Switching Block lanczos precision to explicitly single

Adding sample run script and input file
Checking block lanczos
2026-10-25 07:03:19 +00:00 · 2022-03-08 10:18:36 -08:00 · 2021-09-05 23:04:41 -04:00 · 2021-09-03 17:38:10 -04:00 · 2020-10-01 15:35:29 -04:00 · 2020-10-01 15:35:03 -04:00
511 changed files with 18532 additions and 43109 deletions
@@ -10,6 +10,8 @@
 *~
 *#
 *.sublime-*
+.ctags
+tags

 # Precompiled Headers #
 #######################
@@ -88,7 +90,6 @@ Thumbs.db
 # build directory #
 ###################
 build*/*
-Documentation/_build

 # IDE related files #
 #####################
@@ -44,22 +44,14 @@ directory
 #ifdef __NVCC__
 //disables nvcc specific warning in json.hpp
 #pragma clang diagnostic ignored "-Wdeprecated-register"
-
-#if (__CUDACC_VER_MAJOR__ >= 11) && (__CUDACC_VER_MINOR__ >= 5)
- //disables nvcc specific warning in json.hpp
-#pragma nv_diag_suppress unsigned_compare_with_zero
-#pragma nv_diag_suppress cast_to_qualified_type
- //disables nvcc specific warning in many files
-#pragma nv_diag_suppress esa_on_defaulted_function_ignored
-#pragma nv_diag_suppress extra_semicolon
-#else
- //disables nvcc specific warning in json.hpp
 #pragma diag_suppress unsigned_compare_with_zero
 #pragma diag_suppress cast_to_qualified_type
+
 //disables nvcc specific warning in many files
 #pragma diag_suppress esa_on_defaulted_function_ignored
 #pragma diag_suppress extra_semicolon
-#endif
+
+//Eigen only
 #endif

 // Disable vectorisation in Eigen on the Power8/9 and PowerPC
@@ -44,10 +44,9 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 #include <Grid/GridStd.h>
 #include <Grid/threads/Pragmas.h>
 #include <Grid/perfmon/Timer.h>
-//#include <Grid/perfmon/PerfCount.h>
+#include <Grid/perfmon/PerfCount.h>
 #include <Grid/util/Util.h>
 #include <Grid/log/Log.h>
-#include <Grid/perfmon/Tracing.h>
 #include <Grid/allocator/Allocator.h>
 #include <Grid/simd/Simd.h>
 #include <Grid/threads/ThreadReduction.h>
@@ -36,7 +36,6 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 #include <Grid/GridCore.h>
 #include <Grid/qcd/QCD.h>
 #include <Grid/qcd/spin/Spin.h>
-#include <Grid/qcd/gparity/Gparity.h>
 #include <Grid/qcd/utils/Utils.h>
 #include <Grid/qcd/representations/Representations.h>
 NAMESPACE_CHECK(GridQCDCore);
@@ -16,7 +16,6 @@
 #include <functional>
 #include <stdio.h>
 #include <stdlib.h>
-#include <strings.h>
 #include <stdio.h>
 #include <signal.h>
 #include <ctime>
@@ -14,11 +14,7 @@
 /* NVCC save and restore compile environment*/
 #ifdef __NVCC__
 #pragma push
-#if (__CUDACC_VER_MAJOR__ >= 11) && (__CUDACC_VER_MINOR__ >= 5)
-#pragma nv_diag_suppress code_is_unreachable
-#else
 #pragma diag_suppress code_is_unreachable
-#endif
 #pragma push_macro("__CUDA_ARCH__")
 #pragma push_macro("__NVCC__")
 #pragma push_macro("__CUDACC__")
@@ -54,7 +54,6 @@ NAMESPACE_CHECK(BiCGSTAB);
 #include <Grid/algorithms/iterative/SchurRedBlack.h>
 #include <Grid/algorithms/iterative/ConjugateGradientMultiShift.h>
 #include <Grid/algorithms/iterative/ConjugateGradientMixedPrec.h>
-#include <Grid/algorithms/iterative/ConjugateGradientMultiShiftMixedPrec.h>
 #include <Grid/algorithms/iterative/BiCGSTABMixedPrec.h>
 #include <Grid/algorithms/iterative/BlockConjugateGradient.h>
 #include <Grid/algorithms/iterative/ConjugateGradientReliableUpdate.h>
@@ -262,7 +262,7 @@ public:
 	autoView( Tnp_v , (*Tnp), AcceleratorWrite);
 	autoView( Tnm_v , (*Tnm), AcceleratorWrite);
 	const int Nsimd = CComplex::Nsimd();
-	accelerator_for(ss, FineGrid->oSites(), Nsimd, {
+	accelerator_forNB(ss, FineGrid->oSites(), Nsimd, {
 	  coalescedWrite(y_v[ss],xscale*y_v(ss)+mscale*Tn_v(ss));
 	  coalescedWrite(Tnp_v[ss],2.0*y_v(ss)-Tnm_v(ss));
        });
@@ -324,9 +324,9 @@ public:
  GridBase*        _cbgrid;
  int hermitian;

-  CartesianStencil<siteVector,siteVector,DefaultImplParams> Stencil; 
-  CartesianStencil<siteVector,siteVector,DefaultImplParams> StencilEven;
-  CartesianStencil<siteVector,siteVector,DefaultImplParams> StencilOdd;
+  CartesianStencil<siteVector,siteVector,int> Stencil; 
+  CartesianStencil<siteVector,siteVector,int> StencilEven;
+  CartesianStencil<siteVector,siteVector,int> StencilOdd;

  std::vector<CoarseMatrix> A;
  std::vector<CoarseMatrix> Aeven;
@@ -358,7 +358,7 @@ public:
    autoView( in_v , in, AcceleratorRead);
    autoView( out_v , out, AcceleratorWrite);
    autoView( Stencil_v  , Stencil, AcceleratorRead);
-    int npoint = geom.npoint;
+    auto& geom_v = geom;
    typedef LatticeView<Cobj> Aview;
      
    Vector<Aview> AcceleratorViewContainer;
@@ -380,7 +380,7 @@ public:
      int ptype;
      StencilEntry *SE;

-      for(int point=0;point<npoint;point++){
+      for(int point=0;point<geom_v.npoint;point++){

 	SE=Stencil_v.GetEntry(ptype,point,ss);
 	  
@@ -424,7 +424,7 @@ public:
    autoView( in_v , in, AcceleratorRead);
    autoView( out_v , out, AcceleratorWrite);
    autoView( Stencil_v  , Stencil, AcceleratorRead);
-    int npoint = geom.npoint;
+    auto& geom_v = geom;
    typedef LatticeView<Cobj> Aview;

    Vector<Aview> AcceleratorViewContainer;
@@ -442,8 +442,6 @@ public:
    for(int p=0; p<geom.npoint; p++)
      points[p] = geom.points_dagger[p];

-    auto points_p = &points[0];
-
    RealD* dag_factor_p = &dag_factor[0];

    accelerator_for(sss, Grid()->oSites()*nbasis, Nsimd, {
@@ -454,8 +452,8 @@ public:
      int ptype;
      StencilEntry *SE;

-      for(int p=0;p<npoint;p++){
-        int point = points_p[p];
+      for(int p=0;p<geom_v.npoint;p++){
+        int point = points[p];

 	SE=Stencil_v.GetEntry(ptype,point,ss);

@@ -631,7 +629,7 @@ public:
    assert(Aself != nullptr);
  }

-  void DselfInternal(CartesianStencil<siteVector,siteVector,DefaultImplParams> &st, CoarseMatrix &a,
+  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);
@@ -694,7 +692,7 @@ public:
    }
  }

-  void DhopInternal(CartesianStencil<siteVector,siteVector,DefaultImplParams> &st, std::vector<CoarseMatrix> &a,
+  void DhopInternal(CartesianStencil<siteVector,siteVector,int> &st, std::vector<CoarseMatrix> &a,
                    const CoarseVector &in, CoarseVector &out, int dag) {
    SimpleCompressor<siteVector> compressor;

@@ -710,8 +708,6 @@ public:
    for(int p=0; p<npoint; p++)
      points[p] = (dag && !hermitian) ? geom.points_dagger[p] : p;

-    auto points_p = &points[0];
-
    Vector<Aview> AcceleratorViewContainer;
    for(int p=0;p<npoint;p++) AcceleratorViewContainer.push_back(a[p].View(AcceleratorRead));
    Aview *Aview_p = & AcceleratorViewContainer[0];
@@ -732,7 +728,7 @@ public:
        StencilEntry *SE;

        for(int p=0;p<npoint;p++){
-          int point = points_p[p];
+          int point = points[p];
          SE=st_v.GetEntry(ptype,point,ss);

          if(SE->_is_local) {
@@ -758,7 +754,7 @@ public:
        StencilEntry *SE;

        for(int p=0;p<npoint;p++){
-          int point = points_p[p];
+          int point = points[p];
          SE=st_v.GetEntry(ptype,point,ss);

          if(SE->_is_local) {
@@ -784,9 +780,9 @@ public:
    _cbgrid(new GridRedBlackCartesian(&CoarseGrid)),
    geom(CoarseGrid._ndimension),
    hermitian(hermitian_),
-    Stencil(&CoarseGrid,geom.npoint,Even,geom.directions,geom.displacements),
-    StencilEven(_cbgrid,geom.npoint,Even,geom.directions,geom.displacements),
-    StencilOdd(_cbgrid,geom.npoint,Odd,geom.directions,geom.displacements),
+    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),
@@ -804,9 +800,9 @@ public:
    _cbgrid(&CoarseRBGrid),
    geom(CoarseGrid._ndimension),
    hermitian(hermitian_),
-    Stencil(&CoarseGrid,geom.npoint,Even,geom.directions,geom.displacements),
-    StencilEven(&CoarseRBGrid,geom.npoint,Even,geom.directions,geom.displacements),
-    StencilOdd(&CoarseRBGrid,geom.npoint,Odd,geom.directions,geom.displacements),
+    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),
@@ -136,7 +136,7 @@ public:
    flops=0;
    usec =0;
    Coordinate layout(Nd,1);
-    sgrid = new GridCartesian(dimensions,layout,processors,*grid);
+    sgrid = new GridCartesian(dimensions,layout,processors);
  };
    
  ~FFT ( void)  {
@@ -182,7 +182,7 @@ public:
    pencil_gd[dim] = G*processors[dim];
      
    // Pencil global vol LxLxGxLxL per node
-    GridCartesian pencil_g(pencil_gd,layout,processors,*vgrid);
+    GridCartesian pencil_g(pencil_gd,layout,processors);
      
    // Construct pencils
    typedef typename vobj::scalar_object sobj;
@@ -52,7 +52,6 @@ public:
  virtual void AdjOp  (const Field &in, Field &out) = 0; // Abstract base
  virtual void HermOpAndNorm(const Field &in, Field &out,RealD &n1,RealD &n2)=0;
  virtual void HermOp(const Field &in, Field &out)=0;
-  virtual ~LinearOperatorBase(){};
 };


@@ -508,7 +507,7 @@ class SchurStaggeredOperator :  public SchurOperatorBase<Field> {
  virtual  void MpcDag   (const Field &in, Field &out){
    Mpc(in,out);
  }
-  virtual void MpcDagMpc(const Field &in, Field &out) {
+  virtual void MpcDagMpc(const Field &in, Field &out,RealD &ni,RealD &no) {
    assert(0);// Never need with staggered
  }
 };
@@ -526,22 +525,11 @@ public:
      (*this)(Linop,in[k],out[k]);
    }
  };
-  virtual ~OperatorFunction(){};
 };

 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> {
@@ -587,7 +575,6 @@ class HermOpOperatorFunction : public OperatorFunction<Field> {
 template<typename Field>
 class PlainHermOp : public LinearFunction<Field> {
 public:
-  using LinearFunction<Field>::operator();
  LinearOperatorBase<Field> &_Linop;
      
  PlainHermOp(LinearOperatorBase<Field>& linop) : _Linop(linop) 
@@ -601,7 +588,6 @@ public:
 template<typename Field>
 class FunctionHermOp : public LinearFunction<Field> {
 public:
-  using LinearFunction<Field>::operator(); 
  OperatorFunction<Field>   & _poly;
  LinearOperatorBase<Field> &_Linop;
      
@@ -30,19 +30,13 @@ Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>

 NAMESPACE_BEGIN(Grid);

-template<class Field> using Preconditioner =  LinearFunction<Field> ;
-
-/*
-template<class Field> class Preconditioner :  public LinearFunction<Field> {
-  using LinearFunction<Field>::operator();
+template<class Field> class Preconditioner :  public LinearFunction<Field> { 
  virtual void operator()(const Field &src, Field & psi)=0;
 };
-*/

 template<class Field> class TrivialPrecon :  public Preconditioner<Field> { 
 public:
-  using Preconditioner<Field>::operator();
-  virtual void operator()(const Field &src, Field & psi){
+  void operator()(const Field &src, Field & psi){
    psi = src;
  }
  TrivialPrecon(void){};
@@ -48,7 +48,6 @@ public:
  virtual  void Mdiag    (const Field &in, Field &out)=0;
  virtual  void Mdir     (const Field &in, Field &out,int dir, int disp)=0;
  virtual  void MdirAll  (const Field &in, std::vector<Field> &out)=0;
-  virtual ~SparseMatrixBase() {};
 };

 /////////////////////////////////////////////////////////////////////////////////////////////
@@ -73,7 +72,7 @@ public:
  virtual  void MeooeDag    (const Field &in, Field &out)=0;
  virtual  void MooeeDag    (const Field &in, Field &out)=0;
  virtual  void MooeeInvDag (const Field &in, Field &out)=0;
-  virtual ~CheckerBoardedSparseMatrixBase() {};
+
 };

 NAMESPACE_END(Grid);
@@ -240,12 +240,14 @@ public:
    Field T0(grid); T0 = in;  
    Field T1(grid); 
    Field T2(grid);
+    Field Tout(grid);
    Field y(grid);
      
    Field *Tnm = &T0;
    Field *Tn  = &T1;
    Field *Tnp = &T2;

+    std::cout << GridLogMessage << "Chebyshev() starts"<<std::endl;
    // Tn=T1 = (xscale M + mscale)in
    RealD xscale = 2.0/(hi-lo);
    RealD mscale = -(hi+lo)/(hi-lo);
@@ -254,16 +256,30 @@ public:

    // sum = .5 c[0] T0 + c[1] T1
    //    out = ()*T0 + Coeffs[1]*T1;
-    axpby(out,0.5*Coeffs[0],Coeffs[1],T0,T1);
+    axpby(Tout,0.5*Coeffs[0],Coeffs[1],T0,T1);
    for(int n=2;n<order;n++){

      Linop.HermOp(*Tn,y);
-      axpby(y,xscale,mscale,y,(*Tn));
-      axpby(*Tnp,2.0,-1.0,y,(*Tnm));
+#if 0
+      auto y_v = y.View();
+      auto Tn_v = Tn->View();
+      auto Tnp_v = Tnp->View();
+      auto Tnm_v = Tnm->View();
+      constexpr int Nsimd = vector_type::Nsimd();
+      accelerator_forNB(ss, in.Grid()->oSites(), Nsimd, {
+	  coalescedWrite(y_v[ss],xscale*y_v(ss)+mscale*Tn_v(ss));
+	  coalescedWrite(Tnp_v[ss],2.0*y_v(ss)-Tnm_v(ss));
+      });
      if ( Coeffs[n] != 0.0) {
 	axpy(out,Coeffs[n],*Tnp,out);
      }
-
+#else
+      axpby(y,xscale,mscale,y,(*Tn));
+      axpby(*Tnp,2.0,-1.0,y,(*Tnm));
+      if ( Coeffs[n] != 0.0) {
+	axpy(Tout,Coeffs[n],*Tnp,Tout);
+      }
+#endif
      // Cycle pointers to avoid copies
      Field *swizzle = Tnm;
      Tnm    =Tn;
@@ -271,6 +287,8 @@ public:
      Tnp    =swizzle;
 	  
    }
+    out = Tout;
+    std::cout << GridLogMessage << "Chebyshev() ends"<<std::endl;
  }
 };

@@ -363,24 +381,26 @@ public:
    Field T0(grid); T0 = in;  
    Field T1(grid); 
    Field T2(grid);
+    Field Tout(grid);
    Field  y(grid);
      
    Field *Tnm = &T0;
    Field *Tn  = &T1;
    Field *Tnp = &T2;

+    std::cout << GridLogMessage << "ChebyshevLanczos() starts"<<std::endl;
    // Tn=T1 = (xscale M )*in
    AminusMuSq(Linop,T0,T1);

    // sum = .5 c[0] T0 + c[1] T1
-    out = (0.5*Coeffs[0])*T0 + Coeffs[1]*T1;
+    Tout = (0.5*Coeffs[0])*T0 + Coeffs[1]*T1;
    for(int n=2;n<order;n++){
 	
      AminusMuSq(Linop,*Tn,y);

      *Tnp=2.0*y-(*Tnm);

-      out=out+Coeffs[n]* (*Tnp);
+      Tout=Tout+Coeffs[n]* (*Tnp);

      // Cycle pointers to avoid copies
      Field *swizzle = Tnm;
@@ -389,6 +409,8 @@ public:
      Tnp    =swizzle;
 	  
    }
+    out=Tout;
+    std::cout << GridLogMessage << "ChebyshevLanczos() ends"<<std::endl;
  }
 };
 NAMESPACE_END(Grid);
@@ -36,8 +36,7 @@ NAMESPACE_BEGIN(Grid);
 template<class FieldD, class FieldF, typename std::enable_if< getPrecision<FieldD>::value == 2, int>::type = 0, typename std::enable_if< getPrecision<FieldF>::value == 1, int>::type = 0> 
 class MixedPrecisionBiCGSTAB : public LinearFunction<FieldD> 
 {
-  public:
-    using LinearFunction<FieldD>::operator();
+  public:                                                
    RealD   Tolerance;
    RealD   InnerTolerance; // Initial tolerance for inner CG. Defaults to Tolerance but can be changed
    Integer MaxInnerIterations;
@@ -58,7 +58,6 @@ public:

  void operator()(LinearOperatorBase<Field> &Linop, const Field &src, Field &psi) {

-    GRID_TRACE("ConjugateGradient");
    psi.Checkerboard() = src.Checkerboard();

    conformable(psi, src);
@@ -118,13 +117,9 @@ public:
    GridStopWatch MatrixTimer;
    GridStopWatch SolverTimer;

-    RealD usecs = -usecond();
    SolverTimer.Start();
    int k;
    for (k = 1; k <= MaxIterations; k++) {
-
-      GridStopWatch IterationTimer;
-      IterationTimer.Start();
      c = cp;

      MatrixTimer.Start();
@@ -157,41 +152,31 @@ public:
      LinearCombTimer.Stop();
      LinalgTimer.Stop();

-      IterationTimer.Stop();
-      if ( (k % 500) == 0 ) {
-	std::cout << GridLogMessage << "ConjugateGradient: Iteration " << k
+      std::cout << GridLogIterative << "ConjugateGradient: Iteration " << k
                << " residual " << sqrt(cp/ssq) << " target " << Tolerance << std::endl;
-      } else { 
-	std::cout << GridLogIterative << "ConjugateGradient: Iteration " << k
-		  << " residual " << sqrt(cp/ssq) << " target " << Tolerance << " took " << IterationTimer.Elapsed() << std::endl;
-      }

      // Stopping condition
      if (cp <= rsq) {
-	usecs +=usecond();
        SolverTimer.Stop();
        Linop.HermOpAndNorm(psi, mmp, d, qq);
        p = mmp - src;
-	GridBase *grid = src.Grid();
-	RealD DwfFlops = (1452. )*grid->gSites()*4*k
-   	               + (8+4+8+4+4)*12*grid->gSites()*k; // CG linear algebra
+
        RealD srcnorm = std::sqrt(norm2(src));
        RealD resnorm = std::sqrt(norm2(p));
        RealD true_residual = resnorm / srcnorm;
+
        std::cout << GridLogMessage << "ConjugateGradient Converged on iteration " << k 
 		  << "\tComputed residual " << std::sqrt(cp / ssq)
 		  << "\tTrue residual " << true_residual
 		  << "\tTarget " << Tolerance << std::endl;

-        std::cout << GridLogMessage << "Time breakdown "<<std::endl;
-	std::cout << GridLogMessage << "\tElapsed    " << SolverTimer.Elapsed() <<std::endl;
-	std::cout << GridLogMessage << "\tMatrix     " << MatrixTimer.Elapsed() <<std::endl;
-	std::cout << GridLogMessage << "\tLinalg     " << LinalgTimer.Elapsed() <<std::endl;
-	std::cout << GridLogMessage << "\tInner      " << InnerTimer.Elapsed() <<std::endl;
-	std::cout << GridLogMessage << "\tAxpyNorm   " << AxpyNormTimer.Elapsed() <<std::endl;
-	std::cout << GridLogMessage << "\tLinearComb " << LinearCombTimer.Elapsed() <<std::endl;
-
-	std::cout << GridLogMessage << "\tMobius flop rate " << DwfFlops/ usecs<< " Gflops " <<std::endl;
+        std::cout << GridLogIterative << "Time breakdown "<<std::endl;
+	std::cout << GridLogIterative << "\tElapsed    " << SolverTimer.Elapsed() <<std::endl;
+	std::cout << GridLogIterative << "\tMatrix     " << MatrixTimer.Elapsed() <<std::endl;
+	std::cout << GridLogIterative << "\tLinalg     " << LinalgTimer.Elapsed() <<std::endl;
+	std::cout << GridLogIterative << "\tInner      " << InnerTimer.Elapsed() <<std::endl;
+	std::cout << GridLogIterative << "\tAxpyNorm   " << AxpyNormTimer.Elapsed() <<std::endl;
+	std::cout << GridLogIterative << "\tLinearComb " << LinearCombTimer.Elapsed() <<std::endl;

        if (ErrorOnNoConverge) assert(true_residual / Tolerance < 10000.0);

@@ -35,8 +35,7 @@ NAMESPACE_BEGIN(Grid);
    typename std::enable_if< getPrecision<FieldD>::value == 2, int>::type = 0,
    typename std::enable_if< getPrecision<FieldF>::value == 1, int>::type = 0> 
  class MixedPrecisionConjugateGradient : public LinearFunction<FieldD> {
-  public:
-    using LinearFunction<FieldD>::operator();
+  public:                                                
    RealD   Tolerance;
    RealD   InnerTolerance; //Initial tolerance for inner CG. Defaults to Tolerance but can be changed
    Integer MaxInnerIterations;
@@ -49,7 +48,6 @@ NAMESPACE_BEGIN(Grid);
    Integer TotalInnerIterations; //Number of inner CG iterations
    Integer TotalOuterIterations; //Number of restarts
    Integer TotalFinalStepIterations; //Number of CG iterations in final patch-up step
-    RealD TrueResidual;

    //Option to speed up *inner single precision* solves using a LinearFunction that produces a guess
    LinearFunction<FieldF> *guesser;
@@ -69,7 +67,6 @@ NAMESPACE_BEGIN(Grid);
    }
  
  void operator() (const FieldD &src_d_in, FieldD &sol_d){
-    std::cout << GridLogMessage << "MixedPrecisionConjugateGradient: Starting mixed precision CG with outer tolerance " << Tolerance << " and inner tolerance " << InnerTolerance << std::endl;
    TotalInnerIterations = 0;
 	
    GridStopWatch TotalTimer;
@@ -99,7 +96,6 @@ NAMESPACE_BEGIN(Grid);
    FieldF sol_f(SinglePrecGrid);
    sol_f.Checkerboard() = cb;
    
-    std::cout<<GridLogMessage<<"MixedPrecisionConjugateGradient: Starting initial inner CG with tolerance " << inner_tol << std::endl;
    ConjugateGradient<FieldF> CG_f(inner_tol, MaxInnerIterations);
    CG_f.ErrorOnNoConverge = false;

@@ -133,7 +129,6 @@ NAMESPACE_BEGIN(Grid);
 	(*guesser)(src_f, sol_f);

      //Inner CG
-      std::cout<<GridLogMessage<<"MixedPrecisionConjugateGradient: Outer iteration " << outer_iter << " starting inner CG with tolerance " << inner_tol << std::endl;
      CG_f.Tolerance = inner_tol;
      InnerCGtimer.Start();
      CG_f(Linop_f, src_f, sol_f);
@@ -154,7 +149,6 @@ NAMESPACE_BEGIN(Grid);
    ConjugateGradient<FieldD> CG_d(Tolerance, MaxInnerIterations);
    CG_d(Linop_d, src_d_in, sol_d);
    TotalFinalStepIterations = CG_d.IterationsToComplete;
-    TrueResidual = CG_d.TrueResidual;

    TotalTimer.Stop();
    std::cout<<GridLogMessage<<"MixedPrecisionConjugateGradient: Inner CG iterations " << TotalInnerIterations << " Restarts " << TotalOuterIterations << " Final CG iterations " << TotalFinalStepIterations << std::endl;
@@ -44,7 +44,7 @@ public:

  using OperatorFunction<Field>::operator();

-  //  RealD   Tolerance;
+  RealD   Tolerance;
  Integer MaxIterations;
  Integer IterationsToComplete; //Number of iterations the CG took to finish. Filled in upon completion
  std::vector<int> IterationsToCompleteShift;  // Iterations for this shift
@@ -52,7 +52,7 @@ public:
  MultiShiftFunction shifts;
  std::vector<RealD> TrueResidualShift;

-  ConjugateGradientMultiShift(Integer maxit, const MultiShiftFunction &_shifts) : 
+  ConjugateGradientMultiShift(Integer maxit,MultiShiftFunction &_shifts) : 
    MaxIterations(maxit),
    shifts(_shifts)
  { 
@@ -84,7 +84,6 @@ public:

  void operator() (LinearOperatorBase<Field> &Linop, const Field &src, std::vector<Field> &psi)
  {
-    GRID_TRACE("ConjugateGradientMultiShift");
  
    GridBase *grid = src.Grid();
  
@@ -183,9 +182,6 @@ public:
    for(int s=0;s<nshift;s++) {
      axpby(psi[s],0.,-bs[s]*alpha[s],src,src);
    }
-
-    std::cout << GridLogIterative << "ConjugateGradientMultiShift: initial rn (|src|^2) =" << rn << " qq (|MdagM src|^2) =" << qq << " d ( dot(src, [MdagM + m_0]src) ) =" << d << " c=" << c << std::endl;
-    
  
  ///////////////////////////////////////
  // Timers
@@ -325,8 +321,8 @@ public:

      std::cout << GridLogMessage << "Time Breakdown "<<std::endl;
      std::cout << GridLogMessage << "\tElapsed    " << SolverTimer.Elapsed()     <<std::endl;
-      std::cout << GridLogMessage << "\tAXPY     " << AXPYTimer.Elapsed()     <<std::endl;
-      std::cout << GridLogMessage << "\tMatrix   " << MatrixTimer.Elapsed()     <<std::endl;
+      std::cout << GridLogMessage << "\tAXPY    " << AXPYTimer.Elapsed()     <<std::endl;
+      std::cout << GridLogMessage << "\tMarix    " << MatrixTimer.Elapsed()     <<std::endl;
      std::cout << GridLogMessage << "\tShift    " << ShiftTimer.Elapsed()     <<std::endl;

      IterationsToComplete = k;	
@@ -1,409 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/algorithms/iterative/ConjugateGradientMultiShift.h
-
-    Copyright (C) 2015
-
-Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Christopher Kelly <ckelly@bnl.gov>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#ifndef GRID_CONJUGATE_GRADIENT_MULTI_SHIFT_MIXEDPREC_H
-#define GRID_CONJUGATE_GRADIENT_MULTI_SHIFT_MIXEDPREC_H
-
-NAMESPACE_BEGIN(Grid);
-
-//CK 2020: A variant of the multi-shift conjugate gradient with the matrix multiplication in single precision. 
-//The residual is stored in single precision, but the search directions and solution are stored in double precision. 
-//Every update_freq iterations the residual is corrected in double precision. 
-    
-//For safety the a final regular CG is applied to clean up if necessary
-
-//Linop to add shift to input linop, used in cleanup CG
-namespace ConjugateGradientMultiShiftMixedPrecSupport{
-template<typename Field>
-class ShiftedLinop: public LinearOperatorBase<Field>{
-public:
-  LinearOperatorBase<Field> &linop_base;
-  RealD shift;
-
-  ShiftedLinop(LinearOperatorBase<Field> &_linop_base, RealD _shift): linop_base(_linop_base), shift(_shift){}
-
-  void OpDiag (const Field &in, Field &out){ assert(0); }
-  void OpDir  (const Field &in, Field &out,int dir,int disp){ assert(0); }
-  void OpDirAll  (const Field &in, std::vector<Field> &out){ assert(0); }
-  
-  void Op     (const Field &in, Field &out){ assert(0); }
-  void AdjOp  (const Field &in, Field &out){ assert(0); }
-
-  void HermOp(const Field &in, Field &out){
-    linop_base.HermOp(in, out);
-    axpy(out, shift, in, out);
-  }    
-
-  void HermOpAndNorm(const Field &in, Field &out,RealD &n1,RealD &n2){
-    HermOp(in,out);
-    ComplexD dot = innerProduct(in,out);
-    n1=real(dot);
-    n2=norm2(out);
-  }
-};
-};
-
-
-template<class FieldD, class FieldF,
-	 typename std::enable_if< getPrecision<FieldD>::value == 2, int>::type = 0,
-	 typename std::enable_if< getPrecision<FieldF>::value == 1, int>::type = 0> 
-class ConjugateGradientMultiShiftMixedPrec : public OperatorMultiFunction<FieldD>,
-					     public OperatorFunction<FieldD>
-{
-public:                                                
-
-  using OperatorFunction<FieldD>::operator();
-
-  RealD   Tolerance;
-  Integer MaxIterations;
-  Integer IterationsToComplete; //Number of iterations the CG took to finish. Filled in upon completion
-  std::vector<int> IterationsToCompleteShift;  // Iterations for this shift
-  int verbose;
-  MultiShiftFunction shifts;
-  std::vector<RealD> TrueResidualShift;
-
-  int ReliableUpdateFreq; //number of iterations between reliable updates
-
-  GridBase* SinglePrecGrid; //Grid for single-precision fields
-  LinearOperatorBase<FieldF> &Linop_f; //single precision
-
-  ConjugateGradientMultiShiftMixedPrec(Integer maxit, const MultiShiftFunction &_shifts,
-				       GridBase* _SinglePrecGrid, LinearOperatorBase<FieldF> &_Linop_f,
-				       int _ReliableUpdateFreq
-				       ) : 
-    MaxIterations(maxit),  shifts(_shifts), SinglePrecGrid(_SinglePrecGrid), Linop_f(_Linop_f), ReliableUpdateFreq(_ReliableUpdateFreq)
-  { 
-    verbose=1;
-    IterationsToCompleteShift.resize(_shifts.order);
-    TrueResidualShift.resize(_shifts.order);
-  }
-
-  void operator() (LinearOperatorBase<FieldD> &Linop, const FieldD &src, FieldD &psi)
-  {
-    GridBase *grid = src.Grid();
-    int nshift = shifts.order;
-    std::vector<FieldD> results(nshift,grid);
-    (*this)(Linop,src,results,psi);
-  }
-  void operator() (LinearOperatorBase<FieldD> &Linop, const FieldD &src, std::vector<FieldD> &results, FieldD &psi)
-  {
-    int nshift = shifts.order;
-
-    (*this)(Linop,src,results);
-  
-    psi = shifts.norm*src;
-    for(int i=0;i<nshift;i++){
-      psi = psi + shifts.residues[i]*results[i];
-    }
-
-    return;
-  }
-
-  void operator() (LinearOperatorBase<FieldD> &Linop_d, const FieldD &src_d, std::vector<FieldD> &psi_d)
-  { 
-    GRID_TRACE("ConjugateGradientMultiShiftMixedPrec");
-    GridBase *DoublePrecGrid = src_d.Grid();
-
-    ////////////////////////////////////////////////////////////////////////
-    // Convenience references to the info stored in "MultiShiftFunction"
-    ////////////////////////////////////////////////////////////////////////
-    int nshift = shifts.order;
-
-    std::vector<RealD> &mass(shifts.poles); // Make references to array in "shifts"
-    std::vector<RealD> &mresidual(shifts.tolerances);
-    std::vector<RealD> alpha(nshift,1.0);
-
-    //Double precision search directions
-    FieldD p_d(DoublePrecGrid);
-    std::vector<FieldD> ps_d(nshift, DoublePrecGrid);// Search directions (double precision)
-
-    FieldD tmp_d(DoublePrecGrid);
-    FieldD r_d(DoublePrecGrid);
-    FieldD mmp_d(DoublePrecGrid);
-
-    assert(psi_d.size()==nshift);
-    assert(mass.size()==nshift);
-    assert(mresidual.size()==nshift);
-  
-    // dynamic sized arrays on stack; 2d is a pain with vector
-    RealD  bs[nshift];
-    RealD  rsq[nshift];
-    RealD  rsqf[nshift];
-    RealD  z[nshift][2];
-    int     converged[nshift];
-  
-    const int       primary =0;
-  
-    //Primary shift fields CG iteration
-    RealD a,b,c,d;
-    RealD cp,bp,qq; //prev
-  
-    // Matrix mult fields
-    FieldF p_f(SinglePrecGrid);
-    FieldF mmp_f(SinglePrecGrid);
-
-    // Check lightest mass
-    for(int s=0;s<nshift;s++){
-      assert( mass[s]>= mass[primary] );
-      converged[s]=0;
-    }
-  
-    // Wire guess to zero
-    // Residuals "r" are src
-    // First search direction "p" is also src
-    cp = norm2(src_d);
-
-    // Handle trivial case of zero src.
-    if( cp == 0. ){
-      for(int s=0;s<nshift;s++){
-	psi_d[s] = Zero();
-	IterationsToCompleteShift[s] = 1;
-	TrueResidualShift[s] = 0.;
-      }
-      return;
-    }
-
-    for(int s=0;s<nshift;s++){
-      rsq[s] = cp * mresidual[s] * mresidual[s];
-      rsqf[s] =rsq[s];
-      std::cout<<GridLogMessage<<"ConjugateGradientMultiShiftMixedPrec: shift "<< s <<" target resid "<<rsq[s]<<std::endl;
-      ps_d[s] = src_d;
-    }
-    // r and p for primary
-    p_d = src_d; //primary copy --- make this a reference to ps_d to save axpys
-    r_d = p_d;
-    
-    //MdagM+m[0]
-    precisionChangeFast(p_f,p_d);
-
-    Linop_f.HermOpAndNorm(p_f,mmp_f,d,qq); // mmp = MdagM p        d=real(dot(p, mmp)),  qq=norm2(mmp)
-    precisionChangeFast(tmp_d,mmp_f);
-    Linop_d.HermOpAndNorm(p_d,mmp_d,d,qq); // mmp = MdagM p        d=real(dot(p, mmp)),  qq=norm2(mmp)
-    tmp_d = tmp_d - mmp_d;
-    std::cout << " Testing operators match "<<norm2(mmp_d)<<" f "<<norm2(mmp_f)<<" diff "<< norm2(tmp_d)<<std::endl;
-    //    assert(norm2(tmp_d)< 1.0e-4);
-
-    axpy(mmp_d,mass[0],p_d,mmp_d);
-    RealD rn = norm2(p_d);
-    d += rn*mass[0];
-
-    b = -cp /d;
-  
-    // Set up the various shift variables
-    int       iz=0;
-    z[0][1-iz] = 1.0;
-    z[0][iz]   = 1.0;
-    bs[0]      = b;
-    for(int s=1;s<nshift;s++){
-      z[s][1-iz] = 1.0;
-      z[s][iz]   = 1.0/( 1.0 - b*(mass[s]-mass[0]));
-      bs[s]      = b*z[s][iz]; 
-    }
-  
-    // r += b[0] A.p[0]
-    // c= norm(r)
-    c=axpy_norm(r_d,b,mmp_d,r_d);
-  
-    for(int s=0;s<nshift;s++) {
-      axpby(psi_d[s],0.,-bs[s]*alpha[s],src_d,src_d);
-    }
-  
-    ///////////////////////////////////////
-    // Timers
-    ///////////////////////////////////////
-    GridStopWatch AXPYTimer, ShiftTimer, QRTimer, MatrixTimer, SolverTimer, PrecChangeTimer, CleanupTimer;
-
-    SolverTimer.Start();
-  
-    // Iteration loop
-    int k;
-  
-    for (k=1;k<=MaxIterations;k++){    
-
-      a = c /cp;
-      AXPYTimer.Start();
-      axpy(p_d,a,p_d,r_d); 
-
-      for(int s=0;s<nshift;s++){
-	if ( ! converged[s] ) { 
-	  if (s==0){
-	    axpy(ps_d[s],a,ps_d[s],r_d);
-	  } else{
-	    RealD as =a *z[s][iz]*bs[s] /(z[s][1-iz]*b);
-	    axpby(ps_d[s],z[s][iz],as,r_d,ps_d[s]);
-	  }
-	}
-      }
-      AXPYTimer.Stop();
-
-      PrecChangeTimer.Start();
-      precisionChangeFast(p_f, p_d); //get back single prec search direction for linop
-      PrecChangeTimer.Stop();
-
-      cp=c;
-      MatrixTimer.Start();  
-      Linop_f.HermOp(p_f,mmp_f);
-      MatrixTimer.Stop();  
-
-      PrecChangeTimer.Start();
-      precisionChangeFast(mmp_d, mmp_f); // From Float to Double
-      PrecChangeTimer.Stop();
-
-      AXPYTimer.Start();
-      d=real(innerProduct(p_d,mmp_d));    
-      axpy(mmp_d,mass[0],p_d,mmp_d);
-      AXPYTimer.Stop();
-      RealD rn = norm2(p_d);
-      d += rn*mass[0];
-    
-      bp=b;
-      b=-cp/d;
-
-      // Toggle the recurrence history
-      bs[0] = b;
-      iz = 1-iz;
-      ShiftTimer.Start();
-      for(int s=1;s<nshift;s++){
-	if((!converged[s])){
-	  RealD z0 = z[s][1-iz];
-	  RealD z1 = z[s][iz];
-	  z[s][iz] = z0*z1*bp
-	    / (b*a*(z1-z0) + z1*bp*(1- (mass[s]-mass[0])*b)); 
-	  bs[s] = b*z[s][iz]/z0; // NB sign  rel to Mike
-	}
-      }
-      ShiftTimer.Stop();
-
-      //Update double precision solutions
-      AXPYTimer.Start();
-      for(int s=0;s<nshift;s++){
-	int ss = s;
-	if( (!converged[s]) ) { 
-	  axpy(psi_d[ss],-bs[s]*alpha[s],ps_d[s],psi_d[ss]);
-	}
-      }
-
-      //Perform reliable update if necessary; otherwise update residual from single-prec mmp
-      c = axpy_norm(r_d,b,mmp_d,r_d);
-
-      AXPYTimer.Stop();
-
-      if(k % ReliableUpdateFreq == 0){
-	RealD c_old = c;
-	//Replace r with true residual
-	MatrixTimer.Start();  
-	Linop_d.HermOp(psi_d[0],mmp_d); 
-	MatrixTimer.Stop();  
-
-	AXPYTimer.Start();
-	axpy(mmp_d,mass[0],psi_d[0],mmp_d);
-
-	c = axpy_norm(r_d, -1.0, mmp_d, src_d);
-	AXPYTimer.Stop();
-
-	std::cout<<GridLogMessage<<"ConjugateGradientMultiShiftMixedPrec k="<<k<< ", replaced |r|^2 = "<<c_old <<" with |r|^2 = "<<c<<std::endl;
-      }
-    
-      // Convergence checks
-      int all_converged = 1;
-      for(int s=0;s<nshift;s++){
-      
-	if ( (!converged[s]) ){
-	  IterationsToCompleteShift[s] = k;
-	
-	  RealD css  = c * z[s][iz]* z[s][iz];
-	
-	  if(css<rsqf[s]){
-	    if ( ! converged[s] )
-	      std::cout<<GridLogMessage<<"ConjugateGradientMultiShiftMixedPrec k="<<k<<" Shift "<<s<<" has converged"<<std::endl;
-	    converged[s]=1;
-	  } else {
-	    all_converged=0;
-	  }
-
-	}
-      }
-
-      if ( all_converged ){
-
-	SolverTimer.Stop();
-	std::cout<<GridLogMessage<< "ConjugateGradientMultiShiftMixedPrec: All shifts have converged iteration "<<k<<std::endl;
-	std::cout<<GridLogMessage<< "ConjugateGradientMultiShiftMixedPrec: Checking solutions"<<std::endl;
-      
-	// Check answers 
-	for(int s=0; s < nshift; s++) { 
-	  Linop_d.HermOpAndNorm(psi_d[s],mmp_d,d,qq);
-	  axpy(tmp_d,mass[s],psi_d[s],mmp_d);
-	  axpy(r_d,-alpha[s],src_d,tmp_d);
-	  RealD rn = norm2(r_d);
-	  RealD cn = norm2(src_d);
-	  TrueResidualShift[s] = std::sqrt(rn/cn);
-	  std::cout<<GridLogMessage<<"ConjugateGradientMultiShiftMixedPrec: shift["<<s<<"] true residual "<< TrueResidualShift[s] << " target " << mresidual[s] << std::endl;
-
-	  //If we have not reached the desired tolerance, do a (mixed precision) CG cleanup
-	  if(rn >= rsq[s]){
-	    CleanupTimer.Start();
-	    std::cout<<GridLogMessage<<"ConjugateGradientMultiShiftMixedPrec: performing cleanup step for shift " << s << std::endl;
-
-	    //Setup linear operators for final cleanup
-	    ConjugateGradientMultiShiftMixedPrecSupport::ShiftedLinop<FieldD> Linop_shift_d(Linop_d, mass[s]);
-	    ConjugateGradientMultiShiftMixedPrecSupport::ShiftedLinop<FieldF> Linop_shift_f(Linop_f, mass[s]);
-					       
-	    MixedPrecisionConjugateGradient<FieldD,FieldF> cg(mresidual[s], MaxIterations, MaxIterations, SinglePrecGrid, Linop_shift_f, Linop_shift_d); 
-	    cg(src_d, psi_d[s]);
-	    
-	    TrueResidualShift[s] = cg.TrueResidual;
-	    CleanupTimer.Stop();
-	  }
-	}
-
-	std::cout << GridLogMessage << "ConjugateGradientMultiShiftMixedPrec: Time Breakdown for body"<<std::endl;
-	std::cout << GridLogMessage << "\tSolver    " << SolverTimer.Elapsed()     <<std::endl;
-	std::cout << GridLogMessage << "\t\tAXPY    " << AXPYTimer.Elapsed()     <<std::endl;
-	std::cout << GridLogMessage << "\t\tMatrix    " << MatrixTimer.Elapsed()     <<std::endl;
-	std::cout << GridLogMessage << "\t\tShift    " << ShiftTimer.Elapsed()     <<std::endl;
-	std::cout << GridLogMessage << "\t\tPrecision Change " << PrecChangeTimer.Elapsed()     <<std::endl;
-	std::cout << GridLogMessage << "\tFinal Cleanup " << CleanupTimer.Elapsed()     <<std::endl;
-	std::cout << GridLogMessage << "\tSolver+Cleanup " << SolverTimer.Elapsed() + CleanupTimer.Elapsed() << std::endl;
-
-	IterationsToComplete = k;	
-
-	return;
-      }
-
-   
-    }
-    // ugly hack
-    std::cout<<GridLogMessage<<"CG multi shift did not converge"<<std::endl;
-    //  assert(0);
-  }
-
-};
-NAMESPACE_END(Grid);
-#endif
@@ -73,7 +73,6 @@ public:
  }
    
  void operator()(const FieldD &src, FieldD &psi) {
-    GRID_TRACE("ConjugateGradientReliableUpdate");
    LinearOperatorBase<FieldF> *Linop_f_use = &Linop_f;
    bool using_fallback = false;
      
@@ -33,19 +33,16 @@ namespace Grid {
 template<class Field>
 class ZeroGuesser: public LinearFunction<Field> {
 public:
-  using LinearFunction<Field>::operator();
    virtual void operator()(const Field &src, Field &guess) { guess = Zero(); };
 };
 template<class Field>
 class DoNothingGuesser: public LinearFunction<Field> {
 public:
-  using LinearFunction<Field>::operator();
  virtual void operator()(const Field &src, Field &guess) {  };
 };
 template<class Field>
 class SourceGuesser: public LinearFunction<Field> {
 public:
-  using LinearFunction<Field>::operator();
  virtual void operator()(const Field &src, Field &guess) { guess = src; };
 };

@@ -57,24 +54,15 @@ class DeflatedGuesser: public LinearFunction<Field> {
 private:
  const std::vector<Field> &evec;
  const std::vector<RealD> &eval;
-  const unsigned int       N;

 public:
-  using LinearFunction<Field>::operator();

-  DeflatedGuesser(const std::vector<Field> & _evec,const std::vector<RealD> & _eval)
-  : DeflatedGuesser(_evec, _eval, _evec.size())
-  {}
-
-  DeflatedGuesser(const std::vector<Field> & _evec, const std::vector<RealD> & _eval, const unsigned int _N)
-  : evec(_evec), eval(_eval), N(_N)
-  {
-    assert(evec.size()==eval.size());
-    assert(N <= evec.size());
-  } 
+  DeflatedGuesser(const std::vector<Field> & _evec,const std::vector<RealD> & _eval) : evec(_evec), eval(_eval) {};

  virtual void operator()(const Field &src,Field &guess) {
    guess = Zero();
+    assert(evec.size()==eval.size());
+    auto N = evec.size();
    for (int i=0;i<N;i++) {
      const Field& tmp = evec[i];
      axpy(guess,TensorRemove(innerProduct(tmp,src)) / eval[i],tmp,guess);
@@ -91,7 +79,6 @@ private:
  const std::vector<RealD>       &eval_coarse;
 public:
  
-  using LinearFunction<FineField>::operator();
  LocalCoherenceDeflatedGuesser(const std::vector<FineField>   &_subspace,
 				const std::vector<CoarseField> &_evec_coarse,
 				const std::vector<RealD>       &_eval_coarse)
@@ -113,43 +100,7 @@ public:
    blockPromote(guess_coarse,guess,subspace);
    guess.Checkerboard() = src.Checkerboard();
  };
-
-  void operator()(const std::vector<FineField> &src,std::vector<FineField> &guess) {
-    int Nevec = (int)evec_coarse.size();
-    int Nsrc = (int)src.size();
-    // make temp variables
-    std::vector<CoarseField> src_coarse(Nsrc,evec_coarse[0].Grid());
-    std::vector<CoarseField> guess_coarse(Nsrc,evec_coarse[0].Grid());    
-    //Preporcessing
-    std::cout << GridLogMessage << "Start BlockProject for loop" << std::endl;
-    for (int j=0;j<Nsrc;j++)
-    {
-    guess_coarse[j] = Zero();
-    std::cout << GridLogMessage << "BlockProject iter: " << j << std::endl;
-    blockProject(src_coarse[j],src[j],subspace);
-    }
-    //deflation set up for eigen vector batchsize 1 and source batch size equal number of sources
-    std::cout << GridLogMessage << "Start ProjectAccum for loop" << std::endl;
-    for (int i=0;i<Nevec;i++)
-    {
-      std::cout << GridLogMessage << "ProjectAccum Nvec: " << i << std::endl;
-      const CoarseField & tmp = evec_coarse[i];
-      for (int j=0;j<Nsrc;j++)
-      {
-        axpy(guess_coarse[j],TensorRemove(innerProduct(tmp,src_coarse[j])) / eval_coarse[i],tmp,guess_coarse[j]);
-      }
-    }
-    //postprocessing
-    std::cout << GridLogMessage << "Start BlockPromote for loop" << std::endl;
-    for (int j=0;j<Nsrc;j++)
-    {
-    std::cout << GridLogMessage << "BlockProject iter: " << j << std::endl;
-    blockPromote(guess_coarse[j],guess[j],subspace);
-    guess[j].Checkerboard() = src[j].Checkerboard();
-    }
-  };
-
-  };
+};



@@ -44,7 +44,6 @@ public:
 				  int, MinRes);    // Must restart
 };

-//This class is the input parameter class for some testing programs
 struct LocalCoherenceLanczosParams : Serializable {
 public:
  GRID_SERIALIZABLE_CLASS_MEMBERS(LocalCoherenceLanczosParams,
@@ -68,7 +67,6 @@ public:
 template<class Fobj,class CComplex,int nbasis>
 class ProjectedHermOp : public LinearFunction<Lattice<iVector<CComplex,nbasis > > > {
 public:
-  using LinearFunction<Lattice<iVector<CComplex,nbasis > > >::operator();
  typedef iVector<CComplex,nbasis >           CoarseSiteVector;
  typedef Lattice<CoarseSiteVector>           CoarseField;
  typedef Lattice<CComplex>   CoarseScalar; // used for inner products on fine field
@@ -99,7 +97,6 @@ public:
 template<class Fobj,class CComplex,int nbasis>
 class ProjectedFunctionHermOp : public LinearFunction<Lattice<iVector<CComplex,nbasis > > > {
 public:
-  using LinearFunction<Lattice<iVector<CComplex,nbasis > > >::operator();
  typedef iVector<CComplex,nbasis >           CoarseSiteVector;
  typedef Lattice<CoarseSiteVector>           CoarseField;
  typedef Lattice<CComplex>   CoarseScalar; // used for inner products on fine field
@@ -146,24 +143,16 @@ public:
  LinearOperatorBase<FineField> &_Linop;
  RealD                             _coarse_relax_tol;
  std::vector<FineField>        &_subspace;
-
-  int _largestEvalIdxForReport; //The convergence of the LCL is based on the evals of the coarse grid operator, not those of the underlying fine grid operator
-                                //As a result we do not know what the eval range of the fine operator is until the very end, making tuning the Cheby bounds very difficult
-                                //To work around this issue, every restart we separately reconstruct the fine operator eval for the lowest and highest evec and print these
-                                //out alongside the evals of the coarse operator. To do so we need to know the index of the largest eval (i.e. Nstop-1)
-                                //NOTE: If largestEvalIdxForReport=-1 (default) then this is not performed
  
  ImplicitlyRestartedLanczosSmoothedTester(LinearFunction<CoarseField>   &Poly,
 					   OperatorFunction<FineField>   &smoother,
 					   LinearOperatorBase<FineField> &Linop,
 					   std::vector<FineField>        &subspace,
-					   RealD coarse_relax_tol=5.0e3,
-					   int largestEvalIdxForReport=-1) 
+					   RealD coarse_relax_tol=5.0e3) 
    : _smoother(smoother), _Linop(Linop), _Poly(Poly), _subspace(subspace),
-      _coarse_relax_tol(coarse_relax_tol), _largestEvalIdxForReport(largestEvalIdxForReport)
+      _coarse_relax_tol(coarse_relax_tol)  
  {    };

-  //evalMaxApprox: approximation of largest eval of the fine Chebyshev operator (suitably wrapped by block projection)
  int TestConvergence(int j,RealD eresid,CoarseField &B, RealD &eval,RealD evalMaxApprox)
  {
    CoarseField v(B);
@@ -186,26 +175,12 @@ public:
 	     <<" |H B[i] - eval[i]B[i]|^2 / evalMaxApprox^2 " << std::setw(25) << vv
 	     <<std::endl;

-    if(_largestEvalIdxForReport != -1 && (j==0 || j==_largestEvalIdxForReport)){
-      std::cout<<GridLogIRL << "Estimating true eval of fine grid operator for eval idx " << j << std::endl;
-      RealD tmp_eval;
-      ReconstructEval(j,eresid,B,tmp_eval,1.0); //don't use evalMaxApprox of coarse operator! (cf below)
-    }
-    
    int conv=0;
    if( (vv<eresid*eresid) ) conv = 1;
    return conv;
  }
-
-  //This function is called at the end of the coarse grid Lanczos. It promotes the coarse eigenvector 'B' to the fine grid,
-  //applies a smoother to the result then computes the computes the *fine grid* eigenvalue (output as 'eval').
-
-  //evalMaxApprox should be the approximation of the largest eval of the fine Hermop. However when this function is called by IRL it actually passes the largest eval of the *Chebyshev* operator (as this is the max approx used for the TestConvergence above)
-  //As the largest eval of the Chebyshev is typically several orders of magnitude larger this makes the convergence test pass even when it should not.
-  //We therefore ignore evalMaxApprox here and use a value of 1.0 (note this value is already used by TestCoarse)
-  int ReconstructEval(int j,RealD eresid,CoarseField &B, RealD &eval,RealD evalMaxApprox)  
+  int ReconstructEval(int j,RealD eresid,CoarseField &B, RealD &eval,RealD evalMaxApprox)
  {
-    evalMaxApprox = 1.0; //cf above
    GridBase *FineGrid = _subspace[0].Grid();    
    int checkerboard   = _subspace[0].Checkerboard();
    FineField fB(FineGrid);fB.Checkerboard() =checkerboard;
@@ -224,13 +199,13 @@ public:
    eval   = vnum/vden;
    fv -= eval*fB;
    RealD vv = norm2(fv) / ::pow(evalMaxApprox,2.0);
-    if ( j > nbasis ) eresid = eresid*_coarse_relax_tol;
-    
+
    std::cout.precision(13);
    std::cout<<GridLogIRL  << "[" << std::setw(3)<<j<<"] "
 	     <<"eval = "<<std::setw(25)<< eval << " (" << eval_poly << ")"
-	     <<" |H B[i] - eval[i]B[i]|^2 / evalMaxApprox^2 " << std::setw(25) << vv << " target " << eresid*eresid
+	     <<" |H B[i] - eval[i]B[i]|^2 / evalMaxApprox^2 " << std::setw(25) << vv
 	     <<std::endl;
+    if ( j > nbasis ) eresid = eresid*_coarse_relax_tol;
    if( (vv<eresid*eresid) ) return 1;
    return 0;
  }
@@ -308,10 +283,6 @@ public:
    evals_coarse.resize(0);
  };

-  //The block inner product is the inner product on the fine grid locally summed over the blocks
-  //to give a Lattice<Scalar> on the coarse grid. This function orthnormalizes the fine-grid subspace
-  //vectors under the block inner product. This step must be performed after computing the fine grid
-  //eigenvectors and before computing the coarse grid eigenvectors.    
  void Orthogonalise(void ) {
    CoarseScalar InnerProd(_CoarseGrid);
    std::cout << GridLogMessage <<" Gramm-Schmidt pass 1"<<std::endl;
@@ -355,8 +326,6 @@ public:
    }
  }

-  //While this method serves to check the coarse eigenvectors, it also recomputes the eigenvalues from the smoothed reconstructed eigenvectors
-  //hence the smoother can be tuned after running the coarse Lanczos by using a different smoother here
  void testCoarse(RealD resid,ChebyParams cheby_smooth,RealD relax) 
  {
    assert(evals_fine.size() == nbasis);
@@ -405,31 +374,25 @@ public:
    evals_fine.resize(nbasis);
    subspace.resize(nbasis,_FineGrid);
  }
-
-
-  //cheby_op: Parameters of the fine grid Chebyshev polynomial used for the Lanczos acceleration
-  //cheby_smooth: Parameters of a separate Chebyshev polynomial used after the Lanczos has completed to smooth out high frequency noise in the reconstructed fine grid eigenvectors prior to computing the eigenvalue
-  //relax: Reconstructed eigenvectors (post smoothing) are naturally not as precise as true eigenvectors. This factor acts as a multiplier on the stopping condition when determining whether the results satisfy the user provided stopping condition
  void calcCoarse(ChebyParams cheby_op,ChebyParams cheby_smooth,RealD relax,
 		  int Nstop, int Nk, int Nm,RealD resid, 
 		  RealD MaxIt, RealD betastp, int MinRes)
  {
-    Chebyshev<FineField>                          Cheby(cheby_op); //Chebyshev of fine operator on fine grid
-    ProjectedHermOp<Fobj,CComplex,nbasis>         Op(_FineOp,subspace); //Fine operator on coarse grid with intermediate fine grid conversion
-    ProjectedFunctionHermOp<Fobj,CComplex,nbasis> ChebyOp (Cheby,_FineOp,subspace); //Chebyshev of fine operator on coarse grid with intermediate fine grid conversion
+    Chebyshev<FineField>                          Cheby(cheby_op);
+    ProjectedHermOp<Fobj,CComplex,nbasis>         Op(_FineOp,subspace);
+    ProjectedFunctionHermOp<Fobj,CComplex,nbasis> ChebyOp (Cheby,_FineOp,subspace);
    //////////////////////////////////////////////////////////////////////////////////////////////////
    // create a smoother and see if we can get a cheap convergence test and smooth inside the IRL
    //////////////////////////////////////////////////////////////////////////////////////////////////

-    Chebyshev<FineField>                                           ChebySmooth(cheby_smooth); //lower order Chebyshev of fine operator on fine grid used to smooth regenerated eigenvectors
-    ImplicitlyRestartedLanczosSmoothedTester<Fobj,CComplex,nbasis> ChebySmoothTester(ChebyOp,ChebySmooth,_FineOp,subspace,relax,Nstop-1); 
+    Chebyshev<FineField>                                           ChebySmooth(cheby_smooth);
+    ImplicitlyRestartedLanczosSmoothedTester<Fobj,CComplex,nbasis> ChebySmoothTester(ChebyOp,ChebySmooth,_FineOp,subspace,relax);

    evals_coarse.resize(Nm);
    evec_coarse.resize(Nm,_CoarseGrid);

    CoarseField src(_CoarseGrid);     src=1.0; 

-    //Note the "tester" here is also responsible for generating the fine grid eigenvalues which are output into the "evals_coarse" array
    ImplicitlyRestartedLanczos<CoarseField> IRL(ChebyOp,ChebyOp,ChebySmoothTester,Nstop,Nk,Nm,resid,MaxIt,betastp,MinRes);
    int Nconv=0;
    IRL.calc(evals_coarse,evec_coarse,src,Nconv,false);
@@ -440,14 +403,6 @@ public:
      std::cout << i << " Coarse eval = " << evals_coarse[i]  << std::endl;
    }
  }
-
-  //Get the fine eigenvector 'i' by reconstruction
-  void getFineEvecEval(FineField &evec, RealD &eval, const int i) const{
-    blockPromote(evec_coarse[i],evec,subspace);  
-    eval = evals_coarse[i];
-  }
-    
-    
 };

 NAMESPACE_END(Grid);
@@ -29,8 +29,6 @@ template<class Field> class PowerMethod
      RealD vnum = real(innerProduct(src_n,tmp)); // HermOp. 
      RealD vden = norm2(src_n); 
      RealD na = vnum/vden; 
-
-      std::cout << GridLogIterative << "PowerMethod: Current approximation of largest eigenvalue " << na << std::endl;
      
      if ( (fabs(evalMaxApprox/na - 1.0) < 0.001) || (i==_MAX_ITER_EST_-1) ) { 
 	evalMaxApprox = na; 
@@ -43,7 +43,7 @@ NAMESPACE_BEGIN(Grid);
 template<class Field>
 class PrecGeneralisedConjugateResidual : public LinearFunction<Field> {
 public:                                                
-  using LinearFunction<Field>::operator();
+
  RealD   Tolerance;
  Integer MaxIterations;
  int verbose;
@@ -43,7 +43,7 @@ NAMESPACE_BEGIN(Grid);
 template<class Field>
 class PrecGeneralisedConjugateResidualNonHermitian : public LinearFunction<Field> {
 public:                                                
-  using LinearFunction<Field>::operator();
+
  RealD   Tolerance;
  Integer MaxIterations;
  int verbose;
@@ -119,8 +119,7 @@ public:
  RealD GCRnStep(const Field &src, Field &psi,RealD rsq){

    RealD cp;
-    ComplexD a, b;
-    //    ComplexD zAz;
+    ComplexD a, b, zAz;
    RealD zAAz;
    ComplexD rq;

@@ -147,7 +146,7 @@ public:
    //////////////////////////////////
    MatTimer.Start();
    Linop.Op(psi,Az);
-    //    zAz = innerProduct(Az,psi);
+    zAz = innerProduct(Az,psi);
    zAAz= norm2(Az);
    MatTimer.Stop();
    
@@ -171,7 +170,7 @@ public:

    LinalgTimer.Start();

-    //    zAz = innerProduct(Az,psi);
+    zAz = innerProduct(Az,psi);
    zAAz= norm2(Az);

    //p[0],q[0],qq[0] 
@@ -213,7 +212,7 @@ public:
      MatTimer.Start();
      Linop.Op(z,Az);
      MatTimer.Stop();
-      //      zAz = innerProduct(Az,psi);
+      zAz = innerProduct(Az,psi);
      zAAz= norm2(Az);

      LinalgTimer.Start();
@@ -132,31 +132,6 @@ namespace Grid {
      (*this)(_Matrix,in,out,guess);
    }

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

-      
-      if(useSolnAsInitGuess) {
-        for(int b=0;b<nblock;b++){
+      for(int b=0;b<nblock;b++){
+        if(useSolnAsInitGuess) {
          pickCheckerboard(Odd, sol_o[b], out[b]);
+        } else {
+          guess(src_o[b],sol_o[b]); 
        }
-      } else {
-        guess(src_o, sol_o); 
-      }

-	    if ( subGuess ) { 
-        for(int b=0;b<nblock;b++){
-          guess_save[b] = sol_o[b];
-        }
+	if ( subGuess ) { 
+	  guess_save[b] = sol_o[b];
+	}
      }
      //////////////////////////////////////////////////////////////
      // Call the block solver
@@ -9,30 +9,14 @@ NAMESPACE_BEGIN(Grid);
 #define AccSmall (3)
 #define Shared   (4)
 #define SharedSmall (5)
-#undef GRID_MM_VERBOSE 
 uint64_t total_shared;
 uint64_t total_device;
 uint64_t total_host;;
 void MemoryManager::PrintBytes(void)
 {
-  std::cout << " MemoryManager : ------------------------------------ "<<std::endl;
-  std::cout << " MemoryManager : PrintBytes "<<std::endl;
-  std::cout << " MemoryManager : ------------------------------------ "<<std::endl;
-  std::cout << " MemoryManager : "<<(total_shared>>20)<<" shared      Mbytes "<<std::endl;
-  std::cout << " MemoryManager : "<<(total_device>>20)<<" accelerator Mbytes "<<std::endl;
-  std::cout << " MemoryManager : "<<(total_host>>20)  <<" cpu         Mbytes "<<std::endl;
-  uint64_t cacheBytes;
-  cacheBytes = CacheBytes[Cpu];
-  std::cout << " MemoryManager : "<<(cacheBytes>>20) <<" cpu cache Mbytes "<<std::endl;
-  cacheBytes = CacheBytes[Acc];
-  std::cout << " MemoryManager : "<<(cacheBytes>>20) <<" acc cache Mbytes "<<std::endl;
-  cacheBytes = CacheBytes[Shared];
-  std::cout << " MemoryManager : "<<(cacheBytes>>20) <<" shared cache Mbytes "<<std::endl;
-  
-#ifdef GRID_CUDA
-  cuda_mem();
-#endif
-  
+  std::cout << " MemoryManager : "<<total_shared<<" shared      bytes "<<std::endl;
+  std::cout << " MemoryManager : "<<total_device<<" accelerator bytes "<<std::endl;
+  std::cout << " MemoryManager : "<<total_host  <<" cpu         bytes "<<std::endl;
 }

 //////////////////////////////////////////////////////////////////////
@@ -40,114 +24,88 @@ void MemoryManager::PrintBytes(void)
 //////////////////////////////////////////////////////////////////////
 MemoryManager::AllocationCacheEntry MemoryManager::Entries[MemoryManager::NallocType][MemoryManager::NallocCacheMax];
 int MemoryManager::Victim[MemoryManager::NallocType];
-int MemoryManager::Ncache[MemoryManager::NallocType] = { 2, 8, 8, 16, 8, 16 };
-uint64_t MemoryManager::CacheBytes[MemoryManager::NallocType];
+int MemoryManager::Ncache[MemoryManager::NallocType] = { 8, 32, 8, 32, 8, 32 };
+
 //////////////////////////////////////////////////////////////////////
 // Actual allocation and deallocation utils
 //////////////////////////////////////////////////////////////////////
 void *MemoryManager::AcceleratorAllocate(size_t bytes)
 {
-  total_device+=bytes;
  void *ptr = (void *) Lookup(bytes,Acc);
  if ( ptr == (void *) NULL ) {
    ptr = (void *) acceleratorAllocDevice(bytes);
+    total_device+=bytes;
  }
-#ifdef GRID_MM_VERBOSE
-  std::cout <<"AcceleratorAllocate "<<std::endl;
-  PrintBytes();
-#endif
  return ptr;
 }
 void  MemoryManager::AcceleratorFree    (void *ptr,size_t bytes)
 {
-  total_device-=bytes;
  void *__freeme = Insert(ptr,bytes,Acc);
  if ( __freeme ) {
    acceleratorFreeDevice(__freeme);
+    total_device-=bytes;
+//       PrintBytes();
  }
-#ifdef GRID_MM_VERBOSE
-  std::cout <<"AcceleratorFree "<<std::endl;
-  PrintBytes();
-#endif
 }
 void *MemoryManager::SharedAllocate(size_t bytes)
 {
-  total_shared+=bytes;
  void *ptr = (void *) Lookup(bytes,Shared);
  if ( ptr == (void *) NULL ) {
    ptr = (void *) acceleratorAllocShared(bytes);
+    total_shared+=bytes;
+        std::cout <<GridLogMessage<<"AcceleratorAllocate: allocated Shared pointer "<<std::hex<<ptr<<std::dec<<std::endl;
+//        PrintBytes();
  }
-#ifdef GRID_MM_VERBOSE
-  std::cout <<"SharedAllocate "<<std::endl;
-  PrintBytes();
-#endif
  return ptr;
 }
 void  MemoryManager::SharedFree    (void *ptr,size_t bytes)
 {
-  total_shared-=bytes;
  void *__freeme = Insert(ptr,bytes,Shared);
  if ( __freeme ) {
    acceleratorFreeShared(__freeme);
+    total_shared-=bytes;
+    //    PrintBytes();
  }
-#ifdef GRID_MM_VERBOSE
-  std::cout <<"SharedFree "<<std::endl;
-  PrintBytes();
-#endif
 }
 #ifdef GRID_UVM
 void *MemoryManager::CpuAllocate(size_t bytes)
 {
-  total_host+=bytes;
  void *ptr = (void *) Lookup(bytes,Cpu);
  if ( ptr == (void *) NULL ) {
    ptr = (void *) acceleratorAllocShared(bytes);
+    total_host+=bytes;
+//    std::cout << GridLogMessage<< "MemoryManager:: CpuAllocate  total_host= "<<total_host<<" "<< ptr << std::endl;
  }
-#ifdef GRID_MM_VERBOSE
-  std::cout <<"CpuAllocate "<<std::endl;
-  PrintBytes();
-#endif
  return ptr;
 }
 void  MemoryManager::CpuFree    (void *_ptr,size_t bytes)
 {
-  total_host-=bytes;
  NotifyDeletion(_ptr);
  void *__freeme = Insert(_ptr,bytes,Cpu);
  if ( __freeme ) { 
    acceleratorFreeShared(__freeme);
+//    std::cout << GridLogMessage<< "MemoryManager:: CpuFree  total_host= "<<total_host<<" "<< __freeme << std::endl;
+    total_host-=bytes;
  }
-#ifdef GRID_MM_VERBOSE
-  std::cout <<"CpuFree "<<std::endl;
-  PrintBytes();
-#endif
 }
 #else
 void *MemoryManager::CpuAllocate(size_t bytes)
 {
-  total_host+=bytes;
  void *ptr = (void *) Lookup(bytes,Cpu);
  if ( ptr == (void *) NULL ) {
    ptr = (void *) acceleratorAllocCpu(bytes);
+    total_host+=bytes;
  }
-#ifdef GRID_MM_VERBOSE
-  std::cout <<"CpuAllocate "<<std::endl;
-  PrintBytes();
-#endif
  return ptr;
 }
 void  MemoryManager::CpuFree    (void *_ptr,size_t bytes)
 {
-  total_host-=bytes;
  NotifyDeletion(_ptr);
  void *__freeme = Insert(_ptr,bytes,Cpu);
  if ( __freeme ) { 
    acceleratorFreeCpu(__freeme);
+    total_host-=bytes;
  }
-#ifdef GRID_MM_VERBOSE
-  std::cout <<"CpuFree "<<std::endl;
-  PrintBytes();
-#endif
 }
 #endif

@@ -159,6 +117,7 @@ void MemoryManager::Init(void)

  char * str;
  int Nc;
+  int NcS;
  
  str= getenv("GRID_ALLOC_NCACHE_LARGE");
  if ( str ) {
@@ -224,13 +183,13 @@ void *MemoryManager::Insert(void *ptr,size_t bytes,int type)
 #ifdef ALLOCATION_CACHE
  bool small = (bytes < GRID_ALLOC_SMALL_LIMIT);
  int cache = type + small;
-  return Insert(ptr,bytes,Entries[cache],Ncache[cache],Victim[cache],CacheBytes[cache]);  
+  return Insert(ptr,bytes,Entries[cache],Ncache[cache],Victim[cache]);  
 #else
  return ptr;
 #endif
 }

-void *MemoryManager::Insert(void *ptr,size_t bytes,AllocationCacheEntry *entries,int ncache,int &victim, uint64_t &cacheBytes) 
+void *MemoryManager::Insert(void *ptr,size_t bytes,AllocationCacheEntry *entries,int ncache,int &victim) 
 {
  assert(ncache>0);
 #ifdef GRID_OMP
@@ -254,7 +213,6 @@ void *MemoryManager::Insert(void *ptr,size_t bytes,AllocationCacheEntry *entries

  if ( entries[v].valid ) {
    ret = entries[v].address;
-    cacheBytes -= entries[v].bytes;
    entries[v].valid = 0;
    entries[v].address = NULL;
    entries[v].bytes = 0;
@@ -263,7 +221,6 @@ void *MemoryManager::Insert(void *ptr,size_t bytes,AllocationCacheEntry *entries
  entries[v].address=ptr;
  entries[v].bytes  =bytes;
  entries[v].valid  =1;
-  cacheBytes += bytes;

  return ret;
 }
@@ -273,13 +230,13 @@ void *MemoryManager::Lookup(size_t bytes,int type)
 #ifdef ALLOCATION_CACHE
  bool small = (bytes < GRID_ALLOC_SMALL_LIMIT);
  int cache = type+small;
-  return Lookup(bytes,Entries[cache],Ncache[cache],CacheBytes[cache]);
+  return Lookup(bytes,Entries[cache],Ncache[cache]);
 #else
  return NULL;
 #endif
 }

-void *MemoryManager::Lookup(size_t bytes,AllocationCacheEntry *entries,int ncache,uint64_t & cacheBytes) 
+void *MemoryManager::Lookup(size_t bytes,AllocationCacheEntry *entries,int ncache) 
 {
  assert(ncache>0);
 #ifdef GRID_OMP
@@ -288,7 +245,6 @@ void *MemoryManager::Lookup(size_t bytes,AllocationCacheEntry *entries,int ncach
  for(int e=0;e<ncache;e++){
    if ( entries[e].valid && ( entries[e].bytes == bytes ) ) {
      entries[e].valid = 0;
-      cacheBytes -= entries[e].bytes;
      return entries[e].address;
    }
  }
@@ -36,11 +36,6 @@ NAMESPACE_BEGIN(Grid);

 #define GRID_ALLOC_SMALL_LIMIT (4096)

-#define STRINGIFY(x) #x
-#define TOSTRING(x) STRINGIFY(x)
-#define FILE_LINE __FILE__ ":" TOSTRING(__LINE__)
-#define AUDIT(a) MemoryManager::Audit(FILE_LINE)
-
 /*Pinning pages is costly*/
 ////////////////////////////////////////////////////////////////////////////
 // Advise the LatticeAccelerator class
@@ -87,19 +82,17 @@ private:
  static AllocationCacheEntry Entries[NallocType][NallocCacheMax];
  static int Victim[NallocType];
  static int Ncache[NallocType];
-  static uint64_t CacheBytes[NallocType];

  /////////////////////////////////////////////////
  // Free pool
  /////////////////////////////////////////////////
  static void *Insert(void *ptr,size_t bytes,int type) ;
  static void *Lookup(size_t bytes,int type) ;
-  static void *Insert(void *ptr,size_t bytes,AllocationCacheEntry *entries,int ncache,int &victim,uint64_t &cbytes) ;
-  static void *Lookup(size_t bytes,AllocationCacheEntry *entries,int ncache,uint64_t &cbytes) ;
+  static void *Insert(void *ptr,size_t bytes,AllocationCacheEntry *entries,int ncache,int &victim) ;
+  static void *Lookup(size_t bytes,AllocationCacheEntry *entries,int ncache) ;

  static void PrintBytes(void);
 public:
-  static void Audit(std::string s);
  static void Init(void);
  static void InitMessage(void);
  static void *AcceleratorAllocate(size_t bytes);
@@ -176,7 +169,6 @@ private:

 public:
  static void Print(void);
-  static void PrintState( void* CpuPtr);
  static int   isOpen   (void* CpuPtr);
  static void  ViewClose(void* CpuPtr,ViewMode mode);
  static void *ViewOpen (void* CpuPtr,size_t bytes,ViewMode mode,ViewAdvise hint);
@@ -3,13 +3,8 @@

 #warning "Using explicit device memory copies"
 NAMESPACE_BEGIN(Grid);
-
-#define MAXLINE 512
-static char print_buffer [ MAXLINE ];
-
-#define mprintf(...) snprintf (print_buffer,MAXLINE, __VA_ARGS__ ); std::cout << GridLogMemory << print_buffer;
-#define dprintf(...) snprintf (print_buffer,MAXLINE, __VA_ARGS__ ); std::cout << GridLogMemory << print_buffer;
-//#define dprintf(...) 
+//define dprintf(...) printf ( __VA_ARGS__ ); fflush(stdout);
+#define dprintf(...)


 ////////////////////////////////////////////////////////////
@@ -109,7 +104,7 @@ void MemoryManager::AccDiscard(AcceleratorViewEntry &AccCache)
  ///////////////////////////////////////////////////////////
  assert(AccCache.state!=Empty);
  
-  mprintf("MemoryManager: Discard(%lx) %lx\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);
@@ -117,7 +112,7 @@ void MemoryManager::AccDiscard(AcceleratorViewEntry &AccCache)
    AcceleratorFree((void *)AccCache.AccPtr,AccCache.bytes);
    DeviceBytes   -=AccCache.bytes;
    LRUremove(AccCache);
-    dprintf("MemoryManager: Free(%lx) LRU %ld Total %ld\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);
@@ -131,11 +126,9 @@ void MemoryManager::Evict(AcceleratorViewEntry &AccCache)
  ///////////////////////////////////////////////////////////////////////////
  assert(AccCache.state!=Empty);
  
-  mprintf("MemoryManager: Evict cpu %lx acc %lx cpuLock %ld accLock %ld\n",
-	  (uint64_t)AccCache.CpuPtr,(uint64_t)AccCache.AccPtr,
-	  (uint64_t)AccCache.cpuLock,(uint64_t)AccCache.accLock); 
-  if (AccCache.accLock!=0) return;
-  if (AccCache.cpuLock!=0) return;
+  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) {
    Flush(AccCache);
  }
@@ -144,7 +137,7 @@ void MemoryManager::Evict(AcceleratorViewEntry &AccCache)
    AcceleratorFree((void *)AccCache.AccPtr,AccCache.bytes);
    DeviceBytes   -=AccCache.bytes;
    LRUremove(AccCache);
-    dprintf("MemoryManager: Free(%lx) footprint now %ld \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);
@@ -157,7 +150,7 @@ 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);
-  mprintf("MemoryManager: Flush  %lx -> %lx\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;
@@ -172,7 +165,7 @@ void MemoryManager::Clone(AcceleratorViewEntry &AccCache)
    AccCache.AccPtr=(uint64_t)AcceleratorAllocate(AccCache.bytes);
    DeviceBytes+=AccCache.bytes;
  }
-  mprintf("MemoryManager: Clone %lx <- %lx\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++;
@@ -198,7 +191,6 @@ void MemoryManager::CpuDiscard(AcceleratorViewEntry &AccCache)
 void MemoryManager::ViewClose(void* Ptr,ViewMode mode)
 {
  if( (mode==AcceleratorRead)||(mode==AcceleratorWrite)||(mode==AcceleratorWriteDiscard) ){
-    dprintf("AcceleratorViewClose %lx\n",(uint64_t)Ptr);
    AcceleratorViewClose((uint64_t)Ptr);
  } else if( (mode==CpuRead)||(mode==CpuWrite)){
    CpuViewClose((uint64_t)Ptr);
@@ -210,7 +202,6 @@ void *MemoryManager::ViewOpen(void* _CpuPtr,size_t bytes,ViewMode mode,ViewAdvis
 {
  uint64_t CpuPtr = (uint64_t)_CpuPtr;
  if( (mode==AcceleratorRead)||(mode==AcceleratorWrite)||(mode==AcceleratorWriteDiscard) ){
-    dprintf("AcceleratorViewOpen %lx\n",(uint64_t)CpuPtr);
    return (void *) AcceleratorViewOpen(CpuPtr,bytes,mode,hint);
  } else if( (mode==CpuRead)||(mode==CpuWrite)){
    return (void *)CpuViewOpen(CpuPtr,bytes,mode,hint);
@@ -250,12 +241,11 @@ uint64_t MemoryManager::AcceleratorViewOpen(uint64_t CpuPtr,size_t bytes,ViewMod
  assert(AccCache.cpuLock==0);  // Programming error

  if(AccCache.state!=Empty) {
-    dprintf("ViewOpen found entry %lx %lx : %ld %ld accLock %ld\n",
+    dprintf("ViewOpen found entry %llx %llx : %lld %lld\n",
 		    (uint64_t)AccCache.CpuPtr,
 		    (uint64_t)CpuPtr,
 		    (uint64_t)AccCache.bytes,
-	            (uint64_t)bytes,
-		    (uint64_t)AccCache.accLock);
+		    (uint64_t)bytes);
    assert(AccCache.CpuPtr == CpuPtr);
    assert(AccCache.bytes  ==bytes);
  }
@@ -290,7 +280,6 @@ uint64_t MemoryManager::AcceleratorViewOpen(uint64_t CpuPtr,size_t bytes,ViewMod
      AccCache.state  = Consistent; // Empty + AccRead => Consistent
    }
    AccCache.accLock= 1;
-    dprintf("Copied Empty entry into device accLock= %d\n",AccCache.accLock);
  } else if(AccCache.state==CpuDirty ){
    if(mode==AcceleratorWriteDiscard) {
      CpuDiscard(AccCache);
@@ -303,21 +292,21 @@ uint64_t MemoryManager::AcceleratorViewOpen(uint64_t CpuPtr,size_t bytes,ViewMod
      AccCache.state  = Consistent; // CpuDirty + AccRead => Consistent
    }
    AccCache.accLock++;
-    dprintf("CpuDirty entry into device ++accLock= %d\n",AccCache.accLock);
+    dprintf("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);
+    dprintf("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 ++accLock= %d\n",AccCache.accLock);
+    dprintf("AccDirty entry into device accLock %d\n",AccCache.accLock);
  } else {
    assert(0);
  }
@@ -325,7 +314,6 @@ uint64_t MemoryManager::AcceleratorViewOpen(uint64_t CpuPtr,size_t bytes,ViewMod
  // If view is opened on device remove from LRU
  if(AccCache.LRU_valid==1){
    // must possibly remove from LRU as now locked on GPU
-    dprintf("AccCache entry removed from LRU \n");
    LRUremove(AccCache);
  }

@@ -346,12 +334,10 @@ void MemoryManager::AcceleratorViewClose(uint64_t CpuPtr)
  assert(AccCache.accLock>0);

  AccCache.accLock--;
+
  // Move to LRU queue if not locked and close on device
  if(AccCache.accLock==0) {
-    dprintf("AccleratorViewClose %lx AccLock decremented to %ld move to LRU queue\n",(uint64_t)CpuPtr,(uint64_t)AccCache.accLock);
    LRUinsert(AccCache);
-  } else {
-    dprintf("AccleratorViewClose %lx AccLock decremented to %ld\n",(uint64_t)CpuPtr,(uint64_t)AccCache.accLock);
  }
 }
 void MemoryManager::CpuViewClose(uint64_t CpuPtr)
@@ -443,7 +429,6 @@ void  MemoryManager::NotifyDeletion(void *_ptr)
 }
 void  MemoryManager::Print(void)
 {
-  PrintBytes();
  std::cout << GridLogDebug << "--------------------------------------------" << std::endl;
  std::cout << GridLogDebug << "Memory Manager                             " << std::endl;
  std::cout << GridLogDebug << "--------------------------------------------" << std::endl;
@@ -487,55 +472,6 @@ int   MemoryManager::isOpen   (void* _CpuPtr)
    return 0;
  }
 }
-void MemoryManager::Audit(std::string s)
-{
-  for(auto it=AccViewTable.begin();it!=AccViewTable.end();it++){
-    auto &AccCache = it->second;
-    
-    std::string str;
-    if ( AccCache.state==Empty    ) str = std::string("Empty");
-    if ( AccCache.state==CpuDirty ) str = std::string("CpuDirty");
-    if ( AccCache.state==AccDirty ) str = std::string("AccDirty");
-    if ( AccCache.state==Consistent)str = std::string("Consistent");
-
-    if ( AccCache.cpuLock || AccCache.accLock ) { 
-      std::cout << GridLogError << s<< "\n\t 0x"<<std::hex<<AccCache.CpuPtr<<std::dec
-		<< "\t0x"<<std::hex<<AccCache.AccPtr<<std::dec<<"\t" <<str
-		<< "\t cpuLock  " << AccCache.cpuLock
-		<< "\t accLock  " << AccCache.accLock
-		<< "\t LRUvalid " << AccCache.LRU_valid<<std::endl;
-    }
-
-    assert( AccCache.cpuLock== 0 ) ;
-    assert( AccCache.accLock== 0 ) ;
-  }
-}
-
-void MemoryManager::PrintState(void* _CpuPtr)
-{
-  uint64_t CpuPtr = (uint64_t)_CpuPtr;
-
-  if ( EntryPresent(CpuPtr) ){
-    auto AccCacheIterator = EntryLookup(CpuPtr);
-    auto & AccCache = AccCacheIterator->second;
-    std::string str;
-    if ( AccCache.state==Empty    ) str = std::string("Empty");
-    if ( AccCache.state==CpuDirty ) str = std::string("CpuDirty");
-    if ( AccCache.state==AccDirty ) str = std::string("AccDirty");
-    if ( AccCache.state==Consistent)str = std::string("Consistent");
-    if ( AccCache.state==EvictNext) str = std::string("EvictNext");
-
-    std::cout << GridLogMessage << "CpuAddr\t\tAccAddr\t\tState\t\tcpuLock\taccLock\tLRU_valid "<<std::endl;
-    std::cout << GridLogMessage << "0x"<<std::hex<<AccCache.CpuPtr<<std::dec
-    << "\t0x"<<std::hex<<AccCache.AccPtr<<std::dec<<"\t" <<str
-    << "\t" << AccCache.cpuLock
-    << "\t" << AccCache.accLock
-    << "\t" << AccCache.LRU_valid<<std::endl;
-
-  } else {
-    std::cout << GridLogMessage << "No Entry in AccCache table." << std::endl; 
-  }
-}

 NAMESPACE_END(Grid);

@@ -13,14 +13,9 @@ uint64_t  MemoryManager::DeviceToHostBytes;
 uint64_t  MemoryManager::HostToDeviceXfer;
 uint64_t  MemoryManager::DeviceToHostXfer;

-void  MemoryManager::Audit(void){};
 void  MemoryManager::ViewClose(void* AccPtr,ViewMode mode){};
 void *MemoryManager::ViewOpen(void* CpuPtr,size_t bytes,ViewMode mode,ViewAdvise hint){ return CpuPtr; };
 int   MemoryManager::isOpen   (void* CpuPtr) { return 0;}
-void  MemoryManager::PrintState(void* CpuPtr)
-{
-std::cout << GridLogMessage << "Host<->Device memory movement not currently managed by Grid." << std::endl;
-};
 void  MemoryManager::Print(void){};
 void  MemoryManager::NotifyDeletion(void *ptr){};

@@ -33,8 +33,6 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>

 NAMESPACE_BEGIN(Grid);

-bool Stencil_force_mpi = true;
-
 ///////////////////////////////////////////////////////////////
 // Info that is setup once and indept of cartesian layout
 ///////////////////////////////////////////////////////////////
@@ -35,8 +35,6 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>

 NAMESPACE_BEGIN(Grid);

-extern bool Stencil_force_mpi ;
-
 class CartesianCommunicator : public SharedMemory {

 public:    
@@ -53,11 +51,10 @@ public:
  // Communicator should know nothing of the physics grid, only processor grid.
  ////////////////////////////////////////////
  int              _Nprocessors;     // How many in all
-  int              _processor;       // linear processor rank
-  unsigned long    _ndimension;
-  Coordinate _shm_processors;  // Which dimensions get relayed out over processors lanes.
  Coordinate _processors;      // Which dimensions get relayed out over processors lanes.
+  int              _processor;       // linear processor rank
  Coordinate _processor_coor;  // linear processor coordinate
+  unsigned long    _ndimension;
  static Grid_MPI_Comm      communicator_world;
  Grid_MPI_Comm             communicator;
  std::vector<Grid_MPI_Comm> communicator_halo;
@@ -98,16 +95,14 @@ public:
  int                      BossRank(void)          ;
  int                      ThisRank(void)          ;
  const Coordinate & ThisProcessorCoor(void) ;
-  const Coordinate & ShmGrid(void)  { return _shm_processors; }  ;
  const Coordinate & ProcessorGrid(void)     ;
-  int                ProcessorCount(void)    ;
+  int                      ProcessorCount(void)    ;

  ////////////////////////////////////////////////////////////////////////////////
  // very VERY rarely (Log, serial RNG) we need world without a grid
  ////////////////////////////////////////////////////////////////////////////////
  static int  RankWorld(void) ;
  static void BroadcastWorld(int root,void* data, int bytes);
-  static void BarrierWorld(void);
  
  ////////////////////////////////////////////////////////////
  // Reduction
@@ -131,7 +126,7 @@ public:
  template<class obj> void GlobalSum(obj &o){
    typedef typename obj::scalar_type scalar_type;
    int words = sizeof(obj)/sizeof(scalar_type);
-    scalar_type * ptr = (scalar_type *)& o; // Safe alias 
+    scalar_type * ptr = (scalar_type *)& o;
    GlobalSumVector(ptr,words);
  }
  
@@ -145,17 +140,17 @@ public:
 		      int bytes);
  
  double StencilSendToRecvFrom(void *xmit,
-			       int xmit_to_rank,int do_xmit,
+			       int xmit_to_rank,
 			       void *recv,
-			       int recv_from_rank,int do_recv,
+			       int recv_from_rank,
 			       int bytes,int dir);

  double StencilSendToRecvFromBegin(std::vector<CommsRequest_t> &list,
 				    void *xmit,
-				    int xmit_to_rank,int do_xmit,
+				    int xmit_to_rank,
 				    void *recv,
-				    int recv_from_rank,int do_recv,
-				    int xbytes,int rbytes,int dir);
+				    int recv_from_rank,
+				    int bytes,int dir);
  
  
  void StencilSendToRecvFromComplete(std::vector<CommsRequest_t> &waitall,int i);
@@ -106,7 +106,7 @@ CartesianCommunicator::CartesianCommunicator(const Coordinate &processors)
  // Remap using the shared memory optimising routine
  // The remap creates a comm which must be freed
  ////////////////////////////////////////////////////
-  GlobalSharedMemory::OptimalCommunicator    (processors,optimal_comm,_shm_processors);
+  GlobalSharedMemory::OptimalCommunicator    (processors,optimal_comm);
  InitFromMPICommunicator(processors,optimal_comm);
  SetCommunicator(optimal_comm);
  ///////////////////////////////////////////////////
@@ -124,13 +124,12 @@ CartesianCommunicator::CartesianCommunicator(const Coordinate &processors,const
  int parent_ndimension = parent._ndimension; assert(_ndimension >= parent._ndimension);
  Coordinate parent_processor_coor(_ndimension,0);
  Coordinate parent_processors    (_ndimension,1);
-  Coordinate shm_processors       (_ndimension,1);
+
  // Can make 5d grid from 4d etc...
  int pad = _ndimension-parent_ndimension;
  for(int d=0;d<parent_ndimension;d++){
    parent_processor_coor[pad+d]=parent._processor_coor[d];
    parent_processors    [pad+d]=parent._processors[d];
-    shm_processors       [pad+d]=parent._shm_processors[d];
  }

  //////////////////////////////////////////////////////////////////////////////////////////////////////
@@ -155,7 +154,6 @@ CartesianCommunicator::CartesianCommunicator(const Coordinate &processors,const
    ccoor[d] = parent_processor_coor[d] % processors[d];
    scoor[d] = parent_processor_coor[d] / processors[d];
    ssize[d] = parent_processors[d]     / processors[d];
-    if ( processors[d] < shm_processors[d] ) shm_processors[d] = processors[d]; // subnode splitting.
  }

  // rank within subcomm ; srank is rank of subcomm within blocks of subcomms
@@ -337,23 +335,23 @@ void CartesianCommunicator::SendToRecvFrom(void *xmit,
 }
 // Basic Halo comms primitive
 double CartesianCommunicator::StencilSendToRecvFrom( void *xmit,
-						     int dest, int dox,
+						     int dest,
 						     void *recv,
-						     int from, int dor,
+						     int from,
 						     int bytes,int dir)
 {
  std::vector<CommsRequest_t> list;
-  double offbytes = StencilSendToRecvFromBegin(list,xmit,dest,dox,recv,from,dor,bytes,bytes,dir);
+  double offbytes = StencilSendToRecvFromBegin(list,xmit,dest,recv,from,bytes,dir);
  StencilSendToRecvFromComplete(list,dir);
  return offbytes;
 }

 double CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsRequest_t> &list,
 							 void *xmit,
-							 int dest,int dox,
+							 int dest,
 							 void *recv,
-							 int from,int dor,
-							 int xbytes,int rbytes,int dir)
+							 int from,
+							 int bytes,int dir)
 {
  int ncomm  =communicator_halo.size();
  int commdir=dir%ncomm;
@@ -372,37 +370,30 @@ double CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsReques
  double off_node_bytes=0.0;
  int tag;

-  if ( dor ) {
-    if ( (gfrom ==MPI_UNDEFINED) || Stencil_force_mpi ) {
-      tag= dir+from*32;
-      ierr=MPI_Irecv(recv, rbytes, MPI_CHAR,from,tag,communicator_halo[commdir],&rrq);
-      assert(ierr==0);
-      list.push_back(rrq);
-      off_node_bytes+=rbytes;
-    }
+  if ( gfrom ==MPI_UNDEFINED) {
+    tag= dir+from*32;
+    ierr=MPI_Irecv(recv, bytes, MPI_CHAR,from,tag,communicator_halo[commdir],&rrq);
+    assert(ierr==0);
+    list.push_back(rrq);
+    off_node_bytes+=bytes;
  }
-  
-  if (dox) {
-    if ( (gdest == MPI_UNDEFINED) || Stencil_force_mpi ) {
-      tag= dir+_processor*32;
-      ierr =MPI_Isend(xmit, xbytes, MPI_CHAR,dest,tag,communicator_halo[commdir],&xrq);
-      assert(ierr==0);
-      list.push_back(xrq);
-      off_node_bytes+=xbytes;
-    } else {
-      void *shm = (void *) this->ShmBufferTranslate(dest,recv);
-      assert(shm!=NULL);
-      acceleratorCopyDeviceToDeviceAsynch(xmit,shm,xbytes);
-    }
+
+  if ( gdest == MPI_UNDEFINED ) {
+    tag= dir+_processor*32;
+    ierr =MPI_Isend(xmit, bytes, MPI_CHAR,dest,tag,communicator_halo[commdir],&xrq);
+    assert(ierr==0);
+    list.push_back(xrq);
+    off_node_bytes+=bytes;
+  }
+
+  if ( CommunicatorPolicy == CommunicatorPolicySequential ) {
+    this->StencilSendToRecvFromComplete(list,dir);
  }

  return off_node_bytes;
 }
 void CartesianCommunicator::StencilSendToRecvFromComplete(std::vector<CommsRequest_t> &list,int dir)
 {
-  acceleratorCopySynchronise();
-  StencilBarrier();// Synch shared memory on a single nodes
-
  int nreq=list.size();

  if (nreq==0) return;
@@ -438,10 +429,6 @@ int CartesianCommunicator::RankWorld(void){
  MPI_Comm_rank(communicator_world,&r);
  return r;
 }
-void CartesianCommunicator::BarrierWorld(void){
-  int ierr = MPI_Barrier(communicator_world);
-  assert(ierr==0);
-}
 void CartesianCommunicator::BroadcastWorld(int root,void* data, int bytes)
 {
  int ierr= MPI_Bcast(data,
@@ -45,14 +45,12 @@ void CartesianCommunicator::Init(int *argc, char *** arv)
 CartesianCommunicator::CartesianCommunicator(const Coordinate &processors,const CartesianCommunicator &parent,int &srank) 
  : CartesianCommunicator(processors) 
 {
-  _shm_processors = Coordinate(processors.size(),1);
  srank=0;
  SetCommunicator(communicator_world);
 }

 CartesianCommunicator::CartesianCommunicator(const Coordinate &processors)
 {
-  _shm_processors = Coordinate(processors.size(),1);
  _processors = processors;
  _ndimension = processors.size();  assert(_ndimension>=1);
  _processor_coor.resize(_ndimension);
@@ -104,7 +102,6 @@ int  CartesianCommunicator::RankWorld(void){return 0;}
 void CartesianCommunicator::Barrier(void){}
 void CartesianCommunicator::Broadcast(int root,void* data, int bytes) {}
 void CartesianCommunicator::BroadcastWorld(int root,void* data, int bytes) { }
-void CartesianCommunicator::BarrierWorld(void) { }
 int  CartesianCommunicator::RankFromProcessorCoor(Coordinate &coor) {  return 0;}
 void CartesianCommunicator::ProcessorCoorFromRank(int rank, Coordinate &coor){  coor = _processor_coor; }
 void CartesianCommunicator::ShiftedRanks(int dim,int shift,int &source,int &dest)
@@ -114,19 +111,19 @@ void CartesianCommunicator::ShiftedRanks(int dim,int shift,int &source,int &dest
 }

 double CartesianCommunicator::StencilSendToRecvFrom( void *xmit,
-						     int xmit_to_rank,int dox,
+						     int xmit_to_rank,
 						     void *recv,
-						     int recv_from_rank,int dor,
+						     int recv_from_rank,
 						     int bytes, int dir)
 {
  return 2.0*bytes;
 }
 double CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsRequest_t> &list,
 							 void *xmit,
-							 int xmit_to_rank,int dox,
+							 int xmit_to_rank,
 							 void *recv,
-							 int recv_from_rank,int dor,
-							 int xbytes,int rbytes, int dir)
+							 int recv_from_rank,
+							 int bytes, int dir)
 {
  return 2.0*bytes;
 }
@@ -93,10 +93,9 @@ public:
  // Create an optimal reordered communicator that makes MPI_Cart_create get it right
  //////////////////////////////////////////////////////////////////////////////////////
  static void Init(Grid_MPI_Comm comm); // Typically MPI_COMM_WORLD
-  // Turns MPI_COMM_WORLD into right layout for Cartesian
-  static void OptimalCommunicator            (const Coordinate &processors,Grid_MPI_Comm & optimal_comm,Coordinate &ShmDims); 
-  static void OptimalCommunicatorHypercube   (const Coordinate &processors,Grid_MPI_Comm & optimal_comm,Coordinate &ShmDims); 
-  static void OptimalCommunicatorSharedMemory(const Coordinate &processors,Grid_MPI_Comm & optimal_comm,Coordinate &ShmDims); 
+  static void OptimalCommunicator            (const Coordinate &processors,Grid_MPI_Comm & optimal_comm);  // Turns MPI_COMM_WORLD into right layout for Cartesian
+  static void OptimalCommunicatorHypercube   (const Coordinate &processors,Grid_MPI_Comm & optimal_comm);  // Turns MPI_COMM_WORLD into right layout for Cartesian
+  static void OptimalCommunicatorSharedMemory(const Coordinate &processors,Grid_MPI_Comm & optimal_comm);  // Turns MPI_COMM_WORLD into right layout for Cartesian
  static void GetShmDims(const Coordinate &WorldDims,Coordinate &ShmDims);
  ///////////////////////////////////////////////////
  // Provide shared memory facilities off comm world
@@ -152,7 +152,7 @@ int Log2Size(int TwoToPower,int MAXLOG2)
  }
  return log2size;
 }
-void GlobalSharedMemory::OptimalCommunicator(const Coordinate &processors,Grid_MPI_Comm & optimal_comm,Coordinate &SHM)
+void GlobalSharedMemory::OptimalCommunicator(const Coordinate &processors,Grid_MPI_Comm & optimal_comm)
 {
  //////////////////////////////////////////////////////////////////////////////
  // Look and see if it looks like an HPE 8600 based on hostname conventions
@@ -165,8 +165,8 @@ void GlobalSharedMemory::OptimalCommunicator(const Coordinate &processors,Grid_M
  gethostname(name,namelen);
  int nscan = sscanf(name,"r%di%dn%d",&R,&I,&N) ;

-  if(nscan==3 && HPEhypercube ) OptimalCommunicatorHypercube(processors,optimal_comm,SHM);
-  else                          OptimalCommunicatorSharedMemory(processors,optimal_comm,SHM);
+  if(nscan==3 && HPEhypercube ) OptimalCommunicatorHypercube(processors,optimal_comm);
+  else                          OptimalCommunicatorSharedMemory(processors,optimal_comm);
 }
 static inline int divides(int a,int b)
 {
@@ -221,7 +221,7 @@ void GlobalSharedMemory::GetShmDims(const Coordinate &WorldDims,Coordinate &ShmD
    dim=(dim+1) %ndimension;
  }
 }
-void GlobalSharedMemory::OptimalCommunicatorHypercube(const Coordinate &processors,Grid_MPI_Comm & optimal_comm,Coordinate &SHM)
+void GlobalSharedMemory::OptimalCommunicatorHypercube(const Coordinate &processors,Grid_MPI_Comm & optimal_comm)
 {
  ////////////////////////////////////////////////////////////////
  // Assert power of two shm_size.
@@ -294,8 +294,7 @@ void GlobalSharedMemory::OptimalCommunicatorHypercube(const Coordinate &processo
  Coordinate HyperCoor(ndimension);

  GetShmDims(WorldDims,ShmDims);
-  SHM = ShmDims;
-  
+
  ////////////////////////////////////////////////////////////////
  // Establish torus of processes and nodes with sub-blockings
  ////////////////////////////////////////////////////////////////
@@ -342,7 +341,7 @@ void GlobalSharedMemory::OptimalCommunicatorHypercube(const Coordinate &processo
  int ierr= MPI_Comm_split(WorldComm,0,rank,&optimal_comm);
  assert(ierr==0);
 }
-void GlobalSharedMemory::OptimalCommunicatorSharedMemory(const Coordinate &processors,Grid_MPI_Comm & optimal_comm,Coordinate &SHM)
+void GlobalSharedMemory::OptimalCommunicatorSharedMemory(const Coordinate &processors,Grid_MPI_Comm & optimal_comm)
 {
  ////////////////////////////////////////////////////////////////
  // Identify subblock of ranks on node spreading across dims
@@ -354,8 +353,6 @@ void GlobalSharedMemory::OptimalCommunicatorSharedMemory(const Coordinate &proce
  Coordinate ShmCoor(ndimension);    Coordinate NodeCoor(ndimension);   Coordinate WorldCoor(ndimension);

  GetShmDims(WorldDims,ShmDims);
-  SHM=ShmDims;
-
  ////////////////////////////////////////////////////////////////
  // Establish torus of processes and nodes with sub-blockings
  ////////////////////////////////////////////////////////////////
@@ -516,16 +513,26 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
  ///////////////////////////////////////////////////////////////////////////////////////////////////////////
  // Each MPI rank should allocate our own buffer
  ///////////////////////////////////////////////////////////////////////////////////////////////////////////
+#ifdef GRID_SYCL_LEVEL_ZERO_IPC
+  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_device_mem_alloc_desc_t zeDesc = {};
+  zeMemAllocDevice(zeContext,&zeDesc,bytes,2*1024*1024,zeDevice,&ShmCommBuf);
+  std::cout << WorldRank << header " SharedMemoryMPI.cc zeMemAllocDevice "<< bytes 
+	      << "bytes at "<< std::hex<< ShmCommBuf <<std::dec<<" for comms buffers " <<std::endl;
+#else  
  ShmCommBuf = acceleratorAllocDevice(bytes);
+#endif  
  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 << " - "<<(bytes-1+(uint64_t)ShmCommBuf) <<std::dec<<" for comms buffers " <<std::endl;
+	      << "bytes at "<< std::hex<< ShmCommBuf <<std::dec<<" for comms buffers " <<std::endl;
  }
-  SharedMemoryZero(ShmCommBuf,bytes);
+  //  SharedMemoryZero(ShmCommBuf,bytes);
  std::cout<< "Setting up IPC"<<std::endl;
  ///////////////////////////////////////////////////////////////////////////////////////////////////////////
  // Loop over ranks/gpu's on our node
@@ -536,27 +543,21 @@ 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;
-
+    ze_ipc_mem_handle_t handle;
    if ( r==WorldShmRank ) { 
-      auto err = zeMemGetIpcHandle(zeContext,ShmCommBuf,&ihandle);
+      auto err = zeMemGetIpcHandle(zeContext,ShmCommBuf,&handle);
      if ( err != ZE_RESULT_SUCCESS ) {
-	std::cout << "SharedMemoryMPI.cc zeMemGetIpcHandle failed for rank "<<r<<" "<<std::hex<<err<<std::dec<<std::endl;
+	std::cerr << "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;
+	std::cerr << "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();
+      std::cerr<<"Allocated IpcHandle rank "<<r<<" (hex) ";
+      for(int c=0;c<ZE_MAX_IPC_HANDLE_SIZE;c++){
+	std::cerr<<std::hex<<(uint32_t)((uint8_t)handle.data[c])<<std::dec;
+      }
+      std::cerr<<std::endl;
    }
 #endif
 #ifdef GRID_CUDA
@@ -579,7 +580,6 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
      }
    }
 #endif
-
    //////////////////////////////////////////////////
    // Share this IPC handle across the Shm Comm
    //////////////////////////////////////////////////
@@ -595,31 +595,22 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
    ///////////////////////////////////////////////////////////////
    // If I am not the source, overwrite thisBuf with remote buffer
    ///////////////////////////////////////////////////////////////
-
+    void * thisBuf = ShmCommBuf;
 #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);
+      std::cerr<<"Using IpcHandle rank "<<r<<" ";
+      for(int c=0;c<ZE_MAX_IPC_HANDLE_SIZE;c++){
+	std::cerr<<std::hex<<(uint32_t)((uint8_t)handle.data[c])<<std::dec;
+      }
+      std::cerr<<std::endl;
+      auto err = zeMemOpenIpcHandle(zeContext,zeDevice,handle,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; 
+	std::cerr << "SharedMemoryMPI.cc "<<zeContext<<" "<<zeDevice<<std::endl;
+	std::cerr << "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;
+	std::cerr << "SharedMemoryMPI.cc zeMemOpenIpcHandle succeeded for rank "<<r<<" "<<std::hex<<err<<std::dec<<std::endl;
      }
      assert(thisBuf!=nullptr);
    }
@@ -645,7 +636,6 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
    ///////////////////////////////////////////////////////////////
    // Save a copy of the device buffers
    ///////////////////////////////////////////////////////////////
-    }
    WorldShmCommBufs[r] = thisBuf;
 #else
    WorldShmCommBufs[r] = ShmCommBuf;
@@ -48,10 +48,9 @@ void GlobalSharedMemory::Init(Grid_MPI_Comm comm)
  _ShmSetup=1;
 }

-void GlobalSharedMemory::OptimalCommunicator(const Coordinate &processors,Grid_MPI_Comm & optimal_comm,Coordinate &SHM)
+void GlobalSharedMemory::OptimalCommunicator(const Coordinate &processors,Grid_MPI_Comm & optimal_comm)
 {
  optimal_comm = WorldComm;
-  SHM = Coordinate(processors.size(),1);
 }

 ////////////////////////////////////////////////////////////////////////////////////////////
@@ -46,4 +46,3 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 #include <Grid/lattice/Lattice_unary.h>
 #include <Grid/lattice/Lattice_transfer.h>
 #include <Grid/lattice/Lattice_basis.h>
-#include <Grid/lattice/Lattice_crc.h>
@@ -63,7 +63,7 @@ accelerator_inline vobj predicatedWhere(const iobj &predicate,
  typename std::remove_const<vobj>::type ret;

  typedef typename vobj::scalar_object scalar_object;
-  //  typedef typename vobj::scalar_type scalar_type;
+  typedef typename vobj::scalar_type scalar_type;
  typedef typename vobj::vector_type vector_type;

  const int Nsimd = vobj::vector_type::Nsimd();
@@ -36,7 +36,6 @@ NAMESPACE_BEGIN(Grid);
 //////////////////////////////////////////////////////////////////////////////////////////////////////
 template<class obj1,class obj2,class obj3> inline
 void mult(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const Lattice<obj3> &rhs){
-  GRID_TRACE("mult");
  ret.Checkerboard() = lhs.Checkerboard();
  autoView( ret_v , ret, AcceleratorWrite);
  autoView( lhs_v , lhs, AcceleratorRead);
@@ -54,7 +53,6 @@ void mult(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const Lattice<obj3> &rhs){
  
 template<class obj1,class obj2,class obj3> inline
 void mac(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const Lattice<obj3> &rhs){
-  GRID_TRACE("mac");
  ret.Checkerboard() = lhs.Checkerboard();
  conformable(ret,rhs);
  conformable(lhs,rhs);
@@ -72,7 +70,6 @@ void mac(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const Lattice<obj3> &rhs){
  
 template<class obj1,class obj2,class obj3> inline
 void sub(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const Lattice<obj3> &rhs){
-  GRID_TRACE("sub");
  ret.Checkerboard() = lhs.Checkerboard();
  conformable(ret,rhs);
  conformable(lhs,rhs);
@@ -89,7 +86,6 @@ void sub(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const Lattice<obj3> &rhs){
 }
 template<class obj1,class obj2,class obj3> inline
 void add(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const Lattice<obj3> &rhs){
-  GRID_TRACE("add");
  ret.Checkerboard() = lhs.Checkerboard();
  conformable(ret,rhs);
  conformable(lhs,rhs);
@@ -110,7 +106,6 @@ void add(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const Lattice<obj3> &rhs){
 //////////////////////////////////////////////////////////////////////////////////////////////////////
 template<class obj1,class obj2,class obj3> inline
 void mult(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
-  GRID_TRACE("mult");
  ret.Checkerboard() = lhs.Checkerboard();
  conformable(lhs,ret);
  autoView( ret_v , ret, AcceleratorWrite);
@@ -124,7 +119,6 @@ void mult(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
  
 template<class obj1,class obj2,class obj3> inline
 void mac(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
-  GRID_TRACE("mac");
  ret.Checkerboard() = lhs.Checkerboard();
  conformable(ret,lhs);
  autoView( ret_v , ret, AcceleratorWrite);
@@ -139,7 +133,6 @@ void mac(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
  
 template<class obj1,class obj2,class obj3> inline
 void sub(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
-  GRID_TRACE("sub");
  ret.Checkerboard() = lhs.Checkerboard();
  conformable(ret,lhs);
  autoView( ret_v , ret, AcceleratorWrite);
@@ -153,7 +146,6 @@ void sub(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
 }
 template<class obj1,class obj2,class obj3> inline
 void add(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
-  GRID_TRACE("add");
  ret.Checkerboard() = lhs.Checkerboard();
  conformable(lhs,ret);
  autoView( ret_v , ret, AcceleratorWrite);
@@ -171,7 +163,6 @@ void add(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
 //////////////////////////////////////////////////////////////////////////////////////////////////////
 template<class obj1,class obj2,class obj3> inline
 void mult(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
-  GRID_TRACE("mult");
  ret.Checkerboard() = rhs.Checkerboard();
  conformable(ret,rhs);
  autoView( ret_v , ret, AcceleratorWrite);
@@ -186,7 +177,6 @@ void mult(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
  
 template<class obj1,class obj2,class obj3> inline
 void mac(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
-  GRID_TRACE("mac");
  ret.Checkerboard() = rhs.Checkerboard();
  conformable(ret,rhs);
  autoView( ret_v , ret, AcceleratorWrite);
@@ -201,7 +191,6 @@ void mac(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
  
 template<class obj1,class obj2,class obj3> inline
 void sub(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
-  GRID_TRACE("sub");
  ret.Checkerboard() = rhs.Checkerboard();
  conformable(ret,rhs);
  autoView( ret_v , ret, AcceleratorWrite);
@@ -215,7 +204,6 @@ void sub(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
 }
 template<class obj1,class obj2,class obj3> inline
 void add(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
-  GRID_TRACE("add");
  ret.Checkerboard() = rhs.Checkerboard();
  conformable(ret,rhs);
  autoView( ret_v , ret, AcceleratorWrite);
@@ -230,7 +218,6 @@ void add(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
  
 template<class sobj,class vobj> inline
 void axpy(Lattice<vobj> &ret,sobj a,const Lattice<vobj> &x,const Lattice<vobj> &y){
-  GRID_TRACE("axpy");
  ret.Checkerboard() = x.Checkerboard();
  conformable(ret,x);
  conformable(x,y);
@@ -238,13 +225,12 @@ 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);
  });
 }
 template<class sobj,class vobj> inline
 void axpby(Lattice<vobj> &ret,sobj a,sobj b,const Lattice<vobj> &x,const Lattice<vobj> &y){
-  GRID_TRACE("axpby");
  ret.Checkerboard() = x.Checkerboard();
  conformable(ret,x);
  conformable(x,y);
@@ -260,13 +246,11 @@ void axpby(Lattice<vobj> &ret,sobj a,sobj b,const Lattice<vobj> &x,const Lattice
 template<class sobj,class vobj> inline
 RealD axpy_norm(Lattice<vobj> &ret,sobj a,const Lattice<vobj> &x,const Lattice<vobj> &y)
 {
-  GRID_TRACE("axpy_norm");
    return axpy_norm_fast(ret,a,x,y);
 }
 template<class sobj,class vobj> inline
 RealD axpby_norm(Lattice<vobj> &ret,sobj a,sobj b,const Lattice<vobj> &x,const Lattice<vobj> &y)
 {
-  GRID_TRACE("axpby_norm");
    return axpby_norm_fast(ret,a,b,x,y);
 }

@@ -88,13 +88,6 @@ public:
    LatticeView<vobj> accessor(*( (LatticeAccelerator<vobj> *) this),mode);
    accessor.ViewClose();
  }
-
-  // Helper function to print the state of this object in the AccCache
-  void PrintCacheState(void)
-  {
-    MemoryManager::PrintState(this->_odata);
-  }
-
  /////////////////////////////////////////////////////////////////////////////////
  // Return a view object that may be dereferenced in site loops.
  // The view is trivially copy constructible and may be copied to an accelerator device
@@ -117,7 +110,6 @@ public:
  ////////////////////////////////////////////////////////////////////////////////
  template <typename Op, typename T1> inline Lattice<vobj> & operator=(const LatticeUnaryExpression<Op,T1> &expr)
  {
-    GRID_TRACE("ExpressionTemplateEval");
    GridBase *egrid(nullptr);
    GridFromExpression(egrid,expr);
    assert(egrid!=nullptr);
@@ -141,7 +133,6 @@ public:
  }
  template <typename Op, typename T1,typename T2> inline Lattice<vobj> & operator=(const LatticeBinaryExpression<Op,T1,T2> &expr)
  {
-    GRID_TRACE("ExpressionTemplateEval");
    GridBase *egrid(nullptr);
    GridFromExpression(egrid,expr);
    assert(egrid!=nullptr);
@@ -165,7 +156,6 @@ public:
  }
  template <typename Op, typename T1,typename T2,typename T3> inline Lattice<vobj> & operator=(const LatticeTrinaryExpression<Op,T1,T2,T3> &expr)
  {
-    GRID_TRACE("ExpressionTemplateEval");
    GridBase *egrid(nullptr);
    GridFromExpression(egrid,expr);
    assert(egrid!=nullptr);
@@ -125,7 +125,7 @@ void basisRotate(VField &basis,Matrix& Qt,int j0, int j1, int k0,int k1,int Nm)

 	for(int k=k0; k<k1; ++k){
 	  auto tmp = coalescedRead(Bp[ss*nrot+j]);
-	  coalescedWrite(Bp[ss*nrot+j],tmp+ Qt_p[jj*Nm+k] * coalescedRead(basis_vp[k][sss]));
+	  coalescedWrite(Bp[ss*nrot+j],tmp+ Qt_p[jj*Nm+k] * coalescedRead(basis_v[k][sss]));
 	}
      });

@@ -134,7 +134,7 @@ void basisRotate(VField &basis,Matrix& Qt,int j0, int j1, int k0,int k1,int Nm)
 	int jj  =j0+j;
 	int ss =sj/nrot;
 	int sss=ss+s;
-	coalescedWrite(basis_vp[jj][sss],coalescedRead(Bp[ss*nrot+j]));
+	coalescedWrite(basis_v[jj][sss],coalescedRead(Bp[ss*nrot+j]));
      });
  }
 #endif
@@ -1,55 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/lattice/Lattice_crc.h
-
-    Copyright (C) 2021
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#pragma once
-
-NAMESPACE_BEGIN(Grid);
-
-template<class vobj> void DumpSliceNorm(std::string s,Lattice<vobj> &f,int mu=-1)
-{
-  auto ff = localNorm2(f);
-  if ( mu==-1 ) mu = f.Grid()->Nd()-1;
-  typedef typename vobj::tensor_reduced normtype;
-  typedef typename normtype::scalar_object scalar;
-  std::vector<scalar> sff;
-  sliceSum(ff,sff,mu);
-  for(int t=0;t<sff.size();t++){
-    std::cout << s<<" "<<t<<" "<<sff[t]<<std::endl;
-  }
-}
-
-template<class vobj> uint32_t crc(Lattice<vobj> & buf)
-{
-  autoView( buf_v , buf, CpuRead);
-  return ::crc32(0L,(unsigned char *)&buf_v[0],(size_t)sizeof(vobj)*buf.oSites());
-}
-
-#define CRC(U) std::cout << "FingerPrint "<<__FILE__ <<" "<< __LINE__ <<" "<< #U <<" "<<crc(U)<<std::endl;
-
-NAMESPACE_END(Grid);
-
-
@@ -32,6 +32,7 @@ template<class vobj>
 static void sliceMaddMatrix (Lattice<vobj> &R,Eigen::MatrixXcd &aa,const Lattice<vobj> &X,const Lattice<vobj> &Y,int Orthog,RealD scale=1.0) 
 {    
  typedef typename vobj::scalar_object sobj;
+  typedef typename vobj::scalar_type scalar_type;
  typedef typename vobj::vector_type vector_type;

  int Nblock = X.Grid()->GlobalDimensions()[Orthog];
@@ -81,6 +82,7 @@ template<class vobj>
 static void sliceMulMatrix (Lattice<vobj> &R,Eigen::MatrixXcd &aa,const Lattice<vobj> &X,int Orthog,RealD scale=1.0) 
 {    
  typedef typename vobj::scalar_object sobj;
+  typedef typename vobj::scalar_type scalar_type;
  typedef typename vobj::vector_type vector_type;

  int Nblock = X.Grid()->GlobalDimensions()[Orthog];
@@ -128,6 +130,7 @@ template<class vobj>
 static void sliceInnerProductMatrix(  Eigen::MatrixXcd &mat, const Lattice<vobj> &lhs,const Lattice<vobj> &rhs,int Orthog) 
 {
  typedef typename vobj::scalar_object sobj;
+  typedef typename vobj::scalar_type scalar_type;
  typedef typename vobj::vector_type vector_type;
  
  GridBase *FullGrid  = lhs.Grid();
@@ -96,6 +96,9 @@ void pokeSite(const sobj &s,Lattice<vobj> &l,const Coordinate &site){

  GridBase *grid=l.Grid();

+  typedef typename vobj::scalar_type scalar_type;
+  typedef typename vobj::vector_type vector_type;
+
  int Nsimd = grid->Nsimd();

  assert( l.Checkerboard()== l.Grid()->CheckerBoard(site));
@@ -122,17 +125,14 @@ void pokeSite(const sobj &s,Lattice<vobj> &l,const Coordinate &site){
 //////////////////////////////////////////////////////////
 // Peek a scalar object from the SIMD array
 //////////////////////////////////////////////////////////
-template<class vobj>
-typename vobj::scalar_object peekSite(const Lattice<vobj> &l,const Coordinate &site){
-  typename vobj::scalar_object s;
-  peekSite(s,l,site);
-  return s;
-}        
 template<class vobj,class sobj>
 void peekSite(sobj &s,const Lattice<vobj> &l,const Coordinate &site){
        
  GridBase *grid=l.Grid();

+  typedef typename vobj::scalar_type scalar_type;
+  typedef typename vobj::vector_type vector_type;
+
  int Nsimd = grid->Nsimd();

  assert( l.Checkerboard() == l.Grid()->CheckerBoard(site));
@@ -173,11 +173,11 @@ inline void peekLocalSite(sobj &s,const LatticeView<vobj> &l,Coordinate &site)
  idx= grid->iIndex(site);
  odx= grid->oIndex(site);
  
-  const vector_type *vp = (const vector_type *) &l[odx];
+  scalar_type * vp = (scalar_type *)&l[odx];
  scalar_type * pt = (scalar_type *)&s;
      
  for(int w=0;w<words;w++){
-    pt[w] = getlane(vp[w],idx);
+    pt[w] = vp[idx+w*Nsimd];
  }
      
  return;
@@ -210,10 +210,10 @@ inline void pokeLocalSite(const sobj &s,LatticeView<vobj> &l,Coordinate &site)
  idx= grid->iIndex(site);
  odx= grid->oIndex(site);

-  vector_type * vp = (vector_type *)&l[odx];
+  scalar_type * vp = (scalar_type *)&l[odx];
  scalar_type * pt = (scalar_type *)&s;
  for(int w=0;w<words;w++){
-    putlane(vp[w],pt[w],idx);
+    vp[idx+w*Nsimd] = pt[w];
  }
  return;
 };
@@ -91,7 +91,10 @@ inline typename vobj::scalar_objectD sumD_cpu(const vobj *arg, Integer osites)
  for(int i=0;i<nthread;i++){
    ssum = ssum+sumarray[i];
  } 
-  return ssum;
+  
+  typedef typename vobj::scalar_object ssobj;
+  ssobj ret = ssum;
+  return ret;
 }
 /*
 Threaded max, don't use for now
@@ -139,15 +142,6 @@ inline typename vobj::scalar_objectD sumD(const vobj *arg, Integer osites)
  return sumD_cpu(arg,osites);
 #endif  
 }
-template<class vobj>
-inline typename vobj::scalar_objectD sumD_large(const vobj *arg, Integer osites)
-{
-#if defined(GRID_CUDA)||defined(GRID_HIP)
-  return sumD_gpu_large(arg,osites);
-#else
-  return sumD_cpu(arg,osites);
-#endif  
-}

 template<class vobj>
 inline typename vobj::scalar_object sum(const Lattice<vobj> &arg)
@@ -165,22 +159,6 @@ inline typename vobj::scalar_object sum(const Lattice<vobj> &arg)
  return ssum;
 }

-template<class vobj>
-inline typename vobj::scalar_object sum_large(const Lattice<vobj> &arg)
-{
-#if defined(GRID_CUDA)||defined(GRID_HIP)
-  autoView( arg_v, arg, AcceleratorRead);
-  Integer osites = arg.Grid()->oSites();
-  auto ssum= sum_gpu_large(&arg_v[0],osites);
-#else
-  autoView(arg_v, arg, CpuRead);
-  Integer osites = arg.Grid()->oSites();
-  auto ssum= sum_cpu(&arg_v[0],osites);
-#endif
-  arg.Grid()->GlobalSum(ssum);
-  return ssum;
-}
-
 ////////////////////////////////////////////////////////////////////////////////////////////////////
 // Deterministic Reduction operations
 ////////////////////////////////////////////////////////////////////////////////////////////////////
@@ -219,6 +197,7 @@ template<class vobj> inline RealD maxLocalNorm2(const Lattice<vobj> &arg)
 template<class vobj>
 inline ComplexD rankInnerProduct(const Lattice<vobj> &left,const Lattice<vobj> &right)
 {
+  typedef typename vobj::scalar_type scalar_type;
  typedef typename vobj::vector_typeD vector_type;
  ComplexD  nrm;
  
@@ -228,7 +207,6 @@ inline ComplexD rankInnerProduct(const Lattice<vobj> &left,const Lattice<vobj> &
  const uint64_t sites = grid->oSites();
  
  // Might make all code paths go this way.
-#if 0
  typedef decltype(innerProductD(vobj(),vobj())) inner_t;
  Vector<inner_t> inner_tmp(sites);
  auto inner_tmp_v = &inner_tmp[0];
@@ -238,31 +216,15 @@ inline ComplexD rankInnerProduct(const Lattice<vobj> &left,const Lattice<vobj> &
    autoView( right_v,right, AcceleratorRead);

    // GPU - SIMT lane compliance...
-    accelerator_for( ss, sites, nsimd,{
-	auto x_l = left_v(ss);
-	auto y_l = right_v(ss);
-	coalescedWrite(inner_tmp_v[ss],innerProductD(x_l,y_l));
+    accelerator_for( ss, sites, 1,{
+	auto x_l = left_v[ss];
+	auto y_l = right_v[ss];
+	inner_tmp_v[ss]=innerProductD(x_l,y_l);
    });
  }
-#else
-  typedef decltype(innerProduct(vobj(),vobj())) inner_t;
-  Vector<inner_t> inner_tmp(sites);
-  auto inner_tmp_v = &inner_tmp[0];
-    
-  {
-    autoView( left_v , left, AcceleratorRead);
-    autoView( right_v,right, AcceleratorRead);

-    // GPU - SIMT lane compliance...
-    accelerator_for( ss, sites, nsimd,{
-	auto x_l = left_v(ss);
-	auto y_l = right_v(ss);
-	coalescedWrite(inner_tmp_v[ss],innerProduct(x_l,y_l));
-    });
-  }
-#endif
  // This is in single precision and fails some tests
-  auto anrm = sumD(inner_tmp_v,sites);  
+  auto anrm = sum(inner_tmp_v,sites);  
  nrm = anrm;
  return nrm;
 }
@@ -295,7 +257,8 @@ axpby_norm_fast(Lattice<vobj> &z,sobj a,sobj b,const Lattice<vobj> &x,const Latt
  conformable(z,x);
  conformable(x,y);

-  //  typedef typename vobj::vector_typeD vector_type;
+  typedef typename vobj::scalar_type scalar_type;
+  typedef typename vobj::vector_typeD vector_type;
  RealD  nrm;
  
  GridBase *grid = x.Grid();
@@ -307,29 +270,17 @@ axpby_norm_fast(Lattice<vobj> &z,sobj a,sobj b,const Lattice<vobj> &x,const Latt
  autoView( x_v, x, AcceleratorRead);
  autoView( y_v, y, AcceleratorRead);
  autoView( z_v, z, AcceleratorWrite);
-#if 0
+
  typedef decltype(innerProductD(x_v[0],y_v[0])) inner_t;
  Vector<inner_t> inner_tmp(sites);
  auto inner_tmp_v = &inner_tmp[0];

-  accelerator_for( ss, sites, nsimd,{
-      auto tmp = a*x_v(ss)+b*y_v(ss);
-      coalescedWrite(inner_tmp_v[ss],innerProductD(tmp,tmp));
-      coalescedWrite(z_v[ss],tmp);
+  accelerator_for( ss, sites, 1,{
+      auto tmp = a*x_v[ss]+b*y_v[ss];
+      inner_tmp_v[ss]=innerProductD(tmp,tmp);
+      z_v[ss]=tmp;
  });
  nrm = real(TensorRemove(sum(inner_tmp_v,sites)));
-#else
-  typedef decltype(innerProduct(x_v[0],y_v[0])) inner_t;
-  Vector<inner_t> inner_tmp(sites);
-  auto inner_tmp_v = &inner_tmp[0];
-
-  accelerator_for( ss, sites, nsimd,{
-      auto tmp = a*x_v(ss)+b*y_v(ss);
-      coalescedWrite(inner_tmp_v[ss],innerProduct(tmp,tmp));
-      coalescedWrite(z_v[ss],tmp);
-  });
-  nrm = real(TensorRemove(sumD(inner_tmp_v,sites)));
-#endif
  grid->GlobalSum(nrm);
  return nrm; 
 }
@@ -339,6 +290,7 @@ innerProductNorm(ComplexD& ip, RealD &nrm, const Lattice<vobj> &left,const Latti
 {
  conformable(left,right);

+  typedef typename vobj::scalar_type scalar_type;
  typedef typename vobj::vector_typeD vector_type;
  Vector<ComplexD> tmp(2);

@@ -409,7 +361,6 @@ template<class vobj> inline void sliceSum(const Lattice<vobj> &Data,std::vector<
  // But easily avoided by using double precision fields
  ///////////////////////////////////////////////////////
  typedef typename vobj::scalar_object sobj;
-  typedef typename vobj::scalar_object::scalar_type scalar_type;
  GridBase  *grid = Data.Grid();
  assert(grid!=NULL);

@@ -468,27 +419,20 @@ template<class vobj> inline void sliceSum(const Lattice<vobj> &Data,std::vector<
  }
  
  // sum over nodes.
+  sobj gsum;
  for(int t=0;t<fd;t++){
    int pt = t/ld; // processor plane
    int lt = t%ld;
    if ( pt == grid->_processor_coor[orthogdim] ) {
-      result[t]=lsSum[lt];
+      gsum=lsSum[lt];
    } else {
-      result[t]=Zero();
+      gsum=Zero();
    }

+    grid->GlobalSum(gsum);
+
+    result[t]=gsum;
  }
-  scalar_type * ptr = (scalar_type *) &result[0];
-  int words = fd*sizeof(sobj)/sizeof(scalar_type);
-  grid->GlobalSumVector(ptr, words);
-}
-template<class vobj> inline
-std::vector<typename vobj::scalar_object> 
-sliceSum(const Lattice<vobj> &Data,int orthogdim)
-{
-  std::vector<typename vobj::scalar_object> result;
-  sliceSum(Data,result,orthogdim);
-  return result;
 }

 template<class vobj>
@@ -594,8 +538,7 @@ static void sliceNorm (std::vector<RealD> &sn,const Lattice<vobj> &rhs,int Ortho
 template<class vobj>
 static void sliceMaddVector(Lattice<vobj> &R,std::vector<RealD> &a,const Lattice<vobj> &X,const Lattice<vobj> &Y,
 			    int orthogdim,RealD scale=1.0) 
-{
-  // perhaps easier to just promote A to a field and use regular madd
+{    
  typedef typename vobj::scalar_object sobj;
  typedef typename vobj::scalar_type scalar_type;
  typedef typename vobj::vector_type vector_type;
@@ -626,7 +569,8 @@ static void sliceMaddVector(Lattice<vobj> &R,std::vector<RealD> &a,const Lattice
    for(int l=0;l<Nsimd;l++){
      grid->iCoorFromIindex(icoor,l);
      int ldx =r+icoor[orthogdim]*rd;
-      av.putlane(scalar_type(a[ldx])*zscale,l);
+      scalar_type *as =(scalar_type *)&av;
+      as[l] = scalar_type(a[ldx])*zscale;
    }

    tensor_reduced at; at=av;
@@ -666,6 +610,7 @@ template<class vobj>
 static void sliceMaddMatrix (Lattice<vobj> &R,Eigen::MatrixXcd &aa,const Lattice<vobj> &X,const Lattice<vobj> &Y,int Orthog,RealD scale=1.0) 
 {    
  typedef typename vobj::scalar_object sobj;
+  typedef typename vobj::scalar_type scalar_type;
  typedef typename vobj::vector_type vector_type;

  int Nblock = X.Grid()->GlobalDimensions()[Orthog];
@@ -719,6 +664,7 @@ template<class vobj>
 static void sliceMulMatrix (Lattice<vobj> &R,Eigen::MatrixXcd &aa,const Lattice<vobj> &X,int Orthog,RealD scale=1.0) 
 {    
  typedef typename vobj::scalar_object sobj;
+  typedef typename vobj::scalar_type scalar_type;
  typedef typename vobj::vector_type vector_type;

  int Nblock = X.Grid()->GlobalDimensions()[Orthog];
@@ -772,6 +718,7 @@ template<class vobj>
 static void sliceInnerProductMatrix(  Eigen::MatrixXcd &mat, const Lattice<vobj> &lhs,const Lattice<vobj> &rhs,int Orthog) 
 {
  typedef typename vobj::scalar_object sobj;
+  typedef typename vobj::scalar_type scalar_type;
  typedef typename vobj::vector_type vector_type;
  
  GridBase *FullGrid  = lhs.Grid();
@@ -23,7 +23,7 @@ unsigned int nextPow2(Iterator x) {
 }

 template <class Iterator>
-int getNumBlocksAndThreads(const Iterator n, const size_t sizeofsobj, Iterator &threads, Iterator &blocks) {
+void getNumBlocksAndThreads(const Iterator n, const size_t sizeofsobj, Iterator &threads, Iterator &blocks) {
  
  int device;
 #ifdef GRID_CUDA
@@ -37,13 +37,14 @@ int getNumBlocksAndThreads(const Iterator n, const size_t sizeofsobj, Iterator &
  Iterator sharedMemPerBlock   = gpu_props[device].sharedMemPerBlock;
  Iterator maxThreadsPerBlock  = gpu_props[device].maxThreadsPerBlock;
  Iterator multiProcessorCount = gpu_props[device].multiProcessorCount;
-  /*  
+  
  std::cout << GridLogDebug << "GPU has:" << std::endl;
  std::cout << GridLogDebug << "\twarpSize            = " << warpSize << std::endl;
  std::cout << GridLogDebug << "\tsharedMemPerBlock   = " << sharedMemPerBlock << std::endl;
  std::cout << GridLogDebug << "\tmaxThreadsPerBlock  = " << maxThreadsPerBlock << std::endl;
+  std::cout << GridLogDebug << "\tmaxThreadsPerBlock  = " << warpSize << std::endl;
  std::cout << GridLogDebug << "\tmultiProcessorCount = " << multiProcessorCount << std::endl;
-  */  
+  
  if (warpSize != WARP_SIZE) {
    std::cout << GridLogError << "The warp size of the GPU in use does not match the warp size set when compiling Grid." << std::endl;
    exit(EXIT_FAILURE);
@@ -51,14 +52,10 @@ int getNumBlocksAndThreads(const Iterator n, const size_t sizeofsobj, Iterator &
  
  // let the number of threads in a block be a multiple of 2, starting from warpSize
  threads = warpSize;
-  if ( threads*sizeofsobj > sharedMemPerBlock ) {
-    std::cout << GridLogError << "The object is too large for the shared memory." << std::endl;
-    return 0;
-  }
  while( 2*threads*sizeofsobj < sharedMemPerBlock && 2*threads <= maxThreadsPerBlock ) threads *= 2;
  // keep all the streaming multiprocessors busy
  blocks = nextPow2(multiProcessorCount);
-  return 1;
+  
 }

 template <class sobj, class Iterator>
@@ -198,7 +195,7 @@ __global__ void reduceKernel(const vobj *lat, sobj *buffer, Iterator n) {
 // Possibly promote to double and sum
 /////////////////////////////////////////////////////////////////////////////////////////////////////////
 template <class vobj>
-inline typename vobj::scalar_objectD sumD_gpu_small(const vobj *lat, Integer osites) 
+inline typename vobj::scalar_objectD sumD_gpu(const vobj *lat, Integer osites) 
 {
  typedef typename vobj::scalar_objectD sobj;
  typedef decltype(lat) Iterator;
@@ -207,9 +204,7 @@ inline typename vobj::scalar_objectD sumD_gpu_small(const vobj *lat, Integer osi
  Integer size = osites*nsimd;

  Integer numThreads, numBlocks;
-  int ok = getNumBlocksAndThreads(size, sizeof(sobj), numThreads, numBlocks);
-  assert(ok);
-
+  getNumBlocksAndThreads(size, sizeof(sobj), numThreads, numBlocks);
  Integer smemSize = numThreads * sizeof(sobj);

  Vector<sobj> buffer(numBlocks);
@@ -220,52 +215,6 @@ inline typename vobj::scalar_objectD sumD_gpu_small(const vobj *lat, Integer osi
  auto result = buffer_v[0];
  return result;
 }
-
-template <class vobj>
-inline typename vobj::scalar_objectD sumD_gpu_large(const vobj *lat, Integer osites)
-{
-  typedef typename vobj::vector_type  vector;
-  typedef typename vobj::scalar_typeD scalarD;
-  typedef typename vobj::scalar_objectD sobj;
-  sobj ret;
-  scalarD *ret_p = (scalarD *)&ret;
-  
-  const int words = sizeof(vobj)/sizeof(vector);
-
-  Vector<vector> buffer(osites);
-  vector *dat = (vector *)lat;
-  vector *buf = &buffer[0];
-  iScalar<vector> *tbuf =(iScalar<vector> *)  &buffer[0];
-  for(int w=0;w<words;w++) {
-
-    accelerator_for(ss,osites,1,{
-	buf[ss] = dat[ss*words+w];
-      });
-      
-    ret_p[w] = sumD_gpu_small(tbuf,osites);
-  }
-  return ret;
-}
-
-template <class vobj>
-inline typename vobj::scalar_objectD sumD_gpu(const vobj *lat, Integer osites)
-{
-  typedef typename vobj::scalar_objectD sobj;
-  sobj ret;
-  
-  Integer nsimd= vobj::Nsimd();
-  Integer size = osites*nsimd;
-  Integer numThreads, numBlocks;
-  int ok = getNumBlocksAndThreads(size, sizeof(sobj), numThreads, numBlocks);
-  
-  if ( ok ) {
-    ret = sumD_gpu_small(lat,osites);
-  } else {
-    ret = sumD_gpu_large(lat,osites);
-  }
-  return ret;
-}
-
 /////////////////////////////////////////////////////////////////////////////////////////////////////////
 // Return as same precision as input performing reduction in double precision though
 /////////////////////////////////////////////////////////////////////////////////////////////////////////
@@ -278,13 +227,6 @@ inline typename vobj::scalar_object sum_gpu(const vobj *lat, Integer osites)
  return result;
 }

-template <class vobj>
-inline typename vobj::scalar_object sum_gpu_large(const vobj *lat, Integer osites)
-{
-  typedef typename vobj::scalar_object sobj;
-  sobj result;
-  result = sumD_gpu_large(lat,osites);
-  return result;
-}
+

 NAMESPACE_END(Grid);
@@ -424,32 +424,9 @@ public:
    // MT implementation does not implement fast discard even though
    // in principle this is possible
    ////////////////////////////////////////////////
-#if 1
-    thread_for( lidx, _grid->lSites(), {

-	int gidx;
-	int o_idx;
-	int i_idx;
-	int rank;
-	Coordinate pcoor;
-	Coordinate lcoor;
-	Coordinate gcoor;
-	_grid->LocalIndexToLocalCoor(lidx,lcoor);
-	pcoor=_grid->ThisProcessorCoor();
-	_grid->ProcessorCoorLocalCoorToGlobalCoor(pcoor,lcoor,gcoor);
-	_grid->GlobalCoorToGlobalIndex(gcoor,gidx);
-
-	_grid->GlobalCoorToRankIndex(rank,o_idx,i_idx,gcoor);
-	assert(rank == _grid->ThisRank() );
-	
-	int l_idx=generator_idx(o_idx,i_idx);
-	_generators[l_idx] = master_engine;
-	Skip(_generators[l_idx],gidx); // Skip to next RNG sequence
-    });
-#else
    // Everybody loops over global volume.
    thread_for( gidx, _grid->_gsites, {
-
 	// Where is it?
 	int rank;
 	int o_idx;
@@ -466,7 +443,6 @@ public:
 	  Skip(_generators[l_idx],gidx); // Skip to next RNG sequence
 	}
    });
-#endif
 #else 
    ////////////////////////////////////////////////////////////////
    // Machine and thread decomposition dependent seeding is efficient
@@ -85,76 +85,6 @@ template<class vobj> inline void setCheckerboard(Lattice<vobj> &full,const Latti
  });
 }

-template<class vobj> inline void acceleratorPickCheckerboard(int cb,Lattice<vobj> &half,const Lattice<vobj> &full, int checker_dim_half=0)
-{
-  half.Checkerboard() = cb;
-  autoView(half_v, half, AcceleratorWrite);
-  autoView(full_v, full, AcceleratorRead);
-  Coordinate rdim_full             = full.Grid()->_rdimensions;
-  Coordinate rdim_half             = half.Grid()->_rdimensions;
-  unsigned long ndim_half          = half.Grid()->_ndimension;
-  Coordinate checker_dim_mask_half = half.Grid()->_checker_dim_mask;
-  Coordinate ostride_half          = half.Grid()->_ostride;
-  accelerator_for(ss, full.Grid()->oSites(),full.Grid()->Nsimd(),{
-    
-    Coordinate coor;
-    int cbos;
-    int linear=0;
-
-    Lexicographic::CoorFromIndex(coor,ss,rdim_full);
-    assert(coor.size()==ndim_half);
-
-    for(int d=0;d<ndim_half;d++){ 
-      if(checker_dim_mask_half[d]) linear += coor[d];
-    }
-    cbos = (linear&0x1);
-
-    if (cbos==cb) {
-      int ssh=0;
-      for(int d=0;d<ndim_half;d++) {
-        if (d == checker_dim_half) ssh += ostride_half[d] * ((coor[d] / 2) % rdim_half[d]);
-        else ssh += ostride_half[d] * (coor[d] % rdim_half[d]);
-      }
-      coalescedWrite(half_v[ssh],full_v(ss));
-    }
-  });
-}
-template<class vobj> inline void acceleratorSetCheckerboard(Lattice<vobj> &full,const Lattice<vobj> &half, int checker_dim_half=0)
-{
-  int cb = half.Checkerboard();
-  autoView(half_v , half, AcceleratorRead);
-  autoView(full_v , full, AcceleratorWrite);
-  Coordinate rdim_full             = full.Grid()->_rdimensions;
-  Coordinate rdim_half             = half.Grid()->_rdimensions;
-  unsigned long ndim_half          = half.Grid()->_ndimension;
-  Coordinate checker_dim_mask_half = half.Grid()->_checker_dim_mask;
-  Coordinate ostride_half          = half.Grid()->_ostride;
-  accelerator_for(ss,full.Grid()->oSites(),full.Grid()->Nsimd(),{
-
-    Coordinate coor;
-    int cbos;
-    int linear=0;
-  
-    Lexicographic::CoorFromIndex(coor,ss,rdim_full);
-    assert(coor.size()==ndim_half);
-
-    for(int d=0;d<ndim_half;d++){ 
-      if(checker_dim_mask_half[d]) linear += coor[d];
-    }
-    cbos = (linear&0x1);
-
-    if (cbos==cb) {
-      int ssh=0;
-      for(int d=0;d<ndim_half;d++){
-        if (d == checker_dim_half) ssh += ostride_half[d] * ((coor[d] / 2) % rdim_half[d]);
-        else ssh += ostride_half[d] * (coor[d] % rdim_half[d]);
-      }
-      coalescedWrite(full_v[ss],half_v(ssh));
-    }
-
-  });
-}
-
 ////////////////////////////////////////////////////////////////////////////////////////////
 // Flexible Type Conversion for internal promotion to double as well as graceful
 // treatment of scalar-compatible types
@@ -194,11 +124,11 @@ accelerator_inline void convertType(vComplexD2 & out, const ComplexD & in) {
 #endif

 accelerator_inline void convertType(vComplexF & out, const vComplexD2 & in) {
-  precisionChange(out,in);
+  out.v = Optimization::PrecisionChange::DtoS(in._internal[0].v,in._internal[1].v);
 }

 accelerator_inline void convertType(vComplexD2 & out, const vComplexF & in) {
-  precisionChange(out,in);
+  Optimization::PrecisionChange::StoD(in.v,out._internal[0].v,out._internal[1].v);
 }

 template<typename T1,typename T2>
@@ -434,22 +364,16 @@ inline void blockSum(Lattice<vobj> &coarseData,const Lattice<vobj> &fineData)
  autoView( coarseData_ , coarseData, AcceleratorWrite);
  autoView( fineData_   , fineData, AcceleratorRead);

-  auto coarseData_p = &coarseData_[0];
-  auto fineData_p = &fineData_[0];
-  
  Coordinate fine_rdimensions = fine->_rdimensions;
  Coordinate coarse_rdimensions = coarse->_rdimensions;
-
-  vobj zz = Zero();
  
  accelerator_for(sc,coarse->oSites(),1,{

      // One thread per sub block
      Coordinate coor_c(_ndimension);
      Lexicographic::CoorFromIndex(coor_c,sc,coarse_rdimensions);  // Block coordinate
+      coarseData_[sc]=Zero();

-      vobj cd = zz;
-      
      for(int sb=0;sb<blockVol;sb++){

 	int sf;
@@ -459,11 +383,9 @@ inline void blockSum(Lattice<vobj> &coarseData,const Lattice<vobj> &fineData)
 	for(int d=0;d<_ndimension;d++) coor_f[d]=coor_c[d]*block_r[d] + coor_b[d];
 	Lexicographic::IndexFromCoor(coor_f,sf,fine_rdimensions);

-	cd=cd+fineData_p[sf];
+	coarseData_[sc]=coarseData_[sc]+fineData_[sf];
      }

-      coarseData_p[sc] = cd;
-
    });
  return;
 }
@@ -677,10 +599,10 @@ void localCopyRegion(const Lattice<vobj> &From,Lattice<vobj> & To,Coordinate Fro
      Integer idx_t = 0; for(int d=0;d<nd;d++) idx_t+=ist[d]*(Tcoor[d]/rdt[d]);
      Integer odx_f = 0; for(int d=0;d<nd;d++) odx_f+=osf[d]*(Fcoor[d]%rdf[d]);
      Integer odx_t = 0; for(int d=0;d<nd;d++) odx_t+=ost[d]*(Tcoor[d]%rdt[d]);
-      vector_type * fp = (vector_type *)&f_v[odx_f];
-      vector_type * tp = (vector_type *)&t_v[odx_t];
+      scalar_type * fp = (scalar_type *)&f_v[odx_f];
+      scalar_type * tp = (scalar_type *)&t_v[odx_t];
      for(int w=0;w<words;w++){
-	tp[w].putlane(fp[w].getlane(idx_f),idx_t);
+	tp[idx_t+w*Nsimd] = fp[idx_f+w*Nsimd];  // FIXME IF RRII layout, type pun no worke
      }
    }
  });
@@ -855,7 +777,7 @@ void ExtractSliceLocal(Lattice<vobj> &lowDim,const Lattice<vobj> & higherDim,int


 template<class vobj>
-void Replicate(const Lattice<vobj> &coarse,Lattice<vobj> & fine)
+void Replicate(Lattice<vobj> &coarse,Lattice<vobj> & fine)
 {
  typedef typename vobj::scalar_object sobj;

@@ -1080,23 +1002,6 @@ vectorizeFromRevLexOrdArray( std::vector<sobj> &in, Lattice<vobj> &out)
  });
 }

-template<class VobjOut, class VobjIn>
-void precisionChangeFast(Lattice<VobjOut> &out, const Lattice<VobjIn> &in)
-{
-  typedef typename VobjOut::vector_type Vout;
-  typedef typename VobjIn::vector_type Vin;
-  const int N = sizeof(VobjOut)/sizeof(Vout);
-  conformable(out.Grid(),in.Grid());
-  out.Checkerboard() = in.Checkerboard();
-  int nsimd = out.Grid()->Nsimd();
-  autoView( out_v  , out, AcceleratorWrite);
-  autoView(  in_v ,   in, AcceleratorRead);
-  accelerator_for(idx,out.Grid()->oSites(),1,{
-      Vout *vout = (Vout *)&out_v[idx];
-      Vin  *vin  = (Vin  *)&in_v[idx];
-      precisionChange(vout,vin,N);
-  });
-}
 //Convert a Lattice from one precision to another
 template<class VobjOut, class VobjIn>
 void precisionChange(Lattice<VobjOut> &out, const Lattice<VobjIn> &in)
@@ -1114,7 +1019,7 @@ void precisionChange(Lattice<VobjOut> &out, const Lattice<VobjIn> &in)

  int ndim = out.Grid()->Nd();
  int out_nsimd = out_grid->Nsimd();
-  int in_nsimd = in_grid->Nsimd();
+    
  std::vector<Coordinate > out_icoor(out_nsimd);
      
  for(int lane=0; lane < out_nsimd; lane++){
@@ -65,40 +65,29 @@ GridLogger GridLogSolver (1, "Solver", GridLogColours, "NORMAL");
 GridLogger GridLogError  (1, "Error" , GridLogColours, "RED");
 GridLogger GridLogWarning(1, "Warning", GridLogColours, "YELLOW");
 GridLogger GridLogMessage(1, "Message", GridLogColours, "NORMAL");
-GridLogger GridLogMemory (1, "Memory", GridLogColours, "NORMAL");
-GridLogger GridLogTracing(1, "Tracing", GridLogColours, "NORMAL");
 GridLogger GridLogDebug  (1, "Debug", GridLogColours, "PURPLE");
 GridLogger GridLogPerformance(1, "Performance", GridLogColours, "GREEN");
-GridLogger GridLogDslash     (1, "Dslash", GridLogColours, "BLUE");
 GridLogger GridLogIterative  (1, "Iterative", GridLogColours, "BLUE");
 GridLogger GridLogIntegrator (1, "Integrator", GridLogColours, "BLUE");
-GridLogger GridLogHMC (1, "HMC", GridLogColours, "BLUE");

 void GridLogConfigure(std::vector<std::string> &logstreams) {
-  GridLogError.Active(1);
+  GridLogError.Active(0);
  GridLogWarning.Active(0);
  GridLogMessage.Active(1); // at least the messages should be always on
-  GridLogMemory.Active(0); 
-  GridLogTracing.Active(0); 
  GridLogIterative.Active(0);
  GridLogDebug.Active(0);
  GridLogPerformance.Active(0);
-  GridLogDslash.Active(0);
  GridLogIntegrator.Active(1);
  GridLogColours.Active(0);
-  GridLogHMC.Active(1);

  for (int i = 0; i < logstreams.size(); i++) {
-    if (logstreams[i] == std::string("Tracing"))     GridLogTracing.Active(1);
-    if (logstreams[i] == std::string("Memory"))      GridLogMemory.Active(1);
+    if (logstreams[i] == std::string("Error"))       GridLogError.Active(1);
    if (logstreams[i] == std::string("Warning"))     GridLogWarning.Active(1);
    if (logstreams[i] == std::string("NoMessage"))   GridLogMessage.Active(0);
    if (logstreams[i] == std::string("Iterative"))   GridLogIterative.Active(1);
    if (logstreams[i] == std::string("Debug"))       GridLogDebug.Active(1);
    if (logstreams[i] == std::string("Performance")) GridLogPerformance.Active(1);
-    if (logstreams[i] == std::string("Dslash"))      GridLogDslash.Active(1);
-    if (logstreams[i] == std::string("NoIntegrator"))GridLogIntegrator.Active(0);
-    if (logstreams[i] == std::string("NoHMC"))       GridLogHMC.Active(0);
+    if (logstreams[i] == std::string("Integrator"))  GridLogIntegrator.Active(1);
    if (logstreams[i] == std::string("Colours"))     GridLogColours.Active(1);
  }
 }
@@ -138,8 +138,7 @@ public:
        stream << std::setw(log.topWidth);
      }
      stream << log.topName << log.background()<< " : ";
-      //      stream << log.colour() <<  std::left;
-      stream <<  std::left;
+      stream << log.colour() <<  std::left;
      if (log.chanWidth > 0)
      {
        stream << std::setw(log.chanWidth);
@@ -154,9 +153,9 @@ public:
 	stream << log.evidence()
 	       << now	       << log.background() << " : " ;
      }
-      //      stream << log.colour();
-      stream <<  std::right;
+      stream << log.colour();
      stream.flags(f);
+
      return stream;
    } else { 
      return devnull;
@@ -181,12 +180,8 @@ extern GridLogger GridLogWarning;
 extern GridLogger GridLogMessage;
 extern GridLogger GridLogDebug  ;
 extern GridLogger GridLogPerformance;
-extern GridLogger GridLogDslash;
 extern GridLogger GridLogIterative  ;
 extern GridLogger GridLogIntegrator  ;
-extern GridLogger GridLogHMC;
-extern GridLogger GridLogMemory;
-extern GridLogger GridLogTracing;
 extern Colours    GridLogColours;

 std::string demangle(const char* name) ;
@@ -31,7 +31,6 @@ directory
 #include <fstream>
 #include <iomanip>
 #include <iostream>
-#include <string>
 #include <map>

 #include <pwd.h>
@@ -577,8 +576,6 @@ class ScidacReader : public GridLimeReader {
    std::string rec_name(ILDG_BINARY_DATA);
    while ( limeReaderNextRecord(LimeR) == LIME_SUCCESS ) { 
      if ( !strncmp(limeReaderType(LimeR), rec_name.c_str(),strlen(rec_name.c_str()) )  ) {
-  // in principle should do the line below, but that breaks backard compatibility with old data
-  // skipPastObjectRecord(std::string(GRID_FIELD_NORM));
 	skipPastObjectRecord(std::string(SCIDAC_CHECKSUM));
 	return;
      }
@@ -655,8 +652,7 @@ class IldgWriter : public ScidacWriter {
    // Fill ILDG header data struct
    //////////////////////////////////////////////////////
    ildgFormat ildgfmt ;
-    const std::string stNC = std::to_string( Nc ) ;
-    ildgfmt.field          = std::string("su"+stNC+"gauge");
+    ildgfmt.field     = std::string("su3gauge");

    if ( format == std::string("IEEE32BIG") ) { 
      ildgfmt.precision = 32;
@@ -873,8 +869,7 @@ class IldgReader : public GridLimeReader {
    } else { 

      assert(found_ildgFormat);
-      const std::string stNC = std::to_string( Nc ) ;
-      assert ( ildgFormat_.field == std::string("su"+stNC+"gauge") );
+      assert ( ildgFormat_.field == std::string("su3gauge") );

      ///////////////////////////////////////////////////////////////////////////////////////
      // Populate our Grid metadata as best we can
@@ -882,7 +877,7 @@ class IldgReader : public GridLimeReader {

      std::ostringstream vers; vers << ildgFormat_.version;
      FieldMetaData_.hdr_version = vers.str();
-      FieldMetaData_.data_type = std::string("4D_SU"+stNC+"_GAUGE_"+stNC+"x"+stNC);
+      FieldMetaData_.data_type = std::string("4D_SU3_GAUGE_3X3");

      FieldMetaData_.nd=4;
      FieldMetaData_.dimension.resize(4);
@@ -6,8 +6,8 @@

    Copyright (C) 2015

+
    Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-    Author: Jamie Hudspith <renwick.james.hudspth@gmail.com>

    This program is free software; you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
@@ -182,8 +182,8 @@ class GaugeStatistics
 public:
  void operator()(Lattice<vLorentzColourMatrixD> & data,FieldMetaData &header)
  {
-    header.link_trace = WilsonLoops<Impl>::linkTrace(data);
-    header.plaquette  = WilsonLoops<Impl>::avgPlaquette(data);
+    header.link_trace=WilsonLoops<Impl>::linkTrace(data);
+    header.plaquette =WilsonLoops<Impl>::avgPlaquette(data);
  }
 };
 typedef GaugeStatistics<PeriodicGimplD> PeriodicGaugeStatistics;
@@ -203,24 +203,20 @@ template<> inline void PrepareMetaData<vLorentzColourMatrixD>(Lattice<vLorentzCo
 //////////////////////////////////////////////////////////////////////
 inline void reconstruct3(LorentzColourMatrix & cm)
 {
-  assert( Nc < 4 && Nc > 1 ) ;
+  const int x=0;
+  const int y=1;
+  const int z=2;
  for(int mu=0;mu<Nd;mu++){
-    #if Nc == 2
-      cm(mu)()(1,0) = -adj(cm(mu)()(0,y)) ;
-      cm(mu)()(1,1) =  adj(cm(mu)()(0,x)) ;
-    #else
-      const int x=0 , y=1 , z=2 ; // a little disinenuous labelling
-      cm(mu)()(2,x) = adj(cm(mu)()(0,y)*cm(mu)()(1,z)-cm(mu)()(0,z)*cm(mu)()(1,y)); //x= yz-zy
-      cm(mu)()(2,y) = adj(cm(mu)()(0,z)*cm(mu)()(1,x)-cm(mu)()(0,x)*cm(mu)()(1,z)); //y= zx-xz
-      cm(mu)()(2,z) = adj(cm(mu)()(0,x)*cm(mu)()(1,y)-cm(mu)()(0,y)*cm(mu)()(1,x)); //z= xy-yx
-    #endif
+    cm(mu)()(2,x) = adj(cm(mu)()(0,y)*cm(mu)()(1,z)-cm(mu)()(0,z)*cm(mu)()(1,y)); //x= yz-zy
+    cm(mu)()(2,y) = adj(cm(mu)()(0,z)*cm(mu)()(1,x)-cm(mu)()(0,x)*cm(mu)()(1,z)); //y= zx-xz
+    cm(mu)()(2,z) = adj(cm(mu)()(0,x)*cm(mu)()(1,y)-cm(mu)()(0,y)*cm(mu)()(1,x)); //z= xy-yx
  }
 }

 ////////////////////////////////////////////////////////////////////////////////
 // Some data types for intermediate storage
 ////////////////////////////////////////////////////////////////////////////////
-template<typename vtype> using iLorentzColour2x3 = iVector<iVector<iVector<vtype, Nc>, Nc-1>, Nd >;
+template<typename vtype> using iLorentzColour2x3 = iVector<iVector<iVector<vtype, Nc>, 2>, Nd >;

 typedef iLorentzColour2x3<Complex>  LorentzColour2x3;
 typedef iLorentzColour2x3<ComplexF> LorentzColour2x3F;
@@ -282,6 +278,7 @@ struct GaugeSimpleMunger{

 template <class fobj, class sobj>
 struct GaugeSimpleUnmunger {
+
  void operator()(sobj &in, fobj &out) {
    for (int mu = 0; mu < Nd; mu++) {
      for (int i = 0; i < Nc; i++) {
@@ -320,8 +317,8 @@ template<class fobj,class sobj>
 struct Gauge3x2munger{
  void operator() (fobj &in,sobj &out){
    for(int mu=0;mu<Nd;mu++){
-      for(int i=0;i<Nc-1;i++){
-	for(int j=0;j<Nc;j++){
+      for(int i=0;i<2;i++){
+	for(int j=0;j<3;j++){
 	  out(mu)()(i,j) = in(mu)(i)(j);
 	}}
    }
@@ -333,8 +330,8 @@ template<class fobj,class sobj>
 struct Gauge3x2unmunger{
  void operator() (sobj &in,fobj &out){
    for(int mu=0;mu<Nd;mu++){
-      for(int i=0;i<Nc-1;i++){
-	for(int j=0;j<Nc;j++){
+      for(int i=0;i<2;i++){
+	for(int j=0;j<3;j++){
 	  out(mu)(i)(j) = in(mu)()(i,j);
 	}}
    }
@@ -9,7 +9,6 @@
    Author: Matt Spraggs <matthew.spraggs@gmail.com>
    Author: Peter Boyle <paboyle@ph.ed.ac.uk>
    Author: paboyle <paboyle@ph.ed.ac.uk>
-    Author: Jamie Hudspith <renwick.james.hudspth@gmail.com>

    This program is free software; you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
@@ -31,8 +30,6 @@
 #ifndef GRID_NERSC_IO_H
 #define GRID_NERSC_IO_H

-#include <string>
-
 NAMESPACE_BEGIN(Grid);

 using namespace Grid;
@@ -42,10 +39,8 @@ using namespace Grid;
 ////////////////////////////////////////////////////////////////////////////////
 class NerscIO : public BinaryIO { 
 public:
-  typedef Lattice<vLorentzColourMatrixD> GaugeField;

-  // Enable/disable exiting if the plaquette in the header does not match the value computed (default true)
-  static bool & exitOnReadPlaquetteMismatch(){ static bool v=true; return v; }
+  typedef Lattice<vLorentzColourMatrixD> GaugeField;

  static inline void truncate(std::string file){
    std::ofstream fout(file,std::ios::out);
@@ -150,17 +145,15 @@ public:

    std::string format(header.floating_point);

-    const int ieee32big = (format == std::string("IEEE32BIG"));
-    const int ieee32    = (format == std::string("IEEE32"));
-    const int ieee64big = (format == std::string("IEEE64BIG"));
-    const int ieee64    = (format == std::string("IEEE64") || \
-			   format == std::string("IEEE64LITTLE"));
+    int ieee32big = (format == std::string("IEEE32BIG"));
+    int ieee32    = (format == std::string("IEEE32"));
+    int ieee64big = (format == std::string("IEEE64BIG"));
+    int ieee64    = (format == std::string("IEEE64") || format == std::string("IEEE64LITTLE"));

    uint32_t nersc_csum,scidac_csuma,scidac_csumb;
    // depending on datatype, set up munger;
    // munger is a function of <floating point, Real, data_type>
-    const std::string stNC = std::to_string( Nc ) ;
-    if ( header.data_type == std::string("4D_SU"+stNC+"_GAUGE") ) {
+    if ( header.data_type == std::string("4D_SU3_GAUGE") ) {
      if ( ieee32 || ieee32big ) {
 	BinaryIO::readLatticeObject<vLorentzColourMatrixD, LorentzColour2x3F> 
 	  (Umu,file,Gauge3x2munger<LorentzColour2x3F,LorentzColourMatrix>(), offset,format,
@@ -171,7 +164,7 @@ public:
 	  (Umu,file,Gauge3x2munger<LorentzColour2x3D,LorentzColourMatrix>(),offset,format,
 	   nersc_csum,scidac_csuma,scidac_csumb);
      }
-    } else if ( header.data_type == std::string("4D_SU"+stNC+"_GAUGE_"+stNC+"x"+stNC) ) {
+    } else if ( header.data_type == std::string("4D_SU3_GAUGE_3x3") ) {
      if ( ieee32 || ieee32big ) {
 	BinaryIO::readLatticeObject<vLorentzColourMatrixD,LorentzColourMatrixF>
 	  (Umu,file,GaugeSimpleMunger<LorentzColourMatrixF,LorentzColourMatrix>(),offset,format,
@@ -205,7 +198,7 @@ public:
      std::cerr << " nersc_csum  " <<std::hex<< nersc_csum << " " << header.checksum<< std::dec<< std::endl;
      exit(0);
    }
-    if(exitOnReadPlaquetteMismatch()) assert(fabs(clone.plaquette -header.plaquette ) < 1.0e-5 );
+    assert(fabs(clone.plaquette -header.plaquette ) < 1.0e-1 );
    assert(fabs(clone.link_trace-header.link_trace) < 1.0e-6 );
    assert(nersc_csum == header.checksum );
      
@@ -216,29 +209,27 @@ public:
  template<class GaugeStats=PeriodicGaugeStatistics>
  static inline void writeConfiguration(Lattice<vLorentzColourMatrixD > &Umu,
 					std::string file, 
-					std::string ens_label = std::string("DWF"),
-					std::string ens_id = std::string("UKQCD"),
-					unsigned int sequence_number = 1)
+					std::string ens_label = std::string("DWF"))
  {
-    writeConfiguration(Umu,file,0,1,ens_label,ens_id,sequence_number);
+    writeConfiguration(Umu,file,0,1,ens_label);
  }
  template<class GaugeStats=PeriodicGaugeStatistics>
  static inline void writeConfiguration(Lattice<vLorentzColourMatrixD > &Umu,
 					std::string file, 
 					int two_row,
 					int bits32,
-					std::string ens_label = std::string("DWF"),
-					std::string ens_id = std::string("UKQCD"),
-					unsigned int sequence_number = 1)
+					std::string ens_label = std::string("DWF"))
  {
    typedef vLorentzColourMatrixD vobj;
    typedef typename vobj::scalar_object sobj;

    FieldMetaData header;
-    header.sequence_number = sequence_number;
-    header.ensemble_id     = ens_id;
+    ///////////////////////////////////////////
+    // Following should become arguments
+    ///////////////////////////////////////////
+    header.sequence_number = 1;
+    header.ensemble_id     = std::string("UKQCD");
    header.ensemble_label  = ens_label;
-    header.hdr_version     = "1.0" ;

    typedef LorentzColourMatrixD fobj3D;
    typedef LorentzColour2x3D    fobj2D;
@@ -252,14 +243,10 @@ public:

    uint64_t offset;

-    // Sod it -- always write NcxNc double
-    header.floating_point  = std::string("IEEE64BIG");
-    const std::string stNC = std::to_string( Nc ) ;
-    if( two_row ) {
-      header.data_type = std::string("4D_SU" + stNC + "_GAUGE" );
-    } else {
-      header.data_type = std::string("4D_SU" + stNC + "_GAUGE_" + stNC + "x" + stNC );
-    }
+    // Sod it -- always write 3x3 double
+    header.floating_point = std::string("IEEE64BIG");
+    header.data_type      = std::string("4D_SU3_GAUGE_3x3");
+    GaugeSimpleUnmunger<fobj3D,sobj> munge;
    if ( grid->IsBoss() ) { 
      truncate(file);
      offset = writeHeader(header,file);
@@ -267,15 +254,8 @@ public:
    grid->Broadcast(0,(void *)&offset,sizeof(offset));

    uint32_t nersc_csum,scidac_csuma,scidac_csumb;
-    if( two_row ) {
-      Gauge3x2unmunger<fobj2D,sobj> munge;
-      BinaryIO::writeLatticeObject<vobj,fobj2D>(Umu,file,munge,offset,header.floating_point,
-						nersc_csum,scidac_csuma,scidac_csumb);
-    } else {
-      GaugeSimpleUnmunger<fobj3D,sobj> munge;
-      BinaryIO::writeLatticeObject<vobj,fobj3D>(Umu,file,munge,offset,header.floating_point,
-						nersc_csum,scidac_csuma,scidac_csumb);
-    }
+    BinaryIO::writeLatticeObject<vobj,fobj3D>(Umu,file,munge,offset,header.floating_point,
+					      nersc_csum,scidac_csuma,scidac_csumb);
    header.checksum = nersc_csum;
    if ( grid->IsBoss() ) { 
      writeHeader(header,file);
@@ -307,7 +287,8 @@ public:
    header.plaquette=0.0;
    MachineCharacteristics(header);

-    uint64_t offset;
+	uint64_t offset;
+  
 #ifdef RNG_RANLUX
    header.floating_point = std::string("UINT64");
    header.data_type      = std::string("RANLUX48");
@@ -347,7 +328,7 @@ public:

    GridBase *grid = parallel.Grid();

-    uint64_t offset = readHeader(file,grid,header);
+	uint64_t offset = readHeader(file,grid,header);

    FieldMetaData clone(header);

@@ -27,12 +27,9 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 /*  END LEGAL */

 #include <Grid/GridCore.h>
-
-#include <Grid/perfmon/Timer.h>
 #include <Grid/perfmon/PerfCount.h>
-NAMESPACE_BEGIN(Grid);

-GridTimePoint theProgramStart = GridClock::now();
+NAMESPACE_BEGIN(Grid);

 #define CacheControl(L,O,R) ((PERF_COUNT_HW_CACHE_##L)|(PERF_COUNT_HW_CACHE_OP_##O<<8)| (PERF_COUNT_HW_CACHE_RESULT_##R<<16))
 #define RawConfig(A,B) (A<<8|B)
@@ -72,9 +72,17 @@ static long perf_event_open(struct perf_event_attr *hw_event, pid_t pid,
 inline uint64_t cyclecount(void){ 
  return 0;
 }
+#define __SSC_MARK(mark) __asm__ __volatile__ ("movl %0, %%ebx; .byte 0x64, 0x67, 0x90 " ::"i"(mark):"%ebx")
+#define __SSC_STOP  __SSC_MARK(0x110)
+#define __SSC_START __SSC_MARK(0x111)
+

 #else

+#define __SSC_MARK(mark) 
+#define __SSC_STOP  
+#define __SSC_START 
+
 /*
 * cycle counters arch dependent
 */
@@ -35,8 +35,17 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>

 NAMESPACE_BEGIN(Grid)

-//typedef  std::chrono::system_clock          GridClock;
-typedef  std::chrono::high_resolution_clock   GridClock;
+// Dress the output; use std::chrono
+// C++11 time facilities better?
+inline double usecond(void) {
+  struct timeval tv;
+#ifdef TIMERS_ON
+  gettimeofday(&tv,NULL);
+#endif
+  return 1.0*tv.tv_usec + 1.0e6*tv.tv_sec;
+}
+
+typedef  std::chrono::system_clock          GridClock;
 typedef  std::chrono::time_point<GridClock> GridTimePoint;

 typedef  std::chrono::seconds               GridSecs;
@@ -44,15 +53,6 @@ typedef  std::chrono::milliseconds          GridMillisecs;
 typedef  std::chrono::microseconds          GridUsecs;
 typedef  std::chrono::microseconds          GridTime;

-extern GridTimePoint theProgramStart;
-// Dress the output; use std::chrono
-// C++11 time facilities better?
-inline double usecond(void) {
-  auto usecs = std::chrono::duration_cast<GridUsecs>(GridClock::now()-theProgramStart); 
-  return 1.0*usecs.count();
-}
-
-
 inline std::ostream& operator<< (std::ostream & stream, const GridSecs & time)
 {
  stream << time.count()<<" s";
@@ -1,70 +0,0 @@
-#pragma once
-
-NAMESPACE_BEGIN(Grid);
-
-#ifdef GRID_TRACING_NVTX
-#include <nvToolsExt.h>
-class GridTracer {
-public:
-  GridTracer(const char* name) {
-    nvtxRangePushA(name);
-  }
-  ~GridTracer() {
-    nvtxRangePop();
-  }
-};
-inline void tracePush(const char *name) { nvtxRangePushA(name); }
-inline void tracePop(const char *name) { nvtxRangePop(); }
-inline int  traceStart(const char *name) {  }
-inline void traceStop(int ID) {  }
-#endif
-
-#ifdef GRID_TRACING_ROCTX
-#include <roctracer/roctx.h>
-class GridTracer {
- public:
-  GridTracer(const char* name) {
-    roctxRangePushA(name);
-    std::cout << "roctxRangePush "<<name<<std::endl;
-  }
-  ~GridTracer() {
-    roctxRangePop();
-    std::cout << "roctxRangePop "<<std::endl;
-  }
-};
-inline void tracePush(const char *name) { roctxRangePushA(name); }
-inline void tracePop(const char *name) { roctxRangePop(); }
-inline int  traceStart(const char *name) { roctxRangeStart(name); }
-inline void traceStop(int ID) { roctxRangeStop(ID); }
-#endif
-
-#ifdef GRID_TRACING_TIMER
-class GridTracer {
- public:
-  const char *name;
-  double elapsed;
-  GridTracer(const char* _name) {
-    name = _name;
-    elapsed=-usecond();
-  }
-  ~GridTracer() {
-    elapsed+=usecond();
-    std::cout << GridLogTracing << name << " took " <<elapsed<< " us" <<std::endl;
-  }
-};
-inline void tracePush(const char *name) {  }
-inline void tracePop(const char *name) {  }
-inline int  traceStart(const char *name) { return 0; }
-inline void traceStop(int ID) {  }
-#endif
-
-#ifdef GRID_TRACING_NONE
-#define GRID_TRACE(name) 
-inline void tracePush(const char *name) {  }
-inline void tracePop(const char *name) {  }
-inline int  traceStart(const char *name) { return 0;  }
-inline void traceStop(int ID) {  }
-#else
-#define GRID_TRACE(name) GridTracer uniq_name_using_macros##__COUNTER__(name);
-#endif
-NAMESPACE_END(Grid);
@@ -16,12 +16,8 @@

 #ifdef __NVCC__
 #pragma push
-#if (__CUDACC_VER_MAJOR__ >= 11) && (__CUDACC_VER_MINOR__ >= 5)
-#pragma nv_diag_suppress declared_but_not_referenced // suppress "function was declared but never referenced warning"
-#else
 #pragma diag_suppress declared_but_not_referenced // suppress "function was declared but never referenced warning"
 #endif
-#endif

 #include "pugixml.h"

@@ -63,7 +63,6 @@ static constexpr int Ngp=2; // gparity index range
 #define ColourIndex  (2)
 #define SpinIndex    (1)
 #define LorentzIndex (0)
-#define GparityFlavourIndex (0)

 // Also should make these a named enum type
 static constexpr int DaggerNo=0;
@@ -88,8 +87,6 @@ template<typename T> struct isCoarsened {
 template <typename T> using IfCoarsened    = Invoke<std::enable_if< isCoarsened<T>::value,int> > ;
 template <typename T> using IfNotCoarsened = Invoke<std::enable_if<!isCoarsened<T>::value,int> > ;

-const int GparityFlavourTensorIndex = 3; //TensorLevel counts from the bottom!
-
 // ChrisK very keen to add extra space for Gparity doubling.
 //
 // Also add domain wall index, in a way where Wilson operator 
@@ -113,10 +110,8 @@ template<typename vtype> using iHalfSpinColourVector      = iScalar<iVector<iVec
    template<typename vtype> using iSpinColourSpinColourMatrix  = iScalar<iMatrix<iMatrix<iMatrix<iMatrix<vtype, Nc>, Ns>, Nc>, Ns> >;


-template<typename vtype> using iGparityFlavourVector                = iVector<iScalar<iScalar<vtype> >, Ngp>;
 template<typename vtype> using iGparitySpinColourVector       = iVector<iVector<iVector<vtype, Nc>, Ns>, Ngp >;
 template<typename vtype> using iGparityHalfSpinColourVector   = iVector<iVector<iVector<vtype, Nc>, Nhs>, Ngp >;
-template<typename vtype> using iGparityFlavourMatrix = iMatrix<iScalar<iScalar<vtype> >, Ngp>;

 // Spin matrix
 typedef iSpinMatrix<Complex  >          SpinMatrix;
@@ -126,7 +121,6 @@ typedef iSpinMatrix<ComplexD >          SpinMatrixD;
 typedef iSpinMatrix<vComplex >          vSpinMatrix;
 typedef iSpinMatrix<vComplexF>          vSpinMatrixF;
 typedef iSpinMatrix<vComplexD>          vSpinMatrixD;
-typedef iSpinMatrix<vComplexD2>         vSpinMatrixD2;

 // Colour Matrix
 typedef iColourMatrix<Complex  >        ColourMatrix;
@@ -136,7 +130,6 @@ typedef iColourMatrix<ComplexD >        ColourMatrixD;
 typedef iColourMatrix<vComplex >        vColourMatrix;
 typedef iColourMatrix<vComplexF>        vColourMatrixF;
 typedef iColourMatrix<vComplexD>        vColourMatrixD;
-typedef iColourMatrix<vComplexD2>       vColourMatrixD2;

 // SpinColour matrix
 typedef iSpinColourMatrix<Complex  >    SpinColourMatrix;
@@ -146,7 +139,6 @@ typedef iSpinColourMatrix<ComplexD >    SpinColourMatrixD;
 typedef iSpinColourMatrix<vComplex >    vSpinColourMatrix;
 typedef iSpinColourMatrix<vComplexF>    vSpinColourMatrixF;
 typedef iSpinColourMatrix<vComplexD>    vSpinColourMatrixD;
-typedef iSpinColourMatrix<vComplexD2>   vSpinColourMatrixD2;

 // SpinColourSpinColour matrix
 typedef iSpinColourSpinColourMatrix<Complex  >    SpinColourSpinColourMatrix;
@@ -156,7 +148,6 @@ typedef iSpinColourSpinColourMatrix<ComplexD >    SpinColourSpinColourMatrixD;
 typedef iSpinColourSpinColourMatrix<vComplex >    vSpinColourSpinColourMatrix;
 typedef iSpinColourSpinColourMatrix<vComplexF>    vSpinColourSpinColourMatrixF;
 typedef iSpinColourSpinColourMatrix<vComplexD>    vSpinColourSpinColourMatrixD;
-typedef iSpinColourSpinColourMatrix<vComplexD2>   vSpinColourSpinColourMatrixD2;

 // SpinColourSpinColour matrix
 typedef iSpinColourSpinColourMatrix<Complex  >    SpinColourSpinColourMatrix;
@@ -166,38 +157,24 @@ typedef iSpinColourSpinColourMatrix<ComplexD >    SpinColourSpinColourMatrixD;
 typedef iSpinColourSpinColourMatrix<vComplex >    vSpinColourSpinColourMatrix;
 typedef iSpinColourSpinColourMatrix<vComplexF>    vSpinColourSpinColourMatrixF;
 typedef iSpinColourSpinColourMatrix<vComplexD>    vSpinColourSpinColourMatrixD;
-typedef iSpinColourSpinColourMatrix<vComplexD2>   vSpinColourSpinColourMatrixD2;

 // LorentzColour
 typedef iLorentzColourMatrix<Complex  > LorentzColourMatrix;
 typedef iLorentzColourMatrix<ComplexF > LorentzColourMatrixF;
 typedef iLorentzColourMatrix<ComplexD > LorentzColourMatrixD;

-typedef iLorentzColourMatrix<vComplex >  vLorentzColourMatrix;
-typedef iLorentzColourMatrix<vComplexF>  vLorentzColourMatrixF;
-typedef iLorentzColourMatrix<vComplexD>  vLorentzColourMatrixD;
-typedef iLorentzColourMatrix<vComplexD2> vLorentzColourMatrixD2;
+typedef iLorentzColourMatrix<vComplex > vLorentzColourMatrix;
+typedef iLorentzColourMatrix<vComplexF> vLorentzColourMatrixF;
+typedef iLorentzColourMatrix<vComplexD> vLorentzColourMatrixD;

 // DoubleStored gauge field
 typedef iDoubleStoredColourMatrix<Complex  > DoubleStoredColourMatrix;
 typedef iDoubleStoredColourMatrix<ComplexF > DoubleStoredColourMatrixF;
 typedef iDoubleStoredColourMatrix<ComplexD > DoubleStoredColourMatrixD;

-typedef iDoubleStoredColourMatrix<vComplex >  vDoubleStoredColourMatrix;
-typedef iDoubleStoredColourMatrix<vComplexF>  vDoubleStoredColourMatrixF;
-typedef iDoubleStoredColourMatrix<vComplexD>  vDoubleStoredColourMatrixD;
-typedef iDoubleStoredColourMatrix<vComplexD2> vDoubleStoredColourMatrixD2;
-
-//G-parity flavour matrix
-typedef iGparityFlavourMatrix<Complex> GparityFlavourMatrix;
-typedef iGparityFlavourMatrix<ComplexF> GparityFlavourMatrixF;
-typedef iGparityFlavourMatrix<ComplexD> GparityFlavourMatrixD;
-
-typedef iGparityFlavourMatrix<vComplex>   vGparityFlavourMatrix;
-typedef iGparityFlavourMatrix<vComplexF>  vGparityFlavourMatrixF;
-typedef iGparityFlavourMatrix<vComplexD>  vGparityFlavourMatrixD;
-typedef iGparityFlavourMatrix<vComplexD2> vGparityFlavourMatrixD2;
-
+typedef iDoubleStoredColourMatrix<vComplex > vDoubleStoredColourMatrix;
+typedef iDoubleStoredColourMatrix<vComplexF> vDoubleStoredColourMatrixF;
+typedef iDoubleStoredColourMatrix<vComplexD> vDoubleStoredColourMatrixD;

 // Spin vector
 typedef iSpinVector<Complex >           SpinVector;
@@ -207,7 +184,6 @@ typedef iSpinVector<ComplexD>           SpinVectorD;
 typedef iSpinVector<vComplex >           vSpinVector;
 typedef iSpinVector<vComplexF>           vSpinVectorF;
 typedef iSpinVector<vComplexD>           vSpinVectorD;
-typedef iSpinVector<vComplexD2>          vSpinVectorD2;

 // Colour vector
 typedef iColourVector<Complex >         ColourVector;
@@ -217,7 +193,6 @@ typedef iColourVector<ComplexD>         ColourVectorD;
 typedef iColourVector<vComplex >         vColourVector;
 typedef iColourVector<vComplexF>         vColourVectorF;
 typedef iColourVector<vComplexD>         vColourVectorD;
-typedef iColourVector<vComplexD2>        vColourVectorD2;

 // SpinColourVector
 typedef iSpinColourVector<Complex >     SpinColourVector;
@@ -227,7 +202,6 @@ typedef iSpinColourVector<ComplexD>     SpinColourVectorD;
 typedef iSpinColourVector<vComplex >     vSpinColourVector;
 typedef iSpinColourVector<vComplexF>     vSpinColourVectorF;
 typedef iSpinColourVector<vComplexD>     vSpinColourVectorD;
-typedef iSpinColourVector<vComplexD2>    vSpinColourVectorD2;

 // HalfSpin vector
 typedef iHalfSpinVector<Complex >       HalfSpinVector;
@@ -237,27 +211,15 @@ typedef iHalfSpinVector<ComplexD>       HalfSpinVectorD;
 typedef iHalfSpinVector<vComplex >       vHalfSpinVector;
 typedef iHalfSpinVector<vComplexF>       vHalfSpinVectorF;
 typedef iHalfSpinVector<vComplexD>       vHalfSpinVectorD;
-typedef iHalfSpinVector<vComplexD2>      vHalfSpinVectorD2;

 // HalfSpinColour vector
 typedef iHalfSpinColourVector<Complex > HalfSpinColourVector;
 typedef iHalfSpinColourVector<ComplexF> HalfSpinColourVectorF;
 typedef iHalfSpinColourVector<ComplexD> HalfSpinColourVectorD;
    
-typedef iHalfSpinColourVector<vComplex >  vHalfSpinColourVector;
-typedef iHalfSpinColourVector<vComplexF>  vHalfSpinColourVectorF;
-typedef iHalfSpinColourVector<vComplexD>  vHalfSpinColourVectorD;
-typedef iHalfSpinColourVector<vComplexD2> vHalfSpinColourVectorD2;
-
-//G-parity flavour vector
-typedef iGparityFlavourVector<Complex >         GparityFlavourVector;
-typedef iGparityFlavourVector<ComplexF>         GparityFlavourVectorF;
-typedef iGparityFlavourVector<ComplexD>         GparityFlavourVectorD;
-
-typedef iGparityFlavourVector<vComplex >         vGparityFlavourVector;
-typedef iGparityFlavourVector<vComplexF>         vGparityFlavourVectorF;
-typedef iGparityFlavourVector<vComplexD>         vGparityFlavourVectorD;
-typedef iGparityFlavourVector<vComplexD2>        vGparityFlavourVectorD2;
+typedef iHalfSpinColourVector<vComplex > vHalfSpinColourVector;
+typedef iHalfSpinColourVector<vComplexF> vHalfSpinColourVectorF;
+typedef iHalfSpinColourVector<vComplexD> vHalfSpinColourVectorD;
    
 // singlets
 typedef iSinglet<Complex >         TComplex;     // FIXME This is painful. Tensor singlet complex type.
@@ -267,7 +229,6 @@ typedef iSinglet<ComplexD>         TComplexD;    // FIXME This is painful. Tenso
 typedef iSinglet<vComplex >        vTComplex ;   // what if we don't know the tensor structure
 typedef iSinglet<vComplexF>        vTComplexF;   // what if we don't know the tensor structure
 typedef iSinglet<vComplexD>        vTComplexD;   // what if we don't know the tensor structure
-typedef iSinglet<vComplexD2>       vTComplexD2;   // what if we don't know the tensor structure

 typedef iSinglet<Real >            TReal;        // Shouldn't need these; can I make it work without?
 typedef iSinglet<RealF>            TRealF;       // Shouldn't need these; can I make it work without?
@@ -285,58 +246,47 @@ typedef iSinglet<Integer >         TInteger;
 typedef Lattice<vColourMatrix>          LatticeColourMatrix;
 typedef Lattice<vColourMatrixF>         LatticeColourMatrixF;
 typedef Lattice<vColourMatrixD>         LatticeColourMatrixD;
-typedef Lattice<vColourMatrixD2>        LatticeColourMatrixD2;

 typedef Lattice<vSpinMatrix>            LatticeSpinMatrix;
 typedef Lattice<vSpinMatrixF>           LatticeSpinMatrixF;
 typedef Lattice<vSpinMatrixD>           LatticeSpinMatrixD;
-typedef Lattice<vSpinMatrixD2>          LatticeSpinMatrixD2;

 typedef Lattice<vSpinColourMatrix>      LatticeSpinColourMatrix;
 typedef Lattice<vSpinColourMatrixF>     LatticeSpinColourMatrixF;
 typedef Lattice<vSpinColourMatrixD>     LatticeSpinColourMatrixD;
-typedef Lattice<vSpinColourMatrixD2>    LatticeSpinColourMatrixD2;

 typedef Lattice<vSpinColourSpinColourMatrix>      LatticeSpinColourSpinColourMatrix;
 typedef Lattice<vSpinColourSpinColourMatrixF>     LatticeSpinColourSpinColourMatrixF;
 typedef Lattice<vSpinColourSpinColourMatrixD>     LatticeSpinColourSpinColourMatrixD;
-typedef Lattice<vSpinColourSpinColourMatrixD2>    LatticeSpinColourSpinColourMatrixD2;

-typedef Lattice<vLorentzColourMatrix>   LatticeLorentzColourMatrix;
-typedef Lattice<vLorentzColourMatrixF>  LatticeLorentzColourMatrixF;
-typedef Lattice<vLorentzColourMatrixD>  LatticeLorentzColourMatrixD;
-typedef Lattice<vLorentzColourMatrixD2> LatticeLorentzColourMatrixD2;
+typedef Lattice<vLorentzColourMatrix>  LatticeLorentzColourMatrix;
+typedef Lattice<vLorentzColourMatrixF> LatticeLorentzColourMatrixF;
+typedef Lattice<vLorentzColourMatrixD> LatticeLorentzColourMatrixD;

 // DoubleStored gauge field
-typedef Lattice<vDoubleStoredColourMatrix>   LatticeDoubleStoredColourMatrix;
-typedef Lattice<vDoubleStoredColourMatrixF>  LatticeDoubleStoredColourMatrixF;
-typedef Lattice<vDoubleStoredColourMatrixD>  LatticeDoubleStoredColourMatrixD;
-typedef Lattice<vDoubleStoredColourMatrixD2> LatticeDoubleStoredColourMatrixD2;
+typedef Lattice<vDoubleStoredColourMatrix>  LatticeDoubleStoredColourMatrix;
+typedef Lattice<vDoubleStoredColourMatrixF> LatticeDoubleStoredColourMatrixF;
+typedef Lattice<vDoubleStoredColourMatrixD> LatticeDoubleStoredColourMatrixD;

 typedef Lattice<vSpinVector>            LatticeSpinVector;
 typedef Lattice<vSpinVectorF>           LatticeSpinVectorF;
 typedef Lattice<vSpinVectorD>           LatticeSpinVectorD;
-typedef Lattice<vSpinVectorD2>          LatticeSpinVectorD2;

 typedef Lattice<vColourVector>          LatticeColourVector;
 typedef Lattice<vColourVectorF>         LatticeColourVectorF;
 typedef Lattice<vColourVectorD>         LatticeColourVectorD;
-typedef Lattice<vColourVectorD2>        LatticeColourVectorD2;

 typedef Lattice<vSpinColourVector>      LatticeSpinColourVector;
 typedef Lattice<vSpinColourVectorF>     LatticeSpinColourVectorF;
 typedef Lattice<vSpinColourVectorD>     LatticeSpinColourVectorD;
-typedef Lattice<vSpinColourVectorD2>    LatticeSpinColourVectorD2;

 typedef Lattice<vHalfSpinVector>        LatticeHalfSpinVector;
 typedef Lattice<vHalfSpinVectorF>       LatticeHalfSpinVectorF;
 typedef Lattice<vHalfSpinVectorD>       LatticeHalfSpinVectorD;
-typedef Lattice<vHalfSpinVectorD2>      LatticeHalfSpinVectorD2;

-typedef Lattice<vHalfSpinColourVector>   LatticeHalfSpinColourVector;
-typedef Lattice<vHalfSpinColourVectorF>  LatticeHalfSpinColourVectorF;
-typedef Lattice<vHalfSpinColourVectorD>  LatticeHalfSpinColourVectorD;
-typedef Lattice<vHalfSpinColourVectorD2> LatticeHalfSpinColourVectorD2;
+typedef Lattice<vHalfSpinColourVector>  LatticeHalfSpinColourVector;
+typedef Lattice<vHalfSpinColourVectorF> LatticeHalfSpinColourVectorF;
+typedef Lattice<vHalfSpinColourVectorD> LatticeHalfSpinColourVectorD;

 typedef Lattice<vTReal>            LatticeReal;
 typedef Lattice<vTRealF>           LatticeRealF;
@@ -345,7 +295,6 @@ typedef Lattice<vTRealD>           LatticeRealD;
 typedef Lattice<vTComplex>         LatticeComplex;
 typedef Lattice<vTComplexF>        LatticeComplexF;
 typedef Lattice<vTComplexD>        LatticeComplexD;
-typedef Lattice<vTComplexD2>       LatticeComplexD2;

 typedef Lattice<vTInteger>         LatticeInteger; // Predicates for "where"

@@ -353,42 +302,37 @@ typedef Lattice<vTInteger>         LatticeInteger; // Predicates for "where"
 ///////////////////////////////////////////
 // Physical names for things
 ///////////////////////////////////////////
-typedef LatticeHalfSpinColourVector   LatticeHalfFermion;
-typedef LatticeHalfSpinColourVectorF  LatticeHalfFermionF;
-typedef LatticeHalfSpinColourVectorD  LatticeHalfFermionD;
-typedef LatticeHalfSpinColourVectorD2 LatticeHalfFermionD2;
+typedef LatticeHalfSpinColourVector  LatticeHalfFermion;
+typedef LatticeHalfSpinColourVectorF LatticeHalfFermionF;
+typedef LatticeHalfSpinColourVectorF LatticeHalfFermionD;

 typedef LatticeSpinColourVector      LatticeFermion;
 typedef LatticeSpinColourVectorF     LatticeFermionF;
 typedef LatticeSpinColourVectorD     LatticeFermionD;
-typedef LatticeSpinColourVectorD2    LatticeFermionD2;

 typedef LatticeSpinColourMatrix                LatticePropagator;
 typedef LatticeSpinColourMatrixF               LatticePropagatorF;
 typedef LatticeSpinColourMatrixD               LatticePropagatorD;
-typedef LatticeSpinColourMatrixD2              LatticePropagatorD2;

 typedef LatticeLorentzColourMatrix             LatticeGaugeField;
 typedef LatticeLorentzColourMatrixF            LatticeGaugeFieldF;
 typedef LatticeLorentzColourMatrixD            LatticeGaugeFieldD;
-typedef LatticeLorentzColourMatrixD2           LatticeGaugeFieldD2;

 typedef LatticeDoubleStoredColourMatrix        LatticeDoubledGaugeField;
 typedef LatticeDoubleStoredColourMatrixF       LatticeDoubledGaugeFieldF;
 typedef LatticeDoubleStoredColourMatrixD       LatticeDoubledGaugeFieldD;
-typedef LatticeDoubleStoredColourMatrixD2      LatticeDoubledGaugeFieldD2;

 template<class GF> using LorentzScalar = Lattice<iScalar<typename GF::vector_object::element> >;

+// Uhgg... typing this hurt  ;)
+// (my keyboard got burning hot when I typed this, must be the anti-Fermion)
 typedef Lattice<vColourVector>          LatticeStaggeredFermion;    
 typedef Lattice<vColourVectorF>         LatticeStaggeredFermionF;    
 typedef Lattice<vColourVectorD>         LatticeStaggeredFermionD;    
-typedef Lattice<vColourVectorD2>        LatticeStaggeredFermionD2;    

 typedef Lattice<vColourMatrix>          LatticeStaggeredPropagator; 
 typedef Lattice<vColourMatrixF>         LatticeStaggeredPropagatorF; 
 typedef Lattice<vColourMatrixD>         LatticeStaggeredPropagatorD; 
-typedef Lattice<vColourMatrixD2>        LatticeStaggeredPropagatorD2; 

 //////////////////////////////////////////////////////////////////////////////
 // Peek and Poke named after physics attributes
@@ -40,47 +40,9 @@ class Action

 public:
  bool is_smeared = false;
-  RealD deriv_norm_sum;
-  RealD deriv_max_sum;
-  RealD Fdt_norm_sum;
-  RealD Fdt_max_sum;
-  int   deriv_num;
-  RealD deriv_us;
-  RealD S_us;
-  RealD refresh_us;
-  void  reset_timer(void)        {
-    deriv_us = S_us = refresh_us = 0.0;
-    deriv_norm_sum = deriv_max_sum=0.0;
-    Fdt_max_sum =  Fdt_norm_sum = 0.0;
-    deriv_num=0;
-  }
-  void  deriv_log(RealD nrm, RealD max,RealD Fdt_nrm,RealD Fdt_max) {
-    if ( max > deriv_max_sum ) {
-      deriv_max_sum=max;
-    }
-    deriv_norm_sum+=nrm;
-    if ( Fdt_max > Fdt_max_sum ) {
-      Fdt_max_sum=Fdt_max;
-    }
-    Fdt_norm_sum+=Fdt_nrm; deriv_num++;
-  }
-  RealD deriv_max_average(void)       { return deriv_max_sum; };
-  RealD deriv_norm_average(void)      { return deriv_norm_sum/deriv_num; };
-  RealD Fdt_max_average(void)         { return Fdt_max_sum; };
-  RealD Fdt_norm_average(void)        { return Fdt_norm_sum/deriv_num; };
-  RealD deriv_timer(void)        { return deriv_us; };
-  RealD S_timer(void)            { return S_us; };
-  RealD refresh_timer(void)      { return refresh_us; };
-  void deriv_timer_start(void)   { deriv_us-=usecond(); }
-  void deriv_timer_stop(void)    { deriv_us+=usecond(); }
-  void refresh_timer_start(void) { refresh_us-=usecond(); }
-  void refresh_timer_stop(void)  { refresh_us+=usecond(); }
-  void S_timer_start(void)       { S_us-=usecond(); }
-  void S_timer_stop(void)        { S_us+=usecond(); }
  // Heatbath?
  virtual void refresh(const GaugeField& U, GridSerialRNG &sRNG, GridParallelRNG& pRNG) = 0; // refresh pseudofermions
  virtual RealD S(const GaugeField& U) = 0;                             // evaluate the action
-  virtual RealD Sinitial(const GaugeField& U) { return this->S(U); } ;  // if the refresh computes the action, can cache it. Alternately refreshAndAction() ?
  virtual void deriv(const GaugeField& U, GaugeField& dSdU) = 0;        // evaluate the action derivative
  virtual std::string action_name()    = 0;                             // return the action name
  virtual std::string LogParameters()  = 0;                             // prints action parameters
@@ -37,10 +37,6 @@ NAMESPACE_CHECK(ActionSet);
 #include <Grid/qcd/action/ActionParams.h>
 NAMESPACE_CHECK(ActionParams);

-#include <Grid/qcd/action/filters/MomentumFilter.h>
-#include <Grid/qcd/action/filters/DirichletFilter.h>
-#include <Grid/qcd/action/filters/DDHMCFilter.h>
-
 ////////////////////////////////////////////
 // Gauge Actions
 ////////////////////////////////////////////
@@ -34,45 +34,27 @@ directory

 NAMESPACE_BEGIN(Grid);

-
+// These can move into a params header and be given MacroMagic serialisation
 struct GparityWilsonImplParams {
  Coordinate twists;
-                     //mu=Nd-1 is assumed to be the time direction and a twist value of 1 indicates antiperiodic BCs
-  Coordinate dirichlet; // Blocksize of dirichlet BCs
-  int  partialDirichlet;
-  GparityWilsonImplParams() : twists(Nd, 0) {
-    dirichlet.resize(0);
-    partialDirichlet=0;
-  };
+  GparityWilsonImplParams() : twists(Nd, 0) {};
 };
  
 struct WilsonImplParams {
  bool overlapCommsCompute;
-  Coordinate dirichlet; // Blocksize of dirichlet BCs
-  int  partialDirichlet;
  AcceleratorVector<Real,Nd> twist_n_2pi_L;
  AcceleratorVector<Complex,Nd> boundary_phases;
  WilsonImplParams()  {
-    dirichlet.resize(0);
-    partialDirichlet=0;
    boundary_phases.resize(Nd, 1.0);
      twist_n_2pi_L.resize(Nd, 0.0);
  };
  WilsonImplParams(const AcceleratorVector<Complex,Nd> phi) : boundary_phases(phi), overlapCommsCompute(false) {
    twist_n_2pi_L.resize(Nd, 0.0);
-    partialDirichlet=0;
-    dirichlet.resize(0);
  }
 };

 struct StaggeredImplParams {
-  Coordinate dirichlet; // Blocksize of dirichlet BCs
-  int  partialDirichlet;
-  StaggeredImplParams()
-  {
-    partialDirichlet=0;
-    dirichlet.resize(0);
-  };
+  StaggeredImplParams()  {};
 };
  
  struct OneFlavourRationalParams : Serializable {
@@ -81,11 +63,9 @@ struct StaggeredImplParams {
 				    RealD, hi, 
 				    int,   MaxIter, 
 				    RealD, tolerance, 
-				    RealD, mdtolerance, 
 				    int,   degree, 
 				    int,   precision,
-				    int,   BoundsCheckFreq,
-				    RealD, BoundsCheckTol);
+				    int,   BoundsCheckFreq);
    
  // MaxIter and tolerance, vectors??
    
@@ -96,62 +76,16 @@ struct StaggeredImplParams {
 				RealD tol      = 1.0e-8, 
                           	int _degree    = 10,
 				int _precision = 64,
-				int _BoundsCheckFreq=20,
-				RealD mdtol    = 1.0e-6,
-				double _BoundsCheckTol=1e-6)
+				int _BoundsCheckFreq=20)
      : lo(_lo),
 	hi(_hi),
 	MaxIter(_maxit),
 	tolerance(tol),
-        mdtolerance(mdtol),
 	degree(_degree),
        precision(_precision),
-        BoundsCheckFreq(_BoundsCheckFreq),
-        BoundsCheckTol(_BoundsCheckTol){};
+        BoundsCheckFreq(_BoundsCheckFreq){};
  };
  
-  /*Action parameters for the generalized rational action
-    The approximation is for (M^dag M)^{1/inv_pow}
-    where inv_pow is the denominator of the fractional power.
-    Default inv_pow=2 for square root, making this equivalent to 
-    the OneFlavourRational action
-  */
-    struct RationalActionParams : Serializable {
-    GRID_SERIALIZABLE_CLASS_MEMBERS(RationalActionParams, 
-				    int, inv_pow, 
-				    RealD, lo, //low eigenvalue bound of rational approx
-				    RealD, hi, //high eigenvalue bound of rational approx
-				    int,   MaxIter,  //maximum iterations in msCG
-				    RealD, action_tolerance,  //msCG tolerance in action evaluation
-				    int,   action_degree, //rational approx tolerance in action evaluation
-				    RealD, md_tolerance,  //msCG tolerance in MD integration
-				    int,   md_degree, //rational approx tolerance in MD integration
-				    int,   precision, //precision of floating point arithmetic
-				    int,   BoundsCheckFreq); //frequency the approximation is tested (with Metropolis degree/tolerance); 0 disables the check
-  // constructor 
-  RationalActionParams(int _inv_pow = 2,
-		       RealD _lo      = 0.0, 
-		       RealD _hi      = 1.0, 
-		       int _maxit     = 1000,
-		       RealD _action_tolerance      = 1.0e-8, 
-		       int _action_degree    = 10,
-		       RealD _md_tolerance      = 1.0e-8, 
-		       int _md_degree    = 10,
-		       int _precision = 64,
-		       int _BoundsCheckFreq=20)
-    : inv_pow(_inv_pow), 
-      lo(_lo),
-      hi(_hi),
-      MaxIter(_maxit),
-      action_tolerance(_action_tolerance),
-      action_degree(_action_degree),
-      md_tolerance(_md_tolerance),
-      md_degree(_md_degree),
-      precision(_precision),
-      BoundsCheckFreq(_BoundsCheckFreq){};
-  };
-
-
 NAMESPACE_END(Grid);

 #endif
@@ -68,17 +68,9 @@ public:
  ///////////////////////////////////////////////////////////////
  // Support for MADWF tricks
  ///////////////////////////////////////////////////////////////
-  RealD Mass(void) { return (mass_plus + mass_minus) / 2.0; };
-  RealD MassPlus(void) { return mass_plus; };
-  RealD MassMinus(void) { return mass_minus; };
-
+  RealD Mass(void) { return mass; };
  void  SetMass(RealD _mass) { 
-    mass_plus=mass_minus=_mass; 
-    SetCoefficientsInternal(_zolo_hi,_gamma,_b,_c);  // Reset coeffs
-  } ;
-  void  SetMass(RealD _mass_plus, RealD _mass_minus) { 
-    mass_plus=_mass_plus;
-    mass_minus=_mass_minus;
+    mass=_mass; 
    SetCoefficientsInternal(_zolo_hi,_gamma,_b,_c);  // Reset coeffs
  } ;
  void  P(const FermionField &psi, FermionField &chi);
@@ -116,7 +108,7 @@ public:
  void   MeooeDag5D    (const FermionField &in, FermionField &out);

  //    protected:
-  RealD mass_plus, mass_minus;
+  RealD mass;

  // Save arguments to SetCoefficientsInternal
  Vector<Coeff_t> _gamma;
@@ -183,6 +175,16 @@ public:
 		  GridRedBlackCartesian &FourDimRedBlackGrid,
 		  RealD _mass,RealD _M5,const ImplParams &p= ImplParams());

+  void CayleyReport(void);
+  void CayleyZeroCounters(void);
+
+  double M5Dflops;
+  double M5Dcalls;
+  double M5Dtime;
+
+  double MooeeInvFlops;
+  double MooeeInvCalls;
+  double MooeeInvTime;

 protected:
  virtual void SetCoefficientsZolotarev(RealD zolohi,Approx::zolotarev_data *zdata,RealD b,RealD c);
@@ -1,436 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: ./lib/qcd/action/fermion/WilsonCloverFermionImplementation.h
-
-    Copyright (C) 2017 - 2022
-
-    Author: paboyle <paboyle@ph.ed.ac.uk>
-    Author: Daniel Richtmann <daniel.richtmann@gmail.com>
-    Author: Mattia Bruno <mattia.bruno@cern.ch>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-    *************************************************************************************/
-/*  END LEGAL */
-
-#pragma once
-
-#include <Grid/Grid.h>
-#include <Grid/qcd/spin/Dirac.h>
-#include <Grid/qcd/action/fermion/WilsonCloverHelpers.h>
-
-////////////////////////////////////////////
-// Standard Clover
-//   (4+m0) + csw * clover_term
-// Exp Clover
-//   (4+m0) * exp(csw/(4+m0) clover_term)
-//   = (4+m0) + csw * clover_term + ...
-////////////////////////////////////////////
-
-NAMESPACE_BEGIN(Grid);
-
-
-//////////////////////////////////
-// Generic Standard Clover
-//////////////////////////////////
-
-template<class Impl>
-class CloverHelpers: public WilsonCloverHelpers<Impl> {
-public:
-
-  INHERIT_IMPL_TYPES(Impl);
-  INHERIT_CLOVER_TYPES(Impl);
-
-  typedef WilsonCloverHelpers<Impl> Helpers;
-
-  static void Instantiate(CloverField& CloverTerm, CloverField& CloverTermInv, RealD csw_t, RealD diag_mass) {
-    GridBase *grid = CloverTerm.Grid();
-    CloverTerm += diag_mass;
-
-    int lvol = grid->lSites();
-    int DimRep = Impl::Dimension;
-    {
-      autoView(CTv,CloverTerm,CpuRead);
-      autoView(CTIv,CloverTermInv,CpuWrite);
-      thread_for(site, lvol, {
-        Coordinate lcoor;
-        grid->LocalIndexToLocalCoor(site, lcoor);
-        Eigen::MatrixXcd EigenCloverOp = Eigen::MatrixXcd::Zero(Ns * DimRep, Ns * DimRep);
-        Eigen::MatrixXcd EigenInvCloverOp = Eigen::MatrixXcd::Zero(Ns * DimRep, Ns * DimRep);
-        typename SiteClover::scalar_object Qx = Zero(), Qxinv = Zero();
-        peekLocalSite(Qx, CTv, lcoor);
-
-        for (int j = 0; j < Ns; j++)
-          for (int k = 0; k < Ns; k++)
-            for (int a = 0; a < DimRep; a++)
-              for (int b = 0; b < DimRep; b++){
-                auto zz =  Qx()(j, k)(a, b);
-                EigenCloverOp(a + j * DimRep, b + k * DimRep) = std::complex<double>(zz);
-              }
-
-        EigenInvCloverOp = EigenCloverOp.inverse();
-        for (int j = 0; j < Ns; j++)
-          for (int k = 0; k < Ns; k++)
-            for (int a = 0; a < DimRep; a++)
-              for (int b = 0; b < DimRep; b++)
-                Qxinv()(j, k)(a, b) = EigenInvCloverOp(a + j * DimRep, b + k * DimRep);
-               pokeLocalSite(Qxinv, CTIv, lcoor);
-      });
-    }
-  }
-
-  static GaugeLinkField Cmunu(std::vector<GaugeLinkField> &U, GaugeLinkField &lambda, int mu, int nu) {
-    return Helpers::Cmunu(U, lambda, mu, nu);
-  }
-
-};
-
-
-//////////////////////////////////
-// Generic Exp Clover
-//////////////////////////////////
-
-template<class Impl>
-class ExpCloverHelpers: public WilsonCloverHelpers<Impl> {
-public:
-
-  INHERIT_IMPL_TYPES(Impl);
-  INHERIT_CLOVER_TYPES(Impl);
-
-  template <typename vtype> using iImplClover = iScalar<iMatrix<iMatrix<vtype, Impl::Dimension>, Ns>>;
-  typedef WilsonCloverHelpers<Impl> Helpers;
-
-  // Can this be avoided?
-  static void IdentityTimesC(const CloverField& in, RealD c) {
-    int DimRep = Impl::Dimension;
-
-    autoView(in_v, in, AcceleratorWrite);
-
-    accelerator_for(ss, in.Grid()->oSites(), 1, {
-      for (int sa=0; sa<Ns; sa++)
-        for (int ca=0; ca<DimRep; ca++)
-          in_v[ss]()(sa,sa)(ca,ca) = c;
-    });
-  }
-
-  static int getNMAX(RealD prec, RealD R) {
-    /* compute stop condition for exponential */
-    int NMAX=1;
-    RealD cond=R*R/2.;
-
-    while (cond*std::exp(R)>prec) {
-      NMAX++;
-      cond*=R/(double)(NMAX+1);
-    }
-    return NMAX;
-  }
-
-  static int getNMAX(Lattice<iImplClover<vComplexD2>> &t, RealD R) {return getNMAX(1e-12,R);}
-  static int getNMAX(Lattice<iImplClover<vComplexD>> &t, RealD R) {return getNMAX(1e-12,R);}
-  static int getNMAX(Lattice<iImplClover<vComplexF>> &t, RealD R) {return getNMAX(1e-6,R);}
-
-  static void Instantiate(CloverField& Clover, CloverField& CloverInv, RealD csw_t, RealD diag_mass) {
-    GridBase* grid = Clover.Grid();
-    CloverField ExpClover(grid);
-
-    int NMAX = getNMAX(Clover, 3.*csw_t/diag_mass);
-
-    Clover *= (1.0/diag_mass);
-
-    // Taylor expansion, slow but generic
-    // Horner scheme: a0 + a1 x + a2 x^2 + .. = a0 + x (a1 + x(...))
-    // qN = cN
-    // qn = cn + qn+1 X
-    std::vector<RealD> cn(NMAX+1);
-    cn[0] = 1.0;
-    for (int i=1; i<=NMAX; i++)
-      cn[i] = cn[i-1] / RealD(i);
-
-    ExpClover = Zero();
-    IdentityTimesC(ExpClover, cn[NMAX]);
-    for (int i=NMAX-1; i>=0; i--)
-      ExpClover = ExpClover * Clover + cn[i];
-
-    // prepare inverse
-    CloverInv = (-1.0)*Clover;
-
-    Clover = ExpClover * diag_mass;
-
-    ExpClover = Zero();
-    IdentityTimesC(ExpClover, cn[NMAX]);
-    for (int i=NMAX-1; i>=0; i--)
-      ExpClover = ExpClover * CloverInv + cn[i];
-
-    CloverInv = ExpClover * (1.0/diag_mass);
-
-  }
-
-  static GaugeLinkField Cmunu(std::vector<GaugeLinkField> &U, GaugeLinkField &lambda, int mu, int nu) {
-    assert(0);
-    return lambda;
-  }
-
-};
-
-
-//////////////////////////////////
-// Compact Standard Clover
-//////////////////////////////////
-
-
-template<class Impl>
-class CompactCloverHelpers: public CompactWilsonCloverHelpers<Impl>,
-                            public WilsonCloverHelpers<Impl> {
-public:
-
-  INHERIT_IMPL_TYPES(Impl);
-  INHERIT_CLOVER_TYPES(Impl);
-  INHERIT_COMPACT_CLOVER_TYPES(Impl);
-
-  typedef WilsonCloverHelpers<Impl> Helpers;
-  typedef CompactWilsonCloverHelpers<Impl> CompactHelpers;
-
-  static void MassTerm(CloverField& Clover, RealD diag_mass) {
-    Clover += diag_mass;
-  }
-
-  static void Exponentiate_Clover(CloverDiagonalField& Diagonal,
-                          CloverTriangleField& Triangle,
-                          RealD csw_t, RealD diag_mass) {
-
-    // Do nothing
-  }
-
-  // TODO: implement Cmunu for better performances with compact layout, but don't do it
-  // here, but rather in WilsonCloverHelpers.h -> CompactWilsonCloverHelpers
-  static GaugeLinkField Cmunu(std::vector<GaugeLinkField> &U, GaugeLinkField &lambda, int mu, int nu) {
-    return Helpers::Cmunu(U, lambda, mu, nu);
-  }
-};
-
-//////////////////////////////////
-// Compact Exp Clover
-//////////////////////////////////
-
-template<class Impl>
-class CompactExpCloverHelpers: public CompactWilsonCloverHelpers<Impl> {
-public:
-
-  INHERIT_IMPL_TYPES(Impl);
-  INHERIT_CLOVER_TYPES(Impl);
-  INHERIT_COMPACT_CLOVER_TYPES(Impl);
-
-  template <typename vtype> using iImplClover = iScalar<iMatrix<iMatrix<vtype, Impl::Dimension>, Ns>>;
-  typedef CompactWilsonCloverHelpers<Impl> CompactHelpers;
-
-  static void MassTerm(CloverField& Clover, RealD diag_mass) {
-    // do nothing!
-    // mass term is multiplied to exp(Clover) below
-  }
-
-  static int getNMAX(RealD prec, RealD R) {
-    /* compute stop condition for exponential */
-    int NMAX=1;
-    RealD cond=R*R/2.;
-
-    while (cond*std::exp(R)>prec) {
-      NMAX++;
-      cond*=R/(double)(NMAX+1);
-    }
-    return NMAX;
-  }
-
-  static int getNMAX(Lattice<iImplCloverDiagonal<vComplexD>> &t, RealD R) {return getNMAX(1e-12,R);}
-  static int getNMAX(Lattice<iImplCloverDiagonal<vComplexF>> &t, RealD R) {return getNMAX(1e-6,R);}
-
-  static void ExponentiateHermitean6by6(const iMatrix<ComplexD,6> &arg, const RealD& alpha, const std::vector<RealD>& cN, const int Niter, iMatrix<ComplexD,6>& dest){
-
-  	  typedef iMatrix<ComplexD,6> mat;
-
-  	  RealD qn[6];
-  	  RealD qnold[6];
-  	  RealD p[5];
-  	  RealD trA2, trA3, trA4;
-
-  	  mat A2, A3, A4, A5;
-  	  A2 = alpha * alpha * arg * arg;
-  	  A3 = alpha * arg * A2;
-  	  A4 = A2 * A2;
-  	  A5 = A2 * A3;
-
-  	  trA2 = toReal( trace(A2) );
-  	  trA3 = toReal( trace(A3) );
-  	  trA4 = toReal( trace(A4));
-
-  	  p[0] = toReal( trace(A3 * A3)) / 6.0 - 0.125 * trA4 * trA2 - trA3 * trA3 / 18.0 + trA2 * trA2 * trA2/ 48.0;
-  	  p[1] = toReal( trace(A5)) / 5.0 - trA3 * trA2 / 6.0;
-  	  p[2] = toReal( trace(A4)) / 4.0 - 0.125 * trA2 * trA2;
-  	  p[3] = trA3 / 3.0;
-  	  p[4] = 0.5 * trA2;
-
-  	  qnold[0] = cN[Niter];
-  	  qnold[1] = 0.0;
-  	  qnold[2] = 0.0;
-  	  qnold[3] = 0.0;
-  	  qnold[4] = 0.0;
-  	  qnold[5] = 0.0;
-
-  	  for(int i = Niter-1; i >= 0; i--)
-  	  {
-  	   qn[0] = p[0] * qnold[5] + cN[i];
-  	   qn[1] = p[1] * qnold[5] + qnold[0];
-  	   qn[2] = p[2] * qnold[5] + qnold[1];
-  	   qn[3] = p[3] * qnold[5] + qnold[2];
-  	   qn[4] = p[4] * qnold[5] + qnold[3];
-  	   qn[5] = qnold[4];
-
-  	   qnold[0] = qn[0];
-  	   qnold[1] = qn[1];
-  	   qnold[2] = qn[2];
-  	   qnold[3] = qn[3];
-  	   qnold[4] = qn[4];
-  	   qnold[5] = qn[5];
-  	  }
-
-  	  mat unit(1.0);
-
-  	  dest = (qn[0] * unit + qn[1] * alpha * arg + qn[2] * A2 + qn[3] * A3 + qn[4] * A4 + qn[5] * A5);
-
-    }
-
-  static void Exponentiate_Clover(CloverDiagonalField& Diagonal, CloverTriangleField& Triangle, RealD csw_t, RealD diag_mass) {
-
-    GridBase* grid = Diagonal.Grid();
-    int NMAX = getNMAX(Diagonal, 3.*csw_t/diag_mass);
-
-    //
-    // Implementation completely in Daniel's layout
-    //
-
-    // Taylor expansion with Cayley-Hamilton recursion
-    // underlying Horner scheme as above
-    std::vector<RealD> cn(NMAX+1);
-    cn[0] = 1.0;
-    for (int i=1; i<=NMAX; i++){
-      cn[i] = cn[i-1] / RealD(i);
-    }
-
-      // Taken over from Daniel's implementation
-      conformable(Diagonal, Triangle);
-
-      long lsites = grid->lSites();
-    {
-      typedef typename SiteCloverDiagonal::scalar_object scalar_object_diagonal;
-      typedef typename SiteCloverTriangle::scalar_object scalar_object_triangle;
-      typedef iMatrix<ComplexD,6> mat;
-
-      autoView(diagonal_v,  Diagonal,  CpuRead);
-      autoView(triangle_v,  Triangle,  CpuRead);
-      autoView(diagonalExp_v, Diagonal, CpuWrite);
-      autoView(triangleExp_v, Triangle, CpuWrite);
-
-      thread_for(site, lsites, { // NOTE: Not on GPU because of (peek/poke)LocalSite
-
-    	  mat srcCloverOpUL(0.0); // upper left block
-    	  mat srcCloverOpLR(0.0); // lower right block
-    	  mat ExpCloverOp;
-
-        scalar_object_diagonal diagonal_tmp     = Zero();
-        scalar_object_diagonal diagonal_exp_tmp = Zero();
-        scalar_object_triangle triangle_tmp     = Zero();
-        scalar_object_triangle triangle_exp_tmp = Zero();
-
-        Coordinate lcoor;
-        grid->LocalIndexToLocalCoor(site, lcoor);
-
-        peekLocalSite(diagonal_tmp, diagonal_v, lcoor);
-        peekLocalSite(triangle_tmp, triangle_v, lcoor);
-
-        int block;
-        block = 0;
-        for(int i = 0; i < 6; i++){
-        	for(int j = 0; j < 6; j++){
-        		if (i == j){
-        			srcCloverOpUL(i,j) = static_cast<ComplexD>(TensorRemove(diagonal_tmp()(block)(i)));
-        		}
-        		else{
-        			srcCloverOpUL(i,j) = static_cast<ComplexD>(TensorRemove(CompactHelpers::triangle_elem(triangle_tmp, block, i, j)));
-        		}
-        	}
-        }
-        block = 1;
-        for(int i = 0; i < 6; i++){
-          	for(int j = 0; j < 6; j++){
-           		if (i == j){
-           			srcCloverOpLR(i,j) = static_cast<ComplexD>(TensorRemove(diagonal_tmp()(block)(i)));
-           		}
-           		else{
-           			srcCloverOpLR(i,j) = static_cast<ComplexD>(TensorRemove(CompactHelpers::triangle_elem(triangle_tmp, block, i, j)));
-           		}
-            }
-        }
-
-        // exp(Clover)
-
-        ExponentiateHermitean6by6(srcCloverOpUL,1.0/diag_mass,cn,NMAX,ExpCloverOp);
-
-        block = 0;
-        for(int i = 0; i < 6; i++){
-        	for(int j = 0; j < 6; j++){
-            	if (i == j){
-            		diagonal_exp_tmp()(block)(i) = ExpCloverOp(i,j);
-            	}
-            	else if(i < j){
-            		triangle_exp_tmp()(block)(CompactHelpers::triangle_index(i, j)) = ExpCloverOp(i,j);
-            	}
-           	}
-        }
-
-        ExponentiateHermitean6by6(srcCloverOpLR,1.0/diag_mass,cn,NMAX,ExpCloverOp);
-
-        block = 1;
-        for(int i = 0; i < 6; i++){
-        	for(int j = 0; j < 6; j++){
-              	if (i == j){
-              		diagonal_exp_tmp()(block)(i) = ExpCloverOp(i,j);
-               	}
-               	else if(i < j){
-               		triangle_exp_tmp()(block)(CompactHelpers::triangle_index(i, j)) = ExpCloverOp(i,j);
-               	}
-            }
-        }
-
-        pokeLocalSite(diagonal_exp_tmp, diagonalExp_v, lcoor);
-        pokeLocalSite(triangle_exp_tmp, triangleExp_v, lcoor);
-      });
-    }
-
-    Diagonal *= diag_mass;
-    Triangle *= diag_mass;
-  }
-
-
-  static GaugeLinkField Cmunu(std::vector<GaugeLinkField> &U, GaugeLinkField &lambda, int mu, int nu) {
-    assert(0);
-    return lambda;
-  }
-
-};
-
-
-NAMESPACE_END(Grid);
@@ -1,241 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: ./lib/qcd/action/fermion/CompactWilsonCloverFermion.h
-
-    Copyright (C) 2020 - 2022
-
-    Author: Daniel Richtmann <daniel.richtmann@gmail.com>
-    Author: Nils Meyer <nils.meyer@ur.de>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-    *************************************************************************************/
-/*  END LEGAL */
-
-#pragma once
-
-#include <Grid/qcd/action/fermion/WilsonCloverTypes.h>
-#include <Grid/qcd/action/fermion/WilsonCloverHelpers.h>
-#include <Grid/qcd/action/fermion/CloverHelpers.h>
-
-NAMESPACE_BEGIN(Grid);
-
-// see Grid/qcd/action/fermion/WilsonCloverFermion.h for description
-//
-// Modifications done here:
-//
-// Original: clover term = 12x12 matrix per site
-//
-// But: Only two diagonal 6x6 hermitian blocks are non-zero (also true for original, verified by running)
-// Sufficient to store/transfer only the real parts of the diagonal and one triangular part
-// 2 * (6 + 15 * 2) = 72 real or 36 complex words to be stored/transfered
-//
-// Here: Above but diagonal as complex numbers, i.e., need to store/transfer
-// 2 * (6 * 2 + 15 * 2) = 84 real or 42 complex words
-//
-// Words per site and improvement compared to original (combined with the input and output spinors):
-//
-// - Original: 2*12 + 12*12 = 168 words -> 1.00 x less
-// - Minimal:  2*12 + 36    =  60 words -> 2.80 x less
-// - Here:     2*12 + 42    =  66 words -> 2.55 x less
-//
-// These improvements directly translate to wall-clock time
-//
-// Data layout:
-//
-// - diagonal and triangle part as separate lattice fields,
-//   this was faster than as 1 combined field on all tested machines
-// - diagonal: as expected
-// - triangle: store upper right triangle in row major order
-// - graphical:
-//        0  1  2  3  4
-//           5  6  7  8
-//              9 10 11 = upper right triangle indices
-//                12 13
-//                   14
-//     0
-//        1
-//           2
-//              3       = diagonal indices
-//                 4
-//                    5
-//     0
-//     1  5
-//     2  6  9          = lower left triangle indices
-//     3  7 10 12
-//     4  8 11 13 14
-//
-// Impact on total memory consumption:
-// - Original: (2 * 1 + 8 * 1/2) 12x12 matrices = 6 12x12 matrices = 864 complex words per site
-// - Here:     (2 * 1 + 4 * 1/2) diagonal parts = 4 diagonal parts =  24 complex words per site
-//           + (2 * 1 + 4 * 1/2) triangle parts = 4 triangle parts =  60 complex words per site
-//                                                                 =  84 complex words per site
-
-template<class Impl, class CloverHelpers>
-class CompactWilsonCloverFermion : public WilsonFermion<Impl>,
-                                   public WilsonCloverHelpers<Impl>,
-                                   public CompactWilsonCloverHelpers<Impl> {
-  /////////////////////////////////////////////
-  // Sizes
-  /////////////////////////////////////////////
-
-public:
-
-  INHERIT_COMPACT_CLOVER_SIZES(Impl);
-
-  /////////////////////////////////////////////
-  // Type definitions
-  /////////////////////////////////////////////
-
-public:
-
-  INHERIT_IMPL_TYPES(Impl);
-  INHERIT_CLOVER_TYPES(Impl);
-  INHERIT_COMPACT_CLOVER_TYPES(Impl);
-
-  typedef WilsonFermion<Impl>              WilsonBase;
-  typedef WilsonCloverHelpers<Impl>        Helpers;
-  typedef CompactWilsonCloverHelpers<Impl> CompactHelpers;
-
-  /////////////////////////////////////////////
-  // Constructors
-  /////////////////////////////////////////////
-
-public:
-
-  CompactWilsonCloverFermion(GaugeField& _Umu,
-			    GridCartesian& Fgrid,
-			    GridRedBlackCartesian& Hgrid,
-			    const RealD _mass,
-			    const RealD _csw_r = 0.0,
-			    const RealD _csw_t = 0.0,
-			    const RealD _cF = 1.0,
-			    const WilsonAnisotropyCoefficients& clover_anisotropy = WilsonAnisotropyCoefficients(),
-			    const ImplParams& impl_p = ImplParams());
-
-  /////////////////////////////////////////////
-  // Member functions (implementing interface)
-  /////////////////////////////////////////////
-
-public:
-
-  virtual void Instantiatable() {};
-  int          ConstEE()     override { return 0; };
-  int          isTrivialEE() override { return 0; };
-
-  void Dhop(const FermionField& in, FermionField& out, int dag) override;
-
-  void DhopOE(const FermionField& in, FermionField& out, int dag) override;
-
-  void DhopEO(const FermionField& in, FermionField& out, int dag) override;
-
-  void DhopDir(const FermionField& in, FermionField& out, int dir, int disp) override;
-
-  void DhopDirAll(const FermionField& in, std::vector<FermionField>& out) /* override */;
-
-  void M(const FermionField& in, FermionField& out) override;
-
-  void Mdag(const FermionField& in, FermionField& out) override;
-
-  void Meooe(const FermionField& in, FermionField& out) override;
-
-  void MeooeDag(const FermionField& in, FermionField& out) override;
-
-  void Mooee(const FermionField& in, FermionField& out) override;
-
-  void MooeeDag(const FermionField& in, FermionField& out) override;
-
-  void MooeeInv(const FermionField& in, FermionField& out) override;
-
-  void MooeeInvDag(const FermionField& in, FermionField& out) override;
-
-  void Mdir(const FermionField& in, FermionField& out, int dir, int disp) override;
-
-  void MdirAll(const FermionField& in, std::vector<FermionField>& out) override;
-
-  void MDeriv(GaugeField& force, const FermionField& X, const FermionField& Y, int dag) override;
-
-  void MooDeriv(GaugeField& mat, const FermionField& U, const FermionField& V, int dag) override;
-
-  void MeeDeriv(GaugeField& mat, const FermionField& U, const FermionField& V, int dag) override;
-
-  /////////////////////////////////////////////
-  // Member functions (internals)
-  /////////////////////////////////////////////
-
-  void MooeeInternal(const FermionField&        in,
-                     FermionField&              out,
-                     const CloverDiagonalField& diagonal,
-                     const CloverTriangleField& triangle);
-
-  /////////////////////////////////////////////
-  // Helpers
-  /////////////////////////////////////////////
-
-  void ImportGauge(const GaugeField& _Umu) override;
-
-  /////////////////////////////////////////////
-  // Helpers
-  /////////////////////////////////////////////
-
-private:
-
-  template<class Field>
-  const MaskField* getCorrectMaskField(const Field &in) const {
-    if(in.Grid()->_isCheckerBoarded) {
-      if(in.Checkerboard() == Odd) {
-        return &this->BoundaryMaskOdd;
-      } else {
-        return &this->BoundaryMaskEven;
-      }
-    } else {
-      return &this->BoundaryMask;
-    }
-  }
-
-  template<class Field>
-  void ApplyBoundaryMask(Field& f) {
-    const MaskField* m = getCorrectMaskField(f); assert(m != nullptr);
-    assert(m != nullptr);
-    CompactHelpers::ApplyBoundaryMask(f, *m);
-  }
-
-  /////////////////////////////////////////////
-  // Member Data
-  /////////////////////////////////////////////
-
-public:
-
-  RealD csw_r;
-  RealD csw_t;
-  RealD cF;
-
-  bool open_boundaries;
-
-  CloverDiagonalField Diagonal,    DiagonalEven,    DiagonalOdd;
-  CloverDiagonalField DiagonalInv, DiagonalInvEven, DiagonalInvOdd;
-
-  CloverTriangleField Triangle,    TriangleEven,    TriangleOdd;
-  CloverTriangleField TriangleInv, TriangleInvEven, TriangleInvOdd;
-
-  FermionField Tmp;
-
-  MaskField BoundaryMask, BoundaryMaskEven, BoundaryMaskOdd;
-};
-
-NAMESPACE_END(Grid);
@@ -1,291 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/DWFSlow.h
-
-Copyright (C) 2022
-
-Author: Peter Boyle <pboyle@bnl.gov>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-			   /*  END LEGAL */
-#pragma once
-
-NAMESPACE_BEGIN(Grid);
-
-template <class Impl>
-class DWFSlowFermion : public FermionOperator<Impl>
-{
-public:
-  INHERIT_IMPL_TYPES(Impl);
-
-  ///////////////////////////////////////////////////////////////
-  // Implement the abstract base
-  ///////////////////////////////////////////////////////////////
-  GridBase *GaugeGrid(void) { return _grid4; }
-  GridBase *GaugeRedBlackGrid(void) { return _cbgrid4; }
-  GridBase *FermionGrid(void) { return _grid; }
-  GridBase *FermionRedBlackGrid(void) { return _cbgrid; }
-
-  FermionField _tmp;
-  FermionField &tmp(void) { return _tmp; }
-
-  //////////////////////////////////////////////////////////////////
-  // override multiply; cut number routines if pass dagger argument
-  // and also make interface more uniformly consistent
-  //////////////////////////////////////////////////////////////////
-  virtual void  M(const FermionField &in, FermionField &out)
-  {
-    FermionField tmp(_grid);
-    out = (5.0 - M5) * in;
-    Dhop(in,tmp,DaggerNo);
-    out = out + tmp;
-  }
-  virtual void  Mdag(const FermionField &in, FermionField &out)
-  {
-    FermionField tmp(_grid);
-    out = (5.0 - M5) * in;
-    Dhop(in,tmp,DaggerYes);
-    out = out + tmp;
-  };
-
-  /////////////////////////////////////////////////////////
-  // half checkerboard operations 5D redblack so just site identiy
-  /////////////////////////////////////////////////////////
-  void Meooe(const FermionField &in, FermionField &out)
-  {
-    if ( in.Checkerboard() == Odd ) {
-      this->DhopEO(in,out,DaggerNo);
-    } else {
-      this->DhopOE(in,out,DaggerNo);
-    }
-  }
-  void MeooeDag(const FermionField &in, FermionField &out)
-  {
-    if ( in.Checkerboard() == Odd ) {
-      this->DhopEO(in,out,DaggerYes);
-    } else {
-      this->DhopOE(in,out,DaggerYes);
-    }
-  };
-
-  // allow override for twisted mass and clover
-  virtual void Mooee(const FermionField &in, FermionField &out)
-  {
-    out = (5.0 - M5) * in;
-  }
-  virtual void MooeeDag(const FermionField &in, FermionField &out)
-  {
-    out = (5.0 - M5) * in;
-  }
-  virtual void MooeeInv(const FermionField &in, FermionField &out)
-  {
-    out = (1.0/(5.0 - M5)) * in;
-  };
-  virtual void MooeeInvDag(const FermionField &in, FermionField &out)
-  {
-    out = (1.0/(5.0 - M5)) * in;
-  };
-
-  virtual void  MomentumSpacePropagator(FermionField &out,const FermionField &in,RealD _mass,std::vector<double> twist) {} ;
-
-  ////////////////////////
-  // Derivative interface
-  ////////////////////////
-  // Interface calls an internal routine
-  void DhopDeriv(GaugeField &mat,const FermionField &U,const FermionField &V,int dag)  { assert(0);};
-  void DhopDerivOE(GaugeField &mat,const FermionField &U,const FermionField &V,int dag){ assert(0);};
-  void DhopDerivEO(GaugeField &mat,const FermionField &U,const FermionField &V,int dag){ assert(0);};
-
-  ///////////////////////////////////////////////////////////////
-  // non-hermitian hopping term; half cb or both
-  ///////////////////////////////////////////////////////////////
-  void Dhop(const FermionField &in, FermionField &out, int dag)
-  {
-    FermionField tmp(in.Grid());
-    Dhop5(in,out,MassField,MassField,dag );
-    for(int mu=0;mu<4;mu++){
-      DhopDirU(in,Umu[mu],Umu[mu],tmp,mu,dag );    out = out + tmp;
-    }
-  };
-  void DhopOE(const FermionField &in, FermionField &out, int dag)
-  {
-    FermionField tmp(in.Grid());
-    assert(in.Checkerboard()==Even);
-    Dhop5(in,out,MassFieldOdd,MassFieldEven,dag);
-    for(int mu=0;mu<4;mu++){
-      DhopDirU(in,UmuOdd[mu],UmuEven[mu],tmp,mu,dag );    out = out + tmp;
-    }
-  };
-  void DhopEO(const FermionField &in, FermionField &out, int dag)
-  {
-    FermionField tmp(in.Grid());
-    assert(in.Checkerboard()==Odd);
-    Dhop5(in,out, MassFieldEven,MassFieldOdd ,dag );  
-    for(int mu=0;mu<4;mu++){
-      DhopDirU(in,UmuEven[mu],UmuOdd[mu],tmp,mu,dag );    out = out + tmp;
-    }
-  };
-
-  ///////////////////////////////////////////////////////////////
-  // Multigrid assistance; force term uses too
-  ///////////////////////////////////////////////////////////////
-  void Mdir(const FermionField &in, FermionField &out, int dir, int disp){ assert(0);};
-  void MdirAll(const FermionField &in, std::vector<FermionField> &out)   { assert(0);};
-  void DhopDir(const FermionField &in, FermionField &out, int dir, int disp) { assert(0);};
-  void DhopDirAll(const FermionField &in, std::vector<FermionField> &out)    { assert(0);};
-  void DhopDirCalc(const FermionField &in, FermionField &out, int dirdisp,int gamma, int dag) { assert(0);};
-
-  void DhopDirU(const FermionField &in, const GaugeLinkField &U5e, const GaugeLinkField &U5o, FermionField &out, int mu, int dag)
-  {
-    RealD     sgn= 1.0;
-    if (dag ) sgn=-1.0;
-
-    Gamma::Algebra Gmu [] = {
-			 Gamma::Algebra::GammaX,
-			 Gamma::Algebra::GammaY,
-			 Gamma::Algebra::GammaZ,
-			 Gamma::Algebra::GammaT
-    };
-
-    //    mass is  1,1,1,1,-m has to multiply the round the world term
-    FermionField tmp (in.Grid());
-    tmp = U5e * Cshift(in,mu+1,1);
-    out = tmp - Gamma(Gmu[mu])*tmp*sgn;
-    
-    tmp = Cshift(adj(U5o)*in,mu+1,-1);
-    out = out + tmp + Gamma(Gmu[mu])*tmp*sgn;
-
-    out = -0.5*out;
-  };
-
-  void Dhop5(const FermionField &in, FermionField &out, ComplexField &massE, ComplexField &massO, int dag)
-  {
-    // Mass term.... must multiple the round world with mass = 1,1,1,1, -m
-    RealD     sgn= 1.0;
-    if (dag ) sgn=-1.0;
-
-    Gamma G5(Gamma::Algebra::Gamma5);
-
-    FermionField tmp (in.Grid());
-    tmp = massE*Cshift(in,0,1);
-    out = tmp - G5*tmp*sgn;
-    
-    tmp = Cshift(massO*in,0,-1);
-    out = out + tmp + G5*tmp*sgn;
-    out = -0.5*out;
-  };
-
-  // Constructor
-  DWFSlowFermion(GaugeField &_Umu, GridCartesian &Fgrid,
-		 GridRedBlackCartesian &Hgrid, RealD _mass, RealD _M5)
-    :
-    _grid(&Fgrid),
-    _cbgrid(&Hgrid),
-    _grid4(_Umu.Grid()),
-    Umu(Nd,&Fgrid),
-    UmuEven(Nd,&Hgrid),
-    UmuOdd(Nd,&Hgrid),
-    MassField(&Fgrid),
-    MassFieldEven(&Hgrid),
-    MassFieldOdd(&Hgrid),
-    M5(_M5),
-    mass(_mass),
-    _tmp(&Hgrid)
-    {
-      Ls=Fgrid._fdimensions[0];
-      ImportGauge(_Umu);
-
-      typedef typename FermionField::scalar_type scalar;
-
-      Lattice<iScalar<vInteger> > coor(&Fgrid);
-      LatticeCoordinate(coor, 0); // Scoor
-      ComplexField one(&Fgrid);
-      MassField =scalar(-mass);
-      one       =scalar(1.0);
-      MassField =where(coor==Integer(Ls-1),MassField,one);
-      for(int mu=0;mu<Nd;mu++){
-	pickCheckerboard(Even,UmuEven[mu],Umu[mu]);
-	pickCheckerboard(Odd ,UmuOdd[mu],Umu[mu]);
-      }
-      pickCheckerboard(Even,MassFieldEven,MassField);
-      pickCheckerboard(Odd ,MassFieldOdd,MassField);
-      
-    }
-  
-  // DoubleStore impl dependent
-  void ImportGauge(const GaugeField &_Umu4)
-  {
-    GaugeLinkField U4(_grid4);
-    for(int mu=0;mu<Nd;mu++){
-      U4 = PeekIndex<LorentzIndex>(_Umu4, mu);
-      for(int s=0;s<this->Ls;s++){
-	InsertSlice(U4,Umu[mu],s,0);
-      }
-    }
-  }
-
-  ///////////////////////////////////////////////////////////////
-  // Data members require to support the functionality
-  ///////////////////////////////////////////////////////////////
-
-public:
-  virtual RealD Mass(void) { return mass; }
-  virtual int   isTrivialEE(void) { return 1; };
-  RealD mass;
-  RealD M5;
-  int Ls;
-
-  GridBase *_grid4;
-  GridBase *_grid;
-  GridBase *_cbgrid4;
-  GridBase *_cbgrid;
-
-  // Copy of the gauge field , with even and odd subsets
-  std::vector<GaugeLinkField> Umu;
-  std::vector<GaugeLinkField> UmuEven;
-  std::vector<GaugeLinkField> UmuOdd;
-  ComplexField MassField;
-  ComplexField MassFieldEven;
-  ComplexField MassFieldOdd;
-
-  ///////////////////////////////////////////////////////////////
-  // Conserved current utilities
-  ///////////////////////////////////////////////////////////////
-  void ContractConservedCurrent(PropagatorField &q_in_1,
-                                PropagatorField &q_in_2,
-                                PropagatorField &q_out,
-                                PropagatorField &phys_src,
-                                Current curr_type,
-                                unsigned int mu){}
-  void SeqConservedCurrent(PropagatorField &q_in,
-                           PropagatorField &q_out,
-                           PropagatorField &phys_src,
-                           Current curr_type,
-                           unsigned int mu,
-                           unsigned int tmin,
-			   unsigned int tmax,
-			   ComplexField &lattice_cmplx){}
-};
-
-typedef DWFSlowFermion<WilsonImplF> DWFSlowFermionF;
-typedef DWFSlowFermion<WilsonImplD> DWFSlowFermionD;
-
-NAMESPACE_END(Grid);
@@ -47,14 +47,12 @@ Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
 ////////////////////////////////////////////
 // Fermion operators / actions
 ////////////////////////////////////////////
-#include <Grid/qcd/action/fermion/DWFSlow.h>       // Slow DWF

 #include <Grid/qcd/action/fermion/WilsonFermion.h>       // 4d wilson like
 NAMESPACE_CHECK(Wilson);
 #include <Grid/qcd/action/fermion/WilsonTMFermion.h>       // 4d wilson like
 NAMESPACE_CHECK(WilsonTM);
 #include <Grid/qcd/action/fermion/WilsonCloverFermion.h> // 4d wilson clover fermions
-#include <Grid/qcd/action/fermion/CompactWilsonCloverFermion.h> // 4d compact wilson clover fermions
 NAMESPACE_CHECK(WilsonClover);
 #include <Grid/qcd/action/fermion/WilsonFermion5D.h>     // 5d base used by all 5d overlap types
 NAMESPACE_CHECK(Wilson5D);
@@ -113,151 +111,183 @@ NAMESPACE_CHECK(DWFutils);
 // Cayley 5d
 NAMESPACE_BEGIN(Grid);

-typedef WilsonFermion<WilsonImplD2> WilsonFermionD2;
+typedef WilsonFermion<WilsonImplR> WilsonFermionR;
 typedef WilsonFermion<WilsonImplF> WilsonFermionF;
 typedef WilsonFermion<WilsonImplD> WilsonFermionD;

+typedef WilsonFermion<WilsonImplRL> WilsonFermionRL;
+typedef WilsonFermion<WilsonImplFH> WilsonFermionFH;
+typedef WilsonFermion<WilsonImplDF> WilsonFermionDF;
+
+typedef WilsonFermion<WilsonAdjImplR> WilsonAdjFermionR;
 typedef WilsonFermion<WilsonAdjImplF> WilsonAdjFermionF;
 typedef WilsonFermion<WilsonAdjImplD> WilsonAdjFermionD;

+typedef WilsonFermion<WilsonTwoIndexSymmetricImplR> WilsonTwoIndexSymmetricFermionR;
 typedef WilsonFermion<WilsonTwoIndexSymmetricImplF> WilsonTwoIndexSymmetricFermionF;
 typedef WilsonFermion<WilsonTwoIndexSymmetricImplD> WilsonTwoIndexSymmetricFermionD;

+typedef WilsonFermion<WilsonTwoIndexAntiSymmetricImplR> WilsonTwoIndexAntiSymmetricFermionR;
 typedef WilsonFermion<WilsonTwoIndexAntiSymmetricImplF> WilsonTwoIndexAntiSymmetricFermionF;
 typedef WilsonFermion<WilsonTwoIndexAntiSymmetricImplD> WilsonTwoIndexAntiSymmetricFermionD;

 // Twisted mass fermion
-typedef WilsonTMFermion<WilsonImplD2> WilsonTMFermionD2;
+typedef WilsonTMFermion<WilsonImplR> WilsonTMFermionR;
 typedef WilsonTMFermion<WilsonImplF> WilsonTMFermionF;
 typedef WilsonTMFermion<WilsonImplD> WilsonTMFermionD;

 // Clover fermions
-template <typename WImpl> using WilsonClover = WilsonCloverFermion<WImpl, CloverHelpers<WImpl>>;
-template <typename WImpl> using WilsonExpClover = WilsonCloverFermion<WImpl, ExpCloverHelpers<WImpl>>;
+typedef WilsonCloverFermion<WilsonImplR> WilsonCloverFermionR;
+typedef WilsonCloverFermion<WilsonImplF> WilsonCloverFermionF;
+typedef WilsonCloverFermion<WilsonImplD> WilsonCloverFermionD;

-typedef WilsonClover<WilsonImplD2> WilsonCloverFermionD2;
-typedef WilsonClover<WilsonImplF> WilsonCloverFermionF;
-typedef WilsonClover<WilsonImplD> WilsonCloverFermionD;
+typedef WilsonCloverFermion<WilsonAdjImplR> WilsonCloverAdjFermionR;
+typedef WilsonCloverFermion<WilsonAdjImplF> WilsonCloverAdjFermionF;
+typedef WilsonCloverFermion<WilsonAdjImplD> WilsonCloverAdjFermionD;

-typedef WilsonExpClover<WilsonImplD2> WilsonExpCloverFermionD2;
-typedef WilsonExpClover<WilsonImplF> WilsonExpCloverFermionF;
-typedef WilsonExpClover<WilsonImplD> WilsonExpCloverFermionD;
+typedef WilsonCloverFermion<WilsonTwoIndexSymmetricImplR> WilsonCloverTwoIndexSymmetricFermionR;
+typedef WilsonCloverFermion<WilsonTwoIndexSymmetricImplF> WilsonCloverTwoIndexSymmetricFermionF;
+typedef WilsonCloverFermion<WilsonTwoIndexSymmetricImplD> WilsonCloverTwoIndexSymmetricFermionD;

-typedef WilsonClover<WilsonAdjImplF> WilsonCloverAdjFermionF;
-typedef WilsonClover<WilsonAdjImplD> WilsonCloverAdjFermionD;
-
-typedef WilsonClover<WilsonTwoIndexSymmetricImplF> WilsonCloverTwoIndexSymmetricFermionF;
-typedef WilsonClover<WilsonTwoIndexSymmetricImplD> WilsonCloverTwoIndexSymmetricFermionD;
-
-typedef WilsonClover<WilsonTwoIndexAntiSymmetricImplF> WilsonCloverTwoIndexAntiSymmetricFermionF;
-typedef WilsonClover<WilsonTwoIndexAntiSymmetricImplD> WilsonCloverTwoIndexAntiSymmetricFermionD;
-
-// Compact Clover fermions
-template <typename WImpl> using CompactWilsonClover = CompactWilsonCloverFermion<WImpl, CompactCloverHelpers<WImpl>>;
-template <typename WImpl> using CompactWilsonExpClover = CompactWilsonCloverFermion<WImpl, CompactExpCloverHelpers<WImpl>>;
-
-typedef CompactWilsonClover<WilsonImplD2> CompactWilsonCloverFermionD2;
-typedef CompactWilsonClover<WilsonImplF> CompactWilsonCloverFermionF;
-typedef CompactWilsonClover<WilsonImplD> CompactWilsonCloverFermionD;
-
-typedef CompactWilsonExpClover<WilsonImplD2> CompactWilsonExpCloverFermionD2;
-typedef CompactWilsonExpClover<WilsonImplF> CompactWilsonExpCloverFermionF;
-typedef CompactWilsonExpClover<WilsonImplD> CompactWilsonExpCloverFermionD;
-
-typedef CompactWilsonClover<WilsonAdjImplF> CompactWilsonCloverAdjFermionF;
-typedef CompactWilsonClover<WilsonAdjImplD> CompactWilsonCloverAdjFermionD;
-
-typedef CompactWilsonClover<WilsonTwoIndexSymmetricImplF> CompactWilsonCloverTwoIndexSymmetricFermionF;
-typedef CompactWilsonClover<WilsonTwoIndexSymmetricImplD> CompactWilsonCloverTwoIndexSymmetricFermionD;
-
-typedef CompactWilsonClover<WilsonTwoIndexAntiSymmetricImplF> CompactWilsonCloverTwoIndexAntiSymmetricFermionF;
-typedef CompactWilsonClover<WilsonTwoIndexAntiSymmetricImplD> CompactWilsonCloverTwoIndexAntiSymmetricFermionD;
+typedef WilsonCloverFermion<WilsonTwoIndexAntiSymmetricImplR> WilsonCloverTwoIndexAntiSymmetricFermionR;
+typedef WilsonCloverFermion<WilsonTwoIndexAntiSymmetricImplF> WilsonCloverTwoIndexAntiSymmetricFermionF;
+typedef WilsonCloverFermion<WilsonTwoIndexAntiSymmetricImplD> WilsonCloverTwoIndexAntiSymmetricFermionD;

 // Domain Wall fermions
+typedef DomainWallFermion<WilsonImplR> DomainWallFermionR;
 typedef DomainWallFermion<WilsonImplF> DomainWallFermionF;
 typedef DomainWallFermion<WilsonImplD> DomainWallFermionD;
-typedef DomainWallFermion<WilsonImplD2> DomainWallFermionD2;

-typedef DomainWallEOFAFermion<WilsonImplD2> DomainWallEOFAFermionD2;
+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 MobiusFermion<WilsonImplD2> MobiusFermionD2;
+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 MobiusEOFAFermion<WilsonImplD2> MobiusEOFAFermionD2;
+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 ZMobiusFermion<ZWilsonImplD2> ZMobiusFermionD2;
+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 ScaledShamirFermion<WilsonImplD2> ScaledShamirFermionD2;
+typedef ZMobiusFermion<ZWilsonImplRL> ZMobiusFermionRL;
+typedef ZMobiusFermion<ZWilsonImplFH> ZMobiusFermionFH;
+typedef ZMobiusFermion<ZWilsonImplDF> ZMobiusFermionDF;
+
+// Ls vectorised
+typedef ScaledShamirFermion<WilsonImplR> ScaledShamirFermionR;
 typedef ScaledShamirFermion<WilsonImplF> ScaledShamirFermionF;
 typedef ScaledShamirFermion<WilsonImplD> ScaledShamirFermionD;

-typedef MobiusZolotarevFermion<WilsonImplD2> MobiusZolotarevFermionD2;
+typedef MobiusZolotarevFermion<WilsonImplR> MobiusZolotarevFermionR;
 typedef MobiusZolotarevFermion<WilsonImplF> MobiusZolotarevFermionF;
 typedef MobiusZolotarevFermion<WilsonImplD> MobiusZolotarevFermionD;
-typedef ShamirZolotarevFermion<WilsonImplD2> ShamirZolotarevFermionD2;
+typedef ShamirZolotarevFermion<WilsonImplR> ShamirZolotarevFermionR;
 typedef ShamirZolotarevFermion<WilsonImplF> ShamirZolotarevFermionF;
 typedef ShamirZolotarevFermion<WilsonImplD> ShamirZolotarevFermionD;

-typedef OverlapWilsonCayleyTanhFermion<WilsonImplD2> OverlapWilsonCayleyTanhFermionD2;
+typedef OverlapWilsonCayleyTanhFermion<WilsonImplR> OverlapWilsonCayleyTanhFermionR;
 typedef OverlapWilsonCayleyTanhFermion<WilsonImplF> OverlapWilsonCayleyTanhFermionF;
 typedef OverlapWilsonCayleyTanhFermion<WilsonImplD> OverlapWilsonCayleyTanhFermionD;
-typedef OverlapWilsonCayleyZolotarevFermion<WilsonImplD2> OverlapWilsonCayleyZolotarevFermionD2;
+typedef OverlapWilsonCayleyZolotarevFermion<WilsonImplR> OverlapWilsonCayleyZolotarevFermionR;
 typedef OverlapWilsonCayleyZolotarevFermion<WilsonImplF> OverlapWilsonCayleyZolotarevFermionF;
 typedef OverlapWilsonCayleyZolotarevFermion<WilsonImplD> OverlapWilsonCayleyZolotarevFermionD;

 // Continued fraction
-typedef OverlapWilsonContFracTanhFermion<WilsonImplD2> OverlapWilsonContFracTanhFermionD2;
+typedef OverlapWilsonContFracTanhFermion<WilsonImplR> OverlapWilsonContFracTanhFermionR;
 typedef OverlapWilsonContFracTanhFermion<WilsonImplF> OverlapWilsonContFracTanhFermionF;
 typedef OverlapWilsonContFracTanhFermion<WilsonImplD> OverlapWilsonContFracTanhFermionD;
-typedef OverlapWilsonContFracZolotarevFermion<WilsonImplD2> OverlapWilsonContFracZolotarevFermionD2;
+typedef OverlapWilsonContFracZolotarevFermion<WilsonImplR> OverlapWilsonContFracZolotarevFermionR;
 typedef OverlapWilsonContFracZolotarevFermion<WilsonImplF> OverlapWilsonContFracZolotarevFermionF;
 typedef OverlapWilsonContFracZolotarevFermion<WilsonImplD> OverlapWilsonContFracZolotarevFermionD;

 // Partial fraction
-typedef OverlapWilsonPartialFractionTanhFermion<WilsonImplD2> OverlapWilsonPartialFractionTanhFermionD2;
+typedef OverlapWilsonPartialFractionTanhFermion<WilsonImplR> OverlapWilsonPartialFractionTanhFermionR;
 typedef OverlapWilsonPartialFractionTanhFermion<WilsonImplF> OverlapWilsonPartialFractionTanhFermionF;
 typedef OverlapWilsonPartialFractionTanhFermion<WilsonImplD> OverlapWilsonPartialFractionTanhFermionD;

-typedef OverlapWilsonPartialFractionZolotarevFermion<WilsonImplD2> OverlapWilsonPartialFractionZolotarevFermionD2;
+typedef OverlapWilsonPartialFractionZolotarevFermion<WilsonImplR> OverlapWilsonPartialFractionZolotarevFermionR;
 typedef OverlapWilsonPartialFractionZolotarevFermion<WilsonImplF> OverlapWilsonPartialFractionZolotarevFermionF;
 typedef OverlapWilsonPartialFractionZolotarevFermion<WilsonImplD> OverlapWilsonPartialFractionZolotarevFermionD;

 // Gparity cases; partial list until tested
+typedef WilsonFermion<GparityWilsonImplR>     GparityWilsonFermionR;
 typedef WilsonFermion<GparityWilsonImplF>     GparityWilsonFermionF;
 typedef WilsonFermion<GparityWilsonImplD>     GparityWilsonFermionD;

+typedef WilsonFermion<GparityWilsonImplRL>     GparityWilsonFermionRL;
+typedef WilsonFermion<GparityWilsonImplFH>     GparityWilsonFermionFH;
+typedef WilsonFermion<GparityWilsonImplDF>     GparityWilsonFermionDF;
+
+typedef DomainWallFermion<GparityWilsonImplR> GparityDomainWallFermionR;
 typedef DomainWallFermion<GparityWilsonImplF> GparityDomainWallFermionF;
 typedef DomainWallFermion<GparityWilsonImplD> GparityDomainWallFermionD;

-typedef DomainWallEOFAFermion<GparityWilsonImplR> GparityDomainWallEOFAFermionD2;
+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 WilsonTMFermion<GparityWilsonImplR> GparityWilsonTMFermionD2;
+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 MobiusFermion<GparityWilsonImplR> GparityMobiusFermionD2;
+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 MobiusEOFAFermion<GparityWilsonImplR> GparityMobiusEOFAFermionD2;
+typedef MobiusFermion<GparityWilsonImplRL> GparityMobiusFermionRL;
+typedef MobiusFermion<GparityWilsonImplFH> GparityMobiusFermionFH;
+typedef MobiusFermion<GparityWilsonImplDF> GparityMobiusFermionDF;
+
+typedef MobiusEOFAFermion<GparityWilsonImplR> GparityMobiusEOFAFermionR;
 typedef MobiusEOFAFermion<GparityWilsonImplF> GparityMobiusEOFAFermionF;
 typedef MobiusEOFAFermion<GparityWilsonImplD> GparityMobiusEOFAFermionD;

+typedef MobiusEOFAFermion<GparityWilsonImplRL> GparityMobiusEOFAFermionRL;
+typedef MobiusEOFAFermion<GparityWilsonImplFH> GparityMobiusEOFAFermionFH;
+typedef MobiusEOFAFermion<GparityWilsonImplDF> GparityMobiusEOFAFermionDF;
+
+typedef ImprovedStaggeredFermion<StaggeredImplR> ImprovedStaggeredFermionR;
 typedef ImprovedStaggeredFermion<StaggeredImplF> ImprovedStaggeredFermionF;
 typedef ImprovedStaggeredFermion<StaggeredImplD> ImprovedStaggeredFermionD;

+typedef NaiveStaggeredFermion<StaggeredImplR> NaiveStaggeredFermionR;
 typedef NaiveStaggeredFermion<StaggeredImplF> NaiveStaggeredFermionF;
 typedef NaiveStaggeredFermion<StaggeredImplD> NaiveStaggeredFermionD;

+typedef ImprovedStaggeredFermion5D<StaggeredImplR> ImprovedStaggeredFermion5DR;
 typedef ImprovedStaggeredFermion5D<StaggeredImplF> ImprovedStaggeredFermion5DF;
 typedef ImprovedStaggeredFermion5D<StaggeredImplD> ImprovedStaggeredFermion5DD;

@@ -49,8 +49,6 @@ public:

  virtual FermionField &tmp(void) = 0;

-  virtual void DirichletBlock(const Coordinate & _Block) { assert(0); };
-  
  GridBase * Grid(void)   { return FermionGrid(); };   // this is all the linalg routines need to know
  GridBase * RedBlackGrid(void) { return FermionRedBlackGrid(); };

@@ -30,18 +30,6 @@ directory

 NAMESPACE_BEGIN(Grid);

-/*
-  Policy implementation for G-parity boundary conditions
-
-  Rather than treating the gauge field as a flavored field, the Grid implementation of G-parity treats the gauge field as a regular
-  field with complex conjugate boundary conditions. In order to ensure the second flavor interacts with the conjugate links and the first
-  with the regular links we overload the functionality of doubleStore, whose purpose is to store the gauge field and the barrel-shifted gauge field
-  to avoid communicating links when applying the Dirac operator, such that the double-stored field contains also a flavor index which maps to
-  either the link or the conjugate link. This flavored field is then used by multLink to apply the correct link to a spinor.
-
-  Here the first Nd-1 directions are treated as "spatial", and a twist value of 1 indicates G-parity BCs in that direction. 
-  mu=Nd-1 is assumed to be the time direction and a twist value of 1 indicates antiperiodic BCs
- */
 template <class S, class Representation = FundamentalRepresentation, class Options=CoeffReal>
 class GparityWilsonImpl : public ConjugateGaugeImpl<GaugeImplTypes<S, Representation::Dimension> > {
 public:
@@ -125,7 +113,7 @@ public:
    || ((distance== 1)&&(icoor[direction]==1))
    || ((distance==-1)&&(icoor[direction]==0));

-    permute_lane = permute_lane && SE->_around_the_world && St.parameters.twists[mmu] && mmu < Nd-1; //only if we are going around the world in a spatial direction
+    permute_lane = permute_lane && SE->_around_the_world && St.parameters.twists[mmu]; //only if we are going around the world

    //Apply the links
    int f_upper = permute_lane ? 1 : 0;
@@ -151,10 +139,10 @@ public:
    assert((distance == 1) || (distance == -1));  // nearest neighbour stencil hard code
    assert((sl == 1) || (sl == 2));

-    //If this site is an global boundary site, perform the G-parity flavor twist
-    if ( mmu < Nd-1 && SE->_around_the_world && St.parameters.twists[mmu] ) {
+    if ( SE->_around_the_world && St.parameters.twists[mmu] ) {
+
      if ( sl == 2 ) {
-	//Only do the twist for lanes on the edge of the physical node
+       
 	ExtractBuffer<sobj> vals(Nsimd);

 	extract(chi,vals);
@@ -209,19 +197,6 @@ public:
    reg = memory;
  }

-
-  //Poke 'poke_f0' onto flavor 0 and 'poke_f1' onto flavor 1 in direction mu of the doubled gauge field Uds
-  inline void pokeGparityDoubledGaugeField(DoubledGaugeField &Uds, const GaugeLinkField &poke_f0, const GaugeLinkField &poke_f1, const int mu){
-    autoView(poke_f0_v, poke_f0, CpuRead);
-    autoView(poke_f1_v, poke_f1, CpuRead);
-    autoView(Uds_v, Uds, CpuWrite);
-    thread_foreach(ss,poke_f0_v,{
-	Uds_v[ss](0)(mu) = poke_f0_v[ss]();
-	Uds_v[ss](1)(mu) = poke_f1_v[ss]();
-      });
-  }
-    
-
  inline void DoubleStore(GridBase *GaugeGrid,DoubledGaugeField &Uds,const GaugeField &Umu)
  {
    conformable(Uds.Grid(),GaugeGrid);
@@ -232,19 +207,14 @@ public:
    GaugeLinkField Uconj(GaugeGrid);
   
    Lattice<iScalar<vInteger> > coor(GaugeGrid);
-
-    //Here the first Nd-1 directions are treated as "spatial", and a twist value of 1 indicates G-parity BCs in that direction. 
-    //mu=Nd-1 is assumed to be the time direction and a twist value of 1 indicates antiperiodic BCs        
-    for(int mu=0;mu<Nd-1;mu++){
-
-      if( Params.twists[mu] ){
-	LatticeCoordinate(coor,mu);
-      }
+        
+    for(int mu=0;mu<Nd;mu++){
+          
+      LatticeCoordinate(coor,mu);
          
      U     = PeekIndex<LorentzIndex>(Umu,mu);
      Uconj = conjugate(U);
     
-      // Implement the isospin rotation sign on the boundary between f=1 and f=0
      // This phase could come from a simple bc 1,1,-1,1 ..
      int neglink = GaugeGrid->GlobalDimensions()[mu]-1;
      if ( Params.twists[mu] ) { 
@@ -259,7 +229,7 @@ public:
 	thread_foreach(ss,U_v,{
 	    Uds_v[ss](0)(mu) = U_v[ss]();
 	    Uds_v[ss](1)(mu) = Uconj_v[ss]();
-	});
+	  });
      }
          
      U     = adj(Cshift(U    ,mu,-1));      // correct except for spanning the boundary
@@ -290,38 +260,6 @@ public:
        });
      }
    }
-
-    { //periodic / antiperiodic temporal BCs
-      int mu = Nd-1;
-      int L   = GaugeGrid->GlobalDimensions()[mu];
-      int Lmu = L - 1;
-
-      LatticeCoordinate(coor, mu);
-
-      U = PeekIndex<LorentzIndex>(Umu, mu); //Get t-directed links
-      
-      GaugeLinkField *Upoke = &U;
-
-      if(Params.twists[mu]){ //antiperiodic
-	Utmp =  where(coor == Lmu, -U, U);
-	Upoke = &Utmp;
-      }
-    
-      Uconj = conjugate(*Upoke); //second flavor interacts with conjugate links      
-      pokeGparityDoubledGaugeField(Uds, *Upoke, Uconj, mu);
-
-      //Get the barrel-shifted field
-      Utmp = adj(Cshift(U, mu, -1)); //is a forward shift!
-      Upoke = &Utmp;
-
-      if(Params.twists[mu]){
-	U = where(coor == 0, -Utmp, Utmp);  //boundary phase
-	Upoke = &U;
-      }
-      
-      Uconj = conjugate(*Upoke);
-      pokeGparityDoubledGaugeField(Uds, *Upoke, Uconj, mu + 4);
-    }
  }
      
  inline void InsertForce4D(GaugeField &mat, FermionField &Btilde, FermionField &A, int mu) {
@@ -360,48 +298,28 @@ public:
  inline void extractLinkField(std::vector<GaugeLinkField> &mat, DoubledGaugeField &Uds){
    assert(0);
  }
- 
-  inline void InsertForce5D(GaugeField &mat, FermionField &Btilde, FermionField &Atilde, int mu) {
-    int Ls=Btilde.Grid()->_fdimensions[0];
-    
-    {
-      GridBase *GaugeGrid = mat.Grid();
-      Lattice<iScalar<vInteger> > coor(GaugeGrid);
-
-      if( Params.twists[mu] ){
-	LatticeCoordinate(coor,mu);
-      }
-
-      autoView( mat_v , mat, AcceleratorWrite);
-      autoView( Btilde_v , Btilde, AcceleratorRead);
-      autoView( Atilde_v , Atilde, AcceleratorRead);
-      accelerator_for(sss,mat.Grid()->oSites(), FermionField::vector_type::Nsimd(),{	  
-  	  int sU=sss;
-  	  typedef decltype(coalescedRead(mat_v[sU](mu)() )) ColorMatrixType;
-  	  ColorMatrixType sum;
-  	  zeroit(sum);
-  	  for(int s=0;s<Ls;s++){
-  	    int sF = s+Ls*sU;
-  	    for(int spn=0;spn<Ns;spn++){ //sum over spin
-	      //Flavor 0
-  	      auto bb = coalescedRead(Btilde_v[sF](0)(spn) ); //color vector
-  	      auto aa = coalescedRead(Atilde_v[sF](0)(spn) );
-  	      sum = sum + outerProduct(bb,aa);
-
-  	      //Flavor 1
-  	      bb = coalescedRead(Btilde_v[sF](1)(spn) );
-  	      aa = coalescedRead(Atilde_v[sF](1)(spn) );
-  	      sum = sum + conjugate(outerProduct(bb,aa));
-  	    }
-  	  }	    
-  	  coalescedWrite(mat_v[sU](mu)(), sum);
-  	});
-    }
-  }
-
-
  
+  inline void InsertForce5D(GaugeField &mat, FermionField &Btilde, FermionField &Atilde, int mu) {

+    int Ls = Btilde.Grid()->_fdimensions[0];
+        
+    GaugeLinkField tmp(mat.Grid());
+    tmp = Zero();
+    {
+      autoView( tmp_v , tmp, CpuWrite);
+      autoView( Atilde_v , Atilde, CpuRead);
+      autoView( Btilde_v , Btilde, CpuRead);
+      thread_for(ss,tmp.Grid()->oSites(),{
+	  for (int s = 0; s < Ls; s++) {
+	    int sF = s + Ls * ss;
+	    auto ttmp = traceIndex<SpinIndex>(outerProduct(Btilde_v[sF], Atilde_v[sF]));
+	    tmp_v[ss]() = tmp_v[ss]() + ttmp(0, 0) + conjugate(ttmp(1, 1));
+	  }
+	});
+    }
+    PokeIndex<LorentzIndex>(mat, tmp, mu);
+    return;
+  }
  
 };

@@ -409,8 +327,8 @@ typedef GparityWilsonImpl<vComplex , FundamentalRepresentation,CoeffReal> Gparit
 typedef GparityWilsonImpl<vComplexF, FundamentalRepresentation,CoeffReal> GparityWilsonImplF;  // Float
 typedef GparityWilsonImpl<vComplexD, FundamentalRepresentation,CoeffReal> GparityWilsonImplD;  // Double
 
-//typedef GparityWilsonImpl<vComplex , FundamentalRepresentation,CoeffRealHalfComms> GparityWilsonImplRL;  // Real.. whichever prec
-//typedef GparityWilsonImpl<vComplexF, FundamentalRepresentation,CoeffRealHalfComms> GparityWilsonImplFH;  // Float
-//typedef GparityWilsonImpl<vComplexD, FundamentalRepresentation,CoeffRealHalfComms> GparityWilsonImplDF;  // Double
+typedef GparityWilsonImpl<vComplex , FundamentalRepresentation,CoeffRealHalfComms> GparityWilsonImplRL;  // Real.. whichever prec
+typedef GparityWilsonImpl<vComplexF, FundamentalRepresentation,CoeffRealHalfComms> GparityWilsonImplFH;  // Float
+typedef GparityWilsonImpl<vComplexD, FundamentalRepresentation,CoeffRealHalfComms> GparityWilsonImplDF;  // Double

 NAMESPACE_END(Grid);
@@ -47,6 +47,18 @@ public:
  FermionField _tmp;
  FermionField &tmp(void) { return _tmp; }

+  ////////////////////////////////////////
+  // Performance monitoring
+  ////////////////////////////////////////
+  void Report(void);
+  void ZeroCounters(void);
+  double DhopTotalTime;
+  double DhopCalls;
+  double DhopCommTime;
+  double DhopComputeTime;
+  double DhopComputeTime2;
+  double DhopFaceTime;
+
  ///////////////////////////////////////////////////////////////
  // Implement the abstract base
  ///////////////////////////////////////////////////////////////
@@ -52,6 +52,18 @@ public:
  FermionField _tmp;
  FermionField &tmp(void) { return _tmp; }

+  ////////////////////////////////////////
+  // Performance monitoring
+  ////////////////////////////////////////
+  void Report(void);
+  void ZeroCounters(void);
+  double DhopTotalTime;
+  double DhopCalls;
+  double DhopCommTime;
+  double DhopComputeTime;
+  double DhopComputeTime2;
+  double DhopFaceTime;
+
  ///////////////////////////////////////////////////////////////
  // Implement the abstract base
  ///////////////////////////////////////////////////////////////
@@ -47,6 +47,18 @@ public:
  FermionField _tmp;
  FermionField &tmp(void) { return _tmp; }

+  ////////////////////////////////////////
+  // Performance monitoring
+  ////////////////////////////////////////
+  void Report(void);
+  void ZeroCounters(void);
+  double DhopTotalTime;
+  double DhopCalls;
+  double DhopCommTime;
+  double DhopComputeTime;
+  double DhopComputeTime2;
+  double DhopFaceTime;
+
  ///////////////////////////////////////////////////////////////
  // Implement the abstract base
  ///////////////////////////////////////////////////////////////
@@ -4,11 +4,10 @@

    Source file: ./lib/qcd/action/fermion/WilsonCloverFermion.h

-    Copyright (C) 2017 - 2022
+    Copyright (C) 2017

    Author: Guido Cossu <guido.cossu@ed.ac.uk>
    Author: David Preti <>
-    Author: Daniel Richtmann <daniel.richtmann@gmail.com>

    This program is free software; you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
@@ -30,9 +29,7 @@

 #pragma once

-#include <Grid/qcd/action/fermion/WilsonCloverTypes.h>
-#include <Grid/qcd/action/fermion/WilsonCloverHelpers.h>
-#include <Grid/qcd/action/fermion/CloverHelpers.h>
+#include <Grid/Grid.h>

 NAMESPACE_BEGIN(Grid);

@@ -52,16 +49,19 @@ NAMESPACE_BEGIN(Grid);
 // csw_r = csw_t to recover the isotropic version
 //////////////////////////////////////////////////////////////////

-template<class Impl, class CloverHelpers>
-class WilsonCloverFermion : public WilsonFermion<Impl>,
-                            public WilsonCloverHelpers<Impl>
+template <class Impl>
+class WilsonCloverFermion : public WilsonFermion<Impl>
 {
 public:
+  // Types definitions
  INHERIT_IMPL_TYPES(Impl);
-  INHERIT_CLOVER_TYPES(Impl);
+  template <typename vtype>
+  using iImplClover = iScalar<iMatrix<iMatrix<vtype, Impl::Dimension>, Ns>>;
+  typedef iImplClover<Simd> SiteCloverType;
+  typedef Lattice<SiteCloverType> CloverFieldType;

-  typedef WilsonFermion<Impl>       WilsonBase;
-  typedef WilsonCloverHelpers<Impl> Helpers;
+public:
+  typedef WilsonFermion<Impl> WilsonBase;

  virtual int    ConstEE(void)     { return 0; };
  virtual void Instantiatable(void){};
@@ -72,7 +72,42 @@ public:
                      const RealD _csw_r = 0.0,
                      const RealD _csw_t = 0.0,
                      const WilsonAnisotropyCoefficients &clover_anisotropy = WilsonAnisotropyCoefficients(),
-                      const ImplParams &impl_p = ImplParams());
+                      const ImplParams &impl_p = ImplParams()) : WilsonFermion<Impl>(_Umu,
+                                                                                     Fgrid,
+                                                                                     Hgrid,
+                                                                                     _mass, impl_p, clover_anisotropy),
+                                                                 CloverTerm(&Fgrid),
+                                                                 CloverTermInv(&Fgrid),
+                                                                 CloverTermEven(&Hgrid),
+                                                                 CloverTermOdd(&Hgrid),
+                                                                 CloverTermInvEven(&Hgrid),
+                                                                 CloverTermInvOdd(&Hgrid),
+                                                                 CloverTermDagEven(&Hgrid),
+                                                                 CloverTermDagOdd(&Hgrid),
+                                                                 CloverTermInvDagEven(&Hgrid),
+                                                                 CloverTermInvDagOdd(&Hgrid)
+  {
+    assert(Nd == 4); // require 4 dimensions
+
+    if (clover_anisotropy.isAnisotropic)
+    {
+      csw_r = _csw_r * 0.5 / clover_anisotropy.xi_0;
+      diag_mass = _mass + 1.0 + (Nd - 1) * (clover_anisotropy.nu / clover_anisotropy.xi_0);
+    }
+    else
+    {
+      csw_r = _csw_r * 0.5;
+      diag_mass = 4.0 + _mass;
+    }
+    csw_t = _csw_t * 0.5;
+
+    if (csw_r == 0)
+      std::cout << GridLogWarning << "Initializing WilsonCloverFermion with csw_r = 0" << std::endl;
+    if (csw_t == 0)
+      std::cout << GridLogWarning << "Initializing WilsonCloverFermion with csw_t = 0" << std::endl;
+
+    ImportGauge(_Umu);
+  }

  virtual void M(const FermionField &in, FermionField &out);
  virtual void Mdag(const FermionField &in, FermionField &out);
@@ -89,21 +124,250 @@ public:
  void ImportGauge(const GaugeField &_Umu);

  // Derivative parts unpreconditioned pseudofermions
-  void MDeriv(GaugeField &force, const FermionField &X, const FermionField &Y, int dag);
+  void MDeriv(GaugeField &force, const FermionField &X, const FermionField &Y, int dag)
+  {
+    conformable(X.Grid(), Y.Grid());
+    conformable(X.Grid(), force.Grid());
+    GaugeLinkField force_mu(force.Grid()), lambda(force.Grid());
+    GaugeField clover_force(force.Grid());
+    PropagatorField Lambda(force.Grid());

-public:
+    // Guido: Here we are hitting some performance issues:
+    // need to extract the components of the DoubledGaugeField
+    // for each call
+    // Possible solution
+    // Create a vector object to store them? (cons: wasting space)
+    std::vector<GaugeLinkField> U(Nd, this->Umu.Grid());
+
+    Impl::extractLinkField(U, this->Umu);
+
+    force = Zero();
+    // Derivative of the Wilson hopping term
+    this->DhopDeriv(force, X, Y, dag);
+
+    ///////////////////////////////////////////////////////////
+    // Clover term derivative
+    ///////////////////////////////////////////////////////////
+    Impl::outerProductImpl(Lambda, X, Y);
+    //std::cout << "Lambda:" << Lambda << std::endl;
+
+    Gamma::Algebra sigma[] = {
+        Gamma::Algebra::SigmaXY,
+        Gamma::Algebra::SigmaXZ,
+        Gamma::Algebra::SigmaXT,
+        Gamma::Algebra::MinusSigmaXY,
+        Gamma::Algebra::SigmaYZ,
+        Gamma::Algebra::SigmaYT,
+        Gamma::Algebra::MinusSigmaXZ,
+        Gamma::Algebra::MinusSigmaYZ,
+        Gamma::Algebra::SigmaZT,
+        Gamma::Algebra::MinusSigmaXT,
+        Gamma::Algebra::MinusSigmaYT,
+        Gamma::Algebra::MinusSigmaZT};
+
+    /*
+      sigma_{\mu \nu}=
+      | 0         sigma[0]  sigma[1]  sigma[2] |
+      | sigma[3]    0       sigma[4]  sigma[5] |
+      | sigma[6]  sigma[7]     0      sigma[8] |
+      | sigma[9]  sigma[10] sigma[11]   0      |
+    */
+
+    int count = 0;
+    clover_force = Zero();
+    for (int mu = 0; mu < 4; mu++)
+    {
+      force_mu = Zero();
+      for (int nu = 0; nu < 4; nu++)
+      {
+        if (mu == nu)
+        continue;
+        
+        RealD factor;
+        if (nu == 4 || mu == 4)
+        {
+          factor = 2.0 * csw_t;
+        }
+        else
+        {
+          factor = 2.0 * csw_r;
+        }
+        PropagatorField Slambda = Gamma(sigma[count]) * Lambda; // sigma checked
+        Impl::TraceSpinImpl(lambda, Slambda);                   // traceSpin ok
+        force_mu -= factor*Cmunu(U, lambda, mu, nu);                   // checked
+        count++;
+      }
+
+      pokeLorentz(clover_force, U[mu] * force_mu, mu);
+    }
+    //clover_force *= csw;
+    force += clover_force;
+  }
+
+  // Computing C_{\mu \nu}(x) as in Eq.(B.39) in Zbigniew Sroczynski's PhD thesis
+  GaugeLinkField Cmunu(std::vector<GaugeLinkField> &U, GaugeLinkField &lambda, int mu, int nu)
+  {
+    conformable(lambda.Grid(), U[0].Grid());
+    GaugeLinkField out(lambda.Grid()), tmp(lambda.Grid());
+    // insertion in upper staple
+    // please check redundancy of shift operations
+
+    // C1+
+    tmp = lambda * U[nu];
+    out = Impl::ShiftStaple(Impl::CovShiftForward(tmp, nu, Impl::CovShiftBackward(U[mu], mu, Impl::CovShiftIdentityBackward(U[nu], nu))), mu);
+
+    // C2+
+    tmp = U[mu] * Impl::ShiftStaple(adj(lambda), mu);
+    out += Impl::ShiftStaple(Impl::CovShiftForward(U[nu], nu, Impl::CovShiftBackward(tmp, mu, Impl::CovShiftIdentityBackward(U[nu], nu))), mu);
+
+    // C3+
+    tmp = U[nu] * Impl::ShiftStaple(adj(lambda), nu);
+    out += Impl::ShiftStaple(Impl::CovShiftForward(U[nu], nu, Impl::CovShiftBackward(U[mu], mu, Impl::CovShiftIdentityBackward(tmp, nu))), mu);
+
+    // C4+
+    out += Impl::ShiftStaple(Impl::CovShiftForward(U[nu], nu, Impl::CovShiftBackward(U[mu], mu, Impl::CovShiftIdentityBackward(U[nu], nu))), mu) * lambda;
+
+    // insertion in lower staple
+    // C1-
+    out -= Impl::ShiftStaple(lambda, mu) * Impl::ShiftStaple(Impl::CovShiftBackward(U[nu], nu, Impl::CovShiftBackward(U[mu], mu, U[nu])), mu);
+
+    // C2-
+    tmp = adj(lambda) * U[nu];
+    out -= Impl::ShiftStaple(Impl::CovShiftBackward(tmp, nu, Impl::CovShiftBackward(U[mu], mu, U[nu])), mu);
+
+    // C3-
+    tmp = lambda * U[nu];
+    out -= Impl::ShiftStaple(Impl::CovShiftBackward(U[nu], nu, Impl::CovShiftBackward(U[mu], mu, tmp)), mu);
+
+    // C4-
+    out -= Impl::ShiftStaple(Impl::CovShiftBackward(U[nu], nu, Impl::CovShiftBackward(U[mu], mu, U[nu])), mu) * lambda;
+
+    return out;
+  }
+
+protected:
  // here fixing the 4 dimensions, make it more general?

  RealD csw_r;                                               // Clover coefficient - spatial
  RealD csw_t;                                               // Clover coefficient - temporal
  RealD diag_mass;                                           // Mass term
-  CloverField CloverTerm, CloverTermInv;                     // Clover term
-  CloverField CloverTermEven, CloverTermOdd;                 // Clover term EO
-  CloverField CloverTermInvEven, CloverTermInvOdd;           // Clover term Inv EO
-  CloverField CloverTermDagEven, CloverTermDagOdd;           // Clover term Dag EO
-  CloverField CloverTermInvDagEven, CloverTermInvDagOdd;     // Clover term Inv Dag EO
-};
+  CloverFieldType CloverTerm, CloverTermInv;                 // Clover term
+  CloverFieldType CloverTermEven, CloverTermOdd;             // Clover term EO
+  CloverFieldType CloverTermInvEven, CloverTermInvOdd;       // Clover term Inv EO
+  CloverFieldType CloverTermDagEven, CloverTermDagOdd;       // Clover term Dag EO
+  CloverFieldType CloverTermInvDagEven, CloverTermInvDagOdd; // Clover term Inv Dag EO

+ public:
+  // eventually these can be compressed into 6x6 blocks instead of the 12x12
+  // using the DeGrand-Rossi basis for the gamma matrices
+  CloverFieldType fillCloverYZ(const GaugeLinkField &F)
+  {
+    CloverFieldType T(F.Grid());
+    T = Zero();
+    autoView(T_v,T,AcceleratorWrite);
+    autoView(F_v,F,AcceleratorRead);
+    accelerator_for(i, CloverTerm.Grid()->oSites(),1,
+    {
+      T_v[i]()(0, 1) = timesMinusI(F_v[i]()());
+      T_v[i]()(1, 0) = timesMinusI(F_v[i]()());
+      T_v[i]()(2, 3) = timesMinusI(F_v[i]()());
+      T_v[i]()(3, 2) = timesMinusI(F_v[i]()());
+    });
+
+    return T;
+  }
+
+  CloverFieldType fillCloverXZ(const GaugeLinkField &F)
+  {
+    CloverFieldType T(F.Grid());
+    T = Zero();
+    
+    autoView(T_v, T,AcceleratorWrite);
+    autoView(F_v, F,AcceleratorRead);
+    accelerator_for(i, CloverTerm.Grid()->oSites(),1,
+    {
+      T_v[i]()(0, 1) = -F_v[i]()();
+      T_v[i]()(1, 0) = F_v[i]()();
+      T_v[i]()(2, 3) = -F_v[i]()();
+      T_v[i]()(3, 2) = F_v[i]()();
+    });
+
+    return T;
+  }
+
+  CloverFieldType fillCloverXY(const GaugeLinkField &F)
+  {
+    CloverFieldType T(F.Grid());
+    T = Zero();
+
+    autoView(T_v,T,AcceleratorWrite);
+    autoView(F_v,F,AcceleratorRead);
+    accelerator_for(i, CloverTerm.Grid()->oSites(),1,
+    {
+      T_v[i]()(0, 0) = timesMinusI(F_v[i]()());
+      T_v[i]()(1, 1) = timesI(F_v[i]()());
+      T_v[i]()(2, 2) = timesMinusI(F_v[i]()());
+      T_v[i]()(3, 3) = timesI(F_v[i]()());
+    });
+
+    return T;
+  }
+
+  CloverFieldType fillCloverXT(const GaugeLinkField &F)
+  {
+    CloverFieldType T(F.Grid());
+    T = Zero();
+
+    autoView( T_v , T, AcceleratorWrite);
+    autoView( F_v , F, AcceleratorRead);
+    accelerator_for(i, CloverTerm.Grid()->oSites(),1,
+    {
+      T_v[i]()(0, 1) = timesI(F_v[i]()());
+      T_v[i]()(1, 0) = timesI(F_v[i]()());
+      T_v[i]()(2, 3) = timesMinusI(F_v[i]()());
+      T_v[i]()(3, 2) = timesMinusI(F_v[i]()());
+    });
+
+    return T;
+  }
+
+  CloverFieldType fillCloverYT(const GaugeLinkField &F)
+  {
+    CloverFieldType T(F.Grid());
+    T = Zero();
+    
+    autoView( T_v ,T,AcceleratorWrite);
+    autoView( F_v ,F,AcceleratorRead);
+    accelerator_for(i, CloverTerm.Grid()->oSites(),1,
+    {
+      T_v[i]()(0, 1) = -(F_v[i]()());
+      T_v[i]()(1, 0) = (F_v[i]()());
+      T_v[i]()(2, 3) = (F_v[i]()());
+      T_v[i]()(3, 2) = -(F_v[i]()());
+    });
+
+    return T;
+  }
+
+  CloverFieldType fillCloverZT(const GaugeLinkField &F)
+  {
+    CloverFieldType T(F.Grid());
+
+    T = Zero();
+
+    autoView( T_v , T,AcceleratorWrite);
+    autoView( F_v , F,AcceleratorRead);
+    accelerator_for(i, CloverTerm.Grid()->oSites(),1,
+    {
+      T_v[i]()(0, 0) = timesI(F_v[i]()());
+      T_v[i]()(1, 1) = timesMinusI(F_v[i]()());
+      T_v[i]()(2, 2) = timesMinusI(F_v[i]()());
+      T_v[i]()(3, 3) = timesI(F_v[i]()());
+    });
+
+    return T;
+  }
+};
 NAMESPACE_END(Grid);


@@ -1,763 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: ./lib/qcd/action/fermion/WilsonCloverHelpers.h
-
-    Copyright (C) 2021 - 2022
-
-    Author: Daniel Richtmann <daniel.richtmann@gmail.com>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-
-#pragma once
-
-// Helper routines that implement common clover functionality
-
-NAMESPACE_BEGIN(Grid);
-
-template<class Impl> class WilsonCloverHelpers {
-public:
-
-  INHERIT_IMPL_TYPES(Impl);
-  INHERIT_CLOVER_TYPES(Impl);
-
-  // Computing C_{\mu \nu}(x) as in Eq.(B.39) in Zbigniew Sroczynski's PhD thesis
-  static GaugeLinkField Cmunu(std::vector<GaugeLinkField> &U, GaugeLinkField &lambda, int mu, int nu)
-  {
-    conformable(lambda.Grid(), U[0].Grid());
-    GaugeLinkField out(lambda.Grid()), tmp(lambda.Grid());
-    // insertion in upper staple
-    // please check redundancy of shift operations
-
-    // C1+
-    tmp = lambda * U[nu];
-    out = Impl::ShiftStaple(Impl::CovShiftForward(tmp, nu, Impl::CovShiftBackward(U[mu], mu, Impl::CovShiftIdentityBackward(U[nu], nu))), mu);
-
-    // C2+
-    tmp = U[mu] * Impl::ShiftStaple(adj(lambda), mu);
-    out += Impl::ShiftStaple(Impl::CovShiftForward(U[nu], nu, Impl::CovShiftBackward(tmp, mu, Impl::CovShiftIdentityBackward(U[nu], nu))), mu);
-
-    // C3+
-    tmp = U[nu] * Impl::ShiftStaple(adj(lambda), nu);
-    out += Impl::ShiftStaple(Impl::CovShiftForward(U[nu], nu, Impl::CovShiftBackward(U[mu], mu, Impl::CovShiftIdentityBackward(tmp, nu))), mu);
-
-    // C4+
-    out += Impl::ShiftStaple(Impl::CovShiftForward(U[nu], nu, Impl::CovShiftBackward(U[mu], mu, Impl::CovShiftIdentityBackward(U[nu], nu))), mu) * lambda;
-
-    // insertion in lower staple
-    // C1-
-    out -= Impl::ShiftStaple(lambda, mu) * Impl::ShiftStaple(Impl::CovShiftBackward(U[nu], nu, Impl::CovShiftBackward(U[mu], mu, U[nu])), mu);
-
-    // C2-
-    tmp = adj(lambda) * U[nu];
-    out -= Impl::ShiftStaple(Impl::CovShiftBackward(tmp, nu, Impl::CovShiftBackward(U[mu], mu, U[nu])), mu);
-
-    // C3-
-    tmp = lambda * U[nu];
-    out -= Impl::ShiftStaple(Impl::CovShiftBackward(U[nu], nu, Impl::CovShiftBackward(U[mu], mu, tmp)), mu);
-
-    // C4-
-    out -= Impl::ShiftStaple(Impl::CovShiftBackward(U[nu], nu, Impl::CovShiftBackward(U[mu], mu, U[nu])), mu) * lambda;
-
-    return out;
-  }
-
-  static CloverField fillCloverYZ(const GaugeLinkField &F)
-  {
-    CloverField T(F.Grid());
-    T = Zero();
-    autoView(T_v,T,AcceleratorWrite);
-    autoView(F_v,F,AcceleratorRead);
-    accelerator_for(i, T.Grid()->oSites(),CloverField::vector_type::Nsimd(),
-    {
-      coalescedWrite(T_v[i]()(0, 1), coalescedRead(timesMinusI(F_v[i]()())));
-      coalescedWrite(T_v[i]()(1, 0), coalescedRead(timesMinusI(F_v[i]()())));
-      coalescedWrite(T_v[i]()(2, 3), coalescedRead(timesMinusI(F_v[i]()())));
-      coalescedWrite(T_v[i]()(3, 2), coalescedRead(timesMinusI(F_v[i]()())));
-    });
-
-    return T;
-  }
-
-  static CloverField fillCloverXZ(const GaugeLinkField &F)
-  {
-    CloverField T(F.Grid());
-    T = Zero();
-    
-    autoView(T_v, T,AcceleratorWrite);
-    autoView(F_v, F,AcceleratorRead);
-    accelerator_for(i, T.Grid()->oSites(),CloverField::vector_type::Nsimd(),
-    {
-      coalescedWrite(T_v[i]()(0, 1), coalescedRead(-F_v[i]()()));
-      coalescedWrite(T_v[i]()(1, 0), coalescedRead(F_v[i]()()));
-      coalescedWrite(T_v[i]()(2, 3), coalescedRead(-F_v[i]()()));
-      coalescedWrite(T_v[i]()(3, 2), coalescedRead(F_v[i]()()));
-    });
-
-    return T;
-  }
-
-  static CloverField fillCloverXY(const GaugeLinkField &F)
-  {
-    CloverField T(F.Grid());
-    T = Zero();
-
-    autoView(T_v,T,AcceleratorWrite);
-    autoView(F_v,F,AcceleratorRead);
-    accelerator_for(i, T.Grid()->oSites(),CloverField::vector_type::Nsimd(),
-    {
-      coalescedWrite(T_v[i]()(0, 0), coalescedRead(timesMinusI(F_v[i]()())));
-      coalescedWrite(T_v[i]()(1, 1), coalescedRead(timesI(F_v[i]()())));
-      coalescedWrite(T_v[i]()(2, 2), coalescedRead(timesMinusI(F_v[i]()())));
-      coalescedWrite(T_v[i]()(3, 3), coalescedRead(timesI(F_v[i]()())));
-    });
-
-    return T;
-  }
-
-  static CloverField fillCloverXT(const GaugeLinkField &F)
-  {
-    CloverField T(F.Grid());
-    T = Zero();
-
-    autoView( T_v , T, AcceleratorWrite);
-    autoView( F_v , F, AcceleratorRead);
-    accelerator_for(i, T.Grid()->oSites(),CloverField::vector_type::Nsimd(),
-    {
-      coalescedWrite(T_v[i]()(0, 1), coalescedRead(timesI(F_v[i]()())));
-      coalescedWrite(T_v[i]()(1, 0), coalescedRead(timesI(F_v[i]()())));
-      coalescedWrite(T_v[i]()(2, 3), coalescedRead(timesMinusI(F_v[i]()())));
-      coalescedWrite(T_v[i]()(3, 2), coalescedRead(timesMinusI(F_v[i]()())));
-    });
-
-    return T;
-  }
-
-  static CloverField fillCloverYT(const GaugeLinkField &F)
-  {
-    CloverField T(F.Grid());
-    T = Zero();
-    
-    autoView( T_v ,T,AcceleratorWrite);
-    autoView( F_v ,F,AcceleratorRead);
-    accelerator_for(i, T.Grid()->oSites(),CloverField::vector_type::Nsimd(),
-    {
-      coalescedWrite(T_v[i]()(0, 1), coalescedRead(-(F_v[i]()())));
-      coalescedWrite(T_v[i]()(1, 0), coalescedRead((F_v[i]()())));
-      coalescedWrite(T_v[i]()(2, 3), coalescedRead((F_v[i]()())));
-      coalescedWrite(T_v[i]()(3, 2), coalescedRead(-(F_v[i]()())));
-    });
-
-    return T;
-  }
-
-  static CloverField fillCloverZT(const GaugeLinkField &F)
-  {
-    CloverField T(F.Grid());
-
-    T = Zero();
-
-    autoView( T_v , T,AcceleratorWrite);
-    autoView( F_v , F,AcceleratorRead);
-    accelerator_for(i, T.Grid()->oSites(),CloverField::vector_type::Nsimd(),
-    {
-      coalescedWrite(T_v[i]()(0, 0), coalescedRead(timesI(F_v[i]()())));
-      coalescedWrite(T_v[i]()(1, 1), coalescedRead(timesMinusI(F_v[i]()())));
-      coalescedWrite(T_v[i]()(2, 2), coalescedRead(timesMinusI(F_v[i]()())));
-      coalescedWrite(T_v[i]()(3, 3), coalescedRead(timesI(F_v[i]()())));
-    });
-
-    return T;
-  }
-
-  template<class _Spinor>
-  static accelerator_inline void multClover(_Spinor& phi, const SiteClover& C, const _Spinor& chi) {
-    auto CC = coalescedRead(C);
-    mult(&phi, &CC, &chi);
-  }
-
-  template<class _SpinorField>
-  inline void multCloverField(_SpinorField& out, const CloverField& C, const _SpinorField& phi) {
-    const int Nsimd = SiteSpinor::Nsimd();
-    autoView(out_v, out, AcceleratorWrite);
-    autoView(phi_v, phi, AcceleratorRead);
-    autoView(C_v,   C,   AcceleratorRead);
-    typedef decltype(coalescedRead(out_v[0])) calcSpinor;
-    accelerator_for(sss,out.Grid()->oSites(),Nsimd,{
-      calcSpinor tmp;
-      multClover(tmp,C_v[sss],phi_v(sss));
-      coalescedWrite(out_v[sss],tmp);
-    });
-  }
-};
-
-
-////////////////////////////////////////////////////////
-
-template<class Impl> class CompactWilsonCloverHelpers {
-public:
-
-  INHERIT_COMPACT_CLOVER_SIZES(Impl);
-
-  INHERIT_IMPL_TYPES(Impl);
-  INHERIT_CLOVER_TYPES(Impl);
-  INHERIT_COMPACT_CLOVER_TYPES(Impl);
-
-  #if 0
-  static accelerator_inline typename SiteCloverTriangle::vector_type triangle_elem(const SiteCloverTriangle& triangle, int block, int i, int j) {
-    assert(i != j);
-    if(i < j) {
-      return triangle()(block)(triangle_index(i, j));
-    } else { // i > j
-      return conjugate(triangle()(block)(triangle_index(i, j)));
-    }
-  }
-  #else
-  template<typename vobj>
-  static accelerator_inline vobj triangle_elem(const iImplCloverTriangle<vobj>& triangle, int block, int i, int j) {
-    assert(i != j);
-    if(i < j) {
-      return triangle()(block)(triangle_index(i, j));
-    } else { // i > j
-      return conjugate(triangle()(block)(triangle_index(i, j)));
-    }
-  }
-  #endif
-
-  static accelerator_inline int triangle_index(int i, int j) {
-    if(i == j)
-      return 0;
-    else if(i < j)
-      return Nred * (Nred - 1) / 2 - (Nred - i) * (Nred - i - 1) / 2 + j - i - 1;
-    else // i > j
-      return Nred * (Nred - 1) / 2 - (Nred - j) * (Nred - j - 1) / 2 + i - j - 1;
-  }
-
-  static void MooeeKernel_gpu(int                        Nsite,
-                              int                        Ls,
-                              const FermionField&        in,
-                              FermionField&              out,
-                              const CloverDiagonalField& diagonal,
-                              const CloverTriangleField& triangle) {
-    autoView(diagonal_v, diagonal, AcceleratorRead);
-    autoView(triangle_v, triangle, AcceleratorRead);
-    autoView(in_v,       in,       AcceleratorRead);
-    autoView(out_v,      out,      AcceleratorWrite);
-
-    typedef decltype(coalescedRead(out_v[0])) CalcSpinor;
-
-    const uint64_t NN = Nsite * Ls;
-
-    accelerator_for(ss, NN, Simd::Nsimd(), {
-      int sF = ss;
-      int sU = ss/Ls;
-      CalcSpinor res;
-      CalcSpinor in_t = in_v(sF);
-      auto diagonal_t = diagonal_v(sU);
-      auto triangle_t = triangle_v(sU);
-      for(int block=0; block<Nhs; block++) {
-        int s_start = block*Nhs;
-        for(int i=0; i<Nred; i++) {
-          int si = s_start + i/Nc, ci = i%Nc;
-          res()(si)(ci) = diagonal_t()(block)(i) * in_t()(si)(ci);
-          for(int j=0; j<Nred; j++) {
-            if (j == i) continue;
-            int sj = s_start + j/Nc, cj = j%Nc;
-            res()(si)(ci) = res()(si)(ci) + triangle_elem(triangle_t, block, i, j) * in_t()(sj)(cj);
-          };
-        };
-      };
-      coalescedWrite(out_v[sF], res);
-    });
-  }
-
-  static void MooeeKernel_cpu(int                        Nsite,
-                              int                        Ls,
-                              const FermionField&        in,
-                              FermionField&              out,
-                              const CloverDiagonalField& diagonal,
-                              const CloverTriangleField& triangle) {
-    autoView(diagonal_v, diagonal, CpuRead);
-    autoView(triangle_v, triangle, CpuRead);
-    autoView(in_v,       in,       CpuRead);
-    autoView(out_v,      out,      CpuWrite);
-
-    typedef SiteSpinor CalcSpinor;
-
-#if defined(A64FX) || defined(A64FXFIXEDSIZE)
-#define PREFETCH_CLOVER(BASE) {                                     \
-    uint64_t base;                                                  \
-    int pf_dist_L1 = 1;                                             \
-    int pf_dist_L2 = -5; /* -> penalty -> disable */                \
-                                                                    \
-    if ((pf_dist_L1 >= 0) && (sU + pf_dist_L1 < Nsite)) {           \
-      base = (uint64_t)&diag_t()(pf_dist_L1+BASE)(0);               \
-      svprfd(svptrue_b64(), (int64_t*)(base +    0), SV_PLDL1STRM); \
-      svprfd(svptrue_b64(), (int64_t*)(base +  256), SV_PLDL1STRM); \
-      svprfd(svptrue_b64(), (int64_t*)(base +  512), SV_PLDL1STRM); \
-      svprfd(svptrue_b64(), (int64_t*)(base +  768), SV_PLDL1STRM); \
-      svprfd(svptrue_b64(), (int64_t*)(base + 1024), SV_PLDL1STRM); \
-      svprfd(svptrue_b64(), (int64_t*)(base + 1280), SV_PLDL1STRM); \
-    }                                                               \
-                                                                    \
-    if ((pf_dist_L2 >= 0) && (sU + pf_dist_L2 < Nsite)) {           \
-      base = (uint64_t)&diag_t()(pf_dist_L2+BASE)(0);               \
-      svprfd(svptrue_b64(), (int64_t*)(base +    0), SV_PLDL2STRM); \
-      svprfd(svptrue_b64(), (int64_t*)(base +  256), SV_PLDL2STRM); \
-      svprfd(svptrue_b64(), (int64_t*)(base +  512), SV_PLDL2STRM); \
-      svprfd(svptrue_b64(), (int64_t*)(base +  768), SV_PLDL2STRM); \
-      svprfd(svptrue_b64(), (int64_t*)(base + 1024), SV_PLDL2STRM); \
-      svprfd(svptrue_b64(), (int64_t*)(base + 1280), SV_PLDL2STRM); \
-    }                                                               \
-  }
-// TODO: Implement/generalize this for other architectures
-// I played around a bit on KNL (see below) but didn't bring anything
-// #elif defined(AVX512)
-// #define PREFETCH_CLOVER(BASE) {                              \
-//     uint64_t base;                                           \
-//     int pf_dist_L1 = 1;                                      \
-//     int pf_dist_L2 = +4;                                     \
-//                                                              \
-//     if ((pf_dist_L1 >= 0) && (sU + pf_dist_L1 < Nsite)) {    \
-//       base = (uint64_t)&diag_t()(pf_dist_L1+BASE)(0);        \
-//       _mm_prefetch((const char*)(base +    0), _MM_HINT_T0); \
-//       _mm_prefetch((const char*)(base +   64), _MM_HINT_T0); \
-//       _mm_prefetch((const char*)(base +  128), _MM_HINT_T0); \
-//       _mm_prefetch((const char*)(base +  192), _MM_HINT_T0); \
-//       _mm_prefetch((const char*)(base +  256), _MM_HINT_T0); \
-//       _mm_prefetch((const char*)(base +  320), _MM_HINT_T0); \
-//     }                                                        \
-//                                                              \
-//     if ((pf_dist_L2 >= 0) && (sU + pf_dist_L2 < Nsite)) {    \
-//       base = (uint64_t)&diag_t()(pf_dist_L2+BASE)(0);        \
-//       _mm_prefetch((const char*)(base +    0), _MM_HINT_T1); \
-//       _mm_prefetch((const char*)(base +   64), _MM_HINT_T1); \
-//       _mm_prefetch((const char*)(base +  128), _MM_HINT_T1); \
-//       _mm_prefetch((const char*)(base +  192), _MM_HINT_T1); \
-//       _mm_prefetch((const char*)(base +  256), _MM_HINT_T1); \
-//       _mm_prefetch((const char*)(base +  320), _MM_HINT_T1); \
-//     }                                                        \
-//   }
-#else
-#define PREFETCH_CLOVER(BASE)
-#endif
-
-    const uint64_t NN = Nsite * Ls;
-
-    thread_for(ss, NN, {
-      int sF = ss;
-      int sU = ss/Ls;
-      CalcSpinor res;
-      CalcSpinor in_t = in_v[sF];
-      auto diag_t     = diagonal_v[sU]; // "diag" instead of "diagonal" here to make code below easier to read
-      auto triangle_t = triangle_v[sU];
-
-      // upper half
-      PREFETCH_CLOVER(0);
-
-      auto in_cc_0_0 = conjugate(in_t()(0)(0)); // Nils: reduces number
-      auto in_cc_0_1 = conjugate(in_t()(0)(1)); // of conjugates from
-      auto in_cc_0_2 = conjugate(in_t()(0)(2)); // 30 to 20
-      auto in_cc_1_0 = conjugate(in_t()(1)(0));
-      auto in_cc_1_1 = conjugate(in_t()(1)(1));
-
-      res()(0)(0) =               diag_t()(0)( 0) * in_t()(0)(0)
-                  +           triangle_t()(0)( 0) * in_t()(0)(1)
-                  +           triangle_t()(0)( 1) * in_t()(0)(2)
-                  +           triangle_t()(0)( 2) * in_t()(1)(0)
-                  +           triangle_t()(0)( 3) * in_t()(1)(1)
-                  +           triangle_t()(0)( 4) * in_t()(1)(2);
-
-      res()(0)(1) =           triangle_t()(0)( 0) * in_cc_0_0;
-      res()(0)(1) =               diag_t()(0)( 1) * in_t()(0)(1)
-                  +           triangle_t()(0)( 5) * in_t()(0)(2)
-                  +           triangle_t()(0)( 6) * in_t()(1)(0)
-                  +           triangle_t()(0)( 7) * in_t()(1)(1)
-                  +           triangle_t()(0)( 8) * in_t()(1)(2)
-                  + conjugate(       res()(0)( 1));
-
-      res()(0)(2) =           triangle_t()(0)( 1) * in_cc_0_0
-                  +           triangle_t()(0)( 5) * in_cc_0_1;
-      res()(0)(2) =               diag_t()(0)( 2) * in_t()(0)(2)
-                  +           triangle_t()(0)( 9) * in_t()(1)(0)
-                  +           triangle_t()(0)(10) * in_t()(1)(1)
-                  +           triangle_t()(0)(11) * in_t()(1)(2)
-                  + conjugate(       res()(0)( 2));
-
-      res()(1)(0) =           triangle_t()(0)( 2) * in_cc_0_0
-                  +           triangle_t()(0)( 6) * in_cc_0_1
-                  +           triangle_t()(0)( 9) * in_cc_0_2;
-      res()(1)(0) =               diag_t()(0)( 3) * in_t()(1)(0)
-                  +           triangle_t()(0)(12) * in_t()(1)(1)
-                  +           triangle_t()(0)(13) * in_t()(1)(2)
-                  + conjugate(       res()(1)( 0));
-
-      res()(1)(1) =           triangle_t()(0)( 3) * in_cc_0_0
-                  +           triangle_t()(0)( 7) * in_cc_0_1
-                  +           triangle_t()(0)(10) * in_cc_0_2
-                  +           triangle_t()(0)(12) * in_cc_1_0;
-      res()(1)(1) =               diag_t()(0)( 4) * in_t()(1)(1)
-                  +           triangle_t()(0)(14) * in_t()(1)(2)
-                  + conjugate(       res()(1)( 1));
-
-      res()(1)(2) =           triangle_t()(0)( 4) * in_cc_0_0
-                  +           triangle_t()(0)( 8) * in_cc_0_1
-                  +           triangle_t()(0)(11) * in_cc_0_2
-                  +           triangle_t()(0)(13) * in_cc_1_0
-                  +           triangle_t()(0)(14) * in_cc_1_1;
-      res()(1)(2) =               diag_t()(0)( 5) * in_t()(1)(2)
-                  + conjugate(       res()(1)( 2));
-
-      vstream(out_v[sF]()(0)(0), res()(0)(0));
-      vstream(out_v[sF]()(0)(1), res()(0)(1));
-      vstream(out_v[sF]()(0)(2), res()(0)(2));
-      vstream(out_v[sF]()(1)(0), res()(1)(0));
-      vstream(out_v[sF]()(1)(1), res()(1)(1));
-      vstream(out_v[sF]()(1)(2), res()(1)(2));
-
-      // lower half
-      PREFETCH_CLOVER(1);
-
-      auto in_cc_2_0 = conjugate(in_t()(2)(0));
-      auto in_cc_2_1 = conjugate(in_t()(2)(1));
-      auto in_cc_2_2 = conjugate(in_t()(2)(2));
-      auto in_cc_3_0 = conjugate(in_t()(3)(0));
-      auto in_cc_3_1 = conjugate(in_t()(3)(1));
-
-      res()(2)(0) =               diag_t()(1)( 0) * in_t()(2)(0)
-                  +           triangle_t()(1)( 0) * in_t()(2)(1)
-                  +           triangle_t()(1)( 1) * in_t()(2)(2)
-                  +           triangle_t()(1)( 2) * in_t()(3)(0)
-                  +           triangle_t()(1)( 3) * in_t()(3)(1)
-                  +           triangle_t()(1)( 4) * in_t()(3)(2);
-
-      res()(2)(1) =           triangle_t()(1)( 0) * in_cc_2_0;
-      res()(2)(1) =               diag_t()(1)( 1) * in_t()(2)(1)
-                  +           triangle_t()(1)( 5) * in_t()(2)(2)
-                  +           triangle_t()(1)( 6) * in_t()(3)(0)
-                  +           triangle_t()(1)( 7) * in_t()(3)(1)
-                  +           triangle_t()(1)( 8) * in_t()(3)(2)
-                  + conjugate(       res()(2)( 1));
-
-      res()(2)(2) =           triangle_t()(1)( 1) * in_cc_2_0
-                  +           triangle_t()(1)( 5) * in_cc_2_1;
-      res()(2)(2) =               diag_t()(1)( 2) * in_t()(2)(2)
-                  +           triangle_t()(1)( 9) * in_t()(3)(0)
-                  +           triangle_t()(1)(10) * in_t()(3)(1)
-                  +           triangle_t()(1)(11) * in_t()(3)(2)
-                  + conjugate(       res()(2)( 2));
-
-      res()(3)(0) =           triangle_t()(1)( 2) * in_cc_2_0
-                  +           triangle_t()(1)( 6) * in_cc_2_1
-                  +           triangle_t()(1)( 9) * in_cc_2_2;
-      res()(3)(0) =               diag_t()(1)( 3) * in_t()(3)(0)
-                  +           triangle_t()(1)(12) * in_t()(3)(1)
-                  +           triangle_t()(1)(13) * in_t()(3)(2)
-                  + conjugate(       res()(3)( 0));
-
-      res()(3)(1) =           triangle_t()(1)( 3) * in_cc_2_0
-                  +           triangle_t()(1)( 7) * in_cc_2_1
-                  +           triangle_t()(1)(10) * in_cc_2_2
-                  +           triangle_t()(1)(12) * in_cc_3_0;
-      res()(3)(1) =               diag_t()(1)( 4) * in_t()(3)(1)
-                  +           triangle_t()(1)(14) * in_t()(3)(2)
-                  + conjugate(       res()(3)( 1));
-
-      res()(3)(2) =           triangle_t()(1)( 4) * in_cc_2_0
-                  +           triangle_t()(1)( 8) * in_cc_2_1
-                  +           triangle_t()(1)(11) * in_cc_2_2
-                  +           triangle_t()(1)(13) * in_cc_3_0
-                  +           triangle_t()(1)(14) * in_cc_3_1;
-      res()(3)(2) =               diag_t()(1)( 5) * in_t()(3)(2)
-                  + conjugate(       res()(3)( 2));
-
-      vstream(out_v[sF]()(2)(0), res()(2)(0));
-      vstream(out_v[sF]()(2)(1), res()(2)(1));
-      vstream(out_v[sF]()(2)(2), res()(2)(2));
-      vstream(out_v[sF]()(3)(0), res()(3)(0));
-      vstream(out_v[sF]()(3)(1), res()(3)(1));
-      vstream(out_v[sF]()(3)(2), res()(3)(2));
-    });
-  }
-
-  static void MooeeKernel(int                        Nsite,
-                          int                        Ls,
-                          const FermionField&        in,
-                          FermionField&              out,
-                          const CloverDiagonalField& diagonal,
-                          const CloverTriangleField& triangle) {
-#if defined(GRID_CUDA) || defined(GRID_HIP)
-    MooeeKernel_gpu(Nsite, Ls, in, out, diagonal, triangle);
-#else
-    MooeeKernel_cpu(Nsite, Ls, in, out, diagonal, triangle);
-#endif
-  }
-
-  static void Invert(const CloverDiagonalField& diagonal,
-                     const CloverTriangleField& triangle,
-                     CloverDiagonalField&       diagonalInv,
-                     CloverTriangleField&       triangleInv) {
-    conformable(diagonal, diagonalInv);
-    conformable(triangle, triangleInv);
-    conformable(diagonal, triangle);
-
-    diagonalInv.Checkerboard() = diagonal.Checkerboard();
-    triangleInv.Checkerboard() = triangle.Checkerboard();
-
-    GridBase* grid = diagonal.Grid();
-
-    long lsites = grid->lSites();
-
-    typedef typename SiteCloverDiagonal::scalar_object scalar_object_diagonal;
-    typedef typename SiteCloverTriangle::scalar_object scalar_object_triangle;
-
-    autoView(diagonal_v,  diagonal,  CpuRead);
-    autoView(triangle_v,  triangle,  CpuRead);
-    autoView(diagonalInv_v, diagonalInv, CpuWrite);
-    autoView(triangleInv_v, triangleInv, CpuWrite);
-
-    thread_for(site, lsites, { // NOTE: Not on GPU because of Eigen & (peek/poke)LocalSite
-      Eigen::MatrixXcd clover_inv_eigen = Eigen::MatrixXcd::Zero(Ns*Nc, Ns*Nc);
-      Eigen::MatrixXcd clover_eigen = Eigen::MatrixXcd::Zero(Ns*Nc, Ns*Nc);
-
-      scalar_object_diagonal diagonal_tmp     = Zero();
-      scalar_object_diagonal diagonal_inv_tmp = Zero();
-      scalar_object_triangle triangle_tmp     = Zero();
-      scalar_object_triangle triangle_inv_tmp = Zero();
-
-      Coordinate lcoor;
-      grid->LocalIndexToLocalCoor(site, lcoor);
-
-      peekLocalSite(diagonal_tmp, diagonal_v, lcoor);
-      peekLocalSite(triangle_tmp, triangle_v, lcoor);
-
-      // TODO: can we save time here by inverting the two 6x6 hermitian matrices separately?
-      for (long s_row=0;s_row<Ns;s_row++) {
-        for (long s_col=0;s_col<Ns;s_col++) {
-          if(abs(s_row - s_col) > 1 || s_row + s_col == 3) continue;
-          int block       = s_row / Nhs;
-          int s_row_block = s_row % Nhs;
-          int s_col_block = s_col % Nhs;
-          for (long c_row=0;c_row<Nc;c_row++) {
-            for (long c_col=0;c_col<Nc;c_col++) {
-              int i = s_row_block * Nc + c_row;
-              int j = s_col_block * Nc + c_col;
-              if(i == j)
-                clover_eigen(s_row*Nc+c_row, s_col*Nc+c_col) = static_cast<ComplexD>(TensorRemove(diagonal_tmp()(block)(i)));
-              else
-                clover_eigen(s_row*Nc+c_row, s_col*Nc+c_col) = static_cast<ComplexD>(TensorRemove(triangle_elem(triangle_tmp, block, i, j)));
-            }
-          }
-        }
-      }
-
-      clover_inv_eigen = clover_eigen.inverse();
-
-      for (long s_row=0;s_row<Ns;s_row++) {
-        for (long s_col=0;s_col<Ns;s_col++) {
-          if(abs(s_row - s_col) > 1 || s_row + s_col == 3) continue;
-          int block       = s_row / Nhs;
-          int s_row_block = s_row % Nhs;
-          int s_col_block = s_col % Nhs;
-          for (long c_row=0;c_row<Nc;c_row++) {
-            for (long c_col=0;c_col<Nc;c_col++) {
-              int i = s_row_block * Nc + c_row;
-              int j = s_col_block * Nc + c_col;
-              if(i == j)
-                diagonal_inv_tmp()(block)(i) = clover_inv_eigen(s_row*Nc+c_row, s_col*Nc+c_col);
-              else if(i < j)
-                triangle_inv_tmp()(block)(triangle_index(i, j)) = clover_inv_eigen(s_row*Nc+c_row, s_col*Nc+c_col);
-              else
-                continue;
-            }
-          }
-        }
-      }
-
-      pokeLocalSite(diagonal_inv_tmp, diagonalInv_v, lcoor);
-      pokeLocalSite(triangle_inv_tmp, triangleInv_v, lcoor);
-    });
-  }
-
-  static void ConvertLayout(const CloverField&   full,
-                            CloverDiagonalField& diagonal,
-                            CloverTriangleField& triangle) {
-    conformable(full, diagonal);
-    conformable(full, triangle);
-
-    diagonal.Checkerboard() = full.Checkerboard();
-    triangle.Checkerboard() = full.Checkerboard();
-
-    autoView(full_v,     full,     AcceleratorRead);
-    autoView(diagonal_v, diagonal, AcceleratorWrite);
-    autoView(triangle_v, triangle, AcceleratorWrite);
-
-    // NOTE: this function cannot be 'private' since nvcc forbids this for kernels
-    accelerator_for(ss, full.Grid()->oSites(), 1, {
-      for(int s_row = 0; s_row < Ns; s_row++) {
-        for(int s_col = 0; s_col < Ns; s_col++) {
-          if(abs(s_row - s_col) > 1 || s_row + s_col == 3) continue;
-          int block       = s_row / Nhs;
-          int s_row_block = s_row % Nhs;
-          int s_col_block = s_col % Nhs;
-          for(int c_row = 0; c_row < Nc; c_row++) {
-            for(int c_col = 0; c_col < Nc; c_col++) {
-              int i = s_row_block * Nc + c_row;
-              int j = s_col_block * Nc + c_col;
-              if(i == j)
-                diagonal_v[ss]()(block)(i) = full_v[ss]()(s_row, s_col)(c_row, c_col);
-              else if(i < j)
-                triangle_v[ss]()(block)(triangle_index(i, j)) = full_v[ss]()(s_row, s_col)(c_row, c_col);
-              else
-                continue;
-            }
-          }
-        }
-      }
-    });
-  }
-
-
-  static void ConvertLayout(const CloverDiagonalField& diagonal,
-                            const CloverTriangleField& triangle,
-                            CloverField&               full) {
-    conformable(full, diagonal);
-    conformable(full, triangle);
-
-    full.Checkerboard() = diagonal.Checkerboard();
-
-    full = Zero();
-
-    autoView(diagonal_v, diagonal, AcceleratorRead);
-    autoView(triangle_v, triangle, AcceleratorRead);
-    autoView(full_v,     full,     AcceleratorWrite);
-
-    // NOTE: this function cannot be 'private' since nvcc forbids this for kernels
-    accelerator_for(ss, full.Grid()->oSites(), 1, {
-      for(int s_row = 0; s_row < Ns; s_row++) {
-        for(int s_col = 0; s_col < Ns; s_col++) {
-          if(abs(s_row - s_col) > 1 || s_row + s_col == 3) continue;
-          int block       = s_row / Nhs;
-          int s_row_block = s_row % Nhs;
-          int s_col_block = s_col % Nhs;
-          for(int c_row = 0; c_row < Nc; c_row++) {
-            for(int c_col = 0; c_col < Nc; c_col++) {
-              int i = s_row_block * Nc + c_row;
-              int j = s_col_block * Nc + c_col;
-              if(i == j)
-                full_v[ss]()(s_row, s_col)(c_row, c_col) = diagonal_v[ss]()(block)(i);
-              else
-                full_v[ss]()(s_row, s_col)(c_row, c_col) = triangle_elem(triangle_v[ss], block, i, j);
-            }
-          }
-        }
-      }
-    });
-  }
-
-  static void ModifyBoundaries(CloverDiagonalField& diagonal, CloverTriangleField& triangle, RealD csw_t, RealD cF, RealD diag_mass) {
-    // Checks/grid
-    double t0 = usecond();
-    conformable(diagonal, triangle);
-    GridBase* grid = diagonal.Grid();
-
-    // Determine the boundary coordinates/sites
-    double t1 = usecond();
-    int t_dir = Nd - 1;
-    Lattice<iScalar<vInteger>> t_coor(grid);
-    LatticeCoordinate(t_coor, t_dir);
-    int T = grid->GlobalDimensions()[t_dir];
-
-    // Set off-diagonal parts at boundary to zero -- OK
-    double t2 = usecond();
-    CloverTriangleField zeroTriangle(grid);
-    zeroTriangle.Checkerboard() = triangle.Checkerboard();
-    zeroTriangle = Zero();
-    triangle = where(t_coor == 0,   zeroTriangle, triangle);
-    triangle = where(t_coor == T-1, zeroTriangle, triangle);
-
-    // Set diagonal to unity (scaled correctly) -- OK
-    double t3 = usecond();
-    CloverDiagonalField tmp(grid);
-    tmp.Checkerboard() = diagonal.Checkerboard();
-    tmp                = -1.0 * csw_t + diag_mass;
-    diagonal           = where(t_coor == 0,   tmp, diagonal);
-    diagonal           = where(t_coor == T-1, tmp, diagonal);
-
-    // Correct values next to boundary
-    double t4 = usecond();
-    if(cF != 1.0) {
-      tmp = cF - 1.0;
-      tmp += diagonal;
-      diagonal = where(t_coor == 1,   tmp, diagonal);
-      diagonal = where(t_coor == T-2, tmp, diagonal);
-    }
-
-    // Report timings
-    double t5 = usecond();
-#if 0
-    std::cout << GridLogMessage << "CompactWilsonCloverHelpers::ModifyBoundaries timings:"
-              << " checks = "          << (t1 - t0) / 1e6
-              << ", coordinate = "     << (t2 - t1) / 1e6
-              << ", off-diag zero = "  << (t3 - t2) / 1e6
-              << ", diagonal unity = " << (t4 - t3) / 1e6
-              << ", near-boundary = "  << (t5 - t4) / 1e6
-              << ", total = "          << (t5 - t0) / 1e6
-              << std::endl;
-#endif
-  }
-
-  template<class Field, class Mask>
-  static strong_inline void ApplyBoundaryMask(Field& f, const Mask& m) {
-    conformable(f, m);
-    auto grid  = f.Grid();
-    const uint32_t Nsite = grid->oSites();
-    const uint32_t Nsimd = grid->Nsimd();
-    autoView(f_v, f, AcceleratorWrite);
-    autoView(m_v, m, AcceleratorRead);
-    // NOTE: this function cannot be 'private' since nvcc forbids this for kernels
-    accelerator_for(ss, Nsite, Nsimd, {
-      coalescedWrite(f_v[ss], m_v(ss) * f_v(ss));
-    });
-  }
-
-  template<class MaskField>
-  static void SetupMasks(MaskField& full, MaskField& even, MaskField& odd) {
-    assert(even.Grid()->_isCheckerBoarded && even.Checkerboard() == Even);
-    assert(odd.Grid()->_isCheckerBoarded  && odd.Checkerboard()  == Odd);
-    assert(!full.Grid()->_isCheckerBoarded);
-
-    GridBase* grid = full.Grid();
-    int t_dir = Nd-1;
-    Lattice<iScalar<vInteger>> t_coor(grid);
-    LatticeCoordinate(t_coor, t_dir);
-    int T = grid->GlobalDimensions()[t_dir];
-
-    MaskField zeroMask(grid); zeroMask = Zero();
-    full = 1.0;
-    full = where(t_coor == 0,   zeroMask, full);
-    full = where(t_coor == T-1, zeroMask, full);
-
-    pickCheckerboard(Even, even, full);
-    pickCheckerboard(Odd,  odd,  full);
-  }
-};
-
-NAMESPACE_END(Grid);
@@ -1,90 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: ./lib/qcd/action/fermion/WilsonCloverTypes.h
-
-    Copyright (C) 2021 - 2022
-
-    Author: Daniel Richtmann <daniel.richtmann@gmail.com>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-
-#pragma once
-
-NAMESPACE_BEGIN(Grid);
-
-template<class Impl>
-class WilsonCloverTypes {
-public:
-  INHERIT_IMPL_TYPES(Impl);
-
-  template <typename vtype> using iImplClover = iScalar<iMatrix<iMatrix<vtype, Impl::Dimension>, Ns>>;
-
-  typedef iImplClover<Simd> SiteClover;
-
-  typedef Lattice<SiteClover> CloverField;
-};
-
-template<class Impl>
-class CompactWilsonCloverTypes {
-public:
-  INHERIT_IMPL_TYPES(Impl);
-
-  static constexpr int Nred      = Nc * Nhs;        // 6
-  static constexpr int Nblock    = Nhs;             // 2
-  static constexpr int Ndiagonal = Nred;            // 6
-  static constexpr int Ntriangle = (Nred - 1) * Nc; // 15
-
-  template<typename vtype> using iImplCloverDiagonal = iScalar<iVector<iVector<vtype, Ndiagonal>, Nblock>>;
-  template<typename vtype> using iImplCloverTriangle = iScalar<iVector<iVector<vtype, Ntriangle>, Nblock>>;
-
-  typedef iImplCloverDiagonal<Simd> SiteCloverDiagonal;
-  typedef iImplCloverTriangle<Simd> SiteCloverTriangle;
-  typedef iSinglet<Simd>            SiteMask;
-
-  typedef Lattice<SiteCloverDiagonal> CloverDiagonalField;
-  typedef Lattice<SiteCloverTriangle> CloverTriangleField;
-  typedef Lattice<SiteMask>           MaskField;
-};
-
-#define INHERIT_CLOVER_TYPES(Impl)                                 \
-  typedef typename WilsonCloverTypes<Impl>::SiteClover SiteClover; \
-  typedef typename WilsonCloverTypes<Impl>::CloverField CloverField;
-
-#define INHERIT_COMPACT_CLOVER_TYPES(Impl) \
-  typedef typename CompactWilsonCloverTypes<Impl>::SiteCloverDiagonal  SiteCloverDiagonal; \
-  typedef typename CompactWilsonCloverTypes<Impl>::SiteCloverTriangle  SiteCloverTriangle; \
-  typedef typename CompactWilsonCloverTypes<Impl>::SiteMask            SiteMask; \
-  typedef typename CompactWilsonCloverTypes<Impl>::CloverDiagonalField CloverDiagonalField; \
-  typedef typename CompactWilsonCloverTypes<Impl>::CloverTriangleField CloverTriangleField; \
-  typedef typename CompactWilsonCloverTypes<Impl>::MaskField           MaskField; \
-  /* ugly duplication but needed inside functionality classes */ \
-  template<typename vtype> using iImplCloverDiagonal = \
-    iScalar<iVector<iVector<vtype, CompactWilsonCloverTypes<Impl>::Ndiagonal>, CompactWilsonCloverTypes<Impl>::Nblock>>; \
-  template<typename vtype> using iImplCloverTriangle = \
-    iScalar<iVector<iVector<vtype, CompactWilsonCloverTypes<Impl>::Ntriangle>, CompactWilsonCloverTypes<Impl>::Nblock>>;
-
-#define INHERIT_COMPACT_CLOVER_SIZES(Impl)                                    \
-  static constexpr int Nred      = CompactWilsonCloverTypes<Impl>::Nred;      \
-  static constexpr int Nblock    = CompactWilsonCloverTypes<Impl>::Nblock;    \
-  static constexpr int Ndiagonal = CompactWilsonCloverTypes<Impl>::Ndiagonal; \
-  static constexpr int Ntriangle = CompactWilsonCloverTypes<Impl>::Ntriangle;
-
-NAMESPACE_END(Grid);
@@ -32,218 +32,17 @@ Author: paboyle <paboyle@ph.ed.ac.uk>

 NAMESPACE_BEGIN(Grid);

-///////////////////////////////////////////////////////////////
-// Wilson compressor will need FaceGather policies for:
-// Periodic, Dirichlet, and partial Dirichlet for DWF
-///////////////////////////////////////////////////////////////
-const int dwf_compressor_depth=1;
-#define DWF_COMPRESS
-class FaceGatherPartialDWF
-{
-public:
-#ifdef DWF_COMPRESS
-  static int PartialCompressionFactor(GridBase *grid) {return grid->_fdimensions[0]/(2*dwf_compressor_depth);};
-#else
-  static int PartialCompressionFactor(GridBase *grid) { return 1;}
-#endif
-  template<class vobj,class cobj,class compressor>
-  static void Gather_plane_simple (commVector<std::pair<int,int> >& table,
-				   const Lattice<vobj> &rhs,
-				   cobj *buffer,
-				   compressor &compress,
-				   int off,int so,int partial)
-  {
-    //DWF only hack: If a direction that is OFF node we use Partial Dirichlet
-    //  Shrinks local and remote comms buffers
-    GridBase *Grid = rhs.Grid();
-    int Ls = Grid->_rdimensions[0];
-#ifdef DWF_COMPRESS
-    int depth=dwf_compressor_depth;
-#else 
-    int depth=Ls/2;
-#endif
-    std::pair<int,int> *table_v = & table[0];
-    auto rhs_v = rhs.View(AcceleratorRead);
-    int vol=table.size()/Ls;
-    accelerator_forNB( idx,table.size(), vobj::Nsimd(), {
-	Integer i=idx/Ls;
-	Integer s=idx%Ls;
-	Integer sc=depth+s-(Ls-depth);
-	if(s<depth)     compress.Compress(buffer[off+i+s*vol],rhs_v[so+table_v[idx].second]);
-	if(s>=Ls-depth) compress.Compress(buffer[off+i+sc*vol],rhs_v[so+table_v[idx].second]);
-    });
-    rhs_v.ViewClose();
-  }
-  template<class decompressor,class Decompression>
-  static void DecompressFace(decompressor decompress,Decompression &dd)
-  {
-    auto Ls = dd.dims[0];
-#ifdef DWF_COMPRESS
-    int depth=dwf_compressor_depth;
-#else
-    int depth=Ls/2;
-#endif    
-    // Just pass in the Grid
-    auto kp = dd.kernel_p;
-    auto mp = dd.mpi_p;
-    int size= dd.buffer_size;
-    int vol= size/Ls;
-    accelerator_forNB(o,size,1,{
-	int idx=o/Ls;
-	int   s=o%Ls;
-	if ( s < depth ) {
-	  int oo=s*vol+idx;
-	  kp[o]=mp[oo];
-	} else if ( s >= Ls-depth ) {
-	  int sc = depth + s - (Ls-depth);
-	  int oo=sc*vol+idx; 
-	  kp[o]=mp[oo];
-	} else {
-	  kp[o] = Zero();//fill rest with zero if partial dirichlet
-	}
-    });
-  }
-  ////////////////////////////////////////////////////////////////////////////////////////////
-  // Need to gather *interior portions* for ALL s-slices in simd directions
-  // Do the gather as need to treat SIMD lanes differently, and insert zeroes on receive side
-  // Reorder the fifth dim to be s=Ls-1 , s=0, s=1,...,Ls-2.
-  ////////////////////////////////////////////////////////////////////////////////////////////
-  template<class vobj,class cobj,class compressor>
-  static void Gather_plane_exchange(commVector<std::pair<int,int> >& table,const Lattice<vobj> &rhs,
-				    Vector<cobj *> pointers,int dimension,int plane,int cbmask,
-				    compressor &compress,int type,int partial)
-  {
-    GridBase *Grid = rhs.Grid();
-    int Ls = Grid->_rdimensions[0];
-#ifdef DWF_COMPRESS
-    int depth=dwf_compressor_depth;
-#else
-    int depth = Ls/2;
-#endif
-    
-    // insertion of zeroes...
-    assert( (table.size()&0x1)==0);
-    int num=table.size()/2;
-    int so  = plane*rhs.Grid()->_ostride[dimension]; // base offset for start of plane
-    
-    auto rhs_v = rhs.View(AcceleratorRead);
-    auto p0=&pointers[0][0];
-    auto p1=&pointers[1][0];
-    auto tp=&table[0];
-    int nnum=num/Ls;
-    accelerator_forNB(j, num, vobj::Nsimd(), {
-	//  Reorders both local and remote comms buffers
-	//  
-	int s  = j % Ls;
-	int sp1 = (s+depth)%Ls;  // peri incremented s slice
-	
-	int hxyz= j/Ls;
-
-	int xyz0= hxyz*2; // xyzt part of coor
-	int xyz1= hxyz*2+1;
-	
-	int jj= hxyz + sp1*nnum ; // 0,1,2,3 -> Ls-1 slice , 0-slice, 1-slice ....
-	
-	int kk0= xyz0*Ls + s ; // s=0 goes to s=1
-	int kk1= xyz1*Ls + s ; // s=Ls-1 -> s=0
-	compress.CompressExchange(p0[jj],p1[jj],
-				  rhs_v[so+tp[kk0 ].second], // Same s, consecutive xyz sites
-				  rhs_v[so+tp[kk1 ].second], 
-				  type);
-    });
-    rhs_v.ViewClose();
-  }
-  // Merge routine is for SIMD faces
-  template<class decompressor,class Merger>
-  static void MergeFace(decompressor decompress,Merger &mm)
-  {
-    auto Ls = mm.dims[0];
-#ifdef DWF_COMPRESS
-    int depth=dwf_compressor_depth;
-#else
-    int depth = Ls/2;
-#endif
-    int  num= mm.buffer_size/2; // relate vol and Ls to buffer size
-    auto mp = &mm.mpointer[0];
-    auto vp0= &mm.vpointers[0][0]; // First arg is exchange first
-    auto vp1= &mm.vpointers[1][0];
-    auto type= mm.type;
-    int nnum = num/Ls;
-    accelerator_forNB(o,num,Merger::Nsimd,{
-
-	int  s=o%Ls;
-	int hxyz=o/Ls; // xyzt related component
-	int xyz0=hxyz*2;
-	int xyz1=hxyz*2+1;
-
-	int sp = (s+depth)%Ls; 
-	int jj= hxyz + sp*nnum ; // 0,1,2,3 -> Ls-1 slice , 0-slice, 1-slice ....
-
-	int oo0= s+xyz0*Ls;
-	int oo1= s+xyz1*Ls;
-
-	// same ss0, ss1 pair goes to new layout
-	decompress.Exchange(mp[oo0],mp[oo1],vp0[jj],vp1[jj],type);
-      });
-  }
-};
-class FaceGatherDWFMixedBCs
-{
-public:
-#ifdef DWF_COMPRESS
-  static int PartialCompressionFactor(GridBase *grid) {return grid->_fdimensions[0]/(2*dwf_compressor_depth);};
-#else 
-  static int PartialCompressionFactor(GridBase *grid) {return 1;}
-#endif
-  
-  template<class vobj,class cobj,class compressor>
-  static void Gather_plane_simple (commVector<std::pair<int,int> >& table,
-					 const Lattice<vobj> &rhs,
-					 cobj *buffer,
-					 compressor &compress,
-					 int off,int so,int partial)
-  {
-    //    std::cout << " face gather simple DWF partial "<<partial <<std::endl;
-    if(partial) FaceGatherPartialDWF::Gather_plane_simple(table,rhs,buffer,compress,off,so,partial);
-    else        FaceGatherSimple::Gather_plane_simple(table,rhs,buffer,compress,off,so,partial);
-  }
-  template<class vobj,class cobj,class compressor>
-  static void Gather_plane_exchange(commVector<std::pair<int,int> >& table,const Lattice<vobj> &rhs,
-				    Vector<cobj *> pointers,int dimension,int plane,int cbmask,
-				    compressor &compress,int type,int partial)
-  {
-    //    std::cout << " face gather exch DWF partial "<<partial <<std::endl;
-    if(partial) FaceGatherPartialDWF::Gather_plane_exchange(table,rhs,pointers,dimension, plane,cbmask,compress,type,partial);
-    else        FaceGatherSimple::Gather_plane_exchange    (table,rhs,pointers,dimension, plane,cbmask,compress,type,partial);
-  }
-  template<class decompressor,class Merger>
-  static void MergeFace(decompressor decompress,Merger &mm)
-  {
-    int partial = mm.partial;
-    //    std::cout << " merge DWF partial "<<partial <<std::endl;
-    if ( partial ) FaceGatherPartialDWF::MergeFace(decompress,mm);
-    else           FaceGatherSimple::MergeFace(decompress,mm);
-  }
-
-  template<class decompressor,class Decompression>
-  static void DecompressFace(decompressor decompress,Decompression &dd)
-  {
-    int partial = dd.partial;
-    //    std::cout << " decompress DWF partial "<<partial <<std::endl;
-    if ( partial ) FaceGatherPartialDWF::DecompressFace(decompress,dd);
-    else           FaceGatherSimple::DecompressFace(decompress,dd);
-  }
-};
-
 /////////////////////////////////////////////////////////////////////////////////////////////
-// optimised versions supporting half precision too??? Deprecate
+// optimised versions supporting half precision too
 /////////////////////////////////////////////////////////////////////////////////////////////

+template<class _HCspinor,class _Hspinor,class _Spinor, class projector,typename SFINAE = void >
+class WilsonCompressorTemplate;
+

-//Could make FaceGather a template param, but then behaviour is runtime not compile time
 template<class _HCspinor,class _Hspinor,class _Spinor, class projector>
-class WilsonCompressorTemplate  : public FaceGatherDWFMixedBCs
-//  : public FaceGatherSimple
+class WilsonCompressorTemplate< _HCspinor, _Hspinor, _Spinor, projector,
+				typename std::enable_if<std::is_same<_HCspinor,_Hspinor>::value>::type >
 {
 public:
  
@@ -269,90 +68,140 @@ public:
  /*****************************************************/
  /* Compress includes precision change if mpi data is not same */
  /*****************************************************/
-  accelerator_inline void Compress(SiteHalfSpinor &buf,const SiteSpinor &in) const {
-    typedef decltype(coalescedRead(buf)) sobj;
-    sobj sp;
-    auto sin = coalescedRead(in);
-    projector::Proj(sp,sin,mu,dag);
-    coalescedWrite(buf,sp);
+  template<class _SiteHalfSpinor, class _SiteSpinor>
+  accelerator_inline void Compress(_SiteHalfSpinor *buf,Integer o,const _SiteSpinor &in) const {
+    _SiteHalfSpinor tmp;
+    projector::Proj(tmp,in,mu,dag);
+    vstream(buf[o],tmp);
  }

  /*****************************************************/
  /* Exchange includes precision change if mpi data is not same */
  /*****************************************************/
-  accelerator_inline void Exchange(SiteHalfSpinor &mp0,
-				   SiteHalfSpinor &mp1,
-				   const SiteHalfSpinor & vp0,
-				   const SiteHalfSpinor & vp1,
-				   Integer type) const {
-#ifdef GRID_SIMT
-    exchangeSIMT(mp0,mp1,vp0,vp1,type);
-#else
+  accelerator_inline void Exchange(SiteHalfSpinor *mp,
+				   const SiteHalfSpinor * __restrict__ vp0,
+				   const SiteHalfSpinor * __restrict__ vp1,
+				   Integer type,Integer o) const {
    SiteHalfSpinor tmp1;
    SiteHalfSpinor tmp2;
-    exchange(tmp1,tmp2,vp0,vp1,type);
-    vstream(mp0,tmp1);
-    vstream(mp1,tmp2);
-#endif
+    exchange(tmp1,tmp2,vp0[o],vp1[o],type);
+    vstream(mp[2*o  ],tmp1);
+    vstream(mp[2*o+1],tmp2);
  }
-  

  /*****************************************************/
  /* Have a decompression step if mpi data is not same */
  /*****************************************************/
-  accelerator_inline void Decompress(SiteHalfSpinor &out,
-				     SiteHalfSpinor &in) const {    
-    out = in;
+  accelerator_inline void Decompress(SiteHalfSpinor * __restrict__ out,
+				     SiteHalfSpinor * __restrict__ in, Integer o) const {    
+    assert(0);
  }

  /*****************************************************/
  /* Compress Exchange                                 */
  /*****************************************************/
-  accelerator_inline void CompressExchange(SiteHalfSpinor &out0,
-					   SiteHalfSpinor &out1,
-					   const SiteSpinor &in0,
-					   const SiteSpinor &in1,
-					   Integer type) const
+  accelerator_inline void CompressExchange(SiteHalfSpinor * __restrict__ out0,
+					   SiteHalfSpinor * __restrict__ out1,
+					   const SiteSpinor * __restrict__ in,
+					   Integer j,Integer k, Integer m,Integer type) const
  {
-#ifdef GRID_SIMT
-    typedef SiteSpinor vobj;
-    typedef SiteHalfSpinor hvobj;
-    typedef decltype(coalescedRead(in0))    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 = &in0;
-    const vobj *vp1 = &in1;
-    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,psa);
-    coalescedWrite(out1,psb);
-#else
    SiteHalfSpinor temp1, temp2;
    SiteHalfSpinor temp3, temp4;
-    projector::Proj(temp1,in0,mu,dag);
-    projector::Proj(temp2,in1,mu,dag);
+    projector::Proj(temp1,in[k],mu,dag);
+    projector::Proj(temp2,in[m],mu,dag);
    exchange(temp3,temp4,temp1,temp2,type);
-    vstream(out0,temp3);
-    vstream(out1,temp4);
-#endif
+    vstream(out0[j],temp3);
+    vstream(out1[j],temp4);
  }

  /*****************************************************/
  /* Pass the info to the stencil */
  /*****************************************************/
-  accelerator_inline bool DecompressionStep(void) const {
-    return false;
+  accelerator_inline bool DecompressionStep(void) const { return false; }
+
+};
+
+template<class _HCspinor,class _Hspinor,class _Spinor, class projector>
+class WilsonCompressorTemplate< _HCspinor, _Hspinor, _Spinor, projector,
+				typename std::enable_if<!std::is_same<_HCspinor,_Hspinor>::value>::type >
+{
+public:
+  
+  int mu,dag;  
+
+  void Point(int p) { mu=p; };
+
+  WilsonCompressorTemplate(int _dag=0){
+    dag = _dag;
  }

+  typedef _Spinor         SiteSpinor;
+  typedef _Hspinor     SiteHalfSpinor;
+  typedef _HCspinor SiteHalfCommSpinor;
+  typedef typename SiteHalfCommSpinor::vector_type vComplexLow;
+  typedef typename SiteHalfSpinor::vector_type     vComplexHigh;
+  constexpr static int Nw=sizeof(SiteHalfSpinor)/sizeof(vComplexHigh);
+
+  accelerator_inline int CommDatumSize(void) const {
+    return sizeof(SiteHalfCommSpinor);
+  }
+
+  /*****************************************************/
+  /* Compress includes precision change if mpi data is not same */
+  /*****************************************************/
+  template<class _SiteHalfSpinor, class _SiteSpinor>
+  accelerator_inline void Compress(_SiteHalfSpinor *buf,Integer o,const _SiteSpinor &in) const {
+    _SiteHalfSpinor hsp;
+    SiteHalfCommSpinor *hbuf = (SiteHalfCommSpinor *)buf;
+    projector::Proj(hsp,in,mu,dag);
+    precisionChange((vComplexLow *)&hbuf[o],(vComplexHigh *)&hsp,Nw);
+  }
+
+  /*****************************************************/
+  /* Exchange includes precision change if mpi data is not same */
+  /*****************************************************/
+  accelerator_inline void Exchange(SiteHalfSpinor *mp,
+                       SiteHalfSpinor *vp0,
+                       SiteHalfSpinor *vp1,
+		       Integer type,Integer o) const {
+    SiteHalfSpinor vt0,vt1;
+    SiteHalfCommSpinor *vpp0 = (SiteHalfCommSpinor *)vp0;
+    SiteHalfCommSpinor *vpp1 = (SiteHalfCommSpinor *)vp1;
+    precisionChange((vComplexHigh *)&vt0,(vComplexLow *)&vpp0[o],Nw);
+    precisionChange((vComplexHigh *)&vt1,(vComplexLow *)&vpp1[o],Nw);
+    exchange(mp[2*o],mp[2*o+1],vt0,vt1,type);
+  }
+
+  /*****************************************************/
+  /* Have a decompression step if mpi data is not same */
+  /*****************************************************/
+  accelerator_inline void Decompress(SiteHalfSpinor *out, SiteHalfSpinor *in, Integer o) const {
+    SiteHalfCommSpinor *hin=(SiteHalfCommSpinor *)in;
+    precisionChange((vComplexHigh *)&out[o],(vComplexLow *)&hin[o],Nw);
+  }
+
+  /*****************************************************/
+  /* Compress Exchange                                 */
+  /*****************************************************/
+  accelerator_inline void CompressExchange(SiteHalfSpinor *out0,
+			       SiteHalfSpinor *out1,
+			       const SiteSpinor *in,
+			       Integer j,Integer k, Integer m,Integer type) const {
+    SiteHalfSpinor temp1, temp2,temp3,temp4;
+    SiteHalfCommSpinor *hout0 = (SiteHalfCommSpinor *)out0;
+    SiteHalfCommSpinor *hout1 = (SiteHalfCommSpinor *)out1;
+    projector::Proj(temp1,in[k],mu,dag);
+    projector::Proj(temp2,in[m],mu,dag);
+    exchange(temp3,temp4,temp1,temp2,type);
+    precisionChange((vComplexLow *)&hout0[j],(vComplexHigh *)&temp3,Nw);
+    precisionChange((vComplexLow *)&hout1[j],(vComplexHigh *)&temp4,Nw);
+  }
+
+  /*****************************************************/
+  /* Pass the info to the stencil */
+  /*****************************************************/
+  accelerator_inline bool DecompressionStep(void) const { return true; }
+
 };

 #define DECLARE_PROJ(Projector,Compressor,spProj)			\
@@ -404,7 +253,36 @@ public:
  typedef typename Base::View_type View_type;
  typedef typename Base::StencilVector StencilVector;

-  //  Vector<int> surface_list;
+  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;
+
  WilsonStencil(GridBase *grid,
 		int npoints,
 		int checkerboard,
@@ -412,11 +290,11 @@ public:
 		const std::vector<int> &distances,Parameters p)  
    : CartesianStencil<vobj,cobj,Parameters> (grid,npoints,checkerboard,directions,distances,p) 
  { 
-    //    surface_list.resize(0);
+    ZeroCountersi();
+    surface_list.resize(0);
    this->same_node.resize(npoints);
  };

-  /*
  void BuildSurfaceList(int Ls,int vol4){

    // find same node for SHM
@@ -437,25 +315,32 @@ public:
      }
    }
  }
-  */
-  
+
  template < class compressor>
  void HaloExchangeOpt(const Lattice<vobj> &source,compressor &compress) 
  {
    std::vector<std::vector<CommsRequest_t> > reqs;
    this->HaloExchangeOptGather(source,compress);
+    double t1=usecond();
    // Asynchronous MPI calls multidirectional, Isend etc...
    // Non-overlapped directions within a thread. Asynchronous calls except MPI3, threaded up to comm threads ways.
    this->Communicate();
+    double t2=usecond(); timer1 += t2-t1;
    this->CommsMerge(compress);
+    double t3=usecond(); timer2 += t3-t2;
    this->CommsMergeSHM(compress);
+    double t4=usecond(); timer3 += t4-t3;
  }
  
  template <class compressor>
  void HaloExchangeOptGather(const Lattice<vobj> &source,compressor &compress) 
  {
    this->Prepare();
+    double t0=usecond();
    this->HaloGatherOpt(source,compress);
+    double t1=usecond();
+    timer0 += t1-t0;
+    callsi++;
  }

  template <class compressor>
@@ -467,9 +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;
      
@@ -505,6 +393,8 @@ public:
    }
    this->face_table_computed=1;
    assert(this->u_comm_offset==this->_unified_buffer_size);
+    this->halogtime+=usecond();
+    accelerator_barrier();
  }

 };
@@ -74,6 +74,20 @@ public:
  FermionField _tmp;
  FermionField &tmp(void) { return _tmp; }

+  void Report(void);
+  void ZeroCounters(void);
+  double DhopCalls;
+  double DhopCommTime;
+  double DhopComputeTime;
+  double DhopComputeTime2;
+  double DhopFaceTime;
+  double DhopTotalTime;
+
+  double DerivCalls;
+  double DerivCommTime;
+  double DerivComputeTime;
+  double DerivDhopComputeTime;
+
  //////////////////////////////////////////////////////////////////
  // override multiply; cut number routines if pass dagger argument
  // and also make interface more uniformly consistent
@@ -75,8 +75,19 @@ public:
  FermionField _tmp;
  FermionField &tmp(void) { return _tmp; }

-  int Dirichlet;
-  Coordinate Block; 
+  void Report(void);
+  void ZeroCounters(void);
+  double DhopCalls;
+  double DhopCommTime;
+  double DhopComputeTime;
+  double DhopComputeTime2;
+  double DhopFaceTime;
+  double DhopTotalTime;
+
+  double DerivCalls;
+  double DerivCommTime;
+  double DerivComputeTime;
+  double DerivDhopComputeTime;

  ///////////////////////////////////////////////////////////////
  // Implement the abstract base
@@ -162,10 +173,7 @@ public:
 		  GridCartesian         &FourDimGrid,
 		  GridRedBlackCartesian &FourDimRedBlackGrid,
 		  double _M5,const ImplParams &p= ImplParams());
-
-  virtual void DirichletBlock(const Coordinate & block)
-  {
-  }
+    
  // Constructors
  /*
    WilsonFermion5D(int simd, 
@@ -37,7 +37,7 @@ NAMESPACE_BEGIN(Grid);
 template <class S, class Representation = FundamentalRepresentation,class Options = CoeffReal >
 class WilsonImpl : public PeriodicGaugeImpl<GaugeImplTypes<S, Representation::Dimension > > {
 public:
-  
+
  static const int Dimension = Representation::Dimension;
  static const bool isFundamental = Representation::isFundamental;
  static const bool LsVectorised=false;
@@ -242,13 +242,19 @@ public:
 typedef WilsonImpl<vComplex,  FundamentalRepresentation, CoeffReal > WilsonImplR;  // Real.. whichever prec
 typedef WilsonImpl<vComplexF, FundamentalRepresentation, CoeffReal > WilsonImplF;  // Float
 typedef WilsonImpl<vComplexD, FundamentalRepresentation, CoeffReal > WilsonImplD;  // Double
-typedef WilsonImpl<vComplexD2, FundamentalRepresentation, CoeffReal > WilsonImplD2;  // 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<vComplexD2, FundamentalRepresentation, CoeffComplex > ZWilsonImplD2; // Double

+typedef WilsonImpl<vComplex,  FundamentalRepresentation, CoeffComplexHalfComms > ZWilsonImplRL; // Real.. whichever prec
+typedef WilsonImpl<vComplexF, FundamentalRepresentation, CoeffComplexHalfComms > ZWilsonImplFH; // Float
+typedef WilsonImpl<vComplexD, FundamentalRepresentation, CoeffComplexHalfComms > ZWilsonImplDF; // Double
+ 
 typedef WilsonImpl<vComplex,  AdjointRepresentation, CoeffReal > WilsonAdjImplR;   // Real.. whichever prec
 typedef WilsonImpl<vComplexF, AdjointRepresentation, CoeffReal > WilsonAdjImplF;  // Float
 typedef WilsonImpl<vComplexD, AdjointRepresentation, CoeffReal > WilsonAdjImplD;  // Double
@@ -47,7 +47,7 @@ CayleyFermion5D<Impl>::CayleyFermion5D(GaugeField &_Umu,
 			FiveDimRedBlackGrid,
 			FourDimGrid,
 			FourDimRedBlackGrid,_M5,p),
-  mass_plus(_mass), mass_minus(_mass)
+  mass(_mass)
 { 
 }

@@ -152,13 +152,65 @@ void CayleyFermion5D<Impl>::DminusDag(const FermionField &psi, FermionField &chi
  }
 }

+template<class Impl> void CayleyFermion5D<Impl>::CayleyReport(void)
+{
+  this->Report();
+  Coordinate latt = GridDefaultLatt();          
+  RealD volume = this->Ls;  for(int mu=0;mu<Nd;mu++) volume=volume*latt[mu];
+  RealD NP     = this->_FourDimGrid->_Nprocessors;
+  if ( M5Dcalls > 0 ) {
+    std::cout << GridLogMessage << "#### M5D calls report " << std::endl;
+    std::cout << GridLogMessage << "CayleyFermion5D Number of M5D Calls     : " << M5Dcalls   << std::endl;
+    std::cout << GridLogMessage << "CayleyFermion5D ComputeTime/Calls       : " << M5Dtime / M5Dcalls << " us" << std::endl;
+
+    // Flops = 10.0*(Nc*Ns) *Ls*vol
+    RealD mflops = 10.0*(Nc*Ns)*volume*M5Dcalls/M5Dtime/2; // 2 for red black counting
+    std::cout << GridLogMessage << "Average mflops/s per call                : " << mflops << std::endl;
+    std::cout << GridLogMessage << "Average mflops/s per call per rank       : " << mflops/NP << std::endl;
+
+    // Bytes = sizeof(Real) * (Nc*Ns*Nreim) * Ls * vol * (read+write) (/2 for red black counting)
+    // read = 2 ( psi[ss+s+1] and psi[ss+s-1] count as 1 )
+    // write = 1
+    RealD Gbytes = sizeof(Real) * (Nc*Ns*2) * volume * 3 /2. * 1.e-9;
+    std::cout << GridLogMessage << "Average bandwidth (GB/s)                 : " << Gbytes/M5Dtime*M5Dcalls*1.e6 << std::endl;
+  }
+
+  if ( MooeeInvCalls > 0 ) {
+
+    std::cout << GridLogMessage << "#### MooeeInv calls report " << std::endl;
+    std::cout << GridLogMessage << "CayleyFermion5D Number of MooeeInv Calls     : " << MooeeInvCalls   << std::endl;
+    std::cout << GridLogMessage << "CayleyFermion5D ComputeTime/Calls            : " << MooeeInvTime / MooeeInvCalls << " us" << std::endl;
+#ifdef GRID_CUDA
+    RealD mflops = ( -16.*Nc*Ns+this->Ls*(1.+18.*Nc*Ns) )*volume*MooeeInvCalls/MooeeInvTime/2; // 2 for red black counting
+    std::cout << GridLogMessage << "Average mflops/s per call                : " << mflops << std::endl;
+    std::cout << GridLogMessage << "Average mflops/s per call per rank       : " << mflops/NP << std::endl;
+#else
+    // Flops = MADD * Ls *Ls *4dvol * spin/colour/complex
+    RealD mflops = 2.0*24*this->Ls*volume*MooeeInvCalls/MooeeInvTime/2; // 2 for red black counting
+    std::cout << GridLogMessage << "Average mflops/s per call                : " << mflops << std::endl;
+    std::cout << GridLogMessage << "Average mflops/s per call per rank       : " << mflops/NP << std::endl;
+#endif
+  }
+
+}
+template<class Impl> void CayleyFermion5D<Impl>::CayleyZeroCounters(void)
+{
+  this->ZeroCounters();
+  M5Dflops=0;
+  M5Dcalls=0;
+  M5Dtime=0;
+  MooeeInvFlops=0;
+  MooeeInvCalls=0;
+  MooeeInvTime=0;
+}
+
 template<class Impl>  
 void CayleyFermion5D<Impl>::M5D   (const FermionField &psi, FermionField &chi)
 {
  int Ls=this->Ls;
  Vector<Coeff_t> diag (Ls,1.0);
-  Vector<Coeff_t> upper(Ls,-1.0); upper[Ls-1]=mass_minus;
-  Vector<Coeff_t> lower(Ls,-1.0); lower[0]   =mass_plus;
+  Vector<Coeff_t> upper(Ls,-1.0); upper[Ls-1]=mass;
+  Vector<Coeff_t> lower(Ls,-1.0); lower[0]   =mass;
  M5D(psi,chi,chi,lower,diag,upper);
 }
 template<class Impl>
@@ -168,8 +220,8 @@ void CayleyFermion5D<Impl>::Meooe5D    (const FermionField &psi, FermionField &D
  Vector<Coeff_t> diag = bs;
  Vector<Coeff_t> upper= cs;
  Vector<Coeff_t> lower= cs; 
-  upper[Ls-1]=-mass_minus*upper[Ls-1];
-  lower[0]   =-mass_plus*lower[0];
+  upper[Ls-1]=-mass*upper[Ls-1];
+  lower[0]   =-mass*lower[0];
  M5D(psi,psi,Din,lower,diag,upper);
 }
 // FIXME Redunant with the above routine; check this and eliminate
@@ -183,8 +235,8 @@ template<class Impl> void CayleyFermion5D<Impl>::Meo5D     (const FermionField &
    upper[i]=-ceo[i];
    lower[i]=-ceo[i];
  }
-  upper[Ls-1]=-mass_minus*upper[Ls-1];
-  lower[0]   =-mass_plus*lower[0];
+  upper[Ls-1]=-mass*upper[Ls-1];
+  lower[0]   =-mass*lower[0];
  M5D(psi,psi,chi,lower,diag,upper);
 }
 template<class Impl>
@@ -198,8 +250,8 @@ void CayleyFermion5D<Impl>::Mooee       (const FermionField &psi, FermionField &
    upper[i]=-cee[i];
    lower[i]=-cee[i];
  }
-  upper[Ls-1]=-mass_minus*upper[Ls-1];
-  lower[0]   =-mass_plus*lower[0];
+  upper[Ls-1]=-mass*upper[Ls-1];
+  lower[0]   =-mass*lower[0];
  M5D(psi,psi,chi,lower,diag,upper);
 }
 template<class Impl>
@@ -214,9 +266,9 @@ void CayleyFermion5D<Impl>::MooeeDag    (const FermionField &psi, FermionField &
    // Assemble the 5d matrix
    if ( s==0 ) {
      upper[s] = -cee[s+1] ;
-      lower[s] = mass_minus*cee[Ls-1];
+      lower[s] = mass*cee[Ls-1];
    } else if ( s==(Ls-1)) { 
-      upper[s] = mass_plus*cee[0];
+      upper[s] = mass*cee[0];
      lower[s] = -cee[s-1];
    } else {
      upper[s]=-cee[s+1];
@@ -239,8 +291,8 @@ void CayleyFermion5D<Impl>::M5Ddag (const FermionField &psi, FermionField &chi)
  Vector<Coeff_t> diag(Ls,1.0);
  Vector<Coeff_t> upper(Ls,-1.0);
  Vector<Coeff_t> lower(Ls,-1.0);
-  upper[Ls-1]=-mass_plus*upper[Ls-1];
-  lower[0]   =-mass_minus*lower[0];
+  upper[Ls-1]=-mass*upper[Ls-1];
+  lower[0]   =-mass*lower[0];
  M5Ddag(psi,chi,chi,lower,diag,upper);
 }

@@ -255,9 +307,9 @@ void CayleyFermion5D<Impl>::MeooeDag5D    (const FermionField &psi, FermionField
  for (int s=0;s<Ls;s++){
    if ( s== 0 ) {
      upper[s] = cs[s+1];
-      lower[s] =-mass_minus*cs[Ls-1];
+      lower[s] =-mass*cs[Ls-1];
    } else if ( s==(Ls-1) ) { 
-      upper[s] =-mass_plus*cs[0];
+      upper[s] =-mass*cs[0];
      lower[s] = cs[s-1];
    } else { 
      upper[s] = cs[s+1];
@@ -500,7 +552,7 @@ void CayleyFermion5D<Impl>::SetCoefficientsInternal(RealD zolo_hi,Vector<Coeff_t
      
      lee[i] =-cee[i+1]/bee[i]; // sub-diag entry on the ith column
      
-      leem[i]=mass_minus*cee[Ls-1]/bee[0];
+      leem[i]=mass*cee[Ls-1]/bee[0];
      for(int j=0;j<i;j++) {
 	assert(bee[j+1]!=Coeff_t(0.0));
 	leem[i]*= aee[j]/bee[j+1];
@@ -508,7 +560,7 @@ void CayleyFermion5D<Impl>::SetCoefficientsInternal(RealD zolo_hi,Vector<Coeff_t
      
      uee[i] =-aee[i]/bee[i];   // up-diag entry on the ith row
      
-      ueem[i]=mass_plus;
+      ueem[i]=mass;
      for(int j=1;j<=i;j++) ueem[i]*= cee[j]/bee[j];
      ueem[i]*= aee[0]/bee[0];
      
@@ -521,7 +573,7 @@ void CayleyFermion5D<Impl>::SetCoefficientsInternal(RealD zolo_hi,Vector<Coeff_t
  }
 	
  { 
-    Coeff_t delta_d=mass_minus*cee[Ls-1];
+    Coeff_t delta_d=mass*cee[Ls-1];
    for(int j=0;j<Ls-1;j++) {
      assert(bee[j] != Coeff_t(0.0));
      delta_d *= cee[j]/bee[j];
@@ -590,10 +642,7 @@ void CayleyFermion5D<Impl>::ContractConservedCurrent( PropagatorField &q_in_1,
 						      Current curr_type,
 						      unsigned int mu)
 {
-
-  assert(mass_plus == mass_minus);
-  RealD mass = mass_plus;
-  
+#if (!defined(GRID_HIP))
  Gamma::Algebra Gmu [] = {
    Gamma::Algebra::GammaX,
    Gamma::Algebra::GammaY,
@@ -712,7 +761,7 @@ void CayleyFermion5D<Impl>::ContractConservedCurrent( PropagatorField &q_in_1,
    else          q_out +=     C;
    
  }
-
+#endif
 }

 template <class Impl>
@@ -728,8 +777,6 @@ void CayleyFermion5D<Impl>::SeqConservedCurrent(PropagatorField &q_in,
  assert(mu>=0);
  assert(mu<Nd);

-  assert(mass_plus == mass_minus);
-  RealD mass = mass_plus;

 #if 0
  int tshift = (mu == Nd-1) ? 1 : 0;
@@ -779,8 +826,8 @@ void CayleyFermion5D<Impl>::SeqConservedCurrent(PropagatorField &q_in,
  }
 #endif

+#if (!defined(GRID_HIP))
  int tshift = (mu == Nd-1) ? 1 : 0;
-  unsigned int LLt    = GridDefaultLatt()[Tp];
  ////////////////////////////////////////////////
  // GENERAL CAYLEY CASE
  ////////////////////////////////////////////////
@@ -833,7 +880,7 @@ void CayleyFermion5D<Impl>::SeqConservedCurrent(PropagatorField &q_in,
  }

  std::vector<RealD> G_s(Ls,1.0);
-  RealD sign = 1.0; // sign flip for vector/tadpole
+  Integer sign = 1; // sign flip for vector/tadpole
  if ( curr_type == Current::Axial ) {
    for(int s=0;s<Ls/2;s++){
      G_s[s] = -1.0;
@@ -843,7 +890,7 @@ void CayleyFermion5D<Impl>::SeqConservedCurrent(PropagatorField &q_in,
    auto b=this->_b;
    auto c=this->_c;
    if ( b == 1 && c == 0 ) {
-      sign = -1.0;    
+      sign = -1;    
    }
    else {
      std::cerr << "Error: Tadpole implementation currently unavailable for non-Shamir actions." << std::endl;
@@ -854,8 +901,8 @@ void CayleyFermion5D<Impl>::SeqConservedCurrent(PropagatorField &q_in,
  for(int s=0;s<Ls;s++){

    int sp = (s+1)%Ls;
-    //    int sr = Ls-1-s;
-    //    int srp= (sr+1)%Ls;
+    int sr = Ls-1-s;
+    int srp= (sr+1)%Ls;

    // Mobius parameters
    auto b=this->bs[s];
@@ -887,17 +934,12 @@ void CayleyFermion5D<Impl>::SeqConservedCurrent(PropagatorField &q_in,
    tmp    = Cshift(tmp,mu,-1);
    Impl::multLinkField(Utmp,this->Umu,tmp,mu+Nd); // Adjoint link
    tmp = -G_s[s]*( Utmp + gmu*Utmp );
-    // Mask the time
-    if (tmax == LLt - 1 && tshift == 1){ // quick fix to include timeslice 0 if tmax + tshift is over the last timeslice
-      unsigned int t0 = 0;
-      tmp    = where(((lcoor==t0) || (lcoor>=tmin+tshift)),tmp,zz);
-    } else {
-      tmp    = where((lcoor>=tmin+tshift),tmp,zz);
-    }
+    tmp    = where((lcoor>=tmin+tshift),tmp,zz); // Mask the time 
    L_Q   += where((lcoor<=tmax+tshift),tmp,zz); // Position of current complicated

    InsertSlice(L_Q, q_out, s , 0);
  }
+#endif
 }
 #undef Pp
 #undef Pm
@@ -905,6 +947,88 @@ void CayleyFermion5D<Impl>::SeqConservedCurrent(PropagatorField &q_in,
 #undef TopRowWithSource


+
+#if 0
+template<class Impl>
+void CayleyFermion5D<Impl>::MooeeInternalCompute(int dag, int inv,
+						 Vector<iSinglet<Simd> > & Matp,
+						 Vector<iSinglet<Simd> > & Matm)
+{
+  int Ls=this->Ls;
+
+  GridBase *grid = this->FermionRedBlackGrid();
+  int LLs = grid->_rdimensions[0];
+
+  if ( LLs == Ls ) {
+    return; // Not vectorised in 5th direction
+  }
+
+  Eigen::MatrixXcd Pplus  = Eigen::MatrixXcd::Zero(Ls,Ls);
+  Eigen::MatrixXcd Pminus = Eigen::MatrixXcd::Zero(Ls,Ls);
+  
+  for(int s=0;s<Ls;s++){
+    Pplus(s,s) = bee[s];
+    Pminus(s,s)= bee[s];
+  }
+  
+  for(int s=0;s<Ls-1;s++){
+    Pminus(s,s+1) = -cee[s];
+  }
+  
+  for(int s=0;s<Ls-1;s++){
+    Pplus(s+1,s) = -cee[s+1];
+  }
+  Pplus (0,Ls-1) = mass*cee[0];
+  Pminus(Ls-1,0) = mass*cee[Ls-1];
+  
+  Eigen::MatrixXcd PplusMat ;
+  Eigen::MatrixXcd PminusMat;
+  
+  if ( inv ) {
+    PplusMat =Pplus.inverse();
+    PminusMat=Pminus.inverse();
+  } else { 
+    PplusMat =Pplus;
+    PminusMat=Pminus;
+  }
+  
+  if(dag){
+    PplusMat.adjointInPlace();
+    PminusMat.adjointInPlace();
+  }
+  
+  typedef typename SiteHalfSpinor::scalar_type scalar_type;
+  const int Nsimd=Simd::Nsimd();
+  Matp.resize(Ls*LLs);
+  Matm.resize(Ls*LLs);
+
+  for(int s2=0;s2<Ls;s2++){
+    for(int s1=0;s1<LLs;s1++){
+      int istride = LLs;
+      int ostride = 1;
+      Simd Vp;
+      Simd Vm;
+      scalar_type *sp = (scalar_type *)&Vp;
+      scalar_type *sm = (scalar_type *)&Vm;
+      for(int l=0;l<Nsimd;l++){
+	if ( switcheroo<Coeff_t>::iscomplex() ) {
+	  sp[l] = PplusMat (l*istride+s1*ostride,s2);
+	  sm[l] = PminusMat(l*istride+s1*ostride,s2);
+	} else { 
+	  // if real
+	  scalar_type tmp;
+	  tmp = PplusMat (l*istride+s1*ostride,s2);
+	  sp[l] = scalar_type(tmp.real(),tmp.real());
+	  tmp = PminusMat(l*istride+s1*ostride,s2);
+	  sm[l] = scalar_type(tmp.real(),tmp.real());
+	}
+      }
+      Matp[LLs*s2+s1] = Vp;
+      Matm[LLs*s2+s1] = Vm;
+    }}
+}
+#endif
+
 NAMESPACE_END(Grid);


@@ -63,18 +63,23 @@ CayleyFermion5D<Impl>::M5D(const FermionField &psi_i,

  // 10 = 3 complex mult + 2 complex add
  // Flops = 10.0*(Nc*Ns) *Ls*vol (/2 for red black counting)
-  uint64_t nloop = grid->oSites();
+  M5Dcalls++;
+  M5Dtime-=usecond();
+
+  uint64_t nloop = grid->oSites()/Ls;
  accelerator_for(sss,nloop,Simd::Nsimd(),{
-    uint64_t s = sss%Ls;
-    uint64_t ss= sss-s;
+    uint64_t ss= sss*Ls;
    typedef decltype(coalescedRead(psi[0])) spinor;
    spinor tmp1, tmp2;
-    uint64_t idx_u = ss+((s+1)%Ls);
-    uint64_t idx_l = ss+((s+Ls-1)%Ls);
-    spProj5m(tmp1,psi(idx_u));
-    spProj5p(tmp2,psi(idx_l));
-    coalescedWrite(chi[ss+s],pdiag[s]*phi(ss+s)+pupper[s]*tmp1+plower[s]*tmp2);
+    for(int s=0;s<Ls;s++){
+      uint64_t idx_u = ss+((s+1)%Ls);
+      uint64_t idx_l = ss+((s+Ls-1)%Ls);
+      spProj5m(tmp1,psi(idx_u));
+      spProj5p(tmp2,psi(idx_l));
+      coalescedWrite(chi[ss+s],pdiag[s]*phi(ss+s)+pupper[s]*tmp1+plower[s]*tmp2);
+    }
  });
+  M5Dtime+=usecond();
 }

 template<class Impl>  
@@ -100,18 +105,23 @@ CayleyFermion5D<Impl>::M5Ddag(const FermionField &psi_i,
  int Ls=this->Ls;

  // Flops = 6.0*(Nc*Ns) *Ls*vol
-  uint64_t nloop = grid->oSites();
+  M5Dcalls++;
+  M5Dtime-=usecond();
+
+  uint64_t nloop = grid->oSites()/Ls;
  accelerator_for(sss,nloop,Simd::Nsimd(),{
-    uint64_t s = sss%Ls;
-    uint64_t ss= sss-s;
+    uint64_t ss=sss*Ls;
    typedef decltype(coalescedRead(psi[0])) spinor;
    spinor tmp1,tmp2;
-    uint64_t idx_u = ss+((s+1)%Ls);
-    uint64_t idx_l = ss+((s+Ls-1)%Ls);
-    spProj5p(tmp1,psi(idx_u));
-    spProj5m(tmp2,psi(idx_l));
-    coalescedWrite(chi[ss+s],pdiag[s]*phi(ss+s)+pupper[s]*tmp1+plower[s]*tmp2);
+    for(int s=0;s<Ls;s++){
+      uint64_t idx_u = ss+((s+1)%Ls);
+      uint64_t idx_l = ss+((s+Ls-1)%Ls);
+      spProj5p(tmp1,psi(idx_u));
+      spProj5m(tmp2,psi(idx_l));
+      coalescedWrite(chi[ss+s],pdiag[s]*phi(ss+s)+pupper[s]*tmp1+plower[s]*tmp2);
+    }
  });
+  M5Dtime+=usecond();
 }

 template<class Impl>
@@ -132,6 +142,8 @@ CayleyFermion5D<Impl>::MooeeInv    (const FermionField &psi_i, FermionField &chi
  auto pleem = & leem[0];
  auto pueem = & ueem[0];

+  MooeeInvCalls++;
+  MooeeInvTime-=usecond();
  uint64_t nloop = grid->oSites()/Ls;
  accelerator_for(sss,nloop,Simd::Nsimd(),{
    uint64_t ss=sss*Ls;
@@ -168,6 +180,8 @@ CayleyFermion5D<Impl>::MooeeInv    (const FermionField &psi_i, FermionField &chi
      coalescedWrite(chi[ss+s],res);
    }
  });
+
+  MooeeInvTime+=usecond();
  
 }

@@ -190,6 +204,10 @@ CayleyFermion5D<Impl>::MooeeInvDag (const FermionField &psi_i, FermionField &chi

  assert(psi.Checkerboard() == psi.Checkerboard());

+  MooeeInvCalls++;
+  MooeeInvTime-=usecond();
+
+
  uint64_t nloop = grid->oSites()/Ls;
  accelerator_for(sss,nloop,Simd::Nsimd(),{
    uint64_t ss=sss*Ls;
@@ -226,6 +244,7 @@ CayleyFermion5D<Impl>::MooeeInvDag (const FermionField &psi_i, FermionField &chi
      coalescedWrite(chi[ss+s],res);
    }
  });
+  MooeeInvTime+=usecond();

 }

@@ -94,6 +94,10 @@ CayleyFermion5D<Impl>::M5D(const FermionField &psi_i,
      d_p[ss] = diag[s];
    }}

+
+  M5Dcalls++;
+  M5Dtime-=usecond();
+
  assert(Nc==3);

  thread_loop( (int ss=0;ss<grid->oSites();ss+=LLs),{ // adds LLs
@@ -194,6 +198,7 @@ CayleyFermion5D<Impl>::M5D(const FermionField &psi_i,
    }
 #endif
  });
+  M5Dtime+=usecond();
 }

 template<class Impl>  
@@ -237,6 +242,8 @@ CayleyFermion5D<Impl>::M5Ddag(const FermionField &psi_i,
      d_p[ss] = diag[s];
    }}

+  M5Dcalls++;
+  M5Dtime-=usecond();
  thread_loop( (int ss=0;ss<grid->oSites();ss+=LLs),{ // adds LLs
 #if 0
    alignas(64) SiteHalfSpinor hp;
@@ -332,6 +339,7 @@ CayleyFermion5D<Impl>::M5Ddag(const FermionField &psi_i,
    }
 #endif
  });
+  M5Dtime+=usecond();
 }


@@ -805,6 +813,9 @@ CayleyFermion5D<Impl>::MooeeInternal(const FermionField &psi, FermionField &chi,
  }
  assert(_Matp->size()==Ls*LLs);

+  MooeeInvCalls++;
+  MooeeInvTime-=usecond();
+
  if ( switcheroo<Coeff_t>::iscomplex() ) {
    thread_loop( (auto site=0;site<vol;site++),{
      MooeeInternalZAsm(psi,chi,LLs,site,*_Matp,*_Matm);
@@ -814,7 +825,7 @@ CayleyFermion5D<Impl>::MooeeInternal(const FermionField &psi, FermionField &chi,
      MooeeInternalAsm(psi,chi,LLs,site,*_Matp,*_Matm);
    });
  }
-
+  MooeeInvTime+=usecond();
 }

 NAMESPACE_END(Grid);
@@ -1,373 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: ./lib/qcd/action/fermion/CompactWilsonCloverFermionImplementation.h
-
-    Copyright (C) 2017 - 2022
-
-    Author: paboyle <paboyle@ph.ed.ac.uk>
-    Author: Guido Cossu <guido.cossu@ed.ac.uk>
-    Author: Daniel Richtmann <daniel.richtmann@gmail.com>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-    *************************************************************************************/
-/*  END LEGAL */
-
-#include <Grid/Grid.h>
-#include <Grid/qcd/spin/Dirac.h>
-#include <Grid/qcd/action/fermion/CompactWilsonCloverFermion.h>
-
-
-NAMESPACE_BEGIN(Grid);
-template<class Impl, class CloverHelpers>
-CompactWilsonCloverFermion<Impl, CloverHelpers>::CompactWilsonCloverFermion(GaugeField& _Umu,
-                                                                            GridCartesian& Fgrid,
-                                                                            GridRedBlackCartesian& Hgrid,
-                                                                            const RealD _mass,
-                                                                            const RealD _csw_r,
-                                                                            const RealD _csw_t,
-                                                                            const RealD _cF,
-                                                                            const WilsonAnisotropyCoefficients& clover_anisotropy,
-                                                                            const ImplParams& impl_p)
-  : WilsonBase(_Umu, Fgrid, Hgrid, _mass, impl_p, clover_anisotropy)
-  , csw_r(_csw_r)
-  , csw_t(_csw_t)
-  , cF(_cF)
-  , open_boundaries(impl_p.boundary_phases[Nd-1] == 0.0)
-  , Diagonal(&Fgrid),        Triangle(&Fgrid)
-  , DiagonalEven(&Hgrid),    TriangleEven(&Hgrid)
-  , DiagonalOdd(&Hgrid),     TriangleOdd(&Hgrid)
-  , DiagonalInv(&Fgrid),     TriangleInv(&Fgrid)
-  , DiagonalInvEven(&Hgrid), TriangleInvEven(&Hgrid)
-  , DiagonalInvOdd(&Hgrid),  TriangleInvOdd(&Hgrid)
-  , Tmp(&Fgrid)
-  , BoundaryMask(&Fgrid)
-  , BoundaryMaskEven(&Hgrid), BoundaryMaskOdd(&Hgrid)
-{
-  assert(Nd == 4 && Nc == 3 && Ns == 4 && Impl::Dimension == 3);
-
-  csw_r *= 0.5;
-  csw_t *= 0.5;
-  if (clover_anisotropy.isAnisotropic)
-    csw_r /= clover_anisotropy.xi_0;
-
-  ImportGauge(_Umu);
-  if (open_boundaries) {
-    this->BoundaryMaskEven.Checkerboard() = Even;
-    this->BoundaryMaskOdd.Checkerboard() = Odd;
-    CompactHelpers::SetupMasks(this->BoundaryMask, this->BoundaryMaskEven, this->BoundaryMaskOdd);
-  }
-}
-
-template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl, CloverHelpers>::Dhop(const FermionField& in, FermionField& out, int dag) {
-  WilsonBase::Dhop(in, out, dag);
-  if(open_boundaries) ApplyBoundaryMask(out);
-}
-
-template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl, CloverHelpers>::DhopOE(const FermionField& in, FermionField& out, int dag) {
-  WilsonBase::DhopOE(in, out, dag);
-  if(open_boundaries) ApplyBoundaryMask(out);
-}
-
-template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl, CloverHelpers>::DhopEO(const FermionField& in, FermionField& out, int dag) {
-  WilsonBase::DhopEO(in, out, dag);
-  if(open_boundaries) ApplyBoundaryMask(out);
-}
-
-template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl, CloverHelpers>::DhopDir(const FermionField& in, FermionField& out, int dir, int disp) {
-  WilsonBase::DhopDir(in, out, dir, disp);
-  if(this->open_boundaries) ApplyBoundaryMask(out);
-}
-
-template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl, CloverHelpers>::DhopDirAll(const FermionField& in, std::vector<FermionField>& out) {
-  WilsonBase::DhopDirAll(in, out);
-  if(this->open_boundaries) {
-    for(auto& o : out) ApplyBoundaryMask(o);
-  }
-}
-
-template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl, CloverHelpers>::M(const FermionField& in, FermionField& out) {
-  out.Checkerboard() = in.Checkerboard();
-  WilsonBase::Dhop(in, out, DaggerNo); // call base to save applying bc
-  Mooee(in, Tmp);
-  axpy(out, 1.0, out, Tmp);
-  if(open_boundaries) ApplyBoundaryMask(out);
-}
-
-template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl, CloverHelpers>::Mdag(const FermionField& in, FermionField& out) {
-  out.Checkerboard() = in.Checkerboard();
-  WilsonBase::Dhop(in, out, DaggerYes);  // call base to save applying bc
-  MooeeDag(in, Tmp);
-  axpy(out, 1.0, out, Tmp);
-  if(open_boundaries) ApplyBoundaryMask(out);
-}
-
-template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl, CloverHelpers>::Meooe(const FermionField& in, FermionField& out) {
-  WilsonBase::Meooe(in, out);
-  if(open_boundaries) ApplyBoundaryMask(out);
-}
-
-template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl, CloverHelpers>::MeooeDag(const FermionField& in, FermionField& out) {
-  WilsonBase::MeooeDag(in, out);
-  if(open_boundaries) ApplyBoundaryMask(out);
-}
-
-template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl, CloverHelpers>::Mooee(const FermionField& in, FermionField& out) {
-  if(in.Grid()->_isCheckerBoarded) {
-    if(in.Checkerboard() == Odd) {
-      MooeeInternal(in, out, DiagonalOdd, TriangleOdd);
-    } else {
-      MooeeInternal(in, out, DiagonalEven, TriangleEven);
-    }
-  } else {
-    MooeeInternal(in, out, Diagonal, Triangle);
-  }
-  if(open_boundaries) ApplyBoundaryMask(out);
-}
-
-template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl, CloverHelpers>::MooeeDag(const FermionField& in, FermionField& out) {
-  Mooee(in, out); // blocks are hermitian
-}
-
-template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl, CloverHelpers>::MooeeInv(const FermionField& in, FermionField& out) {
-  if(in.Grid()->_isCheckerBoarded) {
-    if(in.Checkerboard() == Odd) {
-      MooeeInternal(in, out, DiagonalInvOdd, TriangleInvOdd);
-    } else {
-      MooeeInternal(in, out, DiagonalInvEven, TriangleInvEven);
-    }
-  } else {
-    MooeeInternal(in, out, DiagonalInv, TriangleInv);
-  }
-  if(open_boundaries) ApplyBoundaryMask(out);
-}
-
-template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl, CloverHelpers>::MooeeInvDag(const FermionField& in, FermionField& out) {
-  MooeeInv(in, out); // blocks are hermitian
-}
-
-template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl, CloverHelpers>::Mdir(const FermionField& in, FermionField& out, int dir, int disp) {
-  DhopDir(in, out, dir, disp);
-}
-
-template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl, CloverHelpers>::MdirAll(const FermionField& in, std::vector<FermionField>& out) {
-  DhopDirAll(in, out);
-}
-
-template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl, CloverHelpers>::MDeriv(GaugeField& force, const FermionField& X, const FermionField& Y, int dag) {
-  assert(!open_boundaries); // TODO check for changes required for open bc
-
-  // NOTE: code copied from original clover term
-  conformable(X.Grid(), Y.Grid());
-  conformable(X.Grid(), force.Grid());
-  GaugeLinkField force_mu(force.Grid()), lambda(force.Grid());
-  GaugeField clover_force(force.Grid());
-  PropagatorField Lambda(force.Grid());
-
-  // Guido: Here we are hitting some performance issues:
-  // need to extract the components of the DoubledGaugeField
-  // for each call
-  // Possible solution
-  // Create a vector object to store them? (cons: wasting space)
-  std::vector<GaugeLinkField> U(Nd, this->Umu.Grid());
-
-  Impl::extractLinkField(U, this->Umu);
-
-  force = Zero();
-  // Derivative of the Wilson hopping term
-  this->DhopDeriv(force, X, Y, dag);
-
-  ///////////////////////////////////////////////////////////
-  // Clover term derivative
-  ///////////////////////////////////////////////////////////
-  Impl::outerProductImpl(Lambda, X, Y);
-  //std::cout << "Lambda:" << Lambda << std::endl;
-
-  Gamma::Algebra sigma[] = {
-      Gamma::Algebra::SigmaXY,
-      Gamma::Algebra::SigmaXZ,
-      Gamma::Algebra::SigmaXT,
-      Gamma::Algebra::MinusSigmaXY,
-      Gamma::Algebra::SigmaYZ,
-      Gamma::Algebra::SigmaYT,
-      Gamma::Algebra::MinusSigmaXZ,
-      Gamma::Algebra::MinusSigmaYZ,
-      Gamma::Algebra::SigmaZT,
-      Gamma::Algebra::MinusSigmaXT,
-      Gamma::Algebra::MinusSigmaYT,
-      Gamma::Algebra::MinusSigmaZT};
-
-  /*
-    sigma_{\mu \nu}=
-    | 0         sigma[0]  sigma[1]  sigma[2] |
-    | sigma[3]    0       sigma[4]  sigma[5] |
-    | sigma[6]  sigma[7]     0      sigma[8] |
-    | sigma[9]  sigma[10] sigma[11]   0      |
-  */
-
-  int count = 0;
-  clover_force = Zero();
-  for (int mu = 0; mu < 4; mu++)
-  {
-    force_mu = Zero();
-    for (int nu = 0; nu < 4; nu++)
-    {
-      if (mu == nu)
-        continue;
-
-      RealD factor;
-      if (nu == 4 || mu == 4)
-      {
-        factor = 2.0 * csw_t;
-      }
-      else
-      {
-        factor = 2.0 * csw_r;
-      }
-      PropagatorField Slambda = Gamma(sigma[count]) * Lambda; // sigma checked
-      Impl::TraceSpinImpl(lambda, Slambda);                   // traceSpin ok
-      force_mu -= factor*CloverHelpers::Cmunu(U, lambda, mu, nu);   // checked
-      count++;
-    }
-
-    pokeLorentz(clover_force, U[mu] * force_mu, mu);
-  }
-  //clover_force *= csw;
-  force += clover_force;
-}
-
-template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl, CloverHelpers>::MooDeriv(GaugeField& mat, const FermionField& U, const FermionField& V, int dag) {
-  assert(0);
-}
-
-template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl, CloverHelpers>::MeeDeriv(GaugeField& mat, const FermionField& U, const FermionField& V, int dag) {
-  assert(0);
-}
-
-template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl, CloverHelpers>::MooeeInternal(const FermionField&        in,
-                    FermionField&              out,
-                    const CloverDiagonalField& diagonal,
-                    const CloverTriangleField& triangle) {
-  assert(in.Checkerboard() == Odd || in.Checkerboard() == Even);
-  out.Checkerboard() = in.Checkerboard();
-  conformable(in, out);
-  conformable(in, diagonal);
-  conformable(in, triangle);
-
-  CompactHelpers::MooeeKernel(diagonal.oSites(), 1, in, out, diagonal, triangle);
-}
-
-template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl, CloverHelpers>::ImportGauge(const GaugeField& _Umu) {
-  // NOTE: parts copied from original implementation
-
-  // Import gauge into base class
-  double t0 = usecond();
-  WilsonBase::ImportGauge(_Umu); // NOTE: called here and in wilson constructor -> performed twice, but can't avoid that
-
-  // Initialize temporary variables
-  double t1 = usecond();
-  conformable(_Umu.Grid(), this->GaugeGrid());
-  GridBase* grid = _Umu.Grid();
-  typename Impl::GaugeLinkField Bx(grid), By(grid), Bz(grid), Ex(grid), Ey(grid), Ez(grid);
-  CloverField TmpOriginal(grid);
-
-  // Compute the field strength terms mu>nu
-  double t2 = usecond();
-  WilsonLoops<Impl>::FieldStrength(Bx, _Umu, Zdir, Ydir);
-  WilsonLoops<Impl>::FieldStrength(By, _Umu, Zdir, Xdir);
-  WilsonLoops<Impl>::FieldStrength(Bz, _Umu, Ydir, Xdir);
-  WilsonLoops<Impl>::FieldStrength(Ex, _Umu, Tdir, Xdir);
-  WilsonLoops<Impl>::FieldStrength(Ey, _Umu, Tdir, Ydir);
-  WilsonLoops<Impl>::FieldStrength(Ez, _Umu, Tdir, Zdir);
-
-  // Compute the Clover Operator acting on Colour and Spin
-  // multiply here by the clover coefficients for the anisotropy
-  double t3 = usecond();
-  TmpOriginal  = Helpers::fillCloverYZ(Bx) * csw_r;
-  TmpOriginal += Helpers::fillCloverXZ(By) * csw_r;
-  TmpOriginal += Helpers::fillCloverXY(Bz) * csw_r;
-  TmpOriginal += Helpers::fillCloverXT(Ex) * csw_t;
-  TmpOriginal += Helpers::fillCloverYT(Ey) * csw_t;
-  TmpOriginal += Helpers::fillCloverZT(Ez) * csw_t;
-  // Handle mass term based on clover policy
-  CloverHelpers::MassTerm(TmpOriginal, this->diag_mass);
-  
-  // Convert the data layout of the clover term
-  double t4 = usecond();
-  CompactHelpers::ConvertLayout(TmpOriginal, Diagonal, Triangle);
-
-  // Exponentiate the clover (nothing happens in case of the standard clover)
-  double t5 = usecond();
-  CloverHelpers::Exponentiate_Clover(Diagonal, Triangle, csw_t, this->diag_mass);
-
-  // Possible modify the boundary values
-  double t6 = usecond();
-  if(open_boundaries) CompactHelpers::ModifyBoundaries(Diagonal, Triangle, csw_t, cF, this->diag_mass);
-
-  // Invert the Clover term (explicit inversion needed for the improvement in case of open boundary conditions)
-  double t7 = usecond();
-  CompactHelpers::Invert(Diagonal, Triangle, DiagonalInv, TriangleInv);
-
-  // Fill the remaining clover fields
-  double t8 = usecond();
-  pickCheckerboard(Even, DiagonalEven,    Diagonal);
-  pickCheckerboard(Even, TriangleEven,    Triangle);
-  pickCheckerboard(Odd,  DiagonalOdd,     Diagonal);
-  pickCheckerboard(Odd,  TriangleOdd,     Triangle);
-  pickCheckerboard(Even, DiagonalInvEven, DiagonalInv);
-  pickCheckerboard(Even, TriangleInvEven, TriangleInv);
-  pickCheckerboard(Odd,  DiagonalInvOdd,  DiagonalInv);
-  pickCheckerboard(Odd,  TriangleInvOdd,  TriangleInv);
-
-  // Report timings
-  double t9 = usecond();
-
-  std::cout << GridLogDebug << "CompactWilsonCloverFermion::ImportGauge timings:" << std::endl;
-  std::cout << GridLogDebug << "WilsonFermion::Importgauge = " << (t1 - t0) / 1e6 << std::endl;
-  std::cout << GridLogDebug << "allocations =                " << (t2 - t1) / 1e6 << std::endl;
-  std::cout << GridLogDebug << "field strength =             " << (t3 - t2) / 1e6 << std::endl;
-  std::cout << GridLogDebug << "fill clover =                " << (t4 - t3) / 1e6 << std::endl;
-  std::cout << GridLogDebug << "convert =                    " << (t5 - t4) / 1e6 << std::endl;
-  std::cout << GridLogDebug << "exponentiation =             " << (t6 - t5) / 1e6 << std::endl;
-  std::cout << GridLogDebug << "boundaries =                 " << (t7 - t6) / 1e6 << std::endl;
-  std::cout << GridLogDebug << "inversions =                 " << (t8 - t7) / 1e6 << std::endl;
-  std::cout << GridLogDebug << "pick cbs =                   " << (t9 - t8) / 1e6 << std::endl;
-  std::cout << GridLogDebug << "total =                      " << (t9 - t0) / 1e6 << std::endl;
-}
-
-NAMESPACE_END(Grid);
@@ -54,6 +54,8 @@ void DomainWallEOFAFermion<Impl>::M5D(const FermionField& psi_i, const FermionFi
  auto pupper = &upper[0];
  auto plower = &lower[0];
  // Flops = 6.0*(Nc*Ns) *Ls*vol
+  this->M5Dcalls++;
+  this->M5Dtime -= usecond();
  
  auto nloop=grid->oSites()/Ls;
  accelerator_for(sss,nloop,Simd::Nsimd(),{
@@ -69,6 +71,7 @@ void DomainWallEOFAFermion<Impl>::M5D(const FermionField& psi_i, const FermionFi
    }
  });

+  this->M5Dtime += usecond();
 }

 template<class Impl>
@@ -88,6 +91,8 @@ void DomainWallEOFAFermion<Impl>::M5Ddag(const FermionField& psi_i, const Fermio
  auto plower = &lower[0];

  // Flops = 6.0*(Nc*Ns) *Ls*vol
+  this->M5Dcalls++;
+  this->M5Dtime -= usecond();

  auto nloop=grid->oSites()/Ls;
  accelerator_for(sss,nloop,Simd::Nsimd(),{
@@ -103,6 +108,7 @@ void DomainWallEOFAFermion<Impl>::M5Ddag(const FermionField& psi_i, const Fermio
    }
  });

+  this->M5Dtime += usecond();
 }

 template<class Impl>
@@ -121,6 +127,8 @@ void DomainWallEOFAFermion<Impl>::MooeeInv(const FermionField& psi_i, FermionFie
  auto pleem = & this->leem[0];
  auto pueem = & this->ueem[0];

+  this->MooeeInvCalls++;
+  this->MooeeInvTime -= usecond();
  uint64_t nloop=grid->oSites()/Ls;
  accelerator_for(sss,nloop,Simd::Nsimd(),{
    uint64_t ss=sss*Ls;
@@ -156,6 +164,7 @@ void DomainWallEOFAFermion<Impl>::MooeeInv(const FermionField& psi_i, FermionFie
      coalescedWrite(chi[ss+s],res);
    }
  });
+  this->MooeeInvTime += usecond();
 }

 template<class Impl>
@@ -176,6 +185,8 @@ void DomainWallEOFAFermion<Impl>::MooeeInvDag(const FermionField& psi_i, Fermion

  assert(psi.Checkerboard() == psi.Checkerboard());

+  this->MooeeInvCalls++;
+  this->MooeeInvTime -= usecond();
  auto nloop = grid->oSites()/Ls;
  accelerator_for(sss,nloop,Simd::Nsimd(),{
    uint64_t ss=sss*Ls;
@@ -212,6 +223,7 @@ void DomainWallEOFAFermion<Impl>::MooeeInvDag(const FermionField& psi_i, Fermion
    }
  });

+  this->MooeeInvTime += usecond();
 }

 NAMESPACE_END(Grid);
@@ -298,33 +298,45 @@ void ImprovedStaggeredFermion5D<Impl>::DhopInternalOverlappedComms(StencilImpl &
  int LLs = in.Grid()->_rdimensions[0];
  int len =  U.Grid()->oSites();

+  DhopFaceTime-=usecond();
  st.Prepare();
  st.HaloGather(in,compressor);
+  DhopFaceTime+=usecond();

+  DhopCommTime -=usecond();
  std::vector<std::vector<CommsRequest_t> > requests;
  st.CommunicateBegin(requests);

  //  st.HaloExchangeOptGather(in,compressor); // Wilson compressor
+  DhopFaceTime-=usecond();
  st.CommsMergeSHM(compressor);// Could do this inside parallel region overlapped with comms
+  DhopFaceTime+=usecond();

  //////////////////////////////////////////////////////////////////////////////////////////////////////
  // Remove explicit thread mapping introduced for OPA reasons.
  //////////////////////////////////////////////////////////////////////////////////////////////////////
+  DhopComputeTime-=usecond();
  {
    int interior=1;
    int exterior=0;
    Kernels::DhopImproved(st,lo,U,UUU,in,out,dag,interior,exterior);
  }
+  DhopComputeTime+=usecond();

+  DhopFaceTime-=usecond();
  st.CommsMerge(compressor);
+  DhopFaceTime+=usecond();

  st.CommunicateComplete(requests);
+  DhopCommTime +=usecond();

+  DhopComputeTime2-=usecond();
  {
    int interior=0;
    int exterior=1;
    Kernels::DhopImproved(st,lo,U,UUU,in,out,dag,interior,exterior);
  }
+  DhopComputeTime2+=usecond();
 }

 template<class Impl>
@@ -335,14 +347,22 @@ void ImprovedStaggeredFermion5D<Impl>::DhopInternalSerialComms(StencilImpl & st,
  Compressor compressor;
  int LLs = in.Grid()->_rdimensions[0];

+ //double t1=usecond();
+  DhopTotalTime -= usecond();
+  DhopCommTime -= usecond();
  st.HaloExchange(in,compressor);
+  DhopCommTime += usecond();
  
+  DhopComputeTime -= usecond();
  // Dhop takes the 4d grid from U, and makes a 5d index for fermion
  {
    int interior=1;
    int exterior=1;
    Kernels::DhopImproved(st,lo,U,UUU,in,out,dag,interior,exterior);
  }
+  DhopComputeTime += usecond();
+  DhopTotalTime   += usecond();
+
 }
 /*CHANGE END*/

@@ -351,6 +371,7 @@ void ImprovedStaggeredFermion5D<Impl>::DhopInternalSerialComms(StencilImpl & st,
 template<class Impl>
 void ImprovedStaggeredFermion5D<Impl>::DhopOE(const FermionField &in, FermionField &out,int dag)
 {
+  DhopCalls+=1;
  conformable(in.Grid(),FermionRedBlackGrid());    // verifies half grid
  conformable(in.Grid(),out.Grid()); // drops the cb check

@@ -362,6 +383,7 @@ void ImprovedStaggeredFermion5D<Impl>::DhopOE(const FermionField &in, FermionFie
 template<class Impl>
 void ImprovedStaggeredFermion5D<Impl>::DhopEO(const FermionField &in, FermionField &out,int dag)
 {
+  DhopCalls+=1;
  conformable(in.Grid(),FermionRedBlackGrid());    // verifies half grid
  conformable(in.Grid(),out.Grid()); // drops the cb check

@@ -373,6 +395,7 @@ void ImprovedStaggeredFermion5D<Impl>::DhopEO(const FermionField &in, FermionFie
 template<class Impl>
 void ImprovedStaggeredFermion5D<Impl>::Dhop(const FermionField &in, FermionField &out,int dag)
 {
+  DhopCalls+=2;
  conformable(in.Grid(),FermionGrid()); // verifies full grid
  conformable(in.Grid(),out.Grid());

@@ -381,6 +404,58 @@ void ImprovedStaggeredFermion5D<Impl>::Dhop(const FermionField &in, FermionField
  DhopInternal(Stencil,Lebesgue,Umu,UUUmu,in,out,dag);
 }

+template<class Impl>
+void ImprovedStaggeredFermion5D<Impl>::Report(void) 
+{
+  Coordinate latt = GridDefaultLatt();          
+  RealD volume = Ls;  for(int mu=0;mu<Nd;mu++) volume=volume*latt[mu];
+  RealD NP = _FourDimGrid->_Nprocessors;
+  RealD NN = _FourDimGrid->NodeCount();
+
+  std::cout << GridLogMessage << "#### Dhop calls report " << std::endl;
+
+  std::cout << GridLogMessage << "ImprovedStaggeredFermion5D Number of DhopEO Calls   : " 
+	    << DhopCalls   << std::endl;
+  std::cout << GridLogMessage << "ImprovedStaggeredFermion5D TotalTime   /Calls       : " 
+	    << DhopTotalTime   / DhopCalls << " us" << std::endl;
+  std::cout << GridLogMessage << "ImprovedStaggeredFermion5D CommTime    /Calls       : " 
+	    << DhopCommTime    / DhopCalls << " us" << std::endl;
+  std::cout << GridLogMessage << "ImprovedStaggeredFermion5D ComputeTime/Calls        : " 
+	    << DhopComputeTime / DhopCalls << " us" << std::endl;
+
+  // Average the compute time
+  _FourDimGrid->GlobalSum(DhopComputeTime);
+  DhopComputeTime/=NP;
+
+  RealD mflops = 1154*volume*DhopCalls/DhopComputeTime/2; // 2 for red black counting
+  std::cout << GridLogMessage << "Average mflops/s per call                : " << mflops << std::endl;
+  std::cout << GridLogMessage << "Average mflops/s per call per rank       : " << mflops/NP << std::endl;
+  std::cout << GridLogMessage << "Average mflops/s per call per node       : " << mflops/NN << std::endl;
+  
+  RealD Fullmflops = 1154*volume*DhopCalls/(DhopTotalTime)/2; // 2 for red black counting
+  std::cout << GridLogMessage << "Average mflops/s per call (full)         : " << Fullmflops << std::endl;
+  std::cout << GridLogMessage << "Average mflops/s per call per rank (full): " << Fullmflops/NP << std::endl;
+  std::cout << GridLogMessage << "Average mflops/s per call per node (full): " << Fullmflops/NN << std::endl;
+
+  std::cout << GridLogMessage << "ImprovedStaggeredFermion5D Stencil"    <<std::endl;  Stencil.Report();
+  std::cout << GridLogMessage << "ImprovedStaggeredFermion5D StencilEven"<<std::endl;  StencilEven.Report();
+  std::cout << GridLogMessage << "ImprovedStaggeredFermion5D StencilOdd" <<std::endl;  StencilOdd.Report();
+}
+template<class Impl>
+void ImprovedStaggeredFermion5D<Impl>::ZeroCounters(void) 
+{
+  DhopCalls       = 0;
+  DhopTotalTime    = 0;
+  DhopCommTime    = 0;
+  DhopComputeTime = 0;
+  DhopFaceTime    = 0;
+
+
+  Stencil.ZeroCounters();
+  StencilEven.ZeroCounters();
+  StencilOdd.ZeroCounters();
+}
+
 /////////////////////////////////////////////////////////////////////////
 // Implement the general interface. Here we use SAME mass on all slices
 /////////////////////////////////////////////////////////////////////////
@@ -334,6 +334,7 @@ void ImprovedStaggeredFermion<Impl>::DhopDerivEO(GaugeField &mat, const FermionF
 template <class Impl>
 void ImprovedStaggeredFermion<Impl>::Dhop(const FermionField &in, FermionField &out, int dag) 
 {
+  DhopCalls+=2;
  conformable(in.Grid(), _grid);  // verifies full grid
  conformable(in.Grid(), out.Grid());

@@ -345,6 +346,7 @@ void ImprovedStaggeredFermion<Impl>::Dhop(const FermionField &in, FermionField &
 template <class Impl>
 void ImprovedStaggeredFermion<Impl>::DhopOE(const FermionField &in, FermionField &out, int dag) 
 {
+  DhopCalls+=1;
  conformable(in.Grid(), _cbgrid);    // verifies half grid
  conformable(in.Grid(), out.Grid());  // drops the cb check

@@ -357,6 +359,7 @@ void ImprovedStaggeredFermion<Impl>::DhopOE(const FermionField &in, FermionField
 template <class Impl>
 void ImprovedStaggeredFermion<Impl>::DhopEO(const FermionField &in, FermionField &out, int dag) 
 {
+  DhopCalls+=1;
  conformable(in.Grid(), _cbgrid);    // verifies half grid
  conformable(in.Grid(), out.Grid());  // drops the cb check

@@ -415,33 +418,47 @@ void ImprovedStaggeredFermion<Impl>::DhopInternalOverlappedComms(StencilImpl &st
  Compressor compressor; 
  int len =  U.Grid()->oSites();

+  DhopTotalTime   -= usecond();
+
+  DhopFaceTime    -= usecond();
  st.Prepare();
  st.HaloGather(in,compressor);
+  DhopFaceTime    += usecond();

+  DhopCommTime -=usecond();
  std::vector<std::vector<CommsRequest_t> > requests;
  st.CommunicateBegin(requests);

+  DhopFaceTime-=usecond();
  st.CommsMergeSHM(compressor);
+  DhopFaceTime+= usecond();

  //////////////////////////////////////////////////////////////////////////////////////////////////////
  // Removed explicit thread comms
  //////////////////////////////////////////////////////////////////////////////////////////////////////
+  DhopComputeTime    -= usecond();
  {
    int interior=1;
    int exterior=0;
    Kernels::DhopImproved(st,lo,U,UUU,in,out,dag,interior,exterior);
  }
+  DhopComputeTime    += usecond();

  st.CommunicateComplete(requests);
+  DhopCommTime +=usecond();

  // First to enter, last to leave timing
+  DhopFaceTime    -= usecond();
  st.CommsMerge(compressor);
+  DhopFaceTime    -= usecond();

+  DhopComputeTime2    -= usecond();
  {
    int interior=0;
    int exterior=1;
    Kernels::DhopImproved(st,lo,U,UUU,in,out,dag,interior,exterior);
  }
+  DhopComputeTime2    += usecond();
 }


@@ -454,16 +471,78 @@ void ImprovedStaggeredFermion<Impl>::DhopInternalSerialComms(StencilImpl &st, Le
 {
  assert((dag == DaggerNo) || (dag == DaggerYes));

+  DhopTotalTime   -= usecond();
+
+  DhopCommTime    -= usecond();
  Compressor compressor;
  st.HaloExchange(in, compressor);
+  DhopCommTime    += usecond();

+  DhopComputeTime -= usecond();
  {
    int interior=1;
    int exterior=1;
    Kernels::DhopImproved(st,lo,U,UUU,in,out,dag,interior,exterior);
  }
+  DhopComputeTime += usecond();
+  DhopTotalTime   += usecond();
 };

+  ////////////////////////////////////////////////////////////////
+  // Reporting
+  ////////////////////////////////////////////////////////////////
+template<class Impl>
+void ImprovedStaggeredFermion<Impl>::Report(void) 
+{
+  Coordinate latt = _grid->GlobalDimensions();
+  RealD volume = 1;  for(int mu=0;mu<Nd;mu++) volume=volume*latt[mu];
+  RealD NP = _grid->_Nprocessors;
+  RealD NN = _grid->NodeCount();
+
+  std::cout << GridLogMessage << "#### Dhop calls report " << std::endl;
+
+  std::cout << GridLogMessage << "ImprovedStaggeredFermion Number of DhopEO Calls   : " 
+	    << DhopCalls   << std::endl;
+  std::cout << GridLogMessage << "ImprovedStaggeredFermion TotalTime   /Calls       : " 
+	    << DhopTotalTime   / DhopCalls << " us" << std::endl;
+  std::cout << GridLogMessage << "ImprovedStaggeredFermion CommTime    /Calls       : " 
+	    << DhopCommTime    / DhopCalls << " us" << std::endl;
+  std::cout << GridLogMessage << "ImprovedStaggeredFermion ComputeTime/Calls        : " 
+	    << DhopComputeTime / DhopCalls << " us" << std::endl;
+
+  // Average the compute time
+  _grid->GlobalSum(DhopComputeTime);
+  DhopComputeTime/=NP;
+
+  RealD mflops = 1154*volume*DhopCalls/DhopComputeTime/2; // 2 for red black counting
+  std::cout << GridLogMessage << "Average mflops/s per call                : " << mflops << std::endl;
+  std::cout << GridLogMessage << "Average mflops/s per call per rank       : " << mflops/NP << std::endl;
+  std::cout << GridLogMessage << "Average mflops/s per call per node       : " << mflops/NN << std::endl;
+  
+  RealD Fullmflops = 1154*volume*DhopCalls/(DhopTotalTime)/2; // 2 for red black counting
+  std::cout << GridLogMessage << "Average mflops/s per call (full)         : " << Fullmflops << std::endl;
+  std::cout << GridLogMessage << "Average mflops/s per call per rank (full): " << Fullmflops/NP << std::endl;
+  std::cout << GridLogMessage << "Average mflops/s per call per node (full): " << Fullmflops/NN << std::endl;
+
+  std::cout << GridLogMessage << "ImprovedStaggeredFermion Stencil"    <<std::endl;  Stencil.Report();
+  std::cout << GridLogMessage << "ImprovedStaggeredFermion StencilEven"<<std::endl;  StencilEven.Report();
+  std::cout << GridLogMessage << "ImprovedStaggeredFermion StencilOdd" <<std::endl;  StencilOdd.Report();
+}
+template<class Impl>
+void ImprovedStaggeredFermion<Impl>::ZeroCounters(void) 
+{
+  DhopCalls       = 0;
+  DhopTotalTime   = 0;
+  DhopCommTime    = 0;
+  DhopComputeTime = 0;
+  DhopFaceTime    = 0;
+
+  Stencil.ZeroCounters();
+  StencilEven.ZeroCounters();
+  StencilOdd.ZeroCounters();
+}
+
+
 //////////////////////////////////////////////////////// 
 // Conserved current - not yet implemented.
 ////////////////////////////////////////////////////////
@@ -55,6 +55,9 @@ void MobiusEOFAFermion<Impl>::M5D(const FermionField &psi_i, const FermionField
  auto plower = &lower[0];

  // Flops = 6.0*(Nc*Ns) *Ls*vol
+  this->M5Dcalls++;
+  this->M5Dtime -= usecond();
+
  int nloop = grid->oSites()/Ls;
  accelerator_for(sss,nloop,Simd::Nsimd(),{
    uint64_t ss = sss*Ls;
@@ -70,6 +73,7 @@ void MobiusEOFAFermion<Impl>::M5D(const FermionField &psi_i, const FermionField
    }
  });

+  this->M5Dtime += usecond();
 }

 template<class Impl>
@@ -95,6 +99,9 @@ void MobiusEOFAFermion<Impl>::M5D_shift(const FermionField &psi_i, const Fermion
  auto pshift_coeffs = &shift_coeffs[0];

  // Flops = 6.0*(Nc*Ns) *Ls*vol
+  this->M5Dcalls++;
+  this->M5Dtime -= usecond();
+
  int nloop = grid->oSites()/Ls;
  accelerator_for(sss,nloop,Simd::Nsimd(),{
    uint64_t ss = sss*Ls;
@@ -115,6 +122,7 @@ void MobiusEOFAFermion<Impl>::M5D_shift(const FermionField &psi_i, const Fermion
    }
  });

+  this->M5Dtime += usecond();
 }

 template<class Impl>
@@ -135,6 +143,9 @@ void MobiusEOFAFermion<Impl>::M5Ddag(const FermionField &psi_i, const FermionFie
  auto plower = &lower[0];

  // Flops = 6.0*(Nc*Ns) *Ls*vol
+  this->M5Dcalls++;
+  this->M5Dtime -= usecond();
+
  int nloop = grid->oSites()/Ls;
  accelerator_for(sss,nloop,Simd::Nsimd(), {
    uint64_t ss = sss*Ls;
@@ -150,6 +161,8 @@ void MobiusEOFAFermion<Impl>::M5Ddag(const FermionField &psi_i, const FermionFie
      coalescedWrite(chi[ss+s], pdiag[s]*phi(ss+s) + pupper[s]*tmp1 + plower[s]*tmp2);
    }
  });
+
+  this->M5Dtime += usecond();
 }

 template<class Impl>
@@ -173,6 +186,9 @@ void MobiusEOFAFermion<Impl>::M5Ddag_shift(const FermionField &psi_i, const Ferm
  auto pshift_coeffs = &shift_coeffs[0];

  // Flops = 6.0*(Nc*Ns) *Ls*vol
+  this->M5Dcalls++;
+  this->M5Dtime -= usecond();
+
  auto pm = this->pm;

  int nloop = grid->oSites()/Ls;
@@ -201,6 +217,7 @@ void MobiusEOFAFermion<Impl>::M5Ddag_shift(const FermionField &psi_i, const Ferm
    }
  });

+  this->M5Dtime += usecond();
 }

 template<class Impl>
@@ -220,6 +237,9 @@ void MobiusEOFAFermion<Impl>::MooeeInv(const FermionField &psi_i, FermionField &

  if(this->shift != 0.0){ MooeeInv_shift(psi_i,chi_i); return; }

+  this->MooeeInvCalls++;
+  this->MooeeInvTime -= usecond();
+
  int nloop = grid->oSites()/Ls;
  accelerator_for(sss,nloop,Simd::Nsimd(),{
    uint64_t ss=sss*Ls;
@@ -257,6 +277,7 @@ void MobiusEOFAFermion<Impl>::MooeeInv(const FermionField &psi_i, FermionField &
    }
  });
   
+  this->MooeeInvTime += usecond();
 }

 template<class Impl>
@@ -276,6 +297,8 @@ void MobiusEOFAFermion<Impl>::MooeeInv_shift(const FermionField &psi_i, FermionF
  auto pueem= & this->ueem[0];
  auto pMooeeInv_shift_lc   = &MooeeInv_shift_lc[0];
  auto pMooeeInv_shift_norm = &MooeeInv_shift_norm[0];
+  this->MooeeInvCalls++;
+  this->MooeeInvTime -= usecond();

  int nloop = grid->oSites()/Ls;
  accelerator_for(sss,nloop,Simd::Nsimd(),{
@@ -320,6 +343,7 @@ void MobiusEOFAFermion<Impl>::MooeeInv_shift(const FermionField &psi_i, FermionF
      }
  });

+  this->MooeeInvTime += usecond();
 }

 template<class Impl>
@@ -339,6 +363,9 @@ void MobiusEOFAFermion<Impl>::MooeeInvDag(const FermionField &psi_i, FermionFiel
  auto pleem= & this->leem[0];
  auto pueem= & this->ueem[0];

+  this->MooeeInvCalls++;
+  this->MooeeInvTime -= usecond();
+
  int nloop = grid->oSites()/Ls;
  accelerator_for(sss,nloop,Simd::Nsimd(),{
    uint64_t ss=sss*Ls;
@@ -375,6 +402,7 @@ void MobiusEOFAFermion<Impl>::MooeeInvDag(const FermionField &psi_i, FermionFiel
      coalescedWrite(chi[ss+s],res);
    }
  });
+  this->MooeeInvTime += usecond();
 }

 template<class Impl>
@@ -395,6 +423,9 @@ void MobiusEOFAFermion<Impl>::MooeeInvDag_shift(const FermionField &psi_i, Fermi
  auto pMooeeInvDag_shift_lc   = &MooeeInvDag_shift_lc[0];
  auto pMooeeInvDag_shift_norm = &MooeeInvDag_shift_norm[0];

+  this->MooeeInvCalls++;
+  this->MooeeInvTime -= usecond();
+
  int nloop = grid->oSites()/Ls;
  accelerator_for(sss,nloop,Simd::Nsimd(),{
      uint64_t ss=sss*Ls;
@@ -438,6 +469,7 @@ void MobiusEOFAFermion<Impl>::MooeeInvDag_shift(const FermionField &psi_i, Fermi
      }
  });

+  this->MooeeInvTime += usecond();
 }

 NAMESPACE_END(Grid);
@@ -263,6 +263,7 @@ void NaiveStaggeredFermion<Impl>::DhopDerivEO(GaugeField &mat, const FermionFiel
 template <class Impl>
 void NaiveStaggeredFermion<Impl>::Dhop(const FermionField &in, FermionField &out, int dag) 
 {
+  DhopCalls+=2;
  conformable(in.Grid(), _grid);  // verifies full grid
  conformable(in.Grid(), out.Grid());

@@ -274,6 +275,7 @@ void NaiveStaggeredFermion<Impl>::Dhop(const FermionField &in, FermionField &out
 template <class Impl>
 void NaiveStaggeredFermion<Impl>::DhopOE(const FermionField &in, FermionField &out, int dag) 
 {
+  DhopCalls+=1;
  conformable(in.Grid(), _cbgrid);    // verifies half grid
  conformable(in.Grid(), out.Grid());  // drops the cb check

@@ -286,6 +288,7 @@ void NaiveStaggeredFermion<Impl>::DhopOE(const FermionField &in, FermionField &o
 template <class Impl>
 void NaiveStaggeredFermion<Impl>::DhopEO(const FermionField &in, FermionField &out, int dag) 
 {
+  DhopCalls+=1;
  conformable(in.Grid(), _cbgrid);    // verifies half grid
  conformable(in.Grid(), out.Grid());  // drops the cb check

@@ -342,33 +345,47 @@ void NaiveStaggeredFermion<Impl>::DhopInternalOverlappedComms(StencilImpl &st, L
  Compressor compressor; 
  int len =  U.Grid()->oSites();

+  DhopTotalTime   -= usecond();
+
+  DhopFaceTime    -= usecond();
  st.Prepare();
  st.HaloGather(in,compressor);
+  DhopFaceTime    += usecond();

+  DhopCommTime -=usecond();
  std::vector<std::vector<CommsRequest_t> > requests;
  st.CommunicateBegin(requests);

+  DhopFaceTime-=usecond();
  st.CommsMergeSHM(compressor);
+  DhopFaceTime+= usecond();

  //////////////////////////////////////////////////////////////////////////////////////////////////////
  // Removed explicit thread comms
  //////////////////////////////////////////////////////////////////////////////////////////////////////
+  DhopComputeTime    -= usecond();
  {
    int interior=1;
    int exterior=0;
    Kernels::DhopNaive(st,lo,U,in,out,dag,interior,exterior);
  }
+  DhopComputeTime    += usecond();

  st.CommunicateComplete(requests);
+  DhopCommTime +=usecond();

  // First to enter, last to leave timing
+  DhopFaceTime    -= usecond();
  st.CommsMerge(compressor);
+  DhopFaceTime    -= usecond();

+  DhopComputeTime2    -= usecond();
  {
    int interior=0;
    int exterior=1;
    Kernels::DhopNaive(st,lo,U,in,out,dag,interior,exterior);
  }
+  DhopComputeTime2    += usecond();
 }

 template <class Impl>
@@ -379,16 +396,78 @@ void NaiveStaggeredFermion<Impl>::DhopInternalSerialComms(StencilImpl &st, Lebes
 {
  assert((dag == DaggerNo) || (dag == DaggerYes));

+  DhopTotalTime   -= usecond();
+
+  DhopCommTime    -= usecond();
  Compressor compressor;
  st.HaloExchange(in, compressor);
+  DhopCommTime    += usecond();

+  DhopComputeTime -= usecond();
  {
    int interior=1;
    int exterior=1;
    Kernels::DhopNaive(st,lo,U,in,out,dag,interior,exterior);
  }
+  DhopComputeTime += usecond();
+  DhopTotalTime   += usecond();
 };

+  ////////////////////////////////////////////////////////////////
+  // Reporting
+  ////////////////////////////////////////////////////////////////
+template<class Impl>
+void NaiveStaggeredFermion<Impl>::Report(void) 
+{
+  Coordinate latt = _grid->GlobalDimensions();
+  RealD volume = 1;  for(int mu=0;mu<Nd;mu++) volume=volume*latt[mu];
+  RealD NP = _grid->_Nprocessors;
+  RealD NN = _grid->NodeCount();
+
+  std::cout << GridLogMessage << "#### Dhop calls report " << std::endl;
+
+  std::cout << GridLogMessage << "NaiveStaggeredFermion Number of DhopEO Calls   : " 
+	    << DhopCalls   << std::endl;
+  std::cout << GridLogMessage << "NaiveStaggeredFermion TotalTime   /Calls       : " 
+	    << DhopTotalTime   / DhopCalls << " us" << std::endl;
+  std::cout << GridLogMessage << "NaiveStaggeredFermion CommTime    /Calls       : " 
+	    << DhopCommTime    / DhopCalls << " us" << std::endl;
+  std::cout << GridLogMessage << "NaiveStaggeredFermion ComputeTime/Calls        : " 
+	    << DhopComputeTime / DhopCalls << " us" << std::endl;
+
+  // Average the compute time
+  _grid->GlobalSum(DhopComputeTime);
+  DhopComputeTime/=NP;
+
+  RealD mflops = 1154*volume*DhopCalls/DhopComputeTime/2; // 2 for red black counting
+  std::cout << GridLogMessage << "Average mflops/s per call                : " << mflops << std::endl;
+  std::cout << GridLogMessage << "Average mflops/s per call per rank       : " << mflops/NP << std::endl;
+  std::cout << GridLogMessage << "Average mflops/s per call per node       : " << mflops/NN << std::endl;
+  
+  RealD Fullmflops = 1154*volume*DhopCalls/(DhopTotalTime)/2; // 2 for red black counting
+  std::cout << GridLogMessage << "Average mflops/s per call (full)         : " << Fullmflops << std::endl;
+  std::cout << GridLogMessage << "Average mflops/s per call per rank (full): " << Fullmflops/NP << std::endl;
+  std::cout << GridLogMessage << "Average mflops/s per call per node (full): " << Fullmflops/NN << std::endl;
+
+  std::cout << GridLogMessage << "NaiveStaggeredFermion Stencil"    <<std::endl;  Stencil.Report();
+  std::cout << GridLogMessage << "NaiveStaggeredFermion StencilEven"<<std::endl;  StencilEven.Report();
+  std::cout << GridLogMessage << "NaiveStaggeredFermion StencilOdd" <<std::endl;  StencilOdd.Report();
+}
+template<class Impl>
+void NaiveStaggeredFermion<Impl>::ZeroCounters(void) 
+{
+  DhopCalls       = 0;
+  DhopTotalTime   = 0;
+  DhopCommTime    = 0;
+  DhopComputeTime = 0;
+  DhopFaceTime    = 0;
+
+  Stencil.ZeroCounters();
+  StencilEven.ZeroCounters();
+  StencilOdd.ZeroCounters();
+}
+
+
 //////////////////////////////////////////////////////// 
 // Conserved current - not yet implemented.
 ////////////////////////////////////////////////////////
@@ -2,13 +2,12 @@

    Grid physics library, www.github.com/paboyle/Grid

-    Source file: ./lib/qcd/action/fermion/WilsonCloverFermionImplementation.h
+    Source file: ./lib/qcd/action/fermion/WilsonCloverFermion.cc

-    Copyright (C) 2017 - 2022
+    Copyright (C) 2017

    Author: paboyle <paboyle@ph.ed.ac.uk>
    Author: Guido Cossu <guido.cossu@ed.ac.uk>
-    Author: Daniel Richtmann <daniel.richtmann@gmail.com>

    This program is free software; you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
@@ -34,48 +33,9 @@

 NAMESPACE_BEGIN(Grid);

-template<class Impl, class CloverHelpers>
-WilsonCloverFermion<Impl, CloverHelpers>::WilsonCloverFermion(GaugeField&                         _Umu,
-                                               GridCartesian&                      Fgrid,
-                                               GridRedBlackCartesian&              Hgrid,
-                                               const RealD                         _mass,
-                                               const RealD                         _csw_r,
-                                               const RealD                         _csw_t,
-                                               const WilsonAnisotropyCoefficients& clover_anisotropy,
-                                               const ImplParams&                   impl_p)
-  : WilsonFermion<Impl>(_Umu, Fgrid, Hgrid, _mass, impl_p, clover_anisotropy)
-  , CloverTerm(&Fgrid)
-  , CloverTermInv(&Fgrid)
-  , CloverTermEven(&Hgrid)
-  , CloverTermOdd(&Hgrid)
-  , CloverTermInvEven(&Hgrid)
-  , CloverTermInvOdd(&Hgrid)
-  , CloverTermDagEven(&Hgrid)
-  , CloverTermDagOdd(&Hgrid)
-  , CloverTermInvDagEven(&Hgrid)
-  , CloverTermInvDagOdd(&Hgrid) {
-  assert(Nd == 4); // require 4 dimensions
-
-  if(clover_anisotropy.isAnisotropic) {
-    csw_r     = _csw_r * 0.5 / clover_anisotropy.xi_0;
-    diag_mass = _mass + 1.0 + (Nd - 1) * (clover_anisotropy.nu / clover_anisotropy.xi_0);
-  } else {
-    csw_r     = _csw_r * 0.5;
-    diag_mass = 4.0 + _mass;
-  }
-  csw_t = _csw_t * 0.5;
-
-  if(csw_r == 0)
-    std::cout << GridLogWarning << "Initializing WilsonCloverFermion with csw_r = 0" << std::endl;
-  if(csw_t == 0)
-    std::cout << GridLogWarning << "Initializing WilsonCloverFermion with csw_t = 0" << std::endl;
-
-  ImportGauge(_Umu);
-}
-
 // *NOT* EO
-template<class Impl, class CloverHelpers>
-void WilsonCloverFermion<Impl, CloverHelpers>::M(const FermionField &in, FermionField &out)
+template <class Impl>
+void WilsonCloverFermion<Impl>::M(const FermionField &in, FermionField &out)
 {
  FermionField temp(out.Grid());

@@ -89,8 +49,8 @@ void WilsonCloverFermion<Impl, CloverHelpers>::M(const FermionField &in, Fermion
  out += temp;
 }

-template<class Impl, class CloverHelpers>
-void WilsonCloverFermion<Impl, CloverHelpers>::Mdag(const FermionField &in, FermionField &out)
+template <class Impl>
+void WilsonCloverFermion<Impl>::Mdag(const FermionField &in, FermionField &out)
 {
  FermionField temp(out.Grid());

@@ -104,16 +64,13 @@ void WilsonCloverFermion<Impl, CloverHelpers>::Mdag(const FermionField &in, Ferm
  out += temp;
 }

-template<class Impl, class CloverHelpers>
-void WilsonCloverFermion<Impl, CloverHelpers>::ImportGauge(const GaugeField &_Umu)
+template <class Impl>
+void WilsonCloverFermion<Impl>::ImportGauge(const GaugeField &_Umu)
 {
-  double t0 = usecond();
  WilsonFermion<Impl>::ImportGauge(_Umu);
-  double t1 = usecond();
  GridBase *grid = _Umu.Grid();
  typename Impl::GaugeLinkField Bx(grid), By(grid), Bz(grid), Ex(grid), Ey(grid), Ez(grid);

-  double t2 = usecond();
  // Compute the field strength terms mu>nu
  WilsonLoops<Impl>::FieldStrength(Bx, _Umu, Zdir, Ydir);
  WilsonLoops<Impl>::FieldStrength(By, _Umu, Zdir, Xdir);
@@ -122,20 +79,52 @@ void WilsonCloverFermion<Impl, CloverHelpers>::ImportGauge(const GaugeField &_Um
  WilsonLoops<Impl>::FieldStrength(Ey, _Umu, Tdir, Ydir);
  WilsonLoops<Impl>::FieldStrength(Ez, _Umu, Tdir, Zdir);

-  double t3 = usecond();
  // Compute the Clover Operator acting on Colour and Spin
  // multiply here by the clover coefficients for the anisotropy
-  CloverTerm  = Helpers::fillCloverYZ(Bx) * csw_r;
-  CloverTerm += Helpers::fillCloverXZ(By) * csw_r;
-  CloverTerm += Helpers::fillCloverXY(Bz) * csw_r;
-  CloverTerm += Helpers::fillCloverXT(Ex) * csw_t;
-  CloverTerm += Helpers::fillCloverYT(Ey) * csw_t;
-  CloverTerm += Helpers::fillCloverZT(Ez) * csw_t;
-   
-  double t4 = usecond();
-  CloverHelpers::Instantiate(CloverTerm, CloverTermInv, csw_t, this->diag_mass);
+  CloverTerm  = fillCloverYZ(Bx) * csw_r;
+  CloverTerm += fillCloverXZ(By) * csw_r;
+  CloverTerm += fillCloverXY(Bz) * csw_r;
+  CloverTerm += fillCloverXT(Ex) * csw_t;
+  CloverTerm += fillCloverYT(Ey) * csw_t;
+  CloverTerm += fillCloverZT(Ez) * csw_t;
+  CloverTerm += diag_mass;
+
+  int lvol = _Umu.Grid()->lSites();
+  int DimRep = Impl::Dimension;
+
+  {
+    autoView(CTv,CloverTerm,CpuRead);
+    autoView(CTIv,CloverTermInv,CpuWrite);
+    thread_for(site, lvol, {
+      Coordinate lcoor;
+      grid->LocalIndexToLocalCoor(site, lcoor);
+      Eigen::MatrixXcd EigenCloverOp = Eigen::MatrixXcd::Zero(Ns * DimRep, Ns * DimRep);
+      Eigen::MatrixXcd EigenInvCloverOp = Eigen::MatrixXcd::Zero(Ns * DimRep, Ns * DimRep);
+      typename SiteCloverType::scalar_object Qx = Zero(), Qxinv = Zero();
+      peekLocalSite(Qx, CTv, lcoor);
+      //if (csw!=0){
+      for (int j = 0; j < Ns; j++)
+	for (int k = 0; k < Ns; k++)
+	  for (int a = 0; a < DimRep; a++)
+	    for (int b = 0; b < DimRep; b++){
+	      auto zz =  Qx()(j, k)(a, b);
+	      EigenCloverOp(a + j * DimRep, b + k * DimRep) = std::complex<double>(zz);
+	    }
+      //   if (site==0) std::cout << "site =" << site << "\n" << EigenCloverOp << std::endl;
+      
+      EigenInvCloverOp = EigenCloverOp.inverse();
+      //std::cout << EigenInvCloverOp << std::endl;
+      for (int j = 0; j < Ns; j++)
+	for (int k = 0; k < Ns; k++)
+	  for (int a = 0; a < DimRep; a++)
+	    for (int b = 0; b < DimRep; b++)
+	      Qxinv()(j, k)(a, b) = EigenInvCloverOp(a + j * DimRep, b + k * DimRep);
+      //    if (site==0) std::cout << "site =" << site << "\n" << EigenInvCloverOp << std::endl;
+      //  }
+      pokeLocalSite(Qxinv, CTIv, lcoor);
+    });
+  }

-  double t5 = usecond();
  // Separate the even and odd parts
  pickCheckerboard(Even, CloverTermEven, CloverTerm);
  pickCheckerboard(Odd, CloverTermOdd, CloverTerm);
@@ -148,47 +137,37 @@ void WilsonCloverFermion<Impl, CloverHelpers>::ImportGauge(const GaugeField &_Um

  pickCheckerboard(Even, CloverTermInvDagEven, adj(CloverTermInv));
  pickCheckerboard(Odd, CloverTermInvDagOdd, adj(CloverTermInv));
-  double t6 = usecond();
-
-  std::cout << GridLogDebug << "WilsonCloverFermion::ImportGauge timings:" << std::endl;
-  std::cout << GridLogDebug << "WilsonFermion::Importgauge = " << (t1 - t0) / 1e6 << std::endl;
-  std::cout << GridLogDebug << "allocations =                " << (t2 - t1) / 1e6 << std::endl;
-  std::cout << GridLogDebug << "field strength =             " << (t3 - t2) / 1e6 << std::endl;
-  std::cout << GridLogDebug << "fill clover =                " << (t4 - t3) / 1e6 << std::endl;
-  std::cout << GridLogDebug << "instantiation =              " << (t5 - t4) / 1e6 << std::endl;
-  std::cout << GridLogDebug << "pick cbs =                   " << (t6 - t5) / 1e6 << std::endl;
-  std::cout << GridLogDebug << "total =                      " << (t6 - t0) / 1e6 << std::endl;
 }

-template<class Impl, class CloverHelpers>
-void WilsonCloverFermion<Impl, CloverHelpers>::Mooee(const FermionField &in, FermionField &out)
+template <class Impl>
+void WilsonCloverFermion<Impl>::Mooee(const FermionField &in, FermionField &out)
 {
  this->MooeeInternal(in, out, DaggerNo, InverseNo);
 }

-template<class Impl, class CloverHelpers>
-void WilsonCloverFermion<Impl, CloverHelpers>::MooeeDag(const FermionField &in, FermionField &out)
+template <class Impl>
+void WilsonCloverFermion<Impl>::MooeeDag(const FermionField &in, FermionField &out)
 {
  this->MooeeInternal(in, out, DaggerYes, InverseNo);
 }

-template<class Impl, class CloverHelpers>
-void WilsonCloverFermion<Impl, CloverHelpers>::MooeeInv(const FermionField &in, FermionField &out)
+template <class Impl>
+void WilsonCloverFermion<Impl>::MooeeInv(const FermionField &in, FermionField &out)
 {
  this->MooeeInternal(in, out, DaggerNo, InverseYes);
 }

-template<class Impl, class CloverHelpers>
-void WilsonCloverFermion<Impl, CloverHelpers>::MooeeInvDag(const FermionField &in, FermionField &out)
+template <class Impl>
+void WilsonCloverFermion<Impl>::MooeeInvDag(const FermionField &in, FermionField &out)
 {
  this->MooeeInternal(in, out, DaggerYes, InverseYes);
 }

-template<class Impl, class CloverHelpers>
-void WilsonCloverFermion<Impl, CloverHelpers>::MooeeInternal(const FermionField &in, FermionField &out, int dag, int inv)
+template <class Impl>
+void WilsonCloverFermion<Impl>::MooeeInternal(const FermionField &in, FermionField &out, int dag, int inv)
 {
  out.Checkerboard() = in.Checkerboard();
-  CloverField *Clover;
+  CloverFieldType *Clover;
  assert(in.Checkerboard() == Odd || in.Checkerboard() == Even);

  if (dag)
@@ -203,12 +182,12 @@ void WilsonCloverFermion<Impl, CloverHelpers>::MooeeInternal(const FermionField
      {
        Clover = (inv) ? &CloverTermInvDagEven : &CloverTermDagEven;
      }
-      Helpers::multCloverField(out, *Clover, in);
+      out = *Clover * in;
    }
    else
    {
      Clover = (inv) ? &CloverTermInv : &CloverTerm;
-      Helpers::multCloverField(out, *Clover, in); // don't bother with adj, hermitian anyway
+      out = adj(*Clover) * in;
    }
  }
  else
@@ -226,109 +205,29 @@ void WilsonCloverFermion<Impl, CloverHelpers>::MooeeInternal(const FermionField
        //  std::cout << "Calling clover term Even" << std::endl;
        Clover = (inv) ? &CloverTermInvEven : &CloverTermEven;
      }
-      Helpers::multCloverField(out, *Clover, in);
+      out = *Clover * in;
      //  std::cout << GridLogMessage << "*Clover.Checkerboard() "  << (*Clover).Checkerboard() << std::endl;
    }
    else
    {
      Clover = (inv) ? &CloverTermInv : &CloverTerm;
-      Helpers::multCloverField(out, *Clover, in);
+      out = *Clover * in;
    }
  }
+
 } // MooeeInternal

-// Derivative parts unpreconditioned pseudofermions
-template<class Impl, class CloverHelpers>
-void WilsonCloverFermion<Impl, CloverHelpers>::MDeriv(GaugeField &force, const FermionField &X, const FermionField &Y, int dag)
-{
-  conformable(X.Grid(), Y.Grid());
-  conformable(X.Grid(), force.Grid());
-  GaugeLinkField force_mu(force.Grid()), lambda(force.Grid());
-  GaugeField clover_force(force.Grid());
-  PropagatorField Lambda(force.Grid());
-
-  // Guido: Here we are hitting some performance issues:
-  // need to extract the components of the DoubledGaugeField
-  // for each call
-  // Possible solution
-  // Create a vector object to store them? (cons: wasting space)
-  std::vector<GaugeLinkField> U(Nd, this->Umu.Grid());
-
-  Impl::extractLinkField(U, this->Umu);
-
-  force = Zero();
-  // Derivative of the Wilson hopping term
-  this->DhopDeriv(force, X, Y, dag);
-
-  ///////////////////////////////////////////////////////////
-  // Clover term derivative
-  ///////////////////////////////////////////////////////////
-  Impl::outerProductImpl(Lambda, X, Y);
-  //std::cout << "Lambda:" << Lambda << std::endl;
-
-  Gamma::Algebra sigma[] = {
-      Gamma::Algebra::SigmaXY,
-      Gamma::Algebra::SigmaXZ,
-      Gamma::Algebra::SigmaXT,
-      Gamma::Algebra::MinusSigmaXY,
-      Gamma::Algebra::SigmaYZ,
-      Gamma::Algebra::SigmaYT,
-      Gamma::Algebra::MinusSigmaXZ,
-      Gamma::Algebra::MinusSigmaYZ,
-      Gamma::Algebra::SigmaZT,
-      Gamma::Algebra::MinusSigmaXT,
-      Gamma::Algebra::MinusSigmaYT,
-      Gamma::Algebra::MinusSigmaZT};
-
-  /*
-    sigma_{\mu \nu}=
-    | 0         sigma[0]  sigma[1]  sigma[2] |
-    | sigma[3]    0       sigma[4]  sigma[5] |
-    | sigma[6]  sigma[7]     0      sigma[8] |
-    | sigma[9]  sigma[10] sigma[11]   0      |
-  */
-
-  int count = 0;
-  clover_force = Zero();
-  for (int mu = 0; mu < 4; mu++)
-  {
-    force_mu = Zero();
-    for (int nu = 0; nu < 4; nu++)
-    {
-      if (mu == nu)
-      continue;
-
-      RealD factor;
-      if (nu == 4 || mu == 4)
-      {
-        factor = 2.0 * csw_t;
-      }
-      else
-      {
-        factor = 2.0 * csw_r;
-      }
-      PropagatorField Slambda = Gamma(sigma[count]) * Lambda; // sigma checked
-      Impl::TraceSpinImpl(lambda, Slambda);                   // traceSpin ok
-      force_mu -= factor*CloverHelpers::Cmunu(U, lambda, mu, nu);                   // checked
-      count++;
-    }
-
-    pokeLorentz(clover_force, U[mu] * force_mu, mu);
-  }
-  //clover_force *= csw;
-  force += clover_force;
-}

 // Derivative parts
-template<class Impl, class CloverHelpers>
-void WilsonCloverFermion<Impl, CloverHelpers>::MooDeriv(GaugeField &mat, const FermionField &X, const FermionField &Y, int dag)
+template <class Impl>
+void WilsonCloverFermion<Impl>::MooDeriv(GaugeField &mat, const FermionField &X, const FermionField &Y, int dag)
 {
  assert(0);
 }

 // Derivative parts
-template<class Impl, class CloverHelpers>
-void WilsonCloverFermion<Impl, CloverHelpers>::MeeDeriv(GaugeField &mat, const FermionField &U, const FermionField &V, int dag)
+template <class Impl>
+void WilsonCloverFermion<Impl>::MeeDeriv(GaugeField &mat, const FermionField &U, const FermionField &V, int dag)
 {
  assert(0); // not implemented yet
 }
@@ -60,8 +60,7 @@ WilsonFermion5D<Impl>::WilsonFermion5D(GaugeField &_Umu,
  UmuOdd (_FourDimRedBlackGrid),
  Lebesgue(_FourDimGrid),
  LebesgueEvenOdd(_FourDimRedBlackGrid),
-  _tmp(&FiveDimRedBlackGrid),
-  Dirichlet(0)
+  _tmp(&FiveDimRedBlackGrid)
 {
  // some assertions
  assert(FiveDimGrid._ndimension==5);
@@ -92,19 +91,6 @@ WilsonFermion5D<Impl>::WilsonFermion5D(GaugeField &_Umu,
    assert(FourDimRedBlackGrid._simd_layout[d]  ==FourDimGrid._simd_layout[d]);
  }

-  if ( p.dirichlet.size() == Nd+1) {
-    Coordinate block = p.dirichlet;
-    if ( block[0] || block[1] || block[2] || block[3] || block[4] ){
-      Dirichlet = 1;
-      std::cout << GridLogMessage << " WilsonFermion: non-trivial Dirichlet condition "<< block << std::endl;
-      std::cout << GridLogMessage << " WilsonFermion: partial Dirichlet "<< p.partialDirichlet << std::endl;
-      Block = block;
-    }
-  } else {
-    Coordinate block(Nd+1,0);
-    Block = block;
-  }
-
  if (Impl::LsVectorised) { 

    int nsimd = Simd::Nsimd();
@@ -139,38 +125,99 @@ WilsonFermion5D<Impl>::WilsonFermion5D(GaugeField &_Umu,
  StencilEven.BuildSurfaceList(LLs,vol4);
   StencilOdd.BuildSurfaceList(LLs,vol4);

+   //  std::cout << GridLogMessage << " SurfaceLists "<< Stencil.surface_list.size()
+   //                       <<" " << StencilEven.surface_list.size()<<std::endl;
+
 }
+     
+template<class Impl>
+void WilsonFermion5D<Impl>::Report(void)
+{
+  RealD NP     = _FourDimGrid->_Nprocessors;
+  RealD NN     = _FourDimGrid->NodeCount();
+  RealD volume = Ls;  
+  Coordinate latt = _FourDimGrid->GlobalDimensions();
+  for(int mu=0;mu<Nd;mu++) volume=volume*latt[mu];
+
+  if ( DhopCalls > 0 ) {
+    std::cout << GridLogMessage << "#### Dhop calls report " << std::endl;
+    std::cout << GridLogMessage << "WilsonFermion5D Number of DhopEO Calls   : " << DhopCalls   << std::endl;
+    std::cout << GridLogMessage << "WilsonFermion5D TotalTime   /Calls        : " << DhopTotalTime   / DhopCalls << " us" << std::endl;
+    std::cout << GridLogMessage << "WilsonFermion5D CommTime    /Calls        : " << DhopCommTime    / DhopCalls << " us" << std::endl;
+    std::cout << GridLogMessage << "WilsonFermion5D FaceTime    /Calls        : " << DhopFaceTime    / DhopCalls << " us" << std::endl;
+    std::cout << GridLogMessage << "WilsonFermion5D ComputeTime1/Calls        : " << DhopComputeTime / DhopCalls << " us" << std::endl;
+    std::cout << GridLogMessage << "WilsonFermion5D ComputeTime2/Calls        : " << DhopComputeTime2/ DhopCalls << " us" << std::endl;
+
+    // Average the compute time
+    _FourDimGrid->GlobalSum(DhopComputeTime);
+    DhopComputeTime/=NP;
+    RealD mflops = 1344*volume*DhopCalls/DhopComputeTime/2; // 2 for red black counting
+    std::cout << GridLogMessage << "Average mflops/s per call                : " << mflops << std::endl;
+    std::cout << GridLogMessage << "Average mflops/s per call per rank       : " << mflops/NP << std::endl;
+    std::cout << GridLogMessage << "Average mflops/s per call per node       : " << mflops/NN << std::endl;
+
+    RealD Fullmflops = 1344*volume*DhopCalls/(DhopTotalTime)/2; // 2 for red black counting
+    std::cout << GridLogMessage << "Average mflops/s per call (full)         : " << Fullmflops << std::endl;
+    std::cout << GridLogMessage << "Average mflops/s per call per rank (full): " << Fullmflops/NP << std::endl;
+    std::cout << GridLogMessage << "Average mflops/s per call per node (full): " << Fullmflops/NN << std::endl;
+
+   }
+
+  if ( DerivCalls > 0 ) {
+    std::cout << GridLogMessage << "#### Deriv calls report "<< std::endl;
+    std::cout << GridLogMessage << "WilsonFermion5D Number of Deriv Calls    : " <<DerivCalls <<std::endl;
+    std::cout << GridLogMessage << "WilsonFermion5D CommTime/Calls           : " <<DerivCommTime/DerivCalls<<" us" <<std::endl;
+    std::cout << GridLogMessage << "WilsonFermion5D ComputeTime/Calls        : " <<DerivComputeTime/DerivCalls<<" us" <<std::endl;
+    std::cout << GridLogMessage << "WilsonFermion5D Dhop ComputeTime/Calls   : " <<DerivDhopComputeTime/DerivCalls<<" us" <<std::endl;
+    
+    RealD mflops = 144*volume*DerivCalls/DerivDhopComputeTime;
+    std::cout << GridLogMessage << "Average mflops/s per call                : " << mflops << std::endl;
+    std::cout << GridLogMessage << "Average mflops/s per call per node       : " << mflops/NP << std::endl;
+
+    RealD Fullmflops = 144*volume*DerivCalls/(DerivDhopComputeTime+DerivCommTime)/2; // 2 for red black counting
+    std::cout << GridLogMessage << "Average mflops/s per call (full)         : " << Fullmflops << std::endl;
+    std::cout << GridLogMessage << "Average mflops/s per call per node (full): " << Fullmflops/NP << std::endl;  }
+
+  if (DerivCalls > 0 || DhopCalls > 0){
+    std::cout << GridLogMessage << "WilsonFermion5D Stencil"    <<std::endl;  Stencil.Report();
+    std::cout << GridLogMessage << "WilsonFermion5D StencilEven"<<std::endl;  StencilEven.Report();
+    std::cout << GridLogMessage << "WilsonFermion5D StencilOdd" <<std::endl;  StencilOdd.Report();
+  }
+  if ( DhopCalls > 0){
+    std::cout << GridLogMessage << "WilsonFermion5D Stencil     Reporti()"    <<std::endl;  Stencil.Reporti(DhopCalls);
+    std::cout << GridLogMessage << "WilsonFermion5D StencilEven Reporti()"<<std::endl;  StencilEven.Reporti(DhopCalls);
+    std::cout << GridLogMessage << "WilsonFermion5D StencilOdd  Reporti()" <<std::endl;  StencilOdd.Reporti(DhopCalls);
+  }
+}
+
+template<class Impl>
+void WilsonFermion5D<Impl>::ZeroCounters(void) {
+  DhopCalls       = 0;
+  DhopCommTime    = 0;
+  DhopComputeTime = 0;
+  DhopComputeTime2= 0;
+  DhopFaceTime    = 0;
+  DhopTotalTime   = 0;
+
+  DerivCalls       = 0;
+  DerivCommTime    = 0;
+  DerivComputeTime = 0;
+  DerivDhopComputeTime = 0;
+
+  Stencil.ZeroCounters();
+  StencilEven.ZeroCounters();
+  StencilOdd.ZeroCounters();
+  Stencil.ZeroCountersi();
+  StencilEven.ZeroCountersi();
+  StencilOdd.ZeroCountersi();
+}
+

 template<class Impl>
 void WilsonFermion5D<Impl>::ImportGauge(const GaugeField &_Umu)
 {
  GaugeField HUmu(_Umu.Grid());
  HUmu = _Umu*(-0.5);
-  if ( Dirichlet ) {
-
-    if ( this->Params.partialDirichlet ) {
-      std::cout << GridLogMessage << " partialDirichlet BCs " <<Block<<std::endl;
-    } else {
-      std::cout << GridLogMessage << " FULL Dirichlet BCs " <<Block<<std::endl;
-    }
-    
-    std:: cout << GridLogMessage << "Checking block size multiple of rank boundaries for Dirichlet"<<std::endl;
-    for(int d=0;d<Nd;d++) {
-      int GaugeBlock = Block[d+1];
-      int ldim=GaugeGrid()->LocalDimensions()[d];
-      if (GaugeBlock) assert( (GaugeBlock%ldim)==0);
-    }
-
-    if (!this->Params.partialDirichlet) {
-      std::cout << GridLogMessage << " Dirichlet filtering gauge field BCs block " <<Block<<std::endl;
-      Coordinate GaugeBlock(Nd);
-      for(int d=0;d<Nd;d++) GaugeBlock[d] = Block[d+1];
-      DirichletFilter<GaugeField> Filter(GaugeBlock);
-      Filter.applyFilter(HUmu);
-    } else {
-      std::cout << GridLogMessage << " Dirichlet "<< Dirichlet << " NOT filtered gauge field" <<std::endl;
-    }
-  }
  Impl::DoubleStore(GaugeGrid(),Umu,HUmu);
  pickCheckerboard(Even,UmuEven,Umu);
  pickCheckerboard(Odd ,UmuOdd,Umu);
@@ -212,6 +259,7 @@ void WilsonFermion5D<Impl>::DerivInternal(StencilImpl & st,
 					  const FermionField &B,
 					  int dag)
 {
+  DerivCalls++;
  assert((dag==DaggerNo) ||(dag==DaggerYes));

  conformable(st.Grid(),A.Grid());
@@ -222,12 +270,15 @@ void WilsonFermion5D<Impl>::DerivInternal(StencilImpl & st,
  FermionField Btilde(B.Grid());
  FermionField Atilde(B.Grid());

+  DerivCommTime-=usecond();
  st.HaloExchange(B,compressor);
+  DerivCommTime+=usecond();

  Atilde=A;
  int LLs = B.Grid()->_rdimensions[0];


+  DerivComputeTime-=usecond();
  for (int mu = 0; mu < Nd; mu++) {
    ////////////////////////////////////////////////////////////////////////
    // Flip gamma if dag
@@ -239,6 +290,8 @@ void WilsonFermion5D<Impl>::DerivInternal(StencilImpl & st,
    // Call the single hop
    ////////////////////////

+    DerivDhopComputeTime -= usecond();
+
    int Usites = U.Grid()->oSites();

    Kernels::DhopDirKernel(st, U, st.CommBuf(), Ls, Usites, B, Btilde, mu,gamma);
@@ -246,8 +299,10 @@ void WilsonFermion5D<Impl>::DerivInternal(StencilImpl & st,
    ////////////////////////////
    // spin trace outer product
    ////////////////////////////
+    DerivDhopComputeTime += usecond();
    Impl::InsertForce5D(mat, Btilde, Atilde, mu);
  }
+  DerivComputeTime += usecond();
 }

 template<class Impl>
@@ -305,10 +360,12 @@ void WilsonFermion5D<Impl>::DhopInternal(StencilImpl & st, LebesgueOrder &lo,
                                         DoubledGaugeField & U,
                                         const FermionField &in, FermionField &out,int dag)
 {
+  DhopTotalTime-=usecond();
  if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsAndCompute )
    DhopInternalOverlappedComms(st,lo,U,in,out,dag);
  else 
    DhopInternalSerialComms(st,lo,U,in,out,dag);
+  DhopTotalTime+=usecond();
 }


@@ -317,7 +374,6 @@ void WilsonFermion5D<Impl>::DhopInternalOverlappedComms(StencilImpl & st, Lebesg
 							DoubledGaugeField & U,
 							const FermionField &in, FermionField &out,int dag)
 {
-  GRID_TRACE("DhopInternalOverlappedComms");
  Compressor compressor(dag);

  int LLs = in.Grid()->_rdimensions[0];
@@ -326,58 +382,53 @@ void WilsonFermion5D<Impl>::DhopInternalOverlappedComms(StencilImpl & st, Lebesg
  /////////////////////////////
  // Start comms  // Gather intranode and extra node differentiated??
  /////////////////////////////
-  {
-    GRID_TRACE("Gather");
-    st.HaloExchangeOptGather(in,compressor);
-    accelerator_barrier();
-  }
-  
+  DhopFaceTime-=usecond();
+  st.HaloExchangeOptGather(in,compressor);
+  DhopFaceTime+=usecond();
+
+  DhopCommTime -=usecond();
  std::vector<std::vector<CommsRequest_t> > requests;
-  auto id=traceStart("Communicate overlapped");
  st.CommunicateBegin(requests);

  /////////////////////////////
  // Overlap with comms
  /////////////////////////////
-  {
-    GRID_TRACE("MergeSHM");
-    st.CommsMergeSHM(compressor);// Could do this inside parallel region overlapped with comms
-  }
+  DhopFaceTime-=usecond();
+  st.CommsMergeSHM(compressor);// Could do this inside parallel region overlapped with comms
+  DhopFaceTime+=usecond();
      
  /////////////////////////////
  // do the compute interior
  /////////////////////////////
  int Opt = WilsonKernelsStatic::Opt; // Why pass this. Kernels should know
+  DhopComputeTime-=usecond();
  if (dag == DaggerYes) {
-    GRID_TRACE("DhopDagInterior");
    Kernels::DhopDagKernel(Opt,st,U,st.CommBuf(),LLs,U.oSites(),in,out,1,0);
  } else {
-    GRID_TRACE("DhopInterior");
    Kernels::DhopKernel   (Opt,st,U,st.CommBuf(),LLs,U.oSites(),in,out,1,0);
  }
+  DhopComputeTime+=usecond();

  /////////////////////////////
  // Complete comms
  /////////////////////////////
  st.CommunicateComplete(requests);
-  traceStop(id);
+  DhopCommTime   +=usecond();

  /////////////////////////////
  // do the compute exterior
  /////////////////////////////
-  {
-    GRID_TRACE("Merge");
-    st.CommsMerge(compressor);
-  }
-  
+  DhopFaceTime-=usecond();
+  st.CommsMerge(compressor);
+  DhopFaceTime+=usecond();

+  DhopComputeTime2-=usecond();
  if (dag == DaggerYes) {
-    GRID_TRACE("DhopDagExterior");
    Kernels::DhopDagKernel(Opt,st,U,st.CommBuf(),LLs,U.oSites(),in,out,0,1);
  } else {
-    GRID_TRACE("DhopExterior");
    Kernels::DhopKernel   (Opt,st,U,st.CommBuf(),LLs,U.oSites(),in,out,0,1);
  }
+  DhopComputeTime2+=usecond();
 }


@@ -387,30 +438,29 @@ void WilsonFermion5D<Impl>::DhopInternalSerialComms(StencilImpl & st, LebesgueOr
 						    const FermionField &in, 
 						    FermionField &out,int dag)
 {
-  GRID_TRACE("DhopInternalSerialComms");
  Compressor compressor(dag);

  int LLs = in.Grid()->_rdimensions[0];
-
-  {
-    GRID_TRACE("HaloExchange");
-    st.HaloExchangeOpt(in,compressor);
-  }
  
+  DhopCommTime-=usecond();
+  st.HaloExchangeOpt(in,compressor);
+  DhopCommTime+=usecond();
+  
+  DhopComputeTime-=usecond();
  int Opt = WilsonKernelsStatic::Opt;
  if (dag == DaggerYes) {
-    GRID_TRACE("DhopDag");
    Kernels::DhopDagKernel(Opt,st,U,st.CommBuf(),LLs,U.oSites(),in,out);
  } else {
-    GRID_TRACE("Dhop");
    Kernels::DhopKernel(Opt,st,U,st.CommBuf(),LLs,U.oSites(),in,out);
  }
+  DhopComputeTime+=usecond();
 }


 template<class Impl>
 void WilsonFermion5D<Impl>::DhopOE(const FermionField &in, FermionField &out,int dag)
 {
+  DhopCalls++;
  conformable(in.Grid(),FermionRedBlackGrid());    // verifies half grid
  conformable(in.Grid(),out.Grid()); // drops the cb check

@@ -422,6 +472,7 @@ void WilsonFermion5D<Impl>::DhopOE(const FermionField &in, FermionField &out,int
 template<class Impl>
 void WilsonFermion5D<Impl>::DhopEO(const FermionField &in, FermionField &out,int dag)
 {
+  DhopCalls++;
  conformable(in.Grid(),FermionRedBlackGrid());    // verifies half grid
  conformable(in.Grid(),out.Grid()); // drops the cb check

@@ -433,6 +484,7 @@ void WilsonFermion5D<Impl>::DhopEO(const FermionField &in, FermionField &out,int
 template<class Impl>
 void WilsonFermion5D<Impl>::Dhop(const FermionField &in, FermionField &out,int dag)
 {
+  DhopCalls+=2;
  conformable(in.Grid(),FermionGrid()); // verifies full grid
  conformable(in.Grid(),out.Grid());

@@ -487,17 +539,12 @@ void WilsonFermion5D<Impl>::MomentumSpacePropagatorHt_5d(FermionField &out,const
  LatComplex    sk(_grid);  sk = Zero();
  LatComplex    sk2(_grid); sk2= Zero();
  LatComplex    W(_grid); W= Zero();
+  LatComplex    a(_grid); a= Zero();
  LatComplex    one  (_grid); one = ScalComplex(1.0,0.0);
  LatComplex 	cosha(_grid);
  LatComplex 	kmu(_grid);
  LatComplex 	Wea(_grid);
  LatComplex 	Wema(_grid);
-  LatComplex 	ea(_grid);
-  LatComplex 	ema(_grid);
-  LatComplex 	eaLs(_grid);
-  LatComplex 	emaLs(_grid);
-  LatComplex 	ea2Ls(_grid);
-  LatComplex 	ema2Ls(_grid);
  LatComplex 	sinha(_grid);
  LatComplex 	sinhaLs(_grid);
  LatComplex 	coshaLs(_grid);
@@ -532,29 +579,39 @@ void WilsonFermion5D<Impl>::MomentumSpacePropagatorHt_5d(FermionField &out,const
  ////////////////////////////////////////////
  cosha = (one + W*W + sk) / (abs(W)*2.0);

-  ea = (cosha + sqrt(cosha*cosha-one));
-  ema= (cosha - sqrt(cosha*cosha-one));
-  eaLs = pow(ea,Ls);
-  emaLs= pow(ema,Ls);
-  ea2Ls = pow(ea,2.0*Ls);
-  ema2Ls= pow(ema,2.0*Ls);
-  Wea= abs(W) * ea;
-  Wema= abs(W) * ema;
-  //  a=log(ea);
-  
-  sinha = 0.5*(ea - ema);
-  sinhaLs = 0.5*(eaLs-emaLs);
-  coshaLs = 0.5*(eaLs+emaLs);
+  // FIXME Need a Lattice acosh
+
+  {
+    autoView(cosha_v,cosha,CpuRead);
+    autoView(a_v,a,CpuWrite);
+    for(int idx=0;idx<_grid->lSites();idx++){
+      Coordinate lcoor(Nd);
+      Tcomplex cc;
+      //    RealD sgn;
+      _grid->LocalIndexToLocalCoor(idx,lcoor);
+      peekLocalSite(cc,cosha_v,lcoor);
+      assert((double)real(cc)>=1.0);
+      assert(fabs((double)imag(cc))<=1.0e-15);
+      cc = ScalComplex(::acosh(real(cc)),0.0);
+      pokeLocalSite(cc,a_v,lcoor);
+    }
+  }
+
+  Wea = ( exp( a) * abs(W)  );
+  Wema= ( exp(-a) * abs(W)  );
+  sinha = 0.5*(exp( a) - exp(-a));
+  sinhaLs = 0.5*(exp( a*Ls) - exp(-a*Ls));
+  coshaLs = 0.5*(exp( a*Ls) + exp(-a*Ls));

  A = one / (abs(W) * sinha * 2.0) * one / (sinhaLs * 2.0);
-  F = eaLs * (one - Wea + (Wema - one) * mass*mass);
-  F = F + emaLs * (Wema - one + (one - Wea) * mass*mass);
+  F = exp( a*Ls) * (one - Wea + (Wema - one) * mass*mass);
+  F = F + exp(-a*Ls) * (Wema - one + (one - Wea) * mass*mass);
  F = F - abs(W) * sinha * 4.0 * mass;

-  Bpp =  (A/F) * (ema2Ls - one) * (one - Wema) * (one - mass*mass * one);
-  Bmm =  (A/F) * (one - ea2Ls)  * (one - Wea) * (one - mass*mass * one);
-  App =  (A/F) * (ema2Ls - one) * ema * (ema - abs(W)) * (one - mass*mass * one);
-  Amm =  (A/F) * (one - ea2Ls)  * ea  * (ea  - abs(W)) * (one - mass*mass * one);
+  Bpp =  (A/F) * (exp(-a*Ls*2.0) - one) * (one - Wema) * (one - mass*mass * one);
+  Bmm =  (A/F) * (one - exp(a*Ls*2.0)) * (one - Wea) * (one - mass*mass * one);
+  App =  (A/F) * (exp(-a*Ls*2.0) - one) * exp(-a) * (exp(-a) - abs(W)) * (one - mass*mass * one);
+  Amm =  (A/F) * (one - exp(a*Ls*2.0)) * exp(a) * (exp(a) - abs(W)) * (one - mass*mass * one);
  ABpm = (A/F) * abs(W) * sinha * 2.0  * (one + mass * coshaLs * 2.0 + mass*mass * one);

  //P+ source, P- source
@@ -577,29 +634,29 @@ void WilsonFermion5D<Impl>::MomentumSpacePropagatorHt_5d(FermionField &out,const
      buf1_4d = Zero();
      ExtractSlice(buf1_4d, PRsource, (tt-1), 0);
      //G(s,t)
-      bufR_4d = bufR_4d + A * eaLs * pow(ema,f) * signW * buf1_4d + A * emaLs * pow(ea,f) * signW * buf1_4d;
+      bufR_4d = bufR_4d + A * exp(a*Ls) * exp(-a*f) * signW * buf1_4d + A * exp(-a*Ls) * exp(a*f) * signW * buf1_4d;
      //A++*exp(a(s+t))
-      bufR_4d = bufR_4d + App * pow(ea,ss) * pow(ea,tt) * signW * buf1_4d ;
+      bufR_4d = bufR_4d + App * exp(a*ss) * exp(a*tt) * signW * buf1_4d ;
      //A+-*exp(a(s-t))
-      bufR_4d = bufR_4d + ABpm * pow(ea,ss) * pow(ema,tt) * signW * buf1_4d ;
+      bufR_4d = bufR_4d + ABpm * exp(a*ss) * exp(-a*tt) * signW * buf1_4d ;
      //A-+*exp(a(-s+t))
-      bufR_4d = bufR_4d + ABpm * pow(ema,ss) * pow(ea,tt) * signW * buf1_4d ;
+      bufR_4d = bufR_4d + ABpm * exp(-a*ss) * exp(a*tt) * signW * buf1_4d ;
      //A--*exp(a(-s-t))
-      bufR_4d = bufR_4d + Amm * pow(ema,ss) * pow(ema,tt) * signW * buf1_4d ;
+      bufR_4d = bufR_4d + Amm * exp(-a*ss) * exp(-a*tt) * signW * buf1_4d ;

      //GL
      buf2_4d = Zero();
      ExtractSlice(buf2_4d, PLsource, (tt-1), 0);
      //G(s,t)
-      bufL_4d = bufL_4d + A * eaLs * pow(ema,f) * signW * buf2_4d + A * emaLs * pow(ea,f) * signW * buf2_4d;
+      bufL_4d = bufL_4d + A * exp(a*Ls) * exp(-a*f) * signW * buf2_4d + A * exp(-a*Ls) * exp(a*f) * signW * buf2_4d;
      //B++*exp(a(s+t))
-      bufL_4d = bufL_4d + Bpp * pow(ea,ss) * pow(ea,tt) * signW * buf2_4d ;
+      bufL_4d = bufL_4d + Bpp * exp(a*ss) * exp(a*tt) * signW * buf2_4d ;
      //B+-*exp(a(s-t))
-      bufL_4d = bufL_4d + ABpm * pow(ea,ss) * pow(ema,tt) * signW * buf2_4d ;
+      bufL_4d = bufL_4d + ABpm * exp(a*ss) * exp(-a*tt) * signW * buf2_4d ;
      //B-+*exp(a(-s+t))
-      bufL_4d = bufL_4d + ABpm * pow(ema,ss) * pow(ea,tt) * signW * buf2_4d ;
+      bufL_4d = bufL_4d + ABpm * exp(-a*ss) * exp(a*tt) * signW * buf2_4d ;
      //B--*exp(a(-s-t))
-      bufL_4d = bufL_4d + Bmm * pow(ema,ss) * pow(ema,tt) * signW * buf2_4d ;
+      bufL_4d = bufL_4d + Bmm * exp(-a*ss) * exp(-a*tt) * signW * buf2_4d ;
    }
    InsertSlice(bufR_4d, GR, (ss-1), 0);
    InsertSlice(bufL_4d, GL, (ss-1), 0);
@@ -718,12 +775,28 @@ void WilsonFermion5D<Impl>::MomentumSpacePropagatorHt(FermionField &out,const Fe
  W = one - M5 + sk2;

  ////////////////////////////////////////////
-  // Cosh alpha -> exp(+/- alpha)
+  // Cosh alpha -> alpha
  ////////////////////////////////////////////
  cosha =  (one + W*W + sk) / (abs(W)*2.0);

-  Wea = abs(W)*(cosha + sqrt(cosha*cosha-one));
-  Wema= abs(W)*(cosha - sqrt(cosha*cosha-one));
+  // FIXME Need a Lattice acosh
+  {
+  autoView(cosha_v,cosha,CpuRead);
+  autoView(a_v,a,CpuWrite);
+  for(int idx=0;idx<_grid->lSites();idx++){
+    Coordinate lcoor(Nd);
+    Tcomplex cc;
+    //    RealD sgn;
+    _grid->LocalIndexToLocalCoor(idx,lcoor);
+    peekLocalSite(cc,cosha_v,lcoor);
+    assert((double)real(cc)>=1.0);
+    assert(fabs((double)imag(cc))<=1.0e-15);
+    cc = ScalComplex(::acosh(real(cc)),0.0);
+    pokeLocalSite(cc,a_v,lcoor);
+  }}
+  
+  Wea = ( exp( a) * abs(W)  );
+  Wema= ( exp(-a) * abs(W)  );
  
  num   = num + ( one - Wema ) * mass * in;
  denom= ( Wea - one ) + mass*mass * (one - Wema); 
@@ -4,13 +4,12 @@ Grid physics library, www.github.com/paboyle/Grid

 Source file: ./lib/qcd/action/fermion/WilsonFermion.cc

-Copyright (C) 2022
+Copyright (C) 2015

 Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
 Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
 Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Fabian Joswig <fabian.joswig@ed.ac.uk>

 This program is free software; you can redistribute it and/or modify
 it under the terms of the GNU General Public License as published by
@@ -76,6 +75,91 @@ WilsonFermion<Impl>::WilsonFermion(GaugeField &_Umu, GridCartesian &Fgrid,
  StencilOdd.BuildSurfaceList(1,vol4);
 }

+template<class Impl>
+void WilsonFermion<Impl>::Report(void)
+{
+  RealD NP = _grid->_Nprocessors;
+  RealD NN = _grid->NodeCount();
+  RealD volume = 1;
+  Coordinate latt = _grid->GlobalDimensions();
+  for(int mu=0;mu<Nd;mu++) volume=volume*latt[mu];
+
+  if ( DhopCalls > 0 ) {
+    std::cout << GridLogMessage << "#### Dhop calls report " << std::endl;
+    std::cout << GridLogMessage << "WilsonFermion Number of DhopEO Calls   : " << DhopCalls   << std::endl;
+    std::cout << GridLogMessage << "WilsonFermion TotalTime   /Calls        : " << DhopTotalTime   / DhopCalls << " us" << std::endl;
+    std::cout << GridLogMessage << "WilsonFermion CommTime    /Calls        : " << DhopCommTime    / DhopCalls << " us" << std::endl;
+    std::cout << GridLogMessage << "WilsonFermion FaceTime    /Calls        : " << DhopFaceTime    / DhopCalls << " us" << std::endl;
+    std::cout << GridLogMessage << "WilsonFermion ComputeTime1/Calls        : " << DhopComputeTime / DhopCalls << " us" << std::endl;
+    std::cout << GridLogMessage << "WilsonFermion ComputeTime2/Calls        : " << DhopComputeTime2/ DhopCalls << " us" << std::endl;
+
+    // Average the compute time
+    _grid->GlobalSum(DhopComputeTime);
+    DhopComputeTime/=NP;
+    RealD mflops = 1320*volume*DhopCalls/DhopComputeTime/2; // 2 for red black counting
+    std::cout << GridLogMessage << "Average mflops/s per call                : " << mflops << std::endl;
+    std::cout << GridLogMessage << "Average mflops/s per call per rank       : " << mflops/NP << std::endl;
+    std::cout << GridLogMessage << "Average mflops/s per call per node       : " << mflops/NN << std::endl;
+
+    RealD Fullmflops = 1320*volume*DhopCalls/(DhopTotalTime)/2; // 2 for red black counting
+    std::cout << GridLogMessage << "Average mflops/s per call (full)         : " << Fullmflops << std::endl;
+    std::cout << GridLogMessage << "Average mflops/s per call per rank (full): " << Fullmflops/NP << std::endl;
+    std::cout << GridLogMessage << "Average mflops/s per call per node (full): " << Fullmflops/NN << std::endl;
+
+   }
+
+  if ( DerivCalls > 0 ) {
+    std::cout << GridLogMessage << "#### Deriv calls report "<< std::endl;
+    std::cout << GridLogMessage << "WilsonFermion Number of Deriv Calls    : " <<DerivCalls <<std::endl;
+    std::cout << GridLogMessage << "WilsonFermion CommTime/Calls           : " <<DerivCommTime/DerivCalls<<" us" <<std::endl;
+    std::cout << GridLogMessage << "WilsonFermion ComputeTime/Calls        : " <<DerivComputeTime/DerivCalls<<" us" <<std::endl;
+    std::cout << GridLogMessage << "WilsonFermion Dhop ComputeTime/Calls   : " <<DerivDhopComputeTime/DerivCalls<<" us" <<std::endl;
+
+    // how to count flops here?
+    RealD mflops = 144*volume*DerivCalls/DerivDhopComputeTime;
+    std::cout << GridLogMessage << "Average mflops/s per call               ? : " << mflops << std::endl;
+    std::cout << GridLogMessage << "Average mflops/s per call per node      ? : " << mflops/NP << std::endl;
+
+    // how to count flops here?
+    RealD Fullmflops = 144*volume*DerivCalls/(DerivDhopComputeTime+DerivCommTime)/2; // 2 for red black counting
+    std::cout << GridLogMessage << "Average mflops/s per call (full)        ? : " << Fullmflops << std::endl;
+    std::cout << GridLogMessage << "Average mflops/s per call per node (full) ? : " << Fullmflops/NP << std::endl;  }
+
+  if (DerivCalls > 0 || DhopCalls > 0){
+    std::cout << GridLogMessage << "WilsonFermion Stencil"    <<std::endl;  Stencil.Report();
+    std::cout << GridLogMessage << "WilsonFermion StencilEven"<<std::endl;  StencilEven.Report();
+    std::cout << GridLogMessage << "WilsonFermion StencilOdd" <<std::endl;  StencilOdd.Report();
+  }
+  if ( DhopCalls > 0){
+    std::cout << GridLogMessage << "WilsonFermion Stencil     Reporti()"    <<std::endl;  Stencil.Reporti(DhopCalls);
+    std::cout << GridLogMessage << "WilsonFermion StencilEven Reporti()"<<std::endl;  StencilEven.Reporti(DhopCalls);
+    std::cout << GridLogMessage << "WilsonFermion StencilOdd  Reporti()" <<std::endl;  StencilOdd.Reporti(DhopCalls);
+  }
+}
+
+template<class Impl>
+void WilsonFermion<Impl>::ZeroCounters(void) {
+  DhopCalls       = 0; // ok
+  DhopCommTime    = 0;
+  DhopComputeTime = 0;
+  DhopComputeTime2= 0;
+  DhopFaceTime    = 0;
+  DhopTotalTime   = 0;
+
+  DerivCalls       = 0; // ok
+  DerivCommTime    = 0;
+  DerivComputeTime = 0;
+  DerivDhopComputeTime = 0;
+
+  Stencil.ZeroCounters();
+  StencilEven.ZeroCounters();
+  StencilOdd.ZeroCounters();
+  Stencil.ZeroCountersi();
+  StencilEven.ZeroCountersi();
+  StencilOdd.ZeroCountersi();
+}
+
+
 template <class Impl>
 void WilsonFermion<Impl>::ImportGauge(const GaugeField &_Umu)
 {
@@ -235,6 +319,7 @@ template <class Impl>
 void WilsonFermion<Impl>::DerivInternal(StencilImpl &st, DoubledGaugeField &U,
                                        GaugeField &mat, const FermionField &A,
                                        const FermionField &B, int dag) {
+  DerivCalls++;
  assert((dag == DaggerNo) || (dag == DaggerYes));

  Compressor compressor(dag);
@@ -243,8 +328,11 @@ void WilsonFermion<Impl>::DerivInternal(StencilImpl &st, DoubledGaugeField &U,
  FermionField Atilde(B.Grid());
  Atilde = A;

+  DerivCommTime-=usecond();
  st.HaloExchange(B, compressor);
+  DerivCommTime+=usecond();

+  DerivComputeTime-=usecond();
  for (int mu = 0; mu < Nd; mu++) {
    ////////////////////////////////////////////////////////////////////////
    // Flip gamma (1+g)<->(1-g) if dag
@@ -252,6 +340,7 @@ void WilsonFermion<Impl>::DerivInternal(StencilImpl &st, DoubledGaugeField &U,
    int gamma = mu;
    if (!dag) gamma += Nd;

+    DerivDhopComputeTime -= usecond();
    int Ls=1;
    Kernels::DhopDirKernel(st, U, st.CommBuf(), Ls, B.Grid()->oSites(), B, Btilde, mu, gamma);

@@ -259,7 +348,9 @@ void WilsonFermion<Impl>::DerivInternal(StencilImpl &st, DoubledGaugeField &U,
    // spin trace outer product
    //////////////////////////////////////////////////
    Impl::InsertForce4D(mat, Btilde, Atilde, mu);
+    DerivDhopComputeTime += usecond();
  }
+  DerivComputeTime += usecond();
 }

 template <class Impl>
@@ -306,6 +397,7 @@ 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());

@@ -317,6 +409,7 @@ 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

@@ -329,6 +422,7 @@ 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

@@ -393,12 +487,14 @@ void WilsonFermion<Impl>::DhopInternal(StencilImpl &st, LebesgueOrder &lo,
                                       const FermionField &in,
                                       FermionField &out, int dag)
 {
+  DhopTotalTime-=usecond();
 #ifdef GRID_OMP
  if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsAndCompute )
    DhopInternalOverlappedComms(st,lo,U,in,out,dag);
  else
 #endif
    DhopInternalSerial(st,lo,U,in,out,dag);
+  DhopTotalTime+=usecond();
 }

 template <class Impl>
@@ -407,7 +503,6 @@ void WilsonFermion<Impl>::DhopInternalOverlappedComms(StencilImpl &st, LebesgueO
 						      const FermionField &in,
 						      FermionField &out, int dag)
 {
-  GRID_TRACE("DhopOverlapped");
  assert((dag == DaggerNo) || (dag == DaggerYes));

  Compressor compressor(dag);
@@ -418,55 +513,53 @@ void WilsonFermion<Impl>::DhopInternalOverlappedComms(StencilImpl &st, LebesgueO
  /////////////////////////////
  std::vector<std::vector<CommsRequest_t> > requests;
  st.Prepare();
-  {
-    GRID_TRACE("Gather");
-    st.HaloGather(in,compressor);
-  }
+  DhopFaceTime-=usecond();
+  st.HaloGather(in,compressor);
+  DhopFaceTime+=usecond();

-  tracePush("Communication");
+  DhopCommTime -=usecond();
  st.CommunicateBegin(requests);

  /////////////////////////////
  // Overlap with comms
  /////////////////////////////
-  {
-    GRID_TRACE("MergeSHM");
-    st.CommsMergeSHM(compressor);
-  }
+  DhopFaceTime-=usecond();
+  st.CommsMergeSHM(compressor);
+  DhopFaceTime+=usecond();

  /////////////////////////////
  // do the compute interior
  /////////////////////////////
  int Opt = WilsonKernelsStatic::Opt;
+  DhopComputeTime-=usecond();
  if (dag == DaggerYes) {
-    GRID_TRACE("DhopDagInterior");
    Kernels::DhopDagKernel(Opt,st,U,st.CommBuf(),1,U.oSites(),in,out,1,0);
  } else {
-    GRID_TRACE("DhopInterior");
    Kernels::DhopKernel(Opt,st,U,st.CommBuf(),1,U.oSites(),in,out,1,0);
  }
+  DhopComputeTime+=usecond();

  /////////////////////////////
  // Complete comms
  /////////////////////////////
  st.CommunicateComplete(requests);
-  tracePop("Communication");
+  DhopCommTime   +=usecond();
+
+  DhopFaceTime-=usecond();
+  st.CommsMerge(compressor);
+  DhopFaceTime+=usecond();

-  {
-    GRID_TRACE("Merge");
-    st.CommsMerge(compressor);
-  }
  /////////////////////////////
  // do the compute exterior
  /////////////////////////////

+  DhopComputeTime2-=usecond();
  if (dag == DaggerYes) {
-    GRID_TRACE("DhopDagExterior");
    Kernels::DhopDagKernel(Opt,st,U,st.CommBuf(),1,U.oSites(),in,out,0,1);
  } else {
-    GRID_TRACE("DhopExterior");
    Kernels::DhopKernel(Opt,st,U,st.CommBuf(),1,U.oSites(),in,out,0,1);
  }
+  DhopComputeTime2+=usecond();
 };


@@ -476,22 +569,20 @@ void WilsonFermion<Impl>::DhopInternalSerial(StencilImpl &st, LebesgueOrder &lo,
                                       const FermionField &in,
                                       FermionField &out, int dag)
 {
-  GRID_TRACE("DhopSerial");
  assert((dag == DaggerNo) || (dag == DaggerYes));
  Compressor compressor(dag);
-  {
-    GRID_TRACE("HaloExchange");
-    st.HaloExchange(in, compressor);
-  }
+  DhopCommTime-=usecond();
+  st.HaloExchange(in, compressor);
+  DhopCommTime+=usecond();

+  DhopComputeTime-=usecond();
  int Opt = WilsonKernelsStatic::Opt;
  if (dag == DaggerYes) {
-    GRID_TRACE("DhopDag");
    Kernels::DhopDagKernel(Opt,st,U,st.CommBuf(),1,U.oSites(),in,out);
  } else {
-    GRID_TRACE("Dhop");
    Kernels::DhopKernel(Opt,st,U,st.CommBuf(),1,U.oSites(),in,out);
  }
+  DhopComputeTime+=usecond();
 };
 /*Change ends */

@@ -508,47 +599,11 @@ void WilsonFermion<Impl>::ContractConservedCurrent(PropagatorField &q_in_1,
                                                   Current curr_type,
                                                   unsigned int mu)
 {
-  if(curr_type != Current::Vector)
-  {
-    std::cout << GridLogError << "Only the conserved vector current is implemented so far." << std::endl;
-    exit(1);
-  }
-
  Gamma g5(Gamma::Algebra::Gamma5);
  conformable(_grid, q_in_1.Grid());
  conformable(_grid, q_in_2.Grid());
  conformable(_grid, q_out.Grid());
-  auto UGrid= this->GaugeGrid();
-
-  PropagatorField tmp_shifted(UGrid);
-  PropagatorField g5Lg5(UGrid);
-  PropagatorField R(UGrid);
-  PropagatorField gmuR(UGrid);
-
-    Gamma::Algebra Gmu [] = {
-    Gamma::Algebra::GammaX,
-    Gamma::Algebra::GammaY,
-    Gamma::Algebra::GammaZ,
-    Gamma::Algebra::GammaT,
-  };
-  Gamma gmu=Gamma(Gmu[mu]);
-
-  g5Lg5=g5*q_in_1*g5;
-  tmp_shifted=Cshift(q_in_2,mu,1);
-  Impl::multLinkField(R,this->Umu,tmp_shifted,mu);
-  gmuR=gmu*R;
-
-  q_out=adj(g5Lg5)*R;
-  q_out-=adj(g5Lg5)*gmuR;
-
-  tmp_shifted=Cshift(q_in_1,mu,1);
-  Impl::multLinkField(g5Lg5,this->Umu,tmp_shifted,mu);
-  g5Lg5=g5*g5Lg5*g5;
-  R=q_in_2;
-  gmuR=gmu*R;
-
-  q_out-=adj(g5Lg5)*R;
-  q_out-=adj(g5Lg5)*gmuR;
+  assert(0);
 }


@@ -562,51 +617,9 @@ void WilsonFermion<Impl>::SeqConservedCurrent(PropagatorField &q_in,
                                              unsigned int tmax,
 					      ComplexField &lattice_cmplx)
 {
-  if(curr_type != Current::Vector)
-  {
-    std::cout << GridLogError << "Only the conserved vector current is implemented so far." << std::endl;
-    exit(1);
-  }
-
-  int tshift = (mu == Nd-1) ? 1 : 0;
-  unsigned int LLt    = GridDefaultLatt()[Tp];
  conformable(_grid, q_in.Grid());
  conformable(_grid, q_out.Grid());
-  auto UGrid= this->GaugeGrid();
-
-  PropagatorField tmp(UGrid);
-  PropagatorField Utmp(UGrid);
-  PropagatorField L(UGrid);
-  PropagatorField zz (UGrid);
-  zz=Zero();
-  LatticeInteger lcoor(UGrid); LatticeCoordinate(lcoor,Nd-1);
-
-    Gamma::Algebra Gmu [] = {
-    Gamma::Algebra::GammaX,
-    Gamma::Algebra::GammaY,
-    Gamma::Algebra::GammaZ,
-    Gamma::Algebra::GammaT,
-  };
-  Gamma gmu=Gamma(Gmu[mu]);
-
-  tmp = Cshift(q_in,mu,1);
-  Impl::multLinkField(Utmp,this->Umu,tmp,mu);
-  tmp = ( Utmp*lattice_cmplx - gmu*Utmp*lattice_cmplx ); // Forward hop
-  tmp = where((lcoor>=tmin),tmp,zz); // Mask the time
-  q_out = where((lcoor<=tmax),tmp,zz); // Position of current complicated
-
-  tmp = q_in *lattice_cmplx;
-  tmp = Cshift(tmp,mu,-1);
-  Impl::multLinkField(Utmp,this->Umu,tmp,mu+Nd); // Adjoint link
-  tmp = -( Utmp + gmu*Utmp );
-  // Mask the time
-  if (tmax == LLt - 1 && tshift == 1){ // quick fix to include timeslice 0 if tmax + tshift is over the last timeslice
-    unsigned int t0 = 0;
-    tmp = where(((lcoor==t0) || (lcoor>=tmin+tshift)),tmp,zz);
-  } else {
-    tmp = where((lcoor>=tmin+tshift),tmp,zz);
-  }
-  q_out+= where((lcoor<=tmax+tshift),tmp,zz); // Position of current complicated
+  assert(0);
 }

 NAMESPACE_END(Grid);
@@ -73,17 +73,17 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>

-//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-//template<> void
-//WilsonKernels<WilsonImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
+template<> void
+WilsonKernels<WilsonImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>

-//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-//template<> void
-//WilsonKernels<ZWilsonImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
+template<> void
+WilsonKernels<ZWilsonImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>


 #undef INTERIOR_AND_EXTERIOR
@@ -102,17 +102,17 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldVi
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>

-//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-//template<> void
-//WilsonKernels<WilsonImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
+template<> void
+WilsonKernels<WilsonImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>

-//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-//template<> void
-//WilsonKernels<ZWilsonImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
+template<> void
+WilsonKernels<ZWilsonImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>


 #undef INTERIOR_AND_EXTERIOR
@@ -131,17 +131,17 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldVi
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>

-//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-//template<> void
-//WilsonKernels<WilsonImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
+template<> void
+WilsonKernels<WilsonImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>

-//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-//template<> void
-//WilsonKernels<ZWilsonImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
+template<> void
+WilsonKernels<ZWilsonImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>



@@ -165,17 +165,17 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldVi
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>

-//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-//template<> void
-//WilsonKernels<WilsonImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
+template<> void
+WilsonKernels<WilsonImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>

-//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-//template<> void
-//WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
+template<> void
+WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>


 #undef INTERIOR_AND_EXTERIOR
@@ -194,17 +194,17 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFiel
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>

-//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-//template<> void
-//WilsonKernels<WilsonImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
+template<> void
+WilsonKernels<WilsonImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>

-//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-//template<> void
-//WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
+template<> void
+WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>


 #undef INTERIOR_AND_EXTERIOR
@@ -223,17 +223,17 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFiel
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>

-//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-//template<> void
-//WilsonKernels<WilsonImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
+template<> void
+WilsonKernels<WilsonImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>

-//#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-//template<> void
-//WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
+template<> void
+WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>



@@ -280,17 +280,17 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>

-// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-// template<> void
-// WilsonKernels<WilsonImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-// 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
+template<> void
+WilsonKernels<WilsonImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>

-// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-// template<> void
-// WilsonKernels<ZWilsonImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-// 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
+template<> void
+WilsonKernels<ZWilsonImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>


 #undef INTERIOR_AND_EXTERIOR
@@ -309,17 +309,17 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldVi
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>

-// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-// template<> void
-// WilsonKernels<WilsonImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-// 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
+template<> void
+WilsonKernels<WilsonImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>

-// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-// template<> void
-// WilsonKernels<ZWilsonImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-// 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
+template<> void
+WilsonKernels<ZWilsonImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>


 #undef INTERIOR_AND_EXTERIOR
@@ -338,17 +338,17 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldVi
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>

-// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-// template<> void
-// WilsonKernels<WilsonImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-// 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
+template<> void
+WilsonKernels<WilsonImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>

-// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-// template<> void
-// WilsonKernels<ZWilsonImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-// 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
+template<> void
+WilsonKernels<ZWilsonImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>


 /////////////////////////////////////////////////////////////////
@@ -371,17 +371,17 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldVi
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>

-// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-// template<> void
-// WilsonKernels<WilsonImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-// 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
+template<> void
+WilsonKernels<WilsonImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>

-// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-// template<> void
-// WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-// 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
+template<> void
+WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>


 #undef INTERIOR_AND_EXTERIOR
@@ -400,17 +400,17 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFiel
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>

-// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-// template<> void
-// WilsonKernels<WilsonImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-// 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
+template<> void
+WilsonKernels<WilsonImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>

-// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-// template<> void
-// WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-// 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
+template<> void
+WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>


 #undef INTERIOR_AND_EXTERIOR
@@ -429,17 +429,17 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFiel
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>

-// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-// template<> void
-// WilsonKernels<WilsonImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-// 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
+template<> void
+WilsonKernels<WilsonImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>

-// #pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
-// template<> void
-// WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-// 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-// #include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
+#pragma GCC optimize ("-O3", "-fno-schedule-insns", "-fno-schedule-insns2")
+template<> void
+WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>



@@ -74,15 +74,15 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

-//template<> void
-//WilsonKernels<WilsonImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//
-//template<> void
-//WilsonKernels<ZWilsonImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<WilsonImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+
+template<> void 
+WilsonKernels<ZWilsonImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

 #undef INTERIOR_AND_EXTERIOR
 #define INTERIOR
@@ -97,15 +97,15 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldVi
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

-//template<> void
-//WilsonKernels<WilsonImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//
-//template<> void
-//WilsonKernels<ZWilsonImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<WilsonImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+
+template<> void 
+WilsonKernels<ZWilsonImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>


 #undef INTERIOR_AND_EXTERIOR
@@ -121,15 +121,15 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldVi
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

-//template<> void
-//WilsonKernels<WilsonImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//
-//template<> void
-//WilsonKernels<ZWilsonImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<WilsonImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+
+template<> void 
+WilsonKernels<ZWilsonImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
      
 /////////////////////////////////////////////////////////////////
 // XYZT vectorised, dag Kernel, single
@@ -148,15 +148,15 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldVi
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

-//template<> void
-//WilsonKernels<WilsonImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//
-//template<> void
-//WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<WilsonImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+
+template<> void 
+WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

 #undef INTERIOR_AND_EXTERIOR
 #define INTERIOR
@@ -171,15 +171,15 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFiel
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

-//template<> void
-//WilsonKernels<WilsonImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//
-//template<> void
-//WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<WilsonImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+
+template<> void 
+WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

 #undef INTERIOR_AND_EXTERIOR
 #undef INTERIOR
@@ -194,15 +194,15 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFiel
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 				    
-//template<> void
-//WilsonKernels<WilsonImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//
-//template<> void
-//WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<WilsonImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+				    
+template<> void 
+WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 				    
 #undef MAYBEPERM
 #undef MULT_2SPIN
@@ -228,14 +228,14 @@ WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSite(StencilView &st, DoubledGaugeF
 							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

-//template<> void
-//WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

 #undef INTERIOR_AND_EXTERIOR
 #define INTERIOR
@@ -249,14 +249,14 @@ WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSiteInt(StencilView &st, DoubledGau
 							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

-//template<> void
-//WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

 #undef INTERIOR_AND_EXTERIOR
 #undef INTERIOR
@@ -273,15 +273,15 @@ WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSiteExt(StencilView &st, DoubledGau
 							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 				    
-//template<> void
-//WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//
-//template<> void
-//WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+				    
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 				    
 /////////////////////////////////////////////////////////////////
 // Ls vectorised, dag Kernel, single
@@ -299,14 +299,14 @@ WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSiteDag(StencilView &st, DoubledGau
 							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

-//template<> void
-//WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
-//							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
-//							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

 #undef INTERIOR_AND_EXTERIOR
 #define INTERIOR
@@ -320,14 +320,14 @@ WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSiteDagInt(StencilView &st, Doubled
 							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

-//template<> void
-//WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
-//							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
-//							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

 #undef INTERIOR_AND_EXTERIOR
 #undef INTERIOR
@@ -341,14 +341,14 @@ WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSiteDagExt(StencilView &st, Doubled
 							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

-//template<> void
-//WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
-//							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
-//							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

 #endif  // VEC 5D

@@ -392,14 +392,14 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

-//template<> void
-//WilsonKernels<WilsonImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZWilsonImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<WilsonImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZWilsonImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

 #undef INTERIOR_AND_EXTERIOR
 #define INTERIOR
@@ -413,14 +413,14 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldVi
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

-//template<> void
-//WilsonKernels<WilsonImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZWilsonImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<WilsonImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZWilsonImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

 #undef INTERIOR_AND_EXTERIOR
 #undef INTERIOR
@@ -434,14 +434,14 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldVi
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
      
-//template<> void
-//WilsonKernels<WilsonImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZWilsonImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<WilsonImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZWilsonImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
      
 /////////////////////////////////////////////////////////////////
 // XYZT vectorised, dag Kernel, single
@@ -459,14 +459,14 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldVi
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

-//template<> void
-//WilsonKernels<WilsonImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<WilsonImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

 #undef INTERIOR_AND_EXTERIOR
 #define INTERIOR
@@ -480,14 +480,14 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFiel
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

-//template<> void
-//WilsonKernels<WilsonImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<WilsonImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

 #undef INTERIOR_AND_EXTERIOR
 #undef INTERIOR
@@ -501,14 +501,14 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFiel
 						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 				    
-//template<> void
-//WilsonKernels<WilsonImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<WilsonImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 				    
 #undef MAYBEPERM
 #undef MULT_2SPIN
@@ -533,14 +533,14 @@ WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSite(StencilView &st, DoubledGaugeF
 							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

-//template<> void
-//WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

 #undef INTERIOR_AND_EXTERIOR
 #define INTERIOR
@@ -554,14 +554,14 @@ WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSiteInt(StencilView &st, DoubledGau
 							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

-//template<> void
-//WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

 #undef INTERIOR_AND_EXTERIOR
 #undef INTERIOR
@@ -577,14 +577,14 @@ WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSiteExt(StencilView &st, DoubledGau
 							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 				    
-//template<> void
-//WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-//							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
 				    
 /////////////////////////////////////////////////////////////////
 // Ls vectorised, dag Kernel, single
@@ -602,14 +602,14 @@ WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSiteDag(StencilView &st, DoubledGau
 							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

-//template<> void
-//WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
-//							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
-//							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

 #undef INTERIOR_AND_EXTERIOR
 #define INTERIOR
@@ -623,14 +623,14 @@ WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSiteDagInt(StencilView &st, Doubled
 							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

-//template<> void
-//WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
-//							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
-//							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

 #undef INTERIOR_AND_EXTERIOR
 #undef INTERIOR
@@ -645,14 +645,14 @@ WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSiteDagExt(StencilView &st, Doubled
 							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
 #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

-//template<> void
-//WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
-//							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-//template<> void
-//WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
-//							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-//#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
+#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>

 #endif  // VEC 5D

@@ -77,23 +77,23 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 #define REGISTER

 #ifdef GRID_SIMT
-#define LOAD_CHIMU(Ptype)		\
+#define LOAD_CHIMU(ptype)		\
  {const SiteSpinor & ref (in[offset]);	\
-    Chimu_00=coalescedReadPermute<Ptype>(ref()(0)(0),perm,lane);	\
-    Chimu_01=coalescedReadPermute<Ptype>(ref()(0)(1),perm,lane);		\
-    Chimu_02=coalescedReadPermute<Ptype>(ref()(0)(2),perm,lane);		\
-    Chimu_10=coalescedReadPermute<Ptype>(ref()(1)(0),perm,lane);		\
-    Chimu_11=coalescedReadPermute<Ptype>(ref()(1)(1),perm,lane);		\
-    Chimu_12=coalescedReadPermute<Ptype>(ref()(1)(2),perm,lane);		\
-    Chimu_20=coalescedReadPermute<Ptype>(ref()(2)(0),perm,lane);		\
-    Chimu_21=coalescedReadPermute<Ptype>(ref()(2)(1),perm,lane);		\
-    Chimu_22=coalescedReadPermute<Ptype>(ref()(2)(2),perm,lane);		\
-    Chimu_30=coalescedReadPermute<Ptype>(ref()(3)(0),perm,lane);		\
-    Chimu_31=coalescedReadPermute<Ptype>(ref()(3)(1),perm,lane);		\
-    Chimu_32=coalescedReadPermute<Ptype>(ref()(3)(2),perm,lane);	}
+    Chimu_00=coalescedReadPermute<ptype>(ref()(0)(0),perm,lane);	\
+    Chimu_01=coalescedReadPermute<ptype>(ref()(0)(1),perm,lane);		\
+    Chimu_02=coalescedReadPermute<ptype>(ref()(0)(2),perm,lane);		\
+    Chimu_10=coalescedReadPermute<ptype>(ref()(1)(0),perm,lane);		\
+    Chimu_11=coalescedReadPermute<ptype>(ref()(1)(1),perm,lane);		\
+    Chimu_12=coalescedReadPermute<ptype>(ref()(1)(2),perm,lane);		\
+    Chimu_20=coalescedReadPermute<ptype>(ref()(2)(0),perm,lane);		\
+    Chimu_21=coalescedReadPermute<ptype>(ref()(2)(1),perm,lane);		\
+    Chimu_22=coalescedReadPermute<ptype>(ref()(2)(2),perm,lane);		\
+    Chimu_30=coalescedReadPermute<ptype>(ref()(3)(0),perm,lane);		\
+    Chimu_31=coalescedReadPermute<ptype>(ref()(3)(1),perm,lane);		\
+    Chimu_32=coalescedReadPermute<ptype>(ref()(3)(2),perm,lane);	}
 #define PERMUTE_DIR(dir) ;
 #else
-#define LOAD_CHIMU(Ptype)		\
+#define LOAD_CHIMU(ptype)		\
  {const SiteSpinor & ref (in[offset]);	\
    Chimu_00=ref()(0)(0);\
    Chimu_01=ref()(0)(1);\
@@ -109,12 +109,12 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
    Chimu_32=ref()(3)(2);}

 #define PERMUTE_DIR(dir)			\
-  permute##dir(Chi_00,Chi_00);			\
-  permute##dir(Chi_01,Chi_01);			\
-  permute##dir(Chi_02,Chi_02);			\
-  permute##dir(Chi_10,Chi_10);			\
-  permute##dir(Chi_11,Chi_11);			\
-  permute##dir(Chi_12,Chi_12);
+  permute##dir(Chi_00,Chi_00);	\
+      permute##dir(Chi_01,Chi_01);\
+      permute##dir(Chi_02,Chi_02);\
+      permute##dir(Chi_10,Chi_10);	\
+      permute##dir(Chi_11,Chi_11);\
+      permute##dir(Chi_12,Chi_12);

 #endif

@@ -371,91 +371,88 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
  result_32-= UChi_12;

 #define HAND_STENCIL_LEGB(PROJ,PERM,DIR,RECON)	\
-  {int ptype;					\
-   SE=st.GetEntry(ptype,DIR,ss);		\
-   auto offset = SE->_offset;			\
-   auto local  = SE->_is_local;			\
-   auto perm   = SE->_permute;			\
-   if ( local ) {				\
-     LOAD_CHIMU(PERM);				\
-     PROJ;					\
-     if ( perm) {				\
-       PERMUTE_DIR(PERM);			\
-     }						\
-   } else {					\
-     LOAD_CHI;					\
-   }						\
-   acceleratorSynchronise();			\
-   MULT_2SPIN(DIR);				\
-   RECON;					}
+  SE=st.GetEntry(ptype,DIR,ss);			\
+  offset = SE->_offset;				\
+  local  = SE->_is_local;			\
+  perm   = SE->_permute;			\
+  if ( local ) {				\
+    LOAD_CHIMU(PERM);				\
+    PROJ;					\
+    if ( perm) {				\
+      PERMUTE_DIR(PERM);			\
+    }						\
+  } else {					\
+    LOAD_CHI;					\
+  }						\
+  acceleratorSynchronise();			\
+  MULT_2SPIN(DIR);				\
+  RECON;					

-#define HAND_STENCIL_LEG(PROJ,PERM,DIR,RECON)		\
-  { SE=&st_p[DIR+8*ss];						\
-  auto ptype=st_perm[DIR];					\
-  auto offset = SE->_offset;					\
-  auto local  = SE->_is_local;					\
-  auto perm   = SE->_permute;					\
-  if ( local ) {						\
-    LOAD_CHIMU(PERM);						\
-    PROJ;							\
-    if ( perm) {						\
-      PERMUTE_DIR(PERM);					\
-    }								\
-  } else {							\
-    LOAD_CHI;							\
-  }								\
-  acceleratorSynchronise();					\
-  MULT_2SPIN(DIR);						\
-  RECON;					}
+#define HAND_STENCIL_LEG(PROJ,PERM,DIR,RECON)	\
+  SE=&st_p[DIR+8*ss];				\
+  ptype=st_perm[DIR];				\
+  offset = SE->_offset;				\
+  local  = SE->_is_local;			\
+  perm   = SE->_permute;			\
+  if ( local ) {				\
+    LOAD_CHIMU(PERM);				\
+    PROJ;					\
+    if ( perm) {				\
+      PERMUTE_DIR(PERM);			\
+    }						\
+  } else {					\
+    LOAD_CHI;					\
+  }						\
+  acceleratorSynchronise();			\
+  MULT_2SPIN(DIR);				\
+  RECON;					

 #define HAND_STENCIL_LEGA(PROJ,PERM,DIR,RECON)				\
-  { SE=&st_p[DIR+8*ss];							\
-    auto ptype=st_perm[DIR];						\
-    /*SE=st.GetEntry(ptype,DIR,ss);*/					\
-    auto offset = SE->_offset;						\
-    auto perm   = SE->_permute;						\
-    LOAD_CHIMU(PERM);							\
-    PROJ;								\
-    MULT_2SPIN(DIR);							\
-    RECON;					}
+  SE=&st_p[DIR+8*ss];							\
+  ptype=st_perm[DIR];							\
+ /*SE=st.GetEntry(ptype,DIR,ss);*/					\
+  offset = SE->_offset;				\
+  perm   = SE->_permute;			\
+  LOAD_CHIMU(PERM);				\
+  PROJ;						\
+  MULT_2SPIN(DIR);				\
+  RECON;					

 #define HAND_STENCIL_LEG_INT(PROJ,PERM,DIR,RECON)	\
-  { int ptype;						\
-  SE=st.GetEntry(ptype,DIR,ss);				\
-  auto offset = SE->_offset;					\
-  auto local  = SE->_is_local;					\
-  auto perm   = SE->_permute;					\
-  if ( local ) {						\
-    LOAD_CHIMU(PERM);						\
-    PROJ;							\
-    if ( perm) {						\
-      PERMUTE_DIR(PERM);					\
-    }								\
-  } else if ( st.same_node[DIR] ) {				\
-    LOAD_CHI;							\
-  }								\
-  acceleratorSynchronise();					\
-  if (local || st.same_node[DIR] ) {				\
-    MULT_2SPIN(DIR);						\
-    RECON;							\
-  }								\
-  acceleratorSynchronise();			}
+  SE=st.GetEntry(ptype,DIR,ss);			\
+  offset = SE->_offset;				\
+  local  = SE->_is_local;			\
+  perm   = SE->_permute;			\
+  if ( local ) {				\
+    LOAD_CHIMU(PERM);				\
+    PROJ;					\
+    if ( perm) {				\
+      PERMUTE_DIR(PERM);			\
+    }						\
+  } else if ( st.same_node[DIR] ) {		\
+    LOAD_CHI;					\
+  }						\
+  acceleratorSynchronise();			\
+  if (local || st.same_node[DIR] ) {		\
+    MULT_2SPIN(DIR);				\
+    RECON;					\
+  }						\
+  acceleratorSynchronise();			

 #define HAND_STENCIL_LEG_EXT(PROJ,PERM,DIR,RECON)	\
-  { int ptype;						\
-  SE=st.GetEntry(ptype,DIR,ss);				\
-  auto offset = SE->_offset;				\
-  if((!SE->_is_local)&&(!st.same_node[DIR]) ) {		\
-    LOAD_CHI;						\
-    MULT_2SPIN(DIR);					\
-    RECON;						\
-    nmu++;						\
-  }							\
-  acceleratorSynchronise();			}
+  SE=st.GetEntry(ptype,DIR,ss);			\
+  offset = SE->_offset;				\
+  if((!SE->_is_local)&&(!st.same_node[DIR]) ) {	\
+    LOAD_CHI;					\
+    MULT_2SPIN(DIR);				\
+    RECON;					\
+    nmu++;					\
+  }						\
+  acceleratorSynchronise();			

-#define HAND_RESULT(ss)					\
-  {							\
-    SiteSpinor & ref (out[ss]);				\
+#define HAND_RESULT(ss)				\
+  {						\
+    SiteSpinor & ref (out[ss]);			\
    coalescedWrite(ref()(0)(0),result_00,lane);		\
    coalescedWrite(ref()(0)(1),result_01,lane);		\
    coalescedWrite(ref()(0)(2),result_02,lane);		\
@@ -566,6 +563,7 @@ WilsonKernels<Impl>::HandDhopSiteSycl(StencilVector st_perm,StencilEntry *st_p,

  HAND_DECLARATIONS(Simt);

+  int offset,local,perm, ptype;
  StencilEntry *SE;
  HAND_STENCIL_LEG(XM_PROJ,3,Xp,XM_RECON);
  HAND_STENCIL_LEG(YM_PROJ,2,Yp,YM_RECON_ACCUM);
@@ -595,7 +593,9 @@ WilsonKernels<Impl>::HandDhopSite(StencilView &st, DoubledGaugeFieldView &U,Site

  HAND_DECLARATIONS(Simt);

+  int offset,local,perm, ptype;
  StencilEntry *SE;
+
  HAND_STENCIL_LEG(XM_PROJ,3,Xp,XM_RECON);
  HAND_STENCIL_LEG(YM_PROJ,2,Yp,YM_RECON_ACCUM);
  HAND_STENCIL_LEG(ZM_PROJ,1,Zp,ZM_RECON_ACCUM);
@@ -623,6 +623,8 @@ void WilsonKernels<Impl>::HandDhopSiteDag(StencilView &st,DoubledGaugeFieldView
  HAND_DECLARATIONS(Simt);

  StencilEntry *SE;
+  int offset,local,perm, ptype;
+  
  HAND_STENCIL_LEG(XP_PROJ,3,Xp,XP_RECON);
  HAND_STENCIL_LEG(YP_PROJ,2,Yp,YP_RECON_ACCUM);
  HAND_STENCIL_LEG(ZP_PROJ,1,Zp,ZP_RECON_ACCUM);
@@ -638,8 +640,8 @@ template<class Impl>  accelerator_inline void
 WilsonKernels<Impl>::HandDhopSiteInt(StencilView &st,DoubledGaugeFieldView &U,SiteHalfSpinor  *buf,
 					  int ss,int sU,const FermionFieldView &in, FermionFieldView &out)
 {
-  //  auto st_p = st._entries_p;						
-  //  auto st_perm = st._permute_type;					
+  auto st_p = st._entries_p;						
+  auto st_perm = st._permute_type;					
 // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
  typedef typename Simd::scalar_type S;
  typedef typename Simd::vector_type V;
@@ -650,6 +652,7 @@ WilsonKernels<Impl>::HandDhopSiteInt(StencilView &st,DoubledGaugeFieldView &U,Si

  HAND_DECLARATIONS(Simt);

+  int offset,local,perm, ptype;
  StencilEntry *SE;
  ZERO_RESULT;
  HAND_STENCIL_LEG_INT(XM_PROJ,3,Xp,XM_RECON_ACCUM);
@@ -667,8 +670,8 @@ template<class Impl> accelerator_inline
 void WilsonKernels<Impl>::HandDhopSiteDagInt(StencilView &st,DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
 						  int ss,int sU,const FermionFieldView &in, FermionFieldView &out)
 {
-  //  auto st_p = st._entries_p;						
-  //  auto st_perm = st._permute_type;					
+  auto st_p = st._entries_p;						
+  auto st_perm = st._permute_type;					
  typedef typename Simd::scalar_type S;
  typedef typename Simd::vector_type V;
  typedef decltype( coalescedRead( in[0]()(0)(0) )) Simt;
@@ -679,6 +682,7 @@ void WilsonKernels<Impl>::HandDhopSiteDagInt(StencilView &st,DoubledGaugeFieldVi
  HAND_DECLARATIONS(Simt);

  StencilEntry *SE;
+  int offset,local,perm, ptype;
  ZERO_RESULT;
  HAND_STENCIL_LEG_INT(XP_PROJ,3,Xp,XP_RECON_ACCUM);
  HAND_STENCIL_LEG_INT(YP_PROJ,2,Yp,YP_RECON_ACCUM);
@@ -695,8 +699,8 @@ template<class Impl>  accelerator_inline void
 WilsonKernels<Impl>::HandDhopSiteExt(StencilView &st,DoubledGaugeFieldView &U,SiteHalfSpinor  *buf,
 					  int ss,int sU,const FermionFieldView &in, FermionFieldView &out)
 {
-  //  auto st_p = st._entries_p;						
-  //  auto st_perm = st._permute_type;					
+  auto st_p = st._entries_p;						
+  auto st_perm = st._permute_type;					
 // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
  typedef typename Simd::scalar_type S;
  typedef typename Simd::vector_type V;
@@ -707,7 +711,7 @@ WilsonKernels<Impl>::HandDhopSiteExt(StencilView &st,DoubledGaugeFieldView &U,Si

  HAND_DECLARATIONS(Simt);

-  //  int offset, ptype;
+  int offset, ptype;
  StencilEntry *SE;
  int nmu=0;
  ZERO_RESULT;
@@ -726,8 +730,8 @@ template<class Impl>  accelerator_inline
 void WilsonKernels<Impl>::HandDhopSiteDagExt(StencilView &st,DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
 						  int ss,int sU,const FermionFieldView &in, FermionFieldView &out)
 {
-  //  auto st_p = st._entries_p;						
-  //  auto st_perm = st._permute_type;					
+  auto st_p = st._entries_p;						
+  auto st_perm = st._permute_type;					
  typedef typename Simd::scalar_type S;
  typedef typename Simd::vector_type V;
  typedef decltype( coalescedRead( in[0]()(0)(0) )) Simt;
@@ -738,7 +742,7 @@ void WilsonKernels<Impl>::HandDhopSiteDagExt(StencilView &st,DoubledGaugeFieldVi
  HAND_DECLARATIONS(Simt);

  StencilEntry *SE;
-  //  int offset, ptype;
+  int offset, ptype;
  int nmu=0;
  ZERO_RESULT;
  HAND_STENCIL_LEG_EXT(XP_PROJ,3,Xp,XP_RECON_ACCUM);
@@ -72,15 +72,20 @@ accelerator_inline void get_stencil(StencilEntry * mem, StencilEntry &chip)
  if (SE->_is_local) {						\
    int perm= SE->_permute;					\
    auto tmp = coalescedReadPermute(in[SE->_offset],ptype,perm,lane);	\
-    spProj(chi,tmp);							\
-    Impl::multLink(Uchi, U[sU], chi, Dir, SE, st);			\
-    Recon(result, Uchi);						\
-  }									\
+    spProj(chi,tmp);						\
+  } else if ( st.same_node[Dir] ) {				\
+    chi = coalescedRead(buf[SE->_offset],lane);			\
+  }								\
+  acceleratorSynchronise();						\
+  if (SE->_is_local || st.same_node[Dir] ) {			\
+    Impl::multLink(Uchi, U[sU], chi, Dir, SE, st);		\
+    Recon(result, Uchi);					\
+  }								\
  acceleratorSynchronise();

 #define GENERIC_STENCIL_LEG_EXT(Dir,spProj,Recon)		\
  SE = st.GetEntry(ptype, Dir, sF);				\
-  if (!SE->_is_local ) {		\
+  if ((!SE->_is_local) && (!st.same_node[Dir]) ) {		\
    auto chi = coalescedRead(buf[SE->_offset],lane);		\
    Impl::multLink(Uchi, U[sU], chi, Dir, SE, st);		\
    Recon(result, Uchi);					\
@@ -411,6 +416,19 @@ 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)						\
  const uint64_t    NN = Nsite*Ls;					\
@@ -422,16 +440,6 @@ void WilsonKernels<Impl>::DhopDirKernel( StencilImpl &st, DoubledGaugeField &U,S

 #define KERNEL_CALL(A) KERNEL_CALLNB(A); accelerator_barrier();

-#define KERNEL_CALL_EXT(A)						\
-  const uint64_t    NN = Nsite*Ls;					\
-  const uint64_t    sz = st.surface_list.size();			\
-  auto ptr = &st.surface_list[0];					\
-  accelerator_forNB( ss, sz, Simd::Nsimd(), {				\
-      int sF = ptr[ss];							\
-      int sU = ss/Ls;							\
-      WilsonKernels<Impl>::A(st_v,U_v,buf,sF,sU,in_v,out_v);		\
-    });									
-
 #define ASM_CALL(A)							\
  thread_for( ss, Nsite, {						\
    int sU = ss;							\
@@ -452,7 +460,7 @@ 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(HandDhopSiteSycl);    return; }
+     if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL_TMP(HandDhopSiteSycl);    return; }
 #else
     if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSite);    return;}
 #endif     
@@ -490,7 +498,6 @@ void WilsonKernels<Impl>::DhopKernel(int Opt,StencilImpl &st,  DoubledGaugeField
 #ifndef GRID_CUDA
     if (Opt == WilsonKernelsStatic::OptInlineAsm  ) {  ASM_CALL(AsmDhopSiteDag);     return;}
 #endif
-     acceleratorFenceComputeStream();
   } else if( interior ) {
     if (Opt == WilsonKernelsStatic::OptGeneric    ) { KERNEL_CALL(GenericDhopSiteDagInt); return;}
     if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSiteDagInt);    return;}
@@ -498,13 +505,11 @@ void WilsonKernels<Impl>::DhopKernel(int Opt,StencilImpl &st,  DoubledGaugeField
     if (Opt == WilsonKernelsStatic::OptInlineAsm  ) {  ASM_CALL(AsmDhopSiteDagInt);     return;}
 #endif
   } else if( exterior ) {
-     acceleratorFenceComputeStream();
     if (Opt == WilsonKernelsStatic::OptGeneric    ) { KERNEL_CALL(GenericDhopSiteDagExt); return;}
     if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSiteDagExt);    return;}
 #ifndef GRID_CUDA
     if (Opt == WilsonKernelsStatic::OptInlineAsm  ) {  ASM_CALL(AsmDhopSiteDagExt);     return;}
 #endif
-     acceleratorFenceComputeStream();
   }
   assert(0 && " Kernel optimisation case not covered ");
  }
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: ./lib/ qcd/action/fermion/instantiation/CompactWilsonCloverFermionInstantiation.cc.master
-
-    Copyright (C) 2017 - 2022
-
-    Author: paboyle <paboyle@ph.ed.ac.uk>
-    Author: Guido Cossu <guido.cossu@ed.ac.uk>
-    Author: Daniel Richtmann <daniel.richtmann@gmail.com>
-    Author: Mattia Bruno <mattia.bruno@cern.ch>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-    *************************************************************************************/
-/*  END LEGAL */
-
-#include <Grid/Grid.h>
-#include <Grid/qcd/spin/Dirac.h>
-#include <Grid/qcd/action/fermion/CompactWilsonCloverFermion.h>
-#include <Grid/qcd/action/fermion/implementation/CompactWilsonCloverFermionImplementation.h>
-#include <Grid/qcd/action/fermion/CloverHelpers.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class CompactWilsonCloverFermion<IMPLEMENTATION, CompactCloverHelpers<IMPLEMENTATION>>; 
-template class CompactWilsonCloverFermion<IMPLEMENTATION, CompactExpCloverHelpers<IMPLEMENTATION>>; 
-
-NAMESPACE_END(Grid);
Author	SHA1	Message	Date
Chulwoo Jung	b2493d6d25	Switching Block lanczos precision to explicitly single Adding sample run script and input file	2022-03-08 10:18:36 -08:00
Chulwoo Jung	8fd16686dc	Checking block lanczos deleted some diag outputs	2021-09-05 23:04:41 -04:00
Chulwoo Jung	23b9c6b5f5	Merge branch 'develop' of https://github.com/paboyle/Grid into feature/block_lanczos	2021-09-03 17:38:10 -04:00
Chulwoo Jung	43943008bf	Merge branch 'develop' of https://github.com/paboyle/Grid into feature/block_lanczos	2020-10-01 15:35:29 -04:00
Chulwoo Jung	8ee26f9112	Checking in before pulling develop	2020-10-01 15:35:03 -04:00
Chulwoo Jung	f91e3af97f	Checking in before trying to reduce memory footprint	2020-08-08 22:11:14 -04:00
Chulwoo Jung	43298ef681	Make precision switchable for cublas routines	2020-07-22 14:49:32 -04:00
Chulwoo Jung	7e70df27e4	Confirmed double precision working	2020-07-21 00:48:46 -04:00
Chulwoo Jung	c55d657736	Merge branch 'dev-BlockLanczosOpt' of https://github.com/yongchull/Grid into feature/block_lanczos	2020-07-20 16:36:46 -04:00
Yong-Chull Jang	b89b1280d5	use gemm twice to complete the Gram Schmidt	2020-03-31 05:39:31 -04:00
Yong-Chull Jang	ac7090e6d3	block Lanczos cublas buffer is set at the inital step; buffer width is fixed to the block size then cublas Zgemm is called multiple times	2020-03-30 22:25:50 -04:00
Yong-Chull Jang	02edbe624f	first working version of Gram Schmidt using cublas gemm; explicit data type and site vector size has to be removed	2020-03-30 18:36:21 -04:00
Yong-Chull Jang	9266b89ad8	fix rngs issue; block Lanczos is working	2020-03-25 15:45:50 -04:00
Yong-Chull Jang	2db7e6f8ab	merge manually Block Lanczos files from Chulwoo's update (last state = commit 731a05 + untracked files) to develop branch; namespace QCD is removed; FIXME: multiple starting vectors result in nan after initial orthogonalization	2020-03-24 01:03:24 -04:00