Updated TODO list

Curious if we can deprecate dependencly on BLAS.
Will see when we get 48^3 running on our BG/Q port

TODO

@@ -1,23 +1,30 @@
 TODO:
 ---------------
 
-Peter's work list:
-2)- Precision conversion and sort out localConvert      <-- 
-3)- Remove DenseVector, DenseMatrix; Use Eigen instead. <-- started 
-4)- Binary I/O speed up & x-strips
--- Profile CG, BlockCG, etc... Flop count/rate -- PARTIAL, time but no flop/s yet
--- Physical propagator interface
--- Conserved currents
--- GaugeFix into central location
--- Multigrid Wilson and DWF, compare to other Multigrid implementations
--- HDCR resume
+Large item work list:
+1)- MultiRHS with spread out extra dim -- Go through filesystem with SciDAC I/O
+
+2)- Christoph's local basis expansion Lanczos
+3)- BG/Q port and check
+4)- Precision conversion and sort out localConvert      <-- partial
+  - Consistent linear solver flop count/rate -- PARTIAL, time but no flop/s yet
+5)- Physical propagator interface
+6)- Conserved currents
+7)- Multigrid Wilson and DWF, compare to other Multigrid implementations
+8)- HDCR resume
 
 Recent DONE 
+-- Lanczos Remove DenseVector, DenseMatrix; Use Eigen instead. <-- DONE
+-- GaugeFix into central location                      <-- DONE
+-- Scidac and Ildg metadata handling                   <-- DONE
+-- Binary I/O MPI2 IO                                  <-- DONE
+-- Binary I/O speed up & x-strips                      <-- DONE
 -- Cut down the exterior overhead                      <-- DONE
 -- Interior legs from SHM comms                        <-- DONE
 -- Half-precision comms                                <-- DONE
--- Merge high precision reduction into develop        
--- multiRHS DWF; benchmark on Cori/BNL for comms elimination
+-- Merge high precision reduction into develop         <-- DONE
+-- BlockCG, BCGrQ                                      <-- DONE
+-- multiRHS DWF; benchmark on Cori/BNL for comms elimination <-- DONE
    -- slice* linalg routines for multiRHS, BlockCG    
 
 -----

benchmarks/Benchmark_comms.cc

@@ -31,6 +31,32 @@ using namespace std;
 using namespace Grid;
 using namespace Grid::QCD;
 
+struct time_statistics{
+  double mean;
+  double err;
+  double min;
+  double max;
+
+  void statistics(std::vector<double> v){
+      double sum = std::accumulate(v.begin(), v.end(), 0.0);
+      mean = sum / v.size();
+
+      std::vector<double> diff(v.size());
+      std::transform(v.begin(), v.end(), diff.begin(), [=](double x) { return x - mean; });
+      double sq_sum = std::inner_product(diff.begin(), diff.end(), diff.begin(), 0.0);
+      err = std::sqrt(sq_sum / (v.size()*(v.size() - 1)));
+
+      auto result = std::minmax_element(v.begin(), v.end());
+      min = *result.first;
+      max = *result.second;
+}
+};
+
+void header(){
+  std::cout <<GridLogMessage << " L  "<<"\t"<<" Ls  "<<"\t"
+            <<std::setw(11)<<"bytes"<<"MB/s uni (err/min/max)"<<"\t\t"<<"MB/s bidi (err/min/max)"<<std::endl;
+};
+
 int main (int argc, char ** argv)
 {
   Grid_init(&argc,&argv);
@@ -40,15 +66,19 @@ int main (int argc, char ** argv)
   int threads = GridThread::GetThreads();
   std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
 
-  int Nloop=10;
+  int Nloop=100;
   int nmu=0;
+  int maxlat=24;
   for(int mu=0;mu<Nd;mu++) if (mpi_layout[mu]>1) nmu++;
 
+  std::cout << GridLogMessage << "Number of iterations to average: "<< Nloop << std::endl;
+  std::vector<double> t_time(Nloop);
+  time_statistics timestat;
+
   std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
   std::cout<<GridLogMessage << "= Benchmarking concurrent halo exchange in "<<nmu<<" dimensions"<<std::endl;
   std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
-  std::cout<<GridLogMessage << "  L  "<<"\t\t"<<" Ls  "<<"\t\t"<<"bytes"<<"\t\t"<<"MB/s uni"<<"\t\t"<<"MB/s bidi"<<std::endl;
-  int maxlat=24;
+  header();
   for(int lat=4;lat<=maxlat;lat+=4){
     for(int Ls=8;Ls<=32;Ls*=2){
 
@@ -58,6 +88,9 @@ int main (int argc, char ** argv)
       				    lat*mpi_layout[3]});
 
       GridCartesian     Grid(latt_size,simd_layout,mpi_layout);
+      RealD Nrank = Grid._Nprocessors;
+      RealD Nnode = Grid.NodeCount();
+      RealD ppn = Nrank/Nnode;
 
       std::vector<std::vector<HalfSpinColourVectorD> > xbuf(8,std::vector<HalfSpinColourVectorD>(lat*lat*lat*Ls));
       std::vector<std::vector<HalfSpinColourVectorD> > rbuf(8,std::vector<HalfSpinColourVectorD>(lat*lat*lat*Ls));
@@ -65,8 +98,8 @@ int main (int argc, char ** argv)
       int ncomm;
       int bytes=lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD);
 
-      double start=usecond();
       for(int i=0;i<Nloop;i++){
+      double start=usecond();
 
 	std::vector<CartesianCommunicator::CommsRequest_t> requests;
 
@@ -102,18 +135,24 @@ int main (int argc, char ** argv)
 	}
 	Grid.SendToRecvFromComplete(requests);
 	Grid.Barrier();
-
+	double stop=usecond();
+	t_time[i] = stop-start; // microseconds
       }
-      double stop=usecond();
 
-      double dbytes    = bytes;
-      double xbytes    = Nloop*dbytes*2.0*ncomm;
+      timestat.statistics(t_time);
+
+      double dbytes    = bytes*ppn;
+      double xbytes    = dbytes*2.0*ncomm;
       double rbytes    = xbytes;
       double bidibytes = xbytes+rbytes;
 
-      double time = stop-start; // microseconds
+      std::cout<<GridLogMessage << std::setw(4) << lat<<"\t"<<Ls<<"\t"
+               <<std::setw(11) << bytes<< std::fixed << std::setprecision(1) << std::setw(7)
+               <<std::right<< xbytes/timestat.mean<<"  "<< xbytes*timestat.err/(timestat.mean*timestat.mean)<< " "
+               <<xbytes/timestat.max <<" "<< xbytes/timestat.min  
+               << "\t\t"<<std::setw(7)<< bidibytes/timestat.mean<< "  " << bidibytes*timestat.err/(timestat.mean*timestat.mean) << " "
+               << bidibytes/timestat.max << " " << bidibytes/timestat.min << std::endl;
 
-      std::cout<<GridLogMessage << lat<<"\t\t"<<Ls<<"\t\t"<<bytes<<"\t\t"<<xbytes/time<<"\t\t"<<bidibytes/time<<std::endl;
     }
   }    
 
@@ -121,8 +160,7 @@ int main (int argc, char ** argv)
   std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
   std::cout<<GridLogMessage << "= Benchmarking sequential halo exchange in "<<nmu<<" dimensions"<<std::endl;
   std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
-  std::cout<<GridLogMessage << "  L  "<<"\t\t"<<" Ls  "<<"\t\t"<<"bytes"<<"\t\t"<<"MB/s uni"<<"\t\t"<<"MB/s bidi"<<std::endl;
-
+  header();
 
   for(int lat=4;lat<=maxlat;lat+=4){
     for(int Ls=8;Ls<=32;Ls*=2){
@@ -130,6 +168,9 @@ int main (int argc, char ** argv)
       std::vector<int> latt_size  ({lat,lat,lat,lat});
 
       GridCartesian     Grid(latt_size,simd_layout,mpi_layout);
+      RealD Nrank = Grid._Nprocessors;
+      RealD Nnode = Grid.NodeCount();
+      RealD ppn = Nrank/Nnode;
 
       std::vector<std::vector<HalfSpinColourVectorD> > xbuf(8,std::vector<HalfSpinColourVectorD>(lat*lat*lat*Ls));
       std::vector<std::vector<HalfSpinColourVectorD> > rbuf(8,std::vector<HalfSpinColourVectorD>(lat*lat*lat*Ls));
@@ -138,8 +179,8 @@ int main (int argc, char ** argv)
       int ncomm;
       int bytes=lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD);
 
-      double start=usecond();
       for(int i=0;i<Nloop;i++){
+      double start=usecond();
     
 	ncomm=0;
 	for(int mu=0;mu<4;mu++){
@@ -178,27 +219,34 @@ int main (int argc, char ** argv)
 	  }
 	}
 	Grid.Barrier();
+	double stop=usecond();
+	t_time[i] = stop-start; // microseconds
+
       }
 
-      double stop=usecond();
+      timestat.statistics(t_time);
       
-      double dbytes    = bytes;
-      double xbytes    = Nloop*dbytes*2.0*ncomm;
+      double dbytes    = bytes*ppn;
+      double xbytes    = dbytes*2.0*ncomm;
       double rbytes    = xbytes;
       double bidibytes = xbytes+rbytes;
 
-      double time = stop-start;
+    std::cout<<GridLogMessage << std::setw(4) << lat<<"\t"<<Ls<<"\t"
+               <<std::setw(11) << bytes<< std::fixed << std::setprecision(1) << std::setw(7)
+               <<std::right<< xbytes/timestat.mean<<"  "<< xbytes*timestat.err/(timestat.mean*timestat.mean)<< " "
+               <<xbytes/timestat.max <<" "<< xbytes/timestat.min  
+               << "\t\t"<<std::setw(7)<< bidibytes/timestat.mean<< "  " << bidibytes*timestat.err/(timestat.mean*timestat.mean) << " "
+               << bidibytes/timestat.max << " " << bidibytes/timestat.min << std::endl;
 
-      std::cout<<GridLogMessage << lat<<"\t\t"<<Ls<<"\t\t"<<bytes<<"\t\t"<<xbytes/time<<"\t\t"<<bidibytes/time<<std::endl;
+      
     }
   }  
 
 
-  Nloop=10;
   std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
   std::cout<<GridLogMessage << "= Benchmarking concurrent STENCIL halo exchange in "<<nmu<<" dimensions"<<std::endl;
   std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
-  std::cout<<GridLogMessage << "  L  "<<"\t\t"<<" Ls  "<<"\t\t"<<"bytes"<<"\t\t"<<"MB/s uni"<<"\t\t"<<"MB/s bidi"<<std::endl;
+  header();
 
   for(int lat=4;lat<=maxlat;lat+=4){
     for(int Ls=8;Ls<=32;Ls*=2){
@@ -209,6 +257,9 @@ int main (int argc, char ** argv)
       				    lat*mpi_layout[3]});
 
       GridCartesian     Grid(latt_size,simd_layout,mpi_layout);
+      RealD Nrank = Grid._Nprocessors;
+      RealD Nnode = Grid.NodeCount();
+      RealD ppn = Nrank/Nnode;
 
       std::vector<HalfSpinColourVectorD *> xbuf(8);
       std::vector<HalfSpinColourVectorD *> rbuf(8);
@@ -216,70 +267,83 @@ int main (int argc, char ** argv)
       for(int d=0;d<8;d++){
 	xbuf[d] = (HalfSpinColourVectorD *)Grid.ShmBufferMalloc(lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD));
 	rbuf[d] = (HalfSpinColourVectorD *)Grid.ShmBufferMalloc(lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD));
+	bzero((void *)xbuf[d],lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD));
+	bzero((void *)rbuf[d],lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD));
       }
 
       int ncomm;
       int bytes=lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD);
 
-      double start=usecond();
+      double dbytes;
       for(int i=0;i<Nloop;i++){
+	double start=usecond();
+
+	dbytes=0;
+	ncomm=0;
 
 	std::vector<CartesianCommunicator::CommsRequest_t> requests;
 
-	ncomm=0;
 	for(int mu=0;mu<4;mu++){
 	
+
 	  if (mpi_layout[mu]>1 ) {
 	  
 	    ncomm++;
 	    int comm_proc=1;
 	    int xmit_to_rank;
 	    int recv_from_rank;
-	    
 	    Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank);
-	    Grid.StencilSendToRecvFromBegin(requests,
-					    (void *)&xbuf[mu][0],
-					    xmit_to_rank,
-					    (void *)&rbuf[mu][0],
-					    recv_from_rank,
-					    bytes);
+	    dbytes+=
+	      Grid.StencilSendToRecvFromBegin(requests,
+					      (void *)&xbuf[mu][0],
+					      xmit_to_rank,
+					      (void *)&rbuf[mu][0],
+					      recv_from_rank,
+					      bytes);
 	
 	    comm_proc = mpi_layout[mu]-1;
 	  
 	    Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank);
-	    Grid.StencilSendToRecvFromBegin(requests,
-					    (void *)&xbuf[mu+4][0],
-					    xmit_to_rank,
-					    (void *)&rbuf[mu+4][0],
-					    recv_from_rank,
-					    bytes);
+	    dbytes+=
+	      Grid.StencilSendToRecvFromBegin(requests,
+					      (void *)&xbuf[mu+4][0],
+					      xmit_to_rank,
+					      (void *)&rbuf[mu+4][0],
+					      recv_from_rank,
+					      bytes);
 	  
 	  }
 	}
 	Grid.StencilSendToRecvFromComplete(requests);
 	Grid.Barrier();
-
+	double stop=usecond();
+	t_time[i] = stop-start; // microseconds
+	
       }
-      double stop=usecond();
 
-      double dbytes    = bytes;
-      double xbytes    = Nloop*dbytes*2.0*ncomm;
-      double rbytes    = xbytes;
-      double bidibytes = xbytes+rbytes;
+      timestat.statistics(t_time);
+
+      dbytes=dbytes*ppn;
+      double xbytes    = dbytes*0.5;
+      double rbytes    = dbytes*0.5;
+      double bidibytes = dbytes;
+
+      std::cout<<GridLogMessage << std::setw(4) << lat<<"\t"<<Ls<<"\t"
+               <<std::setw(11) << bytes<< std::fixed << std::setprecision(1) << std::setw(7)
+               <<std::right<< xbytes/timestat.mean<<"  "<< xbytes*timestat.err/(timestat.mean*timestat.mean)<< " "
+               <<xbytes/timestat.max <<" "<< xbytes/timestat.min  
+               << "\t\t"<<std::setw(7)<< bidibytes/timestat.mean<< "  " << bidibytes*timestat.err/(timestat.mean*timestat.mean) << " "
+               << bidibytes/timestat.max << " " << bidibytes/timestat.min << std::endl;
 
-      double time = stop-start; // microseconds
 
-      std::cout<<GridLogMessage << lat<<"\t\t"<<Ls<<"\t\t"<<bytes<<"\t\t"<<xbytes/time<<"\t\t"<<bidibytes/time<<std::endl;
     }
   }    
 
 
-
-  Nloop=100;
   std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
   std::cout<<GridLogMessage << "= Benchmarking sequential STENCIL halo exchange in "<<nmu<<" dimensions"<<std::endl;
   std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
-  std::cout<<GridLogMessage << "  L  "<<"\t\t"<<" Ls  "<<"\t\t"<<"bytes"<<"\t\t"<<"MB/s uni"<<"\t\t"<<"MB/s bidi"<<std::endl;
+  header();
 
   for(int lat=4;lat<=maxlat;lat+=4){
     for(int Ls=8;Ls<=32;Ls*=2){
@@ -290,6 +354,9 @@ int main (int argc, char ** argv)
       				    lat*mpi_layout[3]});
 
       GridCartesian     Grid(latt_size,simd_layout,mpi_layout);
+      RealD Nrank = Grid._Nprocessors;
+      RealD Nnode = Grid.NodeCount();
+      RealD ppn = Nrank/Nnode;
 
       std::vector<HalfSpinColourVectorD *> xbuf(8);
       std::vector<HalfSpinColourVectorD *> rbuf(8);
@@ -297,16 +364,18 @@ int main (int argc, char ** argv)
       for(int d=0;d<8;d++){
 	xbuf[d] = (HalfSpinColourVectorD *)Grid.ShmBufferMalloc(lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD));
 	rbuf[d] = (HalfSpinColourVectorD *)Grid.ShmBufferMalloc(lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD));
+	bzero((void *)xbuf[d],lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD));
+	bzero((void *)rbuf[d],lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD));
       }
 
       int ncomm;
       int bytes=lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD);
-
-      double start=usecond();
+      double dbytes;
       for(int i=0;i<Nloop;i++){
+	double start=usecond();
 
 	std::vector<CartesianCommunicator::CommsRequest_t> requests;
-
+	dbytes=0;
 	ncomm=0;
 	for(int mu=0;mu<4;mu++){
 	
@@ -318,42 +387,52 @@ int main (int argc, char ** argv)
 	    int recv_from_rank;
 	    
 	    Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank);
-	    Grid.StencilSendToRecvFromBegin(requests,
-					    (void *)&xbuf[mu][0],
-					    xmit_to_rank,
-					    (void *)&rbuf[mu][0],
-					    recv_from_rank,
-					    bytes);
+	    dbytes+=
+	      Grid.StencilSendToRecvFromBegin(requests,
+					      (void *)&xbuf[mu][0],
+					      xmit_to_rank,
+					      (void *)&rbuf[mu][0],
+					      recv_from_rank,
+					      bytes);
 	    Grid.StencilSendToRecvFromComplete(requests);
 	    requests.resize(0);
 
 	    comm_proc = mpi_layout[mu]-1;
 	  
 	    Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank);
-	    Grid.StencilSendToRecvFromBegin(requests,
-					    (void *)&xbuf[mu+4][0],
-					    xmit_to_rank,
-					    (void *)&rbuf[mu+4][0],
-					    recv_from_rank,
-					    bytes);
+	    dbytes+=
+	      Grid.StencilSendToRecvFromBegin(requests,
+					      (void *)&xbuf[mu+4][0],
+					      xmit_to_rank,
+					      (void *)&rbuf[mu+4][0],
+					      recv_from_rank,
+					      bytes);
 	    Grid.StencilSendToRecvFromComplete(requests);
 	    requests.resize(0);
 	  
 	  }
 	}
 	Grid.Barrier();
-
+	double stop=usecond();
+	t_time[i] = stop-start; // microseconds
+	
       }
-      double stop=usecond();
 
-      double dbytes    = bytes;
-      double xbytes    = Nloop*dbytes*2.0*ncomm;
-      double rbytes    = xbytes;
-      double bidibytes = xbytes+rbytes;
+      timestat.statistics(t_time);
 
-      double time = stop-start; // microseconds
+      dbytes=dbytes*ppn;
+      double xbytes    = dbytes*0.5;
+      double rbytes    = dbytes*0.5;
+      double bidibytes = dbytes;
 
-      std::cout<<GridLogMessage << lat<<"\t\t"<<Ls<<"\t\t"<<bytes<<"\t\t"<<xbytes/time<<"\t\t"<<bidibytes/time<<std::endl;
+
+      std::cout<<GridLogMessage << std::setw(4) << lat<<"\t"<<Ls<<"\t"
+               <<std::setw(11) << bytes<< std::fixed << std::setprecision(1) << std::setw(7)
+               <<std::right<< xbytes/timestat.mean<<"  "<< xbytes*timestat.err/(timestat.mean*timestat.mean)<< " "
+               <<xbytes/timestat.max <<" "<< xbytes/timestat.min  
+               << "\t\t"<<std::setw(7)<< bidibytes/timestat.mean<< "  " << bidibytes*timestat.err/(timestat.mean*timestat.mean) << " "
+               << bidibytes/timestat.max << " " << bidibytes/timestat.min << std::endl;
+ 
     }
   }    
 

benchmarks/Benchmark_memory_bandwidth.cc

@@ -55,8 +55,8 @@ int main (int argc, char ** argv)
   std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
   std::cout<<GridLogMessage << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s"<<"\t\t"<<"Gflop/s"<<"\t\t seconds"<<std::endl;
   std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
-  uint64_t lmax=44;
-#define NLOOP (1*lmax*lmax*lmax*lmax/vol)
+  uint64_t lmax=64;
+#define NLOOP (100*lmax*lmax*lmax*lmax/vol)
   for(int lat=4;lat<=lmax;lat+=4){
 
       std::vector<int> latt_size  ({lat*mpi_layout[0],lat*mpi_layout[1],lat*mpi_layout[2],lat*mpi_layout[3]});

benchmarks/Benchmark_su3.cc

@@ -35,9 +35,9 @@ using namespace Grid::QCD;
 int main (int argc, char ** argv)
 {
   Grid_init(&argc,&argv);
-#define LMAX (32)
+#define LMAX (64)
 
-  int Nloop=200;
+  int Nloop=20;
 
   std::vector<int> simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd());
   std::vector<int> mpi_layout  = GridDefaultMpi();

bootstrap.sh

@@ -1,4 +1,4 @@
-]#!/usr/bin/env bash
+#!/usr/bin/env bash
 
 EIGEN_URL='http://bitbucket.org/eigen/eigen/get/3.3.3.tar.bz2'
 

configure.ac

@@ -27,7 +27,7 @@ AX_GXX_VERSION
 AC_DEFINE_UNQUOTED([GXX_VERSION],["$GXX_VERSION"],
       [version of g++ that will compile the code])
 
-CXXFLAGS="-O3 $CXXFLAGS"
+CXXFLAGS="-g $CXXFLAGS"
 
 
 ############### Checks for typedefs, structures, and compiler characteristics
@@ -184,6 +184,10 @@ AC_SEARCH_LIBS([limeCreateReader], [lime],
 In order to use ILGG file format please install or provide the correct path to your installation
 Info at: http://usqcd.jlab.org/usqcd-docs/c-lime/)])
 
+AC_SEARCH_LIBS([crc32], [z],
+               [AC_DEFINE([HAVE_ZLIB], [1], [Define to 1 if you have the `LIBZ' library])]
+               [have_zlib=true],
+	       [AC_MSG_ERROR(zlib library was not found in your system.)])
 
 AC_SEARCH_LIBS([H5Fopen], [hdf5_cpp],
                [AC_DEFINE([HAVE_HDF5], [1], [Define to 1 if you have the `HDF5' library])]

extras/Hadrons/Modules/MAction/DWF.hpp

@@ -48,7 +48,8 @@ public:
                                     std::string, gauge,
                                     unsigned int, Ls,
                                     double      , mass,
-                                    double      , M5);
+                                    double      , M5,
+                                    std::string , boundary);
 };
 
 template <typename FImpl>
@@ -116,14 +117,19 @@ void TDWF<FImpl>::execute(void)
                  << par().mass << ", M5= " << par().M5 << " and Ls= "
                  << par().Ls << " using gauge field '" << par().gauge << "'"
                  << std::endl;
+    LOG(Message) << "Fermion boundary conditions: " << par().boundary 
+                 << std::endl;
     env().createGrid(par().Ls);
     auto &U      = *env().template getObject<LatticeGaugeField>(par().gauge);
     auto &g4     = *env().getGrid();
     auto &grb4   = *env().getRbGrid();
     auto &g5     = *env().getGrid(par().Ls);
     auto &grb5   = *env().getRbGrid(par().Ls);
+    std::vector<Complex> boundary = strToVec<Complex>(par().boundary);
+    typename DomainWallFermion<FImpl>::ImplParams implParams(boundary);
     FMat *fMatPt = new DomainWallFermion<FImpl>(U, g5, grb5, g4, grb4,
-                                                par().mass, par().M5);
+                                                par().mass, par().M5,
+                                                implParams);
     env().setObject(getName(), fMatPt);
 }
 

extras/Hadrons/Modules/MAction/Wilson.hpp

@@ -46,7 +46,8 @@ class WilsonPar: Serializable
 public:
     GRID_SERIALIZABLE_CLASS_MEMBERS(WilsonPar,
                                     std::string, gauge,
-                                    double     , mass);
+                                    double     , mass,
+                                    std::string, boundary);
 };
 
 template <typename FImpl>
@@ -112,10 +113,15 @@ void TWilson<FImpl>::execute()
 {
     LOG(Message) << "Setting up TWilson fermion matrix with m= " << par().mass
                  << " using gauge field '" << par().gauge << "'" << std::endl;
+    LOG(Message) << "Fermion boundary conditions: " << par().boundary 
+                 << std::endl;
     auto &U      = *env().template getObject<LatticeGaugeField>(par().gauge);
     auto &grid   = *env().getGrid();
     auto &gridRb = *env().getRbGrid();
-    FMat *fMatPt = new WilsonFermion<FImpl>(U, grid, gridRb, par().mass);
+    std::vector<Complex> boundary = strToVec<Complex>(par().boundary);
+    typename WilsonFermion<FImpl>::ImplParams implParams(boundary);
+    FMat *fMatPt = new WilsonFermion<FImpl>(U, grid, gridRb, par().mass,
+                                            implParams);
     env().setObject(getName(), fMatPt);
 }
 

extras/Hadrons/Modules/MContraction/Meson.hpp

@@ -131,12 +131,11 @@ std::vector<std::string> TMeson<FImpl1, FImpl2>::getOutput(void)
 template <typename FImpl1, typename FImpl2>
 void TMeson<FImpl1, FImpl2>::parseGammaString(std::vector<GammaPair> &gammaList)
 {
+    gammaList.clear();
     // Determine gamma matrices to insert at source/sink.
     if (par().gammas.compare("all") == 0)
     {
         // Do all contractions.
-        unsigned int n_gam = Ns * Ns;
-        gammaList.resize(n_gam*n_gam);
         for (unsigned int i = 1; i < Gamma::nGamma; i += 2)
         {
             for (unsigned int j = 1; j < Gamma::nGamma; j += 2)

extras/Hadrons/Modules/MGauge/Load.cc

@@ -65,7 +65,7 @@ void TLoad::setup(void)
 // execution ///////////////////////////////////////////////////////////////////
 void TLoad::execute(void)
 {
-    NerscField  header;
+    FieldMetaData  header;
     std::string fileName = par().file + "."
                            + std::to_string(env().getTrajectory());
     
@@ -74,5 +74,5 @@ void TLoad::execute(void)
     LatticeGaugeField &U = *env().createLattice<LatticeGaugeField>(getName());
     NerscIO::readConfiguration(U, header, fileName);
     LOG(Message) << "NERSC header:" << std::endl;
-    dump_nersc_header(header, LOG(Message));
+    dump_meta_data(header, LOG(Message));
 }

lib/Grid.h

@@ -41,7 +41,9 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 #include <Grid/GridCore.h>
 #include <Grid/GridQCDcore.h>
 #include <Grid/qcd/action/Action.h>
+#include <Grid/qcd/utils/GaugeFix.h>
 #include <Grid/qcd/smearing/Smearing.h>
+#include <Grid/parallelIO/MetaData.h>
 #include <Grid/qcd/hmc/HMC_aggregate.h>
 
 #endif

lib/GridStd.h

@@ -7,6 +7,7 @@
 #include <cassert>
 #include <complex>
 #include <vector>
+#include <string>
 #include <iostream>
 #include <iomanip>
 #include <random>
@@ -18,6 +19,7 @@
 #include <ctime>
 #include <sys/time.h>
 #include <chrono>
+#include <zlib.h>
 
 ///////////////////
 // Grid config

lib/algorithms/densematrix/DenseMatrix.h

@@ -1,137 +0,0 @@
-    /*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/algorithms/iterative/DenseMatrix.h
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-    *************************************************************************************/
-    /*  END LEGAL */
-#ifndef GRID_DENSE_MATRIX_H
-#define GRID_DENSE_MATRIX_H
-
-namespace Grid {
-    /////////////////////////////////////////////////////////////
-    // Matrix untils
-    /////////////////////////////////////////////////////////////
-
-template<class T> using DenseVector = std::vector<T>;
-template<class T> using DenseMatrix = DenseVector<DenseVector<T> >;
-
-template<class T> void Size(DenseVector<T> & vec, int &N) 
-{ 
-  N= vec.size();
-}
-template<class T> void Size(DenseMatrix<T> & mat, int &N,int &M) 
-{ 
-  N= mat.size();
-  M= mat[0].size();
-}
-
-template<class T> void SizeSquare(DenseMatrix<T> & mat, int &N) 
-{ 
-  int M; Size(mat,N,M);
-  assert(N==M);
-}
-
-template<class T> void Resize(DenseVector<T > & mat, int N) { 
-  mat.resize(N);
-}
-template<class T> void Resize(DenseMatrix<T > & mat, int N, int M) { 
-  mat.resize(N);
-  for(int i=0;i<N;i++){
-    mat[i].resize(M);
-  }
-}
-template<class T> void Fill(DenseMatrix<T> & mat, T&val) { 
-  int N,M;
-  Size(mat,N,M);
-  for(int i=0;i<N;i++){
-  for(int j=0;j<M;j++){
-    mat[i][j] = val;
-  }}
-}
-
-/** Transpose of a matrix **/
-template<class T> DenseMatrix<T> Transpose(DenseMatrix<T> & mat){
-  int N,M;
-  Size(mat,N,M);
-  DenseMatrix<T> C; Resize(C,M,N);
-  for(int i=0;i<M;i++){
-  for(int j=0;j<N;j++){
-    C[i][j] = mat[j][i];
-  }} 
-  return C;
-}
-/** Set DenseMatrix to unit matrix **/
-template<class T> void Unity(DenseMatrix<T> &A){
-  int N;  SizeSquare(A,N);
-  for(int i=0;i<N;i++){
-    for(int j=0;j<N;j++){
-      if ( i==j ) A[i][j] = 1;
-      else        A[i][j] = 0;
-    } 
-  } 
-}
-
-/** Add C * I to matrix **/
-template<class T>
-void PlusUnit(DenseMatrix<T> & A,T c){
-  int dim;  SizeSquare(A,dim);
-  for(int i=0;i<dim;i++){A[i][i] = A[i][i] + c;} 
-}
-
-/** return the Hermitian conjugate of matrix **/
-template<class T>
-DenseMatrix<T> HermitianConj(DenseMatrix<T> &mat){
-
-  int dim; SizeSquare(mat,dim);
-
-  DenseMatrix<T> C; Resize(C,dim,dim);
-
-  for(int i=0;i<dim;i++){
-    for(int j=0;j<dim;j++){
-      C[i][j] = conj(mat[j][i]);
-    } 
-  } 
-  return C;
-}
-/**Get a square submatrix**/
-template <class T>
-DenseMatrix<T> GetSubMtx(DenseMatrix<T> &A,int row_st, int row_end, int col_st, int col_end)
-{
-  DenseMatrix<T> H; Resize(H,row_end - row_st,col_end-col_st);
-
-  for(int i = row_st; i<row_end; i++){
-  for(int j = col_st; j<col_end; j++){
-    H[i-row_st][j-col_st]=A[i][j];
-  }}
-  return H;
-}
-
-}
-
-#include "Householder.h"
-#include "Francis.h"
-
-#endif
-

lib/algorithms/densematrix/Francis.h

@@ -1,525 +0,0 @@
-    /*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/algorithms/iterative/Francis.h
-
-    Copyright (C) 2015
-
-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 */
-#ifndef FRANCIS_H
-#define FRANCIS_H
-
-#include <cstdlib>
-#include <string>
-#include <cmath>
-#include <iostream>
-#include <sstream>
-#include <stdexcept>
-#include <fstream>
-#include <complex>
-#include <algorithm>
-
-//#include <timer.h>
-//#include <lapacke.h>
-//#include <Eigen/Dense>
-
-namespace Grid {
-
-template <class T> int SymmEigensystem(DenseMatrix<T > &Ain, DenseVector<T> &evals, DenseMatrix<T> &evecs, RealD small);
-template <class T> int     Eigensystem(DenseMatrix<T > &Ain, DenseVector<T> &evals, DenseMatrix<T> &evecs, RealD small);
-
-/**
-  Find the eigenvalues of an upper hessenberg matrix using the Francis QR algorithm.
-H =
-      x  x  x  x  x  x  x  x  x
-      x  x  x  x  x  x  x  x  x
-      0  x  x  x  x  x  x  x  x
-      0  0  x  x  x  x  x  x  x
-      0  0  0  x  x  x  x  x  x
-      0  0  0  0  x  x  x  x  x
-      0  0  0  0  0  x  x  x  x
-      0  0  0  0  0  0  x  x  x
-      0  0  0  0  0  0  0  x  x
-Factorization is P T P^H where T is upper triangular (mod cc blocks) and P is orthagonal/unitary.
-**/
-template <class T>
-int QReigensystem(DenseMatrix<T> &Hin, DenseVector<T> &evals, DenseMatrix<T> &evecs, RealD small)
-{
-  DenseMatrix<T> H = Hin; 
-
-  int N ; SizeSquare(H,N);
-  int M = N;
-
-  Fill(evals,0);
-  Fill(evecs,0);
-
-  T s,t,x=0,y=0,z=0;
-  T u,d;
-  T apd,amd,bc;
-  DenseVector<T> p(N,0);
-  T nrm = Norm(H);    ///DenseMatrix Norm
-  int n, m;
-  int e = 0;
-  int it = 0;
-  int tot_it = 0;
-  int l = 0;
-  int r = 0;
-  DenseMatrix<T> P; Resize(P,N,N); Unity(P);
-  DenseVector<int> trows(N,0);
-
-  /// Check if the matrix is really hessenberg, if not abort
-  RealD sth = 0;
-  for(int j=0;j<N;j++){
-    for(int i=j+2;i<N;i++){
-      sth = abs(H[i][j]);
-      if(sth > small){
-	std::cout << "Non hessenberg H = " << sth << " > " << small << std::endl;
-	exit(1);
-      }
-    }
-  }
-
-  do{
-    std::cout << "Francis QR Step N = " << N << std::endl;
-    /** Check for convergence
-      x  x  x  x  x
-      0  x  x  x  x
-      0  0  x  x  x
-      0  0  x  x  x
-      0  0  0  0  x
-      for this matrix l = 4
-     **/
-    do{
-      l = Chop_subdiag(H,nrm,e,small);
-      r = 0;    ///May have converged on more than one eval
-      ///Single eval
-      if(l == N-1){
-        evals[e] = H[l][l];
-        N--; e++; r++; it = 0;
-      }
-      ///RealD eval
-      if(l == N-2){
-        trows[l+1] = 1;    ///Needed for UTSolve
-        apd = H[l][l] + H[l+1][l+1];
-        amd = H[l][l] - H[l+1][l+1];
-        bc =  (T)4.0*H[l+1][l]*H[l][l+1];
-        evals[e]   = (T)0.5*( apd + sqrt(amd*amd + bc) );
-        evals[e+1] = (T)0.5*( apd - sqrt(amd*amd + bc) );
-        N-=2; e+=2; r++; it = 0;
-      }
-    } while(r>0);
-
-    if(N ==0) break;
-
-    DenseVector<T > ck; Resize(ck,3);
-    DenseVector<T> v;   Resize(v,3);
-
-    for(int m = N-3; m >= l; m--){
-      ///Starting vector essentially random shift.
-      if(it%10 == 0 && N >= 3 && it > 0){
-        s = (T)1.618033989*( abs( H[N-1][N-2] ) + abs( H[N-2][N-3] ) );
-        t = (T)0.618033989*( abs( H[N-1][N-2] ) + abs( H[N-2][N-3] ) );
-        x = H[m][m]*H[m][m] + H[m][m+1]*H[m+1][m] - s*H[m][m] + t;
-        y = H[m+1][m]*(H[m][m] + H[m+1][m+1] - s);
-        z = H[m+1][m]*H[m+2][m+1];
-      }
-      ///Starting vector implicit Q theorem
-      else{
-        s = (H[N-2][N-2] + H[N-1][N-1]);
-        t = (H[N-2][N-2]*H[N-1][N-1] - H[N-2][N-1]*H[N-1][N-2]);
-        x = H[m][m]*H[m][m] + H[m][m+1]*H[m+1][m] - s*H[m][m] + t;
-        y = H[m+1][m]*(H[m][m] + H[m+1][m+1] - s);
-        z = H[m+1][m]*H[m+2][m+1];
-      }
-      ck[0] = x; ck[1] = y; ck[2] = z;
-
-      if(m == l) break;
-
-      /** Some stupid thing from numerical recipies, seems to work**/
-      // PAB.. for heaven's sake quote page, purpose, evidence it works.
-      //       what sort of comment is that!?!?!?
-      u=abs(H[m][m-1])*(abs(y)+abs(z));
-      d=abs(x)*(abs(H[m-1][m-1])+abs(H[m][m])+abs(H[m+1][m+1]));
-      if ((T)abs(u+d) == (T)abs(d) ){
-	l = m; break;
-      }
-
-      //if (u < small){l = m; break;}
-    }
-    if(it > 100000){
-     std::cout << "QReigensystem: bugger it got stuck after 100000 iterations" << std::endl;
-     std::cout << "got " << e << " evals " << l << " " << N << std::endl;
-      exit(1);
-    }
-    normalize(ck);    ///Normalization cancels in PHP anyway
-    T beta;
-    Householder_vector<T >(ck, 0, 2, v, beta);
-    Householder_mult<T >(H,v,beta,0,l,l+2,0);
-    Householder_mult<T >(H,v,beta,0,l,l+2,1);
-    ///Accumulate eigenvector
-    Householder_mult<T >(P,v,beta,0,l,l+2,1);
-    int sw = 0;      ///Are we on the last row?
-    for(int k=l;k<N-2;k++){
-      x = H[k+1][k];
-      y = H[k+2][k];
-      z = (T)0.0;
-      if(k+3 <= N-1){
-	z = H[k+3][k];
-      } else{
-	sw = 1; 
-	v[2] = (T)0.0;
-      }
-      ck[0] = x; ck[1] = y; ck[2] = z;
-      normalize(ck);
-      Householder_vector<T >(ck, 0, 2-sw, v, beta);
-      Householder_mult<T >(H,v, beta,0,k+1,k+3-sw,0);
-      Householder_mult<T >(H,v, beta,0,k+1,k+3-sw,1);
-      ///Accumulate eigenvector
-      Householder_mult<T >(P,v, beta,0,k+1,k+3-sw,1);
-    }
-    it++;
-    tot_it++;
-  }while(N > 1);
-  N = evals.size();
-  ///Annoying - UT solves in reverse order;
-  DenseVector<T> tmp; Resize(tmp,N);
-  for(int i=0;i<N;i++){
-    tmp[i] = evals[N-i-1];
-  } 
-  evals = tmp;
-  UTeigenvectors(H, trows, evals, evecs);
-  for(int i=0;i<evals.size();i++){evecs[i] = P*evecs[i]; normalize(evecs[i]);}
-  return tot_it;
-}
-
-template <class T>
-int my_Wilkinson(DenseMatrix<T> &Hin, DenseVector<T> &evals, DenseMatrix<T> &evecs, RealD small)
-{
-  /**
-  Find the eigenvalues of an upper Hessenberg matrix using the Wilkinson QR algorithm.
-  H =
-  x  x  0  0  0  0
-  x  x  x  0  0  0
-  0  x  x  x  0  0
-  0  0  x  x  x  0
-  0  0  0  x  x  x
-  0  0  0  0  x  x
-  Factorization is P T P^H where T is upper triangular (mod cc blocks) and P is orthagonal/unitary.  **/
-  return my_Wilkinson(Hin, evals, evecs, small, small);
-}
-
-template <class T>
-int my_Wilkinson(DenseMatrix<T> &Hin, DenseVector<T> &evals, DenseMatrix<T> &evecs, RealD small, RealD tol)
-{
-  int N; SizeSquare(Hin,N);
-  int M = N;
-
-  ///I don't want to modify the input but matricies must be passed by reference
-  //Scale a matrix by its "norm"
-  //RealD Hnorm = abs( Hin.LargestDiag() ); H =  H*(1.0/Hnorm);
-  DenseMatrix<T> H;  H = Hin;
-  
-  RealD Hnorm = abs(Norm(Hin));
-  H = H * (1.0 / Hnorm);
-
-  // TODO use openmp and memset
-  Fill(evals,0);
-  Fill(evecs,0);
-
-  T s, t, x = 0, y = 0, z = 0;
-  T u, d;
-  T apd, amd, bc;
-  DenseVector<T> p; Resize(p,N); Fill(p,0);
-
-  T nrm = Norm(H);    ///DenseMatrix Norm
-  int n, m;
-  int e = 0;
-  int it = 0;
-  int tot_it = 0;
-  int l = 0;
-  int r = 0;
-  DenseMatrix<T> P; Resize(P,N,N);
-  Unity(P);
-  DenseVector<int> trows(N, 0);
-  /// Check if the matrix is really symm tridiag
-  RealD sth = 0;
-  for(int j = 0; j < N; ++j)
-  {
-    for(int i = j + 2; i < N; ++i)
-    {
-      if(abs(H[i][j]) > tol || abs(H[j][i]) > tol)
-      {
-	std::cout << "Non Tridiagonal H(" << i << ","<< j << ") = |" << Real( real( H[j][i] ) ) << "| > " << tol << std::endl;
-	std::cout << "Warning tridiagonalize and call again" << std::endl;
-        // exit(1); // see what is going on
-        //return;
-      }
-    }
-  }
-
-  do{
-    do{
-      //Jasper
-      //Check if the subdiagonal term is small enough (<small)
-      //if true then it is converged.
-      //check start from H.dim - e - 1
-      //How to deal with more than 2 are converged?
-      //What if Chop_symm_subdiag return something int the middle?
-      //--------------
-      l = Chop_symm_subdiag(H,nrm, e, small);
-      r = 0;    ///May have converged on more than one eval
-      //Jasper
-      //In this case
-      // x  x  0  0  0  0
-      // x  x  x  0  0  0
-      // 0  x  x  x  0  0
-      // 0  0  x  x  x  0
-      // 0  0  0  x  x  0
-      // 0  0  0  0  0  x  <- l
-      //--------------
-      ///Single eval
-      if(l == N - 1)
-      {
-        evals[e] = H[l][l];
-        N--;
-        e++;
-        r++;
-        it = 0;
-      }
-      //Jasper
-      // x  x  0  0  0  0
-      // x  x  x  0  0  0
-      // 0  x  x  x  0  0
-      // 0  0  x  x  0  0
-      // 0  0  0  0  x  x  <- l
-      // 0  0  0  0  x  x
-      //--------------
-      ///RealD eval
-      if(l == N - 2)
-      {
-        trows[l + 1] = 1;    ///Needed for UTSolve
-        apd = H[l][l] + H[l + 1][ l + 1];
-        amd = H[l][l] - H[l + 1][l + 1];
-        bc =  (T) 4.0 * H[l + 1][l] * H[l][l + 1];
-        evals[e] = (T) 0.5 * (apd + sqrt(amd * amd + bc));
-        evals[e + 1] = (T) 0.5 * (apd - sqrt(amd * amd + bc));
-        N -= 2;
-        e += 2;
-        r++;
-        it = 0;
-      }
-    }while(r > 0);
-    //Jasper
-    //Already converged
-    //--------------
-    if(N == 0) break;
-
-    DenseVector<T> ck,v; Resize(ck,2); Resize(v,2);
-
-    for(int m = N - 3; m >= l; m--)
-    {
-      ///Starting vector essentially random shift.
-      if(it%10 == 0 && N >= 3 && it > 0)
-      {
-        t = abs(H[N - 1][N - 2]) + abs(H[N - 2][N - 3]);
-        x = H[m][m] - t;
-        z = H[m + 1][m];
-      } else {
-      ///Starting vector implicit Q theorem
-        d = (H[N - 2][N - 2] - H[N - 1][N - 1]) * (T) 0.5;
-        t =  H[N - 1][N - 1] - H[N - 1][N - 2] * H[N - 1][N - 2] 
-	  / (d + sign(d) * sqrt(d * d + H[N - 1][N - 2] * H[N - 1][N - 2]));
-        x = H[m][m] - t;
-        z = H[m + 1][m];
-      }
-      //Jasper
-      //why it is here????
-      //-----------------------
-      if(m == l)
-        break;
-
-      u = abs(H[m][m - 1]) * (abs(y) + abs(z));
-      d = abs(x) * (abs(H[m - 1][m - 1]) + abs(H[m][m]) + abs(H[m + 1][m + 1]));
-      if ((T)abs(u + d) == (T)abs(d))
-      {
-        l = m;
-        break;
-      }
-    }
-    //Jasper
-    if(it > 1000000)
-    {
-      std::cout << "Wilkinson: bugger it got stuck after 100000 iterations" << std::endl;
-      std::cout << "got " << e << " evals " << l << " " << N << std::endl;
-      exit(1);
-    }
-    //
-    T s, c;
-    Givens_calc<T>(x, z, c, s);
-    Givens_mult<T>(H, l, l + 1, c, -s, 0);
-    Givens_mult<T>(H, l, l + 1, c,  s, 1);
-    Givens_mult<T>(P, l, l + 1, c,  s, 1);
-    //
-    for(int k = l; k < N - 2; ++k)
-    {
-      x = H.A[k + 1][k];
-      z = H.A[k + 2][k];
-      Givens_calc<T>(x, z, c, s);
-      Givens_mult<T>(H, k + 1, k + 2, c, -s, 0);
-      Givens_mult<T>(H, k + 1, k + 2, c,  s, 1);
-      Givens_mult<T>(P, k + 1, k + 2, c,  s, 1);
-    }
-    it++;
-    tot_it++;
-  }while(N > 1);
-
-  N = evals.size();
-  ///Annoying - UT solves in reverse order;
-  DenseVector<T> tmp(N);
-  for(int i = 0; i < N; ++i)
-    tmp[i] = evals[N-i-1];
-  evals = tmp;
-  //
-  UTeigenvectors(H, trows, evals, evecs);
-  //UTSymmEigenvectors(H, trows, evals, evecs);
-  for(int i = 0; i < evals.size(); ++i)
-  {
-    evecs[i] = P * evecs[i];
-    normalize(evecs[i]);
-    evals[i] = evals[i] * Hnorm;
-  }
-  // // FIXME this is to test
-  // Hin.write("evecs3", evecs);
-  // Hin.write("evals3", evals);
-  // // check rsd
-  // for(int i = 0; i < M; i++) {
-  //   vector<T> Aevec = Hin * evecs[i];
-  //   RealD norm2(0.);
-  //   for(int j = 0; j < M; j++) {
-  //     norm2 += (Aevec[j] - evals[i] * evecs[i][j]) * (Aevec[j] - evals[i] * evecs[i][j]);
-  //   }
-  // }
-  return tot_it;
-}
-
-template <class T>
-void Hess(DenseMatrix<T > &A, DenseMatrix<T> &Q, int start){
-
-  /**
-  turn a matrix A =
-  x  x  x  x  x
-  x  x  x  x  x
-  x  x  x  x  x
-  x  x  x  x  x
-  x  x  x  x  x
-  into
-  x  x  x  x  x
-  x  x  x  x  x
-  0  x  x  x  x
-  0  0  x  x  x
-  0  0  0  x  x
-  with householder rotations
-  Slow.
-  */
-  int N ; SizeSquare(A,N);
-  DenseVector<T > p; Resize(p,N); Fill(p,0);
-
-  for(int k=start;k<N-2;k++){
-    //cerr << "hess" << k << std::endl;
-    DenseVector<T > ck,v; Resize(ck,N-k-1); Resize(v,N-k-1);
-    for(int i=k+1;i<N;i++){ck[i-k-1] = A(i,k);}  ///kth column
-    normalize(ck);    ///Normalization cancels in PHP anyway
-    T beta;
-    Householder_vector<T >(ck, 0, ck.size()-1, v, beta);  ///Householder vector
-    Householder_mult<T>(A,v,beta,start,k+1,N-1,0);  ///A -> PA
-    Householder_mult<T >(A,v,beta,start,k+1,N-1,1);  ///PA -> PAP^H
-    ///Accumulate eigenvector
-    Householder_mult<T >(Q,v,beta,start,k+1,N-1,1);  ///Q -> QP^H
-  }
-  /*for(int l=0;l<N-2;l++){
-    for(int k=l+2;k<N;k++){
-    A(0,k,l);
-    }
-    }*/
-}
-
-template <class T>
-void Tri(DenseMatrix<T > &A, DenseMatrix<T> &Q, int start){
-///Tridiagonalize a matrix
-  int N; SizeSquare(A,N);
-  Hess(A,Q,start);
-  /*for(int l=0;l<N-2;l++){
-    for(int k=l+2;k<N;k++){
-    A(0,l,k);
-    }
-    }*/
-}
-
-template <class T>
-void ForceTridiagonal(DenseMatrix<T> &A){
-///Tridiagonalize a matrix
-  int N ; SizeSquare(A,N);
-  for(int l=0;l<N-2;l++){
-    for(int k=l+2;k<N;k++){
-      A[l][k]=0;
-      A[k][l]=0;
-    }
-  }
-}
-
-template <class T>
-int my_SymmEigensystem(DenseMatrix<T > &Ain, DenseVector<T> &evals, DenseVector<DenseVector<T> > &evecs, RealD small){
-  ///Solve a symmetric eigensystem, not necessarily in tridiagonal form
-  int N; SizeSquare(Ain,N);
-  DenseMatrix<T > A; A = Ain;
-  DenseMatrix<T > Q; Resize(Q,N,N); Unity(Q);
-  Tri(A,Q,0);
-  int it = my_Wilkinson<T>(A, evals, evecs, small);
-  for(int k=0;k<N;k++){evecs[k] = Q*evecs[k];}
-  return it;
-}
-
-
-template <class T>
-int Wilkinson(DenseMatrix<T> &Ain, DenseVector<T> &evals, DenseVector<DenseVector<T> > &evecs, RealD small){
-  return my_Wilkinson(Ain, evals, evecs, small);
-}
-
-template <class T>
-int SymmEigensystem(DenseMatrix<T> &Ain, DenseVector<T> &evals, DenseVector<DenseVector<T> > &evecs, RealD small){
-  return my_SymmEigensystem(Ain, evals, evecs, small);
-}
-
-template <class T>
-int Eigensystem(DenseMatrix<T > &Ain, DenseVector<T> &evals, DenseVector<DenseVector<T> > &evecs, RealD small){
-///Solve a general eigensystem, not necessarily in tridiagonal form
-  int N = Ain.dim;
-  DenseMatrix<T > A(N); A = Ain;
-  DenseMatrix<T > Q(N);Q.Unity();
-  Hess(A,Q,0);
-  int it = QReigensystem<T>(A, evals, evecs, small);
-  for(int k=0;k<N;k++){evecs[k] = Q*evecs[k];}
-  return it;
-}
-
-}
-#endif

lib/algorithms/densematrix/Householder.h

@@ -1,242 +0,0 @@
-    /*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/algorithms/iterative/Householder.h
-
-    Copyright (C) 2015
-
-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 */
-#ifndef HOUSEHOLDER_H
-#define HOUSEHOLDER_H
-
-#define TIMER(A) std::cout << GridLogMessage << __FUNC__ << " file "<< __FILE__ <<" line " << __LINE__ << std::endl;
-#define ENTER()  std::cout << GridLogMessage << "ENTRY "<<__FUNC__ << " file "<< __FILE__ <<" line " << __LINE__ << std::endl;
-#define LEAVE()  std::cout << GridLogMessage << "EXIT  "<<__FUNC__ << " file "<< __FILE__ <<" line " << __LINE__ << std::endl;
-
-#include <cstdlib>
-#include <string>
-#include <cmath>
-#include <iostream>
-#include <sstream>
-#include <stdexcept>
-#include <fstream>
-#include <complex>
-#include <algorithm>
-
-namespace Grid {
-/** Comparison function for finding the max element in a vector **/
-template <class T> bool cf(T i, T j) { 
-  return abs(i) < abs(j); 
-}
-
-/** 
-	Calculate a real Givens angle 
- **/
-template <class T> inline void Givens_calc(T y, T z, T &c, T &s){
-
-  RealD mz = (RealD)abs(z);
-  
-  if(mz==0.0){
-    c = 1; s = 0;
-  }
-  if(mz >= (RealD)abs(y)){
-    T t = -y/z;
-    s = (T)1.0 / sqrt ((T)1.0 + t * t);
-    c = s * t;
-  } else {
-    T t = -z/y;
-    c = (T)1.0 / sqrt ((T)1.0 + t * t);
-    s = c * t;
-  }
-}
-
-template <class T> inline void Givens_mult(DenseMatrix<T> &A,  int i, int k, T c, T s, int dir)
-{
-  int q ; SizeSquare(A,q);
-
-  if(dir == 0){
-    for(int j=0;j<q;j++){
-      T nu = A[i][j];
-      T w  = A[k][j];
-      A[i][j] = (c*nu + s*w);
-      A[k][j] = (-s*nu + c*w);
-    }
-  }
-
-  if(dir == 1){
-    for(int j=0;j<q;j++){
-      T nu = A[j][i];
-      T w  = A[j][k];
-      A[j][i] = (c*nu - s*w);
-      A[j][k] = (s*nu + c*w);
-    }
-  }
-}
-
-/**
-	from input = x;
-	Compute the complex Householder vector, v, such that
-	P = (I - b v transpose(v) )
-	b = 2/v.v
-
-	P | x |    | x | k = 0
-	| x |    | 0 | 
-	| x | =  | 0 |
-	| x |    | 0 | j = 3
-	| x |	   | x |
-
-	These are the "Unreduced" Householder vectors.
-
- **/
-template <class T> inline void Householder_vector(DenseVector<T> input, int k, int j, DenseVector<T> &v, T &beta)
-{
-  int N ; Size(input,N);
-  T m = *max_element(input.begin() + k, input.begin() + j + 1, cf<T> );
-
-  if(abs(m) > 0.0){
-    T alpha = 0;
-
-    for(int i=k; i<j+1; i++){
-      v[i] = input[i]/m;
-      alpha = alpha + v[i]*conj(v[i]);
-    }
-    alpha = sqrt(alpha);
-    beta = (T)1.0/(alpha*(alpha + abs(v[k]) ));
-
-    if(abs(v[k]) > 0.0)  v[k] = v[k] + (v[k]/abs(v[k]))*alpha;
-    else                 v[k] = -alpha;
-  } else{
-    for(int i=k; i<j+1; i++){
-      v[i] = 0.0;
-    } 
-  }
-}
-
-/**
-	from input = x;
-	Compute the complex Householder vector, v, such that
-	P = (I - b v transpose(v) )
-	b = 2/v.v
-
-	Px = alpha*e_dir
-
-	These are the "Unreduced" Householder vectors.
-
- **/
-
-template <class T> inline void Householder_vector(DenseVector<T> input, int k, int j, int dir, DenseVector<T> &v, T &beta)
-{
-  int N = input.size();
-  T m = *max_element(input.begin() + k, input.begin() + j + 1, cf);
-  
-  if(abs(m) > 0.0){
-    T alpha = 0;
-
-    for(int i=k; i<j+1; i++){
-      v[i] = input[i]/m;
-      alpha = alpha + v[i]*conj(v[i]);
-    }
-    
-    alpha = sqrt(alpha);
-    beta = 1.0/(alpha*(alpha + abs(v[dir]) ));
-	
-    if(abs(v[dir]) > 0.0) v[dir] = v[dir] + (v[dir]/abs(v[dir]))*alpha;
-    else                  v[dir] = -alpha;
-  }else{
-    for(int i=k; i<j+1; i++){
-      v[i] = 0.0;
-    } 
-  }
-}
-
-/**
-	Compute the product PA if trans = 0
-	AP if trans = 1
-	P = (I - b v transpose(v) )
-	b = 2/v.v
-	start at element l of matrix A
-	v is of length j - k + 1 of v are nonzero
- **/
-
-template <class T> inline void Householder_mult(DenseMatrix<T> &A , DenseVector<T> v, T beta, int l, int k, int j, int trans)
-{
-  int N ; SizeSquare(A,N);
-
-  if(abs(beta) > 0.0){
-    for(int p=l; p<N; p++){
-      T s = 0;
-      if(trans==0){
-	for(int i=k;i<j+1;i++) s += conj(v[i-k])*A[i][p];
-	s *= beta;
-	for(int i=k;i<j+1;i++){ A[i][p] = A[i][p]-s*conj(v[i-k]);}
-      } else {
-	for(int i=k;i<j+1;i++){ s += conj(v[i-k])*A[p][i];}
-	s *= beta;
-	for(int i=k;i<j+1;i++){ A[p][i]=A[p][i]-s*conj(v[i-k]);}
-      }
-    }
-  }
-}
-
-/**
-	Compute the product PA if trans = 0
-	AP if trans = 1
-	P = (I - b v transpose(v) )
-	b = 2/v.v
-	start at element l of matrix A
-	v is of length j - k + 1 of v are nonzero
-	A is tridiagonal
- **/
-template <class T> inline void Householder_mult_tri(DenseMatrix<T> &A , DenseVector<T> v, T beta, int l, int M, int k, int j, int trans)
-{
-  if(abs(beta) > 0.0){
-
-    int N ; SizeSquare(A,N);
-
-    DenseMatrix<T> tmp; Resize(tmp,N,N); Fill(tmp,0); 
-
-    T s;
-    for(int p=l; p<M; p++){
-      s = 0;
-      if(trans==0){
-	for(int i=k;i<j+1;i++) s = s + conj(v[i-k])*A[i][p];
-      }else{
-	for(int i=k;i<j+1;i++) s = s + v[i-k]*A[p][i];
-      }
-      s = beta*s;
-      if(trans==0){
-	for(int i=k;i<j+1;i++) tmp[i][p] = tmp(i,p) - s*v[i-k];
-      }else{
-	for(int i=k;i<j+1;i++) tmp[p][i] = tmp[p][i] - s*conj(v[i-k]);
-      }
-    }
-    for(int p=l; p<M; p++){
-      if(trans==0){
-	for(int i=k;i<j+1;i++) A[i][p] = A[i][p] + tmp[i][p];
-      }else{
-	for(int i=k;i<j+1;i++) A[p][i] = A[p][i] + tmp[p][i];
-      }
-    }
-  }
-}
-}
-#endif

lib/algorithms/iterative/BlockConjugateGradient.h

@@ -33,6 +33,8 @@ directory
 
 namespace Grid {
 
+enum BlockCGtype { BlockCG, BlockCGrQ, CGmultiRHS };
+
 //////////////////////////////////////////////////////////////////////////
 // Block conjugate gradient. Dimension zero should be the block direction
 //////////////////////////////////////////////////////////////////////////
@@ -40,25 +42,273 @@ template <class Field>
 class BlockConjugateGradient : public OperatorFunction<Field> {
  public:
 
+
   typedef typename Field::scalar_type scomplex;
 
-  const int blockDim = 0;
-
+  int blockDim ;
   int Nblock;
+
+  BlockCGtype CGtype;
   bool ErrorOnNoConverge;  // throw an assert when the CG fails to converge.
                            // Defaults true.
   RealD Tolerance;
   Integer MaxIterations;
   Integer IterationsToComplete; //Number of iterations the CG took to finish. Filled in upon completion
   
-  BlockConjugateGradient(RealD tol, Integer maxit, bool err_on_no_conv = true)
-    : Tolerance(tol),
-    MaxIterations(maxit),
-    ErrorOnNoConverge(err_on_no_conv){};
+  BlockConjugateGradient(BlockCGtype cgtype,int _Orthog,RealD tol, Integer maxit, bool err_on_no_conv = true)
+    : Tolerance(tol), CGtype(cgtype),   blockDim(_Orthog),  MaxIterations(maxit), ErrorOnNoConverge(err_on_no_conv)
+  {};
 
+////////////////////////////////////////////////////////////////////////////////////////////////////
+// Thin QR factorisation (google it)
+////////////////////////////////////////////////////////////////////////////////////////////////////
+void ThinQRfact (Eigen::MatrixXcd &m_rr,
+		 Eigen::MatrixXcd &C,
+		 Eigen::MatrixXcd &Cinv,
+		 Field & Q,
+		 const Field & R)
+{
+  int Orthog = blockDim; // First dimension is block dim; this is an assumption
+  ////////////////////////////////////////////////////////////////////////////////////////////////////
+  //Dimensions
+  // R_{ferm x Nblock} =  Q_{ferm x Nblock} x  C_{Nblock x Nblock} -> ferm x Nblock
+  //
+  // Rdag R = m_rr = Herm = L L^dag        <-- Cholesky decomposition (LLT routine in Eigen)
+  //
+  //   Q  C = R => Q = R C^{-1}
+  //
+  // Want  Ident = Q^dag Q = C^{-dag} R^dag R C^{-1} = C^{-dag} L L^dag C^{-1} = 1_{Nblock x Nblock} 
+  //
+  // Set C = L^{dag}, and then Q^dag Q = ident 
+  //
+  // Checks:
+  // Cdag C = Rdag R ; passes.
+  // QdagQ  = 1      ; passes
+  ////////////////////////////////////////////////////////////////////////////////////////////////////
+  sliceInnerProductMatrix(m_rr,R,R,Orthog);
+
+  ////////////////////////////////////////////////////////////////////////////////////////////////////
+  // Cholesky from Eigen
+  // There exists a ldlt that is documented as more stable
+  ////////////////////////////////////////////////////////////////////////////////////////////////////
+  Eigen::MatrixXcd L    = m_rr.llt().matrixL(); 
+
+  C    = L.adjoint();
+  Cinv = C.inverse();
+
+  ////////////////////////////////////////////////////////////////////////////////////////////////////
+  // Q = R C^{-1}
+  //
+  // Q_j  = R_i Cinv(i,j) 
+  //
+  // NB maddMatrix conventions are Right multiplication X[j] a[j,i] already
+  ////////////////////////////////////////////////////////////////////////////////////////////////////
+  // FIXME:: make a sliceMulMatrix to avoid zero vector
+  sliceMulMatrix(Q,Cinv,R,Orthog);
+}
+////////////////////////////////////////////////////////////////////////////////////////////////////
+// Call one of several implementations
+////////////////////////////////////////////////////////////////////////////////////////////////////
 void operator()(LinearOperatorBase<Field> &Linop, const Field &Src, Field &Psi) 
 {
-  int Orthog = 0; // First dimension is block dim
+  if ( CGtype == BlockCGrQ ) {
+    BlockCGrQsolve(Linop,Src,Psi);
+  } else if (CGtype == BlockCG ) {
+    BlockCGsolve(Linop,Src,Psi);
+  } else if (CGtype == CGmultiRHS ) {
+    CGmultiRHSsolve(Linop,Src,Psi);
+  } else {
+    assert(0);
+  }
+}
+
+////////////////////////////////////////////////////////////////////////////
+// BlockCGrQ implementation:
+//--------------------------
+// X is guess/Solution
+// B is RHS
+// Solve A X_i = B_i    ;        i refers to Nblock index
+////////////////////////////////////////////////////////////////////////////
+void BlockCGrQsolve(LinearOperatorBase<Field> &Linop, const Field &B, Field &X) 
+{
+  int Orthog = blockDim; // First dimension is block dim; this is an assumption
+  Nblock = B._grid->_fdimensions[Orthog];
+
+  std::cout<<GridLogMessage<<" Block Conjugate Gradient : Orthog "<<Orthog<<" Nblock "<<Nblock<<std::endl;
+
+  X.checkerboard = B.checkerboard;
+  conformable(X, B);
+
+  Field tmp(B);
+  Field Q(B);
+  Field D(B);
+  Field Z(B);
+  Field AD(B);
+
+  Eigen::MatrixXcd m_DZ     = Eigen::MatrixXcd::Identity(Nblock,Nblock);
+  Eigen::MatrixXcd m_M      = Eigen::MatrixXcd::Identity(Nblock,Nblock);
+  Eigen::MatrixXcd m_rr     = Eigen::MatrixXcd::Zero(Nblock,Nblock);
+
+  Eigen::MatrixXcd m_C      = Eigen::MatrixXcd::Zero(Nblock,Nblock);
+  Eigen::MatrixXcd m_Cinv   = Eigen::MatrixXcd::Zero(Nblock,Nblock);
+  Eigen::MatrixXcd m_S      = Eigen::MatrixXcd::Zero(Nblock,Nblock);
+  Eigen::MatrixXcd m_Sinv   = Eigen::MatrixXcd::Zero(Nblock,Nblock);
+
+  Eigen::MatrixXcd m_tmp    = Eigen::MatrixXcd::Identity(Nblock,Nblock);
+  Eigen::MatrixXcd m_tmp1   = Eigen::MatrixXcd::Identity(Nblock,Nblock);
+
+  // Initial residual computation & set up
+  std::vector<RealD> residuals(Nblock);
+  std::vector<RealD> ssq(Nblock);
+
+  sliceNorm(ssq,B,Orthog);
+  RealD sssum=0;
+  for(int b=0;b<Nblock;b++) sssum+=ssq[b];
+
+  sliceNorm(residuals,B,Orthog);
+  for(int b=0;b<Nblock;b++){ assert(std::isnan(residuals[b])==0); }
+
+  sliceNorm(residuals,X,Orthog);
+  for(int b=0;b<Nblock;b++){ assert(std::isnan(residuals[b])==0); }
+
+  /************************************************************************
+   * Block conjugate gradient rQ (Sebastien Birk Thesis, after Dubrulle 2001)
+   ************************************************************************
+   * Dimensions:
+   *
+   *   X,B==(Nferm x Nblock)
+   *   A==(Nferm x Nferm)
+   *  
+   * Nferm = Nspin x Ncolour x Ncomplex x Nlattice_site
+   * 
+   * QC = R = B-AX, D = Q     ; QC => Thin QR factorisation (google it)
+   * for k: 
+   *   Z  = AD
+   *   M  = [D^dag Z]^{-1}
+   *   X  = X + D MC
+   *   QS = Q - ZM
+   *   D  = Q + D S^dag
+   *   C  = S C
+   */
+  ///////////////////////////////////////
+  // Initial block: initial search dir is guess
+  ///////////////////////////////////////
+  std::cout << GridLogMessage<<"BlockCGrQ algorithm initialisation " <<std::endl;
+
+  //1.  QC = R = B-AX, D = Q     ; QC => Thin QR factorisation (google it)
+
+  Linop.HermOp(X, AD);
+  tmp = B - AD;  
+  ThinQRfact (m_rr, m_C, m_Cinv, Q, tmp);
+  D=Q;
+
+  std::cout << GridLogMessage<<"BlockCGrQ computed initial residual and QR fact " <<std::endl;
+
+  ///////////////////////////////////////
+  // Timers
+  ///////////////////////////////////////
+  GridStopWatch sliceInnerTimer;
+  GridStopWatch sliceMaddTimer;
+  GridStopWatch QRTimer;
+  GridStopWatch MatrixTimer;
+  GridStopWatch SolverTimer;
+  SolverTimer.Start();
+
+  int k;
+  for (k = 1; k <= MaxIterations; k++) {
+
+    //3. Z  = AD
+    MatrixTimer.Start();
+    Linop.HermOp(D, Z);      
+    MatrixTimer.Stop();
+
+    //4. M  = [D^dag Z]^{-1}
+    sliceInnerTimer.Start();
+    sliceInnerProductMatrix(m_DZ,D,Z,Orthog);
+    sliceInnerTimer.Stop();
+    m_M       = m_DZ.inverse();
+
+    //5. X  = X + D MC
+    m_tmp     = m_M * m_C;
+    sliceMaddTimer.Start();
+    sliceMaddMatrix(X,m_tmp, D,X,Orthog);     
+    sliceMaddTimer.Stop();
+
+    //6. QS = Q - ZM
+    sliceMaddTimer.Start();
+    sliceMaddMatrix(tmp,m_M,Z,Q,Orthog,-1.0);
+    sliceMaddTimer.Stop();
+    QRTimer.Start();
+    ThinQRfact (m_rr, m_S, m_Sinv, Q, tmp);
+    QRTimer.Stop();
+    
+    //7. D  = Q + D S^dag
+    m_tmp = m_S.adjoint();
+    sliceMaddTimer.Start();
+    sliceMaddMatrix(D,m_tmp,D,Q,Orthog);
+    sliceMaddTimer.Stop();
+
+    //8. C  = S C
+    m_C = m_S*m_C;
+    
+    /*********************
+     * convergence monitor
+     *********************
+     */
+    m_rr = m_C.adjoint() * m_C;
+
+    RealD max_resid=0;
+    RealD rrsum=0;
+    RealD rr;
+
+    for(int b=0;b<Nblock;b++) {
+      rrsum+=real(m_rr(b,b));
+      rr = real(m_rr(b,b))/ssq[b];
+      if ( rr > max_resid ) max_resid = rr;
+    }
+
+    std::cout << GridLogIterative << "\titeration "<<k<<" rr_sum "<<rrsum<<" ssq_sum "<< sssum
+	      <<" ave "<<std::sqrt(rrsum/sssum) << " max "<< max_resid <<std::endl;
+
+    if ( max_resid < Tolerance*Tolerance ) { 
+
+      SolverTimer.Stop();
+
+      std::cout << GridLogMessage<<"BlockCGrQ converged in "<<k<<" iterations"<<std::endl;
+
+      for(int b=0;b<Nblock;b++){
+	std::cout << GridLogMessage<< "\t\tblock "<<b<<" computed resid "
+		  << std::sqrt(real(m_rr(b,b))/ssq[b])<<std::endl;
+      }
+      std::cout << GridLogMessage<<"\tMax residual is "<<std::sqrt(max_resid)<<std::endl;
+
+      Linop.HermOp(X, AD);
+      AD = AD-B;
+      std::cout << GridLogMessage <<"\t True residual is " << std::sqrt(norm2(AD)/norm2(B)) <<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 << "\tInnerProd  " << sliceInnerTimer.Elapsed() <<std::endl;
+      std::cout << GridLogMessage << "\tMaddMatrix " << sliceMaddTimer.Elapsed()  <<std::endl;
+      std::cout << GridLogMessage << "\tThinQRfact " << QRTimer.Elapsed()  <<std::endl;
+	    
+      IterationsToComplete = k;
+      return;
+    }
+
+  }
+  std::cout << GridLogMessage << "BlockConjugateGradient(rQ) did NOT converge" << std::endl;
+
+  if (ErrorOnNoConverge) assert(0);
+  IterationsToComplete = k;
+}
+//////////////////////////////////////////////////////////////////////////
+// Block conjugate gradient; Original O'Leary Dimension zero should be the block direction
+//////////////////////////////////////////////////////////////////////////
+void BlockCGsolve(LinearOperatorBase<Field> &Linop, const Field &Src, Field &Psi) 
+{
+  int Orthog = blockDim; // First dimension is block dim; this is an assumption
   Nblock = Src._grid->_fdimensions[Orthog];
 
   std::cout<<GridLogMessage<<" Block Conjugate Gradient : Orthog "<<Orthog<<" Nblock "<<Nblock<<std::endl;
@@ -162,8 +412,9 @@ void operator()(LinearOperatorBase<Field> &Linop, const Field &Src, Field &Psi)
      *********************
      */
     RealD max_resid=0;
+    RealD rr;
     for(int b=0;b<Nblock;b++){
-      RealD rr = real(m_rr(b,b))/ssq[b];
+      rr = real(m_rr(b,b))/ssq[b];
       if ( rr > max_resid ) max_resid = rr;
     }
     
@@ -173,13 +424,14 @@ void operator()(LinearOperatorBase<Field> &Linop, const Field &Src, Field &Psi)
 
       std::cout << GridLogMessage<<"BlockCG converged in "<<k<<" iterations"<<std::endl;
       for(int b=0;b<Nblock;b++){
-	std::cout << GridLogMessage<< "\t\tblock "<<b<<" resid "<< std::sqrt(real(m_rr(b,b))/ssq[b])<<std::endl;
+	std::cout << GridLogMessage<< "\t\tblock "<<b<<" computed resid "
+		  << std::sqrt(real(m_rr(b,b))/ssq[b])<<std::endl;
       }
       std::cout << GridLogMessage<<"\tMax residual is "<<std::sqrt(max_resid)<<std::endl;
 
       Linop.HermOp(Psi, AP);
       AP = AP-Src;
-      std::cout << GridLogMessage <<"\tTrue residual is " << std::sqrt(norm2(AP)/norm2(Src)) <<std::endl;
+      std::cout << GridLogMessage <<"\t True residual is " << std::sqrt(norm2(AP)/norm2(Src)) <<std::endl;
 
       std::cout << GridLogMessage << "Time Breakdown "<<std::endl;
       std::cout << GridLogMessage << "\tElapsed    " << SolverTimer.Elapsed()     <<std::endl;
@@ -197,35 +449,13 @@ void operator()(LinearOperatorBase<Field> &Linop, const Field &Src, Field &Psi)
   if (ErrorOnNoConverge) assert(0);
   IterationsToComplete = k;
 }
-};
-
-
 //////////////////////////////////////////////////////////////////////////
 // multiRHS conjugate gradient. Dimension zero should be the block direction
+// Use this for spread out across nodes
 //////////////////////////////////////////////////////////////////////////
-template <class Field>
-class MultiRHSConjugateGradient : public OperatorFunction<Field> {
- public:
-
-  typedef typename Field::scalar_type scomplex;
-
-  const int blockDim = 0;
-
-  int Nblock;
-  bool ErrorOnNoConverge;  // throw an assert when the CG fails to converge.
-                           // Defaults true.
-  RealD Tolerance;
-  Integer MaxIterations;
-  Integer IterationsToComplete; //Number of iterations the CG took to finish. Filled in upon completion
-  
-   MultiRHSConjugateGradient(RealD tol, Integer maxit, bool err_on_no_conv = true)
-    : Tolerance(tol),
-    MaxIterations(maxit),
-    ErrorOnNoConverge(err_on_no_conv){};
-
-void operator()(LinearOperatorBase<Field> &Linop, const Field &Src, Field &Psi) 
+void CGmultiRHSsolve(LinearOperatorBase<Field> &Linop, const Field &Src, Field &Psi) 
 {
-  int Orthog = 0; // First dimension is block dim
+  int Orthog = blockDim; // First dimension is block dim
   Nblock = Src._grid->_fdimensions[Orthog];
 
   std::cout<<GridLogMessage<<"MultiRHS Conjugate Gradient : Orthog "<<Orthog<<" Nblock "<<Nblock<<std::endl;
@@ -285,12 +515,10 @@ void operator()(LinearOperatorBase<Field> &Linop, const Field &Src, Field &Psi)
     MatrixTimer.Stop();
 
     // Alpha
-    //    sliceInnerProductVectorTest(v_pAp_test,P,AP,Orthog);
     sliceInnerTimer.Start();
     sliceInnerProductVector(v_pAp,P,AP,Orthog);
     sliceInnerTimer.Stop();
     for(int b=0;b<Nblock;b++){
-      //      std::cout << " "<< v_pAp[b]<<" "<< v_pAp_test[b]<<std::endl;
       v_alpha[b] = v_rr[b]/real(v_pAp[b]);
     }
 
@@ -332,7 +560,7 @@ void operator()(LinearOperatorBase<Field> &Linop, const Field &Src, Field &Psi)
 
       std::cout << GridLogMessage<<"MultiRHS solver converged in " <<k<<" iterations"<<std::endl;
       for(int b=0;b<Nblock;b++){
-	std::cout << GridLogMessage<< "\t\tBlock "<<b<<" resid "<< std::sqrt(v_rr[b]/ssq[b])<<std::endl;
+	std::cout << GridLogMessage<< "\t\tBlock "<<b<<" computed resid "<< std::sqrt(v_rr[b]/ssq[b])<<std::endl;
       }
       std::cout << GridLogMessage<<"\tMax residual is "<<std::sqrt(max_resid)<<std::endl;
 
@@ -358,9 +586,8 @@ void operator()(LinearOperatorBase<Field> &Linop, const Field &Src, Field &Psi)
   if (ErrorOnNoConverge) assert(0);
   IterationsToComplete = k;
 }
+
 };
 
-
-
 }
 #endif

lib/algorithms/iterative/EigenSort.h

@@ -1,81 +0,0 @@
-    /*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./lib/algorithms/iterative/EigenSort.h
-
-    Copyright (C) 2015
-
-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 */
-#ifndef GRID_EIGENSORT_H
-#define GRID_EIGENSORT_H
-
-
-namespace Grid {
-    /////////////////////////////////////////////////////////////
-    // Eigen sorter to begin with
-    /////////////////////////////////////////////////////////////
-
-template<class Field>
-class SortEigen {
- private:
-  
-//hacking for testing for now
- private:
-  static bool less_lmd(RealD left,RealD right){
-    return left > right;
-  }  
-  static bool less_pair(std::pair<RealD,Field const*>& left,
-                        std::pair<RealD,Field const*>& right){
-    return left.first > (right.first);
-  }  
-  
-  
- public:
-
-  void push(DenseVector<RealD>& lmd,
-            DenseVector<Field>& evec,int N) {
-    DenseVector<Field> cpy(lmd.size(),evec[0]._grid);
-    for(int i=0;i<lmd.size();i++) cpy[i] = evec[i];
-    
-    DenseVector<std::pair<RealD, Field const*> > emod(lmd.size());    
-    for(int i=0;i<lmd.size();++i)
-      emod[i] = std::pair<RealD,Field const*>(lmd[i],&cpy[i]);
-
-    partial_sort(emod.begin(),emod.begin()+N,emod.end(),less_pair);
-
-    typename DenseVector<std::pair<RealD, Field const*> >::iterator it = emod.begin();
-    for(int i=0;i<N;++i){
-      lmd[i]=it->first;
-      evec[i]=*(it->second);
-      ++it;
-    }
-  }
-  void push(DenseVector<RealD>& lmd,int N) {
-    std::partial_sort(lmd.begin(),lmd.begin()+N,lmd.end(),less_lmd);
-  }
-  bool saturated(RealD lmd, RealD thrs) {
-    return fabs(lmd) > fabs(thrs);
-  }
-};
-
-}
-#endif

lib/cartesian/Cartesian_base.h

@@ -50,7 +50,6 @@ public:
 
     GridBase(const std::vector<int> & processor_grid) : CartesianCommunicator(processor_grid) {};
 
-
     // Physics Grid information.
     std::vector<int> _simd_layout;// Which dimensions get relayed out over simd lanes.
     std::vector<int> _fdimensions;// (full) Global dimensions of array prior to cb removal
@@ -63,13 +62,12 @@ public:
     int _isites;
     int _fsites;                  // _isites*_osites = product(dimensions).
     int _gsites;
-    std::vector<int> _slice_block;   // subslice information
+    std::vector<int> _slice_block;// subslice information
     std::vector<int> _slice_stride;
     std::vector<int> _slice_nblock;
 
-    // Might need these at some point
-    //    std::vector<int> _lstart;     // local start of array in gcoors. _processor_coor[d]*_ldimensions[d]
-    //    std::vector<int> _lend;       // local end of array in gcoors    _processor_coor[d]*_ldimensions[d]+_ldimensions_[d]-1
+    std::vector<int> _lstart;     // local start of array in gcoors _processor_coor[d]*_ldimensions[d]
+    std::vector<int> _lend  ;     // local end of array in gcoors   _processor_coor[d]*_ldimensions[d]+_ldimensions_[d]-1
 
 public:
 
@@ -176,6 +174,7 @@ public:
     inline int gSites(void) const { return _isites*_osites*_Nprocessors; }; 
     inline int Nd    (void) const { return _ndimension;};
 
+    inline const std::vector<int> LocalStarts(void)             { return _lstart;    };
     inline const std::vector<int> &FullDimensions(void)         { return _fdimensions;};
     inline const std::vector<int> &GlobalDimensions(void)       { return _gdimensions;};
     inline const std::vector<int> &LocalDimensions(void)        { return _ldimensions;};

lib/cartesian/Cartesian_full.h

@@ -76,6 +76,8 @@ public:
         _ldimensions.resize(_ndimension);
         _rdimensions.resize(_ndimension);
         _simd_layout.resize(_ndimension);
+	_lstart.resize(_ndimension);
+	_lend.resize(_ndimension);
             
         _ostride.resize(_ndimension);
         _istride.resize(_ndimension);
@@ -94,8 +96,10 @@ public:
 	  // Use a reduced simd grid
 	  _ldimensions[d]= _gdimensions[d]/_processors[d];  //local dimensions
 	  _rdimensions[d]= _ldimensions[d]/_simd_layout[d]; //overdecomposition
-	  _osites *= _rdimensions[d];
-	  _isites *= _simd_layout[d];
+	  _lstart[d]     = _processor_coor[d]*_ldimensions[d];
+	  _lend[d]       = _processor_coor[d]*_ldimensions[d]+_ldimensions[d]-1;
+	  _osites  *= _rdimensions[d];
+	  _isites  *= _simd_layout[d];
                 
 	  // Addressing support
 	  if ( d==0 ) {

lib/cartesian/Cartesian_red_black.h

@@ -151,6 +151,8 @@ public:
       _ldimensions.resize(_ndimension);
       _rdimensions.resize(_ndimension);
       _simd_layout.resize(_ndimension);
+      _lstart.resize(_ndimension);
+      _lend.resize(_ndimension);
       
       _ostride.resize(_ndimension);
       _istride.resize(_ndimension);
@@ -169,6 +171,8 @@ public:
 	  _gdimensions[d] = _gdimensions[d]/2; // Remove a checkerboard
 	}
 	_ldimensions[d] = _gdimensions[d]/_processors[d];
+	_lstart[d]     = _processor_coor[d]*_ldimensions[d];
+	_lend[d]       = _processor_coor[d]*_ldimensions[d]+_ldimensions[d]-1;
 
 	// Use a reduced simd grid
 	_simd_layout[d] = simd_layout[d];

lib/communicator/Communicator_base.cc

@@ -60,6 +60,7 @@ void CartesianCommunicator::ShmBufferFreeAll(void) {
 /////////////////////////////////
 // Grid information queries
 /////////////////////////////////
+int                      CartesianCommunicator::Dimensions(void)         { return _ndimension; };
 int                      CartesianCommunicator::IsBoss(void)            { return _processor==0; };
 int                      CartesianCommunicator::BossRank(void)          { return 0; };
 int                      CartesianCommunicator::ThisRank(void)          { return _processor; };
@@ -91,6 +92,7 @@ void CartesianCommunicator::GlobalSumVector(ComplexD *c,int N)
 #if !defined( GRID_COMMS_MPI3) && !defined (GRID_COMMS_MPI3L)
 
 int                      CartesianCommunicator::NodeCount(void)    { return ProcessorCount();};
+int                      CartesianCommunicator::RankCount(void)    { return ProcessorCount();};
 
 double CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsRequest_t> &list,
 						       void *xmit,

lib/communicator/Communicator_base.h

@@ -148,6 +148,7 @@ class CartesianCommunicator {
   int  RankFromProcessorCoor(std::vector<int> &coor);
   void ProcessorCoorFromRank(int rank,std::vector<int> &coor);
   
+  int                      Dimensions(void)        ;
   int                      IsBoss(void)            ;
   int                      BossRank(void)          ;
   int                      ThisRank(void)          ;
@@ -155,6 +156,7 @@ class CartesianCommunicator {
   const std::vector<int> & ProcessorGrid(void)     ;
   int                      ProcessorCount(void)    ;
   int                      NodeCount(void)    ;
+  int                      RankCount(void)    ;
 
   ////////////////////////////////////////////////////////////////////////////////
   // very VERY rarely (Log, serial RNG) we need world without a grid
@@ -175,6 +177,8 @@ class CartesianCommunicator {
   void GlobalSumVector(ComplexF *c,int N);
   void GlobalSum(ComplexD &c);
   void GlobalSumVector(ComplexD *c,int N);
+  void GlobalXOR(uint32_t &);
+  void GlobalXOR(uint64_t &);
   
   template<class obj> void GlobalSum(obj &o){
     typedef typename obj::scalar_type scalar_type;

lib/communicator/Communicator_mpi.cc

@@ -83,6 +83,14 @@ void CartesianCommunicator::GlobalSum(uint64_t &u){
   int ierr=MPI_Allreduce(MPI_IN_PLACE,&u,1,MPI_UINT64_T,MPI_SUM,communicator);
   assert(ierr==0);
 }
+void CartesianCommunicator::GlobalXOR(uint32_t &u){
+  int ierr=MPI_Allreduce(MPI_IN_PLACE,&u,1,MPI_UINT32_T,MPI_BXOR,communicator);
+  assert(ierr==0);
+}
+void CartesianCommunicator::GlobalXOR(uint64_t &u){
+  int ierr=MPI_Allreduce(MPI_IN_PLACE,&u,1,MPI_UINT64_T,MPI_BXOR,communicator);
+  assert(ierr==0);
+}
 void CartesianCommunicator::GlobalSum(float &f){
   int ierr=MPI_Allreduce(MPI_IN_PLACE,&f,1,MPI_FLOAT,MPI_SUM,communicator);
   assert(ierr==0);

lib/communicator/Communicator_mpi3.cc

@@ -65,6 +65,7 @@ std::vector<int> CartesianCommunicator::MyGroup;
 std::vector<void *> CartesianCommunicator::ShmCommBufs;
 
 int CartesianCommunicator::NodeCount(void)    { return GroupSize;};
+int CartesianCommunicator::RankCount(void)    { return WorldSize;};
 
 
 #undef FORCE_COMMS
@@ -509,6 +510,14 @@ void CartesianCommunicator::GlobalSum(uint64_t &u){
   int ierr=MPI_Allreduce(MPI_IN_PLACE,&u,1,MPI_UINT64_T,MPI_SUM,communicator);
   assert(ierr==0);
 }
+void CartesianCommunicator::GlobalXOR(uint32_t &u){
+  int ierr=MPI_Allreduce(MPI_IN_PLACE,&u,1,MPI_UINT32_T,MPI_BXOR,communicator);
+  assert(ierr==0);
+}
+void CartesianCommunicator::GlobalXOR(uint64_t &u){
+  int ierr=MPI_Allreduce(MPI_IN_PLACE,&u,1,MPI_UINT64_T,MPI_BXOR,communicator);
+  assert(ierr==0);
+}
 void CartesianCommunicator::GlobalSum(float &f){
   int ierr=MPI_Allreduce(MPI_IN_PLACE,&f,1,MPI_FLOAT,MPI_SUM,communicator);
   assert(ierr==0);

lib/communicator/Communicator_none.cc

@@ -59,6 +59,8 @@ void CartesianCommunicator::GlobalSum(double &){}
 void CartesianCommunicator::GlobalSum(uint32_t &){}
 void CartesianCommunicator::GlobalSum(uint64_t &){}
 void CartesianCommunicator::GlobalSumVector(double *,int N){}
+void CartesianCommunicator::GlobalXOR(uint32_t &){}
+void CartesianCommunicator::GlobalXOR(uint64_t &){}
 
 void CartesianCommunicator::SendRecvPacket(void *xmit,
 					   void *recv,

lib/lattice/Lattice_reduction.h

@@ -1,4 +1,4 @@
- /*************************************************************************************
+/*************************************************************************************
     Grid physics library, www.github.com/paboyle/Grid 
     Source file: ./lib/lattice/Lattice_reduction.h
     Copyright (C) 2015
@@ -328,6 +328,8 @@ static void sliceMaddVector(Lattice<vobj> &R,std::vector<RealD> &a,const Lattice
   typedef typename vobj::vector_type vector_type;
   typedef typename vobj::tensor_reduced tensor_reduced;
   
+  scalar_type zscale(scale);
+
   GridBase *grid  = X._grid;
 
   int Nsimd  =grid->Nsimd();
@@ -353,7 +355,7 @@ static void sliceMaddVector(Lattice<vobj> &R,std::vector<RealD> &a,const Lattice
       grid->iCoorFromIindex(icoor,l);
       int ldx =r+icoor[orthogdim]*rd;
       scalar_type *as =(scalar_type *)&av;
-      as[l] = scalar_type(a[ldx])*scale;
+      as[l] = scalar_type(a[ldx])*zscale;
     }
 
     tensor_reduced at; at=av;
@@ -367,71 +369,6 @@ static void sliceMaddVector(Lattice<vobj> &R,std::vector<RealD> &a,const Lattice
   }
 };
 
-
-/*
-template<class vobj>
-static void sliceMaddVectorSlow (Lattice<vobj> &R,std::vector<RealD> &a,const Lattice<vobj> &X,const Lattice<vobj> &Y,
-			     int Orthog,RealD scale=1.0) 
-{    
-  // FIXME: Implementation is slow
-  // Best base the linear combination by constructing a 
-  // set of vectors of size grid->_rdimensions[Orthog].
-  typedef typename vobj::scalar_object sobj;
-  typedef typename vobj::scalar_type scalar_type;
-  typedef typename vobj::vector_type vector_type;
-  
-  int Nblock = X._grid->GlobalDimensions()[Orthog];
-  
-  GridBase *FullGrid  = X._grid;
-  GridBase *SliceGrid = makeSubSliceGrid(FullGrid,Orthog);
-  
-  Lattice<vobj> Xslice(SliceGrid);
-  Lattice<vobj> Rslice(SliceGrid);
-  // If we based this on Cshift it would work for spread out
-  // but it would be even slower
-  for(int i=0;i<Nblock;i++){
-    ExtractSlice(Rslice,Y,i,Orthog);
-    ExtractSlice(Xslice,X,i,Orthog);
-    Rslice = Rslice + Xslice*(scale*a[i]);
-    InsertSlice(Rslice,R,i,Orthog);
-  }
-};
-template<class vobj>
-static void sliceInnerProductVectorSlow( std::vector<ComplexD> & vec, const Lattice<vobj> &lhs,const Lattice<vobj> &rhs,int Orthog) 
-  {
-    // FIXME: Implementation is slow
-    // Look at localInnerProduct implementation,
-    // and do inside a site loop with block strided iterators
-    typedef typename vobj::scalar_object sobj;
-    typedef typename vobj::scalar_type scalar_type;
-    typedef typename vobj::vector_type vector_type;
-    typedef typename vobj::tensor_reduced scalar;
-    typedef typename scalar::scalar_object  scomplex;
-  
-    int Nblock = lhs._grid->GlobalDimensions()[Orthog];
-    vec.resize(Nblock);
-    std::vector<scomplex> sip(Nblock);
-    Lattice<scalar> IP(lhs._grid); 
-    IP=localInnerProduct(lhs,rhs);
-    sliceSum(IP,sip,Orthog);
-  
-    for(int ss=0;ss<Nblock;ss++){
-      vec[ss] = TensorRemove(sip[ss]);
-    }
-  }
-*/
-
-//////////////////////////////////////////////////////////////////////////////////////////
-// FIXME: Implementation is slow
-// If we based this on Cshift it would work for spread out
-// but it would be even slower
-//
-// Repeated extract slice is inefficient
-//
-// Best base the linear combination by constructing a 
-// set of vectors of size grid->_rdimensions[Orthog].
-//////////////////////////////////////////////////////////////////////////////////////////
-
 inline GridBase         *makeSubSliceGrid(const GridBase *BlockSolverGrid,int Orthog)
 {
   int NN    = BlockSolverGrid->_ndimension;
@@ -451,7 +388,6 @@ inline GridBase         *makeSubSliceGrid(const GridBase *BlockSolverGrid,int Or
   return (GridBase *)new GridCartesian(latt_phys,simd_phys,mpi_phys); 
 }
 
-
 template<class vobj>
 static void sliceMaddMatrix (Lattice<vobj> &R,Eigen::MatrixXcd &aa,const Lattice<vobj> &X,const Lattice<vobj> &Y,int Orthog,RealD scale=1.0) 
 {    
@@ -460,28 +396,103 @@ static void sliceMaddMatrix (Lattice<vobj> &R,Eigen::MatrixXcd &aa,const Lattice
   typedef typename vobj::vector_type vector_type;
 
   int Nblock = X._grid->GlobalDimensions()[Orthog];
-  
+
   GridBase *FullGrid  = X._grid;
   GridBase *SliceGrid = makeSubSliceGrid(FullGrid,Orthog);
-  
+
   Lattice<vobj> Xslice(SliceGrid);
   Lattice<vobj> Rslice(SliceGrid);
-  
-  for(int i=0;i<Nblock;i++){
-    ExtractSlice(Rslice,Y,i,Orthog);
-    for(int j=0;j<Nblock;j++){
-      ExtractSlice(Xslice,X,j,Orthog);
-      Rslice = Rslice + Xslice*(scale*aa(j,i));
-    }
-    InsertSlice(Rslice,R,i,Orthog);
+
+  assert( FullGrid->_simd_layout[Orthog]==1);
+  int nh =  FullGrid->_ndimension;
+  int nl = SliceGrid->_ndimension;
+
+  //FIXME package in a convenient iterator
+  //Should loop over a plane orthogonal to direction "Orthog"
+  int stride=FullGrid->_slice_stride[Orthog];
+  int block =FullGrid->_slice_block [Orthog];
+  int nblock=FullGrid->_slice_nblock[Orthog];
+  int ostride=FullGrid->_ostride[Orthog];
+#pragma omp parallel 
+  {
+    std::vector<vobj> s_x(Nblock);
+
+#pragma omp for collapse(2)
+    for(int n=0;n<nblock;n++){
+    for(int b=0;b<block;b++){
+      int o  = n*stride + b;
+
+      for(int i=0;i<Nblock;i++){
+	s_x[i] = X[o+i*ostride];
+      }
+
+      vobj dot;
+      for(int i=0;i<Nblock;i++){
+	dot = Y[o+i*ostride];
+	for(int j=0;j<Nblock;j++){
+	  dot = dot + s_x[j]*(scale*aa(j,i));
+	}
+	R[o+i*ostride]=dot;
+      }
+    }}
   }
 };
 
+template<class vobj>
+static void sliceMulMatrix (Lattice<vobj> &R,Eigen::MatrixXcd &aa,const Lattice<vobj> &X,int Orthog,RealD scale=1.0) 
+{    
+  typedef typename vobj::scalar_object sobj;
+  typedef typename vobj::scalar_type scalar_type;
+  typedef typename vobj::vector_type vector_type;
+
+  int Nblock = X._grid->GlobalDimensions()[Orthog];
+
+  GridBase *FullGrid  = X._grid;
+  GridBase *SliceGrid = makeSubSliceGrid(FullGrid,Orthog);
+
+  Lattice<vobj> Xslice(SliceGrid);
+  Lattice<vobj> Rslice(SliceGrid);
+
+  assert( FullGrid->_simd_layout[Orthog]==1);
+  int nh =  FullGrid->_ndimension;
+  int nl = SliceGrid->_ndimension;
+
+  //FIXME package in a convenient iterator
+  //Should loop over a plane orthogonal to direction "Orthog"
+  int stride=FullGrid->_slice_stride[Orthog];
+  int block =FullGrid->_slice_block [Orthog];
+  int nblock=FullGrid->_slice_nblock[Orthog];
+  int ostride=FullGrid->_ostride[Orthog];
+#pragma omp parallel 
+  {
+    std::vector<vobj> s_x(Nblock);
+
+#pragma omp for collapse(2)
+    for(int n=0;n<nblock;n++){
+    for(int b=0;b<block;b++){
+      int o  = n*stride + b;
+
+      for(int i=0;i<Nblock;i++){
+	s_x[i] = X[o+i*ostride];
+      }
+
+      vobj dot;
+      for(int i=0;i<Nblock;i++){
+	dot = s_x[0]*(scale*aa(0,i));
+	for(int j=1;j<Nblock;j++){
+	  dot = dot + s_x[j]*(scale*aa(j,i));
+	}
+	R[o+i*ostride]=dot;
+      }
+    }}
+  }
+
+};
+
+
 template<class vobj>
 static void sliceInnerProductMatrix(  Eigen::MatrixXcd &mat, const Lattice<vobj> &lhs,const Lattice<vobj> &rhs,int Orthog) 
 {
-  // FIXME: Implementation is slow
-  // Not sure of best solution.. think about it
   typedef typename vobj::scalar_object sobj;
   typedef typename vobj::scalar_type scalar_type;
   typedef typename vobj::vector_type vector_type;
@@ -495,22 +506,49 @@ static void sliceInnerProductMatrix(  Eigen::MatrixXcd &mat, const Lattice<vobj>
   Lattice<vobj> Rslice(SliceGrid);
   
   mat = Eigen::MatrixXcd::Zero(Nblock,Nblock);
-  
-  for(int i=0;i<Nblock;i++){
-    ExtractSlice(Lslice,lhs,i,Orthog);
-    for(int j=0;j<Nblock;j++){
-      ExtractSlice(Rslice,rhs,j,Orthog);
-      mat(i,j) = innerProduct(Lslice,Rslice);
-    }
+
+  assert( FullGrid->_simd_layout[Orthog]==1);
+  int nh =  FullGrid->_ndimension;
+  int nl = SliceGrid->_ndimension;
+
+  //FIXME package in a convenient iterator
+  //Should loop over a plane orthogonal to direction "Orthog"
+  int stride=FullGrid->_slice_stride[Orthog];
+  int block =FullGrid->_slice_block [Orthog];
+  int nblock=FullGrid->_slice_nblock[Orthog];
+  int ostride=FullGrid->_ostride[Orthog];
+
+  typedef typename vobj::vector_typeD vector_typeD;
+
+#pragma omp parallel 
+  {
+    std::vector<vobj> Left(Nblock);
+    std::vector<vobj> Right(Nblock);
+    Eigen::MatrixXcd  mat_thread = Eigen::MatrixXcd::Zero(Nblock,Nblock);
+
+#pragma omp for collapse(2)
+    for(int n=0;n<nblock;n++){
+    for(int b=0;b<block;b++){
+
+      int o  = n*stride + b;
+
+      for(int i=0;i<Nblock;i++){
+	Left [i] = lhs[o+i*ostride];
+	Right[i] = rhs[o+i*ostride];
+      }
+
+      for(int i=0;i<Nblock;i++){
+      for(int j=0;j<Nblock;j++){
+	auto tmp = innerProduct(Left[i],Right[j]);
+	vector_typeD rtmp = TensorRemove(tmp);
+	mat_thread(i,j) += Reduce(rtmp);
+      }}
+    }}
+#pragma omp critical
+    {
+      mat += mat_thread;
+    }  
   }
-#undef FORCE_DIAG
-#ifdef FORCE_DIAG
-  for(int i=0;i<Nblock;i++){
-    for(int j=0;j<Nblock;j++){
-      if ( i != j ) mat(i,j)=0.0;
-    }
-  }
-#endif
   return;
 }
 

lib/lattice/Lattice_transfer.h

@@ -551,7 +551,10 @@ void Replicate(Lattice<vobj> &coarse,Lattice<vobj> & fine)
 
 //Copy SIMD-vectorized lattice to array of scalar objects in lexicographic order
 template<typename vobj, typename sobj>
-typename std::enable_if<isSIMDvectorized<vobj>::value && !isSIMDvectorized<sobj>::value, void>::type unvectorizeToLexOrdArray(std::vector<sobj> &out, const Lattice<vobj> &in){
+typename std::enable_if<isSIMDvectorized<vobj>::value && !isSIMDvectorized<sobj>::value, void>::type 
+unvectorizeToLexOrdArray(std::vector<sobj> &out, const Lattice<vobj> &in)
+{
+
   typedef typename vobj::vector_type vtype;
   
   GridBase* in_grid = in._grid;
@@ -590,6 +593,54 @@ typename std::enable_if<isSIMDvectorized<vobj>::value && !isSIMDvectorized<sobj>
     extract1(in_vobj, out_ptrs, 0);
   }
 }
+//Copy SIMD-vectorized lattice to array of scalar objects in lexicographic order
+template<typename vobj, typename sobj>
+typename std::enable_if<isSIMDvectorized<vobj>::value 
+                    && !isSIMDvectorized<sobj>::value, void>::type 
+vectorizeFromLexOrdArray( std::vector<sobj> &in, Lattice<vobj> &out)
+{
+
+  typedef typename vobj::vector_type vtype;
+  
+  GridBase* grid = out._grid;
+  assert(in.size()==grid->lSites());
+  
+  int ndim     = grid->Nd();
+  int nsimd    = vtype::Nsimd();
+
+  std::vector<std::vector<int> > icoor(nsimd);
+      
+  for(int lane=0; lane < nsimd; lane++){
+    icoor[lane].resize(ndim);
+    grid->iCoorFromIindex(icoor[lane],lane);
+  }
+  
+  parallel_for(uint64_t oidx = 0; oidx < grid->oSites(); oidx++){ //loop over outer index
+    //Assemble vector of pointers to output elements
+    std::vector<sobj*> ptrs(nsimd);
+
+    std::vector<int> ocoor(ndim);
+    grid->oCoorFromOindex(ocoor, oidx);
+
+    std::vector<int> lcoor(grid->Nd());
+      
+    for(int lane=0; lane < nsimd; lane++){
+
+      for(int mu=0;mu<ndim;mu++){
+	lcoor[mu] = ocoor[mu] + grid->_rdimensions[mu]*icoor[lane][mu];
+      }
+
+      int lex;
+      Lexicographic::IndexFromCoor(lcoor, lex, grid->_ldimensions);
+      ptrs[lane] = &in[lex];
+    }
+    
+    //pack from those ptrs
+    vobj vecobj;
+    merge1(vecobj, ptrs, 0);
+    out._odata[oidx] = vecobj; 
+  }
+}
 
 //Convert a Lattice from one precision to another
 template<class VobjOut, class VobjIn>
@@ -615,7 +666,7 @@ void precisionChange(Lattice<VobjOut> &out, const Lattice<VobjIn> &in){
   std::vector<SobjOut> in_slex_conv(in_grid->lSites());
   unvectorizeToLexOrdArray(in_slex_conv, in);
     
-  parallel_for(int out_oidx=0;out_oidx<out_grid->oSites();out_oidx++){
+  parallel_for(uint64_t out_oidx=0;out_oidx<out_grid->oSites();out_oidx++){
     std::vector<int> out_ocoor(ndim);
     out_grid->oCoorFromOindex(out_ocoor, out_oidx);
 

lib/lattice/Lattice_unary.h

@@ -62,14 +62,20 @@ namespace Grid {
     return ret;
   }
 
-  template<class obj> Lattice<obj> expMat(const Lattice<obj> &rhs, ComplexD alpha, Integer Nexp = DEFAULT_MAT_EXP){
+  template<class obj> Lattice<obj> expMat(const Lattice<obj> &rhs, RealD alpha, Integer Nexp = DEFAULT_MAT_EXP){
     Lattice<obj> ret(rhs._grid);
     ret.checkerboard = rhs.checkerboard;
     conformable(ret,rhs);
     parallel_for(int ss=0;ss<rhs._grid->oSites();ss++){
       ret._odata[ss]=Exponentiate(rhs._odata[ss],alpha, Nexp);
     }
+
     return ret;
+
+    
+    
+
+    
   }
 
 

lib/parallelIO/IldgIO.h

@@ -27,6 +27,7 @@ directory
 #ifndef GRID_ILDG_IO_H
 #define GRID_ILDG_IO_H
 
+#ifdef HAVE_LIME
 #include <algorithm>
 #include <fstream>
 #include <iomanip>
@@ -37,213 +38,677 @@ directory
 #include <sys/utsname.h>
 #include <unistd.h>
 
-#ifdef HAVE_LIME
-
-extern "C" {  // for linkage
+//C-Lime is a must have for this functionality
+extern "C" {  
 #include "lime.h"
 }
 
 namespace Grid {
 namespace QCD {
 
-inline void ILDGGrid(GridBase *grid, ILDGField &header) {
-  assert(grid->_ndimension == 4);  // emit error if not
-  header.dimension.resize(4);
-  header.boundary.resize(4);
-  for (int d = 0; d < 4; d++) {
-    header.dimension[d] = grid->_fdimensions[d];
-    // Read boundary conditions from ... ?
-    header.boundary[d] = std::string("periodic");
-  }
-}
+  /////////////////////////////////
+  // Encode word types as strings
+  /////////////////////////////////
+ template<class word> inline std::string ScidacWordMnemonic(void){ return std::string("unknown"); }
+ template<> inline std::string ScidacWordMnemonic<double>  (void){ return std::string("D"); }
+ template<> inline std::string ScidacWordMnemonic<float>   (void){ return std::string("F"); }
+ template<> inline std::string ScidacWordMnemonic< int32_t>(void){ return std::string("I32_t"); }
+ template<> inline std::string ScidacWordMnemonic<uint32_t>(void){ return std::string("U32_t"); }
+ template<> inline std::string ScidacWordMnemonic< int64_t>(void){ return std::string("I64_t"); }
+ template<> inline std::string ScidacWordMnemonic<uint64_t>(void){ return std::string("U64_t"); }
 
-inline void ILDGChecksum(uint32_t *buf, uint32_t buf_size_bytes,
-                         uint32_t &csum) {
-  BinaryIO::Uint32Checksum(buf, buf_size_bytes, csum);
-}
+  /////////////////////////////////////////
+  // Encode a generic tensor as a string
+  /////////////////////////////////////////
+ template<class vobj> std::string ScidacRecordTypeString(int &colors, int &spins, int & typesize,int &datacount) { 
 
-//////////////////////////////////////////////////////////////////////
-// Utilities ; these are QCD aware
-//////////////////////////////////////////////////////////////////////
-template <class GaugeField>
-inline void ILDGStatistics(GaugeField &data, ILDGField &header) {
-  // How to convert data precision etc...
-  header.link_trace = Grid::QCD::WilsonLoops<PeriodicGimplR>::linkTrace(data);
-  header.plaquette = Grid::QCD::WilsonLoops<PeriodicGimplR>::avgPlaquette(data);
-  // header.polyakov =
-}
+   typedef typename getPrecision<vobj>::real_scalar_type stype;
 
-// Forcing QCD here
-template <class fobj, class sobj>
-struct ILDGMunger {
-  void operator()(fobj &in, sobj &out, uint32_t &csum) {
-    for (int mu = 0; mu < 4; mu++) {
-      for (int i = 0; i < 3; i++) {
-        for (int j = 0; j < 3; j++) {
-          out(mu)()(i, j) = in(mu)()(i, j);
-        }
-      }
-    }
-    ILDGChecksum((uint32_t *)&in, sizeof(in), csum);
-  };
-};
+   int _ColourN       = indexRank<ColourIndex,vobj>();
+   int _ColourScalar  =  isScalar<ColourIndex,vobj>();
+   int _ColourVector  =  isVector<ColourIndex,vobj>();
+   int _ColourMatrix  =  isMatrix<ColourIndex,vobj>();
 
-template <class fobj, class sobj>
-struct ILDGUnmunger {
-  void operator()(sobj &in, fobj &out, uint32_t &csum) {
-    for (int mu = 0; mu < 4; mu++) {
-      for (int i = 0; i < 3; i++) {
-        for (int j = 0; j < 3; j++) {
-          out(mu)()(i, j) = in(mu)()(i, j);
-        }
-      }
-    }
-    ILDGChecksum((uint32_t *)&out, sizeof(out), csum);
-  };
-};
+   int _SpinN       = indexRank<SpinIndex,vobj>();
+   int _SpinScalar  =  isScalar<SpinIndex,vobj>();
+   int _SpinVector  =  isVector<SpinIndex,vobj>();
+   int _SpinMatrix  =  isMatrix<SpinIndex,vobj>();
 
-////////////////////////////////////////////////////////////////////////////////
-// Write and read from fstream; compute header offset for payload
-////////////////////////////////////////////////////////////////////////////////
-enum ILDGstate {ILDGread, ILDGwrite};
+   int _LorentzN       = indexRank<LorentzIndex,vobj>();
+   int _LorentzScalar  =  isScalar<LorentzIndex,vobj>();
+   int _LorentzVector  =  isVector<LorentzIndex,vobj>();
+   int _LorentzMatrix  =  isMatrix<LorentzIndex,vobj>();
 
-class ILDGIO : public BinaryIO {
-  FILE *File;
-  LimeWriter *LimeW;
-  LimeRecordHeader *LimeHeader;
-  LimeReader *LimeR;
-  std::string filename;
+   std::stringstream stream;
+
+   stream << "GRID_";
+   stream << ScidacWordMnemonic<stype>();
+
+   //   std::cout << " Lorentz N/S/V/M : " << _LorentzN<<" "<<_LorentzScalar<<"/"<<_LorentzVector<<"/"<<_LorentzMatrix<<std::endl;
+   //   std::cout << " Spin    N/S/V/M : " << _SpinN   <<" "<<_SpinScalar   <<"/"<<_SpinVector   <<"/"<<_SpinMatrix<<std::endl;
+   //   std::cout << " Colour  N/S/V/M : " << _ColourN <<" "<<_ColourScalar <<"/"<<_ColourVector <<"/"<<_ColourMatrix<<std::endl;
+
+   if ( _LorentzVector )   stream << "_LorentzVector"<<_LorentzN;
+   if ( _LorentzMatrix )   stream << "_LorentzMatrix"<<_LorentzN;
+
+   if ( _SpinVector )   stream << "_SpinVector"<<_SpinN;
+   if ( _SpinMatrix )   stream << "_SpinMatrix"<<_SpinN;
+
+   if ( _ColourVector )   stream << "_ColourVector"<<_ColourN;
+   if ( _ColourMatrix )   stream << "_ColourMatrix"<<_ColourN;
+
+   if ( _ColourScalar && _LorentzScalar && _SpinScalar )   stream << "_Complex";
 
 
+   typesize = sizeof(typename vobj::scalar_type);
+
+   if ( _ColourMatrix ) typesize*= _ColourN*_ColourN;
+   else                 typesize*= _ColourN;
+
+   if ( _SpinMatrix )   typesize*= _SpinN*_SpinN;
+   else                 typesize*= _SpinN;
+
+   colors    = _ColourN;
+   spins     = _SpinN;
+   datacount = _LorentzN;
+
+   return stream.str();
+ }
+ 
+ template<class vobj> std::string ScidacRecordTypeString(Lattice<vobj> & lat,int &colors, int &spins, int & typesize,int &datacount) { 
+   return ScidacRecordTypeString<vobj>(colors,spins,typesize,datacount);
+ };
+
+
+ ////////////////////////////////////////////////////////////
+ // Helper to fill out metadata
+ ////////////////////////////////////////////////////////////
+ template<class vobj> void ScidacMetaData(Lattice<vobj> & field,
+					  FieldMetaData &header,
+					  scidacRecord & _scidacRecord,
+					  scidacFile   & _scidacFile) 
+ {
+   typedef typename getPrecision<vobj>::real_scalar_type stype;
+
+   /////////////////////////////////////
+   // Pull Grid's metadata
+   /////////////////////////////////////
+   PrepareMetaData(field,header);
+
+   /////////////////////////////////////
+   // Scidac Private File structure
+   /////////////////////////////////////
+   _scidacFile              = scidacFile(field._grid);
+
+   /////////////////////////////////////
+   // Scidac Private Record structure
+   /////////////////////////////////////
+   scidacRecord sr;
+   sr.datatype   = ScidacRecordTypeString(field,sr.colors,sr.spins,sr.typesize,sr.datacount);
+   sr.date       = header.creation_date;
+   sr.precision  = ScidacWordMnemonic<stype>();
+   sr.recordtype = GRID_IO_FIELD;
+
+   _scidacRecord = sr;
+
+   std::cout << GridLogMessage << "Build SciDAC datatype " <<sr.datatype<<std::endl;
+ }
+ 
+ ///////////////////////////////////////////////////////
+ // Scidac checksum
+ ///////////////////////////////////////////////////////
+ static int scidacChecksumVerify(scidacChecksum &scidacChecksum_,uint32_t scidac_csuma,uint32_t scidac_csumb)
+ {
+   uint32_t scidac_checksuma = stoull(scidacChecksum_.suma,0,16);
+   uint32_t scidac_checksumb = stoull(scidacChecksum_.sumb,0,16);
+   if ( scidac_csuma !=scidac_checksuma) return 0;
+   if ( scidac_csumb !=scidac_checksumb) return 0;
+    return 1;
+ }
+
+////////////////////////////////////////////////////////////////////////////////////
+// Lime, ILDG and Scidac I/O classes
+////////////////////////////////////////////////////////////////////////////////////
+class GridLimeReader : public BinaryIO {
  public:
-  ILDGIO(std::string file, ILDGstate RW) {
-      filename = file;
-    if (RW == ILDGwrite){
-      File = fopen(file.c_str(), "w");
-      // check if opened correctly
+   ///////////////////////////////////////////////////
+   // FIXME: format for RNG? Now just binary out instead
+   ///////////////////////////////////////////////////
 
-      LimeW = limeCreateWriter(File);
-    } else {
-      File = fopen(file.c_str(), "r");
-      // check if opened correctly
+   FILE       *File;
+   LimeReader *LimeR;
+   std::string filename;
 
-      LimeR = limeCreateReader(File);
+   /////////////////////////////////////////////
+   // Open the file
+   /////////////////////////////////////////////
+   void open(std::string &_filename) 
+   {
+     filename= _filename;
+     File = fopen(filename.c_str(), "r");
+     LimeR = limeCreateReader(File);
+   }
+   /////////////////////////////////////////////
+   // Close the file
+   /////////////////////////////////////////////
+   void close(void){
+     fclose(File);
+     //     limeDestroyReader(LimeR);
+   }
+
+  ////////////////////////////////////////////
+  // Read a generic lattice field and verify checksum
+  ////////////////////////////////////////////
+  template<class vobj>
+  void readLimeLatticeBinaryObject(Lattice<vobj> &field,std::string record_name)
+  {
+    typedef typename vobj::scalar_object sobj;
+    scidacChecksum scidacChecksum_;
+    uint32_t nersc_csum,scidac_csuma,scidac_csumb;
+
+    std::string format = getFormatString<vobj>();
+
+    while ( limeReaderNextRecord(LimeR) == LIME_SUCCESS ) { 
+
+      std::cout << GridLogMessage << limeReaderType(LimeR) <<std::endl;
+	
+      if ( strncmp(limeReaderType(LimeR), record_name.c_str(),strlen(record_name.c_str()) )  ) {
+
+
+	off_t offset= ftell(File);
+	BinarySimpleMunger<sobj,sobj> munge;
+	BinaryIO::readLatticeObject< sobj, sobj >(field, filename, munge, offset, format,nersc_csum,scidac_csuma,scidac_csumb);
+
+	/////////////////////////////////////////////
+	// Insist checksum is next record
+	/////////////////////////////////////////////
+	readLimeObject(scidacChecksum_,std::string("scidacChecksum"),record_name);
+
+	/////////////////////////////////////////////
+	// Verify checksums
+	/////////////////////////////////////////////
+	scidacChecksumVerify(scidacChecksum_,scidac_csuma,scidac_csumb);
+	return;
+      }
     }
   }
+  ////////////////////////////////////////////
+  // Read a generic serialisable object
+  ////////////////////////////////////////////
+  template<class serialisable_object>
+  void readLimeObject(serialisable_object &object,std::string object_name,std::string record_name)
+  {
+    std::string xmlstring;
+    // should this be a do while; can we miss a first record??
+    while ( limeReaderNextRecord(LimeR) == LIME_SUCCESS ) { 
 
-  ~ILDGIO() { fclose(File); }
+      uint64_t nbytes = limeReaderBytes(LimeR);//size of this record (configuration)
 
-  int createHeader(std::string message, int MB, int ME, size_t PayloadSize, LimeWriter* L){
+      if ( strncmp(limeReaderType(LimeR), record_name.c_str(),strlen(record_name.c_str()) )  ) {
+	std::vector<char> xmlc(nbytes+1,'\0');
+	limeReaderReadData((void *)&xmlc[0], &nbytes, LimeR);    
+	XmlReader RD(&xmlc[0],"");
+	read(RD,object_name,object);
+	return;
+      }
+
+    }  
+    assert(0);
+  }
+};
+
+class GridLimeWriter : public BinaryIO {
+ public:
+   ///////////////////////////////////////////////////
+   // FIXME: format for RNG? Now just binary out instead
+   ///////////////////////////////////////////////////
+
+   FILE       *File;
+   LimeWriter *LimeW;
+   std::string filename;
+
+   void open(std::string &_filename) { 
+     filename= _filename;
+     File = fopen(filename.c_str(), "w");
+     LimeW = limeCreateWriter(File); assert(LimeW != NULL );
+   }
+   /////////////////////////////////////////////
+   // Close the file
+   /////////////////////////////////////////////
+   void close(void) {
+     fclose(File);
+     //  limeDestroyWriter(LimeW);
+   }
+  ///////////////////////////////////////////////////////
+  // Lime utility functions
+  ///////////////////////////////////////////////////////
+  int createLimeRecordHeader(std::string message, int MB, int ME, size_t PayloadSize)
+  {
     LimeRecordHeader *h;
     h = limeCreateHeader(MB, ME, const_cast<char *>(message.c_str()), PayloadSize);
-    int status = limeWriteRecordHeader(h, L);
-    if (status < 0) {
-      std::cerr << "ILDG Header error\n";
-      return status;
-    }
+    assert(limeWriteRecordHeader(h, LimeW) >= 0);
     limeDestroyHeader(h);
     return LIME_SUCCESS;
   }
+  ////////////////////////////////////////////
+  // Write a generic serialisable object
+  ////////////////////////////////////////////
+  template<class serialisable_object>
+  void writeLimeObject(int MB,int ME,serialisable_object &object,std::string object_name,std::string record_name)
+  {
+    std::string xmlstring;
+    {
+      XmlWriter WR("","");
+      write(WR,object_name,object);
+      xmlstring = WR.XmlString();
+    }
+    uint64_t nbytes = xmlstring.size();
+    int err;
+    LimeRecordHeader *h = limeCreateHeader(MB, ME,(char *)record_name.c_str(), nbytes); assert(h!= NULL);
 
-  unsigned int writeHeader(ILDGField &header) {
-    // write header in LIME
-    n_uint64_t nbytes;
-    int MB_flag = 1, ME_flag = 0;
+    err=limeWriteRecordHeader(h, LimeW);                    assert(err>=0);
+    err=limeWriteRecordData(&xmlstring[0], &nbytes, LimeW); assert(err>=0);
+    err=limeWriterCloseRecord(LimeW);                       assert(err>=0);
+    limeDestroyHeader(h);
+  }
+  ////////////////////////////////////////////
+  // Write a generic lattice field and csum
+  ////////////////////////////////////////////
+  template<class vobj>
+  void writeLimeLatticeBinaryObject(Lattice<vobj> &field,std::string record_name)
+  {
+    ////////////////////////////////////////////
+    // Create record header
+    ////////////////////////////////////////////
+    typedef typename vobj::scalar_object sobj;
+    int err;
+    uint32_t nersc_csum,scidac_csuma,scidac_csumb;
+    uint64_t PayloadSize = sizeof(sobj) * field._grid->_gsites;
+    createLimeRecordHeader(record_name, 0, 0, PayloadSize);
 
-    char message[] = "ildg-format";
-    nbytes = strlen(message);
-    LimeHeader = limeCreateHeader(MB_flag, ME_flag, message, nbytes);
-    limeWriteRecordHeader(LimeHeader, LimeW);
-    limeDestroyHeader(LimeHeader);
-    // save the xml header here
-    // use the xml_writer to c++ streams in pugixml
-    // and convert to char message
-    limeWriteRecordData(message, &nbytes, LimeW);
-    limeWriterCloseRecord(LimeW);
+    ////////////////////////////////////////////////////////////////////
+    // NB: FILE and iostream are jointly writing disjoint sequences in the
+    // the same file through different file handles (integer units).
+    // 
+    // These are both buffered, so why I think this code is right is as follows.
+    //
+    // i)  write record header to FILE *File, telegraphing the size. 
+    // ii) ftell reads the offset from FILE *File .
+    // iii) iostream / MPI Open independently seek this offset. Write sequence direct to disk.
+    //      Closes iostream and flushes.
+    // iv) fseek on FILE * to end of this disjoint section.
+    //  v) Continue writing scidac record.
+    ////////////////////////////////////////////////////////////////////
+    off_t offset = ftell(File);
+    std::string format = getFormatString<vobj>();
+    BinarySimpleMunger<sobj,sobj> munge;
+    BinaryIO::writeLatticeObject<vobj,sobj>(field, filename, munge, offset, format,nersc_csum,scidac_csuma,scidac_csumb);
+    err=limeWriterCloseRecord(LimeW);  assert(err>=0);
+    ////////////////////////////////////////
+    // Write checksum element, propagaing forward from the BinaryIO
+    // Always pair a checksum with a binary object, and close message
+    ////////////////////////////////////////
+    scidacChecksum checksum;
+    std::stringstream streama; streama << std::hex << scidac_csuma;
+    std::stringstream streamb; streamb << std::hex << scidac_csumb;
+    checksum.suma= streama.str();
+    checksum.sumb= streamb.str();
+    std::cout << GridLogMessage<<" writing scidac checksums "<<std::hex<<scidac_csuma<<"/"<<scidac_csumb<<std::dec<<std::endl;
+    writeLimeObject(0,1,checksum,std::string("scidacChecksum"    ),std::string(SCIDAC_CHECKSUM));
+  }
+};
 
-    return 0;
-  }
-
-  unsigned int readHeader(ILDGField &header) {
-    return 0;
+class ScidacWriter : public GridLimeWriter {
+ public:
+
+   template<class SerialisableUserFile>
+   void writeScidacFileRecord(GridBase *grid,SerialisableUserFile &_userFile)
+   {
+     scidacFile    _scidacFile(grid);
+     writeLimeObject(1,0,_scidacFile,_scidacFile.SerialisableClassName(),std::string(SCIDAC_PRIVATE_FILE_XML));
+     writeLimeObject(0,1,_userFile,_userFile.SerialisableClassName(),std::string(SCIDAC_FILE_XML));
+   }
+  ////////////////////////////////////////////////
+  // Write generic lattice field in scidac format
+  ////////////////////////////////////////////////
+   template <class vobj, class userRecord>
+  void writeScidacFieldRecord(Lattice<vobj> &field,userRecord _userRecord) 
+  {
+    typedef typename vobj::scalar_object sobj;
+    uint64_t nbytes;
+    GridBase * grid = field._grid;
+
+    ////////////////////////////////////////
+    // fill the Grid header
+    ////////////////////////////////////////
+    FieldMetaData header;
+    scidacRecord  _scidacRecord;
+    scidacFile    _scidacFile;
+
+    ScidacMetaData(field,header,_scidacRecord,_scidacFile);
+
+    //////////////////////////////////////////////
+    // Fill the Lime file record by record
+    //////////////////////////////////////////////
+    writeLimeObject(1,0,header ,std::string("FieldMetaData"),std::string(GRID_FORMAT)); // Open message 
+    writeLimeObject(0,0,_userRecord,_userRecord.SerialisableClassName(),std::string(SCIDAC_RECORD_XML));
+    writeLimeObject(0,0,_scidacRecord,_scidacRecord.SerialisableClassName(),std::string(SCIDAC_PRIVATE_RECORD_XML));
+    writeLimeLatticeBinaryObject(field,std::string(ILDG_BINARY_DATA));      // Closes message with checksum
+  }
+};
+
+class IldgWriter : public ScidacWriter {
+ public:
+
+  ///////////////////////////////////
+  // A little helper
+  ///////////////////////////////////
+  void writeLimeIldgLFN(std::string &LFN)
+  {
+    uint64_t PayloadSize = LFN.size();
+    int err;
+    createLimeRecordHeader(ILDG_DATA_LFN, 0 , 0, PayloadSize);
+    err=limeWriteRecordData(const_cast<char*>(LFN.c_str()), &PayloadSize,LimeW); assert(err>=0);
+    err=limeWriterCloseRecord(LimeW); assert(err>=0);
   }
 
+  ////////////////////////////////////////////////////////////////
+  // Special ILDG operations ; gauge configs only.
+  // Don't require scidac records EXCEPT checksum
+  // Use Grid MetaData object if present.
+  ////////////////////////////////////////////////////////////////
   template <class vsimd>
-  uint32_t readConfiguration(Lattice<iLorentzColourMatrix<vsimd> > &Umu) {
-    typedef Lattice<iLorentzColourMatrix<vsimd> > GaugeField;
-    typedef LorentzColourMatrixD sobjd;
-    typedef LorentzColourMatrixF sobjf;
-    typedef iLorentzColourMatrix<vsimd> itype;
-    typedef LorentzColourMatrix sobj;
-    GridBase *grid = Umu._grid;
-
-    ILDGField header;
-    readHeader(header);
-
-    // now just the conf, ignore the header
-    std::string format = std::string("IEEE64BIG");
-    do {limeReaderNextRecord(LimeR);}
-    while (strncmp(limeReaderType(LimeR), "ildg-binary-data",16));
-
-    n_uint64_t nbytes = limeReaderBytes(LimeR);//size of this record (configuration)
-
-
-    ILDGtype ILDGt(true, LimeR);
-    // this is special for double prec data, just for the moment
-    uint32_t csum = BinaryIO::readObjectParallel< itype, sobjd >(
-       Umu, filename, ILDGMunger<sobjd, sobj>(), 0, format, ILDGt);
-
-    // Check configuration 
-    // todo
-
-    return csum;
-  }
-
-  template <class vsimd>
-  uint32_t writeConfiguration(Lattice<iLorentzColourMatrix<vsimd> > &Umu, std::string format) {
+  void writeConfiguration(Lattice<iLorentzColourMatrix<vsimd> > &Umu,int sequence,std::string LFN,std::string description) 
+  {
+    GridBase * grid = Umu._grid;
     typedef Lattice<iLorentzColourMatrix<vsimd> > GaugeField;
     typedef iLorentzColourMatrix<vsimd> vobj;
     typedef typename vobj::scalar_object sobj;
-    typedef LorentzColourMatrixD fobj;
 
-    ILDGField header;
-    // fill the header
-    header.floating_point = format;
+    uint64_t nbytes;
 
-    ILDGUnmunger<fobj, sobj> munge;
-    unsigned int offset = writeHeader(header);
+    ////////////////////////////////////////
+    // fill the Grid header
+    ////////////////////////////////////////
+    FieldMetaData header;
+    scidacRecord  _scidacRecord;
+    scidacFile    _scidacFile;
 
-    BinaryIO::Uint32Checksum<vobj, fobj>(Umu, munge, header.checksum);
+    ScidacMetaData(Umu,header,_scidacRecord,_scidacFile);
 
-    // Write data record header
-    n_uint64_t PayloadSize = sizeof(fobj) * Umu._grid->_gsites;
-    createHeader("ildg-binary-data", 0, 1, PayloadSize, LimeW);
+    std::string format = header.floating_point;
+    header.ensemble_id    = description;
+    header.ensemble_label = description;
+    header.sequence_number = sequence;
+    header.ildg_lfn = LFN;
 
-    ILDGtype ILDGt(true, LimeW);
-    uint32_t csum = BinaryIO::writeObjectParallel<vobj, fobj>(
-       Umu, filename, munge, 0, header.floating_point, ILDGt);
+    assert ( (format == std::string("IEEE32BIG"))  
+           ||(format == std::string("IEEE64BIG")) );
 
-    limeWriterCloseRecord(LimeW);
+    //////////////////////////////////////////////////////
+    // Fill ILDG header data struct
+    //////////////////////////////////////////////////////
+    ildgFormat ildgfmt ;
+    ildgfmt.field     = std::string("su3gauge");
 
-    // Last record
-    // the logical file name LNF
-    // look into documentation on how to generate this string
-    std::string LNF = "empty"; 
+    if ( format == std::string("IEEE32BIG") ) { 
+      ildgfmt.precision = 32;
+    } else { 
+      ildgfmt.precision = 64;
+    }
+    ildgfmt.version = 1.0;
+    ildgfmt.lx = header.dimension[0];
+    ildgfmt.ly = header.dimension[1];
+    ildgfmt.lz = header.dimension[2];
+    ildgfmt.lt = header.dimension[3];
+    assert(header.nd==4);
+    assert(header.nd==header.dimension.size());
 
+    //////////////////////////////////////////////////////////////////////////////
+    // Fill the USQCD info field
+    //////////////////////////////////////////////////////////////////////////////
+    usqcdInfo info;
+    info.version=1.0;
+    info.plaq   = header.plaquette;
+    info.linktr = header.link_trace;
 
-    PayloadSize = sizeof(LNF);
-    createHeader("ildg-binary-lfn", 1 , 1, PayloadSize, LimeW);
-    limeWriteRecordData(const_cast<char*>(LNF.c_str()), &PayloadSize, LimeW);
-
-    limeWriterCloseRecord(LimeW);
-
-    return csum;
+    std::cout << GridLogMessage << " Writing config; IldgIO "<<std::endl;
+    //////////////////////////////////////////////
+    // Fill the Lime file record by record
+    //////////////////////////////////////////////
+    writeLimeObject(1,0,header ,std::string("FieldMetaData"),std::string(GRID_FORMAT)); // Open message 
+    writeLimeObject(0,0,_scidacFile,_scidacFile.SerialisableClassName(),std::string(SCIDAC_PRIVATE_FILE_XML));
+    writeLimeObject(0,1,info,info.SerialisableClassName(),std::string(SCIDAC_FILE_XML));
+    writeLimeObject(1,0,_scidacRecord,_scidacRecord.SerialisableClassName(),std::string(SCIDAC_PRIVATE_RECORD_XML));
+    writeLimeObject(0,0,info,info.SerialisableClassName(),std::string(SCIDAC_RECORD_XML));
+    writeLimeObject(0,0,ildgfmt,std::string("ildgFormat")   ,std::string(ILDG_FORMAT)); // rec
+    writeLimeIldgLFN(header.ildg_lfn);                                                 // rec
+    writeLimeLatticeBinaryObject(Umu,std::string(ILDG_BINARY_DATA));      // Closes message with checksum
+    //    limeDestroyWriter(LimeW);
+    fclose(File);
   }
-
-  // format for RNG? Now just binary out
 };
-}
-}
+
+class IldgReader : public GridLimeReader {
+ public:
+
+  ////////////////////////////////////////////////////////////////
+  // Read either Grid/SciDAC/ILDG configuration
+  // Don't require scidac records EXCEPT checksum
+  // Use Grid MetaData object if present.
+  // Else use ILDG MetaData object if present.
+  // Else use SciDAC MetaData object if present.
+  ////////////////////////////////////////////////////////////////
+  template <class vsimd>
+  void readConfiguration(Lattice<iLorentzColourMatrix<vsimd> > &Umu, FieldMetaData &FieldMetaData_) {
+
+    typedef Lattice<iLorentzColourMatrix<vsimd> > GaugeField;
+    typedef typename GaugeField::vector_object  vobj;
+    typedef typename vobj::scalar_object sobj;
+
+    typedef LorentzColourMatrixF fobj;
+    typedef LorentzColourMatrixD dobj;
+
+    GridBase *grid = Umu._grid;
+
+    std::vector<int> dims = Umu._grid->FullDimensions();
+
+    assert(dims.size()==4);
+
+    // Metadata holders
+    ildgFormat     ildgFormat_    ;
+    std::string    ildgLFN_       ;
+    scidacChecksum scidacChecksum_; 
+    usqcdInfo      usqcdInfo_     ;
+
+    // track what we read from file
+    int found_ildgFormat    =0;
+    int found_ildgLFN       =0;
+    int found_scidacChecksum=0;
+    int found_usqcdInfo     =0;
+    int found_ildgBinary =0;
+    int found_FieldMetaData =0;
+
+    uint32_t nersc_csum;
+    uint32_t scidac_csuma;
+    uint32_t scidac_csumb;
+
+    // Binary format
+    std::string format;
+
+    //////////////////////////////////////////////////////////////////////////
+    // Loop over all records
+    // -- Order is poorly guaranteed except ILDG header preceeds binary section.
+    // -- Run like an event loop.
+    // -- Impose trust hierarchy. Grid takes precedence & look for ILDG, and failing
+    //    that Scidac. 
+    // -- Insist on Scidac checksum record.
+    //////////////////////////////////////////////////////////////////////////
+
+    while ( limeReaderNextRecord(LimeR) == LIME_SUCCESS ) { 
+
+      uint64_t nbytes = limeReaderBytes(LimeR);//size of this record (configuration)
+      
+      //////////////////////////////////////////////////////////////////
+      // If not BINARY_DATA read a string and parse
+      //////////////////////////////////////////////////////////////////
+      if ( strncmp(limeReaderType(LimeR), ILDG_BINARY_DATA,strlen(ILDG_BINARY_DATA) )  ) {
+	
+	// Copy out the string
+	std::vector<char> xmlc(nbytes+1,'\0');
+	limeReaderReadData((void *)&xmlc[0], &nbytes, LimeR);    
+	std::cout << GridLogMessage<< "Non binary record :" <<limeReaderType(LimeR) <<std::endl; //<<"\n"<<(&xmlc[0])<<std::endl;
+
+	//////////////////////////////////
+	// ILDG format record
+	if ( !strncmp(limeReaderType(LimeR), ILDG_FORMAT,strlen(ILDG_FORMAT)) ) { 
+
+	  XmlReader RD(&xmlc[0],"");
+	  read(RD,"ildgFormat",ildgFormat_);
+
+	  if ( ildgFormat_.precision == 64 ) format = std::string("IEEE64BIG");
+	  if ( ildgFormat_.precision == 32 ) format = std::string("IEEE32BIG");
+
+	  assert( ildgFormat_.lx == dims[0]);
+	  assert( ildgFormat_.ly == dims[1]);
+	  assert( ildgFormat_.lz == dims[2]);
+	  assert( ildgFormat_.lt == dims[3]);
+
+	  found_ildgFormat = 1;
+	}
+
+	if ( !strncmp(limeReaderType(LimeR), ILDG_DATA_LFN,strlen(ILDG_DATA_LFN)) ) {
+	  FieldMetaData_.ildg_lfn = std::string(&xmlc[0]);
+	  found_ildgLFN = 1;
+	}
+
+	if ( !strncmp(limeReaderType(LimeR), GRID_FORMAT,strlen(ILDG_FORMAT)) ) { 
+
+	  XmlReader RD(&xmlc[0],"");
+	  read(RD,"FieldMetaData",FieldMetaData_);
+
+	  format = FieldMetaData_.floating_point;
+
+	  assert(FieldMetaData_.dimension[0] == dims[0]);
+	  assert(FieldMetaData_.dimension[1] == dims[1]);
+	  assert(FieldMetaData_.dimension[2] == dims[2]);
+	  assert(FieldMetaData_.dimension[3] == dims[3]);
+
+	  found_FieldMetaData = 1;
+	}
+
+	if ( !strncmp(limeReaderType(LimeR), SCIDAC_RECORD_XML,strlen(SCIDAC_RECORD_XML)) ) { 
+	  std::string xmls(&xmlc[0]);
+	  // is it a USQCD info field
+	  if ( xmls.find(std::string("usqcdInfo")) != std::string::npos ) { 
+	    std::cout << GridLogMessage<<"...found a usqcdInfo field"<<std::endl;
+	    XmlReader RD(&xmlc[0],"");
+	    read(RD,"usqcdInfo",usqcdInfo_);
+	    found_usqcdInfo = 1;
+	  }
+	}
+
+	if ( !strncmp(limeReaderType(LimeR), SCIDAC_CHECKSUM,strlen(SCIDAC_CHECKSUM)) ) { 
+	  XmlReader RD(&xmlc[0],"");
+	  read(RD,"scidacChecksum",scidacChecksum_);
+	  found_scidacChecksum = 1;
+	}
+
+      } else {  
+	/////////////////////////////////
+	// Binary data
+	/////////////////////////////////
+	std::cout << GridLogMessage << "ILDG Binary record found : "  ILDG_BINARY_DATA << std::endl;
+	off_t offset= ftell(File);
+
+	if ( format == std::string("IEEE64BIG") ) {
+	  GaugeSimpleMunger<dobj, sobj> munge;
+	  BinaryIO::readLatticeObject< vobj, dobj >(Umu, filename, munge, offset, format,nersc_csum,scidac_csuma,scidac_csumb);
+	} else { 
+	  GaugeSimpleMunger<fobj, sobj> munge;
+	  BinaryIO::readLatticeObject< vobj, fobj >(Umu, filename, munge, offset, format,nersc_csum,scidac_csuma,scidac_csumb);
+	}
+
+	found_ildgBinary = 1;
+      }
+
+    }
+
+    //////////////////////////////////////////////////////
+    // Minimally must find binary segment and checksum
+    // Since this is an ILDG reader require ILDG format
+    //////////////////////////////////////////////////////
+    assert(found_ildgBinary);
+    assert(found_ildgFormat);
+    assert(found_scidacChecksum);
+
+    // Must find something with the lattice dimensions
+    assert(found_FieldMetaData||found_ildgFormat);
+
+    if ( found_FieldMetaData ) {
+
+      std::cout << GridLogMessage<<"Grid MetaData was record found: configuration was probably written by Grid ! Yay ! "<<std::endl;
+
+    } else { 
+
+      assert(found_ildgFormat);
+      assert ( ildgFormat_.field == std::string("su3gauge") );
+
+      ///////////////////////////////////////////////////////////////////////////////////////
+      // Populate our Grid metadata as best we can
+      ///////////////////////////////////////////////////////////////////////////////////////
+
+      std::ostringstream vers; vers << ildgFormat_.version;
+      FieldMetaData_.hdr_version = vers.str();
+      FieldMetaData_.data_type = std::string("4D_SU3_GAUGE_3X3");
+
+      FieldMetaData_.nd=4;
+      FieldMetaData_.dimension.resize(4);
+
+      FieldMetaData_.dimension[0] = ildgFormat_.lx ;
+      FieldMetaData_.dimension[1] = ildgFormat_.ly ;
+      FieldMetaData_.dimension[2] = ildgFormat_.lz ;
+      FieldMetaData_.dimension[3] = ildgFormat_.lt ;
+
+      if ( found_usqcdInfo ) { 
+	FieldMetaData_.plaquette = usqcdInfo_.plaq;
+	FieldMetaData_.link_trace= usqcdInfo_.linktr;
+	std::cout << GridLogMessage <<"This configuration was probably written by USQCD "<<std::endl;
+	std::cout << GridLogMessage <<"USQCD xml record Plaquette : "<<FieldMetaData_.plaquette<<std::endl;
+	std::cout << GridLogMessage <<"USQCD xml record LinkTrace : "<<FieldMetaData_.link_trace<<std::endl;
+      } else { 
+	FieldMetaData_.plaquette = 0.0;
+	FieldMetaData_.link_trace= 0.0;
+	std::cout << GridLogWarning << "This configuration is unsafe with no plaquette records that can verify it !!! "<<std::endl;
+      }
+    }
+
+    ////////////////////////////////////////////////////////////
+    // Really really want to mandate a scidac checksum
+    ////////////////////////////////////////////////////////////
+    if ( found_scidacChecksum ) {
+      FieldMetaData_.scidac_checksuma = stoull(scidacChecksum_.suma,0,16);
+      FieldMetaData_.scidac_checksumb = stoull(scidacChecksum_.sumb,0,16);
+      scidacChecksumVerify(scidacChecksum_,scidac_csuma,scidac_csumb);
+      assert( scidac_csuma ==FieldMetaData_.scidac_checksuma);
+      assert( scidac_csumb ==FieldMetaData_.scidac_checksumb);
+      std::cout << GridLogMessage<<"SciDAC checksums match " << std::endl;
+    } else { 
+      std::cout << GridLogWarning<<"SciDAC checksums not found. This is unsafe. " << std::endl;
+      assert(0); // Can I insist always checksum ?
+    }
+
+    if ( found_FieldMetaData || found_usqcdInfo ) {
+      FieldMetaData checker;
+      GaugeStatistics(Umu,checker);
+      assert(fabs(checker.plaquette  - FieldMetaData_.plaquette )<1.0e-5);
+      assert(fabs(checker.link_trace - FieldMetaData_.link_trace)<1.0e-5);
+      std::cout << GridLogMessage<<"Plaquette and link trace match " << std::endl;
+    }
+  }
+ };
+
+}}
 
 //HAVE_LIME
 #endif

lib/parallelIO/IldgIOtypes.h

@@ -34,47 +34,198 @@ extern "C" { // for linkage
 
 namespace Grid {
 
-struct ILDGtype {
-  bool is_ILDG;
-  LimeWriter* LW;
-  LimeReader* LR;
+/////////////////////////////////////////////////////////////////////////////////
+// Data representation of records that enter ILDG and SciDac formats
+/////////////////////////////////////////////////////////////////////////////////
 
-  ILDGtype(bool is, LimeWriter* L) : is_ILDG(is), LW(L), LR(NULL) {}
-  ILDGtype(bool is, LimeReader* L) : is_ILDG(is), LW(NULL), LR(L) {}
-  ILDGtype() : is_ILDG(false), LW(NULL), LR(NULL) {}
-};
+#define GRID_FORMAT      "grid-format"
+#define ILDG_FORMAT      "ildg-format"
+#define ILDG_BINARY_DATA "ildg-binary-data"
+#define ILDG_DATA_LFN    "ildg-data-lfn"
+#define SCIDAC_CHECKSUM           "scidac-checksum"
+#define SCIDAC_PRIVATE_FILE_XML   "scidac-private-file-xml"
+#define SCIDAC_FILE_XML           "scidac-file-xml"
+#define SCIDAC_PRIVATE_RECORD_XML "scidac-private-record-xml"
+#define SCIDAC_RECORD_XML         "scidac-record-xml"
+#define SCIDAC_BINARY_DATA        "scidac-binary-data"
+// Unused SCIDAC records names; could move to support this functionality
+#define SCIDAC_SITELIST           "scidac-sitelist"
 
-class ILDGField {
+  ////////////////////////////////////////////////////////////
+  const int GRID_IO_SINGLEFILE = 0; // hardcode lift from QIO compat
+  const int GRID_IO_MULTIFILE  = 1; // hardcode lift from QIO compat
+  const int GRID_IO_FIELD      = 0; // hardcode lift from QIO compat
+  const int GRID_IO_GLOBAL     = 1; // hardcode lift from QIO compat
+  ////////////////////////////////////////////////////////////
+
+/////////////////////////////////////////////////////////////////////////////////
+// QIO uses mandatory "private" records fixed format
+// Private is in principle "opaque" however it can't be changed now because that would break existing 
+// file compatability, so should be correct to assume the undocumented but defacto file structure.
+/////////////////////////////////////////////////////////////////////////////////
+
+////////////////////////
+// Scidac private file xml
+// <?xml version="1.0" encoding="UTF-8"?><scidacFile><version>1.1</version><spacetime>4</spacetime><dims>16 16 16 32 </dims><volfmt>0</volfmt></scidacFile>
+////////////////////////
+struct scidacFile : Serializable {
  public:
-  // header strings (not in order)
-  std::vector<int> dimension;
-  std::vector<std::string> boundary;
-  int data_start;
-  std::string hdr_version;
-  std::string storage_format;
-  // Checks on data
-  double link_trace;
-  double plaquette;
-  uint32_t checksum;
-  unsigned int sequence_number;
-  std::string data_type;
-  std::string ensemble_id;
-  std::string ensemble_label;
-  std::string creator;
-  std::string creator_hardware;
-  std::string creation_date;
-  std::string archive_date;
-  std::string floating_point;
-};
-}
-#else
-namespace Grid {
+  GRID_SERIALIZABLE_CLASS_MEMBERS(scidacFile,
+                                  double, version,
+                                  int, spacetime,
+				  std::string, dims, // must convert to int
+                                  int, volfmt);
 
-struct ILDGtype {
-  bool is_ILDG;
-  ILDGtype() : is_ILDG(false) {}
-};
-}
+  std::vector<int> getDimensions(void) { 
+    std::stringstream stream(dims);
+    std::vector<int> dimensions;
+    int n;
+    while(stream >> n){
+      dimensions.push_back(n);
+    }
+    return dimensions;
+  }
 
+  void setDimensions(std::vector<int> dimensions) { 
+    char delimiter = ' ';
+    std::stringstream stream;
+    for(int i=0;i<dimensions.size();i++){ 
+      stream << dimensions[i];
+      if ( i != dimensions.size()-1) { 
+	stream << delimiter <<std::endl;
+      }
+    }
+    dims = stream.str();
+  }
+
+  // Constructor provides Grid
+  scidacFile() =default; // default constructor
+  scidacFile(GridBase * grid){
+    version      = 1.0;
+    spacetime    = grid->_ndimension;
+    setDimensions(grid->FullDimensions()); 
+    volfmt       = GRID_IO_SINGLEFILE;
+  }
+
+};
+
+///////////////////////////////////////////////////////////////////////
+// scidac-private-record-xml : example
+// <scidacRecord>
+// <version>1.1</version><date>Tue Jul 26 21:14:44 2011 UTC</date><recordtype>0</recordtype>
+// <datatype>QDP_D3_ColorMatrix</datatype><precision>D</precision><colors>3</colors><spins>4</spins>
+// <typesize>144</typesize><datacount>4</datacount>
+// </scidacRecord>
+///////////////////////////////////////////////////////////////////////
+
+struct scidacRecord : Serializable {
+ public:
+  GRID_SERIALIZABLE_CLASS_MEMBERS(scidacRecord,
+                                  double, version,
+                                  std::string, date,
+				  int, recordtype,
+				  std::string, datatype,
+				  std::string, precision,
+				  int, colors,
+				  int, spins,
+				  int, typesize,
+				  int, datacount);
+
+  scidacRecord() { version =1.0; }
+
+};
+
+////////////////////////
+// ILDG format
+////////////////////////
+struct ildgFormat : Serializable {
+public:
+  GRID_SERIALIZABLE_CLASS_MEMBERS(ildgFormat,
+				  double, version,
+				  std::string, field,
+				  int, precision,
+				  int, lx,
+				  int, ly,
+				  int, lz,
+				  int, lt);
+  ildgFormat() { version=1.0; };
+};
+////////////////////////
+// USQCD info
+////////////////////////
+struct usqcdInfo : Serializable { 
+ public:
+  GRID_SERIALIZABLE_CLASS_MEMBERS(usqcdInfo,
+				  double, version,
+				  double, plaq,
+				  double, linktr,
+				  std::string, info);
+  usqcdInfo() { 
+    version=1.0; 
+  };
+};
+////////////////////////
+// Scidac Checksum
+////////////////////////
+struct scidacChecksum : Serializable { 
+ public:
+  GRID_SERIALIZABLE_CLASS_MEMBERS(scidacChecksum,
+				  double, version,
+				  std::string, suma,
+				  std::string, sumb);
+  scidacChecksum() { 
+    version=1.0; 
+  };
+};
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+// Type:           scidac-file-xml         <title>MILC ILDG archival gauge configuration</title>
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+// Type:           
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+
+////////////////////////
+// Scidac private file xml 
+// <?xml version="1.0" encoding="UTF-8"?><scidacFile><version>1.1</version><spacetime>4</spacetime><dims>16 16 16 32 </dims><volfmt>0</volfmt></scidacFile> 
+////////////////////////                                                                                                                                                                              
+
+#if 0
+////////////////////////////////////////////////////////////////////////////////////////
+// From http://www.physics.utah.edu/~detar/scidac/qio_2p3.pdf
+////////////////////////////////////////////////////////////////////////////////////////
+struct usqcdPropFile : Serializable { 
+ public:
+  GRID_SERIALIZABLE_CLASS_MEMBERS(usqcdPropFile,
+				  double, version,
+				  std::string, type,
+				  std::string, info);
+  usqcdPropFile() { 
+    version=1.0; 
+  };
+};
+struct usqcdSourceInfo : Serializable { 
+ public:
+  GRID_SERIALIZABLE_CLASS_MEMBERS(usqcdSourceInfo,
+				  double, version,
+				  std::string, info);
+  usqcdSourceInfo() { 
+    version=1.0; 
+  };
+};
+struct usqcdPropInfo : Serializable { 
+ public:
+  GRID_SERIALIZABLE_CLASS_MEMBERS(usqcdPropInfo,
+				  double, version,
+				  int, spin,
+				  int, color,
+				  std::string, info);
+  usqcdPropInfo() { 
+    version=1.0; 
+  };
+};
+#endif
+
+}
 #endif
 #endif

lib/parallelIO/MetaData.h

@@ -0,0 +1,325 @@
+/*************************************************************************************
+
+    Grid physics library, www.github.com/paboyle/Grid 
+
+    Source file: ./lib/parallelIO/NerscIO.h
+
+    Copyright (C) 2015
+
+
+    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 <algorithm>
+#include <iostream>
+#include <iomanip>
+#include <fstream>
+#include <map>
+#include <unistd.h>
+#include <sys/utsname.h>
+#include <pwd.h>
+
+namespace Grid {
+
+  ///////////////////////////////////////////////////////
+  // Precision mapping
+  ///////////////////////////////////////////////////////
+  template<class vobj> static std::string getFormatString (void)
+  {
+    std::string format;
+    typedef typename getPrecision<vobj>::real_scalar_type stype;
+    if ( sizeof(stype) == sizeof(float) ) {
+      format = std::string("IEEE32BIG");
+    }
+    if ( sizeof(stype) == sizeof(double) ) {
+      format = std::string("IEEE64BIG");
+    }
+    return format;
+  }
+  ////////////////////////////////////////////////////////////////////////////////
+  // header specification/interpretation
+  ////////////////////////////////////////////////////////////////////////////////
+    class FieldMetaData : Serializable {
+    public:
+
+      GRID_SERIALIZABLE_CLASS_MEMBERS(FieldMetaData,
+				      int, nd,
+				      std::vector<int>, dimension,
+				      std::vector<std::string>, boundary,
+				      int, data_start,
+				      std::string, hdr_version,
+				      std::string, storage_format,
+				      double, link_trace,
+				      double, plaquette,
+				      uint32_t, checksum,
+				      uint32_t, scidac_checksuma,
+				      uint32_t, scidac_checksumb,
+				      unsigned int, sequence_number,
+				      std::string, data_type,
+				      std::string, ensemble_id,
+				      std::string, ensemble_label,
+				      std::string, ildg_lfn,
+				      std::string, creator,
+				      std::string, creator_hardware,
+				      std::string, creation_date,
+				      std::string, archive_date,
+				      std::string, floating_point);
+      FieldMetaData(void) { 
+	nd=4;
+	dimension.resize(4);
+	boundary.resize(4);
+      }
+    };
+
+
+
+  namespace QCD {
+
+    using namespace Grid;
+
+
+    //////////////////////////////////////////////////////////////////////
+    // Bit and Physical Checksumming and QA of data
+    //////////////////////////////////////////////////////////////////////
+    inline void GridMetaData(GridBase *grid,FieldMetaData &header)
+    {
+      int nd = grid->_ndimension;
+      header.nd = nd;
+      header.dimension.resize(nd);
+      header.boundary.resize(nd);
+      for(int d=0;d<nd;d++) {
+	header.dimension[d] = grid->_fdimensions[d];
+      }
+      for(int d=0;d<nd;d++) {
+	header.boundary[d] = std::string("PERIODIC");
+      }
+    }
+
+    inline void MachineCharacteristics(FieldMetaData &header)
+    {
+      // Who
+      struct passwd *pw = getpwuid (getuid());
+      if (pw) header.creator = std::string(pw->pw_name); 
+
+      // When
+      std::time_t t = std::time(nullptr);
+      std::tm tm_ = *std::localtime(&t);
+      std::ostringstream oss; 
+      //      oss << std::put_time(&tm_, "%c %Z");
+      header.creation_date = oss.str();
+      header.archive_date  = header.creation_date;
+
+      // What
+      struct utsname name;  uname(&name);
+      header.creator_hardware = std::string(name.nodename)+"-";
+      header.creator_hardware+= std::string(name.machine)+"-";
+      header.creator_hardware+= std::string(name.sysname)+"-";
+      header.creator_hardware+= std::string(name.release);
+    }
+
+#define dump_meta_data(field, s)					\
+      s << "BEGIN_HEADER"      << std::endl;				\
+      s << "HDR_VERSION = "    << field.hdr_version    << std::endl;	\
+      s << "DATATYPE = "       << field.data_type      << std::endl;	\
+      s << "STORAGE_FORMAT = " << field.storage_format << std::endl;	\
+      for(int i=0;i<4;i++){						\
+	s << "DIMENSION_" << i+1 << " = " << field.dimension[i] << std::endl ; \
+      }									\
+      s << "LINK_TRACE = " << std::setprecision(10) << field.link_trace << std::endl; \
+      s << "PLAQUETTE  = " << std::setprecision(10) << field.plaquette  << std::endl; \
+      for(int i=0;i<4;i++){						\
+	s << "BOUNDARY_"<<i+1<<" = " << field.boundary[i] << std::endl;	\
+      }									\
+									\
+      s << "CHECKSUM = "<< std::hex << std::setw(10) << field.checksum << std::dec<<std::endl; \
+      s << "SCIDAC_CHECKSUMA = "<< std::hex << std::setw(10) << field.scidac_checksuma << std::dec<<std::endl; \
+      s << "SCIDAC_CHECKSUMB = "<< std::hex << std::setw(10) << field.scidac_checksumb << std::dec<<std::endl; \
+      s << "ENSEMBLE_ID = "     << field.ensemble_id      << std::endl;	\
+      s << "ENSEMBLE_LABEL = "  << field.ensemble_label   << std::endl;	\
+      s << "SEQUENCE_NUMBER = " << field.sequence_number  << std::endl;	\
+      s << "CREATOR = "         << field.creator          << std::endl;	\
+      s << "CREATOR_HARDWARE = "<< field.creator_hardware << std::endl;	\
+      s << "CREATION_DATE = "   << field.creation_date    << std::endl;	\
+      s << "ARCHIVE_DATE = "    << field.archive_date     << std::endl;	\
+      s << "FLOATING_POINT = "  << field.floating_point   << std::endl;	\
+      s << "END_HEADER"         << std::endl;
+
+template<class vobj> inline void PrepareMetaData(Lattice<vobj> & field, FieldMetaData &header)
+{
+  GridBase *grid = field._grid;
+  std::string format = getFormatString<vobj>();
+   header.floating_point = format;
+   header.checksum = 0x0; // Nersc checksum unused in ILDG, Scidac
+   GridMetaData(grid,header); 
+   MachineCharacteristics(header);
+ }
+ inline void GaugeStatistics(Lattice<vLorentzColourMatrixF> & data,FieldMetaData &header)
+ {
+   // How to convert data precision etc...
+   header.link_trace=Grid::QCD::WilsonLoops<PeriodicGimplF>::linkTrace(data);
+   header.plaquette =Grid::QCD::WilsonLoops<PeriodicGimplF>::avgPlaquette(data);
+ }
+ inline void GaugeStatistics(Lattice<vLorentzColourMatrixD> & data,FieldMetaData &header)
+ {
+   // How to convert data precision etc...
+   header.link_trace=Grid::QCD::WilsonLoops<PeriodicGimplD>::linkTrace(data);
+   header.plaquette =Grid::QCD::WilsonLoops<PeriodicGimplD>::avgPlaquette(data);
+ }
+ template<> inline void PrepareMetaData<vLorentzColourMatrixF>(Lattice<vLorentzColourMatrixF> & field, FieldMetaData &header)
+ {
+   
+   GridBase *grid = field._grid;
+   std::string format = getFormatString<vLorentzColourMatrixF>();
+   header.floating_point = format;
+   header.checksum = 0x0; // Nersc checksum unused in ILDG, Scidac
+   GridMetaData(grid,header); 
+   GaugeStatistics(field,header);
+   MachineCharacteristics(header);
+ }
+ template<> inline void PrepareMetaData<vLorentzColourMatrixD>(Lattice<vLorentzColourMatrixD> & field, FieldMetaData &header)
+ {
+   GridBase *grid = field._grid;
+   std::string format = getFormatString<vLorentzColourMatrixD>();
+   header.floating_point = format;
+   header.checksum = 0x0; // Nersc checksum unused in ILDG, Scidac
+   GridMetaData(grid,header); 
+   GaugeStatistics(field,header);
+   MachineCharacteristics(header);
+ }
+
+    //////////////////////////////////////////////////////////////////////
+    // Utilities ; these are QCD aware
+    //////////////////////////////////////////////////////////////////////
+    inline void reconstruct3(LorentzColourMatrix & cm)
+    {
+      const int x=0;
+      const int y=1;
+      const int z=2;
+      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
+      }
+    }
+
+    ////////////////////////////////////////////////////////////////////////////////
+    // Some data types for intermediate storage
+    ////////////////////////////////////////////////////////////////////////////////
+    template<typename vtype> using iLorentzColour2x3 = iVector<iVector<iVector<vtype, Nc>, 2>, Nd >;
+
+    typedef iLorentzColour2x3<Complex>  LorentzColour2x3;
+    typedef iLorentzColour2x3<ComplexF> LorentzColour2x3F;
+    typedef iLorentzColour2x3<ComplexD> LorentzColour2x3D;
+
+/////////////////////////////////////////////////////////////////////////////////
+// Simple classes for precision conversion
+/////////////////////////////////////////////////////////////////////////////////
+template <class fobj, class sobj>
+struct BinarySimpleUnmunger {
+  typedef typename getPrecision<fobj>::real_scalar_type fobj_stype;
+  typedef typename getPrecision<sobj>::real_scalar_type sobj_stype;
+  
+  void operator()(sobj &in, fobj &out) {
+    // take word by word and transform accoding to the status
+    fobj_stype *out_buffer = (fobj_stype *)&out;
+    sobj_stype *in_buffer = (sobj_stype *)&in;
+    size_t fobj_words = sizeof(out) / sizeof(fobj_stype);
+    size_t sobj_words = sizeof(in) / sizeof(sobj_stype);
+    assert(fobj_words == sobj_words);
+    
+    for (unsigned int word = 0; word < sobj_words; word++)
+      out_buffer[word] = in_buffer[word];  // type conversion on the fly
+    
+  }
+};
+
+template <class fobj, class sobj>
+struct BinarySimpleMunger {
+  typedef typename getPrecision<fobj>::real_scalar_type fobj_stype;
+  typedef typename getPrecision<sobj>::real_scalar_type sobj_stype;
+
+  void operator()(fobj &in, sobj &out) {
+    // take word by word and transform accoding to the status
+    fobj_stype *in_buffer = (fobj_stype *)&in;
+    sobj_stype *out_buffer = (sobj_stype *)&out;
+    size_t fobj_words = sizeof(in) / sizeof(fobj_stype);
+    size_t sobj_words = sizeof(out) / sizeof(sobj_stype);
+    assert(fobj_words == sobj_words);
+    
+    for (unsigned int word = 0; word < sobj_words; word++)
+      out_buffer[word] = in_buffer[word];  // type conversion on the fly
+    
+  }
+};
+
+
+    template<class fobj,class sobj>
+    struct GaugeSimpleMunger{
+      void operator()(fobj &in, sobj &out) {
+        for (int mu = 0; mu < Nd; mu++) {
+          for (int i = 0; i < Nc; i++) {
+          for (int j = 0; j < Nc; j++) {
+	    out(mu)()(i, j) = in(mu)()(i, j);
+	  }}
+        }
+      };
+    };
+
+    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++) {
+          for (int j = 0; j < Nc; j++) {
+	    out(mu)()(i, j) = in(mu)()(i, j);
+	  }}
+        }
+      };
+    };
+
+    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++){
+	    out(mu)()(i,j) = in(mu)(i)(j);
+	  }}
+	}
+	reconstruct3(out);
+      }
+    };
+
+    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++){
+	    out(mu)(i)(j) = in(mu)()(i,j);
+	  }}
+	}
+      }
+    };
+  }
+
+
+}

lib/parallelIO/NerscIO.h

@@ -30,182 +30,11 @@
 #ifndef GRID_NERSC_IO_H
 #define GRID_NERSC_IO_H
 
-#include <algorithm>
-#include <iostream>
-#include <iomanip>
-#include <fstream>
-#include <map>
-
-#include <unistd.h>
-#include <sys/utsname.h>
-#include <pwd.h>
-
 namespace Grid {
   namespace QCD {
 
     using namespace Grid;
 
-    ////////////////////////////////////////////////////////////////////////////////
-    // Some data types for intermediate storage
-    ////////////////////////////////////////////////////////////////////////////////
-    template<typename vtype> using iLorentzColour2x3 = iVector<iVector<iVector<vtype, Nc>, 2>, 4 >;
-
-    typedef iLorentzColour2x3<Complex>  LorentzColour2x3;
-    typedef iLorentzColour2x3<ComplexF> LorentzColour2x3F;
-    typedef iLorentzColour2x3<ComplexD> LorentzColour2x3D;
-
-    ////////////////////////////////////////////////////////////////////////////////
-    // header specification/interpretation
-    ////////////////////////////////////////////////////////////////////////////////
-    class NerscField {
-    public:
-      // header strings (not in order)
-      int dimension[4];
-      std::string boundary[4]; 
-      int data_start;
-      std::string hdr_version;
-      std::string storage_format;
-      // Checks on data
-      double link_trace;
-      double plaquette;
-      uint32_t checksum;
-      unsigned int sequence_number;
-      std::string data_type;
-      std::string ensemble_id ;
-      std::string ensemble_label ;
-      std::string creator ;
-      std::string creator_hardware ;
-      std::string creation_date ;
-      std::string archive_date ;
-      std::string floating_point;
-    };
-
-    //////////////////////////////////////////////////////////////////////
-    // Bit and Physical Checksumming and QA of data
-    //////////////////////////////////////////////////////////////////////
-
-    inline void NerscGrid(GridBase *grid,NerscField &header)
-    {
-      assert(grid->_ndimension==4);
-      for(int d=0;d<4;d++) {
-	header.dimension[d] = grid->_fdimensions[d];
-      }
-      for(int d=0;d<4;d++) {
-	header.boundary[d] = std::string("PERIODIC");
-      }
-    }
-    template<class GaugeField>
-    inline void NerscStatistics(GaugeField & data,NerscField &header)
-    {
-      // How to convert data precision etc...
-      header.link_trace=Grid::QCD::WilsonLoops<PeriodicGimplR>::linkTrace(data);
-      header.plaquette =Grid::QCD::WilsonLoops<PeriodicGimplR>::avgPlaquette(data);
-    }
-
-    inline void NerscMachineCharacteristics(NerscField &header)
-    {
-      // Who
-      struct passwd *pw = getpwuid (getuid());
-      if (pw) header.creator = std::string(pw->pw_name); 
-
-      // When
-      std::time_t t = std::time(nullptr);
-      std::tm tm = *std::localtime(&t);
-      std::ostringstream oss; 
-      //  oss << std::put_time(&tm, "%c %Z");
-      header.creation_date = oss.str();
-      header.archive_date  = header.creation_date;
-
-      // What
-      struct utsname name;  uname(&name);
-      header.creator_hardware = std::string(name.nodename)+"-";
-      header.creator_hardware+= std::string(name.machine)+"-";
-      header.creator_hardware+= std::string(name.sysname)+"-";
-      header.creator_hardware+= std::string(name.release);
-
-    }
-    //////////////////////////////////////////////////////////////////////
-    // Utilities ; these are QCD aware
-    //////////////////////////////////////////////////////////////////////
-    inline void NerscChecksum(uint32_t *buf,uint32_t buf_size_bytes,uint32_t &csum)
-    {
-      BinaryIO::Uint32Checksum(buf,buf_size_bytes,csum);
-    }
-    inline void reconstruct3(LorentzColourMatrix & cm)
-    {
-      const int x=0;
-      const int y=1;
-      const int z=2;
-      for(int mu=0;mu<4;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
-      }
-    }
-
-    template<class fobj,class sobj>
-    struct NerscSimpleMunger{
-      void operator()(fobj &in, sobj &out, uint32_t &csum) {
-        for (int mu = 0; mu < Nd; mu++) {
-          for (int i = 0; i < Nc; i++) {
-            for (int j = 0; j < Nc; j++) {
-              out(mu)()(i, j) = in(mu)()(i, j);
-            }
-          }
-        }
-        NerscChecksum((uint32_t *)&in, sizeof(in), csum);
-      };
-    };
-
-    template <class fobj, class sobj>
-    struct NerscSimpleUnmunger {
-      void operator()(sobj &in, fobj &out, uint32_t &csum) {
-        for (int mu = 0; mu < Nd; mu++) {
-          for (int i = 0; i < Nc; i++) {
-            for (int j = 0; j < Nc; j++) {
-              out(mu)()(i, j) = in(mu)()(i, j);
-            }
-          }
-        }
-        NerscChecksum((uint32_t *)&out, sizeof(out), csum);
-      };
-    };
-
-    template<class fobj,class sobj>
-    struct Nersc3x2munger{
-      void operator() (fobj &in,sobj &out,uint32_t &csum){
-     
-	NerscChecksum((uint32_t *)&in,sizeof(in),csum); 
-
-	for(int mu=0;mu<4;mu++){
-	  for(int i=0;i<2;i++){
-	    for(int j=0;j<3;j++){
-	      out(mu)()(i,j) = in(mu)(i)(j);
-	    }}
-	}
-	reconstruct3(out);
-      }
-    };
-
-    template<class fobj,class sobj>
-    struct Nersc3x2unmunger{
-
-      void operator() (sobj &in,fobj &out,uint32_t &csum){
-
-
-	for(int mu=0;mu<4;mu++){
-	  for(int i=0;i<2;i++){
-	    for(int j=0;j<3;j++){
-	      out(mu)(i)(j) = in(mu)()(i,j);
-	    }}
-	}
-
-	NerscChecksum((uint32_t *)&out,sizeof(out),csum); 
-
-      }
-    };
-
-
     ////////////////////////////////////////////////////////////////////////////////
     // Write and read from fstream; comput header offset for payload
     ////////////////////////////////////////////////////////////////////////////////
@@ -216,42 +45,17 @@ namespace Grid {
 	std::ofstream fout(file,std::ios::out);
       }
   
-#define dump_nersc_header(field, s)					\
-      s << "BEGIN_HEADER"      << std::endl;				\
-      s << "HDR_VERSION = "    << field.hdr_version    << std::endl;	\
-      s << "DATATYPE = "       << field.data_type      << std::endl;	\
-      s << "STORAGE_FORMAT = " << field.storage_format << std::endl;	\
-      for(int i=0;i<4;i++){						\
-	s << "DIMENSION_" << i+1 << " = " << field.dimension[i] << std::endl ; \
-      }									\
-      s << "LINK_TRACE = " << std::setprecision(10) << field.link_trace << std::endl; \
-      s << "PLAQUETTE  = " << std::setprecision(10) << field.plaquette  << std::endl; \
-      for(int i=0;i<4;i++){						\
-	s << "BOUNDARY_"<<i+1<<" = " << field.boundary[i] << std::endl;	\
-      }									\
-									\
-      s << "CHECKSUM = "<< std::hex << std::setw(10) << field.checksum << std::dec<<std::endl; \
-      s << "ENSEMBLE_ID = "     << field.ensemble_id      << std::endl;	\
-      s << "ENSEMBLE_LABEL = "  << field.ensemble_label   << std::endl;	\
-      s << "SEQUENCE_NUMBER = " << field.sequence_number  << std::endl;	\
-      s << "CREATOR = "         << field.creator          << std::endl;	\
-      s << "CREATOR_HARDWARE = "<< field.creator_hardware << std::endl;	\
-      s << "CREATION_DATE = "   << field.creation_date    << std::endl;	\
-      s << "ARCHIVE_DATE = "    << field.archive_date     << std::endl;	\
-      s << "FLOATING_POINT = "  << field.floating_point   << std::endl;	\
-      s << "END_HEADER"         << std::endl;
-  
-      static inline unsigned int writeHeader(NerscField &field,std::string file)
+      static inline unsigned int writeHeader(FieldMetaData &field,std::string file)
       {
       std::ofstream fout(file,std::ios::out|std::ios::in);
       fout.seekp(0,std::ios::beg);
-      dump_nersc_header(field, fout);
+      dump_meta_data(field, fout);
       field.data_start = fout.tellp();
       return field.data_start;
     }
 
       // for the header-reader
-      static inline int readHeader(std::string file,GridBase *grid,  NerscField &field)
+      static inline int readHeader(std::string file,GridBase *grid,  FieldMetaData &field)
       {
       int offset=0;
       std::map<std::string,std::string> header;
@@ -323,21 +127,21 @@ namespace Grid {
       return field.data_start;
     }
 
-      /////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-      // Now the meat: the object readers
-      /////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-#define PARALLEL_READ
-#define PARALLEL_WRITE
+    /////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+    // Now the meat: the object readers
+    /////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 
-      template<class vsimd>
-      static inline void readConfiguration(Lattice<iLorentzColourMatrix<vsimd> > &Umu,NerscField& header,std::string file)
-      {
+    template<class vsimd>
+    static inline void readConfiguration(Lattice<iLorentzColourMatrix<vsimd> > &Umu,
+					 FieldMetaData& header,
+					 std::string file)
+    {
       typedef Lattice<iLorentzColourMatrix<vsimd> > GaugeField;
 
       GridBase *grid = Umu._grid;
       int offset = readHeader(file,Umu._grid,header);
 
-      NerscField clone(header);
+      FieldMetaData clone(header);
 
       std::string format(header.floating_point);
 
@@ -346,76 +150,78 @@ namespace Grid {
       int ieee64big = (format == std::string("IEEE64BIG"));
       int ieee64    = (format == std::string("IEEE64"));
 
-      uint32_t csum;
+      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") ) {
-      if ( ieee32 || ieee32big ) {
-#ifdef PARALLEL_READ
-	csum=BinaryIO::readObjectParallel<iLorentzColourMatrix<vsimd>, LorentzColour2x3F> 
-	  (Umu,file,Nersc3x2munger<LorentzColour2x3F,LorentzColourMatrix>(), offset,format);
-#else
-	csum=BinaryIO::readObjectSerial<iLorentzColourMatrix<vsimd>, LorentzColour2x3F> 
-	  (Umu,file,Nersc3x2munger<LorentzColour2x3F,LorentzColourMatrix>(), offset,format);
-#endif
-      }
-      if ( ieee64 || ieee64big ) {
-#ifdef PARALLEL_READ
-	csum=BinaryIO::readObjectParallel<iLorentzColourMatrix<vsimd>, LorentzColour2x3D> 
-	  (Umu,file,Nersc3x2munger<LorentzColour2x3D,LorentzColourMatrix>(),offset,format);
-#else 
-	csum=BinaryIO::readObjectSerial<iLorentzColourMatrix<vsimd>, LorentzColour2x3D> 
-	  (Umu,file,Nersc3x2munger<LorentzColour2x3D,LorentzColourMatrix>(),offset,format);
-#endif
-      }
-      } else if ( header.data_type == std::string("4D_SU3_GAUGE_3x3") ) {
 	if ( ieee32 || ieee32big ) {
-#ifdef PARALLEL_READ
-	  csum=BinaryIO::readObjectParallel<iLorentzColourMatrix<vsimd>,LorentzColourMatrixF>
-	    (Umu,file,NerscSimpleMunger<LorentzColourMatrixF,LorentzColourMatrix>(),offset,format);
-#else
-	  csum=BinaryIO::readObjectSerial<iLorentzColourMatrix<vsimd>,LorentzColourMatrixF>
-	    (Umu,file,NerscSimpleMunger<LorentzColourMatrixF,LorentzColourMatrix>(),offset,format);
-#endif
+	  BinaryIO::readLatticeObject<iLorentzColourMatrix<vsimd>, LorentzColour2x3F> 
+	    (Umu,file,Gauge3x2munger<LorentzColour2x3F,LorentzColourMatrix>(), offset,format,
+	     nersc_csum,scidac_csuma,scidac_csumb);
 	}
 	if ( ieee64 || ieee64big ) {
-#ifdef PARALLEL_READ
-	  csum=BinaryIO::readObjectParallel<iLorentzColourMatrix<vsimd>,LorentzColourMatrixD>
-	    (Umu,file,NerscSimpleMunger<LorentzColourMatrixD,LorentzColourMatrix>(),offset,format);
-#else
-	  csum=BinaryIO::readObjectSerial<iLorentzColourMatrix<vsimd>,LorentzColourMatrixD>
-	    (Umu,file,NerscSimpleMunger<LorentzColourMatrixD,LorentzColourMatrix>(),offset,format);
-#endif
+	  BinaryIO::readLatticeObject<iLorentzColourMatrix<vsimd>, LorentzColour2x3D> 
+	    (Umu,file,Gauge3x2munger<LorentzColour2x3D,LorentzColourMatrix>(),offset,format,
+	     nersc_csum,scidac_csuma,scidac_csumb);
+	}
+      } else if ( header.data_type == std::string("4D_SU3_GAUGE_3x3") ) {
+	if ( ieee32 || ieee32big ) {
+	  BinaryIO::readLatticeObject<iLorentzColourMatrix<vsimd>,LorentzColourMatrixF>
+	    (Umu,file,GaugeSimpleMunger<LorentzColourMatrixF,LorentzColourMatrix>(),offset,format,
+	     nersc_csum,scidac_csuma,scidac_csumb);
+	}
+	if ( ieee64 || ieee64big ) {
+	  BinaryIO::readLatticeObject<iLorentzColourMatrix<vsimd>,LorentzColourMatrixD>
+	    (Umu,file,GaugeSimpleMunger<LorentzColourMatrixD,LorentzColourMatrix>(),offset,format,
+	     nersc_csum,scidac_csuma,scidac_csumb);
 	}
       } else {
 	assert(0);
       }
 
-      NerscStatistics<GaugeField>(Umu,clone);
+      GaugeStatistics(Umu,clone);
 
-      std::cout<<GridLogMessage <<"NERSC Configuration "<<file<<" checksum "<<std::hex<<            csum<< std::dec
+      std::cout<<GridLogMessage <<"NERSC Configuration "<<file<<" checksum "<<std::hex<<nersc_csum<< std::dec
 	       <<" header   "<<std::hex<<header.checksum<<std::dec <<std::endl;
       std::cout<<GridLogMessage <<"NERSC Configuration "<<file<<" plaquette "<<clone.plaquette
 	       <<" header    "<<header.plaquette<<std::endl;
       std::cout<<GridLogMessage <<"NERSC Configuration "<<file<<" link_trace "<<clone.link_trace
 	       <<" header    "<<header.link_trace<<std::endl;
+
+      if ( fabs(clone.plaquette -header.plaquette ) >=  1.0e-5 ) { 
+	std::cout << " Plaquette mismatch "<<std::endl;
+	std::cout << Umu[0]<<std::endl;
+	std::cout << Umu[1]<<std::endl;
+      }
+      if ( nersc_csum != header.checksum ) { 
+	std::cerr << " checksum mismatch " << std::endl;
+	std::cerr << " plaqs " << clone.plaquette << " " << header.plaquette << std::endl;
+	std::cerr << " trace " << clone.link_trace<< " " << header.link_trace<< std::endl;
+	std::cerr << " nersc_csum  " <<std::hex<< nersc_csum << " " << header.checksum<< std::dec<< std::endl;
+	exit(0);
+      }
       assert(fabs(clone.plaquette -header.plaquette ) < 1.0e-5 );
       assert(fabs(clone.link_trace-header.link_trace) < 1.0e-6 );
-      assert(csum == header.checksum );
-
+      assert(nersc_csum == header.checksum );
+      
       std::cout<<GridLogMessage <<"NERSC Configuration "<<file<< " and plaquette, link trace, and checksum agree"<<std::endl;
-      }
+    }
 
       template<class vsimd>
-      static inline void writeConfiguration(Lattice<iLorentzColourMatrix<vsimd> > &Umu,std::string file, int two_row,int bits32)
+      static inline void writeConfiguration(Lattice<iLorentzColourMatrix<vsimd> > &Umu,
+					    std::string file, 
+					    int two_row,
+					    int bits32)
       {
 	typedef Lattice<iLorentzColourMatrix<vsimd> > GaugeField;
 
 	typedef iLorentzColourMatrix<vsimd> vobj;
 	typedef typename vobj::scalar_object sobj;
 
+	FieldMetaData header;
+	///////////////////////////////////////////
 	// Following should become arguments
-	NerscField header;
+	///////////////////////////////////////////
 	header.sequence_number = 1;
 	header.ensemble_id     = "UKQCD";
 	header.ensemble_label  = "DWF";
@@ -425,45 +231,32 @@ namespace Grid {
   
 	GridBase *grid = Umu._grid;
 
-	NerscGrid(grid,header);
-	NerscStatistics<GaugeField>(Umu,header);
-	NerscMachineCharacteristics(header);
+	GridMetaData(grid,header);
+	assert(header.nd==4);
+	GaugeStatistics(Umu,header);
+	MachineCharacteristics(header);
 
-	uint32_t csum;
 	int offset;
   
 	truncate(file);
 
-	if ( two_row ) { 
+	// 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;
+	offset = writeHeader(header,file);
 
-	  header.floating_point = std::string("IEEE64BIG");
-	  header.data_type      = std::string("4D_SU3_GAUGE");
-	  Nersc3x2unmunger<fobj2D,sobj> munge;
-	  BinaryIO::Uint32Checksum<vobj,fobj2D>(Umu, munge,header.checksum);
-	  offset = writeHeader(header,file);
-#ifdef PARALLEL_WRITE
-	  csum=BinaryIO::writeObjectParallel<vobj,fobj2D>(Umu,file,munge,offset,header.floating_point);
-#else
-	  csum=BinaryIO::writeObjectSerial<vobj,fobj2D>(Umu,file,munge,offset,header.floating_point);
-#endif
-	} else { 
-	  header.floating_point = std::string("IEEE64BIG");
-	  header.data_type      = std::string("4D_SU3_GAUGE_3x3");
-	  NerscSimpleUnmunger<fobj3D,sobj> munge;
-	  BinaryIO::Uint32Checksum<vobj,fobj3D>(Umu, munge,header.checksum);
-	  offset = writeHeader(header,file);
-#ifdef PARALLEL_WRITE
-	  csum=BinaryIO::writeObjectParallel<vobj,fobj3D>(Umu,file,munge,offset,header.floating_point);
-#else
-	  csum=BinaryIO::writeObjectSerial<vobj,fobj3D>(Umu,file,munge,offset,header.floating_point);
-#endif
-	}
+	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);
+	header.checksum = nersc_csum;
+	writeHeader(header,file);
 
-	std::cout<<GridLogMessage <<"Written NERSC Configuration on "<< file << " checksum "<<std::hex<<csum<< std::dec<<" plaq "<< header.plaquette <<std::endl;
+	std::cout<<GridLogMessage <<"Written NERSC Configuration on "<< file << " checksum "
+		 <<std::hex<<header.checksum
+		 <<std::dec<<" plaq "<< header.plaquette <<std::endl;
 
       }
-
-
       ///////////////////////////////
       // RNG state
       ///////////////////////////////
@@ -472,19 +265,19 @@ namespace Grid {
 	typedef typename GridParallelRNG::RngStateType RngStateType;
 
 	// Following should become arguments
-	NerscField header;
+	FieldMetaData header;
 	header.sequence_number = 1;
 	header.ensemble_id     = "UKQCD";
 	header.ensemble_label  = "DWF";
 
 	GridBase *grid = parallel._grid;
 
-	NerscGrid(grid,header);
+	GridMetaData(grid,header);
+	assert(header.nd==4);
 	header.link_trace=0.0;
 	header.plaquette=0.0;
-	NerscMachineCharacteristics(header);
+	MachineCharacteristics(header);
 
-	uint32_t csum;
 	int offset;
   
 #ifdef RNG_RANLUX
@@ -502,15 +295,19 @@ namespace Grid {
 
 	truncate(file);
 	offset = writeHeader(header,file);
-	csum=BinaryIO::writeRNGSerial(serial,parallel,file,offset);
-	header.checksum = csum;
+	uint32_t nersc_csum,scidac_csuma,scidac_csumb;
+	BinaryIO::writeRNG(serial,parallel,file,offset,nersc_csum,scidac_csuma,scidac_csumb);
+	header.checksum = nersc_csum;
 	offset = writeHeader(header,file);
 
-	std::cout<<GridLogMessage <<"Written NERSC RNG STATE "<<file<< " checksum "<<std::hex<<csum<<std::dec<<std::endl;
+	std::cout<<GridLogMessage 
+		 <<"Written NERSC RNG STATE "<<file<< " checksum "
+		 <<std::hex<<header.checksum
+		 <<std::dec<<std::endl;
 
       }
     
-      static inline void readRNGState(GridSerialRNG &serial,GridParallelRNG & parallel,NerscField& header,std::string file)
+      static inline void readRNGState(GridSerialRNG &serial,GridParallelRNG & parallel,FieldMetaData& header,std::string file)
       {
 	typedef typename GridParallelRNG::RngStateType RngStateType;
 
@@ -518,7 +315,7 @@ namespace Grid {
 
 	int offset = readHeader(file,grid,header);
 
-	NerscField clone(header);
+	FieldMetaData clone(header);
 
 	std::string format(header.floating_point);
 	std::string data_type(header.data_type);
@@ -538,15 +335,19 @@ namespace Grid {
 
 	// depending on datatype, set up munger;
 	// munger is a function of <floating point, Real, data_type>
-	uint32_t csum=BinaryIO::readRNGSerial(serial,parallel,file,offset);
+	uint32_t nersc_csum,scidac_csuma,scidac_csumb;
+	BinaryIO::readRNG(serial,parallel,file,offset,nersc_csum,scidac_csuma,scidac_csumb);
 
-	assert(csum == header.checksum );
+	if ( nersc_csum != header.checksum ) { 
+	  std::cerr << "checksum mismatch "<<std::hex<< nersc_csum <<" "<<header.checksum<<std::dec<<std::endl;
+	  exit(0);
+	}
+	assert(nersc_csum == header.checksum );
 
 	std::cout<<GridLogMessage <<"Read NERSC RNG file "<<file<< " format "<< data_type <<std::endl;
       }
 
     };
 
-
   }}
 #endif

lib/qcd/action/fermion/FermionOperatorImpl.h

@@ -644,19 +644,16 @@ class StaggeredImpl : public PeriodicGaugeImpl<GaugeImplTypes<S, Representation:
 
     INHERIT_GIMPL_TYPES(Gimpl);
       
-    template <typename vtype> using iImplScalar            = iScalar<iScalar<iScalar<vtype> > >;
     template <typename vtype> using iImplSpinor            = iScalar<iScalar<iVector<vtype, Dimension> > >;
     template <typename vtype> using iImplHalfSpinor        = iScalar<iScalar<iVector<vtype, Dimension> > >;
     template <typename vtype> using iImplDoubledGaugeField = iVector<iScalar<iMatrix<vtype, Dimension> >, Nds>;
     template <typename vtype> using iImplPropagator        = iScalar<iScalar<iMatrix<vtype, Dimension> > >;
     
-    typedef iImplScalar<Simd>            SiteComplex;
     typedef iImplSpinor<Simd>            SiteSpinor;
     typedef iImplHalfSpinor<Simd>        SiteHalfSpinor;
     typedef iImplDoubledGaugeField<Simd> SiteDoubledGaugeField;
     typedef iImplPropagator<Simd>        SitePropagator;
     
-    typedef Lattice<SiteComplex>           ComplexField;
     typedef Lattice<SiteSpinor>            FermionField;
     typedef Lattice<SiteDoubledGaugeField> DoubledGaugeField;
     typedef Lattice<SitePropagator> PropagatorField;
@@ -775,7 +772,6 @@ class StaggeredImpl : public PeriodicGaugeImpl<GaugeImplTypes<S, Representation:
 
     INHERIT_GIMPL_TYPES(Gimpl);
 
-    template <typename vtype> using iImplScalar            = iScalar<iScalar<iScalar<vtype> > >;
     template <typename vtype> using iImplSpinor            = iScalar<iScalar<iVector<vtype, Dimension> > >;
     template <typename vtype> using iImplHalfSpinor        = iScalar<iScalar<iVector<vtype, Dimension> > >;
     template <typename vtype> using iImplDoubledGaugeField = iVector<iScalar<iMatrix<vtype, Dimension> >, Nds>;
@@ -792,12 +788,10 @@ class StaggeredImpl : public PeriodicGaugeImpl<GaugeImplTypes<S, Representation:
     typedef Lattice<SiteDoubledGaugeField> DoubledGaugeField;
     typedef Lattice<SitePropagator> PropagatorField;
     
-    typedef iImplScalar<Simd>            SiteComplex;
     typedef iImplSpinor<Simd>            SiteSpinor;
     typedef iImplHalfSpinor<Simd>        SiteHalfSpinor;
 
     
-    typedef Lattice<SiteComplex>           ComplexField;
     typedef Lattice<SiteSpinor>            FermionField;
     
     typedef SimpleCompressor<SiteSpinor> Compressor;

lib/qcd/action/gauge/GaugeImplTypes.h

@@ -40,12 +40,15 @@ namespace QCD {
   typedef typename GImpl::Simd Simd;                \
   typedef typename GImpl::LinkField GaugeLinkField; \
   typedef typename GImpl::Field GaugeField;         \
+  typedef typename GImpl::ComplexField ComplexField;\
   typedef typename GImpl::SiteField SiteGaugeField; \
+  typedef typename GImpl::SiteComplex SiteComplex;  \
   typedef typename GImpl::SiteLink SiteGaugeLink;
 
-#define INHERIT_FIELD_TYPES(Impl)             \
-  typedef typename Impl::Simd Simd;           \
-  typedef typename Impl::SiteField SiteField; \
+#define INHERIT_FIELD_TYPES(Impl)		    \
+  typedef typename Impl::Simd Simd;		    \
+  typedef typename Impl::ComplexField ComplexField; \
+  typedef typename Impl::SiteField SiteField;	    \
   typedef typename Impl::Field Field;
 
 // hardcodes the exponential approximation in the template
@@ -53,14 +56,17 @@ template <class S, int Nrepresentation = Nc, int Nexp = 12 > class GaugeImplType
 public:
   typedef S Simd;
 
-  template <typename vtype> using iImplGaugeLink  = iScalar<iScalar<iMatrix<vtype, Nrepresentation>>>;
-  template <typename vtype> using iImplGaugeField = iVector<iScalar<iMatrix<vtype, Nrepresentation>>, Nd>;
+  template <typename vtype> using iImplScalar     = iScalar<iScalar<iScalar<vtype> > >;
+  template <typename vtype> using iImplGaugeLink  = iScalar<iScalar<iMatrix<vtype, Nrepresentation> > >;
+  template <typename vtype> using iImplGaugeField = iVector<iScalar<iMatrix<vtype, Nrepresentation> >, Nd>;
 
+  typedef iImplScalar<Simd>     SiteComplex;
   typedef iImplGaugeLink<Simd>  SiteLink;
   typedef iImplGaugeField<Simd> SiteField;
 
-  typedef Lattice<SiteLink>  LinkField; 
-  typedef Lattice<SiteField> Field;
+  typedef Lattice<SiteComplex> ComplexField;
+  typedef Lattice<SiteLink>    LinkField; 
+  typedef Lattice<SiteField>   Field;
 
   // Guido: we can probably separate the types from the HMC functions
   // this will create 2 kind of implementations
@@ -80,7 +86,7 @@ public:
 
   ///////////////////////////////////////////////////////////
   // Move these to another class
-  // HMC auxiliary functions 
+  // HMC auxiliary functions
   static inline void generate_momenta(Field &P, GridParallelRNG &pRNG) {
     // specific for SU gauge fields
     LinkField Pmu(P._grid);
@@ -92,14 +98,19 @@ public:
   }
 
   static inline Field projectForce(Field &P) { return Ta(P); }
-  
+
   static inline void update_field(Field& P, Field& U, double ep){
-    for (int mu = 0; mu < Nd; mu++) {
-      auto Umu = PeekIndex<LorentzIndex>(U, mu);
-      auto Pmu = PeekIndex<LorentzIndex>(P, mu);
-      Umu = expMat(Pmu, ep, Nexp) * Umu;
-      PokeIndex<LorentzIndex>(U, ProjectOnGroup(Umu), mu);
+    //static std::chrono::duration<double> diff;
+
+    //auto start = std::chrono::high_resolution_clock::now();
+    parallel_for(int ss=0;ss<P._grid->oSites();ss++){
+      for (int mu = 0; mu < Nd; mu++) 
+        U[ss]._internal[mu] = ProjectOnGroup(Exponentiate(P[ss]._internal[mu], ep, Nexp) * U[ss]._internal[mu]);
     }
+    
+    //auto end = std::chrono::high_resolution_clock::now();
+   // diff += end - start;
+   // std::cout << "Time to exponentiate matrix " << diff.count() << " s\n";
   }
 
   static inline RealD FieldSquareNorm(Field& U){

lib/qcd/action/gauge/WilsonGaugeAction.h

@@ -71,14 +71,18 @@ class WilsonGaugeAction : public Action<typename Gimpl::GaugeField> {
 
     RealD factor = 0.5 * beta / RealD(Nc);
 
-    GaugeLinkField Umu(U._grid);
+    //GaugeLinkField Umu(U._grid);
     GaugeLinkField dSdU_mu(U._grid);
     for (int mu = 0; mu < Nd; mu++) {
-      Umu = PeekIndex<LorentzIndex>(U, mu);
+      //Umu = PeekIndex<LorentzIndex>(U, mu);
 
       // Staple in direction mu
-      WilsonLoops<Gimpl>::Staple(dSdU_mu, U, mu);
-      dSdU_mu = Ta(Umu * dSdU_mu) * factor;
+      //WilsonLoops<Gimpl>::Staple(dSdU_mu, U, mu);
+      //dSdU_mu = Ta(Umu * dSdU_mu) * factor;
+
+  
+      WilsonLoops<Gimpl>::StapleMult(dSdU_mu, U, mu);
+      dSdU_mu = Ta(dSdU_mu) * factor;
 
       PokeIndex<LorentzIndex>(dSdU, dSdU_mu, mu);
     }

lib/qcd/action/scalar/ScalarImpl.h

@@ -15,6 +15,8 @@ namespace Grid {
     
     typedef iImplField<Simd> SiteField;
     
+    template <typename vtype> using iImplScalar= iScalar<iScalar<iScalar<vtype   > > >;
+    typedef iImplScalar<Simd> ComplexField;
     
     typedef Lattice<SiteField> Field;
     
@@ -51,13 +53,14 @@ namespace Grid {
   public:
     typedef S Simd;
     
-    template <typename vtype>
-    using iImplField = iScalar<iScalar<iMatrix<vtype, N> > >;
-    
+    template <typename vtype> using iImplField = iScalar<iScalar<iMatrix<vtype, N> > >;
+
     typedef iImplField<Simd> SiteField;
-    
-    
     typedef Lattice<SiteField> Field;
+
+    template <typename vtype> using iImplScalar= iScalar<iScalar<iScalar<vtype   > > >;
+    typedef iImplScalar<Simd> ComplexField;
+    
     
     static inline void generate_momenta(Field& P, GridParallelRNG& pRNG){
       gaussian(pRNG, P);

lib/qcd/hmc/checkpointers/BinaryCheckpointer.h

@@ -62,36 +62,50 @@ class BinaryHmcCheckpointer : public BaseHmcCheckpointer<Impl> {
     fout.close();
   }
 
-  void TrajectoryComplete(int traj, Field &U, GridSerialRNG &sRNG,
-                          GridParallelRNG &pRNG) {
+  void TrajectoryComplete(int traj, Field &U, GridSerialRNG &sRNG, GridParallelRNG &pRNG) {
+
     if ((traj % Params.saveInterval) == 0) {
       std::string config, rng;
       this->build_filenames(traj, Params, config, rng);
 
-      BinaryIO::BinarySimpleUnmunger<sobj_double, sobj> munge;
+      uint32_t nersc_csum;
+      uint32_t scidac_csuma;
+      uint32_t scidac_csumb;
+      
+      BinarySimpleUnmunger<sobj_double, sobj> munge;
       truncate(rng);
-      BinaryIO::writeRNGSerial(sRNG, pRNG, rng, 0);
+      BinaryIO::writeRNG(sRNG, pRNG, rng, 0,nersc_csum,scidac_csuma,scidac_csumb);
       truncate(config);
-      uint32_t csum = BinaryIO::writeObjectParallel<vobj, sobj_double>(
-          U, config, munge, 0, Params.format);
+
+      BinaryIO::writeLatticeObject<vobj, sobj_double>(U, config, munge, 0, Params.format,
+						      nersc_csum,scidac_csuma,scidac_csumb);
 
       std::cout << GridLogMessage << "Written Binary Configuration " << config
-                << " checksum " << std::hex << csum << std::dec << std::endl;
+                << " checksum " << std::hex 
+		<< nersc_csum   <<"/"
+		<< scidac_csuma   <<"/"
+		<< scidac_csumb 
+		<< std::dec << std::endl;
     }
+
   };
 
-  void CheckpointRestore(int traj, Field &U, GridSerialRNG &sRNG,
-                         GridParallelRNG &pRNG) {
+  void CheckpointRestore(int traj, Field &U, GridSerialRNG &sRNG, GridParallelRNG &pRNG) {
     std::string config, rng;
     this->build_filenames(traj, Params, config, rng);
 
-    BinaryIO::BinarySimpleMunger<sobj_double, sobj> munge;
-    BinaryIO::readRNGSerial(sRNG, pRNG, rng, 0);
-    uint32_t csum = BinaryIO::readObjectParallel<vobj, sobj_double>(
-        U, config, munge, 0, Params.format);
+    BinarySimpleMunger<sobj_double, sobj> munge;
 
+    uint32_t nersc_csum;
+    uint32_t scidac_csuma;
+    uint32_t scidac_csumb;
+    BinaryIO::readRNG(sRNG, pRNG, rng, 0,nersc_csum,scidac_csuma,scidac_csumb);
+    BinaryIO::readLatticeObject<vobj, sobj_double>(U, config, munge, 0, Params.format,
+						   nersc_csum,scidac_csuma,scidac_csumb);
+    
     std::cout << GridLogMessage << "Read Binary Configuration " << config
-              << " checksum " << std::hex << csum << std::dec << std::endl;
+              << " checksums " << std::hex << nersc_csum<<"/"<<scidac_csuma<<"/"<<scidac_csumb 
+	      << std::dec << std::endl;
   };
 };
 }

lib/qcd/hmc/checkpointers/ILDGCheckpointer.h

@@ -54,9 +54,9 @@ class ILDGHmcCheckpointer : public BaseHmcCheckpointer<Implementation> {
 
     // check here that the format is valid
     int ieee32big = (Params.format == std::string("IEEE32BIG"));
-    int ieee32 = (Params.format == std::string("IEEE32"));
+    int ieee32    = (Params.format == std::string("IEEE32"));
     int ieee64big = (Params.format == std::string("IEEE64BIG"));
-    int ieee64 = (Params.format == std::string("IEEE64"));
+    int ieee64    = (Params.format == std::string("IEEE64"));
 
     if (!(ieee64big || ieee32 || ieee32big || ieee64)) {
       std::cout << GridLogError << "Unrecognized file format " << Params.format
@@ -74,13 +74,20 @@ class ILDGHmcCheckpointer : public BaseHmcCheckpointer<Implementation> {
     if ((traj % Params.saveInterval) == 0) {
       std::string config, rng;
       this->build_filenames(traj, Params, config, rng);
-
-      ILDGIO IO(config, ILDGwrite);
-      BinaryIO::writeRNGSerial(sRNG, pRNG, rng, 0);
-      uint32_t csum = IO.writeConfiguration(U, Params.format);
+      
+      uint32_t nersc_csum,scidac_csuma,scidac_csumb;
+      BinaryIO::writeRNG(sRNG, pRNG, rng, 0,nersc_csum,scidac_csuma,scidac_csumb);
+      IldgWriter _IldgWriter;
+      _IldgWriter.open(config);
+      _IldgWriter.writeConfiguration(U, traj, config, config);
+      _IldgWriter.close();
 
       std::cout << GridLogMessage << "Written ILDG Configuration on " << config
-                << " checksum " << std::hex << csum << std::dec << std::endl;
+                << " checksum " << std::hex 
+		<< nersc_csum<<"/"
+		<< scidac_csuma<<"/"
+		<< scidac_csumb
+		<< std::dec << std::endl;
     }
   };
 
@@ -89,12 +96,21 @@ class ILDGHmcCheckpointer : public BaseHmcCheckpointer<Implementation> {
     std::string config, rng;
     this->build_filenames(traj, Params, config, rng);
 
-    ILDGIO IO(config, ILDGread);
-    BinaryIO::readRNGSerial(sRNG, pRNG, rng, 0);
-    uint32_t csum = IO.readConfiguration(U);  // format from the header
+    uint32_t nersc_csum,scidac_csuma,scidac_csumb;
+    BinaryIO::readRNG(sRNG, pRNG, rng, 0,nersc_csum,scidac_csuma,scidac_csumb);
+
+    FieldMetaData header;
+    IldgReader _IldgReader;
+    _IldgReader.open(config);
+    _IldgReader.readConfiguration(U,header);  // format from the header
+    _IldgReader.close();
 
     std::cout << GridLogMessage << "Read ILDG Configuration from " << config
-              << " checksum " << std::hex << csum << std::dec << std::endl;
+              << " checksum " << std::hex 
+	      << nersc_csum<<"/"
+	      << scidac_csuma<<"/"
+	      << scidac_csumb
+	      << std::dec << std::endl;
   };
 };
 }

lib/qcd/hmc/checkpointers/NerscCheckpointer.h

@@ -70,7 +70,7 @@ class NerscHmcCheckpointer : public BaseHmcCheckpointer<Gimpl> {
     std::string config, rng;
     this->build_filenames(traj, Params, config, rng);
 
-    NerscField header;
+    FieldMetaData header;
     NerscIO::readRNGState(sRNG, pRNG, header, rng);
     NerscIO::readConfiguration(U, header, config);
   };

lib/qcd/smearing/StoutSmearing.h

@@ -58,6 +58,8 @@ class Smear_Stout : public Smear<Gimpl> {
     SmearBase->smear(C, U);
   };
 
+
+  // Repetion of code here (use the Tensor_exp.h function)
   void exponentiate_iQ(GaugeLinkField& e_iQ, const GaugeLinkField& iQ) const {
     // Put this outside
     // only valid for SU(3) matrices

lib/qcd/smearing/WilsonFlow.h

@@ -36,20 +36,23 @@ namespace QCD {
 template <class Gimpl>
 class WilsonFlow: public Smear<Gimpl>{
     unsigned int Nstep;
-    RealD epsilon;
+    unsigned int measure_interval;
+    mutable RealD epsilon, taus;
+
 
     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); }
 
-
  public:
     INHERIT_GIMPL_TYPES(Gimpl)
 
-    explicit WilsonFlow(unsigned int Nstep, RealD epsilon):
+    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);
@@ -72,7 +75,9 @@ class WilsonFlow: public Smear<Gimpl>{
         // 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;
 };
 
 
@@ -98,23 +103,111 @@ void WilsonFlow<Gimpl>::evolve_step(typename Gimpl::GaugeField &U) const{
     Gimpl::update_field(Z, U, -2.0*epsilon);    // V(t+e) = exp(ep*Z)*W2
 }
 
+template <class Gimpl>
+void WilsonFlow<Gimpl>::evolve_step_adaptive(typename Gimpl::GaugeField &U, RealD maxTau) {
+    if (maxTau - taus < epsilon){
+        epsilon = maxTau-taus;
+    }
+    std::cout << GridLogMessage << "Integration epsilon : " << epsilon << std::endl;
+    GaugeField Z(U._grid);
+    GaugeField Zprime(U._grid);
+    GaugeField tmp(U._grid), Uprime(U._grid);
+    Uprime = U;
+    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
+
+    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
+    
+
+    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
+
+    // Ramos 
+    Gimpl::update_field(Zprime, Uprime, -2.0*epsilon); // 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;
+    std::cout << GridLogMessage << "Adjusting integration step with distance: " << diff << std::endl;
+    
+    epsilon = epsilon*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();
 }
 
+template <class Gimpl>
+RealD WilsonFlow<Gimpl>::energyDensityPlaquette(const GaugeField& U) const {
+    return 2.0 * taus * taus * SG.S(U)/U._grid->gSites();
+}
+
+
+//#define WF_TIMING 
+
+
+
 template <class Gimpl>
 void WilsonFlow<Gimpl>::smear(GaugeField& out, const GaugeField& in) const {
     out = in;
-    for (unsigned int step = 0; step < Nstep; step++) {
+    for (unsigned int step = 1; step <= Nstep; step++) {
+        auto start = std::chrono::high_resolution_clock::now();
+        std::cout << GridLogMessage << "Evolution time :"<< tau(step) << std::endl;
         evolve_step(out);
+        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 << " "
+            << step << "  "
             << energyDensityPlaquette(step,out) << std::endl;
+         if( step % measure_interval == 0){
+         std::cout << GridLogMessage << "[WilsonFlow] Top. charge           : "
+            << step << "  " 
+            << WilsonLoops<PeriodicGimplR>::TopologicalCharge(out) << std::endl;
+        }
     }
 }
 
+template <class Gimpl>
+void WilsonFlow<Gimpl>::smear_adaptive(GaugeField& out, const GaugeField& in, RealD maxTau){
+    out = in;
+    taus = 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 << "  "
+            << energyDensityPlaquette(out) << std::endl;
+         if( step % measure_interval == 0){
+         std::cout << GridLogMessage << "[WilsonFlow] Top. charge           : "
+            << step << "  " 
+            << WilsonLoops<PeriodicGimplR>::TopologicalCharge(out) << std::endl;
+        }
+    } while (taus < maxTau);
+
+
+
+}
+
+
 }  // namespace QCD
 }  // namespace Grid
 

lib/qcd/utils/GaugeFix.h

@@ -0,0 +1,188 @@
+    /*************************************************************************************
+
+    grid` physics library, www.github.com/paboyle/Grid 
+
+    Copyright (C) 2015
+
+Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
+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 Grid;
+using namespace Grid::QCD;
+
+template <class Gimpl> 
+class FourierAcceleratedGaugeFixer  : public Gimpl {
+  public:
+  INHERIT_GIMPL_TYPES(Gimpl);
+
+  typedef typename Gimpl::GaugeLinkField GaugeMat;
+  typedef typename Gimpl::GaugeField GaugeLorentz;
+
+  static void GaugeLinkToLieAlgebraField(const std::vector<GaugeMat> &U,std::vector<GaugeMat> &A) {
+    for(int mu=0;mu<Nd;mu++){
+      Complex cmi(0.0,-1.0);
+      A[mu] = Ta(U[mu]) * cmi;
+    }
+  }
+  static void DmuAmu(const std::vector<GaugeMat> &A,GaugeMat &dmuAmu) {
+    dmuAmu=zero;
+    for(int mu=0;mu<Nd;mu++){
+      dmuAmu = dmuAmu + A[mu] - Cshift(A[mu],mu,-1);
+    }
+  }  
+  static void SteepestDescentGaugeFix(GaugeLorentz &Umu,Real & alpha,int maxiter,Real Omega_tol, Real Phi_tol,bool Fourier=false) {
+    GridBase *grid = Umu._grid;
+
+    Real org_plaq      =WilsonLoops<Gimpl>::avgPlaquette(Umu);
+    Real org_link_trace=WilsonLoops<Gimpl>::linkTrace(Umu); 
+    Real old_trace = org_link_trace;
+    Real trG;
+
+    std::vector<GaugeMat> U(Nd,grid);
+                 GaugeMat dmuAmu(grid);
+
+    for(int i=0;i<maxiter;i++){
+      for(int mu=0;mu<Nd;mu++) U[mu]= PeekIndex<LorentzIndex>(Umu,mu);
+      if ( Fourier==false ) { 
+	trG = SteepestDescentStep(U,alpha,dmuAmu);
+      } else { 
+	trG = FourierAccelSteepestDescentStep(U,alpha,dmuAmu);
+      }
+      for(int mu=0;mu<Nd;mu++) PokeIndex<LorentzIndex>(Umu,U[mu],mu);
+      // Monitor progress and convergence test 
+      // infrequently to minimise cost overhead
+      if ( i %20 == 0 ) { 
+	Real plaq      =WilsonLoops<Gimpl>::avgPlaquette(Umu);
+	Real link_trace=WilsonLoops<Gimpl>::linkTrace(Umu); 
+
+	if (Fourier) 
+	  std::cout << GridLogMessage << "Fourier Iteration "<<i<< " plaq= "<<plaq<< " dmuAmu " << norm2(dmuAmu)<< std::endl;
+	else 
+	  std::cout << GridLogMessage << " Iteration "<<i<< " plaq= "<<plaq<< " dmuAmu " << norm2(dmuAmu)<< std::endl;
+	
+	Real Phi  = 1.0 - old_trace / link_trace ;
+	Real Omega= 1.0 - trG;
+
+
+	std::cout << GridLogMessage << " Iteration "<<i<< " Phi= "<<Phi<< " Omega= " << Omega<< " trG " << trG <<std::endl;
+	if ( (Omega < Omega_tol) && ( ::fabs(Phi) < Phi_tol) ) {
+	  std::cout << GridLogMessage << "Converged ! "<<std::endl;
+	  return;
+	}
+
+	old_trace = link_trace;
+
+      }
+    }
+  };
+  static Real SteepestDescentStep(std::vector<GaugeMat> &U,Real & alpha, GaugeMat & dmuAmu) {
+    GridBase *grid = U[0]._grid;
+
+    std::vector<GaugeMat> A(Nd,grid);
+    GaugeMat g(grid);
+
+    GaugeLinkToLieAlgebraField(U,A);
+    ExpiAlphaDmuAmu(A,g,alpha,dmuAmu);
+
+
+    Real vol = grid->gSites();
+    Real trG = TensorRemove(sum(trace(g))).real()/vol/Nc;
+
+    SU<Nc>::GaugeTransform(U,g);
+
+    return trG;
+  }
+
+  static Real FourierAccelSteepestDescentStep(std::vector<GaugeMat> &U,Real & alpha, GaugeMat & dmuAmu) {
+
+    GridBase *grid = U[0]._grid;
+
+    Real vol = grid->gSites();
+
+    FFT theFFT((GridCartesian *)grid);
+
+    LatticeComplex  Fp(grid);
+    LatticeComplex  psq(grid); psq=zero;
+    LatticeComplex  pmu(grid); 
+    LatticeComplex   one(grid); one = Complex(1.0,0.0);
+
+    GaugeMat g(grid);
+    GaugeMat dmuAmu_p(grid);
+    std::vector<GaugeMat> A(Nd,grid);
+
+    GaugeLinkToLieAlgebraField(U,A);
+
+    DmuAmu(A,dmuAmu);
+
+    theFFT.FFT_all_dim(dmuAmu_p,dmuAmu,FFT::forward);
+
+    //////////////////////////////////
+    // Work out Fp = psq_max/ psq...
+    //////////////////////////////////
+    std::vector<int> latt_size = grid->GlobalDimensions();
+    std::vector<int> coor(grid->_ndimension,0);
+    for(int mu=0;mu<Nd;mu++) {
+
+      Real TwoPiL =  M_PI * 2.0/ latt_size[mu];
+      LatticeCoordinate(pmu,mu);
+      pmu = TwoPiL * pmu ;
+      psq = psq + 4.0*sin(pmu*0.5)*sin(pmu*0.5); 
+    }
+
+    Complex psqMax(16.0);
+    Fp =  psqMax*one/psq;
+
+    /*
+    static int once;
+    if ( once == 0 ) { 
+      std::cout << " Fp " << Fp <<std::endl;
+      once ++;
+      }*/
+
+    pokeSite(TComplex(1.0),Fp,coor);
+
+    dmuAmu_p  = dmuAmu_p * Fp; 
+
+    theFFT.FFT_all_dim(dmuAmu,dmuAmu_p,FFT::backward);
+
+    GaugeMat ciadmam(grid);
+    Complex cialpha(0.0,-alpha);
+    ciadmam = dmuAmu*cialpha;
+    SU<Nc>::taExp(ciadmam,g);
+
+    Real trG = TensorRemove(sum(trace(g))).real()/vol/Nc;
+
+    SU<Nc>::GaugeTransform(U,g);
+
+    return trG;
+  }
+
+  static void ExpiAlphaDmuAmu(const std::vector<GaugeMat> &A,GaugeMat &g,Real & alpha, GaugeMat &dmuAmu) {
+    GridBase *grid = g._grid;
+    Complex cialpha(0.0,-alpha);
+    GaugeMat ciadmam(grid);
+    DmuAmu(A,dmuAmu);
+    ciadmam = dmuAmu*cialpha;
+    SU<Nc>::taExp(ciadmam,g);
+  }  
+};
+

lib/qcd/utils/Utils.h

@@ -12,7 +12,4 @@
 #include <Grid/qcd/utils/SUnAdjoint.h>
 #include <Grid/qcd/utils/SUnTwoIndex.h>
 
-
-
-
 #endif

lib/qcd/utils/WilsonLoops.h

@@ -73,7 +73,7 @@ public:
   //////////////////////////////////////////////////
   // trace of directed plaquette oriented in mu,nu plane
   //////////////////////////////////////////////////
-  static void traceDirPlaquette(LatticeComplex &plaq,
+  static void traceDirPlaquette(ComplexField &plaq,
                                 const std::vector<GaugeMat> &U, const int mu,
                                 const int nu) {
     GaugeMat sp(U[0]._grid);
@@ -83,9 +83,9 @@ public:
   //////////////////////////////////////////////////
   // sum over all planes of plaquette
   //////////////////////////////////////////////////
-  static void sitePlaquette(LatticeComplex &Plaq,
+  static void sitePlaquette(ComplexField &Plaq,
                             const std::vector<GaugeMat> &U) {
-    LatticeComplex sitePlaq(U[0]._grid);
+    ComplexField sitePlaq(U[0]._grid);
     Plaq = zero;
     for (int mu = 1; mu < Nd; mu++) {
       for (int nu = 0; nu < mu; nu++) {
@@ -104,11 +104,11 @@ public:
       U[mu] = PeekIndex<LorentzIndex>(Umu, mu);
     }
 
-    LatticeComplex Plaq(Umu._grid);
+    ComplexField Plaq(Umu._grid);
 
     sitePlaquette(Plaq, U);
-    TComplex Tp = sum(Plaq);
-    Complex p = TensorRemove(Tp);
+    auto Tp = sum(Plaq);
+    auto p = TensorRemove(Tp);
     return p.real();
   }
 
@@ -129,15 +129,15 @@ public:
   static RealD linkTrace(const GaugeLorentz &Umu) {
     std::vector<GaugeMat> U(Nd, Umu._grid);
 
-    LatticeComplex Tr(Umu._grid);
+    ComplexField Tr(Umu._grid);
     Tr = zero;
     for (int mu = 0; mu < Nd; mu++) {
       U[mu] = PeekIndex<LorentzIndex>(Umu, mu);
       Tr = Tr + trace(U[mu]);
     }
 
-    TComplex Tp = sum(Tr);
-    Complex p = TensorRemove(Tp);
+    auto Tp = sum(Tr);
+    auto p = TensorRemove(Tp);
 
     double vol = Umu._grid->gSites();
 
@@ -188,6 +188,32 @@ public:
     }
   }
 
+
+// For the force term
+static void StapleMult(GaugeMat &staple, const GaugeLorentz &Umu, int mu) {
+    GridBase *grid = Umu._grid;
+    std::vector<GaugeMat> U(Nd, grid);
+    for (int d = 0; d < Nd; d++) {
+      // this operation is taking too much time
+      U[d] = PeekIndex<LorentzIndex>(Umu, d);
+    }
+    staple = zero;
+    GaugeMat tmp1(grid);
+    GaugeMat tmp2(grid);
+
+    for (int nu = 0; nu < Nd; nu++) {
+      if (nu != mu) {
+        // this is ~10% faster than the Staple
+        tmp1 = Cshift(U[nu], mu, 1);
+        tmp2 = Cshift(U[mu], nu, 1);
+        staple += tmp1* adj(U[nu]*tmp2);
+        tmp2 = adj(U[mu]*tmp1)*U[nu];
+        staple += Cshift(tmp2, nu, -1);
+      }
+    }
+    staple = U[mu]*staple;
+}
+
   //////////////////////////////////////////////////
   // the sum over all staples on each site
   //////////////////////////////////////////////////
@@ -200,7 +226,6 @@ public:
       U[d] = PeekIndex<LorentzIndex>(Umu, d);
     }
     staple = zero;
-    GaugeMat tmp(grid);
 
     for (int nu = 0; nu < Nd; nu++) {
 
@@ -214,7 +239,7 @@ public:
         //      |
         //    __|
         //
-
+     
         staple += Gimpl::ShiftStaple(
             Gimpl::CovShiftForward(
                 U[nu], nu,
@@ -227,6 +252,7 @@ public:
         // |__
         //
         //
+
         staple += Gimpl::ShiftStaple(
             Gimpl::CovShiftBackward(U[nu], nu,
                                     Gimpl::CovShiftBackward(U[mu], mu, U[nu])), mu);
@@ -289,8 +315,7 @@ public:
       //
       staple = Gimpl::ShiftStaple(
           Gimpl::CovShiftBackward(U[nu], nu,
-                                  Gimpl::CovShiftBackward(U[mu], mu, U[nu])),
-          mu);
+                                  Gimpl::CovShiftBackward(U[mu], mu, U[nu])), mu);
     }
   }
 
@@ -307,10 +332,10 @@ public:
       GaugeMat Vup(Umu._grid), Vdn(Umu._grid);
       StapleUpper(Vup, Umu, mu, nu);
       StapleLower(Vdn, Umu, mu, nu);
-      GaugeMat v = adj(Vup) - adj(Vdn);
+      GaugeMat v = Vup - Vdn;
       GaugeMat u = PeekIndex<LorentzIndex>(Umu, mu);  // some redundant copies
       GaugeMat vu = v*u;
-      FS = 0.25*Ta(u*v + Cshift(vu, mu, +1));
+      FS = 0.25*Ta(u*v + Cshift(vu, mu, -1));
   }
 
   static Real TopologicalCharge(GaugeLorentz &U){
@@ -330,8 +355,8 @@ public:
 
     double coeff = 8.0/(32.0*M_PI*M_PI);
 
-    LatticeComplex qfield = coeff*trace(Bx*Ex + By*Ey + Bz*Ez);
-    TComplex Tq = sum(qfield);
+    ComplexField qfield = coeff*trace(Bx*Ex + By*Ey + Bz*Ez);
+    auto Tq = sum(qfield);
     return TensorRemove(Tq).real();
   }
 
@@ -350,16 +375,16 @@ public:
                adj(Gimpl::CovShiftForward(
                    U[nu], nu, Gimpl::CovShiftForward(U[nu], nu, U[mu])));
   }
-  static void traceDirRectangle(LatticeComplex &rect,
+  static void traceDirRectangle(ComplexField &rect,
                                 const std::vector<GaugeMat> &U, const int mu,
                                 const int nu) {
     GaugeMat sp(U[0]._grid);
     dirRectangle(sp, U, mu, nu);
     rect = trace(sp);
   }
-  static void siteRectangle(LatticeComplex &Rect,
+  static void siteRectangle(ComplexField &Rect,
                             const std::vector<GaugeMat> &U) {
-    LatticeComplex siteRect(U[0]._grid);
+    ComplexField siteRect(U[0]._grid);
     Rect = zero;
     for (int mu = 1; mu < Nd; mu++) {
       for (int nu = 0; nu < mu; nu++) {
@@ -379,12 +404,12 @@ public:
       U[mu] = PeekIndex<LorentzIndex>(Umu, mu);
     }
 
-    LatticeComplex Rect(Umu._grid);
+    ComplexField Rect(Umu._grid);
 
     siteRectangle(Rect, U);
 
-    TComplex Tp = sum(Rect);
-    Complex p = TensorRemove(Tp);
+    auto Tp = sum(Rect);
+    auto p = TensorRemove(Tp);
     return p.real();
   }
   //////////////////////////////////////////////////

lib/serialisation/MacroMagic.h

@@ -110,11 +110,12 @@ THE SOFTWARE.
 
 #define GRID_MACRO_MEMBER(A,B)        A B;
 #define GRID_MACRO_COMP_MEMBER(A,B) result = (result and (lhs. B == rhs. B));
-#define GRID_MACRO_OS_WRITE_MEMBER(A,B) os<< #A <<" "#B <<" = "<< obj. B <<" ; " <<std::endl;
+#define GRID_MACRO_OS_WRITE_MEMBER(A,B) os<< #A <<" " #B << " = " << obj. B << " ; " <<std::endl;
 #define GRID_MACRO_READ_MEMBER(A,B) Grid::read(RD,#B,obj. B);
 #define GRID_MACRO_WRITE_MEMBER(A,B) Grid::write(WR,#B,obj. B);
 
 #define GRID_SERIALIZABLE_CLASS_MEMBERS(cname,...)\
+  std::string SerialisableClassName(void) {return std::string(#cname);}	\
 GRID_MACRO_EVAL(GRID_MACRO_MAP(GRID_MACRO_MEMBER,__VA_ARGS__))\
 template <typename T>\
 static inline void write(Writer<T> &WR,const std::string &s, const cname &obj){ \

lib/serialisation/XmlIO.cc

@@ -32,16 +32,21 @@ using namespace Grid;
 using namespace std;
 
 // Writer implementation ///////////////////////////////////////////////////////
-XmlWriter::XmlWriter(const string &fileName)
-: fileName_(fileName)
+XmlWriter::XmlWriter(const string &fileName, string toplev) : fileName_(fileName)
 {
-  node_ = doc_.append_child();
-  node_.set_name("grid");
+  if ( toplev == std::string("") ) {
+    node_=doc_;
+  } else { 
+    node_=doc_.append_child();
+    node_.set_name(toplev.c_str());
+  }
 }
 
 XmlWriter::~XmlWriter(void)
 {
-  doc_.save_file(fileName_.c_str(), "  ");
+  if ( fileName_ != std::string("") ) { 
+    doc_.save_file(fileName_.c_str(), "  ");
+  }
 }
 
 void XmlWriter::push(const string &s)
@@ -53,21 +58,44 @@ void XmlWriter::pop(void)
 {
   node_ = node_.parent();
 }
-
-// Reader implementation ///////////////////////////////////////////////////////
-XmlReader::XmlReader(const string &fileName)
-: fileName_(fileName)
+std::string XmlWriter::XmlString(void)
 {
-  pugi::xml_parse_result result = doc_.load_file(fileName_.c_str());
-  
-  if ( !result )
-  {
+  std::ostringstream oss; 
+  doc_.save(oss);
+  return oss.str();
+}
+
+XmlReader::XmlReader(const char *xmlstring,string toplev) : fileName_("")
+{
+  pugi::xml_parse_result result;
+  result = doc_.load_string(xmlstring);
+  if ( !result ) {
     cerr << "XML error description: " << result.description() << "\n";
     cerr << "XML error offset     : " << result.offset        << "\n";
     abort();
   }
-  
-  node_ = doc_.child("grid");
+  if ( toplev == std::string("") ) {
+    node_ = doc_;
+  } else { 
+    node_ = doc_.child(toplev.c_str());
+  }
+}
+
+// Reader implementation ///////////////////////////////////////////////////////
+XmlReader::XmlReader(const string &fileName,string toplev) : fileName_(fileName)
+{
+  pugi::xml_parse_result result;
+  result = doc_.load_file(fileName_.c_str());
+  if ( !result ) {
+    cerr << "XML error description: " << result.description() << "\n";
+    cerr << "XML error offset     : " << result.offset        << "\n";
+    abort();
+  }
+  if ( toplev == std::string("") ) {
+    node_ = doc_;
+  } else { 
+    node_ = doc_.child(toplev.c_str());
+  }
 }
 
 bool XmlReader::push(const string &s)

lib/serialisation/XmlIO.h

@@ -44,10 +44,9 @@ namespace Grid
 {
   
   class XmlWriter: public Writer<XmlWriter>
-  {
-    
+  {    
   public:
-    XmlWriter(const std::string &fileName);
+    XmlWriter(const std::string &fileName,std::string toplev = std::string("grid") );
     virtual ~XmlWriter(void);
     void push(const std::string &s);
     void pop(void);
@@ -55,6 +54,7 @@ namespace Grid
     void writeDefault(const std::string &s, const U &x);
     template <typename U>
     void writeDefault(const std::string &s, const std::vector<U> &x);
+    std::string XmlString(void);
   private:
     pugi::xml_document doc_;
     pugi::xml_node     node_;
@@ -64,7 +64,8 @@ namespace Grid
   class XmlReader: public Reader<XmlReader>
   {
   public:
-    XmlReader(const std::string &fileName);
+    XmlReader(const char *xmlstring,std::string toplev = std::string("grid") );
+    XmlReader(const std::string &fileName,std::string toplev = std::string("grid") );
     virtual ~XmlReader(void) = default;
     bool push(const std::string &s);
     void pop(void);
@@ -118,7 +119,7 @@ namespace Grid
     std::string buf;
     
     readDefault(s, buf);
-    std::cout << s << "   " << buf << std::endl;
+    //    std::cout << s << "   " << buf << std::endl;
     fromString(output, buf);
   }
   

lib/simd/Grid_generic.h

@@ -281,8 +281,8 @@ namespace Optimization {
 
   struct PrecisionChange {
     static inline vech StoH (const vecf &a,const vecf &b) {
-#ifdef USE_FP16
       vech ret;
+#ifdef USE_FP16
       vech *ha = (vech *)&a;
       vech *hb = (vech *)&b;
       const int nf = W<float>::r;
@@ -493,6 +493,8 @@ namespace Optimization {
     
     return a;
   }
+
+  #undef acc  // EIGEN compatibility
 }
 
 //////////////////////////////////////////////////////////////////////////////////////

lib/simd/Grid_vector_types.h

@@ -327,18 +327,16 @@ class Grid_simd {
   // provides support
   ///////////////////////////////////////
 
-#if (__GNUC__ == 5 ) || ( ( __GNUC__ == 6 ) && __GNUC_MINOR__ < 3 )
-#pragma GCC push_options 
-#pragma GCC optimize ("O0") 
-#endif
   template <class functor>
   friend inline Grid_simd SimdApply(const functor &func, const Grid_simd &v) {
     Grid_simd ret;
     Grid_simd::conv_t conv;
-
+    Grid_simd::scalar_type s;
+    
     conv.v = v.v;
     for (int i = 0; i < Nsimd(); i++) {
-      conv.s[i] = func(conv.s[i]);
+      s = conv.s[i];
+      conv.s[i] = func(s);
     }
     ret.v = conv.v;
     return ret;
@@ -350,18 +348,18 @@ class Grid_simd {
     Grid_simd ret;
     Grid_simd::conv_t cx;
     Grid_simd::conv_t cy;
+    Grid_simd::scalar_type sx,sy;
 
     cx.v = x.v;
     cy.v = y.v;
     for (int i = 0; i < Nsimd(); i++) {
-      cx.s[i] = func(cx.s[i], cy.s[i]);
+      sx = cx.s[i];
+      sy = cy.s[i];
+      cx.s[i] = func(sx,sy);
     }
     ret.v = cx.v;
     return ret;
   }
-#if (__GNUC__ == 5 ) || ( ( __GNUC__ == 6 ) && __GNUC_MINOR__ < 3 )
-#pragma GCC pop_options
-#endif
   ///////////////////////
   // Exchange 
   // Al Ah , Bl Bh -> Al Bl Ah,Bh
@@ -423,7 +421,6 @@ class Grid_simd {
   
 };  // end of Grid_simd class definition
 
-
 inline void permute(ComplexD &y,ComplexD b, int perm) {  y=b; }
 inline void permute(ComplexF &y,ComplexF b, int perm) {  y=b; }
 inline void permute(RealD &y,RealD b, int perm) {  y=b; }
@@ -754,8 +751,8 @@ inline Grid_simd<std::complex<R>, V> toComplex(const Grid_simd<R, V> &in) {
 
   conv.v = in.v;
   for (int i = 0; i < Rsimd::Nsimd(); i += 2) {
-    assert(conv.s[i + 1] ==
-           conv.s[i]);  // trap any cases where real was not duplicated
+    assert(conv.s[i + 1] == conv.s[i]);  
+    // trap any cases where real was not duplicated
     // indicating the SIMD grids of real and imag assignment did not correctly
     // match
     conv.s[i + 1] = 0.0;  // zero imaginary parts
@@ -833,8 +830,6 @@ inline void precisionChange(vComplexD *out,vComplexF *in,int nvec){ precisionCha
 inline void precisionChange(vComplexD *out,vComplexH *in,int nvec){ precisionChange((vRealD *)out,(vRealH *)in,nvec);}
 inline void precisionChange(vComplexF *out,vComplexH *in,int nvec){ precisionChange((vRealF *)out,(vRealH *)in,nvec);}
 
-
-
 // Check our vector types are of an appropriate size.
 #if defined QPX
 static_assert(2*sizeof(SIMD_Ftype) == sizeof(SIMD_Dtype), "SIMD vector lengths incorrect");
@@ -849,21 +844,14 @@ static_assert(sizeof(SIMD_Ftype) == sizeof(SIMD_Itype), "SIMD vector lengths inc
 /////////////////////////////////////////
 template <typename T>
 struct is_simd : public std::false_type {};
-template <>
-struct is_simd<vRealF> : public std::true_type {};
-template <>
-struct is_simd<vRealD> : public std::true_type {};
-template <>
-struct is_simd<vComplexF> : public std::true_type {};
-template <>
-struct is_simd<vComplexD> : public std::true_type {};
-template <>
-struct is_simd<vInteger> : public std::true_type {};
+template <> struct is_simd<vRealF>     : public std::true_type {};
+template <> struct is_simd<vRealD>     : public std::true_type {};
+template <> struct is_simd<vComplexF>  : public std::true_type {};
+template <> struct is_simd<vComplexD>  : public std::true_type {};
+template <> struct is_simd<vInteger>   : public std::true_type {};
 
-template <typename T>
-using IfSimd = Invoke<std::enable_if<is_simd<T>::value, int> >;
-template <typename T>
-using IfNotSimd = Invoke<std::enable_if<!is_simd<T>::value, unsigned> >;
+template <typename T> using IfSimd    = Invoke<std::enable_if<is_simd<T>::value, int> >;
+template <typename T> using IfNotSimd = Invoke<std::enable_if<!is_simd<T>::value, unsigned> >;
 }
 
 #endif

lib/simd/Grid_vector_unops.h

@@ -179,13 +179,6 @@ inline Grid_simd<S, V> div(const Grid_simd<S, V> &r, Integer y) {
 ////////////////////////////////////////////////////////////////////////////
 // Allows us to assign into **conformable** real vectors from complex
 ////////////////////////////////////////////////////////////////////////////
-//  template < class S, class V >
-//  inline auto ComplexRemove(const Grid_simd<S,V> &c) ->
-//  Grid_simd<Grid_simd<S,V>::Real,V> {
-//    Grid_simd<Grid_simd<S,V>::Real,V> ret;
-//    ret.v = c.v;
-//    return ret;
-//  }
 template <class scalar>
 struct AndFunctor {
   scalar operator()(const scalar &x, const scalar &y) const { return x & y; }

lib/tensors/Tensor_class.h

@@ -156,11 +156,18 @@ class iScalar {
 
   // convert from a something to a scalar via constructor of something arg
   template <class T, typename std::enable_if<!isGridTensor<T>::value, T>::type * = nullptr>
-    strong_inline iScalar<vtype> operator=(T arg) {
+  strong_inline iScalar<vtype> operator=(T arg) {
     _internal = arg;
     return *this;
   }
 
+  // Convert elements
+  template <class ttype>
+  strong_inline iScalar<vtype> operator=(iScalar<ttype> &&arg) {
+    _internal = arg._internal;
+    return *this;
+  }
+
   friend std::ostream &operator<<(std::ostream &stream,const iScalar<vtype> &o) {
     stream << "S {" << o._internal << "}";
     return stream;

lib/tensors/Tensor_exp.h

@@ -37,30 +37,108 @@ namespace Grid {
   /////////////////////////////////////////////// 
 
 
-  template<class vtype> inline iScalar<vtype> Exponentiate(const iScalar<vtype>&r, ComplexD alpha ,  Integer Nexp = DEFAULT_MAT_EXP)
+  template<class vtype> inline iScalar<vtype> Exponentiate(const iScalar<vtype>&r, RealD alpha ,  Integer Nexp = DEFAULT_MAT_EXP)
     {
       iScalar<vtype> ret;
       ret._internal = Exponentiate(r._internal, alpha, Nexp);
       return ret;
     }
 
-
-  template<class vtype,int N, typename std::enable_if< GridTypeMapper<vtype>::TensorLevel == 0 >::type * =nullptr> 
-    inline iMatrix<vtype,N> Exponentiate(const iMatrix<vtype,N> &arg, ComplexD alpha  , Integer Nexp = DEFAULT_MAT_EXP )
+template<class vtype, int N> inline iVector<vtype, N> Exponentiate(const iVector<vtype,N>&r, RealD alpha ,  Integer Nexp = DEFAULT_MAT_EXP)
     {
-      iMatrix<vtype,N> unit(1.0);
-      iMatrix<vtype,N> temp(unit);
-      
-      for(int i=Nexp; i>=1;--i){
-	temp *= alpha/ComplexD(i);
-	temp = unit + temp*arg;
-      }
-      
-      return temp;
-      
+      iVector<vtype, N> ret;
+      for (int i = 0; i < N; i++)
+        ret._internal[i] = Exponentiate(r._internal[i], alpha, Nexp);
+      return ret;
     }
 
 
 
+    // Specialisation: Cayley-Hamilton exponential for SU(3)
+    template<class vtype, typename std::enable_if< GridTypeMapper<vtype>::TensorLevel == 0>::type * =nullptr> 
+    inline iMatrix<vtype,3> Exponentiate(const iMatrix<vtype,3> &arg, RealD alpha  , Integer Nexp = DEFAULT_MAT_EXP )
+    {
+    // for SU(3) 2x faster than the std implementation using Nexp=12
+    // notice that it actually computes
+    // exp ( input matrix )
+    // the i sign is coming from outside
+    // input matrix is anti-hermitian NOT hermitian
+      typedef iMatrix<vtype,3> mat;
+      typedef iScalar<vtype> scalar;
+      mat unit(1.0);
+      mat temp(unit);
+      const Complex one_over_three = 1.0 / 3.0;
+      const Complex one_over_two = 1.0 / 2.0;
+
+      scalar c0, c1, tmp, c0max, theta, u, w;
+      scalar xi0, u2, w2, cosw;
+      scalar fden, h0, h1, h2;
+      scalar e2iu, emiu, ixi0, qt;
+      scalar f0, f1, f2;
+      scalar unity(1.0);
+      
+      mat iQ2 = arg*arg*alpha*alpha;
+      mat iQ3 = arg*iQ2*alpha;   
+      // sign in c0 from the conventions on the Ta
+      scalar imQ3, reQ2;
+      imQ3 = imag( trace(iQ3) );
+      reQ2 = real( trace(iQ2) );
+      c0 = -imQ3 * one_over_three;  
+      c1 = -reQ2 * one_over_two;
+
+      // Cayley Hamilton checks to machine precision, tested
+      tmp = c1 * one_over_three;
+      c0max = 2.0 * pow(tmp, 1.5);
+
+      theta = acos(c0 / c0max) * one_over_three;
+      u = sqrt(tmp) * cos(theta);
+      w = sqrt(c1) * sin(theta);
+
+      xi0 = sin(w) / w;
+      u2 = u * u;
+      w2 = w * w;
+      cosw = cos(w);
+
+      ixi0 = timesI(xi0);
+      emiu = cos(u) - timesI(sin(u));
+      e2iu = cos(2.0 * u) + timesI(sin(2.0 * u));
+
+      h0 = e2iu * (u2 - w2) +
+           emiu * ((8.0 * u2 * cosw) + (2.0 * u * (3.0 * u2 + w2) * ixi0));
+      h1 = e2iu * (2.0 * u) - emiu * ((2.0 * u * cosw) - (3.0 * u2 - w2) * ixi0);
+      h2 = e2iu - emiu * (cosw + (3.0 * u) * ixi0);
+
+      fden = unity / (9.0 * u2 - w2);  // reals
+      f0 = h0 * fden;
+      f1 = h1 * fden;
+      f2 = h2 * fden;
+
+      return (f0 * unit + timesMinusI(f1) * arg*alpha - f2 * iQ2);
+    }
+
+
+
+// General exponential
+template<class vtype,int N, typename std::enable_if< GridTypeMapper<vtype>::TensorLevel == 0 >::type * =nullptr> 
+    inline iMatrix<vtype,N> Exponentiate(const iMatrix<vtype,N> &arg, RealD alpha  , Integer Nexp = DEFAULT_MAT_EXP )
+    {
+    // notice that it actually computes
+    // exp ( input matrix )
+    // the i sign is coming from outside
+    // input matrix is anti-hermitian NOT hermitian
+      typedef iMatrix<vtype,N> mat;
+      mat unit(1.0);
+      mat temp(unit);
+      for(int i=Nexp; i>=1;--i){
+	      temp *= alpha/RealD(i);
+	      temp = unit + temp*arg;
+      }
+      return temp;
+
+    }
+
+
+
+
 }
 #endif

lib/tensors/Tensor_index.h

@@ -47,6 +47,28 @@ template<int Level>
 class TensorIndexRecursion {
 
  public:
+
+  ////////////////////////////////////////////////////
+  // Type Queries
+  ////////////////////////////////////////////////////
+  template<class vtype>       static inline int indexRank(const iScalar<vtype> tmp)  { return TensorIndexRecursion<Level-1>::indexRank(tmp._internal);  }
+  template<class vtype,int N> static inline int indexRank(const iVector<vtype,N> tmp){ return TensorIndexRecursion<Level-1>::indexRank(tmp._internal[0]);  }
+  template<class vtype,int N> static inline int indexRank(const iMatrix<vtype,N> tmp){ return TensorIndexRecursion<Level-1>::indexRank(tmp._internal[0][0]);  }
+
+  template<class vtype>       static inline int isScalar(const iScalar<vtype> tmp)  { return TensorIndexRecursion<Level-1>::isScalar(tmp._internal);  }
+  template<class vtype,int N> static inline int isScalar(const iVector<vtype,N> tmp){ return TensorIndexRecursion<Level-1>::isScalar(tmp._internal[0]);  }
+  template<class vtype,int N> static inline int isScalar(const iMatrix<vtype,N> tmp){ return TensorIndexRecursion<Level-1>::isScalar(tmp._internal[0][0]);  }
+
+  template<class vtype>       static inline int isVector(const iScalar<vtype> tmp)  { return TensorIndexRecursion<Level-1>::isVector(tmp._internal);  }
+  template<class vtype,int N> static inline int isVector(const iVector<vtype,N> tmp){ return TensorIndexRecursion<Level-1>::isVector(tmp._internal[0]);  }
+  template<class vtype,int N> static inline int isVector(const iMatrix<vtype,N> tmp){ return TensorIndexRecursion<Level-1>::isVector(tmp._internal[0][0]);  }
+
+  template<class vtype>       static inline int isMatrix(const iScalar<vtype> tmp)  { return TensorIndexRecursion<Level-1>::isMatrix(tmp._internal);  }
+  template<class vtype,int N> static inline int isMatrix(const iVector<vtype,N> tmp){ return TensorIndexRecursion<Level-1>::isMatrix(tmp._internal[0]);  }
+  template<class vtype,int N> static inline int isMatrix(const iMatrix<vtype,N> tmp){ return TensorIndexRecursion<Level-1>::isMatrix(tmp._internal[0][0]);  }
+  ////////////////////////////////////////////////////
+  // Trace
+  ////////////////////////////////////////////////////
   template<class vtype>
   static auto traceIndex(const iScalar<vtype> arg) ->  iScalar<decltype(TensorIndexRecursion<Level-1>::traceIndex(arg._internal))> 
   {
@@ -215,6 +237,24 @@ class TensorIndexRecursion {
 template<>
 class TensorIndexRecursion<0> {
  public:
+  ////////////////////////////////////////////////////
+  // Type Queries
+  ////////////////////////////////////////////////////
+  template<class vtype>       static inline int indexRank(const iScalar<vtype> tmp)  { return 1; }
+  template<class vtype,int N> static inline int indexRank(const iVector<vtype,N> tmp){ return N; }
+  template<class vtype,int N> static inline int indexRank(const iMatrix<vtype,N> tmp){ return N; }
+
+  template<class vtype>       static inline int isScalar(const iScalar<vtype> tmp)  { return true;}
+  template<class vtype,int N> static inline int isScalar(const iVector<vtype,N> tmp){ return false;}
+  template<class vtype,int N> static inline int isScalar(const iMatrix<vtype,N> tmp){ return false;}
+
+  template<class vtype>       static inline int isVector(const iScalar<vtype> tmp)  { return false;}
+  template<class vtype,int N> static inline int isVector(const iVector<vtype,N> tmp){ return true;}
+  template<class vtype,int N> static inline int isVector(const iMatrix<vtype,N> tmp){ return false;}
+
+  template<class vtype>       static inline int isMatrix(const iScalar<vtype> tmp)  { return false;}
+  template<class vtype,int N> static inline int isMatrix(const iVector<vtype,N> tmp){ return false;}
+  template<class vtype,int N> static inline int isMatrix(const iMatrix<vtype,N> tmp){ return true;}
 
   /////////////////////////////////////////
   // Ends recursion for trace (scalar/vector/matrix)
@@ -302,6 +342,26 @@ class TensorIndexRecursion<0> {
 ////////////////////////////////////////////////////////////////////////////////////////////////////////
 // External wrappers
 ////////////////////////////////////////////////////////////////////////////////////////////////////////
+template<int Level,class vtype> inline int indexRank(void)
+{
+  vtype tmp;
+  return TensorIndexRecursion<Level>::indexRank(tmp);
+}
+template<int Level,class vtype> inline int isScalar(void)
+{
+  vtype tmp;
+  return TensorIndexRecursion<Level>::isScalar(tmp);
+}
+template<int Level,class vtype> inline int isVector(void)
+{
+  vtype tmp;
+  return TensorIndexRecursion<Level>::isVector(tmp);
+}
+template<int Level,class vtype> inline int isMatrix(void)
+{
+  vtype tmp;
+  return TensorIndexRecursion<Level>::isMatrix(tmp);
+}
 
 template<int Level,class vtype> inline auto traceIndex (const vtype &arg) -> RemoveCRV(TensorIndexRecursion<Level>::traceIndex(arg))
 {

lib/tensors/Tensor_traits.h

@@ -281,8 +281,8 @@ namespace Grid {
   template<typename T>
   class getPrecision{
   public:
-    typedef typename getVectorType<T>::type vector_obj; //get the vector_obj (i.e. a grid Tensor) if its a Lattice<vobj>, do nothing otherwise (i.e. if fundamental or grid Tensor)
-  
+    //get the vector_obj (i.e. a grid Tensor) if its a Lattice<vobj>, do nothing otherwise (i.e. if fundamental or grid Tensor)
+    typedef typename getVectorType<T>::type vector_obj; 
     typedef typename GridTypeMapper<vector_obj>::scalar_type scalar_type; //get the associated scalar type. Works on fundamental and tensor types
     typedef typename GridTypeMapper<scalar_type>::Realified real_scalar_type; //remove any std::complex wrapper, should get us to the fundamental type
 

tests/IO/Test_ildg_io.cc

@@ -0,0 +1,99 @@
+    /*************************************************************************************
+
+    Grid physics library, www.github.com/paboyle/Grid 
+
+    Source file: ./tests/Test_nersc_io.cc
+
+    Copyright (C) 2015
+
+Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
+Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+Author: paboyle <paboyle@ph.ed.ac.uk>
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License along
+    with this program; if not, write to the Free Software Foundation, Inc.,
+    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+    See the full license in the file "LICENSE" in the top level distribution directory
+    *************************************************************************************/
+    /*  END LEGAL */
+#include <Grid/Grid.h>
+
+using namespace std;
+using namespace Grid;
+using namespace Grid::QCD;
+
+
+int main (int argc, char ** argv)
+{
+  Grid_init(&argc,&argv);
+
+  std::cout <<GridLogMessage<< " main "<<std::endl;
+
+  std::vector<int> simd_layout = GridDefaultSimd(4,vComplex::Nsimd());
+  std::vector<int> mpi_layout  = GridDefaultMpi();
+  //std::vector<int> latt_size  ({48,48,48,96});
+  //std::vector<int> latt_size  ({32,32,32,32});
+  std::vector<int> latt_size  ({16,16,16,32});
+  std::vector<int> clatt_size  ({4,4,4,8});
+  int orthodir=3;
+  int orthosz =latt_size[orthodir];
+    
+  GridCartesian     Fine(latt_size,simd_layout,mpi_layout);
+  GridCartesian     Coarse(clatt_size,simd_layout,mpi_layout);
+
+
+  GridParallelRNG   pRNGa(&Fine);
+  GridParallelRNG   pRNGb(&Fine);
+  GridSerialRNG     sRNGa;
+  GridSerialRNG     sRNGb;
+
+  std::cout <<GridLogMessage<< " seeding... "<<std::endl;
+  pRNGa.SeedFixedIntegers(std::vector<int>({45,12,81,9}));
+  sRNGa.SeedFixedIntegers(std::vector<int>({45,12,81,9}));
+  std::cout <<GridLogMessage<< " ...done "<<std::endl;
+
+  LatticeGaugeField Umu(&Fine);
+  LatticeGaugeField Umu_diff(&Fine);
+  LatticeGaugeField Umu_saved(&Fine);
+
+  std::vector<LatticeColourMatrix> U(4,&Fine);
+  
+  SU3::HotConfiguration(pRNGa,Umu);
+
+
+  FieldMetaData header;
+
+  std::cout <<GridLogMessage<<"**************************************"<<std::endl;
+  std::cout <<GridLogMessage<<"** Writing out  ILDG conf    *********"<<std::endl;
+  std::cout <<GridLogMessage<<"**************************************"<<std::endl;
+  std::string file("./ckpoint_ildg.4000");
+  IldgWriter _IldgWriter;
+  _IldgWriter.open(file);
+  _IldgWriter.writeConfiguration(Umu,4000,std::string("dummy_ildg_LFN"),std::string("dummy_config"));
+  _IldgWriter.close();
+
+  Umu_saved = Umu;
+  std::cout <<GridLogMessage<<"**************************************"<<std::endl;
+  std::cout <<GridLogMessage<<"** Reading back ILDG conf    *********"<<std::endl;
+  std::cout <<GridLogMessage<<"**************************************"<<std::endl;
+  IldgReader _IldgReader;
+  _IldgReader.open(file);
+  _IldgReader.readConfiguration(Umu,header);
+  _IldgReader.close();
+  Umu_diff = Umu - Umu_saved;
+
+  std::cout <<GridLogMessage<< "norm2 Gauge Diff = "<<norm2(Umu_diff)<<std::endl;
+
+  Grid_finalize();
+}

tests/IO/Test_ildg_read.cc

@@ -0,0 +1,115 @@
+    /*************************************************************************************
+
+    Grid physics library, www.github.com/paboyle/Grid 
+
+    Source file: ./tests/Test_nersc_io.cc
+
+    Copyright (C) 2015
+
+Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
+Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+Author: paboyle <paboyle@ph.ed.ac.uk>
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License along
+    with this program; if not, write to the Free Software Foundation, Inc.,
+    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+    See the full license in the file "LICENSE" in the top level distribution directory
+    *************************************************************************************/
+    /*  END LEGAL */
+#include <Grid/Grid.h>
+
+using namespace std;
+using namespace Grid;
+using namespace Grid::QCD;
+
+
+int main (int argc, char ** argv)
+{
+  Grid_init(&argc,&argv);
+
+
+  std::vector<int> simd_layout = GridDefaultSimd(4,vComplex::Nsimd());
+  std::vector<int> mpi_layout  = GridDefaultMpi();
+  std::vector<int> latt_size = GridDefaultLatt();
+  int orthodir=3;
+  int orthosz =latt_size[orthodir];
+    
+  GridCartesian     Fine(latt_size,simd_layout,mpi_layout);
+
+  LatticeGaugeField Umu(&Fine);
+  std::vector<LatticeColourMatrix> U(4,&Fine);
+  
+  FieldMetaData header;
+  std::string file("./ildg.file");
+  IldgReader IR;
+  IR.open(file);
+  IR.readConfiguration(Umu,header);
+  IR.close();
+
+  for(int mu=0;mu<Nd;mu++){
+    U[mu] = PeekIndex<LorentzIndex>(Umu,mu);
+  }
+
+  // Painful ; fix syntactical niceness
+  LatticeComplex LinkTrace(&Fine);
+  LinkTrace=zero;
+  for(int mu=0;mu<Nd;mu++){
+    LinkTrace = LinkTrace + trace(U[mu]);
+  }
+
+  // (1+2+3)=6 = N(N-1)/2 terms
+  LatticeComplex Plaq(&Fine);
+
+  Plaq = zero;
+
+  for(int mu=1;mu<Nd;mu++){
+    for(int nu=0;nu<mu;nu++){
+      Plaq = Plaq + trace(U[mu]*Cshift(U[nu],mu,1)*adj(Cshift(U[mu],nu,1))*adj(U[nu]));
+    }
+  }
+
+  double vol = Fine.gSites();
+  Complex PlaqScale(1.0/vol/6.0/3.0);
+  std::cout<<GridLogMessage <<"PlaqScale" << PlaqScale<<std::endl;
+
+  std::vector<TComplex> Plaq_T(orthosz);
+  sliceSum(Plaq,Plaq_T,Nd-1);
+  int Nt = Plaq_T.size();
+
+  TComplex Plaq_T_sum; 
+  Plaq_T_sum=zero;
+  for(int t=0;t<Nt;t++){
+    Plaq_T_sum = Plaq_T_sum+Plaq_T[t];
+    Complex Pt=TensorRemove(Plaq_T[t]);
+    std::cout<<GridLogMessage << "sliced ["<<t<<"]" <<Pt*PlaqScale*Real(Nt)<<std::endl;
+  }
+
+  {
+    Complex Pt = TensorRemove(Plaq_T_sum);
+    std::cout<<GridLogMessage << "total " <<Pt*PlaqScale<<std::endl;
+  }  
+
+
+  TComplex Tp = sum(Plaq);
+  Complex p  = TensorRemove(Tp);
+  std::cout<<GridLogMessage << "calculated plaquettes " <<p*PlaqScale<<std::endl;
+
+
+  Complex LinkTraceScale(1.0/vol/4.0/3.0);
+  TComplex Tl = sum(LinkTrace);
+  Complex l  = TensorRemove(Tl);
+  std::cout<<GridLogMessage << "calculated link trace " <<l*LinkTraceScale<<std::endl;
+
+  Grid_finalize();
+}

tests/IO/Test_nersc_io.cc

@@ -38,10 +38,13 @@ int main (int argc, char ** argv)
 {
   Grid_init(&argc,&argv);
 
+  std::cout <<GridLogMessage<< " main "<<std::endl;
 
   std::vector<int> simd_layout = GridDefaultSimd(4,vComplex::Nsimd());
   std::vector<int> mpi_layout  = GridDefaultMpi();
-  std::vector<int> latt_size  ({16,16,16,16});
+  //std::vector<int> latt_size  ({48,48,48,96});
+  //std::vector<int> latt_size  ({32,32,32,32});
+  std::vector<int> latt_size  ({16,16,16,32});
   std::vector<int> clatt_size  ({4,4,4,8});
   int orthodir=3;
   int orthosz =latt_size[orthodir];
@@ -49,30 +52,32 @@ int main (int argc, char ** argv)
   GridCartesian     Fine(latt_size,simd_layout,mpi_layout);
   GridCartesian     Coarse(clatt_size,simd_layout,mpi_layout);
 
+
   GridParallelRNG   pRNGa(&Fine);
   GridParallelRNG   pRNGb(&Fine);
   GridSerialRNG     sRNGa;
   GridSerialRNG     sRNGb;
 
+  std::cout <<GridLogMessage<< " seeding... "<<std::endl;
   pRNGa.SeedFixedIntegers(std::vector<int>({45,12,81,9}));
   sRNGa.SeedFixedIntegers(std::vector<int>({45,12,81,9}));
-  
+  std::cout <<GridLogMessage<< " ...done "<<std::endl;
+
   std::string rfile("./ckpoint_rng.4000");
+  FieldMetaData rngheader;
   NerscIO::writeRNGState(sRNGa,pRNGa,rfile);
-  NerscField rngheader;
   NerscIO::readRNGState (sRNGb,pRNGb,rngheader,rfile);
 
   LatticeComplex tmpa(&Fine); random(pRNGa,tmpa);
   LatticeComplex tmpb(&Fine); random(pRNGb,tmpb);
   tmpa = tmpa - tmpb;
-  std::cout << " difference between restored randoms and orig "<<norm2( tmpa ) <<" / "<< norm2(tmpb)<<std::endl;
+  std::cout <<GridLogMessage<< " difference between restored randoms and orig "<<norm2( tmpa ) <<" / "<< norm2(tmpb)<<std::endl;
 
   ComplexD a,b;
 
   random(sRNGa,a);
   random(sRNGb,b);
-  std::cout << " serial RNG numbers "<<a<<" "<<b<<std::endl;
-
+  std::cout <<GridLogMessage<< " serial RNG numbers "<<a<<" "<<b<<std::endl;
 
   LatticeGaugeField Umu(&Fine);
   LatticeGaugeField Umu_diff(&Fine);
@@ -80,15 +85,20 @@ int main (int argc, char ** argv)
 
   std::vector<LatticeColourMatrix> U(4,&Fine);
   
-  SU3::ColdConfiguration(pRNGa,Umu);
+  SU3::HotConfiguration(pRNGa,Umu);
 
-  NerscField header;
+  FieldMetaData header;
   std::string file("./ckpoint_lat.4000");
 
   int precision32 = 0;
   int tworow      = 0;
   NerscIO::writeConfiguration(Umu,file,tworow,precision32);
+  Umu_saved = Umu;
   NerscIO::readConfiguration(Umu,header,file);
+  Umu_diff = Umu - Umu_saved;
+  //std::cout << "Umu_save "<<Umu_saved[0]<<std::endl;
+  //std::cout << "Umu_read "<<Umu[0]<<std::endl;
+  std::cout <<GridLogMessage<< "norm2 Gauge Diff = "<<norm2(Umu_diff)<<std::endl;
 
   for(int mu=0;mu<Nd;mu++){
     U[mu] = PeekIndex<LorentzIndex>(Umu,mu);
@@ -115,7 +125,6 @@ int main (int argc, char ** argv)
 #endif
   double vol = Fine.gSites();
   Complex PlaqScale(1.0/vol/6.0/3.0);
-  std::cout<<GridLogMessage <<"PlaqScale" << PlaqScale<<std::endl;
 
   std::vector<TComplex> Plaq_T(orthosz);
   sliceSum(Plaq,Plaq_T,Nd-1);
@@ -139,7 +148,6 @@ 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);
   TComplex Tl = sum(LinkTrace);
   Complex l  = TensorRemove(Tl);

tests/IO/Test_nersc_read.cc

@@ -50,7 +50,7 @@ int main (int argc, char ** argv)
   LatticeGaugeField Umu(&Fine);
   std::vector<LatticeColourMatrix> U(4,&Fine);
   
-  NerscField header;
+  FieldMetaData header;
   std::string file("./ckpoint_lat");
   NerscIO::readConfiguration(Umu,header,file);
 

tests/IO/Test_serialisation.cc

@@ -31,6 +31,7 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 
 
 using namespace Grid;
+using namespace Grid::QCD;
 
 GRID_SERIALIZABLE_ENUM(myenum, undef, red, 1, blue, 2, green, 3);
   
@@ -44,8 +45,8 @@ public:
                           double, y,
                           bool , b,
                           std::vector<double>, array,
-                          std::vector<std::vector<double>>, twodimarray,
-                          std::vector<std::vector<std::vector<Complex>>>, cmplx3darray
+                          std::vector<std::vector<double> >, twodimarray,
+                          std::vector<std::vector<std::vector<Complex> > >, cmplx3darray
                           );
   myclass() {}
   myclass(int i)
@@ -237,7 +238,7 @@ int main(int argc,char **argv)
     std::cout << "Loaded (JSON) -----------------" << std::endl;
     std::cout << jcopy1 << std::endl << jveccopy1 << std::endl;
   }
-
+  
 /* 
   // This is still work in progress
   {

tests/Test_dwf_mixedcg_prec_halfcomms.cc

@@ -0,0 +1,110 @@
+    /*************************************************************************************
+
+    Grid physics library, www.github.com/paboyle/Grid 
+
+    Source file: ./tests/Test_dwf_cg_prec.cc
+
+    Copyright (C) 2015
+
+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;
+using namespace Grid::QCD;
+
+template<class d>
+struct scal {
+  d internal;
+};
+
+  Gamma::Algebra Gmu [] = {
+    Gamma::Algebra::GammaX,
+    Gamma::Algebra::GammaY,
+    Gamma::Algebra::GammaZ,
+    Gamma::Algebra::GammaT
+  };
+
+int main (int argc, char ** argv)
+{
+  Grid_init(&argc,&argv);
+
+  const int Ls=24;
+
+  GridCartesian         * UGrid   = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplexD::Nsimd()),GridDefaultMpi());
+  GridRedBlackCartesian * UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
+  GridCartesian         * FGrid   = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
+  GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid);
+
+  GridCartesian         * UGrid_f   = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplexF::Nsimd()),GridDefaultMpi());
+  GridRedBlackCartesian * UrbGrid_f = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid_f);
+  GridCartesian         * FGrid_f   = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid_f);
+  GridRedBlackCartesian * FrbGrid_f = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid_f);
+  
+  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);
+
+  LatticeFermionD    src(FGrid); random(RNG5,src);
+  LatticeFermionD result(FGrid); result=zero;
+  LatticeGaugeFieldD Umu(UGrid);
+  LatticeGaugeFieldF Umu_f(UGrid_f); 
+  
+  SU3::HotConfiguration(RNG4,Umu);
+
+  precisionChange(Umu_f,Umu);
+  
+  RealD mass=0.1;
+  RealD M5=1.8;
+  DomainWallFermionD Ddwf(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
+  DomainWallFermionFH Ddwf_f(Umu_f,*FGrid_f,*FrbGrid_f,*UGrid_f,*UrbGrid_f,mass,M5);
+
+
+  LatticeFermionD    src_o(FrbGrid);
+  LatticeFermionD result_o(FrbGrid);
+  LatticeFermionD result_o_2(FrbGrid);
+  pickCheckerboard(Odd,src_o,src);
+  result_o.checkerboard = Odd;
+  result_o = zero;
+  result_o_2.checkerboard = Odd;
+  result_o_2 = zero;
+
+  SchurDiagMooeeOperator<DomainWallFermionD,LatticeFermionD> HermOpEO(Ddwf);
+  SchurDiagMooeeOperator<DomainWallFermionFH,LatticeFermionF> HermOpEO_f(Ddwf_f);
+
+  std::cout << "Starting mixed CG" << std::endl;
+  MixedPrecisionConjugateGradient<LatticeFermionD,LatticeFermionF> mCG(1.0e-8, 10000, 50, FrbGrid_f, HermOpEO_f, HermOpEO);
+  mCG.InnerTolerance = 3.0e-5;
+  mCG(src_o,result_o);
+
+  std::cout << "Starting regular CG" << std::endl;
+  ConjugateGradient<LatticeFermionD> CG(1.0e-8,10000);
+  CG(HermOpEO,src_o,result_o_2);
+
+  LatticeFermionD diff_o(FrbGrid);
+  RealD diff = axpy_norm(diff_o, -1.0, result_o, result_o_2);
+
+  std::cout << "Diff between mixed and regular CG: " << diff << std::endl;
+
+  
+  Grid_finalize();
+}

tests/core/Test_GaugeAction.cc

@@ -73,7 +73,7 @@ int main (int argc, char ** argv)
 
   std::vector<LatticeColourMatrix> U(4,&Fine);
   
-  NerscField header;
+  FieldMetaData header;
   
   std::string file("./ckpoint_lat.4000");
   NerscIO::readConfiguration(Umu,header,file);

tests/core/Test_RectPlaq.cc

@@ -90,7 +90,7 @@ int main (int argc, char ** argv)
 
   std::vector<LatticeColourMatrix> U(4,&Fine);
   
-  NerscField header;
+  FieldMetaData header;
   
   std::string file("./ckpoint_lat.4000");
   NerscIO::readConfiguration(Umu,header,file);

tests/core/Test_fft_gfix.cc

@@ -28,212 +28,6 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
     /*  END LEGAL */
 #include <Grid/Grid.h>
 
-using namespace Grid;
-using namespace Grid::QCD;
-
-template <class Gimpl> 
-class FourierAcceleratedGaugeFixer  : public Gimpl {
-  public:
-  INHERIT_GIMPL_TYPES(Gimpl);
-
-  typedef typename Gimpl::GaugeLinkField GaugeMat;
-  typedef typename Gimpl::GaugeField GaugeLorentz;
-
-  static void GaugeLinkToLieAlgebraField(const std::vector<GaugeMat> &U,std::vector<GaugeMat> &A) {
-    for(int mu=0;mu<Nd;mu++){
-//      ImplComplex cmi(0.0,-1.0);
-      Complex cmi(0.0,-1.0);
-      A[mu] = Ta(U[mu]) * cmi;
-    }
-  }
-  static void DmuAmu(const std::vector<GaugeMat> &A,GaugeMat &dmuAmu) {
-    dmuAmu=zero;
-    for(int mu=0;mu<Nd;mu++){
-      dmuAmu = dmuAmu + A[mu] - Cshift(A[mu],mu,-1);
-    }
-  }  
-  static void SteepestDescentGaugeFix(GaugeLorentz &Umu,Real & alpha,int maxiter,Real Omega_tol, Real Phi_tol) {
-    GridBase *grid = Umu._grid;
-
-    Real org_plaq      =WilsonLoops<Gimpl>::avgPlaquette(Umu);
-    Real org_link_trace=WilsonLoops<Gimpl>::linkTrace(Umu); 
-    Real old_trace = org_link_trace;
-    Real trG;
-
-    std::vector<GaugeMat> U(Nd,grid);
-                 GaugeMat dmuAmu(grid);
-
-    for(int i=0;i<maxiter;i++){
-      for(int mu=0;mu<Nd;mu++) U[mu]= PeekIndex<LorentzIndex>(Umu,mu);
-      //trG = SteepestDescentStep(U,alpha,dmuAmu);
-      trG = FourierAccelSteepestDescentStep(U,alpha,dmuAmu);
-      for(int mu=0;mu<Nd;mu++) PokeIndex<LorentzIndex>(Umu,U[mu],mu);
-      // Monitor progress and convergence test 
-      // infrequently to minimise cost overhead
-      if ( i %20 == 0 ) { 
-	Real plaq      =WilsonLoops<Gimpl>::avgPlaquette(Umu);
-	Real link_trace=WilsonLoops<Gimpl>::linkTrace(Umu); 
-
-	std::cout << GridLogMessage << " Iteration "<<i<< " plaq= "<<plaq<< " dmuAmu " << norm2(dmuAmu)<< std::endl;
-	
-	Real Phi  = 1.0 - old_trace / link_trace ;
-	Real Omega= 1.0 - trG;
-
-
-	std::cout << GridLogMessage << " Iteration "<<i<< " Phi= "<<Phi<< " Omega= " << Omega<< " trG " << trG <<std::endl;
-	if ( (Omega < Omega_tol) && ( ::fabs(Phi) < Phi_tol) ) {
-	  std::cout << GridLogMessage << "Converged ! "<<std::endl;
-	  return;
-	}
-
-	old_trace = link_trace;
-
-      }
-    }
-  };
-  static Real SteepestDescentStep(std::vector<GaugeMat> &U,Real & alpha, GaugeMat & dmuAmu) {
-    GridBase *grid = U[0]._grid;
-
-    std::vector<GaugeMat> A(Nd,grid);
-    GaugeMat g(grid);
-
-    GaugeLinkToLieAlgebraField(U,A);
-    ExpiAlphaDmuAmu(A,g,alpha,dmuAmu);
-
-
-    Real vol = grid->gSites();
-    Real trG = TensorRemove(sum(trace(g))).real()/vol/Nc;
-
-    SU<Nc>::GaugeTransform(U,g);
-
-    return trG;
-  }
-
-  static Real FourierAccelSteepestDescentStep(std::vector<GaugeMat> &U,Real & alpha, GaugeMat & dmuAmu) {
-
-    GridBase *grid = U[0]._grid;
-
-    Real vol = grid->gSites();
-
-    FFT theFFT((GridCartesian *)grid);
-
-    LatticeComplex  Fp(grid);
-    LatticeComplex  psq(grid); psq=zero;
-    LatticeComplex  pmu(grid); 
-    LatticeComplex   one(grid); one = Complex(1.0,0.0);
-
-    GaugeMat g(grid);
-    GaugeMat dmuAmu_p(grid);
-    std::vector<GaugeMat> A(Nd,grid);
-
-    GaugeLinkToLieAlgebraField(U,A);
-
-    DmuAmu(A,dmuAmu);
-
-    theFFT.FFT_all_dim(dmuAmu_p,dmuAmu,FFT::forward);
-
-    //////////////////////////////////
-    // Work out Fp = psq_max/ psq...
-    //////////////////////////////////
-    std::vector<int> latt_size = grid->GlobalDimensions();
-    std::vector<int> coor(grid->_ndimension,0);
-    for(int mu=0;mu<Nd;mu++) {
-
-      Real TwoPiL =  M_PI * 2.0/ latt_size[mu];
-      LatticeCoordinate(pmu,mu);
-      pmu = TwoPiL * pmu ;
-      psq = psq + 4.0*sin(pmu*0.5)*sin(pmu*0.5); 
-    }
-
-    Complex psqMax(16.0);
-    Fp =  psqMax*one/psq;
-
-    /*
-    static int once;
-    if ( once == 0 ) { 
-      std::cout << " Fp " << Fp <<std::endl;
-      once ++;
-      }*/
-
-    pokeSite(TComplex(1.0),Fp,coor);
-
-    dmuAmu_p  = dmuAmu_p * Fp; 
-
-    theFFT.FFT_all_dim(dmuAmu,dmuAmu_p,FFT::backward);
-
-    GaugeMat ciadmam(grid);
-    Complex cialpha(0.0,-alpha);
-    ciadmam = dmuAmu*cialpha;
-    SU<Nc>::taExp(ciadmam,g);
-
-    Real trG = TensorRemove(sum(trace(g))).real()/vol/Nc;
-
-    SU<Nc>::GaugeTransform(U,g);
-
-    return trG;
-  }
-
-  static void ExpiAlphaDmuAmu(const std::vector<GaugeMat> &A,GaugeMat &g,Real & alpha, GaugeMat &dmuAmu) {
-    GridBase *grid = g._grid;
-    Complex cialpha(0.0,-alpha);
-    GaugeMat ciadmam(grid);
-    DmuAmu(A,dmuAmu);
-    ciadmam = dmuAmu*cialpha;
-    SU<Nc>::taExp(ciadmam,g);
-  }  
-/*
-  ////////////////////////////////////////////////////////////////
-  // NB The FT for fields living on links has an extra phase in it
-  // Could add these to the FFT class as a later task since this code
-  // might be reused elsewhere ????
-  ////////////////////////////////////////////////////////////////
-  static void InverseFourierTransformAmu(FFT &theFFT,const std::vector<GaugeMat> &Ap,std::vector<GaugeMat> &Ax) {
-    GridBase * grid = theFFT.Grid();
-    std::vector<int> latt_size = grid->GlobalDimensions();
-
-    ComplexField  pmu(grid);
-    ComplexField  pha(grid);
-    GaugeMat      Apha(grid);
-
-    Complex ci(0.0,1.0);
-
-    for(int mu=0;mu<Nd;mu++){
-
-      Real TwoPiL =  M_PI * 2.0/ latt_size[mu];
-      LatticeCoordinate(pmu,mu);
-      pmu = TwoPiL * pmu ;
-      pha = exp(pmu *  (0.5 *ci)); // e(ipmu/2) since Amu(x+mu/2)
-
-      Apha = Ap[mu] * pha;
-
-      theFFT.FFT_all_dim(Apha,Ax[mu],FFT::backward);
-    }
-  }
-  static void FourierTransformAmu(FFT & theFFT,const std::vector<GaugeMat> &Ax,std::vector<GaugeMat> &Ap) {
-    GridBase * grid = theFFT.Grid();
-    std::vector<int> latt_size = grid->GlobalDimensions();
-
-    ComplexField  pmu(grid);
-    ComplexField  pha(grid);
-    Complex ci(0.0,1.0);
-    
-    // Sign convention for FFTW calls:
-    // A(x)= Sum_p e^ipx A(p) / V
-    // A(p)= Sum_p e^-ipx A(x)
-
-    for(int mu=0;mu<Nd;mu++){
-      Real TwoPiL =  M_PI * 2.0/ latt_size[mu];
-      LatticeCoordinate(pmu,mu);
-      pmu = TwoPiL * pmu ;
-      pha = exp(-pmu *  (0.5 *ci)); // e(+ipmu/2) since Amu(x+mu/2)
-
-      theFFT.FFT_all_dim(Ax[mu],Ap[mu],FFT::backward);
-      Ap[mu] = Ap[mu] * pha;
-    }
-  }
-*/
-};
-
 int main (int argc, char ** argv)
 {
   std::vector<int> seeds({1,2,3,4});
@@ -264,22 +58,24 @@ int main (int argc, char ** argv)
   std::cout<< "*****************************************************************" <<std::endl;
 
   LatticeGaugeField   Umu(&GRID);
+  LatticeGaugeField   Urnd(&GRID);
   LatticeGaugeField   Uorg(&GRID);
   LatticeColourMatrix   g(&GRID); // Gauge xform
 
   
   SU3::ColdConfiguration(pRNG,Umu); // Unit gauge
   Uorg=Umu;
+  Urnd=Umu;
+
+  SU3::RandomGaugeTransform(pRNG,Urnd,g); // Unit gauge
 
-  SU3::RandomGaugeTransform(pRNG,Umu,g); // Unit gauge
   Real plaq=WilsonLoops<PeriodicGimplR>::avgPlaquette(Umu);
   std::cout << " Initial plaquette "<<plaq << std::endl;
 
-
-
   Real alpha=0.1;
-  FourierAcceleratedGaugeFixer<PeriodicGimplR>::SteepestDescentGaugeFix(Umu,alpha,10000,1.0e-10, 1.0e-10);
 
+  Umu = Urnd;
+  FourierAcceleratedGaugeFixer<PeriodicGimplR>::SteepestDescentGaugeFix(Umu,alpha,10000,1.0e-12, 1.0e-12,false);
 
   plaq=WilsonLoops<PeriodicGimplR>::avgPlaquette(Umu);
   std::cout << " Final plaquette "<<plaq << std::endl;
@@ -288,14 +84,28 @@ int main (int argc, char ** argv)
   std::cout << " Norm Difference "<< norm2(Uorg) << std::endl;
 
 
-  //  std::cout<< "*****************************************************************" <<std::endl;
-  //  std::cout<< "* Testing Fourier accelerated fixing                            *" <<std::endl;
-  //  std::cout<< "*****************************************************************" <<std::endl;
+  std::cout<< "*****************************************************************" <<std::endl;
+  std::cout<< "* Testing Fourier accelerated fixing                            *" <<std::endl;
+  std::cout<< "*****************************************************************" <<std::endl;
+  Umu=Urnd;
+  FourierAcceleratedGaugeFixer<PeriodicGimplR>::SteepestDescentGaugeFix(Umu,alpha,10000,1.0e-12, 1.0e-12,true);
 
-  //  std::cout<< "*****************************************************************" <<std::endl;
-  //  std::cout<< "* Testing non-unit configuration                                *" <<std::endl;
-  //  std::cout<< "*****************************************************************" <<std::endl;
+  plaq=WilsonLoops<PeriodicGimplR>::avgPlaquette(Umu);
+  std::cout << " Final plaquette "<<plaq << std::endl;
 
+  std::cout<< "*****************************************************************" <<std::endl;
+  std::cout<< "* Testing non-unit configuration                                *" <<std::endl;
+  std::cout<< "*****************************************************************" <<std::endl;
+
+  SU3::HotConfiguration(pRNG,Umu); // Unit gauge
+
+  plaq=WilsonLoops<PeriodicGimplR>::avgPlaquette(Umu);
+  std::cout << " Initial plaquette "<<plaq << std::endl;
+
+  FourierAcceleratedGaugeFixer<PeriodicGimplR>::SteepestDescentGaugeFix(Umu,alpha,10000,1.0e-12, 1.0e-12,true);
+
+  plaq=WilsonLoops<PeriodicGimplR>::avgPlaquette(Umu);
+  std::cout << " Final plaquette "<<plaq << std::endl;
 
 
   Grid_finalize();

tests/core/Test_main.cc

@@ -336,7 +336,7 @@ int main(int argc, char **argv) {
 
       std::cout << GridLogMessage << "norm cMmat : " << norm2(cMat)
                 << std::endl;
-      cMat = expMat(cMat, ComplexD(1.0, 0.0));
+      cMat = expMat(cMat,1.0);// ComplexD(1.0, 0.0));
       std::cout << GridLogMessage << "norm expMat: " << norm2(cMat)
                 << std::endl;
       peekSite(cm, cMat, mysite);

tests/debug/Test_cayley_ldop_cr.cc

@@ -67,7 +67,7 @@ int main (int argc, char ** argv)
   LatticeFermion    err(FGrid);
   LatticeGaugeField Umu(UGrid); 
 
-  NerscField header;
+  FieldMetaData header;
   std::string file("./ckpoint_lat.400");
   NerscIO::readConfiguration(Umu,header,file);
 

tests/debug/Test_synthetic_lanczos.cc

@@ -133,8 +133,8 @@ int main (int argc, char ** argv)
   int Nconv;
   RealD eresid = 1.0e-6;
 
-  ImplicitlyRestartedLanczos<LatticeComplex> IRL(HermOp,X,Nk,Nm,eresid,Nit);
-  ImplicitlyRestartedLanczos<LatticeComplex> ChebyIRL(HermOp,Cheby,Nk,Nm,eresid,Nit);
+  ImplicitlyRestartedLanczos<LatticeComplex> IRL(HermOp,X,Nk,Nk,Nm,eresid,Nit);
+  ImplicitlyRestartedLanczos<LatticeComplex> ChebyIRL(HermOp,Cheby,Nk,Nk,Nm,eresid,Nit);
 
   LatticeComplex src(grid); gaussian(RNG,src);
   {

tests/forces/Test_contfrac_force.cc

@@ -139,7 +139,7 @@ int main (int argc, char ** argv)
 
   }
 
-  Complex dSpred    = sum(dS);
+  ComplexD dSpred    = sum(dS);
 
   std::cout << GridLogMessage << " S      "<<S<<std::endl;
   std::cout << GridLogMessage << " Sprime "<<Sprime<<std::endl;

tests/forces/Test_dwf_force.cc

@@ -150,7 +150,7 @@ int main (int argc, char ** argv)
 
   }
 
-  Complex dSpred    = sum(dS);
+  ComplexD dSpred    = sum(dS);
 
   std::cout << GridLogMessage << " S      "<<S<<std::endl;
   std::cout << GridLogMessage << " Sprime "<<Sprime<<std::endl;

tests/forces/Test_dwf_gpforce.cc

@@ -194,9 +194,9 @@ int main (int argc, char ** argv)
 
   }
 
-  Complex dSpred    = sum(dS);
-  Complex dSm       = sum(dSmom);
-  Complex dSm2      = sum(dSmom2);
+  ComplexD dSpred    = sum(dS);
+  ComplexD dSm       = sum(dSmom);
+  ComplexD dSm2      = sum(dSmom2);
 
 
   std::cout << GridLogMessage <<"Initial mom hamiltonian is "<< Hmom <<std::endl;

tests/forces/Test_gp_rect_force.cc

@@ -113,7 +113,7 @@ int main (int argc, char ** argv)
     dS = dS - trace(mommu*UdSdUmu)*dt*2.0;
 
   }
-  Complex dSpred    = sum(dS);
+  ComplexD dSpred    = sum(dS);
 
   std::cout << GridLogMessage << " S      "<<S<<std::endl;
   std::cout << GridLogMessage << " Sprime "<<Sprime<<std::endl;

tests/forces/Test_gpdwf_force.cc

@@ -143,7 +143,7 @@ int main (int argc, char ** argv)
     dS = dS+trace(mommu*forcemu)*dt;
   }
 
-  Complex dSpred    = sum(dS);
+  ComplexD dSpred    = sum(dS);
 
   // From TwoFlavourPseudoFermion:
   //////////////////////////////////////////////////////

tests/forces/Test_gpwilson_force.cc

@@ -143,7 +143,7 @@ int main (int argc, char ** argv)
     dS = dS+trace(mommu*forcemu)*dt;
   }
 
-  Complex dSpred    = sum(dS);
+  ComplexD dSpred    = sum(dS);
 
   std::cout << GridLogMessage << " S      "<<S<<std::endl;
   std::cout << GridLogMessage << " Sprime "<<Sprime<<std::endl;

tests/forces/Test_laplacian_force.cc

@@ -128,7 +128,7 @@ int main (int argc, char ** argv)
     dS = dS + trace(mommu*UdSdUmu)*dt*2.0;
   }
 
-  Complex dSpred    = sum(dS);
+  ComplexD dSpred    = sum(dS);
 
   std::cout << GridLogMessage << " S      "<<S<<std::endl;
   std::cout << GridLogMessage << " Sprime "<<Sprime<<std::endl;

tests/forces/Test_mobius_force.cc

@@ -141,7 +141,7 @@ int main (int argc, char ** argv)
 
   }
 
-  Complex dSpred    = sum(dS);
+  ComplexD dSpred    = sum(dS);
 
   std::cout << GridLogMessage << " -- S         "<<S<<std::endl;
   std::cout << GridLogMessage << " -- Sprime    "<<Sprime<<std::endl;

tests/forces/Test_partfrac_force.cc

@@ -141,7 +141,7 @@ int main (int argc, char ** argv)
 
   }
 
-  Complex dSpred    = sum(dS);
+  ComplexD dSpred    = sum(dS);
 
   std::cout << GridLogMessage << " S      "<<S<<std::endl;
   std::cout << GridLogMessage << " Sprime "<<Sprime<<std::endl;

tests/forces/Test_rect_force.cc

@@ -112,7 +112,7 @@ int main (int argc, char ** argv)
     dS = dS - trace(mommu*UdSdUmu)*dt*2.0;
 
   }
-  Complex dSpred    = sum(dS);
+  ComplexD dSpred    = sum(dS);
 
   std::cout << GridLogMessage << " S      "<<S<<std::endl;
   std::cout << GridLogMessage << " Sprime "<<Sprime<<std::endl;

tests/forces/Test_wilson_force.cc

@@ -178,9 +178,9 @@ int main (int argc, char ** argv)
 
   }
 
-  Complex dSpred    = sum(dS);
-  Complex dSm       = sum(dSmom);
-  Complex dSm2      = sum(dSmom2);
+  ComplexD dSpred    = sum(dS);
+  ComplexD dSm       = sum(dSmom);
+  ComplexD dSm2      = sum(dSmom2);
 
 
   std::cout << GridLogMessage <<"Initial mom hamiltonian is "<< Hmom <<std::endl;

tests/forces/Test_zmobius_force.cc

@@ -155,7 +155,7 @@ int main (int argc, char ** argv)
 
   }
 
-  Complex dSpred    = sum(dS);
+  ComplexD dSpred    = sum(dS);
 
   std::cout << GridLogMessage << " S      "<<S<<std::endl;
   std::cout << GridLogMessage << " Sprime "<<Sprime<<std::endl;

tests/hadrons/Test_hadrons_meson_3pt.cc

@@ -61,6 +61,10 @@ int main(int argc, char *argv[])
     
     // gauge field
     application.createModule<MGauge::Unit>("gauge");
+    
+    // set fermion boundary conditions to be periodic space, antiperiodic time.
+    std::string boundary = "1 1 1 -1";
+
     for (unsigned int i = 0; i < flavour.size(); ++i)
     {
         // actions
@@ -69,6 +73,7 @@ int main(int argc, char *argv[])
         actionPar.Ls    = 12;
         actionPar.M5    = 1.8;
         actionPar.mass  = mass[i];
+        actionPar.boundary = boundary;
         application.createModule<MAction::DWF>("DWF_" + flavour[i], actionPar);
         
         // solvers

tests/hadrons/Test_hadrons_rarekaon.cc

@@ -98,6 +98,10 @@ int main(int argc, char *argv[])
         gaugePar.file = configStem;
         application.createModule<MGauge::Load>("gauge", gaugePar);
     }
+    
+    // set fermion boundary conditions to be periodic space, antiperiodic time.
+    std::string boundary = "1 1 1 -1";
+
     for (unsigned int i = 0; i < flavour.size(); ++i)
     {
         // actions
@@ -106,6 +110,7 @@ int main(int argc, char *argv[])
         actionPar.Ls    = 16;
         actionPar.M5    = 1.8;
         actionPar.mass  = mass[i];
+        actionPar.boundary = boundary;
         application.createModule<MAction::DWF>("DWF_" + flavour[i], actionPar);
 
         // solvers

tests/hadrons/Test_hadrons_spectrum.cc

@@ -63,6 +63,10 @@ int main(int argc, char *argv[])
     MSource::Point::Par ptPar;
     ptPar.position = "0 0 0 0";
     application.createModule<MSource::Point>("pt", ptPar);
+    
+    // set fermion boundary conditions to be periodic space, antiperiodic time.
+    std::string boundary = "1 1 1 -1";
+
     for (unsigned int i = 0; i < flavour.size(); ++i)
     {
         // actions
@@ -71,6 +75,7 @@ int main(int argc, char *argv[])
         actionPar.Ls    = 12;
         actionPar.M5    = 1.8;
         actionPar.mass  = mass[i];
+        actionPar.boundary = boundary;
         application.createModule<MAction::DWF>("DWF_" + flavour[i], actionPar);
         
         // solvers

tests/smearing/Test_WilsonFlow.cc

@@ -28,6 +28,38 @@ directory
 /*  END LEGAL */
 #include <Grid/Grid.h>
 
+namespace Grid{
+  struct WFParameters: Serializable {
+    GRID_SERIALIZABLE_CLASS_MEMBERS(WFParameters,
+            int, steps,
+            double, step_size,
+            int, meas_interval,
+            double, maxTau); // for the adaptive algorithm
+       
+
+    template <class ReaderClass >
+    WFParameters(Reader<ReaderClass>& Reader){
+      read(Reader, "WilsonFlow", *this);
+    }
+
+  };
+
+  struct ConfParameters: Serializable {
+    GRID_SERIALIZABLE_CLASS_MEMBERS(ConfParameters,
+           std::string, conf_prefix,
+            std::string, rng_prefix,
+				    int, StartConfiguration,
+				    int, EndConfiguration,
+            int, Skip);
+  
+    template <class ReaderClass >
+    ConfParameters(Reader<ReaderClass>& Reader){
+      read(Reader, "Configurations", *this);
+    }
+
+  };
+}
+
 int main(int argc, char **argv) {
   using namespace Grid;
   using namespace Grid::QCD;
@@ -42,22 +74,38 @@ int main(int argc, char **argv) {
   GridRedBlackCartesian     RBGrid(latt_size, simd_layout, mpi_layout);
 
   std::vector<int> seeds({1, 2, 3, 4, 5});
+  GridSerialRNG sRNG;
   GridParallelRNG pRNG(&Grid);
   pRNG.SeedFixedIntegers(seeds);
 
   LatticeGaugeField Umu(&Grid), Uflow(&Grid);
   SU<Nc>::HotConfiguration(pRNG, Umu);
+  
+  typedef Grid::JSONReader       Serialiser;
+  Serialiser Reader("input.json");
+  WFParameters WFPar(Reader);
+  ConfParameters CPar(Reader);
+  CheckpointerParameters CPPar(CPar.conf_prefix, CPar.rng_prefix);
+  BinaryHmcCheckpointer<PeriodicGimplR> CPBin(CPPar);
+
+  for (int conf = CPar.StartConfiguration; conf <= CPar.EndConfiguration; conf+= CPar.Skip){
+
+  CPBin.CheckpointRestore(conf, Umu, sRNG, pRNG);
 
   std::cout << std::setprecision(15);
-  std::cout << GridLogMessage << "Plaquette: "
+  std::cout << GridLogMessage << "Initial plaquette: "
     << WilsonLoops<PeriodicGimplR>::avgPlaquette(Umu) << std::endl;
 
-  WilsonFlow<PeriodicGimplR> WF(200, 0.01);
+  WilsonFlow<PeriodicGimplR> WF(WFPar.steps, WFPar.step_size, WFPar.meas_interval);
 
-  WF.smear(Uflow, Umu);
+  WF.smear_adaptive(Uflow, Umu, WFPar.maxTau);
 
   RealD WFlow_plaq = WilsonLoops<PeriodicGimplR>::avgPlaquette(Uflow);
-  std::cout << GridLogMessage << "Plaquette: "<< WFlow_plaq << std::endl;
+  RealD WFlow_TC   = WilsonLoops<PeriodicGimplR>::TopologicalCharge(Uflow);
+  RealD WFlow_T0   = WF.energyDensityPlaquette(Uflow);
+  std::cout << GridLogMessage << "Plaquette          "<< conf << "   " << WFlow_plaq << std::endl;
+  std::cout << GridLogMessage << "T0                 "<< conf << "   " << WFlow_T0 << std::endl;
+  std::cout << GridLogMessage << "TopologicalCharge  "<< conf << "   " << WFlow_TC   << std::endl;
 
   std::cout<< GridLogMessage << " Admissibility check:\n";
   const double sp_adm = 0.067;                // admissible threshold
@@ -73,6 +121,32 @@ int main(int argc, char **argv) {
   std::cout<< GridLogMessage << "   (sp_admissible = "<< sp_adm <<")\n";
   //std::cout<< GridLogMessage << "   sp_admissible - sp_max = "<<sp_adm-sp_max <<"\n";
   std::cout<< GridLogMessage << "   sp_admissible - sp_ave = "<<sp_adm-sp_ave <<"\n";
-
+  }
   Grid_finalize();
 }  // main
+
+
+/*
+Input file example
+
+
+JSON
+
+{
+    "WilsonFlow":{
+	"steps": 200,
+	"step_size": 0.01,
+	"meas_interval": 50,
+  "maxTau": 2.0
+    },
+    "Configurations":{
+	"conf_prefix": "ckpoint_lat",
+	"rng_prefix": "ckpoint_rng",
+	"StartConfiguration": 3000,
+	"EndConfiguration": 3000,
+	"Skip": 5
+    }
+}
+
+
+*/

tests/solver/Test_dwf_hdcr.cc

@@ -516,7 +516,7 @@ int main (int argc, char ** argv)
   LatticeColourMatrix U(UGrid);
   LatticeColourMatrix zz(UGrid);
 
-  NerscField header;
+  FieldMetaData header;
   std::string file("./ckpoint_lat.4000");
   NerscIO::readConfiguration(Umu,header,file);
 

tests/solver/Test_dwf_lanczos.cc

@@ -54,7 +54,7 @@ int main (int argc, char ** argv)
   GridParallelRNG          RNG5rb(FrbGrid);  RNG5.SeedFixedIntegers(seeds5);
 
   LatticeGaugeField Umu(UGrid); 
-  SU3::TepidConfiguration(RNG4, Umu);
+  SU3::HotConfiguration(RNG4, Umu);
 
   std::vector<LatticeColourMatrix> U(4,UGrid);
   for(int mu=0;mu<Nd;mu++){
@@ -92,16 +92,15 @@ int main (int argc, char ** argv)
 
   
   std::vector<RealD>          eval(Nm);
-  FermionField    src(FrbGrid); gaussian(RNG5rb,src);
+  FermionField    src(FrbGrid); 
+  gaussian(RNG5rb,src);
   std::vector<FermionField> evec(Nm,FrbGrid);
   for(int i=0;i<1;i++){
-    std::cout << i<<" / "<< Nm<< " grid pointer "<<evec[i]._grid<<std::endl;
+    std::cout << GridLogMessage <<i<<" / "<< Nm<< " grid pointer "<<evec[i]._grid<<std::endl;
   };
 
   int Nconv;
-  IRL.calc(eval,evec,
-	   src,
-	   Nconv);
+  IRL.calc(eval,evec,src,Nconv);
 
 
   Grid_finalize();

tests/solver/Test_staggered_block_cg_unprec.cc

@@ -51,7 +51,7 @@ int main (int argc, char ** argv)
   typedef typename ImprovedStaggeredFermion5DR::ComplexField ComplexField; 
   typename ImprovedStaggeredFermion5DR::ImplParams params; 
 
-  const int Ls=4;
+  const int Ls=8;
 
   Grid_init(&argc,&argv);
 
@@ -74,17 +74,19 @@ int main (int argc, char ** argv)
 
   LatticeGaugeField Umu(UGrid); SU3::HotConfiguration(pRNG,Umu);
 
-  RealD mass=0.01;
+  RealD mass=0.003;
   ImprovedStaggeredFermion5DR Ds(Umu,Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass);
   MdagMLinearOperator<ImprovedStaggeredFermion5DR,FermionField> HermOp(Ds);
 
   ConjugateGradient<FermionField> CG(1.0e-8,10000);
-  BlockConjugateGradient<FermionField> BCG(1.0e-8,10000);
-  MultiRHSConjugateGradient<FermionField> mCG(1.0e-8,10000);
+  int blockDim = 0;
+  BlockConjugateGradient<FermionField>    BCGrQ(BlockCGrQ,blockDim,1.0e-8,10000);
+  BlockConjugateGradient<FermionField>    BCG  (BlockCG,blockDim,1.0e-8,10000);
+  BlockConjugateGradient<FermionField>    mCG  (CGmultiRHS,blockDim,1.0e-8,10000);
 
-  std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
+  std::cout << GridLogMessage << "****************************************************************** "<<std::endl;
   std::cout << GridLogMessage << " Calling 4d CG "<<std::endl;
-  std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
+  std::cout << GridLogMessage << "****************************************************************** "<<std::endl;
   ImprovedStaggeredFermionR Ds4d(Umu,Umu,*UGrid,*UrbGrid,mass);
   MdagMLinearOperator<ImprovedStaggeredFermionR,FermionField> HermOp4d(Ds4d);
   FermionField src4d(UGrid); random(pRNG,src4d);
@@ -111,7 +113,7 @@ int main (int argc, char ** argv)
   std::cout << GridLogMessage << " Calling Block CG for "<<Ls <<" right hand sides" <<std::endl;
   std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
   result=zero;
-  BCG(HermOp,src,result);
+  BCGrQ(HermOp,src,result);
   std::cout << GridLogMessage << "************************************************************************ "<<std::endl;