Global edit with change to View usage. autoView() creates a wrapper object that closes the view when scope closes.

Grid/algorithms/Algorithms.h

@@ -29,9 +29,11 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 #ifndef GRID_ALGORITHMS_H
 #define GRID_ALGORITHMS_H
 
+NAMESPACE_CHECK(algorithms);
 #include <Grid/algorithms/SparseMatrix.h>
 #include <Grid/algorithms/LinearOperator.h>
 #include <Grid/algorithms/Preconditioner.h>
+NAMESPACE_CHECK(SparseMatrix);
 
 #include <Grid/algorithms/approx/Zolotarev.h>
 #include <Grid/algorithms/approx/Chebyshev.h>
@@ -41,10 +43,12 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 #include <Grid/algorithms/approx/Forecast.h>
 #include <Grid/algorithms/approx/RemezGeneral.h>
 #include <Grid/algorithms/approx/ZMobius.h>
-
+NAMESPACE_CHECK(approx);
 #include <Grid/algorithms/iterative/Deflation.h>
 #include <Grid/algorithms/iterative/ConjugateGradient.h>
+NAMESPACE_CHECK(ConjGrad);
 #include <Grid/algorithms/iterative/BiCGSTAB.h>
+NAMESPACE_CHECK(BiCGSTAB);
 #include <Grid/algorithms/iterative/ConjugateResidual.h>
 #include <Grid/algorithms/iterative/NormalEquations.h>
 #include <Grid/algorithms/iterative/SchurRedBlack.h>
@@ -62,7 +66,9 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 #include <Grid/algorithms/iterative/ImplicitlyRestartedLanczos.h>
 #include <Grid/algorithms/iterative/PowerMethod.h>
 
+NAMESPACE_CHECK(PowerMethod);
 #include <Grid/algorithms/CoarsenedMatrix.h>
+NAMESPACE_CHECK(CoarsendMatrix);
 #include <Grid/algorithms/FFT.h>
 
 #endif

Grid/algorithms/CoarsenedMatrix.h

@@ -186,10 +186,10 @@ public:
 	
 	hermop.HermOp(*Tn,y);
 
-	auto y_v = y.View(AcceleratorWrite);
+	autoView( y_v , y, AcceleratorWrite);
-	auto Tn_v = Tn->View(AcceleratorWrite);
+	autoView( Tn_v , (*Tn), AcceleratorWrite);
-	auto Tnp_v = Tnp->View(AcceleratorWrite);
+	autoView( Tnp_v , (*Tnp), AcceleratorWrite);
-	auto Tnm_v = Tnm->View(AcceleratorWrite);
+	autoView( Tnm_v , (*Tnm), AcceleratorWrite);
 	const int Nsimd = CComplex::Nsimd();
 	accelerator_forNB(ss, FineGrid->oSites(), Nsimd, {
 	  coalescedWrite(y_v[ss],xscale*y_v(ss)+mscale*Tn_v(ss));
@@ -246,13 +246,14 @@ public:
   CartesianStencil<siteVector,siteVector,int> Stencil; 
 
   std::vector<CoarseMatrix> A;
-      
+    
   ///////////////////////
   // Interface
   ///////////////////////
   GridBase * Grid(void)         { return _grid; };   // this is all the linalg routines need to know
 
-  RealD M (const CoarseVector &in, CoarseVector &out){
+  RealD M (const CoarseVector &in, CoarseVector &out)
+  {
 
     conformable(_grid,in.Grid());
     conformable(in.Grid(),out.Grid());
@@ -263,12 +264,13 @@ public:
     double comms_usec = -usecond();
     Stencil.HaloExchange(in,compressor);
     comms_usec += usecond();
-
+  
-    auto in_v = in.View(AcceleratorRead);
+    autoView( in_v , in, AcceleratorRead);
-    auto out_v = out.View(AcceleratorWrite);
+    autoView( out_v , out, AcceleratorWrite);
     typedef LatticeView<Cobj> Aview;
-
+      
     Vector<Aview> AcceleratorViewContainer;
+  
     for(int p=0;p<geom.npoint;p++) AcceleratorViewContainer.push_back(A[p].View(AcceleratorRead));
     Aview *Aview_p = & AcceleratorViewContainer[0];
 
@@ -307,13 +309,15 @@ public:
 	}
       }
       coalescedWrite(out_v[ss](b),res);
-    });
+      });
     usecs +=usecond();
 
     double nrm_usec=-usecond();
     RealD Nout= norm2(out);
     nrm_usec+=usecond();
 
+    for(int p=0;p<geom.npoint;p++) AcceleratorViewContainer[p].ViewClose();
+
     return Nout;
   };
 
@@ -346,8 +350,8 @@ public:
     for(int p=0;p<geom.npoint;p++) AcceleratorViewContainer.push_back(A[p].View(AcceleratorRead));
     Aview *Aview_p = & AcceleratorViewContainer[0];
 
-    auto out_v = out.View(AcceleratorWrite);
+    autoView( out_v , out, AcceleratorWrite);
-    auto in_v  = in.View(AcceleratorRead);
+    autoView( in_v  , in, AcceleratorRead);
 
     const int Nsimd = CComplex::Nsimd();
     typedef decltype(coalescedRead(in_v[0])) calcVector;
@@ -375,6 +379,7 @@ public:
       }
       coalescedWrite(out_v[ss](b),res);
     });
+    for(int p=0;p<geom.npoint;p++) AcceleratorViewContainer[p].ViewClose();
   }
   void MdirAll(const CoarseVector &in,std::vector<CoarseVector> &out)
   {
@@ -542,10 +547,10 @@ public:
 	    
 	    blockMaskedInnerProduct(oZProj,omask,Subspace.subspace[j],Mphi);
 	    
-	    auto iZProj_v = iZProj.View(AcceleratorRead) ;
+	    autoView( iZProj_v , iZProj, AcceleratorRead) ;
-	    auto oZProj_v = oZProj.View(AcceleratorRead) ;
+	    autoView( oZProj_v , oZProj, AcceleratorRead) ;
-	    auto A_p     =  A[p].View(AcceleratorWrite);
+	    autoView( A_p     ,  A[p], AcceleratorWrite);
-	    auto A_self  = A[self_stencil].View(AcceleratorWrite);
+	    autoView( A_self  , A[self_stencil], AcceleratorWrite);
 
 	    accelerator_for(ss, Grid()->oSites(), Fobj::Nsimd(),{ coalescedWrite(A_p[ss](j,i),oZProj_v(ss)); });
 	    //      if( disp!= 0 ) { accelerator_for(ss, Grid()->oSites(), Fobj::Nsimd(),{ coalescedWrite(A_p[ss](j,i),oZProj_v(ss)); });}
@@ -563,11 +568,11 @@ public:
 	mult(tmp,phi,oddmask );  linop.Op(tmp,Mphio);
 
 	{
-	  auto tmp_      = tmp.View(AcceleratorWrite);
+	  autoView( tmp_      , tmp, AcceleratorWrite);
-	  auto evenmask_ = evenmask.View(AcceleratorRead);
+	  autoView( evenmask_ , evenmask, AcceleratorRead);
-	  auto oddmask_  =  oddmask.View(AcceleratorRead);
+	  autoView( oddmask_  ,  oddmask, AcceleratorRead);
-	  auto Mphie_    =  Mphie.View(AcceleratorRead);
+	  autoView( Mphie_    ,  Mphie, AcceleratorRead);
-	  auto Mphio_    =  Mphio.View(AcceleratorRead);
+	  autoView( Mphio_    ,  Mphio, AcceleratorRead);
 	  accelerator_for(ss, FineGrid->oSites(), Fobj::Nsimd(),{ 
 	      coalescedWrite(tmp_[ss],evenmask_(ss)*Mphie_(ss) + oddmask_(ss)*Mphio_(ss));
 	    });
@@ -575,8 +580,8 @@ public:
 
 	blockProject(SelfProj,tmp,Subspace.subspace);
 
-	auto SelfProj_ = SelfProj.View(AcceleratorRead);
+	autoView( SelfProj_ , SelfProj, AcceleratorRead);
-	auto A_self  = A[self_stencil].View(AcceleratorWrite);
+	autoView( A_self  , A[self_stencil], AcceleratorWrite);
 
 	accelerator_for(ss, Grid()->oSites(), Fobj::Nsimd(),{
 	  for(int j=0;j<nbasis;j++){

Grid/algorithms/FFT.h

@@ -36,7 +36,6 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 #endif
 #endif
 
-
 NAMESPACE_BEGIN(Grid);
 
 template<class scalar> struct FFTW { };
@@ -190,7 +189,7 @@ public:
     typedef typename sobj::scalar_type   scalar;
       
     Lattice<sobj> pgbuf(&pencil_g);
-    auto pgbuf_v = pgbuf.View(CpuWrite);
+    autoView(pgbuf_v , pgbuf, CpuWrite);
 
     typedef typename FFTW<scalar>::FFTW_scalar FFTW_scalar;
     typedef typename FFTW<scalar>::FFTW_plan   FFTW_plan;

Grid/algorithms/iterative/BiCGSTAB.h

@@ -122,12 +122,14 @@ class BiCGSTAB : public OperatorFunction<Field>
 
         LinearCombTimer.Start();
         bo = beta * omega;
-        auto p_v = p.View(AcceleratorWrite);
+	{
-        auto r_v = r.View(AcceleratorWrite);
+	  autoView( p_v , p, AcceleratorWrite);
-        auto v_v = v.View(AcceleratorWrite);
+	  autoView( r_v , r, AcceleratorRead);
-        accelerator_for(ss, p_v.size(), Field::vector_object::Nsimd(),{
+	  autoView( v_v , v, AcceleratorRead);
-          coalescedWrite(p_v[ss], beta*p_v(ss) - bo*v_v(ss) + r_v(ss));
+	  accelerator_for(ss, p_v.size(), Field::vector_object::Nsimd(),{
-        });
+	      coalescedWrite(p_v[ss], beta*p_v(ss) - bo*v_v(ss) + r_v(ss));
+	    });
+	}
         LinearCombTimer.Stop();
         LinalgTimer.Stop();
 
@@ -142,16 +144,20 @@ class BiCGSTAB : public OperatorFunction<Field>
         alpha = rho / Calpha.real();
 
         LinearCombTimer.Start();
-        auto h_v = h.View(AcceleratorWrite);
+	{
-        auto psi_v = psi.View(AcceleratorWrite);
+	  autoView( p_v , p, AcceleratorRead);
-        accelerator_for(ss, h_v.size(), Field::vector_object::Nsimd(),{
+	  autoView( r_v , r, AcceleratorRead);
-          coalescedWrite(h_v[ss], alpha*p_v(ss) + psi_v(ss));
+	  autoView( v_v , v, AcceleratorRead);
-        });
+	  autoView( psi_v,psi, AcceleratorRead);
-        
+	  autoView( h_v  ,  h, AcceleratorWrite);
-        auto s_v = s.View(AcceleratorWrite);
+	  autoView( s_v  ,  s, AcceleratorWrite);
-        accelerator_for(ss, s_v.size(), Field::vector_object::Nsimd(),{
+	  accelerator_for(ss, h_v.size(), Field::vector_object::Nsimd(),{
-          coalescedWrite(s_v[ss], -alpha*v_v(ss) + r_v(ss));
+	      coalescedWrite(h_v[ss], alpha*p_v(ss) + psi_v(ss));
-        });
+	    });
+	  accelerator_for(ss, s_v.size(), Field::vector_object::Nsimd(),{
+	      coalescedWrite(s_v[ss], -alpha*v_v(ss) + r_v(ss));
+ 	  });
+        }
         LinearCombTimer.Stop();
         LinalgTimer.Stop();
 
@@ -166,13 +172,19 @@ class BiCGSTAB : public OperatorFunction<Field>
         omega = Comega.real() / norm2(t);
 
         LinearCombTimer.Start();
-        auto t_v = t.View(AcceleratorWrite);
+	{
-        accelerator_for(ss, psi_v.size(), Field::vector_object::Nsimd(),{
+	  autoView( psi_v,psi, AcceleratorWrite);
-          coalescedWrite(psi_v[ss], h_v(ss) + omega * s_v(ss));
+	  autoView( r_v , r, AcceleratorWrite);
-          coalescedWrite(r_v[ss], -omega * t_v(ss) + s_v(ss));
+	  autoView( h_v , h, AcceleratorRead);
-        });
+	  autoView( s_v , s, AcceleratorRead);
+	  autoView( t_v , t, AcceleratorRead);
+	  accelerator_for(ss, psi_v.size(), Field::vector_object::Nsimd(),{
+	      coalescedWrite(psi_v[ss], h_v(ss) + omega * s_v(ss));
+	      coalescedWrite(r_v[ss], -omega * t_v(ss) + s_v(ss));
+	    });
+	}
         LinearCombTimer.Stop();
-
+	
         cp = norm2(r);
         LinalgTimer.Stop();
 

Grid/algorithms/iterative/ConjugateGradient.h

@@ -141,16 +141,16 @@ public:
 
       LinearCombTimer.Start();
       {
-	auto psi_v = psi.View(AcceleratorWrite);
+	autoView( psi_v , psi, AcceleratorWrite);
-	auto p_v   = p.View(AcceleratorWrite);
+	autoView( p_v   , p,   AcceleratorWrite);
-	auto r_v   = r.View(AcceleratorWrite);
+	autoView( r_v   , r,   AcceleratorWrite);
 	accelerator_for(ss,p_v.size(), Field::vector_object::Nsimd(),{
 	    coalescedWrite(psi_v[ss], a      *  p_v(ss) + psi_v(ss));
 	    coalescedWrite(p_v[ss]  , b      *  p_v(ss) + r_v  (ss));
-	  });
+	});
-	LinearCombTimer.Stop();
-	LinalgTimer.Stop();
       }
+      LinearCombTimer.Stop();
+      LinalgTimer.Stop();
 
       std::cout << GridLogIterative << "ConjugateGradient: Iteration " << k
                 << " residual " << sqrt(cp/ssq) << " target " << Tolerance << std::endl;

Grid/algorithms/iterative/ImplicitlyRestartedLanczos.h

@@ -57,16 +57,17 @@ void basisOrthogonalize(std::vector<Field> &basis,Field &w,int k)
 template<class Field>
 void basisRotate(std::vector<Field> &basis,Eigen::MatrixXd& Qt,int j0, int j1, int k0,int k1,int Nm) 
 {
-  typedef decltype(basis[0].View(CpuWrite)) View;
-  auto tmp_v = basis[0].View(CpuWrite);
-  Vector<View> basis_v(basis.size(),tmp_v);
-  View *basis_vp = &basis_v[0];
-  typedef typename Field::vector_object vobj;
   GridBase* grid = basis[0].Grid();
 
-  for(int k=0;k<basis.size();k++){
+  typedef typename Field::vector_object vobj;
-    basis_v[k] = basis[k].View(CpuWrite);
+  typedef decltype(basis[0].View(CpuWrite)) View;
-  }
+
+  Vector<View> basis_v; basis_v.reserve(basis.size());
+
+  for(int k=0;k<basis.size();k++) basis_v.push_back(basis[k].View(CpuWrite));
+
+  View *basis_vp = &basis_v[0];
+
 #if 1
   std::vector < vobj , commAllocator<vobj> > Bt(thread_max() * Nm); // Thread private
   thread_region
@@ -142,6 +143,7 @@ void basisRotate(std::vector<Field> &basis,Eigen::MatrixXd& Qt,int j0, int j1, i
       coalescedWrite(basis_vp[jj][sss],coalescedRead(Bp[ss*nrot+j]));
     });
   }
+  for(int k=0;k<basis.size();k++) basis_v[k].ViewClose();
 #endif
 }
 
@@ -149,20 +151,22 @@ void basisRotate(std::vector<Field> &basis,Eigen::MatrixXd& Qt,int j0, int j1, i
 template<class Field>
 void basisRotateJ(Field &result,std::vector<Field> &basis,Eigen::MatrixXd& Qt,int j, int k0,int k1,int Nm) 
 {
-  typedef decltype(basis[0].View(AcceleratorWrite)) View;
-  typedef typename Field::vector_object vobj;
   GridBase* grid = basis[0].Grid();
+  typedef typename Field::vector_object vobj;
+  typedef decltype(basis[0].View(AcceleratorWrite)) View;
 
   result.Checkerboard() = basis[0].Checkerboard();
-  auto result_v=result.View(AcceleratorWrite);
+
-  Vector<View> basis_v(basis.size(),result_v);
+  autoView(result_v,result, AcceleratorWrite);
+  Vector<View> basis_v; basis_v.reserve(basis.size());
   View * basis_vp = &basis_v[0];
-  for(int k=0;k<basis.size();k++){
+
-    basis_v[k] = basis[k].View(AcceleratorRead);
+  for(int k=0;k<basis.size();k++) basis_v.push_back(basis[k].View(AcceleratorRead));
-  }
+
-  Vector<double> Qt_jv(Nm);
+  Vector<double> Qt_jv(Nm);  double * Qt_j = & Qt_jv[0];
-  double * Qt_j = & Qt_jv[0];
+
   for(int k=0;k<Nm;++k) Qt_j[k]=Qt(j,k);
+
   accelerator_for(ss, grid->oSites(),vobj::Nsimd(),{
     auto B=coalescedRead(basis_vp[k0][ss]);
     B=Zero();
@@ -171,6 +175,7 @@ void basisRotateJ(Field &result,std::vector<Field> &basis,Eigen::MatrixXd& Qt,in
     }
     coalescedWrite(result_v[ss], B);
   });
+  for(int k=0;k<basis.size();k++) basis_v[k].ViewClose();
 }
 
 template<class Field>

Grid/allocator/MemoryManager.h

@@ -169,8 +169,9 @@ private:
 
  public:
   static void Print(void);
-  static void  ViewClose(void* AccPtr,ViewMode mode);
+  static int   isOpen   (void* CpuPtr);
-  static void *ViewOpen(void* CpuPtr,size_t bytes,ViewMode mode,ViewAdvise hint);
+  static void  ViewClose(void* CpuPtr,ViewMode mode);
+  static void *ViewOpen (void* CpuPtr,size_t bytes,ViewMode mode,ViewAdvise hint);
 
 };
 

Grid/allocator/MemoryManagerCache.cc

@@ -461,6 +461,17 @@ void  MemoryManager::Print(void)
   std::cout << GridLogDebug << "--------------------------------------------" << std::endl;
 
 };
+int   MemoryManager::isOpen   (void* _CpuPtr) 
+{ 
+  uint64_t CpuPtr = (uint64_t)_CpuPtr;
+  if ( EntryPresent(CpuPtr) ){
+    auto AccCacheIterator = EntryLookup(CpuPtr);
+    auto & AccCache = AccCacheIterator->second;
+    return AccCache.cpuLock+AccCache.accLock;
+  } else { 
+    return 0;
+  }
+}
 
 NAMESPACE_END(Grid);
 

Grid/allocator/MemoryManagerShared.cc

@@ -9,6 +9,7 @@ NAMESPACE_BEGIN(Grid);
 
 void  MemoryManager::ViewClose(void* AccPtr,ViewMode mode){};
 void *MemoryManager::ViewOpen(void* CpuPtr,size_t bytes,ViewMode mode,ViewAdvise hint){ return CpuPtr; };
+int   MemoryManager::isOpen   (void* CpuPtr) { return 0;}
 void  MemoryManager::Print(void){};
 void  MemoryManager::NotifyDeletion(void *ptr){};
 

Grid/cshift/Cshift_common.h

@@ -72,12 +72,14 @@ Gather_plane_simple (const Lattice<vobj> &rhs,commVector<vobj> &buffer,int dimen
        }
      }
   }
-  auto rhs_v = rhs.View(AcceleratorRead);
+  {
-  auto buffer_p = & buffer[0];
+    autoView(rhs_v , rhs, AcceleratorRead);
-  auto table = &Cshift_table[0];
+    auto buffer_p = & buffer[0];
-  accelerator_for(i,ent,1,{
+    auto table = &Cshift_table[0];
-    buffer_p[table[i].first]=rhs_v[table[i].second];
+    accelerator_for(i,ent,1,{
-  });
+      buffer_p[table[i].first]=rhs_v[table[i].second];
+    });
+  }
 }
 
 ///////////////////////////////////////////////////////////////////
@@ -100,8 +102,8 @@ Gather_plane_extract(const Lattice<vobj> &rhs,
   int e2=rhs.Grid()->_slice_block[dimension];
   int n1=rhs.Grid()->_slice_stride[dimension];
 
-  auto rhs_v = rhs.View(AcceleratorRead);
   if ( cbmask ==0x3){
+    autoView(rhs_v , rhs, AcceleratorRead);
     accelerator_for2d(n,e1,b,e2,1,{
 	int o      =   n*n1;
 	int offset = b+n*e2;
@@ -110,8 +112,8 @@ Gather_plane_extract(const Lattice<vobj> &rhs,
 	extract<vobj>(temp,pointers,offset);
       });
   } else { 
+    autoView(rhs_v , rhs, AcceleratorRead);
 
-  
     Coordinate rdim=rhs.Grid()->_rdimensions;
     Coordinate cdm =rhs.Grid()->_checker_dim_mask;
     std::cout << " Dense packed buffer WARNING " <<std::endl; // Does this get called twice once for each cb?
@@ -179,12 +181,14 @@ template<class vobj> void Scatter_plane_simple (Lattice<vobj> &rhs,commVector<vo
     }
   }
   
-  auto rhs_v = rhs.View(AcceleratorWrite);
+  {
-  auto buffer_p = & buffer[0];
+    autoView( rhs_v, rhs, AcceleratorWrite);
-  auto table = &Cshift_table[0];
+    auto buffer_p = & buffer[0];
-  accelerator_for(i,ent,1,{
+    auto table = &Cshift_table[0];
-    rhs_v[table[i].first]=buffer_p[table[i].second];
+    accelerator_for(i,ent,1,{
-  });
+	rhs_v[table[i].first]=buffer_p[table[i].second];
+    });
+  }
 }
 
 //////////////////////////////////////////////////////
@@ -204,7 +208,7 @@ template<class vobj> void Scatter_plane_merge(Lattice<vobj> &rhs,ExtractPointerA
   int e2=rhs.Grid()->_slice_block[dimension];
 
   if(cbmask ==0x3 ) {
-    auto rhs_v = rhs.View(AcceleratorWrite);
+    autoView( rhs_v , rhs, AcceleratorWrite);
     accelerator_for2d(n,e1,b,e2,1,{
 	int o      = n*rhs.Grid()->_slice_stride[dimension];
 	int offset = b+n*rhs.Grid()->_slice_block[dimension];
@@ -216,7 +220,7 @@ 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;
-    auto rhs_v = rhs.View(CpuWrite);
+    autoView( rhs_v, rhs, CpuWrite);
     for(int n=0;n<e1;n++){
       for(int b=0;b<e2;b++){
 	int o      = n*rhs.Grid()->_slice_stride[dimension];
@@ -272,13 +276,14 @@ template<class vobj> void Copy_plane(Lattice<vobj>& lhs,const Lattice<vobj> &rhs
     }
   }
 
-  auto rhs_v = rhs.View(AcceleratorRead);
+  {
-  auto lhs_v = lhs.View(AcceleratorWrite);
+    autoView(rhs_v , rhs, AcceleratorRead);
-  auto table = &Cshift_table[0];
+    autoView(lhs_v , lhs, AcceleratorWrite);
-  accelerator_for(i,ent,1,{
+    auto table = &Cshift_table[0];
-    lhs_v[table[i].first]=rhs_v[table[i].second];
+    accelerator_for(i,ent,1,{
-  });
+      lhs_v[table[i].first]=rhs_v[table[i].second];
-
+    });
+  }
 }
 
 template<class vobj> void Copy_plane_permute(Lattice<vobj>& lhs,const Lattice<vobj> &rhs, int dimension,int lplane,int rplane,int cbmask,int permute_type)
@@ -315,12 +320,14 @@ template<class vobj> void Copy_plane_permute(Lattice<vobj>& lhs,const Lattice<vo
     }}
   }
 
-  auto rhs_v = rhs.View(AcceleratorRead);
+  {
-  auto lhs_v = lhs.View(AcceleratorWrite);
+    autoView( rhs_v, rhs, AcceleratorRead);
-  auto table = &Cshift_table[0];
+    autoView( lhs_v, lhs, AcceleratorWrite);
-  accelerator_for(i,ent,1,{
+    auto table = &Cshift_table[0];
-    permute(lhs_v[table[i].first],rhs_v[table[i].second],permute_type);
+    accelerator_for(i,ent,1,{
-  });
+      permute(lhs_v[table[i].first],rhs_v[table[i].second],permute_type);
+    });
+  }
 }
 
 //////////////////////////////////////////////////////

Grid/lattice/Lattice_ET.h

@@ -87,12 +87,14 @@ sobj eval(const uint64_t ss, const sobj &arg)
 }
 
 template <class lobj> accelerator_inline 
-const lobj & eval(const uint64_t ss, const LatticeExprView<lobj> &arg) 
+const lobj & eval(const uint64_t ss, const LatticeView<lobj> &arg) 
 {
   return arg[ss];
 }
 
 // What needs this?
+// Cannot be legal on accelerator
+// Comparison must convert
 #if 1
 template <class lobj> accelerator_inline 
 const lobj & eval(const uint64_t ss, const Lattice<lobj> &arg) 

Grid/lattice/Lattice_arith.h

@@ -36,9 +36,9 @@ NAMESPACE_BEGIN(Grid);
 template<class obj1,class obj2,class obj3> inline
 void mult(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const Lattice<obj3> &rhs){
   ret.Checkerboard() = lhs.Checkerboard();
-  auto ret_v = ret.View(AcceleratorWrite);
+  autoView( ret_v , ret, AcceleratorWrite);
-  auto lhs_v = lhs.View(AcceleratorRead);
+  autoView( lhs_v , lhs, AcceleratorRead);
-  auto rhs_v = rhs.View(AcceleratorRead);
+  autoView( rhs_v , rhs, AcceleratorRead);
   conformable(ret,rhs);
   conformable(lhs,rhs);
   accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
@@ -55,9 +55,9 @@ void mac(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const Lattice<obj3> &rhs){
   ret.Checkerboard() = lhs.Checkerboard();
   conformable(ret,rhs);
   conformable(lhs,rhs);
-  auto ret_v = ret.View(AcceleratorWrite);
+  autoView( ret_v , ret, AcceleratorWrite);
-  auto lhs_v = lhs.View(AcceleratorRead);
+  autoView( lhs_v , lhs, AcceleratorRead);
-  auto rhs_v = rhs.View(AcceleratorRead);
+  autoView( rhs_v , rhs, AcceleratorRead);
   accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
     decltype(coalescedRead(obj1())) tmp;
     auto lhs_t=lhs_v(ss);
@@ -72,9 +72,9 @@ void sub(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const Lattice<obj3> &rhs){
   ret.Checkerboard() = lhs.Checkerboard();
   conformable(ret,rhs);
   conformable(lhs,rhs);
-  auto ret_v = ret.View(AcceleratorWrite);
+  autoView( ret_v , ret, AcceleratorWrite);
-  auto lhs_v = lhs.View(AcceleratorRead);
+  autoView( lhs_v , lhs, AcceleratorRead);
-  auto rhs_v = rhs.View(AcceleratorRead);
+  autoView( rhs_v , rhs, AcceleratorRead);
   accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
     decltype(coalescedRead(obj1())) tmp;
     auto lhs_t=lhs_v(ss);
@@ -88,9 +88,9 @@ void add(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const Lattice<obj3> &rhs){
   ret.Checkerboard() = lhs.Checkerboard();
   conformable(ret,rhs);
   conformable(lhs,rhs);
-  auto ret_v = ret.View(AcceleratorWrite);
+  autoView( ret_v , ret, AcceleratorWrite);
-  auto lhs_v = lhs.View(AcceleratorRead);
+  autoView( lhs_v , lhs, AcceleratorRead);
-  auto rhs_v = rhs.View(AcceleratorRead);
+  autoView( rhs_v , rhs, AcceleratorRead);
   accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
     decltype(coalescedRead(obj1())) tmp;
     auto lhs_t=lhs_v(ss);
@@ -107,8 +107,8 @@ template<class obj1,class obj2,class obj3> inline
 void mult(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
   ret.Checkerboard() = lhs.Checkerboard();
   conformable(lhs,ret);
-  auto ret_v = ret.View(AcceleratorWrite);
+  autoView( ret_v , ret, AcceleratorWrite);
-  auto lhs_v = lhs.View(AcceleratorRead);
+  autoView( lhs_v , lhs, AcceleratorRead);
   accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
     decltype(coalescedRead(obj1())) tmp;
     mult(&tmp,&lhs_v(ss),&rhs);
@@ -120,8 +120,8 @@ template<class obj1,class obj2,class obj3> inline
 void mac(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
   ret.Checkerboard() = lhs.Checkerboard();
   conformable(ret,lhs);
-  auto ret_v = ret.View(AcceleratorWrite);
+  autoView( ret_v , ret, AcceleratorWrite);
-  auto lhs_v = lhs.View(AcceleratorRead);
+  autoView( lhs_v , lhs, AcceleratorRead);
   accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
     decltype(coalescedRead(obj1())) tmp;
     auto lhs_t=lhs_v(ss);
@@ -134,8 +134,8 @@ template<class obj1,class obj2,class obj3> inline
 void sub(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
   ret.Checkerboard() = lhs.Checkerboard();
   conformable(ret,lhs);
-  auto ret_v = ret.View(AcceleratorWrite);
+  autoView( ret_v , ret, AcceleratorWrite);
-  auto lhs_v = lhs.View(AcceleratorRead);
+  autoView( lhs_v , lhs, AcceleratorRead);
   accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
     decltype(coalescedRead(obj1())) tmp;
     auto lhs_t=lhs_v(ss);
@@ -147,8 +147,8 @@ template<class obj1,class obj2,class obj3> inline
 void add(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
   ret.Checkerboard() = lhs.Checkerboard();
   conformable(lhs,ret);
-  auto ret_v = ret.View(AcceleratorWrite);
+  autoView( ret_v , ret, AcceleratorWrite);
-  auto lhs_v = lhs.View(AcceleratorRead);
+  autoView( lhs_v , lhs, AcceleratorRead);
   accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
     decltype(coalescedRead(obj1())) tmp;
     auto lhs_t=lhs_v(ss);
@@ -164,8 +164,8 @@ template<class obj1,class obj2,class obj3> inline
 void mult(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
   ret.Checkerboard() = rhs.Checkerboard();
   conformable(ret,rhs);
-  auto ret_v = ret.View(AcceleratorWrite);
+  autoView( ret_v , ret, AcceleratorWrite);
-  auto rhs_v = lhs.View(AcceleratorRead);
+  autoView( rhs_v , lhs, AcceleratorRead);
   accelerator_for(ss,rhs_v.size(),obj1::Nsimd(),{
     decltype(coalescedRead(obj1())) tmp;
     auto rhs_t=rhs_v(ss);
@@ -178,8 +178,8 @@ template<class obj1,class obj2,class obj3> inline
 void mac(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
   ret.Checkerboard() = rhs.Checkerboard();
   conformable(ret,rhs);
-  auto ret_v = ret.View(AcceleratorWrite);
+  autoView( ret_v , ret, AcceleratorWrite);
-  auto rhs_v = lhs.View(AcceleratorRead);
+  autoView( rhs_v , lhs, AcceleratorRead);
   accelerator_for(ss,rhs_v.size(),obj1::Nsimd(),{
     decltype(coalescedRead(obj1())) tmp;
     auto rhs_t=rhs_v(ss);
@@ -192,8 +192,8 @@ template<class obj1,class obj2,class obj3> inline
 void sub(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
   ret.Checkerboard() = rhs.Checkerboard();
   conformable(ret,rhs);
-  auto ret_v = ret.View(AcceleratorWrite);
+  autoView( ret_v , ret, AcceleratorWrite);
-  auto rhs_v = lhs.View(AcceleratorRead);
+  autoView( rhs_v , lhs, AcceleratorRead);
   accelerator_for(ss,rhs_v.size(),obj1::Nsimd(),{
     decltype(coalescedRead(obj1())) tmp;
     auto rhs_t=rhs_v(ss);
@@ -205,8 +205,8 @@ template<class obj1,class obj2,class obj3> inline
 void add(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
   ret.Checkerboard() = rhs.Checkerboard();
   conformable(ret,rhs);
-  auto ret_v = ret.View(AcceleratorWrite);
+  autoView( ret_v , ret, AcceleratorWrite);
-  auto rhs_v = lhs.View(AcceleratorRead);
+  autoView( rhs_v , lhs, AcceleratorRead);
   accelerator_for(ss,rhs_v.size(),obj1::Nsimd(),{
     decltype(coalescedRead(obj1())) tmp;
     auto rhs_t=rhs_v(ss);
@@ -220,9 +220,9 @@ void axpy(Lattice<vobj> &ret,sobj a,const Lattice<vobj> &x,const Lattice<vobj> &
   ret.Checkerboard() = x.Checkerboard();
   conformable(ret,x);
   conformable(x,y);
-  auto ret_v = ret.View(AcceleratorWrite);
+  autoView( ret_v , ret, AcceleratorWrite);
-  auto x_v = x.View(AcceleratorRead);
+  autoView( x_v , x, AcceleratorRead);
-  auto y_v = y.View(AcceleratorRead);
+  autoView( y_v , y, AcceleratorRead);
   accelerator_for(ss,x_v.size(),vobj::Nsimd(),{
     auto tmp = a*x_v(ss)+y_v(ss);
     coalescedWrite(ret_v[ss],tmp);
@@ -233,9 +233,9 @@ void axpby(Lattice<vobj> &ret,sobj a,sobj b,const Lattice<vobj> &x,const Lattice
   ret.Checkerboard() = x.Checkerboard();
   conformable(ret,x);
   conformable(x,y);
-  auto ret_v = ret.View(AcceleratorWrite);
+  autoView( ret_v , ret, AcceleratorWrite);
-  auto x_v = x.View(AcceleratorRead);
+  autoView( x_v , x, AcceleratorRead);
-  auto y_v = y.View(AcceleratorRead);
+  autoView( y_v , y, AcceleratorRead);
   accelerator_for(ss,x_v.size(),vobj::Nsimd(),{
     auto tmp = a*x_v(ss)+b*y_v(ss);
     coalescedWrite(ret_v[ss],tmp);

Grid/lattice/Lattice_base.h

@@ -84,6 +84,7 @@ public:
   /////////////////////////////////////////////////////////////////////////////////
   void SetViewMode(ViewMode mode) {
     LatticeView<vobj> accessor(*( (LatticeAccelerator<vobj> *) this),mode);
+    accessor.ViewClose();
   }
   /////////////////////////////////////////////////////////////////////////////////
   // Return a view object that may be dereferenced in site loops.
@@ -123,6 +124,7 @@ public:
       auto tmp = eval(ss,exprCopy);
       vstream(me[ss],tmp);
     });
+    me.ViewClose();
     ExpressionViewClose(exprCopy);
     return *this;
   }
@@ -145,6 +147,7 @@ public:
       auto tmp = eval(ss,exprCopy);
       vstream(me[ss],tmp);
     });
+    me.ViewClose();
     ExpressionViewClose(exprCopy);
     return *this;
   }
@@ -166,6 +169,7 @@ public:
       auto tmp = eval(ss,exprCopy);
       vstream(me[ss],tmp);
     });
+    me.ViewClose();
     ExpressionViewClose(exprCopy);
     return *this;
   }
@@ -221,6 +225,7 @@ public:
     thread_for(ss,me.size(),{
 	me[ss]= r;
     });
+    me.ViewClose();
     return *this;
   }
 
@@ -278,6 +283,7 @@ public:
     accelerator_for(ss,me.size(),vobj::Nsimd(),{
       coalescedWrite(me[ss],him(ss));
     });
+    me.ViewClose();    him.ViewClose();
     return *this;
   }
 
@@ -292,6 +298,7 @@ public:
     accelerator_for(ss,me.size(),vobj::Nsimd(),{
       coalescedWrite(me[ss],him(ss));
     });
+    me.ViewClose();    him.ViewClose();
     return *this;
   }
   ///////////////////////////////////////////

Grid/lattice/Lattice_comparison.h

@@ -78,9 +78,9 @@ template<class vfunctor,class lobj,class robj>
 inline Lattice<vPredicate> LLComparison(vfunctor op,const Lattice<lobj> &lhs,const Lattice<robj> &rhs)
 {
   Lattice<vPredicate> ret(rhs.Grid());
-  auto lhs_v = lhs.View(CpuRead);
+  autoView( lhs_v, lhs, CpuRead);
-  auto rhs_v = rhs.View(CpuRead);
+  autoView( rhs_v, rhs, CpuRead);
-  auto ret_v = ret.View(CpuWrite);
+  autoView( ret_v, ret, CpuWrite);
   thread_for( ss, rhs_v.size(), {
       ret_v[ss]=op(lhs_v[ss],rhs_v[ss]);
   });
@@ -93,8 +93,8 @@ template<class vfunctor,class lobj,class robj>
 inline Lattice<vPredicate> LSComparison(vfunctor op,const Lattice<lobj> &lhs,const robj &rhs)
 {
   Lattice<vPredicate> ret(lhs.Grid());
-  auto lhs_v = lhs.View(CpuRead);
+  autoView( lhs_v, lhs, CpuRead);
-  auto ret_v = ret.View(CpuWrite);
+  autoView( ret_v, ret, CpuWrite);
   thread_for( ss, lhs_v.size(), {
     ret_v[ss]=op(lhs_v[ss],rhs);
   });
@@ -107,8 +107,8 @@ template<class vfunctor,class lobj,class robj>
 inline Lattice<vPredicate> SLComparison(vfunctor op,const lobj &lhs,const Lattice<robj> &rhs)
 {
   Lattice<vPredicate> ret(rhs.Grid());
-  auto rhs_v = rhs.View(CpuRead);
+  autoView( rhs_v, rhs, CpuRead);
-  auto ret_v = ret.View(CpuWrite);
+  autoView( ret_v, ret, CpuWrite);
   thread_for( ss, rhs_v.size(), {
     ret_v[ss]=op(lhs,rhs_v[ss]);
   });

Grid/lattice/Lattice_coordinate.h

@@ -37,7 +37,7 @@ template<class iobj> inline void LatticeCoordinate(Lattice<iobj> &l,int mu)
   GridBase *grid = l.Grid();
   int Nsimd = grid->iSites();
 
-  auto l_v = l.View(CpuWrite);
+  autoView(l_v, l, CpuWrite);
   thread_for( o, grid->oSites(), {
     vector_type vI;
     Coordinate gcoor;

Grid/lattice/Lattice_local.h

@@ -43,8 +43,8 @@ template<class vobj>
 inline auto localNorm2 (const Lattice<vobj> &rhs)-> Lattice<typename vobj::tensor_reduced>
 {
   Lattice<typename vobj::tensor_reduced> ret(rhs.Grid());
-  auto rhs_v = rhs.View(AcceleratorRead);
+  autoView( rhs_v , rhs, AcceleratorRead);
-  auto ret_v = ret.View(AcceleratorWrite);
+  autoView( ret_v , ret, AcceleratorWrite);
   accelerator_for(ss,rhs_v.size(),vobj::Nsimd(),{
     coalescedWrite(ret_v[ss],innerProduct(rhs_v(ss),rhs_v(ss)));
   });
@@ -56,9 +56,9 @@ template<class vobj>
 inline auto localInnerProduct (const Lattice<vobj> &lhs,const Lattice<vobj> &rhs) -> Lattice<typename vobj::tensor_reduced>
 {
   Lattice<typename vobj::tensor_reduced> ret(rhs.Grid());
-  auto lhs_v = lhs.View(AcceleratorRead);
+  autoView( lhs_v , lhs, AcceleratorRead);
-  auto rhs_v = rhs.View(AcceleratorRead);
+  autoView( rhs_v , rhs, AcceleratorRead);
-  auto ret_v = ret.View(AcceleratorWrite);
+  autoView( ret_v , ret, AcceleratorWrite);
   accelerator_for(ss,rhs_v.size(),vobj::Nsimd(),{
     coalescedWrite(ret_v[ss],innerProduct(lhs_v(ss),rhs_v(ss)));
   });
@@ -73,9 +73,9 @@ inline auto outerProduct (const Lattice<ll> &lhs,const Lattice<rr> &rhs) -> Latt
   typedef decltype(coalescedRead(ll())) sll;
   typedef decltype(coalescedRead(rr())) srr;
   Lattice<decltype(outerProduct(ll(),rr()))> ret(rhs.Grid());
-  auto lhs_v = lhs.View(AcceleratorRead);
+  autoView( lhs_v , lhs, AcceleratorRead);
-  auto rhs_v = rhs.View(AcceleratorRead);
+  autoView( rhs_v , rhs, AcceleratorRead);
-  auto ret_v = ret.View(AcceleratorWrite);
+  autoView( ret_v , ret, AcceleratorWrite);
   accelerator_for(ss,rhs_v.size(),1,{
     // FIXME had issues with scalar version of outer 
     // Use vector [] operator and don't read coalesce this loop

Grid/lattice/Lattice_matrix_reduction.h

@@ -51,9 +51,9 @@ static void sliceMaddMatrix (Lattice<vobj> &R,Eigen::MatrixXcd &aa,const Lattice
   int block =FullGrid->_slice_block [Orthog];
   int nblock=FullGrid->_slice_nblock[Orthog];
   int ostride=FullGrid->_ostride[Orthog];
-  auto X_v = X.View(CpuRead);
+  autoView( X_v , X, CpuRead);
-  auto Y_v = Y.View(CpuRead);
+  autoView( Y_v , Y, CpuRead);
-  auto R_v = R.View(CpuWrite);
+  autoView( R_v , R, CpuWrite);
   thread_region
   {
     std::vector<vobj> s_x(Nblock);
@@ -97,8 +97,8 @@ static void sliceMulMatrix (Lattice<vobj> &R,Eigen::MatrixXcd &aa,const Lattice<
   int nblock=FullGrid->_slice_nblock[Orthog];
   int ostride=FullGrid->_ostride[Orthog];
 
-  auto X_v = X.View(CpuRead);
+  autoView( X_v , X, CpuRead);
-  auto R_v = R.View(CpuWrite);
+  autoView( R_v , R, CpuWrite);
 
   thread_region
   {
@@ -156,8 +156,8 @@ static void sliceInnerProductMatrix(  Eigen::MatrixXcd &mat, const Lattice<vobj>
   int ostride=FullGrid->_ostride[Orthog];
 
   typedef typename vobj::vector_typeD vector_typeD;
-  auto lhs_v = lhs.View(CpuRead);
+  autoView( lhs_v , lhs, CpuRead);
-  auto rhs_v = rhs.View(CpuRead);
+  autoView( rhs_v , rhs, CpuRead);
   thread_region {
     std::vector<vobj> Left(Nblock);
     std::vector<vobj> Right(Nblock);

Grid/lattice/Lattice_peekpoke.h

@@ -46,8 +46,8 @@ auto PeekIndex(const Lattice<vobj> &lhs,int i) -> Lattice<decltype(peekIndex<Ind
 {
   Lattice<decltype(peekIndex<Index>(vobj(),i))> ret(lhs.Grid());
   ret.Checkerboard()=lhs.Checkerboard();
-  auto ret_v = ret.View(CpuWrite);
+  autoView( ret_v, ret, CpuWrite);
-  auto lhs_v = lhs.View(CpuRead);
+  autoView( lhs_v, lhs, CpuRead);
   thread_for( ss, lhs_v.size(), {
     ret_v[ss] = peekIndex<Index>(lhs_v[ss],i);
   });
@@ -58,8 +58,8 @@ auto PeekIndex(const Lattice<vobj> &lhs,int i,int j) -> Lattice<decltype(peekInd
 {
   Lattice<decltype(peekIndex<Index>(vobj(),i,j))> ret(lhs.Grid());
   ret.Checkerboard()=lhs.Checkerboard();
-  auto ret_v = ret.View(CpuWrite);
+  autoView( ret_v, ret, CpuWrite);
-  auto lhs_v = lhs.View(CpuRead);
+  autoView( lhs_v, lhs, CpuRead);
   thread_for( ss, lhs_v.size(), {
     ret_v[ss] = peekIndex<Index>(lhs_v[ss],i,j);
   });
@@ -72,8 +72,8 @@ auto PeekIndex(const Lattice<vobj> &lhs,int i,int j) -> Lattice<decltype(peekInd
 template<int Index,class vobj>  
 void PokeIndex(Lattice<vobj> &lhs,const Lattice<decltype(peekIndex<Index>(vobj(),0))> & rhs,int i)
 {
-  auto rhs_v = rhs.View(CpuRead);
+  autoView( rhs_v, rhs, CpuRead);
-  auto lhs_v = lhs.View(CpuWrite);
+  autoView( lhs_v, lhs, CpuWrite);
   thread_for( ss, lhs_v.size(), {
     pokeIndex<Index>(lhs_v[ss],rhs_v[ss],i);
   });
@@ -81,8 +81,8 @@ void PokeIndex(Lattice<vobj> &lhs,const Lattice<decltype(peekIndex<Index>(vobj()
 template<int Index,class vobj> 
 void PokeIndex(Lattice<vobj> &lhs,const Lattice<decltype(peekIndex<Index>(vobj(),0,0))> & rhs,int i,int j)
 {
-  auto rhs_v = rhs.View(CpuRead);
+  autoView( rhs_v, rhs, CpuRead);
-  auto lhs_v = lhs.View(CpuWrite);
+  autoView( lhs_v, lhs, CpuWrite);
   thread_for( ss, lhs_v.size(), {
     pokeIndex<Index>(lhs_v[ss],rhs_v[ss],i,j);
   });
@@ -111,7 +111,7 @@ void pokeSite(const sobj &s,Lattice<vobj> &l,const Coordinate &site){
 
   // extract-modify-merge cycle is easiest way and this is not perf critical
   ExtractBuffer<sobj> buf(Nsimd);
-  auto l_v = l.View(CpuWrite);
+  autoView( l_v , l, CpuWrite);
   if ( rank == grid->ThisRank() ) {
     extract(l_v[odx],buf);
     buf[idx] = s;
@@ -141,7 +141,7 @@ void peekSite(sobj &s,const Lattice<vobj> &l,const Coordinate &site){
   grid->GlobalCoorToRankIndex(rank,odx,idx,site);
 
   ExtractBuffer<sobj> buf(Nsimd);
-  auto l_v = l.View(CpuWrite);
+  autoView( l_v , l, CpuWrite);
   extract(l_v[odx],buf);
 
   s = buf[idx];
@@ -173,7 +173,7 @@ inline void peekLocalSite(sobj &s,const Lattice<vobj> &l,Coordinate &site){
   idx= grid->iIndex(site);
   odx= grid->oIndex(site);
   
-  auto l_v = l.View(CpuRead);
+  autoView( l_v , l, CpuRead);
   scalar_type * vp = (scalar_type *)&l_v[odx];
   scalar_type * pt = (scalar_type *)&s;
       
@@ -202,7 +202,7 @@ inline void pokeLocalSite(const sobj &s,Lattice<vobj> &l,Coordinate &site){
   idx= grid->iIndex(site);
   odx= grid->oIndex(site);
 
-  auto l_v = l.View(CpuWrite);
+  autoView( l_v , l, CpuWrite);
   scalar_type * vp = (scalar_type *)&l_v[odx];
   scalar_type * pt = (scalar_type *)&s;
   for(int w=0;w<words;w++){

Grid/lattice/Lattice_reality.h

@@ -40,8 +40,8 @@ NAMESPACE_BEGIN(Grid);
 
 template<class vobj> inline Lattice<vobj> adj(const Lattice<vobj> &lhs){
   Lattice<vobj> ret(lhs.Grid());
-  auto lhs_v = lhs.View(AcceleratorRead);
+  autoView( lhs_v, lhs, AcceleratorRead);
-  auto ret_v = ret.View(AcceleratorWrite);
+  autoView( ret_v, ret, AcceleratorWrite);
   accelerator_for( ss, lhs_v.size(), vobj::Nsimd(), {
     coalescedWrite(ret_v[ss], adj(lhs_v(ss)));
   });
@@ -50,8 +50,8 @@ template<class vobj> inline Lattice<vobj> adj(const Lattice<vobj> &lhs){
 
 template<class vobj> inline Lattice<vobj> conjugate(const Lattice<vobj> &lhs){
   Lattice<vobj> ret(lhs.Grid());
-  auto lhs_v = lhs.View(AcceleratorRead);
+  autoView( lhs_v, lhs, AcceleratorRead);
-  auto ret_v = ret.View(AcceleratorWrite);
+  autoView( ret_v, ret, AcceleratorWrite);
   accelerator_for( ss, lhs_v.size(), vobj::Nsimd(), {
     coalescedWrite( ret_v[ss] , conjugate(lhs_v(ss)));
   });

Grid/lattice/Lattice_reduction.h

@@ -79,11 +79,11 @@ template<class vobj>
 inline typename vobj::scalar_object sum(const Lattice<vobj> &arg)
 {
 #if defined(GRID_CUDA)||defined(GRID_HIP)
-  auto arg_v = arg.View(AcceleratorRead);
+  autoView( arg_v, arg, AcceleratorRead);
   Integer osites = arg.Grid()->oSites();
   auto ssum= sum_gpu(&arg_v[0],osites);
 #else
-  auto arg_v = arg.View(CpuRead);
+  autoView(arg_v, arg, CpuRead);
   Integer osites = arg.Grid()->oSites();
   auto ssum= sum_cpu(&arg_v[0],osites);
 #endif  
@@ -113,8 +113,8 @@ inline ComplexD innerProduct(const Lattice<vobj> &left,const Lattice<vobj> &righ
   const uint64_t sites = grid->oSites();
   
   // Might make all code paths go this way.
-  auto left_v = left.View(AcceleratorRead);
+  autoView( left_v , left, AcceleratorRead);
-  auto right_v=right.View(AcceleratorRead);
+  autoView( right_v,right, AcceleratorRead);
 
   // GPU - SIMT lane compliance...
   typedef decltype(innerProduct(left_v[0],right_v[0])) inner_t;
@@ -168,9 +168,9 @@ axpby_norm_fast(Lattice<vobj> &z,sobj a,sobj b,const Lattice<vobj> &x,const Latt
   const uint64_t sites = grid->oSites();
   
   // GPU
-  auto x_v=x.View(AcceleratorRead);
+  autoView( x_v, x, AcceleratorRead);
-  auto y_v=y.View(AcceleratorRead);
+  autoView( y_v, y, AcceleratorRead);
-  auto z_v=z.View(AcceleratorWrite);
+  autoView( z_v, z, AcceleratorWrite);
 
   typedef decltype(innerProduct(x_v[0],y_v[0])) inner_t;
   Vector<inner_t> inner_tmp(sites);
@@ -257,7 +257,7 @@ template<class vobj> inline void sliceSum(const Lattice<vobj> &Data,std::vector<
 
   // sum over reduced dimension planes, breaking out orthog dir
   // Parallel over orthog direction
-  auto Data_v=Data.View(CpuRead);
+  autoView( Data_v, Data, CpuRead);
   thread_for( r,rd, {
     int so=r*grid->_ostride[orthogdim]; // base offset for start of plane 
     for(int n=0;n<e1;n++){
@@ -335,8 +335,8 @@ static void sliceInnerProductVector( std::vector<ComplexD> & result, const Latti
   int e2=    grid->_slice_block [orthogdim];
   int stride=grid->_slice_stride[orthogdim];
 
-  auto lhv=lhs.View(CpuRead);
+  autoView( lhv, lhs, CpuRead);
-  auto rhv=rhs.View(CpuRead);
+  autoView( rhv, rhs, CpuRead);
   thread_for( r,rd,{
 
     int so=r*grid->_ostride[orthogdim]; // base offset for start of plane 
@@ -443,9 +443,9 @@ static void sliceMaddVector(Lattice<vobj> &R,std::vector<RealD> &a,const Lattice
 
     tensor_reduced at; at=av;
 
-    auto Rv=R.View(CpuWrite);
+    autoView( Rv, R, CpuWrite);
-    auto Xv=X.View(CpuRead);
+    autoView( Xv, X, CpuRead);
-    auto Yv=Y.View(CpuRead);
+    autoView( Yv, Y, CpuRead);
     thread_for2d( n, e1, b,e2, {
 	int ss= so+n*stride+b;
 	Rv[ss] = at*Xv[ss]+Yv[ss];
@@ -501,9 +501,9 @@ static void sliceMaddMatrix (Lattice<vobj> &R,Eigen::MatrixXcd &aa,const Lattice
   int nblock=FullGrid->_slice_nblock[Orthog];
   int ostride=FullGrid->_ostride[Orthog];
 
-  auto X_v=X.View(CpuRead);
+  autoView( X_v, X, CpuRead);
-  auto Y_v=Y.View(CpuRead);
+  autoView( Y_v, Y, CpuRead);
-  auto R_v=R.View(CpuWrite);
+  autoView( R_v, R, CpuWrite);
   thread_region
   {
     Vector<vobj> s_x(Nblock);
@@ -554,8 +554,8 @@ static void sliceMulMatrix (Lattice<vobj> &R,Eigen::MatrixXcd &aa,const Lattice<
   int block =FullGrid->_slice_block [Orthog];
   int nblock=FullGrid->_slice_nblock[Orthog];
   int ostride=FullGrid->_ostride[Orthog];
-  auto R_v = R.View(CpuWrite);
+  autoView( R_v, R, CpuWrite);
-  auto X_v = X.View(CpuRead);
+  autoView( X_v, X, CpuRead);
   thread_region
   {
     std::vector<vobj> s_x(Nblock);
@@ -613,8 +613,8 @@ static void sliceInnerProductMatrix(  Eigen::MatrixXcd &mat, const Lattice<vobj>
 
   typedef typename vobj::vector_typeD vector_typeD;
 
-  auto lhs_v=lhs.View(CpuRead);
+  autoView( lhs_v, lhs, CpuRead);
-  auto rhs_v=rhs.View(CpuRead);
+  autoView( rhs_v, rhs, CpuRead);
   thread_region
   {
     std::vector<vobj> Left(Nblock);

Grid/lattice/Lattice_rng.h

@@ -375,7 +375,7 @@ public:
     int osites = _grid->oSites();  // guaranteed to be <= l.Grid()->oSites() by a factor multiplicity
     int words  = sizeof(scalar_object) / sizeof(scalar_type);
 
-    auto l_v = l.View(CpuWrite);
+    autoView(l_v, l, CpuWrite);
     thread_for( ss, osites, {
       ExtractBuffer<scalar_object> buf(Nsimd);
       for (int m = 0; m < multiplicity; m++) {  // Draw from same generator multiplicity times

Grid/lattice/Lattice_trace.h

@@ -41,8 +41,8 @@ template<class vobj>
 inline auto trace(const Lattice<vobj> &lhs)  -> Lattice<decltype(trace(vobj()))>
 {
   Lattice<decltype(trace(vobj()))> ret(lhs.Grid());
-  auto ret_v = ret.View(AcceleratorWrite);
+  autoView(ret_v , ret, AcceleratorWrite);
-  auto lhs_v = lhs.View(AcceleratorRead);
+  autoView(lhs_v , lhs, AcceleratorRead);
   accelerator_for( ss, lhs_v.size(), vobj::Nsimd(), {
     coalescedWrite(ret_v[ss], trace(lhs_v(ss)));
   });
@@ -56,8 +56,8 @@ template<int Index,class vobj>
 inline auto TraceIndex(const Lattice<vobj> &lhs) -> Lattice<decltype(traceIndex<Index>(vobj()))>
 {
   Lattice<decltype(traceIndex<Index>(vobj()))> ret(lhs.Grid());
-  auto ret_v = ret.View(AcceleratorWrite);
+  autoView( ret_v , ret, AcceleratorWrite);
-  auto lhs_v = lhs.View(AcceleratorRead);
+  autoView( lhs_v , lhs, AcceleratorRead);
   accelerator_for( ss, lhs_v.size(), vobj::Nsimd(), {
     coalescedWrite(ret_v[ss], traceIndex<Index>(lhs_v(ss)));
   });

Grid/lattice/Lattice_transfer.h

@@ -46,11 +46,12 @@ inline void subdivides(GridBase *coarse,GridBase *fine)
 ////////////////////////////////////////////////////////////////////////////////////////////
 // remove and insert a half checkerboard
 ////////////////////////////////////////////////////////////////////////////////////////////
-template<class vobj> inline void pickCheckerboard(int cb,Lattice<vobj> &half,const Lattice<vobj> &full){
+template<class vobj> inline void pickCheckerboard(int cb,Lattice<vobj> &half,const Lattice<vobj> &full)
+{
   half.Checkerboard() = cb;
 
-  auto half_v = half.View(CpuWrite);
+  autoView( half_v, half, CpuWrite);
-  auto full_v = full.View(CpuRead);
+  autoView( full_v, full, CpuRead);
   thread_for(ss, full.Grid()->oSites(),{
     int cbos;
     Coordinate coor;
@@ -63,10 +64,11 @@ template<class vobj> inline void pickCheckerboard(int cb,Lattice<vobj> &half,con
     }
   });
 }
-template<class vobj> inline void setCheckerboard(Lattice<vobj> &full,const Lattice<vobj> &half){
+template<class vobj> inline void setCheckerboard(Lattice<vobj> &full,const Lattice<vobj> &half)
+{
   int cb = half.Checkerboard();
-  auto half_v = half.View(CpuRead);
+  autoView( half_v , half, CpuRead);
-  auto full_v = full.View(CpuWrite);
+  autoView( full_v , full, CpuWrite);
   thread_for(ss,full.Grid()->oSites(),{
 
     Coordinate coor;
@@ -92,79 +94,15 @@ inline void blockProject(Lattice<iVector<CComplex,nbasis > > &coarseData,
 
   Lattice<CComplex> ip(coarse); 
 
-  auto coarseData_ = coarseData.View(AcceleratorWrite);
+  autoView( coarseData_ , coarseData, AcceleratorWrite);
-  auto ip_         = ip.View(AcceleratorWrite);
+  autoView( ip_         , ip,         AcceleratorWrite);
   for(int v=0;v<nbasis;v++) {
     blockInnerProduct(ip,Basis[v],fineData);
     accelerator_for( sc, coarse->oSites(), vobj::Nsimd(), {
 	coalescedWrite(coarseData_[sc](v),ip_(sc));
-      });
+    });
   }
 }
-#if 0
-template<class vobj,class CComplex,int nbasis>
-inline void blockProject1(Lattice<iVector<CComplex,nbasis > > &coarseData,
-			 const             Lattice<vobj>   &fineData,
-			 const std::vector<Lattice<vobj> > &Basis)
-{
-  typedef iVector<CComplex,nbasis > coarseSiteData;
-  coarseSiteData elide;
-  typedef decltype(coalescedRead(elide)) ScalarComplex;
-  GridBase * fine  = fineData.Grid();
-  GridBase * coarse= coarseData.Grid();
-  int  _ndimension = coarse->_ndimension;
-
-  // checks
-  assert( nbasis == Basis.size() );
-  subdivides(coarse,fine); 
-  for(int i=0;i<nbasis;i++){
-    conformable(Basis[i],fineData);
-  }
-
-  Coordinate block_r      (_ndimension);
-  
-  for(int d=0 ; d<_ndimension;d++){
-    block_r[d] = fine->_rdimensions[d] / coarse->_rdimensions[d];
-    assert(block_r[d]*coarse->_rdimensions[d] == fine->_rdimensions[d]);
-  }
-  int blockVol = fine->oSites()/coarse->oSites();
-
-  coarseData=Zero();
-
-  auto fineData_   = fineData.View(AcceleratorRead);
-  auto coarseData_ = coarseData.View(AcceleratorWrite);
-  ////////////////////////////////////////////////////////////////////////////////////////////////////////
-  // To make this lock free, loop over coars parallel, and then loop over fine associated with coarse.
-  // Otherwise do fine inner product per site, and make the update atomic
-  ////////////////////////////////////////////////////////////////////////////////////////////////////////
-  accelerator_for( sci, nbasis*coarse->oSites(), vobj::Nsimd(), {
-
-    auto sc=sci/nbasis;
-    auto i=sci%nbasis;
-    auto Basis_      = Basis[i].View(AcceleratorRead);
-
-    Coordinate coor_c(_ndimension);
-    Lexicographic::CoorFromIndex(coor_c,sc,coarse->_rdimensions);  // Block coordinate
-
-    int sf;
-    decltype(innerProduct(Basis_(sf),fineData_(sf))) reduce=Zero();
-
-    for(int sb=0;sb<blockVol;sb++){
-
-      Coordinate coor_b(_ndimension);
-      Coordinate coor_f(_ndimension);
-
-      Lexicographic::CoorFromIndex(coor_b,sb,block_r);
-      for(int d=0;d<_ndimension;d++) coor_f[d]=coor_c[d]*block_r[d]+coor_b[d];
-      Lexicographic::IndexFromCoor(coor_f,sf,fine->_rdimensions);
-      
-      reduce=reduce+innerProduct(Basis_(sf),fineData_(sf));
-    }
-    coalescedWrite(coarseData_[sc](i),reduce);
-  });
-  return;
-}
-#endif
 
 template<class vobj,class CComplex>
 inline void blockZAXPY(Lattice<vobj> &fineZ,
@@ -191,10 +129,10 @@ inline void blockZAXPY(Lattice<vobj> &fineZ,
     assert(block_r[d]*coarse->_rdimensions[d]==fine->_rdimensions[d]);
   }
 
-  auto fineZ_  = fineZ.View(AcceleratorWrite);
+  autoView( fineZ_  , fineZ, AcceleratorWrite);
-  auto fineX_  = fineX.View(AcceleratorRead);
+  autoView( fineX_  , fineX, AcceleratorRead);
-  auto fineY_  = fineY.View(AcceleratorRead);
+  autoView( fineY_  , fineY, AcceleratorRead);
-  auto coarseA_= coarseA.View(AcceleratorRead);
+  autoView( coarseA_, coarseA, AcceleratorRead);
 
   accelerator_for(sf, fine->oSites(), CComplex::Nsimd(), {
     
@@ -229,8 +167,8 @@ inline void blockInnerProduct(Lattice<CComplex> &CoarseInner,
   // Precision promotion?
   fine_inner = localInnerProduct(fineX,fineY);
   blockSum(coarse_inner,fine_inner);
-  auto CoarseInner_  = CoarseInner.View(AcceleratorWrite);
+  autoView( CoarseInner_  , CoarseInner, AcceleratorWrite);
-  auto coarse_inner_ = coarse_inner.View(AcceleratorRead);
+  autoView( coarse_inner_ , coarse_inner, AcceleratorRead);
   accelerator_for(ss, coarse->oSites(), 1, {
     CoarseInner_[ss] = coarse_inner_[ss];
   });
@@ -265,8 +203,8 @@ inline void blockSum(Lattice<vobj> &coarseData,const Lattice<vobj> &fineData)
 
   // Turn this around to loop threaded over sc and interior loop 
   // over sf would thread better
-  auto coarseData_ = coarseData.View(AcceleratorWrite);
+  autoView( coarseData_ , coarseData, AcceleratorWrite);
-  auto fineData_   = fineData.View(AcceleratorRead);
+  autoView( fineData_   , fineData, AcceleratorRead);
 
   accelerator_for(sc,coarse->oSites(),1,{
 
@@ -359,8 +297,8 @@ inline void blockPromote(const Lattice<iVector<CComplex,nbasis > > &coarseData,
   for(int d=0 ; d<_ndimension;d++){
     block_r[d] = fine->_rdimensions[d] / coarse->_rdimensions[d];
   }
-  auto fineData_   = fineData.View(AcceleratorWrite);
+  autoView( fineData_   , fineData, AcceleratorWrite);
-  auto coarseData_ = coarseData.View(AcceleratorRead);
+  autoView( coarseData_ , coarseData, AcceleratorRead);
 
   // Loop with a cache friendly loop ordering
   accelerator_for(sf,fine->oSites(),1,{
@@ -373,7 +311,7 @@ inline void blockPromote(const Lattice<iVector<CComplex,nbasis > > &coarseData,
     Lexicographic::IndexFromCoor(coor_c,sc,coarse->_rdimensions);
 
     for(int i=0;i<nbasis;i++) {
-      /*      auto basis_ = Basis[i].View( );*/
+      /*      auto basis_ = Basis[i],  );*/
       if(i==0) fineData_[sf]=coarseData_[sc](i) *basis_[sf]);
       else     fineData_[sf]=fineData_[sf]+coarseData_[sc](i)*basis_[sf]);
     }
@@ -394,8 +332,8 @@ inline void blockPromote(const Lattice<iVector<CComplex,nbasis > > &coarseData,
   for(int i=0;i<nbasis;i++) {
     Lattice<iScalar<CComplex> > ip = PeekIndex<0>(coarseData,i);
     Lattice<CComplex> cip(coarse);
-    auto cip_ = cip.View(AcceleratorWrite);
+    autoView( cip_ , cip, AcceleratorWrite);
-    auto  ip_ =  ip.View(AcceleratorRead);
+    autoView(  ip_ ,  ip, AcceleratorRead);
     accelerator_forNB(sc,coarse->oSites(),CComplex::Nsimd(),{
 	coalescedWrite(cip_[sc], ip_(sc)());
     });
@@ -469,8 +407,8 @@ void localCopyRegion(const Lattice<vobj> &From,Lattice<vobj> & To,Coordinate Fro
   Coordinate rdt = Tg->_rdimensions;
   Coordinate ist = Tg->_istride;
   Coordinate ost = Tg->_ostride;
-  auto t_v = To.View(AcceleratorWrite);
+  autoView( t_v , To, AcceleratorWrite);
-  auto f_v = From.View(AcceleratorRead);
+  autoView( f_v , From, AcceleratorRead);
   accelerator_for(idx,Fg->lSites(),1,{
     sobj s;
     Coordinate Fcoor(nd);
@@ -717,7 +655,7 @@ unvectorizeToLexOrdArray(std::vector<sobj> &out, const Lattice<vobj> &in)
   }
 
   //loop over outer index
-  auto in_v  = in.View(CpuRead);
+  autoView( in_v  , in, CpuRead);
   thread_for(in_oidx,in_grid->oSites(),{
     //Assemble vector of pointers to output elements
     ExtractPointerArray<sobj> out_ptrs(in_nsimd);
@@ -810,7 +748,7 @@ vectorizeFromLexOrdArray( std::vector<sobj> &in, Lattice<vobj> &out)
     icoor[lane].resize(ndim);
     grid->iCoorFromIindex(icoor[lane],lane);
   }
-  auto out_v = out.View(CpuWrite);
+  autoView( out_v , out, CpuWrite);
   thread_for(oidx, grid->oSites(),{
     //Assemble vector of pointers to output elements
     ExtractPointerArray<sobj> ptrs(nsimd);
@@ -913,7 +851,7 @@ void precisionChange(Lattice<VobjOut> &out, const Lattice<VobjIn> &in)
   std::vector<SobjOut> in_slex_conv(in_grid->lSites());
   unvectorizeToLexOrdArray(in_slex_conv, in);
     
-  auto out_v = out.View(CpuWrite);
+  autoView( out_v , out, CpuWrite);
   thread_for(out_oidx,out_grid->oSites(),{
     Coordinate out_ocoor(ndim);
     out_grid->oCoorFromOindex(out_ocoor, out_oidx);

Grid/lattice/Lattice_transpose.h

@@ -41,8 +41,8 @@ NAMESPACE_BEGIN(Grid);
 template<class vobj>
 inline Lattice<vobj> transpose(const Lattice<vobj> &lhs){
   Lattice<vobj> ret(lhs.Grid());
-  auto ret_v = ret.View(AcceleratorWrite);
+  autoView( ret_v, ret, AcceleratorWrite);
-  auto lhs_v = lhs.View(AcceleratorRead);
+  autoView( lhs_v, lhs, AcceleratorRead);
   accelerator_for(ss,lhs_v.size(),vobj::Nsimd(),{
     coalescedWrite(ret_v[ss], transpose(lhs_v(ss)));
   });
@@ -56,8 +56,8 @@ template<int Index,class vobj>
 inline auto TransposeIndex(const Lattice<vobj> &lhs) -> Lattice<decltype(transposeIndex<Index>(vobj()))>
 {
   Lattice<decltype(transposeIndex<Index>(vobj()))> ret(lhs.Grid());
-  auto ret_v = ret.View(AcceleratorWrite);
+  autoView( ret_v, ret, AcceleratorWrite);
-  auto lhs_v = lhs.View(AcceleratorRead);
+  autoView( lhs_v, lhs, AcceleratorRead);
   accelerator_for(ss,lhs_v.size(),vobj::Nsimd(),{
     coalescedWrite(ret_v[ss] , transposeIndex<Index>(lhs_v(ss)));
   });

Grid/lattice/Lattice_unary.h

@@ -35,8 +35,8 @@ NAMESPACE_BEGIN(Grid);
 
 template<class obj> Lattice<obj> pow(const Lattice<obj> &rhs_i,RealD y){
   Lattice<obj> ret_i(rhs_i.Grid());
-  auto rhs = rhs_i.View(AcceleratorRead);
+  autoView( rhs, rhs_i, AcceleratorRead);
-  auto ret = ret_i.View(AcceleratorWrite);
+  autoView( ret, ret_i, AcceleratorWrite);
   ret.Checkerboard() = rhs.Checkerboard();
   accelerator_for(ss,rhs.size(),1,{
       ret[ss]=pow(rhs[ss],y);
@@ -45,8 +45,8 @@ template<class obj> Lattice<obj> pow(const Lattice<obj> &rhs_i,RealD y){
 }
 template<class obj> Lattice<obj> mod(const Lattice<obj> &rhs_i,Integer y){
   Lattice<obj> ret_i(rhs_i.Grid());
-  auto rhs = rhs_i.View(AcceleratorRead);
+  autoView( rhs , rhs_i, AcceleratorRead);
-  auto ret = ret_i.View(AcceleratorWrite);
+  autoView( ret , ret_i, AcceleratorWrite);
   ret.Checkerboard() = rhs.Checkerboard();
   accelerator_for(ss,rhs.size(),obj::Nsimd(),{
     coalescedWrite(ret[ss],mod(rhs(ss),y));
@@ -56,8 +56,8 @@ template<class obj> Lattice<obj> mod(const Lattice<obj> &rhs_i,Integer y){
 
 template<class obj> Lattice<obj> div(const Lattice<obj> &rhs_i,Integer y){
   Lattice<obj> ret_i(rhs_i.Grid());
-  auto ret = ret_i.View(AcceleratorWrite);
+  autoView( ret , ret_i, AcceleratorWrite);
-  auto rhs = rhs_i.View(AcceleratorRead);
+  autoView( rhs , rhs_i, AcceleratorRead);
   ret.Checkerboard() = rhs_i.Checkerboard();
   accelerator_for(ss,rhs.size(),obj::Nsimd(),{
     coalescedWrite(ret[ss],div(rhs(ss),y));
@@ -67,8 +67,8 @@ template<class obj> Lattice<obj> div(const Lattice<obj> &rhs_i,Integer y){
 
 template<class obj> Lattice<obj> expMat(const Lattice<obj> &rhs_i, RealD alpha, Integer Nexp = DEFAULT_MAT_EXP){
   Lattice<obj> ret_i(rhs_i.Grid());
-  auto rhs = rhs_i.View(AcceleratorRead);
+  autoView( rhs , rhs_i, AcceleratorRead);
-  auto ret = ret_i.View(AcceleratorWrite);
+  autoView( ret , ret_i, AcceleratorWrite);
   ret.Checkerboard() = rhs.Checkerboard();
   accelerator_for(ss,rhs.size(),obj::Nsimd(),{
     coalescedWrite(ret[ss],Exponentiate(rhs(ss),alpha, Nexp));

Grid/lattice/Lattice_view.h

@@ -25,6 +25,7 @@ void accelerator_inline conformable(GridBase *lhs,GridBase *rhs)
 template<class vobj> class LatticeAccelerator : public LatticeBase
 {
 protected:
+  //public:
   GridBase *_grid;
   int checkerboard;
   vobj     *_odata;    // A managed pointer
@@ -47,7 +48,7 @@ public:
 // The copy constructor for this will need to be used by device lambda functions
 /////////////////////////////////////////////////////////////////////////////////////////
 template<class vobj> 
-class LatticeExprView : public LatticeAccelerator<vobj>
+class LatticeView : public LatticeAccelerator<vobj>
 {
 public:
   // Rvalue
@@ -68,7 +69,12 @@ public:
   accelerator_inline uint64_t end(void)   const { return this->_odata_size; };
   accelerator_inline uint64_t size(void)  const { return this->_odata_size; };
 
-  LatticeExprView(const LatticeAccelerator<vobj> &refer_to_me) : LatticeAccelerator<vobj> (refer_to_me){}
+  LatticeView(const LatticeAccelerator<vobj> &refer_to_me) : LatticeAccelerator<vobj> (refer_to_me){}
+  LatticeView(const LatticeView<vobj> &refer_to_me) = default; // Trivially copyable
+  LatticeView(const LatticeAccelerator<vobj> &refer_to_me,ViewMode mode) : LatticeAccelerator<vobj> (refer_to_me)
+  {
+    this->ViewOpen(mode);
+  }
 
   // Host functions
   void ViewOpen(ViewMode mode)
@@ -89,46 +95,20 @@ public:
   }
 
 };
-
+// Little autoscope assister
-
+template<class View> 
-///////////////////////////////////////////////////////////////////////
+class ViewCloser
-// An object to be stored in a shared_ptr to clean up after last view.
-// UserView constructor,destructor updates view manager
-// Non-copyable object??? Second base with copy/= deleted?
-///////////////////////////////////////////////////////////////////////
-class MemViewDeleter {
- public:
-  void *cpu_ptr;
-  ViewMode mode;
-  ~MemViewDeleter(){
-    MemoryManager::ViewClose(cpu_ptr,mode);    
-  }
-};
-template<class vobj> 
-class LatticeView : public LatticeExprView<vobj>
 {
-#ifndef GRID_UVM
+  View v;  // Take a copy of view and call view close when I go out of scope automatically
-  std::shared_ptr<MemViewDeleter> Deleter;
+ public:
-#endif
+  ViewCloser(View &_v) : v(_v) {};
-public:
+  ~ViewCloser() { v.ViewClose(); }
-#ifdef GRID_UVM
- LatticeView(const LatticeAccelerator<vobj> &refer_to_me,ViewMode mode) : 
-    LatticeExprView<vobj> (refer_to_me)
-  {
-  }
-#else
- LatticeView(const LatticeView<vobj> &orig) : LatticeExprView<vobj>(orig) {  }
- LatticeView(const LatticeAccelerator<vobj> &refer_to_me,ViewMode mode) : 
-    LatticeExprView<vobj> (refer_to_me), Deleter(new MemViewDeleter)
-  {
-    //    std::cout << "FIXME - copy shared pointer? View Open in LatticeView"<<std::hex<<this->_odata<<std::dec<<" mode "<<mode <<std::endl;
-    this->ViewOpen(mode);
-    Deleter->cpu_ptr = this->cpu_ptr;
-    Deleter->mode    = mode;
-  }
-#endif
 };
 
+#define autoView(l_v,l,mode)				\
+	  auto l_v = l.View(mode);			\
+	  ViewCloser<decltype(l_v)> _autoView##l_v(l_v);
+
 /////////////////////////////////////////////////////////////////////////////////////////
 // Lattice expression types used by ET to assemble the AST
 // 
@@ -142,7 +122,7 @@ template <typename T> using is_lattice = std::is_base_of<LatticeBase, T>;
 template <typename T> using is_lattice_expr = std::is_base_of<LatticeExpressionBase,T >;
 
 template<class T, bool isLattice> struct ViewMapBase { typedef T Type; };
-template<class T>                 struct ViewMapBase<T,true> { typedef LatticeExprView<typename T::vector_object> Type; };
+template<class T>                 struct ViewMapBase<T,true> { typedef LatticeView<typename T::vector_object> Type; };
 template<class T> using ViewMap = ViewMapBase<T,std::is_base_of<LatticeBase, T>::value >;
 
 template <typename Op, typename _T1>                           

Grid/qcd/action/fermion/GparityWilsonImpl.h

@@ -232,15 +232,17 @@ public:
       if ( Params.twists[mu] ) { 
 	Uconj = where(coor==neglink,-Uconj,Uconj);
       }
-	  
+
-      auto U_v = U.View(CpuRead);
+      {
-      auto Uds_v = Uds.View(CpuWrite);
+	autoView( U_v , U, CpuRead);
-      auto Uconj_v = Uconj.View(CpuRead);
+	autoView( Uconj_v , Uconj, CpuRead);
-      auto Utmp_v= Utmp.View(CpuWrite);
+	autoView( Uds_v , Uds, CpuWrite);
-      thread_foreach(ss,U_v,{
+	autoView( Utmp_v, Utmp, CpuWrite);
-	Uds_v[ss](0)(mu) = U_v[ss]();
+	thread_foreach(ss,U_v,{
-	Uds_v[ss](1)(mu) = Uconj_v[ss]();
+	    Uds_v[ss](0)(mu) = U_v[ss]();
-      });
+	    Uds_v[ss](1)(mu) = Uconj_v[ss]();
+	  });
+      }
           
       U     = adj(Cshift(U    ,mu,-1));      // correct except for spanning the boundary
       Uconj = adj(Cshift(Uconj,mu,-1));
@@ -250,19 +252,25 @@ public:
 	Utmp = where(coor==0,Uconj,Utmp);
       }
 
-      thread_foreach(ss,Utmp_v,{
+      {
-	Uds_v[ss](0)(mu+4) = Utmp_v[ss]();
+	autoView( Uds_v , Uds, CpuWrite);
-      });
+	autoView( Utmp_v, Utmp, CpuWrite);
-          
+	thread_foreach(ss,Utmp_v,{
+	    Uds_v[ss](0)(mu+4) = Utmp_v[ss]();
+	  });
+      }
       Utmp = Uconj;
       if ( Params.twists[mu] ) { 
 	Utmp = where(coor==0,U,Utmp);
       }
-	  
+
-      thread_foreach(ss,Utmp_v,{
+      {	  
-        Uds_v[ss](1)(mu+4) = Utmp_v[ss]();
+	autoView( Uds_v , Uds, CpuWrite);
-      });
+	autoView( Utmp_v, Utmp, CpuWrite);
-          
+	thread_foreach(ss,Utmp_v,{
+	    Uds_v[ss](1)(mu+4) = Utmp_v[ss]();
+        });
+      }
     }
   }
       
@@ -272,11 +280,14 @@ public:
     GaugeLinkField link(mat.Grid());
     // use lorentz for flavour as hack.
     auto tmp = TraceIndex<SpinIndex>(outerProduct(Btilde, A));
-    auto link_v = link.View(CpuWrite);
+
-    auto tmp_v = tmp.View(CpuRead);
+    {
-    thread_foreach(ss,tmp_v,{
+      autoView( link_v , link, CpuWrite);
-      link_v[ss]() = tmp_v[ss](0, 0) + conjugate(tmp_v[ss](1, 1));
+      autoView( tmp_v , tmp, CpuRead);
-    });
+      thread_foreach(ss,tmp_v,{
+	  link_v[ss]() = tmp_v[ss](0, 0) + conjugate(tmp_v[ss](1, 1));
+	});
+    }
     PokeIndex<LorentzIndex>(mat, link, mu);
     return;
   }
@@ -306,16 +317,18 @@ public:
         
     GaugeLinkField tmp(mat.Grid());
     tmp = Zero();
-    auto tmp_v = tmp.View(CpuWrite);
+    {
-    auto Atilde_v = Atilde.View(CpuRead);
+      autoView( tmp_v , tmp, CpuWrite);
-    auto Btilde_v = Btilde.View(CpuRead);
+      autoView( Atilde_v , Atilde, CpuRead);
-    thread_for(ss,tmp.Grid()->oSites(),{
+      autoView( Btilde_v , Btilde, CpuRead);
-      for (int s = 0; s < Ls; s++) {
+      thread_for(ss,tmp.Grid()->oSites(),{
-	int sF = s + Ls * ss;
+	  for (int s = 0; s < Ls; s++) {
-	auto ttmp = traceIndex<SpinIndex>(outerProduct(Btilde_v[sF], Atilde_v[sF]));
+	    int sF = s + Ls * ss;
-	tmp_v[ss]() = tmp_v[ss]() + ttmp(0, 0) + conjugate(ttmp(1, 1));
+	    auto ttmp = traceIndex<SpinIndex>(outerProduct(Btilde_v[sF], Atilde_v[sF]));
-      }
+	    tmp_v[ss]() = tmp_v[ss]() + ttmp(0, 0) + conjugate(ttmp(1, 1));
-    });
+	  }
+	});
+    }
     PokeIndex<LorentzIndex>(mat, tmp, mu);
     return;
   }

Grid/qcd/action/fermion/WilsonCloverFermion.h

@@ -264,8 +264,8 @@ private:
   {
     CloverFieldType T(F.Grid());
     T = Zero();
-    auto T_v = T.View(CpuWrite);
+    autoView(T_v,T,CpuWrite);
-    auto F_v = F.View(CpuRead);
+    autoView(F_v,F,CpuRead);
     thread_for(i, CloverTerm.Grid()->oSites(),
     {
       T_v[i]()(0, 1) = timesMinusI(F_v[i]()());
@@ -282,8 +282,8 @@ private:
     CloverFieldType T(F.Grid());
     T = Zero();
     
-    auto T_v = T.View(CpuWrite);
+    autoView(T_v, T,CpuWrite);
-    auto F_v = F.View(CpuRead);
+    autoView(F_v, F,CpuRead);
     thread_for(i, CloverTerm.Grid()->oSites(),
     {
       T_v[i]()(0, 1) = -F_v[i]()();
@@ -300,8 +300,8 @@ private:
     CloverFieldType T(F.Grid());
     T = Zero();
 
-    auto T_v = T.View(CpuWrite);
+    autoView(T_v,T,CpuWrite);
-    auto F_v = F.View(CpuRead);
+    autoView(F_v,F,CpuRead);
     thread_for(i, CloverTerm.Grid()->oSites(),
     {
       T_v[i]()(0, 0) = timesMinusI(F_v[i]()());
@@ -318,8 +318,8 @@ private:
     CloverFieldType T(F.Grid());
     T = Zero();
 
-    auto T_v = T.View(CpuWrite);
+    autoView( T_v , T, CpuWrite);
-    auto F_v = F.View(CpuRead);
+    autoView( F_v , F, CpuRead);
     thread_for(i, CloverTerm.Grid()->oSites(),
     {
       T_v[i]()(0, 1) = timesI(F_v[i]()());
@@ -336,8 +336,8 @@ private:
     CloverFieldType T(F.Grid());
     T = Zero();
     
-    auto T_v = T.View(CpuWrite);
+    autoView( T_v ,T,CpuWrite);
-    auto F_v = F.View(CpuRead);
+    autoView( F_v ,F,CpuRead);
     thread_for(i, CloverTerm.Grid()->oSites(),
     {
       T_v[i]()(0, 1) = -(F_v[i]()());
@@ -355,8 +355,8 @@ private:
 
     T = Zero();
 
-    auto T_v = T.View(CpuWrite);
+    autoView( T_v , T,CpuWrite);
-    auto F_v = F.View(CpuRead);
+    autoView( F_v , F,CpuRead);
     thread_for(i, CloverTerm.Grid()->oSites(),
     {
       T_v[i]()(0, 0) = timesI(F_v[i]()());

Grid/qcd/action/fermion/WilsonImpl.h

@@ -106,9 +106,9 @@ public:
 			    const _SpinorField & phi,
 			    int mu)
   {
-    auto out_v= out.View(CpuWrite);
+    autoView( out_v, out, CpuWrite);
-    auto phi_v= phi.View(CpuRead);
+    autoView( phi_v, phi, CpuRead);
-    auto Umu_v= Umu.View(CpuRead);
+    autoView( Umu_v, Umu, CpuRead);
     thread_for(sss,out.Grid()->oSites(),{
 	multLink(out_v[sss],Umu_v[sss],phi_v[sss],mu);
     });
@@ -191,18 +191,19 @@ public:
     int Ls=Btilde.Grid()->_fdimensions[0];
     GaugeLinkField tmp(mat.Grid());
     tmp = Zero();
-    auto tmp_v = tmp.View(CpuWrite);
+    {
-    auto Btilde_v = Btilde.View(CpuRead);
+      autoView( tmp_v , tmp, CpuWrite);
-    auto Atilde_v = Atilde.View(CpuRead);
+      autoView( Btilde_v , Btilde, CpuRead);
-    thread_for(sss,tmp.Grid()->oSites(),{
+      autoView( Atilde_v , Atilde, CpuRead);
-      int sU=sss;
+      thread_for(sss,tmp.Grid()->oSites(),{
-      for(int s=0;s<Ls;s++){
+	  int sU=sss;
-	int sF = s+Ls*sU;
+	  for(int s=0;s<Ls;s++){
-	tmp_v[sU] = tmp_v[sU]+ traceIndex<SpinIndex>(outerProduct(Btilde_v[sF],Atilde_v[sF])); // ordering here
+	    int sF = s+Ls*sU;
-      }
+	    tmp_v[sU] = tmp_v[sU]+ traceIndex<SpinIndex>(outerProduct(Btilde_v[sF],Atilde_v[sF])); // ordering here
-    });
+	  }
+	});
+    }
     PokeIndex<LorentzIndex>(mat,tmp,mu);
-      
   }
 };
 

Grid/qcd/action/fermion/implementation/CayleyFermion5Dcache.h

@@ -50,9 +50,9 @@ CayleyFermion5D<Impl>::M5D(const FermionField &psi_i,
   
   chi_i.Checkerboard()=psi_i.Checkerboard();
   GridBase *grid=psi_i.Grid();
-  auto psi = psi_i.View(AcceleratorRead);
+  autoView(psi , psi_i,AcceleratorRead);
-  auto phi = phi_i.View(AcceleratorRead);
+  autoView(phi , phi_i,AcceleratorRead);
-  auto chi = chi_i.View(AcceleratorWrite);
+  autoView(chi , chi_i,AcceleratorWrite);
   assert(phi.Checkerboard() == psi.Checkerboard());
 
   auto pdiag = &diag[0];
@@ -93,9 +93,9 @@ CayleyFermion5D<Impl>::M5Ddag(const FermionField &psi_i,
 {
   chi_i.Checkerboard()=psi_i.Checkerboard();
   GridBase *grid=psi_i.Grid();
-  auto psi = psi_i.View(AcceleratorRead);
+  autoView(psi , psi_i,AcceleratorRead);
-  auto phi = phi_i.View(AcceleratorRead);
+  autoView(phi , phi_i,AcceleratorRead);
-  auto chi = chi_i.View(AcceleratorWrite);
+  autoView(chi , chi_i,AcceleratorWrite);
   assert(phi.Checkerboard() == psi.Checkerboard());
 
   auto pdiag = &diag[0];
@@ -131,8 +131,8 @@ CayleyFermion5D<Impl>::MooeeInv    (const FermionField &psi_i, FermionField &chi
   chi_i.Checkerboard()=psi_i.Checkerboard();
   GridBase *grid=psi_i.Grid();
 
-  auto psi = psi_i.View(AcceleratorRead);
+  autoView(psi , psi_i,AcceleratorRead);
-  auto chi = chi_i.View(AcceleratorWrite);
+  autoView(chi , chi_i,AcceleratorWrite);
 
   int Ls=this->Ls;
 
@@ -193,8 +193,8 @@ CayleyFermion5D<Impl>::MooeeInvDag (const FermionField &psi_i, FermionField &chi
   GridBase *grid=psi_i.Grid();
   int Ls=this->Ls;
 
-  auto psi = psi_i.View(AcceleratorRead);
+  autoView(psi , psi_i,AcceleratorRead);
-  auto chi = chi_i.View(AcceleratorWrite);
+  autoView(chi , chi_i,AcceleratorWrite);
 
   auto plee  = & lee [0];
   auto pdee  = & dee [0];

Grid/qcd/action/fermion/implementation/CayleyFermion5Dvec.h

@@ -65,9 +65,9 @@ CayleyFermion5D<Impl>::M5D(const FermionField &psi_i,
   EnableIf<Impl::LsVectorised&&EnableBool,int> sfinae=0;
   chi_i.Checkerboard()=psi_i.Checkerboard();
   GridBase *grid=psi_i.Grid();
-  auto psi = psi_i.View(CpuRead);
+  autoView(psi, psi_i,CpuRead);
-  auto phi = phi_i.View(CpuRead);
+  autoView(phi, phi_i,CpuRead);
-  auto chi = chi_i.View(CpuWrite);
+  autoView(chi, chi_i,CpuWrite);
   int Ls   = this->Ls;
   int LLs  = grid->_rdimensions[0];
   const int nsimd= Simd::Nsimd();
@@ -213,9 +213,9 @@ CayleyFermion5D<Impl>::M5Ddag(const FermionField &psi_i,
   EnableIf<Impl::LsVectorised&&EnableBool,int> sfinae=0;
   chi_i.Checkerboard()=psi_i.Checkerboard();
   GridBase *grid=psi_i.Grid();
-  auto psi=psi_i.View(CpuRead);
+  autoView(psi,psi_i,CpuRead);
-  auto phi=phi_i.View(CpuRead);
+  autoView(phi,phi_i,CpuRead);
-  auto chi=chi_i.View(CpuWrite);
+  autoView(chi,chi_i,CpuWrite);
   int Ls   = this->Ls;
   int LLs  = grid->_rdimensions[0];
   int nsimd= Simd::Nsimd();
@@ -357,8 +357,8 @@ CayleyFermion5D<Impl>::MooeeInternalAsm(const FermionField &psi_i, FermionField
 					Vector<iSinglet<Simd> > &Matm)
 {
   EnableIf<Impl::LsVectorised&&EnableBool,int> sfinae=0;
-  auto psi = psi_i.View(CpuRead);
+  autoView(psi , psi_i,CpuRead);
-  auto chi = chi_i.View(CpuWrite);
+  autoView(chi , chi_i,CpuWrite);
 #ifndef AVX512
   {
     SiteHalfSpinor BcastP;
@@ -535,8 +535,8 @@ CayleyFermion5D<Impl>::MooeeInternalZAsm(const FermionField &psi_i, FermionField
   EnableIf<Impl::LsVectorised,int> sfinae=0;
 #ifndef AVX512
   {
-    auto psi = psi_i.View(CpuRead);
+    autoView(psi , psi_i,CpuRead);
-    auto chi = chi_i.View(CpuWrite);
+    autoView(chi , chi_i,CpuWrite);
 
     SiteHalfSpinor BcastP;
     SiteHalfSpinor BcastM;
@@ -586,8 +586,8 @@ CayleyFermion5D<Impl>::MooeeInternalZAsm(const FermionField &psi_i, FermionField
   }
 #else
   {
-    auto psi = psi_i.View(CpuRead);
+    autoView(psi , psi_i,CpuRead);
-    auto chi = chi_i.View(CpuWrite);
+    autoView(chi , chi_i,CpuWrite);
     // pointers
     //  MASK_REGS;
 #define Chi_00 %zmm0

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

@@ -46,9 +46,9 @@ void DomainWallEOFAFermion<Impl>::M5D(const FermionField& psi_i, const FermionFi
   chi_i.Checkerboard() = psi_i.Checkerboard();
   int Ls = this->Ls;
   GridBase* grid = psi_i.Grid();
-  auto phi = phi_i.View(AcceleratorRead);
+  autoView( phi , phi_i, AcceleratorRead);
-  auto psi = psi_i.View(AcceleratorRead);
+  autoView( psi , psi_i, AcceleratorRead);
-  auto chi = chi_i.View(AcceleratorWrite);
+  autoView( chi , chi_i, AcceleratorWrite);
   assert(phi.Checkerboard() == psi.Checkerboard());
   auto pdiag = &diag[0];
   auto pupper = &upper[0];
@@ -82,9 +82,9 @@ void DomainWallEOFAFermion<Impl>::M5Ddag(const FermionField& psi_i, const Fermio
   GridBase* grid = psi_i.Grid();
   int Ls = this->Ls;
 
-  auto psi = psi_i.View(AcceleratorRead);
+  autoView( psi , psi_i, AcceleratorRead);
-  auto phi = phi_i.View(AcceleratorRead);
+  autoView( phi , phi_i, AcceleratorRead);
-  auto chi = chi_i.View(AcceleratorWrite);
+  autoView( chi , chi_i, AcceleratorWrite);
   assert(phi.Checkerboard() == psi.Checkerboard());
   auto pdiag = &diag[0];
   auto pupper = &upper[0];
@@ -116,8 +116,8 @@ void DomainWallEOFAFermion<Impl>::MooeeInv(const FermionField& psi_i, FermionFie
 {
   chi_i.Checkerboard() = psi_i.Checkerboard();
   GridBase* grid = psi_i.Grid();
-  auto psi=psi_i.View(AcceleratorRead);
+  autoView( psi, psi_i, AcceleratorRead);
-  auto chi=chi_i.View(AcceleratorWrite);
+  autoView( chi, chi_i, AcceleratorWrite);
   int Ls = this->Ls;
 
   auto plee  = & this->lee[0];
@@ -172,8 +172,8 @@ void DomainWallEOFAFermion<Impl>::MooeeInvDag(const FermionField& psi_i, Fermion
 {
   chi_i.Checkerboard() = psi_i.Checkerboard();
   GridBase* grid = psi_i.Grid();
-  auto psi = psi_i.View(AcceleratorRead);
+  autoView( psi, psi_i, AcceleratorRead);
-  auto chi = chi_i.View(AcceleratorWrite);
+  autoView( chi, chi_i, AcceleratorWrite);
   int Ls = this->Ls;
 
   auto plee  = & this->lee[0];

Grid/qcd/action/fermion/implementation/ImprovedStaggeredFermion5DImplementation.h

@@ -221,10 +221,10 @@ void ImprovedStaggeredFermion5D<Impl>::DhopDir(const FermionField &in, FermionFi
 
   Compressor compressor;
   Stencil.HaloExchange(in,compressor);
-  auto Umu_v   =   Umu.View(CpuRead);
+  autoView( Umu_v   ,   Umu, CpuRead);
-  auto UUUmu_v = UUUmu.View(CpuRead);
+  autoView( UUUmu_v , UUUmu, CpuRead);
-  auto in_v    =  in.View(CpuRead);
+  autoView( in_v    ,  in, CpuRead);
-  auto out_v   = out.View(CpuWrite);
+  autoView( out_v   , out, CpuWrite);
   thread_for( ss,Umu.Grid()->oSites(),{
     for(int s=0;s<Ls;s++){
       int sU=ss;

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

@@ -250,10 +250,10 @@ void ImprovedStaggeredFermion<Impl>::DerivInternal(StencilImpl &st, DoubledGauge
     ////////////////////////
     // Call the single hop
     ////////////////////////
-    auto U_v   = U.View(CpuRead);
+    autoView( U_v   , U, CpuRead);
-    auto UUU_v = UUU.View(CpuRead);
+    autoView( UUU_v , UUU, CpuRead);
-    auto B_v      = B.View(CpuWrite);
+    autoView( B_v      , B, CpuWrite);
-    auto Btilde_v = Btilde.View(CpuWrite);
+    autoView( Btilde_v , Btilde, CpuWrite);
     thread_for(sss,B.Grid()->oSites(),{
       Kernels::DhopDirKernel(st, U_v, UUU_v, st.CommBuf(), sss, sss, B_v, Btilde_v, mu,1);
     });
@@ -378,10 +378,10 @@ void ImprovedStaggeredFermion<Impl>::DhopDir(const FermionField &in, FermionFiel
 
   Compressor compressor;
   Stencil.HaloExchange(in, compressor);
-  auto Umu_v   =   Umu.View(CpuRead);
+  autoView( Umu_v   ,   Umu, CpuRead);
-  auto UUUmu_v = UUUmu.View(CpuRead);
+  autoView( UUUmu_v , UUUmu, CpuRead);
-  auto in_v    =  in.View(CpuRead);
+  autoView( in_v    ,  in, CpuRead);
-  auto out_v   = out.View(CpuWrite);
+  autoView( out_v   , out, CpuWrite);
   thread_for( sss, in.Grid()->oSites(),{
     Kernels::DhopDirKernel(Stencil, Umu_v, UUUmu_v, Stencil.CommBuf(), sss, sss, in_v, out_v, dir, disp);
   });

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

@@ -44,9 +44,9 @@ void MobiusEOFAFermion<Impl>::M5D(const FermionField &psi_i, const FermionField
   chi_i.Checkerboard() = psi_i.Checkerboard();
   GridBase *grid = psi_i.Grid();
   int Ls = this->Ls;
-  auto psi = psi_i.View(AcceleratorRead);
+  autoView(psi , psi_i, AcceleratorRead);
-  auto phi = phi_i.View(AcceleratorRead);
+  autoView(phi , phi_i, AcceleratorRead);
-  auto chi = chi_i.View(AcceleratorWrite);
+  autoView(chi , chi_i, AcceleratorWrite);
 
   assert(phi.Checkerboard() == psi.Checkerboard());
 
@@ -84,9 +84,9 @@ void MobiusEOFAFermion<Impl>::M5D_shift(const FermionField &psi_i, const Fermion
   chi_i.Checkerboard() = psi_i.Checkerboard();
   GridBase *grid = psi_i.Grid();
   int Ls = this->Ls;
-  auto psi = psi_i.View(AcceleratorRead);
+  autoView(psi , psi_i, AcceleratorRead);
-  auto phi = phi_i.View(AcceleratorRead);
+  autoView(phi , phi_i, AcceleratorRead);
-  auto chi = chi_i.View(AcceleratorWrite);
+  autoView(chi , chi_i, AcceleratorWrite);
 
   auto pm  = this->pm;
   int shift_s = (pm == 1) ? (Ls-1) : 0; // s-component modified by shift operator
@@ -132,9 +132,9 @@ void MobiusEOFAFermion<Impl>::M5Ddag(const FermionField &psi_i, const FermionFie
   chi_i.Checkerboard() = psi_i.Checkerboard();
   GridBase *grid = psi_i.Grid();
   int Ls = this->Ls;
-  auto psi = psi_i.View(AcceleratorRead);
+  autoView(psi , psi_i, AcceleratorRead);
-  auto phi = phi_i.View(AcceleratorRead);
+  autoView(phi , phi_i, AcceleratorRead);
-  auto chi = chi_i.View(AcceleratorWrite);
+  autoView(chi , chi_i, AcceleratorWrite);
 
   assert(phi.Checkerboard() == psi.Checkerboard());
 
@@ -174,9 +174,9 @@ void MobiusEOFAFermion<Impl>::M5Ddag_shift(const FermionField &psi_i, const Ferm
   GridBase *grid = psi_i.Grid();
   int Ls = this->Ls;
   int shift_s = (this->pm == 1) ? (Ls-1) : 0; // s-component modified by shift operator
-  auto psi = psi_i.View(AcceleratorRead);
+  autoView(psi , psi_i, AcceleratorRead);
-  auto phi = phi_i.View(AcceleratorRead);
+  autoView(phi , phi_i, AcceleratorRead);
-  auto chi = chi_i.View(AcceleratorWrite);
+  autoView(chi , chi_i, AcceleratorWrite);
 
   assert(phi.Checkerboard() == psi.Checkerboard());
 
@@ -226,8 +226,8 @@ void MobiusEOFAFermion<Impl>::MooeeInv(const FermionField &psi_i, FermionField &
   chi_i.Checkerboard() = psi_i.Checkerboard();
   GridBase *grid = psi_i.Grid();
   int Ls = this->Ls;
-  auto psi = psi_i.View(AcceleratorRead);
+  autoView(psi , psi_i, AcceleratorRead);
-  auto chi = chi_i.View(AcceleratorWrite);
+  autoView(chi , chi_i, AcceleratorWrite);
 
   auto plee = & this->lee [0];
   auto pdee = & this->dee [0];
@@ -286,8 +286,8 @@ void MobiusEOFAFermion<Impl>::MooeeInv_shift(const FermionField &psi_i, FermionF
   chi_i.Checkerboard() = psi_i.Checkerboard();
   GridBase *grid = psi_i.Grid();
   int Ls = this->Ls;
-  auto psi = psi_i.View(AcceleratorRead);
+  autoView(psi , psi_i, AcceleratorRead);
-  auto chi = chi_i.View(AcceleratorWrite);
+  autoView(chi , chi_i, AcceleratorWrite);
 
   auto pm = this->pm;
   auto plee = & this->lee [0];
@@ -354,8 +354,8 @@ void MobiusEOFAFermion<Impl>::MooeeInvDag(const FermionField &psi_i, FermionFiel
   chi_i.Checkerboard() = psi_i.Checkerboard();
   GridBase *grid = psi_i.Grid();
   int Ls = this->Ls;
-  auto psi = psi_i.View(AcceleratorRead);
+  autoView(psi , psi_i, AcceleratorRead);
-  auto chi = chi_i.View(AcceleratorWrite);
+  autoView(chi , chi_i, AcceleratorWrite);
 
   auto plee = & this->lee [0];
   auto pdee = & this->dee [0];
@@ -410,8 +410,8 @@ void MobiusEOFAFermion<Impl>::MooeeInvDag_shift(const FermionField &psi_i, Fermi
 {
   chi_i.Checkerboard() = psi_i.Checkerboard();
   GridBase *grid = psi_i.Grid();
-  auto psi = psi_i.View(AcceleratorRead);
+  autoView(psi , psi_i, AcceleratorRead);
-  auto chi = chi_i.View(AcceleratorWrite);
+  autoView(chi , chi_i, AcceleratorWrite);
   int Ls = this->Ls;
 
   auto pm = this->pm;

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

@@ -208,9 +208,9 @@ void NaiveStaggeredFermion<Impl>::DerivInternal(StencilImpl &st, DoubledGaugeFie
     ////////////////////////
     // Call the single hop
     ////////////////////////
-    auto U_v   = U.View(CpuRead);
+    autoView( U_v      , U, CpuRead);
-    auto B_v      = B.View(CpuWrite);
+    autoView( B_v      , B, CpuWrite);
-    auto Btilde_v = Btilde.View(CpuWrite);
+    autoView( Btilde_v , Btilde, CpuWrite);
     thread_for(sss,B.Grid()->oSites(),{
       Kernels::DhopDirKernel(st, U_v, U_v, st.CommBuf(), sss, sss, B_v, Btilde_v, mu,1);
     });
@@ -315,9 +315,9 @@ void NaiveStaggeredFermion<Impl>::DhopDir(const FermionField &in, FermionField &
 
   Compressor compressor;
   Stencil.HaloExchange(in, compressor);
-  auto Umu_v   =   Umu.View(CpuRead);
+  autoView( Umu_v   ,  Umu, CpuRead);
-  auto in_v    =  in.View(CpuRead);
+  autoView( in_v    ,  in, CpuRead);
-  auto out_v   = out.View(CpuWrite);
+  autoView( out_v   , out, CpuWrite);
   //  thread_for( sss, in.Grid()->oSites(),{
   //    Kernels::DhopDirKernel(Stencil, Umu_v, Stencil.CommBuf(), sss, sss, in_v, out_v, dir, disp);
   //  });

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

@@ -261,11 +261,11 @@ void StaggeredKernels<Impl>::DhopImproved(StencilImpl &st, LebesgueOrder &lo,
   GridBase *FGrid=in.Grid();  
   GridBase *UGrid=U.Grid();  
   typedef StaggeredKernels<Impl> ThisKernel;
-  auto UUU_v = UUU.View(AcceleratorRead);
+  autoView( UUU_v , UUU, AcceleratorRead);
-  auto U_v   =   U.View(AcceleratorRead);
+  autoView( U_v   ,   U, AcceleratorRead);
-  auto in_v  =  in.View(AcceleratorRead);
+  autoView( in_v  ,  in, AcceleratorRead);
-  auto out_v = out.View(AcceleratorWrite);
+  autoView( out_v , out, AcceleratorWrite);
-  auto st_v  =  st.View(AcceleratorRead);
+  autoView( st_v  ,  st, AcceleratorRead);
   SiteSpinor * buf = st.CommBuf();
     
   int Ls=1;
@@ -301,11 +301,11 @@ void StaggeredKernels<Impl>::DhopNaive(StencilImpl &st, LebesgueOrder &lo,
   GridBase *FGrid=in.Grid();  
   GridBase *UGrid=U.Grid();  
   typedef StaggeredKernels<Impl> ThisKernel;
-  auto UUU_v=   U.View(AcceleratorRead);
+  autoView( UUU_v ,   U, AcceleratorRead);
-  auto U_v   =   U.View(AcceleratorRead);
+  autoView( U_v   ,   U, AcceleratorRead);
-  auto in_v  =  in.View(AcceleratorRead);
+  autoView( in_v  ,  in, AcceleratorRead);
-  auto out_v = out.View(AcceleratorWrite);
+  autoView( out_v , out, AcceleratorWrite);
-  auto st_v  =  st.View(AcceleratorRead);
+  autoView( st_v  ,  st, AcceleratorRead);
   SiteSpinor * buf = st.CommBuf();
 
   int Ls=1;

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

@@ -475,12 +475,12 @@ void WilsonFermion<Impl>::ContractConservedCurrent(PropagatorField &q_in_1,
   // Inefficient comms method but not performance critical.
   tmp1 = Cshift(q_in_1, mu, 1);
   tmp2 = Cshift(q_in_2, mu, 1);
-  auto tmp1_v  =  tmp1.View(CpuWrite);
+  autoView( tmp1_v  ,  tmp1, CpuWrite);
-  auto tmp2_v  =  tmp2.View(CpuWrite);
+  autoView( tmp2_v  ,  tmp2, CpuWrite);
-  auto q_in_1_v=q_in_1.View(CpuRead);
+  autoView( q_in_1_v,q_in_1, CpuRead);
-  auto q_in_2_v=q_in_2.View(CpuRead);
+  autoView( q_in_2_v,q_in_2, CpuRead);
-  auto q_out_v = q_out.View(CpuRead);
+  autoView( q_out_v , q_out, CpuRead);
-  auto Umu_v   =   Umu.View(CpuRead);
+  autoView( Umu_v   ,   Umu, CpuRead);
   thread_for(sU, Umu.Grid()->oSites(),{
       Kernels::ContractConservedCurrentSiteFwd(tmp1_v[sU],
 					       q_in_2_v[sU],
@@ -526,11 +526,11 @@ void WilsonFermion<Impl>::SeqConservedCurrent(PropagatorField &q_in,
   tmp    = lattice_cmplx*q_in;
   tmpBwd = Cshift(tmp, mu, -1);
 
-  auto coords_v = coords.View(CpuRead);
+  autoView( coords_v , coords, CpuRead);
-  auto tmpFwd_v = tmpFwd.View(CpuRead);
+  autoView( tmpFwd_v , tmpFwd, CpuRead);
-  auto tmpBwd_v = tmpBwd.View(CpuRead);
+  autoView( tmpBwd_v , tmpBwd, CpuRead);
-  auto Umu_v    = Umu.View(CpuRead);
+  autoView( Umu_v    , Umu, CpuRead);
-  auto q_out_v  = q_out.View(CpuWrite);
+  autoView( q_out_v  , q_out, CpuWrite);
 
   thread_for(sU, Umu.Grid()->oSites(), {
 

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

@@ -348,18 +348,18 @@ template <class Impl>
 void WilsonKernels<Impl>::DhopDirAll( StencilImpl &st, DoubledGaugeField &U,SiteHalfSpinor *buf, int Ls,
 				      int Nsite, const FermionField &in, std::vector<FermionField> &out) 
 {
-   auto U_v   = U.View(AcceleratorRead);
+   autoView(U_v  ,U,AcceleratorRead);
-   auto in_v  = in.View(AcceleratorRead);
+   autoView(in_v ,in,AcceleratorRead);
-   auto st_v  = st.View(AcceleratorRead);
+   autoView(st_v ,st,AcceleratorRead);
 
-   auto out_Xm = out[0].View(AcceleratorWrite);
+   autoView(out_Xm,out[0],AcceleratorWrite);
-   auto out_Ym = out[1].View(AcceleratorWrite);
+   autoView(out_Ym,out[1],AcceleratorWrite);
-   auto out_Zm = out[2].View(AcceleratorWrite);
+   autoView(out_Zm,out[2],AcceleratorWrite);
-   auto out_Tm = out[3].View(AcceleratorWrite);
+   autoView(out_Tm,out[3],AcceleratorWrite);
-   auto out_Xp = out[4].View(AcceleratorWrite);
+   autoView(out_Xp,out[4],AcceleratorWrite);
-   auto out_Yp = out[5].View(AcceleratorWrite);
+   autoView(out_Yp,out[5],AcceleratorWrite);
-   auto out_Zp = out[6].View(AcceleratorWrite);
+   autoView(out_Zp,out[6],AcceleratorWrite);
-   auto out_Tp = out[7].View(AcceleratorWrite);
+   autoView(out_Tp,out[7],AcceleratorWrite);
    auto CBp=st.CommBuf();
    accelerator_forNB(sss,Nsite*Ls,Simd::Nsimd(),{
       int sU=sss/Ls;				
@@ -383,10 +383,10 @@ void WilsonKernels<Impl>::DhopDirKernel( StencilImpl &st, DoubledGaugeField &U,S
   assert(dirdisp<=7);
   assert(dirdisp>=0);
 
-   auto U_v   = U.View(AcceleratorRead);
+   autoView(U_v  ,U  ,AcceleratorRead);
-   auto in_v  = in.View(AcceleratorRead);
+   autoView(in_v ,in ,AcceleratorRead);
-   auto out_v = out.View(AcceleratorWrite);
+   autoView(out_v,out,AcceleratorWrite);
-   auto st_v  = st.View(AcceleratorRead);
+   autoView(st_v ,st ,AcceleratorRead);
    auto CBp=st.CommBuf();			
 #define LoopBody(Dir)				\
    case Dir :					\
@@ -438,10 +438,10 @@ void WilsonKernels<Impl>::DhopKernel(int Opt,StencilImpl &st,  DoubledGaugeField
 				     int Ls, int Nsite, const FermionField &in, FermionField &out,
 				     int interior,int exterior) 
 {
-    auto U_v   =   U.View(AcceleratorRead);
+    autoView(U_v  ,  U,AcceleratorRead);
-    auto in_v  =  in.View(AcceleratorRead);
+    autoView(in_v , in,AcceleratorRead);
-    auto out_v = out.View(AcceleratorWrite);
+    autoView(out_v,out,AcceleratorWrite);
-    auto st_v  =  st.View(AcceleratorRead);
+    autoView(st_v , st,AcceleratorRead);
 
    if( interior && exterior ) { 
      if (Opt == WilsonKernelsStatic::OptGeneric    ) { KERNEL_CALL(GenericDhopSite); return;}
@@ -469,10 +469,10 @@ void WilsonKernels<Impl>::DhopKernel(int Opt,StencilImpl &st,  DoubledGaugeField
 					  int Ls, int Nsite, const FermionField &in, FermionField &out,
 					  int interior,int exterior) 
   {
-    auto U_v   = U.View(AcceleratorRead);
+    autoView(U_v  ,U,AcceleratorRead);
-    auto in_v  = in.View(AcceleratorRead);
+    autoView(in_v ,in,AcceleratorRead);
-    auto out_v = out.View(AcceleratorWrite);
+    autoView(out_v,out,AcceleratorWrite);
-    auto st_v  = st.View(AcceleratorRead);
+    autoView(st_v ,st,AcceleratorRead);
 
    if( interior && exterior ) { 
      if (Opt == WilsonKernelsStatic::OptGeneric    ) { KERNEL_CALL(GenericDhopSiteDag); return;}

Grid/qcd/action/gauge/GaugeImplTypes.h

@@ -86,8 +86,8 @@ public:
 
   // Move this elsewhere? FIXME
   static inline void AddLink(Field &U, LinkField &W, int mu) { // U[mu] += W
-    auto U_v = U.View(CpuWrite);
+    autoView(U_v,U,CpuWrite);
-    auto W_v = W.View(CpuRead);
+    autoView(W_v,W,CpuRead);
     thread_for( ss, U.Grid()->oSites(), {
       U_v[ss](mu) = U_v[ss](mu) + W_v[ss]();
     });
@@ -131,15 +131,15 @@ public:
     //static std::chrono::duration<double> diff;
 
     //auto start = std::chrono::high_resolution_clock::now();
-    auto U_v = U.View(CpuWrite);
+    autoView(U_v,U,CpuWrite);
-    auto P_v = P.View(CpuRead);
+    autoView(P_v,P,CpuRead);
     thread_for(ss, P.Grid()->oSites(),{
       for (int mu = 0; mu < Nd; mu++) {
         U_v[ss](mu) = ProjectOnGroup(Exponentiate(P_v[ss](mu), ep, Nexp) * U_v[ss](mu));
       }
     });
     
-    //auto end = std::chrono::high_resolution_clock::now();
+   //auto end = std::chrono::high_resolution_clock::now();
    // diff += end - start;
    // std::cout << "Time to exponentiate matrix " << diff.count() << " s\n";
   }

Grid/qcd/action/scalar/ScalarInteractionAction.h

@@ -89,8 +89,8 @@ public:
     action = (2.0 * Ndim + mass_square) * phisquared - lambda * phisquared * phisquared;
     
     
-    auto p_v = p.View(CpuRead);
+    autoView( p_v      , p, CpuRead);
-    auto action_v = action.View(CpuWrite);
+    autoView( action_v , action, CpuWrite);
     for (int mu = 0; mu < Ndim; mu++)
     {
       //  pshift = Cshift(p, mu, +1);  // not efficient, implement with stencils
@@ -146,8 +146,8 @@ public:
     for (int point = 0; point < npoint; point++)
     {
 
-      auto p_v = p.View(CpuRead);
+      autoView( p_v , p, CpuRead);
-      auto force_v = force.View(CpuWrite);
+      autoView( force_v , force, CpuWrite);
             
       int permute_type;
       StencilEntry *SE;

Grid/qcd/modules/Registration.h

@@ -81,7 +81,7 @@ static Registrar<OneFlavourRatioEOFModule<FermionImplementationPolicy>,
 static Registrar< ConjugateGradientModule<WilsonFermionR::FermionField>,   
                   HMC_SolverModuleFactory<solver_string, WilsonFermionR::FermionField, Serialiser> > __CGWFmodXMLInit("ConjugateGradient"); 
 static Registrar< BiCGSTABModule<WilsonFermionR::FermionField>,   
-                  HMC_SolverModuleFactory<solver_string, WilsonFermionR::FermionField, Serialiser> > __CGWFmodXMLInit("BiCGSTAB"); 
+                  HMC_SolverModuleFactory<solver_string, WilsonFermionR::FermionField, Serialiser> > __BiCGWFmodXMLInit("BiCGSTAB"); 
 static Registrar< ConjugateResidualModule<WilsonFermionR::FermionField>,   
                   HMC_SolverModuleFactory<solver_string, WilsonFermionR::FermionField, Serialiser> > __CRWFmodXMLInit("ConjugateResidual"); 
 

Grid/qcd/utils/A2Autils.h

@@ -185,13 +185,14 @@ void A2Autils<FImpl>::MesonField(TensorType &mat,
 
 	for(int i=0;i<Lblock;i++){
 
-	  auto lhs_v = lhs_wi[i].View(CpuRead);
+	  // Recreate view potentially expensive outside fo UVM mode
+	  autoView(lhs_v,lhs_wi[i],CpuRead);
 	  auto left = conjugate(lhs_v[ss]);
-
 	  for(int j=0;j<Rblock;j++){
 
 	    SpinMatrix_v vv;
-	    auto rhs_v = rhs_vj[j].View(CpuRead);
+	    // Recreate view potentially expensive outside fo UVM mode
+	    autoView(rhs_v,rhs_vj[j],CpuRead);
 	    auto right = rhs_v[ss];
 	    for(int s1=0;s1<Ns;s1++){
 	    for(int s2=0;s2<Ns;s2++){
@@ -204,11 +205,10 @@ void A2Autils<FImpl>::MesonField(TensorType &mat,
 	    int base = Nmom*i+Nmom*Lblock*j+Nmom*Lblock*Rblock*r;
 	    for ( int m=0;m<Nmom;m++){
 	      int idx = m+base;
-	      auto mom_v = mom[m].View(CpuRead);
+	      autoView(mom_v,mom[m],CpuRead);
 	      auto phase = mom_v[ss];
 	      mac(&lvSum[idx],&vv,&phase);
 	    }
-	  
 	  }
 	}
       }
@@ -371,7 +371,7 @@ void A2Autils<FImpl>::PionFieldXX(Eigen::Tensor<ComplexD,3> &mat,
 
 	for(int i=0;i<Lblock;i++){
 
-	  auto wi_v = wi[i].View(CpuRead);
+	  autoView(wi_v,wi[i],CpuRead);
 	  auto w = conjugate(wi_v[ss]);
 	  if (g5) {
 	    w()(2)(0) = - w()(2)(0);
@@ -383,7 +383,7 @@ void A2Autils<FImpl>::PionFieldXX(Eigen::Tensor<ComplexD,3> &mat,
 	  }
 	  for(int j=0;j<Rblock;j++){
 	    
-	    auto vj_v=vj[j].View(CpuRead);
+	    autoView(vj_v,vj[j],CpuRead);
 	    auto v  = vj_v[ss];
 	    auto vv = v()(0)(0);
 
@@ -518,12 +518,12 @@ void A2Autils<FImpl>::PionFieldWVmom(Eigen::Tensor<ComplexD,4> &mat,
 
 	for(int i=0;i<Lblock;i++){
 
-	  auto wi_v = wi[i].View(CpuRead);
+	  autoView(wi_v,wi[i],CpuRead);
 	  auto w = conjugate(wi_v[ss]);
 
 	  for(int j=0;j<Rblock;j++){
-	    
+
-	    auto vj_v = vj[j].View(CpuRead);
+	    autoView(vj_v,vj[j],CpuRead);
 	    auto v = vj_v[ss];
 
 	    auto vv = w()(0)(0) * v()(0)(0)// Gamma5 Dirac basis explicitly written out
@@ -544,7 +544,7 @@ void A2Autils<FImpl>::PionFieldWVmom(Eigen::Tensor<ComplexD,4> &mat,
 	    int base = Nmom*i+Nmom*Lblock*j+Nmom*Lblock*Rblock*r;
 	    for ( int m=0;m<Nmom;m++){
 	      int idx = m+base;
-	      auto mom_v = mom[m].View(CpuRead);
+	      autoView(mom_v,mom[m],CpuRead);
 	      auto phase = mom_v[ss];
 	      mac(&lvSum[idx],&vv,&phase()()());
 	    }
@@ -730,13 +730,13 @@ void A2Autils<FImpl>::AslashField(TensorType &mat,
 
             for(int i=0;i<Lblock;i++)
             {
-  	        auto wi_v = lhs_wi[i].View(CpuRead);
+  	        autoView(wi_v,lhs_wi[i],CpuRead);
                 auto left = conjugate(wi_v[ss]);
 
                 for(int j=0;j<Rblock;j++)
                 {
                     SpinMatrix_v vv;
-		    auto vj_v  = rhs_vj[j].View(CpuRead);
+		    autoView(vj_v,rhs_vj[j],CpuRead);
                     auto right = vj_v[ss];
 
                     for(int s1=0;s1<Ns;s1++)
@@ -752,8 +752,8 @@ void A2Autils<FImpl>::AslashField(TensorType &mat,
 
                     for ( int m=0;m<Nem;m++)
                     {
-  		        auto emB0_v = emB0[m].View(CpuRead);
+  		        autoView(emB0_v,emB0[m],CpuRead);
-  		        auto emB1_v = emB1[m].View(CpuRead);
+		        autoView(emB1_v,emB1[m],CpuRead);
                         int idx  = m+base;
                         auto b0  = emB0_v[ss];
                         auto b1  = emB1_v[ss];
@@ -1014,21 +1014,21 @@ A2Autils<FImpl>::ContractWWVV(std::vector<PropagatorField> &WWVV,
     for(int d_o=0;d_o<N_d;d_o+=d_unroll){
       for(int t=0;t<N_t;t++){
       for(int s=0;s<N_s;s++){
-  auto vs_v = vs[s].View(CpuRead);
+	autoView(vs_v,vs[s],CpuRead);
-  auto tmp1 = vs_v[ss];
+	auto tmp1 = vs_v[ss];
-  vobj tmp2 = Zero();
+	vobj tmp2 = Zero();
-  vobj tmp3 = Zero();
+	vobj tmp3 = Zero();
-  for(int d=d_o;d<MIN(d_o+d_unroll,N_d);d++){
+	for(int d=d_o;d<MIN(d_o+d_unroll,N_d);d++){
-    auto vd_v = vd[d].View(CpuRead);
+	  autoView(vd_v,vd[d],CpuRead);
-    Scalar_v coeff = WW_sd(t,s,d);
+	  Scalar_v coeff = WW_sd(t,s,d);
-    tmp3 = conjugate(vd_v[ss]);
+	  tmp3 = conjugate(vd_v[ss]);
-    mac(&tmp2, &coeff, &tmp3);
+	  mac(&tmp2, &coeff, &tmp3);
-  }
+	}
 
-  //////////////////////////
+	//////////////////////////
-  // Fast outer product of tmp1 with a sum of terms suppressed by d_unroll
+	// Fast outer product of tmp1 with a sum of terms suppressed by d_unroll
-  //////////////////////////
+	//////////////////////////
-  OuterProductWWVV(WWVV[t], tmp1, tmp2, Ns, ss);
+	OuterProductWWVV(WWVV[t], tmp1, tmp2, Ns, ss);
 
       }}
     }
@@ -1067,21 +1067,20 @@ A2Autils<FImpl>::ContractWWVV(std::vector<PropagatorField> &WWVV,
     thread_for(ss,grid->oSites(),{
       for(int d_o=0;d_o<N_d;d_o+=d_unroll){
         for(int s=0;s<N_s;s++){
-    auto vs_v = vs[s].View(CpuRead);
+	  autoView(vs_v,vs[s],CpuRead);
-    auto tmp1 = vs_v[ss];
+	  auto tmp1 = vs_v[ss];
-    vobj tmp2 = Zero();
+	  vobj tmp2 = Zero();
-    vobj tmp3 = Zero();
+	  vobj tmp3 = Zero();
-    for(int d=d_o;d<MIN(d_o+d_unroll,N_d);d++){
+	  for(int d=d_o;d<MIN(d_o+d_unroll,N_d);d++){
-      auto vd_v = vd[d].View(CpuRead);
+	    autoView(vd_v,vd[d],CpuRead);
-      Scalar_v coeff = buf(s,d);
+	    Scalar_v coeff = buf(s,d);
-      tmp3 = conjugate(vd_v[ss]);
+	    tmp3 = conjugate(vd_v[ss]);
-      mac(&tmp2, &coeff, &tmp3);
+	    mac(&tmp2, &coeff, &tmp3);
-    }
+	  }
-
+	  //////////////////////////
-    //////////////////////////
+	  // Fast outer product of tmp1 with a sum of terms suppressed by d_unroll
-    // Fast outer product of tmp1 with a sum of terms suppressed by d_unroll
+	  //////////////////////////
-    //////////////////////////
+	  OuterProductWWVV(WWVV[t], tmp1, tmp2, Ns, ss);
-    OuterProductWWVV(WWVV[t], tmp1, tmp2, Ns, ss);
       }}
     });
   }
@@ -1093,7 +1092,7 @@ inline void A2Autils<FImpl>::OuterProductWWVV(PropagatorField &WWVV,
                                              const vobj &rhs,
                                              const int Ns, const int ss)
 {
-  auto WWVV_v = WWVV.View(CpuWrite);
+  autoView(WWVV_v,WWVV,CpuWrite);
   for (int s1 = 0; s1 < Ns; s1++){
     for (int s2 = 0; s2 < Ns; s2++){
       WWVV_v[ss]()(s1,s2)(0, 0) += lhs()(s1)(0) * rhs()(s2)(0);
@@ -1122,10 +1121,10 @@ void A2Autils<FImpl>::ContractFourQuarkColourDiagonal(const PropagatorField &WWV
 
   GridBase *grid = WWVV0.Grid();
 
-  auto WWVV0_v = WWVV0.View(CpuRead);
+  autoView(WWVV0_v , WWVV0,CpuRead);
-  auto WWVV1_v = WWVV1.View(CpuRead);
+  autoView(WWVV1_v , WWVV1,CpuRead);
-  auto O_trtr_v= O_trtr.View(CpuWrite);
+  autoView(O_trtr_v, O_trtr,CpuWrite);
-  auto O_fig8_v= O_fig8.View(CpuWrite);
+  autoView(O_fig8_v, O_fig8,CpuWrite);
   thread_for(ss,grid->oSites(),{
 
     typedef typename ComplexField::vector_object vobj;
@@ -1166,10 +1165,10 @@ void A2Autils<FImpl>::ContractFourQuarkColourMix(const PropagatorField &WWVV0,
 
   GridBase *grid = WWVV0.Grid();
 
-  auto WWVV0_v = WWVV0.View(CpuRead);
+  autoView( WWVV0_v , WWVV0,CpuRead);
-  auto WWVV1_v = WWVV1.View(CpuRead);
+  autoView( WWVV1_v , WWVV1,CpuRead);
-  auto O_trtr_v= O_trtr.View(CpuWrite);
+  autoView( O_trtr_v, O_trtr,CpuWrite);
-  auto O_fig8_v= O_fig8.View(CpuWrite);
+  autoView( O_fig8_v, O_fig8,CpuWrite);
 
   thread_for(ss,grid->oSites(),{
 

Grid/qcd/utils/BaryonUtils.h

@@ -273,10 +273,10 @@ void BaryonUtils<FImpl>::ContractBaryons(const PropagatorField &q1_left,
   for (int ie=0; ie < 6 ; ie++)
     wick_contraction[ie] = (quarks_left[0] == quarks_right[epsilon[ie][0]] && quarks_left[1] == quarks_right[epsilon[ie][1]] && quarks_left[2] == quarks_right[epsilon[ie][2]]) ? 1 : 0;
 
-  auto vbaryon_corr= baryon_corr.View(CpuWrite);
+  autoView(vbaryon_corr, baryon_corr,CpuWrite);
-  auto v1 = q1_left.View(CpuRead);
+  autoView( v1 , q1_left, CpuRead);
-  auto v2 = q2_left.View(CpuRead);
+  autoView( v2 , q2_left, CpuRead);
-  auto v3 = q3_left.View(CpuRead);
+  autoView( v3 , q3_left, CpuRead);
 
  // accelerator_for(ss, grid->oSites(), grid->Nsimd(), {
   thread_for(ss,grid->oSites(),{
@@ -560,10 +560,10 @@ void BaryonUtils<FImpl>::Sigma_to_Nucleon_Eye(const PropagatorField &qq_loop,
 {
   GridBase *grid = qs_ti.Grid();
 
-  auto vcorr= stn_corr.View(CpuWrite);
+  autoView( vcorr, stn_corr, CpuWrite);
-  auto vq_loop = qq_loop.View(CpuRead);
+  autoView( vq_loop , qq_loop, CpuRead);
-  auto vd_tf = qd_tf.View(CpuRead);
+  autoView( vd_tf , qd_tf, CpuRead);
-  auto vs_ti = qs_ti.View(CpuRead);
+  autoView( vs_ti , qs_ti, CpuRead);
 
  // accelerator_for(ss, grid->oSites(), grid->Nsimd(), {
   thread_for(ss,grid->oSites(),{
@@ -597,12 +597,11 @@ void BaryonUtils<FImpl>::Sigma_to_Nucleon_NonEye(const PropagatorField &qq_ti,
 {
   GridBase *grid = qs_ti.Grid();
 
-  auto vcorr= stn_corr.View(CpuWrite);
+  autoView( vcorr , stn_corr, CpuWrite);
-  auto vq_ti = qq_ti.View(CpuRead);
+  autoView( vq_ti , qq_ti, CpuRead);
-  auto vq_tf = qq_tf.View(CpuRead);
+  autoView( vq_tf , qq_tf, CpuRead);
-  auto vd_tf = qd_tf.View(CpuRead);
+  autoView( vd_tf , qd_tf, CpuRead);
-  auto vs_ti = qs_ti.View(CpuRead);
+  autoView( vs_ti , qs_ti, CpuRead);
-
  // accelerator_for(ss, grid->oSites(), grid->Nsimd(), {
   thread_for(ss,grid->oSites(),{
     auto Dq_ti = vq_ti[ss];

Grid/qcd/utils/LinalgUtils.h

@@ -47,8 +47,8 @@ void axpibg5x(Lattice<vobj> &z,const Lattice<vobj> &x,Coeff a,Coeff b)
   GridBase *grid=x.Grid();
 
   Gamma G5(Gamma::Algebra::Gamma5);
-  auto x_v = x.View(AcceleratorRead);
+  autoView(x_v, x, AcceleratorRead);
-  auto z_v = z.View(AcceleratorWrite);
+  autoView(z_v, z, AcceleratorWrite);
   accelerator_for( ss, x_v.size(),vobj::Nsimd(), {
     auto tmp = a*x_v(ss) + G5*(b*timesI(x_v(ss)));
     coalescedWrite(z_v[ss],tmp);
@@ -63,9 +63,9 @@ void axpby_ssp(Lattice<vobj> &z, Coeff a,const Lattice<vobj> &x,Coeff b,const La
   conformable(x,z);
   GridBase *grid=x.Grid();
   int Ls = grid->_rdimensions[0];
-  auto x_v = x.View(AcceleratorRead);
+  autoView( x_v, x, AcceleratorRead);
-  auto y_v = y.View(AcceleratorRead);
+  autoView( y_v, y, AcceleratorRead);
-  auto z_v = z.View(AcceleratorWrite);
+  autoView( z_v, z, AcceleratorWrite);
   // FIXME -- need a new class of accelerator_loop to implement this
   //
   uint64_t nloop = grid->oSites()/Ls;
@@ -85,9 +85,9 @@ void ag5xpby_ssp(Lattice<vobj> &z,Coeff a,const Lattice<vobj> &x,Coeff b,const L
   GridBase *grid=x.Grid();
   int Ls = grid->_rdimensions[0];
   Gamma G5(Gamma::Algebra::Gamma5);
-  auto x_v = x.View(AcceleratorRead);
+  autoView( x_v, x, AcceleratorRead);
-  auto y_v = y.View(AcceleratorRead);
+  autoView( y_v, y, AcceleratorRead);
-  auto z_v = z.View(AcceleratorWrite);
+  autoView( z_v, z, AcceleratorWrite);
   uint64_t nloop = grid->oSites()/Ls;
   accelerator_for(sss,nloop,vobj::Nsimd(),{
     uint64_t ss = sss*Ls;
@@ -104,9 +104,9 @@ void axpbg5y_ssp(Lattice<vobj> &z,Coeff a,const Lattice<vobj> &x,Coeff b,const L
   conformable(x,z);
   GridBase *grid=x.Grid();
   int Ls = grid->_rdimensions[0];
-  auto x_v = x.View(AcceleratorRead);
+  autoView( x_v, x, AcceleratorRead);
-  auto y_v = y.View(AcceleratorRead);
+  autoView( y_v, y, AcceleratorRead);
-  auto z_v = z.View(AcceleratorWrite);
+  autoView( z_v, z, AcceleratorWrite);
   Gamma G5(Gamma::Algebra::Gamma5);
   uint64_t nloop = grid->oSites()/Ls;
   accelerator_for(sss,nloop,vobj::Nsimd(),{
@@ -125,9 +125,9 @@ void ag5xpbg5y_ssp(Lattice<vobj> &z,Coeff a,const Lattice<vobj> &x,Coeff b,const
   GridBase *grid=x.Grid();
   int Ls = grid->_rdimensions[0];
 
-  auto x_v = x.View(AcceleratorRead);
+  autoView( x_v, x, AcceleratorRead);
-  auto y_v = y.View(AcceleratorRead);
+  autoView( y_v, y, AcceleratorRead);
-  auto z_v = z.View(AcceleratorWrite);
+  autoView( z_v, z, AcceleratorWrite);
   Gamma G5(Gamma::Algebra::Gamma5);
   uint64_t nloop = grid->oSites()/Ls;
   accelerator_for(sss,nloop,vobj::Nsimd(),{
@@ -147,9 +147,9 @@ void axpby_ssp_pminus(Lattice<vobj> &z,Coeff a,const Lattice<vobj> &x,Coeff b,co
   GridBase *grid=x.Grid();
   int Ls = grid->_rdimensions[0];
 
-  auto x_v = x.View(AcceleratorRead);
+  autoView( x_v, x, AcceleratorRead);
-  auto y_v = y.View(AcceleratorRead);
+  autoView( y_v, y, AcceleratorRead);
-  auto z_v = z.View(AcceleratorWrite);
+  autoView( z_v, z, AcceleratorWrite);
   uint64_t nloop = grid->oSites()/Ls;
   accelerator_for(sss,nloop,vobj::Nsimd(),{
     uint64_t ss = sss*Ls;
@@ -168,9 +168,9 @@ void axpby_ssp_pplus(Lattice<vobj> &z,Coeff a,const Lattice<vobj> &x,Coeff b,con
   conformable(x,z);
   GridBase *grid=x.Grid();
   int Ls = grid->_rdimensions[0];
-  auto x_v = x.View(AcceleratorRead);
+  autoView( x_v, x, AcceleratorRead);
-  auto y_v = y.View(AcceleratorRead);
+  autoView( y_v, y, AcceleratorRead);
-  auto z_v = z.View(AcceleratorWrite);
+  autoView( z_v, z, AcceleratorWrite);
   uint64_t nloop = grid->oSites()/Ls;
   accelerator_for(sss,nloop,vobj::Nsimd(),{
     uint64_t ss = sss*Ls;
@@ -189,8 +189,8 @@ void G5R5(Lattice<vobj> &z,const Lattice<vobj> &x)
   conformable(x,z);
   int Ls = grid->_rdimensions[0];
   Gamma G5(Gamma::Algebra::Gamma5);
-  auto x_v = x.View(AcceleratorRead);
+  autoView( x_v, x, AcceleratorRead);
-  auto z_v = z.View(AcceleratorWrite);
+  autoView( z_v, z, AcceleratorWrite);
   uint64_t nloop = grid->oSites()/Ls;
   accelerator_for(sss,nloop,vobj::Nsimd(),{
     uint64_t ss = sss*Ls;
@@ -222,8 +222,8 @@ void G5C(Lattice<iVector<CComplex, nbasis>> &z, const Lattice<iVector<CComplex,
   static_assert(nbasis % 2 == 0, "");
   int nb = nbasis / 2;
 
-  auto z_v = z.View(AcceleratorWrite);
+  autoView( z_v, z, AcceleratorWrite);
-  auto x_v = x.View(AcceleratorRead);
+  autoView( x_v, x, AcceleratorRead);
   accelerator_for(ss,grid->oSites(),CComplex::Nsimd(),
   {
     for(int n = 0; n < nb; ++n) {

Grid/qcd/utils/SUn.h

@@ -222,10 +222,10 @@ public:
     conformable(subgroup, Determinant);
     int i0, i1;
     su2SubGroupIndex(i0, i1, su2_index);
-    auto subgroup_v = subgroup.View(CpuWrite);
-    auto source_v   = source.View(CpuRead);
-    auto Determinant_v = Determinant.View(CpuWrite);
 
+    autoView( subgroup_v , subgroup,CpuWrite);
+    autoView( source_v   , source,CpuRead);
+    autoView( Determinant_v , Determinant,CpuWrite);
     thread_for(ss, grid->oSites(), {
 
       subgroup_v[ss]()()(0, 0) = source_v[ss]()()(i0, i0);
@@ -257,8 +257,8 @@ public:
     su2SubGroupIndex(i0, i1, su2_index);
 
     dest = 1.0;  // start out with identity
-    auto dest_v = dest.View(CpuWrite);
+    autoView( dest_v , dest, CpuWrite);
-    auto subgroup_v = subgroup.View(CpuRead);
+    autoView( subgroup_v, subgroup, CpuRead);
     thread_for(ss, grid->oSites(),
     {
       dest_v[ss]()()(i0, i0) = subgroup_v[ss]()()(0, 0);
@@ -266,6 +266,7 @@ public:
       dest_v[ss]()()(i1, i0) = subgroup_v[ss]()()(1, 0);
       dest_v[ss]()()(i1, i1) = subgroup_v[ss]()()(1, 1);
     });
+
   }
 
   ///////////////////////////////////////////////
@@ -608,8 +609,8 @@ public:
 
   // reunitarise??
   template <typename LatticeMatrixType>
-  static void LieRandomize(GridParallelRNG &pRNG, LatticeMatrixType &out,
+  static void LieRandomize(GridParallelRNG &pRNG, LatticeMatrixType &out, double scale = 1.0) 
-                           double scale = 1.0) {
+  {
     GridBase *grid = out.Grid();
 
     typedef typename LatticeMatrixType::vector_type vector_type;
@@ -618,8 +619,7 @@ public:
     typedef iSinglet<vector_type> vTComplexType;
 
     typedef Lattice<vTComplexType> LatticeComplexType;
-    typedef typename GridTypeMapper<
+    typedef typename GridTypeMapper<typename LatticeMatrixType::vector_object>::scalar_object MatrixType;
-      typename LatticeMatrixType::vector_object>::scalar_object MatrixType;
 
     LatticeComplexType ca(grid);
     LatticeMatrixType lie(grid);
@@ -629,6 +629,7 @@ public:
     MatrixType ta;
 
     lie = Zero();
+
     for (int a = 0; a < AdjointDimension; a++) {
       random(pRNG, ca);
 
@@ -640,6 +641,7 @@ public:
       la = ci * ca * ta;
 
       lie = lie + la;  // e^{i la ta}
+
     }
     taExp(lie, out);
   }

Grid/stencil/Stencil.h

@@ -67,6 +67,7 @@ void Gather_plane_simple_table (Vector<std::pair<int,int> >& table,const Lattice
 {
   int num=table.size();
   std::pair<int,int> *table_v = & table[0];
+
   auto rhs_v = rhs.View(AcceleratorRead);
   accelerator_forNB( i,num, vobj::Nsimd(), {
     typedef decltype(coalescedRead(buffer[0])) compressed_t;
@@ -75,6 +76,7 @@ void Gather_plane_simple_table (Vector<std::pair<int,int> >& table,const Lattice
     compress.Compress(&tmp_c,0,rhs_v(so+table_v[i].second));
     coalescedWrite(buffer[off+o],tmp_c);
   });
+  rhs_v.ViewClose();
 // Further optimisatoin: i) software prefetch the first element of the next table entry, prefetch the table
 }
 
@@ -104,6 +106,7 @@ void Gather_plane_exchange_table(Vector<std::pair<int,int> >& table,const Lattic
 			      so+tp[2*j+1].second,
 			      type);
   });
+  rhs_v.ViewClose();
 }
 
 struct StencilEntry { 
@@ -181,31 +184,30 @@ class CartesianStencilAccelerator {
 template<class vobj,class cobj,class Parameters>
 class CartesianStencilView : public CartesianStencilAccelerator<vobj,cobj,Parameters> 
 {
-#ifndef GRID_UVM
+ private:
-  std::shared_ptr<MemViewDeleter> Deleter;
+  int *closed;
-#endif
+  StencilEntry *cpu_ptr;
+  ViewMode      mode;
  public:
-  // 
+  // default copy constructor
-#ifdef GRID_UVM
+  CartesianStencilView (const CartesianStencilView &refer_to_me) = default;
-  CartesianStencilView (const CartesianStencilAccelerator<vobj,cobj,Parameters> &refer_to_me,ViewMode mode) 
-    : CartesianStencilAccelerator<vobj,cobj,Parameters>(refer_to_me){};
-#else
-  CartesianStencilView (const CartesianStencilView &refer_to_me) 
-    : CartesianStencilAccelerator<vobj,cobj,Parameters>(refer_to_me), Deleter(refer_to_me.Deleter)
-  { }
-  CartesianStencilView (const CartesianStencilAccelerator<vobj,cobj,Parameters> &refer_to_me,ViewMode mode) 
-    : CartesianStencilAccelerator<vobj,cobj,Parameters>(refer_to_me), Deleter(new MemViewDeleter)
-    {
-      Deleter->cpu_ptr =(void *)this->_entries_p;
-      Deleter->mode    = mode;
-      this->_entries_p =(StencilEntry *)
 
+  CartesianStencilView (const CartesianStencilAccelerator<vobj,cobj,Parameters> &refer_to_me,ViewMode _mode) 
+    : CartesianStencilAccelerator<vobj,cobj,Parameters>(refer_to_me),
+    cpu_ptr(this->_entries_p),
+    mode(_mode)
+  {
+    this->_entries_p =(StencilEntry *)
       MemoryManager::ViewOpen(this->_entries_p,
-				this->_npoints*this->_osites*sizeof(StencilEntry),
+			      this->_npoints*this->_osites*sizeof(StencilEntry),
-				mode,
+			      mode,
-				AdviseDefault);    
+			      AdviseDefault);    
-    }
+  }
-#endif
+  
+  void ViewClose(void)
+  {
+    MemoryManager::ViewClose(this->cpu_ptr,this->mode);    
+  }
   
 };
 

benchmarks/Benchmark_ITT.cc

@@ -252,9 +252,9 @@ public:
       double start=usecond();
       for(int i=0;i<Nloop;i++){
 	z=a*x-y;
-	auto x_v = x.View(CpuWrite);
+	autoView( x_v , x, CpuWrite);
-	auto y_v = y.View(CpuWrite);
+	autoView( y_v , y, CpuWrite);
-	auto z_v = z.View(CpuRead);
+	autoView( z_v , z, CpuRead);
         x_v[0]=z_v[0]; // force serial dependency to prevent optimise away
         y_v[4]=z_v[4];
       }
@@ -534,11 +534,13 @@ public:
     {
       LatticeGaugeField Umu5d(FGrid); 
       std::vector<LatticeColourMatrix> U(4,FGrid);
-      auto Umu_v = Umu.View(CpuRead);
+      {
-      auto Umu5d_v = Umu5d.View(CpuWrite);
+	autoView( Umu_v  , Umu  , CpuRead);
-      for(int ss=0;ss<Umu.Grid()->oSites();ss++){
+	autoView( Umu5d_v, Umu5d, CpuWrite);
-	for(int s=0;s<Ls;s++){
+	for(int ss=0;ss<Umu.Grid()->oSites();ss++){
-	  Umu5d_v[Ls*ss+s] = Umu_v[ss];
+	  for(int s=0;s<Ls;s++){
+	    Umu5d_v[Ls*ss+s] = Umu_v[ss];
+	  }
 	}
       }
       ref = Zero();

benchmarks/Benchmark_dwf.cc

@@ -129,8 +129,8 @@ int main (int argc, char ** argv)
   LatticeGaugeField Umu5d(FGrid); 
   std::vector<LatticeColourMatrix> U(4,FGrid);
   {
-    auto Umu5d_v = Umu5d.View(CpuWrite);
+    autoView( Umu5d_v, Umu5d, CpuWrite);
-    auto Umu_v = Umu.View(CpuRead);
+    autoView( Umu_v  , Umu  , CpuRead);
     for(int ss=0;ss<Umu.Grid()->oSites();ss++){
       for(int s=0;s<Ls;s++){
 	Umu5d_v[Ls*ss+s] = Umu_v[ss];
@@ -258,8 +258,8 @@ int main (int argc, char ** argv)
       //    ref =  src - Gamma(Gamma::Algebra::GammaX)* src ; // 1+gamma_x
       tmp = U[mu]*Cshift(src,mu+1,1);
       {
-	auto ref_v = ref.View(CpuWrite);
+	autoView( ref_v, ref, CpuWrite);
-	auto tmp_v = tmp.View(CpuRead);
+	autoView( tmp_v, tmp, CpuRead);
 	for(int i=0;i<ref_v.size();i++){
 	  ref_v[i]+= tmp_v[i] + Gamma(Gmu[mu])*tmp_v[i]; ;
 	}
@@ -268,8 +268,8 @@ int main (int argc, char ** argv)
       tmp =adj(U[mu])*src;
       tmp =Cshift(tmp,mu+1,-1);
       {
-	auto ref_v = ref.View(CpuWrite);
+	autoView( ref_v, ref, CpuWrite);
-	auto tmp_v = tmp.View(CpuRead);
+	autoView( tmp_v, tmp, CpuRead);
 	for(int i=0;i<ref_v.size();i++){
 	  ref_v[i]+= tmp_v[i] - Gamma(Gmu[mu])*tmp_v[i]; ;
 	}

benchmarks/Benchmark_dwf_sweep.cc

@@ -130,11 +130,13 @@ void benchDw(std::vector<int> & latt4, int Ls, int threads,int report )
   LatticeGaugeField Umu5d(FGrid); 
 
   // replicate across fifth dimension
-  auto Umu5d_v = Umu5d.View(CpuWrite);
+  {
-  auto Umu_v   = Umu.View(CpuRead);
+    autoView(Umu5d_v , Umu5d, CpuWrite);
-  for(int ss=0;ss<Umu.Grid()->oSites();ss++){
+    autoView(  Umu_v , Umu,   CpuRead);
-    for(int s=0;s<Ls;s++){
+    for(int ss=0;ss<Umu.Grid()->oSites();ss++){
-      Umu5d_v[Ls*ss+s] = Umu_v[ss];
+      for(int s=0;s<Ls;s++){
+	Umu5d_v[Ls*ss+s] = Umu_v[ss];
+      }
     }
   }
 

benchmarks/Benchmark_memory_asynch.cc

@@ -79,7 +79,7 @@ int main (int argc, char ** argv)
 
     double start=usecond();
     thread_for(t,threads,{
-      auto x_t = x[t].View(CpuRead);
+      autoView( x_t , x[t],CpuRead);
       sum[t] = x_t[0];
       for(int i=0;i<Nloop;i++){
 	for(auto ss=x_t.begin();ss<x_t.end();ss++){

benchmarks/Benchmark_memory_bandwidth.cc

@@ -177,9 +177,7 @@ int main (int argc, char ** argv)
       Real nn;      
       double start=usecond();
       for(int i=0;i<Nloop;i++){
-	auto x_v = x.View(CpuWrite);
 	nn=norm2(x);
-	vsplat(x_v[0]._internal[0],nn);
       }
       double stop=usecond();
       double time = (stop-start)/Nloop*1000;

benchmarks/Benchmark_meson_field.cc

@@ -85,11 +85,11 @@ void sliceInnerProductMesonField(std::vector< std::vector<ComplexD> > &mat,
       for(int b=0;b<e2;b++){
 	int ss= so+n*stride+b;
 	for(int i=0;i<Lblock;i++){
-	  auto lhs_v = lhs[i].View(CpuRead);
+	  autoView(lhs_v, lhs[i], CpuRead);
 	  auto left = conjugate(lhs_v[ss]);
 	  for(int j=0;j<Rblock;j++){
 	    int idx = i+Lblock*j+Lblock*Rblock*r;
-	    auto rhs_v = rhs[j].View(CpuRead);
+	    autoView(rhs_v, rhs[j], CpuRead);
 	    auto right = rhs_v[ss];
 	    vector_type vv = left()(0)(0) * right()(0)(0)
 	      +              left()(0)(1) * right()(0)(1)
@@ -221,12 +221,12 @@ void sliceInnerProductMesonFieldGamma(std::vector< std::vector<ComplexD> > &mat,
       for(int b=0;b<e2;b++){
 	int ss= so+n*stride+b;
 	for(int i=0;i<Lblock;i++){
- 	  auto lhs_v=lhs[i].View(CpuRead);
+ 	  autoView(lhs_v,lhs[i],CpuRead);
 	  auto left = conjugate(lhs_v[ss]);
 	  for(int j=0;j<Rblock;j++){
 	  for(int mu=0;mu<Ngamma;mu++){
 	    
-	    auto rhs_v = rhs[j].View(CpuRead);
+	    autoView(rhs_v,rhs[j],CpuRead);
 	    auto right = Gamma(gammas[mu])*rhs_v[ss];
 
 	      vector_type vv = left()(0)(0) * right()(0)(0)
@@ -370,12 +370,12 @@ void sliceInnerProductMesonFieldGamma1(std::vector< std::vector<ComplexD> > &mat
 	int ss= so+n*stride+b;
 	for(int i=0;i<Lblock;i++){
 	  
-	  auto lhs_v=lhs[i].View(CpuRead);
+	  autoView(lhs_v,lhs[i],CpuRead);
 	  auto left = conjugate(lhs_v[ss]);
 	  for(int j=0;j<Rblock;j++){
 
 	    SpinMatrix_v vv;
-	    auto rhs_v = rhs[j].View(CpuRead);
+	    autoView(rhs_v,rhs[j],CpuRead);
 	    auto right = rhs_v[ss];
 	    for(int s1=0;s1<Ns;s1++){
 	    for(int s2=0;s2<Ns;s2++){
@@ -518,12 +518,12 @@ void sliceInnerProductMesonFieldGammaMom(std::vector< std::vector<ComplexD> > &m
 
 	for(int i=0;i<Lblock;i++){
 
-	  auto lhs_v = lhs[i].View(CpuRead);
+	  autoView(lhs_v,lhs[i],CpuRead);
 	  auto left = conjugate(lhs_v[ss]);
 	  for(int j=0;j<Rblock;j++){
 
 	    SpinMatrix_v vv;
-	    auto rhs_v = rhs[j].View(CpuRead);
+	    autoView(rhs_v,rhs[j],CpuRead);
 	    auto right = rhs_v[ss];
 	    for(int s1=0;s1<Ns;s1++){
 	    for(int s2=0;s2<Ns;s2++){
@@ -537,7 +537,7 @@ void sliceInnerProductMesonFieldGammaMom(std::vector< std::vector<ComplexD> > &m
 	    // Trigger unroll
 	    for ( int m=0;m<Nmom;m++){
 	      int idx = m+base;
-	      auto mom_v = mom[m].View(CpuRead);
+	      autoView(mom_v,mom[m],CpuRead);
 	      auto phase = mom_v[ss];
 	      mac(&lvSum[idx],&vv,&phase);
 	    }

benchmarks/Benchmark_su3_gpu.cc

@@ -66,9 +66,9 @@ int main (int argc, char ** argv)
       LatticeColourMatrix x(&Grid);// random(pRNG,x);
       LatticeColourMatrix y(&Grid);// random(pRNG,y);
 
-      auto x_v = x.View(AcceleratorRead);
+      autoView( x_v , x, AcceleratorRead);
-      auto y_v = y.View(AcceleratorRead);
+      autoView( y_v , y, AcceleratorRead);
-      auto z_v = z.View(AcceleratorWrite);
+      autoView( z_v , z, AcceleratorWrite);
       const uint64_t Nsite = x_v.size();
       const uint64_t nsimd = vComplex::Nsimd();
       for(int64_t i=0;i<Nwarm;i++){
@@ -116,9 +116,9 @@ int main (int argc, char ** argv)
       LatticeColourMatrix x(&Grid);// random(pRNG,x);
       LatticeColourMatrix y(&Grid);// random(pRNG,y);
 
-      auto x_v = x.View(AcceleratorWrite);
+      autoView( x_v , x, AcceleratorWrite);
-      auto y_v = y.View(AcceleratorRead);
+      autoView( y_v , y, AcceleratorRead);
-      auto z_v = z.View(AcceleratorRead);
+      autoView( z_v , z, AcceleratorRead);
       const uint64_t Nsite = x_v.size();
       const uint64_t nsimd = vComplex::Nsimd();
       for(int64_t i=0;i<Nwarm;i++){
@@ -167,9 +167,9 @@ int main (int argc, char ** argv)
       LatticeColourMatrix x(&Grid);// random(pRNG,x);
       LatticeColourMatrix y(&Grid);// random(pRNG,y);
 
-      auto x_v = x.View(AcceleratorRead);
+      autoView( x_v , x, AcceleratorRead);
-      auto y_v = y.View(AcceleratorRead);
+      autoView( y_v , y, AcceleratorRead);
-      auto z_v = z.View(AcceleratorWrite);
+      autoView( z_v , z, AcceleratorWrite);
       const uint64_t Nsite = x_v.size();
       const uint64_t nsimd = vComplex::Nsimd();
       for(int64_t i=0;i<Nwarm;i++){
@@ -220,10 +220,10 @@ int main (int argc, char ** argv)
       LatticeColourMatrix y(&Grid);// random(pRNG,y);
       LatticeColourMatrix w(&Grid);// random(pRNG,y);
 
-      auto x_v = x.View(AcceleratorRead);
+      autoView( x_v , x, AcceleratorRead);
-      auto y_v = y.View(AcceleratorRead);
+      autoView( y_v , y, AcceleratorRead);
-      auto z_v = z.View(AcceleratorRead);
+      autoView( z_v , z, AcceleratorRead);
-      auto w_v = w.View(AcceleratorWrite);
+      autoView( w_v , w, AcceleratorWrite);
       const uint64_t Nsite = x_v.size();
       const uint64_t nsimd = vComplex::Nsimd();
       for(int64_t i=0;i<Nwarm;i++){

benchmarks/Benchmark_wilson.cc

@@ -125,8 +125,8 @@ int main (int argc, char ** argv)
       //    ref =  src + Gamma(Gamma::Algebra::GammaX)* src ; // 1-gamma_x
       tmp = U[mu]*Cshift(src,mu,1);
       {
-	auto ref_v = ref.View(CpuWrite);
+	autoView( ref_v, ref, CpuWrite);
-	auto tmp_v = tmp.View(CpuWrite);
+	autoView( tmp_v, tmp, CpuWrite);
 	for(int i=0;i<ref_v.size();i++){
 	  ref_v[i]+= tmp_v[i] - Gamma(Gmu[mu])*tmp_v[i]; ;
 	}
@@ -135,8 +135,8 @@ int main (int argc, char ** argv)
       tmp =adj(U[mu])*src;
       tmp =Cshift(tmp,mu,-1);
       {
-	auto ref_v = ref.View(CpuWrite);
+	autoView( ref_v, ref, CpuWrite);
-	auto tmp_v = tmp.View(CpuWrite);
+	autoView( tmp_v, tmp, CpuWrite);
 	for(int i=0;i<ref_v.size();i++){
 	  ref_v[i]+= tmp_v[i] + Gamma(Gmu[mu])*tmp_v[i]; ;
 	}
@@ -187,8 +187,8 @@ int main (int argc, char ** argv)
   for(int ss=0;ss<0;ss++ ){
     for(int i=0;i<Ns;i++){
       for(int j=0;j<Nc;j++){
-	auto ref_v = ref.View(CpuWrite);
+	autoView( ref_v, ref, CpuWrite);
-	auto result_v = result.View(CpuWrite);
+	autoView( result_v, result, CpuWrite);
 	ComplexF * ref_p = (ComplexF *)&ref_v[ss]()(i)(j);
 	ComplexF * res_p = (ComplexF *)&result_v[ss]()(i)(j);
 	std::cout<<GridLogMessage << ss<< " "<<i<<" "<<j<<" "<< (*ref_p)<<" " <<(*res_p)<<std::endl;
@@ -204,8 +204,8 @@ int main (int argc, char ** argv)
       //    ref =  src - Gamma(Gamma::Algebra::GammaX)* src ; // 1+gamma_x
       tmp = U[mu]*Cshift(src,mu,1);
       {
-	auto ref_v = ref.View(CpuWrite);
+	autoView( ref_v, ref, CpuWrite);
-	auto tmp_v = tmp.View(CpuWrite);
+	autoView( tmp_v, tmp, CpuWrite);
 	for(int i=0;i<ref_v.size();i++){
 	  ref_v[i]+= tmp_v[i] + Gamma(Gmu[mu])*tmp_v[i]; ;
 	}
@@ -214,8 +214,8 @@ int main (int argc, char ** argv)
       tmp =adj(U[mu])*src;
       tmp =Cshift(tmp,mu,-1);
       {
-	auto ref_v = ref.View(CpuWrite);
+	autoView( ref_v, ref, CpuWrite);
-	auto tmp_v = tmp.View(CpuWrite);
+	autoView( tmp_v, tmp, CpuWrite);
 	for(int i=0;i<ref_v.size();i++){
 	  ref_v[i]+= tmp_v[i] - Gamma(Gmu[mu])*tmp_v[i]; ;
 	}

tests/Test_dwf_mixedcg_prec.cc

@@ -29,7 +29,6 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 
 using namespace std;
 using namespace Grid;
- ;
 
 template<class d>
 struct scal {
@@ -51,6 +50,7 @@ int main (int argc, char ** argv)
 
   std::cout << GridLogMessage << "::::: NB: to enable a quick bit reproducibility check use the --checksums flag. " << std::endl;
 
+  { 
   GridCartesian         * UGrid   = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplexD::Nsimd()),GridDefaultMpi());
   GridRedBlackCartesian * UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
   GridCartesian         * FGrid   = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
@@ -100,6 +100,8 @@ int main (int argc, char ** argv)
   ConjugateGradient<LatticeFermionD> CG(1.0e-8,10000);
   CG(HermOpEO,src_o,result_o_2);
 
+  MemoryManager::Print();
+
   LatticeFermionD diff_o(FrbGrid);
   RealD diff = axpy_norm(diff_o, -1.0, result_o, result_o_2);
 
@@ -130,7 +132,9 @@ int main (int argc, char ** argv)
   std::cout << GridLogMessage << " CG checksums "<<std::hex << scidac_csuma << " "<<scidac_csumb<<std::endl;
   }
   #endif
-
+  }
   
+  MemoryManager::Print();
+
   Grid_finalize();
 }

tests/Test_general_stencil.cc

@@ -107,9 +107,8 @@ int main(int argc, char ** argv)
 	// Implement a stencil code that should agree with cshift!
 	for(int i=0;i<Check.Grid()->oSites();i++){
 	  auto SE = gStencil.GetEntry(0,i);
-	  auto check = Check.View(CpuWrite);
+	  autoView(check, Check, CpuWrite);
-	  auto foo   = Foo.View(CpuRead);
+	  autoView(  foo, Foo, CpuRead);
-
 	  // Encapsulate in a general wrapper
 	  check[i] = foo[SE->_offset];                   auto tmp=check[i];
 	  if (SE->_permute & 0x1 ) { permute(check[i],tmp,0); tmp=check[i];}
@@ -147,8 +146,8 @@ int main(int argc, char ** argv)
 	}}}}
 
 	if (nrm > 1.0e-4) {
-	  auto check = Check.View(CpuRead);
+	  autoView( check , Check, CpuRead);
-	  auto bar   = Bar.View(CpuRead);
+	  autoView(   bar ,   Bar, CpuRead);
 	  for(int i=0;i<check.size();i++){
 	    std::cout << i<<" Check "<<check[i]<< "\n"<<i<<" Bar "<<bar[i]<<std::endl;
 	  }

tests/Test_stencil.cc

@@ -109,8 +109,8 @@ int main(int argc, char ** argv) {
 	  StencilEntry *SE;
 	  SE = myStencil.GetEntry(permute_type,0,i);
 	  
-	  auto check = Check.View(CpuWrite);
+	  autoView( check , Check, CpuWrite);
-	  auto foo   = Foo.View(CpuRead);
+	  autoView( foo   , Foo, CpuRead);
 	  if ( SE->_is_local && SE->_permute )
 	    permute(check[i],foo[SE->_offset],permute_type);
 	  else if (SE->_is_local)
@@ -151,8 +151,8 @@ int main(int argc, char ** argv) {
 	}}}}
 
 	if (nrm > 1.0e-4) {
-	  auto check = Check.View(CpuRead);
+	  autoView( check , Check, CpuRead);
-	  auto bar   = Bar.View(CpuRead);
+	  autoView( bar   , Bar, CpuRead);
 	  for(int i=0;i<check.size();i++){
 	    std::cout << i<<" Check "<<check[i]<< "\n"<<i<<" Bar "<<bar[i]<<std::endl;
 	  }
@@ -210,8 +210,8 @@ int main(int argc, char ** argv) {
 	  SE = EStencil.GetEntry(permute_type,0,i);
 	  //	  std::cout << "Even source "<< i<<" -> " <<SE->_offset << " "<< SE->_is_local<<std::endl;
 
-	  auto ocheck = OCheck.View(CpuWrite);
+	  autoView( ocheck , OCheck, CpuWrite);
-	  auto efoo   = EFoo.View(CpuRead);
+	  autoView( efoo   , EFoo, CpuRead);
 	  if ( SE->_is_local && SE->_permute )
 	    permute(ocheck[i],efoo[SE->_offset],permute_type);
 	  else if (SE->_is_local)
@@ -226,8 +226,8 @@ int main(int argc, char ** argv) {
 	  SE = OStencil.GetEntry(permute_type,0,i);
 	  //	  std::cout << "ODD source "<< i<<" -> " <<SE->_offset << " "<< SE->_is_local<<std::endl;
 
-	  auto echeck = ECheck.View(CpuWrite);
+	  autoView( echeck , ECheck, CpuWrite);
-	  auto ofoo   = OFoo.View(CpuRead);
+	  autoView( ofoo   , OFoo, CpuRead);
 	  if ( SE->_is_local && SE->_permute )
 	    permute(echeck[i],ofoo[SE->_offset],permute_type);
 	  else if (SE->_is_local)

tests/core/Test_staggered5D.cc

@@ -89,8 +89,8 @@ int main (int argc, char ** argv)
   ////////////////////////////////////
   LatticeGaugeField Umu5d(FGrid); 
   {
-    auto umu5d = Umu5d.View(CpuWrite);
+    autoView(umu5d, Umu5d, CpuWrite);
-    auto umu   = Umu.View(CpuRead);
+    autoView(  umu, Umu  , CpuRead);
     for(int ss=0;ss<Umu.Grid()->oSites();ss++){
       for(int s=0;s<Ls;s++){
 	umu5d[Ls*ss+s] = umu[ss];

tests/debug/Test_cayley_mres.cc

@@ -570,8 +570,8 @@ void  TestConserved1(Action & Ddwf, Action & Ddwfrev,
       LatticeGaugeField Umu5d(FGrid); 
       std::vector<LatticeColourMatrix> U(4,FGrid);
       {
-	auto Umu5d_v = Umu5d.View(CpuWrite);
+        autoView( Umu5d_v , Umu5d, CpuWrite);
-	auto Umu_v = Umu.View(CpuRead);
+        autoView( Umu_v   , Umu  , CpuRead);
 	for(int ss=0;ss<Umu.Grid()->oSites();ss++){
 	  for(int s=0;s<Ls;s++){
 	    Umu5d_v[Ls*ss+s] = Umu_v[ss];
@@ -597,8 +597,8 @@ void  TestConserved1(Action & Ddwf, Action & Ddwfrev,
       {
 	RealD diag = 5.0 - Ddwf.M5;
 	mass = Ddwf.mass;
-	auto psi=result5.View(CpuRead);
+	autoView( psi,result5,CpuRead);
-	auto chi=tmp.View(CpuWrite);
+	autoView( chi,tmp,   CpuWrite);
 	thread_for(sss,UGrid->oSites(),{
 	  uint64_t ss= sss*Ls;
 	  typedef vSpinColourVector spinor;

tests/forces/Test_contfrac_force.cc

@@ -98,9 +98,9 @@ int main (int argc, char ** argv)
     PokeIndex<LorentzIndex>(mom,mommu,mu);
 
     // fourth order exponential approx
-    auto mom_v = mom.View(CpuRead);
+    autoView( mom_v, mom, CpuRead);
-    auto U_v   = U.View(CpuRead);
+    autoView( U_v , U, CpuRead);
-    auto Uprime_v = Uprime.View(CpuWrite);
+    autoView(Uprime_v, Uprime, CpuWrite);
     thread_foreach( i,mom_v,{
       Uprime_v[i](mu) =
 	  U_v[i](mu)

tests/forces/Test_dwf_force.cc

@@ -100,9 +100,9 @@ int main (int argc, char ** argv)
 
     // fourth order exponential approx
 
-    auto mom_v    = mom.View(CpuRead);
+    autoView( mom_v, mom, CpuRead);
-    auto U_v      = U.View(CpuRead);
+    autoView( U_v , U, CpuRead);
-    auto Uprime_v = Uprime.View(CpuWrite);
+    autoView(Uprime_v, Uprime, CpuWrite);
 
     thread_foreach( i,mom_v,{
       Uprime_v[i](mu) =	  U_v[i](mu)

tests/forces/Test_dwf_force_eofa.cc

@@ -110,9 +110,9 @@ int main (int argc, char** argv)
     PokeIndex<LorentzIndex>(mom, mommu, mu);
 
     // fourth order exponential approx
-    auto mom_v    = mom.View(CpuRead);
+    autoView( mom_v, mom, CpuRead);
-    auto U_v      = U.View(CpuRead);
+    autoView( U_v , U, CpuRead);
-    auto Uprime_v = Uprime.View(CpuWrite);
+    autoView(Uprime_v, Uprime, CpuWrite);
 
     thread_foreach(i,mom_v,{
       Uprime_v[i](mu) =	  U_v[i](mu)

tests/forces/Test_dwf_gpforce.cc

@@ -119,9 +119,9 @@ int main (int argc, char ** argv)
     PokeIndex<LorentzIndex>(mom,mommu,mu);
 
     // fourth order exponential approx
-    auto mom_v    = mom.View(CpuRead);
+    autoView( mom_v, mom, CpuRead);
-    auto U_v      = U.View(CpuRead);
+    autoView( U_v , U, CpuRead);
-    auto Uprime_v = Uprime.View(CpuWrite);
+    autoView(Uprime_v, Uprime, CpuWrite);
 
     thread_foreach(i,mom_v,{
       Uprime_v[i](mu) =	  U_v[i](mu)

tests/forces/Test_dwf_gpforce_eofa.cc

@@ -114,9 +114,9 @@ int main (int argc, char** argv)
     PokeIndex<LorentzIndex>(mom, mommu, mu);
 
     // fourth order exponential approx
-    auto mom_v    = mom.View(CpuRead);
+    autoView( mom_v, mom, CpuRead);
-    auto U_v      = U.View(CpuRead);
+    autoView( U_v , U, CpuRead);
-    auto Uprime_v = Uprime.View(CpuWrite);
+    autoView(Uprime_v, Uprime, CpuWrite);
 
     thread_foreach(i,mom_v,{
       Uprime_v[i](mu) =	  U_v[i](mu)

tests/forces/Test_gp_plaq_force.cc

@@ -85,9 +85,9 @@ int main (int argc, char ** argv)
     PokeIndex<LorentzIndex>(mom,mommu,mu);
 
     // fourth order exponential approx
-    auto Uprime_v = Uprime.View(CpuWrite);
+    autoView(Uprime_v, Uprime, CpuWrite);
-    auto U_v = U.View(CpuRead);
+    autoView( U_v , U, CpuRead);
-    auto mom_v = mom.View(CpuRead);
+    autoView( mom_v, mom, CpuRead);
     thread_foreach(i,mom_v,{ // exp(pmu dt) * Umu
       Uprime_v[i](mu) = U_v[i](mu) + mom_v[i](mu)*U_v[i](mu)*dt ;
     });

tests/forces/Test_gp_rect_force.cc

@@ -87,9 +87,9 @@ int main (int argc, char ** argv)
     PokeIndex<LorentzIndex>(mom,mommu,mu);
 
     // fourth order exponential approx
-    auto mom_v   = mom.View(CpuRead);
+    autoView( mom_v, mom, CpuRead);
-    auto Uprime_v= Uprime.View(CpuWrite);
+    autoView(Uprime_v, Uprime, CpuWrite);
-    auto U_v     = U.View(CpuRead);
+    autoView( U_v , U, CpuRead);
     thread_foreach(i,mom_v,{ // exp(pmu dt) * Umu
       Uprime_v[i](mu) = U_v[i](mu) + mom_v[i](mu)*U_v[i](mu)*dt ;
     });

tests/forces/Test_gpdwf_force.cc

@@ -105,9 +105,9 @@ int main (int argc, char ** argv)
     PokeIndex<LorentzIndex>(mom,mommu,mu);
 
     // fourth order exponential approx
-    auto U_v = U.View(CpuRead);
+    autoView( U_v , U, CpuRead);
-    auto mom_v = mom.View(CpuRead);
+    autoView( mom_v, mom, CpuRead);
-    auto Uprime_v = Uprime.View(CpuWrite);
+    autoView(Uprime_v, Uprime, CpuWrite);
     thread_foreach(i,mom_v,{
       Uprime_v[i](mu) = U_v[i](mu)
 	+ mom_v[i](mu)*U_v[i](mu)*dt 

tests/forces/Test_gpwilson_force.cc

@@ -99,9 +99,9 @@ int main (int argc, char ** argv)
     PokeIndex<LorentzIndex>(mom,mommu,mu);
 
     // fourth order exponential approx
-    auto mom_v    = mom.View(CpuRead);
+    autoView( mom_v, mom, CpuRead);
-    auto U_v      = U.View(CpuRead);
+    autoView( U_v , U, CpuRead);
-    auto Uprime_v = Uprime.View(CpuWrite);
+    autoView(Uprime_v, Uprime, CpuWrite);
 
     thread_foreach(i,mom_v,{
       Uprime_v[i](mu) =	  U_v[i](mu)

tests/forces/Test_mobius_force.cc

@@ -101,9 +101,9 @@ int main (int argc, char ** argv)
     PokeIndex<LorentzIndex>(mom,mommu,mu);
 
     // fourth order exponential approx
-    auto U_v      = U.View(CpuRead);
+    autoView( U_v , U, CpuRead);
-    auto mom_v    = mom.View(CpuRead);
+    autoView( mom_v, mom, CpuRead);
-    auto Uprime_v = Uprime.View(CpuWrite);
+    autoView(Uprime_v, Uprime, CpuWrite);
     thread_foreach(i,mom_v,{
       Uprime_v[i](mu) =  U_v[i](mu)
 	+ mom_v[i](mu)*U_v[i](mu)*dt 

tests/forces/Test_mobius_force_eofa.cc

@@ -112,9 +112,9 @@ int main (int argc, char** argv)
     PokeIndex<LorentzIndex>(mom, mommu, mu);
 
     // fourth order exponential approx
-    auto mom_v    = mom.View(CpuRead);
+    autoView( mom_v, mom, CpuRead);
-    auto U_v      = U.View(CpuRead);
+    autoView( U_v , U, CpuRead);
-    auto Uprime_v = Uprime.View(CpuWrite);
+    autoView(Uprime_v, Uprime, CpuWrite);
 
     thread_foreach(i,mom_v,{
       Uprime_v[i](mu) =	  U_v[i](mu)

tests/forces/Test_mobius_gpforce_eofa.cc

@@ -115,9 +115,9 @@ int main (int argc, char** argv)
     SU3::GaussianFundamentalLieAlgebraMatrix(RNG4, mommu); // Traceless antihermitian momentum; gaussian in lie alg
 
     PokeIndex<LorentzIndex>(mom, mommu, mu);
-    auto U_v      = U.View(CpuRead);
+    autoView( U_v , U, CpuRead);
-    auto mom_v    = mom.View(CpuRead);
+    autoView( mom_v, mom, CpuRead);
-    auto Uprime_v = Uprime.View(CpuWrite);
+    autoView(Uprime_v, Uprime, CpuWrite);
     // fourth order exponential approx
     thread_foreach( i, mom_v,{
       Uprime_v[i](mu) = U_v[i](mu) + mom_v[i](mu)*U_v[i](mu)*dt + mom_v[i](mu) *mom_v[i](mu) *U_v[i](mu)*(dt*dt/2.0)

tests/forces/Test_partfrac_force.cc

@@ -101,9 +101,9 @@ int main (int argc, char ** argv)
     PokeIndex<LorentzIndex>(mom,mommu,mu);
 
     // fourth order exponential approx
-    auto U_v = U.View(CpuRead);
+    autoView( U_v , U, CpuRead);
-    auto mom_v = mom.View(CpuRead);
+    autoView( mom_v, mom, CpuRead);
-    auto Uprime_v = Uprime.View(CpuWrite);
+    autoView(Uprime_v, Uprime, CpuWrite);
     thread_foreach(i,mom_v,{
       Uprime_v[i](mu) =  U_v[i](mu)
 	+ mom_v[i](mu)*U_v[i](mu)*dt 

tests/forces/Test_rect_force.cc

@@ -87,9 +87,9 @@ int main (int argc, char ** argv)
     PokeIndex<LorentzIndex>(mom,mommu,mu);
 
     // fourth order exponential approx
-    auto Uprime_v = Uprime.View(CpuWrite);
+    autoView(Uprime_v, Uprime, CpuWrite);
-    auto U_v      = U.View(CpuRead);
+    autoView( U_v , U, CpuRead);
-    auto mom_v    = mom.View(CpuRead);
+    autoView( mom_v, mom, CpuRead);
     thread_foreach(i,mom_v,{ // exp(pmu dt) * Umu
       Uprime_v[i](mu) = U_v[i](mu) + mom_v[i](mu)*U_v[i](mu)*dt ;
     });

tests/forces/Test_wilson_force.cc

@@ -105,9 +105,9 @@ int main (int argc, char ** argv)
     PokeIndex<LorentzIndex>(mom,mommu,mu);
 
     // fourth order exponential approx
-    auto U_v = U.View(CpuRead);
+    autoView( U_v , U, CpuRead);
-    auto mom_v = mom.View(CpuRead);
+    autoView( mom_v, mom, CpuRead);
-    auto Uprime_v = Uprime.View(CpuWrite);
+    autoView(Uprime_v, Uprime, CpuWrite);
     thread_foreach( i,mom_v,{
       Uprime_v[i](mu)  =	  U_v[i](mu);
       Uprime_v[i](mu) += mom_v[i](mu)*U_v[i](mu)*dt ;

tests/forces/Test_wilsonclover_force.cc

@@ -105,9 +105,9 @@ int main(int argc, char **argv)
     Hmom -= real(sum(trace(mommu * mommu)));
     PokeIndex<LorentzIndex>(mom, mommu, mu);
 
-    auto Uprime_v = Uprime.View(CpuWrite);
+    autoView(Uprime_v, Uprime, CpuWrite);
-    auto U_v      = U.View(CpuRead);
+    autoView( U_v , U, CpuRead);
-    auto mom_v    = mom.View(CpuRead);
+    autoView( mom_v, mom, CpuRead);
     thread_foreach(ss,mom_v,
     {
       Uprime_v[ss]._internal[mu] = ProjectOnGroup(Exponentiate(mom_v[ss]._internal[mu], dt, 12) * U_v[ss]._internal[mu]);

tests/forces/Test_zmobius_force.cc

@@ -114,9 +114,9 @@ int main (int argc, char ** argv)
     PokeIndex<LorentzIndex>(mom,mommu,mu);
 
     // fourth order exponential approx
-    auto mom_v    = mom.View(CpuRead);
+    autoView( mom_v, mom, CpuRead);
-    auto U_v      = U.View(CpuRead);
+    autoView( U_v , U, CpuRead);
-    auto Uprime_v = Uprime.View(CpuWrite);
+    autoView(Uprime_v, Uprime, CpuWrite);
 
     thread_foreach(i,mom_v,{
       Uprime_v[i](mu) =	  U_v[i](mu)

tests/solver/Test_dwf_hdcr.cc

@@ -300,8 +300,8 @@ int main (int argc, char ** argv)
     int nb=nbasisc/2;
     CoarseAggregates.CreateSubspaceChebyshev(CRNG,PosdefLdop,nb,12.0,0.02,500,100,100,0.0);
     for(int n=0;n<nb;n++){
-      auto subspace    = CoarseAggregates.subspace[n].View(CpuRead);
+      autoView( subspace,   CoarseAggregates.subspace[n]   ,CpuRead);
-      auto subspace_g5 = CoarseAggregates.subspace[n+nb].View(CpuWrite);
+      autoView( subspace_g5,CoarseAggregates.subspace[n+nb],CpuWrite);
       for(int nn=0;nn<nb;nn++){
 	for(int site=0;site<Coarse5d->oSites();site++){
 	  subspace_g5[site](nn)   = subspace[site](nn);