Merge branch 'feature/hadrons' into feature/qed-fvol

# Conflicts:
#	Makefile.am
#	configure.ac
#	lib/qcd/action/gauge/Photon.h

.gitignore

@@ -47,7 +47,9 @@ Config.h.in
 config.log
 config.status
 .deps
-*.inc
+Make.inc
+eigen.inc
+Eigen.inc
 
 # http://www.gnu.org/software/autoconf #
 ########################################

Makefile.am

@@ -1,5 +1,12 @@
 # additional include paths necessary to compile the C++ library
-SUBDIRS = lib benchmarks tests programs
+SUBDIRS = lib benchmarks tests extras
+
+include $(top_srcdir)/doxygen.inc
+
+tests: all
+	$(MAKE) -C tests tests
+
+.PHONY: tests doxygen-run doxygen-doc $(DX_PS_GOAL) $(DX_PDF_GOAL)
 
 AM_CXXFLAGS += -I$(top_builddir)/include
 ACLOCAL_AMFLAGS = -I m4

README

@@ -1,44 +0,0 @@
-This library provides data parallel C++ container classes with internal memory layout
-that is transformed to map efficiently to SIMD architectures. CSHIFT facilities
-are provided, similar to HPF and cmfortran, and user control is given over the mapping of
-array indices to both MPI tasks and SIMD processing elements.
-
-* Identically shaped arrays then be processed with perfect data parallelisation.
-* Such identically shapped arrays are called conformable arrays.
-
-The transformation is based on the observation that Cartesian array processing involves
-identical processing to be performed on different regions of the Cartesian array.
-
-The library will (eventually) both geometrically decompose into MPI tasks and across SIMD lanes.
-
-Data parallel array operations can then be specified with a SINGLE data parallel paradigm, but
-optimally use MPI, OpenMP and SIMD parallelism under the hood. This is a significant simplification
-for most programmers.
-
-The layout transformations are parametrised by the SIMD vector length. This adapts according to the architecture.
-Presently SSE2 (128 bit) AVX, AVX2 (256 bit) and IMCI and AVX512 (512 bit) targets are supported.
-
-These are presented as 
-
-  vRealF, vRealD, vComplexF, vComplexD 
-
-internal vector data types. These may be useful in themselves for other programmers.
-The corresponding scalar types are named
-
-  RealF, RealD, ComplexF, ComplexD
-
-MPI parallelism is UNIMPLEMENTED and for now only OpenMP and SIMD parallelism is present in the library.
-
-   You can give `configure' initial values for configuration parameters
-by setting variables in the command line or in the environment.  Here
-is are examples:
-
-     ./configure CXX=clang++ CXXFLAGS="-std=c++11 -O3 -msse4" --enable-simd=SSE4
-
-     ./configure CXX=clang++ CXXFLAGS="-std=c++11 -O3 -mavx" --enable-simd=AVX1
-
-     ./configure CXX=clang++ CXXFLAGS="-std=c++11 -O3 -mavx2" --enable-simd=AVX2
-
-     ./configure CXX=icpc CXXFLAGS="-std=c++11 -O3 -mmic" --enable-simd=AVX512 --host=none
-     
-     

README

@@ -0,0 +1 @@
+README.md

README.md

@@ -106,6 +106,7 @@ To minimise the build time, only the tests at the root of the `tests` directory
 ``` bash
 make -C tests/<subdir> tests
 ```
+If you want to build all the tests at once just use `make tests`.
 
 ### Build configuration options
 
@@ -115,17 +116,19 @@ make -C tests/<subdir> tests
 - `--with-fftw=<path>`: look for FFTW in the UNIX prefix `<path>`
 - `--enable-lapack[=<path>]`: enable LAPACK support in Lanczos eigensolver. A UNIX prefix containing the library can be specified (optional).
 - `--enable-mkl[=<path>]`: use Intel MKL for FFT (and LAPACK if enabled) routines. A UNIX prefix containing the library can be specified (optional).
-- `--enable-numa`: ???
+- `--enable-numa`: enable NUMA first touch optimisation
 - `--enable-simd=<code>`: setup Grid for the SIMD target `<code>` (default: `GEN`). A list of possible SIMD targets is detailed in a section below.
+- `--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 `).
 - `--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`)
 
 ### Possible communication interfaces
 
-The following options can be use with the `--enable-simd=` option to target different communication interfaces:
+The following options can be use with the `--enable-comms=` option to target different communication interfaces:
 
 | `<comm>`       | Description                                                   |
 | -------------- | ------------------------------------------------------------- |
@@ -135,7 +138,7 @@ The following options can be use with the `--enable-simd=` option to target diff
 | `mpi3l[-auto]` | MPI communications using MPI 3 shared memory and leader model |
 | `shmem `       | Cray SHMEM communications                                     |
 
-For the MPI interfaces the optional `-auto` suffix instructs the `configure` scripts to determine all the necessary compilation and linking flags. This is done by extracting the informations from the MPI wrapper specified in the environment variable `MPICXX` (if not specified `configure` will scan though a list of default names).
+For the MPI interfaces the optional `-auto` suffix instructs the `configure` scripts to determine all the necessary compilation and linking flags. This is done by extracting the informations from the MPI wrapper specified in the environment variable `MPICXX` (if not specified `configure` will scan though a list of default names). The `-auto` suffix is not supported by the Cray environment wrapper scripts. Use the standard versions instead.  
 
 ### Possible SIMD types
 
@@ -164,6 +167,7 @@ Alternatively, some CPU codenames can be directly used:
 - We currently support AVX512 only for the Intel compiler. Support for GCC and clang will appear in future versions of Grid when the AVX512 support within GCC and clang will be more advanced.
 - For BG/Q only [bgclang](http://trac.alcf.anl.gov/projects/llvm-bgq) is supported. We do not presently plan to support more compilers for this platform.
 - BG/Q performances are currently rather poor. This is being investigated for future versions.
+- The vector size for the `GEN` target can be specified with the `configure` script option `--enable-gen-simd-width`.
 
 ### Build setup for Intel Knights Landing platform
 

benchmarks/Benchmark_comms.cc

@@ -193,6 +193,7 @@ int main (int argc, char ** argv)
     }
   }  
 
+
   Nloop=100;
   std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
   std::cout<<GridLogMessage << "= Benchmarking concurrent STENCIL halo exchange in "<<nmu<<" dimensions"<<std::endl;
@@ -271,5 +272,90 @@ int main (int argc, char ** argv)
       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;
+
+  for(int lat=4;lat<=maxlat;lat+=2){
+    for(int Ls=1;Ls<=16;Ls*=2){
+
+      std::vector<int> latt_size  ({lat*mpi_layout[0],
+      				    lat*mpi_layout[1],
+      				    lat*mpi_layout[2],
+      				    lat*mpi_layout[3]});
+
+      GridCartesian     Grid(latt_size,simd_layout,mpi_layout);
+
+      std::vector<HalfSpinColourVectorD *> xbuf(8);
+      std::vector<HalfSpinColourVectorD *> rbuf(8);
+      Grid.ShmBufferFreeAll();
+      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));
+      }
+
+      int ncomm;
+      int bytes=lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD);
+
+      double start=usecond();
+      for(int i=0;i<Nloop;i++){
+
+	std::vector<CartesianCommunicator::CommsRequest_t> requests;
+
+	ncomm=0;
+	for(int mu=0;mu<4;mu++){
+	
+	  if (mpi_layout[mu]>1 ) {
+	  
+	    ncomm++;
+	    int comm_proc=1;
+	    int xmit_to_rank;
+	    int recv_from_rank;
+	    
+	    Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank);
+	    Grid.StencilSendToRecvFromBegin(requests,
+					    (void *)&xbuf[mu][0],
+					    xmit_to_rank,
+					    (void *)&rbuf[mu][0],
+					    recv_from_rank,
+					    bytes);
+	    //	    Grid.StencilSendToRecvFromComplete(requests);
+	    //	    requests.resize(0);
+
+	    comm_proc = mpi_layout[mu]-1;
+	  
+	    Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank);
+	    Grid.StencilSendToRecvFromBegin(requests,
+					    (void *)&xbuf[mu+4][0],
+					    xmit_to_rank,
+					    (void *)&rbuf[mu+4][0],
+					    recv_from_rank,
+					    bytes);
+	    Grid.StencilSendToRecvFromComplete(requests);
+	    requests.resize(0);
+	  
+	  }
+	}
+	Grid.Barrier();
+
+      }
+      double stop=usecond();
+
+      double dbytes    = bytes;
+      double xbytes    = Nloop*dbytes*2.0*ncomm;
+      double rbytes    = xbytes;
+      double bidibytes = xbytes+rbytes;
+
+      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;
+    }
+  }    
+
   Grid_finalize();
 }

benchmarks/Benchmark_dwf.cc

@@ -57,7 +57,7 @@ int main (int argc, char ** argv)
   std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
 
   std::vector<int> latt4 = GridDefaultLatt();
-  const int Ls=16;
+  const int Ls=8;
   GridCartesian         * UGrid   = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());
   GridRedBlackCartesian * UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
   GridCartesian         * FGrid   = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
@@ -138,7 +138,7 @@ int main (int argc, char ** argv)
 
   int ncall =100;
   if (1) {
-
+    FGrid->Barrier();
     Dw.ZeroCounters();
     double t0=usecond();
     for(int i=0;i<ncall;i++){
@@ -147,6 +147,7 @@ int main (int argc, char ** argv)
       __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;
@@ -158,7 +159,7 @@ int main (int argc, char ** argv)
     std::cout<<GridLogMessage << "mflop/s per rank =  "<< flops/(t1-t0)/NP<<std::endl;
     err = ref-result; 
     std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<<std::endl;
-    assert (norm2(err)< 1.0e-5 );
+    assert (norm2(err)< 1.0e-4 );
     Dw.Report();
   }
 
@@ -193,6 +194,7 @@ int main (int argc, char ** argv)
       pokeSite(tmp,ssrc,site);
     }}}}}
     std::cout<<GridLogMessage<< "src norms "<< norm2(src)<<" " <<norm2(ssrc)<<std::endl;
+    FGrid->Barrier();
     double t0=usecond();
     sDw.ZeroCounters();
     for(int i=0;i<ncall;i++){
@@ -201,6 +203,7 @@ int main (int argc, char ** argv)
       __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;
 
@@ -211,12 +214,12 @@ int main (int argc, char ** argv)
   
     if(0){
       for(int i=0;i< PerformanceCounter::NumTypes(); i++ ){
-  sDw.Dhop(ssrc,sresult,0);
-  PerformanceCounter Counter(i);
-  Counter.Start();
-  sDw.Dhop(ssrc,sresult,0);
-  Counter.Stop();
-  Counter.Report();
+	sDw.Dhop(ssrc,sresult,0);
+	PerformanceCounter Counter(i);
+	Counter.Start();
+	sDw.Dhop(ssrc,sresult,0);
+	Counter.Stop();
+	Counter.Report();
       }
     }
 
@@ -240,7 +243,7 @@ int main (int argc, char ** argv)
       }
     }}}}}
     std::cout<<GridLogMessage<<" difference between normal and simd is "<<sum<<std::endl;
-    assert (sum< 1.0e-5 );
+    assert (sum< 1.0e-4 );
 
 
     if (1) {
@@ -271,6 +274,7 @@ int main (int argc, char ** argv)
       if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptInlineAsm ) std::cout << GridLogMessage<< "* Using Asm Nc=3   WilsonKernels" <<std::endl;
       std::cout << GridLogMessage<< "*********************************************************" <<std::endl;
 
+      FGrid->Barrier();
       sDw.ZeroCounters();
       sDw.stat.init("DhopEO");
       double t0=usecond();
@@ -278,6 +282,7 @@ int main (int argc, char ** argv)
         sDw.DhopEO(ssrc_o, sr_e, DaggerNo);
       }
       double t1=usecond();
+      FGrid->Barrier();
       sDw.stat.print();
 
       double volume=Ls;  for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
@@ -301,7 +306,7 @@ int main (int argc, char ** argv)
 
       error+= norm2(ssrc_o);
       std::cout<<GridLogMessage << "sO norm diff   "<< norm2(ssrc_o)<< "  vec nrm"<<norm2(sr_o) <<std::endl;
-      if(error>1.0e-5) { 
+      if(error>1.0e-4) { 
 	setCheckerboard(ssrc,ssrc_o);
 	setCheckerboard(ssrc,ssrc_e);
 	std::cout<< ssrc << std::endl;
@@ -337,7 +342,7 @@ int main (int argc, char ** argv)
   std::cout<<GridLogMessage << "norm ref    "<< norm2(ref)<<std::endl;
   err = ref-result; 
   std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<<std::endl;
-  assert(norm2(err)<1.0e-5);
+  assert(norm2(err)<1.0e-4);
   LatticeFermion src_e (FrbGrid);
   LatticeFermion src_o (FrbGrid);
   LatticeFermion r_e   (FrbGrid);
@@ -363,11 +368,13 @@ int main (int argc, char ** argv)
   std::cout << GridLogMessage<< "*********************************************************" <<std::endl;
   {
     Dw.ZeroCounters();
+    FGrid->Barrier();
     double t0=usecond();
     for(int i=0;i<ncall;i++){
       Dw.DhopEO(src_o,r_e,DaggerNo);
     }
     double t1=usecond();
+    FGrid->Barrier();
     
     double volume=Ls;  for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
     double flops=(1344.0*volume*ncall)/2;
@@ -389,14 +396,14 @@ int main (int argc, char ** argv)
 
   err = r_eo-result; 
   std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<<std::endl;
-  assert(norm2(err)<1.0e-5);
+  assert(norm2(err)<1.0e-4);
 
   pickCheckerboard(Even,src_e,err);
   pickCheckerboard(Odd,src_o,err);
   std::cout<<GridLogMessage << "norm diff even  "<< norm2(src_e)<<std::endl;
   std::cout<<GridLogMessage << "norm diff odd   "<< norm2(src_o)<<std::endl;
-  assert(norm2(src_e)<1.0e-5);
-  assert(norm2(src_o)<1.0e-5);
+  assert(norm2(src_e)<1.0e-4);
+  assert(norm2(src_o)<1.0e-4);
 
   Grid_finalize();
 }

benchmarks/Benchmark_dwf_sweep.cc

@@ -69,8 +69,8 @@ int main (int argc, char ** argv)
   std::cout<<GridLogMessage << "Volume \t\t\tProcs \t Dw \t eoDw \t sDw \t eosDw (Mflop/s)  "<<std::endl;
   std::cout<<GridLogMessage << "=========================================================================="<<std::endl;
 
-  int Lmax=32;
-  int dmin=0;
+  int Lmax=16;
+  int dmin=2;
   if ( getenv("LMAX") ) Lmax=atoi(getenv("LMAX"));
   if ( getenv("DMIN") ) dmin=atoi(getenv("DMIN"));
   for (int L=8;L<=Lmax;L*=2){

benchmarks/Benchmark_mooee.cc

@@ -0,0 +1,183 @@
+    /*************************************************************************************
+
+    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>
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License along
+    with this program; if not, write to the Free Software Foundation, Inc.,
+    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+    See the full license in the file "LICENSE" in the top level distribution directory
+    *************************************************************************************/
+    /*  END LEGAL */
+#include <Grid/Grid.h>
+
+using namespace std;
+using namespace Grid;
+using namespace Grid::QCD;
+
+
+int main (int argc, char ** argv)
+{
+  Grid_init(&argc,&argv);
+
+  int threads = GridThread::GetThreads();
+  std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
+
+  std::vector<int> latt4 = GridDefaultLatt();
+  const int Ls=16;
+  GridCartesian         * UGrid   = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());
+  GridRedBlackCartesian * UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
+  GridCartesian         * FGrid   = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
+  GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid);
+
+  std::cout << GridLogMessage << "Making Vec5d innermost grids"<<std::endl;
+  GridCartesian         * sUGrid   = SpaceTimeGrid::makeFourDimDWFGrid(GridDefaultLatt(),GridDefaultMpi());
+  GridRedBlackCartesian * sUrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(sUGrid);
+  GridCartesian         * sFGrid   = SpaceTimeGrid::makeFiveDimDWFGrid(Ls,UGrid);
+  GridRedBlackCartesian * sFrbGrid = SpaceTimeGrid::makeFiveDimDWFRedBlackGrid(Ls,UGrid);
+
+  std::vector<int> seeds4({1,2,3,4});
+  std::vector<int> seeds5({5,6,7,8});
+
+  GridParallelRNG          RNG4(UGrid);  RNG4.SeedFixedIntegers(seeds4);
+  std::cout << GridLogMessage << "Seeded"<<std::endl;
+
+  LatticeGaugeField Umu(UGrid); SU3::HotConfiguration(RNG4,Umu);
+
+  std::cout << GridLogMessage << "made random gauge fields"<<std::endl;
+
+  RealD mass=0.1;
+  RealD M5  =1.8;
+  RealD NP = UGrid->_Nprocessors;
+
+
+  if (1)
+  {
+    const int ncall=1000;
+
+    std::cout << GridLogMessage<< "*********************************************************" <<std::endl;
+    std::cout << GridLogMessage<< "* Benchmarking DomainWallFermionR::Dhop "<<std::endl;
+    std::cout << GridLogMessage<< "*********************************************************" <<std::endl;
+
+    GridParallelRNG RNG5(FGrid);
+    LatticeFermion src(FGrid); random(RNG5,src);
+    LatticeFermion result(FGrid);
+
+    DomainWallFermionR Dw(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
+    double t0,t1;
+
+    LatticeFermion r_eo(FGrid);
+    LatticeFermion src_e (FrbGrid);
+    LatticeFermion src_o (FrbGrid);
+    LatticeFermion r_e   (FrbGrid);
+    LatticeFermion r_o   (FrbGrid);
+    
+    pickCheckerboard(Even,src_e,src);
+    pickCheckerboard(Odd,src_o,src);
+    
+    setCheckerboard(r_eo,src_o);
+    setCheckerboard(r_eo,src_e);
+    
+    r_e = zero;
+    r_o = zero;
+
+
+#define BENCH_DW(A,in,out)			\
+    Dw.CayleyZeroCounters();			\
+    Dw. A (in,out);				\
+    FGrid->Barrier();				\
+    t0=usecond();				\
+    for(int i=0;i<ncall;i++){			\
+      Dw. A (in,out);				\
+    }						\
+    t1=usecond();				\
+    FGrid->Barrier();				\
+    Dw.CayleyReport();					\
+    std::cout<<GridLogMessage << "Called " #A " "<< (t1-t0)/ncall<<" us"<<std::endl;\
+    std::cout<<GridLogMessage << "******************"<<std::endl;
+
+#define BENCH_DW_MEO(A,in,out)			\
+    Dw.CayleyZeroCounters();			\
+    Dw. A (in,out,0);				\
+    FGrid->Barrier();				\
+    t0=usecond();				\
+    for(int i=0;i<ncall;i++){			\
+      Dw. A (in,out,0);				\
+    }						\
+    t1=usecond();				\
+    FGrid->Barrier();				\
+    Dw.CayleyReport();					\
+    std::cout<<GridLogMessage << "Called " #A " "<< (t1-t0)/ncall<<" us"<<std::endl;\
+    std::cout<<GridLogMessage << "******************"<<std::endl;
+
+    BENCH_DW_MEO(Dhop    ,src,result);
+    BENCH_DW_MEO(DhopEO  ,src_o,r_e);
+    BENCH_DW(Meooe   ,src_o,r_e);
+    BENCH_DW(Mooee   ,src_o,r_o);
+    BENCH_DW(MooeeInv,src_o,r_o);
+
+  }
+
+  if (1)
+  {
+    const int ncall=1000;
+
+    std::cout << GridLogMessage<< "*********************************************************" <<std::endl;
+    std::cout << GridLogMessage<< "* Benchmarking DomainWallFermionVec5dR::Dhop "<<std::endl;
+    std::cout << GridLogMessage<< "*********************************************************" <<std::endl;
+
+    GridParallelRNG RNG5(sFGrid);
+    LatticeFermion src(sFGrid); random(RNG5,src);
+    LatticeFermion sref(sFGrid);
+    LatticeFermion result(sFGrid);
+
+    std::cout<<GridLogMessage << "Constructing Vec5D Dw "<<std::endl;
+    DomainWallFermionVec5dR Dw(Umu,*sFGrid,*sFrbGrid,*sUGrid,*sUrbGrid,mass,M5);
+
+    std::cout<<GridLogMessage << "Calling Dhop "<<std::endl;
+    FGrid->Barrier();
+
+    double t0,t1;
+
+    LatticeFermion r_eo(sFGrid);
+    LatticeFermion src_e (sFrbGrid);
+    LatticeFermion src_o (sFrbGrid);
+    LatticeFermion r_e   (sFrbGrid);
+    LatticeFermion r_o   (sFrbGrid);
+    
+    pickCheckerboard(Even,src_e,src);
+    pickCheckerboard(Odd,src_o,src);
+    
+    setCheckerboard(r_eo,src_o);
+    setCheckerboard(r_eo,src_e);
+    
+    r_e = zero;
+    r_o = zero;
+
+    BENCH_DW_MEO(Dhop    ,src,result);
+    BENCH_DW_MEO(DhopEO  ,src_o,r_e);
+    BENCH_DW(Meooe   ,src_o,r_e);
+    BENCH_DW(Mooee   ,src_o,r_o);
+    BENCH_DW(MooeeInv,src_o,r_o);
+
+  }
+
+  Grid_finalize();
+}

benchmarks/Makefile.am

@@ -1 +1,11 @@
 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

benchmarks/simple_simd_test.cc

@@ -0,0 +1,11 @@
+#include <Grid/Grid.h>
+
+Grid::vRealD add(const Grid::vRealD &x, const Grid::vRealD &y)
+{
+	return x+y;
+}
+
+Grid::vRealD sub(const Grid::vRealD &x, const Grid::vRealD &y)
+{
+	return x-y;
+}

benchmarks/simple_su3_expr.cc

@@ -25,7 +25,7 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
     See the full license in the file "LICENSE" in the top level distribution directory
     *************************************************************************************/
     /*  END LEGAL */
-#include <Grid.h>
+#include <Grid/Grid.h>
 
 using namespace std;
 using namespace Grid;

benchmarks/simple_su3_test.cc

@@ -25,7 +25,7 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
     See the full license in the file "LICENSE" in the top level distribution directory
     *************************************************************************************/
     /*  END LEGAL */
-#include <Grid.h>
+#include <Grid/Grid.h>
 
 using namespace std;
 using namespace Grid;

configure.ac

@@ -149,8 +149,14 @@ CXXFLAGS=$CXXFLAGS_CPY
 LDFLAGS=$LDFLAGS_CPY
 
 ############### SIMD instruction selection
-AC_ARG_ENABLE([simd],[AC_HELP_STRING([--enable-simd=<code>],
-	            [select SIMD target (cf. README.md)])], [ac_SIMD=${enable_simd}], [ac_SIMD=GEN])
+AC_ARG_ENABLE([simd],[AC_HELP_STRING([--enable-simd=code],
+              [select SIMD target (cf. README.md)])], [ac_SIMD=${enable_simd}], [ac_SIMD=GEN])
+
+AC_ARG_ENABLE([gen-simd-width],
+            [AS_HELP_STRING([--enable-gen-simd-width=size],
+            [size (in bytes) of the generic SIMD vectors (default: 32)])],
+            [ac_gen_simd_width=$enable_gen_simd_width],
+            [ac_gen_simd_width=32])
 
 case ${ax_cv_cxx_compiler_vendor} in
   clang|gnu)
@@ -180,7 +186,10 @@ case ${ax_cv_cxx_compiler_vendor} in
         AC_DEFINE([AVX512],[1],[AVX512 intrinsics])
         SIMD_FLAGS='-march=knl';;
       GEN)
-        AC_DEFINE([GENERIC_VEC],[1],[generic vector code])
+        AC_DEFINE([GEN],[1],[generic vector code])
+        AC_DEFINE_UNQUOTED([GEN_SIMD_WIDTH],[$ac_gen_simd_width],
+                           [generic SIMD vector width (in bytes)])
+        SIMD_GEN_WIDTH_MSG=" (width= $ac_gen_simd_width)"
         SIMD_FLAGS='';;
       QPX|BGQ)
         AC_DEFINE([QPX],[1],[QPX intrinsics for BG/Q])
@@ -197,8 +206,8 @@ case ${ax_cv_cxx_compiler_vendor} in
         AC_DEFINE([AVX1],[1],[AVX intrinsics])
         SIMD_FLAGS='-mavx -xavx';;
       AVXFMA)
-        AC_DEFINE([AVXFMA],[1],[AVX intrinsics with FMA4])
-        SIMD_FLAGS='-mavx -mfma';;
+        AC_DEFINE([AVXFMA],[1],[AVX intrinsics with FMA3])
+        SIMD_FLAGS='-mavx -fma';;
       AVX2)
         AC_DEFINE([AVX2],[1],[AVX2 intrinsics])
         SIMD_FLAGS='-march=core-avx2 -xcore-avx2';;
@@ -212,7 +221,10 @@ case ${ax_cv_cxx_compiler_vendor} in
         AC_DEFINE([AVX512],[1],[AVX512 intrinsics for Knights Landing])
         SIMD_FLAGS='-xmic-avx512';;
       GEN)
-        AC_DEFINE([GENERIC_VEC],[1],[generic vector code])
+        AC_DEFINE([GEN],[1],[generic vector code])
+        AC_DEFINE_UNQUOTED([GEN_SIMD_WIDTH],[$ac_gen_simd_width],
+                           [generic SIMD vector width (in bytes)])
+        SIMD_GEN_WIDTH_MSG=" (width= $ac_gen_simd_width)"
         SIMD_FLAGS='';;
       *)
         AC_MSG_ERROR(["SIMD option ${ac_SIMD} not supported by the Intel compiler"]);;
@@ -278,7 +290,7 @@ esac
 case ${ac_COMMS} in
     *-auto)
         LX_FIND_MPI
-        if test "x$have_CXX_mpi" = 'xno'; then AC_MSG_ERROR(["MPI not found"]); fi
+        if test "x$have_CXX_mpi" = 'xno'; then AC_MSG_ERROR(["The configure could not find the MPI compilation flags. N.B. The -auto mode is not supported by Cray wrappers. Use the non -auto version in this case."]); fi
         AM_CXXFLAGS="$MPI_CXXFLAGS $AM_CXXFLAGS"
         AM_CFLAGS="$MPI_CFLAGS $AM_CFLAGS"
         AM_LDFLAGS="`echo $MPI_CXXLDFLAGS | sed -E 's/-l@<:@^ @:>@+//g'` $AM_LDFLAGS"
@@ -342,12 +354,17 @@ esac
 AM_CONDITIONAL(BUILD_CHROMA_REGRESSION,[ test "X${ac_CHROMA}X" == "XyesX" ])
 
 ############### Doxygen
-AC_PROG_DOXYGEN
-
-if test -n "$DOXYGEN"
-then
-AC_CONFIG_FILES([docs/doxy.cfg])
-fi
+DX_DOXYGEN_FEATURE([OFF])
+DX_DOT_FEATURE([OFF])
+DX_HTML_FEATURE([ON])
+DX_CHM_FEATURE([OFF])
+DX_CHI_FEATURE([OFF])
+DX_MAN_FEATURE([OFF])
+DX_RTF_FEATURE([OFF])
+DX_XML_FEATURE([OFF])
+DX_PDF_FEATURE([OFF])
+DX_PS_FEATURE([OFF])
+DX_INIT_DOXYGEN([$PACKAGE_NAME], [doxygen.cfg])
 
 ############### Ouput
 cwd=`pwd -P`; cd ${srcdir}; abs_srcdir=`pwd -P`; cd ${cwd}
@@ -364,12 +381,14 @@ 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(benchmarks/Makefile)
-AC_CONFIG_FILES(programs/Makefile)
-AC_CONFIG_FILES(programs/qed-fvol/Makefile)
+AC_CONFIG_FILES(extras/Makefile)
+AC_CONFIG_FILES(extras/Hadrons/Makefile)
+AC_CONFIG_FILES(extras/qed-fvol/Makefile)
 AC_OUTPUT
 
 echo "~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
@@ -384,7 +403,7 @@ os (target)                 : $target_os
 compiler vendor             : ${ax_cv_cxx_compiler_vendor}
 compiler version            : ${ax_cv_gxx_version}
 ----- BUILD OPTIONS -----------------------------------
-SIMD                        : ${ac_SIMD}
+SIMD                        : ${ac_SIMD}${SIMD_GEN_WIDTH_MSG}
 Threading                   : ${ac_openmp} 
 Communications type         : ${comms_type}
 Default precision           : ${ac_PRECISION}
@@ -392,8 +411,7 @@ 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`
-build DOXYGEN documentation : `if test "x$enable_doc" = xyes; then echo yes; else echo no; fi`
-graphs and diagrams         : `if test "x$enable_dot" = xyes; 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 -------------------------------------
 CXXFLAGS:
 `echo ${AM_CXXFLAGS} ${CXXFLAGS} | tr ' ' '\n' | sed 's/^-/    -/g'`

doxygen.inc

@@ -0,0 +1,184 @@
+# Copyright (C) 2004 Oren Ben-Kiki
+# This file is distributed under the same terms as the Automake macro files.
+
+# Generate automatic documentation using Doxygen. Goals and variables values
+# are controlled by the various DX_COND_??? conditionals set by autoconf.
+#
+# The provided goals are:
+# doxygen-doc: Generate all doxygen documentation.
+# doxygen-run: Run doxygen, which will generate some of the documentation
+#              (HTML, CHM, CHI, MAN, RTF, XML) but will not do the post
+#              processing required for the rest of it (PS, PDF, and some MAN).
+# doxygen-man: Rename some doxygen generated man pages.
+# doxygen-ps: Generate doxygen PostScript documentation.
+# doxygen-pdf: Generate doxygen PDF documentation.
+#
+# Note that by default these are not integrated into the automake goals. If
+# doxygen is used to generate man pages, you can achieve this integration by
+# setting man3_MANS to the list of man pages generated and then adding the
+# dependency:
+#
+#   $(man3_MANS): doxygen-doc
+#
+# This will cause make to run doxygen and generate all the documentation.
+#
+# The following variable is intended for use in Makefile.am:
+#
+# DX_CLEANFILES = everything to clean.
+#
+# This is usually added to MOSTLYCLEANFILES.
+
+## --------------------------------- ##
+## Format-independent Doxygen rules. ##
+## --------------------------------- ##
+
+if DX_COND_doc
+
+## ------------------------------- ##
+## Rules specific for HTML output. ##
+## ------------------------------- ##
+
+if DX_COND_html
+
+DX_CLEAN_HTML = @DX_DOCDIR@/html
+
+endif DX_COND_html
+
+## ------------------------------ ##
+## Rules specific for CHM output. ##
+## ------------------------------ ##
+
+if DX_COND_chm
+
+DX_CLEAN_CHM = @DX_DOCDIR@/chm
+
+if DX_COND_chi
+
+DX_CLEAN_CHI = @DX_DOCDIR@/@PACKAGE@.chi
+
+endif DX_COND_chi
+
+endif DX_COND_chm
+
+## ------------------------------ ##
+## Rules specific for MAN output. ##
+## ------------------------------ ##
+
+if DX_COND_man
+
+DX_CLEAN_MAN = @DX_DOCDIR@/man
+
+endif DX_COND_man
+
+## ------------------------------ ##
+## Rules specific for RTF output. ##
+## ------------------------------ ##
+
+if DX_COND_rtf
+
+DX_CLEAN_RTF = @DX_DOCDIR@/rtf
+
+endif DX_COND_rtf
+
+## ------------------------------ ##
+## Rules specific for XML output. ##
+## ------------------------------ ##
+
+if DX_COND_xml
+
+DX_CLEAN_XML = @DX_DOCDIR@/xml
+
+endif DX_COND_xml
+
+## ----------------------------- ##
+## Rules specific for PS output. ##
+## ----------------------------- ##
+
+if DX_COND_ps
+
+DX_CLEAN_PS = @DX_DOCDIR@/@PACKAGE@.ps
+
+DX_PS_GOAL = doxygen-ps
+
+doxygen-ps: @DX_DOCDIR@/@PACKAGE@.ps
+
+@DX_DOCDIR@/@PACKAGE@.ps: @DX_DOCDIR@/@PACKAGE@.tag
+	cd @DX_DOCDIR@/latex; \
+	rm -f *.aux *.toc *.idx *.ind *.ilg *.log *.out; \
+	$(DX_LATEX) refman.tex; \
+	$(MAKEINDEX_PATH) refman.idx; \
+	$(DX_LATEX) refman.tex; \
+	countdown=5; \
+	while $(DX_EGREP) 'Rerun (LaTeX|to get cross-references right)' \
+	                  refman.log > /dev/null 2>&1 \
+	   && test $$countdown -gt 0; do \
+	    $(DX_LATEX) refman.tex; \
+	    countdown=`expr $$countdown - 1`; \
+	done; \
+	$(DX_DVIPS) -o ../@PACKAGE@.ps refman.dvi
+
+endif DX_COND_ps
+
+## ------------------------------ ##
+## Rules specific for PDF output. ##
+## ------------------------------ ##
+
+if DX_COND_pdf
+
+DX_CLEAN_PDF = @DX_DOCDIR@/@PACKAGE@.pdf
+
+DX_PDF_GOAL = doxygen-pdf
+
+doxygen-pdf: @DX_DOCDIR@/@PACKAGE@.pdf
+
+@DX_DOCDIR@/@PACKAGE@.pdf: @DX_DOCDIR@/@PACKAGE@.tag
+	cd @DX_DOCDIR@/latex; \
+	rm -f *.aux *.toc *.idx *.ind *.ilg *.log *.out; \
+	$(DX_PDFLATEX) refman.tex; \
+	$(DX_MAKEINDEX) refman.idx; \
+	$(DX_PDFLATEX) refman.tex; \
+	countdown=5; \
+	while $(DX_EGREP) 'Rerun (LaTeX|to get cross-references right)' \
+	                  refman.log > /dev/null 2>&1 \
+	   && test $$countdown -gt 0; do \
+	    $(DX_PDFLATEX) refman.tex; \
+	    countdown=`expr $$countdown - 1`; \
+	done; \
+	mv refman.pdf ../@PACKAGE@.pdf
+
+endif DX_COND_pdf
+
+## ------------------------------------------------- ##
+## Rules specific for LaTeX (shared for PS and PDF). ##
+## ------------------------------------------------- ##
+
+if DX_COND_latex
+
+DX_CLEAN_LATEX = @DX_DOCDIR@/latex
+
+endif DX_COND_latex
+
+.INTERMEDIATE: doxygen-run $(DX_PS_GOAL) $(DX_PDF_GOAL)
+
+doxygen-run: @DX_DOCDIR@/@PACKAGE@.tag
+
+doxygen-doc: doxygen-run $(DX_PS_GOAL) $(DX_PDF_GOAL)
+
+@DX_DOCDIR@/@PACKAGE@.tag: $(DX_CONFIG) $(pkginclude_HEADERS)
+	rm -rf @DX_DOCDIR@
+	$(DX_ENV) $(DX_DOXYGEN) $(srcdir)/$(DX_CONFIG)
+
+DX_CLEANFILES = \
+    @DX_DOCDIR@/@PACKAGE@.tag \
+    -r \
+    $(DX_CLEAN_HTML) \
+    $(DX_CLEAN_CHM) \
+    $(DX_CLEAN_CHI) \
+    $(DX_CLEAN_MAN) \
+    $(DX_CLEAN_RTF) \
+    $(DX_CLEAN_XML) \
+    $(DX_CLEAN_PS) \
+    $(DX_CLEAN_PDF) \
+    $(DX_CLEAN_LATEX)
+
+endif DX_COND_doc

extras/Hadrons/Application.cc

@@ -0,0 +1,312 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Application.cc
+
+Copyright (C) 2015
+Copyright (C) 2016
+
+Author: Antonin Portelli <antonin.portelli@me.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/Application.hpp>
+#include <Grid/Hadrons/GeneticScheduler.hpp>
+
+using namespace Grid;
+using namespace QCD;
+using namespace Hadrons;
+
+#define BIG_SEP "==============="
+#define SEP     "---------------"
+
+/******************************************************************************
+ *                       Application implementation                           *
+ ******************************************************************************/
+// constructors ////////////////////////////////////////////////////////////////
+Application::Application(void)
+: env_(Environment::getInstance())
+{
+    LOG(Message) << "Modules available:" << std::endl;
+    auto list = ModuleFactory::getInstance().getBuilderList();
+    for (auto &m: list)
+    {
+        LOG(Message) << "  " << m << std::endl;
+    }
+    auto dim = GridDefaultLatt(), mpi = GridDefaultMpi(), loc(dim);
+    locVol_ = 1;
+    for (unsigned int d = 0; d < dim.size(); ++d)
+    {
+        loc[d]  /= mpi[d];
+        locVol_ *= loc[d];
+    }
+    LOG(Message) << "Global lattice: " << dim << std::endl;
+    LOG(Message) << "MPI partition : " << mpi << std::endl;
+    LOG(Message) << "Local lattice : " << loc << std::endl;
+}
+
+Application::Application(const Application::GlobalPar &par)
+: Application()
+{
+    setPar(par);
+}
+
+Application::Application(const std::string parameterFileName)
+: Application()
+{
+    parameterFileName_ = parameterFileName;
+}
+
+// access //////////////////////////////////////////////////////////////////////
+void Application::setPar(const Application::GlobalPar &par)
+{
+    par_ = par;
+    env_.setSeed(strToVec<int>(par_.seed));
+}
+
+const Application::GlobalPar & Application::getPar(void)
+{
+    return par_;
+}
+
+// execute /////////////////////////////////////////////////////////////////////
+void Application::run(void)
+{
+    if (!parameterFileName_.empty() and (env_.getNModule() == 0))
+    {
+        parseParameterFile(parameterFileName_);
+    }
+    if (!scheduled_)
+    {
+        schedule();
+    }
+    printSchedule();
+    configLoop();
+}
+
+// parse parameter file ////////////////////////////////////////////////////////
+class ObjectId: Serializable
+{
+public:
+    GRID_SERIALIZABLE_CLASS_MEMBERS(ObjectId,
+                                    std::string, name,
+                                    std::string, type);
+};
+
+void Application::parseParameterFile(const std::string parameterFileName)
+{
+    XmlReader reader(parameterFileName);
+    GlobalPar par;
+    ObjectId  id;
+    
+    LOG(Message) << "Building application from '" << parameterFileName << "'..." << std::endl;
+    read(reader, "parameters", par);
+    setPar(par);
+    push(reader, "modules");
+    push(reader, "module");
+    do
+    {
+        read(reader, "id", id);
+        env_.createModule(id.name, id.type, reader);
+    } while (reader.nextElement("module"));
+    pop(reader);
+    pop(reader);
+}
+
+void Application::saveParameterFile(const std::string parameterFileName)
+{
+    XmlWriter          writer(parameterFileName);
+    ObjectId           id;
+    const unsigned int nMod = env_.getNModule();
+    
+    LOG(Message) << "Saving application to '" << parameterFileName << "'..." << std::endl;
+    write(writer, "parameters", getPar());
+    push(writer, "modules");
+    for (unsigned int i = 0; i < nMod; ++i)
+    {
+        push(writer, "module");
+        id.name = env_.getModuleName(i);
+        id.type = env_.getModule(i)->getRegisteredName();
+        write(writer, "id", id);
+        env_.getModule(i)->saveParameters(writer, "options");
+        pop(writer);
+    }
+    pop(writer);
+    pop(writer);
+}
+
+// schedule computation ////////////////////////////////////////////////////////
+#define MEM_MSG(size)\
+sizeString((size)*locVol_) << " (" << sizeString(size)  << "/site)"
+
+#define DEFINE_MEMPEAK \
+auto memPeak = [this](const std::vector<unsigned int> &program)\
+{\
+    unsigned int memPeak;\
+    bool         msg;\
+    \
+    msg = HadronsLogMessage.isActive();\
+    HadronsLogMessage.Active(false);\
+    env_.dryRun(true);\
+    memPeak = env_.executeProgram(program);\
+    env_.dryRun(false);\
+    env_.freeAll();\
+    HadronsLogMessage.Active(true);\
+    \
+    return memPeak;\
+}
+
+void Application::schedule(void)
+{
+    DEFINE_MEMPEAK;
+    
+    // build module dependency graph
+    LOG(Message) << "Building module graph..." << std::endl;
+    auto graph = env_.makeModuleGraph();
+    auto con = graph.getConnectedComponents();
+    
+    // constrained topological sort using a genetic algorithm
+    LOG(Message) << "Scheduling computation..." << std::endl;
+    LOG(Message) << "               #module= " << graph.size() << std::endl;
+    LOG(Message) << "       population size= " << par_.genetic.popSize << std::endl;
+    LOG(Message) << "       max. generation= " << par_.genetic.maxGen << std::endl;
+    LOG(Message) << "  max. cst. generation= " << par_.genetic.maxCstGen << std::endl;
+    LOG(Message) << "         mutation rate= " << par_.genetic.mutationRate << std::endl;
+    
+    unsigned int                               k = 0, gen, prevPeak, nCstPeak = 0;
+    std::random_device                         rd;
+    GeneticScheduler<unsigned int>::Parameters par;
+    
+    par.popSize      = par_.genetic.popSize;
+    par.mutationRate = par_.genetic.mutationRate;
+    par.seed         = rd();
+    memPeak_         = 0;
+    CartesianCommunicator::BroadcastWorld(0, &(par.seed), sizeof(par.seed));
+    for (unsigned int i = 0; i < con.size(); ++i)
+    {
+        GeneticScheduler<unsigned int> scheduler(con[i], memPeak, par);
+        
+        gen = 0;
+        do
+        {
+            LOG(Debug) << "Generation " << gen << ":" << std::endl;
+            scheduler.nextGeneration();
+            if (gen != 0)
+            {
+                if (prevPeak == scheduler.getMinValue())
+                {
+                    nCstPeak++;
+                }
+                else
+                {
+                    nCstPeak = 0;
+                }
+            }
+            
+            prevPeak = scheduler.getMinValue();
+            if (gen % 10 == 0)
+            {
+                LOG(Iterative) << "Generation " << gen << ": "
+                               << MEM_MSG(scheduler.getMinValue()) << std::endl;
+            }
+            
+            gen++;
+        } while ((gen < par_.genetic.maxGen)
+                 and (nCstPeak < par_.genetic.maxCstGen));
+        auto &t = scheduler.getMinSchedule();
+        if (scheduler.getMinValue() > memPeak_)
+        {
+            memPeak_ = scheduler.getMinValue();
+        }
+        for (unsigned int j = 0; j < t.size(); ++j)
+        {
+            program_.push_back(t[j]);
+        }
+    }
+    scheduled_ = true;
+}
+
+void Application::saveSchedule(const std::string filename)
+{
+    TextWriter               writer(filename);
+    std::vector<std::string> program;
+    
+    if (!scheduled_)
+    {
+        HADRON_ERROR("Computation not scheduled");
+    }
+    LOG(Message) << "Saving current schedule to '" << filename << "'..."
+                 << std::endl;
+    for (auto address: program_)
+    {
+        program.push_back(env_.getModuleName(address));
+    }
+    write(writer, "schedule", program);
+}
+
+void Application::loadSchedule(const std::string filename)
+{
+    DEFINE_MEMPEAK;
+    
+    TextReader               reader(filename);
+    std::vector<std::string> program;
+    
+    LOG(Message) << "Loading schedule from '" << filename << "'..."
+                 << std::endl;
+    read(reader, "schedule", program);
+    program_.clear();
+    for (auto &name: program)
+    {
+        program_.push_back(env_.getModuleAddress(name));
+    }
+    scheduled_ = true;
+    memPeak_   = memPeak(program_);
+}
+
+void Application::printSchedule(void)
+{
+    if (!scheduled_)
+    {
+        HADRON_ERROR("Computation not scheduled");
+    }
+    LOG(Message) << "Schedule (memory peak: " << MEM_MSG(memPeak_) << "):"
+                 << std::endl;
+    for (unsigned int i = 0; i < program_.size(); ++i)
+    {
+        LOG(Message) << std::setw(4) << i + 1 << ": "
+                     << env_.getModuleName(program_[i]) << std::endl;
+    }
+}
+
+// loop on configurations //////////////////////////////////////////////////////
+void Application::configLoop(void)
+{
+    auto range = par_.trajCounter;
+    
+    for (unsigned int t = range.start; t < range.end; t += range.step)
+    {
+        LOG(Message) << BIG_SEP << " Starting measurement for trajectory " << t
+                     << " " << BIG_SEP << std::endl;
+        env_.setTrajectory(t);
+        env_.executeProgram(program_);
+        env_.freeAll();
+    }
+    LOG(Message) << BIG_SEP << " End of measurement " << BIG_SEP << std::endl;
+}

extras/Hadrons/Application.hpp

@@ -0,0 +1,130 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Application.hpp
+
+Copyright (C) 2015
+Copyright (C) 2016
+
+Author: Antonin Portelli <antonin.portelli@me.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_Application_hpp_
+#define Hadrons_Application_hpp_
+
+#include <Grid/Hadrons/Global.hpp>
+#include <Grid/Hadrons/Environment.hpp>
+#include <Grid/Hadrons/ModuleFactory.hpp>
+#include <Grid/Hadrons/Modules.hpp>
+
+BEGIN_HADRONS_NAMESPACE
+
+/******************************************************************************
+ *                         Main program manager                               *
+ ******************************************************************************/
+class Application
+{
+public:
+    class TrajRange: Serializable
+    {
+    public:
+        GRID_SERIALIZABLE_CLASS_MEMBERS(TrajRange,
+                                        unsigned int, start,
+                                        unsigned int, end,
+                                        unsigned int, step);
+    };
+    class GeneticPar: Serializable
+    {
+    public:
+        GeneticPar(void):
+            popSize{20}, maxGen{1000}, maxCstGen{100}, mutationRate{.1} {};
+    public:
+        GRID_SERIALIZABLE_CLASS_MEMBERS(GeneticPar,
+                                        unsigned int, popSize,
+                                        unsigned int, maxGen,
+                                        unsigned int, maxCstGen,
+                                        double      , mutationRate);
+    };
+    class GlobalPar: Serializable
+    {
+    public:
+        GRID_SERIALIZABLE_CLASS_MEMBERS(GlobalPar,
+                                        TrajRange,   trajCounter,
+                                        GeneticPar,  genetic,
+                                        std::string, seed);
+    };
+public:
+    // constructors
+    Application(void);
+    Application(const GlobalPar &par);
+    Application(const std::string parameterFileName);
+    // destructor
+    virtual ~Application(void) = default;
+    // access
+    void              setPar(const GlobalPar &par);
+    const GlobalPar & getPar(void);
+    // module creation
+    template <typename M>
+    void createModule(const std::string name);
+    template <typename M>
+    void createModule(const std::string name, const typename M::Par &par);
+    // execute
+    void run(void);
+    // XML parameter file I/O
+    void parseParameterFile(const std::string parameterFileName);
+    void saveParameterFile(const std::string parameterFileName);
+    // schedule computation
+    void schedule(void);
+    void saveSchedule(const std::string filename);
+    void loadSchedule(const std::string filename);
+    void printSchedule(void);
+    // loop on configurations
+    void configLoop(void);
+private:
+    long unsigned int         locVol_;
+    std::string               parameterFileName_{""};
+    GlobalPar                 par_;
+    Environment               &env_;
+    std::vector<unsigned int> program_;
+    Environment::Size         memPeak_;
+    bool                      scheduled_{false};
+};
+
+/******************************************************************************
+ *                     Application template implementation                    *
+ ******************************************************************************/
+// module creation /////////////////////////////////////////////////////////////
+template <typename M>
+void Application::createModule(const std::string name)
+{
+    env_.createModule<M>(name);
+}
+
+template <typename M>
+void Application::createModule(const std::string name,
+                               const typename M::Par &par)
+{
+    env_.createModule<M>(name, par);
+}
+
+END_HADRONS_NAMESPACE
+
+#endif // Hadrons_Application_hpp_

extras/Hadrons/Environment.cc

@@ -0,0 +1,702 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Environment.cc
+
+Copyright (C) 2015
+Copyright (C) 2016
+
+Author: Antonin Portelli <antonin.portelli@me.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/Environment.hpp>
+#include <Grid/Hadrons/Module.hpp>
+#include <Grid/Hadrons/ModuleFactory.hpp>
+
+using namespace Grid;
+using namespace QCD;
+using namespace Hadrons;
+
+/******************************************************************************
+ *                       Environment implementation                           *
+ ******************************************************************************/
+// constructor /////////////////////////////////////////////////////////////////
+Environment::Environment(void)
+{
+    grid4d_.reset(SpaceTimeGrid::makeFourDimGrid(
+        GridDefaultLatt(), GridDefaultSimd(Nd, vComplex::Nsimd()),
+        GridDefaultMpi()));
+    gridRb4d_.reset(SpaceTimeGrid::makeFourDimRedBlackGrid(grid4d_.get()));
+    auto loc = getGrid()->LocalDimensions();
+    locVol_ = 1;
+    for (unsigned int d = 0; d < loc.size(); ++d)
+    {
+        locVol_ *= loc[d];
+    }
+    rng4d_.reset(new GridParallelRNG(grid4d_.get()));
+}
+
+// dry run /////////////////////////////////////////////////////////////////////
+void Environment::dryRun(const bool isDry)
+{
+    dryRun_ = isDry;
+}
+
+bool Environment::isDryRun(void) const
+{
+    return dryRun_;
+}
+
+// trajectory number ///////////////////////////////////////////////////////////
+void Environment::setTrajectory(const unsigned int traj)
+{
+    traj_ = traj;
+}
+
+unsigned int Environment::getTrajectory(void) const
+{
+    return traj_;
+}
+
+// grids ///////////////////////////////////////////////////////////////////////
+void Environment::createGrid(const unsigned int Ls)
+{
+    if (grid5d_.find(Ls) == grid5d_.end())
+    {
+        auto g = getGrid();
+        
+        grid5d_[Ls].reset(SpaceTimeGrid::makeFiveDimGrid(Ls, g));
+        gridRb5d_[Ls].reset(SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls, g));
+    }
+}
+
+GridCartesian * Environment::getGrid(const unsigned int Ls) const
+{
+    try
+    {
+        if (Ls == 1)
+        {
+            return grid4d_.get();
+        }
+        else
+        {
+            return grid5d_.at(Ls).get();
+        }
+    }
+    catch(std::out_of_range &)
+    {
+        HADRON_ERROR("no grid with Ls= " << Ls);
+    }
+}
+
+GridRedBlackCartesian * Environment::getRbGrid(const unsigned int Ls) const
+{
+    try
+    {
+        if (Ls == 1)
+        {
+            return gridRb4d_.get();
+        }
+        else
+        {
+            return gridRb5d_.at(Ls).get();
+        }
+    }
+    catch(std::out_of_range &)
+    {
+        HADRON_ERROR("no red-black 5D grid with Ls= " << Ls);
+    }
+}
+
+// random number generator /////////////////////////////////////////////////////
+void Environment::setSeed(const std::vector<int> &seed)
+{
+    rng4d_->SeedFixedIntegers(seed);
+}
+
+GridParallelRNG * Environment::get4dRng(void) const
+{
+    return rng4d_.get();
+}
+
+// module management ///////////////////////////////////////////////////////////
+void Environment::pushModule(Environment::ModPt &pt)
+{
+    std::string name = pt->getName();
+    
+    if (!hasModule(name))
+    {
+        std::vector<unsigned int> inputAddress;
+        unsigned int              address;
+        ModuleInfo                m;
+        
+        m.data = std::move(pt);
+        m.type = typeIdPt(*m.data.get());
+        m.name = name;
+        auto input  = m.data->getInput();
+        for (auto &in: input)
+        {
+            if (!hasObject(in))
+            {
+                addObject(in , -1);
+            }
+            m.input.push_back(objectAddress_[in]);
+        }
+        auto output = m.data->getOutput();
+        module_.push_back(std::move(m));
+        address              = static_cast<unsigned int>(module_.size() - 1);
+        moduleAddress_[name] = address;
+        for (auto &out: output)
+        {
+            if (!hasObject(out))
+            {
+                addObject(out, address);
+            }
+            else
+            {
+                if (object_[objectAddress_[out]].module < 0)
+                {
+                    object_[objectAddress_[out]].module = address;
+                }
+                else
+                {
+                    HADRON_ERROR("object '" + out
+                                 + "' is already produced by module '"
+                                 + module_[object_[getObjectAddress(out)].module].name
+                                 + "' (while pushing module '" + name + "')");
+                }
+            }
+        }
+    }
+    else
+    {
+        HADRON_ERROR("module '" + name + "' already exists");
+    }
+}
+
+unsigned int Environment::getNModule(void) const
+{
+    return module_.size();
+}
+
+void Environment::createModule(const std::string name, const std::string type,
+                               XmlReader &reader)
+{
+    auto &factory = ModuleFactory::getInstance();
+    auto pt       = factory.create(type, name);
+    
+    pt->parseParameters(reader, "options");
+    pushModule(pt);
+}
+
+ModuleBase * Environment::getModule(const unsigned int address) const
+{
+    if (hasModule(address))
+    {
+        return module_[address].data.get();
+    }
+    else
+    {
+        HADRON_ERROR("no module with address " + std::to_string(address));
+    }
+}
+
+ModuleBase * Environment::getModule(const std::string name) const
+{
+    return getModule(getModuleAddress(name));
+}
+
+unsigned int Environment::getModuleAddress(const std::string name) const
+{
+    if (hasModule(name))
+    {
+        return moduleAddress_.at(name);
+    }
+    else
+    {
+        HADRON_ERROR("no module with name '" + name + "'");
+    }
+}
+
+std::string Environment::getModuleName(const unsigned int address) const
+{
+    if (hasModule(address))
+    {
+        return module_[address].name;
+    }
+    else
+    {
+        HADRON_ERROR("no module with address " + std::to_string(address));
+    }
+}
+
+std::string Environment::getModuleType(const unsigned int address) const
+{
+    if (hasModule(address))
+    {
+        return typeName(module_[address].type);
+    }
+    else
+    {
+        HADRON_ERROR("no module with address " + std::to_string(address));
+    }
+}
+
+std::string Environment::getModuleType(const std::string name) const
+{
+    return getModuleType(getModuleAddress(name));
+}
+
+bool Environment::hasModule(const unsigned int address) const
+{
+    return (address < module_.size());
+}
+
+bool Environment::hasModule(const std::string name) const
+{
+    return (moduleAddress_.find(name) != moduleAddress_.end());
+}
+
+Graph<unsigned int> Environment::makeModuleGraph(void) const
+{
+    Graph<unsigned int> moduleGraph;
+    
+    for (unsigned int i = 0; i < module_.size(); ++i)
+    {
+        moduleGraph.addVertex(i);
+        for (auto &j: module_[i].input)
+        {
+            moduleGraph.addEdge(object_[j].module, i);
+        }
+    }
+    
+    return moduleGraph;
+}
+
+#define BIG_SEP "==============="
+#define SEP     "---------------"
+#define MEM_MSG(size)\
+sizeString((size)*locVol_) << " (" << sizeString(size)  << "/site)"
+
+Environment::Size
+Environment::executeProgram(const std::vector<unsigned int> &p)
+{
+    Size                                memPeak = 0, sizeBefore, sizeAfter;
+    std::vector<std::set<unsigned int>> freeProg;
+    bool                                continueCollect, nothingFreed;
+    
+    // build garbage collection schedule
+    freeProg.resize(p.size());
+    for (unsigned int i = 0; i < object_.size(); ++i)
+    {
+        auto pred = [i, this](const unsigned int j)
+        {
+            auto &in = module_[j].input;
+            auto it  = std::find(in.begin(), in.end(), i);
+            
+            return (it != in.end()) or (j == object_[i].module);
+        };
+        auto it = std::find_if(p.rbegin(), p.rend(), pred);
+        if (it != p.rend())
+        {
+            freeProg[p.rend() - it - 1].insert(i);
+        }
+    }
+    
+    // program execution
+    for (unsigned int i = 0; i < p.size(); ++i)
+    {
+        // execute module
+        if (!isDryRun())
+        {
+            LOG(Message) << SEP << " Measurement step " << i+1 << "/"
+                         << p.size() << " (module '" << module_[p[i]].name
+                         << "') " << SEP << std::endl;
+        }
+        (*module_[p[i]].data)();
+        sizeBefore = getTotalSize();
+        // print used memory after execution
+        if (!isDryRun())
+        {
+            LOG(Message) << "Allocated objects: " << MEM_MSG(sizeBefore)
+                         << std::endl;
+        }
+        if (sizeBefore > memPeak)
+        {
+            memPeak = sizeBefore;
+        }
+        // garbage collection for step i
+        if (!isDryRun())
+        {
+            LOG(Message) << "Garbage collection..." << std::endl;
+        }
+        nothingFreed = true;
+        do
+        {
+            continueCollect = false;
+            auto toFree = freeProg[i];
+            for (auto &j: toFree)
+            {
+                // continue garbage collection while there are still
+                // objects without owners
+                continueCollect = continueCollect or !hasOwners(j);
+                if(freeObject(j))
+                {
+                    // if an object has been freed, remove it from
+                    // the garbage collection schedule
+                    freeProg[i].erase(j);
+                    nothingFreed = false;
+                }
+            }
+        } while (continueCollect);
+        // any remaining objects in step i garbage collection schedule
+        // is scheduled for step i + 1
+        if (i + 1 < p.size())
+        {
+            for (auto &j: freeProg[i])
+            {
+                freeProg[i + 1].insert(j);
+            }
+        }
+        // print used memory after garbage collection if necessary
+        if (!isDryRun())
+        {
+            sizeAfter = getTotalSize();
+            if (sizeBefore != sizeAfter)
+            {
+                LOG(Message) << "Allocated objects: " << MEM_MSG(sizeAfter)
+                             << std::endl;
+            }
+            else
+            {
+                LOG(Message) << "Nothing to free" << std::endl;
+            }
+        }
+    }
+    
+    return memPeak;
+}
+
+Environment::Size Environment::executeProgram(const std::vector<std::string> &p)
+{
+    std::vector<unsigned int> pAddress;
+    
+    for (auto &n: p)
+    {
+        pAddress.push_back(getModuleAddress(n));
+    }
+    
+    return executeProgram(pAddress);
+}
+
+// general memory management ///////////////////////////////////////////////////
+void Environment::addObject(const std::string name, const int moduleAddress)
+{
+    ObjInfo info;
+    
+    info.name   = name;
+    info.module = moduleAddress;
+    object_.push_back(std::move(info));
+    objectAddress_[name] = static_cast<unsigned int>(object_.size() - 1);
+}
+
+void Environment::registerObject(const unsigned int address,
+                                 const unsigned int size, const unsigned int Ls)
+{
+    if (!hasRegisteredObject(address))
+    {
+        if (hasObject(address))
+        {
+            object_[address].size         = size;
+            object_[address].Ls           = Ls;
+            object_[address].isRegistered = true;
+        }
+        else
+        {
+            HADRON_ERROR("no object with address " + std::to_string(address));
+        }
+    }
+    else
+    {
+        HADRON_ERROR("object with address " + std::to_string(address)
+                     + " already registered");
+    }
+}
+
+void Environment::registerObject(const std::string name,
+                                 const unsigned int size, const unsigned int Ls)
+{
+    registerObject(getObjectAddress(name), size, Ls);
+}
+
+unsigned int Environment::getObjectAddress(const std::string name) const
+{
+    if (hasObject(name))
+    {
+        return objectAddress_.at(name);
+    }
+    else
+    {
+        HADRON_ERROR("no object with name '" + name + "'");
+    }
+}
+
+std::string Environment::getObjectName(const unsigned int address) const
+{
+    if (hasObject(address))
+    {
+        return object_[address].name;
+    }
+    else
+    {
+        HADRON_ERROR("no object with address " + std::to_string(address));
+    }
+}
+
+std::string Environment::getObjectType(const unsigned int address) const
+{
+    if (hasRegisteredObject(address))
+    {
+        return typeName(object_[address].type);
+    }
+    else if (hasObject(address))
+    {
+        HADRON_ERROR("object with address " + std::to_string(address)
+                     + " exists but is not registered");
+    }
+    else
+    {
+        HADRON_ERROR("no object with address " + std::to_string(address));
+    }
+}
+
+std::string Environment::getObjectType(const std::string name) const
+{
+    return getObjectType(getObjectAddress(name));
+}
+
+Environment::Size Environment::getObjectSize(const unsigned int address) const
+{
+    if (hasRegisteredObject(address))
+    {
+        return object_[address].size;
+    }
+    else if (hasObject(address))
+    {
+        HADRON_ERROR("object with address " + std::to_string(address)
+                     + " exists but is not registered");
+    }
+    else
+    {
+        HADRON_ERROR("no object with address " + std::to_string(address));
+    }
+}
+
+Environment::Size Environment::getObjectSize(const std::string name) const
+{
+    return getObjectSize(getObjectAddress(name));
+}
+
+unsigned int Environment::getObjectLs(const unsigned int address) const
+{
+    if (hasRegisteredObject(address))
+    {
+        return object_[address].Ls;
+    }
+    else if (hasObject(address))
+    {
+        HADRON_ERROR("object with address " + std::to_string(address)
+                     + " exists but is not registered");
+    }
+    else
+    {
+        HADRON_ERROR("no object with address " + std::to_string(address));
+    }
+}
+
+unsigned int Environment::getObjectLs(const std::string name) const
+{
+    return getObjectLs(getObjectAddress(name));
+}
+
+bool Environment::hasObject(const unsigned int address) const
+{
+    return (address < object_.size());
+}
+
+bool Environment::hasObject(const std::string name) const
+{
+    auto it = objectAddress_.find(name);
+    
+    return ((it != objectAddress_.end()) and hasObject(it->second));
+}
+
+bool Environment::hasRegisteredObject(const unsigned int address) const
+{
+    if (hasObject(address))
+    {
+        return object_[address].isRegistered;
+    }
+    else
+    {
+        return false;
+    }
+}
+
+bool Environment::hasRegisteredObject(const std::string name) const
+{
+    if (hasObject(name))
+    {
+        return hasRegisteredObject(getObjectAddress(name));
+    }
+    else
+    {
+        return false;
+    }
+}
+
+bool Environment::isObject5d(const unsigned int address) const
+{
+    return (getObjectLs(address) > 1);
+}
+
+bool Environment::isObject5d(const std::string name) const
+{
+    return (getObjectLs(name) > 1);
+}
+
+Environment::Size Environment::getTotalSize(void) const
+{
+    Environment::Size size = 0;
+    
+    for (auto &o: object_)
+    {
+        if (o.isRegistered)
+        {
+            size += o.size;
+        }
+    }
+    
+    return size;
+}
+
+void Environment::addOwnership(const unsigned int owner,
+                               const unsigned int property)
+{
+    if (hasObject(property))
+    {
+        object_[property].owners.insert(owner);
+    }
+    else
+    {
+        HADRON_ERROR("no object with address " + std::to_string(property));
+    }
+    if (hasObject(owner))
+    {
+        object_[owner].properties.insert(property);
+    }
+    else
+    {
+        HADRON_ERROR("no object with address " + std::to_string(owner));
+    }
+}
+
+void Environment::addOwnership(const std::string owner,
+                               const std::string property)
+{
+    addOwnership(getObjectAddress(owner), getObjectAddress(property));
+}
+
+bool Environment::hasOwners(const unsigned int address) const
+{
+    
+    if (hasObject(address))
+    {
+        return (!object_[address].owners.empty());
+    }
+    else
+    {
+        HADRON_ERROR("no object with address " + std::to_string(address));
+    }
+}
+
+bool Environment::hasOwners(const std::string name) const
+{
+    return hasOwners(getObjectAddress(name));
+}
+
+bool Environment::freeObject(const unsigned int address)
+{
+    if (!hasOwners(address))
+    {
+        if (!isDryRun() and object_[address].isRegistered)
+        {
+            LOG(Message) << "Destroying object '" << object_[address].name
+                         << "'" << std::endl;
+        }
+        for (auto &p: object_[address].properties)
+        {
+            object_[p].owners.erase(address);
+        }
+        object_[address].size         = 0;
+        object_[address].Ls           = 0;
+        object_[address].isRegistered = false;
+        object_[address].type         = nullptr;
+        object_[address].owners.clear();
+        object_[address].properties.clear();
+        object_[address].data.reset(nullptr);
+        
+        return true;
+    }
+    else
+    {
+        return false;
+    }
+}
+
+bool Environment::freeObject(const std::string name)
+{
+    return freeObject(getObjectAddress(name));
+}
+
+void Environment::freeAll(void)
+{
+    for (unsigned int i = 0; i < object_.size(); ++i)
+    {
+        freeObject(i);
+    }
+}
+
+void Environment::printContent(void)
+{
+    LOG(Message) << "Modules: " << std::endl;
+    for (unsigned int i = 0; i < module_.size(); ++i)
+    {
+        LOG(Message) << std::setw(4) << i << ": "
+                     << getModuleName(i) << std::endl;
+    }
+    LOG(Message) << "Objects: " << std::endl;
+    for (unsigned int i = 0; i < object_.size(); ++i)
+    {
+        LOG(Message) << std::setw(4) << i << ": "
+                     << getObjectName(i) << std::endl;
+    }
+}

extras/Hadrons/Environment.hpp

@@ -0,0 +1,381 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Environment.hpp
+
+Copyright (C) 2015
+Copyright (C) 2016
+
+Author: Antonin Portelli <antonin.portelli@me.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_Environment_hpp_
+#define Hadrons_Environment_hpp_
+
+#include <Grid/Hadrons/Global.hpp>
+#include <Grid/Hadrons/Graph.hpp>
+
+#ifndef SITE_SIZE_TYPE
+#define SITE_SIZE_TYPE unsigned int
+#endif
+
+BEGIN_HADRONS_NAMESPACE
+
+/******************************************************************************
+ *                         Global environment                                 *
+ ******************************************************************************/
+// forward declaration of Module
+class ModuleBase;
+
+class Object
+{
+public:
+    Object(void) = default;
+    virtual ~Object(void) = default;
+};
+
+template <typename T>
+class Holder: public Object
+{
+public:
+    Holder(void) = default;
+    Holder(T *pt);
+    virtual ~Holder(void) = default;
+    T &       get(void) const;
+    T *       getPt(void) const;
+    void      reset(T *pt);
+private:
+    std::unique_ptr<T> objPt_{nullptr};
+};
+
+class Environment
+{
+    SINGLETON(Environment);
+public:
+    typedef SITE_SIZE_TYPE                         Size;
+    typedef std::unique_ptr<ModuleBase>            ModPt;
+    typedef std::unique_ptr<GridCartesian>         GridPt;
+    typedef std::unique_ptr<GridRedBlackCartesian> GridRbPt;
+    typedef std::unique_ptr<GridParallelRNG>       RngPt;
+    typedef std::unique_ptr<LatticeBase>           LatticePt;
+private:
+    struct ModuleInfo
+    {
+        const std::type_info      *type{nullptr};
+        std::string               name;
+        ModPt                     data{nullptr};
+        std::vector<unsigned int> input;
+    };
+    struct ObjInfo
+    {
+        Size                    size{0};
+        unsigned int            Ls{0};
+        bool                    isRegistered{false};
+        const std::type_info    *type{nullptr};
+        std::string             name;
+        int                     module{-1};
+        std::set<unsigned int>  owners, properties;
+        std::unique_ptr<Object> data{nullptr};
+    };
+public:
+    // dry run
+    void                    dryRun(const bool isDry);
+    bool                    isDryRun(void) const;
+    // trajectory number
+    void                    setTrajectory(const unsigned int traj);
+    unsigned int            getTrajectory(void) const;
+    // grids
+    void                    createGrid(const unsigned int Ls);
+    GridCartesian *         getGrid(const unsigned int Ls = 1) const;
+    GridRedBlackCartesian * getRbGrid(const unsigned int Ls = 1) const;
+    // random number generator
+    void                    setSeed(const std::vector<int> &seed);
+    GridParallelRNG *       get4dRng(void) const;
+    // module management
+    void                    pushModule(ModPt &pt);
+    template <typename M>
+    void                    createModule(const std::string name);
+    template <typename M>
+    void                    createModule(const std::string name,
+                                         const typename M::Par &par);
+    void                    createModule(const std::string name,
+                                         const std::string type,
+                                         XmlReader &reader);
+    unsigned int            getNModule(void) const;
+    ModuleBase *            getModule(const unsigned int address) const;
+    ModuleBase *            getModule(const std::string name) const;
+    template <typename M>
+    M *                     getModule(const unsigned int address) const;
+    template <typename M>
+    M *                     getModule(const std::string name) const;
+    unsigned int            getModuleAddress(const std::string name) const;
+    std::string             getModuleName(const unsigned int address) const;
+    std::string             getModuleType(const unsigned int address) const;
+    std::string             getModuleType(const std::string name) const;
+    bool                    hasModule(const unsigned int address) const;
+    bool                    hasModule(const std::string name) const;
+    Graph<unsigned int>     makeModuleGraph(void) const;
+    Size                    executeProgram(const std::vector<unsigned int> &p);
+    Size                    executeProgram(const std::vector<std::string> &p);
+    // general memory management
+    void                    addObject(const std::string name,
+                                      const int moduleAddress);
+    void                    registerObject(const unsigned int address,
+                                           const unsigned int size,
+                                           const unsigned int Ls = 1);
+    void                    registerObject(const std::string name,
+                                           const unsigned int size,
+                                           const unsigned int Ls = 1);
+    template <typename T>
+    unsigned int            lattice4dSize(void) const;
+    template <typename T>
+    void                    registerLattice(const unsigned int address,
+                                            const unsigned int Ls = 1);
+    template <typename T>
+    void                    registerLattice(const std::string name,
+                                            const unsigned int Ls = 1);
+    template <typename T>
+    void                    setObject(const unsigned int address, T *object);
+    template <typename T>
+    void                    setObject(const std::string name, T *object);
+    template <typename T>
+    T *                     getObject(const unsigned int address) const;
+    template <typename T>
+    T *                     getObject(const std::string name) const;
+    template <typename T>
+    T *                     createLattice(const unsigned int address);
+    template <typename T>
+    T *                     createLattice(const std::string name);
+    unsigned int            getObjectAddress(const std::string name) const;
+    std::string             getObjectName(const unsigned int address) const;
+    std::string             getObjectType(const unsigned int address) const;
+    std::string             getObjectType(const std::string name) const;
+    Size                    getObjectSize(const unsigned int address) const;
+    Size                    getObjectSize(const std::string name) const;
+    unsigned int            getObjectLs(const unsigned int address) const;
+    unsigned int            getObjectLs(const std::string name) const;
+    bool                    hasObject(const unsigned int address) const;
+    bool                    hasObject(const std::string name) const;
+    bool                    hasRegisteredObject(const unsigned int address) const;
+    bool                    hasRegisteredObject(const std::string name) const;
+    bool                    isObject5d(const unsigned int address) const;
+    bool                    isObject5d(const std::string name) const;
+    Environment::Size       getTotalSize(void) const;
+    void                    addOwnership(const unsigned int owner,
+                                         const unsigned int property);
+    void                    addOwnership(const std::string owner,
+                                         const std::string property);
+    bool                    hasOwners(const unsigned int address) const;
+    bool                    hasOwners(const std::string name) const;
+    bool                    freeObject(const unsigned int address);
+    bool                    freeObject(const std::string name);
+    void                    freeAll(void);
+    void                    printContent(void);
+private:
+    // general
+    bool                                   dryRun_{false};
+    unsigned int                           traj_, locVol_;
+    // grids
+    GridPt                                 grid4d_;
+    std::map<unsigned int, GridPt>         grid5d_;
+    GridRbPt                               gridRb4d_;
+    std::map<unsigned int, GridRbPt>       gridRb5d_;
+    // random number generator
+    RngPt                                  rng4d_;
+    // module and related maps
+    std::vector<ModuleInfo>                module_;
+    std::map<std::string, unsigned int>    moduleAddress_;
+    // lattice store
+    std::map<unsigned int, LatticePt>      lattice_;
+    // object store
+    std::vector<ObjInfo>                   object_;
+    std::map<std::string, unsigned int>    objectAddress_;
+};
+
+/******************************************************************************
+ *                       Holder template implementation                       *
+ ******************************************************************************/
+// constructor /////////////////////////////////////////////////////////////////
+template <typename T>
+Holder<T>::Holder(T *pt)
+: objPt_(pt)
+{}
+
+// access //////////////////////////////////////////////////////////////////////
+template <typename T>
+T & Holder<T>::get(void) const
+{
+    return &objPt_.get();
+}
+
+template <typename T>
+T * Holder<T>::getPt(void) const
+{
+    return objPt_.get();
+}
+
+template <typename T>
+void Holder<T>::reset(T *pt)
+{
+    objPt_.reset(pt);
+}
+
+/******************************************************************************
+ *                     Environment template implementation                    *
+ ******************************************************************************/
+// module management ///////////////////////////////////////////////////////////
+template <typename M>
+void Environment::createModule(const std::string name)
+{
+    ModPt pt(new M(name));
+    
+    pushModule(pt);
+}
+
+template <typename M>
+void Environment::createModule(const std::string name,
+                               const typename M::Par &par)
+{
+    ModPt pt(new M(name));
+    
+    static_cast<M *>(pt.get())->setPar(par);
+    pushModule(pt);
+}
+
+template <typename M>
+M * Environment::getModule(const unsigned int address) const
+{
+    if (auto *pt = dynamic_cast<M *>(getModule(address)))
+    {
+        return pt;
+    }
+    else
+    {
+        HADRON_ERROR("module '" + module_[address].name
+                     + "' does not have type " + typeid(M).name()
+                     + "(object type: " + getModuleType(address) + ")");
+    }
+}
+
+template <typename M>
+M * Environment::getModule(const std::string name) const
+{
+    return getModule<M>(getModuleAddress(name));
+}
+
+template <typename T>
+unsigned int Environment::lattice4dSize(void) const
+{
+    return sizeof(typename T::vector_object)/getGrid()->Nsimd();
+}
+
+template <typename T>
+void Environment::registerLattice(const unsigned int address,
+                                  const unsigned int Ls)
+{
+    createGrid(Ls);
+    registerObject(address, Ls*lattice4dSize<T>(), Ls);
+}
+
+template <typename T>
+void Environment::registerLattice(const std::string name, const unsigned int Ls)
+{
+    createGrid(Ls);
+    registerObject(name, Ls*lattice4dSize<T>(), Ls);
+}
+
+template <typename T>
+void Environment::setObject(const unsigned int address, T *object)
+{
+    if (hasRegisteredObject(address))
+    {
+        object_[address].data.reset(new Holder<T>(object));
+        object_[address].type = &typeid(T);
+    }
+    else if (hasObject(address))
+    {
+        HADRON_ERROR("object with address " + std::to_string(address) +
+                     " exists but is not registered");
+    }
+    else
+    {
+        HADRON_ERROR("no object with address " + std::to_string(address));
+    }
+}
+
+template <typename T>
+void Environment::setObject(const std::string name, T *object)
+{
+    setObject(getObjectAddress(name), object);
+}
+
+template <typename T>
+T * Environment::getObject(const unsigned int address) const
+{
+    if (hasRegisteredObject(address))
+    {
+        if (auto h = dynamic_cast<Holder<T> *>(object_[address].data.get()))
+        {
+            return h->getPt();
+        }
+        else
+        {
+            HADRON_ERROR("object with address " + std::to_string(address) +
+                         " does not have type '" + typeid(T).name() +
+                         "' (has type '" + getObjectType(address) + "')");
+        }
+    }
+    else if (hasObject(address))
+    {
+        HADRON_ERROR("object with address " + std::to_string(address) +
+                     " exists but is not registered");
+    }
+    else
+    {
+        HADRON_ERROR("no object with address " + std::to_string(address));
+    }
+}
+
+template <typename T>
+T * Environment::getObject(const std::string name) const
+{
+    return getObject<T>(getObjectAddress(name));
+}
+
+template <typename T>
+T * Environment::createLattice(const unsigned int address)
+{
+    GridCartesian *g = getGrid(getObjectLs(address));
+    
+    setObject(address, new T(g));
+    
+    return getObject<T>(address);
+}
+
+template <typename T>
+T * Environment::createLattice(const std::string name)
+{
+    return createLattice<T>(getObjectAddress(name));
+}
+
+END_HADRONS_NAMESPACE
+
+#endif // Hadrons_Environment_hpp_

extras/Hadrons/Factory.hpp

@@ -0,0 +1,106 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Factory.hpp
+
+Copyright (C) 2015
+Copyright (C) 2016
+
+Author: Antonin Portelli <antonin.portelli@me.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_Factory_hpp_
+#define Hadrons_Factory_hpp_
+
+#include <Grid/Hadrons/Global.hpp>
+
+BEGIN_HADRONS_NAMESPACE
+
+/******************************************************************************
+ *                        abstract factory class                              *
+ ******************************************************************************/
+template <typename T>
+class Factory
+{
+public:
+    typedef std::function<std::unique_ptr<T>(const std::string)> Func;
+public:
+    // constructor
+    Factory(void) = default;
+    // destructor
+    virtual ~Factory(void) = default;
+    // registration
+    void registerBuilder(const std::string type, const Func &f);
+    // get builder list
+    std::vector<std::string> getBuilderList(void) const;
+    // factory
+    std::unique_ptr<T> create(const std::string type,
+                              const std::string name) const;
+private:
+    std::map<std::string, Func> builder_;
+};
+
+/******************************************************************************
+ *                         template implementation                            *
+ ******************************************************************************/
+// registration ////////////////////////////////////////////////////////////////
+template <typename T>
+void Factory<T>::registerBuilder(const std::string type, const Func &f)
+{
+    builder_[type] = f;
+}
+
+// get module list /////////////////////////////////////////////////////////////
+template <typename T>
+std::vector<std::string> Factory<T>::getBuilderList(void) const
+{
+    std::vector<std::string> list;
+    
+    for (auto &b: builder_)
+    {
+        list.push_back(b.first);
+    }
+    
+    return list;
+}
+
+// factory /////////////////////////////////////////////////////////////////////
+template <typename T>
+std::unique_ptr<T> Factory<T>::create(const std::string type,
+                                      const std::string name) const
+{
+    Func func;
+    
+    try
+    {
+        func = builder_.at(type);
+    }
+    catch (std::out_of_range &)
+    {
+        HADRON_ERROR("object of type '" + type + "' unknown");
+    }
+    
+    return func(name);
+}
+
+END_HADRONS_NAMESPACE
+
+#endif // Hadrons_Factory_hpp_

extras/Hadrons/GeneticScheduler.hpp

@@ -0,0 +1,329 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/GeneticScheduler.hpp
+
+Copyright (C) 2015
+Copyright (C) 2016
+
+Author: Antonin Portelli <antonin.portelli@me.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_GeneticScheduler_hpp_
+#define Hadrons_GeneticScheduler_hpp_
+
+#include <Grid/Hadrons/Global.hpp>
+#include <Grid/Hadrons/Graph.hpp>
+
+BEGIN_HADRONS_NAMESPACE
+
+/******************************************************************************
+ *                   Scheduler based on a genetic algorithm                   *
+ ******************************************************************************/
+template <typename T>
+class GeneticScheduler
+{
+public:
+    typedef std::vector<T>                   Gene;
+    typedef std::pair<Gene *, Gene *>        GenePair;
+    typedef std::function<int(const Gene &)> ObjFunc;
+    struct Parameters
+    {
+        double       mutationRate;
+        unsigned int popSize, seed;
+    };
+public:
+    // constructor
+    GeneticScheduler(Graph<T> &graph, const ObjFunc &func,
+                     const Parameters &par);
+    // destructor
+    virtual ~GeneticScheduler(void) = default;
+    // access
+    const Gene & getMinSchedule(void);
+    int          getMinValue(void);
+    // breed a new generation
+    void nextGeneration(void);
+    // heuristic benchmarks
+    void benchmarkCrossover(const unsigned int nIt);
+    // print population
+    friend std::ostream & operator<<(std::ostream &out,
+                                     const GeneticScheduler<T> &s)
+    {
+        out << "[";
+        for (auto &p: s.population_)
+        {
+            out << p.first << ", ";
+        }
+        out << "\b\b]";
+        
+        return out;
+    }
+private:
+    // evolution steps
+    void initPopulation(void);
+    void doCrossover(void);
+    void doMutation(void);
+    // genetic operators
+    GenePair selectPair(void);
+    void     crossover(Gene &c1, Gene &c2, const Gene &p1, const Gene &p2);
+    void     mutation(Gene &m, const Gene &c);
+    
+private:
+    Graph<T>                 &graph_;
+    const ObjFunc            &func_;
+    const Parameters         par_;
+    std::multimap<int, Gene> population_;
+    std::mt19937             gen_;
+};
+
+/******************************************************************************
+ *                       template implementation                              *
+ ******************************************************************************/
+// constructor /////////////////////////////////////////////////////////////////
+template <typename T>
+GeneticScheduler<T>::GeneticScheduler(Graph<T> &graph, const ObjFunc &func,
+                                      const Parameters &par)
+: graph_(graph)
+, func_(func)
+, par_(par)
+{
+    gen_.seed(par_.seed);
+}
+
+// access //////////////////////////////////////////////////////////////////////
+template <typename T>
+const typename GeneticScheduler<T>::Gene &
+GeneticScheduler<T>::getMinSchedule(void)
+{
+    return population_.begin()->second;
+}
+
+template <typename T>
+int GeneticScheduler<T>::getMinValue(void)
+{
+    return population_.begin()->first;
+}
+
+// breed a new generation //////////////////////////////////////////////////////
+template <typename T>
+void GeneticScheduler<T>::nextGeneration(void)
+{
+    // random initialization of the population if necessary
+    if (population_.size() != par_.popSize)
+    {
+        initPopulation();
+    }
+    LOG(Debug) << "Starting population:\n" << *this << std::endl;
+    
+    // random mutations
+    PARALLEL_FOR_LOOP
+    for (unsigned int i = 0; i < par_.popSize; ++i)
+    {
+        doMutation();
+    }
+    LOG(Debug) << "After mutations:\n" << *this << std::endl;
+    
+    // mating
+    PARALLEL_FOR_LOOP
+    for (unsigned int i = 0; i < par_.popSize/2; ++i)
+    {
+        doCrossover();
+    }
+    LOG(Debug) << "After mating:\n" << *this << std::endl;
+    
+    // grim reaper
+    auto it = population_.begin();
+    
+    std::advance(it, par_.popSize);
+    population_.erase(it, population_.end());
+    LOG(Debug) << "After grim reaper:\n" << *this << std::endl;
+}
+
+// evolution steps /////////////////////////////////////////////////////////////
+template <typename T>
+void GeneticScheduler<T>::initPopulation(void)
+{
+    population_.clear();
+    for (unsigned int i = 0; i < par_.popSize; ++i)
+    {
+        auto p = graph_.topoSort(gen_);
+        
+        population_.emplace(func_(p), p);
+    }
+}
+
+template <typename T>
+void GeneticScheduler<T>::doCrossover(void)
+{
+    auto p = selectPair();
+    Gene &p1 = *(p.first), &p2 = *(p.second);
+    Gene c1, c2;
+    
+    crossover(c1, c2, p1, p2);
+    PARALLEL_CRITICAL
+    {
+        population_.emplace(func_(c1), c1);
+        population_.emplace(func_(c2), c2);
+    }
+}
+
+template <typename T>
+void GeneticScheduler<T>::doMutation(void)
+{
+    std::uniform_real_distribution<double>      mdis(0., 1.);
+    std::uniform_int_distribution<unsigned int> pdis(0, population_.size() - 1);
+    
+    if (mdis(gen_) < par_.mutationRate)
+    {
+        Gene m;
+        auto it = population_.begin();
+        
+        std::advance(it, pdis(gen_));
+        mutation(m, it->second);
+        PARALLEL_CRITICAL
+        {
+            population_.emplace(func_(m), m);
+        }
+    }
+}
+
+// genetic operators ///////////////////////////////////////////////////////////
+template <typename T>
+typename GeneticScheduler<T>::GenePair GeneticScheduler<T>::selectPair(void)
+{
+    std::vector<double> prob;
+    unsigned int        ind;
+    Gene                *p1, *p2;
+    
+    for (auto &c: population_)
+    {
+        prob.push_back(1./c.first);
+    }
+    do
+    {
+        double probCpy;
+        
+        std::discrete_distribution<unsigned int> dis1(prob.begin(), prob.end());
+        auto rIt = population_.begin();
+        ind = dis1(gen_);
+        std::advance(rIt, ind);
+        p1 = &(rIt->second);
+        probCpy   = prob[ind];
+        prob[ind] = 0.;
+        std::discrete_distribution<unsigned int> dis2(prob.begin(), prob.end());
+        rIt = population_.begin();
+        std::advance(rIt, dis2(gen_));
+        p2 = &(rIt->second);
+        prob[ind] = probCpy;
+    } while (p1 == p2);
+    
+    return std::make_pair(p1, p2);
+}
+
+template <typename T>
+void GeneticScheduler<T>::crossover(Gene &c1, Gene &c2, const Gene &p1,
+                                    const Gene &p2)
+{
+    Gene                                        buf;
+    std::uniform_int_distribution<unsigned int> dis(0, p1.size() - 1);
+    unsigned int                                cut = dis(gen_);
+    
+    c1.clear();
+    buf = p2;
+    for (unsigned int i = 0; i < cut; ++i)
+    {
+        c1.push_back(p1[i]);
+        buf.erase(std::find(buf.begin(), buf.end(), p1[i]));
+    }
+    for (unsigned int i = 0; i < buf.size(); ++i)
+    {
+        c1.push_back(buf[i]);
+    }
+    c2.clear();
+    buf = p2;
+    for (unsigned int i = cut; i < p1.size(); ++i)
+    {
+        buf.erase(std::find(buf.begin(), buf.end(), p1[i]));
+    }
+    for (unsigned int i = 0; i < buf.size(); ++i)
+    {
+        c2.push_back(buf[i]);
+    }
+    for (unsigned int i = cut; i < p1.size(); ++i)
+    {
+        c2.push_back(p1[i]);
+    }
+}
+
+template <typename T>
+void GeneticScheduler<T>::mutation(Gene &m, const Gene &c)
+{
+    Gene                                        buf;
+    std::uniform_int_distribution<unsigned int> dis(0, c.size() - 1);
+    unsigned int                                cut = dis(gen_);
+    Graph<T>                                    g1 = graph_, g2 = graph_;
+    
+    for (unsigned int i = 0; i < cut; ++i)
+    {
+        g1.removeVertex(c[i]);
+    }
+    for (unsigned int i = cut; i < c.size(); ++i)
+    {
+        g2.removeVertex(c[i]);
+    }
+    if (g1.size() > 0)
+    {
+        buf = g1.topoSort(gen_);
+    }
+    if (g2.size() > 0)
+    {
+        m = g2.topoSort(gen_);
+    }
+    for (unsigned int i = cut; i < c.size(); ++i)
+    {
+        m.push_back(buf[i - cut]);
+    }
+}
+
+template <typename T>
+void GeneticScheduler<T>::benchmarkCrossover(const unsigned int nIt)
+{
+    Gene   p1, p2, c1, c2;
+    double neg = 0., eq = 0., pos = 0., total;
+    int    improvement;
+    
+    LOG(Message) << "Benchmarking crossover..." << std::endl;
+    for (unsigned int i = 0; i < nIt; ++i)
+    {
+        p1 = graph_.topoSort(gen_);
+        p2 = graph_.topoSort(gen_);
+        crossover(c1, c2, p1, p2);
+        improvement = (func_(c1) + func_(c2) - func_(p1) - func_(p2))/2;
+        if (improvement < 0) neg++; else if (improvement == 0) eq++; else pos++;
+    }
+    total = neg + eq + pos;
+    LOG(Message) << "  -: " << neg/total << "  =: " << eq/total
+                 << "  +: " << pos/total << std::endl;
+}
+
+END_HADRONS_NAMESPACE
+
+#endif // Hadrons_GeneticScheduler_hpp_

extras/Hadrons/Global.cc

@@ -0,0 +1,82 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Global.cc
+
+Copyright (C) 2015
+Copyright (C) 2016
+
+Author: Antonin Portelli <antonin.portelli@me.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/Global.hpp>
+
+using namespace Grid;
+using namespace QCD;
+using namespace Hadrons;
+
+HadronsLogger Hadrons::HadronsLogError(1,"Error");
+HadronsLogger Hadrons::HadronsLogWarning(1,"Warning");
+HadronsLogger Hadrons::HadronsLogMessage(1,"Message");
+HadronsLogger Hadrons::HadronsLogIterative(1,"Iterative");
+HadronsLogger Hadrons::HadronsLogDebug(1,"Debug");
+
+// pretty size formatting //////////////////////////////////////////////////////
+std::string Hadrons::sizeString(long unsigned int bytes)
+
+{
+    constexpr unsigned int bufSize = 256;
+    const char             *suffixes[7] = {"", "K", "M", "G", "T", "P", "E"};
+    char                   buf[256];
+    long unsigned int      s     = 0;
+    double                 count = bytes;
+    
+    while (count >= 1024 && s < 7)
+    {
+        s++;
+        count /= 1024;
+    }
+    if (count - floor(count) == 0.0)
+    {
+        snprintf(buf, bufSize, "%d %sB", (int)count, suffixes[s]);
+    }
+    else
+    {
+        snprintf(buf, bufSize, "%.1f %sB", count, suffixes[s]);
+    }
+    
+    return std::string(buf);
+}
+
+// type utilities //////////////////////////////////////////////////////////////
+constexpr unsigned int maxNameSize = 1024u;
+
+std::string Hadrons::typeName(const std::type_info *info)
+{
+    char        *buf;
+    std::string name;
+    
+    buf  = abi::__cxa_demangle(info->name(), nullptr, nullptr, nullptr);
+    name = buf;
+    free(buf);
+    
+    return name;
+}

extras/Hadrons/Global.hpp

@@ -0,0 +1,150 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Global.hpp
+
+Copyright (C) 2015
+Copyright (C) 2016
+
+Author: Antonin Portelli <antonin.portelli@me.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_Global_hpp_
+#define Hadrons_Global_hpp_
+
+#include <set>
+#include <stack>
+#include <Grid/Grid.h>
+#include <cxxabi.h>
+
+#define BEGIN_HADRONS_NAMESPACE \
+namespace Grid {\
+using namespace QCD;\
+namespace Hadrons {\
+using Grid::operator<<;
+#define END_HADRONS_NAMESPACE }}
+
+#define BEGIN_MODULE_NAMESPACE(name)\
+namespace name {\
+using Grid::operator<<;
+#define END_MODULE_NAMESPACE }
+
+/* the 'using Grid::operator<<;' statement prevents a very nasty compilation
+ * error with GCC 5 (clang & GCC 6 compile fine without it).
+ */
+
+// FIXME: find a way to do that in a more general fashion
+#ifndef FIMPL
+#define FIMPL WilsonImplR
+#endif
+
+BEGIN_HADRONS_NAMESPACE
+
+// type aliases
+#define TYPE_ALIASES(FImpl, suffix)\
+typedef FermionOperator<FImpl>                       FMat##suffix;             \
+typedef typename FImpl::FermionField                 FermionField##suffix;     \
+typedef typename FImpl::PropagatorField              PropagatorField##suffix;  \
+typedef typename FImpl::SitePropagator               SitePropagator##suffix;   \
+typedef typename FImpl::DoubledGaugeField            DoubledGaugeField##suffix;\
+typedef std::function<void(FermionField##suffix &,                             \
+                      const FermionField##suffix &)> SolverFn##suffix;
+
+// logger
+class HadronsLogger: public Logger
+{
+public:
+    HadronsLogger(int on, std::string nm): Logger("Hadrons", on, nm,
+                                                  GridLogColours, "BLACK"){};
+};
+
+#define LOG(channel) std::cout << HadronsLog##channel
+#define HADRON_ERROR(msg)\
+LOG(Error) << msg << " (" << __FUNCTION__ << " at " << __FILE__ << ":"\
+           << __LINE__ << ")" << std::endl;\
+abort();
+
+#define DEBUG_VAR(var) LOG(Debug) << #var << "= " << (var) << std::endl;
+
+extern HadronsLogger HadronsLogError;
+extern HadronsLogger HadronsLogWarning;
+extern HadronsLogger HadronsLogMessage;
+extern HadronsLogger HadronsLogIterative;
+extern HadronsLogger HadronsLogDebug;
+
+// singleton pattern
+#define SINGLETON(name)\
+public:\
+    name(const name &e) = delete;\
+    void operator=(const name &e) = delete;\
+    static name & getInstance(void)\
+    {\
+        static name e;\
+        return e;\
+    }\
+private:\
+    name(void);
+
+#define SINGLETON_DEFCTOR(name)\
+public:\
+    name(const name &e) = delete;\
+    void operator=(const name &e) = delete;\
+    static name & getInstance(void)\
+    {\
+        static name e;\
+        return e;\
+    }\
+private:\
+    name(void) = default;
+
+// pretty size formating
+std::string sizeString(long unsigned int bytes);
+
+// type utilities
+template <typename T>
+const std::type_info * typeIdPt(const T &x)
+{
+    return &typeid(x);
+}
+
+std::string typeName(const std::type_info *info);
+
+template <typename T>
+const std::type_info * typeIdPt(void)
+{
+    return &typeid(T);
+}
+
+template <typename T>
+std::string typeName(const T &x)
+{
+    return typeName(typeIdPt(x));
+}
+
+template <typename T>
+std::string typeName(void)
+{
+    return typeName(typeIdPt<T>());
+}
+
+END_HADRONS_NAMESPACE
+
+#endif // Hadrons_Global_hpp_

extras/Hadrons/Graph.hpp

@@ -0,0 +1,760 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Graph.hpp
+
+Copyright (C) 2015
+Copyright (C) 2016
+
+Author: Antonin Portelli <antonin.portelli@me.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_Graph_hpp_
+#define Hadrons_Graph_hpp_
+
+#include <Grid/Hadrons/Global.hpp>
+
+BEGIN_HADRONS_NAMESPACE
+
+/******************************************************************************
+ *                          Oriented graph class                              *
+ ******************************************************************************/
+// I/O for edges
+template <typename T>
+std::ostream & operator<<(std::ostream &out, const std::pair<T, T> &e)
+{
+    out << "\""  << e.first << "\" -> \"" << e.second << "\"";
+    
+    return out;
+}
+
+// main class
+template <typename T>
+class Graph
+{
+public:
+    typedef std::pair<T, T> Edge;
+public:
+    // constructor
+    Graph(void);
+    // destructor
+    virtual ~Graph(void) = default;
+    // access
+    void           addVertex(const T &value);
+    void           addEdge(const Edge &e);
+    void           addEdge(const T &start, const T &end);
+    std::vector<T> getVertices(void) const;
+    void           removeVertex(const T &value);
+    void           removeEdge(const Edge &e);
+    void           removeEdge(const T &start, const T &end);
+    unsigned int   size(void) const;
+    // tests
+    bool gotValue(const T &value) const;
+    // graph topological manipulations
+    std::vector<T>              getAdjacentVertices(const T &value) const;
+    std::vector<T>              getChildren(const T &value) const;
+    std::vector<T>              getParents(const T &value) const;
+    std::vector<T>              getRoots(void) const;
+    std::vector<Graph<T>>       getConnectedComponents(void) const;
+    std::vector<T>              topoSort(void);
+    template <typename Gen>
+    std::vector<T>              topoSort(Gen &gen);
+    std::vector<std::vector<T>> allTopoSort(void);
+    // I/O
+    friend std::ostream & operator<<(std::ostream &out, const Graph<T> &g)
+    {
+        out << "{";
+        for (auto &e: g.edgeSet_)
+        {
+            out << e << ", ";
+        }
+        if (g.edgeSet_.size() != 0)
+        {
+            out << "\b\b";
+        }
+        out << "}";
+        
+        return out;
+    }
+private:
+    // vertex marking
+    void      mark(const T &value, const bool doMark = true);
+    void      markAll(const bool doMark = true);
+    void      unmark(const T &value);
+    void      unmarkAll(void);
+    bool      isMarked(const T &value) const;
+    const T * getFirstMarked(const bool isMarked = true) const;
+    template <typename Gen>
+    const T * getRandomMarked(const bool isMarked, Gen &gen);
+    const T * getFirstUnmarked(void) const;
+    template <typename Gen>
+    const T * getRandomUnmarked(Gen &gen);
+    // prune marked/unmarked vertices
+    void removeMarked(const bool isMarked = true);
+    void removeUnmarked(void);
+    // depth-first search marking
+    void depthFirstSearch(void);
+    void depthFirstSearch(const T &root);
+private:
+    std::map<T, bool>  isMarked_;
+    std::set<Edge>     edgeSet_;
+};
+
+// build depedency matrix from topological sorts
+template <typename T>
+std::map<T, std::map<T, bool>>
+makeDependencyMatrix(const std::vector<std::vector<T>> &topSort);
+
+/******************************************************************************
+ *                       template implementation                              *
+ ******************************************************************************
+ * in all the following V is the number of vertex and E is the number of edge
+ * in the worst case E = V^2
+ */
+
+// constructor /////////////////////////////////////////////////////////////////
+template <typename T>
+Graph<T>::Graph(void)
+{}
+
+// access //////////////////////////////////////////////////////////////////////
+// complexity: log(V)
+template <typename T>
+void Graph<T>::addVertex(const T &value)
+{
+    isMarked_[value] = false;
+}
+
+// complexity: O(log(V))
+template <typename T>
+void Graph<T>::addEdge(const Edge &e)
+{
+    addVertex(e.first);
+    addVertex(e.second);
+    edgeSet_.insert(e);
+}
+
+// complexity: O(log(V))
+template <typename T>
+void Graph<T>::addEdge(const T &start, const T &end)
+{
+    addEdge(Edge(start, end));
+}
+
+template <typename T>
+std::vector<T> Graph<T>::getVertices(void) const
+{
+    std::vector<T> vertex;
+    
+    for (auto &v: isMarked_)
+    {
+        vertex.push_back(v.first);
+    }
+    
+    return vertex;
+}
+
+// complexity: O(V*log(V))
+template <typename T>
+void Graph<T>::removeVertex(const T &value)
+{
+    // remove vertex from the mark table
+    auto vIt = isMarked_.find(value);
+    
+    if (vIt != isMarked_.end())
+    {
+        isMarked_.erase(vIt);
+    }
+    else
+    {
+        HADRON_ERROR("vertex " << value << " does not exists");
+    }
+
+    // remove all edges containing the vertex
+    auto pred = [&value](const Edge &e)
+    {
+        return ((e.first == value) or (e.second == value));
+    };
+    auto eIt = find_if(edgeSet_.begin(), edgeSet_.end(), pred);
+    
+    while (eIt != edgeSet_.end())
+    {
+        edgeSet_.erase(eIt);
+        eIt = find_if(edgeSet_.begin(), edgeSet_.end(), pred);
+    }
+}
+
+// complexity: O(log(V))
+template <typename T>
+void Graph<T>::removeEdge(const Edge &e)
+{
+    auto eIt = edgeSet_.find(e);
+    
+    if (eIt != edgeSet_.end())
+    {
+        edgeSet_.erase(eIt);
+    }
+    else
+    {
+        HADRON_ERROR("edge "  << e << " does not exists");
+    }
+}
+
+// complexity: O(log(V))
+template <typename T>
+void Graph<T>::removeEdge(const T &start, const T &end)
+{
+    removeEdge(Edge(start, end));
+}
+
+// complexity: O(1)
+template <typename T>
+unsigned int Graph<T>::size(void) const
+{
+    return isMarked_.size();
+}
+
+// tests ///////////////////////////////////////////////////////////////////////
+// complexity: O(log(V))
+template <typename T>
+bool Graph<T>::gotValue(const T &value) const
+{
+    auto it = isMarked_.find(value);
+    
+    if (it == isMarked_.end())
+    {
+        return false;
+    }
+    else
+    {
+        return true;
+    }
+}
+
+// vertex marking //////////////////////////////////////////////////////////////
+// complexity: O(log(V))
+template <typename T>
+void Graph<T>::mark(const T &value, const bool doMark)
+{
+    if (gotValue(value))
+    {
+        isMarked_[value] = doMark;
+    }
+    else
+    {
+        HADRON_ERROR("vertex " << value << " does not exists");
+    }
+}
+
+// complexity: O(V*log(V))
+template <typename T>
+void Graph<T>::markAll(const bool doMark)
+{
+    for (auto &v: isMarked_)
+    {
+        mark(v.first, doMark);
+    }
+}
+
+// complexity: O(log(V))
+template <typename T>
+void Graph<T>::unmark(const T &value)
+{
+    mark(value, false);
+}
+
+// complexity: O(V*log(V))
+template <typename T>
+void Graph<T>::unmarkAll(void)
+{
+    markAll(false);
+}
+
+// complexity: O(log(V))
+template <typename T>
+bool Graph<T>::isMarked(const T &value) const
+{
+    if (gotValue(value))
+    {
+        return isMarked_.at(value);
+    }
+    else
+    {
+        HADRON_ERROR("vertex " << value << " does not exists");
+        
+        return false;
+    }
+}
+
+// complexity: O(log(V))
+template <typename T>
+const T * Graph<T>::getFirstMarked(const bool isMarked) const
+{
+    auto pred = [&isMarked](const std::pair<T, bool> &v)
+    {
+        return (v.second == isMarked);
+    };
+    auto vIt = std::find_if(isMarked_.begin(), isMarked_.end(), pred);
+    
+    if (vIt != isMarked_.end())
+    {
+        return &(vIt->first);
+    }
+    else
+    {
+        return nullptr;
+    }
+}
+
+// complexity: O(log(V))
+template <typename T>
+template <typename Gen>
+const T * Graph<T>::getRandomMarked(const bool isMarked, Gen &gen)
+{
+    auto pred = [&isMarked](const std::pair<T, bool> &v)
+    {
+        return (v.second == isMarked);
+    };
+    std::uniform_int_distribution<unsigned int> dis(0, size() - 1);
+    auto                                        rIt = isMarked_.begin();
+    
+    std::advance(rIt, dis(gen));
+    auto vIt = std::find_if(rIt, isMarked_.end(), pred);
+    if (vIt != isMarked_.end())
+    {
+        return &(vIt->first);
+    }
+    else
+    {
+        vIt = std::find_if(isMarked_.begin(), rIt, pred);
+        if (vIt != rIt)
+        {
+            return &(vIt->first);
+        }
+        else
+        {
+            return nullptr;
+        }
+    }
+}
+
+// complexity: O(log(V))
+template <typename T>
+const T * Graph<T>::getFirstUnmarked(void) const
+{
+    return getFirstMarked(false);
+}
+
+// complexity: O(log(V))
+template <typename T>
+template <typename Gen>
+const T * Graph<T>::getRandomUnmarked(Gen &gen)
+{
+    return getRandomMarked(false, gen);
+}
+
+// prune marked/unmarked vertices //////////////////////////////////////////////
+// complexity: O(V^2*log(V))
+template <typename T>
+void Graph<T>::removeMarked(const bool isMarked)
+{
+    auto isMarkedCopy = isMarked_;
+    
+    for (auto &v: isMarkedCopy)
+    {
+        if (v.second == isMarked)
+        {
+            removeVertex(v.first);
+        }
+    }
+}
+
+// complexity: O(V^2*log(V))
+template <typename T>
+void Graph<T>::removeUnmarked(void)
+{
+    removeMarked(false);
+}
+
+// depth-first search marking //////////////////////////////////////////////////
+// complexity: O(V*log(V))
+template <typename T>
+void Graph<T>::depthFirstSearch(void)
+{
+    depthFirstSearch(isMarked_.begin()->first);
+}
+
+// complexity: O(V*log(V))
+template <typename T>
+void Graph<T>::depthFirstSearch(const T &root)
+{
+    std::vector<T> adjacentVertex;
+    
+    mark(root);
+    adjacentVertex = getAdjacentVertices(root);
+    for (auto &v: adjacentVertex)
+    {
+        if (!isMarked(v))
+        {
+            depthFirstSearch(v);
+        }
+    }
+}
+
+// graph topological manipulations /////////////////////////////////////////////
+// complexity: O(V*log(V))
+template <typename T>
+std::vector<T> Graph<T>::getAdjacentVertices(const T &value) const
+{
+    std::vector<T> adjacentVertex;
+    
+    auto pred = [&value](const Edge &e)
+    {
+        return ((e.first == value) or (e.second == value));
+    };
+    auto eIt = find_if(edgeSet_.begin(), edgeSet_.end(), pred);
+    
+    while (eIt != edgeSet_.end())
+    {
+        if (eIt->first == value)
+        {
+            adjacentVertex.push_back((*eIt).second);
+        }
+        else if (eIt->second == value)
+        {
+            adjacentVertex.push_back((*eIt).first);
+        }
+        eIt = find_if(++eIt, edgeSet_.end(), pred);
+    }
+    
+    return adjacentVertex;
+}
+
+// complexity: O(V*log(V))
+template <typename T>
+std::vector<T> Graph<T>::getChildren(const T &value) const
+{
+    std::vector<T> child;
+    
+    auto pred = [&value](const Edge &e)
+    {
+        return (e.first == value);
+    };
+    auto eIt = find_if(edgeSet_.begin(), edgeSet_.end(), pred);
+    
+    while (eIt != edgeSet_.end())
+    {
+        child.push_back((*eIt).second);
+        eIt = find_if(++eIt, edgeSet_.end(), pred);
+    }
+    
+    return child;
+}
+
+// complexity: O(V*log(V))
+template <typename T>
+std::vector<T> Graph<T>::getParents(const T &value) const
+{
+    std::vector<T> parent;
+    
+    auto pred = [&value](const Edge &e)
+    {
+        return (e.second == value);
+    };
+    auto eIt = find_if(edgeSet_.begin(), edgeSet_.end(), pred);
+    
+    while (eIt != edgeSet_.end())
+    {
+        parent.push_back((*eIt).first);
+        eIt = find_if(++eIt, edgeSet_.end(), pred);
+    }
+    
+    return parent;
+}
+
+// complexity: O(V^2*log(V))
+template <typename T>
+std::vector<T> Graph<T>::getRoots(void) const
+{
+    std::vector<T> root;
+    
+    for (auto &v: isMarked_)
+    {
+        auto parent = getParents(v.first);
+        
+        if (parent.size() == 0)
+        {
+            root.push_back(v.first);
+        }
+    }
+    
+    return root;
+}
+
+// complexity: O(V^2*log(V))
+template <typename T>
+std::vector<Graph<T>> Graph<T>::getConnectedComponents(void) const
+{
+    std::vector<Graph<T>> res;
+    Graph<T>              copy(*this);
+    
+    while (copy.size() > 0)
+    {
+        copy.depthFirstSearch();
+        res.push_back(copy);
+        res.back().removeUnmarked();
+        res.back().unmarkAll();
+        copy.removeMarked();
+        copy.unmarkAll();
+    }
+    
+    return res;
+}
+
+// topological sort using a directed DFS algorithm
+// complexity: O(V*log(V))
+template <typename T>
+std::vector<T> Graph<T>::topoSort(void)
+{
+    std::stack<T>     buf;
+    std::vector<T>    res;
+    const T           *vPt;
+    std::map<T, bool> tmpMarked(isMarked_);
+
+    // visit function
+    std::function<void(const T &)> visit = [&](const T &v)
+    {
+        if (tmpMarked.at(v))
+        {
+            HADRON_ERROR("cannot topologically sort a cyclic graph");
+        }
+        if (!isMarked(v))
+        {
+            std::vector<T> child = getChildren(v);
+
+            tmpMarked[v] = true;
+            for (auto &c: child)
+            {
+                visit(c);
+            }
+            mark(v);
+            tmpMarked[v] = false;
+            buf.push(v);
+        }
+    };
+    
+    // reset temporary marks
+    for (auto &v: tmpMarked)
+    {
+        tmpMarked.at(v.first) = false;
+    }
+    
+    // loop on unmarked vertices
+    unmarkAll();
+    vPt = getFirstUnmarked();
+    while (vPt)
+    {
+        visit(*vPt);
+        vPt = getFirstUnmarked();
+    }
+    unmarkAll();
+    
+    // create result vector
+    while (!buf.empty())
+    {
+        res.push_back(buf.top());
+        buf.pop();
+    }
+    
+    return res;
+}
+
+// random version of the topological sort
+// complexity: O(V*log(V))
+template <typename T>
+template <typename Gen>
+std::vector<T> Graph<T>::topoSort(Gen &gen)
+{
+    std::stack<T>     buf;
+    std::vector<T>    res;
+    const T           *vPt;
+    std::map<T, bool> tmpMarked(isMarked_);
+    
+    // visit function
+    std::function<void(const T &)> visit = [&](const T &v)
+    {
+        if (tmpMarked.at(v))
+        {
+            HADRON_ERROR("cannot topologically sort a cyclic graph");
+        }
+        if (!isMarked(v))
+        {
+            std::vector<T> child = getChildren(v);
+            
+            tmpMarked[v] = true;
+            std::shuffle(child.begin(), child.end(), gen);
+            for (auto &c: child)
+            {
+                visit(c);
+            }
+            mark(v);
+            tmpMarked[v] = false;
+            buf.push(v);
+        }
+    };
+    
+    // reset temporary marks
+    for (auto &v: tmpMarked)
+    {
+        tmpMarked.at(v.first) = false;
+    }
+    
+    // loop on unmarked vertices
+    unmarkAll();
+    vPt = getRandomUnmarked(gen);
+    while (vPt)
+    {
+        visit(*vPt);
+        vPt = getRandomUnmarked(gen);
+    }
+    unmarkAll();
+    
+    // create result vector
+    while (!buf.empty())
+    {
+        res.push_back(buf.top());
+        buf.pop();
+    }
+    
+    return res;
+}
+
+// generate all possible topological sorts
+// Y. L. Varol & D. Rotem, Comput. J. 24(1), pp. 83–84, 1981
+// http://comjnl.oupjournals.org/cgi/doi/10.1093/comjnl/24.1.83
+// complexity: O(V*log(V)) (from the paper, but really ?)
+template <typename T>
+std::vector<std::vector<T>> Graph<T>::allTopoSort(void)
+{
+    std::vector<std::vector<T>>    res;
+    std::map<T, std::map<T, bool>> iMat;
+    
+    // create incidence matrix
+    for (auto &v1: isMarked_)
+    for (auto &v2: isMarked_)
+    {
+        iMat[v1.first][v2.first] = false;
+    }
+    for (auto &v: isMarked_)
+    {
+        auto cVec = getChildren(v.first);
+        
+        for (auto &c: cVec)
+        {
+            iMat[v.first][c] = true;
+        }
+    }
+    
+    // generate initial topological sort
+    res.push_back(topoSort());
+    
+    // generate all other topological sorts by permutation
+    std::vector<T>            p = res[0];
+    const unsigned int        n = size();
+    std::vector<unsigned int> loc(n);
+    unsigned int              i, k, k1;
+    T                         obj_k, obj_k1;
+    bool                      isFinal;
+    
+    for (unsigned int j = 0; j < n; ++j)
+    {
+        loc[j] = j;
+    }
+    i = 0;
+    while (i < n-1)
+    {
+        k      = loc[i];
+        k1     = k + 1;
+        obj_k  = p[k];
+        if (k1 >= n)
+        {
+            isFinal = true;
+            obj_k1  = obj_k;
+        }
+        else
+        {
+            isFinal = false;
+            obj_k1  = p[k1];
+        }
+        if (iMat[res[0][i]][obj_k1] or isFinal)
+        {
+            for (unsigned int l = k; l >= i + 1; --l)
+            {
+                p[l]   = p[l-1];
+            }
+            p[i]   = obj_k;
+            loc[i] = i;
+            i++;
+        }
+        else
+        {
+            p[k]   = obj_k1;
+            p[k1]  = obj_k;
+            loc[i] = k1;
+            i      = 0;
+            res.push_back(p);
+        }
+    }
+    
+    return res;
+}
+
+// build depedency matrix from topological sorts ///////////////////////////////
+// complexity: something like O(V^2*log(V!))
+template <typename T>
+std::map<T, std::map<T, bool>>
+makeDependencyMatrix(const std::vector<std::vector<T>> &topSort)
+{
+    std::map<T, std::map<T, bool>> m;
+    const std::vector<T>           &vList = topSort[0];
+    
+    for (auto &v1: vList)
+    for (auto &v2: vList)
+    {
+        bool dep = true;
+        
+        for (auto &t: topSort)
+        {
+            auto i1 = std::find(t.begin(), t.end(), v1);
+            auto i2 = std::find(t.begin(), t.end(), v2);
+            
+            dep = dep and (i1 - i2 > 0);
+            if (!dep) break;
+        }
+        m[v1][v2] = dep;
+    }
+    
+    return m;
+}
+
+END_HADRONS_NAMESPACE
+
+#endif // Hadrons_Graph_hpp_

extras/Hadrons/HadronsXmlRun.cc

@@ -0,0 +1,80 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/HadronsXmlRun.cc
+
+Copyright (C) 2015
+Copyright (C) 2016
+
+Author: Antonin Portelli <antonin.portelli@me.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/Application.hpp>
+
+using namespace Grid;
+using namespace QCD;
+using namespace Hadrons;
+
+int main(int argc, char *argv[])
+{
+    // parse command line
+    std::string parameterFileName, scheduleFileName = "";
+    
+    if (argc < 2)
+    {
+        std::cerr << "usage: " << argv[0] << " <parameter file> [<precomputed schedule>] [Grid options]";
+        std::cerr << std::endl;
+        std::exit(EXIT_FAILURE);
+    }
+    parameterFileName = argv[1];
+    if (argc > 2)
+    {
+        if (argv[2][0] != '-')
+        {
+            scheduleFileName = argv[2];
+        }
+    }
+    
+    // initialization
+    Grid_init(&argc, &argv);
+    HadronsLogError.Active(GridLogError.isActive());
+    HadronsLogWarning.Active(GridLogWarning.isActive());
+    HadronsLogMessage.Active(GridLogMessage.isActive());
+    HadronsLogIterative.Active(GridLogIterative.isActive());
+    HadronsLogDebug.Active(GridLogDebug.isActive());
+    LOG(Message) << "Grid initialized" << std::endl;
+    
+    // execution
+    Application application(parameterFileName);
+    
+    application.parseParameterFile(parameterFileName);
+    if (!scheduleFileName.empty())
+    {
+        application.loadSchedule(scheduleFileName);
+    }
+    application.run();
+    
+    // epilogue
+    LOG(Message) << "Grid is finalizing now" << std::endl;
+    Grid_finalize();
+    
+    return EXIT_SUCCESS;
+}

extras/Hadrons/HadronsXmlSchedule.cc

@@ -0,0 +1,72 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/HadronsXmlSchedule.cc
+
+Copyright (C) 2015
+Copyright (C) 2016
+
+Author: Antonin Portelli <antonin.portelli@me.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/Application.hpp>
+
+using namespace Grid;
+using namespace QCD;
+using namespace Hadrons;
+
+int main(int argc, char *argv[])
+{
+    // parse command line
+    std::string parameterFileName, scheduleFileName;
+    
+    if (argc < 3)
+    {
+        std::cerr << "usage: " << argv[0] << " <parameter file> <schedule output> [Grid options]";
+        std::cerr << std::endl;
+        std::exit(EXIT_FAILURE);
+    }
+    parameterFileName = argv[1];
+    scheduleFileName  = argv[2];
+    
+    // initialization
+    Grid_init(&argc, &argv);
+    HadronsLogError.Active(GridLogError.isActive());
+    HadronsLogWarning.Active(GridLogWarning.isActive());
+    HadronsLogMessage.Active(GridLogMessage.isActive());
+    HadronsLogIterative.Active(GridLogIterative.isActive());
+    HadronsLogDebug.Active(GridLogDebug.isActive());
+    LOG(Message) << "Grid initialized" << std::endl;
+    
+    // execution
+    Application application;
+    
+    application.parseParameterFile(parameterFileName);
+    application.schedule();
+    application.printSchedule();
+    application.saveSchedule(scheduleFileName);
+    
+    // epilogue
+    LOG(Message) << "Grid is finalizing now" << std::endl;
+    Grid_finalize();
+    
+    return EXIT_SUCCESS;
+}

extras/Hadrons/Makefile.am

@@ -0,0 +1,29 @@
+lib_LIBRARIES = libHadrons.a
+bin_PROGRAMS  = HadronsXmlRun HadronsXmlSchedule
+
+include modules.inc
+
+libHadrons_a_SOURCES = \
+    $(modules_cc)      \
+    Application.cc     \
+    Environment.cc     \
+    Global.cc          \
+    Module.cc
+libHadrons_adir = $(pkgincludedir)/Hadrons
+nobase_libHadrons_a_HEADERS = \
+	$(modules_hpp)            \
+	Application.hpp           \
+	Environment.hpp           \
+	Factory.hpp               \
+	GeneticScheduler.hpp      \
+	Global.hpp                \
+	Graph.hpp                 \
+	Module.hpp                \
+	Modules.hpp               \
+	ModuleFactory.hpp
+
+HadronsXmlRun_SOURCES = HadronsXmlRun.cc
+HadronsXmlRun_LDADD   = libHadrons.a -lGrid
+
+HadronsXmlSchedule_SOURCES = HadronsXmlSchedule.cc
+HadronsXmlSchedule_LDADD   = libHadrons.a -lGrid

extras/Hadrons/Module.cc

@@ -0,0 +1,71 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Module.cc
+
+Copyright (C) 2015
+Copyright (C) 2016
+
+Author: Antonin Portelli <antonin.portelli@me.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/Module.hpp>
+
+using namespace Grid;
+using namespace QCD;
+using namespace Hadrons;
+
+/******************************************************************************
+ *                       ModuleBase implementation                            *
+ ******************************************************************************/
+// constructor /////////////////////////////////////////////////////////////////
+ModuleBase::ModuleBase(const std::string name)
+: name_(name)
+, env_(Environment::getInstance())
+{}
+
+// access //////////////////////////////////////////////////////////////////////
+std::string ModuleBase::getName(void) const
+{
+    return name_;
+}
+
+Environment & ModuleBase::env(void) const
+{
+    return env_;
+}
+
+// get factory registration name if available
+std::string ModuleBase::getRegisteredName(void)
+{
+    HADRON_ERROR("module '" + getName() + "' has a type not registered"
+                 + " in the factory");
+}
+
+// execution ///////////////////////////////////////////////////////////////////
+void ModuleBase::operator()(void)
+{
+    setup();
+    if (!env().isDryRun())
+    {
+        execute();
+    }
+}

extras/Hadrons/Module.hpp

@@ -0,0 +1,198 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Module.hpp
+
+Copyright (C) 2015
+Copyright (C) 2016
+
+Author: Antonin Portelli <antonin.portelli@me.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_Module_hpp_
+#define Hadrons_Module_hpp_
+
+#include <Grid/Hadrons/Global.hpp>
+#include <Grid/Hadrons/Environment.hpp>
+
+BEGIN_HADRONS_NAMESPACE
+
+// module registration macros
+#define MODULE_REGISTER(mod, base)\
+class mod: public base\
+{\
+public:\
+    typedef base Base;\
+    using Base::Base;\
+    virtual std::string getRegisteredName(void)\
+    {\
+        return std::string(#mod);\
+    }\
+};\
+class mod##ModuleRegistrar\
+{\
+public:\
+    mod##ModuleRegistrar(void)\
+    {\
+        ModuleFactory &modFac = ModuleFactory::getInstance();\
+        modFac.registerBuilder(#mod, [&](const std::string name)\
+                              {\
+                                  return std::unique_ptr<mod>(new mod(name));\
+                              });\
+    }\
+};\
+static mod##ModuleRegistrar mod##ModuleRegistrarInstance;
+
+#define MODULE_REGISTER_NS(mod, base, ns)\
+class mod: public base\
+{\
+public:\
+    typedef base Base;\
+    using Base::Base;\
+    virtual std::string getRegisteredName(void)\
+    {\
+        return std::string(#ns "::" #mod);\
+    }\
+};\
+class ns##mod##ModuleRegistrar\
+{\
+public:\
+    ns##mod##ModuleRegistrar(void)\
+    {\
+        ModuleFactory &modFac = ModuleFactory::getInstance();\
+        modFac.registerBuilder(#ns "::" #mod, [&](const std::string name)\
+                              {\
+                                  return std::unique_ptr<ns::mod>(new ns::mod(name));\
+                              });\
+    }\
+};\
+static ns##mod##ModuleRegistrar ns##mod##ModuleRegistrarInstance;
+
+#define ARG(...) __VA_ARGS__
+
+/******************************************************************************
+ *                            Module class                                    *
+ ******************************************************************************/
+// base class
+class ModuleBase
+{
+public:
+    // constructor
+    ModuleBase(const std::string name);
+    // destructor
+    virtual ~ModuleBase(void) = default;
+    // access
+    std::string getName(void) const;
+    Environment &env(void) const;
+    // get factory registration name if available
+    virtual std::string getRegisteredName(void);
+    // dependencies/products
+    virtual std::vector<std::string> getInput(void) = 0;
+    virtual std::vector<std::string> getOutput(void) = 0;
+    // parse parameters
+    virtual void parseParameters(XmlReader &reader, const std::string name) = 0;
+    virtual void saveParameters(XmlWriter &writer, const std::string name) = 0;
+    // setup
+    virtual void setup(void) {};
+    // execution
+    void operator()(void);
+    virtual void execute(void) = 0;
+private:
+    std::string name_;
+    Environment &env_;
+};
+
+// derived class, templating the parameter class
+template <typename P>
+class Module: public ModuleBase
+{
+public:
+    typedef P Par;
+public:
+    // constructor
+    Module(const std::string name);
+    // destructor
+    virtual ~Module(void) = default;
+    // parse parameters
+    virtual void parseParameters(XmlReader &reader, const std::string name);
+    virtual void saveParameters(XmlWriter &writer, const std::string name);
+    // parameter access
+    const P & par(void) const;
+    void      setPar(const P &par);
+private:
+    P par_;
+};
+
+// no parameter type
+class NoPar {};
+
+template <>
+class Module<NoPar>: public ModuleBase
+{
+public:
+    // constructor
+    Module(const std::string name): ModuleBase(name) {};
+    // destructor
+    virtual ~Module(void) = default;
+    // parse parameters (do nothing)
+    virtual void parseParameters(XmlReader &reader, const std::string name) {};
+    virtual void saveParameters(XmlWriter &writer, const std::string name)
+    {
+        push(writer, "options");
+        pop(writer);
+    };
+};
+
+/******************************************************************************
+ *                           Template implementation                          *
+ ******************************************************************************/
+template <typename P>
+Module<P>::Module(const std::string name)
+: ModuleBase(name)
+{}
+
+template <typename P>
+void Module<P>::parseParameters(XmlReader &reader, const std::string name)
+{
+    read(reader, name, par_);
+}
+
+template <typename P>
+void Module<P>::saveParameters(XmlWriter &writer, const std::string name)
+{
+    write(writer, name, par_);
+}
+
+template <typename P>
+const P & Module<P>::par(void) const
+{
+    return par_;
+}
+
+template <typename P>
+void Module<P>::setPar(const P &par)
+{
+    par_ = par;
+}
+
+END_HADRONS_NAMESPACE
+
+#endif // Hadrons_Module_hpp_

extras/Hadrons/ModuleFactory.hpp

@@ -0,0 +1,49 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/ModuleFactory.hpp
+
+Copyright (C) 2015
+Copyright (C) 2016
+
+Author: Antonin Portelli <antonin.portelli@me.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_ModuleFactory_hpp_
+#define Hadrons_ModuleFactory_hpp_
+
+#include <Grid/Hadrons/Global.hpp>
+#include <Grid/Hadrons/Factory.hpp>
+#include <Grid/Hadrons/Module.hpp>
+
+BEGIN_HADRONS_NAMESPACE
+
+/******************************************************************************
+ *                            ModuleFactory                                   *
+ ******************************************************************************/
+class ModuleFactory: public Factory<ModuleBase>
+{
+    SINGLETON_DEFCTOR(ModuleFactory)
+};
+
+END_HADRONS_NAMESPACE
+
+#endif // Hadrons_ModuleFactory_hpp_

extras/Hadrons/Modules.hpp

@@ -0,0 +1,40 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Modules.hpp
+
+Copyright (C) 2015
+Copyright (C) 2016
+
+Author: Antonin Portelli <antonin.portelli@me.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/MAction/DWF.hpp>
+#include <Grid/Hadrons/Modules/MAction/Wilson.hpp>
+#include <Grid/Hadrons/Modules/MContraction/Baryon.hpp>
+#include <Grid/Hadrons/Modules/MContraction/Meson.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/MSolver/RBPrecCG.hpp>
+#include <Grid/Hadrons/Modules/MSource/Point.hpp>
+#include <Grid/Hadrons/Modules/MSource/SeqGamma.hpp>
+#include <Grid/Hadrons/Modules/MSource/Z2.hpp>
+#include <Grid/Hadrons/Modules/Quark.hpp>

extras/Hadrons/Modules/MAction/DWF.hpp

@@ -0,0 +1,134 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Modules/MAction/DWF.hpp
+
+Copyright (C) 2015
+Copyright (C) 2016
+
+Author: Antonin Portelli <antonin.portelli@me.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_DWF_hpp_
+#define Hadrons_DWF_hpp_
+
+#include <Grid/Hadrons/Global.hpp>
+#include <Grid/Hadrons/Module.hpp>
+#include <Grid/Hadrons/ModuleFactory.hpp>
+
+BEGIN_HADRONS_NAMESPACE
+
+/******************************************************************************
+ *                     Domain wall quark action                               *
+ ******************************************************************************/
+BEGIN_MODULE_NAMESPACE(MAction)
+
+class DWFPar: Serializable
+{
+public:
+    GRID_SERIALIZABLE_CLASS_MEMBERS(DWFPar,
+                                    std::string, gauge,
+                                    unsigned int, Ls,
+                                    double      , mass,
+                                    double      , M5);
+};
+
+template <typename FImpl>
+class TDWF: public Module<DWFPar>
+{
+public:
+    TYPE_ALIASES(FImpl,);
+public:
+    // constructor
+    TDWF(const std::string name);
+    // destructor
+    virtual ~TDWF(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(DWF, TDWF<FIMPL>, MAction);
+
+/******************************************************************************
+ *                        DWF template implementation                         *
+ ******************************************************************************/
+// constructor /////////////////////////////////////////////////////////////////
+template <typename FImpl>
+TDWF<FImpl>::TDWF(const std::string name)
+: Module<DWFPar>(name)
+{}
+
+// dependencies/products ///////////////////////////////////////////////////////
+template <typename FImpl>
+std::vector<std::string> TDWF<FImpl>::getInput(void)
+{
+    std::vector<std::string> in = {par().gauge};
+    
+    return in;
+}
+
+template <typename FImpl>
+std::vector<std::string> TDWF<FImpl>::getOutput(void)
+{
+    std::vector<std::string> out = {getName()};
+    
+    return out;
+}
+
+// setup ///////////////////////////////////////////////////////////////////////
+template <typename FImpl>
+void TDWF<FImpl>::setup(void)
+{
+    unsigned int size;
+    
+    size = 2*env().template lattice4dSize<typename FImpl::DoubledGaugeField>();
+    env().registerObject(getName(), size, par().Ls);
+}
+
+// execution ///////////////////////////////////////////////////////////////////
+template <typename FImpl>
+void TDWF<FImpl>::execute(void)
+{
+    LOG(Message) << "Setting up domain wall fermion matrix with m= "
+                 << par().mass << ", M5= " << par().M5 << " and Ls= "
+                 << par().Ls << " using gauge field '" << par().gauge << "'"
+                 << 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);
+    FMat *fMatPt = new DomainWallFermion<FImpl>(U, g5, grb5, g4, grb4,
+                                                par().mass, par().M5);
+    env().setObject(getName(), fMatPt);
+}
+
+END_MODULE_NAMESPACE
+
+END_HADRONS_NAMESPACE
+
+#endif // Hadrons_DWF_hpp_

extras/Hadrons/Modules/MAction/Wilson.hpp

@@ -0,0 +1,126 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Modules/MAction/Wilson.hpp
+
+Copyright (C) 2015
+Copyright (C) 2016
+
+Author: Antonin Portelli <antonin.portelli@me.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_Wilson_hpp_
+#define Hadrons_Wilson_hpp_
+
+#include <Grid/Hadrons/Global.hpp>
+#include <Grid/Hadrons/Module.hpp>
+#include <Grid/Hadrons/ModuleFactory.hpp>
+
+BEGIN_HADRONS_NAMESPACE
+
+/******************************************************************************
+ *                            TWilson quark action                            *
+ ******************************************************************************/
+BEGIN_MODULE_NAMESPACE(MAction)
+
+class WilsonPar: Serializable
+{
+public:
+    GRID_SERIALIZABLE_CLASS_MEMBERS(WilsonPar,
+                                    std::string, gauge,
+                                    double     , mass);
+};
+
+template <typename FImpl>
+class TWilson: public Module<WilsonPar>
+{
+public:
+    TYPE_ALIASES(FImpl,);
+public:
+    // constructor
+    TWilson(const std::string name);
+    // destructor
+    virtual ~TWilson(void) = default;
+    // dependencies/products
+    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(Wilson, TWilson<FIMPL>, MAction);
+
+/******************************************************************************
+ *                     TWilson template implementation                        *
+ ******************************************************************************/
+// constructor /////////////////////////////////////////////////////////////////
+template <typename FImpl>
+TWilson<FImpl>::TWilson(const std::string name)
+: Module<WilsonPar>(name)
+{}
+
+// dependencies/products ///////////////////////////////////////////////////////
+template <typename FImpl>
+std::vector<std::string> TWilson<FImpl>::getInput(void)
+{
+    std::vector<std::string> in = {par().gauge};
+    
+    return in;
+}
+
+template <typename FImpl>
+std::vector<std::string> TWilson<FImpl>::getOutput(void)
+{
+    std::vector<std::string> out = {getName()};
+    
+    return out;
+}
+
+// setup ///////////////////////////////////////////////////////////////////////
+template <typename FImpl>
+void TWilson<FImpl>::setup(void)
+{
+    unsigned int size;
+    
+    size = 2*env().template lattice4dSize<typename FImpl::DoubledGaugeField>();
+    env().registerObject(getName(), size);
+}
+
+// execution ///////////////////////////////////////////////////////////////////
+template <typename FImpl>
+void TWilson<FImpl>::execute()
+{
+    LOG(Message) << "Setting up TWilson fermion matrix with m= " << par().mass
+                 << " using gauge field '" << par().gauge << "'" << 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);
+    env().setObject(getName(), fMatPt);
+}
+
+END_MODULE_NAMESPACE
+
+END_HADRONS_NAMESPACE
+
+#endif // Hadrons_Wilson_hpp_

extras/Hadrons/Modules/MContraction/Baryon.hpp

@@ -0,0 +1,131 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Modules/MContraction/Baryon.hpp
+
+Copyright (C) 2015
+Copyright (C) 2016
+
+Author: Antonin Portelli <antonin.portelli@me.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_Baryon_hpp_
+#define Hadrons_Baryon_hpp_
+
+#include <Grid/Hadrons/Global.hpp>
+#include <Grid/Hadrons/Module.hpp>
+#include <Grid/Hadrons/ModuleFactory.hpp>
+
+BEGIN_HADRONS_NAMESPACE
+
+/******************************************************************************
+ *                               Baryon                                       *
+ ******************************************************************************/
+BEGIN_MODULE_NAMESPACE(MContraction)
+
+class BaryonPar: Serializable
+{
+public:
+    GRID_SERIALIZABLE_CLASS_MEMBERS(BaryonPar,
+                                    std::string, q1,
+                                    std::string, q2,
+                                    std::string, q3,
+                                    std::string, output);
+};
+
+template <typename FImpl1, typename FImpl2, typename FImpl3>
+class TBaryon: public Module<BaryonPar>
+{
+public:
+    TYPE_ALIASES(FImpl1, 1);
+    TYPE_ALIASES(FImpl2, 2);
+    TYPE_ALIASES(FImpl3, 3);
+    class Result: Serializable
+    {
+    public:
+        GRID_SERIALIZABLE_CLASS_MEMBERS(Result,
+                                        std::vector<std::vector<std::vector<Complex>>>, corr);
+    };
+public:
+    // constructor
+    TBaryon(const std::string name);
+    // destructor
+    virtual ~TBaryon(void) = default;
+    // dependency relation
+    virtual std::vector<std::string> getInput(void);
+    virtual std::vector<std::string> getOutput(void);
+    // execution
+    virtual void execute(void);
+};
+
+MODULE_REGISTER_NS(Baryon, ARG(TBaryon<FIMPL, FIMPL, FIMPL>), MContraction);
+
+/******************************************************************************
+ *                         TBaryon implementation                             *
+ ******************************************************************************/
+// constructor /////////////////////////////////////////////////////////////////
+template <typename FImpl1, typename FImpl2, typename FImpl3>
+TBaryon<FImpl1, FImpl2, FImpl3>::TBaryon(const std::string name)
+: Module<BaryonPar>(name)
+{}
+
+// dependencies/products ///////////////////////////////////////////////////////
+template <typename FImpl1, typename FImpl2, typename FImpl3>
+std::vector<std::string> TBaryon<FImpl1, FImpl2, FImpl3>::getInput(void)
+{
+    std::vector<std::string> input = {par().q1, par().q2, par().q3};
+    
+    return input;
+}
+
+template <typename FImpl1, typename FImpl2, typename FImpl3>
+std::vector<std::string> TBaryon<FImpl1, FImpl2, FImpl3>::getOutput(void)
+{
+    std::vector<std::string> out = {getName()};
+    
+    return out;
+}
+
+// execution ///////////////////////////////////////////////////////////////////
+template <typename FImpl1, typename FImpl2, typename FImpl3>
+void TBaryon<FImpl1, FImpl2, FImpl3>::execute(void)
+{
+    LOG(Message) << "Computing baryon contractions '" << getName() << "' using"
+                 << " quarks '" << par().q1 << "', '" << par().q2 << "', and '"
+                 << par().q3 << "'" << std::endl;
+    
+    XmlWriter             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);
+    LatticeComplex        c(env().getGrid());
+    Result                result;
+    
+    // FIXME: do contractions
+    
+    write(writer, "meson", result);
+}
+
+END_MODULE_NAMESPACE
+
+END_HADRONS_NAMESPACE
+
+#endif // Hadrons_Baryon_hpp_

extras/Hadrons/Modules/MContraction/Meson.hpp

@@ -0,0 +1,148 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Modules/MContraction/Meson.hpp
+
+Copyright (C) 2015
+Copyright (C) 2016
+
+Author: Antonin Portelli <antonin.portelli@me.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_Meson_hpp_
+#define Hadrons_Meson_hpp_
+
+#include <Grid/Hadrons/Global.hpp>
+#include <Grid/Hadrons/Module.hpp>
+#include <Grid/Hadrons/ModuleFactory.hpp>
+
+BEGIN_HADRONS_NAMESPACE
+
+/******************************************************************************
+ *                                TMeson                                       *
+ ******************************************************************************/
+BEGIN_MODULE_NAMESPACE(MContraction)
+
+class MesonPar: Serializable
+{
+public:
+    GRID_SERIALIZABLE_CLASS_MEMBERS(MesonPar,
+                                    std::string, q1,
+                                    std::string, q2,
+                                    std::string, output);
+};
+
+template <typename FImpl1, typename FImpl2>
+class TMeson: public Module<MesonPar>
+{
+public:
+    TYPE_ALIASES(FImpl1, 1);
+    TYPE_ALIASES(FImpl2, 2);
+    class Result: Serializable
+    {
+    public:
+        GRID_SERIALIZABLE_CLASS_MEMBERS(Result,
+                                        std::vector<std::vector<std::vector<Complex>>>, corr);
+    };
+public:
+    // constructor
+    TMeson(const std::string name);
+    // destructor
+    virtual ~TMeson(void) = default;
+    // dependencies/products
+    virtual std::vector<std::string> getInput(void);
+    virtual std::vector<std::string> getOutput(void);
+    // execution
+    virtual void execute(void);
+};
+
+MODULE_REGISTER_NS(Meson, ARG(TMeson<FIMPL, FIMPL>), MContraction);
+
+/******************************************************************************
+ *                           TMeson implementation                            *
+ ******************************************************************************/
+// constructor /////////////////////////////////////////////////////////////////
+template <typename FImpl1, typename FImpl2>
+TMeson<FImpl1, FImpl2>::TMeson(const std::string name)
+: Module<MesonPar>(name)
+{}
+
+// dependencies/products ///////////////////////////////////////////////////////
+template <typename FImpl1, typename FImpl2>
+std::vector<std::string> TMeson<FImpl1, FImpl2>::getInput(void)
+{
+    std::vector<std::string> input = {par().q1, par().q2};
+    
+    return input;
+}
+
+template <typename FImpl1, typename FImpl2>
+std::vector<std::string> TMeson<FImpl1, FImpl2>::getOutput(void)
+{
+    std::vector<std::string> output = {getName()};
+    
+    return output;
+}
+
+// execution ///////////////////////////////////////////////////////////////////
+template <typename FImpl1, typename FImpl2>
+void TMeson<FImpl1, FImpl2>::execute(void)
+{
+    LOG(Message) << "Computing meson contractions '" << getName() << "' using"
+                 << " quarks '" << par().q1 << "' and '" << par().q2 << "'"
+                 << std::endl;
+    
+    XmlWriter             writer(par().output);
+    PropagatorField1      &q1 = *env().template getObject<PropagatorField1>(par().q1);
+    PropagatorField2      &q2 = *env().template getObject<PropagatorField2>(par().q2);
+    LatticeComplex        c(env().getGrid());
+    SpinMatrix            g[Ns*Ns], g5;
+    std::vector<TComplex> buf;
+    Result                result;
+    
+    g5 = makeGammaProd(Ns*Ns - 1);
+    result.corr.resize(Ns*Ns);
+    for (unsigned int i = 0; i < Ns*Ns; ++i)
+    {
+        g[i] = makeGammaProd(i);
+    }
+    for (unsigned int iSink = 0; iSink < Ns*Ns; ++iSink)
+    {
+        result.corr[iSink].resize(Ns*Ns);
+        for (unsigned int iSrc = 0; iSrc < Ns*Ns; ++iSrc)
+        {
+            c = trace(g[iSink]*q1*g[iSrc]*g5*adj(q2)*g5);
+            sliceSum(c, buf, Tp);
+            result.corr[iSink][iSrc].resize(buf.size());
+            for (unsigned int t = 0; t < buf.size(); ++t)
+            {
+                result.corr[iSink][iSrc][t] = TensorRemove(buf[t]);
+            }
+        }
+    }
+    write(writer, "meson", result);
+}
+
+END_MODULE_NAMESPACE
+
+END_HADRONS_NAMESPACE
+
+#endif // Hadrons_Meson_hpp_

extras/Hadrons/Modules/MGauge/Load.cc

@@ -0,0 +1,78 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Modules/MGauge/Load.cc
+
+Copyright (C) 2015
+Copyright (C) 2016
+
+Author: Antonin Portelli <antonin.portelli@me.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/MGauge/Load.hpp>
+
+using namespace Grid;
+using namespace Hadrons;
+using namespace MGauge;
+
+/******************************************************************************
+*                           TLoad implementation                               *
+******************************************************************************/
+// constructor /////////////////////////////////////////////////////////////////
+TLoad::TLoad(const std::string name)
+: Module<LoadPar>(name)
+{}
+
+// dependencies/products ///////////////////////////////////////////////////////
+std::vector<std::string> TLoad::getInput(void)
+{
+    std::vector<std::string> in;
+    
+    return in;
+}
+
+std::vector<std::string> TLoad::getOutput(void)
+{
+    std::vector<std::string> out = {getName()};
+    
+    return out;
+}
+
+// setup ///////////////////////////////////////////////////////////////////////
+void TLoad::setup(void)
+{
+    env().registerLattice<LatticeGaugeField>(getName());
+}
+
+// execution ///////////////////////////////////////////////////////////////////
+void TLoad::execute(void)
+{
+    NerscField  header;
+    std::string fileName = par().file + "."
+                           + std::to_string(env().getTrajectory());
+    
+    LOG(Message) << "Loading NERSC configuration from file '" << fileName
+                 << "'" << std::endl;
+    LatticeGaugeField &U = *env().createLattice<LatticeGaugeField>(getName());
+    NerscIO::readConfiguration(U, header, fileName);
+    LOG(Message) << "NERSC header:" << std::endl;
+    dump_nersc_header(header, LOG(Message));
+}

extras/Hadrons/Modules/MGauge/Load.hpp

@@ -0,0 +1,73 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Modules/MGauge/Load.hpp
+
+Copyright (C) 2015
+Copyright (C) 2016
+
+Author: Antonin Portelli <antonin.portelli@me.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_Load_hpp_
+#define Hadrons_Load_hpp_
+
+#include <Grid/Hadrons/Global.hpp>
+#include <Grid/Hadrons/Module.hpp>
+#include <Grid/Hadrons/ModuleFactory.hpp>
+
+BEGIN_HADRONS_NAMESPACE
+
+/******************************************************************************
+ *                         Load a NERSC configuration                         *
+ ******************************************************************************/
+BEGIN_MODULE_NAMESPACE(MGauge)
+
+class LoadPar: Serializable
+{
+public:
+    GRID_SERIALIZABLE_CLASS_MEMBERS(LoadPar,
+                                    std::string, file);
+};
+
+class TLoad: public Module<LoadPar>
+{
+public:
+    // constructor
+    TLoad(const std::string name);
+    // destructor
+    virtual ~TLoad(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(Load, TLoad, MGauge);
+
+END_MODULE_NAMESPACE
+
+END_HADRONS_NAMESPACE
+
+#endif // Hadrons_Load_hpp_

extras/Hadrons/Modules/MGauge/Random.cc

@@ -0,0 +1,69 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Modules/MGauge/Random.cc
+
+Copyright (C) 2015
+Copyright (C) 2016
+
+Author: Antonin Portelli <antonin.portelli@me.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/MGauge/Random.hpp>
+
+using namespace Grid;
+using namespace Hadrons;
+using namespace MGauge;
+
+/******************************************************************************
+*                           TRandom implementation                            *
+******************************************************************************/
+// constructor /////////////////////////////////////////////////////////////////
+TRandom::TRandom(const std::string name)
+: Module<NoPar>(name)
+{}
+
+// dependencies/products ///////////////////////////////////////////////////////
+std::vector<std::string> TRandom::getInput(void)
+{
+    return std::vector<std::string>();
+}
+
+std::vector<std::string> TRandom::getOutput(void)
+{
+    std::vector<std::string> out = {getName()};
+    
+    return out;
+}
+
+// setup ///////////////////////////////////////////////////////////////////////
+void TRandom::setup(void)
+{
+    env().registerLattice<LatticeGaugeField>(getName());
+}
+
+// execution ///////////////////////////////////////////////////////////////////
+void TRandom::execute(void)
+{
+    LOG(Message) << "Generating random gauge configuration" << std::endl;
+    LatticeGaugeField &U = *env().createLattice<LatticeGaugeField>(getName());
+    SU3::HotConfiguration(*env().get4dRng(), U);
+}

extras/Hadrons/Modules/MGauge/Random.hpp

@@ -0,0 +1,66 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Modules/MGauge/Random.hpp
+
+Copyright (C) 2015
+Copyright (C) 2016
+
+Author: Antonin Portelli <antonin.portelli@me.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_Random_hpp_
+#define Hadrons_Random_hpp_
+
+#include <Grid/Hadrons/Global.hpp>
+#include <Grid/Hadrons/Module.hpp>
+#include <Grid/Hadrons/ModuleFactory.hpp>
+
+BEGIN_HADRONS_NAMESPACE
+
+/******************************************************************************
+ *                             Random gauge                                   *
+ ******************************************************************************/
+BEGIN_MODULE_NAMESPACE(MGauge)
+
+class TRandom: public Module<NoPar>
+{
+public:
+    // constructor
+    TRandom(const std::string name);
+    // destructor
+    virtual ~TRandom(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(Random, TRandom, MGauge);
+
+END_MODULE_NAMESPACE
+
+END_HADRONS_NAMESPACE
+
+#endif // Hadrons_Random_hpp_

extras/Hadrons/Modules/MGauge/Unit.cc

@@ -0,0 +1,69 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Modules/MGauge/Unit.cc
+
+Copyright (C) 2015
+Copyright (C) 2016
+
+Author: Antonin Portelli <antonin.portelli@me.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/MGauge/Unit.hpp>
+
+using namespace Grid;
+using namespace Hadrons;
+using namespace MGauge;
+
+/******************************************************************************
+*                            TUnit implementation                             *
+******************************************************************************/
+// constructor /////////////////////////////////////////////////////////////////
+TUnit::TUnit(const std::string name)
+: Module<NoPar>(name)
+{}
+
+// dependencies/products ///////////////////////////////////////////////////////
+std::vector<std::string> TUnit::getInput(void)
+{
+    return std::vector<std::string>();
+}
+
+std::vector<std::string> TUnit::getOutput(void)
+{
+    std::vector<std::string> out = {getName()};
+    
+    return out;
+}
+
+// setup ///////////////////////////////////////////////////////////////////////
+void TUnit::setup(void)
+{
+    env().registerLattice<LatticeGaugeField>(getName());
+}
+
+// execution ///////////////////////////////////////////////////////////////////
+void TUnit::execute(void)
+{
+    LOG(Message) << "Creating unit gauge configuration" << std::endl;
+    LatticeGaugeField &U = *env().createLattice<LatticeGaugeField>(getName());
+    SU3::ColdConfiguration(*env().get4dRng(), U);
+}

extras/Hadrons/Modules/MGauge/Unit.hpp

@@ -0,0 +1,66 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Modules/MGauge/Unit.hpp
+
+Copyright (C) 2015
+Copyright (C) 2016
+
+Author: Antonin Portelli <antonin.portelli@me.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_Unit_hpp_
+#define Hadrons_Unit_hpp_
+
+#include <Grid/Hadrons/Global.hpp>
+#include <Grid/Hadrons/Module.hpp>
+#include <Grid/Hadrons/ModuleFactory.hpp>
+
+BEGIN_HADRONS_NAMESPACE
+
+/******************************************************************************
+ *                              Unit gauge                                    *
+ ******************************************************************************/
+BEGIN_MODULE_NAMESPACE(MGauge)
+
+class TUnit: public Module<NoPar>
+{
+public:
+    // constructor
+    TUnit(const std::string name);
+    // destructor
+    virtual ~TUnit(void) = default;
+    // dependencies/products
+    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(Unit, TUnit, MGauge);
+
+END_MODULE_NAMESPACE
+
+END_HADRONS_NAMESPACE
+
+#endif // Hadrons_Unit_hpp_

extras/Hadrons/Modules/MSolver/RBPrecCG.hpp

@@ -0,0 +1,132 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Modules/MSolver/RBPrecCG.hpp
+
+Copyright (C) 2015
+Copyright (C) 2016
+
+Author: Antonin Portelli <antonin.portelli@me.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_RBPrecCG_hpp_
+#define Hadrons_RBPrecCG_hpp_
+
+#include <Grid/Hadrons/Global.hpp>
+#include <Grid/Hadrons/Module.hpp>
+#include <Grid/Hadrons/ModuleFactory.hpp>
+
+BEGIN_HADRONS_NAMESPACE
+
+/******************************************************************************
+ *                     Schur red-black preconditioned CG                      *
+ ******************************************************************************/
+BEGIN_MODULE_NAMESPACE(MSolver)
+
+class RBPrecCGPar: Serializable
+{
+public:
+    GRID_SERIALIZABLE_CLASS_MEMBERS(RBPrecCGPar,
+                                    std::string, action,
+                                    double     , residual);
+};
+
+template <typename FImpl>
+class TRBPrecCG: public Module<RBPrecCGPar>
+{
+public:
+    TYPE_ALIASES(FImpl,);
+public:
+    // constructor
+    TRBPrecCG(const std::string name);
+    // destructor
+    virtual ~TRBPrecCG(void) = default;
+    // dependencies/products
+    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(RBPrecCG, TRBPrecCG<FIMPL>, MSolver);
+
+/******************************************************************************
+ *                      TRBPrecCG template implementation                     *
+ ******************************************************************************/
+// constructor /////////////////////////////////////////////////////////////////
+template <typename FImpl>
+TRBPrecCG<FImpl>::TRBPrecCG(const std::string name)
+: Module(name)
+{}
+
+// dependencies/products ///////////////////////////////////////////////////////
+template <typename FImpl>
+std::vector<std::string> TRBPrecCG<FImpl>::getInput(void)
+{
+    std::vector<std::string> in = {par().action};
+    
+    return in;
+}
+
+template <typename FImpl>
+std::vector<std::string> TRBPrecCG<FImpl>::getOutput(void)
+{
+    std::vector<std::string> out = {getName()};
+    
+    return out;
+}
+
+// setup ///////////////////////////////////////////////////////////////////////
+template <typename FImpl>
+void TRBPrecCG<FImpl>::setup(void)
+{
+    auto Ls = env().getObjectLs(par().action);
+    
+    env().registerObject(getName(), 0, Ls);
+    env().addOwnership(getName(), par().action);
+}
+
+// execution ///////////////////////////////////////////////////////////////////
+template <typename FImpl>
+void TRBPrecCG<FImpl>::execute(void)
+{
+    auto &mat   = *(env().template getObject<FMat>(par().action));
+    auto solver = [&mat, this](FermionField &sol, const FermionField &source)
+    {
+        ConjugateGradient<FermionField>           cg(par().residual, 10000);
+        SchurRedBlackDiagMooeeSolve<FermionField> schurSolver(cg);
+        
+        schurSolver(mat, source, sol);
+    };
+    
+    LOG(Message) << "setting up Schur red-black preconditioned CG for"
+                 << " action '" << par().action << "' with residual "
+                 << par().residual << std::endl;
+    env().setObject(getName(), new SolverFn(solver));
+}
+
+END_MODULE_NAMESPACE
+
+END_HADRONS_NAMESPACE
+
+#endif // Hadrons_RBPrecCG_hpp_

extras/Hadrons/Modules/MSource/Point.hpp

@@ -0,0 +1,135 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Modules/MSource/Point.hpp
+
+Copyright (C) 2015
+Copyright (C) 2016
+
+Author: Antonin Portelli <antonin.portelli@me.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_Point_hpp_
+#define Hadrons_Point_hpp_
+
+#include <Grid/Hadrons/Global.hpp>
+#include <Grid/Hadrons/Module.hpp>
+#include <Grid/Hadrons/ModuleFactory.hpp>
+
+BEGIN_HADRONS_NAMESPACE
+
+/*
+ 
+ Point source
+ ------------
+ * src_x = delta_x,position
+ 
+ * options:
+ - position: space-separated integer sequence (e.g. "0 1 1 0")
+ 
+ */
+
+/******************************************************************************
+ *                                  TPoint                                     *
+ ******************************************************************************/
+BEGIN_MODULE_NAMESPACE(MSource)
+
+class PointPar: Serializable
+{
+public:
+    GRID_SERIALIZABLE_CLASS_MEMBERS(PointPar,
+                                    std::string, position);
+};
+
+template <typename FImpl>
+class TPoint: public Module<PointPar>
+{
+public:
+    TYPE_ALIASES(FImpl,);
+public:
+    // constructor
+    TPoint(const std::string name);
+    // destructor
+    virtual ~TPoint(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(Point, TPoint<FIMPL>, MSource);
+
+/******************************************************************************
+ *                       TPoint template implementation                       *
+ ******************************************************************************/
+// constructor /////////////////////////////////////////////////////////////////
+template <typename FImpl>
+TPoint<FImpl>::TPoint(const std::string name)
+: Module<PointPar>(name)
+{}
+
+// dependencies/products ///////////////////////////////////////////////////////
+template <typename FImpl>
+std::vector<std::string> TPoint<FImpl>::getInput(void)
+{
+    std::vector<std::string> in;
+    
+    return in;
+}
+
+template <typename FImpl>
+std::vector<std::string> TPoint<FImpl>::getOutput(void)
+{
+    std::vector<std::string> out = {getName()};
+    
+    return out;
+}
+
+// setup ///////////////////////////////////////////////////////////////////////
+template <typename FImpl>
+void TPoint<FImpl>::setup(void)
+{
+    env().template registerLattice<PropagatorField>(getName());
+}
+
+// execution ///////////////////////////////////////////////////////////////////
+template <typename FImpl>
+void TPoint<FImpl>::execute(void)
+{
+    std::vector<int> position = strToVec<int>(par().position);
+    typename SitePropagator::scalar_object id;
+    
+    LOG(Message) << "Creating point source at position [" << par().position
+                 << "]" << std::endl;
+    PropagatorField &src = *env().template createLattice<PropagatorField>(getName());
+    id  = 1.;
+    src = zero;
+    pokeSite(id, src, position);
+}
+
+END_MODULE_NAMESPACE
+
+END_HADRONS_NAMESPACE
+
+#endif // Hadrons_Point_hpp_

extras/Hadrons/Modules/MSource/SeqGamma.hpp

@@ -0,0 +1,164 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Modules/MSource/SeqGamma.hpp
+
+Copyright (C) 2015
+Copyright (C) 2016
+
+Author: Antonin Portelli <antonin.portelli@me.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_SeqGamma_hpp_
+#define Hadrons_SeqGamma_hpp_
+
+#include <Grid/Hadrons/Global.hpp>
+#include <Grid/Hadrons/Module.hpp>
+#include <Grid/Hadrons/ModuleFactory.hpp>
+
+BEGIN_HADRONS_NAMESPACE
+
+/*
+ 
+ Sequential source
+ -----------------------------
+ * src_x = q_x * theta(x_3 - tA) * theta(tB - x_3) * gamma * exp(i x.mom)
+ 
+ * options:
+ - q: input propagator (string)
+ - tA: begin timeslice (integer)
+ - tB: end timesilce (integer)
+ - gamma: gamma product to insert (integer)
+ - mom: momentum insertion, space-separated float sequence (e.g ".1 .2 1. 0.")
+ 
+ */
+
+/******************************************************************************
+ *                         SeqGamma                                 *
+ ******************************************************************************/
+BEGIN_MODULE_NAMESPACE(MSource)
+
+class SeqGammaPar: Serializable
+{
+public:
+    GRID_SERIALIZABLE_CLASS_MEMBERS(SeqGammaPar,
+                                    std::string,  q,
+                                    unsigned int, tA,
+                                    unsigned int, tB,
+                                    unsigned int, gamma,
+                                    std::string,  mom);
+};
+
+template <typename FImpl>
+class TSeqGamma: public Module<SeqGammaPar>
+{
+public:
+    TYPE_ALIASES(FImpl,);
+public:
+    // constructor
+    TSeqGamma(const std::string name);
+    // destructor
+    virtual ~TSeqGamma(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(SeqGamma, TSeqGamma<FIMPL>, MSource);
+
+/******************************************************************************
+ *                         TSeqGamma implementation                           *
+ ******************************************************************************/
+// constructor /////////////////////////////////////////////////////////////////
+template <typename FImpl>
+TSeqGamma<FImpl>::TSeqGamma(const std::string name)
+: Module<SeqGammaPar>(name)
+{}
+
+// dependencies/products ///////////////////////////////////////////////////////
+template <typename FImpl>
+std::vector<std::string> TSeqGamma<FImpl>::getInput(void)
+{
+    std::vector<std::string> in = {par().q};
+    
+    return in;
+}
+
+template <typename FImpl>
+std::vector<std::string> TSeqGamma<FImpl>::getOutput(void)
+{
+    std::vector<std::string> out = {getName()};
+    
+    return out;
+}
+
+// setup ///////////////////////////////////////////////////////////////////////
+template <typename FImpl>
+void TSeqGamma<FImpl>::setup(void)
+{
+    env().template registerLattice<PropagatorField>(getName());
+}
+
+// execution ///////////////////////////////////////////////////////////////////
+template <typename FImpl>
+void TSeqGamma<FImpl>::execute(void)
+{
+    if (par().tA == par().tB)
+    {
+        LOG(Message) << "Generating gamma_" << par().gamma
+                     << " sequential source at t= " << par().tA << std::endl;
+    }
+    else
+    {
+        LOG(Message) << "Generating gamma_" << par().gamma
+                     << " sequential source for "
+                     << par().tA << " <= t <= " << par().tB << std::endl;
+    }
+    PropagatorField &src = *env().template createLattice<PropagatorField>(getName());
+    PropagatorField &q   = *env().template getObject<PropagatorField>(par().q);
+    Lattice<iScalar<vInteger>> t(env().getGrid());
+    LatticeComplex             ph(env().getGrid()), coor(env().getGrid());
+    SpinMatrix                 g;
+    std::vector<Real>          p;
+    Complex                    i(0.0,1.0);
+    
+    g  = makeGammaProd(par().gamma);
+    p  = strToVec<Real>(par().mom);
+    ph = zero;
+    for(unsigned int mu = 0; mu < Nd; mu++)
+    {
+        LatticeCoordinate(coor, mu);
+        ph = ph + p[mu]*coor;
+    }
+    ph = exp(i*ph);
+    LatticeCoordinate(t, Tp);
+    src = where((t >= par().tA) and (t <= par().tB), g*ph*q, 0.*q);
+}
+
+END_MODULE_NAMESPACE
+
+END_HADRONS_NAMESPACE
+
+#endif // Hadrons_SeqGamma_hpp_

extras/Hadrons/Modules/MSource/Z2.hpp

@@ -0,0 +1,151 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Modules/MSource/Z2.hpp
+
+Copyright (C) 2015
+Copyright (C) 2016
+
+Author: Antonin Portelli <antonin.portelli@me.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_Z2_hpp_
+#define Hadrons_Z2_hpp_
+
+#include <Grid/Hadrons/Global.hpp>
+#include <Grid/Hadrons/Module.hpp>
+#include <Grid/Hadrons/ModuleFactory.hpp>
+
+BEGIN_HADRONS_NAMESPACE
+
+/*
+ 
+ Z_2 stochastic source
+ -----------------------------
+ * src_x = eta_x * theta(x_3 - tA) * theta(tB - x_3)
+ 
+ the eta_x are independent uniform random numbers in {+/- 1 +/- i}
+ 
+ * options:
+ - tA: begin timeslice (integer)
+ - tB: end timesilce (integer)
+ 
+ */
+ 
+/******************************************************************************
+ *                          Z2 stochastic source                              *
+ ******************************************************************************/
+BEGIN_MODULE_NAMESPACE(MSource)
+
+class Z2Par: Serializable
+{
+public:
+    GRID_SERIALIZABLE_CLASS_MEMBERS(Z2Par,
+                                    unsigned int, tA,
+                                    unsigned int, tB);
+};
+
+template <typename FImpl>
+class TZ2: public Module<Z2Par>
+{
+public:
+    TYPE_ALIASES(FImpl,);
+public:
+    // constructor
+    TZ2(const std::string name);
+    // destructor
+    virtual ~TZ2(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(Z2, TZ2<FIMPL>, MSource);
+
+/******************************************************************************
+ *                       TZ2 template implementation                          *
+ ******************************************************************************/
+// constructor /////////////////////////////////////////////////////////////////
+template <typename FImpl>
+TZ2<FImpl>::TZ2(const std::string name)
+: Module<Z2Par>(name)
+{}
+
+// dependencies/products ///////////////////////////////////////////////////////
+template <typename FImpl>
+std::vector<std::string> TZ2<FImpl>::getInput(void)
+{
+    std::vector<std::string> in;
+    
+    return in;
+}
+
+template <typename FImpl>
+std::vector<std::string> TZ2<FImpl>::getOutput(void)
+{
+    std::vector<std::string> out = {getName()};
+    
+    return out;
+}
+
+// setup ///////////////////////////////////////////////////////////////////////
+template <typename FImpl>
+void TZ2<FImpl>::setup(void)
+{
+    env().template registerLattice<PropagatorField>(getName());
+}
+
+// execution ///////////////////////////////////////////////////////////////////
+template <typename FImpl>
+void TZ2<FImpl>::execute(void)
+{
+    Lattice<iScalar<vInteger>> t(env().getGrid());
+    LatticeComplex             eta(env().getGrid());
+    Complex                    shift(1., 1.);
+    
+    if (par().tA == par().tB)
+    {
+        LOG(Message) << "Generating Z_2 wall source at t= " << par().tA
+        << std::endl;
+    }
+    else
+    {
+        LOG(Message) << "Generating Z_2 band for " << par().tA << " <= t <= "
+        << par().tB << std::endl;
+    }
+    PropagatorField &src = *env().template createLattice<PropagatorField>(getName());
+    LatticeCoordinate(t, Tp);
+    bernoulli(*env().get4dRng(), eta);
+    eta = (2.*eta - shift)*(1./::sqrt(2.));
+    eta = where((t >= par().tA) and (t <= par().tB), eta, 0.*eta);
+    src = 1.;
+    src = src*eta;
+}
+
+END_MODULE_NAMESPACE
+
+END_HADRONS_NAMESPACE
+
+#endif // Hadrons_Z2_hpp_

extras/Hadrons/Modules/Quark.hpp

@@ -0,0 +1,185 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid 
+
+Source file: extras/Hadrons/Modules/Quark.hpp
+
+Copyright (C) 2015
+Copyright (C) 2016
+
+Author: Antonin Portelli <antonin.portelli@me.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_Quark_hpp_
+#define Hadrons_Quark_hpp_
+
+#include <Grid/Hadrons/Global.hpp>
+#include <Grid/Hadrons/Module.hpp>
+#include <Grid/Hadrons/ModuleFactory.hpp>
+
+BEGIN_HADRONS_NAMESPACE
+
+/******************************************************************************
+ *                               TQuark                                       *
+ ******************************************************************************/
+class QuarkPar: Serializable
+{
+public:
+    GRID_SERIALIZABLE_CLASS_MEMBERS(QuarkPar,
+                                    std::string, source,
+                                    std::string, solver);
+};
+
+template <typename FImpl>
+class TQuark: public Module<QuarkPar>
+{
+public:
+    TYPE_ALIASES(FImpl,);
+public:
+    // constructor
+    TQuark(const std::string name);
+    // destructor
+    virtual ~TQuark(void) = default;
+    // dependencies/products
+    virtual std::vector<std::string> getInput(void);
+    virtual std::vector<std::string> getOutput(void);
+    // setup
+    virtual void setup(void);
+    // execution
+    virtual void execute(void);
+private:
+    unsigned int Ls_;
+    SolverFn     *solver_{nullptr};
+};
+
+MODULE_REGISTER(Quark, TQuark<FIMPL>);
+
+/******************************************************************************
+ *                          TQuark implementation                             *
+ ******************************************************************************/
+// constructor /////////////////////////////////////////////////////////////////
+template <typename FImpl>
+TQuark<FImpl>::TQuark(const std::string name)
+: Module(name)
+{}
+
+// dependencies/products ///////////////////////////////////////////////////////
+template <typename FImpl>
+std::vector<std::string> TQuark<FImpl>::getInput(void)
+{
+    std::vector<std::string> in = {par().source, par().solver};
+    
+    return in;
+}
+
+template <typename FImpl>
+std::vector<std::string> TQuark<FImpl>::getOutput(void)
+{
+    std::vector<std::string> out = {getName(), getName() + "_5d"};
+    
+    return out;
+}
+
+// setup ///////////////////////////////////////////////////////////////////////
+template <typename FImpl>
+void TQuark<FImpl>::setup(void)
+{
+    Ls_ = env().getObjectLs(par().solver);
+    env().template registerLattice<PropagatorField>(getName());
+    if (Ls_ > 1)
+    {
+        env().template registerLattice<PropagatorField>(getName() + "_5d", Ls_);
+    }
+}
+
+// execution ///////////////////////////////////////////////////////////////////
+template <typename FImpl>
+void TQuark<FImpl>::execute(void)
+{
+    LOG(Message) << "Computing quark propagator '" << getName() << "'"
+                 << std::endl;
+    
+    FermionField    source(env().getGrid(Ls_)), sol(env().getGrid(Ls_)),
+                    tmp(env().getGrid());
+    std::string     propName = (Ls_ == 1) ? getName() : (getName() + "_5d");
+    PropagatorField &prop    = *env().template createLattice<PropagatorField>(propName);
+    PropagatorField &fullSrc = *env().template getObject<PropagatorField>(par().source);
+    SolverFn        &solver  = *env().template getObject<SolverFn>(par().solver);
+    if (Ls_ > 1)
+    {
+        env().template createLattice<PropagatorField>(getName());
+    }
+    
+    LOG(Message) << "Inverting using solver '" << par().solver
+                 << "' on source '" << par().source << "'" << std::endl;
+    for (unsigned int s = 0; s < Ns; ++s)
+    for (unsigned int c = 0; c < Nc; ++c)
+    {
+        LOG(Message) << "Inversion for spin= " << s << ", color= " << c
+        << std::endl;
+        // source conversion for 4D sources
+        if (!env().isObject5d(par().source))
+        {
+            if (Ls_ == 1)
+            {
+                PropToFerm(source, fullSrc, s, c);
+            }
+            else
+            {
+                source = zero;
+                PropToFerm(tmp, fullSrc, s, c);
+                InsertSlice(tmp, source, 0, 0);
+                InsertSlice(tmp, source, Ls_-1, 0);
+                axpby_ssp_pplus(source, 0., source, 1., source, 0, 0);
+                axpby_ssp_pminus(source, 0., source, 1., source, Ls_-1, Ls_-1);
+            }
+        }
+        // source conversion for 5D sources
+        else
+        {
+            if (Ls_ != env().getObjectLs(par().source))
+            {
+                HADRON_ERROR("Ls mismatch between quark action and source");
+            }
+            else
+            {
+                PropToFerm(source, fullSrc, s, c);
+            }
+        }
+        sol = zero;
+        solver(sol, source);
+        FermToProp(prop, sol, s, c);
+        // create 4D propagators from 5D one if necessary
+        if (Ls_ > 1)
+        {
+            PropagatorField &p4d =
+                *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);
+            ExtractSlice(tmp, sol, 0, 0);
+            FermToProp(p4d, tmp, s, c);
+        }
+    }
+}
+
+END_HADRONS_NAMESPACE
+
+#endif // Hadrons_Quark_hpp_

extras/Hadrons/Modules/templates/Module.cc.template

@@ -0,0 +1,39 @@
+#include <Grid/Hadrons/Modules/___FILEBASENAME___.hpp>
+
+using namespace Grid;
+using namespace Hadrons;
+
+/******************************************************************************
+*                  T___FILEBASENAME___ implementation                             *
+******************************************************************************/
+// constructor /////////////////////////////////////////////////////////////////
+T___FILEBASENAME___::T___FILEBASENAME___(const std::string name)
+: Module<___FILEBASENAME___Par>(name)
+{}
+
+// dependencies/products ///////////////////////////////////////////////////////
+std::vector<std::string> T___FILEBASENAME___::getInput(void)
+{
+    std::vector<std::string> in;
+    
+    return in;
+}
+
+std::vector<std::string> T___FILEBASENAME___::getOutput(void)
+{
+    std::vector<std::string> out = {getName()};
+    
+    return out;
+}
+
+// setup ///////////////////////////////////////////////////////////////////////
+void T___FILEBASENAME___::setup(void)
+{
+
+}
+
+// execution ///////////////////////////////////////////////////////////////////
+void T___FILEBASENAME___::execute(void)
+{
+
+}

extras/Hadrons/Modules/templates/Module.hpp.template

@@ -0,0 +1,40 @@
+#ifndef Hadrons____FILEBASENAME____hpp_
+#define Hadrons____FILEBASENAME____hpp_
+
+#include <Grid/Hadrons/Global.hpp>
+#include <Grid/Hadrons/Module.hpp>
+#include <Grid/Hadrons/ModuleFactory.hpp>
+
+BEGIN_HADRONS_NAMESPACE
+
+/******************************************************************************
+ *                         ___FILEBASENAME___                                 *
+ ******************************************************************************/
+class ___FILEBASENAME___Par: Serializable
+{
+public:
+    GRID_SERIALIZABLE_CLASS_MEMBERS(___FILEBASENAME___Par, 
+                                    unsigned int, i);
+};
+
+class T___FILEBASENAME___: public Module<___FILEBASENAME___Par>
+{
+public:
+    // constructor
+    T___FILEBASENAME___(const std::string name);
+    // destructor
+    virtual ~T___FILEBASENAME___(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(___FILEBASENAME___, T___FILEBASENAME___);
+
+END_HADRONS_NAMESPACE
+
+#endif // Hadrons____FILEBASENAME____hpp_

extras/Hadrons/Modules/templates/Module_in_NS.cc.template

@@ -0,0 +1,40 @@
+#include <Grid/Hadrons/Modules/___NAMESPACE___/___FILEBASENAME___.hpp>
+
+using namespace Grid;
+using namespace Hadrons;
+using namespace ___NAMESPACE___;
+
+/******************************************************************************
+*                  T___FILEBASENAME___ implementation                             *
+******************************************************************************/
+// constructor /////////////////////////////////////////////////////////////////
+T___FILEBASENAME___::T___FILEBASENAME___(const std::string name)
+: Module<___FILEBASENAME___Par>(name)
+{}
+
+// dependencies/products ///////////////////////////////////////////////////////
+std::vector<std::string> T___FILEBASENAME___::getInput(void)
+{
+    std::vector<std::string> in;
+    
+    return in;
+}
+
+std::vector<std::string> T___FILEBASENAME___::getOutput(void)
+{
+    std::vector<std::string> out = {getName()};
+    
+    return out;
+}
+
+// setup ///////////////////////////////////////////////////////////////////////
+void T___FILEBASENAME___::setup(void)
+{
+
+}
+
+// execution ///////////////////////////////////////////////////////////////////
+void T___FILEBASENAME___::execute(void)
+{
+
+}

extras/Hadrons/Modules/templates/Module_in_NS.hpp.template

@@ -0,0 +1,44 @@
+#ifndef Hadrons____FILEBASENAME____hpp_
+#define Hadrons____FILEBASENAME____hpp_
+
+#include <Grid/Hadrons/Global.hpp>
+#include <Grid/Hadrons/Module.hpp>
+#include <Grid/Hadrons/ModuleFactory.hpp>
+
+BEGIN_HADRONS_NAMESPACE
+
+/******************************************************************************
+ *                         ___FILEBASENAME___                                 *
+ ******************************************************************************/
+BEGIN_MODULE_NAMESPACE(___NAMESPACE___)
+
+class ___FILEBASENAME___Par: Serializable
+{
+public:
+    GRID_SERIALIZABLE_CLASS_MEMBERS(___FILEBASENAME___Par,
+                                    unsigned int, i);
+};
+
+class T___FILEBASENAME___: public Module<___FILEBASENAME___Par>
+{
+public:
+    // constructor
+    T___FILEBASENAME___(const std::string name);
+    // destructor
+    virtual ~T___FILEBASENAME___(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(___FILEBASENAME___, T___FILEBASENAME___, ___NAMESPACE___);
+
+END_MODULE_NAMESPACE
+
+END_HADRONS_NAMESPACE
+
+#endif // Hadrons____FILEBASENAME____hpp_

extras/Hadrons/Modules/templates/Module_tmp.hpp.template

@@ -0,0 +1,81 @@
+#ifndef Hadrons____FILEBASENAME____hpp_
+#define Hadrons____FILEBASENAME____hpp_
+
+#include <Grid/Hadrons/Global.hpp>
+#include <Grid/Hadrons/Module.hpp>
+#include <Grid/Hadrons/ModuleFactory.hpp>
+
+BEGIN_HADRONS_NAMESPACE
+
+/******************************************************************************
+ *                         ___FILEBASENAME___                                 *
+ ******************************************************************************/
+class ___FILEBASENAME___Par: Serializable
+{
+public:
+    GRID_SERIALIZABLE_CLASS_MEMBERS(___FILEBASENAME___Par,
+                                    unsigned int, i);
+};
+
+template <typename FImpl>
+class T___FILEBASENAME___: public Module<___FILEBASENAME___Par>
+{
+public:
+    // constructor
+    T___FILEBASENAME___(const std::string name);
+    // destructor
+    virtual ~T___FILEBASENAME___(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(___FILEBASENAME___, T___FILEBASENAME___<FIMPL>);
+
+/******************************************************************************
+ *                 T___FILEBASENAME___ implementation                             *
+ ******************************************************************************/
+// constructor /////////////////////////////////////////////////////////////////
+template <typename FImpl>
+T___FILEBASENAME___<FImpl>::T___FILEBASENAME___(const std::string name)
+: Module<___FILEBASENAME___Par>(name)
+{}
+
+// dependencies/products ///////////////////////////////////////////////////////
+template <typename FImpl>
+std::vector<std::string> T___FILEBASENAME___<FImpl>::getInput(void)
+{
+    std::vector<std::string> in;
+    
+    return in;
+}
+
+template <typename FImpl>
+std::vector<std::string> T___FILEBASENAME___<FImpl>::getOutput(void)
+{
+    std::vector<std::string> out = {getName()};
+    
+    return out;
+}
+
+// setup ///////////////////////////////////////////////////////////////////////
+template <typename FImpl>
+void T___FILEBASENAME___<FImpl>::setup(void)
+{
+    
+}
+
+// execution ///////////////////////////////////////////////////////////////////
+template <typename FImpl>
+void T___FILEBASENAME___<FImpl>::execute(void)
+{
+    
+}
+
+END_HADRONS_NAMESPACE
+
+#endif // Hadrons____FILEBASENAME____hpp_

extras/Hadrons/Modules/templates/Module_tmp_in_NS.hpp.template

@@ -0,0 +1,85 @@
+#ifndef Hadrons____FILEBASENAME____hpp_
+#define Hadrons____FILEBASENAME____hpp_
+
+#include <Grid/Hadrons/Global.hpp>
+#include <Grid/Hadrons/Module.hpp>
+#include <Grid/Hadrons/ModuleFactory.hpp>
+
+BEGIN_HADRONS_NAMESPACE
+
+/******************************************************************************
+ *                         ___FILEBASENAME___                                 *
+ ******************************************************************************/
+BEGIN_MODULE_NAMESPACE(___NAMESPACE___)
+
+class ___FILEBASENAME___Par: Serializable
+{
+public:
+    GRID_SERIALIZABLE_CLASS_MEMBERS(___FILEBASENAME___Par,
+                                    unsigned int, i);
+};
+
+template <typename FImpl>
+class T___FILEBASENAME___: public Module<___FILEBASENAME___Par>
+{
+public:
+    // constructor
+    T___FILEBASENAME___(const std::string name);
+    // destructor
+    virtual ~T___FILEBASENAME___(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(___FILEBASENAME___, T___FILEBASENAME___<FIMPL>, ___NAMESPACE___);
+
+/******************************************************************************
+ *                 T___FILEBASENAME___ implementation                             *
+ ******************************************************************************/
+// constructor /////////////////////////////////////////////////////////////////
+template <typename FImpl>
+T___FILEBASENAME___<FImpl>::T___FILEBASENAME___(const std::string name)
+: Module<___FILEBASENAME___Par>(name)
+{}
+
+// dependencies/products ///////////////////////////////////////////////////////
+template <typename FImpl>
+std::vector<std::string> T___FILEBASENAME___<FImpl>::getInput(void)
+{
+    std::vector<std::string> in;
+    
+    return in;
+}
+
+template <typename FImpl>
+std::vector<std::string> T___FILEBASENAME___<FImpl>::getOutput(void)
+{
+    std::vector<std::string> out = {getName()};
+    
+    return out;
+}
+
+// setup ///////////////////////////////////////////////////////////////////////
+template <typename FImpl>
+void T___FILEBASENAME___<FImpl>::setup(void)
+{
+    
+}
+
+// execution ///////////////////////////////////////////////////////////////////
+template <typename FImpl>
+void T___FILEBASENAME___<FImpl>::execute(void)
+{
+    
+}
+
+END_MODULE_NAMESPACE
+
+END_HADRONS_NAMESPACE
+
+#endif // Hadrons____FILEBASENAME____hpp_

extras/Hadrons/add_module.sh

@@ -0,0 +1,31 @@
+#!/usr/bin/env bash
+
+if (( $# != 1 && $# != 2)); then
+    echo "usage: `basename $0` <module name> [<namespace>]" 1>&2
+    exit 1
+fi
+NAME=$1
+NS=$2
+
+if (( $# == 1 )); then
+	if [ -e "Modules/${NAME}.cc" ] || [ -e "Modules/${NAME}.hpp" ]; then
+	    echo "error: files Modules/${NAME}.* already exists" 1>&2
+	    exit 1
+	fi
+	sed "s/___FILEBASENAME___/${NAME}/g" Modules/templates/Module.cc.template > Modules/${NAME}.cc
+	sed "s/___FILEBASENAME___/${NAME}/g" Modules/templates/Module.hpp.template > Modules/${NAME}.hpp
+elif (( $# == 2 )); then
+	mkdir -p Modules/${NS}
+	if [ -e "Modules/${NS}/${NAME}.cc" ] || [ -e "Modules/${NS}/${NAME}.hpp" ]; then
+	    echo "error: files Modules/${NS}/${NAME}.* already exists" 1>&2
+	    exit 1
+	fi
+	TMPCC=".${NS}.${NAME}.tmp.cc"
+	TMPHPP=".${NS}.${NAME}.tmp.hpp"
+	sed "s/___FILEBASENAME___/${NAME}/g" Modules/templates/Module_in_NS.cc.template  > ${TMPCC}
+	sed "s/___FILEBASENAME___/${NAME}/g" Modules/templates/Module_in_NS.hpp.template > ${TMPHPP}
+	sed "s/___NAMESPACE___/${NS}/g" ${TMPCC}  > Modules/${NS}/${NAME}.cc
+	sed "s/___NAMESPACE___/${NS}/g" ${TMPHPP} > Modules/${NS}/${NAME}.hpp
+	rm -f ${TMPCC} ${TMPHPP}
+fi
+./make_module_list.sh

extras/Hadrons/add_module_template.sh

@@ -0,0 +1,28 @@
+#!/usr/bin/env bash
+
+if (( $# != 1 && $# != 2)); then
+    echo "usage: `basename $0` <module name> [<namespace>]" 1>&2
+    exit 1
+fi
+NAME=$1
+NS=$2
+
+if (( $# == 1 )); then
+	if [ -e "Modules/${NAME}.cc" ] || [ -e "Modules/${NAME}.hpp" ]; then
+	    echo "error: files Modules/${NAME}.* already exists" 1>&2
+	    exit 1
+	fi
+	sed "s/___FILEBASENAME___/${NAME}/g" Modules/templates/Module_tmp.hpp.template > Modules/${NAME}.hpp
+elif (( $# == 2 )); then
+	mkdir -p Modules/${NS}
+	if [ -e "Modules/${NS}/${NAME}.cc" ] || [ -e "Modules/${NS}/${NAME}.hpp" ]; then
+	    echo "error: files Modules/${NS}/${NAME}.* already exists" 1>&2
+	    exit 1
+	fi
+	TMPCC=".${NS}.${NAME}.tmp.cc"
+	TMPHPP=".${NS}.${NAME}.tmp.hpp"
+	sed "s/___FILEBASENAME___/${NAME}/g" Modules/templates/Module_tmp_in_NS.hpp.template > ${TMPHPP}
+	sed "s/___NAMESPACE___/${NS}/g" ${TMPHPP} > Modules/${NS}/${NAME}.hpp
+	rm -f ${TMPCC} ${TMPHPP}
+fi
+./make_module_list.sh

extras/Hadrons/make_module_list.sh

@@ -0,0 +1,12 @@
+#!/usr/bin/env bash
+
+echo 'modules_cc =\' > modules.inc
+find Modules -name '*.cc' -type f -print | sed 's/^/  /;$q;s/$/ \\/' >> modules.inc
+echo '' >> modules.inc
+echo 'modules_hpp =\' >> modules.inc
+find Modules -name '*.hpp' -type f -print | sed 's/^/  /;$q;s/$/ \\/' >> modules.inc
+echo '' >> modules.inc
+rm -f Modules.hpp
+for f in `find Modules -name '*.hpp'`; do
+	echo "#include <Grid/Hadrons/${f}>" >> Modules.hpp
+done

extras/Hadrons/modules.inc

@@ -0,0 +1,19 @@
+modules_cc =\
+  Modules/MGauge/Load.cc \
+  Modules/MGauge/Random.cc \
+  Modules/MGauge/Unit.cc
+
+modules_hpp =\
+  Modules/MAction/DWF.hpp \
+  Modules/MAction/Wilson.hpp \
+  Modules/MContraction/Baryon.hpp \
+  Modules/MContraction/Meson.hpp \
+  Modules/MGauge/Load.hpp \
+  Modules/MGauge/Random.hpp \
+  Modules/MGauge/Unit.hpp \
+  Modules/MSolver/RBPrecCG.hpp \
+  Modules/MSource/Point.hpp \
+  Modules/MSource/SeqGamma.hpp \
+  Modules/MSource/Z2.hpp \
+  Modules/Quark.hpp
+

extras/Makefile.am

@@ -0,0 +1 @@
+SUBDIRS = Hadrons

lib/FFT.h

@@ -29,8 +29,12 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 #ifndef _GRID_FFT_H_
 #define _GRID_FFT_H_
 
-#ifdef HAVE_FFTW	
-#include <Grid/fftw/fftw3.h>
+#ifdef HAVE_FFTW
+#ifdef USE_MKL
+#include <fftw/fftw3.h>
+#else
+#include <fftw3.h>
+#endif
 #endif
 
 
@@ -122,7 +126,8 @@ namespace Grid {
     
     double Flops(void) {return flops;}
     double MFlops(void) {return flops/usec;}
-    
+    double USec(void)   {return (double)usec;}    
+
     FFT ( GridCartesian * grid ) :
     vgrid(grid),
     Nd(grid->_ndimension),
@@ -226,28 +231,41 @@ namespace Grid {
       std::vector<int> lcoor(Nd), gcoor(Nd);
       result = source;
       for(int p=0;p<processors[dim];p++) {
-        for(int idx=0;idx<sgrid->lSites();idx++) {
-          sgrid->LocalIndexToLocalCoor(idx,lcoor);
+        PARALLEL_REGION
+        {
+          std::vector<int> cbuf(Nd);
           sobj s;
-          peekLocalSite(s,result,lcoor);
-          lcoor[dim]+=p*L;
-          pokeLocalSite(s,pgbuf,lcoor);
+          
+          PARALLEL_FOR_LOOP_INTERN
+          for(int idx=0;idx<sgrid->lSites();idx++) {
+            sgrid->LocalIndexToLocalCoor(idx,cbuf);
+            peekLocalSite(s,result,cbuf);
+            cbuf[dim]+=p*L;
+            pokeLocalSite(s,pgbuf,cbuf);
+          }
+        }
+        if (p != processors[dim] - 1)
+        {
+          result = Cshift(result,dim,L);
         }
-        result = Cshift(result,dim,L);
       }
       
       // Loop over orthog coords
       int NN=pencil_g.lSites();
       GridStopWatch timer;
       timer.Start();
-      //PARALLEL_FOR_LOOP
-      for(int idx=0;idx<NN;idx++) {
-        pencil_g.LocalIndexToLocalCoor(idx,lcoor);
+      PARALLEL_REGION
+      {
+        std::vector<int> cbuf(Nd);
         
-        if ( lcoor[dim] == 0 ) {  // restricts loop to plane at lcoor[dim]==0
-          FFTW_scalar *in = (FFTW_scalar *)&pgbuf._odata[idx];
-          FFTW_scalar *out= (FFTW_scalar *)&pgbuf._odata[idx];
-          FFTW<scalar>::fftw_execute_dft(p,in,out);
+        PARALLEL_FOR_LOOP_INTERN
+        for(int idx=0;idx<NN;idx++) {
+          pencil_g.LocalIndexToLocalCoor(idx, cbuf);
+          if ( cbuf[dim] == 0 ) {  // restricts loop to plane at lcoor[dim]==0
+            FFTW_scalar *in = (FFTW_scalar *)&pgbuf._odata[idx];
+            FFTW_scalar *out= (FFTW_scalar *)&pgbuf._odata[idx];
+            FFTW<scalar>::fftw_execute_dft(p,in,out);
+          }
         }
       }
       timer.Stop();
@@ -261,15 +279,21 @@ namespace Grid {
       
       // writing out result
       int pc = processor_coor[dim];
-      for(int idx=0;idx<sgrid->lSites();idx++) {
-        sgrid->LocalIndexToLocalCoor(idx,lcoor);
-        gcoor = lcoor;
+      PARALLEL_REGION
+      {
+        std::vector<int> clbuf(Nd), cgbuf(Nd);
         sobj s;
-        gcoor[dim] = lcoor[dim]+L*pc;
-        peekLocalSite(s,pgbuf,gcoor);
-		s = s * div;
-        pokeLocalSite(s,result,lcoor);
+        
+        PARALLEL_FOR_LOOP_INTERN
+        for(int idx=0;idx<sgrid->lSites();idx++) {
+          sgrid->LocalIndexToLocalCoor(idx,clbuf);
+          cgbuf = clbuf;
+          cgbuf[dim] = clbuf[dim]+L*pc;
+          peekLocalSite(s,pgbuf,cgbuf);
+          pokeLocalSite(s,result,clbuf);
+        }
       }
+      result = result*div;
       
       // destroying plan
       FFTW<scalar>::fftw_destroy_plan(p);

lib/Hadrons

@@ -0,0 +1 @@
+../extras/Hadrons

lib/Init.cc

@@ -369,7 +369,7 @@ void Grid_init(int *argc,char ***argv)
   
 void Grid_finalize(void)
 {
-#if defined (GRID_COMMS_MPI) || defined (GRID_COMMS_MPI3)
+#if defined (GRID_COMMS_MPI) || defined (GRID_COMMS_MPI3) 
   MPI_Finalize();
   Grid_unquiesce_nodes();
 #endif

lib/Log.cc

@@ -93,7 +93,7 @@ void GridLogConfigure(std::vector<std::string> &logstreams) {
 ////////////////////////////////////////////////////////////
 void Grid_quiesce_nodes(void) {
   int me = 0;
-#if defined(GRID_COMMS_MPI) || defined(GRID_COMMS_MPI3)
+#if defined(GRID_COMMS_MPI) || defined(GRID_COMMS_MPI3) || defined(GRID_COMMS_MPI3L)
   MPI_Comm_rank(MPI_COMM_WORLD, &me);
 #endif
 #ifdef GRID_COMMS_SHMEM

lib/Log.h

@@ -110,8 +110,8 @@ public:
   friend std::ostream& operator<< (std::ostream& stream, Logger& log){
 
     if ( log.active ) {
-      stream << log.background()<< log.topName << log.background()<< " : ";
-      stream << log.colour() <<std::setw(14) << std::left << log.name << log.background() << " : ";
+      stream << log.background()<< std::setw(10) << std::left << log.topName << log.background()<< " : ";
+      stream << log.colour() << std::setw(14) << std::left << log.name << log.background() << " : ";
       if ( log.timestamp ) {
 	StopWatch.Stop();
 	GridTime now = StopWatch.Elapsed();

lib/Threads.h

@@ -38,14 +38,21 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 #ifdef GRID_OMP
 #include <omp.h>
 #ifdef GRID_NUMA
-#define PARALLEL_FOR_LOOP _Pragma("omp parallel for schedule(static)")
+#define PARALLEL_FOR_LOOP        _Pragma("omp parallel for schedule(static)")
+#define PARALLEL_FOR_LOOP_INTERN _Pragma("omp for schedule(static)")
 #else
-#define PARALLEL_FOR_LOOP _Pragma("omp parallel for schedule(runtime)")
+#define PARALLEL_FOR_LOOP        _Pragma("omp parallel for schedule(runtime)")
+#define PARALLEL_FOR_LOOP_INTERN _Pragma("omp for schedule(runtime)")
 #endif
 #define PARALLEL_NESTED_LOOP2 _Pragma("omp parallel for collapse(2)")
+#define PARALLEL_REGION       _Pragma("omp parallel")
+#define PARALLEL_CRITICAL     _Pragma("omp critical")
 #else
-#define PARALLEL_FOR_LOOP 
+#define PARALLEL_FOR_LOOP
+#define PARALLEL_FOR_LOOP_INTERN
 #define PARALLEL_NESTED_LOOP2
+#define PARALLEL_REGION
+#define PARALLEL_CRITICAL
 #endif
 
 namespace Grid {

lib/algorithms/CoarsenedMatrix.h

@@ -282,7 +282,7 @@ PARALLEL_FOR_LOOP
 	  } else if(SE->_is_local) { 
 	    nbr = in._odata[SE->_offset];
 	  } else {
-	    nbr = Stencil.comm_buf[SE->_offset];
+	    nbr = Stencil.CommBuf()[SE->_offset];
 	  }
 	  res = res + A[point]._odata[ss]*nbr;
 	}

lib/algorithms/iterative/ConjugateGradient.h

@@ -154,7 +154,7 @@ class ConjugateGradient : public OperatorFunction<Field> {
                   << LinalgTimer.Elapsed();
         std::cout << std::endl;
 
-        if (ErrorOnNoConverge) assert(true_residual / Tolerance < 1000.0);
+        if (ErrorOnNoConverge) assert(true_residual / Tolerance < 10000.0);
 
         return;
       }

lib/algorithms/iterative/ImplicitlyRestartedLanczos.h

@@ -1080,10 +1080,10 @@ say con = 2
 **/
 
 template<class T>
-static void Lock(DenseMatrix<T> &H, 	///Hess mtx	
-		 DenseMatrix<T> &Q, 	///Lock Transform
-		 T val, 		///value to be locked
-		 int con, 	///number already locked
+static void Lock(DenseMatrix<T> &H, 	// Hess mtx	
+		 DenseMatrix<T> &Q, 	// Lock Transform
+		 T val, 		// value to be locked
+		 int con, 	// number already locked
 		 RealD small,
 		 int dfg,
 		 bool herm)

lib/communicator/Communicator_mpi3_leader.cc

@@ -39,6 +39,10 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 ///
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 #include <semaphore.h>
+#include <fcntl.h>
+#include <unistd.h>
+#include <limits.h>
+
 typedef sem_t *Grid_semaphore;
 
 #define SEM_INIT(S)      S = sem_open(sem_name,0,0600,0); assert ( S != SEM_FAILED );

lib/communicator/Communicator_none.cc

@@ -97,7 +97,7 @@ void CartesianCommunicator::Barrier(void){}
 void CartesianCommunicator::Broadcast(int root,void* data, int bytes) {}
 void CartesianCommunicator::BroadcastWorld(int root,void* data, int bytes) { }
 int  CartesianCommunicator::RankFromProcessorCoor(std::vector<int> &coor) {  return 0;}
-void CartesianCommunicator::ProcessorCoorFromRank(int rank, std::vector<int> &coor){  assert(0);}
+void CartesianCommunicator::ProcessorCoorFromRank(int rank, std::vector<int> &coor){ coor = _processor_coor ;}
 void CartesianCommunicator::ShiftedRanks(int dim,int shift,int &source,int &dest)
 {
   source =0;

lib/fftw/fftw3.h

@@ -1,412 +0,0 @@
-/*
- * Copyright (c) 2003, 2007-14 Matteo Frigo
- * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
- *
- * The following statement of license applies *only* to this header file,
- * and *not* to the other files distributed with FFTW or derived therefrom:
- * 
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- *
- * 1. Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- *
- * 2. Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in the
- *    documentation and/or other materials provided with the distribution.
- *
- * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS
- * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
- * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
- * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY
- * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
- * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE
- * GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
- * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
- * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
- * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
- * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
- */
-
-/***************************** NOTE TO USERS *********************************
- *
- *                 THIS IS A HEADER FILE, NOT A MANUAL
- *
- *    If you want to know how to use FFTW, please read the manual,
- *    online at http://www.fftw.org/doc/ and also included with FFTW.
- *    For a quick start, see the manual's tutorial section.
- *
- *   (Reading header files to learn how to use a library is a habit
- *    stemming from code lacking a proper manual.  Arguably, it's a
- *    *bad* habit in most cases, because header files can contain
- *    interfaces that are not part of the public, stable API.)
- *
- ****************************************************************************/
-
-#ifndef FFTW3_H
-#define FFTW3_H
-
-#include <stdio.h>
-
-#ifdef __cplusplus
-extern "C"
-{
-#endif /* __cplusplus */
-
-/* If <complex.h> is included, use the C99 complex type.  Otherwise
-   define a type bit-compatible with C99 complex */
-#if !defined(FFTW_NO_Complex) && defined(_Complex_I) && defined(complex) && defined(I)
-#  define FFTW_DEFINE_COMPLEX(R, C) typedef R _Complex C
-#else
-#  define FFTW_DEFINE_COMPLEX(R, C) typedef R C[2]
-#endif
-
-#define FFTW_CONCAT(prefix, name) prefix ## name
-#define FFTW_MANGLE_DOUBLE(name) FFTW_CONCAT(fftw_, name)
-#define FFTW_MANGLE_FLOAT(name) FFTW_CONCAT(fftwf_, name)
-#define FFTW_MANGLE_LONG_DOUBLE(name) FFTW_CONCAT(fftwl_, name)
-#define FFTW_MANGLE_QUAD(name) FFTW_CONCAT(fftwq_, name)
-
-/* IMPORTANT: for Windows compilers, you should add a line
-        #define FFTW_DLL
-   here and in kernel/ifftw.h if you are compiling/using FFTW as a
-   DLL, in order to do the proper importing/exporting, or
-   alternatively compile with -DFFTW_DLL or the equivalent
-   command-line flag.  This is not necessary under MinGW/Cygwin, where
-   libtool does the imports/exports automatically. */
-#if defined(FFTW_DLL) && (defined(_WIN32) || defined(__WIN32__))
-   /* annoying Windows syntax for shared-library declarations */
-#  if defined(COMPILING_FFTW) /* defined in api.h when compiling FFTW */
-#    define FFTW_EXTERN extern __declspec(dllexport) 
-#  else /* user is calling FFTW; import symbol */
-#    define FFTW_EXTERN extern __declspec(dllimport) 
-#  endif
-#else
-#  define FFTW_EXTERN extern
-#endif
-
-enum fftw_r2r_kind_do_not_use_me {
-     FFTW_R2HC=0, FFTW_HC2R=1, FFTW_DHT=2,
-     FFTW_REDFT00=3, FFTW_REDFT01=4, FFTW_REDFT10=5, FFTW_REDFT11=6,
-     FFTW_RODFT00=7, FFTW_RODFT01=8, FFTW_RODFT10=9, FFTW_RODFT11=10
-};
-
-struct fftw_iodim_do_not_use_me {
-     int n;                     /* dimension size */
-     int is;			/* input stride */
-     int os;			/* output stride */
-};
-
-#include <stddef.h> /* for ptrdiff_t */
-struct fftw_iodim64_do_not_use_me {
-     ptrdiff_t n;                     /* dimension size */
-     ptrdiff_t is;			/* input stride */
-     ptrdiff_t os;			/* output stride */
-};
-
-typedef void (*fftw_write_char_func_do_not_use_me)(char c, void *);
-typedef int (*fftw_read_char_func_do_not_use_me)(void *);
-
-/*
-  huge second-order macro that defines prototypes for all API
-  functions.  We expand this macro for each supported precision
- 
-  X: name-mangling macro
-  R: real data type
-  C: complex data type
-*/
-
-#define FFTW_DEFINE_API(X, R, C)					   \
-									   \
-FFTW_DEFINE_COMPLEX(R, C);						   \
-									   \
-typedef struct X(plan_s) *X(plan);					   \
-									   \
-typedef struct fftw_iodim_do_not_use_me X(iodim);			   \
-typedef struct fftw_iodim64_do_not_use_me X(iodim64);			   \
-									   \
-typedef enum fftw_r2r_kind_do_not_use_me X(r2r_kind);			   \
-									   \
-typedef fftw_write_char_func_do_not_use_me X(write_char_func);		   \
-typedef fftw_read_char_func_do_not_use_me X(read_char_func);		   \
-									   \
-FFTW_EXTERN void X(execute)(const X(plan) p);				   \
-									   \
-FFTW_EXTERN X(plan) X(plan_dft)(int rank, const int *n,			   \
-		    C *in, C *out, int sign, unsigned flags);		   \
-									   \
-FFTW_EXTERN X(plan) X(plan_dft_1d)(int n, C *in, C *out, int sign,	   \
-		       unsigned flags);					   \
-FFTW_EXTERN X(plan) X(plan_dft_2d)(int n0, int n1,			   \
-		       C *in, C *out, int sign, unsigned flags);	   \
-FFTW_EXTERN X(plan) X(plan_dft_3d)(int n0, int n1, int n2,		   \
-		       C *in, C *out, int sign, unsigned flags);	   \
-									   \
-FFTW_EXTERN X(plan) X(plan_many_dft)(int rank, const int *n,		   \
-                         int howmany,					   \
-                         C *in, const int *inembed,			   \
-                         int istride, int idist,			   \
-                         C *out, const int *onembed,			   \
-                         int ostride, int odist,			   \
-                         int sign, unsigned flags);			   \
-									   \
-FFTW_EXTERN X(plan) X(plan_guru_dft)(int rank, const X(iodim) *dims,	   \
-			 int howmany_rank,				   \
-			 const X(iodim) *howmany_dims,			   \
-			 C *in, C *out,					   \
-			 int sign, unsigned flags);			   \
-FFTW_EXTERN X(plan) X(plan_guru_split_dft)(int rank, const X(iodim) *dims, \
-			 int howmany_rank,				   \
-			 const X(iodim) *howmany_dims,			   \
-			 R *ri, R *ii, R *ro, R *io,			   \
-			 unsigned flags);				   \
-									   \
-FFTW_EXTERN X(plan) X(plan_guru64_dft)(int rank,			   \
-                         const X(iodim64) *dims,			   \
-			 int howmany_rank,				   \
-			 const X(iodim64) *howmany_dims,		   \
-			 C *in, C *out,					   \
-			 int sign, unsigned flags);			   \
-FFTW_EXTERN X(plan) X(plan_guru64_split_dft)(int rank,			   \
-                         const X(iodim64) *dims,			   \
-			 int howmany_rank,				   \
-			 const X(iodim64) *howmany_dims,		   \
-			 R *ri, R *ii, R *ro, R *io,			   \
-			 unsigned flags);				   \
-									   \
-FFTW_EXTERN void X(execute_dft)(const X(plan) p, C *in, C *out);	   \
-FFTW_EXTERN void X(execute_split_dft)(const X(plan) p, R *ri, R *ii,	   \
-                                      R *ro, R *io);			   \
-									   \
-FFTW_EXTERN X(plan) X(plan_many_dft_r2c)(int rank, const int *n,	   \
-                             int howmany,				   \
-                             R *in, const int *inembed,			   \
-                             int istride, int idist,			   \
-                             C *out, const int *onembed,		   \
-                             int ostride, int odist,			   \
-                             unsigned flags);				   \
-									   \
-FFTW_EXTERN X(plan) X(plan_dft_r2c)(int rank, const int *n,		   \
-                        R *in, C *out, unsigned flags);			   \
-									   \
-FFTW_EXTERN X(plan) X(plan_dft_r2c_1d)(int n,R *in,C *out,unsigned flags); \
-FFTW_EXTERN X(plan) X(plan_dft_r2c_2d)(int n0, int n1,			   \
-			   R *in, C *out, unsigned flags);		   \
-FFTW_EXTERN X(plan) X(plan_dft_r2c_3d)(int n0, int n1,			   \
-			   int n2,					   \
-			   R *in, C *out, unsigned flags);		   \
-									   \
-									   \
-FFTW_EXTERN X(plan) X(plan_many_dft_c2r)(int rank, const int *n,	   \
-			     int howmany,				   \
-			     C *in, const int *inembed,			   \
-			     int istride, int idist,			   \
-			     R *out, const int *onembed,		   \
-			     int ostride, int odist,			   \
-			     unsigned flags);				   \
-									   \
-FFTW_EXTERN X(plan) X(plan_dft_c2r)(int rank, const int *n,		   \
-                        C *in, R *out, unsigned flags);			   \
-									   \
-FFTW_EXTERN X(plan) X(plan_dft_c2r_1d)(int n,C *in,R *out,unsigned flags); \
-FFTW_EXTERN X(plan) X(plan_dft_c2r_2d)(int n0, int n1,			   \
-			   C *in, R *out, unsigned flags);		   \
-FFTW_EXTERN X(plan) X(plan_dft_c2r_3d)(int n0, int n1,			   \
-			   int n2,					   \
-			   C *in, R *out, unsigned flags);		   \
-									   \
-FFTW_EXTERN X(plan) X(plan_guru_dft_r2c)(int rank, const X(iodim) *dims,   \
-			     int howmany_rank,				   \
-			     const X(iodim) *howmany_dims,		   \
-			     R *in, C *out,				   \
-			     unsigned flags);				   \
-FFTW_EXTERN X(plan) X(plan_guru_dft_c2r)(int rank, const X(iodim) *dims,   \
-			     int howmany_rank,				   \
-			     const X(iodim) *howmany_dims,		   \
-			     C *in, R *out,				   \
-			     unsigned flags);				   \
-									   \
-FFTW_EXTERN X(plan) X(plan_guru_split_dft_r2c)(				   \
-                             int rank, const X(iodim) *dims,		   \
-			     int howmany_rank,				   \
-			     const X(iodim) *howmany_dims,		   \
-			     R *in, R *ro, R *io,			   \
-			     unsigned flags);				   \
-FFTW_EXTERN X(plan) X(plan_guru_split_dft_c2r)(				   \
-                             int rank, const X(iodim) *dims,		   \
-			     int howmany_rank,				   \
-			     const X(iodim) *howmany_dims,		   \
-			     R *ri, R *ii, R *out,			   \
-			     unsigned flags);				   \
-									   \
-FFTW_EXTERN X(plan) X(plan_guru64_dft_r2c)(int rank,			   \
-                             const X(iodim64) *dims,			   \
-			     int howmany_rank,				   \
-			     const X(iodim64) *howmany_dims,		   \
-			     R *in, C *out,				   \
-			     unsigned flags);				   \
-FFTW_EXTERN X(plan) X(plan_guru64_dft_c2r)(int rank,			   \
-                             const X(iodim64) *dims,			   \
-			     int howmany_rank,				   \
-			     const X(iodim64) *howmany_dims,		   \
-			     C *in, R *out,				   \
-			     unsigned flags);				   \
-									   \
-FFTW_EXTERN X(plan) X(plan_guru64_split_dft_r2c)(			   \
-                             int rank, const X(iodim64) *dims,		   \
-			     int howmany_rank,				   \
-			     const X(iodim64) *howmany_dims,		   \
-			     R *in, R *ro, R *io,			   \
-			     unsigned flags);				   \
-FFTW_EXTERN X(plan) X(plan_guru64_split_dft_c2r)(			   \
-                             int rank, const X(iodim64) *dims,		   \
-			     int howmany_rank,				   \
-			     const X(iodim64) *howmany_dims,		   \
-			     R *ri, R *ii, R *out,			   \
-			     unsigned flags);				   \
-									   \
-FFTW_EXTERN void X(execute_dft_r2c)(const X(plan) p, R *in, C *out);	   \
-FFTW_EXTERN void X(execute_dft_c2r)(const X(plan) p, C *in, R *out);	   \
-									   \
-FFTW_EXTERN void X(execute_split_dft_r2c)(const X(plan) p,		   \
-                                          R *in, R *ro, R *io);		   \
-FFTW_EXTERN void X(execute_split_dft_c2r)(const X(plan) p,		   \
-                                          R *ri, R *ii, R *out);	   \
-									   \
-FFTW_EXTERN X(plan) X(plan_many_r2r)(int rank, const int *n,		   \
-                         int howmany,					   \
-                         R *in, const int *inembed,			   \
-                         int istride, int idist,			   \
-                         R *out, const int *onembed,			   \
-                         int ostride, int odist,			   \
-                         const X(r2r_kind) *kind, unsigned flags);	   \
-									   \
-FFTW_EXTERN X(plan) X(plan_r2r)(int rank, const int *n, R *in, R *out,	   \
-                    const X(r2r_kind) *kind, unsigned flags);		   \
-									   \
-FFTW_EXTERN X(plan) X(plan_r2r_1d)(int n, R *in, R *out,		   \
-                       X(r2r_kind) kind, unsigned flags);		   \
-FFTW_EXTERN X(plan) X(plan_r2r_2d)(int n0, int n1, R *in, R *out,	   \
-                       X(r2r_kind) kind0, X(r2r_kind) kind1,		   \
-                       unsigned flags);					   \
-FFTW_EXTERN X(plan) X(plan_r2r_3d)(int n0, int n1, int n2,		   \
-                       R *in, R *out, X(r2r_kind) kind0,		   \
-                       X(r2r_kind) kind1, X(r2r_kind) kind2,		   \
-                       unsigned flags);					   \
-									   \
-FFTW_EXTERN X(plan) X(plan_guru_r2r)(int rank, const X(iodim) *dims,	   \
-                         int howmany_rank,				   \
-                         const X(iodim) *howmany_dims,			   \
-                         R *in, R *out,					   \
-                         const X(r2r_kind) *kind, unsigned flags);	   \
-									   \
-FFTW_EXTERN X(plan) X(plan_guru64_r2r)(int rank, const X(iodim64) *dims,   \
-                         int howmany_rank,				   \
-                         const X(iodim64) *howmany_dims,		   \
-                         R *in, R *out,					   \
-                         const X(r2r_kind) *kind, unsigned flags);	   \
-									   \
-FFTW_EXTERN void X(execute_r2r)(const X(plan) p, R *in, R *out);	   \
-									   \
-FFTW_EXTERN void X(destroy_plan)(X(plan) p);				   \
-FFTW_EXTERN void X(forget_wisdom)(void);				   \
-FFTW_EXTERN void X(cleanup)(void);					   \
-									   \
-FFTW_EXTERN void X(set_timelimit)(double t);				   \
-									   \
-FFTW_EXTERN void X(plan_with_nthreads)(int nthreads);			   \
-FFTW_EXTERN int X(init_threads)(void);					   \
-FFTW_EXTERN void X(cleanup_threads)(void);				   \
-									   \
-FFTW_EXTERN int X(export_wisdom_to_filename)(const char *filename);	   \
-FFTW_EXTERN void X(export_wisdom_to_file)(FILE *output_file);		   \
-FFTW_EXTERN char *X(export_wisdom_to_string)(void);			   \
-FFTW_EXTERN void X(export_wisdom)(X(write_char_func) write_char,   	   \
-                                  void *data);				   \
-FFTW_EXTERN int X(import_system_wisdom)(void);				   \
-FFTW_EXTERN int X(import_wisdom_from_filename)(const char *filename);	   \
-FFTW_EXTERN int X(import_wisdom_from_file)(FILE *input_file);		   \
-FFTW_EXTERN int X(import_wisdom_from_string)(const char *input_string);	   \
-FFTW_EXTERN int X(import_wisdom)(X(read_char_func) read_char, void *data); \
-									   \
-FFTW_EXTERN void X(fprint_plan)(const X(plan) p, FILE *output_file);	   \
-FFTW_EXTERN void X(print_plan)(const X(plan) p);			   \
-FFTW_EXTERN char *X(sprint_plan)(const X(plan) p);			   \
-									   \
-FFTW_EXTERN void *X(malloc)(size_t n);					   \
-FFTW_EXTERN R *X(alloc_real)(size_t n);					   \
-FFTW_EXTERN C *X(alloc_complex)(size_t n);				   \
-FFTW_EXTERN void X(free)(void *p);					   \
-									   \
-FFTW_EXTERN void X(flops)(const X(plan) p,				   \
-                          double *add, double *mul, double *fmas);	   \
-FFTW_EXTERN double X(estimate_cost)(const X(plan) p);			   \
-FFTW_EXTERN double X(cost)(const X(plan) p);				   \
-									   \
-FFTW_EXTERN int X(alignment_of)(R *p);                                     \
-FFTW_EXTERN const char X(version)[];                                       \
-FFTW_EXTERN const char X(cc)[];						   \
-FFTW_EXTERN const char X(codelet_optim)[];
-
-
-/* end of FFTW_DEFINE_API macro */
-
-FFTW_DEFINE_API(FFTW_MANGLE_DOUBLE, double, fftw_complex)
-FFTW_DEFINE_API(FFTW_MANGLE_FLOAT, float, fftwf_complex)
-FFTW_DEFINE_API(FFTW_MANGLE_LONG_DOUBLE, long double, fftwl_complex)
-
-/* __float128 (quad precision) is a gcc extension on i386, x86_64, and ia64
-   for gcc >= 4.6 (compiled in FFTW with --enable-quad-precision) */
-#if (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 6)) \
- && !(defined(__ICC) || defined(__INTEL_COMPILER)) \
- && (defined(__i386__) || defined(__x86_64__) || defined(__ia64__))
-#  if !defined(FFTW_NO_Complex) && defined(_Complex_I) && defined(complex) && defined(I)
-/* note: __float128 is a typedef, which is not supported with the _Complex
-         keyword in gcc, so instead we use this ugly __attribute__ version.
-         However, we can't simply pass the __attribute__ version to
-         FFTW_DEFINE_API because the __attribute__ confuses gcc in pointer
-         types.  Hence redefining FFTW_DEFINE_COMPLEX.  Ugh. */
-#    undef FFTW_DEFINE_COMPLEX
-#    define FFTW_DEFINE_COMPLEX(R, C) typedef _Complex float __attribute__((mode(TC))) C
-#  endif
-FFTW_DEFINE_API(FFTW_MANGLE_QUAD, __float128, fftwq_complex)
-#endif
-
-#define FFTW_FORWARD (-1)
-#define FFTW_BACKWARD (+1)
-
-#define FFTW_NO_TIMELIMIT (-1.0)
-
-/* documented flags */
-#define FFTW_MEASURE (0U)
-#define FFTW_DESTROY_INPUT (1U << 0)
-#define FFTW_UNALIGNED (1U << 1)
-#define FFTW_CONSERVE_MEMORY (1U << 2)
-#define FFTW_EXHAUSTIVE (1U << 3) /* NO_EXHAUSTIVE is default */
-#define FFTW_PRESERVE_INPUT (1U << 4) /* cancels FFTW_DESTROY_INPUT */
-#define FFTW_PATIENT (1U << 5) /* IMPATIENT is default */
-#define FFTW_ESTIMATE (1U << 6)
-#define FFTW_WISDOM_ONLY (1U << 21)
-
-/* undocumented beyond-guru flags */
-#define FFTW_ESTIMATE_PATIENT (1U << 7)
-#define FFTW_BELIEVE_PCOST (1U << 8)
-#define FFTW_NO_DFT_R2HC (1U << 9)
-#define FFTW_NO_NONTHREADED (1U << 10)
-#define FFTW_NO_BUFFERING (1U << 11)
-#define FFTW_NO_INDIRECT_OP (1U << 12)
-#define FFTW_ALLOW_LARGE_GENERIC (1U << 13) /* NO_LARGE_GENERIC is default */
-#define FFTW_NO_RANK_SPLITS (1U << 14)
-#define FFTW_NO_VRANK_SPLITS (1U << 15)
-#define FFTW_NO_VRECURSE (1U << 16)
-#define FFTW_NO_SIMD (1U << 17)
-#define FFTW_NO_SLOW (1U << 18)
-#define FFTW_NO_FIXED_RADIX_LARGE_N (1U << 19)
-#define FFTW_ALLOW_PRUNING (1U << 20)
-
-#ifdef __cplusplus
-}  /* extern "C" */
-#endif /* __cplusplus */
-
-#endif /* FFTW3_H */

lib/qcd/action/Actions.h

@@ -57,7 +57,6 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 ////////////////////////////////////////////
 // Gauge Actions
 ////////////////////////////////////////////
-#include <Grid/qcd/action/gauge/Photon.h>
 #include <Grid/qcd/action/gauge/WilsonGaugeAction.h>
 #include <Grid/qcd/action/gauge/PlaqPlusRectangleAction.h>
 
@@ -196,6 +195,7 @@ typedef WilsonTMFermion<WilsonImplD> WilsonTMFermionD;
 typedef DomainWallFermion<WilsonImplR> DomainWallFermionR;
 typedef DomainWallFermion<WilsonImplF> DomainWallFermionF;
 typedef DomainWallFermion<WilsonImplD> DomainWallFermionD;
+
 typedef MobiusFermion<WilsonImplR> MobiusFermionR;
 typedef MobiusFermion<WilsonImplF> MobiusFermionF;
 typedef MobiusFermion<WilsonImplD> MobiusFermionD;
@@ -204,6 +204,20 @@ typedef ZMobiusFermion<ZWilsonImplR> ZMobiusFermionR;
 typedef ZMobiusFermion<ZWilsonImplF> ZMobiusFermionF;
 typedef ZMobiusFermion<ZWilsonImplD> ZMobiusFermionD;
 
+// Ls vectorised 
+typedef DomainWallFermion<DomainWallVec5dImplR> DomainWallFermionVec5dR;
+typedef DomainWallFermion<DomainWallVec5dImplF> DomainWallFermionVec5dF;
+typedef DomainWallFermion<DomainWallVec5dImplD> DomainWallFermionVec5dD;
+
+typedef MobiusFermion<DomainWallVec5dImplR> MobiusFermionVec5dR;
+typedef MobiusFermion<DomainWallVec5dImplF> MobiusFermionVec5dF;
+typedef MobiusFermion<DomainWallVec5dImplD> MobiusFermionVec5dD;
+
+typedef ZMobiusFermion<ZDomainWallVec5dImplR> ZMobiusFermionVec5dR;
+typedef ZMobiusFermion<ZDomainWallVec5dImplF> ZMobiusFermionVec5dF;
+typedef ZMobiusFermion<ZDomainWallVec5dImplD> ZMobiusFermionVec5dD;
+
+
 typedef ScaledShamirFermion<WilsonImplR> ScaledShamirFermionR;
 typedef ScaledShamirFermion<WilsonImplF> ScaledShamirFermionF;
 typedef ScaledShamirFermion<WilsonImplD> ScaledShamirFermionD;
@@ -255,6 +269,7 @@ typedef MobiusFermion<GparityWilsonImplF> GparityMobiusFermionF;
 typedef MobiusFermion<GparityWilsonImplD> GparityMobiusFermionD;
 
 
+
   }}
 ///////////////////////////////////////////////////////////////////////////////
 // G5 herm -- this has to live in QCD since dirac matrix is not in the broader sector of code

lib/qcd/action/fermion/CayleyFermion5D.cc

@@ -62,6 +62,50 @@ void CayleyFermion5D<Impl>::Dminus(const FermionField &psi, FermionField &chi)
     axpby_ssp(chi,Coeff_t(1.0),psi,-cs[s],tmp,s,s);// chi = (1-c[s] D_W) psi
   }
 }
+
+
+template<class Impl> void CayleyFermion5D<Impl>::CayleyReport(void)
+{
+  this->Report();
+  std::vector<int> latt = GridDefaultLatt();          
+  RealD volume = this->Ls;  for(int mu=0;mu<Nd;mu++) volume=volume*latt[mu];
+  RealD NP     = this->_FourDimGrid->_Nprocessors;
+  if ( M5Dcalls > 0 ) {
+    std::cout << GridLogMessage << "#### M5D calls report " << std::endl;
+    std::cout << GridLogMessage << "CayleyFermion5D Number of M5D Calls     : " << M5Dcalls   << std::endl;
+    std::cout << GridLogMessage << "CayleyFermion5D ComputeTime/Calls       : " << M5Dtime / M5Dcalls << " us" << std::endl;
+
+    // Flops = 6.0*(Nc*Ns) *Ls*vol
+    RealD mflops = 6.0*12*volume*M5Dcalls/M5Dtime/2; // 2 for red black counting
+    std::cout << GridLogMessage << "Average mflops/s per call                : " << mflops << std::endl;
+    std::cout << GridLogMessage << "Average mflops/s per call per rank       : " << mflops/NP << std::endl;
+  }
+
+  if ( MooeeInvCalls > 0 ) {
+
+    std::cout << GridLogMessage << "#### MooeeInv calls report " << std::endl;
+    std::cout << GridLogMessage << "CayleyFermion5D Number of MooeeInv Calls     : " << MooeeInvCalls   << std::endl;
+    std::cout << GridLogMessage << "CayleyFermion5D ComputeTime/Calls            : " << MooeeInvTime / MooeeInvCalls << " us" << std::endl;
+
+    // Flops = 9*12*Ls*vol/2
+    RealD mflops = 9.0*12*volume*MooeeInvCalls/MooeeInvTime/2; // 2 for red black counting
+    std::cout << GridLogMessage << "Average mflops/s per call                : " << mflops << std::endl;
+    std::cout << GridLogMessage << "Average mflops/s per call per rank       : " << mflops/NP << std::endl;
+  }
+
+}
+template<class Impl> void CayleyFermion5D<Impl>::CayleyZeroCounters(void)
+{
+  this->ZeroCounters();
+  M5Dflops=0;
+  M5Dcalls=0;
+  M5Dtime=0;
+  MooeeInvFlops=0;
+  MooeeInvCalls=0;
+  MooeeInvTime=0;
+}
+
+
 template<class Impl>  
 void CayleyFermion5D<Impl>::DminusDag(const FermionField &psi, FermionField &chi)
 {

lib/qcd/action/fermion/CayleyFermion5D.h

@@ -120,6 +120,18 @@ namespace Grid {
 		      GridRedBlackCartesian &FourDimRedBlackGrid,
 		      RealD _mass,RealD _M5,const ImplParams &p= ImplParams());
 
+      
+
+     void CayleyReport(void);
+     void CayleyZeroCounters(void);
+
+     double M5Dflops;
+     double M5Dcalls;
+     double M5Dtime;
+
+     double MooeeInvFlops;
+     double MooeeInvCalls;
+     double MooeeInvTime;
 
     protected:
       void SetCoefficientsZolotarev(RealD zolohi,Approx::zolotarev_data *zdata,RealD b,RealD c);

lib/qcd/action/fermion/CayleyFermion5Dcache.cc

@@ -51,6 +51,9 @@ void CayleyFermion5D<Impl>::M5D(const FermionField &psi,
   GridBase *grid=psi._grid;
   assert(phi.checkerboard == psi.checkerboard);
   chi.checkerboard=psi.checkerboard;
+  // Flops = 6.0*(Nc*Ns) *Ls*vol
+  M5Dcalls++;
+  M5Dtime-=usecond();
 PARALLEL_FOR_LOOP
   for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
     for(int s=0;s<Ls;s++){
@@ -76,6 +79,7 @@ PARALLEL_FOR_LOOP
       }
     }
   }
+  M5Dtime+=usecond();
 }
 
 template<class Impl>  
@@ -91,6 +95,9 @@ void CayleyFermion5D<Impl>::M5Ddag(const FermionField &psi,
   assert(phi.checkerboard == psi.checkerboard);
   chi.checkerboard=psi.checkerboard;
 
+  // Flops = 6.0*(Nc*Ns) *Ls*vol
+  M5Dcalls++;
+  M5Dtime-=usecond();
 PARALLEL_FOR_LOOP
   for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
     auto tmp = psi._odata[0];
@@ -116,6 +123,7 @@ PARALLEL_FOR_LOOP
       }
     }
   }
+  M5Dtime+=usecond();
 }
 
 template<class Impl>
@@ -126,10 +134,14 @@ void CayleyFermion5D<Impl>::MooeeInv    (const FermionField &psi, FermionField &
 
   chi.checkerboard=psi.checkerboard;
 
+  MooeeInvCalls++;
+  MooeeInvTime-=usecond();
+
 PARALLEL_FOR_LOOP
   for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
     auto tmp = psi._odata[0];
 
+    // flops = 12*2*Ls + 12*2*Ls + 3*12*Ls + 12*2*Ls  = 12*Ls * (9) = 108*Ls flops
     // Apply (L^{\prime})^{-1}
     chi[ss]=psi[ss]; // chi[0]=psi[0]
     for(int s=1;s<Ls;s++){
@@ -155,6 +167,9 @@ PARALLEL_FOR_LOOP
       chi[ss+s] = chi[ss+s] - uee[s]*tmp;
     }
   }
+
+  MooeeInvTime+=usecond();
+
 }
 
 template<class Impl>
@@ -166,6 +181,8 @@ void CayleyFermion5D<Impl>::MooeeInvDag (const FermionField &psi, FermionField &
   assert(psi.checkerboard == psi.checkerboard);
   chi.checkerboard=psi.checkerboard;
 
+  MooeeInvCalls++;
+  MooeeInvTime-=usecond();
 
 PARALLEL_FOR_LOOP
   for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
@@ -197,6 +214,9 @@ PARALLEL_FOR_LOOP
       chi[ss+s] = chi[ss+s] - lee[s]*tmp;
     }
   }
+
+  MooeeInvTime+=usecond();
+
 }
 
 #ifdef CAYLEY_DPERP_CACHE

lib/qcd/action/fermion/CayleyFermion5Dvec.cc

@@ -60,7 +60,7 @@ void CayleyFermion5D<Impl>::M5D(const FermionField &psi,
   GridBase *grid=psi._grid;
   int Ls   = this->Ls;
   int LLs  = grid->_rdimensions[0];
-  int nsimd= Simd::Nsimd();
+  const int nsimd= Simd::Nsimd();
 
   Vector<iSinglet<Simd> > u(LLs);
   Vector<iSinglet<Simd> > l(LLs);
@@ -86,35 +86,138 @@ void CayleyFermion5D<Impl>::M5D(const FermionField &psi,
     d_p[ss] = diag[s];
   }}
 
+
+  M5Dcalls++;
+  M5Dtime-=usecond();
+
+  assert(Nc==3);
+
 PARALLEL_FOR_LOOP
   for(int ss=0;ss<grid->oSites();ss+=LLs){ // adds LLs
+#if 0
+      alignas(64) SiteHalfSpinor hp;
+      alignas(64) SiteHalfSpinor hm;
+      alignas(64) SiteSpinor fp;
+      alignas(64) SiteSpinor fm;
 
-    alignas(64) SiteHalfSpinor hp;
-    alignas(64) SiteHalfSpinor hm;
-    alignas(64) SiteSpinor fp;
-    alignas(64) SiteSpinor fm;
+      for(int v=0;v<LLs;v++){
 
-    for(int v=0;v<LLs;v++){
+	int vp=(v+1)%LLs;
+	int vm=(v+LLs-1)%LLs;
 
-      int vp=(v+1)%LLs;
-      int vm=(v+LLs-1)%LLs;
+	spProj5m(hp,psi[ss+vp]);
+	spProj5p(hm,psi[ss+vm]);
 
-      spProj5m(hp,psi[ss+vp]);
-      spProj5p(hm,psi[ss+vm]);
-      
-      if ( vp<=v ) rotate(hp,hp,1);
-      if ( vm>=v ) rotate(hm,hm,nsimd-1);
+	if ( vp<=v ) rotate(hp,hp,1);
+	if ( vm>=v ) rotate(hm,hm,nsimd-1);
+	
+	hp=0.5*hp;
+        hm=0.5*hm;
 
-      hp=hp*0.5;
-      hm=hm*0.5;
-      spRecon5m(fp,hp);
-      spRecon5p(fm,hm);
+	spRecon5m(fp,hp);
+	spRecon5p(fm,hm);
 
-      chi[ss+v] = d[v]*phi[ss+v]+u[v]*fp;
-      chi[ss+v] = chi[ss+v]     +l[v]*fm;
+	chi[ss+v] = d[v]*phi[ss+v];
+	chi[ss+v] = chi[ss+v]     +u[v]*fp;
+	chi[ss+v] = chi[ss+v]     +l[v]*fm;
 
-    }
+      }
+#else
+      for(int v=0;v<LLs;v++){
+
+	vprefetch(psi[ss+v+LLs]);
+	//	vprefetch(phi[ss+v+LLs]);
+
+	int vp= (v==LLs-1) ? 0     : v+1;
+	int vm= (v==0    ) ? LLs-1 : v-1;
+	
+	Simd hp_00 = psi[ss+vp]()(2)(0); 
+	Simd hp_01 = psi[ss+vp]()(2)(1); 
+	Simd hp_02 = psi[ss+vp]()(2)(2); 
+	Simd hp_10 = psi[ss+vp]()(3)(0); 
+	Simd hp_11 = psi[ss+vp]()(3)(1); 
+	Simd hp_12 = psi[ss+vp]()(3)(2); 
+	
+	Simd hm_00 = psi[ss+vm]()(0)(0); 
+	Simd hm_01 = psi[ss+vm]()(0)(1); 
+	Simd hm_02 = psi[ss+vm]()(0)(2); 
+	Simd hm_10 = psi[ss+vm]()(1)(0); 
+	Simd hm_11 = psi[ss+vm]()(1)(1); 
+	Simd hm_12 = psi[ss+vm]()(1)(2); 
+
+	//	if ( ss==0) std::cout << " hp_00 " <<hp_00<<std::endl;
+	//	if ( ss==0) std::cout << " hm_00 " <<hm_00<<std::endl;
+
+	if ( vp<=v ) {
+	  hp_00.v = Optimization::Rotate::tRotate<2>(hp_00.v);
+	  hp_01.v = Optimization::Rotate::tRotate<2>(hp_01.v);
+	  hp_02.v = Optimization::Rotate::tRotate<2>(hp_02.v);
+	  hp_10.v = Optimization::Rotate::tRotate<2>(hp_10.v);
+	  hp_11.v = Optimization::Rotate::tRotate<2>(hp_11.v);
+	  hp_12.v = Optimization::Rotate::tRotate<2>(hp_12.v);
+	}
+	if ( vm>=v ) {
+	  hm_00.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_00.v);
+	  hm_01.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_01.v);
+	  hm_02.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_02.v);
+	  hm_10.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_10.v);
+	  hm_11.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_11.v);
+	  hm_12.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_12.v);
+	}
+
+	/*
+	if ( ss==0) std::cout << " dphi_00 " <<d[v]()()() * phi[ss+v]()(0)(0) <<std::endl;
+	if ( ss==0) std::cout << " dphi_10 " <<d[v]()()() * phi[ss+v]()(1)(0) <<std::endl;
+	if ( ss==0) std::cout << " dphi_20 " <<d[v]()()() * phi[ss+v]()(2)(0) <<std::endl;
+	if ( ss==0) std::cout << " dphi_30 " <<d[v]()()() * phi[ss+v]()(3)(0) <<std::endl;
+	*/	
+	Simd p_00  = d[v]()()() * phi[ss+v]()(0)(0)  + l[v]()()()*hm_00; 
+	Simd p_01  = d[v]()()() * phi[ss+v]()(0)(1)  + l[v]()()()*hm_01; 
+	Simd p_02  = d[v]()()() * phi[ss+v]()(0)(2)  + l[v]()()()*hm_02; 
+	Simd p_10  = d[v]()()() * phi[ss+v]()(1)(0)  + l[v]()()()*hm_10; 
+	Simd p_11  = d[v]()()() * phi[ss+v]()(1)(1)  + l[v]()()()*hm_11; 
+	Simd p_12  = d[v]()()() * phi[ss+v]()(1)(2)  + l[v]()()()*hm_12; 
+	Simd p_20  = d[v]()()() * phi[ss+v]()(2)(0)  + u[v]()()()*hp_00; 
+	Simd p_21  = d[v]()()() * phi[ss+v]()(2)(1)  + u[v]()()()*hp_01; 
+	Simd p_22  = d[v]()()() * phi[ss+v]()(2)(2)  + u[v]()()()*hp_02;  
+	Simd p_30  = d[v]()()() * phi[ss+v]()(3)(0)  + u[v]()()()*hp_10; 
+	Simd p_31  = d[v]()()() * phi[ss+v]()(3)(1)  + u[v]()()()*hp_11; 
+	Simd p_32  = d[v]()()() * phi[ss+v]()(3)(2)  + u[v]()()()*hp_12; 
+
+	
+	//	if ( ss==0){
+	/*
+	std::cout << ss<<" "<< v<< " good "<< chi[ss+v]()(0)(0) << " bad "<<p_00<<" diff "<<chi[ss+v]()(0)(0)-p_00<<std::endl;
+	std::cout << ss<<" "<< v<< " good "<< chi[ss+v]()(0)(1) << " bad "<<p_01<<" diff "<<chi[ss+v]()(0)(1)-p_01<<std::endl;
+	std::cout << ss<<" "<< v<< " good "<< chi[ss+v]()(0)(2) << " bad "<<p_02<<" diff "<<chi[ss+v]()(0)(2)-p_02<<std::endl;
+	std::cout << ss<<" "<< v<< " good "<< chi[ss+v]()(1)(0) << " bad "<<p_10<<" diff "<<chi[ss+v]()(1)(0)-p_10<<std::endl;
+	std::cout << ss<<" "<< v<< " good "<< chi[ss+v]()(1)(1) << " bad "<<p_11<<" diff "<<chi[ss+v]()(1)(1)-p_11<<std::endl;
+	std::cout << ss<<" "<< v<< " good "<< chi[ss+v]()(1)(2) << " bad "<<p_12<<" diff "<<chi[ss+v]()(1)(2)-p_12<<std::endl;
+	std::cout << ss<<" "<< v<< " good "<< chi[ss+v]()(2)(0) << " bad "<<p_20<<" diff "<<chi[ss+v]()(2)(0)-p_20<<std::endl;
+	std::cout << ss<<" "<< v<< " good "<< chi[ss+v]()(2)(1) << " bad "<<p_21<<" diff "<<chi[ss+v]()(2)(1)-p_21<<std::endl;
+	std::cout << ss<<" "<< v<< " good "<< chi[ss+v]()(2)(2) << " bad "<<p_22<<" diff "<<chi[ss+v]()(2)(2)-p_22<<std::endl;
+	std::cout << ss<<" "<< v<< " good "<< chi[ss+v]()(3)(0) << " bad "<<p_30<<" diff "<<chi[ss+v]()(3)(0)-p_30<<std::endl;
+	std::cout << ss<<" "<< v<< " good "<< chi[ss+v]()(3)(1) << " bad "<<p_31<<" diff "<<chi[ss+v]()(3)(1)-p_31<<std::endl;
+	std::cout << ss<<" "<< v<< " good "<< chi[ss+v]()(3)(2) << " bad "<<p_32<<" diff "<<chi[ss+v]()(3)(2)-p_32<<std::endl;
+	}
+	*/
+	vstream(chi[ss+v]()(0)(0),p_00);
+	vstream(chi[ss+v]()(0)(1),p_01);
+	vstream(chi[ss+v]()(0)(2),p_02);
+	vstream(chi[ss+v]()(1)(0),p_10);
+	vstream(chi[ss+v]()(1)(1),p_11);
+	vstream(chi[ss+v]()(1)(2),p_12);
+	vstream(chi[ss+v]()(2)(0),p_20);
+	vstream(chi[ss+v]()(2)(1),p_21);
+	vstream(chi[ss+v]()(2)(2),p_22);
+	vstream(chi[ss+v]()(3)(0),p_30);
+	vstream(chi[ss+v]()(3)(1),p_31);
+	vstream(chi[ss+v]()(3)(2),p_32);
+
+      }
+#endif
   }
+  M5Dtime+=usecond();
 }
 
 template<class Impl>  
@@ -154,6 +257,8 @@ void CayleyFermion5D<Impl>::M5Ddag(const FermionField &psi,
     d_p[ss] = diag[s];
   }}
 
+  M5Dcalls++;
+  M5Dtime-=usecond();
 PARALLEL_FOR_LOOP
   for(int ss=0;ss<grid->oSites();ss+=LLs){ // adds LLs
 
@@ -183,8 +288,8 @@ PARALLEL_FOR_LOOP
 
     }
   }
+  M5Dtime+=usecond();
 }
-
 template<class Impl>
 void CayleyFermion5D<Impl>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv)
 {
@@ -250,13 +355,11 @@ void CayleyFermion5D<Impl>::MooeeInternal(const FermionField &psi, FermionField
     }
   }
   
+  MooeeInvCalls++;
+  MooeeInvTime-=usecond();
   // Dynamic allocate on stack to get per thread without serialised heap acces
-PARALLEL_FOR_LOOP
-  for(auto site=0;site<vol;site++){
-    
-    //    SiteHalfSpinor *SitePplus =(SiteHalfSpinor *) alloca(LLs*sizeof(SiteHalfSpinor));
-    //    SiteHalfSpinor *SitePminus=(SiteHalfSpinor *) alloca(LLs*sizeof(SiteHalfSpinor));
-    //    SiteSpinor     *SiteChi   =(SiteSpinor *)     alloca(LLs*sizeof(SiteSpinor));
+#pragma omp parallel  
+  {
 
     Vector<SiteHalfSpinor> SitePplus(LLs);
     Vector<SiteHalfSpinor> SitePminus(LLs);
@@ -267,6 +370,9 @@ PARALLEL_FOR_LOOP
     SiteHalfSpinor BcastP;
     SiteHalfSpinor BcastM;
 
+#pragma omp for 
+  for(auto site=0;site<vol;site++){
+
     for(int s=0;s<LLs;s++){
       int lex = s+LLs*site;
       spProj5p(SitePplus[s] ,psi[lex]);
@@ -294,6 +400,8 @@ PARALLEL_FOR_LOOP
       chi[lex] = SiteChi[s]*0.5;
     }
   }
+  }
+  MooeeInvTime+=usecond();
 }
 
 INSTANTIATE_DPERP(DomainWallVec5dImplD);

lib/qcd/action/fermion/FermionOperatorImpl.h

@@ -48,8 +48,10 @@ namespace QCD {
   //    typedef typename XXX         GaugeField;
   //    typedef typename XXX      GaugeActField;
   //    typedef typename XXX       FermionField;
+  //    typedef typename XXX    PropagatorField;
   //    typedef typename XXX  DoubledGaugeField;
   //    typedef typename XXX         SiteSpinor;
+  //    typedef typename XXX     SitePropagator;
   //    typedef typename XXX     SiteHalfSpinor;	
   //    typedef typename XXX         Compressor;	
   //
@@ -95,13 +97,15 @@ namespace QCD {
   
 #define INHERIT_FIMPL_TYPES(Impl)\
   typedef typename Impl::FermionField           FermionField;		\
+  typedef typename Impl::PropagatorField     PropagatorField;		\
   typedef typename Impl::DoubledGaugeField DoubledGaugeField;		\
   typedef typename Impl::SiteSpinor               SiteSpinor;		\
+  typedef typename Impl::SitePropagator       SitePropagator;		\
   typedef typename Impl::SiteHalfSpinor       SiteHalfSpinor;		\
   typedef typename Impl::Compressor               Compressor;		\
   typedef typename Impl::StencilImpl             StencilImpl;		\
-  typedef typename Impl::ImplParams ImplParams;				\
-  typedef typename Impl::Coeff_t       Coeff_t;
+  typedef typename Impl::ImplParams               ImplParams;	        \
+  typedef typename Impl::Coeff_t                     Coeff_t;           \
   
 #define INHERIT_IMPL_TYPES(Base) \
   INHERIT_GIMPL_TYPES(Base)	 \
@@ -127,14 +131,17 @@ namespace QCD {
     INHERIT_GIMPL_TYPES(Gimpl);
       
     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 iImplDoubledGaugeField = iVector<iScalar<iMatrix<vtype, Dimension> >, Nds>;
     
     typedef iImplSpinor<Simd>            SiteSpinor;
+    typedef iImplPropagator<Simd>        SitePropagator;
     typedef iImplHalfSpinor<Simd>        SiteHalfSpinor;
     typedef iImplDoubledGaugeField<Simd> SiteDoubledGaugeField;
     
     typedef Lattice<SiteSpinor>            FermionField;
+    typedef Lattice<SitePropagator>        PropagatorField;
     typedef Lattice<SiteDoubledGaugeField> DoubledGaugeField;
     
     typedef WilsonCompressor<SiteHalfSpinor, SiteSpinor> Compressor;
@@ -216,14 +223,17 @@ class DomainWallVec5dImpl :  public PeriodicGaugeImpl< GaugeImplTypes< S,Nrepres
   INHERIT_GIMPL_TYPES(Gimpl);
   
   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 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> > >;
   
   typedef iImplSpinor<Simd> SiteSpinor;
+  typedef iImplPropagator<Simd> SitePropagator;
   typedef iImplHalfSpinor<Simd> SiteHalfSpinor;
   typedef Lattice<SiteSpinor> FermionField;
+  typedef Lattice<SitePropagator> PropagatorField;
   
   // Make the doubled gauge field a *scalar*
   typedef iImplDoubledGaugeField<typename Simd::scalar_type>  SiteDoubledGaugeField;  // This is a scalar
@@ -315,14 +325,17 @@ class GparityWilsonImpl : public ConjugateGaugeImpl<GaugeImplTypes<S, Nrepresent
  INHERIT_GIMPL_TYPES(Gimpl);
       
  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 iImplHalfSpinor        = iVector<iVector<iVector<vtype, Nrepresentation>, Nhs>, 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 iImplDoubledGaugeField<Simd> SiteDoubledGaugeField;
  
  typedef Lattice<SiteSpinor> FermionField;
+ typedef Lattice<SitePropagator> PropagatorField;
  typedef Lattice<SiteDoubledGaugeField> DoubledGaugeField;
  
  typedef WilsonCompressor<SiteHalfSpinor, SiteSpinor> Compressor;

lib/qcd/action/fermion/WilsonFermion5D.cc

@@ -194,6 +194,11 @@ void WilsonFermion5D<Impl>::Report(void)
     std::cout << GridLogMessage << "Average mflops/s per call                : " << mflops << std::endl;
     std::cout << GridLogMessage << "Average mflops/s per call per rank       : " << mflops/NP << std::endl;
 
+    RealD Fullmflops = 1344*volume*DhopCalls/(DhopComputeTime+DhopCommTime)/2; // 2 for red black counting
+    std::cout << GridLogMessage << "Average mflops/s per call (full)         : " << Fullmflops << std::endl;
+    std::cout << GridLogMessage << "Average mflops/s per call per rank (full): " << Fullmflops/NP << std::endl;
+
+
    }
 
   if ( DerivCalls > 0 ) {
@@ -209,12 +214,15 @@ void WilsonFermion5D<Impl>::Report(void)
     RealD mflops = 144*volume*DerivCalls/DerivDhopComputeTime;
     std::cout << GridLogMessage << "Average mflops/s per call                : " << mflops << std::endl;
     std::cout << GridLogMessage << "Average mflops/s per call per node       : " << mflops/NP << std::endl;
-  }
+
+    RealD Fullmflops = 144*volume*DerivCalls/(DerivDhopComputeTime+DerivCommTime)/2; // 2 for red black counting
+    std::cout << GridLogMessage << "Average mflops/s per call (full)         : " << Fullmflops << std::endl;
+    std::cout << GridLogMessage << "Average mflops/s per call per node (full): " << Fullmflops/NP << std::endl;  }
 
   if (DerivCalls > 0 || DhopCalls > 0){
-    std::cout << GridLogMessage << "WilsonFermion5D Stencil"<<std::endl;  Stencil.Report();
+    std::cout << GridLogMessage << "WilsonFermion5D Stencil"    <<std::endl;  Stencil.Report();
     std::cout << GridLogMessage << "WilsonFermion5D StencilEven"<<std::endl;  StencilEven.Report();
-    std::cout << GridLogMessage << "WilsonFermion5D StencilOdd"<<std::endl;  StencilOdd.Report();
+    std::cout << GridLogMessage << "WilsonFermion5D StencilOdd" <<std::endl;  StencilOdd.Report();
   }
 }
 

lib/qcd/action/fermion/WilsonKernels.h

@@ -61,14 +61,8 @@ public:
     switch(Opt) {
 #ifdef AVX512
     case OptInlineAsm:
-      for (int site = 0; site < Ns; site++) {
-	for (int s = 0; s < Ls; s++) {
-	  WilsonKernels<Impl>::DiracOptAsmDhopSite(st,lo,U,buf,sF,sU,Ls,Ns,in,out);
-	  sF++;
-	}
-	sU++;
-      }
-      break;
+       WilsonKernels<Impl>::DiracOptAsmDhopSite(st,lo,U,buf,sF,sU,Ls,Ns,in,out);
+       break;
 #endif
     case OptHandUnroll:
       for (int site = 0; site < Ns; site++) {
@@ -115,13 +109,7 @@ public:
     switch(Opt) {
 #ifdef AVX512
     case OptInlineAsm:
-      for (int site = 0; site < Ns; site++) {
-	for (int s = 0; s < Ls; s++) {
-	  WilsonKernels<Impl>::DiracOptAsmDhopSiteDag(st,lo,U,buf,sF,sU,Ls,Ns,in,out);
-	  sF++;
-	}
-	sU++;
-      }
+      WilsonKernels<Impl>::DiracOptAsmDhopSiteDag(st,lo,U,buf,sF,sU,Ls,Ns,in,out);
       break;
 #endif
     case OptHandUnroll:

lib/qcd/action/fermion/WilsonKernelsAsm.cc

@@ -10,6 +10,7 @@
 
 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
@@ -53,24 +54,26 @@ WilsonKernels<Impl >::DiracOptAsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,
 }
 
 #if defined(AVX512) 
-    
+#include <simd/Intel512wilson.h>
+
     ///////////////////////////////////////////////////////////
     // If we are AVX512 specialise the single precision routine
     ///////////////////////////////////////////////////////////
-    
-#include <simd/Intel512wilson.h>
+
 #include <simd/Intel512single.h>
     
-static Vector<vComplexF> signs;
-    
-  int setupSigns(void ){
-    Vector<vComplexF> bother(2);
+static Vector<vComplexF> signsF;
+
+  template<typename vtype>    
+  int setupSigns(Vector<vtype>& signs ){
+    Vector<vtype> bother(2);
     signs = bother;
     vrsign(signs[0]);
     visign(signs[1]);
     return 1;
   }
-  static int signInit = setupSigns();
+
+  static int signInitF = setupSigns(signsF);
   
 #define label(A)  ilabel(A)
 #define ilabel(A) ".globl\n"  #A ":\n" 
@@ -78,6 +81,8 @@ static Vector<vComplexF> signs;
 #define MAYBEPERM(A,perm) if (perm) { A ; }
 #define MULT_2SPIN(ptr,pf) MULT_ADDSUB_2SPIN(ptr,pf)
 #define FX(A) WILSONASM_ ##A
+#define COMPLEX_TYPE vComplexF
+#define signs signsF
   
 #undef KERNEL_DAG
 template<> void 
@@ -98,8 +103,8 @@ WilsonKernels<WilsonImplF>::DiracOptAsmDhopSiteDag(StencilImpl &st,LebesgueOrder
 #undef FX 
 #define FX(A) DWFASM_ ## A
 #define MAYBEPERM(A,B) 
-#define VMOVIDUP(A,B,C)                                  VBCASTIDUPf(A,B,C)
-#define VMOVRDUP(A,B,C)                                  VBCASTRDUPf(A,B,C)
+//#define VMOVIDUP(A,B,C)                                  VBCASTIDUPf(A,B,C)
+//#define VMOVRDUP(A,B,C)                                  VBCASTRDUPf(A,B,C)
 #define MULT_2SPIN(ptr,pf) MULT_ADDSUB_2SPIN_LS(ptr,pf)
 				    
 #undef KERNEL_DAG
@@ -113,8 +118,71 @@ template<> void
 WilsonKernels<DomainWallVec5dImplF>::DiracOptAsmDhopSiteDag(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_TYPE
+#undef signs
+#undef VMOVRDUP
+#undef MAYBEPERM
+#undef MULT_2SPIN
+#undef FX 
+	
+///////////////////////////////////////////////////////////
+// If we are AVX512 specialise the double precision routine
+///////////////////////////////////////////////////////////
+
+#include <simd/Intel512double.h>
+    
+static Vector<vComplexD> signsD;
+#define signs signsD
+static int signInitD = setupSigns(signsD);
+    
+#define MAYBEPERM(A,perm) if (perm) { A ; }
+#define MULT_2SPIN(ptr,pf) MULT_ADDSUB_2SPIN(ptr,pf)
+#define FX(A) WILSONASM_ ##A
+#define COMPLEX_TYPE vComplexD
+  
+#undef KERNEL_DAG
+template<> void 
+WilsonKernels<WilsonImplD>::DiracOptAsmDhopSite(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>
+      
+#define KERNEL_DAG
+template<> void 
+WilsonKernels<WilsonImplD>::DiracOptAsmDhopSiteDag(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>
 				    
-#endif
+#undef VMOVIDUP
+#undef VMOVRDUP
+#undef MAYBEPERM
+#undef MULT_2SPIN
+#undef FX 
+#define FX(A) DWFASM_ ## A
+#define MAYBEPERM(A,B) 
+//#define VMOVIDUP(A,B,C)                                  VBCASTIDUPd(A,B,C)
+//#define VMOVRDUP(A,B,C)                                  VBCASTRDUPd(A,B,C)
+#define MULT_2SPIN(ptr,pf) MULT_ADDSUB_2SPIN_LS(ptr,pf)
+				    
+#undef KERNEL_DAG
+template<> void 
+WilsonKernels<DomainWallVec5dImplD>::DiracOptAsmDhopSite(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>
+				    
+#define KERNEL_DAG
+template<> void 
+WilsonKernels<DomainWallVec5dImplD>::DiracOptAsmDhopSiteDag(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_TYPE
+#undef signs
+#undef VMOVRDUP
+#undef MAYBEPERM
+#undef MULT_2SPIN
+#undef FX 
+
+#endif //AVX512
 
 #define INSTANTIATE_ASM(A)\
 template void WilsonKernels<A>::DiracOptAsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,\

lib/qcd/action/fermion/WilsonKernelsAsmBody.h

@@ -5,7 +5,9 @@
   const uint64_t plocal =(uint64_t) & in._odata[0];
 
   //  vComplexF isigns[2] = { signs[0], signs[1] };
-  vComplexF *isigns = &signs[0];
+  //COMPLEX_TYPE is vComplexF of vComplexD depending 
+  //on the chosen precision
+  COMPLEX_TYPE *isigns = &signs[0];
 
   MASK_REGS;
   int nmax=U._grid->oSites();

lib/qcd/hmc/HmcRunner.h

@@ -116,7 +116,7 @@ class NerscHmcRunnerTemplate {
     NoSmearing<Gimpl> SmearingPolicy;
     typedef MinimumNorm2<GaugeField, NoSmearing<Gimpl>, RepresentationsPolicy >
         IntegratorType;  // change here to change the algorithm
-    IntegratorParameters MDpar(20, 1.0);
+    IntegratorParameters MDpar(40, 1.0);
     IntegratorType MDynamics(UGrid, MDpar, TheAction, SmearingPolicy);
 
     // Checkpoint strategy

lib/serialisation/BaseIO.h

@@ -67,9 +67,8 @@ namespace Grid {
     return os;
   }
   
-  class Serializable {};
-  
   // static polymorphism implemented using CRTP idiom
+  class Serializable;
   
   // Static abstract writer
   template <typename T>
@@ -122,6 +121,27 @@ namespace Grid {
     T *upcast;
   };
   
+  // serializable base class
+  class Serializable
+  {
+  public:
+    template <typename T>
+    static inline void write(Writer<T> &WR,const std::string &s,
+                             const Serializable &obj)
+    {}
+    
+    template <typename T>
+    static inline void read(Reader<T> &RD,const std::string &s,
+                            Serializable &obj)
+    {}
+    
+    friend inline std::ostream & operator<<(std::ostream &os,
+                                            const Serializable &obj)
+    {
+      return os;
+    }
+  };
+  
   // Generic writer interface
   template <typename T>
   inline void push(Writer<T> &w, const std::string &s)

lib/simd/Grid_avx.h

@@ -167,7 +167,7 @@ namespace Optimization {
     }
     //Integer
     inline __m256i operator()(__m256i a, __m256i b){
-#if defined (AVX1) || defined (AVXFMA4)
+#if defined (AVX1) || defined (AVXFMA) || defined (AVXFMA4)
           __m128i a0,a1;
           __m128i b0,b1;
           a0 = _mm256_extractf128_si256(a,0);
@@ -195,7 +195,7 @@ namespace Optimization {
     }
     //Integer
     inline __m256i operator()(__m256i a, __m256i b){
-#if defined (AVX1) || defined (AVXFMA4)
+#if defined (AVX1) || defined (AVXFMA) || defined (AVXFMA4)
           __m128i a0,a1;
           __m128i b0,b1;
           a0 = _mm256_extractf128_si256(a,0);
@@ -216,7 +216,7 @@ namespace Optimization {
   struct MultComplex{
     // Complex float
     inline __m256 operator()(__m256 a, __m256 b){
-#if defined (AVX1) 
+#if defined (AVX1)
       __m256 ymm0,ymm1,ymm2;
       ymm0 = _mm256_shuffle_ps(a,a,_MM_SELECT_FOUR_FOUR(2,2,0,0)); // ymm0 <- ar ar,
       ymm0 = _mm256_mul_ps(ymm0,b);                       // ymm0 <- ar bi, ar br
@@ -233,7 +233,7 @@ namespace Optimization {
       a_imag = _mm256_mul_ps( a_imag,tmp  );  // (Ai, Ai) * (Bi, Br) = Ai Bi, Ai Br
       return _mm256_maddsub_ps( a_real, b, a_imag ); // Ar Br , Ar Bi   +- Ai Bi             = ArBr-AiBi , ArBi+AiBr
 #endif
-#if defined (AVX2)
+#if defined (AVX2)  || defined (AVXFMA)
       __m256 a_real = _mm256_moveldup_ps( a ); // Ar Ar
       __m256 a_imag = _mm256_movehdup_ps( a ); // Ai Ai
       a_imag = _mm256_mul_ps( a_imag, _mm256_shuffle_ps( b,b, _MM_SELECT_FOUR_FOUR(2,3,0,1) ));  // (Ai, Ai) * (Bi, Br) = Ai Bi, Ai Br
@@ -264,7 +264,7 @@ namespace Optimization {
 	IF IMM0[3] = 0
 	THEN DEST[255:192]=SRC2[191:128] ELSE DEST[255:192]=SRC2[255:192] FI; // Ox5 r<->i   ; 0xC unchanged
       */
-#if defined (AVX1) 
+#if defined (AVX1)
       __m256d ymm0,ymm1,ymm2;
       ymm0 = _mm256_shuffle_pd(a,a,0x0); // ymm0 <- ar ar, ar,ar b'00,00
       ymm0 = _mm256_mul_pd(ymm0,b);      // ymm0 <- ar bi, ar br
@@ -279,7 +279,7 @@ namespace Optimization {
       a_imag = _mm256_mul_pd( a_imag, _mm256_permute_pd( b, 0x5 ) );  // (Ai, Ai) * (Bi, Br) = Ai Bi, Ai Br
       return _mm256_maddsub_pd( a_real, b, a_imag ); // Ar Br , Ar Bi   +- Ai Bi             = ArBr-AiBi , ArBi+AiBr
 #endif
-#if defined (AVX2)
+#if defined (AVX2) || defined (AVXFMA)
       __m256d a_real = _mm256_movedup_pd( a ); // Ar Ar
       __m256d a_imag = _mm256_shuffle_pd(a,a,0xF);//aiai
       a_imag = _mm256_mul_pd( a_imag, _mm256_permute_pd( b, 0x5 ) );  // (Ai, Ai) * (Bi, Br) = Ai Bi, Ai Br
@@ -320,7 +320,7 @@ namespace Optimization {
 #if defined (AVXFMA4)
       a= _mm256_macc_ps(b,c,a);
 #endif
-#if defined (AVX2)
+#if defined (AVX2) || defined (AVXFMA)
       a= _mm256_fmadd_ps( b, c, a);
 #endif
     }
@@ -332,7 +332,7 @@ namespace Optimization {
 #if defined (AVXFMA4)
       a= _mm256_macc_pd(b,c,a);
 #endif
-#if defined (AVX2)
+#if defined (AVX2) || defined (AVXFMA)
       a= _mm256_fmadd_pd( b, c, a);
 #endif
     }
@@ -347,7 +347,7 @@ namespace Optimization {
     }
     // Integer
     inline __m256i operator()(__m256i a, __m256i b){
-#if defined (AVX1) 
+#if defined (AVX1) || defined (AVXFMA)
       __m128i a0,a1;
       __m128i b0,b1;
       a0 = _mm256_extractf128_si256(a,0);

lib/simd/Grid_avx512.h

@@ -27,15 +27,6 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
     See the full license in the file "LICENSE" in the top level distribution directory
     *************************************************************************************/
     /*  END LEGAL */
-//----------------------------------------------------------------------
-/*! @file Grid_knc.h
-  @brief Optimization libraries for AVX512 instructions set for KNC
-
-  Using intrinsics
-*/
-// Time-stamp: <2015-06-09 14:27:28 neo>
-//----------------------------------------------------------------------
-
 #include <immintrin.h>
 
 
@@ -95,13 +86,13 @@ namespace Optimization {
   struct Vstream{
     //Float
     inline void operator()(float * a, __m512 b){
-      //_mm512_stream_ps(a,b);
-      _mm512_store_ps(a,b);
+      _mm512_stream_ps(a,b);
+      //      _mm512_store_ps(a,b);
     }
     //Double
     inline void operator()(double * a, __m512d b){
-      //_mm512_stream_pd(a,b);
-      _mm512_store_pd(a,b);
+      _mm512_stream_pd(a,b);
+      //      _mm512_store_pd(a,b);
     }
 
   };
@@ -382,7 +373,6 @@ namespace Optimization {
   // Some Template specialization
 
   // Hack for CLANG until mm512_reduce_add_ps etc... are implemented in GCC and Clang releases
-
 #ifndef __INTEL_COMPILER
 #warning "Slow reduction due to incomplete reduce intrinsics"
   //Complex float Reduce

lib/simd/Grid_generic.h

@@ -6,8 +6,7 @@
 
     Copyright (C) 2015
 
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: neo <cossu@post.kek.jp>
+Author: Antonin Portelli <antonin.portelli@me.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
@@ -27,133 +26,352 @@ Author: neo <cossu@post.kek.jp>
     *************************************************************************************/
     /*  END LEGAL */
 
+static_assert(GEN_SIMD_WIDTH % 16u == 0, "SIMD vector size is not an integer multiple of 16 bytes");
+
+//#define VECTOR_LOOPS
+
+// playing with compiler pragmas
+#ifdef VECTOR_LOOPS
+#ifdef __clang__
+#define VECTOR_FOR(i, w, inc)\
+_Pragma("clang loop unroll(full) vectorize(enable) interleave(enable) vectorize_width(w)")\
+for (unsigned int i = 0; i < w; i += inc)
+#elif defined __INTEL_COMPILER
+#define VECTOR_FOR(i, w, inc)\
+_Pragma("simd vectorlength(w*8)")\
+for (unsigned int i = 0; i < w; i += inc)
+#else
+#define VECTOR_FOR(i, w, inc)\
+for (unsigned int i = 0; i < w; i += inc)
+#endif
+#else
+#define VECTOR_FOR(i, w, inc)\
+for (unsigned int i = 0; i < w; i += inc)
+#endif
+
 namespace Grid {
 namespace Optimization {
 
-  template<class vtype>
-  union uconv {
-    float f;
-    vtype v;
+  // type traits giving the number of elements for each vector type
+  template <typename T> struct W;
+  template <> struct W<double> {
+    constexpr static unsigned int c = GEN_SIMD_WIDTH/16u;
+    constexpr static unsigned int r = GEN_SIMD_WIDTH/8u;
   };
-
-  union u128f {
-    float v;
-    float f[4];
-  };
-  union u128d {
-    double v;
-    double f[2];
+  template <> struct W<float> {
+    constexpr static unsigned int c = GEN_SIMD_WIDTH/8u;
+    constexpr static unsigned int r = GEN_SIMD_WIDTH/4u;
   };
   
+  // SIMD vector types
+  template <typename T>
+  struct vec {
+    alignas(GEN_SIMD_WIDTH) T v[W<T>::r];
+  };
+  
+  typedef vec<float>   vecf;
+  typedef vec<double>  vecd;
+  
   struct Vsplat{
-    //Complex float
-    inline u128f operator()(float a, float b){
-      u128f out; 
-      out.f[0] = a;
-      out.f[1] = b;
-      out.f[2] = a;
-      out.f[3] = b;
+    // Complex
+    template <typename T>
+    inline vec<T> operator()(T a, T b){
+      vec<T> out;
+      
+      VECTOR_FOR(i, W<T>::r, 2)
+      {
+        out.v[i]   = a;
+        out.v[i+1] = b;
+      }
+
       return out;
     }
-    // Real float
-    inline u128f operator()(float a){
-      u128f out; 
-      out.f[0] = a;
-      out.f[1] = a;
-      out.f[2] = a;
-      out.f[3] = a;
+    
+    // Real
+    template <typename T>
+    inline vec<T> operator()(T a){
+      vec<T> out;
+      
+      VECTOR_FOR(i, W<T>::r, 1)
+      {
+        out.v[i] = a;
+      }
+      
       return out;
     }
-    //Complex double
-    inline u128d operator()(double a, double b){
-      u128d out; 
-      out.f[0] = a;
-      out.f[1] = b;
-      return out;
-    }
-    //Real double
-    inline u128d operator()(double a){
-      u128d out; 
-      out.f[0] = a;
-      out.f[1] = a;
-      return out;
-    }
-    //Integer
+    
+    // Integer
     inline int operator()(Integer a){
       return a;
     }
   };
 
   struct Vstore{
-    //Float 
-    inline void operator()(u128f a, float* F){
-      memcpy(F,a.f,4*sizeof(float));
-    }
-    //Double
-    inline void operator()(u128d a, double* D){
-      memcpy(D,a.f,2*sizeof(double));
+    // Real
+    template <typename T>
+    inline void operator()(vec<T> a, T *D){
+      *((vec<T> *)D) = a;
     }
     //Integer
-    inline void operator()(int a, Integer* I){
-      I[0] = a;
+    inline void operator()(int a, Integer *I){
+      *I = a;
     }
 
   };
 
   struct Vstream{
-    //Float
-    inline void operator()(float * a, u128f b){
-      memcpy(a,b.f,4*sizeof(float));
+    // Real
+    template <typename T>
+    inline void operator()(T * a, vec<T> b){
+      *((vec<T> *)a) = b;
     }
-    //Double
-    inline void operator()(double * a, u128d b){
-      memcpy(a,b.f,2*sizeof(double));
-    }
-
-
   };
 
   struct Vset{
-    // Complex float 
-    inline u128f operator()(Grid::ComplexF *a){
-      u128f out; 
-      out.f[0] = a[0].real();
-      out.f[1] = a[0].imag();
-      out.f[2] = a[1].real();
-      out.f[3] = a[1].imag();
+    // Complex
+    template <typename T>
+    inline vec<T> operator()(std::complex<T> *a){
+      vec<T> out;
+      
+      VECTOR_FOR(i, W<T>::c, 1)
+      {
+        out.v[2*i]   = a[i].real();
+        out.v[2*i+1] = a[i].imag();
+      }
+      
       return out;
     }
-    // Complex double 
-    inline u128d operator()(Grid::ComplexD *a){
-      u128d out; 
-      out.f[0] = a[0].real();
-      out.f[1] = a[0].imag();
-      return out;
-    }
-    // Real float 
-    inline u128f operator()(float *a){
-      u128f out; 
-      out.f[0] = a[0];
-      out.f[1] = a[1];
-      out.f[2] = a[2];
-      out.f[3] = a[3];
-      return out;
-    }
-    // Real double
-    inline u128d operator()(double *a){
-      u128d out; 
-      out.f[0] = a[0];
-      out.f[1] = a[1];
+    
+    // Real
+    template <typename T>
+    inline vec<T> operator()(T *a){
+      vec<T> out;
+      
+      out = *((vec<T> *)a);
+      
       return out;
     }
+
     // Integer
     inline int operator()(Integer *a){
-      return a[0];
+      return *a;
     }
-
-
   };
 
+  /////////////////////////////////////////////////////
+  // Arithmetic operations
+  /////////////////////////////////////////////////////
+  struct Sum{
+    // Complex/Real
+    template <typename T>
+    inline vec<T> operator()(vec<T> a, vec<T> b){
+      vec<T> out;
+      
+      VECTOR_FOR(i, W<T>::r, 1)
+      {
+        out.v[i] = a.v[i] + b.v[i];
+      }
+      
+      return out;
+    }
+    
+    //I nteger
+    inline int operator()(int a, int b){
+      return a + b;
+    }
+  };
+
+  struct Sub{
+    // Complex/Real
+    template <typename T>
+    inline vec<T> operator()(vec<T> a, vec<T> b){
+      vec<T> out;
+      
+      VECTOR_FOR(i, W<T>::r, 1)
+      {
+        out.v[i] = a.v[i] - b.v[i];
+      }
+      
+      return out;
+    }
+    
+    //Integer
+    inline int operator()(int a, int b){
+      return a-b;
+    }
+  };
+
+  struct Mult{
+    // Real
+    template <typename T>
+    inline vec<T> operator()(vec<T> a, vec<T> b){
+      vec<T> out;
+      
+      VECTOR_FOR(i, W<T>::r, 1)
+      {
+        out.v[i] = a.v[i]*b.v[i];
+      }
+      
+      return out;
+    }
+    
+    // Integer
+    inline int operator()(int a, int b){
+      return a*b;
+    }
+  };
+  
+  #define cmul(a, b, c, i)\
+  c[i]   = a[i]*b[i]   - a[i+1]*b[i+1];\
+  c[i+1] = a[i]*b[i+1] + a[i+1]*b[i];
+  
+  struct MultComplex{
+    // Complex
+    template <typename T>
+    inline vec<T> operator()(vec<T> a, vec<T> b){
+      vec<T> out;
+      
+      VECTOR_FOR(i, W<T>::c, 1)
+      {
+        cmul(a.v, b.v, out.v, 2*i);
+      }      
+      
+      return out;
+    }
+  };
+  
+  #undef cmul
+
+  struct Div{
+    // Real
+    template <typename T>
+    inline vec<T> operator()(vec<T> a, vec<T> b){
+      vec<T> out;
+      
+      VECTOR_FOR(i, W<T>::r, 1)
+      {
+        out.v[i] = a.v[i]/b.v[i];
+      }
+      
+      return out;
+    }
+  };
+  
+  #define conj(a, b, i)\
+  b[i]   = a[i];\
+  b[i+1] = -a[i+1];
+  
+  struct Conj{
+    // Complex
+    template <typename T>
+    inline vec<T> operator()(vec<T> a){
+      vec<T> out;
+      
+      VECTOR_FOR(i, W<T>::c, 1)
+      {
+        conj(a.v, out.v, 2*i);
+      }
+      
+      return out;
+    }
+  };
+  
+  #undef conj
+
+  #define timesmi(a, b, i)\
+  b[i]   = a[i+1];\
+  b[i+1] = -a[i];
+  
+  struct TimesMinusI{
+    // Complex
+    template <typename T>
+    inline vec<T> operator()(vec<T> a, vec<T> b){
+      vec<T> out;
+      
+      VECTOR_FOR(i, W<T>::c, 1)
+      {
+        timesmi(a.v, out.v, 2*i);
+      }
+      
+      return out;
+    }
+  };
+
+  #undef timesmi
+  
+  #define timesi(a, b, i)\
+  b[i]   = -a[i+1];\
+  b[i+1] = a[i];
+  
+  struct TimesI{
+    // Complex
+    template <typename T>
+    inline vec<T> operator()(vec<T> a, vec<T> b){
+      vec<T> out;
+      
+      VECTOR_FOR(i, W<T>::c, 1)
+      {
+        timesi(a.v, out.v, 2*i);
+      }
+      
+      return out;
+    }
+  };
+  
+  #undef timesi
+
+  //////////////////////////////////////////////
+  // Some Template specialization
+  #define perm(a, b, n, w)\
+  unsigned int _mask = w >> (n + 1);\
+  VECTOR_FOR(i, w, 1)\
+  {\
+    b[i] = a[i^_mask];\
+  }
+  
+  #define DECL_PERMUTE_N(n)\
+  template <typename T>\
+  static inline vec<T> Permute##n(vec<T> in) {\
+    vec<T> out;\
+    perm(in.v, out.v, n, W<T>::r);\
+    return out;\
+  }
+  
+  struct Permute{
+    DECL_PERMUTE_N(0);
+    DECL_PERMUTE_N(1);
+    DECL_PERMUTE_N(2);
+    DECL_PERMUTE_N(3);
+  };
+  
+  #undef perm
+  #undef DECL_PERMUTE_N
+  
+  #define rot(a, b, n, w)\
+  VECTOR_FOR(i, w, 1)\
+  {\
+    b[i] = a[(i + n)%w];\
+  }
+  
+  struct Rotate{
+    template <typename T>
+    static inline vec<T> rotate(vec<T> in, int n){
+      vec<T> out;
+      
+      rot(in.v, out.v, n, W<T>::r);
+      
+      return out;
+    }
+  };
+
+  #undef rot
+  
+  #define acc(v, a, off, step, n)\
+  for (unsigned int i = off; i < n; i += step)\
+  {\
+    a += v[i];\
+  }
+  
   template <typename Out_type, typename In_type>
   struct Reduce{
     //Need templated class to overload output type
@@ -164,316 +382,67 @@ namespace Optimization {
       return 0;
     }
   };
-
-  /////////////////////////////////////////////////////
-  // Arithmetic operations
-  /////////////////////////////////////////////////////
-  struct Sum{
-    //Complex/Real float
-    inline u128f operator()(u128f a, u128f b){
-      u128f out;
-      out.f[0] = a.f[0] + b.f[0];
-      out.f[1] = a.f[1] + b.f[1];
-      out.f[2] = a.f[2] + b.f[2];
-      out.f[3] = a.f[3] + b.f[3];
-      return out;
-    }
-    //Complex/Real double
-    inline u128d operator()(u128d a, u128d b){
-      u128d out;
-      out.f[0] = a.f[0] + b.f[0];
-      out.f[1] = a.f[1] + b.f[1];
-      return out;
-    }
-    //Integer
-    inline int operator()(int a, int b){
-      return a + b;
-    }
-  };
-
-  struct Sub{
-    //Complex/Real float
-    inline u128f operator()(u128f a, u128f b){
-      u128f out;
-      out.f[0] = a.f[0] - b.f[0];
-      out.f[1] = a.f[1] - b.f[1];
-      out.f[2] = a.f[2] - b.f[2];
-      out.f[3] = a.f[3] - b.f[3];
-      return out;
-    }
-    //Complex/Real double
-    inline u128d operator()(u128d a, u128d b){
-      u128d out;
-      out.f[0] = a.f[0] - b.f[0];
-      out.f[1] = a.f[1] - b.f[1];
-      return out;
-    }
-    //Integer
-    inline int operator()(int a, int b){
-      return a-b;
-    }
-  };
-
-  struct MultComplex{
-    // Complex float
-    inline u128f operator()(u128f a, u128f b){
-      u128f out;
-      out.f[0] = a.f[0]*b.f[0] - a.f[1]*b.f[1];
-      out.f[1] = a.f[0]*b.f[1] + a.f[1]*b.f[0];
-      out.f[2] = a.f[2]*b.f[2] - a.f[3]*b.f[3];
-      out.f[3] = a.f[2]*b.f[3] + a.f[3]*b.f[2];
-      return out;
-    }
-    // Complex double
-    inline u128d operator()(u128d a, u128d b){
-      u128d out;
-      out.f[0] = a.f[0]*b.f[0] - a.f[1]*b.f[1];
-      out.f[1] = a.f[0]*b.f[1] + a.f[1]*b.f[0];
-      return out;
-    }
-  };
-
-  struct Mult{
-    //CK: Appear unneeded
-    // inline float  mac(float a, float b,double c){
-    //   return 0;
-    // }
-    // inline double mac(double a, double b,double c){
-    //   return 0;
-    // }
-
-    // Real float
-    inline u128f operator()(u128f a, u128f b){
-      u128f out;
-      out.f[0] = a.f[0]*b.f[0];
-      out.f[1] = a.f[1]*b.f[1];
-      out.f[2] = a.f[2]*b.f[2];
-      out.f[3] = a.f[3]*b.f[3];
-      return out;
-    }
-    // Real double
-    inline u128d operator()(u128d a, u128d b){
-      u128d out;
-      out.f[0] = a.f[0]*b.f[0];
-      out.f[1] = a.f[1]*b.f[1];
-      return out;
-    }
-    // Integer
-    inline int operator()(int a, int b){
-      return a*b;
-    }
-  };
-
-  struct Conj{
-    // Complex single
-    inline u128f operator()(u128f in){
-      u128f out;
-      out.f[0] = in.f[0];
-      out.f[1] = -in.f[1];
-      out.f[2] = in.f[2];
-      out.f[3] = -in.f[3];
-      return out;
-    }
-    // Complex double
-    inline u128d operator()(u128d in){
-      u128d out;
-      out.f[0] = in.f[0];
-      out.f[1] = -in.f[1];
-      return out;
-    }
-    // do not define for integer input
-  };
-
-  struct TimesMinusI{
-    //Complex single
-    inline u128f operator()(u128f in, u128f ret){ //note ret is ignored
-      u128f out;
-      out.f[0] = in.f[1];
-      out.f[1] = -in.f[0];
-      out.f[2] = in.f[3];
-      out.f[3] = -in.f[2];
-      return out;
-    }
-    //Complex double
-    inline u128d operator()(u128d in, u128d ret){
-      u128d out;
-      out.f[0] = in.f[1];
-      out.f[1] = -in.f[0];
-      return out;
-    }
-  };
-
-  struct TimesI{
-    //Complex single
-    inline u128f operator()(u128f in, u128f ret){ //note ret is ignored
-      u128f out;
-      out.f[0] = -in.f[1];
-      out.f[1] = in.f[0];
-      out.f[2] = -in.f[3];
-      out.f[3] = in.f[2];
-      return out;
-    }
-    //Complex double
-    inline u128d operator()(u128d in, u128d ret){
-      u128d out;
-      out.f[0] = -in.f[1];
-      out.f[1] = in.f[0];
-      return out;
-    }
-  };
-
-  //////////////////////////////////////////////
-  // Some Template specialization
-  struct Permute{
-    //We just have to mirror the permutes of Grid_sse4.h
-    static inline u128f Permute0(u128f in){ //AB CD -> CD AB
-      u128f out;
-      out.f[0] = in.f[2];
-      out.f[1] = in.f[3];
-      out.f[2] = in.f[0];
-      out.f[3] = in.f[1];
-      return out;
-    };
-    static inline u128f Permute1(u128f in){ //AB CD -> BA DC
-      u128f out;
-      out.f[0] = in.f[1];
-      out.f[1] = in.f[0];
-      out.f[2] = in.f[3];
-      out.f[3] = in.f[2];
-      return out;
-    };
-    static inline u128f Permute2(u128f in){
-      return in;
-    };
-    static inline u128f Permute3(u128f in){
-      return in;
-    };
-
-    static inline u128d Permute0(u128d in){ //AB -> BA
-      u128d out;
-      out.f[0] = in.f[1];
-      out.f[1] = in.f[0];
-      return out;      
-    };
-    static inline u128d Permute1(u128d in){
-      return in;
-    };
-    static inline u128d Permute2(u128d in){
-      return in;
-    };
-    static inline u128d Permute3(u128d in){
-      return in;
-    };
-
-  };
   
-  template < typename vtype > 
-    void permute(vtype &a, vtype b, int perm) {
-   };
-    
-  struct Rotate{
-
-    static inline u128f rotate(u128f in,int n){
-      u128f out;
-      switch(n){
-      case 0:
-        out.f[0] = in.f[0];
-        out.f[1] = in.f[1];
-        out.f[2] = in.f[2];
-        out.f[3] = in.f[3];
-        break;
-      case 1:
-        out.f[0] = in.f[1];
-        out.f[1] = in.f[2];
-        out.f[2] = in.f[3];
-        out.f[3] = in.f[0];
-        break;
-      case 2:
-        out.f[0] = in.f[2];
-        out.f[1] = in.f[3];
-        out.f[2] = in.f[0];
-        out.f[3] = in.f[1];
-        break;
-      case 3:
-        out.f[0] = in.f[3];
-        out.f[1] = in.f[0];
-        out.f[2] = in.f[1];
-        out.f[3] = in.f[2];
-        break;
-      default: assert(0);
-      }
-      return out;
-    }
-    static inline u128d rotate(u128d in,int n){
-      u128d out;
-      switch(n){
-      case 0:
-        out.f[0] = in.f[0];
-        out.f[1] = in.f[1];
-        break;
-      case 1:
-        out.f[0] = in.f[1];
-        out.f[1] = in.f[0];
-        break;
-      default: assert(0);
-      }
-      return out;
-    }
-  };
-
   //Complex float Reduce
-  template<>
-  inline Grid::ComplexF Reduce<Grid::ComplexF, u128f>::operator()(u128f in){ //2 complex
-    return Grid::ComplexF(in.f[0] + in.f[2], in.f[1] + in.f[3]);
+  template <>
+  inline Grid::ComplexF Reduce<Grid::ComplexF, vecf>::operator()(vecf in){
+    float a = 0.f, b = 0.f;
+    
+    acc(in.v, a, 0, 2, W<float>::r);
+    acc(in.v, b, 1, 2, W<float>::r);
+    
+    return Grid::ComplexF(a, b);
   }
+  
   //Real float Reduce
   template<>
-  inline Grid::RealF Reduce<Grid::RealF, u128f>::operator()(u128f in){ //4 floats
-    return in.f[0] + in.f[1] + in.f[2] + in.f[3];
+  inline Grid::RealF Reduce<Grid::RealF, vecf>::operator()(vecf in){
+    float a = 0.;
+    
+    acc(in.v, a, 0, 1, W<float>::r);
+    
+    return a;
   }
   
-  
   //Complex double Reduce
   template<>
-  inline Grid::ComplexD Reduce<Grid::ComplexD, u128d>::operator()(u128d in){ //1 complex
-    return Grid::ComplexD(in.f[0],in.f[1]);
+  inline Grid::ComplexD Reduce<Grid::ComplexD, vecd>::operator()(vecd in){
+    double a = 0., b = 0.;
+    
+    acc(in.v, a, 0, 2, W<double>::r);
+    acc(in.v, b, 1, 2, W<double>::r);
+    
+    return Grid::ComplexD(a, b);
   }
   
   //Real double Reduce
   template<>
-  inline Grid::RealD Reduce<Grid::RealD, u128d>::operator()(u128d in){ //2 doubles
-    return in.f[0] + in.f[1];
+  inline Grid::RealD Reduce<Grid::RealD, vecd>::operator()(vecd in){
+    double a = 0.f;
+    
+    acc(in.v, a, 0, 1, W<double>::r);
+    
+    return a;
   }
 
   //Integer Reduce
   template<>
   inline Integer Reduce<Integer, int>::operator()(int in){
-    // FIXME unimplemented
-   printf("Reduce : Missing integer implementation -> FIX\n");
-    assert(0);
+    return in;
   }
 }
 
 //////////////////////////////////////////////////////////////////////////////////////
 // Here assign types 
 
-  typedef Optimization::u128f SIMD_Ftype;  // Single precision type
-  typedef Optimization::u128d SIMD_Dtype; // Double precision type
+  typedef Optimization::vecf SIMD_Ftype; // Single precision type
+  typedef Optimization::vecd SIMD_Dtype; // Double precision type
   typedef int SIMD_Itype; // Integer type
 
   // prefetch utilities
   inline void v_prefetch0(int size, const char *ptr){};
   inline void prefetch_HINT_T0(const char *ptr){};
 
-
-
-  // Gpermute function
-  template < typename VectorSIMD > 
-    inline void Gpermute(VectorSIMD &y,const VectorSIMD &b, int perm ) {
-    Optimization::permute(y.v,b.v,perm);
-  }
-
-
   // Function name aliases
   typedef Optimization::Vsplat   VsplatSIMD;
   typedef Optimization::Vstore   VstoreSIMD;
@@ -481,16 +450,13 @@ namespace Optimization {
   typedef Optimization::Vstream  VstreamSIMD;
   template <typename S, typename T> using ReduceSIMD = Optimization::Reduce<S,T>;
 
- 
-
-
   // Arithmetic operations
   typedef Optimization::Sum         SumSIMD;
   typedef Optimization::Sub         SubSIMD;
+  typedef Optimization::Div         DivSIMD;
   typedef Optimization::Mult        MultSIMD;
   typedef Optimization::MultComplex MultComplexSIMD;
   typedef Optimization::Conj        ConjSIMD;
   typedef Optimization::TimesMinusI TimesMinusISIMD;
   typedef Optimization::TimesI      TimesISIMD;
-
 }

lib/simd/Grid_imci.h

@@ -26,14 +26,6 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
     See the full license in the file "LICENSE" in the top level distribution directory
     *************************************************************************************/
     /*  END LEGAL */
-//----------------------------------------------------------------------
-/*! @file Grid_knc.h
-  @brief Optimization libraries for AVX512 instructions set for KNC
-
-  Using intrinsics
-*/
-// Time-stamp: <2015-06-09 14:27:28 neo>
-//----------------------------------------------------------------------
 
 #include <immintrin.h>
 #include <zmmintrin.h>

lib/simd/Grid_qpx.h

@@ -244,7 +244,22 @@ namespace Optimization {
       return a*b;
     }
   };
-  
+
+  struct Div{
+    // Real double
+    inline vector4double operator()(vector4double a, vector4double b){
+      return vec_swdiv(a, b);
+    }
+
+    // Real float
+    FLOAT_WRAP_2(operator(), inline)
+
+    // Integer
+    inline int operator()(int a, int b){
+      return a/b;
+    }
+  };
+
   struct Conj{
     // Complex double
     inline vector4double operator()(vector4double v){
@@ -413,6 +428,7 @@ template <typename S, typename T> using ReduceSIMD = Optimization::Reduce<S,T>;
 typedef Optimization::Sum         SumSIMD;
 typedef Optimization::Sub         SubSIMD;
 typedef Optimization::Mult        MultSIMD;
+typedef Optimization::Div         DivSIMD;
 typedef Optimization::MultComplex MultComplexSIMD;
 typedef Optimization::Conj        ConjSIMD;
 typedef Optimization::TimesMinusI TimesMinusISIMD;

lib/simd/Grid_vector_types.h

@@ -38,13 +38,13 @@ directory
 #ifndef GRID_VECTOR_TYPES
 #define GRID_VECTOR_TYPES
 
-#ifdef GENERIC_VEC
+#ifdef GEN
 #include "Grid_generic.h"
 #endif
 #ifdef SSE4
 #include "Grid_sse4.h"
 #endif
-#if defined(AVX1) || defined(AVX2) || defined(AVXFMA4)
+#if defined(AVX1) || defined (AVXFMA) || defined(AVX2) || defined(AVXFMA4)
 #include "Grid_avx.h"
 #endif
 #if defined AVX512
@@ -130,7 +130,7 @@ class Grid_simd {
 
   Vector_type v;
 
-  static inline int Nsimd(void) {
+  static inline constexpr int Nsimd(void) {
     return sizeof(Vector_type) / sizeof(Scalar_type);
   }
 

lib/stencil/Lebesgue.cc

@@ -65,7 +65,7 @@ void LebesgueOrder::CartesianBlocking(void)
 {
   _LebesgueReorder.resize(0);
 
-  std::cout << GridLogDebug << " CartesianBlocking ";
+  //    std::cout << GridLogDebug << " CartesianBlocking ";
   //    for(int d=0;d<Block.size();d++) std::cout <<Block[d]<<" ";
   //    std::cout<<std::endl; 
 

m4/ac_prog_doxygen.m4

@@ -1,54 +0,0 @@
-dnl Check for doxygen to create API docs
-dnl
-AC_DEFUN([AC_PROG_DOXYGEN],
-[
-AC_ARG_ENABLE(doxygen,
-AS_HELP_STRING([--enable-doxygen],[enable documentation generation with doxygen (auto)]))
-
-AC_ARG_ENABLE(dot,
-AS_HELP_STRING([--enable-dot],[use 'dot' to generate graphs in doxygen (auto)]))              
-AC_ARG_ENABLE(html-docs,
-AS_HELP_STRING([--enable-html-docs],[enable HTML generation with doxygen (yes)]),
-[],[ enable_html_docs=yes])              
-AC_ARG_ENABLE(latex-docs,
-AS_HELP_STRING([--enable-latex-docs],
-[enable LaTeX documentation generation with doxygen (no)]),[],[enable_latex_docs=no])              
-
-if test "x$enable_doxygen" = xno; then
- enable_doc=no	
-else 
- AC_CHECK_PROG(DOXYGEN, doxygen, doxygen)
- if test x$DOXYGEN = x; then
-  if test "x$enable_doxygen" = xyes; then
-   AC_MSG_ERROR([could not find doxygen])
-  fi
-  enable_doc=no
- else
-  doxy_ver=`doxygen --version`
-  doxy_major=`expr "$doxy_ver" : '\(@<:@0-9@:>@\)\..*'`
-  doxy_minor=`expr "$doxy_ver" : '@<:@0-9@:>@\.\(@<:@0-9@:>@\).*'`
-  if test $doxy_major -eq "1" -a $doxy_minor -ge "3" ; then
-   enable_doc=yes
-   AC_CHECK_PROG(DOT, dot, dot)
-  else
-   AC_MSG_WARN([doxygen version $doxy_ver too old, doxygen will not be used.])
-   enable_doc=no
-  fi
- fi
-fi
-
-AM_CONDITIONAL(DOXYGEN_DOC, test x$enable_doc = xyes)
-
-if test x$DOT = x; then
- if test "x$enable_dot" = xyes; then
-  AC_MSG_ERROR([could not find dot])
- fi
- enable_dot=no
-else
- enable_dot=yes
-fi
-
-AC_SUBST(enable_dot)
-AC_SUBST(enable_html_docs)
-AC_SUBST(enable_latex_docs)
-])

scripts/filelist

@@ -20,15 +20,21 @@ for subdir in $dirs; do
     TESTS=`ls T*.cc`
     TESTLIST=`echo ${TESTS} | sed s/.cc//g `
     PREF=`[ $subdir = '.' ] && echo noinst || echo EXTRA`
-    echo "tests: ${TESTLIST}" > Make.inc
+    SUB=`[ $subdir = '.' ] && echo subtests`
+    echo "tests: ${TESTLIST} ${SUB}" > Make.inc
     echo ${PREF}_PROGRAMS = ${TESTLIST} >> Make.inc
     echo >> Make.inc
+    HADLINK=`[ $subdir = './hadrons' ] && echo '-lHadrons '`
     for f in $TESTS; do
-	BNAME=`basename $f .cc`
-	echo ${BNAME}_SOURCES=$f  >> Make.inc
-	echo ${BNAME}_LDADD=-lGrid>> Make.inc
-	echo >> Make.inc
+	   BNAME=`basename $f .cc`
+	   echo ${BNAME}_SOURCES=$f >> Make.inc
+	   echo ${BNAME}_LDADD=${HADLINK}-lGrid  >> Make.inc
+	   echo >> Make.inc
     done
+    if [ $subdir != '.' ]; then
+       echo CLEANFILES = ${TESTLIST} >> Make.inc
+       echo >> Make.inc
+    fi
 done
 
 # benchmarks Make.inc

tests/Makefile.am

@@ -4,4 +4,9 @@ if BUILD_CHROMA_REGRESSION
   SUBDIRS+= qdpxx
 endif
 
+.PHONY: subtests
+
 include Make.inc
+
+subtests:
+	for d in $(SUBDIRS); do $(MAKE) -C $${d} tests; done

tests/Test_simd.cc

@@ -50,6 +50,12 @@ public:
   template<class vec> void operator()(vec &rr,vec &i1,vec &i2) const { rr = i1*i2;}
   std::string name(void) const { return std::string("Times"); }
 };
+class funcDivide {
+public:
+  funcDivide() {};
+  template<class vec> void operator()(vec &rr,vec &i1,vec &i2) const { rr = i1/i2;}
+  std::string name(void) const { return std::string("Divide"); }
+};
 class funcConj {
 public:
   funcConj() {};
@@ -341,6 +347,7 @@ int main (int argc, char ** argv)
   Tester<RealF,vRealF>(funcPlus());
   Tester<RealF,vRealF>(funcMinus());
   Tester<RealF,vRealF>(funcTimes());
+  Tester<RealF,vRealF>(funcDivide());
   Tester<RealF,vRealF>(funcAdj());
   Tester<RealF,vRealF>(funcConj());
   Tester<RealF,vRealF>(funcInnerProduct());
@@ -371,6 +378,7 @@ int main (int argc, char ** argv)
   Tester<RealD,vRealD>(funcPlus());
   Tester<RealD,vRealD>(funcMinus());
   Tester<RealD,vRealD>(funcTimes());
+  Tester<RealD,vRealD>(funcDivide());
   Tester<RealD,vRealD>(funcAdj());
   Tester<RealD,vRealD>(funcConj());
   Tester<RealD,vRealD>(funcInnerProduct());