diff --git a/Grid/GridCore.h b/Grid/GridCore.h
index a48d2d49..2209f960 100644
--- a/Grid/GridCore.h
+++ b/Grid/GridCore.h
@@ -47,9 +47,9 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 #include <Grid/perfmon/PerfCount.h>
 #include <Grid/util/Util.h>
 #include <Grid/log/Log.h>
-#include <Grid/allocator/AlignedAllocator.h>
+#include <Grid/allocator/Allocator.h>
 #include <Grid/simd/Simd.h>
-#include <Grid/threads/Threads.h>
+#include <Grid/threads/ThreadReduction.h>
 #include <Grid/serialisation/Serialisation.h>
 #include <Grid/util/Sha.h>
 #include <Grid/communicator/Communicator.h> 
diff --git a/Grid/GridStd.h b/Grid/GridStd.h
index 16cfcf50..ecb561ea 100644
--- a/Grid/GridStd.h
+++ b/Grid/GridStd.h
@@ -6,6 +6,7 @@
 ///////////////////
 #include <cassert>
 #include <complex>
+#include <memory>
 #include <vector>
 #include <array>
 #include <string>
diff --git a/Grid/Grid_Eigen_Dense.h b/Grid/Grid_Eigen_Dense.h
index f9bccf2d..9556c03d 100644
--- a/Grid/Grid_Eigen_Dense.h
+++ b/Grid/Grid_Eigen_Dense.h
@@ -18,12 +18,23 @@
 #pragma push_macro("__CUDA_ARCH__")
 #pragma push_macro("__NVCC__")
 #pragma push_macro("__CUDACC__")
+#undef __CUDA_ARCH__
 #undef __NVCC__
 #undef __CUDACC__
-#undef __CUDA_ARCH__
 #define __NVCC__REDEFINE__
+#endif 
+
+/* SYCL save and restore compile environment*/
+#ifdef GRID_SYCL
+#pragma push
+#pragma push_macro("__SYCL_DEVICE_ONLY__")
+#undef __SYCL_DEVICE_ONLY__
+#define EIGEN_DONT_VECTORIZE
+//#undef EIGEN_USE_SYCL
+#define __SYCL__REDEFINE__
 #endif
 
+
 #include <Grid/Eigen/Dense>
 #include <Grid/Eigen/unsupported/CXX11/Tensor>
 
@@ -31,7 +42,13 @@
 #ifdef __NVCC__REDEFINE__
 #pragma pop_macro("__CUDACC__")
 #pragma pop_macro("__NVCC__")
-#pragma pop_macro("__CUDA_ARCH__")
+#pragma pop_macro("GRID_SIMT")
+#pragma pop
+#endif
+
+/*SYCL restore*/
+#ifdef __SYCL__REDEFINE__
+#pragma pop_macro("__SYCL_DEVICE_ONLY__")
 #pragma pop
 #endif
 
@@ -39,3 +56,4 @@
 #pragma GCC diagnostic pop
 #endif
 
+
diff --git a/Grid/Makefile.am b/Grid/Makefile.am
index b88ea4f2..f1fa462e 100644
--- a/Grid/Makefile.am
+++ b/Grid/Makefile.am
@@ -21,7 +21,7 @@ if BUILD_HDF5
   extra_headers+=serialisation/Hdf5Type.h
 endif
 
-all: version-cache
+all: version-cache Version.h
 
 version-cache:
 	@if [ `git status --porcelain | grep -v '??' | wc -l` -gt 0 ]; then\
@@ -42,7 +42,7 @@ version-cache:
 	fi;\
 	rm -f vertmp
 
-Version.h:
+Version.h: version-cache
 	cp version-cache Version.h
 
 .PHONY: version-cache
diff --git a/Grid/algorithms/Algorithms.h b/Grid/algorithms/Algorithms.h
index 48ea194b..7f27784b 100644
--- a/Grid/algorithms/Algorithms.h
+++ b/Grid/algorithms/Algorithms.h
@@ -29,9 +29,11 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 #ifndef GRID_ALGORITHMS_H
 #define GRID_ALGORITHMS_H
 
+NAMESPACE_CHECK(algorithms);
 #include <Grid/algorithms/SparseMatrix.h>
 #include <Grid/algorithms/LinearOperator.h>
 #include <Grid/algorithms/Preconditioner.h>
+NAMESPACE_CHECK(SparseMatrix);
 
 #include <Grid/algorithms/approx/Zolotarev.h>
 #include <Grid/algorithms/approx/Chebyshev.h>
@@ -41,10 +43,12 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 #include <Grid/algorithms/approx/Forecast.h>
 #include <Grid/algorithms/approx/RemezGeneral.h>
 #include <Grid/algorithms/approx/ZMobius.h>
-
+NAMESPACE_CHECK(approx);
 #include <Grid/algorithms/iterative/Deflation.h>
 #include <Grid/algorithms/iterative/ConjugateGradient.h>
+NAMESPACE_CHECK(ConjGrad);
 #include <Grid/algorithms/iterative/BiCGSTAB.h>
+NAMESPACE_CHECK(BiCGSTAB);
 #include <Grid/algorithms/iterative/ConjugateResidual.h>
 #include <Grid/algorithms/iterative/NormalEquations.h>
 #include <Grid/algorithms/iterative/SchurRedBlack.h>
@@ -62,7 +66,9 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 #include <Grid/algorithms/iterative/ImplicitlyRestartedLanczos.h>
 #include <Grid/algorithms/iterative/PowerMethod.h>
 
+NAMESPACE_CHECK(PowerMethod);
 #include <Grid/algorithms/CoarsenedMatrix.h>
+NAMESPACE_CHECK(CoarsendMatrix);
 #include <Grid/algorithms/FFT.h>
 
 #endif
diff --git a/Grid/algorithms/CoarsenedMatrix.h b/Grid/algorithms/CoarsenedMatrix.h
index a6b01986..8d184aea 100644
--- a/Grid/algorithms/CoarsenedMatrix.h
+++ b/Grid/algorithms/CoarsenedMatrix.h
@@ -1,14 +1,3 @@
-    // blockZaxpy in bockPromote - 3s, 5%
-    // noncoalesced linalg in Preconditionoer ~ 3s 5%
-    // Lancos tuning or replace 10-20s ~ 25%, open ended
-    // setup tuning   5s  ~  8%
-    //    -- e.g. ordermin, orderstep tunables.
-    // MdagM path without norm in LinOp code.     few seconds
-
-    // Mdir calc blocking kernels
-    // Fuse kernels in blockMaskedInnerProduct
-    // preallocate Vectors in Cayley 5D ~ few percent few seconds
-
 /*************************************************************************************
 
     Grid physics library, www.github.com/paboyle/Grid 
@@ -91,34 +80,7 @@ public:
     }
     directions   [2*_d]=0;
     displacements[2*_d]=0;
-      
-    //// report back
-    std::cout<<GridLogMessage<<"directions    :";
-    for(int d=0;d<npoint;d++) std::cout<< directions[d]<< " ";
-    std::cout<<std::endl;
-    std::cout<<GridLogMessage<<"displacements :";
-    for(int d=0;d<npoint;d++) std::cout<< displacements[d]<< " ";
-    std::cout<<std::endl;
   }
-  
-  /*
-  // Original cleaner code
-  Geometry(int _d) : dimension(_d), npoint(2*_d+1), directions(npoint), displacements(npoint) {
-  for(int d=0;d<dimension;d++){
-  directions[2*d  ] = d;
-  directions[2*d+1] = d;
-  displacements[2*d  ] = +1;
-  displacements[2*d+1] = -1;
-  }
-  directions   [2*dimension]=0;
-  displacements[2*dimension]=0;
-  }
-  std::vector<int> GetDelta(int point) {
-  std::vector<int> delta(dimension,0);
-  delta[directions[point]] = displacements[point];
-  return delta;
-  };
-  */    
 
 };
   
@@ -149,25 +111,7 @@ public:
     CoarseScalar InnerProd(CoarseGrid); 
     std::cout << GridLogMessage <<" Block Gramm-Schmidt pass 1"<<std::endl;
     blockOrthogonalise(InnerProd,subspace);
-    //    std::cout << GridLogMessage <<" Block Gramm-Schmidt pass 2"<<std::endl; // Really have to do twice? Yuck
-    //    blockOrthogonalise(InnerProd,subspace);
-    //      std::cout << GridLogMessage <<" Gramm-Schmidt checking orthogonality"<<std::endl;
-    //      CheckOrthogonal();
   } 
-  void CheckOrthogonal(void){
-    CoarseVector iProj(CoarseGrid); 
-    CoarseVector eProj(CoarseGrid); 
-    for(int i=0;i<nbasis;i++){
-      blockProject(iProj,subspace[i],subspace);
-      eProj=Zero(); 
-      accelerator_for(ss, CoarseGrid->oSites(),1,{
-	eProj[ss](i)=CComplex(1.0);
-      });
-      eProj=eProj - iProj;
-      std::cout<<GridLogMessage<<"Orthog check error "<<i<<" " << norm2(eProj)<<std::endl;
-    }
-    std::cout<<GridLogMessage <<"CheckOrthog done"<<std::endl;
-  }
   void ProjectToSubspace(CoarseVector &CoarseVec,const FineField &FineVec){
     blockProject(CoarseVec,FineVec,subspace);
   }
@@ -175,11 +119,6 @@ public:
     FineVec.Checkerboard() = subspace[0].Checkerboard();
     blockPromote(CoarseVec,FineVec,subspace);
   }
-  void CreateSubspaceRandom(GridParallelRNG &RNG){
-    for(int i=0;i<nbasis;i++){
-      random(RNG,subspace[i]);
-    }
-  }
 
   virtual void CreateSubspace(GridParallelRNG  &RNG,LinearOperatorBase<FineField> &hermop,int nn=nbasis) {
 
@@ -190,12 +129,12 @@ public:
     FineField Mn(FineGrid);
 
     for(int b=0;b<nn;b++){
-	
+      
       subspace[b] = Zero();
       gaussian(RNG,noise);
       scale = std::pow(norm2(noise),-0.5); 
       noise=noise*scale;
-	
+      
       hermop.Op(noise,Mn); std::cout<<GridLogMessage << "noise   ["<<b<<"] <n|MdagM|n> "<<norm2(Mn)<<std::endl;
 
       for(int i=0;i<1;i++){
@@ -218,7 +157,7 @@ public:
   // World of possibilities here. But have tried quite a lot of experiments (250+ jobs run on Summit)
   // and this is the best I found
   ////////////////////////////////////////////////////////////////////////////////////////////////
-#if 1
+
   virtual void CreateSubspaceChebyshev(GridParallelRNG  &RNG,LinearOperatorBase<FineField> &hermop,
 				       int nn,
 				       double hi,
@@ -280,10 +219,10 @@ public:
 	
 	hermop.HermOp(*Tn,y);
 
-	auto y_v = y.View();
-	auto Tn_v = Tn->View();
-	auto Tnp_v = Tnp->View();
-	auto Tnm_v = Tnm->View();
+	autoView( y_v , y, AcceleratorWrite);
+	autoView( Tn_v , (*Tn), AcceleratorWrite);
+	autoView( Tnp_v , (*Tnp), AcceleratorWrite);
+	autoView( Tnm_v , (*Tnm), AcceleratorWrite);
 	const int Nsimd = CComplex::Nsimd();
 	accelerator_forNB(ss, FineGrid->oSites(), Nsimd, {
 	  coalescedWrite(y_v[ss],xscale*y_v(ss)+mscale*Tn_v(ss));
@@ -313,201 +252,6 @@ public:
     }
     assert(b==nn);
   }
-#endif
-#if 0
-  virtual void CreateSubspaceChebyshev(GridParallelRNG  &RNG,LinearOperatorBase<FineField> &hermop,
-				       int nn,
-				       double hi,
-				       double lo,
-				       int orderfilter,
-				       int ordermin,
-				       int orderstep,
-				       double filterlo
-				       ) {
-
-    RealD scale;
-
-    FineField noise(FineGrid);
-    FineField Mn(FineGrid);
-    FineField tmp(FineGrid);
-    FineField combined(FineGrid);
-
-    // New normalised noise
-    gaussian(RNG,noise);
-    scale = std::pow(norm2(noise),-0.5); 
-    noise=noise*scale;
-
-    // Initial matrix element
-    hermop.Op(noise,Mn); std::cout<<GridLogMessage << "noise <n|MdagM|n> "<<norm2(Mn)<<std::endl;
-
-    int b =0;
-#define FILTERb(llo,hhi,oorder)						\
-    {									\
-      Chebyshev<FineField> Cheb(llo,hhi,oorder);			\
-      Cheb(hermop,noise,Mn);						\
-      scale = std::pow(norm2(Mn),-0.5); Mn=Mn*scale;			\
-      subspace[b]   = Mn;						\
-      hermop.Op(Mn,tmp);						\
-      std::cout<<GridLogMessage << oorder<< " Cheb filt ["<<b<<"] <n|MdagM|n> "<<norm2(tmp)<<std::endl; \
-      b++;								\
-    }									
-
-    //      JacobiPolynomial<FineField> Cheb(0.002,60.0,1500,-0.5,3.5);	\
-
-    RealD alpha=-0.8;
-    RealD beta =-0.8;
-#define FILTER(llo,hhi,oorder)						\
-    {									\
-      Chebyshev<FineField> Cheb(llo,hhi,oorder);			\
-      /* JacobiPolynomial<FineField> Cheb(0.0,60.0,oorder,alpha,beta);*/\
-      Cheb(hermop,noise,Mn);						\
-      scale = std::pow(norm2(Mn),-0.5); Mn=Mn*scale;			\
-      subspace[b]   = Mn;						\
-      hermop.Op(Mn,tmp);						\
-      std::cout<<GridLogMessage << oorder<< "filt ["<<b<<"] <n|MdagM|n> "<<norm2(tmp)<<std::endl; \
-      b++;								\
-    }									
-    
-#define FILTERc(llo,hhi,oorder)				\
-    {							\
-      Chebyshev<FineField> Cheb(llo,hhi,oorder);	\
-      Cheb(hermop,noise,combined);			\
-    }									
-
-    double node = 0.000;
-    FILTERb(lo,hi,orderfilter);// 0
-    //    FILTERc(node,hi,51);// 0
-    noise = Mn;
-    int base = 0;
-    int mult = 100;
-    FILTER(node,hi,base+1*mult);
-    FILTER(node,hi,base+2*mult);
-    FILTER(node,hi,base+3*mult);
-    FILTER(node,hi,base+4*mult);
-    FILTER(node,hi,base+5*mult);
-    FILTER(node,hi,base+6*mult);
-    FILTER(node,hi,base+7*mult);
-    FILTER(node,hi,base+8*mult);
-    FILTER(node,hi,base+9*mult);
-    FILTER(node,hi,base+10*mult);
-    FILTER(node,hi,base+11*mult);
-    FILTER(node,hi,base+12*mult);
-    FILTER(node,hi,base+13*mult);
-    FILTER(node,hi,base+14*mult);
-    FILTER(node,hi,base+15*mult);
-    assert(b==nn);
-  }
-#endif
-
-#if 0
-  virtual void CreateSubspaceChebyshev(GridParallelRNG  &RNG,LinearOperatorBase<FineField> &hermop,
-				       int nn,
-				       double hi,
-				       double lo,
-				       int orderfilter,
-				       int ordermin,
-				       int orderstep,
-				       double filterlo
-				       ) {
-
-    RealD scale;
-
-    FineField noise(FineGrid);
-    FineField Mn(FineGrid);
-    FineField tmp(FineGrid);
-    FineField combined(FineGrid);
-
-    // New normalised noise
-    gaussian(RNG,noise);
-    scale = std::pow(norm2(noise),-0.5); 
-    noise=noise*scale;
-
-    // Initial matrix element
-    hermop.Op(noise,Mn); std::cout<<GridLogMessage << "noise <n|MdagM|n> "<<norm2(Mn)<<std::endl;
-
-    int b =0;
-    {						
-      Chebyshev<FineField> JacobiPoly(0.005,60.,1500);
-      //      JacobiPolynomial<FineField> JacobiPoly(0.002,60.0,1500,-0.5,3.5);
-      //JacobiPolynomial<FineField> JacobiPoly(0.03,60.0,500,-0.5,3.5);
-      //      JacobiPolynomial<FineField> JacobiPoly(0.00,60.0,1000,-0.5,3.5);
-      JacobiPoly(hermop,noise,Mn);
-      scale = std::pow(norm2(Mn),-0.5); Mn=Mn*scale;
-      subspace[b]   = Mn;
-      hermop.Op(Mn,tmp);
-      std::cout<<GridLogMessage << "filt ["<<b<<"] <n|MdagM|n> "<<norm2(tmp)<<std::endl; 
-      b++;
-      //      scale = std::pow(norm2(tmp),-0.5);     tmp=tmp*scale;
-      //      subspace[b]   = tmp;      b++;
-      //    }									
-    }									
-
-#define FILTER(lambda)						\
-    {								\
-      hermop.HermOp(subspace[0],tmp);				\
-      tmp = tmp - lambda *subspace[0];				\
-      scale = std::pow(norm2(tmp),-0.5);			\
-      tmp=tmp*scale;							\
-      subspace[b]   = tmp;						\
-      hermop.Op(subspace[b],tmp);					\
-      std::cout<<GridLogMessage << "filt ["<<b<<"] <n|MdagM|n> "<<norm2(tmp)<<std::endl; \
-      b++;								\
-    }									
-    //      scale = std::pow(norm2(tmp),-0.5);     tmp=tmp*scale;
-    //      subspace[b]   = tmp;      b++;
-    //    }									
-
-    FILTER(2.0e-5);
-    FILTER(2.0e-4);
-    FILTER(4.0e-4);
-    FILTER(8.0e-4);
-    FILTER(8.0e-4);
-
-    FILTER(2.0e-3);
-    FILTER(3.0e-3);
-    FILTER(4.0e-3);
-    FILTER(5.0e-3);
-    FILTER(6.0e-3);
-
-    FILTER(2.5e-3);
-    FILTER(3.5e-3);
-    FILTER(4.5e-3);
-    FILTER(5.5e-3);
-    FILTER(6.5e-3);
-
-    //    FILTER(6.0e-5);//6
-    //    FILTER(7.0e-5);//8
-    //    FILTER(8.0e-5);//9
-    //    FILTER(9.0e-5);//3
-
-    /*
-    //    FILTER(1.0e-4);//10
-    FILTER(2.0e-4);//11
-    //   FILTER(3.0e-4);//12
-    //    FILTER(4.0e-4);//13
-    FILTER(5.0e-4);//14
-
-    FILTER(6.0e-3);//4
-    FILTER(7.0e-4);//1
-    FILTER(8.0e-4);//7
-    FILTER(9.0e-4);//15
-    FILTER(1.0e-3);//2
-
-    FILTER(2.0e-3);//2
-    FILTER(3.0e-3);//2
-    FILTER(4.0e-3);//2
-    FILTER(5.0e-3);//2
-    FILTER(6.0e-3);//2
-
-    FILTER(7.0e-3);//2
-    FILTER(8.0e-3);//2
-    FILTER(1.0e-2);//2
-    */
-    std::cout << GridLogMessage <<"Jacobi filtering done" <<std::endl;
-    assert(b==nn);
-  }
-#endif
-
 
 };
 
@@ -535,42 +279,34 @@ public:
   CartesianStencil<siteVector,siteVector,int> Stencil; 
 
   std::vector<CoarseMatrix> A;
-      
+    
   ///////////////////////
   // Interface
   ///////////////////////
   GridBase * Grid(void)         { return _grid; };   // this is all the linalg routines need to know
 
-  RealD M (const CoarseVector &in, CoarseVector &out){
-
+  void M (const CoarseVector &in, CoarseVector &out)
+  {
     conformable(_grid,in.Grid());
     conformable(in.Grid(),out.Grid());
 
-    //    RealD Nin = norm2(in);
     SimpleCompressor<siteVector> compressor;
 
-    double comms_usec = -usecond();
     Stencil.HaloExchange(in,compressor);
-    comms_usec += usecond();
-
-    auto in_v = in.View();
-    auto out_v = out.View();
+    autoView( in_v , in, AcceleratorRead);
+    autoView( out_v , out, AcceleratorWrite);
     typedef LatticeView<Cobj> Aview;
-
+      
     Vector<Aview> AcceleratorViewContainer;
-    for(int p=0;p<geom.npoint;p++) AcceleratorViewContainer.push_back(A[p].View());
+  
+    for(int p=0;p<geom.npoint;p++) AcceleratorViewContainer.push_back(A[p].View(AcceleratorRead));
     Aview *Aview_p = & AcceleratorViewContainer[0];
 
     const int Nsimd = CComplex::Nsimd();
     typedef decltype(coalescedRead(in_v[0])) calcVector;
     typedef decltype(coalescedRead(in_v[0](0))) calcComplex;
 
-    GridStopWatch ArithmeticTimer;
     int osites=Grid()->oSites();
-    //    double flops = osites*Nsimd*nbasis*nbasis*8.0*geom.npoint;
-    //    double bytes = osites*nbasis*nbasis*geom.npoint*sizeof(CComplex);
-    double usecs =-usecond();
-    // assert(geom.npoint==9);
 
     accelerator_for(sss, Grid()->oSites()*nbasis, Nsimd, {
       int ss = sss/nbasis;
@@ -580,41 +316,28 @@ public:
       int ptype;
       StencilEntry *SE;
 
-      int lane=SIMTlane(Nsimd);
       for(int point=0;point<geom.npoint;point++){
 
 	SE=Stencil.GetEntry(ptype,point,ss);
 	  
 	if(SE->_is_local) { 
-	  nbr = coalescedReadPermute(in_v[SE->_offset],ptype,SE->_permute,lane);
+	  nbr = coalescedReadPermute(in_v[SE->_offset],ptype,SE->_permute);
 	} else {
-	  nbr = coalescedRead(Stencil.CommBuf()[SE->_offset],lane);
+	  nbr = coalescedRead(Stencil.CommBuf()[SE->_offset]);
 	}
-	synchronise();
+	acceleratorSynchronise();
 
 	for(int bb=0;bb<nbasis;bb++) {
 	  res = res + coalescedRead(Aview_p[point][ss](b,bb))*nbr(bb);
 	}
       }
-      coalescedWrite(out_v[ss](b),res,lane);
-    });
-    usecs +=usecond();
+      coalescedWrite(out_v[ss](b),res);
+      });
 
-    double nrm_usec=-usecond();
-    RealD Nout= norm2(out);
-    nrm_usec+=usecond();
-
-    /*
-        std::cout << GridLogMessage << "\tNorm        " << nrm_usec << " us" <<std::endl;
-        std::cout << GridLogMessage << "\tHalo        " << comms_usec << " us" <<std::endl;
-        std::cout << GridLogMessage << "\tMatrix      " << usecs << " us" <<std::endl;
-        std::cout << GridLogMessage << "\t  mflop/s   " << flops/usecs<<std::endl;
-        std::cout << GridLogMessage << "\t  MB/s      " << bytes/usecs<<std::endl;
-    */
-    return Nout;
+    for(int p=0;p<geom.npoint;p++) AcceleratorViewContainer[p].ViewClose();
   };
 
-  RealD Mdag (const CoarseVector &in, CoarseVector &out)
+  void Mdag (const CoarseVector &in, CoarseVector &out)
   {
     if(hermitian) {
       // corresponds to Petrov-Galerkin coarsening
@@ -625,7 +348,6 @@ public:
       G5C(tmp, in); 
       M(tmp, out);
       G5C(out, out);
-      return norm2(out);
     }
   };
   void MdirComms(const CoarseVector &in)
@@ -640,11 +362,11 @@ public:
 
     typedef LatticeView<Cobj> Aview;
     Vector<Aview> AcceleratorViewContainer;
-    for(int p=0;p<geom.npoint;p++) AcceleratorViewContainer.push_back(A[p].View());
+    for(int p=0;p<geom.npoint;p++) AcceleratorViewContainer.push_back(A[p].View(AcceleratorRead));
     Aview *Aview_p = & AcceleratorViewContainer[0];
 
-    auto out_v = out.View();
-    auto in_v  = in.View();
+    autoView( out_v , out, AcceleratorWrite);
+    autoView( in_v  , in, AcceleratorRead);
 
     const int Nsimd = CComplex::Nsimd();
     typedef decltype(coalescedRead(in_v[0])) calcVector;
@@ -658,45 +380,21 @@ public:
       int ptype;
       StencilEntry *SE;
 
-      int lane=SIMTlane(Nsimd);
       SE=Stencil.GetEntry(ptype,point,ss);
 	  
       if(SE->_is_local) { 
-	nbr = coalescedReadPermute(in_v[SE->_offset],ptype,SE->_permute,lane);
+	nbr = coalescedReadPermute(in_v[SE->_offset],ptype,SE->_permute);
       } else {
-	nbr = coalescedRead(Stencil.CommBuf()[SE->_offset],lane);
+	nbr = coalescedRead(Stencil.CommBuf()[SE->_offset]);
       }
-      synchronise();
+      acceleratorSynchronise();
 
       for(int bb=0;bb<nbasis;bb++) {
 	res = res + coalescedRead(Aview_p[point][ss](b,bb))*nbr(bb);
       }
-      coalescedWrite(out_v[ss](b),res,lane);
+      coalescedWrite(out_v[ss](b),res);
     });
-#if 0
-    accelerator_for(ss,Grid()->oSites(),1,{
-
-      siteVector res = Zero();
-      siteVector nbr;
-      int ptype;
-      StencilEntry *SE;
-      
-      SE=Stencil.GetEntry(ptype,point,ss);
-      
-      if(SE->_is_local&&SE->_permute) {
-	permute(nbr,in_v[SE->_offset],ptype);
-      } else if(SE->_is_local) {
-	nbr = in_v[SE->_offset];
-      } else {
-	nbr = Stencil.CommBuf()[SE->_offset];
-      }
-      synchronise();
-
-      res = res + Aview_p[point][ss]*nbr;
-      
-      out_v[ss]=res;
-    });
-#endif
+    for(int p=0;p<geom.npoint;p++) AcceleratorViewContainer[p].ViewClose();
   }
   void MdirAll(const CoarseVector &in,std::vector<CoarseVector> &out)
   {
@@ -864,14 +562,12 @@ public:
 	    
 	    blockMaskedInnerProduct(oZProj,omask,Subspace.subspace[j],Mphi);
 	    
-	    auto iZProj_v = iZProj.View() ;
-	    auto oZProj_v = oZProj.View() ;
-	    auto A_p     =  A[p].View();
-	    auto A_self  = A[self_stencil].View();
+	    autoView( iZProj_v , iZProj, AcceleratorRead) ;
+	    autoView( oZProj_v , oZProj, AcceleratorRead) ;
+	    autoView( A_p     ,  A[p], AcceleratorWrite);
+	    autoView( A_self  , A[self_stencil], AcceleratorWrite);
 
 	    accelerator_for(ss, Grid()->oSites(), Fobj::Nsimd(),{ coalescedWrite(A_p[ss](j,i),oZProj_v(ss)); });
-	    //      if( disp!= 0 ) { accelerator_for(ss, Grid()->oSites(), Fobj::Nsimd(),{ coalescedWrite(A_p[ss](j,i),oZProj_v(ss)); });}
-	    //	    accelerator_for(ss, Grid()->oSites(), Fobj::Nsimd(),{ coalescedWrite(A_self[ss](j,i),A_self(ss)(j,i)+iZProj_v(ss)); });
 
 	  }
 	}
@@ -885,11 +581,11 @@ public:
 	mult(tmp,phi,oddmask );  linop.Op(tmp,Mphio);
 
 	{
-	  auto tmp_      = tmp.View();
-	  auto evenmask_ = evenmask.View();
-	  auto oddmask_  =  oddmask.View();
-	  auto Mphie_    =  Mphie.View();
-	  auto Mphio_    =  Mphio.View();
+	  autoView( tmp_      , tmp, AcceleratorWrite);
+	  autoView( evenmask_ , evenmask, AcceleratorRead);
+	  autoView( oddmask_  ,  oddmask, AcceleratorRead);
+	  autoView( Mphie_    ,  Mphie, AcceleratorRead);
+	  autoView( Mphio_    ,  Mphio, AcceleratorRead);
 	  accelerator_for(ss, FineGrid->oSites(), Fobj::Nsimd(),{ 
 	      coalescedWrite(tmp_[ss],evenmask_(ss)*Mphie_(ss) + oddmask_(ss)*Mphio_(ss));
 	    });
@@ -897,8 +593,8 @@ public:
 
 	blockProject(SelfProj,tmp,Subspace.subspace);
 
-	auto SelfProj_ = SelfProj.View();
-	auto A_self  = A[self_stencil].View();
+	autoView( SelfProj_ , SelfProj, AcceleratorRead);
+	autoView( A_self  , A[self_stencil], AcceleratorWrite);
 
 	accelerator_for(ss, Grid()->oSites(), Fobj::Nsimd(),{
 	  for(int j=0;j<nbasis;j++){
@@ -912,33 +608,8 @@ public:
       std::cout << GridLogMessage << " ForceHermitian, new code "<<std::endl;
       ForceHermitian();
     }
-      // AssertHermitian();
-      // ForceDiagonal();
   }
 
-#if 0
-    ///////////////////////////
-    // test code worth preserving in if block
-    ///////////////////////////
-    std::cout<<GridLogMessage<< " Computed matrix elements "<< self_stencil <<std::endl;
-    for(int p=0;p<geom.npoint;p++){
-      std::cout<<GridLogMessage<< "A["<<p<<"]" << std::endl;
-      std::cout<<GridLogMessage<< A[p] << std::endl;
-    }
-    std::cout<<GridLogMessage<< " picking by block0 "<< self_stencil <<std::endl;
-
-    phi=Subspace.subspace[0];
-    std::vector<int> bc(FineGrid->_ndimension,0);
-
-    blockPick(Grid(),phi,tmp,bc);      // Pick out a block
-    linop.Op(tmp,Mphi);                // Apply big dop
-    blockProject(iProj,Mphi,Subspace.subspace); // project it and print it
-    std::cout<<GridLogMessage<< " Computed matrix elements from block zero only "<<std::endl;
-    std::cout<<GridLogMessage<< iProj <<std::endl;
-    std::cout<<GridLogMessage<<"Computed Coarse Operator"<<std::endl;
-#endif
-
-
   void ForceHermitian(void) {
     CoarseMatrix Diff  (Grid());
     for(int p=0;p<geom.npoint;p++){
@@ -958,27 +629,6 @@ public:
       }
     }
   }
-  void AssertHermitian(void) {
-    CoarseMatrix AA    (Grid());
-    CoarseMatrix AAc   (Grid());
-    CoarseMatrix Diff  (Grid());
-    for(int d=0;d<4;d++){
-	
-      int dd=d+1;
-      AAc = Cshift(A[2*d+1],dd,1);
-      AA  = A[2*d];
-	
-      Diff = AA - adj(AAc);
-
-      std::cout<<GridLogMessage<<"Norm diff dim "<<d<<" "<< norm2(Diff)<<std::endl;
-      std::cout<<GridLogMessage<<"Norm dim "<<d<<" "<< norm2(AA)<<std::endl;
-	  
-    }
-    Diff = A[8] - adj(A[8]);
-    std::cout<<GridLogMessage<<"Norm diff local "<< norm2(Diff)<<std::endl;
-    std::cout<<GridLogMessage<<"Norm local "<< norm2(A[8])<<std::endl;
-  }
-    
 };
 
 NAMESPACE_END(Grid);
diff --git a/Grid/algorithms/FFT.h b/Grid/algorithms/FFT.h
index 0d568c95..ad42f049 100644
--- a/Grid/algorithms/FFT.h
+++ b/Grid/algorithms/FFT.h
@@ -1,4 +1,3 @@
-
 /*************************************************************************************
 
     Grid physics library, www.github.com/paboyle/Grid 
@@ -37,7 +36,6 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 #endif
 #endif
 
-
 NAMESPACE_BEGIN(Grid);
 
 template<class scalar> struct FFTW { };
@@ -191,7 +189,7 @@ public:
     typedef typename sobj::scalar_type   scalar;
       
     Lattice<sobj> pgbuf(&pencil_g);
-    auto pgbuf_v = pgbuf.View();
+    autoView(pgbuf_v , pgbuf, CpuWrite);
 
     typedef typename FFTW<scalar>::FFTW_scalar FFTW_scalar;
     typedef typename FFTW<scalar>::FFTW_plan   FFTW_plan;
@@ -232,15 +230,18 @@ public:
     result = source;
     int pc = processor_coor[dim];
     for(int p=0;p<processors[dim];p++) {
-      thread_for(idx, sgrid->lSites(),{
+      {
+	autoView(r_v,result,CpuRead);
+	autoView(p_v,pgbuf,CpuWrite);
+	thread_for(idx, sgrid->lSites(),{
           Coordinate cbuf(Nd);
           sobj s;
 	  sgrid->LocalIndexToLocalCoor(idx,cbuf);
-	  peekLocalSite(s,result,cbuf);
+	  peekLocalSite(s,r_v,cbuf);
 	  cbuf[dim]+=((pc+p) % processors[dim])*L;
-	  //            cbuf[dim]+=p*L;
-	  pokeLocalSite(s,pgbuf,cbuf);
-      });
+	  pokeLocalSite(s,p_v,cbuf);
+        });
+      }
       if (p != processors[dim] - 1) {
 	result = Cshift(result,dim,L);
       }
@@ -269,15 +270,19 @@ public:
     flops+= flops_call*NN;
       
     // writing out result
-    thread_for(idx,sgrid->lSites(),{
+    {
+      autoView(pgbuf_v,pgbuf,CpuRead);
+      autoView(result_v,result,CpuWrite);
+      thread_for(idx,sgrid->lSites(),{
 	Coordinate clbuf(Nd), cgbuf(Nd);
 	sobj s;
 	sgrid->LocalIndexToLocalCoor(idx,clbuf);
 	cgbuf = clbuf;
 	cgbuf[dim] = clbuf[dim]+L*pc;
-	peekLocalSite(s,pgbuf,cgbuf);
-	pokeLocalSite(s,result,clbuf);
-    });
+	peekLocalSite(s,pgbuf_v,cgbuf);
+	pokeLocalSite(s,result_v,clbuf);
+      });
+    }
     result = result*div;
       
     // destroying plan
diff --git a/Grid/algorithms/LinearOperator.h b/Grid/algorithms/LinearOperator.h
index 50600d2d..1add212c 100644
--- a/Grid/algorithms/LinearOperator.h
+++ b/Grid/algorithms/LinearOperator.h
@@ -43,7 +43,6 @@ NAMESPACE_BEGIN(Grid);
 /////////////////////////////////////////////////////////////////////////////////////////////
 template<class Field> class LinearOperatorBase {
 public:
-
   // Support for coarsening to a multigrid
   virtual void OpDiag (const Field &in, Field &out) = 0; // Abstract base
   virtual void OpDir  (const Field &in, Field &out,int dir,int disp) = 0; // Abstract base
@@ -94,7 +93,10 @@ public:
     _Mat.Mdag(in,out);
   }
   void HermOpAndNorm(const Field &in, Field &out,RealD &n1,RealD &n2){
-    _Mat.MdagM(in,out,n1,n2);
+    _Mat.MdagM(in,out);
+    ComplexD dot = innerProduct(in,out);
+    n1=real(dot);
+    n2=norm2(out);
   }
   void HermOp(const Field &in, Field &out){
     _Mat.MdagM(in,out);
@@ -131,17 +133,14 @@ public:
     assert(0);
   }
   void HermOpAndNorm(const Field &in, Field &out,RealD &n1,RealD &n2){
-    _Mat.MdagM(in,out,n1,n2);
-    out = out + _shift*in;
-
-    ComplexD dot;	
-    dot= innerProduct(in,out);
+    HermOp(in,out);
+    ComplexD dot = innerProduct(in,out);
     n1=real(dot);
     n2=norm2(out);
   }
   void HermOp(const Field &in, Field &out){
-    RealD n1,n2;
-    HermOpAndNorm(in,out,n1,n2);
+    _Mat.MdagM(in,out);
+    out = out + _shift*in;
   }
 };
 
@@ -170,7 +169,7 @@ public:
     _Mat.M(in,out);
   }
   void HermOpAndNorm(const Field &in, Field &out,RealD &n1,RealD &n2){
-    _Mat.M(in,out);
+    HermOp(in,out);
     ComplexD dot= innerProduct(in,out); n1=real(dot);
     n2=norm2(out);
   }
@@ -208,338 +207,305 @@ public:
   }
 };
 
-    //////////////////////////////////////////////////////////
-    // Even Odd Schur decomp operators; there are several
-    // ways to introduce the even odd checkerboarding
-    //////////////////////////////////////////////////////////
+//////////////////////////////////////////////////////////
+// Even Odd Schur decomp operators; there are several
+// ways to introduce the even odd checkerboarding
+//////////////////////////////////////////////////////////
 
-    template<class Field>
-    class SchurOperatorBase :  public LinearOperatorBase<Field> {
-    public:
-      virtual  RealD Mpc      (const Field &in, Field &out) =0;
-      virtual  RealD MpcDag   (const Field &in, Field &out) =0;
-      virtual void MpcDagMpc(const Field &in, Field &out,RealD &ni,RealD &no) {
-      Field tmp(in.Grid());
-      tmp.Checkerboard() = in.Checkerboard();
-	ni=Mpc(in,tmp);
-	no=MpcDag(tmp,out);
-      }
-      virtual void HermOpAndNorm(const Field &in, Field &out,RealD &n1,RealD &n2){
-      out.Checkerboard() = in.Checkerboard();
-	MpcDagMpc(in,out,n1,n2);
-      }
-      virtual void HermOp(const Field &in, Field &out){
-	RealD n1,n2;
-	HermOpAndNorm(in,out,n1,n2);
-      }
-      void Op     (const Field &in, Field &out){
-	Mpc(in,out);
-      }
-      void AdjOp     (const Field &in, Field &out){ 
-	MpcDag(in,out);
-      }
-      // Support for coarsening to a multigrid
-      void OpDiag (const Field &in, Field &out) {
-	assert(0); // must coarsen the unpreconditioned system
-      }
-      void OpDir  (const Field &in, Field &out,int dir,int disp) {
-	assert(0);
-      }
-      void OpDirAll  (const Field &in, std::vector<Field> &out){
-	assert(0);
-      };
-    };
-    template<class Matrix,class Field>
-    class SchurDiagMooeeOperator :  public SchurOperatorBase<Field> {
-    public:
-      Matrix &_Mat;
-      SchurDiagMooeeOperator (Matrix &Mat): _Mat(Mat){};
-      virtual  RealD Mpc      (const Field &in, Field &out) {
-      Field tmp(in.Grid());
-      tmp.Checkerboard() = !in.Checkerboard();
-	//std::cout <<"grid pointers: in._grid="<< in._grid << " out._grid=" << out._grid << "  _Mat.Grid=" << _Mat.Grid() << " _Mat.RedBlackGrid=" << _Mat.RedBlackGrid() << std::endl;
-
-	_Mat.Meooe(in,tmp);
-	_Mat.MooeeInv(tmp,out);
-	_Mat.Meooe(out,tmp);
-
-      //std::cout << "cb in " << in.Checkerboard() << "  cb out " << out.Checkerboard() << std::endl;
-	_Mat.Mooee(in,out);
-	return axpy_norm(out,-1.0,tmp,out);
-      }
-      virtual  RealD MpcDag   (const Field &in, Field &out){
-	Field tmp(in.Grid());
-
-	_Mat.MeooeDag(in,tmp);
-        _Mat.MooeeInvDag(tmp,out);
-	_Mat.MeooeDag(out,tmp);
-
-	_Mat.MooeeDag(in,out);
-	return axpy_norm(out,-1.0,tmp,out);
-      }
-    };
-    template<class Matrix,class Field>
-      class SchurDiagOneOperator :  public SchurOperatorBase<Field> {
-    protected:
-      Matrix &_Mat;
-    public:
-      SchurDiagOneOperator (Matrix &Mat): _Mat(Mat){};
-
-      virtual  RealD Mpc      (const Field &in, Field &out) {
-	Field tmp(in.Grid());
-
-	_Mat.Meooe(in,out);
-	_Mat.MooeeInv(out,tmp);
-	_Mat.Meooe(tmp,out);
-	_Mat.MooeeInv(out,tmp);
-
-	return axpy_norm(out,-1.0,tmp,in);
-      }
-      virtual  RealD MpcDag   (const Field &in, Field &out){
-	Field tmp(in.Grid());
-
-	_Mat.MooeeInvDag(in,out);
-	_Mat.MeooeDag(out,tmp);
-	_Mat.MooeeInvDag(tmp,out);
-	_Mat.MeooeDag(out,tmp);
-
-	return axpy_norm(out,-1.0,tmp,in);
-      }
-    };
-    template<class Matrix,class Field>
-      class SchurDiagTwoOperator :  public SchurOperatorBase<Field> {
-    protected:
-      Matrix &_Mat;
-    public:
-      SchurDiagTwoOperator (Matrix &Mat): _Mat(Mat){};
-
-      virtual  RealD Mpc      (const Field &in, Field &out) {
-	Field tmp(in.Grid());
-
-	_Mat.MooeeInv(in,out);
-	_Mat.Meooe(out,tmp);
-	_Mat.MooeeInv(tmp,out);
-	_Mat.Meooe(out,tmp);
-
-	return axpy_norm(out,-1.0,tmp,in);
-      }
-      virtual  RealD MpcDag   (const Field &in, Field &out){
-	Field tmp(in.Grid());
-
-	_Mat.MeooeDag(in,out);
-	_Mat.MooeeInvDag(out,tmp);
-	_Mat.MeooeDag(tmp,out);
-	_Mat.MooeeInvDag(out,tmp);
-
-	return axpy_norm(out,-1.0,tmp,in);
-      }
-    };
-
-    template<class Field>
-    class NonHermitianSchurOperatorBase :  public LinearOperatorBase<Field> 
-    {
-      public:
-        virtual RealD Mpc      (const Field& in, Field& out) = 0;
-        virtual RealD MpcDag   (const Field& in, Field& out) = 0;
-        virtual void  MpcDagMpc(const Field& in, Field& out, RealD& ni, RealD& no) {
-          Field tmp(in.Grid());
-          tmp.Checkerboard() = in.Checkerboard();
-	        ni = Mpc(in,tmp);
-	        no = MpcDag(tmp,out);
-        }
-        virtual void HermOpAndNorm(const Field& in, Field& out, RealD& n1, RealD& n2) {
-          assert(0);
-        }
-        virtual void HermOp(const Field& in, Field& out) {
-          assert(0);
-        }
-        void Op(const Field& in, Field& out) {
-          Mpc(in, out);
-        }
-        void AdjOp(const Field& in, Field& out) { 
-          MpcDag(in, out);
-        }
-        // Support for coarsening to a multigrid
-        void OpDiag(const Field& in, Field& out) {
-          assert(0); // must coarsen the unpreconditioned system
-        }
-        void OpDir(const Field& in, Field& out, int dir, int disp) {
-          assert(0);
-        }
-    };
-
-    template<class Matrix, class Field>
-    class NonHermitianSchurDiagMooeeOperator :  public NonHermitianSchurOperatorBase<Field> 
-    {
-      public:
-        Matrix& _Mat;
-        NonHermitianSchurDiagMooeeOperator(Matrix& Mat): _Mat(Mat){};
-        virtual RealD Mpc(const Field& in, Field& out) {
-          Field tmp(in.Grid());
-          tmp.Checkerboard() = !in.Checkerboard();
-
-  	      _Mat.Meooe(in, tmp);
-	        _Mat.MooeeInv(tmp, out);
-	        _Mat.Meooe(out, tmp);
-
-	        _Mat.Mooee(in, out);
-	
-          return axpy_norm(out, -1.0, tmp, out);
-        }
-        virtual RealD MpcDag(const Field& in, Field& out) {
-	        Field tmp(in.Grid());
-
-	        _Mat.MeooeDag(in, tmp);
-          _Mat.MooeeInvDag(tmp, out);
-	        _Mat.MeooeDag(out, tmp);
-
-	        _Mat.MooeeDag(in, out);
-	
-          return axpy_norm(out, -1.0, tmp, out);
-      }
-    };
-    
-    template<class Matrix,class Field>
-    class NonHermitianSchurDiagOneOperator : public NonHermitianSchurOperatorBase<Field> 
-    {
-      protected:
-        Matrix &_Mat;
-    
-      public:
-        NonHermitianSchurDiagOneOperator (Matrix& Mat): _Mat(Mat){};
-        virtual RealD Mpc(const Field& in, Field& out) {
-	        Field tmp(in.Grid());
-
-	        _Mat.Meooe(in, out);
-	        _Mat.MooeeInv(out, tmp);
-	        _Mat.Meooe(tmp, out);
-	        _Mat.MooeeInv(out, tmp);
-
-	        return axpy_norm(out, -1.0, tmp, in);
-        }
-        virtual RealD MpcDag(const Field& in, Field& out) {
-	        Field tmp(in.Grid());
-
-	        _Mat.MooeeInvDag(in, out);
-	        _Mat.MeooeDag(out, tmp);
-	        _Mat.MooeeInvDag(tmp, out);
-	        _Mat.MeooeDag(out, tmp);
-
-	        return axpy_norm(out, -1.0, tmp, in);
-        }
-    };
-
-    template<class Matrix, class Field>
-    class NonHermitianSchurDiagTwoOperator : public NonHermitianSchurOperatorBase<Field> 
-    {
-      protected:
-        Matrix& _Mat;
-    
-      public:
-        NonHermitianSchurDiagTwoOperator(Matrix& Mat): _Mat(Mat){};
-
-        virtual RealD Mpc(const Field& in, Field& out) {
-          Field tmp(in.Grid());
-
-	        _Mat.MooeeInv(in, out);
-	        _Mat.Meooe(out, tmp);
-	        _Mat.MooeeInv(tmp, out);
-	        _Mat.Meooe(out, tmp);
-
-	        return axpy_norm(out, -1.0, tmp, in);
-        }
-        virtual RealD MpcDag(const Field& in, Field& out) {
-	        Field tmp(in.Grid());
-
-          _Mat.MeooeDag(in, out);
-          _Mat.MooeeInvDag(out, tmp);
-          _Mat.MeooeDag(tmp, out);
-          _Mat.MooeeInvDag(out, tmp);
-
-          return axpy_norm(out, -1.0, tmp, in);
-        }
-    };
-
-    ///////////////////////////////////////////////////////////////////////////////////////////////////
-    // Left  handed Moo^-1 ; (Moo - Moe Mee^-1 Meo) psi = eta  -->  ( 1 - Moo^-1 Moe Mee^-1 Meo ) psi = Moo^-1 eta
-    // Right handed Moo^-1 ; (Moo - Moe Mee^-1 Meo) Moo^-1 Moo psi = eta  -->  ( 1 - Moe Mee^-1 Meo Moo^-1) phi=eta ; psi = Moo^-1 phi
-    ///////////////////////////////////////////////////////////////////////////////////////////////////
-    template<class Matrix,class Field> using SchurDiagOneRH = SchurDiagTwoOperator<Matrix,Field> ;
-    template<class Matrix,class Field> using SchurDiagOneLH = SchurDiagOneOperator<Matrix,Field> ;
-    ///////////////////////////////////////////////////////////////////////////////////////////////////
-    //  Staggered use
-    ///////////////////////////////////////////////////////////////////////////////////////////////////
-    template<class Matrix,class Field>
-      class SchurStaggeredOperator :  public SchurOperatorBase<Field> {
-    protected:
-      Matrix &_Mat;
-      Field tmp;
-      RealD mass;
-      double tMpc;
-      double tIP;
-      double tMeo;
-      double taxpby_norm;
-      uint64_t ncall;
-public:
-      void Report(void)
-      {
-	std::cout << GridLogMessage << " HermOpAndNorm.Mpc "<< tMpc/ncall<<" usec "<<std::endl;
-	std::cout << GridLogMessage << " HermOpAndNorm.IP  "<< tIP /ncall<<" usec "<<std::endl;
-	std::cout << GridLogMessage << " Mpc.MeoMoe        "<< tMeo/ncall<<" usec "<<std::endl;
-	std::cout << GridLogMessage << " Mpc.axpby_norm    "<< taxpby_norm/ncall<<" usec "<<std::endl;
-      }
-      SchurStaggeredOperator (Matrix &Mat): _Mat(Mat), tmp(_Mat.RedBlackGrid()) 
-      { 
-	assert( _Mat.isTrivialEE() );
-	mass = _Mat.Mass();
-	tMpc=0;
-	tIP =0;
-        tMeo=0;
-        taxpby_norm=0;
-	ncall=0;
-      }
+template<class Field>
+class SchurOperatorBase :  public LinearOperatorBase<Field> {
+ public:
+  virtual  void Mpc      (const Field &in, Field &out) =0;
+  virtual  void MpcDag   (const Field &in, Field &out) =0;
+  virtual  void MpcDagMpc(const Field &in, Field &out) {
+    Field tmp(in.Grid());
+    tmp.Checkerboard() = in.Checkerboard();
+    Mpc(in,tmp);
+    MpcDag(tmp,out);
+  }
   virtual void HermOpAndNorm(const Field &in, Field &out,RealD &n1,RealD &n2){
-	ncall++;
-	tMpc-=usecond();
-    n2 = Mpc(in,out);
-	tMpc+=usecond();
-	tIP-=usecond();
-    ComplexD dot= innerProduct(in,out);
-	tIP+=usecond();
-    n1 = real(dot);
+    out.Checkerboard() = in.Checkerboard();
+    MpcDagMpc(in,out);
+    ComplexD dot= innerProduct(in,out); 
+    n1=real(dot);
+    n2=norm2(out);
   }
   virtual void HermOp(const Field &in, Field &out){
-	ncall++;
-	tMpc-=usecond();
-	_Mat.Meooe(in,out);
-	_Mat.Meooe(out,tmp);
-	tMpc+=usecond();
-	taxpby_norm-=usecond();
-	axpby(out,-1.0,mass*mass,tmp,in);
-	taxpby_norm+=usecond();
+    out.Checkerboard() = in.Checkerboard();
+    MpcDagMpc(in,out);
   }
-  virtual  RealD Mpc      (const Field &in, Field &out) 
-  {
+  void Op     (const Field &in, Field &out){
+    Mpc(in,out);
+  }
+  void AdjOp     (const Field &in, Field &out){ 
+    MpcDag(in,out);
+  }
+  // Support for coarsening to a multigrid
+  void OpDiag (const Field &in, Field &out) {
+    assert(0); // must coarsen the unpreconditioned system
+  }
+  void OpDir  (const Field &in, Field &out,int dir,int disp) {
+    assert(0);
+  }
+  void OpDirAll  (const Field &in, std::vector<Field> &out){
+    assert(0);
+  };
+};
+template<class Matrix,class Field>
+  class SchurDiagMooeeOperator :  public SchurOperatorBase<Field> {
+ public:
+    Matrix &_Mat;
+    SchurDiagMooeeOperator (Matrix &Mat): _Mat(Mat){};
+    virtual  void Mpc      (const Field &in, Field &out) {
+      Field tmp(in.Grid());
+      tmp.Checkerboard() = !in.Checkerboard();
+      
+      _Mat.Meooe(in,tmp);
+      _Mat.MooeeInv(tmp,out);
+      _Mat.Meooe(out,tmp);
+      _Mat.Mooee(in,out);
+      axpy(out,-1.0,tmp,out);
+    }
+    virtual void MpcDag   (const Field &in, Field &out){
+      Field tmp(in.Grid());
+	
+      _Mat.MeooeDag(in,tmp);
+      _Mat.MooeeInvDag(tmp,out);
+      _Mat.MeooeDag(out,tmp);
+      _Mat.MooeeDag(in,out);
+      axpy(out,-1.0,tmp,out);
+    }
+};
+template<class Matrix,class Field>
+  class SchurDiagOneOperator :  public SchurOperatorBase<Field> {
+ protected:
+    Matrix &_Mat;
+ public:
+    SchurDiagOneOperator (Matrix &Mat): _Mat(Mat){};
+    
+    virtual void Mpc      (const Field &in, Field &out) {
+      Field tmp(in.Grid());
 
+      _Mat.Meooe(in,out);
+      _Mat.MooeeInv(out,tmp);
+      _Mat.Meooe(tmp,out);
+      _Mat.MooeeInv(out,tmp);
+      axpy(out,-1.0,tmp,in);
+    }
+    virtual void MpcDag   (const Field &in, Field &out){
+      Field tmp(in.Grid());
+      
+      _Mat.MooeeInvDag(in,out);
+      _Mat.MeooeDag(out,tmp);
+      _Mat.MooeeInvDag(tmp,out);
+      _Mat.MeooeDag(out,tmp);
+      axpy(out,-1.0,tmp,in);
+    }
+};
+template<class Matrix,class Field>
+  class SchurDiagTwoOperator :  public SchurOperatorBase<Field> {
+ protected:
+    Matrix &_Mat;
+ public:
+    SchurDiagTwoOperator (Matrix &Mat): _Mat(Mat){};
+    
+    virtual void Mpc      (const Field &in, Field &out) {
+      Field tmp(in.Grid());
+      
+      _Mat.MooeeInv(in,out);
+      _Mat.Meooe(out,tmp);
+      _Mat.MooeeInv(tmp,out);
+      _Mat.Meooe(out,tmp);
+      
+      axpy(out,-1.0,tmp,in);
+    }
+    virtual  void MpcDag   (const Field &in, Field &out){
+      Field tmp(in.Grid());
+
+      _Mat.MeooeDag(in,out);
+      _Mat.MooeeInvDag(out,tmp);
+      _Mat.MeooeDag(tmp,out);
+      _Mat.MooeeInvDag(out,tmp);
+
+      axpy(out,-1.0,tmp,in);
+    }
+};
+
+template<class Field>
+class NonHermitianSchurOperatorBase :  public LinearOperatorBase<Field> 
+{
+ public:
+  virtual void  Mpc      (const Field& in, Field& out) = 0;
+  virtual void  MpcDag   (const Field& in, Field& out) = 0;
+  virtual void  MpcDagMpc(const Field& in, Field& out) {
+    Field tmp(in.Grid());
+    tmp.Checkerboard() = in.Checkerboard();
+    Mpc(in,tmp);
+    MpcDag(tmp,out);
+  }
+  virtual void HermOpAndNorm(const Field& in, Field& out, RealD& n1, RealD& n2) {
+    assert(0);
+  }
+  virtual void HermOp(const Field& in, Field& out) {
+    assert(0);
+  }
+  void Op(const Field& in, Field& out) {
+    Mpc(in, out);
+  }
+  void AdjOp(const Field& in, Field& out) { 
+    MpcDag(in, out);
+  }
+  // Support for coarsening to a multigrid
+  void OpDiag(const Field& in, Field& out) {
+    assert(0); // must coarsen the unpreconditioned system
+  }
+  void OpDir(const Field& in, Field& out, int dir, int disp) {
+    assert(0);
+  }
+  void OpDirAll(const Field& in, std::vector<Field>& out){
+    assert(0);
+  };
+};
+
+template<class Matrix, class Field>
+class NonHermitianSchurDiagMooeeOperator :  public NonHermitianSchurOperatorBase<Field> 
+{
+ public:
+  Matrix& _Mat;
+ NonHermitianSchurDiagMooeeOperator(Matrix& Mat): _Mat(Mat){};
+  virtual void Mpc(const Field& in, Field& out) {
+    Field tmp(in.Grid());
+    tmp.Checkerboard() = !in.Checkerboard();
+    
+    _Mat.Meooe(in, tmp);
+    _Mat.MooeeInv(tmp, out);
+    _Mat.Meooe(out, tmp);
+    
+    _Mat.Mooee(in, out);
+    
+    axpy(out, -1.0, tmp, out);
+  }
+  virtual void MpcDag(const Field& in, Field& out) {
+    Field tmp(in.Grid());
+    
+    _Mat.MeooeDag(in, tmp);
+    _Mat.MooeeInvDag(tmp, out);
+    _Mat.MeooeDag(out, tmp);
+	  
+    _Mat.MooeeDag(in, out);
+    
+    axpy(out, -1.0, tmp, out);
+  }
+};
+    
+template<class Matrix,class Field>
+class NonHermitianSchurDiagOneOperator : public NonHermitianSchurOperatorBase<Field> 
+{
+ protected:
+  Matrix &_Mat;
+  
+ public:
+  NonHermitianSchurDiagOneOperator (Matrix& Mat): _Mat(Mat){};
+  virtual void Mpc(const Field& in, Field& out) {
+    Field tmp(in.Grid());
+	  
+    _Mat.Meooe(in, out);
+    _Mat.MooeeInv(out, tmp);
+    _Mat.Meooe(tmp, out);
+    _Mat.MooeeInv(out, tmp);
+
+    axpy(out, -1.0, tmp, in);
+  }
+  virtual void MpcDag(const Field& in, Field& out) {
+    Field tmp(in.Grid());
+    
+    _Mat.MooeeInvDag(in, out);
+    _Mat.MeooeDag(out, tmp);
+    _Mat.MooeeInvDag(tmp, out);
+    _Mat.MeooeDag(out, tmp);
+    
+    axpy(out, -1.0, tmp, in);
+  }
+};
+
+template<class Matrix, class Field>
+class NonHermitianSchurDiagTwoOperator : public NonHermitianSchurOperatorBase<Field> 
+{
+ protected:
+  Matrix& _Mat;
+  
+ public:
+ NonHermitianSchurDiagTwoOperator(Matrix& Mat): _Mat(Mat){};
+
+  virtual void Mpc(const Field& in, Field& out) {
+    Field tmp(in.Grid());
+    
+    _Mat.MooeeInv(in, out);
+    _Mat.Meooe(out, tmp);
+    _Mat.MooeeInv(tmp, out);
+    _Mat.Meooe(out, tmp);
+
+    axpy(out, -1.0, tmp, in);
+  }
+  virtual void MpcDag(const Field& in, Field& out) {
+    Field tmp(in.Grid());
+    
+    _Mat.MeooeDag(in, out);
+    _Mat.MooeeInvDag(out, tmp);
+    _Mat.MeooeDag(tmp, out);
+    _Mat.MooeeInvDag(out, tmp);
+
+    axpy(out, -1.0, tmp, in);
+  }
+};
+
+///////////////////////////////////////////////////////////////////////////////////////////////////
+// Left  handed Moo^-1 ; (Moo - Moe Mee^-1 Meo) psi = eta  -->  ( 1 - Moo^-1 Moe Mee^-1 Meo ) psi = Moo^-1 eta
+// Right handed Moo^-1 ; (Moo - Moe Mee^-1 Meo) Moo^-1 Moo psi = eta  -->  ( 1 - Moe Mee^-1 Meo Moo^-1) phi=eta ; psi = Moo^-1 phi
+///////////////////////////////////////////////////////////////////////////////////////////////////
+template<class Matrix,class Field> using SchurDiagOneRH = SchurDiagTwoOperator<Matrix,Field> ;
+template<class Matrix,class Field> using SchurDiagOneLH = SchurDiagOneOperator<Matrix,Field> ;
+///////////////////////////////////////////////////////////////////////////////////////////////////
+//  Staggered use
+///////////////////////////////////////////////////////////////////////////////////////////////////
+template<class Matrix,class Field>
+class SchurStaggeredOperator :  public SchurOperatorBase<Field> {
+ protected:
+  Matrix &_Mat;
+  Field tmp;
+  RealD mass;
+ public:
+  SchurStaggeredOperator (Matrix &Mat): _Mat(Mat), tmp(_Mat.RedBlackGrid()) 
+  { 
+    assert( _Mat.isTrivialEE() );
+    mass = _Mat.Mass();
+  }
+  virtual void HermOpAndNorm(const Field &in, Field &out,RealD &n1,RealD &n2){
+    Mpc(in,out);
+    ComplexD dot= innerProduct(in,out);
+    n1 = real(dot);
+    n2 =0.0;
+  }
+  virtual void HermOp(const Field &in, Field &out){
+    Mpc(in,out);
+    //    _Mat.Meooe(in,out);
+    //    _Mat.Meooe(out,tmp);
+    //    axpby(out,-1.0,mass*mass,tmp,in);
+  }
+  virtual  void Mpc      (const Field &in, Field &out) 
+  {
     Field tmp(in.Grid());
     Field tmp2(in.Grid());
+	
+    //    _Mat.Mooee(in,out);
+    //    _Mat.Mooee(out,tmp);
 
-    //    std::cout << GridLogIterative << " HermOp.Mpc "<<std::endl;
-    _Mat.Mooee(in,out);
-    _Mat.Mooee(out,tmp);
-    //    std::cout << GridLogIterative << " HermOp.MooeeMooee "<<std::endl;
-
-    tMeo-=usecond();
     _Mat.Meooe(in,out);
     _Mat.Meooe(out,tmp);
-    tMeo+=usecond();
-    taxpby_norm-=usecond();
-    RealD nn=axpby_norm(out,-1.0,mass*mass,tmp,in);
-    taxpby_norm+=usecond();
-    return nn;
+    axpby(out,-1.0,mass*mass,tmp,in);
   }
-  virtual  RealD MpcDag   (const Field &in, Field &out){
-    return Mpc(in,out);
+  virtual  void MpcDag   (const Field &in, Field &out){
+    Mpc(in,out);
   }
   virtual void MpcDagMpc(const Field &in, Field &out,RealD &ni,RealD &no) {
     assert(0);// Never need with staggered
@@ -547,7 +513,6 @@ public:
 };
 template<class Matrix,class Field> using SchurStagOperator = SchurStaggeredOperator<Matrix,Field>;
 
-
 /////////////////////////////////////////////////////////////
 // Base classes for functions of operators
 /////////////////////////////////////////////////////////////
diff --git a/Grid/algorithms/SparseMatrix.h b/Grid/algorithms/SparseMatrix.h
index b959f53c..8a265b3f 100644
--- a/Grid/algorithms/SparseMatrix.h
+++ b/Grid/algorithms/SparseMatrix.h
@@ -38,16 +38,12 @@ template<class Field> class SparseMatrixBase {
 public:
   virtual GridBase *Grid(void) =0;
   // Full checkerboar operations
-  virtual RealD M    (const Field &in, Field &out)=0;
-  virtual RealD Mdag (const Field &in, Field &out)=0;
-  virtual void  MdagM(const Field &in, Field &out,RealD &ni,RealD &no) {
-    Field tmp (in.Grid());
-    ni=M(in,tmp);
-    no=Mdag(tmp,out);
-  }
+  virtual void  M    (const Field &in, Field &out)=0;
+  virtual void  Mdag (const Field &in, Field &out)=0;
   virtual void  MdagM(const Field &in, Field &out) {
-    RealD ni, no;
-    MdagM(in,out,ni,no);
+    Field tmp (in.Grid());
+    M(in,tmp);
+    Mdag(tmp,out);
   }
   virtual  void Mdiag    (const Field &in, Field &out)=0;
   virtual  void Mdir     (const Field &in, Field &out,int dir, int disp)=0;
diff --git a/Grid/algorithms/approx/Chebyshev.h b/Grid/algorithms/approx/Chebyshev.h
index 133db2b4..584ed1d5 100644
--- a/Grid/algorithms/approx/Chebyshev.h
+++ b/Grid/algorithms/approx/Chebyshev.h
@@ -234,10 +234,8 @@ public:
 
     GridBase *grid=in.Grid();
 
-    // std::cout << "Chevyshef(): in.Grid()="<<in.Grid()<<std::endl;
-    //std::cout <<" Linop.Grid()="<<Linop.Grid()<<"Linop.RedBlackGrid()="<<Linop.RedBlackGrid()<<std::endl;
-
     int vol=grid->gSites();
+    typedef typename Field::vector_type vector_type;
 
     Field T0(grid); T0 = in;  
     Field T1(grid); 
@@ -258,14 +256,28 @@ public:
     //    out = ()*T0 + Coeffs[1]*T1;
     axpby(out,0.5*Coeffs[0],Coeffs[1],T0,T1);
     for(int n=2;n<order;n++){
-	
+
       Linop.HermOp(*Tn,y);
-      //     y=xscale*y+mscale*(*Tn);
-      //      *Tnp=2.0*y-(*Tnm);
-      //      out=out+Coeffs[n]* (*Tnp);
+#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));
-      axpy(out,Coeffs[n],*Tnp,out);
+      if ( Coeffs[n] != 0.0) {
+	axpy(out,Coeffs[n],*Tnp,out);
+      }
+#endif
       // Cycle pointers to avoid copies
       Field *swizzle = Tnm;
       Tnm    =Tn;
diff --git a/Grid/algorithms/iterative/BiCGSTAB.h b/Grid/algorithms/iterative/BiCGSTAB.h
index 3a7be1ef..f4e5cdda 100644
--- a/Grid/algorithms/iterative/BiCGSTAB.h
+++ b/Grid/algorithms/iterative/BiCGSTAB.h
@@ -122,12 +122,14 @@ class BiCGSTAB : public OperatorFunction<Field>
 
         LinearCombTimer.Start();
         bo = beta * omega;
-        auto p_v = p.View();
-        auto r_v = r.View();
-        auto v_v = v.View();
-        accelerator_for(ss, p_v.size(), Field::vector_object::Nsimd(),{
-          coalescedWrite(p_v[ss], beta*p_v(ss) - bo*v_v(ss) + r_v(ss));
-        });
+	{
+	  autoView( p_v , p, AcceleratorWrite);
+	  autoView( r_v , r, AcceleratorRead);
+	  autoView( v_v , v, AcceleratorRead);
+	  accelerator_for(ss, p_v.size(), Field::vector_object::Nsimd(),{
+	      coalescedWrite(p_v[ss], beta*p_v(ss) - bo*v_v(ss) + r_v(ss));
+	    });
+	}
         LinearCombTimer.Stop();
         LinalgTimer.Stop();
 
@@ -142,16 +144,20 @@ class BiCGSTAB : public OperatorFunction<Field>
         alpha = rho / Calpha.real();
 
         LinearCombTimer.Start();
-        auto h_v = h.View();
-        auto psi_v = psi.View();
-        accelerator_for(ss, h_v.size(), Field::vector_object::Nsimd(),{
-          coalescedWrite(h_v[ss], alpha*p_v(ss) + psi_v(ss));
-        });
-        
-        auto s_v = s.View();
-        accelerator_for(ss, s_v.size(), Field::vector_object::Nsimd(),{
-          coalescedWrite(s_v[ss], -alpha*v_v(ss) + r_v(ss));
-        });
+	{
+	  autoView( p_v , p, AcceleratorRead);
+	  autoView( r_v , r, AcceleratorRead);
+	  autoView( v_v , v, AcceleratorRead);
+	  autoView( psi_v,psi, AcceleratorRead);
+	  autoView( h_v  ,  h, AcceleratorWrite);
+	  autoView( s_v  ,  s, AcceleratorWrite);
+	  accelerator_for(ss, h_v.size(), Field::vector_object::Nsimd(),{
+	      coalescedWrite(h_v[ss], alpha*p_v(ss) + psi_v(ss));
+	    });
+	  accelerator_for(ss, s_v.size(), Field::vector_object::Nsimd(),{
+	      coalescedWrite(s_v[ss], -alpha*v_v(ss) + r_v(ss));
+ 	  });
+        }
         LinearCombTimer.Stop();
         LinalgTimer.Stop();
 
@@ -166,13 +172,19 @@ class BiCGSTAB : public OperatorFunction<Field>
         omega = Comega.real() / norm2(t);
 
         LinearCombTimer.Start();
-        auto t_v = t.View();
-        accelerator_for(ss, psi_v.size(), Field::vector_object::Nsimd(),{
-          coalescedWrite(psi_v[ss], h_v(ss) + omega * s_v(ss));
-          coalescedWrite(r_v[ss], -omega * t_v(ss) + s_v(ss));
-        });
+	{
+	  autoView( psi_v,psi, AcceleratorWrite);
+	  autoView( r_v , r, AcceleratorWrite);
+	  autoView( h_v , h, AcceleratorRead);
+	  autoView( s_v , s, AcceleratorRead);
+	  autoView( t_v , t, AcceleratorRead);
+	  accelerator_for(ss, psi_v.size(), Field::vector_object::Nsimd(),{
+	      coalescedWrite(psi_v[ss], h_v(ss) + omega * s_v(ss));
+	      coalescedWrite(r_v[ss], -omega * t_v(ss) + s_v(ss));
+	    });
+	}
         LinearCombTimer.Stop();
-
+	
         cp = norm2(r);
         LinalgTimer.Stop();
 
diff --git a/Grid/algorithms/iterative/ConjugateGradient.h b/Grid/algorithms/iterative/ConjugateGradient.h
index 3a2544b5..14f3d306 100644
--- a/Grid/algorithms/iterative/ConjugateGradient.h
+++ b/Grid/algorithms/iterative/ConjugateGradient.h
@@ -140,13 +140,15 @@ public:
       b = cp / c;
 
       LinearCombTimer.Start();
-      auto psi_v = psi.View();
-      auto p_v   = p.View();
-      auto r_v   = r.View();
-      accelerator_for(ss,p_v.size(), Field::vector_object::Nsimd(),{
-	  coalescedWrite(psi_v[ss], a      *  p_v(ss) + psi_v(ss));
-	  coalescedWrite(p_v[ss]  , b      *  p_v(ss) + r_v  (ss));
-      });
+      {
+	autoView( psi_v , psi, AcceleratorWrite);
+	autoView( p_v   , p,   AcceleratorWrite);
+	autoView( r_v   , r,   AcceleratorWrite);
+	accelerator_for(ss,p_v.size(), Field::vector_object::Nsimd(),{
+	    coalescedWrite(psi_v[ss], a      *  p_v(ss) + psi_v(ss));
+	    coalescedWrite(p_v[ss]  , b      *  p_v(ss) + r_v  (ss));
+	});
+      }
       LinearCombTimer.Stop();
       LinalgTimer.Stop();
 
diff --git a/Grid/algorithms/iterative/ImplicitlyRestartedLanczos.h b/Grid/algorithms/iterative/ImplicitlyRestartedLanczos.h
index 8bee43cc..3d0a2a75 100644
--- a/Grid/algorithms/iterative/ImplicitlyRestartedLanczos.h
+++ b/Grid/algorithms/iterative/ImplicitlyRestartedLanczos.h
@@ -37,211 +37,6 @@ Author: Christoph Lehner <clehner@bnl.gov>
 
 NAMESPACE_BEGIN(Grid); 
 
-  ////////////////////////////////////////////////////////
-  // Move following 100 LOC to lattice/Lattice_basis.h
-  ////////////////////////////////////////////////////////
-template<class Field>
-void basisOrthogonalize(std::vector<Field> &basis,Field &w,int k) 
-{
-  // If assume basis[j] are already orthonormal,
-  // can take all inner products in parallel saving 2x bandwidth
-  // Save 3x bandwidth on the second line of loop.
-  // perhaps 2.5x speed up.
-  // 2x overall in Multigrid Lanczos  
-  for(int j=0; j<k; ++j){
-    auto ip = innerProduct(basis[j],w);
-    w = w - ip*basis[j];
-  }
-}
-
-template<class Field>
-void basisRotate(std::vector<Field> &basis,Eigen::MatrixXd& Qt,int j0, int j1, int k0,int k1,int Nm) 
-{
-  typedef decltype(basis[0].View()) View;
-  auto tmp_v = basis[0].View();
-  Vector<View> basis_v(basis.size(),tmp_v);
-  typedef typename Field::vector_object vobj;
-  GridBase* grid = basis[0].Grid();
-
-  for(int k=0;k<basis.size();k++){
-    basis_v[k] = basis[k].View();
-  }
-#if 0
-  std::vector < vobj , commAllocator<vobj> > Bt(thread_max() * Nm); // Thread private
-  thread_region
-  {
-    vobj* B = Bt.data() + Nm * thread_num();
-
-    thread_for_in_region(ss, grid->oSites(),{
-      for(int j=j0; j<j1; ++j) B[j]=0.;
-      
-      for(int j=j0; j<j1; ++j){
-	for(int k=k0; k<k1; ++k){
-	  B[j] +=Qt(j,k) * basis_v[k][ss];
-	}
-      }
-      for(int j=j0; j<j1; ++j){
-	basis_v[j][ss] = B[j];
-      }
-    });
-  }
-#else
-
-  int nrot = j1-j0;
-
-
-  uint64_t oSites   =grid->oSites();
-  uint64_t siteBlock=(grid->oSites()+nrot-1)/nrot; // Maximum 1 additional vector overhead
-
-  //  printf("BasisRotate %d %d nrot %d siteBlock %d\n",j0,j1,nrot,siteBlock);
-
-  Vector <vobj> Bt(siteBlock * nrot); 
-  auto Bp=&Bt[0];
-
-  // GPU readable copy of Eigen matrix
-  Vector<double> Qt_jv(Nm*Nm);
-  double *Qt_p = & Qt_jv[0];
-  for(int k=0;k<Nm;++k){
-    for(int j=0;j<Nm;++j){
-      Qt_p[j*Nm+k]=Qt(j,k);
-    }
-  }
-
-  // Block the loop to keep storage footprint down
-  vobj zz=Zero();
-  for(uint64_t s=0;s<oSites;s+=siteBlock){
-
-    // remaining work in this block
-    int ssites=MIN(siteBlock,oSites-s);
-
-    // zero out the accumulators
-    accelerator_for(ss,siteBlock*nrot,vobj::Nsimd(),{
-	auto z=coalescedRead(zz);
-	coalescedWrite(Bp[ss],z);
-    });
-
-    accelerator_for(sj,ssites*nrot,vobj::Nsimd(),{
-	
-      int j =sj%nrot;
-      int jj  =j0+j;
-      int ss =sj/nrot;
-      int sss=ss+s;
-
-      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_v[k][sss]));
-      }
-    });
-
-    accelerator_for(sj,ssites*nrot,vobj::Nsimd(),{
-      int j =sj%nrot;
-      int jj  =j0+j;
-      int ss =sj/nrot;
-      int sss=ss+s;
-      coalescedWrite(basis_v[jj][sss],coalescedRead(Bp[ss*nrot+j]));
-    });
-  }
-#endif
-}
-
-// Extract a single rotated vector
-template<class Field>
-void basisRotateJ(Field &result,std::vector<Field> &basis,Eigen::MatrixXd& Qt,int j, int k0,int k1,int Nm) 
-{
-  typedef decltype(basis[0].View()) View;
-  typedef typename Field::vector_object vobj;
-  GridBase* grid = basis[0].Grid();
-
-  result.Checkerboard() = basis[0].Checkerboard();
-  auto result_v=result.View();
-  Vector<View> basis_v(basis.size(),result_v);
-  for(int k=0;k<basis.size();k++){
-    basis_v[k] = basis[k].View();
-  }
-  vobj zz=Zero();
-  Vector<double> Qt_jv(Nm);
-  double * Qt_j = & Qt_jv[0];
-  for(int k=0;k<Nm;++k) Qt_j[k]=Qt(j,k);
-  accelerator_for(ss, grid->oSites(),vobj::Nsimd(),{
-    auto B=coalescedRead(zz);
-    for(int k=k0; k<k1; ++k){
-      B +=Qt_j[k] * coalescedRead(basis_v[k][ss]);
-    }
-    coalescedWrite(result_v[ss], B);
-  });
-}
-
-template<class Field>
-void basisReorderInPlace(std::vector<Field> &_v,std::vector<RealD>& sort_vals, std::vector<int>& idx) 
-{
-  int vlen = idx.size();
-
-  assert(vlen>=1);
-  assert(vlen<=sort_vals.size());
-  assert(vlen<=_v.size());
-
-  for (size_t i=0;i<vlen;i++) {
-
-    if (idx[i] != i) {
-
-      //////////////////////////////////////
-      // idx[i] is a table of desired sources giving a permutation.
-      // Swap v[i] with v[idx[i]].
-      // Find  j>i for which _vnew[j] = _vold[i],
-      // track the move idx[j] => idx[i]
-      // track the move idx[i] => i
-      //////////////////////////////////////
-      size_t j;
-      for (j=i;j<idx.size();j++)
-	if (idx[j]==i)
-	  break;
-
-      assert(idx[i] > i);     assert(j!=idx.size());      assert(idx[j]==i);
-
-      swap(_v[i],_v[idx[i]]); // should use vector move constructor, no data copy
-      std::swap(sort_vals[i],sort_vals[idx[i]]);
-
-      idx[j] = idx[i];
-      idx[i] = i;
-    }
-  }
-}
-
-inline std::vector<int> basisSortGetIndex(std::vector<RealD>& sort_vals) 
-{
-  std::vector<int> idx(sort_vals.size());
-  std::iota(idx.begin(), idx.end(), 0);
-
-  // sort indexes based on comparing values in v
-  std::sort(idx.begin(), idx.end(), [&sort_vals](int i1, int i2) {
-    return ::fabs(sort_vals[i1]) < ::fabs(sort_vals[i2]);
-  });
-  return idx;
-}
-
-template<class Field>
-void basisSortInPlace(std::vector<Field> & _v,std::vector<RealD>& sort_vals, bool reverse) 
-{
-  std::vector<int> idx = basisSortGetIndex(sort_vals);
-  if (reverse)
-    std::reverse(idx.begin(), idx.end());
-  
-  basisReorderInPlace(_v,sort_vals,idx);
-}
-
-// PAB: faster to compute the inner products first then fuse loops.
-// If performance critical can improve.
-template<class Field>
-void basisDeflate(const std::vector<Field> &_v,const std::vector<RealD>& eval,const Field& src_orig,Field& result) {
-  result = Zero();
-  assert(_v.size()==eval.size());
-  int N = (int)_v.size();
-  for (int i=0;i<N;i++) {
-    Field& tmp = _v[i];
-    axpy(result,TensorRemove(innerProduct(tmp,src_orig)) / eval[i],tmp,result);
-  }
-}
-
 /////////////////////////////////////////////////////////////
 // Implicitly restarted lanczos
 /////////////////////////////////////////////////////////////
diff --git a/Grid/algorithms/iterative/PrecGeneralisedConjugateResidualNonHermitian.h b/Grid/algorithms/iterative/PrecGeneralisedConjugateResidualNonHermitian.h
new file mode 100644
index 00000000..22b7725e
--- /dev/null
+++ b/Grid/algorithms/iterative/PrecGeneralisedConjugateResidualNonHermitian.h
@@ -0,0 +1,241 @@
+/*************************************************************************************
+
+    Grid physics library, www.github.com/paboyle/Grid 
+
+    Source file: ./lib/algorithms/iterative/PrecGeneralisedConjugateResidual.h
+
+    Copyright (C) 2015
+
+Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
+Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License along
+    with this program; if not, write to the Free Software Foundation, Inc.,
+    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+    See the full license in the file "LICENSE" in the top level distribution directory
+*************************************************************************************/
+/*  END LEGAL */
+#ifndef GRID_PREC_GCR_NON_HERM_H
+#define GRID_PREC_GCR_NON_HERM_H
+
+///////////////////////////////////////////////////////////////////////////////////////////////////////
+//VPGCR Abe and Zhang, 2005.
+//INTERNATIONAL JOURNAL OF NUMERICAL ANALYSIS AND MODELING
+//Computing and Information Volume 2, Number 2, Pages 147-161
+//NB. Likely not original reference since they are focussing on a preconditioner variant.
+//    but VPGCR was nicely written up in their paper
+///////////////////////////////////////////////////////////////////////////////////////////////////////
+NAMESPACE_BEGIN(Grid);
+
+#define GCRLogLevel std::cout << GridLogMessage <<std::string(level,'\t')<< " Level "<<level<<" " 
+
+template<class Field>
+class PrecGeneralisedConjugateResidualNonHermitian : public LinearFunction<Field> {
+public:                                                
+
+  RealD   Tolerance;
+  Integer MaxIterations;
+  int verbose;
+  int mmax;
+  int nstep;
+  int steps;
+  int level;
+  GridStopWatch PrecTimer;
+  GridStopWatch MatTimer;
+  GridStopWatch LinalgTimer;
+
+  LinearFunction<Field>     &Preconditioner;
+  LinearOperatorBase<Field> &Linop;
+
+  void Level(int lv) { level=lv; };
+
+  PrecGeneralisedConjugateResidualNonHermitian(RealD tol,Integer maxit,LinearOperatorBase<Field> &_Linop,LinearFunction<Field> &Prec,int _mmax,int _nstep) : 
+    Tolerance(tol), 
+    MaxIterations(maxit),
+    Linop(_Linop),
+    Preconditioner(Prec),
+    mmax(_mmax),
+    nstep(_nstep)
+  { 
+    level=1;
+    verbose=1;
+  };
+
+  void operator() (const Field &src, Field &psi){
+
+    psi=Zero();
+    RealD cp, ssq,rsq;
+    ssq=norm2(src);
+    rsq=Tolerance*Tolerance*ssq;
+      
+    Field r(src.Grid());
+
+    PrecTimer.Reset();
+    MatTimer.Reset();
+    LinalgTimer.Reset();
+
+    GridStopWatch SolverTimer;
+    SolverTimer.Start();
+
+    steps=0;
+    for(int k=0;k<MaxIterations;k++){
+
+      cp=GCRnStep(src,psi,rsq);
+
+      GCRLogLevel <<"PGCR("<<mmax<<","<<nstep<<") "<< steps <<" steps cp = "<<cp<<" target "<<rsq <<std::endl;
+
+      if(cp<rsq) {
+
+	SolverTimer.Stop();
+
+	Linop.Op(psi,r);
+	axpy(r,-1.0,src,r);
+	RealD tr = norm2(r);
+	GCRLogLevel<<"PGCR: Converged on iteration " <<steps
+		 << " computed residual "<<sqrt(cp/ssq)
+		 << " true residual "    <<sqrt(tr/ssq)
+		 << " target "           <<Tolerance <<std::endl;
+
+	GCRLogLevel<<"PGCR Time elapsed: Total  "<< SolverTimer.Elapsed() <<std::endl;
+	return;
+      }
+
+    }
+    GCRLogLevel<<"Variable Preconditioned GCR did not converge"<<std::endl;
+    //    assert(0);
+  }
+
+  RealD GCRnStep(const Field &src, Field &psi,RealD rsq){
+
+    RealD cp;
+    ComplexD a, b, zAz;
+    RealD zAAz;
+    ComplexD rq;
+
+    GridBase *grid = src.Grid();
+
+    Field r(grid);
+    Field z(grid);
+    Field tmp(grid);
+    Field ttmp(grid);
+    Field Az(grid);
+
+    ////////////////////////////////
+    // history for flexible orthog
+    ////////////////////////////////
+    std::vector<Field> q(mmax,grid);
+    std::vector<Field> p(mmax,grid);
+    std::vector<RealD> qq(mmax);
+      
+    GCRLogLevel<< "PGCR nStep("<<nstep<<")"<<std::endl;
+
+    //////////////////////////////////
+    // initial guess x0 is taken as nonzero.
+    // r0=src-A x0 = src
+    //////////////////////////////////
+    MatTimer.Start();
+    Linop.Op(psi,Az);
+    zAz = innerProduct(Az,psi);
+    zAAz= norm2(Az);
+    MatTimer.Stop();
+    
+
+    LinalgTimer.Start();
+    r=src-Az;
+    LinalgTimer.Stop();
+    GCRLogLevel<< "PGCR true residual r = src - A psi   "<<norm2(r) <<std::endl;
+    
+    /////////////////////
+    // p = Prec(r)
+    /////////////////////
+
+    PrecTimer.Start();
+    Preconditioner(r,z);
+    PrecTimer.Stop();
+
+    MatTimer.Start();
+    Linop.Op(z,Az);
+    MatTimer.Stop();
+
+    LinalgTimer.Start();
+
+    zAz = innerProduct(Az,psi);
+    zAAz= norm2(Az);
+
+    //p[0],q[0],qq[0] 
+    p[0]= z;
+    q[0]= Az;
+    qq[0]= zAAz;
+    
+    cp =norm2(r);
+    LinalgTimer.Stop();
+
+    for(int k=0;k<nstep;k++){
+
+      steps++;
+
+      int kp     = k+1;
+      int peri_k = k %mmax;
+      int peri_kp= kp%mmax;
+
+      LinalgTimer.Start();
+      rq= innerProduct(q[peri_k],r); // what if rAr not real?
+      a = rq/qq[peri_k];
+
+      axpy(psi,a,p[peri_k],psi);         
+
+      cp = axpy_norm(r,-a,q[peri_k],r);
+      LinalgTimer.Stop();
+
+      GCRLogLevel<< "PGCR step["<<steps<<"]  resid " << cp << " target " <<rsq<<std::endl; 
+
+      if((k==nstep-1)||(cp<rsq)){
+	return cp;
+      }
+
+
+      PrecTimer.Start();
+      Preconditioner(r,z);// solve Az = r
+      PrecTimer.Stop();
+
+      MatTimer.Start();
+      Linop.Op(z,Az);
+      MatTimer.Stop();
+      zAz = innerProduct(Az,psi);
+      zAAz= norm2(Az);
+
+      LinalgTimer.Start();
+
+      q[peri_kp]=Az;
+      p[peri_kp]=z;
+
+      int northog = ((kp)>(mmax-1))?(mmax-1):(kp);  // if more than mmax done, we orthog all mmax history.
+      for(int back=0;back<northog;back++){
+
+	int peri_back=(k-back)%mmax;   	  assert((k-back)>=0);
+
+	b=-real(innerProduct(q[peri_back],Az))/qq[peri_back];
+	p[peri_kp]=p[peri_kp]+b*p[peri_back];
+	q[peri_kp]=q[peri_kp]+b*q[peri_back];
+
+      }
+      qq[peri_kp]=norm2(q[peri_kp]); // could use axpy_norm
+      LinalgTimer.Stop();
+    }
+    assert(0); // never reached
+    return cp;
+  }
+};
+NAMESPACE_END(Grid);
+#endif
diff --git a/Grid/allocator/AlignedAllocator.cc b/Grid/allocator/AlignedAllocator.cc
index d53c4dc2..0d1707d9 100644
--- a/Grid/allocator/AlignedAllocator.cc
+++ b/Grid/allocator/AlignedAllocator.cc
@@ -6,72 +6,6 @@ NAMESPACE_BEGIN(Grid);
 MemoryStats *MemoryProfiler::stats = nullptr;
 bool         MemoryProfiler::debug = false;
 
-#ifdef GRID_NVCC
-#define SMALL_LIMIT (0)
-#else
-#define SMALL_LIMIT (4096)
-#endif
-
-#ifdef POINTER_CACHE
-int PointerCache::victim;
-
-PointerCache::PointerCacheEntry PointerCache::Entries[PointerCache::Ncache];
-
-void *PointerCache::Insert(void *ptr,size_t bytes) {
-
-  if (bytes < SMALL_LIMIT ) return ptr;
-
-#ifdef GRID_OMP
-  assert(omp_in_parallel()==0);
-#endif 
-
-  void * ret = NULL;
-  int v = -1;
-
-  for(int e=0;e<Ncache;e++) {
-    if ( Entries[e].valid==0 ) {
-      v=e; 
-      break;
-    }
-  }
-
-  if ( v==-1 ) {
-    v=victim;
-    victim = (victim+1)%Ncache;
-  }
-
-  if ( Entries[v].valid ) {
-    ret = Entries[v].address;
-    Entries[v].valid = 0;
-    Entries[v].address = NULL;
-    Entries[v].bytes = 0;
-  }
-
-  Entries[v].address=ptr;
-  Entries[v].bytes  =bytes;
-  Entries[v].valid  =1;
-
-  return ret;
-}
-
-void *PointerCache::Lookup(size_t bytes) {
-
-  if (bytes < SMALL_LIMIT ) return NULL;
-
-#ifdef GRID_OMP
-  assert(omp_in_parallel()==0);
-#endif 
-
-  for(int e=0;e<Ncache;e++){
-    if ( Entries[e].valid && ( Entries[e].bytes == bytes ) ) {
-      Entries[e].valid = 0;
-      return Entries[e].address;
-    }
-  }
-  return NULL;
-}
-#endif
-
 void check_huge_pages(void *Buf,uint64_t BYTES)
 {
 #ifdef __linux__
diff --git a/Grid/allocator/AlignedAllocator.h b/Grid/allocator/AlignedAllocator.h
index 8c189be8..ebb3162b 100644
--- a/Grid/allocator/AlignedAllocator.h
+++ b/Grid/allocator/AlignedAllocator.h
@@ -26,118 +26,10 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
     See the full license in the file "LICENSE" in the top level distribution directory
 *************************************************************************************/
 /*  END LEGAL */
-#ifndef GRID_ALIGNED_ALLOCATOR_H
-#define GRID_ALIGNED_ALLOCATOR_H
-
-#ifdef HAVE_MALLOC_MALLOC_H
-#include <malloc/malloc.h>
-#endif
-#ifdef HAVE_MALLOC_H
-#include <malloc.h>
-#endif
-
-#ifdef HAVE_MM_MALLOC_H
-#include <mm_malloc.h>
-#endif
-
-#define POINTER_CACHE
-#define GRID_ALLOC_ALIGN (2*1024*1024)
+#pragma once
 
 NAMESPACE_BEGIN(Grid);
 
-// Move control to configure.ac and Config.h?
-#ifdef POINTER_CACHE
-class PointerCache {
-private:
-/*Pinning pages is costly*/
-/*Could maintain separate large and small allocation caches*/
-#ifdef GRID_NVCC 
-  static const int Ncache=128;
-#else
-  static const int Ncache=8;
-#endif
-  static int victim;
-
-  typedef struct { 
-    void *address;
-    size_t bytes;
-    int valid;
-  } PointerCacheEntry;
-    
-  static PointerCacheEntry Entries[Ncache];
-
-public:
-
-  static void *Insert(void *ptr,size_t bytes) ;
-  static void *Lookup(size_t bytes) ;
-
-};
-#endif  
-
-std::string sizeString(size_t bytes);
-
-struct MemoryStats
-{
-  size_t totalAllocated{0}, maxAllocated{0}, 
-    currentlyAllocated{0}, totalFreed{0};
-};
-    
-class MemoryProfiler
-{
-public:
-  static MemoryStats *stats;
-  static bool        debug;
-};
-
-#define memString(bytes) std::to_string(bytes) + " (" + sizeString(bytes) + ")"
-#define profilerDebugPrint						\
-  if (MemoryProfiler::stats)						\
-    {									\
-      auto s = MemoryProfiler::stats;					\
-      std::cout << GridLogDebug << "[Memory debug] Stats " << MemoryProfiler::stats << std::endl; \
-      std::cout << GridLogDebug << "[Memory debug] total  : " << memString(s->totalAllocated) \
-		<< std::endl;						\
-      std::cout << GridLogDebug << "[Memory debug] max    : " << memString(s->maxAllocated) \
-		<< std::endl;						\
-      std::cout << GridLogDebug << "[Memory debug] current: " << memString(s->currentlyAllocated) \
-		<< std::endl;						\
-      std::cout << GridLogDebug << "[Memory debug] freed  : " << memString(s->totalFreed) \
-		<< std::endl;						\
-    }
-
-#define profilerAllocate(bytes)						\
-  if (MemoryProfiler::stats)						\
-    {									\
-      auto s = MemoryProfiler::stats;					\
-      s->totalAllocated     += (bytes);					\
-      s->currentlyAllocated += (bytes);					\
-      s->maxAllocated        = std::max(s->maxAllocated, s->currentlyAllocated); \
-    }									\
-  if (MemoryProfiler::debug)						\
-    {									\
-      std::cout << GridLogDebug << "[Memory debug] allocating " << memString(bytes) << std::endl; \
-      profilerDebugPrint;						\
-    }
-
-#define profilerFree(bytes)						\
-  if (MemoryProfiler::stats)						\
-    {									\
-      auto s = MemoryProfiler::stats;					\
-      s->totalFreed         += (bytes);					\
-      s->currentlyAllocated -= (bytes);					\
-    }									\
-  if (MemoryProfiler::debug)						\
-    {									\
-      std::cout << GridLogDebug << "[Memory debug] freeing " << memString(bytes) << std::endl; \
-      profilerDebugPrint;						\
-    }
-
-void check_huge_pages(void *Buf,uint64_t BYTES);
-
-////////////////////////////////////////////////////////////////////
-// A lattice of something, but assume the something is SIMDized.
-////////////////////////////////////////////////////////////////////
-
 template<typename _Tp>
 class alignedAllocator {
 public: 
@@ -161,70 +53,60 @@ public:
   { 
     size_type bytes = __n*sizeof(_Tp);
     profilerAllocate(bytes);
-
-
-#ifdef POINTER_CACHE
-    _Tp *ptr = (_Tp *) PointerCache::Lookup(bytes);
-#else
-    pointer ptr = nullptr;
-#endif
-
-#ifdef GRID_NVCC
-    ////////////////////////////////////
-    // Unified (managed) memory
-    ////////////////////////////////////
-    if ( ptr == (_Tp *) NULL ) {
-      //      printf(" alignedAllocater cache miss %ld bytes ",bytes);      BACKTRACEFP(stdout);
-      auto err = cudaMallocManaged((void **)&ptr,bytes);
-      if( err != cudaSuccess ) {
-	ptr = (_Tp *) NULL;
-	std::cerr << " cudaMallocManaged failed for " << bytes<<" bytes " <<cudaGetErrorString(err)<< std::endl;
-	assert(0);
-      }
-    } 
-    assert( ptr != (_Tp *)NULL);
-#else 
-    //////////////////////////////////////////////////////////////////////////////////////////
-    // 2MB align; could make option probably doesn't need configurability
-    //////////////////////////////////////////////////////////////////////////////////////////
-  #ifdef HAVE_MM_MALLOC_H
-    if ( ptr == (_Tp *) NULL ) ptr = (_Tp *) _mm_malloc(bytes,GRID_ALLOC_ALIGN);
-  #else
-    if ( ptr == (_Tp *) NULL ) ptr = (_Tp *) memalign(GRID_ALLOC_ALIGN,bytes);
-  #endif
-    assert( ptr != (_Tp *)NULL);
-
-    //////////////////////////////////////////////////
-    // First touch optimise in threaded loop 
-    //////////////////////////////////////////////////
-    uint64_t *cp = (uint64_t *)ptr;
-    thread_for(n,bytes/sizeof(uint64_t), { // need only one touch per page
-      cp[n]=0;
-    });
-#endif
+    _Tp *ptr = (_Tp*) MemoryManager::CpuAllocate(bytes);
+    assert( ( (_Tp*)ptr != (_Tp *)NULL ) );
     return ptr;
   }
 
-  void deallocate(pointer __p, size_type __n) { 
+  void deallocate(pointer __p, size_type __n) 
+  { 
     size_type bytes = __n * sizeof(_Tp);
-
     profilerFree(bytes);
+    MemoryManager::CpuFree((void *)__p,bytes);
+  }
 
-#ifdef POINTER_CACHE
-    pointer __freeme = (pointer)PointerCache::Insert((void *)__p,bytes);
-#else 
-    pointer __freeme = __p;
-#endif
+  // FIXME: hack for the copy constructor, eventually it must be avoided
+  //void construct(pointer __p, const _Tp& __val) { new((void *)__p) _Tp(__val); };
+  void construct(pointer __p, const _Tp& __val) { assert(0);};
+  void construct(pointer __p) { };
+  void destroy(pointer __p) { };
+};
+template<typename _Tp>  inline bool operator==(const alignedAllocator<_Tp>&, const alignedAllocator<_Tp>&){ return true; }
+template<typename _Tp>  inline bool operator!=(const alignedAllocator<_Tp>&, const alignedAllocator<_Tp>&){ return false; }
 
-#ifdef GRID_NVCC
-    if ( __freeme ) cudaFree((void *)__freeme);
-#else 
-  #ifdef HAVE_MM_MALLOC_H
-    if ( __freeme ) _mm_free((void *)__freeme); 
-  #else
-    if ( __freeme ) free((void *)__freeme);
-  #endif
-#endif
+template<typename _Tp>
+class uvmAllocator {
+public: 
+  typedef std::size_t     size_type;
+  typedef std::ptrdiff_t  difference_type;
+  typedef _Tp*       pointer;
+  typedef const _Tp* const_pointer;
+  typedef _Tp&       reference;
+  typedef const _Tp& const_reference;
+  typedef _Tp        value_type;
+
+  template<typename _Tp1>  struct rebind { typedef uvmAllocator<_Tp1> other; };
+  uvmAllocator() throw() { }
+  uvmAllocator(const uvmAllocator&) throw() { }
+  template<typename _Tp1> uvmAllocator(const uvmAllocator<_Tp1>&) throw() { }
+  ~uvmAllocator() throw() { }
+  pointer       address(reference __x)       const { return &__x; }
+  size_type  max_size() const throw() { return size_t(-1) / sizeof(_Tp); }
+
+  pointer allocate(size_type __n, const void* _p= 0)
+  { 
+    size_type bytes = __n*sizeof(_Tp);
+    profilerAllocate(bytes);
+    _Tp *ptr = (_Tp*) MemoryManager::SharedAllocate(bytes);
+    assert( ( (_Tp*)ptr != (_Tp *)NULL ) );
+    return ptr;
+  }
+
+  void deallocate(pointer __p, size_type __n) 
+  { 
+    size_type bytes = __n * sizeof(_Tp);
+    profilerFree(bytes);
+    MemoryManager::SharedFree((void *)__p,bytes);
   }
 
   // FIXME: hack for the copy constructor, eventually it must be avoided
@@ -233,17 +115,17 @@ public:
   void construct(pointer __p) { };
   void destroy(pointer __p) { };
 };
-template<typename _Tp>  inline bool operator==(const alignedAllocator<_Tp>&, const alignedAllocator<_Tp>&){ return true; }
-template<typename _Tp>  inline bool operator!=(const alignedAllocator<_Tp>&, const alignedAllocator<_Tp>&){ return false; }
+template<typename _Tp>  inline bool operator==(const uvmAllocator<_Tp>&, const uvmAllocator<_Tp>&){ return true; }
+template<typename _Tp>  inline bool operator!=(const uvmAllocator<_Tp>&, const uvmAllocator<_Tp>&){ return false; }
 
 ////////////////////////////////////////////////////////////////////////////////
 // Template typedefs
 ////////////////////////////////////////////////////////////////////////////////
-template<class T> using commAllocator = alignedAllocator<T>;
-template<class T> using Vector     = std::vector<T,alignedAllocator<T> >;           
-template<class T> using commVector = std::vector<T,alignedAllocator<T> >;
-template<class T> using Matrix     = std::vector<std::vector<T,alignedAllocator<T> > >;
+template<class T> using commAllocator = uvmAllocator<T>;
+template<class T> using Vector     = std::vector<T,uvmAllocator<T> >;           
+template<class T> using commVector = std::vector<T,uvmAllocator<T> >;
+//template<class T> using Matrix     = std::vector<std::vector<T,alignedAllocator<T> > >;
 
 NAMESPACE_END(Grid);
 
-#endif
+
diff --git a/Grid/allocator/Allocator.h b/Grid/allocator/Allocator.h
new file mode 100644
index 00000000..589ea36f
--- /dev/null
+++ b/Grid/allocator/Allocator.h
@@ -0,0 +1,4 @@
+#pragma once
+#include <Grid/allocator/MemoryStats.h>
+#include <Grid/allocator/MemoryManager.h>
+#include <Grid/allocator/AlignedAllocator.h>
diff --git a/Grid/allocator/MemoryManager.cc b/Grid/allocator/MemoryManager.cc
new file mode 100644
index 00000000..e11ce948
--- /dev/null
+++ b/Grid/allocator/MemoryManager.cc
@@ -0,0 +1,244 @@
+#include <Grid/GridCore.h>
+
+NAMESPACE_BEGIN(Grid);
+
+/*Allocation types, saying which pointer cache should be used*/
+#define Cpu      (0)
+#define CpuSmall (1)
+#define Acc      (2)
+#define AccSmall (3)
+#define Shared   (4)
+#define SharedSmall (5)
+uint64_t total_shared;
+uint64_t total_device;
+uint64_t total_host;;
+void MemoryManager::PrintBytes(void)
+{
+  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;
+}
+
+//////////////////////////////////////////////////////////////////////
+// Data tables for recently freed pooiniter caches
+//////////////////////////////////////////////////////////////////////
+MemoryManager::AllocationCacheEntry MemoryManager::Entries[MemoryManager::NallocType][MemoryManager::NallocCacheMax];
+int MemoryManager::Victim[MemoryManager::NallocType];
+int MemoryManager::Ncache[MemoryManager::NallocType] = { 8, 32, 8, 32, 8, 32 };
+
+//////////////////////////////////////////////////////////////////////
+// Actual allocation and deallocation utils
+//////////////////////////////////////////////////////////////////////
+void *MemoryManager::AcceleratorAllocate(size_t bytes)
+{
+  void *ptr = (void *) Lookup(bytes,Acc);
+  if ( ptr == (void *) NULL ) {
+    ptr = (void *) acceleratorAllocDevice(bytes);
+    total_device+=bytes;
+  }
+  return ptr;
+}
+void  MemoryManager::AcceleratorFree    (void *ptr,size_t bytes)
+{
+  void *__freeme = Insert(ptr,bytes,Acc);
+  if ( __freeme ) {
+    acceleratorFreeDevice(__freeme);
+    total_device-=bytes;
+    //    PrintBytes();
+  }
+}
+void *MemoryManager::SharedAllocate(size_t bytes)
+{
+  void *ptr = (void *) Lookup(bytes,Shared);
+  if ( ptr == (void *) NULL ) {
+    ptr = (void *) acceleratorAllocShared(bytes);
+    total_shared+=bytes;
+    //    std::cout <<"AcceleratorAllocate: allocated Shared pointer "<<std::hex<<ptr<<std::dec<<std::endl;
+    //    PrintBytes();
+  }
+  return ptr;
+}
+void  MemoryManager::SharedFree    (void *ptr,size_t bytes)
+{
+  void *__freeme = Insert(ptr,bytes,Shared);
+  if ( __freeme ) {
+    acceleratorFreeShared(__freeme);
+    total_shared-=bytes;
+    //    PrintBytes();
+  }
+}
+#ifdef GRID_UVM
+void *MemoryManager::CpuAllocate(size_t bytes)
+{
+  void *ptr = (void *) Lookup(bytes,Cpu);
+  if ( ptr == (void *) NULL ) {
+    ptr = (void *) acceleratorAllocShared(bytes);
+    total_host+=bytes;
+  }
+  return ptr;
+}
+void  MemoryManager::CpuFree    (void *_ptr,size_t bytes)
+{
+  NotifyDeletion(_ptr);
+  void *__freeme = Insert(_ptr,bytes,Cpu);
+  if ( __freeme ) { 
+    acceleratorFreeShared(__freeme);
+    total_host-=bytes;
+  }
+}
+#else
+void *MemoryManager::CpuAllocate(size_t bytes)
+{
+  void *ptr = (void *) Lookup(bytes,Cpu);
+  if ( ptr == (void *) NULL ) {
+    ptr = (void *) acceleratorAllocCpu(bytes);
+    total_host+=bytes;
+  }
+  return ptr;
+}
+void  MemoryManager::CpuFree    (void *_ptr,size_t bytes)
+{
+  NotifyDeletion(_ptr);
+  void *__freeme = Insert(_ptr,bytes,Cpu);
+  if ( __freeme ) { 
+    acceleratorFreeCpu(__freeme);
+    total_host-=bytes;
+  }
+}
+#endif
+
+//////////////////////////////////////////
+// call only once
+//////////////////////////////////////////
+void MemoryManager::Init(void)
+{
+
+  char * str;
+  int Nc;
+  int NcS;
+  
+  str= getenv("GRID_ALLOC_NCACHE_LARGE");
+  if ( str ) {
+    Nc = atoi(str);
+    if ( (Nc>=0) && (Nc < NallocCacheMax)) {
+      Ncache[Cpu]=Nc;
+      Ncache[Acc]=Nc;
+      Ncache[Shared]=Nc;
+    }
+  }
+
+  str= getenv("GRID_ALLOC_NCACHE_SMALL");
+  if ( str ) {
+    Nc = atoi(str);
+    if ( (Nc>=0) && (Nc < NallocCacheMax)) {
+      Ncache[CpuSmall]=Nc;
+      Ncache[AccSmall]=Nc;
+      Ncache[SharedSmall]=Nc;
+    }
+  }
+  std::cout << GridLogMessage<< "MemoryManager::Init() setting up"<<std::endl;
+#ifdef ALLOCATION_CACHE
+  std::cout << GridLogMessage<< "MemoryManager::Init() cache pool for recent allocations: SMALL "<<Ncache[CpuSmall]<<" LARGE "<<Ncache[Cpu]<<std::endl;
+#endif
+
+#ifdef GRID_UVM
+  std::cout << GridLogMessage<< "MemoryManager::Init() Unified memory space"<<std::endl;
+#ifdef GRID_CUDA
+  std::cout << GridLogMessage<< "MemoryManager::Init() Using cudaMallocManaged"<<std::endl;
+#endif
+#ifdef GRID_HIP
+  std::cout << GridLogMessage<< "MemoryManager::Init() Using hipMallocManaged"<<std::endl;
+#endif
+#ifdef GRID_SYCL
+  std::cout << GridLogMessage<< "MemoryManager::Init() Using SYCL malloc_shared"<<std::endl;
+#endif
+#else
+  std::cout << GridLogMessage<< "MemoryManager::Init() Non unified: Caching accelerator data in dedicated memory"<<std::endl;
+#ifdef GRID_CUDA
+  std::cout << GridLogMessage<< "MemoryManager::Init() Using cudaMalloc"<<std::endl;
+#endif
+#ifdef GRID_HIP
+  std::cout << GridLogMessage<< "MemoryManager::Init() Using hipMalloc"<<std::endl;
+#endif
+#ifdef GRID_SYCL
+  std::cout << GridLogMessage<< "MemoryManager::Init() Using SYCL malloc_device"<<std::endl;
+#endif
+#endif
+}
+
+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]);  
+#else
+  return ptr;
+#endif
+}
+
+void *MemoryManager::Insert(void *ptr,size_t bytes,AllocationCacheEntry *entries,int ncache,int &victim) 
+{
+  assert(ncache>0);
+#ifdef GRID_OMP
+  assert(omp_in_parallel()==0);
+#endif 
+
+  void * ret = NULL;
+  int v = -1;
+
+  for(int e=0;e<ncache;e++) {
+    if ( entries[e].valid==0 ) {
+      v=e; 
+      break;
+    }
+  }
+
+  if ( v==-1 ) {
+    v=victim;
+    victim = (victim+1)%ncache;
+  }
+
+  if ( entries[v].valid ) {
+    ret = entries[v].address;
+    entries[v].valid = 0;
+    entries[v].address = NULL;
+    entries[v].bytes = 0;
+  }
+
+  entries[v].address=ptr;
+  entries[v].bytes  =bytes;
+  entries[v].valid  =1;
+
+  return ret;
+}
+
+void *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]);
+#else
+  return NULL;
+#endif
+}
+
+void *MemoryManager::Lookup(size_t bytes,AllocationCacheEntry *entries,int ncache) 
+{
+  assert(ncache>0);
+#ifdef GRID_OMP
+  assert(omp_in_parallel()==0);
+#endif 
+  for(int e=0;e<ncache;e++){
+    if ( entries[e].valid && ( entries[e].bytes == bytes ) ) {
+      entries[e].valid = 0;
+      return entries[e].address;
+    }
+  }
+  return NULL;
+}
+
+
+NAMESPACE_END(Grid);
+
diff --git a/Grid/allocator/MemoryManager.h b/Grid/allocator/MemoryManager.h
new file mode 100644
index 00000000..f158bade
--- /dev/null
+++ b/Grid/allocator/MemoryManager.h
@@ -0,0 +1,181 @@
+/*************************************************************************************
+
+    Grid physics library, www.github.com/paboyle/Grid 
+
+    Source file: ./lib/MemoryManager.h
+
+    Copyright (C) 2015
+
+Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
+Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License along
+    with this program; if not, write to the Free Software Foundation, Inc.,
+    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+    See the full license in the file "LICENSE" in the top level distribution directory
+*************************************************************************************/
+/*  END LEGAL */
+#pragma once
+#include <list> 
+#include <unordered_map>  
+
+NAMESPACE_BEGIN(Grid);
+
+// Move control to configure.ac and Config.h?
+
+#define ALLOCATION_CACHE
+#define GRID_ALLOC_ALIGN (2*1024*1024)
+#define GRID_ALLOC_SMALL_LIMIT (4096)
+
+/*Pinning pages is costly*/
+////////////////////////////////////////////////////////////////////////////
+// Advise the LatticeAccelerator class
+////////////////////////////////////////////////////////////////////////////
+enum ViewAdvise {
+ AdviseDefault       = 0x0,    // Regular data
+ AdviseInfrequentUse = 0x1     // Advise that the data is used infrequently.  This can
+                               // significantly influence performance of bulk storage.
+ 
+ // AdviseTransient      = 0x2,   // Data will mostly be read.  On some architectures
+                               // enables read-only copies of memory to be kept on
+                               // host and device.
+
+ // AdviseAcceleratorWriteDiscard = 0x4  // Field will be written in entirety on device
+
+};
+
+////////////////////////////////////////////////////////////////////////////
+// View Access Mode
+////////////////////////////////////////////////////////////////////////////
+enum ViewMode {
+  AcceleratorRead  = 0x01,
+  AcceleratorWrite = 0x02,
+  AcceleratorWriteDiscard = 0x04,
+  CpuRead  = 0x08,
+  CpuWrite = 0x10,
+  CpuWriteDiscard = 0x10 // same for now
+};
+
+class MemoryManager {
+private:
+
+  ////////////////////////////////////////////////////////////
+  // For caching recently freed allocations
+  ////////////////////////////////////////////////////////////
+  typedef struct { 
+    void *address;
+    size_t bytes;
+    int valid;
+  } AllocationCacheEntry;
+
+  static const int NallocCacheMax=128; 
+  static const int NallocType=6;
+  static AllocationCacheEntry Entries[NallocType][NallocCacheMax];
+  static int Victim[NallocType];
+  static int Ncache[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) ;
+  static void *Lookup(size_t bytes,AllocationCacheEntry *entries,int ncache) ;
+
+  static void *AcceleratorAllocate(size_t bytes);
+  static void  AcceleratorFree    (void *ptr,size_t bytes);
+  static void PrintBytes(void);
+ public:
+  static void Init(void);
+  static void *SharedAllocate(size_t bytes);
+  static void  SharedFree    (void *ptr,size_t bytes);
+  static void *CpuAllocate(size_t bytes);
+  static void  CpuFree    (void *ptr,size_t bytes);
+
+  ////////////////////////////////////////////////////////
+  // Footprint tracking
+  ////////////////////////////////////////////////////////
+  static uint64_t     DeviceBytes;
+  static uint64_t     DeviceLRUBytes;
+  static uint64_t     DeviceMaxBytes;
+  static uint64_t     HostToDeviceBytes;
+  static uint64_t     DeviceToHostBytes;
+  static uint64_t     HostToDeviceXfer;
+  static uint64_t     DeviceToHostXfer;
+ 
+ private:
+#ifndef GRID_UVM
+  //////////////////////////////////////////////////////////////////////
+  // Data tables for ViewCache
+  //////////////////////////////////////////////////////////////////////
+  typedef std::list<uint64_t> LRU_t;
+  typedef typename LRU_t::iterator LRUiterator;
+  typedef struct { 
+    int        LRU_valid;
+    LRUiterator LRU_entry;
+    uint64_t CpuPtr;
+    uint64_t AccPtr;
+    size_t   bytes;
+    uint32_t transient;
+    uint32_t state;
+    uint32_t accLock;
+    uint32_t cpuLock;
+  } AcceleratorViewEntry;
+  
+  typedef std::unordered_map<uint64_t,AcceleratorViewEntry> AccViewTable_t;
+  typedef typename AccViewTable_t::iterator AccViewTableIterator ;
+
+  static AccViewTable_t AccViewTable;
+  static LRU_t LRU;
+
+  /////////////////////////////////////////////////
+  // Device motion
+  /////////////////////////////////////////////////
+  static void  Create(uint64_t CpuPtr,size_t bytes,ViewMode mode,ViewAdvise hint);
+  static void  EvictVictims(uint64_t bytes); // Frees up <bytes>
+  static void  Evict(AcceleratorViewEntry &AccCache);
+  static void  Flush(AcceleratorViewEntry &AccCache);
+  static void  Clone(AcceleratorViewEntry &AccCache);
+  static void  AccDiscard(AcceleratorViewEntry &AccCache);
+  static void  CpuDiscard(AcceleratorViewEntry &AccCache);
+
+  //  static void  LRUupdate(AcceleratorViewEntry &AccCache);
+  static void  LRUinsert(AcceleratorViewEntry &AccCache);
+  static void  LRUremove(AcceleratorViewEntry &AccCache);
+  
+  // manage entries in the table
+  static int                  EntryPresent(uint64_t CpuPtr);
+  static void                 EntryCreate(uint64_t CpuPtr,size_t bytes,ViewMode mode,ViewAdvise hint);
+  static void                 EntryErase (uint64_t CpuPtr);
+  static AccViewTableIterator EntryLookup(uint64_t CpuPtr);
+  static void                 EntrySet   (uint64_t CpuPtr,AcceleratorViewEntry &entry);
+
+  static void     AcceleratorViewClose(uint64_t AccPtr);
+  static uint64_t AcceleratorViewOpen(uint64_t  CpuPtr,size_t bytes,ViewMode mode,ViewAdvise hint);
+  static void     CpuViewClose(uint64_t Ptr);
+  static uint64_t CpuViewOpen(uint64_t  CpuPtr,size_t bytes,ViewMode mode,ViewAdvise hint);
+#endif
+  static void NotifyDeletion(void * CpuPtr);
+
+ public:
+  static void Print(void);
+  static int   isOpen   (void* CpuPtr);
+  static void  ViewClose(void* CpuPtr,ViewMode mode);
+  static void *ViewOpen (void* CpuPtr,size_t bytes,ViewMode mode,ViewAdvise hint);
+
+};
+
+NAMESPACE_END(Grid);
+
+
diff --git a/Grid/allocator/MemoryManagerCache.cc b/Grid/allocator/MemoryManagerCache.cc
new file mode 100644
index 00000000..5dd7575e
--- /dev/null
+++ b/Grid/allocator/MemoryManagerCache.cc
@@ -0,0 +1,468 @@
+#include <Grid/GridCore.h>
+
+#ifndef GRID_UVM
+
+#warning "Using explicit device memory copies"
+NAMESPACE_BEGIN(Grid);
+#define dprintf(...)
+
+////////////////////////////////////////////////////////////
+// For caching copies of data on device
+////////////////////////////////////////////////////////////
+MemoryManager::AccViewTable_t MemoryManager::AccViewTable;
+MemoryManager::LRU_t MemoryManager::LRU;
+  
+////////////////////////////////////////////////////////
+// Footprint tracking
+////////////////////////////////////////////////////////
+uint64_t  MemoryManager::DeviceBytes;
+uint64_t  MemoryManager::DeviceLRUBytes;
+uint64_t  MemoryManager::DeviceMaxBytes = 1024*1024*128;
+uint64_t  MemoryManager::HostToDeviceBytes;
+uint64_t  MemoryManager::DeviceToHostBytes;
+uint64_t  MemoryManager::HostToDeviceXfer;
+uint64_t  MemoryManager::DeviceToHostXfer;
+
+////////////////////////////////////
+// Priority ordering for unlocked entries
+//  Empty
+//  CpuDirty 
+//  Consistent
+//  AccDirty
+////////////////////////////////////
+#define Empty         (0x0)  /*Entry unoccupied  */
+#define CpuDirty      (0x1)  /*CPU copy is golden, Acc buffer MAY not be allocated*/
+#define Consistent    (0x2)  /*ACC copy AND CPU copy are valid */
+#define AccDirty      (0x4)  /*ACC copy is golden */
+#define EvictNext     (0x8)  /*Priority for eviction*/
+
+/////////////////////////////////////////////////
+// Mechanics of data table maintenance
+/////////////////////////////////////////////////
+int   MemoryManager::EntryPresent(uint64_t CpuPtr)
+{
+  if(AccViewTable.empty()) return 0;
+
+  auto count = AccViewTable.count(CpuPtr);  assert((count==0)||(count==1));
+  return count;
+}
+void  MemoryManager::EntryCreate(uint64_t CpuPtr,size_t bytes,ViewMode mode,ViewAdvise hint)
+{
+  assert(!EntryPresent(CpuPtr));
+  AcceleratorViewEntry AccCache;
+  AccCache.CpuPtr = CpuPtr;
+  AccCache.AccPtr = (uint64_t)NULL;
+  AccCache.bytes  = bytes;
+  AccCache.state  = CpuDirty;
+  AccCache.LRU_valid=0;
+  AccCache.transient=0;
+  AccCache.accLock=0;
+  AccCache.cpuLock=0;
+  AccViewTable[CpuPtr] = AccCache;
+}
+MemoryManager::AccViewTableIterator MemoryManager::EntryLookup(uint64_t CpuPtr)
+{
+  assert(EntryPresent(CpuPtr));
+  auto AccCacheIterator = AccViewTable.find(CpuPtr);
+  assert(AccCacheIterator!=AccViewTable.end());
+  return AccCacheIterator;
+}
+void MemoryManager::EntryErase(uint64_t CpuPtr)
+{
+  auto AccCache = EntryLookup(CpuPtr);
+  AccViewTable.erase(CpuPtr);
+}
+void  MemoryManager::LRUinsert(AcceleratorViewEntry &AccCache)
+{
+  assert(AccCache.LRU_valid==0);
+  if (AccCache.transient) { 
+    LRU.push_back(AccCache.CpuPtr);
+    AccCache.LRU_entry = --LRU.end();
+  } else {
+    LRU.push_front(AccCache.CpuPtr);
+    AccCache.LRU_entry = LRU.begin();
+  }
+  AccCache.LRU_valid = 1;
+  DeviceLRUBytes+=AccCache.bytes;
+}
+void  MemoryManager::LRUremove(AcceleratorViewEntry &AccCache)
+{
+  assert(AccCache.LRU_valid==1);
+  LRU.erase(AccCache.LRU_entry);
+  AccCache.LRU_valid = 0;
+  DeviceLRUBytes-=AccCache.bytes;
+}
+/////////////////////////////////////////////////
+// Accelerator cache motion & consistency logic
+/////////////////////////////////////////////////
+void MemoryManager::AccDiscard(AcceleratorViewEntry &AccCache)
+{
+  ///////////////////////////////////////////////////////////
+  // Remove from Accelerator, remove entry, without flush
+  // Cannot be locked. If allocated Must be in LRU pool.
+  ///////////////////////////////////////////////////////////
+  assert(AccCache.state!=Empty);
+  
+  //  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);
+  if(AccCache.AccPtr) {
+    AcceleratorFree((void *)AccCache.AccPtr,AccCache.bytes);
+    DeviceBytes   -=AccCache.bytes;
+    LRUremove(AccCache);
+    //    dprintf("MemoryManager: Free(%llx) LRU %lld Total %lld\n",(uint64_t)AccCache.AccPtr,DeviceLRUBytes,DeviceBytes);  
+  }
+  uint64_t CpuPtr = AccCache.CpuPtr;
+  EntryErase(CpuPtr);
+}
+
+void MemoryManager::Evict(AcceleratorViewEntry &AccCache)
+{
+  ///////////////////////////////////////////////////////////////////////////
+  // Make CPU consistent, remove from Accelerator, remove entry
+  // Cannot be locked. If allocated must be in LRU pool.
+  ///////////////////////////////////////////////////////////////////////////
+  assert(AccCache.state!=Empty);
+  
+  //  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);
+  }
+  assert(AccCache.CpuPtr!=(uint64_t)NULL);
+  if(AccCache.AccPtr) {
+    AcceleratorFree((void *)AccCache.AccPtr,AccCache.bytes);
+    DeviceBytes   -=AccCache.bytes;
+    LRUremove(AccCache);
+    //    dprintf("MemoryManager: Free(%llx) footprint now %lld \n",(uint64_t)AccCache.AccPtr,DeviceBytes);  
+  }
+  uint64_t CpuPtr = AccCache.CpuPtr;
+  EntryErase(CpuPtr);
+}
+void MemoryManager::Flush(AcceleratorViewEntry &AccCache)
+{
+  assert(AccCache.state==AccDirty);
+  assert(AccCache.cpuLock==0);
+  assert(AccCache.accLock==0);
+  assert(AccCache.AccPtr!=(uint64_t)NULL);
+  assert(AccCache.CpuPtr!=(uint64_t)NULL);
+  acceleratorCopyFromDevice((void *)AccCache.AccPtr,(void *)AccCache.CpuPtr,AccCache.bytes);
+  //  dprintf("MemoryManager: Flush  %llx -> %llx\n",(uint64_t)AccCache.AccPtr,(uint64_t)AccCache.CpuPtr); fflush(stdout);
+  DeviceToHostBytes+=AccCache.bytes;
+  DeviceToHostXfer++;
+  AccCache.state=Consistent;
+}
+void MemoryManager::Clone(AcceleratorViewEntry &AccCache)
+{
+  assert(AccCache.state==CpuDirty);
+  assert(AccCache.cpuLock==0);
+  assert(AccCache.accLock==0);
+  assert(AccCache.CpuPtr!=(uint64_t)NULL);
+  if(AccCache.AccPtr==(uint64_t)NULL){
+    AccCache.AccPtr=(uint64_t)AcceleratorAllocate(AccCache.bytes);
+    DeviceBytes+=AccCache.bytes;
+  }
+  //  dprintf("MemoryManager: Clone %llx <- %llx\n",(uint64_t)AccCache.AccPtr,(uint64_t)AccCache.CpuPtr); fflush(stdout);
+  acceleratorCopyToDevice((void *)AccCache.CpuPtr,(void *)AccCache.AccPtr,AccCache.bytes);
+  HostToDeviceBytes+=AccCache.bytes;
+  HostToDeviceXfer++;
+  AccCache.state=Consistent;
+}
+
+void MemoryManager::CpuDiscard(AcceleratorViewEntry &AccCache)
+{
+  assert(AccCache.state!=Empty);
+  assert(AccCache.cpuLock==0);
+  assert(AccCache.accLock==0);
+  assert(AccCache.CpuPtr!=(uint64_t)NULL);
+  if(AccCache.AccPtr==(uint64_t)NULL){
+    AccCache.AccPtr=(uint64_t)AcceleratorAllocate(AccCache.bytes);
+    DeviceBytes+=AccCache.bytes;
+  }
+  AccCache.state=AccDirty;
+}
+
+/////////////////////////////////////////////////////////////////////////////////
+// View management
+/////////////////////////////////////////////////////////////////////////////////
+void MemoryManager::ViewClose(void* Ptr,ViewMode mode)
+{
+  if( (mode==AcceleratorRead)||(mode==AcceleratorWrite)||(mode==AcceleratorWriteDiscard) ){
+    AcceleratorViewClose((uint64_t)Ptr);
+  } else if( (mode==CpuRead)||(mode==CpuWrite)){
+    CpuViewClose((uint64_t)Ptr);
+  } else { 
+    assert(0);
+  }
+}
+void *MemoryManager::ViewOpen(void* _CpuPtr,size_t bytes,ViewMode mode,ViewAdvise hint)
+{
+  uint64_t CpuPtr = (uint64_t)_CpuPtr;
+  if( (mode==AcceleratorRead)||(mode==AcceleratorWrite)||(mode==AcceleratorWriteDiscard) ){
+    return (void *) AcceleratorViewOpen(CpuPtr,bytes,mode,hint);
+  } else if( (mode==CpuRead)||(mode==CpuWrite)){
+    return (void *)CpuViewOpen(CpuPtr,bytes,mode,hint);
+  } else { 
+    assert(0);
+    return NULL;
+  }
+}
+void  MemoryManager::EvictVictims(uint64_t bytes)
+{
+  while(bytes+DeviceLRUBytes > DeviceMaxBytes){
+    if ( DeviceLRUBytes > 0){
+      assert(LRU.size()>0);
+      uint64_t victim = LRU.back();
+      auto AccCacheIterator = EntryLookup(victim);
+      auto & AccCache = AccCacheIterator->second;
+      Evict(AccCache);
+    }
+  }
+}
+uint64_t MemoryManager::AcceleratorViewOpen(uint64_t CpuPtr,size_t bytes,ViewMode mode,ViewAdvise hint)
+{
+  ////////////////////////////////////////////////////////////////////////////
+  // Find if present, otherwise get or force an empty
+  ////////////////////////////////////////////////////////////////////////////
+  if ( EntryPresent(CpuPtr)==0 ){
+    EvictVictims(bytes);
+    EntryCreate(CpuPtr,bytes,mode,hint);
+  }
+
+  auto AccCacheIterator = EntryLookup(CpuPtr);
+  auto & AccCache = AccCacheIterator->second;
+  
+  assert((mode==AcceleratorRead)||(mode==AcceleratorWrite)||(mode==AcceleratorWriteDiscard));
+
+  assert(AccCache.cpuLock==0);  // Programming error
+
+  if(AccCache.state!=Empty) {
+    assert(AccCache.CpuPtr == CpuPtr);
+    assert(AccCache.bytes  ==bytes);
+  }
+/*
+ *  State transitions and actions
+ *
+ *  Action  State   StateNext         Flush    Clone
+ *
+ *  AccRead  Empty   Consistent        -        Y
+ *  AccWrite Empty   AccDirty          -        Y
+ *  AccRead  CpuDirty Consistent       -        Y
+ *  AccWrite CpuDirty AccDirty         -        Y
+ *  AccRead  Consistent Consistent     -        - 
+ *  AccWrite Consistent AccDirty       -        - 
+ *  AccRead  AccDirty   AccDirty       -        - 
+ *  AccWrite AccDirty   AccDirty       -        - 
+ */
+  if(AccCache.state==Empty) {
+    assert(AccCache.LRU_valid==0);
+    AccCache.CpuPtr = CpuPtr;
+    AccCache.AccPtr = (uint64_t)NULL;
+    AccCache.bytes  = bytes;
+    AccCache.state  = CpuDirty;   // Cpu starts primary
+    if(mode==AcceleratorWriteDiscard){
+      CpuDiscard(AccCache);
+      AccCache.state  = AccDirty;   // Empty + AcceleratorWrite=> AccDirty
+    } else if(mode==AcceleratorWrite){
+      Clone(AccCache);
+      AccCache.state  = AccDirty;   // Empty + AcceleratorWrite=> AccDirty
+    } else {
+      Clone(AccCache);
+      AccCache.state  = Consistent; // Empty + AccRead => Consistent
+    }
+    AccCache.accLock= 1;
+  } else if(AccCache.state==CpuDirty ){
+    if(mode==AcceleratorWriteDiscard) {
+      CpuDiscard(AccCache);
+      AccCache.state  = AccDirty;   // CpuDirty + AcceleratorWrite=> AccDirty
+    } else if(mode==AcceleratorWrite) {
+      Clone(AccCache);
+      AccCache.state  = AccDirty;   // CpuDirty + AcceleratorWrite=> AccDirty
+    } else {
+      Clone(AccCache);
+      AccCache.state  = Consistent; // CpuDirty + AccRead => Consistent
+    }
+    AccCache.accLock++;
+    //    printf("Copied CpuDirty entry into device accLock %d\n",AccCache.accLock);
+  } else if(AccCache.state==Consistent) {
+    if((mode==AcceleratorWrite)||(mode==AcceleratorWriteDiscard))
+      AccCache.state  = AccDirty;   // Consistent + AcceleratorWrite=> AccDirty
+    else
+      AccCache.state  = Consistent; // Consistent + AccRead => Consistent
+    AccCache.accLock++;
+    //    printf("Consistent entry into device accLock %d\n",AccCache.accLock);
+  } else if(AccCache.state==AccDirty) {
+    if((mode==AcceleratorWrite)||(mode==AcceleratorWriteDiscard))
+      AccCache.state  = AccDirty; // AccDirty + AcceleratorWrite=> AccDirty
+    else
+      AccCache.state  = AccDirty; // AccDirty + AccRead => AccDirty
+    AccCache.accLock++;
+    //    printf("AccDirty entry into device accLock %d\n",AccCache.accLock);
+  } else {
+    assert(0);
+  }
+
+  // If view is opened on device remove from LRU
+  if(AccCache.LRU_valid==1){
+    // must possibly remove from LRU as now locked on GPU
+    LRUremove(AccCache);
+  }
+
+  int transient =hint;
+  AccCache.transient= transient? EvictNext : 0;
+
+  return AccCache.AccPtr;
+}
+////////////////////////////////////
+// look up & decrement lock count
+////////////////////////////////////
+void MemoryManager::AcceleratorViewClose(uint64_t CpuPtr)
+{
+  auto AccCacheIterator = EntryLookup(CpuPtr);
+  auto & AccCache = AccCacheIterator->second;
+
+  assert(AccCache.cpuLock==0);
+  assert(AccCache.accLock>0);
+
+  AccCache.accLock--;
+
+  // Move to LRU queue if not locked and close on device
+  if(AccCache.accLock==0) {
+    LRUinsert(AccCache);
+  }
+}
+void MemoryManager::CpuViewClose(uint64_t CpuPtr)
+{
+  auto AccCacheIterator = EntryLookup(CpuPtr);
+  auto & AccCache = AccCacheIterator->second;
+
+  assert(AccCache.cpuLock>0);
+  assert(AccCache.accLock==0);
+
+  AccCache.cpuLock--;
+}
+/*
+ *  Action  State   StateNext         Flush    Clone
+ *
+ *  CpuRead  Empty   CpuDirty          -        -
+ *  CpuWrite Empty   CpuDirty          -        -
+ *  CpuRead  CpuDirty CpuDirty         -        -
+ *  CpuWrite CpuDirty CpuDirty         -        - 
+ *  CpuRead  Consistent Consistent     -        - 
+ *  CpuWrite Consistent CpuDirty       -        - 
+ *  CpuRead  AccDirty   Consistent     Y        -
+ *  CpuWrite AccDirty   CpuDirty       Y        -
+ */
+uint64_t MemoryManager::CpuViewOpen(uint64_t CpuPtr,size_t bytes,ViewMode mode,ViewAdvise transient)
+{
+  ////////////////////////////////////////////////////////////////////////////
+  // Find if present, otherwise get or force an empty
+  ////////////////////////////////////////////////////////////////////////////
+  if ( EntryPresent(CpuPtr)==0 ){
+    EvictVictims(bytes);
+    EntryCreate(CpuPtr,bytes,mode,transient);
+  }
+
+  auto AccCacheIterator = EntryLookup(CpuPtr);
+  auto & AccCache = AccCacheIterator->second;
+  
+  assert((mode==CpuRead)||(mode==CpuWrite));
+  assert(AccCache.accLock==0);  // Programming error
+
+  if(AccCache.state!=Empty) {
+    assert(AccCache.CpuPtr == CpuPtr);
+    assert(AccCache.bytes==bytes);
+  }
+
+  if(AccCache.state==Empty) {
+    AccCache.CpuPtr = CpuPtr;
+    AccCache.AccPtr = (uint64_t)NULL;
+    AccCache.bytes  = bytes;
+    AccCache.state  = CpuDirty; // Empty + CpuRead/CpuWrite => CpuDirty
+    AccCache.accLock= 0;
+    AccCache.cpuLock= 1;
+  } else if(AccCache.state==CpuDirty ){
+    // AccPtr dont care, deferred allocate
+    AccCache.state = CpuDirty; // CpuDirty +CpuRead/CpuWrite => CpuDirty
+    AccCache.cpuLock++;
+  } else if(AccCache.state==Consistent) {
+    assert(AccCache.AccPtr != (uint64_t)NULL);
+    if(mode==CpuWrite)
+      AccCache.state = CpuDirty;   // Consistent +CpuWrite => CpuDirty
+    else 
+      AccCache.state = Consistent; // Consistent +CpuRead  => Consistent
+    AccCache.cpuLock++;
+  } else if(AccCache.state==AccDirty) {
+    assert(AccCache.AccPtr != (uint64_t)NULL);
+    Flush(AccCache);
+    if(mode==CpuWrite) AccCache.state = CpuDirty;   // AccDirty +CpuWrite => CpuDirty, Flush
+    else            AccCache.state = Consistent; // AccDirty +CpuRead  => Consistent, Flush
+    AccCache.cpuLock++;
+  } else {
+    assert(0); // should be unreachable
+  }
+
+  AccCache.transient= transient? EvictNext : 0;
+
+  return AccCache.CpuPtr;
+}
+void  MemoryManager::NotifyDeletion(void *_ptr)
+{
+  // Look up in ViewCache
+  uint64_t ptr = (uint64_t)_ptr;
+  if(EntryPresent(ptr)) {
+    auto e = EntryLookup(ptr);
+    AccDiscard(e->second);
+  }
+}
+void  MemoryManager::Print(void)
+{
+  std::cout << GridLogDebug << "--------------------------------------------" << std::endl;
+  std::cout << GridLogDebug << "Memory Manager                             " << std::endl;
+  std::cout << GridLogDebug << "--------------------------------------------" << std::endl;
+  std::cout << GridLogDebug << DeviceBytes   << " bytes allocated on device " << std::endl;
+  std::cout << GridLogDebug << DeviceLRUBytes<< " bytes evictable on device " << std::endl;
+  std::cout << GridLogDebug << DeviceMaxBytes<< " bytes max on device       " << std::endl;
+  std::cout << GridLogDebug << HostToDeviceXfer << " transfers        to   device " << std::endl;
+  std::cout << GridLogDebug << DeviceToHostXfer << " transfers        from device " << std::endl;
+  std::cout << GridLogDebug << HostToDeviceBytes<< " bytes transfered to   device " << std::endl;
+  std::cout << GridLogDebug << DeviceToHostBytes<< " bytes transfered from device " << std::endl;
+  std::cout << GridLogDebug << AccViewTable.size()<< " vectors " << LRU.size()<<" evictable"<< std::endl;
+  std::cout << GridLogDebug << "--------------------------------------------" << std::endl;
+  std::cout << GridLogDebug << "CpuAddr\t\tAccAddr\t\tState\t\tcpuLock\taccLock\tLRU_valid "<<std::endl;
+  std::cout << GridLogDebug << "--------------------------------------------" << std::endl;
+  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");
+
+    std::cout << GridLogDebug << "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;
+  }
+  std::cout << GridLogDebug << "--------------------------------------------" << std::endl;
+
+};
+int   MemoryManager::isOpen   (void* _CpuPtr) 
+{ 
+  uint64_t CpuPtr = (uint64_t)_CpuPtr;
+  if ( EntryPresent(CpuPtr) ){
+    auto AccCacheIterator = EntryLookup(CpuPtr);
+    auto & AccCache = AccCacheIterator->second;
+    return AccCache.cpuLock+AccCache.accLock;
+  } else { 
+    return 0;
+  }
+}
+
+NAMESPACE_END(Grid);
+
+#endif
diff --git a/Grid/allocator/MemoryManagerShared.cc b/Grid/allocator/MemoryManagerShared.cc
new file mode 100644
index 00000000..537f7c32
--- /dev/null
+++ b/Grid/allocator/MemoryManagerShared.cc
@@ -0,0 +1,24 @@
+#include <Grid/GridCore.h>
+#ifdef GRID_UVM
+
+#warning "Grid is assuming unified virtual memory address space"
+NAMESPACE_BEGIN(Grid);
+/////////////////////////////////////////////////////////////////////////////////
+// View management is 1:1 address space mapping
+/////////////////////////////////////////////////////////////////////////////////
+uint64_t  MemoryManager::DeviceBytes;
+uint64_t  MemoryManager::DeviceLRUBytes;
+uint64_t  MemoryManager::DeviceMaxBytes = 1024*1024*128;
+uint64_t  MemoryManager::HostToDeviceBytes;
+uint64_t  MemoryManager::DeviceToHostBytes;
+uint64_t  MemoryManager::HostToDeviceXfer;
+uint64_t  MemoryManager::DeviceToHostXfer;
+
+void  MemoryManager::ViewClose(void* AccPtr,ViewMode mode){};
+void *MemoryManager::ViewOpen(void* CpuPtr,size_t bytes,ViewMode mode,ViewAdvise hint){ return CpuPtr; };
+int   MemoryManager::isOpen   (void* CpuPtr) { return 0;}
+void  MemoryManager::Print(void){};
+void  MemoryManager::NotifyDeletion(void *ptr){};
+
+NAMESPACE_END(Grid);
+#endif
diff --git a/Grid/allocator/MemoryStats.cc b/Grid/allocator/MemoryStats.cc
new file mode 100644
index 00000000..0d1707d9
--- /dev/null
+++ b/Grid/allocator/MemoryStats.cc
@@ -0,0 +1,67 @@
+#include <Grid/GridCore.h>
+#include <fcntl.h>
+
+NAMESPACE_BEGIN(Grid);
+
+MemoryStats *MemoryProfiler::stats = nullptr;
+bool         MemoryProfiler::debug = false;
+
+void check_huge_pages(void *Buf,uint64_t BYTES)
+{
+#ifdef __linux__
+  int fd = open("/proc/self/pagemap", O_RDONLY);
+  assert(fd >= 0);
+  const int page_size = 4096;
+  uint64_t virt_pfn = (uint64_t)Buf / page_size;
+  off_t offset = sizeof(uint64_t) * virt_pfn;
+  uint64_t npages = (BYTES + page_size-1) / page_size;
+  uint64_t pagedata[npages];
+  uint64_t ret = lseek(fd, offset, SEEK_SET);
+  assert(ret == offset);
+  ret = ::read(fd, pagedata, sizeof(uint64_t)*npages);
+  assert(ret == sizeof(uint64_t) * npages);
+  int nhugepages = npages / 512;
+  int n4ktotal, nnothuge;
+  n4ktotal = 0;
+  nnothuge = 0;
+  for (int i = 0; i < nhugepages; ++i) {
+    uint64_t baseaddr = (pagedata[i*512] & 0x7fffffffffffffULL) * page_size;
+    for (int j = 0; j < 512; ++j) {
+      uint64_t pageaddr = (pagedata[i*512+j] & 0x7fffffffffffffULL) * page_size;
+      ++n4ktotal;
+      if (pageaddr != baseaddr + j * page_size)
+	++nnothuge;
+    }
+  }
+  int rank = CartesianCommunicator::RankWorld();
+  printf("rank %d Allocated %d 4k pages, %d not in huge pages\n", rank, n4ktotal, nnothuge);
+#endif
+}
+
+std::string sizeString(const size_t bytes)
+{
+  constexpr unsigned int bufSize = 256;
+  const char             *suffixes[7] = {"", "K", "M", "G", "T", "P", "E"};
+  char                   buf[256];
+  size_t                 s     = 0;
+  double                 count = bytes;
+  
+  while (count >= 1024 && s < 7)
+    {
+      s++;
+      count /= 1024;
+    }
+  if (count - floor(count) == 0.0)
+    {
+      snprintf(buf, bufSize, "%d %sB", (int)count, suffixes[s]);
+    }
+  else
+    {
+      snprintf(buf, bufSize, "%.1f %sB", count, suffixes[s]);
+    }
+  
+  return std::string(buf);
+}
+
+NAMESPACE_END(Grid);
+
diff --git a/Grid/allocator/MemoryStats.h b/Grid/allocator/MemoryStats.h
new file mode 100644
index 00000000..156c9747
--- /dev/null
+++ b/Grid/allocator/MemoryStats.h
@@ -0,0 +1,95 @@
+/*************************************************************************************
+
+    Grid physics library, www.github.com/paboyle/Grid 
+
+    Source file: ./lib/MemoryStats.h
+
+    Copyright (C) 2015
+
+Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
+Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License along
+    with this program; if not, write to the Free Software Foundation, Inc.,
+    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+    See the full license in the file "LICENSE" in the top level distribution directory
+*************************************************************************************/
+/*  END LEGAL */
+#pragma once
+
+
+NAMESPACE_BEGIN(Grid);
+
+std::string sizeString(size_t bytes);
+
+struct MemoryStats
+{
+  size_t totalAllocated{0}, maxAllocated{0}, 
+    currentlyAllocated{0}, totalFreed{0};
+};
+    
+class MemoryProfiler
+{
+public:
+  static MemoryStats *stats;
+  static bool        debug;
+};
+
+#define memString(bytes) std::to_string(bytes) + " (" + sizeString(bytes) + ")"
+#define profilerDebugPrint						\
+  if (MemoryProfiler::stats)						\
+    {									\
+      auto s = MemoryProfiler::stats;					\
+      std::cout << GridLogDebug << "[Memory debug] Stats " << MemoryProfiler::stats << std::endl; \
+      std::cout << GridLogDebug << "[Memory debug] total  : " << memString(s->totalAllocated) \
+		<< std::endl;						\
+      std::cout << GridLogDebug << "[Memory debug] max    : " << memString(s->maxAllocated) \
+		<< std::endl;						\
+      std::cout << GridLogDebug << "[Memory debug] current: " << memString(s->currentlyAllocated) \
+		<< std::endl;						\
+      std::cout << GridLogDebug << "[Memory debug] freed  : " << memString(s->totalFreed) \
+		<< std::endl;						\
+    }
+
+#define profilerAllocate(bytes)						\
+  if (MemoryProfiler::stats)						\
+    {									\
+      auto s = MemoryProfiler::stats;					\
+      s->totalAllocated     += (bytes);					\
+      s->currentlyAllocated += (bytes);					\
+      s->maxAllocated        = std::max(s->maxAllocated, s->currentlyAllocated); \
+    }									\
+  if (MemoryProfiler::debug)						\
+    {									\
+      std::cout << GridLogDebug << "[Memory debug] allocating " << memString(bytes) << std::endl; \
+      profilerDebugPrint;						\
+    }
+
+#define profilerFree(bytes)						\
+  if (MemoryProfiler::stats)						\
+    {									\
+      auto s = MemoryProfiler::stats;					\
+      s->totalFreed         += (bytes);					\
+      s->currentlyAllocated -= (bytes);					\
+    }									\
+  if (MemoryProfiler::debug)						\
+    {									\
+      std::cout << GridLogDebug << "[Memory debug] freeing " << memString(bytes) << std::endl; \
+      profilerDebugPrint;						\
+    }
+
+void check_huge_pages(void *Buf,uint64_t BYTES);
+
+NAMESPACE_END(Grid);
+
diff --git a/Grid/cartesian/Cartesian_base.h b/Grid/cartesian/Cartesian_base.h
index 87472cc9..ae1fd1fd 100644
--- a/Grid/cartesian/Cartesian_base.h
+++ b/Grid/cartesian/Cartesian_base.h
@@ -81,6 +81,7 @@ public:
 
   bool _isCheckerBoarded; 
   int        LocallyPeriodic;
+  Coordinate _checker_dim_mask;
 
 public:
 
diff --git a/Grid/cartesian/Cartesian_full.h b/Grid/cartesian/Cartesian_full.h
index c083817b..31a67bf0 100644
--- a/Grid/cartesian/Cartesian_full.h
+++ b/Grid/cartesian/Cartesian_full.h
@@ -38,6 +38,7 @@ class GridCartesian: public GridBase {
 
 public:
   int dummy;
+  Coordinate _checker_dim_mask;
   virtual int  CheckerBoardFromOindexTable (int Oindex) {
     return 0;
   }
@@ -104,6 +105,7 @@ public:
     _ldimensions.resize(_ndimension);
     _rdimensions.resize(_ndimension);
     _simd_layout.resize(_ndimension);
+    _checker_dim_mask.resize(_ndimension);;
     _lstart.resize(_ndimension);
     _lend.resize(_ndimension);
 
@@ -114,6 +116,8 @@ public:
 
     for (int d = 0; d < _ndimension; d++)
       {
+	_checker_dim_mask[d]=0;
+
         _fdimensions[d] = dimensions[d];   // Global dimensions
         _gdimensions[d] = _fdimensions[d]; // Global dimensions
         _simd_layout[d] = simd_layout[d];
diff --git a/Grid/cartesian/Cartesian_red_black.h b/Grid/cartesian/Cartesian_red_black.h
index 34f763d2..b71981f5 100644
--- a/Grid/cartesian/Cartesian_red_black.h
+++ b/Grid/cartesian/Cartesian_red_black.h
@@ -35,12 +35,28 @@ static const int CbRed  =0;
 static const int CbBlack=1;
 static const int Even   =CbRed;
 static const int Odd    =CbBlack;
+
+accelerator_inline int RedBlackCheckerBoardFromOindex (int oindex, Coordinate &rdim, Coordinate &chk_dim_msk)
+{
+  int nd=rdim.size();
+  Coordinate coor(nd);
+
+  Lexicographic::CoorFromIndex(coor,oindex,rdim);
+
+  int linear=0;
+  for(int d=0;d<nd;d++){
+    if(chk_dim_msk[d])
+      linear=linear+coor[d];
+  }
+  return (linear&0x1);
+}
+
     
 // Specialise this for red black grids storing half the data like a chess board.
 class GridRedBlackCartesian : public GridBase
 {
 public:
-  Coordinate _checker_dim_mask;
+  //  Coordinate _checker_dim_mask;
   int              _checker_dim;
   std::vector<int> _checker_board;
 
diff --git a/Grid/communicator/Communicator_base.h b/Grid/communicator/Communicator_base.h
index 11dbfcbb..436d75ef 100644
--- a/Grid/communicator/Communicator_base.h
+++ b/Grid/communicator/Communicator_base.h
@@ -114,6 +114,7 @@ public:
   void GlobalSumVector(RealD *,int N);
   void GlobalSum(uint32_t &);
   void GlobalSum(uint64_t &);
+  void GlobalSumVector(uint64_t*,int N);
   void GlobalSum(ComplexF &c);
   void GlobalSumVector(ComplexF *c,int N);
   void GlobalSum(ComplexD &c);
diff --git a/Grid/communicator/Communicator_mpi3.cc b/Grid/communicator/Communicator_mpi3.cc
index e9399ddc..3fe8a297 100644
--- a/Grid/communicator/Communicator_mpi3.cc
+++ b/Grid/communicator/Communicator_mpi3.cc
@@ -275,6 +275,10 @@ void CartesianCommunicator::GlobalSum(uint64_t &u){
   int ierr=MPI_Allreduce(MPI_IN_PLACE,&u,1,MPI_UINT64_T,MPI_SUM,communicator);
   assert(ierr==0);
 }
+void CartesianCommunicator::GlobalSumVector(uint64_t* u,int N){
+  int ierr=MPI_Allreduce(MPI_IN_PLACE,u,N,MPI_UINT64_T,MPI_SUM,communicator);
+  assert(ierr==0);
+}
 void CartesianCommunicator::GlobalXOR(uint32_t &u){
   int ierr=MPI_Allreduce(MPI_IN_PLACE,&u,1,MPI_UINT32_T,MPI_BXOR,communicator);
   assert(ierr==0);
diff --git a/Grid/communicator/Communicator_none.cc b/Grid/communicator/Communicator_none.cc
index b8a15a0e..81900371 100644
--- a/Grid/communicator/Communicator_none.cc
+++ b/Grid/communicator/Communicator_none.cc
@@ -70,9 +70,10 @@ CartesianCommunicator::~CartesianCommunicator(){}
 void CartesianCommunicator::GlobalSum(float &){}
 void CartesianCommunicator::GlobalSumVector(float *,int N){}
 void CartesianCommunicator::GlobalSum(double &){}
+void CartesianCommunicator::GlobalSumVector(double *,int N){}
 void CartesianCommunicator::GlobalSum(uint32_t &){}
 void CartesianCommunicator::GlobalSum(uint64_t &){}
-void CartesianCommunicator::GlobalSumVector(double *,int N){}
+void CartesianCommunicator::GlobalSumVector(uint64_t *,int N){}
 void CartesianCommunicator::GlobalXOR(uint32_t &){}
 void CartesianCommunicator::GlobalXOR(uint64_t &){}
 
diff --git a/Grid/communicator/SharedMemory.cc b/Grid/communicator/SharedMemory.cc
index 5bca9764..de10da3d 100644
--- a/Grid/communicator/SharedMemory.cc
+++ b/Grid/communicator/SharedMemory.cc
@@ -74,7 +74,9 @@ void *SharedMemory::ShmBufferMalloc(size_t bytes){
   if (heap_bytes >= heap_size) {
     std::cout<< " ShmBufferMalloc exceeded shared heap size -- try increasing with --shm <MB> flag" <<std::endl;
     std::cout<< " Parameter specified in units of MB (megabytes) " <<std::endl;
-    std::cout<< " Current value is " << (heap_size/(1024*1024)) <<std::endl;
+    std::cout<< " Current alloc is " << (bytes/(1024*1024)) <<"MB"<<std::endl;
+    std::cout<< " Current bytes is " << (heap_bytes/(1024*1024)) <<"MB"<<std::endl;
+    std::cout<< " Current heap  is " << (heap_size/(1024*1024)) <<"MB"<<std::endl;
     assert(heap_bytes<heap_size);
   }
   //std::cerr << "ShmBufferMalloc "<<std::hex<< ptr<<" - "<<((uint64_t)ptr+bytes)<<std::dec<<std::endl;
diff --git a/Grid/communicator/SharedMemoryMPI.cc b/Grid/communicator/SharedMemoryMPI.cc
index ed465252..45fefc71 100644
--- a/Grid/communicator/SharedMemoryMPI.cc
+++ b/Grid/communicator/SharedMemoryMPI.cc
@@ -29,7 +29,7 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 #include <Grid/GridCore.h>
 #include <pwd.h>
 
-#ifdef GRID_NVCC
+#ifdef GRID_CUDA
 #include <cuda_runtime_api.h>
 #endif
 
@@ -170,17 +170,24 @@ void GlobalSharedMemory::GetShmDims(const Coordinate &WorldDims,Coordinate &ShmD
   std::vector<int> primes({2,3,5});
 
   int dim = 0;
+  int last_dim = ndimension - 1;
   int AutoShmSize = 1;
   while(AutoShmSize != WorldShmSize) {
-    for(int p=0;p<primes.size();p++) {
+    int p;
+    for(p=0;p<primes.size();p++) {
       int prime=primes[p];
       if ( divides(prime,WorldDims[dim]/ShmDims[dim])
         && divides(prime,WorldShmSize/AutoShmSize)  ) {
 	AutoShmSize*=prime;
 	ShmDims[dim]*=prime;
+	last_dim = dim;
 	break;
       }
     }
+    if (p == primes.size() && last_dim == dim) {
+      std::cerr << "GlobalSharedMemory::GetShmDims failed" << std::endl;
+      exit(EXIT_FAILURE);
+    }
     dim=(dim+1) %ndimension;
   }
 }
@@ -413,7 +420,7 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
 ////////////////////////////////////////////////////////////////////////////////////////////
 // Hugetlbfs mapping intended
 ////////////////////////////////////////////////////////////////////////////////////////////
-#ifdef GRID_NVCC
+#ifdef GRID_CUDA
 void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
 {
   void * ShmCommBuf ; 
@@ -433,13 +440,6 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
   //////////////////////////////////////////////////////////////////////////////////////////////////////////
   //  cudaDeviceGetP2PAttribute(&perfRank, cudaDevP2PAttrPerformanceRank, device1, device2);
 
-#ifdef GRID_IBM_SUMMIT
-  // IBM Jsrun makes cuda Device numbering screwy and not match rank
-    std::cout << "IBM Summit or similar - NOT setting device to WorldShmRank"<<std::endl;
-#else
-    std::cout << "setting device to WorldShmRank"<<std::endl;
-    cudaSetDevice(WorldShmRank);
-#endif
   ///////////////////////////////////////////////////////////////////////////////////////////////////////////
   // Each MPI rank should allocate our own buffer
   ///////////////////////////////////////////////////////////////////////////////////////////////////////////
@@ -677,7 +677,7 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
 /////////////////////////////////////////////////////////////////////////
 void GlobalSharedMemory::SharedMemoryZero(void *dest,size_t bytes)
 {
-#ifdef GRID_NVCC
+#ifdef GRID_CUDA
   cudaMemset(dest,0,bytes);
 #else
   bzero(dest,bytes);
@@ -685,7 +685,7 @@ void GlobalSharedMemory::SharedMemoryZero(void *dest,size_t bytes)
 }
 void GlobalSharedMemory::SharedMemoryCopy(void *dest,const void *src,size_t bytes)
 {
-#ifdef GRID_NVCC
+#ifdef GRID_CUDA
   cudaMemcpy(dest,src,bytes,cudaMemcpyDefault);
 #else   
   bcopy(src,dest,bytes);
diff --git a/Grid/cshift/Cshift.h b/Grid/cshift/Cshift.h
index 07ec8412..9150579c 100644
--- a/Grid/cshift/Cshift.h
+++ b/Grid/cshift/Cshift.h
@@ -49,4 +49,29 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 #ifdef GRID_COMMS_SHMEM
 #include <Grid/cshift/Cshift_mpi.h> // uses same implementation of communicator
 #endif 
+
+NAMESPACE_BEGIN(Grid);
+
+template<typename Op, typename T1> 
+auto Cshift(const LatticeUnaryExpression<Op,T1> &expr,int dim,int shift)
+    -> Lattice<decltype(expr.op.func(eval(0, expr.arg1)))> 
+{
+  return Cshift(closure(expr),dim,shift);
+}
+template <class Op, class T1, class T2>
+auto Cshift(const LatticeBinaryExpression<Op,T1,T2> &expr,int dim,int shift)
+  -> Lattice<decltype(expr.op.func(eval(0, expr.arg1),eval(0, expr.arg2)))> 
+{
+  return Cshift(closure(expr),dim,shift);
+}
+template <class Op, class T1, class T2, class T3>
+auto Cshift(const LatticeTrinaryExpression<Op,T1,T2,T3> &expr,int dim,int shift)
+  -> Lattice<decltype(expr.op.func(eval(0, expr.arg1),
+				   eval(0, expr.arg2),
+				   eval(0, expr.arg3)))> 
+{
+  return Cshift(closure(expr),dim,shift);
+}
+NAMESPACE_END(Grid);
+
 #endif
diff --git a/Grid/cshift/Cshift_common.h b/Grid/cshift/Cshift_common.h
index 954342cb..4de2bbe2 100644
--- a/Grid/cshift/Cshift_common.h
+++ b/Grid/cshift/Cshift_common.h
@@ -29,6 +29,8 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 
 NAMESPACE_BEGIN(Grid);
 
+extern Vector<std::pair<int,int> > Cshift_table; 
+
 ///////////////////////////////////////////////////////////////////
 // Gather for when there is no need to SIMD split 
 ///////////////////////////////////////////////////////////////////
@@ -46,16 +48,16 @@ Gather_plane_simple (const Lattice<vobj> &rhs,commVector<vobj> &buffer,int dimen
   int e2=rhs.Grid()->_slice_block[dimension];
   int ent = 0;
 
-  static Vector<std::pair<int,int> > table; table.resize(e1*e2);
+  if(Cshift_table.size()<e1*e2) Cshift_table.resize(e1*e2); // Let it grow to biggest
+
   int stride=rhs.Grid()->_slice_stride[dimension];
 
-  auto rhs_v = rhs.View();
   if ( cbmask == 0x3 ) { 
     for(int n=0;n<e1;n++){
       for(int b=0;b<e2;b++){
 	int o  = n*stride;
 	int bo = n*e2;
-	table[ent++] = std::pair<int,int>(off+bo+b,so+o+b);
+	Cshift_table[ent++] = std::pair<int,int>(off+bo+b,so+o+b);
       }
     }
   } else { 
@@ -65,14 +67,19 @@ Gather_plane_simple (const Lattice<vobj> &rhs,commVector<vobj> &buffer,int dimen
 	 int o  = n*stride;
 	 int ocb=1<<rhs.Grid()->CheckerBoardFromOindex(o+b);
 	 if ( ocb &cbmask ) {
-	   table[ent++]=std::pair<int,int> (off+bo++,so+o+b);
+	   Cshift_table[ent++]=std::pair<int,int> (off+bo++,so+o+b);
 	 }
        }
      }
   }
-  thread_for(i,ent,{
-    buffer[table[i].first]=rhs_v[table[i].second];
-  });
+  {
+    autoView(rhs_v , rhs, AcceleratorRead);
+    auto buffer_p = & buffer[0];
+    auto table = &Cshift_table[0];
+    accelerator_for(i,ent,1,{
+      buffer_p[table[i].first]=rhs_v[table[i].second];
+    });
+  }
 }
 
 ///////////////////////////////////////////////////////////////////
@@ -95,36 +102,38 @@ Gather_plane_extract(const Lattice<vobj> &rhs,
   int e2=rhs.Grid()->_slice_block[dimension];
   int n1=rhs.Grid()->_slice_stride[dimension];
 
-  auto rhs_v = rhs.View();
   if ( cbmask ==0x3){
-    thread_for_collapse(2,n,e1,{
-      for(int b=0;b<e2;b++){
-
+    autoView(rhs_v , rhs, AcceleratorRead);
+    accelerator_for2d(n,e1,b,e2,1,{
 	int o      =   n*n1;
 	int offset = b+n*e2;
 	
 	vobj temp =rhs_v[so+o+b];
 	extract<vobj>(temp,pointers,offset);
-      }
-    });
+      });
   } else { 
+    autoView(rhs_v , rhs, AcceleratorRead);
 
-    // Case of SIMD split AND checker dim cannot currently be hit, except in 
-    // Test_cshift_red_black code.
-    std::cout << " Dense packed buffer WARNING " <<std::endl;
-    thread_for_collapse(2,n,e1,{
-      for(int b=0;b<e2;b++){
+    Coordinate rdim=rhs.Grid()->_rdimensions;
+    Coordinate cdm =rhs.Grid()->_checker_dim_mask;
+    std::cout << " Dense packed buffer WARNING " <<std::endl; // Does this get called twice once for each cb?
+    accelerator_for2d(n,e1,b,e2,1,{
+
+	Coordinate coor;
 
 	int o=n*n1;
-	int ocb=1<<rhs.Grid()->CheckerBoardFromOindex(o+b);
+	int oindex = o+b;
+
+       	int cb = RedBlackCheckerBoardFromOindex(oindex, rdim, cdm);
+
+	int ocb=1<<cb;
 	int offset = b+n*e2;
 
 	if ( ocb & cbmask ) {
 	  vobj temp =rhs_v[so+o+b];
 	  extract<vobj>(temp,pointers,offset);
 	}
-      }
-    });
+      });
   }
 }
 
@@ -145,7 +154,8 @@ template<class vobj> void Scatter_plane_simple (Lattice<vobj> &rhs,commVector<vo
   int e2=rhs.Grid()->_slice_block[dimension];
   int stride=rhs.Grid()->_slice_stride[dimension];
 
-  static std::vector<std::pair<int,int> > table; table.resize(e1*e2);
+  if(Cshift_table.size()<e1*e2) Cshift_table.resize(e1*e2); // Let it grow to biggest
+
   int ent    =0;
 
   if ( cbmask ==0x3 ) {
@@ -154,7 +164,7 @@ template<class vobj> void Scatter_plane_simple (Lattice<vobj> &rhs,commVector<vo
       for(int b=0;b<e2;b++){
 	int o   =n*rhs.Grid()->_slice_stride[dimension];
 	int bo  =n*rhs.Grid()->_slice_block[dimension];
-	table[ent++] = std::pair<int,int>(so+o+b,bo+b);
+	Cshift_table[ent++] = std::pair<int,int>(so+o+b,bo+b);
       }
     }
 
@@ -165,16 +175,20 @@ template<class vobj> void Scatter_plane_simple (Lattice<vobj> &rhs,commVector<vo
 	int o   =n*rhs.Grid()->_slice_stride[dimension];
 	int ocb=1<<rhs.Grid()->CheckerBoardFromOindex(o+b);// Could easily be a table lookup
 	if ( ocb & cbmask ) {
-	  table[ent++]=std::pair<int,int> (so+o+b,bo++);
+	  Cshift_table[ent++]=std::pair<int,int> (so+o+b,bo++);
 	}
       }
     }
   }
   
-  auto rhs_v = rhs.View();
-  thread_for(i,ent,{
-    rhs_v[table[i].first]=buffer[table[i].second];
-  });
+  {
+    autoView( rhs_v, rhs, AcceleratorWrite);
+    auto buffer_p = & buffer[0];
+    auto table = &Cshift_table[0];
+    accelerator_for(i,ent,1,{
+	rhs_v[table[i].first]=buffer_p[table[i].second];
+    });
+  }
 }
 
 //////////////////////////////////////////////////////
@@ -194,21 +208,19 @@ template<class vobj> void Scatter_plane_merge(Lattice<vobj> &rhs,ExtractPointerA
   int e2=rhs.Grid()->_slice_block[dimension];
 
   if(cbmask ==0x3 ) {
-    auto rhs_v = rhs.View();
-    thread_for_collapse(2,n,e1,{
-      for(int b=0;b<e2;b++){
+    autoView( rhs_v , rhs, AcceleratorWrite);
+    accelerator_for2d(n,e1,b,e2,1,{
 	int o      = n*rhs.Grid()->_slice_stride[dimension];
 	int offset = b+n*rhs.Grid()->_slice_block[dimension];
 	merge(rhs_v[so+o+b],pointers,offset);
-      }
-    });
+      });
   } else { 
 
     // Case of SIMD split AND checker dim cannot currently be hit, except in 
     // Test_cshift_red_black code.
     //    std::cout << "Scatter_plane merge assert(0); think this is buggy FIXME "<< std::endl;// think this is buggy FIXME
     std::cout<<" Unthreaded warning -- buffer is not densely packed ??"<<std::endl;
-    auto rhs_v = rhs.View();
+    autoView( rhs_v, rhs, CpuWrite);
     for(int n=0;n<e1;n++){
       for(int b=0;b<e2;b++){
 	int o      = n*rhs.Grid()->_slice_stride[dimension];
@@ -225,6 +237,7 @@ template<class vobj> void Scatter_plane_merge(Lattice<vobj> &rhs,ExtractPointerA
 //////////////////////////////////////////////////////
 // local to node block strided copies
 //////////////////////////////////////////////////////
+
 template<class vobj> void Copy_plane(Lattice<vobj>& lhs,const Lattice<vobj> &rhs, int dimension,int lplane,int rplane,int cbmask)
 {
   int rd = rhs.Grid()->_rdimensions[dimension];
@@ -239,14 +252,16 @@ template<class vobj> void Copy_plane(Lattice<vobj>& lhs,const Lattice<vobj> &rhs
   int e1=rhs.Grid()->_slice_nblock[dimension]; // clearly loop invariant for icpc
   int e2=rhs.Grid()->_slice_block[dimension];
   int stride = rhs.Grid()->_slice_stride[dimension];
-  static std::vector<std::pair<int,int> > table; table.resize(e1*e2);
+
+  if(Cshift_table.size()<e1*e2) Cshift_table.resize(e1*e2); // Let it grow to biggest
+
   int ent=0;
 
   if(cbmask == 0x3 ){
     for(int n=0;n<e1;n++){
       for(int b=0;b<e2;b++){
         int o =n*stride+b;
-	table[ent++] = std::pair<int,int>(lo+o,ro+o);
+	Cshift_table[ent++] = std::pair<int,int>(lo+o,ro+o);
       }
     }
   } else { 
@@ -255,23 +270,24 @@ template<class vobj> void Copy_plane(Lattice<vobj>& lhs,const Lattice<vobj> &rhs
         int o =n*stride+b;
         int ocb=1<<lhs.Grid()->CheckerBoardFromOindex(o);
         if ( ocb&cbmask ) {
-	  table[ent++] = std::pair<int,int>(lo+o,ro+o);
+	  Cshift_table[ent++] = std::pair<int,int>(lo+o,ro+o);
 	}
       }
     }
   }
 
-  auto rhs_v = rhs.View();
-  auto lhs_v = lhs.View();
-  thread_for(i,ent,{
-    lhs_v[table[i].first]=rhs_v[table[i].second];
-  });
-
+  {
+    autoView(rhs_v , rhs, AcceleratorRead);
+    autoView(lhs_v , lhs, AcceleratorWrite);
+    auto table = &Cshift_table[0];
+    accelerator_for(i,ent,1,{
+      lhs_v[table[i].first]=rhs_v[table[i].second];
+    });
+  }
 }
 
 template<class vobj> void Copy_plane_permute(Lattice<vobj>& lhs,const Lattice<vobj> &rhs, int dimension,int lplane,int rplane,int cbmask,int permute_type)
 {
- 
   int rd = rhs.Grid()->_rdimensions[dimension];
 
   if ( !rhs.Grid()->CheckerBoarded(dimension) ) {
@@ -285,29 +301,33 @@ template<class vobj> void Copy_plane_permute(Lattice<vobj>& lhs,const Lattice<vo
   int e2=rhs.Grid()->_slice_block [dimension];
   int stride = rhs.Grid()->_slice_stride[dimension];
 
-  static std::vector<std::pair<int,int> > table;  table.resize(e1*e2);
+  if(Cshift_table.size()<e1*e2) Cshift_table.resize(e1*e2); // Let it grow to biggest
+
   int ent=0;
 
   if ( cbmask == 0x3 ) {
     for(int n=0;n<e1;n++){
     for(int b=0;b<e2;b++){
       int o  =n*stride;
-      table[ent++] = std::pair<int,int>(lo+o+b,ro+o+b);
+      Cshift_table[ent++] = std::pair<int,int>(lo+o+b,ro+o+b);
     }}
   } else {
     for(int n=0;n<e1;n++){
     for(int b=0;b<e2;b++){
       int o  =n*stride;
       int ocb=1<<lhs.Grid()->CheckerBoardFromOindex(o+b);
-      if ( ocb&cbmask ) table[ent++] = std::pair<int,int>(lo+o+b,ro+o+b);
+      if ( ocb&cbmask ) Cshift_table[ent++] = std::pair<int,int>(lo+o+b,ro+o+b);
     }}
   }
 
-  auto rhs_v = rhs.View();
-  auto lhs_v = lhs.View();
-  thread_for(i,ent,{
-    permute(lhs_v[table[i].first],rhs_v[table[i].second],permute_type);
-  });
+  {
+    autoView( rhs_v, rhs, AcceleratorRead);
+    autoView( lhs_v, lhs, AcceleratorWrite);
+    auto table = &Cshift_table[0];
+    accelerator_for(i,ent,1,{
+      permute(lhs_v[table[i].first],rhs_v[table[i].second],permute_type);
+    });
+  }
 }
 
 //////////////////////////////////////////////////////
diff --git a/Grid/cshift/Cshift_table.cc b/Grid/cshift/Cshift_table.cc
new file mode 100644
index 00000000..d46e51c0
--- /dev/null
+++ b/Grid/cshift/Cshift_table.cc
@@ -0,0 +1,4 @@
+#include <Grid/GridCore.h>       
+NAMESPACE_BEGIN(Grid);
+Vector<std::pair<int,int> > Cshift_table; 
+NAMESPACE_END(Grid);
diff --git a/Grid/lattice/Lattice.h b/Grid/lattice/Lattice.h
index 1eea98ed..a3017198 100644
--- a/Grid/lattice/Lattice.h
+++ b/Grid/lattice/Lattice.h
@@ -26,6 +26,7 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
     *************************************************************************************/
     /*  END LEGAL */
 #pragma once
+#include <Grid/lattice/Lattice_view.h>
 #include <Grid/lattice/Lattice_base.h>
 #include <Grid/lattice/Lattice_conformable.h>
 #include <Grid/lattice/Lattice_ET.h>
@@ -35,7 +36,7 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 #include <Grid/lattice/Lattice_local.h>
 #include <Grid/lattice/Lattice_reduction.h>
 #include <Grid/lattice/Lattice_peekpoke.h>
-#include <Grid/lattice/Lattice_reality.h>
+//#include <Grid/lattice/Lattice_reality.h>
 #include <Grid/lattice/Lattice_comparison_utils.h>
 #include <Grid/lattice/Lattice_comparison.h>
 #include <Grid/lattice/Lattice_coordinate.h>
@@ -43,4 +44,4 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 #include <Grid/lattice/Lattice_rng.h>
 #include <Grid/lattice/Lattice_unary.h>
 #include <Grid/lattice/Lattice_transfer.h>
-
+#include <Grid/lattice/Lattice_basis.h>
diff --git a/Grid/lattice/Lattice_ET.h b/Grid/lattice/Lattice_ET.h
index cf7147b9..91b456d9 100644
--- a/Grid/lattice/Lattice_ET.h
+++ b/Grid/lattice/Lattice_ET.h
@@ -9,6 +9,7 @@ Copyright (C) 2015
 Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
 Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 Author: neo <cossu@post.kek.jp>
+Author: Christoph Lehner <christoph@lhnr.de
 
 This program is free software; you can redistribute it and/or modify
 it under the terms of the GNU General Public License as published by
@@ -91,12 +92,18 @@ const lobj & eval(const uint64_t ss, const LatticeView<lobj> &arg)
 {
   return arg[ss];
 }
+
+// What needs this?
+// Cannot be legal on accelerator
+// Comparison must convert
+#if 1
 template <class lobj> accelerator_inline 
 const lobj & eval(const uint64_t ss, const Lattice<lobj> &arg) 
 {
-  auto view = arg.View();
+  auto view = arg.View(AcceleratorRead);
   return view[ss];
 }
+#endif
 
 ///////////////////////////////////////////////////
 // handle nodes in syntax tree- eval one operand
@@ -179,16 +186,12 @@ inline void CBFromExpression(int &cb, const T1 &lat)  // Lattice leaf
   cb = lat.Checkerboard();
 }
 template <class T1,typename std::enable_if<!is_lattice<T1>::value, T1>::type * = nullptr>
-inline void CBFromExpression(int &cb, const T1 &notlat)  // non-lattice leaf
-{
-}
-
+inline void CBFromExpression(int &cb, const T1 &notlat) {} // non-lattice leaf
 template <typename Op, typename T1> inline 
 void CBFromExpression(int &cb,const LatticeUnaryExpression<Op, T1> &expr) 
 {
   CBFromExpression(cb, expr.arg1);  // recurse AST
 }
-
 template <typename Op, typename T1, typename T2> inline 
 void CBFromExpression(int &cb,const LatticeBinaryExpression<Op, T1, T2> &expr) 
 {
@@ -203,6 +206,68 @@ inline void CBFromExpression(int &cb, const LatticeTrinaryExpression<Op, T1, T2,
   CBFromExpression(cb, expr.arg3);  // recurse AST
 }
 
+
+//////////////////////////////////////////////////////////////////////////
+// ViewOpen
+//////////////////////////////////////////////////////////////////////////
+template <class T1,typename std::enable_if<is_lattice<T1>::value, T1>::type * = nullptr>
+inline void ExpressionViewOpen(T1 &lat)  // Lattice leaf
+{
+  lat.ViewOpen(AcceleratorRead);
+}
+template <class T1,typename std::enable_if<!is_lattice<T1>::value, T1>::type * = nullptr>
+  inline void ExpressionViewOpen(T1 &notlat) {}
+
+template <typename Op, typename T1> inline 
+void ExpressionViewOpen(LatticeUnaryExpression<Op, T1> &expr) 
+{  
+  ExpressionViewOpen(expr.arg1); // recurse AST
+}
+
+template <typename Op, typename T1, typename T2> inline 
+void ExpressionViewOpen(LatticeBinaryExpression<Op, T1, T2> &expr) 
+{
+  ExpressionViewOpen(expr.arg1);  // recurse AST
+  ExpressionViewOpen(expr.arg2);  // recurse AST
+}
+template <typename Op, typename T1, typename T2, typename T3>
+inline void ExpressionViewOpen(LatticeTrinaryExpression<Op, T1, T2, T3> &expr) 
+{
+  ExpressionViewOpen(expr.arg1);  // recurse AST
+  ExpressionViewOpen(expr.arg2);  // recurse AST
+  ExpressionViewOpen(expr.arg3);  // recurse AST
+}
+
+//////////////////////////////////////////////////////////////////////////
+// ViewClose
+//////////////////////////////////////////////////////////////////////////
+template <class T1,typename std::enable_if<is_lattice<T1>::value, T1>::type * = nullptr>
+inline void ExpressionViewClose( T1 &lat)  // Lattice leaf
+{
+  lat.ViewClose();
+}
+template <class T1,typename std::enable_if<!is_lattice<T1>::value, T1>::type * = nullptr>
+inline void ExpressionViewClose(T1 &notlat) {}
+
+template <typename Op, typename T1> inline 
+void ExpressionViewClose(LatticeUnaryExpression<Op, T1> &expr) 
+{  
+  ExpressionViewClose(expr.arg1); // recurse AST
+}
+template <typename Op, typename T1, typename T2> inline 
+void ExpressionViewClose(LatticeBinaryExpression<Op, T1, T2> &expr) 
+{
+  ExpressionViewClose(expr.arg1);  // recurse AST
+  ExpressionViewClose(expr.arg2);  // recurse AST
+}
+template <typename Op, typename T1, typename T2, typename T3>
+inline void ExpressionViewClose(LatticeTrinaryExpression<Op, T1, T2, T3> &expr) 
+{
+  ExpressionViewClose(expr.arg1);  // recurse AST
+  ExpressionViewClose(expr.arg2);  // recurse AST
+  ExpressionViewClose(expr.arg3);  // recurse AST
+}
+
 ////////////////////////////////////////////
 // Unary operators and funcs
 ////////////////////////////////////////////
diff --git a/Grid/lattice/Lattice_arith.h b/Grid/lattice/Lattice_arith.h
index 3543d6aa..a3ae1f28 100644
--- a/Grid/lattice/Lattice_arith.h
+++ b/Grid/lattice/Lattice_arith.h
@@ -7,6 +7,7 @@
     Copyright (C) 2015
 
 Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+Author: Christoph Lehner <christoph@lhnr.de>
 
     This program is free software; you can redistribute it and/or modify
     it under the terms of the GNU General Public License as published by
@@ -36,9 +37,9 @@ NAMESPACE_BEGIN(Grid);
 template<class obj1,class obj2,class obj3> inline
 void mult(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const Lattice<obj3> &rhs){
   ret.Checkerboard() = lhs.Checkerboard();
-  auto ret_v = ret.View();
-  auto lhs_v = lhs.View();
-  auto rhs_v = rhs.View();
+  autoView( ret_v , ret, AcceleratorWrite);
+  autoView( lhs_v , lhs, AcceleratorRead);
+  autoView( rhs_v , rhs, AcceleratorRead);
   conformable(ret,rhs);
   conformable(lhs,rhs);
   accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
@@ -55,9 +56,9 @@ void mac(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const Lattice<obj3> &rhs){
   ret.Checkerboard() = lhs.Checkerboard();
   conformable(ret,rhs);
   conformable(lhs,rhs);
-  auto ret_v = ret.View();
-  auto lhs_v = lhs.View();
-  auto rhs_v = rhs.View();
+  autoView( ret_v , ret, AcceleratorWrite);
+  autoView( lhs_v , lhs, AcceleratorRead);
+  autoView( rhs_v , rhs, AcceleratorRead);
   accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
     decltype(coalescedRead(obj1())) tmp;
     auto lhs_t=lhs_v(ss);
@@ -72,9 +73,9 @@ void sub(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const Lattice<obj3> &rhs){
   ret.Checkerboard() = lhs.Checkerboard();
   conformable(ret,rhs);
   conformable(lhs,rhs);
-  auto ret_v = ret.View();
-  auto lhs_v = lhs.View();
-  auto rhs_v = rhs.View();
+  autoView( ret_v , ret, AcceleratorWrite);
+  autoView( lhs_v , lhs, AcceleratorRead);
+  autoView( rhs_v , rhs, AcceleratorRead);
   accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
     decltype(coalescedRead(obj1())) tmp;
     auto lhs_t=lhs_v(ss);
@@ -88,9 +89,9 @@ void add(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const Lattice<obj3> &rhs){
   ret.Checkerboard() = lhs.Checkerboard();
   conformable(ret,rhs);
   conformable(lhs,rhs);
-  auto ret_v = ret.View();
-  auto lhs_v = lhs.View();
-  auto rhs_v = rhs.View();
+  autoView( ret_v , ret, AcceleratorWrite);
+  autoView( lhs_v , lhs, AcceleratorRead);
+  autoView( rhs_v , rhs, AcceleratorRead);
   accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
     decltype(coalescedRead(obj1())) tmp;
     auto lhs_t=lhs_v(ss);
@@ -107,8 +108,8 @@ template<class obj1,class obj2,class obj3> inline
 void mult(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
   ret.Checkerboard() = lhs.Checkerboard();
   conformable(lhs,ret);
-  auto ret_v = ret.View();
-  auto lhs_v = lhs.View();
+  autoView( ret_v , ret, AcceleratorWrite);
+  autoView( lhs_v , lhs, AcceleratorRead);
   accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
     decltype(coalescedRead(obj1())) tmp;
     mult(&tmp,&lhs_v(ss),&rhs);
@@ -120,8 +121,8 @@ template<class obj1,class obj2,class obj3> inline
 void mac(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
   ret.Checkerboard() = lhs.Checkerboard();
   conformable(ret,lhs);
-  auto ret_v = ret.View();
-  auto lhs_v = lhs.View();
+  autoView( ret_v , ret, AcceleratorWrite);
+  autoView( lhs_v , lhs, AcceleratorRead);
   accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
     decltype(coalescedRead(obj1())) tmp;
     auto lhs_t=lhs_v(ss);
@@ -134,8 +135,8 @@ template<class obj1,class obj2,class obj3> inline
 void sub(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
   ret.Checkerboard() = lhs.Checkerboard();
   conformable(ret,lhs);
-  auto ret_v = ret.View();
-  auto lhs_v = lhs.View();
+  autoView( ret_v , ret, AcceleratorWrite);
+  autoView( lhs_v , lhs, AcceleratorRead);
   accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
     decltype(coalescedRead(obj1())) tmp;
     auto lhs_t=lhs_v(ss);
@@ -147,8 +148,8 @@ template<class obj1,class obj2,class obj3> inline
 void add(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
   ret.Checkerboard() = lhs.Checkerboard();
   conformable(lhs,ret);
-  auto ret_v = ret.View();
-  auto lhs_v = lhs.View();
+  autoView( ret_v , ret, AcceleratorWrite);
+  autoView( lhs_v , lhs, AcceleratorRead);
   accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
     decltype(coalescedRead(obj1())) tmp;
     auto lhs_t=lhs_v(ss);
@@ -164,8 +165,8 @@ template<class obj1,class obj2,class obj3> inline
 void mult(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
   ret.Checkerboard() = rhs.Checkerboard();
   conformable(ret,rhs);
-  auto ret_v = ret.View();
-  auto rhs_v = lhs.View();
+  autoView( ret_v , ret, AcceleratorWrite);
+  autoView( rhs_v , lhs, AcceleratorRead);
   accelerator_for(ss,rhs_v.size(),obj1::Nsimd(),{
     decltype(coalescedRead(obj1())) tmp;
     auto rhs_t=rhs_v(ss);
@@ -178,8 +179,8 @@ template<class obj1,class obj2,class obj3> inline
 void mac(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
   ret.Checkerboard() = rhs.Checkerboard();
   conformable(ret,rhs);
-  auto ret_v = ret.View();
-  auto rhs_v = lhs.View();
+  autoView( ret_v , ret, AcceleratorWrite);
+  autoView( rhs_v , lhs, AcceleratorRead);
   accelerator_for(ss,rhs_v.size(),obj1::Nsimd(),{
     decltype(coalescedRead(obj1())) tmp;
     auto rhs_t=rhs_v(ss);
@@ -192,8 +193,8 @@ template<class obj1,class obj2,class obj3> inline
 void sub(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
   ret.Checkerboard() = rhs.Checkerboard();
   conformable(ret,rhs);
-  auto ret_v = ret.View();
-  auto rhs_v = lhs.View();
+  autoView( ret_v , ret, AcceleratorWrite);
+  autoView( rhs_v , lhs, AcceleratorRead);
   accelerator_for(ss,rhs_v.size(),obj1::Nsimd(),{
     decltype(coalescedRead(obj1())) tmp;
     auto rhs_t=rhs_v(ss);
@@ -205,8 +206,8 @@ template<class obj1,class obj2,class obj3> inline
 void add(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
   ret.Checkerboard() = rhs.Checkerboard();
   conformable(ret,rhs);
-  auto ret_v = ret.View();
-  auto rhs_v = lhs.View();
+  autoView( ret_v , ret, AcceleratorWrite);
+  autoView( rhs_v , lhs, AcceleratorRead);
   accelerator_for(ss,rhs_v.size(),obj1::Nsimd(),{
     decltype(coalescedRead(obj1())) tmp;
     auto rhs_t=rhs_v(ss);
@@ -220,9 +221,9 @@ void axpy(Lattice<vobj> &ret,sobj a,const Lattice<vobj> &x,const Lattice<vobj> &
   ret.Checkerboard() = x.Checkerboard();
   conformable(ret,x);
   conformable(x,y);
-  auto ret_v = ret.View();
-  auto x_v = x.View();
-  auto y_v = y.View();
+  autoView( ret_v , ret, AcceleratorWrite);
+  autoView( x_v , x, AcceleratorRead);
+  autoView( y_v , y, AcceleratorRead);
   accelerator_for(ss,x_v.size(),vobj::Nsimd(),{
     auto tmp = a*x_v(ss)+y_v(ss);
     coalescedWrite(ret_v[ss],tmp);
@@ -233,9 +234,9 @@ void axpby(Lattice<vobj> &ret,sobj a,sobj b,const Lattice<vobj> &x,const Lattice
   ret.Checkerboard() = x.Checkerboard();
   conformable(ret,x);
   conformable(x,y);
-  auto ret_v = ret.View();
-  auto x_v = x.View();
-  auto y_v = y.View();
+  autoView( ret_v , ret, AcceleratorWrite);
+  autoView( x_v , x, AcceleratorRead);
+  autoView( y_v , y, AcceleratorRead);
   accelerator_for(ss,x_v.size(),vobj::Nsimd(),{
     auto tmp = a*x_v(ss)+b*y_v(ss);
     coalescedWrite(ret_v[ss],tmp);
diff --git a/Grid/lattice/Lattice_base.h b/Grid/lattice/Lattice_base.h
index ec7c54ec..73b1b6a1 100644
--- a/Grid/lattice/Lattice_base.h
+++ b/Grid/lattice/Lattice_base.h
@@ -9,6 +9,7 @@ Copyright (C) 2015
 Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
 Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 Author: paboyle <paboyle@ph.ed.ac.uk>
+Author: Christoph Lehner <christoph@lhnr.de>
 
 This program is free software; you can redistribute it and/or modify
 it under the terms of the GNU General Public License as published by
@@ -28,6 +29,7 @@ See the full license in the file "LICENSE" in the top level distribution
 directory
 *************************************************************************************/
 			   /*  END LEGAL */
+
 #pragma once 
 
 #define STREAMING_STORES
@@ -36,129 +38,6 @@ NAMESPACE_BEGIN(Grid);
 
 extern int GridCshiftPermuteMap[4][16];
 
-///////////////////////////////////////////////////////////////////
-// Base class which can be used by traits to pick up behaviour
-///////////////////////////////////////////////////////////////////
-class LatticeBase {};
-
-/////////////////////////////////////////////////////////////////////////////////////////
-// Conformable checks; same instance of Grid required
-/////////////////////////////////////////////////////////////////////////////////////////
-void accelerator_inline conformable(GridBase *lhs,GridBase *rhs)
-{
-  assert(lhs == rhs);
-}
-
-////////////////////////////////////////////////////////////////////////////
-// Minimal base class containing only data valid to access from accelerator
-// _odata will be a managed pointer in CUDA
-////////////////////////////////////////////////////////////////////////////
-// Force access to lattice through a view object.
-// prevents writing of code that will not offload to GPU, but perhaps annoyingly
-// strict since host could could in principle direct access through the lattice object
-// Need to decide programming model.
-#define LATTICE_VIEW_STRICT
-template<class vobj> class LatticeAccelerator : public LatticeBase
-{
-protected:
-  GridBase *_grid;
-  int checkerboard;
-  vobj     *_odata;    // A managed pointer
-  uint64_t _odata_size;    
-public:
-  accelerator_inline LatticeAccelerator() : checkerboard(0), _odata(nullptr), _odata_size(0), _grid(nullptr) { }; 
-  accelerator_inline uint64_t oSites(void) const { return _odata_size; };
-  accelerator_inline int  Checkerboard(void) const { return checkerboard; };
-  accelerator_inline int &Checkerboard(void) { return this->checkerboard; }; // can assign checkerboard on a container, not a view
-  accelerator_inline void Conformable(GridBase * &grid) const
-  { 
-    if (grid) conformable(grid, _grid);
-    else      grid = _grid;
-  };
-};
-
-/////////////////////////////////////////////////////////////////////////////////////////
-// A View class which provides accessor to the data.
-// This will be safe to call from accelerator_for and is trivially copy constructible
-// The copy constructor for this will need to be used by device lambda functions
-/////////////////////////////////////////////////////////////////////////////////////////
-template<class vobj> 
-class LatticeView : public LatticeAccelerator<vobj>
-{
-public:
-
-
-  // Rvalue
-#ifdef __CUDA_ARCH__
-  accelerator_inline const typename vobj::scalar_object operator()(size_t i) const { return coalescedRead(this->_odata[i]); }
-#else 
-  accelerator_inline const vobj & operator()(size_t i) const { return this->_odata[i]; }
-#endif
-
-  accelerator_inline const vobj & operator[](size_t i) const { return this->_odata[i]; };
-  accelerator_inline vobj       & operator[](size_t i)       { return this->_odata[i]; };
-
-  accelerator_inline uint64_t begin(void) const { return 0;};
-  accelerator_inline uint64_t end(void)   const { return this->_odata_size; };
-  accelerator_inline uint64_t size(void)  const { return this->_odata_size; };
-
-  LatticeView(const LatticeAccelerator<vobj> &refer_to_me) : LatticeAccelerator<vobj> (refer_to_me)
-  {
-  }
-};
-
-/////////////////////////////////////////////////////////////////////////////////////////
-// Lattice expression types used by ET to assemble the AST
-// 
-// Need to be able to detect code paths according to the whether a lattice object or not
-// so introduce some trait type things
-/////////////////////////////////////////////////////////////////////////////////////////
-
-class LatticeExpressionBase {};
-
-template <typename T> using is_lattice = std::is_base_of<LatticeBase, T>;
-template <typename T> using is_lattice_expr = std::is_base_of<LatticeExpressionBase,T >;
-
-template<class T, bool isLattice> struct ViewMapBase { typedef T Type; };
-template<class T>                 struct ViewMapBase<T,true> { typedef LatticeView<typename T::vector_object> Type; };
-template<class T> using ViewMap = ViewMapBase<T,std::is_base_of<LatticeBase, T>::value >;
-
-template <typename Op, typename _T1>                           
-class LatticeUnaryExpression : public  LatticeExpressionBase 
-{
-public:
-  typedef typename ViewMap<_T1>::Type T1;
-  Op op;
-  T1 arg1;
-  LatticeUnaryExpression(Op _op,const _T1 &_arg1) : op(_op), arg1(_arg1) {};
-};
-
-template <typename Op, typename _T1, typename _T2>              
-class LatticeBinaryExpression : public LatticeExpressionBase 
-{
-public:
-  typedef typename ViewMap<_T1>::Type T1;
-  typedef typename ViewMap<_T2>::Type T2;
-  Op op;
-  T1 arg1;
-  T2 arg2;
-  LatticeBinaryExpression(Op _op,const _T1 &_arg1,const _T2 &_arg2) : op(_op), arg1(_arg1), arg2(_arg2) {};
-};
-
-template <typename Op, typename _T1, typename _T2, typename _T3> 
-class LatticeTrinaryExpression : public LatticeExpressionBase 
-{
-public:
-  typedef typename ViewMap<_T1>::Type T1;
-  typedef typename ViewMap<_T2>::Type T2;
-  typedef typename ViewMap<_T3>::Type T3;
-  Op op;
-  T1 arg1;
-  T2 arg2;
-  T3 arg3;
-  LatticeTrinaryExpression(Op _op,const _T1 &_arg1,const _T2 &_arg2,const _T3 &_arg3) : op(_op), arg1(_arg1), arg2(_arg2), arg3(_arg3) {};
-};
-
 /////////////////////////////////////////////////////////////////////////////////////////
 // The real lattice class, with normal copy and assignment semantics.
 // This contains extra (host resident) grid pointer data that may be accessed by host code
@@ -194,24 +73,33 @@ private:
       dealloc();
       
       this->_odata_size = size;
-      if ( size ) 
+      if ( size )
 	this->_odata      = alloc.allocate(this->_odata_size);
       else 
 	this->_odata      = nullptr;
     }
   }
 public:
+
+  /////////////////////////////////////////////////////////////////////////////////
+  // Can use to make accelerator dirty without copy from host ; useful for temporaries "dont care" prev contents
+  /////////////////////////////////////////////////////////////////////////////////
+  void SetViewMode(ViewMode mode) {
+    LatticeView<vobj> accessor(*( (LatticeAccelerator<vobj> *) this),mode);
+    accessor.ViewClose();
+  }
   /////////////////////////////////////////////////////////////////////////////////
   // Return a view object that may be dereferenced in site loops.
   // The view is trivially copy constructible and may be copied to an accelerator device
   // in device lambdas
   /////////////////////////////////////////////////////////////////////////////////
-  LatticeView<vobj> View (void) const 
+
+  LatticeView<vobj> View (ViewMode mode) const 
   {
-    LatticeView<vobj> accessor(*( (LatticeAccelerator<vobj> *) this));
+    LatticeView<vobj> accessor(*( (LatticeAccelerator<vobj> *) this),mode);
     return accessor;
   }
-  
+
   ~Lattice() { 
     if ( this->_odata_size ) {
       dealloc();
@@ -231,12 +119,16 @@ public:
     CBFromExpression(cb,expr);
     assert( (cb==Odd) || (cb==Even));
     this->checkerboard=cb;
-
-    auto me  = View();
+    
+    auto exprCopy = expr;
+    ExpressionViewOpen(exprCopy);
+    auto me  = View(AcceleratorWriteDiscard);
     accelerator_for(ss,me.size(),1,{
-      auto tmp = eval(ss,expr);
+      auto tmp = eval(ss,exprCopy);
       vstream(me[ss],tmp);
     });
+    me.ViewClose();
+    ExpressionViewClose(exprCopy);
     return *this;
   }
   template <typename Op, typename T1,typename T2> inline Lattice<vobj> & operator=(const LatticeBinaryExpression<Op,T1,T2> &expr)
@@ -251,11 +143,15 @@ public:
     assert( (cb==Odd) || (cb==Even));
     this->checkerboard=cb;
 
-    auto me  = View();
+    auto exprCopy = expr;
+    ExpressionViewOpen(exprCopy);
+    auto me  = View(AcceleratorWriteDiscard);
     accelerator_for(ss,me.size(),1,{
-      auto tmp = eval(ss,expr);
+      auto tmp = eval(ss,exprCopy);
       vstream(me[ss],tmp);
     });
+    me.ViewClose();
+    ExpressionViewClose(exprCopy);
     return *this;
   }
   template <typename Op, typename T1,typename T2,typename T3> inline Lattice<vobj> & operator=(const LatticeTrinaryExpression<Op,T1,T2,T3> &expr)
@@ -269,11 +165,15 @@ public:
     CBFromExpression(cb,expr);
     assert( (cb==Odd) || (cb==Even));
     this->checkerboard=cb;
-    auto me  = View();
+    auto exprCopy = expr;
+    ExpressionViewOpen(exprCopy);
+    auto me  = View(AcceleratorWriteDiscard);
     accelerator_for(ss,me.size(),1,{
-      auto tmp = eval(ss,expr);
+      auto tmp = eval(ss,exprCopy);
       vstream(me[ss],tmp);
     });
+    me.ViewClose();
+    ExpressionViewClose(exprCopy);
     return *this;
   }
   //GridFromExpression is tricky to do
@@ -324,10 +224,11 @@ public:
   }
 
   template<class sobj> inline Lattice<vobj> & operator = (const sobj & r){
-    auto me  = View();
+    auto me  = View(CpuWrite);
     thread_for(ss,me.size(),{
-      me[ss] = r;
+	me[ss]= r;
     });
+    me.ViewClose();
     return *this;
   }
 
@@ -337,11 +238,12 @@ public:
   ///////////////////////////////////////////
   // user defined constructor
   ///////////////////////////////////////////
-  Lattice(GridBase *grid) { 
+  Lattice(GridBase *grid,ViewMode mode=AcceleratorWriteDiscard) { 
     this->_grid = grid;
     resize(this->_grid->oSites());
     assert((((uint64_t)&this->_odata[0])&0xF) ==0);
     this->checkerboard=0;
+    SetViewMode(mode);
   }
   
   //  virtual ~Lattice(void) = default;
@@ -357,7 +259,6 @@ public:
   // copy constructor
   ///////////////////////////////////////////
   Lattice(const Lattice& r){ 
-    //    std::cout << "Lattice constructor(const Lattice &) "<<this<<std::endl; 
     this->_grid = r.Grid();
     resize(this->_grid->oSites());
     *this = r;
@@ -380,11 +281,12 @@ public:
     typename std::enable_if<!std::is_same<robj,vobj>::value,int>::type i=0;
     conformable(*this,r);
     this->checkerboard = r.Checkerboard();
-    auto me =   View();
-    auto him= r.View();
+    auto me =   View(AcceleratorWriteDiscard);
+    auto him= r.View(AcceleratorRead);
     accelerator_for(ss,me.size(),vobj::Nsimd(),{
       coalescedWrite(me[ss],him(ss));
     });
+    me.ViewClose();    him.ViewClose();
     return *this;
   }
 
@@ -394,11 +296,12 @@ public:
   inline Lattice<vobj> & operator = (const Lattice<vobj> & r){
     this->checkerboard = r.Checkerboard();
     conformable(*this,r);
-    auto me =   View();
-    auto him= r.View();
+    auto me =   View(AcceleratorWriteDiscard);
+    auto him= r.View(AcceleratorRead);
     accelerator_for(ss,me.size(),vobj::Nsimd(),{
       coalescedWrite(me[ss],him(ss));
     });
+    me.ViewClose();    him.ViewClose();
     return *this;
   }
   ///////////////////////////////////////////
diff --git a/Grid/lattice/Lattice_basis.h b/Grid/lattice/Lattice_basis.h
new file mode 100644
index 00000000..9f1155eb
--- /dev/null
+++ b/Grid/lattice/Lattice_basis.h
@@ -0,0 +1,226 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid
+
+Source file: ./lib/lattice/Lattice_basis.h
+
+Copyright (C) 2015
+
+Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+Author: paboyle <paboyle@ph.ed.ac.uk>
+Author: Christoph Lehner <christoph@lhnr.de>
+
+This program is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 2 of the License, or
+(at your option) any later version.
+
+This program is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along
+with this program; if not, write to the Free Software Foundation, Inc.,
+51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+See the full license in the file "LICENSE" in the top level distribution
+directory
+*************************************************************************************/
+			   /*  END LEGAL */
+
+#pragma once
+
+NAMESPACE_BEGIN(Grid);
+
+template<class Field>
+void basisOrthogonalize(std::vector<Field> &basis,Field &w,int k) 
+{
+  // If assume basis[j] are already orthonormal,
+  // can take all inner products in parallel saving 2x bandwidth
+  // Save 3x bandwidth on the second line of loop.
+  // perhaps 2.5x speed up.
+  // 2x overall in Multigrid Lanczos  
+  for(int j=0; j<k; ++j){
+    auto ip = innerProduct(basis[j],w);
+    w = w - ip*basis[j];
+  }
+}
+
+template<class VField, class Matrix>
+void basisRotate(VField &basis,Matrix& Qt,int j0, int j1, int k0,int k1,int Nm) 
+{
+  typedef decltype(basis[0]) Field;
+  typedef decltype(basis[0].View(AcceleratorRead)) View;
+
+  Vector<View> basis_v; basis_v.reserve(basis.size());
+  GridBase* grid = basis[0].Grid();
+      
+  for(int k=0;k<basis.size();k++){
+    basis_v.push_back(basis[k].View(AcceleratorWrite));
+  }
+
+
+  View *basis_vp = &basis_v[0];
+
+  int nrot = j1-j0;
+  if (!nrot) // edge case not handled gracefully by Cuda
+    return;
+
+  uint64_t oSites   =grid->oSites();
+  uint64_t siteBlock=(grid->oSites()+nrot-1)/nrot; // Maximum 1 additional vector overhead
+
+  typedef typename std::remove_reference<decltype(basis_v[0][0])>::type vobj;
+
+  Vector <vobj> Bt(siteBlock * nrot); 
+  auto Bp=&Bt[0];
+
+  // GPU readable copy of matrix
+  Vector<double> Qt_jv(Nm*Nm);
+  double *Qt_p = & Qt_jv[0];
+  thread_for(i,Nm*Nm,{
+      int j = i/Nm;
+      int k = i%Nm;
+      Qt_p[i]=Qt(j,k);
+  });
+
+  // Block the loop to keep storage footprint down
+  for(uint64_t s=0;s<oSites;s+=siteBlock){
+
+    // remaining work in this block
+    int ssites=MIN(siteBlock,oSites-s);
+
+    // zero out the accumulators
+    accelerator_for(ss,siteBlock*nrot,vobj::Nsimd(),{
+	decltype(coalescedRead(Bp[ss])) z;
+	z=Zero();
+	coalescedWrite(Bp[ss],z);
+      });
+
+    accelerator_for(sj,ssites*nrot,vobj::Nsimd(),{
+	
+	int j =sj%nrot;
+	int jj  =j0+j;
+	int ss =sj/nrot;
+	int sss=ss+s;
+
+	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_v[k][sss]));
+	}
+      });
+
+    accelerator_for(sj,ssites*nrot,vobj::Nsimd(),{
+	int j =sj%nrot;
+	int jj  =j0+j;
+	int ss =sj/nrot;
+	int sss=ss+s;
+	coalescedWrite(basis_v[jj][sss],coalescedRead(Bp[ss*nrot+j]));
+      });
+  }
+
+  for(int k=0;k<basis.size();k++) basis_v[k].ViewClose();
+}
+
+// Extract a single rotated vector
+template<class Field>
+void basisRotateJ(Field &result,std::vector<Field> &basis,Eigen::MatrixXd& Qt,int j, int k0,int k1,int Nm) 
+{
+  typedef decltype(basis[0].View(AcceleratorRead)) View;
+  typedef typename Field::vector_object vobj;
+  GridBase* grid = basis[0].Grid();
+
+  result.Checkerboard() = basis[0].Checkerboard();
+
+  Vector<View> basis_v; basis_v.reserve(basis.size());
+  for(int k=0;k<basis.size();k++){
+    basis_v.push_back(basis[k].View(AcceleratorRead));
+  }
+  vobj zz=Zero();
+  Vector<double> Qt_jv(Nm);
+  double * Qt_j = & Qt_jv[0];
+  for(int k=0;k<Nm;++k) Qt_j[k]=Qt(j,k);
+
+  autoView(result_v,result,AcceleratorWrite);
+  accelerator_for(ss, grid->oSites(),vobj::Nsimd(),{
+    auto B=coalescedRead(zz);
+    for(int k=k0; k<k1; ++k){
+      B +=Qt_j[k] * coalescedRead(basis_v[k][ss]);
+    }
+    coalescedWrite(result_v[ss], B);
+  });
+  for(int k=0;k<basis.size();k++) basis_v[k].ViewClose();
+}
+
+template<class Field>
+void basisReorderInPlace(std::vector<Field> &_v,std::vector<RealD>& sort_vals, std::vector<int>& idx) 
+{
+  int vlen = idx.size();
+
+  assert(vlen>=1);
+  assert(vlen<=sort_vals.size());
+  assert(vlen<=_v.size());
+
+  for (size_t i=0;i<vlen;i++) {
+
+    if (idx[i] != i) {
+
+      //////////////////////////////////////
+      // idx[i] is a table of desired sources giving a permutation.
+      // Swap v[i] with v[idx[i]].
+      // Find  j>i for which _vnew[j] = _vold[i],
+      // track the move idx[j] => idx[i]
+      // track the move idx[i] => i
+      //////////////////////////////////////
+      size_t j;
+      for (j=i;j<idx.size();j++)
+	if (idx[j]==i)
+	  break;
+
+      assert(idx[i] > i);     assert(j!=idx.size());      assert(idx[j]==i);
+
+      swap(_v[i],_v[idx[i]]); // should use vector move constructor, no data copy
+      std::swap(sort_vals[i],sort_vals[idx[i]]);
+
+      idx[j] = idx[i];
+      idx[i] = i;
+    }
+  }
+}
+
+inline std::vector<int> basisSortGetIndex(std::vector<RealD>& sort_vals) 
+{
+  std::vector<int> idx(sort_vals.size());
+  std::iota(idx.begin(), idx.end(), 0);
+
+  // sort indexes based on comparing values in v
+  std::sort(idx.begin(), idx.end(), [&sort_vals](int i1, int i2) {
+    return ::fabs(sort_vals[i1]) < ::fabs(sort_vals[i2]);
+  });
+  return idx;
+}
+
+template<class Field>
+void basisSortInPlace(std::vector<Field> & _v,std::vector<RealD>& sort_vals, bool reverse) 
+{
+  std::vector<int> idx = basisSortGetIndex(sort_vals);
+  if (reverse)
+    std::reverse(idx.begin(), idx.end());
+  
+  basisReorderInPlace(_v,sort_vals,idx);
+}
+
+// PAB: faster to compute the inner products first then fuse loops.
+// If performance critical can improve.
+template<class Field>
+void basisDeflate(const std::vector<Field> &_v,const std::vector<RealD>& eval,const Field& src_orig,Field& result) {
+  result = Zero();
+  assert(_v.size()==eval.size());
+  int N = (int)_v.size();
+  for (int i=0;i<N;i++) {
+    Field& tmp = _v[i];
+    axpy(result,TensorRemove(innerProduct(tmp,src_orig)) / eval[i],tmp,result);
+  }
+}
+
+NAMESPACE_END(Grid);
diff --git a/Grid/lattice/Lattice_comparison.h b/Grid/lattice/Lattice_comparison.h
index bbed2ef5..6a29be94 100644
--- a/Grid/lattice/Lattice_comparison.h
+++ b/Grid/lattice/Lattice_comparison.h
@@ -78,9 +78,9 @@ template<class vfunctor,class lobj,class robj>
 inline Lattice<vPredicate> LLComparison(vfunctor op,const Lattice<lobj> &lhs,const Lattice<robj> &rhs)
 {
   Lattice<vPredicate> ret(rhs.Grid());
-  auto lhs_v = lhs.View();
-  auto rhs_v = rhs.View();
-  auto ret_v = ret.View();
+  autoView( lhs_v, lhs, CpuRead);
+  autoView( rhs_v, rhs, CpuRead);
+  autoView( ret_v, ret, CpuWrite);
   thread_for( ss, rhs_v.size(), {
       ret_v[ss]=op(lhs_v[ss],rhs_v[ss]);
   });
@@ -93,8 +93,8 @@ template<class vfunctor,class lobj,class robj>
 inline Lattice<vPredicate> LSComparison(vfunctor op,const Lattice<lobj> &lhs,const robj &rhs)
 {
   Lattice<vPredicate> ret(lhs.Grid());
-  auto lhs_v = lhs.View();
-  auto ret_v = ret.View();
+  autoView( lhs_v, lhs, CpuRead);
+  autoView( ret_v, ret, CpuWrite);
   thread_for( ss, lhs_v.size(), {
     ret_v[ss]=op(lhs_v[ss],rhs);
   });
@@ -107,8 +107,8 @@ template<class vfunctor,class lobj,class robj>
 inline Lattice<vPredicate> SLComparison(vfunctor op,const lobj &lhs,const Lattice<robj> &rhs)
 {
   Lattice<vPredicate> ret(rhs.Grid());
-  auto rhs_v = rhs.View();
-  auto ret_v = ret.View();
+  autoView( rhs_v, rhs, CpuRead);
+  autoView( ret_v, ret, CpuWrite);
   thread_for( ss, rhs_v.size(), {
     ret_v[ss]=op(lhs,rhs_v[ss]);
   });
diff --git a/Grid/lattice/Lattice_coordinate.h b/Grid/lattice/Lattice_coordinate.h
index a1abe58d..cd0f11ee 100644
--- a/Grid/lattice/Lattice_coordinate.h
+++ b/Grid/lattice/Lattice_coordinate.h
@@ -37,7 +37,7 @@ template<class iobj> inline void LatticeCoordinate(Lattice<iobj> &l,int mu)
   GridBase *grid = l.Grid();
   int Nsimd = grid->iSites();
 
-  auto l_v = l.View();
+  autoView(l_v, l, CpuWrite);
   thread_for( o, grid->oSites(), {
     vector_type vI;
     Coordinate gcoor;
@@ -51,23 +51,5 @@ template<class iobj> inline void LatticeCoordinate(Lattice<iobj> &l,int mu)
   });
 };
 
-// LatticeCoordinate();
-// FIXME for debug; deprecate this; made obscelete by 
-template<class vobj> void lex_sites(Lattice<vobj> &l){
-  auto l_v = l.View();
-  Real *v_ptr = (Real *)&l_v[0];
-  size_t o_len = l.Grid()->oSites();
-  size_t v_len = sizeof(vobj)/sizeof(vRealF);
-  size_t vec_len = vRealF::Nsimd();
-
-  for(int i=0;i<o_len;i++){
-    for(int j=0;j<v_len;j++){
-      for(int vv=0;vv<vec_len;vv+=2){
-	v_ptr[i*v_len*vec_len+j*vec_len+vv  ]= i+vv*500;
-	v_ptr[i*v_len*vec_len+j*vec_len+vv+1]= i+vv*500;
-      }
-    }}
-}
-
 NAMESPACE_END(Grid);
 
diff --git a/Grid/lattice/Lattice_local.h b/Grid/lattice/Lattice_local.h
index 99455e24..1b31e9b3 100644
--- a/Grid/lattice/Lattice_local.h
+++ b/Grid/lattice/Lattice_local.h
@@ -43,8 +43,8 @@ template<class vobj>
 inline auto localNorm2 (const Lattice<vobj> &rhs)-> Lattice<typename vobj::tensor_reduced>
 {
   Lattice<typename vobj::tensor_reduced> ret(rhs.Grid());
-  auto rhs_v = rhs.View();
-  auto ret_v = ret.View();
+  autoView( rhs_v , rhs, AcceleratorRead);
+  autoView( ret_v , ret, AcceleratorWrite);
   accelerator_for(ss,rhs_v.size(),vobj::Nsimd(),{
     coalescedWrite(ret_v[ss],innerProduct(rhs_v(ss),rhs_v(ss)));
   });
@@ -56,9 +56,9 @@ template<class vobj>
 inline auto localInnerProduct (const Lattice<vobj> &lhs,const Lattice<vobj> &rhs) -> Lattice<typename vobj::tensor_reduced>
 {
   Lattice<typename vobj::tensor_reduced> ret(rhs.Grid());
-  auto lhs_v = lhs.View();
-  auto rhs_v = rhs.View();
-  auto ret_v = ret.View();
+  autoView( lhs_v , lhs, AcceleratorRead);
+  autoView( rhs_v , rhs, AcceleratorRead);
+  autoView( ret_v , ret, AcceleratorWrite);
   accelerator_for(ss,rhs_v.size(),vobj::Nsimd(),{
     coalescedWrite(ret_v[ss],innerProduct(lhs_v(ss),rhs_v(ss)));
   });
@@ -73,9 +73,9 @@ inline auto outerProduct (const Lattice<ll> &lhs,const Lattice<rr> &rhs) -> Latt
   typedef decltype(coalescedRead(ll())) sll;
   typedef decltype(coalescedRead(rr())) srr;
   Lattice<decltype(outerProduct(ll(),rr()))> ret(rhs.Grid());
-  auto lhs_v = lhs.View();
-  auto rhs_v = rhs.View();
-  auto ret_v = ret.View();
+  autoView( lhs_v , lhs, AcceleratorRead);
+  autoView( rhs_v , rhs, AcceleratorRead);
+  autoView( ret_v , ret, AcceleratorWrite);
   accelerator_for(ss,rhs_v.size(),1,{
     // FIXME had issues with scalar version of outer 
     // Use vector [] operator and don't read coalesce this loop
diff --git a/Grid/lattice/Lattice_matrix_reduction.h b/Grid/lattice/Lattice_matrix_reduction.h
index 0980ad8a..7c470fef 100644
--- a/Grid/lattice/Lattice_matrix_reduction.h
+++ b/Grid/lattice/Lattice_matrix_reduction.h
@@ -51,9 +51,9 @@ static void sliceMaddMatrix (Lattice<vobj> &R,Eigen::MatrixXcd &aa,const Lattice
   int block =FullGrid->_slice_block [Orthog];
   int nblock=FullGrid->_slice_nblock[Orthog];
   int ostride=FullGrid->_ostride[Orthog];
-  auto X_v = X.View();
-  auto Y_v = Y.View();
-  auto R_v = R.View();
+  autoView( X_v , X, CpuRead);
+  autoView( Y_v , Y, CpuRead);
+  autoView( R_v , R, CpuWrite);
   thread_region
   {
     std::vector<vobj> s_x(Nblock);
@@ -97,8 +97,8 @@ static void sliceMulMatrix (Lattice<vobj> &R,Eigen::MatrixXcd &aa,const Lattice<
   int nblock=FullGrid->_slice_nblock[Orthog];
   int ostride=FullGrid->_ostride[Orthog];
 
-  auto X_v = X.View();
-  auto R_v = R.View();
+  autoView( X_v , X, CpuRead);
+  autoView( R_v , R, CpuWrite);
 
   thread_region
   {
@@ -156,8 +156,8 @@ static void sliceInnerProductMatrix(  Eigen::MatrixXcd &mat, const Lattice<vobj>
   int ostride=FullGrid->_ostride[Orthog];
 
   typedef typename vobj::vector_typeD vector_typeD;
-  auto lhs_v = lhs.View();
-  auto rhs_v = rhs.View();
+  autoView( lhs_v , lhs, CpuRead);
+  autoView( rhs_v , rhs, CpuRead);
   thread_region {
     std::vector<vobj> Left(Nblock);
     std::vector<vobj> Right(Nblock);
diff --git a/Grid/lattice/Lattice_peekpoke.h b/Grid/lattice/Lattice_peekpoke.h
index feca2f44..c79becf2 100644
--- a/Grid/lattice/Lattice_peekpoke.h
+++ b/Grid/lattice/Lattice_peekpoke.h
@@ -46,9 +46,9 @@ auto PeekIndex(const Lattice<vobj> &lhs,int i) -> Lattice<decltype(peekIndex<Ind
 {
   Lattice<decltype(peekIndex<Index>(vobj(),i))> ret(lhs.Grid());
   ret.Checkerboard()=lhs.Checkerboard();
-  auto ret_v = ret.View();
-  auto lhs_v = lhs.View();
-  thread_for( ss, lhs_v.size(), {
+  autoView( ret_v, ret, AcceleratorWrite);
+  autoView( lhs_v, lhs, AcceleratorRead);
+  accelerator_for( ss, lhs_v.size(), 1, {
     ret_v[ss] = peekIndex<Index>(lhs_v[ss],i);
   });
   return ret;
@@ -58,9 +58,9 @@ auto PeekIndex(const Lattice<vobj> &lhs,int i,int j) -> Lattice<decltype(peekInd
 {
   Lattice<decltype(peekIndex<Index>(vobj(),i,j))> ret(lhs.Grid());
   ret.Checkerboard()=lhs.Checkerboard();
-  auto ret_v = ret.View();
-  auto lhs_v = lhs.View();
-  thread_for( ss, lhs_v.size(), {
+  autoView( ret_v, ret, AcceleratorWrite);
+  autoView( lhs_v, lhs, AcceleratorRead);
+  accelerator_for( ss, lhs_v.size(), 1, {
     ret_v[ss] = peekIndex<Index>(lhs_v[ss],i,j);
   });
   return ret;
@@ -72,18 +72,18 @@ auto PeekIndex(const Lattice<vobj> &lhs,int i,int j) -> Lattice<decltype(peekInd
 template<int Index,class vobj>  
 void PokeIndex(Lattice<vobj> &lhs,const Lattice<decltype(peekIndex<Index>(vobj(),0))> & rhs,int i)
 {
-  auto rhs_v = rhs.View();
-  auto lhs_v = lhs.View();
-  thread_for( ss, lhs_v.size(), {
+  autoView( rhs_v, rhs, AcceleratorRead);
+  autoView( lhs_v, lhs, AcceleratorWrite);
+  accelerator_for( ss, lhs_v.size(), 1, {
     pokeIndex<Index>(lhs_v[ss],rhs_v[ss],i);
   });
 }
 template<int Index,class vobj> 
 void PokeIndex(Lattice<vobj> &lhs,const Lattice<decltype(peekIndex<Index>(vobj(),0,0))> & rhs,int i,int j)
 {
-  auto rhs_v = rhs.View();
-  auto lhs_v = lhs.View();
-  thread_for( ss, lhs_v.size(), {
+  autoView( rhs_v, rhs, AcceleratorRead);
+  autoView( lhs_v, lhs, AcceleratorWrite);
+  accelerator_for( ss, lhs_v.size(), 1, {
     pokeIndex<Index>(lhs_v[ss],rhs_v[ss],i,j);
   });
 }
@@ -111,7 +111,7 @@ void pokeSite(const sobj &s,Lattice<vobj> &l,const Coordinate &site){
 
   // extract-modify-merge cycle is easiest way and this is not perf critical
   ExtractBuffer<sobj> buf(Nsimd);
-  auto l_v = l.View();
+  autoView( l_v , l, CpuWrite);
   if ( rank == grid->ThisRank() ) {
     extract(l_v[odx],buf);
     buf[idx] = s;
@@ -141,7 +141,7 @@ void peekSite(sobj &s,const Lattice<vobj> &l,const Coordinate &site){
   grid->GlobalCoorToRankIndex(rank,odx,idx,site);
 
   ExtractBuffer<sobj> buf(Nsimd);
-  auto l_v = l.View();
+  autoView( l_v , l, CpuWrite);
   extract(l_v[odx],buf);
 
   s = buf[idx];
@@ -151,21 +151,21 @@ void peekSite(sobj &s,const Lattice<vobj> &l,const Coordinate &site){
   return;
 };
 
-
 //////////////////////////////////////////////////////////
 // Peek a scalar object from the SIMD array
 //////////////////////////////////////////////////////////
+// Must be CPU read view
 template<class vobj,class sobj>
-accelerator_inline void peekLocalSite(sobj &s,const Lattice<vobj> &l,Coordinate &site){
-        
-  GridBase *grid = l.Grid();
-
+inline void peekLocalSite(sobj &s,const LatticeView<vobj> &l,Coordinate &site)
+{
+  GridBase *grid = l.getGrid();
+  assert(l.mode==CpuRead);
   typedef typename vobj::scalar_type scalar_type;
   typedef typename vobj::vector_type vector_type;
 
   int Nsimd = grid->Nsimd();
 
-  assert( l.Checkerboard()== l.Grid()->CheckerBoard(site));
+  assert( l.Checkerboard()== grid->CheckerBoard(site));
   assert( sizeof(sobj)*Nsimd == sizeof(vobj));
 
   static const int words=sizeof(vobj)/sizeof(vector_type);
@@ -173,8 +173,7 @@ accelerator_inline void peekLocalSite(sobj &s,const Lattice<vobj> &l,Coordinate
   idx= grid->iIndex(site);
   odx= grid->oIndex(site);
   
-  auto l_v = l.View();
-  scalar_type * vp = (scalar_type *)&l_v[odx];
+  scalar_type * vp = (scalar_type *)&l[odx];
   scalar_type * pt = (scalar_type *)&s;
       
   for(int w=0;w<words;w++){
@@ -183,18 +182,19 @@ accelerator_inline void peekLocalSite(sobj &s,const Lattice<vobj> &l,Coordinate
       
   return;
 };
-
+// Must be CPU write view
 template<class vobj,class sobj>
-accelerator_inline void pokeLocalSite(const sobj &s,Lattice<vobj> &l,Coordinate &site){
-
-  GridBase *grid=l.Grid();
+inline void pokeLocalSite(const sobj &s,LatticeView<vobj> &l,Coordinate &site)
+{
+  GridBase *grid=l.getGrid();
+  assert(l.mode==CpuWrite);
 
   typedef typename vobj::scalar_type scalar_type;
   typedef typename vobj::vector_type vector_type;
 
   int Nsimd = grid->Nsimd();
 
-  assert( l.Checkerboard()== l.Grid()->CheckerBoard(site));
+  assert( l.Checkerboard()== grid->CheckerBoard(site));
   assert( sizeof(sobj)*Nsimd == sizeof(vobj));
 
   static const int words=sizeof(vobj)/sizeof(vector_type);
@@ -202,13 +202,11 @@ accelerator_inline void pokeLocalSite(const sobj &s,Lattice<vobj> &l,Coordinate
   idx= grid->iIndex(site);
   odx= grid->oIndex(site);
 
-  auto l_v = l.View();
-  scalar_type * vp = (scalar_type *)&l_v[odx];
+  scalar_type * vp = (scalar_type *)&l[odx];
   scalar_type * pt = (scalar_type *)&s;
   for(int w=0;w<words;w++){
     vp[idx+w*Nsimd] = pt[w];
   }
-
   return;
 };
 
diff --git a/Grid/lattice/Lattice_reality.h b/Grid/lattice/Lattice_reality.h
index 7373b2f9..61491d6b 100644
--- a/Grid/lattice/Lattice_reality.h
+++ b/Grid/lattice/Lattice_reality.h
@@ -40,8 +40,11 @@ NAMESPACE_BEGIN(Grid);
 
 template<class vobj> inline Lattice<vobj> adj(const Lattice<vobj> &lhs){
   Lattice<vobj> ret(lhs.Grid());
-  auto lhs_v = lhs.View();
-  auto ret_v = ret.View();
+
+  autoView( lhs_v, lhs, AcceleratorRead);
+  autoView( ret_v, ret, AcceleratorWrite);
+
+  ret.Checkerboard()=lhs.Checkerboard();
   accelerator_for( ss, lhs_v.size(), vobj::Nsimd(), {
     coalescedWrite(ret_v[ss], adj(lhs_v(ss)));
   });
@@ -50,8 +53,11 @@ template<class vobj> inline Lattice<vobj> adj(const Lattice<vobj> &lhs){
 
 template<class vobj> inline Lattice<vobj> conjugate(const Lattice<vobj> &lhs){
   Lattice<vobj> ret(lhs.Grid());
-  auto lhs_v = lhs.View();
-  auto ret_v = ret.View();
+
+  autoView( lhs_v, lhs, AcceleratorRead);
+  autoView( ret_v, ret, AcceleratorWrite);
+
+  ret.Checkerboard() = lhs.Checkerboard();
   accelerator_for( ss, lhs_v.size(), vobj::Nsimd(), {
     coalescedWrite( ret_v[ss] , conjugate(lhs_v(ss)));
   });
diff --git a/Grid/lattice/Lattice_reduction.h b/Grid/lattice/Lattice_reduction.h
index 3c5b03e5..c2955485 100644
--- a/Grid/lattice/Lattice_reduction.h
+++ b/Grid/lattice/Lattice_reduction.h
@@ -5,6 +5,7 @@
 Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
 Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 Author: paboyle <paboyle@ph.ed.ac.uk>
+Author: Christoph Lehner <christoph@lhnr.de>
     This program is free software; you can redistribute it and/or modify
     it under the terms of the GNU General Public License as published by
     the Free Software Foundation; either version 2 of the License, or
@@ -24,7 +25,7 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 #include <Grid/Grid_Eigen_Dense.h>
 
 
-#ifdef GRID_NVCC
+#if defined(GRID_CUDA)||defined(GRID_HIP)
 #include <Grid/lattice/Lattice_reduction_gpu.h>
 #endif
 
@@ -38,7 +39,36 @@ inline typename vobj::scalar_object sum_cpu(const vobj *arg, Integer osites)
 {
   typedef typename vobj::scalar_object  sobj;
 
-  const int Nsimd = vobj::Nsimd();
+  //  const int Nsimd = vobj::Nsimd();
+  const int nthread = GridThread::GetThreads();
+
+  Vector<sobj> sumarray(nthread);
+  for(int i=0;i<nthread;i++){
+    sumarray[i]=Zero();
+  }
+  
+  thread_for(thr,nthread, {
+    int nwork, mywork, myoff;
+    nwork = osites;
+    GridThread::GetWork(nwork,thr,mywork,myoff);
+    vobj vvsum=Zero();
+    for(int ss=myoff;ss<mywork+myoff; ss++){
+      vvsum = vvsum + arg[ss];
+    }
+    sumarray[thr]=Reduce(vvsum);
+  });
+  
+  sobj ssum=Zero();  // sum across threads
+  for(int i=0;i<nthread;i++){
+    ssum = ssum+sumarray[i];
+  } 
+  return ssum;
+}
+template<class vobj>
+inline typename vobj::scalar_objectD sumD_cpu(const vobj *arg, Integer osites)
+{
+  typedef typename vobj::scalar_objectD  sobj;
+
   const int nthread = GridThread::GetThreads();
 
   Vector<sobj> sumarray(nthread);
@@ -62,23 +92,43 @@ inline typename vobj::scalar_object sum_cpu(const vobj *arg, Integer osites)
     ssum = ssum+sumarray[i];
   } 
   
-  return ssum;
+  typedef typename vobj::scalar_object ssobj;
+  ssobj ret = ssum;
+  return ret;
 }
+
+
 template<class vobj>
 inline typename vobj::scalar_object sum(const vobj *arg, Integer osites)
 {
-#ifdef GRID_NVCC
+#if defined(GRID_CUDA)||defined(GRID_HIP)
   return sum_gpu(arg,osites);
 #else
   return sum_cpu(arg,osites);
 #endif  
 }
+template<class vobj>
+inline typename vobj::scalar_objectD sumD(const vobj *arg, Integer osites)
+{
+#if defined(GRID_CUDA)||defined(GRID_HIP)
+  return sumD_gpu(arg,osites);
+#else
+  return sumD_cpu(arg,osites);
+#endif  
+}
+
 template<class vobj>
 inline typename vobj::scalar_object sum(const Lattice<vobj> &arg)
 {
-  auto arg_v = arg.View();
+#if defined(GRID_CUDA)||defined(GRID_HIP)
+  autoView( arg_v, arg, AcceleratorRead);
   Integer osites = arg.Grid()->oSites();
-  auto ssum= sum(&arg_v[0],osites);
+  auto ssum= sum_gpu(&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;
 }
@@ -93,55 +143,49 @@ template<class vobj> inline RealD norm2(const Lattice<vobj> &arg){
 
 // Double inner product
 template<class vobj>
-inline ComplexD innerProduct(const Lattice<vobj> &left,const Lattice<vobj> &right)
+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;
   
   GridBase *grid = left.Grid();
-  
-  // Might make all code paths go this way.
-  auto left_v = left.View();
-  auto right_v=right.View();
 
   const uint64_t nsimd = grid->Nsimd();
   const uint64_t sites = grid->oSites();
   
-#ifdef GRID_NVCC
-  // GPU - SIMT lane compliance...
-  typedef decltype(innerProduct(left_v[0],right_v[0])) inner_t;
+  // Might make all code paths go this way.
+  typedef decltype(innerProductD(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);
 
-  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));
-  })
+    // GPU - SIMT lane compliance...
+    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);
+    });
+  }
 
   // This is in single precision and fails some tests
-  // Need a sumD that sums in double
-  nrm = TensorRemove(sumD_gpu(inner_tmp_v,sites));  
-#else
-  // CPU 
-  typedef decltype(innerProductD(left_v[0],right_v[0])) inner_t;
-  Vector<inner_t> inner_tmp(sites);
-  auto inner_tmp_v = &inner_tmp[0];
-  
-  accelerator_for( ss, sites, nsimd,{
-      auto x_l = left_v[ss];
-      auto y_l = right_v[ss];
-      inner_tmp_v[ss]=innerProductD(x_l,y_l);
-  })
-  nrm = TensorRemove(sum(inner_tmp_v,sites));
-#endif
-  grid->GlobalSum(nrm);
-
+  auto anrm = sum(inner_tmp_v,sites);  
+  nrm = anrm;
   return nrm;
 }
 
+template<class vobj>
+inline ComplexD innerProduct(const Lattice<vobj> &left,const Lattice<vobj> &right) {
+  GridBase *grid = left.Grid();
+  ComplexD nrm = rankInnerProduct(left,right);
+  grid->GlobalSum(nrm);
+  return nrm;
+}
+
+
 /////////////////////////
 // Fast axpby_norm
 // z = a x + b y
@@ -167,45 +211,67 @@ axpby_norm_fast(Lattice<vobj> &z,sobj a,sobj b,const Lattice<vobj> &x,const Latt
   
   GridBase *grid = x.Grid();
 
-  auto x_v=x.View();
-  auto y_v=y.View();
-  auto z_v=z.View();
-
   const uint64_t nsimd = grid->Nsimd();
   const uint64_t sites = grid->oSites();
   
-#ifdef GRID_NVCC
   // GPU
-  typedef decltype(innerProduct(x_v[0],y_v[0])) inner_t;
-  Vector<inner_t> inner_tmp(sites);
-  auto inner_tmp_v = &inner_tmp[0];
+  autoView( x_v, x, AcceleratorRead);
+  autoView( y_v, y, AcceleratorRead);
+  autoView( z_v, z, AcceleratorWrite);
 
-  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_gpu(inner_tmp_v,sites)));
-#else
-  // CPU 
   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);
+
+  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;
   });
-  // Already promoted to double
   nrm = real(TensorRemove(sum(inner_tmp_v,sites)));
-#endif
   grid->GlobalSum(nrm);
   return nrm; 
 }
-
  
+template<class vobj> strong_inline void
+innerProductNorm(ComplexD& ip, RealD &nrm, const Lattice<vobj> &left,const Lattice<vobj> &right)
+{
+  conformable(left,right);
+
+  typedef typename vobj::scalar_type scalar_type;
+  typedef typename vobj::vector_typeD vector_type;
+  Vector<ComplexD> tmp(2);
+
+  GridBase *grid = left.Grid();
+
+  const uint64_t nsimd = grid->Nsimd();
+  const uint64_t sites = grid->oSites();
+
+  // GPU
+  typedef decltype(innerProductD(vobj(),vobj())) inner_t;
+  typedef decltype(innerProductD(vobj(),vobj())) norm_t;
+  Vector<inner_t> inner_tmp(sites);
+  Vector<norm_t>  norm_tmp(sites);
+  auto inner_tmp_v = &inner_tmp[0];
+  auto norm_tmp_v = &norm_tmp[0];
+  {
+    autoView(left_v,left, AcceleratorRead);
+    autoView(right_v,right,AcceleratorRead);
+    accelerator_for( ss, sites, 1,{
+	auto left_tmp = left_v[ss];
+	inner_tmp_v[ss]=innerProductD(left_tmp,right_v[ss]);
+        norm_tmp_v [ss]=innerProductD(left_tmp,left_tmp);
+      });
+  }
+
+  tmp[0] = TensorRemove(sum(inner_tmp_v,sites));
+  tmp[1] = TensorRemove(sum(norm_tmp_v,sites));
+
+  grid->GlobalSumVector(&tmp[0],2); // keep norm Complex -> can use GlobalSumVector
+  ip = tmp[0];
+  nrm = real(tmp[1]);
+}
+
 template<class Op,class T1>
 inline auto sum(const LatticeUnaryExpression<Op,T1> & expr)
   ->typename decltype(expr.op.func(eval(0,expr.arg1)))::scalar_object
@@ -271,7 +337,7 @@ template<class vobj> inline void sliceSum(const Lattice<vobj> &Data,std::vector<
 
   // sum over reduced dimension planes, breaking out orthog dir
   // Parallel over orthog direction
-  auto Data_v=Data.View();
+  autoView( Data_v, Data, CpuRead);
   thread_for( r,rd, {
     int so=r*grid->_ostride[orthogdim]; // base offset for start of plane 
     for(int n=0;n<e1;n++){
@@ -349,8 +415,8 @@ static void sliceInnerProductVector( std::vector<ComplexD> & result, const Latti
   int e2=    grid->_slice_block [orthogdim];
   int stride=grid->_slice_stride[orthogdim];
 
-  auto lhv=lhs.View();
-  auto rhv=rhs.View();
+  autoView( lhv, lhs, CpuRead);
+  autoView( rhv, rhs, CpuRead);
   thread_for( r,rd,{
 
     int so=r*grid->_ostride[orthogdim]; // base offset for start of plane 
@@ -457,14 +523,12 @@ static void sliceMaddVector(Lattice<vobj> &R,std::vector<RealD> &a,const Lattice
 
     tensor_reduced at; at=av;
 
-    auto Rv=R.View();
-    auto Xv=X.View();
-    auto Yv=Y.View();
-    thread_for_collapse(2, n, e1, {
-      for(int b=0;b<e2;b++){
+    autoView( Rv, R, CpuWrite);
+    autoView( Xv, X, CpuRead);
+    autoView( Yv, Y, CpuRead);
+    thread_for2d( n, e1, b,e2, {
 	int ss= so+n*stride+b;
 	Rv[ss] = at*Xv[ss]+Yv[ss];
-      }
     });
   }
 };
@@ -517,9 +581,9 @@ static void sliceMaddMatrix (Lattice<vobj> &R,Eigen::MatrixXcd &aa,const Lattice
   int nblock=FullGrid->_slice_nblock[Orthog];
   int ostride=FullGrid->_ostride[Orthog];
 
-  auto X_v=X.View();
-  auto Y_v=Y.View();
-  auto R_v=R.View();
+  autoView( X_v, X, CpuRead);
+  autoView( Y_v, Y, CpuRead);
+  autoView( R_v, R, CpuWrite);
   thread_region
   {
     Vector<vobj> s_x(Nblock);
@@ -564,13 +628,14 @@ static void sliceMulMatrix (Lattice<vobj> &R,Eigen::MatrixXcd &aa,const Lattice<
   //  int nl=1;
 
   //FIXME package in a convenient iterator
+  // thread_for2d_in_region
   //Should loop over a plane orthogonal to direction "Orthog"
   int stride=FullGrid->_slice_stride[Orthog];
   int block =FullGrid->_slice_block [Orthog];
   int nblock=FullGrid->_slice_nblock[Orthog];
   int ostride=FullGrid->_ostride[Orthog];
-  auto R_v = R.View();
-  auto X_v = X.View();
+  autoView( R_v, R, CpuWrite);
+  autoView( X_v, X, CpuRead);
   thread_region
   {
     std::vector<vobj> s_x(Nblock);
@@ -628,8 +693,8 @@ static void sliceInnerProductMatrix(  Eigen::MatrixXcd &mat, const Lattice<vobj>
 
   typedef typename vobj::vector_typeD vector_typeD;
 
-  auto lhs_v=lhs.View();
-  auto rhs_v=rhs.View();
+  autoView( lhs_v, lhs, CpuRead);
+  autoView( rhs_v, rhs, CpuRead);
   thread_region
   {
     std::vector<vobj> Left(Nblock);
diff --git a/Grid/lattice/Lattice_reduction_gpu.h b/Grid/lattice/Lattice_reduction_gpu.h
index c5d75356..5f490507 100644
--- a/Grid/lattice/Lattice_reduction_gpu.h
+++ b/Grid/lattice/Lattice_reduction_gpu.h
@@ -1,7 +1,13 @@
 NAMESPACE_BEGIN(Grid);
 
-#define WARP_SIZE 32
+#ifdef GRID_HIP
+extern hipDeviceProp_t *gpu_props;
+#endif
+#ifdef GRID_CUDA
 extern cudaDeviceProp *gpu_props;
+#endif
+
+#define WARP_SIZE 32
 __device__ unsigned int retirementCount = 0;
 
 template <class Iterator>
@@ -19,7 +25,12 @@ template <class Iterator>
 void getNumBlocksAndThreads(const Iterator n, const size_t sizeofsobj, Iterator &threads, Iterator &blocks) {
   
   int device;
+#ifdef GRID_CUDA
   cudaGetDevice(&device);
+#endif
+#ifdef GRID_HIP
+  hipGetDevice(&device);
+#endif
   
   Iterator warpSize            = gpu_props[device].warpSize;
   Iterator sharedMemPerBlock   = gpu_props[device].sharedMemPerBlock;
@@ -147,7 +158,7 @@ __global__ void reduceKernel(const vobj *lat, sobj *buffer, Iterator n) {
     sobj *smem = (sobj *)shmem_pointer;
     
     // wait until all outstanding memory instructions in this thread are finished
-    __threadfence();
+    acceleratorFence();
     
     if (tid==0) {
       unsigned int ticket = atomicInc(&retirementCount, gridDim.x);
@@ -156,8 +167,8 @@ __global__ void reduceKernel(const vobj *lat, sobj *buffer, Iterator n) {
     }
     
     // each thread must read the correct value of amLast
-    __syncthreads();
-    
+    acceleratorSynchroniseAll();
+
     if (amLast) {
       // reduce buffer[0], ..., buffer[gridDim.x-1]
       Iterator i = tid;
@@ -199,13 +210,7 @@ inline typename vobj::scalar_objectD sumD_gpu(const vobj *lat, Integer osites)
   sobj *buffer_v = &buffer[0];
   
   reduceKernel<<< numBlocks, numThreads, smemSize >>>(lat, buffer_v, size);
-  cudaDeviceSynchronize();
-  
-  cudaError err = cudaGetLastError();
-  if ( cudaSuccess != err ) {
-    printf("Cuda error %s\n",cudaGetErrorString( err ));
-    exit(0);
-  }
+  accelerator_barrier();
   auto result = buffer_v[0];
   return result;
 }
diff --git a/Grid/lattice/Lattice_rng.h b/Grid/lattice/Lattice_rng.h
index 1bb1f087..e5e63716 100644
--- a/Grid/lattice/Lattice_rng.h
+++ b/Grid/lattice/Lattice_rng.h
@@ -375,7 +375,7 @@ public:
     int osites = _grid->oSites();  // guaranteed to be <= l.Grid()->oSites() by a factor multiplicity
     int words  = sizeof(scalar_object) / sizeof(scalar_type);
 
-    auto l_v = l.View();
+    autoView(l_v, l, CpuWrite);
     thread_for( ss, osites, {
       ExtractBuffer<scalar_object> buf(Nsimd);
       for (int m = 0; m < multiplicity; m++) {  // Draw from same generator multiplicity times
@@ -461,8 +461,8 @@ public:
     }
 
     {
-      // Obtain one reseeded generator per thread
-      int Nthread = GridThread::GetThreads();
+      // Obtain one reseeded generator per thread      
+      int Nthread = 32; // Hardwire a good level or parallelism
       std::vector<RngEngine> seeders(Nthread);
       for(int t=0;t<Nthread;t++){
 	seeders[t] = Reseed(master_engine);
diff --git a/Grid/lattice/Lattice_trace.h b/Grid/lattice/Lattice_trace.h
index 93444e0c..5fe5173d 100644
--- a/Grid/lattice/Lattice_trace.h
+++ b/Grid/lattice/Lattice_trace.h
@@ -37,17 +37,19 @@ NAMESPACE_BEGIN(Grid);
 ////////////////////////////////////////////////////////////////////////////////////////////////////
 // Trace
 ////////////////////////////////////////////////////////////////////////////////////////////////////
+/*
 template<class vobj>
 inline auto trace(const Lattice<vobj> &lhs)  -> Lattice<decltype(trace(vobj()))>
 {
   Lattice<decltype(trace(vobj()))> ret(lhs.Grid());
-  auto ret_v = ret.View();
-  auto lhs_v = lhs.View();
+  autoView(ret_v , ret, AcceleratorWrite);
+  autoView(lhs_v , lhs, AcceleratorRead);
   accelerator_for( ss, lhs_v.size(), vobj::Nsimd(), {
     coalescedWrite(ret_v[ss], trace(lhs_v(ss)));
   });
   return ret;
 };
+*/
     
 ////////////////////////////////////////////////////////////////////////////////////////////////////
 // Trace Index level dependent operation
@@ -56,8 +58,8 @@ template<int Index,class vobj>
 inline auto TraceIndex(const Lattice<vobj> &lhs) -> Lattice<decltype(traceIndex<Index>(vobj()))>
 {
   Lattice<decltype(traceIndex<Index>(vobj()))> ret(lhs.Grid());
-  auto ret_v = ret.View();
-  auto lhs_v = lhs.View();
+  autoView( ret_v , ret, AcceleratorWrite);
+  autoView( lhs_v , lhs, AcceleratorRead);
   accelerator_for( ss, lhs_v.size(), vobj::Nsimd(), {
     coalescedWrite(ret_v[ss], traceIndex<Index>(lhs_v(ss)));
   });
diff --git a/Grid/lattice/Lattice_transfer.h b/Grid/lattice/Lattice_transfer.h
index c80e7db2..beceecc9 100644
--- a/Grid/lattice/Lattice_transfer.h
+++ b/Grid/lattice/Lattice_transfer.h
@@ -6,6 +6,7 @@
     Copyright (C) 2015
 
 Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+Author: Christoph Lehner <christoph@lhnr.de>
 
     This program is free software; you can redistribute it and/or modify
     it under the terms of the GNU General Public License as published by
@@ -46,11 +47,12 @@ inline void subdivides(GridBase *coarse,GridBase *fine)
 ////////////////////////////////////////////////////////////////////////////////////////////
 // remove and insert a half checkerboard
 ////////////////////////////////////////////////////////////////////////////////////////////
-template<class vobj> inline void pickCheckerboard(int cb,Lattice<vobj> &half,const Lattice<vobj> &full){
+template<class vobj> inline void pickCheckerboard(int cb,Lattice<vobj> &half,const Lattice<vobj> &full)
+{
   half.Checkerboard() = cb;
 
-  auto half_v = half.View();
-  auto full_v = full.View();
+  autoView( half_v, half, CpuWrite);
+  autoView( full_v, full, CpuRead);
   thread_for(ss, full.Grid()->oSites(),{
     int cbos;
     Coordinate coor;
@@ -63,10 +65,11 @@ template<class vobj> inline void pickCheckerboard(int cb,Lattice<vobj> &half,con
     }
   });
 }
-template<class vobj> inline void setCheckerboard(Lattice<vobj> &full,const Lattice<vobj> &half){
+template<class vobj> inline void setCheckerboard(Lattice<vobj> &full,const Lattice<vobj> &half)
+{
   int cb = half.Checkerboard();
-  auto half_v = half.View();
-  auto full_v = full.View();
+  autoView( half_v , half, CpuRead);
+  autoView( full_v , full, CpuWrite);
   thread_for(ss,full.Grid()->oSites(),{
 
     Coordinate coor;
@@ -81,96 +84,138 @@ template<class vobj> inline void setCheckerboard(Lattice<vobj> &full,const Latti
     }
   });
 }
-  
-template<class vobj,class CComplex,int nbasis>
-inline void blockProject(Lattice<iVector<CComplex,nbasis > > &coarseData,
-			  const             Lattice<vobj>   &fineData,
-			  const std::vector<Lattice<vobj> > &Basis)
-{
-  GridBase * fine  = fineData.Grid();
-  GridBase * coarse= coarseData.Grid();
 
-  Lattice<CComplex> ip(coarse); 
-
-  //  auto fineData_   = fineData.View();
-  auto coarseData_ = coarseData.View();
-  auto ip_         = ip.View();
-  for(int v=0;v<nbasis;v++) {
-    blockInnerProduct(ip,Basis[v],fineData);
-    accelerator_for( sc, coarse->oSites(), vobj::Nsimd(), {
-	coalescedWrite(coarseData_[sc](v),ip_(sc));
-      });
-  }
+////////////////////////////////////////////////////////////////////////////////////////////
+// Flexible Type Conversion for internal promotion to double as well as graceful
+// treatment of scalar-compatible types
+////////////////////////////////////////////////////////////////////////////////////////////
+accelerator_inline void convertType(ComplexD & out, const std::complex<double> & in) {
+  out = in;
 }
 
-template<class vobj,class CComplex,int nbasis>
-inline void blockProject1(Lattice<iVector<CComplex,nbasis > > &coarseData,
-			 const             Lattice<vobj>   &fineData,
-			 const std::vector<Lattice<vobj> > &Basis)
-{
-  typedef iVector<CComplex,nbasis > coarseSiteData;
-  coarseSiteData elide;
-  typedef decltype(coalescedRead(elide)) ScalarComplex;
-  GridBase * fine  = fineData.Grid();
-  GridBase * coarse= coarseData.Grid();
-  int  _ndimension = coarse->_ndimension;
+accelerator_inline void convertType(ComplexF & out, const std::complex<float> & in) {
+  out = in;
+}
 
-  // checks
-  assert( nbasis == Basis.size() );
-  subdivides(coarse,fine); 
-  for(int i=0;i<nbasis;i++){
-    conformable(Basis[i],fineData);
-  }
+#ifdef GRID_SIMT
+accelerator_inline void convertType(vComplexF & out, const ComplexF & in) {
+  ((ComplexF*)&out)[acceleratorSIMTlane(vComplexF::Nsimd())] = in;
+}
+accelerator_inline void convertType(vComplexD & out, const ComplexD & in) {
+  ((ComplexD*)&out)[acceleratorSIMTlane(vComplexD::Nsimd())] = in;
+}
+accelerator_inline void convertType(vComplexD2 & out, const ComplexD & in) {
+  ((ComplexD*)&out)[acceleratorSIMTlane(vComplexD::Nsimd()*2)] = in;
+}
+#endif
 
-  Coordinate block_r      (_ndimension);
-  
-  for(int d=0 ; d<_ndimension;d++){
-    block_r[d] = fine->_rdimensions[d] / coarse->_rdimensions[d];
-    assert(block_r[d]*coarse->_rdimensions[d] == fine->_rdimensions[d]);
-  }
-  int blockVol = fine->oSites()/coarse->oSites();
+accelerator_inline void convertType(vComplexF & out, const vComplexD2 & in) {
+  out.v = Optimization::PrecisionChange::DtoS(in._internal[0].v,in._internal[1].v);
+}
 
-  coarseData=Zero();
+accelerator_inline void convertType(vComplexD2 & out, const vComplexF & in) {
+  Optimization::PrecisionChange::StoD(in.v,out._internal[0].v,out._internal[1].v);
+}
 
-  auto fineData_   = fineData.View();
-  auto coarseData_ = coarseData.View();
-  ////////////////////////////////////////////////////////////////////////////////////////////////////////
-  // To make this lock free, loop over coars parallel, and then loop over fine associated with coarse.
-  // Otherwise do fine inner product per site, and make the update atomic
-  ////////////////////////////////////////////////////////////////////////////////////////////////////////
-  accelerator_for( sci, nbasis*coarse->oSites(), vobj::Nsimd(), {
+template<typename T1,typename T2,int N>
+  accelerator_inline void convertType(iMatrix<T1,N> & out, const iMatrix<T2,N> & in);
+template<typename T1,typename T2,int N>
+  accelerator_inline void convertType(iVector<T1,N> & out, const iVector<T2,N> & in);
 
-    auto sc=sci/nbasis;
-    auto i=sci%nbasis;
-    auto Basis_      = Basis[i].View();
+template<typename T1,typename T2, typename std::enable_if<!isGridScalar<T1>::value, T1>::type* = nullptr>
+accelerator_inline void convertType(T1 & out, const iScalar<T2> & in) {
+  convertType(out,in._internal);
+}
 
-    Coordinate coor_c(_ndimension);
-    Lexicographic::CoorFromIndex(coor_c,sc,coarse->_rdimensions);  // Block coordinate
+template<typename T1,typename T2>
+accelerator_inline void convertType(iScalar<T1> & out, const T2 & in) {
+  convertType(out._internal,in);
+}
 
-    int sf;
-    decltype(innerProduct(Basis_(sf),fineData_(sf))) reduce=Zero();
+template<typename T1,typename T2,int N>
+accelerator_inline void convertType(iMatrix<T1,N> & out, const iMatrix<T2,N> & in) {
+  for (int i=0;i<N;i++)
+    for (int j=0;j<N;j++)
+      convertType(out._internal[i][j],in._internal[i][j]);
+}
 
-    for(int sb=0;sb<blockVol;sb++){
+template<typename T1,typename T2,int N>
+accelerator_inline void convertType(iVector<T1,N> & out, const iVector<T2,N> & in) {
+  for (int i=0;i<N;i++)
+    convertType(out._internal[i],in._internal[i]);
+}
 
-      Coordinate coor_b(_ndimension);
-      Coordinate coor_f(_ndimension);
+template<typename T, typename std::enable_if<isGridFundamental<T>::value, T>::type* = nullptr>
+accelerator_inline void convertType(T & out, const T & in) {
+  out = in;
+}
 
-      Lexicographic::CoorFromIndex(coor_b,sb,block_r);
-      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);
-      
-      reduce=reduce+innerProduct(Basis_(sf),fineData_(sf));
-    }
-    coalescedWrite(coarseData_[sc](i),reduce);
+template<typename T1,typename T2>
+accelerator_inline void convertType(Lattice<T1> & out, const Lattice<T2> & in) {
+  autoView( out_v , out,AcceleratorWrite);
+  autoView( in_v  , in ,AcceleratorRead);
+  accelerator_for(ss,out_v.size(),T1::Nsimd(),{
+      convertType(out_v[ss],in_v(ss));
   });
-  return;
 }
 
-template<class vobj,class CComplex>
-inline void blockZAXPY(Lattice<vobj> &fineZ,
-		       const Lattice<CComplex> &coarseA,
-		       const Lattice<vobj> &fineX,
-		       const Lattice<vobj> &fineY)
+////////////////////////////////////////////////////////////////////////////////////////////
+// precision-promoted local inner product
+////////////////////////////////////////////////////////////////////////////////////////////
+template<class vobj>
+inline auto localInnerProductD(const Lattice<vobj> &lhs,const Lattice<vobj> &rhs)
+-> Lattice<iScalar<decltype(TensorRemove(innerProductD2(lhs.View(CpuRead)[0],rhs.View(CpuRead)[0])))>>
+{
+  autoView( lhs_v , lhs, AcceleratorRead);
+  autoView( rhs_v , rhs, AcceleratorRead);
+
+  typedef decltype(TensorRemove(innerProductD2(lhs_v[0],rhs_v[0]))) t_inner;
+  Lattice<iScalar<t_inner>> ret(lhs.Grid());
+
+  {
+    autoView(ret_v, ret,AcceleratorWrite);
+    accelerator_for(ss,rhs_v.size(),vobj::Nsimd(),{
+      convertType(ret_v[ss],innerProductD2(lhs_v(ss),rhs_v(ss)));
+    });
+  }
+  return ret;
+}
+
+////////////////////////////////////////////////////////////////////////////////////////////
+// block routines
+////////////////////////////////////////////////////////////////////////////////////////////
+template<class vobj,class CComplex,int nbasis,class VLattice>
+inline void blockProject(Lattice<iVector<CComplex,nbasis > > &coarseData,
+			   const             Lattice<vobj>   &fineData,
+			   const VLattice &Basis)
+{
+  GridBase * fine  = fineData.Grid();
+  GridBase * coarse= coarseData.Grid();
+
+  Lattice<iScalar<CComplex>> ip(coarse);
+  Lattice<vobj>     fineDataRed = fineData;
+
+  autoView( coarseData_ , coarseData, AcceleratorWrite);
+  autoView( ip_         , ip,         AcceleratorWrite);
+  for(int v=0;v<nbasis;v++) {
+    blockInnerProductD(ip,Basis[v],fineDataRed); // ip = <basis|fine>
+    accelerator_for( sc, coarse->oSites(), vobj::Nsimd(), {
+	convertType(coarseData_[sc](v),ip_[sc]);
+    });
+
+    // improve numerical stability of projection
+    // |fine> = |fine> - <basis|fine> |basis>
+    ip=-ip;
+    blockZAXPY(fineDataRed,ip,Basis[v],fineDataRed); 
+  }
+}
+
+
+template<class vobj,class vobj2,class CComplex>
+  inline void blockZAXPY(Lattice<vobj> &fineZ,
+			 const Lattice<CComplex> &coarseA,
+			 const Lattice<vobj2> &fineX,
+			 const Lattice<vobj> &fineY)
 {
   GridBase * fine  = fineZ.Grid();
   GridBase * coarse= coarseA.Grid();
@@ -182,7 +227,7 @@ inline void blockZAXPY(Lattice<vobj> &fineZ,
   conformable(fineX,fineZ);
 
   int _ndimension = coarse->_ndimension;
-  
+
   Coordinate  block_r      (_ndimension);
 
   // FIXME merge with subdivide checking routine as this is redundant
@@ -191,29 +236,66 @@ inline void blockZAXPY(Lattice<vobj> &fineZ,
     assert(block_r[d]*coarse->_rdimensions[d]==fine->_rdimensions[d]);
   }
 
-  auto fineZ_  = fineZ.View();
-  auto fineX_  = fineX.View();
-  auto fineY_  = fineY.View();
-  auto coarseA_= coarseA.View();
+  autoView( fineZ_  , fineZ, AcceleratorWrite);
+  autoView( fineX_  , fineX, AcceleratorRead);
+  autoView( fineY_  , fineY, AcceleratorRead);
+  autoView( coarseA_, coarseA, AcceleratorRead);
 
   accelerator_for(sf, fine->oSites(), CComplex::Nsimd(), {
-    
-    int sc;
-    Coordinate coor_c(_ndimension);
-    Coordinate coor_f(_ndimension);
 
-    Lexicographic::CoorFromIndex(coor_f,sf,fine->_rdimensions);
-    for(int d=0;d<_ndimension;d++) coor_c[d]=coor_f[d]/block_r[d];
-    Lexicographic::IndexFromCoor(coor_c,sc,coarse->_rdimensions);
+      int sc;
+      Coordinate coor_c(_ndimension);
+      Coordinate coor_f(_ndimension);
 
-    // z = A x + y
-    coalescedWrite(fineZ_[sf],coarseA_(sc)*fineX_(sf)+fineY_(sf));
+      Lexicographic::CoorFromIndex(coor_f,sf,fine->_rdimensions);
+      for(int d=0;d<_ndimension;d++) coor_c[d]=coor_f[d]/block_r[d];
+      Lexicographic::IndexFromCoor(coor_c,sc,coarse->_rdimensions);
 
-  });
+      // z = A x + y
+#ifdef GRID_SIMT
+      typename vobj2::tensor_reduced::scalar_object cA;
+      typename vobj::scalar_object cAx;
+#else
+      typename vobj2::tensor_reduced cA;
+      vobj cAx;
+#endif
+      convertType(cA,TensorRemove(coarseA_(sc)));
+      auto prod = cA*fineX_(sf);
+      convertType(cAx,prod);
+      coalescedWrite(fineZ_[sf],cAx+fineY_(sf));
+
+    });
 
   return;
 }
+
 template<class vobj,class CComplex>
+  inline void blockInnerProductD(Lattice<CComplex> &CoarseInner,
+				 const Lattice<vobj> &fineX,
+				 const Lattice<vobj> &fineY)
+{
+  typedef iScalar<decltype(TensorRemove(innerProductD2(vobj(),vobj())))> dotp;
+
+  GridBase *coarse(CoarseInner.Grid());
+  GridBase *fine  (fineX.Grid());
+
+  Lattice<dotp> fine_inner(fine); fine_inner.Checkerboard() = fineX.Checkerboard();
+  Lattice<dotp> coarse_inner(coarse);
+
+  // Precision promotion
+  fine_inner = localInnerProductD<vobj>(fineX,fineY);
+  blockSum(coarse_inner,fine_inner);
+  {
+    autoView( CoarseInner_  , CoarseInner,AcceleratorWrite);
+    autoView( coarse_inner_ , coarse_inner,AcceleratorRead);
+    accelerator_for(ss, coarse->oSites(), 1, {
+      convertType(CoarseInner_[ss], TensorRemove(coarse_inner_[ss]));
+    });
+  }
+ 
+}
+
+template<class vobj,class CComplex> // deprecate
 inline void blockInnerProduct(Lattice<CComplex> &CoarseInner,
 			      const Lattice<vobj> &fineX,
 			      const Lattice<vobj> &fineY)
@@ -227,15 +309,17 @@ inline void blockInnerProduct(Lattice<CComplex> &CoarseInner,
   Lattice<dotp> coarse_inner(coarse);
 
   // Precision promotion?
-  auto CoarseInner_  = CoarseInner.View();
-  auto coarse_inner_ = coarse_inner.View();
-
   fine_inner = localInnerProduct(fineX,fineY);
   blockSum(coarse_inner,fine_inner);
-  accelerator_for(ss, coarse->oSites(), 1, {
-    CoarseInner_[ss] = coarse_inner_[ss];
-  });
+  {
+    autoView( CoarseInner_  , CoarseInner, AcceleratorWrite);
+    autoView( coarse_inner_ , coarse_inner, AcceleratorRead);
+    accelerator_for(ss, coarse->oSites(), 1, {
+	CoarseInner_[ss] = coarse_inner_[ss];
+    });
+  }
 }
+
 template<class vobj,class CComplex>
 inline void blockNormalise(Lattice<CComplex> &ip,Lattice<vobj> &fineX)
 {
@@ -248,7 +332,7 @@ inline void blockNormalise(Lattice<CComplex> &ip,Lattice<vobj> &fineX)
 // useful in multigrid project;
 // Generic name : Coarsen?
 template<class vobj>
-inline void blockSum(Lattice<vobj> &coarseData,const Lattice<vobj> &fineData)
+inline void blockSum(Lattice<vobj> &coarseData,const Lattice<vobj> &fineData) 
 {
   GridBase * fine  = fineData.Grid();
   GridBase * coarse= coarseData.Grid();
@@ -256,9 +340,9 @@ inline void blockSum(Lattice<vobj> &coarseData,const Lattice<vobj> &fineData)
   subdivides(coarse,fine); // require they map
 
   int _ndimension = coarse->_ndimension;
-  
+
   Coordinate  block_r      (_ndimension);
-  
+
   for(int d=0 ; d<_ndimension;d++){
     block_r[d] = fine->_rdimensions[d] / coarse->_rdimensions[d];
   }
@@ -266,32 +350,33 @@ inline void blockSum(Lattice<vobj> &coarseData,const Lattice<vobj> &fineData)
 
   // Turn this around to loop threaded over sc and interior loop 
   // over sf would thread better
-  auto coarseData_ = coarseData.View();
-  auto fineData_   = fineData.View();
+  autoView( coarseData_ , coarseData, AcceleratorWrite);
+  autoView( fineData_   , fineData, AcceleratorRead);
 
   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();
+      // One thread per sub block
+      Coordinate coor_c(_ndimension);
+      Lexicographic::CoorFromIndex(coor_c,sc,coarse->_rdimensions);  // Block coordinate
+      coarseData_[sc]=Zero();
 
-    for(int sb=0;sb<blockVol;sb++){
-      
-      int sf;
-      Coordinate coor_b(_ndimension);
-      Coordinate coor_f(_ndimension);
-      Lexicographic::CoorFromIndex(coor_b,sb,block_r);               // Block sub coordinate
-      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);
+      for(int sb=0;sb<blockVol;sb++){
 
-      coarseData_[sc]=coarseData_[sc]+fineData_[sf];
-    }
+	int sf;
+	Coordinate coor_b(_ndimension);
+	Coordinate coor_f(_ndimension);
+	Lexicographic::CoorFromIndex(coor_b,sb,block_r);               // Block sub coordinate
+	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);
 
-  });
+	coarseData_[sc]=coarseData_[sc]+fineData_[sf];
+      }
+
+    });
   return;
 }
 
+
 template<class vobj>
 inline void blockPick(GridBase *coarse,const Lattice<vobj> &unpicked,Lattice<vobj> &picked,Coordinate coor)
 {
@@ -313,8 +398,8 @@ inline void blockPick(GridBase *coarse,const Lattice<vobj> &unpicked,Lattice<vob
   }
 }
 
-template<class vobj,class CComplex>
-inline void blockOrthogonalise(Lattice<CComplex> &ip,std::vector<Lattice<vobj> > &Basis)
+template<class CComplex,class VLattice>
+inline void blockOrthonormalize(Lattice<CComplex> &ip,VLattice &Basis)
 {
   GridBase *coarse = ip.Grid();
   GridBase *fine   = Basis[0].Grid();
@@ -322,23 +407,30 @@ inline void blockOrthogonalise(Lattice<CComplex> &ip,std::vector<Lattice<vobj> >
   int       nbasis = Basis.size() ;
 
   // checks
-  subdivides(coarse,fine); 
+  subdivides(coarse,fine);
   for(int i=0;i<nbasis;i++){
     conformable(Basis[i].Grid(),fine);
   }
 
   for(int v=0;v<nbasis;v++) {
     for(int u=0;u<v;u++) {
-      //Inner product & remove component 
-      blockInnerProduct(ip,Basis[u],Basis[v]);
+      //Inner product & remove component
+      blockInnerProductD(ip,Basis[u],Basis[v]);
       ip = -ip;
-      blockZAXPY<vobj,CComplex> (Basis[v],ip,Basis[u],Basis[v]);
+      blockZAXPY(Basis[v],ip,Basis[u],Basis[v]);
     }
     blockNormalise(ip,Basis[v]);
   }
 }
 
+template<class vobj,class CComplex>
+inline void blockOrthogonalise(Lattice<CComplex> &ip,std::vector<Lattice<vobj> > &Basis) // deprecated inaccurate naming
+{
+  blockOrthonormalize(ip,Basis);
+}
+
 #if 0
+// TODO: CPU optimized version here
 template<class vobj,class CComplex,int nbasis>
 inline void blockPromote(const Lattice<iVector<CComplex,nbasis > > &coarseData,
 			 Lattice<vobj>   &fineData,
@@ -360,8 +452,8 @@ inline void blockPromote(const Lattice<iVector<CComplex,nbasis > > &coarseData,
   for(int d=0 ; d<_ndimension;d++){
     block_r[d] = fine->_rdimensions[d] / coarse->_rdimensions[d];
   }
-  auto fineData_   = fineData.View();
-  auto coarseData_ = coarseData.View();
+  autoView( fineData_   , fineData, AcceleratorWrite);
+  autoView( coarseData_ , coarseData, AcceleratorRead);
 
   // Loop with a cache friendly loop ordering
   accelerator_for(sf,fine->oSites(),1,{
@@ -374,7 +466,7 @@ inline void blockPromote(const Lattice<iVector<CComplex,nbasis > > &coarseData,
     Lexicographic::IndexFromCoor(coor_c,sc,coarse->_rdimensions);
 
     for(int i=0;i<nbasis;i++) {
-      auto basis_ = Basis[i].View();
+      /*      auto basis_ = Basis[i],  );*/
       if(i==0) fineData_[sf]=coarseData_[sc](i) *basis_[sf]);
       else     fineData_[sf]=fineData_[sf]+coarseData_[sc](i)*basis_[sf]);
     }
@@ -383,24 +475,25 @@ inline void blockPromote(const Lattice<iVector<CComplex,nbasis > > &coarseData,
   
 }
 #else
-template<class vobj,class CComplex,int nbasis>
+template<class vobj,class CComplex,int nbasis,class VLattice>
 inline void blockPromote(const Lattice<iVector<CComplex,nbasis > > &coarseData,
 			 Lattice<vobj>   &fineData,
-			 const std::vector<Lattice<vobj> > &Basis)
+			 const VLattice &Basis)
 {
   GridBase * fine  = fineData.Grid();
   GridBase * coarse= coarseData.Grid();
-
   fineData=Zero();
   for(int i=0;i<nbasis;i++) {
     Lattice<iScalar<CComplex> > ip = PeekIndex<0>(coarseData,i);
-    Lattice<CComplex> cip(coarse);
-    auto cip_ = cip.View();
-    auto  ip_ =  ip.View();
-    accelerator_forNB(sc,coarse->oSites(),CComplex::Nsimd(),{
-	coalescedWrite(cip_[sc], ip_(sc)());
-    });
-    blockZAXPY<vobj,CComplex >(fineData,cip,Basis[i],fineData);
+
+    //Lattice<CComplex> cip(coarse);
+    //autoView( cip_ , cip, AcceleratorWrite);
+    //autoView(  ip_ ,  ip, AcceleratorRead);
+    //accelerator_forNB(sc,coarse->oSites(),CComplex::Nsimd(),{
+    //	coalescedWrite(cip_[sc], ip_(sc)());
+    //  });
+    //blockZAXPY<vobj,CComplex >(fineData,cip,Basis[i],fineData);
+    blockZAXPY(fineData,ip,Basis[i],fineData);
   }
 }
 #endif
@@ -427,15 +520,17 @@ void localConvert(const Lattice<vobj> &in,Lattice<vvobj> &out)
     assert(ig->lSites() == og->lSites());
   }
 
+  autoView(in_v,in,CpuRead);
+  autoView(out_v,out,CpuWrite);
   thread_for(idx, ig->lSites(),{
     sobj s;
     ssobj ss;
 
     Coordinate lcoor(ni);
     ig->LocalIndexToLocalCoor(idx,lcoor);
-    peekLocalSite(s,in,lcoor);
+    peekLocalSite(s,in_v,lcoor);
     ss=s;
-    pokeLocalSite(ss,out,lcoor);
+    pokeLocalSite(ss,out_v,lcoor);
   });
 }
 
@@ -470,8 +565,9 @@ void localCopyRegion(const Lattice<vobj> &From,Lattice<vobj> & To,Coordinate Fro
   Coordinate rdt = Tg->_rdimensions;
   Coordinate ist = Tg->_istride;
   Coordinate ost = Tg->_ostride;
-  auto t_v = To.View();
-  auto f_v = From.View();
+
+  autoView( t_v , To, AcceleratorWrite);
+  autoView( f_v , From, AcceleratorRead);
   accelerator_for(idx,Fg->lSites(),1,{
     sobj s;
     Coordinate Fcoor(nd);
@@ -494,8 +590,6 @@ void localCopyRegion(const Lattice<vobj> &From,Lattice<vobj> & To,Coordinate Fro
       for(int w=0;w<words;w++){
 	tp[idx_t+w*Nsimd] = fp[idx_f+w*Nsimd];  // FIXME IF RRII layout, type pun no worke
       }
-      //      peekLocalSite(s,From,Fcoor);
-      //      pokeLocalSite(s,To  ,Tcoor);
     }
   });
 }
@@ -526,6 +620,8 @@ void InsertSlice(const Lattice<vobj> &lowDim,Lattice<vobj> & higherDim,int slice
   }
 
   // the above should guarantee that the operations are local
+  autoView(lowDimv,lowDim,CpuRead);
+  autoView(higherDimv,higherDim,CpuWrite);
   thread_for(idx,lg->lSites(),{
     sobj s;
     Coordinate lcoor(nl);
@@ -538,8 +634,8 @@ void InsertSlice(const Lattice<vobj> &lowDim,Lattice<vobj> & higherDim,int slice
 	hcoor[d]=lcoor[ddl++];
       }
     }
-    peekLocalSite(s,lowDim,lcoor);
-    pokeLocalSite(s,higherDim,hcoor);
+    peekLocalSite(s,lowDimv,lcoor);
+    pokeLocalSite(s,higherDimv,hcoor);
   });
 }
 
@@ -567,6 +663,8 @@ void ExtractSlice(Lattice<vobj> &lowDim,const Lattice<vobj> & higherDim,int slic
     }
   }
   // the above should guarantee that the operations are local
+  autoView(lowDimv,lowDim,CpuWrite);
+  autoView(higherDimv,higherDim,CpuRead);
   thread_for(idx,lg->lSites(),{
     sobj s;
     Coordinate lcoor(nl);
@@ -579,8 +677,8 @@ void ExtractSlice(Lattice<vobj> &lowDim,const Lattice<vobj> & higherDim,int slic
 	hcoor[d]=lcoor[ddl++];
       }
     }
-    peekLocalSite(s,higherDim,hcoor);
-    pokeLocalSite(s,lowDim,lcoor);
+    peekLocalSite(s,higherDimv,hcoor);
+    pokeLocalSite(s,lowDimv,lcoor);
   });
 
 }
@@ -608,6 +706,8 @@ void InsertSliceLocal(const Lattice<vobj> &lowDim, Lattice<vobj> & higherDim,int
   }
 
   // the above should guarantee that the operations are local
+  autoView(lowDimv,lowDim,CpuRead);
+  autoView(higherDimv,higherDim,CpuWrite);
   thread_for(idx,lg->lSites(),{
     sobj s;
     Coordinate lcoor(nl);
@@ -616,8 +716,8 @@ void InsertSliceLocal(const Lattice<vobj> &lowDim, Lattice<vobj> & higherDim,int
     if( lcoor[orthog] == slice_lo ) { 
       hcoor=lcoor;
       hcoor[orthog] = slice_hi;
-      peekLocalSite(s,lowDim,lcoor);
-      pokeLocalSite(s,higherDim,hcoor);
+      peekLocalSite(s,lowDimv,lcoor);
+      pokeLocalSite(s,higherDimv,hcoor);
     }
   });
 }
@@ -645,6 +745,8 @@ void ExtractSliceLocal(Lattice<vobj> &lowDim,const Lattice<vobj> & higherDim,int
   }
 
   // the above should guarantee that the operations are local
+  autoView(lowDimv,lowDim,CpuWrite);
+  autoView(higherDimv,higherDim,CpuRead);
   thread_for(idx,lg->lSites(),{
     sobj s;
     Coordinate lcoor(nl);
@@ -653,8 +755,8 @@ void ExtractSliceLocal(Lattice<vobj> &lowDim,const Lattice<vobj> & higherDim,int
     if( lcoor[orthog] == slice_lo ) { 
       hcoor=lcoor;
       hcoor[orthog] = slice_hi;
-      peekLocalSite(s,higherDim,hcoor);
-      pokeLocalSite(s,lowDim,lcoor);
+      peekLocalSite(s,higherDimv,hcoor);
+      pokeLocalSite(s,lowDimv,lcoor);
     }
   });
 }
@@ -718,7 +820,7 @@ unvectorizeToLexOrdArray(std::vector<sobj> &out, const Lattice<vobj> &in)
   }
 
   //loop over outer index
-  auto in_v  = in.View();
+  autoView( in_v  , in, CpuRead);
   thread_for(in_oidx,in_grid->oSites(),{
     //Assemble vector of pointers to output elements
     ExtractPointerArray<sobj> out_ptrs(in_nsimd);
@@ -811,7 +913,7 @@ vectorizeFromLexOrdArray( std::vector<sobj> &in, Lattice<vobj> &out)
     icoor[lane].resize(ndim);
     grid->iCoorFromIindex(icoor[lane],lane);
   }
-  auto out_v = out.View();
+  autoView( out_v , out, CpuWrite);
   thread_for(oidx, grid->oSites(),{
     //Assemble vector of pointers to output elements
     ExtractPointerArray<sobj> ptrs(nsimd);
@@ -914,7 +1016,7 @@ void precisionChange(Lattice<VobjOut> &out, const Lattice<VobjIn> &in)
   std::vector<SobjOut> in_slex_conv(in_grid->lSites());
   unvectorizeToLexOrdArray(in_slex_conv, in);
     
-  auto out_v = out.View();
+  autoView( out_v , out, CpuWrite);
   thread_for(out_oidx,out_grid->oSites(),{
     Coordinate out_ocoor(ndim);
     out_grid->oCoorFromOindex(out_ocoor, out_oidx);
diff --git a/Grid/lattice/Lattice_transpose.h b/Grid/lattice/Lattice_transpose.h
index 9b0b3483..adfe3380 100644
--- a/Grid/lattice/Lattice_transpose.h
+++ b/Grid/lattice/Lattice_transpose.h
@@ -38,17 +38,19 @@ NAMESPACE_BEGIN(Grid);
 ////////////////////////////////////////////////////////////////////////////////////////////////////
 // Transpose
 ////////////////////////////////////////////////////////////////////////////////////////////////////
+/*
 template<class vobj>
 inline Lattice<vobj> transpose(const Lattice<vobj> &lhs){
   Lattice<vobj> ret(lhs.Grid());
-  auto ret_v = ret.View();
-  auto lhs_v = lhs.View();
+  autoView( ret_v, ret, AcceleratorWrite);
+  autoView( lhs_v, lhs, AcceleratorRead);
   accelerator_for(ss,lhs_v.size(),vobj::Nsimd(),{
     coalescedWrite(ret_v[ss], transpose(lhs_v(ss)));
   });
   return ret;
 };
-    
+*/    
+
 ////////////////////////////////////////////////////////////////////////////////////////////////////
 // Index level dependent transpose
 ////////////////////////////////////////////////////////////////////////////////////////////////////
@@ -56,8 +58,8 @@ template<int Index,class vobj>
 inline auto TransposeIndex(const Lattice<vobj> &lhs) -> Lattice<decltype(transposeIndex<Index>(vobj()))>
 {
   Lattice<decltype(transposeIndex<Index>(vobj()))> ret(lhs.Grid());
-  auto ret_v = ret.View();
-  auto lhs_v = lhs.View();
+  autoView( ret_v, ret, AcceleratorWrite);
+  autoView( lhs_v, lhs, AcceleratorRead);
   accelerator_for(ss,lhs_v.size(),vobj::Nsimd(),{
     coalescedWrite(ret_v[ss] , transposeIndex<Index>(lhs_v(ss)));
   });
diff --git a/Grid/lattice/Lattice_unary.h b/Grid/lattice/Lattice_unary.h
index 591afe72..07424b3d 100644
--- a/Grid/lattice/Lattice_unary.h
+++ b/Grid/lattice/Lattice_unary.h
@@ -35,8 +35,8 @@ NAMESPACE_BEGIN(Grid);
 
 template<class obj> Lattice<obj> pow(const Lattice<obj> &rhs_i,RealD y){
   Lattice<obj> ret_i(rhs_i.Grid());
-  auto rhs = rhs_i.View();
-  auto ret = ret_i.View();
+  autoView( rhs, rhs_i, AcceleratorRead);
+  autoView( ret, ret_i, AcceleratorWrite);
   ret.Checkerboard() = rhs.Checkerboard();
   accelerator_for(ss,rhs.size(),1,{
       ret[ss]=pow(rhs[ss],y);
@@ -45,8 +45,8 @@ template<class obj> Lattice<obj> pow(const Lattice<obj> &rhs_i,RealD y){
 }
 template<class obj> Lattice<obj> mod(const Lattice<obj> &rhs_i,Integer y){
   Lattice<obj> ret_i(rhs_i.Grid());
-  auto rhs = rhs_i.View();
-  auto ret = ret_i.View();
+  autoView( rhs , rhs_i, AcceleratorRead);
+  autoView( ret , ret_i, AcceleratorWrite);
   ret.Checkerboard() = rhs.Checkerboard();
   accelerator_for(ss,rhs.size(),obj::Nsimd(),{
     coalescedWrite(ret[ss],mod(rhs(ss),y));
@@ -56,8 +56,8 @@ template<class obj> Lattice<obj> mod(const Lattice<obj> &rhs_i,Integer y){
 
 template<class obj> Lattice<obj> div(const Lattice<obj> &rhs_i,Integer y){
   Lattice<obj> ret_i(rhs_i.Grid());
-  auto ret = ret_i.View();
-  auto rhs = rhs_i.View();
+  autoView( ret , ret_i, AcceleratorWrite);
+  autoView( rhs , rhs_i, AcceleratorRead);
   ret.Checkerboard() = rhs_i.Checkerboard();
   accelerator_for(ss,rhs.size(),obj::Nsimd(),{
     coalescedWrite(ret[ss],div(rhs(ss),y));
@@ -67,8 +67,8 @@ template<class obj> Lattice<obj> div(const Lattice<obj> &rhs_i,Integer y){
 
 template<class obj> Lattice<obj> expMat(const Lattice<obj> &rhs_i, RealD alpha, Integer Nexp = DEFAULT_MAT_EXP){
   Lattice<obj> ret_i(rhs_i.Grid());
-  auto rhs = rhs_i.View();
-  auto ret = ret_i.View();
+  autoView( rhs , rhs_i, AcceleratorRead);
+  autoView( ret , ret_i, AcceleratorWrite);
   ret.Checkerboard() = rhs.Checkerboard();
   accelerator_for(ss,rhs.size(),obj::Nsimd(),{
     coalescedWrite(ret[ss],Exponentiate(rhs(ss),alpha, Nexp));
diff --git a/Grid/lattice/Lattice_view.h b/Grid/lattice/Lattice_view.h
new file mode 100644
index 00000000..3b76b921
--- /dev/null
+++ b/Grid/lattice/Lattice_view.h
@@ -0,0 +1,168 @@
+#pragma once
+NAMESPACE_BEGIN(Grid);
+///////////////////////////////////////////////////////////////////
+// Base class which can be used by traits to pick up behaviour
+///////////////////////////////////////////////////////////////////
+class LatticeBase {};
+
+/////////////////////////////////////////////////////////////////////////////////////////
+// Conformable checks; same instance of Grid required
+/////////////////////////////////////////////////////////////////////////////////////////
+void accelerator_inline conformable(GridBase *lhs,GridBase *rhs)
+{
+  assert(lhs == rhs);
+}
+
+////////////////////////////////////////////////////////////////////////////
+// Minimal base class containing only data valid to access from accelerator
+// _odata will be a managed pointer in CUDA
+////////////////////////////////////////////////////////////////////////////
+// Force access to lattice through a view object.
+// prevents writing of code that will not offload to GPU, but perhaps annoyingly
+// strict since host could could in principle direct access through the lattice object
+// Need to decide programming model.
+#define LATTICE_VIEW_STRICT
+template<class vobj> class LatticeAccelerator : public LatticeBase
+{
+protected:
+  //public:
+  GridBase *_grid;
+  int checkerboard;
+  vobj     *_odata;    // A managed pointer
+  uint64_t _odata_size;    
+  ViewAdvise advise;
+public:
+  accelerator_inline LatticeAccelerator() : checkerboard(0), _odata(nullptr), _odata_size(0), _grid(nullptr), advise(AdviseDefault) { }; 
+  accelerator_inline uint64_t oSites(void) const { return _odata_size; };
+  accelerator_inline int  Checkerboard(void) const { return checkerboard; };
+  accelerator_inline int &Checkerboard(void) { return this->checkerboard; }; // can assign checkerboard on a container, not a view
+  accelerator_inline ViewAdvise Advise(void) const { return advise; };
+  accelerator_inline ViewAdvise &Advise(void) { return this->advise; }; // can assign advise on a container, not a view
+  accelerator_inline void Conformable(GridBase * &grid) const
+  { 
+    if (grid) conformable(grid, _grid);
+    else      grid = _grid;
+  };
+  // Host only
+  GridBase * getGrid(void) const { return _grid; };
+};
+
+/////////////////////////////////////////////////////////////////////////////////////////
+// A View class which provides accessor to the data.
+// This will be safe to call from accelerator_for and is trivially copy constructible
+// The copy constructor for this will need to be used by device lambda functions
+/////////////////////////////////////////////////////////////////////////////////////////
+template<class vobj> 
+class LatticeView : public LatticeAccelerator<vobj>
+{
+public:
+  // Rvalue
+  ViewMode mode;
+  void * cpu_ptr;
+#ifdef GRID_SIMT
+  accelerator_inline const typename vobj::scalar_object operator()(size_t i) const { 
+    return coalescedRead(this->_odata[i]); 
+  }
+#else 
+  accelerator_inline const vobj & operator()(size_t i) const { return this->_odata[i]; }
+#endif
+
+  accelerator_inline const vobj & operator[](size_t i) const { return this->_odata[i]; };
+  accelerator_inline vobj       & operator[](size_t i)       { return this->_odata[i]; };
+
+  accelerator_inline uint64_t begin(void) const { return 0;};
+  accelerator_inline uint64_t end(void)   const { return this->_odata_size; };
+  accelerator_inline uint64_t size(void)  const { return this->_odata_size; };
+
+  LatticeView(const LatticeAccelerator<vobj> &refer_to_me) : LatticeAccelerator<vobj> (refer_to_me){}
+  LatticeView(const LatticeView<vobj> &refer_to_me) = default; // Trivially copyable
+  LatticeView(const LatticeAccelerator<vobj> &refer_to_me,ViewMode mode) : LatticeAccelerator<vobj> (refer_to_me)
+  {
+    this->ViewOpen(mode);
+  }
+
+  // Host functions
+  void ViewOpen(ViewMode mode)
+  { // Translate the pointer, could save a copy. Could use a "Handle" and not save _odata originally in base
+    //    std::cout << "View Open"<<std::hex<<this->_odata<<std::dec <<std::endl;
+    this->cpu_ptr = (void *)this->_odata;
+    this->mode    = mode;
+    this->_odata  =(vobj *)
+      MemoryManager::ViewOpen(this->cpu_ptr,
+				this->_odata_size*sizeof(vobj),
+				mode,
+				this->advise);    
+  }
+  void ViewClose(void)
+  { // Inform the manager
+    //    std::cout << "View Close"<<std::hex<<this->cpu_ptr<<std::dec <<std::endl;
+    MemoryManager::ViewClose(this->cpu_ptr,this->mode);    
+  }
+
+};
+// Little autoscope assister
+template<class View> 
+class ViewCloser
+{
+  View v;  // Take a copy of view and call view close when I go out of scope automatically
+ public:
+  ViewCloser(View &_v) : v(_v) {};
+  ~ViewCloser() { v.ViewClose(); }
+};
+
+#define autoView(l_v,l,mode)				\
+	  auto l_v = l.View(mode);			\
+	  ViewCloser<decltype(l_v)> _autoView##l_v(l_v);
+
+/////////////////////////////////////////////////////////////////////////////////////////
+// Lattice expression types used by ET to assemble the AST
+// 
+// Need to be able to detect code paths according to the whether a lattice object or not
+// so introduce some trait type things
+/////////////////////////////////////////////////////////////////////////////////////////
+
+class LatticeExpressionBase {};
+
+template <typename T> using is_lattice = std::is_base_of<LatticeBase, T>;
+template <typename T> using is_lattice_expr = std::is_base_of<LatticeExpressionBase,T >;
+
+template<class T, bool isLattice> struct ViewMapBase { typedef T Type; };
+template<class T>                 struct ViewMapBase<T,true> { typedef LatticeView<typename T::vector_object> Type; };
+template<class T> using ViewMap = ViewMapBase<T,std::is_base_of<LatticeBase, T>::value >;
+
+template <typename Op, typename _T1>                           
+class LatticeUnaryExpression : public  LatticeExpressionBase 
+{
+public:
+  typedef typename ViewMap<_T1>::Type T1;
+  Op op;
+  T1 arg1;
+  LatticeUnaryExpression(Op _op,const _T1 &_arg1) : op(_op), arg1(_arg1) {};
+};
+
+template <typename Op, typename _T1, typename _T2>              
+class LatticeBinaryExpression : public LatticeExpressionBase 
+{
+public:
+  typedef typename ViewMap<_T1>::Type T1;
+  typedef typename ViewMap<_T2>::Type T2;
+  Op op;
+  T1 arg1;
+  T2 arg2;
+  LatticeBinaryExpression(Op _op,const _T1 &_arg1,const _T2 &_arg2) : op(_op), arg1(_arg1), arg2(_arg2) {};
+};
+
+template <typename Op, typename _T1, typename _T2, typename _T3> 
+class LatticeTrinaryExpression : public LatticeExpressionBase 
+{
+public:
+  typedef typename ViewMap<_T1>::Type T1;
+  typedef typename ViewMap<_T2>::Type T2;
+  typedef typename ViewMap<_T3>::Type T3;
+  Op op;
+  T1 arg1;
+  T2 arg2;
+  T3 arg3;
+  LatticeTrinaryExpression(Op _op,const _T1 &_arg1,const _T2 &_arg2,const _T3 &_arg3) : op(_op), arg1(_arg1), arg2(_arg2), arg3(_arg3) {};
+};
+NAMESPACE_END(Grid);
diff --git a/Grid/parallelIO/BinaryIO.h b/Grid/parallelIO/BinaryIO.h
index f90c34a9..1f11add9 100644
--- a/Grid/parallelIO/BinaryIO.h
+++ b/Grid/parallelIO/BinaryIO.h
@@ -341,7 +341,7 @@ class BinaryIO {
     int ieee32big = (format == std::string("IEEE32BIG"));
     int ieee32    = (format == std::string("IEEE32"));
     int ieee64big = (format == std::string("IEEE64BIG"));
-    int ieee64    = (format == std::string("IEEE64"));
+    int ieee64    = (format == std::string("IEEE64") || format == std::string("IEEE64LITTLE"));
     assert(ieee64||ieee32|ieee64big||ieee32big);
     assert((ieee64+ieee32+ieee64big+ieee32big)==1);
     //////////////////////////////////////////////////////////////////////////////
diff --git a/Grid/parallelIO/MetaData.h b/Grid/parallelIO/MetaData.h
index 2e211838..4c1cfbdb 100644
--- a/Grid/parallelIO/MetaData.h
+++ b/Grid/parallelIO/MetaData.h
@@ -301,6 +301,30 @@ struct GaugeSimpleUnmunger {
   };
 };
 
+template<class fobj,class sobj>
+struct GaugeDoubleStoredMunger{
+  void operator()(fobj &in, sobj &out) {
+    for (int mu = 0; mu < Nds; mu++) {
+      for (int i = 0; i < Nc; i++) {
+        for (int j = 0; j < Nc; j++) {
+          out(mu)()(i, j) = in(mu)()(i, j);
+        }}
+    }
+  };
+};
+
+template <class fobj, class sobj>
+struct GaugeDoubleStoredUnmunger {
+  void operator()(sobj &in, fobj &out) {
+    for (int mu = 0; mu < Nds; mu++) {
+      for (int i = 0; i < Nc; i++) {
+        for (int j = 0; j < Nc; j++) {
+          out(mu)()(i, j) = in(mu)()(i, j);
+        }}
+    }
+  };
+};
+
 template<class fobj,class sobj>
 struct Gauge3x2munger{
   void operator() (fobj &in,sobj &out){
diff --git a/Grid/parallelIO/NerscIO.h b/Grid/parallelIO/NerscIO.h
index d3b62d1f..5522ba91 100644
--- a/Grid/parallelIO/NerscIO.h
+++ b/Grid/parallelIO/NerscIO.h
@@ -146,7 +146,7 @@ public:
     int ieee32big = (format == std::string("IEEE32BIG"));
     int ieee32    = (format == std::string("IEEE32"));
     int ieee64big = (format == std::string("IEEE64BIG"));
-    int ieee64    = (format == std::string("IEEE64"));
+    int ieee64    = (format == std::string("IEEE64") || format == std::string("IEEE64LITTLE"));
 
     uint32_t nersc_csum,scidac_csuma,scidac_csumb;
     // depending on datatype, set up munger;
diff --git a/Grid/parallelIO/OpenQcdIO.h b/Grid/parallelIO/OpenQcdIO.h
new file mode 100644
index 00000000..00911595
--- /dev/null
+++ b/Grid/parallelIO/OpenQcdIO.h
@@ -0,0 +1,224 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid
+
+Source file: ./lib/parallelIO/OpenQcdIO.h
+
+Copyright (C) 2015 - 2020
+
+Author: Daniel Richtmann <daniel.richtmann@ur.de>
+
+This program is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 2 of the License, or
+(at your option) any later version.
+
+This program is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along
+with this program; if not, write to the Free Software Foundation, Inc.,
+51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+See the full license in the file "LICENSE" in the top level distribution
+directory
+*************************************************************************************/
+/*  END LEGAL */
+#pragma once
+
+NAMESPACE_BEGIN(Grid);
+
+struct OpenQcdHeader : Serializable {
+  GRID_SERIALIZABLE_CLASS_MEMBERS(OpenQcdHeader,
+                                  int,    Nt,
+                                  int,    Nx,
+                                  int,    Ny,
+                                  int,    Nz,
+                                  double, plaq);
+};
+
+class OpenQcdIO : public BinaryIO {
+public:
+  static constexpr double normalisationFactor = Nc; // normalisation difference: grid 18, openqcd 6
+
+  static inline int readHeader(std::string file, GridBase* grid, FieldMetaData& field) {
+    OpenQcdHeader header;
+
+    {
+      std::ifstream fin(file, std::ios::in | std::ios::binary);
+      fin.read(reinterpret_cast<char*>(&header), sizeof(OpenQcdHeader));
+      assert(!fin.fail());
+      field.data_start = fin.tellg();
+      fin.close();
+    }
+
+    header.plaq /= normalisationFactor;
+
+    // sanity check (should trigger on endian issues)
+    assert(0 < header.Nt && header.Nt <= 1024);
+    assert(0 < header.Nx && header.Nx <= 1024);
+    assert(0 < header.Ny && header.Ny <= 1024);
+    assert(0 < header.Nz && header.Nz <= 1024);
+
+    field.dimension[0] = header.Nx;
+    field.dimension[1] = header.Ny;
+    field.dimension[2] = header.Nz;
+    field.dimension[3] = header.Nt;
+
+    std::cout << GridLogDebug << "header: " << header << std::endl;
+    std::cout << GridLogDebug << "grid dimensions: " << grid->_fdimensions << std::endl;
+    std::cout << GridLogDebug << "file dimensions: " << field.dimension << std::endl;
+
+    assert(grid->_ndimension == Nd);
+    for(int d = 0; d < Nd; d++)
+      assert(grid->_fdimensions[d] == field.dimension[d]);
+
+    field.plaquette = header.plaq;
+
+    return field.data_start;
+  }
+
+  template<class vsimd>
+  static inline void readConfiguration(Lattice<iLorentzColourMatrix<vsimd>>& Umu,
+                                       FieldMetaData&                        header,
+                                       std::string                           file) {
+    typedef Lattice<iDoubleStoredColourMatrix<vsimd>> DoubleStoredGaugeField;
+
+    assert(Ns == 4 and Nd == 4 and Nc == 3);
+
+    auto grid = dynamic_cast<GridCartesian*>(Umu.Grid());
+    assert(grid != nullptr); assert(grid->_ndimension == Nd);
+
+    uint64_t offset = readHeader(file, Umu.Grid(), header);
+
+    FieldMetaData clone(header);
+
+    std::string format("IEEE64"); // they always store little endian double precsision
+    uint32_t    nersc_csum, scidac_csuma, scidac_csumb;
+
+    GridCartesian*         grid_openqcd = createOpenQcdGrid(grid);
+    GridRedBlackCartesian* grid_rb      = SpaceTimeGrid::makeFourDimRedBlackGrid(grid);
+
+    typedef DoubleStoredColourMatrixD                                              fobj;
+    typedef typename DoubleStoredGaugeField::vector_object::scalar_object          sobj;
+    typedef typename DoubleStoredGaugeField::vector_object::Realified::scalar_type word;
+
+    word w = 0;
+
+    std::vector<fobj> iodata(grid_openqcd->lSites()); // Munge, checksum, byte order in here
+    std::vector<sobj> scalardata(grid->lSites());
+
+    IOobject(w, grid_openqcd, iodata, file, offset, format, BINARYIO_READ | BINARYIO_LEXICOGRAPHIC,
+             nersc_csum, scidac_csuma, scidac_csumb);
+
+    GridStopWatch timer;
+    timer.Start();
+
+    DoubleStoredGaugeField Umu_ds(grid);
+
+    auto munge = GaugeDoubleStoredMunger<DoubleStoredColourMatrixD, DoubleStoredColourMatrix>();
+
+    Coordinate ldim = grid->LocalDimensions();
+    thread_for(idx_g, grid->lSites(), {
+        Coordinate coor;
+        grid->LocalIndexToLocalCoor(idx_g, coor);
+
+        bool isOdd = grid_rb->CheckerBoard(coor) == Odd;
+
+        if(!isOdd) continue;
+
+        int idx_o = (coor[Tdir] * ldim[Xdir] * ldim[Ydir] * ldim[Zdir]
+                  +  coor[Xdir] * ldim[Ydir] * ldim[Zdir]
+                  +  coor[Ydir] * ldim[Zdir]
+                  +  coor[Zdir])/2;
+
+        munge(iodata[idx_o], scalardata[idx_g]);
+    });
+
+    grid->Barrier(); timer.Stop();
+    std::cout << Grid::GridLogMessage << "OpenQcdIO::readConfiguration: munge overhead " << timer.Elapsed() << std::endl;
+
+    timer.Reset(); timer.Start();
+
+    vectorizeFromLexOrdArray(scalardata, Umu_ds);
+
+    grid->Barrier(); timer.Stop();
+    std::cout << Grid::GridLogMessage << "OpenQcdIO::readConfiguration: vectorize overhead " << timer.Elapsed() << std::endl;
+
+    timer.Reset(); timer.Start();
+
+    undoDoubleStore(Umu, Umu_ds);
+
+    grid->Barrier(); timer.Stop();
+    std::cout << Grid::GridLogMessage << "OpenQcdIO::readConfiguration: redistribute overhead " << timer.Elapsed() << std::endl;
+
+    GaugeStatistics(Umu, clone);
+
+    RealD plaq_diff = fabs(clone.plaquette - header.plaquette);
+
+    // clang-format off
+    std::cout << GridLogMessage << "OpenQcd Configuration " << file
+              << " plaquette " << clone.plaquette
+              << " header " << header.plaquette
+              << " difference " << plaq_diff
+              << std::endl;
+    // clang-format on
+
+    RealD precTol = (getPrecision<vsimd>::value == 1) ? 2e-7 : 2e-15;
+    RealD tol     = precTol * std::sqrt(grid->_Nprocessors); // taken from RQCD chroma code
+
+    if(plaq_diff >= tol)
+      std::cout << " Plaquette mismatch (diff = " << plaq_diff << ", tol = " << tol << ")" << std::endl;
+    assert(plaq_diff < tol);
+
+    std::cout << GridLogMessage << "OpenQcd Configuration " << file << " and plaquette agree" << std::endl;
+  }
+
+  template<class vsimd>
+  static inline void writeConfiguration(Lattice<iLorentzColourMatrix<vsimd>>& Umu,
+                                        std::string                           file) {
+    std::cout << GridLogError << "Writing to openQCD file format is not implemented" << std::endl;
+    exit(EXIT_FAILURE);
+  }
+
+private:
+  static inline GridCartesian* createOpenQcdGrid(GridCartesian* grid) {
+    // exploit GridCartesian to be able to still use IOobject
+    Coordinate gdim  = grid->GlobalDimensions();
+    Coordinate ldim  = grid->LocalDimensions();
+    Coordinate pcoor = grid->ThisProcessorCoor();
+
+    // openqcd does rb on the z direction
+    gdim[Zdir] /= 2;
+    ldim[Zdir] /= 2;
+
+    // and has the order T X Y Z (from slowest to fastest)
+    std::swap(gdim[Xdir], gdim[Zdir]);
+    std::swap(ldim[Xdir], ldim[Zdir]);
+    std::swap(pcoor[Xdir], pcoor[Zdir]);
+
+    GridCartesian* ret   = SpaceTimeGrid::makeFourDimGrid(gdim, grid->_simd_layout, grid->ProcessorGrid());
+    ret->_ldimensions    = ldim;
+    ret->_processor_coor = pcoor;
+    return ret;
+  }
+
+  template<class vsimd>
+  static inline void undoDoubleStore(Lattice<iLorentzColourMatrix<vsimd>>&            Umu,
+                                     Lattice<iDoubleStoredColourMatrix<vsimd>> const& Umu_ds) {
+    conformable(Umu.Grid(), Umu_ds.Grid());
+    Lattice<iColourMatrix<vsimd>> U(Umu.Grid());
+
+    // they store T+, T-, X+, X-, Y+, Y-, Z+, Z-
+    for(int mu_g = 0; mu_g < Nd; ++mu_g) {
+      int mu_o = (mu_g + 1) % Nd;
+      U        = PeekIndex<LorentzIndex>(Umu_ds, 2 * mu_o)
+               + Cshift(PeekIndex<LorentzIndex>(Umu_ds, 2 * mu_o + 1), mu_g, +1);
+      PokeIndex<LorentzIndex>(Umu, U, mu_g);
+    }
+  }
+};
+
+NAMESPACE_END(Grid);
diff --git a/Grid/parallelIO/OpenQcdIOChromaReference.h b/Grid/parallelIO/OpenQcdIOChromaReference.h
new file mode 100644
index 00000000..bab54fe8
--- /dev/null
+++ b/Grid/parallelIO/OpenQcdIOChromaReference.h
@@ -0,0 +1,281 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid
+
+Source file: ./lib/parallelIO/OpenQcdIOChromaReference.h
+
+Copyright (C) 2015 - 2020
+
+Author: Daniel Richtmann <daniel.richtmann@ur.de>
+
+This program is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 2 of the License, or
+(at your option) any later version.
+
+This program is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along
+with this program; if not, write to the Free Software Foundation, Inc.,
+51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+See the full license in the file "LICENSE" in the top level distribution
+directory
+*************************************************************************************/
+/*  END LEGAL */
+#pragma once
+
+#include <ios>
+#include <iostream>
+#include <limits>
+#include <iomanip>
+#include <mpi.h>
+#include <ostream>
+#include <string>
+
+#define CHECK {std::cerr << __FILE__ << " @l " << __LINE__ << ": CHECK" << grid->ThisRank() << std::endl;}
+#define CHECK_VAR(a)   { std::cerr << __FILE__ << "@l" << __LINE__ << " on "<< grid->ThisRank() << ": " << __func__ << " " << #a << "=" << (a) << std::endl; }
+// #undef CHECK
+// #define CHECK
+
+NAMESPACE_BEGIN(Grid);
+
+class ParRdr {
+private:
+  bool const swap;
+
+  MPI_Status status;
+  MPI_File   fp;
+
+  int err;
+
+  MPI_Datatype oddSiteType;
+  MPI_Datatype fileViewType;
+
+  GridBase* grid;
+
+public:
+  ParRdr(MPI_Comm comm, std::string const& filename, GridBase* gridPtr)
+    : swap(false)
+    , grid(gridPtr) {
+    err = MPI_File_open(comm, const_cast<char*>(filename.c_str()), MPI_MODE_RDONLY, MPI_INFO_NULL, &fp);
+    assert(err == MPI_SUCCESS);
+  }
+
+  virtual ~ParRdr() { MPI_File_close(&fp); }
+
+  inline void errInfo(int const err, std::string const& func) {
+    static char estring[MPI_MAX_ERROR_STRING];
+    int         eclass = -1, len = 0;
+    MPI_Error_class(err, &eclass);
+    MPI_Error_string(err, estring, &len);
+    std::cerr << func << " - Error " << eclass << ": " << estring << std::endl;
+  }
+
+  int readHeader(FieldMetaData& field) {
+    assert((grid->_ndimension == Nd) && (Nd == 4));
+    assert(Nc == 3);
+
+    OpenQcdHeader header;
+
+    readBlock(reinterpret_cast<char*>(&header), 0, sizeof(OpenQcdHeader), MPI_CHAR);
+
+    header.plaq /= 3.; // TODO change this into normalizationfactor
+
+    // sanity check (should trigger on endian issues) TODO remove?
+    assert(0 < header.Nt && header.Nt <= 1024);
+    assert(0 < header.Nx && header.Nx <= 1024);
+    assert(0 < header.Ny && header.Ny <= 1024);
+    assert(0 < header.Nz && header.Nz <= 1024);
+
+    field.dimension[0] = header.Nx;
+    field.dimension[1] = header.Ny;
+    field.dimension[2] = header.Nz;
+    field.dimension[3] = header.Nt;
+
+    for(int d = 0; d < Nd; d++)
+      assert(grid->FullDimensions()[d] == field.dimension[d]);
+
+    field.plaquette = header.plaq;
+
+    field.data_start = sizeof(OpenQcdHeader);
+
+    return field.data_start;
+  }
+
+  void readBlock(void* const dest, uint64_t const pos, uint64_t const nbytes, MPI_Datatype const datatype) {
+    err = MPI_File_read_at_all(fp, pos, dest, nbytes, datatype, &status);
+    errInfo(err, "MPI_File_read_at_all");
+    // CHECK_VAR(err)
+
+    int read = -1;
+    MPI_Get_count(&status, datatype, &read);
+    // CHECK_VAR(read)
+    assert(nbytes == (uint64_t)read);
+    assert(err == MPI_SUCCESS);
+  }
+
+  void createTypes() {
+    constexpr int elem_size = Nd * 2 * 2 * Nc * Nc * sizeof(double); // 2_complex 2_fwdbwd
+
+    err = MPI_Type_contiguous(elem_size, MPI_BYTE, &oddSiteType); assert(err == MPI_SUCCESS);
+    err = MPI_Type_commit(&oddSiteType); assert(err == MPI_SUCCESS);
+
+    Coordinate const L = grid->GlobalDimensions();
+    Coordinate const l = grid->LocalDimensions();
+    Coordinate const i = grid->ThisProcessorCoor();
+
+    Coordinate sizes({L[2] / 2, L[1], L[0], L[3]});
+    Coordinate subsizes({l[2] / 2, l[1], l[0], l[3]});
+    Coordinate starts({i[2] * l[2] / 2, i[1] * l[1], i[0] * l[0], i[3] * l[3]});
+
+    err = MPI_Type_create_subarray(grid->_ndimension, &sizes[0], &subsizes[0], &starts[0], MPI_ORDER_FORTRAN, oddSiteType, &fileViewType); assert(err == MPI_SUCCESS);
+    err = MPI_Type_commit(&fileViewType); assert(err == MPI_SUCCESS);
+  }
+
+  void freeTypes() {
+    err = MPI_Type_free(&fileViewType); assert(err == MPI_SUCCESS);
+    err = MPI_Type_free(&oddSiteType); assert(err == MPI_SUCCESS);
+  }
+
+  bool readGauge(std::vector<ColourMatrixD>& domain_buff, FieldMetaData& meta) {
+    auto hdr_offset = readHeader(meta);
+    CHECK
+    createTypes();
+    err = MPI_File_set_view(fp, hdr_offset, oddSiteType, fileViewType, "native", MPI_INFO_NULL); errInfo(err, "MPI_File_set_view0"); assert(err == MPI_SUCCESS);
+    CHECK
+    int const domainSites = grid->lSites();
+    domain_buff.resize(Nd * domainSites); // 2_fwdbwd * 4_Nd * domainSites / 2_onlyodd
+
+    // the actual READ
+    constexpr uint64_t cm_size   = 2 * Nc * Nc * sizeof(double);    // 2_complex
+    constexpr uint64_t os_size   = Nd * 2 * cm_size;                // 2_fwdbwd
+    constexpr uint64_t max_elems = std::numeric_limits<int>::max(); // int adressable elems: floor is fine
+    uint64_t const     n_os      = domainSites / 2;
+
+    for(uint64_t os_idx = 0; os_idx < n_os;) {
+      uint64_t const read_os = os_idx + max_elems <= n_os ? max_elems : n_os - os_idx;
+      uint64_t const cm      = os_idx * Nd * 2;
+      readBlock(&(domain_buff[cm]), os_idx, read_os, oddSiteType);
+      os_idx += read_os;
+    }
+
+    CHECK
+    err = MPI_File_set_view(fp, 0, MPI_BYTE, MPI_BYTE, "native", MPI_INFO_NULL);
+  errInfo(err, "MPI_File_set_view1");
+    assert(err == MPI_SUCCESS);
+    freeTypes();
+
+    std::cout << GridLogMessage << "read sum: " << n_os * os_size << " bytes" << std::endl;
+    return true;
+  }
+};
+
+class OpenQcdIOChromaReference : public BinaryIO {
+public:
+  template<class vsimd>
+  static inline void readConfiguration(Lattice<iLorentzColourMatrix<vsimd>>& Umu,
+                                       Grid::FieldMetaData&                  header,
+                                       std::string                           file) {
+    typedef Lattice<iDoubleStoredColourMatrix<vsimd>> DoubledGaugeField;
+
+    assert(Ns == 4 and Nd == 4 and Nc == 3);
+
+    auto grid = Umu.Grid();
+
+    typedef ColourMatrixD fobj;
+
+    std::vector<fobj> iodata(
+      Nd * grid->lSites()); // actual size = 2*Nd*lsites but have only lsites/2 sites in file
+
+    {
+      ParRdr rdr(MPI_COMM_WORLD, file, grid);
+      rdr.readGauge(iodata, header);
+    } // equivalent to using binaryio
+
+    std::vector<iDoubleStoredColourMatrix<typename vsimd::scalar_type>> Umu_ds_scalar(grid->lSites());
+
+    copyToLatticeObject(Umu_ds_scalar, iodata, grid); // equivalent to munging
+
+    DoubledGaugeField Umu_ds(grid);
+
+    vectorizeFromLexOrdArray(Umu_ds_scalar, Umu_ds);
+
+    redistribute(Umu, Umu_ds); // equivalent to undoDoublestore
+
+    FieldMetaData clone(header);
+
+    GaugeStatistics(Umu, clone);
+
+    RealD plaq_diff = fabs(clone.plaquette - header.plaquette);
+
+    // clang-format off
+    std::cout << GridLogMessage << "OpenQcd Configuration " << file
+              << " plaquette " << clone.plaquette
+              << " header " << header.plaquette
+              << " difference " << plaq_diff
+              << std::endl;
+    // clang-format on
+
+    RealD precTol = (getPrecision<vsimd>::value == 1) ? 2e-7 : 2e-15;
+    RealD tol     = precTol * std::sqrt(grid->_Nprocessors); // taken from RQCD chroma code
+
+    if(plaq_diff >= tol)
+      std::cout << " Plaquette mismatch (diff = " << plaq_diff << ", tol = " << tol << ")" << std::endl;
+    assert(plaq_diff < tol);
+
+    std::cout << GridLogMessage << "OpenQcd Configuration " << file << " and plaquette agree" << std::endl;
+  }
+
+private:
+  template<class vsimd>
+  static inline void redistribute(Lattice<iLorentzColourMatrix<vsimd>>&            Umu,
+                                  Lattice<iDoubleStoredColourMatrix<vsimd>> const& Umu_ds) {
+    Grid::conformable(Umu.Grid(), Umu_ds.Grid());
+    Lattice<iColourMatrix<vsimd>> U(Umu.Grid());
+
+    U = PeekIndex<LorentzIndex>(Umu_ds, 2) + Cshift(PeekIndex<LorentzIndex>(Umu_ds, 3), 0, +1); PokeIndex<LorentzIndex>(Umu, U, 0);
+    U = PeekIndex<LorentzIndex>(Umu_ds, 4) + Cshift(PeekIndex<LorentzIndex>(Umu_ds, 5), 1, +1); PokeIndex<LorentzIndex>(Umu, U, 1);
+    U = PeekIndex<LorentzIndex>(Umu_ds, 6) + Cshift(PeekIndex<LorentzIndex>(Umu_ds, 7), 2, +1); PokeIndex<LorentzIndex>(Umu, U, 2);
+    U = PeekIndex<LorentzIndex>(Umu_ds, 0) + Cshift(PeekIndex<LorentzIndex>(Umu_ds, 1), 3, +1); PokeIndex<LorentzIndex>(Umu, U, 3);
+  }
+
+  static inline void copyToLatticeObject(std::vector<DoubleStoredColourMatrix>& u_fb,
+                                         std::vector<ColourMatrixD> const&      node_buff,
+                                         GridBase*                              grid) {
+    assert(node_buff.size() == Nd * grid->lSites());
+
+    Coordinate const& l = grid->LocalDimensions();
+
+    Coordinate coord(Nd);
+    int&       x = coord[0];
+    int&       y = coord[1];
+    int&       z = coord[2];
+    int&       t = coord[3];
+
+    int buff_idx = 0;
+    for(t = 0; t < l[3]; ++t) // IMPORTANT: openQCD file ordering
+      for(x = 0; x < l[0]; ++x)
+        for(y = 0; y < l[1]; ++y)
+          for(z = 0; z < l[2]; ++z) {
+            if((t + z + y + x) % 2 == 0) continue;
+
+            int local_idx;
+            Lexicographic::IndexFromCoor(coord, local_idx, grid->LocalDimensions());
+            for(int mu = 0; mu < 2 * Nd; ++mu)
+              for(int c1 = 0; c1 < Nc; ++c1) {
+                for(int c2 = 0; c2 < Nc; ++c2) {
+                  u_fb[local_idx](mu)()(c1,c2) = node_buff[mu+buff_idx]()()(c1,c2);
+                }
+              }
+            buff_idx += 2 * Nd;
+          }
+
+    assert(node_buff.size() == buff_idx);
+  }
+};
+
+NAMESPACE_END(Grid);
diff --git a/Grid/perfmon/PerfCount.h b/Grid/perfmon/PerfCount.h
index b8229eec..dd25b41e 100644
--- a/Grid/perfmon/PerfCount.h
+++ b/Grid/perfmon/PerfCount.h
@@ -44,7 +44,7 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 #include <sys/syscall.h>
 #endif
 #ifdef __x86_64__
-#ifdef GRID_NVCC
+#ifdef GRID_CUDA
 accelerator_inline uint64_t __rdtsc(void) {  return 0; }
 accelerator_inline uint64_t __rdpmc(int ) {  return 0; }
 #else
@@ -95,7 +95,8 @@ inline uint64_t cyclecount(void){
 }
 #elif defined __x86_64__
 inline uint64_t cyclecount(void){ 
-  return __rdtsc();
+  uint64_t ret = __rdtsc();
+  return (uint64_t)ret;
 }
 #else
 
@@ -111,7 +112,6 @@ class PerformanceCounter {
 private:
 
   typedef struct { 
-  public:
     uint32_t type;
     uint64_t config;
     const char *name;
diff --git a/Grid/perfmon/Timer.h b/Grid/perfmon/Timer.h
index 88b4e1cc..2a44faee 100644
--- a/Grid/perfmon/Timer.h
+++ b/Grid/perfmon/Timer.h
@@ -110,15 +110,15 @@ public:
 #endif
     accumulator = std::chrono::duration_cast<GridUsecs>(start-start); 
   }
-  GridTime Elapsed(void) {
+  GridTime Elapsed(void) const {
     assert(running == false);
     return std::chrono::duration_cast<GridTime>( accumulator );
   }
-  uint64_t useconds(void){
+  uint64_t useconds(void) const {
     assert(running == false);
     return (uint64_t) accumulator.count();
   }
-  bool isRunning(void){
+  bool isRunning(void) const {
     return running;
   }
 };
diff --git a/Grid/pugixml/pugixml.cc b/Grid/pugixml/pugixml.cc
index e7b395ad..45e6496a 100644
--- a/Grid/pugixml/pugixml.cc
+++ b/Grid/pugixml/pugixml.cc
@@ -12773,7 +12773,7 @@ namespace pugi
 #undef PUGI__THROW_ERROR
 #undef PUGI__CHECK_ERROR
 
-#ifdef GRID_NVCC
+#ifdef GRID_CUDA
 #pragma pop
 #endif
 
diff --git a/Grid/qcd/QCD.h b/Grid/qcd/QCD.h
index 2c8e60da..faacac63 100644
--- a/Grid/qcd/QCD.h
+++ b/Grid/qcd/QCD.h
@@ -133,23 +133,23 @@ typedef iSpinColourMatrix<vComplex >    vSpinColourMatrix;
 typedef iSpinColourMatrix<vComplexF>    vSpinColourMatrixF;
 typedef iSpinColourMatrix<vComplexD>    vSpinColourMatrixD;
 
-    // SpinColourSpinColour matrix
-    typedef iSpinColourSpinColourMatrix<Complex  >    SpinColourSpinColourMatrix;
-    typedef iSpinColourSpinColourMatrix<ComplexF >    SpinColourSpinColourMatrixF;
-    typedef iSpinColourSpinColourMatrix<ComplexD >    SpinColourSpinColourMatrixD;
+// SpinColourSpinColour matrix
+typedef iSpinColourSpinColourMatrix<Complex  >    SpinColourSpinColourMatrix;
+typedef iSpinColourSpinColourMatrix<ComplexF >    SpinColourSpinColourMatrixF;
+typedef iSpinColourSpinColourMatrix<ComplexD >    SpinColourSpinColourMatrixD;
 
-    typedef iSpinColourSpinColourMatrix<vComplex >    vSpinColourSpinColourMatrix;
-    typedef iSpinColourSpinColourMatrix<vComplexF>    vSpinColourSpinColourMatrixF;
-    typedef iSpinColourSpinColourMatrix<vComplexD>    vSpinColourSpinColourMatrixD;
+typedef iSpinColourSpinColourMatrix<vComplex >    vSpinColourSpinColourMatrix;
+typedef iSpinColourSpinColourMatrix<vComplexF>    vSpinColourSpinColourMatrixF;
+typedef iSpinColourSpinColourMatrix<vComplexD>    vSpinColourSpinColourMatrixD;
 
-    // SpinColourSpinColour matrix
-    typedef iSpinColourSpinColourMatrix<Complex  >    SpinColourSpinColourMatrix;
-    typedef iSpinColourSpinColourMatrix<ComplexF >    SpinColourSpinColourMatrixF;
-    typedef iSpinColourSpinColourMatrix<ComplexD >    SpinColourSpinColourMatrixD;
+// SpinColourSpinColour matrix
+typedef iSpinColourSpinColourMatrix<Complex  >    SpinColourSpinColourMatrix;
+typedef iSpinColourSpinColourMatrix<ComplexF >    SpinColourSpinColourMatrixF;
+typedef iSpinColourSpinColourMatrix<ComplexD >    SpinColourSpinColourMatrixD;
 
-    typedef iSpinColourSpinColourMatrix<vComplex >    vSpinColourSpinColourMatrix;
-    typedef iSpinColourSpinColourMatrix<vComplexF>    vSpinColourSpinColourMatrixF;
-    typedef iSpinColourSpinColourMatrix<vComplexD>    vSpinColourSpinColourMatrixD;
+typedef iSpinColourSpinColourMatrix<vComplex >    vSpinColourSpinColourMatrix;
+typedef iSpinColourSpinColourMatrix<vComplexF>    vSpinColourSpinColourMatrixF;
+typedef iSpinColourSpinColourMatrix<vComplexD>    vSpinColourSpinColourMatrixD;
 
 // LorentzColour
 typedef iLorentzColourMatrix<Complex  > LorentzColourMatrix;
@@ -443,16 +443,16 @@ template<class vobj> void pokeLorentz(vobj &lhs,const decltype(peekIndex<Lorentz
 //////////////////////////////////////////////
 // Fermion <-> propagator assignements
 //////////////////////////////////////////////
-    //template <class Prop, class Ferm>
-    template <class Fimpl>
-      void FermToProp(typename Fimpl::PropagatorField &p, const typename Fimpl::FermionField &f, const int s, const int c)
+//template <class Prop, class Ferm>
+template <class Fimpl>
+void FermToProp(typename Fimpl::PropagatorField &p, const typename Fimpl::FermionField &f, const int s, const int c)
 {
   for(int j = 0; j < Ns; ++j)
     {
       auto pjs = peekSpin(p, j, s);
       auto fj  = peekSpin(f, j);
             
-            for(int i = 0; i < Fimpl::Dimension; ++i)
+      for(int i = 0; i < Fimpl::Dimension; ++i)
 	{
 	  pokeColour(pjs, peekColour(fj, i), i, c);
 	}
@@ -460,16 +460,16 @@ template<class vobj> void pokeLorentz(vobj &lhs,const decltype(peekIndex<Lorentz
     }
 }
     
-    //template <class Prop, class Ferm>
-    template <class Fimpl>
-      void PropToFerm(typename Fimpl::FermionField &f, const typename Fimpl::PropagatorField &p, const int s, const int c)
+//template <class Prop, class Ferm>
+template <class Fimpl>
+void PropToFerm(typename Fimpl::FermionField &f, const typename Fimpl::PropagatorField &p, const int s, const int c)
 {
   for(int j = 0; j < Ns; ++j)
     {
       auto pjs = peekSpin(p, j, s);
       auto fj  = peekSpin(f, j);
             
-            for(int i = 0; i < Fimpl::Dimension; ++i)
+      for(int i = 0; i < Fimpl::Dimension; ++i)
 	{
 	  pokeColour(fj, peekColour(pjs, i, c), i);
 	}
diff --git a/Grid/qcd/action/fermion/CayleyFermion5D.h b/Grid/qcd/action/fermion/CayleyFermion5D.h
index c2ccb98b..c7d68d73 100644
--- a/Grid/qcd/action/fermion/CayleyFermion5D.h
+++ b/Grid/qcd/action/fermion/CayleyFermion5D.h
@@ -40,8 +40,8 @@ public:
 public:
 
   // override multiply
-  virtual RealD  M    (const FermionField &in, FermionField &out);
-  virtual RealD  Mdag (const FermionField &in, FermionField &out);
+  virtual void   M    (const FermionField &in, FermionField &out);
+  virtual void   Mdag (const FermionField &in, FermionField &out);
 
   // half checkerboard operations
   virtual void   Meooe       (const FermionField &in, FermionField &out);
@@ -141,7 +141,33 @@ public:
   Vector<iSinglet<Simd> >  MatpInvDag;
   Vector<iSinglet<Simd> >  MatmInvDag;
 
+  ///////////////////////////////////////////////////////////////
+  // Conserved current utilities
+  ///////////////////////////////////////////////////////////////
+
+  // Virtual can't template
+  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);
+
+  void ContractJ5q(PropagatorField &q_in,ComplexField &J5q);
+  void ContractJ5q(FermionField &q_in,ComplexField &J5q);
+
+  ///////////////////////////////////////////////////////////////
   // Constructors
+  ///////////////////////////////////////////////////////////////
   CayleyFermion5D(GaugeField &_Umu,
 		  GridCartesian         &FiveDimGrid,
 		  GridRedBlackCartesian &FiveDimRedBlackGrid,
diff --git a/Grid/qcd/action/fermion/ContinuedFractionFermion5D.h b/Grid/qcd/action/fermion/ContinuedFractionFermion5D.h
index 5aa7bfbd..2300afd3 100644
--- a/Grid/qcd/action/fermion/ContinuedFractionFermion5D.h
+++ b/Grid/qcd/action/fermion/ContinuedFractionFermion5D.h
@@ -41,8 +41,8 @@ public:
 public:
 
   // override multiply
-  virtual RealD  M    (const FermionField &in, FermionField &out);
-  virtual RealD  Mdag (const FermionField &in, FermionField &out);
+  virtual void   M    (const FermionField &in, FermionField &out);
+  virtual void   Mdag (const FermionField &in, FermionField &out);
 
   // half checkerboard operaions
   virtual void   Meooe       (const FermionField &in, FermionField &out);
diff --git a/Grid/qcd/action/fermion/DomainWallEOFAFermion.h b/Grid/qcd/action/fermion/DomainWallEOFAFermion.h
index a2d0e733..bcc97176 100644
--- a/Grid/qcd/action/fermion/DomainWallEOFAFermion.h
+++ b/Grid/qcd/action/fermion/DomainWallEOFAFermion.h
@@ -53,8 +53,8 @@ public:
   virtual void  DtildeInv  (const FermionField& in, FermionField& out);
 
   // override multiply
-  virtual RealD M          (const FermionField& in, FermionField& out);
-  virtual RealD Mdag       (const FermionField& in, FermionField& out);
+  virtual void  M          (const FermionField& in, FermionField& out);
+  virtual void  Mdag       (const FermionField& in, FermionField& out);
 
   // half checkerboard operations
   virtual void  Mooee      (const FermionField& in, FermionField& out);
diff --git a/Grid/qcd/action/fermion/DomainWallVec5dImpl.h b/Grid/qcd/action/fermion/DomainWallVec5dImpl.h
index 890c680b..0c8a0930 100644
--- a/Grid/qcd/action/fermion/DomainWallVec5dImpl.h
+++ b/Grid/qcd/action/fermion/DomainWallVec5dImpl.h
@@ -114,19 +114,22 @@ public:
       U = adj(Cshift(U, mu, -1));
       PokeIndex<LorentzIndex>(Uadj, U, mu);
     }
-    
-    for (int lidx = 0; lidx < GaugeGrid->lSites(); lidx++) {
+
+    autoView(Umu_v,Umu,CpuRead);
+    autoView(Uadj_v,Uadj,CpuRead);
+    autoView(Uds_v,Uds,CpuWrite);
+    thread_for( lidx, GaugeGrid->lSites(), {
       Coordinate lcoor;
       GaugeGrid->LocalIndexToLocalCoor(lidx, lcoor);
       
-      peekLocalSite(ScalarUmu, Umu, lcoor);
+      peekLocalSite(ScalarUmu, Umu_v, lcoor);
       for (int mu = 0; mu < 4; mu++) ScalarUds(mu) = ScalarUmu(mu);
       
-      peekLocalSite(ScalarUmu, Uadj, lcoor);
+      peekLocalSite(ScalarUmu, Uadj_v, lcoor);
       for (int mu = 0; mu < 4; mu++) ScalarUds(mu + 4) = ScalarUmu(mu);
       
-      pokeLocalSite(ScalarUds, Uds, lcoor);
-    }
+      pokeLocalSite(ScalarUds, Uds_v, lcoor);
+    });
   }
       
   inline void InsertForce4D(GaugeField &mat, FermionField &Btilde,FermionField &A, int mu) 
diff --git a/Grid/qcd/action/fermion/Fermion.h b/Grid/qcd/action/fermion/Fermion.h
index fb6f18bb..16252340 100644
--- a/Grid/qcd/action/fermion/Fermion.h
+++ b/Grid/qcd/action/fermion/Fermion.h
@@ -57,6 +57,7 @@ NAMESPACE_CHECK(WilsonClover);
 #include <Grid/qcd/action/fermion/WilsonFermion5D.h>     // 5d base used by all 5d overlap types
 NAMESPACE_CHECK(Wilson5D);
 
+#include <Grid/qcd/action/fermion/NaiveStaggeredFermion.h>
 #include <Grid/qcd/action/fermion/ImprovedStaggeredFermion.h>
 #include <Grid/qcd/action/fermion/ImprovedStaggeredFermion5D.h>
 NAMESPACE_CHECK(Staggered);
@@ -282,11 +283,15 @@ 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;
 
-#ifndef GRID_NVCC
+#ifndef GRID_CUDA
 typedef ImprovedStaggeredFermion5D<StaggeredVec5dImplR> ImprovedStaggeredFermionVec5dR;
 typedef ImprovedStaggeredFermion5D<StaggeredVec5dImplF> ImprovedStaggeredFermionVec5dF;
 typedef ImprovedStaggeredFermion5D<StaggeredVec5dImplD> ImprovedStaggeredFermionVec5dD;
diff --git a/Grid/qcd/action/fermion/FermionOperator.h b/Grid/qcd/action/fermion/FermionOperator.h
index cbc6ca63..570e350d 100644
--- a/Grid/qcd/action/fermion/FermionOperator.h
+++ b/Grid/qcd/action/fermion/FermionOperator.h
@@ -58,8 +58,8 @@ public:
   virtual GridBase *GaugeRedBlackGrid(void)   =0;
 
   // override multiply
-  virtual RealD  M    (const FermionField &in, FermionField &out)=0;
-  virtual RealD  Mdag (const FermionField &in, FermionField &out)=0;
+  virtual void  M    (const FermionField &in, FermionField &out)=0;
+  virtual void  Mdag (const FermionField &in, FermionField &out)=0;
 
   // half checkerboard operaions
   virtual void   Meooe       (const FermionField &in, FermionField &out)=0;
@@ -86,15 +86,14 @@ public:
   virtual void DhopDerivEO(GaugeField &mat,const FermionField &U,const FermionField &V,int dag)=0;
   virtual void DhopDerivOE(GaugeField &mat,const FermionField &U,const FermionField &V,int dag)=0;
 
-
   virtual void  Mdiag  (const FermionField &in, FermionField &out) { Mooee(in,out);};   // Same as Mooee applied to both CB's
   virtual void  Mdir   (const FermionField &in, FermionField &out,int dir,int disp)=0;   // case by case Wilson, Clover, Cayley, ContFrac, PartFrac
   virtual void  MdirAll(const FermionField &in, std::vector<FermionField> &out)=0;   // case by case Wilson, Clover, Cayley, ContFrac, PartFrac
 
 
-      virtual void  MomentumSpacePropagator(FermionField &out,const FermionField &in,RealD _m,std::vector<double> twist) { assert(0);};
+  virtual void  MomentumSpacePropagator(FermionField &out,const FermionField &in,RealD _m,std::vector<double> twist) { assert(0);};
 
-      virtual void  FreePropagator(const FermionField &in,FermionField &out,RealD mass,std::vector<Complex> boundary,std::vector<double> twist) 
+  virtual void  FreePropagator(const FermionField &in,FermionField &out,RealD mass,std::vector<Complex> boundary,std::vector<double> twist) 
       {
 	FFT theFFT((GridCartesian *) in.Grid());
 
@@ -148,15 +147,19 @@ public:
   virtual void ContractConservedCurrent(PropagatorField &q_in_1,
 					PropagatorField &q_in_2,
 					PropagatorField &q_out,
+					PropagatorField &phys_src,
 					Current curr_type,
-					unsigned int mu)=0;
+					unsigned int mu)
+  {assert(0);};
   virtual 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)=0;
+				   ComplexField &lattice_cmplx)
+  {assert(0);};
 
       // Only reimplemented in Wilson5D 
       // Default to just a zero correlation function
diff --git a/Grid/qcd/action/fermion/GparityWilsonImpl.h b/Grid/qcd/action/fermion/GparityWilsonImpl.h
index f87d2260..0b726db9 100644
--- a/Grid/qcd/action/fermion/GparityWilsonImpl.h
+++ b/Grid/qcd/action/fermion/GparityWilsonImpl.h
@@ -38,6 +38,7 @@ public:
  static const bool isFundamental = Representation::isFundamental;
  static const int Nhcs = Options::Nhcs;
  static const bool LsVectorised=false;
+ static const bool isGparity=true;
 
  typedef ConjugateGaugeImpl< GaugeImplTypes<S,Dimension> > Gimpl;
  INHERIT_GIMPL_TYPES(Gimpl);
@@ -46,7 +47,7 @@ public:
  typedef typename Options::template PrecisionMapper<Simd>::LowerPrecVector SimdL;
       
  template <typename vtype> using iImplSpinor            = iVector<iVector<iVector<vtype, Dimension>, Ns>,   Ngp>;
- template <typename vtype> using iImplPropagator        = iVector<iMatrix<iMatrix<vtype, Dimension>, Ns>,   Ngp>;
+ template <typename vtype> using iImplPropagator        = iMatrix<iMatrix<iMatrix<vtype, Dimension>, Ns>,   Ngp>;
  template <typename vtype> using iImplHalfSpinor        = iVector<iVector<iVector<vtype, Dimension>, Nhs>,  Ngp>;
  template <typename vtype> using iImplHalfCommSpinor    = iVector<iVector<iVector<vtype, Dimension>, Nhcs>, Ngp>;
  template <typename vtype> using iImplDoubledGaugeField = iVector<iVector<iScalar<iMatrix<vtype, Dimension> >, Nds>, Ngp>;
@@ -80,6 +81,7 @@ public:
   {
     assert(0);
   } 
+
   template<class _Spinor>
   static accelerator_inline void multLink(_Spinor &phi, 
 					  const SiteDoubledGaugeField &U,
@@ -94,11 +96,11 @@ public:
     int sl        = St._simd_layout[direction];
     Coordinate icoor;
 
-#ifdef __CUDA_ARCH__
+#ifdef GRID_SIMT
     _Spinor tmp;
 
     const int Nsimd =SiteDoubledGaugeField::Nsimd();
-    int s = SIMTlane(Nsimd);
+    int s = acceleratorSIMTlane(Nsimd);
     St.iCoorFromIindex(icoor,s);
 
     int mmu = mu % Nd;
@@ -191,6 +193,16 @@ public:
 #endif   
   }
 
+
+  template<class _SpinorField>
+  inline void multLinkField(_SpinorField & out,
+			    const DoubledGaugeField &Umu,
+			    const _SpinorField & phi,
+			    int mu)
+  {
+    assert(0);
+  }
+
   template <class ref>
   static accelerator_inline void loadLinkElement(Simd &reg, ref &memory) 
   {
@@ -220,15 +232,17 @@ public:
       if ( Params.twists[mu] ) { 
 	Uconj = where(coor==neglink,-Uconj,Uconj);
       }
-	  
-      auto U_v = U.View();
-      auto Uds_v = Uds.View();
-      auto Uconj_v = Uconj.View();
-      auto Utmp_v= Utmp.View();
-      thread_foreach(ss,U_v,{
-	Uds_v[ss](0)(mu) = U_v[ss]();
-	Uds_v[ss](1)(mu) = Uconj_v[ss]();
-      });
+
+      {
+	autoView( U_v , U, CpuRead);
+	autoView( Uconj_v , Uconj, CpuRead);
+	autoView( Uds_v , Uds, CpuWrite);
+	autoView( Utmp_v, Utmp, CpuWrite);
+	thread_foreach(ss,U_v,{
+	    Uds_v[ss](0)(mu) = U_v[ss]();
+	    Uds_v[ss](1)(mu) = Uconj_v[ss]();
+	  });
+      }
           
       U     = adj(Cshift(U    ,mu,-1));      // correct except for spanning the boundary
       Uconj = adj(Cshift(Uconj,mu,-1));
@@ -238,19 +252,25 @@ public:
 	Utmp = where(coor==0,Uconj,Utmp);
       }
 
-      thread_foreach(ss,Utmp_v,{
-	Uds_v[ss](0)(mu+4) = Utmp_v[ss]();
-      });
-          
+      {
+	autoView( Uds_v , Uds, CpuWrite);
+	autoView( Utmp_v, Utmp, CpuWrite);
+	thread_foreach(ss,Utmp_v,{
+	    Uds_v[ss](0)(mu+4) = Utmp_v[ss]();
+	  });
+      }
       Utmp = Uconj;
       if ( Params.twists[mu] ) { 
 	Utmp = where(coor==0,U,Utmp);
       }
-	  
-      thread_foreach(ss,Utmp_v,{
-        Uds_v[ss](1)(mu+4) = Utmp_v[ss]();
-      });
-          
+
+      {	  
+	autoView( Uds_v , Uds, CpuWrite);
+	autoView( Utmp_v, Utmp, CpuWrite);
+	thread_foreach(ss,Utmp_v,{
+	    Uds_v[ss](1)(mu+4) = Utmp_v[ss]();
+        });
+      }
     }
   }
       
@@ -260,11 +280,14 @@ public:
     GaugeLinkField link(mat.Grid());
     // use lorentz for flavour as hack.
     auto tmp = TraceIndex<SpinIndex>(outerProduct(Btilde, A));
-    auto link_v = link.View();
-    auto tmp_v = tmp.View();
-    thread_foreach(ss,tmp_v,{
-      link_v[ss]() = tmp_v[ss](0, 0) + conjugate(tmp_v[ss](1, 1));
-    });
+
+    {
+      autoView( link_v , link, CpuWrite);
+      autoView( tmp_v , tmp, CpuRead);
+      thread_foreach(ss,tmp_v,{
+	  link_v[ss]() = tmp_v[ss](0, 0) + conjugate(tmp_v[ss](1, 1));
+	});
+    }
     PokeIndex<LorentzIndex>(mat, link, mu);
     return;
   }
@@ -294,16 +317,18 @@ public:
         
     GaugeLinkField tmp(mat.Grid());
     tmp = Zero();
-    auto tmp_v = tmp.View();
-    auto Atilde_v = Atilde.View();
-    auto Btilde_v = Btilde.View();
-    thread_for(ss,tmp.Grid()->oSites(),{
-      for (int s = 0; s < Ls; s++) {
-	int sF = s + Ls * ss;
-	auto ttmp = traceIndex<SpinIndex>(outerProduct(Btilde_v[sF], Atilde_v[sF]));
-	tmp_v[ss]() = tmp_v[ss]() + ttmp(0, 0) + conjugate(ttmp(1, 1));
-      }
-    });
+    {
+      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;
   }
diff --git a/Grid/qcd/action/fermion/ImprovedStaggeredFermion.h b/Grid/qcd/action/fermion/ImprovedStaggeredFermion.h
index 5cb95ca6..ecf44ed7 100644
--- a/Grid/qcd/action/fermion/ImprovedStaggeredFermion.h
+++ b/Grid/qcd/action/fermion/ImprovedStaggeredFermion.h
@@ -71,8 +71,8 @@ public:
   // override multiply; cut number routines if pass dagger argument
   // and also make interface more uniformly consistent
   //////////////////////////////////////////////////////////////////
-  RealD M(const FermionField &in, FermionField &out);
-  RealD Mdag(const FermionField &in, FermionField &out);
+  void M(const FermionField &in, FermionField &out);
+  void Mdag(const FermionField &in, FermionField &out);
 
   /////////////////////////////////////////////////////////
   // half checkerboard operations
@@ -185,10 +185,12 @@ public:
   void ContractConservedCurrent(PropagatorField &q_in_1,
                                 PropagatorField &q_in_2,
                                 PropagatorField &q_out,
+                                PropagatorField &src,
                                 Current curr_type,
                                 unsigned int mu);
   void SeqConservedCurrent(PropagatorField &q_in,
                            PropagatorField &q_out,
+                           PropagatorField &srct,
                            Current curr_type,
                            unsigned int mu, 
                            unsigned int tmin,
diff --git a/Grid/qcd/action/fermion/ImprovedStaggeredFermion5D.h b/Grid/qcd/action/fermion/ImprovedStaggeredFermion5D.h
index 8e3d4be5..625eda63 100644
--- a/Grid/qcd/action/fermion/ImprovedStaggeredFermion5D.h
+++ b/Grid/qcd/action/fermion/ImprovedStaggeredFermion5D.h
@@ -1,4 +1,3 @@
-
 /*************************************************************************************
 
     Grid physics library, www.github.com/paboyle/Grid 
@@ -62,8 +61,8 @@ public:
   double DhopCalls;
   double DhopCommTime;
   double DhopComputeTime;
-      double DhopComputeTime2;
-      double DhopFaceTime;
+  double DhopComputeTime2;
+  double DhopFaceTime;
 
   ///////////////////////////////////////////////////////////////
   // Implement the abstract base
@@ -74,8 +73,8 @@ public:
   GridBase *FermionRedBlackGrid(void)    { return _FiveDimRedBlackGrid;}
 
   // full checkerboard operations; leave unimplemented as abstract for now
-  RealD  M    (const FermionField &in, FermionField &out);
-  RealD  Mdag (const FermionField &in, FermionField &out);
+  void  M    (const FermionField &in, FermionField &out);
+  void  Mdag (const FermionField &in, FermionField &out);
 
   // half checkerboard operations
   void   Meooe       (const FermionField &in, FermionField &out);
@@ -217,15 +216,17 @@ public:
   void ContractConservedCurrent(PropagatorField &q_in_1,
 				PropagatorField &q_in_2,
 				PropagatorField &q_out,
+				PropagatorField &src,
 				Current curr_type,
 				unsigned int mu);
   void SeqConservedCurrent(PropagatorField &q_in,
 			   PropagatorField &q_out,
+			   PropagatorField &src,
 			   Current curr_type,
 			   unsigned int mu, 
 			   unsigned int tmin,
-                             unsigned int tmax,
-                 	     ComplexField &lattice_cmplx);
+			   unsigned int tmax,
+			   ComplexField &lattice_cmplx);
 };
 
 NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/MobiusEOFAFermion.h b/Grid/qcd/action/fermion/MobiusEOFAFermion.h
index 6b214233..6e4f79eb 100644
--- a/Grid/qcd/action/fermion/MobiusEOFAFermion.h
+++ b/Grid/qcd/action/fermion/MobiusEOFAFermion.h
@@ -56,8 +56,8 @@ public:
   virtual void  DtildeInv        (const FermionField& in, FermionField& out);
 
   // override multiply
-  virtual RealD M                (const FermionField& in, FermionField& out);
-  virtual RealD Mdag             (const FermionField& in, FermionField& out);
+  virtual void  M                (const FermionField& in, FermionField& out);
+  virtual void  Mdag             (const FermionField& in, FermionField& out);
 
   // half checkerboard operations
   virtual void  Mooee            (const FermionField& in, FermionField& out);
diff --git a/Grid/qcd/action/fermion/MobiusFermion.h b/Grid/qcd/action/fermion/MobiusFermion.h
index 1cbb6609..1e948092 100644
--- a/Grid/qcd/action/fermion/MobiusFermion.h
+++ b/Grid/qcd/action/fermion/MobiusFermion.h
@@ -59,7 +59,7 @@ public:
   {
     RealD eps = 1.0;
 
-    std::cout<<GridLogMessage << "MobiusFermion (b="<<b<<",c="<<c<<") with Ls= "<<this->Ls<<" Tanh approx"<<std::endl;
+    //    std::cout<<GridLogMessage << "MobiusFermion (b="<<b<<",c="<<c<<") with Ls= "<<this->Ls<<" Tanh approx"<<std::endl;
     Approx::zolotarev_data *zdata = Approx::higham(eps,this->Ls);// eps is ignored for higham
     assert(zdata->n==this->Ls);
 	
diff --git a/Grid/qcd/action/fermion/NaiveStaggeredFermion.h b/Grid/qcd/action/fermion/NaiveStaggeredFermion.h
new file mode 100644
index 00000000..ca38a64f
--- /dev/null
+++ b/Grid/qcd/action/fermion/NaiveStaggeredFermion.h
@@ -0,0 +1,194 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid
+
+Source file: ./lib/qcd/action/fermion/ImprovedStaggered.h
+
+Copyright (C) 2015
+
+Author: Azusa Yamaguchi, Peter Boyle
+
+This program is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 2 of the License, or
+(at your option) any later version.
+
+This program is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along
+with this program; if not, write to the Free Software Foundation, Inc.,
+51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+See the full license in the file "LICENSE" in the top level distribution
+directory
+*************************************************************************************/
+			   /*  END LEGAL */
+#ifndef GRID_QCD_NAIVE_STAG_FERMION_H
+#define GRID_QCD_NAIVE_STAG_FERMION_H
+
+NAMESPACE_BEGIN(Grid);
+
+class NaiveStaggeredFermionStatic {
+public:
+  static const std::vector<int> directions;
+  static const std::vector<int> displacements;
+  static const int npoint = 8;
+};
+
+template <class Impl>
+class NaiveStaggeredFermion : public StaggeredKernels<Impl>, public NaiveStaggeredFermionStatic {
+public:
+  INHERIT_IMPL_TYPES(Impl);
+  typedef StaggeredKernels<Impl> Kernels;
+
+  FermionField _tmp;
+  FermionField &tmp(void) { return _tmp; }
+
+  ////////////////////////////////////////
+  // Performance monitoring
+  ////////////////////////////////////////
+  void Report(void);
+  void ZeroCounters(void);
+  double DhopTotalTime;
+  double DhopCalls;
+  double DhopCommTime;
+  double DhopComputeTime;
+  double DhopComputeTime2;
+  double DhopFaceTime;
+
+  ///////////////////////////////////////////////////////////////
+  // Implement the abstract base
+  ///////////////////////////////////////////////////////////////
+  GridBase *GaugeGrid(void) { return _grid; }
+  GridBase *GaugeRedBlackGrid(void) { return _cbgrid; }
+  GridBase *FermionGrid(void) { return _grid; }
+  GridBase *FermionRedBlackGrid(void) { return _cbgrid; }
+
+  //////////////////////////////////////////////////////////////////
+  // override multiply; cut number routines if pass dagger argument
+  // and also make interface more uniformly consistent
+  //////////////////////////////////////////////////////////////////
+  void M(const FermionField &in, FermionField &out);
+  void Mdag(const FermionField &in, FermionField &out);
+
+  /////////////////////////////////////////////////////////
+  // half checkerboard operations
+  /////////////////////////////////////////////////////////
+  void Meooe(const FermionField &in, FermionField &out);
+  void MeooeDag(const FermionField &in, FermionField &out);
+  void Mooee(const FermionField &in, FermionField &out);
+  void MooeeDag(const FermionField &in, FermionField &out);
+  void MooeeInv(const FermionField &in, FermionField &out);
+  void MooeeInvDag(const FermionField &in, FermionField &out);
+
+  ////////////////////////
+  // Derivative interface
+  ////////////////////////
+  // Interface calls an internal routine
+  void DhopDeriv  (GaugeField &mat, const FermionField &U, const FermionField &V, int dag);
+  void DhopDerivOE(GaugeField &mat, const FermionField &U, const FermionField &V, int dag);
+  void DhopDerivEO(GaugeField &mat, const FermionField &U, const FermionField &V, int dag);
+
+  ///////////////////////////////////////////////////////////////
+  // non-hermitian hopping term; half cb or both
+  ///////////////////////////////////////////////////////////////
+  void Dhop  (const FermionField &in, FermionField &out, int dag);
+  void DhopOE(const FermionField &in, FermionField &out, int dag);
+  void DhopEO(const FermionField &in, FermionField &out, int dag);
+
+  ///////////////////////////////////////////////////////////////
+  // Multigrid assistance; force term uses too
+  ///////////////////////////////////////////////////////////////
+  void Mdir(const FermionField &in, FermionField &out, int dir, int disp);
+  void MdirAll(const FermionField &in, std::vector<FermionField> &out);
+  void DhopDir(const FermionField &in, FermionField &out, int dir, int disp);
+
+  ///////////////////////////////////////////////////////////////
+  // Extra methods added by derived
+  ///////////////////////////////////////////////////////////////
+  void DerivInternal(StencilImpl &st, 
+		     DoubledGaugeField &U,
+		     GaugeField &mat, 
+		     const FermionField &A, const FermionField &B, int dag);
+
+  void DhopInternal(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
+                    const FermionField &in, FermionField &out, int dag);
+  void DhopInternalSerialComms(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
+			       const FermionField &in, FermionField &out, int dag);
+  void DhopInternalOverlappedComms(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
+				   const FermionField &in, FermionField &out, int dag);
+
+  //////////////////////////////////////////////////////////////////////////
+  // Grid own interface Constructor
+  //////////////////////////////////////////////////////////////////////////
+  NaiveStaggeredFermion(GaugeField &_U, GridCartesian &Fgrid,
+			GridRedBlackCartesian &Hgrid, RealD _mass,
+			RealD _c1, RealD _u0,
+			const ImplParams &p = ImplParams());
+  NaiveStaggeredFermion(GridCartesian &Fgrid,
+			GridRedBlackCartesian &Hgrid, RealD _mass,
+			RealD _c1, RealD _u0,
+			const ImplParams &p = ImplParams());
+
+  // DoubleStore impl dependent
+  void ImportGauge      (const GaugeField &_U );
+  DoubledGaugeField &GetU(void)   { return Umu ; } ;
+  void CopyGaugeCheckerboards(void);
+
+  ///////////////////////////////////////////////////////////////
+  // Data members require to support the functionality
+  ///////////////////////////////////////////////////////////////
+
+  //    protected:
+public:
+  // any other parameters of action ???
+  virtual int   isTrivialEE(void) { return 1; };
+  virtual RealD Mass(void) { return mass; }
+  RealD mass;
+  RealD u0;
+  RealD c1;
+
+  GridBase *_grid;
+  GridBase *_cbgrid;
+
+  // Defines the stencils for even and odd
+  StencilImpl Stencil;
+  StencilImpl StencilEven;
+  StencilImpl StencilOdd;
+
+  // Copy of the gauge field , with even and odd subsets
+  DoubledGaugeField Umu;
+  DoubledGaugeField UmuEven;
+  DoubledGaugeField UmuOdd;
+
+  LebesgueOrder Lebesgue;
+  LebesgueOrder LebesgueEvenOdd;
+  
+  ///////////////////////////////////////////////////////////////
+  // Conserved current utilities
+  ///////////////////////////////////////////////////////////////
+  void ContractConservedCurrent(PropagatorField &q_in_1,
+                                PropagatorField &q_in_2,
+                                PropagatorField &q_out,
+                                PropagatorField &src,
+                                Current curr_type,
+                                unsigned int mu);
+  void SeqConservedCurrent(PropagatorField &q_in,
+                           PropagatorField &q_out,
+                           PropagatorField &srct,
+                           Current curr_type,
+                           unsigned int mu, 
+                           unsigned int tmin,
+                           unsigned int tmax,
+			   ComplexField &lattice_cmplx);
+};
+
+typedef NaiveStaggeredFermion<StaggeredImplF> NaiveStaggeredFermionF;
+typedef NaiveStaggeredFermion<StaggeredImplD> NaiveStaggeredFermionD;
+
+NAMESPACE_END(Grid);
+
+#endif
diff --git a/Grid/qcd/action/fermion/PartialFractionFermion5D.h b/Grid/qcd/action/fermion/PartialFractionFermion5D.h
index 928abd3f..54f8547f 100644
--- a/Grid/qcd/action/fermion/PartialFractionFermion5D.h
+++ b/Grid/qcd/action/fermion/PartialFractionFermion5D.h
@@ -47,8 +47,8 @@ public:
   void   M_internal(const FermionField &in, FermionField &out,int dag);
 
   // override multiply
-  virtual RealD  M    (const FermionField &in, FermionField &out);
-  virtual RealD  Mdag (const FermionField &in, FermionField &out);
+  virtual void   M    (const FermionField &in, FermionField &out);
+  virtual void   Mdag (const FermionField &in, FermionField &out);
 
   // half checkerboard operaions
   virtual void   Meooe       (const FermionField &in, FermionField &out);
diff --git a/Grid/qcd/action/fermion/StaggeredKernels.h b/Grid/qcd/action/fermion/StaggeredKernels.h
index 6ef0ab9d..30deee06 100644
--- a/Grid/qcd/action/fermion/StaggeredKernels.h
+++ b/Grid/qcd/action/fermion/StaggeredKernels.h
@@ -47,23 +47,34 @@ template<class Impl> class StaggeredKernels : public FermionOperator<Impl> , pub
   INHERIT_IMPL_TYPES(Impl);
   typedef FermionOperator<Impl> Base;
    
-public:
-    
-   void DhopDirKernel(StencilImpl &st, DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU, SiteSpinor * buf,
-		      int sF, int sU, const FermionFieldView &in, FermionFieldView &out, int dir,int disp);
+ public:
+
+  void DhopImproved(StencilImpl &st, LebesgueOrder &lo, 
+		    DoubledGaugeField &U, DoubledGaugeField &UUU, 
+		    const FermionField &in, FermionField &out, int dag, int interior,int exterior);
+  void DhopNaive(StencilImpl &st, LebesgueOrder &lo, 
+		 DoubledGaugeField &U,
+		 const FermionField &in, FermionField &out, int dag, int interior,int exterior);
+  
+  void DhopDirKernel(StencilImpl &st, DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU, SiteSpinor * buf,
+		     int sF, int sU, const FermionFieldView &in, FermionFieldView &out, int dir,int disp);
+ protected:    
 
    ///////////////////////////////////////////////////////////////////////////////////////
    // Generic Nc kernels
    ///////////////////////////////////////////////////////////////////////////////////////
-   void DhopSiteGeneric(StencilImpl &st, LebesgueOrder &lo, 
+   template<int Naik> accelerator_inline
+   void DhopSiteGeneric(StencilView &st, 
 			DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU, 
 			SiteSpinor * buf, int LLs, int sU, 
 			const FermionFieldView &in, FermionFieldView &out,int dag);
-   void DhopSiteGenericInt(StencilImpl &st, LebesgueOrder &lo, 
+   template<int Naik> accelerator_inline
+   void DhopSiteGenericInt(StencilView &st, 
 			   DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU, 
 			   SiteSpinor * buf, int LLs, int sU, 
 			   const FermionFieldView &in, FermionFieldView &out,int dag);
-   void DhopSiteGenericExt(StencilImpl &st, LebesgueOrder &lo, 
+   template<int Naik> accelerator_inline
+   void DhopSiteGenericExt(StencilView &st, 
 			   DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU,
 			   SiteSpinor * buf, int LLs, int sU, 
 			   const FermionFieldView &in, FermionFieldView &out,int dag);
@@ -71,15 +82,18 @@ public:
    ///////////////////////////////////////////////////////////////////////////////////////
    // Nc=3 specific kernels
    ///////////////////////////////////////////////////////////////////////////////////////
-   void DhopSiteHand(StencilImpl &st, LebesgueOrder &lo, 
+   template<int Naik> accelerator_inline
+   void DhopSiteHand(StencilView &st, 
 		     DoubledGaugeFieldView &U,DoubledGaugeFieldView &UUU, 
 		     SiteSpinor * buf, int LLs, int sU, 
 		     const FermionFieldView &in, FermionFieldView &out,int dag);
-   void DhopSiteHandInt(StencilImpl &st, LebesgueOrder &lo, 
+   template<int Naik> accelerator_inline
+   void DhopSiteHandInt(StencilView &st, 
 			DoubledGaugeFieldView &U,DoubledGaugeFieldView &UUU, 
 			SiteSpinor * buf, int LLs, int sU, 
 			const FermionFieldView &in, FermionFieldView &out,int dag);
-   void DhopSiteHandExt(StencilImpl &st, LebesgueOrder &lo, 
+   template<int Naik> accelerator_inline
+   void DhopSiteHandExt(StencilView &st, 
 			DoubledGaugeFieldView &U,DoubledGaugeFieldView &UUU, 
 			SiteSpinor * buf, int LLs, int sU, 
 			const FermionFieldView &in, FermionFieldView &out,int dag);
@@ -87,27 +101,10 @@ public:
    ///////////////////////////////////////////////////////////////////////////////////////
    // Asm Nc=3 specific kernels
    ///////////////////////////////////////////////////////////////////////////////////////
-   void DhopSiteAsm(StencilImpl &st, LebesgueOrder &lo, 
+   void DhopSiteAsm(StencilView &st, 
 		    DoubledGaugeFieldView &U,DoubledGaugeFieldView &UUU, 
 		    SiteSpinor * buf, int LLs, int sU, 
 		    const FermionFieldView &in, FermionFieldView &out,int dag);
-   ///////////////////////////////////////////////////////////////////////////////////////////////////
-   // Generic interface; fan out to right routine
-   ///////////////////////////////////////////////////////////////////////////////////////////////////
-   void DhopSite(StencilImpl &st, LebesgueOrder &lo, 
-		 DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU, 
-		 SiteSpinor * buf, int LLs, int sU,
-		 const FermionFieldView &in, FermionFieldView &out, int interior=1,int exterior=1);
-
-   void DhopSiteDag(StencilImpl &st, LebesgueOrder &lo, 
-		    DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU, 
-		    SiteSpinor * buf, int LLs, int sU,
-		    const FermionFieldView &in, FermionFieldView &out, int interior=1,int exterior=1);
-
-   void DhopSite(StencilImpl &st, LebesgueOrder &lo, 
-		 DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU, 
-		 SiteSpinor * buf, int LLs, int sU,
-		 const FermionFieldView &in, FermionFieldView &out, int dag, int interior,int exterior);
   
 public:
 
diff --git a/Grid/qcd/action/fermion/StaggeredVec5dImpl.h b/Grid/qcd/action/fermion/StaggeredVec5dImpl.h
index 2d4de18e..18fe993c 100644
--- a/Grid/qcd/action/fermion/StaggeredVec5dImpl.h
+++ b/Grid/qcd/action/fermion/StaggeredVec5dImpl.h
@@ -113,20 +113,7 @@ public:
       
   inline void InsertGaugeField(DoubledGaugeField &U_ds,const GaugeLinkField &U,int mu)
   {
-    GridBase *GaugeGrid = U_ds.Grid();
-    thread_for(lidx, GaugeGrid->lSites(),{
-
-	SiteScalarGaugeLink   ScalarU;
-	SiteDoubledGaugeField ScalarUds;
-	
-	Coordinate lcoor;
-	GaugeGrid->LocalIndexToLocalCoor(lidx, lcoor);
-	peekLocalSite(ScalarUds, U_ds, lcoor);
-	
-	peekLocalSite(ScalarU, U, lcoor);
-	ScalarUds(mu) = ScalarU();
-	
-    });
+    assert(0);
   }
   inline void DoubleStore(GridBase *GaugeGrid,
 			  DoubledGaugeField &UUUds, // for Naik term
diff --git a/Grid/qcd/action/fermion/WilsonCloverFermion.h b/Grid/qcd/action/fermion/WilsonCloverFermion.h
index 3847b0d9..91ad6d6d 100644
--- a/Grid/qcd/action/fermion/WilsonCloverFermion.h
+++ b/Grid/qcd/action/fermion/WilsonCloverFermion.h
@@ -109,9 +109,8 @@ public:
     ImportGauge(_Umu);
   }
 
-  virtual RealD M(const FermionField &in, FermionField &out);
-  virtual RealD Mdag(const FermionField &in, FermionField &out);
-
+  virtual void M(const FermionField &in, FermionField &out);
+  virtual void Mdag(const FermionField &in, FermionField &out);
   virtual void Mooee(const FermionField &in, FermionField &out);
   virtual void MooeeDag(const FermionField &in, FermionField &out);
   virtual void MooeeInv(const FermionField &in, FermionField &out);
@@ -258,15 +257,16 @@ private:
   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();
-    auto T_v = T.View();
-    auto F_v = F.View();
-    thread_for(i, CloverTerm.Grid()->oSites(),
+    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]()());
@@ -282,9 +282,9 @@ private:
     CloverFieldType T(F.Grid());
     T = Zero();
     
-    auto T_v = T.View();
-    auto F_v = F.View();
-    thread_for(i, CloverTerm.Grid()->oSites(),
+    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]()();
@@ -300,9 +300,9 @@ private:
     CloverFieldType T(F.Grid());
     T = Zero();
 
-    auto T_v = T.View();
-    auto F_v = F.View();
-    thread_for(i, CloverTerm.Grid()->oSites(),
+    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]()());
@@ -318,9 +318,9 @@ private:
     CloverFieldType T(F.Grid());
     T = Zero();
 
-    auto T_v = T.View();
-    auto F_v = F.View();
-    thread_for(i, CloverTerm.Grid()->oSites(),
+    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]()());
@@ -336,9 +336,9 @@ private:
     CloverFieldType T(F.Grid());
     T = Zero();
     
-    auto T_v = T.View();
-    auto F_v = F.View();
-    thread_for(i, CloverTerm.Grid()->oSites(),
+    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]()());
@@ -355,9 +355,9 @@ private:
 
     T = Zero();
 
-    auto T_v = T.View();
-    auto F_v = F.View();
-    thread_for(i, CloverTerm.Grid()->oSites(),
+    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]()());
diff --git a/Grid/qcd/action/fermion/WilsonFermion.h b/Grid/qcd/action/fermion/WilsonFermion.h
index eb9efa41..bf8926d0 100644
--- a/Grid/qcd/action/fermion/WilsonFermion.h
+++ b/Grid/qcd/action/fermion/WilsonFermion.h
@@ -92,8 +92,8 @@ public:
   // override multiply; cut number routines if pass dagger argument
   // and also make interface more uniformly consistent
   //////////////////////////////////////////////////////////////////
-  virtual RealD M(const FermionField &in, FermionField &out);
-  virtual RealD Mdag(const FermionField &in, FermionField &out);
+  virtual void  M(const FermionField &in, FermionField &out);
+  virtual void  Mdag(const FermionField &in, FermionField &out);
 
   /////////////////////////////////////////////////////////
   // half checkerboard operations
@@ -193,15 +193,17 @@ public:
   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);
+			   unsigned int tmax,
+			   ComplexField &lattice_cmplx);
 };
 
 typedef WilsonFermion<WilsonImplF> WilsonFermionF;
diff --git a/Grid/qcd/action/fermion/WilsonFermion5D.h b/Grid/qcd/action/fermion/WilsonFermion5D.h
index 58b54421..804b1d10 100644
--- a/Grid/qcd/action/fermion/WilsonFermion5D.h
+++ b/Grid/qcd/action/fermion/WilsonFermion5D.h
@@ -1,4 +1,3 @@
-
 /*************************************************************************************
 
     Grid physics library, www.github.com/paboyle/Grid 
@@ -99,8 +98,8 @@ public:
   GridBase *FermionRedBlackGrid(void)    { return _FiveDimRedBlackGrid;}
 
   // full checkerboard operations; leave unimplemented as abstract for now
-  virtual RealD  M    (const FermionField &in, FermionField &out){assert(0); return 0.0;};
-  virtual RealD  Mdag (const FermionField &in, FermionField &out){assert(0); return 0.0;};
+  virtual void   M    (const FermionField &in, FermionField &out){assert(0);};
+  virtual void   Mdag (const FermionField &in, FermionField &out){assert(0);};
 
   // half checkerboard operations; leave unimplemented as abstract for now
   virtual void   Meooe       (const FermionField &in, FermionField &out){assert(0);};
@@ -217,25 +216,7 @@ public:
     
   // Comms buffer
   std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> >  comm_buf;
-    
-  ///////////////////////////////////////////////////////////////
-  // Conserved current utilities
-  ///////////////////////////////////////////////////////////////
-  void ContractConservedCurrent(PropagatorField &q_in_1,
-				PropagatorField &q_in_2,
-				PropagatorField &q_out,
-				Current curr_type, 
-				unsigned int mu);
-  void SeqConservedCurrent(PropagatorField &q_in,
-			   PropagatorField &q_out,
-			   Current curr_type,
-			   unsigned int mu,
-			   unsigned int tmin,
-			   unsigned int tmax,
-			   ComplexField &lattice_cmplx);
 
-  void ContractJ5q(PropagatorField &q_in,ComplexField &J5q);
-  void ContractJ5q(FermionField &q_in,ComplexField &J5q);
 
 };
 
diff --git a/Grid/qcd/action/fermion/WilsonImpl.h b/Grid/qcd/action/fermion/WilsonImpl.h
index 47160730..52e1ee00 100644
--- a/Grid/qcd/action/fermion/WilsonImpl.h
+++ b/Grid/qcd/action/fermion/WilsonImpl.h
@@ -41,6 +41,7 @@ public:
   static const int Dimension = Representation::Dimension;
   static const bool isFundamental = Representation::isFundamental;
   static const bool LsVectorised=false;
+  static const bool isGparity=false;
   static const int Nhcs = Options::Nhcs;
 
   typedef PeriodicGaugeImpl<GaugeImplTypes<S, Dimension > > Gimpl;
@@ -98,8 +99,21 @@ public:
   {
     multLink(phi,U,chi,mu);
   }
-    
-      
+
+  template<class _SpinorField> 
+  inline void multLinkField(_SpinorField & out,
+			    const DoubledGaugeField &Umu,
+			    const _SpinorField & phi,
+			    int mu)
+  {
+    autoView( out_v, out, AcceleratorWrite);
+    autoView( phi_v, phi, AcceleratorRead);
+    autoView( Umu_v, Umu, AcceleratorRead);
+    accelerator_for(sss,out.Grid()->oSites(),1,{
+	multLink(out_v[sss],Umu_v[sss],phi_v[sss],mu);
+    });
+  }
+					   
   template <class ref>
   static accelerator_inline void loadLinkElement(Simd &reg, ref &memory) 
   {
@@ -177,18 +191,19 @@ public:
     int Ls=Btilde.Grid()->_fdimensions[0];
     GaugeLinkField tmp(mat.Grid());
     tmp = Zero();
-    auto tmp_v = tmp.View();
-    auto Btilde_v = Btilde.View();
-    auto Atilde_v = Atilde.View();
-    thread_for(sss,tmp.Grid()->oSites(),{
-      int sU=sss;
-      for(int s=0;s<Ls;s++){
-	int sF = s+Ls*sU;
-	tmp_v[sU] = tmp_v[sU]+ traceIndex<SpinIndex>(outerProduct(Btilde_v[sF],Atilde_v[sF])); // ordering here
-      }
-    });
+    {
+      autoView( tmp_v , tmp, AcceleratorWrite);
+      autoView( Btilde_v , Btilde, AcceleratorRead);
+      autoView( Atilde_v , Atilde, AcceleratorRead);
+      accelerator_for(sss,tmp.Grid()->oSites(),1,{
+	  int sU=sss;
+	  for(int s=0;s<Ls;s++){
+	    int sF = s+Ls*sU;
+	    tmp_v[sU] = tmp_v[sU]+ traceIndex<SpinIndex>(outerProduct(Btilde_v[sF],Atilde_v[sF])); // ordering here
+	  }
+	});
+    }
     PokeIndex<LorentzIndex>(mat,tmp,mu);
-      
   }
 };
 
diff --git a/Grid/qcd/action/fermion/WilsonKernels.h b/Grid/qcd/action/fermion/WilsonKernels.h
index 7348a463..1bac9211 100644
--- a/Grid/qcd/action/fermion/WilsonKernels.h
+++ b/Grid/qcd/action/fermion/WilsonKernels.h
@@ -66,41 +66,6 @@ public:
   static void DhopDirKernel(StencilImpl &st, DoubledGaugeField &U,SiteHalfSpinor * buf,
 			    int Ls, int Nsite, const FermionField &in, FermionField &out, int dirdisp, int gamma);
 
-  //////////////////////////////////////////////////////////////////////////////
-  // Utilities for inserting Wilson conserved current.
-  //////////////////////////////////////////////////////////////////////////////
-  static void ContractConservedCurrentSiteFwd(const SitePropagator &q_in_1,
-                                       const SitePropagator &q_in_2,
-                                       SitePropagator &q_out,
-                                       DoubledGaugeFieldView &U,
-                                       unsigned int sU,
-                                       unsigned int mu,
-                                       bool switch_sign = false);
-
-  static void ContractConservedCurrentSiteBwd(const SitePropagator &q_in_1,
-                                       const SitePropagator &q_in_2,
-                                       SitePropagator &q_out,
-                                       DoubledGaugeFieldView &U,
-                                       unsigned int sU,
-                                       unsigned int mu,
-                                       bool switch_sign = false);
-
-  static void SeqConservedCurrentSiteFwd(const SitePropagator &q_in, 
-                                  SitePropagator &q_out,
-                                  DoubledGaugeFieldView &U,
-                                  unsigned int sU,
-                                  unsigned int mu,
-                                  vPredicate t_mask,
-                                  bool switch_sign = false);
-
-  static void SeqConservedCurrentSiteBwd(const SitePropagator &q_in,
-                                  SitePropagator &q_out,
-                                  DoubledGaugeFieldView &U,
-                                  unsigned int sU,
-                                  unsigned int mu,
-                                  vPredicate t_mask,
-                                  bool switch_sign = false);
-
 private:
 
   static accelerator_inline void DhopDirK(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor * buf,
diff --git a/Grid/qcd/action/fermion/WilsonTMFermion5D.h b/Grid/qcd/action/fermion/WilsonTMFermion5D.h
index 71acf763..982e722a 100644
--- a/Grid/qcd/action/fermion/WilsonTMFermion5D.h
+++ b/Grid/qcd/action/fermion/WilsonTMFermion5D.h
@@ -120,7 +120,8 @@ class WilsonTMFermion5D : public WilsonFermion5D<Impl>
     }
   }
   
-  virtual RealD M(const FermionField &in, FermionField &out) {
+  virtual void M(const FermionField &in, FermionField &out) 
+  {
     out.Checkerboard() = in.Checkerboard();
     this->Dhop(in, out, DaggerNo);
     FermionField tmp(out.Grid());
@@ -129,11 +130,12 @@ class WilsonTMFermion5D : public WilsonFermion5D<Impl>
       ComplexD b(0.0,this->mu[s]);
       axpbg5y_ssp(tmp,a,in,b,in,s,s);
     }
-    return axpy_norm(out, 1.0, tmp, out);
+    axpy(out, 1.0, tmp, out);
   }
   
   // needed for fast PV
-  void update(const std::vector<RealD>& _mass, const std::vector<RealD>& _mu) {
+  void update(const std::vector<RealD>& _mass, const std::vector<RealD>& _mu) 
+  {
     assert(_mass.size() == _mu.size());
     assert(_mass.size() == this->FermionGrid()->_fdimensions[0]);
     this->mass = _mass;
diff --git a/Grid/qcd/action/fermion/implementation/CayleyFermion5DImplementation.h b/Grid/qcd/action/fermion/implementation/CayleyFermion5DImplementation.h
index c3c14ae9..e79b64dc 100644
--- a/Grid/qcd/action/fermion/implementation/CayleyFermion5DImplementation.h
+++ b/Grid/qcd/action/fermion/implementation/CayleyFermion5DImplementation.h
@@ -180,7 +180,7 @@ template<class Impl> void CayleyFermion5D<Impl>::CayleyReport(void)
     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_NVCC
+#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;
@@ -323,7 +323,7 @@ void CayleyFermion5D<Impl>::MeooeDag5D    (const FermionField &psi, FermionField
 }
 
 template<class Impl>
-RealD CayleyFermion5D<Impl>::M    (const FermionField &psi, FermionField &chi)
+void CayleyFermion5D<Impl>::M    (const FermionField &psi, FermionField &chi)
 {
   FermionField Din(psi.Grid());
   
@@ -335,11 +335,10 @@ RealD CayleyFermion5D<Impl>::M    (const FermionField &psi, FermionField &chi)
   axpby(chi,1.0,1.0,chi,psi); 
   
   M5D(psi,chi);
-  return(norm2(chi));
 }
 
 template<class Impl>
-RealD CayleyFermion5D<Impl>::Mdag (const FermionField &psi, FermionField &chi)
+void CayleyFermion5D<Impl>::Mdag (const FermionField &psi, FermionField &chi)
 {
   // Under adjoint
   //D1+        D1- P-    ->   D1+^dag   P+ D2-^dag
@@ -354,7 +353,6 @@ RealD CayleyFermion5D<Impl>::Mdag (const FermionField &psi, FermionField &chi)
   M5Ddag(psi,chi);
   // ((b D_W + D_w hop terms +1) on s-diag
   axpby (chi,1.0,1.0,chi,psi); 
-  return norm2(chi);
 }
 
 // half checkerboard operations
@@ -588,6 +586,356 @@ void CayleyFermion5D<Impl>::SetCoefficientsInternal(RealD zolo_hi,Vector<Coeff_t
   //  this->MooeeInternalCompute(1,inv,MatpInvDag,MatmInvDag);
 }
 
+
+template <class Impl>
+void CayleyFermion5D<Impl>::ContractJ5q(FermionField &q_in,ComplexField &J5q)
+{
+  conformable(this->GaugeGrid(), J5q.Grid());
+  conformable(q_in.Grid(), this->FermionGrid());
+  Gamma G5(Gamma::Algebra::Gamma5);
+  // 4d field
+  int Ls = this->Ls;
+  FermionField psi(this->GaugeGrid());
+  FermionField p_plus (this->GaugeGrid());
+  FermionField p_minus(this->GaugeGrid());
+  FermionField p(this->GaugeGrid());
+
+  ExtractSlice(p_plus , q_in, Ls/2-1 , 0);
+  ExtractSlice(p_minus, q_in, Ls/2   , 0);
+  p_plus = p_plus + G5*p_plus;
+  p_minus= p_minus - G5*p_minus;
+  p=0.5*(p_plus+p_minus);
+  J5q = localInnerProduct(p,p);
+}
+
+template <class Impl>
+void CayleyFermion5D<Impl>::ContractJ5q(PropagatorField &q_in,ComplexField &J5q)
+{
+  conformable(this->GaugeGrid(), J5q.Grid());
+  conformable(q_in.Grid(), this->FermionGrid());
+  Gamma G5(Gamma::Algebra::Gamma5);
+  // 4d field
+  int Ls = this->Ls;
+  PropagatorField psi(this->GaugeGrid());
+  PropagatorField p_plus (this->GaugeGrid());
+  PropagatorField p_minus(this->GaugeGrid());
+  PropagatorField p(this->GaugeGrid());
+
+  ExtractSlice(p_plus , q_in, Ls/2-1 , 0);
+  ExtractSlice(p_minus, q_in, Ls/2   , 0);
+  p_plus = p_plus + G5*p_plus;
+  p_minus= p_minus - G5*p_minus;
+  p=0.5*(p_plus+p_minus);
+  J5q = localInnerProduct(p,p);
+}
+
+#define Pp(Q) (0.5*(Q+g5*Q))
+#define Pm(Q) (0.5*(Q-g5*Q))
+#define Q_4d(Q) (Pm((Q)[0]) + Pp((Q)[Ls-1]))
+#define TopRowWithSource(Q) (phys_src + (1.0-mass)*Q_4d(Q))
+
+template <class Impl> 
+void CayleyFermion5D<Impl>::ContractConservedCurrent( PropagatorField &q_in_1,
+						      PropagatorField &q_in_2,
+						      PropagatorField &q_out,
+						      PropagatorField &phys_src,
+						      Current curr_type,
+						      unsigned int mu)
+{
+#if (!defined(GRID_CUDA)) && (!defined(GRID_HIP))
+  Gamma::Algebra Gmu [] = {
+    Gamma::Algebra::GammaX,
+    Gamma::Algebra::GammaY,
+    Gamma::Algebra::GammaZ,
+    Gamma::Algebra::GammaT,
+    Gamma::Algebra::Gamma5
+  };
+
+  auto UGrid= this->GaugeGrid();
+  auto FGrid= this->FermionGrid();
+  RealD sgn=1.0;
+  if ( curr_type == Current::Axial ) sgn = -1.0;
+
+  int Ls = this->Ls;
+
+  std::vector<PropagatorField> L_Q(Ls,UGrid); 
+  std::vector<PropagatorField> R_Q(Ls,UGrid); 
+  for(int s=0;s<Ls;s++){
+    ExtractSlice(L_Q[s], q_in_1, s , 0);
+    ExtractSlice(R_Q[s], q_in_2, s , 0);
+  }
+
+  Gamma g5(Gamma::Algebra::Gamma5);
+  PropagatorField C(UGrid); 
+  PropagatorField p5d(UGrid); 
+  PropagatorField us_p5d(UGrid); 
+  PropagatorField gp5d(UGrid); 
+  PropagatorField gus_p5d(UGrid); 
+
+  PropagatorField L_TmLsGq0(UGrid); 
+  PropagatorField L_TmLsTmp(UGrid);
+  PropagatorField R_TmLsGq0(UGrid); 
+  PropagatorField R_TmLsTmp(UGrid);
+  {
+    PropagatorField TermA(UGrid);
+    PropagatorField TermB(UGrid);
+    PropagatorField TermC(UGrid);
+    PropagatorField TermD(UGrid);
+    TermA = (Pp(Q_4d(L_Q)));
+    TermB = (Pm(Q_4d(L_Q)));
+    TermC = (Pm(TopRowWithSource(L_Q)));
+    TermD = (Pp(TopRowWithSource(L_Q)));
+
+    L_TmLsGq0 = (TermD - TermA + TermB);
+    L_TmLsTmp = (TermC - TermB + TermA);
+
+    TermA = (Pp(Q_4d(R_Q)));
+    TermB = (Pm(Q_4d(R_Q)));
+    TermC = (Pm(TopRowWithSource(R_Q)));
+    TermD = (Pp(TopRowWithSource(R_Q)));
+
+    R_TmLsGq0 = (TermD - TermA + TermB);
+    R_TmLsTmp = (TermC - TermB + TermA);
+  }
+
+  std::vector<PropagatorField> R_TmLsGq(Ls,UGrid);
+  std::vector<PropagatorField> L_TmLsGq(Ls,UGrid);
+  for(int s=0;s<Ls;s++){
+    R_TmLsGq[s] = (Pm((R_Q)[(s)]) + Pp((R_Q)[((s)-1+Ls)%Ls]));
+    L_TmLsGq[s] = (Pm((L_Q)[(s)]) + Pp((L_Q)[((s)-1+Ls)%Ls]));
+  }
+
+  Gamma gmu=Gamma(Gmu[mu]);
+
+  q_out = Zero();
+  PropagatorField tmp(UGrid); 
+  for(int s=0;s<Ls;s++){
+
+    int sp = (s+1)%Ls;
+    int sr = Ls-1-s;
+    int srp= (sr+1)%Ls;
+
+    // Mobius parameters
+    auto b=this->bs[s];
+    auto c=this->cs[s];
+    auto bpc = 1.0/(b+c);  // -0.5 factor in gauge links
+    if (s == 0) {
+      p5d    =(b*Pm(L_TmLsGq[Ls-1])+ c*Pp(L_TmLsGq[Ls-1]) + b*Pp(L_TmLsTmp)   + c*Pm(L_TmLsTmp     ));
+      tmp    =(b*Pm(R_TmLsGq0)     + c*Pp(R_TmLsGq0     ) + b*Pp(R_TmLsGq[1]) + c*Pm(R_TmLsGq[1]));
+    } else if (s == Ls-1) {
+      p5d    =(b*Pm(L_TmLsGq0)     + c*Pp(L_TmLsGq0     ) + b*Pp(L_TmLsGq[1]) + c*Pm(L_TmLsGq[1]));
+      tmp    =(b*Pm(R_TmLsGq[Ls-1])+ c*Pp(R_TmLsGq[Ls-1]) + b*Pp(R_TmLsTmp)   + c*Pm(R_TmLsTmp   ));
+    } else {
+      p5d    =(b*Pm(L_TmLsGq[sr]) + c*Pp(L_TmLsGq[sr])+ b*Pp(L_TmLsGq[srp])+ c*Pm(L_TmLsGq[srp]));
+      tmp    =(b*Pm(R_TmLsGq[s])  + c*Pp(R_TmLsGq[s]) + b*Pp(R_TmLsGq[sp ])+ c*Pm(R_TmLsGq[sp]));
+    }
+    tmp    = Cshift(tmp,mu,1);
+    Impl::multLinkField(us_p5d,this->Umu,tmp,mu);
+    
+    gp5d=g5*p5d*g5;
+    gus_p5d=gmu*us_p5d;
+
+    C = bpc*(adj(gp5d)*us_p5d);
+    C-= bpc*(adj(gp5d)*gus_p5d);
+
+    if (s == 0) {
+      p5d    =(b*Pm(R_TmLsGq0)     + c*Pp(R_TmLsGq0  )    + b*Pp(R_TmLsGq[1]) + c*Pm(R_TmLsGq[1]));
+      tmp    =(b*Pm(L_TmLsGq[Ls-1])+ c*Pp(L_TmLsGq[Ls-1]) + b*Pp(L_TmLsTmp)   + c*Pm(L_TmLsTmp  ));
+    } else if (s == Ls-1) {
+      p5d    =(b*Pm(R_TmLsGq[Ls-1])+ c*Pp(R_TmLsGq[Ls-1]) + b*Pp(R_TmLsTmp)   + c*Pm(R_TmLsTmp  ));
+      tmp    =(b*Pm(L_TmLsGq0)     + c*Pp(L_TmLsGq0  )    + b*Pp(L_TmLsGq[1]) + c*Pm(L_TmLsGq[1]));
+    } else {
+      p5d    =(b*Pm(R_TmLsGq[s])  + c*Pp(R_TmLsGq[s])  + b*Pp(R_TmLsGq[sp ])+ c*Pm(R_TmLsGq[sp]));
+      tmp    =(b*Pm(L_TmLsGq[sr]) + c*Pp(L_TmLsGq[sr]) + b*Pp(L_TmLsGq[srp])+ c*Pm(L_TmLsGq[srp]));
+    }
+    tmp    = Cshift(tmp,mu,1);
+    Impl::multLinkField(us_p5d,this->Umu,tmp,mu);
+
+    gp5d=gmu*p5d;
+    gus_p5d=g5*us_p5d*g5;
+
+    C-= bpc*(adj(gus_p5d)*gp5d);
+    C-= bpc*(adj(gus_p5d)*p5d);
+
+    if (s < Ls/2) q_out += sgn*C;
+    else          q_out +=     C;
+    
+  }
+#endif
+}
+
+template <class Impl>
+void CayleyFermion5D<Impl>::SeqConservedCurrent(PropagatorField &q_in, 
+                                                PropagatorField &q_out,
+                                                PropagatorField &phys_src,
+                                                Current curr_type, 
+                                                unsigned int mu,
+                                                unsigned int tmin, 
+                                                unsigned int tmax,
+						ComplexField &ph)// Complex phase factor
+{
+  assert(mu>=0);
+  assert(mu<Nd);
+
+
+#if 0
+  int tshift = (mu == Nd-1) ? 1 : 0;
+  ////////////////////////////////////////////////
+  // SHAMIR CASE 
+  ////////////////////////////////////////////////
+  int Ls = this->Ls;
+  auto UGrid= this->GaugeGrid();
+  auto FGrid= this->FermionGrid();
+  Gamma::Algebra Gmu [] = {
+    Gamma::Algebra::GammaX,
+    Gamma::Algebra::GammaY,
+    Gamma::Algebra::GammaZ,
+    Gamma::Algebra::GammaT
+  };
+  Gamma gmu=Gamma(Gmu[mu]);
+
+  PropagatorField L_Q(UGrid); 
+  PropagatorField R_Q(UGrid); 
+
+  PropagatorField tmp(UGrid);
+  PropagatorField Utmp(UGrid);
+  LatticeInteger zz (UGrid);   zz=0.0;
+  LatticeInteger lcoor(UGrid); LatticeCoordinate(lcoor,Nd-1);
+  for (int s=0;s<Ls;s++) {
+
+    RealD G_s = (curr_type == Current::Axial  ) ? ((s < Ls/2) ? -1 : 1) : 1;
+
+    ExtractSlice(R_Q, q_in, s , 0);
+
+    tmp    = Cshift(R_Q,mu,1);
+    Impl::multLinkField(Utmp,this->Umu,tmp,mu);
+    tmp    = G_s*( Utmp*ph - gmu*Utmp*ph ); // Forward hop
+    tmp    = where((lcoor>=tmin),tmp,zz); // Mask the time
+    tmp    = where((lcoor<=tmax),tmp,zz);
+    L_Q = tmp;
+
+    tmp    = R_Q*ph;
+    tmp    = Cshift(tmp,mu,-1);
+    Impl::multLinkField(Utmp,this->Umu,tmp,mu+Nd);// Adjoint link
+    tmp    = -G_s*( Utmp + gmu*Utmp ); 
+    tmp    = where((lcoor>=tmin+tshift),tmp,zz); // Mask the time 
+    tmp    = where((lcoor<=tmax+tshift),tmp,zz); // Position of current complicated
+    L_Q= L_Q+tmp;
+
+    InsertSlice(L_Q, q_out, s , 0);
+  }
+#endif
+
+#if (!defined(GRID_CUDA)) && (!defined(GRID_HIP))
+  int tshift = (mu == Nd-1) ? 1 : 0;
+  ////////////////////////////////////////////////
+  // GENERAL CAYLEY CASE
+  ////////////////////////////////////////////////
+  Gamma::Algebra Gmu [] = {
+    Gamma::Algebra::GammaX,
+    Gamma::Algebra::GammaY,
+    Gamma::Algebra::GammaZ,
+    Gamma::Algebra::GammaT,
+    Gamma::Algebra::Gamma5
+  };
+  Gamma gmu=Gamma(Gmu[mu]);
+  Gamma g5(Gamma::Algebra::Gamma5);
+
+  int Ls = this->Ls;
+  auto UGrid= this->GaugeGrid();
+  auto FGrid= this->FermionGrid();
+
+  std::vector<PropagatorField> R_Q(Ls,UGrid); 
+  PropagatorField L_Q(UGrid); 
+  PropagatorField tmp(UGrid);
+  PropagatorField Utmp(UGrid);
+
+  LatticeInteger zz (UGrid);   zz=0.0;
+  LatticeInteger lcoor(UGrid); LatticeCoordinate(lcoor,Nd-1);
+
+  for(int s=0;s<Ls;s++){
+    ExtractSlice(R_Q[s], q_in, s , 0);
+  }
+
+  PropagatorField R_TmLsGq0(UGrid); 
+  PropagatorField R_TmLsTmp(UGrid);
+  {
+    PropagatorField TermA(UGrid);
+    PropagatorField TermB(UGrid);
+    PropagatorField TermC(UGrid);
+    PropagatorField TermD(UGrid);
+
+    TermA = (Pp(Q_4d(R_Q)));
+    TermB = (Pm(Q_4d(R_Q)));
+    TermC = (Pm(TopRowWithSource(R_Q)));
+    TermD = (Pp(TopRowWithSource(R_Q)));
+
+    R_TmLsGq0 = (TermD - TermA + TermB);
+    R_TmLsTmp = (TermC - TermB + TermA);
+  }
+
+  std::vector<PropagatorField> R_TmLsGq(Ls,UGrid);
+  for(int s=0;s<Ls;s++){
+    R_TmLsGq[s] = (Pm((R_Q)[(s)]) + Pp((R_Q)[((s)-1+Ls)%Ls]));
+  }
+
+  std::vector<RealD> G_s(Ls,1.0);
+  if ( curr_type == Current::Axial ) {
+    for(int s=0;s<Ls/2;s++){
+      G_s[s] = -1.0;
+    }
+  }
+
+  for(int s=0;s<Ls;s++){
+
+    int sp = (s+1)%Ls;
+    int sr = Ls-1-s;
+    int srp= (sr+1)%Ls;
+
+    // Mobius parameters
+    auto b=this->bs[s];
+    auto c=this->cs[s];
+    //    auto bpc = G_s[s]*1.0/(b+c);  // -0.5 factor in gauge links
+
+    if (s == 0) {
+      tmp    =(b*Pm(R_TmLsGq0)     + c*Pp(R_TmLsGq0     ) + b*Pp(R_TmLsGq[1]) + c*Pm(R_TmLsGq[1]));
+    } else if (s == Ls-1) {
+      tmp    =(b*Pm(R_TmLsGq[Ls-1])+ c*Pp(R_TmLsGq[Ls-1]) + b*Pp(R_TmLsTmp)   + c*Pm(R_TmLsTmp   ));
+    } else {
+      tmp    =(b*Pm(R_TmLsGq[s])  + c*Pp(R_TmLsGq[s]) + b*Pp(R_TmLsGq[sp ])+ c*Pm(R_TmLsGq[sp]));
+    }
+
+    tmp    = Cshift(tmp,mu,1);
+    Impl::multLinkField(Utmp,this->Umu,tmp,mu);
+    tmp    = G_s[s]*( Utmp*ph - gmu*Utmp*ph ); // Forward hop
+    tmp    = where((lcoor>=tmin),tmp,zz); // Mask the time 
+    L_Q    = where((lcoor<=tmax),tmp,zz); // Position of current complicated
+
+    if (s == 0) {
+      tmp    =(b*Pm(R_TmLsGq0)     + c*Pp(R_TmLsGq0  )    + b*Pp(R_TmLsGq[1]) + c*Pm(R_TmLsGq[1]));
+    } else if (s == Ls-1) {
+      tmp    =(b*Pm(R_TmLsGq[Ls-1])+ c*Pp(R_TmLsGq[Ls-1]) + b*Pp(R_TmLsTmp)   + c*Pm(R_TmLsTmp  ));
+    } else {
+      tmp    =(b*Pm(R_TmLsGq[s])   + c*Pp(R_TmLsGq[s])    + b*Pp(R_TmLsGq[sp])+ c*Pm(R_TmLsGq[sp]));
+    }
+    tmp    = tmp *ph;
+    tmp    = Cshift(tmp,mu,-1);
+    Impl::multLinkField(Utmp,this->Umu,tmp,mu+Nd); // Adjoint link
+    tmp = -G_s[s]*( Utmp + gmu*Utmp );
+    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
+#undef Q_4d
+#undef TopRowWithSource
+
+
+
 #if 0
 template<class Impl>
 void CayleyFermion5D<Impl>::MooeeInternalCompute(int dag, int inv,
diff --git a/Grid/qcd/action/fermion/implementation/CayleyFermion5Dcache.h b/Grid/qcd/action/fermion/implementation/CayleyFermion5Dcache.h
index dbdf134b..d2537ccf 100644
--- a/Grid/qcd/action/fermion/implementation/CayleyFermion5Dcache.h
+++ b/Grid/qcd/action/fermion/implementation/CayleyFermion5Dcache.h
@@ -50,9 +50,9 @@ CayleyFermion5D<Impl>::M5D(const FermionField &psi_i,
   
   chi_i.Checkerboard()=psi_i.Checkerboard();
   GridBase *grid=psi_i.Grid();
-  auto psi = psi_i.View();
-  auto phi = phi_i.View();
-  auto chi = chi_i.View();
+  autoView(psi , psi_i,AcceleratorRead);
+  autoView(phi , phi_i,AcceleratorRead);
+  autoView(chi , chi_i,AcceleratorWrite);
   assert(phi.Checkerboard() == psi.Checkerboard());
 
   auto pdiag = &diag[0];
@@ -93,9 +93,9 @@ CayleyFermion5D<Impl>::M5Ddag(const FermionField &psi_i,
 {
   chi_i.Checkerboard()=psi_i.Checkerboard();
   GridBase *grid=psi_i.Grid();
-  auto psi = psi_i.View();
-  auto phi = phi_i.View();
-  auto chi = chi_i.View();
+  autoView(psi , psi_i,AcceleratorRead);
+  autoView(phi , phi_i,AcceleratorRead);
+  autoView(chi , chi_i,AcceleratorWrite);
   assert(phi.Checkerboard() == psi.Checkerboard());
 
   auto pdiag = &diag[0];
@@ -131,8 +131,8 @@ CayleyFermion5D<Impl>::MooeeInv    (const FermionField &psi_i, FermionField &chi
   chi_i.Checkerboard()=psi_i.Checkerboard();
   GridBase *grid=psi_i.Grid();
 
-  auto psi = psi_i.View();
-  auto chi = chi_i.View();
+  autoView(psi , psi_i,AcceleratorRead);
+  autoView(chi , chi_i,AcceleratorWrite);
 
   int Ls=this->Ls;
 
@@ -193,8 +193,8 @@ CayleyFermion5D<Impl>::MooeeInvDag (const FermionField &psi_i, FermionField &chi
   GridBase *grid=psi_i.Grid();
   int Ls=this->Ls;
 
-  auto psi = psi_i.View();
-  auto chi = chi_i.View();
+  autoView(psi , psi_i,AcceleratorRead);
+  autoView(chi , chi_i,AcceleratorWrite);
 
   auto plee  = & lee [0];
   auto pdee  = & dee [0];
diff --git a/Grid/qcd/action/fermion/implementation/CayleyFermion5Dvec.h b/Grid/qcd/action/fermion/implementation/CayleyFermion5Dvec.h
index 034ce642..b54f63ad 100644
--- a/Grid/qcd/action/fermion/implementation/CayleyFermion5Dvec.h
+++ b/Grid/qcd/action/fermion/implementation/CayleyFermion5Dvec.h
@@ -65,9 +65,9 @@ CayleyFermion5D<Impl>::M5D(const FermionField &psi_i,
   EnableIf<Impl::LsVectorised&&EnableBool,int> sfinae=0;
   chi_i.Checkerboard()=psi_i.Checkerboard();
   GridBase *grid=psi_i.Grid();
-  auto psi = psi_i.View();
-  auto phi = phi_i.View();
-  auto chi = chi_i.View();
+  autoView(psi, psi_i,CpuRead);
+  autoView(phi, phi_i,CpuRead);
+  autoView(chi, chi_i,CpuWrite);
   int Ls   = this->Ls;
   int LLs  = grid->_rdimensions[0];
   const int nsimd= Simd::Nsimd();
@@ -213,9 +213,9 @@ CayleyFermion5D<Impl>::M5Ddag(const FermionField &psi_i,
   EnableIf<Impl::LsVectorised&&EnableBool,int> sfinae=0;
   chi_i.Checkerboard()=psi_i.Checkerboard();
   GridBase *grid=psi_i.Grid();
-  auto psi=psi_i.View();
-  auto phi=phi_i.View();
-  auto chi=chi_i.View();
+  autoView(psi,psi_i,CpuRead);
+  autoView(phi,phi_i,CpuRead);
+  autoView(chi,chi_i,CpuWrite);
   int Ls   = this->Ls;
   int LLs  = grid->_rdimensions[0];
   int nsimd= Simd::Nsimd();
@@ -357,8 +357,8 @@ CayleyFermion5D<Impl>::MooeeInternalAsm(const FermionField &psi_i, FermionField
 					Vector<iSinglet<Simd> > &Matm)
 {
   EnableIf<Impl::LsVectorised&&EnableBool,int> sfinae=0;
-  auto psi = psi_i.View();
-  auto chi = chi_i.View();
+  autoView(psi , psi_i,CpuRead);
+  autoView(chi , chi_i,CpuWrite);
 #ifndef AVX512
   {
     SiteHalfSpinor BcastP;
@@ -535,8 +535,8 @@ CayleyFermion5D<Impl>::MooeeInternalZAsm(const FermionField &psi_i, FermionField
   EnableIf<Impl::LsVectorised,int> sfinae=0;
 #ifndef AVX512
   {
-    auto psi = psi_i.View();
-    auto chi = chi_i.View();
+    autoView(psi , psi_i,CpuRead);
+    autoView(chi , chi_i,CpuWrite);
 
     SiteHalfSpinor BcastP;
     SiteHalfSpinor BcastM;
@@ -586,8 +586,8 @@ CayleyFermion5D<Impl>::MooeeInternalZAsm(const FermionField &psi_i, FermionField
   }
 #else
   {
-    auto psi = psi_i.View();
-    auto chi = chi_i.View();
+    autoView(psi , psi_i,CpuRead);
+    autoView(chi , chi_i,CpuWrite);
     // pointers
     //  MASK_REGS;
 #define Chi_00 %zmm0
diff --git a/Grid/qcd/action/fermion/implementation/ContinuedFractionFermion5DImplementation.h b/Grid/qcd/action/fermion/implementation/ContinuedFractionFermion5DImplementation.h
index beeb3e00..6687800e 100644
--- a/Grid/qcd/action/fermion/implementation/ContinuedFractionFermion5DImplementation.h
+++ b/Grid/qcd/action/fermion/implementation/ContinuedFractionFermion5DImplementation.h
@@ -94,7 +94,7 @@ void ContinuedFractionFermion5D<Impl>::SetCoefficientsZolotarev(RealD zolo_hi,Ap
 
 
 template<class Impl>
-RealD  ContinuedFractionFermion5D<Impl>::M           (const FermionField &psi, FermionField &chi)
+void ContinuedFractionFermion5D<Impl>::M           (const FermionField &psi, FermionField &chi)
 {
   int Ls = this->Ls;
 
@@ -116,15 +116,14 @@ RealD  ContinuedFractionFermion5D<Impl>::M           (const FermionField &psi, F
     }
     sign=-sign; 
   }
-  return norm2(chi);
 }
 template<class Impl>
-RealD  ContinuedFractionFermion5D<Impl>::Mdag        (const FermionField &psi, FermionField &chi)
+void ContinuedFractionFermion5D<Impl>::Mdag        (const FermionField &psi, FermionField &chi)
 {
   // This matrix is already hermitian. (g5 Dw) = Dw dag g5 = (g5 Dw)dag
   // The rest of matrix is symmetric.
   // Can ignore "dag"
-  return M(psi,chi);
+  M(psi,chi);
 }
 template<class Impl>
 void  ContinuedFractionFermion5D<Impl>::Mdir (const FermionField &psi, FermionField &chi,int dir,int disp){
diff --git a/Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionCache.h b/Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionCache.h
index 46d3fa1f..9a8454ef 100644
--- a/Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionCache.h
+++ b/Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionCache.h
@@ -46,9 +46,9 @@ void DomainWallEOFAFermion<Impl>::M5D(const FermionField& psi_i, const FermionFi
   chi_i.Checkerboard() = psi_i.Checkerboard();
   int Ls = this->Ls;
   GridBase* grid = psi_i.Grid();
-  auto phi = phi_i.View();
-  auto psi = psi_i.View();
-  auto chi = chi_i.View();
+  autoView( phi , phi_i, AcceleratorRead);
+  autoView( psi , psi_i, AcceleratorRead);
+  autoView( chi , chi_i, AcceleratorWrite);
   assert(phi.Checkerboard() == psi.Checkerboard());
   auto pdiag = &diag[0];
   auto pupper = &upper[0];
@@ -82,9 +82,9 @@ void DomainWallEOFAFermion<Impl>::M5Ddag(const FermionField& psi_i, const Fermio
   GridBase* grid = psi_i.Grid();
   int Ls = this->Ls;
 
-  auto psi = psi_i.View();
-  auto phi = phi_i.View();
-  auto chi = chi_i.View();
+  autoView( psi , psi_i, AcceleratorRead);
+  autoView( phi , phi_i, AcceleratorRead);
+  autoView( chi , chi_i, AcceleratorWrite);
   assert(phi.Checkerboard() == psi.Checkerboard());
   auto pdiag = &diag[0];
   auto pupper = &upper[0];
@@ -116,8 +116,8 @@ void DomainWallEOFAFermion<Impl>::MooeeInv(const FermionField& psi_i, FermionFie
 {
   chi_i.Checkerboard() = psi_i.Checkerboard();
   GridBase* grid = psi_i.Grid();
-  auto psi=psi_i.View();
-  auto chi=chi_i.View();
+  autoView( psi, psi_i, AcceleratorRead);
+  autoView( chi, chi_i, AcceleratorWrite);
   int Ls = this->Ls;
 
   auto plee  = & this->lee[0];
@@ -172,8 +172,8 @@ void DomainWallEOFAFermion<Impl>::MooeeInvDag(const FermionField& psi_i, Fermion
 {
   chi_i.Checkerboard() = psi_i.Checkerboard();
   GridBase* grid = psi_i.Grid();
-  auto psi = psi_i.View();
-  auto chi = chi_i.View();
+  autoView( psi, psi_i, AcceleratorRead);
+  autoView( chi, chi_i, AcceleratorWrite);
   int Ls = this->Ls;
 
   auto plee  = & this->lee[0];
diff --git a/Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionImplementation.h b/Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionImplementation.h
index 3684fd6c..64ee4033 100644
--- a/Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionImplementation.h
+++ b/Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionImplementation.h
@@ -89,7 +89,7 @@ void DomainWallEOFAFermion<Impl>::DtildeInv(const FermionField& psi, FermionFiel
 /*****************************************************************************************************/
 
 template<class Impl>
-RealD DomainWallEOFAFermion<Impl>::M(const FermionField& psi, FermionField& chi)
+void DomainWallEOFAFermion<Impl>::M(const FermionField& psi, FermionField& chi)
 {
   FermionField Din(psi.Grid());
 
@@ -97,11 +97,10 @@ RealD DomainWallEOFAFermion<Impl>::M(const FermionField& psi, FermionField& chi)
   this->DW(Din, chi, DaggerNo);
   axpby(chi, 1.0, 1.0, chi, psi);
   this->M5D(psi, chi);
-  return(norm2(chi));
 }
 
 template<class Impl>
-RealD DomainWallEOFAFermion<Impl>::Mdag(const FermionField& psi, FermionField& chi)
+void DomainWallEOFAFermion<Impl>::Mdag(const FermionField& psi, FermionField& chi)
 {
   FermionField Din(psi.Grid());
 
@@ -109,7 +108,6 @@ RealD DomainWallEOFAFermion<Impl>::Mdag(const FermionField& psi, FermionField& c
   this->MeooeDag5D(Din, chi);
   this->M5Ddag(psi, chi);
   axpby(chi, 1.0, 1.0, chi, psi);
-  return(norm2(chi));
 }
 
 /********************************************************************
diff --git a/Grid/qcd/action/fermion/implementation/ImprovedStaggeredFermion5DImplementation.h b/Grid/qcd/action/fermion/implementation/ImprovedStaggeredFermion5DImplementation.h
index fdaa2f71..888691c4 100644
--- a/Grid/qcd/action/fermion/implementation/ImprovedStaggeredFermion5DImplementation.h
+++ b/Grid/qcd/action/fermion/implementation/ImprovedStaggeredFermion5DImplementation.h
@@ -221,10 +221,10 @@ void ImprovedStaggeredFermion5D<Impl>::DhopDir(const FermionField &in, FermionFi
 
   Compressor compressor;
   Stencil.HaloExchange(in,compressor);
-  auto Umu_v   = Umu.View();
-  auto UUUmu_v = UUUmu.View();
-  auto in_v    = in.View();
-  auto out_v   = out.View();
+  autoView( Umu_v   ,   Umu, CpuRead);
+  autoView( UUUmu_v , UUUmu, CpuRead);
+  autoView( in_v    ,  in, CpuRead);
+  autoView( out_v   , out, CpuWrite);
   thread_for( ss,Umu.Grid()->oSites(),{
     for(int s=0;s<Ls;s++){
       int sU=ss;
@@ -281,11 +281,9 @@ void ImprovedStaggeredFermion5D<Impl>::DhopInternal(StencilImpl & st, LebesgueOr
 						    DoubledGaugeField & U,DoubledGaugeField & UUU,
 						    const FermionField &in, FermionField &out,int dag)
 {
-#ifdef GRID_OMP
   if ( StaggeredKernelsStatic::Comms == StaggeredKernelsStatic::CommsAndCompute )
     DhopInternalOverlappedComms(st,lo,U,UUU,in,out,dag);
   else
-#endif
     DhopInternalSerialComms(st,lo,U,UUU,in,out,dag);
 }
 
@@ -294,9 +292,7 @@ void ImprovedStaggeredFermion5D<Impl>::DhopInternalOverlappedComms(StencilImpl &
 								   DoubledGaugeField & U,DoubledGaugeField & UUU,
 								   const FermionField &in, FermionField &out,int dag)
 {
-#ifdef GRID_OMP
   //  assert((dag==DaggerNo) ||(dag==DaggerYes));
-
   Compressor compressor; 
 
   int LLs = in.Grid()->_rdimensions[0];
@@ -305,99 +301,42 @@ void ImprovedStaggeredFermion5D<Impl>::DhopInternalOverlappedComms(StencilImpl &
   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();
 
-  double ctime=0;
-  double ptime=0;
-
   //////////////////////////////////////////////////////////////////////////////////////////////////////
-  // Ugly explicit thread mapping introduced for OPA reasons.
+  // Remove explicit thread mapping introduced for OPA reasons.
   //////////////////////////////////////////////////////////////////////////////////////////////////////
-#pragma omp parallel reduction(max:ctime) reduction(max:ptime)
+  DhopComputeTime-=usecond();
   {
-    int tid = omp_get_thread_num();
-    int nthreads = omp_get_num_threads();
-    int ncomms = CartesianCommunicator::nCommThreads;
-    if (ncomms == -1) ncomms = 1;
-    assert(nthreads > ncomms);
-    if (tid >= ncomms) {
-      double start = usecond();
-      nthreads -= ncomms;
-      int ttid  = tid - ncomms;
-      int n     = U.Grid()->oSites(); // 4d vol
-      int chunk = n / nthreads;
-      int rem   = n % nthreads;
-      int myblock, myn;
-      if (ttid < rem) {
-        myblock = ttid * chunk + ttid;
-        myn = chunk+1;
-      } else {
-        myblock = ttid*chunk + rem;
-        myn = chunk;
-      }
-
-      // do the compute
-      auto   U_v  =   U.View();
-      auto UUU_v  = UUU.View();
-      auto  in_v  =  in.View();
-      auto out_v  = out.View();
-
-      if (dag == DaggerYes) {
-        for (int ss = myblock; ss < myblock+myn; ++ss) {
-          int sU = ss;
-	  // Interior = 1; Exterior = 0; must implement for staggered
-          Kernels::DhopSiteDag(st,lo,U_v,UUU_v,st.CommBuf(),LLs,sU,in_v,out_v,1,0); //<---------
-        }
-      } else {
-        for (int ss = myblock; ss < myblock+myn; ++ss) {
-	  // Interior = 1; Exterior = 0;
-          int sU = ss;
-          Kernels::DhopSite(st,lo,U_v,UUU_v,st.CommBuf(),LLs,sU,in_v,out_v,1,0); //<------------
-        }
-      }
-        ptime = usecond() - start;
-    } else {
-      double start = usecond();
-      st.CommunicateThreaded();
-      ctime = usecond() - start;
-    }
+    int interior=1;
+    int exterior=0;
+    Kernels::DhopImproved(st,lo,U,UUU,in,out,dag,interior,exterior);
   }
-  DhopCommTime += ctime;
-  DhopComputeTime+=ptime;
-
-  // First to enter, last to leave timing
-  st.CollateThreads();
+  DhopComputeTime+=usecond();
 
   DhopFaceTime-=usecond();
   st.CommsMerge(compressor);
   DhopFaceTime+=usecond();
 
-  DhopComputeTime2-=usecond();
+  st.CommunicateComplete(requests);
+  DhopCommTime +=usecond();
 
-  auto   U_v  =   U.View();
-  auto UUU_v  = UUU.View();
-  auto  in_v  =  in.View();
-  auto out_v  = out.View();
-  if (dag == DaggerYes) {
-    int sz=st.surface_list.size();
-    thread_for( ss,sz,{
-      int sU = st.surface_list[ss];
-      Kernels::DhopSiteDag(st,lo,U_v,UUU_v,st.CommBuf(),LLs,sU,in_v,out_v,0,1); //<----------
-    });
-  } else {
-    int sz=st.surface_list.size();
-    thread_for( ss,sz,{
-      int sU = st.surface_list[ss];
-      Kernels::DhopSite(st,lo,U_v,UUU_v,st.CommBuf(),LLs,sU,in_v,out_v,0,1);//<----------
-    });
+  DhopComputeTime2-=usecond();
+  {
+    int interior=0;
+    int exterior=1;
+    Kernels::DhopImproved(st,lo,U,UUU,in,out,dag,interior,exterior);
   }
   DhopComputeTime2+=usecond();
-#else
-  assert(0);
-#endif
-
 }
 
 template<class Impl>
@@ -408,8 +347,6 @@ void ImprovedStaggeredFermion5D<Impl>::DhopInternalSerialComms(StencilImpl & st,
   Compressor compressor;
   int LLs = in.Grid()->_rdimensions[0];
 
-
-
  //double t1=usecond();
   DhopTotalTime -= usecond();
   DhopCommTime -= usecond();
@@ -418,28 +355,13 @@ void ImprovedStaggeredFermion5D<Impl>::DhopInternalSerialComms(StencilImpl & st,
   
   DhopComputeTime -= usecond();
   // Dhop takes the 4d grid from U, and makes a 5d index for fermion
-  auto   U_v  =   U.View();
-  auto UUU_v  = UUU.View();
-  auto  in_v  =  in.View();
-  auto out_v  = out.View();
-  if (dag == DaggerYes) {
-    thread_for( ss,U.Grid()->oSites(),{
-      int sU=ss;
-      Kernels::DhopSiteDag(st, lo, U_v, UUU_v, st.CommBuf(), LLs, sU,in_v, out_v);
-    });
-  } else {
-    thread_for( ss,U.Grid()->oSites(),{
-      int sU=ss;
-      Kernels::DhopSite(st,lo,U_v,UUU_v,st.CommBuf(),LLs,sU,in_v,out_v);
-    });
+  {
+    int interior=1;
+    int exterior=1;
+    Kernels::DhopImproved(st,lo,U,UUU,in,out,dag,interior,exterior);
   }
   DhopComputeTime += usecond();
   DhopTotalTime   += usecond();
- //double t2=usecond();
- //std::cout << __FILE__ << " " << __func__  << " Total Time " << DhopTotalTime << std::endl;
- //std::cout << __FILE__ << " " << __func__  << " Total Time Org " << t2-t1 << std::endl;
- //std::cout << __FILE__ << " " << __func__  << " Comml Time " << DhopCommTime << std::endl;
- //std::cout << __FILE__ << " " << __func__  << " Compute Time " << DhopComputeTime << std::endl;
 
 }
 /*CHANGE END*/
@@ -548,21 +470,24 @@ void ImprovedStaggeredFermion5D<Impl>::MdirAll(const FermionField &in, std::vect
   assert(0);
 }
 template <class Impl>
-RealD ImprovedStaggeredFermion5D<Impl>::M(const FermionField &in, FermionField &out) {
+void ImprovedStaggeredFermion5D<Impl>::M(const FermionField &in, FermionField &out) 
+{
   out.Checkerboard() = in.Checkerboard();
   Dhop(in, out, DaggerNo);
-  return axpy_norm(out, mass, in, out);
+  axpy(out, mass, in, out);
 }
 
 template <class Impl>
-RealD ImprovedStaggeredFermion5D<Impl>::Mdag(const FermionField &in, FermionField &out) {
+void ImprovedStaggeredFermion5D<Impl>::Mdag(const FermionField &in, FermionField &out) 
+{
   out.Checkerboard() = in.Checkerboard();
   Dhop(in, out, DaggerYes);
-  return axpy_norm(out, mass, in, out);
+  axpy(out, mass, in, out);
 }
 
 template <class Impl>
-void ImprovedStaggeredFermion5D<Impl>::Meooe(const FermionField &in, FermionField &out) {
+void ImprovedStaggeredFermion5D<Impl>::Meooe(const FermionField &in, FermionField &out) 
+{
   if (in.Checkerboard() == Odd) {
     DhopEO(in, out, DaggerNo);
   } else {
@@ -570,7 +495,8 @@ void ImprovedStaggeredFermion5D<Impl>::Meooe(const FermionField &in, FermionFiel
   }
 }
 template <class Impl>
-void ImprovedStaggeredFermion5D<Impl>::MeooeDag(const FermionField &in, FermionField &out) {
+void ImprovedStaggeredFermion5D<Impl>::MeooeDag(const FermionField &in, FermionField &out) 
+{
   if (in.Checkerboard() == Odd) {
     DhopEO(in, out, DaggerYes);
   } else {
@@ -579,27 +505,30 @@ void ImprovedStaggeredFermion5D<Impl>::MeooeDag(const FermionField &in, FermionF
 }
 
 template <class Impl>
-void ImprovedStaggeredFermion5D<Impl>::Mooee(const FermionField &in, FermionField &out) {
+void ImprovedStaggeredFermion5D<Impl>::Mooee(const FermionField &in, FermionField &out) 
+{
   out.Checkerboard() = in.Checkerboard();
   typename FermionField::scalar_type scal(mass);
   out = scal * in;
 }
 
 template <class Impl>
-void ImprovedStaggeredFermion5D<Impl>::MooeeDag(const FermionField &in, FermionField &out) {
+void ImprovedStaggeredFermion5D<Impl>::MooeeDag(const FermionField &in, FermionField &out) 
+{
   out.Checkerboard() = in.Checkerboard();
   Mooee(in, out);
 }
 
 template <class Impl>
-void ImprovedStaggeredFermion5D<Impl>::MooeeInv(const FermionField &in, FermionField &out) {
+void ImprovedStaggeredFermion5D<Impl>::MooeeInv(const FermionField &in, FermionField &out) 
+{
   out.Checkerboard() = in.Checkerboard();
   out = (1.0 / (mass)) * in;
 }
 
 template <class Impl>
-void ImprovedStaggeredFermion5D<Impl>::MooeeInvDag(const FermionField &in,
-						   FermionField &out) {
+void ImprovedStaggeredFermion5D<Impl>::MooeeInvDag(const FermionField &in,FermionField &out) 
+{
   out.Checkerboard() = in.Checkerboard();
   MooeeInv(in, out);
 }
@@ -611,6 +540,7 @@ template <class Impl>
 void ImprovedStaggeredFermion5D<Impl>::ContractConservedCurrent(PropagatorField &q_in_1,
 								PropagatorField &q_in_2,
 								PropagatorField &q_out,
+								PropagatorField &src,
 								Current curr_type,
 								unsigned int mu)
 {
@@ -620,11 +550,12 @@ void ImprovedStaggeredFermion5D<Impl>::ContractConservedCurrent(PropagatorField
 template <class Impl>
 void ImprovedStaggeredFermion5D<Impl>::SeqConservedCurrent(PropagatorField &q_in,
 							   PropagatorField &q_out,
+							   PropagatorField &src,
 							   Current curr_type,
 							   unsigned int mu, 
 							   unsigned int tmin,
-                                              unsigned int tmax,
-					      ComplexField &lattice_cmplx)
+							   unsigned int tmax,
+							   ComplexField &lattice_cmplx)
 {
   assert(0);
 
diff --git a/Grid/qcd/action/fermion/implementation/ImprovedStaggeredFermionImplementation.h b/Grid/qcd/action/fermion/implementation/ImprovedStaggeredFermionImplementation.h
index 0b723c47..05d9a17e 100644
--- a/Grid/qcd/action/fermion/implementation/ImprovedStaggeredFermionImplementation.h
+++ b/Grid/qcd/action/fermion/implementation/ImprovedStaggeredFermionImplementation.h
@@ -171,21 +171,24 @@ void ImprovedStaggeredFermion<Impl>::ImportGauge(const GaugeField &_Uthin,const
 /////////////////////////////
 
 template <class Impl>
-RealD ImprovedStaggeredFermion<Impl>::M(const FermionField &in, FermionField &out) {
+void ImprovedStaggeredFermion<Impl>::M(const FermionField &in, FermionField &out) 
+{
   out.Checkerboard() = in.Checkerboard();
   Dhop(in, out, DaggerNo);
-  return axpy_norm(out, mass, in, out);
+  axpy(out, mass, in, out);
 }
 
 template <class Impl>
-RealD ImprovedStaggeredFermion<Impl>::Mdag(const FermionField &in, FermionField &out) {
+void ImprovedStaggeredFermion<Impl>::Mdag(const FermionField &in, FermionField &out) 
+{
   out.Checkerboard() = in.Checkerboard();
   Dhop(in, out, DaggerYes);
-  return axpy_norm(out, mass, in, out);
+  axpy(out, mass, in, out);
 }
 
 template <class Impl>
-void ImprovedStaggeredFermion<Impl>::Meooe(const FermionField &in, FermionField &out) {
+void ImprovedStaggeredFermion<Impl>::Meooe(const FermionField &in, FermionField &out) 
+{
   if (in.Checkerboard() == Odd) {
     DhopEO(in, out, DaggerNo);
   } else {
@@ -193,7 +196,8 @@ void ImprovedStaggeredFermion<Impl>::Meooe(const FermionField &in, FermionField
   }
 }
 template <class Impl>
-void ImprovedStaggeredFermion<Impl>::MeooeDag(const FermionField &in, FermionField &out) {
+void ImprovedStaggeredFermion<Impl>::MeooeDag(const FermionField &in, FermionField &out) 
+{
   if (in.Checkerboard() == Odd) {
     DhopEO(in, out, DaggerYes);
   } else {
@@ -202,27 +206,30 @@ void ImprovedStaggeredFermion<Impl>::MeooeDag(const FermionField &in, FermionFie
 }
 
 template <class Impl>
-void ImprovedStaggeredFermion<Impl>::Mooee(const FermionField &in, FermionField &out) {
+void ImprovedStaggeredFermion<Impl>::Mooee(const FermionField &in, FermionField &out) 
+{
   out.Checkerboard() = in.Checkerboard();
   typename FermionField::scalar_type scal(mass);
   out = scal * in;
 }
 
 template <class Impl>
-void ImprovedStaggeredFermion<Impl>::MooeeDag(const FermionField &in, FermionField &out) {
+void ImprovedStaggeredFermion<Impl>::MooeeDag(const FermionField &in, FermionField &out) 
+{
   out.Checkerboard() = in.Checkerboard();
   Mooee(in, out);
 }
 
 template <class Impl>
-void ImprovedStaggeredFermion<Impl>::MooeeInv(const FermionField &in, FermionField &out) {
+void ImprovedStaggeredFermion<Impl>::MooeeInv(const FermionField &in, FermionField &out) 
+{
   out.Checkerboard() = in.Checkerboard();
   out = (1.0 / (mass)) * in;
 }
 
 template <class Impl>
-void ImprovedStaggeredFermion<Impl>::MooeeInvDag(const FermionField &in,
-						 FermionField &out) {
+void ImprovedStaggeredFermion<Impl>::MooeeInvDag(const FermionField &in,FermionField &out) 
+{
   out.Checkerboard() = in.Checkerboard();
   MooeeInv(in, out);
 }
@@ -234,7 +241,8 @@ void ImprovedStaggeredFermion<Impl>::MooeeInvDag(const FermionField &in,
 template <class Impl>
 void ImprovedStaggeredFermion<Impl>::DerivInternal(StencilImpl &st, DoubledGaugeField &U, DoubledGaugeField &UUU, 
 						   GaugeField & mat,
-						   const FermionField &A, const FermionField &B, int dag) {
+						   const FermionField &A, const FermionField &B, int dag) 
+{
   assert((dag == DaggerNo) || (dag == DaggerYes));
 
   Compressor compressor;
@@ -250,10 +258,10 @@ void ImprovedStaggeredFermion<Impl>::DerivInternal(StencilImpl &st, DoubledGauge
     ////////////////////////
     // Call the single hop
     ////////////////////////
-    auto U_v   = U.View();
-    auto UUU_v = UUU.View();
-    auto B_v   = B.View();
-    auto Btilde_v   = Btilde.View();
+    autoView( U_v   , U, CpuRead);
+    autoView( UUU_v , UUU, CpuRead);
+    autoView( B_v      , B, CpuWrite);
+    autoView( Btilde_v , Btilde, CpuWrite);
     thread_for(sss,B.Grid()->oSites(),{
       Kernels::DhopDirKernel(st, U_v, UUU_v, st.CommBuf(), sss, sss, B_v, Btilde_v, mu,1);
     });
@@ -284,8 +292,8 @@ void ImprovedStaggeredFermion<Impl>::DerivInternal(StencilImpl &st, DoubledGauge
 }
 
 template <class Impl>
-void ImprovedStaggeredFermion<Impl>::DhopDeriv(GaugeField &mat, const FermionField &U, const FermionField &V, int dag) {
-
+void ImprovedStaggeredFermion<Impl>::DhopDeriv(GaugeField &mat, const FermionField &U, const FermionField &V, int dag) 
+{
   conformable(U.Grid(), _grid);
   conformable(U.Grid(), V.Grid());
   conformable(U.Grid(), mat.Grid());
@@ -296,8 +304,8 @@ void ImprovedStaggeredFermion<Impl>::DhopDeriv(GaugeField &mat, const FermionFie
 }
 
 template <class Impl>
-void ImprovedStaggeredFermion<Impl>::DhopDerivOE(GaugeField &mat, const FermionField &U, const FermionField &V, int dag) {
-
+void ImprovedStaggeredFermion<Impl>::DhopDerivOE(GaugeField &mat, const FermionField &U, const FermionField &V, int dag) 
+{
   conformable(U.Grid(), _cbgrid);
   conformable(U.Grid(), V.Grid());
   conformable(U.Grid(), mat.Grid());
@@ -310,8 +318,8 @@ void ImprovedStaggeredFermion<Impl>::DhopDerivOE(GaugeField &mat, const FermionF
 }
 
 template <class Impl>
-void ImprovedStaggeredFermion<Impl>::DhopDerivEO(GaugeField &mat, const FermionField &U, const FermionField &V, int dag) {
-
+void ImprovedStaggeredFermion<Impl>::DhopDerivEO(GaugeField &mat, const FermionField &U, const FermionField &V, int dag) 
+{
   conformable(U.Grid(), _cbgrid);
   conformable(U.Grid(), V.Grid());
   conformable(U.Grid(), mat.Grid());
@@ -378,10 +386,10 @@ void ImprovedStaggeredFermion<Impl>::DhopDir(const FermionField &in, FermionFiel
 
   Compressor compressor;
   Stencil.HaloExchange(in, compressor);
-  auto Umu_v   =   Umu.View();
-  auto UUUmu_v = UUUmu.View();
-  auto in_v    =  in.View();
-  auto out_v   = out.View();
+  autoView( Umu_v   ,   Umu, CpuRead);
+  autoView( UUUmu_v , UUUmu, CpuRead);
+  autoView( in_v    ,  in, CpuRead);
+  autoView( out_v   , out, CpuWrite);
   thread_for( sss, in.Grid()->oSites(),{
     Kernels::DhopDirKernel(Stencil, Umu_v, UUUmu_v, Stencil.CommBuf(), sss, sss, in_v, out_v, dir, disp);
   });
@@ -395,11 +403,9 @@ void ImprovedStaggeredFermion<Impl>::DhopInternal(StencilImpl &st, LebesgueOrder
 						  const FermionField &in,
 						  FermionField &out, int dag) 
 {
-#ifdef GRID_OMP
   if ( StaggeredKernelsStatic::Comms == StaggeredKernelsStatic::CommsAndCompute )
     DhopInternalOverlappedComms(st,lo,U,UUU,in,out,dag);
   else
-#endif
     DhopInternalSerialComms(st,lo,U,UUU,in,out,dag);
 }
 template <class Impl>
@@ -409,7 +415,6 @@ void ImprovedStaggeredFermion<Impl>::DhopInternalOverlappedComms(StencilImpl &st
 								 const FermionField &in,
 								 FermionField &out, int dag) 
 {
-#ifdef GRID_OMP
   Compressor compressor; 
   int len =  U.Grid()->oSites();
 
@@ -418,60 +423,30 @@ void ImprovedStaggeredFermion<Impl>::DhopInternalOverlappedComms(StencilImpl &st
   DhopFaceTime    -= usecond();
   st.Prepare();
   st.HaloGather(in,compressor);
-  st.CommsMergeSHM(compressor);
   DhopFaceTime    += usecond();
 
+  DhopCommTime -=usecond();
+  std::vector<std::vector<CommsRequest_t> > requests;
+  st.CommunicateBegin(requests);
+
+  DhopFaceTime-=usecond();
+  st.CommsMergeSHM(compressor);
+  DhopFaceTime+= usecond();
+
   //////////////////////////////////////////////////////////////////////////////////////////////////////
-  // Ugly explicit thread mapping introduced for OPA reasons.
+  // Removed explicit thread comms
   //////////////////////////////////////////////////////////////////////////////////////////////////////
   DhopComputeTime    -= usecond();
-#pragma omp parallel 
   {
-    int tid = omp_get_thread_num();
-    int nthreads = omp_get_num_threads();
-    int ncomms = CartesianCommunicator::nCommThreads;
-    if (ncomms == -1) ncomms = 1;
-    assert(nthreads > ncomms);
-
-    if (tid >= ncomms) {
-      nthreads -= ncomms;
-      int ttid  = tid - ncomms;
-      int n     = len;
-      int chunk = n / nthreads;
-      int rem   = n % nthreads;
-      int myblock, myn;
-      if (ttid < rem) {
-        myblock = ttid * chunk + ttid;
-        myn = chunk+1;
-      } else {
-        myblock = ttid*chunk + rem;
-        myn = chunk;
-      }
-
-      // do the compute
-      auto U_v   = U.View();
-      auto UUU_v = UUU.View();
-      auto in_v  = in.View();
-      auto out_v = out.View();
-      if (dag == DaggerYes) {
-        for (int ss = myblock; ss < myblock+myn; ++ss) {
-          int sU = ss;
-	  // Interior = 1; Exterior = 0; must implement for staggered
-          Kernels::DhopSiteDag(st,lo,U_v,UUU_v,st.CommBuf(),1,sU,in_v,out_v,1,0); 
-        }
-      } else {
-        for (int ss = myblock; ss < myblock+myn; ++ss) {
-	  // Interior = 1; Exterior = 0;
-          int sU = ss;
-          Kernels::DhopSite(st,lo,U_v,UUU_v,st.CommBuf(),1,sU,in_v,out_v,1,0);
-        }
-      }
-    } else {
-      st.CommunicateThreaded();
-    }
+    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);
@@ -479,28 +454,11 @@ void ImprovedStaggeredFermion<Impl>::DhopInternalOverlappedComms(StencilImpl &st
 
   DhopComputeTime2    -= usecond();
   {
-    auto U_v   = U.View();
-    auto UUU_v = UUU.View();
-    auto in_v  = in.View();
-    auto out_v = out.View();
-    if (dag == DaggerYes) {
-      int sz=st.surface_list.size();
-      thread_for(ss,sz,{
-	int sU = st.surface_list[ss];
-	Kernels::DhopSiteDag(st,lo,U_v,UUU_v,st.CommBuf(),1,sU,in_v,out_v,0,1);
-      });
-    } else {
-      int sz=st.surface_list.size();
-      thread_for(ss,sz,{
-	int sU = st.surface_list[ss];
-	Kernels::DhopSite(st,lo,U_v,UUU_v,st.CommBuf(),1,sU,in_v,out_v,0,1);
-      });
-    }
+    int interior=0;
+    int exterior=1;
+    Kernels::DhopImproved(st,lo,U,UUU,in,out,dag,interior,exterior);
   }
   DhopComputeTime2    += usecond();
-#else
-  assert(0);
-#endif
 }
 
 
@@ -520,19 +478,11 @@ void ImprovedStaggeredFermion<Impl>::DhopInternalSerialComms(StencilImpl &st, Le
   st.HaloExchange(in, compressor);
   DhopCommTime    += usecond();
 
-  auto U_v   =   U.View();
-  auto UUU_v = UUU.View();
-  auto in_v  =  in.View();
-  auto out_v = out.View();
   DhopComputeTime -= usecond();
-  if (dag == DaggerYes) {
-    thread_for(sss, in.Grid()->oSites(),{
-      Kernels::DhopSiteDag(st, lo, U_v, UUU_v, st.CommBuf(), 1, sss, in_v, out_v);
-    });
-  } else {
-    thread_for(sss, in.Grid()->oSites(),{
-      Kernels::DhopSite(st, lo, U_v, UUU_v, st.CommBuf(), 1, sss, in_v, out_v);
-    });
+  {
+    int interior=1;
+    int exterior=1;
+    Kernels::DhopImproved(st,lo,U,UUU,in,out,dag,interior,exterior);
   }
   DhopComputeTime += usecond();
   DhopTotalTime   += usecond();
@@ -600,6 +550,7 @@ template <class Impl>
 void ImprovedStaggeredFermion<Impl>::ContractConservedCurrent(PropagatorField &q_in_1,
 							      PropagatorField &q_in_2,
 							      PropagatorField &q_out,
+							      PropagatorField &src,
 							      Current curr_type,
 							      unsigned int mu)
 {
@@ -609,6 +560,7 @@ void ImprovedStaggeredFermion<Impl>::ContractConservedCurrent(PropagatorField &q
 template <class Impl>
 void ImprovedStaggeredFermion<Impl>::SeqConservedCurrent(PropagatorField &q_in,
                                                          PropagatorField &q_out,
+                                                         PropagatorField &src,
                                                          Current curr_type,
                                                          unsigned int mu, 
                                                          unsigned int tmin,
diff --git a/Grid/qcd/action/fermion/implementation/MobiusEOFAFermionCache.h b/Grid/qcd/action/fermion/implementation/MobiusEOFAFermionCache.h
index f74c7a51..41b9170d 100644
--- a/Grid/qcd/action/fermion/implementation/MobiusEOFAFermionCache.h
+++ b/Grid/qcd/action/fermion/implementation/MobiusEOFAFermionCache.h
@@ -44,9 +44,9 @@ void MobiusEOFAFermion<Impl>::M5D(const FermionField &psi_i, const FermionField
   chi_i.Checkerboard() = psi_i.Checkerboard();
   GridBase *grid = psi_i.Grid();
   int Ls = this->Ls;
-  auto psi = psi_i.View();
-  auto phi = phi_i.View();
-  auto chi = chi_i.View();
+  autoView(psi , psi_i, AcceleratorRead);
+  autoView(phi , phi_i, AcceleratorRead);
+  autoView(chi , chi_i, AcceleratorWrite);
 
   assert(phi.Checkerboard() == psi.Checkerboard());
 
@@ -84,9 +84,9 @@ void MobiusEOFAFermion<Impl>::M5D_shift(const FermionField &psi_i, const Fermion
   chi_i.Checkerboard() = psi_i.Checkerboard();
   GridBase *grid = psi_i.Grid();
   int Ls = this->Ls;
-  auto psi = psi_i.View();
-  auto phi = phi_i.View();
-  auto chi = chi_i.View();
+  autoView(psi , psi_i, AcceleratorRead);
+  autoView(phi , phi_i, AcceleratorRead);
+  autoView(chi , chi_i, AcceleratorWrite);
 
   auto pm  = this->pm;
   int shift_s = (pm == 1) ? (Ls-1) : 0; // s-component modified by shift operator
@@ -132,9 +132,9 @@ void MobiusEOFAFermion<Impl>::M5Ddag(const FermionField &psi_i, const FermionFie
   chi_i.Checkerboard() = psi_i.Checkerboard();
   GridBase *grid = psi_i.Grid();
   int Ls = this->Ls;
-  auto psi = psi_i.View();
-  auto phi = phi_i.View();
-  auto chi = chi_i.View();
+  autoView(psi , psi_i, AcceleratorRead);
+  autoView(phi , phi_i, AcceleratorRead);
+  autoView(chi , chi_i, AcceleratorWrite);
 
   assert(phi.Checkerboard() == psi.Checkerboard());
 
@@ -174,9 +174,9 @@ void MobiusEOFAFermion<Impl>::M5Ddag_shift(const FermionField &psi_i, const Ferm
   GridBase *grid = psi_i.Grid();
   int Ls = this->Ls;
   int shift_s = (this->pm == 1) ? (Ls-1) : 0; // s-component modified by shift operator
-  auto psi = psi_i.View();
-  auto phi = phi_i.View();
-  auto chi = chi_i.View();
+  autoView(psi , psi_i, AcceleratorRead);
+  autoView(phi , phi_i, AcceleratorRead);
+  autoView(chi , chi_i, AcceleratorWrite);
 
   assert(phi.Checkerboard() == psi.Checkerboard());
 
@@ -226,8 +226,8 @@ void MobiusEOFAFermion<Impl>::MooeeInv(const FermionField &psi_i, FermionField &
   chi_i.Checkerboard() = psi_i.Checkerboard();
   GridBase *grid = psi_i.Grid();
   int Ls = this->Ls;
-  auto psi = psi_i.View();
-  auto chi = chi_i.View();
+  autoView(psi , psi_i, AcceleratorRead);
+  autoView(chi , chi_i, AcceleratorWrite);
 
   auto plee = & this->lee [0];
   auto pdee = & this->dee [0];
@@ -286,8 +286,8 @@ void MobiusEOFAFermion<Impl>::MooeeInv_shift(const FermionField &psi_i, FermionF
   chi_i.Checkerboard() = psi_i.Checkerboard();
   GridBase *grid = psi_i.Grid();
   int Ls = this->Ls;
-  auto psi = psi_i.View();
-  auto chi = chi_i.View();
+  autoView(psi , psi_i, AcceleratorRead);
+  autoView(chi , chi_i, AcceleratorWrite);
 
   auto pm = this->pm;
   auto plee = & this->lee [0];
@@ -354,8 +354,8 @@ void MobiusEOFAFermion<Impl>::MooeeInvDag(const FermionField &psi_i, FermionFiel
   chi_i.Checkerboard() = psi_i.Checkerboard();
   GridBase *grid = psi_i.Grid();
   int Ls = this->Ls;
-  auto psi = psi_i.View();
-  auto chi = chi_i.View();
+  autoView(psi , psi_i, AcceleratorRead);
+  autoView(chi , chi_i, AcceleratorWrite);
 
   auto plee = & this->lee [0];
   auto pdee = & this->dee [0];
@@ -410,8 +410,8 @@ void MobiusEOFAFermion<Impl>::MooeeInvDag_shift(const FermionField &psi_i, Fermi
 {
   chi_i.Checkerboard() = psi_i.Checkerboard();
   GridBase *grid = psi_i.Grid();
-  auto psi = psi_i.View();
-  auto chi = chi_i.View();
+  autoView(psi , psi_i, AcceleratorRead);
+  autoView(chi , chi_i, AcceleratorWrite);
   int Ls = this->Ls;
 
   auto pm = this->pm;
diff --git a/Grid/qcd/action/fermion/implementation/MobiusEOFAFermionImplementation.h b/Grid/qcd/action/fermion/implementation/MobiusEOFAFermionImplementation.h
index 256423e6..9b9db178 100644
--- a/Grid/qcd/action/fermion/implementation/MobiusEOFAFermionImplementation.h
+++ b/Grid/qcd/action/fermion/implementation/MobiusEOFAFermionImplementation.h
@@ -166,7 +166,7 @@ void MobiusEOFAFermion<Impl>::DtildeInv(const FermionField& psi, FermionField& c
 /*****************************************************************************************************/
 
 template<class Impl>
-RealD MobiusEOFAFermion<Impl>::M(const FermionField& psi, FermionField& chi)
+void MobiusEOFAFermion<Impl>::M(const FermionField& psi, FermionField& chi)
 {
   FermionField Din(psi.Grid());
 
@@ -174,11 +174,10 @@ RealD MobiusEOFAFermion<Impl>::M(const FermionField& psi, FermionField& chi)
   this->DW(Din, chi, DaggerNo);
   axpby(chi, 1.0, 1.0, chi, psi);
   this->M5D(psi, chi);
-  return(norm2(chi));
 }
 
 template<class Impl>
-RealD MobiusEOFAFermion<Impl>::Mdag(const FermionField& psi, FermionField& chi)
+void MobiusEOFAFermion<Impl>::Mdag(const FermionField& psi, FermionField& chi)
 {
   FermionField Din(psi.Grid());
 
@@ -186,7 +185,6 @@ RealD MobiusEOFAFermion<Impl>::Mdag(const FermionField& psi, FermionField& chi)
   this->MeooeDag5D(Din, chi);
   this->M5Ddag(psi, chi);
   axpby(chi, 1.0, 1.0, chi, psi);
-  return(norm2(chi));
 }
 
 /********************************************************************
diff --git a/Grid/qcd/action/fermion/implementation/NaiveStaggeredFermionImplementation.h b/Grid/qcd/action/fermion/implementation/NaiveStaggeredFermionImplementation.h
new file mode 100644
index 00000000..788e02cf
--- /dev/null
+++ b/Grid/qcd/action/fermion/implementation/NaiveStaggeredFermionImplementation.h
@@ -0,0 +1,499 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid
+
+Source file: ./lib/qcd/action/fermion/ImprovedStaggeredFermion.cc
+
+Copyright (C) 2015
+
+Author: Azusa Yamaguchi, Peter Boyle
+
+This program is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 2 of the License, or
+(at your option) any later version.
+
+This program is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along
+with this program; if not, write to the Free Software Foundation, Inc.,
+51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+See the full license in the file "LICENSE" in the top level distribution
+directory
+*************************************************************************************/
+/*  END LEGAL */
+#include <Grid/Grid.h>
+
+#pragma once 
+
+NAMESPACE_BEGIN(Grid);
+
+/////////////////////////////////
+// Constructor and gauge import
+/////////////////////////////////
+
+template <class Impl>
+NaiveStaggeredFermion<Impl>::NaiveStaggeredFermion(GridCartesian &Fgrid, GridRedBlackCartesian &Hgrid, 
+						   RealD _mass,
+						   RealD _c1, RealD _u0,
+						   const ImplParams &p)
+  : Kernels(p),
+    _grid(&Fgrid),
+    _cbgrid(&Hgrid),
+    Stencil(&Fgrid, npoint, Even, directions, displacements,p),
+    StencilEven(&Hgrid, npoint, Even, directions, displacements,p),  // source is Even
+    StencilOdd(&Hgrid, npoint, Odd, directions, displacements,p),  // source is Odd
+    mass(_mass),
+    Lebesgue(_grid),
+    LebesgueEvenOdd(_cbgrid),
+    Umu(&Fgrid),
+    UmuEven(&Hgrid),
+    UmuOdd(&Hgrid),
+    _tmp(&Hgrid)
+{
+  int vol4;
+  int LLs=1;
+  c1=_c1;
+  u0=_u0;
+  vol4= _grid->oSites();
+  Stencil.BuildSurfaceList(LLs,vol4);
+  vol4= _cbgrid->oSites();
+  StencilEven.BuildSurfaceList(LLs,vol4);
+  StencilOdd.BuildSurfaceList(LLs,vol4);
+}
+
+template <class Impl>
+NaiveStaggeredFermion<Impl>::NaiveStaggeredFermion(GaugeField &_U, GridCartesian &Fgrid,
+						   GridRedBlackCartesian &Hgrid, RealD _mass,
+						   RealD _c1, RealD _u0,
+						   const ImplParams &p)
+  : NaiveStaggeredFermion(Fgrid,Hgrid,_mass,_c1,_u0,p)
+{
+  ImportGauge(_U);
+}
+
+////////////////////////////////////////////////////////////
+// Momentum space propagator should be 
+// https://arxiv.org/pdf/hep-lat/9712010.pdf
+//
+// mom space action.
+//   gamma_mu i ( c1 sin pmu + c2 sin 3 pmu ) + m
+//
+// must track through staggered flavour/spin reduction in literature to 
+// turn to free propagator for the one component chi field, a la page 4/5
+// of above link to implmement fourier based solver.
+////////////////////////////////////////////////////////////
+
+template <class Impl>
+void NaiveStaggeredFermion<Impl>::CopyGaugeCheckerboards(void)
+{
+  pickCheckerboard(Even, UmuEven,  Umu);
+  pickCheckerboard(Odd,  UmuOdd ,  Umu);
+}
+template <class Impl>
+void NaiveStaggeredFermion<Impl>::ImportGauge(const GaugeField &_U) 
+{
+  GaugeLinkField U(GaugeGrid());
+  DoubledGaugeField _UUU(GaugeGrid());
+  ////////////////////////////////////////////////////////
+  // Double Store should take two fields for Naik and one hop separately.
+  // Discard teh Naik as Naive
+  ////////////////////////////////////////////////////////
+  Impl::DoubleStore(GaugeGrid(), _UUU, Umu, _U, _U );
+
+  ////////////////////////////////////////////////////////
+  // Apply scale factors to get the right fermion Kinetic term
+  // Could pass coeffs into the double store to save work.
+  // 0.5 ( U p(x+mu) - Udag(x-mu) p(x-mu) ) 
+  ////////////////////////////////////////////////////////
+  for (int mu = 0; mu < Nd; mu++) {
+
+    U = PeekIndex<LorentzIndex>(Umu, mu);
+    PokeIndex<LorentzIndex>(Umu, U*( 0.5*c1/u0), mu );
+    
+    U = PeekIndex<LorentzIndex>(Umu, mu+4);
+    PokeIndex<LorentzIndex>(Umu, U*(-0.5*c1/u0), mu+4);
+
+  }
+
+  CopyGaugeCheckerboards();
+}
+
+/////////////////////////////
+// Implement the interface
+/////////////////////////////
+
+template <class Impl>
+void NaiveStaggeredFermion<Impl>::M(const FermionField &in, FermionField &out) {
+  out.Checkerboard() = in.Checkerboard();
+  Dhop(in, out, DaggerNo);
+  axpy(out, mass, in, out);
+}
+
+template <class Impl>
+void NaiveStaggeredFermion<Impl>::Mdag(const FermionField &in, FermionField &out) {
+  out.Checkerboard() = in.Checkerboard();
+  Dhop(in, out, DaggerYes);
+  axpy(out, mass, in, out);
+}
+
+template <class Impl>
+void NaiveStaggeredFermion<Impl>::Meooe(const FermionField &in, FermionField &out) {
+  if (in.Checkerboard() == Odd) {
+    DhopEO(in, out, DaggerNo);
+  } else {
+    DhopOE(in, out, DaggerNo);
+  }
+}
+template <class Impl>
+void NaiveStaggeredFermion<Impl>::MeooeDag(const FermionField &in, FermionField &out) {
+  if (in.Checkerboard() == Odd) {
+    DhopEO(in, out, DaggerYes);
+  } else {
+    DhopOE(in, out, DaggerYes);
+  }
+}
+
+template <class Impl>
+void NaiveStaggeredFermion<Impl>::Mooee(const FermionField &in, FermionField &out) {
+  out.Checkerboard() = in.Checkerboard();
+  typename FermionField::scalar_type scal(mass);
+  out = scal * in;
+}
+
+template <class Impl>
+void NaiveStaggeredFermion<Impl>::MooeeDag(const FermionField &in, FermionField &out) {
+  out.Checkerboard() = in.Checkerboard();
+  Mooee(in, out);
+}
+
+template <class Impl>
+void NaiveStaggeredFermion<Impl>::MooeeInv(const FermionField &in, FermionField &out) {
+  out.Checkerboard() = in.Checkerboard();
+  out = (1.0 / (mass)) * in;
+}
+
+template <class Impl>
+void NaiveStaggeredFermion<Impl>::MooeeInvDag(const FermionField &in, FermionField &out) 
+{
+  out.Checkerboard() = in.Checkerboard();
+  MooeeInv(in, out);
+}
+
+///////////////////////////////////
+// Internal
+///////////////////////////////////
+
+template <class Impl>
+void NaiveStaggeredFermion<Impl>::DerivInternal(StencilImpl &st, DoubledGaugeField &U,
+						GaugeField & mat,
+						const FermionField &A, const FermionField &B, int dag) 
+{
+  assert((dag == DaggerNo) || (dag == DaggerYes));
+
+  Compressor compressor;
+
+  FermionField Btilde(B.Grid());
+  FermionField Atilde(B.Grid());
+  Atilde = A;
+
+  st.HaloExchange(B, compressor);
+
+  for (int mu = 0; mu < Nd; mu++) {
+
+    ////////////////////////
+    // Call the single hop
+    ////////////////////////
+    autoView( U_v      , U, CpuRead);
+    autoView( B_v      , B, CpuWrite);
+    autoView( Btilde_v , Btilde, CpuWrite);
+    thread_for(sss,B.Grid()->oSites(),{
+      Kernels::DhopDirKernel(st, U_v, U_v, st.CommBuf(), sss, sss, B_v, Btilde_v, mu,1);
+    });
+
+    assert(0);// need to figure out the force interface with a blasted three link term.
+    
+  }
+}
+
+template <class Impl>
+void NaiveStaggeredFermion<Impl>::DhopDeriv(GaugeField &mat, const FermionField &U, const FermionField &V, int dag) {
+
+  conformable(U.Grid(), _grid);
+  conformable(U.Grid(), V.Grid());
+  conformable(U.Grid(), mat.Grid());
+
+  mat.Checkerboard() = U.Checkerboard();
+
+  DerivInternal(Stencil, Umu, mat, U, V, dag);
+}
+
+template <class Impl>
+void NaiveStaggeredFermion<Impl>::DhopDerivOE(GaugeField &mat, const FermionField &U, const FermionField &V, int dag) {
+
+  conformable(U.Grid(), _cbgrid);
+  conformable(U.Grid(), V.Grid());
+  conformable(U.Grid(), mat.Grid());
+
+  assert(V.Checkerboard() == Even);
+  assert(U.Checkerboard() == Odd);
+  mat.Checkerboard() = Odd;
+
+  DerivInternal(StencilEven, UmuOdd, mat, U, V, dag);
+}
+
+template <class Impl>
+void NaiveStaggeredFermion<Impl>::DhopDerivEO(GaugeField &mat, const FermionField &U, const FermionField &V, int dag) {
+
+  conformable(U.Grid(), _cbgrid);
+  conformable(U.Grid(), V.Grid());
+  conformable(U.Grid(), mat.Grid());
+
+  assert(V.Checkerboard() == Odd);
+  assert(U.Checkerboard() == Even);
+  mat.Checkerboard() = Even;
+
+  DerivInternal(StencilOdd, UmuEven, mat, U, V, dag);
+}
+
+template <class Impl>
+void NaiveStaggeredFermion<Impl>::Dhop(const FermionField &in, FermionField &out, int dag) 
+{
+  DhopCalls+=2;
+  conformable(in.Grid(), _grid);  // verifies full grid
+  conformable(in.Grid(), out.Grid());
+
+  out.Checkerboard() = in.Checkerboard();
+
+  DhopInternal(Stencil, Lebesgue, Umu, in, out, dag);
+}
+
+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
+
+  assert(in.Checkerboard() == Even);
+  out.Checkerboard() = Odd;
+
+  DhopInternal(StencilEven, LebesgueEvenOdd, UmuOdd, in, out, dag);
+}
+
+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
+
+  assert(in.Checkerboard() == Odd);
+  out.Checkerboard() = Even;
+
+  DhopInternal(StencilOdd, LebesgueEvenOdd, UmuEven, in, out, dag);
+}
+
+template <class Impl>
+void NaiveStaggeredFermion<Impl>::Mdir(const FermionField &in, FermionField &out, int dir, int disp) 
+{
+  DhopDir(in, out, dir, disp);
+}
+template <class Impl>
+void NaiveStaggeredFermion<Impl>::MdirAll(const FermionField &in, std::vector<FermionField> &out) 
+{
+  assert(0); // Not implemented yet
+}
+
+template <class Impl>
+void NaiveStaggeredFermion<Impl>::DhopDir(const FermionField &in, FermionField &out, int dir, int disp) 
+{
+
+  Compressor compressor;
+  Stencil.HaloExchange(in, compressor);
+  autoView( Umu_v   ,  Umu, CpuRead);
+  autoView( in_v    ,  in, CpuRead);
+  autoView( out_v   , out, CpuWrite);
+  //  thread_for( sss, in.Grid()->oSites(),{
+  //    Kernels::DhopDirKernel(Stencil, Umu_v, Stencil.CommBuf(), sss, sss, in_v, out_v, dir, disp);
+  //  });
+  assert(0);
+};
+
+
+template <class Impl>
+void NaiveStaggeredFermion<Impl>::DhopInternal(StencilImpl &st, LebesgueOrder &lo,
+					       DoubledGaugeField &U,
+					       const FermionField &in,
+					       FermionField &out, int dag) 
+{
+  if ( StaggeredKernelsStatic::Comms == StaggeredKernelsStatic::CommsAndCompute )
+    DhopInternalOverlappedComms(st,lo,U,in,out,dag);
+  else
+    DhopInternalSerialComms(st,lo,U,in,out,dag);
+}
+template <class Impl>
+void NaiveStaggeredFermion<Impl>::DhopInternalOverlappedComms(StencilImpl &st, LebesgueOrder &lo,
+							      DoubledGaugeField &U,
+							      const FermionField &in,
+							      FermionField &out, int dag) 
+{
+  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>
+void NaiveStaggeredFermion<Impl>::DhopInternalSerialComms(StencilImpl &st, LebesgueOrder &lo,
+							  DoubledGaugeField &U,
+							  const FermionField &in,
+							  FermionField &out, int dag) 
+{
+  assert((dag == DaggerNo) || (dag == DaggerYes));
+
+  DhopTotalTime   -= usecond();
+
+  DhopCommTime    -= usecond();
+  Compressor compressor;
+  st.HaloExchange(in, compressor);
+  DhopCommTime    += usecond();
+
+  DhopComputeTime -= usecond();
+  {
+    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.
+////////////////////////////////////////////////////////
+template <class Impl>
+void NaiveStaggeredFermion<Impl>::ContractConservedCurrent(PropagatorField &q_in_1,
+							      PropagatorField &q_in_2,
+							      PropagatorField &q_out,
+							      PropagatorField &src,
+							      Current curr_type,
+							      unsigned int mu)
+{
+  assert(0);
+}
+
+template <class Impl>
+void NaiveStaggeredFermion<Impl>::SeqConservedCurrent(PropagatorField &q_in,
+                                                         PropagatorField &q_out,
+                                                         PropagatorField &src,
+                                                         Current curr_type,
+                                                         unsigned int mu, 
+                                                         unsigned int tmin,
+                                              unsigned int tmax,
+					      ComplexField &lattice_cmplx)
+{
+  assert(0);
+
+}
+
+NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/implementation/PartialFractionFermion5DImplementation.h b/Grid/qcd/action/fermion/implementation/PartialFractionFermion5DImplementation.h
index edc674cc..0206828b 100644
--- a/Grid/qcd/action/fermion/implementation/PartialFractionFermion5DImplementation.h
+++ b/Grid/qcd/action/fermion/implementation/PartialFractionFermion5DImplementation.h
@@ -269,16 +269,14 @@ void   PartialFractionFermion5D<Impl>::M_internal(const FermionField &psi, Fermi
 }
 
 template<class Impl>
-RealD  PartialFractionFermion5D<Impl>::M    (const FermionField &in, FermionField &out)
+void PartialFractionFermion5D<Impl>::M    (const FermionField &in, FermionField &out)
 {
   M_internal(in,out,DaggerNo);
-  return norm2(out);
 }
 template<class Impl>
-RealD  PartialFractionFermion5D<Impl>::Mdag (const FermionField &in, FermionField &out)
+void PartialFractionFermion5D<Impl>::Mdag (const FermionField &in, FermionField &out)
 {
   M_internal(in,out,DaggerYes);
-  return norm2(out);
 }
 
 template<class Impl>
diff --git a/Grid/qcd/action/fermion/implementation/StaggeredKernelsAsm.h b/Grid/qcd/action/fermion/implementation/StaggeredKernelsAsm.h
index 1a13e73a..63fd2a2f 100644
--- a/Grid/qcd/action/fermion/implementation/StaggeredKernelsAsm.h
+++ b/Grid/qcd/action/fermion/implementation/StaggeredKernelsAsm.h
@@ -618,10 +618,10 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 NAMESPACE_BEGIN(Grid);
 
 template <class Impl>
-void StaggeredKernels<Impl>::DhopSiteAsm(StencilImpl &st, LebesgueOrder &lo, 
+void StaggeredKernels<Impl>::DhopSiteAsm(StencilView &st,
 					 DoubledGaugeFieldView &U,
 					 DoubledGaugeFieldView &UUU,
-					 SiteSpinor *buf, int LLs,
+					 SiteSpinor *buf, int sF,
 					 int sU, const FermionFieldView &in, FermionFieldView &out,int dag) 
 {
   assert(0);
@@ -680,12 +680,13 @@ void StaggeredKernels<Impl>::DhopSiteAsm(StencilImpl &st, LebesgueOrder &lo,
   gauge2 =(uint64_t)&UU[sU]( Z );				\
   gauge3 =(uint64_t)&UU[sU]( T ); 
   
+
   // This is the single precision 5th direction vectorised kernel
 #include <Grid/simd/Intel512single.h>
-template <> void StaggeredKernels<StaggeredVec5dImplF>::DhopSiteAsm(StencilImpl &st, LebesgueOrder &lo, 
+template <> void StaggeredKernels<StaggeredVec5dImplF>::DhopSiteAsm(StencilView &st,
 								    DoubledGaugeFieldView &U,
 								    DoubledGaugeFieldView &UUU,
-								    SiteSpinor *buf, int LLs,
+								    SiteSpinor *buf, int sF,
 								    int sU, const FermionFieldView &in, FermionFieldView &out,int dag) 
 {
 #ifdef AVX512
@@ -702,9 +703,10 @@ template <> void StaggeredKernels<StaggeredVec5dImplF>::DhopSiteAsm(StencilImpl
   StencilEntry *SE2;
   StencilEntry *SE3;
 
-   for(int s=0;s<LLs;s++){
+  //   for(int s=0;s<LLs;s++){
 
-    int sF=s+LLs*sU;
+  //    int sF=s+LLs*sU;
+  {
     // Xp, Yp, Zp, Tp
     PREPARE(Xp,Yp,Zp,Tp,0,U);
     LOAD_CHI(addr0,addr1,addr2,addr3);
@@ -736,10 +738,10 @@ template <> void StaggeredKernels<StaggeredVec5dImplF>::DhopSiteAsm(StencilImpl
 }
 
 #include <Grid/simd/Intel512double.h>
-template <> void StaggeredKernels<StaggeredVec5dImplD>::DhopSiteAsm(StencilImpl &st, LebesgueOrder &lo, 
+template <> void StaggeredKernels<StaggeredVec5dImplD>::DhopSiteAsm(StencilView &st, 
 								    DoubledGaugeFieldView &U,
 								    DoubledGaugeFieldView &UUU,
-								    SiteSpinor *buf, int LLs,
+								    SiteSpinor *buf, int sF,
 								    int sU, const FermionFieldView &in, FermionFieldView &out, int dag) 
 {
 #ifdef AVX512
@@ -756,8 +758,9 @@ template <> void StaggeredKernels<StaggeredVec5dImplD>::DhopSiteAsm(StencilImpl
   StencilEntry *SE2;
   StencilEntry *SE3;
 
-  for(int s=0;s<LLs;s++){
-    int sF=s+LLs*sU;
+  //  for(int s=0;s<LLs;s++){
+  //    int sF=s+LLs*sU;
+  {
     // Xp, Yp, Zp, Tp
     PREPARE(Xp,Yp,Zp,Tp,0,U);
     LOAD_CHI(addr0,addr1,addr2,addr3);
@@ -821,10 +824,10 @@ template <> void StaggeredKernels<StaggeredVec5dImplD>::DhopSiteAsm(StencilImpl
   // This is the single precision 5th direction vectorised kernel
 
 #include <Grid/simd/Intel512single.h>
-template <> void StaggeredKernels<StaggeredImplF>::DhopSiteAsm(StencilImpl &st, LebesgueOrder &lo, 
+template <> void StaggeredKernels<StaggeredImplF>::DhopSiteAsm(StencilView &st, 
 							       DoubledGaugeFieldView &U,
 							       DoubledGaugeFieldView &UUU,
-							       SiteSpinor *buf, int LLs,
+							       SiteSpinor *buf, int sF,
 							       int sU, const FermionFieldView &in, FermionFieldView &out,int dag) 
 {
 #ifdef AVX512
@@ -841,9 +844,9 @@ template <> void StaggeredKernels<StaggeredImplF>::DhopSiteAsm(StencilImpl &st,
   StencilEntry *SE2;
   StencilEntry *SE3;
 
-  for(int s=0;s<LLs;s++){
-    
-    int sF=s+LLs*sU;
+  //  for(int s=0;s<LLs;s++){
+  //    int sF=s+LLs*sU;
+  {
     // Xp, Yp, Zp, Tp
     PREPARE(Xp,Yp,Zp,Tp,0,U);
     LOAD_CHIa(addr0,addr1);
@@ -890,10 +893,10 @@ template <> void StaggeredKernels<StaggeredImplF>::DhopSiteAsm(StencilImpl &st,
 }
 
 #include <Grid/simd/Intel512double.h>
-template <> void StaggeredKernels<StaggeredImplD>::DhopSiteAsm(StencilImpl &st, LebesgueOrder &lo, 
+template <> void StaggeredKernels<StaggeredImplD>::DhopSiteAsm(StencilView &st, 
 							       DoubledGaugeFieldView &U,
 							       DoubledGaugeFieldView &UUU,
-							       SiteSpinor *buf, int LLs,
+							       SiteSpinor *buf, int sF,
 							       int sU, const FermionFieldView &in, FermionFieldView &out,int dag) 
 {
 #ifdef AVX512
@@ -910,9 +913,9 @@ template <> void StaggeredKernels<StaggeredImplD>::DhopSiteAsm(StencilImpl &st,
   StencilEntry *SE2;
   StencilEntry *SE3;
 
-  for(int s=0;s<LLs;s++){
-    
-    int sF=s+LLs*sU;
+  //  for(int s=0;s<LLs;s++){
+  //    int sF=s+LLs*sU;
+  {
     // Xp, Yp, Zp, Tp
     PREPARE(Xp,Yp,Zp,Tp,0,U);
     LOAD_CHIa(addr0,addr1);
diff --git a/Grid/qcd/action/fermion/implementation/StaggeredKernelsHand.h b/Grid/qcd/action/fermion/implementation/StaggeredKernelsHand.h
index 62370d75..ee53de49 100644
--- a/Grid/qcd/action/fermion/implementation/StaggeredKernelsHand.h
+++ b/Grid/qcd/action/fermion/implementation/StaggeredKernelsHand.h
@@ -146,9 +146,10 @@ NAMESPACE_BEGIN(Grid);
 
 
 template <class Impl>
-void StaggeredKernels<Impl>::DhopSiteHand(StencilImpl &st, LebesgueOrder &lo, 
+template <int Naik>
+void StaggeredKernels<Impl>::DhopSiteHand(StencilView &st,
 					  DoubledGaugeFieldView &U,DoubledGaugeFieldView &UUU,
-					  SiteSpinor *buf, int LLs, int sU, 
+					  SiteSpinor *buf, int sF, int sU, 
 					  const FermionFieldView &in, FermionFieldView &out,int dag) 
 {
   typedef typename Simd::scalar_type S;
@@ -181,8 +182,9 @@ void StaggeredKernels<Impl>::DhopSiteHand(StencilImpl &st, LebesgueOrder &lo,
   StencilEntry *SE;
   int skew;
 
-  for(int s=0;s<LLs;s++){
-    int sF=s+LLs*sU;
+  //  for(int s=0;s<LLs;s++){
+  //    int sF=s+LLs*sU;
+  {
 
     skew = 0;
     HAND_STENCIL_LEG_BEGIN(Xp,3,skew,even);  
@@ -193,6 +195,7 @@ void StaggeredKernels<Impl>::DhopSiteHand(StencilImpl &st, LebesgueOrder &lo,
     HAND_STENCIL_LEG      (U,Ym,2,skew,odd);   
     HAND_STENCIL_LEG      (U,Zm,1,skew,even);  
     HAND_STENCIL_LEG      (U,Tm,0,skew,odd);  
+    if (Naik) {
     skew = 8;
     HAND_STENCIL_LEG(UUU,Xp,3,skew,even);  
     HAND_STENCIL_LEG(UUU,Yp,2,skew,odd);   
@@ -202,7 +205,7 @@ void StaggeredKernels<Impl>::DhopSiteHand(StencilImpl &st, LebesgueOrder &lo,
     HAND_STENCIL_LEG(UUU,Ym,2,skew,odd);   
     HAND_STENCIL_LEG(UUU,Zm,1,skew,even);  
     HAND_STENCIL_LEG(UUU,Tm,0,skew,odd);  
-    
+    }    
     if ( dag ) {
       result()()(0) = - even_0 - odd_0;
       result()()(1) = - even_1 - odd_1;
@@ -218,9 +221,10 @@ void StaggeredKernels<Impl>::DhopSiteHand(StencilImpl &st, LebesgueOrder &lo,
 
 
 template <class Impl>
-void StaggeredKernels<Impl>::DhopSiteHandInt(StencilImpl &st, LebesgueOrder &lo, 
+template <int Naik>
+void StaggeredKernels<Impl>::DhopSiteHandInt(StencilView &st, 
 					     DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU,
-					     SiteSpinor *buf, int LLs, int sU, 
+					     SiteSpinor *buf, int sF, int sU, 
 					     const FermionFieldView &in, FermionFieldView &out,int dag) 
 {
   typedef typename Simd::scalar_type S;
@@ -253,8 +257,9 @@ void StaggeredKernels<Impl>::DhopSiteHandInt(StencilImpl &st, LebesgueOrder &lo,
   StencilEntry *SE;
   int skew;
 
-  for(int s=0;s<LLs;s++){
-    int sF=s+LLs*sU;
+  //  for(int s=0;s<LLs;s++){
+  //    int sF=s+LLs*sU;
+  {
 
     even_0 = Zero();    even_1 = Zero();    even_2 = Zero();
      odd_0 = Zero();     odd_1 = Zero();     odd_2 = Zero();
@@ -268,6 +273,7 @@ void StaggeredKernels<Impl>::DhopSiteHandInt(StencilImpl &st, LebesgueOrder &lo,
     HAND_STENCIL_LEG_INT(U,Ym,2,skew,odd);   
     HAND_STENCIL_LEG_INT(U,Zm,1,skew,even);  
     HAND_STENCIL_LEG_INT(U,Tm,0,skew,odd);  
+    if (Naik) {
     skew = 8;
     HAND_STENCIL_LEG_INT(UUU,Xp,3,skew,even);  
     HAND_STENCIL_LEG_INT(UUU,Yp,2,skew,odd);   
@@ -277,7 +283,7 @@ void StaggeredKernels<Impl>::DhopSiteHandInt(StencilImpl &st, LebesgueOrder &lo,
     HAND_STENCIL_LEG_INT(UUU,Ym,2,skew,odd);   
     HAND_STENCIL_LEG_INT(UUU,Zm,1,skew,even);  
     HAND_STENCIL_LEG_INT(UUU,Tm,0,skew,odd);  
-
+    }
     // Assume every site must be connected to at least one interior point. No 1^4 subvols.
     if ( dag ) {
       result()()(0) = - even_0 - odd_0;
@@ -294,9 +300,10 @@ void StaggeredKernels<Impl>::DhopSiteHandInt(StencilImpl &st, LebesgueOrder &lo,
 
 
 template <class Impl>
-void StaggeredKernels<Impl>::DhopSiteHandExt(StencilImpl &st, LebesgueOrder &lo, 
+template <int Naik>
+void StaggeredKernels<Impl>::DhopSiteHandExt(StencilView &st,
 					     DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU,
-					     SiteSpinor *buf, int LLs, int sU, 
+					     SiteSpinor *buf, int sF, int sU, 
 					     const FermionFieldView &in, FermionFieldView &out,int dag) 
 {
   typedef typename Simd::scalar_type S;
@@ -329,8 +336,9 @@ void StaggeredKernels<Impl>::DhopSiteHandExt(StencilImpl &st, LebesgueOrder &lo,
   StencilEntry *SE;
   int skew;
 
-  for(int s=0;s<LLs;s++){
-    int sF=s+LLs*sU;
+  //  for(int s=0;s<LLs;s++){
+  //    int sF=s+LLs*sU;
+  {
 
     even_0 = Zero();    even_1 = Zero();    even_2 = Zero();
      odd_0 = Zero();     odd_1 = Zero();     odd_2 = Zero();
@@ -344,6 +352,7 @@ void StaggeredKernels<Impl>::DhopSiteHandExt(StencilImpl &st, LebesgueOrder &lo,
     HAND_STENCIL_LEG_EXT(U,Ym,2,skew,odd);   
     HAND_STENCIL_LEG_EXT(U,Zm,1,skew,even);  
     HAND_STENCIL_LEG_EXT(U,Tm,0,skew,odd);  
+    if (Naik) {
     skew = 8;
     HAND_STENCIL_LEG_EXT(UUU,Xp,3,skew,even);  
     HAND_STENCIL_LEG_EXT(UUU,Yp,2,skew,odd);   
@@ -353,7 +362,7 @@ void StaggeredKernels<Impl>::DhopSiteHandExt(StencilImpl &st, LebesgueOrder &lo,
     HAND_STENCIL_LEG_EXT(UUU,Ym,2,skew,odd);   
     HAND_STENCIL_LEG_EXT(UUU,Zm,1,skew,even);  
     HAND_STENCIL_LEG_EXT(UUU,Tm,0,skew,odd);  
-
+    }
     // Add sum of all exterior connected stencil legs
     if ( nmu ) { 
       if ( dag ) {
@@ -370,6 +379,7 @@ void StaggeredKernels<Impl>::DhopSiteHandExt(StencilImpl &st, LebesgueOrder &lo,
   }
 }
 
+/*
 #define DHOP_SITE_HAND_INSTANTIATE(IMPL)				\
   template void StaggeredKernels<IMPL>::DhopSiteHand(StencilImpl &st, LebesgueOrder &lo, \
 						     DoubledGaugeFieldView &U,DoubledGaugeFieldView &UUU, \
@@ -385,7 +395,7 @@ void StaggeredKernels<Impl>::DhopSiteHandExt(StencilImpl &st, LebesgueOrder &lo,
 						     DoubledGaugeFieldView &U,DoubledGaugeFieldView &UUU, \
 						     SiteSpinor *buf, int LLs, int sU, \
 						     const FermionFieldView &in, FermionFieldView &out, int dag); \
-
+*/
 #undef LOAD_CHI
 
 NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/implementation/StaggeredKernelsImplementation.h b/Grid/qcd/action/fermion/implementation/StaggeredKernelsImplementation.h
index d301556c..141725a7 100644
--- a/Grid/qcd/action/fermion/implementation/StaggeredKernelsImplementation.h
+++ b/Grid/qcd/action/fermion/implementation/StaggeredKernelsImplementation.h
@@ -37,9 +37,9 @@ NAMESPACE_BEGIN(Grid);
   if (SE->_is_local ) {						\
     if (SE->_permute) {						\
       chi_p = &chi;						\
-      permute(chi,  in[SE->_offset], ptype);		\
+      permute(chi,  in[SE->_offset], ptype);			\
     } else {							\
-      chi_p = &in[SE->_offset];				\
+      chi_p = &in[SE->_offset];					\
     }								\
   } else {							\
     chi_p = &buf[SE->_offset];					\
@@ -51,15 +51,15 @@ NAMESPACE_BEGIN(Grid);
   if (SE->_is_local ) {						\
     if (SE->_permute) {						\
       chi_p = &chi;						\
-      permute(chi,  in[SE->_offset], ptype);		\
+      permute(chi,  in[SE->_offset], ptype);			\
     } else {							\
-      chi_p = &in[SE->_offset];				\
+      chi_p = &in[SE->_offset];					\
     }								\
   } else if ( st.same_node[Dir] ) {				\
     chi_p = &buf[SE->_offset];					\
   }								\
   if (SE->_is_local || st.same_node[Dir] ) {			\
-    multLink(Uchi, U[sU], *chi_p, Dir);			\
+    multLink(Uchi, U[sU], *chi_p, Dir);				\
   }
 
 #define GENERIC_STENCIL_LEG_EXT(U,Dir,skew,multLink)		\
@@ -67,7 +67,7 @@ NAMESPACE_BEGIN(Grid);
   if ((!SE->_is_local) && (!st.same_node[Dir]) ) {		\
     nmu++;							\
     chi_p = &buf[SE->_offset];					\
-    multLink(Uchi, U[sU], *chi_p, Dir);			\
+    multLink(Uchi, U[sU], *chi_p, Dir);				\
   }
 
 template <class Impl>
@@ -78,10 +78,12 @@ StaggeredKernels<Impl>::StaggeredKernels(const ImplParams &p) : Base(p){};
 // Int, Ext, Int+Ext cases for comms overlap
 ////////////////////////////////////////////////////////////////////////////////////
 template <class Impl>
-void StaggeredKernels<Impl>::DhopSiteGeneric(StencilImpl &st, LebesgueOrder &lo, 
+template <int Naik>
+void StaggeredKernels<Impl>::DhopSiteGeneric(StencilView &st, 
 					     DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU,
-					     SiteSpinor *buf, int LLs, int sU, 
-					     const FermionFieldView &in, FermionFieldView &out, int dag) {
+					     SiteSpinor *buf, int sF, int sU, 
+					     const FermionFieldView &in, FermionFieldView &out, int dag) 
+{
   const SiteSpinor *chi_p;
   SiteSpinor chi;
   SiteSpinor Uchi;
@@ -89,8 +91,10 @@ void StaggeredKernels<Impl>::DhopSiteGeneric(StencilImpl &st, LebesgueOrder &lo,
   int ptype;
   int skew;
 
-  for(int s=0;s<LLs;s++){
-    int sF=LLs*sU+s;
+  //  for(int s=0;s<LLs;s++){
+  //
+  //    int sF=LLs*sU+s;
+  {
     skew = 0;
     GENERIC_STENCIL_LEG(U,Xp,skew,Impl::multLink);
     GENERIC_STENCIL_LEG(U,Yp,skew,Impl::multLinkAdd);
@@ -100,6 +104,7 @@ void StaggeredKernels<Impl>::DhopSiteGeneric(StencilImpl &st, LebesgueOrder &lo,
     GENERIC_STENCIL_LEG(U,Ym,skew,Impl::multLinkAdd);
     GENERIC_STENCIL_LEG(U,Zm,skew,Impl::multLinkAdd);
     GENERIC_STENCIL_LEG(U,Tm,skew,Impl::multLinkAdd);
+    if ( Naik ) {
     skew=8;
     GENERIC_STENCIL_LEG(UUU,Xp,skew,Impl::multLinkAdd);
     GENERIC_STENCIL_LEG(UUU,Yp,skew,Impl::multLinkAdd);
@@ -109,6 +114,7 @@ void StaggeredKernels<Impl>::DhopSiteGeneric(StencilImpl &st, LebesgueOrder &lo,
     GENERIC_STENCIL_LEG(UUU,Ym,skew,Impl::multLinkAdd);
     GENERIC_STENCIL_LEG(UUU,Zm,skew,Impl::multLinkAdd);
     GENERIC_STENCIL_LEG(UUU,Tm,skew,Impl::multLinkAdd);
+    }
     if ( dag ) { 
       Uchi = - Uchi;
     } 
@@ -120,9 +126,10 @@ void StaggeredKernels<Impl>::DhopSiteGeneric(StencilImpl &st, LebesgueOrder &lo,
   // Only contributions from interior of our node
   ///////////////////////////////////////////////////
 template <class Impl>
-void StaggeredKernels<Impl>::DhopSiteGenericInt(StencilImpl &st, LebesgueOrder &lo, 
+template <int Naik>
+void StaggeredKernels<Impl>::DhopSiteGenericInt(StencilView &st, 
 						DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU,
-						SiteSpinor *buf, int LLs, int sU, 
+						SiteSpinor *buf, int sF, int sU, 
 						const FermionFieldView &in, FermionFieldView &out,int dag) {
   const SiteSpinor *chi_p;
   SiteSpinor chi;
@@ -131,8 +138,9 @@ void StaggeredKernels<Impl>::DhopSiteGenericInt(StencilImpl &st, LebesgueOrder &
   int ptype;
   int skew ;
 
-  for(int s=0;s<LLs;s++){
-    int sF=LLs*sU+s;
+  //  for(int s=0;s<LLs;s++){
+  //    int sF=LLs*sU+s;
+  {
     skew = 0;
     Uchi=Zero();
     GENERIC_STENCIL_LEG_INT(U,Xp,skew,Impl::multLinkAdd);
@@ -143,6 +151,7 @@ void StaggeredKernels<Impl>::DhopSiteGenericInt(StencilImpl &st, LebesgueOrder &
     GENERIC_STENCIL_LEG_INT(U,Ym,skew,Impl::multLinkAdd);
     GENERIC_STENCIL_LEG_INT(U,Zm,skew,Impl::multLinkAdd);
     GENERIC_STENCIL_LEG_INT(U,Tm,skew,Impl::multLinkAdd);
+    if ( Naik ) {
     skew=8;
     GENERIC_STENCIL_LEG_INT(UUU,Xp,skew,Impl::multLinkAdd);
     GENERIC_STENCIL_LEG_INT(UUU,Yp,skew,Impl::multLinkAdd);
@@ -152,6 +161,7 @@ void StaggeredKernels<Impl>::DhopSiteGenericInt(StencilImpl &st, LebesgueOrder &
     GENERIC_STENCIL_LEG_INT(UUU,Ym,skew,Impl::multLinkAdd);
     GENERIC_STENCIL_LEG_INT(UUU,Zm,skew,Impl::multLinkAdd);
     GENERIC_STENCIL_LEG_INT(UUU,Tm,skew,Impl::multLinkAdd);
+    }
     if ( dag ) {
       Uchi = - Uchi;
     }
@@ -164,9 +174,10 @@ void StaggeredKernels<Impl>::DhopSiteGenericInt(StencilImpl &st, LebesgueOrder &
   // Only contributions from exterior of our node
   ///////////////////////////////////////////////////
 template <class Impl>
-void StaggeredKernels<Impl>::DhopSiteGenericExt(StencilImpl &st, LebesgueOrder &lo, 
+template <int Naik>
+void StaggeredKernels<Impl>::DhopSiteGenericExt(StencilView &st, 
 						DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU,
-						SiteSpinor *buf, int LLs, int sU,
+						SiteSpinor *buf, int sF, int sU,
 						const FermionFieldView &in, FermionFieldView &out,int dag) {
   const SiteSpinor *chi_p;
   //  SiteSpinor chi;
@@ -176,8 +187,9 @@ void StaggeredKernels<Impl>::DhopSiteGenericExt(StencilImpl &st, LebesgueOrder &
   int nmu=0;
   int skew ;
 
-  for(int s=0;s<LLs;s++){
-    int sF=LLs*sU+s;
+  //  for(int s=0;s<LLs;s++){
+  //    int sF=LLs*sU+s;
+  {
     skew = 0;
     Uchi=Zero();
     GENERIC_STENCIL_LEG_EXT(U,Xp,skew,Impl::multLinkAdd);
@@ -188,6 +200,7 @@ void StaggeredKernels<Impl>::DhopSiteGenericExt(StencilImpl &st, LebesgueOrder &
     GENERIC_STENCIL_LEG_EXT(U,Ym,skew,Impl::multLinkAdd);
     GENERIC_STENCIL_LEG_EXT(U,Zm,skew,Impl::multLinkAdd);
     GENERIC_STENCIL_LEG_EXT(U,Tm,skew,Impl::multLinkAdd);
+    if ( Naik ) {
     skew=8;
     GENERIC_STENCIL_LEG_EXT(UUU,Xp,skew,Impl::multLinkAdd);
     GENERIC_STENCIL_LEG_EXT(UUU,Yp,skew,Impl::multLinkAdd);
@@ -197,7 +210,7 @@ void StaggeredKernels<Impl>::DhopSiteGenericExt(StencilImpl &st, LebesgueOrder &
     GENERIC_STENCIL_LEG_EXT(UUU,Ym,skew,Impl::multLinkAdd);
     GENERIC_STENCIL_LEG_EXT(UUU,Zm,skew,Impl::multLinkAdd);
     GENERIC_STENCIL_LEG_EXT(UUU,Tm,skew,Impl::multLinkAdd);
-
+    }
     if ( nmu ) { 
       if ( dag ) { 
 	out[sF] = out[sF] - Uchi;
@@ -211,72 +224,9 @@ void StaggeredKernels<Impl>::DhopSiteGenericExt(StencilImpl &st, LebesgueOrder &
 ////////////////////////////////////////////////////////////////////////////////////
 // Driving / wrapping routine to select right kernel
 ////////////////////////////////////////////////////////////////////////////////////
-
 template <class Impl>
-void StaggeredKernels<Impl>::DhopSiteDag(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU,
-					 SiteSpinor *buf, int LLs, int sU,
-					 const FermionFieldView &in, FermionFieldView &out,
-					 int interior,int exterior)
-{
-  int dag=1;
-  DhopSite(st,lo,U,UUU,buf,LLs,sU,in,out,dag,interior,exterior);
-};
-
-template <class Impl>
-void StaggeredKernels<Impl>::DhopSite(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU,
-				      SiteSpinor *buf, int LLs, int sU,
-				      const FermionFieldView &in, FermionFieldView &out,
-				      int interior,int exterior)
-{
-  int dag=0;
-  DhopSite(st,lo,U,UUU,buf,LLs,sU,in,out,dag,interior,exterior);
-};
-
-template <class Impl>
-void StaggeredKernels<Impl>::DhopSite(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU,
-				      SiteSpinor *buf, int LLs,
-				      int sU, const FermionFieldView &in, FermionFieldView &out,
-				      int dag,int interior,int exterior) 
-{
-  switch(Opt) {
-#ifdef AVX512
-  case OptInlineAsm:
-    if ( interior && exterior ) {
-      DhopSiteAsm(st,lo,U,UUU,buf,LLs,sU,in,out,dag);
-    } else { 
-      std::cout << GridLogError << "Cannot overlap comms and compute with Staggered assembly"<<std::endl;
-      assert(0);
-    }
-    break;
-#endif
-  case OptHandUnroll:
-    if ( interior && exterior ) {
-      DhopSiteHand   (st,lo,U,UUU,buf,LLs,sU,in,out,dag);
-    } else if ( interior ) {
-      DhopSiteHandInt(st,lo,U,UUU,buf,LLs,sU,in,out,dag);
-    } else if ( exterior ) {
-      DhopSiteHandExt(st,lo,U,UUU,buf,LLs,sU,in,out,dag);
-    }
-    break;
-  case OptGeneric:
-    if ( interior && exterior ) {
-      DhopSiteGeneric   (st,lo,U,UUU,buf,LLs,sU,in,out,dag);
-    } else if ( interior ) {
-      DhopSiteGenericInt(st,lo,U,UUU,buf,LLs,sU,in,out,dag);
-    } else if ( exterior ) {
-      DhopSiteGenericExt(st,lo,U,UUU,buf,LLs,sU,in,out,dag);
-    }
-    break;
-  default:
-    std::cout<<"Oops Opt = "<<Opt<<std::endl;
-    assert(0);
-    break;
-  }
-};
-
-template <class Impl>
-void StaggeredKernels<Impl>::DhopDirKernel( StencilImpl &st, DoubledGaugeFieldView &U,  DoubledGaugeFieldView &UUU, SiteSpinor *buf, int sF,
-					    int sU, const FermionFieldView &in, FermionFieldView &out, int dir, int disp) 
+void StaggeredKernels<Impl>::DhopDirKernel(StencilImpl &st, DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU, SiteSpinor * buf,
+					   int sF, int sU, const FermionFieldView &in, FermionFieldView &out, int dir,int disp)
 {
   // Disp should be either +1,-1,+3,-3
   // What about "dag" ?
@@ -285,6 +235,108 @@ void StaggeredKernels<Impl>::DhopDirKernel( StencilImpl &st, DoubledGaugeFieldVi
   assert(0);
 }
 
+#define KERNEL_CALLNB(A,improved)					\
+  const uint64_t    NN = Nsite*Ls;					\
+  accelerator_forNB( ss, NN, Simd::Nsimd(), {				\
+      int sF = ss;							\
+      int sU = ss/Ls;							\
+      ThisKernel:: template A<improved>(st_v,U_v,UUU_v,buf,sF,sU,in_v,out_v,dag); \
+    });
+
+#define KERNEL_CALL(A,improved) KERNEL_CALLNB(A,improved); accelerator_barrier(); 
+
+#define ASM_CALL(A)							\
+  const uint64_t    NN = Nsite*Ls;					\
+  thread_for( ss, NN, {							\
+      int sF = ss;							\
+      int sU = ss/Ls;							\
+      ThisKernel::A(st_v,U_v,UUU_v,buf,sF,sU,in_v,out_v,dag);		\
+  });
+
+template <class Impl>
+void StaggeredKernels<Impl>::DhopImproved(StencilImpl &st, LebesgueOrder &lo, 
+					  DoubledGaugeField &U, DoubledGaugeField &UUU, 
+					  const FermionField &in, FermionField &out, int dag, int interior,int exterior)
+{
+  GridBase *FGrid=in.Grid();  
+  GridBase *UGrid=U.Grid();  
+  typedef StaggeredKernels<Impl> ThisKernel;
+  autoView( UUU_v , UUU, AcceleratorRead);
+  autoView( U_v   ,   U, AcceleratorRead);
+  autoView( in_v  ,  in, AcceleratorRead);
+  autoView( out_v , out, AcceleratorWrite);
+  autoView( st_v  ,  st, AcceleratorRead);
+  SiteSpinor * buf = st.CommBuf();
+    
+  int Ls=1;
+  if(FGrid->Nd()==UGrid->Nd()+1){
+    Ls    = FGrid->_rdimensions[0];
+  }
+  int Nsite = UGrid->oSites();
+
+  if( interior && exterior ) { 
+    if (Opt == OptGeneric    ) { KERNEL_CALL(DhopSiteGeneric,1); return;}
+#ifndef GRID_CUDA
+    if (Opt == OptHandUnroll ) { KERNEL_CALL(DhopSiteHand,1);    return;}
+    if (Opt == OptInlineAsm  ) {  ASM_CALL(DhopSiteAsm);     return;}
+#endif
+  } else if( interior ) {
+    if (Opt == OptGeneric    ) { KERNEL_CALL(DhopSiteGenericInt,1); return;}
+#ifndef GRID_CUDA
+    if (Opt == OptHandUnroll ) { KERNEL_CALL(DhopSiteHandInt,1);    return;}
+#endif
+  } else if( exterior ) { 
+    if (Opt == OptGeneric    ) { KERNEL_CALL(DhopSiteGenericExt,1); return;}
+#ifndef GRID_CUDA
+    if (Opt == OptHandUnroll ) { KERNEL_CALL(DhopSiteHandExt,1);    return;}
+#endif
+  }
+  assert(0 && " Kernel optimisation case not covered ");
+}
+template <class Impl>
+void StaggeredKernels<Impl>::DhopNaive(StencilImpl &st, LebesgueOrder &lo, 
+				       DoubledGaugeField &U,
+				       const FermionField &in, FermionField &out, int dag, int interior,int exterior)
+{
+  GridBase *FGrid=in.Grid();  
+  GridBase *UGrid=U.Grid();  
+  typedef StaggeredKernels<Impl> ThisKernel;
+  autoView( UUU_v ,   U, AcceleratorRead);
+  autoView( U_v   ,   U, AcceleratorRead);
+  autoView( in_v  ,  in, AcceleratorRead);
+  autoView( out_v , out, AcceleratorWrite);
+  autoView( st_v  ,  st, AcceleratorRead);
+  SiteSpinor * buf = st.CommBuf();
+
+  int Ls=1;
+  if(FGrid->Nd()==UGrid->Nd()+1){
+    Ls    = FGrid->_rdimensions[0];
+  }
+  int Nsite = UGrid->oSites();
+  
+  if( interior && exterior ) { 
+    if (Opt == OptGeneric    ) { KERNEL_CALL(DhopSiteGeneric,0); return;}
+#ifndef GRID_CUDA
+    if (Opt == OptHandUnroll ) { KERNEL_CALL(DhopSiteHand,0);    return;}
+#endif
+  } else if( interior ) {
+    if (Opt == OptGeneric    ) { KERNEL_CALL(DhopSiteGenericInt,0); return;}
+#ifndef GRID_CUDA
+    if (Opt == OptHandUnroll ) { KERNEL_CALL(DhopSiteHandInt,0);    return;}
+#endif
+  } else if( exterior ) { 
+    if (Opt == OptGeneric    ) { KERNEL_CALL(DhopSiteGenericExt,0); return;}
+#ifndef GRID_CUDA
+    if (Opt == OptHandUnroll ) { KERNEL_CALL(DhopSiteHandExt,0);    return;}
+#endif
+  }
+}
+
+
+#undef KERNEL_CALLNB
+#undef KERNEL_CALL
+#undef ASM_CALL
+
 NAMESPACE_END(Grid);
 
 
diff --git a/Grid/qcd/action/fermion/implementation/WilsonCloverFermionImplementation.h b/Grid/qcd/action/fermion/implementation/WilsonCloverFermionImplementation.h
index 5744d3bb..df1bce7c 100644
--- a/Grid/qcd/action/fermion/implementation/WilsonCloverFermionImplementation.h
+++ b/Grid/qcd/action/fermion/implementation/WilsonCloverFermionImplementation.h
@@ -35,7 +35,7 @@ NAMESPACE_BEGIN(Grid);
 
 // *NOT* EO
 template <class Impl>
-RealD WilsonCloverFermion<Impl>::M(const FermionField &in, FermionField &out)
+void WilsonCloverFermion<Impl>::M(const FermionField &in, FermionField &out)
 {
   FermionField temp(out.Grid());
 
@@ -47,11 +47,10 @@ RealD WilsonCloverFermion<Impl>::M(const FermionField &in, FermionField &out)
   Mooee(in, temp);
 
   out += temp;
-  return norm2(out);
 }
 
 template <class Impl>
-RealD WilsonCloverFermion<Impl>::Mdag(const FermionField &in, FermionField &out)
+void WilsonCloverFermion<Impl>::Mdag(const FermionField &in, FermionField &out)
 {
   FermionField temp(out.Grid());
 
@@ -63,7 +62,6 @@ RealD WilsonCloverFermion<Impl>::Mdag(const FermionField &in, FermionField &out)
   MooeeDag(in, temp);
 
   out += temp;
-  return norm2(out);
 }
 
 template <class Impl>
@@ -100,46 +98,49 @@ void WilsonCloverFermion<Impl>::ImportGauge(const GaugeField &_Umu)
   Coordinate lcoor;
   typename SiteCloverType::scalar_object Qx = Zero(), Qxinv = Zero();
 
-  for (int site = 0; site < lvol; site++)
   {
-    grid->LocalIndexToLocalCoor(site, lcoor);
-    EigenCloverOp = Eigen::MatrixXcd::Zero(Ns * DimRep, Ns * DimRep);
-    peekLocalSite(Qx, CloverTerm, lcoor);
-    Qxinv = Zero();
-    //if (csw!=0){
-    for (int j = 0; j < Ns; j++)
-      for (int k = 0; k < Ns; k++)
-        for (int a = 0; a < DimRep; a++)
-          for (int b = 0; b < DimRep; b++){
-	    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, CloverTermInv, lcoor);
+    autoView(CTv,CloverTerm,CpuRead);
+    autoView(CTIv,CloverTermInv,CpuWrite);
+    for (int site = 0; site < lvol; site++) {
+      grid->LocalIndexToLocalCoor(site, lcoor);
+      EigenCloverOp = Eigen::MatrixXcd::Zero(Ns * DimRep, Ns * DimRep);
+      peekLocalSite(Qx, CTv, lcoor);
+      Qxinv = Zero();
+      //if (csw!=0){
+      for (int j = 0; j < Ns; j++)
+	for (int k = 0; k < Ns; k++)
+	  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);
+    }
   }
 
   // Separate the even and odd parts
   pickCheckerboard(Even, CloverTermEven, CloverTerm);
   pickCheckerboard(Odd, CloverTermOdd, CloverTerm);
 
-  pickCheckerboard(Even, CloverTermDagEven, adj(CloverTerm));
-  pickCheckerboard(Odd, CloverTermDagOdd, adj(CloverTerm));
+  pickCheckerboard(Even, CloverTermDagEven, closure(adj(CloverTerm)));
+  pickCheckerboard(Odd, CloverTermDagOdd, closure(adj(CloverTerm)));
 
   pickCheckerboard(Even, CloverTermInvEven, CloverTermInv);
   pickCheckerboard(Odd, CloverTermInvOdd, CloverTermInv);
 
-  pickCheckerboard(Even, CloverTermInvDagEven, adj(CloverTermInv));
-  pickCheckerboard(Odd, CloverTermInvDagOdd, adj(CloverTermInv));
+  pickCheckerboard(Even, CloverTermInvDagEven, closure(adj(CloverTermInv)));
+  pickCheckerboard(Odd, CloverTermInvDagOdd, closure(adj(CloverTermInv)));
 }
 
 template <class Impl>
diff --git a/Grid/qcd/action/fermion/implementation/WilsonFermion5DImplementation.h b/Grid/qcd/action/fermion/implementation/WilsonFermion5DImplementation.h
index 613eaa7b..2cc308cc 100644
--- a/Grid/qcd/action/fermion/implementation/WilsonFermion5DImplementation.h
+++ b/Grid/qcd/action/fermion/implementation/WilsonFermion5DImplementation.h
@@ -580,16 +580,21 @@ void WilsonFermion5D<Impl>::MomentumSpacePropagatorHt_5d(FermionField &out,const
   cosha = (one + W*W + sk) / (abs(W)*2.0);
 
   // FIXME Need a Lattice acosh
-  for(int idx=0;idx<_grid->lSites();idx++){
-    Coordinate lcoor(Nd);
-    Tcomplex cc;
-    //    RealD sgn;
-    _grid->LocalIndexToLocalCoor(idx,lcoor);
-    peekLocalSite(cc,cosha,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,lcoor);
+
+  {
+    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)  );
@@ -775,17 +780,20 @@ void WilsonFermion5D<Impl>::MomentumSpacePropagatorHt(FermionField &out,const Fe
   cosha =  (one + W*W + sk) / (abs(W)*2.0);
 
   // 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,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,lcoor);
-  }
+    pokeLocalSite(cc,a_v,lcoor);
+  }}
   
   Wea = ( exp( a) * abs(W)  );
   Wema= ( exp(-a) * abs(W)  );
@@ -861,7 +869,6 @@ void WilsonFermion5D<Impl>::MomentumSpacePropagatorHw(FermionField &out,const Fe
  * Conserved current utilities for Wilson fermions, for contracting propagators
  * to make a conserved current sink or inserting the conserved current 
  * sequentially.
- ******************************************************************************/
 
 // Helper macro to reverse Simd vector. Fixme: slow, generic implementation.
 #define REVERSE_LS(qSite, qSiteRev, Nsimd) \
@@ -877,220 +884,10 @@ void WilsonFermion5D<Impl>::MomentumSpacePropagatorHw(FermionField &out,const Fe
     merge(qSiteRev, qSiteVec); \
 }
 
-//          psi = chiralProjectPlus(Result_s[Ls/2-1]);
-//          psi+= chiralProjectMinus(Result_s[Ls/2]);
-//         PJ5q+=localInnerProduct(psi,psi);
-
-template<class vobj> 
-Lattice<vobj> spProj5p(const Lattice<vobj> & in)
-{
-  GridBase *grid=in.Grid();
-  Gamma G5(Gamma::Algebra::Gamma5);
-  Lattice<vobj> ret(grid);
-  auto ret_v = ret.View();
-  auto in_v  =  in.View();
-  thread_for(ss,grid->oSites(),{
-    ret_v[ss] = in_v[ss] + G5*in_v[ss];
-  });
-  return ret;
-}
-template<class vobj> 
-Lattice<vobj> spProj5m(const Lattice<vobj> & in)
-{
-  Gamma G5(Gamma::Algebra::Gamma5);
-  GridBase *grid=in.Grid();
-  Lattice<vobj> ret(grid);
-  auto ret_v = ret.View();
-  auto in_v  =  in.View();
-  thread_for(ss,grid->oSites(),{
-    ret_v[ss] = in_v[ss] - G5*in_v[ss];
-  });
-  return ret;
-}
-
-template <class Impl>
-void WilsonFermion5D<Impl>::ContractJ5q(FermionField &q_in,ComplexField &J5q)
-{
-  conformable(GaugeGrid(), J5q.Grid());
-  conformable(q_in.Grid(), FermionGrid());
-
-  // 4d field
-  int Ls = this->Ls;
-  FermionField psi(GaugeGrid());
-  FermionField p_plus (GaugeGrid());
-  FermionField p_minus(GaugeGrid());
-  FermionField p(GaugeGrid());
-
-  ExtractSlice(p_plus , q_in, Ls/2   , 0);
-  ExtractSlice(p_minus, q_in, Ls/2-1 , 0);
-  p_plus = spProj5p(p_plus );
-  p_minus= spProj5m(p_minus);
-  p=p_plus+p_minus;
-  J5q = localInnerProduct(p,p);
-}
-
-template <class Impl>
-void WilsonFermion5D<Impl>::ContractJ5q(PropagatorField &q_in,ComplexField &J5q)
-{
-  conformable(GaugeGrid(), J5q.Grid());
-  conformable(q_in.Grid(), FermionGrid());
-
-  // 4d field
-  int Ls = this->Ls;
-  PropagatorField psi(GaugeGrid());
-  PropagatorField p_plus (GaugeGrid());
-  PropagatorField p_minus(GaugeGrid());
-  PropagatorField p(GaugeGrid());
-
-  ExtractSlice(p_plus , q_in, Ls/2   , 0);
-  ExtractSlice(p_minus, q_in, Ls/2-1 , 0);
-  p_plus = spProj5p(p_plus );
-  p_minus= spProj5m(p_minus);
-  p=p_plus+p_minus;
-  J5q = localInnerProduct(p,p);
-}
-
-template <class Impl>
-void WilsonFermion5D<Impl>::ContractConservedCurrent(PropagatorField &q_in_1,
-                                                     PropagatorField &q_in_2,
-                                                     PropagatorField &q_out,
-                                                     Current curr_type,
-                                                     unsigned int mu)
-{
-    conformable(q_in_1.Grid(), FermionGrid());
-    conformable(q_in_1.Grid(), q_in_2.Grid());
-    conformable(_FourDimGrid, q_out.Grid());
-
-    PropagatorField tmp1(FermionGrid()), tmp2(FermionGrid());
-    unsigned int LLs = q_in_1.Grid()->_rdimensions[0];
-    q_out = Zero();
-
-    // Forward, need q1(x + mu, s), q2(x, Ls - 1 - s). Backward, need q1(x, s), 
-    // q2(x + mu, Ls - 1 - s). 5D lattice so shift 4D coordinate mu by one.
-    tmp1 = Cshift(q_in_1, mu + 1, 1);
-    tmp2 = Cshift(q_in_2, mu + 1, 1);
-    auto q_in_1_v = q_in_1.View();
-    auto q_in_2_v = q_in_2.View();
-    auto tmp1_v   = tmp1.View();
-    auto tmp2_v   = tmp2.View();
-    auto q_out_v  = q_out.View();
-    auto Umu_v    = Umu.View();
-    thread_for(sU, Umu.Grid()->oSites(),{
-
-        unsigned int sF1 = sU * LLs;
-        unsigned int sF2 = (sU + 1) * LLs - 1;
-
-        for (unsigned int s = 0; s < LLs; ++s)
-        {
-            bool axial_sign = ((curr_type == Current::Axial) && \
-                               (s < (LLs / 2)));
-            SitePropagator qSite2, qmuSite2;
-
-            // If vectorised in 5th dimension, reverse q2 vector to match up
-            // sites correctly.
-            if (Impl::LsVectorised)
-            {
-                REVERSE_LS(q_in_2_v[sF2], qSite2, Ls / LLs);
-                REVERSE_LS(tmp2_v[sF2], qmuSite2, Ls / LLs);
-            }
-            else
-            {
-                qSite2   = q_in_2_v[sF2];
-                qmuSite2 = tmp2_v[sF2];
-            }
-            Kernels::ContractConservedCurrentSiteFwd(tmp1_v[sF1], 
-                                                     qSite2, 
-                                                     q_out_v[sU],
-                                                     Umu_v, sU, mu, axial_sign);
-            Kernels::ContractConservedCurrentSiteBwd(q_in_1_v[sF1],
-                                                     qmuSite2,
-                                                     q_out_v[sU],
-                                                     Umu_v, sU, mu, axial_sign);
-            sF1++;
-            sF2--;
-        }
-    });
-}
+ ******************************************************************************/
 
 
-template <class Impl>
-void WilsonFermion5D<Impl>::SeqConservedCurrent(PropagatorField &q_in, 
-                                                PropagatorField &q_out,
-                                                Current curr_type, 
-                                                unsigned int mu,
-                                                unsigned int tmin, 
-                                                unsigned int tmax,
-						ComplexField &lattice_cmplx)
-{
-    conformable(q_in.Grid(), FermionGrid());
-    conformable(q_in.Grid(), q_out.Grid());
-    PropagatorField tmp(GaugeGrid()),tmp2(GaugeGrid());
-    unsigned int tshift = (mu == Tp) ? 1 : 0;
-    unsigned int LLs = q_in.Grid()->_rdimensions[0];
-    unsigned int LLt    = GridDefaultLatt()[Tp];
 
-    q_out = Zero();
-    LatticeInteger coords(_FourDimGrid);
-    LatticeCoordinate(coords, Tp);
-    
-    auto q_out_v = q_out.View();
-    auto tmp2_v  = tmp2.View();
-    auto coords_v= coords.View();
-    auto Umu_v   = Umu.View();
-    for (unsigned int s = 0; s < LLs; ++s)
-    {
-        bool axial_sign = ((curr_type == Current::Axial) && (s < (LLs / 2)));
-	bool tadpole_sign = (curr_type == Current::Tadpole);
-	bool switch_sgn = tadpole_sign || axial_sign;
-
-
-        //forward direction: Need q(x + mu, s)*A(x)
-        ExtractSlice(tmp2, q_in, s, 0);  //q(x,s) 
-        tmp = Cshift(tmp2, mu, 1);	 //q(x+mu,s)
-        tmp2 = tmp*lattice_cmplx;	 //q(x+mu,s)*A(x)	
-
-    	thread_for(sU, Umu.Grid()->oSites(),{
-            // Compute the sequential conserved current insertion only if our simd
-            // object contains a timeslice we need.
-            vPredicate t_mask;
-	    t_mask() = ((coords_v[sU] >= tmin) && (coords_v[sU] <= tmax));
-            Integer timeSlices = Reduce(t_mask());
-
-            if (timeSlices > 0)
-            {
-		unsigned int sF = sU * LLs + s;
-                Kernels::SeqConservedCurrentSiteFwd(tmp2_v[sU], 
-						    q_out_v[sF], Umu_v, sU,
-						    mu, t_mask, switch_sgn);
-            }
-
-        });
-
-        //backward direction: Need q(x - mu, s)*A(x-mu)
-        ExtractSlice(tmp2, q_in, s, 0);  //q(x,s)
-        tmp = lattice_cmplx*tmp2;	 //q(x,s)*A(x)
-        tmp2 = Cshift(tmp, mu, -1);	 //q(x-mu,s)*A(x-mu,s)
-
-    	thread_for(sU, Umu.Grid()->oSites(),
-    	{
-	  vPredicate t_mask;
-	  t_mask()= ((coords_v[sU] >= (tmin + tshift)) && (coords_v[sU] <= (tmax + tshift)));
-
-	  //if tmax = LLt-1 (last timeslice) include timeslice 0 if the time is shifted (mu=3)	
-	  unsigned int t0 = 0;
-	  if((tmax==LLt-1) && (tshift==1)) t_mask() = (t_mask() || (coords_v[sU] == t0 ));
-	  
-	  Integer timeSlices = Reduce(t_mask());
-	  
-	  if (timeSlices > 0) {
-	    unsigned int sF = sU * LLs + s; 
-	    Kernels::SeqConservedCurrentSiteBwd(tmp2_v[sU], 
-						q_out_v[sF], Umu_v, sU,
-						mu, t_mask, axial_sign);
-	  }
-	});
-    }
-}
   
 NAMESPACE_END(Grid);
 
diff --git a/Grid/qcd/action/fermion/implementation/WilsonFermionImplementation.h b/Grid/qcd/action/fermion/implementation/WilsonFermionImplementation.h
index cb852a63..4977ea68 100644
--- a/Grid/qcd/action/fermion/implementation/WilsonFermionImplementation.h
+++ b/Grid/qcd/action/fermion/implementation/WilsonFermionImplementation.h
@@ -67,7 +67,12 @@ WilsonFermion<Impl>::WilsonFermion(GaugeField &_Umu, GridCartesian &Fgrid,
     diag_mass = 4.0 + mass;
   }
 
-
+  int vol4;
+  vol4=Fgrid.oSites();
+  Stencil.BuildSurfaceList(1,vol4);
+  vol4=Hgrid.oSites();
+  StencilEven.BuildSurfaceList(1,vol4);
+  StencilOdd.BuildSurfaceList(1,vol4);
 }
 
 template<class Impl>
@@ -187,21 +192,24 @@ void WilsonFermion<Impl>::ImportGauge(const GaugeField &_Umu)
 /////////////////////////////
 
 template <class Impl>
-RealD WilsonFermion<Impl>::M(const FermionField &in, FermionField &out) {
+void WilsonFermion<Impl>::M(const FermionField &in, FermionField &out)
+{
   out.Checkerboard() = in.Checkerboard();
   Dhop(in, out, DaggerNo);
-  return axpy_norm(out, diag_mass, in, out);
+  axpy(out, diag_mass, in, out);
 }
 
 template <class Impl>
-RealD WilsonFermion<Impl>::Mdag(const FermionField &in, FermionField &out) {
+void WilsonFermion<Impl>::Mdag(const FermionField &in, FermionField &out)
+{
   out.Checkerboard() = in.Checkerboard();
   Dhop(in, out, DaggerYes);
-  return axpy_norm(out, diag_mass, in, out);
+  axpy(out, diag_mass, in, out);
 }
 
 template <class Impl>
-void WilsonFermion<Impl>::Meooe(const FermionField &in, FermionField &out) {
+void WilsonFermion<Impl>::Meooe(const FermionField &in, FermionField &out)
+{
   if (in.Checkerboard() == Odd) {
     DhopEO(in, out, DaggerNo);
   } else {
@@ -210,7 +218,8 @@ void WilsonFermion<Impl>::Meooe(const FermionField &in, FermionField &out) {
 }
 
 template <class Impl>
-void WilsonFermion<Impl>::MeooeDag(const FermionField &in, FermionField &out) {
+void WilsonFermion<Impl>::MeooeDag(const FermionField &in, FermionField &out)
+{
   if (in.Checkerboard() == Odd) {
     DhopEO(in, out, DaggerYes);
   } else {
@@ -219,26 +228,30 @@ void WilsonFermion<Impl>::MeooeDag(const FermionField &in, FermionField &out) {
 }
 
 template <class Impl>
-void WilsonFermion<Impl>::Mooee(const FermionField &in, FermionField &out) {
+void WilsonFermion<Impl>::Mooee(const FermionField &in, FermionField &out)
+{
   out.Checkerboard() = in.Checkerboard();
   typename FermionField::scalar_type scal(diag_mass);
   out = scal * in;
 }
 
 template <class Impl>
-void WilsonFermion<Impl>::MooeeDag(const FermionField &in, FermionField &out) {
+void WilsonFermion<Impl>::MooeeDag(const FermionField &in, FermionField &out)
+{
   out.Checkerboard() = in.Checkerboard();
   Mooee(in, out);
 }
 
 template<class Impl>
-void WilsonFermion<Impl>::MooeeInv(const FermionField &in, FermionField &out) {
+void WilsonFermion<Impl>::MooeeInv(const FermionField &in, FermionField &out)
+{
   out.Checkerboard() = in.Checkerboard();
   out = (1.0/(diag_mass))*in;
 }
 
 template<class Impl>
-void WilsonFermion<Impl>::MooeeInvDag(const FermionField &in, FermionField &out) {
+void WilsonFermion<Impl>::MooeeInvDag(const FermionField &in, FermionField &out)
+{
   out.Checkerboard() = in.Checkerboard();
   MooeeInv(in,out);
 }
@@ -341,7 +354,8 @@ void WilsonFermion<Impl>::DerivInternal(StencilImpl &st, DoubledGaugeField &U,
 }
 
 template <class Impl>
-void WilsonFermion<Impl>::DhopDeriv(GaugeField &mat, const FermionField &U, const FermionField &V, int dag) {
+void WilsonFermion<Impl>::DhopDeriv(GaugeField &mat, const FermionField &U, const FermionField &V, int dag)
+{
   conformable(U.Grid(), _grid);
   conformable(U.Grid(), V.Grid());
   conformable(U.Grid(), mat.Grid());
@@ -352,7 +366,8 @@ void WilsonFermion<Impl>::DhopDeriv(GaugeField &mat, const FermionField &U, cons
 }
 
 template <class Impl>
-void WilsonFermion<Impl>::DhopDerivOE(GaugeField &mat, const FermionField &U, const FermionField &V, int dag) {
+void WilsonFermion<Impl>::DhopDerivOE(GaugeField &mat, const FermionField &U, const FermionField &V, int dag)
+{
   conformable(U.Grid(), _cbgrid);
   conformable(U.Grid(), V.Grid());
   //conformable(U.Grid(), mat.Grid()); not general, leaving as a comment (Guido)
@@ -366,7 +381,8 @@ void WilsonFermion<Impl>::DhopDerivOE(GaugeField &mat, const FermionField &U, co
 }
 
 template <class Impl>
-void WilsonFermion<Impl>::DhopDerivEO(GaugeField &mat, const FermionField &U, const FermionField &V, int dag) {
+void WilsonFermion<Impl>::DhopDerivEO(GaugeField &mat, const FermionField &U, const FermionField &V, int dag)
+{
   conformable(U.Grid(), _cbgrid);
   conformable(U.Grid(), V.Grid());
   //conformable(U.Grid(), mat.Grid());
@@ -379,8 +395,8 @@ 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;
+void WilsonFermion<Impl>::Dhop(const FermionField &in, FermionField &out, int dag)
+{
   conformable(in.Grid(), _grid);  // verifies full grid
   conformable(in.Grid(), out.Grid());
 
@@ -390,8 +406,8 @@ 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+=1;
+void WilsonFermion<Impl>::DhopOE(const FermionField &in, FermionField &out, int dag)
+{
   conformable(in.Grid(), _cbgrid);    // verifies half grid
   conformable(in.Grid(), out.Grid());  // drops the cb check
 
@@ -402,8 +418,8 @@ 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+=1;
+void WilsonFermion<Impl>::DhopEO(const FermionField &in, FermionField &out,int dag)
+{
   conformable(in.Grid(), _cbgrid);    // verifies half grid
   conformable(in.Grid(), out.Grid());  // drops the cb check
 
@@ -482,7 +498,8 @@ template <class Impl>
 void WilsonFermion<Impl>::DhopInternalOverlappedComms(StencilImpl &st, LebesgueOrder &lo,
 						      DoubledGaugeField &U,
 						      const FermionField &in,
-						      FermionField &out, int dag) {
+						      FermionField &out, int dag)
+{
   assert((dag == DaggerNo) || (dag == DaggerYes));
 
   Compressor compressor(dag);
@@ -547,7 +564,8 @@ template <class Impl>
 void WilsonFermion<Impl>::DhopInternalSerial(StencilImpl &st, LebesgueOrder &lo,
                                        DoubledGaugeField &U,
                                        const FermionField &in,
-                                       FermionField &out, int dag) {
+                                       FermionField &out, int dag)
+{
   assert((dag == DaggerNo) || (dag == DaggerYes));
   Compressor compressor(dag);
   DhopCommTime-=usecond();
@@ -574,6 +592,7 @@ template <class Impl>
 void WilsonFermion<Impl>::ContractConservedCurrent(PropagatorField &q_in_1,
                                                    PropagatorField &q_in_2,
                                                    PropagatorField &q_out,
+                                                   PropagatorField &src,
                                                    Current curr_type,
                                                    unsigned int mu)
 {
@@ -581,35 +600,14 @@ void WilsonFermion<Impl>::ContractConservedCurrent(PropagatorField &q_in_1,
   conformable(_grid, q_in_1.Grid());
   conformable(_grid, q_in_2.Grid());
   conformable(_grid, q_out.Grid());
-  PropagatorField tmp1(_grid), tmp2(_grid);
-  q_out = Zero();
-
-  // Forward, need q1(x + mu), q2(x). Backward, need q1(x), q2(x + mu).
-  // Inefficient comms method but not performance critical.
-  tmp1 = Cshift(q_in_1, mu, 1);
-  tmp2 = Cshift(q_in_2, mu, 1);
-  auto tmp1_v  =  tmp1.View();
-  auto tmp2_v  =  tmp2.View();
-  auto q_in_1_v=q_in_1.View();
-  auto q_in_2_v=q_in_2.View();
-  auto q_out_v = q_out.View();
-  auto Umu_v   =   Umu.View();
-  thread_for(sU, Umu.Grid()->oSites(),{
-      Kernels::ContractConservedCurrentSiteFwd(tmp1_v[sU],
-					       q_in_2_v[sU],
-					       q_out_v[sU],
-					       Umu_v, sU, mu);
-      Kernels::ContractConservedCurrentSiteBwd(q_in_1_v[sU],
-					       tmp2_v[sU],
-					       q_out_v[sU],
-					       Umu_v, sU, mu);
-  });
+  assert(0);
 }
 
 
 template <class Impl>
 void WilsonFermion<Impl>::SeqConservedCurrent(PropagatorField &q_in,
                                               PropagatorField &q_out,
+                                              PropagatorField &src,
                                               Current curr_type,
                                               unsigned int mu,
                                               unsigned int tmin,
@@ -618,59 +616,7 @@ void WilsonFermion<Impl>::SeqConservedCurrent(PropagatorField &q_in,
 {
   conformable(_grid, q_in.Grid());
   conformable(_grid, q_out.Grid());
-
-  //  Lattice<iSinglet<Simd>> ph(_grid), coor(_grid);
-  Complex i(0.0,1.0);
-  PropagatorField tmpFwd(_grid), tmpBwd(_grid), tmp(_grid);
-  unsigned int tshift = (mu == Tp) ? 1 : 0;
-  unsigned int LLt    = GridDefaultLatt()[Tp];
-
-  q_out = Zero();
-  LatticeInteger coords(_grid);
-  LatticeCoordinate(coords, Tp);
-
-  // Need q(x + mu) and q(x - mu).
-  tmp    = Cshift(q_in, mu, 1);
-  tmpFwd = tmp*lattice_cmplx;
-  tmp    = lattice_cmplx*q_in;
-  tmpBwd = Cshift(tmp, mu, -1);
-
-  auto coords_v = coords.View();
-  auto tmpFwd_v = tmpFwd.View();
-  auto tmpBwd_v = tmpBwd.View();
-  auto Umu_v    = Umu.View();
-  auto q_out_v  = q_out.View();
-
-  thread_for(sU, Umu.Grid()->oSites(), {
-
-    // Compute the sequential conserved current insertion only if our simd
-    // object contains a timeslice we need.
-    vPredicate t_mask;
-    t_mask() = ((coords_v[sU] >= tmin) && (coords_v[sU] <= tmax));
-    Integer timeSlices = Reduce(t_mask());
-
-    if (timeSlices > 0) {
-      Kernels::SeqConservedCurrentSiteFwd(tmpFwd_v[sU],
-					  q_out_v[sU],
-					  Umu_v, sU, mu, t_mask);
-    }
-
-    // Repeat for backward direction.
-    t_mask()     = ((coords_v[sU] >= (tmin + tshift)) &&
-		    (coords_v[sU] <= (tmax + tshift)));
-
-    //if tmax = LLt-1 (last timeslice) include timeslice 0 if the time is shifted (mu=3)
-    unsigned int t0 = 0;
-    if((tmax==LLt-1) && (tshift==1)) t_mask() = (t_mask() || (coords_v[sU] == t0 ));
-
-    timeSlices = Reduce(t_mask());
-
-    if (timeSlices > 0) {
-      Kernels::SeqConservedCurrentSiteBwd(tmpBwd_v[sU],
-					  q_out_v[sU],
-					  Umu_v, sU, mu, t_mask);
-    }
-  });
+  assert(0);
 }
 
 NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h b/Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h
index 81216e03..c2b62416 100644
--- a/Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h
+++ b/Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h
@@ -39,19 +39,21 @@ NAMESPACE_BEGIN(Grid);
 // Generic implementation; move to different file?
 ////////////////////////////////////////////
 
+/*
 accelerator_inline void get_stencil(StencilEntry * mem, StencilEntry &chip)
 {
-#ifdef __CUDA_ARCH__
-  static_assert(sizeof(StencilEntry)==sizeof(uint4),"Unexpected Stencil Entry Size"); 
+#ifdef GRID_SIMT
+  static_assert(sizeof(StencilEntry)==sizeof(uint4),"Unexpected Stencil Entry Size");
   uint4 * mem_pun  = (uint4 *)mem; // force 128 bit loads
   uint4 * chip_pun = (uint4 *)&chip;
   * chip_pun = * mem_pun;
-#else 
+#else
   chip = *mem;
 #endif
   return;
 }
-  
+*/
+
 #define GENERIC_STENCIL_LEG(Dir,spProj,Recon)			\
   SE = st.GetEntry(ptype, Dir, sF);				\
   if (SE->_is_local) {						\
@@ -61,10 +63,10 @@ accelerator_inline void get_stencil(StencilEntry * mem, StencilEntry &chip)
   } else {							\
     chi = coalescedRead(buf[SE->_offset],lane);			\
   }								\
-  synchronise();						\
+  acceleratorSynchronise();						\
   Impl::multLink(Uchi, U[sU], chi, Dir, SE, st);		\
   Recon(result, Uchi);
-  
+
 #define GENERIC_STENCIL_LEG_INT(Dir,spProj,Recon)		\
   SE = st.GetEntry(ptype, Dir, sF);				\
   if (SE->_is_local) {						\
@@ -74,12 +76,12 @@ accelerator_inline void get_stencil(StencilEntry * mem, StencilEntry &chip)
   } else if ( st.same_node[Dir] ) {				\
     chi = coalescedRead(buf[SE->_offset],lane);			\
   }								\
-  synchronise();						\
+  acceleratorSynchronise();						\
   if (SE->_is_local || st.same_node[Dir] ) {			\
     Impl::multLink(Uchi, U[sU], chi, Dir, SE, st);		\
     Recon(result, Uchi);					\
   }								\
-  synchronise();						
+  acceleratorSynchronise();
 
 #define GENERIC_STENCIL_LEG_EXT(Dir,spProj,Recon)		\
   SE = st.GetEntry(ptype, Dir, sF);				\
@@ -89,7 +91,7 @@ accelerator_inline void get_stencil(StencilEntry * mem, StencilEntry &chip)
     Recon(result, Uchi);					\
     nmu++;							\
   }								\
-  synchronise();						
+  acceleratorSynchronise();
 
 #define GENERIC_DHOPDIR_LEG_BODY(Dir,spProj,Recon)		\
     if (SE->_is_local ) {					\
@@ -99,9 +101,9 @@ accelerator_inline void get_stencil(StencilEntry * mem, StencilEntry &chip)
     } else {							\
       chi = coalescedRead(buf[SE->_offset],lane);		\
     }								\
-    synchronise();						\
+    acceleratorSynchronise();					\
     Impl::multLink(Uchi, U[sU], chi, dir, SE, st);		\
-    Recon(result, Uchi);					
+    Recon(result, Uchi);
 
 #define GENERIC_DHOPDIR_LEG(Dir,spProj,Recon)			\
   if (gamma == Dir) {						\
@@ -126,7 +128,7 @@ void WilsonKernels<Impl>::GenericDhopSiteDag(StencilView &st, DoubledGaugeFieldV
   StencilEntry *SE;
   int ptype;
   const int Nsimd = SiteHalfSpinor::Nsimd();
-  const int lane=SIMTlane(Nsimd);
+  const int lane=acceleratorSIMTlane(Nsimd);
   GENERIC_STENCIL_LEG(Xp,spProjXp,spReconXp);
   GENERIC_STENCIL_LEG(Yp,spProjYp,accumReconYp);
   GENERIC_STENCIL_LEG(Zp,spProjZp,accumReconZp);
@@ -141,7 +143,7 @@ void WilsonKernels<Impl>::GenericDhopSiteDag(StencilView &st, DoubledGaugeFieldV
 template <class Impl>
 void WilsonKernels<Impl>::GenericDhopSite(StencilView &st, DoubledGaugeFieldView &U,
 					  SiteHalfSpinor *buf, int sF,
-					  int sU, const FermionFieldView &in, FermionFieldView &out) 
+					  int sU, const FermionFieldView &in, FermionFieldView &out)
 {
   typedef decltype(coalescedRead(buf[0])) calcHalfSpinor;
   typedef decltype(coalescedRead(in[0]))  calcSpinor;
@@ -153,7 +155,7 @@ void WilsonKernels<Impl>::GenericDhopSite(StencilView &st, DoubledGaugeFieldView
   int ptype;
 
   const int Nsimd = SiteHalfSpinor::Nsimd();
-  const int lane=SIMTlane(Nsimd);
+  const int lane=acceleratorSIMTlane(Nsimd);
   GENERIC_STENCIL_LEG(Xm,spProjXp,spReconXp);
   GENERIC_STENCIL_LEG(Ym,spProjYp,accumReconYp);
   GENERIC_STENCIL_LEG(Zm,spProjZp,accumReconZp);
@@ -181,7 +183,7 @@ void WilsonKernels<Impl>::GenericDhopSiteDagInt(StencilView &st,  DoubledGaugeFi
   StencilEntry *SE;
   int ptype;
   const int Nsimd = SiteHalfSpinor::Nsimd();
-  const int lane=SIMTlane(Nsimd);
+  const int lane=acceleratorSIMTlane(Nsimd);
 
   result=Zero();
   GENERIC_STENCIL_LEG_INT(Xp,spProjXp,accumReconXp);
@@ -198,12 +200,12 @@ void WilsonKernels<Impl>::GenericDhopSiteDagInt(StencilView &st,  DoubledGaugeFi
 template <class Impl>
 void WilsonKernels<Impl>::GenericDhopSiteInt(StencilView &st,  DoubledGaugeFieldView &U,
 							 SiteHalfSpinor *buf, int sF,
-							 int sU, const FermionFieldView &in, FermionFieldView &out) 
+							 int sU, const FermionFieldView &in, FermionFieldView &out)
 {
   typedef decltype(coalescedRead(buf[0])) calcHalfSpinor;
   typedef decltype(coalescedRead(in[0]))  calcSpinor;
   const int Nsimd = SiteHalfSpinor::Nsimd();
-  const int lane=SIMTlane(Nsimd);
+  const int lane=acceleratorSIMTlane(Nsimd);
 
   calcHalfSpinor chi;
   //  calcHalfSpinor *chi_p;
@@ -239,7 +241,7 @@ void WilsonKernels<Impl>::GenericDhopSiteDagExt(StencilView &st,  DoubledGaugeFi
   int ptype;
   int nmu=0;
   const int Nsimd = SiteHalfSpinor::Nsimd();
-  const int lane=SIMTlane(Nsimd);
+  const int lane=acceleratorSIMTlane(Nsimd);
   result=Zero();
   GENERIC_STENCIL_LEG_EXT(Xp,spProjXp,accumReconXp);
   GENERIC_STENCIL_LEG_EXT(Yp,spProjYp,accumReconYp);
@@ -249,7 +251,7 @@ void WilsonKernels<Impl>::GenericDhopSiteDagExt(StencilView &st,  DoubledGaugeFi
   GENERIC_STENCIL_LEG_EXT(Ym,spProjYm,accumReconYm);
   GENERIC_STENCIL_LEG_EXT(Zm,spProjZm,accumReconZm);
   GENERIC_STENCIL_LEG_EXT(Tm,spProjTm,accumReconTm);
-  if ( nmu ) { 
+  if ( nmu ) {
     auto out_t = coalescedRead(out[sF],lane);
     out_t = out_t + result;
     coalescedWrite(out[sF],out_t,lane);
@@ -259,7 +261,7 @@ void WilsonKernels<Impl>::GenericDhopSiteDagExt(StencilView &st,  DoubledGaugeFi
 template <class Impl>
 void WilsonKernels<Impl>::GenericDhopSiteExt(StencilView &st,  DoubledGaugeFieldView &U,
 					     SiteHalfSpinor *buf, int sF,
-					     int sU, const FermionFieldView &in, FermionFieldView &out) 
+					     int sU, const FermionFieldView &in, FermionFieldView &out)
 {
   typedef decltype(coalescedRead(buf[0])) calcHalfSpinor;
   typedef decltype(coalescedRead(in[0]))  calcSpinor;
@@ -270,7 +272,7 @@ void WilsonKernels<Impl>::GenericDhopSiteExt(StencilView &st,  DoubledGaugeField
   int ptype;
   int nmu=0;
   const int Nsimd = SiteHalfSpinor::Nsimd();
-  const int lane=SIMTlane(Nsimd);
+  const int lane=acceleratorSIMTlane(Nsimd);
   result=Zero();
   GENERIC_STENCIL_LEG_EXT(Xm,spProjXp,accumReconXp);
   GENERIC_STENCIL_LEG_EXT(Ym,spProjYp,accumReconYp);
@@ -280,7 +282,7 @@ void WilsonKernels<Impl>::GenericDhopSiteExt(StencilView &st,  DoubledGaugeField
   GENERIC_STENCIL_LEG_EXT(Yp,spProjYm,accumReconYm);
   GENERIC_STENCIL_LEG_EXT(Zp,spProjZm,accumReconZm);
   GENERIC_STENCIL_LEG_EXT(Tp,spProjTm,accumReconTm);
-  if ( nmu ) { 
+  if ( nmu ) {
     auto out_t = coalescedRead(out[sF],lane);
     out_t = out_t + result;
     coalescedWrite(out[sF],out_t,lane);
@@ -300,12 +302,12 @@ void WilsonKernels<Impl>::GenericDhopSiteExt(StencilView &st,  DoubledGaugeField
   StencilEntry *SE;							\
   int ptype;								\
   const int Nsimd = SiteHalfSpinor::Nsimd();				\
-  const int lane=SIMTlane(Nsimd);					\
+  const int lane=acceleratorSIMTlane(Nsimd);					\
 									\
   SE = st.GetEntry(ptype, dir, sF);					\
   GENERIC_DHOPDIR_LEG_BODY(Dir,spProj,spRecon);				\
   coalescedWrite(out[sF], result,lane);					\
-  }									
+  }
 
 DhopDirMacro(Xp,spProjXp,spReconXp);
 DhopDirMacro(Yp,spProjYp,spReconYp);
@@ -316,9 +318,9 @@ DhopDirMacro(Ym,spProjYm,spReconYm);
 DhopDirMacro(Zm,spProjZm,spReconZm);
 DhopDirMacro(Tm,spProjTm,spReconTm);
 
-template <class Impl> 
+template <class Impl>
 void WilsonKernels<Impl>::DhopDirK( StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf, int sF,
-				    int sU, const FermionFieldView &in, FermionFieldView &out, int dir, int gamma) 
+				    int sU, const FermionFieldView &in, FermionFieldView &out, int dir, int gamma)
 {
   typedef decltype(coalescedRead(buf[0])) calcHalfSpinor;
   typedef decltype(coalescedRead(in[0]))  calcSpinor;
@@ -328,7 +330,7 @@ void WilsonKernels<Impl>::DhopDirK( StencilView &st, DoubledGaugeFieldView &U,Si
   StencilEntry *SE;
   int ptype;
   const int Nsimd = SiteHalfSpinor::Nsimd();
-  const int lane=SIMTlane(Nsimd);
+  const int lane=acceleratorSIMTlane(Nsimd);
 
   SE = st.GetEntry(ptype, dir, sF);
   GENERIC_DHOPDIR_LEG(Xp,spProjXp,spReconXp);
@@ -344,54 +346,55 @@ void WilsonKernels<Impl>::DhopDirK( StencilView &st, DoubledGaugeFieldView &U,Si
 
 template <class Impl>
 void WilsonKernels<Impl>::DhopDirAll( StencilImpl &st, DoubledGaugeField &U,SiteHalfSpinor *buf, int Ls,
-				      int Nsite, const FermionField &in, std::vector<FermionField> &out) 
+				      int Nsite, const FermionField &in, std::vector<FermionField> &out)
 {
-   auto U_v   = U.View();
-   auto in_v  = in.View();
-   auto st_v  = st.View();
+   autoView(U_v  ,U,AcceleratorRead);
+   autoView(in_v ,in,AcceleratorRead);
+   autoView(st_v ,st,AcceleratorRead);
 
-   auto out_Xm = out[0].View();
-   auto out_Ym = out[1].View();
-   auto out_Zm = out[2].View();
-   auto out_Tm = out[3].View();
-   auto out_Xp = out[4].View();
-   auto out_Yp = out[5].View();
-   auto out_Zp = out[6].View();
-   auto out_Tp = out[7].View();
-
-   accelerator_forNB(sss,Nsite*Ls,Simd::Nsimd(),{
-      int sU=sss/Ls;				
-      int sF =sss;				
-      DhopDirXm(st_v,U_v,st.CommBuf(),sF,sU,in_v,out_Xm,0);
-      DhopDirYm(st_v,U_v,st.CommBuf(),sF,sU,in_v,out_Ym,1);
-      DhopDirZm(st_v,U_v,st.CommBuf(),sF,sU,in_v,out_Zm,2);
-      DhopDirTm(st_v,U_v,st.CommBuf(),sF,sU,in_v,out_Tm,3);
-      DhopDirXp(st_v,U_v,st.CommBuf(),sF,sU,in_v,out_Xp,4);
-      DhopDirYp(st_v,U_v,st.CommBuf(),sF,sU,in_v,out_Yp,5);
-      DhopDirZp(st_v,U_v,st.CommBuf(),sF,sU,in_v,out_Zp,6);
-      DhopDirTp(st_v,U_v,st.CommBuf(),sF,sU,in_v,out_Tp,7);
+   autoView(out_Xm,out[0],AcceleratorWrite);
+   autoView(out_Ym,out[1],AcceleratorWrite);
+   autoView(out_Zm,out[2],AcceleratorWrite);
+   autoView(out_Tm,out[3],AcceleratorWrite);
+   autoView(out_Xp,out[4],AcceleratorWrite);
+   autoView(out_Yp,out[5],AcceleratorWrite);
+   autoView(out_Zp,out[6],AcceleratorWrite);
+   autoView(out_Tp,out[7],AcceleratorWrite);
+   auto CBp=st.CommBuf();
+   accelerator_for(sss,Nsite*Ls,Simd::Nsimd(),{
+      int sU=sss/Ls;
+      int sF =sss;
+      DhopDirXm(st_v,U_v,CBp,sF,sU,in_v,out_Xm,0);
+      DhopDirYm(st_v,U_v,CBp,sF,sU,in_v,out_Ym,1);
+      DhopDirZm(st_v,U_v,CBp,sF,sU,in_v,out_Zm,2);
+      DhopDirTm(st_v,U_v,CBp,sF,sU,in_v,out_Tm,3);
+      DhopDirXp(st_v,U_v,CBp,sF,sU,in_v,out_Xp,4);
+      DhopDirYp(st_v,U_v,CBp,sF,sU,in_v,out_Yp,5);
+      DhopDirZp(st_v,U_v,CBp,sF,sU,in_v,out_Zp,6);
+      DhopDirTp(st_v,U_v,CBp,sF,sU,in_v,out_Tp,7);
    });
 }
 
 
 template <class Impl>
 void WilsonKernels<Impl>::DhopDirKernel( StencilImpl &st, DoubledGaugeField &U,SiteHalfSpinor *buf, int Ls,
-					 int Nsite, const FermionField &in, FermionField &out, int dirdisp, int gamma) 
+					 int Nsite, const FermionField &in, FermionField &out, int dirdisp, int gamma)
 {
   assert(dirdisp<=7);
   assert(dirdisp>=0);
 
-   auto U_v   = U.View();
-   auto in_v  = in.View();
-   auto out_v = out.View();
-   auto st_v  = st.View();
+   autoView(U_v  ,U  ,AcceleratorRead);
+   autoView(in_v ,in ,AcceleratorRead);
+   autoView(out_v,out,AcceleratorWrite);
+   autoView(st_v ,st ,AcceleratorRead);
+   auto CBp=st.CommBuf();
 #define LoopBody(Dir)				\
-   case Dir :			\
-     accelerator_forNB(ss,Nsite,Simd::Nsimd(),{	\
+   case Dir :					\
+     accelerator_for(ss,Nsite,Simd::Nsimd(),{	\
        for(int s=0;s<Ls;s++){			\
 	 int sU=ss;				\
 	 int sF = s+Ls*sU;						\
-	 DhopDir##Dir(st_v,U_v,st.CommBuf(),sF,sU,in_v,out_v,dirdisp);\
+	 DhopDir##Dir(st_v,U_v,CBp,sF,sU,in_v,out_v,dirdisp);\
        }							       \
        });							       \
      break;
@@ -411,7 +414,7 @@ void WilsonKernels<Impl>::DhopDirKernel( StencilImpl &st, DoubledGaugeField &U,S
      break;
    }
 #undef LoopBody
-} 
+}
 
 #define KERNEL_CALLNB(A) \
   const uint64_t    NN = Nsite*Ls;					\
@@ -421,7 +424,7 @@ void WilsonKernels<Impl>::DhopDirKernel( StencilImpl &st, DoubledGaugeField &U,S
       WilsonKernels<Impl>::A(st_v,U_v,buf,sF,sU,in_v,out_v);		\
   });
 
-#define KERNEL_CALL(A) KERNEL_CALLNB(A); accelerator_barrier(); 
+#define KERNEL_CALL(A) KERNEL_CALLNB(A); accelerator_barrier();
 
 #define ASM_CALL(A)							\
   thread_for( ss, Nsite, {						\
@@ -433,30 +436,30 @@ void WilsonKernels<Impl>::DhopDirKernel( StencilImpl &st, DoubledGaugeField &U,S
 template <class Impl>
 void WilsonKernels<Impl>::DhopKernel(int Opt,StencilImpl &st,  DoubledGaugeField &U, SiteHalfSpinor * buf,
 				     int Ls, int Nsite, const FermionField &in, FermionField &out,
-				     int interior,int exterior) 
+				     int interior,int exterior)
 {
-    auto U_v   =   U.View();
-    auto in_v  =  in.View();
-    auto out_v = out.View();
-    auto st_v  =  st.View();
+    autoView(U_v  ,  U,AcceleratorRead);
+    autoView(in_v , in,AcceleratorRead);
+    autoView(out_v,out,AcceleratorWrite);
+    autoView(st_v , st,AcceleratorRead);
 
-   if( interior && exterior ) { 
+   if( interior && exterior ) {
      if (Opt == WilsonKernelsStatic::OptGeneric    ) { KERNEL_CALL(GenericDhopSite); return;}
-#ifndef GRID_NVCC
+#ifndef GRID_CUDA
      if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSite);    return;}
-     if (Opt == WilsonKernelsStatic::OptInlineAsm  ) {  ASM_CALL(AsmDhopSite); /* printf("."); */   return;}
+     if (Opt == WilsonKernelsStatic::OptInlineAsm  ) {  ASM_CALL(AsmDhopSite);    return;}
 #endif
    } else if( interior ) {
      if (Opt == WilsonKernelsStatic::OptGeneric    ) { KERNEL_CALLNB(GenericDhopSiteInt); return;}
-#ifndef GRID_NVCC
+#ifndef GRID_CUDA
      if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALLNB(HandDhopSiteInt);    return;}
-     if (Opt == WilsonKernelsStatic::OptInlineAsm  ) {  ASM_CALL(AsmDhopSiteInt); /* printf("-"); */   return;}
+     if (Opt == WilsonKernelsStatic::OptInlineAsm  ) {  ASM_CALL(AsmDhopSiteInt);    return;}
 #endif
-   } else if( exterior ) { 
+   } else if( exterior ) {
      if (Opt == WilsonKernelsStatic::OptGeneric    ) { KERNEL_CALL(GenericDhopSiteExt); return;}
-#ifndef GRID_NVCC
+#ifndef GRID_CUDA
      if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSiteExt);    return;}
-     if (Opt == WilsonKernelsStatic::OptInlineAsm  ) {  ASM_CALL(AsmDhopSiteExt); /* printf("+"); */   return;}
+     if (Opt == WilsonKernelsStatic::OptInlineAsm  ) {  ASM_CALL(AsmDhopSiteExt);    return;}
 #endif
    }
    assert(0 && " Kernel optimisation case not covered ");
@@ -464,28 +467,28 @@ void WilsonKernels<Impl>::DhopKernel(int Opt,StencilImpl &st,  DoubledGaugeField
   template <class Impl>
   void WilsonKernels<Impl>::DhopDagKernel(int Opt,StencilImpl &st,  DoubledGaugeField &U, SiteHalfSpinor * buf,
 					  int Ls, int Nsite, const FermionField &in, FermionField &out,
-					  int interior,int exterior) 
+					  int interior,int exterior)
   {
-    auto U_v   = U.View();
-    auto in_v  = in.View();
-    auto out_v = out.View();
-    auto st_v  = st.View();
+    autoView(U_v  ,U,AcceleratorRead);
+    autoView(in_v ,in,AcceleratorRead);
+    autoView(out_v,out,AcceleratorWrite);
+    autoView(st_v ,st,AcceleratorRead);
 
-   if( interior && exterior ) { 
+   if( interior && exterior ) {
      if (Opt == WilsonKernelsStatic::OptGeneric    ) { KERNEL_CALL(GenericDhopSiteDag); return;}
-#ifndef GRID_NVCC
+#ifndef GRID_CUDA
      if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSiteDag);    return;}
      if (Opt == WilsonKernelsStatic::OptInlineAsm  ) {  ASM_CALL(AsmDhopSiteDag);     return;}
 #endif
    } else if( interior ) {
      if (Opt == WilsonKernelsStatic::OptGeneric    ) { KERNEL_CALL(GenericDhopSiteDagInt); return;}
-#ifndef GRID_NVCC
+#ifndef GRID_CUDA
      if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSiteDagInt);    return;}
      if (Opt == WilsonKernelsStatic::OptInlineAsm  ) {  ASM_CALL(AsmDhopSiteDagInt);     return;}
 #endif
-   } else if( exterior ) { 
+   } else if( exterior ) {
      if (Opt == WilsonKernelsStatic::OptGeneric    ) { KERNEL_CALL(GenericDhopSiteDagExt); return;}
-#ifndef GRID_NVCC
+#ifndef GRID_CUDA
      if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSiteDagExt);    return;}
      if (Opt == WilsonKernelsStatic::OptInlineAsm  ) {  ASM_CALL(AsmDhopSiteDagExt);     return;}
 #endif
@@ -493,131 +496,8 @@ void WilsonKernels<Impl>::DhopKernel(int Opt,StencilImpl &st,  DoubledGaugeField
    assert(0 && " Kernel optimisation case not covered ");
   }
 
-/*******************************************************************************
- * Conserved current utilities for Wilson fermions, for contracting propagators
- * to make a conserved current sink or inserting the conserved current 
- * sequentially. Common to both 4D and 5D.
- ******************************************************************************/
-// N.B. Functions below assume a -1/2 factor within U.
-#define WilsonCurrentFwd(expr, mu) ((expr - Gamma::gmu[mu]*expr))
-#define WilsonCurrentBwd(expr, mu) ((expr + Gamma::gmu[mu]*expr))
-
-/*******************************************************************************
- * Name: ContractConservedCurrentSiteFwd
- * Operation: (1/2) * q2[x] * U(x) * (g[mu] - 1) * q1[x + mu]
- * Notes: - DoubledGaugeField U assumed to contain -1/2 factor.
- *        - Pass in q_in_1 shifted in +ve mu direction.
- ******************************************************************************/
-template<class Impl>
-void WilsonKernels<Impl>::ContractConservedCurrentSiteFwd(const SitePropagator &q_in_1,
-							  const SitePropagator &q_in_2,
-							  SitePropagator &q_out,
-							  DoubledGaugeFieldView &U,
-							  unsigned int sU,
-							  unsigned int mu,
-							  bool switch_sign)
-{
-  SitePropagator result, tmp;
-  Gamma g5(Gamma::Algebra::Gamma5);
-
-  Impl::multLink(tmp, U[sU], q_in_1, mu);
-
-  result = g5 * adj(q_in_2) * g5 * WilsonCurrentFwd(tmp, mu);
-
-  if (switch_sign) {
-    q_out -= result;
-  } else {
-    q_out += result;
-  }
-}
-
-/*******************************************************************************
- * Name: ContractConservedCurrentSiteBwd
- * Operation: (1/2) * q2[x + mu] * U^dag(x) * (g[mu] + 1) * q1[x]
- * Notes: - DoubledGaugeField U assumed to contain -1/2 factor.
- *        - Pass in q_in_2 shifted in +ve mu direction.
- ******************************************************************************/
-template<class Impl>
-void WilsonKernels<Impl>::ContractConservedCurrentSiteBwd(const SitePropagator &q_in_1,
-							  const SitePropagator &q_in_2,
-							  SitePropagator &q_out,
-							  DoubledGaugeFieldView &U,
-							  unsigned int sU,
-							  unsigned int mu,
-							  bool switch_sign)
-{
-  SitePropagator result, tmp;
-  Gamma g5(Gamma::Algebra::Gamma5);
-
-  Impl::multLink(tmp, U[sU], q_in_1, mu + Nd);
-
-  result = g5 * adj(q_in_2) * g5 * WilsonCurrentBwd(tmp, mu);
-  if (switch_sign) {
-    q_out += result;
-  } else {
-    q_out -= result;
-  }
-}
-
-/*******************************************************************************
- * Name: SeqConservedCurrentSiteFwd
- * Operation: (1/2) * U(x) * (g[mu] - 1) * q[x + mu]
- * Notes: - DoubledGaugeField U assumed to contain -1/2 factor.
- *        - Pass in q_in shifted in +ve mu direction.
- ******************************************************************************/
-template<class Impl>
-void WilsonKernels<Impl>::SeqConservedCurrentSiteFwd(const SitePropagator &q_in,
-                                                     SitePropagator &q_out,
-                                                     DoubledGaugeFieldView &U,
-                                                     unsigned int sU,
-                                                     unsigned int mu,
-                                                     vPredicate t_mask,
-                                                     bool switch_sign)
-{
-  SitePropagator result;
-  
-  Impl::multLink(result, U[sU], q_in, mu);
-  result = WilsonCurrentFwd(result, mu);
-
-  // Zero any unwanted timeslice entries.
-  result = predicatedWhere(t_mask, result, 0.*result);
-  
-  if (switch_sign) {
-    q_out -= result;
-  } else {
-    q_out += result;
-  }
-}
-
-/*******************************************************************************
- * Name: SeqConservedCurrentSiteFwd
- * Operation: (1/2) * U^dag(x) * (g[mu] + 1) * q[x - mu]
- * Notes: - DoubledGaugeField U assumed to contain -1/2 factor.
- *        - Pass in q_in shifted in -ve mu direction.
- ******************************************************************************/
-template<class Impl>
-void WilsonKernels<Impl>::SeqConservedCurrentSiteBwd(const SitePropagator &q_in, 
-                                                     SitePropagator &q_out,
-                                                     DoubledGaugeFieldView &U,
-                                                     unsigned int sU,
-                                                     unsigned int mu,
-                                                     vPredicate t_mask,
-                                                     bool switch_sign)
-{
-  SitePropagator result;
-  Impl::multLink(result, U[sU], q_in, mu + Nd);
-  result = WilsonCurrentBwd(result, mu);
-
-  // Zero any unwanted timeslice entries.
-  result = predicatedWhere(t_mask, result, 0.*result);
-  
-  if (switch_sign) {
-    q_out += result;
-  } else {
-    q_out -= result;
-  }
-}
-
+#undef KERNEL_CALLNB
+#undef KERNEL_CALL
+#undef ASM_CALL
 
 NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/CayleyFermion5DInstantiationGparityWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/CayleyFermion5DInstantiationGparityWilsonImplD.cc
deleted file mode 100644
index 5130db9c..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/CayleyFermion5DInstantiationGparityWilsonImplD.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/CayleyFermion5DImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dcache.h>
-
-			   //#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dvec.h>
-			   //#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dgpu.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class CayleyFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/CayleyFermion5DInstantiationGparityWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/CayleyFermion5DInstantiationGparityWilsonImplD.cc
new file mode 120000
index 00000000..cb1db625
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/CayleyFermion5DInstantiationGparityWilsonImplD.cc
@@ -0,0 +1 @@
+../CayleyFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/ContinuedFractionFermion5DInstantiationGparityWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/ContinuedFractionFermion5DInstantiationGparityWilsonImplD.cc
deleted file mode 100644
index ca0d6cea..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/ContinuedFractionFermion5DInstantiationGparityWilsonImplD.cc
+++ /dev/null
@@ -1,38 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/ContinuedFractionFermion5D.cc
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/ContinuedFractionFermion5D.h>
-#include <Grid/qcd/action/fermion/implementation/ContinuedFractionFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class ContinuedFractionFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/ContinuedFractionFermion5DInstantiationGparityWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/ContinuedFractionFermion5DInstantiationGparityWilsonImplD.cc
new file mode 120000
index 00000000..c2d4b8fc
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/ContinuedFractionFermion5DInstantiationGparityWilsonImplD.cc
@@ -0,0 +1 @@
+../ContinuedFractionFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/DomainWallEOFAFermionInstantiationGparityWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/DomainWallEOFAFermionInstantiationGparityWilsonImplD.cc
deleted file mode 100644
index f7198131..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/DomainWallEOFAFermionInstantiationGparityWilsonImplD.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/DomainWallEOFAFermion.cc
-
-Copyright (C) 2017
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: David Murphy <dmurphy@phys.columbia.edu>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-			   /*  END LEGAL */
-
-#include <Grid/Grid_Eigen_Dense.h>
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/DomainWallEOFAFermion.h>
-#include <Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionCache.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class DomainWallEOFAFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/DomainWallEOFAFermionInstantiationGparityWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/DomainWallEOFAFermionInstantiationGparityWilsonImplD.cc
new file mode 120000
index 00000000..2f550a2b
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/DomainWallEOFAFermionInstantiationGparityWilsonImplD.cc
@@ -0,0 +1 @@
+../DomainWallEOFAFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/MobiusEOFAFermionInstantiationGparityWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/MobiusEOFAFermionInstantiationGparityWilsonImplD.cc
deleted file mode 100644
index ce7eaac9..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/MobiusEOFAFermionInstantiationGparityWilsonImplD.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/MobiusEOFAFermion.cc
-
-Copyright (C) 2017
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: David Murphy <dmurphy@phys.columbia.edu>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-			   /*  END LEGAL */
-
-#include <Grid/Grid_Eigen_Dense.h>
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/MobiusEOFAFermion.h>
-#include <Grid/qcd/action/fermion/implementation/MobiusEOFAFermionImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/MobiusEOFAFermionCache.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class MobiusEOFAFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/MobiusEOFAFermionInstantiationGparityWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/MobiusEOFAFermionInstantiationGparityWilsonImplD.cc
new file mode 120000
index 00000000..7a8f1172
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/MobiusEOFAFermionInstantiationGparityWilsonImplD.cc
@@ -0,0 +1 @@
+../MobiusEOFAFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/PartialFractionFermion5DInstantiationGparityWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/PartialFractionFermion5DInstantiationGparityWilsonImplD.cc
deleted file mode 100644
index 757719b6..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/PartialFractionFermion5DInstantiationGparityWilsonImplD.cc
+++ /dev/null
@@ -1,39 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/PartialFractionFermion5D.cc
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/PartialFractionFermion5D.h>
-#include <Grid/qcd/action/fermion/implementation/PartialFractionFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
- 
-#include "impl.h"
-template class PartialFractionFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/PartialFractionFermion5DInstantiationGparityWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/PartialFractionFermion5DInstantiationGparityWilsonImplD.cc
new file mode 120000
index 00000000..7f4cea71
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/PartialFractionFermion5DInstantiationGparityWilsonImplD.cc
@@ -0,0 +1 @@
+../PartialFractionFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/WilsonCloverFermionInstantiationGparityWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/WilsonCloverFermionInstantiationGparityWilsonImplD.cc
deleted file mode 100644
index af99dfb6..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/WilsonCloverFermionInstantiationGparityWilsonImplD.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: ./lib/qcd/action/fermion/WilsonCloverFermion.cc
-
-    Copyright (C) 2017
-
-    Author: paboyle <paboyle@ph.ed.ac.uk>
-    Author: Guido Cossu <guido.cossu@ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-    *************************************************************************************/
-/*  END LEGAL */
-
-#include <Grid/Grid.h>
-#include <Grid/qcd/spin/Dirac.h>
-#include <Grid/qcd/action/fermion/WilsonCloverFermion.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonCloverFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonCloverFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/WilsonCloverFermionInstantiationGparityWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/WilsonCloverFermionInstantiationGparityWilsonImplD.cc
new file mode 120000
index 00000000..9cc05107
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/WilsonCloverFermionInstantiationGparityWilsonImplD.cc
@@ -0,0 +1 @@
+../WilsonCloverFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/WilsonFermion5DInstantiationGparityWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/WilsonFermion5DInstantiationGparityWilsonImplD.cc
deleted file mode 100644
index 0dac989c..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/WilsonFermion5DInstantiationGparityWilsonImplD.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/WilsonFermion5DInstantiationGparityWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/WilsonFermion5DInstantiationGparityWilsonImplD.cc
new file mode 120000
index 00000000..804d0884
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/WilsonFermion5DInstantiationGparityWilsonImplD.cc
@@ -0,0 +1 @@
+../WilsonFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/WilsonFermionInstantiationGparityWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/WilsonFermionInstantiationGparityWilsonImplD.cc
deleted file mode 100644
index 6fd9c5ca..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/WilsonFermionInstantiationGparityWilsonImplD.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/WilsonFermionInstantiationGparityWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/WilsonFermionInstantiationGparityWilsonImplD.cc
new file mode 120000
index 00000000..5f6ab65e
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/WilsonFermionInstantiationGparityWilsonImplD.cc
@@ -0,0 +1 @@
+../WilsonFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/WilsonKernelsInstantiationGparityWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/WilsonKernelsInstantiationGparityWilsonImplD.cc
deleted file mode 100644
index 75f143cb..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/WilsonKernelsInstantiationGparityWilsonImplD.cc
+++ /dev/null
@@ -1,74 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandGparityImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-// Move these
-
-#include "impl.h"
-
-// G-parity requires more specialised implementation.
-template <>
-void WilsonKernels<IMPLEMENTATION>::ContractConservedCurrentSiteFwd(const SitePropagator &q_in_1, 
-							  const SitePropagator &q_in_2, 
-							  SitePropagator &q_out, 
-							  DoubledGaugeFieldView &U,
-							  unsigned int sU, 
-							  unsigned int mu, 
-							  bool switch_sign)
-{			
-  assert(0);		
-}			
-template <>		
-void WilsonKernels<IMPLEMENTATION>::ContractConservedCurrentSiteBwd( const SitePropagator &q_in_1,
-							   const SitePropagator &q_in_2,
-							   SitePropagator &q_out,
-							   DoubledGaugeFieldView &U,
-							   unsigned int mu,
-							   unsigned int sU,
-							   bool switch_sign)
-{			
-  assert(0);		
-}
-
-HAND_SPECIALISE_GPARITY(IMPLEMENTATION);
-
-
-template class WilsonKernels<IMPLEMENTATION>; 
-
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/WilsonKernelsInstantiationGparityWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/WilsonKernelsInstantiationGparityWilsonImplD.cc
new file mode 120000
index 00000000..87adea48
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/WilsonKernelsInstantiationGparityWilsonImplD.cc
@@ -0,0 +1 @@
+../WilsonKernelsInstantiationGparity.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/WilsonTMFermionInstantiationGparityWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/WilsonTMFermionInstantiationGparityWilsonImplD.cc
deleted file mode 100644
index adfa310c..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/WilsonTMFermionInstantiationGparityWilsonImplD.cc
+++ /dev/null
@@ -1,37 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/WilsonTMFermion.cc
-
-    Copyright (C) 2015
-
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/WilsonTMFermion.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonTMFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonTMFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/WilsonTMFermionInstantiationGparityWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/WilsonTMFermionInstantiationGparityWilsonImplD.cc
new file mode 120000
index 00000000..d5789bcf
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplD/WilsonTMFermionInstantiationGparityWilsonImplD.cc
@@ -0,0 +1 @@
+../WilsonTMFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/CayleyFermion5DInstantiationGparityWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/CayleyFermion5DInstantiationGparityWilsonImplDF.cc
deleted file mode 100644
index 5130db9c..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/CayleyFermion5DInstantiationGparityWilsonImplDF.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/CayleyFermion5DImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dcache.h>
-
-			   //#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dvec.h>
-			   //#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dgpu.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class CayleyFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/CayleyFermion5DInstantiationGparityWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/CayleyFermion5DInstantiationGparityWilsonImplDF.cc
new file mode 120000
index 00000000..cb1db625
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/CayleyFermion5DInstantiationGparityWilsonImplDF.cc
@@ -0,0 +1 @@
+../CayleyFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/ContinuedFractionFermion5DInstantiationGparityWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/ContinuedFractionFermion5DInstantiationGparityWilsonImplDF.cc
deleted file mode 100644
index ca0d6cea..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/ContinuedFractionFermion5DInstantiationGparityWilsonImplDF.cc
+++ /dev/null
@@ -1,38 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/ContinuedFractionFermion5D.cc
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/ContinuedFractionFermion5D.h>
-#include <Grid/qcd/action/fermion/implementation/ContinuedFractionFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class ContinuedFractionFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/ContinuedFractionFermion5DInstantiationGparityWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/ContinuedFractionFermion5DInstantiationGparityWilsonImplDF.cc
new file mode 120000
index 00000000..c2d4b8fc
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/ContinuedFractionFermion5DInstantiationGparityWilsonImplDF.cc
@@ -0,0 +1 @@
+../ContinuedFractionFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/DomainWallEOFAFermionInstantiationGparityWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/DomainWallEOFAFermionInstantiationGparityWilsonImplDF.cc
deleted file mode 100644
index f7198131..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/DomainWallEOFAFermionInstantiationGparityWilsonImplDF.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/DomainWallEOFAFermion.cc
-
-Copyright (C) 2017
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: David Murphy <dmurphy@phys.columbia.edu>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-			   /*  END LEGAL */
-
-#include <Grid/Grid_Eigen_Dense.h>
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/DomainWallEOFAFermion.h>
-#include <Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionCache.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class DomainWallEOFAFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/DomainWallEOFAFermionInstantiationGparityWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/DomainWallEOFAFermionInstantiationGparityWilsonImplDF.cc
new file mode 120000
index 00000000..2f550a2b
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/DomainWallEOFAFermionInstantiationGparityWilsonImplDF.cc
@@ -0,0 +1 @@
+../DomainWallEOFAFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/MobiusEOFAFermionInstantiationGparityWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/MobiusEOFAFermionInstantiationGparityWilsonImplDF.cc
deleted file mode 100644
index ce7eaac9..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/MobiusEOFAFermionInstantiationGparityWilsonImplDF.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/MobiusEOFAFermion.cc
-
-Copyright (C) 2017
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: David Murphy <dmurphy@phys.columbia.edu>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-			   /*  END LEGAL */
-
-#include <Grid/Grid_Eigen_Dense.h>
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/MobiusEOFAFermion.h>
-#include <Grid/qcd/action/fermion/implementation/MobiusEOFAFermionImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/MobiusEOFAFermionCache.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class MobiusEOFAFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/MobiusEOFAFermionInstantiationGparityWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/MobiusEOFAFermionInstantiationGparityWilsonImplDF.cc
new file mode 120000
index 00000000..7a8f1172
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/MobiusEOFAFermionInstantiationGparityWilsonImplDF.cc
@@ -0,0 +1 @@
+../MobiusEOFAFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/PartialFractionFermion5DInstantiationGparityWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/PartialFractionFermion5DInstantiationGparityWilsonImplDF.cc
deleted file mode 100644
index 757719b6..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/PartialFractionFermion5DInstantiationGparityWilsonImplDF.cc
+++ /dev/null
@@ -1,39 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/PartialFractionFermion5D.cc
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/PartialFractionFermion5D.h>
-#include <Grid/qcd/action/fermion/implementation/PartialFractionFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
- 
-#include "impl.h"
-template class PartialFractionFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/PartialFractionFermion5DInstantiationGparityWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/PartialFractionFermion5DInstantiationGparityWilsonImplDF.cc
new file mode 120000
index 00000000..7f4cea71
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/PartialFractionFermion5DInstantiationGparityWilsonImplDF.cc
@@ -0,0 +1 @@
+../PartialFractionFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonCloverFermionInstantiationGparityWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonCloverFermionInstantiationGparityWilsonImplDF.cc
deleted file mode 100644
index af99dfb6..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonCloverFermionInstantiationGparityWilsonImplDF.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: ./lib/qcd/action/fermion/WilsonCloverFermion.cc
-
-    Copyright (C) 2017
-
-    Author: paboyle <paboyle@ph.ed.ac.uk>
-    Author: Guido Cossu <guido.cossu@ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-    *************************************************************************************/
-/*  END LEGAL */
-
-#include <Grid/Grid.h>
-#include <Grid/qcd/spin/Dirac.h>
-#include <Grid/qcd/action/fermion/WilsonCloverFermion.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonCloverFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonCloverFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonCloverFermionInstantiationGparityWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonCloverFermionInstantiationGparityWilsonImplDF.cc
new file mode 120000
index 00000000..9cc05107
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonCloverFermionInstantiationGparityWilsonImplDF.cc
@@ -0,0 +1 @@
+../WilsonCloverFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonFermion5DInstantiationGparityWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonFermion5DInstantiationGparityWilsonImplDF.cc
deleted file mode 100644
index 0dac989c..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonFermion5DInstantiationGparityWilsonImplDF.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonFermion5DInstantiationGparityWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonFermion5DInstantiationGparityWilsonImplDF.cc
new file mode 120000
index 00000000..804d0884
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonFermion5DInstantiationGparityWilsonImplDF.cc
@@ -0,0 +1 @@
+../WilsonFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonFermionInstantiationGparityWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonFermionInstantiationGparityWilsonImplDF.cc
deleted file mode 100644
index 6fd9c5ca..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonFermionInstantiationGparityWilsonImplDF.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonFermionInstantiationGparityWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonFermionInstantiationGparityWilsonImplDF.cc
new file mode 120000
index 00000000..5f6ab65e
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonFermionInstantiationGparityWilsonImplDF.cc
@@ -0,0 +1 @@
+../WilsonFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonKernelsInstantiationGparityWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonKernelsInstantiationGparityWilsonImplDF.cc
deleted file mode 100644
index 75f143cb..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonKernelsInstantiationGparityWilsonImplDF.cc
+++ /dev/null
@@ -1,74 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandGparityImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-// Move these
-
-#include "impl.h"
-
-// G-parity requires more specialised implementation.
-template <>
-void WilsonKernels<IMPLEMENTATION>::ContractConservedCurrentSiteFwd(const SitePropagator &q_in_1, 
-							  const SitePropagator &q_in_2, 
-							  SitePropagator &q_out, 
-							  DoubledGaugeFieldView &U,
-							  unsigned int sU, 
-							  unsigned int mu, 
-							  bool switch_sign)
-{			
-  assert(0);		
-}			
-template <>		
-void WilsonKernels<IMPLEMENTATION>::ContractConservedCurrentSiteBwd( const SitePropagator &q_in_1,
-							   const SitePropagator &q_in_2,
-							   SitePropagator &q_out,
-							   DoubledGaugeFieldView &U,
-							   unsigned int mu,
-							   unsigned int sU,
-							   bool switch_sign)
-{			
-  assert(0);		
-}
-
-HAND_SPECIALISE_GPARITY(IMPLEMENTATION);
-
-
-template class WilsonKernels<IMPLEMENTATION>; 
-
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonKernelsInstantiationGparityWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonKernelsInstantiationGparityWilsonImplDF.cc
new file mode 120000
index 00000000..87adea48
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonKernelsInstantiationGparityWilsonImplDF.cc
@@ -0,0 +1 @@
+../WilsonKernelsInstantiationGparity.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonTMFermionInstantiationGparityWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonTMFermionInstantiationGparityWilsonImplDF.cc
deleted file mode 100644
index adfa310c..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonTMFermionInstantiationGparityWilsonImplDF.cc
+++ /dev/null
@@ -1,37 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/WilsonTMFermion.cc
-
-    Copyright (C) 2015
-
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/WilsonTMFermion.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonTMFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonTMFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonTMFermionInstantiationGparityWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonTMFermionInstantiationGparityWilsonImplDF.cc
new file mode 120000
index 00000000..d5789bcf
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplDF/WilsonTMFermionInstantiationGparityWilsonImplDF.cc
@@ -0,0 +1 @@
+../WilsonTMFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/CayleyFermion5DInstantiationGparityWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/CayleyFermion5DInstantiationGparityWilsonImplF.cc
deleted file mode 100644
index 5130db9c..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/CayleyFermion5DInstantiationGparityWilsonImplF.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/CayleyFermion5DImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dcache.h>
-
-			   //#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dvec.h>
-			   //#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dgpu.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class CayleyFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/CayleyFermion5DInstantiationGparityWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/CayleyFermion5DInstantiationGparityWilsonImplF.cc
new file mode 120000
index 00000000..cb1db625
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/CayleyFermion5DInstantiationGparityWilsonImplF.cc
@@ -0,0 +1 @@
+../CayleyFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/ContinuedFractionFermion5DInstantiationGparityWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/ContinuedFractionFermion5DInstantiationGparityWilsonImplF.cc
deleted file mode 100644
index ca0d6cea..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/ContinuedFractionFermion5DInstantiationGparityWilsonImplF.cc
+++ /dev/null
@@ -1,38 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/ContinuedFractionFermion5D.cc
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/ContinuedFractionFermion5D.h>
-#include <Grid/qcd/action/fermion/implementation/ContinuedFractionFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class ContinuedFractionFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/ContinuedFractionFermion5DInstantiationGparityWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/ContinuedFractionFermion5DInstantiationGparityWilsonImplF.cc
new file mode 120000
index 00000000..c2d4b8fc
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/ContinuedFractionFermion5DInstantiationGparityWilsonImplF.cc
@@ -0,0 +1 @@
+../ContinuedFractionFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/DomainWallEOFAFermionInstantiationGparityWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/DomainWallEOFAFermionInstantiationGparityWilsonImplF.cc
deleted file mode 100644
index f7198131..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/DomainWallEOFAFermionInstantiationGparityWilsonImplF.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/DomainWallEOFAFermion.cc
-
-Copyright (C) 2017
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: David Murphy <dmurphy@phys.columbia.edu>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-			   /*  END LEGAL */
-
-#include <Grid/Grid_Eigen_Dense.h>
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/DomainWallEOFAFermion.h>
-#include <Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionCache.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class DomainWallEOFAFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/DomainWallEOFAFermionInstantiationGparityWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/DomainWallEOFAFermionInstantiationGparityWilsonImplF.cc
new file mode 120000
index 00000000..2f550a2b
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/DomainWallEOFAFermionInstantiationGparityWilsonImplF.cc
@@ -0,0 +1 @@
+../DomainWallEOFAFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/MobiusEOFAFermionInstantiationGparityWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/MobiusEOFAFermionInstantiationGparityWilsonImplF.cc
deleted file mode 100644
index ce7eaac9..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/MobiusEOFAFermionInstantiationGparityWilsonImplF.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/MobiusEOFAFermion.cc
-
-Copyright (C) 2017
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: David Murphy <dmurphy@phys.columbia.edu>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-			   /*  END LEGAL */
-
-#include <Grid/Grid_Eigen_Dense.h>
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/MobiusEOFAFermion.h>
-#include <Grid/qcd/action/fermion/implementation/MobiusEOFAFermionImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/MobiusEOFAFermionCache.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class MobiusEOFAFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/MobiusEOFAFermionInstantiationGparityWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/MobiusEOFAFermionInstantiationGparityWilsonImplF.cc
new file mode 120000
index 00000000..7a8f1172
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/MobiusEOFAFermionInstantiationGparityWilsonImplF.cc
@@ -0,0 +1 @@
+../MobiusEOFAFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/PartialFractionFermion5DInstantiationGparityWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/PartialFractionFermion5DInstantiationGparityWilsonImplF.cc
deleted file mode 100644
index 757719b6..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/PartialFractionFermion5DInstantiationGparityWilsonImplF.cc
+++ /dev/null
@@ -1,39 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/PartialFractionFermion5D.cc
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/PartialFractionFermion5D.h>
-#include <Grid/qcd/action/fermion/implementation/PartialFractionFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
- 
-#include "impl.h"
-template class PartialFractionFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/PartialFractionFermion5DInstantiationGparityWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/PartialFractionFermion5DInstantiationGparityWilsonImplF.cc
new file mode 120000
index 00000000..7f4cea71
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/PartialFractionFermion5DInstantiationGparityWilsonImplF.cc
@@ -0,0 +1 @@
+../PartialFractionFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/WilsonCloverFermionInstantiationGparityWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/WilsonCloverFermionInstantiationGparityWilsonImplF.cc
deleted file mode 100644
index af99dfb6..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/WilsonCloverFermionInstantiationGparityWilsonImplF.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: ./lib/qcd/action/fermion/WilsonCloverFermion.cc
-
-    Copyright (C) 2017
-
-    Author: paboyle <paboyle@ph.ed.ac.uk>
-    Author: Guido Cossu <guido.cossu@ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-    *************************************************************************************/
-/*  END LEGAL */
-
-#include <Grid/Grid.h>
-#include <Grid/qcd/spin/Dirac.h>
-#include <Grid/qcd/action/fermion/WilsonCloverFermion.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonCloverFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonCloverFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/WilsonCloverFermionInstantiationGparityWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/WilsonCloverFermionInstantiationGparityWilsonImplF.cc
new file mode 120000
index 00000000..9cc05107
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/WilsonCloverFermionInstantiationGparityWilsonImplF.cc
@@ -0,0 +1 @@
+../WilsonCloverFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/WilsonFermion5DInstantiationGparityWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/WilsonFermion5DInstantiationGparityWilsonImplF.cc
deleted file mode 100644
index 0dac989c..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/WilsonFermion5DInstantiationGparityWilsonImplF.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/WilsonFermion5DInstantiationGparityWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/WilsonFermion5DInstantiationGparityWilsonImplF.cc
new file mode 120000
index 00000000..804d0884
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/WilsonFermion5DInstantiationGparityWilsonImplF.cc
@@ -0,0 +1 @@
+../WilsonFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/WilsonFermionInstantiationGparityWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/WilsonFermionInstantiationGparityWilsonImplF.cc
deleted file mode 100644
index 6fd9c5ca..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/WilsonFermionInstantiationGparityWilsonImplF.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/WilsonFermionInstantiationGparityWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/WilsonFermionInstantiationGparityWilsonImplF.cc
new file mode 120000
index 00000000..5f6ab65e
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/WilsonFermionInstantiationGparityWilsonImplF.cc
@@ -0,0 +1 @@
+../WilsonFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/WilsonKernelsInstantiationGparityWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/WilsonKernelsInstantiationGparityWilsonImplF.cc
deleted file mode 100644
index 75f143cb..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/WilsonKernelsInstantiationGparityWilsonImplF.cc
+++ /dev/null
@@ -1,74 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandGparityImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-// Move these
-
-#include "impl.h"
-
-// G-parity requires more specialised implementation.
-template <>
-void WilsonKernels<IMPLEMENTATION>::ContractConservedCurrentSiteFwd(const SitePropagator &q_in_1, 
-							  const SitePropagator &q_in_2, 
-							  SitePropagator &q_out, 
-							  DoubledGaugeFieldView &U,
-							  unsigned int sU, 
-							  unsigned int mu, 
-							  bool switch_sign)
-{			
-  assert(0);		
-}			
-template <>		
-void WilsonKernels<IMPLEMENTATION>::ContractConservedCurrentSiteBwd( const SitePropagator &q_in_1,
-							   const SitePropagator &q_in_2,
-							   SitePropagator &q_out,
-							   DoubledGaugeFieldView &U,
-							   unsigned int mu,
-							   unsigned int sU,
-							   bool switch_sign)
-{			
-  assert(0);		
-}
-
-HAND_SPECIALISE_GPARITY(IMPLEMENTATION);
-
-
-template class WilsonKernels<IMPLEMENTATION>; 
-
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/WilsonKernelsInstantiationGparityWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/WilsonKernelsInstantiationGparityWilsonImplF.cc
new file mode 120000
index 00000000..87adea48
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/WilsonKernelsInstantiationGparityWilsonImplF.cc
@@ -0,0 +1 @@
+../WilsonKernelsInstantiationGparity.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/WilsonTMFermionInstantiationGparityWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/WilsonTMFermionInstantiationGparityWilsonImplF.cc
deleted file mode 100644
index adfa310c..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/WilsonTMFermionInstantiationGparityWilsonImplF.cc
+++ /dev/null
@@ -1,37 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/WilsonTMFermion.cc
-
-    Copyright (C) 2015
-
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/WilsonTMFermion.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonTMFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonTMFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/WilsonTMFermionInstantiationGparityWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/WilsonTMFermionInstantiationGparityWilsonImplF.cc
new file mode 120000
index 00000000..d5789bcf
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplF/WilsonTMFermionInstantiationGparityWilsonImplF.cc
@@ -0,0 +1 @@
+../WilsonTMFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/CayleyFermion5DInstantiationGparityWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/CayleyFermion5DInstantiationGparityWilsonImplFH.cc
deleted file mode 100644
index 5130db9c..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/CayleyFermion5DInstantiationGparityWilsonImplFH.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/CayleyFermion5DImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dcache.h>
-
-			   //#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dvec.h>
-			   //#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dgpu.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class CayleyFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/CayleyFermion5DInstantiationGparityWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/CayleyFermion5DInstantiationGparityWilsonImplFH.cc
new file mode 120000
index 00000000..cb1db625
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/CayleyFermion5DInstantiationGparityWilsonImplFH.cc
@@ -0,0 +1 @@
+../CayleyFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/ContinuedFractionFermion5DInstantiationGparityWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/ContinuedFractionFermion5DInstantiationGparityWilsonImplFH.cc
deleted file mode 100644
index ca0d6cea..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/ContinuedFractionFermion5DInstantiationGparityWilsonImplFH.cc
+++ /dev/null
@@ -1,38 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/ContinuedFractionFermion5D.cc
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/ContinuedFractionFermion5D.h>
-#include <Grid/qcd/action/fermion/implementation/ContinuedFractionFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class ContinuedFractionFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/ContinuedFractionFermion5DInstantiationGparityWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/ContinuedFractionFermion5DInstantiationGparityWilsonImplFH.cc
new file mode 120000
index 00000000..c2d4b8fc
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/ContinuedFractionFermion5DInstantiationGparityWilsonImplFH.cc
@@ -0,0 +1 @@
+../ContinuedFractionFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/DomainWallEOFAFermionInstantiationGparityWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/DomainWallEOFAFermionInstantiationGparityWilsonImplFH.cc
deleted file mode 100644
index f7198131..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/DomainWallEOFAFermionInstantiationGparityWilsonImplFH.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/DomainWallEOFAFermion.cc
-
-Copyright (C) 2017
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: David Murphy <dmurphy@phys.columbia.edu>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-			   /*  END LEGAL */
-
-#include <Grid/Grid_Eigen_Dense.h>
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/DomainWallEOFAFermion.h>
-#include <Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionCache.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class DomainWallEOFAFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/DomainWallEOFAFermionInstantiationGparityWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/DomainWallEOFAFermionInstantiationGparityWilsonImplFH.cc
new file mode 120000
index 00000000..2f550a2b
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/DomainWallEOFAFermionInstantiationGparityWilsonImplFH.cc
@@ -0,0 +1 @@
+../DomainWallEOFAFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/MobiusEOFAFermionInstantiationGparityWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/MobiusEOFAFermionInstantiationGparityWilsonImplFH.cc
deleted file mode 100644
index ce7eaac9..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/MobiusEOFAFermionInstantiationGparityWilsonImplFH.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/MobiusEOFAFermion.cc
-
-Copyright (C) 2017
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: David Murphy <dmurphy@phys.columbia.edu>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-			   /*  END LEGAL */
-
-#include <Grid/Grid_Eigen_Dense.h>
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/MobiusEOFAFermion.h>
-#include <Grid/qcd/action/fermion/implementation/MobiusEOFAFermionImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/MobiusEOFAFermionCache.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class MobiusEOFAFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/MobiusEOFAFermionInstantiationGparityWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/MobiusEOFAFermionInstantiationGparityWilsonImplFH.cc
new file mode 120000
index 00000000..7a8f1172
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/MobiusEOFAFermionInstantiationGparityWilsonImplFH.cc
@@ -0,0 +1 @@
+../MobiusEOFAFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/PartialFractionFermion5DInstantiationGparityWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/PartialFractionFermion5DInstantiationGparityWilsonImplFH.cc
deleted file mode 100644
index 757719b6..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/PartialFractionFermion5DInstantiationGparityWilsonImplFH.cc
+++ /dev/null
@@ -1,39 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/PartialFractionFermion5D.cc
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/PartialFractionFermion5D.h>
-#include <Grid/qcd/action/fermion/implementation/PartialFractionFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
- 
-#include "impl.h"
-template class PartialFractionFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/PartialFractionFermion5DInstantiationGparityWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/PartialFractionFermion5DInstantiationGparityWilsonImplFH.cc
new file mode 120000
index 00000000..7f4cea71
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/PartialFractionFermion5DInstantiationGparityWilsonImplFH.cc
@@ -0,0 +1 @@
+../PartialFractionFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonCloverFermionInstantiationGparityWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonCloverFermionInstantiationGparityWilsonImplFH.cc
deleted file mode 100644
index af99dfb6..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonCloverFermionInstantiationGparityWilsonImplFH.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: ./lib/qcd/action/fermion/WilsonCloverFermion.cc
-
-    Copyright (C) 2017
-
-    Author: paboyle <paboyle@ph.ed.ac.uk>
-    Author: Guido Cossu <guido.cossu@ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-    *************************************************************************************/
-/*  END LEGAL */
-
-#include <Grid/Grid.h>
-#include <Grid/qcd/spin/Dirac.h>
-#include <Grid/qcd/action/fermion/WilsonCloverFermion.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonCloverFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonCloverFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonCloverFermionInstantiationGparityWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonCloverFermionInstantiationGparityWilsonImplFH.cc
new file mode 120000
index 00000000..9cc05107
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonCloverFermionInstantiationGparityWilsonImplFH.cc
@@ -0,0 +1 @@
+../WilsonCloverFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonFermion5DInstantiationGparityWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonFermion5DInstantiationGparityWilsonImplFH.cc
deleted file mode 100644
index 0dac989c..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonFermion5DInstantiationGparityWilsonImplFH.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonFermion5DInstantiationGparityWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonFermion5DInstantiationGparityWilsonImplFH.cc
new file mode 120000
index 00000000..804d0884
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonFermion5DInstantiationGparityWilsonImplFH.cc
@@ -0,0 +1 @@
+../WilsonFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonFermionInstantiationGparityWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonFermionInstantiationGparityWilsonImplFH.cc
deleted file mode 100644
index 6fd9c5ca..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonFermionInstantiationGparityWilsonImplFH.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonFermionInstantiationGparityWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonFermionInstantiationGparityWilsonImplFH.cc
new file mode 120000
index 00000000..5f6ab65e
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonFermionInstantiationGparityWilsonImplFH.cc
@@ -0,0 +1 @@
+../WilsonFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonKernelsInstantiationGparityWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonKernelsInstantiationGparityWilsonImplFH.cc
deleted file mode 100644
index 75f143cb..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonKernelsInstantiationGparityWilsonImplFH.cc
+++ /dev/null
@@ -1,74 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandGparityImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-// Move these
-
-#include "impl.h"
-
-// G-parity requires more specialised implementation.
-template <>
-void WilsonKernels<IMPLEMENTATION>::ContractConservedCurrentSiteFwd(const SitePropagator &q_in_1, 
-							  const SitePropagator &q_in_2, 
-							  SitePropagator &q_out, 
-							  DoubledGaugeFieldView &U,
-							  unsigned int sU, 
-							  unsigned int mu, 
-							  bool switch_sign)
-{			
-  assert(0);		
-}			
-template <>		
-void WilsonKernels<IMPLEMENTATION>::ContractConservedCurrentSiteBwd( const SitePropagator &q_in_1,
-							   const SitePropagator &q_in_2,
-							   SitePropagator &q_out,
-							   DoubledGaugeFieldView &U,
-							   unsigned int mu,
-							   unsigned int sU,
-							   bool switch_sign)
-{			
-  assert(0);		
-}
-
-HAND_SPECIALISE_GPARITY(IMPLEMENTATION);
-
-
-template class WilsonKernels<IMPLEMENTATION>; 
-
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonKernelsInstantiationGparityWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonKernelsInstantiationGparityWilsonImplFH.cc
new file mode 120000
index 00000000..87adea48
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonKernelsInstantiationGparityWilsonImplFH.cc
@@ -0,0 +1 @@
+../WilsonKernelsInstantiationGparity.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonTMFermionInstantiationGparityWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonTMFermionInstantiationGparityWilsonImplFH.cc
deleted file mode 100644
index adfa310c..00000000
--- a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonTMFermionInstantiationGparityWilsonImplFH.cc
+++ /dev/null
@@ -1,37 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/WilsonTMFermion.cc
-
-    Copyright (C) 2015
-
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/WilsonTMFermion.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonTMFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonTMFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonTMFermionInstantiationGparityWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonTMFermionInstantiationGparityWilsonImplFH.cc
new file mode 120000
index 00000000..d5789bcf
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/GparityWilsonImplFH/WilsonTMFermionInstantiationGparityWilsonImplFH.cc
@@ -0,0 +1 @@
+../WilsonTMFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/NaiveStaggeredFermionInstantiation.cc b/Grid/qcd/action/fermion/instantiation/NaiveStaggeredFermionInstantiation.cc
new file mode 100644
index 00000000..c424cb2d
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/NaiveStaggeredFermionInstantiation.cc
@@ -0,0 +1,36 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid
+
+Source file: ./lib/qcd/action/fermion/ImprovedStaggeredFermion.cc
+
+Copyright (C) 2015
+
+Author: Azusa Yamaguchi, Peter Boyle
+
+This program is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 2 of the License, or
+(at your option) any later version.
+
+This program is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along
+with this program; if not, write to the Free Software Foundation, Inc.,
+51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+See the full license in the file "LICENSE" in the top level distribution
+directory
+*************************************************************************************/
+			   /*  END LEGAL */
+#include <Grid/Grid.h>
+
+NAMESPACE_BEGIN(Grid);
+
+const std::vector<int> NaiveStaggeredFermionStatic::directions({0, 1, 2, 3, 0, 1, 2, 3});
+const std::vector<int> NaiveStaggeredFermionStatic::displacements({1, 1, 1, 1, -1, -1, -1, -1});
+
+NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/NaiveStaggeredFermionInstantiation.cc.master b/Grid/qcd/action/fermion/instantiation/NaiveStaggeredFermionInstantiation.cc.master
new file mode 100644
index 00000000..75b75678
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/NaiveStaggeredFermionInstantiation.cc.master
@@ -0,0 +1,37 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid
+
+Source file: ./lib/qcd/action/fermion/NaiveStaggeredFermion.cc
+
+Copyright (C) 2015
+
+Author: Azusa Yamaguchi, Peter Boyle
+
+This program is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 2 of the License, or
+(at your option) any later version.
+
+This program is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along
+with this program; if not, write to the Free Software Foundation, Inc.,
+51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+See the full license in the file "LICENSE" in the top level distribution
+directory
+*************************************************************************************/
+			   /*  END LEGAL */
+#include <Grid/Grid.h>
+#include <Grid/qcd/action/fermion/implementation/NaiveStaggeredFermionImplementation.h>
+
+NAMESPACE_BEGIN(Grid);
+
+#include "impl.h"
+template class NaiveStaggeredFermion<IMPLEMENTATION>; 
+
+NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/StaggeredImplD/ImprovedStaggeredFermion5DInstantiationStaggeredImplD.cc b/Grid/qcd/action/fermion/instantiation/StaggeredImplD/ImprovedStaggeredFermion5DInstantiationStaggeredImplD.cc
deleted file mode 100644
index 572b375c..00000000
--- a/Grid/qcd/action/fermion/instantiation/StaggeredImplD/ImprovedStaggeredFermion5DInstantiationStaggeredImplD.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/ImprovedStaggeredFermion5D.cc
-
-    Copyright (C) 2015
-
-Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-
-#include <Grid/qcd/action/fermion/ImprovedStaggeredFermion5D.h>
-#include <Grid/qcd/action/fermion/implementation/ImprovedStaggeredFermion5DImplementation.h>
-#include <Grid/perfmon/PerfCount.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class ImprovedStaggeredFermion5D<IMPLEMENTATION>; 
-  
-NAMESPACE_END(Grid);
-
-
-
-
diff --git a/Grid/qcd/action/fermion/instantiation/StaggeredImplD/ImprovedStaggeredFermion5DInstantiationStaggeredImplD.cc b/Grid/qcd/action/fermion/instantiation/StaggeredImplD/ImprovedStaggeredFermion5DInstantiationStaggeredImplD.cc
new file mode 120000
index 00000000..a8082840
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/StaggeredImplD/ImprovedStaggeredFermion5DInstantiationStaggeredImplD.cc
@@ -0,0 +1 @@
+../ImprovedStaggeredFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/StaggeredImplD/ImprovedStaggeredFermionInstantiationStaggeredImplD.cc b/Grid/qcd/action/fermion/instantiation/StaggeredImplD/ImprovedStaggeredFermionInstantiationStaggeredImplD.cc
deleted file mode 100644
index d35b7349..00000000
--- a/Grid/qcd/action/fermion/instantiation/StaggeredImplD/ImprovedStaggeredFermionInstantiationStaggeredImplD.cc
+++ /dev/null
@@ -1,37 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/ImprovedStaggeredFermion.cc
-
-Copyright (C) 2015
-
-Author: Azusa Yamaguchi, Peter Boyle
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-			   /*  END LEGAL */
-#include <Grid/Grid.h>
-#include <Grid/qcd/action/fermion/implementation/ImprovedStaggeredFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class ImprovedStaggeredFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/StaggeredImplD/ImprovedStaggeredFermionInstantiationStaggeredImplD.cc b/Grid/qcd/action/fermion/instantiation/StaggeredImplD/ImprovedStaggeredFermionInstantiationStaggeredImplD.cc
new file mode 120000
index 00000000..e239b39a
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/StaggeredImplD/ImprovedStaggeredFermionInstantiationStaggeredImplD.cc
@@ -0,0 +1 @@
+../ImprovedStaggeredFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/StaggeredImplD/NaiveStaggeredFermionInstantiationStaggeredImplD.cc b/Grid/qcd/action/fermion/instantiation/StaggeredImplD/NaiveStaggeredFermionInstantiationStaggeredImplD.cc
new file mode 120000
index 00000000..42057f56
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/StaggeredImplD/NaiveStaggeredFermionInstantiationStaggeredImplD.cc
@@ -0,0 +1 @@
+../NaiveStaggeredFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/StaggeredImplD/StaggeredKernelsInstantiationStaggeredImplD.cc b/Grid/qcd/action/fermion/instantiation/StaggeredImplD/StaggeredKernelsInstantiationStaggeredImplD.cc
deleted file mode 100644
index c3acf963..00000000
--- a/Grid/qcd/action/fermion/instantiation/StaggeredImplD/StaggeredKernelsInstantiationStaggeredImplD.cc
+++ /dev/null
@@ -1,39 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Azusa Yamaguchi, Peter Boyle
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/StaggeredKernelsImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/StaggeredKernelsHand.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class StaggeredKernels<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/StaggeredImplD/StaggeredKernelsInstantiationStaggeredImplD.cc b/Grid/qcd/action/fermion/instantiation/StaggeredImplD/StaggeredKernelsInstantiationStaggeredImplD.cc
new file mode 120000
index 00000000..a4d1cbc9
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/StaggeredImplD/StaggeredKernelsInstantiationStaggeredImplD.cc
@@ -0,0 +1 @@
+../StaggeredKernelsInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/StaggeredImplF/ImprovedStaggeredFermion5DInstantiationStaggeredImplF.cc b/Grid/qcd/action/fermion/instantiation/StaggeredImplF/ImprovedStaggeredFermion5DInstantiationStaggeredImplF.cc
deleted file mode 100644
index 572b375c..00000000
--- a/Grid/qcd/action/fermion/instantiation/StaggeredImplF/ImprovedStaggeredFermion5DInstantiationStaggeredImplF.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/ImprovedStaggeredFermion5D.cc
-
-    Copyright (C) 2015
-
-Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-
-#include <Grid/qcd/action/fermion/ImprovedStaggeredFermion5D.h>
-#include <Grid/qcd/action/fermion/implementation/ImprovedStaggeredFermion5DImplementation.h>
-#include <Grid/perfmon/PerfCount.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class ImprovedStaggeredFermion5D<IMPLEMENTATION>; 
-  
-NAMESPACE_END(Grid);
-
-
-
-
diff --git a/Grid/qcd/action/fermion/instantiation/StaggeredImplF/ImprovedStaggeredFermion5DInstantiationStaggeredImplF.cc b/Grid/qcd/action/fermion/instantiation/StaggeredImplF/ImprovedStaggeredFermion5DInstantiationStaggeredImplF.cc
new file mode 120000
index 00000000..a8082840
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/StaggeredImplF/ImprovedStaggeredFermion5DInstantiationStaggeredImplF.cc
@@ -0,0 +1 @@
+../ImprovedStaggeredFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/StaggeredImplF/ImprovedStaggeredFermionInstantiationStaggeredImplF.cc b/Grid/qcd/action/fermion/instantiation/StaggeredImplF/ImprovedStaggeredFermionInstantiationStaggeredImplF.cc
deleted file mode 100644
index d35b7349..00000000
--- a/Grid/qcd/action/fermion/instantiation/StaggeredImplF/ImprovedStaggeredFermionInstantiationStaggeredImplF.cc
+++ /dev/null
@@ -1,37 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/ImprovedStaggeredFermion.cc
-
-Copyright (C) 2015
-
-Author: Azusa Yamaguchi, Peter Boyle
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-			   /*  END LEGAL */
-#include <Grid/Grid.h>
-#include <Grid/qcd/action/fermion/implementation/ImprovedStaggeredFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class ImprovedStaggeredFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/StaggeredImplF/ImprovedStaggeredFermionInstantiationStaggeredImplF.cc b/Grid/qcd/action/fermion/instantiation/StaggeredImplF/ImprovedStaggeredFermionInstantiationStaggeredImplF.cc
new file mode 120000
index 00000000..e239b39a
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/StaggeredImplF/ImprovedStaggeredFermionInstantiationStaggeredImplF.cc
@@ -0,0 +1 @@
+../ImprovedStaggeredFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/StaggeredImplF/NaiveStaggeredFermionInstantiationStaggeredImplF.cc b/Grid/qcd/action/fermion/instantiation/StaggeredImplF/NaiveStaggeredFermionInstantiationStaggeredImplF.cc
new file mode 120000
index 00000000..42057f56
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/StaggeredImplF/NaiveStaggeredFermionInstantiationStaggeredImplF.cc
@@ -0,0 +1 @@
+../NaiveStaggeredFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/StaggeredImplF/StaggeredKernelsInstantiationStaggeredImplF.cc b/Grid/qcd/action/fermion/instantiation/StaggeredImplF/StaggeredKernelsInstantiationStaggeredImplF.cc
deleted file mode 100644
index c3acf963..00000000
--- a/Grid/qcd/action/fermion/instantiation/StaggeredImplF/StaggeredKernelsInstantiationStaggeredImplF.cc
+++ /dev/null
@@ -1,39 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Azusa Yamaguchi, Peter Boyle
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/StaggeredKernelsImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/StaggeredKernelsHand.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class StaggeredKernels<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/StaggeredImplF/StaggeredKernelsInstantiationStaggeredImplF.cc b/Grid/qcd/action/fermion/instantiation/StaggeredImplF/StaggeredKernelsInstantiationStaggeredImplF.cc
new file mode 120000
index 00000000..a4d1cbc9
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/StaggeredImplF/StaggeredKernelsInstantiationStaggeredImplF.cc
@@ -0,0 +1 @@
+../StaggeredKernelsInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonAdjImplD/WilsonCloverFermionInstantiationWilsonAdjImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonAdjImplD/WilsonCloverFermionInstantiationWilsonAdjImplD.cc
deleted file mode 100644
index af99dfb6..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonAdjImplD/WilsonCloverFermionInstantiationWilsonAdjImplD.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: ./lib/qcd/action/fermion/WilsonCloverFermion.cc
-
-    Copyright (C) 2017
-
-    Author: paboyle <paboyle@ph.ed.ac.uk>
-    Author: Guido Cossu <guido.cossu@ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-    *************************************************************************************/
-/*  END LEGAL */
-
-#include <Grid/Grid.h>
-#include <Grid/qcd/spin/Dirac.h>
-#include <Grid/qcd/action/fermion/WilsonCloverFermion.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonCloverFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonCloverFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonAdjImplD/WilsonCloverFermionInstantiationWilsonAdjImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonAdjImplD/WilsonCloverFermionInstantiationWilsonAdjImplD.cc
new file mode 120000
index 00000000..9cc05107
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonAdjImplD/WilsonCloverFermionInstantiationWilsonAdjImplD.cc
@@ -0,0 +1 @@
+../WilsonCloverFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonAdjImplD/WilsonFermionInstantiationWilsonAdjImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonAdjImplD/WilsonFermionInstantiationWilsonAdjImplD.cc
deleted file mode 100644
index 6fd9c5ca..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonAdjImplD/WilsonFermionInstantiationWilsonAdjImplD.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonAdjImplD/WilsonFermionInstantiationWilsonAdjImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonAdjImplD/WilsonFermionInstantiationWilsonAdjImplD.cc
new file mode 120000
index 00000000..5f6ab65e
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonAdjImplD/WilsonFermionInstantiationWilsonAdjImplD.cc
@@ -0,0 +1 @@
+../WilsonFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonAdjImplD/WilsonTMFermionInstantiationWilsonAdjImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonAdjImplD/WilsonTMFermionInstantiationWilsonAdjImplD.cc
deleted file mode 100644
index adfa310c..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonAdjImplD/WilsonTMFermionInstantiationWilsonAdjImplD.cc
+++ /dev/null
@@ -1,37 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/WilsonTMFermion.cc
-
-    Copyright (C) 2015
-
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/WilsonTMFermion.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonTMFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonTMFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonAdjImplD/WilsonTMFermionInstantiationWilsonAdjImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonAdjImplD/WilsonTMFermionInstantiationWilsonAdjImplD.cc
new file mode 120000
index 00000000..d5789bcf
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonAdjImplD/WilsonTMFermionInstantiationWilsonAdjImplD.cc
@@ -0,0 +1 @@
+../WilsonTMFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonAdjImplF/WilsonCloverFermionInstantiationWilsonAdjImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonAdjImplF/WilsonCloverFermionInstantiationWilsonAdjImplF.cc
deleted file mode 100644
index af99dfb6..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonAdjImplF/WilsonCloverFermionInstantiationWilsonAdjImplF.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: ./lib/qcd/action/fermion/WilsonCloverFermion.cc
-
-    Copyright (C) 2017
-
-    Author: paboyle <paboyle@ph.ed.ac.uk>
-    Author: Guido Cossu <guido.cossu@ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-    *************************************************************************************/
-/*  END LEGAL */
-
-#include <Grid/Grid.h>
-#include <Grid/qcd/spin/Dirac.h>
-#include <Grid/qcd/action/fermion/WilsonCloverFermion.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonCloverFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonCloverFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonAdjImplF/WilsonCloverFermionInstantiationWilsonAdjImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonAdjImplF/WilsonCloverFermionInstantiationWilsonAdjImplF.cc
new file mode 120000
index 00000000..9cc05107
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonAdjImplF/WilsonCloverFermionInstantiationWilsonAdjImplF.cc
@@ -0,0 +1 @@
+../WilsonCloverFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonAdjImplF/WilsonFermionInstantiationWilsonAdjImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonAdjImplF/WilsonFermionInstantiationWilsonAdjImplF.cc
deleted file mode 100644
index 6fd9c5ca..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonAdjImplF/WilsonFermionInstantiationWilsonAdjImplF.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonAdjImplF/WilsonFermionInstantiationWilsonAdjImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonAdjImplF/WilsonFermionInstantiationWilsonAdjImplF.cc
new file mode 120000
index 00000000..5f6ab65e
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonAdjImplF/WilsonFermionInstantiationWilsonAdjImplF.cc
@@ -0,0 +1 @@
+../WilsonFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonAdjImplF/WilsonTMFermionInstantiationWilsonAdjImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonAdjImplF/WilsonTMFermionInstantiationWilsonAdjImplF.cc
deleted file mode 100644
index adfa310c..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonAdjImplF/WilsonTMFermionInstantiationWilsonAdjImplF.cc
+++ /dev/null
@@ -1,37 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/WilsonTMFermion.cc
-
-    Copyright (C) 2015
-
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/WilsonTMFermion.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonTMFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonTMFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonAdjImplF/WilsonTMFermionInstantiationWilsonAdjImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonAdjImplF/WilsonTMFermionInstantiationWilsonAdjImplF.cc
new file mode 120000
index 00000000..d5789bcf
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonAdjImplF/WilsonTMFermionInstantiationWilsonAdjImplF.cc
@@ -0,0 +1 @@
+../WilsonTMFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplD/CayleyFermion5DInstantiationWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplD/CayleyFermion5DInstantiationWilsonImplD.cc
deleted file mode 100644
index 5130db9c..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplD/CayleyFermion5DInstantiationWilsonImplD.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/CayleyFermion5DImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dcache.h>
-
-			   //#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dvec.h>
-			   //#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dgpu.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class CayleyFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplD/CayleyFermion5DInstantiationWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplD/CayleyFermion5DInstantiationWilsonImplD.cc
new file mode 120000
index 00000000..cb1db625
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplD/CayleyFermion5DInstantiationWilsonImplD.cc
@@ -0,0 +1 @@
+../CayleyFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplD/ContinuedFractionFermion5DInstantiationWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplD/ContinuedFractionFermion5DInstantiationWilsonImplD.cc
deleted file mode 100644
index ca0d6cea..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplD/ContinuedFractionFermion5DInstantiationWilsonImplD.cc
+++ /dev/null
@@ -1,38 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/ContinuedFractionFermion5D.cc
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/ContinuedFractionFermion5D.h>
-#include <Grid/qcd/action/fermion/implementation/ContinuedFractionFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class ContinuedFractionFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplD/ContinuedFractionFermion5DInstantiationWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplD/ContinuedFractionFermion5DInstantiationWilsonImplD.cc
new file mode 120000
index 00000000..c2d4b8fc
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplD/ContinuedFractionFermion5DInstantiationWilsonImplD.cc
@@ -0,0 +1 @@
+../ContinuedFractionFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplD/DomainWallEOFAFermionInstantiationWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplD/DomainWallEOFAFermionInstantiationWilsonImplD.cc
deleted file mode 100644
index f7198131..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplD/DomainWallEOFAFermionInstantiationWilsonImplD.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/DomainWallEOFAFermion.cc
-
-Copyright (C) 2017
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: David Murphy <dmurphy@phys.columbia.edu>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-			   /*  END LEGAL */
-
-#include <Grid/Grid_Eigen_Dense.h>
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/DomainWallEOFAFermion.h>
-#include <Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionCache.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class DomainWallEOFAFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplD/DomainWallEOFAFermionInstantiationWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplD/DomainWallEOFAFermionInstantiationWilsonImplD.cc
new file mode 120000
index 00000000..2f550a2b
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplD/DomainWallEOFAFermionInstantiationWilsonImplD.cc
@@ -0,0 +1 @@
+../DomainWallEOFAFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplD/MobiusEOFAFermionInstantiationWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplD/MobiusEOFAFermionInstantiationWilsonImplD.cc
deleted file mode 100644
index ce7eaac9..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplD/MobiusEOFAFermionInstantiationWilsonImplD.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/MobiusEOFAFermion.cc
-
-Copyright (C) 2017
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: David Murphy <dmurphy@phys.columbia.edu>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-			   /*  END LEGAL */
-
-#include <Grid/Grid_Eigen_Dense.h>
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/MobiusEOFAFermion.h>
-#include <Grid/qcd/action/fermion/implementation/MobiusEOFAFermionImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/MobiusEOFAFermionCache.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class MobiusEOFAFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplD/MobiusEOFAFermionInstantiationWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplD/MobiusEOFAFermionInstantiationWilsonImplD.cc
new file mode 120000
index 00000000..7a8f1172
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplD/MobiusEOFAFermionInstantiationWilsonImplD.cc
@@ -0,0 +1 @@
+../MobiusEOFAFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplD/PartialFractionFermion5DInstantiationWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplD/PartialFractionFermion5DInstantiationWilsonImplD.cc
deleted file mode 100644
index 757719b6..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplD/PartialFractionFermion5DInstantiationWilsonImplD.cc
+++ /dev/null
@@ -1,39 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/PartialFractionFermion5D.cc
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/PartialFractionFermion5D.h>
-#include <Grid/qcd/action/fermion/implementation/PartialFractionFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
- 
-#include "impl.h"
-template class PartialFractionFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplD/PartialFractionFermion5DInstantiationWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplD/PartialFractionFermion5DInstantiationWilsonImplD.cc
new file mode 120000
index 00000000..7f4cea71
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplD/PartialFractionFermion5DInstantiationWilsonImplD.cc
@@ -0,0 +1 @@
+../PartialFractionFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplD/WilsonCloverFermionInstantiationWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplD/WilsonCloverFermionInstantiationWilsonImplD.cc
deleted file mode 100644
index af99dfb6..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplD/WilsonCloverFermionInstantiationWilsonImplD.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: ./lib/qcd/action/fermion/WilsonCloverFermion.cc
-
-    Copyright (C) 2017
-
-    Author: paboyle <paboyle@ph.ed.ac.uk>
-    Author: Guido Cossu <guido.cossu@ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-    *************************************************************************************/
-/*  END LEGAL */
-
-#include <Grid/Grid.h>
-#include <Grid/qcd/spin/Dirac.h>
-#include <Grid/qcd/action/fermion/WilsonCloverFermion.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonCloverFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonCloverFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplD/WilsonCloverFermionInstantiationWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplD/WilsonCloverFermionInstantiationWilsonImplD.cc
new file mode 120000
index 00000000..9cc05107
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplD/WilsonCloverFermionInstantiationWilsonImplD.cc
@@ -0,0 +1 @@
+../WilsonCloverFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplD/WilsonFermion5DInstantiationWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplD/WilsonFermion5DInstantiationWilsonImplD.cc
deleted file mode 100644
index 0dac989c..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplD/WilsonFermion5DInstantiationWilsonImplD.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplD/WilsonFermion5DInstantiationWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplD/WilsonFermion5DInstantiationWilsonImplD.cc
new file mode 120000
index 00000000..804d0884
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplD/WilsonFermion5DInstantiationWilsonImplD.cc
@@ -0,0 +1 @@
+../WilsonFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplD/WilsonFermionInstantiationWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplD/WilsonFermionInstantiationWilsonImplD.cc
deleted file mode 100644
index 6fd9c5ca..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplD/WilsonFermionInstantiationWilsonImplD.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplD/WilsonFermionInstantiationWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplD/WilsonFermionInstantiationWilsonImplD.cc
new file mode 120000
index 00000000..5f6ab65e
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplD/WilsonFermionInstantiationWilsonImplD.cc
@@ -0,0 +1 @@
+../WilsonFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplD/WilsonTMFermionInstantiationWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplD/WilsonTMFermionInstantiationWilsonImplD.cc
deleted file mode 100644
index adfa310c..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplD/WilsonTMFermionInstantiationWilsonImplD.cc
+++ /dev/null
@@ -1,37 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/WilsonTMFermion.cc
-
-    Copyright (C) 2015
-
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/WilsonTMFermion.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonTMFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonTMFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplD/WilsonTMFermionInstantiationWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplD/WilsonTMFermionInstantiationWilsonImplD.cc
new file mode 120000
index 00000000..d5789bcf
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplD/WilsonTMFermionInstantiationWilsonImplD.cc
@@ -0,0 +1 @@
+../WilsonTMFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplDF/CayleyFermion5DInstantiationWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplDF/CayleyFermion5DInstantiationWilsonImplDF.cc
deleted file mode 100644
index 5130db9c..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplDF/CayleyFermion5DInstantiationWilsonImplDF.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/CayleyFermion5DImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dcache.h>
-
-			   //#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dvec.h>
-			   //#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dgpu.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class CayleyFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplDF/CayleyFermion5DInstantiationWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplDF/CayleyFermion5DInstantiationWilsonImplDF.cc
new file mode 120000
index 00000000..cb1db625
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplDF/CayleyFermion5DInstantiationWilsonImplDF.cc
@@ -0,0 +1 @@
+../CayleyFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplDF/ContinuedFractionFermion5DInstantiationWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplDF/ContinuedFractionFermion5DInstantiationWilsonImplDF.cc
deleted file mode 100644
index ca0d6cea..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplDF/ContinuedFractionFermion5DInstantiationWilsonImplDF.cc
+++ /dev/null
@@ -1,38 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/ContinuedFractionFermion5D.cc
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/ContinuedFractionFermion5D.h>
-#include <Grid/qcd/action/fermion/implementation/ContinuedFractionFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class ContinuedFractionFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplDF/ContinuedFractionFermion5DInstantiationWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplDF/ContinuedFractionFermion5DInstantiationWilsonImplDF.cc
new file mode 120000
index 00000000..c2d4b8fc
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplDF/ContinuedFractionFermion5DInstantiationWilsonImplDF.cc
@@ -0,0 +1 @@
+../ContinuedFractionFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplDF/DomainWallEOFAFermionInstantiationWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplDF/DomainWallEOFAFermionInstantiationWilsonImplDF.cc
deleted file mode 100644
index f7198131..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplDF/DomainWallEOFAFermionInstantiationWilsonImplDF.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/DomainWallEOFAFermion.cc
-
-Copyright (C) 2017
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: David Murphy <dmurphy@phys.columbia.edu>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-			   /*  END LEGAL */
-
-#include <Grid/Grid_Eigen_Dense.h>
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/DomainWallEOFAFermion.h>
-#include <Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionCache.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class DomainWallEOFAFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplDF/DomainWallEOFAFermionInstantiationWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplDF/DomainWallEOFAFermionInstantiationWilsonImplDF.cc
new file mode 120000
index 00000000..2f550a2b
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplDF/DomainWallEOFAFermionInstantiationWilsonImplDF.cc
@@ -0,0 +1 @@
+../DomainWallEOFAFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplDF/MobiusEOFAFermionInstantiationWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplDF/MobiusEOFAFermionInstantiationWilsonImplDF.cc
deleted file mode 100644
index ce7eaac9..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplDF/MobiusEOFAFermionInstantiationWilsonImplDF.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/MobiusEOFAFermion.cc
-
-Copyright (C) 2017
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: David Murphy <dmurphy@phys.columbia.edu>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-			   /*  END LEGAL */
-
-#include <Grid/Grid_Eigen_Dense.h>
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/MobiusEOFAFermion.h>
-#include <Grid/qcd/action/fermion/implementation/MobiusEOFAFermionImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/MobiusEOFAFermionCache.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class MobiusEOFAFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplDF/MobiusEOFAFermionInstantiationWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplDF/MobiusEOFAFermionInstantiationWilsonImplDF.cc
new file mode 120000
index 00000000..7a8f1172
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplDF/MobiusEOFAFermionInstantiationWilsonImplDF.cc
@@ -0,0 +1 @@
+../MobiusEOFAFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplDF/PartialFractionFermion5DInstantiationWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplDF/PartialFractionFermion5DInstantiationWilsonImplDF.cc
deleted file mode 100644
index 757719b6..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplDF/PartialFractionFermion5DInstantiationWilsonImplDF.cc
+++ /dev/null
@@ -1,39 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/PartialFractionFermion5D.cc
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/PartialFractionFermion5D.h>
-#include <Grid/qcd/action/fermion/implementation/PartialFractionFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
- 
-#include "impl.h"
-template class PartialFractionFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplDF/PartialFractionFermion5DInstantiationWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplDF/PartialFractionFermion5DInstantiationWilsonImplDF.cc
new file mode 120000
index 00000000..7f4cea71
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplDF/PartialFractionFermion5DInstantiationWilsonImplDF.cc
@@ -0,0 +1 @@
+../PartialFractionFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplDF/WilsonCloverFermionInstantiationWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplDF/WilsonCloverFermionInstantiationWilsonImplDF.cc
deleted file mode 100644
index af99dfb6..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplDF/WilsonCloverFermionInstantiationWilsonImplDF.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: ./lib/qcd/action/fermion/WilsonCloverFermion.cc
-
-    Copyright (C) 2017
-
-    Author: paboyle <paboyle@ph.ed.ac.uk>
-    Author: Guido Cossu <guido.cossu@ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-    *************************************************************************************/
-/*  END LEGAL */
-
-#include <Grid/Grid.h>
-#include <Grid/qcd/spin/Dirac.h>
-#include <Grid/qcd/action/fermion/WilsonCloverFermion.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonCloverFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonCloverFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplDF/WilsonCloverFermionInstantiationWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplDF/WilsonCloverFermionInstantiationWilsonImplDF.cc
new file mode 120000
index 00000000..9cc05107
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplDF/WilsonCloverFermionInstantiationWilsonImplDF.cc
@@ -0,0 +1 @@
+../WilsonCloverFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplDF/WilsonFermion5DInstantiationWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplDF/WilsonFermion5DInstantiationWilsonImplDF.cc
deleted file mode 100644
index 0dac989c..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplDF/WilsonFermion5DInstantiationWilsonImplDF.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplDF/WilsonFermion5DInstantiationWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplDF/WilsonFermion5DInstantiationWilsonImplDF.cc
new file mode 120000
index 00000000..804d0884
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplDF/WilsonFermion5DInstantiationWilsonImplDF.cc
@@ -0,0 +1 @@
+../WilsonFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplDF/WilsonFermionInstantiationWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplDF/WilsonFermionInstantiationWilsonImplDF.cc
deleted file mode 100644
index 6fd9c5ca..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplDF/WilsonFermionInstantiationWilsonImplDF.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplDF/WilsonFermionInstantiationWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplDF/WilsonFermionInstantiationWilsonImplDF.cc
new file mode 120000
index 00000000..5f6ab65e
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplDF/WilsonFermionInstantiationWilsonImplDF.cc
@@ -0,0 +1 @@
+../WilsonFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplDF/WilsonTMFermionInstantiationWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplDF/WilsonTMFermionInstantiationWilsonImplDF.cc
deleted file mode 100644
index adfa310c..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplDF/WilsonTMFermionInstantiationWilsonImplDF.cc
+++ /dev/null
@@ -1,37 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/WilsonTMFermion.cc
-
-    Copyright (C) 2015
-
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/WilsonTMFermion.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonTMFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonTMFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplDF/WilsonTMFermionInstantiationWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplDF/WilsonTMFermionInstantiationWilsonImplDF.cc
new file mode 120000
index 00000000..d5789bcf
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplDF/WilsonTMFermionInstantiationWilsonImplDF.cc
@@ -0,0 +1 @@
+../WilsonTMFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplF/CayleyFermion5DInstantiationWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplF/CayleyFermion5DInstantiationWilsonImplF.cc
deleted file mode 100644
index 5130db9c..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplF/CayleyFermion5DInstantiationWilsonImplF.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/CayleyFermion5DImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dcache.h>
-
-			   //#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dvec.h>
-			   //#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dgpu.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class CayleyFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplF/CayleyFermion5DInstantiationWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplF/CayleyFermion5DInstantiationWilsonImplF.cc
new file mode 120000
index 00000000..cb1db625
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplF/CayleyFermion5DInstantiationWilsonImplF.cc
@@ -0,0 +1 @@
+../CayleyFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplF/ContinuedFractionFermion5DInstantiationWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplF/ContinuedFractionFermion5DInstantiationWilsonImplF.cc
deleted file mode 100644
index ca0d6cea..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplF/ContinuedFractionFermion5DInstantiationWilsonImplF.cc
+++ /dev/null
@@ -1,38 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/ContinuedFractionFermion5D.cc
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/ContinuedFractionFermion5D.h>
-#include <Grid/qcd/action/fermion/implementation/ContinuedFractionFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class ContinuedFractionFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplF/ContinuedFractionFermion5DInstantiationWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplF/ContinuedFractionFermion5DInstantiationWilsonImplF.cc
new file mode 120000
index 00000000..c2d4b8fc
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplF/ContinuedFractionFermion5DInstantiationWilsonImplF.cc
@@ -0,0 +1 @@
+../ContinuedFractionFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplF/DomainWallEOFAFermionInstantiationWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplF/DomainWallEOFAFermionInstantiationWilsonImplF.cc
deleted file mode 100644
index f7198131..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplF/DomainWallEOFAFermionInstantiationWilsonImplF.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/DomainWallEOFAFermion.cc
-
-Copyright (C) 2017
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: David Murphy <dmurphy@phys.columbia.edu>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-			   /*  END LEGAL */
-
-#include <Grid/Grid_Eigen_Dense.h>
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/DomainWallEOFAFermion.h>
-#include <Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionCache.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class DomainWallEOFAFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplF/DomainWallEOFAFermionInstantiationWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplF/DomainWallEOFAFermionInstantiationWilsonImplF.cc
new file mode 120000
index 00000000..2f550a2b
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplF/DomainWallEOFAFermionInstantiationWilsonImplF.cc
@@ -0,0 +1 @@
+../DomainWallEOFAFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplF/MobiusEOFAFermionInstantiationWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplF/MobiusEOFAFermionInstantiationWilsonImplF.cc
deleted file mode 100644
index ce7eaac9..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplF/MobiusEOFAFermionInstantiationWilsonImplF.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/MobiusEOFAFermion.cc
-
-Copyright (C) 2017
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: David Murphy <dmurphy@phys.columbia.edu>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-			   /*  END LEGAL */
-
-#include <Grid/Grid_Eigen_Dense.h>
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/MobiusEOFAFermion.h>
-#include <Grid/qcd/action/fermion/implementation/MobiusEOFAFermionImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/MobiusEOFAFermionCache.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class MobiusEOFAFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplF/MobiusEOFAFermionInstantiationWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplF/MobiusEOFAFermionInstantiationWilsonImplF.cc
new file mode 120000
index 00000000..7a8f1172
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplF/MobiusEOFAFermionInstantiationWilsonImplF.cc
@@ -0,0 +1 @@
+../MobiusEOFAFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplF/PartialFractionFermion5DInstantiationWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplF/PartialFractionFermion5DInstantiationWilsonImplF.cc
deleted file mode 100644
index 757719b6..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplF/PartialFractionFermion5DInstantiationWilsonImplF.cc
+++ /dev/null
@@ -1,39 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/PartialFractionFermion5D.cc
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/PartialFractionFermion5D.h>
-#include <Grid/qcd/action/fermion/implementation/PartialFractionFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
- 
-#include "impl.h"
-template class PartialFractionFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplF/PartialFractionFermion5DInstantiationWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplF/PartialFractionFermion5DInstantiationWilsonImplF.cc
new file mode 120000
index 00000000..7f4cea71
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplF/PartialFractionFermion5DInstantiationWilsonImplF.cc
@@ -0,0 +1 @@
+../PartialFractionFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplF/WilsonCloverFermionInstantiationWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplF/WilsonCloverFermionInstantiationWilsonImplF.cc
deleted file mode 100644
index af99dfb6..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplF/WilsonCloverFermionInstantiationWilsonImplF.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: ./lib/qcd/action/fermion/WilsonCloverFermion.cc
-
-    Copyright (C) 2017
-
-    Author: paboyle <paboyle@ph.ed.ac.uk>
-    Author: Guido Cossu <guido.cossu@ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-    *************************************************************************************/
-/*  END LEGAL */
-
-#include <Grid/Grid.h>
-#include <Grid/qcd/spin/Dirac.h>
-#include <Grid/qcd/action/fermion/WilsonCloverFermion.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonCloverFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonCloverFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplF/WilsonCloverFermionInstantiationWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplF/WilsonCloverFermionInstantiationWilsonImplF.cc
new file mode 120000
index 00000000..9cc05107
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplF/WilsonCloverFermionInstantiationWilsonImplF.cc
@@ -0,0 +1 @@
+../WilsonCloverFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplF/WilsonFermion5DInstantiationWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplF/WilsonFermion5DInstantiationWilsonImplF.cc
deleted file mode 100644
index 0dac989c..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplF/WilsonFermion5DInstantiationWilsonImplF.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplF/WilsonFermion5DInstantiationWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplF/WilsonFermion5DInstantiationWilsonImplF.cc
new file mode 120000
index 00000000..804d0884
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplF/WilsonFermion5DInstantiationWilsonImplF.cc
@@ -0,0 +1 @@
+../WilsonFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplF/WilsonFermionInstantiationWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplF/WilsonFermionInstantiationWilsonImplF.cc
deleted file mode 100644
index 6fd9c5ca..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplF/WilsonFermionInstantiationWilsonImplF.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplF/WilsonFermionInstantiationWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplF/WilsonFermionInstantiationWilsonImplF.cc
new file mode 120000
index 00000000..5f6ab65e
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplF/WilsonFermionInstantiationWilsonImplF.cc
@@ -0,0 +1 @@
+../WilsonFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplF/WilsonTMFermionInstantiationWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplF/WilsonTMFermionInstantiationWilsonImplF.cc
deleted file mode 100644
index adfa310c..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplF/WilsonTMFermionInstantiationWilsonImplF.cc
+++ /dev/null
@@ -1,37 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/WilsonTMFermion.cc
-
-    Copyright (C) 2015
-
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/WilsonTMFermion.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonTMFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonTMFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplF/WilsonTMFermionInstantiationWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplF/WilsonTMFermionInstantiationWilsonImplF.cc
new file mode 120000
index 00000000..d5789bcf
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplF/WilsonTMFermionInstantiationWilsonImplF.cc
@@ -0,0 +1 @@
+../WilsonTMFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplFH/CayleyFermion5DInstantiationWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplFH/CayleyFermion5DInstantiationWilsonImplFH.cc
deleted file mode 100644
index 5130db9c..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplFH/CayleyFermion5DInstantiationWilsonImplFH.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/CayleyFermion5DImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dcache.h>
-
-			   //#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dvec.h>
-			   //#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dgpu.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class CayleyFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplFH/CayleyFermion5DInstantiationWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplFH/CayleyFermion5DInstantiationWilsonImplFH.cc
new file mode 120000
index 00000000..cb1db625
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplFH/CayleyFermion5DInstantiationWilsonImplFH.cc
@@ -0,0 +1 @@
+../CayleyFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplFH/ContinuedFractionFermion5DInstantiationWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplFH/ContinuedFractionFermion5DInstantiationWilsonImplFH.cc
deleted file mode 100644
index ca0d6cea..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplFH/ContinuedFractionFermion5DInstantiationWilsonImplFH.cc
+++ /dev/null
@@ -1,38 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/ContinuedFractionFermion5D.cc
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/ContinuedFractionFermion5D.h>
-#include <Grid/qcd/action/fermion/implementation/ContinuedFractionFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class ContinuedFractionFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplFH/ContinuedFractionFermion5DInstantiationWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplFH/ContinuedFractionFermion5DInstantiationWilsonImplFH.cc
new file mode 120000
index 00000000..c2d4b8fc
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplFH/ContinuedFractionFermion5DInstantiationWilsonImplFH.cc
@@ -0,0 +1 @@
+../ContinuedFractionFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplFH/DomainWallEOFAFermionInstantiationWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplFH/DomainWallEOFAFermionInstantiationWilsonImplFH.cc
deleted file mode 100644
index f7198131..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplFH/DomainWallEOFAFermionInstantiationWilsonImplFH.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/DomainWallEOFAFermion.cc
-
-Copyright (C) 2017
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: David Murphy <dmurphy@phys.columbia.edu>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-			   /*  END LEGAL */
-
-#include <Grid/Grid_Eigen_Dense.h>
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/DomainWallEOFAFermion.h>
-#include <Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionCache.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class DomainWallEOFAFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplFH/DomainWallEOFAFermionInstantiationWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplFH/DomainWallEOFAFermionInstantiationWilsonImplFH.cc
new file mode 120000
index 00000000..2f550a2b
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplFH/DomainWallEOFAFermionInstantiationWilsonImplFH.cc
@@ -0,0 +1 @@
+../DomainWallEOFAFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplFH/MobiusEOFAFermionInstantiationWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplFH/MobiusEOFAFermionInstantiationWilsonImplFH.cc
deleted file mode 100644
index ce7eaac9..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplFH/MobiusEOFAFermionInstantiationWilsonImplFH.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/MobiusEOFAFermion.cc
-
-Copyright (C) 2017
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: David Murphy <dmurphy@phys.columbia.edu>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-			   /*  END LEGAL */
-
-#include <Grid/Grid_Eigen_Dense.h>
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/MobiusEOFAFermion.h>
-#include <Grid/qcd/action/fermion/implementation/MobiusEOFAFermionImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/MobiusEOFAFermionCache.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class MobiusEOFAFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplFH/MobiusEOFAFermionInstantiationWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplFH/MobiusEOFAFermionInstantiationWilsonImplFH.cc
new file mode 120000
index 00000000..7a8f1172
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplFH/MobiusEOFAFermionInstantiationWilsonImplFH.cc
@@ -0,0 +1 @@
+../MobiusEOFAFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplFH/PartialFractionFermion5DInstantiationWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplFH/PartialFractionFermion5DInstantiationWilsonImplFH.cc
deleted file mode 100644
index 757719b6..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplFH/PartialFractionFermion5DInstantiationWilsonImplFH.cc
+++ /dev/null
@@ -1,39 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/PartialFractionFermion5D.cc
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/PartialFractionFermion5D.h>
-#include <Grid/qcd/action/fermion/implementation/PartialFractionFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
- 
-#include "impl.h"
-template class PartialFractionFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplFH/PartialFractionFermion5DInstantiationWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplFH/PartialFractionFermion5DInstantiationWilsonImplFH.cc
new file mode 120000
index 00000000..7f4cea71
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplFH/PartialFractionFermion5DInstantiationWilsonImplFH.cc
@@ -0,0 +1 @@
+../PartialFractionFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplFH/WilsonCloverFermionInstantiationWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplFH/WilsonCloverFermionInstantiationWilsonImplFH.cc
deleted file mode 100644
index af99dfb6..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplFH/WilsonCloverFermionInstantiationWilsonImplFH.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: ./lib/qcd/action/fermion/WilsonCloverFermion.cc
-
-    Copyright (C) 2017
-
-    Author: paboyle <paboyle@ph.ed.ac.uk>
-    Author: Guido Cossu <guido.cossu@ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-    *************************************************************************************/
-/*  END LEGAL */
-
-#include <Grid/Grid.h>
-#include <Grid/qcd/spin/Dirac.h>
-#include <Grid/qcd/action/fermion/WilsonCloverFermion.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonCloverFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonCloverFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplFH/WilsonCloverFermionInstantiationWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplFH/WilsonCloverFermionInstantiationWilsonImplFH.cc
new file mode 120000
index 00000000..9cc05107
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplFH/WilsonCloverFermionInstantiationWilsonImplFH.cc
@@ -0,0 +1 @@
+../WilsonCloverFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplFH/WilsonFermion5DInstantiationWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplFH/WilsonFermion5DInstantiationWilsonImplFH.cc
deleted file mode 100644
index 0dac989c..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplFH/WilsonFermion5DInstantiationWilsonImplFH.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplFH/WilsonFermion5DInstantiationWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplFH/WilsonFermion5DInstantiationWilsonImplFH.cc
new file mode 120000
index 00000000..804d0884
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplFH/WilsonFermion5DInstantiationWilsonImplFH.cc
@@ -0,0 +1 @@
+../WilsonFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplFH/WilsonFermionInstantiationWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplFH/WilsonFermionInstantiationWilsonImplFH.cc
deleted file mode 100644
index 6fd9c5ca..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplFH/WilsonFermionInstantiationWilsonImplFH.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplFH/WilsonFermionInstantiationWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplFH/WilsonFermionInstantiationWilsonImplFH.cc
new file mode 120000
index 00000000..5f6ab65e
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplFH/WilsonFermionInstantiationWilsonImplFH.cc
@@ -0,0 +1 @@
+../WilsonFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplFH/WilsonTMFermionInstantiationWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplFH/WilsonTMFermionInstantiationWilsonImplFH.cc
deleted file mode 100644
index adfa310c..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonImplFH/WilsonTMFermionInstantiationWilsonImplFH.cc
+++ /dev/null
@@ -1,37 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/WilsonTMFermion.cc
-
-    Copyright (C) 2015
-
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/WilsonTMFermion.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonTMFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonTMFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonImplFH/WilsonTMFermionInstantiationWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/WilsonImplFH/WilsonTMFermionInstantiationWilsonImplFH.cc
new file mode 120000
index 00000000..d5789bcf
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonImplFH/WilsonTMFermionInstantiationWilsonImplFH.cc
@@ -0,0 +1 @@
+../WilsonTMFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonKernelsInstantiationGparity.cc.master b/Grid/qcd/action/fermion/instantiation/WilsonKernelsInstantiationGparity.cc.master
index 75f143cb..1aaf929b 100644
--- a/Grid/qcd/action/fermion/instantiation/WilsonKernelsInstantiationGparity.cc.master
+++ b/Grid/qcd/action/fermion/instantiation/WilsonKernelsInstantiationGparity.cc.master
@@ -40,33 +40,8 @@ NAMESPACE_BEGIN(Grid);
 
 #include "impl.h"
 
-// G-parity requires more specialised implementation.
-template <>
-void WilsonKernels<IMPLEMENTATION>::ContractConservedCurrentSiteFwd(const SitePropagator &q_in_1, 
-							  const SitePropagator &q_in_2, 
-							  SitePropagator &q_out, 
-							  DoubledGaugeFieldView &U,
-							  unsigned int sU, 
-							  unsigned int mu, 
-							  bool switch_sign)
-{			
-  assert(0);		
-}			
-template <>		
-void WilsonKernels<IMPLEMENTATION>::ContractConservedCurrentSiteBwd( const SitePropagator &q_in_1,
-							   const SitePropagator &q_in_2,
-							   SitePropagator &q_out,
-							   DoubledGaugeFieldView &U,
-							   unsigned int mu,
-							   unsigned int sU,
-							   bool switch_sign)
-{			
-  assert(0);		
-}
-
 HAND_SPECIALISE_GPARITY(IMPLEMENTATION);
 
-
 template class WilsonKernels<IMPLEMENTATION>; 
 
 
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplD/WilsonCloverFermionInstantiationWilsonTwoIndexAntiSymmetricImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplD/WilsonCloverFermionInstantiationWilsonTwoIndexAntiSymmetricImplD.cc
deleted file mode 100644
index af99dfb6..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplD/WilsonCloverFermionInstantiationWilsonTwoIndexAntiSymmetricImplD.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: ./lib/qcd/action/fermion/WilsonCloverFermion.cc
-
-    Copyright (C) 2017
-
-    Author: paboyle <paboyle@ph.ed.ac.uk>
-    Author: Guido Cossu <guido.cossu@ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-    *************************************************************************************/
-/*  END LEGAL */
-
-#include <Grid/Grid.h>
-#include <Grid/qcd/spin/Dirac.h>
-#include <Grid/qcd/action/fermion/WilsonCloverFermion.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonCloverFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonCloverFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplD/WilsonCloverFermionInstantiationWilsonTwoIndexAntiSymmetricImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplD/WilsonCloverFermionInstantiationWilsonTwoIndexAntiSymmetricImplD.cc
new file mode 120000
index 00000000..9cc05107
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplD/WilsonCloverFermionInstantiationWilsonTwoIndexAntiSymmetricImplD.cc
@@ -0,0 +1 @@
+../WilsonCloverFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplD/WilsonFermionInstantiationWilsonTwoIndexAntiSymmetricImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplD/WilsonFermionInstantiationWilsonTwoIndexAntiSymmetricImplD.cc
deleted file mode 100644
index 6fd9c5ca..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplD/WilsonFermionInstantiationWilsonTwoIndexAntiSymmetricImplD.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplD/WilsonFermionInstantiationWilsonTwoIndexAntiSymmetricImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplD/WilsonFermionInstantiationWilsonTwoIndexAntiSymmetricImplD.cc
new file mode 120000
index 00000000..5f6ab65e
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplD/WilsonFermionInstantiationWilsonTwoIndexAntiSymmetricImplD.cc
@@ -0,0 +1 @@
+../WilsonFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplD/WilsonTMFermionInstantiationWilsonTwoIndexAntiSymmetricImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplD/WilsonTMFermionInstantiationWilsonTwoIndexAntiSymmetricImplD.cc
deleted file mode 100644
index adfa310c..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplD/WilsonTMFermionInstantiationWilsonTwoIndexAntiSymmetricImplD.cc
+++ /dev/null
@@ -1,37 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/WilsonTMFermion.cc
-
-    Copyright (C) 2015
-
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/WilsonTMFermion.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonTMFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonTMFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplD/WilsonTMFermionInstantiationWilsonTwoIndexAntiSymmetricImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplD/WilsonTMFermionInstantiationWilsonTwoIndexAntiSymmetricImplD.cc
new file mode 120000
index 00000000..d5789bcf
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplD/WilsonTMFermionInstantiationWilsonTwoIndexAntiSymmetricImplD.cc
@@ -0,0 +1 @@
+../WilsonTMFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplF/WilsonCloverFermionInstantiationWilsonTwoIndexAntiSymmetricImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplF/WilsonCloverFermionInstantiationWilsonTwoIndexAntiSymmetricImplF.cc
deleted file mode 100644
index af99dfb6..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplF/WilsonCloverFermionInstantiationWilsonTwoIndexAntiSymmetricImplF.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: ./lib/qcd/action/fermion/WilsonCloverFermion.cc
-
-    Copyright (C) 2017
-
-    Author: paboyle <paboyle@ph.ed.ac.uk>
-    Author: Guido Cossu <guido.cossu@ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-    *************************************************************************************/
-/*  END LEGAL */
-
-#include <Grid/Grid.h>
-#include <Grid/qcd/spin/Dirac.h>
-#include <Grid/qcd/action/fermion/WilsonCloverFermion.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonCloverFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonCloverFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplF/WilsonCloverFermionInstantiationWilsonTwoIndexAntiSymmetricImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplF/WilsonCloverFermionInstantiationWilsonTwoIndexAntiSymmetricImplF.cc
new file mode 120000
index 00000000..9cc05107
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplF/WilsonCloverFermionInstantiationWilsonTwoIndexAntiSymmetricImplF.cc
@@ -0,0 +1 @@
+../WilsonCloverFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplF/WilsonFermionInstantiationWilsonTwoIndexAntiSymmetricImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplF/WilsonFermionInstantiationWilsonTwoIndexAntiSymmetricImplF.cc
deleted file mode 100644
index 6fd9c5ca..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplF/WilsonFermionInstantiationWilsonTwoIndexAntiSymmetricImplF.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplF/WilsonFermionInstantiationWilsonTwoIndexAntiSymmetricImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplF/WilsonFermionInstantiationWilsonTwoIndexAntiSymmetricImplF.cc
new file mode 120000
index 00000000..5f6ab65e
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplF/WilsonFermionInstantiationWilsonTwoIndexAntiSymmetricImplF.cc
@@ -0,0 +1 @@
+../WilsonFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplF/WilsonTMFermionInstantiationWilsonTwoIndexAntiSymmetricImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplF/WilsonTMFermionInstantiationWilsonTwoIndexAntiSymmetricImplF.cc
deleted file mode 100644
index adfa310c..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplF/WilsonTMFermionInstantiationWilsonTwoIndexAntiSymmetricImplF.cc
+++ /dev/null
@@ -1,37 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/WilsonTMFermion.cc
-
-    Copyright (C) 2015
-
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/WilsonTMFermion.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonTMFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonTMFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplF/WilsonTMFermionInstantiationWilsonTwoIndexAntiSymmetricImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplF/WilsonTMFermionInstantiationWilsonTwoIndexAntiSymmetricImplF.cc
new file mode 120000
index 00000000..d5789bcf
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplF/WilsonTMFermionInstantiationWilsonTwoIndexAntiSymmetricImplF.cc
@@ -0,0 +1 @@
+../WilsonTMFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplD/WilsonCloverFermionInstantiationWilsonTwoIndexSymmetricImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplD/WilsonCloverFermionInstantiationWilsonTwoIndexSymmetricImplD.cc
deleted file mode 100644
index af99dfb6..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplD/WilsonCloverFermionInstantiationWilsonTwoIndexSymmetricImplD.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: ./lib/qcd/action/fermion/WilsonCloverFermion.cc
-
-    Copyright (C) 2017
-
-    Author: paboyle <paboyle@ph.ed.ac.uk>
-    Author: Guido Cossu <guido.cossu@ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-    *************************************************************************************/
-/*  END LEGAL */
-
-#include <Grid/Grid.h>
-#include <Grid/qcd/spin/Dirac.h>
-#include <Grid/qcd/action/fermion/WilsonCloverFermion.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonCloverFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonCloverFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplD/WilsonCloverFermionInstantiationWilsonTwoIndexSymmetricImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplD/WilsonCloverFermionInstantiationWilsonTwoIndexSymmetricImplD.cc
new file mode 120000
index 00000000..9cc05107
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplD/WilsonCloverFermionInstantiationWilsonTwoIndexSymmetricImplD.cc
@@ -0,0 +1 @@
+../WilsonCloverFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplD/WilsonFermionInstantiationWilsonTwoIndexSymmetricImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplD/WilsonFermionInstantiationWilsonTwoIndexSymmetricImplD.cc
deleted file mode 100644
index 6fd9c5ca..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplD/WilsonFermionInstantiationWilsonTwoIndexSymmetricImplD.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplD/WilsonFermionInstantiationWilsonTwoIndexSymmetricImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplD/WilsonFermionInstantiationWilsonTwoIndexSymmetricImplD.cc
new file mode 120000
index 00000000..5f6ab65e
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplD/WilsonFermionInstantiationWilsonTwoIndexSymmetricImplD.cc
@@ -0,0 +1 @@
+../WilsonFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplD/WilsonTMFermionInstantiationWilsonTwoIndexSymmetricImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplD/WilsonTMFermionInstantiationWilsonTwoIndexSymmetricImplD.cc
deleted file mode 100644
index adfa310c..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplD/WilsonTMFermionInstantiationWilsonTwoIndexSymmetricImplD.cc
+++ /dev/null
@@ -1,37 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/WilsonTMFermion.cc
-
-    Copyright (C) 2015
-
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/WilsonTMFermion.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonTMFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonTMFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplD/WilsonTMFermionInstantiationWilsonTwoIndexSymmetricImplD.cc b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplD/WilsonTMFermionInstantiationWilsonTwoIndexSymmetricImplD.cc
new file mode 120000
index 00000000..d5789bcf
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplD/WilsonTMFermionInstantiationWilsonTwoIndexSymmetricImplD.cc
@@ -0,0 +1 @@
+../WilsonTMFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplF/WilsonCloverFermionInstantiationWilsonTwoIndexSymmetricImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplF/WilsonCloverFermionInstantiationWilsonTwoIndexSymmetricImplF.cc
deleted file mode 100644
index af99dfb6..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplF/WilsonCloverFermionInstantiationWilsonTwoIndexSymmetricImplF.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: ./lib/qcd/action/fermion/WilsonCloverFermion.cc
-
-    Copyright (C) 2017
-
-    Author: paboyle <paboyle@ph.ed.ac.uk>
-    Author: Guido Cossu <guido.cossu@ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-    *************************************************************************************/
-/*  END LEGAL */
-
-#include <Grid/Grid.h>
-#include <Grid/qcd/spin/Dirac.h>
-#include <Grid/qcd/action/fermion/WilsonCloverFermion.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonCloverFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonCloverFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplF/WilsonCloverFermionInstantiationWilsonTwoIndexSymmetricImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplF/WilsonCloverFermionInstantiationWilsonTwoIndexSymmetricImplF.cc
new file mode 120000
index 00000000..9cc05107
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplF/WilsonCloverFermionInstantiationWilsonTwoIndexSymmetricImplF.cc
@@ -0,0 +1 @@
+../WilsonCloverFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplF/WilsonFermionInstantiationWilsonTwoIndexSymmetricImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplF/WilsonFermionInstantiationWilsonTwoIndexSymmetricImplF.cc
deleted file mode 100644
index 6fd9c5ca..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplF/WilsonFermionInstantiationWilsonTwoIndexSymmetricImplF.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplF/WilsonFermionInstantiationWilsonTwoIndexSymmetricImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplF/WilsonFermionInstantiationWilsonTwoIndexSymmetricImplF.cc
new file mode 120000
index 00000000..5f6ab65e
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplF/WilsonFermionInstantiationWilsonTwoIndexSymmetricImplF.cc
@@ -0,0 +1 @@
+../WilsonFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplF/WilsonTMFermionInstantiationWilsonTwoIndexSymmetricImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplF/WilsonTMFermionInstantiationWilsonTwoIndexSymmetricImplF.cc
deleted file mode 100644
index adfa310c..00000000
--- a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplF/WilsonTMFermionInstantiationWilsonTwoIndexSymmetricImplF.cc
+++ /dev/null
@@ -1,37 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/WilsonTMFermion.cc
-
-    Copyright (C) 2015
-
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/WilsonTMFermion.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonTMFermionImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonTMFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplF/WilsonTMFermionInstantiationWilsonTwoIndexSymmetricImplF.cc b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplF/WilsonTMFermionInstantiationWilsonTwoIndexSymmetricImplF.cc
new file mode 120000
index 00000000..d5789bcf
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplF/WilsonTMFermionInstantiationWilsonTwoIndexSymmetricImplF.cc
@@ -0,0 +1 @@
+../WilsonTMFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/CayleyFermion5DInstantiationZWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/CayleyFermion5DInstantiationZWilsonImplD.cc
deleted file mode 100644
index 5130db9c..00000000
--- a/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/CayleyFermion5DInstantiationZWilsonImplD.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/CayleyFermion5DImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dcache.h>
-
-			   //#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dvec.h>
-			   //#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dgpu.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class CayleyFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/CayleyFermion5DInstantiationZWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/CayleyFermion5DInstantiationZWilsonImplD.cc
new file mode 120000
index 00000000..cb1db625
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/CayleyFermion5DInstantiationZWilsonImplD.cc
@@ -0,0 +1 @@
+../CayleyFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/ContinuedFractionFermion5DInstantiationZWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/ContinuedFractionFermion5DInstantiationZWilsonImplD.cc
deleted file mode 100644
index ca0d6cea..00000000
--- a/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/ContinuedFractionFermion5DInstantiationZWilsonImplD.cc
+++ /dev/null
@@ -1,38 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/ContinuedFractionFermion5D.cc
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/ContinuedFractionFermion5D.h>
-#include <Grid/qcd/action/fermion/implementation/ContinuedFractionFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class ContinuedFractionFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/ContinuedFractionFermion5DInstantiationZWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/ContinuedFractionFermion5DInstantiationZWilsonImplD.cc
new file mode 120000
index 00000000..c2d4b8fc
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/ContinuedFractionFermion5DInstantiationZWilsonImplD.cc
@@ -0,0 +1 @@
+../ContinuedFractionFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/DomainWallEOFAFermionInstantiationZWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/DomainWallEOFAFermionInstantiationZWilsonImplD.cc
deleted file mode 100644
index f7198131..00000000
--- a/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/DomainWallEOFAFermionInstantiationZWilsonImplD.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/DomainWallEOFAFermion.cc
-
-Copyright (C) 2017
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: David Murphy <dmurphy@phys.columbia.edu>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-			   /*  END LEGAL */
-
-#include <Grid/Grid_Eigen_Dense.h>
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/DomainWallEOFAFermion.h>
-#include <Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionCache.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class DomainWallEOFAFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/DomainWallEOFAFermionInstantiationZWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/DomainWallEOFAFermionInstantiationZWilsonImplD.cc
new file mode 120000
index 00000000..2f550a2b
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/DomainWallEOFAFermionInstantiationZWilsonImplD.cc
@@ -0,0 +1 @@
+../DomainWallEOFAFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/MobiusEOFAFermionInstantiationZWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/MobiusEOFAFermionInstantiationZWilsonImplD.cc
deleted file mode 100644
index ce7eaac9..00000000
--- a/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/MobiusEOFAFermionInstantiationZWilsonImplD.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/MobiusEOFAFermion.cc
-
-Copyright (C) 2017
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: David Murphy <dmurphy@phys.columbia.edu>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-			   /*  END LEGAL */
-
-#include <Grid/Grid_Eigen_Dense.h>
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/MobiusEOFAFermion.h>
-#include <Grid/qcd/action/fermion/implementation/MobiusEOFAFermionImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/MobiusEOFAFermionCache.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class MobiusEOFAFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/MobiusEOFAFermionInstantiationZWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/MobiusEOFAFermionInstantiationZWilsonImplD.cc
new file mode 120000
index 00000000..7a8f1172
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/MobiusEOFAFermionInstantiationZWilsonImplD.cc
@@ -0,0 +1 @@
+../MobiusEOFAFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/PartialFractionFermion5DInstantiationZWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/PartialFractionFermion5DInstantiationZWilsonImplD.cc
deleted file mode 100644
index 757719b6..00000000
--- a/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/PartialFractionFermion5DInstantiationZWilsonImplD.cc
+++ /dev/null
@@ -1,39 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/PartialFractionFermion5D.cc
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/PartialFractionFermion5D.h>
-#include <Grid/qcd/action/fermion/implementation/PartialFractionFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
- 
-#include "impl.h"
-template class PartialFractionFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/PartialFractionFermion5DInstantiationZWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/PartialFractionFermion5DInstantiationZWilsonImplD.cc
new file mode 120000
index 00000000..7f4cea71
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/PartialFractionFermion5DInstantiationZWilsonImplD.cc
@@ -0,0 +1 @@
+../PartialFractionFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/WilsonFermion5DInstantiationZWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/WilsonFermion5DInstantiationZWilsonImplD.cc
deleted file mode 100644
index 0dac989c..00000000
--- a/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/WilsonFermion5DInstantiationZWilsonImplD.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/WilsonFermion5DInstantiationZWilsonImplD.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/WilsonFermion5DInstantiationZWilsonImplD.cc
new file mode 120000
index 00000000..804d0884
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/ZWilsonImplD/WilsonFermion5DInstantiationZWilsonImplD.cc
@@ -0,0 +1 @@
+../WilsonFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/CayleyFermion5DInstantiationZWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/CayleyFermion5DInstantiationZWilsonImplDF.cc
deleted file mode 100644
index 5130db9c..00000000
--- a/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/CayleyFermion5DInstantiationZWilsonImplDF.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/CayleyFermion5DImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dcache.h>
-
-			   //#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dvec.h>
-			   //#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dgpu.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class CayleyFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/CayleyFermion5DInstantiationZWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/CayleyFermion5DInstantiationZWilsonImplDF.cc
new file mode 120000
index 00000000..cb1db625
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/CayleyFermion5DInstantiationZWilsonImplDF.cc
@@ -0,0 +1 @@
+../CayleyFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/ContinuedFractionFermion5DInstantiationZWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/ContinuedFractionFermion5DInstantiationZWilsonImplDF.cc
deleted file mode 100644
index ca0d6cea..00000000
--- a/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/ContinuedFractionFermion5DInstantiationZWilsonImplDF.cc
+++ /dev/null
@@ -1,38 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/ContinuedFractionFermion5D.cc
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/ContinuedFractionFermion5D.h>
-#include <Grid/qcd/action/fermion/implementation/ContinuedFractionFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class ContinuedFractionFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/ContinuedFractionFermion5DInstantiationZWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/ContinuedFractionFermion5DInstantiationZWilsonImplDF.cc
new file mode 120000
index 00000000..c2d4b8fc
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/ContinuedFractionFermion5DInstantiationZWilsonImplDF.cc
@@ -0,0 +1 @@
+../ContinuedFractionFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/DomainWallEOFAFermionInstantiationZWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/DomainWallEOFAFermionInstantiationZWilsonImplDF.cc
deleted file mode 100644
index f7198131..00000000
--- a/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/DomainWallEOFAFermionInstantiationZWilsonImplDF.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/DomainWallEOFAFermion.cc
-
-Copyright (C) 2017
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: David Murphy <dmurphy@phys.columbia.edu>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-			   /*  END LEGAL */
-
-#include <Grid/Grid_Eigen_Dense.h>
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/DomainWallEOFAFermion.h>
-#include <Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionCache.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class DomainWallEOFAFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/DomainWallEOFAFermionInstantiationZWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/DomainWallEOFAFermionInstantiationZWilsonImplDF.cc
new file mode 120000
index 00000000..2f550a2b
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/DomainWallEOFAFermionInstantiationZWilsonImplDF.cc
@@ -0,0 +1 @@
+../DomainWallEOFAFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/MobiusEOFAFermionInstantiationZWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/MobiusEOFAFermionInstantiationZWilsonImplDF.cc
deleted file mode 100644
index ce7eaac9..00000000
--- a/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/MobiusEOFAFermionInstantiationZWilsonImplDF.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/MobiusEOFAFermion.cc
-
-Copyright (C) 2017
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: David Murphy <dmurphy@phys.columbia.edu>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-			   /*  END LEGAL */
-
-#include <Grid/Grid_Eigen_Dense.h>
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/MobiusEOFAFermion.h>
-#include <Grid/qcd/action/fermion/implementation/MobiusEOFAFermionImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/MobiusEOFAFermionCache.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class MobiusEOFAFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/MobiusEOFAFermionInstantiationZWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/MobiusEOFAFermionInstantiationZWilsonImplDF.cc
new file mode 120000
index 00000000..7a8f1172
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/MobiusEOFAFermionInstantiationZWilsonImplDF.cc
@@ -0,0 +1 @@
+../MobiusEOFAFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/PartialFractionFermion5DInstantiationZWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/PartialFractionFermion5DInstantiationZWilsonImplDF.cc
deleted file mode 100644
index 757719b6..00000000
--- a/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/PartialFractionFermion5DInstantiationZWilsonImplDF.cc
+++ /dev/null
@@ -1,39 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/PartialFractionFermion5D.cc
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/PartialFractionFermion5D.h>
-#include <Grid/qcd/action/fermion/implementation/PartialFractionFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
- 
-#include "impl.h"
-template class PartialFractionFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/PartialFractionFermion5DInstantiationZWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/PartialFractionFermion5DInstantiationZWilsonImplDF.cc
new file mode 120000
index 00000000..7f4cea71
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/PartialFractionFermion5DInstantiationZWilsonImplDF.cc
@@ -0,0 +1 @@
+../PartialFractionFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/WilsonFermion5DInstantiationZWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/WilsonFermion5DInstantiationZWilsonImplDF.cc
deleted file mode 100644
index 0dac989c..00000000
--- a/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/WilsonFermion5DInstantiationZWilsonImplDF.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/WilsonFermion5DInstantiationZWilsonImplDF.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/WilsonFermion5DInstantiationZWilsonImplDF.cc
new file mode 120000
index 00000000..804d0884
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/WilsonFermion5DInstantiationZWilsonImplDF.cc
@@ -0,0 +1 @@
+../WilsonFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/CayleyFermion5DInstantiationZWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/CayleyFermion5DInstantiationZWilsonImplF.cc
deleted file mode 100644
index 5130db9c..00000000
--- a/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/CayleyFermion5DInstantiationZWilsonImplF.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/CayleyFermion5DImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dcache.h>
-
-			   //#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dvec.h>
-			   //#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dgpu.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class CayleyFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/CayleyFermion5DInstantiationZWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/CayleyFermion5DInstantiationZWilsonImplF.cc
new file mode 120000
index 00000000..cb1db625
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/CayleyFermion5DInstantiationZWilsonImplF.cc
@@ -0,0 +1 @@
+../CayleyFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/ContinuedFractionFermion5DInstantiationZWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/ContinuedFractionFermion5DInstantiationZWilsonImplF.cc
deleted file mode 100644
index ca0d6cea..00000000
--- a/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/ContinuedFractionFermion5DInstantiationZWilsonImplF.cc
+++ /dev/null
@@ -1,38 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/ContinuedFractionFermion5D.cc
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/ContinuedFractionFermion5D.h>
-#include <Grid/qcd/action/fermion/implementation/ContinuedFractionFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class ContinuedFractionFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/ContinuedFractionFermion5DInstantiationZWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/ContinuedFractionFermion5DInstantiationZWilsonImplF.cc
new file mode 120000
index 00000000..c2d4b8fc
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/ContinuedFractionFermion5DInstantiationZWilsonImplF.cc
@@ -0,0 +1 @@
+../ContinuedFractionFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/DomainWallEOFAFermionInstantiationZWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/DomainWallEOFAFermionInstantiationZWilsonImplF.cc
deleted file mode 100644
index f7198131..00000000
--- a/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/DomainWallEOFAFermionInstantiationZWilsonImplF.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/DomainWallEOFAFermion.cc
-
-Copyright (C) 2017
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: David Murphy <dmurphy@phys.columbia.edu>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-			   /*  END LEGAL */
-
-#include <Grid/Grid_Eigen_Dense.h>
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/DomainWallEOFAFermion.h>
-#include <Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionCache.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class DomainWallEOFAFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/DomainWallEOFAFermionInstantiationZWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/DomainWallEOFAFermionInstantiationZWilsonImplF.cc
new file mode 120000
index 00000000..2f550a2b
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/DomainWallEOFAFermionInstantiationZWilsonImplF.cc
@@ -0,0 +1 @@
+../DomainWallEOFAFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/MobiusEOFAFermionInstantiationZWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/MobiusEOFAFermionInstantiationZWilsonImplF.cc
deleted file mode 100644
index ce7eaac9..00000000
--- a/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/MobiusEOFAFermionInstantiationZWilsonImplF.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/MobiusEOFAFermion.cc
-
-Copyright (C) 2017
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: David Murphy <dmurphy@phys.columbia.edu>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-			   /*  END LEGAL */
-
-#include <Grid/Grid_Eigen_Dense.h>
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/MobiusEOFAFermion.h>
-#include <Grid/qcd/action/fermion/implementation/MobiusEOFAFermionImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/MobiusEOFAFermionCache.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class MobiusEOFAFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/MobiusEOFAFermionInstantiationZWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/MobiusEOFAFermionInstantiationZWilsonImplF.cc
new file mode 120000
index 00000000..7a8f1172
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/MobiusEOFAFermionInstantiationZWilsonImplF.cc
@@ -0,0 +1 @@
+../MobiusEOFAFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/PartialFractionFermion5DInstantiationZWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/PartialFractionFermion5DInstantiationZWilsonImplF.cc
deleted file mode 100644
index 757719b6..00000000
--- a/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/PartialFractionFermion5DInstantiationZWilsonImplF.cc
+++ /dev/null
@@ -1,39 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/PartialFractionFermion5D.cc
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/PartialFractionFermion5D.h>
-#include <Grid/qcd/action/fermion/implementation/PartialFractionFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
- 
-#include "impl.h"
-template class PartialFractionFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/PartialFractionFermion5DInstantiationZWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/PartialFractionFermion5DInstantiationZWilsonImplF.cc
new file mode 120000
index 00000000..7f4cea71
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/PartialFractionFermion5DInstantiationZWilsonImplF.cc
@@ -0,0 +1 @@
+../PartialFractionFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/WilsonFermion5DInstantiationZWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/WilsonFermion5DInstantiationZWilsonImplF.cc
deleted file mode 100644
index 0dac989c..00000000
--- a/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/WilsonFermion5DInstantiationZWilsonImplF.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/WilsonFermion5DInstantiationZWilsonImplF.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/WilsonFermion5DInstantiationZWilsonImplF.cc
new file mode 120000
index 00000000..804d0884
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/ZWilsonImplF/WilsonFermion5DInstantiationZWilsonImplF.cc
@@ -0,0 +1 @@
+../WilsonFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/CayleyFermion5DInstantiationZWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/CayleyFermion5DInstantiationZWilsonImplFH.cc
deleted file mode 100644
index 5130db9c..00000000
--- a/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/CayleyFermion5DInstantiationZWilsonImplFH.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/CayleyFermion5DImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dcache.h>
-
-			   //#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dvec.h>
-			   //#include <Grid/qcd/action/fermion/implementation/CayleyFermion5Dgpu.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class CayleyFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/CayleyFermion5DInstantiationZWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/CayleyFermion5DInstantiationZWilsonImplFH.cc
new file mode 120000
index 00000000..cb1db625
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/CayleyFermion5DInstantiationZWilsonImplFH.cc
@@ -0,0 +1 @@
+../CayleyFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/ContinuedFractionFermion5DInstantiationZWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/ContinuedFractionFermion5DInstantiationZWilsonImplFH.cc
deleted file mode 100644
index ca0d6cea..00000000
--- a/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/ContinuedFractionFermion5DInstantiationZWilsonImplFH.cc
+++ /dev/null
@@ -1,38 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/ContinuedFractionFermion5D.cc
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/ContinuedFractionFermion5D.h>
-#include <Grid/qcd/action/fermion/implementation/ContinuedFractionFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class ContinuedFractionFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/ContinuedFractionFermion5DInstantiationZWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/ContinuedFractionFermion5DInstantiationZWilsonImplFH.cc
new file mode 120000
index 00000000..c2d4b8fc
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/ContinuedFractionFermion5DInstantiationZWilsonImplFH.cc
@@ -0,0 +1 @@
+../ContinuedFractionFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/DomainWallEOFAFermionInstantiationZWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/DomainWallEOFAFermionInstantiationZWilsonImplFH.cc
deleted file mode 100644
index f7198131..00000000
--- a/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/DomainWallEOFAFermionInstantiationZWilsonImplFH.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/DomainWallEOFAFermion.cc
-
-Copyright (C) 2017
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: David Murphy <dmurphy@phys.columbia.edu>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-			   /*  END LEGAL */
-
-#include <Grid/Grid_Eigen_Dense.h>
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/DomainWallEOFAFermion.h>
-#include <Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionCache.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class DomainWallEOFAFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/DomainWallEOFAFermionInstantiationZWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/DomainWallEOFAFermionInstantiationZWilsonImplFH.cc
new file mode 120000
index 00000000..2f550a2b
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/DomainWallEOFAFermionInstantiationZWilsonImplFH.cc
@@ -0,0 +1 @@
+../DomainWallEOFAFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/MobiusEOFAFermionInstantiationZWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/MobiusEOFAFermionInstantiationZWilsonImplFH.cc
deleted file mode 100644
index ce7eaac9..00000000
--- a/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/MobiusEOFAFermionInstantiationZWilsonImplFH.cc
+++ /dev/null
@@ -1,44 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/MobiusEOFAFermion.cc
-
-Copyright (C) 2017
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: David Murphy <dmurphy@phys.columbia.edu>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-			   /*  END LEGAL */
-
-#include <Grid/Grid_Eigen_Dense.h>
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/MobiusEOFAFermion.h>
-#include <Grid/qcd/action/fermion/implementation/MobiusEOFAFermionImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/MobiusEOFAFermionCache.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class MobiusEOFAFermion<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/MobiusEOFAFermionInstantiationZWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/MobiusEOFAFermionInstantiationZWilsonImplFH.cc
new file mode 120000
index 00000000..7a8f1172
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/MobiusEOFAFermionInstantiationZWilsonImplFH.cc
@@ -0,0 +1 @@
+../MobiusEOFAFermionInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/PartialFractionFermion5DInstantiationZWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/PartialFractionFermion5DInstantiationZWilsonImplFH.cc
deleted file mode 100644
index 757719b6..00000000
--- a/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/PartialFractionFermion5DInstantiationZWilsonImplFH.cc
+++ /dev/null
@@ -1,39 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/qcd/action/fermion/PartialFractionFermion5D.cc
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/PartialFractionFermion5D.h>
-#include <Grid/qcd/action/fermion/implementation/PartialFractionFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
- 
-#include "impl.h"
-template class PartialFractionFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/PartialFractionFermion5DInstantiationZWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/PartialFractionFermion5DInstantiationZWilsonImplFH.cc
new file mode 120000
index 00000000..7f4cea71
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/PartialFractionFermion5DInstantiationZWilsonImplFH.cc
@@ -0,0 +1 @@
+../PartialFractionFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/WilsonFermion5DInstantiationZWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/WilsonFermion5DInstantiationZWilsonImplFH.cc
deleted file mode 100644
index 0dac989c..00000000
--- a/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/WilsonFermion5DInstantiationZWilsonImplFH.cc
+++ /dev/null
@@ -1,40 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonFermion5DImplementation.h>
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonFermion5D<IMPLEMENTATION>; 
-
-NAMESPACE_END(Grid);
-
diff --git a/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/WilsonFermion5DInstantiationZWilsonImplFH.cc b/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/WilsonFermion5DInstantiationZWilsonImplFH.cc
new file mode 120000
index 00000000..804d0884
--- /dev/null
+++ b/Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/WilsonFermion5DInstantiationZWilsonImplFH.cc
@@ -0,0 +1 @@
+../WilsonFermion5DInstantiation.cc.master
\ No newline at end of file
diff --git a/Grid/qcd/action/fermion/instantiation/generate_instantiations.sh b/Grid/qcd/action/fermion/instantiation/generate_instantiations.sh
index 330dcfa8..72a9eaf9 100755
--- a/Grid/qcd/action/fermion/instantiation/generate_instantiations.sh
+++ b/Grid/qcd/action/fermion/instantiation/generate_instantiations.sh
@@ -88,6 +88,7 @@ done
 CC_LIST=" \
   ImprovedStaggeredFermion5DInstantiation \
   ImprovedStaggeredFermionInstantiation \
+  NaiveStaggeredFermionInstantiation \
   StaggeredKernelsInstantiation "
 
 for impl in $STAG_IMPL_LIST
diff --git a/Grid/qcd/action/gauge/GaugeImplTypes.h b/Grid/qcd/action/gauge/GaugeImplTypes.h
index b9a5296d..9b7d5a60 100644
--- a/Grid/qcd/action/gauge/GaugeImplTypes.h
+++ b/Grid/qcd/action/gauge/GaugeImplTypes.h
@@ -86,9 +86,9 @@ public:
 
   // Move this elsewhere? FIXME
   static inline void AddLink(Field &U, LinkField &W, int mu) { // U[mu] += W
-    auto U_v = U.View();
-    auto W_v = W.View();
-    thread_for( ss, U.Grid()->oSites(), {
+    autoView(U_v,U,AcceleratorWrite);
+    autoView(W_v,W,AcceleratorRead);
+    accelerator_for( ss, U.Grid()->oSites(), 1, {
       U_v[ss](mu) = U_v[ss](mu) + W_v[ss]();
     });
   }
@@ -131,15 +131,14 @@ public:
     //static std::chrono::duration<double> diff;
 
     //auto start = std::chrono::high_resolution_clock::now();
-    auto U_v = U.View();
-    auto P_v = P.View();
-    thread_for(ss, P.Grid()->oSites(),{
+    autoView(U_v,U,AcceleratorWrite);
+    autoView(P_v,P,AcceleratorRead);
+    accelerator_for(ss, P.Grid()->oSites(),1,{
       for (int mu = 0; mu < Nd; mu++) {
         U_v[ss](mu) = ProjectOnGroup(Exponentiate(P_v[ss](mu), ep, Nexp) * U_v[ss](mu));
       }
     });
-    
-    //auto end = std::chrono::high_resolution_clock::now();
+   //auto end = std::chrono::high_resolution_clock::now();
    // diff += end - start;
    // std::cout << "Time to exponentiate matrix " << diff.count() << " s\n";
   }
diff --git a/Grid/qcd/action/gauge/GaugeImplementations.h b/Grid/qcd/action/gauge/GaugeImplementations.h
index a14aec1b..19bc5aa6 100644
--- a/Grid/qcd/action/gauge/GaugeImplementations.h
+++ b/Grid/qcd/action/gauge/GaugeImplementations.h
@@ -59,7 +59,7 @@ public:
   }
   static inline GaugeLinkField
   CovShiftIdentityBackward(const GaugeLinkField &Link, int mu) {
-    return Cshift(adj(Link), mu, -1);
+    return Cshift(closure(adj(Link)), mu, -1);
   }
   static inline GaugeLinkField
   CovShiftIdentityForward(const GaugeLinkField &Link, int mu) {
diff --git a/Grid/qcd/action/scalar/ScalarInteractionAction.h b/Grid/qcd/action/scalar/ScalarInteractionAction.h
index 3be84480..5a5f9251 100644
--- a/Grid/qcd/action/scalar/ScalarInteractionAction.h
+++ b/Grid/qcd/action/scalar/ScalarInteractionAction.h
@@ -89,8 +89,8 @@ public:
     action = (2.0 * Ndim + mass_square) * phisquared - lambda * phisquared * phisquared;
     
     
-    auto p_v = p.View();
-    auto action_v = action.View();
+    autoView( p_v      , p, CpuRead);
+    autoView( action_v , action, CpuWrite);
     for (int mu = 0; mu < Ndim; mu++)
     {
       //  pshift = Cshift(p, mu, +1);  // not efficient, implement with stencils
@@ -146,8 +146,8 @@ public:
     for (int point = 0; point < npoint; point++)
     {
 
-      auto p_v = p.View();
-      auto force_v = force.View();
+      autoView( p_v , p, CpuRead);
+      autoView( force_v , force, CpuWrite);
             
       int permute_type;
       StencilEntry *SE;
diff --git a/Grid/qcd/hmc/HMC_aggregate.h b/Grid/qcd/hmc/HMC_aggregate.h
index e4d2ce83..cb510953 100644
--- a/Grid/qcd/hmc/HMC_aggregate.h
+++ b/Grid/qcd/hmc/HMC_aggregate.h
@@ -39,6 +39,10 @@ directory
 #include <Grid/parallelIO/IldgIOtypes.h>
 #include <Grid/parallelIO/IldgIO.h>
 #include <Grid/parallelIO/NerscIO.h>
+#include <Grid/parallelIO/OpenQcdIO.h>
+#if !defined(GRID_COMMS_NONE)
+#include <Grid/parallelIO/OpenQcdIOChromaReference.h>
+#endif
 NAMESPACE_CHECK(Ildg);
 
 #include <Grid/qcd/hmc/checkpointers/CheckPointers.h>
diff --git a/Grid/qcd/modules/Registration.h b/Grid/qcd/modules/Registration.h
index ec28f020..28a9fdae 100644
--- a/Grid/qcd/modules/Registration.h
+++ b/Grid/qcd/modules/Registration.h
@@ -80,8 +80,9 @@ static Registrar<OneFlavourRatioEOFModule<FermionImplementationPolicy>,
 
 static Registrar< ConjugateGradientModule<WilsonFermionR::FermionField>,   
                   HMC_SolverModuleFactory<solver_string, WilsonFermionR::FermionField, Serialiser> > __CGWFmodXMLInit("ConjugateGradient"); 
+
 static Registrar< BiCGSTABModule<WilsonFermionR::FermionField>,   
-                  HMC_SolverModuleFactory<solver_string, WilsonFermionR::FermionField, Serialiser> > __CGWFmodXMLInit("BiCGSTAB"); 
+                  HMC_SolverModuleFactory<solver_string, WilsonFermionR::FermionField, Serialiser> > __BiCGWFmodXMLInit("BiCGSTAB"); 
 static Registrar< ConjugateResidualModule<WilsonFermionR::FermionField>,   
                   HMC_SolverModuleFactory<solver_string, WilsonFermionR::FermionField, Serialiser> > __CRWFmodXMLInit("ConjugateResidual"); 
 
diff --git a/Grid/qcd/smearing/GaugeConfiguration.h b/Grid/qcd/smearing/GaugeConfiguration.h
index f4d00c72..0ff7fc25 100644
--- a/Grid/qcd/smearing/GaugeConfiguration.h
+++ b/Grid/qcd/smearing/GaugeConfiguration.h
@@ -49,7 +49,7 @@ public:
 
 private:
   const unsigned int smearingLevels;
-  Smear_Stout<Gimpl> StoutSmearing;
+  Smear_Stout<Gimpl> *StoutSmearing;
   std::vector<GaugeField> SmearedSet;
 
   // Member functions
@@ -72,7 +72,7 @@ private:
       previous_u = *ThinLinks;
       for (int smearLvl = 0; smearLvl < smearingLevels; ++smearLvl)
       {
-        StoutSmearing.smear(SmearedSet[smearLvl], previous_u);
+        StoutSmearing->smear(SmearedSet[smearLvl], previous_u);
         previous_u = SmearedSet[smearLvl];
 
         // For debug purposes
@@ -93,7 +93,7 @@ private:
     GaugeLinkField SigmaKPrime_mu(grid);
     GaugeLinkField GaugeKmu(grid), Cmu(grid);
 
-    StoutSmearing.BaseSmear(C, GaugeK);
+    StoutSmearing->BaseSmear(C, GaugeK);
     SigmaK = Zero();
     iLambda = Zero();
 
@@ -107,7 +107,7 @@ private:
       pokeLorentz(SigmaK, SigmaKPrime_mu * e_iQ + adj(Cmu) * iLambda_mu, mu);
       pokeLorentz(iLambda, iLambda_mu, mu);
     }
-    StoutSmearing.derivative(SigmaK, iLambda,
+    StoutSmearing->derivative(SigmaK, iLambda,
                              GaugeK);  // derivative of SmearBase
     return SigmaK;
   }
@@ -144,14 +144,14 @@ private:
     // Exponential
     iQ2 = iQ * iQ;
     iQ3 = iQ * iQ2;
-    StoutSmearing.set_uw(u, w, iQ2, iQ3);
-    StoutSmearing.set_fj(f0, f1, f2, u, w);
+    StoutSmearing->set_uw(u, w, iQ2, iQ3);
+    StoutSmearing->set_fj(f0, f1, f2, u, w);
     e_iQ = f0 * unity + timesMinusI(f1) * iQ - f2 * iQ2;
 
     // Getting B1, B2, Gamma and Lambda
     // simplify this part, reduntant calculations in set_fj
-    xi0 = StoutSmearing.func_xi0(w);
-    xi1 = StoutSmearing.func_xi1(w);
+    xi0 = StoutSmearing->func_xi0(w);
+    xi1 = StoutSmearing->func_xi1(w);
     u2 = u * u;
     w2 = w * w;
     cosw = cos(w);
@@ -219,7 +219,7 @@ public:
   /* Standard constructor */
   SmearedConfiguration(GridCartesian* UGrid, unsigned int Nsmear,
                        Smear_Stout<Gimpl>& Stout)
-      : smearingLevels(Nsmear), StoutSmearing(Stout), ThinLinks(NULL)
+      : smearingLevels(Nsmear), StoutSmearing(&Stout), ThinLinks(NULL)
   {
     for (unsigned int i = 0; i < smearingLevels; ++i)
       SmearedSet.push_back(*(new GaugeField(UGrid)));
@@ -227,7 +227,7 @@ public:
 
   /*! For just thin links */
   SmearedConfiguration()
-    : smearingLevels(0), StoutSmearing(), SmearedSet(), ThinLinks(NULL) {}
+    : smearingLevels(0), StoutSmearing(nullptr), SmearedSet(), ThinLinks(NULL) {}
 
   // attach the smeared routines to the thin links U and fill the smeared set
   void set_Field(GaugeField &U)
diff --git a/Grid/qcd/utils/A2Autils.h b/Grid/qcd/utils/A2Autils.h
index c7c7d329..b63d8571 100644
--- a/Grid/qcd/utils/A2Autils.h
+++ b/Grid/qcd/utils/A2Autils.h
@@ -185,13 +185,14 @@ void A2Autils<FImpl>::MesonField(TensorType &mat,
 
 	for(int i=0;i<Lblock;i++){
 
-	  auto lhs_v = lhs_wi[i].View();
+	  // Recreate view potentially expensive outside fo UVM mode
+	  autoView(lhs_v,lhs_wi[i],CpuRead);
 	  auto left = conjugate(lhs_v[ss]);
-
 	  for(int j=0;j<Rblock;j++){
 
 	    SpinMatrix_v vv;
-	    auto rhs_v = rhs_vj[j].View();
+	    // Recreate view potentially expensive outside fo UVM mode
+	    autoView(rhs_v,rhs_vj[j],CpuRead);
 	    auto right = rhs_v[ss];
 	    for(int s1=0;s1<Ns;s1++){
 	    for(int s2=0;s2<Ns;s2++){
@@ -204,11 +205,10 @@ void A2Autils<FImpl>::MesonField(TensorType &mat,
 	    int base = Nmom*i+Nmom*Lblock*j+Nmom*Lblock*Rblock*r;
 	    for ( int m=0;m<Nmom;m++){
 	      int idx = m+base;
-	      auto mom_v = mom[m].View();
+	      autoView(mom_v,mom[m],CpuRead);
 	      auto phase = mom_v[ss];
 	      mac(&lvSum[idx],&vv,&phase);
 	    }
-	  
 	  }
 	}
       }
@@ -371,7 +371,7 @@ void A2Autils<FImpl>::PionFieldXX(Eigen::Tensor<ComplexD,3> &mat,
 
 	for(int i=0;i<Lblock;i++){
 
-	  auto wi_v = wi[i].View();
+	  autoView(wi_v,wi[i],CpuRead);
 	  auto w = conjugate(wi_v[ss]);
 	  if (g5) {
 	    w()(2)(0) = - w()(2)(0);
@@ -383,7 +383,7 @@ void A2Autils<FImpl>::PionFieldXX(Eigen::Tensor<ComplexD,3> &mat,
 	  }
 	  for(int j=0;j<Rblock;j++){
 	    
-	    auto vj_v=vj[j].View();
+	    autoView(vj_v,vj[j],CpuRead);
 	    auto v  = vj_v[ss];
 	    auto vv = v()(0)(0);
 
@@ -518,12 +518,12 @@ void A2Autils<FImpl>::PionFieldWVmom(Eigen::Tensor<ComplexD,4> &mat,
 
 	for(int i=0;i<Lblock;i++){
 
-	  auto wi_v = wi[i].View();
+	  autoView(wi_v,wi[i],CpuRead);
 	  auto w = conjugate(wi_v[ss]);
 
 	  for(int j=0;j<Rblock;j++){
-	    
-	    auto vj_v = vj[j].View();
+
+	    autoView(vj_v,vj[j],CpuRead);
 	    auto v = vj_v[ss];
 
 	    auto vv = w()(0)(0) * v()(0)(0)// Gamma5 Dirac basis explicitly written out
@@ -544,7 +544,7 @@ void A2Autils<FImpl>::PionFieldWVmom(Eigen::Tensor<ComplexD,4> &mat,
 	    int base = Nmom*i+Nmom*Lblock*j+Nmom*Lblock*Rblock*r;
 	    for ( int m=0;m<Nmom;m++){
 	      int idx = m+base;
-	      auto mom_v = mom[m].View();
+	      autoView(mom_v,mom[m],CpuRead);
 	      auto phase = mom_v[ss];
 	      mac(&lvSum[idx],&vv,&phase()()());
 	    }
@@ -730,13 +730,13 @@ void A2Autils<FImpl>::AslashField(TensorType &mat,
 
             for(int i=0;i<Lblock;i++)
             {
-  	        auto wi_v = lhs_wi[i].View();
+  	        autoView(wi_v,lhs_wi[i],CpuRead);
                 auto left = conjugate(wi_v[ss]);
 
                 for(int j=0;j<Rblock;j++)
                 {
                     SpinMatrix_v vv;
-		    auto vj_v  = rhs_vj[j].View();
+		    autoView(vj_v,rhs_vj[j],CpuRead);
                     auto right = vj_v[ss];
 
                     for(int s1=0;s1<Ns;s1++)
@@ -752,8 +752,8 @@ void A2Autils<FImpl>::AslashField(TensorType &mat,
 
                     for ( int m=0;m<Nem;m++)
                     {
-  		        auto emB0_v = emB0[m].View();
-  		        auto emB1_v = emB1[m].View();
+  		        autoView(emB0_v,emB0[m],CpuRead);
+		        autoView(emB1_v,emB1[m],CpuRead);
                         int idx  = m+base;
                         auto b0  = emB0_v[ss];
                         auto b1  = emB1_v[ss];
@@ -1014,21 +1014,21 @@ A2Autils<FImpl>::ContractWWVV(std::vector<PropagatorField> &WWVV,
     for(int d_o=0;d_o<N_d;d_o+=d_unroll){
       for(int t=0;t<N_t;t++){
       for(int s=0;s<N_s;s++){
-  auto vs_v = vs[s].View();
-  auto tmp1 = vs_v[ss];
-  vobj tmp2 = Zero();
-  vobj tmp3 = Zero();
-  for(int d=d_o;d<MIN(d_o+d_unroll,N_d);d++){
-    auto vd_v = vd[d].View();
-    Scalar_v coeff = WW_sd(t,s,d);
-    tmp3 = conjugate(vd_v[ss]);
-    mac(&tmp2, &coeff, &tmp3);
-  }
+	autoView(vs_v,vs[s],CpuRead);
+	auto tmp1 = vs_v[ss];
+	vobj tmp2 = Zero();
+	vobj tmp3 = Zero();
+	for(int d=d_o;d<MIN(d_o+d_unroll,N_d);d++){
+	  autoView(vd_v,vd[d],CpuRead);
+	  Scalar_v coeff = WW_sd(t,s,d);
+	  tmp3 = conjugate(vd_v[ss]);
+	  mac(&tmp2, &coeff, &tmp3);
+	}
 
-  //////////////////////////
-  // Fast outer product of tmp1 with a sum of terms suppressed by d_unroll
-  //////////////////////////
-  OuterProductWWVV(WWVV[t], tmp1, tmp2, Ns, ss);
+	//////////////////////////
+	// Fast outer product of tmp1 with a sum of terms suppressed by d_unroll
+	//////////////////////////
+	OuterProductWWVV(WWVV[t], tmp1, tmp2, Ns, ss);
 
       }}
     }
@@ -1067,21 +1067,20 @@ A2Autils<FImpl>::ContractWWVV(std::vector<PropagatorField> &WWVV,
     thread_for(ss,grid->oSites(),{
       for(int d_o=0;d_o<N_d;d_o+=d_unroll){
         for(int s=0;s<N_s;s++){
-    auto vs_v = vs[s].View();
-    auto tmp1 = vs_v[ss];
-    vobj tmp2 = Zero();
-    vobj tmp3 = Zero();
-    for(int d=d_o;d<MIN(d_o+d_unroll,N_d);d++){
-      auto vd_v = vd[d].View();
-      Scalar_v coeff = buf(s,d);
-      tmp3 = conjugate(vd_v[ss]);
-      mac(&tmp2, &coeff, &tmp3);
-    }
-
-    //////////////////////////
-    // Fast outer product of tmp1 with a sum of terms suppressed by d_unroll
-    //////////////////////////
-    OuterProductWWVV(WWVV[t], tmp1, tmp2, Ns, ss);
+	  autoView(vs_v,vs[s],CpuRead);
+	  auto tmp1 = vs_v[ss];
+	  vobj tmp2 = Zero();
+	  vobj tmp3 = Zero();
+	  for(int d=d_o;d<MIN(d_o+d_unroll,N_d);d++){
+	    autoView(vd_v,vd[d],CpuRead);
+	    Scalar_v coeff = buf(s,d);
+	    tmp3 = conjugate(vd_v[ss]);
+	    mac(&tmp2, &coeff, &tmp3);
+	  }
+	  //////////////////////////
+	  // Fast outer product of tmp1 with a sum of terms suppressed by d_unroll
+	  //////////////////////////
+	  OuterProductWWVV(WWVV[t], tmp1, tmp2, Ns, ss);
       }}
     });
   }
@@ -1093,7 +1092,7 @@ inline void A2Autils<FImpl>::OuterProductWWVV(PropagatorField &WWVV,
                                              const vobj &rhs,
                                              const int Ns, const int ss)
 {
-  auto WWVV_v = WWVV.View();
+  autoView(WWVV_v,WWVV,CpuWrite);
   for (int s1 = 0; s1 < Ns; s1++){
     for (int s2 = 0; s2 < Ns; s2++){
       WWVV_v[ss]()(s1,s2)(0, 0) += lhs()(s1)(0) * rhs()(s2)(0);
@@ -1122,10 +1121,10 @@ void A2Autils<FImpl>::ContractFourQuarkColourDiagonal(const PropagatorField &WWV
 
   GridBase *grid = WWVV0.Grid();
 
-  auto WWVV0_v = WWVV0.View();
-  auto WWVV1_v = WWVV1.View();
-  auto O_trtr_v= O_trtr.View();
-  auto O_fig8_v= O_fig8.View();
+  autoView(WWVV0_v , WWVV0,CpuRead);
+  autoView(WWVV1_v , WWVV1,CpuRead);
+  autoView(O_trtr_v, O_trtr,CpuWrite);
+  autoView(O_fig8_v, O_fig8,CpuWrite);
   thread_for(ss,grid->oSites(),{
 
     typedef typename ComplexField::vector_object vobj;
@@ -1166,10 +1165,10 @@ void A2Autils<FImpl>::ContractFourQuarkColourMix(const PropagatorField &WWVV0,
 
   GridBase *grid = WWVV0.Grid();
 
-  auto WWVV0_v = WWVV0.View();
-  auto WWVV1_v = WWVV1.View();
-  auto O_trtr_v= O_trtr.View();
-  auto O_fig8_v= O_fig8.View();
+  autoView( WWVV0_v , WWVV0,CpuRead);
+  autoView( WWVV1_v , WWVV1,CpuRead);
+  autoView( O_trtr_v, O_trtr,CpuWrite);
+  autoView( O_fig8_v, O_fig8,CpuWrite);
 
   thread_for(ss,grid->oSites(),{
 
diff --git a/Grid/qcd/utils/BaryonUtils.h b/Grid/qcd/utils/BaryonUtils.h
index d65b9176..b268b684 100644
--- a/Grid/qcd/utils/BaryonUtils.h
+++ b/Grid/qcd/utils/BaryonUtils.h
@@ -7,6 +7,7 @@
  Copyright (C) 2019
  
  Author: Felix Erben <felix.erben@ed.ac.uk>
+ Author: Raoul Hodgson <raoul.hodgson@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
@@ -46,7 +47,7 @@ public:
   typedef typename SpinMatrixField::vector_object sobj;
 
   static const int epsilon[6][3] ;
-  static const Complex epsilon_sgn[6];
+  static const Real epsilon_sgn[6];
 
   private: 
   template <class mobj, class robj>
@@ -58,9 +59,12 @@ public:
 				 const Gamma GammaA_right,
 				 const Gamma GammaB_right,
 				 const int parity,
-				 const int * wick_contractions,
+				 const bool * wick_contractions,
   				 robj &result);
   public:
+  static void Wick_Contractions(std::string qi, 
+                 std::string qf, 
+                 bool* wick_contractions);
   static void ContractBaryons(const PropagatorField &q1_left,
 				 const PropagatorField &q2_left,
 				 const PropagatorField &q3_left,
@@ -68,8 +72,7 @@ public:
 				 const Gamma GammaB_left,
 				 const Gamma GammaA_right,
 				 const Gamma GammaB_right,
-				 const char * quarks_left,
-				 const char * quarks_right,
+				 const bool* wick_contractions,
 				 const int parity,
 				 ComplexField &baryon_corr);
   template <class mobj, class robj>
@@ -80,10 +83,59 @@ public:
 				 const Gamma GammaB_left,
 				 const Gamma GammaA_right,
 				 const Gamma GammaB_right,
-				 const char * quarks_left,
-				 const char * quarks_right,
+				 const bool* wick_contractions,
 				 const int parity,
+				 const int nt,
 				 robj &result);
+  private:
+  template <class mobj, class mobj2, class robj>
+  static void Baryon_Gamma_3pt_Group1_Site(
+           const mobj &Dq1_ti,
+           const mobj2 &Dq2_spec,
+           const mobj2 &Dq3_spec,
+           const mobj &Dq4_tf,
+                   const Gamma GammaJ,
+                   const Gamma GammaBi,
+                   const Gamma GammaBf,
+           int wick_contraction,
+           robj &result);
+
+  template <class mobj, class mobj2, class robj>
+  static void Baryon_Gamma_3pt_Group2_Site(
+           const mobj2 &Dq1_spec,
+           const mobj &Dq2_ti,
+           const mobj2 &Dq3_spec,
+           const mobj &Dq4_tf,
+                   const Gamma GammaJ,
+                   const Gamma GammaBi,
+                   const Gamma GammaBf,
+           int wick_contraction,
+           robj &result);
+
+  template <class mobj, class mobj2, class robj>
+  static void Baryon_Gamma_3pt_Group3_Site(
+           const mobj2 &Dq1_spec,
+           const mobj2 &Dq2_spec,
+           const mobj &Dq3_ti,
+           const mobj &Dq4_tf,
+                   const Gamma GammaJ,
+                   const Gamma GammaBi,
+                   const Gamma GammaBf,
+           int wick_contraction,
+           robj &result);
+  public:
+  template <class mobj>
+  static void Baryon_Gamma_3pt(
+           const PropagatorField &q_ti,
+           const mobj &Dq_spec1,
+           const mobj &Dq_spec2,
+           const PropagatorField &q_tf,
+           int group,
+           int wick_contraction,
+                   const Gamma GammaJ,
+                   const Gamma GammaBi,
+                   const Gamma GammaBf,
+           SpinMatrixField &stn_corr);
   private: 
   template <class mobj, class mobj2, class robj>
   static void Sigma_to_Nucleon_Q1_Eye_site(const mobj &Dq_loop,
@@ -151,101 +203,152 @@ public:
 
 template <class FImpl> 
 const int BaryonUtils<FImpl>::epsilon[6][3] = {{0,1,2},{1,2,0},{2,0,1},{0,2,1},{2,1,0},{1,0,2}};
-template <class FImpl> 
+/*template <class FImpl> 
 const Complex BaryonUtils<FImpl>::epsilon_sgn[6] = {Complex(1),
 						    Complex(1),
 						    Complex(1),
 						    Complex(-1),
 						    Complex(-1),
 						    Complex(-1)};
+*/
+template <class FImpl> 
+const Real BaryonUtils<FImpl>::epsilon_sgn[6] = {1.,1.,1.,-1.,-1.,-1.};
 
+//This is the old version
 template <class FImpl>
 template <class mobj, class robj>
 void BaryonUtils<FImpl>::baryon_site(const mobj &D1,
-						 const mobj &D2,
-						 const mobj &D3,
-				                 const Gamma GammaA_left,
-				                 const Gamma GammaB_left,
-				                 const Gamma GammaA_right,
-		                 		 const Gamma GammaB_right,
-						 const int parity,
-						 const int * wick_contraction,
-						 robj &result)
+                const mobj &D2,
+                const mobj &D3,
+                         const Gamma GammaA_i,
+                         const Gamma GammaB_i,
+                         const Gamma GammaA_f,
+                         const Gamma GammaB_f,
+                const int parity,
+                const bool * wick_contraction,
+                robj &result)
 {
 
-  Gamma g4(Gamma::Algebra::GammaT); //needed for parity P_\pm = 0.5*(1 \pm \gamma_4)
+    Gamma g4(Gamma::Algebra::GammaT); //needed for parity P_\pm = 0.5*(1 \pm \gamma_4)
+    
+    auto D1_GAi =  D1 * GammaA_i;
+    auto D1_GAi_g4 = D1_GAi * g4;
+    auto D1_GAi_P = 0.5*(D1_GAi + (Real)parity * D1_GAi_g4);
+    auto GAf_D1_GAi_P = GammaA_f * D1_GAi_P;
+    auto GBf_D1_GAi_P = GammaB_f * D1_GAi_P;
 
-    auto gD1a = GammaA_left * GammaA_right * D1;
-    auto gD1b = GammaA_left * g4 * GammaA_right * D1;
-    auto pD1 = 0.5* (gD1a + (double)parity * gD1b);
-    auto gD3 = GammaB_right * D3;
+    auto D2_GBi = D2 * GammaB_i;
+    auto GBf_D2_GBi = GammaB_f * D2_GBi;
+    auto GAf_D2_GBi = GammaA_f * D2_GBi;
 
-    for (int ie_left=0; ie_left < 6 ; ie_left++){
-      int a_left = epsilon[ie_left][0]; //a
-      int b_left = epsilon[ie_left][1]; //b
-      int c_left = epsilon[ie_left][2]; //c
-      for (int ie_right=0; ie_right < 6 ; ie_right++){
-        int a_right = epsilon[ie_right][0]; //a'
-        int b_right = epsilon[ie_right][1]; //b'
-        int c_right = epsilon[ie_right][2]; //c'
+    auto GBf_D3 = GammaB_f * D3;
+    auto GAf_D3 = GammaA_f * D3;
+
+    for (int ie_f=0; ie_f < 6 ; ie_f++){
+        int a_f = epsilon[ie_f][0]; //a
+        int b_f = epsilon[ie_f][1]; //b
+        int c_f = epsilon[ie_f][2]; //c
+    for (int ie_i=0; ie_i < 6 ; ie_i++){
+        int a_i = epsilon[ie_i][0]; //a'
+        int b_i = epsilon[ie_i][1]; //b'
+        int c_i = epsilon[ie_i][2]; //c'
+
+        Real ee = epsilon_sgn[ie_f] * epsilon_sgn[ie_i];
         //This is the \delta_{456}^{123} part
-	if (wick_contraction[0]){
-          auto D2g = D2 * GammaB_left;
-	  for (int alpha_right=0; alpha_right<Ns; alpha_right++){
-	  for (int beta_left=0; beta_left<Ns; beta_left++){
-	  for (int gamma_left=0; gamma_left<Ns; gamma_left++){
-	    result()()() += epsilon_sgn[ie_left] * epsilon_sgn[ie_right] * pD1()(gamma_left,gamma_left)(c_right,c_left)*D2g()(alpha_right,beta_left)(a_right,a_left)*gD3()(alpha_right,beta_left)(b_right,b_left);
-          }}}
-  	}	  
+        if (wick_contraction[0]){
+            for (int rho=0; rho<Ns; rho++){
+                auto GAf_D1_GAi_P_rr_cc = GAf_D1_GAi_P()(rho,rho)(c_f,c_i);
+                for (int alpha_f=0; alpha_f<Ns; alpha_f++){
+                for (int beta_i=0; beta_i<Ns; beta_i++){
+                    result()()() += ee  * GAf_D1_GAi_P_rr_cc
+                                        * D2_GBi    ()(alpha_f,beta_i)(a_f,a_i)
+                                        * GBf_D3    ()(alpha_f,beta_i)(b_f,b_i);
+                }}
+            }
+        }   
         //This is the \delta_{456}^{231} part
-	if (wick_contraction[1]){
-          auto pD1g = pD1 * GammaB_left;
-	  for (int alpha_right=0; alpha_right<Ns; alpha_right++){
-	  for (int beta_left=0; beta_left<Ns; beta_left++){
-	  for (int gamma_left=0; gamma_left<Ns; gamma_left++){
-	    result()()() += epsilon_sgn[ie_left] * epsilon_sgn[ie_right] * pD1g()(gamma_left,beta_left)(c_right,a_left)*D2()(alpha_right,beta_left)(a_right,b_left)*gD3()(alpha_right,gamma_left)(b_right,c_left);
-          }}}
-        }	  
+        if (wick_contraction[1]){
+            for (int rho=0; rho<Ns; rho++){
+            for (int alpha_f=0; alpha_f<Ns; alpha_f++){
+                auto D1_GAi_P_ar_ac = D1_GAi_P()(alpha_f,rho)(a_f,c_i);
+                for (int beta_i=0; beta_i<Ns; beta_i++){
+                    result()()() += ee  * D1_GAi_P_ar_ac
+                                        * GBf_D2_GBi    ()(alpha_f,beta_i)(b_f,a_i)
+                                        * GAf_D3        ()(rho,beta_i)(c_f,b_i);
+                }
+            }}
+        }   
         //This is the \delta_{456}^{312} part
-	if (wick_contraction[2]){
-          auto gD3g = gD3 * GammaB_left;
-	  for (int alpha_right=0; alpha_right<Ns; alpha_right++){
-	  for (int beta_left=0; beta_left<Ns; beta_left++){
-	  for (int gamma_left=0; gamma_left<Ns; gamma_left++){
-	    result()()() += epsilon_sgn[ie_left] * epsilon_sgn[ie_right] * pD1()(gamma_left,beta_left)(c_right,b_left)*D2()(alpha_right,gamma_left)(a_right,c_left)*gD3g()(alpha_right,beta_left)(b_right,a_left);
-          }}}
-        }	  
+        if (wick_contraction[2]){
+            for (int rho=0; rho<Ns; rho++){
+                for (int alpha_f=0; alpha_f<Ns; alpha_f++){
+                auto GBf_D1_GAi_P_ar_bc = GBf_D1_GAi_P()(alpha_f,rho)(b_f,c_i);
+                for (int beta_i=0; beta_i<Ns; beta_i++){
+                    result()()() += ee  * GBf_D1_GAi_P_ar_bc
+                                        * GAf_D2_GBi    ()(rho,beta_i)(c_f,a_i)
+                                        * D3            ()(alpha_f,beta_i)(a_f,b_i);
+                }
+            }}
+        }   
         //This is the \delta_{456}^{132} part
-	if (wick_contraction[3]){
-          auto gD3g = gD3 * GammaB_left;
-	  for (int alpha_right=0; alpha_right<Ns; alpha_right++){
-	  for (int beta_left=0; beta_left<Ns; beta_left++){
-	  for (int gamma_left=0; gamma_left<Ns; gamma_left++){
-	    result()()() -= epsilon_sgn[ie_left] * epsilon_sgn[ie_right] * pD1()(gamma_left,gamma_left)(c_right,c_left)*D2()(alpha_right,beta_left)(a_right,b_left)*gD3g()(alpha_right,beta_left)(b_right,a_left);
-          }}}
-        }	  
+        if (wick_contraction[3]){
+            for (int rho=0; rho<Ns; rho++){
+                auto GAf_D1_GAi_P_rr_cc = GAf_D1_GAi_P()(rho,rho)(c_f,c_i);
+                for (int alpha_f=0; alpha_f<Ns; alpha_f++){
+                for (int beta_i=0; beta_i<Ns; beta_i++){
+                    result()()() -= ee  * GAf_D1_GAi_P_rr_cc
+                                        * GBf_D2_GBi    ()(alpha_f,beta_i)(b_f,a_i)
+                                        * D3            ()(alpha_f,beta_i)(a_f,b_i);
+              }
+        }}
+        }   
         //This is the \delta_{456}^{321} part
-	if (wick_contraction[4]){
-          auto D2g = D2 * GammaB_left;
-	  for (int alpha_right=0; alpha_right<Ns; alpha_right++){
-	  for (int beta_left=0; beta_left<Ns; beta_left++){
-	  for (int gamma_left=0; gamma_left<Ns; gamma_left++){
-	    result()()() -= epsilon_sgn[ie_left] * epsilon_sgn[ie_right] * pD1()(gamma_left,beta_left)(c_right,b_left)*D2g()(alpha_right,beta_left)(a_right,a_left)*gD3()(alpha_right,gamma_left)(b_right,c_left);
-          }}}
-        }	  
+        if (wick_contraction[4]){
+            for (int rho=0; rho<Ns; rho++){
+            for (int alpha_f=0; alpha_f<Ns; alpha_f++){
+                auto GBf_D1_GAi_P_ar_bc = GBf_D1_GAi_P()(alpha_f,rho)(b_f,c_i);
+                for (int beta_i=0; beta_i<Ns; beta_i++){
+                    result()()() -= ee  * GBf_D1_GAi_P_ar_bc
+                                        * D2_GBi    ()(alpha_f,beta_i)(a_f,a_i)
+                                        * GAf_D3    ()(rho,beta_i)(c_f,b_i);
+                }
+            }}
+        }   
         //This is the \delta_{456}^{213} part
-	if (wick_contraction[5]){
-          auto pD1g = pD1 * GammaB_left;
-	  for (int alpha_right=0; alpha_right<Ns; alpha_right++){
-	  for (int beta_left=0; beta_left<Ns; beta_left++){
-	  for (int gamma_left=0; gamma_left<Ns; gamma_left++){
-	    result()()() -= epsilon_sgn[ie_left] * epsilon_sgn[ie_right] * pD1g()(gamma_left,beta_left)(c_right,a_left)*D2()(alpha_right,gamma_left)(a_right,c_left)*gD3()(alpha_right,beta_left)(b_right,b_left);
-          }}}
-        }	  
-      }
+        if (wick_contraction[5]){
+            for (int rho=0; rho<Ns; rho++){
+            for (int alpha_f=0; alpha_f<Ns; alpha_f++){
+                auto D1_GAi_P_ar_ac = D1_GAi_P()(alpha_f,rho)(a_f,c_i);
+                for (int beta_i=0; beta_i<Ns; beta_i++){
+                    result()()() -= ee  * D1_GAi_P_ar_ac
+                                        * GAf_D2_GBi    ()(rho,beta_i)(c_f,a_i)
+                                        * GBf_D3        ()(alpha_f,beta_i)(b_f,b_i);
+                }
+            }}
+        }
+    }}
+}
+
+/* Computes which wick contractions should be performed for a    *
+ * baryon 2pt function given the initial and finals state quark  *
+ * flavours.                                                     *
+ * The array wick_contractions must be of length 6               */
+template<class FImpl>
+void BaryonUtils<FImpl>::Wick_Contractions(std::string qi, std::string qf, bool* wick_contractions) {
+    const int epsilon[6][3] = {{0,1,2},{1,2,0},{2,0,1},{0,2,1},{2,1,0},{1,0,2}};
+    for (int ie=0; ie < 6 ; ie++) {
+        wick_contractions[ie] = (qi.size() == 3 && qf.size() == 3
+                                    && qi[0] == qf[epsilon[ie][0]] 
+                                    && qi[1] == qf[epsilon[ie][1]] 
+                                    && qi[2] == qf[epsilon[ie][2]]);
     }
 }
 
+/* The array wick_contractions must be of length 6. The order     * 
+ * corresponds to the to that shown in the Hadrons documentation  *
+ * at https://aportelli.github.io/Hadrons-doc/#/mcontraction      *
+ * This can be computed from the quark flavours using the         *
+ * Wick_Contractions function above                               */
 template<class FImpl>
 void BaryonUtils<FImpl>::ContractBaryons(const PropagatorField &q1_left,
 						 const PropagatorField &q2_left,
@@ -254,43 +357,61 @@ void BaryonUtils<FImpl>::ContractBaryons(const PropagatorField &q1_left,
 				                 const Gamma GammaB_left,
 				                 const Gamma GammaA_right,
 		                 		 const Gamma GammaB_right,
-						 const char * quarks_left,
-						 const char * quarks_right,
+						 const bool* wick_contractions,
 						 const int parity,
 						 ComplexField &baryon_corr)
 {
-  std::cout << "Contraction <" << quarks_right[0] << quarks_right[1] << quarks_right[2] << "|" << quarks_left[0] << quarks_left[1] << quarks_left[2] << ">" << std::endl;
-    std::cout << "GammaA (left) " << (GammaA_left.g) <<  std::endl;
-    std::cout << "GammaB (left) " << (GammaB_left.g) <<  std::endl;
-    std::cout << "GammaA (right) " << (GammaA_right.g) <<  std::endl;
-    std::cout << "GammaB (right) " << (GammaB_right.g) <<  std::endl;
+
+  assert(Ns==4 && "Baryon code only implemented for N_spin = 4");
+  assert(Nc==3 && "Baryon code only implemented for N_colour = 3");
+
+  std::cout << "GammaA (left) " << (GammaA_left.g) <<  std::endl;
+  std::cout << "GammaB (left) " << (GammaB_left.g) <<  std::endl;
+  std::cout << "GammaA (right) " << (GammaA_right.g) <<  std::endl;
+  std::cout << "GammaB (right) " << (GammaB_right.g) <<  std::endl;
  
   assert(parity==1 || parity == -1 && "Parity must be +1 or -1");
 
   GridBase *grid = q1_left.Grid();
+  
+  autoView(vbaryon_corr, baryon_corr,CpuWrite);
+  autoView( v1 , q1_left, CpuRead);
+  autoView( v2 , q2_left, CpuRead);
+  autoView( v3 , q3_left, CpuRead);
 
-  int wick_contraction[6];
-  for (int ie=0; ie < 6 ; ie++)
-    wick_contraction[ie] = (quarks_left[0] == quarks_right[epsilon[ie][0]] && quarks_left[1] == quarks_right[epsilon[ie][1]] && quarks_left[2] == quarks_right[epsilon[ie][2]]) ? 1 : 0;
-
-  auto vbaryon_corr= baryon_corr.View();
-  auto v1 = q1_left.View();
-  auto v2 = q2_left.View();
-  auto v3 = q3_left.View();
-
- // accelerator_for(ss, grid->oSites(), grid->Nsimd(), {
-  thread_for(ss,grid->oSites(),{
-  //for(int ss=0; ss < grid->oSites(); ss++){
+  Real bytes =0.;
+  bytes += grid->oSites() * (432.*sizeof(vComplex) + 126.*sizeof(int) + 36.*sizeof(Real));
+  for (int ie=0; ie < 6 ; ie++){
+    if(ie==0 or ie==3){
+       bytes += grid->oSites() * (4.*sizeof(int) + 4752.*sizeof(vComplex)) * wick_contractions[ie];
+    }
+    else{
+       bytes += grid->oSites() * (64.*sizeof(int) + 5184.*sizeof(vComplex)) * wick_contractions[ie];
+    }
+  }
+  Real t=0.;
+  t =-usecond();
 
+  accelerator_for(ss, grid->oSites(), grid->Nsimd(), {
     auto D1 = v1[ss];
     auto D2 = v2[ss];
     auto D3 = v3[ss];
-
     vobj result=Zero();
-    baryon_site(D1,D2,D3,GammaA_left,GammaB_left,GammaA_right,GammaB_right,parity,wick_contraction,result);
+    baryon_site(D1,D2,D3,GammaA_left,GammaB_left,GammaA_right,GammaB_right,parity,wick_contractions,result);
     vbaryon_corr[ss] = result; 
   }  );//end loop over lattice sites
+
+  t += usecond();
+
+  std::cout << std::setw(10) << bytes/t*1.0e6/1024/1024/1024 << " GB/s " << std::endl;
+
 }
+
+/* The array wick_contractions must be of length 6. The order     * 
+ * corresponds to the to that shown in the Hadrons documentation  *
+ * at https://aportelli.github.io/Hadrons-doc/#/mcontraction      *
+ * This can also be computed from the quark flavours using the    *
+ * Wick_Contractions function above                               */
 template <class FImpl>
 template <class mobj, class robj>
 void BaryonUtils<FImpl>::ContractBaryons_Sliced(const mobj &D1,
@@ -300,30 +421,363 @@ void BaryonUtils<FImpl>::ContractBaryons_Sliced(const mobj &D1,
 				                 const Gamma GammaB_left,
 				                 const Gamma GammaA_right,
 		                 		 const Gamma GammaB_right,
-						 const char * quarks_left,
-						 const char * quarks_right,
+						 const bool* wick_contractions,
 						 const int parity,
+						 const int nt,
 						 robj &result)
 {
-  std::cout << "Contraction <" << quarks_right[0] << quarks_right[1] << quarks_right[2] << "|" << quarks_left[0] << quarks_left[1] << quarks_left[2] << ">" << std::endl;
-    std::cout << "GammaA (left) " << (GammaA_left.g) <<  std::endl;
-    std::cout << "GammaB (left) " << (GammaB_left.g) <<  std::endl;
-    std::cout << "GammaA (right) " << (GammaA_right.g) <<  std::endl;
-    std::cout << "GammaB (right) " << (GammaB_right.g) <<  std::endl;
+
+  assert(Ns==4 && "Baryon code only implemented for N_spin = 4");
+  assert(Nc==3 && "Baryon code only implemented for N_colour = 3");
+
+  std::cout << "GammaA (left) " << (GammaA_left.g) <<  std::endl;
+  std::cout << "GammaB (left) " << (GammaB_left.g) <<  std::endl;
+  std::cout << "GammaA (right) " << (GammaA_right.g) <<  std::endl;
+  std::cout << "GammaB (right) " << (GammaB_right.g) <<  std::endl;
  
   assert(parity==1 || parity == -1 && "Parity must be +1 or -1");
 
-  int wick_contraction[6];
-  for (int ie=0; ie < 6 ; ie++)
-    wick_contraction[ie] = (quarks_left[0] == quarks_right[epsilon[ie][0]] && quarks_left[1] == quarks_right[epsilon[ie][1]] && quarks_left[2] == quarks_right[epsilon[ie][2]]) ? 1 : 0;
-
-     result=Zero();
-     baryon_site(D1,D2,D3,GammaA_left,GammaB_left,GammaA_right,GammaB_right,parity,wick_contraction,result);
+  for (int t=0; t<nt; t++) {
+    baryon_site(D1[t],D2[t],D3[t],GammaA_left,GammaB_left,GammaA_right,GammaB_right,parity,wick_contractions,result[t]);
+  }
 }
 
 /***********************************************************************
  * End of Baryon 2pt-function code.                                    *
  *                                                                     *
+ * The following code is for baryonGamma3pt function                   *
+ **********************************************************************/
+
+/* Dq1_ti is a quark line from t_i to t_J
+ * Dq2_spec is a quark line from t_i to t_f
+ * Dq3_spec is a quark line from t_i to t_f
+ * Dq4_tf is a quark line from t_f to t_J */
+template<class FImpl>
+template <class mobj, class mobj2, class robj>
+void BaryonUtils<FImpl>::Baryon_Gamma_3pt_Group1_Site(
+                        const mobj &Dq1_ti,
+                        const mobj2 &Dq2_spec,
+                        const mobj2 &Dq3_spec,
+                        const mobj &Dq4_tf,
+                                const Gamma GammaJ,
+                                const Gamma GammaBi,
+                                const Gamma GammaBf,
+                        int wick_contraction,
+                        robj &result)
+{
+    Gamma g5(Gamma::Algebra::Gamma5); 
+
+    auto adjD4_g_D1     = g5 * adj(Dq4_tf) * g5 * GammaJ * Dq1_ti;
+    auto Gf_adjD4_g_D1  = GammaBf * adjD4_g_D1;
+    auto D2_Gi          = Dq2_spec * GammaBi;
+    auto Gf_D2_Gi       = GammaBf * D2_Gi;
+    auto Gf_D3          = GammaBf * Dq3_spec;
+
+    int a_f, b_f, c_f;
+    int a_i, b_i, c_i;
+
+    Real ee;
+
+    for (int ie_f=0; ie_f < 6 ; ie_f++){
+        a_f = epsilon[ie_f][0]; //a
+        b_f = epsilon[ie_f][1]; //b
+        c_f = epsilon[ie_f][2]; //c
+    for (int ie_i=0; ie_i < 6 ; ie_i++){
+        a_i = epsilon[ie_i][0]; //a'
+        b_i = epsilon[ie_i][1]; //b'
+        c_i = epsilon[ie_i][2]; //c'
+
+        ee = epsilon_sgn[ie_f] * epsilon_sgn[ie_i];
+
+        for (int alpha_f=0; alpha_f<Ns; alpha_f++){
+        for (int beta_i=0; beta_i<Ns; beta_i++){
+            auto D2_Gi_ab_aa        = D2_Gi     ()(alpha_f,beta_i)(a_f,a_i);
+            auto Gf_D3_ab_bb        = Gf_D3     ()(alpha_f,beta_i)(b_f,b_i);
+            auto Gf_D2_Gi_ab_ba     = Gf_D2_Gi  ()(alpha_f,beta_i)(b_f,a_i);
+            auto Dq3_spec_ab_ab     = Dq3_spec  ()(alpha_f,beta_i)(a_f,b_i);
+
+            for (int gamma_i=0; gamma_i<Ns; gamma_i++){
+                auto ee_adjD4_g_D1_ag_ac        = ee * adjD4_g_D1   ()(alpha_f,gamma_i)(a_f,c_i);
+                auto ee_Gf_adjD4_g_D1_ag_bc     = ee * Gf_adjD4_g_D1()(alpha_f,gamma_i)(b_f,c_i);
+            for (int gamma_f=0; gamma_f<Ns; gamma_f++){
+                auto ee_adjD4_g_D1_gg_cc        = ee * adjD4_g_D1   ()(gamma_f,gamma_i)(c_f,c_i);
+                auto Dq3_spec_gb_cb             = Dq3_spec          ()(gamma_f,beta_i)(c_f,b_i);
+                auto D2_Gi_gb_ca                = D2_Gi             ()(gamma_f,beta_i)(c_f,a_i);
+
+
+                if(wick_contraction == 1) { // Do contraction I1
+                    result()(gamma_f,gamma_i)() -= ee_adjD4_g_D1_gg_cc
+                                                        * D2_Gi_ab_aa
+                                                        * Gf_D3_ab_bb;
+                }
+                if(wick_contraction == 2) { // Do contraction I2
+                    result()(gamma_f,gamma_i)() -= ee_adjD4_g_D1_ag_ac
+                                                        * Gf_D2_Gi_ab_ba
+                                                        * Dq3_spec_gb_cb;
+                }
+                if(wick_contraction == 3) { // Do contraction I3
+                    result()(gamma_f,gamma_i)() -= ee_Gf_adjD4_g_D1_ag_bc
+                                                        * D2_Gi_gb_ca
+                                                        * Dq3_spec_ab_ab;
+                }
+                if(wick_contraction == 4) { // Do contraction I4
+                    result()(gamma_f,gamma_i)() += ee_adjD4_g_D1_gg_cc
+                                                        * Gf_D2_Gi_ab_ba
+                                                        * Dq3_spec_ab_ab;
+                }
+                if(wick_contraction == 5) { // Do contraction I5
+                    result()(gamma_f,gamma_i)() += ee_Gf_adjD4_g_D1_ag_bc
+                                                        * D2_Gi_ab_aa
+                                                        * Dq3_spec_gb_cb;
+                }
+                if(wick_contraction == 6) { // Do contraction I6
+                    result()(gamma_f,gamma_i)() += ee_adjD4_g_D1_ag_ac
+                                                        * D2_Gi_gb_ca
+                                                        * Gf_D3_ab_bb;
+                }
+            }}
+        }}
+    }}
+}
+
+/* Dq1_spec is a quark line from t_i to t_f
+ * Dq2_ti is a quark line from t_i to t_J
+ * Dq3_spec is a quark line from t_i to t_f
+ * Dq4_tf is a quark line from t_f to t_J */
+template<class FImpl>
+template <class mobj, class mobj2, class robj>
+void BaryonUtils<FImpl>::Baryon_Gamma_3pt_Group2_Site(
+                        const mobj2 &Dq1_spec,
+                        const mobj &Dq2_ti,
+                        const mobj2 &Dq3_spec,
+                        const mobj &Dq4_tf,
+                                const Gamma GammaJ,
+                                const Gamma GammaBi,
+                                const Gamma GammaBf,
+                        int wick_contraction,
+                        robj &result)
+{
+    Gamma g5(Gamma::Algebra::Gamma5); 
+
+    auto adjD4_g_D2_Gi      = g5 * adj(Dq4_tf) * g5 * GammaJ * Dq2_ti * GammaBi;
+    auto Gf_adjD4_g_D2_Gi   = GammaBf * adjD4_g_D2_Gi;
+    auto Gf_D1              = GammaBf * Dq1_spec;
+    auto Gf_D3              = GammaBf * Dq3_spec;
+
+    int a_f, b_f, c_f;
+    int a_i, b_i, c_i;
+
+    Real ee;
+
+    for (int ie_f=0; ie_f < 6 ; ie_f++){
+        a_f = epsilon[ie_f][0]; //a
+        b_f = epsilon[ie_f][1]; //b
+        c_f = epsilon[ie_f][2]; //c
+    for (int ie_i=0; ie_i < 6 ; ie_i++){
+        a_i = epsilon[ie_i][0]; //a'
+        b_i = epsilon[ie_i][1]; //b'
+        c_i = epsilon[ie_i][2]; //c'
+
+        ee = epsilon_sgn[ie_f] * epsilon_sgn[ie_i];
+
+        for (int alpha_f=0; alpha_f<Ns; alpha_f++){
+        for (int beta_i=0; beta_i<Ns; beta_i++){
+            auto adjD4_g_D2_Gi_ab_aa        = adjD4_g_D2_Gi     ()(alpha_f,beta_i)(a_f,a_i);
+            auto Gf_D3_ab_bb                = Gf_D3             ()(alpha_f,beta_i)(b_f,b_i);
+            auto Gf_adjD4_g_D2_Gi_ab_ba     = Gf_adjD4_g_D2_Gi  ()(alpha_f,beta_i)(b_f,a_i);
+            auto Dq3_spec_ab_ab             = Dq3_spec          ()(alpha_f,beta_i)(a_f,b_i);
+
+            for (int gamma_i=0; gamma_i<Ns; gamma_i++){ 
+                auto ee_Dq1_spec_ag_ac      = ee * Dq1_spec     ()(alpha_f,gamma_i)(a_f,c_i);
+                auto ee_Gf_D1_ag_bc         = ee * Gf_D1        ()(alpha_f,gamma_i)(b_f,c_i);
+            for (int gamma_f=0; gamma_f<Ns; gamma_f++){
+                auto ee_Dq1_spec_gg_cc      = ee * Dq1_spec     ()(gamma_f,gamma_i)(c_f,c_i);
+                auto Dq3_spec_gb_cb         = Dq3_spec          ()(gamma_f,beta_i)(c_f,b_i);
+                auto adjD4_g_D2_Gi_gb_ca    = adjD4_g_D2_Gi     ()(gamma_f,beta_i)(c_f,a_i);
+
+                if(wick_contraction == 1) { // Do contraction II1
+                    result()(gamma_f,gamma_i)() -= ee_Dq1_spec_gg_cc
+                                                        * adjD4_g_D2_Gi_ab_aa
+                                                        * Gf_D3_ab_bb;
+                }
+                if(wick_contraction == 2) { // Do contraction II2
+                    result()(gamma_f,gamma_i)() -= ee_Dq1_spec_ag_ac
+                                                        * Gf_adjD4_g_D2_Gi_ab_ba
+                                                        * Dq3_spec_gb_cb;
+                }
+                if(wick_contraction == 3) { // Do contraction II3
+                    result()(gamma_f,gamma_i)() -= ee_Gf_D1_ag_bc
+                                                        * adjD4_g_D2_Gi_gb_ca
+                                                        * Dq3_spec_ab_ab;
+                }
+                if(wick_contraction == 4) { // Do contraction II4
+                    result()(gamma_f,gamma_i)() += ee_Dq1_spec_gg_cc
+                                                        * Gf_adjD4_g_D2_Gi_ab_ba
+                                                        * Dq3_spec_ab_ab;
+                }
+                if(wick_contraction == 5) { // Do contraction II5
+                    result()(gamma_f,gamma_i)() += ee_Gf_D1_ag_bc
+                                                        * adjD4_g_D2_Gi_ab_aa
+                                                        * Dq3_spec_gb_cb;
+                }
+                if(wick_contraction == 6) { // Do contraction II6
+                    result()(gamma_f,gamma_i)() += ee_Dq1_spec_ag_ac
+                                                        * adjD4_g_D2_Gi_gb_ca
+                                                        * Gf_D3_ab_bb;
+                }
+            }}
+        }}
+    }}
+}
+
+/* Dq1_spec is a quark line from t_i to t_f
+ * Dq2_spec is a quark line from t_i to t_f
+ * Dq3_ti is a quark line from t_i to t_J
+ * Dq4_tf is a quark line from t_f to t_J */
+template<class FImpl>
+template <class mobj, class mobj2, class robj>
+void BaryonUtils<FImpl>::Baryon_Gamma_3pt_Group3_Site(
+                        const mobj2 &Dq1_spec,
+                        const mobj2 &Dq2_spec,
+                        const mobj &Dq3_ti,
+                        const mobj &Dq4_tf,
+                                const Gamma GammaJ,
+                                const Gamma GammaBi,
+                                const Gamma GammaBf,
+                        int wick_contraction,
+                        robj &result)
+{
+    Gamma g5(Gamma::Algebra::Gamma5);
+
+    auto adjD4_g_D3     = g5 * adj(Dq4_tf) * g5 * GammaJ * Dq3_ti;
+    auto Gf_adjD4_g_D3  = GammaBf * adjD4_g_D3;
+    auto Gf_D1          = GammaBf * Dq1_spec;
+    auto D2_Gi          = Dq2_spec * GammaBi;
+    auto Gf_D2_Gi       = GammaBf * D2_Gi;
+
+    int a_f, b_f, c_f;
+    int a_i, b_i, c_i;
+
+    Real ee;
+
+    for (int ie_f=0; ie_f < 6 ; ie_f++){
+        a_f = epsilon[ie_f][0]; //a
+        b_f = epsilon[ie_f][1]; //b
+        c_f = epsilon[ie_f][2]; //c
+    for (int ie_i=0; ie_i < 6 ; ie_i++){
+        a_i = epsilon[ie_i][0]; //a'
+        b_i = epsilon[ie_i][1]; //b'
+        c_i = epsilon[ie_i][2]; //c'
+
+        ee = epsilon_sgn[ie_f] * epsilon_sgn[ie_i];
+
+        for (int alpha_f=0; alpha_f<Ns; alpha_f++){
+        for (int beta_i=0; beta_i<Ns; beta_i++){
+            auto D2_Gi_ab_aa            = D2_Gi         ()(alpha_f,beta_i)(a_f,a_i);
+            auto Gf_adjD4_g_D3_ab_bb    = Gf_adjD4_g_D3 ()(alpha_f,beta_i)(b_f,b_i);
+            auto Gf_D2_Gi_ab_ba         = Gf_D2_Gi      ()(alpha_f,beta_i)(b_f,a_i);
+            auto adjD4_g_D3_ab_ab       = adjD4_g_D3    ()(alpha_f,beta_i)(a_f,b_i);
+
+            for (int gamma_i=0; gamma_i<Ns; gamma_i++) {
+                auto ee_Dq1_spec_ag_ac  = ee * Dq1_spec ()(alpha_f,gamma_i)(a_f,c_i);
+                auto ee_Gf_D1_ag_bc     = ee * Gf_D1    ()(alpha_f,gamma_i)(b_f,c_i);
+            for (int gamma_f=0; gamma_f<Ns; gamma_f++) {
+                auto ee_Dq1_spec_gg_cc  = ee * Dq1_spec ()(gamma_f,gamma_i)(c_f,c_i);
+                auto adjD4_g_D3_gb_cb   = adjD4_g_D3    ()(gamma_f,beta_i)(c_f,b_i);
+                auto D2_Gi_gb_ca        = D2_Gi         ()(gamma_f,beta_i)(c_f,a_i);
+
+                if(wick_contraction == 1) { // Do contraction III1
+                    result()(gamma_f,gamma_i)() -= ee_Dq1_spec_gg_cc
+                                                        * D2_Gi_ab_aa
+                                                        * Gf_adjD4_g_D3_ab_bb;
+                }
+                if(wick_contraction == 2) { // Do contraction III2
+                    result()(gamma_f,gamma_i)() -= ee_Dq1_spec_ag_ac
+                                                        * Gf_D2_Gi_ab_ba
+                                                        * adjD4_g_D3_gb_cb;
+                }
+                if(wick_contraction == 3) { // Do contraction III3
+                    result()(gamma_f,gamma_i)() -= ee_Gf_D1_ag_bc
+                                                        * D2_Gi_gb_ca
+                                                        * adjD4_g_D3_ab_ab;
+                }
+                if(wick_contraction == 4) { // Do contraction III4
+                    result()(gamma_f,gamma_i)() += ee_Dq1_spec_gg_cc
+                                                        * Gf_D2_Gi_ab_ba
+                                                        * adjD4_g_D3_ab_ab;
+                }
+                if(wick_contraction == 5) { // Do contraction III5
+                    result()(gamma_f,gamma_i)() += ee_Gf_D1_ag_bc
+                                                        * D2_Gi_ab_aa
+                                                        * adjD4_g_D3_gb_cb;
+                }
+                if(wick_contraction == 6) { // Do contraction III6
+                    result()(gamma_f,gamma_i)() += ee_Dq1_spec_ag_ac
+                                                        * D2_Gi_gb_ca
+                                                        * Gf_adjD4_g_D3_ab_bb;
+                }
+            }}
+        }}
+    }}
+}
+
+/* The group indicates which inital state quarks the current is  * 
+ * connected to. It must be in the range 1-3.                    *
+ * The wick_contraction must be in the range 1-6 correspond to   *
+ * the contractions given in the Hadrons documentation at        *
+ * https://aportelli.github.io/Hadrons-doc/#/mcontraction        */
+template<class FImpl>
+template <class mobj>
+void BaryonUtils<FImpl>::Baryon_Gamma_3pt(
+                        const PropagatorField &q_ti,
+                        const mobj &Dq_spec1,
+                        const mobj &Dq_spec2,
+                        const PropagatorField &q_tf,
+                        int group,
+                        int wick_contraction,
+                                const Gamma GammaJ,
+                                const Gamma GammaBi,
+                                const Gamma GammaBf,
+                        SpinMatrixField &stn_corr)
+{
+    GridBase *grid = q_tf.Grid();
+
+    autoView( vcorr, stn_corr, CpuWrite);
+    autoView( vq_ti , q_ti, CpuRead);
+    autoView( vq_tf , q_tf, CpuRead);
+
+    if (group == 1) {
+        accelerator_for(ss, grid->oSites(), grid->Nsimd(), {
+            auto Dq_ti = vq_ti[ss];
+            auto Dq_tf = vq_tf[ss];
+            sobj result=Zero();
+            Baryon_Gamma_3pt_Group1_Site(Dq_ti,Dq_spec1,Dq_spec2,Dq_tf,GammaJ,GammaBi,GammaBf,wick_contraction,result);
+            vcorr[ss] += result; 
+        });//end loop over lattice sites
+    } else if (group == 2) {
+        accelerator_for(ss, grid->oSites(), grid->Nsimd(), {
+            auto Dq_ti = vq_ti[ss];
+            auto Dq_tf = vq_tf[ss];
+            sobj result=Zero();
+            Baryon_Gamma_3pt_Group2_Site(Dq_spec1,Dq_ti,Dq_spec2,Dq_tf,GammaJ,GammaBi,GammaBf,wick_contraction,result);
+            vcorr[ss] += result; 
+        });//end loop over lattice sites
+    } else if (group == 3) {
+        accelerator_for(ss, grid->oSites(), grid->Nsimd(), {
+            auto Dq_ti = vq_ti[ss];
+            auto Dq_tf = vq_tf[ss];
+            sobj result=Zero();
+            Baryon_Gamma_3pt_Group3_Site(Dq_spec1,Dq_spec2,Dq_ti,Dq_tf,GammaJ,GammaBi,GammaBf,wick_contraction,result);
+
+            vcorr[ss] += result; 
+        });//end loop over lattice sites
+    }
+}
+
+
+/***********************************************************************
+ * End of BaryonGamma3pt-function code.                                *
+ *																	   *
  * The following code is for Sigma -> N rare hypeon decays             *
  **********************************************************************/
 
@@ -558,15 +1012,18 @@ void BaryonUtils<FImpl>::Sigma_to_Nucleon_Eye(const PropagatorField &qq_loop,
 						 const std::string op,
 						 SpinMatrixField &stn_corr)
 {
+
+  assert(Ns==4 && "Baryon code only implemented for N_spin = 4");
+  assert(Nc==3 && "Baryon code only implemented for N_colour = 3");
+
   GridBase *grid = qs_ti.Grid();
 
-  auto vcorr= stn_corr.View();
-  auto vq_loop = qq_loop.View();
-  auto vd_tf = qd_tf.View();
-  auto vs_ti = qs_ti.View();
+  autoView( vcorr, stn_corr, CpuWrite);
+  autoView( vq_loop , qq_loop, CpuRead);
+  autoView( vd_tf , qd_tf, CpuRead);
+  autoView( vs_ti , qs_ti, CpuRead);
 
- // accelerator_for(ss, grid->oSites(), grid->Nsimd(), {
-  thread_for(ss,grid->oSites(),{
+  accelerator_for(ss, grid->oSites(), grid->Nsimd(), {
     auto Dq_loop = vq_loop[ss];
     auto Dd_tf = vd_tf[ss];
     auto Ds_ti = vs_ti[ss];
@@ -595,14 +1052,17 @@ void BaryonUtils<FImpl>::Sigma_to_Nucleon_NonEye(const PropagatorField &qq_ti,
 						 const std::string op,
 						 SpinMatrixField &stn_corr)
 {
+
+  assert(Ns==4 && "Baryon code only implemented for N_spin = 4");
+  assert(Nc==3 && "Baryon code only implemented for N_colour = 3");
+
   GridBase *grid = qs_ti.Grid();
 
-  auto vcorr= stn_corr.View();
-  auto vq_ti = qq_ti.View();
-  auto vq_tf = qq_tf.View();
-  auto vd_tf = qd_tf.View();
-  auto vs_ti = qs_ti.View();
-
+  autoView( vcorr , stn_corr, CpuWrite);
+  autoView( vq_ti , qq_ti, CpuRead);
+  autoView( vq_tf , qq_tf, CpuRead);
+  autoView( vd_tf , qd_tf, CpuRead);
+  autoView( vs_ti , qs_ti, CpuRead);
  // accelerator_for(ss, grid->oSites(), grid->Nsimd(), {
   thread_for(ss,grid->oSites(),{
     auto Dq_ti = vq_ti[ss];
diff --git a/Grid/qcd/utils/CovariantCshift.h b/Grid/qcd/utils/CovariantCshift.h
index ed96f3bf..6ac69150 100644
--- a/Grid/qcd/utils/CovariantCshift.h
+++ b/Grid/qcd/utils/CovariantCshift.h
@@ -52,6 +52,26 @@ namespace PeriodicBC {
     tmp = adj(Link)*field;
     return Cshift(tmp,mu,-1);// moves towards positive mu
   }
+
+  template<class gauge,typename Op, typename T1> auto
+    CovShiftForward(const Lattice<gauge> &Link, 
+		    int mu,
+		    const LatticeUnaryExpression<Op,T1> &expr)
+    -> Lattice<decltype(expr.op.func(eval(0, expr.arg1)))> 
+  {
+    Lattice<decltype(expr.op.func(eval(0, expr.arg1)))> arg(expr);
+    return CovShiftForward(Link,mu,arg);
+  }
+  template<class gauge,typename Op, typename T1> auto
+    CovShiftBackward(const Lattice<gauge> &Link, 
+		     int mu,
+		     const LatticeUnaryExpression<Op,T1> &expr)
+    -> Lattice<decltype(expr.op.func(eval(0, expr.arg1)))> 
+  {
+    Lattice<decltype(expr.op.func(eval(0, expr.arg1)))> arg(expr);
+    return CovShiftForward(Link,mu,arg);
+  }
+
 }
 
 
@@ -122,6 +142,26 @@ namespace ConjugateBC {
     return Cshift(tmp,mu,-1);// moves towards positive mu
   }
 
+  template<class gauge,typename Op, typename T1> auto
+    CovShiftForward(const Lattice<gauge> &Link, 
+		    int mu,
+		    const LatticeUnaryExpression<Op,T1> &expr)
+    -> Lattice<decltype(expr.op.func(eval(0, expr.arg1)))> 
+  {
+    Lattice<decltype(expr.op.func(eval(0, expr.arg1)))> arg(expr);
+    return CovShiftForward(Link,mu,arg);
+  }
+  template<class gauge,typename Op, typename T1> auto
+    CovShiftBackward(const Lattice<gauge> &Link, 
+		     int mu,
+		     const LatticeUnaryExpression<Op,T1> &expr)
+    -> Lattice<decltype(expr.op.func(eval(0, expr.arg1)))> 
+  {
+    Lattice<decltype(expr.op.func(eval(0, expr.arg1)))> arg(expr);
+    return CovShiftForward(Link,mu,arg);
+  }
+
+
 }
 
 
diff --git a/Grid/qcd/utils/LinalgUtils.h b/Grid/qcd/utils/LinalgUtils.h
index 56f8f164..1e016e4e 100644
--- a/Grid/qcd/utils/LinalgUtils.h
+++ b/Grid/qcd/utils/LinalgUtils.h
@@ -47,8 +47,8 @@ void axpibg5x(Lattice<vobj> &z,const Lattice<vobj> &x,Coeff a,Coeff b)
   GridBase *grid=x.Grid();
 
   Gamma G5(Gamma::Algebra::Gamma5);
-  auto x_v = x.View();
-  auto z_v = z.View();
+  autoView(x_v, x, AcceleratorRead);
+  autoView(z_v, z, AcceleratorWrite);
   accelerator_for( ss, x_v.size(),vobj::Nsimd(), {
     auto tmp = a*x_v(ss) + G5*(b*timesI(x_v(ss)));
     coalescedWrite(z_v[ss],tmp);
@@ -63,9 +63,9 @@ void axpby_ssp(Lattice<vobj> &z, Coeff a,const Lattice<vobj> &x,Coeff b,const La
   conformable(x,z);
   GridBase *grid=x.Grid();
   int Ls = grid->_rdimensions[0];
-  auto x_v = x.View();
-  auto y_v = y.View();
-  auto z_v = z.View();
+  autoView( x_v, x, AcceleratorRead);
+  autoView( y_v, y, AcceleratorRead);
+  autoView( z_v, z, AcceleratorWrite);
   // FIXME -- need a new class of accelerator_loop to implement this
   //
   uint64_t nloop = grid->oSites()/Ls;
@@ -85,9 +85,9 @@ void ag5xpby_ssp(Lattice<vobj> &z,Coeff a,const Lattice<vobj> &x,Coeff b,const L
   GridBase *grid=x.Grid();
   int Ls = grid->_rdimensions[0];
   Gamma G5(Gamma::Algebra::Gamma5);
-  auto x_v = x.View();
-  auto y_v = y.View();
-  auto z_v = z.View();
+  autoView( x_v, x, AcceleratorRead);
+  autoView( y_v, y, AcceleratorRead);
+  autoView( z_v, z, AcceleratorWrite);
   uint64_t nloop = grid->oSites()/Ls;
   accelerator_for(sss,nloop,vobj::Nsimd(),{
     uint64_t ss = sss*Ls;
@@ -104,9 +104,9 @@ void axpbg5y_ssp(Lattice<vobj> &z,Coeff a,const Lattice<vobj> &x,Coeff b,const L
   conformable(x,z);
   GridBase *grid=x.Grid();
   int Ls = grid->_rdimensions[0];
-  auto x_v = x.View();
-  auto y_v = y.View();
-  auto z_v = z.View();
+  autoView( x_v, x, AcceleratorRead);
+  autoView( y_v, y, AcceleratorRead);
+  autoView( z_v, z, AcceleratorWrite);
   Gamma G5(Gamma::Algebra::Gamma5);
   uint64_t nloop = grid->oSites()/Ls;
   accelerator_for(sss,nloop,vobj::Nsimd(),{
@@ -125,9 +125,9 @@ void ag5xpbg5y_ssp(Lattice<vobj> &z,Coeff a,const Lattice<vobj> &x,Coeff b,const
   GridBase *grid=x.Grid();
   int Ls = grid->_rdimensions[0];
 
-  auto x_v = x.View();
-  auto y_v = y.View();
-  auto z_v = z.View();
+  autoView( x_v, x, AcceleratorRead);
+  autoView( y_v, y, AcceleratorRead);
+  autoView( z_v, z, AcceleratorWrite);
   Gamma G5(Gamma::Algebra::Gamma5);
   uint64_t nloop = grid->oSites()/Ls;
   accelerator_for(sss,nloop,vobj::Nsimd(),{
@@ -147,9 +147,9 @@ void axpby_ssp_pminus(Lattice<vobj> &z,Coeff a,const Lattice<vobj> &x,Coeff b,co
   GridBase *grid=x.Grid();
   int Ls = grid->_rdimensions[0];
 
-  auto x_v = x.View();
-  auto y_v = y.View();
-  auto z_v = z.View();
+  autoView( x_v, x, AcceleratorRead);
+  autoView( y_v, y, AcceleratorRead);
+  autoView( z_v, z, AcceleratorWrite);
   uint64_t nloop = grid->oSites()/Ls;
   accelerator_for(sss,nloop,vobj::Nsimd(),{
     uint64_t ss = sss*Ls;
@@ -168,9 +168,9 @@ void axpby_ssp_pplus(Lattice<vobj> &z,Coeff a,const Lattice<vobj> &x,Coeff b,con
   conformable(x,z);
   GridBase *grid=x.Grid();
   int Ls = grid->_rdimensions[0];
-  auto x_v = x.View();
-  auto y_v = y.View();
-  auto z_v = z.View();
+  autoView( x_v, x, AcceleratorRead);
+  autoView( y_v, y, AcceleratorRead);
+  autoView( z_v, z, AcceleratorWrite);
   uint64_t nloop = grid->oSites()/Ls;
   accelerator_for(sss,nloop,vobj::Nsimd(),{
     uint64_t ss = sss*Ls;
@@ -189,8 +189,8 @@ void G5R5(Lattice<vobj> &z,const Lattice<vobj> &x)
   conformable(x,z);
   int Ls = grid->_rdimensions[0];
   Gamma G5(Gamma::Algebra::Gamma5);
-  auto x_v = x.View();
-  auto z_v = z.View();
+  autoView( x_v, x, AcceleratorRead);
+  autoView( z_v, z, AcceleratorWrite);
   uint64_t nloop = grid->oSites()/Ls;
   accelerator_for(sss,nloop,vobj::Nsimd(),{
     uint64_t ss = sss*Ls;
@@ -222,8 +222,8 @@ void G5C(Lattice<iVector<CComplex, nbasis>> &z, const Lattice<iVector<CComplex,
   static_assert(nbasis % 2 == 0, "");
   int nb = nbasis / 2;
 
-  auto z_v = z.View();
-  auto x_v = x.View();
+  autoView( z_v, z, AcceleratorWrite);
+  autoView( x_v, x, AcceleratorRead);
   accelerator_for(ss,grid->oSites(),CComplex::Nsimd(),
   {
     for(int n = 0; n < nb; ++n) {
diff --git a/Grid/qcd/utils/SUn.h b/Grid/qcd/utils/SUn.h
index 7ad80d00..0cc0cc1a 100644
--- a/Grid/qcd/utils/SUn.h
+++ b/Grid/qcd/utils/SUn.h
@@ -222,11 +222,11 @@ public:
     conformable(subgroup, Determinant);
     int i0, i1;
     su2SubGroupIndex(i0, i1, su2_index);
-    auto subgroup_v = subgroup.View();
-    auto source_v   = source.View();
-    auto Determinant_v = Determinant.View();
 
-    thread_for(ss, grid->oSites(), {
+    autoView( subgroup_v , subgroup,AcceleratorWrite);
+    autoView( source_v   , source,AcceleratorRead);
+    autoView( Determinant_v , Determinant,AcceleratorWrite);
+    accelerator_for(ss, grid->oSites(), 1, {
 
       subgroup_v[ss]()()(0, 0) = source_v[ss]()()(i0, i0);
       subgroup_v[ss]()()(0, 1) = source_v[ss]()()(i0, i1);
@@ -257,15 +257,16 @@ public:
     su2SubGroupIndex(i0, i1, su2_index);
 
     dest = 1.0;  // start out with identity
-    auto dest_v = dest.View();
-    auto subgroup_v = subgroup.View();
-    thread_for(ss, grid->oSites(),
+    autoView( dest_v , dest, AcceleratorWrite);
+    autoView( subgroup_v, subgroup, AcceleratorRead);
+    accelerator_for(ss, grid->oSites(),1,
     {
       dest_v[ss]()()(i0, i0) = subgroup_v[ss]()()(0, 0);
       dest_v[ss]()()(i0, i1) = subgroup_v[ss]()()(0, 1);
       dest_v[ss]()()(i1, i0) = subgroup_v[ss]()()(1, 0);
       dest_v[ss]()()(i1, i1) = subgroup_v[ss]()()(1, 1);
     });
+
   }
 
   ///////////////////////////////////////////////
@@ -608,8 +609,8 @@ public:
 
   // reunitarise??
   template <typename LatticeMatrixType>
-  static void LieRandomize(GridParallelRNG &pRNG, LatticeMatrixType &out,
-                           double scale = 1.0) {
+  static void LieRandomize(GridParallelRNG &pRNG, LatticeMatrixType &out, double scale = 1.0) 
+  {
     GridBase *grid = out.Grid();
 
     typedef typename LatticeMatrixType::vector_type vector_type;
@@ -618,8 +619,7 @@ public:
     typedef iSinglet<vector_type> vTComplexType;
 
     typedef Lattice<vTComplexType> LatticeComplexType;
-    typedef typename GridTypeMapper<
-      typename LatticeMatrixType::vector_object>::scalar_object MatrixType;
+    typedef typename GridTypeMapper<typename LatticeMatrixType::vector_object>::scalar_object MatrixType;
 
     LatticeComplexType ca(grid);
     LatticeMatrixType lie(grid);
@@ -629,6 +629,7 @@ public:
     MatrixType ta;
 
     lie = Zero();
+
     for (int a = 0; a < AdjointDimension; a++) {
       random(pRNG, ca);
 
@@ -640,6 +641,7 @@ public:
       la = ci * ca * ta;
 
       lie = lie + la;  // e^{i la ta}
+
     }
     taExp(lie, out);
   }
diff --git a/Grid/qcd/utils/WilsonLoops.h b/Grid/qcd/utils/WilsonLoops.h
index 0367c9fa..fdd53698 100644
--- a/Grid/qcd/utils/WilsonLoops.h
+++ b/Grid/qcd/utils/WilsonLoops.h
@@ -485,7 +485,7 @@ public:
 
         // Up staple    ___ ___
         //             |       |
-        tmp = Cshift(adj(U[nu]), nu, -1);
+        tmp = Cshift(closure(adj(U[nu])), nu, -1);
         tmp = adj(U2[mu]) * tmp;
         tmp = Cshift(tmp, mu, -2);
 
@@ -519,7 +519,7 @@ public:
         //
         //      |  |
 
-        tmp = Cshift(adj(U2[nu]), nu, -2);
+        tmp = Cshift(closure(adj(U2[nu])), nu, -2);
         tmp = Gimpl::CovShiftBackward(U[mu], mu, tmp);
         tmp = U2[nu] * Cshift(tmp, nu, 2);
         Stap += Cshift(tmp, mu, 1);
diff --git a/Grid/serialisation/BaseIO.h b/Grid/serialisation/BaseIO.h
index bf424fc7..49406201 100644
--- a/Grid/serialisation/BaseIO.h
+++ b/Grid/serialisation/BaseIO.h
@@ -87,11 +87,7 @@ namespace Grid {
     template<typename Scalar_, typename Dimensions_, int Options_, typename IndexType>
     struct is_tensor_fixed<Eigen::TensorFixedSize<Scalar_, Dimensions_, Options_, IndexType>>
         : public std::true_type {};
-    template<typename Scalar_, typename Dimensions_, int Options_, typename IndexType,
-              int MapOptions_, template <class> class MapPointer_>
-    struct is_tensor_fixed<Eigen::TensorMap<Eigen::TensorFixedSize<Scalar_, Dimensions_,
-                                            Options_, IndexType>, MapOptions_, MapPointer_>>
-        : public std::true_type {};
+    template<typename T> struct is_tensor_fixed<Eigen::TensorMap<T>> : public std::true_type {};
 
     // Is this a variable-size Eigen tensor
     template<typename T, typename V = void> struct is_tensor_variable : public std::false_type {};
diff --git a/Grid/serialisation/MacroMagic.h b/Grid/serialisation/MacroMagic.h
index 7866327e..0495b91e 100644
--- a/Grid/serialisation/MacroMagic.h
+++ b/Grid/serialisation/MacroMagic.h
@@ -114,7 +114,8 @@ THE SOFTWARE.
 #define GRID_MACRO_WRITE_MEMBER(A,B) ::Grid::write(WR,#B,obj. B);
 
 #define GRID_SERIALIZABLE_CLASS_MEMBERS(cname,...)\
-  std::string SerialisableClassName(void) const {return std::string(#cname);}	\
+static inline std::string SerialisableClassName(void) {return std::string(#cname);}	\
+static constexpr bool isEnum = false; \
 GRID_MACRO_EVAL(GRID_MACRO_MAP(GRID_MACRO_MEMBER,__VA_ARGS__))\
 template <typename T>\
 static inline void write(Writer<T> &WR,const std::string &s, const cname &obj){ \
@@ -162,6 +163,8 @@ public:\
 public:\
   accelerator name(void)     : value_(undefname) {};		\
   accelerator name(int value): value_(value) {};			\
+  static inline std::string SerialisableClassName(void) {return std::string(#name);}\
+  static constexpr bool isEnum = true; \
   template <typename T>\
   static inline void write(::Grid::Writer<T> &WR,const std::string &s, const name &obj) \
   {\
diff --git a/Grid/serialisation/Serialisation.h b/Grid/serialisation/Serialisation.h
index c95226b1..e14120af 100644
--- a/Grid/serialisation/Serialisation.h
+++ b/Grid/serialisation/Serialisation.h
@@ -36,7 +36,7 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 #include "BinaryIO.h"
 #include "TextIO.h"
 #include "XmlIO.h"
-#ifndef GRID_NVCC
+#if (!defined(GRID_CUDA)) && (!defined(GRID_HIP))
 #include "JSON_IO.h"
 #endif
 
diff --git a/Grid/serialisation/VectorUtils.h b/Grid/serialisation/VectorUtils.h
index a5a73992..dd5ff0b8 100644
--- a/Grid/serialisation/VectorUtils.h
+++ b/Grid/serialisation/VectorUtils.h
@@ -432,12 +432,10 @@ namespace Grid {
   std::vector<T> strToVec(const std::string s)
   {
     std::istringstream sstr(s);
-    T                  buf;
     std::vector<T>     v;
     
-    while(!sstr.eof())
+    for(T buf; sstr >> buf;)
     {
-      sstr >> buf;
       v.push_back(buf);
     }
     
diff --git a/Grid/simd/Grid_gpu_vec.h b/Grid/simd/Grid_gpu_vec.h
index 4584fb36..b9c6a81b 100644
--- a/Grid/simd/Grid_gpu_vec.h
+++ b/Grid/simd/Grid_gpu_vec.h
@@ -32,7 +32,12 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 */
 //----------------------------------------------------------------------
 
+#ifdef GRID_CUDA
 #include <cuda_fp16.h>
+#endif
+#ifdef GRID_HIP
+#include <hip/hip_fp16.h>
+#endif
 
 namespace Grid {
 
@@ -142,7 +147,7 @@ typedef GpuVector<NSIMD_Integer,  Integer     > GpuVectorI;
 accelerator_inline float half2float(half h)
 {
   float f;
-#ifdef __CUDA_ARCH__
+#ifdef GRID_SIMT
   f = __half2float(h);
 #else 
   //f = __half2float(h);
@@ -156,7 +161,7 @@ accelerator_inline float half2float(half h)
 accelerator_inline half float2half(float f)
 {
   half h;
-#ifdef __CUDA_ARCH__
+#ifdef GRID_SIMT
   h = __float2half(f);
 #else
   Grid_half hh = sfw_float_to_half(f);
diff --git a/Grid/simd/Simd.h b/Grid/simd/Simd.h
index bc8cd2fd..37aee2ed 100644
--- a/Grid/simd/Simd.h
+++ b/Grid/simd/Simd.h
@@ -31,7 +31,7 @@ directory
 #ifndef GRID_SIMD_H
 #define GRID_SIMD_H
 
-#ifdef GRID_NVCC
+#if defined(GRID_CUDA) || defined(GRID_HIP)
 #include <thrust/complex.h>
 #endif
 
@@ -65,7 +65,7 @@ typedef RealD   Real;
 typedef RealF  Real;
 #endif
 
-#ifdef GRID_NVCC
+#if defined(GRID_CUDA) || defined(GRID_HIP)
 typedef thrust::complex<RealF> ComplexF;
 typedef thrust::complex<RealD> ComplexD;
 typedef thrust::complex<Real>  Complex;
diff --git a/Grid/stencil/Stencil.h b/Grid/stencil/Stencil.h
index a56d256d..1e198972 100644
--- a/Grid/stencil/Stencil.h
+++ b/Grid/stencil/Stencil.h
@@ -67,7 +67,8 @@ void Gather_plane_simple_table (Vector<std::pair<int,int> >& table,const Lattice
 {
   int num=table.size();
   std::pair<int,int> *table_v = & table[0];
-  auto rhs_v = rhs.View();
+
+  auto rhs_v = rhs.View(AcceleratorRead);
   accelerator_forNB( i,num, vobj::Nsimd(), {
     typedef decltype(coalescedRead(buffer[0])) compressed_t;
     compressed_t   tmp_c;
@@ -75,6 +76,7 @@ void Gather_plane_simple_table (Vector<std::pair<int,int> >& table,const Lattice
     compress.Compress(&tmp_c,0,rhs_v(so+table_v[i].second));
     coalescedWrite(buffer[off+o],tmp_c);
   });
+  rhs_v.ViewClose();
 // Further optimisatoin: i) software prefetch the first element of the next table entry, prefetch the table
 }
 
@@ -94,7 +96,7 @@ void Gather_plane_exchange_table(Vector<std::pair<int,int> >& table,const Lattic
   int num=table.size()/2;
   int so  = plane*rhs.Grid()->_ostride[dimension]; // base offset for start of plane
 
-  auto rhs_v = rhs.View();
+  auto rhs_v = rhs.View(AcceleratorRead);
   auto p0=&pointers[0][0];
   auto p1=&pointers[1][0];
   auto tp=&table[0];
@@ -104,10 +106,11 @@ void Gather_plane_exchange_table(Vector<std::pair<int,int> >& table,const Lattic
 			      so+tp[2*j+1].second,
 			      type);
   });
+  rhs_v.ViewClose();
 }
 
 struct StencilEntry {
-#ifdef GRID_NVCC
+#ifdef GRID_CUDA
   uint64_t _byte_offset;       // 8 bytes
   uint32_t _offset;            // 4 bytes
 #else
@@ -122,7 +125,7 @@ struct StencilEntry {
 // Could pack to 8 + 4 + 4 = 128 bit and use
 
 template<class vobj,class cobj,class Parameters>
-class CartesianStencilView {
+class CartesianStencilAccelerator {
  public:
   typedef AcceleratorVector<int,STENCIL_MAX> StencilVector;
 
@@ -130,14 +133,15 @@ class CartesianStencilView {
   ////////////////////////////////////////
   // Basic Grid and stencil info
   ////////////////////////////////////////
-  int                               _checkerboard;
-  int                               _npoints; // Move to template param?
+  int           _checkerboard;
+  int           _npoints; // Move to template param?
+  int           _osites;
   StencilVector _directions;
   StencilVector _distances;
   StencilVector _comm_buf_size;
   StencilVector _permute_type;
   StencilVector same_node;
-  Coordinate                         _simd_layout;
+  Coordinate    _simd_layout;
   Parameters    parameters;
   StencilEntry*  _entries_p;
   cobj* u_recv_buf_p;
@@ -175,13 +179,43 @@ class CartesianStencilView {
   {
     Lexicographic::CoorFromIndex(coor,lane,this->_simd_layout);
   }
+};
+
+template<class vobj,class cobj,class Parameters>
+class CartesianStencilView : public CartesianStencilAccelerator<vobj,cobj,Parameters>
+{
+ private:
+  int *closed;
+  StencilEntry *cpu_ptr;
+  ViewMode      mode;
+ public:
+  // default copy constructor
+  CartesianStencilView (const CartesianStencilView &refer_to_me) = default;
+
+  CartesianStencilView (const CartesianStencilAccelerator<vobj,cobj,Parameters> &refer_to_me,ViewMode _mode)
+    : CartesianStencilAccelerator<vobj,cobj,Parameters>(refer_to_me),
+    cpu_ptr(this->_entries_p),
+    mode(_mode)
+  {
+    this->_entries_p =(StencilEntry *)
+      MemoryManager::ViewOpen(this->_entries_p,
+			      this->_npoints*this->_osites*sizeof(StencilEntry),
+			      mode,
+			      AdviseDefault);
+  }
+
+  void ViewClose(void)
+  {
+    MemoryManager::ViewClose(this->cpu_ptr,this->mode);
+  }
 
 };
+
 ////////////////////////////////////////
 // The Stencil Class itself
 ////////////////////////////////////////
 template<class vobj,class cobj,class Parameters>
-class CartesianStencil : public CartesianStencilView<vobj,cobj,Parameters> { // Stencil runs along coordinate axes only; NO diagonal fill in.
+class CartesianStencil : public CartesianStencilAccelerator<vobj,cobj,Parameters> { // Stencil runs along coordinate axes only; NO diagonal fill in.
 public:
 
   typedef typename cobj::vector_type vector_type;
@@ -226,8 +260,8 @@ public:
   // Generalise as required later if needed
   ////////////////////////////////////////////////////////////////////////
 
-  View_type View(void) const {
-    View_type accessor(*( (View_type *) this));
+  View_type View(ViewMode mode) const {
+    View_type accessor(*( (View_type *) this),mode);
     return accessor;
   }
 
@@ -662,9 +696,9 @@ public:
     _unified_buffer_size=0;
     surface_list.resize(0);
 
-    int osites  = _grid->oSites();
+    this->_osites  = _grid->oSites();
 
-    _entries.resize(this->_npoints* osites);
+    _entries.resize(this->_npoints* this->_osites);
     this->_entries_p = &_entries[0];
     for(int ii=0;ii<npoints;ii++){
 
diff --git a/Grid/tensors/Tensor_SIMT.h b/Grid/tensors/Tensor_SIMT.h
index c18b8484..ec57a679 100644
--- a/Grid/tensors/Tensor_SIMT.h
+++ b/Grid/tensors/Tensor_SIMT.h
@@ -31,22 +31,11 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 
 NAMESPACE_BEGIN(Grid);
 
-//accelerator_inline void SIMTsynchronise(void) 
-accelerator_inline void synchronise(void) 
-{
-#ifdef __CUDA_ARCH__
-//  __syncthreads();
-  __syncwarp();
-#endif
-  return;
-}
 
-#ifndef __CUDA_ARCH__
+#ifndef GRID_SIMT
 //////////////////////////////////////////
 // Trivial mapping of vectors on host
 //////////////////////////////////////////
-accelerator_inline int SIMTlane(int Nsimd) { return 0; } // CUDA specific
-
 template<class vobj> accelerator_inline
 vobj coalescedRead(const vobj & __restrict__ vec,int lane=0)
 {
@@ -66,7 +55,6 @@ vobj coalescedReadPermute(const vobj & __restrict__ vec,int ptype,int doperm,int
 template<class vobj> accelerator_inline
 void coalescedWrite(vobj & __restrict__ vec,const vobj & __restrict__ extracted,int lane=0)
 {
-  //  vstream(vec, extracted);
   vec = extracted;
 }
 template<class vobj> accelerator_inline
@@ -75,25 +63,24 @@ void coalescedWriteNonTemporal(vobj & __restrict__ vec,const vobj & __restrict__
   vstream(vec, extracted);
 }
 #else
-accelerator_inline int SIMTlane(int Nsimd) { return threadIdx.y; } // CUDA specific
 
 //////////////////////////////////////////
 // Extract and insert slices on the GPU
 //////////////////////////////////////////
 template<class vobj> accelerator_inline
-typename vobj::scalar_object coalescedRead(const vobj & __restrict__ vec,int lane=SIMTlane(vobj::Nsimd()))
+typename vobj::scalar_object coalescedRead(const vobj & __restrict__ vec,int lane=acceleratorSIMTlane(vobj::Nsimd()))
 {
   return extractLane(lane,vec);
 }
 template<class vobj> accelerator_inline
-typename vobj::scalar_object coalescedReadPermute(const vobj & __restrict__ vec,int ptype,int doperm,int lane=SIMTlane(vobj::Nsimd()))
+typename vobj::scalar_object coalescedReadPermute(const vobj & __restrict__ vec,int ptype,int doperm,int lane=acceleratorSIMTlane(vobj::Nsimd()))
 {
   int mask = vobj::Nsimd() >> (ptype + 1);		
   int plane= doperm ? lane ^ mask : lane;
   return extractLane(plane,vec);
 }
 template<class vobj> accelerator_inline
-void coalescedWrite(vobj & __restrict__ vec,const typename vobj::scalar_object & __restrict__ extracted,int lane=SIMTlane(vobj::Nsimd()))
+void coalescedWrite(vobj & __restrict__ vec,const typename vobj::scalar_object & __restrict__ extracted,int lane=acceleratorSIMTlane(vobj::Nsimd()))
 {
   insertLane(lane,vec,extracted);
 }
diff --git a/Grid/tensors/Tensor_class.h b/Grid/tensors/Tensor_class.h
index 75e42721..36becc49 100644
--- a/Grid/tensors/Tensor_class.h
+++ b/Grid/tensors/Tensor_class.h
@@ -6,6 +6,7 @@ Copyright (C) 2015
 Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
 Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 Author: Michael Marshall <michael.marshall@ed.ac.au>
+Author: Christoph Lehner <christoph@lhnr.de>
 
 This program is free software; you can redistribute it and/or modify
 it under the terms of the GNU General Public License as published by
@@ -55,8 +56,23 @@ class GridTensorBase {};
   using Complexified    = typename Traits::Complexified; \
   using Realified       = typename Traits::Realified; \
   using DoublePrecision = typename Traits::DoublePrecision; \
+  using DoublePrecision2= typename Traits::DoublePrecision2; \
   static constexpr int TensorLevel = Traits::TensorLevel
 
+///////////////////////////////////////////////////////////
+// Allows to turn scalar<scalar<scalar<double>>>> back to double.
+///////////////////////////////////////////////////////////
+template <class T>
+accelerator_inline typename std::enable_if<!isGridTensor<T>::value, T>::type
+TensorRemove(T arg) {
+  return arg;
+}
+template <class vtype>
+accelerator_inline auto TensorRemove(iScalar<vtype> arg)
+  -> decltype(TensorRemove(arg._internal)) {
+  return TensorRemove(arg._internal);
+}
+
 template <class vtype>
 class iScalar {
 public:
@@ -133,9 +149,10 @@ public:
   operator ComplexD() const {
     return (TensorRemove(_internal));
   }
+  //             instantiation of "Grid::iScalar<vtype>::operator Grid::RealD() const [with vtype=Grid::Real, U=Grid::Real, V=Grid::RealD, <unnamed>=0, <unnamed>=0U]" 
   template <class U = vtype, class V = scalar_type, IfReal<V> = 0,IfNotSimd<U> = 0> accelerator_inline
   operator RealD() const {
-    return TensorRemove(_internal);
+    return (RealD) TensorRemove(_internal);
   }
   template <class U = vtype, class V = scalar_type, IfInteger<V> = 0, IfNotSimd<U> = 0> accelerator_inline
   operator Integer() const {
@@ -167,20 +184,6 @@ public:
   strong_inline       scalar_type * end()         { return begin() + Traits::count; }
 };
 
-///////////////////////////////////////////////////////////
-// Allows to turn scalar<scalar<scalar<double>>>> back to double.
-///////////////////////////////////////////////////////////
-template <class T>
-accelerator_inline typename std::enable_if<!isGridTensor<T>::value, T>::type
-TensorRemove(T arg) {
-  return arg;
-}
-template <class vtype>
-accelerator_inline auto TensorRemove(iScalar<vtype> arg)
-  -> decltype(TensorRemove(arg._internal)) {
-  return TensorRemove(arg._internal);
-}
-
 template <class vtype, int N>
 class iVector {
 public:
diff --git a/Grid/tensors/Tensor_exp.h b/Grid/tensors/Tensor_exp.h
index 11d37f9c..0a1d6389 100644
--- a/Grid/tensors/Tensor_exp.h
+++ b/Grid/tensors/Tensor_exp.h
@@ -55,7 +55,7 @@ template<class vtype, int N> accelerator_inline iVector<vtype, N> Exponentiate(c
 
 
 // Specialisation: Cayley-Hamilton exponential for SU(3)
-#ifndef GRID_NVCC
+#ifndef GRID_CUDA
 template<class vtype, typename std::enable_if< GridTypeMapper<vtype>::TensorLevel == 0>::type * =nullptr> 
 accelerator_inline iMatrix<vtype,3> Exponentiate(const iMatrix<vtype,3> &arg, RealD alpha  , Integer Nexp = DEFAULT_MAT_EXP )
 {
diff --git a/Grid/tensors/Tensor_inner.h b/Grid/tensors/Tensor_inner.h
index 03f72966..fd651cae 100644
--- a/Grid/tensors/Tensor_inner.h
+++ b/Grid/tensors/Tensor_inner.h
@@ -8,6 +8,7 @@
 
 Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
 Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+Author: Christoph Lehner <christoph@lhnr.de>
 
     This program is free software; you can redistribute it and/or modify
     it under the terms of the GNU General Public License as published by
@@ -194,6 +195,79 @@ auto innerProductD (const iScalar<l>& lhs,const iScalar<r>& rhs) -> iScalar<decl
   ret._internal = innerProductD(lhs._internal,rhs._internal);
   return ret;
 }
+
+
+//////////////////////////////////////
+// innerProductD2: precision promotion without inner sum
+//////////////////////////////////////
+
+accelerator_inline vComplexD2 TensorRemove(const vComplexD2 & x) { return x; };
+accelerator_inline vRealD2 TensorRemove(const vRealD2 & x) { return x; };
+
+accelerator_inline ComplexD innerProductD2(const ComplexF &l,const ComplexF &r){  return innerProduct(l,r); }
+accelerator_inline ComplexD innerProductD2(const ComplexD &l,const ComplexD &r){  return innerProduct(l,r); }
+accelerator_inline RealD    innerProductD2(const RealD    &l,const RealD    &r){  return innerProduct(l,r); }
+accelerator_inline RealD    innerProductD2(const RealF    &l,const RealF    &r){  return innerProduct(l,r); }
+
+accelerator_inline vComplexD innerProductD2(const vComplexD &l,const vComplexD &r){  return innerProduct(l,r); }
+accelerator_inline vRealD    innerProductD2(const vRealD    &l,const vRealD    &r){  return innerProduct(l,r); }
+
+accelerator_inline vComplexD2 innerProductD2(const vComplexF &l,const vComplexF &r)
+{  
+  vComplexD la,lb;
+  vComplexD ra,rb;
+  Optimization::PrecisionChange::StoD(l.v,la.v,lb.v);
+  Optimization::PrecisionChange::StoD(r.v,ra.v,rb.v);
+  vComplexD2 ret;
+  ret._internal[0] = innerProduct(la,ra);
+  ret._internal[1] = innerProduct(lb,rb);
+  return ret;
+}
+accelerator_inline vRealD2 innerProductD2(const vRealF &l,const vRealF &r)
+{  
+  vRealD la,lb;
+  vRealD ra,rb;
+  Optimization::PrecisionChange::StoD(l.v,la.v,lb.v);
+  Optimization::PrecisionChange::StoD(r.v,ra.v,rb.v);
+  vRealD2 ret;
+  ret._internal[0]=innerProduct(la,ra);
+  ret._internal[1]=innerProduct(lb,rb); 
+  return ret;
+}
+
+// Now do it for vector, matrix, scalar
+template<class l,class r,int N> accelerator_inline
+  auto innerProductD2 (const iVector<l,N>& lhs,const iVector<r,N>& rhs) -> iScalar<decltype(innerProductD2(lhs._internal[0],rhs._internal[0]))>
+{
+  typedef decltype(innerProductD2(lhs._internal[0],rhs._internal[0])) ret_t;
+  iScalar<ret_t> ret;
+  zeroit(ret);
+  for(int c1=0;c1<N;c1++){
+    ret._internal += innerProductD2(lhs._internal[c1],rhs._internal[c1]);
+  }
+  return ret;
+}
+template<class l,class r,int N> accelerator_inline
+  auto innerProductD2 (const iMatrix<l,N>& lhs,const iMatrix<r,N>& rhs) -> iScalar<decltype(innerProductD2(lhs._internal[0][0],rhs._internal[0][0]))>
+{
+  typedef decltype(innerProductD2(lhs._internal[0][0],rhs._internal[0][0])) ret_t;
+  iScalar<ret_t> ret;
+  ret=Zero();
+  for(int c1=0;c1<N;c1++){
+    for(int c2=0;c2<N;c2++){
+      ret._internal+=innerProductD2(lhs._internal[c1][c2],rhs._internal[c1][c2]);
+    }}
+  return ret;
+}
+template<class l,class r> accelerator_inline
+  auto innerProductD2 (const iScalar<l>& lhs,const iScalar<r>& rhs) -> iScalar<decltype(innerProductD2(lhs._internal,rhs._internal))>
+{
+  typedef decltype(innerProductD2(lhs._internal,rhs._internal)) ret_t;
+  iScalar<ret_t> ret;
+  ret._internal = innerProductD2(lhs._internal,rhs._internal);
+  return ret;
+}
+
 //////////////////////
 // Keep same precison
 //////////////////////
diff --git a/Grid/tensors/Tensor_traits.h b/Grid/tensors/Tensor_traits.h
index 9067d43d..04d7343e 100644
--- a/Grid/tensors/Tensor_traits.h
+++ b/Grid/tensors/Tensor_traits.h
@@ -6,6 +6,7 @@ Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
 Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 Author: Christopher Kelly <ckelly@phys.columbia.edu>
 Author: Michael Marshall <michael.marshall@ed.ac.au>
+Author: Christoph Lehner <christoph@lhnr.de>
     This program is free software; you can redistribute it and/or modify
     it under the terms of the GNU General Public License as published by
     the Free Software Foundation; either version 2 of the License, or
@@ -37,6 +38,60 @@ NAMESPACE_BEGIN(Grid);
   template<class T, int N> struct isGridTensor<iVector<T, N>> : public std::true_type  { static constexpr bool notvalue = false; };
   template<class T, int N> struct isGridTensor<iMatrix<T, N>> : public std::true_type  { static constexpr bool notvalue = false; };
 
+  // Traits to identify scalars
+  template<typename T>     struct isGridScalar                : public std::false_type { static constexpr bool notvalue = true; };
+  template<class T>        struct isGridScalar<iScalar<T>>    : public std::true_type  { static constexpr bool notvalue = false; };
+
+  // Store double-precision data in single-precision grids for precision promoted localInnerProductD
+  template<typename T>
+  class TypePair {
+  public:
+    T _internal[2];
+    TypePair<T>& operator=(const Grid::Zero& o) {
+      _internal[0] = Zero();
+      _internal[1] = Zero();
+      return *this;
+    }
+
+    TypePair<T> operator+(const TypePair<T>& o) const {
+      TypePair<T> r;
+      r._internal[0] = _internal[0] + o._internal[0];
+      r._internal[1] = _internal[1] + o._internal[1];
+      return r;
+    }
+
+    TypePair<T>& operator+=(const TypePair<T>& o) {
+      _internal[0] += o._internal[0];
+      _internal[1] += o._internal[1];
+      return *this;
+    }
+
+    friend accelerator_inline void add(TypePair<T>* ret, const TypePair<T>* a, const TypePair<T>* b) {
+      add(&ret->_internal[0],&a->_internal[0],&b->_internal[0]);
+      add(&ret->_internal[1],&a->_internal[1],&b->_internal[1]);
+    }
+  };
+  typedef TypePair<ComplexD> ComplexD2;
+  typedef TypePair<RealD> RealD2;
+  typedef TypePair<vComplexD> vComplexD2;
+  typedef TypePair<vRealD> vRealD2;
+
+  // Traits to identify fundamental data types
+  template<typename T>     struct isGridFundamental                : public std::false_type { static constexpr bool notvalue = true; };
+  template<>               struct isGridFundamental<vComplexF>     : public std::true_type  { static constexpr bool notvalue = false; };
+  template<>               struct isGridFundamental<vComplexD>     : public std::true_type  { static constexpr bool notvalue = false; };
+  template<>               struct isGridFundamental<vRealF>        : public std::true_type  { static constexpr bool notvalue = false; };
+  template<>               struct isGridFundamental<vRealD>        : public std::true_type  { static constexpr bool notvalue = false; };
+  template<>               struct isGridFundamental<ComplexF>      : public std::true_type  { static constexpr bool notvalue = false; };
+  template<>               struct isGridFundamental<ComplexD>      : public std::true_type  { static constexpr bool notvalue = false; };
+  template<>               struct isGridFundamental<RealF>         : public std::true_type  { static constexpr bool notvalue = false; };
+  template<>               struct isGridFundamental<RealD>         : public std::true_type  { static constexpr bool notvalue = false; };
+  template<>               struct isGridFundamental<vComplexD2>    : public std::true_type  { static constexpr bool notvalue = false; };
+  template<>               struct isGridFundamental<vRealD2>       : public std::true_type  { static constexpr bool notvalue = false; };
+  template<>               struct isGridFundamental<ComplexD2>     : public std::true_type  { static constexpr bool notvalue = false; };
+  template<>               struct isGridFundamental<RealD2>        : public std::true_type  { static constexpr bool notvalue = false; };
+
+
 //////////////////////////////////////////////////////////////////////////////////
 // Want to recurse: GridTypeMapper<Matrix<vComplexD> >::scalar_type == ComplexD.
 // Use of a helper class like this allows us to template specialise and "dress"
@@ -81,6 +136,7 @@ NAMESPACE_BEGIN(Grid);
     typedef ComplexF Complexified;
     typedef RealF Realified;
     typedef RealD DoublePrecision;
+    typedef RealD2 DoublePrecision2;
   };
   template<> struct GridTypeMapper<RealD> : public GridTypeMapper_Base {
     typedef RealD scalar_type;
@@ -93,6 +149,20 @@ NAMESPACE_BEGIN(Grid);
     typedef ComplexD Complexified;
     typedef RealD Realified;
     typedef RealD DoublePrecision;
+    typedef RealD DoublePrecision2;
+  };
+  template<> struct GridTypeMapper<RealD2> : public GridTypeMapper_Base {
+    typedef RealD2 scalar_type;
+    typedef RealD2 scalar_typeD;
+    typedef RealD2 vector_type;
+    typedef RealD2 vector_typeD;
+    typedef RealD2 tensor_reduced;
+    typedef RealD2 scalar_object;
+    typedef RealD2 scalar_objectD;
+    typedef ComplexD2 Complexified;
+    typedef RealD2 Realified;
+    typedef RealD2 DoublePrecision;
+    typedef RealD2 DoublePrecision2;
   };
   template<> struct GridTypeMapper<ComplexF> : public GridTypeMapper_Base {
     typedef ComplexF scalar_type;
@@ -105,6 +175,7 @@ NAMESPACE_BEGIN(Grid);
     typedef ComplexF Complexified;
     typedef RealF Realified;
     typedef ComplexD DoublePrecision;
+    typedef ComplexD2 DoublePrecision2;
   };
   template<> struct GridTypeMapper<ComplexD> : public GridTypeMapper_Base {
     typedef ComplexD scalar_type;
@@ -117,6 +188,20 @@ NAMESPACE_BEGIN(Grid);
     typedef ComplexD Complexified;
     typedef RealD Realified;
     typedef ComplexD DoublePrecision;
+    typedef ComplexD DoublePrecision2;
+  };
+  template<> struct GridTypeMapper<ComplexD2> : public GridTypeMapper_Base {
+    typedef ComplexD2 scalar_type;
+    typedef ComplexD2 scalar_typeD;
+    typedef ComplexD2 vector_type;
+    typedef ComplexD2 vector_typeD;
+    typedef ComplexD2 tensor_reduced;
+    typedef ComplexD2 scalar_object;
+    typedef ComplexD2 scalar_objectD;
+    typedef ComplexD2 Complexified;
+    typedef RealD2 Realified;
+    typedef ComplexD2 DoublePrecision;
+    typedef ComplexD2 DoublePrecision2;
   };
   template<> struct GridTypeMapper<Integer> : public GridTypeMapper_Base {
     typedef Integer scalar_type;
@@ -129,6 +214,7 @@ NAMESPACE_BEGIN(Grid);
     typedef void Complexified;
     typedef void Realified;
     typedef void DoublePrecision;
+    typedef void DoublePrecision2;
   };
 
   template<> struct GridTypeMapper<vRealF> : public GridTypeMapper_Base {
@@ -142,6 +228,7 @@ NAMESPACE_BEGIN(Grid);
     typedef vComplexF Complexified;
     typedef vRealF Realified;
     typedef vRealD DoublePrecision;
+    typedef vRealD2 DoublePrecision2;
   };
   template<> struct GridTypeMapper<vRealD> : public GridTypeMapper_Base {
     typedef RealD  scalar_type;
@@ -154,6 +241,20 @@ NAMESPACE_BEGIN(Grid);
     typedef vComplexD Complexified;
     typedef vRealD Realified;
     typedef vRealD DoublePrecision;
+    typedef vRealD DoublePrecision2;
+  };
+  template<> struct GridTypeMapper<vRealD2> : public GridTypeMapper_Base {
+    typedef RealD2  scalar_type;
+    typedef RealD2  scalar_typeD;
+    typedef vRealD2 vector_type;
+    typedef vRealD2 vector_typeD;
+    typedef vRealD2 tensor_reduced;
+    typedef RealD2  scalar_object;
+    typedef RealD2  scalar_objectD;
+    typedef vComplexD2 Complexified;
+    typedef vRealD2 Realified;
+    typedef vRealD2 DoublePrecision;
+    typedef vRealD2 DoublePrecision2;
   };
   template<> struct GridTypeMapper<vRealH> : public GridTypeMapper_Base {
     // Fixme this is incomplete until Grid supports fp16 or bfp16 arithmetic types
@@ -167,6 +268,7 @@ NAMESPACE_BEGIN(Grid);
     typedef vComplexH Complexified;
     typedef vRealH Realified;
     typedef vRealD DoublePrecision;
+    typedef vRealD DoublePrecision2;
   };
   template<> struct GridTypeMapper<vComplexH> : public GridTypeMapper_Base {
     // Fixme this is incomplete until Grid supports fp16 or bfp16 arithmetic types
@@ -180,6 +282,7 @@ NAMESPACE_BEGIN(Grid);
     typedef vComplexH Complexified;
     typedef vRealH Realified;
     typedef vComplexD DoublePrecision;
+    typedef vComplexD DoublePrecision2;
   };
   template<> struct GridTypeMapper<vComplexF> : public GridTypeMapper_Base {
     typedef ComplexF  scalar_type;
@@ -192,6 +295,7 @@ NAMESPACE_BEGIN(Grid);
     typedef vComplexF Complexified;
     typedef vRealF Realified;
     typedef vComplexD DoublePrecision;
+    typedef vComplexD2 DoublePrecision2;
   };
   template<> struct GridTypeMapper<vComplexD> : public GridTypeMapper_Base {
     typedef ComplexD  scalar_type;
@@ -204,6 +308,20 @@ NAMESPACE_BEGIN(Grid);
     typedef vComplexD Complexified;
     typedef vRealD Realified;
     typedef vComplexD DoublePrecision;
+    typedef vComplexD DoublePrecision2;
+  };
+  template<> struct GridTypeMapper<vComplexD2> : public GridTypeMapper_Base {
+    typedef ComplexD2  scalar_type;
+    typedef ComplexD2  scalar_typeD;
+    typedef vComplexD2 vector_type;
+    typedef vComplexD2 vector_typeD;
+    typedef vComplexD2 tensor_reduced;
+    typedef ComplexD2  scalar_object;
+    typedef ComplexD2  scalar_objectD;
+    typedef vComplexD2 Complexified;
+    typedef vRealD2 Realified;
+    typedef vComplexD2 DoublePrecision;
+    typedef vComplexD2 DoublePrecision2;
   };
   template<> struct GridTypeMapper<vInteger> : public GridTypeMapper_Base {
     typedef  Integer scalar_type;
@@ -216,6 +334,7 @@ NAMESPACE_BEGIN(Grid);
     typedef void Complexified;
     typedef void Realified;
     typedef void DoublePrecision;
+    typedef void DoublePrecision2;
   };
 
 #define GridTypeMapper_RepeatedTypes \
@@ -234,6 +353,7 @@ NAMESPACE_BEGIN(Grid);
     using Complexified    = iScalar<typename BaseTraits::Complexified>;
     using Realified       = iScalar<typename BaseTraits::Realified>;
     using DoublePrecision = iScalar<typename BaseTraits::DoublePrecision>;
+    using DoublePrecision2= iScalar<typename BaseTraits::DoublePrecision2>;
     static constexpr int Rank = BaseTraits::Rank + 1;
     static constexpr std::size_t count = BaseTraits::count;
     static constexpr int Dimension(int dim) {
@@ -248,6 +368,7 @@ NAMESPACE_BEGIN(Grid);
     using Complexified    = iVector<typename BaseTraits::Complexified,    N>;
     using Realified       = iVector<typename BaseTraits::Realified,       N>;
     using DoublePrecision = iVector<typename BaseTraits::DoublePrecision, N>;
+    using DoublePrecision2= iVector<typename BaseTraits::DoublePrecision2, N>;
     static constexpr int Rank = BaseTraits::Rank + 1;
     static constexpr std::size_t count = BaseTraits::count * N;
     static constexpr int Dimension(int dim) {
@@ -262,6 +383,7 @@ NAMESPACE_BEGIN(Grid);
     using Complexified    = iMatrix<typename BaseTraits::Complexified,    N>;
     using Realified       = iMatrix<typename BaseTraits::Realified,       N>;
     using DoublePrecision = iMatrix<typename BaseTraits::DoublePrecision, N>;
+    using DoublePrecision2= iMatrix<typename BaseTraits::DoublePrecision2, N>;
     static constexpr int Rank = BaseTraits::Rank + 2;
     static constexpr std::size_t count = BaseTraits::count * N * N;
     static constexpr int Dimension(int dim) {
diff --git a/Grid/threads/Accelerator.cc b/Grid/threads/Accelerator.cc
new file mode 100644
index 00000000..2c4ad9df
--- /dev/null
+++ b/Grid/threads/Accelerator.cc
@@ -0,0 +1,207 @@
+#include <Grid/GridCore.h>
+
+NAMESPACE_BEGIN(Grid);
+uint32_t accelerator_threads=2;
+uint32_t acceleratorThreads(void)       {return accelerator_threads;};
+void     acceleratorThreads(uint32_t t) {accelerator_threads = t;};
+
+#ifdef GRID_CUDA
+cudaDeviceProp *gpu_props;
+void acceleratorInit(void)
+{
+  int nDevices = 1;
+  cudaGetDeviceCount(&nDevices);
+  gpu_props = new cudaDeviceProp[nDevices];
+
+  char * localRankStr = NULL;
+  int rank = 0, world_rank=0; 
+#define ENV_LOCAL_RANK_OMPI    "OMPI_COMM_WORLD_LOCAL_RANK"
+#define ENV_LOCAL_RANK_MVAPICH "MV2_COMM_WORLD_LOCAL_RANK"
+#define ENV_RANK_OMPI          "OMPI_COMM_WORLD_RANK"
+#define ENV_RANK_MVAPICH       "MV2_COMM_WORLD_RANK"
+  // We extract the local rank initialization using an environment variable
+  if ((localRankStr = getenv(ENV_LOCAL_RANK_OMPI)) != NULL)
+  {
+    rank = atoi(localRankStr);		
+  }
+  if ((localRankStr = getenv(ENV_LOCAL_RANK_MVAPICH)) != NULL)
+  {
+    rank = atoi(localRankStr);		
+  }
+  if ((localRankStr = getenv(ENV_RANK_OMPI   )) != NULL) { world_rank = atoi(localRankStr);}
+  if ((localRankStr = getenv(ENV_RANK_MVAPICH)) != NULL) { world_rank = atoi(localRankStr);}
+
+  size_t totalDeviceMem=0;
+  for (int i = 0; i < nDevices; i++) {
+
+#define GPU_PROP_FMT(canMapHostMemory,FMT)     printf("AcceleratorCudaInit:   " #canMapHostMemory ": " FMT" \n",prop.canMapHostMemory);
+#define GPU_PROP(canMapHostMemory)             GPU_PROP_FMT(canMapHostMemory,"%d");
+    cudaGetDeviceProperties(&gpu_props[i], i);
+    cudaDeviceProp prop; 
+    prop = gpu_props[i];
+    totalDeviceMem = prop.totalGlobalMem;
+    if ( world_rank == 0) {
+      printf("AcceleratorCudaInit: ========================\n");
+      printf("AcceleratorCudaInit: Device Number    : %d\n", i);
+      printf("AcceleratorCudaInit: ========================\n");
+      printf("AcceleratorCudaInit: Device identifier: %s\n", prop.name);
+
+      GPU_PROP_FMT(totalGlobalMem,"%lld");
+      GPU_PROP(managedMemory);
+      GPU_PROP(isMultiGpuBoard);
+      GPU_PROP(warpSize);
+      //      GPU_PROP(unifiedAddressing);
+      //      GPU_PROP(l2CacheSize);
+      //      GPU_PROP(singleToDoublePrecisionPerfRatio);
+    }
+  }
+  MemoryManager::DeviceMaxBytes = (8*totalDeviceMem)/10; // Assume 80% ours
+#undef GPU_PROP_FMT    
+#undef GPU_PROP
+
+#ifdef GRID_IBM_SUMMIT
+  // IBM Jsrun makes cuda Device numbering screwy and not match rank
+  if ( world_rank == 0 )  printf("AcceleratorCudaInit: IBM Summit or similar - NOT setting device to node rank\n");
+#else
+  if ( world_rank == 0 )  printf("AcceleratorCudaInit: setting device to node rank\n");
+  cudaSetDevice(rank);
+#endif
+  if ( world_rank == 0 )  printf("AcceleratorCudaInit: ================================================\n");
+}
+#endif
+
+#ifdef GRID_HIP
+hipDeviceProp_t *gpu_props;
+void acceleratorInit(void)
+{
+  int nDevices = 1;
+  hipGetDeviceCount(&nDevices);
+  gpu_props = new hipDeviceProp_t[nDevices];
+
+  char * localRankStr = NULL;
+  int rank = 0, world_rank=0; 
+#define ENV_LOCAL_RANK_OMPI    "OMPI_COMM_WORLD_LOCAL_RANK"
+#define ENV_LOCAL_RANK_MVAPICH "MV2_COMM_WORLD_LOCAL_RANK"
+#define ENV_RANK_OMPI          "OMPI_COMM_WORLD_RANK"
+#define ENV_RANK_MVAPICH       "MV2_COMM_WORLD_RANK"
+  // We extract the local rank initialization using an environment variable
+  if ((localRankStr = getenv(ENV_LOCAL_RANK_OMPI)) != NULL)
+  {
+    rank = atoi(localRankStr);		
+  }
+  if ((localRankStr = getenv(ENV_LOCAL_RANK_MVAPICH)) != NULL)
+  {
+    rank = atoi(localRankStr);		
+  }
+  if ((localRankStr = getenv(ENV_RANK_OMPI   )) != NULL) { world_rank = atoi(localRankStr);}
+  if ((localRankStr = getenv(ENV_RANK_MVAPICH)) != NULL) { world_rank = atoi(localRankStr);}
+
+  for (int i = 0; i < nDevices; i++) {
+
+#define GPU_PROP_FMT(canMapHostMemory,FMT)     printf("AcceleratorHipInit:   " #canMapHostMemory ": " FMT" \n",prop.canMapHostMemory);
+#define GPU_PROP(canMapHostMemory)             GPU_PROP_FMT(canMapHostMemory,"%d");
+    
+    hipGetDeviceProperties(&gpu_props[i], i);
+    if ( world_rank == 0) {
+      hipDeviceProp_t prop; 
+      prop = gpu_props[i];
+      printf("AcceleratorHipInit: ========================\n");
+      printf("AcceleratorHipInit: Device Number    : %d\n", i);
+      printf("AcceleratorHipInit: ========================\n");
+      printf("AcceleratorHipInit: Device identifier: %s\n", prop.name);
+
+      //      GPU_PROP(managedMemory);
+      GPU_PROP(isMultiGpuBoard);
+      GPU_PROP(warpSize);
+      //      GPU_PROP(unifiedAddressing);
+      //      GPU_PROP(l2CacheSize);
+      //      GPU_PROP(singleToDoublePrecisionPerfRatio);
+    }
+  }
+#undef GPU_PROP_FMT    
+#undef GPU_PROP
+#ifdef GRID_IBM_SUMMIT
+  // IBM Jsrun makes cuda Device numbering screwy and not match rank
+  if ( world_rank == 0 )  printf("AcceleratorHipInit: IBM Summit or similar - NOT setting device to node rank\n");
+#else
+  if ( world_rank == 0 )  printf("AcceleratorHipInit: setting device to node rank\n");
+  hipSetDevice(rank);
+#endif
+  if ( world_rank == 0 )  printf("AcceleratorHipInit: ================================================\n");
+}
+#endif
+
+
+#ifdef GRID_SYCL
+
+cl::sycl::queue *theGridAccelerator;
+
+void acceleratorInit(void)
+{
+  int nDevices = 1;
+  cl::sycl::gpu_selector selector;
+  cl::sycl::device selectedDevice { selector };
+  theGridAccelerator = new sycl::queue (selectedDevice);
+
+  char * localRankStr = NULL;
+  int rank = 0, world_rank=0; 
+#define ENV_LOCAL_RANK_OMPI    "OMPI_COMM_WORLD_LOCAL_RANK"
+#define ENV_LOCAL_RANK_MVAPICH "MV2_COMM_WORLD_LOCAL_RANK"
+#define ENV_RANK_OMPI          "OMPI_COMM_WORLD_RANK"
+#define ENV_RANK_MVAPICH       "MV2_COMM_WORLD_RANK"
+  // We extract the local rank initialization using an environment variable
+  if ((localRankStr = getenv(ENV_LOCAL_RANK_OMPI)) != NULL)
+  {
+    rank = atoi(localRankStr);		
+  }
+  if ((localRankStr = getenv(ENV_LOCAL_RANK_MVAPICH)) != NULL)
+  {
+    rank = atoi(localRankStr);		
+  }
+  if ((localRankStr = getenv(ENV_RANK_OMPI   )) != NULL) { world_rank = atoi(localRankStr);}
+  if ((localRankStr = getenv(ENV_RANK_MVAPICH)) != NULL) { world_rank = atoi(localRankStr);}
+
+  auto devices = cl::sycl::device::get_devices();
+  for(int d = 0;d<devices.size();d++){
+
+#define GPU_PROP_STR(prop) \
+    printf("AcceleratorSyclInit:   " #prop ": %s \n",devices[d].get_info<cl::sycl::info::device::prop>().c_str());
+
+#define GPU_PROP_FMT(prop,FMT) \
+    printf("AcceleratorSyclInit:   " #prop ": " FMT" \n",devices[d].get_info<cl::sycl::info::device::prop>());
+
+#define GPU_PROP(prop)             GPU_PROP_FMT(prop,"%ld");
+
+    GPU_PROP_STR(vendor);
+    GPU_PROP_STR(version);
+    //    GPU_PROP_STR(device_type);
+    /*
+    GPU_PROP(max_compute_units);
+    GPU_PROP(native_vector_width_char);
+    GPU_PROP(native_vector_width_short);
+    GPU_PROP(native_vector_width_int);
+    GPU_PROP(native_vector_width_long);
+    GPU_PROP(native_vector_width_float);
+    GPU_PROP(native_vector_width_double);
+    GPU_PROP(native_vector_width_half);
+    GPU_PROP(address_bits);
+    GPU_PROP(half_fp_config);
+    GPU_PROP(single_fp_config);
+    */
+    //    GPU_PROP(double_fp_config);
+    GPU_PROP(global_mem_size);
+
+  }
+  if ( world_rank == 0 ) {
+    auto name = theGridAccelerator->get_device().get_info<sycl::info::device::name>();
+    printf("AcceleratorSyclInit: Selected device is %s\n",name.c_str());
+    printf("AcceleratorSyclInit: ================================================\n");
+  }
+}
+#endif
+
+#if (!defined(GRID_CUDA)) && (!defined(GRID_SYCL))&& (!defined(GRID_HIP))
+void acceleratorInit(void){}
+#endif
+
+NAMESPACE_END(Grid);
diff --git a/Grid/threads/Accelerator.h b/Grid/threads/Accelerator.h
new file mode 100644
index 00000000..74a3ea22
--- /dev/null
+++ b/Grid/threads/Accelerator.h
@@ -0,0 +1,426 @@
+/*************************************************************************************
+
+    Grid physics library, www.github.com/paboyle/Grid 
+
+    Source file: ./lib/Accelerator.h
+
+    Copyright (C) 2015
+
+Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+Author: paboyle <paboyle@ph.ed.ac.uk>
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License along
+    with this program; if not, write to the Free Software Foundation, Inc.,
+    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+    See the full license in the file "LICENSE" in the top level distribution directory
+*************************************************************************************/
+/*  END LEGAL */
+#pragma once
+
+#include <string.h>
+
+#ifdef HAVE_MALLOC_MALLOC_H
+#include <malloc/malloc.h>
+#endif
+#ifdef HAVE_MALLOC_H
+#include <malloc.h>
+#endif
+#ifdef HAVE_MM_MALLOC_H
+#include <mm_malloc.h>
+#endif
+
+NAMESPACE_BEGIN(Grid);
+
+//////////////////////////////////////////////////////////////////////////////////
+// Accelerator primitives; fall back to threading if not CUDA or SYCL
+//////////////////////////////////////////////////////////////////////////////////
+//
+// Function attributes
+//
+//    accelerator
+//    accelerator_inline
+//
+// Parallel looping
+// 
+//    accelerator_for
+//    accelerator_forNB 
+//    uint32_t accelerator_barrier();         // device synchronise
+//
+// Parallelism control: Number of threads in thread block is acceleratorThreads*Nsimd
+//
+//    uint32_t acceleratorThreads(void);   
+//    void     acceleratorThreads(uint32_t);
+//
+// Warp control and info:
+//
+//    acceleratorInit;
+//    void     acceleratorSynchronise(void); // synch warp etc..
+//    int      acceleratorSIMTlane(int Nsimd);
+//
+// Memory management:
+//
+//    void *acceleratorAllocShared(size_t bytes);
+//    void acceleratorFreeShared(void *ptr);
+//
+//    void *acceleratorAllocDevice(size_t bytes);
+//    void acceleratorFreeDevice(void *ptr);
+//
+//    void *acceleratorCopyToDevice(void *from,void *to,size_t bytes);
+//    void *acceleratorCopyFromDevice(void *from,void *to,size_t bytes);
+//
+//////////////////////////////////////////////////////////////////////////////////
+
+uint32_t acceleratorThreads(void);   
+void     acceleratorThreads(uint32_t);
+void     acceleratorInit(void);
+
+//////////////////////////////////////////////
+// CUDA acceleration
+//////////////////////////////////////////////
+
+#ifdef GRID_CUDA
+
+#ifdef __CUDA_ARCH__
+#define GRID_SIMT
+#endif
+
+#define accelerator        __host__ __device__
+#define accelerator_inline __host__ __device__ inline
+
+accelerator_inline int acceleratorSIMTlane(int Nsimd) {
+#ifdef GRID_SIMT
+  return threadIdx.z; 
+#else
+  return 0;
+#endif
+} // CUDA specific
+
+#define accelerator_for2dNB( iter1, num1, iter2, num2, nsimd, ... )	\
+  {									\
+    typedef uint64_t Iterator;						\
+    auto lambda = [=] accelerator					\
+      (Iterator iter1,Iterator iter2,Iterator lane) mutable {		\
+      __VA_ARGS__;							\
+    };									\
+    int nt=acceleratorThreads();					\
+    dim3 cu_threads(acceleratorThreads(),1,nsimd);			\
+    dim3 cu_blocks ((num1+nt-1)/nt,num2,1);				\
+    LambdaApply<<<cu_blocks,cu_threads>>>(num1,num2,nsimd,lambda);	\
+  }
+
+template<typename lambda>  __global__
+void LambdaApply(uint64_t num1, uint64_t num2, uint64_t num3, lambda Lambda)
+{
+  uint64_t x = threadIdx.x + blockDim.x*blockIdx.x;
+  uint64_t y = threadIdx.y + blockDim.y*blockIdx.y;
+  uint64_t z = threadIdx.z;
+  if ( (x < num1) && (y<num2) && (z<num3) ) {
+    Lambda(x,y,z);
+  }
+}
+
+#define accelerator_barrier(dummy)					\
+  {									\
+    cudaDeviceSynchronize();						\
+    cudaError err = cudaGetLastError();					\
+    if ( cudaSuccess != err ) {						\
+      printf("Cuda error %s \n", cudaGetErrorString( err ));		\
+      puts(__FILE__);							\
+      printf("Line %d\n",__LINE__);					\
+    }									\
+  }
+
+inline void *acceleratorAllocShared(size_t bytes)
+{
+  void *ptr=NULL;
+  auto err = cudaMallocManaged((void **)&ptr,bytes);
+  if( err != cudaSuccess ) {
+    ptr = (void *) NULL;
+    printf(" cudaMallocManaged failed for %d %s \n",bytes,cudaGetErrorString(err));
+  }
+  return ptr;
+};
+inline void *acceleratorAllocDevice(size_t bytes)
+{
+  void *ptr=NULL;
+  auto err = cudaMalloc((void **)&ptr,bytes);
+  if( err != cudaSuccess ) {
+    ptr = (void *) NULL;
+    printf(" cudaMalloc failed for %d %s \n",bytes,cudaGetErrorString(err));
+  }
+  return ptr;
+};
+inline void acceleratorFreeShared(void *ptr){ cudaFree(ptr);};
+inline void acceleratorFreeDevice(void *ptr){ cudaFree(ptr);};
+inline void acceleratorCopyToDevice(void *from,void *to,size_t bytes)  { cudaMemcpy(to,from,bytes, cudaMemcpyHostToDevice);}
+inline void acceleratorCopyFromDevice(void *from,void *to,size_t bytes){ cudaMemcpy(to,from,bytes, cudaMemcpyDeviceToHost);}
+#endif
+
+//////////////////////////////////////////////
+// SyCL acceleration
+//////////////////////////////////////////////
+
+#ifdef GRID_SYCL
+NAMESPACE_END(Grid);
+#include <CL/sycl.hpp>
+#include <CL/sycl/usm.hpp>
+NAMESPACE_BEGIN(Grid);
+
+extern cl::sycl::queue *theGridAccelerator;
+
+#ifdef __SYCL_DEVICE_ONLY__
+#define GRID_SIMT
+#endif
+
+#define accelerator 
+#define accelerator_inline strong_inline
+
+accelerator_inline int acceleratorSIMTlane(int Nsimd) {
+#ifdef GRID_SIMT
+ return __spirv::initLocalInvocationId<3, cl::sycl::id<3>>()[2]; 
+#else
+ return 0;
+#endif
+} // SYCL specific
+
+#define accelerator_for2dNB( iter1, num1, iter2, num2, nsimd, ... )	\
+  theGridAccelerator->submit([&](cl::sycl::handler &cgh) {		\
+      unsigned long nt=acceleratorThreads();				\
+      unsigned long unum1 = num1;					\
+      unsigned long unum2 = num2;					\
+      cl::sycl::range<3> local {nt,1,nsimd};				\
+      cl::sycl::range<3> global{unum1,unum2,nsimd};			\
+      cgh.parallel_for<class dslash>(					\
+      cl::sycl::nd_range<3>(global,local), \
+      [=] (cl::sycl::nd_item<3> item) mutable {       \
+      auto iter1    = item.get_global_id(0);	      \
+      auto iter2    = item.get_global_id(1);	      \
+      auto lane     = item.get_global_id(2);	      \
+      { __VA_ARGS__ };				      \
+     });	   			              \
+    });
+
+#define accelerator_barrier(dummy) theGridAccelerator->wait();
+
+inline void *acceleratorAllocShared(size_t bytes){ return malloc_shared(bytes,*theGridAccelerator);};
+inline void *acceleratorAllocDevice(size_t bytes){ return malloc_device(bytes,*theGridAccelerator);};
+inline void acceleratorFreeShared(void *ptr){free(ptr,*theGridAccelerator);};
+inline void acceleratorFreeDevice(void *ptr){free(ptr,*theGridAccelerator);};
+inline void acceleratorCopyToDevice(void *from,void *to,size_t bytes)  { theGridAccelerator->memcpy(to,from,bytes); theGridAccelerator->wait();}
+inline void acceleratorCopyFromDevice(void *from,void *to,size_t bytes){ theGridAccelerator->memcpy(to,from,bytes); theGridAccelerator->wait();}
+
+#endif
+
+//////////////////////////////////////////////
+// HIP acceleration
+//////////////////////////////////////////////
+#ifdef GRID_HIP
+NAMESPACE_END(Grid);
+#include <hip/hip_runtime.h>
+NAMESPACE_BEGIN(Grid);
+
+#ifdef __HIP_DEVICE_COMPILE__
+#define GRID_SIMT
+#endif
+
+#define accelerator        __host__ __device__
+#define accelerator_inline __host__ __device__ inline
+
+/*These routines define mapping from thread grid to loop & vector lane indexing */
+accelerator_inline int acceleratorSIMTlane(int Nsimd) {
+#ifdef GRID_SIMT
+  return hipThreadIdx_z; 
+#else
+  return 0;
+#endif
+} // HIP specific
+
+#define accelerator_for2dNB( iter1, num1, iter2, num2, nsimd, ... )	\
+  {									\
+    typedef uint64_t Iterator;						\
+    auto lambda = [=] accelerator					\
+      (Iterator iter1,Iterator iter2,Iterator lane ) mutable {		\
+      { __VA_ARGS__;}							\
+    };									\
+    int nt=acceleratorThreads();					\
+    dim3 hip_threads(nt,1,nsimd);					\
+    dim3 hip_blocks ((num1+nt-1)/nt,num2,1);				\
+    hipLaunchKernelGGL(LambdaApply,hip_blocks,hip_threads,		\
+		       0,0,						\
+		       num1,num2,nsimd,lambda);				\
+  }
+
+template<typename lambda>  __global__
+void LambdaApply(uint64_t numx, uint64_t numy, uint64_t numz, lambda Lambda)
+{
+  uint64_t x = hipThreadIdx_x + hipBlockDim_x*hipBlockIdx_x;
+  uint64_t y = hipThreadIdx_y + hipBlockDim_y*hipBlockIdx_y;
+  uint64_t z = hipThreadIdx_z ;//+ hipBlockDim_z*hipBlockIdx_z;
+  if ( (x < numx) && (y<numy) && (z<numz) ) {
+    Lambda(x,y,z);
+  }
+}
+
+#define accelerator_barrier(dummy)				\
+  {								\
+    hipDeviceSynchronize();					\
+    auto err = hipGetLastError();				\
+    if ( err != hipSuccess ) {					\
+      printf("After hipDeviceSynchronize() : HIP error %s \n", hipGetErrorString( err )); \
+      puts(__FILE__);							\
+      printf("Line %d\n",__LINE__);				\
+      exit(0);							\
+    }								\
+  }
+
+inline void *acceleratorAllocShared(size_t bytes)
+{
+#if 0
+  void *ptr=NULL;
+  auto err = hipMallocManaged((void **)&ptr,bytes);
+  if( err != hipSuccess ) {
+    ptr = (void *) NULL;
+    printf(" hipMallocManaged failed for %d %s \n",bytes,hipGetErrorString(err));
+  }
+  return ptr;
+#else
+  return malloc(bytes);
+#endif
+};
+
+inline void *acceleratorAllocDevice(size_t bytes)
+{
+  void *ptr=NULL;
+  auto err = hipMalloc((void **)&ptr,bytes);
+  if( err != hipSuccess ) {
+    ptr = (void *) NULL;
+    printf(" hipMalloc failed for %d %s \n",bytes,hipGetErrorString(err));
+  }
+  return ptr;
+};
+
+inline void acceleratorFreeShared(void *ptr){ free(ptr);};
+inline void acceleratorFreeDevice(void *ptr){ hipFree(ptr);};
+inline void acceleratorCopyToDevice(void *from,void *to,size_t bytes)  { hipMemcpy(to,from,bytes, hipMemcpyHostToDevice);}
+inline void acceleratorCopyFromDevice(void *from,void *to,size_t bytes){ hipMemcpy(to,from,bytes, hipMemcpyDeviceToHost);}
+
+#endif
+
+//////////////////////////////////////////////
+// Common on all GPU targets
+//////////////////////////////////////////////
+#if defined(GRID_SYCL) || defined(GRID_CUDA) || defined(GRID_HIP)
+#define accelerator_forNB( iter1, num1, nsimd, ... ) accelerator_for2dNB( iter1, num1, iter2, 1, nsimd, {__VA_ARGS__} );
+
+#define accelerator_for( iter, num, nsimd, ... )		\
+  accelerator_forNB(iter, num, nsimd, { __VA_ARGS__ } );	\
+  accelerator_barrier(dummy);
+
+#define accelerator_for2d(iter1, num1, iter2, num2, nsimd, ... )	\
+  accelerator_for2dNB(iter1, num1, iter2, num2, nsimd, { __VA_ARGS__ } ); \
+  accelerator_barrier(dummy);
+
+#endif
+
+//////////////////////////////////////////////
+// CPU Target - No accelerator just thread instead
+//////////////////////////////////////////////
+#define GRID_ALLOC_ALIGN (2*1024*1024) // 2MB aligned 
+#if ( (!defined(GRID_SYCL)) && (!defined(GRID_CUDA)) && (!defined(GRID_HIP)) )
+
+#undef GRID_SIMT
+
+#define accelerator 
+#define accelerator_inline strong_inline
+#define accelerator_for(iterator,num,nsimd, ... )   thread_for(iterator, num, { __VA_ARGS__ });
+#define accelerator_forNB(iterator,num,nsimd, ... ) thread_for(iterator, num, { __VA_ARGS__ });
+#define accelerator_barrier(dummy) 
+#define accelerator_for2d(iter1, num1, iter2, num2, nsimd, ... ) thread_for2d(iter1,num1,iter2,num2,{ __VA_ARGS__ });
+
+accelerator_inline int acceleratorSIMTlane(int Nsimd) { return 0; } // CUDA specific
+inline void acceleratorCopyToDevice(void *from,void *to,size_t bytes)  { memcpy(to,from,bytes);}
+inline void acceleratorCopyFromDevice(void *from,void *to,size_t bytes){ memcpy(to,from,bytes);}
+
+#ifdef HAVE_MM_MALLOC_H
+inline void *acceleratorAllocShared(size_t bytes){return _mm_malloc(bytes,GRID_ALLOC_ALIGN);};
+inline void *acceleratorAllocDevice(size_t bytes){return _mm_malloc(bytes,GRID_ALLOC_ALIGN);};
+inline void acceleratorFreeShared(void *ptr){_mm_free(ptr);};
+inline void acceleratorFreeDevice(void *ptr){_mm_free(ptr);};
+#else
+inline void *acceleratorAllocShared(size_t bytes){return memalign(GRID_ALLOC_ALIGN,bytes);};
+inline void *acceleratorAllocDevice(size_t bytes){return memalign(GRID_ALLOC_ALIGN,bytes);};
+inline void acceleratorFreeShared(void *ptr){free(ptr);};
+inline void acceleratorFreeDevice(void *ptr){free(ptr);};
+#endif
+
+#endif // CPU target
+
+#ifdef HAVE_MM_MALLOC_H
+inline void *acceleratorAllocCpu(size_t bytes){return _mm_malloc(bytes,GRID_ALLOC_ALIGN);};
+inline void acceleratorFreeCpu  (void *ptr){_mm_free(ptr);};
+#else
+inline void *acceleratorAllocCpu(size_t bytes){return memalign(GRID_ALLOC_ALIGN,bytes);};
+inline void acceleratorFreeCpu  (void *ptr){free(ptr);};
+#endif
+
+///////////////////////////////////////////////////
+// Synchronise across local threads for divergence resynch
+///////////////////////////////////////////////////
+accelerator_inline void acceleratorSynchronise(void) 
+{
+#ifdef GRID_SIMT
+#ifdef GRID_CUDA
+  __syncwarp();
+#endif
+#ifdef GRID_SYCL
+  // No barrier call on SYCL??  // Option get __spir:: stuff to do warp barrier
+#endif
+#ifdef GRID_HIP
+  __syncthreads();
+#endif
+#endif
+  return;
+}
+accelerator_inline void acceleratorSynchroniseAll(void) 
+{
+#ifdef GRID_SIMT
+#ifdef GRID_CUDA
+  __syncthreads();
+#endif
+#ifdef GRID_SYCL
+  // No barrier call on SYCL??  // Option get __spir:: stuff to do warp barrier
+#endif
+#ifdef GRID_HIP
+  __syncthreads();
+#endif
+#endif
+  return;
+}
+accelerator_inline void acceleratorFence(void) 
+{
+#ifdef GRID_SIMT
+#ifdef GRID_CUDA
+  __threadfence();
+#endif
+#ifdef GRID_SYCL
+  // FIXMEE
+#endif
+#ifdef GRID_HIP
+  __threadfence();
+#endif
+#endif
+  return;
+}
+
+NAMESPACE_END(Grid);
diff --git a/Grid/threads/Pragmas.h b/Grid/threads/Pragmas.h
index d05f8ee9..f311bf9b 100644
--- a/Grid/threads/Pragmas.h
+++ b/Grid/threads/Pragmas.h
@@ -2,7 +2,7 @@
 
     Grid physics library, www.github.com/paboyle/Grid 
 
-    Source file: ./lib/Threads.h
+    Source file: ./lib/Pragmas.h
 
     Copyright (C) 2015
 
@@ -28,107 +28,5 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 /*  END LEGAL */
 #pragma once
 
-#ifndef MAX
-#define MAX(x,y) ((x)>(y)?(x):(y))
-#define MIN(x,y) ((x)>(y)?(y):(x))
-#endif
-
-#define strong_inline     __attribute__((always_inline)) inline
-#define UNROLL  _Pragma("unroll")
-
-//////////////////////////////////////////////////////////////////////////////////
-// New primitives; explicit host thread calls, and accelerator data parallel calls
-//////////////////////////////////////////////////////////////////////////////////
-
-#ifdef _OPENMP
-#define GRID_OMP
-#include <omp.h>
-#endif
-
-#ifdef GRID_OMP
-#define DO_PRAGMA_(x) _Pragma (#x)
-#define DO_PRAGMA(x) DO_PRAGMA_(x)
-#define thread_num(a) omp_get_thread_num()
-#define thread_max(a) omp_get_max_threads()
-#else 
-#define DO_PRAGMA_(x) 
-#define DO_PRAGMA(x) 
-#define thread_num(a) (0)
-#define thread_max(a) (1)
-#endif
-
-#define thread_for( i, num, ... )                           DO_PRAGMA(omp parallel for schedule(static)) for ( uint64_t i=0;i<num;i++) { __VA_ARGS__ } ;
-#define thread_foreach( i, container, ... )                 DO_PRAGMA(omp parallel for schedule(static)) for ( uint64_t i=container.begin();i<container.end();i++) { __VA_ARGS__ } ;
-#define thread_for_in_region( i, num, ... )                 DO_PRAGMA(omp for schedule(static))          for ( uint64_t i=0;i<num;i++) { __VA_ARGS__ } ;
-#define thread_for_collapse2( i, num, ... )                 DO_PRAGMA(omp parallel for collapse(2))      for ( uint64_t i=0;i<num;i++) { __VA_ARGS__ } ;
-#define thread_for_collapse( N , i, num, ... )              DO_PRAGMA(omp parallel for collapse ( N ) )  for ( uint64_t i=0;i<num;i++) { __VA_ARGS__ } ;
-#define thread_for_collapse_in_region( N , i, num, ... )    DO_PRAGMA(omp for collapse ( N ))            for ( uint64_t i=0;i<num;i++) { __VA_ARGS__ } ;
-#define thread_region                                       DO_PRAGMA(omp parallel)
-#define thread_critical                                     DO_PRAGMA(omp critical)
-
-
-//////////////////////////////////////////////////////////////////////////////////
-// Accelerator primitives; fall back to threading
-//////////////////////////////////////////////////////////////////////////////////
-#ifdef __NVCC__
-#define GRID_NVCC
-#endif
-
-#ifdef GRID_NVCC
-
-extern uint32_t gpu_threads;
-
-#define accelerator        __host__ __device__
-#define accelerator_inline __host__ __device__ inline
-
-template<typename lambda>  __global__
-void LambdaApplySIMT(uint64_t Isites, uint64_t Osites, lambda Lambda)
-{
-  uint64_t isite = threadIdx.y;
-  uint64_t osite = threadIdx.x+blockDim.x*blockIdx.x;
-  if ( (osite <Osites) && (isite<Isites) ) {
-    Lambda(isite,osite);
-  }
-}
-
-/////////////////////////////////////////////////////////////////
-// Internal only really... but need to call when 
-/////////////////////////////////////////////////////////////////
-#define accelerator_barrier(dummy)				\
-  {								\
-    cudaDeviceSynchronize();					\
-    cudaError err = cudaGetLastError();				\
-    if ( cudaSuccess != err ) {					\
-      printf("Cuda error %s \n", cudaGetErrorString( err )); \
-      puts(__FILE__); \
-      printf("Line %d\n",__LINE__);					\
-      exit(0);							\
-    }								\
-  }
-
-// Copy the for_each_n style ; Non-blocking variant
-#define accelerator_forNB( iterator, num, nsimd, ... )			\
-  {									\
-    typedef uint64_t Iterator;						\
-    auto lambda = [=] accelerator (Iterator lane,Iterator iterator) mutable { \
-      __VA_ARGS__;							\
-    };									\
-    dim3 cu_threads(gpu_threads,nsimd);					\
-    dim3 cu_blocks ((num+gpu_threads-1)/gpu_threads);			\
-    LambdaApplySIMT<<<cu_blocks,cu_threads>>>(nsimd,num,lambda);	\
-  }
-
-// Copy the for_each_n style ; Non-blocking variant (default
-#define accelerator_for( iterator, num, nsimd, ... )		\
-  accelerator_forNB(iterator, num, nsimd, { __VA_ARGS__ } );	\
-  accelerator_barrier(dummy);
-
-#else
-
-#define accelerator 
-#define accelerator_inline strong_inline
-#define accelerator_for(iterator,num,nsimd, ... )   thread_for(iterator, num, { __VA_ARGS__ });
-#define accelerator_forNB(iterator,num,nsimd, ... ) thread_for(iterator, num, { __VA_ARGS__ });
-#define accelerator_barrier(dummy) 
-
-#endif
+#include <Grid/threads/Threads.h>
+#include <Grid/threads/Accelerator.h>
diff --git a/Grid/threads/ThreadReduction.h b/Grid/threads/ThreadReduction.h
new file mode 100644
index 00000000..f0d24d50
--- /dev/null
+++ b/Grid/threads/ThreadReduction.h
@@ -0,0 +1,127 @@
+/*************************************************************************************
+
+    Grid physics library, www.github.com/paboyle/Grid 
+
+    Source file: ./lib/ThreadReduction.h
+
+    Copyright (C) 2015
+
+Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+Author: paboyle <paboyle@ph.ed.ac.uk>
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License along
+    with this program; if not, write to the Free Software Foundation, Inc.,
+    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+    See the full license in the file "LICENSE" in the top level distribution directory
+*************************************************************************************/
+/*  END LEGAL */
+#pragma once 
+
+// Introduce a class to gain deterministic bit reproducible reduction.
+// make static; perhaps just a namespace is required.
+NAMESPACE_BEGIN(Grid);
+
+class GridThread {
+public:
+  static int _threads;
+  static int _hyperthreads;
+  static int _cores;
+
+  static void SetCores(int cr) { 
+#ifdef GRID_OMP
+    _cores = cr;
+#else 
+    _cores = 1;
+#endif
+  }
+  static void SetThreads(int thr) { 
+#ifdef GRID_OMP
+    _threads = MIN(thr,omp_get_max_threads()) ;
+    omp_set_num_threads(_threads);
+#else 
+    _threads = 1;
+#endif
+  };
+  static void SetMaxThreads(void) { 
+#ifdef GRID_OMP
+    _threads = omp_get_max_threads();
+    omp_set_num_threads(_threads);
+#else 
+    _threads = 1;
+#endif
+  };
+  static int GetHyperThreads(void) { assert(_threads%_cores ==0); return _threads/_cores; };
+  static int GetCores(void)   { return _cores; };
+  static int GetThreads(void) { return _threads; };
+  static int SumArraySize(void) {return _threads;};
+
+  static void GetWork(int nwork, int me, int & mywork, int & myoff){
+    GetWork(nwork,me,mywork,myoff,_threads);
+  }
+  static void GetWork(int nwork, int me, int & mywork, int & myoff,int units){
+    int basework = nwork/units;
+    int backfill = units-(nwork%units);
+    if ( me >= units ) { 
+      mywork = myoff = 0;
+    } else { 
+      mywork = (nwork+me)/units;
+      myoff  = basework * me;
+      if ( me > backfill ) 
+	myoff+= (me-backfill);
+    }
+    return;
+  };
+
+  static void GetWorkBarrier(int nwork, int &me, int & mywork, int & myoff){
+    me     = ThreadBarrier();
+    GetWork(nwork,me,mywork,myoff);
+  };
+
+  static int  ThreadBarrier(void) {
+#ifdef GRID_OMP
+#pragma omp barrier
+    return omp_get_thread_num();
+#else
+    return 0;
+#endif
+  };
+  
+  template<class obj> static void ThreadSum( std::vector<obj> &sum_array,obj &val,int me){
+    sum_array[me] = val;
+    val=Zero();
+    ThreadBarrier();
+    for(int i=0;i<_threads;i++) val+= sum_array[i];
+    ThreadBarrier();
+  }
+
+  static void bcopy(const void *src, void *dst, size_t len) {
+#ifdef GRID_OMP
+#pragma omp parallel 
+    {
+      const char *c_src =(char *) src;
+      char *c_dest=(char *) dst;
+      int me,mywork,myoff;
+      GridThread::GetWorkBarrier(len,me, mywork,myoff);
+      bcopy(&c_src[myoff],&c_dest[myoff],mywork);
+    }
+#else 
+    bcopy(src,dst,len);
+#endif
+  }
+
+
+};
+
+NAMESPACE_END(Grid);
+
diff --git a/Grid/threads/Threads.h b/Grid/threads/Threads.h
index 29cae060..a9fa13ea 100644
--- a/Grid/threads/Threads.h
+++ b/Grid/threads/Threads.h
@@ -28,101 +28,47 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 /*  END LEGAL */
 #pragma once 
 
+#ifndef MAX
+#define MAX(x,y) ((x)>(y)?(x):(y))
+#define MIN(x,y) ((x)>(y)?(y):(x))
+#endif
 
-// Introduce a class to gain deterministic bit reproducible reduction.
-// make static; perhaps just a namespace is required.
-NAMESPACE_BEGIN(Grid);
+#define strong_inline     __attribute__((always_inline)) inline
+#define UNROLL  _Pragma("unroll")
 
-class GridThread {
-public:
-  static int _threads;
-  static int _hyperthreads;
-  static int _cores;
+//////////////////////////////////////////////////////////////////////////////////
+// New primitives; explicit host thread calls, and accelerator data parallel calls
+//////////////////////////////////////////////////////////////////////////////////
+
+#ifdef _OPENMP
+#define GRID_OMP
+#include <omp.h>
+#endif
 
-  static void SetCores(int cr) { 
 #ifdef GRID_OMP
-    _cores = cr;
+#define DO_PRAGMA_(x) _Pragma (#x)
+#define DO_PRAGMA(x) DO_PRAGMA_(x)
+#define thread_num(a) omp_get_thread_num()
+#define thread_max(a) omp_get_max_threads()
 #else 
-    _cores = 1;
+#define DO_PRAGMA_(x) 
+#define DO_PRAGMA(x) 
+#define thread_num(a) (0)
+#define thread_max(a) (1)
 #endif
-  }
-  static void SetThreads(int thr) { 
-#ifdef GRID_OMP
-    _threads = MIN(thr,omp_get_max_threads()) ;
-    omp_set_num_threads(_threads);
-#else 
-    _threads = 1;
-#endif
-  };
-  static void SetMaxThreads(void) { 
-#ifdef GRID_OMP
-    _threads = omp_get_max_threads();
-    omp_set_num_threads(_threads);
-#else 
-    _threads = 1;
-#endif
-  };
-  static int GetHyperThreads(void) { assert(_threads%_cores ==0); return _threads/_cores; };
-  static int GetCores(void)   { return _cores; };
-  static int GetThreads(void) { return _threads; };
-  static int SumArraySize(void) {return _threads;};
 
-  static void GetWork(int nwork, int me, int & mywork, int & myoff){
-    GetWork(nwork,me,mywork,myoff,_threads);
-  }
-  static void GetWork(int nwork, int me, int & mywork, int & myoff,int units){
-    int basework = nwork/units;
-    int backfill = units-(nwork%units);
-    if ( me >= units ) { 
-      mywork = myoff = 0;
-    } else { 
-      mywork = (nwork+me)/units;
-      myoff  = basework * me;
-      if ( me > backfill ) 
-	myoff+= (me-backfill);
-    }
-    return;
-  };
-
-  static void GetWorkBarrier(int nwork, int &me, int & mywork, int & myoff){
-    me     = ThreadBarrier();
-    GetWork(nwork,me,mywork,myoff);
-  };
-
-  static int  ThreadBarrier(void) {
-#ifdef GRID_OMP
-#pragma omp barrier
-    return omp_get_thread_num();
-#else
-    return 0;
-#endif
-  };
-  
-  template<class obj> static void ThreadSum( std::vector<obj> &sum_array,obj &val,int me){
-    sum_array[me] = val;
-    val=Zero();
-    ThreadBarrier();
-    for(int i=0;i<_threads;i++) val+= sum_array[i];
-    ThreadBarrier();
-  }
-
-  static void bcopy(const void *src, void *dst, size_t len) {
-#ifdef GRID_OMP
-#pragma omp parallel 
-    {
-      const char *c_src =(char *) src;
-      char *c_dest=(char *) dst;
-      int me,mywork,myoff;
-      GridThread::GetWorkBarrier(len,me, mywork,myoff);
-      bcopy(&c_src[myoff],&c_dest[myoff],mywork);
-    }
-#else 
-    bcopy(src,dst,len);
-#endif
-  }
-
-
-};
-
-NAMESPACE_END(Grid);
+#define thread_for( i, num, ... )                           DO_PRAGMA(omp parallel for schedule(static)) for ( uint64_t i=0;i<num;i++) { __VA_ARGS__ } ;
+#define thread_for2d( i1, n1,i2,n2, ... )  \
+  DO_PRAGMA(omp parallel for collapse(2))  \
+  for ( uint64_t i1=0;i1<n1;i1++) {	   \
+  for ( uint64_t i2=0;i2<n2;i2++) {	   \
+  { __VA_ARGS__ } ;			   \
+  }}
+#define thread_foreach( i, container, ... )                 DO_PRAGMA(omp parallel for schedule(static)) for ( uint64_t i=container.begin();i<container.end();i++) { __VA_ARGS__ } ;
+#define thread_for_in_region( i, num, ... )                 DO_PRAGMA(omp for schedule(static))          for ( uint64_t i=0;i<num;i++) { __VA_ARGS__ } ;
+#define thread_for_collapse2( i, num, ... )                 DO_PRAGMA(omp parallel for collapse(2))      for ( uint64_t i=0;i<num;i++) { __VA_ARGS__ } ;
+#define thread_for_collapse( N , i, num, ... )              DO_PRAGMA(omp parallel for collapse ( N ) )  for ( uint64_t i=0;i<num;i++) { __VA_ARGS__ } ;
+#define thread_for_collapse_in_region( N , i, num, ... )    DO_PRAGMA(omp for collapse ( N ))            for ( uint64_t i=0;i<num;i++) { __VA_ARGS__ } ;
+#define thread_region                                       DO_PRAGMA(omp parallel)
+#define thread_critical                                     DO_PRAGMA(omp critical)
 
diff --git a/Grid/util/Coordinate.h b/Grid/util/Coordinate.h
index c319a14d..7f1d31c0 100644
--- a/Grid/util/Coordinate.h
+++ b/Grid/util/Coordinate.h
@@ -52,14 +52,14 @@ public:
   accelerator_inline size_type size(void) const { return _size; };
   accelerator_inline void  clear(void) { resize(0);}
   accelerator_inline void  resize(size_type sz) {
+#ifndef GRID_HIP
     assert(sz>=0);
     assert(sz<=MaxEntries);
+#endif
     _size = sz;
   }
   accelerator_inline void  resize(size_type sz,const value &val) {
-    assert(sz>=0);
-    assert(sz<=MaxEntries);
-    _size = sz;
+    resize(sz);
     for(int s=0;s<sz;s++) _data[s]=val;
   }
   accelerator_inline pointer begin(void)                   { return &_data[0]; } 
@@ -67,7 +67,7 @@ public:
   accelerator_inline pointer end  (void)                   { return &_data[_size]; } 
   accelerator_inline const_pointer end  (void) const       { return &_data[_size]; } 
   accelerator_inline void push_back(const value &val)      { resize(_size+1); _data[_size-1] = val;}
-  accelerator_inline AcceleratorVector()                   { _size = 0; }
+  accelerator_inline AcceleratorVector()                   { resize(0); }
   accelerator_inline AcceleratorVector(size_type sz)           { resize(sz); }
   accelerator_inline AcceleratorVector(size_type sz,const value &val) { resize(sz,val); }
   AcceleratorVector(const std::vector<value> &copyme) { 
diff --git a/Grid/util/Init.cc b/Grid/util/Init.cc
index 570f4234..656e29a9 100644
--- a/Grid/util/Init.cc
+++ b/Grid/util/Init.cc
@@ -73,8 +73,6 @@ feenableexcept (unsigned int excepts)
 }
 #endif
 
-uint32_t gpu_threads=8;
-
 NAMESPACE_BEGIN(Grid);
 
 //////////////////////////////////////////////////////
@@ -192,16 +190,12 @@ void GridParseLayout(char **argv,int argc,
     assert(ompthreads.size()==1);
     GridThread::SetThreads(ompthreads[0]);
   }
-  if( GridCmdOptionExists(argv,argv+argc,"--gpu-threads") ){
+  if( GridCmdOptionExists(argv,argv+argc,"--accelerator-threads") ){
     std::vector<int> gputhreads(0);
-#ifndef GRID_NVCC
-    std::cout << GridLogWarning << "'--gpu-threads' option used but Grid was"
-              << " not compiled with GPU support" << std::endl;
-#endif
-    arg= GridCmdOptionPayload(argv,argv+argc,"--gpu-threads");
+    arg= GridCmdOptionPayload(argv,argv+argc,"--accelerator-threads");
     GridCmdOptionIntVector(arg,gputhreads);
     assert(gputhreads.size()==1);
-    gpu_threads=gputhreads[0];
+    acceleratorThreads(gputhreads[0]);
   }
 
   if( GridCmdOptionExists(argv,argv+argc,"--cores") ){
@@ -241,8 +235,6 @@ static int Grid_is_initialised;
 /////////////////////////////////////////////////////////
 void GridBanner(void)
 {
-  static int printed =0;
-  if( !printed ) {
     std::cout <<std::endl;
     std::cout  << "__|__|__|__|__|__|__|__|__|__|__|__|__|__|__"<<std::endl; 
     std::cout  << "__|__|__|__|__|__|__|__|__|__|__|__|__|__|__"<<std::endl; 
@@ -278,67 +270,6 @@ void GridBanner(void)
     std::cout << "Build " << GRID_BUILD_STR(GRID_BUILD_REF) << std::endl;
 #endif
     std::cout << std::endl;
-    printed=1;
-  }
-}
-#ifdef GRID_NVCC
-cudaDeviceProp *gpu_props;
-#endif
-void GridGpuInit(void)
-{
-#ifdef GRID_NVCC
-  int nDevices = 1;
-  cudaGetDeviceCount(&nDevices);
-  gpu_props = new cudaDeviceProp[nDevices];
-
-  char * localRankStr = NULL;
-  int rank = 0, world_rank=0; 
-#define ENV_LOCAL_RANK_OMPI    "OMPI_COMM_WORLD_LOCAL_RANK"
-#define ENV_LOCAL_RANK_MVAPICH "MV2_COMM_WORLD_LOCAL_RANK"
-#define ENV_RANK_OMPI          "OMPI_COMM_WORLD_RANK"
-#define ENV_RANK_MVAPICH       "MV2_COMM_WORLD_RANK"
-  // We extract the local rank initialization using an environment variable
-  if ((localRankStr = getenv(ENV_LOCAL_RANK_OMPI)) != NULL)
-  {
-    rank = atoi(localRankStr);		
-  }
-  if ((localRankStr = getenv(ENV_LOCAL_RANK_MVAPICH)) != NULL)
-  {
-    rank = atoi(localRankStr);		
-  }
-  if ((localRankStr = getenv(ENV_RANK_OMPI   )) != NULL) { world_rank = atoi(localRankStr);}
-  if ((localRankStr = getenv(ENV_RANK_MVAPICH)) != NULL) { world_rank = atoi(localRankStr);}
-
-  if ( world_rank == 0 ) {
-    GridBanner();
-  }
-
-  for (int i = 0; i < nDevices; i++) {
-
-#define GPU_PROP_FMT(canMapHostMemory,FMT)     printf("GpuInit:   " #canMapHostMemory ": " FMT" \n",prop.canMapHostMemory);
-#define GPU_PROP(canMapHostMemory)             GPU_PROP_FMT(canMapHostMemory,"%d");
-    
-    cudaGetDeviceProperties(&gpu_props[i], i);
-    if ( world_rank == 0) {
-      cudaDeviceProp prop; 
-      prop = gpu_props[i];
-      printf("GpuInit: ========================\n");
-      printf("GpuInit: Device Number    : %d\n", i);
-      printf("GpuInit: ========================\n");
-      printf("GpuInit: Device identifier: %s\n", prop.name);
-
-      GPU_PROP(managedMemory);
-      GPU_PROP(isMultiGpuBoard);
-      GPU_PROP(warpSize);
-      //      GPU_PROP(unifiedAddressing);
-      //      GPU_PROP(l2CacheSize);
-      //      GPU_PROP(singleToDoublePrecisionPerfRatio);
-    }
-  }
-  if ( world_rank == 0 ) {
-    printf("GpuInit: ================================================\n");
-  }
-#endif
 }
 
 void Grid_init(int *argc,char ***argv)
@@ -353,7 +284,7 @@ void Grid_init(int *argc,char ***argv)
   //////////////////////////////////////////////////////////
   // Early intialisation necessities without rank knowledge
   //////////////////////////////////////////////////////////
-  GridGpuInit(); // Must come first to set device prior to MPI init
+  acceleratorInit(); // Must come first to set device prior to MPI init due to Omnipath Driver
 
   if( GridCmdOptionExists(*argv,*argv+*argc,"--shm") ){
     int MB;
@@ -363,6 +294,14 @@ void Grid_init(int *argc,char ***argv)
     GlobalSharedMemory::MAX_MPI_SHM_BYTES = MB64*1024LL*1024LL;
   }
 
+  if( GridCmdOptionExists(*argv,*argv+*argc,"--device-mem") ){
+    int MB;
+    arg= GridCmdOptionPayload(*argv,*argv+*argc,"--device-mem");
+    GridCmdOptionInt(arg,MB);
+    uint64_t MB64 = MB;
+    MemoryManager::DeviceMaxBytes = MB64*1024LL*1024LL;
+  }
+
   if( GridCmdOptionExists(*argv,*argv+*argc,"--hypercube") ){
     int enable;
     arg= GridCmdOptionPayload(*argv,*argv+*argc,"--hypercube");
@@ -379,6 +318,11 @@ void Grid_init(int *argc,char ***argv)
     Grid_debug_handler_init();
   }
 
+  //////////////////////////////////////////////////////////
+  // Memory manager
+  //////////////////////////////////////////////////////////
+  MemoryManager::Init();
+
   //////////////////////////////////////////////////////////
   // MPI initialisation
   //////////////////////////////////////////////////////////
@@ -417,11 +361,18 @@ void Grid_init(int *argc,char ***argv)
   std::cout << GridLogMessage << "MPI is initialised and logging filters activated "<<std::endl;
   std::cout << GridLogMessage << "================================================ "<<std::endl;
 
+
+  /////////////////////////////////////////////////////////
+  // Reporting
+  /////////////////////////////////////////////////////////
   std::cout << GridLogMessage << "Requested "<< GlobalSharedMemory::MAX_MPI_SHM_BYTES <<" byte stencil comms buffers "<<std::endl;
   if ( GlobalSharedMemory::Hugepages) {
     std::cout << GridLogMessage << "Mapped stencil comms buffers as MAP_HUGETLB "<<std::endl;
   }
 
+#ifndef GRID_UVM
+  std::cout << GridLogMessage << "MemoryManager Cache "<< MemoryManager::DeviceMaxBytes <<" bytes "<<std::endl;
+#endif
 
   if( GridCmdOptionExists(*argv,*argv+*argc,"--debug-mem") ){
     MemoryProfiler::debug = true;
diff --git a/Grid/util/Init.h b/Grid/util/Init.h
index f7f032ba..dad963a0 100644
--- a/Grid/util/Init.h
+++ b/Grid/util/Init.h
@@ -56,6 +56,7 @@ std::string GridCmdVectorIntToString(const VectorInt & vec);
 void GridCmdOptionCSL(std::string str,std::vector<std::string> & vec);
 template<class VectorInt>
 void GridCmdOptionIntVector(std::string &str,VectorInt & vec);
+void GridCmdOptionInt(std::string &str,int & val);
 
 
 void GridParseLayout(char **argv,int argc,
diff --git a/benchmarks/Benchmark_IO.cc b/benchmarks/Benchmark_IO.cc
index 3d3b0ce0..c8c0937f 100644
--- a/benchmarks/Benchmark_IO.cc
+++ b/benchmarks/Benchmark_IO.cc
@@ -14,6 +14,7 @@ std::string filestem(const int l)
 
 int main (int argc, char ** argv)
 {
+#ifdef HAVE_LIME
   Grid_init(&argc,&argv);
 
   int64_t threads = GridThread::GetThreads();
@@ -42,6 +43,6 @@ int main (int argc, char ** argv)
   }
 
   Grid_finalize();
-
+#endif
   return EXIT_SUCCESS;
 }
diff --git a/benchmarks/Benchmark_IO.hpp b/benchmarks/Benchmark_IO.hpp
index 91fcb61f..d3416353 100644
--- a/benchmarks/Benchmark_IO.hpp
+++ b/benchmarks/Benchmark_IO.hpp
@@ -2,7 +2,7 @@
 #define Benchmark_IO_hpp_
 
 #include <Grid/Grid.h>
-
+#ifdef HAVE_LIME
 #define MSG std::cout << GridLogMessage
 #define SEP \
 "============================================================================="
@@ -104,4 +104,5 @@ void readBenchmark(const Coordinate &latt, const std::string filename,
 
 }
 
+#endif //LIME
 #endif // Benchmark_IO_hpp_
diff --git a/benchmarks/Benchmark_IO_vs_dir.cc b/benchmarks/Benchmark_IO_vs_dir.cc
index cb4831ed..6e6c9ae0 100644
--- a/benchmarks/Benchmark_IO_vs_dir.cc
+++ b/benchmarks/Benchmark_IO_vs_dir.cc
@@ -8,6 +8,7 @@ using namespace Grid;
 
 int main (int argc, char ** argv)
 {
+#ifdef HAVE_LIME
   std::vector<std::string> dir;
   unsigned int             Ls;
   bool                     rb;
@@ -73,6 +74,6 @@ int main (int argc, char ** argv)
   }
 
   Grid_finalize();
-
+#endif
   return EXIT_SUCCESS;
 }
diff --git a/benchmarks/Benchmark_ITT.cc b/benchmarks/Benchmark_ITT.cc
index 7ad4a147..dc09549c 100644
--- a/benchmarks/Benchmark_ITT.cc
+++ b/benchmarks/Benchmark_ITT.cc
@@ -30,7 +30,6 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 
 using namespace Grid;
 
-
 std::vector<int> L_list;
 std::vector<int> Ls_list;
 std::vector<double> mflop_list;
@@ -76,7 +75,6 @@ struct controls {
   int Opt;
   int CommsOverlap;
   Grid::CartesianCommunicator::CommunicatorPolicy_t CommsAsynch;
-  //  int HugePages;
 };
 
 class Benchmark {
@@ -119,14 +117,15 @@ public:
     std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
     comms_header();
 
-    for(int lat=4;lat<=maxlat;lat+=4){
-      for(int Ls=8;Ls<=8;Ls*=2){
+    for(int lat=16;lat<=maxlat;lat+=8){
+      //      for(int Ls=8;Ls<=8;Ls*=2){
+      { int Ls=12;
 
 	Coordinate latt_size  ({lat*mpi_layout[0],
 	      lat*mpi_layout[1],
 	      lat*mpi_layout[2],
 	      lat*mpi_layout[3]});
-
+	std::cout << GridLogMessage<< latt_size <<std::endl;
 	GridCartesian     Grid(latt_size,simd_layout,mpi_layout);
 	RealD Nrank = Grid._Nprocessors;
 	RealD Nnode = Grid.NodeCount();
@@ -184,9 +183,6 @@ public:
 	}
 
 	timestat.statistics(t_time);
-	//	for(int i=0;i<t_time.size();i++){
-	//	  std::cout << i<<" "<<t_time[i]<<std::endl;
-	//	}
 
 	dbytes=dbytes*ppn;
 	double xbytes    = dbytes*0.5;
@@ -199,8 +195,6 @@ public:
 		 <<xbytes/timestat.max <<" "<< xbytes/timestat.min  
 		 << "\t\t"<<std::setw(7)<< bidibytes/timestat.mean<< "  " << bidibytes*timestat.err/(timestat.mean*timestat.mean) << " "
 		 << bidibytes/timestat.max << " " << bidibytes/timestat.min << std::endl;
-
- 
 	
 	    }
     }    
@@ -227,14 +221,15 @@ public:
     uint64_t NN;
 
 
-  uint64_t lmax=48;
+  uint64_t lmax=32;
 #define NLOOP (100*lmax*lmax*lmax*lmax/lat/lat/lat/lat)
 
     GridSerialRNG          sRNG;      sRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9}));
-    for(int lat=8;lat<=lmax;lat+=4){
+    for(int lat=8;lat<=lmax;lat+=8){
 
       Coordinate latt_size  ({lat*mpi_layout[0],lat*mpi_layout[1],lat*mpi_layout[2],lat*mpi_layout[3]});
       int64_t vol= latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
+
       GridCartesian     Grid(latt_size,simd_layout,mpi_layout);
 
       //      NP= Grid.RankCount();
@@ -242,9 +237,9 @@ public:
 
       Vec rn ; random(sRNG,rn);
 
-      LatticeVec z(&Grid); z=rn;
-      LatticeVec x(&Grid); x=rn;
-      LatticeVec y(&Grid); y=rn;
+      LatticeVec z(&Grid); z=Zero();
+      LatticeVec x(&Grid); x=Zero();
+      LatticeVec y(&Grid); y=Zero();
       double a=2.0;
 
       uint64_t Nloop=NLOOP;
@@ -252,9 +247,9 @@ public:
       double start=usecond();
       for(int i=0;i<Nloop;i++){
 	z=a*x-y;
-	auto x_v = x.View();
-	auto y_v = y.View();
-	auto z_v = z.View();
+	autoView( x_v , x, CpuWrite);
+	autoView( y_v , y, CpuWrite);
+	autoView( z_v , z, CpuRead);
         x_v[0]=z_v[0]; // force serial dependency to prevent optimise away
         y_v[4]=z_v[4];
       }
@@ -270,191 +265,8 @@ public:
     }
   };
 
-#if 0
-  static double DWF5(int Ls,int L)
-  {
-    //    RealD mass=0.1;
-    RealD M5  =1.8;
 
-    double mflops;
-    double mflops_best = 0;
-    double mflops_worst= 0;
-    std::vector<double> mflops_all;
-
-    ///////////////////////////////////////////////////////
-    // Set/Get the layout & grid size
-    ///////////////////////////////////////////////////////
-    int threads = GridThread::GetThreads();
-    Coordinate mpi = GridDefaultMpi(); assert(mpi.size()==4);
-    Coordinate local({L,L,L,L});
-
-    GridCartesian         * TmpGrid   = SpaceTimeGrid::makeFourDimGrid(Coordinate({64,64,64,64}), 
-								       GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());
-    uint64_t NP = TmpGrid->RankCount();
-    uint64_t NN = TmpGrid->NodeCount();
-    NN_global=NN;
-    uint64_t SHM=NP/NN;
-
-    Coordinate internal;
-    if      ( SHM == 1 )   internal = Coordinate({1,1,1,1});
-    else if ( SHM == 2 )   internal = Coordinate({2,1,1,1});
-    else if ( SHM == 4 )   internal = Coordinate({2,2,1,1});
-    else if ( SHM == 8 )   internal = Coordinate({2,2,2,1});
-    else assert(0);
-
-    Coordinate nodes({mpi[0]/internal[0],mpi[1]/internal[1],mpi[2]/internal[2],mpi[3]/internal[3]});
-    Coordinate latt4({local[0]*nodes[0],local[1]*nodes[1],local[2]*nodes[2],local[3]*nodes[3]});
-
-    ///////// Welcome message ////////////
-    std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
-    std::cout<<GridLogMessage << "Benchmark DWF Ls vec on "<<L<<"^4 local volume "<<std::endl;
-    std::cout<<GridLogMessage << "* Global volume  : "<<GridCmdVectorIntToString(latt4)<<std::endl;
-    std::cout<<GridLogMessage << "* Ls             : "<<Ls<<std::endl;
-    std::cout<<GridLogMessage << "* MPI ranks      : "<<GridCmdVectorIntToString(mpi)<<std::endl;
-    std::cout<<GridLogMessage << "* Intranode      : "<<GridCmdVectorIntToString(internal)<<std::endl;
-    std::cout<<GridLogMessage << "* nodes          : "<<GridCmdVectorIntToString(nodes)<<std::endl;
-    std::cout<<GridLogMessage << "* Using "<<threads<<" threads"<<std::endl;
-    std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
-
-    ///////// Lattice Init ////////////
-    GridCartesian         * UGrid    = SpaceTimeGrid::makeFourDimGrid(latt4, GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());
-    GridRedBlackCartesian * UrbGrid  = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
-    GridCartesian         * sUGrid   = SpaceTimeGrid::makeFourDimDWFGrid(latt4,GridDefaultMpi());
-    GridRedBlackCartesian * sUrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(sUGrid);
-    GridCartesian         * sFGrid   = SpaceTimeGrid::makeFiveDimDWFGrid(Ls,UGrid);
-    GridRedBlackCartesian * sFrbGrid = SpaceTimeGrid::makeFiveDimDWFRedBlackGrid(Ls,UGrid);
-
-    ///////// RNG Init ////////////
-    std::vector<int> seeds4({1,2,3,4});
-    std::vector<int> seeds5({5,6,7,8});
-    GridParallelRNG          RNG4(UGrid);  RNG4.SeedFixedIntegers(seeds4);
-    GridParallelRNG          RNG5(sFGrid);  RNG5.SeedFixedIntegers(seeds5);
-    std::cout << GridLogMessage << "Initialised RNGs" << std::endl;
-
-    ///////// Source preparation ////////////
-    LatticeFermion src   (sFGrid); 
-    LatticeFermion tmp   (sFGrid);
-    std::cout << GridLogMessage << "allocated src and tmp" << std::endl;
-    random(RNG5,src);
-    std::cout << GridLogMessage << "intialised random source" << std::endl;
-
-    RealD N2 = 1.0/::sqrt(norm2(src));
-    src = src*N2;
-    
-    LatticeGaugeField Umu(UGrid);  SU3::HotConfiguration(RNG4,Umu); 
-
-    WilsonFermion5DR sDw(Umu,*sFGrid,*sFrbGrid,*sUGrid,*sUrbGrid,M5);
-    LatticeFermion src_e (sFrbGrid);
-    LatticeFermion src_o (sFrbGrid);
-    LatticeFermion r_e   (sFrbGrid);
-    LatticeFermion r_o   (sFrbGrid);
-    LatticeFermion r_eo  (sFGrid);
-    LatticeFermion err   (sFGrid);
-    {
-
-      pickCheckerboard(Even,src_e,src);
-      pickCheckerboard(Odd,src_o,src);
-
-#if defined(AVX512) 
-      const int num_cases = 6;
-      std::string fmt("A/S ; A/O ; U/S ; U/O ; G/S ; G/O ");
-#else
-      const int num_cases = 4;
-      std::string fmt("U/S ; U/O ; G/S ; G/O ");
-#endif
-      controls Cases [] = {
-#ifdef AVX512
-	{  WilsonKernelsStatic::OptInlineAsm ,  WilsonKernelsStatic::CommsThenCompute ,CartesianCommunicator::CommunicatorPolicySequential  },
-	{  WilsonKernelsStatic::OptInlineAsm ,  WilsonKernelsStatic::CommsAndCompute  ,CartesianCommunicator::CommunicatorPolicySequential  },
-#endif
-	{  WilsonKernelsStatic::OptHandUnroll,  WilsonKernelsStatic::CommsThenCompute ,CartesianCommunicator::CommunicatorPolicySequential  },
-	{  WilsonKernelsStatic::OptHandUnroll,  WilsonKernelsStatic::CommsAndCompute  ,CartesianCommunicator::CommunicatorPolicySequential  },
-	{  WilsonKernelsStatic::OptGeneric   ,  WilsonKernelsStatic::CommsThenCompute ,CartesianCommunicator::CommunicatorPolicySequential  },
-	{  WilsonKernelsStatic::OptGeneric   ,  WilsonKernelsStatic::CommsAndCompute  ,CartesianCommunicator::CommunicatorPolicySequential  }
-      }; 
-
-      for(int c=0;c<num_cases;c++) {
-
-	 WilsonKernelsStatic::Comms = Cases[c].CommsOverlap;
-	 WilsonKernelsStatic::Opt   = Cases[c].Opt;
-	CartesianCommunicator::SetCommunicatorPolicy(Cases[c].CommsAsynch);
-
-	std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
-	if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptGeneric   ) std::cout << GridLogMessage<< "* Using GENERIC Nc WilsonKernels" <<std::endl;
-	if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptHandUnroll) std::cout << GridLogMessage<< "* Using Nc=3       WilsonKernels" <<std::endl;
-	if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptInlineAsm ) std::cout << GridLogMessage<< "* Using Asm Nc=3   WilsonKernels" <<std::endl;
-	if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsAndCompute ) std::cout << GridLogMessage<< "* Using Overlapped Comms/Compute" <<std::endl;
-	if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsThenCompute) std::cout << GridLogMessage<< "* Using sequential comms compute" <<std::endl;
-	if ( sizeof(Real)==4 )   std::cout << GridLogMessage<< "* SINGLE precision "<<std::endl;
-	if ( sizeof(Real)==8 )   std::cout << GridLogMessage<< "* DOUBLE precision "<<std::endl;
-	std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
-
-	int nwarm = 100;
-	uint64_t ncall = 1000;
-
-	double t0=usecond();
-	sFGrid->Barrier();
-	for(int i=0;i<nwarm;i++){
-	  sDw.DhopEO(src_o,r_e,DaggerNo);
-	}
-	sFGrid->Barrier();
-	double t1=usecond();
-
-	sDw.ZeroCounters();
-	time_statistics timestat;
-	std::vector<double> t_time(ncall);
-	for(uint64_t i=0;i<ncall;i++){
-	  t0=usecond();
-	  sDw.DhopEO(src_o,r_e,DaggerNo);
-	  t1=usecond();
-	  t_time[i] = t1-t0;
-	}
-	sFGrid->Barrier();
-	
-	double volume=Ls;  for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
-	double flops=(1344.0*volume)/2;
-	double mf_hi, mf_lo, mf_err;
-
-	timestat.statistics(t_time);
-	mf_hi = flops/timestat.min;
-	mf_lo = flops/timestat.max;
-	mf_err= flops/timestat.min * timestat.err/timestat.mean;
-
-	mflops = flops/timestat.mean;
-	mflops_all.push_back(mflops);
-	if ( mflops_best == 0   ) mflops_best = mflops;
-	if ( mflops_worst== 0   ) mflops_worst= mflops;
-	if ( mflops>mflops_best ) mflops_best = mflops;
-	if ( mflops<mflops_worst) mflops_worst= mflops;
-
-	std::cout<<GridLogMessage << std::fixed << std::setprecision(1)<<"sDeo mflop/s =   "<< mflops << " ("<<mf_err<<") " << mf_lo<<"-"<<mf_hi <<std::endl;
-	std::cout<<GridLogMessage << std::fixed << std::setprecision(1)<<"sDeo mflop/s per rank   "<< mflops/NP<<std::endl;
-	std::cout<<GridLogMessage << std::fixed << std::setprecision(1)<<"sDeo mflop/s per node   "<< mflops/NN<<std::endl;
-
-	sDw.Report();
-
-      }
-      double robust = mflops_worst/mflops_best;;
-      std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
-      std::cout<<GridLogMessage << L<<"^4 x "<<Ls<< " sDeo Best  mflop/s        =   "<< mflops_best << " ; " << mflops_best/NN<<" per node " <<std::endl;
-      std::cout<<GridLogMessage << L<<"^4 x "<<Ls<< " sDeo Worst mflop/s        =   "<< mflops_worst<< " ; " << mflops_worst/NN<<" per node " <<std::endl;
-
-      std::cout<<GridLogMessage <<std::setprecision(3)<< L<<"^4 x "<<Ls<< " Performance Robustness   =   "<< robust <<std::endl;
-      std::cout<<GridLogMessage <<fmt << std::endl;
-      std::cout<<GridLogMessage;
-
-      for(int i=0;i<mflops_all.size();i++){
-	std::cout<<mflops_all[i]/NN<<" ; " ;
-      }
-      std::cout<<std::endl;
-      std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
-
-    }
-    return mflops_best;
-  }
-#endif
-
-  static double DWF(int Ls,int L, double & robust)
+  static double DWF(int Ls,int L)
   {
     RealD mass=0.1;
     RealD M5  =1.8;
@@ -471,37 +283,30 @@ public:
     Coordinate mpi = GridDefaultMpi(); assert(mpi.size()==4);
     Coordinate local({L,L,L,L});
 
-    GridCartesian         * TmpGrid   = SpaceTimeGrid::makeFourDimGrid(Coordinate({64,64,64,64}), 
-								       GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());
+    GridCartesian         * TmpGrid   = SpaceTimeGrid::makeFourDimGrid(Coordinate({72,72,72,72}), 
+								       GridDefaultSimd(Nd,vComplex::Nsimd()),
+								       GridDefaultMpi());
     uint64_t NP = TmpGrid->RankCount();
     uint64_t NN = TmpGrid->NodeCount();
     NN_global=NN;
     uint64_t SHM=NP/NN;
 
-    Coordinate internal;
-    if      ( SHM == 1 )   internal = Coordinate({1,1,1,1});
-    else if ( SHM == 2 )   internal = Coordinate({2,1,1,1});
-    else if ( SHM == 4 )   internal = Coordinate({2,2,1,1});
-    else if ( SHM == 8 )   internal = Coordinate({2,2,2,1});
-    else assert(0);
-
-    Coordinate nodes({mpi[0]/internal[0],mpi[1]/internal[1],mpi[2]/internal[2],mpi[3]/internal[3]});
-    Coordinate latt4({local[0]*nodes[0],local[1]*nodes[1],local[2]*nodes[2],local[3]*nodes[3]});
+    Coordinate latt4({local[0]*mpi[0],local[1]*mpi[1],local[2]*mpi[2],local[3]*mpi[3]});
 
     ///////// Welcome message ////////////
     std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
     std::cout<<GridLogMessage << "Benchmark DWF on "<<L<<"^4 local volume "<<std::endl;
     std::cout<<GridLogMessage << "* Global volume  : "<<GridCmdVectorIntToString(latt4)<<std::endl;
     std::cout<<GridLogMessage << "* Ls             : "<<Ls<<std::endl;
-    std::cout<<GridLogMessage << "* MPI ranks      : "<<GridCmdVectorIntToString(mpi)<<std::endl;
-    std::cout<<GridLogMessage << "* Intranode      : "<<GridCmdVectorIntToString(internal)<<std::endl;
-    std::cout<<GridLogMessage << "* nodes          : "<<GridCmdVectorIntToString(nodes)<<std::endl;
+    std::cout<<GridLogMessage << "* ranks          : "<<NP  <<std::endl;
+    std::cout<<GridLogMessage << "* nodes          : "<<NN  <<std::endl;
+    std::cout<<GridLogMessage << "* ranks/node     : "<<SHM <<std::endl;
+    std::cout<<GridLogMessage << "* ranks geom     : "<<GridCmdVectorIntToString(mpi)<<std::endl;
     std::cout<<GridLogMessage << "* Using "<<threads<<" threads"<<std::endl;
     std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
 
-
     ///////// Lattice Init ////////////
-    GridCartesian         * UGrid   = SpaceTimeGrid::makeFourDimGrid(latt4, GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());
+    GridCartesian         * UGrid   = SpaceTimeGrid::makeFourDimGrid(latt4, GridDefaultSimd(Nd,vComplexF::Nsimd()),GridDefaultMpi());
     GridRedBlackCartesian * UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
     GridCartesian         * FGrid   = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
     GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid);
@@ -514,74 +319,31 @@ public:
     GridParallelRNG          RNG5(FGrid);  RNG5.SeedFixedIntegers(seeds5);
     std::cout << GridLogMessage << "Initialised RNGs" << std::endl;
 
+    typedef DomainWallFermionF Action;
+    typedef typename Action::FermionField Fermion;
+    typedef LatticeGaugeFieldF Gauge;
+    
     ///////// Source preparation ////////////
-    LatticeFermion src   (FGrid); random(RNG5,src);
-    LatticeFermion ref   (FGrid);
-    LatticeFermion tmp   (FGrid);
+    Gauge Umu(UGrid);  SU3::HotConfiguration(RNG4,Umu); 
+    Fermion src   (FGrid); random(RNG5,src);
+    Fermion src_e (FrbGrid);
+    Fermion src_o (FrbGrid);
+    Fermion r_e   (FrbGrid);
+    Fermion r_o   (FrbGrid);
+    Fermion r_eo  (FGrid);
+    Action Dw(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
 
-    RealD N2 = 1.0/::sqrt(norm2(src));
-    std::cout<<GridLogMessage << "Normalising src  "<< N2 <<std::endl;
-    src = src*N2;
-    
-    LatticeGaugeField Umu(UGrid);  SU3::HotConfiguration(RNG4,Umu); 
-    
-
-    DomainWallFermionR Dw(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
-
-    ////////////////////////////////////
-    // Naive wilson implementation
-    ////////////////////////////////////
-    {
-      LatticeGaugeField Umu5d(FGrid); 
-      std::vector<LatticeColourMatrix> U(4,FGrid);
-      auto Umu_v = Umu.View();
-      auto Umu5d_v = Umu5d.View();
-      for(int ss=0;ss<Umu.Grid()->oSites();ss++){
-	for(int s=0;s<Ls;s++){
-	  Umu5d_v[Ls*ss+s] = Umu_v[ss];
-	}
-      }
-      ref = Zero();
-      for(int mu=0;mu<Nd;mu++){
-	U[mu] = PeekIndex<LorentzIndex>(Umu5d,mu);
-      }
-      for(int mu=0;mu<Nd;mu++){
-	
-	tmp = U[mu]*Cshift(src,mu+1,1);
-	ref=ref + tmp - Gamma(Gmu[mu])*tmp;
-	
-	tmp =adj(U[mu])*src;
-	tmp =Cshift(tmp,mu+1,-1);
-	ref=ref + tmp + Gamma(Gmu[mu])*tmp;
-      }
-      ref = -0.5*ref;
-    }
-
-    LatticeFermion src_e (FrbGrid);
-    LatticeFermion src_o (FrbGrid);
-    LatticeFermion r_e   (FrbGrid);
-    LatticeFermion r_o   (FrbGrid);
-    LatticeFermion r_eo  (FGrid);
-    LatticeFermion err   (FGrid);
     {
 
       pickCheckerboard(Even,src_e,src);
       pickCheckerboard(Odd,src_o,src);
 
-#if defined(AVX512) 
-      const int num_cases = 6;
-      std::string fmt("A/S ; A/O ; U/S ; U/O ; G/S ; G/O ");
-#else
       const int num_cases = 4;
-      std::string fmt("U/S ; U/O ; G/S ; G/O ");
-#endif
+      std::string fmt("G/S/C ; G/O/C ; G/S/S ; G/O/S ");
+
       controls Cases [] = {
-#ifdef AVX512
-	{  WilsonKernelsStatic::OptInlineAsm ,  WilsonKernelsStatic::CommsThenCompute ,CartesianCommunicator::CommunicatorPolicySequential  },
-	{  WilsonKernelsStatic::OptInlineAsm ,  WilsonKernelsStatic::CommsAndCompute  ,CartesianCommunicator::CommunicatorPolicySequential  },
-#endif
-	{  WilsonKernelsStatic::OptHandUnroll,  WilsonKernelsStatic::CommsThenCompute ,CartesianCommunicator::CommunicatorPolicySequential  },
-	{  WilsonKernelsStatic::OptHandUnroll,  WilsonKernelsStatic::CommsAndCompute  ,CartesianCommunicator::CommunicatorPolicySequential  },
+	{  WilsonKernelsStatic::OptGeneric   ,  WilsonKernelsStatic::CommsThenCompute ,CartesianCommunicator::CommunicatorPolicyConcurrent  },
+	{  WilsonKernelsStatic::OptGeneric   ,  WilsonKernelsStatic::CommsAndCompute  ,CartesianCommunicator::CommunicatorPolicyConcurrent  },
 	{  WilsonKernelsStatic::OptGeneric   ,  WilsonKernelsStatic::CommsThenCompute ,CartesianCommunicator::CommunicatorPolicySequential  },
 	{  WilsonKernelsStatic::OptGeneric   ,  WilsonKernelsStatic::CommsAndCompute  ,CartesianCommunicator::CommunicatorPolicySequential  }
       }; 
@@ -594,15 +356,12 @@ public:
 
 	std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
 	if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptGeneric   ) std::cout << GridLogMessage<< "* Using GENERIC Nc WilsonKernels" <<std::endl;
-	if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptHandUnroll) std::cout << GridLogMessage<< "* Using Nc=3       WilsonKernels" <<std::endl;
-	if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptInlineAsm ) std::cout << GridLogMessage<< "* Using Asm Nc=3   WilsonKernels" <<std::endl;
 	if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsAndCompute ) std::cout << GridLogMessage<< "* Using Overlapped Comms/Compute" <<std::endl;
-	if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsThenCompute) std::cout << GridLogMessage<< "* Using sequential comms compute" <<std::endl;
-	if ( sizeof(Real)==4 )   std::cout << GridLogMessage<< "* SINGLE precision "<<std::endl;
-	if ( sizeof(Real)==8 )   std::cout << GridLogMessage<< "* DOUBLE precision "<<std::endl;
+	if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsThenCompute) std::cout << GridLogMessage<< "* Using sequential Comms/Compute" <<std::endl;
+	std::cout << GridLogMessage<< "* SINGLE precision "<<std::endl;
 	std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
 
-	int nwarm = 200;
+	int nwarm = 10;
 	double t0=usecond();
 	FGrid->Barrier();
 	for(int i=0;i<nwarm;i++){
@@ -610,9 +369,7 @@ public:
 	}
 	FGrid->Barrier();
 	double t1=usecond();
-	//	uint64_t ncall = (uint64_t) 2.5*1000.0*1000.0*nwarm/(t1-t0);
-	//	if (ncall < 500) ncall = 500;
-	uint64_t ncall = 1000;
+	uint64_t ncall = 50;
 
 	FGrid->Broadcast(0,&ncall,sizeof(ncall));
 
@@ -649,24 +406,11 @@ public:
 	std::cout<<GridLogMessage << std::fixed << std::setprecision(1)<<"Deo mflop/s per rank   "<< mflops/NP<<std::endl;
 	std::cout<<GridLogMessage << std::fixed << std::setprecision(1)<<"Deo mflop/s per node   "<< mflops/NN<<std::endl;
 
-	Dw.Report();
-
-	Dw.DhopEO(src_o,r_e,DaggerNo);
-	Dw.DhopOE(src_e,r_o,DaggerNo);
-	setCheckerboard(r_eo,r_o);
-	setCheckerboard(r_eo,r_e);
-	err = r_eo-ref; 
-	RealD absref = norm2(ref);
-	RealD abserr = norm2(err);
-	std::cout<<GridLogMessage << "norm diff   "<< abserr << " / " << absref<<std::endl;
-	assert(abserr<1.0e-4);
-
       }
-      robust = mflops_worst/mflops_best;
+
       std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
       std::cout<<GridLogMessage << L<<"^4 x "<<Ls<< " Deo Best  mflop/s        =   "<< mflops_best << " ; " << mflops_best/NN<<" per node " <<std::endl;
       std::cout<<GridLogMessage << L<<"^4 x "<<Ls<< " Deo Worst mflop/s        =   "<< mflops_worst<< " ; " << mflops_worst/NN<<" per node " <<std::endl;
-      std::cout<<GridLogMessage << std::fixed<<std::setprecision(3)<< L<<"^4 x "<<Ls<< " Performance Robustness   =   "<< robust  <<std::endl;
       std::cout<<GridLogMessage <<fmt << std::endl;
       std::cout<<GridLogMessage ;
 
@@ -680,8 +424,166 @@ public:
     return mflops_best;
   }
 
+
+  static double Staggered(int L)
+  {
+    double mflops;
+    double mflops_best = 0;
+    double mflops_worst= 0;
+    std::vector<double> mflops_all;
+
+    ///////////////////////////////////////////////////////
+    // Set/Get the layout & grid size
+    ///////////////////////////////////////////////////////
+    int threads = GridThread::GetThreads();
+    Coordinate mpi = GridDefaultMpi(); assert(mpi.size()==4);
+    Coordinate local({L,L,L,L});
+    
+    GridCartesian         * TmpGrid   = SpaceTimeGrid::makeFourDimGrid(Coordinate({72,72,72,72}), 
+								       GridDefaultSimd(Nd,vComplex::Nsimd()),
+								       GridDefaultMpi());
+    uint64_t NP = TmpGrid->RankCount();
+    uint64_t NN = TmpGrid->NodeCount();
+    NN_global=NN;
+    uint64_t SHM=NP/NN;
+
+    Coordinate latt4({local[0]*mpi[0],local[1]*mpi[1],local[2]*mpi[2],local[3]*mpi[3]});
+
+    ///////// Welcome message ////////////
+    std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
+    std::cout<<GridLogMessage << "Benchmark ImprovedStaggered on "<<L<<"^4 local volume "<<std::endl;
+    std::cout<<GridLogMessage << "* Global volume  : "<<GridCmdVectorIntToString(latt4)<<std::endl;
+    std::cout<<GridLogMessage << "* ranks          : "<<NP  <<std::endl;
+    std::cout<<GridLogMessage << "* nodes          : "<<NN  <<std::endl;
+    std::cout<<GridLogMessage << "* ranks/node     : "<<SHM <<std::endl;
+    std::cout<<GridLogMessage << "* ranks geom     : "<<GridCmdVectorIntToString(mpi)<<std::endl;
+    std::cout<<GridLogMessage << "* Using "<<threads<<" threads"<<std::endl;
+    std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
+
+    ///////// Lattice Init ////////////
+    GridCartesian         * FGrid   = SpaceTimeGrid::makeFourDimGrid(latt4, GridDefaultSimd(Nd,vComplexF::Nsimd()),GridDefaultMpi());
+    GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(FGrid);
+    
+    ///////// RNG Init ////////////
+    std::vector<int> seeds4({1,2,3,4});
+    GridParallelRNG          RNG4(FGrid);  RNG4.SeedFixedIntegers(seeds4);
+    std::cout << GridLogMessage << "Initialised RNGs" << std::endl;
+
+    RealD mass=0.1;
+    RealD c1=9.0/8.0;
+    RealD c2=-1.0/24.0;
+    RealD u0=1.0;
+
+    typedef ImprovedStaggeredFermionF Action;
+    typedef typename Action::FermionField Fermion; 
+    typedef LatticeGaugeFieldF Gauge;
+    
+    Gauge Umu(FGrid);  SU3::HotConfiguration(RNG4,Umu); 
+
+    typename Action::ImplParams params;
+    Action Ds(Umu,Umu,*FGrid,*FrbGrid,mass,c1,c2,u0,params);
+
+    ///////// Source preparation ////////////
+    Fermion src   (FGrid); random(RNG4,src);
+    Fermion src_e (FrbGrid);
+    Fermion src_o (FrbGrid);
+    Fermion r_e   (FrbGrid);
+    Fermion r_o   (FrbGrid);
+    Fermion r_eo  (FGrid);
+  
+    {
+
+      pickCheckerboard(Even,src_e,src);
+      pickCheckerboard(Odd,src_o,src);
+    
+      const int num_cases = 4;
+      std::string fmt("G/S/C ; G/O/C ; G/S/S ; G/O/S ");
+      
+      controls Cases [] = {
+	{  StaggeredKernelsStatic::OptGeneric   ,  StaggeredKernelsStatic::CommsThenCompute ,CartesianCommunicator::CommunicatorPolicyConcurrent  },
+	{  StaggeredKernelsStatic::OptGeneric   ,  StaggeredKernelsStatic::CommsAndCompute  ,CartesianCommunicator::CommunicatorPolicyConcurrent  },
+	{  StaggeredKernelsStatic::OptGeneric   ,  StaggeredKernelsStatic::CommsThenCompute ,CartesianCommunicator::CommunicatorPolicySequential  },
+	{  StaggeredKernelsStatic::OptGeneric   ,  StaggeredKernelsStatic::CommsAndCompute  ,CartesianCommunicator::CommunicatorPolicySequential  }
+      }; 
+
+      for(int c=0;c<num_cases;c++) {
+	
+	StaggeredKernelsStatic::Comms = Cases[c].CommsOverlap;
+	StaggeredKernelsStatic::Opt   = Cases[c].Opt;
+	CartesianCommunicator::SetCommunicatorPolicy(Cases[c].CommsAsynch);
+      
+	std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
+	if ( StaggeredKernelsStatic::Opt == StaggeredKernelsStatic::OptGeneric   ) std::cout << GridLogMessage<< "* Using GENERIC Nc StaggeredKernels" <<std::endl;
+	if ( StaggeredKernelsStatic::Comms == StaggeredKernelsStatic::CommsAndCompute ) std::cout << GridLogMessage<< "* Using Overlapped Comms/Compute" <<std::endl;
+	if ( StaggeredKernelsStatic::Comms == StaggeredKernelsStatic::CommsThenCompute) std::cout << GridLogMessage<< "* Using sequential Comms/Compute" <<std::endl;
+	std::cout << GridLogMessage<< "* SINGLE precision "<<std::endl;
+	std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
+	
+	int nwarm = 10;
+	double t0=usecond();
+	FGrid->Barrier();
+	for(int i=0;i<nwarm;i++){
+	  Ds.DhopEO(src_o,r_e,DaggerNo);
+	}
+	FGrid->Barrier();
+	double t1=usecond();
+	uint64_t ncall = 500;
+
+	FGrid->Broadcast(0,&ncall,sizeof(ncall));
+
+	//	std::cout << GridLogMessage << " Estimate " << ncall << " calls per second"<<std::endl;
+	Ds.ZeroCounters();
+
+	time_statistics timestat;
+	std::vector<double> t_time(ncall);
+	for(uint64_t i=0;i<ncall;i++){
+	  t0=usecond();
+	  Ds.DhopEO(src_o,r_e,DaggerNo);
+	  t1=usecond();
+	  t_time[i] = t1-t0;
+	}
+	FGrid->Barrier();
+	
+	double volume=1;  for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
+	double flops=(1146.0*volume)/2;
+	double mf_hi, mf_lo, mf_err;
+	
+	timestat.statistics(t_time);
+	mf_hi = flops/timestat.min;
+	mf_lo = flops/timestat.max;
+	mf_err= flops/timestat.min * timestat.err/timestat.mean;
+
+	mflops = flops/timestat.mean;
+	mflops_all.push_back(mflops);
+	if ( mflops_best == 0   ) mflops_best = mflops;
+	if ( mflops_worst== 0   ) mflops_worst= mflops;
+	if ( mflops>mflops_best ) mflops_best = mflops;
+	if ( mflops<mflops_worst) mflops_worst= mflops;
+	
+	std::cout<<GridLogMessage << std::fixed << std::setprecision(1)<<"Deo mflop/s =   "<< mflops << " ("<<mf_err<<") " << mf_lo<<"-"<<mf_hi <<std::endl;
+	std::cout<<GridLogMessage << std::fixed << std::setprecision(1)<<"Deo mflop/s per rank   "<< mflops/NP<<std::endl;
+	std::cout<<GridLogMessage << std::fixed << std::setprecision(1)<<"Deo mflop/s per node   "<< mflops/NN<<std::endl;
+      
+      }
+
+      std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
+      std::cout<<GridLogMessage << L<<"^4  Deo Best  mflop/s        =   "<< mflops_best << " ; " << mflops_best/NN<<" per node " <<std::endl;
+      std::cout<<GridLogMessage << L<<"^4  Deo Worst mflop/s        =   "<< mflops_worst<< " ; " << mflops_worst/NN<<" per node " <<std::endl;
+      std::cout<<GridLogMessage <<fmt << std::endl;
+      std::cout<<GridLogMessage ;
+
+      for(int i=0;i<mflops_all.size();i++){
+	std::cout<<mflops_all[i]/NN<<" ; " ;
+      }
+      std::cout<<std::endl;
+    }
+    std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
+    return mflops_best;
+  }
 };
 
+
+
 int main (int argc, char ** argv)
 {
   Grid_init(&argc,&argv);
@@ -696,62 +598,50 @@ int main (int argc, char ** argv)
 
   int do_memory=1;
   int do_comms =1;
-  int do_su3   =0;
-  int do_wilson=1;
-  int do_dwf   =1;
 
-  if ( do_su3 ) {
-    // empty for now
-  }
-#if 1
   int sel=2;
-  Coordinate L_list({8,12,16,24});
-#else
-  int sel=1;
-  Coordinate L_list({8,12});
-#endif
+  std::vector<int> L_list({16,24,32});
   int selm1=sel-1;
-  std::vector<double> robust_list;
 
   std::vector<double> wilson;
   std::vector<double> dwf4;
-  std::vector<double> dwf5;
+  std::vector<double> staggered;
 
-  if ( do_wilson ) {
-    int Ls=1;
-    std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
-    std::cout<<GridLogMessage << " Wilson dslash 4D vectorised" <<std::endl;
-    std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
-    for(int l=0;l<L_list.size();l++){
-      double robust;
-      wilson.push_back(Benchmark::DWF(Ls,L_list[l],robust));
-    }
+  int Ls=1;
+  std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
+  std::cout<<GridLogMessage << " Wilson dslash 4D vectorised" <<std::endl;
+  std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
+  for(int l=0;l<L_list.size();l++){
+    wilson.push_back(Benchmark::DWF(Ls,L_list[l]));
   }
 
-  int Ls=16;
-  if ( do_dwf ) {
-    std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
-    std::cout<<GridLogMessage << " Domain wall dslash 4D vectorised" <<std::endl;
-    std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
-    for(int l=0;l<L_list.size();l++){
-      double robust;
-      double result = Benchmark::DWF(Ls,L_list[l],robust) ;
-      dwf4.push_back(result);
-      robust_list.push_back(robust);
-    }
+  Ls=12;
+  std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
+  std::cout<<GridLogMessage << " Domain wall dslash 4D vectorised" <<std::endl;
+  std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
+  for(int l=0;l<L_list.size();l++){
+    double result = Benchmark::DWF(Ls,L_list[l]) ;
+    dwf4.push_back(result);
   }
 
-  if ( do_dwf ) {
+  /*
+  std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
+  std::cout<<GridLogMessage << " Improved Staggered dslash 4D vectorised" <<std::endl;
+  std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
+  for(int l=0;l<L_list.size();l++){
+    double result = Benchmark::Staggered(L_list[l]) ;
+    staggered.push_back(result);
+  }
+  */
 
   std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
   std::cout<<GridLogMessage << " Summary table Ls="<<Ls <<std::endl;
   std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
-  std::cout<<GridLogMessage << "L \t\t Wilson \t DWF4 " <<std::endl;
+  std::cout<<GridLogMessage << "L \t\t Wilson \t\t DWF4 \t\tt Staggered" <<std::endl;
   for(int l=0;l<L_list.size();l++){
-    std::cout<<GridLogMessage << L_list[l] <<" \t\t "<< wilson[l]<<" \t "<<dwf4[l] <<std::endl;
+    std::cout<<GridLogMessage << L_list[l] <<" \t\t "<< wilson[l]<<" \t\t "<<dwf4[l] <<std::endl;
   }
   std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
-  }
 
   int NN=NN_global;
   if ( do_memory ) {
@@ -768,24 +658,20 @@ int main (int argc, char ** argv)
     Benchmark::Comms();
   }
 
-  if ( do_dwf ) {
-  std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
-  std::cout<<GridLogMessage << " Per Node Summary table Ls="<<Ls <<std::endl;
-  std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
-  std::cout<<GridLogMessage << " L \t\t Wilson\t\t DWF4  " <<std::endl;
-  for(int l=0;l<L_list.size();l++){
-    std::cout<<GridLogMessage << L_list[l] <<" \t\t "<< wilson[l]/NN<<" \t "<<dwf4[l]/NN<<std::endl;
-  }
-  std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
+    std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
+    std::cout<<GridLogMessage << " Per Node Summary table Ls="<<Ls <<std::endl;
+    std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
+    std::cout<<GridLogMessage << " L \t\t Wilson\t\t DWF4  " <<std::endl;
+    for(int l=0;l<L_list.size();l++){
+      std::cout<<GridLogMessage << L_list[l] <<" \t\t "<< wilson[l]/NN<<" \t "<<dwf4[l]/NN<<std::endl;
+    }
+    std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
 
-  std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
-  std::cout<<GridLogMessage << " Comparison point     result: "  << 0.5*(dwf4[sel]+dwf4[selm1])/NN << " Mflop/s per node"<<std::endl;
-  std::cout<<GridLogMessage << " Comparison point is 0.5*("<<dwf4[sel]/NN<<"+"<<dwf4[selm1]/NN << ") "<<std::endl;
-  std::cout<<std::setprecision(3);
-  std::cout<<GridLogMessage << " Comparison point robustness: "  << robust_list[sel] <<std::endl;
-  std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
-
-  }
+    std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
+    std::cout<<GridLogMessage << " Comparison point     result: "  << 0.5*(dwf4[sel]+dwf4[selm1])/NN << " Mflop/s per node"<<std::endl;
+    std::cout<<GridLogMessage << " Comparison point is 0.5*("<<dwf4[sel]/NN<<"+"<<dwf4[selm1]/NN << ") "<<std::endl;
+    std::cout<<std::setprecision(3);
+    std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
 
   Grid_finalize();
 }
diff --git a/benchmarks/Benchmark_dwf.cc b/benchmarks/Benchmark_dwf.cc
index b76e58c8..2ef5921d 100644
--- a/benchmarks/Benchmark_dwf.cc
+++ b/benchmarks/Benchmark_dwf.cc
@@ -21,7 +21,7 @@
     *************************************************************************************/
     /*  END LEGAL */
 #include <Grid/Grid.h>
-#ifdef GRID_NVCC
+#ifdef GRID_CUDA
 #define CUDA_PROFILE
 #endif
 
@@ -129,8 +129,8 @@ int main (int argc, char ** argv)
   LatticeGaugeField Umu5d(FGrid);
   std::vector<LatticeColourMatrix> U(4,FGrid);
   {
-    auto Umu5d_v = Umu5d.View();
-    auto Umu_v = Umu.View();
+    autoView( Umu5d_v, Umu5d, CpuWrite);
+    autoView( Umu_v  , Umu  , CpuRead);
     for(int ss=0;ss<Umu.Grid()->oSites();ss++){
       for(int s=0;s<Ls;s++){
 	Umu5d_v[Ls*ss+s] = Umu_v[ss];
@@ -272,8 +272,8 @@ int main (int argc, char ** argv)
       //    ref =  src - Gamma(Gamma::Algebra::GammaX)* src ; // 1+gamma_x
       tmp = U[mu]*Cshift(src,mu+1,1);
       {
-	auto ref_v = ref.View();
-	auto tmp_v = tmp.View();
+	autoView( ref_v, ref, CpuWrite);
+	autoView( tmp_v, tmp, CpuRead);
 	for(int i=0;i<ref_v.size();i++){
 	  ref_v[i]+= tmp_v[i] + Gamma(Gmu[mu])*tmp_v[i]; ;
 	}
@@ -282,8 +282,8 @@ int main (int argc, char ** argv)
       tmp =adj(U[mu])*src;
       tmp =Cshift(tmp,mu+1,-1);
       {
-	auto ref_v = ref.View();
-	auto tmp_v = tmp.View();
+	autoView( ref_v, ref, CpuWrite);
+	autoView( tmp_v, tmp, CpuRead);
 	for(int i=0;i<ref_v.size();i++){
 	  ref_v[i]+= tmp_v[i] - Gamma(Gmu[mu])*tmp_v[i]; ;
 	}
diff --git a/benchmarks/Benchmark_dwf_sweep.cc b/benchmarks/Benchmark_dwf_sweep.cc
index 3f108b16..2f13b667 100644
--- a/benchmarks/Benchmark_dwf_sweep.cc
+++ b/benchmarks/Benchmark_dwf_sweep.cc
@@ -130,11 +130,13 @@ void benchDw(std::vector<int> & latt4, int Ls, int threads,int report )
   LatticeGaugeField Umu5d(FGrid); 
 
   // replicate across fifth dimension
-  auto Umu5d_v = Umu5d.View();
-  auto Umu_v   = Umu.View();
-  for(int ss=0;ss<Umu.Grid()->oSites();ss++){
-    for(int s=0;s<Ls;s++){
-      Umu5d_v[Ls*ss+s] = Umu_v[ss];
+  {
+    autoView(Umu5d_v , Umu5d, CpuWrite);
+    autoView(  Umu_v , Umu,   CpuRead);
+    for(int ss=0;ss<Umu.Grid()->oSites();ss++){
+      for(int s=0;s<Ls;s++){
+	Umu5d_v[Ls*ss+s] = Umu_v[ss];
+      }
     }
   }
 
diff --git a/benchmarks/Benchmark_memory_asynch.cc b/benchmarks/Benchmark_memory_asynch.cc
index a29d281b..97825144 100644
--- a/benchmarks/Benchmark_memory_asynch.cc
+++ b/benchmarks/Benchmark_memory_asynch.cc
@@ -79,7 +79,7 @@ int main (int argc, char ** argv)
 
     double start=usecond();
     thread_for(t,threads,{
-      auto x_t = x[t].View();
+      autoView( x_t , x[t],CpuRead);
       sum[t] = x_t[0];
       for(int i=0;i<Nloop;i++){
 	for(auto ss=x_t.begin();ss<x_t.end();ss++){
diff --git a/benchmarks/Benchmark_memory_bandwidth.cc b/benchmarks/Benchmark_memory_bandwidth.cc
index 9b921a77..10fd48a1 100644
--- a/benchmarks/Benchmark_memory_bandwidth.cc
+++ b/benchmarks/Benchmark_memory_bandwidth.cc
@@ -177,9 +177,7 @@ int main (int argc, char ** argv)
       Real nn;      
       double start=usecond();
       for(int i=0;i<Nloop;i++){
-	auto x_v = x.View();
 	nn=norm2(x);
-	vsplat(x_v[0]._internal[0],nn);
       }
       double stop=usecond();
       double time = (stop-start)/Nloop*1000;
diff --git a/benchmarks/Benchmark_meson_field.cc b/benchmarks/Benchmark_meson_field.cc
index d37acb7f..cca33716 100644
--- a/benchmarks/Benchmark_meson_field.cc
+++ b/benchmarks/Benchmark_meson_field.cc
@@ -85,11 +85,11 @@ void sliceInnerProductMesonField(std::vector< std::vector<ComplexD> > &mat,
       for(int b=0;b<e2;b++){
 	int ss= so+n*stride+b;
 	for(int i=0;i<Lblock;i++){
-	  auto lhs_v = lhs[i].View();
+	  autoView(lhs_v, lhs[i], CpuRead);
 	  auto left = conjugate(lhs_v[ss]);
 	  for(int j=0;j<Rblock;j++){
 	    int idx = i+Lblock*j+Lblock*Rblock*r;
-	    auto rhs_v = rhs[j].View();
+	    autoView(rhs_v, rhs[j], CpuRead);
 	    auto right = rhs_v[ss];
 	    vector_type vv = left()(0)(0) * right()(0)(0)
 	      +              left()(0)(1) * right()(0)(1)
@@ -221,12 +221,12 @@ void sliceInnerProductMesonFieldGamma(std::vector< std::vector<ComplexD> > &mat,
       for(int b=0;b<e2;b++){
 	int ss= so+n*stride+b;
 	for(int i=0;i<Lblock;i++){
- 	  auto lhs_v=lhs[i].View();
+ 	  autoView(lhs_v,lhs[i],CpuRead);
 	  auto left = conjugate(lhs_v[ss]);
 	  for(int j=0;j<Rblock;j++){
 	  for(int mu=0;mu<Ngamma;mu++){
 	    
-	    auto rhs_v = rhs[j].View();
+	    autoView(rhs_v,rhs[j],CpuRead);
 	    auto right = Gamma(gammas[mu])*rhs_v[ss];
 
 	      vector_type vv = left()(0)(0) * right()(0)(0)
@@ -370,12 +370,12 @@ void sliceInnerProductMesonFieldGamma1(std::vector< std::vector<ComplexD> > &mat
 	int ss= so+n*stride+b;
 	for(int i=0;i<Lblock;i++){
 	  
-	  auto lhs_v=lhs[i].View();
+	  autoView(lhs_v,lhs[i],CpuRead);
 	  auto left = conjugate(lhs_v[ss]);
 	  for(int j=0;j<Rblock;j++){
 
 	    SpinMatrix_v vv;
-	    auto rhs_v = rhs[j].View();
+	    autoView(rhs_v,rhs[j],CpuRead);
 	    auto right = rhs_v[ss];
 	    for(int s1=0;s1<Ns;s1++){
 	    for(int s2=0;s2<Ns;s2++){
@@ -518,12 +518,12 @@ void sliceInnerProductMesonFieldGammaMom(std::vector< std::vector<ComplexD> > &m
 
 	for(int i=0;i<Lblock;i++){
 
-	  auto lhs_v = lhs[i].View();
+	  autoView(lhs_v,lhs[i],CpuRead);
 	  auto left = conjugate(lhs_v[ss]);
 	  for(int j=0;j<Rblock;j++){
 
 	    SpinMatrix_v vv;
-	    auto rhs_v = rhs[j].View();
+	    autoView(rhs_v,rhs[j],CpuRead);
 	    auto right = rhs_v[ss];
 	    for(int s1=0;s1<Ns;s1++){
 	    for(int s2=0;s2<Ns;s2++){
@@ -537,7 +537,7 @@ void sliceInnerProductMesonFieldGammaMom(std::vector< std::vector<ComplexD> > &m
 	    // Trigger unroll
 	    for ( int m=0;m<Nmom;m++){
 	      int idx = m+base;
-	      auto mom_v = mom[m].View();
+	      autoView(mom_v,mom[m],CpuRead);
 	      auto phase = mom_v[ss];
 	      mac(&lvSum[idx],&vv,&phase);
 	    }
diff --git a/benchmarks/Benchmark_schur.cc b/benchmarks/Benchmark_schur.cc
new file mode 100644
index 00000000..8171998a
--- /dev/null
+++ b/benchmarks/Benchmark_schur.cc
@@ -0,0 +1,176 @@
+    /*************************************************************************************
+
+    Grid physics library, www.github.com/paboyle/Grid 
+
+    Source file: ./benchmarks/Benchmark_dwf.cc
+
+    Copyright (C) 2015
+
+Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+Author: paboyle <paboyle@ph.ed.ac.uk>
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License along
+    with this program; if not, write to the Free Software Foundation, Inc.,
+    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+    See the full license in the file "LICENSE" in the top level distribution directory
+    *************************************************************************************/
+    /*  END LEGAL */
+#include <Grid/Grid.h>
+
+using namespace std;
+using namespace Grid;
+
+  Gamma::Algebra Gmu [] = {
+    Gamma::Algebra::GammaX,
+    Gamma::Algebra::GammaY,
+    Gamma::Algebra::GammaZ,
+    Gamma::Algebra::GammaT
+  };
+
+void benchDw(std::vector<int> & L, int Ls);
+
+int main (int argc, char ** argv)
+{
+  Grid_init(&argc,&argv);
+
+
+  const int Ls=12;
+  std::vector< std::vector<int> > latts;
+#if 1
+  latts.push_back(std::vector<int> ({24,24,24,24}) );
+  latts.push_back(std::vector<int> ({48,24,24,24}) );
+  latts.push_back(std::vector<int> ({96,24,24,24}) );
+  latts.push_back(std::vector<int> ({96,48,24,24}) );
+  //  latts.push_back(std::vector<int> ({96,48,48,24}) );
+  //  latts.push_back(std::vector<int> ({96,48,48,48}) );
+#else
+  //  latts.push_back(std::vector<int> ({96,48,48,48}) );
+  latts.push_back(std::vector<int> ({96,96,96,192}) );
+#endif
+
+  std::cout << GridLogMessage<< "*****************************************************************" <<std::endl;
+  std::cout << GridLogMessage<< "* Kernel options --dslash-generic, --dslash-unroll, --dslash-asm" <<std::endl;
+  std::cout << GridLogMessage<< "*****************************************************************" <<std::endl;
+
+  if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptGeneric   ) std::cout << GridLogMessage<< "* Using GENERIC Nc WilsonKernels" <<std::endl;
+  if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptHandUnroll) std::cout << GridLogMessage<< "* Using Nc=3       WilsonKernels" <<std::endl;
+  if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptInlineAsm ) std::cout << GridLogMessage<< "* Using Asm Nc=3   WilsonKernels" <<std::endl;
+  std::cout << GridLogMessage<< "*****************************************************************" <<std::endl;
+
+  int threads = GridThread::GetThreads();
+  std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
+  std::cout<<GridLogMessage << "=========================================================================="<<std::endl;
+  std::cout<<GridLogMessage << "= Benchmarking DWF"<<std::endl;
+  std::cout<<GridLogMessage << "=========================================================================="<<std::endl;
+  std::cout<<GridLogMessage << "Volume \t\t\tProcs \t SchurDiagOne "<<std::endl;
+  std::cout<<GridLogMessage << "=========================================================================="<<std::endl;
+
+  for (int l=0;l<latts.size();l++){
+    std::vector<int> latt4 = latts[l];
+    std::cout << GridLogMessage <<"\t";
+    for(int d=0;d<Nd;d++){
+      std::cout<<latt4[d]<<"x";
+    }
+    std::cout <<Ls<<"\t" ;
+    benchDw (latt4,Ls);
+  }
+  std::cout<<GridLogMessage << "=========================================================================="<<std::endl;
+  Grid_finalize();
+}
+
+
+void benchDw(std::vector<int> & latt4, int Ls)
+{
+  /////////////////////////////////////////////////////////////////////////////////////
+  // for Nc=3
+  /////////////////////////////////////////////////////////////////////////////////////
+  // Dw :  Ls*24*(7+48)= Ls*1320 
+  //
+  // M5D:  Ls*(4*2*Nc mul + 4*2*Nc madd ) = 3*4*2*Nc*Ls = Ls*72
+  // Meo:  Ls*24*(7+48) + Ls*72 = Ls*1392 
+  //
+  // Mee:  3*Ns*2*Nc*Ls  // Chroma 6*N5*Nc*Ns 
+  //
+  // LeemInv : 2*2*Nc*madd*Ls
+  // LeeInv  : 2*2*Nc*madd*Ls
+  // DeeInv  : 4*2*Nc*mul *Ls
+  // UeeInv  : 2*2*Nc*madd*Ls
+  // UeemInv : 2*2*Nc*madd*Ls = Nc*Ls*(8+8+8+8+8) = 40*Nc*Ls// Chroma (10*N5 - 8)*Nc*Ns ~ (40 N5 - 32)Nc flops
+  // QUDA counts as dense LsxLs real matrix x Ls x NcNsNreim => Nc*4*2 x Ls^2 FMA = 16Nc Ls^2 flops
+  // Mpc => 1452*cbvol*2*Ls flops // 
+  //     => (1344+Ls*48)*Ls*cbvol*2 flops QUDA = 1920 @Ls=12 and 2112 @Ls=16
+  /////////////////////////////////////////////////////////////////////////////////////
+  GridCartesian         * UGrid   = SpaceTimeGrid::makeFourDimGrid(latt4, GridDefaultSimd(Nd,vComplexF::Nsimd()),GridDefaultMpi());
+  GridRedBlackCartesian * UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
+  GridCartesian         * FGrid   = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
+  GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid);
+  //  long unsigned int single_site_flops     = 8*Nc*(7+16*Nc)*Ls;
+  long unsigned int single_site_mpc_flops = 8*Nc*(7+16*Nc)*2*Ls + 40*Nc*2*Ls + 4*Nc*2*Ls;
+  long unsigned int single_site_quda_flops = 8*Nc*(7+16*Nc)*2*Ls + 16*Nc*Ls*Ls + 4*Nc*2*Ls;
+  std::vector<int> seeds4({1,2,3,4});
+  std::vector<int> seeds5({5,6,7,8});
+
+
+  ColourMatrixF cm = ComplexF(1.0,0.0);
+
+  int ncall=300;
+  RealD mass=0.1;
+  RealD M5  =1.8;
+  RealD NP = UGrid->_Nprocessors;
+  double volume=1;  for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
+
+  LatticeGaugeFieldF Umu(UGrid); Umu=Zero();
+  MobiusFermionF Dw(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,1.5,0.5);
+  
+  LatticeFermionF src_o (FrbGrid); src_o=1.0;
+  LatticeFermionF r_o   (FrbGrid); r_o=Zero();
+
+  int order =151;
+  SchurDiagOneOperator<MobiusFermionF,LatticeFermionF>  Mpc(Dw);
+  Chebyshev<LatticeFermionF>      Cheby(0.0,60.0,order);
+
+  {
+    Mpc.Mpc(src_o,r_o);
+    Mpc.Mpc(src_o,r_o);
+    Mpc.Mpc(src_o,r_o);
+
+    double t0=usecond();
+    for(int i=0;i<ncall;i++){
+      Mpc.Mpc(src_o,r_o);
+    }
+    double t1=usecond();
+
+    double flops=(single_site_mpc_flops*volume*ncall); // Mpc has 1 - Moo^-1 Moe Mee^-1 Meo  so CB cancels.
+    std::cout <<"\t"<<NP<< "\t"<<flops/(t1-t0);
+    flops=(single_site_quda_flops*volume*ncall);
+    std::cout <<"\t"<<flops/(t1-t0)<<"\t"<<(t1-t0)/1000./1000.<<" s\t";
+
+    // Cheby uses MpcDagMpc so 2x flops
+    for(int i=0;i<1;i++){
+    Cheby(Mpc,src_o,r_o);
+    t0=usecond();
+    Cheby(Mpc,src_o,r_o);
+    t1=usecond();
+    flops=(single_site_mpc_flops*volume*2*order);
+    std::cout <<"\t"<<flops/(t1-t0);
+    flops=(single_site_quda_flops*volume*2*order);
+    std::cout <<"\t"<<flops/(t1-t0) << "\t" << (t1-t0)/1000./1000. <<" s";
+    std::cout <<std::endl;
+    }
+  }
+  //  Dw.Report();
+}
+
+
+
diff --git a/benchmarks/Benchmark_staggered.cc b/benchmarks/Benchmark_staggered.cc
index 93086927..34e1e470 100644
--- a/benchmarks/Benchmark_staggered.cc
+++ b/benchmarks/Benchmark_staggered.cc
@@ -88,25 +88,6 @@ int main (int argc, char ** argv)
     U[mu] = PeekIndex<LorentzIndex>(Umu,mu);
   }
   ref = Zero();
-  /*  
-  { // Naive wilson implementation
-    ref = Zero();
-    for(int mu=0;mu<Nd;mu++){
-      //    ref =  src + Gamma(Gamma::GammaX)* src ; // 1-gamma_x
-      tmp = U[mu]*Cshift(src,mu,1);
-      for(int i=0;i<ref._odata.size();i++){
-	ref[i]+= tmp[i] - Gamma(Gmu[mu])*tmp[i]; ;
-      }
-
-      tmp =adj(U[mu])*src;
-      tmp =Cshift(tmp,mu,-1);
-      for(int i=0;i<ref._odata.size();i++){
-	ref[i]+= tmp[i] + Gamma(Gmu[mu])*tmp[i]; ;
-      }
-    }
-  }
-  ref = -0.5*ref;
-  */
 
   RealD mass=0.1;
   RealD c1=9.0/8.0;
@@ -125,10 +106,7 @@ int main (int argc, char ** argv)
   
   std::cout<<GridLogMessage << "Called Ds"<<std::endl;
   std::cout<<GridLogMessage << "norm result "<< norm2(result)<<std::endl;
-  std::cout<<GridLogMessage << "norm ref    "<< norm2(ref)<<std::endl;
   std::cout<<GridLogMessage << "mflop/s =   "<< flops/(t1-t0)<<std::endl;
-  err = ref-result; 
-  std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<<std::endl;
 
   Grid_finalize();
 }
diff --git a/benchmarks/Benchmark_staggeredF.cc b/benchmarks/Benchmark_staggeredF.cc
new file mode 100644
index 00000000..f7beed2d
--- /dev/null
+++ b/benchmarks/Benchmark_staggeredF.cc
@@ -0,0 +1,114 @@
+    /*************************************************************************************
+
+    Grid physics library, www.github.com/paboyle/Grid 
+
+    Source file: ./benchmarks/Benchmark_staggered.cc
+
+    Copyright (C) 2015
+
+Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+Author: paboyle <paboyle@ph.ed.ac.uk>
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License along
+    with this program; if not, write to the Free Software Foundation, Inc.,
+    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+    See the full license in the file "LICENSE" in the top level distribution directory
+    *************************************************************************************/
+    /*  END LEGAL */
+#include <Grid/Grid.h>
+
+using namespace std;
+using namespace Grid;
+ ;
+
+int main (int argc, char ** argv)
+{
+  Grid_init(&argc,&argv);
+
+  Coordinate latt_size   = GridDefaultLatt();
+  Coordinate simd_layout = GridDefaultSimd(Nd,vComplexF::Nsimd());
+  Coordinate mpi_layout  = GridDefaultMpi();
+  GridCartesian               Grid(latt_size,simd_layout,mpi_layout);
+  GridRedBlackCartesian     RBGrid(&Grid);
+
+  int threads = GridThread::GetThreads();
+  std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
+  std::cout<<GridLogMessage << "Grid floating point word size is REALF"<< sizeof(RealF)<<std::endl;
+  std::cout<<GridLogMessage << "Grid floating point word size is REALD"<< sizeof(RealD)<<std::endl;
+  std::cout<<GridLogMessage << "Grid floating point word size is REAL"<< sizeof(Real)<<std::endl;
+
+  std::vector<int> seeds({1,2,3,4});
+  GridParallelRNG          pRNG(&Grid);
+  pRNG.SeedFixedIntegers(seeds);
+  //  pRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9});
+
+  typedef typename ImprovedStaggeredFermionF::FermionField FermionField; 
+  typename ImprovedStaggeredFermionF::ImplParams params; 
+
+  FermionField src   (&Grid); random(pRNG,src);
+  FermionField result(&Grid); result=Zero();
+  FermionField    ref(&Grid);    ref=Zero();
+  FermionField    tmp(&Grid);    tmp=Zero();
+  FermionField    err(&Grid);    tmp=Zero();
+  LatticeGaugeFieldF Umu(&Grid); random(pRNG,Umu);
+  std::vector<LatticeColourMatrixF> U(4,&Grid);
+
+  double volume=1;
+  for(int mu=0;mu<Nd;mu++){
+    volume=volume*latt_size[mu];
+  }  
+
+  // Only one non-zero (y)
+#if 0
+  Umu=Zero();
+  Complex cone(1.0,0.0);
+  for(int nn=0;nn<Nd;nn++){
+    random(pRNG,U[nn]);
+    if(1) {
+      if (nn!=2) { U[nn]=Zero(); std::cout<<GridLogMessage << "zeroing gauge field in dir "<<nn<<std::endl; }
+      //      else       { U[nn]= cone;std::cout<<GridLogMessage << "unit gauge field in dir "<<nn<<std::endl; }
+      else       { std::cout<<GridLogMessage << "random gauge field in dir "<<nn<<std::endl; }
+    }
+    PokeIndex<LorentzIndex>(Umu,U[nn],nn);
+  }
+#endif
+
+  for(int mu=0;mu<Nd;mu++){
+    U[mu] = PeekIndex<LorentzIndex>(Umu,mu);
+  }
+
+  RealD mass=0.1;
+  RealD c1=9.0/8.0;
+  RealD c2=-1.0/24.0;
+  RealD u0=1.0;
+  ImprovedStaggeredFermionF Ds(Umu,Umu,Grid,RBGrid,mass,c1,c2,u0,params);
+  
+  std::cout<<GridLogMessage << "Calling Ds"<<std::endl;
+  int ncall=1000;
+  for(int i=0;i<ncall;i++){
+    Ds.Dhop(src,result,0);
+  }
+  double t0=usecond();
+  for(int i=0;i<ncall;i++){
+    Ds.Dhop(src,result,0);
+  }
+  double t1=usecond();
+  double flops=(16*(3*(6+8+8)) + 15*3*2)*volume*ncall; // == 66*16 +  == 1146
+  
+  std::cout<<GridLogMessage << "Called Ds"<<std::endl;
+  std::cout<<GridLogMessage << "norm result "<< norm2(result)<<std::endl;
+  std::cout<<GridLogMessage << "mflop/s =   "<< flops/(t1-t0)<<std::endl;
+
+  Grid_finalize();
+}
diff --git a/benchmarks/Benchmark_su3_gpu.cc b/benchmarks/Benchmark_su3_gpu.cc
index c87f501e..93592ca2 100644
--- a/benchmarks/Benchmark_su3_gpu.cc
+++ b/benchmarks/Benchmark_su3_gpu.cc
@@ -41,7 +41,7 @@ int main (int argc, char ** argv)
 #define LADD (8)
 
   int64_t Nwarm=20;
-  int64_t Nloop=500;
+  int64_t Nloop=50;
 
   Coordinate simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd());
   Coordinate mpi_layout  = GridDefaultMpi();
@@ -66,9 +66,9 @@ int main (int argc, char ** argv)
       LatticeColourMatrix x(&Grid);// random(pRNG,x);
       LatticeColourMatrix y(&Grid);// random(pRNG,y);
 
-      auto x_v = x.View();
-      auto y_v = y.View();
-      auto z_v = z.View();
+      autoView( x_v , x, AcceleratorRead);
+      autoView( y_v , y, AcceleratorRead);
+      autoView( z_v , z, AcceleratorWrite);
       const uint64_t Nsite = x_v.size();
       const uint64_t nsimd = vComplex::Nsimd();
       for(int64_t i=0;i<Nwarm;i++){
@@ -116,9 +116,9 @@ int main (int argc, char ** argv)
       LatticeColourMatrix x(&Grid);// random(pRNG,x);
       LatticeColourMatrix y(&Grid);// random(pRNG,y);
 
-      auto x_v = x.View();
-      auto y_v = y.View();
-      auto z_v = z.View();
+      autoView( x_v , x, AcceleratorWrite);
+      autoView( y_v , y, AcceleratorRead);
+      autoView( z_v , z, AcceleratorRead);
       const uint64_t Nsite = x_v.size();
       const uint64_t nsimd = vComplex::Nsimd();
       for(int64_t i=0;i<Nwarm;i++){
@@ -167,9 +167,9 @@ int main (int argc, char ** argv)
       LatticeColourMatrix x(&Grid);// random(pRNG,x);
       LatticeColourMatrix y(&Grid);// random(pRNG,y);
 
-      auto x_v = x.View();
-      auto y_v = y.View();
-      auto z_v = z.View();
+      autoView( x_v , x, AcceleratorRead);
+      autoView( y_v , y, AcceleratorRead);
+      autoView( z_v , z, AcceleratorWrite);
       const uint64_t Nsite = x_v.size();
       const uint64_t nsimd = vComplex::Nsimd();
       for(int64_t i=0;i<Nwarm;i++){
@@ -220,10 +220,10 @@ int main (int argc, char ** argv)
       LatticeColourMatrix y(&Grid);// random(pRNG,y);
       LatticeColourMatrix w(&Grid);// random(pRNG,y);
 
-      auto x_v = x.View();
-      auto y_v = y.View();
-      auto z_v = z.View();
-      auto w_v = z.View();
+      autoView( x_v , x, AcceleratorRead);
+      autoView( y_v , y, AcceleratorRead);
+      autoView( z_v , z, AcceleratorRead);
+      autoView( w_v , w, AcceleratorWrite);
       const uint64_t Nsite = x_v.size();
       const uint64_t nsimd = vComplex::Nsimd();
       for(int64_t i=0;i<Nwarm;i++){
diff --git a/benchmarks/Benchmark_wilson.cc b/benchmarks/Benchmark_wilson.cc
index 25d9402d..0456c3c3 100644
--- a/benchmarks/Benchmark_wilson.cc
+++ b/benchmarks/Benchmark_wilson.cc
@@ -125,8 +125,8 @@ int main (int argc, char ** argv)
       //    ref =  src + Gamma(Gamma::Algebra::GammaX)* src ; // 1-gamma_x
       tmp = U[mu]*Cshift(src,mu,1);
       {
-	auto ref_v = ref.View();
-	auto tmp_v = tmp.View();
+	autoView( ref_v, ref, CpuWrite);
+	autoView( tmp_v, tmp, CpuWrite);
 	for(int i=0;i<ref_v.size();i++){
 	  ref_v[i]+= tmp_v[i] - Gamma(Gmu[mu])*tmp_v[i]; ;
 	}
@@ -135,8 +135,8 @@ int main (int argc, char ** argv)
       tmp =adj(U[mu])*src;
       tmp =Cshift(tmp,mu,-1);
       {
-	auto ref_v = ref.View();
-	auto tmp_v = tmp.View();
+	autoView( ref_v, ref, CpuWrite);
+	autoView( tmp_v, tmp, CpuWrite);
 	for(int i=0;i<ref_v.size();i++){
 	  ref_v[i]+= tmp_v[i] + Gamma(Gmu[mu])*tmp_v[i]; ;
 	}
@@ -209,8 +209,8 @@ int main (int argc, char ** argv)
   for(int ss=0;ss<0;ss++ ){
     for(int i=0;i<Ns;i++){
       for(int j=0;j<Nc;j++){
-	auto ref_v = ref.View();
-	auto result_v = result.View();
+	autoView( ref_v, ref, CpuWrite);
+	autoView( result_v, result, CpuWrite);
 	ComplexF * ref_p = (ComplexF *)&ref_v[ss]()(i)(j);
 	ComplexF * res_p = (ComplexF *)&result_v[ss]()(i)(j);
 	std::cout<<GridLogMessage << ss<< " "<<i<<" "<<j<<" "<< (*ref_p)<<" " <<(*res_p)<<std::endl;
@@ -226,8 +226,8 @@ int main (int argc, char ** argv)
       //    ref =  src - Gamma(Gamma::Algebra::GammaX)* src ; // 1+gamma_x
       tmp = U[mu]*Cshift(src,mu,1);
       {
-	auto ref_v = ref.View();
-	auto tmp_v = tmp.View();
+	autoView( ref_v, ref, CpuWrite);
+	autoView( tmp_v, tmp, CpuWrite);
 	for(int i=0;i<ref_v.size();i++){
 	  ref_v[i]+= tmp_v[i] + Gamma(Gmu[mu])*tmp_v[i]; ;
 	}
@@ -236,8 +236,8 @@ int main (int argc, char ** argv)
       tmp =adj(U[mu])*src;
       tmp =Cshift(tmp,mu,-1);
       {
-	auto ref_v = ref.View();
-	auto tmp_v = tmp.View();
+	autoView( ref_v, ref, CpuWrite);
+	autoView( tmp_v, tmp, CpuWrite);
 	for(int i=0;i<ref_v.size();i++){
 	  ref_v[i]+= tmp_v[i] - Gamma(Gmu[mu])*tmp_v[i]; ;
 	}
diff --git a/configure.ac b/configure.ac
index e232a12c..878d56e6 100644
--- a/configure.ac
+++ b/configure.ac
@@ -138,7 +138,7 @@ esac
 ############### SUMMIT JSRUN
 AC_ARG_ENABLE([summit],
     [AC_HELP_STRING([--enable-summit=yes|no], [enable IBMs jsrun resource manager for SUMMIT])],
-    [ac_JSRUN=${enable_summit}], [ac_SUMMIT=no])
+    [ac_SUMMIT=${enable_summit}], [ac_SUMMIT=no])
 case ${ac_SUMMIT} in
     no);;
     yes)
@@ -147,6 +147,41 @@ case ${ac_SUMMIT} in
       AC_DEFINE([GRID_IBM_SUMMIT],[1],[Let JSRUN manage the GPU device allocation]);;
 esac
 
+############### SYCL/CUDA/HIP/none
+AC_ARG_ENABLE([accelerator],
+    [AC_HELP_STRING([--enable-accelerator=cuda|sycl|hip|none], [enable none,cuda,sycl,hip acceleration])],
+    [ac_ACCELERATOR=${enable_accelerator}], [ac_ACCELERATOR=none])
+case ${ac_ACCELERATOR} in
+    cuda)
+      echo CUDA acceleration
+      AC_DEFINE([GRID_CUDA],[1],[Use CUDA offload]);;
+    sycl)
+      echo SYCL acceleration
+      AC_DEFINE([GRID_SYCL],[1],[Use SYCL offload]);;
+    hip)
+      echo HIP acceleration
+      AC_DEFINE([GRID_HIP],[1],[Use HIP offload]);;
+    none)
+      echo NO acceleration
+    ;;
+    *)
+      AC_MSG_ERROR(["Acceleration not suppoorted ${ac_ACCELERATOR}"]);;
+esac
+
+############### UNIFIED MEMORY
+AC_ARG_ENABLE([unified],
+    [AC_HELP_STRING([--enable-unified=yes|no], [enable unified address space for accelerator loops])],
+    [ac_UNIFIED=${enable_unified}], [ac_UNIFIED=yes])
+case ${ac_UNIFIED} in
+    yes)
+      echo Unified memory for accelerator loops
+      AC_DEFINE([GRID_UVM],[1],[Use unified address space]);;
+    no)
+      echo Manual memory copy for accelerator loops;;
+    *)
+      AC_MSG_ERROR(["Unified virtual memory option not suppoorted ${ac_UNIFIED}"]);;
+esac
+
 ############### Intel libraries
 AC_ARG_ENABLE([mkl],
     [AC_HELP_STRING([--enable-mkl=yes|no|prefix], [enable Intel MKL for LAPACK & FFTW])],
@@ -226,7 +261,7 @@ AC_SEARCH_LIBS([fftw_execute], [fftw3],
 AC_SEARCH_LIBS([limeCreateReader], [lime],
                [AC_DEFINE([HAVE_LIME], [1], [Define to 1 if you have the `LIME' library])]
                [have_lime=true],
-	             [AC_MSG_ERROR(LIME library was not found in your system.)])
+	             [AC_MSG_WARN(LIME library was not found in your system.)])
 
 AC_SEARCH_LIBS([SHA256_Init], [crypto],
                [AC_DEFINE([HAVE_CRYPTO], [1], [Define to 1 if you have the `OpenSSL' library])]
@@ -274,18 +309,33 @@ case ${ac_gen_scalar} in
 esac
 
 ##################### Compiler dependent choices
-case ${CXX} in
+
+#Strip any optional compiler arguments from nvcc call (eg -ccbin) for compiler comparison
+CXXBASE=${CXX}
+CXXTEST=${CXX}
+if echo "${CXX}" | grep -q "nvcc"; then
+  CXXTEST="nvcc"
+fi
+
+case ${CXXTEST} in
   nvcc)
 #    CXX="nvcc -keep -v -x cu "
 #    CXXLD="nvcc -v -link"
-    CXX="nvcc -x cu "
-    CXXLD="nvcc -link"
+    CXX="${CXXBASE} -x cu "
+    CXXLD="${CXXBASE} -link"
 #    CXXFLAGS="$CXXFLAGS -Xcompiler -fno-strict-aliasing -Xcompiler -Wno-unusable-partial-specialization --expt-extended-lambda --expt-relaxed-constexpr"
     CXXFLAGS="$CXXFLAGS -Xcompiler -fno-strict-aliasing --expt-extended-lambda --expt-relaxed-constexpr"
     if test $ac_openmp = yes; then
        CXXFLAGS="$CXXFLAGS -Xcompiler -fopenmp"
     fi
     ;;
+  hipcc)
+    CXXFLAGS="$CXXFLAGS -Xcompiler -fno-strict-aliasing --expt-extended-lambda --expt-relaxed-constexpr"
+    CXXLD=${CXX}
+    if test $ac_openmp = yes; then
+       CXXFLAGS="$CXXFLAGS -Xcompiler -fopenmp"
+    fi
+    ;;
   *)
     CXXLD=${CXX}
     CXXFLAGS="$CXXFLAGS -fno-strict-aliasing"
@@ -420,6 +470,14 @@ case ${ac_SHM} in
 
      shmopen)
      AC_DEFINE([GRID_MPI3_SHMOPEN],[1],[GRID_MPI3_SHMOPEN] )
+     CXXFLAGS_CPY=$CXXFLAGS
+     LDFLAGS_CPY=$LDFLAGS
+     CXXFLAGS="$AM_CXXFLAGS $CXXFLAGS"
+     LDFLAGS="$AM_LDFLAGS $LDFLAGS"
+     AC_SEARCH_LIBS([shm_unlink], [rt], [],
+                    [AC_MSG_ERROR("no library found for shm_unlink")])
+     CXXFLAGS=$CXXFLAGS_CPY
+     LDFLAGS=$LDFLAGS_CPY
      ;;
 
      shmget)
@@ -547,6 +605,7 @@ DX_INIT_DOXYGEN([$PACKAGE_NAME], [doxygen.cfg])
 ############### Ouput
 cwd=`pwd -P`; cd ${srcdir}; abs_srcdir=`pwd -P`; cd ${cwd}
 GRID_CXX="$CXX"
+GRID_CXXLD="$CXXLD"
 GRID_CXXFLAGS="$AM_CXXFLAGS $CXXFLAGS"
 GRID_LDFLAGS="$AM_LDFLAGS $LDFLAGS"
 GRID_LIBS=$LIBS
@@ -561,6 +620,7 @@ AC_SUBST([AM_CFLAGS])
 AC_SUBST([AM_CXXFLAGS])
 AC_SUBST([AM_LDFLAGS])
 AC_SUBST([GRID_CXX])
+AC_SUBST([GRID_CXXLD])
 AC_SUBST([GRID_CXXFLAGS])
 AC_SUBST([GRID_LDFLAGS])
 AC_SUBST([GRID_LIBS])
@@ -584,6 +644,8 @@ compiler version            : ${ax_cv_gxx_version}
 ----- BUILD OPTIONS -----------------------------------
 SIMD                        : ${ac_SIMD}${SIMD_GEN_WIDTH_MSG}
 Threading                   : ${ac_openmp}
+Acceleration                : ${ac_ACCELERATOR}
+Unified virtual memory      : ${ac_UNIFIED}
 Communications type         : ${comms_type}
 Shared memory allocator     : ${ac_SHM}
 Shared memory mmap path     : ${ac_SHMPATH}
diff --git a/grid-config.in b/grid-config.in
index f39b01bd..af0816a5 100755
--- a/grid-config.in
+++ b/grid-config.in
@@ -19,6 +19,8 @@ Known values for OPTION are:
   --help       display this help and exit
   --version    output version information
   --git        print git revision
+  --cxx        print c++ compiler (may include some flags and spaces)
+  --cxxld      print c++ linker   (may include some flags and spaces)
 
 EOF
   
@@ -65,6 +67,10 @@ while test $# -gt 0; do
       echo @GRID_CXX@
     ;;
     
+    --cxxld)
+      echo @GRID_CXXLD@
+    ;;
+    
     --ldflags)
       echo @GRID_LDFLAGS@
     ;;
diff --git a/tests/IO/Test_openqcd_io.cc b/tests/IO/Test_openqcd_io.cc
new file mode 100644
index 00000000..765509a9
--- /dev/null
+++ b/tests/IO/Test_openqcd_io.cc
@@ -0,0 +1,84 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid
+
+Source file: ./tests/io/Test_openqcd_io.cc
+
+Copyright (C) 2015 - 2020
+
+Author: Daniel Richtmann <daniel.richtmann@ur.de>
+
+This program is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 2 of the License, or
+(at your option) any later version.
+
+This program is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along
+with this program; if not, write to the Free Software Foundation, Inc.,
+51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+See the full license in the file "LICENSE" in the top level distribution directory
+*************************************************************************************/
+/*  END LEGAL */
+
+#include <Grid/Grid.h>
+
+
+
+using namespace Grid;
+
+int main(int argc, char** argv) {
+#if !defined(GRID_COMMS_NONE)
+  Grid_init(&argc, &argv);
+
+  auto simd_layout = GridDefaultSimd(Nd, vComplex::Nsimd());
+  auto mpi_layout  = GridDefaultMpi();
+  auto latt_size   = GridDefaultLatt();
+
+  GridCartesian grid(latt_size, simd_layout, mpi_layout);
+
+  GridParallelRNG pRNG(&grid);
+
+  pRNG.SeedFixedIntegers(std::vector<int>({45, 12, 81, 9}));
+
+  LatticeGaugeField Umu_ref(&grid);
+  LatticeGaugeField Umu_me(&grid);
+  LatticeGaugeField Umu_diff(&grid);
+
+  FieldMetaData header_ref;
+  FieldMetaData header_me;
+
+  Umu_ref = Zero();
+  Umu_me  = Zero();
+
+  std::string file("/home/daniel/configs/openqcd/test_16x8_pbcn6");
+
+  if(GridCmdOptionExists(argv, argv + argc, "--config")) {
+    file = GridCmdOptionPayload(argv, argv + argc, "--config");
+    std::cout << "file: " << file << std::endl;
+    assert(!file.empty());
+  }
+
+  OpenQcdIOChromaReference::readConfiguration(Umu_ref, header_ref, file);
+  OpenQcdIO::readConfiguration(Umu_me, header_me, file);
+
+  std::cout << GridLogMessage << header_ref << std::endl;
+  std::cout << GridLogMessage << header_me << std::endl;
+
+  Umu_diff = Umu_ref - Umu_me;
+
+  // clang-format off
+  std::cout << GridLogMessage
+            << "norm2(Umu_ref) = " << norm2(Umu_ref)
+            << " norm2(Umu_me) = " << norm2(Umu_me)
+            << " norm2(Umu_diff) = " << norm2(Umu_diff) << std::endl;
+  // clang-format on
+
+  Grid_finalize();
+#endif
+}
diff --git a/tests/Test_dwf_mixedcg_prec.cc b/tests/Test_dwf_mixedcg_prec.cc
index 0b8463d9..be881db9 100644
--- a/tests/Test_dwf_mixedcg_prec.cc
+++ b/tests/Test_dwf_mixedcg_prec.cc
@@ -29,7 +29,6 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 
 using namespace std;
 using namespace Grid;
- ;
 
 template<class d>
 struct scal {
@@ -51,6 +50,7 @@ int main (int argc, char ** argv)
 
   std::cout << GridLogMessage << "::::: NB: to enable a quick bit reproducibility check use the --checksums flag. " << std::endl;
 
+  { 
   GridCartesian         * UGrid   = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplexD::Nsimd()),GridDefaultMpi());
   GridRedBlackCartesian * UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
   GridCartesian         * FGrid   = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
@@ -100,6 +100,8 @@ int main (int argc, char ** argv)
   ConjugateGradient<LatticeFermionD> CG(1.0e-8,10000);
   CG(HermOpEO,src_o,result_o_2);
 
+  MemoryManager::Print();
+
   LatticeFermionD diff_o(FrbGrid);
   RealD diff = axpy_norm(diff_o, -1.0, result_o, result_o_2);
 
@@ -130,7 +132,9 @@ int main (int argc, char ** argv)
   std::cout << GridLogMessage << " CG checksums "<<std::hex << scidac_csuma << " "<<scidac_csumb<<std::endl;
   }
   #endif
-
+  }
   
+  MemoryManager::Print();
+
   Grid_finalize();
 }
diff --git a/tests/Test_general_stencil.cc b/tests/Test_general_stencil.cc
index f03677d0..74b6fb4f 100644
--- a/tests/Test_general_stencil.cc
+++ b/tests/Test_general_stencil.cc
@@ -107,9 +107,8 @@ int main(int argc, char ** argv)
 	// Implement a stencil code that should agree with cshift!
 	for(int i=0;i<Check.Grid()->oSites();i++){
 	  auto SE = gStencil.GetEntry(0,i);
-	  auto check = Check.View();
-	  auto foo   = Foo.View();
-
+	  autoView(check, Check, CpuWrite);
+	  autoView(  foo, Foo, CpuRead);
 	  // Encapsulate in a general wrapper
 	  check[i] = foo[SE->_offset];                   auto tmp=check[i];
 	  if (SE->_permute & 0x1 ) { permute(check[i],tmp,0); tmp=check[i];}
@@ -147,8 +146,8 @@ int main(int argc, char ** argv)
 	}}}}
 
 	if (nrm > 1.0e-4) {
-	  auto check = Check.View();
-	  auto bar   = Bar.View();
+	  autoView( check , Check, CpuRead);
+	  autoView(   bar ,   Bar, CpuRead);
 	  for(int i=0;i<check.size();i++){
 	    std::cout << i<<" Check "<<check[i]<< "\n"<<i<<" Bar "<<bar[i]<<std::endl;
 	  }
diff --git a/tests/Test_innerproduct_norm.cc b/tests/Test_innerproduct_norm.cc
new file mode 100644
index 00000000..a8718c6b
--- /dev/null
+++ b/tests/Test_innerproduct_norm.cc
@@ -0,0 +1,126 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid
+
+Source file: ./tests/Test_innerproduct_norm.cc
+
+Copyright (C) 2015
+
+Author: Daniel Richtmann <daniel.richtmann@ur.de>
+
+This program is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 2 of the License, or
+(at your option) any later version.
+
+This program is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along
+with this program; if not, write to the Free Software Foundation, Inc.,
+51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+See the full license in the file "LICENSE" in the top level distribution directory
+*************************************************************************************/
+/*  END LEGAL */
+#include <Grid/Grid.h>
+
+using namespace Grid;
+
+int main(int argc, char** argv) {
+  Grid_init(&argc, &argv);
+
+  const int nIter = 100;
+
+  // clang-format off
+  GridCartesian *Grid_d = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd, vComplexD::Nsimd()), GridDefaultMpi());
+  GridCartesian *Grid_f = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd, vComplexF::Nsimd()), GridDefaultMpi());
+  // clang-format on
+
+  GridParallelRNG pRNG_d(Grid_d);
+  GridParallelRNG pRNG_f(Grid_f);
+
+  std::vector<int> seeds_d({1, 2, 3, 4});
+  std::vector<int> seeds_f({5, 6, 7, 8});
+
+  pRNG_d.SeedFixedIntegers(seeds_d);
+  pRNG_f.SeedFixedIntegers(seeds_f);
+
+  // clang-format off
+  LatticeFermionD x_d(Grid_d); random(pRNG_d, x_d);
+  LatticeFermionD y_d(Grid_d); random(pRNG_d, y_d);
+  LatticeFermionF x_f(Grid_f); random(pRNG_f, x_f);
+  LatticeFermionF y_f(Grid_f); random(pRNG_f, y_f);
+  // clang-format on
+
+  GridStopWatch sw_ref;
+  GridStopWatch sw_res;
+
+  { // double precision
+    ComplexD ip_d_ref, ip_d_res, diff_ip_d;
+    RealD    norm2_d_ref, norm2_d_res, diff_norm2_d;
+
+    sw_ref.Reset();
+    sw_ref.Start();
+    for(int i = 0; i < nIter; ++i) {
+      ip_d_ref    = innerProduct(x_d, y_d);
+      norm2_d_ref = norm2(x_d);
+    }
+    sw_ref.Stop();
+
+    sw_res.Reset();
+    sw_res.Start();
+    for(int i = 0; i < nIter; ++i) { innerProductNorm(ip_d_res, norm2_d_res, x_d, y_d); }
+    sw_res.Stop();
+
+    diff_ip_d    = ip_d_ref - ip_d_res;
+    diff_norm2_d = norm2_d_ref - norm2_d_res;
+
+    // clang-format off
+    std::cout << GridLogMessage << "Double: ip_ref = " << ip_d_ref << " ip_res = " << ip_d_res << " diff = " << diff_ip_d << std::endl;
+    std::cout << GridLogMessage << "Double: norm2_ref = " << norm2_d_ref << " norm2_res = " << norm2_d_res << " diff = " << diff_norm2_d << std::endl;
+    std::cout << GridLogMessage << "Double: time_ref = " << sw_ref.Elapsed() << " time_res = " << sw_res.Elapsed() << std::endl;
+    // clang-format on
+
+    assert(diff_ip_d == 0.);
+    assert(diff_norm2_d == 0.);
+
+    std::cout << GridLogMessage << "Double: all checks passed" << std::endl;
+  }
+
+  { // single precision
+    ComplexD ip_f_ref, ip_f_res, diff_ip_f;
+    RealD    norm2_f_ref, norm2_f_res, diff_norm2_f;
+
+    sw_ref.Reset();
+    sw_ref.Start();
+    for(int i = 0; i < nIter; ++i) {
+      ip_f_ref    = innerProduct(x_f, y_f);
+      norm2_f_ref = norm2(x_f);
+    }
+    sw_ref.Stop();
+
+    sw_res.Reset();
+    sw_res.Start();
+    for(int i = 0; i < nIter; ++i) { innerProductNorm(ip_f_res, norm2_f_res, x_f, y_f); }
+    sw_res.Stop();
+
+    diff_ip_f    = ip_f_ref - ip_f_res;
+    diff_norm2_f = norm2_f_ref - norm2_f_res;
+
+    // clang-format off
+    std::cout << GridLogMessage << "Single: ip_ref = " << ip_f_ref << " ip_res = " << ip_f_res << " diff = " << diff_ip_f << std::endl;
+    std::cout << GridLogMessage << "Single: norm2_ref = " << norm2_f_ref << " norm2_res = " << norm2_f_res << " diff = " << diff_norm2_f << std::endl;
+    std::cout << GridLogMessage << "Single: time_ref = " << sw_ref.Elapsed() << " time_res = " << sw_res.Elapsed() << std::endl;
+    // clang-format on
+
+    assert(diff_ip_f == 0.);
+    assert(diff_norm2_f == 0.);
+
+    std::cout << GridLogMessage << "Single: all checks passed" << std::endl;
+  }
+
+  Grid_finalize();
+}
diff --git a/tests/Test_stencil.cc b/tests/Test_stencil.cc
index 0bd97fea..16a9138a 100644
--- a/tests/Test_stencil.cc
+++ b/tests/Test_stencil.cc
@@ -109,8 +109,8 @@ int main(int argc, char ** argv) {
 	  StencilEntry *SE;
 	  SE = myStencil.GetEntry(permute_type,0,i);
 	  
-	  auto check = Check.View();
-	  auto foo   = Foo.View();
+	  autoView( check , Check, CpuWrite);
+	  autoView( foo   , Foo, CpuRead);
 	  if ( SE->_is_local && SE->_permute )
 	    permute(check[i],foo[SE->_offset],permute_type);
 	  else if (SE->_is_local)
@@ -151,8 +151,8 @@ int main(int argc, char ** argv) {
 	}}}}
 
 	if (nrm > 1.0e-4) {
-	  auto check = Check.View();
-	  auto bar   = Bar.View();
+	  autoView( check , Check, CpuRead);
+	  autoView( bar   , Bar, CpuRead);
 	  for(int i=0;i<check.size();i++){
 	    std::cout << i<<" Check "<<check[i]<< "\n"<<i<<" Bar "<<bar[i]<<std::endl;
 	  }
@@ -210,8 +210,8 @@ int main(int argc, char ** argv) {
 	  SE = EStencil.GetEntry(permute_type,0,i);
 	  //	  std::cout << "Even source "<< i<<" -> " <<SE->_offset << " "<< SE->_is_local<<std::endl;
 
-	  auto ocheck = OCheck.View();
-	  auto efoo   = EFoo.View();
+	  autoView( ocheck , OCheck, CpuWrite);
+	  autoView( efoo   , EFoo, CpuRead);
 	  if ( SE->_is_local && SE->_permute )
 	    permute(ocheck[i],efoo[SE->_offset],permute_type);
 	  else if (SE->_is_local)
@@ -226,8 +226,8 @@ int main(int argc, char ** argv) {
 	  SE = OStencil.GetEntry(permute_type,0,i);
 	  //	  std::cout << "ODD source "<< i<<" -> " <<SE->_offset << " "<< SE->_is_local<<std::endl;
 
-	  auto echeck = ECheck.View();
-	  auto ofoo   = OFoo.View();
+	  autoView( echeck , ECheck, CpuWrite);
+	  autoView( ofoo   , OFoo, CpuRead);
 	  if ( SE->_is_local && SE->_permute )
 	    permute(echeck[i],ofoo[SE->_offset],permute_type);
 	  else if (SE->_is_local)
diff --git a/tests/core/Test_contfrac_even_odd.cc b/tests/core/Test_contfrac_even_odd.cc
index 25affd00..5311f869 100644
--- a/tests/core/Test_contfrac_even_odd.cc
+++ b/tests/core/Test_contfrac_even_odd.cc
@@ -238,11 +238,11 @@ void  TestWhat(What & Ddwf,
   RealD t1,t2;
 
   SchurDiagMooeeOperator<What,LatticeFermion> HermOpEO(Ddwf);
-  HermOpEO.MpcDagMpc(chi_e,dchi_e,t1,t2);
-  HermOpEO.MpcDagMpc(chi_o,dchi_o,t1,t2);
+  HermOpEO.MpcDagMpc(chi_e,dchi_e);
+  HermOpEO.MpcDagMpc(chi_o,dchi_o);
   
-  HermOpEO.MpcDagMpc(phi_e,dphi_e,t1,t2);
-  HermOpEO.MpcDagMpc(phi_o,dphi_o,t1,t2);
+  HermOpEO.MpcDagMpc(phi_e,dphi_e);
+  HermOpEO.MpcDagMpc(phi_o,dphi_o);
 
   pDce = innerProduct(phi_e,dchi_e);
   pDco = innerProduct(phi_o,dchi_o);
diff --git a/tests/core/Test_cshift_red_black.cc b/tests/core/Test_cshift_red_black.cc
index 34325072..4fdd5fc0 100644
--- a/tests/core/Test_cshift_red_black.cc
+++ b/tests/core/Test_cshift_red_black.cc
@@ -82,7 +82,7 @@ int main (int argc, char ** argv)
   pickCheckerboard(Odd,Uo,U);
 
   //  std::cout<<GridLogMessage << U<<std::endl;
-  std::cout<<GridLogMessage<< U <<std::endl;
+  //  std::cout<<GridLogMessage<< U <<std::endl;
   std::cout<<GridLogMessage << "U " <<norm2(U)<<std::endl;
   std::cout<<GridLogMessage << "Ue " <<norm2(Ue)<<std::endl;
   std::cout<<GridLogMessage << "Uo " <<norm2(Uo)<<std::endl;
diff --git a/tests/core/Test_cshift_rotate.cc b/tests/core/Test_cshift_rotate.cc
index 9e572b70..80c4813c 100644
--- a/tests/core/Test_cshift_rotate.cc
+++ b/tests/core/Test_cshift_rotate.cc
@@ -69,6 +69,7 @@ int main (int argc, char ** argv)
 
 	ShiftU  = Cshift(U,dir,shift);    // Shift everything
 
+	std::cout<<GridLogMessage<<"Shifted by "<<shift<<" in direction"<<dir<<" checking the AE35 unit" <<std::endl;
 	/*
 	std::cout << "U[0]" << U[0]<<std::endl;
 	std::cout << "U[1]" << U[1]<<std::endl;
diff --git a/tests/core/Test_dwf_eofa_even_odd.cc b/tests/core/Test_dwf_eofa_even_odd.cc
index 1d2f2909..01fff9ea 100644
--- a/tests/core/Test_dwf_eofa_even_odd.cc
+++ b/tests/core/Test_dwf_eofa_even_odd.cc
@@ -218,11 +218,11 @@ int main (int argc, char ** argv)
     RealD t1,t2;
 
     SchurDiagMooeeOperator<DomainWallEOFAFermionR,LatticeFermion> HermOpEO(Ddwf);
-    HermOpEO.MpcDagMpc(chi_e, dchi_e, t1, t2);
-    HermOpEO.MpcDagMpc(chi_o, dchi_o, t1, t2);
+    HermOpEO.MpcDagMpc(chi_e, dchi_e);
+    HermOpEO.MpcDagMpc(chi_o, dchi_o);
 
-    HermOpEO.MpcDagMpc(phi_e, dphi_e, t1, t2);
-    HermOpEO.MpcDagMpc(phi_o, dphi_o, t1, t2);
+    HermOpEO.MpcDagMpc(phi_e, dphi_e);
+    HermOpEO.MpcDagMpc(phi_o, dphi_o);
 
     pDce = innerProduct(phi_e, dchi_e);
     pDco = innerProduct(phi_o, dchi_o);
diff --git a/tests/core/Test_dwf_even_odd.cc b/tests/core/Test_dwf_even_odd.cc
index d654e588..6093ee8f 100644
--- a/tests/core/Test_dwf_even_odd.cc
+++ b/tests/core/Test_dwf_even_odd.cc
@@ -216,11 +216,11 @@ int main (int argc, char ** argv)
 
 
   SchurDiagMooeeOperator<DomainWallFermionR,LatticeFermion> HermOpEO(Ddwf);
-  HermOpEO.MpcDagMpc(chi_e,dchi_e,t1,t2);
-  HermOpEO.MpcDagMpc(chi_o,dchi_o,t1,t2);
+  HermOpEO.MpcDagMpc(chi_e,dchi_e);
+  HermOpEO.MpcDagMpc(chi_o,dchi_o);
 
-  HermOpEO.MpcDagMpc(phi_e,dphi_e,t1,t2);
-  HermOpEO.MpcDagMpc(phi_o,dphi_o,t1,t2);
+  HermOpEO.MpcDagMpc(phi_e,dphi_e);
+  HermOpEO.MpcDagMpc(phi_o,dphi_o);
 
   pDce = innerProduct(phi_e,dchi_e);
   pDco = innerProduct(phi_o,dchi_o);
diff --git a/tests/core/Test_gpwilson_even_odd.cc b/tests/core/Test_gpwilson_even_odd.cc
index ac4cde99..bf37f4d5 100644
--- a/tests/core/Test_gpwilson_even_odd.cc
+++ b/tests/core/Test_gpwilson_even_odd.cc
@@ -201,11 +201,11 @@ int main (int argc, char ** argv)
   RealD t1,t2;
 
   SchurDiagMooeeOperator<GparityWilsonFermionR,FermionField> HermOpEO(Dw);
-  HermOpEO.MpcDagMpc(chi_e,dchi_e,t1,t2);
-  HermOpEO.MpcDagMpc(chi_o,dchi_o,t1,t2);
+  HermOpEO.MpcDagMpc(chi_e,dchi_e);
+  HermOpEO.MpcDagMpc(chi_o,dchi_o);
 
-  HermOpEO.MpcDagMpc(phi_e,dphi_e,t1,t2);
-  HermOpEO.MpcDagMpc(phi_o,dphi_o,t1,t2);
+  HermOpEO.MpcDagMpc(phi_e,dphi_e);
+  HermOpEO.MpcDagMpc(phi_o,dphi_o);
 
   pDce = innerProduct(phi_e,dchi_e);
   pDco = innerProduct(phi_o,dchi_o);
diff --git a/tests/core/Test_main.cc b/tests/core/Test_main.cc
index cd2373ce..08752a46 100644
--- a/tests/core/Test_main.cc
+++ b/tests/core/Test_main.cc
@@ -73,7 +73,7 @@ int main(int argc, char **argv) {
     omp_set_num_threads(omp);
 #endif
 
-    for (int lat = 8; lat <= 16; lat += 40) {
+    for (int lat = 16; lat <= 16; lat += 40) {
       std::cout << "Lat " << lat << std::endl;
 
       latt_size[0] = lat;
@@ -159,15 +159,17 @@ int main(int argc, char **argv) {
       LatticeColourMatrix newFoo = Foo; 
       // confirm correctness of copy constructor
       Bar = Foo - newFoo;
-      std::cout << "Copy constructor diff check: "; 
+      std::cout << "Copy constructor diff check: \n"; 
       double test_cc = norm2(Bar);
       if (test_cc < 1e-5){
         std::cout << "OK\n";
-    }
-      else{
+      } else{
+	std::cout << "Foo\n"<<Foo<<std::endl;
+	std::cout << "newFoo\n"<<newFoo<<std::endl;
+	std::cout << "Bar\n"<<Bar<<std::endl;
         std::cout << "fail\n";
         abort();
-    }
+      }
 
       // Norm2 check
       LatticeReal BarReal(&Fine);
diff --git a/tests/core/Test_mobius_eofa_even_odd.cc b/tests/core/Test_mobius_eofa_even_odd.cc
index bfd53c72..68091229 100644
--- a/tests/core/Test_mobius_eofa_even_odd.cc
+++ b/tests/core/Test_mobius_eofa_even_odd.cc
@@ -220,11 +220,11 @@ int main (int argc, char ** argv)
     RealD t1,t2;
 
     SchurDiagMooeeOperator<MobiusEOFAFermionR,LatticeFermion> HermOpEO(Ddwf);
-    HermOpEO.MpcDagMpc(chi_e, dchi_e, t1, t2);
-    HermOpEO.MpcDagMpc(chi_o, dchi_o, t1, t2);
+    HermOpEO.MpcDagMpc(chi_e, dchi_e);
+    HermOpEO.MpcDagMpc(chi_o, dchi_o);
 
-    HermOpEO.MpcDagMpc(phi_e, dphi_e, t1, t2);
-    HermOpEO.MpcDagMpc(phi_o, dphi_o, t1, t2);
+    HermOpEO.MpcDagMpc(phi_e, dphi_e);
+    HermOpEO.MpcDagMpc(phi_o, dphi_o);
 
     pDce = innerProduct(phi_e, dchi_e);
     pDco = innerProduct(phi_o, dchi_o);
diff --git a/tests/core/Test_mobius_even_odd.cc b/tests/core/Test_mobius_even_odd.cc
index 0a035dc8..7f808cac 100644
--- a/tests/core/Test_mobius_even_odd.cc
+++ b/tests/core/Test_mobius_even_odd.cc
@@ -266,11 +266,11 @@ int main (int argc, char ** argv)
 
 
   SchurDiagMooeeOperator<MobiusFermionR,LatticeFermion> HermOpEO(Ddwf);
-  HermOpEO.MpcDagMpc(chi_e,dchi_e,t1,t2);
-  HermOpEO.MpcDagMpc(chi_o,dchi_o,t1,t2);
+  HermOpEO.MpcDagMpc(chi_e,dchi_e);
+  HermOpEO.MpcDagMpc(chi_o,dchi_o);
 
-  HermOpEO.MpcDagMpc(phi_e,dphi_e,t1,t2);
-  HermOpEO.MpcDagMpc(phi_o,dphi_o,t1,t2);
+  HermOpEO.MpcDagMpc(phi_e,dphi_e);
+  HermOpEO.MpcDagMpc(phi_o,dphi_o);
 
   pDce = innerProduct(phi_e,dchi_e);
   pDco = innerProduct(phi_o,dchi_o);
diff --git a/tests/core/Test_staggered.cc b/tests/core/Test_staggered.cc
index c85d4090..1f42ff0d 100644
--- a/tests/core/Test_staggered.cc
+++ b/tests/core/Test_staggered.cc
@@ -270,11 +270,11 @@ int main (int argc, char ** argv)
   pickCheckerboard(Odd ,phi_o,phi);
 
   SchurDiagMooeeOperator<ImprovedStaggeredFermionR,FermionField> HermOpEO(Ds);
-  HermOpEO.MpcDagMpc(chi_e,dchi_e,t1,t2);
-  HermOpEO.MpcDagMpc(chi_o,dchi_o,t1,t2);
+  HermOpEO.MpcDagMpc(chi_e,dchi_e);
+  HermOpEO.MpcDagMpc(chi_o,dchi_o);
 
-  HermOpEO.MpcDagMpc(phi_e,dphi_e,t1,t2);
-  HermOpEO.MpcDagMpc(phi_o,dphi_o,t1,t2);
+  HermOpEO.MpcDagMpc(phi_e,dphi_e);
+  HermOpEO.MpcDagMpc(phi_o,dphi_o);
 
   pDce = innerProduct(phi_e,dchi_e);
   pDco = innerProduct(phi_o,dchi_o);
diff --git a/tests/core/Test_staggered5D.cc b/tests/core/Test_staggered5D.cc
index 7055d183..3d175890 100644
--- a/tests/core/Test_staggered5D.cc
+++ b/tests/core/Test_staggered5D.cc
@@ -88,14 +88,15 @@ int main (int argc, char ** argv)
   // replicate across fifth dimension
   ////////////////////////////////////
   LatticeGaugeField Umu5d(FGrid); 
-  auto umu5d = Umu5d.View();
-  auto umu   = Umu.View();
-  for(int ss=0;ss<Umu.Grid()->oSites();ss++){
-    for(int s=0;s<Ls;s++){
-      umu5d[Ls*ss+s] = umu[ss];
+  {
+    autoView(umu5d, Umu5d, CpuWrite);
+    autoView(  umu, Umu  , CpuRead);
+    for(int ss=0;ss<Umu.Grid()->oSites();ss++){
+      for(int s=0;s<Ls;s++){
+	umu5d[Ls*ss+s] = umu[ss];
+      }
     }
   }
-
   std::vector<LatticeColourMatrix> U(4,FGrid);
 
   for(int mu=0;mu<Nd;mu++){
@@ -289,11 +290,11 @@ int main (int argc, char ** argv)
   pickCheckerboard(Odd ,phi_o,phi);
 
   SchurDiagMooeeOperator<ImprovedStaggeredFermion5DR,FermionField> HermOpEO(Ds);
-  HermOpEO.MpcDagMpc(chi_e,dchi_e,t1,t2);
-  HermOpEO.MpcDagMpc(chi_o,dchi_o,t1,t2);
+  HermOpEO.MpcDagMpc(chi_e,dchi_e);
+  HermOpEO.MpcDagMpc(chi_o,dchi_o);
 
-  HermOpEO.MpcDagMpc(phi_e,dphi_e,t1,t2);
-  HermOpEO.MpcDagMpc(phi_o,dphi_o,t1,t2);
+  HermOpEO.MpcDagMpc(phi_e,dphi_e);
+  HermOpEO.MpcDagMpc(phi_o,dphi_o);
 
   pDce = innerProduct(phi_e,dchi_e);
   pDco = innerProduct(phi_o,dchi_o);
diff --git a/tests/core/Test_staggered_naive.cc b/tests/core/Test_staggered_naive.cc
new file mode 100644
index 00000000..f96bac93
--- /dev/null
+++ b/tests/core/Test_staggered_naive.cc
@@ -0,0 +1,282 @@
+    /*************************************************************************************
+
+    Grid physics library, www.github.com/paboyle/Grid 
+
+    Source file: ./benchmarks/Benchmark_wilson.cc
+
+    Copyright (C) 2015
+
+Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+Author: paboyle <paboyle@ph.ed.ac.uk>
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License along
+    with this program; if not, write to the Free Software Foundation, Inc.,
+    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+    See the full license in the file "LICENSE" in the top level distribution directory
+    *************************************************************************************/
+    /*  END LEGAL */
+#include <Grid/Grid.h>
+
+using namespace std;
+using namespace Grid;
+ ;
+
+int main (int argc, char ** argv)
+{
+  Grid_init(&argc,&argv);
+
+  Coordinate latt_size   = GridDefaultLatt();
+  Coordinate simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd());
+  Coordinate mpi_layout  = GridDefaultMpi();
+  GridCartesian               Grid(latt_size,simd_layout,mpi_layout);
+  GridRedBlackCartesian     RBGrid(&Grid);
+
+  int threads = GridThread::GetThreads();
+  std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
+  std::cout<<GridLogMessage << "Grid floating point word size is REALF"<< sizeof(RealF)<<std::endl;
+  std::cout<<GridLogMessage << "Grid floating point word size is REALD"<< sizeof(RealD)<<std::endl;
+  std::cout<<GridLogMessage << "Grid floating point word size is REAL"<< sizeof(Real)<<std::endl;
+
+  std::vector<int> seeds({1,2,3,4});
+  GridParallelRNG          pRNG(&Grid);
+  pRNG.SeedFixedIntegers(seeds);
+  //  pRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9});
+
+  typedef typename NaiveStaggeredFermionR::FermionField FermionField; 
+  typedef typename NaiveStaggeredFermionR::ComplexField ComplexField; 
+  typename NaiveStaggeredFermionR::ImplParams params; 
+
+  FermionField src   (&Grid); random(pRNG,src);
+  FermionField result(&Grid); result=Zero();
+  FermionField    ref(&Grid);    ref=Zero();
+  FermionField    tmp(&Grid);    tmp=Zero();
+  FermionField    err(&Grid);    tmp=Zero();
+  FermionField phi   (&Grid); random(pRNG,phi);
+  FermionField chi   (&Grid); random(pRNG,chi);
+  LatticeGaugeField Umu(&Grid); SU3::HotConfiguration(pRNG,Umu);
+  std::vector<LatticeColourMatrix> U(4,&Grid);
+
+
+  double volume=1;
+  for(int mu=0;mu<Nd;mu++){
+    volume=volume*latt_size[mu];
+  }  
+
+  // Only one non-zero (y)
+  for(int mu=0;mu<Nd;mu++){
+    U[mu] = PeekIndex<LorentzIndex>(Umu,mu);
+  /* Debug force unit
+    U[mu] = 1.0;
+    PokeIndex<LorentzIndex>(Umu,U[mu],mu);
+  */
+  }
+
+  ref = Zero();
+
+  RealD mass=0.1;
+  RealD c1=9.0/8.0;
+  RealD u0=1.0;
+
+  { // Simple improved staggered implementation
+    ref = Zero();
+    RealD c1tad = 0.5*c1/u0;
+
+    Lattice<iScalar<vInteger> > coor(&Grid);
+
+    Lattice<iScalar<vInteger> > x(&Grid); LatticeCoordinate(x,0);
+    Lattice<iScalar<vInteger> > y(&Grid); LatticeCoordinate(y,1);
+    Lattice<iScalar<vInteger> > z(&Grid); LatticeCoordinate(z,2);
+    Lattice<iScalar<vInteger> > t(&Grid); LatticeCoordinate(t,3);
+
+    Lattice<iScalar<vInteger> > lin_z(&Grid); lin_z=x+y;
+    Lattice<iScalar<vInteger> > lin_t(&Grid); lin_t=x+y+z;
+
+    for(int mu=0;mu<Nd;mu++){
+
+      // Staggered Phase.
+      ComplexField phases(&Grid);	phases=1.0;
+  
+      if ( mu == 1 ) phases = where( mod(x    ,2)==(Integer)0, phases,-phases);
+      if ( mu == 2 ) phases = where( mod(lin_z,2)==(Integer)0, phases,-phases);
+      if ( mu == 3 ) phases = where( mod(lin_t,2)==(Integer)0, phases,-phases);
+
+      tmp = PeriodicBC::CovShiftForward(U[mu],mu,src);
+      ref = ref +c1tad*tmp*phases; // Forward 1 hop
+
+      tmp = PeriodicBC::CovShiftBackward(U[mu],mu,src);
+      ref = ref -c1tad*tmp*phases; // Backward 1 hop
+
+    }
+    //    ref = ref + mass * src;
+  }
+
+  NaiveStaggeredFermionR Ds(Umu,Grid,RBGrid,mass,c1,u0,params);
+  
+
+  std::cout<<GridLogMessage<<"=========================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"= Testing Dhop against cshift implementation         "<<std::endl;
+  std::cout<<GridLogMessage<<"=========================================================="<<std::endl;
+
+  std::cout<<GridLogMessage << "Calling Ds"<<std::endl;
+  int ncall=1000;
+  double t0=usecond();
+  for(int i=0;i<ncall;i++){
+    Ds.Dhop(src,result,0);
+  }
+  double t1=usecond();
+  double t2;
+  double flops=(16*(3*(6+8+8)) + 15*3*2)*volume*ncall; // == 66*16 +  == 1146
+  
+  std::cout<<GridLogMessage << "Called Ds"<<std::endl;
+  std::cout<<GridLogMessage << "norm result "<< norm2(result)<<std::endl;
+  std::cout<<GridLogMessage << "norm ref    "<< norm2(ref)<<std::endl;
+  std::cout<<GridLogMessage << "mflop/s =   "<< flops/(t1-t0)<<std::endl;
+
+  err = ref-result; 
+  std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<<std::endl;
+
+  std::cout<<GridLogMessage<<"=========================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"= Testing that Deo + Doe = Dunprec "<<std::endl;
+  std::cout<<GridLogMessage<<"=========================================================="<<std::endl;
+
+  FermionField src_e   (&RBGrid);
+  FermionField src_o   (&RBGrid);
+  FermionField r_e   (&RBGrid);
+  FermionField r_o   (&RBGrid);
+  FermionField r_eo  (&Grid);
+  pickCheckerboard(Even,src_e,src);
+  pickCheckerboard(Odd,src_o,src);
+
+  Ds.Meooe(src_e,r_o);  std::cout<<GridLogMessage<<"Applied Meo"<<std::endl;
+  Ds.Meooe(src_o,r_e);  std::cout<<GridLogMessage<<"Applied Moe"<<std::endl;
+  Ds.Dhop (src,ref,DaggerNo);
+
+  setCheckerboard(r_eo,r_o);
+  setCheckerboard(r_eo,r_e);
+
+  err= ref - r_eo;
+  std::cout<<GridLogMessage << "EO norm diff   "<< norm2(err)<< " "<<norm2(ref)<< " " << norm2(r_eo) <<std::endl;
+
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"= Test Ddagger is the dagger of D by requiring                "<<std::endl;
+  std::cout<<GridLogMessage<<"=  < phi | Deo | chi > * = < chi | Deo^dag| phi>  "<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
+
+  FermionField chi_e   (&RBGrid);
+  FermionField chi_o   (&RBGrid);
+
+  FermionField dchi_e  (&RBGrid);
+  FermionField dchi_o  (&RBGrid);
+
+  FermionField phi_e   (&RBGrid);
+  FermionField phi_o   (&RBGrid);
+
+  FermionField dphi_e  (&RBGrid);
+  FermionField dphi_o  (&RBGrid);
+
+  pickCheckerboard(Even,chi_e,chi);
+  pickCheckerboard(Odd ,chi_o,chi);
+  pickCheckerboard(Even,phi_e,phi);
+  pickCheckerboard(Odd ,phi_o,phi);
+
+  Ds.Meooe(chi_e,dchi_o);
+  Ds.Meooe(chi_o,dchi_e);
+  Ds.MeooeDag(phi_e,dphi_o);
+  Ds.MeooeDag(phi_o,dphi_e);
+
+  ComplexD pDce = innerProduct(phi_e,dchi_e);
+  ComplexD pDco = innerProduct(phi_o,dchi_o);
+  ComplexD cDpe = innerProduct(chi_e,dphi_e);
+  ComplexD cDpo = innerProduct(chi_o,dphi_o);
+
+  std::cout<<GridLogMessage <<"e "<<pDce<<" "<<cDpe <<std::endl;
+  std::cout<<GridLogMessage <<"o "<<pDco<<" "<<cDpo <<std::endl;
+
+  std::cout<<GridLogMessage <<"pDce - conj(cDpo) "<< pDce-conj(cDpo) <<std::endl;
+  std::cout<<GridLogMessage <<"pDco - conj(cDpe) "<< pDco-conj(cDpe) <<std::endl;
+  std::cout<<GridLogMessage <<"e "<<pDce<<" "<<cDpe <<std::endl;
+  std::cout<<GridLogMessage <<"o "<<pDco<<" "<<cDpo <<std::endl;
+
+  std::cout<<GridLogMessage <<"pDce - conj(cDpo) "<< pDce-conj(cDpo) <<std::endl;
+  std::cout<<GridLogMessage <<"pDco - conj(cDpe) "<< pDco-conj(cDpe) <<std::endl;
+
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"= Test MeeInv Mee = 1                                         "<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
+
+  pickCheckerboard(Even,chi_e,chi);
+  pickCheckerboard(Odd ,chi_o,chi);
+
+  Ds.Mooee(chi_e,src_e);
+  Ds.MooeeInv(src_e,phi_e);
+
+  Ds.Mooee(chi_o,src_o);
+  Ds.MooeeInv(src_o,phi_o);
+  
+  setCheckerboard(phi,phi_e);
+  setCheckerboard(phi,phi_o);
+
+  err = phi-chi;
+  std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<< std::endl;
+
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"= Test MeeInvDag MeeDag = 1                                   "<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
+
+  pickCheckerboard(Even,chi_e,chi);
+  pickCheckerboard(Odd ,chi_o,chi);
+
+  Ds.MooeeDag(chi_e,src_e);
+  Ds.MooeeInvDag(src_e,phi_e);
+
+  Ds.MooeeDag(chi_o,src_o);
+  Ds.MooeeInvDag(src_o,phi_o);
+  
+  setCheckerboard(phi,phi_e);
+  setCheckerboard(phi,phi_o);
+
+  err = phi-chi;
+  std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<< std::endl;
+
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"= Test MpcDagMpc is Hermitian              "<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
+  
+  random(pRNG,phi);
+  random(pRNG,chi);
+  pickCheckerboard(Even,chi_e,chi);
+  pickCheckerboard(Odd ,chi_o,chi);
+  pickCheckerboard(Even,phi_e,phi);
+  pickCheckerboard(Odd ,phi_o,phi);
+
+  SchurDiagMooeeOperator<NaiveStaggeredFermionR,FermionField> HermOpEO(Ds);
+  HermOpEO.MpcDagMpc(chi_e,dchi_e,t1,t2);
+  HermOpEO.MpcDagMpc(chi_o,dchi_o,t1,t2);
+
+  HermOpEO.MpcDagMpc(phi_e,dphi_e,t1,t2);
+  HermOpEO.MpcDagMpc(phi_o,dphi_o,t1,t2);
+
+  pDce = innerProduct(phi_e,dchi_e);
+  pDco = innerProduct(phi_o,dchi_o);
+  cDpe = innerProduct(chi_e,dphi_e);
+  cDpo = innerProduct(chi_o,dphi_o);
+
+  std::cout<<GridLogMessage <<"e "<<pDce<<" "<<cDpe <<std::endl;
+  std::cout<<GridLogMessage <<"o "<<pDco<<" "<<cDpo <<std::endl;
+
+  std::cout<<GridLogMessage <<"pDce - conj(cDpo) "<< pDco-conj(cDpo) <<std::endl;
+  std::cout<<GridLogMessage <<"pDco - conj(cDpe) "<< pDce-conj(cDpe) <<std::endl;
+
+  Grid_finalize();
+}
diff --git a/tests/core/Test_wilson_even_odd.cc b/tests/core/Test_wilson_even_odd.cc
index 8f325f1c..dc49cf81 100644
--- a/tests/core/Test_wilson_even_odd.cc
+++ b/tests/core/Test_wilson_even_odd.cc
@@ -207,11 +207,11 @@ int main (int argc, char ** argv)
   RealD t1,t2;
 
   SchurDiagMooeeOperator<WilsonFermionR,LatticeFermion> HermOpEO(Dw);
-  HermOpEO.MpcDagMpc(chi_e,dchi_e,t1,t2);
-  HermOpEO.MpcDagMpc(chi_o,dchi_o,t1,t2);
+  HermOpEO.MpcDagMpc(chi_e,dchi_e);
+  HermOpEO.MpcDagMpc(chi_o,dchi_o);
 
-  HermOpEO.MpcDagMpc(phi_e,dphi_e,t1,t2);
-  HermOpEO.MpcDagMpc(phi_o,dphi_o,t1,t2);
+  HermOpEO.MpcDagMpc(phi_e,dphi_e);
+  HermOpEO.MpcDagMpc(phi_o,dphi_o);
 
   pDce = innerProduct(phi_e,dchi_e);
   pDco = innerProduct(phi_o,dchi_o);
diff --git a/tests/core/Test_wilson_twisted_mass_even_odd.cc b/tests/core/Test_wilson_twisted_mass_even_odd.cc
index 58b0b60f..ba80fd0e 100644
--- a/tests/core/Test_wilson_twisted_mass_even_odd.cc
+++ b/tests/core/Test_wilson_twisted_mass_even_odd.cc
@@ -208,11 +208,11 @@ int main (int argc, char ** argv)
   RealD t1,t2;
 
   SchurDiagMooeeOperator<WilsonTMFermionR,LatticeFermion> HermOpEO(Dw);
-  HermOpEO.MpcDagMpc(chi_e,dchi_e,t1,t2);
-  HermOpEO.MpcDagMpc(chi_o,dchi_o,t1,t2);
+  HermOpEO.MpcDagMpc(chi_e,dchi_e);
+  HermOpEO.MpcDagMpc(chi_o,dchi_o);
 
-  HermOpEO.MpcDagMpc(phi_e,dphi_e,t1,t2);
-  HermOpEO.MpcDagMpc(phi_o,dphi_o,t1,t2);
+  HermOpEO.MpcDagMpc(phi_e,dphi_e);
+  HermOpEO.MpcDagMpc(phi_o,dphi_o);
 
   pDce = innerProduct(phi_e,dchi_e);
   pDco = innerProduct(phi_o,dchi_o);
diff --git a/tests/core/Test_zmobius_even_odd.cc b/tests/core/Test_zmobius_even_odd.cc
index 1150930b..a52e9bc2 100644
--- a/tests/core/Test_zmobius_even_odd.cc
+++ b/tests/core/Test_zmobius_even_odd.cc
@@ -280,11 +280,11 @@ int main (int argc, char ** argv)
 
 
   SchurDiagMooeeOperator<ZMobiusFermionR,LatticeFermion> HermOpEO(Ddwf);
-  HermOpEO.MpcDagMpc(chi_e,dchi_e,t1,t2);
-  HermOpEO.MpcDagMpc(chi_o,dchi_o,t1,t2);
+  HermOpEO.MpcDagMpc(chi_e,dchi_e);
+  HermOpEO.MpcDagMpc(chi_o,dchi_o);
 
-  HermOpEO.MpcDagMpc(phi_e,dphi_e,t1,t2);
-  HermOpEO.MpcDagMpc(phi_o,dphi_o,t1,t2);
+  HermOpEO.MpcDagMpc(phi_e,dphi_e);
+  HermOpEO.MpcDagMpc(phi_o,dphi_o);
 
   pDce = innerProduct(phi_e,dchi_e);
   pDco = innerProduct(phi_o,dchi_o);
diff --git a/tests/debug/Test_cayley_mres.cc b/tests/debug/Test_cayley_mres.cc
new file mode 100644
index 00000000..2ad605b8
--- /dev/null
+++ b/tests/debug/Test_cayley_mres.cc
@@ -0,0 +1,622 @@
+/*************************************************************************************
+    Grid physics library, www.github.com/paboyle/Grid 
+
+    Source file: ./tests/Test_cayley_cg.cc
+
+    Copyright (C) 2015
+
+Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+Author: paboyle <paboyle@ph.ed.ac.uk>
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License along
+    with this program; if not, write to the Free Software Foundation, Inc.,
+    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+    See the full license in the file "LICENSE" in the top level distribution directory
+    *************************************************************************************/
+    /*  END LEGAL */
+#include <Grid/Grid.h>
+#include <Grid/qcd/action/fermion/Reconstruct5Dprop.h>
+
+using namespace std;
+using namespace Grid;
+
+
+template<class What> 
+void  TestConserved(What & Ddwf, What & Ddwfrev, 
+		    LatticeGaugeField &Umu,
+		    GridCartesian         * FGrid,	       GridRedBlackCartesian * FrbGrid,
+		    GridCartesian         * UGrid,	       GridRedBlackCartesian * UrbGrid,
+		    RealD mass, RealD M5,
+		    GridParallelRNG *RNG4,
+		    GridParallelRNG *RNG5);
+
+  Gamma::Algebra Gmu [] = {
+    Gamma::Algebra::GammaX,
+    Gamma::Algebra::GammaY,
+    Gamma::Algebra::GammaZ,
+    Gamma::Algebra::GammaT,
+    Gamma::Algebra::Gamma5
+  };
+
+int main (int argc, char ** argv)
+{
+  Grid_init(&argc,&argv);
+
+  int threads = GridThread::GetThreads();
+  std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
+
+  const int Ls=10;
+  std::vector < ComplexD  > omegas;
+  std::vector < ComplexD  > omegasrev(Ls);
+
+#if 1
+  omegas.push_back( std::complex<double>(1.45806438985048,-0) );
+  omegas.push_back( std::complex<double>(0.830951166685955,-0) );
+  omegas.push_back( std::complex<double>(0.341985020453729,-0) );
+  omegas.push_back( std::complex<double>(0.126074299502912,-0) );
+  //  omegas.push_back( std::complex<double>(0.0686324988446592,0.0550658530827402) );
+  //  omegas.push_back( std::complex<double>(0.0686324988446592,-0.0550658530827402) );
+  omegas.push_back( std::complex<double>(0.0686324988446592,0));
+  omegas.push_back( std::complex<double>(0.0686324988446592,0));
+  omegas.push_back( std::complex<double>(0.0990136651962626,-0) );
+  omegas.push_back( std::complex<double>(0.21137902619029,-0) );
+  omegas.push_back( std::complex<double>(0.542352409156791,-0) );
+  omegas.push_back( std::complex<double>(1.18231318389348,-0) );
+#else 
+  omegas.push_back( std::complex<double>(0.8,0.0));
+  omegas.push_back( std::complex<double>(1.1,0.0));
+  omegas.push_back( std::complex<double>(1.2,0.0));
+  omegas.push_back( std::complex<double>(1.3,0.0));
+  omegas.push_back( std::complex<double>(0.5,0.2));
+  omegas.push_back( std::complex<double>(0.5,-0.2));
+  omegas.push_back( std::complex<double>(0.8,0.0));
+  omegas.push_back( std::complex<double>(1.1,0.0));
+  omegas.push_back( std::complex<double>(1.2,0.0));
+  omegas.push_back( std::complex<double>(1.3,0.0));
+#endif
+
+  GridCartesian         * UGrid   = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), 
+								   GridDefaultSimd(Nd,vComplex::Nsimd()),
+								   GridDefaultMpi());
+  GridRedBlackCartesian * UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
+  GridCartesian         * FGrid   = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
+  GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid);
+
+
+  GridCartesian         * UGridF   = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), 
+								    GridDefaultSimd(Nd,vComplexF::Nsimd()),
+								    GridDefaultMpi());
+  GridRedBlackCartesian * UrbGridF = SpaceTimeGrid::makeFourDimRedBlackGrid(UGridF);
+  GridCartesian         * FGridF   = SpaceTimeGrid::makeFiveDimGrid(Ls,UGridF);
+  GridRedBlackCartesian * FrbGridF = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGridF);
+
+
+  std::vector<int> seeds4({1,2,3,4});
+  std::vector<int> seeds5({5,6,7,8});
+  GridParallelRNG          RNG5(FGrid);  RNG5.SeedFixedIntegers(seeds5);
+  GridParallelRNG          RNG4(UGrid);  RNG4.SeedFixedIntegers(seeds4);
+
+  LatticeGaugeField Umu(UGrid);
+  SU3::ColdConfiguration(Umu);
+  //  SU3::HotConfiguration(RNG4,Umu);
+
+  RealD mass=0.3;
+  RealD M5  =1.0;
+  std::cout<<GridLogMessage <<"======================"<<std::endl;
+  std::cout<<GridLogMessage <<"DomainWallFermion test"<<std::endl;
+  std::cout<<GridLogMessage <<"======================"<<std::endl;
+  DomainWallFermionR Ddwf(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
+  TestConserved<DomainWallFermionR>(Ddwf,Ddwf,Umu,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,&RNG4,&RNG5);
+
+  RealD b=1.5;// Scale factor b+c=2, b-c=1
+  RealD c=0.5;
+  //  std::vector<ComplexD> gamma(Ls,ComplexD(1.0,0.0));
+
+  std::cout<<GridLogMessage <<"======================"<<std::endl;
+  std::cout<<GridLogMessage <<"MobiusFermion test"<<std::endl;
+  std::cout<<GridLogMessage <<"======================"<<std::endl;
+  MobiusFermionR Dmob(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,b,c);
+  TestConserved<MobiusFermionR>(Dmob,Dmob,Umu,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,&RNG4,&RNG5);
+
+  std::cout<<GridLogMessage <<"======================"<<std::endl;
+  std::cout<<GridLogMessage <<"ScaledShamirFermion test"<<std::endl;
+  std::cout<<GridLogMessage <<"======================"<<std::endl;
+  ScaledShamirFermionR Dsham(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,2.0);
+  TestConserved<ScaledShamirFermionR>(Dsham,Dsham,Umu,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,&RNG4,&RNG5);
+
+  std::cout<<GridLogMessage <<"======================"<<std::endl;
+  std::cout<<GridLogMessage <<"ZMobiusFermion test"<<std::endl;
+  std::cout<<GridLogMessage <<"======================"<<std::endl;
+  for(int s=0;s<Ls;s++) omegasrev[s]=conjugate(omegas[Ls-1-s]);
+  //  for(int s=0;s<Ls;s++) omegasrev[s]=omegas[Ls-1-s];
+  ZMobiusFermionR ZDmob(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,omegas,b,c);
+  ZMobiusFermionR ZDmobrev(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,omegasrev,b,c);
+
+  TestConserved<ZMobiusFermionR>(ZDmob,ZDmobrev,Umu,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,&RNG4,&RNG5);
+
+  Grid_finalize();
+}
+
+
+
+template<class Action> 
+void  TestConserved(Action & Ddwf, 
+		    Action & Ddwfrev, 
+		    LatticeGaugeField &Umu,
+		    GridCartesian         * FGrid,	       GridRedBlackCartesian * FrbGrid,
+		    GridCartesian         * UGrid,	       GridRedBlackCartesian * UrbGrid,
+		    RealD mass, RealD M5,
+		    GridParallelRNG *RNG4,
+		    GridParallelRNG *RNG5)
+{
+  int Ls=Ddwf.Ls;
+
+  LatticePropagator phys_src(UGrid); 
+
+  std::vector<LatticeColourMatrix> U(4,UGrid);
+  
+  LatticePropagator seqsrc(FGrid); 
+  LatticePropagator prop5(FGrid); 
+  LatticePropagator prop5rev(FGrid); 
+  LatticePropagator prop4(UGrid); 
+  LatticePropagator Axial_mu(UGrid); 
+  LatticePropagator Vector_mu(UGrid); 
+  LatticeComplex    PA (UGrid); 
+  LatticeComplex    SV (UGrid); 
+  LatticeComplex    VV (UGrid); 
+  LatticeComplex    PJ5q(UGrid);
+  LatticeComplex    PP (UGrid);
+  LatticePropagator seqprop(UGrid); 
+
+  SpinColourMatrix kronecker; kronecker=1.0;
+  Coordinate coor({0,0,0,0});
+  phys_src=Zero();
+  pokeSite(kronecker,phys_src,coor);
+  
+  MdagMLinearOperator<Action,LatticeFermion> HermOp(Ddwf);
+  MdagMLinearOperator<Action,LatticeFermion> HermOprev(Ddwfrev);
+  ConjugateGradient<LatticeFermion> CG(1.0e-16,100000);
+  for(int s=0;s<Nd;s++){
+    for(int c=0;c<Nc;c++){
+      LatticeFermion src4  (UGrid); 
+      PropToFerm<Action>(src4,phys_src,s,c);
+
+      LatticeFermion src5  (FGrid); 
+      Ddwf.ImportPhysicalFermionSource(src4,src5);
+
+      LatticeFermion result5(FGrid); result5=Zero();
+
+      // CGNE
+      LatticeFermion Mdagsrc5  (FGrid); 
+      Ddwf.Mdag(src5,Mdagsrc5);
+      CG(HermOp,Mdagsrc5,result5);
+      FermToProp<Action>(prop5,result5,s,c);
+
+      LatticeFermion result4(UGrid);
+      Ddwf.ExportPhysicalFermionSolution(result5,result4);
+      FermToProp<Action>(prop4,result4,s,c);
+
+      Ddwfrev.ImportPhysicalFermionSource(src4,src5);
+      Ddwfrev.Mdag(src5,Mdagsrc5);
+      CG(HermOprev,Mdagsrc5,result5);
+      FermToProp<Action>(prop5rev,result5,s,c);
+    }
+  }
+
+#if 1
+  auto curr = Current::Axial;
+  const int mu_J=Nd-1;
+#else
+  auto curr = Current::Vector;
+  const int mu_J=0;
+#endif
+  const int t_J=0;
+
+  LatticeComplex    ph (UGrid); ph=1.0;
+
+  Ddwf.SeqConservedCurrent(prop5,
+			   seqsrc,
+			   phys_src,
+			   curr,
+			   mu_J,
+			   t_J,
+			   t_J,// whole lattice
+			   ph);
+
+  for(int s=0;s<Nd;s++){
+    for(int c=0;c<Nc;c++){
+
+      LatticeFermion src5  (FGrid); 
+      PropToFerm<Action>(src5,seqsrc,s,c);
+
+      LatticeFermion result5(FGrid); result5=Zero();
+
+      // CGNE
+      LatticeFermion Mdagsrc5  (FGrid); 
+      Ddwf.Mdag(src5,Mdagsrc5);
+      CG(HermOp,Mdagsrc5,result5);
+
+      LatticeFermion result4(UGrid);
+      Ddwf.ExportPhysicalFermionSolution(result5,result4);
+      FermToProp<Action>(seqprop,result4,s,c);
+    }
+  }
+  
+  Gamma g5(Gamma::Algebra::Gamma5);
+  Gamma gT(Gamma::Algebra::GammaT);
+
+  std::vector<TComplex> sumPA;
+  std::vector<TComplex> sumSV;
+  std::vector<TComplex> sumVV;
+  std::vector<TComplex> sumPP;
+  std::vector<TComplex> sumPJ5q;
+  
+  Ddwf.ContractConservedCurrent(prop5rev,prop5,Axial_mu,phys_src,Current::Axial,Tdir);
+  Ddwf.ContractConservedCurrent(prop5rev,prop5,Vector_mu,phys_src,Current::Vector,Tdir);
+  Ddwf.ContractJ5q(prop5,PJ5q);
+  
+  PA       = trace(g5*Axial_mu);
+  SV       = trace(Vector_mu);
+  VV       = trace(gT*Vector_mu);
+  PP       = trace(adj(prop4)*prop4);
+  
+  // Spatial sum
+  sliceSum(PA,sumPA,Tdir);
+  sliceSum(SV,sumSV,Tdir);
+  sliceSum(VV,sumVV,Tdir);
+  sliceSum(PP,sumPP,Tdir);
+  sliceSum(PJ5q,sumPJ5q,Tdir);
+
+  int Nt=sumPA.size();
+  for(int t=0;t<Nt;t++){
+    std::cout <<" SV "<<real(TensorRemove(sumSV[t]));
+    std::cout <<" VV "<<real(TensorRemove(sumVV[t]))<<std::endl;
+  }
+  for(int t=0;t<Nt;t++){
+    std::cout <<" PAc "<<real(TensorRemove(sumPA[t]));
+    std::cout <<" PJ5q "<<real(TensorRemove(sumPJ5q[t]));
+    std::cout <<" Ward Identity defect " <<real(TensorRemove(sumPA[t]-sumPA[(t-1+Nt)%Nt] - 2.0*(Ddwf.mass*sumPP[t] + sumPJ5q[t]) ))<<"\n";
+  }
+  
+  ///////////////////////////////
+  // 3pt vs 2pt check
+  ///////////////////////////////
+  { 
+    Gamma::Algebra        gA = (curr == Current::Axial) ? Gamma::Algebra::Gamma5 : Gamma::Algebra::Identity;
+    Gamma                 g(gA);
+
+    LatticePropagator cur(UGrid); 
+    LatticePropagator tmp(UGrid); 
+    LatticeComplex c(UGrid);
+    SpinColourMatrix qSite;
+    peekSite(qSite, seqprop, coor);
+
+    Complex               test_S, test_V, check_S, check_V;
+
+    std::vector<TComplex> check_buf;
+
+    test_S = trace(qSite*g);
+    test_V = trace(qSite*g*Gamma::gmu[mu_J]);
+
+    Ddwf.ContractConservedCurrent(prop5rev,prop5,cur,phys_src,curr,mu_J);
+    
+    c = trace(cur*g);
+    sliceSum(c, check_buf, Tp);
+    check_S = TensorRemove(check_buf[t_J]);
+
+    auto gmu=Gamma::gmu[mu_J];
+    c = trace(cur*g*gmu);
+    sliceSum(c, check_buf, Tp);
+    check_V = TensorRemove(check_buf[t_J]);
+
+    
+    std::cout<<GridLogMessage << std::setprecision(14)<<"Test S  = " << abs(test_S)   << std::endl;
+    std::cout<<GridLogMessage << "Test V  = " << abs(test_V) << std::endl;
+    std::cout<<GridLogMessage << "Check S = " << abs(check_S) << std::endl;
+    std::cout<<GridLogMessage << "Check V = " << abs(check_V) << std::endl;
+
+    // Check difference = 0
+    check_S = check_S - test_S;
+    check_V = check_V - test_V;
+
+    std::cout<<GridLogMessage << "Consistency check for sequential conserved " <<std::endl;
+    std::cout<<GridLogMessage << "Diff S  = " << abs(check_S) << std::endl;
+    std::cout<<GridLogMessage << "Diff V  = " << abs(check_V) << std::endl;
+  }
+
+}
+
+      /*
+#if 0
+template<class Action> 
+void  TestConserved1(Action & Ddwf, Action & Ddwfrev, 
+		       LatticeGaugeField &Umu,
+		       GridCartesian         * FGrid,	       GridRedBlackCartesian * FrbGrid,
+		       GridCartesian         * UGrid,	       GridRedBlackCartesian * UrbGrid,
+		       RealD mass, RealD M5,
+		       GridParallelRNG *RNG4,
+		       GridParallelRNG *RNG5)
+{
+  int Ls=Ddwf.Ls;
+
+  LatticePropagator phys_src(UGrid); 
+
+  std::vector<LatticeColourMatrix> U(4,UGrid);
+  
+  LatticePropagator prop5(FGrid); 
+  LatticePropagator prop5rev(FGrid); 
+  LatticePropagator prop4(UGrid); 
+  LatticePropagator Axial_mu(UGrid); 
+  LatticeComplex    PA (UGrid); 
+  LatticeComplex    PAxyz(UGrid); 
+  LatticeComplex    PJ5q(UGrid);
+  LatticeComplex    PP (UGrid);
+  std::vector<LatticePropagator> prop(Ls,UGrid);
+  std::vector<LatticePropagator> proprev(Ls,UGrid);
+
+  SpinColourMatrix kronecker; kronecker=1.0;
+  std::cout << kronecker << std::endl;
+  phys_src=Zero();
+  pokeSite(kronecker,phys_src,Coordinate({0,0,0,0}));
+  
+  MdagMLinearOperator<Action,LatticeFermion> HermOp(Ddwf);
+  MdagMLinearOperator<Action,LatticeFermion> HermOprev(Ddwfrev);
+  ConjugateGradient<LatticeFermion> CG(1.0e-12,10000);
+
+  for(int s=0;s<Nd;s++){
+    for(int c=0;c<Nc;c++){
+      LatticeFermion src4  (UGrid); 
+      PropToFerm<Action>(src4,phys_src,s,c);
+
+      LatticeFermion src5  (FGrid); 
+      Ddwf.ImportPhysicalFermionSource(src4,src5);
+
+      LatticeFermion result5(FGrid); result5=Zero();
+
+      // CGNE
+      LatticeFermion Mdagsrc5  (FGrid); 
+      Ddwf.Mdag(src5,Mdagsrc5);
+      CG(HermOp,Mdagsrc5,result5);
+      FermToProp<Action>(prop5,result5,s,c);
+
+      LatticeFermion result4(UGrid);
+      Ddwf.ExportPhysicalFermionSolution(result5,result4);
+      FermToProp<Action>(prop4,result4,s,c);
+
+      Ddwfrev.Mdag(src5,Mdagsrc5);
+      CG(HermOprev,Mdagsrc5,result5);
+      FermToProp<Action>(prop5rev,result5,s,c);
+    }
+  }
+
+  for(int s=0;s<Ls;s++){
+    ExtractSlice(prop[s], prop5, s , 0);
+    ExtractSlice(proprev[s], prop5rev, s , 0);
+  }
+
+  Gamma g5(Gamma::Algebra::Gamma5);
+  LatticeComplex    C(UGrid); 
+  std::vector<LatticeComplex> PAmu(Nd,UGrid); 
+  LatticePropagator p5d(UGrid); 
+  LatticePropagator us_p5d(UGrid); 
+  LatticePropagator gp5d(UGrid); 
+  LatticePropagator gus_p5d(UGrid); 
+
+#define Pp(Q) (0.5*(Q+g5*Q))
+#define Pm(Q) (0.5*(Q-g5*Q))
+#define Q_4d(Q) (Pm((Q)[0]) + Pp((Q)[Ls-1]))
+#define TopRowWithSource(Q) (phys_src + (1.0-mass)*Q_4d(Q))
+
+  std::vector<LatticePropagator> L_Q(Ls,UGrid); L_Q=proprev; // shorthand name
+  std::vector<LatticePropagator> R_Q(Ls,UGrid); R_Q=prop; // shorthand name
+
+  LatticePropagator L_TmLsGq0(UGrid); 
+  LatticePropagator L_TmLsTmp(UGrid);
+  LatticePropagator R_TmLsGq0(UGrid); 
+  LatticePropagator R_TmLsTmp(UGrid);
+  {
+    LatticePropagator TermA(UGrid);
+    LatticePropagator TermB(UGrid);
+    LatticePropagator TermC(UGrid);
+    LatticePropagator TermD(UGrid);
+    TermA = (Pp(Q_4d(L_Q)));
+    TermB = (Pm(Q_4d(L_Q)));
+    TermC = (Pm(TopRowWithSource(L_Q)));
+    TermD = (Pp(TopRowWithSource(L_Q)));
+
+    L_TmLsGq0 = (TermD - TermA + TermB);
+    L_TmLsTmp = (TermC - TermB + TermA);
+
+    TermA = (Pp(Q_4d(R_Q)));
+    TermB = (Pm(Q_4d(R_Q)));
+    TermC = (Pm(TopRowWithSource(R_Q)));
+    TermD = (Pp(TopRowWithSource(R_Q)));
+
+    R_TmLsGq0 = (TermD - TermA + TermB);
+    R_TmLsTmp = (TermC - TermB + TermA);
+  }
+
+  std::vector<LatticePropagator> R_TmLsGq(Ls,UGrid);
+  std::vector<LatticePropagator> L_TmLsGq(Ls,UGrid);
+  for(int s=0;s<Ls;s++){
+    R_TmLsGq[s] = (Pm((R_Q)[(s)]) + Pp((R_Q)[((s)-1+Ls)%Ls]));
+    L_TmLsGq[s] = (Pm((L_Q)[(s)]) + Pp((L_Q)[((s)-1+Ls)%Ls]));
+  }
+  
+  for(int mu=0;mu<Nd;mu++){
+    PAmu[mu]=Zero();
+    Gamma gmu=Gamma(Gmu[mu]);
+
+    for(int s=0;s<Ls;s++){
+
+      int sp = (s+1)%Ls;
+      int sr = Ls-1-s;
+      int srp= (sr+1)%Ls;
+
+      // Mobius parameters
+      auto b=Ddwf.bs[s];
+      auto c=Ddwf.cs[s];
+      assert(Ddwfrev.bs[sr]==Ddwf.bs[s]);
+      assert(Ddwfrev.cs[sr]==Ddwf.cs[s]);
+
+      LatticePropagator tmp(UGrid); 
+
+      if (s == 0) {
+	p5d    =(b*Pm(L_TmLsGq[Ls-1])+ c*Pp(L_TmLsGq[Ls-1]) + b*Pp(L_TmLsTmp)   + c*Pm(L_TmLsTmp     ));
+	tmp    =(b*Pm(R_TmLsGq0)     + c*Pp(R_TmLsGq0     ) + b*Pp(R_TmLsGq[1]) + c*Pm(R_TmLsGq[1]));
+	us_p5d = peekLorentz(Umu,mu)*Cshift(tmp,mu,1);
+      } else if (s == Ls-1) {
+	p5d    =(b*Pm(L_TmLsGq0)     + c*Pp(L_TmLsGq0     ) + b*Pp(L_TmLsGq[1]) + c*Pm(L_TmLsGq[1]));
+	tmp    =(b*Pm(R_TmLsGq[Ls-1])+ c*Pp(R_TmLsGq[Ls-1]) + b*Pp(R_TmLsTmp)   + c*Pm(R_TmLsTmp   ));
+	us_p5d = peekLorentz(Umu,mu)*Cshift(tmp,mu,1);
+      } else {
+	p5d    =(b*Pm(L_TmLsGq[sr]) + c*Pp(L_TmLsGq[sr])+ b*Pp(L_TmLsGq[srp])+ c*Pm(L_TmLsGq[srp]));
+	tmp    =(b*Pm(R_TmLsGq[s])  + c*Pp(R_TmLsGq[s]) + b*Pp(R_TmLsGq[sp ])+ c*Pm(R_TmLsGq[sp]));
+	us_p5d = peekLorentz(Umu,mu)*Cshift(tmp,mu,1);
+      }
+              
+      gp5d=g5*p5d;
+      gus_p5d=gmu*us_p5d;
+      auto bpc = 0.5/(b+c);
+      C = bpc*localInnerProduct(gp5d,gus_p5d);
+      C-= bpc*localInnerProduct(gp5d,us_p5d);
+
+      if (s == 0) {
+	p5d    =(b*Pm(R_TmLsGq0)     + c*Pp(R_TmLsGq0  )    + b*Pp(R_TmLsGq[1]) + c*Pm(R_TmLsGq[1]));
+	tmp    =(b*Pm(L_TmLsGq[Ls-1])+ c*Pp(L_TmLsGq[Ls-1]) + b*Pp(L_TmLsTmp)   + c*Pm(L_TmLsTmp  ));
+	us_p5d = peekLorentz(Umu,mu)*Cshift(tmp,mu,1);
+      } else if (s == Ls-1) {
+	p5d    =(b*Pm(R_TmLsGq[Ls-1])+ c*Pp(R_TmLsGq[Ls-1]) + b*Pp(R_TmLsTmp)   + c*Pm(R_TmLsTmp  ));
+	tmp    =(b*Pm(L_TmLsGq0)     + c*Pp(L_TmLsGq0  )    + b*Pp(L_TmLsGq[1]) + c*Pm(L_TmLsGq[1]));
+	us_p5d = peekLorentz(Umu,mu)*Cshift(tmp,mu,1);
+      } else {
+	p5d    =(b*Pm(R_TmLsGq[s])  + c*Pp(R_TmLsGq[s])  + b*Pp(R_TmLsGq[sp ])+ c*Pm(R_TmLsGq[sp]));
+	tmp    =(b*Pm(L_TmLsGq[sr]) + c*Pp(L_TmLsGq[sr]) + b*Pp(L_TmLsGq[srp])+ c*Pm(L_TmLsGq[srp]));
+	us_p5d = peekLorentz(Umu,mu)*Cshift(tmp,mu,1);
+      }
+
+      gp5d=gmu*p5d;
+      gus_p5d=g5*us_p5d;
+
+      bpc = 0.5/(b+c);
+      C+= bpc*localInnerProduct(gus_p5d,gp5d);
+      C+= bpc*localInnerProduct(gus_p5d,p5d);
+
+      if (s < Ls/2) PAmu[mu] -= C;
+      else          PAmu[mu] += C;
+              
+    }
+  }
+
+  std::cout << "done "<<std::endl;
+  LatticePropagator psi(UGrid);
+  psi = (prop[Ls/2-1]+g5*prop[Ls/2-1] +prop[Ls/2]  -g5*prop[Ls/2]   )*0.5;
+  PJ5q=localInnerProduct(psi,psi);
+  std::cout << " J5qref "<<norm2(PJ5q)<<std::endl;
+
+  std::cout << " DmuAmu "<<std::endl;
+  LatticeComplex Defect(UGrid);
+  Defect = Zero();
+  for(int mu=0;mu<Nd;mu++) {
+    Defect = Defect + PAmu[mu]-Cshift(PAmu[mu],mu,-1);
+  }
+  Ddwf.ContractConservedCurrent(prop5rev,prop5,Axial_mu,phys_src,Current::Axial,Tdir);
+  PA       = trace(g5*Axial_mu);
+  PP       = trace(adj(prop4)*prop4);
+
+  Defect = Defect - 2.0*Ddwf.mass* PP;
+  Defect = Defect - 2.0*PJ5q;
+
+  std::vector<TComplex> sumPAref;
+  std::vector<TComplex> sumPA;
+  std::vector<TComplex> sumPP;
+  std::vector<TComplex> sumPJ5qref;
+  std::vector<TComplex> sumPJ5q;
+  std::vector<TComplex> sumDefect;
+
+  // Spatial sum
+  sliceSum(PAmu[Tdir],sumPAref,Tdir);
+  sliceSum(PA,sumPA,Tdir);
+  sliceSum(PJ5q,sumPJ5q,Tdir);
+  sliceSum(PP,sumPP,Tdir);
+  sliceSum(Defect,sumDefect,Tdir);
+  
+  Ddwf.ContractJ5q(prop5,PJ5q);
+  sliceSum(PJ5q,sumPJ5qref,Tdir);
+
+  int Nt=sumPA.size();
+  for(int t=0;t<Nt;t++){
+    std::cout <<t<<" PAc reference "<<real(TensorRemove(sumPAref[t]));
+    std::cout    <<" PAc action    "<<real(TensorRemove(sumPA[t]));
+    std::cout    <<" PJ5q ref      "<<real(TensorRemove(sumPJ5qref[t]));
+    std::cout    <<" PJ5q action   "<<real(TensorRemove(sumPJ5q[t]));
+    std::cout <<"WTI defects "<<real(TensorRemove(sumPAref[t]-sumPAref[(t-1+Nt)%Nt] - 2.0*(Ddwf.mass*sumPP[t] + sumPJ5q[t]) ))<<",";
+    std::cout <<real(TensorRemove(sumPA[t]-sumPA[(t-1+Nt)%Nt] - 2.0*(Ddwf.mass*sumPP[t] + sumPJ5q[t]) ))<<"\n";
+  }
+}
+#endif
+      // Verify solution with independent true residual
+      LatticeGaugeField Umu5d(FGrid); 
+      std::vector<LatticeColourMatrix> U(4,FGrid);
+      {
+        autoView( Umu5d_v , Umu5d, CpuWrite);
+        autoView( Umu_v   , Umu  , CpuRead);
+	for(int ss=0;ss<Umu.Grid()->oSites();ss++){
+	  for(int s=0;s<Ls;s++){
+	    Umu5d_v[Ls*ss+s] = Umu_v[ss];
+	  }
+	}
+      }
+      for(int mu=0;mu<Nd;mu++){
+	U[mu] = PeekIndex<LorentzIndex>(Umu5d,mu);
+      }
+      LatticeFermion ref(FGrid);
+      LatticeFermion tmp(FGrid);
+      ref = Zero();
+      for(int mu=0;mu<Nd;mu++){
+        tmp = U[mu]*Cshift(result5,mu+1,1);
+	ref=ref + tmp - Gamma(Gmu[mu])*tmp;
+
+	tmp =adj(U[mu])*result5;
+	tmp =Cshift(tmp,mu+1,-1);
+	ref=ref + tmp + Gamma(Gmu[mu])*tmp;
+      }
+      ref = -0.5*ref;
+      // Dperp
+      {
+	RealD diag = 5.0 - Ddwf.M5;
+	mass = Ddwf.mass;
+	autoView( psi,result5,CpuRead);
+	autoView( chi,tmp,   CpuWrite);
+	thread_for(sss,UGrid->oSites(),{
+	  uint64_t ss= sss*Ls;
+	  typedef vSpinColourVector spinor;
+	  spinor tmp1, tmp2;
+	  for(int s=0;s<Ls;s++){
+	    uint64_t idx_u = ss+((s+1)%Ls);
+	    uint64_t idx_l = ss+((s+Ls-1)%Ls);
+	    spProj5m(tmp1,psi(idx_u));
+	    spProj5p(tmp2,psi(idx_l));
+	    double pu = (s==(Ls-1)) ? mass: -1.0;
+	    double pl = (s==0) ? mass: -1.0;
+	    chi[ss+s]=diag*psi(ss+s)+pu*tmp1+pl*tmp2;
+	  }
+ 	});
+      }
+      ref = ref + tmp;
+      ref = ref - src5;
+      std::cout << "residual "<< norm2(ref)<< std::endl;
+      std::cout << "src      "<< norm2(src5)<< std::endl;
+      std::cout << "result   "<< norm2(result5)<< std::endl;
+      */
diff --git a/tests/forces/Test_contfrac_force.cc b/tests/forces/Test_contfrac_force.cc
index 4eeb8c27..cb30faad 100644
--- a/tests/forces/Test_contfrac_force.cc
+++ b/tests/forces/Test_contfrac_force.cc
@@ -98,9 +98,9 @@ int main (int argc, char ** argv)
     PokeIndex<LorentzIndex>(mom,mommu,mu);
 
     // fourth order exponential approx
-    auto mom_v = mom.View();
-    auto Uprime_v = Uprime.View();
-    auto U_v = U.View();
+    autoView( mom_v, mom, CpuRead);
+    autoView( U_v , U, CpuRead);
+    autoView(Uprime_v, Uprime, CpuWrite);
     thread_foreach( i,mom_v,{
       Uprime_v[i](mu) =
 	  U_v[i](mu)
diff --git a/tests/forces/Test_dwf_force.cc b/tests/forces/Test_dwf_force.cc
index 009f50b3..81a1b8c4 100644
--- a/tests/forces/Test_dwf_force.cc
+++ b/tests/forces/Test_dwf_force.cc
@@ -100,9 +100,9 @@ int main (int argc, char ** argv)
 
     // fourth order exponential approx
 
-    auto mom_v    = mom.View();
-    auto U_v      = U.View();
-    auto Uprime_v = Uprime.View();
+    autoView( mom_v, mom, CpuRead);
+    autoView( U_v , U, CpuRead);
+    autoView(Uprime_v, Uprime, CpuWrite);
 
     thread_foreach( i,mom_v,{
       Uprime_v[i](mu) =	  U_v[i](mu)
diff --git a/tests/forces/Test_dwf_force_eofa.cc b/tests/forces/Test_dwf_force_eofa.cc
index 670e7589..0b0ba346 100644
--- a/tests/forces/Test_dwf_force_eofa.cc
+++ b/tests/forces/Test_dwf_force_eofa.cc
@@ -110,9 +110,9 @@ int main (int argc, char** argv)
     PokeIndex<LorentzIndex>(mom, mommu, mu);
 
     // fourth order exponential approx
-    auto mom_v    = mom.View();
-    auto U_v      = U.View();
-    auto Uprime_v = Uprime.View();
+    autoView( mom_v, mom, CpuRead);
+    autoView( U_v , U, CpuRead);
+    autoView(Uprime_v, Uprime, CpuWrite);
 
     thread_foreach(i,mom_v,{
       Uprime_v[i](mu) =	  U_v[i](mu)
diff --git a/tests/forces/Test_dwf_gpforce.cc b/tests/forces/Test_dwf_gpforce.cc
index d762e22a..b39fdd14 100644
--- a/tests/forces/Test_dwf_gpforce.cc
+++ b/tests/forces/Test_dwf_gpforce.cc
@@ -119,9 +119,9 @@ int main (int argc, char ** argv)
     PokeIndex<LorentzIndex>(mom,mommu,mu);
 
     // fourth order exponential approx
-    auto mom_v    = mom.View();
-    auto U_v      = U.View();
-    auto Uprime_v = Uprime.View();
+    autoView( mom_v, mom, CpuRead);
+    autoView( U_v , U, CpuRead);
+    autoView(Uprime_v, Uprime, CpuWrite);
 
     thread_foreach(i,mom_v,{
       Uprime_v[i](mu) =	  U_v[i](mu)
diff --git a/tests/forces/Test_dwf_gpforce_eofa.cc b/tests/forces/Test_dwf_gpforce_eofa.cc
index 66ae9dcf..58258a5e 100644
--- a/tests/forces/Test_dwf_gpforce_eofa.cc
+++ b/tests/forces/Test_dwf_gpforce_eofa.cc
@@ -114,9 +114,9 @@ int main (int argc, char** argv)
     PokeIndex<LorentzIndex>(mom, mommu, mu);
 
     // fourth order exponential approx
-    auto mom_v    = mom.View();
-    auto U_v      = U.View();
-    auto Uprime_v = Uprime.View();
+    autoView( mom_v, mom, CpuRead);
+    autoView( U_v , U, CpuRead);
+    autoView(Uprime_v, Uprime, CpuWrite);
 
     thread_foreach(i,mom_v,{
       Uprime_v[i](mu) =	  U_v[i](mu)
diff --git a/tests/forces/Test_gp_plaq_force.cc b/tests/forces/Test_gp_plaq_force.cc
index c4e214bb..21f0b9d0 100644
--- a/tests/forces/Test_gp_plaq_force.cc
+++ b/tests/forces/Test_gp_plaq_force.cc
@@ -85,9 +85,9 @@ int main (int argc, char ** argv)
     PokeIndex<LorentzIndex>(mom,mommu,mu);
 
     // fourth order exponential approx
-    auto Uprime_v = Uprime.View();
-    auto U_v = U.View();
-    auto mom_v = mom.View();
+    autoView(Uprime_v, Uprime, CpuWrite);
+    autoView( U_v , U, CpuRead);
+    autoView( mom_v, mom, CpuRead);
     thread_foreach(i,mom_v,{ // exp(pmu dt) * Umu
       Uprime_v[i](mu) = U_v[i](mu) + mom_v[i](mu)*U_v[i](mu)*dt ;
     });
diff --git a/tests/forces/Test_gp_rect_force.cc b/tests/forces/Test_gp_rect_force.cc
index 2573af6a..bb4ea6de 100644
--- a/tests/forces/Test_gp_rect_force.cc
+++ b/tests/forces/Test_gp_rect_force.cc
@@ -87,9 +87,9 @@ int main (int argc, char ** argv)
     PokeIndex<LorentzIndex>(mom,mommu,mu);
 
     // fourth order exponential approx
-    auto mom_v   = mom.View();
-    auto Uprime_v= Uprime.View();
-    auto U_v     = U.View();
+    autoView( mom_v, mom, CpuRead);
+    autoView(Uprime_v, Uprime, CpuWrite);
+    autoView( U_v , U, CpuRead);
     thread_foreach(i,mom_v,{ // exp(pmu dt) * Umu
       Uprime_v[i](mu) = U_v[i](mu) + mom_v[i](mu)*U_v[i](mu)*dt ;
     });
diff --git a/tests/forces/Test_gpdwf_force.cc b/tests/forces/Test_gpdwf_force.cc
index 09a1dc4b..bdc332d9 100644
--- a/tests/forces/Test_gpdwf_force.cc
+++ b/tests/forces/Test_gpdwf_force.cc
@@ -105,9 +105,9 @@ int main (int argc, char ** argv)
     PokeIndex<LorentzIndex>(mom,mommu,mu);
 
     // fourth order exponential approx
-    auto U_v = U.View();
-    auto mom_v = mom.View();
-    auto Uprime_v = Uprime.View();
+    autoView( U_v , U, CpuRead);
+    autoView( mom_v, mom, CpuRead);
+    autoView(Uprime_v, Uprime, CpuWrite);
     thread_foreach(i,mom_v,{
       Uprime_v[i](mu) = U_v[i](mu)
 	+ mom_v[i](mu)*U_v[i](mu)*dt 
diff --git a/tests/forces/Test_gpwilson_force.cc b/tests/forces/Test_gpwilson_force.cc
index cd30d898..1c85a5d9 100644
--- a/tests/forces/Test_gpwilson_force.cc
+++ b/tests/forces/Test_gpwilson_force.cc
@@ -99,9 +99,9 @@ int main (int argc, char ** argv)
     PokeIndex<LorentzIndex>(mom,mommu,mu);
 
     // fourth order exponential approx
-    auto mom_v    = mom.View();
-    auto U_v      = U.View();
-    auto Uprime_v = Uprime.View();
+    autoView( mom_v, mom, CpuRead);
+    autoView( U_v , U, CpuRead);
+    autoView(Uprime_v, Uprime, CpuWrite);
 
     thread_foreach(i,mom_v,{
       Uprime_v[i](mu) =	  U_v[i](mu)
diff --git a/tests/forces/Test_mobius_force.cc b/tests/forces/Test_mobius_force.cc
index a1c4e930..11e69652 100644
--- a/tests/forces/Test_mobius_force.cc
+++ b/tests/forces/Test_mobius_force.cc
@@ -101,9 +101,9 @@ int main (int argc, char ** argv)
     PokeIndex<LorentzIndex>(mom,mommu,mu);
 
     // fourth order exponential approx
-    auto U_v      = U.View();
-    auto mom_v    = mom.View();
-    auto Uprime_v = Uprime.View();
+    autoView( U_v , U, CpuRead);
+    autoView( mom_v, mom, CpuRead);
+    autoView(Uprime_v, Uprime, CpuWrite);
     thread_foreach(i,mom_v,{
       Uprime_v[i](mu) =  U_v[i](mu)
 	+ mom_v[i](mu)*U_v[i](mu)*dt 
diff --git a/tests/forces/Test_mobius_force_eofa.cc b/tests/forces/Test_mobius_force_eofa.cc
index f71e2d41..f85501fa 100644
--- a/tests/forces/Test_mobius_force_eofa.cc
+++ b/tests/forces/Test_mobius_force_eofa.cc
@@ -112,9 +112,9 @@ int main (int argc, char** argv)
     PokeIndex<LorentzIndex>(mom, mommu, mu);
 
     // fourth order exponential approx
-    auto mom_v    = mom.View();
-    auto U_v      = U.View();
-    auto Uprime_v = Uprime.View();
+    autoView( mom_v, mom, CpuRead);
+    autoView( U_v , U, CpuRead);
+    autoView(Uprime_v, Uprime, CpuWrite);
 
     thread_foreach(i,mom_v,{
       Uprime_v[i](mu) =	  U_v[i](mu)
diff --git a/tests/forces/Test_mobius_gpforce_eofa.cc b/tests/forces/Test_mobius_gpforce_eofa.cc
index 4975f36d..68163e63 100644
--- a/tests/forces/Test_mobius_gpforce_eofa.cc
+++ b/tests/forces/Test_mobius_gpforce_eofa.cc
@@ -115,9 +115,9 @@ int main (int argc, char** argv)
     SU3::GaussianFundamentalLieAlgebraMatrix(RNG4, mommu); // Traceless antihermitian momentum; gaussian in lie alg
 
     PokeIndex<LorentzIndex>(mom, mommu, mu);
-    auto U_v      = U.View();
-    auto mom_v    = mom.View();
-    auto Uprime_v = Uprime.View();
+    autoView( U_v , U, CpuRead);
+    autoView( mom_v, mom, CpuRead);
+    autoView(Uprime_v, Uprime, CpuWrite);
     // fourth order exponential approx
     thread_foreach( i, mom_v,{
       Uprime_v[i](mu) = U_v[i](mu) + mom_v[i](mu)*U_v[i](mu)*dt + mom_v[i](mu) *mom_v[i](mu) *U_v[i](mu)*(dt*dt/2.0)
diff --git a/tests/forces/Test_partfrac_force.cc b/tests/forces/Test_partfrac_force.cc
index 3ea2c6aa..17dce530 100644
--- a/tests/forces/Test_partfrac_force.cc
+++ b/tests/forces/Test_partfrac_force.cc
@@ -101,9 +101,9 @@ int main (int argc, char ** argv)
     PokeIndex<LorentzIndex>(mom,mommu,mu);
 
     // fourth order exponential approx
-    auto U_v = U.View();
-    auto mom_v = mom.View();
-    auto Uprime_v = Uprime.View();
+    autoView( U_v , U, CpuRead);
+    autoView( mom_v, mom, CpuRead);
+    autoView(Uprime_v, Uprime, CpuWrite);
     thread_foreach(i,mom_v,{
       Uprime_v[i](mu) =  U_v[i](mu)
 	+ mom_v[i](mu)*U_v[i](mu)*dt 
diff --git a/tests/forces/Test_rect_force.cc b/tests/forces/Test_rect_force.cc
index 9a78de24..ed72f2c0 100644
--- a/tests/forces/Test_rect_force.cc
+++ b/tests/forces/Test_rect_force.cc
@@ -87,9 +87,9 @@ int main (int argc, char ** argv)
     PokeIndex<LorentzIndex>(mom,mommu,mu);
 
     // fourth order exponential approx
-    auto Uprime_v = Uprime.View();
-    auto U_v      = U.View();
-    auto mom_v    = mom.View();
+    autoView(Uprime_v, Uprime, CpuWrite);
+    autoView( U_v , U, CpuRead);
+    autoView( mom_v, mom, CpuRead);
     thread_foreach(i,mom_v,{ // exp(pmu dt) * Umu
       Uprime_v[i](mu) = U_v[i](mu) + mom_v[i](mu)*U_v[i](mu)*dt ;
     });
diff --git a/tests/forces/Test_wilson_force.cc b/tests/forces/Test_wilson_force.cc
index 47f1516a..c8b3a7f4 100644
--- a/tests/forces/Test_wilson_force.cc
+++ b/tests/forces/Test_wilson_force.cc
@@ -105,9 +105,9 @@ int main (int argc, char ** argv)
     PokeIndex<LorentzIndex>(mom,mommu,mu);
 
     // fourth order exponential approx
-    auto U_v = U.View();
-    auto mom_v = mom.View();
-    auto Uprime_v = Uprime.View();
+    autoView( U_v , U, CpuRead);
+    autoView( mom_v, mom, CpuRead);
+    autoView(Uprime_v, Uprime, CpuWrite);
     thread_foreach( i,mom_v,{
       Uprime_v[i](mu)  =	  U_v[i](mu);
       Uprime_v[i](mu) += mom_v[i](mu)*U_v[i](mu)*dt ;
diff --git a/tests/forces/Test_wilsonclover_force.cc b/tests/forces/Test_wilsonclover_force.cc
index d9ace23c..f26f0ac9 100644
--- a/tests/forces/Test_wilsonclover_force.cc
+++ b/tests/forces/Test_wilsonclover_force.cc
@@ -105,9 +105,9 @@ int main(int argc, char **argv)
     Hmom -= real(sum(trace(mommu * mommu)));
     PokeIndex<LorentzIndex>(mom, mommu, mu);
 
-    auto Uprime_v = Uprime.View();
-    auto U_v      = U.View();
-    auto mom_v    = mom.View();
+    autoView(Uprime_v, Uprime, CpuWrite);
+    autoView( U_v , U, CpuRead);
+    autoView( mom_v, mom, CpuRead);
     thread_foreach(ss,mom_v,
     {
       Uprime_v[ss]._internal[mu] = ProjectOnGroup(Exponentiate(mom_v[ss]._internal[mu], dt, 12) * U_v[ss]._internal[mu]);
diff --git a/tests/forces/Test_zmobius_force.cc b/tests/forces/Test_zmobius_force.cc
index 2730885f..e24ae601 100644
--- a/tests/forces/Test_zmobius_force.cc
+++ b/tests/forces/Test_zmobius_force.cc
@@ -114,9 +114,9 @@ int main (int argc, char ** argv)
     PokeIndex<LorentzIndex>(mom,mommu,mu);
 
     // fourth order exponential approx
-    auto mom_v    = mom.View();
-    auto U_v      = U.View();
-    auto Uprime_v = Uprime.View();
+    autoView( mom_v, mom, CpuRead);
+    autoView( U_v , U, CpuRead);
+    autoView(Uprime_v, Uprime, CpuWrite);
 
     thread_foreach(i,mom_v,{
       Uprime_v[i](mu) =	  U_v[i](mu)
diff --git a/tests/hmc/Test_hmc_WilsonMixedRepresentationsFermionGauge.cc b/tests/hmc/Test_hmc_WilsonMixedRepresentationsFermionGauge.cc
index 6fa90f32..3b8cdda6 100644
--- a/tests/hmc/Test_hmc_WilsonMixedRepresentationsFermionGauge.cc
+++ b/tests/hmc/Test_hmc_WilsonMixedRepresentationsFermionGauge.cc
@@ -35,7 +35,7 @@ directory
 
 
 int main(int argc, char **argv) {
-#ifndef GRID_NVCC
+#ifndef GRID_CUDA
   using namespace Grid;
 
 
diff --git a/tests/hmc/Test_multishift_sqrt.cc b/tests/hmc/Test_multishift_sqrt.cc
index 834e6fc8..31697c12 100644
--- a/tests/hmc/Test_multishift_sqrt.cc
+++ b/tests/hmc/Test_multishift_sqrt.cc
@@ -31,7 +31,6 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 
 using namespace std;
 using namespace Grid;
- ;
 
 template<class Field> class DumbOperator  : public LinearOperatorBase<Field> {
 public:
@@ -57,6 +56,7 @@ public:
   // Support for coarsening to a multigrid
   void OpDiag (const Field &in, Field &out) {};
   void OpDir  (const Field &in, Field &out,int dir,int disp){};
+  void OpDirAll  (const Field &in, std::vector<Field> &out) {};
 
   void Op     (const Field &in, Field &out){
     out = scale * in;
@@ -104,7 +104,7 @@ int main (int argc, char ** argv)
 						       GridDefaultMpi());
 
   double     lo=0.001;
-  double     hi=1.0;
+  double     hi=20.0;
   int precision=64;
   int    degree=10;
   AlgRemez remez(lo,hi,precision);
diff --git a/tests/solver/Test_dwf_fpgcr.cc b/tests/solver/Test_dwf_fpgcr.cc
index 226bd933..156f678a 100644
--- a/tests/solver/Test_dwf_fpgcr.cc
+++ b/tests/solver/Test_dwf_fpgcr.cc
@@ -70,9 +70,6 @@ int main (int argc, char ** argv)
 
   SU3::HotConfiguration(RNG4,Umu);
 
-  TrivialPrecon<LatticeFermion> simple;
-
-  PrecGeneralisedConjugateResidual<LatticeFermion> PGCR(1.0e-6,10000,simple,4,160);
 
   ConjugateResidual<LatticeFermion> CR(1.0e-6,10000);
 
@@ -86,15 +83,19 @@ int main (int argc, char ** argv)
   std::cout<<GridLogMessage<<"* Solving with MdagM VPGCR "<<std::endl;
   std::cout<<GridLogMessage<<"*********************************************************"<<std::endl;
   MdagMLinearOperator<DomainWallFermionR,LatticeFermion> HermOp(Ddwf);
+  TrivialPrecon<LatticeFermion> simple;
+  PrecGeneralisedConjugateResidual<LatticeFermion> PGCR(1.0e-6,10000,HermOp,simple,4,160);
+
   result=Zero();
-  PGCR(HermOp,src,result);
+  PGCR(src,result);
 
   std::cout<<GridLogMessage<<"*********************************************************"<<std::endl;
   std::cout<<GridLogMessage<<"* Solving with g5-VPGCR "<<std::endl;
   std::cout<<GridLogMessage<<"*********************************************************"<<std::endl;
   Gamma5R5HermitianLinearOperator<DomainWallFermionR,LatticeFermion> g5HermOp(Ddwf);
+  PrecGeneralisedConjugateResidual<LatticeFermion> PGCR5(1.0e-6,10000,g5HermOp,simple,4,160);
   result=Zero();
-  PGCR(g5HermOp,src,result);
+  PGCR5(src,result);
 
   std::cout<<GridLogMessage<<"*********************************************************"<<std::endl;
   std::cout<<GridLogMessage<<"* Solving with MdagM-CR "<<std::endl;
diff --git a/tests/solver/Test_dwf_hdcr.cc b/tests/solver/Test_dwf_hdcr.cc
index 873530ff..8e083231 100644
--- a/tests/solver/Test_dwf_hdcr.cc
+++ b/tests/solver/Test_dwf_hdcr.cc
@@ -1,5 +1,3 @@
-
-
 /*************************************************************************************
 
     Grid physics library, www.github.com/paboyle/Grid 
@@ -302,8 +300,8 @@ int main (int argc, char ** argv)
     int nb=nbasisc/2;
     CoarseAggregates.CreateSubspaceChebyshev(CRNG,PosdefLdop,nb,12.0,0.02,500,100,100,0.0);
     for(int n=0;n<nb;n++){
-      auto subspace    = CoarseAggregates.subspace[n].View();
-      auto subspace_g5 = CoarseAggregates.subspace[n+nb].View();
+      autoView( subspace,   CoarseAggregates.subspace[n]   ,CpuRead);
+      autoView( subspace_g5,CoarseAggregates.subspace[n+nb],CpuWrite);
       for(int nn=0;nn<nb;nn++){
 	for(int site=0;site<Coarse5d->oSites();site++){
 	  subspace_g5[site](nn)   = subspace[site](nn);
diff --git a/tests/solver/Test_dwf_multigrid.cc b/tests/solver/Test_dwf_multigrid.cc
new file mode 100644
index 00000000..9e11c160
--- /dev/null
+++ b/tests/solver/Test_dwf_multigrid.cc
@@ -0,0 +1,594 @@
+/*************************************************************************************
+
+    Grid physics library, www.github.com/paboyle/Grid 
+
+    Source file: ./tests/Test_dwf_hdcr.cc
+
+    Copyright (C) 2015
+
+Author: Antonin Portelli <antonin.portelli@me.com>
+Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+Author: paboyle <paboyle@ph.ed.ac.uk>
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License along
+    with this program; if not, write to the Free Software Foundation, Inc.,
+    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+    See the full license in the file "LICENSE" in the top level distribution directory
+    *************************************************************************************/
+    /*  END LEGAL */
+#include <Grid/Grid.h>
+#include <Grid/algorithms/iterative/PrecGeneralisedConjugateResidual.h>
+#include <Grid/algorithms/iterative/PrecGeneralisedConjugateResidualNonHermitian.h>
+#include <Grid/algorithms/iterative/BiCGSTAB.h>
+
+using namespace std;
+using namespace Grid;
+/* Params
+ * Grid: 
+ * block1(4)
+ * block2(4)
+ * 
+ * Subspace
+ * * Fine  : Subspace(nbasis,hi,lo,order,first,step) -- 32, 60,0.02,500,100,100
+ * * Coarse: Subspace(nbasis,hi,lo,order,first,step) -- 32, 18,0.02,500,100,100
+
+ * Smoother:
+ * * Fine: Cheby(hi, lo, order)            --  60,0.5,10
+ * * Coarse: Cheby(hi, lo, order)          --  12,0.1,4
+
+ * Lanczos:
+ * CoarseCoarse IRL( Nk, Nm, Nstop, poly(lo,hi,order))   24,36,24,0.002,4.0,61 
+ */
+
+template<class Field> class SolverWrapper : public LinearFunction<Field> {
+private:
+  LinearOperatorBase<Field> & _Matrix;
+  OperatorFunction<Field> & _Solver;
+  LinearFunction<Field>   & _Guess;
+public:
+
+  /////////////////////////////////////////////////////
+  // Wrap the usual normal equations trick
+  /////////////////////////////////////////////////////
+  SolverWrapper(LinearOperatorBase<Field> &Matrix,
+	      OperatorFunction<Field> &Solver,
+	      LinearFunction<Field> &Guess) 
+   :  _Matrix(Matrix), _Solver(Solver), _Guess(Guess) {}; 
+
+  void operator() (const Field &in, Field &out){
+ 
+    _Guess(in,out);
+    _Solver(_Matrix,in,out);  // Mdag M out = Mdag in
+
+  }     
+};
+
+
+// Must use a non-hermitian solver
+template<class Matrix,class Field>
+class PVdagMLinearOperator : public LinearOperatorBase<Field> {
+  Matrix &_Mat;
+  Matrix &_PV;
+public:
+  PVdagMLinearOperator(Matrix &Mat,Matrix &PV): _Mat(Mat),_PV(PV){};
+
+  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){
+    Field tmp(in.Grid());
+    _Mat.M(in,tmp);
+    _PV.Mdag(tmp,out);
+  }
+  void AdjOp     (const Field &in, Field &out){
+    Field tmp(in.Grid());
+    _PV.M(tmp,out);
+    _Mat.Mdag(in,tmp);
+  }
+  void HermOpAndNorm(const Field &in, Field &out,RealD &n1,RealD &n2){
+    assert(0);
+  }
+  void HermOp(const Field &in, Field &out){
+    assert(0);
+  }
+};
+
+
+RealD InverseApproximation(RealD x){
+  return 1.0/x;
+}
+
+template<class Field,class Matrix> class ChebyshevSmoother : public LinearFunction<Field>
+{
+public:
+  typedef LinearOperatorBase<Field>                            FineOperator;
+  Matrix         & _SmootherMatrix;
+  FineOperator   & _SmootherOperator;
+  
+  Chebyshev<Field> Cheby;
+
+  ChebyshevSmoother(RealD _lo,RealD _hi,int _ord, FineOperator &SmootherOperator,Matrix &SmootherMatrix) :
+    _SmootherOperator(SmootherOperator),
+    _SmootherMatrix(SmootherMatrix),
+    Cheby(_lo,_hi,_ord,InverseApproximation)
+  {};
+
+  void operator() (const Field &in, Field &out) 
+  {
+    Field tmp(in.Grid());
+    MdagMLinearOperator<Matrix,Field>   MdagMOp(_SmootherMatrix); 
+    _SmootherOperator.AdjOp(in,tmp);
+    Cheby(MdagMOp,tmp,out);         
+  }
+};
+
+template<class Field,class Matrix> class MirsSmoother : public LinearFunction<Field>
+{
+public:
+  typedef LinearOperatorBase<Field>                            FineOperator;
+  Matrix         & SmootherMatrix;
+  FineOperator   & SmootherOperator;
+  RealD tol;
+  RealD shift;
+  int   maxit;
+
+  MirsSmoother(RealD _shift,RealD _tol,int _maxit,FineOperator &_SmootherOperator,Matrix &_SmootherMatrix) :
+    shift(_shift),tol(_tol),maxit(_maxit),
+    SmootherOperator(_SmootherOperator),
+    SmootherMatrix(_SmootherMatrix)
+  {};
+
+  void operator() (const Field &in, Field &out) 
+  {
+    ZeroGuesser<Field> Guess;
+    ConjugateGradient<Field>  CG(tol,maxit,false);
+ 
+    Field src(in.Grid());
+
+    ShiftedMdagMLinearOperator<SparseMatrixBase<Field>,Field> MdagMOp(SmootherMatrix,shift);
+    SmootherOperator.AdjOp(in,src);
+    Guess(src,out);
+    CG(MdagMOp,src,out); 
+  }
+};
+
+#define GridLogLevel std::cout << GridLogMessage <<std::string(level,'\t')<< " Level "<<level <<" "
+
+template<class Fobj,class CComplex,int nbasis, class CoarseSolver>
+class HDCRPreconditioner : public LinearFunction< Lattice<Fobj> > {
+public:
+
+  typedef Aggregation<Fobj,CComplex,nbasis> Aggregates;
+  typedef CoarsenedMatrix<Fobj,CComplex,nbasis> CoarseOperator;
+  typedef typename Aggregation<Fobj,CComplex,nbasis>::CoarseVector CoarseVector;
+  typedef typename Aggregation<Fobj,CComplex,nbasis>::CoarseMatrix CoarseMatrix;
+  typedef typename Aggregation<Fobj,CComplex,nbasis>::FineField    FineField;
+  typedef LinearOperatorBase<FineField>                            FineOperator;
+  typedef LinearFunction    <FineField>                            FineSmoother;
+
+  Aggregates     & _Aggregates;
+  FineOperator   & _FineOperator;
+  FineSmoother   & _Smoother;
+  CoarseSolver   & _CoarseSolve;
+
+  int    level;  void Level(int lv) {level = lv; };
+
+
+  HDCRPreconditioner(Aggregates &Agg,
+		     FineOperator &Fine,
+		     FineSmoother &Smoother,
+		     CoarseSolver &CoarseSolve_)
+    : _Aggregates(Agg),
+      _FineOperator(Fine),
+      _Smoother(Smoother),
+      _CoarseSolve(CoarseSolve_),
+      level(1)  {  }
+
+  virtual void operator()(const FineField &in, FineField & out) 
+  {
+    auto CoarseGrid = _Aggregates.CoarseGrid;
+    CoarseVector Csrc(CoarseGrid);
+    CoarseVector Csol(CoarseGrid);
+    FineField vec1(in.Grid());
+    FineField vec2(in.Grid());
+
+    double t;
+    // Fine Smoother
+    t=-usecond();
+    _Smoother(in,out);
+    t+=usecond();
+    GridLogLevel << "Smoother took "<< t/1000.0<< "ms" <<std::endl;
+
+    // Update the residual
+    _FineOperator.Op(out,vec1);  sub(vec1, in ,vec1);   
+
+    // Fine to Coarse 
+    t=-usecond();
+    _Aggregates.ProjectToSubspace  (Csrc,vec1);
+    t+=usecond();
+    GridLogLevel << "Project to coarse took "<< t/1000.0<< "ms" <<std::endl;
+
+    // Coarse correction
+    t=-usecond();
+    _CoarseSolve(Csrc,Csol);
+    t+=usecond();
+    GridLogLevel << "Coarse solve took "<< t/1000.0<< "ms" <<std::endl;
+
+    // Coarse to Fine
+    t=-usecond();
+    _Aggregates.PromoteFromSubspace(Csol,vec1); 
+    add(out,out,vec1);
+    t+=usecond();
+    GridLogLevel << "Promote to this level took "<< t/1000.0<< "ms" <<std::endl;
+
+    // Residual
+    _FineOperator.Op(out,vec1);  sub(vec1 ,in , vec1);  
+
+    // Fine Smoother
+    t=-usecond();
+    _Smoother(vec1,vec2);
+    t+=usecond();
+    GridLogLevel << "Smoother took "<< t/1000.0<< "ms" <<std::endl;
+
+    add( out,out,vec2);
+  }
+};
+
+/*
+template<class Fobj,class CComplex,int nbasis, class Guesser, class CoarseSolver>
+class MultiGridPreconditioner : public LinearFunction< Lattice<Fobj> > {
+public:
+
+  typedef Aggregation<Fobj,CComplex,nbasis> Aggregates;
+  typedef CoarsenedMatrix<Fobj,CComplex,nbasis> CoarseOperator;
+  typedef typename Aggregation<Fobj,CComplex,nbasis>::CoarseVector CoarseVector;
+  typedef typename Aggregation<Fobj,CComplex,nbasis>::CoarseMatrix CoarseMatrix;
+  typedef typename Aggregation<Fobj,CComplex,nbasis>::FineField    FineField;
+  typedef LinearOperatorBase<FineField>                            FineOperator;
+  typedef LinearFunction    <FineField>                            FineSmoother;
+
+  Aggregates     & _Aggregates;
+  CoarseOperator & _CoarseOperator;
+  FineOperator   & _FineOperator;
+  Guesser        & _Guess;
+  FineSmoother   & _Smoother;
+  CoarseSolver   & _CoarseSolve;
+
+  int    level;  void Level(int lv) {level = lv; };
+
+
+  MultiGridPreconditioner(Aggregates &Agg, CoarseOperator &Coarse, 
+			  FineOperator &Fine,
+			  FineSmoother &Smoother,
+			  Guesser &Guess_,
+			  CoarseSolver &CoarseSolve_)
+    : _Aggregates(Agg),
+      _CoarseOperator(Coarse),
+      _FineOperator(Fine),
+      _Smoother(Smoother),
+      _Guess(Guess_),
+      _CoarseSolve(CoarseSolve_),
+      level(1)  {  }
+
+  virtual void operator()(const FineField &in, FineField & out) 
+  {
+    CoarseVector Csrc(_CoarseOperator.Grid());
+    CoarseVector Csol(_CoarseOperator.Grid()); 
+    FineField vec1(in.Grid());
+    FineField vec2(in.Grid());
+
+    double t;
+    // Fine Smoother
+    t=-usecond();
+    _Smoother(in,out);
+    t+=usecond();
+    GridLogLevel << "Smoother took "<< t/1000.0<< "ms" <<std::endl;
+
+    // Update the residual
+    _FineOperator.Op(out,vec1);  sub(vec1, in ,vec1);   
+
+    // Fine to Coarse 
+    t=-usecond();
+    _Aggregates.ProjectToSubspace  (Csrc,vec1);
+    t+=usecond();
+    GridLogLevel << "Project to coarse took "<< t/1000.0<< "ms" <<std::endl;
+
+    // Coarse correction
+    t=-usecond();
+    _CoarseSolve(Csrc,Csol);
+    t+=usecond();
+    GridLogLevel << "Coarse solve took "<< t/1000.0<< "ms" <<std::endl;
+
+    // Coarse to Fine
+    t=-usecond();
+    _Aggregates.PromoteFromSubspace(Csol,vec1); 
+    add(out,out,vec1);
+    t+=usecond();
+    GridLogLevel << "Promote to this level took "<< t/1000.0<< "ms" <<std::endl;
+
+    // Residual
+    _FineOperator.Op(out,vec1);  sub(vec1 ,in , vec1);  
+
+    // Fine Smoother
+    t=-usecond();
+    _Smoother(vec1,vec2);
+    t+=usecond();
+    GridLogLevel << "Smoother took "<< t/1000.0<< "ms" <<std::endl;
+
+    add( out,out,vec2);
+  }
+};
+*/
+
+int main (int argc, char ** argv)
+{
+  Grid_init(&argc,&argv);
+
+  const int Ls=16;
+
+  GridCartesian         * UGrid   = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());
+  GridRedBlackCartesian * UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
+
+  GridCartesian         * FGrid   = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
+  GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid);
+
+  ///////////////////////////////////////////////////
+  // Construct a coarsened grid; utility for this?
+  ///////////////////////////////////////////////////
+  std::vector<int> block ({2,2,2,2});
+  std::vector<int> blockc ({2,2,2,2});
+  const int nbasis= 32;
+  const int nbasisc= 32;
+  auto clatt = GridDefaultLatt();
+  for(int d=0;d<clatt.size();d++){
+    clatt[d] = clatt[d]/block[d];
+  }
+  auto cclatt = clatt;
+  for(int d=0;d<clatt.size();d++){
+    cclatt[d] = clatt[d]/blockc[d];
+  }
+
+  GridCartesian *Coarse4d =  SpaceTimeGrid::makeFourDimGrid(clatt, GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());;
+  GridCartesian *Coarse5d =  SpaceTimeGrid::makeFiveDimGrid(1,Coarse4d);
+  GridCartesian *CoarseCoarse4d =  SpaceTimeGrid::makeFourDimGrid(cclatt, GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());;
+  GridCartesian *CoarseCoarse5d =  SpaceTimeGrid::makeFiveDimGrid(1,CoarseCoarse4d);
+
+  std::vector<int> seeds4({1,2,3,4});
+  std::vector<int> seeds5({5,6,7,8});
+  std::vector<int> cseeds({5,6,7,8});
+  GridParallelRNG          RNG5(FGrid);   RNG5.SeedFixedIntegers(seeds5);
+  GridParallelRNG          RNG4(UGrid);   RNG4.SeedFixedIntegers(seeds4);
+  GridParallelRNG          CRNG(Coarse5d);CRNG.SeedFixedIntegers(cseeds);
+  LatticeFermion    src(FGrid); gaussian(RNG5,src);// src=src+g5*src;
+  LatticeFermion result(FGrid); 
+  LatticeGaugeField Umu(UGrid); 
+
+  FieldMetaData header;
+  std::string file("./ckpoint_lat.4000");
+  NerscIO::readConfiguration(Umu,header,file);
+
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "Building g5R5 hermitian DWF operator" <<std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  RealD mass=0.001;
+  RealD M5=1.8;
+  DomainWallFermionR Ddwf(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
+  DomainWallFermionR Dpv (Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,1.0,M5);
+
+  typedef Aggregation<vSpinColourVector,vTComplex,nbasis>              Subspace;
+  typedef CoarsenedMatrix<vSpinColourVector,vTComplex,nbasis>          CoarseOperator;
+  typedef CoarseOperator::CoarseVector                                 CoarseVector;
+  typedef CoarseOperator::siteVector siteVector;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "Calling Aggregation class to build subspace" <<std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  MdagMLinearOperator<DomainWallFermionR,LatticeFermion> HermDefOp(Ddwf);
+
+  Subspace Aggregates(Coarse5d,FGrid,0);
+
+  assert ( (nbasis & 0x1)==0);
+  {
+    int nb=nbasis/2;
+    Aggregates.CreateSubspaceChebyshev(RNG5,HermDefOp,nb,60.0,0.02,500,100,100,0.0);
+    for(int n=0;n<nb;n++){
+      G5R5(Aggregates.subspace[n+nb],Aggregates.subspace[n]);
+    }
+    LatticeFermion A(FGrid);
+    LatticeFermion B(FGrid);
+    for(int n=0;n<nb;n++){
+      A = Aggregates.subspace[n];
+      B = Aggregates.subspace[n+nb];
+      Aggregates.subspace[n]   = A+B; // 1+G5 // eigen value of G5R5 is +1
+      Aggregates.subspace[n+nb]= A-B; // 1-G5 // eigen value of G5R5 is -1
+    }
+  }
+
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << " Will coarsen G5R5 M and G5R5 Mpv in G5R5 compatible way " <<std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  
+  typedef CoarsenedMatrix<vSpinColourVector,vTComplex,nbasis>    Level1Op;
+  typedef CoarsenedMatrix<siteVector,iScalar<vTComplex>,nbasisc> Level2Op;
+
+  Gamma5R5HermitianLinearOperator<DomainWallFermionR,LatticeFermion> HermIndefOp(Ddwf);
+  Gamma5R5HermitianLinearOperator<DomainWallFermionR,LatticeFermion> HermIndefOpPV(Dpv);
+
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "Building coarse representation of Indef operator" <<std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+
+  Level1Op LDOp(*Coarse5d,1);   LDOp.CoarsenOperator(FGrid,HermIndefOp,Aggregates);
+  Level1Op LDOpPV(*Coarse5d,1); LDOpPV.CoarsenOperator(FGrid,HermIndefOpPV,Aggregates);
+
+
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << " Testing fine and coarse solvers " <<std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+
+  CoarseVector c_src(Coarse5d); c_src=1.0;
+  CoarseVector c_res(Coarse5d);
+
+  LatticeFermion f_src(FGrid); f_src=1.0;
+  LatticeFermion f_res(FGrid);
+
+  LatticeFermion f_src_e(FrbGrid); f_src_e=1.0;
+  LatticeFermion f_res_e(FrbGrid);
+
+  RealD tol=1.0e-8;
+  int MaxIt = 10000;
+
+  BiCGSTAB<CoarseVector>                     CoarseBiCGSTAB(tol,MaxIt);
+  ConjugateGradient<CoarseVector>            CoarseCG(tol,MaxIt);
+  //  GeneralisedMinimalResidual<CoarseVector>   CoarseGMRES(tol,MaxIt,20);
+
+  BiCGSTAB<LatticeFermion>                   FineBiCGSTAB(tol,MaxIt);
+  ConjugateGradient<LatticeFermion>          FineCG(tol,MaxIt);
+  //  GeneralisedMinimalResidual<LatticeFermion> FineGMRES(tol,MaxIt,20);
+  
+  MdagMLinearOperator<DomainWallFermionR,LatticeFermion>    FineMdagM(Ddwf);     //  M^\dag M
+  PVdagMLinearOperator<DomainWallFermionR,LatticeFermion>   FinePVdagM(Ddwf,Dpv);//  M_{pv}^\dag M
+  SchurDiagMooeeOperator<DomainWallFermionR,LatticeFermion> FineDiagMooee(Ddwf); //  M_ee - Meo Moo^-1 Moe 
+  SchurDiagOneOperator<DomainWallFermionR,LatticeFermion>   FineDiagOne(Ddwf);   //  1 - M_ee^{-1} Meo Moo^{-1} Moe e
+
+   MdagMLinearOperator<Level1Op,CoarseVector> CoarseMdagM(LDOp);
+  PVdagMLinearOperator<Level1Op,CoarseVector> CoarsePVdagM(LDOp,LDOpPV);
+
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "Fine CG unprec "<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+
+  f_res=Zero();
+  FineCG(FineMdagM,f_src,f_res);
+
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "Fine CG prec DiagMooee "<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  
+  f_res_e=Zero();
+  FineCG(FineDiagMooee,f_src_e,f_res_e);
+
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "Fine CG prec DiagOne "<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+
+  f_res_e=Zero();
+  FineCG(FineDiagOne,f_src_e,f_res_e);
+
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "Fine BiCGSTAB unprec "<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+
+  f_res=Zero();
+  FineBiCGSTAB(FinePVdagM,f_src,f_res);
+
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "Coarse BiCGSTAB "<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  
+  c_res=Zero();
+  CoarseBiCGSTAB(CoarsePVdagM,c_src,c_res);
+
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "Coarse CG unprec "<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+
+  c_res=Zero();
+  CoarseCG(CoarseMdagM,c_src,c_res);
+
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << " Running Coarse grid Lanczos "<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+
+  Chebyshev<CoarseVector>      IRLCheby(0.03,12.0,71);  // 1 iter
+  FunctionHermOp<CoarseVector> IRLOpCheby(IRLCheby,CoarseMdagM);
+  PlainHermOp<CoarseVector>    IRLOp    (CoarseMdagM);
+  int Nk=64;
+  int Nm=128;
+  int Nstop=Nk;
+  ImplicitlyRestartedLanczos<CoarseVector> IRL(IRLOpCheby,IRLOp,Nstop,Nk,Nm,1.0e-3,20);
+
+  int Nconv;
+  std::vector<RealD>            eval(Nm);
+  std::vector<CoarseVector>     evec(Nm,Coarse5d);
+  IRL.calc(eval,evec,c_src,Nconv);
+
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << " Running Coarse grid deflated solver              "<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+
+  DeflatedGuesser<CoarseVector> DeflCoarseGuesser(evec,eval);
+  NormalEquations<CoarseVector> DeflCoarseCGNE   (LDOp,CoarseCG,DeflCoarseGuesser);
+  c_res=Zero();
+  DeflCoarseCGNE(c_src,c_res);
+
+
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << " Running HDCR                                     "<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+
+  ConjugateGradient<CoarseVector>  CoarseMgridCG(0.001,1000);     
+  ChebyshevSmoother<LatticeFermion,DomainWallFermionR> FineSmoother(0.5,60.0,10,HermIndefOp,Ddwf);
+
+  typedef HDCRPreconditioner<vSpinColourVector,  vTComplex,nbasis, NormalEquations<CoarseVector> >   TwoLevelHDCR;
+  TwoLevelHDCR TwoLevelPrecon(Aggregates,
+			      HermIndefOp,
+			      FineSmoother,
+			      DeflCoarseCGNE);
+  TwoLevelPrecon.Level(1);
+  //  PrecGeneralisedConjugateResidual<LatticeFermion> l1PGCR(1.0e-8,100,HermIndefOp,TwoLevelPrecon,16,16);
+  PrecGeneralisedConjugateResidualNonHermitian<LatticeFermion> l1PGCR(1.0e-8,100,HermIndefOp,TwoLevelPrecon,16,16);
+  l1PGCR.Level(1);
+
+  f_res=Zero();
+
+  CoarseCG.Tolerance=0.02;
+  l1PGCR(f_src,f_res);
+
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << " Running Multigrid                                "<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+
+  BiCGSTAB<CoarseVector>    CoarseMgridBiCGSTAB(0.01,1000);     
+  BiCGSTAB<LatticeFermion>  FineMgridBiCGSTAB(0.0,24);
+  ZeroGuesser<CoarseVector> CoarseZeroGuesser;
+  ZeroGuesser<LatticeFermion> FineZeroGuesser;
+
+  SolverWrapper<LatticeFermion> FineBiCGSmoother(  FinePVdagM,  FineMgridBiCGSTAB,  FineZeroGuesser);
+  SolverWrapper<CoarseVector> CoarsePVdagMSolver(CoarsePVdagM,CoarseMgridBiCGSTAB,CoarseZeroGuesser);
+  typedef HDCRPreconditioner<vSpinColourVector, vTComplex,nbasis, SolverWrapper<CoarseVector> >  TwoLevelMG;
+
+  TwoLevelMG _TwoLevelMG(Aggregates,
+			 FinePVdagM,
+			 FineBiCGSmoother,
+			 CoarsePVdagMSolver);
+  _TwoLevelMG.Level(1);
+
+  PrecGeneralisedConjugateResidualNonHermitian<LatticeFermion> pvPGCR(1.0e-8,100,FinePVdagM,_TwoLevelMG,16,16);
+  pvPGCR.Level(1);
+
+  f_res=Zero();
+  pvPGCR(f_src,f_res);
+
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "Done "<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  Grid_finalize();
+  
+}
diff --git a/tests/solver/Test_hw_multigrid.cc b/tests/solver/Test_hw_multigrid.cc
new file mode 100644
index 00000000..b728faa7
--- /dev/null
+++ b/tests/solver/Test_hw_multigrid.cc
@@ -0,0 +1,375 @@
+/*************************************************************************************
+
+    Grid physics library, www.github.com/paboyle/Grid 
+
+    Source file: ./tests/Test_dwf_hdcr.cc
+
+    Copyright (C) 2015
+
+Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License along
+    with this program; if not, write to the Free Software Foundation, Inc.,
+    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+    See the full license in the file "LICENSE" in the top level distribution directory
+    *************************************************************************************/
+    /*  END LEGAL */
+#include <Grid/Grid.h>
+#include <Grid/algorithms/iterative/PrecGeneralisedConjugateResidual.h>
+//#include <Grid/algorithms/iterative/PrecGeneralisedConjugateResidualNonHermitian.h>
+#include <Grid/algorithms/iterative/BiCGSTAB.h>
+
+using namespace std;
+using namespace Grid;
+/* Params
+ * Grid: 
+ * block1(4)
+ * block2(4)
+ * 
+ * Subspace
+ * * Fine  : Subspace(nbasis,hi,lo,order,first,step) -- 32, 60,0.02,500,100,100
+ * * Coarse: Subspace(nbasis,hi,lo,order,first,step) -- 32, 18,0.02,500,100,100
+
+ * Smoother:
+ * * Fine: Cheby(hi, lo, order)            --  60,0.5,10
+ * * Coarse: Cheby(hi, lo, order)          --  12,0.1,4
+
+ * Lanczos:
+ * CoarseCoarse IRL( Nk, Nm, Nstop, poly(lo,hi,order))   24,36,24,0.002,4.0,61 
+ */
+
+template<class Field> class SolverWrapper : public LinearFunction<Field> {
+private:
+  LinearOperatorBase<Field> & _Matrix;
+  OperatorFunction<Field> & _Solver;
+  LinearFunction<Field>   & _Guess;
+public:
+
+  /////////////////////////////////////////////////////
+  // Wrap the usual normal equations trick
+  /////////////////////////////////////////////////////
+  SolverWrapper(LinearOperatorBase<Field> &Matrix,
+	      OperatorFunction<Field> &Solver,
+	      LinearFunction<Field> &Guess) 
+   :  _Matrix(Matrix), _Solver(Solver), _Guess(Guess) {}; 
+
+  void operator() (const Field &in, Field &out){
+ 
+    _Guess(in,out);
+    _Solver(_Matrix,in,out);  // Mdag M out = Mdag in
+
+  }     
+};
+
+
+// Must use a non-hermitian solver
+template<class Matrix,class Field>
+class PVdagMLinearOperator : public LinearOperatorBase<Field> {
+  Matrix &_Mat;
+  Matrix &_PV;
+public:
+  PVdagMLinearOperator(Matrix &Mat,Matrix &PV): _Mat(Mat),_PV(PV){};
+
+  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){
+    Field tmp(in.Grid());
+    _Mat.M(in,tmp);
+    _PV.Mdag(tmp,out);
+  }
+  void AdjOp     (const Field &in, Field &out){
+    Field tmp(in.Grid());
+    _PV.M(tmp,out);
+    _Mat.Mdag(in,tmp);
+  }
+  void HermOpAndNorm(const Field &in, Field &out,RealD &n1,RealD &n2){
+    assert(0);
+  }
+  void HermOp(const Field &in, Field &out){
+    assert(0);
+  }
+};
+
+
+RealD InverseApproximation(RealD x){
+  return 1.0/x;
+}
+
+template<class Field,class Matrix> class ChebyshevSmoother : public LinearFunction<Field>
+{
+public:
+  typedef LinearOperatorBase<Field>                            FineOperator;
+  Matrix         & _SmootherMatrix;
+  FineOperator   & _SmootherOperator;
+  
+  Chebyshev<Field> Cheby;
+
+  ChebyshevSmoother(RealD _lo,RealD _hi,int _ord, FineOperator &SmootherOperator,Matrix &SmootherMatrix) :
+    _SmootherOperator(SmootherOperator),
+    _SmootherMatrix(SmootherMatrix),
+    Cheby(_lo,_hi,_ord,InverseApproximation)
+  {};
+
+  void operator() (const Field &in, Field &out) 
+  {
+    Field tmp(in.Grid());
+    MdagMLinearOperator<Matrix,Field>   MdagMOp(_SmootherMatrix); 
+    _SmootherOperator.AdjOp(in,tmp);
+    Cheby(MdagMOp,tmp,out);         
+  }
+};
+
+template<class Field,class Matrix> class MirsSmoother : public LinearFunction<Field>
+{
+public:
+  typedef LinearOperatorBase<Field>                            FineOperator;
+  Matrix         & SmootherMatrix;
+  FineOperator   & SmootherOperator;
+  RealD tol;
+  RealD shift;
+  int   maxit;
+
+  MirsSmoother(RealD _shift,RealD _tol,int _maxit,FineOperator &_SmootherOperator,Matrix &_SmootherMatrix) :
+    shift(_shift),tol(_tol),maxit(_maxit),
+    SmootherOperator(_SmootherOperator),
+    SmootherMatrix(_SmootherMatrix)
+  {};
+
+  void operator() (const Field &in, Field &out) 
+  {
+    ZeroGuesser<Field> Guess;
+    ConjugateGradient<Field>  CG(tol,maxit,false);
+ 
+    Field src(in.Grid());
+
+    ShiftedMdagMLinearOperator<SparseMatrixBase<Field>,Field> MdagMOp(SmootherMatrix,shift);
+    SmootherOperator.AdjOp(in,src);
+    Guess(src,out);
+    CG(MdagMOp,src,out); 
+  }
+};
+
+#define GridLogLevel std::cout << GridLogMessage <<std::string(level,'\t')<< " Level "<<level <<" "
+
+template<class Fobj,class CComplex,int nbasis, class CoarseSolver>
+class HDCRPreconditioner : public LinearFunction< Lattice<Fobj> > {
+public:
+
+  typedef Aggregation<Fobj,CComplex,nbasis> Aggregates;
+  typedef CoarsenedMatrix<Fobj,CComplex,nbasis> CoarseOperator;
+  typedef typename Aggregation<Fobj,CComplex,nbasis>::CoarseVector CoarseVector;
+  typedef typename Aggregation<Fobj,CComplex,nbasis>::CoarseMatrix CoarseMatrix;
+  typedef typename Aggregation<Fobj,CComplex,nbasis>::FineField    FineField;
+  typedef LinearOperatorBase<FineField>                            FineOperator;
+  typedef LinearFunction    <FineField>                            FineSmoother;
+
+  Aggregates     & _Aggregates;
+  FineOperator   & _FineOperator;
+  FineSmoother   & _Smoother;
+  CoarseSolver   & _CoarseSolve;
+
+  int    level;  void Level(int lv) {level = lv; };
+
+
+  HDCRPreconditioner(Aggregates &Agg,
+		     FineOperator &Fine,
+		     FineSmoother &Smoother,
+		     CoarseSolver &CoarseSolve_)
+    : _Aggregates(Agg),
+      _FineOperator(Fine),
+      _Smoother(Smoother),
+      _CoarseSolve(CoarseSolve_),
+      level(1)  {  }
+
+  virtual void operator()(const FineField &in, FineField & out) 
+  {
+    auto CoarseGrid = _Aggregates.CoarseGrid;
+    CoarseVector Csrc(CoarseGrid);
+    CoarseVector Csol(CoarseGrid);
+    FineField vec1(in.Grid());
+    FineField vec2(in.Grid());
+
+    double t;
+    // Fine Smoother
+    t=-usecond();
+    _Smoother(in,out);
+    t+=usecond();
+    GridLogLevel << "Smoother took "<< t/1000.0<< "ms" <<std::endl;
+
+    // Update the residual
+    _FineOperator.Op(out,vec1);  sub(vec1, in ,vec1);   
+
+    // Fine to Coarse 
+    t=-usecond();
+    _Aggregates.ProjectToSubspace  (Csrc,vec1);
+    t+=usecond();
+    GridLogLevel << "Project to coarse took "<< t/1000.0<< "ms" <<std::endl;
+
+    // Coarse correction
+    t=-usecond();
+    _CoarseSolve(Csrc,Csol);
+    t+=usecond();
+    GridLogLevel << "Coarse solve took "<< t/1000.0<< "ms" <<std::endl;
+
+    // Coarse to Fine
+    t=-usecond();
+    _Aggregates.PromoteFromSubspace(Csol,vec1); 
+    add(out,out,vec1);
+    t+=usecond();
+    GridLogLevel << "Promote to this level took "<< t/1000.0<< "ms" <<std::endl;
+
+    // Residual
+    _FineOperator.Op(out,vec1);  sub(vec1 ,in , vec1);  
+
+    // Fine Smoother
+    t=-usecond();
+    _Smoother(vec1,vec2);
+    t+=usecond();
+    GridLogLevel << "Smoother took "<< t/1000.0<< "ms" <<std::endl;
+
+    add( out,out,vec2);
+  }
+};
+
+int main (int argc, char ** argv)
+{
+  Grid_init(&argc,&argv);
+
+  const int Ls=16;
+
+  GridCartesian         * UGrid   = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());
+  GridRedBlackCartesian * UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
+
+  GridCartesian         * FGrid   = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
+  GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid);
+
+  ///////////////////////////////////////////////////
+  // Construct a coarsened grid; utility for this?
+  ///////////////////////////////////////////////////
+  std::vector<int> block ({2,2,2,2});
+  const int nbasis= 8;
+
+  auto clatt = GridDefaultLatt();
+  for(int d=0;d<clatt.size();d++){
+    clatt[d] = clatt[d]/block[d];
+  }
+
+  GridCartesian *Coarse4d =  SpaceTimeGrid::makeFourDimGrid(clatt, GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());;
+  GridCartesian *Coarse5d =  SpaceTimeGrid::makeFiveDimGrid(Ls,Coarse4d);
+
+  std::vector<int> seeds({1,2,3,4});
+  GridParallelRNG          RNG5(FGrid);   RNG5.SeedFixedIntegers(seeds);
+  GridParallelRNG          RNG4(UGrid);   RNG4.SeedFixedIntegers(seeds);
+  GridParallelRNG          CRNG(Coarse5d);CRNG.SeedFixedIntegers(seeds);
+
+  LatticeGaugeField Umu(UGrid); 
+  FieldMetaData header;
+  std::string file("./ckpoint_lat.4000");
+  NerscIO::readConfiguration(Umu,header,file);
+
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "Building g5R5 hermitian DWF operator" <<std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  RealD mass=0.001;
+  RealD M5=1.8;
+  WilsonFermionR    Dw(Umu,*UGrid,*UrbGrid,-M5);
+  DomainWallFermionR Ddwf(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
+  DomainWallFermionR Dpv (Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,1.0,M5);
+
+  typedef Aggregation<vSpinColourVector,vTComplex,nbasis>              Subspace;
+  typedef CoarsenedMatrix<vSpinColourVector,vTComplex,nbasis>          CoarseOperator;
+  typedef CoarseOperator::CoarseVector                                 CoarseVector;
+  typedef CoarseOperator::siteVector siteVector;
+
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "Calling Aggregation class to build subspace" <<std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  MdagMLinearOperator<WilsonFermionR,LatticeFermion> SubspaceOp(Dw);
+
+  Subspace Aggregates4D(Coarse4d,UGrid,0);
+  Subspace Aggregates5D(Coarse5d,FGrid,0);
+
+  assert ( (nbasis & 0x1)==0);
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << " 4D subspace build                                " <<std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  int nb=nbasis/2;
+  Gamma g5(Gamma::Algebra::Gamma5);
+  Aggregates4D.CreateSubspaceChebyshev(RNG4,SubspaceOp,nb,60.0,0.02,500,100,100,0.0);
+  for(int n=0;n<nb;n++){
+    Aggregates4D.subspace[n+nb]= Aggregates4D.subspace[n] - g5 * Aggregates4D.subspace[n];
+    Aggregates4D.subspace[n]   = Aggregates4D.subspace[n] + g5 * Aggregates4D.subspace[n];
+  }
+
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << " Promote to 5D basis                              " <<std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  for(int n=0;n<nbasis;n++){
+    for(int s=0;s<Ls;s++){
+      InsertSlice(Aggregates4D.subspace[n],Aggregates5D.subspace[n],s,0);
+    }
+  }
+
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << " Coarsen the operator                          " <<std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  typedef CoarsenedMatrix<vSpinColourVector,vTComplex,nbasis>    Level1Op;
+
+  NonHermitianLinearOperator<DomainWallFermionR,LatticeFermion>  LinOpDwf(Ddwf);
+
+  Level1Op LDOp  (*Coarse5d,0);   
+  
+  std::cout<<GridLogMessage << " Callinig Coarsen the operator                          " <<std::endl;
+  LDOp.CoarsenOperator(FGrid,LinOpDwf,Aggregates5D);
+
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "Coarse CG unprec "<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+
+  CoarseVector c_src(Coarse5d); c_src=1.0;
+  CoarseVector c_res(Coarse5d);
+
+  LatticeFermion f_src(FGrid); f_src=1.0;
+  LatticeFermion f_res(FGrid);
+
+  RealD tol=1.0e-8;
+  int MaxIt = 10000;
+
+  MdagMLinearOperator<Level1Op,CoarseVector> CoarseMdagM(LDOp);
+  BiCGSTAB<CoarseVector>                     CoarseBiCGSTAB(tol,MaxIt);
+  ConjugateGradient<CoarseVector>            CoarseCG(tol,MaxIt);
+
+  c_res=Zero();
+  CoarseCG(CoarseMdagM,c_src,c_res);
+
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << " Solve                                            " <<std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  
+  LatticeFermion    src(FGrid); gaussian(RNG5,src);
+  LatticeFermion result(FGrid); 
+
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "Done "<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  Grid_finalize();
+  
+}
diff --git a/tests/solver/Test_staggered_block_cg_prec.cc b/tests/solver/Test_staggered_block_cg_prec.cc
index 95156939..2499fc8a 100644
--- a/tests/solver/Test_staggered_block_cg_prec.cc
+++ b/tests/solver/Test_staggered_block_cg_prec.cc
@@ -128,9 +128,7 @@ int main (int argc, char ** argv)
     std::cout<<GridLogMessage << "usec    =   "<< (t2-t1)<<std::endl;
     std::cout<<GridLogMessage << "flops   =   "<< flops<<std::endl;
     std::cout<<GridLogMessage << "mflop/s =   "<< flops/(t2-t1)<<std::endl;
-    HermOp4d.Report();
   }
-  Ds4d.Report();
   std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
 
 
@@ -148,9 +146,7 @@ int main (int argc, char ** argv)
     std::cout<<GridLogMessage << "usec    =   "<< (t2-t1)<<std::endl;
     std::cout<<GridLogMessage << "flops   =   "<< flops<<std::endl;
     std::cout<<GridLogMessage << "mflop/s =   "<< flops/(t2-t1)<<std::endl;
-    HermOp.Report();
   }
-  Ds.Report();
   std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
 
   std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
@@ -167,10 +163,8 @@ int main (int argc, char ** argv)
     std::cout<<GridLogMessage << "usec    =   "<< (t2-t1)<<std::endl;
     std::cout<<GridLogMessage << "flops   =   "<< flops<<std::endl;
     std::cout<<GridLogMessage << "mflop/s =   "<< flops/(t2-t1)<<std::endl;
-    HermOp.Report();
   }
 
-  Ds.Report();
   std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
 
   std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
@@ -187,9 +181,7 @@ int main (int argc, char ** argv)
     std::cout<<GridLogMessage << "usec    =   "<< (t2-t1)<<std::endl;
     std::cout<<GridLogMessage << "flops   =   "<< flops<<std::endl;
     std::cout<<GridLogMessage << "mflop/s =   "<< flops/(t2-t1)<<std::endl;
-    HermOp.Report();
   }
-  Ds.Report();
   std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
 
   std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
@@ -206,9 +198,7 @@ int main (int argc, char ** argv)
     std::cout<<GridLogMessage << "usec    =   "<< (t2-t1)<<std::endl;
     std::cout<<GridLogMessage << "flops   =   "<< flops<<std::endl;
     std::cout<<GridLogMessage << "mflop/s =   "<< flops/(t2-t1)<<std::endl;
-    HermOp.Report();
   }
-  Ds.Report();
   std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
 
   std::cout << GridLogMessage << "****************************************************************** "<<std::endl;
@@ -232,7 +222,6 @@ int main (int argc, char ** argv)
     std::cout<<GridLogMessage << "usec    =   "<< (t2-t1)<<std::endl;
     std::cout<<GridLogMessage << "flops   =   "<< flops<<std::endl;
     std::cout<<GridLogMessage << "mflop/s =   "<< flops/(t2-t1)<<std::endl;
-    //    HermOp4d.Report();
   }
 
 
diff --git a/tests/solver/Test_zMADWF_prec.cc b/tests/solver/Test_zMADWF_prec.cc
index cbcdd055..13b4c3b2 100644
--- a/tests/solver/Test_zMADWF_prec.cc
+++ b/tests/solver/Test_zMADWF_prec.cc
@@ -220,7 +220,7 @@ void run(const TestParams &params){
 
   GridStopWatch CGTimer;
   
-  typename RunParamsOuter::HermOpType<MobiusFermionD> HermOpEO_outer(D_outer);
+  typename RunParamsOuter::template HermOpType<MobiusFermionD> HermOpEO_outer(D_outer);
 
   CGTimer.Start();
   CG_outer(HermOpEO_outer, src_o_outer, result_o_outer);