Merge branch 'rmhmc_fix' into feature/rmhmc

Grid/Grid_Eigen_Dense.h

@@ -42,7 +42,7 @@
 #ifdef __NVCC__REDEFINE__
 #pragma pop_macro("__CUDACC__")
 #pragma pop_macro("__NVCC__")
-#pragma pop_macro("__CUDA_ARCH__")
+#pragma pop_macro("GRID_SIMT")
 #pragma pop
 #endif
 

Grid/algorithms/CoarsenedMatrix.h

@@ -49,11 +49,13 @@ inline void blockMaskedInnerProduct(Lattice<CComplex> &CoarseInner,
   Lattice<dotp> fine_inner_msk(fine);
 
   // Multiply could be fused with innerProduct
+  // Single block sum kernel could do both masks.
   fine_inner = localInnerProduct(fineX,fineY);
   mult(fine_inner_msk, fine_inner,FineMask);
   blockSum(CoarseInner,fine_inner_msk);
 }
 
+
 class Geometry {
 public:
   int npoint;
@@ -78,12 +80,8 @@ public:
     }
     directions   [2*_d]=0;
     displacements[2*_d]=0;
-
-    std::cout <<GridLogMessage << "Geometry "<<std::endl;
-    for(int p=0;p<npoint;p++){
-      std::cout <<GridLogMessage << "point " <<p<<" dir "<<directions[p]<<" delta " <<displacements[p]<<std::endl;
-    }
   }
+
 };
   
 template<class Fobj,class CComplex,int nbasis>
@@ -104,8 +102,8 @@ public:
   Aggregation(GridBase *_CoarseGrid,GridBase *_FineGrid,int _checkerboard) : 
     CoarseGrid(_CoarseGrid),
     FineGrid(_FineGrid),
-    checkerboard(_checkerboard),
-    subspace(nbasis,_FineGrid)
+    subspace(nbasis,_FineGrid),
+    checkerboard(_checkerboard)
   {
   };
   
@@ -287,8 +285,6 @@ public:
   ///////////////////////
   GridBase * Grid(void)         { return _grid; };   // this is all the linalg routines need to know
 
-  virtual std::vector<int> Directions(void)   { return geom.directions; };
-  virtual std::vector<int> Displacements(void){ return geom.displacements; };
   void M (const CoarseVector &in, CoarseVector &out)
   {
     conformable(_grid,in.Grid());
@@ -312,9 +308,6 @@ public:
 
     int osites=Grid()->oSites();
 
-    autoView(st,Stencil,AcceleratorRead);
-    siteVector *CBp=Stencil.CommBuf();
-
     accelerator_for(sss, Grid()->oSites()*nbasis, Nsimd, {
       int ss = sss/nbasis;
       int b  = sss%nbasis;
@@ -325,12 +318,12 @@ public:
 
       for(int point=0;point<geom.npoint;point++){
 
-	SE=st.GetEntry(ptype,point,ss);
+	SE=Stencil.GetEntry(ptype,point,ss);
 	  
 	if(SE->_is_local) { 
 	  nbr = coalescedReadPermute(in_v[SE->_offset],ptype,SE->_permute);
 	} else {
-	  nbr = coalescedRead(CBp[SE->_offset]);
+	  nbr = coalescedRead(Stencil.CommBuf()[SE->_offset]);
 	}
 	acceleratorSynchronise();
 
@@ -339,7 +332,7 @@ public:
 	}
       }
       coalescedWrite(out_v[ss](b),res);
-    });
+      });
 
     for(int p=0;p<geom.npoint;p++) AcceleratorViewContainer[p].ViewClose();
   };
@@ -416,23 +409,38 @@ public:
       MdirCalc(in,out[p],p);
     }
   };
-  void Mdir(const CoarseVector &in, CoarseVector &out, int dir, int disp)
-  {
+  void Mdir(const CoarseVector &in, CoarseVector &out, int dir, int disp){
+
     this->MdirComms(in);
 
     int ndim = in.Grid()->Nd();
 
-    int point=-1;
-    for(int p=0;p<geom.npoint;p++){
-      if( (dir==geom.directions[p])&&(disp==geom.displacements[p])) point=p;
-    }
-    assert(point!=-1);// Must find
+    //////////////
+    // 4D action like wilson
+    // 0+ => 0 
+    // 0- => 1
+    // 1+ => 2 
+    // 1- => 3
+    // etc..
+    //////////////
+    // 5D action like DWF
+    // 1+ => 0 
+    // 1- => 1
+    // 2+ => 2 
+    // 2- => 3
+    // etc..
+    auto point = [dir, disp, ndim](){
+      if(dir == 0 and disp == 0)
+	return 8;
+      else if ( ndim==4 ) { 
+	return (4 * dir + 1 - disp) / 2;
+      } else { 
+	return (4 * (dir-1) + 1 - disp) / 2;
+      }
+    }();
 
-    std::cout <<GridLogMessage << "Mdir point "<<point<<" dir "<<dir<<" disp "<<disp  <<std::endl;
-    for(int p=0;p<geom.npoint;p++){
-      std::cout <<GridLogMessage << "point " <<p<<" dir "<<geom.directions[p]<<" delta " <<geom.displacements[p]<<std::endl;
-    }
     MdirCalc(in,out,point);
+
   };
 
   void Mdiag(const CoarseVector &in, CoarseVector &out)
@@ -448,58 +456,10 @@ public:
     geom(CoarseGrid._ndimension),
     hermitian(hermitian_),
     Stencil(&CoarseGrid,geom.npoint,Even,geom.directions,geom.displacements,0),
-    A(geom.npoint,&CoarseGrid)
+      A(geom.npoint,&CoarseGrid)
   {
   };
 
-  void Test(Aggregation<Fobj,CComplex,nbasis> &_Aggregates,GridBase *FineGrid,LinearOperatorBase<Lattice<Fobj> > &linop)
-  {
-    typedef Lattice<Fobj> FineField;
-    CoarseVector Cin(_grid);
-    CoarseVector Cout(_grid);
-    CoarseVector CFout(_grid);
-
-    FineField Fin(FineGrid);
-    FineField Fout(FineGrid);
-
-
-    std::vector<int> seeds({1,2,3,4,5});
-    GridParallelRNG RNG(_grid);  RNG.SeedFixedIntegers(seeds);
-    gaussian(RNG,Cin);
-
-    _Aggregates.PromoteFromSubspace(Cin,Fin);
-    _Aggregates.ProjectToSubspace(Cin,Fin);
-
-    std::cout << GridLogMessage<< "************  "<<std::endl;
-    std::cout << GridLogMessage<< " Testing M  "<<std::endl;
-    std::cout << GridLogMessage<< "************  "<<std::endl;
-    // Coarse operator
-    this->M(Cin,Cout);
-    // Fine projected operator
-    _Aggregates.PromoteFromSubspace(Cin,Fin);
-    linop.Op(Fin,Fout);
-    _Aggregates.ProjectToSubspace(CFout,Fout);
-
-    CFout = CFout-Cout;
-    RealD diff = norm2(CFout);
-    std::cout << GridLogMessage<< " diff  "<<diff<<std::endl;
-    assert(diff<1.0e-5);
-
-    std::cout << GridLogMessage<< "************  "<<std::endl;
-    std::cout << GridLogMessage<< " Testing Mdag  "<<std::endl;
-    std::cout << GridLogMessage<< "************  "<<std::endl;
-    // Coarse operator
-    Mdag(Cin,Cout);
-    // Fine operator
-    linop.AdjOp(Fin,Fout);
-    _Aggregates.ProjectToSubspace(CFout,Fout);
-
-    CFout = CFout-Cout;
-    diff = norm2(CFout);
-    std::cout << GridLogMessage<< " diff  "<<diff<<std::endl; 
-    assert(diff<1.0e-5);
-  }
-
   void CoarsenOperator(GridBase *FineGrid,LinearOperatorBase<Lattice<Fobj> > &linop,
 		       Aggregation<Fobj,CComplex,nbasis> & Subspace)
   {
@@ -536,19 +496,8 @@ public:
 
     CoarseScalar InnerProd(Grid()); 
 
-    std::cout << GridLogMessage<< "CoarsenMatrix Orthog " << std::endl;
     // Orthogonalise the subblocks over the basis
     blockOrthogonalise(InnerProd,Subspace.subspace);
-    std::cout << GridLogMessage<< "CoarsenMatrix Orthog done " << std::endl;
-
-    auto OpDirections    = linop.Directions();
-    auto OpDisplacements = linop.Displacements();
-
-    std::cout<<" Coarsening an operator with "<< OpDirections.size()<<" terms "<<std::endl;
-    for(int p=0;p<OpDirections.size();p++) {
-      assert(OpDirections[p]==geom.directions[p]);
-      assert(OpDisplacements[p]==geom.displacements[p]);
-    }
 
     // Compute the matrix elements of linop between this orthonormal
     // set of vectors.
@@ -584,27 +533,13 @@ public:
     evenmask = where(mod(bcb,2)==(Integer)0,one,zero);
     oddmask  = one-evenmask;
 
-    /*
-    {
-      phi=Subspace.subspace[0];
-      linop.OpDirAll(phi,Mphi_p);
-      for(int p=0;p<geom.npoint-1;p++){
-	int dir=geom.directions[p];
-	int disp=geom.displacements[p];
-	linop.OpDir(phi,Mphi,dir,disp);
-	Mphi=Mphi-Mphi_p[p];
-	std::cout << GridLogMessage <<" Direction mapping check " <<norm2(Mphi)<<std::endl;
-      }
-    }
-*/
     assert(self_stencil!=-1);
-    int lhermitian=hermitian;
 
     for(int i=0;i<nbasis;i++){
 
       phi=Subspace.subspace[i];
 
-      std::cout << GridLogMessage<< "CoarsenMatrix vector "<<i << std::endl;
+      //      std::cout << GridLogMessage<< "CoarsenMatrix vector "<<i << std::endl;
       linop.OpDirAll(phi,Mphi_p);
       linop.OpDiag  (phi,Mphi_p[geom.npoint-1]);
 
@@ -615,7 +550,7 @@ public:
 	int dir   = geom.directions[p];
 	int disp  = geom.displacements[p];
 
-	if ( (disp==-1) || (!lhermitian ) ) {
+	if ( (disp==-1) || (!hermitian ) ) {
 
 	  ////////////////////////////////////////////////////////////////////////
 	  // Pick out contributions coming from this cell and neighbour cell
@@ -633,23 +568,11 @@ public:
 	    autoView( A_self  , A[self_stencil], AcceleratorWrite);
 
 	    accelerator_for(ss, Grid()->oSites(), Fobj::Nsimd(),{ coalescedWrite(A_p[ss](j,i),oZProj_v(ss)); });
-	    if ( lhermitian && (disp==-1) ) {
-	      for(int pp=0;pp<geom.npoint;pp++){// Find the opposite link and set <j|A|i> = <i|A|j>*
-		int dirp   = geom.directions[pp];
-		int dispp  = geom.displacements[pp];
-		if ( (dirp==dir) && (dispp==1) ){
-		  auto sft = conjugate(Cshift(oZProj,dir,1));
-		  autoView( sft_v    ,  sft  , AcceleratorWrite);
-		  autoView( A_pp     ,  A[pp], AcceleratorWrite);
-		  accelerator_for(ss, Grid()->oSites(), Fobj::Nsimd(),{ coalescedWrite(A_pp[ss](i,j),sft_v(ss)); });
-		}
-	      }
-	    }
+
 	  }
 	}
       }
 
-      std::cout << GridLogMessage<< "CoarsenMatrix Diag "<<std::endl;
       ///////////////////////////////////////////
       // Faster alternate self coupling.. use hermiticity to save 2x
       ///////////////////////////////////////////
@@ -681,35 +604,31 @@ public:
 
       }
     }
-
-    MemoryManager::PrintBytes();
-
-    // Auto self test
-    Test( Subspace,FineGrid,linop);
-
-#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<< "\n"<<A[p] << std::endl;
+    if(hermitian) {
+      std::cout << GridLogMessage << " ForceHermitian, new code "<<std::endl;
+      ForceHermitian();
     }
-    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++){
+      int dir   = geom.directions[p];
+      int disp  = geom.displacements[p];
+      if(disp==-1) {
+	// Find the opposite link
+	for(int pp=0;pp<geom.npoint;pp++){
+	  int dirp   = geom.directions[pp];
+	  int dispp  = geom.displacements[pp];
+	  if ( (dirp==dir) && (dispp==1) ){
+	    //	    Diff = adj(Cshift(A[p],dir,1)) - A[pp]; 
+	    //	    std::cout << GridLogMessage<<" Replacing stencil leg "<<pp<<" with leg "<<p<< " diff "<<norm2(Diff) <<std::endl;
+	    A[pp] = adj(Cshift(A[p],dir,1));
+	  }
+	}
+      }
+    }
+  }
 };
 
 NAMESPACE_END(Grid);

Grid/algorithms/LinearOperator.h

@@ -52,9 +52,6 @@ public:
   virtual void AdjOp  (const Field &in, Field &out) = 0; // Abstract base
   virtual void HermOpAndNorm(const Field &in, Field &out,RealD &n1,RealD &n2)=0;
   virtual void HermOp(const Field &in, Field &out)=0;
-
-  virtual std::vector<int> Directions(void)   =0;
-  virtual std::vector<int> Displacements(void)=0;
 };
 
 
@@ -79,9 +76,6 @@ class MdagMLinearOperator : public LinearOperatorBase<Field> {
 public:
   MdagMLinearOperator(Matrix &Mat): _Mat(Mat){};
 
-  virtual std::vector<int> Directions(void)   { return _Mat.Directions();};
-  virtual std::vector<int> Displacements(void){ return _Mat.Displacements();};
-
   // Support for coarsening to a multigrid
   void OpDiag (const Field &in, Field &out) {
     _Mat.Mdiag(in,out);
@@ -117,8 +111,6 @@ class ShiftedMdagMLinearOperator : public LinearOperatorBase<Field> {
   Matrix &_Mat;
   RealD _shift;
 public:
-  virtual std::vector<int> Directions(void)   { return _Mat.Directions();};
-  virtual std::vector<int> Displacements(void){ return _Mat.Displacements();};
   ShiftedMdagMLinearOperator(Matrix &Mat,RealD shift): _Mat(Mat), _shift(shift){};
   // Support for coarsening to a multigrid
   void OpDiag (const Field &in, Field &out) {
@@ -159,8 +151,6 @@ template<class Matrix,class Field>
 class HermitianLinearOperator : public LinearOperatorBase<Field> {
   Matrix &_Mat;
 public:
-  virtual std::vector<int> Directions(void)   { return _Mat.Directions();};
-  virtual std::vector<int> Displacements(void){ return _Mat.Displacements();};
   HermitianLinearOperator(Matrix &Mat): _Mat(Mat){};
   // Support for coarsening to a multigrid
   void OpDiag (const Field &in, Field &out) {
@@ -192,8 +182,6 @@ template<class Matrix,class Field>
 class NonHermitianLinearOperator : public LinearOperatorBase<Field> {
   Matrix &_Mat;
 public:
-  virtual std::vector<int> Directions(void)   { return _Mat.Directions();};
-  virtual std::vector<int> Displacements(void){ return _Mat.Displacements();};
   NonHermitianLinearOperator(Matrix &Mat): _Mat(Mat){};
   // Support for coarsening to a multigrid
   void OpDiag (const Field &in, Field &out) {
@@ -267,8 +255,6 @@ template<class Matrix,class Field>
   class SchurDiagMooeeOperator :  public SchurOperatorBase<Field> {
  public:
     Matrix &_Mat;
-    virtual std::vector<int> Directions(void)   { return _Mat.Directions();};
-    virtual std::vector<int> Displacements(void){ return _Mat.Displacements();};
     SchurDiagMooeeOperator (Matrix &Mat): _Mat(Mat){};
     virtual  void Mpc      (const Field &in, Field &out) {
       Field tmp(in.Grid());
@@ -295,8 +281,6 @@ template<class Matrix,class Field>
  protected:
     Matrix &_Mat;
  public:
-    virtual std::vector<int> Directions(void)   { return _Mat.Directions();};
-    virtual std::vector<int> Displacements(void){ return _Mat.Displacements();};
     SchurDiagOneOperator (Matrix &Mat): _Mat(Mat){};
     
     virtual void Mpc      (const Field &in, Field &out) {
@@ -323,8 +307,6 @@ template<class Matrix,class Field>
  protected:
     Matrix &_Mat;
  public:
-    virtual std::vector<int> Directions(void)   { return _Mat.Directions();};
-    virtual std::vector<int> Displacements(void){ return _Mat.Displacements();};
     SchurDiagTwoOperator (Matrix &Mat): _Mat(Mat){};
     
     virtual void Mpc      (const Field &in, Field &out) {
@@ -390,8 +372,6 @@ class NonHermitianSchurDiagMooeeOperator :  public NonHermitianSchurOperatorBase
 {
  public:
   Matrix& _Mat;
-  virtual std::vector<int> Directions(void)   { return _Mat.Directions();};
-  virtual std::vector<int> Displacements(void){ return _Mat.Displacements();};
  NonHermitianSchurDiagMooeeOperator(Matrix& Mat): _Mat(Mat){};
   virtual void Mpc(const Field& in, Field& out) {
     Field tmp(in.Grid());
@@ -425,8 +405,6 @@ class NonHermitianSchurDiagOneOperator : public NonHermitianSchurOperatorBase<Fi
   Matrix &_Mat;
   
  public:
-  virtual std::vector<int> Directions(void)   { return _Mat.Directions();};
-  virtual std::vector<int> Displacements(void){ return _Mat.Displacements();};
   NonHermitianSchurDiagOneOperator (Matrix& Mat): _Mat(Mat){};
   virtual void Mpc(const Field& in, Field& out) {
     Field tmp(in.Grid());
@@ -457,8 +435,6 @@ class NonHermitianSchurDiagTwoOperator : public NonHermitianSchurOperatorBase<Fi
   Matrix& _Mat;
   
  public:
-  virtual std::vector<int> Directions(void)   { return _Mat.Directions();};
-  virtual std::vector<int> Displacements(void){ return _Mat.Displacements();};
  NonHermitianSchurDiagTwoOperator(Matrix& Mat): _Mat(Mat){};
 
   virtual void Mpc(const Field& in, Field& out) {
@@ -499,8 +475,6 @@ class SchurStaggeredOperator :  public SchurOperatorBase<Field> {
   Field tmp;
   RealD mass;
  public:
-  virtual std::vector<int> Directions(void)   { return _Mat.Directions();};
-  virtual std::vector<int> Displacements(void){ return _Mat.Displacements();};
   SchurStaggeredOperator (Matrix &Mat): _Mat(Mat), tmp(_Mat.RedBlackGrid()) 
   { 
     assert( _Mat.isTrivialEE() );

Grid/algorithms/SparseMatrix.h

@@ -48,8 +48,6 @@ public:
   virtual  void Mdiag    (const Field &in, Field &out)=0;
   virtual  void Mdir     (const Field &in, Field &out,int dir, int disp)=0;
   virtual  void MdirAll  (const Field &in, std::vector<Field> &out)=0;
-  virtual std::vector<int> Directions(void)   =0;
-  virtual std::vector<int> Displacements(void)=0;
 };
 
 /////////////////////////////////////////////////////////////////////////////////////////////
@@ -75,8 +73,6 @@ public:
   virtual  void MooeeDag    (const Field &in, Field &out)=0;
   virtual  void MooeeInvDag (const Field &in, Field &out)=0;
 
-  virtual std::vector<int> Directions(void)   =0;
-  virtual std::vector<int> Displacements(void)=0;
 };
 
 NAMESPACE_END(Grid);

Grid/algorithms/iterative/AdefGeneric.h

@@ -1,4 +1,4 @@
-   /*************************************************************************************
+    /*************************************************************************************
 
     Grid physics library, www.github.com/paboyle/Grid 
 
@@ -28,7 +28,6 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 #ifndef GRID_ALGORITHMS_ITERATIVE_GENERIC_PCG
 #define GRID_ALGORITHMS_ITERATIVE_GENERIC_PCG
 
-NAMESPACE_BEGIN(Grid);
   /*
    * Compared to Tang-2009:  P=Pleft. P^T = PRight Q=MssInv. 
    * Script A = SolverMatrix 
@@ -51,54 +50,53 @@ NAMESPACE_BEGIN(Grid);
    * Vout = x
    */
 
-
-template<class Field, class CoarseField, class Aggregates>
+// abstract base
+template<class Field, class CoarseField>
 class TwoLevelFlexiblePcg : public LinearFunction<Field>
 {
  public:
-
   int verbose;
-
   RealD   Tolerance;
   Integer MaxIterations;
-  const int mmax = 4;
-  GridBase *FineGrid;
-  GridBase *CoarseGrid;
+  const int mmax = 5;
+  GridBase *grid;
+  GridBase *coarsegrid;
 
-  LinearOperatorBase<Field>   &_Linop;
-  LinearFunction<Field>     &_Smoother;
-  LinearFunction<CoarseField> &_CoarseSolver;
-  Aggregates                  &_Aggregates;
+  LinearOperatorBase<Field>   *_Linop
+  OperatorFunction<Field>     *_Smoother,
+  LinearFunction<CoarseField> *_CoarseSolver;
+
+  // Need somthing that knows how to get from Coarse to fine and back again
   
   // more most opertor functions
   TwoLevelFlexiblePcg(RealD tol,
-		      Integer maxit,
-		      LinearOperatorBase<Field> *Linop,
-		      LinearFunction<Field>   *Smoother,
-		      LinearFunction<CoarseField> *CoarseSolver,
-		      Aggregates *AggP
-		      ) : 
-  Tolerance(tol), 
-    MaxIterations(maxit),
-    _Linop(*Linop),
-    _Smoother(*Smoother),
-    _CoarseSolver(*CoarseSolver),
-    _Aggregates(*AggP)
+		     Integer maxit,
+		     LinearOperatorBase<Field> *Linop,
+		     LinearOperatorBase<Field> *SmootherLinop,
+		     OperatorFunction<Field>   *Smoother,
+		     OperatorFunction<CoarseField>  CoarseLinop
+		     ) : 
+      Tolerance(tol), 
+      MaxIterations(maxit),
+      _Linop(Linop),
+      _PreconditionerLinop(PrecLinop),
+      _Preconditioner(Preconditioner)
   { 
-    CoarseGrid=_Aggregates.CoarseGrid;
-    FineGrid=_Aggregates.FineGrid;
     verbose=0;
   };
 
   // The Pcg routine is common to all, but the various matrices differ from derived 
   // implementation to derived implmentation
+  void operator() (const Field &src, Field &psi){
   void operator() (const Field &src, Field &psi){
 
     psi.Checkerboard() = src.Checkerboard();
+    grid             = src.Grid();
 
+    RealD f;
     RealD rtzp,rtz,a,d,b;
-    //    RealD rptzp;
-    //    RealD tn;
+    RealD rptzp;
+    RealD tn;
     RealD guess = norm2(psi);
     RealD ssq   = norm2(src);
     RealD rsq   = ssq*Tolerance*Tolerance;
@@ -106,15 +104,15 @@ class TwoLevelFlexiblePcg : public LinearFunction<Field>
     /////////////////////////////
     // Set up history vectors
     /////////////////////////////
-    std::vector<Field> p  (mmax,FineGrid);
-    std::vector<Field> mmp(mmax,FineGrid);
+    std::vector<Field> p  (mmax,grid);
+    std::vector<Field> mmp(mmax,grid);
     std::vector<RealD> pAp(mmax);
 
-    Field x  (FineGrid); x = psi;
-    Field z  (FineGrid);
-    Field tmp(FineGrid);
-    Field r  (FineGrid);
-    Field mu (FineGrid);
+    Field x  (grid); x = psi;
+    Field z  (grid);
+    Field tmp(grid);
+    Field r  (grid);
+    Field mu (grid);
   
     //////////////////////////
     // x0 = Vstart -- possibly modify guess
@@ -123,13 +121,13 @@ class TwoLevelFlexiblePcg : public LinearFunction<Field>
     Vstart(x,src);
 
     // r0 = b -A x0
-    _Linop.HermOp(x,mmp[0]); // Shouldn't this be something else?
+    HermOp(x,mmp); // Shouldn't this be something else?
     axpy (r, -1.0,mmp[0], src);    // Recomputes r=src-Ax0
 
     //////////////////////////////////
     // Compute z = M1 x
     //////////////////////////////////
-    M1(r,z);
+    M1(r,z,tmp,mp,SmootherMirs);
     rtzp =real(innerProduct(r,z));
 
     ///////////////////////////////////////
@@ -145,7 +143,7 @@ class TwoLevelFlexiblePcg : public LinearFunction<Field>
       int peri_kp = (k+1) % mmax;
 
       rtz=rtzp;
-      d= M3(p[peri_k],mmp[peri_k]);
+      d= M3(p[peri_k],mp,mmp[peri_k],tmp);
       a = rtz/d;
     
       // Memorise this
@@ -155,13 +153,13 @@ class TwoLevelFlexiblePcg : public LinearFunction<Field>
       RealD rn = axpy_norm(r,-a,mmp[peri_k],r);
 
       // Compute z = M x
-      M1(r,z);
+      M1(r,z,tmp,mp);
 
       rtzp =real(innerProduct(r,z));
 
       M2(z,mu); // ADEF-2 this is identity. Axpy possible to eliminate
 
-      p[peri_kp]=mu;
+      p[peri_kp]=p[peri_k];
 
       // Standard search direction  p -> z + b p    ; b = 
       b = (rtzp)/rtz;
@@ -183,7 +181,7 @@ class TwoLevelFlexiblePcg : public LinearFunction<Field>
       // Stopping condition
       if ( rn <= rsq ) { 
 
-	_Linop.HermOp(x,mmp[0]); // Shouldn't this be something else?
+	HermOp(x,mmp); // Shouldn't this be something else?
 	axpy(tmp,-1.0,src,mmp[0]);
 	
 	RealD psinorm = sqrt(norm2(x));
@@ -192,8 +190,7 @@ class TwoLevelFlexiblePcg : public LinearFunction<Field>
 	RealD true_residual = tmpnorm/srcnorm;
 	std::cout<<GridLogMessage<<"TwoLevelfPcg:   true residual is "<<true_residual<<std::endl;
 	std::cout<<GridLogMessage<<"TwoLevelfPcg: target residual was"<<Tolerance<<std::endl;
-
-	return;
+	return k;
       }
     }
     // Non-convergence
@@ -202,40 +199,48 @@ class TwoLevelFlexiblePcg : public LinearFunction<Field>
 
  public:
 
-  virtual void M1(Field & in, Field & out) 
-  {// the smoother
+  virtual void M(Field & in,Field & out,Field & tmp) {
+
+  }
+
+  virtual void M1(Field & in, Field & out) {// the smoother
+
     // [PTM+Q] in = [1 - Q A] M in + Q in = Min + Q [ in -A Min]
-    Field tmp(FineGrid);
-    Field Min(FineGrid);
+    Field tmp(grid);
+    Field Min(grid);
 
-    CoarseField PleftProj(CoarseGrid);
-    CoarseField PleftMss_proj(CoarseGrid);
+    PcgM(in,Min); // Smoother call
 
-    _Smoother(in,Min); // Smoother call
-
-    _Linop.HermOp(Min,out);
+    HermOp(Min,out);
     axpy(tmp,-1.0,out,in);          // tmp  = in - A Min
 
-    _Aggregates.ProjectToSubspace(PleftProj,tmp);     
-    _CoarseSolver(PleftProj,PleftMss_proj); // Ass^{-1} [in - A Min]_s
-    _Aggregates.PromoteFromSubspace(PleftMss_proj,tmp);// tmp = Q[in - A Min]  
+    ProjectToSubspace(tmp,PleftProj);     
+    ApplyInverse(PleftProj,PleftMss_proj); // Ass^{-1} [in - A Min]_s
+    PromoteFromSubspace(PleftMss_proj,tmp);// tmp = Q[in - A Min]  
     axpy(out,1.0,Min,tmp); // Min+tmp
   }
 
-  virtual void M2(const Field & in, Field & out) 
-  {
+  virtual void M2(const Field & in, Field & out) {
     out=in;
+    // Must override for Def2 only
+    //  case PcgDef2:
+    //    Pright(in,out);
+    //    break;
   }
 
-  virtual RealD M3(const Field & p, Field & mmp)
-  {
+  virtual RealD M3(const Field & p, Field & mmp){
     double d,dd;
-    _Linop.HermOpAndNorm(p,mmp,d,dd);
+    HermOpAndNorm(p,mmp,d,dd);
     return dd;
+    // Must override for Def1 only
+    //  case PcgDef1:
+    //    d=linop_d->Mprec(p,mmp,tmp,0,1);// Dag no
+    //      linop_d->Mprec(mmp,mp,tmp,1);// Dag yes
+    //    Pleft(mp,mmp);
+    //    d=real(linop_d->inner(p,mmp));
   }
 
-  virtual void Vstart(Field & x,const Field & src)
-  {
+  virtual void VstartDef2(Field & xconst Field & src){
     //case PcgDef2:
     //case PcgAdef2: 
     //case PcgAdef2f:
@@ -251,79 +256,142 @@ class TwoLevelFlexiblePcg : public LinearFunction<Field>
     //                   = src_s - (A guess)_s - src_s  + (A guess)_s 
     //                   = 0 
     ///////////////////////////////////
-    Field r(FineGrid);
-    Field mmp(FineGrid);
-
-    CoarseField PleftProj(CoarseGrid);
-    CoarseField PleftMss_proj(CoarseGrid);
+    Field r(grid);
+    Field mmp(grid);
     
-    _Linop.HermOp(x,mmp);
+    HermOp(x,mmp);
     axpy (r, -1.0, mmp, src);        // r_{-1} = src - A x
-    _Aggregates.ProjectToSubspace(PleftProj,r);     
-    _CoarseSolver(PleftProj,PleftMss_proj); // Ass^{-1} r_s
-    _Aggregates.PromoteFromSubspace(PleftMss_proj,mmp);  
+    ProjectToSubspace(r,PleftProj);     
+    ApplyInverseCG(PleftProj,PleftMss_proj); // Ass^{-1} r_s
+    PromoteFromSubspace(PleftMss_proj,mmp);  
     x=x+mmp;
 
   }
 
+  virtual void Vstart(Field & x,const Field & src){
+    return;
+  }
+
   /////////////////////////////////////////////////////////////////////
   // Only Def1 has non-trivial Vout. Override in Def1
   /////////////////////////////////////////////////////////////////////
   virtual void   Vout  (Field & in, Field & out,Field & src){
     out = in;
+    //case PcgDef1:
+    //    //Qb + PT x
+    //    ProjectToSubspace(src,PleftProj);     
+    //    ApplyInverse(PleftProj,PleftMss_proj); // Ass^{-1} r_s
+    //    PromoteFromSubspace(PleftMss_proj,tmp);  
+    //    
+    //    Pright(in,out);
+    //    
+    //    linop_d->axpy(out,tmp,out,1.0);
+    //    break;
   }
 
   ////////////////////////////////////////////////////////////////////////////////////////////////
   // Pright and Pleft are common to all implementations
   ////////////////////////////////////////////////////////////////////////////////////////////////
-  virtual void Pright(Field & in,Field & out)
-  {
+  virtual void Pright(Field & in,Field & out){
     // P_R  = [ 1              0 ] 
     //        [ -Mss^-1 Msb    0 ] 
-    Field in_sbar(FineGrid);
+    Field in_sbar(grid);
 
-    CoarseField PleftProj(CoarseGrid);
-    CoarseField PleftMss_proj(CoarseGrid);
-
-    _Aggregates.ProjectToSubspace(PleftProj,in);     
-    _Aggregates.PromoteFromSubspace(PleftProj,out);  
+    ProjectToSubspace(in,PleftProj);     
+    PromoteFromSubspace(PleftProj,out);  
     axpy(in_sbar,-1.0,out,in);       // in_sbar = in - in_s 
 
-    _Linop.HermOp(in_sbar,out);
-    _Aggregates.ProjectToSubspace(PleftProj,out);           // Mssbar in_sbar  (project)
+    HermOp(in_sbar,out);
+    ProjectToSubspace(out,PleftProj);           // Mssbar in_sbar  (project)
 
-    _CoarseSolver(PleftProj,PleftMss_proj); // Mss^{-1} Mssbar 
-    _Aggregates.PromoteFromSubspace(PleftMss_proj,out);     // 
+    ApplyInverse     (PleftProj,PleftMss_proj); // Mss^{-1} Mssbar 
+    PromoteFromSubspace(PleftMss_proj,out);     // 
 
     axpy(out,-1.0,out,in_sbar);     // in_sbar - Mss^{-1} Mssbar in_sbar
   }
-  virtual void Pleft (Field & in,Field & out)
-  {
+  virtual void Pleft (Field & in,Field & out){
     // P_L  = [ 1  -Mbs Mss^-1] 
     //        [ 0   0         ] 
-    Field in_sbar(FineGrid);
-    Field    tmp2(FineGrid);
-    Field    Mtmp(FineGrid);
+    Field in_sbar(grid);
+    Field    tmp2(grid);
+    Field    Mtmp(grid);
 
-    CoarseField PleftProj(CoarseGrid);
-    CoarseField PleftMss_proj(CoarseGrid);
-
-    _Aggregates.ProjectToSubspace(PleftProj,in);     
-    _Aggregates.PromoteFromSubspace(PleftProj,out);  
+    ProjectToSubspace(in,PleftProj);     
+    PromoteFromSubspace(PleftProj,out);  
     axpy(in_sbar,-1.0,out,in);      // in_sbar = in - in_s
 
-    _CoarseSolver(PleftProj,PleftMss_proj); // Mss^{-1} in_s
-    _Aggregates.PromoteFromSubspace(PleftMss_proj,out);
+    ApplyInverse(PleftProj,PleftMss_proj); // Mss^{-1} in_s
+    PromoteFromSubspace(PleftMss_proj,out);
 
-    _Linop.HermOp(out,Mtmp);
+    HermOp(out,Mtmp);
 
-    _Aggregates.ProjectToSubspace(PleftProj,Mtmp);      // Msbar s Mss^{-1}
-    _Aggregates.PromoteFromSubspace(PleftProj,tmp2);
+    ProjectToSubspace(Mtmp,PleftProj);      // Msbar s Mss^{-1}
+    PromoteFromSubspace(PleftProj,tmp2);
 
     axpy(out,-1.0,tmp2,Mtmp);
     axpy(out,-1.0,out,in_sbar);     // in_sbar - Msbars Mss^{-1} in_s
   }
-};
-NAMESPACE_END(Grid);
+}
 
+template<class Field>
+class TwoLevelFlexiblePcgADef2 : public TwoLevelFlexiblePcg<Field> {
+ public:
+  virtual void M(Field & in,Field & out,Field & tmp){
+
+  } 
+  virtual void M1(Field & in, Field & out,Field & tmp,Field & mp){
+
+  }
+  virtual void M2(Field & in, Field & out){
+
+  }
+  virtual RealD M3(Field & p, Field & mp,Field & mmp, Field & tmp){
+
+  }
+  virtual void Vstart(Field & in, Field & src, Field & r, Field & mp, Field & mmp, Field & tmp){
+
+  }
+}
+/*
+template<class Field>
+class TwoLevelFlexiblePcgAD : public TwoLevelFlexiblePcg<Field> {
+ public:
+  virtual void M(Field & in,Field & out,Field & tmp); 
+  virtual void M1(Field & in, Field & out,Field & tmp,Field & mp);
+  virtual void M2(Field & in, Field & out);
+  virtual RealD M3(Field & p, Field & mp,Field & mmp, Field & tmp);
+  virtual void Vstart(Field & in, Field & src, Field & r, Field & mp, Field & mmp, Field & tmp);
+}
+
+template<class Field>
+class TwoLevelFlexiblePcgDef1 : public TwoLevelFlexiblePcg<Field> {
+ public:
+  virtual void M(Field & in,Field & out,Field & tmp); 
+  virtual void M1(Field & in, Field & out,Field & tmp,Field & mp);
+  virtual void M2(Field & in, Field & out);
+  virtual RealD M3(Field & p, Field & mp,Field & mmp, Field & tmp);
+  virtual void Vstart(Field & in, Field & src, Field & r, Field & mp, Field & mmp, Field & tmp);
+  virtual void   Vout  (Field & in, Field & out,Field & src,Field & tmp);
+}
+
+template<class Field>
+class TwoLevelFlexiblePcgDef2 : public TwoLevelFlexiblePcg<Field> {
+ public:
+  virtual void M(Field & in,Field & out,Field & tmp); 
+  virtual void M1(Field & in, Field & out,Field & tmp,Field & mp);
+  virtual void M2(Field & in, Field & out);
+  virtual RealD M3(Field & p, Field & mp,Field & mmp, Field & tmp);
+  virtual void Vstart(Field & in, Field & src, Field & r, Field & mp, Field & mmp, Field & tmp);
+}
+
+template<class Field>
+class TwoLevelFlexiblePcgV11: public TwoLevelFlexiblePcg<Field> {
+ public:
+  virtual void M(Field & in,Field & out,Field & tmp); 
+  virtual void M1(Field & in, Field & out,Field & tmp,Field & mp);
+  virtual void M2(Field & in, Field & out);
+  virtual RealD M3(Field & p, Field & mp,Field & mmp, Field & tmp);
+  virtual void Vstart(Field & in, Field & src, Field & r, Field & mp, Field & mmp, Field & tmp);
+}
+*/
 #endif

Grid/algorithms/iterative/Deflation.h

@@ -60,8 +60,6 @@ public:
   DeflatedGuesser(const std::vector<Field> & _evec,const std::vector<RealD> & _eval) : evec(_evec), eval(_eval) {};
 
   virtual void operator()(const Field &src,Field &guess) {
-    RealD t=-usecond();
-    
     guess = Zero();
     assert(evec.size()==eval.size());
     auto N = evec.size();
@@ -70,8 +68,6 @@ public:
       axpy(guess,TensorRemove(innerProduct(tmp,src)) / eval[i],tmp,guess);
     }
     guess.Checkerboard() = src.Checkerboard();
-    t+=usecond();
-    std::cout<<GridLogMessage<<"\t\t\t" << "Deflated guess took "<< t/1000.0<< "ms" <<std::endl;
   }
 };
 

Grid/algorithms/iterative/PrecConjugateResidual.h

@@ -59,7 +59,7 @@ public:
     GridBase *grid = src.Grid();
     Field r(grid),  p(grid), Ap(grid), Ar(grid), z(grid);
       
-    psi=Zero();
+    psi=zero;
     r  = src;
     Preconditioner(r,p);
 

Grid/allocator/AlignedAllocator.h

@@ -53,11 +53,7 @@ public:
   { 
     size_type bytes = __n*sizeof(_Tp);
     profilerAllocate(bytes);
-#ifdef GRID_UVM
-    _Tp *ptr = (_Tp*) MemoryManager::SharedAllocate(bytes);
-#else 
     _Tp *ptr = (_Tp*) MemoryManager::CpuAllocate(bytes);
-#endif
     assert( ( (_Tp*)ptr != (_Tp *)NULL ) );
     return ptr;
   }
@@ -66,14 +62,11 @@ public:
   { 
     size_type bytes = __n * sizeof(_Tp);
     profilerFree(bytes);
-#ifdef GRID_UVM
-    MemoryManager::SharedFree((void *)__p,bytes);
-#else
     MemoryManager::CpuFree((void *)__p,bytes);
-#endif
   }
 
-  // FIXME: hack for the copy constructor: it must be avoided to avoid single thread loop
+  // 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) { };
@@ -81,9 +74,6 @@ public:
 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; }
 
-//////////////////////////////////////////////////////////////////////////////////////
-// Unified virtual memory
-//////////////////////////////////////////////////////////////////////////////////////
 template<typename _Tp>
 class uvmAllocator {
 public: 
@@ -119,63 +109,22 @@ public:
     MemoryManager::SharedFree((void *)__p,bytes);
   }
 
+  // FIXME: hack for the copy constructor, eventually it must be avoided
   void construct(pointer __p, const _Tp& __val) { new((void *)__p) _Tp(__val); };
+  //void construct(pointer __p, const _Tp& __val) { };
   void construct(pointer __p) { };
   void destroy(pointer __p) { };
 };
 template<typename _Tp>  inline bool operator==(const uvmAllocator<_Tp>&, const uvmAllocator<_Tp>&){ return true; }
 template<typename _Tp>  inline bool operator!=(const uvmAllocator<_Tp>&, const uvmAllocator<_Tp>&){ return false; }
 
-////////////////////////////////////////////////////////////////////////////////
-// Device memory
-////////////////////////////////////////////////////////////////////////////////
-template<typename _Tp>
-class devAllocator {
-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 devAllocator<_Tp1> other; };
-  devAllocator() throw() { }
-  devAllocator(const devAllocator&) throw() { }
-  template<typename _Tp1> devAllocator(const devAllocator<_Tp1>&) throw() { }
-  ~devAllocator() 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::AcceleratorAllocate(bytes);
-    assert( ( (_Tp*)ptr != (_Tp *)NULL ) );
-    return ptr;
-  }
-
-  void deallocate(pointer __p, size_type __n) 
-  { 
-    size_type bytes = __n * sizeof(_Tp);
-    profilerFree(bytes);
-    MemoryManager::AcceleratorFree((void *)__p,bytes);
-  }
-  void construct(pointer __p, const _Tp& __val) { };
-  void construct(pointer __p) { };
-  void destroy(pointer __p) { };
-};
-template<typename _Tp>  inline bool operator==(const devAllocator<_Tp>&, const devAllocator<_Tp>&){ return true; }
-template<typename _Tp>  inline bool operator!=(const devAllocator<_Tp>&, const devAllocator<_Tp>&){ return false; }
-
 ////////////////////////////////////////////////////////////////////////////////
 // Template typedefs
 ////////////////////////////////////////////////////////////////////////////////
-//template<class T> using commAllocator = devAllocator<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,devAllocator<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);
 

Grid/allocator/MemoryManager.cc

@@ -9,13 +9,11 @@ NAMESPACE_BEGIN(Grid);
 #define AccSmall (3)
 #define Shared   (4)
 #define SharedSmall (5)
-uint64_t total_cache;
 uint64_t total_shared;
 uint64_t total_device;
 uint64_t total_host;;
 void MemoryManager::PrintBytes(void)
 {
-  std::cout << " MemoryManager : "<<total_cache <<" cache       bytes "<<std::endl;
   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;
@@ -37,8 +35,6 @@ void *MemoryManager::AcceleratorAllocate(size_t bytes)
   if ( ptr == (void *) NULL ) {
     ptr = (void *) acceleratorAllocDevice(bytes);
     total_device+=bytes;
-  } else {
-    //    std::cout <<"AcceleratorAllocate: cache hit Device pointer "<<std::hex<<ptr<<std::dec<<" "<<bytes<<std::endl;
   }
   return ptr;
 }
@@ -57,10 +53,8 @@ void *MemoryManager::SharedAllocate(size_t bytes)
   if ( ptr == (void *) NULL ) {
     ptr = (void *) acceleratorAllocShared(bytes);
     total_shared+=bytes;
-    //    std::cout <<"SharedAllocate: allocated Shared pointer "<<std::hex<<ptr<<std::dec<<std::endl;
+    //    std::cout <<"AcceleratorAllocate: allocated Shared pointer "<<std::hex<<ptr<<std::dec<<std::endl;
     //    PrintBytes();
-  } else {
-    //    std::cout <<"SharedAllocate: cache hit Shared pointer "<<std::hex<<ptr<<std::dec<<" "<<bytes<<std::endl;
   }
   return ptr;
 }
@@ -80,9 +74,6 @@ void *MemoryManager::CpuAllocate(size_t bytes)
   if ( ptr == (void *) NULL ) {
     ptr = (void *) acceleratorAllocShared(bytes);
     total_host+=bytes;
-    //    std::cout <<"CpuAllocate: allocated Cpu pointer "<<std::hex<<ptr<<std::dec<<std::endl;
-  } else {
-    //    std::cout <<"CpufAllocate: cache hit Cpu pointer "<<std::hex<<ptr<<std::dec<<" "<<bytes<<std::endl;
   }
   return ptr;
 }
@@ -129,7 +120,7 @@ void MemoryManager::Init(void)
   str= getenv("GRID_ALLOC_NCACHE_LARGE");
   if ( str ) {
     Nc = atoi(str);
-    if ( (Nc>=0) && (Nc <= NallocCacheMax)) {
+    if ( (Nc>=0) && (Nc < NallocCacheMax)) {
       Ncache[Cpu]=Nc;
       Ncache[Acc]=Nc;
       Ncache[Shared]=Nc;
@@ -139,26 +130,17 @@ void MemoryManager::Init(void)
   str= getenv("GRID_ALLOC_NCACHE_SMALL");
   if ( str ) {
     Nc = atoi(str);
-    if ( (Nc>=0) && (Nc <= NallocCacheMax)) {
+    if ( (Nc>=0) && (Nc < NallocCacheMax)) {
       Ncache[CpuSmall]=Nc;
       Ncache[AccSmall]=Nc;
       Ncache[SharedSmall]=Nc;
     }
   }
-
-}
-
-void MemoryManager::InitMessage(void) {
-
-#ifndef GRID_UVM
-  std::cout << GridLogMessage << "MemoryManager Cache "<< MemoryManager::DeviceMaxBytes <<" bytes "<<std::endl;
-#endif
-  
   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
@@ -182,7 +164,6 @@ void MemoryManager::InitMessage(void) {
   std::cout << GridLogMessage<< "MemoryManager::Init() Using SYCL malloc_device"<<std::endl;
 #endif
 #endif
-
 }
 
 void *MemoryManager::Insert(void *ptr,size_t bytes,int type) 
@@ -220,7 +201,6 @@ void *MemoryManager::Insert(void *ptr,size_t bytes,AllocationCacheEntry *entries
 
   if ( entries[v].valid ) {
     ret = entries[v].address;
-    total_cache-=entries[v].bytes;
     entries[v].valid = 0;
     entries[v].address = NULL;
     entries[v].bytes = 0;
@@ -229,7 +209,6 @@ void *MemoryManager::Insert(void *ptr,size_t bytes,AllocationCacheEntry *entries
   entries[v].address=ptr;
   entries[v].bytes  =bytes;
   entries[v].valid  =1;
-  total_cache+=entries[v].bytes;
 
   return ret;
 }
@@ -254,7 +233,6 @@ void *MemoryManager::Lookup(size_t bytes,AllocationCacheEntry *entries,int ncach
   for(int e=0;e<ncache;e++){
     if ( entries[e].valid && ( entries[e].bytes == bytes ) ) {
       entries[e].valid = 0;
-      total_cache-=bytes;
       return entries[e].address;
     }
   }

Grid/allocator/MemoryManager.h

@@ -93,12 +93,11 @@ private:
   static void *Insert(void *ptr,size_t bytes,AllocationCacheEntry *entries,int ncache,int &victim) ;
   static void *Lookup(size_t bytes,AllocationCacheEntry *entries,int ncache) ;
 
- public:
-  static void PrintBytes(void);
-  static void Init(void);
-  static void InitMessage(void);
   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);

Grid/communicator/Communicator_mpi3.cc

@@ -1,6 +1,6 @@
 /*************************************************************************************
 
-    Grid physics library, www.github.com/paboyle/Grid
+    Grid physics library, www.github.com/paboyle/Grid 
 
     Source file: ./lib/communicator/Communicator_mpi.cc
 
@@ -35,7 +35,7 @@ Grid_MPI_Comm       CartesianCommunicator::communicator_world;
 ////////////////////////////////////////////
 // First initialise of comms system
 ////////////////////////////////////////////
-void CartesianCommunicator::Init(int *argc, char ***argv)
+void CartesianCommunicator::Init(int *argc, char ***argv) 
 {
 
   int flag;
@@ -43,16 +43,8 @@ void CartesianCommunicator::Init(int *argc, char ***argv)
 
   MPI_Initialized(&flag); // needed to coexist with other libs apparently
   if ( !flag ) {
-
-#if defined (TOFU) // FUGAKU, credits go to Issaku Kanamori
-    nCommThreads=1;
-    // wrong results here too
-    // For now: comms-overlap leads to wrong results in Benchmark_wilson even on single node MPI runs
-    // other comms schemes are ok
-    MPI_Init_thread(argc,argv,MPI_THREAD_SERIALIZED,&provided);
-#else
     MPI_Init_thread(argc,argv,MPI_THREAD_MULTIPLE,&provided);
-#endif
+
     //If only 1 comms thread we require any threading mode other than SINGLE, but for multiple comms threads we need MULTIPLE
     if( (nCommThreads == 1) && (provided == MPI_THREAD_SINGLE) ) {
       assert(0);
@@ -99,7 +91,7 @@ void  CartesianCommunicator::ProcessorCoorFromRank(int rank, Coordinate &coor)
 ////////////////////////////////////////////////////////////////////////////////////////////////////////
 // Initialises from communicator_world
 ////////////////////////////////////////////////////////////////////////////////////////////////////////
-CartesianCommunicator::CartesianCommunicator(const Coordinate &processors)
+CartesianCommunicator::CartesianCommunicator(const Coordinate &processors) 
 {
   MPI_Comm optimal_comm;
   ////////////////////////////////////////////////////
@@ -118,7 +110,7 @@ CartesianCommunicator::CartesianCommunicator(const Coordinate &processors)
 //////////////////////////////////
 // Try to subdivide communicator
 //////////////////////////////////
-CartesianCommunicator::CartesianCommunicator(const Coordinate &processors,const CartesianCommunicator &parent,int &srank)
+CartesianCommunicator::CartesianCommunicator(const Coordinate &processors,const CartesianCommunicator &parent,int &srank)    
 {
   _ndimension = processors.size();  assert(_ndimension>=1);
   int parent_ndimension = parent._ndimension; assert(_ndimension >= parent._ndimension);
@@ -135,7 +127,7 @@ CartesianCommunicator::CartesianCommunicator(const Coordinate &processors,const
   //////////////////////////////////////////////////////////////////////////////////////////////////////
   // split the communicator
   //////////////////////////////////////////////////////////////////////////////////////////////////////
-  //  int Nparent = parent._processors ;
+  //  int Nparent = parent._processors ; 
   int Nparent;
   MPI_Comm_size(parent.communicator,&Nparent);
 
@@ -157,13 +149,13 @@ CartesianCommunicator::CartesianCommunicator(const Coordinate &processors,const
   }
 
   // rank within subcomm ; srank is rank of subcomm within blocks of subcomms
-  int crank;
+  int crank;  
   // Mpi uses the reverse Lexico convention to us; so reversed routines called
   Lexicographic::IndexFromCoorReversed(ccoor,crank,processors); // processors is the split grid dimensions
   Lexicographic::IndexFromCoorReversed(scoor,srank,ssize);      // ssize is the number of split grids
 
   MPI_Comm comm_split;
-  if ( Nchild > 1 ) {
+  if ( Nchild > 1 ) { 
 
     ////////////////////////////////////////////////////////////////
     // Split the communicator
@@ -188,11 +180,11 @@ CartesianCommunicator::CartesianCommunicator(const Coordinate &processors,const
   SetCommunicator(comm_split);
 
   ///////////////////////////////////////////////
-  // Free the temp communicator
+  // Free the temp communicator 
   ///////////////////////////////////////////////
   MPI_Comm_free(&comm_split);
 
-  if(0){
+  if(0){ 
     std::cout << " ndim " <<_ndimension<<" " << parent._ndimension << std::endl;
     for(int d=0;d<processors.size();d++){
       std::cout << d<< " " << _processor_coor[d] <<" " <<  ccoor[d]<<std::endl;
@@ -253,7 +245,7 @@ CartesianCommunicator::~CartesianCommunicator()
     for(int i=0;i<communicator_halo.size();i++){
       MPI_Comm_free(&communicator_halo[i]);
     }
-  }
+  }  
 }
 void CartesianCommunicator::GlobalSum(uint32_t &u){
   int ierr=MPI_Allreduce(MPI_IN_PLACE,&u,1,MPI_UINT32_T,MPI_SUM,communicator);
@@ -302,28 +294,60 @@ void CartesianCommunicator::SendToRecvFrom(void *xmit,
 					   int bytes)
 {
   std::vector<CommsRequest_t> reqs(0);
-  unsigned long  xcrc = crc32(0L, Z_NULL, 0);
-  unsigned long  rcrc = crc32(0L, Z_NULL, 0);
-
+  //    unsigned long  xcrc = crc32(0L, Z_NULL, 0);
+  //    unsigned long  rcrc = crc32(0L, Z_NULL, 0);
+  //    xcrc = crc32(xcrc,(unsigned char *)xmit,bytes);
+  SendToRecvFromBegin(reqs,xmit,dest,recv,from,bytes);
+  SendToRecvFromComplete(reqs);
+  //    rcrc = crc32(rcrc,(unsigned char *)recv,bytes);
+  //    printf("proc %d SendToRecvFrom %d bytes %lx %lx\n",_processor,bytes,xcrc,rcrc);
+}
+void CartesianCommunicator::SendRecvPacket(void *xmit,
+					   void *recv,
+					   int sender,
+					   int receiver,
+					   int bytes)
+{
+  MPI_Status stat;
+  assert(sender != receiver);
+  int tag = sender;
+  if ( _processor == sender ) {
+    MPI_Send(xmit, bytes, MPI_CHAR,receiver,tag,communicator);
+  }
+  if ( _processor == receiver ) { 
+    MPI_Recv(recv, bytes, MPI_CHAR,sender,tag,communicator,&stat);
+  }
+}
+// Basic Halo comms primitive
+void CartesianCommunicator::SendToRecvFromBegin(std::vector<CommsRequest_t> &list,
+						void *xmit,
+						int dest,
+						void *recv,
+						int from,
+						int bytes)
+{
   int myrank = _processor;
   int ierr;
 
-  // Enforce no UVM in comms, device or host OK
-  assert(acceleratorIsCommunicable(xmit));
-  assert(acceleratorIsCommunicable(recv));
+  if ( CommunicatorPolicy == CommunicatorPolicyConcurrent ) { 
+    MPI_Request xrq;
+    MPI_Request rrq;
 
-  // Give the CPU to MPI immediately; can use threads to overlap optionally
-  //  printf("proc %d SendToRecvFrom %d bytes Sendrecv \n",_processor,bytes);
-  ierr=MPI_Sendrecv(xmit,bytes,MPI_CHAR,dest,myrank,
-		    recv,bytes,MPI_CHAR,from, from,
-		    communicator,MPI_STATUS_IGNORE);
-  assert(ierr==0);
-
-  //  xcrc = crc32(xcrc,(unsigned char *)xmit,bytes);
-  //  rcrc = crc32(rcrc,(unsigned char *)recv,bytes);
-  //  printf("proc %d SendToRecvFrom %d bytes xcrc %lx rcrc %lx\n",_processor,bytes,xcrc,rcrc); fflush
+    ierr =MPI_Irecv(recv, bytes, MPI_CHAR,from,from,communicator,&rrq);
+    ierr|=MPI_Isend(xmit, bytes, MPI_CHAR,dest,_processor,communicator,&xrq);
+    
+    assert(ierr==0);
+    list.push_back(xrq);
+    list.push_back(rrq);
+  } else { 
+    // Give the CPU to MPI immediately; can use threads to overlap optionally
+    ierr=MPI_Sendrecv(xmit,bytes,MPI_CHAR,dest,myrank,
+		      recv,bytes,MPI_CHAR,from, from,
+		      communicator,MPI_STATUS_IGNORE);
+    assert(ierr==0);
+  }
 }
-// Basic Halo comms primitive
+
 double CartesianCommunicator::StencilSendToRecvFrom( void *xmit,
 						     int dest,
 						     void *recv,
@@ -343,7 +367,7 @@ double CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsReques
 							 int from,
 							 int bytes,int dir)
 {
-  int ncomm  =communicator_halo.size();
+  int ncomm  =communicator_halo.size(); 
   int commdir=dir%ncomm;
 
   MPI_Request xrq;
@@ -373,13 +397,21 @@ double CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsReques
     off_node_bytes+=bytes;
   }
 
-  if ( CommunicatorPolicy == CommunicatorPolicySequential ) {
+  if ( CommunicatorPolicy == CommunicatorPolicySequential ) { 
     this->StencilSendToRecvFromComplete(list,dir);
   }
 
   return off_node_bytes;
 }
-void CartesianCommunicator::StencilSendToRecvFromComplete(std::vector<CommsRequest_t> &list,int dir)
+void CartesianCommunicator::StencilSendToRecvFromComplete(std::vector<CommsRequest_t> &waitall,int dir)
+{
+  SendToRecvFromComplete(waitall);
+}
+void CartesianCommunicator::StencilBarrier(void)
+{
+  MPI_Barrier  (ShmComm);
+}
+void CartesianCommunicator::SendToRecvFromComplete(std::vector<CommsRequest_t> &list)
 {
   int nreq=list.size();
 
@@ -390,13 +422,6 @@ void CartesianCommunicator::StencilSendToRecvFromComplete(std::vector<CommsReque
   assert(ierr==0);
   list.resize(0);
 }
-void CartesianCommunicator::StencilBarrier(void)
-{
-  MPI_Barrier  (ShmComm);
-}
-//void CartesianCommunicator::SendToRecvFromComplete(std::vector<CommsRequest_t> &list)
-//{
-//}
 void CartesianCommunicator::Barrier(void)
 {
   int ierr = MPI_Barrier(communicator);
@@ -411,8 +436,8 @@ void CartesianCommunicator::Broadcast(int root,void* data, int bytes)
 		     communicator);
   assert(ierr==0);
 }
-int CartesianCommunicator::RankWorld(void){
-  int r;
+int CartesianCommunicator::RankWorld(void){ 
+  int r; 
   MPI_Comm_rank(communicator_world,&r);
   return r;
 }
@@ -445,7 +470,7 @@ void CartesianCommunicator::AllToAll(void  *in,void *out,uint64_t words,uint64_t
   // When 24*4 bytes multiples get 50x 10^9 >>> 2x10^9 Y2K bug.
   // (Turns up on 32^3 x 64 Gparity too)
   MPI_Datatype object;
-  int iwords;
+  int iwords; 
   int ibytes;
   iwords = words;
   ibytes = bytes;
@@ -458,3 +483,5 @@ void CartesianCommunicator::AllToAll(void  *in,void *out,uint64_t words,uint64_t
 }
 
 NAMESPACE_END(Grid);
+
+

Grid/communicator/SharedMemoryMPI.cc

@@ -47,12 +47,7 @@ void GlobalSharedMemory::Init(Grid_MPI_Comm comm)
   /////////////////////////////////////////////////////////////////////
   // Split into groups that can share memory
   /////////////////////////////////////////////////////////////////////
-#ifndef GRID_MPI3_SHM_NONE
   MPI_Comm_split_type(comm, MPI_COMM_TYPE_SHARED, 0, MPI_INFO_NULL,&WorldShmComm);
-#else
-  MPI_Comm_split(comm, WorldRank, 0, &WorldShmComm);
-#endif
-
   MPI_Comm_rank(WorldShmComm     ,&WorldShmRank);
   MPI_Comm_size(WorldShmComm     ,&WorldShmSize);
 
@@ -448,11 +443,7 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
   ///////////////////////////////////////////////////////////////////////////////////////////////////////////
   // Each MPI rank should allocate our own buffer
   ///////////////////////////////////////////////////////////////////////////////////////////////////////////
-#ifndef GRID_MPI3_SHM_NONE
   auto err =  cudaMalloc(&ShmCommBuf, bytes);
-#else
-  auto err =  cudaMallocManaged(&ShmCommBuf, bytes);
-#endif
   if ( err !=  cudaSuccess) {
     std::cerr << " SharedMemoryMPI.cc cudaMallocManaged failed for " << bytes<<" bytes " <<cudaGetErrorString(err)<< std::endl;
     exit(EXIT_FAILURE);  
@@ -470,8 +461,7 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
   // Loop over ranks/gpu's on our node
   ///////////////////////////////////////////////////////////////////////////////////////////////////////////
   for(int r=0;r<WorldShmSize;r++){
-
-#ifndef GRID_MPI3_SHM_NONE
+    
     //////////////////////////////////////////////////
     // If it is me, pass around the IPC access key
     //////////////////////////////////////////////////
@@ -511,9 +501,6 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
     // Save a copy of the device buffers
     ///////////////////////////////////////////////////////////////
     WorldShmCommBufs[r] = thisBuf;
-#else
-    WorldShmCommBufs[r] = ShmCommBuf;
-#endif
   }
 
   _ShmAllocBytes=bytes;
@@ -718,11 +705,7 @@ void SharedMemory::SetCommunicator(Grid_MPI_Comm comm)
   /////////////////////////////////////////////////////////////////////
   // Split into groups that can share memory
   /////////////////////////////////////////////////////////////////////
-#ifndef GRID_MPI3_SHM_NONE
   MPI_Comm_split_type(comm, MPI_COMM_TYPE_SHARED, 0, MPI_INFO_NULL,&ShmComm);
-#else
-  MPI_Comm_split(comm, rank, 0, &ShmComm);
-#endif
   MPI_Comm_rank(ShmComm     ,&ShmRank);
   MPI_Comm_size(ShmComm     ,&ShmSize);
   ShmCommBufs.resize(ShmSize);

Grid/cshift/Cshift.h

@@ -52,8 +52,23 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 
 NAMESPACE_BEGIN(Grid);
 
-template<class Expression,typename std::enable_if<is_lattice_expr<Expression>::value,void>::type * = nullptr> 
-auto Cshift(const Expression &expr,int dim,int shift)  -> decltype(closure(expr)) 
+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);
 }

Grid/cshift/Cshift_common.h

@@ -76,8 +76,8 @@ Gather_plane_simple (const Lattice<vobj> &rhs,commVector<vobj> &buffer,int dimen
     autoView(rhs_v , rhs, AcceleratorRead);
     auto buffer_p = & buffer[0];
     auto table = &Cshift_table[0];
-    accelerator_for(i,ent,vobj::Nsimd(),{
-	coalescedWrite(buffer_p[table[i].first],coalescedRead(rhs_v[table[i].second]));
+    accelerator_for(i,ent,1,{
+      buffer_p[table[i].first]=rhs_v[table[i].second];
     });
   }
 }
@@ -185,8 +185,8 @@ template<class vobj> void Scatter_plane_simple (Lattice<vobj> &rhs,commVector<vo
     autoView( rhs_v, rhs, AcceleratorWrite);
     auto buffer_p = & buffer[0];
     auto table = &Cshift_table[0];
-    accelerator_for(i,ent,vobj::Nsimd(),{
-	coalescedWrite(rhs_v[table[i].first],coalescedRead(buffer_p[table[i].second]));
+    accelerator_for(i,ent,1,{
+	rhs_v[table[i].first]=buffer_p[table[i].second];
     });
   }
 }
@@ -209,11 +209,9 @@ template<class vobj> void Scatter_plane_merge(Lattice<vobj> &rhs,ExtractPointerA
 
   if(cbmask ==0x3 ) {
     autoView( rhs_v , rhs, AcceleratorWrite);
-    int _slice_stride = rhs.Grid()->_slice_stride[dimension];
-    int _slice_block = rhs.Grid()->_slice_block[dimension];
     accelerator_for2d(n,e1,b,e2,1,{
-	int o      = n*_slice_stride;
-	int offset = b+n*_slice_block;
+	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 { 
@@ -222,7 +220,6 @@ template<class vobj> void Scatter_plane_merge(Lattice<vobj> &rhs,ExtractPointerA
     // 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;
-    assert(0); // This will fail if hit on GPU
     autoView( rhs_v, rhs, CpuWrite);
     for(int n=0;n<e1;n++){
       for(int b=0;b<e2;b++){
@@ -283,8 +280,8 @@ template<class vobj> void Copy_plane(Lattice<vobj>& lhs,const Lattice<vobj> &rhs
     autoView(rhs_v , rhs, AcceleratorRead);
     autoView(lhs_v , lhs, AcceleratorWrite);
     auto table = &Cshift_table[0];
-    accelerator_for(i,ent,vobj::Nsimd(),{
-      coalescedWrite(lhs_v[table[i].first],coalescedRead(rhs_v[table[i].second]));
+    accelerator_for(i,ent,1,{
+      lhs_v[table[i].first]=rhs_v[table[i].second];
     });
   }
 }

Grid/lattice/Lattice.h

@@ -36,8 +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_real_imag.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>

Grid/lattice/Lattice_ET.h

@@ -42,24 +42,9 @@ NAMESPACE_BEGIN(Grid);
 ////////////////////////////////////////////////////
 // Predicated where support
 ////////////////////////////////////////////////////
-#ifdef GRID_SIMT
-// drop to scalar in SIMT; cleaner in fact
 template <class iobj, class vobj, class robj>
-accelerator_inline vobj predicatedWhere(const iobj &predicate, 
-					const vobj &iftrue, 
-					const robj &iffalse) 
-{
-  Integer mask = TensorRemove(predicate);
-  typename std::remove_const<vobj>::type ret= iffalse;
-  if (mask) ret=iftrue;
-  return ret;
-}
-#else
-template <class iobj, class vobj, class robj>
-accelerator_inline vobj predicatedWhere(const iobj &predicate, 
-					const vobj &iftrue, 
-					const robj &iffalse) 
-{
+accelerator_inline vobj predicatedWhere(const iobj &predicate, const vobj &iftrue,
+                            const robj &iffalse) {
   typename std::remove_const<vobj>::type ret;
 
   typedef typename vobj::scalar_object scalar_object;
@@ -83,7 +68,6 @@ accelerator_inline vobj predicatedWhere(const iobj &predicate,
   merge(ret, falsevals);
   return ret;
 }
-#endif
 
 /////////////////////////////////////////////////////
 //Specialization of getVectorType for lattices
@@ -97,62 +81,32 @@ struct getVectorType<Lattice<T> >{
 //--  recursive evaluation of expressions; --
 // handle leaves of syntax tree
 ///////////////////////////////////////////////////
-template<class sobj,
-  typename std::enable_if<!is_lattice<sobj>::value&&!is_lattice_expr<sobj>::value,sobj>::type * = nullptr> 
-accelerator_inline 
+template<class sobj> accelerator_inline 
 sobj eval(const uint64_t ss, const sobj &arg)
 {
   return arg;
 }
-template <class lobj> accelerator_inline 
-auto eval(const uint64_t ss, const LatticeView<lobj> &arg) -> decltype(arg(ss))
-{
-  return arg(ss);
-}
 
-////////////////////////////////////////////
-//--  recursive evaluation of expressions; --
-// whole vector return, used only for expression return type inference
-///////////////////////////////////////////////////
-template<class sobj> accelerator_inline 
-sobj vecEval(const uint64_t ss, const sobj &arg)
-{
-  return arg;
-}
 template <class lobj> accelerator_inline 
-const lobj & vecEval(const uint64_t ss, const LatticeView<lobj> &arg) 
+const lobj & eval(const uint64_t ss, const LatticeView<lobj> &arg) 
 {
   return arg[ss];
 }
 
-///////////////////////////////////////////////////
-// handle nodes in syntax tree- eval one operand
-// vecEval needed (but never called as all expressions offloaded) to infer the return type
-// in SIMT contexts of closure.
-///////////////////////////////////////////////////
-template <typename Op, typename T1> accelerator_inline 
-auto vecEval(const uint64_t ss, const LatticeUnaryExpression<Op, T1> &expr)  
-  -> decltype(expr.op.func( vecEval(ss, expr.arg1)))
+// 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) 
 {
-  return expr.op.func( vecEval(ss, expr.arg1) );
-}
-// vecEval two operands
-template <typename Op, typename T1, typename T2> accelerator_inline
-auto vecEval(const uint64_t ss, const LatticeBinaryExpression<Op, T1, T2> &expr)  
-  -> decltype(expr.op.func( vecEval(ss,expr.arg1),vecEval(ss,expr.arg2)))
-{
-  return expr.op.func( vecEval(ss,expr.arg1), vecEval(ss,expr.arg2) );
-}
-// vecEval three operands
-template <typename Op, typename T1, typename T2, typename T3> accelerator_inline
-auto vecEval(const uint64_t ss, const LatticeTrinaryExpression<Op, T1, T2, T3> &expr)  
-  -> decltype(expr.op.func(vecEval(ss, expr.arg1), vecEval(ss, expr.arg2), vecEval(ss, expr.arg3)))
-{
-  return expr.op.func(vecEval(ss, expr.arg1), vecEval(ss, expr.arg2), vecEval(ss, expr.arg3));
+  auto view = arg.View(AcceleratorRead);
+  return view[ss];
 }
+#endif
 
 ///////////////////////////////////////////////////
-// handle nodes in syntax tree- eval one operand coalesced
+// handle nodes in syntax tree- eval one operand
 ///////////////////////////////////////////////////
 template <typename Op, typename T1> accelerator_inline 
 auto eval(const uint64_t ss, const LatticeUnaryExpression<Op, T1> &expr)  
@@ -160,41 +114,23 @@ auto eval(const uint64_t ss, const LatticeUnaryExpression<Op, T1> &expr)
 {
   return expr.op.func( eval(ss, expr.arg1) );
 }
+///////////////////////
 // eval two operands
+///////////////////////
 template <typename Op, typename T1, typename T2> accelerator_inline
 auto eval(const uint64_t ss, const LatticeBinaryExpression<Op, T1, T2> &expr)  
   -> decltype(expr.op.func( eval(ss,expr.arg1),eval(ss,expr.arg2)))
 {
   return expr.op.func( eval(ss,expr.arg1), eval(ss,expr.arg2) );
 }
+///////////////////////
 // eval three operands
+///////////////////////
 template <typename Op, typename T1, typename T2, typename T3> accelerator_inline
 auto eval(const uint64_t ss, const LatticeTrinaryExpression<Op, T1, T2, T3> &expr)  
-  -> decltype(expr.op.func(eval(ss, expr.arg1), 
-			   eval(ss, expr.arg2), 
-			   eval(ss, expr.arg3)))
+  -> decltype(expr.op.func(eval(ss, expr.arg1), eval(ss, expr.arg2), eval(ss, expr.arg3)))
 {
-#ifdef GRID_SIMT
-  // Handles Nsimd (vInteger) != Nsimd(ComplexD)
-  typedef decltype(vecEval(ss, expr.arg2)) rvobj;
-  typedef typename std::remove_reference<rvobj>::type vobj;
-
-  const int Nsimd = vobj::vector_type::Nsimd();
-
-  auto vpred = vecEval(ss,expr.arg1);
-
-  ExtractBuffer<Integer> mask(Nsimd);
-  extract<vInteger, Integer>(TensorRemove(vpred), mask);
-
-  int s = acceleratorSIMTlane(Nsimd);
-  return expr.op.func(mask[s],
-		      eval(ss, expr.arg2), 
-		      eval(ss, expr.arg3));
-#else
-  return expr.op.func(eval(ss, expr.arg1),
-		      eval(ss, expr.arg2), 
-		      eval(ss, expr.arg3));
-#endif
+  return expr.op.func(eval(ss, expr.arg1), eval(ss, expr.arg2), eval(ss, expr.arg3));
 }
 
 //////////////////////////////////////////////////////////////////////////
@@ -292,7 +228,7 @@ template <typename Op, typename T1, typename T2> inline
 void ExpressionViewOpen(LatticeBinaryExpression<Op, T1, T2> &expr) 
 {
   ExpressionViewOpen(expr.arg1);  // recurse AST
-  ExpressionViewOpen(expr.arg2);  // rrecurse AST
+  ExpressionViewOpen(expr.arg2);  // recurse AST
 }
 template <typename Op, typename T1, typename T2, typename T3>
 inline void ExpressionViewOpen(LatticeTrinaryExpression<Op, T1, T2, T3> &expr) 
@@ -336,8 +272,9 @@ inline void ExpressionViewClose(LatticeTrinaryExpression<Op, T1, T2, T3> &expr)
 // Unary operators and funcs
 ////////////////////////////////////////////
 #define GridUnopClass(name, ret)					\
+  template <class arg>							\
   struct name {								\
-    template<class _arg> static auto accelerator_inline func(const _arg a) -> decltype(ret) { return ret; } \
+    static auto accelerator_inline func(const arg a) -> decltype(ret) { return ret; } \
   };
 
 GridUnopClass(UnarySub, -a);
@@ -348,6 +285,8 @@ GridUnopClass(UnaryTrace, trace(a));
 GridUnopClass(UnaryTranspose, transpose(a));
 GridUnopClass(UnaryTa, Ta(a));
 GridUnopClass(UnaryProjectOnGroup, ProjectOnGroup(a));
+GridUnopClass(UnaryReal, real(a));
+GridUnopClass(UnaryImag, imag(a));
 GridUnopClass(UnaryToReal, toReal(a));
 GridUnopClass(UnaryToComplex, toComplex(a));
 GridUnopClass(UnaryTimesI, timesI(a));
@@ -366,10 +305,10 @@ GridUnopClass(UnaryExp, exp(a));
 // Binary operators
 ////////////////////////////////////////////
 #define GridBinOpClass(name, combination)			\
+  template <class left, class right>				\
   struct name {							\
-    template <class _left, class _right>			\
     static auto accelerator_inline				\
-    func(const _left &lhs, const _right &rhs)			\
+    func(const left &lhs, const right &rhs)			\
       -> decltype(combination) const				\
     {								\
       return combination;					\
@@ -389,10 +328,10 @@ GridBinOpClass(BinaryOrOr, lhs || rhs);
 // Trinary conditional op
 ////////////////////////////////////////////////////
 #define GridTrinOpClass(name, combination)				\
+  template <class predicate, class left, class right>			\
   struct name {								\
-    template <class _predicate,class _left, class _right>		\
     static auto accelerator_inline					\
-    func(const _predicate &pred, const _left &lhs, const _right &rhs)	\
+    func(const predicate &pred, const left &lhs, const right &rhs)	\
       -> decltype(combination) const					\
     {									\
       return combination;						\
@@ -400,17 +339,17 @@ GridBinOpClass(BinaryOrOr, lhs || rhs);
   };
 
 GridTrinOpClass(TrinaryWhere,
-		(predicatedWhere<
-		 typename std::remove_reference<_predicate>::type, 
-		 typename std::remove_reference<_left>::type,
-		 typename std::remove_reference<_right>::type>(pred, lhs,rhs)));
+		(predicatedWhere<predicate, 
+		 typename std::remove_reference<left>::type,
+		 typename std::remove_reference<right>::type>(pred, lhs,rhs)));
 
 ////////////////////////////////////////////
 // Operator syntactical glue
 ////////////////////////////////////////////
-#define GRID_UNOP(name)   name
-#define GRID_BINOP(name)  name
-#define GRID_TRINOP(name) name
+
+#define GRID_UNOP(name)   name<decltype(eval(0, arg))>
+#define GRID_BINOP(name)  name<decltype(eval(0, lhs)), decltype(eval(0, rhs))>
+#define GRID_TRINOP(name) name<decltype(eval(0, pred)), decltype(eval(0, lhs)), decltype(eval(0, rhs))>
 
 #define GRID_DEF_UNOP(op, name)						\
   template <typename T1, typename std::enable_if<is_lattice<T1>::value||is_lattice_expr<T1>::value,T1>::type * = nullptr> \
@@ -462,6 +401,8 @@ GRID_DEF_UNOP(trace, UnaryTrace);
 GRID_DEF_UNOP(transpose, UnaryTranspose);
 GRID_DEF_UNOP(Ta, UnaryTa);
 GRID_DEF_UNOP(ProjectOnGroup, UnaryProjectOnGroup);
+GRID_DEF_UNOP(real, UnaryReal);
+GRID_DEF_UNOP(imag, UnaryImag);
 GRID_DEF_UNOP(toReal, UnaryToReal);
 GRID_DEF_UNOP(toComplex, UnaryToComplex);
 GRID_DEF_UNOP(timesI, UnaryTimesI);
@@ -494,36 +435,29 @@ GRID_DEF_TRINOP(where, TrinaryWhere);
 /////////////////////////////////////////////////////////////
 template <class Op, class T1>
 auto closure(const LatticeUnaryExpression<Op, T1> &expr)
-  -> Lattice<decltype(expr.op.func(vecEval(0, expr.arg1)))> 
+  -> Lattice<decltype(expr.op.func(eval(0, expr.arg1)))> 
 {
-  Lattice<decltype(expr.op.func(vecEval(0, expr.arg1)))> ret(expr);
+  Lattice<decltype(expr.op.func(eval(0, expr.arg1)))> ret(expr);
   return ret;
 }
 template <class Op, class T1, class T2>
 auto closure(const LatticeBinaryExpression<Op, T1, T2> &expr)
-  -> Lattice<decltype(expr.op.func(vecEval(0, expr.arg1),vecEval(0, expr.arg2)))> 
+  -> Lattice<decltype(expr.op.func(eval(0, expr.arg1),eval(0, expr.arg2)))> 
 {
-  Lattice<decltype(expr.op.func(vecEval(0, expr.arg1),vecEval(0, expr.arg2)))> ret(expr);
+  Lattice<decltype(expr.op.func(eval(0, expr.arg1),eval(0, expr.arg2)))> ret(expr);
   return ret;
 }
 template <class Op, class T1, class T2, class T3>
 auto closure(const LatticeTrinaryExpression<Op, T1, T2, T3> &expr)
-  -> Lattice<decltype(expr.op.func(vecEval(0, expr.arg1),
-				   vecEval(0, expr.arg2),
-				   vecEval(0, expr.arg3)))> 
+  -> Lattice<decltype(expr.op.func(eval(0, expr.arg1),
+				   eval(0, expr.arg2),
+				   eval(0, expr.arg3)))> 
 {
-  Lattice<decltype(expr.op.func(vecEval(0, expr.arg1),
-				vecEval(0, expr.arg2),
-			        vecEval(0, expr.arg3)))>  ret(expr);
+  Lattice<decltype(expr.op.func(eval(0, expr.arg1),
+				eval(0, expr.arg2),
+				eval(0, expr.arg3)))>  ret(expr);
   return ret;
 }
-#define EXPRESSION_CLOSURE(function)					\
-  template<class Expression,typename std::enable_if<is_lattice_expr<Expression>::value,void>::type * = nullptr> \
-    auto function(Expression &expr) -> decltype(function(closure(expr))) \
-  {									\
-    return function(closure(expr));					\
-  }
-
 
 #undef GRID_UNOP
 #undef GRID_BINOP

Grid/lattice/Lattice_base.h

@@ -123,9 +123,9 @@ public:
     auto exprCopy = expr;
     ExpressionViewOpen(exprCopy);
     auto me  = View(AcceleratorWriteDiscard);
-    accelerator_for(ss,me.size(),vobj::Nsimd(),{
+    accelerator_for(ss,me.size(),1,{
       auto tmp = eval(ss,exprCopy);
-      coalescedWrite(me[ss],tmp);
+      vstream(me[ss],tmp);
     });
     me.ViewClose();
     ExpressionViewClose(exprCopy);
@@ -146,9 +146,9 @@ public:
     auto exprCopy = expr;
     ExpressionViewOpen(exprCopy);
     auto me  = View(AcceleratorWriteDiscard);
-    accelerator_for(ss,me.size(),vobj::Nsimd(),{
+    accelerator_for(ss,me.size(),1,{
       auto tmp = eval(ss,exprCopy);
-      coalescedWrite(me[ss],tmp);
+      vstream(me[ss],tmp);
     });
     me.ViewClose();
     ExpressionViewClose(exprCopy);
@@ -168,9 +168,9 @@ public:
     auto exprCopy = expr;
     ExpressionViewOpen(exprCopy);
     auto me  = View(AcceleratorWriteDiscard);
-    accelerator_for(ss,me.size(),vobj::Nsimd(),{
+    accelerator_for(ss,me.size(),1,{
       auto tmp = eval(ss,exprCopy);
-      coalescedWrite(me[ss],tmp);
+      vstream(me[ss],tmp);
     });
     me.ViewClose();
     ExpressionViewClose(exprCopy);

Grid/lattice/Lattice_basis.h

@@ -54,34 +54,13 @@ void basisRotate(VField &basis,Matrix& Qt,int j0, int j1, int k0,int k1,int Nm)
   typedef decltype(basis[0].View(AcceleratorRead)) View;
 
   Vector<View> basis_v; basis_v.reserve(basis.size());
-  typedef typename std::remove_reference<decltype(basis_v[0][0])>::type vobj;
-  typedef typename std::remove_reference<decltype(Qt(0,0))>::type Coeff_t;
   GridBase* grid = basis[0].Grid();
       
   for(int k=0;k<basis.size();k++){
     basis_v.push_back(basis[k].View(AcceleratorWrite));
   }
 
-#if ( (!defined(GRID_SYCL)) && (!defined(GRID_CUDA)) && (!defined(GRID_HIP)) )
-  int max_threads = thread_max();
-  Vector < vobj > Bt(Nm * max_threads);
-  thread_region
-    {
-      vobj* B = &Bt[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
+
   View *basis_vp = &basis_v[0];
 
   int nrot = j1-j0;
@@ -91,12 +70,14 @@ void basisRotate(VField &basis,Matrix& Qt,int j0, int j1, int k0,int k1,int Nm)
   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<Coeff_t> Qt_jv(Nm*Nm);
-  Coeff_t *Qt_p = & Qt_jv[0];
+  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;
@@ -137,7 +118,6 @@ void basisRotate(VField &basis,Matrix& Qt,int j0, int j1, int k0,int k1,int Nm)
 	coalescedWrite(basis_v[jj][sss],coalescedRead(Bp[ss*nrot+j]));
       });
   }
-#endif
 
   for(int k=0;k<basis.size();k++) basis_v[k].ViewClose();
 }

Grid/lattice/Lattice_comparison.h

@@ -42,6 +42,34 @@ NAMESPACE_BEGIN(Grid);
 
 typedef iScalar<vInteger> vPredicate ;
 
+/*
+template <class iobj, class vobj, class robj> accelerator_inline 
+vobj predicatedWhere(const iobj &predicate, const vobj &iftrue, const robj &iffalse) 
+{
+  typename std::remove_const<vobj>::type ret;
+
+  typedef typename vobj::scalar_object scalar_object;
+  typedef typename vobj::scalar_type scalar_type;
+  typedef typename vobj::vector_type vector_type;
+
+  const int Nsimd = vobj::vector_type::Nsimd();
+
+  ExtractBuffer<Integer> mask(Nsimd);
+  ExtractBuffer<scalar_object> truevals(Nsimd);
+  ExtractBuffer<scalar_object> falsevals(Nsimd);
+
+  extract(iftrue, truevals);
+  extract(iffalse, falsevals);
+  extract<vInteger, Integer>(TensorRemove(predicate), mask);
+
+  for (int s = 0; s < Nsimd; s++) {
+    if (mask[s]) falsevals[s] = truevals[s];
+  }
+
+  merge(ret, falsevals);
+  return ret;
+}
+*/
 //////////////////////////////////////////////////////////////////////////
 // compare lattice to lattice
 //////////////////////////////////////////////////////////////////////////

Grid/lattice/Lattice_peekpoke.h

@@ -182,14 +182,6 @@ inline void peekLocalSite(sobj &s,const LatticeView<vobj> &l,Coordinate &site)
       
   return;
 };
-template<class vobj,class sobj>
-inline void peekLocalSite(sobj &s,const Lattice<vobj> &l,Coordinate &site)
-{
-  autoView(lv,l,CpuRead);
-  peekLocalSite(s,lv,site);
-  return;
-};
-
 // Must be CPU write view
 template<class vobj,class sobj>
 inline void pokeLocalSite(const sobj &s,LatticeView<vobj> &l,Coordinate &site)
@@ -218,14 +210,6 @@ inline void pokeLocalSite(const sobj &s,LatticeView<vobj> &l,Coordinate &site)
   return;
 };
 
-template<class vobj,class sobj>
-inline void pokeLocalSite(const sobj &s, Lattice<vobj> &l,Coordinate &site)
-{
-  autoView(lv,l,CpuWrite);
-  pokeLocalSite(s,lv,site);
-  return;
-};
-
 NAMESPACE_END(Grid);
 #endif
 

Grid/lattice/Lattice_real_imag.h

@@ -1,79 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/lattice/Lattice_reality.h
-
-    Copyright (C) 2015
-
-Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: neo <cossu@post.kek.jp>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#ifndef GRID_LATTICE_REAL_IMAG_H
-#define GRID_LATTICE_REAL_IMAG_H
-
-
-// FIXME .. this is the sector of the code 
-// I am most worried about the directions
-// The choice of burying complex in the SIMD
-// is making the use of "real" and "imag" very cumbersome
-
-NAMESPACE_BEGIN(Grid);
-
-template<class vobj> inline Lattice<vobj> real(const Lattice<vobj> &lhs){
-  Lattice<vobj> ret(lhs.Grid());
-
-  autoView( lhs_v, lhs, AcceleratorRead);
-  autoView( ret_v, ret, AcceleratorWrite);
-
-  ret.Checkerboard()=lhs.Checkerboard();
-  accelerator_for( ss, lhs_v.size(), 1, {
-    ret_v[ss] =real(lhs_v[ss]);
-  });
-  return ret;
-};
-template<class vobj> inline Lattice<vobj> imag(const Lattice<vobj> &lhs){
-  Lattice<vobj> ret(lhs.Grid());
-
-  autoView( lhs_v, lhs, AcceleratorRead);
-  autoView( ret_v, ret, AcceleratorWrite);
-
-  ret.Checkerboard()=lhs.Checkerboard();
-  accelerator_for( ss, lhs_v.size(), 1, {
-    ret_v[ss] =imag(lhs_v[ss]);
-  });
-  return ret;
-};
-
-template<class Expression,typename std::enable_if<is_lattice_expr<Expression>::value,void>::type * = nullptr> 
-  auto real(const Expression &expr) -> decltype(real(closure(expr)))		
-{									
-  return real(closure(expr));					
-}
-template<class Expression,typename std::enable_if<is_lattice_expr<Expression>::value,void>::type * = nullptr> 
-  auto imag(const Expression &expr) -> decltype(imag(closure(expr)))		
-{									
-  return imag(closure(expr));					
-}
-
-NAMESPACE_END(Grid);
-
-#endif

Grid/lattice/Lattice_transfer.h

@@ -240,8 +240,6 @@ template<class vobj,class vobj2,class CComplex>
   autoView( fineX_  , fineX, AcceleratorRead);
   autoView( fineY_  , fineY, AcceleratorRead);
   autoView( coarseA_, coarseA, AcceleratorRead);
-  Coordinate fine_rdimensions = fine->_rdimensions;
-  Coordinate coarse_rdimensions = coarse->_rdimensions;
 
   accelerator_for(sf, fine->oSites(), CComplex::Nsimd(), {
 
@@ -249,9 +247,9 @@ template<class vobj,class vobj2,class CComplex>
       Coordinate coor_c(_ndimension);
       Coordinate coor_f(_ndimension);
 
-      Lexicographic::CoorFromIndex(coor_f,sf,fine_rdimensions);
+      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);
+      Lexicographic::IndexFromCoor(coor_c,sc,coarse->_rdimensions);
 
       // z = A x + y
 #ifdef GRID_SIMT
@@ -355,14 +353,11 @@ inline void blockSum(Lattice<vobj> &coarseData,const Lattice<vobj> &fineData)
   autoView( coarseData_ , coarseData, AcceleratorWrite);
   autoView( fineData_   , fineData, AcceleratorRead);
 
-  Coordinate fine_rdimensions = fine->_rdimensions;
-  Coordinate coarse_rdimensions = coarse->_rdimensions;
-  
   accelerator_for(sc,coarse->oSites(),1,{
 
       // One thread per sub block
       Coordinate coor_c(_ndimension);
-      Lexicographic::CoorFromIndex(coor_c,sc,coarse_rdimensions);  // Block coordinate
+      Lexicographic::CoorFromIndex(coor_c,sc,coarse->_rdimensions);  // Block coordinate
       coarseData_[sc]=Zero();
 
       for(int sb=0;sb<blockVol;sb++){
@@ -372,7 +367,7 @@ inline void blockSum(Lattice<vobj> &coarseData,const Lattice<vobj> &fineData)
 	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);
+	Lexicographic::IndexFromCoor(coor_f,sf,fine->_rdimensions);
 
 	coarseData_[sc]=coarseData_[sc]+fineData_[sf];
       }

Grid/lattice/Lattice_view.h

@@ -52,7 +52,6 @@ public:
 // 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
 /////////////////////////////////////////////////////////////////////////////////////////
-#undef LATTICE_BOUNDS_CHECK
 template<class vobj> 
 class LatticeView : public LatticeAccelerator<vobj>
 {
@@ -62,36 +61,14 @@ public:
   void * cpu_ptr;
 #ifdef GRID_SIMT
   accelerator_inline const typename vobj::scalar_object operator()(size_t i) const { 
-#ifdef LATTICE_BOUNDS_CHECK
-    assert(i<this->_odata_size);
-    assert(i>=0);
-#endif
     return coalescedRead(this->_odata[i]); 
   }
 #else 
-  accelerator_inline const vobj & operator()(size_t i) const {
-#ifdef LATTICE_BOUNDS_CHECK
-    assert(i<this->_odata_size);
-    assert(i>=0);
-#endif
-    return this->_odata[i];
-  }
+  accelerator_inline const vobj & operator()(size_t i) const { return this->_odata[i]; }
 #endif
 
-  accelerator_inline const vobj & operator[](size_t i) const { 
-#ifdef LATTICE_BOUNDS_CHECK
-    assert(i<this->_odata_size);
-    assert(i>=0);
-#endif
-    return this->_odata[i]; 
-  };
-  accelerator_inline vobj       & operator[](size_t i)       { 
-#ifdef LATTICE_BOUNDS_CHECK
-    assert(i<this->_odata_size);
-    assert(i>=0);
-#endif
-    return this->_odata[i]; 
-  };
+  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; };

Grid/qcd/QCD.h

@@ -77,16 +77,9 @@ const int SpinorIndex = 2;
 template<typename T> struct isSpinor {
   static constexpr bool value = (SpinorIndex==T::TensorLevel);
 };
-const int CoarseIndex = 4;
-template<typename T> struct isCoarsened {
-  static constexpr bool value = (CoarseIndex<=T::TensorLevel);
-};
 template <typename T> using IfSpinor    = Invoke<std::enable_if< isSpinor<T>::value,int> > ;
 template <typename T> using IfNotSpinor = Invoke<std::enable_if<!isSpinor<T>::value,int> > ;
 
-template <typename T> using IfCoarsened    = Invoke<std::enable_if< isCoarsened<T>::value,int> > ;
-template <typename T> using IfNotCoarsened = Invoke<std::enable_if<!isCoarsened<T>::value,int> > ;
-
 // ChrisK very keen to add extra space for Gparity doubling.
 //
 // Also add domain wall index, in a way where Wilson operator 

Grid/qcd/action/fermion/FermionOperator.h

@@ -89,8 +89,7 @@ public:
   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 std::vector<int> Directions(void)   =0;
-  virtual std::vector<int> Displacements(void)=0;
+
 
   virtual void  MomentumSpacePropagator(FermionField &out,const FermionField &in,RealD _m,std::vector<double> twist) { assert(0);};
 

Grid/qcd/action/fermion/ImprovedStaggeredFermion.h

@@ -44,9 +44,6 @@ public:
   INHERIT_IMPL_TYPES(Impl);
   typedef StaggeredKernels<Impl> Kernels;
 
-  virtual std::vector<int> Directions(void)   { return this->directions; };
-  virtual std::vector<int> Displacements(void){ return this->displacements;};
-
   FermionField _tmp;
   FermionField &tmp(void) { return _tmp; }
 

Grid/qcd/action/fermion/ImprovedStaggeredFermion5D.h

@@ -49,9 +49,6 @@ public:
   INHERIT_IMPL_TYPES(Impl);
   typedef StaggeredKernels<Impl> Kernels;
 
-  virtual std::vector<int> Directions(void)   { return this->directions; };
-  virtual std::vector<int> Displacements(void){ return this->displacements;};
-
   FermionField _tmp;
   FermionField &tmp(void) { return _tmp; }
 
@@ -211,7 +208,7 @@ public:
   LebesgueOrder LebesgueEvenOdd;
     
   // Comms buffer
-  //  std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> >  comm_buf;
+  std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> >  comm_buf;
     
   ///////////////////////////////////////////////////////////////
   // Conserved current utilities

Grid/qcd/action/fermion/NaiveStaggeredFermion.h

@@ -47,9 +47,6 @@ public:
   FermionField _tmp;
   FermionField &tmp(void) { return _tmp; }
 
-  virtual std::vector<int> Directions(void)   { return this->directions; };
-  virtual std::vector<int> Displacements(void){ return this->displacements;};
-
   ////////////////////////////////////////
   // Performance monitoring
   ////////////////////////////////////////

Grid/qcd/action/fermion/WilsonFermion.h

@@ -50,22 +50,19 @@ public:
   double, nu);
 
   WilsonAnisotropyCoefficients():
-    isAnisotropic(false),
-    t_direction(Nd-1),
-    xi_0(1.0),
+    isAnisotropic(false), 
+    t_direction(Nd-1), 
+    xi_0(1.0), 
     nu(1.0){}
 };
 
 template <class Impl>
-class WilsonFermion : public WilsonKernels<Impl>, public WilsonFermionStatic
+class WilsonFermion : public WilsonKernels<Impl>, public WilsonFermionStatic 
 {
 public:
   INHERIT_IMPL_TYPES(Impl);
   typedef WilsonKernels<Impl> Kernels;
 
-  virtual std::vector<int> Directions(void)   { return this->directions; };
-  virtual std::vector<int> Displacements(void){ return this->displacements;};
-
   ///////////////////////////////////////////////////////////////
   // Implement the abstract base
   ///////////////////////////////////////////////////////////////
@@ -77,20 +74,6 @@ public:
   FermionField _tmp;
   FermionField &tmp(void) { return _tmp; }
 
-  void Report(void);
-  void ZeroCounters(void);
-  double DhopCalls;
-  double DhopCommTime;
-  double DhopComputeTime;
-  double DhopComputeTime2;
-  double DhopFaceTime;
-  double DhopTotalTime;
-
-  double DerivCalls;
-  double DerivCommTime;
-  double DerivComputeTime;
-  double DerivDhopComputeTime;
-
   //////////////////////////////////////////////////////////////////
   // override multiply; cut number routines if pass dagger argument
   // and also make interface more uniformly consistent
@@ -155,7 +138,7 @@ public:
   // Constructor
   WilsonFermion(GaugeField &_Umu, GridCartesian &Fgrid,
                 GridRedBlackCartesian &Hgrid, RealD _mass,
-                const ImplParams &p = ImplParams(),
+                const ImplParams &p = ImplParams(), 
                 const WilsonAnisotropyCoefficients &anis = WilsonAnisotropyCoefficients() );
 
   // DoubleStore impl dependent
@@ -187,9 +170,9 @@ public:
 
   LebesgueOrder Lebesgue;
   LebesgueOrder LebesgueEvenOdd;
-
+  
   WilsonAnisotropyCoefficients anisotropyCoeff;
-
+  
   ///////////////////////////////////////////////////////////////
   // Conserved current utilities
   ///////////////////////////////////////////////////////////////
@@ -203,7 +186,7 @@ public:
                            PropagatorField &q_out,
                            PropagatorField &phys_src,
                            Current curr_type,
-                           unsigned int mu,
+                           unsigned int mu, 
                            unsigned int tmin,
 			   unsigned int tmax,
 			   ComplexField &lattice_cmplx);
@@ -213,3 +196,5 @@ typedef WilsonFermion<WilsonImplF> WilsonFermionF;
 typedef WilsonFermion<WilsonImplD> WilsonFermionD;
 
 NAMESPACE_END(Grid);
+
+

Grid/qcd/action/fermion/WilsonFermion5D.h

@@ -72,9 +72,6 @@ public:
   typedef WilsonKernels<Impl> Kernels;
   PmuStat stat;
 
-  virtual std::vector<int> Directions(void)   { return this->directions; };
-  virtual std::vector<int> Displacements(void){ return this->displacements;};
-
   FermionField _tmp;
   FermionField &tmp(void) { return _tmp; }
 
@@ -218,7 +215,7 @@ public:
   LebesgueOrder LebesgueEvenOdd;
     
   // Comms buffer
-  //  std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> >  comm_buf;
+  std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> >  comm_buf;
 
 
 };

Grid/qcd/action/fermion/g5HermitianLinop.h

@@ -79,8 +79,6 @@ public:
     _Mat.M(in,tmp);
     G5R5(out,tmp);
   }
-  virtual std::vector<int> Directions(void)   { return _Mat.Directions();};
-  virtual std::vector<int> Displacements(void){ return _Mat.Displacements();};
 };
 
 
@@ -129,8 +127,6 @@ public:
     _Mat.M(in,tmp);
     out=g5*tmp;
   }
-  virtual std::vector<int> Directions(void)   { return _Mat.Directions();};
-  virtual std::vector<int> Displacements(void){ return _Mat.Displacements();};
 };
 
 NAMESPACE_END(Grid);

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

@@ -133,14 +133,14 @@ void WilsonCloverFermion<Impl>::ImportGauge(const GaugeField &_Umu)
   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>

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

@@ -43,7 +43,7 @@ WilsonFermion<Impl>::WilsonFermion(GaugeField &_Umu, GridCartesian &Fgrid,
                                    GridRedBlackCartesian &Hgrid, RealD _mass,
                                    const ImplParams &p,
                                    const WilsonAnisotropyCoefficients &anis)
-  :
+  : 
     Kernels(p),
     _grid(&Fgrid),
     _cbgrid(&Hgrid),
@@ -75,93 +75,8 @@ WilsonFermion<Impl>::WilsonFermion(GaugeField &_Umu, GridCartesian &Fgrid,
   StencilOdd.BuildSurfaceList(1,vol4);
 }
 
-template<class Impl>
-void WilsonFermion<Impl>::Report(void)
-{
-  RealD NP = _grid->_Nprocessors;
-  RealD NN = _grid->NodeCount();
-  RealD volume = 1;
-  Coordinate latt = _grid->GlobalDimensions();
-  for(int mu=0;mu<Nd;mu++) volume=volume*latt[mu];
-
-  if ( DhopCalls > 0 ) {
-    std::cout << GridLogMessage << "#### Dhop calls report " << std::endl;
-    std::cout << GridLogMessage << "WilsonFermion Number of DhopEO Calls   : " << DhopCalls   << std::endl;
-    std::cout << GridLogMessage << "WilsonFermion TotalTime   /Calls        : " << DhopTotalTime   / DhopCalls << " us" << std::endl;
-    std::cout << GridLogMessage << "WilsonFermion CommTime    /Calls        : " << DhopCommTime    / DhopCalls << " us" << std::endl;
-    std::cout << GridLogMessage << "WilsonFermion FaceTime    /Calls        : " << DhopFaceTime    / DhopCalls << " us" << std::endl;
-    std::cout << GridLogMessage << "WilsonFermion ComputeTime1/Calls        : " << DhopComputeTime / DhopCalls << " us" << std::endl;
-    std::cout << GridLogMessage << "WilsonFermion ComputeTime2/Calls        : " << DhopComputeTime2/ DhopCalls << " us" << std::endl;
-
-    // Average the compute time
-    _grid->GlobalSum(DhopComputeTime);
-    DhopComputeTime/=NP;
-    RealD mflops = 1320*volume*DhopCalls/DhopComputeTime/2; // 2 for red black counting
-    std::cout << GridLogMessage << "Average mflops/s per call                : " << mflops << std::endl;
-    std::cout << GridLogMessage << "Average mflops/s per call per rank       : " << mflops/NP << std::endl;
-    std::cout << GridLogMessage << "Average mflops/s per call per node       : " << mflops/NN << std::endl;
-
-    RealD Fullmflops = 1320*volume*DhopCalls/(DhopTotalTime)/2; // 2 for red black counting
-    std::cout << GridLogMessage << "Average mflops/s per call (full)         : " << Fullmflops << std::endl;
-    std::cout << GridLogMessage << "Average mflops/s per call per rank (full): " << Fullmflops/NP << std::endl;
-    std::cout << GridLogMessage << "Average mflops/s per call per node (full): " << Fullmflops/NN << std::endl;
-
-   }
-
-  if ( DerivCalls > 0 ) {
-    std::cout << GridLogMessage << "#### Deriv calls report "<< std::endl;
-    std::cout << GridLogMessage << "WilsonFermion Number of Deriv Calls    : " <<DerivCalls <<std::endl;
-    std::cout << GridLogMessage << "WilsonFermion CommTime/Calls           : " <<DerivCommTime/DerivCalls<<" us" <<std::endl;
-    std::cout << GridLogMessage << "WilsonFermion ComputeTime/Calls        : " <<DerivComputeTime/DerivCalls<<" us" <<std::endl;
-    std::cout << GridLogMessage << "WilsonFermion Dhop ComputeTime/Calls   : " <<DerivDhopComputeTime/DerivCalls<<" us" <<std::endl;
-
-    // how to count flops here?
-    RealD mflops = 144*volume*DerivCalls/DerivDhopComputeTime;
-    std::cout << GridLogMessage << "Average mflops/s per call               ? : " << mflops << std::endl;
-    std::cout << GridLogMessage << "Average mflops/s per call per node      ? : " << mflops/NP << std::endl;
-
-    // how to count flops here?
-    RealD Fullmflops = 144*volume*DerivCalls/(DerivDhopComputeTime+DerivCommTime)/2; // 2 for red black counting
-    std::cout << GridLogMessage << "Average mflops/s per call (full)        ? : " << Fullmflops << std::endl;
-    std::cout << GridLogMessage << "Average mflops/s per call per node (full) ? : " << Fullmflops/NP << std::endl;  }
-
-  if (DerivCalls > 0 || DhopCalls > 0){
-    std::cout << GridLogMessage << "WilsonFermion Stencil"    <<std::endl;  Stencil.Report();
-    std::cout << GridLogMessage << "WilsonFermion StencilEven"<<std::endl;  StencilEven.Report();
-    std::cout << GridLogMessage << "WilsonFermion StencilOdd" <<std::endl;  StencilOdd.Report();
-  }
-  if ( DhopCalls > 0){
-    std::cout << GridLogMessage << "WilsonFermion Stencil     Reporti()"    <<std::endl;  Stencil.Reporti(DhopCalls);
-    std::cout << GridLogMessage << "WilsonFermion StencilEven Reporti()"<<std::endl;  StencilEven.Reporti(DhopCalls);
-    std::cout << GridLogMessage << "WilsonFermion StencilOdd  Reporti()" <<std::endl;  StencilOdd.Reporti(DhopCalls);
-  }
-}
-
-template<class Impl>
-void WilsonFermion<Impl>::ZeroCounters(void) {
-  DhopCalls       = 0; // ok
-  DhopCommTime    = 0;
-  DhopComputeTime = 0;
-  DhopComputeTime2= 0;
-  DhopFaceTime    = 0;
-  DhopTotalTime   = 0;
-
-  DerivCalls       = 0; // ok
-  DerivCommTime    = 0;
-  DerivComputeTime = 0;
-  DerivDhopComputeTime = 0;
-
-  Stencil.ZeroCounters();
-  StencilEven.ZeroCounters();
-  StencilOdd.ZeroCounters();
-  Stencil.ZeroCountersi();
-  StencilEven.ZeroCountersi();
-  StencilOdd.ZeroCountersi();
-}
-
-
 template <class Impl>
-void WilsonFermion<Impl>::ImportGauge(const GaugeField &_Umu)
+void WilsonFermion<Impl>::ImportGauge(const GaugeField &_Umu) 
 {
   GaugeField HUmu(_Umu.Grid());
 
@@ -192,7 +107,7 @@ void WilsonFermion<Impl>::ImportGauge(const GaugeField &_Umu)
 /////////////////////////////
 
 template <class Impl>
-void 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);
@@ -200,7 +115,7 @@ void WilsonFermion<Impl>::M(const FermionField &in, FermionField &out)
 }
 
 template <class Impl>
-void 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);
@@ -208,7 +123,7 @@ void WilsonFermion<Impl>::Mdag(const FermionField &in, FermionField &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);
@@ -218,7 +133,7 @@ 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);
@@ -226,9 +141,9 @@ void WilsonFermion<Impl>::MeooeDag(const FermionField &in, FermionField &out)
     DhopOE(in, out, DaggerYes);
   }
 }
-
+  
 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);
@@ -236,80 +151,80 @@ void WilsonFermion<Impl>::Mooee(const FermionField &in, FermionField &out)
 }
 
 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);
 }
 template<class Impl>
 void WilsonFermion<Impl>::MomentumSpacePropagator(FermionField &out, const FermionField &in,RealD _m,std::vector<double> twist)
-{
+{  
   typedef typename FermionField::vector_type vector_type;
   typedef typename FermionField::scalar_type ScalComplex;
   typedef Lattice<iSinglet<vector_type> > LatComplex;
-
-  // what type LatticeComplex
+  
+  // what type LatticeComplex 
   conformable(_grid,out.Grid());
-
+  
   Gamma::Algebra Gmu [] = {
     Gamma::Algebra::GammaX,
     Gamma::Algebra::GammaY,
     Gamma::Algebra::GammaZ,
     Gamma::Algebra::GammaT
   };
-
+  
   Coordinate latt_size   = _grid->_fdimensions;
-
+  
   FermionField   num  (_grid); num  = Zero();
   LatComplex    wilson(_grid); wilson= Zero();
   LatComplex     one  (_grid); one = ScalComplex(1.0,0.0);
-
+  
   LatComplex denom(_grid); denom= Zero();
-  LatComplex kmu(_grid);
+  LatComplex kmu(_grid); 
   ScalComplex ci(0.0,1.0);
   // momphase = n * 2pi / L
   for(int mu=0;mu<Nd;mu++) {
-
+    
     LatticeCoordinate(kmu,mu);
-
+    
     RealD TwoPiL =  M_PI * 2.0/ latt_size[mu];
-
+    
     kmu = TwoPiL * kmu;
     kmu = kmu + TwoPiL * one * twist[mu];//momentum for twisted boundary conditions
-
+    
     wilson = wilson + 2.0*sin(kmu*0.5)*sin(kmu*0.5); // Wilson term
-
+    
     num = num - sin(kmu)*ci*(Gamma(Gmu[mu])*in);    // derivative term
-
+    
     denom=denom + sin(kmu)*sin(kmu);
   }
-
+  
   wilson = wilson + _m;     // 2 sin^2 k/2 + m
-
+  
   num   = num + wilson*in;     // -i gmu sin k + 2 sin^2 k/2 + m
-
+  
   denom= denom+wilson*wilson; // sin^2 k + (2 sin^2 k/2 + m)^2
-
+  
   denom= one/denom;
-
+  
   out = num*denom; // [ -i gmu sin k + 2 sin^2 k/2 + m] / [ sin^2 k + (2 sin^2 k/2 + m)^2 ]
-
+  
 }
-
+  
 
 ///////////////////////////////////
 // Internal
@@ -319,7 +234,6 @@ template <class Impl>
 void WilsonFermion<Impl>::DerivInternal(StencilImpl &st, DoubledGaugeField &U,
                                         GaugeField &mat, const FermionField &A,
                                         const FermionField &B, int dag) {
-  DerivCalls++;
   assert((dag == DaggerNo) || (dag == DaggerYes));
 
   Compressor compressor(dag);
@@ -328,11 +242,8 @@ void WilsonFermion<Impl>::DerivInternal(StencilImpl &st, DoubledGaugeField &U,
   FermionField Atilde(B.Grid());
   Atilde = A;
 
-  DerivCommTime-=usecond();
   st.HaloExchange(B, compressor);
-  DerivCommTime+=usecond();
 
-  DerivComputeTime-=usecond();
   for (int mu = 0; mu < Nd; mu++) {
     ////////////////////////////////////////////////////////////////////////
     // Flip gamma (1+g)<->(1-g) if dag
@@ -340,7 +251,6 @@ void WilsonFermion<Impl>::DerivInternal(StencilImpl &st, DoubledGaugeField &U,
     int gamma = mu;
     if (!dag) gamma += Nd;
 
-    DerivDhopComputeTime -= usecond();
     int Ls=1;
     Kernels::DhopDirKernel(st, U, st.CommBuf(), Ls, B.Grid()->oSites(), B, Btilde, mu, gamma);
 
@@ -348,13 +258,11 @@ void WilsonFermion<Impl>::DerivInternal(StencilImpl &st, DoubledGaugeField &U,
     // spin trace outer product
     //////////////////////////////////////////////////
     Impl::InsertForce4D(mat, Btilde, Atilde, mu);
-    DerivDhopComputeTime += usecond();
   }
-  DerivComputeTime += usecond();
 }
 
 template <class Impl>
-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());
@@ -366,13 +274,13 @@ 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)
   // Motivation: look at the SchurDiff operator
-
+  
   assert(V.Checkerboard() == Even);
   assert(U.Checkerboard() == Odd);
   mat.Checkerboard() = Odd;
@@ -381,7 +289,7 @@ 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());
@@ -395,7 +303,7 @@ void WilsonFermion<Impl>::DhopDerivEO(GaugeField &mat, const FermionField &U, co
 }
 
 template <class Impl>
-void WilsonFermion<Impl>::Dhop(const FermionField &in, FermionField &out, int dag)
+void WilsonFermion<Impl>::Dhop(const FermionField &in, FermionField &out, int dag) 
 {
   conformable(in.Grid(), _grid);  // verifies full grid
   conformable(in.Grid(), out.Grid());
@@ -406,7 +314,7 @@ void WilsonFermion<Impl>::Dhop(const FermionField &in, FermionField &out, int da
 }
 
 template <class Impl>
-void WilsonFermion<Impl>::DhopOE(const FermionField &in, FermionField &out, int dag)
+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
@@ -418,7 +326,7 @@ void WilsonFermion<Impl>::DhopOE(const FermionField &in, FermionField &out, int
 }
 
 template <class Impl>
-void WilsonFermion<Impl>::DhopEO(const FermionField &in, FermionField &out,int dag)
+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
@@ -430,18 +338,18 @@ void WilsonFermion<Impl>::DhopEO(const FermionField &in, FermionField &out,int d
 }
 
 template <class Impl>
-void WilsonFermion<Impl>::Mdir(const FermionField &in, FermionField &out, int dir, int disp)
+void WilsonFermion<Impl>::Mdir(const FermionField &in, FermionField &out, int dir, int disp) 
 {
   DhopDir(in, out, dir, disp);
 }
 template <class Impl>
-void WilsonFermion<Impl>::MdirAll(const FermionField &in, std::vector<FermionField> &out)
+void WilsonFermion<Impl>::MdirAll(const FermionField &in, std::vector<FermionField> &out) 
 {
   DhopDirAll(in, out);
 }
 
 template <class Impl>
-void WilsonFermion<Impl>::DhopDir(const FermionField &in, FermionField &out, int dir, int disp)
+void WilsonFermion<Impl>::DhopDir(const FermionField &in, FermionField &out, int dir, int disp) 
 {
   Compressor compressor(DaggerNo);
   Stencil.HaloExchange(in, compressor);
@@ -453,12 +361,12 @@ void WilsonFermion<Impl>::DhopDir(const FermionField &in, FermionField &out, int
   DhopDirCalc(in, out, dirdisp, gamma, DaggerNo);
 };
 template <class Impl>
-void WilsonFermion<Impl>::DhopDirAll(const FermionField &in, std::vector<FermionField> &out)
+void WilsonFermion<Impl>::DhopDirAll(const FermionField &in, std::vector<FermionField> &out) 
 {
   Compressor compressor(DaggerNo);
   Stencil.HaloExchange(in, compressor);
 
-  assert((out.size()==8)||(out.size()==9));
+  assert((out.size()==8)||(out.size()==9)); 
   for(int dir=0;dir<Nd;dir++){
     for(int disp=-1;disp<=1;disp+=2){
 
@@ -471,7 +379,7 @@ void WilsonFermion<Impl>::DhopDirAll(const FermionField &in, std::vector<Fermion
   }
 }
 template <class Impl>
-void WilsonFermion<Impl>::DhopDirCalc(const FermionField &in, FermionField &out,int dirdisp, int gamma, int dag)
+void WilsonFermion<Impl>::DhopDirCalc(const FermionField &in, FermionField &out,int dirdisp, int gamma, int dag) 
 {
   int Ls=1;
   uint64_t Nsite=in.oSites();
@@ -482,23 +390,22 @@ template <class Impl>
 void WilsonFermion<Impl>::DhopInternal(StencilImpl &st, LebesgueOrder &lo,
                                        DoubledGaugeField &U,
                                        const FermionField &in,
-                                       FermionField &out, int dag)
+                                       FermionField &out, int dag) 
 {
-  DhopTotalTime-=usecond();
 #ifdef GRID_OMP
   if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsAndCompute )
     DhopInternalOverlappedComms(st,lo,U,in,out,dag);
   else
-#endif
+#endif 
     DhopInternalSerial(st,lo,U,in,out,dag);
-  DhopTotalTime+=usecond();
+
 }
 
 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));
 
@@ -510,53 +417,38 @@ void WilsonFermion<Impl>::DhopInternalOverlappedComms(StencilImpl &st, LebesgueO
   /////////////////////////////
   std::vector<std::vector<CommsRequest_t> > requests;
   st.Prepare();
-  DhopFaceTime-=usecond();
   st.HaloGather(in,compressor);
-  DhopFaceTime+=usecond();
-
-  DhopCommTime -=usecond();
   st.CommunicateBegin(requests);
 
   /////////////////////////////
   // Overlap with comms
   /////////////////////////////
-  DhopFaceTime-=usecond();
   st.CommsMergeSHM(compressor);
-  DhopFaceTime+=usecond();
 
   /////////////////////////////
   // do the compute interior
   /////////////////////////////
   int Opt = WilsonKernelsStatic::Opt;
-  DhopComputeTime-=usecond();
   if (dag == DaggerYes) {
     Kernels::DhopDagKernel(Opt,st,U,st.CommBuf(),1,U.oSites(),in,out,1,0);
   } else {
     Kernels::DhopKernel(Opt,st,U,st.CommBuf(),1,U.oSites(),in,out,1,0);
-  }
-  DhopComputeTime+=usecond();
+  } 
 
   /////////////////////////////
   // Complete comms
   /////////////////////////////
   st.CommunicateComplete(requests);
-  DhopCommTime   +=usecond();
-
-  DhopFaceTime-=usecond();
   st.CommsMerge(compressor);
-  DhopFaceTime+=usecond();
 
   /////////////////////////////
   // do the compute exterior
   /////////////////////////////
-
-  DhopComputeTime2-=usecond();
   if (dag == DaggerYes) {
     Kernels::DhopDagKernel(Opt,st,U,st.CommBuf(),1,U.oSites(),in,out,0,1);
   } else {
     Kernels::DhopKernel(Opt,st,U,st.CommBuf(),1,U.oSites(),in,out,0,1);
   }
-  DhopComputeTime2+=usecond();
 };
 
 
@@ -564,28 +456,24 @@ 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();
   st.HaloExchange(in, compressor);
-  DhopCommTime+=usecond();
 
-  DhopComputeTime-=usecond();
   int Opt = WilsonKernelsStatic::Opt;
   if (dag == DaggerYes) {
     Kernels::DhopDagKernel(Opt,st,U,st.CommBuf(),1,U.oSites(),in,out);
   } else {
     Kernels::DhopKernel(Opt,st,U,st.CommBuf(),1,U.oSites(),in,out);
   }
-  DhopComputeTime+=usecond();
 };
 /*Change ends */
 
 /*******************************************************************************
  * Conserved current utilities for Wilson fermions, for contracting propagators
- * to make a conserved current sink or inserting the conserved current
+ * to make a conserved current sink or inserting the conserved current 
  * sequentially.
  ******************************************************************************/
 template <class Impl>
@@ -605,12 +493,12 @@ void WilsonFermion<Impl>::ContractConservedCurrent(PropagatorField &q_in_1,
 
 
 template <class Impl>
-void WilsonFermion<Impl>::SeqConservedCurrent(PropagatorField &q_in,
+void WilsonFermion<Impl>::SeqConservedCurrent(PropagatorField &q_in, 
                                               PropagatorField &q_out,
                                               PropagatorField &src,
                                               Current curr_type,
                                               unsigned int mu,
-                                              unsigned int tmin,
+                                              unsigned int tmin, 
                                               unsigned int tmax,
 					      ComplexField &lattice_cmplx)
 {

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

@@ -1,574 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-
-
-    Source file: ./lib/qcd/action/fermion/WilsonKernelsAsmA64FX.h
-
-    Copyright (C) 2020
-
-Author: Nils Meyer  <nils.meyer@ur.de>  Regensburg University
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#pragma once
-
-//#if defined(A64FXASM)
-#if defined(A64FX)
-
-// safety include
-#include <arm_sve.h>
-
-// undefine everything related to kernels
-#include <simd/Fujitsu_A64FX_undef.h>
-
-// enable A64FX body
-#define WILSONKERNELSASMBODYA64FX
-//#pragma message("A64FX Dslash: WilsonKernelsAsmBodyA64FX.h")
-
-    ///////////////////////////////////////////////////////////
-    // If we are A64FX specialise the single precision routine
-    ///////////////////////////////////////////////////////////
-#if defined(DSLASHINTRIN)
-//#pragma message ("A64FX Dslash: intrin")
-#include <simd/Fujitsu_A64FX_intrin_single.h>
-#else
-#pragma message ("A64FX Dslash: asm")
-#include <simd/Fujitsu_A64FX_asm_single.h>
-#endif
-
-/// Switch off the 5d vectorised code optimisations
-#undef DWFVEC5D
-
-/////////////////////////////////////////////////////////////////
-// XYZT vectorised, undag Kernel, single
-/////////////////////////////////////////////////////////////////
-#undef KERNEL_DAG
-#define INTERIOR_AND_EXTERIOR
-#undef INTERIOR
-#undef EXTERIOR
-template<> void
-WilsonKernels<WilsonImplF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<ZWilsonImplF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<WilsonImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<ZWilsonImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-#undef INTERIOR_AND_EXTERIOR
-#define INTERIOR
-#undef EXTERIOR
-template<> void
-WilsonKernels<WilsonImplF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<ZWilsonImplF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<WilsonImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<ZWilsonImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-#undef INTERIOR_AND_EXTERIOR
-#undef INTERIOR
-#define EXTERIOR
-template<> void
-WilsonKernels<WilsonImplF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<ZWilsonImplF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<WilsonImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<ZWilsonImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-
-/////////////////////////////////////////////////////////////////
-// XYZT vectorised, dag Kernel, single
-/////////////////////////////////////////////////////////////////
-#define KERNEL_DAG
-#define INTERIOR_AND_EXTERIOR
-#undef INTERIOR
-#undef EXTERIOR
-template<> void
-WilsonKernels<WilsonImplF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<ZWilsonImplF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<WilsonImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-#undef INTERIOR_AND_EXTERIOR
-#define INTERIOR
-#undef EXTERIOR
-template<> void
-WilsonKernels<WilsonImplF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<ZWilsonImplF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<WilsonImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-#undef INTERIOR_AND_EXTERIOR
-#undef INTERIOR
-#define EXTERIOR
-template<> void
-WilsonKernels<WilsonImplF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<ZWilsonImplF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<WilsonImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-
-// undefine
-#include <simd/Fujitsu_A64FX_undef.h>
-
-///////////////////////////////////////////////////////////
-// If we are A64FX specialise the double precision routine
-///////////////////////////////////////////////////////////
-
-#if defined(DSLASHINTRIN)
-#include <simd/Fujitsu_A64FX_intrin_double.h>
-#else
-#include <simd/Fujitsu_A64FX_asm_double.h>
-#endif
-
-// former KNL
-//#define MAYBEPERM(A,perm) if (perm) { A ; }
-//#define MULT_2SPIN(ptr,pf) MULT_ADDSUB_2SPIN(ptr,pf)
-//#define COMPLEX_SIGNS(isigns) vComplexD *isigns = &signsD[0];
-
-
-#define INTERIOR_AND_EXTERIOR
-#undef  INTERIOR
-#undef  EXTERIOR
-
-/////////////////////////////////////////////////////////////////
-// XYZT vectorised, undag Kernel, double
-/////////////////////////////////////////////////////////////////
-#undef KERNEL_DAG
-#define INTERIOR_AND_EXTERIOR
-#undef INTERIOR
-#undef EXTERIOR
-template<> void
-WilsonKernels<WilsonImplD>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<ZWilsonImplD>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<WilsonImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<ZWilsonImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-#undef INTERIOR_AND_EXTERIOR
-#define INTERIOR
-#undef EXTERIOR
-template<> void
-WilsonKernels<WilsonImplD>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<ZWilsonImplD>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<WilsonImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<ZWilsonImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-#undef INTERIOR_AND_EXTERIOR
-#undef INTERIOR
-#define EXTERIOR
-template<> void
-WilsonKernels<WilsonImplD>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<ZWilsonImplD>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<WilsonImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<ZWilsonImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-/////////////////////////////////////////////////////////////////
-// XYZT vectorised, dag Kernel, double
-/////////////////////////////////////////////////////////////////
-#define KERNEL_DAG
-#define INTERIOR_AND_EXTERIOR
-#undef INTERIOR
-#undef EXTERIOR
-template<> void
-WilsonKernels<WilsonImplD>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<ZWilsonImplD>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<WilsonImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-#undef INTERIOR_AND_EXTERIOR
-#define INTERIOR
-#undef EXTERIOR
-template<> void
-WilsonKernels<WilsonImplD>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<ZWilsonImplD>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<WilsonImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-#undef INTERIOR_AND_EXTERIOR
-#undef INTERIOR
-#define EXTERIOR
-template<> void
-WilsonKernels<WilsonImplD>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<ZWilsonImplD>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<WilsonImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-template<> void
-WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
-						int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
-#if defined (WILSONKERNELSASMBODYA64FX)
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
-#else
-#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
-#endif
-
-
-
-// undefs
-#undef WILSONKERNELSASMBODYA64FX
-#include <simd/Fujitsu_A64FX_undef.h>
-
-#endif //A64FXASM

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

@@ -1,380 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: WilsonKernelsAsmBodyA64FX.h
-
-    Copyright (C) 2020
-
-Author:  Nils Meyer  <nils.meyer@ur.de>  Regensburg University
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#ifdef KERNEL_DAG
-#define DIR0_PROJ    XP_PROJ
-#define DIR1_PROJ    YP_PROJ
-#define DIR2_PROJ    ZP_PROJ
-#define DIR3_PROJ    TP_PROJ
-#define DIR4_PROJ    XM_PROJ
-#define DIR5_PROJ    YM_PROJ
-#define DIR6_PROJ    ZM_PROJ
-#define DIR7_PROJ    TM_PROJ
-#define DIR0_RECON   XP_RECON
-#define DIR1_RECON   YP_RECON_ACCUM
-#define DIR2_RECON   ZP_RECON_ACCUM
-#define DIR3_RECON   TP_RECON_ACCUM
-#define DIR4_RECON   XM_RECON_ACCUM
-#define DIR5_RECON   YM_RECON_ACCUM
-#define DIR6_RECON   ZM_RECON_ACCUM
-#define DIR7_RECON   TM_RECON_ACCUM
-#else
-#define DIR0_PROJ    XM_PROJ
-#define DIR1_PROJ    YM_PROJ
-#define DIR2_PROJ    ZM_PROJ
-#define DIR3_PROJ    TM_PROJ
-#define DIR4_PROJ    XP_PROJ
-#define DIR5_PROJ    YP_PROJ
-#define DIR6_PROJ    ZP_PROJ
-#define DIR7_PROJ    TP_PROJ
-#define DIR0_RECON   XM_RECON
-#define DIR1_RECON   YM_RECON_ACCUM
-#define DIR2_RECON   ZM_RECON_ACCUM
-#define DIR3_RECON   TM_RECON_ACCUM
-#define DIR4_RECON   XP_RECON_ACCUM
-#define DIR5_RECON   YP_RECON_ACCUM
-#define DIR6_RECON   ZP_RECON_ACCUM
-#define DIR7_RECON   TP_RECON_ACCUM
-#endif
-
-//using namespace std;
-
-#undef SHOW
-//#define SHOW
-
-#undef WHERE
-
-#ifdef INTERIOR_AND_EXTERIOR
-#define WHERE "INT_AND_EXT"
-#endif
-
-#ifdef INTERIOR
-#define WHERE "INT"
-#endif
-
-#ifdef EXTERIOR
-#define WHERE "EXT"
-#endif
-
-//#pragma message("here")
-
-
-
-////////////////////////////////////////////////////////////////////////////////
-// Comms then compute kernel
-////////////////////////////////////////////////////////////////////////////////
-#ifdef INTERIOR_AND_EXTERIOR
-
-#define ASM_LEG(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON)			\
-      basep = st.GetPFInfo(nent,plocal); nent++;			\
-      if ( local ) {							            \
-    LOAD_CHIMU(base);                                       \
-    LOAD_TABLE(PERMUTE_DIR);                                \
-    PROJ;							                        \
-    MAYBEPERM(PERMUTE_DIR,perm);					        \
-      } else {								                \
-	LOAD_CHI(base);							                \
-      }									                    \
-      base = st.GetInfo(ptype,local,perm,NxtDir,ent,plocal); ent++;	\
-    MULT_2SPIN_1(Dir);					                    \
-    PREFETCH_CHIMU(base);                                   \
-    PREFETCH_CHIMU_L2(basep);                               \
-    /* PREFETCH_GAUGE_L1(NxtDir); */                        \
-    MULT_2SPIN_2;					                        \
-    if (s == 0) {                                           \
-      if ((Dir == 0) || (Dir == 4)) { PREFETCH_GAUGE_L2(Dir); } \
-    }                                                       \
-    RECON;								                    \
-
-#define ASM_LEG_XP(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON)	    \
-  base = st.GetInfo(ptype,local,perm,Dir,ent,plocal); ent++; \
-  PREFETCH1_CHIMU(base);						            \
-  ASM_LEG(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON)
-
-#define RESULT(base,basep) SAVE_RESULT(base,basep);
-
-#endif
-
-////////////////////////////////////////////////////////////////////////////////
-// Pre comms kernel -- prefetch like normal because it is mostly right
-////////////////////////////////////////////////////////////////////////////////
-#ifdef INTERIOR
-
-#define ASM_LEG(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON)			\
-      basep = st.GetPFInfo(nent,plocal); nent++;			\
-      if ( local ) {							            \
-    LOAD_CHIMU(base);                                       \
-    LOAD_TABLE(PERMUTE_DIR);                                \
-    PROJ;							                        \
-    MAYBEPERM(PERMUTE_DIR,perm);					        \
-      }else if ( st.same_node[Dir] ) {LOAD_CHI(base);}	    \
-      base = st.GetInfo(ptype,local,perm,NxtDir,ent,plocal); ent++;	\
-      if ( local || st.same_node[Dir] ) {				    \
-    MULT_2SPIN_1(Dir);					                    \
-    PREFETCH_CHIMU(base);                                   \
-    /* PREFETCH_GAUGE_L1(NxtDir); */                        \
-    MULT_2SPIN_2;					                        \
-    if (s == 0) {                                           \
-       if ((Dir == 0) || (Dir == 4)) { PREFETCH_GAUGE_L2(Dir); } \
-    }                                                       \
-    RECON;								                    \
-    PREFETCH_CHIMU_L2(basep);                               \
-      } else { PREFETCH_CHIMU(base); }								                    \
-
-#define ASM_LEG_XP(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON)			\
-  base = st.GetInfo(ptype,local,perm,Dir,ent,plocal); ent++;		\
-  PREFETCH1_CHIMU(base);						\
-  ASM_LEG(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON)
-
-#define RESULT(base,basep) SAVE_RESULT(base,basep);
-
-#endif
-////////////////////////////////////////////////////////////////////////////////
-// Post comms kernel
-////////////////////////////////////////////////////////////////////////////////
-#ifdef EXTERIOR
-
-
-#define ASM_LEG(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON)			\
-  base = st.GetInfo(ptype,local,perm,Dir,ent,plocal); ent++; \
-  if((!local)&&(!st.same_node[Dir]) ) {					    \
-    LOAD_CHI(base);							                \
-    MULT_2SPIN_1(Dir);					                    \
-    PREFETCH_CHIMU(base);                                   \
-    /* PREFETCH_GAUGE_L1(NxtDir); */                        \
-    MULT_2SPIN_2;					                        \
-    if (s == 0) {                                           \
-      if ((Dir == 0) || (Dir == 4)) { PREFETCH_GAUGE_L2(Dir); } \
-    }                                                       \
-    RECON;								                    \
-    nmu++;								                    \
-  }
-
-#define ASM_LEG_XP(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON)	    \
-  nmu=0;								                    \
-  base = st.GetInfo(ptype,local,perm,Dir,ent,plocal); ent++;\
-  if((!local)&&(!st.same_node[Dir]) ) {					    \
-    LOAD_CHI(base);							                \
-    MULT_2SPIN_1(Dir);					                    \
-    PREFETCH_CHIMU(base);                                   \
-    /* PREFETCH_GAUGE_L1(NxtDir); */                        \
-    MULT_2SPIN_2;					                        \
-    if (s == 0) {                                           \
-      if ((Dir == 0) || (Dir == 4)) { PREFETCH_GAUGE_L2(Dir); } \
-    }                                                       \
-    RECON;								                    \
-    nmu++;								                    \
-  }
-
-#define RESULT(base,basep) if (nmu){ ADD_RESULT(base,base);}
-
-#endif
-{
-  int nmu;
-  int local,perm, ptype;
-  uint64_t base;
-  uint64_t basep;
-  const uint64_t plocal =(uint64_t) & in[0];
-
-  MASK_REGS;
-  int nmax=U.oSites();
-  for(int site=0;site<Ns;site++) {
-#ifndef EXTERIOR
-    //    int sU =lo.Reorder(ssU);
-    int sU =ssU;
-    int ssn=ssU+1;     if(ssn>=nmax) ssn=0;
-    //    int sUn=lo.Reorder(ssn);
-    int sUn=ssn;
-    LOCK_GAUGE(0);
-#else
-    int sU =ssU;
-    int ssn=ssU+1;     if(ssn>=nmax) ssn=0;
-    int sUn=ssn;
-#endif
-    for(int s=0;s<Ls;s++) {
-      ss =sU*Ls+s;
-      ssn=sUn*Ls+s;
-      int  ent=ss*8;// 2*Ndim
-      int nent=ssn*8;
-
-      uint64_t delta_base, delta_base_p;
-
-   ASM_LEG_XP(Xp,Yp,PERMUTE_DIR3,DIR0_PROJ,DIR0_RECON);
-
-#ifdef SHOW
-      float rescale = 64. * 12.;
-      std::cout << "=================================================================" << std::endl;
-      std::cout << "ss = " << ss << "   ssn = " << ssn << std::endl;
-      std::cout << "sU = " << sU << "   ssU = " << ssU << std::endl;
-      std::cout << " " << std::endl;
-
-
-      std::cout << "Dir = " << Xp << "        "  << WHERE<< std::endl;
-
-      std::cout << "ent  nent  local  perm       = " << ent << "  " << nent << "  " << local << "  "  << perm << std::endl;
-      std::cout << "st.same_node[Dir] = " << st.same_node[Xp] << std::endl;
-      std::cout << "base              = " << (base - plocal)/rescale << std::endl;
-      std::cout << "Basep             = " << (basep - plocal)/rescale << std::endl;
-      //printf("U                 = %llu\n", (uint64_t)&[sU](Dir));
-      std::cout << "----------------------------------------------------" << std::endl;
-#endif
-
-      ASM_LEG(Yp,Zp,PERMUTE_DIR2,DIR1_PROJ,DIR1_RECON);
-
-#ifdef SHOW
-      std::cout << "Dir = " << Yp << "        "  << WHERE<< std::endl;
-
-      std::cout << "ent  nent  local  perm       = " << ent << "  " << nent << "  " << local << "  "  << perm << std::endl;
-      std::cout << "st.same_node[Dir] = " << st.same_node[Yp] << std::endl;
-      std::cout << "base              = " << (base - plocal)/rescale << std::endl;
-      std::cout << "Basep             = " << (basep - plocal)/rescale << std::endl;
-      //printf("U                 = %llu\n", (uint64_t)&[sU](Dir));
-      std::cout << "----------------------------------------------------" << std::endl;
-#endif
-
-      ASM_LEG(Zp,Tp,PERMUTE_DIR1,DIR2_PROJ,DIR2_RECON);
-
-#ifdef SHOW
-      std::cout << "Dir = " << Zp << "        "  << WHERE<< std::endl;
-
-      std::cout << "ent  nent  local  perm       = " << ent << "  " << nent << "  " << local << "  "  << perm << std::endl;
-      std::cout << "st.same_node[Dir] = " << st.same_node[Zp] << std::endl;
-      std::cout << "base              = " << (base - plocal)/rescale << std::endl;
-      std::cout << "Basep             = " << (basep - plocal)/rescale << std::endl;
-      //printf("U                 = %llu\n", (uint64_t)&[sU](Dir));
-      std::cout << "----------------------------------------------------" << std::endl;
-#endif
-
-      ASM_LEG(Tp,Xm,PERMUTE_DIR0,DIR3_PROJ,DIR3_RECON);
-
-#ifdef SHOW
-      std::cout << "Dir = " << Tp << "        "  << WHERE<< std::endl;
-
-      std::cout << "ent  nent  local  perm       = " << ent << "  " << nent << "  " << local << "  "  << perm << std::endl;
-      std::cout << "st.same_node[Dir] = " << st.same_node[Tp] << std::endl;
-      std::cout << "base              = " << (base - plocal)/rescale << std::endl;
-      std::cout << "Basep             = " << (basep - plocal)/rescale << std::endl;
-      //printf("U                 = %llu\n", (uint64_t)&[sU](Dir));
-      std::cout << "----------------------------------------------------" << std::endl;
-#endif
-
-      ASM_LEG(Xm,Ym,PERMUTE_DIR3,DIR4_PROJ,DIR4_RECON);
-
-#ifdef SHOW
-      std::cout << "Dir = " << Xm << "        "  << WHERE<< std::endl;
-
-      std::cout << "ent  nent  local  perm       = " << ent << "  " << nent << "  " << local << "  "  << perm << std::endl;
-      std::cout << "st.same_node[Dir] = " << st.same_node[Xm] << std::endl;
-      std::cout << "base              = " << (base - plocal)/rescale << std::endl;
-      std::cout << "Basep             = " << (basep - plocal)/rescale << std::endl;
-      //printf("U                 = %llu\n", (uint64_t)&[sU](Dir));
-      std::cout << "----------------------------------------------------" << std::endl;
-#endif
-
-      ASM_LEG(Ym,Zm,PERMUTE_DIR2,DIR5_PROJ,DIR5_RECON);
-
-#ifdef SHOW
-      std::cout << "Dir = " << Ym << "        "  << WHERE<< std::endl;
-
-      std::cout << "ent  nent  local  perm       = " << ent << "  " << nent << "  " << local << "  "  << perm << std::endl;
-      std::cout << "st.same_node[Dir] = " << st.same_node[Ym] << std::endl;
-      std::cout << "base              = " << (base - plocal)/rescale << std::endl;
-      std::cout << "Basep             = " << (basep - plocal)/rescale << std::endl;
-      //printf("U                 = %llu\n", (uint64_t)&[sU](Dir));
-      std::cout << "----------------------------------------------------" << std::endl;
-#endif
-
-      ASM_LEG(Zm,Tm,PERMUTE_DIR1,DIR6_PROJ,DIR6_RECON);
-
-#ifdef SHOW
-      std::cout << "Dir = " << Zm << "        "  << WHERE<< std::endl;
-
-      std::cout << "ent  nent  local  perm       = " << ent << "  " << nent << "  " << local << "  "  << perm << std::endl;
-      std::cout << "st.same_node[Dir] = " << st.same_node[Zm] << std::endl;
-      std::cout << "base              = " << (base - plocal)/rescale << std::endl;
-      std::cout << "Basep             = " << (basep - plocal)/rescale << std::endl;
-      //printf("U                 = %llu\n", (uint64_t)&[sU](Dir));
-      std::cout << "----------------------------------------------------" << std::endl;
-#endif
-
-      ASM_LEG(Tm,Xp,PERMUTE_DIR0,DIR7_PROJ,DIR7_RECON);
-
-#ifdef SHOW
-      std::cout << "Dir = " << Tm << "        "  << WHERE<< std::endl;
-
-      std::cout << "ent  nent  local  perm       = " << ent << "  " << nent << "  " << local << "  "  << perm << std::endl;
-      std::cout << "st.same_node[Dir] = " << st.same_node[Tm] << std::endl;
-      std::cout << "base              = " << (base - plocal)/rescale << std::endl;
-      std::cout << "Basep             = " << (basep - plocal)/rescale << std::endl;
-      //printf("U                 = %llu\n", (uint64_t)&[sU](Dir));
-      std::cout << "----------------------------------------------------" << std::endl;
-#endif
-
-#ifdef EXTERIOR
-      if (nmu==0) break;
-      //      if (nmu!=0) std::cout << "EXT "<<sU<<std::endl;
-#endif
-      base = (uint64_t) &out[ss];
-      basep= st.GetPFInfo(nent,plocal); ent++;
-      basep = (uint64_t) &out[ssn];
-      RESULT(base,basep);
-
-#ifdef SHOW
-      std::cout << "Dir = FINAL        " <<  WHERE<< std::endl;;
-
-      base_ss = base;
-      std::cout << "base              = " << (base - (uint64_t) &out[0])/rescale << std::endl;
-      std::cout << "Basep             = " << (basep - plocal)/rescale << std::endl;
-      //printf("U                 = %llu\n", (uint64_t)&[sU](Dir));
-      std::cout << "----------------------------------------------------" << std::endl;
-#endif
-
-    }
-    ssU++;
-    UNLOCK_GAUGE(0);
-  }
-}
-
-#undef DIR0_PROJ
-#undef DIR1_PROJ
-#undef DIR2_PROJ
-#undef DIR3_PROJ
-#undef DIR4_PROJ
-#undef DIR5_PROJ
-#undef DIR6_PROJ
-#undef DIR7_PROJ
-#undef DIR0_RECON
-#undef DIR1_RECON
-#undef DIR2_RECON
-#undef DIR3_RECON
-#undef DIR4_RECON
-#undef DIR5_RECON
-#undef DIR6_RECON
-#undef DIR7_RECON
-#undef ASM_LEG
-#undef ASM_LEG_XP
-#undef RESULT

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

@@ -1,943 +0,0 @@
-    /*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: ./lib/qcd/action/fermion/WilsonKernelsHand.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 */
-
-#pragma once
-
-#include <Grid/qcd/action/fermion/FermionCore.h>
-
-
-#undef LOAD_CHIMU
-#undef LOAD_CHI
-#undef MULT_2SPIN
-#undef PERMUTE_DIR
-#undef XP_PROJ
-#undef YP_PROJ
-#undef ZP_PROJ
-#undef TP_PROJ
-#undef XM_PROJ
-#undef YM_PROJ
-#undef ZM_PROJ
-#undef TM_PROJ
-#undef XP_RECON
-#undef XP_RECON_ACCUM
-#undef XM_RECON
-#undef XM_RECON_ACCUM
-#undef YP_RECON_ACCUM
-#undef YM_RECON_ACCUM
-#undef ZP_RECON_ACCUM
-#undef ZM_RECON_ACCUM
-#undef TP_RECON_ACCUM
-#undef TM_RECON_ACCUM
-#undef ZERO_RESULT
-#undef Chimu_00
-#undef Chimu_01
-#undef Chimu_02
-#undef Chimu_10
-#undef Chimu_11
-#undef Chimu_12
-#undef Chimu_20
-#undef Chimu_21
-#undef Chimu_22
-#undef Chimu_30
-#undef Chimu_31
-#undef Chimu_32
-#undef HAND_STENCIL_LEG
-#undef HAND_STENCIL_LEG_INT
-#undef HAND_STENCIL_LEG_EXT
-#undef HAND_RESULT
-#undef HAND_RESULT_INT
-#undef HAND_RESULT_EXT
-
-#define REGISTER
-
-#define LOAD_CHIMU \
-  {const SiteSpinor & ref (in[offset]);	\
-    Chimu_00=ref()(0)(0);\
-    Chimu_01=ref()(0)(1);\
-    Chimu_02=ref()(0)(2);\
-    Chimu_10=ref()(1)(0);\
-    Chimu_11=ref()(1)(1);\
-    Chimu_12=ref()(1)(2);\
-    Chimu_20=ref()(2)(0);\
-    Chimu_21=ref()(2)(1);\
-    Chimu_22=ref()(2)(2);\
-    Chimu_30=ref()(3)(0);\
-    Chimu_31=ref()(3)(1);\
-    Chimu_32=ref()(3)(2);\
-    std::cout << std::endl << "DEBUG -- LOAD_CHIMU" << std::endl; \
-    std::cout << "Chimu_00 -- " <<  Chimu_00 << std::endl; \
-    std::cout << "Chimu_01 -- " <<  Chimu_01 << std::endl; \
-    std::cout << "Chimu_02 -- " <<  Chimu_02 << std::endl; \
-    std::cout << "Chimu_10 -- " <<  Chimu_10 << std::endl; \
-    std::cout << "Chimu_11 -- " <<  Chimu_11 << std::endl; \
-    std::cout << "Chimu_12 -- " <<  Chimu_12 << std::endl; \
-    std::cout << "Chimu_20 -- " <<  Chimu_20 << std::endl; \
-    std::cout << "Chimu_21 -- " <<  Chimu_21 << std::endl; \
-    std::cout << "Chimu_22 -- " <<  Chimu_22 << std::endl; \
-    std::cout << "Chimu_30 -- " <<  Chimu_30 << std::endl; \
-    std::cout << "Chimu_31 -- " <<  Chimu_31 << std::endl; \
-    std::cout << "Chimu_32 -- " <<  Chimu_32 << std::endl; \
-}
-
-#define LOAD_CHI\
-  {const SiteHalfSpinor &ref(buf[offset]);	\
-    Chi_00 = ref()(0)(0);\
-    Chi_01 = ref()(0)(1);\
-    Chi_02 = ref()(0)(2);\
-    Chi_10 = ref()(1)(0);\
-    Chi_11 = ref()(1)(1);\
-    Chi_12 = ref()(1)(2);\
-    std::cout << std::endl << "DEBUG -- LOAD_CHI" << std::endl; \
-    std::cout << "Chi_00 -- " <<  Chi_00 << std::endl; \
-    std::cout << "Chi_01 -- " <<  Chi_01 << std::endl; \
-    std::cout << "Chi_02 -- " <<  Chi_02 << std::endl; \
-    std::cout << "Chi_10 -- " <<  Chi_10 << std::endl; \
-    std::cout << "Chi_11 -- " <<  Chi_11 << std::endl; \
-    std::cout << "Chi_12 -- " <<  Chi_12 << std::endl; \
-  }
-
-// To splat or not to splat depends on the implementation
-#define MULT_2SPIN(A)\
-  {auto & ref(U[sU](A));			\
-   Impl::loadLinkElement(U_00,ref()(0,0));	\
-   Impl::loadLinkElement(U_10,ref()(1,0));	\
-   Impl::loadLinkElement(U_20,ref()(2,0));	\
-   Impl::loadLinkElement(U_01,ref()(0,1));	\
-   Impl::loadLinkElement(U_11,ref()(1,1));	\
-   Impl::loadLinkElement(U_21,ref()(2,1));	\
-    UChi_00 = U_00*Chi_00;\
-    UChi_10 = U_00*Chi_10;\
-    UChi_01 = U_10*Chi_00;\
-    UChi_11 = U_10*Chi_10;\
-    UChi_02 = U_20*Chi_00;\
-    UChi_12 = U_20*Chi_10;\
-    UChi_00+= U_01*Chi_01;\
-    UChi_10+= U_01*Chi_11;\
-    UChi_01+= U_11*Chi_01;\
-    UChi_11+= U_11*Chi_11;\
-    UChi_02+= U_21*Chi_01;\
-    UChi_12+= U_21*Chi_11;\
-    Impl::loadLinkElement(U_00,ref()(0,2));	\
-    Impl::loadLinkElement(U_10,ref()(1,2));	\
-    Impl::loadLinkElement(U_20,ref()(2,2));	\
-    UChi_00+= U_00*Chi_02;\
-    UChi_10+= U_00*Chi_12;\
-    UChi_01+= U_10*Chi_02;\
-    UChi_11+= U_10*Chi_12;\
-    UChi_02+= U_20*Chi_02;\
-    UChi_12+= U_20*Chi_12;\
-    std::cout << std::endl << "DEBUG -- MULT_2SPIN" << std::endl; \
-    std::cout << "UChi_00 -- " <<  UChi_00 << std::endl; \
-    std::cout << "UChi_01 -- " <<  UChi_01 << std::endl; \
-    std::cout << "UChi_02 -- " <<  UChi_02 << std::endl; \
-    std::cout << "UChi_10 -- " <<  UChi_10 << std::endl; \
-    std::cout << "UChi_11 -- " <<  UChi_11 << std::endl; \
-    std::cout << "UChi_12 -- " <<  UChi_12 << std::endl; \
-    }
-
-
-#define PERMUTE_DIR(dir)			\
-std::cout << std::endl << "DEBUG -- PERM PRE" << std::endl; \
-std::cout << "Chi_00 -- " <<  Chi_00 << std::endl; \
-std::cout << "Chi_01 -- " <<  Chi_01 << std::endl; \
-std::cout << "Chi_02 -- " <<  Chi_02 << std::endl; \
-std::cout << "Chi_10 -- " <<  Chi_10 << std::endl; \
-std::cout << "Chi_11 -- " <<  Chi_11 << std::endl; \
-std::cout << "Chi_12 -- " <<  Chi_12 << std::endl; \
-      permute##dir(Chi_00,Chi_00);\
-      permute##dir(Chi_01,Chi_01);\
-      permute##dir(Chi_02,Chi_02);\
-      permute##dir(Chi_10,Chi_10);\
-      permute##dir(Chi_11,Chi_11);\
-      permute##dir(Chi_12,Chi_12);\
-  std::cout << std::endl << "DEBUG -- PERM POST" << std::endl; \
-  std::cout << "Chi_00 -- " <<  Chi_00 << std::endl; \
-  std::cout << "Chi_01 -- " <<  Chi_01 << std::endl; \
-  std::cout << "Chi_02 -- " <<  Chi_02 << std::endl; \
-  std::cout << "Chi_10 -- " <<  Chi_10 << std::endl; \
-  std::cout << "Chi_11 -- " <<  Chi_11 << std::endl; \
-  std::cout << "Chi_12 -- " <<  Chi_12 << std::endl;
-
-//      hspin(0)=fspin(0)+timesI(fspin(3));
-//      hspin(1)=fspin(1)+timesI(fspin(2));
-#define XP_PROJ \
-    Chi_00 = Chimu_00+timesI(Chimu_30);\
-    Chi_01 = Chimu_01+timesI(Chimu_31);\
-    Chi_02 = Chimu_02+timesI(Chimu_32);\
-    Chi_10 = Chimu_10+timesI(Chimu_20);\
-    Chi_11 = Chimu_11+timesI(Chimu_21);\
-    Chi_12 = Chimu_12+timesI(Chimu_22);\
-    std::cout << std::endl << "DEBUG -- XP_PROJ" << std::endl; \
-    std::cout << "Chi_00 -- " <<  Chi_00 << std::endl; \
-    std::cout << "Chi_01 -- " <<  Chi_01 << std::endl; \
-    std::cout << "Chi_02 -- " <<  Chi_02 << std::endl; \
-    std::cout << "Chi_10 -- " <<  Chi_10 << std::endl; \
-    std::cout << "Chi_11 -- " <<  Chi_11 << std::endl; \
-    std::cout << "Chi_12 -- " <<  Chi_12 << std::endl;
-
-#define YP_PROJ \
-    Chi_00 = Chimu_00-Chimu_30;\
-    Chi_01 = Chimu_01-Chimu_31;\
-    Chi_02 = Chimu_02-Chimu_32;\
-    Chi_10 = Chimu_10+Chimu_20;\
-    Chi_11 = Chimu_11+Chimu_21;\
-    Chi_12 = Chimu_12+Chimu_22;\
-    std::cout << std::endl << "DEBUG -- YP_PROJ" << std::endl; \
-    std::cout << "Chi_00 -- " <<  Chi_00 << std::endl; \
-    std::cout << "Chi_01 -- " <<  Chi_01 << std::endl; \
-    std::cout << "Chi_02 -- " <<  Chi_02 << std::endl; \
-    std::cout << "Chi_10 -- " <<  Chi_10 << std::endl; \
-    std::cout << "Chi_11 -- " <<  Chi_11 << std::endl; \
-    std::cout << "Chi_12 -- " <<  Chi_12 << std::endl;
-
-#define ZP_PROJ \
-  Chi_00 = Chimu_00+timesI(Chimu_20);		\
-  Chi_01 = Chimu_01+timesI(Chimu_21);		\
-  Chi_02 = Chimu_02+timesI(Chimu_22);		\
-  Chi_10 = Chimu_10-timesI(Chimu_30);		\
-  Chi_11 = Chimu_11-timesI(Chimu_31);		\
-  Chi_12 = Chimu_12-timesI(Chimu_32);\
-  std::cout << std::endl << "DEBUG -- ZP_PROJ" << std::endl; \
-  std::cout << "Chi_00 -- " <<  Chi_00 << std::endl; \
-  std::cout << "Chi_01 -- " <<  Chi_01 << std::endl; \
-  std::cout << "Chi_02 -- " <<  Chi_02 << std::endl; \
-  std::cout << "Chi_10 -- " <<  Chi_10 << std::endl; \
-  std::cout << "Chi_11 -- " <<  Chi_11 << std::endl; \
-  std::cout << "Chi_12 -- " <<  Chi_12 << std::endl;
-
-#define TP_PROJ \
-  Chi_00 = Chimu_00+Chimu_20;		\
-  Chi_01 = Chimu_01+Chimu_21;		\
-  Chi_02 = Chimu_02+Chimu_22;		\
-  Chi_10 = Chimu_10+Chimu_30;		\
-  Chi_11 = Chimu_11+Chimu_31;		\
-  Chi_12 = Chimu_12+Chimu_32;\
-  std::cout << std::endl << "DEBUG -- TP_PROJ" << std::endl; \
-  std::cout << "Chi_00 -- " <<  Chi_00 << std::endl; \
-  std::cout << "Chi_01 -- " <<  Chi_01 << std::endl; \
-  std::cout << "Chi_02 -- " <<  Chi_02 << std::endl; \
-  std::cout << "Chi_10 -- " <<  Chi_10 << std::endl; \
-  std::cout << "Chi_11 -- " <<  Chi_11 << std::endl; \
-  std::cout << "Chi_12 -- " <<  Chi_12 << std::endl;
-
-
-//      hspin(0)=fspin(0)-timesI(fspin(3));
-//      hspin(1)=fspin(1)-timesI(fspin(2));
-#define XM_PROJ \
-    Chi_00 = Chimu_00-timesI(Chimu_30);\
-    Chi_01 = Chimu_01-timesI(Chimu_31);\
-    Chi_02 = Chimu_02-timesI(Chimu_32);\
-    Chi_10 = Chimu_10-timesI(Chimu_20);\
-    Chi_11 = Chimu_11-timesI(Chimu_21);\
-    Chi_12 = Chimu_12-timesI(Chimu_22);\
-    std::cout << std::endl << "DEBUG -- XM_PROJ" << std::endl; \
-    std::cout << "Chi_00 -- " <<  Chi_00 << std::endl; \
-    std::cout << "Chi_01 -- " <<  Chi_01 << std::endl; \
-    std::cout << "Chi_02 -- " <<  Chi_02 << std::endl; \
-    std::cout << "Chi_10 -- " <<  Chi_10 << std::endl; \
-    std::cout << "Chi_11 -- " <<  Chi_11 << std::endl; \
-    std::cout << "Chi_12 -- " <<  Chi_12 << std::endl;
-
-#define YM_PROJ \
-    Chi_00 = Chimu_00+Chimu_30;\
-    Chi_01 = Chimu_01+Chimu_31;\
-    Chi_02 = Chimu_02+Chimu_32;\
-    Chi_10 = Chimu_10-Chimu_20;\
-    Chi_11 = Chimu_11-Chimu_21;\
-    Chi_12 = Chimu_12-Chimu_22;\
-    std::cout << std::endl << "DEBUG -- YM_PROJ" << std::endl; \
-    std::cout << "Chi_00 -- " <<  Chi_00 << std::endl; \
-    std::cout << "Chi_01 -- " <<  Chi_01 << std::endl; \
-    std::cout << "Chi_02 -- " <<  Chi_02 << std::endl; \
-    std::cout << "Chi_10 -- " <<  Chi_10 << std::endl; \
-    std::cout << "Chi_11 -- " <<  Chi_11 << std::endl; \
-    std::cout << "Chi_12 -- " <<  Chi_12 << std::endl;
-
-#define ZM_PROJ \
-  Chi_00 = Chimu_00-timesI(Chimu_20);		\
-  Chi_01 = Chimu_01-timesI(Chimu_21);		\
-  Chi_02 = Chimu_02-timesI(Chimu_22);		\
-  Chi_10 = Chimu_10+timesI(Chimu_30);		\
-  Chi_11 = Chimu_11+timesI(Chimu_31);		\
-  Chi_12 = Chimu_12+timesI(Chimu_32);\
-  std::cout << std::endl << "DEBUG -- ZM_PROJ" << std::endl; \
-  std::cout << "Chi_00 -- " <<  Chi_00 << std::endl; \
-  std::cout << "Chi_01 -- " <<  Chi_01 << std::endl; \
-  std::cout << "Chi_02 -- " <<  Chi_02 << std::endl; \
-  std::cout << "Chi_10 -- " <<  Chi_10 << std::endl; \
-  std::cout << "Chi_11 -- " <<  Chi_11 << std::endl; \
-  std::cout << "Chi_12 -- " <<  Chi_12 << std::endl;
-
-#define TM_PROJ \
-  Chi_00 = Chimu_00-Chimu_20;		\
-  Chi_01 = Chimu_01-Chimu_21;		\
-  Chi_02 = Chimu_02-Chimu_22;		\
-  Chi_10 = Chimu_10-Chimu_30;		\
-  Chi_11 = Chimu_11-Chimu_31;		\
-  Chi_12 = Chimu_12-Chimu_32;\
-  std::cout << std::endl << "DEBUG -- TM_PROJ" << std::endl; \
-  std::cout << "Chi_00 -- " <<  Chi_00 << std::endl; \
-  std::cout << "Chi_01 -- " <<  Chi_01 << std::endl; \
-  std::cout << "Chi_02 -- " <<  Chi_02 << std::endl; \
-  std::cout << "Chi_10 -- " <<  Chi_10 << std::endl; \
-  std::cout << "Chi_11 -- " <<  Chi_11 << std::endl; \
-  std::cout << "Chi_12 -- " <<  Chi_12 << std::endl;
-
-//      fspin(0)=hspin(0);
-//      fspin(1)=hspin(1);
-//      fspin(2)=timesMinusI(hspin(1));
-//      fspin(3)=timesMinusI(hspin(0));
-#define XP_RECON\
-  result_00 = UChi_00;\
-  result_01 = UChi_01;\
-  result_02 = UChi_02;\
-  result_10 = UChi_10;\
-  result_11 = UChi_11;\
-  result_12 = UChi_12;\
-  result_20 = timesMinusI(UChi_10);\
-  result_21 = timesMinusI(UChi_11);\
-  result_22 = timesMinusI(UChi_12);\
-  result_30 = timesMinusI(UChi_00);\
-  result_31 = timesMinusI(UChi_01);\
-  result_32 = timesMinusI(UChi_02);\
-  std::cout << std::endl << "DEBUG -- XP_RECON" << std::endl; \
-  std::cout << "result_00 -- " <<  result_00 << std::endl; \
-  std::cout << "result_01 -- " <<  result_01 << std::endl; \
-  std::cout << "result_02 -- " <<  result_02 << std::endl; \
-  std::cout << "result_10 -- " <<  result_10 << std::endl; \
-  std::cout << "result_11 -- " <<  result_11 << std::endl; \
-  std::cout << "result_12 -- " <<  result_12 << std::endl; \
-  std::cout << "result_20 -- " <<  result_20 << std::endl; \
-  std::cout << "result_21 -- " <<  result_21 << std::endl; \
-  std::cout << "result_22 -- " <<  result_22 << std::endl; \
-  std::cout << "result_30 -- " <<  result_30 << std::endl; \
-  std::cout << "result_31 -- " <<  result_31 << std::endl; \
-  std::cout << "result_32 -- " <<  result_32 << std::endl;
-
-#define XP_RECON_ACCUM\
-  result_00+=UChi_00;\
-  result_01+=UChi_01;\
-  result_02+=UChi_02;\
-  result_10+=UChi_10;\
-  result_11+=UChi_11;\
-  result_12+=UChi_12;\
-  result_20-=timesI(UChi_10);\
-  result_21-=timesI(UChi_11);\
-  result_22-=timesI(UChi_12);\
-  result_30-=timesI(UChi_00);\
-  result_31-=timesI(UChi_01);\
-  result_32-=timesI(UChi_02);\
-  std::cout << std::endl << "DEBUG -- XP_RECON_ACCUM" << std::endl; \
-  std::cout << "result_00 -- " <<  result_00 << std::endl; \
-  std::cout << "result_01 -- " <<  result_01 << std::endl; \
-  std::cout << "result_02 -- " <<  result_02 << std::endl; \
-  std::cout << "result_10 -- " <<  result_10 << std::endl; \
-  std::cout << "result_11 -- " <<  result_11 << std::endl; \
-  std::cout << "result_12 -- " <<  result_12 << std::endl; \
-  std::cout << "result_20 -- " <<  result_20 << std::endl; \
-  std::cout << "result_21 -- " <<  result_21 << std::endl; \
-  std::cout << "result_22 -- " <<  result_22 << std::endl; \
-  std::cout << "result_30 -- " <<  result_30 << std::endl; \
-  std::cout << "result_31 -- " <<  result_31 << std::endl; \
-  std::cout << "result_32 -- " <<  result_32 << std::endl;
-
-#define XM_RECON\
-  result_00 = UChi_00;\
-  result_01 = UChi_01;\
-  result_02 = UChi_02;\
-  result_10 = UChi_10;\
-  result_11 = UChi_11;\
-  result_12 = UChi_12;\
-  result_20 = timesI(UChi_10);\
-  result_21 = timesI(UChi_11);\
-  result_22 = timesI(UChi_12);\
-  result_30 = timesI(UChi_00);\
-  result_31 = timesI(UChi_01);\
-  result_32 = timesI(UChi_02);\
-  std::cout << std::endl << "DEBUG -- XM_RECON" << std::endl; \
-  std::cout << "result_00 -- " <<  result_00 << std::endl; \
-  std::cout << "result_01 -- " <<  result_01 << std::endl; \
-  std::cout << "result_02 -- " <<  result_02 << std::endl; \
-  std::cout << "result_10 -- " <<  result_10 << std::endl; \
-  std::cout << "result_11 -- " <<  result_11 << std::endl; \
-  std::cout << "result_12 -- " <<  result_12 << std::endl; \
-  std::cout << "result_20 -- " <<  result_20 << std::endl; \
-  std::cout << "result_21 -- " <<  result_21 << std::endl; \
-  std::cout << "result_22 -- " <<  result_22 << std::endl; \
-  std::cout << "result_30 -- " <<  result_30 << std::endl; \
-  std::cout << "result_31 -- " <<  result_31 << std::endl; \
-  std::cout << "result_32 -- " <<  result_32 << std::endl;
-
-#define XM_RECON_ACCUM\
-  result_00+= UChi_00;\
-  result_01+= UChi_01;\
-  result_02+= UChi_02;\
-  result_10+= UChi_10;\
-  result_11+= UChi_11;\
-  result_12+= UChi_12;\
-  result_20+= timesI(UChi_10);\
-  result_21+= timesI(UChi_11);\
-  result_22+= timesI(UChi_12);\
-  result_30+= timesI(UChi_00);\
-  result_31+= timesI(UChi_01);\
-  result_32+= timesI(UChi_02);\
-  std::cout << std::endl << "DEBUG -- XM_RECON_ACCUM" << std::endl; \
-  std::cout << "result_00 -- " <<  result_00 << std::endl; \
-  std::cout << "result_01 -- " <<  result_01 << std::endl; \
-  std::cout << "result_02 -- " <<  result_02 << std::endl; \
-  std::cout << "result_10 -- " <<  result_10 << std::endl; \
-  std::cout << "result_11 -- " <<  result_11 << std::endl; \
-  std::cout << "result_12 -- " <<  result_12 << std::endl; \
-  std::cout << "result_20 -- " <<  result_20 << std::endl; \
-  std::cout << "result_21 -- " <<  result_21 << std::endl; \
-  std::cout << "result_22 -- " <<  result_22 << std::endl; \
-  std::cout << "result_30 -- " <<  result_30 << std::endl; \
-  std::cout << "result_31 -- " <<  result_31 << std::endl; \
-  std::cout << "result_32 -- " <<  result_32 << std::endl;
-
-#define YP_RECON_ACCUM\
-  result_00+= UChi_00;\
-  result_01+= UChi_01;\
-  result_02+= UChi_02;\
-  result_10+= UChi_10;\
-  result_11+= UChi_11;\
-  result_12+= UChi_12;\
-  result_20+= UChi_10;\
-  result_21+= UChi_11;\
-  result_22+= UChi_12;\
-  result_30-= UChi_00;\
-  result_31-= UChi_01;\
-  result_32-= UChi_02;\
-  std::cout << std::endl << "DEBUG -- YP_RECON_ACCUM" << std::endl; \
-  std::cout << "result_00 -- " <<  result_00 << std::endl; \
-  std::cout << "result_01 -- " <<  result_01 << std::endl; \
-  std::cout << "result_02 -- " <<  result_02 << std::endl; \
-  std::cout << "result_10 -- " <<  result_10 << std::endl; \
-  std::cout << "result_11 -- " <<  result_11 << std::endl; \
-  std::cout << "result_12 -- " <<  result_12 << std::endl; \
-  std::cout << "result_20 -- " <<  result_20 << std::endl; \
-  std::cout << "result_21 -- " <<  result_21 << std::endl; \
-  std::cout << "result_22 -- " <<  result_22 << std::endl; \
-  std::cout << "result_30 -- " <<  result_30 << std::endl; \
-  std::cout << "result_31 -- " <<  result_31 << std::endl; \
-  std::cout << "result_32 -- " <<  result_32 << std::endl;
-
-#define YM_RECON_ACCUM\
-  result_00+= UChi_00;\
-  result_01+= UChi_01;\
-  result_02+= UChi_02;\
-  result_10+= UChi_10;\
-  result_11+= UChi_11;\
-  result_12+= UChi_12;\
-  result_20-= UChi_10;\
-  result_21-= UChi_11;\
-  result_22-= UChi_12;\
-  result_30+= UChi_00;\
-  result_31+= UChi_01;\
-  result_32+= UChi_02;\
-  std::cout << std::endl << "DEBUG -- YM_RECON_ACCUM" << std::endl; \
-  std::cout << "result_00 -- " <<  result_00 << std::endl; \
-  std::cout << "result_01 -- " <<  result_01 << std::endl; \
-  std::cout << "result_02 -- " <<  result_02 << std::endl; \
-  std::cout << "result_10 -- " <<  result_10 << std::endl; \
-  std::cout << "result_11 -- " <<  result_11 << std::endl; \
-  std::cout << "result_12 -- " <<  result_12 << std::endl; \
-  std::cout << "result_20 -- " <<  result_20 << std::endl; \
-  std::cout << "result_21 -- " <<  result_21 << std::endl; \
-  std::cout << "result_22 -- " <<  result_22 << std::endl; \
-  std::cout << "result_30 -- " <<  result_30 << std::endl; \
-  std::cout << "result_31 -- " <<  result_31 << std::endl; \
-  std::cout << "result_32 -- " <<  result_32 << std::endl;
-
-#define ZP_RECON_ACCUM\
-  result_00+= UChi_00;\
-  result_01+= UChi_01;\
-  result_02+= UChi_02;\
-  result_10+= UChi_10;\
-  result_11+= UChi_11;\
-  result_12+= UChi_12;\
-  result_20-= timesI(UChi_00);			\
-  result_21-= timesI(UChi_01);			\
-  result_22-= timesI(UChi_02);			\
-  result_30+= timesI(UChi_10);			\
-  result_31+= timesI(UChi_11);			\
-  result_32+= timesI(UChi_12);\
-  std::cout << std::endl << "DEBUG -- ZP_RECON_ACCUM" << std::endl; \
-  std::cout << "result_00 -- " <<  result_00 << std::endl; \
-  std::cout << "result_01 -- " <<  result_01 << std::endl; \
-  std::cout << "result_02 -- " <<  result_02 << std::endl; \
-  std::cout << "result_10 -- " <<  result_10 << std::endl; \
-  std::cout << "result_11 -- " <<  result_11 << std::endl; \
-  std::cout << "result_12 -- " <<  result_12 << std::endl; \
-  std::cout << "result_20 -- " <<  result_20 << std::endl; \
-  std::cout << "result_21 -- " <<  result_21 << std::endl; \
-  std::cout << "result_22 -- " <<  result_22 << std::endl; \
-  std::cout << "result_30 -- " <<  result_30 << std::endl; \
-  std::cout << "result_31 -- " <<  result_31 << std::endl; \
-  std::cout << "result_32 -- " <<  result_32 << std::endl;
-
-#define ZM_RECON_ACCUM\
-  result_00+= UChi_00;\
-  result_01+= UChi_01;\
-  result_02+= UChi_02;\
-  result_10+= UChi_10;\
-  result_11+= UChi_11;\
-  result_12+= UChi_12;\
-  result_20+= timesI(UChi_00);			\
-  result_21+= timesI(UChi_01);			\
-  result_22+= timesI(UChi_02);			\
-  result_30-= timesI(UChi_10);			\
-  result_31-= timesI(UChi_11);			\
-  result_32-= timesI(UChi_12);\
-  std::cout << std::endl << "DEBUG -- ZM_RECON_ACCUM" << std::endl; \
-  std::cout << "result_00 -- " <<  result_00 << std::endl; \
-  std::cout << "result_01 -- " <<  result_01 << std::endl; \
-  std::cout << "result_02 -- " <<  result_02 << std::endl; \
-  std::cout << "result_10 -- " <<  result_10 << std::endl; \
-  std::cout << "result_11 -- " <<  result_11 << std::endl; \
-  std::cout << "result_12 -- " <<  result_12 << std::endl; \
-  std::cout << "result_20 -- " <<  result_20 << std::endl; \
-  std::cout << "result_21 -- " <<  result_21 << std::endl; \
-  std::cout << "result_22 -- " <<  result_22 << std::endl; \
-  std::cout << "result_30 -- " <<  result_30 << std::endl; \
-  std::cout << "result_31 -- " <<  result_31 << std::endl; \
-  std::cout << "result_32 -- " <<  result_32 << std::endl;
-
-#define TP_RECON_ACCUM\
-  result_00+= UChi_00;\
-  result_01+= UChi_01;\
-  result_02+= UChi_02;\
-  result_10+= UChi_10;\
-  result_11+= UChi_11;\
-  result_12+= UChi_12;\
-  result_20+= UChi_00;			\
-  result_21+= UChi_01;			\
-  result_22+= UChi_02;			\
-  result_30+= UChi_10;			\
-  result_31+= UChi_11;			\
-  result_32+= UChi_12;\
-  std::cout << std::endl << "DEBUG -- TP_RECON_ACCUM" << std::endl; \
-  std::cout << "result_00 -- " <<  result_00 << std::endl; \
-  std::cout << "result_01 -- " <<  result_01 << std::endl; \
-  std::cout << "result_02 -- " <<  result_02 << std::endl; \
-  std::cout << "result_10 -- " <<  result_10 << std::endl; \
-  std::cout << "result_11 -- " <<  result_11 << std::endl; \
-  std::cout << "result_12 -- " <<  result_12 << std::endl; \
-  std::cout << "result_20 -- " <<  result_20 << std::endl; \
-  std::cout << "result_21 -- " <<  result_21 << std::endl; \
-  std::cout << "result_22 -- " <<  result_22 << std::endl; \
-  std::cout << "result_30 -- " <<  result_30 << std::endl; \
-  std::cout << "result_31 -- " <<  result_31 << std::endl; \
-  std::cout << "result_32 -- " <<  result_32 << std::endl;
-
-#define TM_RECON_ACCUM\
-  result_00+= UChi_00;\
-  result_01+= UChi_01;\
-  result_02+= UChi_02;\
-  result_10+= UChi_10;\
-  result_11+= UChi_11;\
-  result_12+= UChi_12;\
-  result_20-= UChi_00;	\
-  result_21-= UChi_01;	\
-  result_22-= UChi_02;	\
-  result_30-= UChi_10;	\
-  result_31-= UChi_11;	\
-  result_32-= UChi_12;\
-  std::cout << std::endl << "DEBUG -- TM_RECON_ACCUM" << std::endl; \
-  std::cout << "result_00 -- " <<  result_00 << std::endl; \
-  std::cout << "result_01 -- " <<  result_01 << std::endl; \
-  std::cout << "result_02 -- " <<  result_02 << std::endl; \
-  std::cout << "result_10 -- " <<  result_10 << std::endl; \
-  std::cout << "result_11 -- " <<  result_11 << std::endl; \
-  std::cout << "result_12 -- " <<  result_12 << std::endl; \
-  std::cout << "result_20 -- " <<  result_20 << std::endl; \
-  std::cout << "result_21 -- " <<  result_21 << std::endl; \
-  std::cout << "result_22 -- " <<  result_22 << std::endl; \
-  std::cout << "result_30 -- " <<  result_30 << std::endl; \
-  std::cout << "result_31 -- " <<  result_31 << std::endl; \
-  std::cout << "result_32 -- " <<  result_32 << std::endl;
-
-#define HAND_STENCIL_LEG(PROJ,PERM,DIR,RECON)	\
-  SE=st.GetEntry(ptype,DIR,ss);			\
-  offset = SE->_offset;				\
-  local  = SE->_is_local;			\
-  perm   = SE->_permute;			\
-  if ( local ) {				\
-    LOAD_CHIMU;					\
-    PROJ;					\
-    if ( perm) {				\
-      PERMUTE_DIR(PERM);			\
-    }						\
-  } else {					\
-    LOAD_CHI;					\
-  }						\
-  MULT_2SPIN(DIR);				\
-  RECON;
-
-#define HAND_STENCIL_LEG_INT(PROJ,PERM,DIR,RECON)	\
-  SE=st.GetEntry(ptype,DIR,ss);			\
-  offset = SE->_offset;				\
-  local  = SE->_is_local;			\
-  perm   = SE->_permute;			\
-  if ( local ) {				\
-    LOAD_CHIMU;					\
-    PROJ;					\
-    if ( perm) {				\
-      PERMUTE_DIR(PERM);			\
-    }						\
-  } else if ( st.same_node[DIR] ) {		\
-    LOAD_CHI;					\
-  }						\
-  if (local || st.same_node[DIR] ) {		\
-    MULT_2SPIN(DIR);				\
-    RECON;					\
-  }
-
-#define HAND_STENCIL_LEG_EXT(PROJ,PERM,DIR,RECON)	\
-  SE=st.GetEntry(ptype,DIR,ss);			\
-  offset = SE->_offset;				\
-  if((!SE->_is_local)&&(!st.same_node[DIR]) ) {	\
-    LOAD_CHI;					\
-    MULT_2SPIN(DIR);				\
-    RECON;					\
-    nmu++;					\
-  }
-
-#define HAND_RESULT(ss)				\
-  {						\
-    SiteSpinor & ref (out[ss]);		\
-    vstream(ref()(0)(0),result_00);		\
-    vstream(ref()(0)(1),result_01);		\
-    vstream(ref()(0)(2),result_02);		\
-    vstream(ref()(1)(0),result_10);		\
-    vstream(ref()(1)(1),result_11);		\
-    vstream(ref()(1)(2),result_12);		\
-    vstream(ref()(2)(0),result_20);		\
-    vstream(ref()(2)(1),result_21);		\
-    vstream(ref()(2)(2),result_22);		\
-    vstream(ref()(3)(0),result_30);		\
-    vstream(ref()(3)(1),result_31);		\
-    vstream(ref()(3)(2),result_32);		\
-    std::cout << std::endl << "DEBUG -- RESULT" << std::endl; \
-    std::cout << "result_00 -- " <<  result_00 << std::endl; \
-    std::cout << "result_01 -- " <<  result_01 << std::endl; \
-    std::cout << "result_02 -- " <<  result_02 << std::endl; \
-    std::cout << "result_10 -- " <<  result_10 << std::endl; \
-    std::cout << "result_11 -- " <<  result_11 << std::endl; \
-    std::cout << "result_12 -- " <<  result_12 << std::endl; \
-    std::cout << "result_20 -- " <<  result_20 << std::endl; \
-    std::cout << "result_21 -- " <<  result_21 << std::endl; \
-    std::cout << "result_22 -- " <<  result_22 << std::endl; \
-    std::cout << "result_30 -- " <<  result_30 << std::endl; \
-    std::cout << "result_31 -- " <<  result_31 << std::endl; \
-    std::cout << "result_32 -- " <<  result_32 << std::endl;\
-  }
-
-#define HAND_RESULT_EXT(ss)			\
-  if (nmu){					\
-    SiteSpinor & ref (out[ss]);		\
-    ref()(0)(0)+=result_00;		\
-    ref()(0)(1)+=result_01;		\
-    ref()(0)(2)+=result_02;		\
-    ref()(1)(0)+=result_10;		\
-    ref()(1)(1)+=result_11;		\
-    ref()(1)(2)+=result_12;		\
-    ref()(2)(0)+=result_20;		\
-    ref()(2)(1)+=result_21;		\
-    ref()(2)(2)+=result_22;		\
-    ref()(3)(0)+=result_30;		\
-    ref()(3)(1)+=result_31;		\
-    ref()(3)(2)+=result_32;		\
-    std::cout << std::endl << "DEBUG -- RESULT EXT" << std::endl; \
-    std::cout << "result_00 -- " <<  result_00 << std::endl; \
-    std::cout << "result_01 -- " <<  result_01 << std::endl; \
-    std::cout << "result_02 -- " <<  result_02 << std::endl; \
-    std::cout << "result_10 -- " <<  result_10 << std::endl; \
-    std::cout << "result_11 -- " <<  result_11 << std::endl; \
-    std::cout << "result_12 -- " <<  result_12 << std::endl; \
-    std::cout << "result_20 -- " <<  result_20 << std::endl; \
-    std::cout << "result_21 -- " <<  result_21 << std::endl; \
-    std::cout << "result_22 -- " <<  result_22 << std::endl; \
-    std::cout << "result_30 -- " <<  result_30 << std::endl; \
-    std::cout << "result_31 -- " <<  result_31 << std::endl; \
-    std::cout << "result_32 -- " <<  result_32 << std::endl;\
-  }
-
-
-#define HAND_DECLARATIONS(a)			\
-  Simd result_00;				\
-  Simd result_01;				\
-  Simd result_02;				\
-  Simd result_10;				\
-  Simd result_11;				\
-  Simd result_12;				\
-  Simd result_20;				\
-  Simd result_21;				\
-  Simd result_22;				\
-  Simd result_30;				\
-  Simd result_31;				\
-  Simd result_32;				\
-  Simd Chi_00;					\
-  Simd Chi_01;					\
-  Simd Chi_02;					\
-  Simd Chi_10;					\
-  Simd Chi_11;					\
-  Simd Chi_12;					\
-  Simd UChi_00;					\
-  Simd UChi_01;					\
-  Simd UChi_02;					\
-  Simd UChi_10;					\
-  Simd UChi_11;					\
-  Simd UChi_12;					\
-  Simd U_00;					\
-  Simd U_10;					\
-  Simd U_20;					\
-  Simd U_01;					\
-  Simd U_11;					\
-  Simd U_21;\
-  Simd debugreg;\
-  svbool_t pg1;        \
-  pg1 = svptrue_b64();        \
-
-#define ZERO_RESULT				\
-  result_00=Zero();				\
-  result_01=Zero();				\
-  result_02=Zero();				\
-  result_10=Zero();				\
-  result_11=Zero();				\
-  result_12=Zero();				\
-  result_20=Zero();				\
-  result_21=Zero();				\
-  result_22=Zero();				\
-  result_30=Zero();				\
-  result_31=Zero();				\
-  result_32=Zero();
-
-#define Chimu_00 Chi_00
-#define Chimu_01 Chi_01
-#define Chimu_02 Chi_02
-#define Chimu_10 Chi_10
-#define Chimu_11 Chi_11
-#define Chimu_12 Chi_12
-#define Chimu_20 UChi_00
-#define Chimu_21 UChi_01
-#define Chimu_22 UChi_02
-#define Chimu_30 UChi_10
-#define Chimu_31 UChi_11
-#define Chimu_32 UChi_12
-
-NAMESPACE_BEGIN(Grid);
-
-template<class Impl> void
-WilsonKernels<Impl>::HandDhopSite(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor  *buf,
-				  int ss,int sU,const FermionFieldView &in, FermionFieldView &out)
-{
-// T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
-  typedef typename Simd::scalar_type S;
-  typedef typename Simd::vector_type V;
-
-  HAND_DECLARATIONS(ignore);
-
-  int offset,local,perm, ptype;
-  StencilEntry *SE;
-
-  HAND_STENCIL_LEG(XM_PROJ,3,Xp,XM_RECON);
-  HAND_STENCIL_LEG(YM_PROJ,2,Yp,YM_RECON_ACCUM);
-  HAND_STENCIL_LEG(ZM_PROJ,1,Zp,ZM_RECON_ACCUM);
-  HAND_STENCIL_LEG(TM_PROJ,0,Tp,TM_RECON_ACCUM);
-  HAND_STENCIL_LEG(XP_PROJ,3,Xm,XP_RECON_ACCUM);
-  HAND_STENCIL_LEG(YP_PROJ,2,Ym,YP_RECON_ACCUM);
-  HAND_STENCIL_LEG(ZP_PROJ,1,Zm,ZP_RECON_ACCUM);
-  HAND_STENCIL_LEG(TP_PROJ,0,Tm,TP_RECON_ACCUM);
-  HAND_RESULT(ss);
-}
-
-template<class Impl>
-void WilsonKernels<Impl>::HandDhopSiteDag(StencilView &st,DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
-					  int ss,int sU,const FermionFieldView &in, FermionFieldView &out)
-{
-  typedef typename Simd::scalar_type S;
-  typedef typename Simd::vector_type V;
-
-  HAND_DECLARATIONS(ignore);
-
-  StencilEntry *SE;
-  int offset,local,perm, ptype;
-
-  HAND_STENCIL_LEG(XP_PROJ,3,Xp,XP_RECON);
-  HAND_STENCIL_LEG(YP_PROJ,2,Yp,YP_RECON_ACCUM);
-  HAND_STENCIL_LEG(ZP_PROJ,1,Zp,ZP_RECON_ACCUM);
-  HAND_STENCIL_LEG(TP_PROJ,0,Tp,TP_RECON_ACCUM);
-  HAND_STENCIL_LEG(XM_PROJ,3,Xm,XM_RECON_ACCUM);
-  HAND_STENCIL_LEG(YM_PROJ,2,Ym,YM_RECON_ACCUM);
-  HAND_STENCIL_LEG(ZM_PROJ,1,Zm,ZM_RECON_ACCUM);
-  HAND_STENCIL_LEG(TM_PROJ,0,Tm,TM_RECON_ACCUM);
-  HAND_RESULT(ss);
-}
-
-template<class Impl> void
-WilsonKernels<Impl>::HandDhopSiteInt(StencilView &st,DoubledGaugeFieldView &U,SiteHalfSpinor  *buf,
-					  int ss,int sU,const FermionFieldView &in, FermionFieldView &out)
-{
-// T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
-  typedef typename Simd::scalar_type S;
-  typedef typename Simd::vector_type V;
-
-  HAND_DECLARATIONS(ignore);
-
-  int offset,local,perm, ptype;
-  StencilEntry *SE;
-  ZERO_RESULT;
-  HAND_STENCIL_LEG_INT(XM_PROJ,3,Xp,XM_RECON_ACCUM);
-  HAND_STENCIL_LEG_INT(YM_PROJ,2,Yp,YM_RECON_ACCUM);
-  HAND_STENCIL_LEG_INT(ZM_PROJ,1,Zp,ZM_RECON_ACCUM);
-  HAND_STENCIL_LEG_INT(TM_PROJ,0,Tp,TM_RECON_ACCUM);
-  HAND_STENCIL_LEG_INT(XP_PROJ,3,Xm,XP_RECON_ACCUM);
-  HAND_STENCIL_LEG_INT(YP_PROJ,2,Ym,YP_RECON_ACCUM);
-  HAND_STENCIL_LEG_INT(ZP_PROJ,1,Zm,ZP_RECON_ACCUM);
-  HAND_STENCIL_LEG_INT(TP_PROJ,0,Tm,TP_RECON_ACCUM);
-  HAND_RESULT(ss);
-}
-
-template<class Impl>
-void WilsonKernels<Impl>::HandDhopSiteDagInt(StencilView &st,DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
-						  int ss,int sU,const FermionFieldView &in, FermionFieldView &out)
-{
-  typedef typename Simd::scalar_type S;
-  typedef typename Simd::vector_type V;
-
-  HAND_DECLARATIONS(ignore);
-
-  StencilEntry *SE;
-  int offset,local,perm, ptype;
-  ZERO_RESULT;
-  HAND_STENCIL_LEG_INT(XP_PROJ,3,Xp,XP_RECON_ACCUM);
-  HAND_STENCIL_LEG_INT(YP_PROJ,2,Yp,YP_RECON_ACCUM);
-  HAND_STENCIL_LEG_INT(ZP_PROJ,1,Zp,ZP_RECON_ACCUM);
-  HAND_STENCIL_LEG_INT(TP_PROJ,0,Tp,TP_RECON_ACCUM);
-  HAND_STENCIL_LEG_INT(XM_PROJ,3,Xm,XM_RECON_ACCUM);
-  HAND_STENCIL_LEG_INT(YM_PROJ,2,Ym,YM_RECON_ACCUM);
-  HAND_STENCIL_LEG_INT(ZM_PROJ,1,Zm,ZM_RECON_ACCUM);
-  HAND_STENCIL_LEG_INT(TM_PROJ,0,Tm,TM_RECON_ACCUM);
-  HAND_RESULT(ss);
-}
-
-template<class Impl> void
-WilsonKernels<Impl>::HandDhopSiteExt(StencilView &st,DoubledGaugeFieldView &U,SiteHalfSpinor  *buf,
-					  int ss,int sU,const FermionFieldView &in, FermionFieldView &out)
-{
-// T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
-  typedef typename Simd::scalar_type S;
-  typedef typename Simd::vector_type V;
-
-  HAND_DECLARATIONS(ignore);
-
-  int offset, ptype;
-  StencilEntry *SE;
-  int nmu=0;
-  ZERO_RESULT;
-  HAND_STENCIL_LEG_EXT(XM_PROJ,3,Xp,XM_RECON_ACCUM);
-  HAND_STENCIL_LEG_EXT(YM_PROJ,2,Yp,YM_RECON_ACCUM);
-  HAND_STENCIL_LEG_EXT(ZM_PROJ,1,Zp,ZM_RECON_ACCUM);
-  HAND_STENCIL_LEG_EXT(TM_PROJ,0,Tp,TM_RECON_ACCUM);
-  HAND_STENCIL_LEG_EXT(XP_PROJ,3,Xm,XP_RECON_ACCUM);
-  HAND_STENCIL_LEG_EXT(YP_PROJ,2,Ym,YP_RECON_ACCUM);
-  HAND_STENCIL_LEG_EXT(ZP_PROJ,1,Zm,ZP_RECON_ACCUM);
-  HAND_STENCIL_LEG_EXT(TP_PROJ,0,Tm,TP_RECON_ACCUM);
-  HAND_RESULT_EXT(ss);
-}
-
-template<class Impl>
-void WilsonKernels<Impl>::HandDhopSiteDagExt(StencilView &st,DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
-						  int ss,int sU,const FermionFieldView &in, FermionFieldView &out)
-{
-  typedef typename Simd::scalar_type S;
-  typedef typename Simd::vector_type V;
-
-  HAND_DECLARATIONS(ignore);
-
-  StencilEntry *SE;
-  int offset, ptype;
-  int nmu=0;
-  ZERO_RESULT;
-  HAND_STENCIL_LEG_EXT(XP_PROJ,3,Xp,XP_RECON_ACCUM);
-  HAND_STENCIL_LEG_EXT(YP_PROJ,2,Yp,YP_RECON_ACCUM);
-  HAND_STENCIL_LEG_EXT(ZP_PROJ,1,Zp,ZP_RECON_ACCUM);
-  HAND_STENCIL_LEG_EXT(TP_PROJ,0,Tp,TP_RECON_ACCUM);
-  HAND_STENCIL_LEG_EXT(XM_PROJ,3,Xm,XM_RECON_ACCUM);
-  HAND_STENCIL_LEG_EXT(YM_PROJ,2,Ym,YM_RECON_ACCUM);
-  HAND_STENCIL_LEG_EXT(ZM_PROJ,1,Zm,ZM_RECON_ACCUM);
-  HAND_STENCIL_LEG_EXT(TM_PROJ,0,Tm,TM_RECON_ACCUM);
-  HAND_RESULT_EXT(ss);
-}
-
-////////////// Wilson ; uses this implementation /////////////////////
-
-NAMESPACE_END(Grid);
-#undef LOAD_CHIMU
-#undef LOAD_CHI
-#undef MULT_2SPIN
-#undef PERMUTE_DIR
-#undef XP_PROJ
-#undef YP_PROJ
-#undef ZP_PROJ
-#undef TP_PROJ
-#undef XM_PROJ
-#undef YM_PROJ
-#undef ZM_PROJ
-#undef TM_PROJ
-#undef XP_RECON
-#undef XP_RECON_ACCUM
-#undef XM_RECON
-#undef XM_RECON_ACCUM
-#undef YP_RECON_ACCUM
-#undef YM_RECON_ACCUM
-#undef ZP_RECON_ACCUM
-#undef ZM_RECON_ACCUM
-#undef TP_RECON_ACCUM
-#undef TM_RECON_ACCUM
-#undef ZERO_RESULT
-#undef Chimu_00
-#undef Chimu_01
-#undef Chimu_02
-#undef Chimu_10
-#undef Chimu_11
-#undef Chimu_12
-#undef Chimu_20
-#undef Chimu_21
-#undef Chimu_22
-#undef Chimu_30
-#undef Chimu_31
-#undef Chimu_32
-#undef HAND_STENCIL_LEG
-#undef HAND_STENCIL_LEG_INT
-#undef HAND_STENCIL_LEG_EXT
-#undef HAND_RESULT
-#undef HAND_RESULT_INT
-#undef HAND_RESULT_EXT

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

@@ -43,11 +43,11 @@ NAMESPACE_BEGIN(Grid);
 accelerator_inline void get_stencil(StencilEntry * mem, StencilEntry &chip)
 {
 #ifdef GRID_SIMT
-  static_assert(sizeof(StencilEntry)==sizeof(uint4),"Unexpected Stencil Entry Size");
+  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;
@@ -66,7 +66,7 @@ accelerator_inline void get_stencil(StencilEntry * mem, StencilEntry &chip)
   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) {						\
@@ -81,7 +81,7 @@ accelerator_inline void get_stencil(StencilEntry * mem, StencilEntry &chip)
     Impl::multLink(Uchi, U[sU], chi, Dir, SE, st);		\
     Recon(result, Uchi);					\
   }								\
-  acceleratorSynchronise();
+  acceleratorSynchronise();						
 
 #define GENERIC_STENCIL_LEG_EXT(Dir,spProj,Recon)		\
   SE = st.GetEntry(ptype, Dir, sF);				\
@@ -91,7 +91,7 @@ accelerator_inline void get_stencil(StencilEntry * mem, StencilEntry &chip)
     Recon(result, Uchi);					\
     nmu++;							\
   }								\
-  acceleratorSynchronise();
+  acceleratorSynchronise();						
 
 #define GENERIC_DHOPDIR_LEG_BODY(Dir,spProj,Recon)		\
     if (SE->_is_local ) {					\
@@ -103,7 +103,7 @@ accelerator_inline void get_stencil(StencilEntry * mem, StencilEntry &chip)
     }								\
     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) {						\
@@ -143,7 +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;
@@ -200,7 +200,7 @@ 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;
@@ -251,7 +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);
@@ -261,7 +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;
@@ -282,7 +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);
@@ -307,7 +307,7 @@ void WilsonKernels<Impl>::GenericDhopSiteExt(StencilView &st,  DoubledGaugeField
   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);
@@ -318,9 +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;
@@ -346,7 +346,7 @@ 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) 
 {
    autoView(U_v  ,U,AcceleratorRead);
    autoView(in_v ,in,AcceleratorRead);
@@ -362,8 +362,8 @@ void WilsonKernels<Impl>::DhopDirAll( StencilImpl &st, DoubledGaugeField &U,Site
    autoView(out_Tp,out[7],AcceleratorWrite);
    auto CBp=st.CommBuf();
    accelerator_for(sss,Nsite*Ls,Simd::Nsimd(),{
-      int sU=sss/Ls;
-      int sF =sss;
+      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);
@@ -378,7 +378,7 @@ void WilsonKernels<Impl>::DhopDirAll( StencilImpl &st, DoubledGaugeField &U,Site
 
 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);
@@ -387,7 +387,7 @@ void WilsonKernels<Impl>::DhopDirKernel( StencilImpl &st, DoubledGaugeField &U,S
    autoView(in_v ,in ,AcceleratorRead);
    autoView(out_v,out,AcceleratorWrite);
    autoView(st_v ,st ,AcceleratorRead);
-   auto CBp=st.CommBuf();
+   auto CBp=st.CommBuf();			
 #define LoopBody(Dir)				\
    case Dir :					\
      accelerator_for(ss,Nsite,Simd::Nsimd(),{	\
@@ -414,7 +414,7 @@ void WilsonKernels<Impl>::DhopDirKernel( StencilImpl &st, DoubledGaugeField &U,S
      break;
    }
 #undef LoopBody
-}
+} 
 
 #define KERNEL_CALLNB(A) \
   const uint64_t    NN = Nsite*Ls;					\
@@ -424,7 +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, {						\
@@ -436,14 +436,14 @@ 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) 
 {
     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_CUDA
      if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSite);    return;}
@@ -455,7 +455,7 @@ void WilsonKernels<Impl>::DhopKernel(int Opt,StencilImpl &st,  DoubledGaugeField
      if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALLNB(HandDhopSiteInt);    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_CUDA
      if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSiteExt);    return;}
@@ -467,14 +467,14 @@ 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) 
   {
     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_CUDA
      if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSiteDag);    return;}
@@ -486,7 +486,7 @@ void WilsonKernels<Impl>::DhopKernel(int Opt,StencilImpl &st,  DoubledGaugeField
      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_CUDA
      if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSiteDagExt);    return;}
@@ -501,3 +501,4 @@ void WilsonKernels<Impl>::DhopKernel(int Opt,StencilImpl &st,  DoubledGaugeField
 #undef ASM_CALL
 
 NAMESPACE_END(Grid);
+

Grid/qcd/action/fermion/instantiation/WilsonAdjImplD/WilsonKernelsInstantiationWilsonAdjImplD.cc

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin 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/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/instantiation/WilsonAdjImplD/WilsonKernelsInstantiationWilsonAdjImplD.cc

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

Grid/qcd/action/fermion/instantiation/WilsonAdjImplF/WilsonKernelsInstantiationWilsonAdjImplF.cc

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin 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/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/instantiation/WilsonAdjImplF/WilsonKernelsInstantiationWilsonAdjImplF.cc

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

Grid/qcd/action/fermion/instantiation/WilsonImplD/WilsonKernelsInstantiationWilsonImplD.cc

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin 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/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/instantiation/WilsonImplD/WilsonKernelsInstantiationWilsonImplD.cc

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

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

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin 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/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

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

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

Grid/qcd/action/fermion/instantiation/WilsonImplF/WilsonKernelsInstantiationWilsonImplF.cc

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin 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/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/instantiation/WilsonImplF/WilsonKernelsInstantiationWilsonImplF.cc

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

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

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin 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/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

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

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

Grid/qcd/action/fermion/instantiation/WilsonKernelsInstantiation.cc.master

@@ -4,12 +4,11 @@ Grid physics library, www.github.com/paboyle/Grid
 
 Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
 
-Copyright (C) 2015, 2020
+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>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
 
 This program is free software; you can redistribute it and/or modify
 it under the terms of the GNU General Public License as published by
@@ -35,17 +34,14 @@ directory
 
 #ifndef AVX512
 #ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
 #include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
 #endif
 #endif
-#endif
-#endif
 
 NAMESPACE_BEGIN(Grid);
 
 #include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
+template class WilsonKernels<IMPLEMENTATION>; 
 
 NAMESPACE_END(Grid);
+

Grid/qcd/action/fermion/instantiation/WilsonKernelsInstantiationAsm.cc

@@ -37,7 +37,6 @@ directory
 ////////////////////////////////////////////////////////////////////////
 NAMESPACE_BEGIN(Grid);
 #include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmAvx512.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmA64FX.h>
 #include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmQPX.h>
 NAMESPACE_END(Grid);
 

Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplD/WilsonKernelsInstantiationWilsonTwoIndexAntiSymmetricImplD.cc

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin 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/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplD/WilsonKernelsInstantiationWilsonTwoIndexAntiSymmetricImplD.cc

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

Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplF/WilsonKernelsInstantiationWilsonTwoIndexAntiSymmetricImplF.cc

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin 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/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplF/WilsonKernelsInstantiationWilsonTwoIndexAntiSymmetricImplF.cc

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

Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplD/WilsonKernelsInstantiationWilsonTwoIndexSymmetricImplD.cc

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin 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/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplD/WilsonKernelsInstantiationWilsonTwoIndexSymmetricImplD.cc

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

Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplF/WilsonKernelsInstantiationWilsonTwoIndexSymmetricImplF.cc

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin 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/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplF/WilsonKernelsInstantiationWilsonTwoIndexSymmetricImplF.cc

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

Grid/qcd/action/fermion/instantiation/ZWilsonImplD/WilsonKernelsInstantiationZWilsonImplD.cc

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin 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/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/instantiation/ZWilsonImplD/WilsonKernelsInstantiationZWilsonImplD.cc

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

Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/WilsonKernelsInstantiationZWilsonImplDF.cc

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin 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/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/instantiation/ZWilsonImplDF/WilsonKernelsInstantiationZWilsonImplDF.cc

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

Grid/qcd/action/fermion/instantiation/ZWilsonImplF/WilsonKernelsInstantiationZWilsonImplF.cc

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin 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/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/instantiation/ZWilsonImplF/WilsonKernelsInstantiationZWilsonImplF.cc

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

Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/WilsonKernelsInstantiationZWilsonImplFH.cc

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin 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/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/instantiation/ZWilsonImplFH/WilsonKernelsInstantiationZWilsonImplFH.cc

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

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) {

Grid/qcd/hmc/GenericHMCrunner.h

@@ -140,7 +140,17 @@ private:
 
     // Can move this outside?
     typedef IntegratorType<SmearingPolicy> TheIntegrator;
-    TheIntegrator MDynamics(UGrid, Parameters.MD, TheAction, Smearing);
+    // Metric
+    //TrivialMetric<typename Implementation::Field> Mtr;
+    ConjugateGradient<LatticeGaugeField> CG(1.0e-8,10000);
+    LaplacianParams LapPar(0.0001, 1.0, 10000, 1e-8, 12, 64);
+//    RealD Kappa = 1.2;
+    RealD Kappa = Parameters.Kappa;
+    std::cout << GridLogMessage << "Kappa = " << Kappa << std::endl;
+
+    // Better to pass the generalised momenta to the integrator
+    LaplacianAdjointField<PeriodicGimplR> Laplacian(UGrid, CG, LapPar, Kappa);
+    TheIntegrator MDynamics(UGrid, Parameters.MD, TheAction, Smearing, Laplacian);
 
     if (Parameters.StartingType == "HotStart") {
       // Hot start

Grid/qcd/hmc/HMC.h

@@ -53,6 +53,7 @@ struct HMCparameters: Serializable {
                                   bool, MetropolisTest,
                                   Integer, NoMetropolisUntil,
                                   std::string, StartingType,
+				  RealD, Kappa,
                                   IntegratorParameters, MD)
 
   HMCparameters() {

Grid/qcd/hmc/integrators/Integrator.h

@@ -73,7 +73,8 @@ protected:
   double t_U;  // Track time passing on each level and for U and for P
   std::vector<double> t_P;  
 
-  MomentaField P;
+//  MomentaField P;
+  GeneralisedMomenta<FieldImplementation > P;
   SmearingPolicy& Smearer;
   RepresentationPolicy Representations;
   IntegratorParameters Params;
@@ -83,7 +84,7 @@ protected:
   void update_P(Field& U, int level, double ep) 
   {
     t_P[level] += ep;
-    update_P(P, U, level, ep);
+    update_P(P.Mom, U, level, ep);
 
     std::cout << GridLogIntegrator << "[" << level << "] P " << " dt " << ep << " : t_P " << t_P[level] << std::endl;
   }
@@ -111,6 +112,21 @@ protected:
     // input U actually not used in the fundamental case
     // Fundamental updates, include smearing
 
+    // Generalised momenta  
+    // Derivative of the kinetic term must be computed before
+    // Mom is the momenta and gets updated by the 
+    // actions derivatives
+    MomentaField MomDer(P.Mom.Grid());
+    P.M.ImportGauge(U);
+    P.DerivativeU(P.Mom, MomDer);
+    Mom -= MomDer * ep;
+
+    // Auxiliary fields
+    P.update_auxiliary_momenta(ep*0.5);
+    P.AuxiliaryFieldsDerivative(MomDer);
+    Mom -= MomDer * ep;
+    P.update_auxiliary_momenta(ep*0.5);
+
     for (int a = 0; a < as[level].actions.size(); ++a) {
       double start_full = usecond();
       Field force(U.Grid());
@@ -137,9 +153,83 @@ protected:
     as[level].apply(update_P_hireps, Representations, Mom, U, ep);
   }
 
+  void implicit_update_P(Field& U, int level, double ep, bool intermediate = false) {
+    t_P[level] += ep;
+
+    std::cout << GridLogIntegrator << "[" << level << "] P "
+              << " dt " << ep << " : t_P " << t_P[level] << std::endl;
+    // Fundamental updates, include smearing
+    MomentaField Msum(P.Mom.Grid());
+    Msum = Zero();
+    for (int a = 0; a < as[level].actions.size(); ++a) {
+      // Compute the force terms for the lagrangian part
+      // We need to compute the derivative of the actions
+      // only once
+      Field force(U.Grid());
+      conformable(U.Grid(), P.Mom.Grid());
+      Field& Us = Smearer.get_U(as[level].actions.at(a)->is_smeared);
+      as[level].actions.at(a)->deriv(Us, force);  // deriv should NOT include Ta
+
+      std::cout << GridLogIntegrator << "Smearing (on/off): " << as[level].actions.at(a)->is_smeared << std::endl;
+      if (as[level].actions.at(a)->is_smeared) Smearer.smeared_force(force);
+      force = FieldImplementation::projectForce(force);  // Ta for gauge fields
+      Real force_abs = std::sqrt(norm2(force) / U.Grid()->gSites());
+      std::cout << GridLogIntegrator << "|Force| site average: " << force_abs
+                << std::endl;
+      Msum += force;
+    }
+
+    MomentaField NewMom = P.Mom;
+    MomentaField OldMom = P.Mom;
+    double threshold = 1e-8;
+    P.M.ImportGauge(U);
+    MomentaField MomDer(P.Mom.Grid());
+    MomentaField MomDer1(P.Mom.Grid());
+    MomentaField AuxDer(P.Mom.Grid());
+    MomDer1 = Zero();
+    MomentaField diff(P.Mom.Grid());
+    double factor = 2.0;
+    if (intermediate){
+      P.DerivativeU(P.Mom, MomDer1);
+      factor = 1.0;
+    }
+
+    // Auxiliary fields
+    P.update_auxiliary_momenta(ep*0.5);
+    P.AuxiliaryFieldsDerivative(AuxDer);
+    Msum += AuxDer;
+    
+
+    // Here run recursively
+    int counter = 1;
+    RealD RelativeError;
+    do {
+      std::cout << GridLogIntegrator << "UpdateP implicit step "<< counter << std::endl;
+
+      // Compute the derivative of the kinetic term
+      // with respect to the gauge field
+      P.DerivativeU(NewMom, MomDer);
+      Real force_abs = std::sqrt(norm2(MomDer) / U.Grid()->gSites());
+      std::cout << GridLogIntegrator << "|Force| laplacian site average: " << force_abs
+                << std::endl;
+
+      NewMom = P.Mom - ep* 0.5 * (2.0*Msum + factor*MomDer + MomDer1);// simplify
+      diff = NewMom - OldMom;
+      counter++;
+      RelativeError = std::sqrt(norm2(diff))/std::sqrt(norm2(NewMom));
+      std::cout << GridLogIntegrator << "UpdateP RelativeError: " << RelativeError << std::endl;
+      OldMom = NewMom;
+    } while (RelativeError > threshold);
+
+    P.Mom = NewMom;
+
+    // update the auxiliary fields momenta    
+    P.update_auxiliary_momenta(ep*0.5);
+  }
+
   void update_U(Field& U, double ep) 
   {
-    update_U(P, U, ep);
+    update_U(P.Mom, U, ep);
 
     t_U += ep;
     int fl = levels - 1;
@@ -158,15 +248,64 @@ protected:
     Representations.update(U);  // void functions if fundamental representation
   }
 
+  void implicit_update_U(Field&U, double ep){
+    t_U += ep;
+    int fl = levels - 1;
+    std::cout << GridLogIntegrator << "   " << "[" << fl << "] U " << " dt " << ep << " : t_U " << t_U << std::endl;
+
+    MomentaField Mom1(P.Mom.Grid());
+    MomentaField Mom2(P.Mom.Grid());
+    RealD RelativeError;
+    Field diff(U.Grid());
+    Real threshold = 1e-8;
+    int counter = 1;
+    int MaxCounter = 100;
+
+    Field OldU = U;
+    Field NewU = U;
+
+    P.M.ImportGauge(U);
+    P.DerivativeP(Mom1); // first term in the derivative 
+
+    P.update_auxiliary_fields(ep*0.5);
+
+
+    MomentaField sum=Mom1;
+    do {
+      std::cout << GridLogIntegrator << "UpdateU implicit step "<< counter << std::endl;
+      
+      P.DerivativeP(Mom2); // second term in the derivative, on the updated U
+      sum = (Mom1 + Mom2);
+
+      for (int mu = 0; mu < Nd; mu++) {
+        auto Umu = PeekIndex<LorentzIndex>(U, mu);
+        auto Pmu = PeekIndex<LorentzIndex>(sum, mu);
+        Umu = expMat(Pmu, ep * 0.5, 12) * Umu;
+        PokeIndex<LorentzIndex>(NewU, ProjectOnGroup(Umu), mu);
+      }
+
+      diff = NewU - OldU;
+      RelativeError = std::sqrt(norm2(diff))/std::sqrt(norm2(NewU));
+      std::cout << GridLogIntegrator << "UpdateU RelativeError: " << RelativeError << std::endl;
+      
+      P.M.ImportGauge(NewU);
+      OldU = NewU; // some redundancy to be eliminated
+      counter++;
+    } while (RelativeError > threshold && counter < MaxCounter);
+
+    U = NewU;
+    P.update_auxiliary_fields(ep*0.5);
+  }
+
   virtual void step(Field& U, int level, int first, int last) = 0;
 
 public:
   Integrator(GridBase* grid, IntegratorParameters Par,
              ActionSet<Field, RepresentationPolicy>& Aset,
-             SmearingPolicy& Sm)
+             SmearingPolicy& Sm, Metric<MomentaField>& M)
     : Params(Par),
       as(Aset),
-      P(grid),
+      P(grid, M),
       levels(Aset.size()),
       Smearer(Sm),
       Representations(grid) 
@@ -203,7 +342,9 @@ public:
 
   void reverse_momenta()
   {
-    P *= -1.0;
+//    P *= -1.0;
+    P.Mom *= -1.0;
+    P.AuxMom *= -1.0;
   }
 
   // to be used by the actionlevel class to iterate
@@ -223,10 +364,13 @@ public:
   // Initialization of momenta and actions
   void refresh(Field& U, GridParallelRNG& pRNG) 
   {
-    assert(P.Grid() == U.Grid());
+    assert(P.Mom.Grid() == U.Grid());
     std::cout << GridLogIntegrator << "Integrator refresh\n";
 
-    FieldImplementation::generate_momenta(P, pRNG);
+//    FieldImplementation::generate_momenta(P.Mom, pRNG);
+    P.M.ImportGauge(U);
+    P.MomentaDistribution(pRNG);
+
 
     // Update the smeared fields, can be implemented as observer
     // necessary to keep the fields updated even after a reject
@@ -272,9 +416,11 @@ public:
 
     std::cout << GridLogIntegrator << "Integrator action\n";
 
-    RealD H = - FieldImplementation::FieldSquareNorm(P)/HMC_MOMENTUM_DENOMINATOR; // - trace (P*P)/denom
-
+//    RealD H = - FieldImplementation::FieldSquareNorm(P)/HMC_MOMENTUM_DENOMINATOR; // - trace (P*P)/denom
+    P.M.ImportGauge(U);
+    RealD H = - P.MomentaAction();
     RealD Hterm;
+    std::cout << GridLogMessage << "Momentum action H_p = " << H << "\n";
 
     // Actions
     for (int level = 0; level < as.size(); ++level) {
@@ -301,9 +447,9 @@ public:
       t_P[level] = 0;
     }
 
-    for (int stp = 0; stp < Params.MDsteps; ++stp) {  // MD step
-      int first_step = (stp == 0);
-      int last_step = (stp == Params.MDsteps - 1);
+    for (int step = 0; step < Params.MDsteps; ++step) {  // MD step
+      int first_step = (step == 0);
+      int last_step = (step == Params.MDsteps - 1);
       this->step(U, 0, first_step, last_step);
     }
 

Grid/qcd/hmc/integrators/Integrator_algorithm.h

@@ -101,8 +101,8 @@ public:
 
   std::string integrator_name(){return "LeapFrog";}
 
-  LeapFrog(GridBase* grid, IntegratorParameters Par, ActionSet<Field, RepresentationPolicy>& Aset, SmearingPolicy& Sm)
-    : Integrator<FieldImplementation, SmearingPolicy, RepresentationPolicy>(grid, Par, Aset, Sm){};
+  LeapFrog(GridBase* grid, IntegratorParameters Par, ActionSet<Field, RepresentationPolicy>& Aset, SmearingPolicy& Sm, Metric<Field>& M)
+    : Integrator<FieldImplementation, SmearingPolicy, RepresentationPolicy>(grid, Par, Aset, Sm,M){};
 
   void step(Field& U, int level, int _first, int _last) {
     int fl = this->as.size() - 1;
@@ -144,8 +144,8 @@ private:
 public:
   INHERIT_FIELD_TYPES(FieldImplementation);
 
-  MinimumNorm2(GridBase* grid, IntegratorParameters Par, ActionSet<Field, RepresentationPolicy>& Aset, SmearingPolicy& Sm)
-    : Integrator<FieldImplementation, SmearingPolicy, RepresentationPolicy>(grid, Par, Aset, Sm){};
+  MinimumNorm2(GridBase* grid, IntegratorParameters Par, ActionSet<Field, RepresentationPolicy>& Aset, SmearingPolicy& Sm, Metric<Field>& M)
+    : Integrator<FieldImplementation, SmearingPolicy, RepresentationPolicy>(grid, Par, Aset, Sm,M){};
 
   std::string integrator_name(){return "MininumNorm2";}
 
@@ -207,9 +207,9 @@ public:
   // Looks like dH scales as dt^4. tested wilson/wilson 2 level.
   ForceGradient(GridBase* grid, IntegratorParameters Par,
                 ActionSet<Field, RepresentationPolicy>& Aset,
-                SmearingPolicy& Sm)
+                SmearingPolicy& Sm, Metric<Field>& M)
     : Integrator<FieldImplementation, SmearingPolicy, RepresentationPolicy>(
-									    grid, Par, Aset, Sm){};
+									    grid, Par, Aset, Sm,M){};
 
   std::string integrator_name(){return "ForceGradient";}
   
@@ -271,6 +271,139 @@ public:
   }
 };
 
+////////////////////////////////
+// Riemannian Manifold HMC
+// Girolami et al
+////////////////////////////////
+
+
+
+// correct
+template <class FieldImplementation, class SmearingPolicy,
+          class RepresentationPolicy =
+              Representations<FundamentalRepresentation> >
+class ImplicitLeapFrog : public Integrator<FieldImplementation, SmearingPolicy,
+                                           RepresentationPolicy> {
+ public:
+  typedef ImplicitLeapFrog<FieldImplementation, SmearingPolicy, RepresentationPolicy>
+      Algorithm;
+  INHERIT_FIELD_TYPES(FieldImplementation);
+
+  // Riemannian manifold metric operator
+  // Hermitian operator Fisher
+
+  std::string integrator_name(){return "ImplicitLeapFrog";}
+
+  ImplicitLeapFrog(GridBase* grid, IntegratorParameters Par,
+           ActionSet<Field, RepresentationPolicy>& Aset, SmearingPolicy& Sm, Metric<Field>& M)
+      : Integrator<FieldImplementation, SmearingPolicy, RepresentationPolicy>(
+            grid, Par, Aset, Sm, M){};
+
+  void step(Field& U, int level, int _first, int _last) {
+    int fl = this->as.size() - 1;
+    // level  : current level
+    // fl     : final level
+    // eps    : current step size
+
+    // Get current level step size
+    RealD eps = this->Params.trajL/this->Params.MDsteps;
+    for (int l = 0; l <= level; ++l) eps /= this->as[l].multiplier;
+
+    int multiplier = this->as[level].multiplier;
+    for (int e = 0; e < multiplier; ++e) {
+      int first_step = _first && (e == 0);
+      int last_step = _last && (e == multiplier - 1);
+
+      if (first_step) {  // initial half step
+       this->implicit_update_P(U, level, eps / 2.0);
+      }
+
+      if (level == fl) {  // lowest level
+        this->implicit_update_U(U, eps);
+      } else {  // recursive function call
+        this->step(U, level + 1, first_step, last_step);
+      }
+
+      //int mm = last_step ? 1 : 2;
+      if (last_step){
+        this->update_P(U, level, eps / 2.0);
+      } else {
+      this->implicit_update_P(U, level, eps, true);// works intermediate step
+      //  this->update_P(U, level, eps); // looks not reversible
+      }
+    }
+  }
+};
+
+
+// This is not completely tested
+template <class FieldImplementation, class SmearingPolicy,
+          class RepresentationPolicy =
+              Representations<FundamentalRepresentation> >
+class ImplicitMinimumNorm2 : public Integrator<FieldImplementation, SmearingPolicy,
+                                       RepresentationPolicy> {
+ private:
+  const RealD lambda = 0.1931833275037836;
+
+ public:
+  INHERIT_FIELD_TYPES(FieldImplementation);
+
+  ImplicitMinimumNorm2(GridBase* grid, IntegratorParameters Par,
+               ActionSet<Field, RepresentationPolicy>& Aset, SmearingPolicy& Sm, Metric<Field>& M)
+      : Integrator<FieldImplementation, SmearingPolicy, RepresentationPolicy>(
+            grid, Par, Aset, Sm, M){};
+
+  std::string integrator_name(){return "ImplicitMininumNorm2";}
+
+  void step(Field& U, int level, int _first, int _last) {
+    // level  : current level
+    // fl     : final level
+    // eps    : current step size
+
+    int fl = this->as.size() - 1;
+
+    RealD eps = this->Params.trajL/this->Params.MDsteps * 2.0;
+    for (int l = 0; l <= level; ++l) eps /= 2.0 * this->as[l].multiplier;
+
+    // Nesting:  2xupdate_U of size eps/2
+    // Next level is eps/2/multiplier
+
+    int multiplier = this->as[level].multiplier;
+    for (int e = 0; e < multiplier; ++e) {  // steps per step
+
+      int first_step = _first && (e == 0);
+      int last_step = _last && (e == multiplier - 1);
+
+      if (first_step) {  // initial half step
+        this->implicit_update_P(U, level, lambda * eps);
+      }
+
+      if (level == fl) {  // lowest level
+        this->implicit_update_U(U, 0.5 * eps);
+      } else {  // recursive function call
+        this->step(U, level + 1, first_step, 0);
+      }
+
+      this->implicit_update_P(U, level, (1.0 - 2.0 * lambda) * eps, true);
+
+      if (level == fl) {  // lowest level
+        this->implicit_update_U(U, 0.5 * eps);
+      } else {  // recursive function call
+        this->step(U, level + 1, 0, last_step);
+      }
+
+      //int mm = (last_step) ? 1 : 2;
+      //this->update_P(U, level, lambda * eps * mm);
+
+      if (last_step) {
+        this->update_P(U, level, eps * lambda);
+      } else {
+        this->implicit_update_P(U, level, lambda * eps*2.0, true);
+      }
+    }
+  }
+};
+
 NAMESPACE_END(Grid);
 
 #endif  // INTEGRATOR_INCLUDED

Grid/qcd/spin/TwoSpinor.h

@@ -128,6 +128,7 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void s
 }
 template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void spProjTm (iVector<vtype,Nhs> &hspin,const iVector<vtype,Ns> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
   hspin(0)=fspin(0)-fspin(2);
   hspin(1)=fspin(1)-fspin(3);
 }
@@ -137,50 +138,40 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void s
  *  0 0 -1  0
  *  0 0  0 -1
  */
+
 template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void spProj5p (iVector<vtype,Nhs> &hspin,const iVector<vtype,Ns> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
   hspin(0)=fspin(0);
   hspin(1)=fspin(1);
 }
 
 template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void spProj5m (iVector<vtype,Nhs> &hspin,const iVector<vtype,Ns> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
   hspin(0)=fspin(2);
   hspin(1)=fspin(3);
 }
   
+//  template<class vtype> accelerator_inline void fspProj5p (iVector<vtype,Ns> &rfspin,const iVector<vtype,Ns> &fspin)
 template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void spProj5p (iVector<vtype,Ns> &rfspin,const iVector<vtype,Ns> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
   rfspin(0)=fspin(0);
   rfspin(1)=fspin(1);
   rfspin(2)=Zero();
   rfspin(3)=Zero();
 }
+//  template<class vtype> accelerator_inline void fspProj5m (iVector<vtype,Ns> &rfspin,const iVector<vtype,Ns> &fspin)
 template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void spProj5m (iVector<vtype,Ns> &rfspin,const iVector<vtype,Ns> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
   rfspin(0)=Zero();
   rfspin(1)=Zero();
   rfspin(2)=fspin(2);
   rfspin(3)=fspin(3);
 }
 
-template<class vtype,int N,IfCoarsened<iVector<vtype,N> > = 0> accelerator_inline void spProj5p (iVector<vtype,N> &rfspin,const iVector<vtype,N> &fspin)
-{
-  const int hN = N>>1;
-  for(int s=0;s<hN;s++){
-    rfspin(s)=fspin(s);
-    rfspin(s+hN)=Zero();
-  }
-}
-template<class vtype,int N,IfCoarsened<iVector<vtype,N> > = 0> accelerator_inline void spProj5m (iVector<vtype,N> &rfspin,const iVector<vtype,N> &fspin)
-{
-  const int hN = N>>1;
-  for(int s=0;s<hN;s++){
-    rfspin(s)=Zero();
-    rfspin(s+hN)=fspin(s+hN);
-  }
-}
-
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 // Reconstruction routines to move back again to four spin
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////
@@ -192,6 +183,7 @@ template<class vtype,int N,IfCoarsened<iVector<vtype,N> > = 0> accelerator_inlin
  */
 template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void spReconXp (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
   fspin(0)=hspin(0);
   fspin(1)=hspin(1);
   fspin(2)=timesMinusI(hspin(1));
@@ -199,6 +191,7 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void s
 }
 template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void spReconXm (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
   fspin(0)=hspin(0);
   fspin(1)=hspin(1);
   fspin(2)=timesI(hspin(1));
@@ -206,6 +199,7 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void s
 }
 template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void accumReconXp (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
   fspin(0)+=hspin(0);
   fspin(1)+=hspin(1);
   fspin(2)-=timesI(hspin(1));
@@ -213,6 +207,7 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void a
 }
 template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void accumReconXm (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
   fspin(0)+=hspin(0);
   fspin(1)+=hspin(1);
   fspin(2)+=timesI(hspin(1));
@@ -226,6 +221,7 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void a
 
 template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void spReconYp (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
   fspin(0)=hspin(0);
   fspin(1)=hspin(1);
   fspin(2)= hspin(1);
@@ -233,6 +229,7 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void s
 }
 template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void spReconYm (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
   fspin(0)=hspin(0);
   fspin(1)=hspin(1);
   fspin(2)=-hspin(1);
@@ -240,6 +237,7 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void s
 }
 template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void accumReconYp (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
   fspin(0)+=hspin(0);
   fspin(1)+=hspin(1);
   fspin(2)+=hspin(1);
@@ -247,6 +245,7 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void a
 }
 template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void accumReconYm (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
   fspin(0)+=hspin(0);
   fspin(1)+=hspin(1);
   fspin(2)-=hspin(1);
@@ -261,6 +260,7 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void a
  */
 template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void spReconZp (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
   fspin(0)=hspin(0);
   fspin(1)=hspin(1);
   fspin(2)=timesMinusI(hspin(0));
@@ -268,6 +268,7 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void s
 }
 template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void spReconZm (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
   fspin(0)=hspin(0);
   fspin(1)=hspin(1);
   fspin(2)=     timesI(hspin(0));
@@ -275,6 +276,7 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void s
 }
 template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void accumReconZp (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
   fspin(0)+=hspin(0);
   fspin(1)+=hspin(1);
   fspin(2)-=timesI(hspin(0));
@@ -282,6 +284,7 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void a
 }
 template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void accumReconZm (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
   fspin(0)+=hspin(0);
   fspin(1)+=hspin(1);
   fspin(2)+=timesI(hspin(0));
@@ -295,6 +298,7 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void a
  */
 template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void spReconTp (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
   fspin(0)=hspin(0);
   fspin(1)=hspin(1);
   fspin(2)=hspin(0);
@@ -302,6 +306,7 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void s
 }
 template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void spReconTm (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
   fspin(0)=hspin(0);
   fspin(1)=hspin(1);
   fspin(2)=-hspin(0);
@@ -309,6 +314,7 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void s
 }
 template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void accumReconTp (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
   fspin(0)+=hspin(0);
   fspin(1)+=hspin(1);
   fspin(2)+=hspin(0);
@@ -316,6 +322,7 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void a
 }
 template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void accumReconTm (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
   fspin(0)+=hspin(0);
   fspin(1)+=hspin(1);
   fspin(2)-=hspin(0);
@@ -329,6 +336,7 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void a
  */
 template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void spRecon5p (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
   fspin(0)=hspin(0)+hspin(0); // add is lower latency than mul
   fspin(1)=hspin(1)+hspin(1); // probably no measurable diffence though
   fspin(2)=Zero();
@@ -336,6 +344,7 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void s
 }
 template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void spRecon5m (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
   fspin(0)=Zero();
   fspin(1)=Zero();
   fspin(2)=hspin(0)+hspin(0);
@@ -343,6 +352,7 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void s
 }
 template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void accumRecon5p (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
   fspin(0)+=hspin(0)+hspin(0);
   fspin(1)+=hspin(1)+hspin(1);
 }
@@ -362,6 +372,7 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> accelerator_inline void a
 //////////
 template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spProjXp (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
   for(int i=0;i<N;i++) {
     spProjXp(hspin._internal[i],fspin._internal[i]);
   }
@@ -415,21 +426,26 @@ template<class rtype,class vtype,int N> accelerator_inline void accumReconXp (iM
     }}
 }
 
+
+
 ////////
 // Xm
 ////////
 template<class rtype,class vtype> accelerator_inline void spProjXm (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
   spProjXm(hspin._internal,fspin._internal);
 }
 template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spProjXm (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
   for(int i=0;i<N;i++) {
     spProjXm(hspin._internal[i],fspin._internal[i]);
   }
 }
 template<class rtype,class vtype,int N> accelerator_inline void spProjXm (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
   for(int i=0;i<N;i++){ 
     for(int j=0;j<N;j++){
       spProjXm(hspin._internal[i][j],fspin._internal[i][j]);
@@ -439,16 +455,19 @@ template<class rtype,class vtype,int N> accelerator_inline void spProjXm (iMatri
 
 template<class rtype,class vtype> accelerator_inline void spReconXm (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
   spReconXm(hspin._internal,fspin._internal);
 }
 template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spReconXm (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
   for(int i=0;i<N;i++) {
     spReconXm(hspin._internal[i],fspin._internal[i]);
   }
 }
 template<class rtype,class vtype,int N> accelerator_inline void spReconXm (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
   for(int i=0;i<N;i++){ 
     for(int j=0;j<N;j++){
       spReconXm(hspin._internal[i][j],fspin._internal[i][j]);
@@ -457,37 +476,45 @@ template<class rtype,class vtype,int N> accelerator_inline void spReconXm (iMatr
 
 template<class rtype,class vtype> accelerator_inline void accumReconXm (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
   accumReconXm(hspin._internal,fspin._internal);
 }
 template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void accumReconXm (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
   for(int i=0;i<N;i++) {
     accumReconXm(hspin._internal[i],fspin._internal[i]);
   }
 }
 template<class rtype,class vtype,int N> accelerator_inline void accumReconXm (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
   for(int i=0;i<N;i++){ 
     for(int j=0;j<N;j++){
       accumReconXm(hspin._internal[i][j],fspin._internal[i][j]);
     }}
 }
 
+
+
 ////////
 // Yp
 ////////
 template<class rtype,class vtype> accelerator_inline void spProjYp (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
   spProjYp(hspin._internal,fspin._internal);
 }
 template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spProjYp (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
   for(int i=0;i<N;i++) {
     spProjYp(hspin._internal[i],fspin._internal[i]);
   }
 }
 template<class rtype,class vtype,int N> accelerator_inline void spProjYp (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
   for(int i=0;i<N;i++){ 
     for(int j=0;j<N;j++){
       spProjYp(hspin._internal[i][j],fspin._internal[i][j]);
@@ -497,16 +524,19 @@ template<class rtype,class vtype,int N> accelerator_inline void spProjYp (iMatri
 
 template<class rtype,class vtype> accelerator_inline void spReconYp (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
   spReconYp(hspin._internal,fspin._internal);
 }
 template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spReconYp (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
   for(int i=0;i<N;i++) {
     spReconYp(hspin._internal[i],fspin._internal[i]);
   }
 }
 template<class rtype,class vtype,int N> accelerator_inline void spReconYp (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
   for(int i=0;i<N;i++){ 
     for(int j=0;j<N;j++){
       spReconYp(hspin._internal[i][j],fspin._internal[i][j]);
@@ -515,55 +545,66 @@ template<class rtype,class vtype,int N> accelerator_inline void spReconYp (iMatr
 
 template<class rtype,class vtype> accelerator_inline void accumReconYp (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
   accumReconYp(hspin._internal,fspin._internal);
 }
 template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void accumReconYp (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
   for(int i=0;i<N;i++) {
     accumReconYp(hspin._internal[i],fspin._internal[i]);
   }
 }
 template<class rtype,class vtype,int N> accelerator_inline void accumReconYp (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
   for(int i=0;i<N;i++){ 
     for(int j=0;j<N;j++){
       accumReconYp(hspin._internal[i][j],fspin._internal[i][j]);
     }}
 }
 
+
 ////////
 // Ym
 ////////
 template<class rtype,class vtype> accelerator_inline void spProjYm (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
   spProjYm(hspin._internal,fspin._internal);
 }
 template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spProjYm (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
   for(int i=0;i<N;i++) {
     spProjYm(hspin._internal[i],fspin._internal[i]);
   }
 }
 template<class rtype,class vtype,int N> accelerator_inline void spProjYm (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
   for(int i=0;i<N;i++){ 
     for(int j=0;j<N;j++){
       spProjYm(hspin._internal[i][j],fspin._internal[i][j]);
     }}
 }
 
+
 template<class rtype,class vtype> accelerator_inline void spReconYm (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
   spReconYm(hspin._internal,fspin._internal);
 }
 template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spReconYm (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,const iVector<vtype,N> >::type *temp;
   for(int i=0;i<N;i++) {
     spReconYm(hspin._internal[i],fspin._internal[i]);
   }
 }
 template<class rtype,class vtype,int N> accelerator_inline void spReconYm (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
   for(int i=0;i<N;i++){ 
     for(int j=0;j<N;j++){
       spReconYm(hspin._internal[i][j],fspin._internal[i][j]);
@@ -572,16 +613,19 @@ template<class rtype,class vtype,int N> accelerator_inline void spReconYm (iMatr
 
 template<class rtype,class vtype> accelerator_inline void accumReconYm (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
   accumReconYm(hspin._internal,fspin._internal);
 }
 template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void accumReconYm (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
   for(int i=0;i<N;i++) {
     accumReconYm(hspin._internal[i],fspin._internal[i]);
   }
 }
 template<class rtype,class vtype,int N> accelerator_inline void accumReconYm (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
   for(int i=0;i<N;i++){ 
     for(int j=0;j<N;j++){
       accumReconYm(hspin._internal[i][j],fspin._internal[i][j]);
@@ -594,57 +638,66 @@ template<class rtype,class vtype,int N> accelerator_inline void accumReconYm (iM
 ////////
 template<class rtype,class vtype> accelerator_inline void spProjZp (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
   spProjZp(hspin._internal,fspin._internal);
 }
 template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spProjZp (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
   for(int i=0;i<N;i++) {
     spProjZp(hspin._internal[i],fspin._internal[i]);
   }
 }
 template<class rtype,class vtype,int N> accelerator_inline void spProjZp (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
   for(int i=0;i<N;i++){ 
     for(int j=0;j<N;j++){
       spProjZp(hspin._internal[i][j],fspin._internal[i][j]);
-  }}
+    }}
 }
 
 
 template<class rtype,class vtype> accelerator_inline void spReconZp (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
   spReconZp(hspin._internal,fspin._internal);
 }
 template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spReconZp (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
   for(int i=0;i<N;i++) {
     spReconZp(hspin._internal[i],fspin._internal[i]);
   }
 }
 template<class rtype,class vtype,int N> accelerator_inline void spReconZp (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
   for(int i=0;i<N;i++){ 
     for(int j=0;j<N;j++){
       spReconZp(hspin._internal[i][j],fspin._internal[i][j]);
-  }}
+    }}
 }
 
 template<class rtype,class vtype> accelerator_inline void accumReconZp (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
   accumReconZp(hspin._internal,fspin._internal);
 }
 template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void accumReconZp (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
   for(int i=0;i<N;i++) {
     accumReconZp(hspin._internal[i],fspin._internal[i]);
   }
 }
 template<class rtype,class vtype,int N> accelerator_inline void accumReconZp (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
   for(int i=0;i<N;i++){ 
     for(int j=0;j<N;j++){
       accumReconZp(hspin._internal[i][j],fspin._internal[i][j]);
-  }}
+    }}
 }
 
 
@@ -653,53 +706,62 @@ template<class rtype,class vtype,int N> accelerator_inline void accumReconZp (iM
 ////////
 template<class rtype,class vtype> accelerator_inline void spProjZm (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
   spProjZm(hspin._internal,fspin._internal);
 }
 template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spProjZm (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
   for(int i=0;i<N;i++) {
     spProjZm(hspin._internal[i],fspin._internal[i]);
   }
 }
 template<class rtype,class vtype,int N> accelerator_inline void spProjZm (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
   for(int i=0;i<N;i++){ 
     for(int j=0;j<N;j++){
       spProjZm(hspin._internal[i][j],fspin._internal[i][j]);
-  }}
+    }}
 }
 
 
 template<class rtype,class vtype> accelerator_inline void spReconZm (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
   spReconZm(hspin._internal,fspin._internal);
 }
 template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spReconZm (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
   for(int i=0;i<N;i++) {
     spReconZm(hspin._internal[i],fspin._internal[i]);
   }
 }
 template<class rtype,class vtype,int N> accelerator_inline void spReconZm (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
   for(int i=0;i<N;i++){ 
     for(int j=0;j<N;j++){
       spReconZm(hspin._internal[i][j],fspin._internal[i][j]);
-  }}
+    }}
 }
 
 template<class rtype,class vtype> accelerator_inline void accumReconZm (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
   accumReconZm(hspin._internal,fspin._internal);
 }
 template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void accumReconZm (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
   for(int i=0;i<N;i++) {
     accumReconZm(hspin._internal[i],fspin._internal[i]);
   }
 }
 template<class rtype,class vtype,int N> accelerator_inline void accumReconZm (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
   for(int i=0;i<N;i++){ 
     for(int j=0;j<N;j++){
       accumReconZm(hspin._internal[i][j],fspin._internal[i][j]);
@@ -712,35 +774,41 @@ template<class rtype,class vtype,int N> accelerator_inline void accumReconZm (iM
 ////////
 template<class rtype,class vtype> accelerator_inline void spProjTp (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
   spProjTp(hspin._internal,fspin._internal);
 }
 template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spProjTp (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
   for(int i=0;i<N;i++) {
     spProjTp(hspin._internal[i],fspin._internal[i]);
   }
 }
 template<class rtype,class vtype,int N> accelerator_inline void spProjTp (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
   for(int i=0;i<N;i++){ 
     for(int j=0;j<N;j++){
       spProjTp(hspin._internal[i][j],fspin._internal[i][j]);
-  }}
+    }}
 }
 
 
 template<class rtype,class vtype> accelerator_inline void spReconTp (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
   spReconTp(hspin._internal,fspin._internal);
 }
 template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spReconTp (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
   for(int i=0;i<N;i++) {
     spReconTp(hspin._internal[i],fspin._internal[i]);
   }
 }
 template<class rtype,class vtype,int N> accelerator_inline void spReconTp (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
   for(int i=0;i<N;i++){ 
     for(int j=0;j<N;j++){
       spReconTp(hspin._internal[i][j],fspin._internal[i][j]);
@@ -749,37 +817,44 @@ template<class rtype,class vtype,int N> accelerator_inline void spReconTp (iMatr
 
 template<class rtype,class vtype> accelerator_inline void accumReconTp (iScalar<rtype> &hspin, iScalar<vtype> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
   accumReconTp(hspin._internal,fspin._internal);
 }
 template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void accumReconTp (iVector<rtype,N> &hspin, const iVector<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
   for(int i=0;i<N;i++) {
     accumReconTp(hspin._internal[i],fspin._internal[i]);
   }
 }
 template<class rtype,class vtype,int N> accelerator_inline void accumReconTp (iMatrix<rtype,N> &hspin, const iMatrix<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
   for(int i=0;i<N;i++){ 
     for(int j=0;j<N;j++){
       accumReconTp(hspin._internal[i][j],fspin._internal[i][j]);
     }}
 }
 
+
 ////////
 // Tm
 ////////
 template<class rtype,class vtype> accelerator_inline void spProjTm (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
   spProjTm(hspin._internal,fspin._internal);
 }
 template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spProjTm (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
   for(int i=0;i<N;i++) {
     spProjTm(hspin._internal[i],fspin._internal[i]);
   }
 }
 template<class rtype,class vtype,int N> accelerator_inline void spProjTm (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
   for(int i=0;i<N;i++){ 
     for(int j=0;j<N;j++){
       spProjTm(hspin._internal[i][j],fspin._internal[i][j]);
@@ -789,16 +864,19 @@ template<class rtype,class vtype,int N> accelerator_inline void spProjTm (iMatri
 
 template<class rtype,class vtype> accelerator_inline void spReconTm (iScalar<rtype> &hspin, const iScalar<vtype> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
   spReconTm(hspin._internal,fspin._internal);
 }
 template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spReconTm (iVector<rtype,N> &hspin, const iVector<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
   for(int i=0;i<N;i++) {
     spReconTm(hspin._internal[i],fspin._internal[i]);
   }
 }
 template<class rtype,class vtype,int N> accelerator_inline void spReconTm (iMatrix<rtype,N> &hspin, const iMatrix<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
   for(int i=0;i<N;i++){ 
     for(int j=0;j<N;j++){
       spReconTm(hspin._internal[i][j],fspin._internal[i][j]);
@@ -807,37 +885,44 @@ template<class rtype,class vtype,int N> accelerator_inline void spReconTm (iMatr
 
 template<class rtype,class vtype> accelerator_inline void accumReconTm (iScalar<rtype> &hspin, const iScalar<vtype> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
   accumReconTm(hspin._internal,fspin._internal);
 }
 template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void accumReconTm (iVector<rtype,N> &hspin, const iVector<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
   for(int i=0;i<N;i++) {
     accumReconTm(hspin._internal[i],fspin._internal[i]);
   }
 }
 template<class rtype,class vtype,int N> accelerator_inline void accumReconTm (iMatrix<rtype,N> &hspin, const iMatrix<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
   for(int i=0;i<N;i++){ 
     for(int j=0;j<N;j++){
       accumReconTm(hspin._internal[i][j],fspin._internal[i][j]);
     }}
 }
 
+
 ////////
 // 5p
 ////////
-template<class rtype,class vtype,IfNotCoarsened<iScalar<vtype> > = 0> accelerator_inline void spProj5p (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
+template<class rtype,class vtype> accelerator_inline void spProj5p (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
   spProj5p(hspin._internal,fspin._internal);
 }
 template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spProj5p (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
   for(int i=0;i<N;i++) {
     spProj5p(hspin._internal[i],fspin._internal[i]);
   }
 }
-template<class rtype,class vtype,int N,IfNotCoarsened<iScalar<vtype> > = 0> accelerator_inline void spProj5p (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
+template<class rtype,class vtype,int N> accelerator_inline void spProj5p (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
   for(int i=0;i<N;i++){ 
     for(int j=0;j<N;j++){
       spProj5p(hspin._internal[i][j],fspin._internal[i][j]);
@@ -846,16 +931,19 @@ template<class rtype,class vtype,int N,IfNotCoarsened<iScalar<vtype> > = 0> acce
 
 template<class rtype,class vtype> accelerator_inline void spRecon5p (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
   spRecon5p(hspin._internal,fspin._internal);
 }
 template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spRecon5p (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
   for(int i=0;i<N;i++) {
     spRecon5p(hspin._internal[i],fspin._internal[i]);
   }
 }
 template<class rtype,class vtype,int N> accelerator_inline void spRecon5p (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
   for(int i=0;i<N;i++){ 
     for(int j=0;j<N;j++){
       spRecon5p(hspin._internal[i][j],fspin._internal[i][j]);
@@ -864,16 +952,19 @@ template<class rtype,class vtype,int N> accelerator_inline void spRecon5p (iMatr
 
 template<class rtype,class vtype> accelerator_inline void accumRecon5p (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
   accumRecon5p(hspin._internal,fspin._internal);
 }
 template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void accumRecon5p (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
   for(int i=0;i<N;i++) {
     accumRecon5p(hspin._internal[i],fspin._internal[i]);
   }
 }
 template<class rtype,class vtype,int N> accelerator_inline void accumRecon5p (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
   for(int i=0;i<N;i++){ 
     for(int j=0;j<N;j++){
       accumRecon5p(hspin._internal[i][j],fspin._internal[i][j]);
@@ -881,18 +972,24 @@ template<class rtype,class vtype,int N> accelerator_inline void accumRecon5p (iM
 }
 
 // four spinor projectors for chiral proj
-template<class vtype,IfNotCoarsened<iScalar<vtype> > = 0> accelerator_inline void spProj5p (iScalar<vtype> &hspin,const iScalar<vtype> &fspin)
+//  template<class vtype> accelerator_inline void fspProj5p (iScalar<vtype> &hspin,const iScalar<vtype> &fspin)
+template<class vtype> accelerator_inline void spProj5p (iScalar<vtype> &hspin,const iScalar<vtype> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
   spProj5p(hspin._internal,fspin._internal);
 }
-template<class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0,IfNotCoarsened<iScalar<vtype> > = 0> accelerator_inline void spProj5p (iVector<vtype,N> &hspin,const iVector<vtype,N> &fspin)
+//  template<class vtype,int N> accelerator_inline void fspProj5p (iVector<vtype,N> &hspin,iVector<vtype,N> &fspin)
+template<class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spProj5p (iVector<vtype,N> &hspin,const iVector<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
   for(int i=0;i<N;i++) {
     spProj5p(hspin._internal[i],fspin._internal[i]);
   }
 }
-template<class vtype,int N,IfNotCoarsened<iScalar<vtype> > = 0> accelerator_inline void spProj5p (iMatrix<vtype,N> &hspin,const iMatrix<vtype,N> &fspin)
+//  template<class vtype,int N> accelerator_inline void fspProj5p (iMatrix<vtype,N> &hspin,iMatrix<vtype,N> &fspin)
+template<class vtype,int N> accelerator_inline void spProj5p (iMatrix<vtype,N> &hspin,const iMatrix<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
   for(int i=0;i<N;i++){ 
     for(int j=0;j<N;j++){
       spProj5p(hspin._internal[i][j],fspin._internal[i][j]);
@@ -904,17 +1001,17 @@ template<class vtype,int N,IfNotCoarsened<iScalar<vtype> > = 0> accelerator_inli
 // 5m
 ////////
 
-template<class rtype,class vtype,IfNotCoarsened<iScalar<vtype> > = 0> accelerator_inline void spProj5m (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
+template<class rtype,class vtype> accelerator_inline void spProj5m (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
 {
   spProj5m(hspin._internal,fspin._internal);
 }
-template<class rtype,class vtype,int N,IfNotSpinor<iVector<rtype,N> > = 0,IfNotCoarsened<iScalar<vtype> > = 0> accelerator_inline void spProj5m (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
+template<class rtype,class vtype,int N,IfNotSpinor<iVector<rtype,N> > = 0> accelerator_inline void spProj5m (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
 {
   for(int i=0;i<N;i++) {
     spProj5m(hspin._internal[i],fspin._internal[i]);
   }
 }
-template<class rtype,class vtype,int N,IfNotCoarsened<iScalar<vtype> > = 0> accelerator_inline void spProj5m (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
+template<class rtype,class vtype,int N> accelerator_inline void spProj5m (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
 {
   for(int i=0;i<N;i++){ 
     for(int j=0;j<N;j++){
@@ -924,34 +1021,40 @@ template<class rtype,class vtype,int N,IfNotCoarsened<iScalar<vtype> > = 0> acce
 
 template<class rtype,class vtype> accelerator_inline void spRecon5m (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
   spRecon5m(hspin._internal,fspin._internal);
 }
 template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spRecon5m (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
   for(int i=0;i<N;i++) {
     spRecon5m(hspin._internal[i],fspin._internal[i]);
   }
 }
 template<class rtype,class vtype,int N> accelerator_inline void spRecon5m (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
   for(int i=0;i<N;i++){ 
     for(int j=0;j<N;j++){
       spRecon5m(hspin._internal[i][j],fspin._internal[i][j]);
-  }}
+    }}
 }
 
 template<class rtype,class vtype> accelerator_inline void accumRecon5m (iScalar<rtype> &hspin,const iScalar<vtype> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
   accumRecon5m(hspin._internal,fspin._internal);
 }
 template<class rtype,class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void accumRecon5m (iVector<rtype,N> &hspin,const iVector<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
   for(int i=0;i<N;i++) {
     accumRecon5m(hspin._internal[i],fspin._internal[i]);
   }
 }
 template<class rtype,class vtype,int N> accelerator_inline void accumRecon5m (iMatrix<rtype,N> &hspin,const iMatrix<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
   for(int i=0;i<N;i++){ 
     for(int j=0;j<N;j++){
       accumRecon5m(hspin._internal[i][j],fspin._internal[i][j]);
@@ -960,18 +1063,24 @@ template<class rtype,class vtype,int N> accelerator_inline void accumRecon5m (iM
 
 
 // four spinor projectors for chiral proj
-template<class vtype,IfNotCoarsened<iScalar<vtype> > = 0> accelerator_inline void spProj5m (iScalar<vtype> &hspin,const iScalar<vtype> &fspin)
+//  template<class vtype> accelerator_inline void fspProj5m (iScalar<vtype> &hspin,const iScalar<vtype> &fspin)
+template<class vtype> accelerator_inline void spProj5m (iScalar<vtype> &hspin,const iScalar<vtype> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,SpinorIndex>::notvalue,iScalar<vtype> >::type *temp;
   spProj5m(hspin._internal,fspin._internal);
 }
-template<class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0,IfNotCoarsened<iScalar<vtype> > = 0> accelerator_inline void spProj5m (iVector<vtype,N> &hspin,const iVector<vtype,N> &fspin)
+//  template<class vtype,int N> accelerator_inline void fspProj5m (iVector<vtype,N> &hspin,iVector<vtype,N> &fspin)
+template<class vtype,int N,IfNotSpinor<iVector<vtype,N> > = 0> accelerator_inline void spProj5m (iVector<vtype,N> &hspin,const iVector<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iVector<vtype,N>,SpinorIndex>::notvalue,iVector<vtype,N> >::type *temp;
   for(int i=0;i<N;i++) {
     spProj5m(hspin._internal[i],fspin._internal[i]);
   }
 }
-template<class vtype,int N,IfNotCoarsened<iScalar<vtype> > = 0> accelerator_inline void spProj5m (iMatrix<vtype,N> &hspin,const iMatrix<vtype,N> &fspin)
+//  template<class vtype,int N> accelerator_inline void fspProj5m (iMatrix<vtype,N> &hspin,iMatrix<vtype,N> &fspin)
+template<class vtype,int N> accelerator_inline void spProj5m (iMatrix<vtype,N> &hspin,const iMatrix<vtype,N> &fspin)
 {
+  //typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,SpinorIndex>::notvalue,iMatrix<vtype,N> >::type *temp;
   for(int i=0;i<N;i++){ 
     for(int j=0;j<N;j++){
       spProj5m(hspin._internal[i][j],fspin._internal[i][j]);

Grid/qcd/utils/CovariantCshift.h

@@ -53,21 +53,23 @@ namespace PeriodicBC {
     return Cshift(tmp,mu,-1);// moves towards positive mu
   }
 
-  template<class gauge,class Expr,typename std::enable_if<is_lattice_expr<Expr>::value,void>::type * = nullptr>
-    auto  CovShiftForward(const Lattice<gauge> &Link, 
-			  int mu,
-			  const Expr &expr) -> decltype(closure(expr))
+  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)))> 
   {
-    auto arg = closure(expr);
+    Lattice<decltype(expr.op.func(eval(0, expr.arg1)))> arg(expr);
     return CovShiftForward(Link,mu,arg);
   }
-  template<class gauge,class Expr,typename std::enable_if<is_lattice_expr<Expr>::value,void>::type * = nullptr>
-    auto  CovShiftBackward(const Lattice<gauge> &Link, 
-			   int mu,
-			   const Expr &expr) -> decltype(closure(expr))
+  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)))> 
   {
-    auto arg = closure(expr);
-    return CovShiftBackward(Link,mu,arg);
+    Lattice<decltype(expr.op.func(eval(0, expr.arg1)))> arg(expr);
+    return CovShiftForward(Link,mu,arg);
   }
 
 }
@@ -140,23 +142,26 @@ namespace ConjugateBC {
     return Cshift(tmp,mu,-1);// moves towards positive mu
   }
 
-  template<class gauge,class Expr,typename std::enable_if<is_lattice_expr<Expr>::value,void>::type * = nullptr>
-    auto  CovShiftForward(const Lattice<gauge> &Link, 
-			  int mu,
-			  const Expr &expr) -> decltype(closure(expr))
+  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)))> 
   {
-    auto arg = closure(expr);
+    Lattice<decltype(expr.op.func(eval(0, expr.arg1)))> arg(expr);
     return CovShiftForward(Link,mu,arg);
   }
-  template<class gauge,class Expr,typename std::enable_if<is_lattice_expr<Expr>::value,void>::type * = nullptr>
-    auto  CovShiftBackward(const Lattice<gauge> &Link, 
-			   int mu,
-			   const Expr &expr)  -> decltype(closure(expr))
+  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)))> 
   {
-    auto arg = closure(expr);
-    return CovShiftBackward(Link,mu,arg);
+    Lattice<decltype(expr.op.func(eval(0, expr.arg1)))> arg(expr);
+    return CovShiftForward(Link,mu,arg);
   }
 
+
 }
 
 

Grid/qcd/utils/LinalgUtils.h

@@ -154,8 +154,8 @@ void axpby_ssp_pminus(Lattice<vobj> &z,Coeff a,const Lattice<vobj> &x,Coeff b,co
   accelerator_for(sss,nloop,vobj::Nsimd(),{
     uint64_t ss = sss*Ls;
     decltype(coalescedRead(y_v[ss+sp])) tmp;
-    spProj5m(tmp,y_v(ss+sp)); 
-   tmp = a*x_v(ss+s)+b*tmp;
+    spProj5m(tmp,y_v(ss+sp));
+    tmp = a*x_v(ss+s)+b*tmp;
     coalescedWrite(z_v[ss+s],tmp);
   });
 }
@@ -188,6 +188,7 @@ void G5R5(Lattice<vobj> &z,const Lattice<vobj> &x)
   z.Checkerboard() = x.Checkerboard();
   conformable(x,z);
   int Ls = grid->_rdimensions[0];
+  Gamma G5(Gamma::Algebra::Gamma5);
   autoView( x_v, x, AcceleratorRead);
   autoView( z_v, z, AcceleratorWrite);
   uint64_t nloop = grid->oSites()/Ls;
@@ -195,13 +196,7 @@ void G5R5(Lattice<vobj> &z,const Lattice<vobj> &x)
     uint64_t ss = sss*Ls;
     for(int s=0;s<Ls;s++){
       int sp = Ls-1-s;
-      auto tmp = x_v(ss+s);
-      decltype(tmp) tmp_p;
-      decltype(tmp) tmp_m;
-      spProj5p(tmp_p,tmp);
-      spProj5m(tmp_m,tmp);
-      // Use of spProj5m, 5p captures the coarse space too
-      coalescedWrite(z_v[ss+sp],tmp_p - tmp_m);
+      coalescedWrite(z_v[ss+sp],G5*x_v(ss+s));
     }
   });
 }
@@ -213,20 +208,10 @@ void G5C(Lattice<vobj> &z, const Lattice<vobj> &x)
   z.Checkerboard() = x.Checkerboard();
   conformable(x, z);
 
-  autoView( x_v, x, AcceleratorRead);
-  autoView( z_v, z, AcceleratorWrite);
-  uint64_t nloop = grid->oSites();
-  accelerator_for(ss,nloop,vobj::Nsimd(),{
-    auto tmp = x_v(ss);
-    decltype(tmp) tmp_p;
-    decltype(tmp) tmp_m;
-    spProj5p(tmp_p,tmp);
-    spProj5m(tmp_m,tmp);
-    coalescedWrite(z_v[ss],tmp_p - tmp_m);
-  });
+  Gamma G5(Gamma::Algebra::Gamma5);
+  z = G5 * x;
 }
 
-/*
 template<class CComplex, int nbasis>
 void G5C(Lattice<iVector<CComplex, nbasis>> &z, const Lattice<iVector<CComplex, nbasis>> &x)
 {
@@ -249,7 +234,6 @@ void G5C(Lattice<iVector<CComplex, nbasis>> &z, const Lattice<iVector<CComplex,
     }
   });
 }
-*/
 
 NAMESPACE_END(Grid);
 

Grid/qcd/utils/SUnAdjoint.h

@@ -39,7 +39,7 @@ public:
   typedef iSUnAdjointMatrix<ComplexF> AMatrixF;
   typedef iSUnAdjointMatrix<ComplexD> AMatrixD;
 
-  typedef iSUnAdjointMatrix<vComplex>  vAMatrix;
+  typedef iSUnAdjointMatrix<vComplex> vAMatrix;
   typedef iSUnAdjointMatrix<vComplexF> vAMatrixF;
   typedef iSUnAdjointMatrix<vComplexD> vAMatrixD;
 
@@ -47,9 +47,14 @@ public:
   typedef Lattice<vAMatrixF> LatticeAdjMatrixF;
   typedef Lattice<vAMatrixD> LatticeAdjMatrixD;
 
-  typedef Lattice<iVector<iScalar<iMatrix<vComplex, Dimension> >, Nd> >  LatticeAdjField;
-  typedef Lattice<iVector<iScalar<iMatrix<vComplexF, Dimension> >, Nd> > LatticeAdjFieldF;
-  typedef Lattice<iVector<iScalar<iMatrix<vComplexD, Dimension> >, Nd> > LatticeAdjFieldD;
+  typedef Lattice<iVector<iScalar<iMatrix<vComplex, Dimension> >, Nd> >
+  LatticeAdjField;
+  typedef Lattice<iVector<iScalar<iMatrix<vComplexF, Dimension> >, Nd> >
+  LatticeAdjFieldF;
+  typedef Lattice<iVector<iScalar<iMatrix<vComplexD, Dimension> >, Nd> >
+  LatticeAdjFieldD;
+
+
 
 
   template <class cplx>
@@ -123,9 +128,7 @@ public:
   }
 
   // Projects the algebra components a lattice matrix (of dimension ncol*ncol -1 )
-  static void projectOnAlgebra(typename SU<ncolour>::LatticeAlgebraVector &h_out, const LatticeAdjMatrix &in, Real scale = 1.0) 
-  {
-    
+  static void projectOnAlgebra(typename SU<ncolour>::LatticeAlgebraVector &h_out, const LatticeAdjMatrix &in, Real scale = 1.0) {
     conformable(h_out, in);
     h_out = Zero();
     AMatrix iTa;
@@ -133,7 +136,7 @@ public:
 
     for (int a = 0; a < Dimension; a++) {
       generator(a, iTa);
-      LatticeComplex tmp = real(trace(iTa * in)) * coefficient;
+      auto tmp = real(trace(iTa * in)) * coefficient;
       pokeColour(h_out, tmp, a);
     }
   }

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);

Grid/simd/Fujitsu_A64FX_asm_double.h

@@ -1,779 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: Fujitsu_A64FX_asm_double.h
-
-    Copyright (C) 2020
-
-Author: Nils Meyer <nils.meyer@ur.de>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#define LOAD_CHIMU(base)               LOAD_CHIMU_INTERLEAVED_A64FXd(base)  
-#define PREFETCH_CHIMU_L1(A)           PREFETCH_CHIMU_L1_INTERNAL_A64FXd(A)  
-#define PREFETCH_GAUGE_L1(A)           PREFETCH_GAUGE_L1_INTERNAL_A64FXd(A)  
-#define PREFETCH_CHIMU_L2(A)           PREFETCH_CHIMU_L2_INTERNAL_A64FXd(A)  
-#define PREFETCH_GAUGE_L2(A)           PREFETCH_GAUGE_L2_INTERNAL_A64FXd(A)  
-#define PF_GAUGE(A)  
-#define PREFETCH_RESULT_L2_STORE(A)    PREFETCH_RESULT_L2_STORE_INTERNAL_A64FXd(A)  
-#define PREFETCH_RESULT_L1_STORE(A)    PREFETCH_RESULT_L1_STORE_INTERNAL_A64FXd(A)  
-#define PREFETCH1_CHIMU(A)             PREFETCH_CHIMU_L1(A)  
-#define PREFETCH_CHIMU(A)              PREFETCH_CHIMU_L1(A)  
-#define LOCK_GAUGE(A)  
-#define UNLOCK_GAUGE(A)  
-#define MASK_REGS                      DECLARATIONS_A64FXd  
-#define SAVE_RESULT(A,B)               RESULT_A64FXd(A); PREFETCH_RESULT_L2_STORE(B)  
-#define MULT_2SPIN_1(Dir)              MULT_2SPIN_1_A64FXd(Dir)  
-#define MULT_2SPIN_2                   MULT_2SPIN_2_A64FXd  
-#define LOAD_CHI(base)                 LOAD_CHI_A64FXd(base)  
-#define ADD_RESULT(base,basep)         LOAD_CHIMU(base); ADD_RESULT_INTERNAL_A64FXd; RESULT_A64FXd(base)  
-#define XP_PROJ                        XP_PROJ_A64FXd  
-#define YP_PROJ                        YP_PROJ_A64FXd  
-#define ZP_PROJ                        ZP_PROJ_A64FXd  
-#define TP_PROJ                        TP_PROJ_A64FXd  
-#define XM_PROJ                        XM_PROJ_A64FXd  
-#define YM_PROJ                        YM_PROJ_A64FXd  
-#define ZM_PROJ                        ZM_PROJ_A64FXd  
-#define TM_PROJ                        TM_PROJ_A64FXd  
-#define XP_RECON                       XP_RECON_A64FXd  
-#define XM_RECON                       XM_RECON_A64FXd  
-#define XM_RECON_ACCUM                 XM_RECON_ACCUM_A64FXd  
-#define YM_RECON_ACCUM                 YM_RECON_ACCUM_A64FXd  
-#define ZM_RECON_ACCUM                 ZM_RECON_ACCUM_A64FXd  
-#define TM_RECON_ACCUM                 TM_RECON_ACCUM_A64FXd  
-#define XP_RECON_ACCUM                 XP_RECON_ACCUM_A64FXd  
-#define YP_RECON_ACCUM                 YP_RECON_ACCUM_A64FXd  
-#define ZP_RECON_ACCUM                 ZP_RECON_ACCUM_A64FXd  
-#define TP_RECON_ACCUM                 TP_RECON_ACCUM_A64FXd  
-#define PERMUTE_DIR0                   0  
-#define PERMUTE_DIR1                   1  
-#define PERMUTE_DIR2                   2  
-#define PERMUTE_DIR3                   3  
-#define PERMUTE                        PERMUTE_A64FXd;  
-#define LOAD_TABLE(Dir)                if (Dir == 0) { LOAD_TABLE0; } else if (Dir == 1) { LOAD_TABLE1; } else if (Dir == 2) { LOAD_TABLE2; }  
-#define MAYBEPERM(Dir,perm)            if (Dir != 3) { if (perm) { PERMUTE; } }  
-// DECLARATIONS
-#define DECLARATIONS_A64FXd  \
-    const uint64_t lut[4][8] = { \
-        {4, 5, 6, 7, 0, 1, 2, 3}, \
-        {2, 3, 0, 1, 6, 7, 4, 5}, \
-        {1, 0, 3, 2, 5, 4, 7, 6}, \
-        {0, 1, 2, 4, 5, 6, 7, 8} };\
-asm ( \
-    "fmov z31.d , 0 \n\t" \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// RESULT
-#define RESULT_A64FXd(base)  \
-{ \
-asm ( \
-    "str z0, [%[storeptr], -6, mul vl] \n\t" \
-    "str z1, [%[storeptr], -5, mul vl] \n\t" \
-    "str z2, [%[storeptr], -4, mul vl] \n\t" \
-    "str z3, [%[storeptr], -3, mul vl] \n\t" \
-    "str z4, [%[storeptr], -2, mul vl] \n\t" \
-    "str z5, [%[storeptr], -1, mul vl] \n\t" \
-    "str z6, [%[storeptr], 0, mul vl] \n\t" \
-    "str z7, [%[storeptr], 1, mul vl] \n\t" \
-    "str z8, [%[storeptr], 2, mul vl] \n\t" \
-    "str z9, [%[storeptr], 3, mul vl] \n\t" \
-    "str z10, [%[storeptr], 4, mul vl] \n\t" \
-    "str z11, [%[storeptr], 5, mul vl] \n\t" \
-    :  \
-    : [storeptr] "r" (base + 2 * 3 * 64) \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31","memory" \
-); \
-}
-// PREFETCH_CHIMU_L2 (prefetch to L2)
-#define PREFETCH_CHIMU_L2_INTERNAL_A64FXd(base)  \
-{ \
-asm ( \
-    "prfd PLDL2STRM, p5, [%[fetchptr], 0, mul vl] \n\t" \
-    "prfd PLDL2STRM, p5, [%[fetchptr], 4, mul vl] \n\t" \
-    "prfd PLDL2STRM, p5, [%[fetchptr], 8, mul vl] \n\t" \
-    :  \
-    : [fetchptr] "r" (base) \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31","memory" \
-); \
-}
-// PREFETCH_CHIMU_L1 (prefetch to L1)
-#define PREFETCH_CHIMU_L1_INTERNAL_A64FXd(base)  \
-{ \
-asm ( \
-    "prfd PLDL1STRM, p5, [%[fetchptr], 0, mul vl] \n\t" \
-    "prfd PLDL1STRM, p5, [%[fetchptr], 4, mul vl] \n\t" \
-    "prfd PLDL1STRM, p5, [%[fetchptr], 8, mul vl] \n\t" \
-    :  \
-    : [fetchptr] "r" (base) \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31","memory" \
-); \
-}
-// PREFETCH_GAUGE_L2 (prefetch to L2)
-#define PREFETCH_GAUGE_L2_INTERNAL_A64FXd(A)  \
-{ \
-    const auto & ref(U[sUn](A)); uint64_t baseU = (uint64_t)&ref + 3 * 3 * 64; \
-asm ( \
-    "prfd PLDL2STRM, p5, [%[fetchptr], -4, mul vl] \n\t" \
-    "prfd PLDL2STRM, p5, [%[fetchptr], 0, mul vl] \n\t" \
-    "prfd PLDL2STRM, p5, [%[fetchptr], 4, mul vl] \n\t" \
-    "prfd PLDL2STRM, p5, [%[fetchptr], 8, mul vl] \n\t" \
-    "prfd PLDL2STRM, p5, [%[fetchptr], 12, mul vl] \n\t" \
-    "prfd PLDL2STRM, p5, [%[fetchptr], 16, mul vl] \n\t" \
-    "prfd PLDL2STRM, p5, [%[fetchptr], 20, mul vl] \n\t" \
-    "prfd PLDL2STRM, p5, [%[fetchptr], 24, mul vl] \n\t" \
-    "prfd PLDL2STRM, p5, [%[fetchptr], 28, mul vl] \n\t" \
-    :  \
-    : [fetchptr] "r" (baseU) \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31","memory" \
-); \
-}
-// PREFETCH_GAUGE_L1 (prefetch to L1)
-#define PREFETCH_GAUGE_L1_INTERNAL_A64FXd(A)  \
-{ \
-    const auto & ref(U[sU](A)); uint64_t baseU = (uint64_t)&ref; \
-asm ( \
-    "prfd PLDL1STRM, p5, [%[fetchptr], 0, mul vl] \n\t" \
-    "prfd PLDL1STRM, p5, [%[fetchptr], 4, mul vl] \n\t" \
-    "prfd PLDL1STRM, p5, [%[fetchptr], 8, mul vl] \n\t" \
-    :  \
-    : [fetchptr] "r" (baseU) \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31","memory" \
-); \
-}
-// LOAD_CHI
-#define LOAD_CHI_A64FXd(base)  \
-{ \
-asm ( \
-    "ldr z12, [%[fetchptr], 0, mul vl] \n\t" \
-    "ldr z13, [%[fetchptr], 1, mul vl] \n\t" \
-    "ldr z14, [%[fetchptr], 2, mul vl] \n\t" \
-    "ldr z15, [%[fetchptr], 3, mul vl] \n\t" \
-    "ldr z16, [%[fetchptr], 4, mul vl] \n\t" \
-    "ldr z17, [%[fetchptr], 5, mul vl] \n\t" \
-    :  \
-    : [fetchptr] "r" (base) \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31","memory" \
-); \
-}
-// LOAD_CHIMU
-#define LOAD_CHIMU_INTERLEAVED_A64FXd(base)  \
-{ \
-asm ( \
-    "ptrue p5.d \n\t" \
-    "ldr z12, [%[fetchptr], -6, mul vl] \n\t" \
-    "ldr z21, [%[fetchptr], 3, mul vl] \n\t" \
-    "ldr z15, [%[fetchptr], -3, mul vl] \n\t" \
-    "ldr z18, [%[fetchptr], 0, mul vl] \n\t" \
-    "ldr z13, [%[fetchptr], -5, mul vl] \n\t" \
-    "ldr z22, [%[fetchptr], 4, mul vl] \n\t" \
-    "ldr z16, [%[fetchptr], -2, mul vl] \n\t" \
-    "ldr z19, [%[fetchptr], 1, mul vl] \n\t" \
-    "ldr z14, [%[fetchptr], -4, mul vl] \n\t" \
-    "ldr z23, [%[fetchptr], 5, mul vl] \n\t" \
-    "ldr z17, [%[fetchptr], -1, mul vl] \n\t" \
-    "ldr z20, [%[fetchptr], 2, mul vl] \n\t" \
-    :  \
-    : [fetchptr] "r" (base + 2 * 3 * 64) \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31","memory" \
-); \
-}
-// LOAD_CHIMU_0213
-#define LOAD_CHIMU_0213_A64FXd  \
-{ \
-    const SiteSpinor & ref(in[offset]); \
-asm ( \
-    "ptrue p5.d \n\t" \
-    "ldr z12, [%[fetchptr], -6, mul vl] \n\t" \
-    "ldr z18, [%[fetchptr], 0, mul vl] \n\t" \
-    "ldr z13, [%[fetchptr], -5, mul vl] \n\t" \
-    "ldr z19, [%[fetchptr], 1, mul vl] \n\t" \
-    "ldr z14, [%[fetchptr], -4, mul vl] \n\t" \
-    "ldr z20, [%[fetchptr], 2, mul vl] \n\t" \
-    "ldr z15, [%[fetchptr], -3, mul vl] \n\t" \
-    "ldr z21, [%[fetchptr], 3, mul vl] \n\t" \
-    "ldr z16, [%[fetchptr], -2, mul vl] \n\t" \
-    "ldr z22, [%[fetchptr], 4, mul vl] \n\t" \
-    "ldr z17, [%[fetchptr], -1, mul vl] \n\t" \
-    "ldr z23, [%[fetchptr], 5, mul vl] \n\t" \
-    :  \
-    : [fetchptr] "r" (&ref[2][0]) \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31","memory" \
-); \
-}
-// LOAD_CHIMU_0312
-#define LOAD_CHIMU_0312_A64FXd  \
-{ \
-    const SiteSpinor & ref(in[offset]); \
-asm ( \
-    "ptrue p5.d \n\t" \
-    "ldr z12, [%[fetchptr], -6, mul vl] \n\t" \
-    "ldr z21, [%[fetchptr], 3, mul vl] \n\t" \
-    "ldr z13, [%[fetchptr], -5, mul vl] \n\t" \
-    "ldr z22, [%[fetchptr], 4, mul vl] \n\t" \
-    "ldr z14, [%[fetchptr], -4, mul vl] \n\t" \
-    "ldr z23, [%[fetchptr], 5, mul vl] \n\t" \
-    "ldr z15, [%[fetchptr], -3, mul vl] \n\t" \
-    "ldr z18, [%[fetchptr], 0, mul vl] \n\t" \
-    "ldr z16, [%[fetchptr], -2, mul vl] \n\t" \
-    "ldr z19, [%[fetchptr], 1, mul vl] \n\t" \
-    "ldr z17, [%[fetchptr], -1, mul vl] \n\t" \
-    "ldr z20, [%[fetchptr], 2, mul vl] \n\t" \
-    :  \
-    : [fetchptr] "r" (&ref[2][0]) \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31","memory" \
-); \
-}
-// LOAD_TABLE0
-#define LOAD_TABLE0  \
-asm ( \
-    "ldr z30, [%[tableptr], %[index], mul vl] \n\t" \
-    :  \
-    : [tableptr] "r" (&lut[0]),[index] "i" (0) \
-    : "memory","cc","p5","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// LOAD_TABLE1
-#define LOAD_TABLE1  \
-asm ( \
-    "ldr z30, [%[tableptr], %[index], mul vl] \n\t" \
-    :  \
-    : [tableptr] "r" (&lut[0]),[index] "i" (1) \
-    : "memory","cc","p5","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// LOAD_TABLE2
-#define LOAD_TABLE2  \
-asm ( \
-    "ldr z30, [%[tableptr], %[index], mul vl] \n\t" \
-    :  \
-    : [tableptr] "r" (&lut[0]),[index] "i" (2) \
-    : "memory","cc","p5","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// LOAD_TABLE3
-#define LOAD_TABLE3  \
-asm ( \
-    "ldr z30, [%[tableptr], %[index], mul vl] \n\t" \
-    :  \
-    : [tableptr] "r" (&lut[0]),[index] "i" (3) \
-    : "memory","cc","p5","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// PERMUTE
-#define PERMUTE_A64FXd  \
-asm ( \
-    "tbl z12.d, { z12.d }, z30.d \n\t"  \
-    "tbl z13.d, { z13.d }, z30.d \n\t"  \
-    "tbl z14.d, { z14.d }, z30.d \n\t"  \
-    "tbl z15.d, { z15.d }, z30.d \n\t"  \
-    "tbl z16.d, { z16.d }, z30.d \n\t"  \
-    "tbl z17.d, { z17.d }, z30.d \n\t"  \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// LOAD_GAUGE
-#define LOAD_GAUGE  \
-    const auto & ref(U[sU](A)); uint64_t baseU = (uint64_t)&ref; \
-{ \
-asm ( \
-    "ptrue p5.d \n\t" \
-    "ldr z24, [%[fetchptr], -6, mul vl] \n\t" \
-    "ldr z25, [%[fetchptr], -3, mul vl] \n\t" \
-    "ldr z26, [%[fetchptr], 0, mul vl] \n\t" \
-    "ldr z27, [%[fetchptr], -5, mul vl] \n\t" \
-    "ldr z28, [%[fetchptr], -2, mul vl] \n\t" \
-    "ldr z29, [%[fetchptr], 1, mul vl] \n\t" \
-    :  \
-    : [fetchptr] "r" (baseU + 2 * 3 * 64) \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31","memory" \
-); \
-}
-// MULT_2SPIN
-#define MULT_2SPIN_1_A64FXd(A)  \
-{ \
-    const auto & ref(U[sU](A)); uint64_t baseU = (uint64_t)&ref; \
-asm ( \
-    "ldr z24, [%[fetchptr], -6, mul vl] \n\t" \
-    "ldr z25, [%[fetchptr], -3, mul vl] \n\t" \
-    "ldr z26, [%[fetchptr], 0, mul vl] \n\t" \
-    "ldr z27, [%[fetchptr], -5, mul vl] \n\t" \
-    "ldr z28, [%[fetchptr], -2, mul vl] \n\t" \
-    "ldr z29, [%[fetchptr], 1, mul vl] \n\t" \
-    "movprfx z18.d, p5/m, z31.d \n\t" \
-    "fcmla z18.d, p5/m, z24.d, z12.d, 0 \n\t" \
-    "movprfx z21.d, p5/m, z31.d \n\t" \
-    "fcmla z21.d, p5/m, z24.d, z15.d, 0 \n\t" \
-    "movprfx z19.d, p5/m, z31.d \n\t" \
-    "fcmla z19.d, p5/m, z25.d, z12.d, 0 \n\t" \
-    "movprfx z22.d, p5/m, z31.d \n\t" \
-    "fcmla z22.d, p5/m, z25.d, z15.d, 0 \n\t" \
-    "movprfx z20.d, p5/m, z31.d \n\t" \
-    "fcmla z20.d, p5/m, z26.d, z12.d, 0 \n\t" \
-    "movprfx z23.d, p5/m, z31.d \n\t" \
-    "fcmla z23.d, p5/m, z26.d, z15.d, 0 \n\t" \
-    "fcmla z18.d, p5/m, z24.d, z12.d, 90 \n\t" \
-    "fcmla z21.d, p5/m, z24.d, z15.d, 90 \n\t" \
-    "fcmla z19.d, p5/m, z25.d, z12.d, 90 \n\t" \
-    "fcmla z22.d, p5/m, z25.d, z15.d, 90 \n\t" \
-    "fcmla z20.d, p5/m, z26.d, z12.d, 90 \n\t" \
-    "fcmla z23.d, p5/m, z26.d, z15.d, 90 \n\t" \
-    "ldr z24, [%[fetchptr], -4, mul vl] \n\t" \
-    "ldr z25, [%[fetchptr], -1, mul vl] \n\t" \
-    "ldr z26, [%[fetchptr], 2, mul vl] \n\t" \
-    :  \
-    : [fetchptr] "r" (baseU + 2 * 3 * 64) \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31","memory" \
-); \
-}
-// MULT_2SPIN_BACKEND
-#define MULT_2SPIN_2_A64FXd  \
-{ \
-asm ( \
-    "fcmla z18.d, p5/m, z27.d, z13.d, 0 \n\t" \
-    "fcmla z21.d, p5/m, z27.d, z16.d, 0 \n\t" \
-    "fcmla z19.d, p5/m, z28.d, z13.d, 0 \n\t" \
-    "fcmla z22.d, p5/m, z28.d, z16.d, 0 \n\t" \
-    "fcmla z20.d, p5/m, z29.d, z13.d, 0 \n\t" \
-    "fcmla z23.d, p5/m, z29.d, z16.d, 0 \n\t" \
-    "fcmla z18.d, p5/m, z27.d, z13.d, 90 \n\t" \
-    "fcmla z21.d, p5/m, z27.d, z16.d, 90 \n\t" \
-    "fcmla z19.d, p5/m, z28.d, z13.d, 90 \n\t" \
-    "fcmla z22.d, p5/m, z28.d, z16.d, 90 \n\t" \
-    "fcmla z20.d, p5/m, z29.d, z13.d, 90 \n\t" \
-    "fcmla z23.d, p5/m, z29.d, z16.d, 90 \n\t" \
-    "fcmla z18.d, p5/m, z24.d, z14.d, 0 \n\t" \
-    "fcmla z21.d, p5/m, z24.d, z17.d, 0 \n\t" \
-    "fcmla z19.d, p5/m, z25.d, z14.d, 0 \n\t" \
-    "fcmla z22.d, p5/m, z25.d, z17.d, 0 \n\t" \
-    "fcmla z20.d, p5/m, z26.d, z14.d, 0 \n\t" \
-    "fcmla z23.d, p5/m, z26.d, z17.d, 0 \n\t" \
-    "fcmla z18.d, p5/m, z24.d, z14.d, 90 \n\t" \
-    "fcmla z21.d, p5/m, z24.d, z17.d, 90 \n\t" \
-    "fcmla z19.d, p5/m, z25.d, z14.d, 90 \n\t" \
-    "fcmla z22.d, p5/m, z25.d, z17.d, 90 \n\t" \
-    "fcmla z20.d, p5/m, z26.d, z14.d, 90 \n\t" \
-    "fcmla z23.d, p5/m, z26.d, z17.d, 90 \n\t" \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); \
-}
-// XP_PROJ
-#define XP_PROJ_A64FXd  \
-{ \
-asm ( \
-    "fcadd z12.d, p5/m, z12.d, z21.d, 90 \n\t" \
-    "fcadd z13.d, p5/m, z13.d, z22.d, 90 \n\t" \
-    "fcadd z14.d, p5/m, z14.d, z23.d, 90 \n\t" \
-    "fcadd z15.d, p5/m, z15.d, z18.d, 90 \n\t" \
-    "fcadd z16.d, p5/m, z16.d, z19.d, 90 \n\t" \
-    "fcadd z17.d, p5/m, z17.d, z20.d, 90 \n\t" \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); \
-}
-// XP_RECON
-#define XP_RECON_A64FXd  \
-asm ( \
-    "movprfx z6.d, p5/m, z31.d \n\t" \
-    "fcadd z6.d, p5/m, z6.d, z21.d, 270 \n\t" \
-    "movprfx z7.d, p5/m, z31.d \n\t" \
-    "fcadd z7.d, p5/m, z7.d, z22.d, 270 \n\t" \
-    "movprfx z8.d, p5/m, z31.d \n\t" \
-    "fcadd z8.d, p5/m, z8.d, z23.d, 270 \n\t" \
-    "movprfx z9.d, p5/m, z31.d \n\t" \
-    "fcadd z9.d, p5/m, z9.d, z18.d, 270 \n\t" \
-    "movprfx z10.d, p5/m, z31.d \n\t" \
-    "fcadd z10.d, p5/m, z10.d, z19.d, 270 \n\t" \
-    "movprfx z11.d, p5/m, z31.d \n\t" \
-    "fcadd z11.d, p5/m, z11.d, z20.d, 270 \n\t" \
-    "mov z0.d, p5/m, z18.d \n\t" \
-    "mov z1.d, p5/m, z19.d \n\t" \
-    "mov z2.d, p5/m, z20.d \n\t" \
-    "mov z3.d, p5/m, z21.d \n\t" \
-    "mov z4.d, p5/m, z22.d \n\t" \
-    "mov z5.d, p5/m, z23.d \n\t" \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// XP_RECON_ACCUM
-#define XP_RECON_ACCUM_A64FXd  \
-asm ( \
-    "fcadd z9.d, p5/m, z9.d, z18.d, 270 \n\t" \
-    "fadd z0.d, p5/m, z0.d, z18.d \n\t"  \
-    "fcadd z10.d, p5/m, z10.d, z19.d, 270 \n\t" \
-    "fadd z1.d, p5/m, z1.d, z19.d \n\t"  \
-    "fcadd z11.d, p5/m, z11.d, z20.d, 270 \n\t" \
-    "fadd z2.d, p5/m, z2.d, z20.d \n\t"  \
-    "fcadd z6.d, p5/m, z6.d, z21.d, 270 \n\t" \
-    "fadd z3.d, p5/m, z3.d, z21.d \n\t"  \
-    "fcadd z7.d, p5/m, z7.d, z22.d, 270 \n\t" \
-    "fadd z4.d, p5/m, z4.d, z22.d \n\t"  \
-    "fcadd z8.d, p5/m, z8.d, z23.d, 270 \n\t" \
-    "fadd z5.d, p5/m, z5.d, z23.d \n\t"  \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// YP_PROJ
-#define YP_PROJ_A64FXd  \
-{ \
-asm ( \
-    "fsub z12.d, p5/m, z12.d, z21.d \n\t" \
-    "fsub z13.d, p5/m, z13.d, z22.d \n\t" \
-    "fsub z14.d, p5/m, z14.d, z23.d \n\t" \
-    "fadd z15.d, p5/m, z15.d, z18.d \n\t"  \
-    "fadd z16.d, p5/m, z16.d, z19.d \n\t"  \
-    "fadd z17.d, p5/m, z17.d, z20.d \n\t"  \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); \
-}
-// ZP_PROJ
-#define ZP_PROJ_A64FXd  \
-{ \
-asm ( \
-    "fcadd z12.d, p5/m, z12.d, z18.d, 90 \n\t" \
-    "fcadd z13.d, p5/m, z13.d, z19.d, 90 \n\t" \
-    "fcadd z14.d, p5/m, z14.d, z20.d, 90 \n\t" \
-    "fcadd z15.d, p5/m, z15.d, z21.d, 270 \n\t" \
-    "fcadd z16.d, p5/m, z16.d, z22.d, 270 \n\t" \
-    "fcadd z17.d, p5/m, z17.d, z23.d, 270 \n\t" \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); \
-}
-// TP_PROJ
-#define TP_PROJ_A64FXd  \
-{ \
-asm ( \
-    "fadd z12.d, p5/m, z12.d, z18.d \n\t"  \
-    "fadd z13.d, p5/m, z13.d, z19.d \n\t"  \
-    "fadd z14.d, p5/m, z14.d, z20.d \n\t"  \
-    "fadd z15.d, p5/m, z15.d, z21.d \n\t"  \
-    "fadd z16.d, p5/m, z16.d, z22.d \n\t"  \
-    "fadd z17.d, p5/m, z17.d, z23.d \n\t"  \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); \
-}
-// XM_PROJ
-#define XM_PROJ_A64FXd  \
-{ \
-asm ( \
-    "fcadd z12.d, p5/m, z12.d, z21.d, 270 \n\t" \
-    "fcadd z13.d, p5/m, z13.d, z22.d, 270 \n\t" \
-    "fcadd z14.d, p5/m, z14.d, z23.d, 270 \n\t" \
-    "fcadd z15.d, p5/m, z15.d, z18.d, 270 \n\t" \
-    "fcadd z16.d, p5/m, z16.d, z19.d, 270 \n\t" \
-    "fcadd z17.d, p5/m, z17.d, z20.d, 270 \n\t" \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); \
-}
-// XM_RECON
-#define XM_RECON_A64FXd  \
-asm ( \
-    "movprfx z6.d, p5/m, z31.d \n\t" \
-    "fcadd z6.d, p5/m, z6.d, z21.d, 90 \n\t" \
-    "movprfx z7.d, p5/m, z31.d \n\t" \
-    "fcadd z7.d, p5/m, z7.d, z22.d, 90 \n\t" \
-    "movprfx z8.d, p5/m, z31.d \n\t" \
-    "fcadd z8.d, p5/m, z8.d, z23.d, 90 \n\t" \
-    "movprfx z9.d, p5/m, z31.d \n\t" \
-    "fcadd z9.d, p5/m, z9.d, z18.d, 90 \n\t" \
-    "movprfx z10.d, p5/m, z31.d \n\t" \
-    "fcadd z10.d, p5/m, z10.d, z19.d, 90 \n\t" \
-    "movprfx z11.d, p5/m, z31.d \n\t" \
-    "fcadd z11.d, p5/m, z11.d, z20.d, 90 \n\t" \
-    "mov z0.d, p5/m, z18.d \n\t" \
-    "mov z1.d, p5/m, z19.d \n\t" \
-    "mov z2.d, p5/m, z20.d \n\t" \
-    "mov z3.d, p5/m, z21.d \n\t" \
-    "mov z4.d, p5/m, z22.d \n\t" \
-    "mov z5.d, p5/m, z23.d \n\t" \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// YM_PROJ
-#define YM_PROJ_A64FXd  \
-{ \
-asm ( \
-    "fadd z12.d, p5/m, z12.d, z21.d \n\t"  \
-    "fadd z13.d, p5/m, z13.d, z22.d \n\t"  \
-    "fadd z14.d, p5/m, z14.d, z23.d \n\t"  \
-    "fsub z15.d, p5/m, z15.d, z18.d \n\t" \
-    "fsub z16.d, p5/m, z16.d, z19.d \n\t" \
-    "fsub z17.d, p5/m, z17.d, z20.d \n\t" \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); \
-}
-// ZM_PROJ
-#define ZM_PROJ_A64FXd  \
-{ \
-asm ( \
-    "fcadd z12.d, p5/m, z12.d, z18.d, 270 \n\t" \
-    "fcadd z13.d, p5/m, z13.d, z19.d, 270 \n\t" \
-    "fcadd z14.d, p5/m, z14.d, z20.d, 270 \n\t" \
-    "fcadd z15.d, p5/m, z15.d, z21.d, 90 \n\t" \
-    "fcadd z16.d, p5/m, z16.d, z22.d, 90 \n\t" \
-    "fcadd z17.d, p5/m, z17.d, z23.d, 90 \n\t" \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); \
-}
-// TM_PROJ
-#define TM_PROJ_A64FXd  \
-{ \
-asm ( \
-    "ptrue p5.d \n\t" \
-    "fsub z12.d, p5/m, z12.d, z18.d \n\t" \
-    "fsub z13.d, p5/m, z13.d, z19.d \n\t" \
-    "fsub z14.d, p5/m, z14.d, z20.d \n\t" \
-    "fsub z15.d, p5/m, z15.d, z21.d \n\t" \
-    "fsub z16.d, p5/m, z16.d, z22.d \n\t" \
-    "fsub z17.d, p5/m, z17.d, z23.d \n\t" \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); \
-}
-// XM_RECON_ACCUM
-#define XM_RECON_ACCUM_A64FXd  \
-asm ( \
-    "fcadd z9.d, p5/m, z9.d, z18.d, 90 \n\t" \
-    "fcadd z10.d, p5/m, z10.d, z19.d, 90 \n\t" \
-    "fcadd z11.d, p5/m, z11.d, z20.d, 90 \n\t" \
-    "fcadd z6.d, p5/m, z6.d, z21.d, 90 \n\t" \
-    "fcadd z7.d, p5/m, z7.d, z22.d, 90 \n\t" \
-    "fcadd z8.d, p5/m, z8.d, z23.d, 90 \n\t" \
-    "fadd z0.d, p5/m, z0.d, z18.d \n\t"  \
-    "fadd z1.d, p5/m, z1.d, z19.d \n\t"  \
-    "fadd z2.d, p5/m, z2.d, z20.d \n\t"  \
-    "fadd z3.d, p5/m, z3.d, z21.d \n\t"  \
-    "fadd z4.d, p5/m, z4.d, z22.d \n\t"  \
-    "fadd z5.d, p5/m, z5.d, z23.d \n\t"  \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// YP_RECON_ACCUM
-#define YP_RECON_ACCUM_A64FXd  \
-asm ( \
-    "fadd z0.d, p5/m, z0.d, z18.d \n\t"  \
-    "fsub z9.d, p5/m, z9.d, z18.d \n\t" \
-    "fadd z1.d, p5/m, z1.d, z19.d \n\t"  \
-    "fsub z10.d, p5/m, z10.d, z19.d \n\t" \
-    "fadd z2.d, p5/m, z2.d, z20.d \n\t"  \
-    "fsub z11.d, p5/m, z11.d, z20.d \n\t" \
-    "fadd z3.d, p5/m, z3.d, z21.d \n\t"  \
-    "fadd z6.d, p5/m, z6.d, z21.d \n\t"  \
-    "fadd z4.d, p5/m, z4.d, z22.d \n\t"  \
-    "fadd z7.d, p5/m, z7.d, z22.d \n\t"  \
-    "fadd z5.d, p5/m, z5.d, z23.d \n\t"  \
-    "fadd z8.d, p5/m, z8.d, z23.d \n\t"  \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// YM_RECON_ACCUM
-#define YM_RECON_ACCUM_A64FXd  \
-asm ( \
-    "fadd z0.d, p5/m, z0.d, z18.d \n\t"  \
-    "fadd z9.d, p5/m, z9.d, z18.d \n\t"  \
-    "fadd z1.d, p5/m, z1.d, z19.d \n\t"  \
-    "fadd z10.d, p5/m, z10.d, z19.d \n\t"  \
-    "fadd z2.d, p5/m, z2.d, z20.d \n\t"  \
-    "fadd z11.d, p5/m, z11.d, z20.d \n\t"  \
-    "fadd z3.d, p5/m, z3.d, z21.d \n\t"  \
-    "fsub z6.d, p5/m, z6.d, z21.d \n\t" \
-    "fadd z4.d, p5/m, z4.d, z22.d \n\t"  \
-    "fsub z7.d, p5/m, z7.d, z22.d \n\t" \
-    "fadd z5.d, p5/m, z5.d, z23.d \n\t"  \
-    "fsub z8.d, p5/m, z8.d, z23.d \n\t" \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// ZP_RECON_ACCUM
-#define ZP_RECON_ACCUM_A64FXd  \
-asm ( \
-    "fcadd z6.d, p5/m, z6.d, z18.d, 270 \n\t" \
-    "fadd z0.d, p5/m, z0.d, z18.d \n\t"  \
-    "fcadd z7.d, p5/m, z7.d, z19.d, 270 \n\t" \
-    "fadd z1.d, p5/m, z1.d, z19.d \n\t"  \
-    "fcadd z8.d, p5/m, z8.d, z20.d, 270 \n\t" \
-    "fadd z2.d, p5/m, z2.d, z20.d \n\t"  \
-    "fcadd z9.d, p5/m, z9.d, z21.d, 90 \n\t" \
-    "fadd z3.d, p5/m, z3.d, z21.d \n\t"  \
-    "fcadd z10.d, p5/m, z10.d, z22.d, 90 \n\t" \
-    "fadd z4.d, p5/m, z4.d, z22.d \n\t"  \
-    "fcadd z11.d, p5/m, z11.d, z23.d, 90 \n\t" \
-    "fadd z5.d, p5/m, z5.d, z23.d \n\t"  \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// ZM_RECON_ACCUM
-#define ZM_RECON_ACCUM_A64FXd  \
-asm ( \
-    "fcadd z6.d, p5/m, z6.d, z18.d, 90 \n\t" \
-    "fadd z0.d, p5/m, z0.d, z18.d \n\t"  \
-    "fcadd z7.d, p5/m, z7.d, z19.d, 90 \n\t" \
-    "fadd z1.d, p5/m, z1.d, z19.d \n\t"  \
-    "fcadd z8.d, p5/m, z8.d, z20.d, 90 \n\t" \
-    "fadd z2.d, p5/m, z2.d, z20.d \n\t"  \
-    "fcadd z9.d, p5/m, z9.d, z21.d, 270 \n\t" \
-    "fadd z3.d, p5/m, z3.d, z21.d \n\t"  \
-    "fcadd z10.d, p5/m, z10.d, z22.d, 270 \n\t" \
-    "fadd z4.d, p5/m, z4.d, z22.d \n\t"  \
-    "fcadd z11.d, p5/m, z11.d, z23.d, 270 \n\t" \
-    "fadd z5.d, p5/m, z5.d, z23.d \n\t"  \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// TP_RECON_ACCUM
-#define TP_RECON_ACCUM_A64FXd  \
-asm ( \
-    "fadd z0.d, p5/m, z0.d, z18.d \n\t"  \
-    "fadd z6.d, p5/m, z6.d, z18.d \n\t"  \
-    "fadd z1.d, p5/m, z1.d, z19.d \n\t"  \
-    "fadd z7.d, p5/m, z7.d, z19.d \n\t"  \
-    "fadd z2.d, p5/m, z2.d, z20.d \n\t"  \
-    "fadd z8.d, p5/m, z8.d, z20.d \n\t"  \
-    "fadd z3.d, p5/m, z3.d, z21.d \n\t"  \
-    "fadd z9.d, p5/m, z9.d, z21.d \n\t"  \
-    "fadd z4.d, p5/m, z4.d, z22.d \n\t"  \
-    "fadd z10.d, p5/m, z10.d, z22.d \n\t"  \
-    "fadd z5.d, p5/m, z5.d, z23.d \n\t"  \
-    "fadd z11.d, p5/m, z11.d, z23.d \n\t"  \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// TM_RECON_ACCUM
-#define TM_RECON_ACCUM_A64FXd  \
-asm ( \
-    "fadd z0.d, p5/m, z0.d, z18.d \n\t"  \
-    "fsub z6.d, p5/m, z6.d, z18.d \n\t" \
-    "fadd z1.d, p5/m, z1.d, z19.d \n\t"  \
-    "fsub z7.d, p5/m, z7.d, z19.d \n\t" \
-    "fadd z2.d, p5/m, z2.d, z20.d \n\t"  \
-    "fsub z8.d, p5/m, z8.d, z20.d \n\t" \
-    "fadd z3.d, p5/m, z3.d, z21.d \n\t"  \
-    "fsub z9.d, p5/m, z9.d, z21.d \n\t" \
-    "fadd z4.d, p5/m, z4.d, z22.d \n\t"  \
-    "fsub z10.d, p5/m, z10.d, z22.d \n\t" \
-    "fadd z5.d, p5/m, z5.d, z23.d \n\t"  \
-    "fsub z11.d, p5/m, z11.d, z23.d \n\t" \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// ZERO_PSI
-#define ZERO_PSI_A64FXd  \
-asm ( \
-    "ptrue p5.d \n\t" \
-    "fmov z0.d , 0 \n\t" \
-    "fmov z1.d , 0 \n\t" \
-    "fmov z2.d , 0 \n\t" \
-    "fmov z3.d , 0 \n\t" \
-    "fmov z4.d , 0 \n\t" \
-    "fmov z5.d , 0 \n\t" \
-    "fmov z6.d , 0 \n\t" \
-    "fmov z7.d , 0 \n\t" \
-    "fmov z8.d , 0 \n\t" \
-    "fmov z9.d , 0 \n\t" \
-    "fmov z10.d , 0 \n\t" \
-    "fmov z11.d , 0 \n\t" \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// PREFETCH_RESULT_L2_STORE (prefetch store to L2)
-#define PREFETCH_RESULT_L2_STORE_INTERNAL_A64FXd(base)  \
-{ \
-asm ( \
-    "prfd PSTL2STRM, p5, [%[fetchptr], 0, mul vl] \n\t" \
-    "prfd PSTL2STRM, p5, [%[fetchptr], 4, mul vl] \n\t" \
-    "prfd PSTL2STRM, p5, [%[fetchptr], 8, mul vl] \n\t" \
-    :  \
-    : [fetchptr] "r" (base) \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31","memory" \
-); \
-}
-// PREFETCH_RESULT_L1_STORE (prefetch store to L1)
-#define PREFETCH_RESULT_L1_STORE_INTERNAL_A64FXd(base)  \
-{ \
-asm ( \
-    "prfd PSTL1STRM, p5, [%[fetchptr], 0, mul vl] \n\t" \
-    "prfd PSTL1STRM, p5, [%[fetchptr], 4, mul vl] \n\t" \
-    "prfd PSTL1STRM, p5, [%[fetchptr], 8, mul vl] \n\t" \
-    :  \
-    : [fetchptr] "r" (base) \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31","memory" \
-); \
-}
-// ADD_RESULT_INTERNAL
-#define ADD_RESULT_INTERNAL_A64FXd  \
-asm ( \
-    "fadd z0.d, p5/m, z0.d, z12.d \n\t"  \
-    "fadd z1.d, p5/m, z1.d, z13.d \n\t"  \
-    "fadd z2.d, p5/m, z2.d, z14.d \n\t"  \
-    "fadd z3.d, p5/m, z3.d, z15.d \n\t"  \
-    "fadd z4.d, p5/m, z4.d, z16.d \n\t"  \
-    "fadd z5.d, p5/m, z5.d, z17.d \n\t"  \
-    "fadd z6.d, p5/m, z6.d, z18.d \n\t"  \
-    "fadd z7.d, p5/m, z7.d, z19.d \n\t"  \
-    "fadd z8.d, p5/m, z8.d, z20.d \n\t"  \
-    "fadd z9.d, p5/m, z9.d, z21.d \n\t"  \
-    "fadd z10.d, p5/m, z10.d, z22.d \n\t"  \
-    "fadd z11.d, p5/m, z11.d, z23.d \n\t"  \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-

Grid/simd/Fujitsu_A64FX_asm_single.h

@@ -1,779 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: Fujitsu_A64FX_asm_single.h
-
-    Copyright (C) 2020
-
-Author: Nils Meyer <nils.meyer@ur.de>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#define LOAD_CHIMU(base)               LOAD_CHIMU_INTERLEAVED_A64FXf(base)  
-#define PREFETCH_CHIMU_L1(A)           PREFETCH_CHIMU_L1_INTERNAL_A64FXf(A)  
-#define PREFETCH_GAUGE_L1(A)           PREFETCH_GAUGE_L1_INTERNAL_A64FXf(A)  
-#define PREFETCH_CHIMU_L2(A)           PREFETCH_CHIMU_L2_INTERNAL_A64FXf(A)  
-#define PREFETCH_GAUGE_L2(A)           PREFETCH_GAUGE_L2_INTERNAL_A64FXf(A)  
-#define PF_GAUGE(A)  
-#define PREFETCH_RESULT_L2_STORE(A)    PREFETCH_RESULT_L2_STORE_INTERNAL_A64FXf(A)  
-#define PREFETCH_RESULT_L1_STORE(A)    PREFETCH_RESULT_L1_STORE_INTERNAL_A64FXf(A)  
-#define PREFETCH1_CHIMU(A)             PREFETCH_CHIMU_L1(A)  
-#define PREFETCH_CHIMU(A)              PREFETCH_CHIMU_L1(A)  
-#define LOCK_GAUGE(A)  
-#define UNLOCK_GAUGE(A)  
-#define MASK_REGS                      DECLARATIONS_A64FXf  
-#define SAVE_RESULT(A,B)               RESULT_A64FXf(A); PREFETCH_RESULT_L2_STORE(B)  
-#define MULT_2SPIN_1(Dir)              MULT_2SPIN_1_A64FXf(Dir)  
-#define MULT_2SPIN_2                   MULT_2SPIN_2_A64FXf  
-#define LOAD_CHI(base)                 LOAD_CHI_A64FXf(base)  
-#define ADD_RESULT(base,basep)         LOAD_CHIMU(base); ADD_RESULT_INTERNAL_A64FXf; RESULT_A64FXf(base)  
-#define XP_PROJ                        XP_PROJ_A64FXf  
-#define YP_PROJ                        YP_PROJ_A64FXf  
-#define ZP_PROJ                        ZP_PROJ_A64FXf  
-#define TP_PROJ                        TP_PROJ_A64FXf  
-#define XM_PROJ                        XM_PROJ_A64FXf  
-#define YM_PROJ                        YM_PROJ_A64FXf  
-#define ZM_PROJ                        ZM_PROJ_A64FXf  
-#define TM_PROJ                        TM_PROJ_A64FXf  
-#define XP_RECON                       XP_RECON_A64FXf  
-#define XM_RECON                       XM_RECON_A64FXf  
-#define XM_RECON_ACCUM                 XM_RECON_ACCUM_A64FXf  
-#define YM_RECON_ACCUM                 YM_RECON_ACCUM_A64FXf  
-#define ZM_RECON_ACCUM                 ZM_RECON_ACCUM_A64FXf  
-#define TM_RECON_ACCUM                 TM_RECON_ACCUM_A64FXf  
-#define XP_RECON_ACCUM                 XP_RECON_ACCUM_A64FXf  
-#define YP_RECON_ACCUM                 YP_RECON_ACCUM_A64FXf  
-#define ZP_RECON_ACCUM                 ZP_RECON_ACCUM_A64FXf  
-#define TP_RECON_ACCUM                 TP_RECON_ACCUM_A64FXf  
-#define PERMUTE_DIR0                   0  
-#define PERMUTE_DIR1                   1  
-#define PERMUTE_DIR2                   2  
-#define PERMUTE_DIR3                   3  
-#define PERMUTE                        PERMUTE_A64FXf;  
-#define LOAD_TABLE(Dir)                if (Dir == 0) { LOAD_TABLE0; } else if (Dir == 1) { LOAD_TABLE1 } else if (Dir == 2) { LOAD_TABLE2; } else if (Dir == 3) { LOAD_TABLE3; }  
-#define MAYBEPERM(A,perm)              if (perm) { PERMUTE; }  
-// DECLARATIONS
-#define DECLARATIONS_A64FXf  \
-    const uint32_t lut[4][16] = { \
-        {8, 9, 10, 11, 12, 13, 14, 15, 0, 1, 2, 3, 4, 5, 6, 7}, \
-        {4, 5, 6, 7, 0, 1, 2, 3, 12, 13, 14, 15, 8, 9, 10, 11}, \
-        {2, 3, 0, 1, 6, 7, 4, 5, 10, 11, 8, 9, 14, 15, 12, 13}, \
-        {1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14} }; \
-asm ( \
-    "fmov z31.s , 0 \n\t" \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// RESULT
-#define RESULT_A64FXf(base)  \
-{ \
-asm ( \
-    "str z0, [%[storeptr], -6, mul vl] \n\t" \
-    "str z1, [%[storeptr], -5, mul vl] \n\t" \
-    "str z2, [%[storeptr], -4, mul vl] \n\t" \
-    "str z3, [%[storeptr], -3, mul vl] \n\t" \
-    "str z4, [%[storeptr], -2, mul vl] \n\t" \
-    "str z5, [%[storeptr], -1, mul vl] \n\t" \
-    "str z6, [%[storeptr], 0, mul vl] \n\t" \
-    "str z7, [%[storeptr], 1, mul vl] \n\t" \
-    "str z8, [%[storeptr], 2, mul vl] \n\t" \
-    "str z9, [%[storeptr], 3, mul vl] \n\t" \
-    "str z10, [%[storeptr], 4, mul vl] \n\t" \
-    "str z11, [%[storeptr], 5, mul vl] \n\t" \
-    :  \
-    : [storeptr] "r" (base + 2 * 3 * 64) \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31","memory" \
-); \
-}
-// PREFETCH_CHIMU_L2 (prefetch to L2)
-#define PREFETCH_CHIMU_L2_INTERNAL_A64FXf(base)  \
-{ \
-asm ( \
-    "prfd PLDL2STRM, p5, [%[fetchptr], 0, mul vl] \n\t" \
-    "prfd PLDL2STRM, p5, [%[fetchptr], 4, mul vl] \n\t" \
-    "prfd PLDL2STRM, p5, [%[fetchptr], 8, mul vl] \n\t" \
-    :  \
-    : [fetchptr] "r" (base) \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31","memory" \
-); \
-}
-// PREFETCH_CHIMU_L1 (prefetch to L1)
-#define PREFETCH_CHIMU_L1_INTERNAL_A64FXf(base)  \
-{ \
-asm ( \
-    "prfd PLDL1STRM, p5, [%[fetchptr], 0, mul vl] \n\t" \
-    "prfd PLDL1STRM, p5, [%[fetchptr], 4, mul vl] \n\t" \
-    "prfd PLDL1STRM, p5, [%[fetchptr], 8, mul vl] \n\t" \
-    :  \
-    : [fetchptr] "r" (base) \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31","memory" \
-); \
-}
-// PREFETCH_GAUGE_L2 (prefetch to L2)
-#define PREFETCH_GAUGE_L2_INTERNAL_A64FXf(A)  \
-{ \
-    const auto & ref(U[sUn](A)); uint64_t baseU = (uint64_t)&ref + 3 * 3 * 64; \
-asm ( \
-    "prfd PLDL2STRM, p5, [%[fetchptr], -4, mul vl] \n\t" \
-    "prfd PLDL2STRM, p5, [%[fetchptr], 0, mul vl] \n\t" \
-    "prfd PLDL2STRM, p5, [%[fetchptr], 4, mul vl] \n\t" \
-    "prfd PLDL2STRM, p5, [%[fetchptr], 8, mul vl] \n\t" \
-    "prfd PLDL2STRM, p5, [%[fetchptr], 12, mul vl] \n\t" \
-    "prfd PLDL2STRM, p5, [%[fetchptr], 16, mul vl] \n\t" \
-    "prfd PLDL2STRM, p5, [%[fetchptr], 20, mul vl] \n\t" \
-    "prfd PLDL2STRM, p5, [%[fetchptr], 24, mul vl] \n\t" \
-    "prfd PLDL2STRM, p5, [%[fetchptr], 28, mul vl] \n\t" \
-    :  \
-    : [fetchptr] "r" (baseU) \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31","memory" \
-); \
-}
-// PREFETCH_GAUGE_L1 (prefetch to L1)
-#define PREFETCH_GAUGE_L1_INTERNAL_A64FXf(A)  \
-{ \
-    const auto & ref(U[sU](A)); uint64_t baseU = (uint64_t)&ref; \
-asm ( \
-    "prfd PLDL1STRM, p5, [%[fetchptr], 0, mul vl] \n\t" \
-    "prfd PLDL1STRM, p5, [%[fetchptr], 4, mul vl] \n\t" \
-    "prfd PLDL1STRM, p5, [%[fetchptr], 8, mul vl] \n\t" \
-    :  \
-    : [fetchptr] "r" (baseU) \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31","memory" \
-); \
-}
-// LOAD_CHI
-#define LOAD_CHI_A64FXf(base)  \
-{ \
-asm ( \
-    "ldr z12, [%[fetchptr], 0, mul vl] \n\t" \
-    "ldr z13, [%[fetchptr], 1, mul vl] \n\t" \
-    "ldr z14, [%[fetchptr], 2, mul vl] \n\t" \
-    "ldr z15, [%[fetchptr], 3, mul vl] \n\t" \
-    "ldr z16, [%[fetchptr], 4, mul vl] \n\t" \
-    "ldr z17, [%[fetchptr], 5, mul vl] \n\t" \
-    :  \
-    : [fetchptr] "r" (base) \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31","memory" \
-); \
-}
-// LOAD_CHIMU
-#define LOAD_CHIMU_INTERLEAVED_A64FXf(base)  \
-{ \
-asm ( \
-    "ptrue p5.s \n\t" \
-    "ldr z12, [%[fetchptr], -6, mul vl] \n\t" \
-    "ldr z21, [%[fetchptr], 3, mul vl] \n\t" \
-    "ldr z15, [%[fetchptr], -3, mul vl] \n\t" \
-    "ldr z18, [%[fetchptr], 0, mul vl] \n\t" \
-    "ldr z13, [%[fetchptr], -5, mul vl] \n\t" \
-    "ldr z22, [%[fetchptr], 4, mul vl] \n\t" \
-    "ldr z16, [%[fetchptr], -2, mul vl] \n\t" \
-    "ldr z19, [%[fetchptr], 1, mul vl] \n\t" \
-    "ldr z14, [%[fetchptr], -4, mul vl] \n\t" \
-    "ldr z23, [%[fetchptr], 5, mul vl] \n\t" \
-    "ldr z17, [%[fetchptr], -1, mul vl] \n\t" \
-    "ldr z20, [%[fetchptr], 2, mul vl] \n\t" \
-    :  \
-    : [fetchptr] "r" (base + 2 * 3 * 64) \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31","memory" \
-); \
-}
-// LOAD_CHIMU_0213
-#define LOAD_CHIMU_0213_A64FXf  \
-{ \
-    const SiteSpinor & ref(in[offset]); \
-asm ( \
-    "ptrue p5.s \n\t" \
-    "ldr z12, [%[fetchptr], -6, mul vl] \n\t" \
-    "ldr z18, [%[fetchptr], 0, mul vl] \n\t" \
-    "ldr z13, [%[fetchptr], -5, mul vl] \n\t" \
-    "ldr z19, [%[fetchptr], 1, mul vl] \n\t" \
-    "ldr z14, [%[fetchptr], -4, mul vl] \n\t" \
-    "ldr z20, [%[fetchptr], 2, mul vl] \n\t" \
-    "ldr z15, [%[fetchptr], -3, mul vl] \n\t" \
-    "ldr z21, [%[fetchptr], 3, mul vl] \n\t" \
-    "ldr z16, [%[fetchptr], -2, mul vl] \n\t" \
-    "ldr z22, [%[fetchptr], 4, mul vl] \n\t" \
-    "ldr z17, [%[fetchptr], -1, mul vl] \n\t" \
-    "ldr z23, [%[fetchptr], 5, mul vl] \n\t" \
-    :  \
-    : [fetchptr] "r" (&ref[2][0]) \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31","memory" \
-); \
-}
-// LOAD_CHIMU_0312
-#define LOAD_CHIMU_0312_A64FXf  \
-{ \
-    const SiteSpinor & ref(in[offset]); \
-asm ( \
-    "ptrue p5.s \n\t" \
-    "ldr z12, [%[fetchptr], -6, mul vl] \n\t" \
-    "ldr z21, [%[fetchptr], 3, mul vl] \n\t" \
-    "ldr z13, [%[fetchptr], -5, mul vl] \n\t" \
-    "ldr z22, [%[fetchptr], 4, mul vl] \n\t" \
-    "ldr z14, [%[fetchptr], -4, mul vl] \n\t" \
-    "ldr z23, [%[fetchptr], 5, mul vl] \n\t" \
-    "ldr z15, [%[fetchptr], -3, mul vl] \n\t" \
-    "ldr z18, [%[fetchptr], 0, mul vl] \n\t" \
-    "ldr z16, [%[fetchptr], -2, mul vl] \n\t" \
-    "ldr z19, [%[fetchptr], 1, mul vl] \n\t" \
-    "ldr z17, [%[fetchptr], -1, mul vl] \n\t" \
-    "ldr z20, [%[fetchptr], 2, mul vl] \n\t" \
-    :  \
-    : [fetchptr] "r" (&ref[2][0]) \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31","memory" \
-); \
-}
-// LOAD_TABLE0
-#define LOAD_TABLE0  \
-asm ( \
-    "ldr z30, [%[tableptr], %[index], mul vl] \n\t" \
-    :  \
-    : [tableptr] "r" (&lut[0]),[index] "i" (0) \
-    : "memory","cc","p5","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// LOAD_TABLE1
-#define LOAD_TABLE1  \
-asm ( \
-    "ldr z30, [%[tableptr], %[index], mul vl] \n\t" \
-    :  \
-    : [tableptr] "r" (&lut[0]),[index] "i" (1) \
-    : "memory","cc","p5","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// LOAD_TABLE2
-#define LOAD_TABLE2  \
-asm ( \
-    "ldr z30, [%[tableptr], %[index], mul vl] \n\t" \
-    :  \
-    : [tableptr] "r" (&lut[0]),[index] "i" (2) \
-    : "memory","cc","p5","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// LOAD_TABLE3
-#define LOAD_TABLE3  \
-asm ( \
-    "ldr z30, [%[tableptr], %[index], mul vl] \n\t" \
-    :  \
-    : [tableptr] "r" (&lut[0]),[index] "i" (3) \
-    : "memory","cc","p5","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// PERMUTE
-#define PERMUTE_A64FXf  \
-asm ( \
-    "tbl z12.s, { z12.s }, z30.s \n\t"  \
-    "tbl z13.s, { z13.s }, z30.s \n\t"  \
-    "tbl z14.s, { z14.s }, z30.s \n\t"  \
-    "tbl z15.s, { z15.s }, z30.s \n\t"  \
-    "tbl z16.s, { z16.s }, z30.s \n\t"  \
-    "tbl z17.s, { z17.s }, z30.s \n\t"  \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// LOAD_GAUGE
-#define LOAD_GAUGE  \
-    const auto & ref(U[sU](A)); uint64_t baseU = (uint64_t)&ref; \
-{ \
-asm ( \
-    "ptrue p5.s \n\t" \
-    "ldr z24, [%[fetchptr], -6, mul vl] \n\t" \
-    "ldr z25, [%[fetchptr], -3, mul vl] \n\t" \
-    "ldr z26, [%[fetchptr], 0, mul vl] \n\t" \
-    "ldr z27, [%[fetchptr], -5, mul vl] \n\t" \
-    "ldr z28, [%[fetchptr], -2, mul vl] \n\t" \
-    "ldr z29, [%[fetchptr], 1, mul vl] \n\t" \
-    :  \
-    : [fetchptr] "r" (baseU + 2 * 3 * 64) \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31","memory" \
-); \
-}
-// MULT_2SPIN
-#define MULT_2SPIN_1_A64FXf(A)  \
-{ \
-    const auto & ref(U[sU](A)); uint64_t baseU = (uint64_t)&ref; \
-asm ( \
-    "ldr z24, [%[fetchptr], -6, mul vl] \n\t" \
-    "ldr z25, [%[fetchptr], -3, mul vl] \n\t" \
-    "ldr z26, [%[fetchptr], 0, mul vl] \n\t" \
-    "ldr z27, [%[fetchptr], -5, mul vl] \n\t" \
-    "ldr z28, [%[fetchptr], -2, mul vl] \n\t" \
-    "ldr z29, [%[fetchptr], 1, mul vl] \n\t" \
-    "movprfx z18.s, p5/m, z31.s \n\t" \
-    "fcmla z18.s, p5/m, z24.s, z12.s, 0 \n\t" \
-    "movprfx z21.s, p5/m, z31.s \n\t" \
-    "fcmla z21.s, p5/m, z24.s, z15.s, 0 \n\t" \
-    "movprfx z19.s, p5/m, z31.s \n\t" \
-    "fcmla z19.s, p5/m, z25.s, z12.s, 0 \n\t" \
-    "movprfx z22.s, p5/m, z31.s \n\t" \
-    "fcmla z22.s, p5/m, z25.s, z15.s, 0 \n\t" \
-    "movprfx z20.s, p5/m, z31.s \n\t" \
-    "fcmla z20.s, p5/m, z26.s, z12.s, 0 \n\t" \
-    "movprfx z23.s, p5/m, z31.s \n\t" \
-    "fcmla z23.s, p5/m, z26.s, z15.s, 0 \n\t" \
-    "fcmla z18.s, p5/m, z24.s, z12.s, 90 \n\t" \
-    "fcmla z21.s, p5/m, z24.s, z15.s, 90 \n\t" \
-    "fcmla z19.s, p5/m, z25.s, z12.s, 90 \n\t" \
-    "fcmla z22.s, p5/m, z25.s, z15.s, 90 \n\t" \
-    "fcmla z20.s, p5/m, z26.s, z12.s, 90 \n\t" \
-    "fcmla z23.s, p5/m, z26.s, z15.s, 90 \n\t" \
-    "ldr z24, [%[fetchptr], -4, mul vl] \n\t" \
-    "ldr z25, [%[fetchptr], -1, mul vl] \n\t" \
-    "ldr z26, [%[fetchptr], 2, mul vl] \n\t" \
-    :  \
-    : [fetchptr] "r" (baseU + 2 * 3 * 64) \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31","memory" \
-); \
-}
-// MULT_2SPIN_BACKEND
-#define MULT_2SPIN_2_A64FXf  \
-{ \
-asm ( \
-    "fcmla z18.s, p5/m, z27.s, z13.s, 0 \n\t" \
-    "fcmla z21.s, p5/m, z27.s, z16.s, 0 \n\t" \
-    "fcmla z19.s, p5/m, z28.s, z13.s, 0 \n\t" \
-    "fcmla z22.s, p5/m, z28.s, z16.s, 0 \n\t" \
-    "fcmla z20.s, p5/m, z29.s, z13.s, 0 \n\t" \
-    "fcmla z23.s, p5/m, z29.s, z16.s, 0 \n\t" \
-    "fcmla z18.s, p5/m, z27.s, z13.s, 90 \n\t" \
-    "fcmla z21.s, p5/m, z27.s, z16.s, 90 \n\t" \
-    "fcmla z19.s, p5/m, z28.s, z13.s, 90 \n\t" \
-    "fcmla z22.s, p5/m, z28.s, z16.s, 90 \n\t" \
-    "fcmla z20.s, p5/m, z29.s, z13.s, 90 \n\t" \
-    "fcmla z23.s, p5/m, z29.s, z16.s, 90 \n\t" \
-    "fcmla z18.s, p5/m, z24.s, z14.s, 0 \n\t" \
-    "fcmla z21.s, p5/m, z24.s, z17.s, 0 \n\t" \
-    "fcmla z19.s, p5/m, z25.s, z14.s, 0 \n\t" \
-    "fcmla z22.s, p5/m, z25.s, z17.s, 0 \n\t" \
-    "fcmla z20.s, p5/m, z26.s, z14.s, 0 \n\t" \
-    "fcmla z23.s, p5/m, z26.s, z17.s, 0 \n\t" \
-    "fcmla z18.s, p5/m, z24.s, z14.s, 90 \n\t" \
-    "fcmla z21.s, p5/m, z24.s, z17.s, 90 \n\t" \
-    "fcmla z19.s, p5/m, z25.s, z14.s, 90 \n\t" \
-    "fcmla z22.s, p5/m, z25.s, z17.s, 90 \n\t" \
-    "fcmla z20.s, p5/m, z26.s, z14.s, 90 \n\t" \
-    "fcmla z23.s, p5/m, z26.s, z17.s, 90 \n\t" \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); \
-}
-// XP_PROJ
-#define XP_PROJ_A64FXf  \
-{ \
-asm ( \
-    "fcadd z12.s, p5/m, z12.s, z21.s, 90 \n\t" \
-    "fcadd z13.s, p5/m, z13.s, z22.s, 90 \n\t" \
-    "fcadd z14.s, p5/m, z14.s, z23.s, 90 \n\t" \
-    "fcadd z15.s, p5/m, z15.s, z18.s, 90 \n\t" \
-    "fcadd z16.s, p5/m, z16.s, z19.s, 90 \n\t" \
-    "fcadd z17.s, p5/m, z17.s, z20.s, 90 \n\t" \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); \
-}
-// XP_RECON
-#define XP_RECON_A64FXf  \
-asm ( \
-    "movprfx z6.s, p5/m, z31.s \n\t" \
-    "fcadd z6.s, p5/m, z6.s, z21.s, 270 \n\t" \
-    "movprfx z7.s, p5/m, z31.s \n\t" \
-    "fcadd z7.s, p5/m, z7.s, z22.s, 270 \n\t" \
-    "movprfx z8.s, p5/m, z31.s \n\t" \
-    "fcadd z8.s, p5/m, z8.s, z23.s, 270 \n\t" \
-    "movprfx z9.s, p5/m, z31.s \n\t" \
-    "fcadd z9.s, p5/m, z9.s, z18.s, 270 \n\t" \
-    "movprfx z10.s, p5/m, z31.s \n\t" \
-    "fcadd z10.s, p5/m, z10.s, z19.s, 270 \n\t" \
-    "movprfx z11.s, p5/m, z31.s \n\t" \
-    "fcadd z11.s, p5/m, z11.s, z20.s, 270 \n\t" \
-    "mov z0.s, p5/m, z18.s \n\t" \
-    "mov z1.s, p5/m, z19.s \n\t" \
-    "mov z2.s, p5/m, z20.s \n\t" \
-    "mov z3.s, p5/m, z21.s \n\t" \
-    "mov z4.s, p5/m, z22.s \n\t" \
-    "mov z5.s, p5/m, z23.s \n\t" \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// XP_RECON_ACCUM
-#define XP_RECON_ACCUM_A64FXf  \
-asm ( \
-    "fcadd z9.s, p5/m, z9.s, z18.s, 270 \n\t" \
-    "fadd z0.s, p5/m, z0.s, z18.s \n\t"  \
-    "fcadd z10.s, p5/m, z10.s, z19.s, 270 \n\t" \
-    "fadd z1.s, p5/m, z1.s, z19.s \n\t"  \
-    "fcadd z11.s, p5/m, z11.s, z20.s, 270 \n\t" \
-    "fadd z2.s, p5/m, z2.s, z20.s \n\t"  \
-    "fcadd z6.s, p5/m, z6.s, z21.s, 270 \n\t" \
-    "fadd z3.s, p5/m, z3.s, z21.s \n\t"  \
-    "fcadd z7.s, p5/m, z7.s, z22.s, 270 \n\t" \
-    "fadd z4.s, p5/m, z4.s, z22.s \n\t"  \
-    "fcadd z8.s, p5/m, z8.s, z23.s, 270 \n\t" \
-    "fadd z5.s, p5/m, z5.s, z23.s \n\t"  \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// YP_PROJ
-#define YP_PROJ_A64FXf  \
-{ \
-asm ( \
-    "fsub z12.s, p5/m, z12.s, z21.s \n\t" \
-    "fsub z13.s, p5/m, z13.s, z22.s \n\t" \
-    "fsub z14.s, p5/m, z14.s, z23.s \n\t" \
-    "fadd z15.s, p5/m, z15.s, z18.s \n\t"  \
-    "fadd z16.s, p5/m, z16.s, z19.s \n\t"  \
-    "fadd z17.s, p5/m, z17.s, z20.s \n\t"  \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); \
-}
-// ZP_PROJ
-#define ZP_PROJ_A64FXf  \
-{ \
-asm ( \
-    "fcadd z12.s, p5/m, z12.s, z18.s, 90 \n\t" \
-    "fcadd z13.s, p5/m, z13.s, z19.s, 90 \n\t" \
-    "fcadd z14.s, p5/m, z14.s, z20.s, 90 \n\t" \
-    "fcadd z15.s, p5/m, z15.s, z21.s, 270 \n\t" \
-    "fcadd z16.s, p5/m, z16.s, z22.s, 270 \n\t" \
-    "fcadd z17.s, p5/m, z17.s, z23.s, 270 \n\t" \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); \
-}
-// TP_PROJ
-#define TP_PROJ_A64FXf  \
-{ \
-asm ( \
-    "fadd z12.s, p5/m, z12.s, z18.s \n\t"  \
-    "fadd z13.s, p5/m, z13.s, z19.s \n\t"  \
-    "fadd z14.s, p5/m, z14.s, z20.s \n\t"  \
-    "fadd z15.s, p5/m, z15.s, z21.s \n\t"  \
-    "fadd z16.s, p5/m, z16.s, z22.s \n\t"  \
-    "fadd z17.s, p5/m, z17.s, z23.s \n\t"  \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); \
-}
-// XM_PROJ
-#define XM_PROJ_A64FXf  \
-{ \
-asm ( \
-    "fcadd z12.s, p5/m, z12.s, z21.s, 270 \n\t" \
-    "fcadd z13.s, p5/m, z13.s, z22.s, 270 \n\t" \
-    "fcadd z14.s, p5/m, z14.s, z23.s, 270 \n\t" \
-    "fcadd z15.s, p5/m, z15.s, z18.s, 270 \n\t" \
-    "fcadd z16.s, p5/m, z16.s, z19.s, 270 \n\t" \
-    "fcadd z17.s, p5/m, z17.s, z20.s, 270 \n\t" \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); \
-}
-// XM_RECON
-#define XM_RECON_A64FXf  \
-asm ( \
-    "movprfx z6.s, p5/m, z31.s \n\t" \
-    "fcadd z6.s, p5/m, z6.s, z21.s, 90 \n\t" \
-    "movprfx z7.s, p5/m, z31.s \n\t" \
-    "fcadd z7.s, p5/m, z7.s, z22.s, 90 \n\t" \
-    "movprfx z8.s, p5/m, z31.s \n\t" \
-    "fcadd z8.s, p5/m, z8.s, z23.s, 90 \n\t" \
-    "movprfx z9.s, p5/m, z31.s \n\t" \
-    "fcadd z9.s, p5/m, z9.s, z18.s, 90 \n\t" \
-    "movprfx z10.s, p5/m, z31.s \n\t" \
-    "fcadd z10.s, p5/m, z10.s, z19.s, 90 \n\t" \
-    "movprfx z11.s, p5/m, z31.s \n\t" \
-    "fcadd z11.s, p5/m, z11.s, z20.s, 90 \n\t" \
-    "mov z0.s, p5/m, z18.s \n\t" \
-    "mov z1.s, p5/m, z19.s \n\t" \
-    "mov z2.s, p5/m, z20.s \n\t" \
-    "mov z3.s, p5/m, z21.s \n\t" \
-    "mov z4.s, p5/m, z22.s \n\t" \
-    "mov z5.s, p5/m, z23.s \n\t" \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// YM_PROJ
-#define YM_PROJ_A64FXf  \
-{ \
-asm ( \
-    "fadd z12.s, p5/m, z12.s, z21.s \n\t"  \
-    "fadd z13.s, p5/m, z13.s, z22.s \n\t"  \
-    "fadd z14.s, p5/m, z14.s, z23.s \n\t"  \
-    "fsub z15.s, p5/m, z15.s, z18.s \n\t" \
-    "fsub z16.s, p5/m, z16.s, z19.s \n\t" \
-    "fsub z17.s, p5/m, z17.s, z20.s \n\t" \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); \
-}
-// ZM_PROJ
-#define ZM_PROJ_A64FXf  \
-{ \
-asm ( \
-    "fcadd z12.s, p5/m, z12.s, z18.s, 270 \n\t" \
-    "fcadd z13.s, p5/m, z13.s, z19.s, 270 \n\t" \
-    "fcadd z14.s, p5/m, z14.s, z20.s, 270 \n\t" \
-    "fcadd z15.s, p5/m, z15.s, z21.s, 90 \n\t" \
-    "fcadd z16.s, p5/m, z16.s, z22.s, 90 \n\t" \
-    "fcadd z17.s, p5/m, z17.s, z23.s, 90 \n\t" \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); \
-}
-// TM_PROJ
-#define TM_PROJ_A64FXf  \
-{ \
-asm ( \
-    "ptrue p5.s \n\t" \
-    "fsub z12.s, p5/m, z12.s, z18.s \n\t" \
-    "fsub z13.s, p5/m, z13.s, z19.s \n\t" \
-    "fsub z14.s, p5/m, z14.s, z20.s \n\t" \
-    "fsub z15.s, p5/m, z15.s, z21.s \n\t" \
-    "fsub z16.s, p5/m, z16.s, z22.s \n\t" \
-    "fsub z17.s, p5/m, z17.s, z23.s \n\t" \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); \
-}
-// XM_RECON_ACCUM
-#define XM_RECON_ACCUM_A64FXf  \
-asm ( \
-    "fcadd z9.s, p5/m, z9.s, z18.s, 90 \n\t" \
-    "fcadd z10.s, p5/m, z10.s, z19.s, 90 \n\t" \
-    "fcadd z11.s, p5/m, z11.s, z20.s, 90 \n\t" \
-    "fcadd z6.s, p5/m, z6.s, z21.s, 90 \n\t" \
-    "fcadd z7.s, p5/m, z7.s, z22.s, 90 \n\t" \
-    "fcadd z8.s, p5/m, z8.s, z23.s, 90 \n\t" \
-    "fadd z0.s, p5/m, z0.s, z18.s \n\t"  \
-    "fadd z1.s, p5/m, z1.s, z19.s \n\t"  \
-    "fadd z2.s, p5/m, z2.s, z20.s \n\t"  \
-    "fadd z3.s, p5/m, z3.s, z21.s \n\t"  \
-    "fadd z4.s, p5/m, z4.s, z22.s \n\t"  \
-    "fadd z5.s, p5/m, z5.s, z23.s \n\t"  \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// YP_RECON_ACCUM
-#define YP_RECON_ACCUM_A64FXf  \
-asm ( \
-    "fadd z0.s, p5/m, z0.s, z18.s \n\t"  \
-    "fsub z9.s, p5/m, z9.s, z18.s \n\t" \
-    "fadd z1.s, p5/m, z1.s, z19.s \n\t"  \
-    "fsub z10.s, p5/m, z10.s, z19.s \n\t" \
-    "fadd z2.s, p5/m, z2.s, z20.s \n\t"  \
-    "fsub z11.s, p5/m, z11.s, z20.s \n\t" \
-    "fadd z3.s, p5/m, z3.s, z21.s \n\t"  \
-    "fadd z6.s, p5/m, z6.s, z21.s \n\t"  \
-    "fadd z4.s, p5/m, z4.s, z22.s \n\t"  \
-    "fadd z7.s, p5/m, z7.s, z22.s \n\t"  \
-    "fadd z5.s, p5/m, z5.s, z23.s \n\t"  \
-    "fadd z8.s, p5/m, z8.s, z23.s \n\t"  \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// YM_RECON_ACCUM
-#define YM_RECON_ACCUM_A64FXf  \
-asm ( \
-    "fadd z0.s, p5/m, z0.s, z18.s \n\t"  \
-    "fadd z9.s, p5/m, z9.s, z18.s \n\t"  \
-    "fadd z1.s, p5/m, z1.s, z19.s \n\t"  \
-    "fadd z10.s, p5/m, z10.s, z19.s \n\t"  \
-    "fadd z2.s, p5/m, z2.s, z20.s \n\t"  \
-    "fadd z11.s, p5/m, z11.s, z20.s \n\t"  \
-    "fadd z3.s, p5/m, z3.s, z21.s \n\t"  \
-    "fsub z6.s, p5/m, z6.s, z21.s \n\t" \
-    "fadd z4.s, p5/m, z4.s, z22.s \n\t"  \
-    "fsub z7.s, p5/m, z7.s, z22.s \n\t" \
-    "fadd z5.s, p5/m, z5.s, z23.s \n\t"  \
-    "fsub z8.s, p5/m, z8.s, z23.s \n\t" \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// ZP_RECON_ACCUM
-#define ZP_RECON_ACCUM_A64FXf  \
-asm ( \
-    "fcadd z6.s, p5/m, z6.s, z18.s, 270 \n\t" \
-    "fadd z0.s, p5/m, z0.s, z18.s \n\t"  \
-    "fcadd z7.s, p5/m, z7.s, z19.s, 270 \n\t" \
-    "fadd z1.s, p5/m, z1.s, z19.s \n\t"  \
-    "fcadd z8.s, p5/m, z8.s, z20.s, 270 \n\t" \
-    "fadd z2.s, p5/m, z2.s, z20.s \n\t"  \
-    "fcadd z9.s, p5/m, z9.s, z21.s, 90 \n\t" \
-    "fadd z3.s, p5/m, z3.s, z21.s \n\t"  \
-    "fcadd z10.s, p5/m, z10.s, z22.s, 90 \n\t" \
-    "fadd z4.s, p5/m, z4.s, z22.s \n\t"  \
-    "fcadd z11.s, p5/m, z11.s, z23.s, 90 \n\t" \
-    "fadd z5.s, p5/m, z5.s, z23.s \n\t"  \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// ZM_RECON_ACCUM
-#define ZM_RECON_ACCUM_A64FXf  \
-asm ( \
-    "fcadd z6.s, p5/m, z6.s, z18.s, 90 \n\t" \
-    "fadd z0.s, p5/m, z0.s, z18.s \n\t"  \
-    "fcadd z7.s, p5/m, z7.s, z19.s, 90 \n\t" \
-    "fadd z1.s, p5/m, z1.s, z19.s \n\t"  \
-    "fcadd z8.s, p5/m, z8.s, z20.s, 90 \n\t" \
-    "fadd z2.s, p5/m, z2.s, z20.s \n\t"  \
-    "fcadd z9.s, p5/m, z9.s, z21.s, 270 \n\t" \
-    "fadd z3.s, p5/m, z3.s, z21.s \n\t"  \
-    "fcadd z10.s, p5/m, z10.s, z22.s, 270 \n\t" \
-    "fadd z4.s, p5/m, z4.s, z22.s \n\t"  \
-    "fcadd z11.s, p5/m, z11.s, z23.s, 270 \n\t" \
-    "fadd z5.s, p5/m, z5.s, z23.s \n\t"  \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// TP_RECON_ACCUM
-#define TP_RECON_ACCUM_A64FXf  \
-asm ( \
-    "fadd z0.s, p5/m, z0.s, z18.s \n\t"  \
-    "fadd z6.s, p5/m, z6.s, z18.s \n\t"  \
-    "fadd z1.s, p5/m, z1.s, z19.s \n\t"  \
-    "fadd z7.s, p5/m, z7.s, z19.s \n\t"  \
-    "fadd z2.s, p5/m, z2.s, z20.s \n\t"  \
-    "fadd z8.s, p5/m, z8.s, z20.s \n\t"  \
-    "fadd z3.s, p5/m, z3.s, z21.s \n\t"  \
-    "fadd z9.s, p5/m, z9.s, z21.s \n\t"  \
-    "fadd z4.s, p5/m, z4.s, z22.s \n\t"  \
-    "fadd z10.s, p5/m, z10.s, z22.s \n\t"  \
-    "fadd z5.s, p5/m, z5.s, z23.s \n\t"  \
-    "fadd z11.s, p5/m, z11.s, z23.s \n\t"  \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// TM_RECON_ACCUM
-#define TM_RECON_ACCUM_A64FXf  \
-asm ( \
-    "fadd z0.s, p5/m, z0.s, z18.s \n\t"  \
-    "fsub z6.s, p5/m, z6.s, z18.s \n\t" \
-    "fadd z1.s, p5/m, z1.s, z19.s \n\t"  \
-    "fsub z7.s, p5/m, z7.s, z19.s \n\t" \
-    "fadd z2.s, p5/m, z2.s, z20.s \n\t"  \
-    "fsub z8.s, p5/m, z8.s, z20.s \n\t" \
-    "fadd z3.s, p5/m, z3.s, z21.s \n\t"  \
-    "fsub z9.s, p5/m, z9.s, z21.s \n\t" \
-    "fadd z4.s, p5/m, z4.s, z22.s \n\t"  \
-    "fsub z10.s, p5/m, z10.s, z22.s \n\t" \
-    "fadd z5.s, p5/m, z5.s, z23.s \n\t"  \
-    "fsub z11.s, p5/m, z11.s, z23.s \n\t" \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// ZERO_PSI
-#define ZERO_PSI_A64FXf  \
-asm ( \
-    "ptrue p5.s \n\t" \
-    "fmov z0.s , 0 \n\t" \
-    "fmov z1.s , 0 \n\t" \
-    "fmov z2.s , 0 \n\t" \
-    "fmov z3.s , 0 \n\t" \
-    "fmov z4.s , 0 \n\t" \
-    "fmov z5.s , 0 \n\t" \
-    "fmov z6.s , 0 \n\t" \
-    "fmov z7.s , 0 \n\t" \
-    "fmov z8.s , 0 \n\t" \
-    "fmov z9.s , 0 \n\t" \
-    "fmov z10.s , 0 \n\t" \
-    "fmov z11.s , 0 \n\t" \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-
-// PREFETCH_RESULT_L2_STORE (prefetch store to L2)
-#define PREFETCH_RESULT_L2_STORE_INTERNAL_A64FXf(base)  \
-{ \
-asm ( \
-    "prfd PSTL2STRM, p5, [%[fetchptr], 0, mul vl] \n\t" \
-    "prfd PSTL2STRM, p5, [%[fetchptr], 4, mul vl] \n\t" \
-    "prfd PSTL2STRM, p5, [%[fetchptr], 8, mul vl] \n\t" \
-    :  \
-    : [fetchptr] "r" (base) \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31","memory" \
-); \
-}
-// PREFETCH_RESULT_L1_STORE (prefetch store to L1)
-#define PREFETCH_RESULT_L1_STORE_INTERNAL_A64FXf(base)  \
-{ \
-asm ( \
-    "prfd PSTL1STRM, p5, [%[fetchptr], 0, mul vl] \n\t" \
-    "prfd PSTL1STRM, p5, [%[fetchptr], 4, mul vl] \n\t" \
-    "prfd PSTL1STRM, p5, [%[fetchptr], 8, mul vl] \n\t" \
-    :  \
-    : [fetchptr] "r" (base) \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31","memory" \
-); \
-}
-// ADD_RESULT_INTERNAL
-#define ADD_RESULT_INTERNAL_A64FXf  \
-asm ( \
-    "fadd z0.s, p5/m, z0.s, z12.s \n\t"  \
-    "fadd z1.s, p5/m, z1.s, z13.s \n\t"  \
-    "fadd z2.s, p5/m, z2.s, z14.s \n\t"  \
-    "fadd z3.s, p5/m, z3.s, z15.s \n\t"  \
-    "fadd z4.s, p5/m, z4.s, z16.s \n\t"  \
-    "fadd z5.s, p5/m, z5.s, z17.s \n\t"  \
-    "fadd z6.s, p5/m, z6.s, z18.s \n\t"  \
-    "fadd z7.s, p5/m, z7.s, z19.s \n\t"  \
-    "fadd z8.s, p5/m, z8.s, z20.s \n\t"  \
-    "fadd z9.s, p5/m, z9.s, z21.s \n\t"  \
-    "fadd z10.s, p5/m, z10.s, z22.s \n\t"  \
-    "fadd z11.s, p5/m, z11.s, z23.s \n\t"  \
-    :  \
-    :  \
-    : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
-); 
-

Grid/simd/Fujitsu_A64FX_intrin_double.h

@@ -1,601 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: Fujitsu_A64FX_intrin_double.h
-
-    Copyright (C) 2020
-
-Author: Nils Meyer <nils.meyer@ur.de>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#define LOAD_CHIMU(base)               LOAD_CHIMU_INTERLEAVED_A64FXd(base)  
-#define PREFETCH_CHIMU_L1(A)           PREFETCH_CHIMU_L1_INTERNAL_A64FXd(A)  
-#define PREFETCH_GAUGE_L1(A)           PREFETCH_GAUGE_L1_INTERNAL_A64FXd(A)  
-#define PREFETCH_CHIMU_L2(A)           PREFETCH_CHIMU_L2_INTERNAL_A64FXd(A)  
-#define PREFETCH_GAUGE_L2(A)           PREFETCH_GAUGE_L2_INTERNAL_A64FXd(A)  
-#define PF_GAUGE(A)  
-#define PREFETCH_RESULT_L2_STORE(A)    PREFETCH_RESULT_L2_STORE_INTERNAL_A64FXd(A)  
-#define PREFETCH_RESULT_L1_STORE(A)    PREFETCH_RESULT_L1_STORE_INTERNAL_A64FXd(A)  
-#define PREFETCH1_CHIMU(A)             PREFETCH_CHIMU_L1(A)  
-#define PREFETCH_CHIMU(A)              PREFETCH_CHIMU_L1(A)  
-#define LOCK_GAUGE(A)  
-#define UNLOCK_GAUGE(A)  
-#define MASK_REGS                      DECLARATIONS_A64FXd  
-#define SAVE_RESULT(A,B)               RESULT_A64FXd(A); PREFETCH_RESULT_L2_STORE(B)  
-#define MULT_2SPIN_1(Dir)              MULT_2SPIN_1_A64FXd(Dir)  
-#define MULT_2SPIN_2                   MULT_2SPIN_2_A64FXd  
-#define LOAD_CHI(base)                 LOAD_CHI_A64FXd(base)  
-#define ADD_RESULT(base,basep)         LOAD_CHIMU(base); ADD_RESULT_INTERNAL_A64FXd; RESULT_A64FXd(base)  
-#define XP_PROJ                        XP_PROJ_A64FXd  
-#define YP_PROJ                        YP_PROJ_A64FXd  
-#define ZP_PROJ                        ZP_PROJ_A64FXd  
-#define TP_PROJ                        TP_PROJ_A64FXd  
-#define XM_PROJ                        XM_PROJ_A64FXd  
-#define YM_PROJ                        YM_PROJ_A64FXd  
-#define ZM_PROJ                        ZM_PROJ_A64FXd  
-#define TM_PROJ                        TM_PROJ_A64FXd  
-#define XP_RECON                       XP_RECON_A64FXd  
-#define XM_RECON                       XM_RECON_A64FXd  
-#define XM_RECON_ACCUM                 XM_RECON_ACCUM_A64FXd  
-#define YM_RECON_ACCUM                 YM_RECON_ACCUM_A64FXd  
-#define ZM_RECON_ACCUM                 ZM_RECON_ACCUM_A64FXd  
-#define TM_RECON_ACCUM                 TM_RECON_ACCUM_A64FXd  
-#define XP_RECON_ACCUM                 XP_RECON_ACCUM_A64FXd  
-#define YP_RECON_ACCUM                 YP_RECON_ACCUM_A64FXd  
-#define ZP_RECON_ACCUM                 ZP_RECON_ACCUM_A64FXd  
-#define TP_RECON_ACCUM                 TP_RECON_ACCUM_A64FXd  
-#define PERMUTE_DIR0                   0  
-#define PERMUTE_DIR1                   1  
-#define PERMUTE_DIR2                   2  
-#define PERMUTE_DIR3                   3  
-#define PERMUTE                        PERMUTE_A64FXd;  
-#define LOAD_TABLE(Dir)                if (Dir == 0) { LOAD_TABLE0; } else if (Dir == 1) { LOAD_TABLE1; } else if (Dir == 2) { LOAD_TABLE2; }  
-#define MAYBEPERM(Dir,perm)            if (Dir != 3) { if (perm) { PERMUTE; } }  
-// DECLARATIONS
-#define DECLARATIONS_A64FXd  \
-    const uint64_t lut[4][8] = { \
-        {4, 5, 6, 7, 0, 1, 2, 3}, \
-        {2, 3, 0, 1, 6, 7, 4, 5}, \
-        {1, 0, 3, 2, 5, 4, 7, 6}, \
-        {0, 1, 2, 4, 5, 6, 7, 8} };\
-    svfloat64_t result_00;        \
-    svfloat64_t result_01;        \
-    svfloat64_t result_02;        \
-    svfloat64_t result_10;        \
-    svfloat64_t result_11;        \
-    svfloat64_t result_12;        \
-    svfloat64_t result_20;        \
-    svfloat64_t result_21;        \
-    svfloat64_t result_22;        \
-    svfloat64_t result_30;        \
-    svfloat64_t result_31;        \
-    svfloat64_t result_32;        \
-    svfloat64_t Chi_00;        \
-    svfloat64_t Chi_01;        \
-    svfloat64_t Chi_02;        \
-    svfloat64_t Chi_10;        \
-    svfloat64_t Chi_11;        \
-    svfloat64_t Chi_12;        \
-    svfloat64_t UChi_00;        \
-    svfloat64_t UChi_01;        \
-    svfloat64_t UChi_02;        \
-    svfloat64_t UChi_10;        \
-    svfloat64_t UChi_11;        \
-    svfloat64_t UChi_12;        \
-    svfloat64_t U_00;        \
-    svfloat64_t U_10;        \
-    svfloat64_t U_20;        \
-    svfloat64_t U_01;        \
-    svfloat64_t U_11;        \
-    svfloat64_t U_21;        \
-    svbool_t pg1;        \
-    pg1 = svptrue_b64();        \
-    svuint64_t table0; \
-    svfloat64_t zero0;        \
-    zero0 = svdup_f64(0.); 
-
-#define Chimu_00 Chi_00  
-#define Chimu_01 Chi_01  
-#define Chimu_02 Chi_02  
-#define Chimu_10 Chi_10  
-#define Chimu_11 Chi_11  
-#define Chimu_12 Chi_12  
-#define Chimu_20 UChi_00  
-#define Chimu_21 UChi_01  
-#define Chimu_22 UChi_02  
-#define Chimu_30 UChi_10  
-#define Chimu_31 UChi_11  
-#define Chimu_32 UChi_12  
-// RESULT
-#define RESULT_A64FXd(base)  \
-{ \
-    svst1(pg1, (float64_t*)(base + 2 * 3 * 64 + -6 * 64), result_00);  \
-    svst1(pg1, (float64_t*)(base + 2 * 3 * 64 + -5 * 64), result_01);  \
-    svst1(pg1, (float64_t*)(base + 2 * 3 * 64 + -4 * 64), result_02);  \
-    svst1(pg1, (float64_t*)(base + 2 * 3 * 64 + -3 * 64), result_10);  \
-    svst1(pg1, (float64_t*)(base + 2 * 3 * 64 + -2 * 64), result_11);  \
-    svst1(pg1, (float64_t*)(base + 2 * 3 * 64 + -1 * 64), result_12);  \
-    svst1(pg1, (float64_t*)(base + 2 * 3 * 64 + 0 * 64), result_20);  \
-    svst1(pg1, (float64_t*)(base + 2 * 3 * 64 + 1 * 64), result_21);  \
-    svst1(pg1, (float64_t*)(base + 2 * 3 * 64 + 2 * 64), result_22);  \
-    svst1(pg1, (float64_t*)(base + 2 * 3 * 64 + 3 * 64), result_30);  \
-    svst1(pg1, (float64_t*)(base + 2 * 3 * 64 + 4 * 64), result_31);  \
-    svst1(pg1, (float64_t*)(base + 2 * 3 * 64 + 5 * 64), result_32);  \
-}
-// PREFETCH_CHIMU_L2 (prefetch to L2)
-#define PREFETCH_CHIMU_L2_INTERNAL_A64FXd(base)  \
-{ \
-    svprfd(pg1, (int64_t*)(base + 0), SV_PLDL2STRM); \
-    svprfd(pg1, (int64_t*)(base + 256), SV_PLDL2STRM); \
-    svprfd(pg1, (int64_t*)(base + 512), SV_PLDL2STRM); \
-}
-// PREFETCH_CHIMU_L1 (prefetch to L1)
-#define PREFETCH_CHIMU_L1_INTERNAL_A64FXd(base)  \
-{ \
-    svprfd(pg1, (int64_t*)(base + 0), SV_PLDL1STRM); \
-    svprfd(pg1, (int64_t*)(base + 256), SV_PLDL1STRM); \
-    svprfd(pg1, (int64_t*)(base + 512), SV_PLDL1STRM); \
-}
-// PREFETCH_GAUGE_L2 (prefetch to L2)
-#define PREFETCH_GAUGE_L2_INTERNAL_A64FXd(A)  \
-{ \
-    const auto & ref(U[sUn](A)); uint64_t baseU = (uint64_t)&ref + 3 * 3 * 64; \
-    svprfd(pg1, (int64_t*)(baseU + -256), SV_PLDL2STRM); \
-    svprfd(pg1, (int64_t*)(baseU + 0), SV_PLDL2STRM); \
-    svprfd(pg1, (int64_t*)(baseU + 256), SV_PLDL2STRM); \
-    svprfd(pg1, (int64_t*)(baseU + 512), SV_PLDL2STRM); \
-    svprfd(pg1, (int64_t*)(baseU + 768), SV_PLDL2STRM); \
-    svprfd(pg1, (int64_t*)(baseU + 1024), SV_PLDL2STRM); \
-    svprfd(pg1, (int64_t*)(baseU + 1280), SV_PLDL2STRM); \
-    svprfd(pg1, (int64_t*)(baseU + 1536), SV_PLDL2STRM); \
-    svprfd(pg1, (int64_t*)(baseU + 1792), SV_PLDL2STRM); \
-}
-// PREFETCH_GAUGE_L1 (prefetch to L1)
-#define PREFETCH_GAUGE_L1_INTERNAL_A64FXd(A)  \
-{ \
-    const auto & ref(U[sU](A)); uint64_t baseU = (uint64_t)&ref; \
-    svprfd(pg1, (int64_t*)(baseU + 0), SV_PLDL1STRM); \
-    svprfd(pg1, (int64_t*)(baseU + 256), SV_PLDL1STRM); \
-    svprfd(pg1, (int64_t*)(baseU + 512), SV_PLDL1STRM); \
-}
-// LOAD_CHI
-#define LOAD_CHI_A64FXd(base)  \
-{ \
-    Chi_00 = svld1(pg1, (float64_t*)(base + 0 * 64));  \
-    Chi_01 = svld1(pg1, (float64_t*)(base + 1 * 64));  \
-    Chi_02 = svld1(pg1, (float64_t*)(base + 2 * 64));  \
-    Chi_10 = svld1(pg1, (float64_t*)(base + 3 * 64));  \
-    Chi_11 = svld1(pg1, (float64_t*)(base + 4 * 64));  \
-    Chi_12 = svld1(pg1, (float64_t*)(base + 5 * 64));  \
-}
-// LOAD_CHIMU
-#define LOAD_CHIMU_INTERLEAVED_A64FXd(base)  \
-{ \
-    Chimu_00 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + -6 * 64));  \
-    Chimu_30 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + 3 * 64));  \
-    Chimu_10 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + -3 * 64));  \
-    Chimu_20 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + 0 * 64));  \
-    Chimu_01 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + -5 * 64));  \
-    Chimu_31 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + 4 * 64));  \
-    Chimu_11 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + -2 * 64));  \
-    Chimu_21 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + 1 * 64));  \
-    Chimu_02 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + -4 * 64));  \
-    Chimu_32 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + 5 * 64));  \
-    Chimu_12 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + -1 * 64));  \
-    Chimu_22 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + 2 * 64));  \
-}
-// LOAD_CHIMU_0213
-#define LOAD_CHIMU_0213_A64FXd  \
-{ \
-    const SiteSpinor & ref(in[offset]); \
-    Chimu_00 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + -6 * 64));  \
-    Chimu_20 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + 0 * 64));  \
-    Chimu_01 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + -5 * 64));  \
-    Chimu_21 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + 1 * 64));  \
-    Chimu_02 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + -4 * 64));  \
-    Chimu_22 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + 2 * 64));  \
-    Chimu_10 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + -3 * 64));  \
-    Chimu_30 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + 3 * 64));  \
-    Chimu_11 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + -2 * 64));  \
-    Chimu_31 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + 4 * 64));  \
-    Chimu_12 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + -1 * 64));  \
-    Chimu_32 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + 5 * 64));  \
-}
-// LOAD_CHIMU_0312
-#define LOAD_CHIMU_0312_A64FXd  \
-{ \
-    const SiteSpinor & ref(in[offset]); \
-    Chimu_00 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + -6 * 64));  \
-    Chimu_30 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + 3 * 64));  \
-    Chimu_01 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + -5 * 64));  \
-    Chimu_31 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + 4 * 64));  \
-    Chimu_02 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + -4 * 64));  \
-    Chimu_32 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + 5 * 64));  \
-    Chimu_10 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + -3 * 64));  \
-    Chimu_20 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + 0 * 64));  \
-    Chimu_11 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + -2 * 64));  \
-    Chimu_21 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + 1 * 64));  \
-    Chimu_12 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + -1 * 64));  \
-    Chimu_22 = svld1(pg1, (float64_t*)(base + 2 * 3 * 64 + 2 * 64));  \
-}
-// LOAD_TABLE0
-#define LOAD_TABLE0  \
-    table0 = svld1(pg1, (uint64_t*)&lut[0]);  
-
-// LOAD_TABLE1
-#define LOAD_TABLE1  \
-    table0 = svld1(pg1, (uint64_t*)&lut[1]);  
-
-// LOAD_TABLE2
-#define LOAD_TABLE2  \
-    table0 = svld1(pg1, (uint64_t*)&lut[2]);  
-
-// LOAD_TABLE3
-#define LOAD_TABLE3  \
-    table0 = svld1(pg1, (uint64_t*)&lut[3]);  
-
-// PERMUTE
-#define PERMUTE_A64FXd  \
-    Chi_00 = svtbl(Chi_00, table0);    \
-    Chi_01 = svtbl(Chi_01, table0);    \
-    Chi_02 = svtbl(Chi_02, table0);    \
-    Chi_10 = svtbl(Chi_10, table0);    \
-    Chi_11 = svtbl(Chi_11, table0);    \
-    Chi_12 = svtbl(Chi_12, table0);    
-
-// LOAD_GAUGE
-#define LOAD_GAUGE  \
-    const auto & ref(U[sU](A)); uint64_t baseU = (uint64_t)&ref; \
-{ \
-    U_00 = svld1(pg1, (float64_t*)(baseU + 2 * 3 * 64 + -6 * 64));  \
-    U_10 = svld1(pg1, (float64_t*)(baseU + 2 * 3 * 64 + -3 * 64));  \
-    U_20 = svld1(pg1, (float64_t*)(baseU + 2 * 3 * 64 + 0 * 64));  \
-    U_01 = svld1(pg1, (float64_t*)(baseU + 2 * 3 * 64 + -5 * 64));  \
-    U_11 = svld1(pg1, (float64_t*)(baseU + 2 * 3 * 64 + -2 * 64));  \
-    U_21 = svld1(pg1, (float64_t*)(baseU + 2 * 3 * 64 + 1 * 64));  \
-}
-// MULT_2SPIN
-#define MULT_2SPIN_1_A64FXd(A)  \
-{ \
-    const auto & ref(U[sU](A)); uint64_t baseU = (uint64_t)&ref; \
-    U_00 = svld1(pg1, (float64_t*)(baseU + 2 * 3 * 64 + -6 * 64));  \
-    U_10 = svld1(pg1, (float64_t*)(baseU + 2 * 3 * 64 + -3 * 64));  \
-    U_20 = svld1(pg1, (float64_t*)(baseU + 2 * 3 * 64 + 0 * 64));  \
-    U_01 = svld1(pg1, (float64_t*)(baseU + 2 * 3 * 64 + -5 * 64));  \
-    U_11 = svld1(pg1, (float64_t*)(baseU + 2 * 3 * 64 + -2 * 64));  \
-    U_21 = svld1(pg1, (float64_t*)(baseU + 2 * 3 * 64 + 1 * 64));  \
-    UChi_00 = svcmla_x(pg1, zero0, U_00, Chi_00, 0); \
-    UChi_10 = svcmla_x(pg1, zero0, U_00, Chi_10, 0); \
-    UChi_01 = svcmla_x(pg1, zero0, U_10, Chi_00, 0); \
-    UChi_11 = svcmla_x(pg1, zero0, U_10, Chi_10, 0); \
-    UChi_02 = svcmla_x(pg1, zero0, U_20, Chi_00, 0); \
-    UChi_12 = svcmla_x(pg1, zero0, U_20, Chi_10, 0); \
-    UChi_00 = svcmla_x(pg1, UChi_00, U_00, Chi_00, 90); \
-    UChi_10 = svcmla_x(pg1, UChi_10, U_00, Chi_10, 90); \
-    UChi_01 = svcmla_x(pg1, UChi_01, U_10, Chi_00, 90); \
-    UChi_11 = svcmla_x(pg1, UChi_11, U_10, Chi_10, 90); \
-    UChi_02 = svcmla_x(pg1, UChi_02, U_20, Chi_00, 90); \
-    UChi_12 = svcmla_x(pg1, UChi_12, U_20, Chi_10, 90); \
-    U_00 = svld1(pg1, (float64_t*)(baseU + 2 * 3 * 64 + -4 * 64));  \
-    U_10 = svld1(pg1, (float64_t*)(baseU + 2 * 3 * 64 + -1 * 64));  \
-    U_20 = svld1(pg1, (float64_t*)(baseU + 2 * 3 * 64 + 2 * 64));  \
-}
-// MULT_2SPIN_BACKEND
-#define MULT_2SPIN_2_A64FXd  \
-{ \
-    UChi_00 = svcmla_x(pg1, UChi_00, U_01, Chi_01, 0); \
-    UChi_10 = svcmla_x(pg1, UChi_10, U_01, Chi_11, 0); \
-    UChi_01 = svcmla_x(pg1, UChi_01, U_11, Chi_01, 0); \
-    UChi_11 = svcmla_x(pg1, UChi_11, U_11, Chi_11, 0); \
-    UChi_02 = svcmla_x(pg1, UChi_02, U_21, Chi_01, 0); \
-    UChi_12 = svcmla_x(pg1, UChi_12, U_21, Chi_11, 0); \
-    UChi_00 = svcmla_x(pg1, UChi_00, U_01, Chi_01, 90); \
-    UChi_10 = svcmla_x(pg1, UChi_10, U_01, Chi_11, 90); \
-    UChi_01 = svcmla_x(pg1, UChi_01, U_11, Chi_01, 90); \
-    UChi_11 = svcmla_x(pg1, UChi_11, U_11, Chi_11, 90); \
-    UChi_02 = svcmla_x(pg1, UChi_02, U_21, Chi_01, 90); \
-    UChi_12 = svcmla_x(pg1, UChi_12, U_21, Chi_11, 90); \
-    UChi_00 = svcmla_x(pg1, UChi_00, U_00, Chi_02, 0); \
-    UChi_10 = svcmla_x(pg1, UChi_10, U_00, Chi_12, 0); \
-    UChi_01 = svcmla_x(pg1, UChi_01, U_10, Chi_02, 0); \
-    UChi_11 = svcmla_x(pg1, UChi_11, U_10, Chi_12, 0); \
-    UChi_02 = svcmla_x(pg1, UChi_02, U_20, Chi_02, 0); \
-    UChi_12 = svcmla_x(pg1, UChi_12, U_20, Chi_12, 0); \
-    UChi_00 = svcmla_x(pg1, UChi_00, U_00, Chi_02, 90); \
-    UChi_10 = svcmla_x(pg1, UChi_10, U_00, Chi_12, 90); \
-    UChi_01 = svcmla_x(pg1, UChi_01, U_10, Chi_02, 90); \
-    UChi_11 = svcmla_x(pg1, UChi_11, U_10, Chi_12, 90); \
-    UChi_02 = svcmla_x(pg1, UChi_02, U_20, Chi_02, 90); \
-    UChi_12 = svcmla_x(pg1, UChi_12, U_20, Chi_12, 90); \
-}
-// XP_PROJ
-#define XP_PROJ_A64FXd  \
-{ \
-    Chi_00 = svcadd_x(pg1, Chimu_00, Chimu_30, 90);   \
-    Chi_01 = svcadd_x(pg1, Chimu_01, Chimu_31, 90);   \
-    Chi_02 = svcadd_x(pg1, Chimu_02, Chimu_32, 90);   \
-    Chi_10 = svcadd_x(pg1, Chimu_10, Chimu_20, 90);   \
-    Chi_11 = svcadd_x(pg1, Chimu_11, Chimu_21, 90);   \
-    Chi_12 = svcadd_x(pg1, Chimu_12, Chimu_22, 90);   \
-}
-// XP_RECON
-#define XP_RECON_A64FXd  \
-    result_20 = svcadd_x(pg1, zero0, UChi_10, 270);   \
-    result_21 = svcadd_x(pg1, zero0, UChi_11, 270);   \
-    result_22 = svcadd_x(pg1, zero0, UChi_12, 270);   \
-    result_30 = svcadd_x(pg1, zero0, UChi_00, 270);   \
-    result_31 = svcadd_x(pg1, zero0, UChi_01, 270);   \
-    result_32 = svcadd_x(pg1, zero0, UChi_02, 270);   \
-    result_00 = UChi_00;        \
-    result_01 = UChi_01;        \
-    result_02 = UChi_02;        \
-    result_10 = UChi_10;        \
-    result_11 = UChi_11;        \
-    result_12 = UChi_12;        
-
-// XP_RECON_ACCUM
-#define XP_RECON_ACCUM_A64FXd  \
-    result_30 = svcadd_x(pg1, result_30, UChi_00, 270);   \
-    result_00 = svadd_x(pg1, result_00, UChi_00); \
-    result_31 = svcadd_x(pg1, result_31, UChi_01, 270);   \
-    result_01 = svadd_x(pg1, result_01, UChi_01); \
-    result_32 = svcadd_x(pg1, result_32, UChi_02, 270);   \
-    result_02 = svadd_x(pg1, result_02, UChi_02); \
-    result_20 = svcadd_x(pg1, result_20, UChi_10, 270);   \
-    result_10 = svadd_x(pg1, result_10, UChi_10); \
-    result_21 = svcadd_x(pg1, result_21, UChi_11, 270);   \
-    result_11 = svadd_x(pg1, result_11, UChi_11); \
-    result_22 = svcadd_x(pg1, result_22, UChi_12, 270);   \
-    result_12 = svadd_x(pg1, result_12, UChi_12); 
-
-// YP_PROJ
-#define YP_PROJ_A64FXd  \
-{ \
-    Chi_00 = svsub_x(pg1, Chimu_00, Chimu_30);  \
-    Chi_01 = svsub_x(pg1, Chimu_01, Chimu_31);  \
-    Chi_02 = svsub_x(pg1, Chimu_02, Chimu_32);  \
-    Chi_10 = svadd_x(pg1, Chimu_10, Chimu_20);  \
-    Chi_11 = svadd_x(pg1, Chimu_11, Chimu_21);  \
-    Chi_12 = svadd_x(pg1, Chimu_12, Chimu_22);  \
-}
-// ZP_PROJ
-#define ZP_PROJ_A64FXd  \
-{ \
-    Chi_00 = svcadd_x(pg1, Chimu_00, Chimu_20, 90);   \
-    Chi_01 = svcadd_x(pg1, Chimu_01, Chimu_21, 90);   \
-    Chi_02 = svcadd_x(pg1, Chimu_02, Chimu_22, 90);   \
-    Chi_10 = svcadd_x(pg1, Chimu_10, Chimu_30, 270);   \
-    Chi_11 = svcadd_x(pg1, Chimu_11, Chimu_31, 270);   \
-    Chi_12 = svcadd_x(pg1, Chimu_12, Chimu_32, 270);   \
-}
-// TP_PROJ
-#define TP_PROJ_A64FXd  \
-{ \
-    Chi_00 = svadd_x(pg1, Chimu_00, Chimu_20);  \
-    Chi_01 = svadd_x(pg1, Chimu_01, Chimu_21);  \
-    Chi_02 = svadd_x(pg1, Chimu_02, Chimu_22);  \
-    Chi_10 = svadd_x(pg1, Chimu_10, Chimu_30);  \
-    Chi_11 = svadd_x(pg1, Chimu_11, Chimu_31);  \
-    Chi_12 = svadd_x(pg1, Chimu_12, Chimu_32);  \
-}
-// XM_PROJ
-#define XM_PROJ_A64FXd  \
-{ \
-    Chi_00 = svcadd_x(pg1, Chimu_00, Chimu_30, 270);   \
-    Chi_01 = svcadd_x(pg1, Chimu_01, Chimu_31, 270);   \
-    Chi_02 = svcadd_x(pg1, Chimu_02, Chimu_32, 270);   \
-    Chi_10 = svcadd_x(pg1, Chimu_10, Chimu_20, 270);   \
-    Chi_11 = svcadd_x(pg1, Chimu_11, Chimu_21, 270);   \
-    Chi_12 = svcadd_x(pg1, Chimu_12, Chimu_22, 270);   \
-}
-// XM_RECON
-#define XM_RECON_A64FXd  \
-    result_20 = svcadd_x(pg1, zero0, UChi_10, 90);   \
-    result_21 = svcadd_x(pg1, zero0, UChi_11, 90);   \
-    result_22 = svcadd_x(pg1, zero0, UChi_12, 90);   \
-    result_30 = svcadd_x(pg1, zero0, UChi_00, 90);   \
-    result_31 = svcadd_x(pg1, zero0, UChi_01, 90);   \
-    result_32 = svcadd_x(pg1, zero0, UChi_02, 90);   \
-    result_00 = UChi_00;        \
-    result_01 = UChi_01;        \
-    result_02 = UChi_02;        \
-    result_10 = UChi_10;        \
-    result_11 = UChi_11;        \
-    result_12 = UChi_12;        
-
-// YM_PROJ
-#define YM_PROJ_A64FXd  \
-{ \
-    Chi_00 = svadd_x(pg1, Chimu_00, Chimu_30);  \
-    Chi_01 = svadd_x(pg1, Chimu_01, Chimu_31);  \
-    Chi_02 = svadd_x(pg1, Chimu_02, Chimu_32);  \
-    Chi_10 = svsub_x(pg1, Chimu_10, Chimu_20);  \
-    Chi_11 = svsub_x(pg1, Chimu_11, Chimu_21);  \
-    Chi_12 = svsub_x(pg1, Chimu_12, Chimu_22);  \
-}
-// ZM_PROJ
-#define ZM_PROJ_A64FXd  \
-{ \
-    Chi_00 = svcadd_x(pg1, Chimu_00, Chimu_20, 270);   \
-    Chi_01 = svcadd_x(pg1, Chimu_01, Chimu_21, 270);   \
-    Chi_02 = svcadd_x(pg1, Chimu_02, Chimu_22, 270);   \
-    Chi_10 = svcadd_x(pg1, Chimu_10, Chimu_30, 90);   \
-    Chi_11 = svcadd_x(pg1, Chimu_11, Chimu_31, 90);   \
-    Chi_12 = svcadd_x(pg1, Chimu_12, Chimu_32, 90);   \
-}
-// TM_PROJ
-#define TM_PROJ_A64FXd  \
-{ \
-    Chi_00 = svsub_x(pg1, Chimu_00, Chimu_20);  \
-    Chi_01 = svsub_x(pg1, Chimu_01, Chimu_21);  \
-    Chi_02 = svsub_x(pg1, Chimu_02, Chimu_22);  \
-    Chi_10 = svsub_x(pg1, Chimu_10, Chimu_30);  \
-    Chi_11 = svsub_x(pg1, Chimu_11, Chimu_31);  \
-    Chi_12 = svsub_x(pg1, Chimu_12, Chimu_32);  \
-}
-// XM_RECON_ACCUM
-#define XM_RECON_ACCUM_A64FXd  \
-    result_30 = svcadd_x(pg1, result_30, UChi_00, 90);   \
-    result_31 = svcadd_x(pg1, result_31, UChi_01, 90);   \
-    result_32 = svcadd_x(pg1, result_32, UChi_02, 90);   \
-    result_20 = svcadd_x(pg1, result_20, UChi_10, 90);   \
-    result_21 = svcadd_x(pg1, result_21, UChi_11, 90);   \
-    result_22 = svcadd_x(pg1, result_22, UChi_12, 90);   \
-    result_00 = svadd_x(pg1, result_00, UChi_00); \
-    result_01 = svadd_x(pg1, result_01, UChi_01); \
-    result_02 = svadd_x(pg1, result_02, UChi_02); \
-    result_10 = svadd_x(pg1, result_10, UChi_10); \
-    result_11 = svadd_x(pg1, result_11, UChi_11); \
-    result_12 = svadd_x(pg1, result_12, UChi_12); 
-
-// YP_RECON_ACCUM
-#define YP_RECON_ACCUM_A64FXd  \
-    result_00 = svadd_x(pg1, result_00, UChi_00); \
-    result_30 = svsub_x(pg1, result_30, UChi_00); \
-    result_01 = svadd_x(pg1, result_01, UChi_01); \
-    result_31 = svsub_x(pg1, result_31, UChi_01); \
-    result_02 = svadd_x(pg1, result_02, UChi_02); \
-    result_32 = svsub_x(pg1, result_32, UChi_02); \
-    result_10 = svadd_x(pg1, result_10, UChi_10); \
-    result_20 = svadd_x(pg1, result_20, UChi_10); \
-    result_11 = svadd_x(pg1, result_11, UChi_11); \
-    result_21 = svadd_x(pg1, result_21, UChi_11); \
-    result_12 = svadd_x(pg1, result_12, UChi_12); \
-    result_22 = svadd_x(pg1, result_22, UChi_12); 
-
-// YM_RECON_ACCUM
-#define YM_RECON_ACCUM_A64FXd  \
-    result_00 = svadd_x(pg1, result_00, UChi_00); \
-    result_30 = svadd_x(pg1, result_30, UChi_00); \
-    result_01 = svadd_x(pg1, result_01, UChi_01); \
-    result_31 = svadd_x(pg1, result_31, UChi_01); \
-    result_02 = svadd_x(pg1, result_02, UChi_02); \
-    result_32 = svadd_x(pg1, result_32, UChi_02); \
-    result_10 = svadd_x(pg1, result_10, UChi_10); \
-    result_20 = svsub_x(pg1, result_20, UChi_10); \
-    result_11 = svadd_x(pg1, result_11, UChi_11); \
-    result_21 = svsub_x(pg1, result_21, UChi_11); \
-    result_12 = svadd_x(pg1, result_12, UChi_12); \
-    result_22 = svsub_x(pg1, result_22, UChi_12); 
-
-// ZP_RECON_ACCUM
-#define ZP_RECON_ACCUM_A64FXd  \
-    result_20 = svcadd_x(pg1, result_20, UChi_00, 270);   \
-    result_00 = svadd_x(pg1, result_00, UChi_00); \
-    result_21 = svcadd_x(pg1, result_21, UChi_01, 270);   \
-    result_01 = svadd_x(pg1, result_01, UChi_01); \
-    result_22 = svcadd_x(pg1, result_22, UChi_02, 270);   \
-    result_02 = svadd_x(pg1, result_02, UChi_02); \
-    result_30 = svcadd_x(pg1, result_30, UChi_10, 90);   \
-    result_10 = svadd_x(pg1, result_10, UChi_10); \
-    result_31 = svcadd_x(pg1, result_31, UChi_11, 90);   \
-    result_11 = svadd_x(pg1, result_11, UChi_11); \
-    result_32 = svcadd_x(pg1, result_32, UChi_12, 90);   \
-    result_12 = svadd_x(pg1, result_12, UChi_12); 
-
-// ZM_RECON_ACCUM
-#define ZM_RECON_ACCUM_A64FXd  \
-    result_20 = svcadd_x(pg1, result_20, UChi_00, 90);   \
-    result_00 = svadd_x(pg1, result_00, UChi_00); \
-    result_21 = svcadd_x(pg1, result_21, UChi_01, 90);   \
-    result_01 = svadd_x(pg1, result_01, UChi_01); \
-    result_22 = svcadd_x(pg1, result_22, UChi_02, 90);   \
-    result_02 = svadd_x(pg1, result_02, UChi_02); \
-    result_30 = svcadd_x(pg1, result_30, UChi_10, 270);   \
-    result_10 = svadd_x(pg1, result_10, UChi_10); \
-    result_31 = svcadd_x(pg1, result_31, UChi_11, 270);   \
-    result_11 = svadd_x(pg1, result_11, UChi_11); \
-    result_32 = svcadd_x(pg1, result_32, UChi_12, 270);   \
-    result_12 = svadd_x(pg1, result_12, UChi_12); 
-
-// TP_RECON_ACCUM
-#define TP_RECON_ACCUM_A64FXd  \
-    result_00 = svadd_x(pg1, result_00, UChi_00); \
-    result_20 = svadd_x(pg1, result_20, UChi_00); \
-    result_01 = svadd_x(pg1, result_01, UChi_01); \
-    result_21 = svadd_x(pg1, result_21, UChi_01); \
-    result_02 = svadd_x(pg1, result_02, UChi_02); \
-    result_22 = svadd_x(pg1, result_22, UChi_02); \
-    result_10 = svadd_x(pg1, result_10, UChi_10); \
-    result_30 = svadd_x(pg1, result_30, UChi_10); \
-    result_11 = svadd_x(pg1, result_11, UChi_11); \
-    result_31 = svadd_x(pg1, result_31, UChi_11); \
-    result_12 = svadd_x(pg1, result_12, UChi_12); \
-    result_32 = svadd_x(pg1, result_32, UChi_12); 
-
-// TM_RECON_ACCUM
-#define TM_RECON_ACCUM_A64FXd  \
-    result_00 = svadd_x(pg1, result_00, UChi_00); \
-    result_20 = svsub_x(pg1, result_20, UChi_00); \
-    result_01 = svadd_x(pg1, result_01, UChi_01); \
-    result_21 = svsub_x(pg1, result_21, UChi_01); \
-    result_02 = svadd_x(pg1, result_02, UChi_02); \
-    result_22 = svsub_x(pg1, result_22, UChi_02); \
-    result_10 = svadd_x(pg1, result_10, UChi_10); \
-    result_30 = svsub_x(pg1, result_30, UChi_10); \
-    result_11 = svadd_x(pg1, result_11, UChi_11); \
-    result_31 = svsub_x(pg1, result_31, UChi_11); \
-    result_12 = svadd_x(pg1, result_12, UChi_12); \
-    result_32 = svsub_x(pg1, result_32, UChi_12); 
-
-// ZERO_PSI
-#define ZERO_PSI_A64FXd  \
-    result_00 = svdup_f64(0.); \
-    result_01 = svdup_f64(0.); \
-    result_02 = svdup_f64(0.); \
-    result_10 = svdup_f64(0.); \
-    result_11 = svdup_f64(0.); \
-    result_12 = svdup_f64(0.); \
-    result_20 = svdup_f64(0.); \
-    result_21 = svdup_f64(0.); \
-    result_22 = svdup_f64(0.); \
-    result_30 = svdup_f64(0.); \
-    result_31 = svdup_f64(0.); \
-    result_32 = svdup_f64(0.); 
-
-// PREFETCH_RESULT_L2_STORE (prefetch store to L2)
-#define PREFETCH_RESULT_L2_STORE_INTERNAL_A64FXd(base)  \
-{ \
-    svprfd(pg1, (int64_t*)(base + 0), SV_PSTL2STRM); \
-    svprfd(pg1, (int64_t*)(base + 256), SV_PSTL2STRM); \
-    svprfd(pg1, (int64_t*)(base + 512), SV_PSTL2STRM); \
-}
-// PREFETCH_RESULT_L1_STORE (prefetch store to L1)
-#define PREFETCH_RESULT_L1_STORE_INTERNAL_A64FXd(base)  \
-{ \
-    svprfd(pg1, (int64_t*)(base + 0), SV_PSTL1STRM); \
-    svprfd(pg1, (int64_t*)(base + 256), SV_PSTL1STRM); \
-    svprfd(pg1, (int64_t*)(base + 512), SV_PSTL1STRM); \
-}
-// ADD_RESULT_INTERNAL
-#define ADD_RESULT_INTERNAL_A64FXd  \
-    result_00 = svadd_x(pg1, result_00, Chimu_00); \
-    result_01 = svadd_x(pg1, result_01, Chimu_01); \
-    result_02 = svadd_x(pg1, result_02, Chimu_02); \
-    result_10 = svadd_x(pg1, result_10, Chimu_10); \
-    result_11 = svadd_x(pg1, result_11, Chimu_11); \
-    result_12 = svadd_x(pg1, result_12, Chimu_12); \
-    result_20 = svadd_x(pg1, result_20, Chimu_20); \
-    result_21 = svadd_x(pg1, result_21, Chimu_21); \
-    result_22 = svadd_x(pg1, result_22, Chimu_22); \
-    result_30 = svadd_x(pg1, result_30, Chimu_30); \
-    result_31 = svadd_x(pg1, result_31, Chimu_31); \
-    result_32 = svadd_x(pg1, result_32, Chimu_32); 
-

Grid/simd/Fujitsu_A64FX_intrin_single.h

@@ -1,601 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: Fujitsu_A64FX_intrin_single.h
-
-    Copyright (C) 2020
-
-Author: Nils Meyer <nils.meyer@ur.de>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-#define LOAD_CHIMU(base)               LOAD_CHIMU_INTERLEAVED_A64FXf(base)  
-#define PREFETCH_CHIMU_L1(A)           PREFETCH_CHIMU_L1_INTERNAL_A64FXf(A)  
-#define PREFETCH_GAUGE_L1(A)           PREFETCH_GAUGE_L1_INTERNAL_A64FXf(A)  
-#define PREFETCH_CHIMU_L2(A)           PREFETCH_CHIMU_L2_INTERNAL_A64FXf(A)  
-#define PREFETCH_GAUGE_L2(A)           PREFETCH_GAUGE_L2_INTERNAL_A64FXf(A)  
-#define PF_GAUGE(A)  
-#define PREFETCH_RESULT_L2_STORE(A)    PREFETCH_RESULT_L2_STORE_INTERNAL_A64FXf(A)  
-#define PREFETCH_RESULT_L1_STORE(A)    PREFETCH_RESULT_L1_STORE_INTERNAL_A64FXf(A)  
-#define PREFETCH1_CHIMU(A)             PREFETCH_CHIMU_L1(A)  
-#define PREFETCH_CHIMU(A)              PREFETCH_CHIMU_L1(A)  
-#define LOCK_GAUGE(A)  
-#define UNLOCK_GAUGE(A)  
-#define MASK_REGS                      DECLARATIONS_A64FXf  
-#define SAVE_RESULT(A,B)               RESULT_A64FXf(A); PREFETCH_RESULT_L2_STORE(B)  
-#define MULT_2SPIN_1(Dir)              MULT_2SPIN_1_A64FXf(Dir)  
-#define MULT_2SPIN_2                   MULT_2SPIN_2_A64FXf  
-#define LOAD_CHI(base)                 LOAD_CHI_A64FXf(base)  
-#define ADD_RESULT(base,basep)         LOAD_CHIMU(base); ADD_RESULT_INTERNAL_A64FXf; RESULT_A64FXf(base)  
-#define XP_PROJ                        XP_PROJ_A64FXf  
-#define YP_PROJ                        YP_PROJ_A64FXf  
-#define ZP_PROJ                        ZP_PROJ_A64FXf  
-#define TP_PROJ                        TP_PROJ_A64FXf  
-#define XM_PROJ                        XM_PROJ_A64FXf  
-#define YM_PROJ                        YM_PROJ_A64FXf  
-#define ZM_PROJ                        ZM_PROJ_A64FXf  
-#define TM_PROJ                        TM_PROJ_A64FXf  
-#define XP_RECON                       XP_RECON_A64FXf  
-#define XM_RECON                       XM_RECON_A64FXf  
-#define XM_RECON_ACCUM                 XM_RECON_ACCUM_A64FXf  
-#define YM_RECON_ACCUM                 YM_RECON_ACCUM_A64FXf  
-#define ZM_RECON_ACCUM                 ZM_RECON_ACCUM_A64FXf  
-#define TM_RECON_ACCUM                 TM_RECON_ACCUM_A64FXf  
-#define XP_RECON_ACCUM                 XP_RECON_ACCUM_A64FXf  
-#define YP_RECON_ACCUM                 YP_RECON_ACCUM_A64FXf  
-#define ZP_RECON_ACCUM                 ZP_RECON_ACCUM_A64FXf  
-#define TP_RECON_ACCUM                 TP_RECON_ACCUM_A64FXf  
-#define PERMUTE_DIR0                   0  
-#define PERMUTE_DIR1                   1  
-#define PERMUTE_DIR2                   2  
-#define PERMUTE_DIR3                   3  
-#define PERMUTE                        PERMUTE_A64FXf;  
-#define LOAD_TABLE(Dir)                if (Dir == 0) { LOAD_TABLE0; } else if (Dir == 1) { LOAD_TABLE1 } else if (Dir == 2) { LOAD_TABLE2; } else if (Dir == 3) { LOAD_TABLE3; }  
-#define MAYBEPERM(A,perm)              if (perm) { PERMUTE; }  
-// DECLARATIONS
-#define DECLARATIONS_A64FXf  \
-    const uint32_t lut[4][16] = { \
-        {8, 9, 10, 11, 12, 13, 14, 15, 0, 1, 2, 3, 4, 5, 6, 7}, \
-        {4, 5, 6, 7, 0, 1, 2, 3, 12, 13, 14, 15, 8, 9, 10, 11}, \
-        {2, 3, 0, 1, 6, 7, 4, 5, 10, 11, 8, 9, 14, 15, 12, 13}, \
-        {1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14} }; \
-    svfloat32_t result_00;        \
-    svfloat32_t result_01;        \
-    svfloat32_t result_02;        \
-    svfloat32_t result_10;        \
-    svfloat32_t result_11;        \
-    svfloat32_t result_12;        \
-    svfloat32_t result_20;        \
-    svfloat32_t result_21;        \
-    svfloat32_t result_22;        \
-    svfloat32_t result_30;        \
-    svfloat32_t result_31;        \
-    svfloat32_t result_32;        \
-    svfloat32_t Chi_00;        \
-    svfloat32_t Chi_01;        \
-    svfloat32_t Chi_02;        \
-    svfloat32_t Chi_10;        \
-    svfloat32_t Chi_11;        \
-    svfloat32_t Chi_12;        \
-    svfloat32_t UChi_00;        \
-    svfloat32_t UChi_01;        \
-    svfloat32_t UChi_02;        \
-    svfloat32_t UChi_10;        \
-    svfloat32_t UChi_11;        \
-    svfloat32_t UChi_12;        \
-    svfloat32_t U_00;        \
-    svfloat32_t U_10;        \
-    svfloat32_t U_20;        \
-    svfloat32_t U_01;        \
-    svfloat32_t U_11;        \
-    svfloat32_t U_21;        \
-    svbool_t pg1;        \
-    pg1 = svptrue_b32();        \
-    svuint32_t table0; \
-    svfloat32_t zero0;        \
-    zero0 = svdup_f32(0.); 
-
-#define Chimu_00 Chi_00  
-#define Chimu_01 Chi_01  
-#define Chimu_02 Chi_02  
-#define Chimu_10 Chi_10  
-#define Chimu_11 Chi_11  
-#define Chimu_12 Chi_12  
-#define Chimu_20 UChi_00  
-#define Chimu_21 UChi_01  
-#define Chimu_22 UChi_02  
-#define Chimu_30 UChi_10  
-#define Chimu_31 UChi_11  
-#define Chimu_32 UChi_12  
-// RESULT
-#define RESULT_A64FXf(base)  \
-{ \
-    svst1(pg1, (float32_t*)(base + 2 * 3 * 64 + -6 * 64), result_00);  \
-    svst1(pg1, (float32_t*)(base + 2 * 3 * 64 + -5 * 64), result_01);  \
-    svst1(pg1, (float32_t*)(base + 2 * 3 * 64 + -4 * 64), result_02);  \
-    svst1(pg1, (float32_t*)(base + 2 * 3 * 64 + -3 * 64), result_10);  \
-    svst1(pg1, (float32_t*)(base + 2 * 3 * 64 + -2 * 64), result_11);  \
-    svst1(pg1, (float32_t*)(base + 2 * 3 * 64 + -1 * 64), result_12);  \
-    svst1(pg1, (float32_t*)(base + 2 * 3 * 64 + 0 * 64), result_20);  \
-    svst1(pg1, (float32_t*)(base + 2 * 3 * 64 + 1 * 64), result_21);  \
-    svst1(pg1, (float32_t*)(base + 2 * 3 * 64 + 2 * 64), result_22);  \
-    svst1(pg1, (float32_t*)(base + 2 * 3 * 64 + 3 * 64), result_30);  \
-    svst1(pg1, (float32_t*)(base + 2 * 3 * 64 + 4 * 64), result_31);  \
-    svst1(pg1, (float32_t*)(base + 2 * 3 * 64 + 5 * 64), result_32);  \
-}
-// PREFETCH_CHIMU_L2 (prefetch to L2)
-#define PREFETCH_CHIMU_L2_INTERNAL_A64FXf(base)  \
-{ \
-    svprfd(pg1, (int64_t*)(base + 0), SV_PLDL2STRM); \
-    svprfd(pg1, (int64_t*)(base + 256), SV_PLDL2STRM); \
-    svprfd(pg1, (int64_t*)(base + 512), SV_PLDL2STRM); \
-}
-// PREFETCH_CHIMU_L1 (prefetch to L1)
-#define PREFETCH_CHIMU_L1_INTERNAL_A64FXf(base)  \
-{ \
-    svprfd(pg1, (int64_t*)(base + 0), SV_PLDL1STRM); \
-    svprfd(pg1, (int64_t*)(base + 256), SV_PLDL1STRM); \
-    svprfd(pg1, (int64_t*)(base + 512), SV_PLDL1STRM); \
-}
-// PREFETCH_GAUGE_L2 (prefetch to L2)
-#define PREFETCH_GAUGE_L2_INTERNAL_A64FXf(A)  \
-{ \
-    const auto & ref(U[sUn](A)); uint64_t baseU = (uint64_t)&ref + 3 * 3 * 64; \
-    svprfd(pg1, (int64_t*)(baseU + -256), SV_PLDL2STRM); \
-    svprfd(pg1, (int64_t*)(baseU + 0), SV_PLDL2STRM); \
-    svprfd(pg1, (int64_t*)(baseU + 256), SV_PLDL2STRM); \
-    svprfd(pg1, (int64_t*)(baseU + 512), SV_PLDL2STRM); \
-    svprfd(pg1, (int64_t*)(baseU + 768), SV_PLDL2STRM); \
-    svprfd(pg1, (int64_t*)(baseU + 1024), SV_PLDL2STRM); \
-    svprfd(pg1, (int64_t*)(baseU + 1280), SV_PLDL2STRM); \
-    svprfd(pg1, (int64_t*)(baseU + 1536), SV_PLDL2STRM); \
-    svprfd(pg1, (int64_t*)(baseU + 1792), SV_PLDL2STRM); \
-}
-// PREFETCH_GAUGE_L1 (prefetch to L1)
-#define PREFETCH_GAUGE_L1_INTERNAL_A64FXf(A)  \
-{ \
-    const auto & ref(U[sU](A)); uint64_t baseU = (uint64_t)&ref; \
-    svprfd(pg1, (int64_t*)(baseU + 0), SV_PLDL1STRM); \
-    svprfd(pg1, (int64_t*)(baseU + 256), SV_PLDL1STRM); \
-    svprfd(pg1, (int64_t*)(baseU + 512), SV_PLDL1STRM); \
-}
-// LOAD_CHI
-#define LOAD_CHI_A64FXf(base)  \
-{ \
-    Chi_00 = svld1(pg1, (float32_t*)(base + 0 * 64));  \
-    Chi_01 = svld1(pg1, (float32_t*)(base + 1 * 64));  \
-    Chi_02 = svld1(pg1, (float32_t*)(base + 2 * 64));  \
-    Chi_10 = svld1(pg1, (float32_t*)(base + 3 * 64));  \
-    Chi_11 = svld1(pg1, (float32_t*)(base + 4 * 64));  \
-    Chi_12 = svld1(pg1, (float32_t*)(base + 5 * 64));  \
-}
-// LOAD_CHIMU
-#define LOAD_CHIMU_INTERLEAVED_A64FXf(base)  \
-{ \
-    Chimu_00 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + -6 * 64));  \
-    Chimu_30 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + 3 * 64));  \
-    Chimu_10 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + -3 * 64));  \
-    Chimu_20 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + 0 * 64));  \
-    Chimu_01 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + -5 * 64));  \
-    Chimu_31 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + 4 * 64));  \
-    Chimu_11 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + -2 * 64));  \
-    Chimu_21 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + 1 * 64));  \
-    Chimu_02 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + -4 * 64));  \
-    Chimu_32 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + 5 * 64));  \
-    Chimu_12 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + -1 * 64));  \
-    Chimu_22 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + 2 * 64));  \
-}
-// LOAD_CHIMU_0213
-#define LOAD_CHIMU_0213_A64FXf  \
-{ \
-    const SiteSpinor & ref(in[offset]); \
-    Chimu_00 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + -6 * 64));  \
-    Chimu_20 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + 0 * 64));  \
-    Chimu_01 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + -5 * 64));  \
-    Chimu_21 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + 1 * 64));  \
-    Chimu_02 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + -4 * 64));  \
-    Chimu_22 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + 2 * 64));  \
-    Chimu_10 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + -3 * 64));  \
-    Chimu_30 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + 3 * 64));  \
-    Chimu_11 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + -2 * 64));  \
-    Chimu_31 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + 4 * 64));  \
-    Chimu_12 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + -1 * 64));  \
-    Chimu_32 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + 5 * 64));  \
-}
-// LOAD_CHIMU_0312
-#define LOAD_CHIMU_0312_A64FXf  \
-{ \
-    const SiteSpinor & ref(in[offset]); \
-    Chimu_00 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + -6 * 64));  \
-    Chimu_30 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + 3 * 64));  \
-    Chimu_01 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + -5 * 64));  \
-    Chimu_31 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + 4 * 64));  \
-    Chimu_02 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + -4 * 64));  \
-    Chimu_32 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + 5 * 64));  \
-    Chimu_10 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + -3 * 64));  \
-    Chimu_20 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + 0 * 64));  \
-    Chimu_11 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + -2 * 64));  \
-    Chimu_21 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + 1 * 64));  \
-    Chimu_12 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + -1 * 64));  \
-    Chimu_22 = svld1(pg1, (float32_t*)(base + 2 * 3 * 64 + 2 * 64));  \
-}
-// LOAD_TABLE0
-#define LOAD_TABLE0  \
-    table0 = svld1(pg1, (uint32_t*)&lut[0]);  
-
-// LOAD_TABLE1
-#define LOAD_TABLE1  \
-    table0 = svld1(pg1, (uint32_t*)&lut[1]);  
-
-// LOAD_TABLE2
-#define LOAD_TABLE2  \
-    table0 = svld1(pg1, (uint32_t*)&lut[2]);  
-
-// LOAD_TABLE3
-#define LOAD_TABLE3  \
-    table0 = svld1(pg1, (uint32_t*)&lut[3]);  
-
-// PERMUTE
-#define PERMUTE_A64FXf  \
-    Chi_00 = svtbl(Chi_00, table0);    \
-    Chi_01 = svtbl(Chi_01, table0);    \
-    Chi_02 = svtbl(Chi_02, table0);    \
-    Chi_10 = svtbl(Chi_10, table0);    \
-    Chi_11 = svtbl(Chi_11, table0);    \
-    Chi_12 = svtbl(Chi_12, table0);    
-
-// LOAD_GAUGE
-#define LOAD_GAUGE  \
-    const auto & ref(U[sU](A)); uint64_t baseU = (uint64_t)&ref; \
-{ \
-    U_00 = svld1(pg1, (float32_t*)(baseU + 2 * 3 * 64 + -6 * 64));  \
-    U_10 = svld1(pg1, (float32_t*)(baseU + 2 * 3 * 64 + -3 * 64));  \
-    U_20 = svld1(pg1, (float32_t*)(baseU + 2 * 3 * 64 + 0 * 64));  \
-    U_01 = svld1(pg1, (float32_t*)(baseU + 2 * 3 * 64 + -5 * 64));  \
-    U_11 = svld1(pg1, (float32_t*)(baseU + 2 * 3 * 64 + -2 * 64));  \
-    U_21 = svld1(pg1, (float32_t*)(baseU + 2 * 3 * 64 + 1 * 64));  \
-}
-// MULT_2SPIN
-#define MULT_2SPIN_1_A64FXf(A)  \
-{ \
-    const auto & ref(U[sU](A)); uint64_t baseU = (uint64_t)&ref; \
-    U_00 = svld1(pg1, (float32_t*)(baseU + 2 * 3 * 64 + -6 * 64));  \
-    U_10 = svld1(pg1, (float32_t*)(baseU + 2 * 3 * 64 + -3 * 64));  \
-    U_20 = svld1(pg1, (float32_t*)(baseU + 2 * 3 * 64 + 0 * 64));  \
-    U_01 = svld1(pg1, (float32_t*)(baseU + 2 * 3 * 64 + -5 * 64));  \
-    U_11 = svld1(pg1, (float32_t*)(baseU + 2 * 3 * 64 + -2 * 64));  \
-    U_21 = svld1(pg1, (float32_t*)(baseU + 2 * 3 * 64 + 1 * 64));  \
-    UChi_00 = svcmla_x(pg1, zero0, U_00, Chi_00, 0); \
-    UChi_10 = svcmla_x(pg1, zero0, U_00, Chi_10, 0); \
-    UChi_01 = svcmla_x(pg1, zero0, U_10, Chi_00, 0); \
-    UChi_11 = svcmla_x(pg1, zero0, U_10, Chi_10, 0); \
-    UChi_02 = svcmla_x(pg1, zero0, U_20, Chi_00, 0); \
-    UChi_12 = svcmla_x(pg1, zero0, U_20, Chi_10, 0); \
-    UChi_00 = svcmla_x(pg1, UChi_00, U_00, Chi_00, 90); \
-    UChi_10 = svcmla_x(pg1, UChi_10, U_00, Chi_10, 90); \
-    UChi_01 = svcmla_x(pg1, UChi_01, U_10, Chi_00, 90); \
-    UChi_11 = svcmla_x(pg1, UChi_11, U_10, Chi_10, 90); \
-    UChi_02 = svcmla_x(pg1, UChi_02, U_20, Chi_00, 90); \
-    UChi_12 = svcmla_x(pg1, UChi_12, U_20, Chi_10, 90); \
-    U_00 = svld1(pg1, (float32_t*)(baseU + 2 * 3 * 64 + -4 * 64));  \
-    U_10 = svld1(pg1, (float32_t*)(baseU + 2 * 3 * 64 + -1 * 64));  \
-    U_20 = svld1(pg1, (float32_t*)(baseU + 2 * 3 * 64 + 2 * 64));  \
-}
-// MULT_2SPIN_BACKEND
-#define MULT_2SPIN_2_A64FXf  \
-{ \
-    UChi_00 = svcmla_x(pg1, UChi_00, U_01, Chi_01, 0); \
-    UChi_10 = svcmla_x(pg1, UChi_10, U_01, Chi_11, 0); \
-    UChi_01 = svcmla_x(pg1, UChi_01, U_11, Chi_01, 0); \
-    UChi_11 = svcmla_x(pg1, UChi_11, U_11, Chi_11, 0); \
-    UChi_02 = svcmla_x(pg1, UChi_02, U_21, Chi_01, 0); \
-    UChi_12 = svcmla_x(pg1, UChi_12, U_21, Chi_11, 0); \
-    UChi_00 = svcmla_x(pg1, UChi_00, U_01, Chi_01, 90); \
-    UChi_10 = svcmla_x(pg1, UChi_10, U_01, Chi_11, 90); \
-    UChi_01 = svcmla_x(pg1, UChi_01, U_11, Chi_01, 90); \
-    UChi_11 = svcmla_x(pg1, UChi_11, U_11, Chi_11, 90); \
-    UChi_02 = svcmla_x(pg1, UChi_02, U_21, Chi_01, 90); \
-    UChi_12 = svcmla_x(pg1, UChi_12, U_21, Chi_11, 90); \
-    UChi_00 = svcmla_x(pg1, UChi_00, U_00, Chi_02, 0); \
-    UChi_10 = svcmla_x(pg1, UChi_10, U_00, Chi_12, 0); \
-    UChi_01 = svcmla_x(pg1, UChi_01, U_10, Chi_02, 0); \
-    UChi_11 = svcmla_x(pg1, UChi_11, U_10, Chi_12, 0); \
-    UChi_02 = svcmla_x(pg1, UChi_02, U_20, Chi_02, 0); \
-    UChi_12 = svcmla_x(pg1, UChi_12, U_20, Chi_12, 0); \
-    UChi_00 = svcmla_x(pg1, UChi_00, U_00, Chi_02, 90); \
-    UChi_10 = svcmla_x(pg1, UChi_10, U_00, Chi_12, 90); \
-    UChi_01 = svcmla_x(pg1, UChi_01, U_10, Chi_02, 90); \
-    UChi_11 = svcmla_x(pg1, UChi_11, U_10, Chi_12, 90); \
-    UChi_02 = svcmla_x(pg1, UChi_02, U_20, Chi_02, 90); \
-    UChi_12 = svcmla_x(pg1, UChi_12, U_20, Chi_12, 90); \
-}
-// XP_PROJ
-#define XP_PROJ_A64FXf  \
-{ \
-    Chi_00 = svcadd_x(pg1, Chimu_00, Chimu_30, 90);   \
-    Chi_01 = svcadd_x(pg1, Chimu_01, Chimu_31, 90);   \
-    Chi_02 = svcadd_x(pg1, Chimu_02, Chimu_32, 90);   \
-    Chi_10 = svcadd_x(pg1, Chimu_10, Chimu_20, 90);   \
-    Chi_11 = svcadd_x(pg1, Chimu_11, Chimu_21, 90);   \
-    Chi_12 = svcadd_x(pg1, Chimu_12, Chimu_22, 90);   \
-}
-// XP_RECON
-#define XP_RECON_A64FXf  \
-    result_20 = svcadd_x(pg1, zero0, UChi_10, 270);   \
-    result_21 = svcadd_x(pg1, zero0, UChi_11, 270);   \
-    result_22 = svcadd_x(pg1, zero0, UChi_12, 270);   \
-    result_30 = svcadd_x(pg1, zero0, UChi_00, 270);   \
-    result_31 = svcadd_x(pg1, zero0, UChi_01, 270);   \
-    result_32 = svcadd_x(pg1, zero0, UChi_02, 270);   \
-    result_00 = UChi_00;        \
-    result_01 = UChi_01;        \
-    result_02 = UChi_02;        \
-    result_10 = UChi_10;        \
-    result_11 = UChi_11;        \
-    result_12 = UChi_12;        
-
-// XP_RECON_ACCUM
-#define XP_RECON_ACCUM_A64FXf  \
-    result_30 = svcadd_x(pg1, result_30, UChi_00, 270);   \
-    result_00 = svadd_x(pg1, result_00, UChi_00); \
-    result_31 = svcadd_x(pg1, result_31, UChi_01, 270);   \
-    result_01 = svadd_x(pg1, result_01, UChi_01); \
-    result_32 = svcadd_x(pg1, result_32, UChi_02, 270);   \
-    result_02 = svadd_x(pg1, result_02, UChi_02); \
-    result_20 = svcadd_x(pg1, result_20, UChi_10, 270);   \
-    result_10 = svadd_x(pg1, result_10, UChi_10); \
-    result_21 = svcadd_x(pg1, result_21, UChi_11, 270);   \
-    result_11 = svadd_x(pg1, result_11, UChi_11); \
-    result_22 = svcadd_x(pg1, result_22, UChi_12, 270);   \
-    result_12 = svadd_x(pg1, result_12, UChi_12); 
-
-// YP_PROJ
-#define YP_PROJ_A64FXf  \
-{ \
-    Chi_00 = svsub_x(pg1, Chimu_00, Chimu_30);  \
-    Chi_01 = svsub_x(pg1, Chimu_01, Chimu_31);  \
-    Chi_02 = svsub_x(pg1, Chimu_02, Chimu_32);  \
-    Chi_10 = svadd_x(pg1, Chimu_10, Chimu_20);  \
-    Chi_11 = svadd_x(pg1, Chimu_11, Chimu_21);  \
-    Chi_12 = svadd_x(pg1, Chimu_12, Chimu_22);  \
-}
-// ZP_PROJ
-#define ZP_PROJ_A64FXf  \
-{ \
-    Chi_00 = svcadd_x(pg1, Chimu_00, Chimu_20, 90);   \
-    Chi_01 = svcadd_x(pg1, Chimu_01, Chimu_21, 90);   \
-    Chi_02 = svcadd_x(pg1, Chimu_02, Chimu_22, 90);   \
-    Chi_10 = svcadd_x(pg1, Chimu_10, Chimu_30, 270);   \
-    Chi_11 = svcadd_x(pg1, Chimu_11, Chimu_31, 270);   \
-    Chi_12 = svcadd_x(pg1, Chimu_12, Chimu_32, 270);   \
-}
-// TP_PROJ
-#define TP_PROJ_A64FXf  \
-{ \
-    Chi_00 = svadd_x(pg1, Chimu_00, Chimu_20);  \
-    Chi_01 = svadd_x(pg1, Chimu_01, Chimu_21);  \
-    Chi_02 = svadd_x(pg1, Chimu_02, Chimu_22);  \
-    Chi_10 = svadd_x(pg1, Chimu_10, Chimu_30);  \
-    Chi_11 = svadd_x(pg1, Chimu_11, Chimu_31);  \
-    Chi_12 = svadd_x(pg1, Chimu_12, Chimu_32);  \
-}
-// XM_PROJ
-#define XM_PROJ_A64FXf  \
-{ \
-    Chi_00 = svcadd_x(pg1, Chimu_00, Chimu_30, 270);   \
-    Chi_01 = svcadd_x(pg1, Chimu_01, Chimu_31, 270);   \
-    Chi_02 = svcadd_x(pg1, Chimu_02, Chimu_32, 270);   \
-    Chi_10 = svcadd_x(pg1, Chimu_10, Chimu_20, 270);   \
-    Chi_11 = svcadd_x(pg1, Chimu_11, Chimu_21, 270);   \
-    Chi_12 = svcadd_x(pg1, Chimu_12, Chimu_22, 270);   \
-}
-// XM_RECON
-#define XM_RECON_A64FXf  \
-    result_20 = svcadd_x(pg1, zero0, UChi_10, 90);   \
-    result_21 = svcadd_x(pg1, zero0, UChi_11, 90);   \
-    result_22 = svcadd_x(pg1, zero0, UChi_12, 90);   \
-    result_30 = svcadd_x(pg1, zero0, UChi_00, 90);   \
-    result_31 = svcadd_x(pg1, zero0, UChi_01, 90);   \
-    result_32 = svcadd_x(pg1, zero0, UChi_02, 90);   \
-    result_00 = UChi_00;        \
-    result_01 = UChi_01;        \
-    result_02 = UChi_02;        \
-    result_10 = UChi_10;        \
-    result_11 = UChi_11;        \
-    result_12 = UChi_12;        
-
-// YM_PROJ
-#define YM_PROJ_A64FXf  \
-{ \
-    Chi_00 = svadd_x(pg1, Chimu_00, Chimu_30);  \
-    Chi_01 = svadd_x(pg1, Chimu_01, Chimu_31);  \
-    Chi_02 = svadd_x(pg1, Chimu_02, Chimu_32);  \
-    Chi_10 = svsub_x(pg1, Chimu_10, Chimu_20);  \
-    Chi_11 = svsub_x(pg1, Chimu_11, Chimu_21);  \
-    Chi_12 = svsub_x(pg1, Chimu_12, Chimu_22);  \
-}
-// ZM_PROJ
-#define ZM_PROJ_A64FXf  \
-{ \
-    Chi_00 = svcadd_x(pg1, Chimu_00, Chimu_20, 270);   \
-    Chi_01 = svcadd_x(pg1, Chimu_01, Chimu_21, 270);   \
-    Chi_02 = svcadd_x(pg1, Chimu_02, Chimu_22, 270);   \
-    Chi_10 = svcadd_x(pg1, Chimu_10, Chimu_30, 90);   \
-    Chi_11 = svcadd_x(pg1, Chimu_11, Chimu_31, 90);   \
-    Chi_12 = svcadd_x(pg1, Chimu_12, Chimu_32, 90);   \
-}
-// TM_PROJ
-#define TM_PROJ_A64FXf  \
-{ \
-    Chi_00 = svsub_x(pg1, Chimu_00, Chimu_20);  \
-    Chi_01 = svsub_x(pg1, Chimu_01, Chimu_21);  \
-    Chi_02 = svsub_x(pg1, Chimu_02, Chimu_22);  \
-    Chi_10 = svsub_x(pg1, Chimu_10, Chimu_30);  \
-    Chi_11 = svsub_x(pg1, Chimu_11, Chimu_31);  \
-    Chi_12 = svsub_x(pg1, Chimu_12, Chimu_32);  \
-}
-// XM_RECON_ACCUM
-#define XM_RECON_ACCUM_A64FXf  \
-    result_30 = svcadd_x(pg1, result_30, UChi_00, 90);   \
-    result_31 = svcadd_x(pg1, result_31, UChi_01, 90);   \
-    result_32 = svcadd_x(pg1, result_32, UChi_02, 90);   \
-    result_20 = svcadd_x(pg1, result_20, UChi_10, 90);   \
-    result_21 = svcadd_x(pg1, result_21, UChi_11, 90);   \
-    result_22 = svcadd_x(pg1, result_22, UChi_12, 90);   \
-    result_00 = svadd_x(pg1, result_00, UChi_00); \
-    result_01 = svadd_x(pg1, result_01, UChi_01); \
-    result_02 = svadd_x(pg1, result_02, UChi_02); \
-    result_10 = svadd_x(pg1, result_10, UChi_10); \
-    result_11 = svadd_x(pg1, result_11, UChi_11); \
-    result_12 = svadd_x(pg1, result_12, UChi_12); 
-
-// YP_RECON_ACCUM
-#define YP_RECON_ACCUM_A64FXf  \
-    result_00 = svadd_x(pg1, result_00, UChi_00); \
-    result_30 = svsub_x(pg1, result_30, UChi_00); \
-    result_01 = svadd_x(pg1, result_01, UChi_01); \
-    result_31 = svsub_x(pg1, result_31, UChi_01); \
-    result_02 = svadd_x(pg1, result_02, UChi_02); \
-    result_32 = svsub_x(pg1, result_32, UChi_02); \
-    result_10 = svadd_x(pg1, result_10, UChi_10); \
-    result_20 = svadd_x(pg1, result_20, UChi_10); \
-    result_11 = svadd_x(pg1, result_11, UChi_11); \
-    result_21 = svadd_x(pg1, result_21, UChi_11); \
-    result_12 = svadd_x(pg1, result_12, UChi_12); \
-    result_22 = svadd_x(pg1, result_22, UChi_12); 
-
-// YM_RECON_ACCUM
-#define YM_RECON_ACCUM_A64FXf  \
-    result_00 = svadd_x(pg1, result_00, UChi_00); \
-    result_30 = svadd_x(pg1, result_30, UChi_00); \
-    result_01 = svadd_x(pg1, result_01, UChi_01); \
-    result_31 = svadd_x(pg1, result_31, UChi_01); \
-    result_02 = svadd_x(pg1, result_02, UChi_02); \
-    result_32 = svadd_x(pg1, result_32, UChi_02); \
-    result_10 = svadd_x(pg1, result_10, UChi_10); \
-    result_20 = svsub_x(pg1, result_20, UChi_10); \
-    result_11 = svadd_x(pg1, result_11, UChi_11); \
-    result_21 = svsub_x(pg1, result_21, UChi_11); \
-    result_12 = svadd_x(pg1, result_12, UChi_12); \
-    result_22 = svsub_x(pg1, result_22, UChi_12); 
-
-// ZP_RECON_ACCUM
-#define ZP_RECON_ACCUM_A64FXf  \
-    result_20 = svcadd_x(pg1, result_20, UChi_00, 270);   \
-    result_00 = svadd_x(pg1, result_00, UChi_00); \
-    result_21 = svcadd_x(pg1, result_21, UChi_01, 270);   \
-    result_01 = svadd_x(pg1, result_01, UChi_01); \
-    result_22 = svcadd_x(pg1, result_22, UChi_02, 270);   \
-    result_02 = svadd_x(pg1, result_02, UChi_02); \
-    result_30 = svcadd_x(pg1, result_30, UChi_10, 90);   \
-    result_10 = svadd_x(pg1, result_10, UChi_10); \
-    result_31 = svcadd_x(pg1, result_31, UChi_11, 90);   \
-    result_11 = svadd_x(pg1, result_11, UChi_11); \
-    result_32 = svcadd_x(pg1, result_32, UChi_12, 90);   \
-    result_12 = svadd_x(pg1, result_12, UChi_12); 
-
-// ZM_RECON_ACCUM
-#define ZM_RECON_ACCUM_A64FXf  \
-    result_20 = svcadd_x(pg1, result_20, UChi_00, 90);   \
-    result_00 = svadd_x(pg1, result_00, UChi_00); \
-    result_21 = svcadd_x(pg1, result_21, UChi_01, 90);   \
-    result_01 = svadd_x(pg1, result_01, UChi_01); \
-    result_22 = svcadd_x(pg1, result_22, UChi_02, 90);   \
-    result_02 = svadd_x(pg1, result_02, UChi_02); \
-    result_30 = svcadd_x(pg1, result_30, UChi_10, 270);   \
-    result_10 = svadd_x(pg1, result_10, UChi_10); \
-    result_31 = svcadd_x(pg1, result_31, UChi_11, 270);   \
-    result_11 = svadd_x(pg1, result_11, UChi_11); \
-    result_32 = svcadd_x(pg1, result_32, UChi_12, 270);   \
-    result_12 = svadd_x(pg1, result_12, UChi_12); 
-
-// TP_RECON_ACCUM
-#define TP_RECON_ACCUM_A64FXf  \
-    result_00 = svadd_x(pg1, result_00, UChi_00); \
-    result_20 = svadd_x(pg1, result_20, UChi_00); \
-    result_01 = svadd_x(pg1, result_01, UChi_01); \
-    result_21 = svadd_x(pg1, result_21, UChi_01); \
-    result_02 = svadd_x(pg1, result_02, UChi_02); \
-    result_22 = svadd_x(pg1, result_22, UChi_02); \
-    result_10 = svadd_x(pg1, result_10, UChi_10); \
-    result_30 = svadd_x(pg1, result_30, UChi_10); \
-    result_11 = svadd_x(pg1, result_11, UChi_11); \
-    result_31 = svadd_x(pg1, result_31, UChi_11); \
-    result_12 = svadd_x(pg1, result_12, UChi_12); \
-    result_32 = svadd_x(pg1, result_32, UChi_12); 
-
-// TM_RECON_ACCUM
-#define TM_RECON_ACCUM_A64FXf  \
-    result_00 = svadd_x(pg1, result_00, UChi_00); \
-    result_20 = svsub_x(pg1, result_20, UChi_00); \
-    result_01 = svadd_x(pg1, result_01, UChi_01); \
-    result_21 = svsub_x(pg1, result_21, UChi_01); \
-    result_02 = svadd_x(pg1, result_02, UChi_02); \
-    result_22 = svsub_x(pg1, result_22, UChi_02); \
-    result_10 = svadd_x(pg1, result_10, UChi_10); \
-    result_30 = svsub_x(pg1, result_30, UChi_10); \
-    result_11 = svadd_x(pg1, result_11, UChi_11); \
-    result_31 = svsub_x(pg1, result_31, UChi_11); \
-    result_12 = svadd_x(pg1, result_12, UChi_12); \
-    result_32 = svsub_x(pg1, result_32, UChi_12); 
-
-// ZERO_PSI
-#define ZERO_PSI_A64FXf  \
-    result_00 = svdup_f32(0.); \
-    result_01 = svdup_f32(0.); \
-    result_02 = svdup_f32(0.); \
-    result_10 = svdup_f32(0.); \
-    result_11 = svdup_f32(0.); \
-    result_12 = svdup_f32(0.); \
-    result_20 = svdup_f32(0.); \
-    result_21 = svdup_f32(0.); \
-    result_22 = svdup_f32(0.); \
-    result_30 = svdup_f32(0.); \
-    result_31 = svdup_f32(0.); \
-    result_32 = svdup_f32(0.); 
-
-// PREFETCH_RESULT_L2_STORE (prefetch store to L2)
-#define PREFETCH_RESULT_L2_STORE_INTERNAL_A64FXf(base)  \
-{ \
-    svprfd(pg1, (int64_t*)(base + 0), SV_PSTL2STRM); \
-    svprfd(pg1, (int64_t*)(base + 256), SV_PSTL2STRM); \
-    svprfd(pg1, (int64_t*)(base + 512), SV_PSTL2STRM); \
-}
-// PREFETCH_RESULT_L1_STORE (prefetch store to L1)
-#define PREFETCH_RESULT_L1_STORE_INTERNAL_A64FXf(base)  \
-{ \
-    svprfd(pg1, (int64_t*)(base + 0), SV_PSTL1STRM); \
-    svprfd(pg1, (int64_t*)(base + 256), SV_PSTL1STRM); \
-    svprfd(pg1, (int64_t*)(base + 512), SV_PSTL1STRM); \
-}
-// ADD_RESULT_INTERNAL
-#define ADD_RESULT_INTERNAL_A64FXf  \
-    result_00 = svadd_x(pg1, result_00, Chimu_00); \
-    result_01 = svadd_x(pg1, result_01, Chimu_01); \
-    result_02 = svadd_x(pg1, result_02, Chimu_02); \
-    result_10 = svadd_x(pg1, result_10, Chimu_10); \
-    result_11 = svadd_x(pg1, result_11, Chimu_11); \
-    result_12 = svadd_x(pg1, result_12, Chimu_12); \
-    result_20 = svadd_x(pg1, result_20, Chimu_20); \
-    result_21 = svadd_x(pg1, result_21, Chimu_21); \
-    result_22 = svadd_x(pg1, result_22, Chimu_22); \
-    result_30 = svadd_x(pg1, result_30, Chimu_30); \
-    result_31 = svadd_x(pg1, result_31, Chimu_31); \
-    result_32 = svadd_x(pg1, result_32, Chimu_32); 
-

Grid/simd/Fujitsu_A64FX_undef.h

@@ -1,76 +0,0 @@
-/*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: Fujitsu_A64FX_undef.h
-
-    Copyright (C) 2020
-
-Author: Nils Meyer <nils.meyer@ur.de>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-
-#undef LOAD_CHIMU
-#undef PREFETCH_CHIMU_L1
-#undef PREFETCH_GAUGE_L1
-#undef PREFETCH_CHIMU_L2
-#undef PREFETCH_GAUGE_L2
-#undef PREFETCH_GAUGE_L1_INTERNAL
-#undef PREFETCH1_CHIMU
-#undef PREFETCH_CHIMU
-#undef PREFETCH_RESULT_L2_STORE
-#undef PREFETCH_RESULT_L1_STORE
-#undef LOAD_GAUGE
-#undef LOCK_GAUGE
-#undef UNLOCK_GAUGE
-#undef MASK_REGS
-#undef SAVE_RESULT
-#undef ADD_RESULT
-#undef MULT_2SPIN_1
-#undef MULT_2SPIN_2
-#undef MAYBEPERM
-#undef LOAD_CHI
-#undef XP_PROJ
-#undef YP_PROJ
-#undef ZP_PROJ
-#undef TP_PROJ
-#undef XM_PROJ
-#undef YM_PROJ
-#undef ZM_PROJ
-#undef TM_PROJ
-#undef XP_RECON
-#undef XM_RECON
-#undef XM_RECON_ACCUM
-#undef YM_RECON_ACCUM
-#undef ZM_RECON_ACCUM
-#undef TM_RECON_ACCUM
-#undef XP_RECON_ACCUM
-#undef YP_RECON_ACCUM
-#undef ZP_RECON_ACCUM
-#undef TP_RECON_ACCUM
-#undef PERMUTE
-#undef PERMUTE_DIR0
-#undef PERMUTE_DIR1
-#undef PERMUTE_DIR2
-#undef PERMUTE_DIR3
-#undef LOAD_TABLE
-#undef LOAD_TABLE0
-#undef LOAD_TABLE1
-#undef LOAD_TABLE2
-#undef LOAD_TABLE3

Grid/simd/Grid_a64fx-2.h

@@ -1,942 +0,0 @@
-    /*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: Grid_a64fx-2.h
-
-    Copyright (C) 2020
-
-    Author: Nils Meyer          <nils.meyer@ur.de>
-
-    with support from Arm
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-    *************************************************************************************/
-    /*  END LEGAL */
-
-/////////////////////////////////////////////////////
-// Using SVE ACLE
-/////////////////////////////////////////////////////
-
-static_assert(GEN_SIMD_WIDTH % 64u == 0, "A64FX SIMD vector size is 64 bytes");
-
-NAMESPACE_BEGIN(Grid);
-NAMESPACE_BEGIN(Optimization);
-
-  // type traits giving the number of elements for each vector type
-  template <typename T> struct W;
-  template <> struct W<double> {
-    constexpr static unsigned int c = GEN_SIMD_WIDTH/16u;
-    constexpr static unsigned int r = GEN_SIMD_WIDTH/8u;
-  };
-  template <> struct W<float> {
-    constexpr static unsigned int c = GEN_SIMD_WIDTH/8u;
-    constexpr static unsigned int r = GEN_SIMD_WIDTH/4u;
-  };
-  template <> struct W<Integer> {
-    constexpr static unsigned int r = GEN_SIMD_WIDTH/4u;
-  };
-  template <> struct W<uint16_t> {
-    constexpr static unsigned int c = GEN_SIMD_WIDTH/4u;
-    constexpr static unsigned int r = GEN_SIMD_WIDTH/2u;
-  };
-  template <> struct W<uint64_t> {
-    constexpr static unsigned int c = GEN_SIMD_WIDTH/16u;
-    constexpr static unsigned int r = GEN_SIMD_WIDTH/8u;
-  };
-
-  #ifdef ARMCLANGCOMPAT
-  // SIMD vector immediate types
-  template <typename T>
-  struct vec_imm {
-    alignas(GEN_SIMD_WIDTH) T v[W<T>::r];
-  };
-
-  // SIMD vector types
-  template <typename T>
-  struct vec {
-    alignas(GEN_SIMD_WIDTH) T v[W<T>::r];
-    vec() = default;
-    vec(const vec &rhs) { this->operator=(rhs); }
-    vec(const vec_imm<T> &rhs) {
-      // v = rhs.v
-      svst1(svptrue_b8(), (T*)this, svld1(svptrue_b8(), (T*)rhs.v));
-    }
-
-    inline vec &operator=(const vec &rhs) {
-      // v = rhs.v
-      svst1(svptrue_b8(), (T*)this, svld1(svptrue_b8(), (T*)rhs.v));
-      return *this;
-    };
-  };
-
-  #else //  no ARMCLANGCOMPAT
-  #define vec_imm vec
-  // SIMD vector types
-  template <typename T>
-  struct vec {
-    alignas(GEN_SIMD_WIDTH) T v[W<T>::r];
-  };
-  #endif
-
-  typedef vec<float>     vecf;
-  typedef vec<double>    vecd;
-  typedef vec<uint16_t>  vech; // half precision comms
-  typedef vec<Integer>   veci;
-
-NAMESPACE_END(Optimization)
-NAMESPACE_END(Grid)
-
-// low-level API
-NAMESPACE_BEGIN(Grid);
-NAMESPACE_BEGIN(Optimization);
-
-template <typename T>
-struct acle{};
-
-template <>
-struct acle<double>{
-  typedef svfloat64_t vt;
-  typedef svfloat64x2_t vt2;
-  typedef svfloat64x4_t vt4;
-  typedef float64_t pt;
-  typedef uint64_t uint;
-  typedef svuint64_t svuint;
-
-  static inline svbool_t pg1(){return svptrue_b64();}
-  static inline svbool_t pg2(){return svptrue_pat_b64(SV_VL4);}
-  static inline svbool_t pg4(){return svptrue_pat_b64(SV_VL2);}
-  static inline vec<uint64_t> tbl_swap(){
-    //const vec<uint64_t> t = {1, 0, 3, 2, 5, 4, 7, 6};
-    const vec_imm<uint64_t> t = {1, 0, 3, 2, 5, 4, 7, 6};
-    return t;
-  }
-  static inline vec<uint64_t> tbl0(){
-    //const vec<uint64_t> t = {4, 5, 6, 7, 0, 1, 2, 3};
-    const vec_imm<uint64_t> t = {4, 5, 6, 7, 0, 1, 2, 3};
-    return t;
-  }
-  static inline vec<uint64_t> tbl1(){
-    //const vec<uint64_t> t = {2, 3, 0, 1, 6, 7, 4, 5};
-    const vec_imm<uint64_t> t = {2, 3, 0, 1, 6, 7, 4, 5};
-    return t;
-  }
-  static inline vec<uint64_t> tbl_exch1a(){ // Exchange1
-    //const vec<uint64_t> t = {0, 1, 4, 5, 2, 3, 6, 7};
-    const vec_imm<uint64_t> t = {0, 1, 4, 5, 2, 3, 6, 7};
-    return t;
-  }
-  static inline vec<uint64_t> tbl_exch1b(){ // Exchange1
-    //const vec<uint64_t> t = {2, 3, 6, 7, 0, 1, 4, 5};
-    const vec_imm<uint64_t> t = {2, 3, 6, 7, 0, 1, 4, 5};
-    return t;
-  }
-  static inline vec<uint64_t> tbl_exch1c(){ // Exchange1
-    //const vec<uint64_t> t = {4, 5, 0, 1, 6, 7, 2, 3};
-    const vec_imm<uint64_t> t = {4, 5, 0, 1, 6, 7, 2, 3};
-    return t;
-  }
-  static inline svbool_t pg_even(){return svzip1_b64(svptrue_b64(), svpfalse_b());}
-  static inline svbool_t pg_odd() {return svzip1_b64(svpfalse_b(), svptrue_b64());}
-  static inline svfloat64_t zero(){return svdup_f64(0.);}
-};
-
-template <>
-struct acle<float>{
-  typedef svfloat32_t vt;
-  typedef svfloat32x2_t vt2;
-  typedef float32_t pt;
-  typedef uint32_t uint;
-  typedef svuint32_t svuint;
-
-  static inline svbool_t pg1(){return svptrue_b32();}
-  static inline svbool_t pg2(){return svptrue_pat_b32(SV_VL8);}
-  // exchange neighboring elements
-  static inline vec<uint32_t> tbl_swap(){
-    //const vec<uint32_t> t = {1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14};
-    const vec_imm<uint32_t> t = {1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14};
-    return t;
-  }
-  static inline vec<uint32_t> tbl0(){
-    //const vec<uint32_t> t = {8, 9, 10, 11, 12, 13, 14, 15, 0, 1, 2, 3, 4, 5, 6, 7};
-    const vec_imm<uint32_t> t = {8, 9, 10, 11, 12, 13, 14, 15, 0, 1, 2, 3, 4, 5, 6, 7};
-    return t;
-  }
-  static inline vec<uint32_t> tbl1(){
-    //const vec<uint32_t> t = {4, 5, 6, 7, 0, 1, 2, 3, 12, 13, 14, 15, 8, 9, 10, 11};
-    const vec_imm<uint32_t> t = {4, 5, 6, 7, 0, 1, 2, 3, 12, 13, 14, 15, 8, 9, 10, 11};
-    return t;
-  }
-  static inline vec<uint32_t> tbl2(){
-    //const vec<uint32_t> t = {2, 3, 0, 1, 6, 7, 4, 5, 10, 11, 8, 9, 14, 15, 12, 13};
-    const vec_imm<uint32_t> t = {2, 3, 0, 1, 6, 7, 4, 5, 10, 11, 8, 9, 14, 15, 12, 13};
-    return t;
-  }
-  static inline vec<uint32_t> tbl_exch1a(){ // Exchange1
-    //const vec<uint32_t> t = {0, 1, 2, 3, 8, 9, 10, 11, 4, 5, 6, 7, 12, 13, 14, 15 };
-    const vec_imm<uint32_t> t = {0, 1, 2, 3, 8, 9, 10, 11, 4, 5, 6, 7, 12, 13, 14, 15 };
-    return t;
-  }
-  static inline vec<uint32_t> tbl_exch1b(){ // Exchange1
-    //const vec<uint32_t> t = {4, 5, 6, 7, 12, 13, 14, 15, 0, 1, 2, 3, 8, 9, 10, 11 };
-    const vec_imm<uint32_t> t = {4, 5, 6, 7, 12, 13, 14, 15, 0, 1, 2, 3, 8, 9, 10, 11 };
-    return t;
-  }
-  static inline vec<uint32_t> tbl_exch1c(){ // Exchange1
-    //const vec<uint32_t> t = {8, 9, 10, 11, 0, 1, 2, 3, 12, 13, 14, 15, 4, 5, 6, 7};
-    const vec_imm<uint32_t> t = {8, 9, 10, 11, 0, 1, 2, 3, 12, 13, 14, 15, 4, 5, 6, 7};
-    return t;
-  }
-  static inline svbool_t pg_even(){return svzip1_b32(svptrue_b32(), svpfalse_b());}
-  static inline svbool_t pg_odd() {return svzip1_b32(svpfalse_b(), svptrue_b32());}
-  static inline svfloat32_t zero(){return svdup_f32(0.);}
-};
-
-template <>
-struct acle<uint16_t>{
-  typedef svfloat16_t vt;
-  typedef float16_t pt;
-  typedef uint16_t uint;
-  typedef svuint16_t svuint;
-
-  static inline svbool_t pg1(){return svptrue_b16();}
-  static inline svbool_t pg2(){return svptrue_pat_b16(SV_VL16);}
-  static inline svbool_t pg_even(){return svzip1_b16(svptrue_b16(), svpfalse_b());}
-  static inline svbool_t pg_odd() {return svzip1_b16(svpfalse_b(), svptrue_b16());}
-  static inline svfloat16_t zero(){return svdup_f16(0.);}
-};
-
-template <>
-struct acle<Integer>{
-  typedef svuint32_t vt;
-  typedef svuint32x2_t vt2;
-  typedef Integer pt;
-  typedef uint32_t uint;
-  typedef svuint32_t svuint;
-
-  //static inline svbool_t pg1(){return svptrue_b16();}
-  static inline svbool_t pg1(){return svptrue_b32();}
-  static inline svbool_t pg2(){return svptrue_pat_b32(SV_VL8);}
-  static inline svbool_t pg_even(){return svzip1_b32(svptrue_b32(), svpfalse_b());}
-  static inline svbool_t pg_odd() {return svzip1_b32(svpfalse_b(), svptrue_b32());}
-};
-
-// ---------------------------------------------------
-
-struct Vsplat{
-  // Complex float
-  inline vecf operator()(float a, float b){
-    vecf out;
-    svbool_t pg1 = acle<float>::pg1();
-    typename acle<float>::vt a_v = svdup_f32(a);
-    typename acle<float>::vt b_v = svdup_f32(b);
-    typename acle<float>::vt r_v = svzip1(a_v, b_v);
-    svst1(pg1, out.v, r_v);
-    return out;
-  }
-
-  // Real float
-  inline vecf operator()(float a){
-    vecf out;
-    svbool_t pg1 = acle<float>::pg1();
-    typename acle<float>::vt r_v = svdup_f32(a);
-    svst1(pg1, out.v, r_v);
-    return out;
-  }
-
- // Complex double
-  inline vecd operator()(double a, double b){
-    vecd out;
-    svbool_t pg1 = acle<double>::pg1();
-    typename acle<double>::vt a_v = svdup_f64(a);
-    typename acle<double>::vt b_v = svdup_f64(b);
-    typename acle<double>::vt r_v = svzip1(a_v, b_v);
-    svst1(pg1, out.v, r_v);
-    return out;
-  }
-
-  // Real double
-  inline vecd operator()(double a){
-    vecd out;
-    svbool_t pg1 = acle<double>::pg1();
-    typename acle<double>::vt r_v = svdup_f64(a);
-    svst1(pg1, out.v, r_v);
-    return out;
-  }
-
-  // Integer
-  inline vec<Integer> operator()(Integer a){
-    vec<Integer> out;
-    svbool_t pg1 = acle<Integer>::pg1();
-    // Add check whether Integer is really a uint32_t???
-    typename acle<Integer>::vt r_v = svdup_u32(a);
-    svst1(pg1, out.v, r_v);
-    return out;
-  }
-};
-
-struct Vstore{
-  // Real
-  template <typename T>
-  inline void operator()(vec<T> a, T *D){
-    svbool_t pg1 = acle<T>::pg1();
-    typename acle<T>::vt a_v = svld1(pg1, (typename acle<T>::pt*)&a.v);
-    svst1(pg1, D, a_v);
-  }
-};
-
-struct Vstream{
-  // Real
-  template <typename T>
-  inline void operator()(T * a, vec<T> b){
-    svbool_t pg1 = acle<T>::pg1();
-    typename acle<T>::vt b_v = svld1(pg1, b.v);
-    svstnt1(pg1, a, b_v);
-    //svst1(pg1, a, b_v);
-  }
-};
-
-  struct Vset{
-    // Complex
-    template <typename T>
-    inline vec<T> operator()(std::complex<T> *a){
-      vec<T> out;
-      svbool_t pg1 = acle<T>::pg1();
-      typename acle<T>::vt a_v = svld1(pg1, (T*)a);
-      svst1(pg1, out.v, a_v);
-
-      return out;
-    }
-
-    // Real
-    template <typename T>
-    inline vec<T> operator()(T *a){
-      vec<T> out;
-      svbool_t pg1 = acle<T>::pg1();
-      typename acle<T>::vt a_v = svld1(pg1, a);
-      svst1(pg1, out.v, a_v);
-
-      return out;
-    }
-  };
-
-/////////////////////////////////////////////////////
-// Arithmetic operations
-/////////////////////////////////////////////////////
-
-struct Sum{
-  template <typename T>
-  inline vec<T> operator()(vec<T> a, vec<T> b){
-    vec<T> out;
-    svbool_t pg1 = acle<T>::pg1();
-    typename acle<T>::vt a_v = svld1(pg1, a.v);
-    typename acle<T>::vt b_v = svld1(pg1, b.v);
-    typename acle<T>::vt r_v = svadd_x(pg1, a_v, b_v);
-    svst1(pg1, out.v, r_v);
-
-    return out;
-  }
-};
-
-struct Sub{
-  template <typename T>
-  inline vec<T> operator()(vec<T> a, vec<T> b){
-    vec<T> out;
-    svbool_t pg1 = acle<T>::pg1();
-    typename acle<T>::vt a_v = svld1(pg1, a.v);
-    typename acle<T>::vt b_v = svld1(pg1, b.v);
-    typename acle<T>::vt r_v = svsub_x(pg1, a_v, b_v);
-    svst1(pg1, out.v, r_v);
-
-    return out;
-  }
-};
-
-struct Mult{
-  template <typename T>
-  inline vec<T> operator()(vec<T> a, vec<T> b, vec<T> c){
-    vec<T> out;
-    svbool_t pg1 = acle<T>::pg1();
-    typename acle<T>::vt a_v = svld1(pg1, a.v);
-    typename acle<T>::vt b_v = svld1(pg1, b.v);
-    typename acle<T>::vt c_v = svld1(pg1, c.v);
-    typename acle<T>::vt r_v = svmla_x(pg1, c_v, a_v, b_v);
-    svst1(pg1, out.v, r_v);
-
-    return out;
-  }
-  template <typename T>
-  inline vec<T> operator()(vec<T> a, vec<T> b){
-    vec<T> out;
-    svbool_t pg1 = acle<T>::pg1();
-    typename acle<T>::vt a_v = svld1(pg1, a.v);
-    typename acle<T>::vt b_v = svld1(pg1, b.v);
-    typename acle<T>::vt r_v = svmul_x(pg1, a_v, b_v);
-    svst1(pg1, out.v, r_v);
-
-    return out;
-  }
-};
-
-struct MultRealPart{
-  template <typename T>
-  inline vec<T> operator()(vec<T> a, vec<T> b){
-    vec<T> out;
-    svbool_t pg1 = acle<T>::pg1();
-    typename acle<T>::vt a_v  = svld1(pg1, a.v);
-    typename acle<T>::vt b_v  = svld1(pg1, b.v);
-
-    // using FCMLA
-    typename acle<T>::vt z_v = acle<T>::zero();
-    typename acle<T>::vt r_v = svcmla_x(pg1, z_v, a_v, b_v, 0);
-
-    svst1(pg1, out.v, r_v);
-
-    return out;
-  }
-};
-
-struct MaddRealPart{
-  template <typename T>
-  inline vec<T> operator()(vec<T> a, vec<T> b, vec<T> c){
-    vec<T> out;
-    svbool_t pg1 = acle<T>::pg1();
-    typename acle<T>::vt a_v  = svld1(pg1, a.v);
-    typename acle<T>::vt b_v  = svld1(pg1, b.v);
-    typename acle<T>::vt c_v  = svld1(pg1, c.v);
-
-    // using FCMLA
-    typename acle<T>::vt r_v = svcmla_x(pg1, c_v, a_v, b_v, 0);
-
-    svst1(pg1, out.v, r_v);
-
-    return out;
-  }
-};
-
-struct MultComplex{
-  // Complex a*b
-  template <typename T>
-  inline vec<T> operator()(vec<T> a, vec<T> b){
-    vec<T> out;
-    svbool_t pg1 = acle<T>::pg1();
-    typename acle<T>::vt a_v = svld1(pg1, a.v);
-    typename acle<T>::vt b_v = svld1(pg1, b.v);
-    typename acle<T>::vt z_v = acle<T>::zero();
-
-    // using FCMLA
-    typename acle<T>::vt r_v = svcmla_x(pg1, z_v, a_v, b_v, 0);
-    r_v = svcmla_x(pg1, r_v, a_v, b_v, 90);
-
-    svst1(pg1, out.v, r_v);
-
-    return out;
-  }
-};
-
-struct MultAddComplex{
-  // Complex a*b+c
-  template <typename T>
-  inline vec<T> operator()(vec<T> a, vec<T> b, vec<T> c){
-    vec<T> out;
-    svbool_t pg1 = acle<T>::pg1();
-    typename acle<T>::vt a_v = svld1(pg1, a.v);
-    typename acle<T>::vt b_v = svld1(pg1, b.v);
-    typename acle<T>::vt c_v = svld1(pg1, c.v);;
-
-    // using FCMLA
-    typename acle<T>::vt r_v = svcmla_x(pg1, c_v, a_v, b_v, 0);
-    r_v = svcmla_x(pg1, r_v, a_v, b_v, 90);
-    svst1(pg1, out.v, r_v);
-
-    return out;
-  }
-};
-
-struct Div{
-  // Real
-  template <typename T>
-  inline vec<T> operator()(vec<T> a, vec<T> b){
-    vec<T> out;
-    svbool_t pg1 = acle<T>::pg1();
-    typename acle<T>::vt a_v = svld1(pg1, a.v);
-    typename acle<T>::vt b_v = svld1(pg1, b.v);
-    typename acle<T>::vt r_v = svdiv_x(pg1, a_v, b_v);
-    svst1(pg1, out.v, r_v);
-
-    return out;
-  }
-};
-
-struct Conj{
-  // Complex
-  template <typename T>
-  inline vec<T> operator()(vec<T> a){
-    vec<T> out;
-    svbool_t pg1 = acle<T>::pg1();
-    svbool_t pg_odd = acle<T>::pg_odd();
-    typename acle<T>::vt a_v = svld1(pg1, a.v);
-    //typename acle<T>::vt r_v = svneg_x(pg_odd, a_v);
-    typename acle<T>::vt r_v = svneg_m(a_v, pg_odd, a_v);
-    svst1(pg1, out.v, r_v);
-
-    return out;
-  }
-};
-
-struct TimesMinusI{
-  // Complex
-  template <typename T>
-  inline vec<T> operator()(vec<T> a, vec<T> b){
-    vec<T> out;
-    const vec<typename acle<T>::uint> tbl_swap = acle<T>::tbl_swap();
-    svbool_t pg1 = acle<T>::pg1();
-    svbool_t pg_odd = acle<T>::pg_odd();
-
-    typename acle<T>::svuint tbl_swap_v = svld1(pg1, tbl_swap.v);
-    typename acle<T>::vt a_v = svld1(pg1, a.v);
-    a_v = svtbl(a_v, tbl_swap_v);
-    typename acle<T>::vt r_v = svneg_m(a_v, pg_odd, a_v);
-    svst1(pg1, out.v, r_v);
-
-    return out;
-  }
-};
-
-struct TimesI{
-  // Complex
-  template <typename T>
-  inline vec<T> operator()(vec<T> a, vec<T> b){
-    vec<T> out;
-    const vec<typename acle<T>::uint> tbl_swap = acle<T>::tbl_swap();
-    svbool_t pg1 = acle<T>::pg1();
-    svbool_t pg_even = acle<T>::pg_even();
-
-    typename acle<T>::svuint tbl_swap_v = svld1(pg1, tbl_swap.v);
-    typename acle<T>::vt a_v = svld1(pg1, a.v);
-    a_v = svtbl(a_v, tbl_swap_v);
-    //typename acle<T>::vt r_v = svneg_x(pg_even, a_v);
-    typename acle<T>::vt r_v = svneg_m(a_v, pg_even, a_v);
-    svst1(pg1, out.v, r_v);
-
-    return out;
-  }
-};
-
-struct PrecisionChange {
-  static inline vech StoH (const vecf &sa,const vecf &sb) {
-    vech ret;
-    svbool_t pg1s = acle<float>::pg1();
-    svbool_t pg1h = acle<uint16_t>::pg1();
-    typename acle<float>::vt sa_v = svld1(pg1s, sa.v);
-    typename acle<float>::vt sb_v = svld1(pg1s, sb.v);
-    typename acle<uint16_t>::vt ha_v = svcvt_f16_x(pg1s, sa_v);
-    typename acle<uint16_t>::vt hb_v = svcvt_f16_x(pg1s, sb_v);
-    typename acle<uint16_t>::vt r_v = svuzp1(ha_v, hb_v);
-    svst1(pg1h, (typename acle<uint16_t>::pt*)&ret.v, r_v);
-
-    return ret;
-  }
-  static inline void HtoS(vech h,vecf &sa,vecf &sb) {
-    svbool_t pg1h = acle<uint16_t>::pg1();
-    svbool_t pg1s = acle<float>::pg1();
-    typename acle<uint16_t>::vt h_v = svld1(pg1h, (typename acle<uint16_t>::pt*)&h.v);
-    typename acle<uint16_t>::vt ha_v = svzip1(h_v, h_v);
-    typename acle<uint16_t>::vt hb_v = svzip2(h_v, h_v);
-    typename acle<float>::vt sa_v = svcvt_f32_x(pg1s, ha_v);
-    typename acle<float>::vt sb_v = svcvt_f32_x(pg1s, hb_v);
-    svst1(pg1s, sa.v, sa_v);
-    svst1(pg1s, sb.v, sb_v);
-  }
-  static inline vecf DtoS (vecd a,vecd b) {
-    vecf ret;
-    svbool_t pg1d = acle<double>::pg1();
-    svbool_t pg1s = acle<float>::pg1();
-    typename acle<double>::vt a_v = svld1(pg1d, a.v);
-    typename acle<double>::vt b_v = svld1(pg1d, b.v);
-    typename acle<float>::vt sa_v = svcvt_f32_x(pg1d, a_v);
-    typename acle<float>::vt sb_v = svcvt_f32_x(pg1d, b_v);
-    typename acle<float>::vt r_v = svuzp1(sa_v, sb_v);
-    svst1(pg1s, ret.v, r_v);
-
-    return ret;
-  }
-  static inline void StoD (vecf s,vecd &a,vecd &b) {
-    svbool_t pg1s = acle<float>::pg1();
-    svbool_t pg1d = acle<double>::pg1();
-    typename acle<float>::vt s_v = svld1(pg1s, s.v);
-    typename acle<float>::vt sa_v = svzip1(s_v, s_v);
-    typename acle<float>::vt sb_v = svzip2(s_v, s_v);
-    typename acle<double>::vt a_v = svcvt_f64_x(pg1d, sa_v);
-    typename acle<double>::vt b_v = svcvt_f64_x(pg1d, sb_v);
-    svst1(pg1d, a.v, a_v);
-    svst1(pg1d, b.v, b_v);
-  }
-  static inline vech DtoH (vecd a,vecd b,vecd c,vecd d) {
-    vech ret;
-    svbool_t pg1d = acle<double>::pg1();
-    svbool_t pg1h = acle<uint16_t>::pg1();
-    typename acle<double>::vt a_v = svld1(pg1d, a.v);
-    typename acle<double>::vt b_v = svld1(pg1d, b.v);
-    typename acle<double>::vt c_v = svld1(pg1d, c.v);
-    typename acle<double>::vt d_v = svld1(pg1d, d.v);
-    typename acle<uint16_t>::vt ha_v = svcvt_f16_x(pg1d, a_v);
-    typename acle<uint16_t>::vt hb_v = svcvt_f16_x(pg1d, b_v);
-    typename acle<uint16_t>::vt hc_v = svcvt_f16_x(pg1d, c_v);
-    typename acle<uint16_t>::vt hd_v = svcvt_f16_x(pg1d, d_v);
-    typename acle<uint16_t>::vt hab_v = svuzp1(ha_v, hb_v);
-    typename acle<uint16_t>::vt hcd_v = svuzp1(hc_v, hd_v);
-    typename acle<uint16_t>::vt r_v = svuzp1(hab_v, hcd_v);
-    svst1(pg1h, (typename acle<uint16_t>::pt*)&ret.v, r_v);
-
-    return ret;
-/*
-    vecf sa,sb;
-    sa = DtoS(a,b);
-    sb = DtoS(c,d);
-    return StoH(sa,sb);
-*/
-  }
-  static inline void HtoD(vech h,vecd &a,vecd &b,vecd &c,vecd &d) {
-    svbool_t pg1h = acle<uint16_t>::pg1();
-    svbool_t pg1d = acle<double>::pg1();
-    typename acle<uint16_t>::vt h_v = svld1(pg1h, (typename acle<uint16_t>::pt*)&h.v);
-    typename acle<uint16_t>::vt sa_v = svzip1(h_v, h_v);
-    typename acle<uint16_t>::vt sb_v = svzip2(h_v, h_v);
-    typename acle<uint16_t>::vt da_v = svzip1(sa_v, sa_v);
-    typename acle<uint16_t>::vt db_v = svzip2(sa_v, sa_v);
-    typename acle<uint16_t>::vt dc_v = svzip1(sb_v, sb_v);
-    typename acle<uint16_t>::vt dd_v = svzip2(sb_v, sb_v);
-    typename acle<double>::vt a_v = svcvt_f64_x(pg1d, da_v);
-    typename acle<double>::vt b_v = svcvt_f64_x(pg1d, db_v);
-    typename acle<double>::vt c_v = svcvt_f64_x(pg1d, dc_v);
-    typename acle<double>::vt d_v = svcvt_f64_x(pg1d, dd_v);
-    svst1(pg1d, a.v, a_v);
-    svst1(pg1d, b.v, b_v);
-    svst1(pg1d, c.v, c_v);
-    svst1(pg1d, d.v, d_v);
-/*
-    vecf sa,sb;
-    HtoS(h,sa,sb);
-    StoD(sa,a,b);
-    StoD(sb,c,d);
-*/
-  }
-};
-
-struct Exchange{
-
-  // Exchange0 is valid for arbitrary SVE vector length
-  template <typename T>
-  static inline void Exchange0(vec<T> &out1, vec<T> &out2, const vec<T> &in1, const vec<T> &in2){
-    svbool_t pg1 = acle<T>::pg1();
-    typename acle<T>::vt a1_v = svld1(pg1, in1.v);
-    typename acle<T>::vt a2_v = svld1(pg1, in2.v);
-    typename acle<T>::vt r1_v = svext(a1_v, a1_v, (uint64_t)W<T>::c);
-    r1_v = svext(r1_v, a2_v, (uint64_t)W<T>::c);
-    typename acle<T>::vt r2_v = svext(a2_v, a2_v, (uint64_t)W<T>::c);
-    r2_v = svext(a1_v, r2_v, (uint64_t)W<T>::c);
-    svst1(pg1, out1.v, r1_v);
-    svst1(pg1, out2.v, r2_v);
-  }
-
-  template <typename T>
-  static inline void Exchange1(vec<T> &out1, vec<T> &out2, const vec<T> &in1, const vec<T> &in2){
-    // this one is tricky; svtrn2q* from SVE2 fits best, but it is not available in SVE1
-    // alternative: use 4-el structure; expect translation into ldp + stp -> SFI
-    svbool_t pg1 = acle<T>::pg1();
-    const vec<typename acle<T>::uint> tbl_exch1a = acle<T>::tbl_exch1a();
-    const vec<typename acle<T>::uint> tbl_exch1b = acle<T>::tbl_exch1b();
-    const vec<typename acle<T>::uint> tbl_exch1c = acle<T>::tbl_exch1c();
-
-    typename acle<T>::svuint tbl_exch1a_v = svld1(pg1, tbl_exch1a.v);
-    typename acle<T>::svuint tbl_exch1b_v = svld1(pg1, tbl_exch1b.v);
-    typename acle<T>::svuint tbl_exch1c_v = svld1(pg1, tbl_exch1c.v);
-
-    typename acle<T>::vt in1_v  = svld1(pg1, in1.v);
-    typename acle<T>::vt in2_v  = svld1(pg1, in2.v);
-
-    typename acle<T>::vt a1_v   = svtbl(in1_v, tbl_exch1a_v);
-    typename acle<T>::vt a2_v   = svtbl(in2_v, tbl_exch1b_v);
-    typename acle<T>::vt b1_v   = svext(a2_v, a1_v, (uint64_t)(W<T>::r / 2u));
-    typename acle<T>::vt b2_v   = svext(a1_v, a2_v, (uint64_t)(W<T>::r / 2u));
-    typename acle<T>::vt out1_v = svtbl(b1_v, tbl_exch1c_v);
-    typename acle<T>::vt out2_v = svtbl(b2_v, tbl_exch1a_v);
-
-    svst1(pg1, out1.v, out1_v);
-    svst1(pg1, out2.v, out2_v);
-  }
-
-  template <typename T>
-  static inline void Exchange2(vec<T> &out1, vec<T> &out2, const vec<T> &in1, const vec<T> &in2){
-    svbool_t pg1 = acle<double>::pg1();
-    typename acle<double>::vt a1_v = svld1(pg1, (typename acle<double>::pt*)in1.v);
-    typename acle<double>::vt a2_v = svld1(pg1, (typename acle<double>::pt*)in2.v);
-    typename acle<double>::vt r1_v = svtrn1(a1_v, a2_v);
-    typename acle<double>::vt r2_v = svtrn2(a1_v, a2_v);
-    svst1(pg1, (typename acle<double>::pt*)out1.v, r1_v);
-    svst1(pg1, (typename acle<double>::pt*)out2.v, r2_v);
-  }
-
-  static inline void Exchange3(vecf &out1, vecf &out2, const vecf &in1, const vecf &in2){
-    svbool_t pg1 = acle<float>::pg1();
-    typename acle<float>::vt a1_v = svld1(pg1, in1.v);
-    typename acle<float>::vt a2_v = svld1(pg1, in2.v);
-    typename acle<float>::vt r1_v = svtrn1(a1_v, a2_v);
-    typename acle<float>::vt r2_v = svtrn2(a1_v, a2_v);
-    svst1(pg1, out1.v, r1_v);
-    svst1(pg1, out2.v, r2_v);
-  }
-
-  static inline void Exchange3(vecd &out1, vecd &out2, const vecd &in1, const vecd &in2){
-    assert(0);
-    return;
-  }
-};
-
-struct Permute{
-
-  // Permute0 is valid for any SVE vector width
-  template <typename T>
-  static inline vec<T> Permute0(vec<T> in) {
-    vec<T> out;
-    svbool_t pg1 = acle<T>::pg1();
-    typename acle<T>::vt a_v = svld1(pg1, in.v);
-    typename acle<T>::vt r_v = svext(a_v, a_v, (uint64_t)(W<T>::r / 2u));
-    svst1(pg1, out.v, r_v);
-
-    return out;
-  }
-
-  static inline vecd Permute1(vecd in) {
-    vecd out;
-    const vec<typename acle<double>::uint> tbl_swap = acle<double>::tbl1();
-    svbool_t pg1 = acle<double>::pg1();
-    typename acle<double>::vt a_v = svld1(pg1, in.v);
-    typename acle<double>::svuint tbl_swap_v = svld1(pg1, tbl_swap.v);
-    typename acle<double>::vt r_v = svtbl(a_v, tbl_swap_v);
-    svst1(pg1, out.v, r_v);
-
-    return out;
-  }
-
-  static inline vecf Permute1(vecf in) {
-    vecf out;
-    const vec<typename acle<float>::uint> tbl_swap = acle<float>::tbl1();
-    svbool_t pg1 = acle<float>::pg1();
-    typename acle<float>::vt a_v = svld1(pg1, in.v);
-    typename acle<float>::svuint tbl_swap_v = svld1(pg1, tbl_swap.v);
-    typename acle<float>::vt r_v = svtbl(a_v, tbl_swap_v);
-    svst1(pg1, out.v, r_v);
-
-    return out;
-  }
-
-  static inline vecd Permute2(vecd in) {
-    vecd out;
-    const vec<typename acle<double>::uint> tbl_swap = acle<double>::tbl_swap();
-    svbool_t pg1 = acle<double>::pg1();
-    typename acle<double>::vt a_v = svld1(pg1, in.v);
-    typename acle<double>::svuint tbl_swap_v = svld1(pg1, tbl_swap.v);
-    typename acle<double>::vt r_v = svtbl(a_v, tbl_swap_v);
-    svst1(pg1, out.v, r_v);
-
-    return out;
-  }
-
-  static inline vecf Permute2(vecf in) {
-    vecf out;
-    const vec<typename acle<float>::uint> tbl_swap = acle<float>::tbl2();
-    svbool_t pg1 = acle<float>::pg1();
-    typename acle<float>::vt a_v = svld1(pg1, in.v);
-    typename acle<float>::svuint tbl_swap_v = svld1(pg1, tbl_swap.v);
-    typename acle<float>::vt r_v = svtbl(a_v, tbl_swap_v);
-    svst1(pg1, out.v, r_v);
-
-    return out;
-  }
-
-  static inline vecf Permute3(vecf in) {
-    vecf out;
-    const vec<typename acle<float>::uint> tbl_swap = acle<float>::tbl_swap();
-    svbool_t pg1 = acle<float>::pg1();
-    typename acle<float>::vt a_v = svld1(pg1, in.v);
-    typename acle<float>::svuint tbl_swap_v = svld1(pg1, tbl_swap.v);
-    typename acle<float>::vt r_v = svtbl(a_v, tbl_swap_v);
-    svst1(pg1, out.v, r_v);
-
-    return out;
-  }
-
-  static inline vecd Permute3(vecd in) {
-    return in;
-  }
-
-};
-
-struct Rotate{
-
-  template <int n, typename T> static inline vec<T> tRotate(vec<T> in){
-    vec<T> out;
-    svbool_t pg1 = acle<T>::pg1();
-    typename acle<T>::vt a_v = svld1(pg1, in.v);
-    typename acle<T>::vt r_v = svext(a_v, a_v, (uint64_t)(n%W<T>::r));
-    svst1(pg1, out.v, r_v);
-
-    return out;
-  }
-
-  template <typename T>
-  static inline vec<T> rotate(vec<T> in, int n){
-
-    switch(n){
-    case 0:  return tRotate<0,  T>(in); break;
-    case 1:  return tRotate<1,  T>(in); break;
-    case 2:  return tRotate<2,  T>(in); break;
-    case 3:  return tRotate<3,  T>(in); break;
-    case 4:  return tRotate<4,  T>(in); break;
-    case 5:  return tRotate<5,  T>(in); break;
-    case 6:  return tRotate<6,  T>(in); break;
-    case 7:  return tRotate<7,  T>(in); break;
-
-    case 8:  return tRotate<8,  T>(in); break;
-    case 9:  return tRotate<9,  T>(in); break;
-    case 10: return tRotate<10, T>(in); break;
-    case 11: return tRotate<11, T>(in); break;
-    case 12: return tRotate<12, T>(in); break;
-    case 13: return tRotate<13, T>(in); break;
-    case 14: return tRotate<14, T>(in); break;
-    case 15: return tRotate<15, T>(in); break;
-    default: assert(0);
-    }
-  }
-};
-
-// tree-based reduction
-#define svred(pg, v)\
-svaddv(pg, v);
-
-// left-to-right reduction
-// #define svred(pg, v)\
-// svadda(pg, 0, v)
-
-template <typename Out_type, typename In_type>
-struct Reduce{
-  //Need templated class to overload output type
-  //General form must generate error if compiled
-  inline Out_type operator()(In_type in){
-    printf("Error, using wrong Reduce function\n");
-    exit(1);
-    return 0;
-  }
-};
-
-//Complex float Reduce
-template <>
-inline Grid::ComplexF Reduce<Grid::ComplexF, vecf>::operator()(vecf in){
-  svbool_t pg1 = acle<float>::pg1();
-  svbool_t pg_even = acle<float>::pg_even();
-  svbool_t pg_odd  = acle<float>::pg_odd();
-  typename acle<float>::vt a_v = svld1(pg1, in.v);
-  float a = svred(pg_even, a_v);
-  float b = svred(pg_odd, a_v);
-
-  return Grid::ComplexF(a, b);
-
-}
-
-//Real float Reduce
-template <>
-inline Grid::RealF Reduce<Grid::RealF, vecf>::operator()(vecf in){
-  svbool_t pg1 = acle<float>::pg1();
-  typename acle<float>::vt a_v = svld1(pg1, in.v);
-  float a = svred(pg1, a_v);
-
-  return a;
-}
-
-//Complex double Reduce
-template <>
-inline Grid::ComplexD Reduce<Grid::ComplexD, vecd>::operator()(vecd in){
-  svbool_t pg1 = acle<double>::pg1();
-  svbool_t pg_even = acle<double>::pg_even();
-  svbool_t pg_odd  = acle<double>::pg_odd();
-  typename acle<double>::vt a_v = svld1(pg1, in.v);
-  double a = svred(pg_even, a_v);
-  double b = svred(pg_odd, a_v);
-
-  return Grid::ComplexD(a, b);
-}
-
-//Real double Reduce
-template <>
-inline Grid::RealD Reduce<Grid::RealD, vecd>::operator()(vecd in){
-  svbool_t pg1 = acle<double>::pg1();
-  typename acle<double>::vt a_v = svld1(pg1, in.v);
-  double a = svred(pg1, a_v);
-
-  return a;
-}
-
-//Integer Reduce
-template <>
-inline Integer Reduce<Integer, veci>::operator()(veci in){
-  svbool_t pg1 = acle<Integer>::pg1();
-  typename acle<Integer>::vt a_v = svld1(pg1, in.v);
-  Integer a = svred(pg1, a_v);
-
-  return a;
-}
-
-#undef svred
-#undef vec_imm
-
-NAMESPACE_END(Optimization)
-
-//////////////////////////////////////////////////////////////////////////////////////
-// Here assign types
-
-typedef Optimization::vech SIMD_Htype; // Reduced precision type
-typedef Optimization::vecf SIMD_Ftype; // Single precision type
-typedef Optimization::vecd SIMD_Dtype; // Double precision type
-typedef Optimization::veci SIMD_Itype; // Integer type
-
-// prefetch utilities
-inline void v_prefetch0(int size, const char *ptr){};
-inline void prefetch_HINT_T0(const char *ptr){};
-
-// Function name aliases
-typedef Optimization::Vsplat   VsplatSIMD;
-typedef Optimization::Vstore   VstoreSIMD;
-typedef Optimization::Vset     VsetSIMD;
-typedef Optimization::Vstream  VstreamSIMD;
-template <typename S, typename T> using ReduceSIMD = Optimization::Reduce<S,T>;
-
-// Arithmetic operations
-typedef Optimization::Sum            SumSIMD;
-typedef Optimization::Sub            SubSIMD;
-typedef Optimization::Div            DivSIMD;
-typedef Optimization::Mult           MultSIMD;
-typedef Optimization::MultComplex    MultComplexSIMD;
-typedef Optimization::MultAddComplex MultAddComplexSIMD;
-typedef Optimization::MultRealPart   MultRealPartSIMD;
-typedef Optimization::MaddRealPart   MaddRealPartSIMD;
-typedef Optimization::Conj           ConjSIMD;
-typedef Optimization::TimesMinusI    TimesMinusISIMD;
-typedef Optimization::TimesI         TimesISIMD;
-
-NAMESPACE_END(Grid)

Grid/simd/Grid_a64fx-fixedsize.h

@@ -1,769 +0,0 @@
-    /*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid
-
-    Source file: Grid_a64fx-fixedsize.h
-
-    Copyright (C) 2020
-
-    Author: Nils Meyer         <nils.meyer@ur.de>           Regensburg University
-
-    with support from Arm
-            Richard Sandiford  <richard.sandiford@arm.com>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-    *************************************************************************************/
-    /*  END LEGAL */
-
-/////////////////////////////////////////////////////
-// Using SVE ACLE with fixed-size data types
-/////////////////////////////////////////////////////
-
-
-// gcc 10 features
-#if __ARM_FEATURE_SVE_BITS==512
-/* gcc 10.0.1 and gcc 10.1 bug using ACLE data types  CAS-159553-Y1K4C6
-   workaround: use gcc's internal data types, bugfix expected for gcc 10.2
-typedef svbool_t    pred __attribute__((arm_sve_vector_bits(512)));
-typedef svfloat16_t vech __attribute__((arm_sve_vector_bits(512)));
-typedef svfloat32_t vecf __attribute__((arm_sve_vector_bits(512)));
-typedef svfloat64_t vecd __attribute__((arm_sve_vector_bits(512)));
-typedef svuint32_t  veci __attribute__((arm_sve_vector_bits(512)));
-typedef svuint32_t  lutf __attribute__((arm_sve_vector_bits(512))); // LUTs for float
-typedef svuint64_t  lutd __attribute__((arm_sve_vector_bits(512))); // LUTs for double
-*/
-typedef __SVBool_t    pred __attribute__((arm_sve_vector_bits(512)));
-typedef __SVFloat16_t vech __attribute__((arm_sve_vector_bits(512)));
-typedef __SVFloat32_t vecf __attribute__((arm_sve_vector_bits(512)));
-typedef __SVFloat64_t vecd __attribute__((arm_sve_vector_bits(512)));
-typedef __SVUint32_t  veci __attribute__((arm_sve_vector_bits(512)));
-typedef __SVUint32_t  lutf __attribute__((arm_sve_vector_bits(512))); // LUTs for float
-typedef __SVUint64_t  lutd __attribute__((arm_sve_vector_bits(512))); // LUTs for double
-#else
-#pragma error("Oops. Illegal SVE vector size!?")
-#endif /* __ARM_FEATURE_SVE_BITS */
-
-// low-level API
-NAMESPACE_BEGIN(Grid);
-NAMESPACE_BEGIN(Optimization);
-
-// convenience union types for tables eliminating loads
-union ulutf {
-  lutf v;
-  uint32_t s[16];
-};
-union ulutd {
-  lutd v;
-  uint64_t s[8];
-};
-
-template <typename T>
-struct acle{};
-
-template <>
-struct acle<double>{
-  static inline lutd tbl_swap(){
-    const ulutd t = { .s = {1, 0, 3, 2, 5, 4, 7, 6} };
-    return t.v;
-  }
-  static inline lutd tbl0(){
-    const ulutd t = { .s = {4, 5, 6, 7, 0, 1, 2, 3} };
-    return t.v;
-  }
-  static inline lutd tbl1(){
-    const ulutd t = { .s = {2, 3, 0, 1, 6, 7, 4, 5} };
-    return t.v;
-  }
-  static inline lutd tbl_exch1a(){ // Exchange1
-    const ulutd t = { .s = {0, 1, 4, 5, 2, 3, 6, 7} };
-    return t.v;
-  }
-  static inline lutd tbl_exch1b(){ // Exchange1
-    const ulutd t = { .s = {2, 3, 6, 7, 0, 1, 4, 5} };
-    return t.v;
-  }
-  static inline lutd tbl_exch1c(){ // Exchange1
-    const ulutd t = { .s = {4, 5, 0, 1, 6, 7, 2, 3} };
-    return t.v;
-  }
-  static inline pred pg1(){return svptrue_b64();}
-  static inline pred pg_even(){return svzip1_b64(svptrue_b64(), svpfalse_b());}
-  static inline pred pg_odd() {return svzip1_b64(svpfalse_b(), svptrue_b64());}
-  static inline vecd zero(){return svdup_f64(0.);}
-};
-
-template <>
-struct acle<float>{
-  // exchange neighboring elements
-  static inline lutf tbl_swap(){
-    const ulutf t = { .s = {1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14} };
-    return t.v;
-  }
-  static inline lutf tbl0(){
-    const ulutf t = { .s = {8, 9, 10, 11, 12, 13, 14, 15, 0, 1, 2, 3, 4, 5, 6, 7} };
-    return t.v;
-  }
-  static inline lutf tbl1(){
-    const ulutf t = { .s = {4, 5, 6, 7, 0, 1, 2, 3, 12, 13, 14, 15, 8, 9, 10, 11} };
-    return t.v;
-  }
-  static inline lutf tbl2(){
-    const ulutf t = { .s = {2, 3, 0, 1, 6, 7, 4, 5, 10, 11, 8, 9, 14, 15, 12, 13} };
-    return t.v;
-  }
-  static inline lutf tbl_exch1a(){ // Exchange1
-    const ulutf t = { .s = {0, 1, 2, 3, 8, 9, 10, 11, 4, 5, 6, 7, 12, 13, 14, 15 } };
-    return t.v;
-  }
-  static inline lutf tbl_exch1b(){ // Exchange1
-    const ulutf t = { .s = {4, 5, 6, 7, 12, 13, 14, 15, 0, 1, 2, 3, 8, 9, 10, 11 } };
-    return t.v;
-  }
-  static inline lutf tbl_exch1c(){ // Exchange1
-    const ulutf t = { .s = {8, 9, 10, 11, 0, 1, 2, 3, 12, 13, 14, 15, 4, 5, 6, 7} };
-    return t.v;
-  }
-  static inline pred pg1(){return svptrue_b32();}
-  static inline pred pg_even(){return svzip1_b32(svptrue_b32(), svpfalse_b());}
-  static inline pred pg_odd() {return svzip1_b32(svpfalse_b(), svptrue_b32());}
-  static inline vecf zero(){return svdup_f32(0.);}
-};
-
-template <>
-struct acle<uint16_t>{
-  static inline pred pg1(){return svptrue_b16();}
-  static inline pred pg_even(){return svzip1_b16(svptrue_b16(), svpfalse_b());}
-  static inline pred pg_odd() {return svzip1_b16(svpfalse_b(), svptrue_b16());}
-  static inline vech zero(){return svdup_f16(0.);}
-};
-
-template <>
-struct acle<Integer>{
-  //static inline svbool_t pg1(){return svptrue_b16();}
-  static inline pred pg1(){return svptrue_b32();}
-  static inline pred pg_even(){return svzip1_b32(svptrue_b32(), svpfalse_b());}
-  static inline pred pg_odd() {return svzip1_b32(svpfalse_b(), svptrue_b32());}
-};
-
-// ---------------------------------------------------
-
-struct Vsplat{
-  // Complex float
-  inline vecf operator()(float a, float b){
-    vecf a_v = svdup_f32(a);
-    vecf b_v = svdup_f32(b);
-    return svzip1(a_v, b_v);
-  }
-  // Real float
-  inline vecf operator()(float a){
-    return svdup_f32(a);
-  }
-  // Complex double
-  inline vecd operator()(double a, double b){
-    vecd a_v = svdup_f64(a);
-    vecd b_v = svdup_f64(b);
-    return svzip1(a_v, b_v);
-  }
-  // Real double
-  inline vecd operator()(double a){
-    return svdup_f64(a);
-  }
-  // Integer
-  inline veci operator()(Integer a){
-    return svdup_u32(a);
-  }
-};
-
-struct Vstore{
-  // Real float
-  inline void operator()(vecf a, float *D){
-    pred pg1 = acle<float>::pg1();
-    svst1(pg1, D, a);
-  }
-  // Real double
-  inline void operator()(vecd a, double *D){
-    pred pg1 = acle<double>::pg1();
-    svst1(pg1, D, a);
-  }
-  // Real float
-  inline void operator()(veci a, Integer *D){
-    pred pg1 = acle<Integer>::pg1();
-    svst1(pg1, D, a);
-  }
-};
-
-struct Vstream{
-  // Real float
-  inline void operator()(float * a, vecf b){
-    pred pg1 = acle<float>::pg1();
-    svstnt1(pg1, a, b);
-    //svst1(pg1, a, b);
-  }
-  // Real double
-  inline void operator()(double * a, vecd b){
-    pred pg1 = acle<double>::pg1();
-    svstnt1(pg1, a, b);
-    //svst1(pg1, a, b);
-  }
-};
-
-struct Vset{
-  // Complex float
-  inline vecf operator()(Grid::ComplexF *a){
-    pred pg1 = acle<float>::pg1();
-    return svld1(pg1, (float*)a);
-  }
-  // Complex double
-  inline vecd operator()(Grid::ComplexD *a){
-    pred pg1 = acle<double>::pg1();
-    return svld1(pg1, (double*)a);
-  }
-  // Real float
-  inline vecf operator()(float *a){
-    pred pg1 = acle<float>::pg1();
-    return svld1(pg1, a);
-  }
-  // Real double
-  inline vecd operator()(double *a){
-    pred pg1 = acle<double>::pg1();
-    return svld1(pg1, a);
-  }
-  // Integer
-  inline veci operator()(Integer *a){
-    pred pg1 = acle<Integer>::pg1();
-    return svld1(pg1, a);
-  }
-};
-
-/////////////////////////////////////////////////////
-// Arithmetic operations
-/////////////////////////////////////////////////////
-
-struct Sum{
-  // Complex/real float
-  inline vecf operator()(vecf a, vecf b){
-    pred pg1 = acle<float>::pg1();
-    return svadd_x(pg1, a, b);
-  }
-  // Complex/real double
-  inline vecd operator()(vecd a, vecd b){
-    pred pg1 = acle<double>::pg1();
-    return svadd_x(pg1, a, b);
-  }
-  // Integer
-  inline veci operator()(veci a, veci b){
-    pred pg1 = acle<Integer>::pg1();
-    return svadd_x(pg1, a, b);
-  }
-};
-
-struct Sub{
-  // Complex/real float
-  inline vecf operator()(vecf a, vecf b){
-    pred pg1 = acle<float>::pg1();
-    return svsub_x(pg1, a, b);
-  }
-  // Complex/real double
-  inline vecd operator()(vecd a, vecd b){
-    pred pg1 = acle<double>::pg1();
-    return svsub_x(pg1, a, b);
-  }
-  // Integer
-  inline veci operator()(veci a, veci b){
-    pred pg1 = acle<Integer>::pg1();
-    return svsub_x(pg1, a, b);
-  }
-
-};
-
-struct Mult{
-  // Real float fma
-  inline vecf operator()(vecf a, vecf b, vecf c){
-    pred pg1 = acle<float>::pg1();
-    return svmad_x(pg1, b, c, a);
-  }
-  // Real double fma
-  inline vecd operator()(vecd a, vecd b, vecd c){
-    pred pg1 = acle<double>::pg1();
-    return svmad_x(pg1, b, c, a);
-  }
-  // Real float
-  inline vecf operator()(vecf a, vecf b){
-    pred pg1 = acle<float>::pg1();
-    return svmul_x(pg1, a, b);
-  }
-  // Real double
-  inline vecd operator()(vecd a, vecd b){
-    pred pg1 = acle<double>::pg1();
-    return svmul_x(pg1, a, b);
-  }
-  // Integer
-  inline veci operator()(veci a, veci b){
-    pred pg1 = acle<Integer>::pg1();
-    return svmul_x(pg1, a, b);
-  }
-};
-
-struct MultRealPart{
-  // Complex float
-  inline vecf operator()(vecf a, vecf b){
-    pred pg1 = acle<float>::pg1();
-    // using FCMLA
-    vecf z_v = acle<float>::zero();
-    return svcmla_x(pg1, z_v, a, b, 0);
-  }
-  // Complex double
-  inline vecd operator()(vecd a, vecd b){
-    pred pg1 = acle<double>::pg1();
-    // using FCMLA
-    vecd z_v = acle<double>::zero();
-    return svcmla_x(pg1, z_v, a, b, 0);
-  }
-};
-
-struct MaddRealPart{
-  // Complex float
-  inline vecf operator()(vecf a, vecf b, vecf c){
-    pred pg1 = acle<float>::pg1();
-    // using FCMLA
-    return svcmla_x(pg1, c, a, b, 0);
-  }
-  // Complex double
-  inline vecd operator()(vecd a, vecd b, vecd c){
-    pred pg1 = acle<double>::pg1();
-    // using FCMLA
-    return svcmla_x(pg1, c, a, b, 0);
-  }
-};
-
-struct MultComplex{
-  // Complex a*b
-  // Complex float
-  inline vecf operator()(vecf a, vecf b){
-    pred pg1 = acle<float>::pg1();
-    vecf z = acle<float>::zero();
-    // using FCMLA
-    vecf r_v = svcmla_x(pg1, z, a, b, 0);
-    return svcmla_x(pg1, r_v, a, b, 90);
-  }
-  // Complex double
-  inline vecd operator()(vecd a, vecd b){
-    pred pg1 = acle<double>::pg1();
-    vecd z = acle<double>::zero();
-    // using FCMLA
-    vecd r_v = svcmla_x(pg1, z, a, b, 0);
-    return svcmla_x(pg1, r_v, a, b, 90);
-  }
-};
-
-struct MultAddComplex{
-  // Complex a*b+c
-  // Complex float
-  inline vecf operator()(vecf a, vecf b, vecf c){
-    pred pg1 = acle<float>::pg1();
-    // using FCMLA
-    vecf r_v = svcmla_x(pg1, c, a, b, 0);
-    return svcmla_x(pg1, r_v, a, b, 90);
-  }
-  // Complex double
-  inline vecd operator()(vecd a, vecd b, vecd c){
-    pred pg1 = acle<double>::pg1();
-    // using FCMLA
-    vecd r_v = svcmla_x(pg1, c, a, b, 0);
-    return svcmla_x(pg1, r_v, a, b, 90);
-  }
-};
-
-struct Div{
-  // Real float
-  inline vecf operator()(vecf a, vecf b){
-    pred pg1 = acle<float>::pg1();
-    return svdiv_x(pg1, a, b);
-  }
-  // Real double
-  inline vecd operator()(vecd a, vecd b){
-    pred pg1 = acle<double>::pg1();
-    return svdiv_x(pg1, a, b);
-  }
-};
-
-struct Conj{
-  // Complex float
-  inline vecf operator()(vecf a){
-    pred pg_odd = acle<float>::pg_odd();
-    //return svneg_x(pg_odd, a);  this is unsafe
-    return svneg_m(a, pg_odd, a);
-  }
-  // Complex double
-  inline vecd operator()(vecd a){
-    pred pg_odd = acle<double>::pg_odd();
-    //return svneg_x(pg_odd, a);  this is unsafe
-    return svneg_m(a, pg_odd, a);
-  }
-};
-
-struct TimesMinusI{
-  // Complex float
-  inline vecf operator()(vecf a, vecf b){
-    lutf tbl_swap = acle<float>::tbl_swap();
-    pred pg1 = acle<float>::pg1();
-    pred pg_odd = acle<float>::pg_odd();
-
-    vecf a_v = svtbl(a, tbl_swap);
-    //return svneg_x(pg_odd, a_v);  this is unsafe
-    return svneg_m(a_v, pg_odd, a_v);
-  }
-  // Complex double
-  inline vecd operator()(vecd a, vecd b){
-    lutd tbl_swap = acle<double>::tbl_swap();
-    pred pg1 = acle<double>::pg1();
-    pred pg_odd = acle<double>::pg_odd();
-
-    vecd a_v = svtbl(a, tbl_swap);
-    //return svneg_x(pg_odd, a_v);  this is unsafe
-    return svneg_m(a_v, pg_odd, a_v);
-  }
-};
-
-struct TimesI{
-  // Complex float
-  inline vecf operator()(vecf a, vecf b){
-    lutf tbl_swap = acle<float>::tbl_swap();
-    pred pg1 = acle<float>::pg1();
-    pred pg_even = acle<float>::pg_even();
-
-    vecf a_v = svtbl(a, tbl_swap);
-    //return svneg_x(pg_even, a_v);  this is unsafe
-    return svneg_m(a_v, pg_even, a_v);
-  }
-  // Complex double
-  inline vecd operator()(vecd a, vecd b){
-    lutd tbl_swap = acle<double>::tbl_swap();
-    pred pg1 = acle<double>::pg1();
-    pred pg_even = acle<double>::pg_even();
-
-    vecd a_v = svtbl(a, tbl_swap);
-    //return svneg_x(pg_even, a_v);  this is unsafe
-    return svneg_m(a_v, pg_even, a_v);
-  }
-};
-
-struct PrecisionChange {
-  static inline vech StoH (vecf sa, vecf sb) {
-    pred pg1s = acle<float>::pg1();
-    vech ha_v = svcvt_f16_x(pg1s, sa);
-    vech hb_v = svcvt_f16_x(pg1s, sb);
-    return svuzp1(ha_v, hb_v);
-  }
-  static inline void HtoS(vech h,vecf &sa,vecf &sb) {
-    pred pg1s = acle<float>::pg1();
-    vech ha_v = svzip1(h, h);
-    vech hb_v = svzip2(h, h);
-    sa = svcvt_f32_x(pg1s, ha_v);
-    sb = svcvt_f32_x(pg1s, hb_v);
-  }
-  static inline vecf DtoS (vecd a,vecd b) {
-    pred pg1d = acle<double>::pg1();
-    vecf sa_v = svcvt_f32_x(pg1d, a);
-    vecf sb_v = svcvt_f32_x(pg1d, b);
-    return svuzp1(sa_v, sb_v);
-  }
-  static inline void StoD (vecf s,vecd &a,vecd &b) {
-    pred pg1d = acle<double>::pg1();
-    vecf sa_v = svzip1(s, s);
-    vecf sb_v = svzip2(s, s);
-    a = svcvt_f64_x(pg1d, sa_v);
-    b = svcvt_f64_x(pg1d, sb_v);
-  }
-  static inline vech DtoH (vecd a,vecd b,vecd c,vecd d) {
-    pred pg1d = acle<double>::pg1();
-    pred pg1h = acle<uint16_t>::pg1();
-    vech ha_v = svcvt_f16_x(pg1d, a);
-    vech hb_v = svcvt_f16_x(pg1d, b);
-    vech hc_v = svcvt_f16_x(pg1d, c);
-    vech hd_v = svcvt_f16_x(pg1d, d);
-    vech hab_v = svuzp1(ha_v, hb_v);
-    vech hcd_v = svuzp1(hc_v, hd_v);
-    return svuzp1(hab_v, hcd_v);
-
-/*
-    vecf sa,sb;
-    sa = DtoS(a,b);
-    sb = DtoS(c,d);
-    return StoH(sa,sb);
-*/
-  }
-  static inline void HtoD(vech h,vecd &a,vecd &b,vecd &c,vecd &d) {
-    pred pg1h = acle<uint16_t>::pg1();
-    pred pg1d = acle<double>::pg1();
-    vech sa_v = svzip1(h, h);
-    vech sb_v = svzip2(h, h);
-    vech da_v = svzip1(sa_v, sa_v);
-    vech db_v = svzip2(sa_v, sa_v);
-    vech dc_v = svzip1(sb_v, sb_v);
-    vech dd_v = svzip2(sb_v, sb_v);
-    a = svcvt_f64_x(pg1d, da_v);
-    b = svcvt_f64_x(pg1d, db_v);
-    c = svcvt_f64_x(pg1d, dc_v);
-    d = svcvt_f64_x(pg1d, dd_v);
-
-/*
-    vecf sa,sb;
-    HtoS(h,sa,sb);
-    StoD(sa,a,b);
-    StoD(sb,c,d);
-*/
-  }
-};
-
-struct Exchange{
-  // float
-  static inline void Exchange0(vecf &out1, vecf &out2, vecf in1, vecf in2){
-    vecf r1_v = svext(in1, in1, (uint64_t)8u);
-    vecf r2_v = svext(in2, in2, (uint64_t)8u);
-    out1 = svext(r1_v, in2, (uint64_t)8u);
-    out2 = svext(in1, r2_v, (uint64_t)8u);
-  }
-  static inline void Exchange1(vecf &out1, vecf &out2, vecf in1, vecf in2){
-    // this one is tricky; svtrn2q* from SVE2 fits best, but it is not available in SVE1
-    // alternative: use 4-el structure; expect translation into 4x ldp + 4x stp -> SFI
-    lutf tbl_exch1a = acle<float>::tbl_exch1a();
-    lutf tbl_exch1b = acle<float>::tbl_exch1b();
-    lutf tbl_exch1c = acle<float>::tbl_exch1c();
-
-    vecf a1_v = svtbl(in1, tbl_exch1a);
-    vecf a2_v = svtbl(in2, tbl_exch1b);
-    vecf b1_v  = svext(a2_v, a1_v, (uint64_t)8u);
-    vecf b2_v  = svext(a1_v, a2_v, (uint64_t)8u);
-    out1 = svtbl(b1_v, tbl_exch1c);
-    out2 = svtbl(b2_v, tbl_exch1a);
-  }
-  static inline void Exchange2(vecf &out1, vecf &out2, vecf in1, vecf in2){
-    out1 = (vecf)svtrn1((vecd)in1, (vecd)in2);
-    out2 = (vecf)svtrn2((vecd)in1, (vecd)in2);
-  }
-  static inline void Exchange3(vecf &out1, vecf &out2, vecf in1, vecf in2){
-    out1 = svtrn1(in1, in2);
-    out2 = svtrn2(in1, in2);
-  }
-
-  // double
-  static inline void Exchange0(vecd &out1, vecd &out2, vecd in1, vecd in2){
-    vecd r1_v = svext(in1, in1, (uint64_t)4u);
-    vecd r2_v = svext(in2, in2, (uint64_t)4u);
-    out1 = svext(r1_v, in2, (uint64_t)4u);
-    out2 = svext(in1, r2_v, (uint64_t)4u);
-  }
-  static inline void Exchange1(vecd &out1, vecd &out2, vecd in1, vecd in2){
-    // this one is tricky; svtrn2q* from SVE2 fits best, but it is not available in SVE1
-    // alternative: use 4-el structure; expect translation into 4x ldp + 4x stp -> SFI
-    lutd tbl_exch1a = acle<double>::tbl_exch1a();
-    lutd tbl_exch1b = acle<double>::tbl_exch1b();
-    lutd tbl_exch1c = acle<double>::tbl_exch1c();
-
-    vecd a1_v = svtbl(in1, tbl_exch1a);
-    vecd a2_v = svtbl(in2, tbl_exch1b);
-    vecd b1_v = svext(a2_v, a1_v, (uint64_t)4u);
-    vecd b2_v = svext(a1_v, a2_v, (uint64_t)4u);
-    out1 = svtbl(b1_v, tbl_exch1c);
-    out2 = svtbl(b2_v, tbl_exch1a);
-  }
-  static inline void Exchange2(vecd &out1, vecd &out2, vecd in1, vecd in2){
-    out1 = svtrn1(in1, in2);
-    out2 = svtrn2(in1, in2);
-  }
-  static inline void Exchange3(vecd &out1, vecd &out2, vecd in1, vecd in2){
-    assert(0);
-    return;
-  }
-};
-
-#undef VECTOR_FOR
-
-struct Permute{
-  // float
-  static inline vecf Permute0(vecf in) {
-    return svext(in, in, (uint64_t)8u);
-  }
-  static inline vecf Permute1(vecf in) {
-    lutf tbl_swap = acle<float>::tbl1();
-    return svtbl(in, tbl_swap);
-  }
-  static inline vecf Permute2(vecf in) {
-    lutf tbl_swap = acle<float>::tbl2();
-    return svtbl(in, tbl_swap);
-  }
-  static inline vecf Permute3(vecf in) {
-    lutf tbl_swap = acle<float>::tbl_swap();
-    return svtbl(in, tbl_swap);
-  }
-
-  // double
-  static inline vecd Permute0(vecd in) {
-    return svext(in, in, (uint64_t)(8u / 2u));
-  }
-  static inline vecd Permute1(vecd in) {
-    lutd tbl_swap = acle<double>::tbl1();
-    return svtbl(in, tbl_swap);
-  }
-  static inline vecd Permute2(vecd in) {
-    lutd tbl_swap = acle<double>::tbl_swap();
-    return svtbl(in, tbl_swap);
-  }
-  static inline vecd Permute3(vecd in) {
-    return in;
-  }
-};
-
-struct Rotate{
-
-  static inline vecf rotate(vecf in, int n){
-    switch(n){
-    case 0:  return tRotate<0>(in); break;
-    case 1:  return tRotate<1>(in); break;
-    case 2:  return tRotate<2>(in); break;
-    case 3:  return tRotate<3>(in); break;
-    case 4:  return tRotate<4>(in); break;
-    case 5:  return tRotate<5>(in); break;
-    case 6:  return tRotate<6>(in); break;
-    case 7:  return tRotate<7>(in); break;
-
-    case 8:  return tRotate<8>(in); break;
-    case 9:  return tRotate<9>(in); break;
-    case 10: return tRotate<10>(in); break;
-    case 11: return tRotate<11>(in); break;
-    case 12: return tRotate<12>(in); break;
-    case 13: return tRotate<13>(in); break;
-    case 14: return tRotate<14>(in); break;
-    case 15: return tRotate<15>(in); break;
-    default: assert(0);
-    }
-  }
-  static inline vecd rotate(vecd in, int n){
-    switch(n){
-    case 0:  return tRotate<0>(in); break;
-    case 1:  return tRotate<1>(in); break;
-    case 2:  return tRotate<2>(in); break;
-    case 3:  return tRotate<3>(in); break;
-    case 4:  return tRotate<4>(in); break;
-    case 5:  return tRotate<5>(in); break;
-    case 6:  return tRotate<6>(in); break;
-    case 7:  return tRotate<7>(in); break;
-    default: assert(0);
-    }
-  }
-
-  template <int n> static inline vecf tRotate(vecf in){
-    return svext(in, in, (uint64_t)n);
-  }
-  template <int n> static inline vecd tRotate(vecd in){
-    return svext(in, in, (uint64_t)n);
-  }
-};
-
-// tree-based reduction
-#define svred(pg, v)\
-svaddv(pg, v);
-
-// left-to-right reduction
-// #define svred(pg, v)\
-// svadda(pg, 0, v)
-
-template <typename Out_type, typename In_type>
-struct Reduce{
-  //Need templated class to overload output type
-  //General form must generate error if compiled
-  inline Out_type operator()(In_type in){
-    printf("Error, using wrong Reduce function\n");
-    //exit(1);
-    return 0;
-  }
-};
-//Complex float Reduce
-template <>
-inline Grid::ComplexF Reduce<Grid::ComplexF, vecf>::operator()(vecf in){
-  pred pg_even = acle<float>::pg_even();
-  pred pg_odd  = acle<float>::pg_odd();
-  float a = svred(pg_even, in);
-  float b = svred(pg_odd, in);
-  return Grid::ComplexF(a, b);
-}
-//Real float Reduce
-template <>
-inline Grid::RealF Reduce<Grid::RealF, vecf>::operator()(vecf in){
-  pred pg1 = acle<float>::pg1();
-  return svred(pg1, in);
-}
-//Complex double Reduce
-template <>
-inline Grid::ComplexD Reduce<Grid::ComplexD, vecd>::operator()(vecd in){
-  pred pg_even = acle<double>::pg_even();
-  pred pg_odd  = acle<double>::pg_odd();
-  double a = svred(pg_even, in);
-  double b = svred(pg_odd, in);
-  return Grid::ComplexD(a, b);
-}
-//Real double Reduce
-template <>
-inline Grid::RealD Reduce<Grid::RealD, vecd>::operator()(vecd in){
-  pred pg1 = acle<double>::pg1();
-  return svred(pg1, in);
-}
-//Integer Reduce
-template <>
-inline Integer Reduce<Integer, veci>::operator()(veci in){
-  pred pg1 = acle<Integer>::pg1();
-  return svred(pg1, in);
-}
-
-#undef svred
-
-NAMESPACE_END(Optimization);
-
-//////////////////////////////////////////////////////////////////////////////////////
-// Here assign types
-
-typedef vech SIMD_Htype; // Reduced precision type
-typedef vecf SIMD_Ftype; // Single precision type
-typedef vecd SIMD_Dtype; // Double precision type
-typedef veci SIMD_Itype; // Integer type
-
-// prefetch utilities
-inline void v_prefetch0(int size, const char *ptr){};
-inline void prefetch_HINT_T0(const char *ptr){};
-
-// Function name aliases
-typedef Optimization::Vsplat   VsplatSIMD;
-typedef Optimization::Vstore   VstoreSIMD;
-typedef Optimization::Vset     VsetSIMD;
-typedef Optimization::Vstream  VstreamSIMD;
-template <typename S, typename T> using ReduceSIMD = Optimization::Reduce<S,T>;
-
-// Arithmetic operations
-typedef Optimization::Sum            SumSIMD;
-typedef Optimization::Sub            SubSIMD;
-typedef Optimization::Div            DivSIMD;
-typedef Optimization::Mult           MultSIMD;
-typedef Optimization::MultComplex    MultComplexSIMD;
-typedef Optimization::MultAddComplex MultAddComplexSIMD;
-typedef Optimization::MultRealPart   MultRealPartSIMD;
-typedef Optimization::MaddRealPart   MaddRealPartSIMD;
-typedef Optimization::Conj           ConjSIMD;
-typedef Optimization::TimesMinusI    TimesMinusISIMD;
-typedef Optimization::TimesI         TimesISIMD;
-
-NAMESPACE_END(Grid);

Grid/simd/Grid_vector_types.h

@@ -1,6 +1,6 @@
 /*************************************************************************************
 
-    Grid physics library, www.github.com/paboyle/Grid
+    Grid physics library, www.github.com/paboyle/Grid 
 
     Source file: ./lib/simd/Grid_vector_types.h
 
@@ -73,7 +73,7 @@ accelerator_inline Grid_half sfw_float_to_half(float ff) {
   const FP32 denorm_magic = { ((127 - 15) + (23 - 10) + 1) << 23 };
   unsigned int sign_mask = 0x80000000u;
   Grid_half o;
-
+    
   o.x = static_cast<unsigned short>(0x0u);
   unsigned int sign = f.u & sign_mask;
   f.u ^= sign;
@@ -93,7 +93,7 @@ accelerator_inline Grid_half sfw_float_to_half(float ff) {
       o.x = static_cast<unsigned short>(f.u - denorm_magic.u);
     } else {
       unsigned int mant_odd = (f.u >> 13) & 1; // resulting mantissa is odd
-
+	
       // update exponent, rounding bias part 1
       f.u += ((unsigned int)(15 - 127) << 23) + 0xfff;
       // rounding bias part 2
@@ -101,7 +101,7 @@ accelerator_inline Grid_half sfw_float_to_half(float ff) {
       // take the bits!
       o.x = static_cast<unsigned short>(f.u >> 13);
     }
-  }
+  } 
   o.x |= static_cast<unsigned short>(sign >> 16);
   return o;
 }
@@ -110,63 +110,9 @@ accelerator_inline Grid_half sfw_float_to_half(float ff) {
 #ifdef GPU_VEC
 #include "Grid_gpu_vec.h"
 #endif
-/*
 #ifdef GEN
 #include "Grid_generic.h"
 #endif
-*/
-
-#ifdef GEN
-  #if defined(A64FX) || defined(A64FXFIXEDSIZE) // breakout A64FX SVE ACLE here
-    #include <arm_sve.h>
-    #if defined(A64FX) // VLA
-      #pragma message("building A64FX / SVE ACLE VLA")
-      #if defined(ARMCLANGCOMPAT)
-        #pragma message("applying data types patch")
-      #endif
-      #include "Grid_a64fx-2.h"
-    #endif
-    #if defined(A64FXFIXEDSIZE) // fixed size data types
-      #pragma message("building for A64FX / SVE ACLE fixed size")
-      #include "Grid_a64fx-fixedsize.h"
-    #endif
-  #else
-    //#pragma message("building GEN") // generic
-    #include "Grid_generic.h"
-  #endif
-#endif
-
-#ifdef A64FX
-  #include <arm_sve.h>
-  #ifdef __ARM_FEATURE_SVE_BITS
-    //#pragma message("building A64FX SVE VLS")
-    #include "Grid_a64fx-fixedsize.h"
-  #else
-    #pragma message("building A64FX SVE VLA")
-    #if defined(ARMCLANGCOMPAT)
-      #pragma message("applying data types patch")
-    #endif
-    #include "Grid_a64fx-2.h"
-  #endif
-#endif
-
-/*
-#ifdef A64FXVLA
-#pragma message("building A64FX VLA")
-#if defined(ARMCLANGCOMPAT)
-  #pragma message("applying data types patch")
-#endif
-#include <arm_sve.h>
-#include "Grid_a64fx-2.h"
-#endif
-
-#ifdef A64FXVLS
-#pragma message("building A64FX VLS")
-#include <arm_sve.h>
-#include "Grid_a64fx-fixedsize.h"
-#endif
-*/
-
 #ifdef SSE4
 #include "Grid_sse4.h"
 #endif
@@ -217,12 +163,6 @@ template <typename T> struct is_complex : public std::false_type {};
 template <> struct is_complex<ComplexD> : public std::true_type {};
 template <> struct is_complex<ComplexF> : public std::true_type {};
 
-template <typename T> struct is_ComplexD : public std::false_type {};
-template <> struct is_ComplexD<ComplexD> : public std::true_type {};
-
-template <typename T> struct is_ComplexF : public std::false_type {};
-template <> struct is_ComplexF<ComplexF> : public std::true_type {};
-
 template<typename T, typename V=void> struct is_real : public std::false_type {};
 template<typename T> struct is_real<T, typename std::enable_if<std::is_floating_point<T>::value,
   void>::type> : public std::true_type {};
@@ -230,7 +170,7 @@ template<typename T> struct is_real<T, typename std::enable_if<std::is_floating_
 template<typename T, typename V=void> struct is_integer : public std::false_type {};
 template<typename T> struct is_integer<T, typename std::enable_if<std::is_integral<T>::value,
   void>::type> : public std::true_type {};
-
+  
 template <typename T>              using IfReal    = Invoke<std::enable_if<is_real<T>::value, int> >;
 template <typename T>              using IfComplex = Invoke<std::enable_if<is_complex<T>::value, int> >;
 template <typename T>              using IfInteger = Invoke<std::enable_if<is_integer<T>::value, int> >;
@@ -283,69 +223,6 @@ public:
     return sizeof(Vector_type) / sizeof(Scalar_type);
   }
 
-  #ifdef ARMCLANGCOMPAT
-    template <class S = Scalar_type>
-    accelerator_inline Grid_simd &operator=(const Grid_simd<typename std::enable_if<!is_complex<S>::value, S>::type, Vector_type> &&rhs) {
-      //v = rhs.v;
-      svst1(svptrue_b8(), (Scalar_type*)this, svld1(svptrue_b8(), (Scalar_type*)&(rhs.v)));
-      return *this;
-    };
-
-    template <class S = Scalar_type>
-    accelerator_inline Grid_simd &operator=(const Grid_simd<typename std::enable_if<!is_complex<S>::value, S>::type, Vector_type> &rhs) {
-      //v = rhs.v;
-      svst1(svptrue_b8(), (Scalar_type*)this, svld1(svptrue_b8(), (Scalar_type*)&(rhs.v)));
-      return *this;
-    };
-
-    /*
-    template <class S = Scalar_type>
-    accelerator_inline Grid_simd &operator=(const Grid_simd<typename std::enable_if<is_complex<S>::value, S>::type, Vector_type> &&rhs) {
-      //v = rhs.v;
-      svst1(svptrue_b8(), (int8_t*)this, svld1(svptrue_b8(), (int8_t*)&(rhs.v)));
-      return *this;
-    };
-
-    template <class S = Scalar_type>
-    accelerator_inline Grid_simd &operator=(const Grid_simd<typename std::enable_if<is_complex<S>::value, S>::type, Vector_type> &rhs) {
-      //v = rhs.v;
-      svst1(svptrue_b8(), (int8_t*)this, svld1(svptrue_b8(), (int8_t*)&(rhs.v)));
-      return *this;
-    };
-    */
-
-    // ComplexF
-    template <class S = Scalar_type>
-    accelerator_inline Grid_simd &operator=(const Grid_simd<typename std::enable_if<is_ComplexF<S>::value, S>::type, Vector_type> &&rhs) {
-      //v = rhs.v;
-      svst1(svptrue_b32(), (float*)this, svld1(svptrue_b32(), (float*)&(rhs.v)));
-      return *this;
-    };
-
-    template <class S = Scalar_type>
-    accelerator_inline Grid_simd &operator=(const Grid_simd<typename std::enable_if<is_ComplexF<S>::value, S>::type, Vector_type> &rhs) {
-      //v = rhs.v;
-      svst1(svptrue_b32(), (float*)this, svld1(svptrue_b32(), (float*)&(rhs.v)));
-      return *this;
-    };
-
-    // ComplexD
-    template <class S = Scalar_type>
-    accelerator_inline Grid_simd &operator=(const Grid_simd<typename std::enable_if<is_ComplexD<S>::value, S>::type, Vector_type> &&rhs) {
-      //v = rhs.v;
-      svst1(svptrue_b64(), (double*)this, svld1(svptrue_b64(), (double*)&(rhs.v)));
-      return *this;
-    };
-
-    template <class S = Scalar_type>
-    accelerator_inline Grid_simd &operator=(const Grid_simd<typename std::enable_if<is_ComplexD<S>::value, S>::type, Vector_type> &rhs) {
-      //v = rhs.v;
-      svst1(svptrue_b64(), (double*)this, svld1(svptrue_b64(), (double*)&(rhs.v)));
-      return *this;
-    };
-
-  #else
-
   accelerator_inline Grid_simd &operator=(const Grid_simd &&rhs) {
     v = rhs.v;
     return *this;
@@ -355,23 +232,10 @@ public:
     return *this;
   };  // faster than not declaring it and leaving to the compiler
 
-  #endif
 
   accelerator Grid_simd() = default;
-
-  #ifdef ARMCLANGCOMPAT
-    template <class S = Scalar_type>
-    accelerator_inline Grid_simd(const Grid_simd<typename std::enable_if<!is_complex<S>::value, S>::type, Vector_type> &rhs) { this->operator=(rhs); }
-    template <class S = Scalar_type>
-    accelerator_inline Grid_simd(const Grid_simd<typename std::enable_if<!is_complex<S>::value, S>::type, Vector_type> &&rhs) { this->operator=(rhs); }
-    template <class S = Scalar_type>
-    accelerator_inline Grid_simd(const Grid_simd<typename std::enable_if<is_complex<S>::value, S>::type, Vector_type> &rhs) { this->operator=(rhs); }
-    template <class S = Scalar_type>
-    accelerator_inline Grid_simd(const Grid_simd<typename std::enable_if<is_complex<S>::value, S>::type, Vector_type> &&rhs) { this->operator=(rhs); }
-  #else
-    accelerator_inline Grid_simd(const Grid_simd &rhs) : v(rhs.v){};  // compiles in movaps
-    accelerator_inline Grid_simd(const Grid_simd &&rhs) : v(rhs.v){};
-  #endif
+  accelerator_inline Grid_simd(const Grid_simd &rhs) : v(rhs.v){};  // compiles in movaps
+  accelerator_inline Grid_simd(const Grid_simd &&rhs) : v(rhs.v){};
   accelerator_inline Grid_simd(const Real a) { vsplat(*this, Scalar_type(a)); };
   // Enable if complex type
   template <typename S = Scalar_type> accelerator_inline
@@ -394,21 +258,12 @@ public:
   ///////////////////////////////////////////////
 
   // FIXME -- alias this to an accelerator_inline MAC struct.
-
-  #if defined(A64FX) || defined(A64FXFIXEDSIZE)
-  friend accelerator_inline void mac(Grid_simd *__restrict__ y,
-				     const Grid_simd *__restrict__ a,
-				     const Grid_simd *__restrict__ x) {
-    *y = fxmac((*a), (*x), (*y));
-  };
-  #else
   friend accelerator_inline void mac(Grid_simd *__restrict__ y,
 				     const Grid_simd *__restrict__ a,
 				     const Grid_simd *__restrict__ x) {
     *y = (*a) * (*x) + (*y);
   };
-  #endif
-
+  
   friend accelerator_inline void mult(Grid_simd *__restrict__ y,
 				      const Grid_simd *__restrict__ l,
 				      const Grid_simd *__restrict__ r) {
@@ -557,7 +412,7 @@ public:
     Grid_simd ret;
     Grid_simd::conv_t conv;
     Grid_simd::scalar_type s;
-
+    
     conv.v = v.v;
     for (int i = 0; i < Nsimd(); i++) {
       s = conv.s[i];
@@ -586,7 +441,7 @@ public:
     return ret;
   }
   ///////////////////////
-  // Exchange
+  // Exchange 
   // Al Ah , Bl Bh -> Al Bl Ah,Bh
   ///////////////////////
   friend accelerator_inline void exchange(Grid_simd &out1,Grid_simd &out2,Grid_simd in1,Grid_simd in2,int n)
@@ -597,20 +452,20 @@ public:
       Optimization::Exchange::Exchange2(out1.v,out2.v,in1.v,in2.v);
     } else if(n==1) {
       Optimization::Exchange::Exchange1(out1.v,out2.v,in1.v,in2.v);
-    } else if(n==0) {
+    } else if(n==0) { 
       Optimization::Exchange::Exchange0(out1.v,out2.v,in1.v,in2.v);
     }
   }
-  friend accelerator_inline void exchange0(Grid_simd &out1,Grid_simd &out2,Grid_simd in1,Grid_simd in2){
+  friend accelerator_inline void exchange0(Grid_simd &out1,Grid_simd &out2,Grid_simd in1,Grid_simd in2){    
     Optimization::Exchange::Exchange0(out1.v,out2.v,in1.v,in2.v);
   }
-  friend accelerator_inline void exchange1(Grid_simd &out1,Grid_simd &out2,Grid_simd in1,Grid_simd in2){
+  friend accelerator_inline void exchange1(Grid_simd &out1,Grid_simd &out2,Grid_simd in1,Grid_simd in2){    
     Optimization::Exchange::Exchange1(out1.v,out2.v,in1.v,in2.v);
   }
-  friend accelerator_inline void exchange2(Grid_simd &out1,Grid_simd &out2,Grid_simd in1,Grid_simd in2){
+  friend accelerator_inline void exchange2(Grid_simd &out1,Grid_simd &out2,Grid_simd in1,Grid_simd in2){    
     Optimization::Exchange::Exchange2(out1.v,out2.v,in1.v,in2.v);
   }
-  friend accelerator_inline void exchange3(Grid_simd &out1,Grid_simd &out2,Grid_simd in1,Grid_simd in2){
+  friend accelerator_inline void exchange3(Grid_simd &out1,Grid_simd &out2,Grid_simd in1,Grid_simd in2){    
     Optimization::Exchange::Exchange3(out1.v,out2.v,in1.v,in2.v);
   }
   ////////////////////////////////////////////////////////////////////
@@ -635,7 +490,7 @@ public:
       int dist = perm & 0xF;
       y = rotate(b, dist);
       return;
-    }
+    } 
     else if(perm==3) permute3(y, b);
     else if(perm==2) permute2(y, b);
     else if(perm==1) permute1(y, b);
@@ -709,29 +564,29 @@ accelerator_inline Grid_simd<S, V> rotate(Grid_simd<S, V> b, int nrot) {
   ret.v = Optimization::Rotate::rotate(b.v, 2 * nrot);
   return ret;
 }
-template <class S, class V, IfNotComplex<S> =0>
+template <class S, class V, IfNotComplex<S> =0> 
 accelerator_inline void rotate( Grid_simd<S,V> &ret,Grid_simd<S,V> b,int nrot)
 {
   nrot = nrot % Grid_simd<S,V>::Nsimd();
   ret.v = Optimization::Rotate::rotate(b.v,nrot);
 }
-template <class S, class V, IfComplex<S> =0>
+template <class S, class V, IfComplex<S> =0> 
 accelerator_inline void rotate(Grid_simd<S,V> &ret,Grid_simd<S,V> b,int nrot)
 {
   nrot = nrot % Grid_simd<S,V>::Nsimd();
   ret.v = Optimization::Rotate::rotate(b.v,2*nrot);
 }
 
-template <class S, class V>
+template <class S, class V> 
 accelerator_inline void vbroadcast(Grid_simd<S,V> &ret,const Grid_simd<S,V> &src,int lane){
   S* typepun =(S*) &src;
   vsplat(ret,typepun[lane]);
-}
-template <class S, class V, IfComplex<S> =0>
+}    
+template <class S, class V, IfComplex<S> =0> 
 accelerator_inline void rbroadcast(Grid_simd<S,V> &ret,const Grid_simd<S,V> &src,int lane){
   S* typepun =(S*) &src;
   ret.v = unary<V>(real(typepun[lane]), VsplatSIMD());
-}
+}    
 
 
 
@@ -886,27 +741,6 @@ accelerator_inline Grid_simd<S, V> operator*(Grid_simd<S, V> a, Grid_simd<S, V>
   return ret;
 };
 
-// ---------------- A64FX MAC -------------------
-// Distinguish between complex types and others
-#if defined(A64FX) || defined(A64FXFIXEDSIZE)
-template <class S, class V, IfComplex<S> = 0>
-accelerator_inline Grid_simd<S, V> fxmac(Grid_simd<S, V> a, Grid_simd<S, V> b, Grid_simd<S, V> c) {
-  Grid_simd<S, V> ret;
-  ret.v = trinary<V>(a.v, b.v, c.v, MultAddComplexSIMD());
-  return ret;
-};
-
-// Real/Integer types
-template <class S, class V, IfNotComplex<S> = 0>
-accelerator_inline Grid_simd<S, V> fxmac(Grid_simd<S, V> a, Grid_simd<S, V> b, Grid_simd<S, V> c) {
-  Grid_simd<S, V> ret;
-  ret.v = trinary<V>(a.v, b.v, c.v, MultSIMD());
-  return ret;
-};
-#endif
-// ----------------------------------------------
-
-
 // Distinguish between complex types and others
 template <class S, class V, IfComplex<S> = 0>
 accelerator_inline Grid_simd<S, V> operator/(Grid_simd<S, V> a, Grid_simd<S, V> b) {
@@ -1043,7 +877,7 @@ accelerator_inline typename toComplexMapper<Rsimd>::Complexified toComplex(const
 
   conv.v = in.v;
   for (int i = 0; i < Rsimd::Nsimd(); i += 2) {
-    assert(conv.s[i + 1] == conv.s[i]);
+    assert(conv.s[i + 1] == conv.s[i]);  
     // trap any cases where real was not duplicated
     // indicating the SIMD grids of real and imag assignment did not correctly
     // match
@@ -1085,14 +919,6 @@ accelerator_inline void precisionChange(vRealD    *out,vRealF    *in,int nvec)
   for(int m=0;m*2<nvec;m++){
     int n=m*2;
     Optimization::PrecisionChange::StoD(in[m].v,out[n].v,out[n+1].v);
-    // Bug in gcc 10.0.1 and gcc 10.1 using fixed-size SVE ACLE data types  CAS-159553-Y1K4C6
-    // function call results in compile-time error:
-    // In function ‘void Grid::precisionChange(Grid::vRealD*, Grid::vRealF*, int)’:
-    // .../Grid_vector_types.h:961:56: error:
-    // cannot bind non-const lvalue reference of type ‘vecd&’ {aka ‘svfloat64_t&’}
-    // to an rvalue of type ‘vecd’ {aka ‘svfloat64_t’}
-    // 961 |     Optimization::PrecisionChange::StoD(in[m].v,out[n].v,out[n+1].v);
-    //  |                                                 ~~~~~~~^
   }
 }
 accelerator_inline void precisionChange(vRealD    *out,vRealH    *in,int nvec)

Grid/simd/Simd.h

@@ -93,11 +93,6 @@ accelerator_inline ComplexF pow(const ComplexF& r,RealF y){ return(std::pow(r,y)
 using std::abs;
 using std::pow;
 using std::sqrt;
-using std::log;
-using std::exp;
-using std::sin;
-using std::cos;
-
 
 accelerator_inline RealF    conjugate(const RealF  & r){ return r; }
 accelerator_inline RealD    conjugate(const RealD  & r){ return r; }

Grid/stencil/Stencil.h

@@ -1,6 +1,6 @@
 /*************************************************************************************
 
-     Grid physics library, www.github.com/paboyle/Grid
+     Grid physics library, www.github.com/paboyle/Grid 
 
      Source file: ./lib/Stencil.h
 
@@ -41,13 +41,13 @@
 // Stencil based code will exchange haloes and use a table lookup for neighbours.
 // This will be done with generality to allow easier efficient implementations.
 // Overlap of comms and compute is enabled by tabulating off-node connected,
-//
+// 
 // Generic services
 // 0) Prebuild neighbour tables
 // 1) Compute sizes of all haloes/comms buffers; allocate them.
 // 2) Gather all faces, and communicate.
 // 3) Loop over result sites, giving nbr index/offnode info for each
-//
+// 
 //////////////////////////////////////////////////////////////////////////////////////////
 
 NAMESPACE_BEGIN(Grid);
@@ -59,10 +59,10 @@ NAMESPACE_BEGIN(Grid);
 void Gather_plane_table_compute (GridBase *grid,int dimension,int plane,int cbmask,
 				 int off,Vector<std::pair<int,int> > & table);
 
-template<class vobj,class cobj,class compressor>
+template<class vobj,class cobj,class compressor> 
 void Gather_plane_simple_table (Vector<std::pair<int,int> >& table,const Lattice<vobj> &rhs,cobj *buffer,compressor &compress, int off,int so)   __attribute__((noinline));
 
-template<class vobj,class cobj,class compressor>
+template<class vobj,class cobj,class compressor> 
 void Gather_plane_simple_table (Vector<std::pair<int,int> >& table,const Lattice<vobj> &rhs,cobj *buffer,compressor &compress, int off,int so)
 {
   int num=table.size();
@@ -94,13 +94,13 @@ void Gather_plane_exchange_table(Vector<std::pair<int,int> >& table,const Lattic
 {
   assert( (table.size()&0x1)==0);
   int num=table.size()/2;
-  int so  = plane*rhs.Grid()->_ostride[dimension]; // base offset for start of plane
+  int so  = plane*rhs.Grid()->_ostride[dimension]; // base offset for start of plane 
 
   auto rhs_v = rhs.View(AcceleratorRead);
   auto p0=&pointers[0][0];
   auto p1=&pointers[1][0];
   auto tp=&table[0];
-  accelerator_forNB(j, num, 1, {
+  accelerator_forNB(j, num, 1, { 
       compress.CompressExchange(p0,p1, &rhs_v[0], j,
 			      so+tp[2*j  ].second,
 			      so+tp[2*j+1].second,
@@ -109,20 +109,20 @@ void Gather_plane_exchange_table(Vector<std::pair<int,int> >& table,const Lattic
   rhs_v.ViewClose();
 }
 
-struct StencilEntry {
+struct StencilEntry { 
 #ifdef GRID_CUDA
-  uint64_t _byte_offset;       // 8 bytes
-  uint32_t _offset;            // 4 bytes
+  uint64_t _byte_offset;       // 8 bytes 
+  uint32_t _offset;            // 4 bytes 
 #else
-  uint64_t _byte_offset;       // 8 bytes
+  uint64_t _byte_offset;       // 8 bytes 
   uint64_t _offset;            // 8 bytes (8 ever required?)
 #endif
-  uint8_t _is_local;           // 1 bytes
+  uint8_t _is_local;           // 1 bytes 
   uint8_t _permute;            // 1 bytes
   uint8_t _around_the_world;   // 1 bytes
   uint8_t _pad;   // 1 bytes
 };
-// Could pack to 8 + 4 + 4 = 128 bit and use
+// Could pack to 8 + 4 + 4 = 128 bit and use 
 
 template<class vobj,class cobj,class Parameters>
 class CartesianStencilAccelerator {
@@ -149,18 +149,18 @@ class CartesianStencilAccelerator {
 
   accelerator_inline cobj *CommBuf(void) { return u_recv_buf_p; }
 
-  accelerator_inline int GetNodeLocal(int osite,int point) {
+  accelerator_inline int GetNodeLocal(int osite,int point) { 
     return this->_entries_p[point+this->_npoints*osite]._is_local;
   }
-  accelerator_inline StencilEntry * GetEntry(int &ptype,int point,int osite) {
-    ptype = this->_permute_type[point]; return & this->_entries_p[point+this->_npoints*osite];
+  accelerator_inline StencilEntry * GetEntry(int &ptype,int point,int osite) { 
+    ptype = this->_permute_type[point]; return & this->_entries_p[point+this->_npoints*osite]; 
   }
 
   accelerator_inline uint64_t GetInfo(int &ptype,int &local,int &perm,int point,int ent,uint64_t base) {
     uint64_t cbase = (uint64_t)&u_recv_buf_p[0];
     local = this->_entries_p[ent]._is_local;
     perm  = this->_entries_p[ent]._permute;
-    if (perm)  ptype = this->_permute_type[point];
+    if (perm)  ptype = this->_permute_type[point]; 
     if (local) {
       return  base + this->_entries_p[ent]._byte_offset;
     } else {
@@ -175,14 +175,14 @@ class CartesianStencilAccelerator {
     else       return cbase + this->_entries_p[ent]._byte_offset;
   }
 
-  accelerator_inline void iCoorFromIindex(Coordinate &coor,int lane)
+  accelerator_inline void iCoorFromIindex(Coordinate &coor,int lane) 
   {
     Lexicographic::CoorFromIndex(coor,lane,this->_simd_layout);
   }
 };
 
 template<class vobj,class cobj,class Parameters>
-class CartesianStencilView : public CartesianStencilAccelerator<vobj,cobj,Parameters>
+class CartesianStencilView : public CartesianStencilAccelerator<vobj,cobj,Parameters> 
 {
  private:
   int *closed;
@@ -192,7 +192,7 @@ class CartesianStencilView : public CartesianStencilAccelerator<vobj,cobj,Parame
   // default copy constructor
   CartesianStencilView (const CartesianStencilView &refer_to_me) = default;
 
-  CartesianStencilView (const CartesianStencilAccelerator<vobj,cobj,Parameters> &refer_to_me,ViewMode _mode)
+  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)
@@ -201,14 +201,14 @@ class CartesianStencilView : public CartesianStencilAccelerator<vobj,cobj,Parame
       MemoryManager::ViewOpen(this->_entries_p,
 			      this->_npoints*this->_osites*sizeof(StencilEntry),
 			      mode,
-			      AdviseDefault);
+			      AdviseDefault);    
   }
-
+  
   void ViewClose(void)
   {
-    MemoryManager::ViewClose(this->cpu_ptr,this->mode);
+    MemoryManager::ViewClose(this->cpu_ptr,this->mode);    
   }
-
+  
 };
 
 ////////////////////////////////////////
@@ -245,12 +245,12 @@ public:
     cobj * mpi_p;
     Integer buffer_size;
   };
-
+  
 
 protected:
   GridBase *                        _grid;
 
-public:
+public: 
   GridBase *Grid(void) const { return _grid; }
 
   ////////////////////////////////////////////////////////////////////////
@@ -264,7 +264,7 @@ public:
     View_type accessor(*( (View_type *) this),mode);
     return accessor;
   }
-
+  
   int face_table_computed;
   std::vector<Vector<std::pair<int,int> > > face_table ;
   Vector<int> surface_list;
@@ -314,7 +314,7 @@ public:
   ////////////////////////////////////////
   // Stencil query
   ////////////////////////////////////////
-  inline int SameNode(int point) {
+  inline int SameNode(int point) { 
 
     int dimension    = this->_directions[point];
     int displacement = this->_distances[point];
@@ -338,7 +338,7 @@ public:
     // FIXME  this logic needs to be sorted for three link term
     //    assert( (displacement==1) || (displacement==-1));
     // Present hack only works for >= 4^4 subvol per node
-    _grid->ShiftedRanks(dimension,nbr_proc,xmit_to_rank,recv_from_rank);
+    _grid->ShiftedRanks(dimension,nbr_proc,xmit_to_rank,recv_from_rank); 
 
     void *shm = (void *) _grid->ShmBufferTranslate(recv_from_rank,this->u_recv_buf_p);
 
@@ -378,7 +378,7 @@ public:
       comm_time_thr[mythread] += comm_leave_thr[mythread] - comm_enter_thr[mythread];
     }
   }
-
+  
   void CollateThreads(void)
   {
     int nthreads = CartesianCommunicator::nCommThreads;
@@ -402,7 +402,7 @@ public:
       if ( (t0 > 0.0) && ( t0 < first ) ) first = t0;   // min time seen
 
       if ( t1 > last ) last = t1;                       // max time seen
-
+      
     }
     commtime+= last-first;
   }
@@ -464,30 +464,30 @@ public:
       this->CommunicateBegin(reqs);
       this->CommunicateComplete(reqs);
     }
-  }
-
-  template<class compressor> void HaloExchange(const Lattice<vobj> &source,compressor &compress)
+  }    
+  
+  template<class compressor> void HaloExchange(const Lattice<vobj> &source,compressor &compress) 
   {
     Prepare();
     HaloGather(source,compress);
     Communicate();
-    CommsMergeSHM(compress);
-    CommsMerge(compress);
+    CommsMergeSHM(compress); 
+    CommsMerge(compress); 
   }
-
+  
   template<class compressor> int HaloGatherDir(const Lattice<vobj> &source,compressor &compress,int point,int & face_idx)
   {
     int dimension    = this->_directions[point];
     int displacement = this->_distances[point];
-
+    
     int fd = _grid->_fdimensions[dimension];
     int rd = _grid->_rdimensions[dimension];
-
+    
     // Map to always positive shift modulo global full dimension.
     int shift = (displacement+fd)%fd;
 
     assert (source.Checkerboard()== this->_checkerboard);
-
+    
     // the permute type
     int simd_layout     = _grid->_simd_layout[dimension];
     int comm_dim        = _grid->_processors[dimension] >1 ;
@@ -505,7 +505,7 @@ public:
 	  auto tmp  = GatherSimd(source,dimension,shift,0x3,compress,face_idx);
 	  is_same_node = is_same_node && tmp;
 	  splicetime+=usecond();
-	} else {
+	} else { 
 	  nosplicetime-=usecond();
 	  auto tmp  = Gather(source,dimension,shift,0x3,compress,face_idx);
 	  is_same_node = is_same_node && tmp;
@@ -531,7 +531,7 @@ public:
     }
     return is_same_node;
   }
-
+  
   template<class compressor>
   void HaloGather(const Lattice<vobj> &source,compressor &compress)
   {
@@ -542,9 +542,9 @@ public:
     // conformable(source.Grid(),_grid);
     assert(source.Grid()==_grid);
     halogtime-=usecond();
-
+    
     u_comm_offset=0;
-
+    
     // Gather all comms buffers
     int face_idx=0;
     for(int point = 0 ; point < this->_npoints; point++) {
@@ -557,16 +557,16 @@ public:
     accelerator_barrier();
     halogtime+=usecond();
   }
-
+ 
   /////////////////////////
   // Implementation
   /////////////////////////
   void Prepare(void)
   {
-    Decompressions.resize(0);
-    DecompressionsSHM.resize(0);
-    Mergers.resize(0);
-    MergersSHM.resize(0);
+    Decompressions.resize(0); 
+    DecompressionsSHM.resize(0); 
+    Mergers.resize(0); 
+    MergersSHM.resize(0); 
     Packets.resize(0);
     calls++;
   }
@@ -595,22 +595,22 @@ public:
     mv.push_back(m);
   }
   template<class decompressor>  void CommsMerge(decompressor decompress)    {
-    CommsMerge(decompress,Mergers,Decompressions);
+    CommsMerge(decompress,Mergers,Decompressions); 
   }
   template<class decompressor>  void CommsMergeSHM(decompressor decompress) {
-    mpi3synctime-=usecond();
+    mpi3synctime-=usecond();    
     _grid->StencilBarrier();// Synch shared memory on a single nodes
-    mpi3synctime+=usecond();
-    shmmergetime-=usecond();
+    mpi3synctime+=usecond();    
+    shmmergetime-=usecond();    
     CommsMerge(decompress,MergersSHM,DecompressionsSHM);
-    shmmergetime+=usecond();
+    shmmergetime+=usecond();    
   }
 
   template<class decompressor>
-  void CommsMerge(decompressor decompress,std::vector<Merge> &mm,std::vector<Decompress> &dd) {
+  void CommsMerge(decompressor decompress,std::vector<Merge> &mm,std::vector<Decompress> &dd) { 
 
     mergetime-=usecond();
-    for(int i=0;i<mm.size();i++){
+    for(int i=0;i<mm.size();i++){	
       auto mp = &mm[i].mpointer[0];
       auto vp0= &mm[i].vpointers[0][0];
       auto vp1= &mm[i].vpointers[1][0];
@@ -622,7 +622,7 @@ public:
     mergetime+=usecond();
 
     decompresstime-=usecond();
-    for(int i=0;i<dd.size();i++){
+    for(int i=0;i<dd.size();i++){	
       auto kp = dd[i].kernel_p;
       auto mp = dd[i].mpi_p;
       accelerator_forNB(o,dd[i].buffer_size,1,{
@@ -638,7 +638,7 @@ public:
     for(int i=0;i<_entries.size();i++){
       if( _entries[i]._is_local ) {
 	_entries[i]._byte_offset = _entries[i]._offset*sizeof(vobj);
-      } else {
+      } else { 
 	_entries[i]._byte_offset = _entries[i]._offset*sizeof(cobj);
       }
     }
@@ -653,15 +653,15 @@ public:
     for(int point=0;point<this->_npoints;point++){
       this->same_node[point] = this->SameNode(point);
     }
-
+    
     for(int site = 0 ;site< vol4;site++){
       int local = 1;
       for(int point=0;point<this->_npoints;point++){
-	if( (!this->GetNodeLocal(site*Ls,point)) && (!this->same_node[point]) ){
+	if( (!this->GetNodeLocal(site*Ls,point)) && (!this->same_node[point]) ){ 
 	  local = 0;
 	}
       }
-      if(local == 0) {
+      if(local == 0) { 
 	surface_list.push_back(site);
       }
     }
@@ -672,11 +672,11 @@ public:
 		   int checkerboard,
 		   const std::vector<int> &directions,
 		   const std::vector<int> &distances,
-		   Parameters p)
-    : shm_bytes_thr(npoints),
-      comm_bytes_thr(npoints),
+		   Parameters p) 
+    : shm_bytes_thr(npoints), 
+      comm_bytes_thr(npoints), 
       comm_enter_thr(npoints),
-      comm_leave_thr(npoints),
+      comm_leave_thr(npoints), 
       comm_time_thr(npoints)
   {
     face_table_computed=0;
@@ -687,7 +687,7 @@ public:
     /////////////////////////////////////
     this->_npoints = npoints;
     this->_comm_buf_size.resize(npoints),
-    this->_permute_type.resize(npoints),
+    this->_permute_type.resize(npoints), 
     this->_simd_layout = _grid->_simd_layout; // copy simd_layout to give access to Accelerator Kernels
     this->_directions = StencilVector(directions);
     this->_distances  = StencilVector(distances);
@@ -697,24 +697,24 @@ public:
     surface_list.resize(0);
 
     this->_osites  = _grid->oSites();
-
+    
     _entries.resize(this->_npoints* this->_osites);
     this->_entries_p = &_entries[0];
     for(int ii=0;ii<npoints;ii++){
-
+      
       int i = ii; // reverse direction to get SIMD comms done first
       int point = i;
-
+      
       int dimension    = directions[i];
       int displacement = distances[i];
       int shift = displacement;
-
+      
       int fd = _grid->_fdimensions[dimension];
       int rd = _grid->_rdimensions[dimension];
       this->_permute_type[point]=_grid->PermuteType(dimension);
-
+      
       this->_checkerboard = checkerboard;
-
+      
       //////////////////////////
       // the permute type
       //////////////////////////
@@ -724,25 +724,25 @@ public:
       int rotate_dim      = _grid->_simd_layout[dimension]>2;
 
       assert ( (rotate_dim && comm_dim) == false) ; // Do not think spread out is supported
-
+      
       int sshift[2];
-
+      
       //////////////////////////
       // Underlying approach. For each local site build
-      // up a table containing the npoint "neighbours" and whether they
+      // up a table containing the npoint "neighbours" and whether they 
       // live in lattice or a comms buffer.
       //////////////////////////
       if ( !comm_dim ) {
 	sshift[0] = _grid->CheckerBoardShiftForCB(this->_checkerboard,dimension,shift,Even);
 	sshift[1] = _grid->CheckerBoardShiftForCB(this->_checkerboard,dimension,shift,Odd);
-
+	
 	if ( sshift[0] == sshift[1] ) {
 	  Local(point,dimension,shift,0x3);
 	} else {
 	  Local(point,dimension,shift,0x1);// if checkerboard is unfavourable take two passes
 	  Local(point,dimension,shift,0x2);// both with block stride loop iteration
 	}
-      } else {
+      } else { 
 	// All permute extract done in comms phase prior to Stencil application
 	//        So tables are the same whether comm_dim or splice_dim
 	sshift[0] = _grid->CheckerBoardShiftForCB(this->_checkerboard,dimension,shift,Even);
@@ -784,23 +784,23 @@ public:
     int ld = _grid->_ldimensions[dimension];
     int gd = _grid->_gdimensions[dimension];
     int ly = _grid->_simd_layout[dimension];
-
+    
     // Map to always positive shift modulo global full dimension.
     int shift = (shiftpm+fd)%fd;
 
     // the permute type
     int permute_dim =_grid->PermuteDim(dimension);
-
-    for(int x=0;x<rd;x++){
-
+    
+    for(int x=0;x<rd;x++){       
+      
       //      int o   = 0;
       int bo  = x * _grid->_ostride[dimension];
-
+      
       int cb= (cbmask==0x2)? Odd : Even;
-
+      
       int sshift = _grid->CheckerBoardShiftForCB(this->_checkerboard,dimension,shift,cb);
       int sx     = (x+sshift)%rd;
-
+      
       int wraparound=0;
       if ( (shiftpm==-1) && (sx>x)  ) {
 	wraparound = 1;
@@ -808,7 +808,7 @@ public:
       if ( (shiftpm== 1) && (sx<x)  ) {
 	wraparound = 1;
       }
-
+      
       int permute_slice=0;
       if(permute_dim){
 	int wrap = sshift/rd; wrap=wrap % ly; // but it is local anyway
@@ -816,66 +816,66 @@ public:
 	if ( x< rd-num ) permute_slice=wrap;
 	else permute_slice = (wrap+1)%ly;
       }
-
+      
       CopyPlane(point,dimension,x,sx,cbmask,permute_slice,wraparound);
-
+      
     }
   }
-
+  
   void Comms     (int point,int dimension,int shiftpm,int cbmask)
   {
     GridBase *grid=_grid;
     const int Nsimd = grid->Nsimd();
-
+    
     int fd              = _grid->_fdimensions[dimension];
     int ld              = _grid->_ldimensions[dimension];
     int rd              = _grid->_rdimensions[dimension];
     int pd              = _grid->_processors[dimension];
     int simd_layout     = _grid->_simd_layout[dimension];
     int comm_dim        = _grid->_processors[dimension] >1 ;
-
+    
     assert(comm_dim==1);
     int shift = (shiftpm + fd) %fd;
     assert(shift>=0);
     assert(shift<fd);
-
+    
     // done in reduced dims, so SIMD factored
-    int buffer_size = _grid->_slice_nblock[dimension]*_grid->_slice_block[dimension];
+    int buffer_size = _grid->_slice_nblock[dimension]*_grid->_slice_block[dimension]; 
 
     this->_comm_buf_size[point] = buffer_size; // Size of _one_ plane. Multiple planes may be gathered and
 
     // send to one or more remote nodes.
-
+    
     int cb= (cbmask==0x2)? Odd : Even;
     int sshift= _grid->CheckerBoardShiftForCB(this->_checkerboard,dimension,shift,cb);
-
-    for(int x=0;x<rd;x++){
-
+    
+    for(int x=0;x<rd;x++){       
+      
       int permute_type=grid->PermuteType(dimension);
-
+      
       int sx        =  (x+sshift)%rd;
-
+      
       int offnode = 0;
       if ( simd_layout > 1 ) {
-
+	
 	for(int i=0;i<Nsimd;i++){
-
+	  
 	  int inner_bit = (Nsimd>>(permute_type+1));
 	  int ic= (i&inner_bit)? 1:0;
 	  int my_coor          = rd*ic + x;
 	  int nbr_coor         = my_coor+sshift;
 	  int nbr_proc = ((nbr_coor)/ld) % pd;// relative shift in processors
-
-	  if ( nbr_proc ) {
+	  
+	  if ( nbr_proc ) { 
 	    offnode =1;
 	  }
 	}
-
-      } else {
+	
+      } else { 
 	int comm_proc = ((x+sshift)/rd)%pd;
 	offnode = (comm_proc!= 0);
       }
-
+      
       int wraparound=0;
       if ( (shiftpm==-1) && (sx>x) && (grid->_processor_coor[dimension]==0) ) {
 	wraparound = 1;
@@ -884,24 +884,24 @@ public:
 	wraparound = 1;
       }
       if (!offnode) {
-
+	
 	int permute_slice=0;
-	CopyPlane(point,dimension,x,sx,cbmask,permute_slice,wraparound);
-
+	CopyPlane(point,dimension,x,sx,cbmask,permute_slice,wraparound); 
+	
       } else {
 
 	int words = buffer_size;
 	if (cbmask != 0x3) words=words>>1;
-
+	
 	//	int rank           = grid->_processor;
 	//	int recv_from_rank;
 	//	int xmit_to_rank;
-
+	
 	int unified_buffer_offset = _unified_buffer_size;
 	_unified_buffer_size    += words;
-
+	
 	ScatterPlane(point,dimension,x,cbmask,unified_buffer_offset,wraparound); // permute/extract/merge is done in comms phase
-
+	
       }
     }
   }
@@ -909,13 +909,13 @@ public:
   void CopyPlane(int point, int dimension,int lplane,int rplane,int cbmask,int permute,int wrap)
   {
     int rd = _grid->_rdimensions[dimension];
-
+    
     if ( !_grid->CheckerBoarded(dimension) ) {
-
+      
       int o   = 0;                                     // relative offset to base within plane
-      int ro  = rplane*_grid->_ostride[dimension]; // base offset for start of plane
+      int ro  = rplane*_grid->_ostride[dimension]; // base offset for start of plane 
       int lo  = lplane*_grid->_ostride[dimension]; // offset in buffer
-
+      
       // Simple block stride gather of SIMD objects
       for(int n=0;n<_grid->_slice_nblock[dimension];n++){
 	for(int b=0;b<_grid->_slice_block[dimension];b++){
@@ -927,18 +927,18 @@ public:
 	}
 	o +=_grid->_slice_stride[dimension];
       }
-
+      
     } else {
-
-      int ro  = rplane*_grid->_ostride[dimension]; // base offset for start of plane
-      int lo  = lplane*_grid->_ostride[dimension]; // base offset for start of plane
+      
+      int ro  = rplane*_grid->_ostride[dimension]; // base offset for start of plane 
+      int lo  = lplane*_grid->_ostride[dimension]; // base offset for start of plane 
       int o   = 0;                                     // relative offset to base within plane
-
+      
       for(int n=0;n<_grid->_slice_nblock[dimension];n++){
 	for(int b=0;b<_grid->_slice_block[dimension];b++){
-
+	  
 	  int ocb=1<<_grid->CheckerBoardFromOindex(o+b);
-
+	  
 	  if ( ocb&cbmask ) {
 	    int idx = point+(lo+o+b)*this->_npoints;
 	    _entries[idx]._offset =ro+o+b;
@@ -946,24 +946,24 @@ public:
 	    _entries[idx]._permute=permute;
 	    _entries[idx]._around_the_world=wrap;
 	  }
-
+	  
 	}
 	o +=_grid->_slice_stride[dimension];
       }
-
+      
     }
   }
   // Routine builds up integer table for each site in _offsets, _is_local, _permute
   void ScatterPlane (int point,int dimension,int plane,int cbmask,int offset, int wrap)
   {
     int rd = _grid->_rdimensions[dimension];
-
+    
     if ( !_grid->CheckerBoarded(dimension) ) {
-
-      int so  = plane*_grid->_ostride[dimension]; // base offset for start of plane
+      
+      int so  = plane*_grid->_ostride[dimension]; // base offset for start of plane 
       int o   = 0;                                    // relative offset to base within plane
       int bo  = 0;                                    // offset in buffer
-
+      
       // Simple block stride gather of SIMD objects
       for(int n=0;n<_grid->_slice_nblock[dimension];n++){
 	for(int b=0;b<_grid->_slice_block[dimension];b++){
@@ -975,16 +975,16 @@ public:
 	}
 	o +=_grid->_slice_stride[dimension];
       }
-
-    } else {
-
-      int so  = plane*_grid->_ostride[dimension]; // base offset for start of plane
+      
+    } else { 
+      
+      int so  = plane*_grid->_ostride[dimension]; // base offset for start of plane 
       int o   = 0;                                      // relative offset to base within plane
       int bo  = 0;                                      // offset in buffer
-
+      
       for(int n=0;n<_grid->_slice_nblock[dimension];n++){
 	for(int b=0;b<_grid->_slice_block[dimension];b++){
-
+	  
 	  int ocb=1<<_grid->CheckerBoardFromOindex(o+b);// Could easily be a table lookup
 	  if ( ocb & cbmask ) {
 	    int idx = point+(so+o+b)*this->_npoints;
@@ -998,16 +998,16 @@ public:
       }
     }
   }
-
+  
   template<class compressor>
   int Gather(const Lattice<vobj> &rhs,int dimension,int shift,int cbmask,compressor & compress,int &face_idx)
   {
     typedef typename cobj::vector_type vector_type;
     typedef typename cobj::scalar_type scalar_type;
-
+    
     assert(rhs.Grid()==_grid);
     //	  conformable(_grid,rhs.Grid());
-
+    
     int fd              = _grid->_fdimensions[dimension];
     int rd              = _grid->_rdimensions[dimension];
     int pd              = _grid->_processors[dimension];
@@ -1019,37 +1019,37 @@ public:
     assert(shift<fd);
 
     int buffer_size = _grid->_slice_nblock[dimension]*_grid->_slice_block[dimension];
-
+    
     int cb= (cbmask==0x2)? Odd : Even;
     int sshift= _grid->CheckerBoardShiftForCB(rhs.Checkerboard(),dimension,shift,cb);
-
+    
     int shm_receive_only = 1;
-    for(int x=0;x<rd;x++){
-
+    for(int x=0;x<rd;x++){       
+      
       int sx        = (x+sshift)%rd;
       int comm_proc = ((x+sshift)/rd)%pd;
-
+      
       if (comm_proc) {
 
 	int words = buffer_size;
 	if (cbmask != 0x3) words=words>>1;
-
+	
 	int bytes =  words * compress.CommDatumSize();
-
-	int so  = sx*rhs.Grid()->_ostride[dimension]; // base offset for start of plane
+	
+	int so  = sx*rhs.Grid()->_ostride[dimension]; // base offset for start of plane 
 	if ( !face_table_computed ) {
 	  face_table.resize(face_idx+1);
 	  Gather_plane_table_compute ((GridBase *)_grid,dimension,sx,cbmask,u_comm_offset,face_table[face_idx]);
 	}
-
+	
 	//      	int rank           = _grid->_processor;
 	int recv_from_rank;
 	int xmit_to_rank;
 	_grid->ShiftedRanks(dimension,comm_proc,xmit_to_rank,recv_from_rank);
-
+	
 	assert (xmit_to_rank   != _grid->ThisRank());
 	assert (recv_from_rank != _grid->ThisRank());
-
+	
 	/////////////////////////////////////////////////////////
 	// try the direct copy if possible
 	/////////////////////////////////////////////////////////
@@ -1062,13 +1062,13 @@ public:
 	}
 
 	send_buf = (cobj *)_grid->ShmBufferTranslate(xmit_to_rank,recv_buf);
-	if ( send_buf==NULL ) {
+	if ( send_buf==NULL ) { 
 	  send_buf = this->u_send_buf_p;
-	}
-
+	} 
+	
 	// Find out if we get the direct copy.
 	void *success = (void *) _grid->ShmBufferTranslate(recv_from_rank,this->u_send_buf_p);
-	if (success==NULL) {
+	if (success==NULL) { 
 	  // we found a packet that comes from MPI and contributes to this leg of stencil
 	  shm_receive_only = 0;
 	}
@@ -1077,9 +1077,9 @@ public:
 	assert(send_buf!=NULL);
 	Gather_plane_simple_table(face_table[face_idx],rhs,send_buf,compress,u_comm_offset,so);  face_idx++;
 	gathertime+=usecond();
-
+	
 	if ( compress.DecompressionStep() ) {
-
+	  
 	  if ( shm_receive_only ) { // Early decompress before MPI is finished is possible
 	    AddDecompress(&this->u_recv_buf_p[u_comm_offset],
 			  &recv_buf[u_comm_offset],
@@ -1108,7 +1108,7 @@ public:
     }
     return shm_receive_only;
   }
-
+  
   template<class compressor>
   int  GatherSimd(const Lattice<vobj> &rhs,int dimension,int shift,int cbmask,compressor &compress,int & face_idx)
   {
@@ -1136,7 +1136,7 @@ public:
     ///////////////////////////////////////////////
     int buffer_size = _grid->_slice_nblock[dimension]*_grid->_slice_block[dimension];
     //    int words = sizeof(cobj)/sizeof(vector_type);
-
+    
     assert(cbmask==0x3); // Fixme think there is a latent bug if not true
                          // This assert will trap it if ever hit. Not hit normally so far
     int reduced_buffer_size = buffer_size;
@@ -1152,22 +1152,22 @@ public:
     ///////////////////////////////////////////
     // Work out what to send where
     ///////////////////////////////////////////
-
+    
     int cb    = (cbmask==0x2)? Odd : Even;
     int sshift= _grid->CheckerBoardShiftForCB(rhs.Checkerboard(),dimension,shift,cb);
-
+    
     // loop over outer coord planes orthog to dim
     int shm_receive_only = 1;
-    for(int x=0;x<rd;x++){
-
+    for(int x=0;x<rd;x++){       
+      
       int any_offnode = ( ((x+sshift)%fd) >= rd );
 
       if ( any_offnode ) {
-
-	for(int i=0;i<maxl;i++){
+	
+	for(int i=0;i<maxl;i++){       
 	  spointers[i] = (cobj *) &u_simd_send_buf[i][u_comm_offset];
 	}
-
+	
 	int sx   = (x+sshift)%rd;
 
 	if ( !face_table_computed ) {
@@ -1202,13 +1202,13 @@ public:
 
 	    int recv_from_rank;
 	    int xmit_to_rank;
-
-	    _grid->ShiftedRanks(dimension,nbr_proc,xmit_to_rank,recv_from_rank);
-
+	    
+	    _grid->ShiftedRanks(dimension,nbr_proc,xmit_to_rank,recv_from_rank); 
+ 
 	    // shm == receive pointer         if offnode
 	    // shm == Translate[send pointer] if on node -- my view of his send pointer
 	    cobj *shm = (cobj *) _grid->ShmBufferTranslate(recv_from_rank,sp);
-	    if (shm==NULL) {
+	    if (shm==NULL) { 
 	      shm = rp;
 	      // we found a packet that comes from MPI and contributes to this shift.
 	      // is_same_node is only used in the WilsonStencil, and gets set for this point in the stencil.
@@ -1222,15 +1222,15 @@ public:
 
 	    AddPacket((void *)sp,(void *)rp,xmit_to_rank,recv_from_rank,bytes);
 
-
-	  } else {
-
+	    
+	  } else { 
+	    
 	    rpointers[i] = sp;
-
+	    
 	  }
 	}
 
-	if ( shm_receive_only ) {
+	if ( shm_receive_only ) { 
 	  AddMerge(&this->u_recv_buf_p[u_comm_offset],rpointers,reduced_buffer_size,permute_type,MergersSHM);
 	} else {
 	  AddMerge(&this->u_recv_buf_p[u_comm_offset],rpointers,reduced_buffer_size,permute_type,Mergers);
@@ -1265,9 +1265,9 @@ public:
     shm_bytes = 0.;
     calls = 0.;
   };
-
+  
   void Report(void) {
-#define AVERAGE(A)
+#define AVERAGE(A) 
 #define PRINTIT(A) AVERAGE(A); std::cout << GridLogMessage << " Stencil " << #A << " "<< A/calls<<std::endl;
     RealD NP = _grid->_Nprocessors;
     RealD NN = _grid->NodeCount();
@@ -1284,7 +1284,7 @@ public:
       }
     }
     if (threaded) commtime += t;
-
+    
     _grid->GlobalSum(commtime);    commtime/=NP;
     if ( calls > 0. ) {
       std::cout << GridLogMessage << " Stencil calls "<<calls<<std::endl;
@@ -1307,8 +1307,8 @@ public:
 	std::cout << GridLogMessage << " Stencil SHM " << (shm_bytes)/gatheralltime/1000.*NP/NN << " GB/s per node"<<std::endl;
 
 	auto all_bytes = comms_bytes+shm_bytes;
-	std::cout << GridLogMessage << " Stencil SHM all " << (all_bytes)/gatheralltime/1000. << " GB/s per rank"<<std::endl;
-	std::cout << GridLogMessage << " Stencil SHM all " << (all_bytes)/gatheralltime/1000.*NP/NN << " GB/s per node"<<std::endl;
+	std::cout << GridLogMessage << " Stencil SHM all" << (all_bytes)/gatheralltime/1000. << " GB/s per rank"<<std::endl;
+	std::cout << GridLogMessage << " Stencil SHM all" << (all_bytes)/gatheralltime/1000.*NP/NN << " GB/s per node"<<std::endl;
 
 	auto membytes = (shm_bytes + comms_bytes/2) // read/write
 	              + (shm_bytes+comms_bytes)/2 * sizeof(vobj)/sizeof(cobj);
@@ -1326,7 +1326,7 @@ public:
 #undef PRINTIT
 #undef AVERAGE
   };
-
+  
 };
 NAMESPACE_END(Grid);
 

Grid/tensors/Tensor_index.h

@@ -272,7 +272,7 @@ public:
   static auto traceIndex(const iVector<vtype,N> arg) ->  iScalar<RemoveCRV(arg._internal[0])>
   {
     iScalar<RemoveCRV(arg._internal[0])> ret;
-    zeroit(ret);
+    ret._internal=Zero();
     for(int i=0;i<N;i++){
       ret._internal = ret._internal+ arg._internal[i];
     }

Grid/tensors/Tensor_traits.h

@@ -190,36 +190,6 @@ NAMESPACE_BEGIN(Grid);
     typedef ComplexD DoublePrecision;
     typedef ComplexD DoublePrecision2;
   };
-
-#ifdef GRID_CUDA
-  template<> struct GridTypeMapper<std::complex<float> > : public GridTypeMapper_Base {
-    typedef std::complex<float> scalar_type;
-    typedef std::complex<double> scalar_typeD;
-    typedef scalar_type vector_type;
-    typedef scalar_typeD vector_typeD;
-    typedef scalar_type tensor_reduced;
-    typedef scalar_type scalar_object;
-    typedef scalar_typeD scalar_objectD;
-    typedef scalar_type Complexified;
-    typedef RealF Realified;
-    typedef scalar_typeD DoublePrecision;
-    typedef scalar_typeD DoublePrecision2;
-  };
-  template<> struct GridTypeMapper<std::complex<double> > : public GridTypeMapper_Base {
-    typedef std::complex<double> scalar_type;
-    typedef std::complex<double> scalar_typeD;
-    typedef scalar_type vector_type;
-    typedef scalar_typeD vector_typeD;
-    typedef scalar_type tensor_reduced;
-    typedef scalar_type scalar_object;
-    typedef scalar_typeD scalar_objectD;
-    typedef scalar_type Complexified;
-    typedef RealD Realified;
-    typedef scalar_typeD DoublePrecision;
-    typedef scalar_typeD DoublePrecision2;
-  };
-#endif
-
   template<> struct GridTypeMapper<ComplexD2> : public GridTypeMapper_Base {
     typedef ComplexD2 scalar_type;
     typedef ComplexD2 scalar_typeD;

Grid/threads/Accelerator.cc

@@ -16,53 +16,40 @@ void acceleratorInit(void)
   char * localRankStr = NULL;
   int rank = 0, world_rank=0; 
 #define ENV_LOCAL_RANK_OMPI    "OMPI_COMM_WORLD_LOCAL_RANK"
-#define ENV_RANK_OMPI          "OMPI_COMM_WORLD_RANK"
-#define ENV_LOCAL_RANK_SLURM   "SLURM_LOCALID"
-#define ENV_RANK_SLURM         "SLURM_PROCID"
 #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) {
-    printf("OPENMPI detected\n");
+  if ((localRankStr = getenv(ENV_LOCAL_RANK_OMPI)) != NULL)
+  {
     rank = atoi(localRankStr);		
-  } else if ((localRankStr = getenv(ENV_LOCAL_RANK_MVAPICH)) != NULL) {
-    printf("MVAPICH detected\n");
+  }
+  if ((localRankStr = getenv(ENV_LOCAL_RANK_MVAPICH)) != NULL)
+  {
     rank = atoi(localRankStr);		
-  } else if ((localRankStr = getenv(ENV_LOCAL_RANK_SLURM)) != NULL) {
-    printf("SLURM detected\n");
-    rank = atoi(localRankStr);		
-  } else { 
-    printf("MPI version is unknown - bad things may happen\n");
   }
   if ((localRankStr = getenv(ENV_RANK_OMPI   )) != NULL) { world_rank = atoi(localRankStr);}
   if ((localRankStr = getenv(ENV_RANK_MVAPICH)) != NULL) { world_rank = atoi(localRankStr);}
-  if ((localRankStr = getenv(ENV_RANK_SLURM  )) != NULL) { world_rank = atoi(localRankStr);}
 
   size_t totalDeviceMem=0;
   for (int i = 0; i < nDevices; i++) {
 
-#define GPU_PROP_FMT(canMapHostMemory,FMT)     printf("AcceleratorCudaInit[%d]:   " #canMapHostMemory ": " FMT" \n",rank,prop.canMapHostMemory);
+#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) {
-#ifndef GRID_IBM_SUMMIT
-      if ( i==rank ) {
-	printf("AcceleratorCudaInit[%d]: ========================\n",rank);
-	printf("AcceleratorCudaInit[%d]: Device Number    : %d\n", rank,i);
-	printf("AcceleratorCudaInit[%d]: ========================\n",rank);
-	printf("AcceleratorCudaInit[%d]: Device identifier: %s\n",rank, prop.name);
+      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(pciBusID);
-	GPU_PROP(pciDeviceID);
-      }
-#endif
+      GPU_PROP_FMT(totalGlobalMem,"%lld");
+      GPU_PROP(managedMemory);
+      GPU_PROP(isMultiGpuBoard);
+      GPU_PROP(warpSize);
       //      GPU_PROP(unifiedAddressing);
       //      GPU_PROP(l2CacheSize);
       //      GPU_PROP(singleToDoublePrecisionPerfRatio);
@@ -74,9 +61,9 @@ void acceleratorInit(void)
 
 #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 - use default device\n");
+  if ( world_rank == 0 )  printf("AcceleratorCudaInit: IBM Summit or similar - NOT setting device to node rank\n");
 #else
-  printf("AcceleratorCudaInit: rank %d setting device to node rank %d\n",world_rank,rank);
+  if ( world_rank == 0 )  printf("AcceleratorCudaInit: setting device to node rank\n");
   cudaSetDevice(rank);
 #endif
   if ( world_rank == 0 )  printf("AcceleratorCudaInit: ================================================\n");

Grid/threads/Accelerator.h

@@ -70,7 +70,6 @@ NAMESPACE_BEGIN(Grid);
 //
 // Memory management:
 //
-//    int   acceleratorIsCommunicable(void *pointer);
 //    void *acceleratorAllocShared(size_t bytes);
 //    void acceleratorFreeShared(void *ptr);
 //
@@ -91,7 +90,6 @@ void     acceleratorInit(void);
 //////////////////////////////////////////////
 
 #ifdef GRID_CUDA
-#include <cuda.h>
 
 #ifdef __CUDA_ARCH__
 #define GRID_SIMT
@@ -151,9 +149,6 @@ inline void *acceleratorAllocShared(size_t bytes)
     ptr = (void *) NULL;
     printf(" cudaMallocManaged failed for %d %s \n",bytes,cudaGetErrorString(err));
   }
-  //  size_t free,total;
-  //  cudaMemGetInfo(&free,&total);
-  //  std::cout << "Malloc managed "<<bytes<<" "<<free<<"/"<<total<<std::endl;
   return ptr;
 };
 inline void *acceleratorAllocDevice(size_t bytes)
@@ -164,25 +159,12 @@ inline void *acceleratorAllocDevice(size_t bytes)
     ptr = (void *) NULL;
     printf(" cudaMalloc failed for %d %s \n",bytes,cudaGetErrorString(err));
   }
-  //  size_t free,total;
-  //  cudaMemGetInfo(&free,&total);
-  //  std::cout << "Malloc device "<<bytes<<" "<<free<<"/"<<total<<std::endl;
   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);}
-inline int  acceleratorIsCommunicable(void *ptr)
-{
-  int uvm;
-  auto 
-  cuerr = cuPointerGetAttribute( &uvm, CU_POINTER_ATTRIBUTE_IS_MANAGED, (CUdeviceptr) ptr);
-  assert(cuerr == cudaSuccess );
-  if(uvm) return 0;
-  else    return 1;
-}
-
 #endif
 
 //////////////////////////////////////////////
@@ -237,15 +219,6 @@ 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();}
-inline int  acceleratorIsCommunicable(void *ptr)
-{
-#if 0
-  auto uvm = cl::sycl::usm::get_pointer_type(ptr, theGridAccelerator->get_context());
-  if ( uvm = cl::sycl::usm::alloc::shared ) return 1;
-  else return 0;
-#endif
-  return 1;
-}
 
 #endif
 
@@ -325,7 +298,6 @@ inline void *acceleratorAllocShared(size_t bytes)
   return malloc(bytes);
 #endif
 };
-inline int  acceleratorIsCommunicable(void *ptr){ return 1; }
 
 inline void *acceleratorAllocDevice(size_t bytes)
 {
@@ -380,7 +352,6 @@ accelerator_inline int acceleratorSIMTlane(int Nsimd) { return 0; } // CUDA spec
 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);}
 
-inline int  acceleratorIsCommunicable(void *ptr){ return 1; }
 #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);};

Grid/util/Coordinate.h

@@ -99,10 +99,10 @@ inline std::ostream & operator<<(std::ostream &os, const AcceleratorVector<T,_nd
 {
   os << "[";
   for(int s=0;s<v.size();s++) {
-    os << v[s];
-    if( s < (v.size()-1) ){
-      os << " ";
-    }
+    os << v[s] << " ";
+  }
+  if (v.size() > 0) {
+    os << "\b";
   }
   os << "]";
   return os;

Grid/util/Init.cc

@@ -318,13 +318,6 @@ void Grid_init(int *argc,char ***argv)
     Grid_debug_handler_init();
   }
 
-#if defined(A64FX)
-  if( GridCmdOptionExists(*argv,*argv+*argc,"--comms-overlap") ){
-    std::cout << "Option --comms-overlap currently not supported on QPACE4. Exiting." << std::endl;
-    exit(EXIT_FAILURE);
-  }
-#endif
-
   //////////////////////////////////////////////////////////
   // Memory manager
   //////////////////////////////////////////////////////////
@@ -377,7 +370,9 @@ void Grid_init(int *argc,char ***argv)
     std::cout << GridLogMessage << "Mapped stencil comms buffers as MAP_HUGETLB "<<std::endl;
   }
 
-  MemoryManager::InitMessage();
+#ifndef GRID_UVM
+  std::cout << GridLogMessage << "MemoryManager Cache "<< MemoryManager::DeviceMaxBytes <<" bytes "<<std::endl;
+#endif
 
   if( GridCmdOptionExists(*argv,*argv+*argc,"--debug-mem") ){
     MemoryProfiler::debug = true;
@@ -472,7 +467,7 @@ void Grid_init(int *argc,char ***argv)
   if( GridCmdOptionExists(*argv,*argv+*argc,"--lebesgue") ){
     LebesgueOrder::UseLebesgueOrder=1;
   }
-  CartesianCommunicator::nCommThreads = 1;
+  CartesianCommunicator::nCommThreads = -1;
   if( GridCmdOptionExists(*argv,*argv+*argc,"--comms-threads") ){
     arg= GridCmdOptionPayload(*argv,*argv+*argc,"--comms-threads");
     GridCmdOptionInt(arg,CartesianCommunicator::nCommThreads);

SVE_README.txt

@@ -1,89 +0,0 @@
-* gcc 10.1 prebuild, QPACE4 interactive login w/ MPI
-
-scl enable gcc-toolset-10 bash
-module load mpi/openmpi-aarch64
-
-../configure --enable-simd=A64FX --enable-comms=mpi3 --enable-shm=shmget CXX=mpicxx CC=mpicc
-
-
-================================== deprecated ================================================
-
-* gcc 10.1 prebuild, QPACE4 interactive login
-
-scl enable gcc-toolset-10 bash
-
-../configure --without-hdf5 --enable-gen-simd-width=64 --enable-simd=GEN --enable-precision=double --enable-comms=none --enable-openmp CXX=g++ CC=gcc CXXFLAGS="-std=c++11 -march=armv8-a+sve -msve-vector-bits=512 -fno-gcse -DA64FXFIXEDSIZE -DA64FXASM -DDSLASHINTRIN"
-
-* gcc 10.1 prebuild w/ MPI, QPACE4 interactive login
-
-scl enable gcc-toolset-10 bash
-module load mpi/openmpi-aarch64
-
-../configure --without-hdf5 --enable-gen-simd-width=64 --enable-simd=GEN --enable-precision=double --enable-comms=mpi-auto --enable-shm=shmget --enable-openmp CXX=mpicxx CC=mpicc CXXFLAGS="-std=c++11 -march=armv8-a+sve -msve-vector-bits=512 -fno-gcse -DA64FXFIXEDSIZE -DA64FXASM -DDSLASHINTRIN"
-
-------------------------------------------------------------------------------
-
-* armclang 20.2 (qp4)
-
-../configure --without-hdf5 --enable-gen-simd-width=64 --enable-simd=GEN --enable-precision=double --enable-comms=none --enable-openmp CXX=armclang++ CC=armclang CXXFLAGS="-std=c++11 -mcpu=a64fx -DA64FX -DARMCLANGCOMPAT -DA64FXASM -DDSLASHINTRIN"
-
-------------------------------------------------------------------------------
-
-* gcc 10.0.1 VLA (merlin)
-
-../configure --with-lime=/home/men04359/lime/c-lime --without-hdf5 --enable-gen-simd-width=64 --enable-simd=GEN --enable-precision=double --enable-comms=none --enable-openmp CXX=g++-10.0.1 CC=gcc-10.0.1 CXXFLAGS="-std=c++11 -march=armv8-a+sve -msve-vector-bits=512 -fno-gcse -DA64FX -DA64FXASM -DDSLASHINTRIN" LDFLAGS=-static GRID_LDFLAGS=-static MPI_CXXLDFLAGS=-static
-
-
-* gcc 10.0.1 fixed-size ACLE (merlin)
-
-../configure --with-lime=/home/men04359/lime/c-lime --without-hdf5 --enable-gen-simd-width=64 --enable-simd=GEN --enable-precision=double --enable-comms=none --enable-openmp CXX=g++-10.0.1 CC=gcc-10.0.1 CXXFLAGS="-std=c++11 -march=armv8-a+sve -msve-vector-bits=512 -fno-gcse -DA64FXFIXEDSIZE -DA64FXASM -DDSLASHINTRIN"
-
-
-* gcc 10.0.1 fixed-size ACLE (fjt) w/ MPI
-
-export OMPI_CC=gcc-10.0.1
-export OMPI_CXX=g++-10.0.1
-export MPICH_CC=gcc-10.0.1
-export MPICH_CXX=g++-10.0.1
-
-$ ../configure --without-hdf5 --enable-gen-simd-width=64 --enable-simd=GEN --enable-precision=double --enable-comms=mpi3 --enable-openmp CXX=mpiFCC CC=mpifcc CXXFLAGS="-std=c++11 -march=armv8-a+sve -msve-vector-bits=512 -fno-gcse -DA64FXFIXEDSIZE -DA64FXASM -DDSLASHINTRIN -DTOFU -I/opt/FJSVxtclanga/tcsds-1.2.25/include/mpi/fujitsu -lrt" LDFLAGS="-L/opt/FJSVxtclanga/tcsds-1.2.25/lib64 -lrt"
-
---------------------------------------------------------
-
-* armclang 20.0 VLA (merlin)
-
-../configure --with-lime=/home/men04359/lime/c-lime --without-hdf5 --enable-gen-simd-width=64 --enable-simd=GEN --enable-precision=double --enable-comms=none --enable-openmp CXX=armclang++ CC=armclang CXXFLAGS="-std=c++11 -fno-unroll-loops -mllvm -vectorizer-min-trip-count=2 -march=armv8-a+sve -DARMCLANGCOMPAT -DA64FX -DA64FXASM -DDSLASHINTRIN" LDFLAGS=-static GRID_LDFLAGS=-static MPI_CXXLDFLAGS=-static
-
-TODO check ARMCLANGCOMPAT
-
-
-* armclang 20.1 VLA (merlin)
-
-../configure --with-lime=/home/men04359/lime/c-lime --without-hdf5 --enable-gen-simd-width=64 --enable-simd=GEN --enable-precision=double --enable-comms=none --enable-openmp CXX=armclang++ CC=armclang CXXFLAGS="-std=c++11 -mcpu=a64fx -DARMCLANGCOMPAT -DA64FX -DA64FXASM -DDSLASHINTRIN" LDFLAGS=-static GRID_LDFLAGS=-static MPI_CXXLDFLAGS=-static
-
-TODO check ARMCLANGCOMPAT
-
-
-* armclang 20.1 VLA (fjt cluster)
-
-../configure --with-lime=$HOME/local --without-hdf5 --enable-gen-simd-width=64 --enable-simd=GEN --enable-precision=double --enable-comms=none --enable-openmp CXX=armclang++ CC=armclang CXXFLAGS="-std=c++11 -mcpu=a64fx -DARMCLANGCOMPAT -DA64FX -DA64FXASM -DDSLASHINTRIN -DTOFU"
-
-TODO check ARMCLANGCOMPAT
-
-
-* armclang 20.1 VLA w/MPI (fjt cluster)
-
-../configure --with-lime=$HOME/local --without-hdf5 --enable-gen-simd-width=64 --enable-simd=GEN --enable-precision=double --enable-comms=mpi3 --enable-openmp CXX=mpiFCC CC=mpifcc CXXFLAGS="-std=c++11 -mcpu=a64fx -DA64FX -DA64FXASM -DDSLASHINTRIN -DTOFU -I/opt/FJSVxtclanga/tcsds-1.2.25/include/mpi/fujitsu -lrt" LDFLAGS="-L/opt/FJSVxtclanga/tcsds-1.2.25/lib64"
-
-No ARMCLANGCOMPAT -> still correct ?
-
---------------------------------------------------------
-
-* Fujitsu fcc
-
-../configure --with-lime=$HOME/grid-a64fx/lime/c-lime --without-hdf5 --enable-gen-simd-width=64 --enable-simd=GEN --enable-precision=double --enable-comms=none --enable-openmp --with-mpfr=/home/users/gre/gre-1/grid-a64fx/mpfr-build/install CXX=FCC CC=fcc CXXFLAGS="-Nclang -Kfast -DA64FX -DA64FXASM -DDSLASHINTRIN"
-
-
-* Fujitsu fcc w/ MPI
-
-../configure --with-lime=$HOME/grid-a64fx/lime/c-lime --without-hdf5 --enable-gen-simd-width=64 --enable-simd=GEN --enable-precision=double --enable-comms=mpi --enable-openmp --with-mpfr=/home/users/gre/gre-1/grid-a64fx/mpfr-build/install CXX=mpiFCC CC=mpifcc CXXFLAGS="-Nclang -Kfast -DA64FX -DA64FXASM -DDSLASHINTRIN -DTOFU"

benchmarks/Benchmark_comms.cc

@@ -74,6 +74,90 @@ int main (int argc, char ** argv)
   std::vector<double> t_time(Nloop);
   time_statistics timestat;
 
+  std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
+  std::cout<<GridLogMessage << "= Benchmarking concurrent halo exchange in "<<nmu<<" dimensions"<<std::endl;
+  std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
+  header();
+  for(int lat=8;lat<=maxlat;lat+=4){
+    for(int Ls=8;Ls<=8;Ls*=2){
+
+      Coordinate latt_size  ({lat*mpi_layout[0],
+	                      lat*mpi_layout[1],
+      			      lat*mpi_layout[2],
+      			      lat*mpi_layout[3]});
+
+      GridCartesian     Grid(latt_size,simd_layout,mpi_layout);
+      RealD Nrank = Grid._Nprocessors;
+      RealD Nnode = Grid.NodeCount();
+      RealD ppn = Nrank/Nnode;
+
+      std::vector<Vector<HalfSpinColourVectorD> > xbuf(8);	
+      std::vector<Vector<HalfSpinColourVectorD> > rbuf(8);
+
+      int ncomm;
+      int bytes=lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD);
+      for(int mu=0;mu<8;mu++){
+	xbuf[mu].resize(lat*lat*lat*Ls);
+	rbuf[mu].resize(lat*lat*lat*Ls);
+	//	std::cout << " buffers " << std::hex << (uint64_t)&xbuf[mu][0] <<" " << (uint64_t)&rbuf[mu][0] <<std::endl;
+      }
+
+      for(int i=0;i<Nloop;i++){
+      double start=usecond();
+
+	std::vector<CommsRequest_t> requests;
+
+	ncomm=0;
+	for(int mu=0;mu<4;mu++){
+	
+	  if (mpi_layout[mu]>1 ) {
+	  
+	    ncomm++;
+	    int comm_proc=1;
+	    int xmit_to_rank;
+	    int recv_from_rank;
+	    Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank);
+	    Grid.SendToRecvFromBegin(requests,
+				   (void *)&xbuf[mu][0],
+				   xmit_to_rank,
+				   (void *)&rbuf[mu][0],
+				   recv_from_rank,
+				   bytes);
+	
+	    comm_proc = mpi_layout[mu]-1;
+	  
+	    Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank);
+	    Grid.SendToRecvFromBegin(requests,
+				     (void *)&xbuf[mu+4][0],
+				     xmit_to_rank,
+				     (void *)&rbuf[mu+4][0],
+				     recv_from_rank,
+				     bytes);
+	  
+	  }
+	}
+	Grid.SendToRecvFromComplete(requests);
+	Grid.Barrier();
+	double stop=usecond();
+	t_time[i] = stop-start; // microseconds
+      }
+
+      timestat.statistics(t_time);
+
+      double dbytes    = bytes*ppn;
+      double xbytes    = dbytes*2.0*ncomm;
+      double rbytes    = xbytes;
+      double bidibytes = xbytes+rbytes;
+
+      std::cout<<GridLogMessage << std::setw(4) << lat<<"\t"<<Ls<<"\t"
+               <<std::setw(11) << bytes<< std::fixed << std::setprecision(1) << std::setw(7)
+               <<std::right<< xbytes/timestat.mean<<"  "<< xbytes*timestat.err/(timestat.mean*timestat.mean)<< " "
+               <<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;
+
+    }
+  }    
 
 
   std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
@@ -122,22 +206,26 @@ int main (int argc, char ** argv)
 	    {
 	      std::vector<CommsRequest_t> requests;
 	      Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank);
-	      Grid.SendToRecvFrom((void *)&xbuf[mu][0],
-				  xmit_to_rank,
-				  (void *)&rbuf[mu][0],
-				  recv_from_rank,
-				  bytes);
+	      Grid.SendToRecvFromBegin(requests,
+				       (void *)&xbuf[mu][0],
+				       xmit_to_rank,
+				       (void *)&rbuf[mu][0],
+				       recv_from_rank,
+				       bytes);
+	      Grid.SendToRecvFromComplete(requests);
 	    }
 
 	    comm_proc = mpi_layout[mu]-1;
 	    {
 	      std::vector<CommsRequest_t> requests;
 	      Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank);
-	      Grid.SendToRecvFrom((void *)&xbuf[mu+4][0],
-				  xmit_to_rank,
-				  (void *)&rbuf[mu+4][0],
-				  recv_from_rank,
-				  bytes);
+	      Grid.SendToRecvFromBegin(requests,
+				       (void *)&xbuf[mu+4][0],
+				       xmit_to_rank,
+				       (void *)&rbuf[mu+4][0],
+				       recv_from_rank,
+				       bytes);
+	      Grid.SendToRecvFromComplete(requests);
 	    }
 	  }
 	}

benchmarks/Benchmark_dwf.cc

@@ -1,5 +1,5 @@
  /*************************************************************************************
-    Grid physics library, www.github.com/paboyle/Grid
+    Grid physics library, www.github.com/paboyle/Grid 
     Source file: ./benchmarks/Benchmark_dwf.cc
     Copyright (C) 2015
 
@@ -77,7 +77,7 @@ int main (int argc, char ** argv)
 
   std::vector<int> seeds4({1,2,3,4});
   std::vector<int> seeds5({5,6,7,8});
-
+  
   std::cout << GridLogMessage << "Initialising 4d RNG" << std::endl;
   GridParallelRNG          RNG4(UGrid);  RNG4.SeedUniqueString(std::string("The 4D RNG"));
   std::cout << GridLogMessage << "Initialising 5d RNG" << std::endl;
@@ -107,8 +107,8 @@ int main (int argc, char ** argv)
   LatticeFermion    err(FGrid);
 
   std::cout << GridLogMessage << "Drawing gauge field" << std::endl;
-  LatticeGaugeField Umu(UGrid);
-  SU3::HotConfiguration(RNG4,Umu);
+  LatticeGaugeField Umu(UGrid); 
+  SU3::HotConfiguration(RNG4,Umu); 
   std::cout << GridLogMessage << "Random gauge initialised " << std::endl;
 #if 0
   Umu=1.0;
@@ -126,7 +126,7 @@ int main (int argc, char ** argv)
   // Naive wilson implementation
   ////////////////////////////////////
   // replicate across fifth dimension
-  LatticeGaugeField Umu5d(FGrid);
+  LatticeGaugeField Umu5d(FGrid); 
   std::vector<LatticeColourMatrix> U(4,FGrid);
   {
     autoView( Umu5d_v, Umu5d, CpuWrite);
@@ -197,38 +197,24 @@ int main (int argc, char ** argv)
     }
     double t1=usecond();
     FGrid->Barrier();
-
+    
     double volume=Ls;  for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
     double flops=single_site_flops*volume*ncall;
 
-    auto nsimd = vComplex::Nsimd();
-    auto simdwidth = sizeof(vComplex);
-
-    // RF: Nd Wilson * Ls, Nd gauge * Ls, Nc colors
-    double data_rf = volume * ((2*Nd+1)*Nd*Nc + 2*Nd*Nc*Nc) * simdwidth / nsimd * ncall / (1024.*1024.*1024.);
-
-    // mem: Nd Wilson * Ls, Nd gauge, Nc colors
-    double data_mem = (volume * (2*Nd+1)*Nd*Nc + (volume/Ls) *2*Nd*Nc*Nc) * simdwidth / nsimd * ncall / (1024.*1024.*1024.);
-
     std::cout<<GridLogMessage << "Called Dw "<<ncall<<" times in "<<t1-t0<<" us"<<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;
     std::cout<<GridLogMessage << "mflop/s per rank =  "<< flops/(t1-t0)/NP<<std::endl;
     std::cout<<GridLogMessage << "mflop/s per node =  "<< flops/(t1-t0)/NN<<std::endl;
-    std::cout<<GridLogMessage << "RF  GiB/s (base 2) =   "<< 1000000. * data_rf/((t1-t0))<<std::endl;
-    std::cout<<GridLogMessage << "mem GiB/s (base 2) =   "<< 1000000. * data_mem/((t1-t0))<<std::endl;
-    err = ref-result;
+    err = ref-result; 
     std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<<std::endl;
     //exit(0);
 
-    if(( norm2(err)>1.0e-4) ) {
-      /*
+    if(( norm2(err)>1.0e-4) ) { 
       std::cout << "RESULT\n " << result<<std::endl;
       std::cout << "REF   \n " << ref   <<std::endl;
       std::cout << "ERR   \n " << err   <<std::endl;
-      */
-      std::cout<<GridLogMessage << "WRONG RESULT" << std::endl;
       FGrid->Barrier();
       exit(-1);
     }
@@ -249,7 +235,7 @@ int main (int argc, char ** argv)
     }
     double t1=usecond();
     FGrid->Barrier();
-
+    
     double volume=Ls;  for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
     double flops=single_site_flops*volume*ncall;
 
@@ -257,7 +243,7 @@ int main (int argc, char ** argv)
     std::cout<<GridLogMessage << "mflop/s =   "<< flops/(t1-t0)<<std::endl;
     std::cout<<GridLogMessage << "mflop/s per rank =  "<< flops/(t1-t0)/NP<<std::endl;
     std::cout<<GridLogMessage << "mflop/s per node =  "<< flops/(t1-t0)/NN<<std::endl;
-    err = ref-result;
+    err = ref-result; 
     std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<<std::endl;
 
     assert (norm2(err)< 1.0e-3 );
@@ -297,14 +283,12 @@ int main (int argc, char ** argv)
   std::cout<<GridLogMessage << "Called DwDag"<<std::endl;
   std::cout<<GridLogMessage << "norm dag result "<< norm2(result)<<std::endl;
   std::cout<<GridLogMessage << "norm dag ref    "<< norm2(ref)<<std::endl;
-  err = ref-result;
+  err = ref-result; 
   std::cout<<GridLogMessage << "norm dag diff   "<< norm2(err)<<std::endl;
   if((norm2(err)>1.0e-4)){
-/*
 	std::cout<< "DAG RESULT\n "  <<ref     << std::endl;
 	std::cout<< "DAG sRESULT\n " <<result  << std::endl;
 	std::cout<< "DAG ERR   \n "  << err    <<std::endl;
-*/
   }
   LatticeFermion src_e (FrbGrid);
   LatticeFermion src_o (FrbGrid);
@@ -351,7 +335,7 @@ int main (int argc, char ** argv)
     }
     double t1=usecond();
     FGrid->Barrier();
-
+    
     double volume=Ls;  for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
     double flops=(single_site_flops*volume*ncall)/2.0;
 
@@ -371,14 +355,12 @@ int main (int argc, char ** argv)
   setCheckerboard(r_eo,r_o);
   setCheckerboard(r_eo,r_e);
 
-  err = r_eo-result;
+  err = r_eo-result; 
   std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<<std::endl;
   if((norm2(err)>1.0e-4)){
-    /*
 	std::cout<< "Deo RESULT\n " <<r_eo << std::endl;
 	std::cout<< "Deo REF\n " <<result  << std::endl;
 	std::cout<< "Deo ERR   \n " << err <<std::endl;
-    */
   }
 
   pickCheckerboard(Even,src_e,err);
@@ -391,3 +373,4 @@ int main (int argc, char ** argv)
   Grid_finalize();
   exit(0);
 }
+

benchmarks/Benchmark_su3.cc

@@ -191,7 +191,7 @@ int main (int argc, char ** argv)
   std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
   std::cout<<GridLogMessage << "= Benchmarking SU3xSU3  CovShiftForward(z,x,y)"<<std::endl;
   std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
-  std::cout<<GridLogMessage << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GB/s (incl Cshift)\t\t GFlop/s"<<std::endl;
+  std::cout<<GridLogMessage << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
   std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
 
   for(int lat=LMIN;lat<=LMAX;lat+=LADD){
@@ -216,16 +216,15 @@ int main (int argc, char ** argv)
 	
 	
 	    double bytes=3*vol*Nc*Nc*sizeof(Complex);
-	    double ncbytes=5*vol*Nc*Nc*sizeof(Complex);
 	    double flops=Nc*Nc*(6+8+8)*vol;
-	    std::cout<<GridLogMessage<<std::setprecision(3) << lat<<"\t\t"<<bytes<<"   \t\t"<<bytes/time<<"\t\t"<<ncbytes/time<<"\t\t" << flops/time<<std::endl;
+	    std::cout<<GridLogMessage<<std::setprecision(3) << lat<<"\t\t"<<bytes<<"   \t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
       }
   }
 #if 1
   std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
   std::cout<<GridLogMessage << "= Benchmarking SU3xSU3  z= x * Cshift(y)"<<std::endl;
   std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
-  std::cout<<GridLogMessage << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GB/s (incl Cshift)\t\t GFlop/s"<<std::endl;
+  std::cout<<GridLogMessage << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
   std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
 
   for(int lat=LMIN;lat<=LMAX;lat+=LADD){
@@ -259,11 +258,10 @@ int main (int argc, char ** argv)
 	tmult  = tmult /Nloop;
 	
 	double bytes=3*vol*Nc*Nc*sizeof(Complex);
-	double ncbytes=5*vol*Nc*Nc*sizeof(Complex);
 	double flops=Nc*Nc*(6+8+8)*vol;
 	std::cout<<GridLogMessage<<std::setprecision(3) << "total us "<<time<<" shift "<<tshift <<" mult "<<tmult<<std::endl;
 	time = time * 1000; // convert to NS for GB/s
-	std::cout<<GridLogMessage<<std::setprecision(3) << lat<<"\t\t"<<bytes<<"   \t\t"<<bytes/time<<"\t\t" <<ncbytes/time<<"\t\t" << flops/time<<std::endl;
+	std::cout<<GridLogMessage<<std::setprecision(3) << lat<<"\t\t"<<bytes<<"   \t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
       }
     }
 #endif

benchmarks/Benchmark_wilson.cc

@@ -1,6 +1,6 @@
     /*************************************************************************************
 
-    Grid physics library, www.github.com/paboyle/Grid
+    Grid physics library, www.github.com/paboyle/Grid 
 
     Source file: ./benchmarks/Benchmark_wilson.cc
 
@@ -98,7 +98,7 @@ int main (int argc, char ** argv)
   double volume=1;
   for(int mu=0;mu<Nd;mu++){
     volume=volume*latt_size[mu];
-  }
+  }  
 
   // Only one non-zero (y)
 #if 0
@@ -118,7 +118,7 @@ int main (int argc, char ** argv)
   for(int mu=0;mu<Nd;mu++){
     U[mu] = PeekIndex<LorentzIndex>(Umu,mu);
   }
-
+  
   { // Naive wilson implementation
     ref = Zero();
     for(int mu=0;mu<Nd;mu++){
@@ -146,32 +146,22 @@ int main (int argc, char ** argv)
   ref = -0.5*ref;
   RealD mass=0.1;
 
-  typename WilsonFermionR::ImplParams params;
+  typename WilsonFermionR::ImplParams params; 
 
   WilsonFermionR Dw(Umu,Grid,RBGrid,mass,params);
-
+  
   std::cout<<GridLogMessage << "Calling Dw"<<std::endl;
   int ncall=1000;
-  //int ncall=1;
-
-  // Counters
-  Dw.ZeroCounters();
-  Grid.Barrier();
-
   double t0=usecond();
   for(int i=0;i<ncall;i++){
     Dw.Dhop(src,result,0);
   }
-
-  // Counters
-  Grid.Barrier();
-
   double t1=usecond();
   double flops=single_site_flops*volume*ncall;
-
+  
   if (perfProfiling){
   std::cout<<GridLogMessage << "Profiling Dw with perf"<<std::endl;
-
+    
   System::profile("kernel", [&]() {
     for(int i=0;i<ncall;i++){
       Dw.Dhop(src,result,0);
@@ -183,28 +173,15 @@ int main (int argc, char ** argv)
 
   }
 
-  auto nsimd = vComplex::Nsimd();
-  auto simdwidth = sizeof(vComplex);
-
-  std::cout<<GridLogMessage << "Nsimd "<< nsimd << std::endl;
-  std::cout<<GridLogMessage << "Simd width "<< simdwidth << std::endl;
-
-  // RF: Nd Wilson, Nd gauge, Nc colors
-  double data = volume * ((2*Nd+1)*Nd*Nc + 2*Nd*Nc*Nc) * simdwidth / nsimd * ncall / (1024.*1024.*1024.);
-
+  
   std::cout<<GridLogMessage << "Called Dw"<<std::endl;
   std::cout<<GridLogMessage << "flops per site " << single_site_flops << 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;
-  std::cout<<GridLogMessage << "RF  GiB/s (base 2) =   "<< 1000000. * data/(t1-t0)<<std::endl;
-  err = ref-result;
+  err = ref-result; 
   std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<<std::endl;
 
-  Dw.Report();
-  
-  // guard
-  double err0 = norm2(err);
 
   //  for(int ss=0;ss<10;ss++ ){
   for(int ss=0;ss<0;ss++ ){
@@ -250,13 +227,8 @@ int main (int argc, char ** argv)
   std::cout<<GridLogMessage << "Called DwDag"<<std::endl;
   std::cout<<GridLogMessage << "norm result "<< norm2(result)<<std::endl;
   std::cout<<GridLogMessage << "norm ref    "<< norm2(ref)<<std::endl;
-  err = ref-result;
+  err = ref-result; 
   std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<<std::endl;
 
-  // guard
-  double err1 = norm2(err);
-  assert(fabs(err0) < 1.0e-3);
-  assert(fabs(err1) < 1.0e-3);
-
   Grid_finalize();
 }

benchmarks/Benchmark_wilson_sweep.cc

@@ -89,7 +89,6 @@ int main (int argc, char ** argv)
 
 	  std::cout << GridLogMessage;
 	  std::cout << latt_size;
-	  std::cout << "\t\t";
 
 	  GridCartesian           Grid(latt_size,simd_layout,mpi_layout);
 	  GridRedBlackCartesian RBGrid(&Grid);
@@ -133,10 +132,7 @@ void bench_wilson (
   for(int i=0; i<ncall; i++) { Dw.Dhop(src,result,dag); }
   double t1    = usecond();
   double flops = single_site_flops * volume * ncall;
-  double data_tp = (volume * 180 * 64 * ncall) / 1000.; // / (1024.*1024.*1024.);
-  //std::cout << flops/(t1-t0) << " (" << data_tp/(t1-t0) << " MB/s)  \t";
   std::cout << flops/(t1-t0) << "\t\t";
-
 }
 
 void bench_wilson_eo (