diff --git a/benchmarks/Benchmark_comms.cc b/benchmarks/Benchmark_comms.cc
index 00526893..6696c8eb 100644
--- a/benchmarks/Benchmark_comms.cc
+++ b/benchmarks/Benchmark_comms.cc
@@ -166,18 +166,18 @@ int main (int argc, char ** argv)
   }  
 
 
+
   std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
-  std::cout<<GridLogMessage << "= Benchmarking concurrent STENCIL halo exchange in "<<nmu<<" dimensions"<<std::endl;
+  std::cout<<GridLogMessage << "= Benchmarking sequential STENCIL halo exchange in "<<nmu<<" dimensions"<<std::endl;
   std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
   header();
-
   for(int lat=8;lat<=maxlat;lat+=4){
     for(int Ls=8;Ls<=8;Ls*=2){
 
       Coordinate latt_size  ({lat*mpi_layout[0],
-	                      lat*mpi_layout[1],
-      			      lat*mpi_layout[2],
-      			      lat*mpi_layout[3]});
+    	                      lat*mpi_layout[1],
+                              lat*mpi_layout[2],
+	                      lat*mpi_layout[3]});
 
       GridCartesian     Grid(latt_size,simd_layout,mpi_layout);
       RealD Nrank = Grid._Nprocessors;
@@ -193,101 +193,6 @@ int main (int argc, char ** argv)
 	rbuf[d] = (HalfSpinColourVectorD *)Grid.ShmBufferMalloc(bytes);
       }
 
-      int ncomm;
-
-      double dbytes;
-      for(int i=0;i<Nloop;i++){
-	double start=usecond();
-
-	dbytes=0;
-	ncomm=0;
-
-	std::vector<CommsRequest_t> requests;
-
-	for(int mu=0;mu<4;mu++){
-	
-
-	  if (mpi_layout[mu]>1 ) {
-	  
-	    ncomm++;
-	    int comm_proc=1;
-	    int xmit_to_rank;
-	    int recv_from_rank;
-	    Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank);
-	    dbytes+=
-	      Grid.StencilSendToRecvFromBegin(requests,
-					      (void *)&xbuf[mu][0],
-					      xmit_to_rank,1,
-					      (void *)&rbuf[mu][0],
-					      recv_from_rank,1,
-					      bytes,bytes,mu);
-	
-	    comm_proc = mpi_layout[mu]-1;
-	  
-	    Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank);
-	    dbytes+=
-	      Grid.StencilSendToRecvFromBegin(requests,
-					      (void *)&xbuf[mu+4][0],
-					      xmit_to_rank,1,
-					      (void *)&rbuf[mu+4][0],
-					      recv_from_rank,1,
-					      bytes,bytes,mu+4);
-	  
-	  }
-	}
-	Grid.StencilSendToRecvFromComplete(requests,0);
-	Grid.Barrier();
-	double stop=usecond();
-	t_time[i] = stop-start; // microseconds
-	
-      }
-
-      timestat.statistics(t_time);
-
-      dbytes=dbytes*ppn;
-      double xbytes    = dbytes*0.5;
-      //      double rbytes    = dbytes*0.5;
-      double bidibytes = dbytes;
-
-      std::cout<<GridLogMessage << std::setw(4) << lat<<"\t"<<Ls<<"\t"
-               <<std::setw(11) << bytes<< std::fixed << std::setprecision(1) << std::setw(7)
-               <<std::right<< xbytes/timestat.mean<<"  "<< xbytes*timestat.err/(timestat.mean*timestat.mean)<< " "
-               <<xbytes/timestat.max <<" "<< xbytes/timestat.min  
-               << "\t\t"<<std::setw(7)<< bidibytes/timestat.mean<< "  " << bidibytes*timestat.err/(timestat.mean*timestat.mean) << " "
-               << bidibytes/timestat.max << " " << bidibytes/timestat.min << std::endl;
-
-
-    }
-  }    
-
-
-  std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
-  std::cout<<GridLogMessage << "= Benchmarking sequential STENCIL halo exchange in "<<nmu<<" dimensions"<<std::endl;
-  std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
-  header();
-
-  for(int lat=8;lat<=maxlat;lat+=4){
-    for(int Ls=8;Ls<=8;Ls*=2){
-
-      Coordinate latt_size  ({lat*mpi_layout[0],
-      			      lat*mpi_layout[1],
-      			      lat*mpi_layout[2],
-      			      lat*mpi_layout[3]});
-
-      GridCartesian     Grid(latt_size,simd_layout,mpi_layout);
-      RealD Nrank = Grid._Nprocessors;
-      RealD Nnode = Grid.NodeCount();
-      RealD ppn = Nrank/Nnode;
-
-      std::vector<HalfSpinColourVectorD *> xbuf(8);
-      std::vector<HalfSpinColourVectorD *> rbuf(8);
-      Grid.ShmBufferFreeAll();
-      uint64_t bytes=lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD);
-      for(int d=0;d<8;d++){
-	xbuf[d] = (HalfSpinColourVectorD *)Grid.ShmBufferMalloc(bytes);
-	rbuf[d] = (HalfSpinColourVectorD *)Grid.ShmBufferMalloc(bytes);
-      }
-
       int ncomm;
       double dbytes;
       for(int i=0;i<Nloop;i++){
@@ -296,45 +201,34 @@ int main (int argc, char ** argv)
 	std::vector<CommsRequest_t> requests;
 	dbytes=0;
 	ncomm=0;
-	for(int mu=0;mu<4;mu++){
-	
+
+	for(int dir=0;dir<8;dir++) {
+
+	  double tbytes;
+	  int mu =dir % 4;
+
 	  if (mpi_layout[mu]>1 ) {
 	  
 	    ncomm++;
-	    int comm_proc=1;
 	    int xmit_to_rank;
 	    int recv_from_rank;
-	    
-	    Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank);
-	    dbytes+=
-	      Grid.StencilSendToRecvFromBegin(requests,
-					      (void *)&xbuf[mu][0],
-					      xmit_to_rank,1,
-					      (void *)&rbuf[mu][0],
-					      recv_from_rank,1,
-					      bytes,bytes,mu);
-	    Grid.StencilSendToRecvFromComplete(requests,mu);
-	    requests.resize(0);
+	    if ( dir == mu ) { 
+	      int comm_proc=1;
+	      Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank);
+	    } else { 
+	      int comm_proc = mpi_layout[mu]-1;
+	      Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank);
+	    }
+            int tid = omp_get_thread_num();
+	    tbytes= Grid.StencilSendToRecvFrom((void *)&xbuf[dir][0], xmit_to_rank,1,
+					       (void *)&rbuf[dir][0], recv_from_rank,1, bytes,tid);
 
-	    comm_proc = mpi_layout[mu]-1;
-	  
-	    Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank);
-	    dbytes+=
-	      Grid.StencilSendToRecvFromBegin(requests,
-					      (void *)&xbuf[mu+4][0],
-					      xmit_to_rank,1,
-					      (void *)&rbuf[mu+4][0],
-					      recv_from_rank,1,
-					      bytes,bytes,mu+4);
-	    Grid.StencilSendToRecvFromComplete(requests,mu+4);
-	    requests.resize(0);
-	  
+	    dbytes+=tbytes;
 	  }
-	}
+        }
 	Grid.Barrier();
 	double stop=usecond();
 	t_time[i] = stop-start; // microseconds
-	
       }
 
       timestat.statistics(t_time);
diff --git a/benchmarks/Benchmark_dwf.cc b/benchmarks/Benchmark_dwf.cc
index 55135322..29772141 100644
--- a/benchmarks/Benchmark_dwf.cc
+++ b/benchmarks/Benchmark_dwf.cc
@@ -32,18 +32,18 @@
 using namespace std;
 using namespace Grid;
 
-template<class d>
-struct scal {
-  d internal;
+////////////////////////
+/// Move to domains ////
+////////////////////////
+
+Gamma::Algebra Gmu [] = {
+			 Gamma::Algebra::GammaX,
+			 Gamma::Algebra::GammaY,
+			 Gamma::Algebra::GammaZ,
+			 Gamma::Algebra::GammaT
 };
 
-  Gamma::Algebra Gmu [] = {
-    Gamma::Algebra::GammaX,
-    Gamma::Algebra::GammaY,
-    Gamma::Algebra::GammaZ,
-    Gamma::Algebra::GammaT
-  };
-
+void Benchmark(int Ls, Coordinate Dirichlet,bool Sloppy);
 
 int main (int argc, char ** argv)
 {
@@ -52,39 +52,108 @@ int main (int argc, char ** argv)
 
   int threads = GridThread::GetThreads();
 
-  Coordinate latt4 = GridDefaultLatt();
-  int Ls=8;
-  for(int i=0;i<argc;i++)
+  int Ls=16;
+  for(int i=0;i<argc;i++) {
     if(std::string(argv[i]) == "-Ls"){
       std::stringstream ss(argv[i+1]); ss >> Ls;
     }
+  }
 
+  //////////////////
+  // With comms
+  //////////////////
+  Coordinate Dirichlet(Nd+1,0);
+
+  std::cout << "\n\n\n\n\n\n" <<std::endl;
+  std::cout << GridLogMessage<< "++++++++++++++++++++++++++++++++++++++++++++++++" <<std::endl;
+  std::cout << GridLogMessage<< " Testing with full communication " <<std::endl;
+  std::cout << GridLogMessage<< "++++++++++++++++++++++++++++++++++++++++++++++++" <<std::endl;
+  
+  Benchmark(Ls,Dirichlet,false);
+
+  std::cout << "\n\n\n\n\n\n" <<std::endl;
+  std::cout << GridLogMessage<< "++++++++++++++++++++++++++++++++++++++++++++++++" <<std::endl;
+  std::cout << GridLogMessage<< " Testing with sloppy communication " <<std::endl;
+  std::cout << GridLogMessage<< "++++++++++++++++++++++++++++++++++++++++++++++++" <<std::endl;
+  
+  Benchmark(Ls,Dirichlet,true);
+
+  //////////////////
+  // Domain decomposed
+  //////////////////
+  /*
+  Coordinate latt4  = GridDefaultLatt();
+  Coordinate mpi    = GridDefaultMpi();
+  Coordinate CommDim(Nd);
+  Coordinate shm;
+  GlobalSharedMemory::GetShmDims(mpi,shm);
+
+
+  std::cout << "\n\n\n\n\n\n" <<std::endl;
+  std::cout << GridLogMessage<< "++++++++++++++++++++++++++++++++++++++++++++++++" <<std::endl;
+  //  std::cout << GridLogMessage<< " Testing without internode communication " <<std::endl;
+  std::cout << GridLogMessage<< "++++++++++++++++++++++++++++++++++++++++++++++++" <<std::endl;
+
+  for(int d=0;d<Nd;d++) CommDim[d]= (mpi[d]/shm[d])>1 ? 1 : 0;
+  Dirichlet[0] = 0;
+  Dirichlet[1] = CommDim[0]*latt4[0]/mpi[0] * shm[0];
+  Dirichlet[2] = CommDim[1]*latt4[1]/mpi[1] * shm[1];
+  Dirichlet[3] = CommDim[2]*latt4[2]/mpi[2] * shm[2];
+  Dirichlet[4] = CommDim[3]*latt4[3]/mpi[3] * shm[3];
+
+  Benchmark(Ls,Dirichlet,false);
+
+  std::cout << "\n\n\n\n\n\n" <<std::endl;
+
+  std::cout << GridLogMessage<< "++++++++++++++++++++++++++++++++++++++++++++++++" <<std::endl;
+  std::cout << GridLogMessage<< " Testing with sloppy communication " <<std::endl;
+  std::cout << GridLogMessage<< "++++++++++++++++++++++++++++++++++++++++++++++++" <<std::endl;
+
+  for(int d=0;d<Nd;d++) CommDim[d]= mpi[d]>1 ? 1 : 0;
+  
+  Benchmark(Ls,Dirichlet,true);
+  */
+  
+  Grid_finalize();
+  exit(0);
+}
+void Benchmark(int Ls, Coordinate Dirichlet,bool sloppy)
+{
+  Coordinate latt4 = GridDefaultLatt();
   GridLogLayout();
 
   long unsigned int single_site_flops = 8*Nc*(7+16*Nc);
 
-
-  GridCartesian         * UGrid   = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());
+  std::vector<int> seeds4({1,2,3,4});
+  std::vector<int> seeds5({5,6,7,8});
+#undef SINGLE
+#ifdef SINGLE
+  typedef vComplexF          Simd;
+  typedef LatticeFermionF    FermionField;
+  typedef LatticeGaugeFieldF GaugeField;
+  typedef LatticeColourMatrixF ColourMatrixField;
+  typedef DomainWallFermionF FermionAction;
+#else
+  typedef vComplexD          Simd;
+  typedef LatticeFermionD    FermionField;
+  typedef LatticeGaugeFieldD GaugeField;
+  typedef LatticeColourMatrixD ColourMatrixField;
+  typedef DomainWallFermionD FermionAction;
+#endif
+  
+  GridCartesian         * UGrid   = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,Simd::Nsimd()),GridDefaultMpi());
   GridRedBlackCartesian * UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
   GridCartesian         * FGrid   = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
   GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid);
 
-  std::cout << GridLogMessage << "Making s innermost grids"<<std::endl;
-  GridCartesian         * sUGrid   = SpaceTimeGrid::makeFourDimDWFGrid(GridDefaultLatt(),GridDefaultMpi());
-  GridRedBlackCartesian * sUrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(sUGrid);
-  GridCartesian         * sFGrid   = SpaceTimeGrid::makeFiveDimDWFGrid(Ls,UGrid);
-  GridRedBlackCartesian * sFrbGrid = SpaceTimeGrid::makeFiveDimDWFRedBlackGrid(Ls,UGrid);
-
-  std::vector<int> seeds4({1,2,3,4});
-  std::vector<int> seeds5({5,6,7,8});
-
   std::cout << GridLogMessage << "Initialising 4d RNG" << std::endl;
   GridParallelRNG          RNG4(UGrid);  RNG4.SeedUniqueString(std::string("The 4D RNG"));
+
   std::cout << GridLogMessage << "Initialising 5d RNG" << std::endl;
   GridParallelRNG          RNG5(FGrid);  RNG5.SeedUniqueString(std::string("The 5D RNG"));
-  std::cout << GridLogMessage << "Initialised RNGs" << std::endl;
 
-  LatticeFermion src   (FGrid); random(RNG5,src);
+ 
+  FermionField src   (FGrid); random(RNG5,src);
 #if 0
   src = Zero();
   {
@@ -100,46 +169,39 @@ int main (int argc, char ** argv)
   src = src*N2;
 #endif
 
-
-  LatticeFermion result(FGrid); result=Zero();
-  LatticeFermion    ref(FGrid);    ref=Zero();
-  LatticeFermion    tmp(FGrid);
-  LatticeFermion    err(FGrid);
+  FermionField result(FGrid); result=Zero();
+  FermionField    ref(FGrid);    ref=Zero();
+  FermionField    tmp(FGrid);
+  FermionField    err(FGrid);
 
   std::cout << GridLogMessage << "Drawing gauge field" << std::endl;
-  LatticeGaugeField Umu(UGrid);
+  GaugeField Umu(UGrid);
+  GaugeField UmuCopy(UGrid);
   SU<Nc>::HotConfiguration(RNG4,Umu);
+  //  SU<Nc>::ColdConfiguration(Umu);
+  UmuCopy=Umu;
   std::cout << GridLogMessage << "Random gauge initialised " << std::endl;
-#if 0
-  Umu=1.0;
-  for(int mu=0;mu<Nd;mu++){
-    LatticeColourMatrix ttmp(UGrid);
-    ttmp = PeekIndex<LorentzIndex>(Umu,mu);
-    //    if (mu !=2 ) ttmp = 0;
-    //    ttmp = ttmp* pow(10.0,mu);
-    PokeIndex<LorentzIndex>(Umu,ttmp,mu);
-  }
-  std::cout << GridLogMessage << "Forced to diagonal " << std::endl;
-#endif
 
+  ////////////////////////////////////
+  // Apply BCs
+  ////////////////////////////////////
+  Coordinate Block(4);
+  for(int d=0;d<4;d++)  Block[d]= Dirichlet[d+1];
+
+  std::cout << GridLogMessage << "Applying BCs for Dirichlet Block5 " << Dirichlet << std::endl;
+  std::cout << GridLogMessage << "Applying BCs for Dirichlet Block4 " << Block << std::endl;
+
+  DirichletFilter<GaugeField> Filter(Block);
+  Filter.applyFilter(Umu);
+  
   ////////////////////////////////////
   // Naive wilson implementation
   ////////////////////////////////////
-  // replicate across fifth dimension
-  LatticeGaugeField Umu5d(FGrid);
-  std::vector<LatticeColourMatrix> U(4,FGrid);
-  {
-    autoView( Umu5d_v, Umu5d, CpuWrite);
-    autoView( Umu_v  , Umu  , CpuRead);
-    for(int ss=0;ss<Umu.Grid()->oSites();ss++){
-      for(int s=0;s<Ls;s++){
-	Umu5d_v[Ls*ss+s] = Umu_v[ss];
-      }
-    }
-  }
+  std::vector<ColourMatrixField> U(4,UGrid);
   for(int mu=0;mu<Nd;mu++){
-    U[mu] = PeekIndex<LorentzIndex>(Umu5d,mu);
+    U[mu] = PeekIndex<LorentzIndex>(Umu,mu);
   }
+
   std::cout << GridLogMessage << "Setting up Cshift based reference " << std::endl;
 
   if (1)
@@ -147,10 +209,28 @@ int main (int argc, char ** argv)
     ref = Zero();
     for(int mu=0;mu<Nd;mu++){
 
-      tmp = U[mu]*Cshift(src,mu+1,1);
+      tmp = Cshift(src,mu+1,1);
+      {
+	autoView( tmp_v  , tmp  , CpuWrite);
+	autoView( U_v  , U[mu]  , CpuRead);
+	for(int ss=0;ss<U[mu].Grid()->oSites();ss++){
+	  for(int s=0;s<Ls;s++){
+	    tmp_v[Ls*ss+s] = U_v[ss]*tmp_v[Ls*ss+s];
+	  }
+	}
+      }
       ref=ref + tmp - Gamma(Gmu[mu])*tmp;
 
-      tmp =adj(U[mu])*src;
+      {
+	autoView( tmp_v  , tmp  , CpuWrite);
+	autoView( U_v  , U[mu]  , CpuRead);
+	autoView( src_v, src    , CpuRead);
+	for(int ss=0;ss<U[mu].Grid()->oSites();ss++){
+	  for(int s=0;s<Ls;s++){
+	    tmp_v[Ls*ss+s] = adj(U_v[ss])*src_v[Ls*ss+s];
+	  }
+	}
+      }
       tmp =Cshift(tmp,mu+1,-1);
       ref=ref + tmp + Gamma(Gmu[mu])*tmp;
     }
@@ -167,11 +247,9 @@ int main (int argc, char ** argv)
   std::cout << GridLogMessage<< "* Kernel options --dslash-generic, --dslash-unroll, --dslash-asm" <<std::endl;
   std::cout << GridLogMessage<< "*****************************************************************" <<std::endl;
   std::cout << GridLogMessage<< "*****************************************************************" <<std::endl;
-  std::cout << GridLogMessage<< "* Benchmarking DomainWallFermionD::Dhop                  "<<std::endl;
-  std::cout << GridLogMessage<< "* Vectorising space-time by "<<vComplex::Nsimd()<<std::endl;
-  std::cout << GridLogMessage<< "* VComplex size is "<<sizeof(vComplex)<< " B"<<std::endl;
-  if ( sizeof(Real)==4 )   std::cout << GridLogMessage<< "* SINGLE precision "<<std::endl;
-  if ( sizeof(Real)==8 )   std::cout << GridLogMessage<< "* DOUBLE precision "<<std::endl;
+  std::cout << GridLogMessage<< "* Benchmarking DomainWallFermionR::Dhop                  "<<std::endl;
+  std::cout << GridLogMessage<< "* Vectorising space-time by "<<Simd::Nsimd()<<std::endl;
+  std::cout << GridLogMessage<< "* VComplex size is "<<sizeof(Simd)<< " B"<<std::endl;
 #ifdef GRID_OMP
   if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsAndCompute ) std::cout << GridLogMessage<< "* Using Overlapped Comms/Compute" <<std::endl;
   if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsThenCompute) std::cout << GridLogMessage<< "* Using sequential comms compute" <<std::endl;
@@ -181,9 +259,15 @@ int main (int argc, char ** argv)
   if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptInlineAsm ) std::cout << GridLogMessage<< "* Using Asm Nc=3   WilsonKernels" <<std::endl;
   std::cout << GridLogMessage<< "*****************************************************************" <<std::endl;
 
-  DomainWallFermionD Dw(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
-  int ncall =1000;
-
+  FermionAction::ImplParams p;
+  p.dirichlet=Dirichlet;
+  FermionAction Dw(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,p);
+  Dw.SloppyComms(sloppy);
+  Dw.ImportGauge(Umu);
+  
+  int ncall =300;
+  RealD n2e;
+  
   if (1) {
     FGrid->Barrier();
     Dw.Dhop(src,result,0);
@@ -198,8 +282,8 @@ int main (int argc, char ** argv)
     double volume=Ls;  for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
     double flops=single_site_flops*volume*ncall;
 
-    auto nsimd = vComplex::Nsimd();
-    auto simdwidth = sizeof(vComplex);
+    auto nsimd = Simd::Nsimd();
+    auto simdwidth = sizeof(Simd);
 
     // RF: Nd Wilson * Ls, Nd gauge * Ls, Nc colors
     double data_rf = volume * ((2*Nd+1)*Nd*Nc + 2*Nd*Nc*Nc) * simdwidth / nsimd * ncall / (1024.*1024.*1024.);
@@ -208,28 +292,27 @@ int main (int argc, char ** argv)
     double data_mem = (volume * (2*Nd+1)*Nd*Nc + (volume/Ls) *2*Nd*Nc*Nc) * simdwidth / nsimd * ncall / (1024.*1024.*1024.);
 
     std::cout<<GridLogMessage << "Called Dw "<<ncall<<" times in "<<t1-t0<<" us"<<std::endl;
-    //    std::cout<<GridLogMessage << "norm result "<< norm2(result)<<std::endl;
-    //    std::cout<<GridLogMessage << "norm ref    "<< norm2(ref)<<std::endl;
     std::cout<<GridLogMessage << "mflop/s =   "<< flops/(t1-t0)<<std::endl;
     std::cout<<GridLogMessage << "mflop/s per rank =  "<< flops/(t1-t0)/NP<<std::endl;
     std::cout<<GridLogMessage << "mflop/s per node =  "<< flops/(t1-t0)/NN<<std::endl;
-    std::cout<<GridLogMessage << "RF  GiB/s (base 2) =   "<< 1000000. * data_rf/((t1-t0))<<std::endl;
-    std::cout<<GridLogMessage << "mem GiB/s (base 2) =   "<< 1000000. * data_mem/((t1-t0))<<std::endl;
     err = ref-result;
-    std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<<std::endl;
-    //exit(0);
+    n2e = norm2(err);
+    std::cout<<GridLogMessage << "norm diff   "<< n2e<< "  Line "<<__LINE__ <<std::endl;
 
-    if(( norm2(err)>1.0e-4) ) {
-      /*
-      std::cout << "RESULT\n " << result<<std::endl;
-      std::cout << "REF   \n " << ref   <<std::endl;
-      std::cout << "ERR   \n " << err   <<std::endl;
-      */
+    if(( n2e>1.0e-4) ) {
       std::cout<<GridLogMessage << "WRONG RESULT" << std::endl;
       FGrid->Barrier();
+      std::cout<<GridLogMessage << "RESULT" << std::endl;
+      //      std::cout << result<<std::endl;
+      std::cout << norm2(result)<<std::endl;
+      std::cout<<GridLogMessage << "REF" << std::endl;
+      std::cout << norm2(ref)<<std::endl;
+      std::cout<<GridLogMessage << "ERR" << std::endl;
+      std::cout << norm2(err)<<std::endl;
+      FGrid->Barrier();
       exit(-1);
     }
-    assert (norm2(err)< 1.0e-4 );
+    assert (n2e< 1.0e-4 );
   }
 
   if (1)
@@ -238,16 +321,30 @@ int main (int argc, char ** argv)
     for(int mu=0;mu<Nd;mu++){
 
       //    ref =  src - Gamma(Gamma::Algebra::GammaX)* src ; // 1+gamma_x
-      tmp = U[mu]*Cshift(src,mu+1,1);
+      tmp = Cshift(src,mu+1,1);
       {
 	autoView( ref_v, ref, CpuWrite);
 	autoView( tmp_v, tmp, CpuRead);
-	for(int i=0;i<ref_v.size();i++){
-	  ref_v[i]+= tmp_v[i] + Gamma(Gmu[mu])*tmp_v[i]; ;
+	autoView( U_v  , U[mu]  , CpuRead);
+	for(int ss=0;ss<U[mu].Grid()->oSites();ss++){
+	  for(int s=0;s<Ls;s++){
+	    int i=s+Ls*ss;
+	    ref_v[i]+= U_v[ss]*(tmp_v[i] + Gamma(Gmu[mu])*tmp_v[i]); ;
+	  }
 	}
       }
-
-      tmp =adj(U[mu])*src;
+      
+      {
+	autoView( tmp_v  , tmp  , CpuWrite);
+	autoView( U_v  , U[mu]  , CpuRead);
+	autoView( src_v, src    , CpuRead);
+	for(int ss=0;ss<U[mu].Grid()->oSites();ss++){
+	  for(int s=0;s<Ls;s++){
+	    tmp_v[Ls*ss+s] = adj(U_v[ss])*src_v[Ls*ss+s];
+	  }
+	}
+      }
+      //      tmp =adj(U[mu])*src;
       tmp =Cshift(tmp,mu+1,-1);
       {
 	autoView( ref_v, ref, CpuWrite);
@@ -259,27 +356,27 @@ int main (int argc, char ** argv)
     }
     ref = -0.5*ref;
   }
-  //  dump=1;
-  Dw.Dhop(src,result,1);
+
+  Dw.Dhop(src,result,DaggerYes);
+
+  std::cout << GridLogMessage << "----------------------------------------------------------------" << std::endl;
   std::cout << GridLogMessage << "Compare to naive wilson implementation Dag to verify correctness" << std::endl;
+  std::cout << GridLogMessage << "----------------------------------------------------------------" << std::endl;
+
   std::cout<<GridLogMessage << "Called DwDag"<<std::endl;
   std::cout<<GridLogMessage << "norm dag result "<< norm2(result)<<std::endl;
   std::cout<<GridLogMessage << "norm dag ref    "<< norm2(ref)<<std::endl;
   err = ref-result;
-  std::cout<<GridLogMessage << "norm dag diff   "<< norm2(err)<<std::endl;
-  if((norm2(err)>1.0e-4)){
-/*
-	std::cout<< "DAG RESULT\n "  <<ref     << std::endl;
-	std::cout<< "DAG sRESULT\n " <<result  << std::endl;
-	std::cout<< "DAG ERR   \n "  << err    <<std::endl;
-*/
-  }
-  LatticeFermion src_e (FrbGrid);
-  LatticeFermion src_o (FrbGrid);
-  LatticeFermion r_e   (FrbGrid);
-  LatticeFermion r_o   (FrbGrid);
-  LatticeFermion r_eo  (FGrid);
+  n2e= norm2(err);
+  std::cout<<GridLogMessage << "norm dag diff   "<< n2e<< "  Line "<<__LINE__ <<std::endl;
 
+  assert((n2e)<1.0e-4);
+  
+  FermionField src_e (FrbGrid);
+  FermionField src_o (FrbGrid);
+  FermionField r_e   (FrbGrid);
+  FermionField r_o   (FrbGrid);
+  FermionField r_eo  (FGrid);
 
   std::cout<<GridLogMessage << "Calling Deo and Doe and //assert Deo+Doe == Dunprec"<<std::endl;
   pickCheckerboard(Even,src_e,src);
@@ -291,10 +388,8 @@ int main (int argc, char ** argv)
 
   // S-direction is INNERMOST and takes no part in the parity.
   std::cout << GridLogMessage<< "*********************************************************" <<std::endl;
-  std::cout << GridLogMessage<< "* Benchmarking DomainWallFermionD::DhopEO                "<<std::endl;
-  std::cout << GridLogMessage<< "* Vectorising space-time by "<<vComplex::Nsimd()<<std::endl;
-  if ( sizeof(Real)==4 )   std::cout << GridLogMessage<< "* SINGLE precision "<<std::endl;
-  if ( sizeof(Real)==8 )   std::cout << GridLogMessage<< "* DOUBLE precision "<<std::endl;
+  std::cout << GridLogMessage<< "* Benchmarking DomainWallFermion::DhopEO                "<<std::endl;
+  std::cout << GridLogMessage<< "* Vectorising space-time by "<<Simd::Nsimd()<<std::endl;
 #ifdef GRID_OMP
   if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsAndCompute ) std::cout << GridLogMessage<< "* Using Overlapped Comms/Compute" <<std::endl;
   if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsThenCompute) std::cout << GridLogMessage<< "* Using sequential comms compute" <<std::endl;
@@ -308,13 +403,7 @@ int main (int argc, char ** argv)
     Dw.DhopEO(src_o,r_e,DaggerNo);
     double t0=usecond();
     for(int i=0;i<ncall;i++){
-#ifdef CUDA_PROFILE
-      if(i==10) cudaProfilerStart();
-#endif
       Dw.DhopEO(src_o,r_e,DaggerNo);
-#ifdef CUDA_PROFILE
-      if(i==20) cudaProfilerStop();
-#endif
     }
     double t1=usecond();
     FGrid->Barrier();
@@ -338,14 +427,9 @@ int main (int argc, char ** argv)
   setCheckerboard(r_eo,r_e);
 
   err = r_eo-result;
-  std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<<std::endl;
-  if((norm2(err)>1.0e-4)){
-    /*
-	std::cout<< "Deo RESULT\n " <<r_eo << std::endl;
-	std::cout<< "Deo REF\n " <<result  << std::endl;
-	std::cout<< "Deo ERR   \n " << err <<std::endl;
-    */
-  }
+  n2e= norm2(err);
+  std::cout<<GridLogMessage << "norm diff   "<< n2e<<std::endl;
+  assert(n2e<1.0e-4);
 
   pickCheckerboard(Even,src_e,err);
   pickCheckerboard(Odd,src_o,err);
@@ -354,6 +438,4 @@ int main (int argc, char ** argv)
 
   assert(norm2(src_e)<1.0e-4);
   assert(norm2(src_o)<1.0e-4);
-  Grid_finalize();
-  exit(0);
 }