Updated TODO list

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

.gitignore

@@ -92,6 +92,7 @@ build*/*
 #####################
 *.xcodeproj/*
 build.sh
+.vscode
 
 # Eigen source #
 ################
@@ -106,6 +107,10 @@ lib/fftw/*
 m4/lt*
 m4/libtool.m4
 
+# github pages #
+################
+gh-pages/
+
 # Buck files #
 ##############
 .buck*
@@ -117,3 +122,4 @@ make-bin-BUCK.sh
 #####################
 lib/qcd/spin/gamma-gen/*.h
 lib/qcd/spin/gamma-gen/*.cc
+

.travis.yml

@@ -10,6 +10,8 @@ matrix:
       osx_image: xcode8.3
       compiler: clang
     - compiler: gcc
+      dist: trusty
+      sudo: required
       addons:
         apt:
           sources:
@@ -24,6 +26,8 @@ matrix:
             - binutils-dev
       env: VERSION=-4.9
     - compiler: gcc
+      dist: trusty
+      sudo: required
       addons:
         apt:
           sources:
@@ -38,6 +42,7 @@ matrix:
             - binutils-dev
       env: VERSION=-5
     - compiler: clang
+      dist: trusty
       addons:
         apt:
           sources:
@@ -52,6 +57,7 @@ matrix:
             - binutils-dev
       env: CLANG_LINK=http://llvm.org/releases/3.8.0/clang+llvm-3.8.0-x86_64-linux-gnu-ubuntu-14.04.tar.xz
     - compiler: clang
+      dist: trusty
       addons:
         apt:
           sources:
@@ -78,6 +84,10 @@ install:
     - export CC=$CC$VERSION
     - export CXX=$CXX$VERSION
     - echo $PATH
+    - which autoconf
+    - autoconf  --version
+    - which automake
+    - automake  --version
     - which $CC
     - $CC  --version
     - which $CXX
@@ -95,9 +105,10 @@ script:
     - ../configure --enable-precision=double --enable-simd=SSE4 --enable-comms=none
     - make -j4
     - ./benchmarks/Benchmark_dwf --threads 1 --debug-signals
+    - make check
     - echo make clean
-    - if [[ "$TRAVIS_OS_NAME" == "linux" ]]; then ../configure --enable-precision=single --enable-simd=SSE4 --enable-comms=mpi-auto CXXFLAGS='-DMPI_UINT32_T=MPI_UNSIGNED -DMPI_UINT64_T=MPI_UNSIGNED_LONG'; fi
-    - if [[ "$TRAVIS_OS_NAME" == "linux" ]]; then make -j4; fi
-    - if [[ "$TRAVIS_OS_NAME" == "linux" ]]; then mpirun.openmpi -n 2 ./benchmarks/Benchmark_dwf --threads 1 --mpi 2.1.1.1; fi
+    - if [[ "$TRAVIS_OS_NAME" == "linux" ]] && [[ "$CC" == "clang" ]]; then ../configure --enable-precision=single --enable-simd=SSE4 --enable-comms=mpi-auto ; fi
+    - if [[ "$TRAVIS_OS_NAME" == "linux" ]] && [[ "$CC" == "clang" ]]; then make -j4; fi
+    - if [[ "$TRAVIS_OS_NAME" == "linux" ]] && [[ "$CC" == "clang" ]]; then mpirun.openmpi -n 2 ./benchmarks/Benchmark_dwf --threads 1 --mpi 2.1.1.1; fi
 
 

Makefile.am

@@ -3,10 +3,15 @@ SUBDIRS = lib benchmarks tests extras
 
 include $(top_srcdir)/doxygen.inc
 
-tests: all
-	$(MAKE) -C tests tests
+bin_SCRIPTS=grid-config
 
-.PHONY: tests doxygen-run doxygen-doc $(DX_PS_GOAL) $(DX_PDF_GOAL)
+
+.PHONY: bench check tests doxygen-run doxygen-doc $(DX_PS_GOAL) $(DX_PDF_GOAL)
+
+tests-local: all
+bench-local: all
+check-local: all
 
 AM_CXXFLAGS += -I$(top_builddir)/include
+
 ACLOCAL_AMFLAGS = -I m4

README.md

@@ -22,6 +22,26 @@ Last update Nov 2016.
 
 _Please do not send pull requests to the `master` branch which is reserved for releases._
 
+### Compilers
+
+Intel ICPC v16.0.3 and later
+
+Clang v3.5 and later (need 3.8 and later for OpenMP)
+
+GCC   v4.9.x (recommended)
+
+GCC   v6.3 and later
+
+### Important: 
+
+Some versions of GCC appear to have a bug under high optimisation (-O2, -O3).
+
+The safety of these compiler versions cannot be guaranteed at this time. Follow Issue 100 for details and updates.
+
+GCC   v5.x
+
+GCC   v6.1, v6.2
+
 ### Bug report
 
 _To help us tracking and solving more efficiently issues with Grid, please report problems using the issue system of GitHub rather than sending emails to Grid developers._
@@ -32,7 +52,7 @@ When you file an issue, please go though the following checklist:
 2. Give a description of the target platform (CPU, network, compiler). Please give the full CPU part description, using for example `cat /proc/cpuinfo | grep 'model name' | uniq` (Linux) or `sysctl machdep.cpu.brand_string` (macOS) and the full output the `--version` option of your compiler.
 3. Give the exact `configure` command used.
 4. Attach `config.log`.
-5. Attach `config.summary`.
+5. Attach `grid.config.summary`.
 6. Attach the output of `make V=1`.
 7. Describe the issue and any previous attempt to solve it. If relevant, show how to reproduce the issue using a minimal working example.
 
@@ -95,10 +115,10 @@ install Grid. Other options are detailed in the next section, you can also use `
 `CXX`, `CXXFLAGS`, `LDFLAGS`, ... environment variables can be modified to
 customise the build.
 
-Finally, you can build and install Grid:
+Finally, you can build, check, and install Grid:
 
 ``` bash
-make; make install
+make; make check; make install
 ```
 
 To minimise the build time, only the tests at the root of the `tests` directory are built by default. If you want to build tests in the sub-directory `<subdir>` you can execute:
@@ -121,7 +141,7 @@ If you want to build all the tests at once just use `make tests`.
 - `--enable-gen-simd-width=<size>`: select the size (in bytes) of the generic SIMD vector type (default: 32 bytes).
 - `--enable-precision={single|double}`: set the default precision (default: `double`).
 - `--enable-precision=<comm>`: Use `<comm>` for message passing (default: `none`). A list of possible SIMD targets is detailed in a section below.
-- `--enable-rng={ranlux48|mt19937}`: choose the RNG (default: `ranlux48 `).
+- `--enable-rng={sitmo|ranlux48|mt19937}`: choose the RNG (default: `sitmo `).
 - `--disable-timers`: disable system dependent high-resolution timers.
 - `--enable-chroma`: enable Chroma regression tests.
 - `--enable-doxygen-doc`: enable the Doxygen documentation generation (build with `make doxygen-doc`)
@@ -159,7 +179,6 @@ Alternatively, some CPU codenames can be directly used:
 
 | `<code>`    | Description                            |
 | ----------- | -------------------------------------- |
-| `KNC`       | [Intel Xeon Phi codename Knights Corner](http://ark.intel.com/products/codename/57721/Knights-Corner) |
 | `KNL`       | [Intel Xeon Phi codename Knights Landing](http://ark.intel.com/products/codename/48999/Knights-Landing) |
 | `BGQ`       | Blue Gene/Q                            |
 

TODO

@@ -1,26 +1,31 @@
 TODO:
 ---------------
 
-Peter's work list:
+Large item work list:
+1)- MultiRHS with spread out extra dim -- Go through filesystem with SciDAC I/O
 
--- Remove DenseVector, DenseMatrix; Use Eigen instead. <-- started
--- Merge high precision reduction into develop         <-- done
--- Precision conversion and sort out localConvert      <-- 
--- Physical propagator interface
+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
 
--- multiRHS DWF; benchmark on Cori/BNL for comms elimination
-   -- slice* linalg routines for multiRHS, BlockCG        <-- started
-
--- Profile CG, BlockCG, etc... Flop count/rate
--- Binary I/O speed up & x-strips
--- Half-precision comms                                <-- started
--- GaugeFix into central location
--- FFTfix in sensible place
--- Multigrid Wilson and DWF, compare to other Multigrid implementations
--- quaternions                 -- Might not need
-
-
--- Conserved currents
+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         <-- DONE
+-- BlockCG, BCGrQ                                      <-- DONE
+-- multiRHS DWF; benchmark on Cori/BNL for comms elimination <-- DONE
+   -- slice* linalg routines for multiRHS, BlockCG    
 
 -----
 * Forces; the UdSdU  term in gauge force term is half of what I think it should

VERSION

@@ -1,6 +1,5 @@
-Version : 0.6.0
+Version : 0.7.0
 
-- AVX512, AVX2, AVX, SSE good
-- Clang 3.5 and above, ICPC v16 and above, GCC 4.9 and above
-- MPI and MPI3
-- HiRep, Smearing, Generic gauge group
+- Clang 3.5 and above, ICPC v16 and above, GCC 6.3 and above recommended
+- MPI and MPI3 comms optimisations for KNL and OPA finished
+- Half precision comms

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 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);
@@ -216,27 +267,33 @@ 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);
+	    dbytes+=
 	      Grid.StencilSendToRecvFromBegin(requests,
 					      (void *)&xbuf[mu][0],
 					      xmit_to_rank,
@@ -247,6 +304,7 @@ int main (int argc, char ** argv)
 	    comm_proc = mpi_layout[mu]-1;
 	  
 	    Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank);
+	    dbytes+=
 	      Grid.StencilSendToRecvFromBegin(requests,
 					      (void *)&xbuf[mu+4][0],
 					      xmit_to_rank,
@@ -258,28 +316,34 @@ int main (int argc, char ** argv)
 	}
 	Grid.StencilSendToRecvFromComplete(requests);
 	Grid.Barrier();
-
-      }
 	double stop=usecond();
+	t_time[i] = stop-start; // microseconds
 	
-      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++){
 	
@@ -318,6 +387,7 @@ int main (int argc, char ** argv)
 	    int recv_from_rank;
 	    
 	    Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank);
+	    dbytes+=
 	      Grid.StencilSendToRecvFromBegin(requests,
 					      (void *)&xbuf[mu][0],
 					      xmit_to_rank,
@@ -330,6 +400,7 @@ int main (int argc, char ** argv)
 	    comm_proc = mpi_layout[mu]-1;
 	  
 	    Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank);
+	    dbytes+=
 	      Grid.StencilSendToRecvFromBegin(requests,
 					      (void *)&xbuf[mu+4][0],
 					      xmit_to_rank,
@@ -342,18 +413,26 @@ int main (int argc, char ** argv)
 	  }
 	}
 	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 << 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;
     }
   }    
 

benchmarks/Benchmark_dwf.cc

@@ -1,28 +1,22 @@
  /*************************************************************************************
-
     Grid physics library, www.github.com/paboyle/Grid 
-
     Source file: ./benchmarks/Benchmark_dwf.cc
-
     Copyright (C) 2015
 
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: paboyle <paboyle@ph.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 */
@@ -151,9 +145,7 @@ int main (int argc, char ** argv)
   RealD M5  =1.8;
 
   RealD NP = UGrid->_Nprocessors;
-
-  std::cout << GridLogMessage << "Creating action operator " << std::endl;
-  DomainWallFermionR Dw(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
+  RealD NN = UGrid->NodeCount();
 
   std::cout << GridLogMessage<< "*****************************************************************" <<std::endl;
   std::cout << GridLogMessage<< "* Kernel options --dslash-generic, --dslash-unroll, --dslash-asm" <<std::endl;
@@ -163,16 +155,22 @@ int main (int argc, char ** argv)
   std::cout << GridLogMessage<< "* Vectorising space-time by "<<vComplex::Nsimd()<<std::endl;
   if ( sizeof(Real)==4 )   std::cout << GridLogMessage<< "* SINGLE precision "<<std::endl;
   if ( sizeof(Real)==8 )   std::cout << GridLogMessage<< "* DOUBLE precision "<<std::endl;
+#ifdef GRID_OMP
+  if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsAndCompute ) std::cout << GridLogMessage<< "* Using Overlapped Comms/Compute" <<std::endl;
+  if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsThenCompute) std::cout << GridLogMessage<< "* Using sequential comms compute" <<std::endl;
+#endif
   if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptGeneric   ) std::cout << GridLogMessage<< "* Using GENERIC Nc WilsonKernels" <<std::endl;
   if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptHandUnroll) std::cout << GridLogMessage<< "* Using Nc=3       WilsonKernels" <<std::endl;
   if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptInlineAsm ) std::cout << GridLogMessage<< "* Using Asm Nc=3   WilsonKernels" <<std::endl;
   std::cout << GridLogMessage<< "*****************************************************************" <<std::endl;
 
+  DomainWallFermionR Dw(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
   int ncall =1000;
   if (1) {
     FGrid->Barrier();
     Dw.ZeroCounters();
     Dw.Dhop(src,result,0);
+    std::cout<<GridLogMessage<<"Called warmup"<<std::endl;
     double t0=usecond();
     for(int i=0;i<ncall;i++){
       __SSC_START;
@@ -190,6 +188,7 @@ int main (int argc, char ** argv)
     //    std::cout<<GridLogMessage << "norm ref    "<< norm2(ref)<<std::endl;
     std::cout<<GridLogMessage << "mflop/s =   "<< flops/(t1-t0)<<std::endl;
     std::cout<<GridLogMessage << "mflop/s per rank =  "<< flops/(t1-t0)/NP<<std::endl;
+    std::cout<<GridLogMessage << "mflop/s per node =  "<< flops/(t1-t0)/NN<<std::endl;
     err = ref-result; 
     std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<<std::endl;
 
@@ -206,6 +205,34 @@ int main (int argc, char ** argv)
     Dw.Report();
   }
 
+  DomainWallFermionRL DwH(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
+  if (1) {
+    FGrid->Barrier();
+    DwH.ZeroCounters();
+    DwH.Dhop(src,result,0);
+    double t0=usecond();
+    for(int i=0;i<ncall;i++){
+      __SSC_START;
+      DwH.Dhop(src,result,0);
+      __SSC_STOP;
+    }
+    double t1=usecond();
+    FGrid->Barrier();
+    
+    double volume=Ls;  for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
+    double flops=1344*volume*ncall;
+
+    std::cout<<GridLogMessage << "Called half prec comms Dw "<<ncall<<" times in "<<t1-t0<<" us"<<std::endl;
+    std::cout<<GridLogMessage << "mflop/s =   "<< flops/(t1-t0)<<std::endl;
+    std::cout<<GridLogMessage << "mflop/s per rank =  "<< flops/(t1-t0)/NP<<std::endl;
+    std::cout<<GridLogMessage << "mflop/s per node =  "<< flops/(t1-t0)/NN<<std::endl;
+    err = ref-result; 
+    std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<<std::endl;
+
+    assert (norm2(err)< 1.0e-3 );
+    DwH.Report();
+  }
+
   if (1)
   {
 
@@ -214,6 +241,10 @@ int main (int argc, char ** argv)
     std::cout << GridLogMessage<< "* Vectorising fifth dimension by "<<vComplex::Nsimd()<<std::endl;
     if ( sizeof(Real)==4 )   std::cout << GridLogMessage<< "* SINGLE precision "<<std::endl;
     if ( sizeof(Real)==8 )   std::cout << GridLogMessage<< "* DOUBLE precision "<<std::endl;
+#ifdef GRID_OMP
+  if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsAndCompute ) std::cout << GridLogMessage<< "* Using Overlapped Comms/Compute" <<std::endl;
+  if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsThenCompute) std::cout << GridLogMessage<< "* Using sequential comms compute" <<std::endl;
+#endif
     if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptGeneric   ) std::cout << GridLogMessage<< "* Using GENERIC Nc WilsonKernels" <<std::endl;
     if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptHandUnroll) std::cout << GridLogMessage<< "* Using Nc=3       WilsonKernels" <<std::endl;
     if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptInlineAsm ) std::cout << GridLogMessage<< "* Using Asm Nc=3   WilsonKernels" <<std::endl;
@@ -245,6 +276,7 @@ int main (int argc, char ** argv)
     std::cout<<GridLogMessage << "Called Dw s_inner "<<ncall<<" times in "<<t1-t0<<" us"<<std::endl;
     std::cout<<GridLogMessage << "mflop/s =   "<< flops/(t1-t0)<<std::endl;
     std::cout<<GridLogMessage << "mflop/s per rank =  "<< flops/(t1-t0)/NP<<std::endl;
+    std::cout<<GridLogMessage << "mflop/s per node =  "<< flops/(t1-t0)/NN<<std::endl;
     //    std::cout<<GridLogMessage<< "res norms "<< norm2(result)<<" " <<norm2(sresult)<<std::endl;
     sDw.Report();
     RealD sum=0;
@@ -277,6 +309,10 @@ int main (int argc, char ** argv)
       std::cout << GridLogMessage<< "* Vectorising fifth dimension by "<<vComplex::Nsimd()<<std::endl;
       if ( sizeof(Real)==4 )   std::cout << GridLogMessage<< "* SINGLE precision "<<std::endl;
       if ( sizeof(Real)==8 )   std::cout << GridLogMessage<< "* DOUBLE precision "<<std::endl;
+#ifdef GRID_OMP
+  if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsAndCompute ) std::cout << GridLogMessage<< "* Using Overlapped Comms/Compute" <<std::endl;
+  if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsThenCompute) std::cout << GridLogMessage<< "* Using sequential comms compute" <<std::endl;
+#endif
       if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptGeneric   ) 
 	std::cout << GridLogMessage<< "* Using GENERIC Nc WilsonKernels" <<std::endl;
       if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptHandUnroll) 
@@ -316,6 +352,7 @@ int main (int argc, char ** argv)
 
       std::cout<<GridLogMessage << "sDeo mflop/s =   "<< flops/(t1-t0)<<std::endl;
       std::cout<<GridLogMessage << "sDeo mflop/s per rank   "<< flops/(t1-t0)/NP<<std::endl;
+      std::cout<<GridLogMessage << "sDeo mflop/s per node   "<< flops/(t1-t0)/NN<<std::endl;
       sDw.Report();
 
       sDw.DhopEO(ssrc_o,sr_e,DaggerNo);
@@ -394,14 +431,15 @@ int main (int argc, char ** argv)
 
 
   // S-direction is INNERMOST and takes no part in the parity.
-  static int Opt;  // these are a temporary hack
-  static int Comms;  // these are a temporary hack
-
   std::cout << GridLogMessage<< "*********************************************************" <<std::endl;
   std::cout << GridLogMessage<< "* Benchmarking DomainWallFermionR::DhopEO                "<<std::endl;
   std::cout << GridLogMessage<< "* Vectorising space-time by "<<vComplex::Nsimd()<<std::endl;
   if ( sizeof(Real)==4 )   std::cout << GridLogMessage<< "* SINGLE precision "<<std::endl;
   if ( sizeof(Real)==8 )   std::cout << GridLogMessage<< "* DOUBLE precision "<<std::endl;
+#ifdef GRID_OMP
+  if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsAndCompute ) std::cout << GridLogMessage<< "* Using Overlapped Comms/Compute" <<std::endl;
+  if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsThenCompute) std::cout << GridLogMessage<< "* Using sequential comms compute" <<std::endl;
+#endif
   if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptGeneric   ) std::cout << GridLogMessage<< "* Using GENERIC Nc WilsonKernels" <<std::endl;
   if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptHandUnroll) std::cout << GridLogMessage<< "* Using Nc=3       WilsonKernels" <<std::endl;
   if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptInlineAsm ) std::cout << GridLogMessage<< "* Using Asm Nc=3   WilsonKernels" <<std::endl;
@@ -422,6 +460,7 @@ int main (int argc, char ** argv)
 
     std::cout<<GridLogMessage << "Deo mflop/s =   "<< flops/(t1-t0)<<std::endl;
     std::cout<<GridLogMessage << "Deo mflop/s per rank   "<< flops/(t1-t0)/NP<<std::endl;
+    std::cout<<GridLogMessage << "Deo mflop/s per node   "<< flops/(t1-t0)/NN<<std::endl;
     Dw.Report();
   }
   Dw.DhopEO(src_o,r_e,DaggerNo);
@@ -453,3 +492,4 @@ int main (int argc, char ** argv)
 
   Grid_finalize();
 }
+

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

@@ -35,8 +35,9 @@ using namespace Grid::QCD;
 int main (int argc, char ** argv)
 {
   Grid_init(&argc,&argv);
+#define LMAX (64)
 
-  int Nloop=1000;
+  int Nloop=20;
 
   std::vector<int> simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd());
   std::vector<int> mpi_layout  = GridDefaultMpi();
@@ -50,7 +51,7 @@ int main (int argc, char ** argv)
   std::cout<<GridLogMessage << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
   std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
 
-  for(int lat=2;lat<=32;lat+=2){
+  for(int lat=2;lat<=LMAX;lat+=2){
 
       std::vector<int> latt_size  ({lat*mpi_layout[0],lat*mpi_layout[1],lat*mpi_layout[2],lat*mpi_layout[3]});
       int vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
@@ -82,7 +83,7 @@ int main (int argc, char ** argv)
   std::cout<<GridLogMessage << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
   std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
 
-  for(int lat=2;lat<=32;lat+=2){
+  for(int lat=2;lat<=LMAX;lat+=2){
 
       std::vector<int> latt_size  ({lat*mpi_layout[0],lat*mpi_layout[1],lat*mpi_layout[2],lat*mpi_layout[3]});
       int vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
@@ -113,7 +114,7 @@ int main (int argc, char ** argv)
   std::cout<<GridLogMessage << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
   std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
 
-  for(int lat=2;lat<=32;lat+=2){
+  for(int lat=2;lat<=LMAX;lat+=2){
 
       std::vector<int> latt_size  ({lat*mpi_layout[0],lat*mpi_layout[1],lat*mpi_layout[2],lat*mpi_layout[3]});
       int vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
@@ -144,7 +145,7 @@ int main (int argc, char ** argv)
   std::cout<<GridLogMessage << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
   std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
 
-  for(int lat=2;lat<=32;lat+=2){
+  for(int lat=2;lat<=LMAX;lat+=2){
 
       std::vector<int> latt_size  ({lat*mpi_layout[0],lat*mpi_layout[1],lat*mpi_layout[2],lat*mpi_layout[3]});
       int vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];

benchmarks/Makefile.am

@@ -1,11 +1,7 @@
 include Make.inc
 
-simple: simple_su3_test.o simple_su3_expr.o simple_simd_test.o
-
-EXTRA_LIBRARIES = libsimple_su3_test.a libsimple_su3_expr.a libsimple_simd_test.a
-
-libsimple_su3_test_a_SOURCES = simple_su3_test.cc
-
-libsimple_su3_expr_a_SOURCES = simple_su3_expr.cc
-
-libsimple_simd_test_a_SOURCES = simple_simd_test.cc
+bench-local: all
+	./Benchmark_su3
+	./Benchmark_memory_bandwidth
+	./Benchmark_wilson
+	./Benchmark_dwf --dslash-unroll

bootstrap.sh

@@ -1,6 +1,6 @@
 #!/usr/bin/env bash
 
-EIGEN_URL='http://bitbucket.org/eigen/eigen/get/3.2.9.tar.bz2'
+EIGEN_URL='http://bitbucket.org/eigen/eigen/get/3.3.3.tar.bz2'
 
 echo "-- deploying Eigen source..."
 wget ${EIGEN_URL} --no-check-certificate

configure.ac

@@ -1,16 +1,19 @@
 AC_PREREQ([2.63])
-AC_INIT([Grid], [0.6.0], [https://github.com/paboyle/Grid], [Grid])
+AC_INIT([Grid], [0.7.0], [https://github.com/paboyle/Grid], [Grid])
 AC_CANONICAL_BUILD
 AC_CANONICAL_HOST
 AC_CANONICAL_TARGET
-AM_INIT_AUTOMAKE(subdir-objects)
+AM_INIT_AUTOMAKE([subdir-objects 1.13])
+AM_EXTRA_RECURSIVE_TARGETS([tests bench])
 AC_CONFIG_MACRO_DIR([m4])
 AC_CONFIG_SRCDIR([lib/Grid.h])
 AC_CONFIG_HEADERS([lib/Config.h],[sed -i 's|PACKAGE_|GRID_|' lib/Config.h])
 m4_ifdef([AM_SILENT_RULES], [AM_SILENT_RULES([yes])])
 
+################ Get git info
+#AC_REVISION([m4_esyscmd_s([./scripts/configure.commit])])
+
 ############### Checks for programs
-CXXFLAGS="-O3 $CXXFLAGS"
 AC_PROG_CXX
 AC_PROG_RANLIB
 
@@ -24,6 +27,9 @@ AX_GXX_VERSION
 AC_DEFINE_UNQUOTED([GXX_VERSION],["$GXX_VERSION"],
       [version of g++ that will compile the code])
 
+CXXFLAGS="-g $CXXFLAGS"
+
+
 ############### Checks for typedefs, structures, and compiler characteristics
 AC_TYPE_SIZE_T
 AC_TYPE_UINT32_T
@@ -67,6 +73,13 @@ AC_ARG_WITH([fftw],
             [AM_CXXFLAGS="-I$with_fftw/include $AM_CXXFLAGS"]
             [AM_LDFLAGS="-L$with_fftw/lib $AM_LDFLAGS"])
 
+############### LIME
+AC_ARG_WITH([lime],
+            [AS_HELP_STRING([--with-lime=prefix],
+            [try this for a non-standard install prefix of the LIME library])],
+            [AM_CXXFLAGS="-I$with_lime/include $AM_CXXFLAGS"]
+            [AM_LDFLAGS="-L$with_lime/lib $AM_LDFLAGS"])
+
 ############### lapack
 AC_ARG_ENABLE([lapack],
     [AC_HELP_STRING([--enable-lapack=yes|no|prefix], [enable LAPACK])],
@@ -84,16 +97,15 @@ case ${ac_LAPACK} in
 esac
 
 ############### FP16 conversions
-AC_ARG_ENABLE([fp16],
-    [AC_HELP_STRING([--enable-fp16=yes|no], [enable fp16 comms])], 
-    [ac_FP16=${enable_fp16}], [ac_FP16=no])
-case ${ac_FP16} in
-    no)
-        ;;
+AC_ARG_ENABLE([sfw-fp16],
+    [AC_HELP_STRING([--enable-sfw-fp16=yes|no], [enable software fp16 comms])],
+    [ac_SFW_FP16=${enable_sfw_fp16}], [ac_SFW_FP16=yes])
+case ${ac_SFW_FP16} in
     yes)
-        AC_DEFINE([USE_FP16],[1],[conversion to fp16]);;
+      AC_DEFINE([SFW_FP16],[1],[software conversion to fp16]);;
+    no);;
     *)
-	;;
+      AC_MSG_ERROR(["SFW FP16 option not supported ${ac_SFW_FP16}"]);;
 esac
 
 ############### MKL
@@ -165,6 +177,18 @@ AC_SEARCH_LIBS([fftw_execute], [fftw3],
                [AC_DEFINE([HAVE_FFTW], [1], [Define to 1 if you have the `FFTW' library])]
                [have_fftw=true])
 
+AC_SEARCH_LIBS([limeCreateReader], [lime],
+               [AC_DEFINE([HAVE_LIME], [1], [Define to 1 if you have the `LIME' library])]
+               [have_lime=true],
+	       [AC_MSG_WARN(C-LIME library was not found in your system.
+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])]
                [have_hdf5=true]
@@ -189,7 +213,14 @@ case ${ax_cv_cxx_compiler_vendor} in
     case ${ac_SIMD} in
       SSE4)
         AC_DEFINE([SSE4],[1],[SSE4 intrinsics])
+	case ${ac_SFW_FP16} in
+	  yes)
 	  SIMD_FLAGS='-msse4.2';;
+	  no)
+	  SIMD_FLAGS='-msse4.2 -mf16c';;
+	  *)
+          AC_MSG_ERROR(["SFW_FP16 must be either yes or no value ${ac_SFW_FP16} "]);;
+	esac;;
       AVX)
         AC_DEFINE([AVX1],[1],[AVX intrinsics])
         SIMD_FLAGS='-mavx -mf16c';;
@@ -397,33 +428,31 @@ DX_INIT_DOXYGEN([$PACKAGE_NAME], [doxygen.cfg])
 
 ############### Ouput
 cwd=`pwd -P`; cd ${srcdir}; abs_srcdir=`pwd -P`; cd ${cwd}
+GRID_CXXFLAGS="$AM_CXXFLAGS $CXXFLAGS"
+GRID_LDFLAGS="$AM_LDFLAGS $LDFLAGS"
+GRID_LIBS=$LIBS
+GRID_SHORT_SHA=`git rev-parse --short HEAD`
+GRID_SHA=`git rev-parse HEAD`
+GRID_BRANCH=`git rev-parse --abbrev-ref HEAD`
 AM_CXXFLAGS="-I${abs_srcdir}/include $AM_CXXFLAGS"
 AM_CFLAGS="-I${abs_srcdir}/include $AM_CFLAGS"
 AM_LDFLAGS="-L${cwd}/lib $AM_LDFLAGS"
 AC_SUBST([AM_CFLAGS])
 AC_SUBST([AM_CXXFLAGS])
 AC_SUBST([AM_LDFLAGS])
-AC_CONFIG_FILES(Makefile)
-AC_CONFIG_FILES(lib/Makefile)
-AC_CONFIG_FILES(tests/Makefile)
-AC_CONFIG_FILES(tests/IO/Makefile)
-AC_CONFIG_FILES(tests/core/Makefile)
-AC_CONFIG_FILES(tests/debug/Makefile)
-AC_CONFIG_FILES(tests/forces/Makefile)
-AC_CONFIG_FILES(tests/hadrons/Makefile)
-AC_CONFIG_FILES(tests/hmc/Makefile)
-AC_CONFIG_FILES(tests/solver/Makefile)
-AC_CONFIG_FILES(tests/qdpxx/Makefile)
-AC_CONFIG_FILES(tests/testu01/Makefile)
-AC_CONFIG_FILES(benchmarks/Makefile)
-AC_CONFIG_FILES(extras/Makefile)
-AC_CONFIG_FILES(extras/Hadrons/Makefile)
-AC_OUTPUT
+AC_SUBST([GRID_CXXFLAGS])
+AC_SUBST([GRID_LDFLAGS])
+AC_SUBST([GRID_LIBS])
+AC_SUBST([GRID_SHA])
+AC_SUBST([GRID_BRANCH])
+
+git_commit=`cd $srcdir && ./scripts/configure.commit`
 
 echo "~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 Summary of configuration for $PACKAGE v$VERSION
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
+----- GIT VERSION -------------------------------------
+$git_commit
 ----- PLATFORM ----------------------------------------
 architecture (build)        : $build_cpu
 os (build)                  : $build_os
@@ -436,10 +465,12 @@ SIMD                        : ${ac_SIMD}${SIMD_GEN_WIDTH_MSG}
 Threading                   : ${ac_openmp}
 Communications type         : ${comms_type}
 Default precision           : ${ac_PRECISION}
+Software FP16 conversion    : ${ac_SFW_FP16}
 RNG choice                  : ${ac_RNG}
 GMP                         : `if test "x$have_gmp" = xtrue; then echo yes; else echo no; fi`
 LAPACK                      : ${ac_LAPACK}
 FFTW                        : `if test "x$have_fftw" = xtrue; then echo yes; else echo no; fi`
+LIME (ILDG support)         : `if test "x$have_lime" = xtrue; then echo yes; else echo no; fi`
 HDF5                        : `if test "x$have_hdf5" = xtrue; then echo yes; else echo no; fi`
 build DOXYGEN documentation : `if test "$DX_FLAG_doc" = '1'; then echo yes; else echo no; fi`
 ----- BUILD FLAGS -------------------------------------
@@ -449,7 +480,32 @@ LDFLAGS:
 `echo ${AM_LDFLAGS} ${LDFLAGS} | tr ' ' '\n' | sed 's/^-/    -/g'`
 LIBS:
 `echo ${LIBS} | tr ' ' '\n' | sed 's/^-/    -/g'`
--------------------------------------------------------" > config.summary
+-------------------------------------------------------" > grid.configure.summary
+
+GRID_SUMMARY="`cat grid.configure.summary`"
+AM_SUBST_NOTMAKE([GRID_SUMMARY])
+AC_SUBST([GRID_SUMMARY])
+
+AC_CONFIG_FILES([grid-config], [chmod +x grid-config])
+AC_CONFIG_FILES(Makefile)
+AC_CONFIG_FILES(lib/Makefile)
+AC_CONFIG_FILES(tests/Makefile)
+AC_CONFIG_FILES(tests/IO/Makefile)
+AC_CONFIG_FILES(tests/core/Makefile)
+AC_CONFIG_FILES(tests/debug/Makefile)
+AC_CONFIG_FILES(tests/forces/Makefile)
+AC_CONFIG_FILES(tests/hadrons/Makefile)
+AC_CONFIG_FILES(tests/hmc/Makefile)
+AC_CONFIG_FILES(tests/solver/Makefile)
+AC_CONFIG_FILES(tests/smearing/Makefile)
+AC_CONFIG_FILES(tests/qdpxx/Makefile)
+AC_CONFIG_FILES(tests/testu01/Makefile)
+AC_CONFIG_FILES(benchmarks/Makefile)
+AC_CONFIG_FILES(extras/Makefile)
+AC_CONFIG_FILES(extras/Hadrons/Makefile)
+AC_OUTPUT
+
 echo ""
-cat config.summary
+cat grid.configure.summary
 echo ""
+

extras/Hadrons/Application.cc

@@ -162,7 +162,8 @@ void Application::saveParameterFile(const std::string parameterFileName)
 sizeString((size)*locVol_) << " (" << sizeString(size)  << "/site)"
 
 #define DEFINE_MEMPEAK \
-auto memPeak = [this](const std::vector<unsigned int> &program)\
+GeneticScheduler<unsigned int>::ObjFunc memPeak = \
+[this](const std::vector<unsigned int> &program)\
 {\
     unsigned int memPeak;\
     bool         msg;\

extras/Hadrons/Global.hpp

@@ -145,6 +145,15 @@ std::string typeName(void)
     return typeName(typeIdPt<T>());
 }
 
+// default writers/readers
+#ifdef HAVE_HDF5
+typedef Hdf5Reader CorrReader;
+typedef Hdf5Writer CorrWriter;
+#else
+typedef XmlReader CorrReader;
+typedef XmlWriter CorrWriter;
+#endif
+
 END_HADRONS_NAMESPACE
 
 #endif // Hadrons_Global_hpp_

extras/Hadrons/Modules.hpp

@@ -29,12 +29,20 @@ See the full license in the file "LICENSE" in the top level distribution directo
 #include <Grid/Hadrons/Modules/MAction/DWF.hpp>
 #include <Grid/Hadrons/Modules/MAction/Wilson.hpp>
 #include <Grid/Hadrons/Modules/MContraction/Baryon.hpp>
+#include <Grid/Hadrons/Modules/MContraction/DiscLoop.hpp>
+#include <Grid/Hadrons/Modules/MContraction/Gamma3pt.hpp>
 #include <Grid/Hadrons/Modules/MContraction/Meson.hpp>
+#include <Grid/Hadrons/Modules/MContraction/WeakHamiltonian.hpp>
+#include <Grid/Hadrons/Modules/MContraction/WeakHamiltonianEye.hpp>
+#include <Grid/Hadrons/Modules/MContraction/WeakHamiltonianNonEye.hpp>
+#include <Grid/Hadrons/Modules/MContraction/WeakNeutral4ptDisc.hpp>
 #include <Grid/Hadrons/Modules/MGauge/Load.hpp>
 #include <Grid/Hadrons/Modules/MGauge/Random.hpp>
 #include <Grid/Hadrons/Modules/MGauge/Unit.hpp>
+#include <Grid/Hadrons/Modules/MLoop/NoiseLoop.hpp>
 #include <Grid/Hadrons/Modules/MSolver/RBPrecCG.hpp>
 #include <Grid/Hadrons/Modules/MSource/Point.hpp>
 #include <Grid/Hadrons/Modules/MSource/SeqGamma.hpp>
+#include <Grid/Hadrons/Modules/MSource/Wall.hpp>
 #include <Grid/Hadrons/Modules/MSource/Z2.hpp>
 #include <Grid/Hadrons/Modules/Quark.hpp>

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/Baryon.hpp

@@ -112,7 +112,7 @@ void TBaryon<FImpl1, FImpl2, FImpl3>::execute(void)
                  << " quarks '" << par().q1 << "', '" << par().q2 << "', and '"
                  << par().q3 << "'" << std::endl;
     
-    XmlWriter             writer(par().output);
+    CorrWriter             writer(par().output);
     PropagatorField1      &q1 = *env().template getObject<PropagatorField1>(par().q1);
     PropagatorField2      &q2 = *env().template getObject<PropagatorField2>(par().q2);
     PropagatorField3      &q3 = *env().template getObject<PropagatorField3>(par().q2);

extras/Hadrons/Modules/MContraction/DiscLoop.hpp

@@ -0,0 +1,144 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Modules/MContraction/DiscLoop.hpp
+
+Copyright (C) 2017
+
+Author: Andrew Lawson    <andrew.lawson1991@gmail.com>
+
+This program is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+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 Hadrons_DiscLoop_hpp_
+#define Hadrons_DiscLoop_hpp_
+
+#include <Grid/Hadrons/Global.hpp>
+#include <Grid/Hadrons/Module.hpp>
+#include <Grid/Hadrons/ModuleFactory.hpp>
+
+BEGIN_HADRONS_NAMESPACE
+
+/******************************************************************************
+ *                                DiscLoop                                    *
+ ******************************************************************************/
+BEGIN_MODULE_NAMESPACE(MContraction)
+
+class DiscLoopPar: Serializable
+{
+public:
+    GRID_SERIALIZABLE_CLASS_MEMBERS(DiscLoopPar,
+                                    std::string,    q_loop,
+                                    Gamma::Algebra, gamma,
+                                    std::string,    output);
+};
+
+template <typename FImpl>
+class TDiscLoop: public Module<DiscLoopPar>
+{
+    TYPE_ALIASES(FImpl,);
+    class Result: Serializable
+    {
+    public:
+        GRID_SERIALIZABLE_CLASS_MEMBERS(Result,
+                                        Gamma::Algebra, gamma,
+                                        std::vector<Complex>, corr);
+    };
+public:
+    // constructor
+    TDiscLoop(const std::string name);
+    // destructor
+    virtual ~TDiscLoop(void) = default;
+    // dependency relation
+    virtual std::vector<std::string> getInput(void);
+    virtual std::vector<std::string> getOutput(void);
+    // setup
+    virtual void setup(void);
+    // execution
+    virtual void execute(void);
+};
+
+MODULE_REGISTER_NS(DiscLoop, TDiscLoop<FIMPL>, MContraction);
+
+/******************************************************************************
+ *                       TDiscLoop implementation                             *
+ ******************************************************************************/
+// constructor /////////////////////////////////////////////////////////////////
+template <typename FImpl>
+TDiscLoop<FImpl>::TDiscLoop(const std::string name)
+: Module<DiscLoopPar>(name)
+{}
+
+// dependencies/products ///////////////////////////////////////////////////////
+template <typename FImpl>
+std::vector<std::string> TDiscLoop<FImpl>::getInput(void)
+{
+    std::vector<std::string> in = {par().q_loop};
+    
+    return in;
+}
+
+template <typename FImpl>
+std::vector<std::string> TDiscLoop<FImpl>::getOutput(void)
+{
+    std::vector<std::string> out = {getName()};
+    
+    return out;
+}
+
+// setup ///////////////////////////////////////////////////////////////////////
+template <typename FImpl>
+void TDiscLoop<FImpl>::setup(void)
+{
+    
+}
+
+// execution ///////////////////////////////////////////////////////////////////
+template <typename FImpl>
+void TDiscLoop<FImpl>::execute(void)
+{
+    LOG(Message) << "Computing disconnected loop contraction '" << getName() 
+                 << "' using '" << par().q_loop << "' with " << par().gamma 
+                 << " insertion." << std::endl;
+
+    CorrWriter            writer(par().output);
+    PropagatorField       &q_loop = *env().template getObject<PropagatorField>(par().q_loop);
+    LatticeComplex        c(env().getGrid());
+    Gamma                 gamma(par().gamma);
+    std::vector<TComplex> buf;
+    Result                result;
+
+    c = trace(gamma*q_loop);
+    sliceSum(c, buf, Tp);
+
+    result.gamma = par().gamma;
+    result.corr.resize(buf.size());
+    for (unsigned int t = 0; t < buf.size(); ++t)
+    {
+        result.corr[t] = TensorRemove(buf[t]);
+    }
+
+    write(writer, "disc", result);
+}
+
+END_MODULE_NAMESPACE
+
+END_HADRONS_NAMESPACE
+
+#endif // Hadrons_DiscLoop_hpp_

extras/Hadrons/Modules/MContraction/Gamma3pt.hpp

@@ -0,0 +1,170 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Modules/MContraction/Gamma3pt.hpp
+
+Copyright (C) 2017
+
+Author: Andrew Lawson    <andrew.lawson1991@gmail.com>
+
+This program is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+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 Hadrons_Gamma3pt_hpp_
+#define Hadrons_Gamma3pt_hpp_
+
+#include <Grid/Hadrons/Global.hpp>
+#include <Grid/Hadrons/Module.hpp>
+#include <Grid/Hadrons/ModuleFactory.hpp>
+
+BEGIN_HADRONS_NAMESPACE
+
+/*
+ * 3pt contraction with gamma matrix insertion.
+ *
+ * Schematic:
+ *
+ *             q2           q3
+ *        /----<------*------<----¬
+ *       /          gamma          \
+ *      /                           \
+ *   i *                            * f
+ *      \                          /
+ *       \                        /
+ *        \----------->----------/
+ *                   q1
+ *
+ *      trace(g5*q1*adj(q2)*g5*gamma*q3)
+ */
+
+/******************************************************************************
+ *                               Gamma3pt                                     *
+ ******************************************************************************/
+BEGIN_MODULE_NAMESPACE(MContraction)
+
+class Gamma3ptPar: Serializable
+{
+public:
+    GRID_SERIALIZABLE_CLASS_MEMBERS(Gamma3ptPar,
+                                    std::string,    q1,
+                                    std::string,    q2,
+                                    std::string,    q3,
+                                    Gamma::Algebra, gamma,
+                                    std::string,    output);
+};
+
+template <typename FImpl1, typename FImpl2, typename FImpl3>
+class TGamma3pt: public Module<Gamma3ptPar>
+{
+    TYPE_ALIASES(FImpl1, 1);
+    TYPE_ALIASES(FImpl2, 2);
+    TYPE_ALIASES(FImpl3, 3);
+    class Result: Serializable
+    {
+    public:
+        GRID_SERIALIZABLE_CLASS_MEMBERS(Result,
+                                        Gamma::Algebra, gamma,
+                                        std::vector<Complex>, corr);
+    };
+public:
+    // constructor
+    TGamma3pt(const std::string name);
+    // destructor
+    virtual ~TGamma3pt(void) = default;
+    // dependency relation
+    virtual std::vector<std::string> getInput(void);
+    virtual std::vector<std::string> getOutput(void);
+    // setup
+    virtual void setup(void);
+    // execution
+    virtual void execute(void);
+};
+
+MODULE_REGISTER_NS(Gamma3pt, ARG(TGamma3pt<FIMPL, FIMPL, FIMPL>), MContraction);
+
+/******************************************************************************
+ *                       TGamma3pt implementation                             *
+ ******************************************************************************/
+// constructor /////////////////////////////////////////////////////////////////
+template <typename FImpl1, typename FImpl2, typename FImpl3>
+TGamma3pt<FImpl1, FImpl2, FImpl3>::TGamma3pt(const std::string name)
+: Module<Gamma3ptPar>(name)
+{}
+
+// dependencies/products ///////////////////////////////////////////////////////
+template <typename FImpl1, typename FImpl2, typename FImpl3>
+std::vector<std::string> TGamma3pt<FImpl1, FImpl2, FImpl3>::getInput(void)
+{
+    std::vector<std::string> in = {par().q1, par().q2, par().q3};
+    
+    return in;
+}
+
+template <typename FImpl1, typename FImpl2, typename FImpl3>
+std::vector<std::string> TGamma3pt<FImpl1, FImpl2, FImpl3>::getOutput(void)
+{
+    std::vector<std::string> out = {getName()};
+    
+    return out;
+}
+
+// setup ///////////////////////////////////////////////////////////////////////
+template <typename FImpl1, typename FImpl2, typename FImpl3>
+void TGamma3pt<FImpl1, FImpl2, FImpl3>::setup(void)
+{
+    
+}
+
+// execution ///////////////////////////////////////////////////////////////////
+template <typename FImpl1, typename FImpl2, typename FImpl3>
+void TGamma3pt<FImpl1, FImpl2, FImpl3>::execute(void)
+{
+    LOG(Message) << "Computing 3pt contractions '" << getName() << "' using"
+                 << " quarks '" << par().q1 << "', '" << par().q2 << "' and '"
+                 << par().q3 << "', with " << par().gamma << " insertion." 
+                 << std::endl;
+
+    CorrWriter            writer(par().output);
+    PropagatorField1      &q1 = *env().template getObject<PropagatorField1>(par().q1);
+    PropagatorField2      &q2 = *env().template getObject<PropagatorField2>(par().q2);
+    PropagatorField3      &q3 = *env().template getObject<PropagatorField3>(par().q3);
+    LatticeComplex        c(env().getGrid());
+    Gamma                 g5(Gamma::Algebra::Gamma5);
+    Gamma                 gamma(par().gamma);
+    std::vector<TComplex> buf;
+    Result                result;
+
+    c = trace(g5*q1*adj(q2)*(g5*gamma)*q3);
+    sliceSum(c, buf, Tp);
+
+    result.gamma = par().gamma;
+    result.corr.resize(buf.size());
+    for (unsigned int t = 0; t < buf.size(); ++t)
+    {
+        result.corr[t] = TensorRemove(buf[t]);
+    }
+
+    write(writer, "gamma3pt", result);
+}
+
+END_MODULE_NAMESPACE
+
+END_HADRONS_NAMESPACE
+
+#endif // Hadrons_Gamma3pt_hpp_

extras/Hadrons/Modules/MContraction/Meson.hpp

@@ -6,8 +6,10 @@ Source file: extras/Hadrons/Modules/MContraction/Meson.hpp
 
 Copyright (C) 2015
 Copyright (C) 2016
+Copyright (C) 2017
 
 Author: Antonin Portelli <antonin.portelli@me.com>
+        Andrew Lawson    <andrew.lawson1991@gmail.com>
 
 This program is free software; you can redistribute it and/or modify
 it under the terms of the GNU General Public License as published by
@@ -36,20 +38,39 @@ See the full license in the file "LICENSE" in the top level distribution directo
 
 BEGIN_HADRONS_NAMESPACE
 
+/*
+ 
+ Meson contractions
+ -----------------------------
+ 
+ * options:
+ - q1: input propagator 1 (string)
+ - q2: input propagator 2 (string)
+ - gammas: gamma products to insert at sink & source, pairs of gamma matrices 
+           (space-separated strings) in angled brackets (i.e. <g_sink g_src>),
+           in a sequence (e.g. "<Gamma5 Gamma5><Gamma5 GammaT>").
+
+           Special values: "all" - perform all possible contractions.
+ - mom: momentum insertion, space-separated float sequence (e.g ".1 .2 1. 0."),
+        given as multiples of (2*pi) / L.
+*/
+
 /******************************************************************************
  *                                TMeson                                       *
  ******************************************************************************/
 BEGIN_MODULE_NAMESPACE(MContraction)
 
+typedef std::pair<Gamma::Algebra, Gamma::Algebra> GammaPair;
+
 class MesonPar: Serializable
 {
 public:
     GRID_SERIALIZABLE_CLASS_MEMBERS(MesonPar,
                                     std::string, q1,
                                     std::string, q2,
-                                    std::string,    output,
-                                    Gamma::Algebra, gammaSource,
-                                    Gamma::Algebra, gammaSink);
+                                    std::string, gammas,
+                                    std::string, mom,
+                                    std::string, output);
 };
 
 template <typename FImpl1, typename FImpl2>
@@ -61,7 +82,10 @@ public:
     class Result: Serializable
     {
     public:
-        GRID_SERIALIZABLE_CLASS_MEMBERS(Result, std::vector<Complex>, corr);
+        GRID_SERIALIZABLE_CLASS_MEMBERS(Result,
+                                        Gamma::Algebra, gamma_snk,
+                                        Gamma::Algebra, gamma_src,
+                                        std::vector<Complex>, corr);
     };
 public:
     // constructor
@@ -71,6 +95,7 @@ public:
     // dependencies/products
     virtual std::vector<std::string> getInput(void);
     virtual std::vector<std::string> getOutput(void);
+    virtual void parseGammaString(std::vector<GammaPair> &gammaList);
     // execution
     virtual void execute(void);
 };
@@ -103,6 +128,31 @@ std::vector<std::string> TMeson<FImpl1, FImpl2>::getOutput(void)
     return output;
 }
 
+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.
+        for (unsigned int i = 1; i < Gamma::nGamma; i += 2)
+        {
+            for (unsigned int j = 1; j < Gamma::nGamma; j += 2)
+            {
+                gammaList.push_back(std::make_pair((Gamma::Algebra)i, 
+                                                   (Gamma::Algebra)j));
+            }
+        }
+    }
+    else
+    {
+        // Parse individual contractions from input string.
+        gammaList = strToVec<GammaPair>(par().gammas);
+    }
+}
+
+
 // execution ///////////////////////////////////////////////////////////////////
 template <typename FImpl1, typename FImpl2>
 void TMeson<FImpl1, FImpl2>::execute(void)
@@ -111,21 +161,44 @@ void TMeson<FImpl1, FImpl2>::execute(void)
                  << " quarks '" << par().q1 << "' and '" << par().q2 << "'"
                  << std::endl;
     
-    XmlWriter             writer(par().output);
+    CorrWriter              writer(par().output);
     PropagatorField1       &q1 = *env().template getObject<PropagatorField1>(par().q1);
     PropagatorField2       &q2 = *env().template getObject<PropagatorField2>(par().q2);
     LatticeComplex         c(env().getGrid());
-    Gamma                 gSrc(par().gammaSource), gSnk(par().gammaSink);
     Gamma                  g5(Gamma::Algebra::Gamma5);
+    std::vector<GammaPair> gammaList;
     std::vector<TComplex>  buf;
-    Result                result;
+    std::vector<Result>    result;
+    std::vector<Real>      p;
 
-    c = trace(gSnk*q1*adj(gSrc)*g5*adj(q2)*g5);
+    p  = strToVec<Real>(par().mom);
+    LatticeComplex         ph(env().getGrid()), coor(env().getGrid());
+    Complex                i(0.0,1.0);
+    ph = zero;
+    for(unsigned int mu = 0; mu < env().getNd(); mu++)
+    {
+        LatticeCoordinate(coor, mu);
+        ph = ph + p[mu]*coor*((1./(env().getGrid()->_fdimensions[mu])));
+    }
+    ph = exp((Real)(2*M_PI)*i*ph);
+    
+    parseGammaString(gammaList);
+
+    result.resize(gammaList.size());
+    for (unsigned int i = 0; i < result.size(); ++i)
+    {
+        Gamma gSnk(gammaList[i].first);
+        Gamma gSrc(gammaList[i].second);
+        c = trace((g5*gSnk)*q1*(adj(gSrc)*g5)*adj(q2))*ph;
         sliceSum(c, buf, Tp);
-    result.corr.resize(buf.size());
+
+        result[i].gamma_snk = gammaList[i].first;
+        result[i].gamma_src = gammaList[i].second;
+        result[i].corr.resize(buf.size());
         for (unsigned int t = 0; t < buf.size(); ++t)
         {
-        result.corr[t] = TensorRemove(buf[t]);
+            result[i].corr[t] = TensorRemove(buf[t]);
+        }
     }
     write(writer, "meson", result);
 }

extras/Hadrons/Modules/MContraction/WeakHamiltonian.hpp

@@ -0,0 +1,114 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Modules/MContraction/WeakHamiltonian.hpp
+
+Copyright (C) 2017
+
+Author: Andrew Lawson    <andrew.lawson1991@gmail.com>
+
+This program is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+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 Hadrons_WeakHamiltonian_hpp_
+#define Hadrons_WeakHamiltonian_hpp_
+
+#include <Grid/Hadrons/Global.hpp>
+#include <Grid/Hadrons/Module.hpp>
+#include <Grid/Hadrons/ModuleFactory.hpp>
+
+BEGIN_HADRONS_NAMESPACE
+
+/******************************************************************************
+ *                         WeakHamiltonian                                    *
+ ******************************************************************************/
+BEGIN_MODULE_NAMESPACE(MContraction)
+
+/*******************************************************************************
+ * Utilities for contractions involving the Weak Hamiltonian.
+ ******************************************************************************/
+//// Sum and store correlator.
+#define MAKE_DIAG(exp, buf, res, n)\
+sliceSum(exp, buf, Tp);\
+res.name = (n);\
+res.corr.resize(buf.size());\
+for (unsigned int t = 0; t < buf.size(); ++t)\
+{\
+    res.corr[t] = TensorRemove(buf[t]);\
+}
+
+//// Contraction of mu index: use 'mu' variable in exp.
+#define SUM_MU(buf,exp)\
+buf = zero;\
+for (unsigned int mu = 0; mu < ndim; ++mu)\
+{\
+    buf += exp;\
+}
+
+enum 
+{
+  i_V = 0,
+  i_A = 1,
+  n_i = 2
+};
+
+class WeakHamiltonianPar: Serializable
+{
+public:
+    GRID_SERIALIZABLE_CLASS_MEMBERS(WeakHamiltonianPar,
+                                    std::string, q1,
+                                    std::string, q2,
+                                    std::string, q3,
+                                    std::string, q4,
+                                    std::string, output);
+};
+
+#define MAKE_WEAK_MODULE(modname)\
+class T##modname: public Module<WeakHamiltonianPar>\
+{\
+public:\
+    TYPE_ALIASES(FIMPL,)\
+    class Result: Serializable\
+    {\
+    public:\
+        GRID_SERIALIZABLE_CLASS_MEMBERS(Result,\
+                                        std::string, name,\
+                                        std::vector<Complex>, corr);\
+    };\
+public:\
+    /* constructor */ \
+    T##modname(const std::string name);\
+    /* destructor */ \
+    virtual ~T##modname(void) = default;\
+    /* dependency relation */ \
+    virtual std::vector<std::string> getInput(void);\
+    virtual std::vector<std::string> getOutput(void);\
+    /* setup */ \
+    virtual void setup(void);\
+    /* execution */ \
+    virtual void execute(void);\
+    std::vector<std::string> VA_label = {"V", "A"};\
+};\
+MODULE_REGISTER_NS(modname, T##modname, MContraction);
+
+END_MODULE_NAMESPACE
+
+END_HADRONS_NAMESPACE
+
+#endif // Hadrons_WeakHamiltonian_hpp_

extras/Hadrons/Modules/MContraction/WeakHamiltonianEye.cc

@@ -0,0 +1,137 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Modules/MContraction/WeakHamiltonianEye.cc
+
+Copyright (C) 2017
+
+Author: Andrew Lawson    <andrew.lawson1991@gmail.com>
+
+This program is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+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/Hadrons/Modules/MContraction/WeakHamiltonianEye.hpp>
+
+using namespace Grid;
+using namespace Hadrons;
+using namespace MContraction;
+
+/*
+ * Weak Hamiltonian current-current contractions, Eye-type.
+ * 
+ * These contractions are generated by the Q1 and Q2 operators in the physical
+ * basis (see e.g. Fig 3 of arXiv:1507.03094).
+ * 
+ * Schematics:        q4                 |                  
+ *                  /-<-¬                |                             
+ *                 /     \               |             q2           q3
+ *                 \     /               |        /----<------*------<----¬                        
+ *            q2    \   /    q3          |       /          /-*-¬          \
+ *       /-----<-----* *-----<----¬      |      /          /     \          \
+ *    i *            H_W           * f   |   i *           \     /  q4      * f
+ *       \                        /      |      \           \->-/          /   
+ *        \                      /       |       \                        /       
+ *         \---------->---------/        |        \----------->----------/        
+ *                   q1                  |                   q1                  
+ *                                       |
+ *                Saucer (S)             |                  Eye (E)
+ * 
+ * S: trace(q3*g5*q1*adj(q2)*g5*gL[mu][p_1]*q4*gL[mu][p_2])
+ * E: trace(q3*g5*q1*adj(q2)*g5*gL[mu][p_1])*trace(q4*gL[mu][p_2])
+ */
+
+/******************************************************************************
+ *                  TWeakHamiltonianEye implementation                        *
+ ******************************************************************************/
+// constructor /////////////////////////////////////////////////////////////////
+TWeakHamiltonianEye::TWeakHamiltonianEye(const std::string name)
+: Module<WeakHamiltonianPar>(name)
+{}
+
+// dependencies/products ///////////////////////////////////////////////////////
+std::vector<std::string> TWeakHamiltonianEye::getInput(void)
+{
+    std::vector<std::string> in = {par().q1, par().q2, par().q3, par().q4};
+    
+    return in;
+}
+
+std::vector<std::string> TWeakHamiltonianEye::getOutput(void)
+{
+    std::vector<std::string> out = {getName()};
+    
+    return out;
+}
+
+// setup ///////////////////////////////////////////////////////////////////////
+void TWeakHamiltonianEye::setup(void)
+{
+
+}
+
+// execution ///////////////////////////////////////////////////////////////////
+void TWeakHamiltonianEye::execute(void)
+{
+    LOG(Message) << "Computing Weak Hamiltonian (Eye type) contractions '" 
+                 << getName() << "' using quarks '" << par().q1 << "', '" 
+                 << par().q2 << ", '" << par().q3 << "' and '" << par().q4 
+                 << "'." << std::endl;
+
+    CorrWriter             writer(par().output);
+    PropagatorField &q1 = *env().template getObject<PropagatorField>(par().q1);
+    PropagatorField &q2 = *env().template getObject<PropagatorField>(par().q2);
+    PropagatorField &q3 = *env().template getObject<PropagatorField>(par().q3);
+    PropagatorField &q4 = *env().template getObject<PropagatorField>(par().q4);
+    Gamma g5            = Gamma(Gamma::Algebra::Gamma5);
+    LatticeComplex        expbuf(env().getGrid());
+    std::vector<TComplex> corrbuf;
+    std::vector<Result>   result(n_eye_diag);
+    unsigned int ndim   = env().getNd();
+
+    PropagatorField              tmp1(env().getGrid());
+    LatticeComplex               tmp2(env().getGrid());
+    std::vector<PropagatorField> S_body(ndim, tmp1);
+    std::vector<PropagatorField> S_loop(ndim, tmp1);
+    std::vector<LatticeComplex>  E_body(ndim, tmp2);
+    std::vector<LatticeComplex>  E_loop(ndim, tmp2);
+
+    // Setup for S-type contractions.
+    for (int mu = 0; mu < ndim; ++mu)
+    {
+        S_body[mu] = MAKE_SE_BODY(q1, q2, q3, GammaL(Gamma::gmu[mu]));
+        S_loop[mu] = MAKE_SE_LOOP(q4, GammaL(Gamma::gmu[mu]));
+    }
+
+    // Perform S-type contractions.    
+    SUM_MU(expbuf, trace(S_body[mu]*S_loop[mu]))
+    MAKE_DIAG(expbuf, corrbuf, result[S_diag], "HW_S")
+
+    // Recycle sub-expressions for E-type contractions.
+    for (unsigned int mu = 0; mu < ndim; ++mu)
+    {
+        E_body[mu] = trace(S_body[mu]);
+        E_loop[mu] = trace(S_loop[mu]);
+    }
+
+    // Perform E-type contractions.
+    SUM_MU(expbuf, E_body[mu]*E_loop[mu])
+    MAKE_DIAG(expbuf, corrbuf, result[E_diag], "HW_E")
+
+    write(writer, "HW_Eye", result);
+}

extras/Hadrons/Modules/MContraction/WeakHamiltonianEye.hpp

@@ -0,0 +1,58 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Modules/MContraction/WeakHamiltonianEye.hpp
+
+Copyright (C) 2017
+
+Author: Andrew Lawson    <andrew.lawson1991@gmail.com>
+
+This program is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+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 Hadrons_WeakHamiltonianEye_hpp_
+#define Hadrons_WeakHamiltonianEye_hpp_
+
+#include <Grid/Hadrons/Modules/MContraction/WeakHamiltonian.hpp>
+
+BEGIN_HADRONS_NAMESPACE
+
+/******************************************************************************
+ *                         WeakHamiltonianEye                                 *
+ ******************************************************************************/
+BEGIN_MODULE_NAMESPACE(MContraction)
+
+enum
+{
+    S_diag = 0,
+    E_diag = 1,
+    n_eye_diag = 2
+};
+
+// Saucer and Eye subdiagram contractions.
+#define MAKE_SE_BODY(Q_1, Q_2, Q_3, gamma) (Q_3*g5*Q_1*adj(Q_2)*g5*gamma)
+#define MAKE_SE_LOOP(Q_loop, gamma) (Q_loop*gamma)
+
+MAKE_WEAK_MODULE(WeakHamiltonianEye)
+
+END_MODULE_NAMESPACE
+
+END_HADRONS_NAMESPACE
+
+#endif // Hadrons_WeakHamiltonianEye_hpp_

extras/Hadrons/Modules/MContraction/WeakHamiltonianNonEye.cc

@@ -0,0 +1,139 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Modules/MContraction/WeakHamiltonianNonEye.cc
+
+Copyright (C) 2017
+
+Author: Andrew Lawson    <andrew.lawson1991@gmail.com>
+
+This program is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+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/Hadrons/Modules/MContraction/WeakHamiltonianNonEye.hpp>
+
+using namespace Grid;
+using namespace Hadrons;
+using namespace MContraction;
+
+/*
+ * Weak Hamiltonian current-current contractions, Non-Eye-type.
+ * 
+ * These contractions are generated by the Q1 and Q2 operators in the physical
+ * basis (see e.g. Fig 3 of arXiv:1507.03094).
+ * 
+ * Schematic:     
+ *            q2             q3          |           q2              q3
+ *          /--<--¬       /--<--¬        |        /--<--¬         /--<--¬       
+ *         /       \     /       \       |       /       \       /       \      
+ *        /         \   /         \      |      /         \     /         \     
+ *       /           \ /           \     |     /           \   /           \    
+ *    i *             * H_W         *  f |  i *             * * H_W         * f 
+ *      \             *             |    |     \           /   \           /
+ *       \           / \           /     |      \         /     \         /    
+ *        \         /   \         /      |       \       /       \       /  
+ *         \       /     \       /       |        \-->--/         \-->--/      
+ *          \-->--/       \-->--/        |          q1               q4 
+ *            q1             q4          |
+ *                Connected (C)          |                 Wing (W)
+ *
+ * C: trace(q1*adj(q2)*g5*gL[mu]*q3*adj(q4)*g5*gL[mu])
+ * W: trace(q1*adj(q2)*g5*gL[mu])*trace(q3*adj(q4)*g5*gL[mu])
+ * 
+ */
+
+/******************************************************************************
+ *                  TWeakHamiltonianNonEye implementation                     *
+ ******************************************************************************/
+// constructor /////////////////////////////////////////////////////////////////
+TWeakHamiltonianNonEye::TWeakHamiltonianNonEye(const std::string name)
+: Module<WeakHamiltonianPar>(name)
+{}
+
+// dependencies/products ///////////////////////////////////////////////////////
+std::vector<std::string> TWeakHamiltonianNonEye::getInput(void)
+{
+    std::vector<std::string> in = {par().q1, par().q2, par().q3, par().q4};
+    
+    return in;
+}
+
+std::vector<std::string> TWeakHamiltonianNonEye::getOutput(void)
+{
+    std::vector<std::string> out = {getName()};
+    
+    return out;
+}
+
+// setup ///////////////////////////////////////////////////////////////////////
+void TWeakHamiltonianNonEye::setup(void)
+{
+
+}
+
+// execution ///////////////////////////////////////////////////////////////////
+void TWeakHamiltonianNonEye::execute(void)
+{
+    LOG(Message) << "Computing Weak Hamiltonian (Non-Eye type) contractions '" 
+                 << getName() << "' using quarks '" << par().q1 << "', '" 
+                 << par().q2 << ", '" << par().q3 << "' and '" << par().q4 
+                 << "'." << std::endl;
+    
+    CorrWriter             writer(par().output);
+    PropagatorField &q1 = *env().template getObject<PropagatorField>(par().q1);
+    PropagatorField &q2 = *env().template getObject<PropagatorField>(par().q2);
+    PropagatorField &q3 = *env().template getObject<PropagatorField>(par().q3);
+    PropagatorField &q4 = *env().template getObject<PropagatorField>(par().q4);
+    Gamma g5            = Gamma(Gamma::Algebra::Gamma5);
+    LatticeComplex        expbuf(env().getGrid());
+    std::vector<TComplex> corrbuf;
+    std::vector<Result>   result(n_noneye_diag); 
+    unsigned int ndim   = env().getNd();
+
+    PropagatorField              tmp1(env().getGrid());
+    LatticeComplex               tmp2(env().getGrid());
+    std::vector<PropagatorField> C_i_side_loop(ndim, tmp1);
+    std::vector<PropagatorField> C_f_side_loop(ndim, tmp1);
+    std::vector<LatticeComplex>  W_i_side_loop(ndim, tmp2);
+    std::vector<LatticeComplex>  W_f_side_loop(ndim, tmp2);
+
+    // Setup for C-type contractions.
+    for (int mu = 0; mu < ndim; ++mu)
+    {
+        C_i_side_loop[mu] = MAKE_CW_SUBDIAG(q1, q2, GammaL(Gamma::gmu[mu]));
+        C_f_side_loop[mu] = MAKE_CW_SUBDIAG(q3, q4, GammaL(Gamma::gmu[mu]));
+    }
+
+    // Perform C-type contractions.    
+    SUM_MU(expbuf, trace(C_i_side_loop[mu]*C_f_side_loop[mu]))
+    MAKE_DIAG(expbuf, corrbuf, result[C_diag], "HW_C")
+
+    // Recycle sub-expressions for W-type contractions.
+    for (unsigned int mu = 0; mu < ndim; ++mu)
+    {
+        W_i_side_loop[mu] = trace(C_i_side_loop[mu]);
+        W_f_side_loop[mu] = trace(C_f_side_loop[mu]);
+    }
+
+    // Perform W-type contractions.
+    SUM_MU(expbuf, W_i_side_loop[mu]*W_f_side_loop[mu])
+    MAKE_DIAG(expbuf, corrbuf, result[W_diag], "HW_W")
+
+    write(writer, "HW_NonEye", result);
+}

extras/Hadrons/Modules/MContraction/WeakHamiltonianNonEye.hpp

@@ -0,0 +1,57 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Modules/MContraction/WeakHamiltonianNonEye.hpp
+
+Copyright (C) 2017
+
+Author: Andrew Lawson    <andrew.lawson1991@gmail.com>
+
+This program is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+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 Hadrons_WeakHamiltonianNonEye_hpp_
+#define Hadrons_WeakHamiltonianNonEye_hpp_
+
+#include <Grid/Hadrons/Modules/MContraction/WeakHamiltonian.hpp>
+
+BEGIN_HADRONS_NAMESPACE
+
+/******************************************************************************
+ *                         WeakHamiltonianNonEye                              *
+ ******************************************************************************/
+BEGIN_MODULE_NAMESPACE(MContraction)
+
+enum
+{
+    W_diag = 0,
+    C_diag = 1,
+    n_noneye_diag = 2
+};
+
+// Wing and Connected subdiagram contractions
+#define MAKE_CW_SUBDIAG(Q_1, Q_2, gamma) (Q_1*adj(Q_2)*g5*gamma)
+
+MAKE_WEAK_MODULE(WeakHamiltonianNonEye)
+
+END_MODULE_NAMESPACE
+
+END_HADRONS_NAMESPACE
+
+#endif // Hadrons_WeakHamiltonianNonEye_hpp_

extras/Hadrons/Modules/MContraction/WeakNeutral4ptDisc.cc

@@ -0,0 +1,135 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Modules/MContraction/WeakNeutral4ptDisc.cc
+
+Copyright (C) 2017
+
+Author: Andrew Lawson    <andrew.lawson1991@gmail.com>
+
+This program is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+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/Hadrons/Modules/MContraction/WeakNeutral4ptDisc.hpp>
+
+using namespace Grid;
+using namespace Hadrons;
+using namespace MContraction;
+
+/*
+ * Weak Hamiltonian + current contractions, disconnected topology for neutral 
+ * mesons.
+ * 
+ * These contractions are generated by operators Q_1,...,10 of the dS=1 Weak
+ * Hamiltonian in the physical basis and an additional current J (see e.g. 
+ * Fig 11 of arXiv:1507.03094).
+ * 
+ * Schematic:
+ *                        
+ *           q2          q4             q3
+ *       /--<--¬     /---<--¬       /---<--¬
+ *     /         \ /         \     /        \
+ *  i *           * H_W      |  J *          * f
+ *     \         / \         /     \        /
+ *      \--->---/   \-------/       \------/
+ *          q1 
+ * 
+ * options
+ * - q1: input propagator 1 (string)
+ * - q2: input propagator 2 (string)
+ * - q3: input propagator 3 (string), assumed to be sequential propagator 
+ * - q4: input propagator 4 (string), assumed to be a loop
+ * 
+ * type 1: trace(q1*adj(q2)*g5*gL[mu])*trace(loop*gL[mu])*trace(q3*g5)
+ * type 2: trace(q1*adj(q2)*g5*gL[mu]*loop*gL[mu])*trace(q3*g5)
+ */
+
+/*******************************************************************************
+ *                  TWeakNeutral4ptDisc implementation                         *
+ ******************************************************************************/
+// constructor /////////////////////////////////////////////////////////////////
+TWeakNeutral4ptDisc::TWeakNeutral4ptDisc(const std::string name)
+: Module<WeakHamiltonianPar>(name)
+{}
+
+// dependencies/products ///////////////////////////////////////////////////////
+std::vector<std::string> TWeakNeutral4ptDisc::getInput(void)
+{
+    std::vector<std::string> in = {par().q1, par().q2, par().q3, par().q4};
+    
+    return in;
+}
+
+std::vector<std::string> TWeakNeutral4ptDisc::getOutput(void)
+{
+    std::vector<std::string> out = {getName()};
+    
+    return out;
+}
+
+// setup ///////////////////////////////////////////////////////////////////////
+void TWeakNeutral4ptDisc::setup(void)
+{
+
+}
+
+// execution ///////////////////////////////////////////////////////////////////
+void TWeakNeutral4ptDisc::execute(void)
+{
+    LOG(Message) << "Computing Weak Hamiltonian neutral disconnected contractions '" 
+                 << getName() << "' using quarks '" << par().q1 << "', '" 
+                 << par().q2 << ", '" << par().q3 << "' and '" << par().q4 
+                 << "'." << std::endl;
+
+    CorrWriter             writer(par().output);
+    PropagatorField &q1 = *env().template getObject<PropagatorField>(par().q1);
+    PropagatorField &q2 = *env().template getObject<PropagatorField>(par().q2);
+    PropagatorField &q3 = *env().template getObject<PropagatorField>(par().q3);
+    PropagatorField &q4 = *env().template getObject<PropagatorField>(par().q4);
+    Gamma g5            = Gamma(Gamma::Algebra::Gamma5);
+    LatticeComplex        expbuf(env().getGrid());
+    std::vector<TComplex> corrbuf;
+    std::vector<Result>   result(n_neut_disc_diag);
+    unsigned int ndim   = env().getNd();
+
+    PropagatorField              tmp(env().getGrid());
+    std::vector<PropagatorField> meson(ndim, tmp);
+    std::vector<PropagatorField> loop(ndim, tmp);
+    LatticeComplex               curr(env().getGrid());
+
+    // Setup for type 1 contractions.
+    for (int mu = 0; mu < ndim; ++mu)
+    {
+        meson[mu] = MAKE_DISC_MESON(q1, q2, GammaL(Gamma::gmu[mu]));
+        loop[mu] = MAKE_DISC_LOOP(q4, GammaL(Gamma::gmu[mu]));
+    }
+    curr = MAKE_DISC_CURR(q3, GammaL(Gamma::Algebra::Gamma5));
+
+    // Perform type 1 contractions.    
+    SUM_MU(expbuf, trace(meson[mu]*loop[mu]))
+    expbuf *= curr;
+    MAKE_DIAG(expbuf, corrbuf, result[neut_disc_1_diag], "HW_disc0_1")
+
+    // Perform type 2 contractions.
+    SUM_MU(expbuf, trace(meson[mu])*trace(loop[mu]))
+    expbuf *= curr;
+    MAKE_DIAG(expbuf, corrbuf, result[neut_disc_2_diag], "HW_disc0_2")
+
+    write(writer, "HW_disc0", result);
+}

extras/Hadrons/Modules/MContraction/WeakNeutral4ptDisc.hpp

@@ -0,0 +1,59 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Modules/MContraction/WeakNeutral4ptDisc.hpp
+
+Copyright (C) 2017
+
+Author: Andrew Lawson    <andrew.lawson1991@gmail.com>
+
+This program is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+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 Hadrons_WeakNeutral4ptDisc_hpp_
+#define Hadrons_WeakNeutral4ptDisc_hpp_
+
+#include <Grid/Hadrons/Modules/MContraction/WeakHamiltonian.hpp>
+
+BEGIN_HADRONS_NAMESPACE
+
+/******************************************************************************
+ *                         WeakNeutral4ptDisc                                 *
+ ******************************************************************************/
+BEGIN_MODULE_NAMESPACE(MContraction)
+
+enum
+{
+    neut_disc_1_diag = 0,
+    neut_disc_2_diag = 1,
+    n_neut_disc_diag = 2
+};
+
+// Neutral 4pt disconnected subdiagram contractions.
+#define MAKE_DISC_MESON(Q_1, Q_2, gamma) (Q_1*adj(Q_2)*g5*gamma)
+#define MAKE_DISC_LOOP(Q_LOOP, gamma) (Q_LOOP*gamma)
+#define MAKE_DISC_CURR(Q_c, gamma) (trace(Q_c*gamma))
+
+MAKE_WEAK_MODULE(WeakNeutral4ptDisc)
+
+END_MODULE_NAMESPACE
+
+END_HADRONS_NAMESPACE
+
+#endif // Hadrons_WeakNeutral4ptDisc_hpp_

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));
 }

extras/Hadrons/Modules/MLoop/NoiseLoop.hpp

@@ -0,0 +1,132 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Modules/MLoop/NoiseLoop.hpp
+
+Copyright (C) 2016
+
+Author: Andrew Lawson <andrew.lawson1991@gmail.com>
+
+This program is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+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 Hadrons_NoiseLoop_hpp_
+#define Hadrons_NoiseLoop_hpp_
+
+#include <Grid/Hadrons/Global.hpp>
+#include <Grid/Hadrons/Module.hpp>
+#include <Grid/Hadrons/ModuleFactory.hpp>
+
+BEGIN_HADRONS_NAMESPACE
+
+/*
+ 
+ Noise loop propagator
+ -----------------------------
+ * loop_x = q_x * adj(eta_x)
+ 
+ * options:
+ - q = Result of inversion on noise source.
+ - eta = noise source.
+
+ */
+
+
+/******************************************************************************
+ *                         NoiseLoop                                          *
+ ******************************************************************************/
+BEGIN_MODULE_NAMESPACE(MLoop)
+
+class NoiseLoopPar: Serializable
+{
+public:
+    GRID_SERIALIZABLE_CLASS_MEMBERS(NoiseLoopPar,
+                                    std::string, q,
+                                    std::string, eta);
+};
+
+template <typename FImpl>
+class TNoiseLoop: public Module<NoiseLoopPar>
+{
+public:
+    TYPE_ALIASES(FImpl,);
+public:
+    // constructor
+    TNoiseLoop(const std::string name);
+    // destructor
+    virtual ~TNoiseLoop(void) = default;
+    // dependency relation
+    virtual std::vector<std::string> getInput(void);
+    virtual std::vector<std::string> getOutput(void);
+    // setup
+    virtual void setup(void);
+    // execution
+    virtual void execute(void);
+};
+
+MODULE_REGISTER_NS(NoiseLoop, TNoiseLoop<FIMPL>, MLoop);
+
+/******************************************************************************
+ *                 TNoiseLoop implementation                                  *
+ ******************************************************************************/
+// constructor /////////////////////////////////////////////////////////////////
+template <typename FImpl>
+TNoiseLoop<FImpl>::TNoiseLoop(const std::string name)
+: Module<NoiseLoopPar>(name)
+{}
+
+// dependencies/products ///////////////////////////////////////////////////////
+template <typename FImpl>
+std::vector<std::string> TNoiseLoop<FImpl>::getInput(void)
+{
+    std::vector<std::string> in = {par().q, par().eta};
+    
+    return in;
+}
+
+template <typename FImpl>
+std::vector<std::string> TNoiseLoop<FImpl>::getOutput(void)
+{
+    std::vector<std::string> out = {getName()};
+    
+    return out;
+}
+
+// setup ///////////////////////////////////////////////////////////////////////
+template <typename FImpl>
+void TNoiseLoop<FImpl>::setup(void)
+{
+    env().template registerLattice<PropagatorField>(getName());
+}
+
+// execution ///////////////////////////////////////////////////////////////////
+template <typename FImpl>
+void TNoiseLoop<FImpl>::execute(void)
+{
+    PropagatorField &loop = *env().template createLattice<PropagatorField>(getName());
+    PropagatorField &q    = *env().template getObject<PropagatorField>(par().q);
+    PropagatorField &eta  = *env().template getObject<PropagatorField>(par().eta);
+    loop = q*adj(eta);
+}
+
+END_MODULE_NAMESPACE
+
+END_HADRONS_NAMESPACE
+
+#endif // Hadrons_NoiseLoop_hpp_

extras/Hadrons/Modules/MSource/SeqGamma.hpp

@@ -6,6 +6,7 @@ Source file: extras/Hadrons/Modules/MSource/SeqGamma.hpp
 
 Copyright (C) 2015
 Copyright (C) 2016
+Copyright (C) 2017
 
 Author: Antonin Portelli <antonin.portelli@me.com>
 
@@ -149,9 +150,9 @@ void TSeqGamma<FImpl>::execute(void)
     for(unsigned int mu = 0; mu < env().getNd(); mu++)
     {
         LatticeCoordinate(coor, mu);
-        ph = ph + p[mu]*coor;
+        ph = ph + p[mu]*coor*((1./(env().getGrid()->_fdimensions[mu])));
     }
-    ph = exp(i*ph);
+    ph = exp((Real)(2*M_PI)*i*ph);
     LatticeCoordinate(t, Tp);
     src = where((t >= par().tA) and (t <= par().tB), ph*(g*q), 0.*q);
 }

extras/Hadrons/Modules/MSource/Wall.hpp

@@ -0,0 +1,147 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Modules/MSource/Wall.hpp
+
+Copyright (C) 2017
+
+Author: Andrew Lawson <andrew.lawson1991@gmail.com>
+
+This program is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+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 Hadrons_WallSource_hpp_
+#define Hadrons_WallSource_hpp_
+
+#include <Grid/Hadrons/Global.hpp>
+#include <Grid/Hadrons/Module.hpp>
+#include <Grid/Hadrons/ModuleFactory.hpp>
+
+BEGIN_HADRONS_NAMESPACE
+
+/*
+ 
+ Wall source
+ -----------------------------
+ * src_x = delta(x_3 - tW) * exp(i x.mom)
+ 
+ * options:
+ - tW: source timeslice (integer)
+ - mom: momentum insertion, space-separated float sequence (e.g ".1 .2 1. 0.")
+ 
+ */
+
+/******************************************************************************
+ *                         Wall                                               *
+ ******************************************************************************/
+BEGIN_MODULE_NAMESPACE(MSource)
+
+class WallPar: Serializable
+{
+public:
+    GRID_SERIALIZABLE_CLASS_MEMBERS(WallPar,
+                                    unsigned int, tW,
+                                    std::string, mom);
+};
+
+template <typename FImpl>
+class TWall: public Module<WallPar>
+{
+public:
+    TYPE_ALIASES(FImpl,);
+public:
+    // constructor
+    TWall(const std::string name);
+    // destructor
+    virtual ~TWall(void) = default;
+    // dependency relation
+    virtual std::vector<std::string> getInput(void);
+    virtual std::vector<std::string> getOutput(void);
+    // setup
+    virtual void setup(void);
+    // execution
+    virtual void execute(void);
+};
+
+MODULE_REGISTER_NS(Wall, TWall<FIMPL>, MSource);
+
+/******************************************************************************
+ *                 TWall implementation                                       *
+ ******************************************************************************/
+// constructor /////////////////////////////////////////////////////////////////
+template <typename FImpl>
+TWall<FImpl>::TWall(const std::string name)
+: Module<WallPar>(name)
+{}
+
+// dependencies/products ///////////////////////////////////////////////////////
+template <typename FImpl>
+std::vector<std::string> TWall<FImpl>::getInput(void)
+{
+    std::vector<std::string> in;
+    
+    return in;
+}
+
+template <typename FImpl>
+std::vector<std::string> TWall<FImpl>::getOutput(void)
+{
+    std::vector<std::string> out = {getName()};
+    
+    return out;
+}
+
+// setup ///////////////////////////////////////////////////////////////////////
+template <typename FImpl>
+void TWall<FImpl>::setup(void)
+{
+    env().template registerLattice<PropagatorField>(getName());
+}
+
+// execution ///////////////////////////////////////////////////////////////////
+template <typename FImpl>
+void TWall<FImpl>::execute(void)
+{    
+    LOG(Message) << "Generating wall source at t = " << par().tW 
+                 << " with momentum " << par().mom << std::endl;
+    
+    PropagatorField &src = *env().template createLattice<PropagatorField>(getName());
+    Lattice<iScalar<vInteger>> t(env().getGrid());
+    LatticeComplex             ph(env().getGrid()), coor(env().getGrid());
+    std::vector<Real>          p;
+    Complex                    i(0.0,1.0);
+    
+    p  = strToVec<Real>(par().mom);
+    ph = zero;
+    for(unsigned int mu = 0; mu < Nd; mu++)
+    {
+        LatticeCoordinate(coor, mu);
+        ph = ph + p[mu]*coor*((1./(env().getGrid()->_fdimensions[mu])));
+    }
+    ph = exp((Real)(2*M_PI)*i*ph);
+    LatticeCoordinate(t, Tp);
+    src = 1.;
+    src = where((t == par().tW), src*ph, 0.*src);
+}
+
+END_MODULE_NAMESPACE
+
+END_HADRONS_NAMESPACE
+
+#endif // Hadrons_WallSource_hpp_

extras/Hadrons/Modules/Quark.hpp

@@ -173,7 +173,7 @@ void TQuark<FImpl>::execute(void)
                 *env().template getObject<PropagatorField>(getName());
             
             axpby_ssp_pminus(sol, 0., sol, 1., sol, 0, 0);
-            axpby_ssp_pplus(sol, 0., sol, 1., sol, 0, Ls_-1);
+            axpby_ssp_pplus(sol, 1., sol, 1., sol, 0, Ls_-1);
             ExtractSlice(tmp, sol, 0, 0);
             FermToProp(p4d, tmp, s, c);
         }

extras/Hadrons/modules.inc

@@ -1,4 +1,7 @@
 modules_cc =\
+  Modules/MContraction/WeakHamiltonianEye.cc \
+  Modules/MContraction/WeakHamiltonianNonEye.cc \
+  Modules/MContraction/WeakNeutral4ptDisc.cc \
   Modules/MGauge/Load.cc \
   Modules/MGauge/Random.cc \
   Modules/MGauge/Unit.cc
@@ -7,13 +10,21 @@ modules_hpp =\
   Modules/MAction/DWF.hpp \
   Modules/MAction/Wilson.hpp \
   Modules/MContraction/Baryon.hpp \
+  Modules/MContraction/DiscLoop.hpp \
+  Modules/MContraction/Gamma3pt.hpp \
   Modules/MContraction/Meson.hpp \
+  Modules/MContraction/WeakHamiltonian.hpp \
+  Modules/MContraction/WeakHamiltonianEye.hpp \
+  Modules/MContraction/WeakHamiltonianNonEye.hpp \
+  Modules/MContraction/WeakNeutral4ptDisc.hpp \
   Modules/MGauge/Load.hpp \
   Modules/MGauge/Random.hpp \
   Modules/MGauge/Unit.hpp \
+  Modules/MLoop/NoiseLoop.hpp \
   Modules/MSolver/RBPrecCG.hpp \
   Modules/MSource/Point.hpp \
   Modules/MSource/SeqGamma.hpp \
+  Modules/MSource/Wall.hpp \
   Modules/MSource/Z2.hpp \
   Modules/Quark.hpp
 

gcc-bug-report/README

@@ -21,3 +21,16 @@ problem. The test case works with icpc and with clang++, but fails consistently
 current variants.
 
 Peter
+
+
+************
+
+Second GCC bug reported, see Issue 100.
+
+https://wandbox.org/permlink/tzssJza6R9XnqANw
+https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80652
+
+Getting Travis fails under gcc-5 for Test_simd, now that I added more comprehensive testing to the
+CI test suite. The limitations of Travis runtime limits & weak cores are being shown.
+
+Travis uses 5.4.1 for g++-5.

grid-config.in

@@ -0,0 +1,86 @@
+#! /bin/sh
+
+prefix=@prefix@
+exec_prefix=@exec_prefix@
+includedir=@includedir@
+
+usage()
+{
+  cat <<EOF
+Usage: grid-config [OPTION]
+
+Known values for OPTION are:
+
+  --prefix     show Grid installation prefix
+  --cxxflags   print pre-processor and compiler flags
+  --ldflags    print library linking flags
+  --libs       print library linking information
+  --summary    print full build summary
+  --help       display this help and exit
+  --version    output version information
+  --git        print git revision
+
+EOF
+  
+  exit $1
+}
+
+if test $# -eq 0; then
+  usage 1
+fi
+
+cflags=false
+libs=false
+
+while test $# -gt 0; do
+  case "$1" in
+    -*=*) optarg=`echo "$1" | sed 's/[-_a-zA-Z0-9]*=//'` ;;
+    *) optarg= ;;
+  esac
+  
+  case "$1" in
+    --prefix)
+      echo $prefix
+    ;;
+    
+    --version)
+      echo @VERSION@
+      exit 0
+    ;;
+    
+    --git)
+      echo "@GRID_BRANCH@ @GRID_SHA@"
+      exit 0
+    ;;
+    
+    --help)
+      usage 0
+    ;;
+    
+    --cxxflags)
+      echo @GRID_CXXFLAGS@
+    ;;
+    
+    --ldflags)
+      echo @GRID_LDFLAGS@
+    ;;
+    
+    --libs)
+      echo @GRID_LIBS@
+    ;;
+    
+    --summary)
+      echo ""
+      echo "@GRID_SUMMARY@"
+      echo ""
+    ;;
+    
+    *)
+      usage
+      exit 1
+    ;;
+  esac
+  shift
+done
+
+exit 0

lib/DisableWarnings.h

@@ -0,0 +1,37 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid
+
+Source file: ./lib/DisableWarnings.h
+
+Copyright (C) 2016
+
+Author: Guido Cossu <guido.cossu@ed.ac.uk>
+
+This program is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 2 of the License, or
+(at your option) any later version.
+
+This program is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along
+with this program; if not, write to the Free Software Foundation, Inc.,
+51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+See the full license in the file "LICENSE" in the top level distribution
+directory
+*************************************************************************************/
+/*  END LEGAL */
+
+#ifndef DISABLE_WARNINGS_H
+#define DISABLE_WARNINGS_H
+
+ //disables and intel compiler specific warning (in json.hpp)
+#pragma warning disable 488  
+
+
+#endif

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/GridCore.h

@@ -38,28 +38,7 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 #ifndef GRID_BASE_H
 #define GRID_BASE_H
 
-///////////////////
-// Std C++ dependencies
-///////////////////
-#include <cassert>
-#include <complex>
-#include <vector>
-#include <iostream>
-#include <iomanip>
-#include <random>
-#include <functional>
-#include <stdio.h>
-#include <stdlib.h>
-#include <stdio.h>
-#include <signal.h>
-#include <ctime>
-#include <sys/time.h>
-#include <chrono>
-
-///////////////////
-// Grid headers
-///////////////////
-#include "Config.h"
+#include <Grid/GridStd.h>
 
 #include <Grid/perfmon/Timer.h>
 #include <Grid/perfmon/PerfCount.h>

lib/GridStd.h

@@ -0,0 +1,29 @@
+#ifndef GRID_STD_H
+#define GRID_STD_H
+
+///////////////////
+// Std C++ dependencies
+///////////////////
+#include <cassert>
+#include <complex>
+#include <vector>
+#include <string>
+#include <iostream>
+#include <iomanip>
+#include <random>
+#include <functional>
+#include <stdio.h>
+#include <stdlib.h>
+#include <stdio.h>
+#include <signal.h>
+#include <ctime>
+#include <sys/time.h>
+#include <chrono>
+#include <zlib.h>
+
+///////////////////
+// Grid config
+///////////////////
+#include "Config.h"
+
+#endif /* GRID_STD_H */

lib/Grid_Eigen_Dense.h

@@ -0,0 +1,9 @@
+#pragma once
+#if defined __GNUC__
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wdeprecated-declarations"
+#endif
+#include <Grid/Eigen/Dense>
+#if defined __GNUC__
+#pragma GCC diagnostic pop
+#endif

lib/algorithms/approx/Chebyshev.h

@@ -197,8 +197,9 @@ namespace Grid {
     void operator() (LinearOperatorBase<Field> &Linop, const Field &in, Field &out) {
 
       GridBase *grid=in._grid;
-//std::cout << "Chevyshef(): in._grid="<<in._grid<<std::endl;
-//<<" Linop.Grid()="<<Linop.Grid()<<"Linop.RedBlackGrid()="<<Linop.RedBlackGrid()<<std::endl;
+
+      // std::cout << "Chevyshef(): in._grid="<<in._grid<<std::endl;
+      //std::cout <<" Linop.Grid()="<<Linop.Grid()<<"Linop.RedBlackGrid()="<<Linop.RedBlackGrid()<<std::endl;
 
       int vol=grid->gSites();
 

lib/algorithms/approx/Remez.h

@@ -16,7 +16,7 @@
 #define INCLUDED_ALG_REMEZ_H
 
 #include <stddef.h>
-#include <Config.h>
+#include <Grid/GridStd.h>
 
 #ifdef HAVE_LIBGMP
 #include "bigfloat.h"

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/ConjugateGradient.h

@@ -123,8 +123,11 @@ class ConjugateGradient : public OperatorFunction<Field> {
       p = p * b + r;
 
       LinalgTimer.Stop();
+
       std::cout << GridLogIterative << "ConjugateGradient: Iteration " << k
                 << " residual " << cp << " target " << rsq << std::endl;
+      std::cout << GridLogDebug << "a = "<< a << " b_pred = "<< b_pred << "  b = "<< b << std::endl;
+      std::cout << GridLogDebug << "qq = "<< qq << " d = "<< d << "  c = "<< c << std::endl;
 
       // Stopping condition
       if (cp <= rsq) {
@@ -132,8 +135,6 @@ class ConjugateGradient : public OperatorFunction<Field> {
         Linop.HermOpAndNorm(psi, mmp, d, qq);
         p = mmp - src;
 
-        RealD mmpnorm = sqrt(norm2(mmp));
-        RealD psinorm = sqrt(norm2(psi));
         RealD srcnorm = sqrt(norm2(src));
         RealD resnorm = sqrt(norm2(p));
         RealD true_residual = resnorm / srcnorm;
@@ -157,8 +158,10 @@ class ConjugateGradient : public OperatorFunction<Field> {
     }
     std::cout << GridLogMessage << "ConjugateGradient did NOT converge"
               << std::endl;
+
     if (ErrorOnNoConverge) assert(0);
     IterationsToComplete = k;
+
   }
 };
 }

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

@@ -6,8 +6,9 @@
 
     Copyright (C) 2015
 
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: paboyle <paboyle@ph.ed.ac.uk>
+    Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+    Author: paboyle <paboyle@ph.ed.ac.uk>
+    Author: Guido Cossu <guido.cossu@ed.ac.uk>
 
     This program is free software; you can redistribute it and/or modify
     it under the terms of the GNU General Public License as published by
@@ -49,7 +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
@@ -62,13 +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:
 
@@ -121,6 +120,12 @@ public:
       Lexicographic::CoorFromIndex(coor,Oindex,_rdimensions);
     }
 
+    inline void InOutCoorToLocalCoor (std::vector<int> &ocoor, std::vector<int> &icoor, std::vector<int> &lcoor) {
+      lcoor.resize(_ndimension);
+      for (int d = 0; d < _ndimension; d++)
+        lcoor[d] = ocoor[d] + _rdimensions[d] * icoor[d];
+    }
+
     //////////////////////////////////////////////////////////
     // SIMD lane addressing
     //////////////////////////////////////////////////////////
@@ -128,6 +133,7 @@ public:
     {
       Lexicographic::CoorFromIndex(coor,lane,_simd_layout);
     }
+
     inline int PermuteDim(int dimension){
       return _simd_layout[dimension]>1;
     }
@@ -168,11 +174,30 @@ 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;};
     inline const std::vector<int> &VirtualLocalDimensions(void) { return _ldimensions;};
 
+    ////////////////////////////////////////////////////////////////
+    // Utility to print the full decomposition details 
+    ////////////////////////////////////////////////////////////////
+
+    void show_decomposition(){
+      std::cout << GridLogMessage << "Full Dimensions    : " << _fdimensions << std::endl;
+      std::cout << GridLogMessage << "Global Dimensions  : " << _gdimensions << std::endl;
+      std::cout << GridLogMessage << "Local Dimensions   : " << _ldimensions << std::endl;
+      std::cout << GridLogMessage << "Reduced Dimensions : " << _rdimensions << std::endl;
+      std::cout << GridLogMessage << "Outer strides      : " << _ostride << std::endl;
+      std::cout << GridLogMessage << "Inner strides      : " << _istride << std::endl;
+      std::cout << GridLogMessage << "iSites             : " << _isites << std::endl;
+      std::cout << GridLogMessage << "oSites             : " << _osites << std::endl;
+      std::cout << GridLogMessage << "lSites             : " << lSites() << std::endl;        
+      std::cout << GridLogMessage << "gSites             : " << gSites() << std::endl;
+      std::cout << GridLogMessage << "Nd                 : " << _ndimension << std::endl;             
+    } 
+
     ////////////////////////////////////////////////////////////////
     // Global addressing
     ////////////////////////////////////////////////////////////////
@@ -184,6 +209,9 @@ public:
       assert(lidx<lSites());
       Lexicographic::CoorFromIndex(lcoor,lidx,_ldimensions);
     }
+
+
+
     void GlobalCoorToGlobalIndex(const std::vector<int> & gcoor,int & gidx){
       gidx=0;
       int mult=1;

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);
@@ -94,6 +96,8 @@ public:
 	  // Use a reduced simd grid
 	  _ldimensions[d]= _gdimensions[d]/_processors[d];  //local dimensions
 	  _rdimensions[d]= _ldimensions[d]/_simd_layout[d]; //overdecomposition
+	  _lstart[d]     = _processor_coor[d]*_ldimensions[d];
+	  _lend[d]       = _processor_coor[d]*_ldimensions[d]+_ldimensions[d]-1;
 	  _osites  *= _rdimensions[d];
 	  _isites  *= _simd_layout[d];
                 

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/cshift/Cshift_common.h

@@ -30,21 +30,11 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 
 namespace Grid {
 
-template<class vobj>
-class SimpleCompressor {
-public:
-  void Point(int) {};
-
-  vobj operator() (const vobj &arg) {
-    return arg;
-  }
-};
-
 ///////////////////////////////////////////////////////////////////
-// Gather for when there is no need to SIMD split with compression
+// Gather for when there is no need to SIMD split 
 ///////////////////////////////////////////////////////////////////
-template<class vobj,class cobj,class compressor> void 
-Gather_plane_simple (const Lattice<vobj> &rhs,commVector<cobj> &buffer,int dimension,int plane,int cbmask,compressor &compress, int off=0)
+template<class vobj> void 
+Gather_plane_simple (const Lattice<vobj> &rhs,commVector<vobj> &buffer,int dimension,int plane,int cbmask, int off=0)
 {
   int rd = rhs._grid->_rdimensions[dimension];
 
@@ -62,7 +52,7 @@ Gather_plane_simple (const Lattice<vobj> &rhs,commVector<cobj> &buffer,int dimen
       for(int b=0;b<e2;b++){
 	int o  = n*stride;
 	int bo = n*e2;
-	buffer[off+bo+b]=compress(rhs._odata[so+o+b]);
+	buffer[off+bo+b]=rhs._odata[so+o+b];
       }
     }
   } else { 
@@ -78,17 +68,16 @@ Gather_plane_simple (const Lattice<vobj> &rhs,commVector<cobj> &buffer,int dimen
        }
      }
      parallel_for(int i=0;i<table.size();i++){
-       buffer[off+table[i].first]=compress(rhs._odata[so+table[i].second]);
+       buffer[off+table[i].first]=rhs._odata[so+table[i].second];
      }
   }
 }
 
-
 ///////////////////////////////////////////////////////////////////
-// Gather for when there *is* need to SIMD split with compression
+// Gather for when there *is* need to SIMD split 
 ///////////////////////////////////////////////////////////////////
-template<class cobj,class vobj,class compressor> void 
-Gather_plane_extract(const Lattice<vobj> &rhs,std::vector<typename cobj::scalar_object *> pointers,int dimension,int plane,int cbmask,compressor &compress)
+template<class vobj> void 
+Gather_plane_extract(const Lattice<vobj> &rhs,std::vector<typename vobj::scalar_object *> pointers,int dimension,int plane,int cbmask)
 {
   int rd = rhs._grid->_rdimensions[dimension];
 
@@ -109,8 +98,8 @@ Gather_plane_extract(const Lattice<vobj> &rhs,std::vector<typename cobj::scalar_
 	int o      =   n*n1;
 	int offset = b+n*e2;
 	
-	cobj temp =compress(rhs._odata[so+o+b]);
-	extract<cobj>(temp,pointers,offset);
+	vobj temp =rhs._odata[so+o+b];
+	extract<vobj>(temp,pointers,offset);
 
       }
     }
@@ -127,32 +116,14 @@ Gather_plane_extract(const Lattice<vobj> &rhs,std::vector<typename cobj::scalar_
 	int offset = b+n*e2;
 
 	if ( ocb & cbmask ) {
-	  cobj temp =compress(rhs._odata[so+o+b]);
-	  extract<cobj>(temp,pointers,offset);
+	  vobj temp =rhs._odata[so+o+b];
+	  extract<vobj>(temp,pointers,offset);
 	}
       }
     }
   }
 }
 
-//////////////////////////////////////////////////////
-// Gather for when there is no need to SIMD split
-//////////////////////////////////////////////////////
-template<class vobj> void Gather_plane_simple (const Lattice<vobj> &rhs,commVector<vobj> &buffer, int dimension,int plane,int cbmask)
-{
-  SimpleCompressor<vobj> dontcompress;
-  Gather_plane_simple (rhs,buffer,dimension,plane,cbmask,dontcompress);
-}
-
-//////////////////////////////////////////////////////
-// Gather for when there *is* need to SIMD split
-//////////////////////////////////////////////////////
-template<class vobj> void Gather_plane_extract(const Lattice<vobj> &rhs,std::vector<typename vobj::scalar_object *> pointers,int dimension,int plane,int cbmask)
-{
-  SimpleCompressor<vobj> dontcompress;
-  Gather_plane_extract<vobj,vobj,decltype(dontcompress)>(rhs,pointers,dimension,plane,cbmask,dontcompress);
-}
-
 //////////////////////////////////////////////////////
 // Scatter for when there is no need to SIMD split
 //////////////////////////////////////////////////////
@@ -200,7 +171,7 @@ template<class vobj> void Scatter_plane_simple (Lattice<vobj> &rhs,commVector<vo
 //////////////////////////////////////////////////////
 // Scatter for when there *is* need to SIMD split
 //////////////////////////////////////////////////////
- template<class vobj,class cobj> void Scatter_plane_merge(Lattice<vobj> &rhs,std::vector<cobj *> pointers,int dimension,int plane,int cbmask)
+template<class vobj> void Scatter_plane_merge(Lattice<vobj> &rhs,std::vector<typename vobj::scalar_object *> pointers,int dimension,int plane,int cbmask)
 {
   int rd = rhs._grid->_rdimensions[dimension];
 

lib/cshift/Cshift_mpi.h

@@ -154,13 +154,7 @@ template<class vobj> void Cshift_comms(Lattice<vobj> &ret,const Lattice<vobj> &r
 			   recv_from_rank,
 			   bytes);
       grid->Barrier();
-      /*
-      for(int i=0;i<send_buf.size();i++){
-	assert(recv_buf.size()==buffer_size);
-	assert(send_buf.size()==buffer_size);
-	std::cout << "SendRecv_Cshift_comms ["<<i<<" "<< dimension<<"] snd "<<send_buf[i]<<" rcv " << recv_buf[i] << "  0x" << cbmask<<std::endl;
-      }
-      */
+
       Scatter_plane_simple (ret,recv_buf,dimension,x,cbmask);
     }
   }
@@ -246,13 +240,6 @@ template<class vobj> void  Cshift_comms_simd(Lattice<vobj> &ret,const Lattice<vo
 			     (void *)&recv_buf_extract[i][0],
 			     recv_from_rank,
 			     bytes);
-	/*
-	for(int w=0;w<recv_buf_extract[i].size();w++){
-	  assert(recv_buf_extract[i].size()==buffer_size);
-	  assert(send_buf_extract[i].size()==buffer_size);
-	  std::cout << "SendRecv_Cshift_comms ["<<w<<" "<< dimension<<"] recv "<<recv_buf_extract[i][w]<<" send " << send_buf_extract[nbr_lane][w]  << cbmask<<std::endl;
-	}
-	*/	
 	grid->Barrier();
 	rpointers[i] = &recv_buf_extract[i][0];
       } else { 

lib/lattice/Lattice_base.h

@@ -261,12 +261,22 @@ public:
   
   virtual ~Lattice(void) = default;
     
+  void reset(GridBase* grid) {
+    if (_grid != grid) {
+      _grid = grid;
+      _odata.resize(grid->oSites());
+      checkerboard = 0;
+    }
+  }
+  
+
   template<class sobj> strong_inline Lattice<vobj> & operator = (const sobj & r){
     parallel_for(int ss=0;ss<_grid->oSites();ss++){
       this->_odata[ss]=r;
     }
     return *this;
   }
+  
   template<class robj> strong_inline Lattice<vobj> & operator = (const Lattice<robj> & r){
     this->checkerboard = r.checkerboard;
     conformable(*this,r);
@@ -291,7 +301,7 @@ public:
     *this = (*this)+r;
     return *this;
   }
- }; // class Lattice
+}; // class Lattice
   
   template<class vobj> std::ostream& operator<< (std::ostream& stream, const Lattice<vobj> &o){
     std::vector<int> gcoor;

lib/lattice/Lattice_reduction.h

@@ -1,36 +1,28 @@
-    /*************************************************************************************
-
+/*************************************************************************************
     Grid physics library, www.github.com/paboyle/Grid 
-
     Source file: ./lib/lattice/Lattice_reduction.h
-
     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 */
 #ifndef GRID_LATTICE_REDUCTION_H
 #define GRID_LATTICE_REDUCTION_H
 
-#include <Grid/Eigen/Dense>
+#include <Grid/Grid_Eigen_Dense.h>
 
 namespace Grid {
 #ifdef GRID_WARN_SUBOPTIMAL
@@ -336,6 +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();
@@ -361,7 +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;
@@ -375,74 +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;
@@ -462,7 +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) 
 {    
@@ -478,21 +403,96 @@ static void sliceMaddMatrix (Lattice<vobj> &R,Eigen::MatrixXcd &aa,const Lattice
   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++){
-    ExtractSlice(Rslice,Y,i,Orthog);
-    for(int j=0;j<Nblock;j++){
-      ExtractSlice(Xslice,X,j,Orthog);
-      Rslice = Rslice + Xslice*(scale*aa(j,i));
+	s_x[i] = X[o+i*ostride];
       }
-    InsertSlice(Rslice,R,i,Orthog);
+
+      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;
@@ -507,24 +507,53 @@ static void sliceInnerProductMatrix(  Eigen::MatrixXcd &mat, const Lattice<vobj>
   
   mat = Eigen::MatrixXcd::Zero(Nblock,Nblock);
 
+  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++){
-    ExtractSlice(Lslice,lhs,i,Orthog);
-    for(int j=0;j<Nblock;j++){
-      ExtractSlice(Rslice,rhs,j,Orthog);
-      mat(i,j) = innerProduct(Lslice,Rslice);
+	Left [i] = lhs[o+i*ostride];
+	Right[i] = rhs[o+i*ostride];
       }
-  }
-#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;
+	auto tmp = innerProduct(Left[i],Right[j]);
+	vector_typeD rtmp = TensorRemove(tmp);
+	mat_thread(i,j) += Reduce(rtmp);
+      }}
+    }}
+#pragma omp critical
+    {
+      mat += mat_thread;
     }  
   }
-#endif
   return;
 }
 
 } /*END NAMESPACE GRID*/
 #endif
 
+
+

lib/lattice/Lattice_rng.h

@@ -6,8 +6,8 @@
 
     Copyright (C) 2015
 
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: paboyle <paboyle@ph.ed.ac.uk>
+    Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+    Author: Guido Cossu <guido.cossu@ed.ac.uk>
 
     This program is free software; you can redistribute it and/or modify
     it under the terms of the GNU General Public License as published by
@@ -75,6 +75,55 @@ namespace Grid {
     return multiplicity;
   }
 
+  
+// merge of April 11 2017
+//<<<<<<< HEAD
+
+
+  // this function is necessary for the LS vectorised field
+  inline int RNGfillable_general(GridBase *coarse,GridBase *fine)
+  {
+    int rngdims = coarse->_ndimension;
+    
+    // trivially extended in higher dims, with locality guaranteeing RNG state is local to node
+    int lowerdims   = fine->_ndimension - coarse->_ndimension;  assert(lowerdims >= 0);
+    // assumes that the higher dimensions are not using more processors
+    // all further divisions are local
+    for(int d=0;d<lowerdims;d++) assert(fine->_processors[d]==1);
+    for(int d=0;d<rngdims;d++) assert(coarse->_processors[d] == fine->_processors[d+lowerdims]);
+    
+
+    // then divide the number of local sites
+    // check that the total number of sims agree, meanse the iSites are the same
+    assert(fine->Nsimd() == coarse->Nsimd());
+
+    // check that the two grids divide cleanly
+    assert( (fine->lSites() / coarse->lSites() ) * coarse->lSites() == fine->lSites() );
+
+    return fine->lSites() / coarse->lSites();
+  }
+
+  /*
+  // Wrap seed_seq to give common interface with random_device
+  class fixedSeed {
+  public:
+    typedef std::seed_seq::result_type result_type;
+    std::seed_seq src;
+    
+    fixedSeed(const std::vector<int> &seeds) : src(seeds.begin(),seeds.end()) {};
+
+    result_type operator () (void){
+      std::vector<result_type> list(1);
+      src.generate(list.begin(),list.end());
+      return list[0];
+    }
+
+  };
+
+=======
+>>>>>>> develop
+  */
+  
   // real scalars are one component
   template<class scalar,class distribution,class generator> 
   void fillScalar(scalar &s,distribution &dist,generator & gen)
@@ -109,7 +158,7 @@ namespace Grid {
 #ifdef RNG_SITMO
     typedef sitmo::prng_engine 	RngEngine;
     typedef uint64_t    	RngStateType;
-    static const int    	RngStateCount = 4;
+    static const int    	RngStateCount = 13;
 #endif
 
     std::vector<RngEngine>                             _generators;
@@ -284,17 +333,19 @@ namespace Grid {
   };
 
   class GridParallelRNG : public GridRNGbase {
+
+    double _time_counter;
+
   public:
     GridBase *_grid;
-    int _vol;
-  public:
+    unsigned int _vol;
 
-    int generator_idx(int os,int is){
+    int generator_idx(int os,int is) {
       return is*_grid->oSites()+os;
     }
 
     GridParallelRNG(GridBase *grid) : GridRNGbase() {
-      _grid=grid;
+      _grid = grid;
       _vol  =_grid->iSites()*_grid->oSites();
 
       _generators.resize(_vol);
@@ -310,32 +361,33 @@ namespace Grid {
       typedef typename vobj::scalar_type scalar_type;
       typedef typename vobj::vector_type vector_type;
 
-      int multiplicity = RNGfillable(_grid,l._grid);
+      double inner_time_counter = usecond();
 
-      int     Nsimd =_grid->Nsimd();
-      int     osites=_grid->oSites();
-      int words=sizeof(scalar_object)/sizeof(scalar_type);
+      int multiplicity = RNGfillable_general(_grid, l._grid); // l has finer or same grid
+      int Nsimd  = _grid->Nsimd();  // guaranteed to be the same for l._grid too
+      int osites = _grid->oSites();  // guaranteed to be <= l._grid->oSites() by a factor multiplicity
+      int words  = sizeof(scalar_object) / sizeof(scalar_type);
 
       parallel_for(int ss=0;ss<osites;ss++){
-
         std::vector<scalar_object> buf(Nsimd);
-	for(int m=0;m<multiplicity;m++) {// Draw from same generator multiplicity times
+        for (int m = 0; m < multiplicity; m++) {  // Draw from same generator multiplicity times
 
-	  int sm=multiplicity*ss+m;      // Maps the generator site to the fine site
+          int sm = multiplicity * ss + m;  // Maps the generator site to the fine site
 
-	  for(int si=0;si<Nsimd;si++){
-	    int gdx = generator_idx(ss,si); // index of generator state
+          for (int si = 0; si < Nsimd; si++) {
+            
+            int gdx = generator_idx(ss, si);  // index of generator state
             scalar_type *pointer = (scalar_type *)&buf[si];
             dist[gdx].reset();
-	    for(int idx=0;idx<words;idx++){
-	      fillScalar(pointer[idx],dist[gdx],_generators[gdx]);
+            for (int idx = 0; idx < words; idx++) 
+              fillScalar(pointer[idx], dist[gdx], _generators[gdx]);
+          }
+          // merge into SIMD lanes, FIXME suboptimal implementation
+          merge(l._odata[sm], buf);
         }
       }
 
-	  // merge into SIMD lanes
-	  merge(l._odata[sm],buf);
-	}
-      }
+      _time_counter += usecond()- inner_time_counter;
     };
 
     void SeedFixedIntegers(const std::vector<int> &seeds){
@@ -412,6 +464,12 @@ namespace Grid {
       }
 #endif
     }
+
+    void Report(){
+      std::cout << GridLogMessage << "Time spent in the fill() routine by GridParallelRNG: "<< _time_counter/1e3 << " ms" << std::endl;
+    }
+
+
     ////////////////////////////////////////////////////////////////////////
     // Support for rigorous test of RNG's
     // Return uniform random uint32_t from requested site generator
@@ -419,7 +477,6 @@ namespace Grid {
     uint32_t GlobalU01(int gsite){
 
       uint32_t the_number;
-
       // who
       std::vector<int> gcoor;
       int rank,o_idx,i_idx;

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/log/Log.cc

@@ -30,6 +30,7 @@ directory
 *************************************************************************************/
 /*  END LEGAL */
 #include <Grid/GridCore.h>
+#include <Grid/util/CompilerCompatible.h>
 
 #include <cxxabi.h>
 #include <memory>

lib/parallelIO/IldgIO.h

@@ -0,0 +1,716 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid
+
+Source file: ./lib/parallelIO/IldgIO.h
+
+Copyright (C) 2015
+
+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_ILDG_IO_H
+#define GRID_ILDG_IO_H
+
+#ifdef HAVE_LIME
+#include <algorithm>
+#include <fstream>
+#include <iomanip>
+#include <iostream>
+#include <map>
+
+#include <pwd.h>
+#include <sys/utsname.h>
+#include <unistd.h>
+
+//C-Lime is a must have for this functionality
+extern "C" {  
+#include "lime.h"
+}
+
+namespace Grid {
+namespace QCD {
+
+  /////////////////////////////////
+  // 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"); }
+
+  /////////////////////////////////////////
+  // Encode a generic tensor as a string
+  /////////////////////////////////////////
+ template<class vobj> std::string ScidacRecordTypeString(int &colors, int &spins, int & typesize,int &datacount) { 
+
+   typedef typename getPrecision<vobj>::real_scalar_type stype;
+
+   int _ColourN       = indexRank<ColourIndex,vobj>();
+   int _ColourScalar  =  isScalar<ColourIndex,vobj>();
+   int _ColourVector  =  isVector<ColourIndex,vobj>();
+   int _ColourMatrix  =  isMatrix<ColourIndex,vobj>();
+
+   int _SpinN       = indexRank<SpinIndex,vobj>();
+   int _SpinScalar  =  isScalar<SpinIndex,vobj>();
+   int _SpinVector  =  isVector<SpinIndex,vobj>();
+   int _SpinMatrix  =  isMatrix<SpinIndex,vobj>();
+
+   int _LorentzN       = indexRank<LorentzIndex,vobj>();
+   int _LorentzScalar  =  isScalar<LorentzIndex,vobj>();
+   int _LorentzVector  =  isVector<LorentzIndex,vobj>();
+   int _LorentzMatrix  =  isMatrix<LorentzIndex,vobj>();
+
+   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:
+   ///////////////////////////////////////////////////
+   // FIXME: format for RNG? Now just binary out instead
+   ///////////////////////////////////////////////////
+
+   FILE       *File;
+   LimeReader *LimeR;
+   std::string filename;
+
+   /////////////////////////////////////////////
+   // 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 ) { 
+
+      uint64_t nbytes = limeReaderBytes(LimeR);//size of this record (configuration)
+
+      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);
+    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);
+
+    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);
+
+    ////////////////////////////////////////////////////////////////////
+    // 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));
+  }
+};
+
+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>
+  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;
+
+    uint64_t nbytes;
+
+    ////////////////////////////////////////
+    // fill the Grid header
+    ////////////////////////////////////////
+    FieldMetaData header;
+    scidacRecord  _scidacRecord;
+    scidacFile    _scidacFile;
+
+    ScidacMetaData(Umu,header,_scidacRecord,_scidacFile);
+
+    std::string format = header.floating_point;
+    header.ensemble_id    = description;
+    header.ensemble_label = description;
+    header.sequence_number = sequence;
+    header.ildg_lfn = LFN;
+
+    assert ( (format == std::string("IEEE32BIG"))  
+           ||(format == std::string("IEEE64BIG")) );
+
+    //////////////////////////////////////////////////////
+    // Fill ILDG header data struct
+    //////////////////////////////////////////////////////
+    ildgFormat ildgfmt ;
+    ildgfmt.field     = std::string("su3gauge");
+
+    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;
+
+    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);
+  }
+};
+
+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
+
+#endif

lib/parallelIO/IldgIOtypes.h

@@ -0,0 +1,231 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid
+
+Source file: ./lib/parallelIO/IldgIO.h
+
+Copyright (C) 2015
+
+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_ILDGTYPES_IO_H
+#define GRID_ILDGTYPES_IO_H
+
+#ifdef HAVE_LIME
+extern "C" { // for linkage
+#include "lime.h"
+}
+
+namespace Grid {
+
+/////////////////////////////////////////////////////////////////////////////////
+// Data representation of records that enter ILDG and SciDac formats
+/////////////////////////////////////////////////////////////////////////////////
+
+#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"
+
+  ////////////////////////////////////////////////////////////
+  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:
+  GRID_SERIALIZABLE_CLASS_MEMBERS(scidacFile,
+                                  double, version,
+                                  int, spacetime,
+				  std::string, dims, // must convert to int
+                                  int, volfmt);
+
+  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

@@ -1,4 +1,4 @@
-    /*************************************************************************************
+/*************************************************************************************
 
     Grid physics library, www.github.com/paboyle/Grid 
 
@@ -6,9 +6,9 @@
 
     Copyright (C) 2015
 
-Author: Matt Spraggs <matthew.spraggs@gmail.com>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: paboyle <paboyle@ph.ed.ac.uk>
+    Author: Matt Spraggs <matthew.spraggs@gmail.com>
+    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
@@ -25,232 +25,38 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
     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 */
+*************************************************************************************/
+/*  END LEGAL */
 #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 {
+  namespace QCD {
 
-using namespace Grid;
+    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<4;mu++){
-      for(int i=0;i<3;i++){
-      for(int j=0;j<3;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
-////////////////////////////////////////////////////////////////////////////////
-class NerscIO : public BinaryIO { 
+    ////////////////////////////////////////////////////////////////////////////////
+    // Write and read from fstream; comput header offset for payload
+    ////////////////////////////////////////////////////////////////////////////////
+    class NerscIO : public BinaryIO { 
     public:
 
       static inline void truncate(std::string file){
 	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)
-{
+      // for the header-reader
+      static inline int readHeader(std::string file,GridBase *grid,  FieldMetaData &field)
+      {
       int offset=0;
       std::map<std::string,std::string> header;
       std::string line;
@@ -319,23 +125,23 @@ static inline int readHeader(std::string file,GridBase *grid,  NerscField &field
       field.floating_point   = header["FLOATING_POINT"];
 
       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);
 
@@ -344,148 +150,134 @@ static inline void readConfiguration(Lattice<iLorentzColourMatrix<vsimd> > &Umu,
       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
+	  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>, LorentzColour2x3D> 
-      	(Umu,file,Nersc3x2munger<LorentzColour2x3D,LorentzColourMatrix>(),offset,format);
-#else 
-      csum=BinaryIO::readObjectSerial<iLorentzColourMatrix<vsimd>, LorentzColour2x3D> 
-      	(Umu,file,Nersc3x2munger<LorentzColour2x3D,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 ) {
-#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>,LorentzColourMatrixF>
+	    (Umu,file,GaugeSimpleMunger<LorentzColourMatrixF,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>,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)
-{
+      template<class vsimd>
+      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";
 
 	typedef LorentzColourMatrixD fobj3D;
 	typedef LorentzColour2x3D    fobj2D;
-  typedef LorentzColourMatrixF fobj3f;
-  typedef LorentzColour2x3F    fobj2f;
   
 	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 ) { 
-
-    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 { 
+	// Sod it -- always write 3x3 double
 	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);
+	GaugeSimpleUnmunger<fobj3D,sobj> munge;
 	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
-  }
 
-  std::cout<<GridLogMessage <<"Written NERSC Configuration "<<file<< " checksum "<<std::hex<<csum<< std::dec<<" plaq "<< header.plaquette <<std::endl;
+	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<<header.checksum
+		 <<std::dec<<" plaq "<< header.plaquette <<std::endl;
 
       }
-
-
       ///////////////////////////////
       // RNG state
       ///////////////////////////////
-static inline void writeRNGState(GridSerialRNG &serial,GridParallelRNG &parallel,std::string file)
-{
+      static inline void writeRNGState(GridSerialRNG &serial,GridParallelRNG &parallel,std::string file)
+      {
 	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
@@ -503,23 +295,27 @@ static inline void writeRNGState(GridSerialRNG &serial,GridParallelRNG &parallel
 
 	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;
 
 	GridBase *grid = parallel._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);
@@ -539,15 +335,19 @@ static inline void readRNGState(GridSerialRNG &serial,GridParallelRNG & parallel
 
 	// 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/perfmon/PerfCount.h

@@ -205,13 +205,14 @@ public:
   void Stop(void) {
     count=0;
     cycles=0;
-    size_t ign;
 #ifdef __linux__
+    ssize_t ign;
     if ( fd!= -1) {
       ::ioctl(fd, PERF_EVENT_IOC_DISABLE, 0);
       ::ioctl(cyclefd, PERF_EVENT_IOC_DISABLE, 0);
       ign=::read(fd, &count, sizeof(long long));
-      ign=::read(cyclefd, &cycles, sizeof(long long));
+      ign+=::read(cyclefd, &cycles, sizeof(long long));
+      assert(ign=2*sizeof(long long));
     }
     elapsed = cyclecount() - begin;
 #else

lib/qcd/LatticeTheories.h

@@ -0,0 +1,124 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid
+
+Source file: ./lib/qcd/QCD.h
+
+Copyright (C) 2015
+
+Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
+Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
+Author: neo <cossu@post.kek.jp>
+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_LT_H
+#define GRID_LT_H
+namespace Grid{
+
+// First steps in the complete generalization of the Physics part
+// Design not final
+namespace LatticeTheories {
+
+template <int Dimensions>
+struct LatticeTheory {
+  static const int Nd = Dimensions;
+  static const int Nds = Dimensions * 2;  // double stored field
+  template <typename vtype>
+  using iSinglet = iScalar<iScalar<iScalar<vtype> > >;
+};
+
+template <int Dimensions, int Colours>
+struct LatticeGaugeTheory : public LatticeTheory<Dimensions> {
+  static const int Nds = Dimensions * 2;
+  static const int Nd = Dimensions;
+  static const int Nc = Colours;
+
+  template <typename vtype> 
+  using iColourMatrix = iScalar<iScalar<iMatrix<vtype, Nc> > >;
+  template <typename vtype>
+  using iLorentzColourMatrix = iVector<iScalar<iMatrix<vtype, Nc> >, Nd>;
+  template <typename vtype>
+  using iDoubleStoredColourMatrix = iVector<iScalar<iMatrix<vtype, Nc> >, Nds>;
+  template <typename vtype>
+  using iColourVector = iScalar<iScalar<iVector<vtype, Nc> > >;
+};
+
+template <int Dimensions, int Colours, int Spin>
+struct FermionicLatticeGaugeTheory
+    : public LatticeGaugeTheory<Dimensions, Colours> {
+  static const int Nd = Dimensions;
+  static const int Nds = Dimensions * 2;
+  static const int Nc = Colours;
+  static const int Ns = Spin;
+
+  template <typename vtype>
+  using iSpinMatrix = iScalar<iMatrix<iScalar<vtype>, Ns> >;
+  template <typename vtype>
+  using iSpinColourMatrix = iScalar<iMatrix<iMatrix<vtype, Nc>, Ns> >;
+  template <typename vtype>
+  using iSpinVector = iScalar<iVector<iScalar<vtype>, Ns> >;
+  template <typename vtype>
+  using iSpinColourVector = iScalar<iVector<iVector<vtype, Nc>, Ns> >;
+  // These 2 only if Spin is a multiple of 2
+  static const int Nhs = Spin / 2;
+  template <typename vtype>
+  using iHalfSpinVector = iScalar<iVector<iScalar<vtype>, Nhs> >;
+  template <typename vtype>
+  using iHalfSpinColourVector = iScalar<iVector<iVector<vtype, Nc>, Nhs> >;
+
+  //tests
+  typedef iColourMatrix<Complex> ColourMatrix;
+  typedef iColourMatrix<ComplexF> ColourMatrixF;
+  typedef iColourMatrix<ComplexD> ColourMatrixD;
+
+
+};
+
+// Examples, not complete now.
+struct QCD : public FermionicLatticeGaugeTheory<4, 3, 4> {
+    static const int Xp = 0;
+    static const int Yp = 1;
+    static const int Zp = 2;
+    static const int Tp = 3;
+    static const int Xm = 4;
+    static const int Ym = 5;
+    static const int Zm = 6;
+    static const int Tm = 7;
+
+    typedef FermionicLatticeGaugeTheory FLGT;
+
+    typedef FLGT::iSpinMatrix<Complex  >          SpinMatrix;
+    typedef FLGT::iSpinMatrix<ComplexF >          SpinMatrixF;
+    typedef FLGT::iSpinMatrix<ComplexD >          SpinMatrixD;
+
+};
+struct QED : public FermionicLatticeGaugeTheory<4, 1, 4> {//fill
+};
+
+template <int Dimensions>
+struct Scalar : public LatticeTheory<Dimensions> {};
+
+};  // LatticeTheories
+
+} // Grid
+
+#endif

lib/qcd/QCD.h

@@ -32,9 +32,12 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 #ifndef GRID_QCD_BASE_H
 #define GRID_QCD_BASE_H
 namespace Grid{
-
 namespace QCD {
 
+    static const int Xdir = 0;
+    static const int Ydir = 1;
+    static const int Zdir = 2;
+    static const int Tdir = 3;
 
     static const int Xp = 0;
     static const int Yp = 1;
@@ -492,6 +495,38 @@ namespace QCD {
 }   //namespace QCD
 } // Grid
 
+/*
+<<<<<<< HEAD
+#include <Grid/qcd/utils/SpaceTimeGrid.h>
+#include <Grid/qcd/spin/Dirac.h>
+#include <Grid/qcd/spin/TwoSpinor.h>
+#include <Grid/qcd/utils/LinalgUtils.h>
+#include <Grid/qcd/utils/CovariantCshift.h>
+
+// Include representations  
+#include <Grid/qcd/utils/SUn.h>
+#include <Grid/qcd/utils/SUnAdjoint.h>
+#include <Grid/qcd/utils/SUnTwoIndex.h>
+#include <Grid/qcd/representations/hmc_types.h>
+
+// Scalar field
+#include <Grid/qcd/utils/ScalarObjs.h>
+
+#include <Grid/qcd/action/Actions.h>
+
+#include <Grid/qcd/smearing/Smearing.h>
+
+#include <Grid/qcd/hmc/integrators/Integrator.h>
+#include <Grid/qcd/hmc/integrators/Integrator_algorithm.h>
+#include <Grid/qcd/observables/hmc_observable.h>
+#include <Grid/qcd/hmc/HMC.h>
+
+
+//#include <Grid/qcd/modules/mods.h>
+=======
+
+>>>>>>> develop
+*/
 
 
 #endif

lib/qcd/action/ActionBase.h

@@ -4,10 +4,11 @@ Grid physics library, www.github.com/paboyle/Grid
 
 Source file: ./lib/qcd/action/ActionBase.h
 
-Copyright (C) 2015
+Copyright (C) 2015-2016
 
 Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 Author: neo <cossu@post.kek.jp>
+Author: Guido Cossu <guido.cossu@ed.ac.uk>
 
 This program is free software; you can redistribute it and/or modify
 it under the terms of the GNU General Public License as published by
@@ -27,128 +28,29 @@ See the full license in the file "LICENSE" in the top level distribution
 directory
 *************************************************************************************/
 /*  END LEGAL */
-#ifndef QCD_ACTION_BASE
-#define QCD_ACTION_BASE
+
+#ifndef ACTION_BASE_H
+#define ACTION_BASE_H
+
 namespace Grid {
 namespace QCD {
 
-template <class GaugeField>
-class Action {
+template <class GaugeField >
+class Action 
+{
+
  public:
   bool is_smeared = false;
-  // Boundary conditions? // Heatbath?
-  virtual void refresh(const GaugeField& U,
-                       GridParallelRNG& pRNG) = 0;  // refresh pseudofermions
+  // Heatbath?
+  virtual void refresh(const GaugeField& U, GridParallelRNG& pRNG) = 0; // refresh pseudofermions
   virtual RealD S(const GaugeField& U) = 0;                             // evaluate the action
-  virtual void deriv(const GaugeField& U,
-                     GaugeField& dSdU) = 0;  // evaluate the action derivative
-  virtual ~Action(){};
+  virtual void deriv(const GaugeField& U, GaugeField& dSdU) = 0;        // evaluate the action derivative
+  virtual std::string action_name()    = 0;                             // return the action name
+  virtual std::string LogParameters()  = 0;                             // prints action parameters
+  virtual ~Action(){}
 };
 
-// Indexing of tuple types
-template <class T, class Tuple>
-struct Index;
-
-template <class T, class... Types>
-struct Index<T, std::tuple<T, Types...>> {
-  static const std::size_t value = 0;
-};
-
-template <class T, class U, class... Types>
-struct Index<T, std::tuple<U, Types...>> {
-  static const std::size_t value = 1 + Index<T, std::tuple<Types...>>::value;
-};
-
-/*
-template <class GaugeField>
-struct ActionLevel {
- public:
-  typedef Action<GaugeField>*
-      ActPtr;  // now force the same colours as the rest of the code
-
-  //Add supported representations here
-
-
-  unsigned int multiplier;
-
-  std::vector<ActPtr> actions;
-
-  ActionLevel(unsigned int mul = 1) : actions(0), multiplier(mul) {
-    assert(mul >= 1);
-  };
-
-  void push_back(ActPtr ptr) { actions.push_back(ptr); }
-};
-*/
-
-template <class GaugeField, class Repr = NoHirep >
-struct ActionLevel {
- public:
-  unsigned int multiplier; 
-
-  // Fundamental repr actions separated because of the smearing
-  typedef Action<GaugeField>* ActPtr;
-
-  // construct a tuple of vectors of the actions for the corresponding higher
-  // representation fields
-  typedef typename AccessTypes<Action, Repr>::VectorCollection action_collection;
-  action_collection actions_hirep;
-  typedef typename  AccessTypes<Action, Repr>::FieldTypeCollection action_hirep_types;
-
-  std::vector<ActPtr>& actions;
-
-  // Temporary conversion between ActionLevel and ActionLevelHirep
-  //ActionLevelHirep(ActionLevel<GaugeField>& AL ):actions(AL.actions), multiplier(AL.multiplier){}
-
-  ActionLevel(unsigned int mul = 1) : actions(std::get<0>(actions_hirep)), multiplier(mul) {
-    // initialize the hirep vectors to zero.
-    //apply(this->resize, actions_hirep, 0); //need a working resize
-    assert(mul >= 1);
-  };
-
-  //void push_back(ActPtr ptr) { actions.push_back(ptr); }
-
-
-
-  template < class Field >
-  void push_back(Action<Field>* ptr) {
-    // insert only in the correct vector
-    std::get< Index < Field, action_hirep_types>::value >(actions_hirep).push_back(ptr);
-  };
-
- 
-
-  template < class ActPtr>
-  static void resize(ActPtr ap, unsigned int n){
-    ap->resize(n);
-
-  }
-
-  //template <std::size_t I>
-  //auto getRepresentation(Repr& R)->decltype(std::get<I>(R).U)  {return std::get<I>(R).U;}
-
-  // Loop on tuple for a callable function
-  template <std::size_t I = 1, typename Callable, typename ...Args>
-  inline typename std::enable_if<I == std::tuple_size<action_collection>::value, void>::type apply(
-      Callable, Repr& R,Args&...) const {}
-
-  template <std::size_t I = 1, typename Callable, typename ...Args>
-  inline typename std::enable_if<I < std::tuple_size<action_collection>::value, void>::type apply(
-      Callable fn, Repr& R, Args&... arguments) const {
-    fn(std::get<I>(actions_hirep), std::get<I>(R.rep), arguments...);
-    apply<I + 1>(fn, R, arguments...);
-  }  
-
-};
-
-
-//template <class GaugeField>
-//using ActionSet = std::vector<ActionLevel<GaugeField> >;
-
-template <class GaugeField, class R>
-using ActionSet = std::vector<ActionLevel<GaugeField, R> >;
-
 }
 }
 
-#endif
+#endif // ACTION_BASE_H

lib/qcd/action/ActionCore.h

@@ -31,15 +31,31 @@ directory
 #define QCD_ACTION_CORE
 
 #include <Grid/qcd/action/ActionBase.h>
+#include <Grid/qcd/action/ActionSet.h>
 #include <Grid/qcd/action/ActionParams.h>
 
 ////////////////////////////////////////////
 // Gauge Actions
 ////////////////////////////////////////////
 #include <Grid/qcd/action/gauge/Gauge.h>
+
 ////////////////////////////////////////////
 // Fermion prereqs
 ////////////////////////////////////////////
 #include <Grid/qcd/action/fermion/FermionCore.h>
 
+////////////////////////////////////////////
+// Scalar Actions
+////////////////////////////////////////////
+#include <Grid/qcd/action/scalar/Scalar.h>
+
+////////////////////////////////////////////
+// Utility functions
+////////////////////////////////////////////
+#include <Grid/qcd/utils/Metric.h>
+#include <Grid/qcd/utils/CovariantLaplacian.h>
+
+
+
+
 #endif

lib/qcd/action/ActionParams.h

@@ -1,31 +1,34 @@
-    /*************************************************************************************
+/*************************************************************************************
 
-    Grid physics library, www.github.com/paboyle/Grid 
+Grid physics library, www.github.com/paboyle/Grid
 
-    Source file: ./lib/qcd/action/ActionParams.h
+Source file: ./lib/qcd/action/ActionParams.h
 
-    Copyright (C) 2015
+Copyright (C) 2015
 
 Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 Author: paboyle <paboyle@ph.ed.ac.uk>
+Author: Guido Cossu <guido.cossu@ed.ac.uk>
 
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
+This program is 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.
+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.
+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 */
 
-    See the full license in the file "LICENSE" in the top level distribution directory
-    *************************************************************************************/
-    /*  END LEGAL */
 #ifndef GRID_QCD_ACTION_PARAMS_H
 #define GRID_QCD_ACTION_PARAMS_H
 
@@ -36,32 +39,54 @@ namespace QCD {
   struct GparityWilsonImplParams {
     bool overlapCommsCompute;
     std::vector<int> twists;
-      GparityWilsonImplParams () : twists(Nd,0), overlapCommsCompute(false) {};
-
+    GparityWilsonImplParams() : twists(Nd, 0), overlapCommsCompute(false){};
   };
   
   struct WilsonImplParams {
     bool overlapCommsCompute;
-      WilsonImplParams() : overlapCommsCompute(false) {};
+    std::vector<Complex> boundary_phases;
+    WilsonImplParams() : overlapCommsCompute(false) {
+      boundary_phases.resize(Nd, 1.0);
+    };
+    WilsonImplParams(const std::vector<Complex> phi)
+      : boundary_phases(phi), overlapCommsCompute(false) {}
   };
 
   struct StaggeredImplParams {
     StaggeredImplParams()  {};
   };
   
-    struct OneFlavourRationalParams { 
-      RealD  lo;
-      RealD  hi;
-      int MaxIter;   // Vector?
-      RealD tolerance; // Vector? 
-      int    degree=10;
-      int precision=64;
+  struct OneFlavourRationalParams : Serializable {
+    GRID_SERIALIZABLE_CLASS_MEMBERS(OneFlavourRationalParams, 
+				    RealD, lo, 
+				    RealD, hi, 
+				    int,   MaxIter, 
+				    RealD, tolerance, 
+				    int,   degree, 
+				    int,   precision);
     
-      OneFlavourRationalParams (RealD _lo,RealD _hi,int _maxit,RealD tol=1.0e-8,int _degree = 10,int _precision=64) :
-        lo(_lo), hi(_hi), MaxIter(_maxit), tolerance(tol), degree(_degree), precision(_precision)
-      {};
+    // MaxIter and tolerance, vectors??
+    
+    // constructor 
+    OneFlavourRationalParams(	RealD _lo      = 0.0, 
+				RealD _hi      = 1.0, 
+				int _maxit     = 1000,
+				RealD tol      = 1.0e-8, 
+                           	int _degree    = 10,
+				int _precision = 64)
+      : lo(_lo),
+	hi(_hi),
+	MaxIter(_maxit),
+	tolerance(tol),
+	degree(_degree),
+	precision(_precision){};
   };
   
-}}
+  
+}
+}
+
+
+
 
 #endif

lib/qcd/action/ActionSet.h

@@ -0,0 +1,116 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid
+
+Source file: ./lib/qcd/action/ActionSet.h
+
+Copyright (C) 2015
+
+Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+Author: neo <cossu@post.kek.jp>
+
+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 ACTION_SET_H
+#define ACTION_SET_H
+
+namespace Grid {
+
+// Should drop this namespace here
+namespace QCD {
+
+//////////////////////////////////
+// Indexing of tuple types
+//////////////////////////////////
+
+template <class T, class Tuple>
+struct Index;
+
+template <class T, class... Types>
+struct Index<T, std::tuple<T, Types...>> {
+  static const std::size_t value = 0;
+};
+
+template <class T, class U, class... Types>
+struct Index<T, std::tuple<U, Types...>> {
+  static const std::size_t value = 1 + Index<T, std::tuple<Types...>>::value;
+};
+
+
+////////////////////////////////////////////
+// Action Level
+// Action collection 
+// in a integration level
+// (for multilevel integration schemes)
+////////////////////////////////////////////
+
+template <class Field, class Repr = NoHirep >
+struct ActionLevel {
+ public:
+  unsigned int multiplier;
+
+  // Fundamental repr actions separated because of the smearing
+  typedef Action<Field>* ActPtr;
+
+  // construct a tuple of vectors of the actions for the corresponding higher
+  // representation fields
+  typedef typename AccessTypes<Action, Repr>::VectorCollection action_collection;
+  typedef typename  AccessTypes<Action, Repr>::FieldTypeCollection action_hirep_types;
+
+  action_collection actions_hirep;
+  std::vector<ActPtr>& actions;
+
+  explicit ActionLevel(unsigned int mul = 1) : 
+  actions(std::get<0>(actions_hirep)), multiplier(mul) {
+    // initialize the hirep vectors to zero.
+    // apply(this->resize, actions_hirep, 0); //need a working resize
+    assert(mul >= 1);
+  }
+
+  template < class GenField >
+  void push_back(Action<GenField>* ptr) {
+    // insert only in the correct vector
+    std::get< Index < GenField, action_hirep_types>::value >(actions_hirep).push_back(ptr);
+  };
+
+  template <class ActPtr>
+  static void resize(ActPtr ap, unsigned int n) {
+    ap->resize(n);
+  }
+
+  // Loop on tuple for a callable function
+  template <std::size_t I = 1, typename Callable, typename ...Args>
+  inline typename std::enable_if<I == std::tuple_size<action_collection>::value, void>::type apply(Callable, Repr& R,Args&...) const {}
+
+  template <std::size_t I = 1, typename Callable, typename ...Args>
+  inline typename std::enable_if<I < std::tuple_size<action_collection>::value, void>::type apply(Callable fn, Repr& R, Args&... arguments) const {
+    fn(std::get<I>(actions_hirep), std::get<I>(R.rep), arguments...);
+    apply<I + 1>(fn, R, arguments...);
+  }  
+
+};
+
+// Define the ActionSet
+template <class GaugeField, class R>
+using ActionSet = std::vector<ActionLevel<GaugeField, R> >;
+
+} // QCD
+} // Grid
+
+#endif  // ACTION_SET_H

lib/qcd/action/fermion/CayleyFermion5D.cc

@@ -29,7 +29,7 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
     *************************************************************************************/
     /*  END LEGAL */
 
-#include <Grid/Eigen/Dense>
+#include <Grid/Grid_Eigen_Dense.h>
 #include <Grid/qcd/action/fermion/FermionCore.h>
 #include <Grid/qcd/action/fermion/CayleyFermion5D.h>
 
@@ -320,7 +320,7 @@ void CayleyFermion5D<Impl>::MDeriv  (GaugeField &mat,const FermionField &U,const
     this->DhopDeriv(mat,U,Din,dag);
   } else {
     //      U d/du [D_w D5]^dag V = U D5^dag d/du DW^dag Y // implicit adj on U in call
-    MeooeDag5D(U,Din);
+    Meooe5D(U,Din);
     this->DhopDeriv(mat,Din,V,dag);
   }
 };
@@ -335,7 +335,7 @@ void CayleyFermion5D<Impl>::MoeDeriv(GaugeField &mat,const FermionField &U,const
     this->DhopDerivOE(mat,U,Din,dag);
   } else {
     //      U d/du [D_w D5]^dag V = U D5^dag d/du DW^dag Y // implicit adj on U in call
-      MeooeDag5D(U,Din);
+    Meooe5D(U,Din);
     this->DhopDerivOE(mat,Din,V,dag);
   }
 };
@@ -350,7 +350,7 @@ void CayleyFermion5D<Impl>::MeoDeriv(GaugeField &mat,const FermionField &U,const
     this->DhopDerivEO(mat,U,Din,dag);
   } else {
     //      U d/du [D_w D5]^dag V = U D5^dag d/du DW^dag Y // implicit adj on U in call
-    MeooeDag5D(U,Din);
+    Meooe5D(U,Din);
     this->DhopDerivEO(mat,Din,V,dag);
   }
 };
@@ -380,6 +380,8 @@ void CayleyFermion5D<Impl>::SetCoefficientsInternal(RealD zolo_hi,std::vector<Co
   ///////////////////////////////////////////////////////////
   // The Cayley coeffs (unprec)
   ///////////////////////////////////////////////////////////
+  assert(gamma.size()==Ls);
+
   omega.resize(Ls);
   bs.resize(Ls);
   cs.resize(Ls);
@@ -412,6 +414,7 @@ void CayleyFermion5D<Impl>::SetCoefficientsInternal(RealD zolo_hi,std::vector<Co
   for(int i=0; i < Ls; i++){
     as[i] = 1.0;
     omega[i] = gamma[i]*zolo_hi; //NB reciprocal relative to Chroma NEF code
+    //    assert(fabs(omega[i])>0.0);
     bs[i] = 0.5*(bpc/omega[i] + bmc);
     cs[i] = 0.5*(bpc/omega[i] - bmc);
   }
@@ -426,11 +429,11 @@ void CayleyFermion5D<Impl>::SetCoefficientsInternal(RealD zolo_hi,std::vector<Co
   
   for(int i=0;i<Ls;i++){
     bee[i]=as[i]*(bs[i]*(4.0-this->M5) +1.0);     
+    //    assert(fabs(bee[i])>0.0);
     cee[i]=as[i]*(1.0-cs[i]*(4.0-this->M5));
     beo[i]=as[i]*bs[i];
     ceo[i]=-as[i]*cs[i];
   }
-  
   aee.resize(Ls);
   aeo.resize(Ls);
   for(int i=0;i<Ls;i++){
@@ -474,14 +477,16 @@ void CayleyFermion5D<Impl>::SetCoefficientsInternal(RealD zolo_hi,std::vector<Co
 	
   { 
     Coeff_t delta_d=mass*cee[Ls-1];
-    for(int j=0;j<Ls-1;j++) delta_d *= cee[j]/bee[j];
+    for(int j=0;j<Ls-1;j++) {
+      //      assert(fabs(bee[j])>0.0);
+      delta_d *= cee[j]/bee[j];
+    }
     dee[Ls-1] += delta_d;
   }  
 
   int inv=1;
   this->MooeeInternalCompute(0,inv,MatpInv,MatmInv);
   this->MooeeInternalCompute(1,inv,MatpInvDag,MatmInvDag);
-
 }
 
 
@@ -495,7 +500,9 @@ void CayleyFermion5D<Impl>::MooeeInternalCompute(int dag, int inv,
   GridBase *grid = this->FermionRedBlackGrid();
   int LLs = grid->_rdimensions[0];
 
-  if ( LLs == Ls ) return; // Not vectorised in 5th direction
+  if ( LLs == Ls ) {
+    return; // Not vectorised in 5th direction
+  }
 
   Eigen::MatrixXcd Pplus  = Eigen::MatrixXcd::Zero(Ls,Ls);
   Eigen::MatrixXcd Pminus = Eigen::MatrixXcd::Zero(Ls,Ls);

lib/qcd/action/fermion/CayleyFermion5Dcache.cc

@@ -237,6 +237,13 @@ void CayleyFermion5D<Impl>::MooeeInvDag (const FermionField &psi, FermionField &
   INSTANTIATE_DPERP(GparityWilsonImplD);
   INSTANTIATE_DPERP(ZWilsonImplF);
   INSTANTIATE_DPERP(ZWilsonImplD);
+
+  INSTANTIATE_DPERP(WilsonImplFH);
+  INSTANTIATE_DPERP(WilsonImplDF);
+  INSTANTIATE_DPERP(GparityWilsonImplFH);
+  INSTANTIATE_DPERP(GparityWilsonImplDF);
+  INSTANTIATE_DPERP(ZWilsonImplFH);
+  INSTANTIATE_DPERP(ZWilsonImplDF);
 #endif
 
 }}

lib/qcd/action/fermion/CayleyFermion5Ddense.cc

@@ -29,7 +29,7 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
     *************************************************************************************/
     /*  END LEGAL */
 
-#include <Grid/Eigen/Dense>
+#include <Grid/Grid_Eigen_Dense.h>
 #include <Grid/qcd/action/fermion/FermionCore.h>
 #include <Grid/qcd/action/fermion/CayleyFermion5D.h>
 
@@ -137,6 +137,20 @@ template void CayleyFermion5D<WilsonImplF>::MooeeInternal(const FermionField &ps
 template void CayleyFermion5D<WilsonImplD>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
 template void CayleyFermion5D<ZWilsonImplF>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
 template void CayleyFermion5D<ZWilsonImplD>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
+
+INSTANTIATE_DPERP(GparityWilsonImplFH);
+INSTANTIATE_DPERP(GparityWilsonImplDF);
+INSTANTIATE_DPERP(WilsonImplFH);
+INSTANTIATE_DPERP(WilsonImplDF);
+INSTANTIATE_DPERP(ZWilsonImplFH);
+INSTANTIATE_DPERP(ZWilsonImplDF);
+
+template void CayleyFermion5D<GparityWilsonImplFH>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
+template void CayleyFermion5D<GparityWilsonImplDF>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
+template void CayleyFermion5D<WilsonImplFH>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
+template void CayleyFermion5D<WilsonImplDF>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
+template void CayleyFermion5D<ZWilsonImplFH>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
+template void CayleyFermion5D<ZWilsonImplDF>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
 #endif
 
 }}

lib/qcd/action/fermion/CayleyFermion5Dssp.cc

@@ -37,7 +37,6 @@ namespace Grid {
 namespace QCD {
 
   // FIXME -- make a version of these routines with site loop outermost for cache reuse.
-
   // Pminus fowards
   // Pplus  backwards
 template<class Impl>  
@@ -152,6 +151,13 @@ void CayleyFermion5D<Impl>::MooeeInvDag (const FermionField &psi, FermionField &
   INSTANTIATE_DPERP(GparityWilsonImplD);
   INSTANTIATE_DPERP(ZWilsonImplF);
   INSTANTIATE_DPERP(ZWilsonImplD);
+
+  INSTANTIATE_DPERP(WilsonImplFH);
+  INSTANTIATE_DPERP(WilsonImplDF);
+  INSTANTIATE_DPERP(GparityWilsonImplFH);
+  INSTANTIATE_DPERP(GparityWilsonImplDF);
+  INSTANTIATE_DPERP(ZWilsonImplFH);
+  INSTANTIATE_DPERP(ZWilsonImplDF);
 #endif
 
 }

lib/qcd/action/fermion/CayleyFermion5Dvec.cc

@@ -808,10 +808,21 @@ INSTANTIATE_DPERP(DomainWallVec5dImplF);
 INSTANTIATE_DPERP(ZDomainWallVec5dImplD);
 INSTANTIATE_DPERP(ZDomainWallVec5dImplF);
 
+INSTANTIATE_DPERP(DomainWallVec5dImplDF);
+INSTANTIATE_DPERP(DomainWallVec5dImplFH);
+INSTANTIATE_DPERP(ZDomainWallVec5dImplDF);
+INSTANTIATE_DPERP(ZDomainWallVec5dImplFH);
+
 template void CayleyFermion5D<DomainWallVec5dImplF>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
 template void CayleyFermion5D<DomainWallVec5dImplD>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
 template void CayleyFermion5D<ZDomainWallVec5dImplF>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
 template void CayleyFermion5D<ZDomainWallVec5dImplD>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
 
+template void CayleyFermion5D<DomainWallVec5dImplFH>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
+template void CayleyFermion5D<DomainWallVec5dImplDF>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
+template void CayleyFermion5D<ZDomainWallVec5dImplFH>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
+template void CayleyFermion5D<ZDomainWallVec5dImplDF>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
+
+
 
 }}

lib/qcd/action/fermion/DomainWallFermion.h

@@ -68,7 +68,7 @@ namespace Grid {
 	Approx::zolotarev_data *zdata = Approx::higham(eps,this->Ls);// eps is ignored for higham
 	assert(zdata->n==this->Ls);
 	
-	//	std::cout<<GridLogMessage << "DomainWallFermion with Ls="<<this->Ls<<std::endl;
+	std::cout<<GridLogMessage << "DomainWallFermion with Ls="<<this->Ls<<std::endl;
 	// Call base setter
 	this->SetCoefficientsTanh(zdata,1.0,0.0);
 

lib/qcd/action/fermion/Fermion.h

@@ -1,4 +1,4 @@
-    /*************************************************************************************
+/*************************************************************************************
 
     Grid physics library, www.github.com/paboyle/Grid 
 
@@ -89,6 +89,10 @@ typedef WilsonFermion<WilsonImplR> WilsonFermionR;
 typedef WilsonFermion<WilsonImplF> WilsonFermionF;
 typedef WilsonFermion<WilsonImplD> WilsonFermionD;
 
+typedef WilsonFermion<WilsonImplRL> WilsonFermionRL;
+typedef WilsonFermion<WilsonImplFH> WilsonFermionFH;
+typedef WilsonFermion<WilsonImplDF> WilsonFermionDF;
+
 typedef WilsonFermion<WilsonAdjImplR> WilsonAdjFermionR;
 typedef WilsonFermion<WilsonAdjImplF> WilsonAdjFermionF;
 typedef WilsonFermion<WilsonAdjImplD> WilsonAdjFermionD;
@@ -105,27 +109,50 @@ typedef DomainWallFermion<WilsonImplR> DomainWallFermionR;
 typedef DomainWallFermion<WilsonImplF> DomainWallFermionF;
 typedef DomainWallFermion<WilsonImplD> DomainWallFermionD;
 
+typedef DomainWallFermion<WilsonImplRL> DomainWallFermionRL;
+typedef DomainWallFermion<WilsonImplFH> DomainWallFermionFH;
+typedef DomainWallFermion<WilsonImplDF> DomainWallFermionDF;
+
 typedef MobiusFermion<WilsonImplR> MobiusFermionR;
 typedef MobiusFermion<WilsonImplF> MobiusFermionF;
 typedef MobiusFermion<WilsonImplD> MobiusFermionD;
 
+typedef MobiusFermion<WilsonImplRL> MobiusFermionRL;
+typedef MobiusFermion<WilsonImplFH> MobiusFermionFH;
+typedef MobiusFermion<WilsonImplDF> MobiusFermionDF;
+
 typedef ZMobiusFermion<ZWilsonImplR> ZMobiusFermionR;
 typedef ZMobiusFermion<ZWilsonImplF> ZMobiusFermionF;
 typedef ZMobiusFermion<ZWilsonImplD> ZMobiusFermionD;
 
+typedef ZMobiusFermion<ZWilsonImplRL> ZMobiusFermionRL;
+typedef ZMobiusFermion<ZWilsonImplFH> ZMobiusFermionFH;
+typedef ZMobiusFermion<ZWilsonImplDF> ZMobiusFermionDF;
+
 // Ls vectorised 
 typedef DomainWallFermion<DomainWallVec5dImplR> DomainWallFermionVec5dR;
 typedef DomainWallFermion<DomainWallVec5dImplF> DomainWallFermionVec5dF;
 typedef DomainWallFermion<DomainWallVec5dImplD> DomainWallFermionVec5dD;
 
+typedef DomainWallFermion<DomainWallVec5dImplRL> DomainWallFermionVec5dRL;
+typedef DomainWallFermion<DomainWallVec5dImplFH> DomainWallFermionVec5dFH;
+typedef DomainWallFermion<DomainWallVec5dImplDF> DomainWallFermionVec5dDF;
+
 typedef MobiusFermion<DomainWallVec5dImplR> MobiusFermionVec5dR;
 typedef MobiusFermion<DomainWallVec5dImplF> MobiusFermionVec5dF;
 typedef MobiusFermion<DomainWallVec5dImplD> MobiusFermionVec5dD;
 
+typedef MobiusFermion<DomainWallVec5dImplRL> MobiusFermionVec5dRL;
+typedef MobiusFermion<DomainWallVec5dImplFH> MobiusFermionVec5dFH;
+typedef MobiusFermion<DomainWallVec5dImplDF> MobiusFermionVec5dDF;
+
 typedef ZMobiusFermion<ZDomainWallVec5dImplR> ZMobiusFermionVec5dR;
 typedef ZMobiusFermion<ZDomainWallVec5dImplF> ZMobiusFermionVec5dF;
 typedef ZMobiusFermion<ZDomainWallVec5dImplD> ZMobiusFermionVec5dD;
 
+typedef ZMobiusFermion<ZDomainWallVec5dImplRL> ZMobiusFermionVec5dRL;
+typedef ZMobiusFermion<ZDomainWallVec5dImplFH> ZMobiusFermionVec5dFH;
+typedef ZMobiusFermion<ZDomainWallVec5dImplDF> ZMobiusFermionVec5dDF;
 
 typedef ScaledShamirFermion<WilsonImplR> ScaledShamirFermionR;
 typedef ScaledShamirFermion<WilsonImplF> ScaledShamirFermionF;
@@ -166,17 +193,35 @@ typedef OverlapWilsonPartialFractionZolotarevFermion<WilsonImplD> OverlapWilsonP
 typedef WilsonFermion<GparityWilsonImplR>     GparityWilsonFermionR;
 typedef WilsonFermion<GparityWilsonImplF>     GparityWilsonFermionF;
 typedef WilsonFermion<GparityWilsonImplD>     GparityWilsonFermionD;
+
+typedef WilsonFermion<GparityWilsonImplRL>     GparityWilsonFermionRL;
+typedef WilsonFermion<GparityWilsonImplFH>     GparityWilsonFermionFH;
+typedef WilsonFermion<GparityWilsonImplDF>     GparityWilsonFermionDF;
+
 typedef DomainWallFermion<GparityWilsonImplR> GparityDomainWallFermionR;
 typedef DomainWallFermion<GparityWilsonImplF> GparityDomainWallFermionF;
 typedef DomainWallFermion<GparityWilsonImplD> GparityDomainWallFermionD;
 
+typedef DomainWallFermion<GparityWilsonImplRL> GparityDomainWallFermionRL;
+typedef DomainWallFermion<GparityWilsonImplFH> GparityDomainWallFermionFH;
+typedef DomainWallFermion<GparityWilsonImplDF> GparityDomainWallFermionDF;
+
 typedef WilsonTMFermion<GparityWilsonImplR> GparityWilsonTMFermionR;
 typedef WilsonTMFermion<GparityWilsonImplF> GparityWilsonTMFermionF;
 typedef WilsonTMFermion<GparityWilsonImplD> GparityWilsonTMFermionD;
+
+typedef WilsonTMFermion<GparityWilsonImplRL> GparityWilsonTMFermionRL;
+typedef WilsonTMFermion<GparityWilsonImplFH> GparityWilsonTMFermionFH;
+typedef WilsonTMFermion<GparityWilsonImplDF> GparityWilsonTMFermionDF;
+
 typedef MobiusFermion<GparityWilsonImplR> GparityMobiusFermionR;
 typedef MobiusFermion<GparityWilsonImplF> GparityMobiusFermionF;
 typedef MobiusFermion<GparityWilsonImplD> GparityMobiusFermionD;
 
+typedef MobiusFermion<GparityWilsonImplRL> GparityMobiusFermionRL;
+typedef MobiusFermion<GparityWilsonImplFH> GparityMobiusFermionFH;
+typedef MobiusFermion<GparityWilsonImplDF> GparityMobiusFermionDF;
+
 typedef ImprovedStaggeredFermion<StaggeredImplR> ImprovedStaggeredFermionR;
 typedef ImprovedStaggeredFermion<StaggeredImplF> ImprovedStaggeredFermionF;
 typedef ImprovedStaggeredFermion<StaggeredImplD> ImprovedStaggeredFermionD;

lib/qcd/action/fermion/FermionCore.h

@@ -55,7 +55,14 @@ Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
   template class A<ZWilsonImplF>;		\
   template class A<ZWilsonImplD>;		\
   template class A<GparityWilsonImplF>;		\
-  template class A<GparityWilsonImplD>;		
+  template class A<GparityWilsonImplD>;		\
+  template class A<WilsonImplFH>;		\
+  template class A<WilsonImplDF>;		\
+  template class A<ZWilsonImplFH>;		\
+  template class A<ZWilsonImplDF>;		\
+  template class A<GparityWilsonImplFH>;		\
+  template class A<GparityWilsonImplDF>;		
+
 
 #define AdjointFermOpTemplateInstantiate(A) \
   template class A<WilsonAdjImplF>; \
@@ -69,7 +76,11 @@ Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
   template class A<DomainWallVec5dImplF>;	\
   template class A<DomainWallVec5dImplD>;	\
   template class A<ZDomainWallVec5dImplF>;	\
-  template class A<ZDomainWallVec5dImplD>;	
+  template class A<ZDomainWallVec5dImplD>;	\
+  template class A<DomainWallVec5dImplFH>;	\
+  template class A<DomainWallVec5dImplDF>;	\
+  template class A<ZDomainWallVec5dImplFH>;	\
+  template class A<ZDomainWallVec5dImplDF>;	
 
 #define FermOpTemplateInstantiate(A) \
  FermOp4dVecTemplateInstantiate(A) \

lib/qcd/action/fermion/FermionOperatorImpl.h

@@ -35,7 +35,6 @@ directory
 namespace Grid {
 namespace QCD {
 
-
   //////////////////////////////////////////////
   // Template parameter class constructs to package
   // externally control Fermion implementations
@@ -90,6 +89,52 @@ namespace QCD {
   //  }
   //////////////////////////////////////////////
 
+  template <class T> struct SamePrecisionMapper {
+    typedef T HigherPrecVector ;
+    typedef T LowerPrecVector ;
+  };
+  template <class T> struct LowerPrecisionMapper {  };
+  template <> struct LowerPrecisionMapper<vRealF> {
+    typedef vRealF HigherPrecVector ;
+    typedef vRealH LowerPrecVector ;
+  };
+  template <> struct LowerPrecisionMapper<vRealD> {
+    typedef vRealD HigherPrecVector ;
+    typedef vRealF LowerPrecVector ;
+  };
+  template <> struct LowerPrecisionMapper<vComplexF> {
+    typedef vComplexF HigherPrecVector ;
+    typedef vComplexH LowerPrecVector ;
+  };
+  template <> struct LowerPrecisionMapper<vComplexD> {
+    typedef vComplexD HigherPrecVector ;
+    typedef vComplexF LowerPrecVector ;
+  };
+
+  struct CoeffReal {
+  public:
+    typedef RealD _Coeff_t;
+    static const int Nhcs = 2;
+    template<class Simd> using PrecisionMapper = SamePrecisionMapper<Simd>;
+  };
+  struct CoeffRealHalfComms {
+  public:
+    typedef RealD _Coeff_t;
+    static const int Nhcs = 1;
+    template<class Simd> using PrecisionMapper = LowerPrecisionMapper<Simd>;
+  };
+  struct CoeffComplex {
+  public:
+    typedef ComplexD _Coeff_t;
+    static const int Nhcs = 2;
+    template<class Simd> using PrecisionMapper = SamePrecisionMapper<Simd>;
+  };
+  struct CoeffComplexHalfComms {
+  public:
+    typedef ComplexD _Coeff_t;
+    static const int Nhcs = 1;
+    template<class Simd> using PrecisionMapper = LowerPrecisionMapper<Simd>;
+  };
 
   ////////////////////////////////////////////////////////////////////////
   // Implementation dependent fermion types
@@ -114,43 +159,48 @@ namespace QCD {
   /////////////////////////////////////////////////////////////////////////////
   // Single flavour four spinors with colour index
   /////////////////////////////////////////////////////////////////////////////
-  template <class S, class Representation = FundamentalRepresentation,class _Coeff_t = RealD >
+  template <class S, class Representation = FundamentalRepresentation,class Options = CoeffReal >
   class WilsonImpl : public PeriodicGaugeImpl<GaugeImplTypes<S, Representation::Dimension > > {
-
     public:
 
     static const int Dimension = Representation::Dimension;
+    static const bool LsVectorised=false;
+    static const int Nhcs = Options::Nhcs;
+
     typedef PeriodicGaugeImpl<GaugeImplTypes<S, Dimension > > Gimpl;
+    INHERIT_GIMPL_TYPES(Gimpl);
       
     //Necessary?
     constexpr bool is_fundamental() const{return Dimension == Nc ? 1 : 0;}
     
-    const bool LsVectorised=false;
-    typedef _Coeff_t Coeff_t;
-
-    INHERIT_GIMPL_TYPES(Gimpl);
+    typedef typename Options::_Coeff_t Coeff_t;
+    typedef typename Options::template PrecisionMapper<Simd>::LowerPrecVector SimdL;
       
     template <typename vtype> using iImplSpinor            = iScalar<iVector<iVector<vtype, Dimension>, Ns> >;
     template <typename vtype> using iImplPropagator        = iScalar<iMatrix<iMatrix<vtype, Dimension>, Ns> >;
     template <typename vtype> using iImplHalfSpinor        = iScalar<iVector<iVector<vtype, Dimension>, Nhs> >;
+    template <typename vtype> using iImplHalfCommSpinor    = iScalar<iVector<iVector<vtype, Dimension>, Nhcs> >;
     template <typename vtype> using iImplDoubledGaugeField = iVector<iScalar<iMatrix<vtype, Dimension> >, Nds>;
     
     typedef iImplSpinor<Simd>            SiteSpinor;
     typedef iImplPropagator<Simd>        SitePropagator;
     typedef iImplHalfSpinor<Simd>        SiteHalfSpinor;
+    typedef iImplHalfCommSpinor<SimdL>   SiteHalfCommSpinor;
     typedef iImplDoubledGaugeField<Simd> SiteDoubledGaugeField;
     
     typedef Lattice<SiteSpinor>            FermionField;
     typedef Lattice<SitePropagator>        PropagatorField;
     typedef Lattice<SiteDoubledGaugeField> DoubledGaugeField;
     
-    typedef WilsonCompressor<SiteHalfSpinor, SiteSpinor> Compressor;
+    typedef WilsonCompressor<SiteHalfCommSpinor,SiteHalfSpinor, SiteSpinor> Compressor;
     typedef WilsonImplParams ImplParams;
     typedef WilsonStencil<SiteSpinor, SiteHalfSpinor> StencilImpl;
     
     ImplParams Params;
     
-    WilsonImpl(const ImplParams &p = ImplParams()) : Params(p){};
+    WilsonImpl(const ImplParams &p = ImplParams()) : Params(p){
+      assert(Params.boundary_phases.size() == Nd);
+    };
       
     bool overlapCommsCompute(void) { return Params.overlapCommsCompute; };
       
@@ -170,14 +220,32 @@ namespace QCD {
       
     inline void DoubleStore(GridBase *GaugeGrid,
                             DoubledGaugeField &Uds,
-			    const GaugeField &Umu) {
+                            const GaugeField &Umu) 
+    {
+      typedef typename Simd::scalar_type scalar_type;
+
       conformable(Uds._grid, GaugeGrid);
       conformable(Umu._grid, GaugeGrid);
+
       GaugeLinkField U(GaugeGrid);
+      GaugeLinkField tmp(GaugeGrid);
+
+      Lattice<iScalar<vInteger> > coor(GaugeGrid);
       for (int mu = 0; mu < Nd; mu++) {
+
+	      auto pha = Params.boundary_phases[mu];
+	      scalar_type phase( real(pha),imag(pha) );
+
+        int Lmu = GaugeGrid->GlobalDimensions()[mu] - 1;
+
+        LatticeCoordinate(coor, mu);
+
         U = PeekIndex<LorentzIndex>(Umu, mu);
-	PokeIndex<LorentzIndex>(Uds, U, mu);
+        tmp = where(coor == Lmu, phase * U, U);
+        PokeIndex<LorentzIndex>(Uds, tmp, mu);
+
         U = adj(Cshift(U, mu, -1));
+        U = where(coor == 0, conjugate(phase) * U, U); 
         PokeIndex<LorentzIndex>(Uds, U, mu + 4);
       }
     }
@@ -209,21 +277,24 @@ namespace QCD {
   ////////////////////////////////////////////////////////////////////////////////////
   // Single flavour four spinors with colour index, 5d redblack
   ////////////////////////////////////////////////////////////////////////////////////
-
-template<class S,int Nrepresentation=Nc,class _Coeff_t = RealD>
+template<class S,int Nrepresentation=Nc, class Options=CoeffReal>
 class DomainWallVec5dImpl :  public PeriodicGaugeImpl< GaugeImplTypes< S,Nrepresentation> > { 
   public:
 
-  static const int Dimension = Nrepresentation;
-  const bool LsVectorised=true;
-  typedef _Coeff_t Coeff_t;      
   typedef PeriodicGaugeImpl<GaugeImplTypes<S, Nrepresentation> > Gimpl;
-  
   INHERIT_GIMPL_TYPES(Gimpl);
 
+  static const int Dimension = Nrepresentation;
+  static const bool LsVectorised=true;
+  static const int Nhcs = Options::Nhcs;
+      
+  typedef typename Options::_Coeff_t Coeff_t;      
+  typedef typename Options::template PrecisionMapper<Simd>::LowerPrecVector SimdL;
+  
   template <typename vtype> using iImplSpinor            = iScalar<iVector<iVector<vtype, Nrepresentation>, Ns> >;
   template <typename vtype> using iImplPropagator        = iScalar<iMatrix<iMatrix<vtype, Nrepresentation>, Ns> >;
   template <typename vtype> using iImplHalfSpinor        = iScalar<iVector<iVector<vtype, Nrepresentation>, Nhs> >;
+  template <typename vtype> using iImplHalfCommSpinor    = iScalar<iVector<iVector<vtype, Nrepresentation>, Nhcs> >;
   template <typename vtype> using iImplDoubledGaugeField = iVector<iScalar<iMatrix<vtype, Nrepresentation> >, Nds>;
   template <typename vtype> using iImplGaugeField        = iVector<iScalar<iMatrix<vtype, Nrepresentation> >, Nd>;
   template <typename vtype> using iImplGaugeLink         = iScalar<iScalar<iMatrix<vtype, Nrepresentation> > >;
@@ -231,10 +302,10 @@ class DomainWallVec5dImpl :  public PeriodicGaugeImpl< GaugeImplTypes< S,Nrepres
   typedef iImplSpinor<Simd>            SiteSpinor;
   typedef iImplPropagator<Simd>        SitePropagator;
   typedef iImplHalfSpinor<Simd>        SiteHalfSpinor;
+  typedef iImplHalfCommSpinor<SimdL>   SiteHalfCommSpinor;
   typedef Lattice<SiteSpinor>          FermionField;
   typedef Lattice<SitePropagator>      PropagatorField;
 
-
   /////////////////////////////////////////////////
   // Make the doubled gauge field a *scalar*
   /////////////////////////////////////////////////
@@ -243,7 +314,7 @@ class DomainWallVec5dImpl :  public PeriodicGaugeImpl< GaugeImplTypes< S,Nrepres
   typedef iImplGaugeLink<typename Simd::scalar_type>          SiteScalarGaugeLink;  // scalar
   typedef Lattice<SiteDoubledGaugeField>                      DoubledGaugeField;
       
-  typedef WilsonCompressor<SiteHalfSpinor, SiteSpinor> Compressor;
+  typedef WilsonCompressor<SiteHalfCommSpinor,SiteHalfSpinor, SiteSpinor> Compressor;
   typedef WilsonImplParams ImplParams;
   typedef WilsonStencil<SiteSpinor, SiteHalfSpinor> StencilImpl;
   
@@ -302,44 +373,89 @@ class DomainWallVec5dImpl :  public PeriodicGaugeImpl< GaugeImplTypes< S,Nrepres
     assert(0);
   }
 
-  inline void InsertForce5D(GaugeField &mat, FermionField &Btilde,FermionField &Atilde, int mu) 
-  {
+  inline void InsertForce5D(GaugeField &mat, FermionField &Btilde, FermionField &Atilde, int mu) {
+
     assert(0);
+    // Following lines to be revised after Peter's addition of half prec
+    // missing put lane...
+    /*
+    typedef decltype(traceIndex<SpinIndex>(outerProduct(Btilde[0], Atilde[0]))) result_type;
+    unsigned int LLs = Btilde._grid->_rdimensions[0];
+    conformable(Atilde._grid,Btilde._grid);
+    GridBase* grid = mat._grid;
+    GridBase* Bgrid = Btilde._grid;
+    unsigned int dimU = grid->Nd();
+    unsigned int dimF = Bgrid->Nd();
+    GaugeLinkField tmp(grid); 
+    tmp = zero;
+    
+    // FIXME 
+    // Current implementation works, thread safe, probably suboptimal
+    // Passing through the local coordinate for grid transformation
+    // the force grid is in general very different from the Ls vectorized grid
+
+    PARALLEL_FOR_LOOP
+    for (int so = 0; so < grid->oSites(); so++) {
+      std::vector<typename result_type::scalar_object> vres(Bgrid->Nsimd());
+      std::vector<int> ocoor;  grid->oCoorFromOindex(ocoor,so); 
+      for (int si = 0; si < tmp._grid->iSites(); si++){
+        typename result_type::scalar_object scalar_object; scalar_object = zero;
+        std::vector<int> local_coor;      
+        std::vector<int> icoor; grid->iCoorFromIindex(icoor,si);
+        grid->InOutCoorToLocalCoor(ocoor, icoor, local_coor);
+        for (int s = 0; s < LLs; s++) {
+          std::vector<int> slocal_coor(dimF);
+          slocal_coor[0] = s;
+          for (int s4d = 1; s4d< dimF; s4d++) slocal_coor[s4d] = local_coor[s4d-1];
+          int sF = Bgrid->oIndexReduced(slocal_coor);  
+          assert(sF < Bgrid->oSites());
+
+          extract(traceIndex<SpinIndex>(outerProduct(Btilde[sF], Atilde[sF])), vres); 
+          // sum across the 5d dimension
+          for (auto v : vres) scalar_object += v;  
+        }
+        tmp._odata[so].putlane(scalar_object, si);
+      }
+    }
+    PokeIndex<LorentzIndex>(mat, tmp, mu);
+    */
   }
 };
     
     ////////////////////////////////////////////////////////////////////////////////////////
     // Flavour doubled spinors; is Gparity the only? what about C*?
     ////////////////////////////////////////////////////////////////////////////////////////
-    
-template <class S, int Nrepresentation,class _Coeff_t = RealD>
+template <class S, int Nrepresentation, class Options=CoeffReal>
 class GparityWilsonImpl : public ConjugateGaugeImpl<GaugeImplTypes<S, Nrepresentation> > {
  public:
 
  static const int Dimension = Nrepresentation;
+ static const int Nhcs = Options::Nhcs;
+ static const bool LsVectorised=false;
 
- const bool LsVectorised=false;
-
- typedef _Coeff_t Coeff_t;
  typedef ConjugateGaugeImpl< GaugeImplTypes<S,Nrepresentation> > Gimpl;
- 
  INHERIT_GIMPL_TYPES(Gimpl);
 
+ typedef typename Options::_Coeff_t Coeff_t;
+ typedef typename Options::template PrecisionMapper<Simd>::LowerPrecVector SimdL;
+      
  template <typename vtype> using iImplSpinor            = iVector<iVector<iVector<vtype, Nrepresentation>, Ns>,   Ngp>;
- template <typename vtype> using iImplPropagator        = iVector<iMatrix<iMatrix<vtype, Nrepresentation>, Ns>, Ngp >;
+ template <typename vtype> using iImplPropagator        = iVector<iMatrix<iMatrix<vtype, Nrepresentation>, Ns>,   Ngp>;
  template <typename vtype> using iImplHalfSpinor        = iVector<iVector<iVector<vtype, Nrepresentation>, Nhs>,  Ngp>;
+ template <typename vtype> using iImplHalfCommSpinor    = iVector<iVector<iVector<vtype, Nrepresentation>, Nhcs>, Ngp>;
  template <typename vtype> using iImplDoubledGaugeField = iVector<iVector<iScalar<iMatrix<vtype, Nrepresentation> >, Nds>, Ngp>;
 
  typedef iImplSpinor<Simd>            SiteSpinor;
  typedef iImplPropagator<Simd>        SitePropagator;
  typedef iImplHalfSpinor<Simd>        SiteHalfSpinor;
+ typedef iImplHalfCommSpinor<SimdL>   SiteHalfCommSpinor;
  typedef iImplDoubledGaugeField<Simd> SiteDoubledGaugeField;
 
  typedef Lattice<SiteSpinor> FermionField;
  typedef Lattice<SitePropagator> PropagatorField;
  typedef Lattice<SiteDoubledGaugeField> DoubledGaugeField;
  
- typedef WilsonCompressor<SiteHalfSpinor, SiteSpinor> Compressor;
+ typedef WilsonCompressor<SiteHalfCommSpinor,SiteHalfSpinor, SiteSpinor> Compressor;
  typedef WilsonStencil<SiteSpinor, SiteHalfSpinor> StencilImpl;
  
  typedef GparityWilsonImplParams ImplParams;
@@ -459,6 +575,7 @@ class GparityWilsonImpl : public ConjugateGaugeImpl<GaugeImplTypes<S, Nrepresent
        Utmp = where(coor==0,Uconj,Utmp);
      }
 
+	  
      parallel_for(auto ss=U.begin();ss<U.end();ss++){
        Uds[ss](0)(mu+4) = Utmp[ss]();
      }
@@ -475,7 +592,6 @@ class GparityWilsonImpl : public ConjugateGaugeImpl<GaugeImplTypes<S, Nrepresent
    }
  }
       
-      
  inline void InsertForce4D(GaugeField &mat, FermionField &Btilde, FermionField &A, int mu) {
 
    // DhopDir provides U or Uconj depending on coor/flavour.
@@ -483,7 +599,7 @@ class GparityWilsonImpl : public ConjugateGaugeImpl<GaugeImplTypes<S, Nrepresent
    // use lorentz for flavour as hack.
    auto tmp = TraceIndex<SpinIndex>(outerProduct(Btilde, A));
    parallel_for(auto ss = tmp.begin(); ss < tmp.end(); ss++) {
-     link[ss]() = tmp[ss](0, 0) - conjugate(tmp[ss](1, 1));
+     link[ss]() = tmp[ss](0, 0) + conjugate(tmp[ss](1, 1));
    }
    PokeIndex<LorentzIndex>(mat, link, mu);
    return;
@@ -508,40 +624,36 @@ class GparityWilsonImpl : public ConjugateGaugeImpl<GaugeImplTypes<S, Nrepresent
 
 };
 
-
-  /////////////////////////////////////////////////////////////////////////////
-  // Single flavour one component spinors with colour index
-  /////////////////////////////////////////////////////////////////////////////
-  template <class S, class Representation = FundamentalRepresentation >
-  class StaggeredImpl : public PeriodicGaugeImpl<GaugeImplTypes<S, Representation::Dimension > > {
+/////////////////////////////////////////////////////////////////////////////
+// Single flavour one component spinors with colour index
+/////////////////////////////////////////////////////////////////////////////
+template <class S, class Representation = FundamentalRepresentation >
+class StaggeredImpl : public PeriodicGaugeImpl<GaugeImplTypes<S, Representation::Dimension > > {
 
     public:
 
     typedef RealD  _Coeff_t ;
     static const int Dimension = Representation::Dimension;
+    static const bool LsVectorised=false;
     typedef PeriodicGaugeImpl<GaugeImplTypes<S, Dimension > > Gimpl;
       
     //Necessary?
     constexpr bool is_fundamental() const{return Dimension == Nc ? 1 : 0;}
     
-    const bool LsVectorised=false;
     typedef _Coeff_t Coeff_t;
 
     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;
@@ -641,8 +753,6 @@ class GparityWilsonImpl : public ConjugateGaugeImpl<GaugeImplTypes<S, Nrepresent
     }
   };
 
-
-
   /////////////////////////////////////////////////////////////////////////////
   // Single flavour one component spinors with colour index. 5d vec
   /////////////////////////////////////////////////////////////////////////////
@@ -651,20 +761,17 @@ class GparityWilsonImpl : public ConjugateGaugeImpl<GaugeImplTypes<S, Nrepresent
 
     public:
 
-    typedef RealD  _Coeff_t ;
     static const int Dimension = Representation::Dimension;
+    static const bool LsVectorised=true;
+    typedef RealD   Coeff_t ;
     typedef PeriodicGaugeImpl<GaugeImplTypes<S, Dimension > > Gimpl;
       
     //Necessary?
     constexpr bool is_fundamental() const{return Dimension == Nc ? 1 : 0;}
 
-    const bool LsVectorised=true;
-
-    typedef _Coeff_t Coeff_t;
 
     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>;
@@ -681,12 +788,10 @@ class GparityWilsonImpl : public ConjugateGaugeImpl<GaugeImplTypes<S, Nrepresent
     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;
@@ -823,43 +928,61 @@ class GparityWilsonImpl : public ConjugateGaugeImpl<GaugeImplTypes<S, Nrepresent
     }
   };
 
+typedef WilsonImpl<vComplex,  FundamentalRepresentation, CoeffReal > WilsonImplR;  // Real.. whichever prec
+typedef WilsonImpl<vComplexF, FundamentalRepresentation, CoeffReal > WilsonImplF;  // Float
+typedef WilsonImpl<vComplexD, FundamentalRepresentation, CoeffReal > WilsonImplD;  // Double
 
+typedef WilsonImpl<vComplex,  FundamentalRepresentation, CoeffRealHalfComms > WilsonImplRL;  // Real.. whichever prec
+typedef WilsonImpl<vComplexF, FundamentalRepresentation, CoeffRealHalfComms > WilsonImplFH;  // Float
+typedef WilsonImpl<vComplexD, FundamentalRepresentation, CoeffRealHalfComms > WilsonImplDF;  // Double
 
- typedef WilsonImpl<vComplex,  FundamentalRepresentation > WilsonImplR;   // Real.. whichever prec
- typedef WilsonImpl<vComplexF, FundamentalRepresentation > WilsonImplF;  // Float
- typedef WilsonImpl<vComplexD, FundamentalRepresentation > WilsonImplD;  // Double
+typedef WilsonImpl<vComplex,  FundamentalRepresentation, CoeffComplex > ZWilsonImplR; // Real.. whichever prec
+typedef WilsonImpl<vComplexF, FundamentalRepresentation, CoeffComplex > ZWilsonImplF; // Float
+typedef WilsonImpl<vComplexD, FundamentalRepresentation, CoeffComplex > ZWilsonImplD; // Double
 
- typedef WilsonImpl<vComplex,  FundamentalRepresentation, ComplexD > ZWilsonImplR; // Real.. whichever prec
- typedef WilsonImpl<vComplexF, FundamentalRepresentation, ComplexD > ZWilsonImplF; // Float
- typedef WilsonImpl<vComplexD, FundamentalRepresentation, ComplexD > ZWilsonImplD; // Double
+typedef WilsonImpl<vComplex,  FundamentalRepresentation, CoeffComplexHalfComms > ZWilsonImplRL; // Real.. whichever prec
+typedef WilsonImpl<vComplexF, FundamentalRepresentation, CoeffComplexHalfComms > ZWilsonImplFH; // Float
+typedef WilsonImpl<vComplexD, FundamentalRepresentation, CoeffComplexHalfComms > ZWilsonImplDF; // Double
  
- typedef WilsonImpl<vComplex,  AdjointRepresentation > WilsonAdjImplR;   // Real.. whichever prec
- typedef WilsonImpl<vComplexF, AdjointRepresentation > WilsonAdjImplF;  // Float
- typedef WilsonImpl<vComplexD, AdjointRepresentation > WilsonAdjImplD;  // Double
+typedef WilsonImpl<vComplex,  AdjointRepresentation, CoeffReal > WilsonAdjImplR;   // Real.. whichever prec
+typedef WilsonImpl<vComplexF, AdjointRepresentation, CoeffReal > WilsonAdjImplF;  // Float
+typedef WilsonImpl<vComplexD, AdjointRepresentation, CoeffReal > WilsonAdjImplD;  // Double
  
- typedef WilsonImpl<vComplex,  TwoIndexSymmetricRepresentation > WilsonTwoIndexSymmetricImplR;   // Real.. whichever prec
- typedef WilsonImpl<vComplexF, TwoIndexSymmetricRepresentation > WilsonTwoIndexSymmetricImplF;  // Float
- typedef WilsonImpl<vComplexD, TwoIndexSymmetricRepresentation > WilsonTwoIndexSymmetricImplD;  // Double
+typedef WilsonImpl<vComplex,  TwoIndexSymmetricRepresentation, CoeffReal > WilsonTwoIndexSymmetricImplR;   // Real.. whichever prec
+typedef WilsonImpl<vComplexF, TwoIndexSymmetricRepresentation, CoeffReal > WilsonTwoIndexSymmetricImplF;  // Float
+typedef WilsonImpl<vComplexD, TwoIndexSymmetricRepresentation, CoeffReal > WilsonTwoIndexSymmetricImplD;  // Double
  
- typedef DomainWallVec5dImpl<vComplex ,Nc> DomainWallVec5dImplR; // Real.. whichever prec
- typedef DomainWallVec5dImpl<vComplexF,Nc> DomainWallVec5dImplF; // Float
- typedef DomainWallVec5dImpl<vComplexD,Nc> DomainWallVec5dImplD; // Double
+typedef DomainWallVec5dImpl<vComplex ,Nc, CoeffReal> DomainWallVec5dImplR; // Real.. whichever prec
+typedef DomainWallVec5dImpl<vComplexF,Nc, CoeffReal> DomainWallVec5dImplF; // Float
+typedef DomainWallVec5dImpl<vComplexD,Nc, CoeffReal> DomainWallVec5dImplD; // Double
  
- typedef DomainWallVec5dImpl<vComplex ,Nc,ComplexD> ZDomainWallVec5dImplR; // Real.. whichever prec
- typedef DomainWallVec5dImpl<vComplexF,Nc,ComplexD> ZDomainWallVec5dImplF; // Float
- typedef DomainWallVec5dImpl<vComplexD,Nc,ComplexD> ZDomainWallVec5dImplD; // Double
+typedef DomainWallVec5dImpl<vComplex ,Nc, CoeffRealHalfComms> DomainWallVec5dImplRL; // Real.. whichever prec
+typedef DomainWallVec5dImpl<vComplexF,Nc, CoeffRealHalfComms> DomainWallVec5dImplFH; // Float
+typedef DomainWallVec5dImpl<vComplexD,Nc, CoeffRealHalfComms> DomainWallVec5dImplDF; // Double
  
- typedef GparityWilsonImpl<vComplex , Nc> GparityWilsonImplR;  // Real.. whichever prec
- typedef GparityWilsonImpl<vComplexF, Nc> GparityWilsonImplF;  // Float
- typedef GparityWilsonImpl<vComplexD, Nc> GparityWilsonImplD;  // Double
+typedef DomainWallVec5dImpl<vComplex ,Nc,CoeffComplex> ZDomainWallVec5dImplR; // Real.. whichever prec
+typedef DomainWallVec5dImpl<vComplexF,Nc,CoeffComplex> ZDomainWallVec5dImplF; // Float
+typedef DomainWallVec5dImpl<vComplexD,Nc,CoeffComplex> ZDomainWallVec5dImplD; // Double
  
- typedef StaggeredImpl<vComplex,  FundamentalRepresentation > StaggeredImplR;   // Real.. whichever prec
- typedef StaggeredImpl<vComplexF, FundamentalRepresentation > StaggeredImplF;  // Float
- typedef StaggeredImpl<vComplexD, FundamentalRepresentation > StaggeredImplD;  // Double
+typedef DomainWallVec5dImpl<vComplex ,Nc,CoeffComplexHalfComms> ZDomainWallVec5dImplRL; // Real.. whichever prec
+typedef DomainWallVec5dImpl<vComplexF,Nc,CoeffComplexHalfComms> ZDomainWallVec5dImplFH; // Float
+typedef DomainWallVec5dImpl<vComplexD,Nc,CoeffComplexHalfComms> ZDomainWallVec5dImplDF; // Double
  
- typedef StaggeredVec5dImpl<vComplex,  FundamentalRepresentation > StaggeredVec5dImplR;   // Real.. whichever prec
- typedef StaggeredVec5dImpl<vComplexF, FundamentalRepresentation > StaggeredVec5dImplF;  // Float
- typedef StaggeredVec5dImpl<vComplexD, FundamentalRepresentation > StaggeredVec5dImplD;  // Double
+typedef GparityWilsonImpl<vComplex , Nc,CoeffReal> GparityWilsonImplR;  // Real.. whichever prec
+typedef GparityWilsonImpl<vComplexF, Nc,CoeffReal> GparityWilsonImplF;  // Float
+typedef GparityWilsonImpl<vComplexD, Nc,CoeffReal> GparityWilsonImplD;  // Double
+ 
+typedef GparityWilsonImpl<vComplex , Nc,CoeffRealHalfComms> GparityWilsonImplRL;  // Real.. whichever prec
+typedef GparityWilsonImpl<vComplexF, Nc,CoeffRealHalfComms> GparityWilsonImplFH;  // Float
+typedef GparityWilsonImpl<vComplexD, Nc,CoeffRealHalfComms> GparityWilsonImplDF;  // Double
+
+typedef StaggeredImpl<vComplex,  FundamentalRepresentation > StaggeredImplR;   // Real.. whichever prec
+typedef StaggeredImpl<vComplexF, FundamentalRepresentation > StaggeredImplF;  // Float
+typedef StaggeredImpl<vComplexD, FundamentalRepresentation > StaggeredImplD;  // Double
+
+typedef StaggeredVec5dImpl<vComplex,  FundamentalRepresentation > StaggeredVec5dImplR;   // Real.. whichever prec
+typedef StaggeredVec5dImpl<vComplexF, FundamentalRepresentation > StaggeredVec5dImplF;  // Float
+typedef StaggeredVec5dImpl<vComplexD, FundamentalRepresentation > StaggeredVec5dImplD;  // Double
 
 }}
 

lib/qcd/action/fermion/WilsonCompressor.h

@@ -33,220 +33,314 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 namespace Grid {
 namespace QCD {
 
-  template<class SiteHalfSpinor,class SiteSpinor>
-  class WilsonCompressor {
+/////////////////////////////////////////////////////////////////////////////////////////////
+// optimised versions supporting half precision too
+/////////////////////////////////////////////////////////////////////////////////////////////
+
+template<class _HCspinor,class _Hspinor,class _Spinor, class projector,typename SFINAE = void >
+class WilsonCompressorTemplate;
+
+
+template<class _HCspinor,class _Hspinor,class _Spinor, class projector>
+class WilsonCompressorTemplate< _HCspinor, _Hspinor, _Spinor, projector,
+  typename std::enable_if<std::is_same<_HCspinor,_Hspinor>::value>::type >
+{
  public:
-    int mu;
-    int dag;
   
-    WilsonCompressor(int _dag){
-      mu=0;
-      dag=_dag;
-      assert((dag==0)||(dag==1));
-    }
-    void Point(int p) { 
-      mu=p;
-    };
+  int mu,dag;  
 
-    inline SiteHalfSpinor operator () (const SiteSpinor &in) {
-      SiteHalfSpinor ret;
-      int mudag=mu;
-      if (!dag) {
-	mudag=(mu+Nd)%(2*Nd);
+  void Point(int p) { mu=p; };
+
+  WilsonCompressorTemplate(int _dag=0){
+    dag = _dag;
   }
-      switch(mudag) {
-      case Xp:
-	spProjXp(ret,in);
-	break;
-      case Yp:
-	spProjYp(ret,in);
-	break;
-      case Zp:
-	spProjZp(ret,in);
-	break;
-      case Tp:
-	spProjTp(ret,in);
-	break;
-      case Xm:
-	spProjXm(ret,in);
-	break;
-      case Ym:
-	spProjYm(ret,in);
-	break;
-      case Zm:
-	spProjZm(ret,in);
-	break;
-      case Tm:
-	spProjTm(ret,in);
-	break;
-      default: 
+
+  typedef _Spinor         SiteSpinor;
+  typedef _Hspinor     SiteHalfSpinor;
+  typedef _HCspinor SiteHalfCommSpinor;
+  typedef typename SiteHalfCommSpinor::vector_type vComplexLow;
+  typedef typename SiteHalfSpinor::vector_type     vComplexHigh;
+  constexpr static int Nw=sizeof(SiteHalfSpinor)/sizeof(vComplexHigh);
+
+  inline int CommDatumSize(void) {
+    return sizeof(SiteHalfCommSpinor);
+  }
+
+  /*****************************************************/
+  /* Compress includes precision change if mpi data is not same */
+  /*****************************************************/
+  inline void Compress(SiteHalfSpinor *buf,Integer o,const SiteSpinor &in) {
+    projector::Proj(buf[o],in,mu,dag);
+  }
+
+  /*****************************************************/
+  /* Exchange includes precision change if mpi data is not same */
+  /*****************************************************/
+  inline void Exchange(SiteHalfSpinor *mp,
+                       SiteHalfSpinor *vp0,
+                       SiteHalfSpinor *vp1,
+		       Integer type,Integer o){
+    exchange(mp[2*o],mp[2*o+1],vp0[o],vp1[o],type);
+  }
+
+  /*****************************************************/
+  /* Have a decompression step if mpi data is not same */
+  /*****************************************************/
+  inline void Decompress(SiteHalfSpinor *out,
+			 SiteHalfSpinor *in, Integer o) {    
     assert(0);
-	break;
   }
-      return ret;
-    }
-  };
 
-  /////////////////////////
-  // optimised versions
-  /////////////////////////
+  /*****************************************************/
+  /* Compress Exchange                                 */
+  /*****************************************************/
+  inline void CompressExchange(SiteHalfSpinor *out0,
+			       SiteHalfSpinor *out1,
+			       const SiteSpinor *in,
+			       Integer j,Integer k, Integer m,Integer type){
+    SiteHalfSpinor temp1, temp2,temp3,temp4;
+    projector::Proj(temp1,in[k],mu,dag);
+    projector::Proj(temp2,in[m],mu,dag);
+    exchange(out0[j],out1[j],temp1,temp2,type);
+  }
 
-  template<class SiteHalfSpinor,class SiteSpinor>
-  class WilsonXpCompressor {
-  public:
-    inline SiteHalfSpinor operator () (const SiteSpinor &in) {
-      SiteHalfSpinor ret;
-      spProjXp(ret,in);
-      return ret;
-    }
-  };
-  template<class SiteHalfSpinor,class SiteSpinor>
-  class WilsonYpCompressor {
-  public:
-    inline SiteHalfSpinor operator () (const SiteSpinor &in) {
-      SiteHalfSpinor ret;
-      spProjYp(ret,in);
-      return ret;
-    }
-  };
-  template<class SiteHalfSpinor,class SiteSpinor>
-  class WilsonZpCompressor {
-  public:
-    inline SiteHalfSpinor operator () (const SiteSpinor &in) {
-      SiteHalfSpinor ret;
-      spProjZp(ret,in);
-      return ret;
-    }
-  };
-  template<class SiteHalfSpinor,class SiteSpinor>
-  class WilsonTpCompressor {
-  public:
-    inline SiteHalfSpinor operator () (const SiteSpinor &in) {
-      SiteHalfSpinor ret;
-      spProjTp(ret,in);
-      return ret;
-    }
-  };
+  /*****************************************************/
+  /* Pass the info to the stencil */
+  /*****************************************************/
+  inline bool DecompressionStep(void) { return false; }
 
-  template<class SiteHalfSpinor,class SiteSpinor>
-  class WilsonXmCompressor {
-  public:
-    inline SiteHalfSpinor operator () (const SiteSpinor &in) {
-      SiteHalfSpinor ret;
-      spProjXm(ret,in);
-      return ret;
-    }
-  };
-  template<class SiteHalfSpinor,class SiteSpinor>
-  class WilsonYmCompressor {
-  public:
-    inline SiteHalfSpinor operator () (const SiteSpinor &in) {
-      SiteHalfSpinor ret;
-      spProjYm(ret,in);
-      return ret;
-    }
-  };
-  template<class SiteHalfSpinor,class SiteSpinor>
-  class WilsonZmCompressor {
-  public:
-    inline SiteHalfSpinor operator () (const SiteSpinor &in) {
-      SiteHalfSpinor ret;
-      spProjZm(ret,in);
-      return ret;
-    }
-  };
-  template<class SiteHalfSpinor,class SiteSpinor>
-  class WilsonTmCompressor {
-  public:
-    inline SiteHalfSpinor operator () (const SiteSpinor &in) {
-      SiteHalfSpinor ret;
-      spProjTm(ret,in);
-      return ret;
-    }
-  };
+};
 
-    // Fast comms buffer manipulation which should inline right through (avoid direction
-    // dependent logic that prevents inlining
-  template<class vobj,class cobj>
-  class WilsonStencil : public CartesianStencil<vobj,cobj> {
+template<class _HCspinor,class _Hspinor,class _Spinor, class projector>
+class WilsonCompressorTemplate< _HCspinor, _Hspinor, _Spinor, projector,
+  typename std::enable_if<!std::is_same<_HCspinor,_Hspinor>::value>::type >
+{
  public:
   
+  int mu,dag;  
+
+  void Point(int p) { mu=p; };
+
+  WilsonCompressorTemplate(int _dag=0){
+    dag = _dag;
+  }
+
+  typedef _Spinor         SiteSpinor;
+  typedef _Hspinor     SiteHalfSpinor;
+  typedef _HCspinor SiteHalfCommSpinor;
+  typedef typename SiteHalfCommSpinor::vector_type vComplexLow;
+  typedef typename SiteHalfSpinor::vector_type     vComplexHigh;
+  constexpr static int Nw=sizeof(SiteHalfSpinor)/sizeof(vComplexHigh);
+
+  inline int CommDatumSize(void) {
+    return sizeof(SiteHalfCommSpinor);
+  }
+
+  /*****************************************************/
+  /* Compress includes precision change if mpi data is not same */
+  /*****************************************************/
+  inline void Compress(SiteHalfSpinor *buf,Integer o,const SiteSpinor &in) {
+    SiteHalfSpinor hsp;
+    SiteHalfCommSpinor *hbuf = (SiteHalfCommSpinor *)buf;
+    projector::Proj(hsp,in,mu,dag);
+    precisionChange((vComplexLow *)&hbuf[o],(vComplexHigh *)&hsp,Nw);
+  }
+
+  /*****************************************************/
+  /* Exchange includes precision change if mpi data is not same */
+  /*****************************************************/
+  inline void Exchange(SiteHalfSpinor *mp,
+                       SiteHalfSpinor *vp0,
+                       SiteHalfSpinor *vp1,
+		       Integer type,Integer o){
+    SiteHalfSpinor vt0,vt1;
+    SiteHalfCommSpinor *vpp0 = (SiteHalfCommSpinor *)vp0;
+    SiteHalfCommSpinor *vpp1 = (SiteHalfCommSpinor *)vp1;
+    precisionChange((vComplexHigh *)&vt0,(vComplexLow *)&vpp0[o],Nw);
+    precisionChange((vComplexHigh *)&vt1,(vComplexLow *)&vpp1[o],Nw);
+    exchange(mp[2*o],mp[2*o+1],vt0,vt1,type);
+  }
+
+  /*****************************************************/
+  /* Have a decompression step if mpi data is not same */
+  /*****************************************************/
+  inline void Decompress(SiteHalfSpinor *out,
+			 SiteHalfSpinor *in, Integer o){
+    SiteHalfCommSpinor *hin=(SiteHalfCommSpinor *)in;
+    precisionChange((vComplexHigh *)&out[o],(vComplexLow *)&hin[o],Nw);
+  }
+
+  /*****************************************************/
+  /* Compress Exchange                                 */
+  /*****************************************************/
+  inline void CompressExchange(SiteHalfSpinor *out0,
+			       SiteHalfSpinor *out1,
+			       const SiteSpinor *in,
+			       Integer j,Integer k, Integer m,Integer type){
+    SiteHalfSpinor temp1, temp2,temp3,temp4;
+    SiteHalfCommSpinor *hout0 = (SiteHalfCommSpinor *)out0;
+    SiteHalfCommSpinor *hout1 = (SiteHalfCommSpinor *)out1;
+    projector::Proj(temp1,in[k],mu,dag);
+    projector::Proj(temp2,in[m],mu,dag);
+    exchange(temp3,temp4,temp1,temp2,type);
+    precisionChange((vComplexLow *)&hout0[j],(vComplexHigh *)&temp3,Nw);
+    precisionChange((vComplexLow *)&hout1[j],(vComplexHigh *)&temp4,Nw);
+  }
+
+  /*****************************************************/
+  /* Pass the info to the stencil */
+  /*****************************************************/
+  inline bool DecompressionStep(void) { return true; }
+
+};
+
+#define DECLARE_PROJ(Projector,Compressor,spProj)			\
+  class Projector {							\
+  public:								\
+    template<class hsp,class fsp>					\
+    static void Proj(hsp &result,const fsp &in,int mu,int dag){			\
+      spProj(result,in);						\
+    }									\
+  };									\
+template<typename HCS,typename HS,typename S> using Compressor = WilsonCompressorTemplate<HCS,HS,S,Projector>;
+
+DECLARE_PROJ(WilsonXpProjector,WilsonXpCompressor,spProjXp);
+DECLARE_PROJ(WilsonYpProjector,WilsonYpCompressor,spProjYp);
+DECLARE_PROJ(WilsonZpProjector,WilsonZpCompressor,spProjZp);
+DECLARE_PROJ(WilsonTpProjector,WilsonTpCompressor,spProjTp);
+DECLARE_PROJ(WilsonXmProjector,WilsonXmCompressor,spProjXm);
+DECLARE_PROJ(WilsonYmProjector,WilsonYmCompressor,spProjYm);
+DECLARE_PROJ(WilsonZmProjector,WilsonZmCompressor,spProjZm);
+DECLARE_PROJ(WilsonTmProjector,WilsonTmCompressor,spProjTm);
+
+class WilsonProjector {
+ public:
+  template<class hsp,class fsp>
+  static void Proj(hsp &result,const fsp &in,int mu,int dag){
+    int mudag=dag? mu : (mu+Nd)%(2*Nd);
+    switch(mudag) {
+    case Xp:	spProjXp(result,in);	break;
+    case Yp:	spProjYp(result,in);	break;
+    case Zp:	spProjZp(result,in);	break;
+    case Tp:	spProjTp(result,in);	break;
+    case Xm:	spProjXm(result,in);	break;
+    case Ym:	spProjYm(result,in);	break;
+    case Zm:	spProjZm(result,in);	break;
+    case Tm:	spProjTm(result,in);	break;
+    default: 	assert(0);	        break;
+    }
+  }
+};
+template<typename HCS,typename HS,typename S> using WilsonCompressor = WilsonCompressorTemplate<HCS,HS,S,WilsonProjector>;
+
+// Fast comms buffer manipulation which should inline right through (avoid direction
+// dependent logic that prevents inlining
+template<class vobj,class cobj>
+class WilsonStencil : public CartesianStencil<vobj,cobj> {
+public:
+
   typedef CartesianCommunicator::CommsRequest_t CommsRequest_t;
 
+  std::vector<int> same_node;
+  std::vector<int> surface_list;
+
   WilsonStencil(GridBase *grid,
 		int npoints,
 		int checkerboard,
 		const std::vector<int> &directions,
-		const std::vector<int> &distances)  : CartesianStencil<vobj,cobj> (grid,npoints,checkerboard,directions,distances) 
-      {    };
+		const std::vector<int> &distances)  
+    : CartesianStencil<vobj,cobj> (grid,npoints,checkerboard,directions,distances) ,
+    same_node(npoints)
+  { 
+    surface_list.resize(0);
+  };
+
+  void BuildSurfaceList(int Ls,int vol4){
+
+    // find same node for SHM
+    // Here we know the distance is 1 for WilsonStencil
+    for(int point=0;point<this->_npoints;point++){
+      same_node[point] = this->SameNode(point);
+      //      std::cout << " dir " <<point<<" same_node " <<same_node[point]<<std::endl;
+    }
+    
+    for(int site = 0 ;site< vol4;site++){
+      int local = 1;
+      for(int point=0;point<this->_npoints;point++){
+	if( (!this->GetNodeLocal(site*Ls,point)) && (!same_node[point]) ){ 
+	  local = 0;
+	}
+      }
+      if(local == 0) { 
+	surface_list.push_back(site);
+      }
+    }
+  }
 
   template < class compressor>
   void HaloExchangeOpt(const Lattice<vobj> &source,compressor &compress) 
   {
     std::vector<std::vector<CommsRequest_t> > reqs;
-      HaloExchangeOptGather(source,compress);
+    this->HaloExchangeOptGather(source,compress);
     this->CommunicateBegin(reqs);
-      this->calls++;
     this->CommunicateComplete(reqs);
-      this->CommsMerge();
+    this->CommsMerge(compress);
+    this->CommsMergeSHM(compress);
   }
   
-    template < class compressor>
+  template <class compressor>
   void HaloExchangeOptGather(const Lattice<vobj> &source,compressor &compress) 
   {
-      this->calls++;
-      this->Mergers.resize(0); 
-      this->Packets.resize(0);
+    this->Prepare();
     this->HaloGatherOpt(source,compress);
   }
 
-
-    template < class compressor>
+  template <class compressor>
   void HaloGatherOpt(const Lattice<vobj> &source,compressor &compress)
   {
+    // Strategy. Inherit types from Compressor.
+    // Use types to select the write direction by directon compressor
+    typedef typename compressor::SiteSpinor         SiteSpinor;
+    typedef typename compressor::SiteHalfSpinor     SiteHalfSpinor;
+    typedef typename compressor::SiteHalfCommSpinor SiteHalfCommSpinor;
+
     this->_grid->StencilBarrier();
-      // conformable(source._grid,_grid);
+
     assert(source._grid==this->_grid);
     this->halogtime-=usecond();
     
     this->u_comm_offset=0;
       
+    WilsonXpCompressor<SiteHalfCommSpinor,SiteHalfSpinor,SiteSpinor> XpCompress; 
+    WilsonYpCompressor<SiteHalfCommSpinor,SiteHalfSpinor,SiteSpinor> YpCompress; 
+    WilsonZpCompressor<SiteHalfCommSpinor,SiteHalfSpinor,SiteSpinor> ZpCompress; 
+    WilsonTpCompressor<SiteHalfCommSpinor,SiteHalfSpinor,SiteSpinor> TpCompress;
+    WilsonXmCompressor<SiteHalfCommSpinor,SiteHalfSpinor,SiteSpinor> XmCompress; 
+    WilsonYmCompressor<SiteHalfCommSpinor,SiteHalfSpinor,SiteSpinor> YmCompress; 
+    WilsonZmCompressor<SiteHalfCommSpinor,SiteHalfSpinor,SiteSpinor> ZmCompress; 
+    WilsonTmCompressor<SiteHalfCommSpinor,SiteHalfSpinor,SiteSpinor> TmCompress;
+
     int dag = compress.dag;
-      
-      WilsonXpCompressor<cobj,vobj> XpCompress; 
-      WilsonYpCompressor<cobj,vobj> YpCompress; 
-      WilsonZpCompressor<cobj,vobj> ZpCompress; 
-      WilsonTpCompressor<cobj,vobj> TpCompress;
-      WilsonXmCompressor<cobj,vobj> XmCompress;
-      WilsonYmCompressor<cobj,vobj> YmCompress;
-      WilsonZmCompressor<cobj,vobj> ZmCompress;
-      WilsonTmCompressor<cobj,vobj> TmCompress;
-
-      // Gather all comms buffers
-      //    for(int point = 0 ; point < _npoints; point++) {
-      //      compress.Point(point);
-      //      HaloGatherDir(source,compress,point,face_idx);
-      //    }
     int face_idx=0;
     if ( dag ) { 
       //	std::cout << " Optimised Dagger compress " <<std::endl;
-	this->HaloGatherDir(source,XpCompress,Xp,face_idx);
-	this->HaloGatherDir(source,YpCompress,Yp,face_idx);
-	this->HaloGatherDir(source,ZpCompress,Zp,face_idx);
-	this->HaloGatherDir(source,TpCompress,Tp,face_idx);
-	this->HaloGatherDir(source,XmCompress,Xm,face_idx);
-	this->HaloGatherDir(source,YmCompress,Ym,face_idx);
-	this->HaloGatherDir(source,ZmCompress,Zm,face_idx);
-	this->HaloGatherDir(source,TmCompress,Tm,face_idx);
+      assert(same_node[Xp]==this->HaloGatherDir(source,XpCompress,Xp,face_idx));
+      assert(same_node[Yp]==this->HaloGatherDir(source,YpCompress,Yp,face_idx));
+      assert(same_node[Zp]==this->HaloGatherDir(source,ZpCompress,Zp,face_idx));
+      assert(same_node[Tp]==this->HaloGatherDir(source,TpCompress,Tp,face_idx));
+      assert(same_node[Xm]==this->HaloGatherDir(source,XmCompress,Xm,face_idx));
+      assert(same_node[Ym]==this->HaloGatherDir(source,YmCompress,Ym,face_idx));
+      assert(same_node[Zm]==this->HaloGatherDir(source,ZmCompress,Zm,face_idx));
+      assert(same_node[Tm]==this->HaloGatherDir(source,TmCompress,Tm,face_idx));
     } else {
-	this->HaloGatherDir(source,XmCompress,Xp,face_idx);
-	this->HaloGatherDir(source,YmCompress,Yp,face_idx);
-	this->HaloGatherDir(source,ZmCompress,Zp,face_idx);
-	this->HaloGatherDir(source,TmCompress,Tp,face_idx);
-	this->HaloGatherDir(source,XpCompress,Xm,face_idx);
-	this->HaloGatherDir(source,YpCompress,Ym,face_idx);
-	this->HaloGatherDir(source,ZpCompress,Zm,face_idx);
-	this->HaloGatherDir(source,TpCompress,Tm,face_idx);
+      assert(same_node[Xp]==this->HaloGatherDir(source,XmCompress,Xp,face_idx));
+      assert(same_node[Yp]==this->HaloGatherDir(source,YmCompress,Yp,face_idx));
+      assert(same_node[Zp]==this->HaloGatherDir(source,ZmCompress,Zp,face_idx));
+      assert(same_node[Tp]==this->HaloGatherDir(source,TmCompress,Tp,face_idx));
+      assert(same_node[Xm]==this->HaloGatherDir(source,XpCompress,Xm,face_idx));
+      assert(same_node[Ym]==this->HaloGatherDir(source,YpCompress,Ym,face_idx));
+      assert(same_node[Zm]==this->HaloGatherDir(source,ZpCompress,Zm,face_idx));
+      assert(same_node[Tm]==this->HaloGatherDir(source,TpCompress,Tm,face_idx));
     }
     this->face_table_computed=1;
     assert(this->u_comm_offset==this->_unified_buffer_size);
@@ -255,6 +349,5 @@ namespace QCD {
 
  };
 
-
 }} // namespace close
 #endif

lib/qcd/action/fermion/WilsonFermion.cc

@@ -230,8 +230,7 @@ void WilsonFermion<Impl>::DerivInternal(StencilImpl &st, DoubledGaugeField &U,
 }
 
 template <class Impl>
-void WilsonFermion<Impl>::DhopDeriv(GaugeField &mat, const FermionField &U,
-                                    const FermionField &V, int dag) {
+void WilsonFermion<Impl>::DhopDeriv(GaugeField &mat, const FermionField &U, const FermionField &V, int dag) {
   conformable(U._grid, _grid);
   conformable(U._grid, V._grid);
   conformable(U._grid, mat._grid);
@@ -242,11 +241,11 @@ void WilsonFermion<Impl>::DhopDeriv(GaugeField &mat, const FermionField &U,
 }
 
 template <class Impl>
-void WilsonFermion<Impl>::DhopDerivOE(GaugeField &mat, const FermionField &U,
-                                      const FermionField &V, int dag) {
+void WilsonFermion<Impl>::DhopDerivOE(GaugeField &mat, const FermionField &U, const FermionField &V, int dag) {
   conformable(U._grid, _cbgrid);
   conformable(U._grid, V._grid);
-  conformable(U._grid, mat._grid);
+  //conformable(U._grid, mat._grid); not general, leaving as a comment (Guido)
+  // Motivation: look at the SchurDiff operator
   
   assert(V.checkerboard == Even);
   assert(U.checkerboard == Odd);
@@ -256,11 +255,10 @@ void WilsonFermion<Impl>::DhopDerivOE(GaugeField &mat, const FermionField &U,
 }
 
 template <class Impl>
-void WilsonFermion<Impl>::DhopDerivEO(GaugeField &mat, const FermionField &U,
-                                      const FermionField &V, int dag) {
+void WilsonFermion<Impl>::DhopDerivEO(GaugeField &mat, const FermionField &U, const FermionField &V, int dag) {
   conformable(U._grid, _cbgrid);
   conformable(U._grid, V._grid);
-  conformable(U._grid, mat._grid);
+  //conformable(U._grid, mat._grid);
 
   assert(V.checkerboard == Odd);
   assert(U.checkerboard == Even);

lib/qcd/action/fermion/WilsonFermion5D.cc

@@ -11,6 +11,7 @@ Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
 Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
 Author: paboyle <paboyle@ph.ed.ac.uk>
+Author: Guido Cossu <guido.cossu@ed.ac.uk>
 
     This program is free software; you can redistribute it and/or modify
     it under the terms of the GNU General Public License as published by
@@ -117,49 +118,19 @@ WilsonFermion5D<Impl>::WilsonFermion5D(GaugeField &_Umu,
     
   // Allocate the required comms buffer
   ImportGauge(_Umu);
+  // Build lists of exterior only nodes
+  int LLs = FiveDimGrid._rdimensions[0];
+  int vol4;
+  vol4=FourDimGrid.oSites();
+  Stencil.BuildSurfaceList(LLs,vol4);
+  vol4=FourDimRedBlackGrid.oSites();
+  StencilEven.BuildSurfaceList(LLs,vol4);
+   StencilOdd.BuildSurfaceList(LLs,vol4);
+
+  std::cout << GridLogMessage << " SurfaceLists "<< Stencil.surface_list.size()
+                       <<" " << StencilEven.surface_list.size()<<std::endl;
+
 }
-  /*
-template<class Impl>
-WilsonFermion5D<Impl>::WilsonFermion5D(int simd,GaugeField &_Umu,
-               GridCartesian         &FiveDimGrid,
-               GridRedBlackCartesian &FiveDimRedBlackGrid,
-               GridCartesian         &FourDimGrid,
-               RealD _M5,const ImplParams &p) :
-{
-  int nsimd = Simd::Nsimd();
-
-  // some assertions
-  assert(FiveDimGrid._ndimension==5);
-  assert(FiveDimRedBlackGrid._ndimension==5);
-  assert(FiveDimRedBlackGrid._checker_dim==0); // Checkerboard the s-direction
-  assert(FourDimGrid._ndimension==4);
-
-  // Dimension zero of the five-d is the Ls direction
-  Ls=FiveDimGrid._fdimensions[0];
-  assert(FiveDimGrid._processors[0]         ==1);
-  assert(FiveDimGrid._simd_layout[0]        ==nsimd);
-
-  assert(FiveDimRedBlackGrid._fdimensions[0]==Ls);
-  assert(FiveDimRedBlackGrid._processors[0] ==1);
-  assert(FiveDimRedBlackGrid._simd_layout[0]==nsimd);
-
-  // Other dimensions must match the decomposition of the four-D fields 
-  for(int d=0;d<4;d++){
-    assert(FiveDimRedBlackGrid._fdimensions[d+1]==FourDimGrid._fdimensions[d]);
-    assert(FiveDimRedBlackGrid._processors[d+1] ==FourDimGrid._processors[d]);
-
-    assert(FourDimGrid._simd_layout[d]=1);
-    assert(FiveDimRedBlackGrid._simd_layout[d+1]==1);
-
-    assert(FiveDimGrid._fdimensions[d+1]        ==FourDimGrid._fdimensions[d]);
-    assert(FiveDimGrid._processors[d+1]         ==FourDimGrid._processors[d]);
-    assert(FiveDimGrid._simd_layout[d+1]        ==FourDimGrid._simd_layout[d]);
-  }
-
-  {
-  }
-}  
-  */
      
 template<class Impl>
 void WilsonFermion5D<Impl>::Report(void)
@@ -296,6 +267,8 @@ void WilsonFermion5D<Impl>::DerivInternal(StencilImpl & st,
   DerivCommTime+=usecond();
 
   Atilde=A;
+  int LLs = B._grid->_rdimensions[0];
+
 
   DerivComputeTime-=usecond();
   for (int mu = 0; mu < Nd; mu++) {
@@ -325,6 +298,9 @@ void WilsonFermion5D<Impl>::DerivInternal(StencilImpl & st,
         ////////////////////////////
       }
     }
+    ////////////////////////////
+    // spin trace outer product
+    ////////////////////////////
     DerivDhopComputeTime += usecond();
     Impl::InsertForce5D(mat, Btilde, Atilde, mu);
   }
@@ -339,7 +315,8 @@ void WilsonFermion5D<Impl>::DhopDeriv(GaugeField &mat,
 {
   conformable(A._grid,FermionGrid());  
   conformable(A._grid,B._grid);
-  conformable(GaugeGrid(),mat._grid);
+
+  //conformable(GaugeGrid(),mat._grid);// this is not general! leaving as a comment
 
   mat.checkerboard = A.checkerboard;
 
@@ -353,7 +330,6 @@ void WilsonFermion5D<Impl>::DhopDerivEO(GaugeField &mat,
                                         int dag)
 {
   conformable(A._grid,FermionRedBlackGrid());
-  conformable(GaugeRedBlackGrid(),mat._grid);
   conformable(A._grid,B._grid);
 
   assert(B.checkerboard==Odd);
@@ -371,7 +347,6 @@ void WilsonFermion5D<Impl>::DhopDerivOE(GaugeField &mat,
                                         int dag)
 {
   conformable(A._grid,FermionRedBlackGrid());
-  conformable(GaugeRedBlackGrid(),mat._grid);
   conformable(A._grid,B._grid);
 
   assert(B.checkerboard==Even);
@@ -396,6 +371,7 @@ void WilsonFermion5D<Impl>::DhopInternal(StencilImpl & st, LebesgueOrder &lo,
   DhopTotalTime+=usecond();
 }
 
+
 template<class Impl>
 void WilsonFermion5D<Impl>::DhopInternalOverlappedComms(StencilImpl & st, LebesgueOrder &lo,
 							DoubledGaugeField & U,
@@ -415,6 +391,15 @@ void WilsonFermion5D<Impl>::DhopInternalOverlappedComms(StencilImpl & st, Lebesg
   DhopFaceTime+=usecond();
   std::vector<std::vector<CommsRequest_t> > reqs;
 
+  // Rely on async comms; start comms before merge of local data
+  DhopCommTime-=usecond();
+  st.CommunicateBegin(reqs);
+
+  DhopFaceTime-=usecond();
+  st.CommsMergeSHM(compressor);
+  DhopFaceTime+=usecond();
+
+  // Perhaps use omp task and region
 #pragma omp parallel 
   { 
     int nthreads = omp_get_num_threads();
@@ -425,8 +410,6 @@ void WilsonFermion5D<Impl>::DhopInternalOverlappedComms(StencilImpl & st, Lebesg
     int sF = LLs * myoff;
 
     if ( me == 0 ) {
-      DhopCommTime-=usecond();
-      st.CommunicateBegin(reqs);
       st.CommunicateComplete(reqs);
       DhopCommTime+=usecond();
     } else { 
@@ -439,28 +422,37 @@ void WilsonFermion5D<Impl>::DhopInternalOverlappedComms(StencilImpl & st, Lebesg
   }
 
   DhopFaceTime-=usecond();
-  st.CommsMerge();
+  st.CommsMerge(compressor);
   DhopFaceTime+=usecond();
 
-#pragma omp parallel 
-  {
-    int nthreads = omp_get_num_threads();
-    int me = omp_get_thread_num();
-    int myoff, mywork;
-
-    GridThread::GetWork(len,me,mywork,myoff,nthreads);
-    int sF = LLs * myoff;
-
-    // Exterior links in stencil
-    if ( me==0 ) DhopComputeTime2-=usecond();
-    if (dag == DaggerYes) Kernels::DhopSiteDag(st,lo,U,st.CommBuf(),sF,myoff,LLs,mywork,in,out,0,1);
-    else                  Kernels::DhopSite   (st,lo,U,st.CommBuf(),sF,myoff,LLs,mywork,in,out,0,1);
-    if ( me==0 ) DhopComputeTime2+=usecond();
-  }// end parallel region
+  // Load imbalance alert. Should use dynamic schedule OMP for loop
+  // Perhaps create a list of only those sites with face work, and 
+  // load balance process the list.
+  DhopComputeTime2-=usecond();
+  if (dag == DaggerYes) {
+    int sz=st.surface_list.size();
+    parallel_for (int ss = 0; ss < sz; ss++) {
+      int sU = st.surface_list[ss];
+      int sF = LLs * sU;
+      Kernels::DhopSiteDag(st,lo,U,st.CommBuf(),sF,sU,LLs,1,in,out,0,1);
+    }
+  } else {
+    int sz=st.surface_list.size();
+    parallel_for (int ss = 0; ss < sz; ss++) {
+      int sU = st.surface_list[ss];
+      int sF = LLs * sU;
+      Kernels::DhopSite(st,lo,U,st.CommBuf(),sF,sU,LLs,1,in,out,0,1);
+    }
+  }
+  DhopComputeTime2+=usecond();
 #else 
   assert(0);
 #endif
+
 }
+
+
+
 template<class Impl>
 void WilsonFermion5D<Impl>::DhopInternalSerialComms(StencilImpl & st, LebesgueOrder &lo,
 					 DoubledGaugeField & U,
@@ -679,7 +671,6 @@ void WilsonFermion5D<Impl>::MomentumSpacePropagatorHw(FermionField &out,const Fe
 
 }
 
-
 FermOpTemplateInstantiate(WilsonFermion5D);
 GparityFermOpTemplateInstantiate(WilsonFermion5D);
   

lib/qcd/action/fermion/WilsonKernels.cc

@@ -33,52 +33,8 @@ directory
 namespace Grid {
 namespace QCD {
 
-  int WilsonKernelsStatic::Opt   = WilsonKernelsStatic::OptGeneric;
-  int WilsonKernelsStatic::Comms = WilsonKernelsStatic::CommsAndCompute;
-
-#ifdef QPX
-#include <spi/include/kernel/location.h>
-#include <spi/include/l1p/types.h>
-#include <hwi/include/bqc/l1p_mmio.h>
-#include <hwi/include/bqc/A2_inlines.h>
-#endif
-
-void bgq_l1p_optimisation(int mode)
-{
-#ifdef QPX
-#undef L1P_CFG_PF_USR
-#define L1P_CFG_PF_USR  (0x3fde8000108ll)   /*  (64 bit reg, 23 bits wide, user/unpriv) */
-
-  uint64_t cfg_pf_usr;
-  if ( mode ) { 
-    cfg_pf_usr =
-        L1P_CFG_PF_USR_ifetch_depth(0)       
-      | L1P_CFG_PF_USR_ifetch_max_footprint(1)   
-      | L1P_CFG_PF_USR_pf_stream_est_on_dcbt 
-      | L1P_CFG_PF_USR_pf_stream_establish_enable
-      | L1P_CFG_PF_USR_pf_stream_optimistic
-      | L1P_CFG_PF_USR_pf_adaptive_throttle(0xF) ;
-    //    if ( sizeof(Float) == sizeof(double) ) {
-      cfg_pf_usr |=  L1P_CFG_PF_USR_dfetch_depth(2)| L1P_CFG_PF_USR_dfetch_max_footprint(3)   ;
-      //    } else {
-      //      cfg_pf_usr |=  L1P_CFG_PF_USR_dfetch_depth(1)| L1P_CFG_PF_USR_dfetch_max_footprint(2)   ;
-      //    }
-  } else { 
-    cfg_pf_usr = L1P_CFG_PF_USR_dfetch_depth(1)
-      | L1P_CFG_PF_USR_dfetch_max_footprint(2)   
-      | L1P_CFG_PF_USR_ifetch_depth(0)       
-      | L1P_CFG_PF_USR_ifetch_max_footprint(1)   
-      | L1P_CFG_PF_USR_pf_stream_est_on_dcbt 
-      | L1P_CFG_PF_USR_pf_stream_establish_enable
-      | L1P_CFG_PF_USR_pf_stream_optimistic
-      | L1P_CFG_PF_USR_pf_stream_prefetch_enable;
-  }
-  *((uint64_t *)L1P_CFG_PF_USR) = cfg_pf_usr;
-
-#endif
-
-}
-
+int WilsonKernelsStatic::Opt   = WilsonKernelsStatic::OptGeneric;
+int WilsonKernelsStatic::Comms = WilsonKernelsStatic::CommsAndCompute;
 
 template <class Impl>
 WilsonKernels<Impl>::WilsonKernels(const ImplParams &p) : Base(p){};
@@ -87,11 +43,71 @@ WilsonKernels<Impl>::WilsonKernels(const ImplParams &p) : Base(p){};
 // Generic implementation; move to different file?
 ////////////////////////////////////////////
   
+#define GENERIC_STENCIL_LEG(Dir,spProj,Recon)			\
+  SE = st.GetEntry(ptype, Dir, sF);				\
+  if (SE->_is_local) {						\
+    chi_p = &chi;						\
+    if (SE->_permute) {						\
+      spProj(tmp, in._odata[SE->_offset]);			\
+      permute(chi, tmp, ptype);					\
+    } else {							\
+      spProj(chi, in._odata[SE->_offset]);			\
+    }								\
+  } else {							\
+    chi_p = &buf[SE->_offset];					\
+  }								\
+  Impl::multLink(Uchi, U._odata[sU], *chi_p, Dir, SE, st);	\
+  Recon(result, Uchi);
+  
+#define GENERIC_STENCIL_LEG_INT(Dir,spProj,Recon)		\
+  SE = st.GetEntry(ptype, Dir, sF);				\
+  if (SE->_is_local) {						\
+    chi_p = &chi;						\
+    if (SE->_permute) {						\
+      spProj(tmp, in._odata[SE->_offset]);			\
+      permute(chi, tmp, ptype);					\
+    } else {							\
+      spProj(chi, in._odata[SE->_offset]);			\
+    }								\
+  } else if ( st.same_node[Dir] ) {				\
+      chi_p = &buf[SE->_offset];				\
+  }								\
+  if (SE->_is_local || st.same_node[Dir] ) {			\
+    Impl::multLink(Uchi, U._odata[sU], *chi_p, Dir, SE, st);	\
+    Recon(result, Uchi);					\
+  }
+
+#define GENERIC_STENCIL_LEG_EXT(Dir,spProj,Recon)		\
+  SE = st.GetEntry(ptype, Dir, sF);				\
+  if ((!SE->_is_local) && (!st.same_node[Dir]) ) {		\
+    chi_p = &buf[SE->_offset];					\
+    Impl::multLink(Uchi, U._odata[sU], *chi_p, Dir, SE, st);	\
+    Recon(result, Uchi);					\
+    nmu++;							\
+  }
+
+#define GENERIC_DHOPDIR_LEG(Dir,spProj,Recon)			\
+  if (gamma == Dir) {						\
+    if (SE->_is_local && SE->_permute) {			\
+      spProj(tmp, in._odata[SE->_offset]);			\
+      permute(chi, tmp, ptype);					\
+    } else if (SE->_is_local) {					\
+      spProj(chi, in._odata[SE->_offset]);			\
+    } else {							\
+      chi = buf[SE->_offset];					\
+    }								\
+    Impl::multLink(Uchi, U._odata[sU], chi, dir, SE, st);	\
+    Recon(result, Uchi);					\
+  }
+
+  ////////////////////////////////////////////////////////////////////
+  // All legs kernels ; comms then compute
+  ////////////////////////////////////////////////////////////////////
 template <class Impl>
 void WilsonKernels<Impl>::GenericDhopSiteDag(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
 					     SiteHalfSpinor *buf, int sF,
-						     int sU, const FermionField &in, FermionField &out,
-						     int interior,int exterior) {
+					     int sU, const FermionField &in, FermionField &out)
+{
   SiteHalfSpinor tmp;
   SiteHalfSpinor chi;
   SiteHalfSpinor *chi_p;
@@ -100,174 +116,22 @@ void WilsonKernels<Impl>::GenericDhopSiteDag(StencilImpl &st, LebesgueOrder &lo,
   StencilEntry *SE;
   int ptype;
 
-  ///////////////////////////
-  // Xp
-  ///////////////////////////
-  SE = st.GetEntry(ptype, Xp, sF);
-
-  if (SE->_is_local) {
-    chi_p = &chi;
-    if (SE->_permute) {
-      spProjXp(tmp, in._odata[SE->_offset]);
-      permute(chi, tmp, ptype);
-    } else {
-      spProjXp(chi, in._odata[SE->_offset]);
-    }
-  } else {
-    chi_p = &buf[SE->_offset];
-  }
-
-  Impl::multLink(Uchi, U._odata[sU], *chi_p, Xp, SE, st);
-  spReconXp(result, Uchi);
-
-  ///////////////////////////
-  // Yp
-  ///////////////////////////
-  SE = st.GetEntry(ptype, Yp, sF);
-
-  if (SE->_is_local) {
-    chi_p = &chi;
-    if (SE->_permute) {
-      spProjYp(tmp, in._odata[SE->_offset]);
-      permute(chi, tmp, ptype);
-    } else {
-      spProjYp(chi, in._odata[SE->_offset]);
-    }
-  } else {
-    chi_p = &buf[SE->_offset];
-  }
-
-  Impl::multLink(Uchi, U._odata[sU], *chi_p, Yp, SE, st);
-  accumReconYp(result, Uchi);
-
-  ///////////////////////////
-  // Zp
-  ///////////////////////////
-  SE = st.GetEntry(ptype, Zp, sF);
-
-  if (SE->_is_local) {
-    chi_p = &chi;
-    if (SE->_permute) {
-      spProjZp(tmp, in._odata[SE->_offset]);
-      permute(chi, tmp, ptype);
-    } else {
-      spProjZp(chi, in._odata[SE->_offset]);
-    }
-  } else {
-    chi_p = &buf[SE->_offset];
-  }
-
-  Impl::multLink(Uchi, U._odata[sU], *chi_p, Zp, SE, st);
-  accumReconZp(result, Uchi);
-
-  ///////////////////////////
-  // Tp
-  ///////////////////////////
-  SE = st.GetEntry(ptype, Tp, sF);
-
-  if (SE->_is_local) {
-    chi_p = &chi;
-    if (SE->_permute) {
-      spProjTp(tmp, in._odata[SE->_offset]);
-      permute(chi, tmp, ptype);
-    } else {
-      spProjTp(chi, in._odata[SE->_offset]);
-    }
-  } else {
-    chi_p = &buf[SE->_offset];
-  }
-
-  Impl::multLink(Uchi, U._odata[sU], *chi_p, Tp, SE, st);
-  accumReconTp(result, Uchi);
-
-  ///////////////////////////
-  // Xm
-  ///////////////////////////
-  SE = st.GetEntry(ptype, Xm, sF);
-
-  if (SE->_is_local) {
-    chi_p = &chi;
-    if (SE->_permute) {
-      spProjXm(tmp, in._odata[SE->_offset]);
-      permute(chi, tmp, ptype);
-    } else {
-      spProjXm(chi, in._odata[SE->_offset]);
-    }
-  } else {
-    chi_p = &buf[SE->_offset];
-  }
-
-  Impl::multLink(Uchi, U._odata[sU], *chi_p, Xm, SE, st);
-  accumReconXm(result, Uchi);
-
-  ///////////////////////////
-  // Ym
-  ///////////////////////////
-  SE = st.GetEntry(ptype, Ym, sF);
-
-  if (SE->_is_local) {
-    chi_p = &chi;
-    if (SE->_permute) {
-      spProjYm(tmp, in._odata[SE->_offset]);
-      permute(chi, tmp, ptype);
-    } else {
-      spProjYm(chi, in._odata[SE->_offset]);
-    }
-  } else {
-    chi_p = &buf[SE->_offset];
-  }
-
-  Impl::multLink(Uchi, U._odata[sU], *chi_p, Ym, SE, st);
-  accumReconYm(result, Uchi);
-
-  ///////////////////////////
-  // Zm
-  ///////////////////////////
-  SE = st.GetEntry(ptype, Zm, sF);
-
-  if (SE->_is_local) {
-    chi_p = &chi;
-    if (SE->_permute) {
-      spProjZm(tmp, in._odata[SE->_offset]);
-      permute(chi, tmp, ptype);
-    } else {
-      spProjZm(chi, in._odata[SE->_offset]);
-    }
-  } else {
-    chi_p = &buf[SE->_offset];
-  }
-
-  Impl::multLink(Uchi, U._odata[sU], *chi_p, Zm, SE, st);
-  accumReconZm(result, Uchi);
-
-  ///////////////////////////
-  // Tm
-  ///////////////////////////
-  SE = st.GetEntry(ptype, Tm, sF);
-
-  if (SE->_is_local) {
-    chi_p = &chi;
-    if (SE->_permute) {
-      spProjTm(tmp, in._odata[SE->_offset]);
-      permute(chi, tmp, ptype);
-    } else {
-      spProjTm(chi, in._odata[SE->_offset]);
-    }
-  } else {
-    chi_p = &buf[SE->_offset];
-  }
-
-  Impl::multLink(Uchi, U._odata[sU], *chi_p, Tm, SE, st);
-  accumReconTm(result, Uchi);
-
+  GENERIC_STENCIL_LEG(Xp,spProjXp,spReconXp);
+  GENERIC_STENCIL_LEG(Yp,spProjYp,accumReconYp);
+  GENERIC_STENCIL_LEG(Zp,spProjZp,accumReconZp);
+  GENERIC_STENCIL_LEG(Tp,spProjTp,accumReconTp);
+  GENERIC_STENCIL_LEG(Xm,spProjXm,accumReconXm);
+  GENERIC_STENCIL_LEG(Ym,spProjYm,accumReconYm);
+  GENERIC_STENCIL_LEG(Zm,spProjZm,accumReconZm);
+  GENERIC_STENCIL_LEG(Tm,spProjTm,accumReconTm);
   vstream(out._odata[sF], result);
 };
 
-// Need controls to do interior, exterior, or both
 template <class Impl>
 void WilsonKernels<Impl>::GenericDhopSite(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
 					  SiteHalfSpinor *buf, int sF,
-						  int sU, const FermionField &in, FermionField &out,int interior,int exterior) {
+					  int sU, const FermionField &in, FermionField &out) 
+{
   SiteHalfSpinor tmp;
   SiteHalfSpinor chi;
   SiteHalfSpinor *chi_p;
@@ -276,168 +140,123 @@ void WilsonKernels<Impl>::GenericDhopSite(StencilImpl &st, LebesgueOrder &lo, Do
   StencilEntry *SE;
   int ptype;
 
-  ///////////////////////////
-  // Xp
-  ///////////////////////////
-  SE = st.GetEntry(ptype, Xm, sF);
-
-  if (SE->_is_local) {
-    chi_p = &chi;
-    if (SE->_permute) {
-      spProjXp(tmp, in._odata[SE->_offset]);
-      permute(chi, tmp, ptype);
-    } else {
-      spProjXp(chi, in._odata[SE->_offset]);
-    }
-  } else {
-    chi_p = &buf[SE->_offset];
-  }
-
-  Impl::multLink(Uchi, U._odata[sU], *chi_p, Xm, SE, st);
-  spReconXp(result, Uchi);
-
-  ///////////////////////////
-  // Yp
-  ///////////////////////////
-  SE = st.GetEntry(ptype, Ym, sF);
-
-  if (SE->_is_local) {
-    chi_p = &chi;
-    if (SE->_permute) {
-      spProjYp(tmp, in._odata[SE->_offset]);
-      permute(chi, tmp, ptype);
-    } else {
-      spProjYp(chi, in._odata[SE->_offset]);
-    }
-  } else {
-    chi_p = &buf[SE->_offset];
-  }
-
-  Impl::multLink(Uchi, U._odata[sU], *chi_p, Ym, SE, st);
-  accumReconYp(result, Uchi);
-
-  ///////////////////////////
-  // Zp
-  ///////////////////////////
-  SE = st.GetEntry(ptype, Zm, sF);
-
-  if (SE->_is_local) {
-    chi_p = &chi;
-    if (SE->_permute) {
-      spProjZp(tmp, in._odata[SE->_offset]);
-      permute(chi, tmp, ptype);
-    } else {
-      spProjZp(chi, in._odata[SE->_offset]);
-    }
-  } else {
-    chi_p = &buf[SE->_offset];
-  }
-
-  Impl::multLink(Uchi, U._odata[sU], *chi_p, Zm, SE, st);
-  accumReconZp(result, Uchi);
-
-  ///////////////////////////
-  // Tp
-  ///////////////////////////
-  SE = st.GetEntry(ptype, Tm, sF);
-
-  if (SE->_is_local) {
-    chi_p = &chi;
-    if (SE->_permute) {
-      spProjTp(tmp, in._odata[SE->_offset]);
-      permute(chi, tmp, ptype);
-    } else {
-      spProjTp(chi, in._odata[SE->_offset]);
-    }
-  } else {
-    chi_p = &buf[SE->_offset];
-  }
-
-  Impl::multLink(Uchi, U._odata[sU], *chi_p, Tm, SE, st);
-  accumReconTp(result, Uchi);
-
-  ///////////////////////////
-  // Xm
-  ///////////////////////////
-  SE = st.GetEntry(ptype, Xp, sF);
-
-  if (SE->_is_local) {
-    chi_p = &chi;
-    if (SE->_permute) {
-      spProjXm(tmp, in._odata[SE->_offset]);
-      permute(chi, tmp, ptype);
-    } else {
-      spProjXm(chi, in._odata[SE->_offset]);
-    }
-  } else {
-    chi_p = &buf[SE->_offset];
-  }
-
-  Impl::multLink(Uchi, U._odata[sU], *chi_p, Xp, SE, st);
-  accumReconXm(result, Uchi);
-
-  ///////////////////////////
-  // Ym
-  ///////////////////////////
-  SE = st.GetEntry(ptype, Yp, sF);
-
-  if (SE->_is_local) {
-    chi_p = &chi;
-    if (SE->_permute) {
-      spProjYm(tmp, in._odata[SE->_offset]);
-      permute(chi, tmp, ptype);
-    } else {
-      spProjYm(chi, in._odata[SE->_offset]);
-    }
-  } else {
-    chi_p = &buf[SE->_offset];
-  }
-
-  Impl::multLink(Uchi, U._odata[sU], *chi_p, Yp, SE, st);
-  accumReconYm(result, Uchi);
-
-  ///////////////////////////
-  // Zm
-  ///////////////////////////
-  SE = st.GetEntry(ptype, Zp, sF);
-
-  if (SE->_is_local) {
-    chi_p = &chi;
-    if (SE->_permute) {
-      spProjZm(tmp, in._odata[SE->_offset]);
-      permute(chi, tmp, ptype);
-    } else {
-      spProjZm(chi, in._odata[SE->_offset]);
-    }
-  } else {
-    chi_p = &buf[SE->_offset];
-  }
-
-  Impl::multLink(Uchi, U._odata[sU], *chi_p, Zp, SE, st);
-  accumReconZm(result, Uchi);
-
-  ///////////////////////////
-  // Tm
-  ///////////////////////////
-  SE = st.GetEntry(ptype, Tp, sF);
-
-  if (SE->_is_local) {
-    chi_p = &chi;
-    if (SE->_permute) {
-      spProjTm(tmp, in._odata[SE->_offset]);
-      permute(chi, tmp, ptype);
-    } else {
-      spProjTm(chi, in._odata[SE->_offset]);
-    }
-  } else {
-    chi_p = &buf[SE->_offset];
-  }
-
-  Impl::multLink(Uchi, U._odata[sU], *chi_p, Tp, SE, st);
-  accumReconTm(result, Uchi);
-
+  GENERIC_STENCIL_LEG(Xm,spProjXp,spReconXp);
+  GENERIC_STENCIL_LEG(Ym,spProjYp,accumReconYp);
+  GENERIC_STENCIL_LEG(Zm,spProjZp,accumReconZp);
+  GENERIC_STENCIL_LEG(Tm,spProjTp,accumReconTp);
+  GENERIC_STENCIL_LEG(Xp,spProjXm,accumReconXm);
+  GENERIC_STENCIL_LEG(Yp,spProjYm,accumReconYm);
+  GENERIC_STENCIL_LEG(Zp,spProjZm,accumReconZm);
+  GENERIC_STENCIL_LEG(Tp,spProjTm,accumReconTm);
   vstream(out._odata[sF], result);
 };
+  ////////////////////////////////////////////////////////////////////
+  // Interior kernels
+  ////////////////////////////////////////////////////////////////////
+template <class Impl>
+void WilsonKernels<Impl>::GenericDhopSiteDagInt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
+						SiteHalfSpinor *buf, int sF,
+						int sU, const FermionField &in, FermionField &out)
+{
+  SiteHalfSpinor tmp;
+  SiteHalfSpinor chi;
+  SiteHalfSpinor *chi_p;
+  SiteHalfSpinor Uchi;
+  SiteSpinor result;
+  StencilEntry *SE;
+  int ptype;
+
+  result=zero;
+  GENERIC_STENCIL_LEG_INT(Xp,spProjXp,accumReconXp);
+  GENERIC_STENCIL_LEG_INT(Yp,spProjYp,accumReconYp);
+  GENERIC_STENCIL_LEG_INT(Zp,spProjZp,accumReconZp);
+  GENERIC_STENCIL_LEG_INT(Tp,spProjTp,accumReconTp);
+  GENERIC_STENCIL_LEG_INT(Xm,spProjXm,accumReconXm);
+  GENERIC_STENCIL_LEG_INT(Ym,spProjYm,accumReconYm);
+  GENERIC_STENCIL_LEG_INT(Zm,spProjZm,accumReconZm);
+  GENERIC_STENCIL_LEG_INT(Tm,spProjTm,accumReconTm);
+  vstream(out._odata[sF], result);
+};
+
+template <class Impl>
+void WilsonKernels<Impl>::GenericDhopSiteInt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
+					     SiteHalfSpinor *buf, int sF,
+					     int sU, const FermionField &in, FermionField &out) 
+{
+  SiteHalfSpinor tmp;
+  SiteHalfSpinor chi;
+  SiteHalfSpinor *chi_p;
+  SiteHalfSpinor Uchi;
+  SiteSpinor result;
+  StencilEntry *SE;
+  int ptype;
+  result=zero;
+  GENERIC_STENCIL_LEG_INT(Xm,spProjXp,accumReconXp);
+  GENERIC_STENCIL_LEG_INT(Ym,spProjYp,accumReconYp);
+  GENERIC_STENCIL_LEG_INT(Zm,spProjZp,accumReconZp);
+  GENERIC_STENCIL_LEG_INT(Tm,spProjTp,accumReconTp);
+  GENERIC_STENCIL_LEG_INT(Xp,spProjXm,accumReconXm);
+  GENERIC_STENCIL_LEG_INT(Yp,spProjYm,accumReconYm);
+  GENERIC_STENCIL_LEG_INT(Zp,spProjZm,accumReconZm);
+  GENERIC_STENCIL_LEG_INT(Tp,spProjTm,accumReconTm);
+  vstream(out._odata[sF], result);
+};
+////////////////////////////////////////////////////////////////////
+// Exterior kernels
+////////////////////////////////////////////////////////////////////
+template <class Impl>
+void WilsonKernels<Impl>::GenericDhopSiteDagExt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
+						SiteHalfSpinor *buf, int sF,
+						int sU, const FermionField &in, FermionField &out)
+{
+  SiteHalfSpinor tmp;
+  SiteHalfSpinor chi;
+  SiteHalfSpinor *chi_p;
+  SiteHalfSpinor Uchi;
+  SiteSpinor result;
+  StencilEntry *SE;
+  int ptype;
+  int nmu=0;
+  result=zero;
+  GENERIC_STENCIL_LEG_EXT(Xp,spProjXp,accumReconXp);
+  GENERIC_STENCIL_LEG_EXT(Yp,spProjYp,accumReconYp);
+  GENERIC_STENCIL_LEG_EXT(Zp,spProjZp,accumReconZp);
+  GENERIC_STENCIL_LEG_EXT(Tp,spProjTp,accumReconTp);
+  GENERIC_STENCIL_LEG_EXT(Xm,spProjXm,accumReconXm);
+  GENERIC_STENCIL_LEG_EXT(Ym,spProjYm,accumReconYm);
+  GENERIC_STENCIL_LEG_EXT(Zm,spProjZm,accumReconZm);
+  GENERIC_STENCIL_LEG_EXT(Tm,spProjTm,accumReconTm);
+  if ( nmu ) { 
+    out._odata[sF] = out._odata[sF] + result; 
+  }
+};
+
+template <class Impl>
+void WilsonKernels<Impl>::GenericDhopSiteExt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
+					     SiteHalfSpinor *buf, int sF,
+					     int sU, const FermionField &in, FermionField &out) 
+{
+  SiteHalfSpinor tmp;
+  SiteHalfSpinor chi;
+  SiteHalfSpinor *chi_p;
+  SiteHalfSpinor Uchi;
+  SiteSpinor result;
+  StencilEntry *SE;
+  int ptype;
+  int nmu=0;
+  result=zero;
+  GENERIC_STENCIL_LEG_EXT(Xm,spProjXp,accumReconXp);
+  GENERIC_STENCIL_LEG_EXT(Ym,spProjYp,accumReconYp);
+  GENERIC_STENCIL_LEG_EXT(Zm,spProjZp,accumReconZp);
+  GENERIC_STENCIL_LEG_EXT(Tm,spProjTp,accumReconTp);
+  GENERIC_STENCIL_LEG_EXT(Xp,spProjXm,accumReconXm);
+  GENERIC_STENCIL_LEG_EXT(Yp,spProjYm,accumReconYm);
+  GENERIC_STENCIL_LEG_EXT(Zp,spProjZm,accumReconZm);
+  GENERIC_STENCIL_LEG_EXT(Tp,spProjTm,accumReconTm);
+  if ( nmu ) { 
+    out._odata[sF] = out._odata[sF] + result; 
+  }
+};
 
 template <class Impl>
 void WilsonKernels<Impl>::DhopDir( StencilImpl &st, DoubledGaugeField &U,SiteHalfSpinor *buf, int sF,
@@ -451,119 +270,14 @@ void WilsonKernels<Impl>::DhopDir( StencilImpl &st, DoubledGaugeField &U,SiteHal
   int ptype;
 
   SE = st.GetEntry(ptype, dir, sF);
-
-  // Xp
-  if (gamma == Xp) {
-    if (SE->_is_local && SE->_permute) {
-      spProjXp(tmp, in._odata[SE->_offset]);
-      permute(chi, tmp, ptype);
-    } else if (SE->_is_local) {
-      spProjXp(chi, in._odata[SE->_offset]);
-    } else {
-      chi = buf[SE->_offset];
-    }
-    Impl::multLink(Uchi, U._odata[sU], chi, dir, SE, st);
-    spReconXp(result, Uchi);
-  }
-
-  // Yp
-  if (gamma == Yp) {
-    if (SE->_is_local && SE->_permute) {
-      spProjYp(tmp, in._odata[SE->_offset]);
-      permute(chi, tmp, ptype);
-    } else if (SE->_is_local) {
-      spProjYp(chi, in._odata[SE->_offset]);
-    } else {
-      chi = buf[SE->_offset];
-    }
-    Impl::multLink(Uchi, U._odata[sU], chi, dir, SE, st);
-    spReconYp(result, Uchi);
-  }
-
-  // Zp
-  if (gamma == Zp) {
-    if (SE->_is_local && SE->_permute) {
-      spProjZp(tmp, in._odata[SE->_offset]);
-      permute(chi, tmp, ptype);
-    } else if (SE->_is_local) {
-      spProjZp(chi, in._odata[SE->_offset]);
-    } else {
-      chi = buf[SE->_offset];
-    }
-    Impl::multLink(Uchi, U._odata[sU], chi, dir, SE, st);
-    spReconZp(result, Uchi);
-  }
-
-  // Tp
-  if (gamma == Tp) {
-    if (SE->_is_local && SE->_permute) {
-      spProjTp(tmp, in._odata[SE->_offset]);
-      permute(chi, tmp, ptype);
-    } else if (SE->_is_local) {
-      spProjTp(chi, in._odata[SE->_offset]);
-    } else {
-      chi = buf[SE->_offset];
-    }
-    Impl::multLink(Uchi, U._odata[sU], chi, dir, SE, st);
-    spReconTp(result, Uchi);
-  }
-
-  // Xm
-  if (gamma == Xm) {
-    if (SE->_is_local && SE->_permute) {
-      spProjXm(tmp, in._odata[SE->_offset]);
-      permute(chi, tmp, ptype);
-    } else if (SE->_is_local) {
-      spProjXm(chi, in._odata[SE->_offset]);
-    } else {
-      chi = buf[SE->_offset];
-    }
-    Impl::multLink(Uchi, U._odata[sU], chi, dir, SE, st);
-    spReconXm(result, Uchi);
-  }
-
-  // Ym
-  if (gamma == Ym) {
-    if (SE->_is_local && SE->_permute) {
-      spProjYm(tmp, in._odata[SE->_offset]);
-      permute(chi, tmp, ptype);
-    } else if (SE->_is_local) {
-      spProjYm(chi, in._odata[SE->_offset]);
-    } else {
-      chi = buf[SE->_offset];
-    }
-    Impl::multLink(Uchi, U._odata[sU], chi, dir, SE, st);
-    spReconYm(result, Uchi);
-  }
-
-  // Zm
-  if (gamma == Zm) {
-    if (SE->_is_local && SE->_permute) {
-      spProjZm(tmp, in._odata[SE->_offset]);
-      permute(chi, tmp, ptype);
-    } else if (SE->_is_local) {
-      spProjZm(chi, in._odata[SE->_offset]);
-    } else {
-      chi = buf[SE->_offset];
-    }
-    Impl::multLink(Uchi, U._odata[sU], chi, dir, SE, st);
-    spReconZm(result, Uchi);
-  }
-
-  // Tm
-  if (gamma == Tm) {
-    if (SE->_is_local && SE->_permute) {
-      spProjTm(tmp, in._odata[SE->_offset]);
-      permute(chi, tmp, ptype);
-    } else if (SE->_is_local) {
-      spProjTm(chi, in._odata[SE->_offset]);
-    } else {
-      chi = buf[SE->_offset];
-    }
-    Impl::multLink(Uchi, U._odata[sU], chi, dir, SE, st);
-    spReconTm(result, Uchi);
-  }
-
+  GENERIC_DHOPDIR_LEG(Xp,spProjXp,spReconXp);
+  GENERIC_DHOPDIR_LEG(Yp,spProjYp,spReconYp);
+  GENERIC_DHOPDIR_LEG(Zp,spProjZp,spReconZp);
+  GENERIC_DHOPDIR_LEG(Tp,spProjTp,spReconTp);
+  GENERIC_DHOPDIR_LEG(Xm,spProjXm,spReconXm);
+  GENERIC_DHOPDIR_LEG(Ym,spProjYm,spReconYm);
+  GENERIC_DHOPDIR_LEG(Zm,spProjZm,spReconZm);
+  GENERIC_DHOPDIR_LEG(Tm,spProjTm,spReconTm);
   vstream(out._odata[sF], result);
 }
 

lib/qcd/action/fermion/WilsonKernels.h

@@ -34,8 +34,6 @@ directory
 namespace Grid {
 namespace QCD {
 
-void bgq_l1p_optimisation(int mode);
-
   ////////////////////////////////////////////////////////////////////////////////////////////////////////////////
   // Helper routines that implement Wilson stencil for a single site.
   // Common to both the WilsonFermion and WilsonFermion5D
@@ -44,9 +42,8 @@ class WilsonKernelsStatic {
  public:
   enum { OptGeneric, OptHandUnroll, OptInlineAsm };
   enum { CommsAndCompute, CommsThenCompute };
-  // S-direction is INNERMOST and takes no part in the parity.
-  static int Opt;  // these are a temporary hack
-  static int Comms;  // these are a temporary hack
+  static int Opt;  
+  static int Comms;
 };
  
 template<class Impl> class WilsonKernels : public FermionOperator<Impl> , public WilsonKernelsStatic { 
@@ -66,7 +63,7 @@ public:
     switch(Opt) {
 #if defined(AVX512) || defined (QPX)
     case OptInlineAsm:
-      if(interior&&exterior) WilsonKernels<Impl>::AsmDhopSite(st,lo,U,buf,sF,sU,Ls,Ns,in,out);
+      if(interior&&exterior) WilsonKernels<Impl>::AsmDhopSite   (st,lo,U,buf,sF,sU,Ls,Ns,in,out);
       else if (interior)     WilsonKernels<Impl>::AsmDhopSiteInt(st,lo,U,buf,sF,sU,Ls,Ns,in,out);
       else if (exterior)     WilsonKernels<Impl>::AsmDhopSiteExt(st,lo,U,buf,sF,sU,Ls,Ns,in,out);
       else assert(0);
@@ -75,7 +72,9 @@ public:
     case OptHandUnroll:
       for (int site = 0; site < Ns; site++) {
 	for (int s = 0; s < Ls; s++) {
-	  if( exterior) WilsonKernels<Impl>::HandDhopSite(st,lo,U,buf,sF,sU,in,out,interior,exterior);
+	  if(interior&&exterior) WilsonKernels<Impl>::HandDhopSite(st,lo,U,buf,sF,sU,in,out);
+	  else if (interior)     WilsonKernels<Impl>::HandDhopSiteInt(st,lo,U,buf,sF,sU,in,out);
+	  else if (exterior)     WilsonKernels<Impl>::HandDhopSiteExt(st,lo,U,buf,sF,sU,in,out);
 	  sF++;
 	}
 	sU++;
@@ -84,7 +83,10 @@ public:
     case OptGeneric:
       for (int site = 0; site < Ns; site++) {
 	for (int s = 0; s < Ls; s++) {
-	  if( exterior) WilsonKernels<Impl>::GenericDhopSite(st,lo,U,buf,sF,sU,in,out,interior,exterior);
+	  if(interior&&exterior) WilsonKernels<Impl>::GenericDhopSite(st,lo,U,buf,sF,sU,in,out);
+	  else if (interior)     WilsonKernels<Impl>::GenericDhopSiteInt(st,lo,U,buf,sF,sU,in,out);
+	  else if (exterior)     WilsonKernels<Impl>::GenericDhopSiteExt(st,lo,U,buf,sF,sU,in,out);
+	  else assert(0);
 	  sF++;
 	}
 	sU++;
@@ -103,7 +105,10 @@ public:
     // no kernel choice  
     for (int site = 0; site < Ns; site++) {
       for (int s = 0; s < Ls; s++) {
-	if( exterior) WilsonKernels<Impl>::GenericDhopSite(st, lo, U, buf, sF, sU, in, out,interior,exterior);
+	if(interior&&exterior) WilsonKernels<Impl>::GenericDhopSite(st,lo,U,buf,sF,sU,in,out);
+	else if (interior)     WilsonKernels<Impl>::GenericDhopSiteInt(st,lo,U,buf,sF,sU,in,out);
+	else if (exterior)     WilsonKernels<Impl>::GenericDhopSiteExt(st,lo,U,buf,sF,sU,in,out);
+	else assert(0);
 	sF++;
       }
       sU++;
@@ -113,13 +118,13 @@ public:
   template <bool EnableBool = true>
   typename std::enable_if<Impl::Dimension == 3 && Nc == 3 && EnableBool,void>::type
   DhopSiteDag(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
-		      int sF, int sU, int Ls, int Ns, const FermionField &in, FermionField &out,int interior=1,int exterior=1) {
-
+	      int sF, int sU, int Ls, int Ns, const FermionField &in, FermionField &out,int interior=1,int exterior=1) 
+{
     bgq_l1p_optimisation(1);
     switch(Opt) {
 #if defined(AVX512) || defined (QPX)
     case OptInlineAsm:
-      if(interior&&exterior) WilsonKernels<Impl>::AsmDhopSiteDag(st,lo,U,buf,sF,sU,Ls,Ns,in,out);
+      if(interior&&exterior) WilsonKernels<Impl>::AsmDhopSiteDag   (st,lo,U,buf,sF,sU,Ls,Ns,in,out);
       else if (interior)     WilsonKernels<Impl>::AsmDhopSiteDagInt(st,lo,U,buf,sF,sU,Ls,Ns,in,out);
       else if (exterior)     WilsonKernels<Impl>::AsmDhopSiteDagExt(st,lo,U,buf,sF,sU,Ls,Ns,in,out);
       else assert(0);
@@ -128,7 +133,10 @@ public:
     case OptHandUnroll:
       for (int site = 0; site < Ns; site++) {
 	for (int s = 0; s < Ls; s++) {
-	  if( exterior) WilsonKernels<Impl>::HandDhopSiteDag(st,lo,U,buf,sF,sU,in,out,interior,exterior);
+	  if(interior&&exterior) WilsonKernels<Impl>::HandDhopSiteDag(st,lo,U,buf,sF,sU,in,out);
+	  else if (interior)     WilsonKernels<Impl>::HandDhopSiteDagInt(st,lo,U,buf,sF,sU,in,out);
+	  else if (exterior)     WilsonKernels<Impl>::HandDhopSiteDagExt(st,lo,U,buf,sF,sU,in,out);
+	  else assert(0);
 	  sF++;
 	}
 	sU++;
@@ -137,7 +145,10 @@ public:
     case OptGeneric:
       for (int site = 0; site < Ns; site++) {
 	for (int s = 0; s < Ls; s++) {
-	  if( exterior) WilsonKernels<Impl>::GenericDhopSiteDag(st,lo,U,buf,sF,sU,in,out,interior,exterior);
+	  if(interior&&exterior) WilsonKernels<Impl>::GenericDhopSiteDag(st,lo,U,buf,sF,sU,in,out);
+	  else if (interior)     WilsonKernels<Impl>::GenericDhopSiteDagInt(st,lo,U,buf,sF,sU,in,out);
+	  else if (exterior)     WilsonKernels<Impl>::GenericDhopSiteDagExt(st,lo,U,buf,sF,sU,in,out);
+	  else assert(0);
 	  sF++;
 	}
 	sU++;
@@ -156,7 +167,10 @@ public:
 
     for (int site = 0; site < Ns; site++) {
       for (int s = 0; s < Ls; s++) {
-	if( exterior) WilsonKernels<Impl>::GenericDhopSiteDag(st,lo,U,buf,sF,sU,in,out,interior,exterior);
+	if(interior&&exterior) WilsonKernels<Impl>::GenericDhopSiteDag(st,lo,U,buf,sF,sU,in,out);
+	else if (interior)     WilsonKernels<Impl>::GenericDhopSiteDagInt(st,lo,U,buf,sF,sU,in,out);
+	else if (exterior)     WilsonKernels<Impl>::GenericDhopSiteDagExt(st,lo,U,buf,sF,sU,in,out);
+	else assert(0);
 	sF++;
       }
       sU++;
@@ -169,10 +183,22 @@ public:
 private:
      // Specialised variants
   void GenericDhopSite(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
-			       int sF, int sU, const FermionField &in, FermionField &out,int interior,int exterior);
+		       int sF, int sU, const FermionField &in, FermionField &out);
       
   void GenericDhopSiteDag(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
-				  int sF, int sU, const FermionField &in, FermionField &out,int interior,int exterior);
+			  int sF, int sU, const FermionField &in, FermionField &out);
+
+  void GenericDhopSiteInt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
+			  int sF, int sU, const FermionField &in, FermionField &out);
+      
+  void GenericDhopSiteDagInt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
+			     int sF, int sU, const FermionField &in, FermionField &out);
+
+  void GenericDhopSiteExt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
+			  int sF, int sU, const FermionField &in, FermionField &out);
+      
+  void GenericDhopSiteDagExt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
+			     int sF, int sU, const FermionField &in, FermionField &out);
 
   void AsmDhopSite(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
 		   int sF, int sU, int Ls, int Ns, const FermionField &in,FermionField &out);
@@ -194,10 +220,22 @@ private:
 
 
   void HandDhopSite(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
-			    int sF, int sU, const FermionField &in, FermionField &out,int interior,int exterior);
+		    int sF, int sU, const FermionField &in, FermionField &out);
 
   void HandDhopSiteDag(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
-			       int sF, int sU, const FermionField &in, FermionField &out,int interior,int exterior);
+		       int sF, int sU, const FermionField &in, FermionField &out);
+      
+  void HandDhopSiteInt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
+		       int sF, int sU, const FermionField &in, FermionField &out);
+  
+  void HandDhopSiteDagInt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
+			  int sF, int sU, const FermionField &in, FermionField &out);
+  
+  void HandDhopSiteExt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
+		       int sF, int sU, const FermionField &in, FermionField &out);
+  
+  void HandDhopSiteDagExt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
+			  int sF, int sU, const FermionField &in, FermionField &out);
   
 public:
 

lib/qcd/action/fermion/WilsonKernelsAsm.cc

@@ -112,5 +112,16 @@ INSTANTIATE_ASM(DomainWallVec5dImplD);
 INSTANTIATE_ASM(ZDomainWallVec5dImplF);
 INSTANTIATE_ASM(ZDomainWallVec5dImplD);
 
+INSTANTIATE_ASM(WilsonImplFH);
+INSTANTIATE_ASM(WilsonImplDF);
+INSTANTIATE_ASM(ZWilsonImplFH);
+INSTANTIATE_ASM(ZWilsonImplDF);
+INSTANTIATE_ASM(GparityWilsonImplFH);
+INSTANTIATE_ASM(GparityWilsonImplDF);
+INSTANTIATE_ASM(DomainWallVec5dImplFH);
+INSTANTIATE_ASM(DomainWallVec5dImplDF);
+INSTANTIATE_ASM(ZDomainWallVec5dImplFH);
+INSTANTIATE_ASM(ZDomainWallVec5dImplDF);
+
 }}
 

lib/qcd/action/fermion/WilsonKernelsAsmAvx512.h

@@ -71,6 +71,16 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSite(StencilImpl &st,LebesgueOrder & lo,Doub
 						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
 #include <qcd/action/fermion/WilsonKernelsAsmBody.h>
 
+template<> void 
+WilsonKernels<WilsonImplFH>::AsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+
+template<> void 
+WilsonKernels<ZWilsonImplFH>::AsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+
 #undef INTERIOR_AND_EXTERIOR
 #define INTERIOR
 #undef EXTERIOR
@@ -84,6 +94,16 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteInt(StencilImpl &st,LebesgueOrder & lo,D
 						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
 #include <qcd/action/fermion/WilsonKernelsAsmBody.h>
 
+template<> void 
+WilsonKernels<WilsonImplFH>::AsmDhopSiteInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+
+template<> void 
+WilsonKernels<ZWilsonImplFH>::AsmDhopSiteInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+
 
 #undef INTERIOR_AND_EXTERIOR
 #undef INTERIOR
@@ -98,6 +118,16 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteExt(StencilImpl &st,LebesgueOrder & lo,D
 						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
 #include <qcd/action/fermion/WilsonKernelsAsmBody.h>
 
+template<> void 
+WilsonKernels<WilsonImplFH>::AsmDhopSiteExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+
+template<> void 
+WilsonKernels<ZWilsonImplFH>::AsmDhopSiteExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+      
 /////////////////////////////////////////////////////////////////
 // XYZT vectorised, dag Kernel, single
 /////////////////////////////////////////////////////////////////
@@ -115,6 +145,16 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,D
 						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
 #include <qcd/action/fermion/WilsonKernelsAsmBody.h>
 
+template<> void 
+WilsonKernels<WilsonImplFH>::AsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+
+template<> void 
+WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+
 #undef INTERIOR_AND_EXTERIOR
 #define INTERIOR
 #undef EXTERIOR
@@ -128,6 +168,16 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteDagInt(StencilImpl &st,LebesgueOrder & l
 						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
 #include <qcd/action/fermion/WilsonKernelsAsmBody.h>
 
+template<> void 
+WilsonKernels<WilsonImplFH>::AsmDhopSiteDagInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+
+template<> void 
+WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDagInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+
 #undef INTERIOR_AND_EXTERIOR
 #undef INTERIOR
 #define EXTERIOR
@@ -141,6 +191,16 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteDagExt(StencilImpl &st,LebesgueOrder & l
 						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
 #include <qcd/action/fermion/WilsonKernelsAsmBody.h>
 				    
+template<> void 
+WilsonKernels<WilsonImplFH>::AsmDhopSiteDagExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+				    
+template<> void 
+WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDagExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+				    
 #undef MAYBEPERM
 #undef MULT_2SPIN
 #define MAYBEPERM(A,B) 
@@ -162,6 +222,15 @@ WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSite(StencilImpl &st,LebesgueOrder
 							 int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
 #include <qcd/action/fermion/WilsonKernelsAsmBody.h>
 
+template<> void 
+WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+
 #undef INTERIOR_AND_EXTERIOR
 #define INTERIOR
 #undef EXTERIOR
@@ -174,6 +243,15 @@ WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSiteInt(StencilImpl &st,LebesgueOrd
 							 int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
 #include <qcd/action/fermion/WilsonKernelsAsmBody.h>
 
+template<> void 
+WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+
 #undef INTERIOR_AND_EXTERIOR
 #undef INTERIOR
 #define EXTERIOR
@@ -189,6 +267,16 @@ WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSiteExt(StencilImpl &st,LebesgueOrd
 							 int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
 #include <qcd/action/fermion/WilsonKernelsAsmBody.h>
 				    
+template<> void 
+WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+				    
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+				    
 /////////////////////////////////////////////////////////////////
 // Ls vectorised, dag Kernel, single
 /////////////////////////////////////////////////////////////////
@@ -205,6 +293,15 @@ WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSiteDag(StencilImpl &st,LebesgueOrd
 							    int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
 #include <qcd/action/fermion/WilsonKernelsAsmBody.h>
 
+template<> void 
+WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+
 #undef INTERIOR_AND_EXTERIOR
 #define INTERIOR
 #undef EXTERIOR
@@ -217,6 +314,15 @@ WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSiteDagInt(StencilImpl &st,Lebesgue
 							    int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
 #include <qcd/action/fermion/WilsonKernelsAsmBody.h>
 
+template<> void 
+WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteDagInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteDagInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+
 #undef INTERIOR_AND_EXTERIOR
 #undef INTERIOR
 #define EXTERIOR
@@ -229,6 +335,15 @@ WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSiteDagExt(StencilImpl &st,Lebesgue
 							    int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
 #include <qcd/action/fermion/WilsonKernelsAsmBody.h>
 
+template<> void 
+WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteDagExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteDagExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+
 #undef COMPLEX_SIGNS
 #undef MAYBEPERM
 #undef MULT_2SPIN
@@ -269,6 +384,15 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSite(StencilImpl &st,LebesgueOrder & lo,Doub
 						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
 #include <qcd/action/fermion/WilsonKernelsAsmBody.h>
 
+template<> void 
+WilsonKernels<WilsonImplDF>::AsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZWilsonImplDF>::AsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+
 #undef INTERIOR_AND_EXTERIOR
 #define INTERIOR
 #undef EXTERIOR
@@ -281,6 +405,15 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteInt(StencilImpl &st,LebesgueOrder & lo,D
 						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
 #include <qcd/action/fermion/WilsonKernelsAsmBody.h>
 
+template<> void 
+WilsonKernels<WilsonImplDF>::AsmDhopSiteInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZWilsonImplDF>::AsmDhopSiteInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+
 #undef INTERIOR_AND_EXTERIOR
 #undef INTERIOR
 #define EXTERIOR
@@ -293,6 +426,15 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteExt(StencilImpl &st,LebesgueOrder & lo,D
 						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
 #include <qcd/action/fermion/WilsonKernelsAsmBody.h>
       
+template<> void 
+WilsonKernels<WilsonImplDF>::AsmDhopSiteExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZWilsonImplDF>::AsmDhopSiteExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+      
 /////////////////////////////////////////////////////////////////
 // XYZT vectorised, dag Kernel, single
 /////////////////////////////////////////////////////////////////
@@ -309,6 +451,15 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,D
 						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
 #include <qcd/action/fermion/WilsonKernelsAsmBody.h>
 
+template<> void 
+WilsonKernels<WilsonImplDF>::AsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+
 #undef INTERIOR_AND_EXTERIOR
 #define INTERIOR
 #undef EXTERIOR
@@ -321,6 +472,15 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteDagInt(StencilImpl &st,LebesgueOrder & l
 						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
 #include <qcd/action/fermion/WilsonKernelsAsmBody.h>
 
+template<> void 
+WilsonKernels<WilsonImplDF>::AsmDhopSiteDagInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+
 #undef INTERIOR_AND_EXTERIOR
 #undef INTERIOR
 #define EXTERIOR
@@ -333,6 +493,15 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteDagExt(StencilImpl &st,LebesgueOrder & l
 						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
 #include <qcd/action/fermion/WilsonKernelsAsmBody.h>
 				    
+template<> void 
+WilsonKernels<WilsonImplDF>::AsmDhopSiteDagExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+						int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+				    
 #undef MAYBEPERM
 #undef MULT_2SPIN
 #define MAYBEPERM(A,B) 
@@ -354,6 +523,15 @@ WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSite(StencilImpl &st,LebesgueOrder
 							 int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
 #include <qcd/action/fermion/WilsonKernelsAsmBody.h>
 
+template<> void 
+WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+
 #undef INTERIOR_AND_EXTERIOR
 #define INTERIOR
 #undef EXTERIOR
@@ -366,6 +544,15 @@ WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSiteInt(StencilImpl &st,LebesgueOrd
 							 int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
 #include <qcd/action/fermion/WilsonKernelsAsmBody.h>
 
+template<> void 
+WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+
 #undef INTERIOR_AND_EXTERIOR
 #undef INTERIOR
 #define EXTERIOR
@@ -380,6 +567,15 @@ WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSiteExt(StencilImpl &st,LebesgueOrd
 							 int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
 #include <qcd/action/fermion/WilsonKernelsAsmBody.h>
 				    
+template<> void 
+WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
+							 int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+				    
 /////////////////////////////////////////////////////////////////
 // Ls vectorised, dag Kernel, single
 /////////////////////////////////////////////////////////////////
@@ -396,6 +592,15 @@ WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSiteDag(StencilImpl &st,LebesgueOrd
 							    int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
 #include <qcd/action/fermion/WilsonKernelsAsmBody.h>
 
+template<> void 
+WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+
 #undef INTERIOR_AND_EXTERIOR
 #define INTERIOR
 #undef EXTERIOR
@@ -408,6 +613,15 @@ WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSiteDagInt(StencilImpl &st,Lebesgue
 							    int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
 #include <qcd/action/fermion/WilsonKernelsAsmBody.h>
 
+template<> void 
+WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteDagInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteDagInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+
 #undef INTERIOR_AND_EXTERIOR
 #undef INTERIOR
 #define EXTERIOR
@@ -420,6 +634,15 @@ WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSiteDagExt(StencilImpl &st,Lebesgue
 							    int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
 #include <qcd/action/fermion/WilsonKernelsAsmBody.h>
 
+template<> void 
+WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteDagExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+template<> void 
+WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteDagExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
+							    int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
+#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
+
 #undef COMPLEX_SIGNS
 #undef MAYBEPERM
 #undef MULT_2SPIN

lib/qcd/action/fermion/WilsonKernelsAsmBody.h

@@ -39,24 +39,26 @@
 ////////////////////////////////////////////////////////////////////////////////
 #ifdef INTERIOR_AND_EXTERIOR
 
-#define ZERO_NMU(A) 
-#define INTERIOR_BLOCK_XP(a,b,PERMUTE_DIR,PROJMEM,RECON) INTERIOR_BLOCK(a,b,PERMUTE_DIR,PROJMEM,RECON)
-#define EXTERIOR_BLOCK_XP(a,b,RECON) EXTERIOR_BLOCK(a,b,RECON)
-
-#define INTERIOR_BLOCK(a,b,PERMUTE_DIR,PROJMEM,RECON)	\
+#define ASM_LEG(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON)			\
+      basep = st.GetPFInfo(nent,plocal); nent++;			\
+      if ( local ) {							\
 	LOAD64(%r10,isigns);						\
-  PROJMEM(base);                                        \
-  MAYBEPERM(PERMUTE_DIR,perm);                                  
-
-#define EXTERIOR_BLOCK(a,b,RECON)             \
-  LOAD_CHI(base);
-
-#define COMMON_BLOCK(a,b,RECON)               \
-  base = st.GetInfo(ptype,local,perm,b,ent,plocal); ent++;     \
+	PROJ(base);							\
+	MAYBEPERM(PERMUTE_DIR,perm);					\
+      } else {								\
+	LOAD_CHI(base);							\
+      }									\
+      base = st.GetInfo(ptype,local,perm,NxtDir,ent,plocal); ent++;	\
       PREFETCH_CHIMU(base);						\
-  MULT_2SPIN_DIR_PF(a,basep);					\
+      MULT_2SPIN_DIR_PF(Dir,basep);					\
       LOAD64(%r10,isigns);						\
-  RECON;                                                        
+      RECON;								\
+
+#define ASM_LEG_XP(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON)			\
+  base = st.GetInfo(ptype,local,perm,Dir,ent,plocal); ent++;		\
+  PF_GAUGE(Xp);								\
+  PREFETCH1_CHIMU(base);						\
+  ASM_LEG(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON) 
 
 #define RESULT(base,basep) SAVE_RESULT(base,basep);
 
@@ -67,62 +69,62 @@
 ////////////////////////////////////////////////////////////////////////////////
 #ifdef INTERIOR
 
-#define COMMON_BLOCK(a,b,RECON)       
-#define ZERO_NMU(A) 
-
-// No accumulate for DIR0
-#define EXTERIOR_BLOCK_XP(a,b,RECON)				\
-  ZERO_PSI;							\
-  base = st.GetInfo(ptype,local,perm,b,ent,plocal); ent++;	
-
-#define EXTERIOR_BLOCK(a,b,RECON)  \
-  base = st.GetInfo(ptype,local,perm,b,ent,plocal); ent++;     
-
-#define INTERIOR_BLOCK_XP(a,b,PERMUTE_DIR,PROJMEM,RECON) INTERIOR_BLOCK(a,b,PERMUTE_DIR,PROJMEM,RECON)
-
-#define INTERIOR_BLOCK(a,b,PERMUTE_DIR,PROJMEM,RECON)		\
+#define ASM_LEG(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON)			\
+      basep = st.GetPFInfo(nent,plocal); nent++;			\
+      if ( local ) {							\
 	LOAD64(%r10,isigns);						\
-  PROJMEM(base);                                                \
+	PROJ(base);							\
 	MAYBEPERM(PERMUTE_DIR,perm);					\
-  base = st.GetInfo(ptype,local,perm,b,ent,plocal); ent++;	\
-  PREFETCH_CHIMU(base);						\
-  MULT_2SPIN_DIR_PF(a,basep);					\
+      }else if ( st.same_node[Dir] ) {LOAD_CHI(base);}			\
+      if ( local || st.same_node[Dir] ) {				\
+	MULT_2SPIN_DIR_PF(Dir,basep);					\
 	LOAD64(%r10,isigns);						\
-  RECON;                                                        
+	RECON;								\
+      }									\
+      base = st.GetInfo(ptype,local,perm,NxtDir,ent,plocal); ent++;	\
+      PREFETCH_CHIMU(base);						\
+
+#define ASM_LEG_XP(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON)			\
+  base = st.GetInfo(ptype,local,perm,Dir,ent,plocal); ent++;		\
+  PF_GAUGE(Xp);								\
+  PREFETCH1_CHIMU(base);						\
+  { ZERO_PSI; }								\
+  ASM_LEG(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON) 
 
 #define RESULT(base,basep) SAVE_RESULT(base,basep);
 
 #endif
-
 ////////////////////////////////////////////////////////////////////////////////
 // Post comms kernel
 ////////////////////////////////////////////////////////////////////////////////
 #ifdef EXTERIOR
 
-#define ZERO_NMU(A) nmu=0;
 
-#define INTERIOR_BLOCK_XP(a,b,PERMUTE_DIR,PROJMEM,RECON) \
-  ZERO_PSI;   base = st.GetInfo(ptype,local,perm,b,ent,plocal); ent++;		
-
-#define EXTERIOR_BLOCK_XP(a,b,RECON) EXTERIOR_BLOCK(a,b,RECON)
-
-#define INTERIOR_BLOCK(a,b,PERMUTE_DIR,PROJMEM,RECON)			\
-  base = st.GetInfo(ptype,local,perm,b,ent,plocal); ent++;		
-
-#define EXTERIOR_BLOCK(a,b,RECON)				\
-    nmu++;							\
+#define ASM_LEG(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON)			\
+  base = st.GetInfo(ptype,local,perm,Dir,ent,plocal); ent++;		\
+  if((!local)&&(!st.same_node[Dir]) ) {					\
     LOAD_CHI(base);							\
-    MULT_2SPIN_DIR_PF(a,base);					\
-    base = st.GetInfo(ptype,local,perm,b,ent,plocal); ent++;	\
+    MULT_2SPIN_DIR_PF(Dir,base);					\
     LOAD64(%r10,isigns);						\
-    RECON;                                                        
+    RECON;								\
+    nmu++;								\
+  }									
 
-#define COMMON_BLOCK(a,b,RECON)			
+#define ASM_LEG_XP(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON)			\
+  nmu=0;								\
+  { ZERO_PSI;}								\
+  base = st.GetInfo(ptype,local,perm,Dir,ent,plocal); ent++;		\
+  if((!local)&&(!st.same_node[Dir]) ) {					\
+    LOAD_CHI(base);							\
+    MULT_2SPIN_DIR_PF(Dir,base);					\
+    LOAD64(%r10,isigns);						\
+    RECON;								\
+    nmu++;								\
+  }
 
 #define RESULT(base,basep) if (nmu){ ADD_RESULT(base,base);}
 
 #endif
-
 {
   int nmu;
   int local,perm, ptype;
@@ -134,11 +136,15 @@
   MASK_REGS;
   int nmax=U._grid->oSites();
   for(int site=0;site<Ns;site++) {
+#ifndef EXTERIOR
     int sU =lo.Reorder(ssU);
     int ssn=ssU+1;     if(ssn>=nmax) ssn=0;
     int sUn=lo.Reorder(ssn);
-#ifndef EXTERIOR
     LOCK_GAUGE(0);
+#else
+    int sU =ssU;
+    int ssn=ssU+1;     if(ssn>=nmax) ssn=0;
+    int sUn=ssn;
 #endif
     for(int s=0;s<Ls;s++) {
       ss =sU*Ls+s;
@@ -146,93 +152,20 @@
       int  ent=ss*8;// 2*Ndim
       int nent=ssn*8;
 
-      ZERO_NMU(0);
-      base  = st.GetInfo(ptype,local,perm,Xp,ent,plocal); ent++;
-#ifndef EXTERIOR
-      PF_GAUGE(Xp); 
-      PREFETCH1_CHIMU(base);
-#endif
-      ////////////////////////////////
-      // Xp
-      ////////////////////////////////
-      basep = st.GetPFInfo(nent,plocal); nent++;
-      if ( local ) {
-	INTERIOR_BLOCK_XP(Xp,Yp,PERMUTE_DIR3,DIR0_PROJMEM,DIR0_RECON);
-      } else { 
-	EXTERIOR_BLOCK_XP(Xp,Yp,DIR0_RECON);
-      }
-      COMMON_BLOCK(Xp,Yp,DIR0_RECON);
-      ////////////////////////////////
-      // Yp
-      ////////////////////////////////
-      basep = st.GetPFInfo(nent,plocal); nent++;
-      if ( local ) {
-	INTERIOR_BLOCK(Yp,Zp,PERMUTE_DIR2,DIR1_PROJMEM,DIR1_RECON);
-      } else { 
-	EXTERIOR_BLOCK(Yp,Zp,DIR1_RECON);
-      }
-      COMMON_BLOCK(Yp,Zp,DIR1_RECON);
-      ////////////////////////////////
-      // Zp
-      ////////////////////////////////
-      basep = st.GetPFInfo(nent,plocal); nent++;
-      if ( local ) {
-	INTERIOR_BLOCK(Zp,Tp,PERMUTE_DIR1,DIR2_PROJMEM,DIR2_RECON);
-      } else { 
-	EXTERIOR_BLOCK(Zp,Tp,DIR2_RECON);
-      }
-      COMMON_BLOCK(Zp,Tp,DIR2_RECON);
-      ////////////////////////////////
-      // Tp
-      ////////////////////////////////
-      basep = st.GetPFInfo(nent,plocal); nent++;
-      if ( local ) {
-	INTERIOR_BLOCK(Tp,Xm,PERMUTE_DIR0,DIR3_PROJMEM,DIR3_RECON);
-      } else { 
-	EXTERIOR_BLOCK(Tp,Xm,DIR3_RECON);
-      }
-      COMMON_BLOCK(Tp,Xm,DIR3_RECON);
-      ////////////////////////////////
-      // Xm
-      ////////////////////////////////
-      //  basep= st.GetPFInfo(nent,plocal); nent++;
-      if ( local ) {
-	INTERIOR_BLOCK(Xm,Ym,PERMUTE_DIR3,DIR4_PROJMEM,DIR4_RECON);
-      } else { 
-	EXTERIOR_BLOCK(Xm,Ym,DIR4_RECON);
-      }
-      COMMON_BLOCK(Xm,Ym,DIR4_RECON);
-      ////////////////////////////////
-      // Ym
-      ////////////////////////////////
-      basep= st.GetPFInfo(nent,plocal); nent++;
-      if ( local ) {
-	INTERIOR_BLOCK(Ym,Zm,PERMUTE_DIR2,DIR5_PROJMEM,DIR5_RECON);
-      } else { 
-	EXTERIOR_BLOCK(Ym,Zm,DIR5_RECON);
-      }
-      COMMON_BLOCK(Ym,Zm,DIR5_RECON);
-      ////////////////////////////////
-      // Zm
-      ////////////////////////////////
-      basep= st.GetPFInfo(nent,plocal); nent++;
-      if ( local ) {
-	INTERIOR_BLOCK(Zm,Tm,PERMUTE_DIR1,DIR6_PROJMEM,DIR6_RECON);
-      } else { 
-	EXTERIOR_BLOCK(Zm,Tm,DIR6_RECON);
-      }
-      COMMON_BLOCK(Zm,Tm,DIR6_RECON);
-      ////////////////////////////////
-      // Tm
-      ////////////////////////////////
-      basep= st.GetPFInfo(nent,plocal); nent++;
-      if ( local ) {
-	INTERIOR_BLOCK(Tm,Xp,PERMUTE_DIR0,DIR7_PROJMEM,DIR7_RECON);
-      } else { 
-	EXTERIOR_BLOCK(Tm,Xp,DIR7_RECON);
-      }
-      COMMON_BLOCK(Tm,Xp,DIR7_RECON);
+   ASM_LEG_XP(Xp,Yp,PERMUTE_DIR3,DIR0_PROJMEM,DIR0_RECON);
+      ASM_LEG(Yp,Zp,PERMUTE_DIR2,DIR1_PROJMEM,DIR1_RECON);
+      ASM_LEG(Zp,Tp,PERMUTE_DIR1,DIR2_PROJMEM,DIR2_RECON);
+      ASM_LEG(Tp,Xm,PERMUTE_DIR0,DIR3_PROJMEM,DIR3_RECON);
 
+      ASM_LEG(Xm,Ym,PERMUTE_DIR3,DIR4_PROJMEM,DIR4_RECON);
+      ASM_LEG(Ym,Zm,PERMUTE_DIR2,DIR5_PROJMEM,DIR5_RECON);
+      ASM_LEG(Zm,Tm,PERMUTE_DIR1,DIR6_PROJMEM,DIR6_RECON);
+      ASM_LEG(Tm,Xp,PERMUTE_DIR0,DIR7_PROJMEM,DIR7_RECON);
+
+#ifdef EXTERIOR
+      if (nmu==0) break;
+      //      if (nmu!=0) std::cout << "EXT "<<sU<<std::endl;
+#endif
       base = (uint64_t) &out._odata[ss];
       basep= st.GetPFInfo(nent,plocal); nent++;
       RESULT(base,basep);
@@ -258,10 +191,6 @@
 #undef DIR5_RECON
 #undef DIR6_RECON
 #undef DIR7_RECON
-#undef EXTERIOR_BLOCK
-#undef INTERIOR_BLOCK
-#undef EXTERIOR_BLOCK_XP
-#undef INTERIOR_BLOCK_XP
-#undef COMMON_BLOCK
-#undef ZERO_NMU
+#undef ASM_LEG
+#undef ASM_LEG_XP
 #undef RESULT

lib/qcd/action/fermion/WilsonKernelsHand.cc

@@ -31,7 +31,7 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 #define REGISTER
 
 #define LOAD_CHIMU \
-  const SiteSpinor & ref (in._odata[offset]);	\
+  {const SiteSpinor & ref (in._odata[offset]);	\
     Chimu_00=ref()(0)(0);\
     Chimu_01=ref()(0)(1);\
     Chimu_02=ref()(0)(2);\
@@ -43,20 +43,20 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
     Chimu_22=ref()(2)(2);\
     Chimu_30=ref()(3)(0);\
     Chimu_31=ref()(3)(1);\
-    Chimu_32=ref()(3)(2);
+    Chimu_32=ref()(3)(2);}
 
 #define LOAD_CHI\
-  const SiteHalfSpinor &ref(buf[offset]);	\
+  {const SiteHalfSpinor &ref(buf[offset]);	\
     Chi_00 = ref()(0)(0);\
     Chi_01 = ref()(0)(1);\
     Chi_02 = ref()(0)(2);\
     Chi_10 = ref()(1)(0);\
     Chi_11 = ref()(1)(1);\
-    Chi_12 = ref()(1)(2);
+    Chi_12 = ref()(1)(2);}
 
 // To splat or not to splat depends on the implementation
 #define MULT_2SPIN(A)\
-   auto & ref(U._odata[sU](A));	\
+  {auto & ref(U._odata[sU](A));			\
    Impl::loadLinkElement(U_00,ref()(0,0));	\
    Impl::loadLinkElement(U_10,ref()(1,0));	\
    Impl::loadLinkElement(U_20,ref()(2,0));	\
@@ -83,7 +83,7 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
     UChi_01+= U_10*Chi_02;\
     UChi_11+= U_10*Chi_12;\
     UChi_02+= U_20*Chi_02;\
-    UChi_12+= U_20*Chi_12;
+    UChi_12+= U_20*Chi_12;}
 
 
 #define PERMUTE_DIR(dir)			\
@@ -307,55 +307,132 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
   result_31-= UChi_11;	\
   result_32-= UChi_12;
 
-namespace Grid {
-namespace QCD {
+#define HAND_STENCIL_LEG(PROJ,PERM,DIR,RECON)	\
+  SE=st.GetEntry(ptype,DIR,ss);			\
+  offset = SE->_offset;				\
+  local  = SE->_is_local;			\
+  perm   = SE->_permute;			\
+  if ( local ) {				\
+    LOAD_CHIMU;					\
+    PROJ;					\
+    if ( perm) {				\
+      PERMUTE_DIR(PERM);			\
+    }						\
+  } else {					\
+    LOAD_CHI;					\
+  }						\
+  MULT_2SPIN(DIR);				\
+  RECON;					
+
+#define HAND_STENCIL_LEG_INT(PROJ,PERM,DIR,RECON)	\
+  SE=st.GetEntry(ptype,DIR,ss);			\
+  offset = SE->_offset;				\
+  local  = SE->_is_local;			\
+  perm   = SE->_permute;			\
+  if ( local ) {				\
+    LOAD_CHIMU;					\
+    PROJ;					\
+    if ( perm) {				\
+      PERMUTE_DIR(PERM);			\
+    }						\
+  } else if ( st.same_node[DIR] ) {		\
+    LOAD_CHI;					\
+  }						\
+  if (local || st.same_node[DIR] ) {		\
+    MULT_2SPIN(DIR);				\
+    RECON;					\
+  }
+
+#define HAND_STENCIL_LEG_EXT(PROJ,PERM,DIR,RECON)	\
+  SE=st.GetEntry(ptype,DIR,ss);			\
+  offset = SE->_offset;				\
+  if((!SE->_is_local)&&(!st.same_node[DIR]) ) {	\
+    LOAD_CHI;					\
+    MULT_2SPIN(DIR);				\
+    RECON;					\
+    nmu++;					\
+  }
+
+#define HAND_RESULT(ss)				\
+  {						\
+    SiteSpinor & ref (out._odata[ss]);		\
+    vstream(ref()(0)(0),result_00);		\
+    vstream(ref()(0)(1),result_01);		\
+    vstream(ref()(0)(2),result_02);		\
+    vstream(ref()(1)(0),result_10);		\
+    vstream(ref()(1)(1),result_11);		\
+    vstream(ref()(1)(2),result_12);		\
+    vstream(ref()(2)(0),result_20);		\
+    vstream(ref()(2)(1),result_21);		\
+    vstream(ref()(2)(2),result_22);		\
+    vstream(ref()(3)(0),result_30);		\
+    vstream(ref()(3)(1),result_31);		\
+    vstream(ref()(3)(2),result_32);		\
+  }
+
+#define HAND_RESULT_EXT(ss)			\
+  if (nmu){					\
+    SiteSpinor & ref (out._odata[ss]);		\
+    ref()(0)(0)+=result_00;		\
+    ref()(0)(1)+=result_01;		\
+    ref()(0)(2)+=result_02;		\
+    ref()(1)(0)+=result_10;		\
+    ref()(1)(1)+=result_11;		\
+    ref()(1)(2)+=result_12;		\
+    ref()(2)(0)+=result_20;		\
+    ref()(2)(1)+=result_21;		\
+    ref()(2)(2)+=result_22;		\
+    ref()(3)(0)+=result_30;		\
+    ref()(3)(1)+=result_31;		\
+    ref()(3)(2)+=result_32;		\
+  }
 
 
-template<class Impl> void 
-WilsonKernels<Impl>::HandDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor  *buf,
-					  int ss,int sU,const FermionField &in, FermionField &out,int interior,int exterior)
-{
-  typedef typename Simd::scalar_type S;
-  typedef typename Simd::vector_type V;
+#define HAND_DECLARATIONS(a)			\
+  Simd result_00;				\
+  Simd result_01;				\
+  Simd result_02;				\
+  Simd result_10;				\
+  Simd result_11;				\
+  Simd result_12;				\
+  Simd result_20;				\
+  Simd result_21;				\
+  Simd result_22;				\
+  Simd result_30;				\
+  Simd result_31;				\
+  Simd result_32;				\
+  Simd Chi_00;					\
+  Simd Chi_01;					\
+  Simd Chi_02;					\
+  Simd Chi_10;					\
+  Simd Chi_11;					\
+  Simd Chi_12;					\
+  Simd UChi_00;					\
+  Simd UChi_01;					\
+  Simd UChi_02;					\
+  Simd UChi_10;					\
+  Simd UChi_11;					\
+  Simd UChi_12;					\
+  Simd U_00;					\
+  Simd U_10;					\
+  Simd U_20;					\
+  Simd U_01;					\
+  Simd U_11;					\
+  Simd U_21;
 
-  REGISTER Simd result_00; // 12 regs on knc
-  REGISTER Simd result_01;
-  REGISTER Simd result_02;
-
-  REGISTER Simd result_10;
-  REGISTER Simd result_11;
-  REGISTER Simd result_12;
-
-  REGISTER Simd result_20;
-  REGISTER Simd result_21;
-  REGISTER Simd result_22;
-
-  REGISTER Simd result_30;
-  REGISTER Simd result_31;
-  REGISTER Simd result_32; // 20 left
-
-  REGISTER Simd Chi_00;    // two spinor; 6 regs
-  REGISTER Simd Chi_01;
-  REGISTER Simd Chi_02;
-
-  REGISTER Simd Chi_10;
-  REGISTER Simd Chi_11;
-  REGISTER Simd Chi_12;   // 14 left
-
-  REGISTER Simd UChi_00;  // two spinor; 6 regs
-  REGISTER Simd UChi_01;
-  REGISTER Simd UChi_02;
-
-  REGISTER Simd UChi_10;
-  REGISTER Simd UChi_11;
-  REGISTER Simd UChi_12;  // 8 left
-
-  REGISTER Simd U_00;  // two rows of U matrix
-  REGISTER Simd U_10;
-  REGISTER Simd U_20;  
-  REGISTER Simd U_01;
-  REGISTER Simd U_11;
-  REGISTER Simd U_21;  // 2 reg left.
+#define ZERO_RESULT				\
+  result_00=zero;				\
+  result_01=zero;				\
+  result_02=zero;				\
+  result_10=zero;				\
+  result_11=zero;				\
+  result_12=zero;				\
+  result_20=zero;				\
+  result_21=zero;				\
+  result_22=zero;				\
+  result_30=zero;				\
+  result_31=zero;				\
+  result_32=zero;			
 
 #define Chimu_00 Chi_00
 #define Chimu_01 Chi_01
@@ -370,475 +447,225 @@ WilsonKernels<Impl>::HandDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGauge
 #define Chimu_31 UChi_11
 #define Chimu_32 UChi_12
 
+namespace Grid {
+namespace QCD {
+
+template<class Impl> void 
+WilsonKernels<Impl>::HandDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor  *buf,
+					  int ss,int sU,const FermionField &in, FermionField &out)
+{
+// T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
+  typedef typename Simd::scalar_type S;
+  typedef typename Simd::vector_type V;
+
+  HAND_DECLARATIONS(ignore);
 
   int offset,local,perm, ptype;
   StencilEntry *SE;
 
-  // Xp
-  SE=st.GetEntry(ptype,Xp,ss);
-  offset = SE->_offset;
-  local  = SE->_is_local;
-  perm   = SE->_permute;
-  
-  if ( local ) {
-    LOAD_CHIMU;
-    XM_PROJ;
-    if ( perm) {
-      PERMUTE_DIR(3); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
-    }
-  } else { 
-    LOAD_CHI;
-  }
-  {
-    MULT_2SPIN(Xp);
-  }
-  XM_RECON;
-  
-  // Yp
-  SE=st.GetEntry(ptype,Yp,ss);
-  offset = SE->_offset;
-  local  = SE->_is_local;
-  perm   = SE->_permute;
-  
-  if ( local ) {
-    LOAD_CHIMU;
-    YM_PROJ;
-    if ( perm) {
-      PERMUTE_DIR(2); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
-    }
-  } else { 
-    LOAD_CHI;
-  }
-  {
-    MULT_2SPIN(Yp);
-  }
-  YM_RECON_ACCUM;
-
-
-  // Zp
-  SE=st.GetEntry(ptype,Zp,ss);
-  offset = SE->_offset;
-  local  = SE->_is_local;
-  perm   = SE->_permute;
-  
-  if ( local ) {
-    LOAD_CHIMU;
-    ZM_PROJ;
-    if ( perm) {
-      PERMUTE_DIR(1); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
-    }
-  } else { 
-    LOAD_CHI;
-  }
-  {
-    MULT_2SPIN(Zp);
-  }
-  ZM_RECON_ACCUM;
-
-  // Tp
-  SE=st.GetEntry(ptype,Tp,ss);
-  offset = SE->_offset;
-  local  = SE->_is_local;
-  perm   = SE->_permute;
-  
-  if ( local ) {
-    LOAD_CHIMU;
-    TM_PROJ;
-    if ( perm) {
-      PERMUTE_DIR(0); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
-    }
-  } else { 
-    LOAD_CHI;
-  }
-  {
-    MULT_2SPIN(Tp);
-  }
-  TM_RECON_ACCUM;
-  
-  // Xm
-  SE=st.GetEntry(ptype,Xm,ss);
-  offset = SE->_offset;
-  local  = SE->_is_local;
-  perm   = SE->_permute;
-  
-  if ( local ) {
-    LOAD_CHIMU;
-    XP_PROJ;
-    if ( perm) {
-      PERMUTE_DIR(3); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
-    }
-  } else { 
-    LOAD_CHI;
-  }
-  {
-    MULT_2SPIN(Xm);
-  }
-  XP_RECON_ACCUM;
-  
-  
-  // Ym
-  SE=st.GetEntry(ptype,Ym,ss);
-  offset = SE->_offset;
-  local  = SE->_is_local;
-  perm   = SE->_permute;
-  
-  if ( local ) {
-    LOAD_CHIMU;
-    YP_PROJ;
-    if ( perm) {
-      PERMUTE_DIR(2); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
-    }
-  } else { 
-    LOAD_CHI;
-  }
-  {
-    MULT_2SPIN(Ym);
-  }
-  YP_RECON_ACCUM;
-
-  // Zm
-  SE=st.GetEntry(ptype,Zm,ss);
-  offset = SE->_offset;
-  local  = SE->_is_local;
-  perm   = SE->_permute;
-  
-  if ( local ) {
-    LOAD_CHIMU;
-    ZP_PROJ;
-    if ( perm) {
-      PERMUTE_DIR(1); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
-    }
-  } else { 
-    LOAD_CHI;
-  }
-  {
-    MULT_2SPIN(Zm);
-  }
-  ZP_RECON_ACCUM;
-
-  // Tm
-  SE=st.GetEntry(ptype,Tm,ss);
-  offset = SE->_offset;
-  local  = SE->_is_local;
-  perm   = SE->_permute;
-  
-  if ( local ) {
-    LOAD_CHIMU;
-    TP_PROJ;
-    if ( perm) {
-      PERMUTE_DIR(0); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
-    }
-  } else { 
-    LOAD_CHI;
-  }
-  {
-    MULT_2SPIN(Tm);
-  }
-  TP_RECON_ACCUM;
-
-  {
-    SiteSpinor & ref (out._odata[ss]);
-    vstream(ref()(0)(0),result_00);
-    vstream(ref()(0)(1),result_01);
-    vstream(ref()(0)(2),result_02);
-    vstream(ref()(1)(0),result_10);
-    vstream(ref()(1)(1),result_11);
-    vstream(ref()(1)(2),result_12);
-    vstream(ref()(2)(0),result_20);
-    vstream(ref()(2)(1),result_21);
-    vstream(ref()(2)(2),result_22);
-    vstream(ref()(3)(0),result_30);
-    vstream(ref()(3)(1),result_31);
-    vstream(ref()(3)(2),result_32);
-  }
+  HAND_STENCIL_LEG(XM_PROJ,3,Xp,XM_RECON);
+  HAND_STENCIL_LEG(YM_PROJ,2,Yp,YM_RECON_ACCUM);
+  HAND_STENCIL_LEG(ZM_PROJ,1,Zp,ZM_RECON_ACCUM);
+  HAND_STENCIL_LEG(TM_PROJ,0,Tp,TM_RECON_ACCUM);
+  HAND_STENCIL_LEG(XP_PROJ,3,Xm,XP_RECON_ACCUM);
+  HAND_STENCIL_LEG(YP_PROJ,2,Ym,YP_RECON_ACCUM);
+  HAND_STENCIL_LEG(ZP_PROJ,1,Zm,ZP_RECON_ACCUM);
+  HAND_STENCIL_LEG(TP_PROJ,0,Tm,TP_RECON_ACCUM);
+  HAND_RESULT(ss);
 }
 
 template<class Impl>
 void WilsonKernels<Impl>::HandDhopSiteDag(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
-						  int ss,int sU,const FermionField &in, FermionField &out,int interior,int exterior)
+						  int ss,int sU,const FermionField &in, FermionField &out)
 {
-  //  std::cout << "Hand op Dhop "<<std::endl;
   typedef typename Simd::scalar_type S;
   typedef typename Simd::vector_type V;
 
-  REGISTER Simd result_00; // 12 regs on knc
-  REGISTER Simd result_01;
-  REGISTER Simd result_02;
-  
-  REGISTER Simd result_10;
-  REGISTER Simd result_11;
-  REGISTER Simd result_12;
-
-  REGISTER Simd result_20;
-  REGISTER Simd result_21;
-  REGISTER Simd result_22;
-
-  REGISTER Simd result_30;
-  REGISTER Simd result_31;
-  REGISTER Simd result_32; // 20 left
-
-  REGISTER Simd Chi_00;    // two spinor; 6 regs
-  REGISTER Simd Chi_01;
-  REGISTER Simd Chi_02;
-
-  REGISTER Simd Chi_10;
-  REGISTER Simd Chi_11;
-  REGISTER Simd Chi_12;   // 14 left
-
-  REGISTER Simd UChi_00;  // two spinor; 6 regs
-  REGISTER Simd UChi_01;
-  REGISTER Simd UChi_02;
-
-  REGISTER Simd UChi_10;
-  REGISTER Simd UChi_11;
-  REGISTER Simd UChi_12;  // 8 left
-
-  REGISTER Simd U_00;  // two rows of U matrix
-  REGISTER Simd U_10;
-  REGISTER Simd U_20;  
-  REGISTER Simd U_01;
-  REGISTER Simd U_11;
-  REGISTER Simd U_21;  // 2 reg left.
-
-#define Chimu_00 Chi_00
-#define Chimu_01 Chi_01
-#define Chimu_02 Chi_02
-#define Chimu_10 Chi_10
-#define Chimu_11 Chi_11
-#define Chimu_12 Chi_12
-#define Chimu_20 UChi_00
-#define Chimu_21 UChi_01
-#define Chimu_22 UChi_02
-#define Chimu_30 UChi_10
-#define Chimu_31 UChi_11
-#define Chimu_32 UChi_12
-
+  HAND_DECLARATIONS(ignore);
 
   StencilEntry *SE;
   int offset,local,perm, ptype;
   
-  // Xp
-  SE=st.GetEntry(ptype,Xp,ss);
-  offset = SE->_offset;
-  local  = SE->_is_local;
-  perm   = SE->_permute;
+  HAND_STENCIL_LEG(XP_PROJ,3,Xp,XP_RECON);
+  HAND_STENCIL_LEG(YP_PROJ,2,Yp,YP_RECON_ACCUM);
+  HAND_STENCIL_LEG(ZP_PROJ,1,Zp,ZP_RECON_ACCUM);
+  HAND_STENCIL_LEG(TP_PROJ,0,Tp,TP_RECON_ACCUM);
+  HAND_STENCIL_LEG(XM_PROJ,3,Xm,XM_RECON_ACCUM);
+  HAND_STENCIL_LEG(YM_PROJ,2,Ym,YM_RECON_ACCUM);
+  HAND_STENCIL_LEG(ZM_PROJ,1,Zm,ZM_RECON_ACCUM);
+  HAND_STENCIL_LEG(TM_PROJ,0,Tm,TM_RECON_ACCUM);
+  HAND_RESULT(ss);
+}
 
-  if ( local ) {
-    LOAD_CHIMU;
-    XP_PROJ;
-    if ( perm) {
-      PERMUTE_DIR(3); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
-    }
-  } else { 
-    LOAD_CHI;
-  }
+template<class Impl> void 
+WilsonKernels<Impl>::HandDhopSiteInt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor  *buf,
+					  int ss,int sU,const FermionField &in, FermionField &out)
+{
+// T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
+  typedef typename Simd::scalar_type S;
+  typedef typename Simd::vector_type V;
 
-  {
-    MULT_2SPIN(Xp);
-  }
-  XP_RECON;
+  HAND_DECLARATIONS(ignore);
 
-  // Yp
-  SE=st.GetEntry(ptype,Yp,ss);
-  offset = SE->_offset;
-  local  = SE->_is_local;
-  perm   = SE->_permute;
+  int offset,local,perm, ptype;
+  StencilEntry *SE;
+  ZERO_RESULT;
+  HAND_STENCIL_LEG_INT(XM_PROJ,3,Xp,XM_RECON_ACCUM);
+  HAND_STENCIL_LEG_INT(YM_PROJ,2,Yp,YM_RECON_ACCUM);
+  HAND_STENCIL_LEG_INT(ZM_PROJ,1,Zp,ZM_RECON_ACCUM);
+  HAND_STENCIL_LEG_INT(TM_PROJ,0,Tp,TM_RECON_ACCUM);
+  HAND_STENCIL_LEG_INT(XP_PROJ,3,Xm,XP_RECON_ACCUM);
+  HAND_STENCIL_LEG_INT(YP_PROJ,2,Ym,YP_RECON_ACCUM);
+  HAND_STENCIL_LEG_INT(ZP_PROJ,1,Zm,ZP_RECON_ACCUM);
+  HAND_STENCIL_LEG_INT(TP_PROJ,0,Tm,TP_RECON_ACCUM);
+  HAND_RESULT(ss);
+}
 
-  if ( local ) {
-    LOAD_CHIMU;
-    YP_PROJ;
-    if ( perm) {
-      PERMUTE_DIR(2); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
-    }
-  } else { 
-    LOAD_CHI;
-  }
-  {
-    MULT_2SPIN(Yp);
-  }
-  YP_RECON_ACCUM;
+template<class Impl>
+void WilsonKernels<Impl>::HandDhopSiteDagInt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
+						  int ss,int sU,const FermionField &in, FermionField &out)
+{
+  typedef typename Simd::scalar_type S;
+  typedef typename Simd::vector_type V;
 
+  HAND_DECLARATIONS(ignore);
 
-  // Zp
-  SE=st.GetEntry(ptype,Zp,ss);
-  offset = SE->_offset;
-  local  = SE->_is_local;
-  perm   = SE->_permute;
+  StencilEntry *SE;
+  int offset,local,perm, ptype;
+  ZERO_RESULT;
+  HAND_STENCIL_LEG_INT(XP_PROJ,3,Xp,XP_RECON_ACCUM);
+  HAND_STENCIL_LEG_INT(YP_PROJ,2,Yp,YP_RECON_ACCUM);
+  HAND_STENCIL_LEG_INT(ZP_PROJ,1,Zp,ZP_RECON_ACCUM);
+  HAND_STENCIL_LEG_INT(TP_PROJ,0,Tp,TP_RECON_ACCUM);
+  HAND_STENCIL_LEG_INT(XM_PROJ,3,Xm,XM_RECON_ACCUM);
+  HAND_STENCIL_LEG_INT(YM_PROJ,2,Ym,YM_RECON_ACCUM);
+  HAND_STENCIL_LEG_INT(ZM_PROJ,1,Zm,ZM_RECON_ACCUM);
+  HAND_STENCIL_LEG_INT(TM_PROJ,0,Tm,TM_RECON_ACCUM);
+  HAND_RESULT(ss);
+}
 
-  if ( local ) {
-    LOAD_CHIMU;
-    ZP_PROJ;
-    if ( perm) {
-      PERMUTE_DIR(1); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
-    }
-  } else { 
-    LOAD_CHI;
-  }
-  {
-    MULT_2SPIN(Zp);
-  }
-  ZP_RECON_ACCUM;
+template<class Impl> void 
+WilsonKernels<Impl>::HandDhopSiteExt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor  *buf,
+					  int ss,int sU,const FermionField &in, FermionField &out)
+{
+// T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
+  typedef typename Simd::scalar_type S;
+  typedef typename Simd::vector_type V;
 
-  // Tp
-  SE=st.GetEntry(ptype,Tp,ss);
-  offset = SE->_offset;
-  local  = SE->_is_local;
-  perm   = SE->_permute;
+  HAND_DECLARATIONS(ignore);
 
-  if ( local ) {
-    LOAD_CHIMU;
-    TP_PROJ;
-    if ( perm) {
-      PERMUTE_DIR(0); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
-    }
-  } else { 
-    LOAD_CHI;
-  }
-  {
-    MULT_2SPIN(Tp);
-  }
-  TP_RECON_ACCUM;
+  int offset,local,perm, ptype;
+  StencilEntry *SE;
+  int nmu=0;
+  ZERO_RESULT;
+  HAND_STENCIL_LEG_EXT(XM_PROJ,3,Xp,XM_RECON_ACCUM);
+  HAND_STENCIL_LEG_EXT(YM_PROJ,2,Yp,YM_RECON_ACCUM);
+  HAND_STENCIL_LEG_EXT(ZM_PROJ,1,Zp,ZM_RECON_ACCUM);
+  HAND_STENCIL_LEG_EXT(TM_PROJ,0,Tp,TM_RECON_ACCUM);
+  HAND_STENCIL_LEG_EXT(XP_PROJ,3,Xm,XP_RECON_ACCUM);
+  HAND_STENCIL_LEG_EXT(YP_PROJ,2,Ym,YP_RECON_ACCUM);
+  HAND_STENCIL_LEG_EXT(ZP_PROJ,1,Zm,ZP_RECON_ACCUM);
+  HAND_STENCIL_LEG_EXT(TP_PROJ,0,Tm,TP_RECON_ACCUM);
+  HAND_RESULT_EXT(ss);
+}
 
-  // Xm
-  SE=st.GetEntry(ptype,Xm,ss);
-  offset = SE->_offset;
-  local  = SE->_is_local;
-  perm   = SE->_permute;
+template<class Impl>
+void WilsonKernels<Impl>::HandDhopSiteDagExt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
+						  int ss,int sU,const FermionField &in, FermionField &out)
+{
+  typedef typename Simd::scalar_type S;
+  typedef typename Simd::vector_type V;
 
-  if ( local ) {
-    LOAD_CHIMU;
-    XM_PROJ;
-    if ( perm) {
-      PERMUTE_DIR(3); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
-    }
-  } else { 
-    LOAD_CHI;
-  }
-  {
-    MULT_2SPIN(Xm);
-  }
-  XM_RECON_ACCUM;
+  HAND_DECLARATIONS(ignore);
 
-  // Ym
-  SE=st.GetEntry(ptype,Ym,ss);
-  offset = SE->_offset;
-  local  = SE->_is_local;
-  perm   = SE->_permute;
-  
-  if ( local ) {
-    LOAD_CHIMU;
-    YM_PROJ;
-    if ( perm) {
-      PERMUTE_DIR(2); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
-    }
-  } else { 
-    LOAD_CHI;
-  }
-  {
-    MULT_2SPIN(Ym);
-  }
-  YM_RECON_ACCUM;
-
-  // Zm
-  SE=st.GetEntry(ptype,Zm,ss);
-  offset = SE->_offset;
-  local  = SE->_is_local;
-  perm   = SE->_permute;
-
-  if ( local ) {
-    LOAD_CHIMU;
-    ZM_PROJ;
-    if ( perm) {
-      PERMUTE_DIR(1); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
-    }
-  } else { 
-    LOAD_CHI;
-  }
-  {
-    MULT_2SPIN(Zm);
-  }
-  ZM_RECON_ACCUM;
-
-  // Tm
-  SE=st.GetEntry(ptype,Tm,ss);
-  offset = SE->_offset;
-  local  = SE->_is_local;
-  perm   = SE->_permute;
-
-  if ( local ) {
-    LOAD_CHIMU;
-    TM_PROJ;
-    if ( perm) {
-      PERMUTE_DIR(0); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
-    }
-  } else { 
-    LOAD_CHI;
-  }
-  {
-    MULT_2SPIN(Tm);
-  }
-  TM_RECON_ACCUM;
-
-  {
-    SiteSpinor & ref (out._odata[ss]);
-    vstream(ref()(0)(0),result_00);
-    vstream(ref()(0)(1),result_01);
-    vstream(ref()(0)(2),result_02);
-    vstream(ref()(1)(0),result_10);
-    vstream(ref()(1)(1),result_11);
-    vstream(ref()(1)(2),result_12);
-    vstream(ref()(2)(0),result_20);
-    vstream(ref()(2)(1),result_21);
-    vstream(ref()(2)(2),result_22);
-    vstream(ref()(3)(0),result_30);
-    vstream(ref()(3)(1),result_31);
-    vstream(ref()(3)(2),result_32);
-  }
+  StencilEntry *SE;
+  int offset,local,perm, ptype;
+  int nmu=0;
+  ZERO_RESULT;
+  HAND_STENCIL_LEG_EXT(XP_PROJ,3,Xp,XP_RECON_ACCUM);
+  HAND_STENCIL_LEG_EXT(YP_PROJ,2,Yp,YP_RECON_ACCUM);
+  HAND_STENCIL_LEG_EXT(ZP_PROJ,1,Zp,ZP_RECON_ACCUM);
+  HAND_STENCIL_LEG_EXT(TP_PROJ,0,Tp,TP_RECON_ACCUM);
+  HAND_STENCIL_LEG_EXT(XM_PROJ,3,Xm,XM_RECON_ACCUM);
+  HAND_STENCIL_LEG_EXT(YM_PROJ,2,Ym,YM_RECON_ACCUM);
+  HAND_STENCIL_LEG_EXT(ZM_PROJ,1,Zm,ZM_RECON_ACCUM);
+  HAND_STENCIL_LEG_EXT(TM_PROJ,0,Tm,TM_RECON_ACCUM);
+  HAND_RESULT_EXT(ss);
 }
 
   ////////////////////////////////////////////////
   // Specialise Gparity to simple implementation
   ////////////////////////////////////////////////
-template<> void 
-WilsonKernels<GparityWilsonImplF>::HandDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,
-							SiteHalfSpinor *buf,
-							int sF,int sU,const FermionField &in, FermionField &out,int internal,int external)
-{
-  assert(0);
-}
-
-template<> void 
-WilsonKernels<GparityWilsonImplF>::HandDhopSiteDag(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,
-							   SiteHalfSpinor *buf,
-							   int sF,int sU,const FermionField &in, FermionField &out,int internal,int external)
-{
-  assert(0);
-}
-
-template<> void 
-WilsonKernels<GparityWilsonImplD>::HandDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
-							int sF,int sU,const FermionField &in, FermionField &out,int internal,int external)
-{
-  assert(0);
-}
-
-template<> void 
-WilsonKernels<GparityWilsonImplD>::HandDhopSiteDag(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
-							   int sF,int sU,const FermionField &in, FermionField &out,int internal,int external)
-{
-  assert(0);
-}
-
+#define HAND_SPECIALISE_EMPTY(IMPL)					\
+  template<> void							\
+  WilsonKernels<IMPL>::HandDhopSite(StencilImpl &st,			\
+				    LebesgueOrder &lo,			\
+				    DoubledGaugeField &U,		\
+				    SiteHalfSpinor *buf,		\
+				    int sF,int sU,			\
+				    const FermionField &in,		\
+				    FermionField &out){ assert(0); }	\
+  template<> void							\
+  WilsonKernels<IMPL>::HandDhopSiteDag(StencilImpl &st,			\
+				    LebesgueOrder &lo,			\
+				    DoubledGaugeField &U,		\
+				    SiteHalfSpinor *buf,		\
+				    int sF,int sU,			\
+				    const FermionField &in,		\
+				    FermionField &out){ assert(0); }	\
+  template<> void							\
+  WilsonKernels<IMPL>::HandDhopSiteInt(StencilImpl &st,			\
+				    LebesgueOrder &lo,			\
+				    DoubledGaugeField &U,		\
+				    SiteHalfSpinor *buf,		\
+				    int sF,int sU,			\
+				    const FermionField &in,		\
+				    FermionField &out){ assert(0); }	\
+  template<> void							\
+  WilsonKernels<IMPL>::HandDhopSiteExt(StencilImpl &st,			\
+				    LebesgueOrder &lo,			\
+				    DoubledGaugeField &U,		\
+				    SiteHalfSpinor *buf,		\
+				    int sF,int sU,			\
+				    const FermionField &in,		\
+				    FermionField &out){ assert(0); }	\
+  template<> void							\
+  WilsonKernels<IMPL>::HandDhopSiteDagInt(StencilImpl &st,	       	\
+				    LebesgueOrder &lo,			\
+				    DoubledGaugeField &U,		\
+				    SiteHalfSpinor *buf,		\
+				    int sF,int sU,			\
+				    const FermionField &in,		\
+				    FermionField &out){ assert(0); }	\
+  template<> void							\
+  WilsonKernels<IMPL>::HandDhopSiteDagExt(StencilImpl &st,	       	\
+				    LebesgueOrder &lo,			\
+				    DoubledGaugeField &U,		\
+				    SiteHalfSpinor *buf,		\
+				    int sF,int sU,			\
+				    const FermionField &in,		\
+				    FermionField &out){ assert(0); }	\
 
+  HAND_SPECIALISE_EMPTY(GparityWilsonImplF);
+  HAND_SPECIALISE_EMPTY(GparityWilsonImplD);
+  HAND_SPECIALISE_EMPTY(GparityWilsonImplFH);
+  HAND_SPECIALISE_EMPTY(GparityWilsonImplDF);
 
 ////////////// Wilson ; uses this implementation /////////////////////
-// Need Nc=3 though //
 
 #define INSTANTIATE_THEM(A) \
 template void WilsonKernels<A>::HandDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,\
-						     int ss,int sU,const FermionField &in, FermionField &out,int interior,int exterior); \
-template void WilsonKernels<A>::HandDhopSiteDag(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,\
-							int ss,int sU,const FermionField &in, FermionField &out,int interior,int exterior);
+					     int ss,int sU,const FermionField &in, FermionField &out); \
+template void WilsonKernels<A>::HandDhopSiteDag(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
+						int ss,int sU,const FermionField &in, FermionField &out);\
+template void WilsonKernels<A>::HandDhopSiteInt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,\
+						int ss,int sU,const FermionField &in, FermionField &out); \
+template void WilsonKernels<A>::HandDhopSiteDagInt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
+						   int ss,int sU,const FermionField &in, FermionField &out); \
+template void WilsonKernels<A>::HandDhopSiteExt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,\
+						int ss,int sU,const FermionField &in, FermionField &out); \
+template void WilsonKernels<A>::HandDhopSiteDagExt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
+						   int ss,int sU,const FermionField &in, FermionField &out); 
 
 INSTANTIATE_THEM(WilsonImplF);
 INSTANTIATE_THEM(WilsonImplD);
@@ -850,5 +677,15 @@ INSTANTIATE_THEM(DomainWallVec5dImplF);
 INSTANTIATE_THEM(DomainWallVec5dImplD);
 INSTANTIATE_THEM(ZDomainWallVec5dImplF);
 INSTANTIATE_THEM(ZDomainWallVec5dImplD);
+INSTANTIATE_THEM(WilsonImplFH);
+INSTANTIATE_THEM(WilsonImplDF);
+INSTANTIATE_THEM(ZWilsonImplFH);
+INSTANTIATE_THEM(ZWilsonImplDF);
+INSTANTIATE_THEM(GparityWilsonImplFH);
+INSTANTIATE_THEM(GparityWilsonImplDF);
+INSTANTIATE_THEM(DomainWallVec5dImplFH);
+INSTANTIATE_THEM(DomainWallVec5dImplDF);
+INSTANTIATE_THEM(ZDomainWallVec5dImplFH);
+INSTANTIATE_THEM(ZDomainWallVec5dImplDF);
 
 }}

lib/qcd/action/gauge/Gauge.h

@@ -29,7 +29,7 @@ directory
 #ifndef GRID_QCD_GAUGE_H
 #define GRID_QCD_GAUGE_H
 
-#include <Grid/qcd/action/gauge/GaugeImpl.h>
+#include <Grid/qcd/action/gauge/GaugeImplementations.h>
 #include <Grid/qcd/utils/WilsonLoops.h>
 #include <Grid/qcd/action/gauge/WilsonGaugeAction.h>
 #include <Grid/qcd/action/gauge/PlaqPlusRectangleAction.h>

lib/qcd/action/gauge/GaugeImplTypes.h

@@ -0,0 +1,153 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid
+
+Source file: ./lib/qcd/action/gauge/GaugeImpl.h
+
+Copyright (C) 2015
+
+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_GAUGE_IMPL_TYPES_H
+#define GRID_GAUGE_IMPL_TYPES_H
+
+namespace Grid {
+namespace QCD {
+
+////////////////////////////////////////////////////////////////////////
+// Implementation dependent gauge types
+////////////////////////////////////////////////////////////////////////
+
+#define INHERIT_GIMPL_TYPES(GImpl)                  \
+  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::ComplexField ComplexField; \
+  typedef typename Impl::SiteField SiteField;	    \
+  typedef typename Impl::Field Field;
+
+// hardcodes the exponential approximation in the template
+template <class S, int Nrepresentation = Nc, int Nexp = 12 > class GaugeImplTypes {
+public:
+  typedef S Simd;
+
+  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<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
+  // probably confusing the users
+  // Now keeping only one class
+
+
+  // Move this elsewhere? FIXME
+  static inline void AddLink(Field &U, LinkField &W,
+                                  int mu) { // U[mu] += W
+    PARALLEL_FOR_LOOP
+    for (auto ss = 0; ss < U._grid->oSites(); ss++) {
+      U._odata[ss]._internal[mu] =
+          U._odata[ss]._internal[mu] + W._odata[ss]._internal;
+    }
+  }
+
+  ///////////////////////////////////////////////////////////
+  // Move these to another class
+  // HMC auxiliary functions
+  static inline void generate_momenta(Field &P, GridParallelRNG &pRNG) {
+    // specific for SU gauge fields
+    LinkField Pmu(P._grid);
+    Pmu = zero;
+    for (int mu = 0; mu < Nd; mu++) {
+      SU<Nrepresentation>::GaussianFundamentalLieAlgebraMatrix(pRNG, Pmu);
+      PokeIndex<LorentzIndex>(P, Pmu, mu);
+    }
+  }
+
+  static inline Field projectForce(Field &P) { return Ta(P); }
+
+  static inline void update_field(Field& P, Field& U, double ep){
+    //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){
+    LatticeComplex Hloc(U._grid);
+    Hloc = zero;
+    for (int mu = 0; mu < Nd; mu++) {
+      auto Umu = PeekIndex<LorentzIndex>(U, mu);
+      Hloc += trace(Umu * Umu);
+    }
+    Complex Hsum = sum(Hloc);
+    return Hsum.real();
+  }
+
+  static inline void HotConfiguration(GridParallelRNG &pRNG, Field &U) {
+    SU<Nc>::HotConfiguration(pRNG, U);
+  }
+
+  static inline void TepidConfiguration(GridParallelRNG &pRNG, Field &U) {
+    SU<Nc>::TepidConfiguration(pRNG, U);
+  }
+
+  static inline void ColdConfiguration(GridParallelRNG &pRNG, Field &U) {
+    SU<Nc>::ColdConfiguration(pRNG, U);
+  }
+};
+
+
+typedef GaugeImplTypes<vComplex, Nc> GimplTypesR;
+typedef GaugeImplTypes<vComplexF, Nc> GimplTypesF;
+typedef GaugeImplTypes<vComplexD, Nc> GimplTypesD;
+
+typedef GaugeImplTypes<vComplex, SU<Nc>::AdjointDimension> GimplAdjointTypesR;
+typedef GaugeImplTypes<vComplexF, SU<Nc>::AdjointDimension> GimplAdjointTypesF;
+typedef GaugeImplTypes<vComplexD, SU<Nc>::AdjointDimension> GimplAdjointTypesD;
+
+
+} // QCD
+} // Grid
+
+#endif // GRID_GAUGE_IMPL_TYPES_H