Merge pull request #409 from giltirn/feature/dirichlet-gparity-stage

Import round 5

Grid/DisableWarnings.h

@@ -44,14 +44,22 @@ directory
 #ifdef __NVCC__
  //disables nvcc specific warning in json.hpp
 #pragma clang diagnostic ignored "-Wdeprecated-register"
+
+#if (__CUDACC_VER_MAJOR__ >= 11) && (__CUDACC_VER_MINOR__ >= 5)
+ //disables nvcc specific warning in json.hpp
+#pragma nv_diag_suppress unsigned_compare_with_zero
+#pragma nv_diag_suppress cast_to_qualified_type
+ //disables nvcc specific warning in many files
+#pragma nv_diag_suppress esa_on_defaulted_function_ignored
+#pragma nv_diag_suppress extra_semicolon
+#else
+ //disables nvcc specific warning in json.hpp
 #pragma diag_suppress unsigned_compare_with_zero
 #pragma diag_suppress cast_to_qualified_type
-
  //disables nvcc specific warning in many files
 #pragma diag_suppress esa_on_defaulted_function_ignored
 #pragma diag_suppress extra_semicolon
-
-//Eigen only
+#endif
 #endif
 
 // Disable vectorisation in Eigen on the Power8/9 and PowerPC

Grid/Grid_Eigen_Dense.h

@@ -14,7 +14,11 @@
 /* NVCC save and restore compile environment*/
 #ifdef __NVCC__
 #pragma push
+#if (__CUDACC_VER_MAJOR__ >= 11) && (__CUDACC_VER_MINOR__ >= 5)
+#pragma nv_diag_suppress code_is_unreachable
+#else
 #pragma diag_suppress code_is_unreachable
+#endif
 #pragma push_macro("__CUDA_ARCH__")
 #pragma push_macro("__NVCC__")
 #pragma push_macro("__CUDACC__")

Grid/algorithms/iterative/ConjugateGradient.h

@@ -120,6 +120,9 @@ public:
     SolverTimer.Start();
     int k;
     for (k = 1; k <= MaxIterations; k++) {
+
+      GridStopWatch IterationTimer;
+      IterationTimer.Start();
       c = cp;
 
       MatrixTimer.Start();
@@ -152,8 +155,14 @@ public:
       LinearCombTimer.Stop();
       LinalgTimer.Stop();
 
-      std::cout << GridLogIterative << "ConjugateGradient: Iteration " << k
+      IterationTimer.Stop();
+      if ( (k % 500) == 0 ) {
+	std::cout << GridLogMessage << "ConjugateGradient: Iteration " << k
                 << " residual " << sqrt(cp/ssq) << " target " << Tolerance << std::endl;
+      } else { 
+	std::cout << GridLogIterative << "ConjugateGradient: Iteration " << k
+		  << " residual " << sqrt(cp/ssq) << " target " << Tolerance << " took " << IterationTimer.Elapsed() << std::endl;
+      }
 
       // Stopping condition
       if (cp <= rsq) {
@@ -170,13 +179,13 @@ public:
 		  << "\tTrue residual " << true_residual
 		  << "\tTarget " << Tolerance << std::endl;
 
-        std::cout << GridLogIterative << "Time breakdown "<<std::endl;
-	std::cout << GridLogIterative << "\tElapsed    " << SolverTimer.Elapsed() <<std::endl;
-	std::cout << GridLogIterative << "\tMatrix     " << MatrixTimer.Elapsed() <<std::endl;
-	std::cout << GridLogIterative << "\tLinalg     " << LinalgTimer.Elapsed() <<std::endl;
-	std::cout << GridLogIterative << "\tInner      " << InnerTimer.Elapsed() <<std::endl;
-	std::cout << GridLogIterative << "\tAxpyNorm   " << AxpyNormTimer.Elapsed() <<std::endl;
-	std::cout << GridLogIterative << "\tLinearComb " << LinearCombTimer.Elapsed() <<std::endl;
+        std::cout << GridLogMessage << "Time breakdown "<<std::endl;
+	std::cout << GridLogMessage << "\tElapsed    " << SolverTimer.Elapsed() <<std::endl;
+	std::cout << GridLogMessage << "\tMatrix     " << MatrixTimer.Elapsed() <<std::endl;
+	std::cout << GridLogMessage << "\tLinalg     " << LinalgTimer.Elapsed() <<std::endl;
+	std::cout << GridLogMessage << "\tInner      " << InnerTimer.Elapsed() <<std::endl;
+	std::cout << GridLogMessage << "\tAxpyNorm   " << AxpyNormTimer.Elapsed() <<std::endl;
+	std::cout << GridLogMessage << "\tLinearComb " << LinearCombTimer.Elapsed() <<std::endl;
 
         if (ErrorOnNoConverge) assert(true_residual / Tolerance < 10000.0);
 

Grid/algorithms/iterative/ConjugateGradientMultiShift.h

@@ -44,7 +44,7 @@ public:
 
   using OperatorFunction<Field>::operator();
 
-  RealD   Tolerance;
+  //  RealD   Tolerance;
   Integer MaxIterations;
   Integer IterationsToComplete; //Number of iterations the CG took to finish. Filled in upon completion
   std::vector<int> IterationsToCompleteShift;  // Iterations for this shift
@@ -324,8 +324,8 @@ public:
 
       std::cout << GridLogMessage << "Time Breakdown "<<std::endl;
       std::cout << GridLogMessage << "\tElapsed    " << SolverTimer.Elapsed()     <<std::endl;
-      std::cout << GridLogMessage << "\tAXPY    " << AXPYTimer.Elapsed()     <<std::endl;
-      std::cout << GridLogMessage << "\tMarix    " << MatrixTimer.Elapsed()     <<std::endl;
+      std::cout << GridLogMessage << "\tAXPY     " << AXPYTimer.Elapsed()     <<std::endl;
+      std::cout << GridLogMessage << "\tMatrix   " << MatrixTimer.Elapsed()     <<std::endl;
       std::cout << GridLogMessage << "\tShift    " << ShiftTimer.Elapsed()     <<std::endl;
 
       IterationsToComplete = k;	

Grid/algorithms/iterative/Deflation.h

@@ -113,7 +113,43 @@ public:
     blockPromote(guess_coarse,guess,subspace);
     guess.Checkerboard() = src.Checkerboard();
   };
-};
+
+  void operator()(const std::vector<FineField> &src,std::vector<FineField> &guess) {
+    int Nevec = (int)evec_coarse.size();
+    int Nsrc = (int)src.size();
+    // make temp variables
+    std::vector<CoarseField> src_coarse(Nsrc,evec_coarse[0].Grid());
+    std::vector<CoarseField> guess_coarse(Nsrc,evec_coarse[0].Grid());    
+    //Preporcessing
+    std::cout << GridLogMessage << "Start BlockProject for loop" << std::endl;
+    for (int j=0;j<Nsrc;j++)
+    {
+    guess_coarse[j] = Zero();
+    std::cout << GridLogMessage << "BlockProject iter: " << j << std::endl;
+    blockProject(src_coarse[j],src[j],subspace);
+    }
+    //deflation set up for eigen vector batchsize 1 and source batch size equal number of sources
+    std::cout << GridLogMessage << "Start ProjectAccum for loop" << std::endl;
+    for (int i=0;i<Nevec;i++)
+    {
+      std::cout << GridLogMessage << "ProjectAccum Nvec: " << i << std::endl;
+      const CoarseField & tmp = evec_coarse[i];
+      for (int j=0;j<Nsrc;j++)
+      {
+        axpy(guess_coarse[j],TensorRemove(innerProduct(tmp,src_coarse[j])) / eval_coarse[i],tmp,guess_coarse[j]);
+      }
+    }
+    //postprocessing
+    std::cout << GridLogMessage << "Start BlockPromote for loop" << std::endl;
+    for (int j=0;j<Nsrc;j++)
+    {
+    std::cout << GridLogMessage << "BlockProject iter: " << j << std::endl;
+    blockPromote(guess_coarse[j],guess[j],subspace);
+    guess[j].Checkerboard() = src[j].Checkerboard();
+    }
+  };
+
+  };
 
 
 

Grid/algorithms/iterative/LocalCoherenceLanczos.h

@@ -146,14 +146,21 @@ public:
   LinearOperatorBase<FineField> &_Linop;
   RealD                             _coarse_relax_tol;
   std::vector<FineField>        &_subspace;
+
+  int _largestEvalIdxForReport; //The convergence of the LCL is based on the evals of the coarse grid operator, not those of the underlying fine grid operator
+                                //As a result we do not know what the eval range of the fine operator is until the very end, making tuning the Cheby bounds very difficult
+                                //To work around this issue, every restart we separately reconstruct the fine operator eval for the lowest and highest evec and print these
+                                //out alongside the evals of the coarse operator. To do so we need to know the index of the largest eval (i.e. Nstop-1)
+                                //NOTE: If largestEvalIdxForReport=-1 (default) then this is not performed
   
   ImplicitlyRestartedLanczosSmoothedTester(LinearFunction<CoarseField>   &Poly,
 					   OperatorFunction<FineField>   &smoother,
 					   LinearOperatorBase<FineField> &Linop,
 					   std::vector<FineField>        &subspace,
-					   RealD coarse_relax_tol=5.0e3) 
+					   RealD coarse_relax_tol=5.0e3,
+					   int largestEvalIdxForReport=-1) 
     : _smoother(smoother), _Linop(Linop), _Poly(Poly), _subspace(subspace),
-      _coarse_relax_tol(coarse_relax_tol)  
+      _coarse_relax_tol(coarse_relax_tol), _largestEvalIdxForReport(largestEvalIdxForReport)
   {    };
 
   //evalMaxApprox: approximation of largest eval of the fine Chebyshev operator (suitably wrapped by block projection)
@@ -179,6 +186,12 @@ public:
 	     <<" |H B[i] - eval[i]B[i]|^2 / evalMaxApprox^2 " << std::setw(25) << vv
 	     <<std::endl;
 
+    if(_largestEvalIdxForReport != -1 && (j==0 || j==_largestEvalIdxForReport)){
+      std::cout<<GridLogIRL << "Estimating true eval of fine grid operator for eval idx " << j << std::endl;
+      RealD tmp_eval;
+      ReconstructEval(j,eresid,B,tmp_eval,1.0); //don't use evalMaxApprox of coarse operator! (cf below)
+    }
+    
     int conv=0;
     if( (vv<eresid*eresid) ) conv = 1;
     return conv;
@@ -409,7 +422,7 @@ public:
     //////////////////////////////////////////////////////////////////////////////////////////////////
 
     Chebyshev<FineField>                                           ChebySmooth(cheby_smooth); //lower order Chebyshev of fine operator on fine grid used to smooth regenerated eigenvectors
-    ImplicitlyRestartedLanczosSmoothedTester<Fobj,CComplex,nbasis> ChebySmoothTester(ChebyOp,ChebySmooth,_FineOp,subspace,relax); 
+    ImplicitlyRestartedLanczosSmoothedTester<Fobj,CComplex,nbasis> ChebySmoothTester(ChebyOp,ChebySmooth,_FineOp,subspace,relax,Nstop-1); 
 
     evals_coarse.resize(Nm);
     evec_coarse.resize(Nm,_CoarseGrid);

Grid/communicator/Communicator_mpi3.cc

@@ -372,7 +372,7 @@ double CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsReques
   double off_node_bytes=0.0;
   int tag;
 
-  if ( dox ) {
+  if ( dor ) {
     if ( (gfrom ==MPI_UNDEFINED) || Stencil_force_mpi ) {
       tag= dir+from*32;
       ierr=MPI_Irecv(recv, bytes, MPI_CHAR,from,tag,communicator_halo[commdir],&rrq);
@@ -382,7 +382,7 @@ double CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsReques
     }
   }
   
-  if (dor) {
+  if (dox) {
     if ( (gdest == MPI_UNDEFINED) || Stencil_force_mpi ) {
       tag= dir+_processor*32;
       ierr =MPI_Isend(xmit, bytes, MPI_CHAR,dest,tag,communicator_halo[commdir],&xrq);
@@ -390,16 +390,15 @@ double CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsReques
       list.push_back(xrq);
       off_node_bytes+=bytes;
     } else {
-    // TODO : make a OMP loop on CPU, call threaded bcopy
       void *shm = (void *) this->ShmBufferTranslate(dest,recv);
       assert(shm!=NULL);
-      //    std::cout <<"acceleratorCopyDeviceToDeviceAsynch"<< std::endl;
       acceleratorCopyDeviceToDeviceAsynch(xmit,shm,bytes);
     }
   }
   
   if ( CommunicatorPolicy == CommunicatorPolicySequential ) {
     this->StencilSendToRecvFromComplete(list,dir);
+    list.resize(0);
   }
 
   return off_node_bytes;

Grid/lattice/Lattice_peekpoke.h

@@ -125,6 +125,12 @@ void pokeSite(const sobj &s,Lattice<vobj> &l,const Coordinate &site){
 //////////////////////////////////////////////////////////
 // Peek a scalar object from the SIMD array
 //////////////////////////////////////////////////////////
+template<class vobj>
+typename vobj::scalar_object peekSite(const Lattice<vobj> &l,const Coordinate &site){
+  typename vobj::scalar_object s;
+  peekSite(s,l,site);
+  return s;
+}        
 template<class vobj,class sobj>
 void peekSite(sobj &s,const Lattice<vobj> &l,const Coordinate &site){
         

Grid/lattice/Lattice_reduction.h

@@ -232,6 +232,7 @@ inline ComplexD rankInnerProduct(const Lattice<vobj> &left,const Lattice<vobj> &
   const uint64_t sites = grid->oSites();
   
   // Might make all code paths go this way.
+#if 0
   typedef decltype(innerProductD(vobj(),vobj())) inner_t;
   Vector<inner_t> inner_tmp(sites);
   auto inner_tmp_v = &inner_tmp[0];
@@ -241,15 +242,31 @@ inline ComplexD rankInnerProduct(const Lattice<vobj> &left,const Lattice<vobj> &
     autoView( right_v,right, AcceleratorRead);
 
     // GPU - SIMT lane compliance...
-    accelerator_for( ss, sites, 1,{
-	auto x_l = left_v[ss];
-	auto y_l = right_v[ss];
-	inner_tmp_v[ss]=innerProductD(x_l,y_l);
+    accelerator_for( ss, sites, nsimd,{
+	auto x_l = left_v(ss);
+	auto y_l = right_v(ss);
+	coalescedWrite(inner_tmp_v[ss],innerProductD(x_l,y_l));
     });
   }
+#else
+  typedef decltype(innerProduct(vobj(),vobj())) inner_t;
+  Vector<inner_t> inner_tmp(sites);
+  auto inner_tmp_v = &inner_tmp[0];
+    
+  {
+    autoView( left_v , left, AcceleratorRead);
+    autoView( right_v,right, AcceleratorRead);
 
+    // GPU - SIMT lane compliance...
+    accelerator_for( ss, sites, nsimd,{
+	auto x_l = left_v(ss);
+	auto y_l = right_v(ss);
+	coalescedWrite(inner_tmp_v[ss],innerProduct(x_l,y_l));
+    });
+  }
+#endif
   // This is in single precision and fails some tests
-  auto anrm = sum(inner_tmp_v,sites);  
+  auto anrm = sumD(inner_tmp_v,sites);  
   nrm = anrm;
   return nrm;
 }
@@ -283,7 +300,7 @@ axpby_norm_fast(Lattice<vobj> &z,sobj a,sobj b,const Lattice<vobj> &x,const Latt
   conformable(x,y);
 
   typedef typename vobj::scalar_type scalar_type;
-  typedef typename vobj::vector_typeD vector_type;
+  //  typedef typename vobj::vector_typeD vector_type;
   RealD  nrm;
   
   GridBase *grid = x.Grid();
@@ -295,17 +312,29 @@ axpby_norm_fast(Lattice<vobj> &z,sobj a,sobj b,const Lattice<vobj> &x,const Latt
   autoView( x_v, x, AcceleratorRead);
   autoView( y_v, y, AcceleratorRead);
   autoView( z_v, z, AcceleratorWrite);
-
+#if 0
   typedef decltype(innerProductD(x_v[0],y_v[0])) inner_t;
   Vector<inner_t> inner_tmp(sites);
   auto inner_tmp_v = &inner_tmp[0];
 
-  accelerator_for( ss, sites, 1,{
-      auto tmp = a*x_v[ss]+b*y_v[ss];
-      inner_tmp_v[ss]=innerProductD(tmp,tmp);
-      z_v[ss]=tmp;
+  accelerator_for( ss, sites, nsimd,{
+      auto tmp = a*x_v(ss)+b*y_v(ss);
+      coalescedWrite(inner_tmp_v[ss],innerProductD(tmp,tmp));
+      coalescedWrite(z_v[ss],tmp);
   });
   nrm = real(TensorRemove(sum(inner_tmp_v,sites)));
+#else
+  typedef decltype(innerProduct(x_v[0],y_v[0])) inner_t;
+  Vector<inner_t> inner_tmp(sites);
+  auto inner_tmp_v = &inner_tmp[0];
+
+  accelerator_for( ss, sites, nsimd,{
+      auto tmp = a*x_v(ss)+b*y_v(ss);
+      coalescedWrite(inner_tmp_v[ss],innerProduct(tmp,tmp));
+      coalescedWrite(z_v[ss],tmp);
+  });
+  nrm = real(TensorRemove(sumD(inner_tmp_v,sites)));
+#endif
   grid->GlobalSum(nrm);
   return nrm; 
 }

Grid/lattice/Lattice_rng.h

@@ -424,9 +424,32 @@ public:
     // MT implementation does not implement fast discard even though
     // in principle this is possible
     ////////////////////////////////////////////////
+#if 1
+    thread_for( lidx, _grid->lSites(), {
 
+	int gidx;
+	int o_idx;
+	int i_idx;
+	int rank;
+	Coordinate pcoor;
+	Coordinate lcoor;
+	Coordinate gcoor;
+	_grid->LocalIndexToLocalCoor(lidx,lcoor);
+	pcoor=_grid->ThisProcessorCoor();
+	_grid->ProcessorCoorLocalCoorToGlobalCoor(pcoor,lcoor,gcoor);
+	_grid->GlobalCoorToGlobalIndex(gcoor,gidx);
+
+	_grid->GlobalCoorToRankIndex(rank,o_idx,i_idx,gcoor);
+	assert(rank == _grid->ThisRank() );
+	
+	int l_idx=generator_idx(o_idx,i_idx);
+	_generators[l_idx] = master_engine;
+	Skip(_generators[l_idx],gidx); // Skip to next RNG sequence
+    });
+#else
     // Everybody loops over global volume.
     thread_for( gidx, _grid->_gsites, {
+
 	// Where is it?
 	int rank;
 	int o_idx;
@@ -443,6 +466,7 @@ public:
 	  Skip(_generators[l_idx],gidx); // Skip to next RNG sequence
 	}
     });
+#endif
 #else 
     ////////////////////////////////////////////////////////////////
     // Machine and thread decomposition dependent seeding is efficient

Grid/log/Log.cc

@@ -65,6 +65,7 @@ GridLogger GridLogSolver (1, "Solver", GridLogColours, "NORMAL");
 GridLogger GridLogError  (1, "Error" , GridLogColours, "RED");
 GridLogger GridLogWarning(1, "Warning", GridLogColours, "YELLOW");
 GridLogger GridLogMessage(1, "Message", GridLogColours, "NORMAL");
+GridLogger GridLogMemory (1, "Memory", GridLogColours, "NORMAL");
 GridLogger GridLogDebug  (1, "Debug", GridLogColours, "PURPLE");
 GridLogger GridLogPerformance(1, "Performance", GridLogColours, "GREEN");
 GridLogger GridLogIterative  (1, "Iterative", GridLogColours, "BLUE");
@@ -72,9 +73,10 @@ GridLogger GridLogIntegrator (1, "Integrator", GridLogColours, "BLUE");
 GridLogger GridLogHMC (1, "HMC", GridLogColours, "BLUE");
 
 void GridLogConfigure(std::vector<std::string> &logstreams) {
-  GridLogError.Active(0);
+  GridLogError.Active(1);
   GridLogWarning.Active(0);
   GridLogMessage.Active(1); // at least the messages should be always on
+  GridLogMemory.Active(0); // at least the messages should be always on
   GridLogIterative.Active(0);
   GridLogDebug.Active(0);
   GridLogPerformance.Active(0);
@@ -83,7 +85,7 @@ void GridLogConfigure(std::vector<std::string> &logstreams) {
   GridLogHMC.Active(1);
 
   for (int i = 0; i < logstreams.size(); i++) {
-    if (logstreams[i] == std::string("Error"))       GridLogError.Active(1);
+    if (logstreams[i] == std::string("Memory"))      GridLogMemory.Active(1);
     if (logstreams[i] == std::string("Warning"))     GridLogWarning.Active(1);
     if (logstreams[i] == std::string("NoMessage"))   GridLogMessage.Active(0);
     if (logstreams[i] == std::string("Iterative"))   GridLogIterative.Active(1);

Grid/log/Log.h

@@ -183,6 +183,7 @@ extern GridLogger GridLogPerformance;
 extern GridLogger GridLogIterative  ;
 extern GridLogger GridLogIntegrator  ;
 extern GridLogger GridLogHMC;
+extern GridLogger GridLogMemory;
 extern Colours    GridLogColours;
 
 std::string demangle(const char* name) ;

Grid/parallelIO/IldgIO.h

@@ -31,6 +31,7 @@ directory
 #include <fstream>
 #include <iomanip>
 #include <iostream>
+#include <string>
 #include <map>
 
 #include <pwd.h>
@@ -654,7 +655,8 @@ class IldgWriter : public ScidacWriter {
     // Fill ILDG header data struct
     //////////////////////////////////////////////////////
     ildgFormat ildgfmt ;
-    ildgfmt.field     = std::string("su3gauge");
+    const std::string stNC = std::to_string( Nc ) ;
+    ildgfmt.field          = std::string("su"+stNC+"gauge");
 
     if ( format == std::string("IEEE32BIG") ) { 
       ildgfmt.precision = 32;
@@ -871,7 +873,8 @@ class IldgReader : public GridLimeReader {
     } else { 
 
       assert(found_ildgFormat);
-      assert ( ildgFormat_.field == std::string("su3gauge") );
+      const std::string stNC = std::to_string( Nc ) ;
+      assert ( ildgFormat_.field == std::string("su"+stNC+"gauge") );
 
       ///////////////////////////////////////////////////////////////////////////////////////
       // Populate our Grid metadata as best we can
@@ -879,7 +882,7 @@ class IldgReader : public GridLimeReader {
 
       std::ostringstream vers; vers << ildgFormat_.version;
       FieldMetaData_.hdr_version = vers.str();
-      FieldMetaData_.data_type = std::string("4D_SU3_GAUGE_3X3");
+      FieldMetaData_.data_type = std::string("4D_SU"+stNC+"_GAUGE_"+stNC+"x"+stNC);
 
       FieldMetaData_.nd=4;
       FieldMetaData_.dimension.resize(4);

Grid/parallelIO/MetaData.h

@@ -6,8 +6,8 @@
 
     Copyright (C) 2015
 
-
     Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+    Author: Jamie Hudspith <renwick.james.hudspth@gmail.com>
 
     This program is free software; you can redistribute it and/or modify
     it under the terms of the GNU General Public License as published by
@@ -182,8 +182,8 @@ class GaugeStatistics
 public:
   void operator()(Lattice<vLorentzColourMatrixD> & data,FieldMetaData &header)
   {
-    header.link_trace=WilsonLoops<Impl>::linkTrace(data);
-    header.plaquette =WilsonLoops<Impl>::avgPlaquette(data);
+    header.link_trace = WilsonLoops<Impl>::linkTrace(data);
+    header.plaquette  = WilsonLoops<Impl>::avgPlaquette(data);
   }
 };
 typedef GaugeStatistics<PeriodicGimplD> PeriodicGaugeStatistics;
@@ -203,20 +203,24 @@ template<> inline void PrepareMetaData<vLorentzColourMatrixD>(Lattice<vLorentzCo
 //////////////////////////////////////////////////////////////////////
 inline void reconstruct3(LorentzColourMatrix & cm)
 {
-  const int x=0;
-  const int y=1;
-  const int z=2;
+  assert( Nc < 4 && Nc > 1 ) ;
   for(int mu=0;mu<Nd;mu++){
-    cm(mu)()(2,x) = adj(cm(mu)()(0,y)*cm(mu)()(1,z)-cm(mu)()(0,z)*cm(mu)()(1,y)); //x= yz-zy
-    cm(mu)()(2,y) = adj(cm(mu)()(0,z)*cm(mu)()(1,x)-cm(mu)()(0,x)*cm(mu)()(1,z)); //y= zx-xz
-    cm(mu)()(2,z) = adj(cm(mu)()(0,x)*cm(mu)()(1,y)-cm(mu)()(0,y)*cm(mu)()(1,x)); //z= xy-yx
+    #if Nc == 2
+      cm(mu)()(1,0) = -adj(cm(mu)()(0,y)) ;
+      cm(mu)()(1,1) =  adj(cm(mu)()(0,x)) ;
+    #else
+      const int x=0 , y=1 , z=2 ; // a little disinenuous labelling
+      cm(mu)()(2,x) = adj(cm(mu)()(0,y)*cm(mu)()(1,z)-cm(mu)()(0,z)*cm(mu)()(1,y)); //x= yz-zy
+      cm(mu)()(2,y) = adj(cm(mu)()(0,z)*cm(mu)()(1,x)-cm(mu)()(0,x)*cm(mu)()(1,z)); //y= zx-xz
+      cm(mu)()(2,z) = adj(cm(mu)()(0,x)*cm(mu)()(1,y)-cm(mu)()(0,y)*cm(mu)()(1,x)); //z= xy-yx
+    #endif
   }
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // Some data types for intermediate storage
 ////////////////////////////////////////////////////////////////////////////////
-template<typename vtype> using iLorentzColour2x3 = iVector<iVector<iVector<vtype, Nc>, 2>, Nd >;
+template<typename vtype> using iLorentzColour2x3 = iVector<iVector<iVector<vtype, Nc>, Nc-1>, Nd >;
 
 typedef iLorentzColour2x3<Complex>  LorentzColour2x3;
 typedef iLorentzColour2x3<ComplexF> LorentzColour2x3F;
@@ -278,7 +282,6 @@ struct GaugeSimpleMunger{
 
 template <class fobj, class sobj>
 struct GaugeSimpleUnmunger {
-
   void operator()(sobj &in, fobj &out) {
     for (int mu = 0; mu < Nd; mu++) {
       for (int i = 0; i < Nc; i++) {
@@ -317,8 +320,8 @@ template<class fobj,class sobj>
 struct Gauge3x2munger{
   void operator() (fobj &in,sobj &out){
     for(int mu=0;mu<Nd;mu++){
-      for(int i=0;i<2;i++){
-	for(int j=0;j<3;j++){
+      for(int i=0;i<Nc-1;i++){
+	for(int j=0;j<Nc;j++){
 	  out(mu)()(i,j) = in(mu)(i)(j);
 	}}
     }
@@ -330,8 +333,8 @@ template<class fobj,class sobj>
 struct Gauge3x2unmunger{
   void operator() (sobj &in,fobj &out){
     for(int mu=0;mu<Nd;mu++){
-      for(int i=0;i<2;i++){
-	for(int j=0;j<3;j++){
+      for(int i=0;i<Nc-1;i++){
+	for(int j=0;j<Nc;j++){
 	  out(mu)(i)(j) = in(mu)()(i,j);
 	}}
     }

Grid/parallelIO/NerscIO.h

@@ -9,6 +9,7 @@
     Author: Matt Spraggs <matthew.spraggs@gmail.com>
     Author: Peter Boyle <paboyle@ph.ed.ac.uk>
     Author: paboyle <paboyle@ph.ed.ac.uk>
+    Author: Jamie Hudspith <renwick.james.hudspth@gmail.com>
 
     This program is free software; you can redistribute it and/or modify
     it under the terms of the GNU General Public License as published by
@@ -30,6 +31,8 @@
 #ifndef GRID_NERSC_IO_H
 #define GRID_NERSC_IO_H
 
+#include <string>
+
 NAMESPACE_BEGIN(Grid);
 
 using namespace Grid;
@@ -147,15 +150,17 @@ public:
 
     std::string format(header.floating_point);
 
-    int ieee32big = (format == std::string("IEEE32BIG"));
-    int ieee32    = (format == std::string("IEEE32"));
-    int ieee64big = (format == std::string("IEEE64BIG"));
-    int ieee64    = (format == std::string("IEEE64") || format == std::string("IEEE64LITTLE"));
+    const int ieee32big = (format == std::string("IEEE32BIG"));
+    const int ieee32    = (format == std::string("IEEE32"));
+    const int ieee64big = (format == std::string("IEEE64BIG"));
+    const int ieee64    = (format == std::string("IEEE64") || \
+			   format == std::string("IEEE64LITTLE"));
 
     uint32_t nersc_csum,scidac_csuma,scidac_csumb;
     // depending on datatype, set up munger;
     // munger is a function of <floating point, Real, data_type>
-    if ( header.data_type == std::string("4D_SU3_GAUGE") ) {
+    const std::string stNC = std::to_string( Nc ) ;
+    if ( header.data_type == std::string("4D_SU"+stNC+"_GAUGE") ) {
       if ( ieee32 || ieee32big ) {
 	BinaryIO::readLatticeObject<vLorentzColourMatrixD, LorentzColour2x3F> 
 	  (Umu,file,Gauge3x2munger<LorentzColour2x3F,LorentzColourMatrix>(), offset,format,
@@ -166,7 +171,7 @@ public:
 	  (Umu,file,Gauge3x2munger<LorentzColour2x3D,LorentzColourMatrix>(),offset,format,
 	   nersc_csum,scidac_csuma,scidac_csumb);
       }
-    } else if ( header.data_type == std::string("4D_SU3_GAUGE_3x3") ) {
+    } else if ( header.data_type == std::string("4D_SU"+stNC+"_GAUGE_"+stNC+"x"+stNC) ) {
       if ( ieee32 || ieee32big ) {
 	BinaryIO::readLatticeObject<vLorentzColourMatrixD,LorentzColourMatrixF>
 	  (Umu,file,GaugeSimpleMunger<LorentzColourMatrixF,LorentzColourMatrix>(),offset,format,
@@ -211,27 +216,29 @@ public:
   template<class GaugeStats=PeriodicGaugeStatistics>
   static inline void writeConfiguration(Lattice<vLorentzColourMatrixD > &Umu,
 					std::string file, 
-					std::string ens_label = std::string("DWF"))
+					std::string ens_label = std::string("DWF"),
+					std::string ens_id = std::string("UKQCD"),
+					unsigned int sequence_number = 1)
   {
-    writeConfiguration(Umu,file,0,1,ens_label);
+    writeConfiguration(Umu,file,0,1,ens_label,ens_id,sequence_number);
   }
   template<class GaugeStats=PeriodicGaugeStatistics>
   static inline void writeConfiguration(Lattice<vLorentzColourMatrixD > &Umu,
 					std::string file, 
 					int two_row,
 					int bits32,
-					std::string ens_label = std::string("DWF"))
+					std::string ens_label = std::string("DWF"),
+					std::string ens_id = std::string("UKQCD"),
+					unsigned int sequence_number = 1)
   {
     typedef vLorentzColourMatrixD vobj;
     typedef typename vobj::scalar_object sobj;
 
     FieldMetaData header;
-    ///////////////////////////////////////////
-    // Following should become arguments
-    ///////////////////////////////////////////
-    header.sequence_number = 1;
-    header.ensemble_id     = std::string("UKQCD");
+    header.sequence_number = sequence_number;
+    header.ensemble_id     = ens_id;
     header.ensemble_label  = ens_label;
+    header.hdr_version     = "1.0" ;
 
     typedef LorentzColourMatrixD fobj3D;
     typedef LorentzColour2x3D    fobj2D;
@@ -245,10 +252,14 @@ public:
 
     uint64_t offset;
 
-    // Sod it -- always write 3x3 double
-    header.floating_point = std::string("IEEE64BIG");
-    header.data_type      = std::string("4D_SU3_GAUGE_3x3");
-    GaugeSimpleUnmunger<fobj3D,sobj> munge;
+    // Sod it -- always write NcxNc double
+    header.floating_point  = std::string("IEEE64BIG");
+    const std::string stNC = std::to_string( Nc ) ;
+    if( two_row ) {
+      header.data_type = std::string("4D_SU" + stNC + "_GAUGE" );
+    } else {
+      header.data_type = std::string("4D_SU" + stNC + "_GAUGE_" + stNC + "x" + stNC );
+    }
     if ( grid->IsBoss() ) { 
       truncate(file);
       offset = writeHeader(header,file);
@@ -256,8 +267,15 @@ public:
     grid->Broadcast(0,(void *)&offset,sizeof(offset));
 
     uint32_t nersc_csum,scidac_csuma,scidac_csumb;
-    BinaryIO::writeLatticeObject<vobj,fobj3D>(Umu,file,munge,offset,header.floating_point,
-					      nersc_csum,scidac_csuma,scidac_csumb);
+    if( two_row ) {
+      Gauge3x2unmunger<fobj2D,sobj> munge;
+      BinaryIO::writeLatticeObject<vobj,fobj2D>(Umu,file,munge,offset,header.floating_point,
+						nersc_csum,scidac_csuma,scidac_csumb);
+    } else {
+      GaugeSimpleUnmunger<fobj3D,sobj> munge;
+      BinaryIO::writeLatticeObject<vobj,fobj3D>(Umu,file,munge,offset,header.floating_point,
+						nersc_csum,scidac_csuma,scidac_csumb);
+    }
     header.checksum = nersc_csum;
     if ( grid->IsBoss() ) { 
       writeHeader(header,file);
@@ -289,8 +307,7 @@ public:
     header.plaquette=0.0;
     MachineCharacteristics(header);
 
-	uint64_t offset;
-  
+    uint64_t offset;
 #ifdef RNG_RANLUX
     header.floating_point = std::string("UINT64");
     header.data_type      = std::string("RANLUX48");
@@ -330,7 +347,7 @@ public:
 
     GridBase *grid = parallel.Grid();
 
-	uint64_t offset = readHeader(file,grid,header);
+    uint64_t offset = readHeader(file,grid,header);
 
     FieldMetaData clone(header);
 

Grid/perfmon/PerfCount.h

@@ -72,17 +72,9 @@ static long perf_event_open(struct perf_event_attr *hw_event, pid_t pid,
 inline uint64_t cyclecount(void){ 
   return 0;
 }
-#define __SSC_MARK(mark) __asm__ __volatile__ ("movl %0, %%ebx; .byte 0x64, 0x67, 0x90 " ::"i"(mark):"%ebx")
-#define __SSC_STOP  __SSC_MARK(0x110)
-#define __SSC_START __SSC_MARK(0x111)
-
 
 #else
 
-#define __SSC_MARK(mark) 
-#define __SSC_STOP  
-#define __SSC_START 
-
 /*
  * cycle counters arch dependent
  */

Grid/perfmon/Timer.h

@@ -39,9 +39,9 @@ NAMESPACE_BEGIN(Grid)
 // C++11 time facilities better?
 inline double usecond(void) {
   struct timeval tv;
-#ifdef TIMERS_ON
+  tv.tv_sec = 0;
+  tv.tv_usec = 0;
   gettimeofday(&tv,NULL);
-#endif
   return 1.0*tv.tv_usec + 1.0e6*tv.tv_sec;
 }
 

Grid/pugixml/pugixml.cc

@@ -16,8 +16,12 @@
 
 #ifdef __NVCC__
 #pragma push
+#if (__CUDACC_VER_MAJOR__ >= 11) && (__CUDACC_VER_MINOR__ >= 5)
+#pragma nv_diag_suppress declared_but_not_referenced // suppress "function was declared but never referenced warning"
+#else
 #pragma diag_suppress declared_but_not_referenced // suppress "function was declared but never referenced warning"
 #endif
+#endif
 
 #include "pugixml.h"
 

Grid/qcd/action/ActionParams.h

@@ -38,24 +38,32 @@ NAMESPACE_BEGIN(Grid);
 struct GparityWilsonImplParams {
   Coordinate twists;
                      //mu=Nd-1 is assumed to be the time direction and a twist value of 1 indicates antiperiodic BCs
-  GparityWilsonImplParams() : twists(Nd, 0) {};
+  Coordinate dirichlet; // Blocksize of dirichlet BCs
+  GparityWilsonImplParams() : twists(Nd, 0), dirichlet(Nd, 0) {};
 };
   
 struct WilsonImplParams {
   bool overlapCommsCompute;
+  Coordinate dirichlet; // Blocksize of dirichlet BCs
   AcceleratorVector<Real,Nd> twist_n_2pi_L;
   AcceleratorVector<Complex,Nd> boundary_phases;
   WilsonImplParams()  {
+    dirichlet.resize(Nd,0);
     boundary_phases.resize(Nd, 1.0);
       twist_n_2pi_L.resize(Nd, 0.0);
   };
   WilsonImplParams(const AcceleratorVector<Complex,Nd> phi) : boundary_phases(phi), overlapCommsCompute(false) {
     twist_n_2pi_L.resize(Nd, 0.0);
+    dirichlet.resize(Nd,0);
   }
 };
 
 struct StaggeredImplParams {
-  StaggeredImplParams()  {};
+  Coordinate dirichlet; // Blocksize of dirichlet BCs
+  StaggeredImplParams()
+  {
+    dirichlet.resize(Nd,0);
+  };
 };
   
   struct OneFlavourRationalParams : Serializable {

Grid/qcd/action/fermion/CayleyFermion5D.h

@@ -68,9 +68,17 @@ public:
   ///////////////////////////////////////////////////////////////
   // Support for MADWF tricks
   ///////////////////////////////////////////////////////////////
-  virtual RealD Mass(void) { return mass; };
+  RealD Mass(void) { return (mass_plus + mass_minus) / 2.0; };
+  RealD MassPlus(void) { return mass_plus; };
+  RealD MassMinus(void) { return mass_minus; };
+
   void  SetMass(RealD _mass) { 
-    mass=_mass; 
+    mass_plus=mass_minus=_mass; 
+    SetCoefficientsInternal(_zolo_hi,_gamma,_b,_c);  // Reset coeffs
+  } ;
+  void  SetMass(RealD _mass_plus, RealD _mass_minus) { 
+    mass_plus=_mass_plus;
+    mass_minus=_mass_minus;
     SetCoefficientsInternal(_zolo_hi,_gamma,_b,_c);  // Reset coeffs
   } ;
   void  P(const FermionField &psi, FermionField &chi);
@@ -108,7 +116,7 @@ public:
   void   MeooeDag5D    (const FermionField &in, FermionField &out);
 
   //    protected:
-  RealD mass;
+  RealD mass_plus, mass_minus;
 
   // Save arguments to SetCoefficientsInternal
   Vector<Coeff_t> _gamma;

Grid/qcd/action/fermion/CloverHelpers.h

@@ -0,0 +1,435 @@
+/*************************************************************************************
+
+    Grid physics library, www.github.com/paboyle/Grid
+
+    Source file: ./lib/qcd/action/fermion/WilsonCloverFermionImplementation.h
+
+    Copyright (C) 2017 - 2022
+
+    Author: paboyle <paboyle@ph.ed.ac.uk>
+    Author: Daniel Richtmann <daniel.richtmann@gmail.com>
+    Author: Mattia Bruno <mattia.bruno@cern.ch>
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License along
+    with this program; if not, write to the Free Software Foundation, Inc.,
+    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+    See the full license in the file "LICENSE" in the top level distribution directory
+    *************************************************************************************/
+/*  END LEGAL */
+
+#pragma once
+
+#include <Grid/Grid.h>
+#include <Grid/qcd/spin/Dirac.h>
+#include <Grid/qcd/action/fermion/WilsonCloverHelpers.h>
+
+////////////////////////////////////////////
+// Standard Clover
+//   (4+m0) + csw * clover_term
+// Exp Clover
+//   (4+m0) * exp(csw/(4+m0) clover_term)
+//   = (4+m0) + csw * clover_term + ...
+////////////////////////////////////////////
+
+NAMESPACE_BEGIN(Grid);
+
+
+//////////////////////////////////
+// Generic Standard Clover
+//////////////////////////////////
+
+template<class Impl>
+class CloverHelpers: public WilsonCloverHelpers<Impl> {
+public:
+
+  INHERIT_IMPL_TYPES(Impl);
+  INHERIT_CLOVER_TYPES(Impl);
+
+  typedef WilsonCloverHelpers<Impl> Helpers;
+
+  static void Instantiate(CloverField& CloverTerm, CloverField& CloverTermInv, RealD csw_t, RealD diag_mass) {
+    GridBase *grid = CloverTerm.Grid();
+    CloverTerm += diag_mass;
+
+    int lvol = grid->lSites();
+    int DimRep = Impl::Dimension;
+    {
+      autoView(CTv,CloverTerm,CpuRead);
+      autoView(CTIv,CloverTermInv,CpuWrite);
+      thread_for(site, lvol, {
+        Coordinate lcoor;
+        grid->LocalIndexToLocalCoor(site, lcoor);
+        Eigen::MatrixXcd EigenCloverOp = Eigen::MatrixXcd::Zero(Ns * DimRep, Ns * DimRep);
+        Eigen::MatrixXcd EigenInvCloverOp = Eigen::MatrixXcd::Zero(Ns * DimRep, Ns * DimRep);
+        typename SiteClover::scalar_object Qx = Zero(), Qxinv = Zero();
+        peekLocalSite(Qx, CTv, lcoor);
+
+        for (int j = 0; j < Ns; j++)
+          for (int k = 0; k < Ns; k++)
+            for (int a = 0; a < DimRep; a++)
+              for (int b = 0; b < DimRep; b++){
+                auto zz =  Qx()(j, k)(a, b);
+                EigenCloverOp(a + j * DimRep, b + k * DimRep) = std::complex<double>(zz);
+              }
+
+        EigenInvCloverOp = EigenCloverOp.inverse();
+        for (int j = 0; j < Ns; j++)
+          for (int k = 0; k < Ns; k++)
+            for (int a = 0; a < DimRep; a++)
+              for (int b = 0; b < DimRep; b++)
+                Qxinv()(j, k)(a, b) = EigenInvCloverOp(a + j * DimRep, b + k * DimRep);
+               pokeLocalSite(Qxinv, CTIv, lcoor);
+      });
+    }
+  }
+
+  static GaugeLinkField Cmunu(std::vector<GaugeLinkField> &U, GaugeLinkField &lambda, int mu, int nu) {
+    return Helpers::Cmunu(U, lambda, mu, nu);
+  }
+
+};
+
+
+//////////////////////////////////
+// Generic Exp Clover
+//////////////////////////////////
+
+template<class Impl>
+class ExpCloverHelpers: public WilsonCloverHelpers<Impl> {
+public:
+
+  INHERIT_IMPL_TYPES(Impl);
+  INHERIT_CLOVER_TYPES(Impl);
+
+  template <typename vtype> using iImplClover = iScalar<iMatrix<iMatrix<vtype, Impl::Dimension>, Ns>>;
+  typedef WilsonCloverHelpers<Impl> Helpers;
+
+  // Can this be avoided?
+  static void IdentityTimesC(const CloverField& in, RealD c) {
+    int DimRep = Impl::Dimension;
+
+    autoView(in_v, in, AcceleratorWrite);
+
+    accelerator_for(ss, in.Grid()->oSites(), 1, {
+      for (int sa=0; sa<Ns; sa++)
+        for (int ca=0; ca<DimRep; ca++)
+          in_v[ss]()(sa,sa)(ca,ca) = c;
+    });
+  }
+
+  static int getNMAX(RealD prec, RealD R) {
+    /* compute stop condition for exponential */
+    int NMAX=1;
+    RealD cond=R*R/2.;
+
+    while (cond*std::exp(R)>prec) {
+      NMAX++;
+      cond*=R/(double)(NMAX+1);
+    }
+    return NMAX;
+  }
+
+  static int getNMAX(Lattice<iImplClover<vComplexD>> &t, RealD R) {return getNMAX(1e-12,R);}
+  static int getNMAX(Lattice<iImplClover<vComplexF>> &t, RealD R) {return getNMAX(1e-6,R);}
+
+  static void Instantiate(CloverField& Clover, CloverField& CloverInv, RealD csw_t, RealD diag_mass) {
+    GridBase* grid = Clover.Grid();
+    CloverField ExpClover(grid);
+
+    int NMAX = getNMAX(Clover, 3.*csw_t/diag_mass);
+
+    Clover *= (1.0/diag_mass);
+
+    // Taylor expansion, slow but generic
+    // Horner scheme: a0 + a1 x + a2 x^2 + .. = a0 + x (a1 + x(...))
+    // qN = cN
+    // qn = cn + qn+1 X
+    std::vector<RealD> cn(NMAX+1);
+    cn[0] = 1.0;
+    for (int i=1; i<=NMAX; i++)
+      cn[i] = cn[i-1] / RealD(i);
+
+    ExpClover = Zero();
+    IdentityTimesC(ExpClover, cn[NMAX]);
+    for (int i=NMAX-1; i>=0; i--)
+      ExpClover = ExpClover * Clover + cn[i];
+
+    // prepare inverse
+    CloverInv = (-1.0)*Clover;
+
+    Clover = ExpClover * diag_mass;
+
+    ExpClover = Zero();
+    IdentityTimesC(ExpClover, cn[NMAX]);
+    for (int i=NMAX-1; i>=0; i--)
+      ExpClover = ExpClover * CloverInv + cn[i];
+
+    CloverInv = ExpClover * (1.0/diag_mass);
+
+  }
+
+  static GaugeLinkField Cmunu(std::vector<GaugeLinkField> &U, GaugeLinkField &lambda, int mu, int nu) {
+    assert(0);
+    return lambda;
+  }
+
+};
+
+
+//////////////////////////////////
+// Compact Standard Clover
+//////////////////////////////////
+
+
+template<class Impl>
+class CompactCloverHelpers: public CompactWilsonCloverHelpers<Impl>,
+                            public WilsonCloverHelpers<Impl> {
+public:
+
+  INHERIT_IMPL_TYPES(Impl);
+  INHERIT_CLOVER_TYPES(Impl);
+  INHERIT_COMPACT_CLOVER_TYPES(Impl);
+
+  typedef WilsonCloverHelpers<Impl> Helpers;
+  typedef CompactWilsonCloverHelpers<Impl> CompactHelpers;
+
+  static void MassTerm(CloverField& Clover, RealD diag_mass) {
+    Clover += diag_mass;
+  }
+
+  static void Exponentiate_Clover(CloverDiagonalField& Diagonal,
+                          CloverTriangleField& Triangle,
+                          RealD csw_t, RealD diag_mass) {
+
+    // Do nothing
+  }
+
+  // TODO: implement Cmunu for better performances with compact layout, but don't do it
+  // here, but rather in WilsonCloverHelpers.h -> CompactWilsonCloverHelpers
+  static GaugeLinkField Cmunu(std::vector<GaugeLinkField> &U, GaugeLinkField &lambda, int mu, int nu) {
+    return Helpers::Cmunu(U, lambda, mu, nu);
+  }
+};
+
+//////////////////////////////////
+// Compact Exp Clover
+//////////////////////////////////
+
+template<class Impl>
+class CompactExpCloverHelpers: public CompactWilsonCloverHelpers<Impl> {
+public:
+
+  INHERIT_IMPL_TYPES(Impl);
+  INHERIT_CLOVER_TYPES(Impl);
+  INHERIT_COMPACT_CLOVER_TYPES(Impl);
+
+  template <typename vtype> using iImplClover = iScalar<iMatrix<iMatrix<vtype, Impl::Dimension>, Ns>>;
+  typedef CompactWilsonCloverHelpers<Impl> CompactHelpers;
+
+  static void MassTerm(CloverField& Clover, RealD diag_mass) {
+    // do nothing!
+    // mass term is multiplied to exp(Clover) below
+  }
+
+  static int getNMAX(RealD prec, RealD R) {
+    /* compute stop condition for exponential */
+    int NMAX=1;
+    RealD cond=R*R/2.;
+
+    while (cond*std::exp(R)>prec) {
+      NMAX++;
+      cond*=R/(double)(NMAX+1);
+    }
+    return NMAX;
+  }
+
+  static int getNMAX(Lattice<iImplCloverDiagonal<vComplexD>> &t, RealD R) {return getNMAX(1e-12,R);}
+  static int getNMAX(Lattice<iImplCloverDiagonal<vComplexF>> &t, RealD R) {return getNMAX(1e-6,R);}
+
+  static void ExponentiateHermitean6by6(const iMatrix<ComplexD,6> &arg, const RealD& alpha, const std::vector<RealD>& cN, const int Niter, iMatrix<ComplexD,6>& dest){
+
+  	  typedef iMatrix<ComplexD,6> mat;
+
+  	  RealD qn[6];
+  	  RealD qnold[6];
+  	  RealD p[5];
+  	  RealD trA2, trA3, trA4;
+
+  	  mat A2, A3, A4, A5;
+  	  A2 = alpha * alpha * arg * arg;
+  	  A3 = alpha * arg * A2;
+  	  A4 = A2 * A2;
+  	  A5 = A2 * A3;
+
+  	  trA2 = toReal( trace(A2) );
+  	  trA3 = toReal( trace(A3) );
+  	  trA4 = toReal( trace(A4));
+
+  	  p[0] = toReal( trace(A3 * A3)) / 6.0 - 0.125 * trA4 * trA2 - trA3 * trA3 / 18.0 + trA2 * trA2 * trA2/ 48.0;
+  	  p[1] = toReal( trace(A5)) / 5.0 - trA3 * trA2 / 6.0;
+  	  p[2] = toReal( trace(A4)) / 4.0 - 0.125 * trA2 * trA2;
+  	  p[3] = trA3 / 3.0;
+  	  p[4] = 0.5 * trA2;
+
+  	  qnold[0] = cN[Niter];
+  	  qnold[1] = 0.0;
+  	  qnold[2] = 0.0;
+  	  qnold[3] = 0.0;
+  	  qnold[4] = 0.0;
+  	  qnold[5] = 0.0;
+
+  	  for(int i = Niter-1; i >= 0; i--)
+  	  {
+  	   qn[0] = p[0] * qnold[5] + cN[i];
+  	   qn[1] = p[1] * qnold[5] + qnold[0];
+  	   qn[2] = p[2] * qnold[5] + qnold[1];
+  	   qn[3] = p[3] * qnold[5] + qnold[2];
+  	   qn[4] = p[4] * qnold[5] + qnold[3];
+  	   qn[5] = qnold[4];
+
+  	   qnold[0] = qn[0];
+  	   qnold[1] = qn[1];
+  	   qnold[2] = qn[2];
+  	   qnold[3] = qn[3];
+  	   qnold[4] = qn[4];
+  	   qnold[5] = qn[5];
+  	  }
+
+  	  mat unit(1.0);
+
+  	  dest = (qn[0] * unit + qn[1] * alpha * arg + qn[2] * A2 + qn[3] * A3 + qn[4] * A4 + qn[5] * A5);
+
+    }
+
+  static void Exponentiate_Clover(CloverDiagonalField& Diagonal, CloverTriangleField& Triangle, RealD csw_t, RealD diag_mass) {
+
+    GridBase* grid = Diagonal.Grid();
+    int NMAX = getNMAX(Diagonal, 3.*csw_t/diag_mass);
+
+    //
+    // Implementation completely in Daniel's layout
+    //
+
+    // Taylor expansion with Cayley-Hamilton recursion
+    // underlying Horner scheme as above
+    std::vector<RealD> cn(NMAX+1);
+    cn[0] = 1.0;
+    for (int i=1; i<=NMAX; i++){
+      cn[i] = cn[i-1] / RealD(i);
+    }
+
+      // Taken over from Daniel's implementation
+      conformable(Diagonal, Triangle);
+
+      long lsites = grid->lSites();
+    {
+      typedef typename SiteCloverDiagonal::scalar_object scalar_object_diagonal;
+      typedef typename SiteCloverTriangle::scalar_object scalar_object_triangle;
+      typedef iMatrix<ComplexD,6> mat;
+
+      autoView(diagonal_v,  Diagonal,  CpuRead);
+      autoView(triangle_v,  Triangle,  CpuRead);
+      autoView(diagonalExp_v, Diagonal, CpuWrite);
+      autoView(triangleExp_v, Triangle, CpuWrite);
+
+      thread_for(site, lsites, { // NOTE: Not on GPU because of (peek/poke)LocalSite
+
+    	  mat srcCloverOpUL(0.0); // upper left block
+    	  mat srcCloverOpLR(0.0); // lower right block
+    	  mat ExpCloverOp;
+
+        scalar_object_diagonal diagonal_tmp     = Zero();
+        scalar_object_diagonal diagonal_exp_tmp = Zero();
+        scalar_object_triangle triangle_tmp     = Zero();
+        scalar_object_triangle triangle_exp_tmp = Zero();
+
+        Coordinate lcoor;
+        grid->LocalIndexToLocalCoor(site, lcoor);
+
+        peekLocalSite(diagonal_tmp, diagonal_v, lcoor);
+        peekLocalSite(triangle_tmp, triangle_v, lcoor);
+
+        int block;
+        block = 0;
+        for(int i = 0; i < 6; i++){
+        	for(int j = 0; j < 6; j++){
+        		if (i == j){
+        			srcCloverOpUL(i,j) = static_cast<ComplexD>(TensorRemove(diagonal_tmp()(block)(i)));
+        		}
+        		else{
+        			srcCloverOpUL(i,j) = static_cast<ComplexD>(TensorRemove(CompactHelpers::triangle_elem(triangle_tmp, block, i, j)));
+        		}
+        	}
+        }
+        block = 1;
+        for(int i = 0; i < 6; i++){
+          	for(int j = 0; j < 6; j++){
+           		if (i == j){
+           			srcCloverOpLR(i,j) = static_cast<ComplexD>(TensorRemove(diagonal_tmp()(block)(i)));
+           		}
+           		else{
+           			srcCloverOpLR(i,j) = static_cast<ComplexD>(TensorRemove(CompactHelpers::triangle_elem(triangle_tmp, block, i, j)));
+           		}
+            }
+        }
+
+        // exp(Clover)
+
+        ExponentiateHermitean6by6(srcCloverOpUL,1.0/diag_mass,cn,NMAX,ExpCloverOp);
+
+        block = 0;
+        for(int i = 0; i < 6; i++){
+        	for(int j = 0; j < 6; j++){
+            	if (i == j){
+            		diagonal_exp_tmp()(block)(i) = ExpCloverOp(i,j);
+            	}
+            	else if(i < j){
+            		triangle_exp_tmp()(block)(CompactHelpers::triangle_index(i, j)) = ExpCloverOp(i,j);
+            	}
+           	}
+        }
+
+        ExponentiateHermitean6by6(srcCloverOpLR,1.0/diag_mass,cn,NMAX,ExpCloverOp);
+
+        block = 1;
+        for(int i = 0; i < 6; i++){
+        	for(int j = 0; j < 6; j++){
+              	if (i == j){
+              		diagonal_exp_tmp()(block)(i) = ExpCloverOp(i,j);
+               	}
+               	else if(i < j){
+               		triangle_exp_tmp()(block)(CompactHelpers::triangle_index(i, j)) = ExpCloverOp(i,j);
+               	}
+            }
+        }
+
+        pokeLocalSite(diagonal_exp_tmp, diagonalExp_v, lcoor);
+        pokeLocalSite(triangle_exp_tmp, triangleExp_v, lcoor);
+      });
+    }
+
+    Diagonal *= diag_mass;
+    Triangle *= diag_mass;
+  }
+
+
+  static GaugeLinkField Cmunu(std::vector<GaugeLinkField> &U, GaugeLinkField &lambda, int mu, int nu) {
+    assert(0);
+    return lambda;
+  }
+
+};
+
+
+NAMESPACE_END(Grid);

Grid/qcd/action/fermion/CompactWilsonCloverFermion.h

@@ -31,6 +31,7 @@
 
 #include <Grid/qcd/action/fermion/WilsonCloverTypes.h>
 #include <Grid/qcd/action/fermion/WilsonCloverHelpers.h>
+#include <Grid/qcd/action/fermion/CloverHelpers.h>
 
 NAMESPACE_BEGIN(Grid);
 
@@ -85,7 +86,7 @@ NAMESPACE_BEGIN(Grid);
 //           + (2 * 1 + 4 * 1/2) triangle parts = 4 triangle parts =  60 complex words per site
 //                                                                 =  84 complex words per site
 
-template<class Impl>
+template<class Impl, class CloverHelpers>
 class CompactWilsonCloverFermion : public WilsonFermion<Impl>,
                                    public WilsonCloverHelpers<Impl>,
                                    public CompactWilsonCloverHelpers<Impl> {

Grid/qcd/action/fermion/Fermion.h

@@ -138,38 +138,52 @@ typedef WilsonTMFermion<WilsonImplF> WilsonTMFermionF;
 typedef WilsonTMFermion<WilsonImplD> WilsonTMFermionD;
 
 // Clover fermions
-typedef WilsonCloverFermion<WilsonImplR> WilsonCloverFermionR;
-typedef WilsonCloverFermion<WilsonImplF> WilsonCloverFermionF;
-typedef WilsonCloverFermion<WilsonImplD> WilsonCloverFermionD;
+template <typename WImpl> using WilsonClover = WilsonCloverFermion<WImpl, CloverHelpers<WImpl>>;
+template <typename WImpl> using WilsonExpClover = WilsonCloverFermion<WImpl, ExpCloverHelpers<WImpl>>;
 
-typedef WilsonCloverFermion<WilsonAdjImplR> WilsonCloverAdjFermionR;
-typedef WilsonCloverFermion<WilsonAdjImplF> WilsonCloverAdjFermionF;
-typedef WilsonCloverFermion<WilsonAdjImplD> WilsonCloverAdjFermionD;
+typedef WilsonClover<WilsonImplR> WilsonCloverFermionR;
+typedef WilsonClover<WilsonImplF> WilsonCloverFermionF;
+typedef WilsonClover<WilsonImplD> WilsonCloverFermionD;
 
-typedef WilsonCloverFermion<WilsonTwoIndexSymmetricImplR> WilsonCloverTwoIndexSymmetricFermionR;
-typedef WilsonCloverFermion<WilsonTwoIndexSymmetricImplF> WilsonCloverTwoIndexSymmetricFermionF;
-typedef WilsonCloverFermion<WilsonTwoIndexSymmetricImplD> WilsonCloverTwoIndexSymmetricFermionD;
+typedef WilsonExpClover<WilsonImplR> WilsonExpCloverFermionR;
+typedef WilsonExpClover<WilsonImplF> WilsonExpCloverFermionF;
+typedef WilsonExpClover<WilsonImplD> WilsonExpCloverFermionD;
 
-typedef WilsonCloverFermion<WilsonTwoIndexAntiSymmetricImplR> WilsonCloverTwoIndexAntiSymmetricFermionR;
-typedef WilsonCloverFermion<WilsonTwoIndexAntiSymmetricImplF> WilsonCloverTwoIndexAntiSymmetricFermionF;
-typedef WilsonCloverFermion<WilsonTwoIndexAntiSymmetricImplD> WilsonCloverTwoIndexAntiSymmetricFermionD;
+typedef WilsonClover<WilsonAdjImplR> WilsonCloverAdjFermionR;
+typedef WilsonClover<WilsonAdjImplF> WilsonCloverAdjFermionF;
+typedef WilsonClover<WilsonAdjImplD> WilsonCloverAdjFermionD;
+
+typedef WilsonClover<WilsonTwoIndexSymmetricImplR> WilsonCloverTwoIndexSymmetricFermionR;
+typedef WilsonClover<WilsonTwoIndexSymmetricImplF> WilsonCloverTwoIndexSymmetricFermionF;
+typedef WilsonClover<WilsonTwoIndexSymmetricImplD> WilsonCloverTwoIndexSymmetricFermionD;
+
+typedef WilsonClover<WilsonTwoIndexAntiSymmetricImplR> WilsonCloverTwoIndexAntiSymmetricFermionR;
+typedef WilsonClover<WilsonTwoIndexAntiSymmetricImplF> WilsonCloverTwoIndexAntiSymmetricFermionF;
+typedef WilsonClover<WilsonTwoIndexAntiSymmetricImplD> WilsonCloverTwoIndexAntiSymmetricFermionD;
 
 // Compact Clover fermions
-typedef CompactWilsonCloverFermion<WilsonImplR> CompactWilsonCloverFermionR;
-typedef CompactWilsonCloverFermion<WilsonImplF> CompactWilsonCloverFermionF;
-typedef CompactWilsonCloverFermion<WilsonImplD> CompactWilsonCloverFermionD;
+template <typename WImpl> using CompactWilsonClover = CompactWilsonCloverFermion<WImpl, CompactCloverHelpers<WImpl>>;
+template <typename WImpl> using CompactWilsonExpClover = CompactWilsonCloverFermion<WImpl, CompactExpCloverHelpers<WImpl>>;
 
-typedef CompactWilsonCloverFermion<WilsonAdjImplR> CompactWilsonCloverAdjFermionR;
-typedef CompactWilsonCloverFermion<WilsonAdjImplF> CompactWilsonCloverAdjFermionF;
-typedef CompactWilsonCloverFermion<WilsonAdjImplD> CompactWilsonCloverAdjFermionD;
+typedef CompactWilsonClover<WilsonImplR> CompactWilsonCloverFermionR;
+typedef CompactWilsonClover<WilsonImplF> CompactWilsonCloverFermionF;
+typedef CompactWilsonClover<WilsonImplD> CompactWilsonCloverFermionD;
 
-typedef CompactWilsonCloverFermion<WilsonTwoIndexSymmetricImplR> CompactWilsonCloverTwoIndexSymmetricFermionR;
-typedef CompactWilsonCloverFermion<WilsonTwoIndexSymmetricImplF> CompactWilsonCloverTwoIndexSymmetricFermionF;
-typedef CompactWilsonCloverFermion<WilsonTwoIndexSymmetricImplD> CompactWilsonCloverTwoIndexSymmetricFermionD;
+typedef CompactWilsonExpClover<WilsonImplR> CompactWilsonExpCloverFermionR;
+typedef CompactWilsonExpClover<WilsonImplF> CompactWilsonExpCloverFermionF;
+typedef CompactWilsonExpClover<WilsonImplD> CompactWilsonExpCloverFermionD;
 
-typedef CompactWilsonCloverFermion<WilsonTwoIndexAntiSymmetricImplR> CompactWilsonCloverTwoIndexAntiSymmetricFermionR;
-typedef CompactWilsonCloverFermion<WilsonTwoIndexAntiSymmetricImplF> CompactWilsonCloverTwoIndexAntiSymmetricFermionF;
-typedef CompactWilsonCloverFermion<WilsonTwoIndexAntiSymmetricImplD> CompactWilsonCloverTwoIndexAntiSymmetricFermionD;
+typedef CompactWilsonClover<WilsonAdjImplR> CompactWilsonCloverAdjFermionR;
+typedef CompactWilsonClover<WilsonAdjImplF> CompactWilsonCloverAdjFermionF;
+typedef CompactWilsonClover<WilsonAdjImplD> CompactWilsonCloverAdjFermionD;
+
+typedef CompactWilsonClover<WilsonTwoIndexSymmetricImplR> CompactWilsonCloverTwoIndexSymmetricFermionR;
+typedef CompactWilsonClover<WilsonTwoIndexSymmetricImplF> CompactWilsonCloverTwoIndexSymmetricFermionF;
+typedef CompactWilsonClover<WilsonTwoIndexSymmetricImplD> CompactWilsonCloverTwoIndexSymmetricFermionD;
+
+typedef CompactWilsonClover<WilsonTwoIndexAntiSymmetricImplR> CompactWilsonCloverTwoIndexAntiSymmetricFermionR;
+typedef CompactWilsonClover<WilsonTwoIndexAntiSymmetricImplF> CompactWilsonCloverTwoIndexAntiSymmetricFermionF;
+typedef CompactWilsonClover<WilsonTwoIndexAntiSymmetricImplD> CompactWilsonCloverTwoIndexAntiSymmetricFermionD;
 
 // Domain Wall fermions
 typedef DomainWallFermion<WilsonImplR> DomainWallFermionR;

Grid/qcd/action/fermion/FermionOperator.h

@@ -49,7 +49,7 @@ public:
 
   virtual FermionField &tmp(void) = 0;
 
-  virtual void DirichletBlock(Coordinate & _Block) { assert(0); };
+  virtual void DirichletBlock(const Coordinate & _Block) { assert(0); };
   
   GridBase * Grid(void)   { return FermionGrid(); };   // this is all the linalg routines need to know
   GridBase * RedBlackGrid(void) { return FermionRedBlackGrid(); };

Grid/qcd/action/fermion/WilsonCloverFermion.h

@@ -32,6 +32,7 @@
 
 #include <Grid/qcd/action/fermion/WilsonCloverTypes.h>
 #include <Grid/qcd/action/fermion/WilsonCloverHelpers.h>
+#include <Grid/qcd/action/fermion/CloverHelpers.h>
 
 NAMESPACE_BEGIN(Grid);
 
@@ -51,7 +52,7 @@ NAMESPACE_BEGIN(Grid);
 // csw_r = csw_t to recover the isotropic version
 //////////////////////////////////////////////////////////////////
 
-template <class Impl>
+template<class Impl, class CloverHelpers>
 class WilsonCloverFermion : public WilsonFermion<Impl>,
                             public WilsonCloverHelpers<Impl>
 {

Grid/qcd/action/fermion/WilsonCloverHelpers.h

@@ -209,6 +209,8 @@ public:
 };
 
 
+////////////////////////////////////////////////////////
+
 template<class Impl> class CompactWilsonCloverHelpers {
 public:
 

Grid/qcd/action/fermion/WilsonCloverTypes.h

@@ -47,8 +47,6 @@ class CompactWilsonCloverTypes {
 public:
   INHERIT_IMPL_TYPES(Impl);
 
-  static_assert(Nd == 4 && Nc == 3 && Ns == 4 && Impl::Dimension == 3, "Wrong dimensions");
-
   static constexpr int Nred      = Nc * Nhs;        // 6
   static constexpr int Nblock    = Nhs;             // 2
   static constexpr int Ndiagonal = Nred;            // 6

Grid/qcd/action/fermion/WilsonCompressor.h

@@ -297,7 +297,7 @@ public:
   void ZeroCountersi(void)  {  }
   void Reporti(int calls)  {  }
 
-  std::vector<int> surface_list;
+  //  Vector<int> surface_list;
 
   WilsonStencil(GridBase *grid,
 		int npoints,
@@ -307,10 +307,11 @@ public:
     : CartesianStencil<vobj,cobj,Parameters> (grid,npoints,checkerboard,directions,distances,p) 
   { 
     ZeroCountersi();
-    surface_list.resize(0);
+    //    surface_list.resize(0);
     this->same_node.resize(npoints);
   };
 
+  /*
   void BuildSurfaceList(int Ls,int vol4){
 
     // find same node for SHM
@@ -331,7 +332,8 @@ public:
       }
     }
   }
-
+  */
+  
   template < class compressor>
   void HaloExchangeOpt(const Lattice<vobj> &source,compressor &compress) 
   {

Grid/qcd/action/fermion/WilsonFermion5D.h

@@ -178,16 +178,8 @@ public:
 		  GridRedBlackCartesian &FourDimRedBlackGrid,
 		  double _M5,const ImplParams &p= ImplParams());
 
-  virtual void DirichletBlock(Coordinate & block)
+  virtual void DirichletBlock(const Coordinate & block)
   {
-    assert(block.size()==Nd+1);
-    if ( block[0] || block[1] || block[2] || block[3] || block[4] ){
-      Dirichlet = 1;
-      Block = block;
-      Stencil.DirichletBlock(block); 
-      StencilEven.DirichletBlock(block); 
-      StencilOdd.DirichletBlock(block);
-    }
   }
   // Constructors
   /*

Grid/qcd/action/fermion/implementation/CayleyFermion5DImplementation.h

@@ -47,7 +47,7 @@ CayleyFermion5D<Impl>::CayleyFermion5D(GaugeField &_Umu,
 			FiveDimRedBlackGrid,
 			FourDimGrid,
 			FourDimRedBlackGrid,_M5,p),
-  mass(_mass)
+  mass_plus(_mass), mass_minus(_mass)
 { 
 }
 
@@ -209,8 +209,8 @@ void CayleyFermion5D<Impl>::M5D   (const FermionField &psi, FermionField &chi)
 {
   int Ls=this->Ls;
   Vector<Coeff_t> diag (Ls,1.0);
-  Vector<Coeff_t> upper(Ls,-1.0); upper[Ls-1]=mass;
-  Vector<Coeff_t> lower(Ls,-1.0); lower[0]   =mass;
+  Vector<Coeff_t> upper(Ls,-1.0); upper[Ls-1]=mass_minus;
+  Vector<Coeff_t> lower(Ls,-1.0); lower[0]   =mass_plus;
   M5D(psi,chi,chi,lower,diag,upper);
 }
 template<class Impl>
@@ -220,8 +220,8 @@ void CayleyFermion5D<Impl>::Meooe5D    (const FermionField &psi, FermionField &D
   Vector<Coeff_t> diag = bs;
   Vector<Coeff_t> upper= cs;
   Vector<Coeff_t> lower= cs; 
-  upper[Ls-1]=-mass*upper[Ls-1];
-  lower[0]   =-mass*lower[0];
+  upper[Ls-1]=-mass_minus*upper[Ls-1];
+  lower[0]   =-mass_plus*lower[0];
   M5D(psi,psi,Din,lower,diag,upper);
 }
 // FIXME Redunant with the above routine; check this and eliminate
@@ -235,8 +235,8 @@ template<class Impl> void CayleyFermion5D<Impl>::Meo5D     (const FermionField &
     upper[i]=-ceo[i];
     lower[i]=-ceo[i];
   }
-  upper[Ls-1]=-mass*upper[Ls-1];
-  lower[0]   =-mass*lower[0];
+  upper[Ls-1]=-mass_minus*upper[Ls-1];
+  lower[0]   =-mass_plus*lower[0];
   M5D(psi,psi,chi,lower,diag,upper);
 }
 template<class Impl>
@@ -250,8 +250,8 @@ void CayleyFermion5D<Impl>::Mooee       (const FermionField &psi, FermionField &
     upper[i]=-cee[i];
     lower[i]=-cee[i];
   }
-  upper[Ls-1]=-mass*upper[Ls-1];
-  lower[0]   =-mass*lower[0];
+  upper[Ls-1]=-mass_minus*upper[Ls-1];
+  lower[0]   =-mass_plus*lower[0];
   M5D(psi,psi,chi,lower,diag,upper);
 }
 template<class Impl>
@@ -266,9 +266,9 @@ void CayleyFermion5D<Impl>::MooeeDag    (const FermionField &psi, FermionField &
     // Assemble the 5d matrix
     if ( s==0 ) {
       upper[s] = -cee[s+1] ;
-      lower[s] = mass*cee[Ls-1];
+      lower[s] = mass_minus*cee[Ls-1];
     } else if ( s==(Ls-1)) { 
-      upper[s] = mass*cee[0];
+      upper[s] = mass_plus*cee[0];
       lower[s] = -cee[s-1];
     } else {
       upper[s]=-cee[s+1];
@@ -291,8 +291,8 @@ void CayleyFermion5D<Impl>::M5Ddag (const FermionField &psi, FermionField &chi)
   Vector<Coeff_t> diag(Ls,1.0);
   Vector<Coeff_t> upper(Ls,-1.0);
   Vector<Coeff_t> lower(Ls,-1.0);
-  upper[Ls-1]=-mass*upper[Ls-1];
-  lower[0]   =-mass*lower[0];
+  upper[Ls-1]=-mass_plus*upper[Ls-1];
+  lower[0]   =-mass_minus*lower[0];
   M5Ddag(psi,chi,chi,lower,diag,upper);
 }
 
@@ -307,9 +307,9 @@ void CayleyFermion5D<Impl>::MeooeDag5D    (const FermionField &psi, FermionField
   for (int s=0;s<Ls;s++){
     if ( s== 0 ) {
       upper[s] = cs[s+1];
-      lower[s] =-mass*cs[Ls-1];
+      lower[s] =-mass_minus*cs[Ls-1];
     } else if ( s==(Ls-1) ) { 
-      upper[s] =-mass*cs[0];
+      upper[s] =-mass_plus*cs[0];
       lower[s] = cs[s-1];
     } else { 
       upper[s] = cs[s+1];
@@ -552,7 +552,7 @@ void CayleyFermion5D<Impl>::SetCoefficientsInternal(RealD zolo_hi,Vector<Coeff_t
       
       lee[i] =-cee[i+1]/bee[i]; // sub-diag entry on the ith column
       
-      leem[i]=mass*cee[Ls-1]/bee[0];
+      leem[i]=mass_minus*cee[Ls-1]/bee[0];
       for(int j=0;j<i;j++) {
 	assert(bee[j+1]!=Coeff_t(0.0));
 	leem[i]*= aee[j]/bee[j+1];
@@ -560,7 +560,7 @@ void CayleyFermion5D<Impl>::SetCoefficientsInternal(RealD zolo_hi,Vector<Coeff_t
       
       uee[i] =-aee[i]/bee[i];   // up-diag entry on the ith row
       
-      ueem[i]=mass;
+      ueem[i]=mass_plus;
       for(int j=1;j<=i;j++) ueem[i]*= cee[j]/bee[j];
       ueem[i]*= aee[0]/bee[0];
       
@@ -573,7 +573,7 @@ void CayleyFermion5D<Impl>::SetCoefficientsInternal(RealD zolo_hi,Vector<Coeff_t
   }
 	
   { 
-    Coeff_t delta_d=mass*cee[Ls-1];
+    Coeff_t delta_d=mass_minus*cee[Ls-1];
     for(int j=0;j<Ls-1;j++) {
       assert(bee[j] != Coeff_t(0.0));
       delta_d *= cee[j]/bee[j];
@@ -642,6 +642,10 @@ void CayleyFermion5D<Impl>::ContractConservedCurrent( PropagatorField &q_in_1,
 						      Current curr_type,
 						      unsigned int mu)
 {
+
+  assert(mass_plus == mass_minus);
+  RealD mass = mass_plus;
+  
 #if (!defined(GRID_HIP))
   Gamma::Algebra Gmu [] = {
     Gamma::Algebra::GammaX,
@@ -777,6 +781,8 @@ void CayleyFermion5D<Impl>::SeqConservedCurrent(PropagatorField &q_in,
   assert(mu>=0);
   assert(mu<Nd);
 
+  assert(mass_plus == mass_minus);
+  RealD mass = mass_plus;
 
 #if 0
   int tshift = (mu == Nd-1) ? 1 : 0;

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

@@ -66,18 +66,17 @@ CayleyFermion5D<Impl>::M5D(const FermionField &psi_i,
   M5Dcalls++;
   M5Dtime-=usecond();
 
-  uint64_t nloop = grid->oSites()/Ls;
+  uint64_t nloop = grid->oSites();
   accelerator_for(sss,nloop,Simd::Nsimd(),{
-    uint64_t ss= sss*Ls;
+    uint64_t s = sss%Ls;
+    uint64_t ss= sss-s;
     typedef decltype(coalescedRead(psi[0])) spinor;
     spinor tmp1, tmp2;
-    for(int s=0;s<Ls;s++){
-      uint64_t idx_u = ss+((s+1)%Ls);
-      uint64_t idx_l = ss+((s+Ls-1)%Ls);
-      spProj5m(tmp1,psi(idx_u));
-      spProj5p(tmp2,psi(idx_l));
-      coalescedWrite(chi[ss+s],pdiag[s]*phi(ss+s)+pupper[s]*tmp1+plower[s]*tmp2);
-    }
+    uint64_t idx_u = ss+((s+1)%Ls);
+    uint64_t idx_l = ss+((s+Ls-1)%Ls);
+    spProj5m(tmp1,psi(idx_u));
+    spProj5p(tmp2,psi(idx_l));
+    coalescedWrite(chi[ss+s],pdiag[s]*phi(ss+s)+pupper[s]*tmp1+plower[s]*tmp2);
   });
   M5Dtime+=usecond();
 }
@@ -108,18 +107,17 @@ CayleyFermion5D<Impl>::M5Ddag(const FermionField &psi_i,
   M5Dcalls++;
   M5Dtime-=usecond();
 
-  uint64_t nloop = grid->oSites()/Ls;
+  uint64_t nloop = grid->oSites();
   accelerator_for(sss,nloop,Simd::Nsimd(),{
-    uint64_t ss=sss*Ls;
+    uint64_t s = sss%Ls;
+    uint64_t ss= sss-s;
     typedef decltype(coalescedRead(psi[0])) spinor;
     spinor tmp1,tmp2;
-    for(int s=0;s<Ls;s++){
-      uint64_t idx_u = ss+((s+1)%Ls);
-      uint64_t idx_l = ss+((s+Ls-1)%Ls);
-      spProj5p(tmp1,psi(idx_u));
-      spProj5m(tmp2,psi(idx_l));
-      coalescedWrite(chi[ss+s],pdiag[s]*phi(ss+s)+pupper[s]*tmp1+plower[s]*tmp2);
-    }
+    uint64_t idx_u = ss+((s+1)%Ls);
+    uint64_t idx_l = ss+((s+Ls-1)%Ls);
+    spProj5p(tmp1,psi(idx_u));
+    spProj5m(tmp2,psi(idx_l));
+    coalescedWrite(chi[ss+s],pdiag[s]*phi(ss+s)+pupper[s]*tmp1+plower[s]*tmp2);
   });
   M5Dtime+=usecond();
 }

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

@@ -32,17 +32,18 @@
 #include <Grid/qcd/spin/Dirac.h>
 #include <Grid/qcd/action/fermion/CompactWilsonCloverFermion.h>
 
+
 NAMESPACE_BEGIN(Grid);
-template<class Impl>
-CompactWilsonCloverFermion<Impl>::CompactWilsonCloverFermion(GaugeField& _Umu,
-                                                             GridCartesian& Fgrid,
-                                                             GridRedBlackCartesian& Hgrid,
-                                                             const RealD _mass,
-                                                             const RealD _csw_r,
-                                                             const RealD _csw_t,
-                                                             const RealD _cF,
-                                                             const WilsonAnisotropyCoefficients& clover_anisotropy,
-                                                             const ImplParams& impl_p)
+template<class Impl, class CloverHelpers>
+CompactWilsonCloverFermion<Impl, CloverHelpers>::CompactWilsonCloverFermion(GaugeField& _Umu,
+                                                                            GridCartesian& Fgrid,
+                                                                            GridRedBlackCartesian& Hgrid,
+                                                                            const RealD _mass,
+                                                                            const RealD _csw_r,
+                                                                            const RealD _csw_t,
+                                                                            const RealD _cF,
+                                                                            const WilsonAnisotropyCoefficients& clover_anisotropy,
+                                                                            const ImplParams& impl_p)
   : WilsonBase(_Umu, Fgrid, Hgrid, _mass, impl_p, clover_anisotropy)
   , csw_r(_csw_r)
   , csw_t(_csw_t)
@@ -58,50 +59,55 @@ CompactWilsonCloverFermion<Impl>::CompactWilsonCloverFermion(GaugeField& _Umu,
   , BoundaryMask(&Fgrid)
   , BoundaryMaskEven(&Hgrid), BoundaryMaskOdd(&Hgrid)
 {
+  assert(Nd == 4 && Nc == 3 && Ns == 4 && Impl::Dimension == 3);
+
   csw_r *= 0.5;
   csw_t *= 0.5;
   if (clover_anisotropy.isAnisotropic)
     csw_r /= clover_anisotropy.xi_0;
 
   ImportGauge(_Umu);
-  if (open_boundaries)
+  if (open_boundaries) {
+    this->BoundaryMaskEven.Checkerboard() = Even;
+    this->BoundaryMaskOdd.Checkerboard() = Odd;
     CompactHelpers::SetupMasks(this->BoundaryMask, this->BoundaryMaskEven, this->BoundaryMaskOdd);
+  }
 }
 
-template<class Impl>
-void CompactWilsonCloverFermion<Impl>::Dhop(const FermionField& in, FermionField& out, int dag) {
+template<class Impl, class CloverHelpers>
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::Dhop(const FermionField& in, FermionField& out, int dag) {
   WilsonBase::Dhop(in, out, dag);
   if(open_boundaries) ApplyBoundaryMask(out);
 }
 
-template<class Impl>
-void CompactWilsonCloverFermion<Impl>::DhopOE(const FermionField& in, FermionField& out, int dag) {
+template<class Impl, class CloverHelpers>
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::DhopOE(const FermionField& in, FermionField& out, int dag) {
   WilsonBase::DhopOE(in, out, dag);
   if(open_boundaries) ApplyBoundaryMask(out);
 }
 
-template<class Impl>
-void CompactWilsonCloverFermion<Impl>::DhopEO(const FermionField& in, FermionField& out, int dag) {
+template<class Impl, class CloverHelpers>
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::DhopEO(const FermionField& in, FermionField& out, int dag) {
   WilsonBase::DhopEO(in, out, dag);
   if(open_boundaries) ApplyBoundaryMask(out);
 }
 
-template<class Impl>
-void CompactWilsonCloverFermion<Impl>::DhopDir(const FermionField& in, FermionField& out, int dir, int disp) {
+template<class Impl, class CloverHelpers>
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::DhopDir(const FermionField& in, FermionField& out, int dir, int disp) {
   WilsonBase::DhopDir(in, out, dir, disp);
   if(this->open_boundaries) ApplyBoundaryMask(out);
 }
 
-template<class Impl>
-void CompactWilsonCloverFermion<Impl>::DhopDirAll(const FermionField& in, std::vector<FermionField>& out) {
+template<class Impl, class CloverHelpers>
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::DhopDirAll(const FermionField& in, std::vector<FermionField>& out) {
   WilsonBase::DhopDirAll(in, out);
   if(this->open_boundaries) {
     for(auto& o : out) ApplyBoundaryMask(o);
   }
 }
 
-template<class Impl>
-void CompactWilsonCloverFermion<Impl>::M(const FermionField& in, FermionField& out) {
+template<class Impl, class CloverHelpers>
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::M(const FermionField& in, FermionField& out) {
   out.Checkerboard() = in.Checkerboard();
   WilsonBase::Dhop(in, out, DaggerNo); // call base to save applying bc
   Mooee(in, Tmp);
@@ -109,8 +115,8 @@ void CompactWilsonCloverFermion<Impl>::M(const FermionField& in, FermionField& o
   if(open_boundaries) ApplyBoundaryMask(out);
 }
 
-template<class Impl>
-void CompactWilsonCloverFermion<Impl>::Mdag(const FermionField& in, FermionField& out) {
+template<class Impl, class CloverHelpers>
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::Mdag(const FermionField& in, FermionField& out) {
   out.Checkerboard() = in.Checkerboard();
   WilsonBase::Dhop(in, out, DaggerYes);  // call base to save applying bc
   MooeeDag(in, Tmp);
@@ -118,20 +124,20 @@ void CompactWilsonCloverFermion<Impl>::Mdag(const FermionField& in, FermionField
   if(open_boundaries) ApplyBoundaryMask(out);
 }
 
-template<class Impl>
-void CompactWilsonCloverFermion<Impl>::Meooe(const FermionField& in, FermionField& out) {
+template<class Impl, class CloverHelpers>
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::Meooe(const FermionField& in, FermionField& out) {
   WilsonBase::Meooe(in, out);
   if(open_boundaries) ApplyBoundaryMask(out);
 }
 
-template<class Impl>
-void CompactWilsonCloverFermion<Impl>::MeooeDag(const FermionField& in, FermionField& out) {
+template<class Impl, class CloverHelpers>
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::MeooeDag(const FermionField& in, FermionField& out) {
   WilsonBase::MeooeDag(in, out);
   if(open_boundaries) ApplyBoundaryMask(out);
 }
 
-template<class Impl>
-void CompactWilsonCloverFermion<Impl>::Mooee(const FermionField& in, FermionField& out) {
+template<class Impl, class CloverHelpers>
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::Mooee(const FermionField& in, FermionField& out) {
   if(in.Grid()->_isCheckerBoarded) {
     if(in.Checkerboard() == Odd) {
       MooeeInternal(in, out, DiagonalOdd, TriangleOdd);
@@ -144,13 +150,13 @@ void CompactWilsonCloverFermion<Impl>::Mooee(const FermionField& in, FermionFiel
   if(open_boundaries) ApplyBoundaryMask(out);
 }
 
-template<class Impl>
-void CompactWilsonCloverFermion<Impl>::MooeeDag(const FermionField& in, FermionField& out) {
+template<class Impl, class CloverHelpers>
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::MooeeDag(const FermionField& in, FermionField& out) {
   Mooee(in, out); // blocks are hermitian
 }
 
-template<class Impl>
-void CompactWilsonCloverFermion<Impl>::MooeeInv(const FermionField& in, FermionField& out) {
+template<class Impl, class CloverHelpers>
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::MooeeInv(const FermionField& in, FermionField& out) {
   if(in.Grid()->_isCheckerBoarded) {
     if(in.Checkerboard() == Odd) {
       MooeeInternal(in, out, DiagonalInvOdd, TriangleInvOdd);
@@ -163,23 +169,23 @@ void CompactWilsonCloverFermion<Impl>::MooeeInv(const FermionField& in, FermionF
   if(open_boundaries) ApplyBoundaryMask(out);
 }
 
-template<class Impl>
-void CompactWilsonCloverFermion<Impl>::MooeeInvDag(const FermionField& in, FermionField& out) {
+template<class Impl, class CloverHelpers>
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::MooeeInvDag(const FermionField& in, FermionField& out) {
   MooeeInv(in, out); // blocks are hermitian
 }
 
-template<class Impl>
-void CompactWilsonCloverFermion<Impl>::Mdir(const FermionField& in, FermionField& out, int dir, int disp) {
+template<class Impl, class CloverHelpers>
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::Mdir(const FermionField& in, FermionField& out, int dir, int disp) {
   DhopDir(in, out, dir, disp);
 }
 
-template<class Impl>
-void CompactWilsonCloverFermion<Impl>::MdirAll(const FermionField& in, std::vector<FermionField>& out) {
+template<class Impl, class CloverHelpers>
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::MdirAll(const FermionField& in, std::vector<FermionField>& out) {
   DhopDirAll(in, out);
 }
 
-template<class Impl>
-void CompactWilsonCloverFermion<Impl>::MDeriv(GaugeField& force, const FermionField& X, const FermionField& Y, int dag) {
+template<class Impl, class CloverHelpers>
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::MDeriv(GaugeField& force, const FermionField& X, const FermionField& Y, int dag) {
   assert(!open_boundaries); // TODO check for changes required for open bc
 
   // NOTE: code copied from original clover term
@@ -251,7 +257,7 @@ void CompactWilsonCloverFermion<Impl>::MDeriv(GaugeField& force, const FermionFi
       }
       PropagatorField Slambda = Gamma(sigma[count]) * Lambda; // sigma checked
       Impl::TraceSpinImpl(lambda, Slambda);                   // traceSpin ok
-      force_mu -= factor*Helpers::Cmunu(U, lambda, mu, nu);   // checked
+      force_mu -= factor*CloverHelpers::Cmunu(U, lambda, mu, nu);   // checked
       count++;
     }
 
@@ -261,18 +267,18 @@ void CompactWilsonCloverFermion<Impl>::MDeriv(GaugeField& force, const FermionFi
   force += clover_force;
 }
 
-template<class Impl>
-void CompactWilsonCloverFermion<Impl>::MooDeriv(GaugeField& mat, const FermionField& U, const FermionField& V, int dag) {
+template<class Impl, class CloverHelpers>
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::MooDeriv(GaugeField& mat, const FermionField& U, const FermionField& V, int dag) {
   assert(0);
 }
 
-template<class Impl>
-void CompactWilsonCloverFermion<Impl>::MeeDeriv(GaugeField& mat, const FermionField& U, const FermionField& V, int dag) {
+template<class Impl, class CloverHelpers>
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::MeeDeriv(GaugeField& mat, const FermionField& U, const FermionField& V, int dag) {
   assert(0);
 }
 
-template<class Impl>
-void CompactWilsonCloverFermion<Impl>::MooeeInternal(const FermionField&        in,
+template<class Impl, class CloverHelpers>
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::MooeeInternal(const FermionField&        in,
                     FermionField&              out,
                     const CloverDiagonalField& diagonal,
                     const CloverTriangleField& triangle) {
@@ -285,8 +291,8 @@ void CompactWilsonCloverFermion<Impl>::MooeeInternal(const FermionField&
   CompactHelpers::MooeeKernel(diagonal.oSites(), 1, in, out, diagonal, triangle);
 }
 
-template<class Impl>
-void CompactWilsonCloverFermion<Impl>::ImportGauge(const GaugeField& _Umu) {
+template<class Impl, class CloverHelpers>
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::ImportGauge(const GaugeField& _Umu) {
   // NOTE: parts copied from original implementation
 
   // Import gauge into base class
@@ -318,22 +324,27 @@ void CompactWilsonCloverFermion<Impl>::ImportGauge(const GaugeField& _Umu) {
   TmpOriginal += Helpers::fillCloverXT(Ex) * csw_t;
   TmpOriginal += Helpers::fillCloverYT(Ey) * csw_t;
   TmpOriginal += Helpers::fillCloverZT(Ez) * csw_t;
-  TmpOriginal += this->diag_mass;
-
+  // Handle mass term based on clover policy
+  CloverHelpers::MassTerm(TmpOriginal, this->diag_mass);
+  
   // Convert the data layout of the clover term
   double t4 = usecond();
   CompactHelpers::ConvertLayout(TmpOriginal, Diagonal, Triangle);
 
-  // Possible modify the boundary values
+  // Exponentiate the clover (nothing happens in case of the standard clover)
   double t5 = usecond();
+  CloverHelpers::Exponentiate_Clover(Diagonal, Triangle, csw_t, this->diag_mass);
+
+  // Possible modify the boundary values
+  double t6 = usecond();
   if(open_boundaries) CompactHelpers::ModifyBoundaries(Diagonal, Triangle, csw_t, cF, this->diag_mass);
 
-  // Invert the clover term in the improved layout
-  double t6 = usecond();
+  // Invert the Clover term (explicit inversion needed for the improvement in case of open boundary conditions)
+  double t7 = usecond();
   CompactHelpers::Invert(Diagonal, Triangle, DiagonalInv, TriangleInv);
 
   // Fill the remaining clover fields
-  double t7 = usecond();
+  double t8 = usecond();
   pickCheckerboard(Even, DiagonalEven,    Diagonal);
   pickCheckerboard(Even, TriangleEven,    Triangle);
   pickCheckerboard(Odd,  DiagonalOdd,     Diagonal);
@@ -344,20 +355,19 @@ void CompactWilsonCloverFermion<Impl>::ImportGauge(const GaugeField& _Umu) {
   pickCheckerboard(Odd,  TriangleInvOdd,  TriangleInv);
 
   // Report timings
-  double t8 = usecond();
-#if 0
-  std::cout << GridLogMessage << "CompactWilsonCloverFermion::ImportGauge timings:"
-            << " WilsonFermion::Importgauge = " << (t1 - t0) / 1e6
-            << ", allocations = "               << (t2 - t1) / 1e6
-            << ", field strength = "            << (t3 - t2) / 1e6
-            << ", fill clover = "               << (t4 - t3) / 1e6
-            << ", convert = "                   << (t5 - t4) / 1e6
-            << ", boundaries = "                << (t6 - t5) / 1e6
-            << ", inversions = "                << (t7 - t6) / 1e6
-            << ", pick cbs = "                  << (t8 - t7) / 1e6
-            << ", total = "                     << (t8 - t0) / 1e6
-            << std::endl;
-#endif
+  double t9 = usecond();
+
+  std::cout << GridLogDebug << "CompactWilsonCloverFermion::ImportGauge timings:" << std::endl;
+  std::cout << GridLogDebug << "WilsonFermion::Importgauge = " << (t1 - t0) / 1e6 << std::endl;
+  std::cout << GridLogDebug << "allocations =                " << (t2 - t1) / 1e6 << std::endl;
+  std::cout << GridLogDebug << "field strength =             " << (t3 - t2) / 1e6 << std::endl;
+  std::cout << GridLogDebug << "fill clover =                " << (t4 - t3) / 1e6 << std::endl;
+  std::cout << GridLogDebug << "convert =                    " << (t5 - t4) / 1e6 << std::endl;
+  std::cout << GridLogDebug << "exponentiation =             " << (t6 - t5) / 1e6 << std::endl;
+  std::cout << GridLogDebug << "boundaries =                 " << (t7 - t6) / 1e6 << std::endl;
+  std::cout << GridLogDebug << "inversions =                 " << (t8 - t7) / 1e6 << std::endl;
+  std::cout << GridLogDebug << "pick cbs =                   " << (t9 - t8) / 1e6 << std::endl;
+  std::cout << GridLogDebug << "total =                      " << (t9 - t0) / 1e6 << std::endl;
 }
 
 NAMESPACE_END(Grid);

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

@@ -34,8 +34,8 @@
 
 NAMESPACE_BEGIN(Grid);
 
-template<class Impl>
-WilsonCloverFermion<Impl>::WilsonCloverFermion(GaugeField&                         _Umu,
+template<class Impl, class CloverHelpers>
+WilsonCloverFermion<Impl, CloverHelpers>::WilsonCloverFermion(GaugeField&                         _Umu,
                                                GridCartesian&                      Fgrid,
                                                GridRedBlackCartesian&              Hgrid,
                                                const RealD                         _mass,
@@ -74,8 +74,8 @@ WilsonCloverFermion<Impl>::WilsonCloverFermion(GaugeField&
 }
 
 // *NOT* EO
-template <class Impl>
-void WilsonCloverFermion<Impl>::M(const FermionField &in, FermionField &out)
+template<class Impl, class CloverHelpers>
+void WilsonCloverFermion<Impl, CloverHelpers>::M(const FermionField &in, FermionField &out)
 {
   FermionField temp(out.Grid());
 
@@ -89,8 +89,8 @@ void WilsonCloverFermion<Impl>::M(const FermionField &in, FermionField &out)
   out += temp;
 }
 
-template <class Impl>
-void WilsonCloverFermion<Impl>::Mdag(const FermionField &in, FermionField &out)
+template<class Impl, class CloverHelpers>
+void WilsonCloverFermion<Impl, CloverHelpers>::Mdag(const FermionField &in, FermionField &out)
 {
   FermionField temp(out.Grid());
 
@@ -104,8 +104,8 @@ void WilsonCloverFermion<Impl>::Mdag(const FermionField &in, FermionField &out)
   out += temp;
 }
 
-template <class Impl>
-void WilsonCloverFermion<Impl>::ImportGauge(const GaugeField &_Umu)
+template<class Impl, class CloverHelpers>
+void WilsonCloverFermion<Impl, CloverHelpers>::ImportGauge(const GaugeField &_Umu)
 {
   double t0 = usecond();
   WilsonFermion<Impl>::ImportGauge(_Umu);
@@ -131,47 +131,11 @@ void WilsonCloverFermion<Impl>::ImportGauge(const GaugeField &_Umu)
   CloverTerm += Helpers::fillCloverXT(Ex) * csw_t;
   CloverTerm += Helpers::fillCloverYT(Ey) * csw_t;
   CloverTerm += Helpers::fillCloverZT(Ez) * csw_t;
-  CloverTerm += diag_mass;
-
+   
   double t4 = usecond();
-  int lvol = _Umu.Grid()->lSites();
-  int DimRep = Impl::Dimension;
+  CloverHelpers::Instantiate(CloverTerm, CloverTermInv, csw_t, this->diag_mass);
 
   double t5 = usecond();
-  {
-    autoView(CTv,CloverTerm,CpuRead);
-    autoView(CTIv,CloverTermInv,CpuWrite);
-    thread_for(site, lvol, {
-      Coordinate lcoor;
-      grid->LocalIndexToLocalCoor(site, lcoor);
-      Eigen::MatrixXcd EigenCloverOp = Eigen::MatrixXcd::Zero(Ns * DimRep, Ns * DimRep);
-      Eigen::MatrixXcd EigenInvCloverOp = Eigen::MatrixXcd::Zero(Ns * DimRep, Ns * DimRep);
-      typename SiteClover::scalar_object Qx = Zero(), Qxinv = Zero();
-      peekLocalSite(Qx, CTv, lcoor);
-      //if (csw!=0){
-      for (int j = 0; j < Ns; j++)
-	for (int k = 0; k < Ns; k++)
-	  for (int a = 0; a < DimRep; a++)
-	    for (int b = 0; b < DimRep; b++){
-	      auto zz =  Qx()(j, k)(a, b);
-	      EigenCloverOp(a + j * DimRep, b + k * DimRep) = std::complex<double>(zz);
-	    }
-      //   if (site==0) std::cout << "site =" << site << "\n" << EigenCloverOp << std::endl;
-      
-      EigenInvCloverOp = EigenCloverOp.inverse();
-      //std::cout << EigenInvCloverOp << std::endl;
-      for (int j = 0; j < Ns; j++)
-	for (int k = 0; k < Ns; k++)
-	  for (int a = 0; a < DimRep; a++)
-	    for (int b = 0; b < DimRep; b++)
-	      Qxinv()(j, k)(a, b) = EigenInvCloverOp(a + j * DimRep, b + k * DimRep);
-      //    if (site==0) std::cout << "site =" << site << "\n" << EigenInvCloverOp << std::endl;
-      //  }
-      pokeLocalSite(Qxinv, CTIv, lcoor);
-    });
-  }
-
-  double t6 = usecond();
   // Separate the even and odd parts
   pickCheckerboard(Even, CloverTermEven, CloverTerm);
   pickCheckerboard(Odd, CloverTermOdd, CloverTerm);
@@ -184,48 +148,44 @@ void WilsonCloverFermion<Impl>::ImportGauge(const GaugeField &_Umu)
 
   pickCheckerboard(Even, CloverTermInvDagEven, adj(CloverTermInv));
   pickCheckerboard(Odd, CloverTermInvDagOdd, adj(CloverTermInv));
-  double t7 = usecond();
+  double t6 = usecond();
 
-#if 0
-  std::cout << GridLogMessage << "WilsonCloverFermion::ImportGauge timings:"
-            << " WilsonFermion::Importgauge = " << (t1 - t0) / 1e6
-            << ", allocations = "               << (t2 - t1) / 1e6
-            << ", field strength = "            << (t3 - t2) / 1e6
-            << ", fill clover = "               << (t4 - t3) / 1e6
-            << ", misc = "                      << (t5 - t4) / 1e6
-            << ", inversions = "                << (t6 - t5) / 1e6
-            << ", pick cbs = "                  << (t7 - t6) / 1e6
-            << ", total = "                     << (t7 - t0) / 1e6
-            << std::endl;
-#endif
+  std::cout << GridLogDebug << "WilsonCloverFermion::ImportGauge timings:" << std::endl;
+  std::cout << GridLogDebug << "WilsonFermion::Importgauge = " << (t1 - t0) / 1e6 << std::endl;
+  std::cout << GridLogDebug << "allocations =                " << (t2 - t1) / 1e6 << std::endl;
+  std::cout << GridLogDebug << "field strength =             " << (t3 - t2) / 1e6 << std::endl;
+  std::cout << GridLogDebug << "fill clover =                " << (t4 - t3) / 1e6 << std::endl;
+  std::cout << GridLogDebug << "instantiation =              " << (t5 - t4) / 1e6 << std::endl;
+  std::cout << GridLogDebug << "pick cbs =                   " << (t6 - t5) / 1e6 << std::endl;
+  std::cout << GridLogDebug << "total =                      " << (t6 - t0) / 1e6 << std::endl;
 }
 
-template <class Impl>
-void WilsonCloverFermion<Impl>::Mooee(const FermionField &in, FermionField &out)
+template<class Impl, class CloverHelpers>
+void WilsonCloverFermion<Impl, CloverHelpers>::Mooee(const FermionField &in, FermionField &out)
 {
   this->MooeeInternal(in, out, DaggerNo, InverseNo);
 }
 
-template <class Impl>
-void WilsonCloverFermion<Impl>::MooeeDag(const FermionField &in, FermionField &out)
+template<class Impl, class CloverHelpers>
+void WilsonCloverFermion<Impl, CloverHelpers>::MooeeDag(const FermionField &in, FermionField &out)
 {
   this->MooeeInternal(in, out, DaggerYes, InverseNo);
 }
 
-template <class Impl>
-void WilsonCloverFermion<Impl>::MooeeInv(const FermionField &in, FermionField &out)
+template<class Impl, class CloverHelpers>
+void WilsonCloverFermion<Impl, CloverHelpers>::MooeeInv(const FermionField &in, FermionField &out)
 {
   this->MooeeInternal(in, out, DaggerNo, InverseYes);
 }
 
-template <class Impl>
-void WilsonCloverFermion<Impl>::MooeeInvDag(const FermionField &in, FermionField &out)
+template<class Impl, class CloverHelpers>
+void WilsonCloverFermion<Impl, CloverHelpers>::MooeeInvDag(const FermionField &in, FermionField &out)
 {
   this->MooeeInternal(in, out, DaggerYes, InverseYes);
 }
 
-template <class Impl>
-void WilsonCloverFermion<Impl>::MooeeInternal(const FermionField &in, FermionField &out, int dag, int inv)
+template<class Impl, class CloverHelpers>
+void WilsonCloverFermion<Impl, CloverHelpers>::MooeeInternal(const FermionField &in, FermionField &out, int dag, int inv)
 {
   out.Checkerboard() = in.Checkerboard();
   CloverField *Clover;
@@ -278,8 +238,8 @@ void WilsonCloverFermion<Impl>::MooeeInternal(const FermionField &in, FermionFie
 } // MooeeInternal
 
 // Derivative parts unpreconditioned pseudofermions
-template <class Impl>
-void WilsonCloverFermion<Impl>::MDeriv(GaugeField &force, const FermionField &X, const FermionField &Y, int dag)
+template<class Impl, class CloverHelpers>
+void WilsonCloverFermion<Impl, CloverHelpers>::MDeriv(GaugeField &force, const FermionField &X, const FermionField &Y, int dag)
 {
   conformable(X.Grid(), Y.Grid());
   conformable(X.Grid(), force.Grid());
@@ -349,7 +309,7 @@ void WilsonCloverFermion<Impl>::MDeriv(GaugeField &force, const FermionField &X,
       }
       PropagatorField Slambda = Gamma(sigma[count]) * Lambda; // sigma checked
       Impl::TraceSpinImpl(lambda, Slambda);                   // traceSpin ok
-      force_mu -= factor*Helpers::Cmunu(U, lambda, mu, nu);                   // checked
+      force_mu -= factor*CloverHelpers::Cmunu(U, lambda, mu, nu);                   // checked
       count++;
     }
 
@@ -360,15 +320,15 @@ void WilsonCloverFermion<Impl>::MDeriv(GaugeField &force, const FermionField &X,
 }
 
 // Derivative parts
-template <class Impl>
-void WilsonCloverFermion<Impl>::MooDeriv(GaugeField &mat, const FermionField &X, const FermionField &Y, int dag)
+template<class Impl, class CloverHelpers>
+void WilsonCloverFermion<Impl, CloverHelpers>::MooDeriv(GaugeField &mat, const FermionField &X, const FermionField &Y, int dag)
 {
   assert(0);
 }
 
 // Derivative parts
-template <class Impl>
-void WilsonCloverFermion<Impl>::MeeDeriv(GaugeField &mat, const FermionField &U, const FermionField &V, int dag)
+template<class Impl, class CloverHelpers>
+void WilsonCloverFermion<Impl, CloverHelpers>::MeeDeriv(GaugeField &mat, const FermionField &U, const FermionField &V, int dag)
 {
   assert(0); // not implemented yet
 }

Grid/qcd/action/fermion/implementation/WilsonFermion5DImplementation.h

@@ -92,6 +92,19 @@ WilsonFermion5D<Impl>::WilsonFermion5D(GaugeField &_Umu,
     assert(FourDimRedBlackGrid._simd_layout[d]  ==FourDimGrid._simd_layout[d]);
   }
 
+  if ( p.dirichlet.size() == Nd+1) {
+    Coordinate block = p.dirichlet;
+    if ( block[0] || block[1] || block[2] || block[3] || block[4] ){
+      Dirichlet = 1;
+      Block = block;
+    }
+  } else {
+    Coordinate block(Nd+1,0);
+    Block = block;
+  }
+
+  ZeroCounters();
+
   if (Impl::LsVectorised) { 
 
     int nsimd = Simd::Nsimd();

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

@@ -4,12 +4,13 @@ Grid physics library, www.github.com/paboyle/Grid
 
 Source file: ./lib/qcd/action/fermion/WilsonFermion.cc
 
-Copyright (C) 2015
+Copyright (C) 2022
 
 Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
 Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
 Author: paboyle <paboyle@ph.ed.ac.uk>
+Author: Fabian Joswig <fabian.joswig@ed.ac.uk>
 
 This program is free software; you can redistribute it and/or modify
 it under the terms of the GNU General Public License as published by
@@ -599,11 +600,47 @@ void WilsonFermion<Impl>::ContractConservedCurrent(PropagatorField &q_in_1,
                                                    Current curr_type,
                                                    unsigned int mu)
 {
+  if(curr_type != Current::Vector)
+  {
+    std::cout << GridLogError << "Only the conserved vector current is implemented so far." << std::endl;
+    exit(1);
+  }
+
   Gamma g5(Gamma::Algebra::Gamma5);
   conformable(_grid, q_in_1.Grid());
   conformable(_grid, q_in_2.Grid());
   conformable(_grid, q_out.Grid());
-  assert(0);
+  auto UGrid= this->GaugeGrid();
+
+  PropagatorField tmp_shifted(UGrid);
+  PropagatorField g5Lg5(UGrid);
+  PropagatorField R(UGrid);
+  PropagatorField gmuR(UGrid);
+
+    Gamma::Algebra Gmu [] = {
+    Gamma::Algebra::GammaX,
+    Gamma::Algebra::GammaY,
+    Gamma::Algebra::GammaZ,
+    Gamma::Algebra::GammaT,
+  };
+  Gamma gmu=Gamma(Gmu[mu]);
+
+  g5Lg5=g5*q_in_1*g5;
+  tmp_shifted=Cshift(q_in_2,mu,1);
+  Impl::multLinkField(R,this->Umu,tmp_shifted,mu);
+  gmuR=gmu*R;
+
+  q_out=adj(g5Lg5)*R;
+  q_out-=adj(g5Lg5)*gmuR;
+
+  tmp_shifted=Cshift(q_in_1,mu,1);
+  Impl::multLinkField(g5Lg5,this->Umu,tmp_shifted,mu);
+  g5Lg5=g5*g5Lg5*g5;
+  R=q_in_2;
+  gmuR=gmu*R;
+
+  q_out-=adj(g5Lg5)*R;
+  q_out-=adj(g5Lg5)*gmuR;
 }
 
 
@@ -617,9 +654,51 @@ void WilsonFermion<Impl>::SeqConservedCurrent(PropagatorField &q_in,
                                               unsigned int tmax,
 					      ComplexField &lattice_cmplx)
 {
+  if(curr_type != Current::Vector)
+  {
+    std::cout << GridLogError << "Only the conserved vector current is implemented so far." << std::endl;
+    exit(1);
+  }
+
+  int tshift = (mu == Nd-1) ? 1 : 0;
+  unsigned int LLt    = GridDefaultLatt()[Tp];
   conformable(_grid, q_in.Grid());
   conformable(_grid, q_out.Grid());
-  assert(0);
+  auto UGrid= this->GaugeGrid();
+
+  PropagatorField tmp(UGrid);
+  PropagatorField Utmp(UGrid);
+  PropagatorField L(UGrid);
+  PropagatorField zz (UGrid);
+  zz=Zero();
+  LatticeInteger lcoor(UGrid); LatticeCoordinate(lcoor,Nd-1);
+
+    Gamma::Algebra Gmu [] = {
+    Gamma::Algebra::GammaX,
+    Gamma::Algebra::GammaY,
+    Gamma::Algebra::GammaZ,
+    Gamma::Algebra::GammaT,
+  };
+  Gamma gmu=Gamma(Gmu[mu]);
+
+  tmp = Cshift(q_in,mu,1);
+  Impl::multLinkField(Utmp,this->Umu,tmp,mu);
+  tmp = ( Utmp*lattice_cmplx - gmu*Utmp*lattice_cmplx ); // Forward hop
+  tmp = where((lcoor>=tmin),tmp,zz); // Mask the time
+  q_out = where((lcoor<=tmax),tmp,zz); // Position of current complicated
+
+  tmp = q_in *lattice_cmplx;
+  tmp = Cshift(tmp,mu,-1);
+  Impl::multLinkField(Utmp,this->Umu,tmp,mu+Nd); // Adjoint link
+  tmp = -( Utmp + gmu*Utmp );
+  // Mask the time
+  if (tmax == LLt - 1 && tshift == 1){ // quick fix to include timeslice 0 if tmax + tshift is over the last timeslice
+    unsigned int t0 = 0;
+    tmp = where(((lcoor==t0) || (lcoor>=tmin+tshift)),tmp,zz);
+  } else {
+    tmp = where((lcoor>=tmin+tshift),tmp,zz);
+  }
+  q_out+= where((lcoor<=tmax+tshift),tmp,zz); // Position of current complicated
 }
 
 NAMESPACE_END(Grid);

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

@@ -440,6 +440,17 @@ void WilsonKernels<Impl>::DhopDirKernel( StencilImpl &st, DoubledGaugeField &U,S
 
 #define KERNEL_CALL(A) KERNEL_CALLNB(A); accelerator_barrier();
 
+#define KERNEL_CALL_EXT(A)						\
+  const uint64_t    NN = Nsite*Ls;					\
+  const uint64_t    sz = st.surface_list.size();			\
+  auto ptr = &st.surface_list[0];					\
+  accelerator_forNB( ss, sz, Simd::Nsimd(), {				\
+      int sF = ptr[ss];							\
+      int sU = ss/Ls;							\
+      WilsonKernels<Impl>::A(st_v,U_v,buf,sF,sU,in_v,out_v);		\
+    });									\
+  accelerator_barrier();
+
 #define ASM_CALL(A)							\
   thread_for( ss, Nsite, {						\
     int sU = ss;							\

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

@@ -9,6 +9,7 @@
     Author: paboyle <paboyle@ph.ed.ac.uk>
     Author: Guido Cossu <guido.cossu@ed.ac.uk>
     Author: Daniel Richtmann <daniel.richtmann@gmail.com>
+    Author: Mattia Bruno <mattia.bruno@cern.ch>
 
     This program is free software; you can redistribute it and/or modify
     it under the terms of the GNU General Public License as published by
@@ -32,10 +33,12 @@
 #include <Grid/qcd/spin/Dirac.h>
 #include <Grid/qcd/action/fermion/CompactWilsonCloverFermion.h>
 #include <Grid/qcd/action/fermion/implementation/CompactWilsonCloverFermionImplementation.h>
+#include <Grid/qcd/action/fermion/CloverHelpers.h>
 
 NAMESPACE_BEGIN(Grid);
 
 #include "impl.h"
-template class CompactWilsonCloverFermion<IMPLEMENTATION>; 
+template class CompactWilsonCloverFermion<IMPLEMENTATION, CompactCloverHelpers<IMPLEMENTATION>>; 
+template class CompactWilsonCloverFermion<IMPLEMENTATION, CompactExpCloverHelpers<IMPLEMENTATION>>; 
 
 NAMESPACE_END(Grid);

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

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

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

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

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

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

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

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

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

@@ -8,7 +8,8 @@
 
     Author: paboyle <paboyle@ph.ed.ac.uk>
     Author: Guido Cossu <guido.cossu@ed.ac.uk>
-
+    Author: Mattia Bruno <mattia.bruno@cern.ch>
+    
     This program is free software; you can redistribute it and/or modify
     it under the terms of the GNU General Public License as published by
     the Free Software Foundation; either version 2 of the License, or
@@ -31,10 +32,12 @@
 #include <Grid/qcd/spin/Dirac.h>
 #include <Grid/qcd/action/fermion/WilsonCloverFermion.h>
 #include <Grid/qcd/action/fermion/implementation/WilsonCloverFermionImplementation.h>
+#include <Grid/qcd/action/fermion/CloverHelpers.h>
 
 NAMESPACE_BEGIN(Grid);
 
 #include "impl.h"
-template class WilsonCloverFermion<IMPLEMENTATION>; 
+template class WilsonCloverFermion<IMPLEMENTATION, CloverHelpers<IMPLEMENTATION>>; 
+template class WilsonCloverFermion<IMPLEMENTATION, ExpCloverHelpers<IMPLEMENTATION>>; 
 
 NAMESPACE_END(Grid);

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

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

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

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

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

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

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

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

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

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

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

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

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

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

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

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

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

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

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

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

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

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

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

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

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

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

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

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

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

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#include <Grid/qcd/action/fermion/FermionCore.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

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

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

Grid/qcd/action/fermion/instantiation/generate_instantiations.sh

@@ -18,6 +18,10 @@ WILSON_IMPL_LIST=" \
 	   GparityWilsonImplF \
 	   GparityWilsonImplD "
 
+COMPACT_WILSON_IMPL_LIST=" \
+	   WilsonImplF \
+	   WilsonImplD "
+
 DWF_IMPL_LIST=" \
 	   WilsonImplF \
 	   WilsonImplD \
@@ -40,13 +44,23 @@ EOF
 
 done
 
-CC_LIST="WilsonCloverFermionInstantiation CompactWilsonCloverFermionInstantiation WilsonFermionInstantiation WilsonKernelsInstantiation WilsonTMFermionInstantiation"
+CC_LIST="WilsonCloverFermionInstantiation WilsonFermionInstantiation WilsonKernelsInstantiation WilsonTMFermionInstantiation"
 
 for impl in $WILSON_IMPL_LIST
 do
 for f in $CC_LIST
 do
-  ln -f -s ../$f.cc.master $impl/$f$impl.cc 
+  ln -f -s ../$f.cc.master $impl/$f$impl.cc
+done
+done
+
+CC_LIST="CompactWilsonCloverFermionInstantiation"
+
+for impl in $COMPACT_WILSON_IMPL_LIST
+do
+for f in $CC_LIST
+do
+  ln -f -s ../$f.cc.master $impl/$f$impl.cc
 done
 done
 
@@ -63,14 +77,14 @@ for impl in $DWF_IMPL_LIST $GDWF_IMPL_LIST
 do
 for f in $CC_LIST
 do
-  ln -f -s ../$f.cc.master $impl/$f$impl.cc 
+  ln -f -s ../$f.cc.master $impl/$f$impl.cc
 done
 done
 
 # overwrite the .cc file in Gparity directories
 for impl in $GDWF_IMPL_LIST
 do
-  ln -f -s ../WilsonKernelsInstantiationGparity.cc.master $impl/WilsonKernelsInstantiation$impl.cc 
+  ln -f -s ../WilsonKernelsInstantiationGparity.cc.master $impl/WilsonKernelsInstantiation$impl.cc
 done
 
 
@@ -84,7 +98,7 @@ for impl in $STAG_IMPL_LIST
 do
 for f in $CC_LIST
 do
-  ln -f -s ../$f.cc.master $impl/$f$impl.cc 
+  ln -f -s ../$f.cc.master $impl/$f$impl.cc
 done
 done
 

Grid/qcd/action/filters/DirichletFilter.h

@@ -53,9 +53,9 @@ struct DirichletFilter: public MomentumFilterBase<MomentaField>
     LatticeInteger coor(grid); 
     LatCM zz(grid); zz = Zero();
     for(int mu=0;mu<Nd;mu++) {
-      if ( (Block[mu]) && (Block[mu] < grid->GlobalDimensions()[mu] ) ) {
+      if ( (Block[mu]) && (Block[mu] <= grid->GlobalDimensions()[mu] ) ) {
 	// If costly could provide Grid earlier and precompute masks
-	std::cout << " Dirichlet in mu="<<mu<<std::endl;
+	std::cout << GridLogMessage << " Dirichlet in mu="<<mu<<std::endl;
 	LatticeCoordinate(coor,mu);
 	auto P_mu = PeekIndex<LorentzIndex>(P, mu);
 	P_mu = where(mod(coor,Block[mu])==Integer(Block[mu]-1),zz,P_mu);

Grid/qcd/action/pseudofermion/OneFlavourRationalRatio.h

@@ -75,16 +75,14 @@ NAMESPACE_BEGIN(Grid);
 	remez.generateApprox(param.degree,1,2);
 	PowerHalf.Init(remez,param.tolerance,false);
 	PowerNegHalf.Init(remez,param.tolerance,true);
+	MDPowerNegHalf.Init(remez,param.mdtolerance,true);
 
 	// MdagM^(+- 1/4)
 	std::cout<<GridLogMessage << "Generating degree "<<param.degree<<" for x^(1/4)"<<std::endl;
 	remez.generateApprox(param.degree,1,4);
    	PowerQuarter.Init(remez,param.tolerance,false);
-	PowerNegQuarter.Init(remez,param.tolerance,true);
-
-	// Derive solves different tol
    	MDPowerQuarter.Init(remez,param.mdtolerance,false);
-	MDPowerNegHalf.Init(remez,param.mdtolerance,true);
+	PowerNegQuarter.Init(remez,param.tolerance,true);
       };
 
       virtual std::string action_name(){return "OneFlavourRatioRationalPseudoFermionAction";}

Grid/qcd/hmc/integrators/Integrator.h

@@ -145,7 +145,7 @@ protected:
 
       MomFilter->applyFilter(force);
       std::cout << GridLogIntegrator << " update_P : Level [" << level <<"]["<<a <<"] "<<name<< std::endl;
-      //      DumpSliceNorm("force ",force,Nd-1);
+      DumpSliceNorm("force ",force,Nd-1);
       
       Real force_abs   = std::sqrt(norm2(force)/U.Grid()->gSites()); //average per-site norm.  nb. norm2(latt) = \sum_x norm2(latt[x]) 
       Real impulse_abs = force_abs * ep * HMC_MOMENTUM_DENOMINATOR;    

Grid/qcd/observables/topological_charge.h

@@ -99,7 +99,7 @@ public:
 	// using wilson flow by default here
 	WilsonFlow<PeriodicGimplR> WF(Pars.Smearing.steps, Pars.Smearing.step_size, Pars.Smearing.meas_interval);
 	WF.smear_adaptive(Usmear, U, Pars.Smearing.maxTau);
-	Real T0   = WF.energyDensityPlaquette(Usmear);
+	Real T0   = WF.energyDensityPlaquette(Pars.Smearing.maxTau, Usmear);
 	std::cout << GridLogMessage << std::setprecision(std::numeric_limits<Real>::digits10 + 1)
 		  << "T0                : [ " << traj << " ] "<< T0 << std::endl;
       }

Grid/qcd/smearing/WilsonFlow.h

@@ -7,6 +7,7 @@ Source file: ./lib/qcd/modules/plaquette.h
 Copyright (C) 2017
 
 Author: Guido Cossu <guido.cossu@ed.ac.uk>
+Author: Christopher Kelly <ckelly@bnl.gov>
 
 This program is free software; you can redistribute it and/or modify
 it under the terms of the GNU General Public License as published by
@@ -33,28 +34,44 @@ NAMESPACE_BEGIN(Grid);
 
 template <class Gimpl>
 class WilsonFlow: public Smear<Gimpl>{
+public:
+  //Store generic measurements to take during smearing process using std::function
+  typedef std::function<void(int, RealD, const typename Gimpl::GaugeField &)> FunctionType;  //int: step,  RealD: flow time,  GaugeField : the gauge field
+  
+private:
   unsigned int Nstep;
-  unsigned int measure_interval;
-  mutable RealD epsilon, taus;
-
+  RealD epsilon; //for regular smearing this is the time step, for adaptive it is the initial time step
+ 
+  std::vector< std::pair<int, FunctionType> > functions; //The int maps to the measurement frequency
 
   mutable WilsonGaugeAction<Gimpl> SG;
 
-  void evolve_step(typename Gimpl::GaugeField&) const;
-  void evolve_step_adaptive(typename Gimpl::GaugeField&, RealD);
-  RealD tau(unsigned int t)const {return epsilon*(t+1.0); }
+  //Evolve the gauge field by 1 step and update tau
+  void evolve_step(typename Gimpl::GaugeField &U, RealD &tau) const;
+  //Evolve the gauge field by 1 step and update tau and the current time step eps
+  void evolve_step_adaptive(typename Gimpl::GaugeField&U, RealD &tau, RealD &eps, RealD maxTau) const;
 
 public:
   INHERIT_GIMPL_TYPES(Gimpl)
 
+  void resetActions(){ functions.clear(); }
+
+  void addMeasurement(int meas_interval, FunctionType meas){ functions.push_back({meas_interval, meas}); }
+
+  //Set the class to perform the default measurements: 
+  //the plaquette energy density every step
+  //the plaquette topological charge every 'topq_meas_interval' steps
+  //and output to stdout
+  void setDefaultMeasurements(int topq_meas_interval = 1);
+
   explicit WilsonFlow(unsigned int Nstep, RealD epsilon, unsigned int interval = 1):
   Nstep(Nstep),
     epsilon(epsilon),
-    measure_interval(interval),
     SG(WilsonGaugeAction<Gimpl>(3.0)) {
     // WilsonGaugeAction with beta 3.0
     assert(epsilon > 0.0);
     LogMessage();
+    setDefaultMeasurements(interval);
   }
 
   void LogMessage() {
@@ -73,9 +90,29 @@ public:
     // undefined for WilsonFlow
   }
 
-  void smear_adaptive(GaugeField&, const GaugeField&, RealD maxTau);
-  RealD energyDensityPlaquette(unsigned int step, const GaugeField& U) const;
-  RealD energyDensityPlaquette(const GaugeField& U) const;
+  void smear_adaptive(GaugeField&, const GaugeField&, RealD maxTau) const;
+
+  //Compute t^2 <E(t)> for time t from the plaquette
+  static RealD energyDensityPlaquette(const RealD t, const GaugeField& U);
+
+  //Compute t^2 <E(t)> for time t from the 1x1 cloverleaf form
+  //t is the Wilson flow time
+  static RealD energyDensityCloverleaf(const RealD t, const GaugeField& U);
+  
+  //Evolve the gauge field by Nstep steps of epsilon and return the energy density computed every interval steps
+  //The smeared field is output as V
+  std::vector<RealD> flowMeasureEnergyDensityPlaquette(GaugeField &V, const GaugeField& U, int measure_interval = 1);
+
+  //Version that does not return the smeared field
+  std::vector<RealD> flowMeasureEnergyDensityPlaquette(const GaugeField& U, int measure_interval = 1);
+
+
+  //Evolve the gauge field by Nstep steps of epsilon and return the Cloverleaf energy density computed every interval steps
+  //The smeared field is output as V
+  std::vector<RealD> flowMeasureEnergyDensityCloverleaf(GaugeField &V, const GaugeField& U, int measure_interval = 1);
+
+  //Version that does not return the smeared field
+  std::vector<RealD> flowMeasureEnergyDensityCloverleaf(const GaugeField& U, int measure_interval = 1);
 };
 
 
@@ -83,7 +120,7 @@ public:
 // Implementations
 ////////////////////////////////////////////////////////////////////////////////
 template <class Gimpl>
-void WilsonFlow<Gimpl>::evolve_step(typename Gimpl::GaugeField &U) const{
+void WilsonFlow<Gimpl>::evolve_step(typename Gimpl::GaugeField &U, RealD &tau) const{
   GaugeField Z(U.Grid());
   GaugeField tmp(U.Grid());
   SG.deriv(U, Z);
@@ -99,12 +136,13 @@ void WilsonFlow<Gimpl>::evolve_step(typename Gimpl::GaugeField &U) const{
   SG.deriv(U, tmp); Z += tmp;                 // 4/3*(17/36*Z0 -8/9*Z1) +Z2
   Z *= 3.0/4.0;                               // Z = 17/36*Z0 -8/9*Z1 +3/4*Z2
   Gimpl::update_field(Z, U, -2.0*epsilon);    // V(t+e) = exp(ep*Z)*W2
+  tau += epsilon;
 }
 
 template <class Gimpl>
-void WilsonFlow<Gimpl>::evolve_step_adaptive(typename Gimpl::GaugeField &U, RealD maxTau) {
-  if (maxTau - taus < epsilon){
-    epsilon = maxTau-taus;
+void WilsonFlow<Gimpl>::evolve_step_adaptive(typename Gimpl::GaugeField &U, RealD &tau, RealD &eps, RealD maxTau) const{
+  if (maxTau - tau < eps){
+    eps = maxTau-tau;
   }
   //std::cout << GridLogMessage << "Integration epsilon : " << epsilon << std::endl;
   GaugeField Z(U.Grid());
@@ -114,95 +152,151 @@ void WilsonFlow<Gimpl>::evolve_step_adaptive(typename Gimpl::GaugeField &U, Real
   SG.deriv(U, Z);
   Zprime = -Z;
   Z *= 0.25;                                  // Z0 = 1/4 * F(U)
-  Gimpl::update_field(Z, U, -2.0*epsilon);    // U = W1 = exp(ep*Z0)*W0
+  Gimpl::update_field(Z, U, -2.0*eps);    // U = W1 = exp(ep*Z0)*W0
 
   Z *= -17.0/8.0;
   SG.deriv(U, tmp); Z += tmp;                 // -17/32*Z0 +Z1
   Zprime += 2.0*tmp;
   Z *= 8.0/9.0;                               // Z = -17/36*Z0 +8/9*Z1
-  Gimpl::update_field(Z, U, -2.0*epsilon);    // U_= W2 = exp(ep*Z)*W1
+  Gimpl::update_field(Z, U, -2.0*eps);    // U_= W2 = exp(ep*Z)*W1
     
 
   Z *= -4.0/3.0;
   SG.deriv(U, tmp); Z += tmp;                 // 4/3*(17/36*Z0 -8/9*Z1) +Z2
   Z *= 3.0/4.0;                               // Z = 17/36*Z0 -8/9*Z1 +3/4*Z2
-  Gimpl::update_field(Z, U, -2.0*epsilon);    // V(t+e) = exp(ep*Z)*W2
+  Gimpl::update_field(Z, U, -2.0*eps);    // V(t+e) = exp(ep*Z)*W2
 
   // Ramos 
-  Gimpl::update_field(Zprime, Uprime, -2.0*epsilon); // V'(t+e) = exp(ep*Z')*W0
+  Gimpl::update_field(Zprime, Uprime, -2.0*eps); // V'(t+e) = exp(ep*Z')*W0
   // Compute distance as norm^2 of the difference
   GaugeField diffU = U - Uprime;
   RealD diff = norm2(diffU);
   // adjust integration step
     
-  taus += epsilon;
+  tau += eps;
   //std::cout << GridLogMessage << "Adjusting integration step with distance: " << diff << std::endl;
     
-  epsilon = epsilon*0.95*std::pow(1e-4/diff,1./3.);
+  eps = eps*0.95*std::pow(1e-4/diff,1./3.);
   //std::cout << GridLogMessage << "New epsilon : " << epsilon << std::endl;
 
 }
 
+
 template <class Gimpl>
-RealD WilsonFlow<Gimpl>::energyDensityPlaquette(unsigned int step, const GaugeField& U) const {
-  RealD td = tau(step);
-  return 2.0 * td * td * SG.S(U)/U.Grid()->gSites();
+RealD WilsonFlow<Gimpl>::energyDensityPlaquette(const RealD t, const GaugeField& U){
+  static WilsonGaugeAction<Gimpl> SG(3.0);
+  return 2.0 * t * t * SG.S(U)/U.Grid()->gSites();
+}
+
+//Compute t^2 <E(t)> for time from the 1x1 cloverleaf form
+template <class Gimpl>
+RealD WilsonFlow<Gimpl>::energyDensityCloverleaf(const RealD t, const GaugeField& U){
+  typedef typename Gimpl::GaugeLinkField GaugeMat;
+  typedef typename Gimpl::GaugeField GaugeLorentz;
+
+  assert(Nd == 4);
+  //E = 1/2 tr( F_munu F_munu )
+  //However as  F_numu = -F_munu, only need to sum the trace of the squares of the following 6 field strengths:
+  //F_01 F_02 F_03   F_12 F_13  F_23
+  GaugeMat F(U.Grid());
+  LatticeComplexD R(U.Grid());
+  R = Zero();
+  
+  for(int mu=0;mu<3;mu++){
+    for(int nu=mu+1;nu<4;nu++){
+      WilsonLoops<Gimpl>::FieldStrength(F, U, mu, nu);
+      R = R + trace(F*F);
+    }
+  }
+  ComplexD out = sum(R);
+  out = t*t*out / RealD(U.Grid()->gSites());
+  return -real(out); //minus sign necessary for +ve energy
+}
+
+
+template <class Gimpl>
+std::vector<RealD> WilsonFlow<Gimpl>::flowMeasureEnergyDensityPlaquette(GaugeField &V, const GaugeField& U, int measure_interval){
+  std::vector<RealD> out;
+  resetActions();
+  addMeasurement(measure_interval, [&out](int step, RealD t, const typename Gimpl::GaugeField &U){ 
+      std::cout << GridLogMessage << "[WilsonFlow] Computing plaquette energy density for step " << step << std::endl;
+      out.push_back( energyDensityPlaquette(t,U) );
+    });      
+  smear(V,U);
+  return out;
 }
 
 template <class Gimpl>
-RealD WilsonFlow<Gimpl>::energyDensityPlaquette(const GaugeField& U) const {
-  return 2.0 * taus * taus * SG.S(U)/U.Grid()->gSites();
+std::vector<RealD> WilsonFlow<Gimpl>::flowMeasureEnergyDensityPlaquette(const GaugeField& U, int measure_interval){
+  GaugeField V(U);
+  return flowMeasureEnergyDensityPlaquette(V,U, measure_interval);
 }
 
+template <class Gimpl>
+std::vector<RealD> WilsonFlow<Gimpl>::flowMeasureEnergyDensityCloverleaf(GaugeField &V, const GaugeField& U, int measure_interval){
+  std::vector<RealD> out;
+  resetActions();
+  addMeasurement(measure_interval, [&out](int step, RealD t, const typename Gimpl::GaugeField &U){ 
+      std::cout << GridLogMessage << "[WilsonFlow] Computing Cloverleaf energy density for step " << step << std::endl;
+      out.push_back( energyDensityCloverleaf(t,U) );
+    });      
+  smear(V,U);
+  return out;
+}
+
+template <class Gimpl>
+std::vector<RealD> WilsonFlow<Gimpl>::flowMeasureEnergyDensityCloverleaf(const GaugeField& U, int measure_interval){
+  GaugeField V(U);
+  return flowMeasureEnergyDensityCloverleaf(V,U, measure_interval);
+}
+
+
 
 //#define WF_TIMING 
-
-
-
 template <class Gimpl>
-void WilsonFlow<Gimpl>::smear(GaugeField& out, const GaugeField& in) const {
+void WilsonFlow<Gimpl>::smear(GaugeField& out, const GaugeField& in) const{
   out = in;
-  for (unsigned int step = 1; step <= Nstep; step++) {
+  RealD taus = 0.;
+  for (unsigned int step = 1; step <= Nstep; step++) { //step indicates the number of smearing steps applied at the time of measurement
     auto start = std::chrono::high_resolution_clock::now();
-    evolve_step(out);
+    evolve_step(out, taus);
     auto end = std::chrono::high_resolution_clock::now();
     std::chrono::duration<double> diff = end - start;
 #ifdef WF_TIMING
     std::cout << "Time to evolve " << diff.count() << " s\n";
 #endif
-    std::cout << GridLogMessage << "[WilsonFlow] Energy density (plaq) : "
-		  << step << "  " << tau(step) << "  " 
-	      << energyDensityPlaquette(step,out) << std::endl;
-    if( step % measure_interval == 0){
-      std::cout << GridLogMessage << "[WilsonFlow] Top. charge           : "
-		<< step << "  " 
-		<< WilsonLoops<PeriodicGimplR>::TopologicalCharge(out) << std::endl;
-    }
+    //Perform measurements
+    for(auto const &meas : functions)
+      if( step % meas.first == 0 ) meas.second(step,taus,out);
   }
 }
 
 template <class Gimpl>
-void WilsonFlow<Gimpl>::smear_adaptive(GaugeField& out, const GaugeField& in, RealD maxTau){
+void WilsonFlow<Gimpl>::smear_adaptive(GaugeField& out, const GaugeField& in, RealD maxTau) const{
   out = in;
-  taus = epsilon;
+  RealD taus = 0.;
+  RealD eps = epsilon;
   unsigned int step = 0;
   do{
     step++;
     //std::cout << GridLogMessage << "Evolution time :"<< taus << std::endl;
-    evolve_step_adaptive(out, maxTau);
-    std::cout << GridLogMessage << "[WilsonFlow] Energy density (plaq) : "
-		  << step << "  " << taus << "  "
-	      << energyDensityPlaquette(out) << std::endl;
-    if( step % measure_interval == 0){
-      std::cout << GridLogMessage << "[WilsonFlow] Top. charge           : "
-		<< step << "  " 
-		<< WilsonLoops<PeriodicGimplR>::TopologicalCharge(out) << std::endl;
-    }
+    evolve_step_adaptive(out, taus, eps, maxTau);
+    //Perform measurements
+    for(auto const &meas : functions)
+      if( step % meas.first == 0 ) meas.second(step,taus,out);
   } while (taus < maxTau);
-
-
-
 }
 
+template <class Gimpl>
+void WilsonFlow<Gimpl>::setDefaultMeasurements(int topq_meas_interval){
+  addMeasurement(1, [](int step, RealD t, const typename Gimpl::GaugeField &U){
+      std::cout << GridLogMessage << "[WilsonFlow] Energy density (plaq) : "  << step << "  " << t << "  " << energyDensityPlaquette(t,U) << std::endl;
+    });
+  addMeasurement(topq_meas_interval, [](int step, RealD t, const typename Gimpl::GaugeField &U){
+      std::cout << GridLogMessage << "[WilsonFlow] Top. charge           : "  << step << "  " << WilsonLoops<Gimpl>::TopologicalCharge(U) << std::endl;
+    });
+}
+
+
 NAMESPACE_END(Grid);
 

Grid/qcd/utils/GaugeFix.h

@@ -72,14 +72,13 @@ public:
 
   //Fix the gauge field Umu
   //0 < alpha < 1 is related to the step size, cf https://arxiv.org/pdf/1405.5812.pdf
-  static void SteepestDescentGaugeFix(GaugeLorentz &Umu, Real alpha,int maxiter,Real Omega_tol, Real Phi_tol,bool Fourier=false,int orthog=-1) {
+  static void SteepestDescentGaugeFix(GaugeLorentz &Umu,Real & alpha,int maxiter,Real Omega_tol, Real Phi_tol,bool Fourier=false,int orthog=-1,bool err_on_no_converge=true) {
     GridBase *grid = Umu.Grid();
     GaugeMat xform(grid);
-    SteepestDescentGaugeFix(Umu,xform,alpha,maxiter,Omega_tol,Phi_tol,Fourier,orthog);
+    SteepestDescentGaugeFix(Umu,xform,alpha,maxiter,Omega_tol,Phi_tol,Fourier,orthog,err_on_no_converge);
   }
-
+  static void SteepestDescentGaugeFix(GaugeLorentz &Umu,GaugeMat &xform,Real & alpha,int maxiter,Real Omega_tol, Real Phi_tol,bool Fourier=false,int orthog=-1,bool err_on_no_converge=true) {
   //Fix the gauge field Umu and also return the gauge transformation from the original gauge field, xform
-  static void SteepestDescentGaugeFix(GaugeLorentz &Umu,GaugeMat &xform, Real alpha,int maxiter,Real Omega_tol, Real Phi_tol,bool Fourier=false,int orthog=-1) {
 
     GridBase *grid = Umu.Grid();
 
@@ -141,7 +140,9 @@ public:
 
       }
     }
-    assert(0 && "Gauge fixing did not converge within the specified number of iterations");
+    std::cout << GridLogError << "Gauge fixing did not converge in " << maxiter << " iterations." << std::endl;
+    if (err_on_no_converge)
+      assert(0 && "Gauge fixing did not converge within the specified number of iterations");
   };
   static Real SteepestDescentStep(std::vector<GaugeMat> &U,GaugeMat &xform, Real alpha, GaugeMat & dmuAmu,int orthog) {
     GridBase *grid = U[0].Grid();

Grid/qcd/utils/WilsonLoops.h

@@ -215,7 +215,7 @@ public:
 
     double vol = Umu.Grid()->gSites();
 
-    return p.real() / vol / 4.0 / 3.0;
+    return p.real() / vol / (4.0 * Nc ) ;
   };
 
   //////////////////////////////////////////////////

Grid/stencil/SimpleCompressor.h

@@ -52,6 +52,11 @@ public:
     return arg;
   }
 };
+class SimpleStencilParams{
+public:
+  Coordinate dirichlet;
+  SimpleStencilParams() {};
+};
 
 NAMESPACE_END(Grid);
 

Grid/stencil/Stencil.h

@@ -131,7 +131,6 @@ class CartesianStencilAccelerator {
   int           _checkerboard;
   int           _npoints; // Move to template param?
   int           _osites;
-  int           _dirichlet;
   StencilVector _directions;
   StencilVector _distances;
   StencilVector _comms_send;
@@ -503,7 +502,6 @@ public:
   }
   void AddCopy(void *from,void * to, Integer bytes)
   {
-    //    std::cout << "Adding CopyReceiveBuffer "<<std::hex<<from<<" "<<to<<std::dec<<" "<<bytes<<std::endl;
     CopyReceiveBuffer obj;
     obj.from_p = from;
     obj.to_p = to;
@@ -517,7 +515,7 @@ public:
       cobj *from=(cobj *)CopyReceiveBuffers[i].from_p;
       cobj *to  =(cobj *)CopyReceiveBuffers[i].to_p;
       Integer words = CopyReceiveBuffers[i].bytes/sizeof(cobj);
-      //    std::cout << "CopyReceiveBuffer "<<std::hex<<from<<" "<<to<<std::dec<<" "<<words*sizeof(cobj)<<std::endl;
+
       accelerator_forNB(j, words, cobj::Nsimd(), {
 	  coalescedWrite(to[j] ,coalescedRead(from [j]));
       });
@@ -543,13 +541,12 @@ public:
 	   &&(CachedTransfers[i].lane       ==lane)
 	   &&(CachedTransfers[i].cb         ==cb)
 	     ){
-	//	std::cout << "Found duplicate plane dir "<<direction<<" plane "<< OrthogPlane<< " simd "<<lane << " relproc "<<DestProc<< " bytes "<<bytes <<std::endl;
+
 	AddCopy(CachedTransfers[i].recv_buf,recv_buf,bytes);
 	return 1;
       }
     }
 
-    //    std::cout << "No duplicate plane dir "<<direction<<" plane "<< OrthogPlane<< " simd "<<lane << " relproc "<<DestProc<<"  bytes "<<bytes<<std::endl;
     CachedTransfers.push_back(obj);
     return 0;
   }
@@ -643,23 +640,23 @@ public:
 	}
       }
       if(local == 0) {
-	surface_list.push_back(site);
+	for(int s=0;s<Ls;s++){
+	  surface_list.push_back(site*Ls+s);
+	}
       }
     }
   }
   /// Introduce a block structure and switch off comms on boundaries
   void DirichletBlock(const Coordinate &dirichlet_block)
   {
-    this->_dirichlet = 1;
     for(int ii=0;ii<this->_npoints;ii++){
       int dimension    = this->_directions[ii];
       int displacement = this->_distances[ii];
-      int shift = displacement;
       int gd = _grid->_gdimensions[dimension];
       int fd = _grid->_fdimensions[dimension];
       int pd = _grid->_processors [dimension];
-      int ld = gd/pd;
       int pc = _grid->_processor_coor[dimension];
+      int ld = fd/pd;
       ///////////////////////////////////////////
       // Figure out dirichlet send and receive
       // on this leg of stencil.
@@ -668,25 +665,25 @@ public:
       int block = dirichlet_block[dimension];
       this->_comms_send[ii] = comm_dim;
       this->_comms_recv[ii] = comm_dim;
-      if ( block ) {
+      if ( block && comm_dim ) {
 	assert(abs(displacement) < ld );
-      
+	// Quiesce communication across block boundaries
 	if( displacement > 0 ) {
 	  // High side, low side
 	  // | <--B--->|
 	  // |    |    |
 	  //           noR
 	  // noS
-	  if ( (ld*(pc+1) ) % block == 0 ) this->_comms_recv[ii] = 0;
-	  if ( ( ld*pc ) % block == 0    ) this->_comms_send[ii] = 0;
+	  if ( ( (ld*(pc+1) ) % block ) == 0 ) this->_comms_recv[ii] = 0;
+	  if ( ( (ld*pc     ) % block ) == 0 ) this->_comms_send[ii] = 0;
 	} else {
 	  // High side, low side
 	  // | <--B--->|
 	  // |    |    |
 	  //           noS
 	  // noR
-	  if ( (ld*(pc+1) ) % block == 0 ) this->_comms_send[ii] = 0;
-	  if ( ( ld*pc ) % block    == 0 ) this->_comms_recv[ii] = 0;
+	  if ( ( (ld*(pc+1) ) % block ) == 0 ) this->_comms_send[ii] = 0;
+	  if ( ( (ld*pc     ) % block ) == 0 ) this->_comms_recv[ii] = 0;
 	}
       }
     }
@@ -698,7 +695,6 @@ public:
 		   const std::vector<int> &distances,
 		   Parameters p)
   {
-    this->_dirichlet = 0;
     face_table_computed=0;
     _grid    = grid;
     this->parameters=p;
@@ -715,6 +711,8 @@ public:
     this->_comms_recv.resize(npoints); 
     this->same_node.resize(npoints);
 
+    if ( p.dirichlet.size() ) DirichletBlock(p.dirichlet); // comms send/recv set up
+
     _unified_buffer_size=0;
     surface_list.resize(0);
 
@@ -734,7 +732,7 @@ public:
       int gd = _grid->_gdimensions[dimension];
       int fd = _grid->_fdimensions[dimension];
       int pd = _grid->_processors [dimension];
-      int ld = gd/pd;
+      //      int ld = gd/pd;
       int rd = _grid->_rdimensions[dimension];
       int pc = _grid->_processor_coor[dimension];
       this->_permute_type[point]=_grid->PermuteType(dimension);
@@ -746,9 +744,6 @@ public:
       int splice_dim      = _grid->_simd_layout[dimension]>1 && (comm_dim);
       int rotate_dim      = _grid->_simd_layout[dimension]>2;
 
-      this->_comms_send[ii] = comm_dim;
-      this->_comms_recv[ii] = comm_dim;
-
       assert ( (rotate_dim && comm_dim) == false) ; // Do not think spread out is supported
 
       int sshift[2];
@@ -878,12 +873,14 @@ public:
     for(int x=0;x<rd;x++){
 
       int permute_type=grid->PermuteType(dimension);
+      int permute_slice;
 
       int sx        =  (x+sshift)%rd;
 
       int offnode = 0;
       if ( simd_layout > 1 ) {
 
+	permute_slice=1;
 	for(int i=0;i<Nsimd;i++){
 
 	  int inner_bit = (Nsimd>>(permute_type+1));
@@ -900,6 +897,7 @@ public:
       } else {
 	int comm_proc = ((x+sshift)/rd)%pd;
 	offnode = (comm_proc!= 0);
+	permute_slice=0;
       }
 
       int wraparound=0;
@@ -911,25 +909,29 @@ public:
       }
 
       // Wrap locally dirichlet support case OR node local
-      if ( (offnode==0) || (comms_recv==0)  ) {
+      if ( offnode==0 ) {
 
-	int permute_slice=0;
+	permute_slice=0;
 	CopyPlane(point,dimension,x,sx,cbmask,permute_slice,wraparound);
-
+	
       } else {
 
+	if ( comms_recv ) {
+
+	  ScatterPlane(point,dimension,x,cbmask,_unified_buffer_size,wraparound); // permute/extract/merge is done in comms phase
+
+	} else { 
+
+	  CopyPlane(point,dimension,x,sx,cbmask,permute_slice,wraparound);
+
+	}
+
+      }
+      
+      if ( offnode ) {
 	int words = buffer_size;
 	if (cbmask != 0x3) words=words>>1;
-
-	//	int rank           = grid->_processor;
-	//	int recv_from_rank;
-	//	int xmit_to_rank;
-
-	int unified_buffer_offset = _unified_buffer_size;
 	_unified_buffer_size    += words;
-
-	ScatterPlane(point,dimension,x,cbmask,unified_buffer_offset,wraparound); // permute/extract/merge is done in comms phase
-
       }
     }
   }
@@ -1060,8 +1062,6 @@ public:
       int comm_proc = ((x+sshift)/rd)%pd;
       
       if (comm_proc) {
-
-	
 	
 	int words = buffer_size;
 	if (cbmask != 0x3) words=words>>1;
@@ -1069,64 +1069,70 @@ public:
 	int bytes =  words * compress.CommDatumSize();
 
 	int so  = sx*rhs.Grid()->_ostride[dimension]; // base offset for start of plane
-	if ( !face_table_computed ) {
-	  face_table.resize(face_idx+1);
-	  std::vector<std::pair<int,int> >  face_table_host ;
-	  Gather_plane_table_compute ((GridBase *)_grid,dimension,sx,cbmask,u_comm_offset,face_table_host);
-	  face_table[face_idx].resize(face_table_host.size());
-	  acceleratorCopyToDevice(&face_table_host[0],
-				  &face_table[face_idx][0],
-				  face_table[face_idx].size()*sizeof(face_table_host[0]));
-	}
+	int comm_off = u_comm_offset;
 
-	//      	int rank           = _grid->_processor;
 	int recv_from_rank;
 	int xmit_to_rank;
+	cobj *recv_buf;
+	cobj *send_buf;
 	_grid->ShiftedRanks(dimension,comm_proc,xmit_to_rank,recv_from_rank);
 
 	assert (xmit_to_rank   != _grid->ThisRank());
 	assert (recv_from_rank != _grid->ThisRank());
 
-	cobj *recv_buf;
-	if ( compress.DecompressionStep() ) {
-	  recv_buf=u_simd_recv_buf[0];
-	} else {
-	  recv_buf=this->u_recv_buf_p;
+	if( comms_send ) {
+
+	  if ( !face_table_computed ) {
+	    face_table.resize(face_idx+1);
+	    std::vector<std::pair<int,int> >  face_table_host ;
+	    Gather_plane_table_compute ((GridBase *)_grid,dimension,sx,cbmask,comm_off,face_table_host);
+	    face_table[face_idx].resize(face_table_host.size());
+	    acceleratorCopyToDevice(&face_table_host[0],
+				    &face_table[face_idx][0],
+				    face_table[face_idx].size()*sizeof(face_table_host[0]));
+	  }
+
+
+	  if ( compress.DecompressionStep() ) {
+	    recv_buf=u_simd_recv_buf[0];
+	  } else {
+	    recv_buf=this->u_recv_buf_p;
+	  }
+
+	  send_buf = this->u_send_buf_p; // Gather locally, must send
+	
+	  ////////////////////////////////////////////////////////
+	  // Gather locally
+	  ////////////////////////////////////////////////////////
+	  assert(send_buf!=NULL);
+
+	  Gather_plane_simple_table(face_table[face_idx],rhs,send_buf,compress,comm_off,so);
 	}
 
-
-	cobj *send_buf;
-	send_buf = this->u_send_buf_p; // Gather locally, must send
-	
-	////////////////////////////////////////////////////////
-	// Gather locally
-	////////////////////////////////////////////////////////
-	assert(send_buf!=NULL);
-	if ( comms_send ) 
-	  Gather_plane_simple_table(face_table[face_idx],rhs,send_buf,compress,u_comm_offset,so);
-	face_idx++;
-
-	int duplicate = CheckForDuplicate(dimension,sx,comm_proc,(void *)&recv_buf[u_comm_offset],0,bytes,cbmask);
+	int duplicate = CheckForDuplicate(dimension,sx,comm_proc,(void *)&recv_buf[comm_off],0,bytes,cbmask);
 	if ( (!duplicate) ) { // Force comms for now
 
 	  ///////////////////////////////////////////////////////////
 	  // Build a list of things to do after we synchronise GPUs
 	  // Start comms now???
 	  ///////////////////////////////////////////////////////////
-	  AddPacket((void *)&send_buf[u_comm_offset],
-		    (void *)&recv_buf[u_comm_offset],
+	  AddPacket((void *)&send_buf[comm_off],
+		    (void *)&recv_buf[comm_off],
 		    xmit_to_rank, comms_send,
 		    recv_from_rank, comms_recv,
 		    bytes);
 	}
 	
-	if ( compress.DecompressionStep()  ) {
-	  AddDecompress(&this->u_recv_buf_p[u_comm_offset],
-			&recv_buf[u_comm_offset],
+	if ( compress.DecompressionStep()  && comms_recv ) {
+	  AddDecompress(&this->u_recv_buf_p[comm_off],
+			&recv_buf[comm_off],
 			words,Decompressions);
 	}
+	
 	u_comm_offset+=words;
-	}
+	face_idx++;
+
+      }
     }
     return 0;
   }
@@ -1155,7 +1161,6 @@ public:
 
 
     int permute_type=_grid->PermuteType(dimension);
-    //    std::cout << "SimdNew permute type "<<permute_type<<std::endl;
 
     ///////////////////////////////////////////////
     // Simd direction uses an extract/merge pair
@@ -1189,8 +1194,9 @@ public:
 
       if ( any_offnode ) {
 
+	int comm_off = u_comm_offset;
 	for(int i=0;i<maxl;i++){
-	  spointers[i] = (cobj *) &u_simd_send_buf[i][u_comm_offset];
+	  spointers[i] = (cobj *) &u_simd_send_buf[i][comm_off];
 	}
 
 	int sx   = (x+sshift)%rd;
@@ -1199,15 +1205,15 @@ public:
 	  face_table.resize(face_idx+1);
 	  std::vector<std::pair<int,int> >  face_table_host ;
 				
-	  Gather_plane_table_compute ((GridBase *)_grid,dimension,sx,cbmask,u_comm_offset,face_table_host);
+	  Gather_plane_table_compute ((GridBase *)_grid,dimension,sx,cbmask,comm_off,face_table_host);
 	  face_table[face_idx].resize(face_table_host.size());
 	  acceleratorCopyToDevice(&face_table_host[0],
 				  &face_table[face_idx][0],
 				  face_table[face_idx].size()*sizeof(face_table_host[0]));
 	}
 
-	//	if ( comms_send )
-	Gather_plane_exchange_table(face_table[face_idx],rhs,spointers,dimension,sx,cbmask,compress,permute_type);
+	if ( comms_send || comms_recv )
+	  Gather_plane_exchange_table(face_table[face_idx],rhs,spointers,dimension,sx,cbmask,compress,permute_type);
 	face_idx++;
 
 	//spointers[0] -- low
@@ -1226,8 +1232,8 @@ public:
 	  int nbr_plane = nbr_ic;
 	  assert (sx == nbr_ox);
 
-	  auto rp = &u_simd_recv_buf[i        ][u_comm_offset];
-	  auto sp = &u_simd_send_buf[nbr_plane][u_comm_offset];
+	  auto rp = &u_simd_recv_buf[i        ][comm_off];
+	  auto sp = &u_simd_send_buf[nbr_plane][comm_off];
 
 	  if(nbr_proc){
 
@@ -1253,9 +1259,12 @@ public:
 	  }
 	}
 
-	AddMerge(&this->u_recv_buf_p[u_comm_offset],rpointers,reduced_buffer_size,permute_type,Mergers);
+	if ( comms_recv ) {
+	  AddMerge(&this->u_recv_buf_p[comm_off],rpointers,reduced_buffer_size,permute_type,Mergers);
+	}
 
 	u_comm_offset     +=buffer_size;
+
       }
     }
     return 0;

Grid/tensors/Tensor_extract_merge.h

@@ -208,5 +208,46 @@ void merge(vobj &vec,const ExtractPointerArray<sobj> &extracted, int offset)
 }
 
 
+
+//////////////////////////////////////////////////////////////////////////////////
+//Copy a single lane of a SIMD tensor type from one object to another
+//Output object must be of the same tensor type but may be of a different precision (i.e. it can have a different root data type)
+///////////////////////////////////////////////////////////////////////////////////
+template<class vobjOut, class vobjIn>
+accelerator_inline 
+void copyLane(vobjOut & __restrict__ vecOut, int lane_out, const vobjIn & __restrict__ vecIn, int lane_in)
+{
+  static_assert( std::is_same<typename vobjOut::DoublePrecision, typename vobjIn::DoublePrecision>::value == 1, "copyLane: tensor types must be the same" ); //if tensor types are same the DoublePrecision type must be the same
+
+  typedef typename vobjOut::vector_type ovector_type;  
+  typedef typename vobjIn::vector_type ivector_type;  
+  constexpr int owords=sizeof(vobjOut)/sizeof(ovector_type);
+  constexpr int iwords=sizeof(vobjIn)/sizeof(ivector_type);
+  static_assert( owords == iwords, "copyLane: Expected number of vector words in input and output objects to be equal" );
+
+  typedef typename vobjOut::scalar_type oscalar_type;  
+  typedef typename vobjIn::scalar_type iscalar_type;  
+  typedef typename ExtractTypeMap<oscalar_type>::extract_type oextract_type;
+  typedef typename ExtractTypeMap<iscalar_type>::extract_type iextract_type;
+
+  typedef oextract_type * opointer;
+  typedef iextract_type * ipointer;
+
+  constexpr int oNsimd=ovector_type::Nsimd();
+  constexpr int iNsimd=ivector_type::Nsimd();
+
+  iscalar_type itmp;
+  oscalar_type otmp;
+
+  opointer __restrict__  op = (opointer)&vecOut;
+  ipointer __restrict__  ip = (ipointer)&vecIn;
+  for(int w=0;w<owords;w++){
+    memcpy( (char*)&itmp, (char*)(ip + lane_in + iNsimd*w), sizeof(iscalar_type) );
+    otmp = itmp; //potential precision change
+    memcpy( (char*)(op + lane_out + oNsimd*w), (char*)&otmp, sizeof(oscalar_type) );
+  }
+}
+
+
 NAMESPACE_END(Grid);
 

Grid/threads/Accelerator.cc

@@ -6,9 +6,17 @@ uint32_t accelerator_threads=2;
 uint32_t acceleratorThreads(void)       {return accelerator_threads;};
 void     acceleratorThreads(uint32_t t) {accelerator_threads = t;};
 
+#define ENV_LOCAL_RANK_OMPI    "OMPI_COMM_WORLD_LOCAL_RANK"
+#define ENV_RANK_OMPI          "OMPI_COMM_WORLD_RANK"
+#define ENV_LOCAL_RANK_SLURM   "SLURM_LOCALID"
+#define ENV_RANK_SLURM         "SLURM_PROCID"
+#define ENV_LOCAL_RANK_MVAPICH "MV2_COMM_WORLD_LOCAL_RANK"
+#define ENV_RANK_MVAPICH       "MV2_COMM_WORLD_RANK"
+
 #ifdef GRID_CUDA
 cudaDeviceProp *gpu_props;
 cudaStream_t copyStream;
+cudaStream_t cpuStream;
 void acceleratorInit(void)
 {
   int nDevices = 1;
@@ -17,12 +25,6 @@ void acceleratorInit(void)
 
   char * localRankStr = NULL;
   int rank = 0, world_rank=0; 
-#define ENV_LOCAL_RANK_OMPI    "OMPI_COMM_WORLD_LOCAL_RANK"
-#define ENV_RANK_OMPI          "OMPI_COMM_WORLD_RANK"
-#define ENV_LOCAL_RANK_SLURM   "SLURM_LOCALID"
-#define ENV_RANK_SLURM         "SLURM_PROCID"
-#define ENV_LOCAL_RANK_MVAPICH "MV2_COMM_WORLD_LOCAL_RANK"
-#define ENV_RANK_MVAPICH       "MV2_COMM_WORLD_RANK"
   if ((localRankStr = getenv(ENV_RANK_OMPI   )) != NULL) { world_rank = atoi(localRankStr);}
   if ((localRankStr = getenv(ENV_RANK_MVAPICH)) != NULL) { world_rank = atoi(localRankStr);}
   if ((localRankStr = getenv(ENV_RANK_SLURM  )) != NULL) { world_rank = atoi(localRankStr);}
@@ -97,6 +99,7 @@ void acceleratorInit(void)
 
   cudaSetDevice(device);
   cudaStreamCreate(&copyStream);
+  cudaStreamCreate(&cpuStream);
   const int len=64;
   char busid[len];
   if( rank == world_rank ) { 
@@ -111,6 +114,7 @@ void acceleratorInit(void)
 #ifdef GRID_HIP
 hipDeviceProp_t *gpu_props;
 hipStream_t copyStream;
+hipStream_t cpuStream;
 void acceleratorInit(void)
 {
   int nDevices = 1;
@@ -119,10 +123,6 @@ void acceleratorInit(void)
 
   char * localRankStr = NULL;
   int rank = 0, world_rank=0; 
-#define ENV_LOCAL_RANK_OMPI    "OMPI_COMM_WORLD_LOCAL_RANK"
-#define ENV_LOCAL_RANK_MVAPICH "MV2_COMM_WORLD_LOCAL_RANK"
-#define ENV_RANK_OMPI          "OMPI_COMM_WORLD_RANK"
-#define ENV_RANK_MVAPICH       "MV2_COMM_WORLD_RANK"
   // We extract the local rank initialization using an environment variable
   if ((localRankStr = getenv(ENV_LOCAL_RANK_OMPI)) != NULL)
   {
@@ -134,8 +134,10 @@ void acceleratorInit(void)
   }
   if ((localRankStr = getenv(ENV_RANK_OMPI   )) != NULL) { world_rank = atoi(localRankStr);}
   if ((localRankStr = getenv(ENV_RANK_MVAPICH)) != NULL) { world_rank = atoi(localRankStr);}
+  if ((localRankStr = getenv(ENV_RANK_SLURM  )) != NULL) { world_rank = atoi(localRankStr);}
 
-  printf("world_rank %d has %d devices\n",world_rank,nDevices);
+  if ( world_rank == 0 ) 
+    printf("world_rank %d has %d devices\n",world_rank,nDevices);
   size_t totalDeviceMem=0;
   for (int i = 0; i < nDevices; i++) {
 
@@ -181,6 +183,7 @@ void acceleratorInit(void)
 #endif
   hipSetDevice(device);
   hipStreamCreate(&copyStream);
+  hipStreamCreate(&cpuStream);
   const int len=64;
   char busid[len];
   if( rank == world_rank ) { 
@@ -208,10 +211,7 @@ void acceleratorInit(void)
   
   char * localRankStr = NULL;
   int rank = 0, world_rank=0; 
-#define ENV_LOCAL_RANK_OMPI    "OMPI_COMM_WORLD_LOCAL_RANK"
-#define ENV_LOCAL_RANK_MVAPICH "MV2_COMM_WORLD_LOCAL_RANK"
-#define ENV_RANK_OMPI          "OMPI_COMM_WORLD_RANK"
-#define ENV_RANK_MVAPICH       "MV2_COMM_WORLD_RANK"
+
   // We extract the local rank initialization using an environment variable
   if ((localRankStr = getenv(ENV_LOCAL_RANK_OMPI)) != NULL)
   {

Grid/threads/Accelerator.h

@@ -107,6 +107,7 @@ void     acceleratorInit(void);
 
 extern int acceleratorAbortOnGpuError;
 extern cudaStream_t copyStream;
+extern cudaStream_t cpuStream;
 
 accelerator_inline int acceleratorSIMTlane(int Nsimd) {
 #ifdef GRID_SIMT
@@ -134,7 +135,7 @@ inline void cuda_mem(void)
     };									\
     dim3 cu_threads(nsimd,acceleratorThreads(),1);			\
     dim3 cu_blocks ((num1+nt-1)/nt,num2,1);				\
-    LambdaApply<<<cu_blocks,cu_threads>>>(num1,num2,nsimd,lambda);	\
+    LambdaApply<<<cu_blocks,cu_threads,0,cpuStream>>>(num1,num2,nsimd,lambda);	\
   }
 
 #define accelerator_for6dNB(iter1, num1,				\
@@ -153,7 +154,7 @@ inline void cuda_mem(void)
     };									\
     dim3 cu_blocks (num1,num2,num3);					\
     dim3 cu_threads(num4,num5,num6);					\
-    Lambda6Apply<<<cu_blocks,cu_threads>>>(num1,num2,num3,num4,num5,num6,lambda); \
+    Lambda6Apply<<<cu_blocks,cu_threads,0,cpuStream>>>(num1,num2,num3,num4,num5,num6,lambda); \
   }
 
 template<typename lambda>  __global__
@@ -189,7 +190,7 @@ void Lambda6Apply(uint64_t num1, uint64_t num2, uint64_t num3,
 
 #define accelerator_barrier(dummy)					\
   {									\
-    cudaDeviceSynchronize();						\
+    cudaStreamSynchronize(cpuStream);					\
     cudaError err = cudaGetLastError();					\
     if ( cudaSuccess != err ) {						\
       printf("accelerator_barrier(): Cuda error %s \n",			\
@@ -339,6 +340,7 @@ NAMESPACE_BEGIN(Grid);
 #define accelerator_inline __host__ __device__ inline
 
 extern hipStream_t copyStream;
+extern hipStream_t cpuStream;
 /*These routines define mapping from thread grid to loop & vector lane indexing */
 accelerator_inline int acceleratorSIMTlane(int Nsimd) {
 #ifdef GRID_SIMT
@@ -360,12 +362,12 @@ accelerator_inline int acceleratorSIMTlane(int Nsimd) {
     dim3 hip_blocks ((num1+nt-1)/nt,num2,1); \
     if(hip_threads.x * hip_threads.y * hip_threads.z <= 64){ \
       hipLaunchKernelGGL(LambdaApply64,hip_blocks,hip_threads,		\
-            0,0,						\
-            num1,num2,nsimd, lambda);				\
+			 0,cpuStream,					\
+			 num1,num2,nsimd, lambda);			\
     } else { \
       hipLaunchKernelGGL(LambdaApply,hip_blocks,hip_threads,		\
-            0,0,						\
-            num1,num2,nsimd, lambda);				\
+			 0,cpuStream,					\
+			 num1,num2,nsimd, lambda);			\
     } \
   }
 
@@ -398,7 +400,7 @@ void LambdaApply(uint64_t numx, uint64_t numy, uint64_t numz, lambda Lambda)
 
 #define accelerator_barrier(dummy)				\
   {								\
-    hipDeviceSynchronize();					\
+    hipStreamSynchronize(cpuStream);					\
     auto err = hipGetLastError();				\
     if ( err != hipSuccess ) {					\
       printf("After hipDeviceSynchronize() : HIP error %s \n", hipGetErrorString( err )); \

HMC/Mobius2p1fIDSDRGparityEOFA_40ID.cc

@@ -0,0 +1,918 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid
+
+Source file: ./HMC/Mobius2p1fIDSDRGparityEOFA.cc
+
+Copyright (C) 2015-2016
+
+Author: Christopher Kelly <ckelly@bnl.gov>
+Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
+
+
+This program is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 2 of the License, or
+(at your option) any later version.
+
+This program is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along
+with this program; if not, write to the Free Software Foundation, Inc.,
+51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+See the full license in the file "LICENSE" in the top level distribution
+directory
+*************************************************************************************/
+/*  END LEGAL */
+#include <Grid/Grid.h>
+
+using namespace Grid;
+
+//Production binary for the 40ID G-parity ensemble
+
+struct RatQuoParameters: Serializable {
+  GRID_SERIALIZABLE_CLASS_MEMBERS(RatQuoParameters,
+				  double, bnd_lo,
+				  double, bnd_hi,
+				  Integer, action_degree,
+				  double, action_tolerance,
+				  Integer, md_degree,
+				  double, md_tolerance,
+				  Integer, reliable_update_freq,
+				  Integer, bnd_check_freq);
+  RatQuoParameters() { 
+    bnd_lo = 1e-2;
+    bnd_hi = 30;
+    action_degree = 10;
+    action_tolerance = 1e-10;
+    md_degree = 10;
+    md_tolerance = 1e-8;
+    bnd_check_freq = 20;
+    reliable_update_freq = 50;
+  }
+
+  void Export(RationalActionParams &into) const{
+    into.lo = bnd_lo;
+    into.hi = bnd_hi;
+    into.action_degree = action_degree;
+    into.action_tolerance = action_tolerance;
+    into.md_degree = md_degree;
+    into.md_tolerance = md_tolerance;
+    into.BoundsCheckFreq = bnd_check_freq;
+  }
+};
+
+struct EOFAparameters: Serializable {
+  GRID_SERIALIZABLE_CLASS_MEMBERS(EOFAparameters,
+				  OneFlavourRationalParams, rat_params,
+				  double, action_tolerance,
+				  double, action_mixcg_inner_tolerance,
+				  double, md_tolerance,
+				  double, md_mixcg_inner_tolerance);
+
+  EOFAparameters() { 
+    action_mixcg_inner_tolerance = 1e-8;
+    action_tolerance = 1e-10;
+    md_tolerance = 1e-8;
+    md_mixcg_inner_tolerance = 1e-8;
+
+    rat_params.lo = 1.0;
+    rat_params.hi = 25.0;
+    rat_params.MaxIter  = 50000;
+    rat_params.tolerance= 1.0e-9;
+    rat_params.degree   = 14;
+    rat_params.precision= 50;
+  }
+};
+
+struct EvolParameters: Serializable {
+  GRID_SERIALIZABLE_CLASS_MEMBERS(EvolParameters,
+                                  Integer, StartTrajectory,
+                                  Integer, Trajectories,
+				  Integer, SaveInterval,
+				  Integer, Steps,
+				  RealD, TrajectoryLength,
+                                  bool, MetropolisTest,
+				  std::string, StartingType,
+				  std::vector<Integer>, GparityDirs,
+				  std::vector<EOFAparameters>, eofa_l,
+				  RatQuoParameters, rat_quo_s,
+				  RatQuoParameters, rat_quo_DSDR);
+
+  EvolParameters() {
+    //For initial thermalization; afterwards user should switch Metropolis on and use StartingType=CheckpointStart
+    MetropolisTest    = false;
+    StartTrajectory   = 0;
+    Trajectories      = 50;
+    SaveInterval = 5;
+    StartingType      = "ColdStart";
+    GparityDirs.resize(3, 1); //1 for G-parity, 0 for periodic
+    Steps = 5;
+    TrajectoryLength = 1.0;
+  }
+};
+
+bool fileExists(const std::string &fn){
+  std::ifstream f(fn);
+  return f.good();
+}
+
+
+
+
+struct LanczosParameters: Serializable {
+  GRID_SERIALIZABLE_CLASS_MEMBERS(LanczosParameters,
+				  double, alpha,
+				  double, beta,
+				  double, mu,
+				  int, ord,
+				  int, n_stop,
+				  int, n_want,
+				  int, n_use,
+				  double, tolerance);
+
+  LanczosParameters() {
+    alpha = 35;
+    beta = 5;
+    mu = 0;
+    ord = 100;
+    n_stop = 10;
+    n_want = 10;
+    n_use = 15;
+    tolerance = 1e-6;
+  }
+};
+
+
+
+template<typename FermionActionD, typename FermionFieldD>
+void computeEigenvalues(std::string param_file,
+			GridCartesian* Grid, GridRedBlackCartesian* rbGrid, const LatticeGaugeFieldD &latt,  //expect lattice to have been initialized to something
+			FermionActionD &action, GridParallelRNG &rng){
+  
+  LanczosParameters params;
+  if(fileExists(param_file)){
+    std::cout << GridLogMessage << " Reading " << param_file << std::endl;
+    Grid::XmlReader rd(param_file);
+    read(rd, "LanczosParameters", params);
+  }else if(!GlobalSharedMemory::WorldRank){
+    std::cout << GridLogMessage << " File " << param_file << " does not exist" << std::endl;
+    std::cout << GridLogMessage << " Writing xml template to " << param_file << ".templ" << std::endl;
+    Grid::XmlWriter wr(param_file + ".templ");
+    write(wr, "LanczosParameters", params);
+  }
+
+  FermionFieldD gauss_o(rbGrid);
+  FermionFieldD gauss(Grid);
+  gaussian(rng, gauss);
+  pickCheckerboard(Odd, gauss_o, gauss);
+
+  action.ImportGauge(latt);
+
+  SchurDiagMooeeOperator<FermionActionD, FermionFieldD> hermop(action);
+  PlainHermOp<FermionFieldD> hermop_wrap(hermop);
+  //ChebyshevLanczos<FermionFieldD> Cheb(params.alpha, params.beta, params.mu, params.ord);
+  assert(params.mu == 0.0);
+
+  Chebyshev<FermionFieldD> Cheb(params.beta*params.beta, params.alpha*params.alpha, params.ord+1);
+  FunctionHermOp<FermionFieldD> Cheb_wrap(Cheb, hermop);
+
+  std::cout << "IRL: alpha=" << params.alpha << " beta=" << params.beta << " mu=" << params.mu << " ord=" << params.ord << std::endl;
+  ImplicitlyRestartedLanczos<FermionFieldD> IRL(Cheb_wrap, hermop_wrap, params.n_stop, params.n_want, params.n_use, params.tolerance, 50000);
+
+  std::vector<RealD> eval(params.n_use);
+  std::vector<FermionFieldD> evec(params.n_use, rbGrid);
+  int Nconv;
+  IRL.calc(eval, evec, gauss_o, Nconv);
+
+  std::cout << "Eigenvalues:" << std::endl;
+  for(int i=0;i<params.n_want;i++){
+    std::cout << i << " " << eval[i] << std::endl;
+  }
+}
+
+
+//Check the quality of the RHMC approx
+//action_or_md toggles checking the action (0), MD (1) or both (2) setups
+template<typename FermionActionD, typename FermionFieldD, typename RHMCtype>
+void checkRHMC(GridCartesian* Grid, GridRedBlackCartesian* rbGrid, const LatticeGaugeFieldD &latt,  //expect lattice to have been initialized to something
+	       FermionActionD &numOp, FermionActionD &denOp, RHMCtype &rhmc, GridParallelRNG &rng,
+	       int inv_pow, const std::string &quark_descr, int action_or_md){
+  assert(action_or_md == 0 || action_or_md == 1 || action_or_md == 2);
+  
+  FermionFieldD gauss_o(rbGrid);
+  FermionFieldD gauss(Grid);
+  gaussian(rng, gauss);
+  pickCheckerboard(Odd, gauss_o, gauss);
+
+  numOp.ImportGauge(latt);
+  denOp.ImportGauge(latt);
+
+  typedef typename FermionActionD::Impl_t FermionImplPolicyD;
+  SchurDifferentiableOperator<FermionImplPolicyD> MdagM(numOp);
+  SchurDifferentiableOperator<FermionImplPolicyD> VdagV(denOp);
+
+  PowerMethod<FermionFieldD> power_method;
+  RealD lambda_max;
+
+  std::cout << "Starting: Get RHMC high bound approx for " << quark_descr << " numerator" << std::endl;
+
+  lambda_max = power_method(MdagM,gauss_o);
+  std::cout << GridLogMessage << "Got lambda_max "<<lambda_max<<std::endl;
+
+  std::cout << "Starting: Get RHMC high bound approx for " << quark_descr << " denominator" << std::endl;
+  lambda_max = power_method(VdagV,gauss_o);
+  std::cout << GridLogMessage << "Got lambda_max "<<lambda_max<<std::endl;
+
+  if(action_or_md == 0 || action_or_md == 2){
+    std::cout << "Starting: Checking quality of RHMC action approx for " << quark_descr << " quark numerator and power -1/" << inv_pow << std::endl;
+    InversePowerBoundsCheck(inv_pow, 50000, 1e16, MdagM,gauss_o, rhmc.ApproxNegPowerAction); //use large tolerance to prevent exit on fail; we are trying to tune here!
+    std::cout << "Finished: Checking quality of RHMC action approx for " << quark_descr << " quark numerator and power -1/" << inv_pow << std::endl;
+
+    std::cout << "Starting: Checking quality of RHMC action approx for " << quark_descr << " quark numerator and power -1/" << 2*inv_pow << std::endl;
+    InversePowerBoundsCheck(2*inv_pow, 50000, 1e16, MdagM,gauss_o, rhmc.ApproxNegHalfPowerAction);
+    std::cout << "Finished: Checking quality of RHMC action approx for " << quark_descr << " quark numerator and power -1/" << 2*inv_pow << std::endl;
+
+    std::cout << "Starting: Checking quality of RHMC action approx for " << quark_descr << " quark denominator and power -1/" << inv_pow << std::endl;
+    InversePowerBoundsCheck(inv_pow, 50000, 1e16, VdagV,gauss_o, rhmc.ApproxNegPowerAction);
+    std::cout << "Finished: Checking quality of RHMC action approx for " << quark_descr << " quark denominator and power -1/" << inv_pow << std::endl;
+
+    std::cout << "Starting: Checking quality of RHMC action approx for " << quark_descr << " quark denominator and power -1/" << 2*inv_pow << std::endl;
+    InversePowerBoundsCheck(2*inv_pow, 50000, 1e16, VdagV,gauss_o, rhmc.ApproxNegHalfPowerAction);
+    std::cout << "Finished: Checking quality of RHMC action approx for " << quark_descr << " quark denominator and power -1/" << 2*inv_pow << std::endl;
+  }
+
+  std::cout << "-------------------------------------------------------------------------------" << std::endl;
+
+  if(action_or_md == 1 || action_or_md == 2){
+    std::cout << "Starting: Checking quality of RHMC MD approx for " << quark_descr << " quark numerator and power -1/" << inv_pow << std::endl;
+    InversePowerBoundsCheck(inv_pow, 50000, 1e16, MdagM,gauss_o, rhmc.ApproxNegPowerMD); 
+    std::cout << "Finished: Checking quality of RHMC MD approx for " << quark_descr << " quark numerator and power -1/" << inv_pow << std::endl;
+
+    std::cout << "Starting: Checking quality of RHMC MD approx for " << quark_descr << " quark numerator and power -1/" << 2*inv_pow << std::endl;
+    InversePowerBoundsCheck(2*inv_pow, 50000, 1e16, MdagM,gauss_o, rhmc.ApproxNegHalfPowerMD);
+    std::cout << "Finished: Checking quality of RHMC MD approx for " << quark_descr << " quark numerator and power -1/" << 2*inv_pow << std::endl;
+
+    std::cout << "Starting: Checking quality of RHMC MD approx for " << quark_descr << " quark denominator and power -1/" << inv_pow << std::endl;
+    InversePowerBoundsCheck(inv_pow, 50000, 1e16, VdagV,gauss_o, rhmc.ApproxNegPowerMD);
+    std::cout << "Finished: Checking quality of RHMC MD approx for " << quark_descr << " quark denominator and power -1/" << inv_pow << std::endl;
+
+    std::cout << "Starting: Checking quality of RHMC MD approx for " << quark_descr << " quark denominator and power -1/" << 2*inv_pow << std::endl;
+    InversePowerBoundsCheck(2*inv_pow, 50000, 1e16, VdagV,gauss_o, rhmc.ApproxNegHalfPowerMD);
+    std::cout << "Finished: Checking quality of RHMC MD approx for " << quark_descr << " quark denominator and power -1/" << 2*inv_pow << std::endl;
+  }
+}
+
+
+template<typename FermionImplPolicy>
+void checkEOFA(ExactOneFlavourRatioPseudoFermionAction<FermionImplPolicy> &EOFA,
+	       GridCartesian* FGrid, GridParallelRNG &rng, const LatticeGaugeFieldD &latt){
+  std::cout << GridLogMessage << "Starting EOFA action/bounds check" << std::endl;
+  typename FermionImplPolicy::FermionField eta(FGrid);
+  RealD scale = std::sqrt(0.5);
+  gaussian(rng,eta); eta = eta * scale;
+
+  //Use the inbuilt check
+  EOFA.refresh(latt, eta);
+  EOFA.S(latt);
+  std::cout << GridLogMessage << "Finished EOFA upper action/bounds check" << std::endl;
+}
+
+
+template<typename FermionImplPolicy>
+class EOFAlinop: public LinearOperatorBase<typename FermionImplPolicy::FermionField>{
+  ExactOneFlavourRatioPseudoFermionAction<FermionImplPolicy> &EOFA;
+  LatticeGaugeFieldD &U;
+public:
+  EOFAlinop(ExactOneFlavourRatioPseudoFermionAction<FermionImplPolicy> &EOFA, LatticeGaugeFieldD &U): EOFA(EOFA), U(U){}
+
+  typedef typename FermionImplPolicy::FermionField Field;
+  void OpDiag (const Field &in, Field &out){ assert(0); }
+  void OpDir  (const Field &in, Field &out,int dir,int disp){ assert(0); }
+  void OpDirAll  (const Field &in, std::vector<Field> &out){ assert(0); } 
+
+  void Op     (const Field &in, Field &out){ assert(0); }
+  void AdjOp  (const Field &in, Field &out){ assert(0); }
+  void HermOpAndNorm(const Field &in, Field &out,RealD &n1,RealD &n2){ assert(0); }
+  void HermOp(const Field &in, Field &out){ EOFA.Meofa(U, in, out); }
+};
+
+template<typename FermionImplPolicy>
+void upperBoundEOFA(ExactOneFlavourRatioPseudoFermionAction<FermionImplPolicy> &EOFA,
+		    GridCartesian* FGrid, GridParallelRNG &rng, LatticeGaugeFieldD &latt){
+  std::cout << GridLogMessage << "Starting EOFA upper bound compute" << std::endl;
+  EOFAlinop<FermionImplPolicy> linop(EOFA, latt);
+  typename FermionImplPolicy::FermionField eta(FGrid);
+  gaussian(rng,eta);
+  PowerMethod<typename FermionImplPolicy::FermionField> power_method;
+  auto lambda_max = power_method(linop,eta);
+  std::cout << GridLogMessage << "Upper bound of EOFA operator " << lambda_max << std::endl;
+}
+
+//Applications of M^{-1} cost the same as M for EOFA!
+template<typename FermionImplPolicy>
+class EOFAinvLinop: public LinearOperatorBase<typename FermionImplPolicy::FermionField>{
+  ExactOneFlavourRatioPseudoFermionAction<FermionImplPolicy> &EOFA;
+  LatticeGaugeFieldD &U;
+public:
+  EOFAinvLinop(ExactOneFlavourRatioPseudoFermionAction<FermionImplPolicy> &EOFA, LatticeGaugeFieldD &U): EOFA(EOFA), U(U){}
+
+  typedef typename FermionImplPolicy::FermionField Field;
+  void OpDiag (const Field &in, Field &out){ assert(0); }
+  void OpDir  (const Field &in, Field &out,int dir,int disp){ assert(0); }
+  void OpDirAll  (const Field &in, std::vector<Field> &out){ assert(0); } 
+
+  void Op     (const Field &in, Field &out){ assert(0); }
+  void AdjOp  (const Field &in, Field &out){ assert(0); }
+  void HermOpAndNorm(const Field &in, Field &out,RealD &n1,RealD &n2){ assert(0); }
+  void HermOp(const Field &in, Field &out){ EOFA.MeofaInv(U, in, out); }
+};
+
+template<typename FermionImplPolicy>
+void lowerBoundEOFA(ExactOneFlavourRatioPseudoFermionAction<FermionImplPolicy> &EOFA,
+		    GridCartesian* FGrid, GridParallelRNG &rng, LatticeGaugeFieldD &latt){
+  std::cout << GridLogMessage << "Starting EOFA lower bound compute using power method on M^{-1}. Inverse of highest eigenvalue is the lowest eigenvalue of M" << std::endl;
+  EOFAinvLinop<FermionImplPolicy> linop(EOFA, latt);
+  typename FermionImplPolicy::FermionField eta(FGrid);
+  gaussian(rng,eta);
+  PowerMethod<typename FermionImplPolicy::FermionField> power_method;
+  auto lambda_max = power_method(linop,eta);
+  std::cout << GridLogMessage << "Lower bound of EOFA operator " << 1./lambda_max << std::endl;
+}
+
+
+NAMESPACE_BEGIN(Grid);
+
+  template<class FermionOperatorD, class FermionOperatorF, class SchurOperatorD, class  SchurOperatorF> 
+  class MixedPrecisionConjugateGradientOperatorFunction : public OperatorFunction<typename FermionOperatorD::FermionField> {
+  public:
+    typedef typename FermionOperatorD::FermionField FieldD;
+    typedef typename FermionOperatorF::FermionField FieldF;
+
+    using OperatorFunction<FieldD>::operator();
+
+    RealD   Tolerance;
+    RealD   InnerTolerance; //Initial tolerance for inner CG. Defaults to Tolerance but can be changed
+    Integer MaxInnerIterations;
+    Integer MaxOuterIterations;
+    GridBase* SinglePrecGrid4; //Grid for single-precision fields
+    GridBase* SinglePrecGrid5; //Grid for single-precision fields
+    RealD OuterLoopNormMult; //Stop the outer loop and move to a final double prec solve when the residual is OuterLoopNormMult * Tolerance
+
+    FermionOperatorF &FermOpF;
+    FermionOperatorD &FermOpD;;
+    SchurOperatorF &LinOpF;
+    SchurOperatorD &LinOpD;
+
+    Integer TotalInnerIterations; //Number of inner CG iterations
+    Integer TotalOuterIterations; //Number of restarts
+    Integer TotalFinalStepIterations; //Number of CG iterations in final patch-up step
+
+    MixedPrecisionConjugateGradientOperatorFunction(RealD tol, 
+						    Integer maxinnerit, 
+						    Integer maxouterit, 
+						    GridBase* _sp_grid4, 
+						    GridBase* _sp_grid5, 
+						    FermionOperatorF &_FermOpF,
+						    FermionOperatorD &_FermOpD,
+						    SchurOperatorF   &_LinOpF,
+						    SchurOperatorD   &_LinOpD): 
+      LinOpF(_LinOpF),
+      LinOpD(_LinOpD),
+      FermOpF(_FermOpF),
+      FermOpD(_FermOpD),
+      Tolerance(tol), 
+      InnerTolerance(tol), 
+      MaxInnerIterations(maxinnerit), 
+      MaxOuterIterations(maxouterit), 
+      SinglePrecGrid4(_sp_grid4),
+      SinglePrecGrid5(_sp_grid5),
+      OuterLoopNormMult(100.) 
+    { 
+    };
+
+    void operator()(LinearOperatorBase<FieldD> &LinOpU, const FieldD &src, FieldD &psi) {
+
+      std::cout << GridLogMessage << " Mixed precision CG wrapper operator() "<<std::endl;
+
+      SchurOperatorD * SchurOpU = static_cast<SchurOperatorD *>(&LinOpU);
+      assert(&(SchurOpU->_Mat)==&(LinOpD._Mat));
+
+      precisionChange(FermOpF.Umu, FermOpD.Umu);
+
+      pickCheckerboard(Even,FermOpF.UmuEven,FermOpF.Umu);
+      pickCheckerboard(Odd ,FermOpF.UmuOdd ,FermOpF.Umu);
+
+      ////////////////////////////////////////////////////////////////////////////////////
+      // Make a mixed precision conjugate gradient
+      ////////////////////////////////////////////////////////////////////////////////////
+      MixedPrecisionConjugateGradient<FieldD,FieldF> MPCG(Tolerance,MaxInnerIterations,MaxOuterIterations,SinglePrecGrid5,LinOpF,LinOpD);
+      MPCG.InnerTolerance = InnerTolerance;
+      std::cout << GridLogMessage << "Calling mixed precision Conjugate Gradient" <<std::endl;
+      MPCG(src,psi);
+    }
+  };
+
+
+
+  template<class FermionOperatorD, class FermionOperatorF, class SchurOperatorD, class  SchurOperatorF> 
+  class MixedPrecisionReliableUpdateConjugateGradientOperatorFunction : public OperatorFunction<typename FermionOperatorD::FermionField> {
+  public:
+    typedef typename FermionOperatorD::FermionField FieldD;
+    typedef typename FermionOperatorF::FermionField FieldF;
+
+    using OperatorFunction<FieldD>::operator();
+
+    RealD Tolerance;
+    Integer MaxIterations;
+
+    RealD Delta; //reliable update parameter
+
+    GridBase* SinglePrecGrid4; //Grid for single-precision fields
+    GridBase* SinglePrecGrid5; //Grid for single-precision fields
+
+    FermionOperatorF &FermOpF;
+    FermionOperatorD &FermOpD;;
+    SchurOperatorF &LinOpF;
+    SchurOperatorD &LinOpD;
+    
+    MixedPrecisionReliableUpdateConjugateGradientOperatorFunction(RealD tol, 
+								  RealD delta,
+								  Integer maxit, 
+								  GridBase* _sp_grid4, 
+								  GridBase* _sp_grid5, 
+								  FermionOperatorF &_FermOpF,
+								  FermionOperatorD &_FermOpD,
+								  SchurOperatorF   &_LinOpF,
+								  SchurOperatorD   &_LinOpD): 
+      LinOpF(_LinOpF),
+      LinOpD(_LinOpD),
+      FermOpF(_FermOpF),
+      FermOpD(_FermOpD),
+      Tolerance(tol), 
+      Delta(delta),
+      MaxIterations(maxit), 
+      SinglePrecGrid4(_sp_grid4),
+      SinglePrecGrid5(_sp_grid5)
+    { 
+    };
+
+    void operator()(LinearOperatorBase<FieldD> &LinOpU, const FieldD &src, FieldD &psi) {
+
+      std::cout << GridLogMessage << " Mixed precision reliable CG update wrapper operator() "<<std::endl;
+
+      SchurOperatorD * SchurOpU = static_cast<SchurOperatorD *>(&LinOpU);
+      assert(&(SchurOpU->_Mat)==&(LinOpD._Mat));
+
+      precisionChange(FermOpF.Umu, FermOpD.Umu);
+
+      pickCheckerboard(Even,FermOpF.UmuEven,FermOpF.Umu);
+      pickCheckerboard(Odd ,FermOpF.UmuOdd ,FermOpF.Umu);
+
+      ////////////////////////////////////////////////////////////////////////////////////
+      // Make a mixed precision conjugate gradient
+      ////////////////////////////////////////////////////////////////////////////////////
+
+      ConjugateGradientReliableUpdate<FieldD,FieldF> MPCG(Tolerance,MaxIterations,Delta,SinglePrecGrid5,LinOpF,LinOpD);
+      std::cout << GridLogMessage << "Calling mixed precision reliable update Conjugate Gradient" <<std::endl;
+      MPCG(src,psi);
+    }
+  };
+
+
+
+NAMESPACE_END(Grid);
+
+
+
+
+
+int main(int argc, char **argv) {
+  Grid_init(&argc, &argv);
+  int threads = GridThread::GetThreads();
+  // here make a routine to print all the relevant information on the run
+  std::cout << GridLogMessage << "Grid is setup to use " << threads << " threads" << std::endl;
+
+  std::string param_file = "params.xml";
+  bool file_load_check = false;
+
+  std::string serial_seeds = "1 2 3 4 5";
+  std::string parallel_seeds = "6 7 8 9 10";
+
+  int i=1;
+  while(i < argc){
+    std::string sarg(argv[i]);
+    if(sarg == "--param_file"){
+      assert(i!=argc-1);
+      param_file = argv[i+1];
+      i+=2;
+    }else if(sarg == "--read_check"){ //check the fields load correctly and pass checksum/plaquette repro
+      file_load_check = true;
+      i++;
+    }else if(sarg == "--set_seeds"){ //set the rng seeds. Expects two vector args, e.g.  --set_seeds 1.2.3.4 5.6.7.8
+      assert(i < argc-2);
+      std::vector<int> tmp;
+      GridCmdOptionIntVector(argv[i+1],tmp);
+      {
+	std::stringstream ss;
+	for(int j=0;j<tmp.size()-1;j++) ss << tmp[j] << " ";
+	ss << tmp.back();
+	serial_seeds = ss.str();
+      }
+      GridCmdOptionIntVector(argv[i+2],tmp);
+      {
+	std::stringstream ss;
+	for(int j=0;j<tmp.size()-1;j++) ss << tmp[j] << " ";
+	ss << tmp.back();
+	parallel_seeds = ss.str();
+      }
+      i+=3;
+      std::cout << GridLogMessage << "Set serial seeds to " << serial_seeds << std::endl;
+      std::cout << GridLogMessage << "Set parallel seeds to " << parallel_seeds << std::endl;
+      
+    }else{
+      i++;
+    }
+  }
+
+  
+  //Read the user parameters
+  EvolParameters user_params;
+  
+  if(fileExists(param_file)){
+    std::cout << GridLogMessage << " Reading " << param_file << std::endl;
+    Grid::XmlReader rd(param_file);
+    read(rd, "Params", user_params);
+  }else if(!GlobalSharedMemory::WorldRank){
+    std::cout << GridLogMessage << " File " << param_file << " does not exist" << std::endl;
+    std::cout << GridLogMessage << " Writing xml template to " << param_file << ".templ" << std::endl;
+    {
+      Grid::XmlWriter wr(param_file + ".templ");
+      write(wr, "Params", user_params);
+    }
+    std::cout << GridLogMessage << " Done" << std::endl;
+    Grid_finalize();
+    return 0;
+  }
+
+  //Check the parameters
+  if(user_params.GparityDirs.size() != Nd-1){
+    std::cerr << "Error in input parameters: expect GparityDirs to have size = " << Nd-1 << std::endl;
+    exit(1);
+  }
+  for(int i=0;i<Nd-1;i++)
+    if(user_params.GparityDirs[i] != 0 && user_params.GparityDirs[i] != 1){
+      std::cerr << "Error in input parameters: expect GparityDirs values to be 0 (periodic) or 1 (G-parity)" << std::endl;
+      exit(1);
+    }
+
+
+  typedef GparityMobiusEOFAFermionD EOFAactionD;
+  typedef GparityMobiusFermionD FermionActionD;
+  typedef typename FermionActionD::Impl_t FermionImplPolicyD;
+  typedef typename FermionActionD::FermionField FermionFieldD;
+
+  typedef GparityMobiusEOFAFermionF EOFAactionF;
+  typedef GparityMobiusFermionF FermionActionF;
+  typedef typename FermionActionF::Impl_t FermionImplPolicyF;
+  typedef typename FermionActionF::FermionField FermionFieldF;
+
+  typedef GeneralEvenOddRatioRationalMixedPrecPseudoFermionAction<FermionImplPolicyD,FermionImplPolicyF> MixedPrecRHMC;
+  typedef GeneralEvenOddRatioRationalPseudoFermionAction<FermionImplPolicyD> DoublePrecRHMC;
+
+  //::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
+  IntegratorParameters MD;
+  typedef ConjugateHMCRunnerD<MinimumNorm2> HMCWrapper; //NB: This is the "Omelyan integrator"
+  MD.name    = std::string("MinimumNorm2");
+
+  // typedef ConjugateHMCRunnerD<ForceGradient> HMCWrapper;
+  // MD.name    = std::string("ForceGradient");
+  
+  MD.MDsteps = user_params.Steps;
+  MD.trajL   = user_params.TrajectoryLength;
+
+  typedef HMCWrapper::ImplPolicy GaugeImplPolicy;
+  
+  HMCparameters HMCparams;
+  HMCparams.StartTrajectory  = user_params.StartTrajectory;
+  HMCparams.Trajectories     = user_params.Trajectories;
+  HMCparams.NoMetropolisUntil= 0;
+  HMCparams.StartingType     = user_params.StartingType;
+  HMCparams.MetropolisTest = user_params.MetropolisTest;
+  HMCparams.MD = MD;
+  HMCWrapper TheHMC(HMCparams);
+
+  // Grid from the command line arguments --grid and --mpi
+  TheHMC.Resources.AddFourDimGrid("gauge"); // use default simd lanes decomposition
+
+  CheckpointerParameters CPparams;
+  CPparams.config_prefix = "ckpoint_lat";
+  CPparams.rng_prefix    = "ckpoint_rng";
+  CPparams.saveInterval  = user_params.SaveInterval;
+  CPparams.format        = "IEEE64BIG";
+  TheHMC.Resources.LoadNerscCheckpointer(CPparams);
+
+  //Note that checkpointing saves the RNG state so that this initialization is required only for the very first configuration
+  RNGModuleParameters RNGpar;
+  RNGpar.serial_seeds = serial_seeds;
+  RNGpar.parallel_seeds = parallel_seeds;
+  TheHMC.Resources.SetRNGSeeds(RNGpar);
+
+  typedef PlaquetteMod<GaugeImplPolicy> PlaqObs;
+  TheHMC.Resources.AddObservable<PlaqObs>();
+  //////////////////////////////////////////////
+  //aiming for ainv=1.723 GeV
+  //                                  me         bob
+  //Estimated  a(ml+mres) [40ID] = 0.001305    0.00131
+  //           a(mh+mres) [40ID] = 0.035910    0.03529
+  //Estimate Ls=12, b+c=2  mres~0.0011
+
+  //1/24/2022 initial mres measurement gives mres=0.001,  adjusted light quark mass to 0.0003 from 0.0001
+  
+  const int Ls      = 12;
+  Real beta         = 1.848;
+  Real light_mass   = 0.0003;
+  Real strange_mass = 0.0342;
+  Real pv_mass      = 1.0;
+  RealD M5  = 1.8;
+  RealD mobius_scale = 2.; //b+c
+
+  RealD mob_bmc = 1.0;
+  RealD mob_b = (mobius_scale + mob_bmc)/2.;
+  RealD mob_c = (mobius_scale - mob_bmc)/2.;
+
+  std::cout << GridLogMessage
+	    << "Ensemble parameters:" << std::endl
+	    << "Ls=" << Ls << std::endl
+	    << "beta=" << beta << std::endl
+	    << "light_mass=" << light_mass << std::endl
+	    << "strange_mass=" << strange_mass << std::endl
+	    << "mobius_scale=" << mobius_scale << std::endl;
+  
+  //Setup the Grids
+  auto UGridD   = TheHMC.Resources.GetCartesian();
+  auto UrbGridD = TheHMC.Resources.GetRBCartesian();
+  auto FGridD     = SpaceTimeGrid::makeFiveDimGrid(Ls,UGridD);
+  auto FrbGridD   = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGridD);
+
+  GridCartesian* UGridF = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd, vComplexF::Nsimd()), GridDefaultMpi());
+  GridRedBlackCartesian* UrbGridF = SpaceTimeGrid::makeFourDimRedBlackGrid(UGridF);
+  auto FGridF     = SpaceTimeGrid::makeFiveDimGrid(Ls,UGridF);
+  auto FrbGridF   = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGridF);
+
+  ConjugateIwasakiGaugeActionD GaugeAction(beta);
+
+  // temporarily need a gauge field
+  LatticeGaugeFieldD Ud(UGridD);
+  LatticeGaugeFieldF Uf(UGridF);
+ 
+  //Setup the BCs
+  FermionActionD::ImplParams Params;
+  for(int i=0;i<Nd-1;i++) Params.twists[i] = user_params.GparityDirs[i]; //G-parity directions
+  Params.twists[Nd-1] = 1; //APBC in time direction
+
+  std::vector<int> dirs4(Nd);
+  for(int i=0;i<Nd-1;i++) dirs4[i] = user_params.GparityDirs[i];
+  dirs4[Nd-1] = 0; //periodic gauge BC in time
+
+  GaugeImplPolicy::setDirections(dirs4); //gauge BC
+
+  //Run optional gauge field checksum checker and exit
+  if(file_load_check){
+    TheHMC.initializeGaugeFieldAndRNGs(Ud);
+    std::cout << GridLogMessage << " Done" << std::endl;
+    Grid_finalize();
+    return 0;
+  }
+
+
+  ////////////////////////////////////
+  // Collect actions
+  ////////////////////////////////////
+  ActionLevel<HMCWrapper::Field> Level1(1); //light quark + strange quark
+  ActionLevel<HMCWrapper::Field> Level2(4); //DSDR
+  ActionLevel<HMCWrapper::Field> Level3(2); //gauge
+
+
+  /////////////////////////////////////////////////////////////
+  // Light EOFA action
+  // have to be careful with the parameters, cf. Test_dwf_gpforce_eofa.cc
+  /////////////////////////////////////////////////////////////
+  typedef SchurDiagMooeeOperator<EOFAactionD,FermionFieldD> EOFAschuropD;
+  typedef SchurDiagMooeeOperator<EOFAactionF,FermionFieldF> EOFAschuropF;
+  typedef ExactOneFlavourRatioMixedPrecHeatbathPseudoFermionAction<FermionImplPolicyD, FermionImplPolicyF> EOFAmixPrecPFaction;
+  typedef MixedPrecisionConjugateGradientOperatorFunction<EOFAactionD, EOFAactionF, EOFAschuropD, EOFAschuropF> EOFA_mxCG;
+  typedef MixedPrecisionReliableUpdateConjugateGradientOperatorFunction<EOFAactionD, EOFAactionF, EOFAschuropD, EOFAschuropF> EOFA_relupCG;
+
+
+  std::vector<RealD> eofa_light_masses = { light_mass ,  0.004,   0.016,   0.064,   0.256    };
+  std::vector<RealD> eofa_pv_masses =    { 0.004       , 0.016,   0.064,   0.256,   1.0      };
+  int n_light_hsb = 5;
+  assert(user_params.eofa_l.size() == n_light_hsb);
+  
+  EOFAmixPrecPFaction* EOFA_pfactions[n_light_hsb];
+
+  for(int i=0;i<n_light_hsb;i++){
+    RealD iml = eofa_light_masses[i];
+    RealD ipv = eofa_pv_masses[i];
+
+    EOFAactionD* LopD = new EOFAactionD(Ud, *FGridD, *FrbGridD, *UGridD, *UrbGridD, iml, iml, ipv, 0.0, -1, M5, mob_b, mob_c, Params);
+    EOFAactionF* LopF = new EOFAactionF(Uf, *FGridF, *FrbGridF, *UGridF, *UrbGridF, iml, iml, ipv, 0.0, -1, M5, mob_b, mob_c, Params);
+    EOFAactionD* RopD = new EOFAactionD(Ud, *FGridD, *FrbGridD, *UGridD, *UrbGridD, ipv, iml, ipv, -1.0, 1, M5, mob_b, mob_c, Params);
+    EOFAactionF* RopF = new EOFAactionF(Uf, *FGridF, *FrbGridF, *UGridF, *UrbGridF, ipv, iml, ipv, -1.0, 1, M5, mob_b, mob_c, Params);
+
+    EOFAschuropD* linopL_D = new EOFAschuropD(*LopD);
+    EOFAschuropD* linopR_D = new EOFAschuropD(*RopD);
+    
+    EOFAschuropF* linopL_F = new EOFAschuropF(*LopF);
+    EOFAschuropF* linopR_F = new EOFAschuropF(*RopF);
+
+#if 1
+    //Note reusing user_params.eofa_l.action(|md)_mixcg_inner_tolerance  as Delta for now
+    EOFA_relupCG* ActionMCG_L = new EOFA_relupCG(user_params.eofa_l[i].action_tolerance, user_params.eofa_l[i].action_mixcg_inner_tolerance, 50000, UGridF, FrbGridF, *LopF, *LopD, *linopL_F, *linopL_D);
+    EOFA_relupCG* ActionMCG_R = new EOFA_relupCG(user_params.eofa_l[i].action_tolerance, user_params.eofa_l[i].action_mixcg_inner_tolerance, 50000, UGridF, FrbGridF, *RopF, *RopD, *linopR_F, *linopR_D);
+
+    EOFA_relupCG* DerivMCG_L = new EOFA_relupCG(user_params.eofa_l[i].md_tolerance, user_params.eofa_l[i].md_mixcg_inner_tolerance, 50000, UGridF, FrbGridF, *LopF, *LopD, *linopL_F, *linopL_D);
+    EOFA_relupCG* DerivMCG_R = new EOFA_relupCG(user_params.eofa_l[i].md_tolerance, user_params.eofa_l[i].md_mixcg_inner_tolerance, 50000, UGridF, FrbGridF, *RopF, *RopD, *linopR_F, *linopR_D);
+
+#else
+    EOFA_mxCG* ActionMCG_L = new EOFA_mxCG(user_params.eofa_l[i].action_tolerance, 50000, 1000, UGridF, FrbGridF, *LopF, *LopD, *linopL_F, *linopL_D);
+    ActionMCG_L->InnerTolerance = user_params.eofa_l[i].action_mixcg_inner_tolerance;
+    
+    EOFA_mxCG* ActionMCG_R = new EOFA_mxCG(user_params.eofa_l[i].action_tolerance, 50000, 1000, UGridF, FrbGridF, *RopF, *RopD, *linopR_F, *linopR_D);
+    ActionMCG_R->InnerTolerance = user_params.eofa_l[i].action_mixcg_inner_tolerance;
+    
+    EOFA_mxCG* DerivMCG_L = new EOFA_mxCG(user_params.eofa_l[i].md_tolerance, 50000, 1000, UGridF, FrbGridF, *LopF, *LopD, *linopL_F, *linopL_D);
+    DerivMCG_L->InnerTolerance = user_params.eofa_l[i].md_mixcg_inner_tolerance;
+    
+    EOFA_mxCG* DerivMCG_R = new EOFA_mxCG(user_params.eofa_l[i].md_tolerance, 50000, 1000, UGridF, FrbGridF, *RopF, *RopD, *linopR_F, *linopR_D);
+    DerivMCG_R->InnerTolerance = user_params.eofa_l[i].md_mixcg_inner_tolerance;
+    
+    std::cout << GridLogMessage << "Set EOFA action solver action tolerance outer=" << ActionMCG_L->Tolerance << " inner=" << ActionMCG_L->InnerTolerance << std::endl;
+    std::cout << GridLogMessage << "Set EOFA MD solver tolerance outer=" << DerivMCG_L->Tolerance << " inner=" << DerivMCG_L->InnerTolerance << std::endl;
+#endif
+
+    EOFAmixPrecPFaction* EOFA = new EOFAmixPrecPFaction(*LopF, *RopF,
+							*LopD, *RopD, 
+							*ActionMCG_L, *ActionMCG_R, 
+							*ActionMCG_L, *ActionMCG_R, 
+							*DerivMCG_L, *DerivMCG_R, 
+							user_params.eofa_l[i].rat_params, true);
+    EOFA_pfactions[i] = EOFA;
+    Level1.push_back(EOFA);
+  }
+
+  ////////////////////////////////////
+  // Strange action
+  ////////////////////////////////////
+  FermionActionD Numerator_sD(Ud,*FGridD,*FrbGridD,*UGridD,*UrbGridD,strange_mass,M5,mob_b,mob_c,Params);
+  FermionActionD Denominator_sD(Ud,*FGridD,*FrbGridD,*UGridD,*UrbGridD, pv_mass,M5,mob_b,mob_c,Params);
+
+  FermionActionF Numerator_sF(Uf,*FGridF,*FrbGridF,*UGridF,*UrbGridF,strange_mass,M5,mob_b,mob_c,Params);
+  FermionActionF Denominator_sF(Uf,*FGridF,*FrbGridF,*UGridF,*UrbGridF, pv_mass,M5,mob_b,mob_c,Params);
+
+  RationalActionParams rat_act_params_s;
+  rat_act_params_s.inv_pow  = 4; // (M^dag M)^{1/4}
+  rat_act_params_s.precision= 60;
+  rat_act_params_s.MaxIter  = 50000;
+  user_params.rat_quo_s.Export(rat_act_params_s);
+  std::cout << GridLogMessage << " Heavy quark bounds check every " << rat_act_params_s.BoundsCheckFreq << " trajectories (avg)" << std::endl;
+
+  //MixedPrecRHMC Quotient_s(Denominator_sD, Numerator_sD, Denominator_sF, Numerator_sF, rat_act_params_s, user_params.rat_quo_s.reliable_update_freq); 
+  DoublePrecRHMC Quotient_s(Denominator_sD, Numerator_sD, rat_act_params_s); 
+  Level1.push_back(&Quotient_s);  
+
+  ///////////////////////////////////
+  // DSDR action
+  ///////////////////////////////////
+  RealD dsdr_mass=-1.8;   
+  //Use same DSDR twists as https://arxiv.org/pdf/1208.4412.pdf
+  RealD dsdr_epsilon_f = 0.02; //numerator (in determinant)
+  RealD dsdr_epsilon_b = 0.5; 
+  GparityWilsonTMFermionD Numerator_DSDR_D(Ud, *UGridD, *UrbGridD, dsdr_mass, dsdr_epsilon_f, Params);
+  GparityWilsonTMFermionF Numerator_DSDR_F(Uf, *UGridF, *UrbGridF, dsdr_mass, dsdr_epsilon_f, Params);
+
+  GparityWilsonTMFermionD Denominator_DSDR_D(Ud, *UGridD, *UrbGridD, dsdr_mass, dsdr_epsilon_b, Params);
+  GparityWilsonTMFermionF Denominator_DSDR_F(Uf, *UGridF, *UrbGridF, dsdr_mass, dsdr_epsilon_b, Params);
+ 
+  RationalActionParams rat_act_params_DSDR;
+  rat_act_params_DSDR.inv_pow  = 2; // (M^dag M)^{1/2}
+  rat_act_params_DSDR.precision= 60;
+  rat_act_params_DSDR.MaxIter  = 50000;
+  user_params.rat_quo_DSDR.Export(rat_act_params_DSDR);
+  std::cout << GridLogMessage << "DSDR quark bounds check every " << rat_act_params_DSDR.BoundsCheckFreq << " trajectories (avg)" << std::endl;
+
+  DoublePrecRHMC Quotient_DSDR(Denominator_DSDR_D, Numerator_DSDR_D, rat_act_params_DSDR);
+  Level2.push_back(&Quotient_DSDR);
+
+  /////////////////////////////////////////////////////////////
+  // Gauge action
+  /////////////////////////////////////////////////////////////
+  Level3.push_back(&GaugeAction);
+
+  TheHMC.TheAction.push_back(Level1);
+  TheHMC.TheAction.push_back(Level2);
+  TheHMC.TheAction.push_back(Level3);
+  std::cout << GridLogMessage << " Action complete "<< std::endl;
+
+
+  //Action tuning
+  bool 
+    tune_rhmc_s=false, eigenrange_s=false, 
+    tune_rhmc_DSDR=false, eigenrange_DSDR=false, 
+    check_eofa=false, 
+    upper_bound_eofa=false, lower_bound_eofa(false);
+
+  std::string lanc_params_s;
+  std::string lanc_params_DSDR;
+  int tune_rhmc_s_action_or_md;
+  int tune_rhmc_DSDR_action_or_md;
+  int eofa_which_hsb;
+
+  for(int i=1;i<argc;i++){
+    std::string sarg(argv[i]);
+    if(sarg == "--tune_rhmc_s"){
+      assert(i < argc-1);
+      tune_rhmc_s=true;
+      tune_rhmc_s_action_or_md = std::stoi(argv[i+1]);
+    }
+    else if(sarg == "--eigenrange_s"){
+      assert(i < argc-1);
+      eigenrange_s=true;
+      lanc_params_s = argv[i+1];
+    }
+    else if(sarg == "--tune_rhmc_DSDR"){
+      assert(i < argc-1);
+      tune_rhmc_DSDR=true;
+      tune_rhmc_DSDR_action_or_md = std::stoi(argv[i+1]);
+    }
+    else if(sarg == "--eigenrange_DSDR"){
+      assert(i < argc-1);
+      eigenrange_DSDR=true;
+      lanc_params_DSDR = argv[i+1];
+    }
+    else if(sarg == "--check_eofa"){
+      assert(i < argc-1);
+      check_eofa = true;
+      eofa_which_hsb = std::stoi(argv[i+1]); //-1 indicates all hasenbusch
+      assert(eofa_which_hsb == -1 || (eofa_which_hsb >= 0 && eofa_which_hsb < n_light_hsb) );
+    }
+    else if(sarg == "--upper_bound_eofa"){
+      assert(i < argc-1);
+      upper_bound_eofa = true;
+      eofa_which_hsb = std::stoi(argv[i+1]);
+      assert(eofa_which_hsb >= 0 && eofa_which_hsb < n_light_hsb);
+    }
+    else if(sarg == "--lower_bound_eofa"){
+      assert(i < argc-1);
+      lower_bound_eofa = true;      
+      eofa_which_hsb = std::stoi(argv[i+1]);
+      assert(eofa_which_hsb >= 0 && eofa_which_hsb < n_light_hsb);
+    }
+  }
+  if(tune_rhmc_s || eigenrange_s || tune_rhmc_DSDR || eigenrange_DSDR ||check_eofa || upper_bound_eofa || lower_bound_eofa) {
+    std::cout << GridLogMessage << "Running checks" << std::endl;
+    TheHMC.initializeGaugeFieldAndRNGs(Ud);
+
+    //std::cout << GridLogMessage << "EOFA action solver action tolerance outer=" << ActionMCG_L.Tolerance << " inner=" << ActionMCG_L.InnerTolerance << std::endl;
+    //std::cout << GridLogMessage << "EOFA MD solver tolerance outer=" << DerivMCG_L.Tolerance << " inner=" << DerivMCG_L.InnerTolerance << std::endl;
+
+    if(check_eofa){
+      if(eofa_which_hsb >= 0){
+	std::cout << GridLogMessage << "Starting checking EOFA Hasenbusch " << eofa_which_hsb << std::endl;
+	checkEOFA(*EOFA_pfactions[eofa_which_hsb], FGridD, TheHMC.Resources.GetParallelRNG(), Ud);
+	std::cout << GridLogMessage << "Finished checking EOFA Hasenbusch " << eofa_which_hsb << std::endl;
+      }else{
+	for(int i=0;i<n_light_hsb;i++){
+	  std::cout << GridLogMessage << "Starting checking EOFA Hasenbusch " << i << std::endl;
+	  checkEOFA(*EOFA_pfactions[i], FGridD, TheHMC.Resources.GetParallelRNG(), Ud);
+	  std::cout << GridLogMessage << "Finished checking EOFA Hasenbusch " << i << std::endl;
+	}
+      }
+    }	  
+    if(upper_bound_eofa) upperBoundEOFA(*EOFA_pfactions[eofa_which_hsb], FGridD, TheHMC.Resources.GetParallelRNG(), Ud);
+    if(lower_bound_eofa) lowerBoundEOFA(*EOFA_pfactions[eofa_which_hsb], FGridD, TheHMC.Resources.GetParallelRNG(), Ud);
+    if(eigenrange_s) computeEigenvalues<FermionActionD, FermionFieldD>(lanc_params_s, FGridD, FrbGridD, Ud, Numerator_sD, TheHMC.Resources.GetParallelRNG());
+    if(tune_rhmc_s) checkRHMC<FermionActionD, FermionFieldD, decltype(Quotient_s)>(FGridD, FrbGridD, Ud, Numerator_sD, Denominator_sD, Quotient_s, TheHMC.Resources.GetParallelRNG(), 4, "strange",  tune_rhmc_s_action_or_md);
+    if(eigenrange_DSDR) computeEigenvalues<GparityWilsonTMFermionD, GparityWilsonTMFermionD::FermionField>(lanc_params_DSDR, UGridD, UrbGridD, Ud, Numerator_DSDR_D, TheHMC.Resources.GetParallelRNG());
+    if(tune_rhmc_DSDR) checkRHMC<GparityWilsonTMFermionD, GparityWilsonTMFermionD::FermionField, decltype(Quotient_DSDR)>(UGridD, UrbGridD, Ud, Numerator_DSDR_D, Denominator_DSDR_D, Quotient_DSDR, TheHMC.Resources.GetParallelRNG(), 2, "DSDR", tune_rhmc_DSDR_action_or_md);
+
+
+    std::cout << GridLogMessage << " Done" << std::endl;
+    Grid_finalize();
+    return 0;
+  }
+
+
+  //Run the HMC
+  std::cout << GridLogMessage << " Running the HMC "<< std::endl;
+  TheHMC.Run();
+
+  std::cout << GridLogMessage << " Done" << std::endl;
+  Grid_finalize();
+  return 0;
+} // main

HMC/Mobius2p1fIDSDRGparityEOFA_48ID.cc

@@ -0,0 +1,873 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid
+
+Source file: ./HMC/Mobius2p1fIDSDRGparityEOFA.cc
+
+Copyright (C) 2015-2016
+
+Author: Christopher Kelly <ckelly@bnl.gov>
+Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
+
+
+This program is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 2 of the License, or
+(at your option) any later version.
+
+This program is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along
+with this program; if not, write to the Free Software Foundation, Inc.,
+51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+See the full license in the file "LICENSE" in the top level distribution
+directory
+*************************************************************************************/
+/*  END LEGAL */
+#include <Grid/Grid.h>
+
+using namespace Grid;
+
+//Production binary for the 40ID G-parity ensemble
+
+struct RatQuoParameters: Serializable {
+  GRID_SERIALIZABLE_CLASS_MEMBERS(RatQuoParameters,
+				  double, bnd_lo,
+				  double, bnd_hi,
+				  Integer, action_degree,
+				  double, action_tolerance,
+				  Integer, md_degree,
+				  double, md_tolerance,
+				  Integer, reliable_update_freq,
+				  Integer, bnd_check_freq);
+  RatQuoParameters() { 
+    bnd_lo = 1e-2;
+    bnd_hi = 30;
+    action_degree = 10;
+    action_tolerance = 1e-10;
+    md_degree = 10;
+    md_tolerance = 1e-8;
+    bnd_check_freq = 20;
+    reliable_update_freq = 50;
+  }
+
+  void Export(RationalActionParams &into) const{
+    into.lo = bnd_lo;
+    into.hi = bnd_hi;
+    into.action_degree = action_degree;
+    into.action_tolerance = action_tolerance;
+    into.md_degree = md_degree;
+    into.md_tolerance = md_tolerance;
+    into.BoundsCheckFreq = bnd_check_freq;
+  }
+};
+
+struct EOFAparameters: Serializable {
+  GRID_SERIALIZABLE_CLASS_MEMBERS(EOFAparameters,
+				  OneFlavourRationalParams, rat_params,
+				  double, action_tolerance,
+				  double, action_mixcg_inner_tolerance,
+				  double, md_tolerance,
+				  double, md_mixcg_inner_tolerance);
+
+  EOFAparameters() { 
+    action_mixcg_inner_tolerance = 1e-8;
+    action_tolerance = 1e-10;
+    md_tolerance = 1e-8;
+    md_mixcg_inner_tolerance = 1e-8;
+
+    rat_params.lo = 1.0;
+    rat_params.hi = 25.0;
+    rat_params.MaxIter  = 10000;
+    rat_params.tolerance= 1.0e-9;
+    rat_params.degree   = 14;
+    rat_params.precision= 50;
+  }
+};
+
+struct EvolParameters: Serializable {
+  GRID_SERIALIZABLE_CLASS_MEMBERS(EvolParameters,
+                                  Integer, StartTrajectory,
+                                  Integer, Trajectories,
+				  Integer, SaveInterval,
+				  Integer, Steps,
+				  RealD, TrajectoryLength,
+                                  bool, MetropolisTest,
+				  std::string, StartingType,
+				  std::vector<Integer>, GparityDirs,
+				  std::vector<EOFAparameters>, eofa_l,
+				  RatQuoParameters, rat_quo_s,
+				  RatQuoParameters, rat_quo_DSDR);
+
+  EvolParameters() {
+    //For initial thermalization; afterwards user should switch Metropolis on and use StartingType=CheckpointStart
+    MetropolisTest    = false;
+    StartTrajectory   = 0;
+    Trajectories      = 50;
+    SaveInterval = 5;
+    StartingType      = "ColdStart";
+    GparityDirs.resize(3, 1); //1 for G-parity, 0 for periodic
+    Steps = 5;
+    TrajectoryLength = 1.0;
+  }
+};
+
+bool fileExists(const std::string &fn){
+  std::ifstream f(fn);
+  return f.good();
+}
+
+
+
+
+struct LanczosParameters: Serializable {
+  GRID_SERIALIZABLE_CLASS_MEMBERS(LanczosParameters,
+				  double, alpha,
+				  double, beta,
+				  double, mu,
+				  int, ord,
+				  int, n_stop,
+				  int, n_want,
+				  int, n_use,
+				  double, tolerance);
+
+  LanczosParameters() {
+    alpha = 35;
+    beta = 5;
+    mu = 0;
+    ord = 100;
+    n_stop = 10;
+    n_want = 10;
+    n_use = 15;
+    tolerance = 1e-6;
+  }
+};
+
+
+
+template<typename FermionActionD, typename FermionFieldD>
+void computeEigenvalues(std::string param_file,
+			GridCartesian* Grid, GridRedBlackCartesian* rbGrid, const LatticeGaugeFieldD &latt,  //expect lattice to have been initialized to something
+			FermionActionD &action, GridParallelRNG &rng){
+  
+  LanczosParameters params;
+  if(fileExists(param_file)){
+    std::cout << GridLogMessage << " Reading " << param_file << std::endl;
+    Grid::XmlReader rd(param_file);
+    read(rd, "LanczosParameters", params);
+  }else if(!GlobalSharedMemory::WorldRank){
+    std::cout << GridLogMessage << " File " << param_file << " does not exist" << std::endl;
+    std::cout << GridLogMessage << " Writing xml template to " << param_file << ".templ" << std::endl;
+    Grid::XmlWriter wr(param_file + ".templ");
+    write(wr, "LanczosParameters", params);
+  }
+
+  FermionFieldD gauss_o(rbGrid);
+  FermionFieldD gauss(Grid);
+  gaussian(rng, gauss);
+  pickCheckerboard(Odd, gauss_o, gauss);
+
+  action.ImportGauge(latt);
+
+  SchurDiagMooeeOperator<FermionActionD, FermionFieldD> hermop(action);
+  PlainHermOp<FermionFieldD> hermop_wrap(hermop);
+  //ChebyshevLanczos<FermionFieldD> Cheb(params.alpha, params.beta, params.mu, params.ord);
+  assert(params.mu == 0.0);
+
+  Chebyshev<FermionFieldD> Cheb(params.beta*params.beta, params.alpha*params.alpha, params.ord+1);
+  FunctionHermOp<FermionFieldD> Cheb_wrap(Cheb, hermop);
+
+  std::cout << "IRL: alpha=" << params.alpha << " beta=" << params.beta << " mu=" << params.mu << " ord=" << params.ord << std::endl;
+  ImplicitlyRestartedLanczos<FermionFieldD> IRL(Cheb_wrap, hermop_wrap, params.n_stop, params.n_want, params.n_use, params.tolerance, 10000);
+
+  std::vector<RealD> eval(params.n_use);
+  std::vector<FermionFieldD> evec(params.n_use, rbGrid);
+  int Nconv;
+  IRL.calc(eval, evec, gauss_o, Nconv);
+
+  std::cout << "Eigenvalues:" << std::endl;
+  for(int i=0;i<params.n_want;i++){
+    std::cout << i << " " << eval[i] << std::endl;
+  }
+}
+
+
+//Check the quality of the RHMC approx
+//action_or_md toggles checking the action (0), MD (1) or both (2) setups
+template<typename FermionActionD, typename FermionFieldD, typename RHMCtype>
+void checkRHMC(GridCartesian* Grid, GridRedBlackCartesian* rbGrid, const LatticeGaugeFieldD &latt,  //expect lattice to have been initialized to something
+	       FermionActionD &numOp, FermionActionD &denOp, RHMCtype &rhmc, GridParallelRNG &rng,
+	       int inv_pow, const std::string &quark_descr, int action_or_md){
+  assert(action_or_md == 0 || action_or_md == 1 || action_or_md == 2);
+  
+  FermionFieldD gauss_o(rbGrid);
+  FermionFieldD gauss(Grid);
+  gaussian(rng, gauss);
+  pickCheckerboard(Odd, gauss_o, gauss);
+
+  numOp.ImportGauge(latt);
+  denOp.ImportGauge(latt);
+
+  typedef typename FermionActionD::Impl_t FermionImplPolicyD;
+  SchurDifferentiableOperator<FermionImplPolicyD> MdagM(numOp);
+  SchurDifferentiableOperator<FermionImplPolicyD> VdagV(denOp);
+
+  PowerMethod<FermionFieldD> power_method;
+  RealD lambda_max;
+
+  std::cout << "Starting: Get RHMC high bound approx for " << quark_descr << " numerator" << std::endl;
+
+  lambda_max = power_method(MdagM,gauss_o);
+  std::cout << GridLogMessage << "Got lambda_max "<<lambda_max<<std::endl;
+
+  std::cout << "Starting: Get RHMC high bound approx for " << quark_descr << " denominator" << std::endl;
+  lambda_max = power_method(VdagV,gauss_o);
+  std::cout << GridLogMessage << "Got lambda_max "<<lambda_max<<std::endl;
+
+  if(action_or_md == 0 || action_or_md == 2){
+    std::cout << "Starting: Checking quality of RHMC action approx for " << quark_descr << " quark numerator and power -1/" << inv_pow << std::endl;
+    InversePowerBoundsCheck(inv_pow, 10000, 1e16, MdagM,gauss_o, rhmc.ApproxNegPowerAction); //use large tolerance to prevent exit on fail; we are trying to tune here!
+    std::cout << "Finished: Checking quality of RHMC action approx for " << quark_descr << " quark numerator and power -1/" << inv_pow << std::endl;
+
+    std::cout << "Starting: Checking quality of RHMC action approx for " << quark_descr << " quark numerator and power -1/" << 2*inv_pow << std::endl;
+    InversePowerBoundsCheck(2*inv_pow, 10000, 1e16, MdagM,gauss_o, rhmc.ApproxNegHalfPowerAction);
+    std::cout << "Finished: Checking quality of RHMC action approx for " << quark_descr << " quark numerator and power -1/" << 2*inv_pow << std::endl;
+
+    std::cout << "Starting: Checking quality of RHMC action approx for " << quark_descr << " quark denominator and power -1/" << inv_pow << std::endl;
+    InversePowerBoundsCheck(inv_pow, 10000, 1e16, VdagV,gauss_o, rhmc.ApproxNegPowerAction);
+    std::cout << "Finished: Checking quality of RHMC action approx for " << quark_descr << " quark denominator and power -1/" << inv_pow << std::endl;
+
+    std::cout << "Starting: Checking quality of RHMC action approx for " << quark_descr << " quark denominator and power -1/" << 2*inv_pow << std::endl;
+    InversePowerBoundsCheck(2*inv_pow, 10000, 1e16, VdagV,gauss_o, rhmc.ApproxNegHalfPowerAction);
+    std::cout << "Finished: Checking quality of RHMC action approx for " << quark_descr << " quark denominator and power -1/" << 2*inv_pow << std::endl;
+  }
+
+  std::cout << "-------------------------------------------------------------------------------" << std::endl;
+
+  if(action_or_md == 1 || action_or_md == 2){
+    std::cout << "Starting: Checking quality of RHMC MD approx for " << quark_descr << " quark numerator and power -1/" << inv_pow << std::endl;
+    InversePowerBoundsCheck(inv_pow, 10000, 1e16, MdagM,gauss_o, rhmc.ApproxNegPowerMD); 
+    std::cout << "Finished: Checking quality of RHMC MD approx for " << quark_descr << " quark numerator and power -1/" << inv_pow << std::endl;
+
+    std::cout << "Starting: Checking quality of RHMC MD approx for " << quark_descr << " quark numerator and power -1/" << 2*inv_pow << std::endl;
+    InversePowerBoundsCheck(2*inv_pow, 10000, 1e16, MdagM,gauss_o, rhmc.ApproxNegHalfPowerMD);
+    std::cout << "Finished: Checking quality of RHMC MD approx for " << quark_descr << " quark numerator and power -1/" << 2*inv_pow << std::endl;
+
+    std::cout << "Starting: Checking quality of RHMC MD approx for " << quark_descr << " quark denominator and power -1/" << inv_pow << std::endl;
+    InversePowerBoundsCheck(inv_pow, 10000, 1e16, VdagV,gauss_o, rhmc.ApproxNegPowerMD);
+    std::cout << "Finished: Checking quality of RHMC MD approx for " << quark_descr << " quark denominator and power -1/" << inv_pow << std::endl;
+
+    std::cout << "Starting: Checking quality of RHMC MD approx for " << quark_descr << " quark denominator and power -1/" << 2*inv_pow << std::endl;
+    InversePowerBoundsCheck(2*inv_pow, 10000, 1e16, VdagV,gauss_o, rhmc.ApproxNegHalfPowerMD);
+    std::cout << "Finished: Checking quality of RHMC MD approx for " << quark_descr << " quark denominator and power -1/" << 2*inv_pow << std::endl;
+  }
+}
+
+
+template<typename FermionImplPolicy>
+void checkEOFA(ExactOneFlavourRatioPseudoFermionAction<FermionImplPolicy> &EOFA,
+	       GridCartesian* FGrid, GridParallelRNG &rng, const LatticeGaugeFieldD &latt){
+  std::cout << GridLogMessage << "Starting EOFA action/bounds check" << std::endl;
+  typename FermionImplPolicy::FermionField eta(FGrid);
+  RealD scale = std::sqrt(0.5);
+  gaussian(rng,eta); eta = eta * scale;
+
+  //Use the inbuilt check
+  EOFA.refresh(latt, eta);
+  EOFA.S(latt);
+  std::cout << GridLogMessage << "Finished EOFA upper action/bounds check" << std::endl;
+}
+
+
+template<typename FermionImplPolicy>
+class EOFAlinop: public LinearOperatorBase<typename FermionImplPolicy::FermionField>{
+  ExactOneFlavourRatioPseudoFermionAction<FermionImplPolicy> &EOFA;
+  LatticeGaugeFieldD &U;
+public:
+  EOFAlinop(ExactOneFlavourRatioPseudoFermionAction<FermionImplPolicy> &EOFA, LatticeGaugeFieldD &U): EOFA(EOFA), U(U){}
+
+  typedef typename FermionImplPolicy::FermionField Field;
+  void OpDiag (const Field &in, Field &out){ assert(0); }
+  void OpDir  (const Field &in, Field &out,int dir,int disp){ assert(0); }
+  void OpDirAll  (const Field &in, std::vector<Field> &out){ assert(0); } 
+
+  void Op     (const Field &in, Field &out){ assert(0); }
+  void AdjOp  (const Field &in, Field &out){ assert(0); }
+  void HermOpAndNorm(const Field &in, Field &out,RealD &n1,RealD &n2){ assert(0); }
+  void HermOp(const Field &in, Field &out){ EOFA.Meofa(U, in, out); }
+};
+
+template<typename FermionImplPolicy>
+void upperBoundEOFA(ExactOneFlavourRatioPseudoFermionAction<FermionImplPolicy> &EOFA,
+		    GridCartesian* FGrid, GridParallelRNG &rng, LatticeGaugeFieldD &latt){
+  std::cout << GridLogMessage << "Starting EOFA upper bound compute" << std::endl;
+  EOFAlinop<FermionImplPolicy> linop(EOFA, latt);
+  typename FermionImplPolicy::FermionField eta(FGrid);
+  gaussian(rng,eta);
+  PowerMethod<typename FermionImplPolicy::FermionField> power_method;
+  auto lambda_max = power_method(linop,eta);
+  std::cout << GridLogMessage << "Upper bound of EOFA operator " << lambda_max << std::endl;
+}
+
+//Applications of M^{-1} cost the same as M for EOFA!
+template<typename FermionImplPolicy>
+class EOFAinvLinop: public LinearOperatorBase<typename FermionImplPolicy::FermionField>{
+  ExactOneFlavourRatioPseudoFermionAction<FermionImplPolicy> &EOFA;
+  LatticeGaugeFieldD &U;
+public:
+  EOFAinvLinop(ExactOneFlavourRatioPseudoFermionAction<FermionImplPolicy> &EOFA, LatticeGaugeFieldD &U): EOFA(EOFA), U(U){}
+
+  typedef typename FermionImplPolicy::FermionField Field;
+  void OpDiag (const Field &in, Field &out){ assert(0); }
+  void OpDir  (const Field &in, Field &out,int dir,int disp){ assert(0); }
+  void OpDirAll  (const Field &in, std::vector<Field> &out){ assert(0); } 
+
+  void Op     (const Field &in, Field &out){ assert(0); }
+  void AdjOp  (const Field &in, Field &out){ assert(0); }
+  void HermOpAndNorm(const Field &in, Field &out,RealD &n1,RealD &n2){ assert(0); }
+  void HermOp(const Field &in, Field &out){ EOFA.MeofaInv(U, in, out); }
+};
+
+template<typename FermionImplPolicy>
+void lowerBoundEOFA(ExactOneFlavourRatioPseudoFermionAction<FermionImplPolicy> &EOFA,
+		    GridCartesian* FGrid, GridParallelRNG &rng, LatticeGaugeFieldD &latt){
+  std::cout << GridLogMessage << "Starting EOFA lower bound compute using power method on M^{-1}. Inverse of highest eigenvalue is the lowest eigenvalue of M" << std::endl;
+  EOFAinvLinop<FermionImplPolicy> linop(EOFA, latt);
+  typename FermionImplPolicy::FermionField eta(FGrid);
+  gaussian(rng,eta);
+  PowerMethod<typename FermionImplPolicy::FermionField> power_method;
+  auto lambda_max = power_method(linop,eta);
+  std::cout << GridLogMessage << "Lower bound of EOFA operator " << 1./lambda_max << std::endl;
+}
+
+
+NAMESPACE_BEGIN(Grid);
+
+  template<class FermionOperatorD, class FermionOperatorF, class SchurOperatorD, class  SchurOperatorF> 
+  class MixedPrecisionConjugateGradientOperatorFunction : public OperatorFunction<typename FermionOperatorD::FermionField> {
+  public:
+    typedef typename FermionOperatorD::FermionField FieldD;
+    typedef typename FermionOperatorF::FermionField FieldF;
+
+    using OperatorFunction<FieldD>::operator();
+
+    RealD   Tolerance;
+    RealD   InnerTolerance; //Initial tolerance for inner CG. Defaults to Tolerance but can be changed
+    Integer MaxInnerIterations;
+    Integer MaxOuterIterations;
+    GridBase* SinglePrecGrid4; //Grid for single-precision fields
+    GridBase* SinglePrecGrid5; //Grid for single-precision fields
+    RealD OuterLoopNormMult; //Stop the outer loop and move to a final double prec solve when the residual is OuterLoopNormMult * Tolerance
+
+    FermionOperatorF &FermOpF;
+    FermionOperatorD &FermOpD;;
+    SchurOperatorF &LinOpF;
+    SchurOperatorD &LinOpD;
+
+    Integer TotalInnerIterations; //Number of inner CG iterations
+    Integer TotalOuterIterations; //Number of restarts
+    Integer TotalFinalStepIterations; //Number of CG iterations in final patch-up step
+
+    MixedPrecisionConjugateGradientOperatorFunction(RealD tol, 
+						    Integer maxinnerit, 
+						    Integer maxouterit, 
+						    GridBase* _sp_grid4, 
+						    GridBase* _sp_grid5, 
+						    FermionOperatorF &_FermOpF,
+						    FermionOperatorD &_FermOpD,
+						    SchurOperatorF   &_LinOpF,
+						    SchurOperatorD   &_LinOpD): 
+      LinOpF(_LinOpF),
+      LinOpD(_LinOpD),
+      FermOpF(_FermOpF),
+      FermOpD(_FermOpD),
+      Tolerance(tol), 
+      InnerTolerance(tol), 
+      MaxInnerIterations(maxinnerit), 
+      MaxOuterIterations(maxouterit), 
+      SinglePrecGrid4(_sp_grid4),
+      SinglePrecGrid5(_sp_grid5),
+      OuterLoopNormMult(100.) 
+    { 
+    };
+
+    void operator()(LinearOperatorBase<FieldD> &LinOpU, const FieldD &src, FieldD &psi) {
+
+      std::cout << GridLogMessage << " Mixed precision CG wrapper operator() "<<std::endl;
+
+      SchurOperatorD * SchurOpU = static_cast<SchurOperatorD *>(&LinOpU);
+      assert(&(SchurOpU->_Mat)==&(LinOpD._Mat));
+
+      precisionChange(FermOpF.Umu, FermOpD.Umu);
+
+      pickCheckerboard(Even,FermOpF.UmuEven,FermOpF.Umu);
+      pickCheckerboard(Odd ,FermOpF.UmuOdd ,FermOpF.Umu);
+
+      ////////////////////////////////////////////////////////////////////////////////////
+      // Make a mixed precision conjugate gradient
+      ////////////////////////////////////////////////////////////////////////////////////
+      MixedPrecisionConjugateGradient<FieldD,FieldF> MPCG(Tolerance,MaxInnerIterations,MaxOuterIterations,SinglePrecGrid5,LinOpF,LinOpD);
+      MPCG.InnerTolerance = InnerTolerance;
+      std::cout << GridLogMessage << "Calling mixed precision Conjugate Gradient" <<std::endl;
+      MPCG(src,psi);
+    }
+  };
+
+
+  template<class FermionOperatorD, class FermionOperatorF, class SchurOperatorD, class  SchurOperatorF> 
+  class MixedPrecisionReliableUpdateConjugateGradientOperatorFunction : public OperatorFunction<typename FermionOperatorD::FermionField> {
+  public:
+    typedef typename FermionOperatorD::FermionField FieldD;
+    typedef typename FermionOperatorF::FermionField FieldF;
+
+    using OperatorFunction<FieldD>::operator();
+
+    RealD Tolerance;
+    Integer MaxIterations;
+
+    RealD Delta; //reliable update parameter
+
+    GridBase* SinglePrecGrid4; //Grid for single-precision fields
+    GridBase* SinglePrecGrid5; //Grid for single-precision fields
+
+    FermionOperatorF &FermOpF;
+    FermionOperatorD &FermOpD;;
+    SchurOperatorF &LinOpF;
+    SchurOperatorD &LinOpD;
+    
+    MixedPrecisionReliableUpdateConjugateGradientOperatorFunction(RealD tol, 
+								  RealD delta,
+								  Integer maxit, 
+								  GridBase* _sp_grid4, 
+								  GridBase* _sp_grid5, 
+								  FermionOperatorF &_FermOpF,
+								  FermionOperatorD &_FermOpD,
+								  SchurOperatorF   &_LinOpF,
+								  SchurOperatorD   &_LinOpD): 
+      LinOpF(_LinOpF),
+      LinOpD(_LinOpD),
+      FermOpF(_FermOpF),
+      FermOpD(_FermOpD),
+      Tolerance(tol), 
+      Delta(delta),
+      MaxIterations(maxit), 
+      SinglePrecGrid4(_sp_grid4),
+      SinglePrecGrid5(_sp_grid5)
+    { 
+    };
+
+    void operator()(LinearOperatorBase<FieldD> &LinOpU, const FieldD &src, FieldD &psi) {
+
+      std::cout << GridLogMessage << " Mixed precision reliable CG update wrapper operator() "<<std::endl;
+
+      SchurOperatorD * SchurOpU = static_cast<SchurOperatorD *>(&LinOpU);
+      assert(&(SchurOpU->_Mat)==&(LinOpD._Mat));
+
+      precisionChange(FermOpF.Umu, FermOpD.Umu);
+
+      pickCheckerboard(Even,FermOpF.UmuEven,FermOpF.Umu);
+      pickCheckerboard(Odd ,FermOpF.UmuOdd ,FermOpF.Umu);
+
+      ////////////////////////////////////////////////////////////////////////////////////
+      // Make a mixed precision conjugate gradient
+      ////////////////////////////////////////////////////////////////////////////////////
+
+      ConjugateGradientReliableUpdate<FieldD,FieldF> MPCG(Tolerance,MaxIterations,Delta,SinglePrecGrid5,LinOpF,LinOpD);
+      std::cout << GridLogMessage << "Calling mixed precision reliable update Conjugate Gradient" <<std::endl;
+      MPCG(src,psi);
+    }
+  };
+
+
+
+NAMESPACE_END(Grid);
+
+
+
+
+
+int main(int argc, char **argv) {
+  Grid_init(&argc, &argv);
+  int threads = GridThread::GetThreads();
+  // here make a routine to print all the relevant information on the run
+  std::cout << GridLogMessage << "Grid is setup to use " << threads << " threads" << std::endl;
+
+  std::string param_file = "params.xml";
+  bool file_load_check = false;
+  for(int i=1;i<argc;i++){
+    std::string sarg(argv[i]);
+    if(sarg == "--param_file"){
+      assert(i!=argc-1);
+      param_file = argv[i+1];
+    }else if(sarg == "--read_check"){ //check the fields load correctly and pass checksum/plaquette repro
+      file_load_check = true;
+    }
+  }
+
+  //Read the user parameters
+  EvolParameters user_params;
+  
+  if(fileExists(param_file)){
+    std::cout << GridLogMessage << " Reading " << param_file << std::endl;
+    Grid::XmlReader rd(param_file);
+    read(rd, "Params", user_params);
+  }else if(!GlobalSharedMemory::WorldRank){
+    std::cout << GridLogMessage << " File " << param_file << " does not exist" << std::endl;
+    std::cout << GridLogMessage << " Writing xml template to " << param_file << ".templ" << std::endl;
+    {
+      Grid::XmlWriter wr(param_file + ".templ");
+      write(wr, "Params", user_params);
+    }
+    std::cout << GridLogMessage << " Done" << std::endl;
+    Grid_finalize();
+    return 0;
+  }
+
+  //Check the parameters
+  if(user_params.GparityDirs.size() != Nd-1){
+    std::cerr << "Error in input parameters: expect GparityDirs to have size = " << Nd-1 << std::endl;
+    exit(1);
+  }
+  for(int i=0;i<Nd-1;i++)
+    if(user_params.GparityDirs[i] != 0 && user_params.GparityDirs[i] != 1){
+      std::cerr << "Error in input parameters: expect GparityDirs values to be 0 (periodic) or 1 (G-parity)" << std::endl;
+      exit(1);
+    }
+
+
+  typedef GparityMobiusEOFAFermionD EOFAactionD;
+  typedef GparityMobiusFermionD FermionActionD;
+  typedef typename FermionActionD::Impl_t FermionImplPolicyD;
+  typedef typename FermionActionD::FermionField FermionFieldD;
+
+  typedef GparityMobiusEOFAFermionF EOFAactionF;
+  typedef GparityMobiusFermionF FermionActionF;
+  typedef typename FermionActionF::Impl_t FermionImplPolicyF;
+  typedef typename FermionActionF::FermionField FermionFieldF;
+
+  typedef GeneralEvenOddRatioRationalMixedPrecPseudoFermionAction<FermionImplPolicyD,FermionImplPolicyF> MixedPrecRHMC;
+  typedef GeneralEvenOddRatioRationalPseudoFermionAction<FermionImplPolicyD> DoublePrecRHMC;
+
+  //::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
+  IntegratorParameters MD;
+  typedef ConjugateHMCRunnerD<MinimumNorm2> HMCWrapper; //NB: This is the "Omelyan integrator"
+  typedef HMCWrapper::ImplPolicy GaugeImplPolicy;
+  MD.name    = std::string("MinimumNorm2");
+  MD.MDsteps = user_params.Steps;
+  MD.trajL   = user_params.TrajectoryLength;
+
+  HMCparameters HMCparams;
+  HMCparams.StartTrajectory  = user_params.StartTrajectory;
+  HMCparams.Trajectories     = user_params.Trajectories;
+  HMCparams.NoMetropolisUntil= 0;
+  HMCparams.StartingType     = user_params.StartingType;
+  HMCparams.MetropolisTest = user_params.MetropolisTest;
+  HMCparams.MD = MD;
+  HMCWrapper TheHMC(HMCparams);
+
+  // Grid from the command line arguments --grid and --mpi
+  TheHMC.Resources.AddFourDimGrid("gauge"); // use default simd lanes decomposition
+
+  CheckpointerParameters CPparams;
+  CPparams.config_prefix = "ckpoint_lat";
+  CPparams.rng_prefix    = "ckpoint_rng";
+  CPparams.saveInterval  = user_params.SaveInterval;
+  CPparams.format        = "IEEE64BIG";
+  TheHMC.Resources.LoadNerscCheckpointer(CPparams);
+
+  //Note that checkpointing saves the RNG state so that this initialization is required only for the very first configuration
+  RNGModuleParameters RNGpar;
+  RNGpar.serial_seeds = "1 2 3 4 5";
+  RNGpar.parallel_seeds = "6 7 8 9 10";
+  TheHMC.Resources.SetRNGSeeds(RNGpar);
+
+  typedef PlaquetteMod<GaugeImplPolicy> PlaqObs;
+  TheHMC.Resources.AddObservable<PlaqObs>();
+  //////////////////////////////////////////////
+
+  //aiming for ainv=2.068             me          Bob
+  //Estimated  a(ml+mres) [48ID] = 0.001048    0.00104 
+  //           a(mh+mres) [48ID] = 0.028847    0.02805
+  //Estimate Ls=12, b+c=2  mres~0.0003
+
+  const int Ls      = 12;
+  Real beta         = 1.946;
+  Real light_mass   = 0.00074;   //0.00104 - mres_approx;
+  Real strange_mass = 0.02775;    //0.02805 - mres_approx
+  Real pv_mass      = 1.0;
+  RealD M5  = 1.8;
+  RealD mobius_scale = 2.; //b+c
+
+  RealD mob_bmc = 1.0;
+  RealD mob_b = (mobius_scale + mob_bmc)/2.;
+  RealD mob_c = (mobius_scale - mob_bmc)/2.;
+
+  //Setup the Grids
+  auto UGridD   = TheHMC.Resources.GetCartesian();
+  auto UrbGridD = TheHMC.Resources.GetRBCartesian();
+  auto FGridD     = SpaceTimeGrid::makeFiveDimGrid(Ls,UGridD);
+  auto FrbGridD   = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGridD);
+
+  GridCartesian* UGridF = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd, vComplexF::Nsimd()), GridDefaultMpi());
+  GridRedBlackCartesian* UrbGridF = SpaceTimeGrid::makeFourDimRedBlackGrid(UGridF);
+  auto FGridF     = SpaceTimeGrid::makeFiveDimGrid(Ls,UGridF);
+  auto FrbGridF   = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGridF);
+
+  ConjugateIwasakiGaugeActionD GaugeAction(beta);
+
+  // temporarily need a gauge field
+  LatticeGaugeFieldD Ud(UGridD);
+  LatticeGaugeFieldF Uf(UGridF);
+ 
+  //Setup the BCs
+  FermionActionD::ImplParams Params;
+  for(int i=0;i<Nd-1;i++) Params.twists[i] = user_params.GparityDirs[i]; //G-parity directions
+  Params.twists[Nd-1] = 1; //APBC in time direction
+
+  std::vector<int> dirs4(Nd);
+  for(int i=0;i<Nd-1;i++) dirs4[i] = user_params.GparityDirs[i];
+  dirs4[Nd-1] = 0; //periodic gauge BC in time
+
+  GaugeImplPolicy::setDirections(dirs4); //gauge BC
+
+  //Run optional gauge field checksum checker and exit
+  if(file_load_check){
+    TheHMC.initializeGaugeFieldAndRNGs(Ud);
+    std::cout << GridLogMessage << " Done" << std::endl;
+    Grid_finalize();
+    return 0;
+  }
+
+
+  ////////////////////////////////////
+  // Collect actions
+  ////////////////////////////////////
+  ActionLevel<HMCWrapper::Field> Level1(1); //light quark + strange quark
+  ActionLevel<HMCWrapper::Field> Level2(4); //DSDR
+  ActionLevel<HMCWrapper::Field> Level3(2); //gauge
+
+
+  /////////////////////////////////////////////////////////////
+  // Light EOFA action
+  // have to be careful with the parameters, cf. Test_dwf_gpforce_eofa.cc
+  /////////////////////////////////////////////////////////////
+  typedef SchurDiagMooeeOperator<EOFAactionD,FermionFieldD> EOFAschuropD;
+  typedef SchurDiagMooeeOperator<EOFAactionF,FermionFieldF> EOFAschuropF;
+  typedef ExactOneFlavourRatioMixedPrecHeatbathPseudoFermionAction<FermionImplPolicyD, FermionImplPolicyF> EOFAmixPrecPFaction;
+  typedef MixedPrecisionConjugateGradientOperatorFunction<EOFAactionD, EOFAactionF, EOFAschuropD, EOFAschuropF> EOFA_mxCG;
+  typedef MixedPrecisionReliableUpdateConjugateGradientOperatorFunction<EOFAactionD, EOFAactionF, EOFAschuropD, EOFAschuropF> EOFA_relupCG;
+  
+  std::vector<RealD> eofa_light_masses = { light_mass ,  0.004,   0.016,   0.064,   0.256    };
+  std::vector<RealD> eofa_pv_masses =    { 0.004       , 0.016,   0.064,   0.256,   1.0      };
+  int n_light_hsb = 5;
+  assert(user_params.eofa_l.size() == n_light_hsb);
+  
+  EOFAmixPrecPFaction* EOFA_pfactions[n_light_hsb];
+
+  for(int i=0;i<n_light_hsb;i++){
+    RealD iml = eofa_light_masses[i];
+    RealD ipv = eofa_pv_masses[i];
+
+    EOFAactionD* LopD = new EOFAactionD(Ud, *FGridD, *FrbGridD, *UGridD, *UrbGridD, iml, iml, ipv, 0.0, -1, M5, mob_b, mob_c, Params);
+    EOFAactionF* LopF = new EOFAactionF(Uf, *FGridF, *FrbGridF, *UGridF, *UrbGridF, iml, iml, ipv, 0.0, -1, M5, mob_b, mob_c, Params);
+    EOFAactionD* RopD = new EOFAactionD(Ud, *FGridD, *FrbGridD, *UGridD, *UrbGridD, ipv, iml, ipv, -1.0, 1, M5, mob_b, mob_c, Params);
+    EOFAactionF* RopF = new EOFAactionF(Uf, *FGridF, *FrbGridF, *UGridF, *UrbGridF, ipv, iml, ipv, -1.0, 1, M5, mob_b, mob_c, Params);
+
+    EOFAschuropD* linopL_D = new EOFAschuropD(*LopD);
+    EOFAschuropD* linopR_D = new EOFAschuropD(*RopD);
+    
+    EOFAschuropF* linopL_F = new EOFAschuropF(*LopF);
+    EOFAschuropF* linopR_F = new EOFAschuropF(*RopF);
+
+#if 1
+    //Note reusing user_params.eofa_l.action(|md)_mixcg_inner_tolerance  as Delta for now
+    EOFA_relupCG* ActionMCG_L = new EOFA_relupCG(user_params.eofa_l[i].action_tolerance, user_params.eofa_l[i].action_mixcg_inner_tolerance, 50000, UGridF, FrbGridF, *LopF, *LopD, *linopL_F, *linopL_D);
+    EOFA_relupCG* ActionMCG_R = new EOFA_relupCG(user_params.eofa_l[i].action_tolerance, user_params.eofa_l[i].action_mixcg_inner_tolerance, 50000, UGridF, FrbGridF, *RopF, *RopD, *linopR_F, *linopR_D);
+
+    EOFA_relupCG* DerivMCG_L = new EOFA_relupCG(user_params.eofa_l[i].md_tolerance, user_params.eofa_l[i].md_mixcg_inner_tolerance, 50000, UGridF, FrbGridF, *LopF, *LopD, *linopL_F, *linopL_D);
+    EOFA_relupCG* DerivMCG_R = new EOFA_relupCG(user_params.eofa_l[i].md_tolerance, user_params.eofa_l[i].md_mixcg_inner_tolerance, 50000, UGridF, FrbGridF, *RopF, *RopD, *linopR_F, *linopR_D);
+
+#else
+    
+    EOFA_mxCG* ActionMCG_L = new EOFA_mxCG(user_params.eofa_l[i].action_tolerance, 10000, 1000, UGridF, FrbGridF, *LopF, *LopD, *linopL_F, *linopL_D);
+    ActionMCG_L->InnerTolerance = user_params.eofa_l[i].action_mixcg_inner_tolerance;
+    
+    EOFA_mxCG* ActionMCG_R = new EOFA_mxCG(user_params.eofa_l[i].action_tolerance, 10000, 1000, UGridF, FrbGridF, *RopF, *RopD, *linopR_F, *linopR_D);
+    ActionMCG_R->InnerTolerance = user_params.eofa_l[i].action_mixcg_inner_tolerance;
+    
+    EOFA_mxCG* DerivMCG_L = new EOFA_mxCG(user_params.eofa_l[i].md_tolerance, 10000, 1000, UGridF, FrbGridF, *LopF, *LopD, *linopL_F, *linopL_D);
+    DerivMCG_L->InnerTolerance = user_params.eofa_l[i].md_mixcg_inner_tolerance;
+    
+    EOFA_mxCG* DerivMCG_R = new EOFA_mxCG(user_params.eofa_l[i].md_tolerance, 10000, 1000, UGridF, FrbGridF, *RopF, *RopD, *linopR_F, *linopR_D);
+    DerivMCG_R->InnerTolerance = user_params.eofa_l[i].md_mixcg_inner_tolerance;
+    
+    std::cout << GridLogMessage << "Set EOFA action solver action tolerance outer=" << ActionMCG_L->Tolerance << " inner=" << ActionMCG_L->InnerTolerance << std::endl;
+    std::cout << GridLogMessage << "Set EOFA MD solver tolerance outer=" << DerivMCG_L->Tolerance << " inner=" << DerivMCG_L->InnerTolerance << std::endl;
+#endif
+
+    
+    EOFAmixPrecPFaction* EOFA = new EOFAmixPrecPFaction(*LopF, *RopF,
+							*LopD, *RopD, 
+							*ActionMCG_L, *ActionMCG_R, 
+							*ActionMCG_L, *ActionMCG_R, 
+							*DerivMCG_L, *DerivMCG_R, 
+							user_params.eofa_l[i].rat_params, true);
+    EOFA_pfactions[i] = EOFA;
+    Level1.push_back(EOFA);
+  }
+
+  ////////////////////////////////////
+  // Strange action
+  ////////////////////////////////////
+  FermionActionD Numerator_sD(Ud,*FGridD,*FrbGridD,*UGridD,*UrbGridD,strange_mass,M5,mob_b,mob_c,Params);
+  FermionActionD Denominator_sD(Ud,*FGridD,*FrbGridD,*UGridD,*UrbGridD, pv_mass,M5,mob_b,mob_c,Params);
+
+  FermionActionF Numerator_sF(Uf,*FGridF,*FrbGridF,*UGridF,*UrbGridF,strange_mass,M5,mob_b,mob_c,Params);
+  FermionActionF Denominator_sF(Uf,*FGridF,*FrbGridF,*UGridF,*UrbGridF, pv_mass,M5,mob_b,mob_c,Params);
+
+  RationalActionParams rat_act_params_s;
+  rat_act_params_s.inv_pow  = 4; // (M^dag M)^{1/4}
+  rat_act_params_s.precision= 60;
+  rat_act_params_s.MaxIter  = 10000;
+  user_params.rat_quo_s.Export(rat_act_params_s);
+  std::cout << GridLogMessage << " Heavy quark bounds check every " << rat_act_params_s.BoundsCheckFreq << " trajectories (avg)" << std::endl;
+
+  //MixedPrecRHMC Quotient_s(Denominator_sD, Numerator_sD, Denominator_sF, Numerator_sF, rat_act_params_s, user_params.rat_quo_s.reliable_update_freq); 
+  DoublePrecRHMC Quotient_s(Denominator_sD, Numerator_sD, rat_act_params_s); 
+  Level1.push_back(&Quotient_s);  
+
+  ///////////////////////////////////
+  // DSDR action
+  ///////////////////////////////////
+  RealD dsdr_mass=-1.8;   
+  //Use same DSDR twists as https://arxiv.org/pdf/1208.4412.pdf
+  RealD dsdr_epsilon_f = 0.02; //numerator (in determinant)
+  RealD dsdr_epsilon_b = 0.5; 
+  GparityWilsonTMFermionD Numerator_DSDR_D(Ud, *UGridD, *UrbGridD, dsdr_mass, dsdr_epsilon_f, Params);
+  GparityWilsonTMFermionF Numerator_DSDR_F(Uf, *UGridF, *UrbGridF, dsdr_mass, dsdr_epsilon_f, Params);
+
+  GparityWilsonTMFermionD Denominator_DSDR_D(Ud, *UGridD, *UrbGridD, dsdr_mass, dsdr_epsilon_b, Params);
+  GparityWilsonTMFermionF Denominator_DSDR_F(Uf, *UGridF, *UrbGridF, dsdr_mass, dsdr_epsilon_b, Params);
+ 
+  RationalActionParams rat_act_params_DSDR;
+  rat_act_params_DSDR.inv_pow  = 2; // (M^dag M)^{1/2}
+  rat_act_params_DSDR.precision= 60;
+  rat_act_params_DSDR.MaxIter  = 10000;
+  user_params.rat_quo_DSDR.Export(rat_act_params_DSDR);
+  std::cout << GridLogMessage << "DSDR quark bounds check every " << rat_act_params_DSDR.BoundsCheckFreq << " trajectories (avg)" << std::endl;
+
+  DoublePrecRHMC Quotient_DSDR(Denominator_DSDR_D, Numerator_DSDR_D, rat_act_params_DSDR);
+  Level2.push_back(&Quotient_DSDR);
+
+  /////////////////////////////////////////////////////////////
+  // Gauge action
+  /////////////////////////////////////////////////////////////
+  Level3.push_back(&GaugeAction);
+
+  TheHMC.TheAction.push_back(Level1);
+  TheHMC.TheAction.push_back(Level2);
+  TheHMC.TheAction.push_back(Level3);
+  std::cout << GridLogMessage << " Action complete "<< std::endl;
+
+
+  //Action tuning
+  bool 
+    tune_rhmc_s=false, eigenrange_s=false, 
+    tune_rhmc_DSDR=false, eigenrange_DSDR=false, 
+    check_eofa=false, 
+    upper_bound_eofa=false, lower_bound_eofa(false);
+
+  std::string lanc_params_s;
+  std::string lanc_params_DSDR;
+  int tune_rhmc_s_action_or_md;
+  int tune_rhmc_DSDR_action_or_md;
+  int eofa_which_hsb;
+
+  for(int i=1;i<argc;i++){
+    std::string sarg(argv[i]);
+    if(sarg == "--tune_rhmc_s"){
+      assert(i < argc-1);
+      tune_rhmc_s=true;
+      tune_rhmc_s_action_or_md = std::stoi(argv[i+1]);
+    }
+    else if(sarg == "--eigenrange_s"){
+      assert(i < argc-1);
+      eigenrange_s=true;
+      lanc_params_s = argv[i+1];
+    }
+    else if(sarg == "--tune_rhmc_DSDR"){
+      assert(i < argc-1);
+      tune_rhmc_DSDR=true;
+      tune_rhmc_DSDR_action_or_md = std::stoi(argv[i+1]);
+    }
+    else if(sarg == "--eigenrange_DSDR"){
+      assert(i < argc-1);
+      eigenrange_DSDR=true;
+      lanc_params_DSDR = argv[i+1];
+    }
+    else if(sarg == "--check_eofa"){
+      assert(i < argc-1);
+      check_eofa = true;
+      eofa_which_hsb = std::stoi(argv[i+1]); //-1 indicates all hasenbusch
+      assert(eofa_which_hsb == -1 || (eofa_which_hsb >= 0 && eofa_which_hsb < n_light_hsb) );
+    }
+    else if(sarg == "--upper_bound_eofa"){
+      assert(i < argc-1);
+      upper_bound_eofa = true;
+      eofa_which_hsb = std::stoi(argv[i+1]);
+      assert(eofa_which_hsb >= 0 && eofa_which_hsb < n_light_hsb);
+    }
+    else if(sarg == "--lower_bound_eofa"){
+      assert(i < argc-1);
+      lower_bound_eofa = true;      
+      eofa_which_hsb = std::stoi(argv[i+1]);
+      assert(eofa_which_hsb >= 0 && eofa_which_hsb < n_light_hsb);
+    }
+  }
+  if(tune_rhmc_s || eigenrange_s || tune_rhmc_DSDR || eigenrange_DSDR ||check_eofa || upper_bound_eofa || lower_bound_eofa) {
+    std::cout << GridLogMessage << "Running checks" << std::endl;
+    TheHMC.initializeGaugeFieldAndRNGs(Ud);
+
+    //std::cout << GridLogMessage << "EOFA action solver action tolerance outer=" << ActionMCG_L.Tolerance << " inner=" << ActionMCG_L.InnerTolerance << std::endl;
+    //std::cout << GridLogMessage << "EOFA MD solver tolerance outer=" << DerivMCG_L.Tolerance << " inner=" << DerivMCG_L.InnerTolerance << std::endl;
+
+
+    if(check_eofa){
+      if(eofa_which_hsb >= 0){
+	std::cout << GridLogMessage << "Starting checking EOFA Hasenbusch " << eofa_which_hsb << std::endl;
+	checkEOFA(*EOFA_pfactions[eofa_which_hsb], FGridD, TheHMC.Resources.GetParallelRNG(), Ud);
+	std::cout << GridLogMessage << "Finished checking EOFA Hasenbusch " << eofa_which_hsb << std::endl;
+      }else{
+	for(int i=0;i<n_light_hsb;i++){
+	  std::cout << GridLogMessage << "Starting checking EOFA Hasenbusch " << i << std::endl;
+	  checkEOFA(*EOFA_pfactions[i], FGridD, TheHMC.Resources.GetParallelRNG(), Ud);
+	  std::cout << GridLogMessage << "Finished checking EOFA Hasenbusch " << i << std::endl;
+	}
+      }
+    }	  
+    if(upper_bound_eofa) upperBoundEOFA(*EOFA_pfactions[eofa_which_hsb], FGridD, TheHMC.Resources.GetParallelRNG(), Ud);
+    if(lower_bound_eofa) lowerBoundEOFA(*EOFA_pfactions[eofa_which_hsb], FGridD, TheHMC.Resources.GetParallelRNG(), Ud);
+    if(eigenrange_s) computeEigenvalues<FermionActionD, FermionFieldD>(lanc_params_s, FGridD, FrbGridD, Ud, Numerator_sD, TheHMC.Resources.GetParallelRNG());
+    if(tune_rhmc_s) checkRHMC<FermionActionD, FermionFieldD, decltype(Quotient_s)>(FGridD, FrbGridD, Ud, Numerator_sD, Denominator_sD, Quotient_s, TheHMC.Resources.GetParallelRNG(), 4, "strange",  tune_rhmc_s_action_or_md);
+    if(eigenrange_DSDR) computeEigenvalues<GparityWilsonTMFermionD, GparityWilsonTMFermionD::FermionField>(lanc_params_DSDR, UGridD, UrbGridD, Ud, Numerator_DSDR_D, TheHMC.Resources.GetParallelRNG());
+    if(tune_rhmc_DSDR) checkRHMC<GparityWilsonTMFermionD, GparityWilsonTMFermionD::FermionField, decltype(Quotient_DSDR)>(UGridD, UrbGridD, Ud, Numerator_DSDR_D, Denominator_DSDR_D, Quotient_DSDR, TheHMC.Resources.GetParallelRNG(), 2, "DSDR", tune_rhmc_DSDR_action_or_md);
+
+
+    std::cout << GridLogMessage << " Done" << std::endl;
+    Grid_finalize();
+    return 0;
+  }
+
+
+  //Run the HMC
+  std::cout << GridLogMessage << " Running the HMC "<< std::endl;
+  TheHMC.Run();
+
+  std::cout << GridLogMessage << " Done" << std::endl;
+  Grid_finalize();
+  return 0;
+} // main

HMC/Mobius2p1f_DD_RHMC.cc

@@ -102,7 +102,7 @@ int main(int argc, char **argv) {
   SFRp.hi       = 30.0;
   SFRp.MaxIter  = 10000;
   SFRp.tolerance= 1.0e-8;
-  SFRp.mdtolerance= 1.0e-6;
+  SFRp.mdtolerance= 1.0e-5;
   SFRp.degree   = 16;
   SFRp.precision= 50;
   SFRp.BoundsCheckFreq=5;
@@ -112,7 +112,7 @@ int main(int argc, char **argv) {
   OFRp.hi       = 30.0;
   OFRp.MaxIter  = 10000;
   OFRp.tolerance= 1.0e-8;
-  OFRp.mdtolerance= 1.0e-6;
+  OFRp.mdtolerance= 1.0e-5;
   OFRp.degree   = 16;
   OFRp.precision= 50;
   OFRp.BoundsCheckFreq=5;
@@ -162,15 +162,17 @@ int main(int argc, char **argv) {
   FermionAction::ImplParams Params(boundary);
 
   double StoppingCondition = 1e-8;
+  double MDStoppingCondition = 1e-6;
   double MaxCGIterations = 30000;
   ConjugateGradient<FermionField>  CG(StoppingCondition,MaxCGIterations);
+  ConjugateGradient<FermionField>  MDCG(MDStoppingCondition,MaxCGIterations);
 
   ////////////////////////////////////
   // Collect actions
   ////////////////////////////////////
   ActionLevel<HMCWrapper::Field> Level1(1);
   ActionLevel<HMCWrapper::Field> Level2(4);
-  ActionLevel<HMCWrapper::Field> Level3(6);
+  ActionLevel<HMCWrapper::Field> Level3(8);
 
   ////////////////////////////////////
   // Strange action
@@ -226,7 +228,7 @@ int main(int argc, char **argv) {
     Numerators.push_back  (new FermionAction(U,*FGrid,*FrbGrid,*GridPtr,*GridRBPtr,light_num[h],M5,b,c, Params));
     Denominators.push_back(new FermionAction(U,*FGrid,*FrbGrid,*GridPtr,*GridRBPtr,light_den[h],M5,b,c, Params));
     if(h!=0) {
-      Quotients.push_back   (new TwoFlavourEvenOddRatioPseudoFermionAction<FermionImplPolicy>(*Numerators[h],*Denominators[h],CG,CG));
+      Quotients.push_back   (new TwoFlavourEvenOddRatioPseudoFermionAction<FermionImplPolicy>(*Numerators[h],*Denominators[h],MDCG,CG));
     } else {
       Bdys.push_back( new OneFlavourEvenOddRatioRationalPseudoFermionAction<FermionImplPolicy>(*Numerators[h],*Denominators[h],OFRp));
       Bdys.push_back( new OneFlavourEvenOddRatioRationalPseudoFermionAction<FermionImplPolicy>(*Numerators[h],*Denominators[h],OFRp));
@@ -241,7 +243,7 @@ int main(int argc, char **argv) {
   for(int h=0;h<nquo-1;h++){
     Level2.push_back(Quotients[h]);
   }
-  Level1.push_back(Quotients[nquo-1]); // PV dirichlet fix on coarse timestep
+  Level2.push_back(Quotients[nquo-1]);
 
   /////////////////////////////////////////////////////////////
   // Gauge action

HMC/Mobius2p1f_DD_RHMC_96I.cc

@@ -0,0 +1,419 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid
+
+Source file: ./tests/Test_hmc_EODWFRatio.cc
+
+Copyright (C) 2015-2016
+
+Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
+Author: Guido Cossu <guido.cossu@ed.ac.uk>
+
+This program is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 2 of the License, or
+(at your option) any later version.
+
+This program is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along
+with this program; if not, write to the Free Software Foundation, Inc.,
+51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+See the full license in the file "LICENSE" in the top level distribution
+directory
+*************************************************************************************/
+/*  END LEGAL */
+#include <Grid/Grid.h>
+
+int main(int argc, char **argv) {
+  using namespace Grid;
+
+  Grid_init(&argc, &argv);
+  int threads = GridThread::GetThreads();
+
+   // Typedefs to simplify notation
+  typedef WilsonImplR FermionImplPolicy;
+  typedef MobiusFermionR FermionAction;
+  typedef typename FermionAction::FermionField FermionField;
+
+  typedef Grid::XmlReader       Serialiser;
+
+  //::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
+  IntegratorParameters MD;
+  //  typedef GenericHMCRunner<LeapFrog> HMCWrapper;
+  //  MD.name    = std::string("Leap Frog");
+  //  typedef GenericHMCRunner<ForceGradient> HMCWrapper;
+  //  MD.name    = std::string("Force Gradient");
+  typedef GenericHMCRunner<MinimumNorm2> HMCWrapper;
+  MD.name    = std::string("MinimumNorm2");
+  MD.MDsteps =  6;
+  MD.trajL   = 1.0;
+
+  HMCparameters HMCparams;
+  HMCparams.StartTrajectory  = 1077;
+  HMCparams.Trajectories     = 1;
+  HMCparams.NoMetropolisUntil=  0;
+  // "[HotStart, ColdStart, TepidStart, CheckpointStart]\n";
+  //  HMCparams.StartingType     =std::string("ColdStart");
+  HMCparams.StartingType     =std::string("CheckpointStart");
+  HMCparams.MD = MD;
+  HMCWrapper TheHMC(HMCparams);
+
+  // Grid from the command line arguments --grid and --mpi
+  TheHMC.Resources.AddFourDimGrid("gauge"); // use default simd lanes decomposition
+
+  CheckpointerParameters CPparams;
+  CPparams.config_prefix = "ckpoint_DDHMC_lat";
+  CPparams.rng_prefix    = "ckpoint_DDHMC_rng";
+  CPparams.saveInterval  = 1;
+  CPparams.format        = "IEEE64BIG";
+  TheHMC.Resources.LoadNerscCheckpointer(CPparams);
+
+  RNGModuleParameters RNGpar;
+  RNGpar.serial_seeds = "1 2 3 4 5";
+  RNGpar.parallel_seeds = "6 7 8 9 10";
+  TheHMC.Resources.SetRNGSeeds(RNGpar);
+
+  // Construct observables
+  // here there is too much indirection
+  typedef PlaquetteMod<HMCWrapper::ImplPolicy> PlaqObs;
+  TheHMC.Resources.AddObservable<PlaqObs>();
+  //////////////////////////////////////////////
+
+  const int Ls      = 12;
+  RealD M5  = 1.8;
+  RealD b   = 1.5;
+  RealD c   = 0.5;
+  //  Real beta         = 2.31;
+  //  Real light_mass   = 5.4e-4;
+  Real beta         = 2.13;
+  Real light_mass   = 7.8e-4;
+  Real strange_mass = 0.02132;
+  Real pv_mass      = 1.0;
+  //  std::vector<Real> hasenbusch({ light_mass, 3.8e-3, 0.0145, 0.045, 0.108, 0.25, 0.51 , pv_mass });
+  std::vector<Real> hasenbusch({ light_mass, 0.0145, 0.045, 0.108, 0.25, 0.51 , pv_mass });
+
+  // FIXME:
+  // Same in MC and MD
+  // Need to mix precision too
+  OneFlavourRationalParams SFRp; // Strange
+  SFRp.lo       = 4.0e-3;
+  SFRp.hi       = 90.0;
+  SFRp.MaxIter  = 60000;
+  SFRp.tolerance= 1.0e-8;
+  SFRp.mdtolerance= 1.0e-4;
+  SFRp.degree   = 12;
+  SFRp.precision= 50;
+  SFRp.BoundsCheckFreq=0;
+
+  OneFlavourRationalParams OFRp; // Up/down
+  OFRp.lo       = 2.0e-5;
+  OFRp.hi       = 90.0;
+  OFRp.MaxIter  = 60000;
+  OFRp.tolerance= 1.0e-7;
+  OFRp.mdtolerance= 1.0e-4;
+  //  OFRp.degree   = 20; converges
+  //  OFRp.degree   = 16;
+  OFRp.degree   = 12;
+  OFRp.precision= 80;
+  OFRp.BoundsCheckFreq=0;
+
+  auto GridPtr   = TheHMC.Resources.GetCartesian();
+  auto GridRBPtr = TheHMC.Resources.GetRBCartesian();
+
+  ////////////////////////////////////////////////////////////////
+  // Domain decomposed
+  ////////////////////////////////////////////////////////////////
+  Coordinate latt4  = GridPtr->GlobalDimensions();
+  Coordinate mpi    = GridPtr->ProcessorGrid();
+  Coordinate shm;
+
+  GlobalSharedMemory::GetShmDims(mpi,shm);
+  
+  Coordinate CommDim(Nd);
+  for(int d=0;d<Nd;d++) CommDim[d]= (mpi[d]/shm[d])>1 ? 1 : 0;
+
+  Coordinate NonDirichlet(Nd+1,0);
+  Coordinate Dirichlet(Nd+1,0);
+  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];
+
+  Coordinate Block4(Nd);
+  //  Block4[0] = Dirichlet[1];
+  //  Block4[1] = Dirichlet[2];
+  //  Block4[2] = Dirichlet[3];
+  Block4[0] = 0;
+  Block4[1] = 0;
+  Block4[2] = 0;
+  Block4[3] = Dirichlet[4];
+
+  int Width=3;
+  TheHMC.Resources.SetMomentumFilter(new DDHMCFilter<WilsonImplR::Field>(Block4,Width));
+
+  //////////////////////////
+  // Fermion Grid
+  //////////////////////////
+  auto FGrid     = SpaceTimeGrid::makeFiveDimGrid(Ls,GridPtr);
+  auto FrbGrid   = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,GridPtr);
+
+  IwasakiGaugeActionR GaugeAction(beta);
+
+  // temporarily need a gauge field
+  LatticeGaugeField U(GridPtr);
+
+  std::cout << GridLogMessage << " Running the HMC "<< std::endl;
+  TheHMC.ReadCommandLine(argc,argv);  // params on CML or from param file
+  TheHMC.initializeGaugeFieldAndRNGs(U);
+
+
+  // These lines are unecessary if BC are all periodic
+  std::vector<Complex> boundary = {1,1,1,-1};
+  FermionAction::ImplParams Params(boundary);
+  Params.dirichlet=NonDirichlet;
+  FermionAction::ImplParams ParamsDir(boundary);
+  ParamsDir.dirichlet=Dirichlet;
+
+  //  double StoppingCondition = 1e-14;
+  //  double MDStoppingCondition = 1e-9;
+  double StoppingCondition = 1e-8;
+  double MDStoppingCondition = 1e-6;
+  double MaxCGIterations = 300000;
+  ConjugateGradient<FermionField>  CG(StoppingCondition,MaxCGIterations);
+  ConjugateGradient<FermionField>  MDCG(MDStoppingCondition,MaxCGIterations);
+
+  ////////////////////////////////////
+  // Collect actions
+  ////////////////////////////////////
+  ActionLevel<HMCWrapper::Field> Level1(1);
+  ActionLevel<HMCWrapper::Field> Level2(4);
+  ActionLevel<HMCWrapper::Field> Level3(8);
+
+  ////////////////////////////////////
+  // Strange action
+  ////////////////////////////////////
+  FermionAction StrangeOp (U,*FGrid,*FrbGrid,*GridPtr,*GridRBPtr,strange_mass,M5,b,c, Params);
+  FermionAction StrangePauliVillarsOp(U,*FGrid,*FrbGrid,*GridPtr,*GridRBPtr,pv_mass,  M5,b,c, Params);
+
+  FermionAction StrangeOpDir (U,*FGrid,*FrbGrid,*GridPtr,*GridRBPtr,strange_mass,M5,b,c, ParamsDir);
+  FermionAction StrangePauliVillarsOpDir(U,*FGrid,*FrbGrid,*GridPtr,*GridRBPtr,pv_mass,  M5,b,c, ParamsDir);
+  
+  OneFlavourEvenOddRatioRationalPseudoFermionAction<FermionImplPolicy> StrangePseudoFermionBdy(StrangeOpDir,StrangeOp,SFRp);
+  OneFlavourEvenOddRatioRationalPseudoFermionAction<FermionImplPolicy> StrangePseudoFermionLocal(StrangePauliVillarsOpDir,StrangeOpDir,SFRp);
+  OneFlavourEvenOddRatioRationalPseudoFermionAction<FermionImplPolicy> StrangePseudoFermionPVBdy(StrangePauliVillarsOp,StrangePauliVillarsOpDir,SFRp);
+  Level1.push_back(&StrangePseudoFermionBdy);
+  Level2.push_back(&StrangePseudoFermionLocal);
+  Level1.push_back(&StrangePseudoFermionPVBdy);
+
+  ////////////////////////////////////
+  // up down action
+  ////////////////////////////////////
+  std::vector<Real> light_den;
+  std::vector<Real> light_num;
+  std::vector<int> dirichlet_den;
+  std::vector<int> dirichlet_num;
+
+  int n_hasenbusch = hasenbusch.size();
+  light_den.push_back(light_mass);  dirichlet_den.push_back(0);
+  for(int h=0;h<n_hasenbusch;h++){
+    light_den.push_back(hasenbusch[h]); dirichlet_den.push_back(1);
+  }
+
+  for(int h=0;h<n_hasenbusch;h++){
+    light_num.push_back(hasenbusch[h]); dirichlet_num.push_back(1);
+  }
+  light_num.push_back(pv_mass);  dirichlet_num.push_back(0);
+
+  std::vector<FermionAction *> Numerators;
+  std::vector<FermionAction *> Denominators;
+  std::vector<TwoFlavourEvenOddRatioPseudoFermionAction<FermionImplPolicy> *> Quotients;
+  std::vector<OneFlavourEvenOddRatioRationalPseudoFermionAction<FermionImplPolicy> *> Bdys;
+  
+  for(int h=0;h<n_hasenbusch+1;h++){
+    std::cout << GridLogMessage
+	      << " 2f quotient Action ";
+    std::cout << "det D("<<light_den[h]<<")";
+    if ( dirichlet_den[h] ) std::cout << "^dirichlet    ";
+    std::cout << "/ det D("<<light_num[h]<<")";
+    if ( dirichlet_num[h] ) std::cout << "^dirichlet    ";
+    std::cout << std::endl;
+
+    FermionAction::ImplParams ParamsNum(boundary);
+    FermionAction::ImplParams ParamsDen(boundary);
+    
+    if ( dirichlet_num[h]==1) ParamsNum.dirichlet = Dirichlet;
+    else                      ParamsNum.dirichlet = NonDirichlet;
+    Numerators.push_back  (new FermionAction(U,*FGrid,*FrbGrid,*GridPtr,*GridRBPtr,light_num[h],M5,b,c, ParamsNum));
+
+    if ( dirichlet_den[h]==1) ParamsDen.dirichlet = Dirichlet;
+    else                      ParamsDen.dirichlet = NonDirichlet;
+    Denominators.push_back(new FermionAction(U,*FGrid,*FrbGrid,*GridPtr,*GridRBPtr,light_den[h],M5,b,c, ParamsDen));
+    
+    if(h!=0) {
+      Quotients.push_back   (new TwoFlavourEvenOddRatioPseudoFermionAction<FermionImplPolicy>(*Numerators[h],*Denominators[h],MDCG,CG));
+    } else {
+      Bdys.push_back( new OneFlavourEvenOddRatioRationalPseudoFermionAction<FermionImplPolicy>(*Numerators[h],*Denominators[h],OFRp));
+      Bdys.push_back( new OneFlavourEvenOddRatioRationalPseudoFermionAction<FermionImplPolicy>(*Numerators[h],*Denominators[h],OFRp));
+    }
+  }
+
+  int nquo=Quotients.size();
+  Level1.push_back(Bdys[0]);
+  Level1.push_back(Bdys[1]);
+  for(int h=0;h<nquo-1;h++){
+    Level2.push_back(Quotients[h]);
+  }
+  Level2.push_back(Quotients[nquo-1]);
+
+  /////////////////////////////////////////////////////////////
+  // Gauge action
+  /////////////////////////////////////////////////////////////
+  Level3.push_back(&GaugeAction);
+  TheHMC.TheAction.push_back(Level1);
+  TheHMC.TheAction.push_back(Level2);
+  TheHMC.TheAction.push_back(Level3);
+  std::cout << GridLogMessage << " Action complete "<< std::endl;
+  /////////////////////////////////////////////////////////////
+
+  if(1){
+    // TODO:
+    // i)  Break high bound, how rapidly does it break? Tune this test.
+    // ii) Break low bound, how rapidly?
+    // iii) Run lanczos
+    // iv)  Have CG return spectral range estimate
+    FermionField vec(StrangeOp.FermionRedBlackGrid());
+    FermionField res(StrangeOp.FermionRedBlackGrid());
+    vec = 1; // Fill with any old junk
+
+    std::cout << "Bounds check on strange operator mass "<< StrangeOp.Mass()<<std::endl;
+    SchurDifferentiableOperator<FermionImplPolicy> SdagS(StrangeOp);
+    HighBoundCheck(SdagS,vec,SFRp.hi);
+    ChebyBoundsCheck(SdagS,vec,SFRp.lo,SFRp.hi);
+    std::cout << "Strange inversion"<<std::endl;
+    res=Zero();
+    //    MDCG(SdagS,vec,res);
+
+
+    std::cout << "Bounds check on light quark operator mass "<< Denominators[0]->Mass() <<std::endl;
+    SchurDifferentiableOperator<FermionImplPolicy> UdagU(*Denominators[0]);
+    HighBoundCheck(UdagU,vec,OFRp.hi);
+    ChebyBoundsCheck(UdagU,vec,OFRp.lo,OFRp.hi);
+    std::cout << "light inversion"<<std::endl;
+    res=Zero();
+    //    MDCG(UdagU,vec,res);
+
+
+    std::cout << "Bounds check on strange dirichlet operator mass "<< StrangeOpDir.Mass()<<std::endl;
+    SchurDifferentiableOperator<FermionImplPolicy> SddagSd(StrangeOpDir);
+    HighBoundCheck(SddagSd,vec,OFRp.hi);
+    ChebyBoundsCheck(SddagSd,vec,OFRp.lo,OFRp.hi);
+    std::cout << "strange dirichlet inversion"<<std::endl;
+    res=Zero();
+    //    MDCG(SddagSd,vec,res);
+
+    std::cout << "Bounds check on light dirichlet operator mass "<< Numerators[0]->Mass()<<std::endl;
+    SchurDifferentiableOperator<FermionImplPolicy> UddagUd(*Numerators[0]);
+    HighBoundCheck(UddagUd,vec,OFRp.hi);
+    ChebyBoundsCheck(UddagUd,vec,OFRp.lo,OFRp.hi);
+    std::cout << "light dirichlet inversion"<<std::endl;
+    res=Zero();
+    //MDCG(UddagUd,vec,res);
+
+    
+    auto grid4= GridPtr;
+    auto rbgrid4= GridRBPtr;
+    auto rbgrid = StrangeOp.FermionRedBlackGrid();
+    auto grid = StrangeOp.FermionGrid();
+    if(1){
+    const int Nstop = 5;
+    const int Nk = 20;
+    const int Np = 20;
+    const int Nm = Nk+Np;
+    const int MaxIt= 10000;
+    int Nconv;
+    RealD resid = 1.0e-5;
+    if(0)
+    {
+      int order = 501;
+      RealD bound = 5.0e-4;
+      std::cout << GridLogMessage << " Lanczos for dirichlet bound " << bound<<" order "<< order<<std::endl;
+      Chebyshev<FermionField> Cheby(bound,90.,order);
+      FunctionHermOp<FermionField> OpCheby(Cheby,UddagUd);
+      PlainHermOp<FermionField> Op     (UddagUd);
+      ImplicitlyRestartedLanczos<FermionField> IRL(OpCheby,Op,Nstop,Nk,Nm,resid,MaxIt);
+      std::vector<RealD>          eval(Nm);
+      std::vector<FermionField> evec(Nm,rbgrid);
+      FermionField    src(rbgrid);src = 1.0;
+      IRL.calc(eval,evec,src,Nconv);
+      
+      FermionField tmp(rbgrid);
+      FermionField ftmp(grid);
+      FermionField ftmp4(grid4);
+      for(int ev=0;ev<evec.size();ev++){
+	Gamma GT(Gamma::Algebra::GammaT);
+	std::cout << " evec " << ev << std::endl;
+	tmp = evec[ev] + GT*evec[ev];
+	DumpSliceNorm(" 1+gammaT ",tmp,Nd);
+	tmp = evec[ev] - GT*evec[ev];
+	DumpSliceNorm(" 1-gammaT ",tmp,Nd);
+      }
+      for(int e=0;e<10;e++){
+	std::cout << " Dirichlet evec "<<e<<std::endl;
+	tmp = evec[e];
+	for(int s=0;s<Ls;s++){
+	  ftmp=Zero();
+	  setCheckerboard(ftmp,tmp);
+	  ExtractSlice(ftmp4,ftmp,s,0);
+	  std::cout << "s-slice "<<s<< " evec[0] " << std::endl;
+	  DumpSliceNorm(" s-slice ",ftmp4,Nd-1);
+	}
+      }
+    }
+    if(1)
+    {
+      int order = 2001;
+      RealD bound = 6.0e-5;
+      std::cout << GridLogMessage << " Lanczos for full operator  bound " << bound<<" order "<< order<<std::endl;
+      Chebyshev<FermionField> Cheby(bound,90.,order);
+      FunctionHermOp<FermionField> OpCheby(Cheby,UdagU);
+      PlainHermOp<FermionField> Op     (UdagU);
+      ImplicitlyRestartedLanczos<FermionField> IRL(OpCheby,Op,Nstop,Nk,Nm,resid,MaxIt);
+      std::vector<RealD>          eval(Nm);
+      std::vector<FermionField> evec(Nm,rbgrid);
+      FermionField    src(rbgrid); src = 1.0;
+      IRL.calc(eval,evec,src,Nconv);
+
+      FermionField tmp(rbgrid);
+      FermionField ftmp(grid);
+      FermionField ftmp4(grid4);
+      for(int e=0;e<evec.size();e++){
+	std::cout << " Full evec "<<e<<std::endl;
+	tmp = evec[e];
+	for(int s=0;s<Ls;s++){
+	  ftmp=Zero();
+	  setCheckerboard(ftmp,tmp);
+	  ExtractSlice(ftmp4,ftmp,s,0);
+	  std::cout << "s-slice "<<s<< " evec[0] " << std::endl;
+	  DumpSliceNorm(" s-slice ",ftmp4,Nd-1);
+	}
+      }
+
+    }
+    Grid_finalize();
+    std::cout << " All done "<<std::endl;
+    exit(EXIT_SUCCESS);
+    }
+  }
+
+  TheHMC.Run();  // no smearing
+
+  Grid_finalize();
+} // main
+
+
+

HMC/Mobius2p1f_DD_RHMC_96I_mixed.cc

@@ -0,0 +1,444 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid
+
+Source file: ./tests/Test_hmc_EODWFRatio.cc
+
+Copyright (C) 2015-2016
+
+Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
+Author: Guido Cossu <guido.cossu@ed.ac.uk>
+
+This program is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 2 of the License, or
+(at your option) any later version.
+
+This program is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along
+with this program; if not, write to the Free Software Foundation, Inc.,
+51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+See the full license in the file "LICENSE" in the top level distribution
+directory
+*************************************************************************************/
+/*  END LEGAL */
+#include <Grid/Grid.h>
+
+NAMESPACE_BEGIN(Grid);
+
+template<class FermionOperatorD, class FermionOperatorF, class SchurOperatorD, class  SchurOperatorF> 
+  class MixedPrecisionConjugateGradientOperatorFunction : public OperatorFunction<typename FermionOperatorD::FermionField> {
+  public:
+    typedef typename FermionOperatorD::FermionField FieldD;
+    typedef typename FermionOperatorF::FermionField FieldF;
+
+    using OperatorFunction<FieldD>::operator();
+
+    RealD   Tolerance;
+    RealD   InnerTolerance; //Initial tolerance for inner CG. Defaults to Tolerance but can be changed
+    Integer MaxInnerIterations;
+    Integer MaxOuterIterations;
+    GridBase* SinglePrecGrid4; //Grid for single-precision fields
+    GridBase* SinglePrecGrid5; //Grid for single-precision fields
+    RealD OuterLoopNormMult; //Stop the outer loop and move to a final double prec solve when the residual is OuterLoopNormMult * Tolerance
+
+    FermionOperatorF &FermOpF;
+    FermionOperatorD &FermOpD;;
+    SchurOperatorF &LinOpF;
+    SchurOperatorD &LinOpD;
+
+    Integer TotalInnerIterations; //Number of inner CG iterations
+    Integer TotalOuterIterations; //Number of restarts
+    Integer TotalFinalStepIterations; //Number of CG iterations in final patch-up step
+
+    MixedPrecisionConjugateGradientOperatorFunction(RealD tol, 
+						    Integer maxinnerit, 
+						    Integer maxouterit, 
+						    GridBase* _sp_grid4, 
+						    GridBase* _sp_grid5, 
+						    FermionOperatorF &_FermOpF,
+						    FermionOperatorD &_FermOpD,
+						    SchurOperatorF   &_LinOpF,
+						    SchurOperatorD   &_LinOpD): 
+      LinOpF(_LinOpF),
+      LinOpD(_LinOpD),
+      FermOpF(_FermOpF),
+      FermOpD(_FermOpD),
+      Tolerance(tol), 
+      InnerTolerance(tol), 
+      MaxInnerIterations(maxinnerit), 
+      MaxOuterIterations(maxouterit), 
+      SinglePrecGrid4(_sp_grid4),
+      SinglePrecGrid5(_sp_grid5),
+      OuterLoopNormMult(100.) 
+    { 
+      /* Debugging instances of objects; references are stored
+      std::cout << GridLogMessage << " Mixed precision CG wrapper LinOpF " <<std::hex<< &LinOpF<<std::dec <<std::endl;
+      std::cout << GridLogMessage << " Mixed precision CG wrapper LinOpD " <<std::hex<< &LinOpD<<std::dec <<std::endl;
+      std::cout << GridLogMessage << " Mixed precision CG wrapper FermOpF " <<std::hex<< &FermOpF<<std::dec <<std::endl;
+      std::cout << GridLogMessage << " Mixed precision CG wrapper FermOpD " <<std::hex<< &FermOpD<<std::dec <<std::endl;
+      */
+    };
+
+    void operator()(LinearOperatorBase<FieldD> &LinOpU, const FieldD &src, FieldD &psi) {
+
+      std::cout << GridLogMessage << " Mixed precision CG wrapper operator() "<<std::endl;
+
+      SchurOperatorD * SchurOpU = static_cast<SchurOperatorD *>(&LinOpU);
+      
+      //      std::cout << GridLogMessage << " Mixed precision CG wrapper operator() FermOpU " <<std::hex<< &(SchurOpU->_Mat)<<std::dec <<std::endl;
+      //      std::cout << GridLogMessage << " Mixed precision CG wrapper operator() FermOpD " <<std::hex<< &(LinOpD._Mat) <<std::dec <<std::endl;
+      // Assumption made in code to extract gauge field
+      // We could avoid storing LinopD reference alltogether ?
+      assert(&(SchurOpU->_Mat)==&(LinOpD._Mat));
+
+      ////////////////////////////////////////////////////////////////////////////////////
+      // Must snarf a single precision copy of the gauge field in Linop_d argument
+      ////////////////////////////////////////////////////////////////////////////////////
+      typedef typename FermionOperatorF::GaugeField GaugeFieldF;
+      typedef typename FermionOperatorF::GaugeLinkField GaugeLinkFieldF;
+      typedef typename FermionOperatorD::GaugeField GaugeFieldD;
+      typedef typename FermionOperatorD::GaugeLinkField GaugeLinkFieldD;
+
+      GridBase * GridPtrF = SinglePrecGrid4;
+      GridBase * GridPtrD = FermOpD.Umu.Grid();
+      GaugeFieldF     U_f  (GridPtrF);
+      GaugeLinkFieldF Umu_f(GridPtrF);
+      //      std::cout << " Dim gauge field "<<GridPtrF->Nd()<<std::endl; // 4d
+      //      std::cout << " Dim gauge field "<<GridPtrD->Nd()<<std::endl; // 4d
+
+      ////////////////////////////////////////////////////////////////////////////////////
+      // Moving this to a Clone method of fermion operator would allow to duplicate the 
+      // physics parameters and decrease gauge field copies
+      ////////////////////////////////////////////////////////////////////////////////////
+      GaugeLinkFieldD Umu_d(GridPtrD);
+      for(int mu=0;mu<Nd*2;mu++){ 
+	Umu_d = PeekIndex<LorentzIndex>(FermOpD.Umu, mu);
+	precisionChange(Umu_f,Umu_d);
+	PokeIndex<LorentzIndex>(FermOpF.Umu, Umu_f, mu);
+      }
+      pickCheckerboard(Even,FermOpF.UmuEven,FermOpF.Umu);
+      pickCheckerboard(Odd ,FermOpF.UmuOdd ,FermOpF.Umu);
+
+      ////////////////////////////////////////////////////////////////////////////////////
+      // Make a mixed precision conjugate gradient
+      ////////////////////////////////////////////////////////////////////////////////////
+      MixedPrecisionConjugateGradient<FieldD,FieldF> MPCG(Tolerance,MaxInnerIterations,MaxOuterIterations,SinglePrecGrid5,LinOpF,LinOpD);
+      std::cout << GridLogMessage << "Calling mixed precision Conjugate Gradient" <<std::endl;
+      MPCG(src,psi);
+    }
+  };
+
+NAMESPACE_END(Grid);
+
+
+int main(int argc, char **argv) {
+  using namespace Grid;
+
+  Grid_init(&argc, &argv);
+  int threads = GridThread::GetThreads();
+
+   // Typedefs to simplify notation
+  typedef WilsonImplR FermionImplPolicy;
+  typedef WilsonImplF FermionImplPolicyF;
+  typedef MobiusFermionR FermionAction;
+  typedef MobiusFermionF FermionActionF;
+  typedef typename FermionAction::FermionField FermionField;
+  typedef typename FermionActionF::FermionField FermionFieldF;
+
+  typedef Grid::XmlReader       Serialiser;
+
+  //::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
+  IntegratorParameters MD;
+  //  typedef GenericHMCRunner<LeapFrog> HMCWrapper;
+  //  MD.name    = std::string("Leap Frog");
+  //  typedef GenericHMCRunner<ForceGradient> HMCWrapper;
+  //  MD.name    = std::string("Force Gradient");
+  typedef GenericHMCRunner<MinimumNorm2> HMCWrapper;
+  MD.name    = std::string("MinimumNorm2");
+  MD.MDsteps =  4;
+  MD.trajL   = 1.0;
+
+  HMCparameters HMCparams;
+  HMCparams.StartTrajectory  = 1077;
+  HMCparams.Trajectories     = 1;
+  HMCparams.NoMetropolisUntil=  0;
+  // "[HotStart, ColdStart, TepidStart, CheckpointStart]\n";
+  //  HMCparams.StartingType     =std::string("ColdStart");
+  HMCparams.StartingType     =std::string("CheckpointStart");
+  HMCparams.MD = MD;
+  HMCWrapper TheHMC(HMCparams);
+
+  // Grid from the command line arguments --grid and --mpi
+  TheHMC.Resources.AddFourDimGrid("gauge"); // use default simd lanes decomposition
+
+  CheckpointerParameters CPparams;
+  CPparams.config_prefix = "ckpoint_DDHMC_lat";
+  CPparams.rng_prefix    = "ckpoint_DDHMC_rng";
+  CPparams.saveInterval  = 1;
+  CPparams.format        = "IEEE64BIG";
+  TheHMC.Resources.LoadNerscCheckpointer(CPparams);
+
+  RNGModuleParameters RNGpar;
+  RNGpar.serial_seeds = "1 2 3 4 5";
+  RNGpar.parallel_seeds = "6 7 8 9 10";
+  TheHMC.Resources.SetRNGSeeds(RNGpar);
+
+  // Construct observables
+  // here there is too much indirection
+  typedef PlaquetteMod<HMCWrapper::ImplPolicy> PlaqObs;
+  TheHMC.Resources.AddObservable<PlaqObs>();
+  //////////////////////////////////////////////
+
+  const int Ls      = 12;
+  RealD M5  = 1.8;
+  RealD b   = 1.5;
+  RealD c   = 0.5;
+  Real beta         = 2.31;
+  //  Real light_mass   = 5.4e-4;
+  Real light_mass   = 7.8e-4;
+  Real strange_mass = 0.02132;
+  Real pv_mass      = 1.0;
+  std::vector<Real> hasenbusch({ light_mass, 3.8e-3, 0.0145, 0.045, 0.108, 0.25, 0.51 , pv_mass });
+
+  // FIXME:
+  // Same in MC and MD
+  // Need to mix precision too
+  OneFlavourRationalParams SFRp; // Strange
+  SFRp.lo       = 4.0e-3;
+  SFRp.hi       = 90.0;
+  SFRp.MaxIter  = 60000;
+  SFRp.tolerance= 1.0e-8;
+  SFRp.mdtolerance= 1.0e-6;
+  SFRp.degree   = 12;
+  SFRp.precision= 50;
+  SFRp.BoundsCheckFreq=0;
+
+  OneFlavourRationalParams OFRp; // Up/down
+  OFRp.lo       = 2.0e-5;
+  OFRp.hi       = 90.0;
+  OFRp.MaxIter  = 60000;
+  OFRp.tolerance= 1.0e-8;
+  OFRp.mdtolerance= 1.0e-6;
+  //  OFRp.degree   = 20; converges
+  //  OFRp.degree   = 16;
+  OFRp.degree   = 12;
+  OFRp.precision= 80;
+  OFRp.BoundsCheckFreq=0;
+
+  auto GridPtr   = TheHMC.Resources.GetCartesian();
+  auto GridRBPtr = TheHMC.Resources.GetRBCartesian();
+
+  typedef SchurDiagMooeeOperator<FermionActionF,FermionFieldF> LinearOperatorF;
+  typedef SchurDiagMooeeOperator<FermionAction ,FermionField > LinearOperatorD;
+  typedef MixedPrecisionConjugateGradientOperatorFunction<MobiusFermionD,MobiusFermionF,LinearOperatorD,LinearOperatorF> MxPCG;
+
+  ////////////////////////////////////////////////////////////////
+  // Domain decomposed
+  ////////////////////////////////////////////////////////////////
+  Coordinate latt4  = GridPtr->GlobalDimensions();
+  Coordinate mpi    = GridPtr->ProcessorGrid();
+  Coordinate shm;
+
+  GlobalSharedMemory::GetShmDims(mpi,shm);
+  
+  Coordinate CommDim(Nd);
+  for(int d=0;d<Nd;d++) CommDim[d]= (mpi[d]/shm[d])>1 ? 1 : 0;
+
+  Coordinate NonDirichlet(Nd+1,0);
+  Coordinate Dirichlet(Nd+1,0);
+  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];
+
+  Coordinate Block4(Nd);
+  Block4[0] = Dirichlet[1];
+  Block4[1] = Dirichlet[2];
+  Block4[2] = Dirichlet[3];
+  Block4[3] = Dirichlet[4];
+
+  int Width=3;
+  TheHMC.Resources.SetMomentumFilter(new DDHMCFilter<WilsonImplR::Field>(Block4,Width));
+
+  //////////////////////////
+  // Fermion Grids
+  //////////////////////////
+  auto FGrid     = SpaceTimeGrid::makeFiveDimGrid(Ls,GridPtr);
+  auto FrbGrid   = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,GridPtr);
+
+  Coordinate simdF = GridDefaultSimd(Nd,vComplexF::Nsimd());
+  auto GridPtrF   = SpaceTimeGrid::makeFourDimGrid(latt4,simdF,mpi);
+  auto GridRBPtrF = SpaceTimeGrid::makeFourDimRedBlackGrid(GridPtrF);
+  auto FGridF     = SpaceTimeGrid::makeFiveDimGrid(Ls,GridPtrF);
+  auto FrbGridF   = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,GridPtrF);
+
+  IwasakiGaugeActionR GaugeAction(beta);
+
+  // temporarily need a gauge field
+  LatticeGaugeField U(GridPtr);
+  LatticeGaugeFieldF UF(GridPtrF);
+
+  std::cout << GridLogMessage << " Running the HMC "<< std::endl;
+  TheHMC.ReadCommandLine(argc,argv);  // params on CML or from param file
+  TheHMC.initializeGaugeFieldAndRNGs(U);
+
+
+  // These lines are unecessary if BC are all periodic
+  std::vector<Complex> boundary = {1,1,1,-1};
+  FermionAction::ImplParams Params(boundary);
+  Params.dirichlet=NonDirichlet;
+  FermionAction::ImplParams ParamsDir(boundary);
+  ParamsDir.dirichlet=Dirichlet;
+
+  //  double StoppingCondition = 1e-14;
+  //  double MDStoppingCondition = 1e-9;
+  double StoppingCondition = 1e-10;
+  double MDStoppingCondition = 1e-7;
+  double MDStoppingConditionLoose = 1e-6;
+  double MaxCGIterations = 300000;
+  ConjugateGradient<FermionField>  CG(StoppingCondition,MaxCGIterations);
+  ConjugateGradient<FermionField>  MDCG(MDStoppingCondition,MaxCGIterations);
+
+  ////////////////////////////////////
+  // Collect actions
+  ////////////////////////////////////
+  ActionLevel<HMCWrapper::Field> Level1(1);
+  ActionLevel<HMCWrapper::Field> Level2(4);
+  ActionLevel<HMCWrapper::Field> Level3(8);
+
+  ////////////////////////////////////
+  // Strange action
+  ////////////////////////////////////
+  FermionAction StrangeOp (U,*FGrid,*FrbGrid,*GridPtr,*GridRBPtr,strange_mass,M5,b,c, Params);
+  FermionAction StrangePauliVillarsOp(U,*FGrid,*FrbGrid,*GridPtr,*GridRBPtr,pv_mass,  M5,b,c, Params);
+
+  FermionAction StrangeOpDir (U,*FGrid,*FrbGrid,*GridPtr,*GridRBPtr,strange_mass,M5,b,c, ParamsDir);
+  FermionAction StrangePauliVillarsOpDir(U,*FGrid,*FrbGrid,*GridPtr,*GridRBPtr,pv_mass,  M5,b,c, ParamsDir);
+  
+  OneFlavourEvenOddRatioRationalPseudoFermionAction<FermionImplPolicy> StrangePseudoFermionBdy(StrangeOpDir,StrangeOp,SFRp);
+  OneFlavourEvenOddRatioRationalPseudoFermionAction<FermionImplPolicy> StrangePseudoFermionLocal(StrangePauliVillarsOpDir,StrangeOpDir,SFRp);
+  OneFlavourEvenOddRatioRationalPseudoFermionAction<FermionImplPolicy> StrangePseudoFermionPVBdy(StrangePauliVillarsOp,StrangePauliVillarsOpDir,SFRp);
+  Level1.push_back(&StrangePseudoFermionBdy);
+  Level2.push_back(&StrangePseudoFermionLocal);
+  Level1.push_back(&StrangePseudoFermionPVBdy);
+
+  ////////////////////////////////////
+  // up down action
+  ////////////////////////////////////
+  std::vector<Real> light_den;
+  std::vector<Real> light_num;
+  std::vector<int> dirichlet_den;
+  std::vector<int> dirichlet_num;
+
+  int n_hasenbusch = hasenbusch.size();
+  light_den.push_back(light_mass);  dirichlet_den.push_back(0);
+  for(int h=0;h<n_hasenbusch;h++){
+    light_den.push_back(hasenbusch[h]); dirichlet_den.push_back(1);
+  }
+
+  for(int h=0;h<n_hasenbusch;h++){
+    light_num.push_back(hasenbusch[h]); dirichlet_num.push_back(1);
+  }
+  light_num.push_back(pv_mass);  dirichlet_num.push_back(0);
+
+  std::vector<FermionAction *> Numerators;
+  std::vector<FermionAction *> Denominators;
+  std::vector<FermionActionF *> DenominatorsF;
+  std::vector<TwoFlavourEvenOddRatioPseudoFermionAction<FermionImplPolicy> *> Quotients;
+  std::vector<OneFlavourEvenOddRatioRationalPseudoFermionAction<FermionImplPolicy> *> Bdys;
+  std::vector<MxPCG *> ActionMPCG;
+  std::vector<MxPCG *> MPCG;
+
+  typedef SchurDiagMooeeOperator<FermionActionF,FermionFieldF> LinearOperatorF;
+  typedef SchurDiagMooeeOperator<FermionAction ,FermionField > LinearOperatorD;
+  std::vector<LinearOperatorD *> LinOpD;
+  std::vector<LinearOperatorF *> LinOpF; 
+  
+  for(int h=0;h<n_hasenbusch+1;h++){
+    std::cout << GridLogMessage
+	      << " 2f quotient Action ";
+    std::cout << "det D("<<light_den[h]<<")";
+    if ( dirichlet_den[h] ) std::cout << "^dirichlet    ";
+    std::cout << "/ det D("<<light_num[h]<<")";
+    if ( dirichlet_num[h] ) std::cout << "^dirichlet    ";
+    std::cout << std::endl;
+
+    FermionAction::ImplParams ParamsNum(boundary);
+    FermionAction::ImplParams ParamsDen(boundary);
+    FermionActionF::ImplParams ParamsDenF(boundary);
+    
+    if ( dirichlet_num[h]==1) ParamsNum.dirichlet = Dirichlet;
+    else                      ParamsNum.dirichlet = NonDirichlet;
+    Numerators.push_back  (new FermionAction(U,*FGrid,*FrbGrid,*GridPtr,*GridRBPtr,light_num[h],M5,b,c, ParamsNum));
+
+    if ( dirichlet_den[h]==1) ParamsDen.dirichlet = Dirichlet;
+    else                      ParamsDen.dirichlet = NonDirichlet;
+
+    Denominators.push_back(new FermionAction(U,*FGrid,*FrbGrid,*GridPtr,*GridRBPtr,light_den[h],M5,b,c, ParamsDen));
+
+    ParamsDenF.dirichlet = ParamsDen.dirichlet;
+    DenominatorsF.push_back(new FermionActionF(UF,*FGridF,*FrbGridF,*GridPtrF,*GridRBPtrF,light_den[h],M5,b,c, ParamsDenF));
+
+    LinOpD.push_back(new LinearOperatorD(*Denominators[h]));
+    LinOpF.push_back(new LinearOperatorF(*DenominatorsF[h]));
+
+    double conv  = MDStoppingCondition;
+    if (h<3) conv= MDStoppingConditionLoose; // Relax on first two hasenbusch factors
+    const int MX_inner = 5000;
+    MPCG.push_back(new MxPCG(conv,
+			     MX_inner,
+			     MaxCGIterations,
+			     GridPtrF,
+			     FrbGridF,
+			     *DenominatorsF[h],*Denominators[h],
+			     *LinOpF[h], *LinOpD[h]) );
+
+    ActionMPCG.push_back(new MxPCG(StoppingCondition,
+				   MX_inner,
+				   MaxCGIterations,
+				   GridPtrF,
+				   FrbGridF,
+				   *DenominatorsF[h],*Denominators[h],
+				   *LinOpF[h], *LinOpD[h]) );
+
+    
+    if(h!=0) {
+      //      Quotients.push_back (new TwoFlavourEvenOddRatioPseudoFermionAction<FermionImplPolicy>(*Numerators[h],*Denominators[h],MDCG,CG));
+      Quotients.push_back (new TwoFlavourEvenOddRatioPseudoFermionAction<FermionImplPolicy>(*Numerators[h],*Denominators[h],*MPCG[h],*ActionMPCG[h],CG));
+    } else {
+      Bdys.push_back( new OneFlavourEvenOddRatioRationalPseudoFermionAction<FermionImplPolicy>(*Numerators[h],*Denominators[h],OFRp));
+      Bdys.push_back( new OneFlavourEvenOddRatioRationalPseudoFermionAction<FermionImplPolicy>(*Numerators[h],*Denominators[h],OFRp));
+    }
+  }
+
+  int nquo=Quotients.size();
+  Level1.push_back(Bdys[0]);
+  Level1.push_back(Bdys[1]);
+  for(int h=0;h<nquo-1;h++){
+    Level2.push_back(Quotients[h]);
+  }
+  Level2.push_back(Quotients[nquo-1]);
+
+  /////////////////////////////////////////////////////////////
+  // Gauge action
+  /////////////////////////////////////////////////////////////
+  Level3.push_back(&GaugeAction);
+  TheHMC.TheAction.push_back(Level1);
+  TheHMC.TheAction.push_back(Level2);
+  TheHMC.TheAction.push_back(Level3);
+  std::cout << GridLogMessage << " Action complete "<< std::endl;
+  /////////////////////////////////////////////////////////////
+
+  TheHMC.Run();  // no smearing
+
+  Grid_finalize();
+} // main
+
+
+

HMC/RNGstate.cc

@@ -0,0 +1,53 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid
+
+Source file:
+
+Copyright (C) 2015-2016
+
+Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
+
+This program is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 2 of the License, or
+(at your option) any later version.
+
+This program is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along
+with this program; if not, write to the Free Software Foundation, Inc.,
+51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+See the full license in the file "LICENSE" in the top level distribution
+directory
+*************************************************************************************/
+/*  END LEGAL */
+#include <Grid/Grid.h>
+
+int main(int argc, char **argv)
+{
+  using namespace Grid;
+
+  Grid_init(&argc, &argv);
+
+  Coordinate latt4  = GridDefaultLatt();
+  Coordinate mpi    = GridDefaultMpi();
+  Coordinate simd   = GridDefaultSimd(Nd,vComplexD::Nsimd());
+
+  GridCartesian         * UGrid   = SpaceTimeGrid::makeFourDimGrid(latt4,simd,mpi);
+
+  GridSerialRNG   sRNG;         sRNG.SeedUniqueString(std::string("The Serial RNG"));
+  GridParallelRNG pRNG(UGrid);  pRNG.SeedUniqueString(std::string("The 4D RNG"));
+
+  std::string rngfile("ckpoint_rng.0");
+  NerscIO::writeRNGState(sRNG, pRNG, rngfile);
+  
+  Grid_finalize();
+}
+
+
+

benchmarks/Benchmark_dwf.cc

@@ -191,9 +191,7 @@ int main (int argc, char ** argv)
     std::cout<<GridLogMessage<<"Called warmup"<<std::endl;
     double t0=usecond();
     for(int i=0;i<ncall;i++){
-      __SSC_START;
       Dw.Dhop(src,result,0);
-      __SSC_STOP;
     }
     double t1=usecond();
     FGrid->Barrier();

benchmarks/Benchmark_dwf_fp32.cc

@@ -249,8 +249,9 @@ void Benchmark(int Ls, Coordinate Dirichlet)
   if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptInlineAsm ) std::cout << GridLogMessage<< "* Using Asm Nc=3   WilsonKernels" <<std::endl;
   std::cout << GridLogMessage<< "*****************************************************************" <<std::endl;
 
-  DomainWallFermionF Dw(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
-  Dw.DirichletBlock(Dirichlet);
+  DomainWallFermionF::ImplParams p;
+  p.dirichlet=Dirichlet;
+  DomainWallFermionF Dw(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,p);
   Dw.ImportGauge(Umu);
   
   int ncall =300;
@@ -261,9 +262,7 @@ void Benchmark(int Ls, Coordinate Dirichlet)
     std::cout<<GridLogMessage<<"Called warmup"<<std::endl;
     double t0=usecond();
     for(int i=0;i<ncall;i++){
-      __SSC_START;
       Dw.Dhop(src,result,0);
-      __SSC_STOP;
     }
     double t1=usecond();
     FGrid->Barrier();

benchmarks/Benchmark_mooee.cc

@@ -81,8 +81,8 @@ int main (int argc, char ** argv)
     Vector<Coeff_t> diag = Dw.bs;
     Vector<Coeff_t> upper= Dw.cs;
     Vector<Coeff_t> lower= Dw.cs;
-    upper[Ls-1]=-Dw.mass*upper[Ls-1];
-    lower[0]   =-Dw.mass*lower[0];
+    upper[Ls-1]=-Dw.mass_minus*upper[Ls-1];
+    lower[0]   =-Dw.mass_plus*lower[0];
     
     LatticeFermion r_eo(FGrid);
     LatticeFermion src_e (FrbGrid);

benchmarks/Benchmark_wilson_sweep.cc

@@ -44,6 +44,13 @@ void bench_wilson (
 		   double const     volume,
 		   int const           dag );
 
+void bench_wilson_eo (
+       LatticeFermion &    src,
+       LatticeFermion & result,
+       WilsonFermionR &     Dw,
+       double const     volume,
+       int const           dag );
+
 int main (int argc, char ** argv)
 {
   Grid_init(&argc,&argv);
@@ -110,8 +117,8 @@ int main (int argc, char ** argv)
 	  bench_wilson(src,result,Dw,volume,DaggerYes);
 	  std::cout << "\t";
     // EO
-	  bench_wilson(src,result,Dw,volume,DaggerNo);
-	  bench_wilson(src,result,Dw,volume,DaggerYes);
+    bench_wilson_eo(src_o,result_e,Dw,volume,DaggerNo);
+    bench_wilson_eo(src_o,result_e,Dw,volume,DaggerYes);
 	  std::cout << std::endl;
 	}
     }

configure.ac

@@ -159,7 +159,7 @@ case ${ac_ZMOBIUS} in
 esac
 ############### Nc
 AC_ARG_ENABLE([Nc],
-    [AC_HELP_STRING([--enable-Nc=2|3|4], [enable number of colours])],
+    [AC_HELP_STRING([--enable-Nc=2|3|4|5], [enable number of colours])],
     [ac_Nc=${enable_Nc}], [ac_Nc=3])
 
 case ${ac_Nc} in
@@ -394,11 +394,10 @@ case ${CXXTEST} in
     fi
     ;;
   hipcc)
-#    CXXFLAGS="$CXXFLAGS -Xcompiler -fno-strict-aliasing --expt-extended-lambda --expt-relaxed-constexpr"
     CXXFLAGS="$CXXFLAGS -fno-strict-aliasing"
     CXXLD=${CXX}
     if test $ac_openmp = yes; then
-       CXXFLAGS="$CXXFLAGS -Xcompiler -fopenmp"
+       CXXFLAGS="$CXXFLAGS -fopenmp"
     fi
     ;;
   dpcpp)
@@ -557,16 +556,19 @@ esac
 AC_ARG_ENABLE([setdevice],[AC_HELP_STRING([--enable-setdevice | --disable-setdevice],
               [Set GPU to rank in node with cudaSetDevice or similar])],[ac_SETDEVICE=${enable_SETDEVICE}],[ac_SETDEVICE=no])
 case ${ac_SETDEVICE} in
-    yes);;
-    no)
+    yes)
+	echo ENABLE SET DEVICE
+    ;;
+    *)
      AC_DEFINE([GRID_DEFAULT_GPU],[1],[GRID_DEFAULT_GPU] )
+   	echo DISABLE SET DEVICE
     ;;
 esac
 
 #########################################################
 ######################  Shared memory intranode #########
 #########################################################
-AC_ARG_ENABLE([shm],[AC_HELP_STRING([--enable-shm=shmopen|shmget|hugetlbfs|shmnone|nvlink|no],
+AC_ARG_ENABLE([shm],[AC_HELP_STRING([--enable-shm=shmopen|shmget|hugetlbfs|shmnone|nvlink|no|none],
               [Select SHM allocation technique])],[ac_SHM=${enable_shm}],[ac_SHM=no])
 
 case ${ac_SHM} in
@@ -586,7 +588,7 @@ case ${ac_SHM} in
      AC_DEFINE([GRID_MPI3_SHMGET],[1],[GRID_MPI3_SHMGET] )
      ;;
 
-     shmnone | no)
+     shmnone | no | none)
      AC_DEFINE([GRID_MPI3_SHM_NONE],[1],[GRID_MPI3_SHM_NONE] )
      ;;
 

examples/Example_Laplacian.cc

@@ -93,14 +93,14 @@ template<class Field> class FreeLaplacianStencil : public SparseMatrixBase<Field
 {
 public:
   typedef typename Field::vector_object siteObject;
-  typedef CartesianStencil<siteObject, siteObject, int> StencilImpl;
+  typedef CartesianStencil<siteObject, siteObject, SimpleStencilParams> StencilImpl;
 
   GridBase *grid;
   StencilImpl Stencil;
   SimpleCompressor<siteObject> Compressor;
   
   FreeLaplacianStencil(GridBase *_grid)
-    : Stencil    (_grid,6,Even,directions,displacements,0), grid(_grid)
+    : Stencil    (_grid,6,Even,directions,displacements,SimpleStencilParams()), grid(_grid)
   {  };
   
   virtual GridBase *Grid(void) { return grid; };
@@ -168,7 +168,8 @@ public:
   typedef iImplDoubledGaugeField<Simd> SiteDoubledGaugeField;
   typedef Lattice<SiteDoubledGaugeField> DoubledGaugeField;
 
-  typedef CartesianStencil<siteObject, siteObject, int> StencilImpl;
+  typedef CartesianStencil<siteObject, siteObject,SimpleStencilParams> StencilImpl;
+  SimpleStencilParams p;
 
   GridBase *grid;
   StencilImpl Stencil;
@@ -177,7 +178,7 @@ public:
   CovariantLaplacianStencil(GaugeField &Umu)
     :
       grid(Umu.Grid()),
-      Stencil    (grid,6,Even,directions,displacements,0),
+      Stencil    (grid,6,Even,directions,displacements,p),
       Uds(grid)
   {
     for (int mu = 0; mu < Nd; mu++) {

systems/Crusher/config-command

@@ -9,6 +9,7 @@
 --with-mpfr=/opt/cray/pe/gcc/mpfr/3.1.4/ \
 --disable-gparity \
 CXX=hipcc MPICXX=mpicxx \
-CXXFLAGS="-fPIC -I/opt/rocm-4.5.0/include/ -std=c++14 -I${MPICH_DIR}/include " \
- LDFLAGS=" -L${MPICH_DIR}/lib -lmpi -L${CRAY_MPICH_ROOTDIR}/gtl/lib -lmpi_gtl_hsa "
-HIPFLAGS = --amdgpu-target=gfx90a
+CXXFLAGS="-fPIC -I{$ROCM_PATH}/include/ -std=c++14 -I${MPICH_DIR}/include " \
+ LDFLAGS=" -L${MPICH_DIR}/lib -lmpi -L${CRAY_MPICH_ROOTDIR}/gtl/lib -lmpi_gtl_hsa -lamdhip64 "
+
+

systems/Crusher/dwf.slurm

@@ -12,19 +12,21 @@
 #SBATCH --gpu-bind=map_gpu:0,1,2,3,7,6,5,4
 
 DIR=.
-module list
+source sourceme.sh
 export MPIR_CVAR_GPU_EAGER_DEVICE_MEM=0
 export MPICH_GPU_SUPPORT_ENABLED=1
-#export MPICH_SMP_SINGLE_COPY_MODE=XPMEM
-export MPICH_SMP_SINGLE_COPY_MODE=NONE
-#export MPICH_SMP_SINGLE_COPY_MODE=CMA
+export MPICH_SMP_SINGLE_COPY_MODE=XPMEM
 export OMP_NUM_THREADS=1
 
 echo MPICH_SMP_SINGLE_COPY_MODE $MPICH_SMP_SINGLE_COPY_MODE
 
-PARAMS=" --accelerator-threads 16 --grid 32.32.32.256 --mpi 1.1.1.8 --comms-overlap --shm 2048 --shm-mpi 0"
-echo $PARAMS
+echo working directory
+pwd
+
+PARAMS=" --accelerator-threads 8 --grid 32.32.32.32 --mpi 1.1.1.1 --comms-sequential --shm 2048 --shm-mpi 0"
+srun --gpus-per-task 1 -n1  ./benchmarks/Benchmark_dwf_fp32 $PARAMS
+
+PARAMS=" --accelerator-threads 8 --grid 64.64.64.32 --mpi 2.2.2.1 --comms-sequential --shm 2048 --shm-mpi 0"
 srun --gpus-per-task 1 -n8  ./benchmarks/Benchmark_dwf_fp32 $PARAMS
 
 
-

systems/Crusher/dwf4.slurm

@@ -7,21 +7,19 @@
 #SBATCH -o DWF.%J
 #SBATCH -e DWF.%J
 #SBATCH -N 1
-#SBATCH -n 4
-#SBATCH --exclusive
+#SBATCH -n 2
+#SBATCH --gpu-bind=map_gpu:0,1
 
 DIR=.
-module list
+source setup.sh
+
+export MPICH_OFI_NIC_POLICY=GPU
 export MPIR_CVAR_GPU_EAGER_DEVICE_MEM=0
 export MPICH_GPU_SUPPORT_ENABLED=1
-#export MPICH_SMP_SINGLE_COPY_MODE=XPMEM
-export MPICH_SMP_SINGLE_COPY_MODE=NONE
-#export MPICH_SMP_SINGLE_COPY_MODE=CMA
-export OMP_NUM_THREADS=4
+export OMP_NUM_THREADS=16
 
 echo MPICH_SMP_SINGLE_COPY_MODE $MPICH_SMP_SINGLE_COPY_MODE
-PARAMS=" --accelerator-threads 8 --grid 32.32.64.64 --mpi 1.1.2.2 --comms-overlap --shm 2048 --shm-mpi 0"
 
-srun --gpus-per-task 1 -n4 ./mpiwrapper.sh ./benchmarks/Benchmark_dwf_fp32 $PARAMS
+srun --gpus-per-task 1 -N1 -n2  ./benchmarks/Benchmark_dwf_fp32 --mpi 1.1.1.2 --grid 16.16.32.64 --shm-mpi 1 --shm 2048 --comms-sequential --accelerator-threads 8
 
 

systems/Crusher/dwf8.slurm

@@ -6,43 +6,23 @@
 #SBATCH -J DWF
 #SBATCH -o DWF.%J
 #SBATCH -e DWF.%J
-#SBATCH -N 8
-#SBATCH -n 64
-#SBATCH --exclusive
-#SBATCH --gpu-bind=map_gpu:0,1,2,3,7,6,5,4
+#SBATCH -N 1
+#SBATCH -n 8
+##SBATCH --gpu-bind=map_gpu:0,1,2,3,7,6,5,4
+#SBATCH --gpu-bind=map_gpu:0,1,2,3,6,7,4,5
 
 DIR=.
-module list
+source setup.sh
+
 export MPICH_OFI_NIC_POLICY=GPU
 export MPIR_CVAR_GPU_EAGER_DEVICE_MEM=0
 export MPICH_GPU_SUPPORT_ENABLED=1
 #export MPICH_SMP_SINGLE_COPY_MODE=XPMEM
 #export MPICH_SMP_SINGLE_COPY_MODE=CMA
-export MPICH_SMP_SINGLE_COPY_MODE=NONE
-export OMP_NUM_THREADS=1
+#export MPICH_SMP_SINGLE_COPY_MODE=NONE
+export OMP_NUM_THREADS=16
 
 echo MPICH_SMP_SINGLE_COPY_MODE $MPICH_SMP_SINGLE_COPY_MODE
 
-PARAMS=" --accelerator-threads 16 --grid 64.64.64.256 --mpi 2.2.2.8 --comms-overlap --shm 2048 --shm-mpi 0"
-echo $PARAMS
-#srun --gpus-per-task 1 -N8 -n64  ./benchmarks/Benchmark_dwf_fp32 $PARAMS > dwf.64.64.64.256.8node
-
-
-PARAMS=" --accelerator-threads 16 --grid 64.64.64.32 --mpi 4.4.4.1 --comms-overlap --shm 2048 --shm-mpi 1"
-echo $PARAMS
-srun --gpus-per-task 1 -N8 -n64  ./benchmarks/Benchmark_dwf_fp32 $PARAMS > dwf.64.64.64.32.8node
-
-PARAMS=" --accelerator-threads 16 --grid 64.64.64.32 --mpi 4.4.4.1 --comms-overlap --shm 2048 --shm-mpi 0"
-echo $PARAMS
-#srun --gpus-per-task 1 -N8 -n64  ./benchmarks/Benchmark_dwf_fp32 $PARAMS > dwf.64.64.64.32.8node.shm0
-
-PARAMS=" --accelerator-threads 16 --grid 64.64.64.32 --mpi 2.2.2.8 --comms-overlap --shm 2048 --shm-mpi 1"
-echo $PARAMS
-#srun --gpus-per-task 1 -N8 -n64  ./benchmarks/Benchmark_ITT $PARAMS > itt.8node
-
-PARAMS=" --accelerator-threads 16 --grid 64.64.64.32 --mpi 2.2.2.8 --comms-overlap --shm 2048 --shm-mpi 0"
-echo $PARAMS
-#srun --gpus-per-task 1 -N8 -n64  ./benchmarks/Benchmark_ITT $PARAMS > itt.8node_shm0
-
-
+srun --gpus-per-task 1 -N1 -n8  ./benchmarks/Benchmark_comms_host_device --mpi 2.2.2.1 --shm-mpi 1 --shm 2048 --comms-sequential --accelerator-threads 8
 

systems/Crusher/sourceme.sh

@@ -1,6 +1,9 @@
 module load PrgEnv-gnu
-module load rocm/4.5.0
+module load rocm/5.1.0
+module load cray-mpich/8.1.16
 module load gmp
-module load cray-fftw
+#module load cray-fftw
 module load craype-accel-amd-gfx90a
 export LD_LIBRARY_PATH=/opt/gcc/mpfr/3.1.4/lib:$LD_LIBRARY_PATH
+#Hack for lib
+export LD_LIBRARY_PATH=`pwd`:$LD_LIBRARY_PATH

systems/Perlmutter/config-command

@@ -1,9 +1,14 @@
+DIR=`pwd`
+PREFIX=$DIR/../Prequisites/install/
 ../../configure \
     --enable-comms=mpi \
     --enable-simd=GPU \
     --enable-shm=nvlink \
     --enable-gen-simd-width=64 \
     --enable-accelerator=cuda \
+    --enable-setdevice \
+    --disable-accelerator-cshift \
+    --with-gmp=$PREFIX \
     --disable-fermion-reps \
     --disable-unified \
     --disable-gparity \

systems/Perlmutter/dwf4.slurm

@@ -1,24 +1,27 @@
 #!/bin/bash
-#SBATCH -A mp13
+#SBATCH -A m3886_g
 #SBATCH -C gpu
-#SBATCH -q regular
+#SBATCH -q debug
 #SBATCH -t 0:20:00
-#SBATCH -n 16
-#SBATCH --ntasks-per-node=4
 #SBATCH -c 32
-#SBATCH --exclusive
+#SBATCH -N 1
+#SBATCH -n 4
+#SBATCH --ntasks-per-node=4
 #SBATCH --gpus-per-task=1
-#SBATCH --gpu-bind=map_gpu:0,1,2,3
+#SBATCH --exclusive
+#SBATCH --gpu-bind=none
 
 export SLURM_CPU_BIND="cores"
-export MPICH_RDMA_ENABLED_CUDA=1
 export MPICH_GPU_SUPPORT_ENABLED=1
-srun ./benchmarks/Benchmark_comms_host_device --mpi 2.2.2.2  --accelerator-threads 8 > comms.4node
+export MPICH_RDMA_ENABLED_CUDA=1
+export MPICH_GPU_IPC_ENABLED=1
+export MPICH_GPU_EAGER_REGISTER_HOST_MEM=0
+export MPICH_GPU_NO_ASYNC_MEMCPY=0
+#export MPICH_SMP_SINGLE_COPY_MODE=CMA
 
-OPT="--comms-overlap --comms-concurrent --shm-mpi 0"
-srun ./benchmarks/Benchmark_dwf_fp32 --mpi 2.2.2.2 --grid 64.64.64.64 --accelerator-threads 8 --shm 2048 $OPT > dwf.64.64.64.64.4node.opt0
-srun ./benchmarks/Benchmark_dwf_fp32 --mpi 2.2.2.2 --grid 48.48.48.48 --accelerator-threads 8 --shm 2048 $OPT > dwf.48.48.48.48.4node.opt0
+OPT="--comms-sequential --shm-mpi 1"
+VOL=64.64.64.64
+srun ./benchmarks/Benchmark_dwf_fp32 --mpi 2.2.1.1 --grid $VOL --accelerator-threads 8 --shm 2048 $OPT
+#srun ./benchmarks/Benchmark_dwf_fp32 --mpi 2.1.1.4 --grid $VOL --accelerator-threads 8 --shm 2048 $OPT
+#srun ./benchmarks/Benchmark_dwf_fp32 --mpi 1.1.1.8 --grid $VOL --accelerator-threads 8 --shm 2048 $OPT
 
-OPT="--comms-overlap --comms-concurrent --shm-mpi 1"
-srun ./benchmarks/Benchmark_dwf_fp32 --mpi 2.2.2.2 --grid 64.64.64.64 --accelerator-threads 8 --shm 2048 $OPT > dwf.64.64.64.64.4node.opt1
-srun ./benchmarks/Benchmark_dwf_fp32 --mpi 2.2.2.2 --grid 48.48.48.48 --accelerator-threads 8 --shm 2048 $OPT > dwf.48.48.48.48.4node.opt1

systems/Perlmutter/sourceme.sh

@@ -1,4 +1,4 @@
 
 export CRAY_ACCEL_TARGET=nvidia80
 
-module load PrgEnv-gnu cpe-cuda cuda
+module load PrgEnv-gnu cpe-cuda cudatoolkit/11.4

systems/Summit/config-command

@@ -2,11 +2,12 @@
 	      --enable-simd=GPU \
 	      --enable-gen-simd-width=32 \
 	      --enable-unified=no \
-	       --enable-shm=nvlink \
+	       --enable-shm=no \
 	       --disable-gparity \
-	       --enable-setdevice \
+	       --disable-setdevice \
 	       --disable-fermion-reps \
 	       --enable-accelerator=cuda \
+	       --enable-accelerator-cshift \
 	       --prefix /ccs/home/paboyle/prefix \
 	       CXX=nvcc \
 	       LDFLAGS=-L/ccs/home/paboyle/prefix/lib/ \

systems/Summit/dwf16.lsf

@@ -1,25 +1,39 @@
 #!/bin/bash
 #BSUB -P LGT104
-#BSUB -W 2:00
+#BSUB -W 0:20
 #BSUB -nnodes 16
 #BSUB -J DWF
 
+
 export OMP_NUM_THREADS=6
 export PAMI_IBV_ADAPTER_AFFINITY=1
 export PAMI_ENABLE_STRIPING=1
-export OPT="--comms-concurrent --comms-overlap "
 
-APP="./benchmarks/Benchmark_comms_host_device  --mpi 4.4.4.3 "
-jsrun --nrs 16 -a6 -g6 -c42 -dpacked -b packed:7 --latency_priority gpu-cpu --smpiargs=-gpu $APP > comms.16node.log
+DIR=.
+source sourceme.sh
 
-APP="./benchmarks/Benchmark_dwf_fp32 --grid 96.96.96.72 --mpi 4.4.4.3 --shm 2048 --shm-force-mpi 1 --device-mem 8000 --shm-force-mpi 1 $OPT "
-jsrun --nrs 16 -a6 -g6 -c42 -dpacked -b packed:7 --latency_priority gpu-cpu --smpiargs=-gpu $APP > dwf.16node.24.log
+echo MPICH_SMP_SINGLE_COPY_MODE $MPICH_SMP_SINGLE_COPY_MODE
 
-APP="./benchmarks/Benchmark_dwf_fp32 --grid 128.128.128.96 --mpi 4.4.4.3 --shm 2048 --shm-force-mpi 1 --device-mem 8000 --shm-force-mpi 1 $OPT "
-jsrun --nrs 16 -a6 -g6 -c42 -dpacked -b packed:7 --latency_priority gpu-cpu --smpiargs=-gpu $APP > dwf.16node.32.log
+VOLS=( 32.32.32.16 32.32.32.64 64.32.32.64 64.32.64.64 64.64.64.64 64.64.64.128  64.64.64.256  64.64.64.512 128.64.64.64.512)
+MPI=( 1.1.1.1      1.1.1.4     2.1.1.4         2.1.2.4     2.2.2.4      2.2.2.8      2.2.2.16      2.2.2.32 4.4.2.32 )
+RANKS=(     1            4           8              16          32          64            128           256 1024)
+NODES=(     1            1           2               4           8           16            32            64  128)
+INTS=(      0            1           2               3           4            5             6             7    8)
 
+for i in 5
+do
+    vol=${VOLS[$i]} 
+    nodes=${NODES[$i]} 
+    mpi=${MPI[$i]} 
+    ranks=${RANKS[$i]} 
 
+    JSRUN="jsrun --nrs $nodes -a4 -g4 -c42 -dpacked -b packed:10 --latency_priority gpu-cpu --smpiargs=-gpu"
 
+    PARAMS=" --accelerator-threads 8 --grid $vol --mpi $mpi --comms-sequential --shm 2048 --shm-mpi 0"
+    $JSRUN ./benchmarks/Benchmark_dwf_fp32 $PARAMS > run.v${vol}.n${nodes}.m${mpi}.seq.ker
 
+    PARAMS=" --accelerator-threads 8 --grid $vol --mpi $mpi --comms-overlap --shm 2048 --shm-mpi 0"
+    $JSRUN ./benchmarks/Benchmark_dwf_fp32 $PARAMS > run.v${vol}.n${nodes}.m${mpi}.over.ker
 
+done
 

tests/IO/Test_field_array_io.cc

@@ -0,0 +1,184 @@
+    /*************************************************************************************
+
+    Grid physics library, www.github.com/paboyle/Grid 
+
+    Source file: ./tests/IO/Test_field_array_io.cc
+
+    Copyright (C) 2015
+
+Author: Christopher Kelly <ckelly@bnl.gov>
+Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License along
+    with this program; if not, write to the Free Software Foundation, Inc.,
+    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+    See the full license in the file "LICENSE" in the top level distribution directory
+    *************************************************************************************/
+    /*  END LEGAL */
+#include <Grid/Grid.h>
+
+using namespace std;
+using namespace Grid;
+
+//This test demonstrates and checks a single-file write of an arbitrary array of fields
+
+uint64_t writeHeader(const uint32_t size, const uint32_t checksum, const std::string &format, const std::string &file){
+  std::ofstream fout(file,std::ios::out|std::ios::in);
+  fout.seekp(0,std::ios::beg);
+  fout << std::setw(10) << size << std::endl;
+  fout << std::hex << std::setw(10) << checksum << std::endl;
+  fout << format << std::endl;
+  return fout.tellp();
+}
+ 
+uint64_t readHeader(uint32_t &size, uint32_t &checksum, std::string &format, const std::string &file){
+  std::ifstream fin(file);
+  std::string line;
+  getline(fin,line);
+  {
+    std::stringstream ss; ss <<line ; ss >> size;
+  }
+  getline(fin,line);
+  {
+    std::stringstream ss; ss <<line ; ss >> std::hex >> checksum;
+  }
+  getline(fin,format);
+  removeWhitespace(format);
+      
+  return fin.tellg();
+}
+ 
+template<typename FieldType>
+void writeFieldArray(const std::string &file, const std::vector<FieldType> &data){
+  typedef typename FieldType::vector_object vobj;
+  typedef typename FieldType::scalar_object sobj;
+  GridBase* grid = data[0].Grid(); //assume all fields have the same Grid
+  BinarySimpleMunger<sobj, sobj> munge; //straight copy
+
+  //We need a 2-pass header write, first to establish the size, the second pass writes the checksum
+  std::string format = getFormatString<typename FieldType::vector_object>();
+
+  uint64_t offset; //leave 64 bits for header
+  if ( grid->IsBoss() ) { 
+    NerscIO::truncate(file);
+    offset = writeHeader(data.size(), 0, format, file);
+  }
+  grid->Broadcast(0,(void *)&offset,sizeof(offset)); //use as a barrier
+
+  std::cout << "Data offset write " << offset << std::endl;
+  std::cout << "Data size write " << data.size() << std::endl;
+  uint64_t field_size = uint64_t(grid->gSites()) * sizeof(sobj);
+  std::cout << "Field size = " << field_size << " B" << std::endl;
+
+  uint32_t checksum = 0;
+  for(int i=0;i<data.size();i++){
+    std::cout << "Data field write " << i << " offset " << offset << std::endl;
+    uint32_t nersc_csum,scidac_csuma,scidac_csumb;
+    BinaryIO::writeLatticeObject<vobj,sobj>(const_cast<FieldType &>(data[i]),file,munge,offset,format,
+					    nersc_csum,scidac_csuma,scidac_csumb);
+    offset += field_size;
+    checksum ^= nersc_csum + 0x9e3779b9 + (checksum<<6) + (checksum>>2);
+  }
+  std::cout << "Write checksum " << checksum << std::endl;
+
+  if ( grid->IsBoss() ) { 
+    writeHeader(data.size(), checksum, format, file);
+  }
+}
+
+
+template<typename FieldType>
+void readFieldArray(std::vector<FieldType> &data, const std::string &file){
+  typedef typename FieldType::vector_object vobj;
+  typedef typename FieldType::scalar_object sobj;
+  assert(data.size() > 0);
+  GridBase* grid = data[0].Grid(); //assume all fields have the same Grid
+  BinarySimpleUnmunger<sobj, sobj> munge; //straight copy
+  
+  uint32_t hdr_checksum, hdr_size;
+  std::string format;
+  uint64_t offset = readHeader(hdr_size, hdr_checksum, format, file);
+  
+  std::cout << "Data offset read " << offset << std::endl;  
+  std::cout << "Data size read " << hdr_size << std::endl;
+  assert(data.size() == hdr_size);
+
+  uint64_t field_size = uint64_t(grid->gSites()) * sizeof(sobj);
+
+  uint32_t checksum = 0;
+
+  for(int i=0;i<data.size();i++){
+    std::cout << "Data field read " << i << " offset " << offset << std::endl;
+    uint32_t nersc_csum,scidac_csuma,scidac_csumb;
+    BinaryIO::readLatticeObject<vobj,sobj>(data[i],file,munge,offset,format,
+					   nersc_csum,scidac_csuma,scidac_csumb);
+    offset += field_size;
+    checksum ^= nersc_csum + 0x9e3779b9 + (checksum<<6) + (checksum>>2);
+  }
+
+  std::cout << "Header checksum " << hdr_checksum << std::endl;    
+  std::cout << "Read checksum " << checksum << std::endl;
+    
+
+  assert( hdr_checksum == checksum );
+}
+
+
+
+
+int main (int argc, char ** argv)
+{
+  Grid_init(&argc,&argv);
+
+  Coordinate latt   = GridDefaultLatt();
+  Coordinate simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd());
+  Coordinate mpi_layout  = GridDefaultMpi();
+
+  const int Ls=8;
+
+  GridCartesian         * UGrid   = SpaceTimeGrid::makeFourDimGrid(latt, simd_layout, mpi_layout);
+  GridRedBlackCartesian * UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
+  GridCartesian         * FGrid   = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
+  GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid);
+
+  std::vector<int> seeds4({1,2,3,4});
+  std::vector<int> seeds5({5,6,7,8});
+  GridParallelRNG RNG5(FGrid);  RNG5.SeedFixedIntegers(seeds5);
+  GridParallelRNG RNG4(UGrid);  RNG4.SeedFixedIntegers(seeds4);
+
+  typedef DomainWallFermionD::FermionField FermionField;
+
+  int nfield = 20;
+  std::vector<FermionField> data(nfield, FGrid);
+
+  for(int i=0;i<data.size();i++)
+    gaussian(RNG5, data[i]);
+  
+  std::string file = "test_field_array_io.0";
+  writeFieldArray(file, data);
+
+  std::vector<FermionField> data_r(nfield, FGrid);
+  readFieldArray(data_r, file);
+  
+  for(int i=0;i<nfield;i++){
+    FermionField diff = data_r[i] - data[i];
+    RealD norm_diff = norm2(diff);
+    std::cout << "Norm2 of difference between stored and loaded data index " << i << " : " << norm_diff << std::endl;
+  }
+  
+  std::cout << "Done" << std::endl;
+
+  Grid_finalize();
+}

tests/IO/Test_nersc_io.cc

@@ -147,7 +147,7 @@ int main (int argc, char ** argv)
   Complex p  = TensorRemove(Tp);
   std::cout<<GridLogMessage << "calculated plaquettes " <<p*PlaqScale<<std::endl;
 
-  Complex LinkTraceScale(1.0/vol/4.0/3.0);
+  Complex LinkTraceScale(1.0/vol/4.0/(Real)Nc);
   TComplex Tl = sum(LinkTrace);
   Complex l  = TensorRemove(Tl);
   std::cout<<GridLogMessage << "calculated link trace " <<l*LinkTraceScale<<std::endl;
@@ -157,8 +157,10 @@ int main (int argc, char ** argv)
   Complex ll= TensorRemove(TcP);
   std::cout<<GridLogMessage << "coarsened plaquettes sum to " <<ll*PlaqScale<<std::endl;
 
-  std::string clone2x3("./ckpoint_clone2x3.4000");
-  std::string clone3x3("./ckpoint_clone3x3.4000");
+  const string stNc   = to_string( Nc   ) ;
+  const string stNcM1 = to_string( Nc-1 ) ;
+  std::string clone2x3("./ckpoint_clone"+stNcM1+"x"+stNc+".4000");
+  std::string clone3x3("./ckpoint_clone"+stNc+"x"+stNc+".4000");
 
   NerscIO::writeConfiguration(Umu,clone3x3,0,precision32);
   NerscIO::writeConfiguration(Umu,clone2x3,1,precision32);

tests/Test_dwf_mixedcg_prec.cc

@@ -46,7 +46,7 @@ int main (int argc, char ** argv)
 {
   Grid_init(&argc,&argv);
 
-  const int Ls=8;
+  const int Ls=12;
 
   std::cout << GridLogMessage << "::::: NB: to enable a quick bit reproducibility check use the --checksums flag. " << std::endl;
 
@@ -94,13 +94,32 @@ int main (int argc, char ** argv)
 
   std::cout << GridLogMessage << "::::::::::::: Starting mixed CG" << std::endl;
   MixedPrecisionConjugateGradient<LatticeFermionD,LatticeFermionF> mCG(1.0e-8, 10000, 50, FrbGrid_f, HermOpEO_f, HermOpEO);
-  mCG(src_o,result_o);
-
+  double t1,t2,flops;
+  int iters;
+  for(int i=0;i<100;i++){
+    result_o = Zero();
+    t1=usecond();
+    mCG(src_o,result_o);
+    t2=usecond();
+    iters = mCG.TotalInnerIterations; //Number of inner CG iterations
+    flops = 1320.0*2*FGrid->gSites()*iters;
+    std::cout << " SinglePrecision iterations/sec "<< iters/(t2-t1)*1000.*1000.<<std::endl;
+    std::cout << " SinglePrecision GF/s "<< flops/(t2-t1)/1000.<<std::endl;
+  }
   std::cout << GridLogMessage << "::::::::::::: Starting regular CG" << std::endl;
   ConjugateGradient<LatticeFermionD> CG(1.0e-8,10000);
-  CG(HermOpEO,src_o,result_o_2);
-
-  MemoryManager::Print();
+  for(int i=0;i<100;i++){
+    result_o_2 = Zero();
+    t1=usecond();
+    CG(HermOpEO,src_o,result_o_2);
+    t2=usecond();
+    iters = CG.IterationsToComplete;
+    flops = 1320.0*2*FGrid->gSites()*iters;
+    std::cout << " DoublePrecision iterations/sec "<< iters/(t2-t1)*1000.*1000.<<std::endl;
+    std::cout << " DoublePrecision GF/s "<< flops/(t2-t1)/1000.<<std::endl;
+  }
+  
+  //  MemoryManager::Print();
 
   LatticeFermionD diff_o(FrbGrid);
   RealD diff = axpy_norm(diff_o, -1.0, result_o, result_o_2);

tests/Test_stencil.cc

@@ -31,7 +31,6 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 
 using namespace std;
 using namespace Grid;
- ;
 
 int main(int argc, char ** argv) {
   Grid_init(&argc, &argv);
@@ -80,7 +79,8 @@ int main(int argc, char ** argv) {
     Foo=lex;
   }
 
-  typedef CartesianStencil<vobj,vobj,int> Stencil;
+  typedef CartesianStencil<vobj,vobj,SimpleStencilParams> Stencil;
+  SimpleStencilParams p;
     for(int dir=0;dir<4;dir++){
       for(int disp=0;disp<Fine._fdimensions[dir];disp++){
 
@@ -90,7 +90,7 @@ int main(int argc, char ** argv) {
 	std::vector<int> directions(npoint,dir);
 	std::vector<int> displacements(npoint,disp);
 
-	Stencil myStencil(&Fine,npoint,0,directions,displacements,0);
+	Stencil myStencil(&Fine,npoint,0,directions,displacements,p);
 	Coordinate ocoor(4);
 	for(int o=0;o<Fine.oSites();o++){
 	  Fine.oCoorFromOindex(ocoor,o);
@@ -183,8 +183,8 @@ int main(int argc, char ** argv) {
 	std::vector<int> directions(npoint,dir);
 	std::vector<int> displacements(npoint,disp);
 
-	Stencil EStencil(&rbFine,npoint,Even,directions,displacements,0);
-	Stencil OStencil(&rbFine,npoint,Odd,directions,displacements,0);
+	Stencil EStencil(&rbFine,npoint,Even,directions,displacements,p);
+	Stencil OStencil(&rbFine,npoint,Odd,directions,displacements,p);
 
 	Coordinate ocoor(4);
 	for(int o=0;o<Fine.oSites();o++){

tests/core/Test_compact_wilson_clover_speedup.cc

@@ -117,8 +117,8 @@ void runBenchmark(int* argc, char*** argv) {
 
   // type definitions
   typedef WilsonImpl<vCoeff_t, FundamentalRepresentation, CoeffReal> WImpl;
-  typedef WilsonCloverFermion<WImpl> WilsonCloverOperator;
-  typedef CompactWilsonCloverFermion<WImpl> CompactWilsonCloverOperator;
+  typedef WilsonCloverFermion<WImpl, CloverHelpers<WImpl>> WilsonCloverOperator;
+  typedef CompactWilsonCloverFermion<WImpl, CompactCloverHelpers<WImpl>> CompactWilsonCloverOperator;
   typedef typename WilsonCloverOperator::FermionField Fermion;
   typedef typename WilsonCloverOperator::GaugeField Gauge;
 

tests/core/Test_lie_generators.cc

@@ -9,6 +9,7 @@ Copyright (C) 2015
 Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
 Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 Author: Guido Cossu <guido.cossu@ed.ac.uk>
+Author: Jamie Hudspith <renwick.james.hudspth@gmail.com>
 
 This program is free software; you can redistribute it and/or modify
 it under the terms of the GNU General Public License as published by
@@ -42,14 +43,14 @@ directory
 
 using namespace std;
 using namespace Grid;
- ;
+;
 
 int main(int argc, char** argv) {
   Grid_init(&argc, &argv);
 
   std::vector<int> latt({4, 4, 4, 8});
   GridCartesian* grid = SpaceTimeGrid::makeFourDimGrid(
-      latt, GridDefaultSimd(Nd, vComplex::Nsimd()), GridDefaultMpi());
+						       latt, GridDefaultSimd(Nd, vComplex::Nsimd()), GridDefaultMpi());
 
   GridRedBlackCartesian* rbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(grid);
 
@@ -60,15 +61,19 @@ int main(int argc, char** argv) {
             << std::endl;
   SU2::printGenerators();
   std::cout << "Dimension of adjoint representation: "<< SU2Adjoint::Dimension << std::endl;
+
+  // guard as this code fails to compile for Nc != 3
+#if (Nc == 3)
+    
   SU2Adjoint::printGenerators();
   SU2::testGenerators();
   SU2Adjoint::testGenerators();
-
+    
   std::cout << GridLogMessage << "*********************************************"
-            << std::endl;
+	    << std::endl;
   std::cout << GridLogMessage << "* Generators for SU(Nc" << std::endl;
   std::cout << GridLogMessage << "*********************************************"
-            << std::endl;
+	    << std::endl;
   SU3::printGenerators();
   std::cout << "Dimension of adjoint representation: "<< SU3Adjoint::Dimension << std::endl;
   SU3Adjoint::printGenerators();
@@ -111,12 +116,10 @@ int main(int argc, char** argv) {
 
   // AdjointRepresentation has the predefined number of colours Nc
   //  Representations<FundamentalRepresentation, AdjointRepresentation, TwoIndexSymmetricRepresentation> RepresentationTypes(grid);  
-
-  
   LatticeGaugeField U(grid), V(grid);
   SU3::HotConfiguration<LatticeGaugeField>(gridRNG, U);
   SU3::HotConfiguration<LatticeGaugeField>(gridRNG, V);
-
+    
   // Adjoint representation
   // Test group structure
   // (U_f * V_f)_r = U_r * V_r
@@ -127,17 +130,17 @@ int main(int argc, char** argv) {
     SU3::LatticeMatrix Vmu = peekLorentz(V,mu);
     pokeLorentz(UV,Umu*Vmu, mu);
   }
-
+    
   AdjRep.update_representation(UV);
   typename AdjointRep<Nc>::LatticeField UVr = AdjRep.U;  // (U_f * V_f)_r 
-
-
+    
+    
   AdjRep.update_representation(U);
   typename AdjointRep<Nc>::LatticeField Ur = AdjRep.U;  // U_r
-
+    
   AdjRep.update_representation(V);
   typename AdjointRep<Nc>::LatticeField Vr = AdjRep.U;  // V_r
-
+    
   typename AdjointRep<Nc>::LatticeField UrVr(grid);
   UrVr = Zero();
   for (int mu = 0; mu < Nd; mu++) {
@@ -145,10 +148,10 @@ int main(int argc, char** argv) {
     typename AdjointRep<Nc>::LatticeMatrix Vrmu = peekLorentz(Vr,mu);
     pokeLorentz(UrVr,Urmu*Vrmu, mu);
   }
-
+    
   typename AdjointRep<Nc>::LatticeField Diff_check = UVr - UrVr;
   std::cout << GridLogMessage << "Group structure SU("<<Nc<<") check difference (Adjoint representation) : " << norm2(Diff_check) << std::endl;
-
+    
   // Check correspondence of algebra and group transformations
   // Create a random vector
   SU3::LatticeAlgebraVector h_adj(grid);
@@ -156,32 +159,31 @@ int main(int argc, char** argv) {
   random(gridRNG,h_adj);
   h_adj = real(h_adj);
   SU_Adjoint<Nc>::AdjointLieAlgebraMatrix(h_adj,Ar);
-
+    
   // Re-extract h_adj
   SU3::LatticeAlgebraVector h_adj2(grid);
   SU_Adjoint<Nc>::projectOnAlgebra(h_adj2, Ar);
   SU3::LatticeAlgebraVector h_diff = h_adj - h_adj2;
   std::cout << GridLogMessage << "Projections structure check vector difference (Adjoint representation) : " << norm2(h_diff) << std::endl;
-
+    
   // Exponentiate
   typename AdjointRep<Nc>::LatticeMatrix Uadj(grid);
   Uadj  = expMat(Ar, 1.0, 16);
-
+    
   typename AdjointRep<Nc>::LatticeMatrix uno(grid);
   uno = 1.0;
   // Check matrix Uadj, must be real orthogonal
   typename AdjointRep<Nc>::LatticeMatrix Ucheck = Uadj - conjugate(Uadj);
   std::cout << GridLogMessage << "Reality check: " << norm2(Ucheck)
-            << std::endl;
-
+	    << std::endl;
+    
   Ucheck = Uadj * adj(Uadj) - uno;
   std::cout << GridLogMessage << "orthogonality check 1: " << norm2(Ucheck)
-            << std::endl;
+	    << std::endl;
   Ucheck = adj(Uadj) * Uadj - uno;
   std::cout << GridLogMessage << "orthogonality check 2: " << norm2(Ucheck)
-            << std::endl;
-      
-
+	    << std::endl;
+    
   // Construct the fundamental matrix in the group
   SU3::LatticeMatrix Af(grid);
   SU3::FundamentalLieAlgebraMatrix(h_adj,Af);
@@ -193,72 +195,65 @@ int main(int argc, char** argv) {
   SU3::LatticeMatrix UnitCheck(grid);
   UnitCheck = Ufund * adj(Ufund) - uno_f;
   std::cout << GridLogMessage << "unitarity check 1: " << norm2(UnitCheck)
-            << std::endl;
+	    << std::endl;
   UnitCheck = adj(Ufund) * Ufund - uno_f;
   std::cout << GridLogMessage << "unitarity check 2: " << norm2(UnitCheck)
-            << std::endl;
-
+	    << std::endl;
 
   // Tranform to the adjoint representation
   U = Zero(); // fill this with only one direction
   pokeLorentz(U,Ufund,0); // the representation transf acts on full gauge fields
-
+    
   AdjRep.update_representation(U);
   Ur = AdjRep.U;  // U_r  
   typename AdjointRep<Nc>::LatticeMatrix Ur0 = peekLorentz(Ur,0); // this should be the same as Uadj
-
+    
   typename AdjointRep<Nc>::LatticeMatrix Diff_check_mat = Ur0 - Uadj;
   std::cout << GridLogMessage << "Projections structure check group difference : " << norm2(Diff_check_mat) << std::endl;
 
-
-
-
   // TwoIndexRep tests
-
   std::cout << GridLogMessage << "*********************************************"
-            << std::endl;
+	    << std::endl;
   std::cout << GridLogMessage << "*********************************************"
-            << std::endl;
+	    << std::endl;
 
 
   
   std::cout << GridLogMessage << "* eS^{ij} base for SU(2)" << std::endl;
   std::cout << GridLogMessage << "*********************************************"
-            << std::endl;
+	    << std::endl;
   std::cout << GridLogMessage << "Dimension of Two Index Symmetric representation: "<< SU2TwoIndexSymm::Dimension << std::endl;
   SU2TwoIndexSymm::printBase();
-      std::cout << GridLogMessage << "*********************************************"
-            << std::endl;
-        std::cout << GridLogMessage << "Generators of Two Index Symmetric representation: "<< SU2TwoIndexSymm::Dimension << std::endl;
+  std::cout << GridLogMessage << "*********************************************"
+	    << std::endl;
+  std::cout << GridLogMessage << "Generators of Two Index Symmetric representation: "<< SU2TwoIndexSymm::Dimension << std::endl;
   SU2TwoIndexSymm::printGenerators();
-        std::cout << GridLogMessage << "Test of Two Index Symmetric Generators: "<< SU2TwoIndexSymm::Dimension << std::endl;
+  std::cout << GridLogMessage << "Test of Two Index Symmetric Generators: "<< SU2TwoIndexSymm::Dimension << std::endl;
   SU2TwoIndexSymm::testGenerators();
   std::cout << GridLogMessage << "*********************************************"
-            << std::endl;
+	    << std::endl;
 
 
   
   std::cout << GridLogMessage << "*********************************************"
-            << std::endl;
+	    << std::endl;
   std::cout << GridLogMessage << "* eAS^{ij} base for SU(2)" << std::endl;
   
   std::cout << GridLogMessage << "*********************************************"
-            << std::endl;
+	    << std::endl;
   std::cout << GridLogMessage << "Dimension of Two Index anti-Symmetric representation: "<< SU2TwoIndexAntiSymm::Dimension << std::endl;
   SU2TwoIndexAntiSymm::printBase();
-      std::cout << GridLogMessage << "*********************************************"
-            << std::endl;
-        std::cout << GridLogMessage << "Dimension of Two Index anti-Symmetric representation: "<< SU2TwoIndexAntiSymm::Dimension << std::endl;
+  std::cout << GridLogMessage << "*********************************************"
+	    << std::endl;
+  std::cout << GridLogMessage << "Dimension of Two Index anti-Symmetric representation: "<< SU2TwoIndexAntiSymm::Dimension << std::endl;
   SU2TwoIndexAntiSymm::printGenerators();
   std::cout << GridLogMessage << "Test of Two Index anti-Symmetric Generators: "<< SU2TwoIndexAntiSymm::Dimension << std::endl;
   SU2TwoIndexAntiSymm::testGenerators();
   
-  
-  
   std::cout << GridLogMessage << "*********************************************"
-      << std::endl;
+	    << std::endl;
   std::cout << GridLogMessage << "Test for the Two Index Symmetric projectors"
-      << std::endl;
+	    << std::endl;
   // Projectors 
   SU3TwoIndexSymm::LatticeTwoIndexMatrix Gauss2(grid);
   random(gridRNG,Gauss2);
@@ -276,13 +271,13 @@ int main(int argc, char** argv) {
   SU3::LatticeAlgebraVector diff2 = ha - hb;
   std::cout << GridLogMessage << "Difference: " << norm2(diff) << std::endl;
   std::cout << GridLogMessage << "*********************************************"
-      << std::endl;
+	    << std::endl;
 
   
-    std::cout << GridLogMessage << "*********************************************"
-      << std::endl;
+  std::cout << GridLogMessage << "*********************************************"
+	    << std::endl;
   std::cout << GridLogMessage << "Test for the Two index anti-Symmetric projectors"
-      << std::endl;
+	    << std::endl;
   // Projectors
   SU3TwoIndexAntiSymm::LatticeTwoIndexMatrix Gauss2a(grid);
   random(gridRNG,Gauss2a);
@@ -300,11 +295,11 @@ int main(int argc, char** argv) {
   SU3::LatticeAlgebraVector diff2a = ha - hb;
   std::cout << GridLogMessage << "Difference: " << norm2(diff2a) << std::endl;
   std::cout << GridLogMessage << "*********************************************"
-      << std::endl;
+	    << std::endl;
     
   
   std::cout << GridLogMessage << "Two index Symmetric: Checking Group Structure"
-      << std::endl;
+	    << std::endl;
   // Testing HMC representation classes
   TwoIndexRep< Nc, Symmetric > TIndexRep(grid);
 
@@ -313,7 +308,7 @@ int main(int argc, char** argv) {
   LatticeGaugeField U2(grid), V2(grid);
   SU3::HotConfiguration<LatticeGaugeField>(gridRNG, U2);
   SU3::HotConfiguration<LatticeGaugeField>(gridRNG, V2);
-  
+    
   LatticeGaugeField UV2(grid);
   UV2 = Zero();
   for (int mu = 0; mu < Nd; mu++) {
@@ -321,16 +316,16 @@ int main(int argc, char** argv) {
     SU3::LatticeMatrix Vmu2 = peekLorentz(V2,mu);
     pokeLorentz(UV2,Umu2*Vmu2, mu);
   }
-  
+    
   TIndexRep.update_representation(UV2);
   typename TwoIndexRep< Nc, Symmetric >::LatticeField UVr2 = TIndexRep.U;  // (U_f * V_f)_r
-  
+    
   TIndexRep.update_representation(U2);
   typename TwoIndexRep< Nc, Symmetric >::LatticeField Ur2 = TIndexRep.U;  // U_r
-  
+    
   TIndexRep.update_representation(V2);
   typename TwoIndexRep< Nc, Symmetric >::LatticeField Vr2 = TIndexRep.U;  // V_r
-  
+    
   typename TwoIndexRep< Nc, Symmetric >::LatticeField Ur2Vr2(grid);
   Ur2Vr2 = Zero();
   for (int mu = 0; mu < Nd; mu++) {
@@ -338,11 +333,11 @@ int main(int argc, char** argv) {
     typename TwoIndexRep< Nc, Symmetric >::LatticeMatrix Vrmu2 = peekLorentz(Vr2,mu);
     pokeLorentz(Ur2Vr2,Urmu2*Vrmu2, mu);
   }
-  
+    
   typename TwoIndexRep< Nc, Symmetric >::LatticeField Diff_check2 = UVr2 - Ur2Vr2;
   std::cout << GridLogMessage << "Group structure SU("<<Nc<<") check difference (Two Index Symmetric): " << norm2(Diff_check2) << std::endl;
-
-  
+    
+    
   // Check correspondence of algebra and group transformations
   // Create a random vector
   SU3::LatticeAlgebraVector h_sym(grid);
@@ -350,34 +345,31 @@ int main(int argc, char** argv) {
   random(gridRNG,h_sym);
   h_sym = real(h_sym);
   SU_TwoIndex<Nc,Symmetric>::TwoIndexLieAlgebraMatrix(h_sym,Ar_sym);
-  
+    
   // Re-extract h_sym
   SU3::LatticeAlgebraVector h_sym2(grid);
   SU_TwoIndex< Nc, Symmetric>::projectOnAlgebra(h_sym2, Ar_sym);
   SU3::LatticeAlgebraVector h_diff_sym = h_sym - h_sym2;
   std::cout << GridLogMessage << "Projections structure check vector difference (Two Index Symmetric): " << norm2(h_diff_sym) << std::endl;
-
-  
+    
   // Exponentiate
   typename TwoIndexRep< Nc, Symmetric>::LatticeMatrix U2iS(grid);
   U2iS  = expMat(Ar_sym, 1.0, 16);
-  
+    
   typename TwoIndexRep< Nc, Symmetric>::LatticeMatrix uno2iS(grid);
   uno2iS = 1.0;
   // Check matrix U2iS, must be real orthogonal
   typename TwoIndexRep< Nc, Symmetric>::LatticeMatrix Ucheck2iS = U2iS - conjugate(U2iS);
   std::cout << GridLogMessage << "Reality check: " << norm2(Ucheck2iS)
-      << std::endl;
-  
+	    << std::endl;
+    
   Ucheck2iS = U2iS * adj(U2iS) - uno2iS;
   std::cout << GridLogMessage << "orthogonality check 1: " << norm2(Ucheck2iS)
-      << std::endl;
+	    << std::endl;
   Ucheck2iS = adj(U2iS) * U2iS - uno2iS;
   std::cout << GridLogMessage << "orthogonality check 2: " << norm2(Ucheck2iS)
-      << std::endl;
-  
-  
-  
+	    << std::endl;
+    
   // Construct the fundamental matrix in the group
   SU3::LatticeMatrix Af_sym(grid);
   SU3::FundamentalLieAlgebraMatrix(h_sym,Af_sym);
@@ -386,147 +378,137 @@ int main(int argc, char** argv) {
   SU3::LatticeMatrix UnitCheck2(grid);
   UnitCheck2 = Ufund2 * adj(Ufund2) - uno_f;
   std::cout << GridLogMessage << "unitarity check 1: " << norm2(UnitCheck2)
-      << std::endl;
+	    << std::endl;
   UnitCheck2 = adj(Ufund2) * Ufund2 - uno_f;
   std::cout << GridLogMessage << "unitarity check 2: " << norm2(UnitCheck2)
-      << std::endl;
-  
-
+	    << std::endl;
+    
+    
   // Tranform to the 2Index Sym representation
   U = Zero(); // fill this with only one direction
   pokeLorentz(U,Ufund2,0); // the representation transf acts on full gauge fields
-  
+    
   TIndexRep.update_representation(U);
   Ur2 = TIndexRep.U;  // U_r  
   typename TwoIndexRep< Nc, Symmetric>::LatticeMatrix Ur02 = peekLorentz(Ur2,0); // this should be the same as U2iS
-  
+    
   typename TwoIndexRep< Nc, Symmetric>::LatticeMatrix Diff_check_mat2 = Ur02 - U2iS;
   std::cout << GridLogMessage << "Projections structure check group difference (Two Index Symmetric): " << norm2(Diff_check_mat2) << std::endl;
-  
-
-
-
+    
   if (TwoIndexRep<Nc, AntiSymmetric >::Dimension != 1){
 
-  std::cout << GridLogMessage << "*********************************************"
-      << std::endl;
+    std::cout << GridLogMessage << "*********************************************"
+	      << std::endl;
     
   
-  std::cout << GridLogMessage << "Two Index anti-Symmetric: Check Group Structure"
-      << std::endl;
-  // Testing HMC representation classes
-  TwoIndexRep< Nc, AntiSymmetric > TIndexRepA(grid);
+    std::cout << GridLogMessage << "Two Index anti-Symmetric: Check Group Structure"
+	      << std::endl;
+    // Testing HMC representation classes
+    TwoIndexRep< Nc, AntiSymmetric > TIndexRepA(grid);
 
 
-  // Test group structure
-  // (U_f * V_f)_r = U_r * V_r
-  LatticeGaugeField U2A(grid), V2A(grid);
-  SU3::HotConfiguration<LatticeGaugeField>(gridRNG, U2A);
-  SU3::HotConfiguration<LatticeGaugeField>(gridRNG, V2A);
+    // Test group structure
+    // (U_f * V_f)_r = U_r * V_r
+    LatticeGaugeField U2A(grid), V2A(grid);
+    SU3::HotConfiguration<LatticeGaugeField>(gridRNG, U2A);
+    SU3::HotConfiguration<LatticeGaugeField>(gridRNG, V2A);
   
-  LatticeGaugeField UV2A(grid);
-  UV2A = Zero();
-  for (int mu = 0; mu < Nd; mu++) {
-    SU3::LatticeMatrix Umu2A = peekLorentz(U2,mu);
-    SU3::LatticeMatrix Vmu2A = peekLorentz(V2,mu);
-    pokeLorentz(UV2A,Umu2A*Vmu2A, mu);
+    LatticeGaugeField UV2A(grid);
+    UV2A = Zero();
+    for (int mu = 0; mu < Nd; mu++) {
+      SU3::LatticeMatrix Umu2A = peekLorentz(U2,mu);
+      SU3::LatticeMatrix Vmu2A = peekLorentz(V2,mu);
+      pokeLorentz(UV2A,Umu2A*Vmu2A, mu);
+    }
+  
+    TIndexRep.update_representation(UV2A);
+    typename TwoIndexRep< Nc, AntiSymmetric >::LatticeField UVr2A = TIndexRepA.U;  // (U_f * V_f)_r
+  
+    TIndexRep.update_representation(U2A);
+    typename TwoIndexRep< Nc, AntiSymmetric >::LatticeField Ur2A = TIndexRepA.U;  // U_r
+  
+    TIndexRep.update_representation(V2A);
+    typename TwoIndexRep< Nc, AntiSymmetric >::LatticeField Vr2A = TIndexRepA.U;  // V_r
+  
+    typename TwoIndexRep< Nc, AntiSymmetric >::LatticeField Ur2Vr2A(grid);
+    Ur2Vr2A = Zero();
+    for (int mu = 0; mu < Nd; mu++) {
+      typename TwoIndexRep< Nc, AntiSymmetric >::LatticeMatrix Urmu2A = peekLorentz(Ur2A,mu);
+      typename TwoIndexRep< Nc, AntiSymmetric >::LatticeMatrix Vrmu2A = peekLorentz(Vr2A,mu);
+      pokeLorentz(Ur2Vr2A,Urmu2A*Vrmu2A, mu);
+    }
+  
+    typename TwoIndexRep< Nc, AntiSymmetric >::LatticeField Diff_check2A = UVr2A - Ur2Vr2A;
+    std::cout << GridLogMessage << "Group structure SU("<<Nc<<") check difference (Two Index anti-Symmetric): " << norm2(Diff_check2A) << std::endl;
+
+  
+    // Check correspondence of algebra and group transformations
+    // Create a random vector
+    SU3::LatticeAlgebraVector h_Asym(grid);
+    typename TwoIndexRep< Nc, AntiSymmetric>::LatticeMatrix Ar_Asym(grid);
+    random(gridRNG,h_Asym);
+    h_Asym = real(h_Asym);
+    SU_TwoIndex< Nc, AntiSymmetric>::TwoIndexLieAlgebraMatrix(h_Asym,Ar_Asym);
+  
+    // Re-extract h_sym
+    SU3::LatticeAlgebraVector h_Asym2(grid);
+    SU_TwoIndex< Nc, AntiSymmetric>::projectOnAlgebra(h_Asym2, Ar_Asym);
+    SU3::LatticeAlgebraVector h_diff_Asym = h_Asym - h_Asym2;
+    std::cout << GridLogMessage << "Projections structure check vector difference (Two Index anti-Symmetric): " << norm2(h_diff_Asym) << std::endl;
+
+  
+    // Exponentiate
+    typename TwoIndexRep< Nc, AntiSymmetric>::LatticeMatrix U2iAS(grid);
+    U2iAS  = expMat(Ar_Asym, 1.0, 16);
+  
+    typename TwoIndexRep< Nc, AntiSymmetric>::LatticeMatrix uno2iAS(grid);
+    uno2iAS = 1.0;
+    // Check matrix U2iS, must be real orthogonal
+    typename TwoIndexRep< Nc, AntiSymmetric>::LatticeMatrix Ucheck2iAS = U2iAS - conjugate(U2iAS);
+    std::cout << GridLogMessage << "Reality check: " << norm2(Ucheck2iAS)
+	      << std::endl;
+  
+    Ucheck2iAS = U2iAS * adj(U2iAS) - uno2iAS;
+    std::cout << GridLogMessage << "orthogonality check 1: " << norm2(Ucheck2iAS)
+	      << std::endl;
+    Ucheck2iAS = adj(U2iAS) * U2iAS - uno2iAS;
+    std::cout << GridLogMessage << "orthogonality check 2: " << norm2(Ucheck2iAS)
+	      << std::endl;
+  
+  
+  
+    // Construct the fundamental matrix in the group
+    SU3::LatticeMatrix Af_Asym(grid);
+    SU3::FundamentalLieAlgebraMatrix(h_Asym,Af_Asym);
+    SU3::LatticeMatrix Ufund2A(grid);
+    Ufund2A  = expMat(Af_Asym, 1.0, 16);
+    SU3::LatticeMatrix UnitCheck2A(grid);
+    UnitCheck2A = Ufund2A * adj(Ufund2A) - uno_f;
+    std::cout << GridLogMessage << "unitarity check 1: " << norm2(UnitCheck2A)
+	      << std::endl;
+    UnitCheck2A = adj(Ufund2A) * Ufund2A - uno_f;
+    std::cout << GridLogMessage << "unitarity check 2: " << norm2(UnitCheck2A)
+	      << std::endl;
+  
+
+    // Tranform to the 2Index Sym representation
+    U = Zero(); // fill this with only one direction
+    pokeLorentz(U,Ufund2A,0); // the representation transf acts on full gauge fields
+  
+    TIndexRepA.update_representation(U);
+    Ur2A = TIndexRepA.U;  // U_r  
+    typename TwoIndexRep< Nc, AntiSymmetric>::LatticeMatrix Ur02A = peekLorentz(Ur2A,0); // this should be the same as U2iS
+  
+    typename TwoIndexRep< Nc, AntiSymmetric>::LatticeMatrix Diff_check_mat2A = Ur02A - U2iAS;
+    std::cout << GridLogMessage << "Projections structure check group difference (Two Index anti-Symmetric): " << norm2(Diff_check_mat2A) << std::endl;
+  
+  } else  {
+    std::cout << GridLogMessage << "Skipping Two Index anti-Symmetric tests "
+      "because representation is trivial (dim = 1)"
+	      << std::endl;
   }
-  
-  TIndexRep.update_representation(UV2A);
-  typename TwoIndexRep< Nc, AntiSymmetric >::LatticeField UVr2A = TIndexRepA.U;  // (U_f * V_f)_r
-  
-  TIndexRep.update_representation(U2A);
-  typename TwoIndexRep< Nc, AntiSymmetric >::LatticeField Ur2A = TIndexRepA.U;  // U_r
-  
-  TIndexRep.update_representation(V2A);
-  typename TwoIndexRep< Nc, AntiSymmetric >::LatticeField Vr2A = TIndexRepA.U;  // V_r
-  
-  typename TwoIndexRep< Nc, AntiSymmetric >::LatticeField Ur2Vr2A(grid);
-  Ur2Vr2A = Zero();
-  for (int mu = 0; mu < Nd; mu++) {
-    typename TwoIndexRep< Nc, AntiSymmetric >::LatticeMatrix Urmu2A = peekLorentz(Ur2A,mu);
-    typename TwoIndexRep< Nc, AntiSymmetric >::LatticeMatrix Vrmu2A = peekLorentz(Vr2A,mu);
-    pokeLorentz(Ur2Vr2A,Urmu2A*Vrmu2A, mu);
-  }
-  
-  typename TwoIndexRep< Nc, AntiSymmetric >::LatticeField Diff_check2A = UVr2A - Ur2Vr2A;
-  std::cout << GridLogMessage << "Group structure SU("<<Nc<<") check difference (Two Index anti-Symmetric): " << norm2(Diff_check2A) << std::endl;
-
-  
-  // Check correspondence of algebra and group transformations
-  // Create a random vector
-  SU3::LatticeAlgebraVector h_Asym(grid);
-  typename TwoIndexRep< Nc, AntiSymmetric>::LatticeMatrix Ar_Asym(grid);
-  random(gridRNG,h_Asym);
-  h_Asym = real(h_Asym);
-  SU_TwoIndex< Nc, AntiSymmetric>::TwoIndexLieAlgebraMatrix(h_Asym,Ar_Asym);
-  
-  // Re-extract h_sym
-  SU3::LatticeAlgebraVector h_Asym2(grid);
-  SU_TwoIndex< Nc, AntiSymmetric>::projectOnAlgebra(h_Asym2, Ar_Asym);
-  SU3::LatticeAlgebraVector h_diff_Asym = h_Asym - h_Asym2;
-  std::cout << GridLogMessage << "Projections structure check vector difference (Two Index anti-Symmetric): " << norm2(h_diff_Asym) << std::endl;
-
-  
-  // Exponentiate
-  typename TwoIndexRep< Nc, AntiSymmetric>::LatticeMatrix U2iAS(grid);
-  U2iAS  = expMat(Ar_Asym, 1.0, 16);
-  
-  typename TwoIndexRep< Nc, AntiSymmetric>::LatticeMatrix uno2iAS(grid);
-  uno2iAS = 1.0;
-  // Check matrix U2iS, must be real orthogonal
-  typename TwoIndexRep< Nc, AntiSymmetric>::LatticeMatrix Ucheck2iAS = U2iAS - conjugate(U2iAS);
-  std::cout << GridLogMessage << "Reality check: " << norm2(Ucheck2iAS)
-      << std::endl;
-  
-  Ucheck2iAS = U2iAS * adj(U2iAS) - uno2iAS;
-  std::cout << GridLogMessage << "orthogonality check 1: " << norm2(Ucheck2iAS)
-      << std::endl;
-  Ucheck2iAS = adj(U2iAS) * U2iAS - uno2iAS;
-  std::cout << GridLogMessage << "orthogonality check 2: " << norm2(Ucheck2iAS)
-      << std::endl;
-  
-  
-  
-  // Construct the fundamental matrix in the group
-  SU3::LatticeMatrix Af_Asym(grid);
-  SU3::FundamentalLieAlgebraMatrix(h_Asym,Af_Asym);
-  SU3::LatticeMatrix Ufund2A(grid);
-  Ufund2A  = expMat(Af_Asym, 1.0, 16);
-  SU3::LatticeMatrix UnitCheck2A(grid);
-  UnitCheck2A = Ufund2A * adj(Ufund2A) - uno_f;
-  std::cout << GridLogMessage << "unitarity check 1: " << norm2(UnitCheck2A)
-      << std::endl;
-  UnitCheck2A = adj(Ufund2A) * Ufund2A - uno_f;
-  std::cout << GridLogMessage << "unitarity check 2: " << norm2(UnitCheck2A)
-      << std::endl;
-  
-
-  // Tranform to the 2Index Sym representation
-  U = Zero(); // fill this with only one direction
-  pokeLorentz(U,Ufund2A,0); // the representation transf acts on full gauge fields
-  
-  TIndexRepA.update_representation(U);
-  Ur2A = TIndexRepA.U;  // U_r  
-  typename TwoIndexRep< Nc, AntiSymmetric>::LatticeMatrix Ur02A = peekLorentz(Ur2A,0); // this should be the same as U2iS
-  
-  typename TwoIndexRep< Nc, AntiSymmetric>::LatticeMatrix Diff_check_mat2A = Ur02A - U2iAS;
-  std::cout << GridLogMessage << "Projections structure check group difference (Two Index anti-Symmetric): " << norm2(Diff_check_mat2A) << std::endl;
-  
-} else  {
-  std::cout << GridLogMessage << "Skipping Two Index anti-Symmetric tests "
-                                 "because representation is trivial (dim = 1)"
-            << std::endl;
-}
-
-
-
-
-
-
-
-
 
+#endif
   
   Grid_finalize();
 }

tests/core/Test_reunitarise.cc

@@ -122,14 +122,15 @@ int main (int argc, char ** argv)
   std::cout << "Determinant defect before projection " <<norm2(detU)<<std::endl;
   tmp = U*adj(U) - ident;
   std::cout << "Unitarity check before projection    " << norm2(tmp)<<std::endl; 
-  
+#if (Nc == 3)
   ProjectSU3(U);
   detU= Determinant(U) ;
   detU= detU -1.0;
   std::cout << "Determinant ProjectSU3 defect " <<norm2(detU)<<std::endl;
   tmp = U*adj(U) - ident;
   std::cout << "Unitarity check after projection    " << norm2(tmp)<<std::endl; 
-
+#endif
+  
   ProjectSUn(UU);
   detUU= Determinant(UU);
   detUU= detUU -1.0;