diff --git a/benchmarks/Benchmark_ITT.cc b/benchmarks/Benchmark_ITT.cc
index 7ad4a147..1bb77aff 100644
--- a/benchmarks/Benchmark_ITT.cc
+++ b/benchmarks/Benchmark_ITT.cc
@@ -30,7 +30,6 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 
 using namespace Grid;
 
-
 std::vector<int> L_list;
 std::vector<int> Ls_list;
 std::vector<double> mflop_list;
@@ -76,7 +75,6 @@ struct controls {
   int Opt;
   int CommsOverlap;
   Grid::CartesianCommunicator::CommunicatorPolicy_t CommsAsynch;
-  //  int HugePages;
 };
 
 class Benchmark {
@@ -119,14 +117,15 @@ public:
     std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
     comms_header();
 
-    for(int lat=4;lat<=maxlat;lat+=4){
-      for(int Ls=8;Ls<=8;Ls*=2){
+    for(int lat=16;lat<=maxlat;lat+=8){
+      //      for(int Ls=8;Ls<=8;Ls*=2){
+      { int Ls=12;
 
 	Coordinate latt_size  ({lat*mpi_layout[0],
 	      lat*mpi_layout[1],
 	      lat*mpi_layout[2],
 	      lat*mpi_layout[3]});
-
+	std::cout << GridLogMessage<< latt_size <<std::endl;
 	GridCartesian     Grid(latt_size,simd_layout,mpi_layout);
 	RealD Nrank = Grid._Nprocessors;
 	RealD Nnode = Grid.NodeCount();
@@ -184,9 +183,6 @@ public:
 	}
 
 	timestat.statistics(t_time);
-	//	for(int i=0;i<t_time.size();i++){
-	//	  std::cout << i<<" "<<t_time[i]<<std::endl;
-	//	}
 
 	dbytes=dbytes*ppn;
 	double xbytes    = dbytes*0.5;
@@ -199,8 +195,6 @@ public:
 		 <<xbytes/timestat.max <<" "<< xbytes/timestat.min  
 		 << "\t\t"<<std::setw(7)<< bidibytes/timestat.mean<< "  " << bidibytes*timestat.err/(timestat.mean*timestat.mean) << " "
 		 << bidibytes/timestat.max << " " << bidibytes/timestat.min << std::endl;
-
- 
 	
 	    }
     }    
@@ -227,14 +221,15 @@ public:
     uint64_t NN;
 
 
-  uint64_t lmax=48;
+  uint64_t lmax=32;
 #define NLOOP (100*lmax*lmax*lmax*lmax/lat/lat/lat/lat)
 
     GridSerialRNG          sRNG;      sRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9}));
-    for(int lat=8;lat<=lmax;lat+=4){
+    for(int lat=8;lat<=lmax;lat+=8){
 
       Coordinate latt_size  ({lat*mpi_layout[0],lat*mpi_layout[1],lat*mpi_layout[2],lat*mpi_layout[3]});
       int64_t vol= latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
+
       GridCartesian     Grid(latt_size,simd_layout,mpi_layout);
 
       //      NP= Grid.RankCount();
@@ -270,191 +265,8 @@ public:
     }
   };
 
-#if 0
-  static double DWF5(int Ls,int L)
-  {
-    //    RealD mass=0.1;
-    RealD M5  =1.8;
 
-    double mflops;
-    double mflops_best = 0;
-    double mflops_worst= 0;
-    std::vector<double> mflops_all;
-
-    ///////////////////////////////////////////////////////
-    // Set/Get the layout & grid size
-    ///////////////////////////////////////////////////////
-    int threads = GridThread::GetThreads();
-    Coordinate mpi = GridDefaultMpi(); assert(mpi.size()==4);
-    Coordinate local({L,L,L,L});
-
-    GridCartesian         * TmpGrid   = SpaceTimeGrid::makeFourDimGrid(Coordinate({64,64,64,64}), 
-								       GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());
-    uint64_t NP = TmpGrid->RankCount();
-    uint64_t NN = TmpGrid->NodeCount();
-    NN_global=NN;
-    uint64_t SHM=NP/NN;
-
-    Coordinate internal;
-    if      ( SHM == 1 )   internal = Coordinate({1,1,1,1});
-    else if ( SHM == 2 )   internal = Coordinate({2,1,1,1});
-    else if ( SHM == 4 )   internal = Coordinate({2,2,1,1});
-    else if ( SHM == 8 )   internal = Coordinate({2,2,2,1});
-    else assert(0);
-
-    Coordinate nodes({mpi[0]/internal[0],mpi[1]/internal[1],mpi[2]/internal[2],mpi[3]/internal[3]});
-    Coordinate latt4({local[0]*nodes[0],local[1]*nodes[1],local[2]*nodes[2],local[3]*nodes[3]});
-
-    ///////// Welcome message ////////////
-    std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
-    std::cout<<GridLogMessage << "Benchmark DWF Ls vec on "<<L<<"^4 local volume "<<std::endl;
-    std::cout<<GridLogMessage << "* Global volume  : "<<GridCmdVectorIntToString(latt4)<<std::endl;
-    std::cout<<GridLogMessage << "* Ls             : "<<Ls<<std::endl;
-    std::cout<<GridLogMessage << "* MPI ranks      : "<<GridCmdVectorIntToString(mpi)<<std::endl;
-    std::cout<<GridLogMessage << "* Intranode      : "<<GridCmdVectorIntToString(internal)<<std::endl;
-    std::cout<<GridLogMessage << "* nodes          : "<<GridCmdVectorIntToString(nodes)<<std::endl;
-    std::cout<<GridLogMessage << "* Using "<<threads<<" threads"<<std::endl;
-    std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
-
-    ///////// Lattice Init ////////////
-    GridCartesian         * UGrid    = SpaceTimeGrid::makeFourDimGrid(latt4, GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());
-    GridRedBlackCartesian * UrbGrid  = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
-    GridCartesian         * sUGrid   = SpaceTimeGrid::makeFourDimDWFGrid(latt4,GridDefaultMpi());
-    GridRedBlackCartesian * sUrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(sUGrid);
-    GridCartesian         * sFGrid   = SpaceTimeGrid::makeFiveDimDWFGrid(Ls,UGrid);
-    GridRedBlackCartesian * sFrbGrid = SpaceTimeGrid::makeFiveDimDWFRedBlackGrid(Ls,UGrid);
-
-    ///////// RNG Init ////////////
-    std::vector<int> seeds4({1,2,3,4});
-    std::vector<int> seeds5({5,6,7,8});
-    GridParallelRNG          RNG4(UGrid);  RNG4.SeedFixedIntegers(seeds4);
-    GridParallelRNG          RNG5(sFGrid);  RNG5.SeedFixedIntegers(seeds5);
-    std::cout << GridLogMessage << "Initialised RNGs" << std::endl;
-
-    ///////// Source preparation ////////////
-    LatticeFermion src   (sFGrid); 
-    LatticeFermion tmp   (sFGrid);
-    std::cout << GridLogMessage << "allocated src and tmp" << std::endl;
-    random(RNG5,src);
-    std::cout << GridLogMessage << "intialised random source" << std::endl;
-
-    RealD N2 = 1.0/::sqrt(norm2(src));
-    src = src*N2;
-    
-    LatticeGaugeField Umu(UGrid);  SU3::HotConfiguration(RNG4,Umu); 
-
-    WilsonFermion5DR sDw(Umu,*sFGrid,*sFrbGrid,*sUGrid,*sUrbGrid,M5);
-    LatticeFermion src_e (sFrbGrid);
-    LatticeFermion src_o (sFrbGrid);
-    LatticeFermion r_e   (sFrbGrid);
-    LatticeFermion r_o   (sFrbGrid);
-    LatticeFermion r_eo  (sFGrid);
-    LatticeFermion err   (sFGrid);
-    {
-
-      pickCheckerboard(Even,src_e,src);
-      pickCheckerboard(Odd,src_o,src);
-
-#if defined(AVX512) 
-      const int num_cases = 6;
-      std::string fmt("A/S ; A/O ; U/S ; U/O ; G/S ; G/O ");
-#else
-      const int num_cases = 4;
-      std::string fmt("U/S ; U/O ; G/S ; G/O ");
-#endif
-      controls Cases [] = {
-#ifdef AVX512
-	{  WilsonKernelsStatic::OptInlineAsm ,  WilsonKernelsStatic::CommsThenCompute ,CartesianCommunicator::CommunicatorPolicySequential  },
-	{  WilsonKernelsStatic::OptInlineAsm ,  WilsonKernelsStatic::CommsAndCompute  ,CartesianCommunicator::CommunicatorPolicySequential  },
-#endif
-	{  WilsonKernelsStatic::OptHandUnroll,  WilsonKernelsStatic::CommsThenCompute ,CartesianCommunicator::CommunicatorPolicySequential  },
-	{  WilsonKernelsStatic::OptHandUnroll,  WilsonKernelsStatic::CommsAndCompute  ,CartesianCommunicator::CommunicatorPolicySequential  },
-	{  WilsonKernelsStatic::OptGeneric   ,  WilsonKernelsStatic::CommsThenCompute ,CartesianCommunicator::CommunicatorPolicySequential  },
-	{  WilsonKernelsStatic::OptGeneric   ,  WilsonKernelsStatic::CommsAndCompute  ,CartesianCommunicator::CommunicatorPolicySequential  }
-      }; 
-
-      for(int c=0;c<num_cases;c++) {
-
-	 WilsonKernelsStatic::Comms = Cases[c].CommsOverlap;
-	 WilsonKernelsStatic::Opt   = Cases[c].Opt;
-	CartesianCommunicator::SetCommunicatorPolicy(Cases[c].CommsAsynch);
-
-	std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
-	if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptGeneric   ) std::cout << GridLogMessage<< "* Using GENERIC Nc WilsonKernels" <<std::endl;
-	if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptHandUnroll) std::cout << GridLogMessage<< "* Using Nc=3       WilsonKernels" <<std::endl;
-	if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptInlineAsm ) std::cout << GridLogMessage<< "* Using Asm Nc=3   WilsonKernels" <<std::endl;
-	if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsAndCompute ) std::cout << GridLogMessage<< "* Using Overlapped Comms/Compute" <<std::endl;
-	if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsThenCompute) std::cout << GridLogMessage<< "* Using sequential comms compute" <<std::endl;
-	if ( sizeof(Real)==4 )   std::cout << GridLogMessage<< "* SINGLE precision "<<std::endl;
-	if ( sizeof(Real)==8 )   std::cout << GridLogMessage<< "* DOUBLE precision "<<std::endl;
-	std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
-
-	int nwarm = 100;
-	uint64_t ncall = 1000;
-
-	double t0=usecond();
-	sFGrid->Barrier();
-	for(int i=0;i<nwarm;i++){
-	  sDw.DhopEO(src_o,r_e,DaggerNo);
-	}
-	sFGrid->Barrier();
-	double t1=usecond();
-
-	sDw.ZeroCounters();
-	time_statistics timestat;
-	std::vector<double> t_time(ncall);
-	for(uint64_t i=0;i<ncall;i++){
-	  t0=usecond();
-	  sDw.DhopEO(src_o,r_e,DaggerNo);
-	  t1=usecond();
-	  t_time[i] = t1-t0;
-	}
-	sFGrid->Barrier();
-	
-	double volume=Ls;  for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
-	double flops=(1344.0*volume)/2;
-	double mf_hi, mf_lo, mf_err;
-
-	timestat.statistics(t_time);
-	mf_hi = flops/timestat.min;
-	mf_lo = flops/timestat.max;
-	mf_err= flops/timestat.min * timestat.err/timestat.mean;
-
-	mflops = flops/timestat.mean;
-	mflops_all.push_back(mflops);
-	if ( mflops_best == 0   ) mflops_best = mflops;
-	if ( mflops_worst== 0   ) mflops_worst= mflops;
-	if ( mflops>mflops_best ) mflops_best = mflops;
-	if ( mflops<mflops_worst) mflops_worst= mflops;
-
-	std::cout<<GridLogMessage << std::fixed << std::setprecision(1)<<"sDeo mflop/s =   "<< mflops << " ("<<mf_err<<") " << mf_lo<<"-"<<mf_hi <<std::endl;
-	std::cout<<GridLogMessage << std::fixed << std::setprecision(1)<<"sDeo mflop/s per rank   "<< mflops/NP<<std::endl;
-	std::cout<<GridLogMessage << std::fixed << std::setprecision(1)<<"sDeo mflop/s per node   "<< mflops/NN<<std::endl;
-
-	sDw.Report();
-
-      }
-      double robust = mflops_worst/mflops_best;;
-      std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
-      std::cout<<GridLogMessage << L<<"^4 x "<<Ls<< " sDeo Best  mflop/s        =   "<< mflops_best << " ; " << mflops_best/NN<<" per node " <<std::endl;
-      std::cout<<GridLogMessage << L<<"^4 x "<<Ls<< " sDeo Worst mflop/s        =   "<< mflops_worst<< " ; " << mflops_worst/NN<<" per node " <<std::endl;
-
-      std::cout<<GridLogMessage <<std::setprecision(3)<< L<<"^4 x "<<Ls<< " Performance Robustness   =   "<< robust <<std::endl;
-      std::cout<<GridLogMessage <<fmt << std::endl;
-      std::cout<<GridLogMessage;
-
-      for(int i=0;i<mflops_all.size();i++){
-	std::cout<<mflops_all[i]/NN<<" ; " ;
-      }
-      std::cout<<std::endl;
-      std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
-
-    }
-    return mflops_best;
-  }
-#endif
-
-  static double DWF(int Ls,int L, double & robust)
+  static double DWF(int Ls,int L)
   {
     RealD mass=0.1;
     RealD M5  =1.8;
@@ -471,37 +283,30 @@ public:
     Coordinate mpi = GridDefaultMpi(); assert(mpi.size()==4);
     Coordinate local({L,L,L,L});
 
-    GridCartesian         * TmpGrid   = SpaceTimeGrid::makeFourDimGrid(Coordinate({64,64,64,64}), 
-								       GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());
+    GridCartesian         * TmpGrid   = SpaceTimeGrid::makeFourDimGrid(Coordinate({72,72,72,72}), 
+								       GridDefaultSimd(Nd,vComplex::Nsimd()),
+								       GridDefaultMpi());
     uint64_t NP = TmpGrid->RankCount();
     uint64_t NN = TmpGrid->NodeCount();
     NN_global=NN;
     uint64_t SHM=NP/NN;
 
-    Coordinate internal;
-    if      ( SHM == 1 )   internal = Coordinate({1,1,1,1});
-    else if ( SHM == 2 )   internal = Coordinate({2,1,1,1});
-    else if ( SHM == 4 )   internal = Coordinate({2,2,1,1});
-    else if ( SHM == 8 )   internal = Coordinate({2,2,2,1});
-    else assert(0);
-
-    Coordinate nodes({mpi[0]/internal[0],mpi[1]/internal[1],mpi[2]/internal[2],mpi[3]/internal[3]});
-    Coordinate latt4({local[0]*nodes[0],local[1]*nodes[1],local[2]*nodes[2],local[3]*nodes[3]});
+    Coordinate latt4({local[0]*mpi[0],local[1]*mpi[1],local[2]*mpi[2],local[3]*mpi[3]});
 
     ///////// Welcome message ////////////
     std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
     std::cout<<GridLogMessage << "Benchmark DWF on "<<L<<"^4 local volume "<<std::endl;
     std::cout<<GridLogMessage << "* Global volume  : "<<GridCmdVectorIntToString(latt4)<<std::endl;
     std::cout<<GridLogMessage << "* Ls             : "<<Ls<<std::endl;
-    std::cout<<GridLogMessage << "* MPI ranks      : "<<GridCmdVectorIntToString(mpi)<<std::endl;
-    std::cout<<GridLogMessage << "* Intranode      : "<<GridCmdVectorIntToString(internal)<<std::endl;
-    std::cout<<GridLogMessage << "* nodes          : "<<GridCmdVectorIntToString(nodes)<<std::endl;
+    std::cout<<GridLogMessage << "* ranks          : "<<NP  <<std::endl;
+    std::cout<<GridLogMessage << "* nodes          : "<<NN  <<std::endl;
+    std::cout<<GridLogMessage << "* ranks/node     : "<<SHM <<std::endl;
+    std::cout<<GridLogMessage << "* ranks geom     : "<<GridCmdVectorIntToString(mpi)<<std::endl;
     std::cout<<GridLogMessage << "* Using "<<threads<<" threads"<<std::endl;
     std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
 
-
     ///////// Lattice Init ////////////
-    GridCartesian         * UGrid   = SpaceTimeGrid::makeFourDimGrid(latt4, GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());
+    GridCartesian         * UGrid   = SpaceTimeGrid::makeFourDimGrid(latt4, GridDefaultSimd(Nd,vComplexF::Nsimd()),GridDefaultMpi());
     GridRedBlackCartesian * UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
     GridCartesian         * FGrid   = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
     GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid);
@@ -514,74 +319,31 @@ public:
     GridParallelRNG          RNG5(FGrid);  RNG5.SeedFixedIntegers(seeds5);
     std::cout << GridLogMessage << "Initialised RNGs" << std::endl;
 
+    typedef DomainWallFermionF Action;
+    typedef typename Action::FermionField Fermion;
+    typedef LatticeGaugeFieldF Gauge;
+    
     ///////// Source preparation ////////////
-    LatticeFermion src   (FGrid); random(RNG5,src);
-    LatticeFermion ref   (FGrid);
-    LatticeFermion tmp   (FGrid);
+    Gauge Umu(UGrid);  SU3::HotConfiguration(RNG4,Umu); 
+    Fermion src   (FGrid); random(RNG5,src);
+    Fermion src_e (FrbGrid);
+    Fermion src_o (FrbGrid);
+    Fermion r_e   (FrbGrid);
+    Fermion r_o   (FrbGrid);
+    Fermion r_eo  (FGrid);
+    Action Dw(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
 
-    RealD N2 = 1.0/::sqrt(norm2(src));
-    std::cout<<GridLogMessage << "Normalising src  "<< N2 <<std::endl;
-    src = src*N2;
-    
-    LatticeGaugeField Umu(UGrid);  SU3::HotConfiguration(RNG4,Umu); 
-    
-
-    DomainWallFermionR Dw(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
-
-    ////////////////////////////////////
-    // Naive wilson implementation
-    ////////////////////////////////////
-    {
-      LatticeGaugeField Umu5d(FGrid); 
-      std::vector<LatticeColourMatrix> U(4,FGrid);
-      auto Umu_v = Umu.View();
-      auto Umu5d_v = Umu5d.View();
-      for(int ss=0;ss<Umu.Grid()->oSites();ss++){
-	for(int s=0;s<Ls;s++){
-	  Umu5d_v[Ls*ss+s] = Umu_v[ss];
-	}
-      }
-      ref = Zero();
-      for(int mu=0;mu<Nd;mu++){
-	U[mu] = PeekIndex<LorentzIndex>(Umu5d,mu);
-      }
-      for(int mu=0;mu<Nd;mu++){
-	
-	tmp = U[mu]*Cshift(src,mu+1,1);
-	ref=ref + tmp - Gamma(Gmu[mu])*tmp;
-	
-	tmp =adj(U[mu])*src;
-	tmp =Cshift(tmp,mu+1,-1);
-	ref=ref + tmp + Gamma(Gmu[mu])*tmp;
-      }
-      ref = -0.5*ref;
-    }
-
-    LatticeFermion src_e (FrbGrid);
-    LatticeFermion src_o (FrbGrid);
-    LatticeFermion r_e   (FrbGrid);
-    LatticeFermion r_o   (FrbGrid);
-    LatticeFermion r_eo  (FGrid);
-    LatticeFermion err   (FGrid);
     {
 
       pickCheckerboard(Even,src_e,src);
       pickCheckerboard(Odd,src_o,src);
 
-#if defined(AVX512) 
-      const int num_cases = 6;
-      std::string fmt("A/S ; A/O ; U/S ; U/O ; G/S ; G/O ");
-#else
       const int num_cases = 4;
-      std::string fmt("U/S ; U/O ; G/S ; G/O ");
-#endif
+      std::string fmt("G/S/C ; G/O/C ; G/S/S ; G/O/S ");
+
       controls Cases [] = {
-#ifdef AVX512
-	{  WilsonKernelsStatic::OptInlineAsm ,  WilsonKernelsStatic::CommsThenCompute ,CartesianCommunicator::CommunicatorPolicySequential  },
-	{  WilsonKernelsStatic::OptInlineAsm ,  WilsonKernelsStatic::CommsAndCompute  ,CartesianCommunicator::CommunicatorPolicySequential  },
-#endif
-	{  WilsonKernelsStatic::OptHandUnroll,  WilsonKernelsStatic::CommsThenCompute ,CartesianCommunicator::CommunicatorPolicySequential  },
-	{  WilsonKernelsStatic::OptHandUnroll,  WilsonKernelsStatic::CommsAndCompute  ,CartesianCommunicator::CommunicatorPolicySequential  },
+	{  WilsonKernelsStatic::OptGeneric   ,  WilsonKernelsStatic::CommsThenCompute ,CartesianCommunicator::CommunicatorPolicyConcurrent  },
+	{  WilsonKernelsStatic::OptGeneric   ,  WilsonKernelsStatic::CommsAndCompute  ,CartesianCommunicator::CommunicatorPolicyConcurrent  },
 	{  WilsonKernelsStatic::OptGeneric   ,  WilsonKernelsStatic::CommsThenCompute ,CartesianCommunicator::CommunicatorPolicySequential  },
 	{  WilsonKernelsStatic::OptGeneric   ,  WilsonKernelsStatic::CommsAndCompute  ,CartesianCommunicator::CommunicatorPolicySequential  }
       }; 
@@ -594,15 +356,12 @@ public:
 
 	std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
 	if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptGeneric   ) std::cout << GridLogMessage<< "* Using GENERIC Nc WilsonKernels" <<std::endl;
-	if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptHandUnroll) std::cout << GridLogMessage<< "* Using Nc=3       WilsonKernels" <<std::endl;
-	if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptInlineAsm ) std::cout << GridLogMessage<< "* Using Asm Nc=3   WilsonKernels" <<std::endl;
 	if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsAndCompute ) std::cout << GridLogMessage<< "* Using Overlapped Comms/Compute" <<std::endl;
-	if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsThenCompute) std::cout << GridLogMessage<< "* Using sequential comms compute" <<std::endl;
-	if ( sizeof(Real)==4 )   std::cout << GridLogMessage<< "* SINGLE precision "<<std::endl;
-	if ( sizeof(Real)==8 )   std::cout << GridLogMessage<< "* DOUBLE precision "<<std::endl;
+	if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsThenCompute) std::cout << GridLogMessage<< "* Using sequential Comms/Compute" <<std::endl;
+	std::cout << GridLogMessage<< "* SINGLE precision "<<std::endl;
 	std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
 
-	int nwarm = 200;
+	int nwarm = 10;
 	double t0=usecond();
 	FGrid->Barrier();
 	for(int i=0;i<nwarm;i++){
@@ -610,9 +369,7 @@ public:
 	}
 	FGrid->Barrier();
 	double t1=usecond();
-	//	uint64_t ncall = (uint64_t) 2.5*1000.0*1000.0*nwarm/(t1-t0);
-	//	if (ncall < 500) ncall = 500;
-	uint64_t ncall = 1000;
+	uint64_t ncall = 50;
 
 	FGrid->Broadcast(0,&ncall,sizeof(ncall));
 
@@ -649,24 +406,11 @@ public:
 	std::cout<<GridLogMessage << std::fixed << std::setprecision(1)<<"Deo mflop/s per rank   "<< mflops/NP<<std::endl;
 	std::cout<<GridLogMessage << std::fixed << std::setprecision(1)<<"Deo mflop/s per node   "<< mflops/NN<<std::endl;
 
-	Dw.Report();
-
-	Dw.DhopEO(src_o,r_e,DaggerNo);
-	Dw.DhopOE(src_e,r_o,DaggerNo);
-	setCheckerboard(r_eo,r_o);
-	setCheckerboard(r_eo,r_e);
-	err = r_eo-ref; 
-	RealD absref = norm2(ref);
-	RealD abserr = norm2(err);
-	std::cout<<GridLogMessage << "norm diff   "<< abserr << " / " << absref<<std::endl;
-	assert(abserr<1.0e-4);
-
       }
-      robust = mflops_worst/mflops_best;
+
       std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
       std::cout<<GridLogMessage << L<<"^4 x "<<Ls<< " Deo Best  mflop/s        =   "<< mflops_best << " ; " << mflops_best/NN<<" per node " <<std::endl;
       std::cout<<GridLogMessage << L<<"^4 x "<<Ls<< " Deo Worst mflop/s        =   "<< mflops_worst<< " ; " << mflops_worst/NN<<" per node " <<std::endl;
-      std::cout<<GridLogMessage << std::fixed<<std::setprecision(3)<< L<<"^4 x "<<Ls<< " Performance Robustness   =   "<< robust  <<std::endl;
       std::cout<<GridLogMessage <<fmt << std::endl;
       std::cout<<GridLogMessage ;
 
@@ -680,8 +424,166 @@ public:
     return mflops_best;
   }
 
+
+  static double Staggered(int L)
+  {
+    double mflops;
+    double mflops_best = 0;
+    double mflops_worst= 0;
+    std::vector<double> mflops_all;
+
+    ///////////////////////////////////////////////////////
+    // Set/Get the layout & grid size
+    ///////////////////////////////////////////////////////
+    int threads = GridThread::GetThreads();
+    Coordinate mpi = GridDefaultMpi(); assert(mpi.size()==4);
+    Coordinate local({L,L,L,L});
+    
+    GridCartesian         * TmpGrid   = SpaceTimeGrid::makeFourDimGrid(Coordinate({72,72,72,72}), 
+								       GridDefaultSimd(Nd,vComplex::Nsimd()),
+								       GridDefaultMpi());
+    uint64_t NP = TmpGrid->RankCount();
+    uint64_t NN = TmpGrid->NodeCount();
+    NN_global=NN;
+    uint64_t SHM=NP/NN;
+
+    Coordinate latt4({local[0]*mpi[0],local[1]*mpi[1],local[2]*mpi[2],local[3]*mpi[3]});
+
+    ///////// Welcome message ////////////
+    std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
+    std::cout<<GridLogMessage << "Benchmark ImprovedStaggered on "<<L<<"^4 local volume "<<std::endl;
+    std::cout<<GridLogMessage << "* Global volume  : "<<GridCmdVectorIntToString(latt4)<<std::endl;
+    std::cout<<GridLogMessage << "* ranks          : "<<NP  <<std::endl;
+    std::cout<<GridLogMessage << "* nodes          : "<<NN  <<std::endl;
+    std::cout<<GridLogMessage << "* ranks/node     : "<<SHM <<std::endl;
+    std::cout<<GridLogMessage << "* ranks geom     : "<<GridCmdVectorIntToString(mpi)<<std::endl;
+    std::cout<<GridLogMessage << "* Using "<<threads<<" threads"<<std::endl;
+    std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
+
+    ///////// Lattice Init ////////////
+    GridCartesian         * FGrid   = SpaceTimeGrid::makeFourDimGrid(latt4, GridDefaultSimd(Nd,vComplexF::Nsimd()),GridDefaultMpi());
+    GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(FGrid);
+    
+    ///////// RNG Init ////////////
+    std::vector<int> seeds4({1,2,3,4});
+    GridParallelRNG          RNG4(FGrid);  RNG4.SeedFixedIntegers(seeds4);
+    std::cout << GridLogMessage << "Initialised RNGs" << std::endl;
+
+    RealD mass=0.1;
+    RealD c1=9.0/8.0;
+    RealD c2=-1.0/24.0;
+    RealD u0=1.0;
+
+    typedef ImprovedStaggeredFermionF Action;
+    typedef typename Action::FermionField Fermion; 
+    typedef LatticeGaugeFieldF Gauge;
+    
+    Gauge Umu(FGrid);  SU3::HotConfiguration(RNG4,Umu); 
+
+    typename Action::ImplParams params;
+    Action Ds(Umu,Umu,*FGrid,*FrbGrid,mass,c1,c2,u0,params);
+
+    ///////// Source preparation ////////////
+    Fermion src   (FGrid); random(RNG4,src);
+    Fermion src_e (FrbGrid);
+    Fermion src_o (FrbGrid);
+    Fermion r_e   (FrbGrid);
+    Fermion r_o   (FrbGrid);
+    Fermion r_eo  (FGrid);
+  
+    {
+
+      pickCheckerboard(Even,src_e,src);
+      pickCheckerboard(Odd,src_o,src);
+    
+      const int num_cases = 4;
+      std::string fmt("G/S/C ; G/O/C ; G/S/S ; G/O/S ");
+      
+      controls Cases [] = {
+	{  StaggeredKernelsStatic::OptGeneric   ,  StaggeredKernelsStatic::CommsThenCompute ,CartesianCommunicator::CommunicatorPolicyConcurrent  },
+	{  StaggeredKernelsStatic::OptGeneric   ,  StaggeredKernelsStatic::CommsAndCompute  ,CartesianCommunicator::CommunicatorPolicyConcurrent  },
+	{  StaggeredKernelsStatic::OptGeneric   ,  StaggeredKernelsStatic::CommsThenCompute ,CartesianCommunicator::CommunicatorPolicySequential  },
+	{  StaggeredKernelsStatic::OptGeneric   ,  StaggeredKernelsStatic::CommsAndCompute  ,CartesianCommunicator::CommunicatorPolicySequential  }
+      }; 
+
+      for(int c=0;c<num_cases;c++) {
+	
+	StaggeredKernelsStatic::Comms = Cases[c].CommsOverlap;
+	StaggeredKernelsStatic::Opt   = Cases[c].Opt;
+	CartesianCommunicator::SetCommunicatorPolicy(Cases[c].CommsAsynch);
+      
+	std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
+	if ( StaggeredKernelsStatic::Opt == StaggeredKernelsStatic::OptGeneric   ) std::cout << GridLogMessage<< "* Using GENERIC Nc StaggeredKernels" <<std::endl;
+	if ( StaggeredKernelsStatic::Comms == StaggeredKernelsStatic::CommsAndCompute ) std::cout << GridLogMessage<< "* Using Overlapped Comms/Compute" <<std::endl;
+	if ( StaggeredKernelsStatic::Comms == StaggeredKernelsStatic::CommsThenCompute) std::cout << GridLogMessage<< "* Using sequential Comms/Compute" <<std::endl;
+	std::cout << GridLogMessage<< "* SINGLE precision "<<std::endl;
+	std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
+	
+	int nwarm = 10;
+	double t0=usecond();
+	FGrid->Barrier();
+	for(int i=0;i<nwarm;i++){
+	  Ds.DhopEO(src_o,r_e,DaggerNo);
+	}
+	FGrid->Barrier();
+	double t1=usecond();
+	uint64_t ncall = 500;
+
+	FGrid->Broadcast(0,&ncall,sizeof(ncall));
+
+	//	std::cout << GridLogMessage << " Estimate " << ncall << " calls per second"<<std::endl;
+	Ds.ZeroCounters();
+
+	time_statistics timestat;
+	std::vector<double> t_time(ncall);
+	for(uint64_t i=0;i<ncall;i++){
+	  t0=usecond();
+	  Ds.DhopEO(src_o,r_e,DaggerNo);
+	  t1=usecond();
+	  t_time[i] = t1-t0;
+	}
+	FGrid->Barrier();
+	
+	double volume=1;  for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
+	double flops=(1146.0*volume)/2;
+	double mf_hi, mf_lo, mf_err;
+	
+	timestat.statistics(t_time);
+	mf_hi = flops/timestat.min;
+	mf_lo = flops/timestat.max;
+	mf_err= flops/timestat.min * timestat.err/timestat.mean;
+
+	mflops = flops/timestat.mean;
+	mflops_all.push_back(mflops);
+	if ( mflops_best == 0   ) mflops_best = mflops;
+	if ( mflops_worst== 0   ) mflops_worst= mflops;
+	if ( mflops>mflops_best ) mflops_best = mflops;
+	if ( mflops<mflops_worst) mflops_worst= mflops;
+	
+	std::cout<<GridLogMessage << std::fixed << std::setprecision(1)<<"Deo mflop/s =   "<< mflops << " ("<<mf_err<<") " << mf_lo<<"-"<<mf_hi <<std::endl;
+	std::cout<<GridLogMessage << std::fixed << std::setprecision(1)<<"Deo mflop/s per rank   "<< mflops/NP<<std::endl;
+	std::cout<<GridLogMessage << std::fixed << std::setprecision(1)<<"Deo mflop/s per node   "<< mflops/NN<<std::endl;
+      
+      }
+
+      std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
+      std::cout<<GridLogMessage << L<<"^4  Deo Best  mflop/s        =   "<< mflops_best << " ; " << mflops_best/NN<<" per node " <<std::endl;
+      std::cout<<GridLogMessage << L<<"^4  Deo Worst mflop/s        =   "<< mflops_worst<< " ; " << mflops_worst/NN<<" per node " <<std::endl;
+      std::cout<<GridLogMessage <<fmt << std::endl;
+      std::cout<<GridLogMessage ;
+
+      for(int i=0;i<mflops_all.size();i++){
+	std::cout<<mflops_all[i]/NN<<" ; " ;
+      }
+      std::cout<<std::endl;
+    }
+    std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
+    return mflops_best;
+  }
 };
 
+
+
 int main (int argc, char ** argv)
 {
   Grid_init(&argc,&argv);
@@ -696,62 +598,50 @@ int main (int argc, char ** argv)
 
   int do_memory=1;
   int do_comms =1;
-  int do_su3   =0;
-  int do_wilson=1;
-  int do_dwf   =1;
 
-  if ( do_su3 ) {
-    // empty for now
-  }
-#if 1
   int sel=2;
-  Coordinate L_list({8,12,16,24});
-#else
-  int sel=1;
-  Coordinate L_list({8,12});
-#endif
+  std::vector<int> L_list({16,24,32});
   int selm1=sel-1;
-  std::vector<double> robust_list;
 
   std::vector<double> wilson;
   std::vector<double> dwf4;
-  std::vector<double> dwf5;
+  std::vector<double> staggered;
 
-  if ( do_wilson ) {
-    int Ls=1;
-    std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
-    std::cout<<GridLogMessage << " Wilson dslash 4D vectorised" <<std::endl;
-    std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
-    for(int l=0;l<L_list.size();l++){
-      double robust;
-      wilson.push_back(Benchmark::DWF(Ls,L_list[l],robust));
-    }
+  int Ls=1;
+  std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
+  std::cout<<GridLogMessage << " Wilson dslash 4D vectorised" <<std::endl;
+  std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
+  for(int l=0;l<L_list.size();l++){
+    wilson.push_back(Benchmark::DWF(Ls,L_list[l]));
   }
 
-  int Ls=16;
-  if ( do_dwf ) {
-    std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
-    std::cout<<GridLogMessage << " Domain wall dslash 4D vectorised" <<std::endl;
-    std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
-    for(int l=0;l<L_list.size();l++){
-      double robust;
-      double result = Benchmark::DWF(Ls,L_list[l],robust) ;
-      dwf4.push_back(result);
-      robust_list.push_back(robust);
-    }
+  Ls=12;
+  std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
+  std::cout<<GridLogMessage << " Domain wall dslash 4D vectorised" <<std::endl;
+  std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
+  for(int l=0;l<L_list.size();l++){
+    double result = Benchmark::DWF(Ls,L_list[l]) ;
+    dwf4.push_back(result);
   }
 
-  if ( do_dwf ) {
+  /*
+  std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
+  std::cout<<GridLogMessage << " Improved Staggered dslash 4D vectorised" <<std::endl;
+  std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
+  for(int l=0;l<L_list.size();l++){
+    double result = Benchmark::Staggered(L_list[l]) ;
+    staggered.push_back(result);
+  }
+  */
 
   std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
   std::cout<<GridLogMessage << " Summary table Ls="<<Ls <<std::endl;
   std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
-  std::cout<<GridLogMessage << "L \t\t Wilson \t DWF4 " <<std::endl;
+  std::cout<<GridLogMessage << "L \t\t Wilson \t\t DWF4 \t\tt Staggered" <<std::endl;
   for(int l=0;l<L_list.size();l++){
-    std::cout<<GridLogMessage << L_list[l] <<" \t\t "<< wilson[l]<<" \t "<<dwf4[l] <<std::endl;
+    std::cout<<GridLogMessage << L_list[l] <<" \t\t "<< wilson[l]<<" \t\t "<<dwf4[l] <<std::endl;
   }
   std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
-  }
 
   int NN=NN_global;
   if ( do_memory ) {
@@ -768,24 +658,20 @@ int main (int argc, char ** argv)
     Benchmark::Comms();
   }
 
-  if ( do_dwf ) {
-  std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
-  std::cout<<GridLogMessage << " Per Node Summary table Ls="<<Ls <<std::endl;
-  std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
-  std::cout<<GridLogMessage << " L \t\t Wilson\t\t DWF4  " <<std::endl;
-  for(int l=0;l<L_list.size();l++){
-    std::cout<<GridLogMessage << L_list[l] <<" \t\t "<< wilson[l]/NN<<" \t "<<dwf4[l]/NN<<std::endl;
-  }
-  std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
+    std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
+    std::cout<<GridLogMessage << " Per Node Summary table Ls="<<Ls <<std::endl;
+    std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
+    std::cout<<GridLogMessage << " L \t\t Wilson\t\t DWF4  " <<std::endl;
+    for(int l=0;l<L_list.size();l++){
+      std::cout<<GridLogMessage << L_list[l] <<" \t\t "<< wilson[l]/NN<<" \t "<<dwf4[l]/NN<<std::endl;
+    }
+    std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
 
-  std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
-  std::cout<<GridLogMessage << " Comparison point     result: "  << 0.5*(dwf4[sel]+dwf4[selm1])/NN << " Mflop/s per node"<<std::endl;
-  std::cout<<GridLogMessage << " Comparison point is 0.5*("<<dwf4[sel]/NN<<"+"<<dwf4[selm1]/NN << ") "<<std::endl;
-  std::cout<<std::setprecision(3);
-  std::cout<<GridLogMessage << " Comparison point robustness: "  << robust_list[sel] <<std::endl;
-  std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
-
-  }
+    std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
+    std::cout<<GridLogMessage << " Comparison point     result: "  << 0.5*(dwf4[sel]+dwf4[selm1])/NN << " Mflop/s per node"<<std::endl;
+    std::cout<<GridLogMessage << " Comparison point is 0.5*("<<dwf4[sel]/NN<<"+"<<dwf4[selm1]/NN << ") "<<std::endl;
+    std::cout<<std::setprecision(3);
+    std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
 
   Grid_finalize();
 }