugeField &mat, FermionField &Btilde, FermionField &Atilde,int mu)
-  {
-#undef USE_OLD_INSERT_FORCE    
+
+  inline void InsertForce5D(GaugeField &mat, FermionField &Btilde, FermionField &Atilde,int mu){
+      
     int Ls=Btilde.Grid()->_fdimensions[0];
-    autoView( mat_v , mat, AcceleratorWrite);
-#ifdef USE_OLD_INSERT_FORCE    
     GaugeLinkField tmp(mat.Grid());
     tmp = Zero();
     {
-      const int Nsimd = SiteSpinor::Nsimd();
       autoView( tmp_v , tmp, AcceleratorWrite);
       autoView( Btilde_v , Btilde, AcceleratorRead);
       autoView( Atilde_v , Atilde, AcceleratorRead);
@@ -212,29 +204,6 @@ public:
 	});
     }
     PokeIndex<LorentzIndex>(mat,tmp,mu);
-#else
-    {
-      const int Nsimd = SiteSpinor::Nsimd();
-      autoView( Btilde_v , Btilde, AcceleratorRead);
-      autoView( Atilde_v , Atilde, AcceleratorRead);
-      accelerator_for(sss,mat.Grid()->oSites(),Nsimd,{
-	  int sU=sss;
-  	  typedef decltype(coalescedRead(mat_v[sU](mu)() )) ColorMatrixType;
-  	  ColorMatrixType sum;
-	  zeroit(sum);  
-	  for(int s=0;s<Ls;s++){
-	    int sF = s+Ls*sU;
-  	    for(int spn=0;spn<Ns;spn++){ //sum over spin
-  	      auto bb = coalescedRead(Btilde_v[sF]()(spn) ); //color vector
-  	      auto aa = coalescedRead(Atilde_v[sF]()(spn) );
-	      auto op = outerProduct(bb,aa);
-  	      sum = sum + op;
-	    }
-	  }
-  	  coalescedWrite(mat_v[sU](mu)(), sum);
-      });
-    }
-#endif    
   }
 };
 
@@ -243,17 +212,17 @@ typedef WilsonImpl<vComplex,  FundamentalRepresentation, CoeffReal > WilsonImplR
 typedef WilsonImpl<vComplexF, FundamentalRepresentation, CoeffReal > WilsonImplF;  // Float
 typedef WilsonImpl<vComplexD, FundamentalRepresentation, CoeffReal > WilsonImplD;  // Double
 
-//typedef WilsonImpl<vComplex,  FundamentalRepresentation, CoeffRealHalfComms > WilsonImplRL;  // Real.. whichever prec
-//typedef WilsonImpl<vComplexF, FundamentalRepresentation, CoeffRealHalfComms > WilsonImplFH;  // Float
-//typedef WilsonImpl<vComplexD, FundamentalRepresentation, CoeffRealHalfComms > WilsonImplDF;  // Double
+typedef WilsonImpl<vComplex,  FundamentalRepresentation, CoeffRealHalfComms > WilsonImplRL;  // Real.. whichever prec
+typedef WilsonImpl<vComplexF, FundamentalRepresentation, CoeffRealHalfComms > WilsonImplFH;  // Float
+typedef WilsonImpl<vComplexD, FundamentalRepresentation, CoeffRealHalfComms > WilsonImplDF;  // Double
 
 typedef WilsonImpl<vComplex,  FundamentalRepresentation, CoeffComplex > ZWilsonImplR; // Real.. whichever prec
 typedef WilsonImpl<vComplexF, FundamentalRepresentation, CoeffComplex > ZWilsonImplF; // Float
 typedef WilsonImpl<vComplexD, FundamentalRepresentation, CoeffComplex > ZWilsonImplD; // Double
 
-//typedef WilsonImpl<vComplex,  FundamentalRepresentation, CoeffComplexHalfComms > ZWilsonImplRL; // Real.. whichever prec
-//typedef WilsonImpl<vComplexF, FundamentalRepresentation, CoeffComplexHalfComms > ZWilsonImplFH; // Float
-//typedef WilsonImpl<vComplexD, FundamentalRepresentation, CoeffComplexHalfComms > ZWilsonImplDF; // Double
+typedef WilsonImpl<vComplex,  FundamentalRepresentation, CoeffComplexHalfComms > ZWilsonImplRL; // Real.. whichever prec
+typedef WilsonImpl<vComplexF, FundamentalRepresentation, CoeffComplexHalfComms > ZWilsonImplFH; // Float
+typedef WilsonImpl<vComplexD, FundamentalRepresentation, CoeffComplexHalfComms > ZWilsonImplDF; // Double
  
 typedef WilsonImpl<vComplex,  AdjointRepresentation, CoeffReal > WilsonAdjImplR;   // Real.. whichever prec
 typedef WilsonImpl<vComplexF, AdjointRepresentation, CoeffReal > WilsonAdjImplF;  // Float

Grid/qcd/action/fermion/WilsonKernels.h

@@ -49,17 +49,9 @@ public:
 
   INHERIT_IMPL_TYPES(Impl);
   typedef FermionOperator<Impl> Base;
-  typedef AcceleratorVector<int,STENCIL_MAX> StencilVector;   
+   
 public:
 
-#ifdef GRID_SYCL
-#define SYCL_HACK
-#endif  
-#ifdef SYCL_HACK
-  static void HandDhopSiteSycl(StencilVector st_perm,StencilEntry *st_p, SiteDoubledGaugeField *U,SiteHalfSpinor  *buf,
-			       int ss,int sU,const SiteSpinor *in, SiteSpinor *out);
-#endif
-  
   static void DhopKernel(int Opt,StencilImpl &st,  DoubledGaugeField &U, SiteHalfSpinor * buf,
 			 int Ls, int Nsite, const FermionField &in, FermionField &out,
 			 int interior=1,int exterior=1) ;

Grid/qcd/action/fermion/implementation/CayleyFermion5DImplementation.h

@@ -642,7 +642,7 @@ void CayleyFermion5D<Impl>::ContractConservedCurrent( PropagatorField &q_in_1,
 						      Current curr_type,
 						      unsigned int mu)
 {
-#if (!defined(GRID_HIP))
+#if (!defined(GRID_CUDA)) && (!defined(GRID_HIP))
   Gamma::Algebra Gmu [] = {
     Gamma::Algebra::GammaX,
     Gamma::Algebra::GammaY,
@@ -826,7 +826,7 @@ void CayleyFermion5D<Impl>::SeqConservedCurrent(PropagatorField &q_in,
   }
 #endif
 
-#if (!defined(GRID_HIP))
+#if (!defined(GRID_CUDA)) && (!defined(GRID_HIP))
   int tshift = (mu == Nd-1) ? 1 : 0;
   ////////////////////////////////////////////////
   // GENERAL CAYLEY CASE
@@ -880,29 +880,17 @@ void CayleyFermion5D<Impl>::SeqConservedCurrent(PropagatorField &q_in,
   }
 
   std::vector<RealD> G_s(Ls,1.0);
-  RealD sign = 1; // sign flip for vector/tadpole
   if ( curr_type == Current::Axial ) {
     for(int s=0;s<Ls/2;s++){
       G_s[s] = -1.0;
     }
   }
-  else if ( curr_type == Current::Tadpole ) {
-    auto b=this->_b;
-    auto c=this->_c;
-    if ( b == 1 && c == 0 ) {
-      sign = -1;    
-    }
-    else {
-      std::cerr << "Error: Tadpole implementation currently unavailable for non-Shamir actions." << std::endl;
-      assert(b==1 && c==0);
-    }
-  }
 
   for(int s=0;s<Ls;s++){
 
     int sp = (s+1)%Ls;
-    //    int sr = Ls-1-s;
-    //    int srp= (sr+1)%Ls;
+    int sr = Ls-1-s;
+    int srp= (sr+1)%Ls;
 
     // Mobius parameters
     auto b=this->bs[s];
@@ -919,7 +907,7 @@ void CayleyFermion5D<Impl>::SeqConservedCurrent(PropagatorField &q_in,
 
     tmp    = Cshift(tmp,mu,1);
     Impl::multLinkField(Utmp,this->Umu,tmp,mu);
-    tmp    = sign*G_s[s]*( Utmp*ph - gmu*Utmp*ph ); // Forward hop
+    tmp    = G_s[s]*( Utmp*ph - gmu*Utmp*ph ); // Forward hop
     tmp    = where((lcoor>=tmin),tmp,zz); // Mask the time 
     L_Q    = where((lcoor<=tmax),tmp,zz); // Position of current complicated
 

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

@@ -680,8 +680,7 @@ void StaggeredKernels<Impl>::DhopSiteAsm(StencilView &st,
   gauge2 =(uint64_t)&UU[sU]( Z );				\
   gauge3 =(uint64_t)&UU[sU]( T ); 
   
-#undef STAG_VEC5D
-#ifdef STAG_VEC5D
+
   // This is the single precision 5th direction vectorised kernel
 #include <Grid/simd/Intel512single.h>
 template <> void StaggeredKernels<StaggeredVec5dImplF>::DhopSiteAsm(StencilView &st,
@@ -791,7 +790,7 @@ template <> void StaggeredKernels<StaggeredVec5dImplD>::DhopSiteAsm(StencilView
 #endif
 }
    
-#endif   
+   
 
 
 #define PERMUTE_DIR3 __asm__ (	\

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

@@ -32,50 +32,25 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 
 NAMESPACE_BEGIN(Grid);
 
-#ifdef GRID_SIMT
-
-#define LOAD_CHI(ptype,b)			\
-  const SiteSpinor & ref (b[offset]);				\
-  Chi_0=coalescedReadPermute<ptype>(ref()()(0),perm,lane);	\
-  Chi_1=coalescedReadPermute<ptype>(ref()()(1),perm,lane);	\
-  Chi_2=coalescedReadPermute<ptype>(ref()()(2),perm,lane);
-
-#define LOAD_CHI_COMMS(b)		\
+#define LOAD_CHI(b)		\
   const SiteSpinor & ref (b[offset]);	\
-  Chi_0=coalescedRead(ref()()(0),lane);	\
-  Chi_1=coalescedRead(ref()()(1),lane);	\
-  Chi_2=coalescedRead(ref()()(2),lane);
-
-#define PERMUTE_DIR(dir)	;
-#else
-#define LOAD_CHI(ptype,b)      LOAD_CHI_COMMS(b)
-
-#define LOAD_CHI_COMMS(b)		\
-  const SiteSpinor & ref (b[offset]);	\
-  Chi_0=ref()()(0);			\
-  Chi_1=ref()()(1);			\
-  Chi_2=ref()()(2);
-
-#define PERMUTE_DIR(dir)			\
-  permute##dir(Chi_0,Chi_0);			\
-  permute##dir(Chi_1,Chi_1);			\
-  permute##dir(Chi_2,Chi_2);
-
-#endif
+    Chi_0=ref()()(0);\
+    Chi_1=ref()()(1);\
+    Chi_2=ref()()(2);
 
 
 // To splat or not to splat depends on the implementation
 #define MULT(A,UChi)				\
   auto & ref(U[sU](A));			\
-    U_00=coalescedRead(ref()(0,0),lane);				\
-    U_10=coalescedRead(ref()(1,0),lane);				\
-    U_20=coalescedRead(ref()(2,0),lane);				\
-    U_01=coalescedRead(ref()(0,1),lane);				\
-    U_11=coalescedRead(ref()(1,1),lane);				\
-    U_21=coalescedRead(ref()(2,1),lane);				\
-    U_02=coalescedRead(ref()(0,2),lane);				\
-    U_12=coalescedRead(ref()(1,2),lane);				\
-    U_22=coalescedRead(ref()(2,2),lane);				\
+   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;\
@@ -88,15 +63,15 @@ NAMESPACE_BEGIN(Grid);
 
 #define MULT_ADD(U,A,UChi)			\
   auto & ref(U[sU](A));			\
-    U_00=coalescedRead(ref()(0,0),lane);				\
-    U_10=coalescedRead(ref()(1,0),lane);				\
-    U_20=coalescedRead(ref()(2,0),lane);				\
-    U_01=coalescedRead(ref()(0,1),lane);				\
-    U_11=coalescedRead(ref()(1,1),lane);				\
-    U_21=coalescedRead(ref()(2,1),lane);				\
-    U_02=coalescedRead(ref()(0,2),lane);				\
-    U_12=coalescedRead(ref()(1,2),lane);				\
-    U_22=coalescedRead(ref()(2,2),lane);				\
+   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;\
@@ -108,18 +83,24 @@ NAMESPACE_BEGIN(Grid);
     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(Perm,in);						\
+    LOAD_CHI(in);					\
     if ( perm) {						\
       PERMUTE_DIR(Perm);					\
     }								\
   } else {							\
-    LOAD_CHI_COMMS(buf);					\
+    LOAD_CHI(buf);						\
   }								
 
 #define HAND_STENCIL_LEG_BEGIN(Dir,Perm,skew,even)		\
@@ -135,18 +116,19 @@ NAMESPACE_BEGIN(Grid);
   }
 
 
+
 #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(Perm,in);					\
+    LOAD_CHI(in);				\
     if ( perm) {					\
       PERMUTE_DIR(Perm);				\
     }							\
   } else if ( st.same_node[Dir] ) {			\
-    LOAD_CHI_COMMS(buf);				\
+    LOAD_CHI(buf);					\
   }							\
   if (local || st.same_node[Dir] ) {		\
     MULT_ADD(U,Dir,even);				\
@@ -158,32 +140,10 @@ NAMESPACE_BEGIN(Grid);
   local  = SE->_is_local;				\
   if ((!local) && (!st.same_node[Dir]) ) {		\
     nmu++;							\
-    { LOAD_CHI_COMMS(buf);	  }				\
+    { LOAD_CHI(buf);	  }					\
     { MULT_ADD(U,Dir,even); }					\
   }								
 
-#define HAND_DECLARATIONS(Simd) \
-  Simd even_0;			\
-  Simd even_1;			\
-  Simd even_2;			\
-  Simd odd_0;			\
-  Simd odd_1;			\
-  Simd odd_2;		        \
-		      		\
-  Simd Chi_0;			\
-  Simd Chi_1;			\
-  Simd Chi_2;			\
-				\
-  Simd U_00;			\
-  Simd U_10;			\
-  Simd U_20;			\
-  Simd U_01;			\
-  Simd U_11;			\
-  Simd U_21;			\
-  Simd U_02;			\
-  Simd U_12;			\
-  Simd U_22;			
-  
 
 template <class Impl>
 template <int Naik> accelerator_inline
@@ -195,14 +155,28 @@ void StaggeredKernels<Impl>::DhopSiteHand(StencilView &st,
   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;
 
-  const int Nsimd = SiteHalfSpinor::Nsimd();
-  const int lane=acceleratorSIMTlane(Nsimd);
-  typedef decltype( coalescedRead( in[0]()()(0) )) Simt;
-  HAND_DECLARATIONS(Simt);
+  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; 
 
-  typedef decltype( coalescedRead( in[0] )) calcSiteSpinor;
-  calcSiteSpinor result;
+  SiteSpinor result;
   int offset,local,perm, ptype;
 
   StencilEntry *SE;
@@ -241,7 +215,7 @@ void StaggeredKernels<Impl>::DhopSiteHand(StencilView &st,
       result()()(1) = even_1 + odd_1;
       result()()(2) = even_2 + odd_2;
     }
-    coalescedWrite(out[sF],result);
+    vstream(out[sF],result);
   }
 }
 
@@ -256,13 +230,28 @@ void StaggeredKernels<Impl>::DhopSiteHandInt(StencilView &st,
   typedef typename Simd::scalar_type S;
   typedef typename Simd::vector_type V;
 
-  const int Nsimd = SiteHalfSpinor::Nsimd();
-  const int lane=acceleratorSIMTlane(Nsimd);
-  typedef decltype( coalescedRead( in[0]()()(0) )) Simt;
-  HAND_DECLARATIONS(Simt);
+  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;
 
-  typedef decltype( coalescedRead( in[0] )) calcSiteSpinor;
-  calcSiteSpinor result;
+  Simd Chi_0;    // two spinor; 6 regs
+  Simd Chi_1;
+  Simd Chi_2;
+  
+  Simd U_00;  // two rows of U matrix
+  Simd U_10;
+  Simd U_20;  
+  Simd U_01;
+  Simd U_11;
+  Simd U_21;  // 2 reg left.
+  Simd U_02;
+  Simd U_12;
+  Simd U_22; 
+
+  SiteSpinor result;
   int offset, ptype, local, perm;
 
   StencilEntry *SE;
@@ -272,8 +261,8 @@ void StaggeredKernels<Impl>::DhopSiteHandInt(StencilView &st,
   //    int sF=s+LLs*sU;
   {
 
-    zeroit(even_0);    zeroit(even_1);    zeroit(even_2);
-    zeroit(odd_0);    zeroit(odd_1);    zeroit(odd_2);
+    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);  
@@ -305,7 +294,7 @@ void StaggeredKernels<Impl>::DhopSiteHandInt(StencilView &st,
       result()()(1) = even_1 + odd_1;
       result()()(2) = even_2 + odd_2;
     }
-    coalescedWrite(out[sF],result);
+    vstream(out[sF],result);
   }
 }
 
@@ -320,13 +309,28 @@ void StaggeredKernels<Impl>::DhopSiteHandExt(StencilView &st,
   typedef typename Simd::scalar_type S;
   typedef typename Simd::vector_type V;
 
-  const int Nsimd = SiteHalfSpinor::Nsimd();
-  const int lane=acceleratorSIMTlane(Nsimd);
-  typedef decltype( coalescedRead( in[0]()()(0) )) Simt;
-  HAND_DECLARATIONS(Simt);
+  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;
 
-  typedef decltype( coalescedRead( in[0] )) calcSiteSpinor;
-  calcSiteSpinor result;
+  Simd Chi_0;    // two spinor; 6 regs
+  Simd Chi_1;
+  Simd Chi_2;
+  
+  Simd U_00;  // two rows of U matrix
+  Simd U_10;
+  Simd U_20;  
+  Simd U_01;
+  Simd U_11;
+  Simd U_21;  // 2 reg left.
+  Simd U_02;
+  Simd U_12;
+  Simd U_22; 
+
+  SiteSpinor result;
   int offset, ptype, local;
 
   StencilEntry *SE;
@@ -336,8 +340,8 @@ void StaggeredKernels<Impl>::DhopSiteHandExt(StencilView &st,
   //    int sF=s+LLs*sU;
   {
 
-    zeroit(even_0);    zeroit(even_1);    zeroit(even_2);
-    zeroit(odd_0);    zeroit(odd_1);    zeroit(odd_2);
+    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);  
@@ -370,7 +374,7 @@ void StaggeredKernels<Impl>::DhopSiteHandExt(StencilView &st,
 	result()()(1) = even_1 + odd_1;
 	result()()(2) = even_2 + odd_2;
       }
-      coalescedWrite(out[sF] , out(sF)+ result);
+      out[sF] = out[sF] + result;
     }
   }
 }
@@ -393,7 +397,6 @@ void StaggeredKernels<Impl>::DhopSiteHandExt(StencilView &st,
 						     const FermionFieldView &in, FermionFieldView &out, int dag); \
 */
 #undef LOAD_CHI
-#undef HAND_DECLARATIONS
 
 NAMESPACE_END(Grid);
 

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

@@ -35,32 +35,39 @@ NAMESPACE_BEGIN(Grid);
 #define GENERIC_STENCIL_LEG(U,Dir,skew,multLink)		\
   SE = st.GetEntry(ptype, Dir+skew, sF);			\
   if (SE->_is_local ) {						\
-    int perm= SE->_permute;						\
-    chi = coalescedReadPermute(in[SE->_offset],ptype,perm,lane);\
+    if (SE->_permute) {						\
+      chi_p = χ						\
+      permute(chi,  in[SE->_offset], ptype);			\
+    } else {							\
+      chi_p = &in[SE->_offset];					\
+    }								\
   } else {							\
-    chi = coalescedRead(buf[SE->_offset],lane);			\
+    chi_p = &buf[SE->_offset];					\
   }								\
-  acceleratorSynchronise();					\
-  multLink(Uchi, U[sU], chi, Dir);			
+  multLink(Uchi, U[sU], *chi_p, Dir);			
 
 #define GENERIC_STENCIL_LEG_INT(U,Dir,skew,multLink)		\
   SE = st.GetEntry(ptype, Dir+skew, sF);			\
   if (SE->_is_local ) {						\
-    int perm= SE->_permute;						\
-    chi = coalescedReadPermute(in[SE->_offset],ptype,perm,lane);\
+    if (SE->_permute) {						\
+      chi_p = χ						\
+      permute(chi,  in[SE->_offset], ptype);			\
+    } else {							\
+      chi_p = &in[SE->_offset];					\
+    }								\
   } else if ( st.same_node[Dir] ) {				\
-    chi = coalescedRead(buf[SE->_offset],lane);                 \
+    chi_p = &buf[SE->_offset];					\
   }								\
   if (SE->_is_local || st.same_node[Dir] ) {			\
-    multLink(Uchi, U[sU], chi, Dir);				\
+    multLink(Uchi, U[sU], *chi_p, Dir);				\
   }
 
 #define GENERIC_STENCIL_LEG_EXT(U,Dir,skew,multLink)		\
   SE = st.GetEntry(ptype, Dir+skew, sF);			\
   if ((!SE->_is_local) && (!st.same_node[Dir]) ) {		\
     nmu++;							\
-    chi = coalescedRead(buf[SE->_offset],lane);			\
-    multLink(Uchi, U[sU], chi, Dir);				\
+    chi_p = &buf[SE->_offset];					\
+    multLink(Uchi, U[sU], *chi_p, Dir);				\
   }
 
 template <class Impl>
@@ -77,14 +84,12 @@ void StaggeredKernels<Impl>::DhopSiteGeneric(StencilView &st,
 					     SiteSpinor *buf, int sF, int sU, 
 					     const FermionFieldView &in, FermionFieldView &out, int dag) 
 {
-  typedef decltype(coalescedRead(in[0])) calcSpinor;
-  calcSpinor chi;
-  calcSpinor Uchi;
+  const SiteSpinor *chi_p;
+  SiteSpinor chi;
+  SiteSpinor Uchi;
   StencilEntry *SE;
   int ptype;
   int skew;
-  const int Nsimd = SiteHalfSpinor::Nsimd();
-  const int lane=acceleratorSIMTlane(Nsimd);
 
   //  for(int s=0;s<LLs;s++){
   //
@@ -113,7 +118,7 @@ void StaggeredKernels<Impl>::DhopSiteGeneric(StencilView &st,
     if ( dag ) { 
       Uchi = - Uchi;
     } 
-    coalescedWrite(out[sF], Uchi,lane);
+    vstream(out[sF], Uchi);
   }
 };
 
@@ -125,16 +130,13 @@ template <int Naik> accelerator_inline
 void StaggeredKernels<Impl>::DhopSiteGenericInt(StencilView &st, 
 						DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU,
 						SiteSpinor *buf, int sF, int sU, 
-						const FermionFieldView &in, FermionFieldView &out,int dag)
-{
-  typedef decltype(coalescedRead(in[0])) calcSpinor;
-  calcSpinor chi;
-  calcSpinor Uchi;
+						const FermionFieldView &in, FermionFieldView &out,int dag) {
+  const SiteSpinor *chi_p;
+  SiteSpinor chi;
+  SiteSpinor Uchi;
   StencilEntry *SE;
   int ptype;
   int skew ;
-  const int Nsimd = SiteHalfSpinor::Nsimd();
-  const int lane=acceleratorSIMTlane(Nsimd);
 
   //  for(int s=0;s<LLs;s++){
   //    int sF=LLs*sU+s;
@@ -163,7 +165,7 @@ void StaggeredKernels<Impl>::DhopSiteGenericInt(StencilView &st,
     if ( dag ) {
       Uchi = - Uchi;
     }
-    coalescedWrite(out[sF], Uchi,lane);
+    vstream(out[sF], Uchi);
   }
 };
 
@@ -176,17 +178,14 @@ template <int Naik> accelerator_inline
 void StaggeredKernels<Impl>::DhopSiteGenericExt(StencilView &st, 
 						DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU,
 						SiteSpinor *buf, int sF, int sU,
-						const FermionFieldView &in, FermionFieldView &out,int dag)
-{
-  typedef decltype(coalescedRead(in[0])) calcSpinor;
-  calcSpinor chi;
-  calcSpinor Uchi;
+						const FermionFieldView &in, FermionFieldView &out,int dag) {
+  const SiteSpinor *chi_p;
+  //  SiteSpinor chi;
+  SiteSpinor Uchi;
   StencilEntry *SE;
   int ptype;
   int nmu=0;
   int skew ;
-  const int Nsimd = SiteHalfSpinor::Nsimd();
-  const int lane=acceleratorSIMTlane(Nsimd);
 
   //  for(int s=0;s<LLs;s++){
   //    int sF=LLs*sU+s;
@@ -212,12 +211,11 @@ void StaggeredKernels<Impl>::DhopSiteGenericExt(StencilView &st,
     GENERIC_STENCIL_LEG_EXT(UUU,Zm,skew,Impl::multLinkAdd);
     GENERIC_STENCIL_LEG_EXT(UUU,Tm,skew,Impl::multLinkAdd);
     }
-    if ( nmu ) {
-      auto _out = coalescedRead(out[sF],lane);
-      if ( dag ) {
-	coalescedWrite(out[sF], _out-Uchi,lane);
+    if ( nmu ) { 
+      if ( dag ) { 
+	out[sF] = out[sF] - Uchi;
       } else { 
-	coalescedWrite(out[sF], _out+Uchi,lane);
+	out[sF] = out[sF] + Uchi;
       }
     }
   }
@@ -263,8 +261,6 @@ void StaggeredKernels<Impl>::DhopImproved(StencilImpl &st, LebesgueOrder &lo,
   GridBase *FGrid=in.Grid();  
   GridBase *UGrid=U.Grid();  
   typedef StaggeredKernels<Impl> ThisKernel;
-  const int Nsimd = SiteHalfSpinor::Nsimd();
-  const int lane=acceleratorSIMTlane(Nsimd);
   autoView( UUU_v , UUU, AcceleratorRead);
   autoView( U_v   ,   U, AcceleratorRead);
   autoView( in_v  ,  in, AcceleratorRead);
@@ -305,8 +301,6 @@ void StaggeredKernels<Impl>::DhopNaive(StencilImpl &st, LebesgueOrder &lo,
   GridBase *FGrid=in.Grid();  
   GridBase *UGrid=U.Grid();  
   typedef StaggeredKernels<Impl> ThisKernel;
-  const int Nsimd = SiteHalfSpinor::Nsimd();
-  const int lane=acceleratorSIMTlane(Nsimd);
   autoView( UUU_v ,   U, AcceleratorRead);
   autoView( U_v   ,   U, AcceleratorRead);
   autoView( in_v  ,  in, AcceleratorRead);

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

@@ -92,16 +92,20 @@ void WilsonCloverFermion<Impl>::ImportGauge(const GaugeField &_Umu)
   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);
+
+  Coordinate lcoor;
+  typename SiteCloverType::scalar_object Qx = Zero(), Qxinv = Zero();
+
   {
     autoView(CTv,CloverTerm,CpuRead);
     autoView(CTIv,CloverTermInv,CpuWrite);
-    thread_for(site, lvol, {
-      Coordinate lcoor;
+    for (int site = 0; site < lvol; site++) {
       grid->LocalIndexToLocalCoor(site, lcoor);
-      Eigen::MatrixXcd EigenCloverOp = Eigen::MatrixXcd::Zero(Ns * DimRep, Ns * DimRep);
-      Eigen::MatrixXcd EigenInvCloverOp = Eigen::MatrixXcd::Zero(Ns * DimRep, Ns * DimRep);
-      typename SiteCloverType::scalar_object Qx = Zero(), Qxinv = Zero();
+      EigenCloverOp = Eigen::MatrixXcd::Zero(Ns * DimRep, Ns * DimRep);
       peekLocalSite(Qx, CTv, lcoor);
+      Qxinv = Zero();
       //if (csw!=0){
       for (int j = 0; j < Ns; j++)
 	for (int k = 0; k < Ns; k++)
@@ -122,7 +126,7 @@ void WilsonCloverFermion<Impl>::ImportGauge(const GaugeField &_Umu)
       //    if (site==0) std::cout << "site =" << site << "\n" << EigenInvCloverOp << std::endl;
       //  }
       pokeLocalSite(Qxinv, CTIv, lcoor);
-    });
+    }
   }
 
   // Separate the even and odd parts

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

@@ -397,7 +397,6 @@ void WilsonFermion<Impl>::DhopDerivEO(GaugeField &mat, const FermionField &U, co
 template <class Impl>
 void WilsonFermion<Impl>::Dhop(const FermionField &in, FermionField &out, int dag)
 {
-  DhopCalls+=2;
   conformable(in.Grid(), _grid);  // verifies full grid
   conformable(in.Grid(), out.Grid());
 
@@ -409,7 +408,6 @@ void WilsonFermion<Impl>::Dhop(const FermionField &in, FermionField &out, int da
 template <class Impl>
 void WilsonFermion<Impl>::DhopOE(const FermionField &in, FermionField &out, int dag)
 {
-  DhopCalls++;
   conformable(in.Grid(), _cbgrid);    // verifies half grid
   conformable(in.Grid(), out.Grid());  // drops the cb check
 
@@ -422,7 +420,6 @@ void WilsonFermion<Impl>::DhopOE(const FermionField &in, FermionField &out, int
 template <class Impl>
 void WilsonFermion<Impl>::DhopEO(const FermionField &in, FermionField &out,int dag)
 {
-  DhopCalls++;
   conformable(in.Grid(), _cbgrid);    // verifies half grid
   conformable(in.Grid(), out.Grid());  // drops the cb check
 

Grid/qcd/action/fermion/implementation/WilsonKernelsAsmA64FX.h

@@ -38,6 +38,9 @@ Author: Nils Meyer  <nils.meyer@ur.de>  Regensburg University
 // undefine everything related to kernels
 #include <simd/Fujitsu_A64FX_undef.h>
 
+// enable A64FX body
+#define WILSONKERNELSASMBODYA64FX
+//#pragma message("A64FX Dslash: WilsonKernelsAsmBodyA64FX.h")
 
     ///////////////////////////////////////////////////////////
     // If we are A64FX specialise the single precision routine
@@ -60,89 +63,119 @@ Author: Nils Meyer  <nils.meyer@ur.de>  Regensburg University
 #define INTERIOR_AND_EXTERIOR
 #undef INTERIOR
 #undef EXTERIOR
-
-#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
 template<> void
 WilsonKernels<WilsonImplF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
-#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
 template<> void
 WilsonKernels<ZWilsonImplF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
-//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-//template<> void
-//WilsonKernels<WilsonImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-
-//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-//template<> void
-//WilsonKernels<ZWilsonImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+template<> void
+WilsonKernels<WilsonImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
+template<> void
+WilsonKernels<ZWilsonImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
 #undef INTERIOR_AND_EXTERIOR
 #define INTERIOR
 #undef EXTERIOR
-
-#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
 template<> void
 WilsonKernels<WilsonImplF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
-#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
 template<> void
 WilsonKernels<ZWilsonImplF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
-//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-//template<> void
-//WilsonKernels<WilsonImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-
-//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-//template<> void
-//WilsonKernels<ZWilsonImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+template<> void
+WilsonKernels<WilsonImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
+template<> void
+WilsonKernels<ZWilsonImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
 #undef INTERIOR_AND_EXTERIOR
 #undef INTERIOR
 #define EXTERIOR
-
-#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
 template<> void
 WilsonKernels<WilsonImplF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
-#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
 template<> void
 WilsonKernels<ZWilsonImplF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
-//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-//template<> void
-//WilsonKernels<WilsonImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-
-//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-//template<> void
-//WilsonKernels<ZWilsonImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+template<> void
+WilsonKernels<WilsonImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
+template<> void
+WilsonKernels<ZWilsonImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
 
 /////////////////////////////////////////////////////////////////
@@ -152,89 +185,119 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldVi
 #define INTERIOR_AND_EXTERIOR
 #undef INTERIOR
 #undef EXTERIOR
-
-#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
 template<> void
 WilsonKernels<WilsonImplF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
-#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
 template<> void
 WilsonKernels<ZWilsonImplF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
-//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-//template<> void
-//WilsonKernels<WilsonImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-
-//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-//template<> void
-//WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+template<> void
+WilsonKernels<WilsonImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
+template<> void
+WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
 #undef INTERIOR_AND_EXTERIOR
 #define INTERIOR
 #undef EXTERIOR
-
-#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
 template<> void
 WilsonKernels<WilsonImplF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
-#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
 template<> void
 WilsonKernels<ZWilsonImplF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
-//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-//template<> void
-//WilsonKernels<WilsonImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-
-//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-//template<> void
-//WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+template<> void
+WilsonKernels<WilsonImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
+template<> void
+WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
 #undef INTERIOR_AND_EXTERIOR
 #undef INTERIOR
 #define EXTERIOR
-
-#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
 template<> void
 WilsonKernels<WilsonImplF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
-#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
 template<> void
 WilsonKernels<ZWilsonImplF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
-//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-//template<> void
-//WilsonKernels<WilsonImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-
-//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-//template<> void
-//WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+template<> void
+WilsonKernels<WilsonImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
+template<> void
+WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
 
 // undefine
@@ -267,89 +330,119 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFiel
 #define INTERIOR_AND_EXTERIOR
 #undef INTERIOR
 #undef EXTERIOR
-
-#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
 template<> void
 WilsonKernels<WilsonImplD>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
-#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
 template<> void
 WilsonKernels<ZWilsonImplD>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
-// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-// template<> void
-// WilsonKernels<WilsonImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-// 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-
-// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-// template<> void
-// WilsonKernels<ZWilsonImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-// 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+template<> void
+WilsonKernels<WilsonImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
+template<> void
+WilsonKernels<ZWilsonImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
 #undef INTERIOR_AND_EXTERIOR
 #define INTERIOR
 #undef EXTERIOR
-
-#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
 template<> void
 WilsonKernels<WilsonImplD>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
-#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
 template<> void
 WilsonKernels<ZWilsonImplD>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
-// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-// template<> void
-// WilsonKernels<WilsonImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-// 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-
-// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-// template<> void
-// WilsonKernels<ZWilsonImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-// 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+template<> void
+WilsonKernels<WilsonImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
+template<> void
+WilsonKernels<ZWilsonImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
 #undef INTERIOR_AND_EXTERIOR
 #undef INTERIOR
 #define EXTERIOR
-
-#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
 template<> void
 WilsonKernels<WilsonImplD>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
-#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
 template<> void
 WilsonKernels<ZWilsonImplD>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
-// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-// template<> void
-// WilsonKernels<WilsonImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-// 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-
-// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-// template<> void
-// WilsonKernels<ZWilsonImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-// 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+template<> void
+WilsonKernels<WilsonImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
+template<> void
+WilsonKernels<ZWilsonImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
 /////////////////////////////////////////////////////////////////
 // XYZT vectorised, dag Kernel, double
@@ -358,93 +451,124 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldVi
 #define INTERIOR_AND_EXTERIOR
 #undef INTERIOR
 #undef EXTERIOR
-
-#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
 template<> void
 WilsonKernels<WilsonImplD>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
-#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
 template<> void
 WilsonKernels<ZWilsonImplD>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
-// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-// template<> void
-// WilsonKernels<WilsonImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-// 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-
-// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-// template<> void
-// WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-// 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+template<> void
+WilsonKernels<WilsonImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
+template<> void
+WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
 #undef INTERIOR_AND_EXTERIOR
 #define INTERIOR
 #undef EXTERIOR
-
-#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
 template<> void
 WilsonKernels<WilsonImplD>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
-#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
 template<> void
 WilsonKernels<ZWilsonImplD>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
-// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-// template<> void
-// WilsonKernels<WilsonImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-// 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-
-// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-// template<> void
-// WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-// 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+template<> void
+WilsonKernels<WilsonImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
+template<> void
+WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
 #undef INTERIOR_AND_EXTERIOR
 #undef INTERIOR
 #define EXTERIOR
-
-#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
 template<> void
 WilsonKernels<WilsonImplD>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
-#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
 template<> void
 WilsonKernels<ZWilsonImplD>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
-// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-// template<> void
-// WilsonKernels<WilsonImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-// 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-
-// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-// template<> void
-// WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-// 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+template<> void
+WilsonKernels<WilsonImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
+template<> void
+WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#if defined (WILSONKERNELSASMBODYA64FX)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#else
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+#endif
 
 
 
 // undefs
+#undef WILSONKERNELSASMBODYA64FX
 #include <simd/Fujitsu_A64FX_undef.h>
 
 #endif //A64FXASM

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

@@ -74,15 +74,15 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
-//template<> void
-//WilsonKernels<WilsonImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//
-//template<> void
-//WilsonKernels<ZWilsonImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<WilsonImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+
+template<> void 
+WilsonKernels<ZWilsonImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
 #undef INTERIOR_AND_EXTERIOR
 #define INTERIOR
@@ -97,15 +97,15 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldVi
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
-//template<> void
-//WilsonKernels<WilsonImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//
-//template<> void
-//WilsonKernels<ZWilsonImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<WilsonImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+
+template<> void 
+WilsonKernels<ZWilsonImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
 
 #undef INTERIOR_AND_EXTERIOR
@@ -121,15 +121,15 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldVi
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
-//template<> void
-//WilsonKernels<WilsonImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//
-//template<> void
-//WilsonKernels<ZWilsonImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<WilsonImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+
+template<> void 
+WilsonKernels<ZWilsonImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
       
 /////////////////////////////////////////////////////////////////
 // XYZT vectorised, dag Kernel, single
@@ -148,15 +148,15 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldVi
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
-//template<> void
-//WilsonKernels<WilsonImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//
-//template<> void
-//WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<WilsonImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+
+template<> void 
+WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
 #undef INTERIOR_AND_EXTERIOR
 #define INTERIOR
@@ -171,15 +171,15 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFiel
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
-//template<> void
-//WilsonKernels<WilsonImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//
-//template<> void
-//WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<WilsonImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+
+template<> void 
+WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
 #undef INTERIOR_AND_EXTERIOR
 #undef INTERIOR
@@ -194,15 +194,15 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFiel
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 				    
-//template<> void
-//WilsonKernels<WilsonImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//
-//template<> void
-//WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<WilsonImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+				    
+template<> void 
+WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 				    
 #undef MAYBEPERM
 #undef MULT_2SPIN
@@ -228,14 +228,14 @@ WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSite(StencilView &st, DoubledGaugeF
 							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
-//template<> void
-//WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
 #undef INTERIOR_AND_EXTERIOR
 #define INTERIOR
@@ -249,14 +249,14 @@ WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSiteInt(StencilView &st, DoubledGau
 							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
-//template<> void
-//WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
 #undef INTERIOR_AND_EXTERIOR
 #undef INTERIOR
@@ -273,15 +273,15 @@ WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSiteExt(StencilView &st, DoubledGau
 							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 				    
-//template<> void
-//WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//
-//template<> void
-//WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+				    
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 				    
 /////////////////////////////////////////////////////////////////
 // Ls vectorised, dag Kernel, single
@@ -299,14 +299,14 @@ WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSiteDag(StencilView &st, DoubledGau
 							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
-//template<> void
-//WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
-//							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
-//							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
 #undef INTERIOR_AND_EXTERIOR
 #define INTERIOR
@@ -320,14 +320,14 @@ WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSiteDagInt(StencilView &st, Doubled
 							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
-//template<> void
-//WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
-//							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
-//							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
 #undef INTERIOR_AND_EXTERIOR
 #undef INTERIOR
@@ -341,14 +341,14 @@ WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSiteDagExt(StencilView &st, Doubled
 							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
-//template<> void
-//WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
-//							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
-//							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
 #endif  // VEC 5D
 
@@ -392,14 +392,14 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
-//template<> void
-//WilsonKernels<WilsonImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZWilsonImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<WilsonImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZWilsonImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
 #undef INTERIOR_AND_EXTERIOR
 #define INTERIOR
@@ -413,14 +413,14 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldVi
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
-//template<> void
-//WilsonKernels<WilsonImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZWilsonImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<WilsonImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZWilsonImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
 #undef INTERIOR_AND_EXTERIOR
 #undef INTERIOR
@@ -434,14 +434,14 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldVi
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
       
-//template<> void
-//WilsonKernels<WilsonImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZWilsonImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<WilsonImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZWilsonImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
       
 /////////////////////////////////////////////////////////////////
 // XYZT vectorised, dag Kernel, single
@@ -459,14 +459,14 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldVi
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
-//template<> void
-//WilsonKernels<WilsonImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<WilsonImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
 #undef INTERIOR_AND_EXTERIOR
 #define INTERIOR
@@ -480,14 +480,14 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFiel
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
-//template<> void
-//WilsonKernels<WilsonImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<WilsonImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
 #undef INTERIOR_AND_EXTERIOR
 #undef INTERIOR
@@ -501,14 +501,14 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFiel
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 				    
-//template<> void
-//WilsonKernels<WilsonImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<WilsonImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 				    
 #undef MAYBEPERM
 #undef MULT_2SPIN
@@ -533,14 +533,14 @@ WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSite(StencilView &st, DoubledGaugeF
 							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
-//template<> void
-//WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
 #undef INTERIOR_AND_EXTERIOR
 #define INTERIOR
@@ -554,14 +554,14 @@ WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSiteInt(StencilView &st, DoubledGau
 							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
-//template<> void
-//WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
 #undef INTERIOR_AND_EXTERIOR
 #undef INTERIOR
@@ -577,14 +577,14 @@ WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSiteExt(StencilView &st, DoubledGau
 							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 				    
-//template<> void
-//WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 				    
 /////////////////////////////////////////////////////////////////
 // Ls vectorised, dag Kernel, single
@@ -602,14 +602,14 @@ WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSiteDag(StencilView &st, DoubledGau
 							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
-//template<> void
-//WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
-//							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
-//							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
 #undef INTERIOR_AND_EXTERIOR
 #define INTERIOR
@@ -623,14 +623,14 @@ WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSiteDagInt(StencilView &st, Doubled
 							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
-//template<> void
-//WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
-//							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
-//							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
 #undef INTERIOR_AND_EXTERIOR
 #undef INTERIOR
@@ -645,14 +645,14 @@ WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSiteDagExt(StencilView &st, Doubled
 							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
-//template<> void
-//WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
-//							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
-//							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 
 #endif  // VEC 5D
 

Grid/qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h

@@ -25,11 +25,6 @@ Author:  Nils Meyer  <nils.meyer@ur.de>  Regensburg University
     See the full license in the file "LICENSE" in the top level distribution directory
 *************************************************************************************/
 /*  END LEGAL */
-
-// GCC 10 messes up SVE instruction scheduling using -O3, but
-// -O3 -fno-schedule-insns -fno-schedule-insns2 does wonders
-// performance now is better than armclang 20.2
-
 #ifdef KERNEL_DAG
 #define DIR0_PROJ    XP_PROJ
 #define DIR1_PROJ    YP_PROJ
@@ -102,7 +97,7 @@ Author:  Nils Meyer  <nils.meyer@ur.de>  Regensburg University
     PROJ;							                        \
     MAYBEPERM(PERMUTE_DIR,perm);					        \
       } else {								                \
-	  LOAD_CHI(base);							                \
+	LOAD_CHI(base);							                \
       }									                    \
       base = st.GetInfo(ptype,local,perm,NxtDir,ent,plocal); ent++;	\
     MULT_2SPIN_1(Dir);					                    \
@@ -115,11 +110,6 @@ Author:  Nils Meyer  <nils.meyer@ur.de>  Regensburg University
     }                                                       \
     RECON;								                    \
 
-/*
-NB: picking PREFETCH_GAUGE_L2(Dir+4); here results in performance penalty
-    though I expected that it would improve on performance
-*/
-
 #define ASM_LEG_XP(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON)	    \
   base = st.GetInfo(ptype,local,perm,Dir,ent,plocal); ent++; \
   PREFETCH1_CHIMU(base);						            \
@@ -136,63 +126,73 @@ NB: picking PREFETCH_GAUGE_L2(Dir+4); here results in performance penalty
 
 #define ASM_LEG(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON)			\
       basep = st.GetPFInfo(nent,plocal); nent++;			\
-      if ( local ) {							\
-  LOAD_CHIMU(base);                                       \
-  LOAD_TABLE(PERMUTE_DIR);                                \
-  PROJ;							                        \
-  MAYBEPERM(PERMUTE_DIR,perm);					        \
-      }else if ( st.same_node[Dir] ) {LOAD_CHI(base);}			\
-      if ( local || st.same_node[Dir] ) {				\
-  MULT_2SPIN_1(Dir);					                    \
-  MULT_2SPIN_2;					                        \
-  RECON;								\
-      }									\
-  base = st.GetInfo(ptype,local,perm,NxtDir,ent,plocal); ent++;	\
-  PREFETCH_CHIMU(base);						\
-  PREFETCH_CHIMU_L2(basep);                               \
+      if ( local ) {							            \
+    LOAD_CHIMU(base);                                       \
+    LOAD_TABLE(PERMUTE_DIR);                                \
+    PROJ;							                        \
+    MAYBEPERM(PERMUTE_DIR,perm);					        \
+      }else if ( st.same_node[Dir] ) {LOAD_CHI(base);}	    \
+      base = st.GetInfo(ptype,local,perm,NxtDir,ent,plocal); ent++;	\
+      if ( local || st.same_node[Dir] ) {				    \
+    MULT_2SPIN_1(Dir);					                    \
+    PREFETCH_CHIMU(base);                                   \
+    /* PREFETCH_GAUGE_L1(NxtDir); */                        \
+    MULT_2SPIN_2;					                        \
+    if (s == 0) {                                           \
+       if ((Dir == 0) || (Dir == 4)) { PREFETCH_GAUGE_L2(Dir); } \
+    }                                                       \
+    RECON;								                    \
+    PREFETCH_CHIMU_L2(basep);                               \
+      } else { PREFETCH_CHIMU(base); }								                    \
 
 #define ASM_LEG_XP(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON)			\
   base = st.GetInfo(ptype,local,perm,Dir,ent,plocal); ent++;		\
   PREFETCH1_CHIMU(base);						\
-  { ZERO_PSI; }								\
   ASM_LEG(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON)
 
 #define RESULT(base,basep) SAVE_RESULT(base,basep);
 
 #endif
-
 ////////////////////////////////////////////////////////////////////////////////
 // Post comms kernel
 ////////////////////////////////////////////////////////////////////////////////
 #ifdef EXTERIOR
 
+
 #define ASM_LEG(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON)			\
-  base = st.GetInfo(ptype,local,perm,Dir,ent,plocal); ent++;		\
-  if((!local)&&(!st.same_node[Dir]) ) {					\
-    LOAD_CHI(base);							\
+  base = st.GetInfo(ptype,local,perm,Dir,ent,plocal); ent++; \
+  if((!local)&&(!st.same_node[Dir]) ) {					    \
+    LOAD_CHI(base);							                \
     MULT_2SPIN_1(Dir);					                    \
+    PREFETCH_CHIMU(base);                                   \
+    /* PREFETCH_GAUGE_L1(NxtDir); */                        \
     MULT_2SPIN_2;					                        \
-    RECON;								\
-    nmu++;								\
+    if (s == 0) {                                           \
+      if ((Dir == 0) || (Dir == 4)) { PREFETCH_GAUGE_L2(Dir); } \
+    }                                                       \
+    RECON;								                    \
+    nmu++;								                    \
   }
 
-#define ASM_LEG_XP(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON)			\
-  nmu=0;								\
-  { ZERO_PSI;}								\
-  base = st.GetInfo(ptype,local,perm,Dir,ent,plocal); ent++;		\
-  if((!local)&&(!st.same_node[Dir]) ) {					\
-    LOAD_CHI(base);							\
+#define ASM_LEG_XP(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON)	    \
+  nmu=0;								                    \
+  base = st.GetInfo(ptype,local,perm,Dir,ent,plocal); ent++;\
+  if((!local)&&(!st.same_node[Dir]) ) {					    \
+    LOAD_CHI(base);							                \
     MULT_2SPIN_1(Dir);					                    \
+    PREFETCH_CHIMU(base);                                   \
+    /* PREFETCH_GAUGE_L1(NxtDir); */                        \
     MULT_2SPIN_2;					                        \
-    RECON;								\
-    nmu++;								\
+    if (s == 0) {                                           \
+      if ((Dir == 0) || (Dir == 4)) { PREFETCH_GAUGE_L2(Dir); } \
+    }                                                       \
+    RECON;								                    \
+    nmu++;								                    \
   }
 
 #define RESULT(base,basep) if (nmu){ ADD_RESULT(base,base);}
 
 #endif
-
-
 {
   int nmu;
   int local,perm, ptype;
@@ -209,6 +209,7 @@ NB: picking PREFETCH_GAUGE_L2(Dir+4); here results in performance penalty
     int ssn=ssU+1;     if(ssn>=nmax) ssn=0;
     //    int sUn=lo.Reorder(ssn);
     int sUn=ssn;
+    LOCK_GAUGE(0);
 #else
     int sU =ssU;
     int ssn=ssU+1;     if(ssn>=nmax) ssn=0;
@@ -294,11 +295,6 @@ NB: picking PREFETCH_GAUGE_L2(Dir+4); here results in performance penalty
       std::cout << "----------------------------------------------------" << std::endl;
 #endif
 
-      // DC ZVA test
-      // { uint64_t basestore = (uint64_t)&out[ss];
-      //   PREFETCH_RESULT_L2_STORE(basestore); }
-
-
       ASM_LEG(Ym,Zm,PERMUTE_DIR2,DIR5_PROJ,DIR5_RECON);
 
 #ifdef SHOW
@@ -312,11 +308,6 @@ NB: picking PREFETCH_GAUGE_L2(Dir+4); here results in performance penalty
       std::cout << "----------------------------------------------------" << std::endl;
 #endif
 
-      // DC ZVA test
-      //{ uint64_t basestore = (uint64_t)&out[ss];
-      //  PREFETCH_RESULT_L2_STORE(basestore); }
-
-
       ASM_LEG(Zm,Tm,PERMUTE_DIR1,DIR6_PROJ,DIR6_RECON);
 
 #ifdef SHOW
@@ -330,11 +321,6 @@ NB: picking PREFETCH_GAUGE_L2(Dir+4); here results in performance penalty
       std::cout << "----------------------------------------------------" << std::endl;
 #endif
 
-      // DC ZVA test
-      //{ uint64_t basestore = (uint64_t)&out[ss];
-      //  PREFETCH_RESULT_L2_STORE(basestore); }
-
-
       ASM_LEG(Tm,Xp,PERMUTE_DIR0,DIR7_PROJ,DIR7_RECON);
 
 #ifdef SHOW
@@ -355,7 +341,6 @@ NB: picking PREFETCH_GAUGE_L2(Dir+4); here results in performance penalty
       base = (uint64_t) &out[ss];
       basep= st.GetPFInfo(nent,plocal); ent++;
       basep = (uint64_t) &out[ssn];
-      //PREFETCH_RESULT_L1_STORE(base);
       RESULT(base,basep);
 
 #ifdef SHOW

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

@@ -76,24 +76,7 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 
 #define REGISTER
 
-#ifdef GRID_SIMT
-#define LOAD_CHIMU(ptype)		\
-  {const SiteSpinor & ref (in[offset]);	\
-    Chimu_00=coalescedReadPermute<ptype>(ref()(0)(0),perm,lane);	\
-    Chimu_01=coalescedReadPermute<ptype>(ref()(0)(1),perm,lane);		\
-    Chimu_02=coalescedReadPermute<ptype>(ref()(0)(2),perm,lane);		\
-    Chimu_10=coalescedReadPermute<ptype>(ref()(1)(0),perm,lane);		\
-    Chimu_11=coalescedReadPermute<ptype>(ref()(1)(1),perm,lane);		\
-    Chimu_12=coalescedReadPermute<ptype>(ref()(1)(2),perm,lane);		\
-    Chimu_20=coalescedReadPermute<ptype>(ref()(2)(0),perm,lane);		\
-    Chimu_21=coalescedReadPermute<ptype>(ref()(2)(1),perm,lane);		\
-    Chimu_22=coalescedReadPermute<ptype>(ref()(2)(2),perm,lane);		\
-    Chimu_30=coalescedReadPermute<ptype>(ref()(3)(0),perm,lane);		\
-    Chimu_31=coalescedReadPermute<ptype>(ref()(3)(1),perm,lane);		\
-    Chimu_32=coalescedReadPermute<ptype>(ref()(3)(2),perm,lane);	}
-#define PERMUTE_DIR(dir) ;
-#else
-#define LOAD_CHIMU(ptype)		\
+#define LOAD_CHIMU \
   {const SiteSpinor & ref (in[offset]);	\
     Chimu_00=ref()(0)(0);\
     Chimu_01=ref()(0)(1);\
@@ -108,54 +91,54 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
     Chimu_31=ref()(3)(1);\
     Chimu_32=ref()(3)(2);}
 
-#define PERMUTE_DIR(dir)			\
-  permute##dir(Chi_00,Chi_00);	\
-      permute##dir(Chi_01,Chi_01);\
-      permute##dir(Chi_02,Chi_02);\
-      permute##dir(Chi_10,Chi_10);	\
-      permute##dir(Chi_11,Chi_11);\
-      permute##dir(Chi_12,Chi_12);
-
-#endif
+#define LOAD_CHI\
+  {const SiteHalfSpinor &ref(buf[offset]);	\
+    Chi_00 = ref()(0)(0);\
+    Chi_01 = ref()(0)(1);\
+    Chi_02 = ref()(0)(2);\
+    Chi_10 = ref()(1)(0);\
+    Chi_11 = ref()(1)(1);\
+    Chi_12 = ref()(1)(2);}
 
+// To splat or not to splat depends on the implementation
 #define MULT_2SPIN(A)\
-  {auto & ref(U[sU](A));						\
-    U_00=coalescedRead(ref()(0,0),lane);				\
-    U_10=coalescedRead(ref()(1,0),lane);				\
-    U_20=coalescedRead(ref()(2,0),lane);				\
-    U_01=coalescedRead(ref()(0,1),lane);				\
-    U_11=coalescedRead(ref()(1,1),lane);				\
-    U_21=coalescedRead(ref()(2,1),lane);				\
-    UChi_00 = U_00*Chi_00;						\
-    UChi_10 = U_00*Chi_10;						\
-    UChi_01 = U_10*Chi_00;						\
-    UChi_11 = U_10*Chi_10;						\
-    UChi_02 = U_20*Chi_00;						\
-    UChi_12 = U_20*Chi_10;						\
-    UChi_00+= U_01*Chi_01;						\
-    UChi_10+= U_01*Chi_11;						\
-    UChi_01+= U_11*Chi_01;						\
-    UChi_11+= U_11*Chi_11;						\
-    UChi_02+= U_21*Chi_01;						\
-    UChi_12+= U_21*Chi_11;						\
-    U_00=coalescedRead(ref()(0,2),lane);				\
-    U_10=coalescedRead(ref()(1,2),lane);				\
-    U_20=coalescedRead(ref()(2,2),lane);				\
-    UChi_00+= U_00*Chi_02;						\
-    UChi_10+= U_00*Chi_12;						\
-    UChi_01+= U_10*Chi_02;						\
-    UChi_11+= U_10*Chi_12;						\
-    UChi_02+= U_20*Chi_02;						\
+  {auto & ref(U[sU](A));			\
+   Impl::loadLinkElement(U_00,ref()(0,0));	\
+   Impl::loadLinkElement(U_10,ref()(1,0));	\
+   Impl::loadLinkElement(U_20,ref()(2,0));	\
+   Impl::loadLinkElement(U_01,ref()(0,1));	\
+   Impl::loadLinkElement(U_11,ref()(1,1));	\
+   Impl::loadLinkElement(U_21,ref()(2,1));	\
+    UChi_00 = U_00*Chi_00;\
+    UChi_10 = U_00*Chi_10;\
+    UChi_01 = U_10*Chi_00;\
+    UChi_11 = U_10*Chi_10;\
+    UChi_02 = U_20*Chi_00;\
+    UChi_12 = U_20*Chi_10;\
+    UChi_00+= U_01*Chi_01;\
+    UChi_10+= U_01*Chi_11;\
+    UChi_01+= U_11*Chi_01;\
+    UChi_11+= U_11*Chi_11;\
+    UChi_02+= U_21*Chi_01;\
+    UChi_12+= U_21*Chi_11;\
+    Impl::loadLinkElement(U_00,ref()(0,2));	\
+    Impl::loadLinkElement(U_10,ref()(1,2));	\
+    Impl::loadLinkElement(U_20,ref()(2,2));	\
+    UChi_00+= U_00*Chi_02;\
+    UChi_10+= U_00*Chi_12;\
+    UChi_01+= U_10*Chi_02;\
+    UChi_11+= U_10*Chi_12;\
+    UChi_02+= U_20*Chi_02;\
     UChi_12+= U_20*Chi_12;}
 
-#define LOAD_CHI				\
-  {const SiteHalfSpinor &ref(buf[offset]);	\
-    Chi_00 = coalescedRead(ref()(0)(0),lane);	\
-    Chi_01 = coalescedRead(ref()(0)(1),lane);	\
-    Chi_02 = coalescedRead(ref()(0)(2),lane);	\
-    Chi_10 = coalescedRead(ref()(1)(0),lane);	\
-    Chi_11 = coalescedRead(ref()(1)(1),lane);	\
-    Chi_12 = coalescedRead(ref()(1)(2),lane);}
+
+#define PERMUTE_DIR(dir)			\
+      permute##dir(Chi_00,Chi_00);\
+      permute##dir(Chi_01,Chi_01);\
+      permute##dir(Chi_02,Chi_02);\
+      permute##dir(Chi_10,Chi_10);\
+      permute##dir(Chi_11,Chi_11);\
+      permute##dir(Chi_12,Chi_12);
 
 //      hspin(0)=fspin(0)+timesI(fspin(3));
 //      hspin(1)=fspin(1)+timesI(fspin(2));
@@ -370,13 +353,13 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
   result_31-= UChi_11;	\
   result_32-= UChi_12;
 
-#define HAND_STENCIL_LEGB(PROJ,PERM,DIR,RECON)	\
+#define HAND_STENCIL_LEG(PROJ,PERM,DIR,RECON)	\
   SE=st.GetEntry(ptype,DIR,ss);			\
   offset = SE->_offset;				\
   local  = SE->_is_local;			\
   perm   = SE->_permute;			\
   if ( local ) {				\
-    LOAD_CHIMU(PERM);				\
+    LOAD_CHIMU;					\
     PROJ;					\
     if ( perm) {				\
       PERMUTE_DIR(PERM);			\
@@ -384,37 +367,6 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
   } else {					\
     LOAD_CHI;					\
   }						\
-  acceleratorSynchronise();			\
-  MULT_2SPIN(DIR);				\
-  RECON;					
-
-#define HAND_STENCIL_LEG(PROJ,PERM,DIR,RECON)	\
-  SE=&st_p[DIR+8*ss];				\
-  ptype=st_perm[DIR];				\
-  offset = SE->_offset;				\
-  local  = SE->_is_local;			\
-  perm   = SE->_permute;			\
-  if ( local ) {				\
-    LOAD_CHIMU(PERM);				\
-    PROJ;					\
-    if ( perm) {				\
-      PERMUTE_DIR(PERM);			\
-    }						\
-  } else {					\
-    LOAD_CHI;					\
-  }						\
-  acceleratorSynchronise();			\
-  MULT_2SPIN(DIR);				\
-  RECON;					
-
-#define HAND_STENCIL_LEGA(PROJ,PERM,DIR,RECON)				\
-  SE=&st_p[DIR+8*ss];							\
-  ptype=st_perm[DIR];							\
- /*SE=st.GetEntry(ptype,DIR,ss);*/					\
-  offset = SE->_offset;				\
-  perm   = SE->_permute;			\
-  LOAD_CHIMU(PERM);				\
-  PROJ;						\
   MULT_2SPIN(DIR);				\
   RECON;					
 
@@ -424,7 +376,7 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
   local  = SE->_is_local;			\
   perm   = SE->_permute;			\
   if ( local ) {				\
-    LOAD_CHIMU(PERM);				\
+    LOAD_CHIMU;					\
     PROJ;					\
     if ( perm) {				\
       PERMUTE_DIR(PERM);			\
@@ -432,12 +384,10 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
   } else if ( st.same_node[DIR] ) {		\
     LOAD_CHI;					\
   }						\
-  acceleratorSynchronise();			\
   if (local || st.same_node[DIR] ) {		\
     MULT_2SPIN(DIR);				\
     RECON;					\
-  }						\
-  acceleratorSynchronise();			
+  }
 
 #define HAND_STENCIL_LEG_EXT(PROJ,PERM,DIR,RECON)	\
   SE=st.GetEntry(ptype,DIR,ss);			\
@@ -447,44 +397,44 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
     MULT_2SPIN(DIR);				\
     RECON;					\
     nmu++;					\
-  }						\
-  acceleratorSynchronise();			
+  }
 
 #define HAND_RESULT(ss)				\
   {						\
-    SiteSpinor & ref (out[ss]);			\
-    coalescedWrite(ref()(0)(0),result_00,lane);		\
-    coalescedWrite(ref()(0)(1),result_01,lane);		\
-    coalescedWrite(ref()(0)(2),result_02,lane);		\
-    coalescedWrite(ref()(1)(0),result_10,lane);		\
-    coalescedWrite(ref()(1)(1),result_11,lane);		\
-    coalescedWrite(ref()(1)(2),result_12,lane);		\
-    coalescedWrite(ref()(2)(0),result_20,lane);		\
-    coalescedWrite(ref()(2)(1),result_21,lane);		\
-    coalescedWrite(ref()(2)(2),result_22,lane);		\
-    coalescedWrite(ref()(3)(0),result_30,lane);		\
-    coalescedWrite(ref()(3)(1),result_31,lane);		\
-    coalescedWrite(ref()(3)(2),result_32,lane);		\
+    SiteSpinor & ref (out[ss]);		\
+    vstream(ref()(0)(0),result_00);		\
+    vstream(ref()(0)(1),result_01);		\
+    vstream(ref()(0)(2),result_02);		\
+    vstream(ref()(1)(0),result_10);		\
+    vstream(ref()(1)(1),result_11);		\
+    vstream(ref()(1)(2),result_12);		\
+    vstream(ref()(2)(0),result_20);		\
+    vstream(ref()(2)(1),result_21);		\
+    vstream(ref()(2)(2),result_22);		\
+    vstream(ref()(3)(0),result_30);		\
+    vstream(ref()(3)(1),result_31);		\
+    vstream(ref()(3)(2),result_32);		\
   }
 
-#define HAND_RESULT_EXT(ss)				\
-  {							\
-    SiteSpinor & ref (out[ss]);				\
-    coalescedWrite(ref()(0)(0),coalescedRead(ref()(0)(0))+result_00,lane);	\
-    coalescedWrite(ref()(0)(1),coalescedRead(ref()(0)(1))+result_01,lane);	\
-    coalescedWrite(ref()(0)(2),coalescedRead(ref()(0)(2))+result_02,lane);	\
-    coalescedWrite(ref()(1)(0),coalescedRead(ref()(1)(0))+result_10,lane);	\
-    coalescedWrite(ref()(1)(1),coalescedRead(ref()(1)(1))+result_11,lane);	\
-    coalescedWrite(ref()(1)(2),coalescedRead(ref()(1)(2))+result_12,lane);	\
-    coalescedWrite(ref()(2)(0),coalescedRead(ref()(2)(0))+result_20,lane);	\
-    coalescedWrite(ref()(2)(1),coalescedRead(ref()(2)(1))+result_21,lane);	\
-    coalescedWrite(ref()(2)(2),coalescedRead(ref()(2)(2))+result_22,lane);	\
-    coalescedWrite(ref()(3)(0),coalescedRead(ref()(3)(0))+result_30,lane);	\
-    coalescedWrite(ref()(3)(1),coalescedRead(ref()(3)(1))+result_31,lane);	\
-    coalescedWrite(ref()(3)(2),coalescedRead(ref()(3)(2))+result_32,lane);	\
+#define HAND_RESULT_EXT(ss)			\
+  if (nmu){					\
+    SiteSpinor & ref (out[ss]);		\
+    ref()(0)(0)+=result_00;		\
+    ref()(0)(1)+=result_01;		\
+    ref()(0)(2)+=result_02;		\
+    ref()(1)(0)+=result_10;		\
+    ref()(1)(1)+=result_11;		\
+    ref()(1)(2)+=result_12;		\
+    ref()(2)(0)+=result_20;		\
+    ref()(2)(1)+=result_21;		\
+    ref()(2)(2)+=result_22;		\
+    ref()(3)(0)+=result_30;		\
+    ref()(3)(1)+=result_31;		\
+    ref()(3)(2)+=result_32;		\
   }
 
-#define HAND_DECLARATIONS(Simd)			\
+
+#define HAND_DECLARATIONS(a)			\
   Simd result_00;				\
   Simd result_01;				\
   Simd result_02;				\
@@ -516,19 +466,19 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
   Simd U_11;					\
   Simd U_21;
 
-#define ZERO_RESULT							\
-  zeroit(result_00);							\
-  zeroit(result_01);							\
-  zeroit(result_02);							\
-  zeroit(result_10);							\
-  zeroit(result_11);							\
-  zeroit(result_12);							\
-  zeroit(result_20);							\
-  zeroit(result_21);							\
-  zeroit(result_22);							\
-  zeroit(result_30);							\
-  zeroit(result_31);							\
-  zeroit(result_32);			
+#define ZERO_RESULT				\
+  result_00=Zero();				\
+  result_01=Zero();				\
+  result_02=Zero();				\
+  result_10=Zero();				\
+  result_11=Zero();				\
+  result_12=Zero();				\
+  result_20=Zero();				\
+  result_21=Zero();				\
+  result_22=Zero();				\
+  result_30=Zero();				\
+  result_31=Zero();				\
+  result_32=Zero();			
 
 #define Chimu_00 Chi_00
 #define Chimu_01 Chi_01
@@ -545,53 +495,15 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 
 NAMESPACE_BEGIN(Grid);
 
-
-#ifdef SYCL_HACK
-template<class Impl> accelerator_inline void 
-WilsonKernels<Impl>::HandDhopSiteSycl(StencilVector st_perm,StencilEntry *st_p, SiteDoubledGaugeField *U,SiteHalfSpinor  *buf,
-				      int ss,int sU,const SiteSpinor *in, SiteSpinor *out)
-{
-// T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
-  typedef typename Simd::scalar_type S;
-  typedef typename Simd::vector_type V;
-  typedef iSinglet<Simd> vCplx;
-  //  typedef decltype( coalescedRead( vCplx()()() )) Simt;
-  typedef decltype( coalescedRead( in[0]()(0)(0) )) Simt;
-
-  const int Nsimd = SiteHalfSpinor::Nsimd();
-  const int lane=acceleratorSIMTlane(Nsimd);
-
-  HAND_DECLARATIONS(Simt);
-
-  int offset,local,perm, ptype;
-  StencilEntry *SE;
-  HAND_STENCIL_LEG(XM_PROJ,3,Xp,XM_RECON);
-  HAND_STENCIL_LEG(YM_PROJ,2,Yp,YM_RECON_ACCUM);
-  HAND_STENCIL_LEG(ZM_PROJ,1,Zp,ZM_RECON_ACCUM);
-  HAND_STENCIL_LEG(TM_PROJ,0,Tp,TM_RECON_ACCUM);
-  HAND_STENCIL_LEG(XP_PROJ,3,Xm,XP_RECON_ACCUM);
-  HAND_STENCIL_LEG(YP_PROJ,2,Ym,YP_RECON_ACCUM);
-  HAND_STENCIL_LEG(ZP_PROJ,1,Zm,ZP_RECON_ACCUM);
-  HAND_STENCIL_LEG(TP_PROJ,0,Tm,TP_RECON_ACCUM);
-  HAND_RESULT(ss);
-}
-#endif
-
 template<class Impl> accelerator_inline void 
 WilsonKernels<Impl>::HandDhopSite(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor  *buf,
 				  int ss,int sU,const FermionFieldView &in, FermionFieldView &out)
 {
-  auto st_p = st._entries_p;						
-  auto st_perm = st._permute_type;					
 // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
   typedef typename Simd::scalar_type S;
   typedef typename Simd::vector_type V;
-  typedef decltype( coalescedRead( in[0]()(0)(0) )) Simt;
 
-  const int Nsimd = SiteHalfSpinor::Nsimd();
-  const int lane=acceleratorSIMTlane(Nsimd);
-
-  HAND_DECLARATIONS(Simt);
+  HAND_DECLARATIONS(ignore);
 
   int offset,local,perm, ptype;
   StencilEntry *SE;
@@ -611,16 +523,10 @@ template<class Impl>  accelerator_inline
 void WilsonKernels<Impl>::HandDhopSiteDag(StencilView &st,DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
 					  int ss,int sU,const FermionFieldView &in, FermionFieldView &out)
 {
-  auto st_p = st._entries_p;						
-  auto st_perm = st._permute_type;					
   typedef typename Simd::scalar_type S;
   typedef typename Simd::vector_type V;
-  typedef decltype( coalescedRead( in[0]()(0)(0) )) Simt;
 
-  const int Nsimd = SiteHalfSpinor::Nsimd();
-  const int lane=acceleratorSIMTlane(Nsimd);
-
-  HAND_DECLARATIONS(Simt);
+  HAND_DECLARATIONS(ignore);
 
   StencilEntry *SE;
   int offset,local,perm, ptype;
@@ -640,17 +546,11 @@ template<class Impl>  accelerator_inline void
 WilsonKernels<Impl>::HandDhopSiteInt(StencilView &st,DoubledGaugeFieldView &U,SiteHalfSpinor  *buf,
 					  int ss,int sU,const FermionFieldView &in, FermionFieldView &out)
 {
-  auto st_p = st._entries_p;						
-  auto st_perm = st._permute_type;					
 // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
   typedef typename Simd::scalar_type S;
   typedef typename Simd::vector_type V;
-  typedef decltype( coalescedRead( in[0]()(0)(0) )) Simt;
 
-  const int Nsimd = SiteHalfSpinor::Nsimd();
-  const int lane=acceleratorSIMTlane(Nsimd);
-
-  HAND_DECLARATIONS(Simt);
+  HAND_DECLARATIONS(ignore);
 
   int offset,local,perm, ptype;
   StencilEntry *SE;
@@ -670,16 +570,10 @@ template<class Impl> accelerator_inline
 void WilsonKernels<Impl>::HandDhopSiteDagInt(StencilView &st,DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
 						  int ss,int sU,const FermionFieldView &in, FermionFieldView &out)
 {
-  auto st_p = st._entries_p;						
-  auto st_perm = st._permute_type;					
   typedef typename Simd::scalar_type S;
   typedef typename Simd::vector_type V;
-  typedef decltype( coalescedRead( in[0]()(0)(0) )) Simt;
 
-  const int Nsimd = SiteHalfSpinor::Nsimd();
-  const int lane=acceleratorSIMTlane(Nsimd);
-
-  HAND_DECLARATIONS(Simt);
+  HAND_DECLARATIONS(ignore);
 
   StencilEntry *SE;
   int offset,local,perm, ptype;
@@ -699,17 +593,11 @@ template<class Impl>  accelerator_inline void
 WilsonKernels<Impl>::HandDhopSiteExt(StencilView &st,DoubledGaugeFieldView &U,SiteHalfSpinor  *buf,
 					  int ss,int sU,const FermionFieldView &in, FermionFieldView &out)
 {
-  auto st_p = st._entries_p;						
-  auto st_perm = st._permute_type;					
 // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
   typedef typename Simd::scalar_type S;
   typedef typename Simd::vector_type V;
-  typedef decltype( coalescedRead( in[0]()(0)(0) )) Simt;
 
-  const int Nsimd = SiteHalfSpinor::Nsimd();
-  const int lane=acceleratorSIMTlane(Nsimd);
-
-  HAND_DECLARATIONS(Simt);
+  HAND_DECLARATIONS(ignore);
 
   int offset, ptype;
   StencilEntry *SE;
@@ -730,16 +618,10 @@ template<class Impl>  accelerator_inline
 void WilsonKernels<Impl>::HandDhopSiteDagExt(StencilView &st,DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
 						  int ss,int sU,const FermionFieldView &in, FermionFieldView &out)
 {
-  auto st_p = st._entries_p;						
-  auto st_perm = st._permute_type;					
   typedef typename Simd::scalar_type S;
   typedef typename Simd::vector_type V;
-  typedef decltype( coalescedRead( in[0]()(0)(0) )) Simt;
 
-  const int Nsimd = SiteHalfSpinor::Nsimd();
-  const int lane=acceleratorSIMTlane(Nsimd);
-
-  HAND_DECLARATIONS(Simt);
+  HAND_DECLARATIONS(ignore);
 
   StencilEntry *SE;
   int offset, ptype;
@@ -800,4 +682,3 @@ NAMESPACE_END(Grid);
 #undef HAND_RESULT
 #undef HAND_RESULT_INT
 #undef HAND_RESULT_EXT
-#undef HAND_DECLARATIONS

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

@@ -416,21 +416,7 @@ void WilsonKernels<Impl>::DhopDirKernel( StencilImpl &st, DoubledGaugeField &U,S
 #undef LoopBody
 }
 
-#define KERNEL_CALL_TMP(A) \
-  const uint64_t    NN = Nsite*Ls;					\
-  auto U_p = & U_v[0];							\
-  auto in_p = & in_v[0];						\
-  auto out_p = & out_v[0];						\
-  auto st_p = st_v._entries_p;						\
-  auto st_perm = st_v._permute_type;					\
-  accelerator_forNB( ss, NN, Simd::Nsimd(), {				\
-      int sF = ss;							\
-      int sU = ss/Ls;							\
-      WilsonKernels<Impl>::A(st_perm,st_p,U_p,buf,sF,sU,in_p,out_p);	\
-    });									\
-  accelerator_barrier();
-
-#define KERNEL_CALLNB(A)						\
+#define KERNEL_CALLNB(A) \
   const uint64_t    NN = Nsite*Ls;					\
   accelerator_forNB( ss, NN, Simd::Nsimd(), {				\
       int sF = ss;							\
@@ -459,24 +445,20 @@ void WilsonKernels<Impl>::DhopKernel(int Opt,StencilImpl &st,  DoubledGaugeField
 
    if( interior && exterior ) {
      if (Opt == WilsonKernelsStatic::OptGeneric    ) { KERNEL_CALL(GenericDhopSite); return;}
-#ifdef SYCL_HACK     
-     if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL_TMP(HandDhopSiteSycl);    return; }
-#else
-     if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSite);    return;}
-#endif     
 #ifndef GRID_CUDA
+     if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSite);    return;}
      if (Opt == WilsonKernelsStatic::OptInlineAsm  ) {  ASM_CALL(AsmDhopSite);    return;}
 #endif
    } else if( interior ) {
      if (Opt == WilsonKernelsStatic::OptGeneric    ) { KERNEL_CALLNB(GenericDhopSiteInt); return;}
-     if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALLNB(HandDhopSiteInt);    return;}
 #ifndef GRID_CUDA
+     if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALLNB(HandDhopSiteInt);    return;}
      if (Opt == WilsonKernelsStatic::OptInlineAsm  ) {  ASM_CALL(AsmDhopSiteInt);    return;}
 #endif
    } else if( exterior ) {
      if (Opt == WilsonKernelsStatic::OptGeneric    ) { KERNEL_CALL(GenericDhopSiteExt); return;}
-     if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSiteExt);    return;}
 #ifndef GRID_CUDA
+     if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSiteExt);    return;}
      if (Opt == WilsonKernelsStatic::OptInlineAsm  ) {  ASM_CALL(AsmDhopSiteExt);    return;}
 #endif
    }
@@ -494,20 +476,20 @@ void WilsonKernels<Impl>::DhopKernel(int Opt,StencilImpl &st,  DoubledGaugeField
 
    if( interior && exterior ) {
      if (Opt == WilsonKernelsStatic::OptGeneric    ) { KERNEL_CALL(GenericDhopSiteDag); return;}
-     if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSiteDag);    return;}
 #ifndef GRID_CUDA
+     if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSiteDag);    return;}
      if (Opt == WilsonKernelsStatic::OptInlineAsm  ) {  ASM_CALL(AsmDhopSiteDag);     return;}
 #endif
    } else if( interior ) {
      if (Opt == WilsonKernelsStatic::OptGeneric    ) { KERNEL_CALL(GenericDhopSiteDagInt); return;}
-     if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSiteDagInt);    return;}
 #ifndef GRID_CUDA
+     if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSiteDagInt);    return;}
      if (Opt == WilsonKernelsStatic::OptInlineAsm  ) {  ASM_CALL(AsmDhopSiteDagInt);     return;}
 #endif
    } else if( exterior ) {
      if (Opt == WilsonKernelsStatic::OptGeneric    ) { KERNEL_CALL(GenericDhopSiteDagExt); return;}
-     if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSiteDagExt);    return;}
 #ifndef GRID_CUDA
+     if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSiteDagExt);    return;}
      if (Opt == WilsonKernelsStatic::OptInlineAsm  ) {  ASM_CALL(AsmDhopSiteDagExt);     return;}
 #endif
    }

Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/CayleyFermion5DInstantiationGparityWilsonImplDF.cc

@@ -0,0 +1 @@
+../CayleyFermion5DInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/ContinuedFractionFermion5DInstantiationGparityWilsonImplDF.cc

@@ -0,0 +1 @@
+../ContinuedFractionFermion5DInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/DomainWallEOFAFermionInstantiationGparityWilsonImplDF.cc

@@ -0,0 +1 @@
+../DomainWallEOFAFermionInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/MobiusEOFAFermionInstantiationGparityWilsonImplDF.cc

@@ -0,0 +1 @@
+../MobiusEOFAFermionInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/PartialFractionFermion5DInstantiationGparityWilsonImplDF.cc

@@ -0,0 +1 @@
+../PartialFractionFermion5DInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonCloverFermionInstantiationGparityWilsonImplDF.cc

@@ -0,0 +1 @@
+../WilsonCloverFermionInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonFermion5DInstantiationGparityWilsonImplDF.cc

@@ -0,0 +1 @@
+../WilsonFermion5DInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonFermionInstantiationGparityWilsonImplDF.cc

@@ -0,0 +1 @@
+../WilsonFermionInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonKernelsInstantiationGparityWilsonImplDF.cc

@@ -0,0 +1 @@
+../WilsonKernelsInstantiationGparity.cc.master

Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonTMFermionInstantiationGparityWilsonImplDF.cc

@@ -0,0 +1 @@
+../WilsonTMFermionInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/impl.h

@@ -0,0 +1 @@
+#define IMPLEMENTATION GparityWilsonImplDF

Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/CayleyFermion5DInstantiationGparityWilsonImplFH.cc

@@ -0,0 +1 @@
+../CayleyFermion5DInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/ContinuedFractionFermion5DInstantiationGparityWilsonImplFH.cc

@@ -0,0 +1 @@
+../ContinuedFractionFermion5DInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/DomainWallEOFAFermionInstantiationGparityWilsonImplFH.cc

@@ -0,0 +1 @@
+../DomainWallEOFAFermionInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/MobiusEOFAFermionInstantiationGparityWilsonImplFH.cc

@@ -0,0 +1 @@
+../MobiusEOFAFermionInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/PartialFractionFermion5DInstantiationGparityWilsonImplFH.cc

@@ -0,0 +1 @@
+../PartialFractionFermion5DInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonCloverFermionInstantiationGparityWilsonImplFH.cc

@@ -0,0 +1 @@
+../WilsonCloverFermionInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonFermion5DInstantiationGparityWilsonImplFH.cc

@@ -0,0 +1 @@
+../WilsonFermion5DInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonFermionInstantiationGparityWilsonImplFH.cc

@@ -0,0 +1 @@
+../WilsonFermionInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonKernelsInstantiationGparityWilsonImplFH.cc

@@ -0,0 +1 @@
+../WilsonKernelsInstantiationGparity.cc.master

Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonTMFermionInstantiationGparityWilsonImplFH.cc

@@ -0,0 +1 @@
+../WilsonTMFermionInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/impl.h

@@ -0,0 +1 @@
+#define IMPLEMENTATION GparityWilsonImplFH

Grid/qcd/action/fermion/instantiation/WilsonImplDF/CayleyFermion5DInstantiationWilsonImplDF.cc

@@ -0,0 +1 @@
+../CayleyFermion5DInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/WilsonImplDF/ContinuedFractionFermion5DInstantiationWilsonImplDF.cc

@@ -0,0 +1 @@
+../ContinuedFractionFermion5DInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/WilsonImplDF/DomainWallEOFAFermionInstantiationWilsonImplDF.cc

@@ -0,0 +1 @@
+../DomainWallEOFAFermionInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/WilsonImplDF/MobiusEOFAFermionInstantiationWilsonImplDF.cc

@@ -0,0 +1 @@
+../MobiusEOFAFermionInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/WilsonImplDF/PartialFractionFermion5DInstantiationWilsonImplDF.cc

@@ -0,0 +1 @@
+../PartialFractionFermion5DInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/WilsonImplDF/WilsonCloverFermionInstantiationWilsonImplDF.cc

@@ -0,0 +1 @@
+../WilsonCloverFermionInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/WilsonImplDF/WilsonFermion5DInstantiationWilsonImplDF.cc

@@ -0,0 +1 @@
+../WilsonFermion5DInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/WilsonImplDF/WilsonFermionInstantiationWilsonImplDF.cc

@@ -0,0 +1 @@
+../WilsonFermionInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/WilsonImplDF/WilsonKernelsInstantiationWilsonImplDF.cc

@@ -2,13 +2,14 @@
 
 Grid physics library, www.github.com/paboyle/Grid
 
-Source file: ./lib/qcd/action/gauge/Gauge.cc
+Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
 
-Copyright (C) 2020
+Copyright (C) 2015, 2020
 
 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: Nils Meyer <nils.meyer@ur.de> Regensburg University
 
 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
@@ -29,10 +30,22 @@ directory
 *************************************************************************************/
 /*  END LEGAL */
 #include <Grid/qcd/action/fermion/FermionCore.h>
+#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
+#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
+
+#ifndef AVX512
+#ifndef QPX
+#ifndef A64FX
+#ifndef A64FXFIXEDSIZE
+#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
+#endif
+#endif
+#endif
+#endif
 
 NAMESPACE_BEGIN(Grid);
 
-std::vector<int> ConjugateGaugeImplBase::_conjDirs;
+#include "impl.h"
+template class WilsonKernels<IMPLEMENTATION>;
 
 NAMESPACE_END(Grid);
-

Grid/qcd/action/fermion/instantiation/WilsonImplDF/WilsonTMFermionInstantiationWilsonImplDF.cc

@@ -0,0 +1 @@
+../WilsonTMFermionInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/WilsonImplDF/impl.h

@@ -0,0 +1 @@
+#define IMPLEMENTATION WilsonImplDF

Grid/qcd/action/fermion/instantiation/WilsonImplFH/CayleyFermion5DInstantiationWilsonImplFH.cc

@@ -0,0 +1 @@
+../CayleyFermion5DInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/WilsonImplFH/ContinuedFractionFermion5DInstantiationWilsonImplFH.cc

@@ -0,0 +1 @@
+../ContinuedFractionFermion5DInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/WilsonImplFH/DomainWallEOFAFermionInstantiationWilsonImplFH.cc

@@ -0,0 +1 @@
+../DomainWallEOFAFermionInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/WilsonImplFH/MobiusEOFAFermionInstantiationWilsonImplFH.cc

@@ -0,0 +1 @@
+../MobiusEOFAFermionInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/WilsonImplFH/PartialFractionFermion5DInstantiationWilsonImplFH.cc

@@ -0,0 +1 @@
+../PartialFractionFermion5DInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/WilsonImplFH/WilsonCloverFermionInstantiationWilsonImplFH.cc

@@ -0,0 +1 @@
+../WilsonCloverFermionInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/WilsonImplFH/WilsonFermion5DInstantiationWilsonImplFH.cc

@@ -0,0 +1 @@
+../WilsonFermion5DInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/WilsonImplFH/WilsonFermionInstantiationWilsonImplFH.cc

@@ -0,0 +1 @@
+../WilsonFermionInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/WilsonImplFH/WilsonKernelsInstantiationWilsonImplFH.cc

@@ -2,11 +2,14 @@
 
 Grid physics library, www.github.com/paboyle/Grid
 
-Source file: ./lib/serialisation/BaseIO.h
+Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
 
-Copyright (C) 2015
+Copyright (C) 2015, 2020
 
-Author: Michael Marshall <michael.marshall@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: Nils Meyer <nils.meyer@ur.de> Regensburg University
 
 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
@@ -22,14 +25,27 @@ 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
+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/implementation/WilsonKernelsImplementation.h>
+#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
 
-#include <Grid/GridCore.h>
+#ifndef AVX512
+#ifndef QPX
+#ifndef A64FX
+#ifndef A64FXFIXEDSIZE
+#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
+#endif
+#endif
+#endif
+#endif
 
-NAMESPACE_BEGIN(Grid)
+NAMESPACE_BEGIN(Grid);
 
-std::uint64_t EigenIO::EigenResizeCounter(0);
+#include "impl.h"
+template class WilsonKernels<IMPLEMENTATION>;
 
-NAMESPACE_END(Grid)
+NAMESPACE_END(Grid);

Grid/qcd/action/fermion/instantiation/WilsonImplFH/WilsonTMFermionInstantiationWilsonImplFH.cc

@@ -0,0 +1 @@
+../WilsonTMFermionInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/WilsonImplFH/impl.h

@@ -0,0 +1 @@
+#define IMPLEMENTATION WilsonImplFH