Checking in working version of Lanczos.

Prepending PACKAGE_ with GRID_ in Config.h

.gitignore

@@ -9,6 +9,7 @@
 ################
 *~
 *#
+*.sublime-*
 
 # Precompiled Headers #
 #######################
@@ -103,4 +104,16 @@ lib/fftw/*
 # libtool macros #
 ##################
 m4/lt*
-m4/libtool.m4
+m4/libtool.m4
+
+# Buck files #
+##############
+.buck*
+buck-out
+BUCK
+make-bin-BUCK.sh
+
+# generated sources #
+#####################
+lib/qcd/spin/gamma-gen/*.h
+lib/qcd/spin/gamma-gen/*.cc

Makefile.am

@@ -1,5 +1,5 @@
 # additional include paths necessary to compile the C++ library
-SUBDIRS = lib benchmarks tests
+SUBDIRS = lib benchmarks tests extras
 
 include $(top_srcdir)/doxygen.inc
 

benchmarks/Benchmark_dwf.cc

@@ -37,11 +37,11 @@ struct scal {
   d internal;
 };
 
-  Gamma::GammaMatrix Gmu [] = {
-    Gamma::GammaX,
-    Gamma::GammaY,
-    Gamma::GammaZ,
-    Gamma::GammaT
+  Gamma::Algebra Gmu [] = {
+    Gamma::Algebra::GammaX,
+    Gamma::Algebra::GammaY,
+    Gamma::Algebra::GammaZ,
+    Gamma::Algebra::GammaT
   };
 
 typedef WilsonFermion5D<DomainWallVec5dImplR> WilsonFermion5DR;
@@ -321,7 +321,7 @@ int main (int argc, char ** argv)
     ref = zero;
     for(int mu=0;mu<Nd;mu++){
 
-      //    ref =  src - Gamma(Gamma::GammaX)* src ; // 1+gamma_x
+      //    ref =  src - Gamma(Gamma::Algebra::GammaX)* src ; // 1+gamma_x
       tmp = U[mu]*Cshift(src,mu+1,1);
       for(int i=0;i<ref._odata.size();i++){
   ref._odata[i]+= tmp._odata[i] + Gamma(Gmu[mu])*tmp._odata[i]; ;

benchmarks/Benchmark_dwf_sweep.cc

@@ -37,11 +37,11 @@ struct scal {
   d internal;
 };
 
-  Gamma::GammaMatrix Gmu [] = {
-    Gamma::GammaX,
-    Gamma::GammaY,
-    Gamma::GammaZ,
-    Gamma::GammaT
+  Gamma::Algebra Gmu [] = {
+    Gamma::Algebra::GammaX,
+    Gamma::Algebra::GammaY,
+    Gamma::Algebra::GammaZ,
+    Gamma::Algebra::GammaT
   };
 
 void benchDw(std::vector<int> & L, int Ls, int threads, int report =0 );

benchmarks/Benchmark_mooee.cc

@@ -113,6 +113,36 @@ int main (int argc, char ** argv)
     std::cout<<GridLogMessage << "Called " #A " "<< (t1-t0)/ncall<<" us"<<std::endl;\
     std::cout<<GridLogMessage << "******************"<<std::endl;
 
+#define BENCH_ZDW(A,in,out)			\
+    zDw.CayleyZeroCounters();			\
+    zDw. A (in,out);				\
+    FGrid->Barrier();				\
+    t0=usecond();				\
+    for(int i=0;i<ncall;i++){			\
+      zDw. A (in,out);				\
+    }						\
+    t1=usecond();				\
+    FGrid->Barrier();				\
+    zDw.CayleyReport();							\
+    std::cout<<GridLogMessage << "Called ZDw " #A " "<< (t1-t0)/ncall<<" us"<<std::endl;\
+    std::cout<<GridLogMessage << "******************"<<std::endl;
+
+#define BENCH_DW_SSC(A,in,out)			\
+    Dw.CayleyZeroCounters();			\
+    Dw. A (in,out);				\
+    FGrid->Barrier();				\
+    t0=usecond();				\
+    for(int i=0;i<ncall;i++){			\
+      __SSC_START ;				\
+      Dw. A (in,out);				\
+      __SSC_STOP ;				\
+    }						\
+    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);				\
@@ -148,9 +178,15 @@ int main (int argc, char ** argv)
     LatticeFermion sref(sFGrid);
     LatticeFermion result(sFGrid);
 
+
     std::cout<<GridLogMessage << "Constructing Vec5D Dw "<<std::endl;
     DomainWallFermionVec5dR Dw(Umu,*sFGrid,*sFrbGrid,*sUGrid,*sUrbGrid,mass,M5);
 
+    RealD b=1.5;// Scale factor b+c=2, b-c=1
+    RealD c=0.5;
+    std::vector<ComplexD> gamma(Ls,std::complex<double>(1.0,0.0));
+    ZMobiusFermionVec5dR zDw(Umu,*sFGrid,*sFrbGrid,*sUGrid,*sUrbGrid,mass,M5,gamma,b,c);
+
     std::cout<<GridLogMessage << "Calling Dhop "<<std::endl;
     FGrid->Barrier();
 
@@ -173,10 +209,13 @@ int main (int argc, char ** argv)
 
     BENCH_DW_MEO(Dhop    ,src,result);
     BENCH_DW_MEO(DhopEO  ,src_o,r_e);
-    BENCH_DW(Meooe   ,src_o,r_e);
+    BENCH_DW_SSC(Meooe   ,src_o,r_e);
     BENCH_DW(Mooee   ,src_o,r_o);
     BENCH_DW(MooeeInv,src_o,r_o);
 
+    BENCH_ZDW(Mooee   ,src_o,r_o);
+    BENCH_ZDW(MooeeInv,src_o,r_o);
+
   }
 
   Grid_finalize();

benchmarks/Benchmark_wilson.cc

@@ -37,11 +37,11 @@ struct scal {
   d internal;
 };
 
-  Gamma::GammaMatrix Gmu [] = {
-    Gamma::GammaX,
-    Gamma::GammaY,
-    Gamma::GammaZ,
-    Gamma::GammaT
+  Gamma::Algebra Gmu [] = {
+    Gamma::Algebra::GammaX,
+    Gamma::Algebra::GammaY,
+    Gamma::Algebra::GammaZ,
+    Gamma::Algebra::GammaT
   };
 
 bool overlapComms = false;
@@ -106,7 +106,7 @@ int main (int argc, char ** argv)
   { // Naive wilson implementation
     ref = zero;
     for(int mu=0;mu<Nd;mu++){
-      //    ref =  src + Gamma(Gamma::GammaX)* src ; // 1-gamma_x
+      //    ref =  src + Gamma(Gamma::Algebra::GammaX)* src ; // 1-gamma_x
       tmp = U[mu]*Cshift(src,mu,1);
       for(int i=0;i<ref._odata.size();i++){
 	ref._odata[i]+= tmp._odata[i] - Gamma(Gmu[mu])*tmp._odata[i]; ;
@@ -159,7 +159,7 @@ int main (int argc, char ** argv)
     ref = zero;
     for(int mu=0;mu<Nd;mu++){
 
-      //    ref =  src - Gamma(Gamma::GammaX)* src ; // 1+gamma_x
+      //    ref =  src - Gamma(Gamma::Algebra::GammaX)* src ; // 1+gamma_x
       tmp = U[mu]*Cshift(src,mu,1);
       for(int i=0;i<ref._odata.size();i++){
 	ref._odata[i]+= tmp._odata[i] + Gamma(Gmu[mu])*tmp._odata[i]; ;

benchmarks/Benchmark_wilson_sweep.cc

@@ -30,11 +30,11 @@ struct scal {
   d internal;
 };
 
-Gamma::GammaMatrix Gmu [] = {
-  Gamma::GammaX,
-  Gamma::GammaY,
-  Gamma::GammaZ,
-  Gamma::GammaT
+Gamma::Algebra Gmu [] = {
+    Gamma::Algebra::GammaX,
+    Gamma::Algebra::GammaY,
+    Gamma::Algebra::GammaZ,
+    Gamma::Algebra::GammaT
 };
 
 bool overlapComms = false;

configure.ac

@@ -6,7 +6,7 @@ AC_CANONICAL_TARGET
 AM_INIT_AUTOMAKE(subdir-objects)
 AC_CONFIG_MACRO_DIR([m4])
 AC_CONFIG_SRCDIR([lib/Grid.h])
-AC_CONFIG_HEADERS([lib/Config.h])
+AC_CONFIG_HEADERS([lib/Config.h],[sed -i 's|PACKAGE_|GRID_|' lib/Config.h])
 m4_ifdef([AM_SILENT_RULES], [AM_SILENT_RULES([yes])])
 
 ############### Checks for programs
@@ -99,6 +99,13 @@ case ${ac_MKL} in
         AC_DEFINE([USE_MKL], [1], [Define to 1 if you use the Intel MKL]);;
 esac
 
+############### HDF5
+AC_ARG_WITH([hdf5],
+    [AS_HELP_STRING([--with-hdf5=prefix],
+    [try this for a non-standard install prefix of the HDF5 library])],
+    [AM_CXXFLAGS="-I$with_hdf5/include $AM_CXXFLAGS"]
+    [AM_LDFLAGS="-L$with_hdf5/lib $AM_LDFLAGS"])
+
 ############### first-touch
 AC_ARG_ENABLE([numa],
     [AC_HELP_STRING([--enable-numa=yes|no|prefix], [enable first touch numa opt])], 
@@ -145,6 +152,12 @@ AC_SEARCH_LIBS([fftw_execute], [fftw3],
                [AC_DEFINE([HAVE_FFTW], [1], [Define to 1 if you have the `FFTW' library])]
                [have_fftw=true])
 
+AC_SEARCH_LIBS([H5Fopen], [hdf5_cpp],
+               [AC_DEFINE([HAVE_HDF5], [1], [Define to 1 if you have the `HDF5' library])]
+               [have_hdf5=true]
+               [LIBS="${LIBS} -lhdf5"], [], [-lhdf5])
+AM_CONDITIONAL(BUILD_HDF5, [ test "${have_hdf5}X" == "trueX" ])
+
 CXXFLAGS=$CXXFLAGS_CPY
 LDFLAGS=$LDFLAGS_CPY
 
@@ -306,7 +319,7 @@ AM_CONDITIONAL(BUILD_COMMS_MPI3L, [ test "${comms_type}X" == "mpi3lX" ] )
 AM_CONDITIONAL(BUILD_COMMS_NONE,  [ test "${comms_type}X" == "noneX" ])
 
 ############### RNG selection
-AC_ARG_ENABLE([rng],[AC_HELP_STRING([--enable-rng=ranlux48|mt19937],\
+AC_ARG_ENABLE([rng],[AC_HELP_STRING([--enable-rng=ranlux48|mt19937|sitmo],\
 	            [Select Random Number Generator to be used])],\
 	            [ac_RNG=${enable_rng}],[ac_RNG=ranlux48])
 
@@ -317,6 +330,9 @@ case ${ac_RNG} in
      mt19937)
       AC_DEFINE([RNG_MT19937],[1],[RNG_MT19937] )
      ;;
+     sitmo)
+      AC_DEFINE([RNG_SITMO],[1],[RNG_SITMO] )
+     ;;
      *)
       AC_MSG_ERROR([${ac_RNG} unsupported --enable-rng option]); 
      ;;
@@ -381,10 +397,13 @@ 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(extras/Makefile)
+AC_CONFIG_FILES(extras/Hadrons/Makefile)
 AC_OUTPUT
 
 echo "~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
@@ -407,6 +426,7 @@ 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`
+HDF5                        : `if test "x$have_hdf5" = xtrue; then echo yes; else echo no; fi`
 build DOXYGEN documentation : `if test "$DX_FLAG_doc" = '1'; then echo yes; else echo no; fi`
 ----- BUILD FLAGS -------------------------------------
 CXXFLAGS:

extras/Hadrons/Application.cc

@@ -0,0 +1,317 @@
+/*************************************************************************************
+
+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)
+{
+    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;
+}
+
+// environment shortcut ////////////////////////////////////////////////////////
+Environment & Application::env(void) const
+{
+    return Environment::getInstance();
+}
+
+// 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_);
+    }
+    LOG(Message) << BIG_SEP << " End of measurement " << BIG_SEP << std::endl;
+    env().freeAll();
+}

extras/Hadrons/Application.hpp

@@ -0,0 +1,132 @@
+/*************************************************************************************
+
+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:
+    // environment shortcut
+    Environment & env(void) const;
+private:
+    long unsigned int         locVol_;
+    std::string               parameterFileName_{""};
+    GlobalPar                 par_;
+    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 +1,743 @@
+/*************************************************************************************
+
+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)
+{
+    nd_ = GridDefaultLatt().size();
+    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);
+    }
+}
+
+unsigned int Environment::getNd(void) const
+{
+    return nd_;
+}
+
+// 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)
+{
+    if (!hasObject(name))
+    {
+        ObjInfo info;
+        
+        info.name   = name;
+        info.module = moduleAddress;
+        object_.push_back(std::move(info));
+        objectAddress_[name] = static_cast<unsigned int>(object_.size() - 1);
+    }
+    else
+    {
+        HADRON_ERROR("object '" + name + "' already exists");
+    }
+}
+
+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)
+{
+    if (!hasObject(name))
+    {
+        addObject(name);
+    }
+    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::hasCreatedObject(const unsigned int address) const
+{
+    if (hasObject(address))
+    {
+        return (object_[address].data != nullptr);
+    }
+    else
+    {
+        return false;
+    }
+}
+
+bool Environment::hasCreatedObject(const std::string name) const
+{
+    if (hasObject(name))
+    {
+        return hasCreatedObject(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 +1,385 @@
+/*************************************************************************************
+
+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;
+    unsigned int            getNd(void) 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 = -1);
+    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                    hasCreatedObject(const unsigned int address) const;
+    bool                    hasCreatedObject(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_;
+    unsigned int                           nd_;
+    // 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_.insert(std::make_pair(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_.insert(std::make_pair(func_(c1), c1));
+        population_.insert(std::make_pair(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_.insert(std::make_pair(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

@@ -1,12 +1,13 @@
 /*************************************************************************************
 
-Grid physics library, www.github.com/paboyle/Grid
+Grid physics library, www.github.com/paboyle/Grid 
 
-Source file: ./lib/Bitwise.cc
+Source file: extras/Hadrons/ModuleFactory.hpp
 
+Copyright (C) 2015
 Copyright (C) 2016
 
-Author: Guido Cossu <guido.cossu@ed.ac.uk>
+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
@@ -22,28 +23,27 @@ 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
+See the full license in the file "LICENSE" in the top level distribution directory
 *************************************************************************************/
 /*  END LEGAL */
-#include <iostream>
-#include <Bitwise.h>
-#include <bitset>
-#include <climits>
 
-namespace Grid {
+#ifndef Hadrons_ModuleFactory_hpp_
+#define Hadrons_ModuleFactory_hpp_
 
-void show_binaryrep(const unsigned char* a, size_t size) {
-  const unsigned char* beg = a;
-  const unsigned char* end = a + size;
-  unsigned int ctr = 0;
-  while (beg != end) {
-    std::cout << std::bitset<CHAR_BIT>(*beg++) << ' ';
-    ctr++;
-    if (ctr % GRID_REAL_BYTES == 0) std::cout << '\n';
-  }
-  std::cout << '\n';
-}
+#include <Grid/Hadrons/Global.hpp>
+#include <Grid/Hadrons/Factory.hpp>
+#include <Grid/Hadrons/Module.hpp>
 
-} // namespace 
+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 +1,137 @@
+/*************************************************************************************
+
+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,
+                                    Gamma::Algebra, gammaSource,
+                                    Gamma::Algebra, gammaSink);
+};
+
+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<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());
+    Gamma                 gSrc(par().gammaSource), gSnk(par().gammaSink);
+    Gamma                 g5(Gamma::Algebra::Gamma5);
+    std::vector<TComplex> buf;
+    Result                result;
+    
+    c = trace(gSnk*q1*adj(gSrc)*g5*adj(q2)*g5);
+    sliceSum(c, buf, Tp);
+    result.corr.resize(buf.size());
+    for (unsigned int t = 0; t < buf.size(); ++t)
+    {
+        result.corr[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 +1,163 @@
+/*************************************************************************************
+
+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,
+                                    Gamma::Algebra, 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());
+    Gamma                      g(par().gamma);
+    std::vector<Real>          p;
+    Complex                    i(0.0,1.0);
+    
+    p  = strToVec<Real>(par().mom);
+    ph = zero;
+    for(unsigned int mu = 0; mu < env().getNd(); 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), ph*(g*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/Algorithms.h

@@ -39,6 +39,7 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 #include <Grid/algorithms/approx/MultiShiftFunction.h>
 
 #include <Grid/algorithms/iterative/ConjugateGradient.h>
+#include <Grid/algorithms/iterative/ConjugateGradientShifted.h>
 #include <Grid/algorithms/iterative/ConjugateResidual.h>
 #include <Grid/algorithms/iterative/NormalEquations.h>
 #include <Grid/algorithms/iterative/SchurRedBlack.h>

lib/AlignedAllocator.cc

@@ -0,0 +1,65 @@
+
+
+
+#include <Grid/Grid.h>
+
+namespace Grid {
+
+int PointerCache::victim;
+
+  PointerCache::PointerCacheEntry PointerCache::Entries[PointerCache::Ncache];
+
+void *PointerCache::Insert(void *ptr,size_t bytes) {
+
+  if (bytes < 4096 ) return NULL;
+
+#ifdef _OPENMP
+  assert(omp_in_parallel()==0);
+#endif 
+  void * ret = NULL;
+  int v = -1;
+
+  for(int e=0;e<Ncache;e++) {
+    if ( Entries[e].valid==0 ) {
+      v=e; 
+      break;
+    }
+  }
+
+  if ( v==-1 ) {
+    v=victim;
+    victim = (victim+1)%Ncache;
+  }
+
+  if ( Entries[v].valid ) {
+    ret = Entries[v].address;
+    Entries[v].valid = 0;
+    Entries[v].address = NULL;
+    Entries[v].bytes = 0;
+  }
+
+  Entries[v].address=ptr;
+  Entries[v].bytes  =bytes;
+  Entries[v].valid  =1;
+
+  return ret;
+}
+
+void *PointerCache::Lookup(size_t bytes) {
+
+ if (bytes < 4096 ) return NULL;
+
+#ifdef _OPENMP
+  assert(omp_in_parallel()==0);
+#endif 
+
+  for(int e=0;e<Ncache;e++){
+    if ( Entries[e].valid && ( Entries[e].bytes == bytes ) ) {
+      Entries[e].valid = 0;
+      return Entries[e].address;
+    }
+  }
+  return NULL;
+}
+
+}

lib/AlignedAllocator.h

@@ -1,4 +1,4 @@
-    /*************************************************************************************
+/*************************************************************************************
 
     Grid physics library, www.github.com/paboyle/Grid 
 
@@ -42,9 +42,32 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 
 namespace Grid {
 
+  class PointerCache {
+  private:
+
+    static const int Ncache=8;
+    static int victim;
+
+    typedef struct { 
+      void *address;
+      size_t bytes;
+      int valid;
+    } PointerCacheEntry;
+    
+    static PointerCacheEntry Entries[Ncache];
+
+  public:
+
+
+    static void *Insert(void *ptr,size_t bytes) ;
+    static void *Lookup(size_t bytes) ;
+
+  };
+
 ////////////////////////////////////////////////////////////////////
 // A lattice of something, but assume the something is SIMDized.
 ////////////////////////////////////////////////////////////////////
+
 template<typename _Tp>
 class alignedAllocator {
 public: 
@@ -66,27 +89,27 @@ public:
 
   pointer allocate(size_type __n, const void* _p= 0)
   { 
+    size_type bytes = __n*sizeof(_Tp);
+
+    _Tp *ptr = (_Tp *) PointerCache::Lookup(bytes);
+    
 #ifdef HAVE_MM_MALLOC_H
-    _Tp * ptr = (_Tp *) _mm_malloc(__n*sizeof(_Tp),128);
+    if ( ptr == (_Tp *) NULL ) ptr = (_Tp *) _mm_malloc(bytes,128);
 #else
-    _Tp * ptr = (_Tp *) memalign(128,__n*sizeof(_Tp));
+    if ( ptr == (_Tp *) NULL ) ptr = (_Tp *) memalign(128,bytes);
 #endif
 
-    _Tp tmp;
-#ifdef GRID_NUMA
-#pragma omp parallel for schedule(static)
-  for(int i=0;i<__n;i++){
-    ptr[i]=tmp;
-  }
-#endif 
     return ptr;
   }
 
-  void deallocate(pointer __p, size_type) { 
+  void deallocate(pointer __p, size_type __n) { 
+    size_type bytes = __n * sizeof(_Tp);
+    pointer __freeme = (pointer)PointerCache::Insert((void *)__p,bytes);
+
 #ifdef HAVE_MM_MALLOC_H
-    _mm_free((void *)__p); 
+    if ( __freeme ) _mm_free((void *)__freeme); 
 #else
-    free((void *)__p);
+    if ( __freeme ) free((void *)__freeme);
 #endif
   }
   void construct(pointer __p, const _Tp& __val) { };

lib/Bitwise.h

@@ -1,76 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/Bitwise.h
-
-Copyright (C) 2016
-
-Author: Guido Cossu <guido.cossu@ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#ifndef GRID_BITWISE_H
-#define GRID_BITWISE_H
-
-#include <cassert>
-#include <cfloat>
-#include <bitset>
-#include <climits>
-#include <Config.h>
-
-#ifdef GRID_DEFAULT_PRECISION_SINGLE
-#define GRID_REAL_BYTES 4
-#endif
-#ifdef GRID_DEFAULT_PRECISION_DOUBLE
-#define GRID_REAL_BYTES 8
-#endif
-
-
-namespace Grid {
-
-void show_binaryrep(const unsigned char* a, size_t size);
-
-template <typename T>
-void show_binaryrep(const T& a) {
-  const char* beg = reinterpret_cast<const char*>(&a);
-  const char* end = beg + sizeof(a);
-  unsigned int ctr = 0;
-  while (beg != end) {
-  	std::cout << std::bitset<CHAR_BIT>(*beg++) << ' ';
-  	ctr++;
-  	if (ctr % GRID_REAL_BYTES == 0) std::cout << '\n';
-  }
-  std::cout << '\n';
-}
-
-template <typename T>
-void bitwise_xor(T& l, T& r, unsigned char* xors) {
-  assert(sizeof(l) == sizeof(r));
-  unsigned char* org = reinterpret_cast<unsigned char*>(&l);
-  unsigned char* cur = reinterpret_cast<unsigned char*>(&r);
-  int words = sizeof(l) / sizeof(*org);
-  unsigned char result = 0;
-  for (int w = 0; w < words; w++) xors[w] = (org[w] ^ cur[w]);
-}
-
-}; // namespace 
-
-
-#endif

lib/Grid.h

@@ -59,14 +59,13 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 ///////////////////
 // Grid headers
 ///////////////////
-#include <Grid/serialisation/Serialisation.h>
 #include "Config.h"
 #include <Grid/Timer.h>
-#include <Grid/Bitwise.h>
 #include <Grid/PerfCount.h>
 #include <Grid/Log.h>
 #include <Grid/AlignedAllocator.h>
 #include <Grid/Simd.h>
+#include <Grid/serialisation/Serialisation.h>
 #include <Grid/Threads.h>
 #include <Grid/Lexicographic.h>
 #include <Grid/Init.h>

lib/Hadrons

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

lib/Init.cc

@@ -41,7 +41,7 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 #include <signal.h>
 #include <iostream>
 #include <iterator>
-#include <Grid.h>
+#include <Grid/Grid.h>
 #include <algorithm>
 #include <iterator>
 #include <cstdlib>

lib/Log.cc

@@ -29,7 +29,7 @@ See the full license in the file "LICENSE" in the top level distribution
 directory
 *************************************************************************************/
 /*  END LEGAL */
-#include <Grid.h>
+#include <Grid/Grid.h>
 
 #include <cxxabi.h>
 
@@ -100,7 +100,7 @@ void Grid_quiesce_nodes(void) {
   me = shmem_my_pe();
 #endif
   if (me) {
-    std::cout.setstate(std::ios::badbit);// mute all nodes except 0
+    std::cout.setstate(std::ios::badbit);
   }
 }
 

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/Makefile.am

@@ -1,4 +1,5 @@
 extra_sources=
+extra_headers=
 if BUILD_COMMS_MPI
   extra_sources+=communicator/Communicator_mpi.cc
   extra_sources+=communicator/Communicator_base.cc
@@ -24,6 +25,12 @@ if BUILD_COMMS_NONE
   extra_sources+=communicator/Communicator_base.cc
 endif
 
+if BUILD_HDF5
+  extra_sources+=serialisation/Hdf5IO.cc 
+  extra_headers+=serialisation/Hdf5IO.h
+  extra_headers+=serialisation/Hdf5Type.h
+endif
+
 #
 # Libraries
 #
@@ -32,6 +39,9 @@ include Eigen.inc
 
 lib_LIBRARIES = libGrid.a
 
-libGrid_a_SOURCES              = $(CCFILES) $(extra_sources)
+CCFILES += $(extra_sources)
+HFILES  += $(extra_headers)
+
+libGrid_a_SOURCES              = $(CCFILES)
 libGrid_adir                   = $(pkgincludedir)
 nobase_dist_pkginclude_HEADERS = $(HFILES) $(eigen_files) Config.h

lib/PerfCount.cc

@@ -26,8 +26,8 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
     *************************************************************************************/
     /*  END LEGAL */
 
-#include <Grid.h>
-#include <PerfCount.h>
+#include <Grid/Grid.h>
+#include <Grid/PerfCount.h>
 
 namespace Grid {
 

lib/PerfCount.h

@@ -205,12 +205,13 @@ public:
   void Stop(void) {
     count=0;
     cycles=0;
+    size_t ign;
 #ifdef __linux__
     if ( fd!= -1) {
       ::ioctl(fd, PERF_EVENT_IOC_DISABLE, 0);
       ::ioctl(cyclefd, PERF_EVENT_IOC_DISABLE, 0);
-      ::read(fd, &count, sizeof(long long));
-      ::read(cyclefd, &cycles, sizeof(long long));
+      ign=::read(fd, &count, sizeof(long long));
+      ign=::read(cyclefd, &cycles, sizeof(long long));
     }
     elapsed = cyclecount() - begin;
 #else

lib/Stat.cc

@@ -1,6 +1,6 @@
-#include <Grid.h>
-#include <PerfCount.h>
-#include <Stat.h>
+#include <Grid/Grid.h>
+#include <Grid/PerfCount.h>
+#include <Grid/Stat.h>
 
 
 namespace Grid { 

lib/Stencil.h

@@ -113,7 +113,7 @@ Gather_plane_simple_table (std::vector<std::pair<int,int> >& table,const Lattice
 {
 PARALLEL_FOR_LOOP     
      for(int i=0;i<table.size();i++){
-       buffer[off+table[i].first]=compress(rhs._odata[so+table[i].second]);
+       vstream(buffer[off+table[i].first],compress(rhs._odata[so+table[i].second]));
      }
 }
 

lib/Threads.h

@@ -38,21 +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_INTERN _Pragma("omp for schedule(static)")
-  #else
-  #define PARALLEL_FOR_LOOP        _Pragma("omp parallel for schedule(runtime)")
-  #define PARALLEL_FOR_LOOP_INTERN _Pragma("omp for schedule(runtime)")
+#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_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_FOR_LOOP_STATIC _Pragma("omp parallel for schedule(static)")
+#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_INTERN
 #define PARALLEL_NESTED_LOOP2
 #define PARALLEL_REGION
-#define PARALLEL_FOR_LOOP_STATIC
+#define PARALLEL_CRITICAL
 #endif
 
 namespace Grid {

lib/algorithms/approx/MultiShiftFunction.cc

@@ -25,7 +25,7 @@ Author: Azusa Yamaguchi <ayamaguc@staffmail.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>
 
 namespace Grid {
 double MultiShiftFunction::approx(double x)

lib/algorithms/approx/Remez.cc

@@ -20,7 +20,7 @@
 #include<iomanip>
 #include<cassert>
 
-#include<algorithms/approx/Remez.h>
+#include<Grid/algorithms/approx/Remez.h>
 
 // Constructor
 AlgRemez::AlgRemez(double lower, double upper, long precision) 

lib/algorithms/iterative/ConjugateGradient.h

@@ -9,7 +9,6 @@ Copyright (C) 2015
 Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
 Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 Author: paboyle <paboyle@ph.ed.ac.uk>
-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
@@ -34,21 +33,6 @@ directory
 
 namespace Grid {
 
-struct CG_state {
-  bool do_repro;
-  std::vector<RealD> residuals;
-
-  CG_state() {reset();}
-
-  void reset(){
-    do_repro = false;
-    residuals.clear();
-  }
-};
-
-
-enum CGexec_mode{ Default, ReproducibilityTest };
-
 /////////////////////////////////////////////////////////////
 // Base classes for iterative processes based on operators
 // single input vec, single output vec.
@@ -61,30 +45,10 @@ class ConjugateGradient : public OperatorFunction<Field> {
                            // Defaults true.
   RealD Tolerance;
   Integer MaxIterations;
-
-  // Reproducibility controls
-  bool ReproTest;
-  CG_state CGState; //to check reproducibility by repeating the CG
-  ReproducibilityState<typename Field::vector_object> ReprTest; // for the inner proucts
-
-  // Constructor
-  ConjugateGradient(RealD tol, Integer maxit, CGexec_mode Mode = Default)
-    : Tolerance(tol),MaxIterations(maxit){
-    switch(Mode)
-    {
-      case Default  : 
-      ErrorOnNoConverge = true;
-      ReproTest = false;
-      case ReproducibilityTest :
-      ErrorOnNoConverge = false;
-      ReproTest = true; 
-    }
-  };
-
-  void set_reproducibility_interval(unsigned int interval){
-    ReprTest.interval = interval;
-  }
-
+  ConjugateGradient(RealD tol, Integer maxit, bool err_on_no_conv = true)
+      : Tolerance(tol),
+        MaxIterations(maxit),
+        ErrorOnNoConverge(err_on_no_conv){};
 
   void operator()(LinearOperatorBase<Field> &Linop, const Field &src,
                   Field &psi) {
@@ -96,37 +60,34 @@ class ConjugateGradient : public OperatorFunction<Field> {
     Field p(src);
     Field mmp(src);
     Field r(src);
-    Field psi_start(psi);// save for the repro test
-
-    if (CGState.do_repro && ReproTest)
-        std::cout << GridLogMessage << "Starting reproducibility test, full check every "
-                  << ReprTest.interval << " calls" << std::endl;
-
-    if(!ReprTest.do_check)
-        ReprTest.reset();
-    ReprTest.enable_reprocheck=ReproTest;
-
-
 
     // Initial residual computation & set up
-    RealD guess = norm2(psi, ReprTest);
+    RealD guess = norm2(psi);
     assert(std::isnan(guess) == 0);
 
-    Linop.HermOpAndNorm(psi, mmp, d, b);// eventually split this for the norm check
     
+    Linop.HermOpAndNorm(psi, mmp, d, b);
+    
+
     r = src - mmp;
     p = r;
 
-    a = norm2(p, ReprTest);
+    a = norm2(p);
     cp = a;
-    ssq = norm2(src, ReprTest);
+    ssq = norm2(src);
 
-    std::cout << GridLogIterative << "ConjugateGradient: guess " << guess << std::endl;
-    std::cout << GridLogIterative << "ConjugateGradient:   src " << ssq << std::endl;
-    std::cout << GridLogIterative << "ConjugateGradient:    mp " << d << std::endl;
-    std::cout << GridLogIterative << "ConjugateGradient:   mmp " << b << std::endl;
-    std::cout << GridLogIterative << "ConjugateGradient:  cp,r " << cp << std::endl;
-    std::cout << GridLogIterative << "ConjugateGradient:     p " << a << std::endl;
+    std::cout << GridLogIterative << std::setprecision(4)
+              << "ConjugateGradient: guess " << guess << std::endl;
+    std::cout << GridLogIterative << std::setprecision(4)
+              << "ConjugateGradient:   src " << ssq << std::endl;
+    std::cout << GridLogIterative << std::setprecision(4)
+              << "ConjugateGradient:    mp " << d << std::endl;
+    std::cout << GridLogIterative << std::setprecision(4)
+              << "ConjugateGradient:   mmp " << b << std::endl;
+    std::cout << GridLogIterative << std::setprecision(4)
+              << "ConjugateGradient:  cp,r " << cp << std::endl;
+    std::cout << GridLogIterative << std::setprecision(4)
+              << "ConjugateGradient:     p " << a << std::endl;
 
     RealD rsq = Tolerance * Tolerance * ssq;
 
@@ -146,10 +107,10 @@ class ConjugateGradient : public OperatorFunction<Field> {
     SolverTimer.Start();
     int k;
     for (k = 1; k <= MaxIterations; k++) {
-      c = cp;// old residual
+      c = cp;
 
       MatrixTimer.Start();
-      Linop.HermOpAndNorm(p, mmp, d, qq);// mmp = Ap, d=pAp
+      Linop.HermOpAndNorm(p, mmp, d, qq);
       MatrixTimer.Stop();
 
       LinalgTimer.Start();
@@ -157,31 +118,14 @@ class ConjugateGradient : public OperatorFunction<Field> {
       //  ComplexD dck  = innerProduct(p,mmp);
 
       a = c / d;
-      b_pred = a * (a * qq - d) / c;// a check
+      b_pred = a * (a * qq - d) / c;
 
-
-      axpy(r, -a, mmp, r);// new residual r = r_old - a * Ap
-      cp = norm2(r, ReprTest); // bookkeeping this norm
-      if (ReproTest && !CGState.do_repro) {
-        CGState.residuals.push_back(cp);  // save residuals state
-                std::cout << GridLogIterative << "ReproTest: Saving state" << std::endl;
-        }
-      if (ReproTest && CGState.do_repro){
-        // check that the residual agrees with the previous run
-        std::cout << GridLogIterative << "ReproTest: Checking state k=" << k << std::endl;
-        if (cp != CGState.residuals[k-1]){
-                std::cout << GridLogMessage << "Failing reproducibility test";
-                std::cout << GridLogMessage << " at k=" << k << std::endl;
-                std::cout << GridLogMessage << "saved residual = " << CGState.residuals[k-1] 
-                        << " cp = " << cp << std::endl;
-                exit(1);  // exit after the first failure
-        }
-      }
+      cp = axpy_norm(r, -a, mmp, r);
       b = cp / c;
 
       // Fuse these loops ; should be really easy
-      psi = a * p + psi; // update solution
-      p = p * b + r;  // update search direction
+      psi = a * p + psi;
+      p = p * b + r;
 
       LinalgTimer.Stop();
       std::cout << GridLogIterative << "ConjugateGradient: Iteration " << k
@@ -212,22 +156,6 @@ class ConjugateGradient : public OperatorFunction<Field> {
 
         if (ErrorOnNoConverge) assert(true_residual / Tolerance < 10000.0);
 
-        if (! (CGState.do_repro && ReproTest)){
-                CGState.do_repro = true;
-                ReprTest.do_check = true;
-                ReprTest.reset_counter();
-                this->operator()(Linop, src, psi_start);// run the repro test
-                if (ReprTest.success)
-                	std::cout << GridLogMessage << "Reproducibility test passed" << std::endl;
-                else{
-                	std::cout << GridLogMessage << "Reproducibility test failed" << std::endl;
-                	exit(1);
-                }
-        }
-
-        // Clear state
-        CGState.reset();
-        ReprTest.reset();
         return;
       }
     }

lib/algorithms/iterative/ConjugateGradientMixedPrec.h

@@ -55,6 +55,9 @@ namespace Grid {
     }
   
     void operator() (const FieldD &src_d_in, FieldD &sol_d){
+	(*this)(src_d_in,sol_d,NULL);
+    }
+    void operator() (const FieldD &src_d_in, FieldD &sol_d, RealD *shift){
       GridStopWatch TotalTimer;
       TotalTimer.Start();
     
@@ -82,7 +85,7 @@ namespace Grid {
       FieldF sol_f(SinglePrecGrid);
       sol_f.checkerboard = cb;
     
-      ConjugateGradient<FieldF> CG_f(inner_tol, MaxInnerIterations);
+      ConjugateGradientShifted<FieldF> CG_f(inner_tol, MaxInnerIterations);
       CG_f.ErrorOnNoConverge = false;
 
       GridStopWatch InnerCGtimer;
@@ -92,6 +95,7 @@ namespace Grid {
       for(Integer outer_iter = 0; outer_iter < MaxOuterIterations; outer_iter++){
 	//Compute double precision rsd and also new RHS vector.
 	Linop_d.HermOp(sol_d, tmp_d);
+	if(shift) axpy(tmp_d,*shift,sol_d,tmp_d);
 	RealD norm = axpy_norm(src_d, -1., tmp_d, src_d_in); //src_d is residual vector
       
 	std::cout<<GridLogMessage<<"MixedPrecisionConjugateGradient: Outer iteration " <<outer_iter<<" residual "<< norm<< " target "<< stop<<std::endl;
@@ -115,7 +119,7 @@ namespace Grid {
 	//Inner CG
 	CG_f.Tolerance = inner_tol;
 	InnerCGtimer.Start();
-	CG_f(Linop_f, src_f, sol_f);
+	CG_f(Linop_f, src_f, sol_f,shift);
 	InnerCGtimer.Stop();
       
 	//Convert sol back to double and add to double prec solution
@@ -129,8 +133,8 @@ namespace Grid {
       //Final trial CG
       std::cout<<GridLogMessage<<"MixedPrecisionConjugateGradient: Starting final patch-up double-precision solve"<<std::endl;
     
-      ConjugateGradient<FieldD> CG_d(Tolerance, MaxInnerIterations);
-      CG_d(Linop_d, src_d_in, sol_d);
+      ConjugateGradientShifted<FieldD> CG_d(Tolerance, MaxInnerIterations);
+      CG_d(Linop_d, src_d_in, sol_d,shift);
 
       TotalTimer.Stop();
       std::cout<<GridLogMessage<<"MixedPrecisionConjugateGradient: Total " << TotalTimer.Elapsed() << " Precision change " << PrecChangeTimer.Elapsed() << " Inner CG total " << InnerCGtimer.Elapsed() << std::endl;

lib/algorithms/iterative/ConjugateGradientMultiShift.h

@@ -45,6 +45,7 @@ public:
     Integer MaxIterations;
     int verbose;
     MultiShiftFunction shifts;
+    int iter;
 
     ConjugateGradientMultiShift(Integer maxit,MultiShiftFunction &_shifts) : 
 	MaxIterations(maxit),
@@ -60,6 +61,7 @@ void operator() (LinearOperatorBase<Field> &Linop, const Field &src, Field &psi)
   std::vector<Field> results(nshift,grid);
   (*this)(Linop,src,results,psi);
 }
+
 void operator() (LinearOperatorBase<Field> &Linop, const Field &src, std::vector<Field> &results, Field &psi)
 {
   int nshift = shifts.order;
@@ -105,11 +107,12 @@ void operator() (LinearOperatorBase<Field> &Linop, const Field &src, std::vector
   RealD a,b,c,d;
   RealD cp,bp,qq; //prev
   
+  int cb=src.checkerboard;
   // Matrix mult fields
   Field r(grid);
-  Field p(grid);
+  Field p(grid); p.checkerboard = src.checkerboard;
   Field tmp(grid);
-  Field mmp(grid);
+  Field mmp(grid);mmp.checkerboard = src.checkerboard;
   
   // Check lightest mass
   for(int s=0;s<nshift;s++){
@@ -132,6 +135,9 @@ void operator() (LinearOperatorBase<Field> &Linop, const Field &src, std::vector
   p=src;
   
   //MdagM+m[0]
+  std::cout << "p.checkerboard " << p.checkerboard
+  << "mmp.checkerboard " << mmp.checkerboard << std::endl;
+
   Linop.HermOpAndNorm(p,mmp,d,qq);
   axpy(mmp,mass[0],p,mmp);
   RealD rn = norm2(p);
@@ -269,6 +275,7 @@ void operator() (LinearOperatorBase<Field> &Linop, const Field &src, std::vector
 	RealD cn = norm2(src);
 	std::cout<<GridLogMessage<<"CGMultiShift: shift["<<s<<"] true residual "<<std::sqrt(rn/cn)<<std::endl;
       }
+      iter = k;
       return;
     }
   }

lib/algorithms/iterative/ConjugateGradientMultiShiftMixedPrec.h

@@ -0,0 +1,404 @@
+    /*************************************************************************************
+
+    Grid physics library, www.github.com/paboyle/Grid 
+
+    Source file: ./lib/algorithms/iterative/ConjugateGradientMultiShiftMixedPrec.h
+
+    Copyright (C) 2015
+
+Author: Chulwoo Jung <chulwoo@quark.phy.bnl.gov>
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License along
+    with this program; if not, write to the Free Software Foundation, Inc.,
+    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+    See the full license in the file "LICENSE" in the top level distribution directory
+    *************************************************************************************/
+    /*  END/ LEGAL */
+#ifndef GRID_CONJUGATE_GRADIENT_MULTI_MIXED_PREC_H
+#define GRID_CONJUGATE_GRADIENT_MULTI_MIXED_PREC_H
+
+namespace Grid {
+
+  //Mixed precision restarted defect correction CG
+  template<class FieldD,class FieldF
+//, typename std::enable_if< getPrecision<FieldD>::value == 2, int>::type = 0
+//, typename std::enable_if< getPrecision<FieldF>::value == 1, int>::type = 0
+> 
+  class MixedPrecisionConjugateGradientMultiShift : public LinearFunction<FieldD> {
+  public:                                                
+//    RealD   Tolerance;
+    Integer MaxInnerIterations;
+    Integer MaxOuterIterations;
+    GridBase* SinglePrecGrid; //Grid for single-precision fields
+    RealD OuterLoopNormMult; //Stop the outer loop and move to a final double prec solve when the residual is OuterLoopNormMult * Tolerance
+    LinearOperatorBase<FieldF> &Linop_f;
+    LinearOperatorBase<FieldD> &Linop_d;
+    MultiShiftFunction shifts;
+    Integer iter;
+
+    //Option to speed up *inner single precision* solves using a LinearFunction that produces a guess
+//    LinearFunction<FieldF> *guesser;
+    
+    MixedPrecisionConjugateGradientMultiShift(GridBase* _sp_grid, LinearOperatorBase<FieldF> &_Linop_f, LinearOperatorBase<FieldD> &_Linop_d, 
+Integer maxinnerit,	MultiShiftFunction &_shifts ) :
+      Linop_f(_Linop_f), Linop_d(_Linop_d),
+      MaxInnerIterations(maxinnerit), SinglePrecGrid(_sp_grid),
+      OuterLoopNormMult(100.), shifts(_shifts) {};
+
+  
+    void operator() (const FieldD &src_d_in, FieldD &sol_d){
+	assert(0); // not yet implemented
+    }
+    void operator() (const FieldD &src_d_in, std::vector<FieldD> &sol_d){
+      GridStopWatch TotalTimer;
+      TotalTimer.Start();
+    
+      int cb = src_d_in.checkerboard;
+
+      int nshift = shifts.order;
+      assert(nshift == sol_d.size());
+      for(int i=0;i<nshift;i++) sol_d[i].checkerboard = cb;
+    
+      RealD src_norm = norm2(src_d_in);
+//      RealD stop = src_norm * Tolerance*Tolerance;
+
+      GridBase* DoublePrecGrid = src_d_in._grid;
+      FieldD tmp_d(DoublePrecGrid); tmp_d.checkerboard = cb;
+    
+      FieldD tmp2_d(DoublePrecGrid); tmp2_d.checkerboard = cb;
+    
+      FieldD src_d(DoublePrecGrid);
+      src_d = src_d_in; //source for next inner iteration, computed from residual during operation
+    
+//      RealD inner_tol = Tolerance;
+  	FieldD psi_d(DoublePrecGrid);psi_d.checkerboard = cb;
+    
+      FieldF src_f(SinglePrecGrid);
+      src_f.checkerboard = cb;
+    
+      std::vector<FieldF> sol_f(nshift,SinglePrecGrid);
+      for(int i=0;i<nshift;i++) sol_f[i].checkerboard = cb;
+    
+//      ConjugateGradientShifted<FieldF> CG_f(inner_tol, MaxInnerIterations);
+      ConjugateGradientMultiShift<FieldF> MSCG(MaxInnerIterations,shifts);
+//      CG_f.ErrorOnNoConverge = false;
+
+      GridStopWatch InnerCGtimer;
+
+      GridStopWatch PrecChangeTimer;
+    
+{
+//	std::cout<<GridLogMessage<<"MixedPrecisionConjugateGradient: Outer iteration " <<outer_iter<<" residual "<< norm<< " target "<< stop<<std::endl;
+
+//	if(norm < OuterLoopNormMult * stop){
+//	  std::cout<<GridLogMessage<<"MixedPrecisionConjugateGradient: Outer iteration converged on iteration " <<outer_iter <<std::endl;
+//	  break;
+//	}
+//	while(norm * inner_tol * inner_tol < stop) inner_tol *= 2;  // inner_tol = sqrt(stop/norm) ??
+
+	PrecChangeTimer.Start();
+	precisionChange(src_f, src_d);
+	PrecChangeTimer.Stop();
+      
+//	zeroit(sol_f);
+
+
+	//Inner CG
+	InnerCGtimer.Start();
+  int if_relup = 0;
+#if 0
+        MSCG(Linop_f,src_f,sol_f);
+#else
+{
+  
+  GridBase *grid = SinglePrecGrid;
+  
+  ////////////////////////////////////////////////////////////////////////
+  // Convenience references to the info stored in "MultiShiftFunction"
+  ////////////////////////////////////////////////////////////////////////
+  int nshift = shifts.order;
+
+
+  std::vector<RealD> &mass(shifts.poles); // Make references to array in "shifts"
+  std::vector<RealD> &mresidual(shifts.tolerances);
+  std::vector<RealD> alpha(nshift,1.);
+  std::vector<FieldF>   ps(nshift,grid);// Search directions
+
+  assert(sol_f.size()==nshift);
+  assert(mass.size()==nshift);
+  assert(mresidual.size()==nshift);
+  
+  // dynamic sized arrays on stack; 2d is a pain with vector
+  RealD  bs[nshift];
+  RealD  rsq[nshift];
+  RealD  z[nshift][2];
+  int     converged[nshift];
+  
+  const int       primary =0;
+  
+  //Primary shift fields CG iteration
+  RealD a,b,c,d;
+  RealD cp,bp,qq; //prev
+  
+  int cb=src_f.checkerboard;
+  // Matrix mult fields
+  FieldF r(grid); r.checkerboard = src_f.checkerboard;
+  FieldF p(grid); p.checkerboard = src_f.checkerboard;
+  FieldF tmp(grid); tmp.checkerboard = src_f.checkerboard;
+  FieldF mmp(grid);mmp.checkerboard = src_f.checkerboard;
+  FieldF psi(grid);psi.checkerboard = src_f.checkerboard;
+    std::cout.precision(12);
+    std::cout<<GridLogMessage<<"norm2(psi_d)= "<<norm2(psi_d)<<std::endl;
+    std::cout<<GridLogMessage<<"norm2(psi)= "<<norm2(psi)<<std::endl;
+  
+  
+  // Check lightest mass
+  for(int s=0;s<nshift;s++){
+    assert( mass[s]>= mass[primary] );
+    converged[s]=0;
+  }
+  
+  // Wire guess to zero
+  // Residuals "r" are src
+  // First search direction "p" is also src
+  cp = norm2(src_f);
+  Real c_relup = cp;
+  for(int s=0;s<nshift;s++){
+    rsq[s] = cp * mresidual[s] * mresidual[s];
+    std::cout<<GridLogMessage<<"MixedPrecisionConjugateGradientMultiShift: shift "<<s
+	     <<" target resid "<<rsq[s]<<std::endl;
+    ps[s] = src_f;
+  }
+  // r and p for primary
+  r=src_f;
+  p=src_f;
+  
+  //MdagM+m[0]
+  std::cout << "p.checkerboard " << p.checkerboard
+  << "mmp.checkerboard " << mmp.checkerboard << std::endl;
+
+  Linop_f.HermOpAndNorm(p,mmp,d,qq);
+  axpy(mmp,mass[0],p,mmp);
+  RealD rn = norm2(p);
+  d += rn*mass[0];
+  
+  // have verified that inner product of 
+  // p and mmp is equal to d after this since
+  // the d computation is tricky
+  //  qq = real(innerProduct(p,mmp));
+  //  std::cout<<GridLogMessage << "debug equal ?  qq "<<qq<<" d "<< d<<std::endl;
+  
+  b = -cp /d;
+  
+  // Set up the various shift variables
+  int       iz=0;
+  z[0][1-iz] = 1.0;
+  z[0][iz]   = 1.0;
+  bs[0]      = b;
+  for(int s=1;s<nshift;s++){
+    z[s][1-iz] = 1.0;
+    z[s][iz]   = 1.0/( 1.0 - b*(mass[s]-mass[0]));
+    bs[s]      = b*z[s][iz]; 
+  }
+  
+  // r += b[0] A.p[0]
+  // c= norm(r)
+  c=axpy_norm(r,b,mmp,r);
+  
+ axpby(psi,0.,-bs[0],src_f,src_f);
+  for(int s=0;s<nshift;s++) {
+    axpby(sol_f[s],0.,-bs[s]*alpha[s],src_f,src_f);
+  }
+  
+  
+  // Iteration loop
+  int k;
+ // inefficient zeroing, please replace!
+//  RealD sol_norm = axpy_norm(sol_d[0],-1.,sol_d[0],sol_d[0]);
+  zeroit(sol_d[0]);
+  std::cout<<GridLogMessage<<"norm(sol_d[0])= "<<norm2(sol_d[0])<<std::endl;
+  
+
+  int all_converged = 1;
+	RealD tmp1,tmp2;
+  for (k=1;k<=MaxOuterIterations;k++){
+    
+    a = c /cp;
+    axpy(p,a,p,r);
+    
+    // Note to self - direction ps is iterated seperately
+    // for each shift. Does not appear to have any scope
+    // for avoiding linear algebra in "single" case.
+    // 
+    // However SAME r is used. Could load "r" and update
+    // ALL ps[s]. 2/3 Bandwidth saving
+    // New Kernel: Load r, vector of coeffs, vector of pointers ps
+    for(int s=0;s<nshift;s++){
+      if ( ! converged[s] ) { 
+	if (s==0){
+	  axpy(ps[s],a,ps[s],r);
+	} else{
+	  RealD as =a *z[s][iz]*bs[s] /(z[s][1-iz]*b);
+	  axpby(ps[s],z[s][iz],as,r,ps[s]);
+	}
+      }
+    }
+    
+    cp=c;
+    
+    Linop_f.HermOpAndNorm(p,mmp,d,qq);
+    axpy(mmp,mass[0],p,mmp);
+    RealD rn = norm2(p);
+    d += rn*mass[0];
+    
+    bp=b;
+    b=-cp/d;
+    
+    c=axpy_norm(r,b,mmp,r);
+
+
+    // Toggle the recurrence history
+    bs[0] = b;
+    iz = 1-iz;
+    for(int s=1;s<nshift;s++){
+      if((!converged[s])){
+	RealD z0 = z[s][1-iz];
+	RealD z1 = z[s][iz];
+	z[s][iz] = z0*z1*bp
+	  / (b*a*(z1-z0) + z1*bp*(1- (mass[s]-mass[0])*b)); 
+	bs[s] = b*z[s][iz]/z0; // NB sign  rel to Mike
+      }
+    }
+    
+    axpy(psi,-bs[0],ps[0],psi);
+    for(int s=0;s<nshift;s++){
+      int ss = s;
+      // Scope for optimisation here in case of "single".
+      // Could load sol_f[0] and pull all ps[s] in.
+      //      if ( single ) ss=primary;
+      // Bandwith saving in single case is Ls * 3 -> 2+Ls, so ~ 3x saving
+      // Pipelined CG gain:
+      //
+      // New Kernel: Load r, vector of coeffs, vector of pointers ps
+      // New Kernel: Load sol_f[0], vector of coeffs, vector of pointers ps
+      // If can predict the coefficient bs then we can fuse these and avoid write reread cyce
+      //  on ps[s].
+      // Before:  3 x npole  + 3 x npole
+      // After :  2 x npole (ps[s])        => 3x speed up of multishift CG.
+      
+      if( (!converged[s]) ) { 
+	axpy(sol_f[ss],-bs[s]*alpha[s],ps[s],sol_f[ss]);
+      }
+    }
+
+
+    if (k%MaxInnerIterations==0){
+//    if (c < 1e-4*c_relup){
+       RealD c_f=c;
+       precisionChange(tmp_d,psi);
+       RealD sol_norm =axpy_norm (psi_d,1.,tmp_d,psi_d);
+       tmp1 = norm2(psi);
+       zeroit(psi);
+       tmp2 = norm2(psi);
+       std::cout<<GridLogMessage<<"k= "<<k<<" norm2(sol)= "<<sol_norm<<" "<<tmp1<<" "<<tmp2<<std::endl;
+//       precisionChange(sol_d[0],sol_f[0]);
+       Linop_d.HermOpAndNorm(psi_d,tmp_d,tmp1,tmp2);
+       axpy(tmp2_d,mass[0],psi_d,tmp_d);
+       axpy(tmp_d,-1.,tmp2_d,src_d);
+       precisionChange(r,tmp_d);
+	c_relup = norm2(r);
+       std::cout<<GridLogMessage<<"k= "<<k<<" norm2(r)= "<<c<<" "<<c_relup<<" "<<c_f<<std::endl;
+	if_relup=1;
+    }
+    
+    // Convergence checks
+  all_converged=1;
+    for(int s=0;s<nshift;s++){
+      
+      if ( (!converged[s]) ){
+	
+	RealD css  = c * z[s][iz]* z[s][iz];
+	
+	if(css<rsq[s]){
+	  if ( ! converged[s] )
+	    std::cout<<GridLogMessage<<"ConjugateGradientMultiShift k="<<k<<" Shift "<<s<<" has converged"<<std::endl;
+	      converged[s]=1;
+	} else {
+		if (k%MaxInnerIterations==0)
+	    std::cout<<GridLogMessage<<"ConjugateGradientMultiShift k="<<k<<" Shift "<<s<<" has not converged "<<css<<"<"<<rsq[s]<<std::endl;
+	  all_converged=0;
+	}
+
+      }
+    }
+    
+#if 0
+    if ( all_converged ){
+      std::cout<<GridLogMessage<< "CGMultiShift: All shifts have converged iteration "<<k<<std::endl;
+#else
+    if ( converged[0] ){
+      std::cout<<GridLogMessage<< "CGMultiShift: Shift 0 have converged iteration, terminating  "<<k<<std::endl;
+#endif
+      
+#if 1
+      for(int s=1; s < nshift; s++) { 
+	Linop_f.HermOpAndNorm(sol_f[s],mmp,d,qq);
+	axpy(tmp,mass[s],sol_f[s],mmp);
+	axpy(r,-alpha[s],src_f,tmp);
+	RealD rn = norm2(r);
+	RealD cn = norm2(src_f);
+	std::cout<<GridLogMessage<<"CGMultiShift: shift["<<s<<"] true residual "<<std::sqrt(rn/cn)<<std::endl;
+      }
+#endif
+     iter = k;
+      break;
+    }
+  }
+  // ugly hack
+  if ( !all_converged )
+  std::cout<<GridLogMessage<<"CG multi shift did not converge"<<std::endl;
+//  assert(0);
+}
+	
+#endif
+	InnerCGtimer.Stop();
+      
+	//Convert sol back to double and add to double prec solution
+	PrecChangeTimer.Start();
+	sol_d[0]=psi_d;
+	for(int i=1;i<nshift;i++)precisionChange(sol_d[i], sol_f[i]);
+      std::cout<<GridLogMessage<< "CGMultiShift: Checking solutions"<<std::endl;
+      // Check answers 
+      for(int s=0; s < nshift; s++) { 
+	RealD tmp1,tmp2;
+       Linop_d.HermOpAndNorm(sol_d[s],tmp_d,tmp1,tmp2);
+       axpy(tmp2_d,shifts.poles[s],sol_d[s],tmp_d);
+       axpy(tmp_d,-1.,src_d,tmp2_d);
+	std::cout<<GridLogMessage<<"CGMultiShift: shift["<<s<<"] true residual "<<std::sqrt(norm2(tmp_d)/norm2(src_d))<<std::endl;
+      }
+	PrecChangeTimer.Stop();
+      
+}
+    
+      //Final trial CG
+ //     std::cout<<GridLogMessage<<"MixedPrecisionConjugateGradient: Starting final patch-up double-precision solve"<<std::endl;
+    
+      TotalTimer.Stop();
+      std::cout<<GridLogMessage<<"MixedPrecisionConjugateGradient: Total " << TotalTimer.Elapsed() << " Precision change " << PrecChangeTimer.Elapsed() << " Inner CG total " << InnerCGtimer.Elapsed() << std::endl;
+    }
+  };
+
+}
+
+#endif

lib/algorithms/iterative/ConjugateGradientShifted.h

@@ -0,0 +1,168 @@
+    /*************************************************************************************
+
+    Grid physics library, www.github.com/paboyle/Grid 
+
+    Source file: ./lib/algorithms/iterative/ConjugateGradient.h
+
+    Copyright (C) 2015
+
+Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
+Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+Author: paboyle <paboyle@ph.ed.ac.uk>
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License along
+    with this program; if not, write to the Free Software Foundation, Inc.,
+    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+    See the full license in the file "LICENSE" in the top level distribution directory
+    *************************************************************************************/
+    /*  END LEGAL */
+#ifndef GRID_CONJUGATE_GRADIENT_SHIFTED_H
+#define GRID_CONJUGATE_GRADIENT_SHIFTED_H
+
+namespace Grid {
+
+    /////////////////////////////////////////////////////////////
+    // Base classes for iterative processes based on operators
+    // single input vec, single output vec.
+    /////////////////////////////////////////////////////////////
+
+  template<class Field> 
+    class ConjugateGradientShifted : public OperatorFunction<Field> {
+public:                                                
+    bool ErrorOnNoConverge; //throw an assert when the CG fails to converge. Defaults true.
+    RealD   Tolerance;
+    Integer MaxIterations;
+    ConjugateGradientShifted(RealD tol,Integer maxit, bool err_on_no_conv = true) : Tolerance(tol), MaxIterations(maxit), ErrorOnNoConverge(err_on_no_conv) { 
+    };
+
+    void operator() (LinearOperatorBase<Field> &Linop,const Field &src, Field &psi ){
+	(*this)(Linop,src,psi,NULL);
+    }
+
+    void operator() (LinearOperatorBase<Field> &Linop,const Field &src, Field &psi, RealD *shift){
+
+      psi.checkerboard = src.checkerboard;
+      conformable(psi,src);
+
+      RealD cp,c,a,d,b,ssq,qq,b_pred;
+      
+      Field   p(src);
+      Field mmp(src);
+      Field   r(src);
+      
+      //Initial residual computation & set up
+      RealD guess = norm2(psi);
+      assert(std::isnan(guess)==0);
+
+      Linop.HermOpAndNorm(psi,mmp,d,b);
+	if(shift) axpy(mmp,*shift,psi,mmp);
+	RealD rn = norm2(psi);
+	if(shift) d += rn*(*shift);
+	RealD d2 = real(innerProduct(psi,mmp));
+	b= norm2(mmp);
+      RealD src_norm=norm2(src);
+      r= src-mmp;
+      p= r;
+      
+      a  =norm2(p);
+      cp =a;
+      ssq=norm2(src);
+
+      std::cout<<GridLogIterative <<std::setprecision(4)<< "ConjugateGradient: guess "<<guess<<std::endl;
+      std::cout<<GridLogIterative <<std::setprecision(4)<< "ConjugateGradient:   src "<<ssq  <<std::endl;
+      std::cout<<GridLogIterative <<std::setprecision(4)<< "ConjugateGradient:    mp "<<d    <<std::endl;
+      std::cout<<GridLogIterative <<std::setprecision(4)<< "ConjugateGradient:   mmp "<<b    <<std::endl;
+      std::cout<<GridLogIterative <<std::setprecision(4)<< "ConjugateGradient:  cp,r "<<cp   <<std::endl;
+      std::cout<<GridLogIterative <<std::setprecision(4)<< "ConjugateGradient:     p "<<a    <<std::endl;
+
+      RealD rsq =  Tolerance* Tolerance*ssq;
+      
+      //Check if guess is really REALLY good :)
+      if ( cp <= rsq ) {
+	return;
+      }
+      
+      std::cout<<GridLogIterative << std::setprecision(4)<< "ConjugateGradient: k=0 residual "<<cp<<" target "<<rsq<<std::endl;
+
+      GridStopWatch LinalgTimer;
+      GridStopWatch MatrixTimer;
+      GridStopWatch SolverTimer;
+
+      SolverTimer.Start();
+      int k;
+      for (k=1;k<=MaxIterations;k++){
+	
+	c=cp;
+
+	MatrixTimer.Start();
+	Linop.HermOpAndNorm(p,mmp,d,qq);
+	MatrixTimer.Stop();
+	LinalgTimer.Start();
+	if(shift) axpy(mmp,*shift,p,mmp);
+	RealD rn = norm2(p);
+	if(shift) d += rn*(*shift);
+	RealD d2 = real(innerProduct(p,mmp));
+	qq = norm2(mmp);
+      if (k%10==1) std::cout<< std::setprecision(4)<< "d:  "<<d<<" d2= "<<d2<<std::endl;
+
+	//	RealD    qqck = norm2(mmp);
+	//	ComplexD dck  = innerProduct(p,mmp);
+      
+	a      = c/d;
+	b_pred = a*(a*qq-d)/c;
+
+	cp = axpy_norm(r,-a,mmp,r);
+	b = cp/c;
+      if (k%10==1) std::cout<< std::setprecision(4)<<"k= "<<k<<" src:  "<<src_norm<<" r= "<<cp<<std::endl;
+	
+	// Fuse these loops ; should be really easy
+	psi= a*p+psi;
+	p  = p*b+r;
+	  
+	LinalgTimer.Stop();
+	std::cout<<GridLogIterative<<"ConjugateGradient: Iteration " <<k<<" residual "<<cp<< " target "<< rsq<<std::endl;
+	
+	// Stopping condition
+	if ( cp <= rsq ) { 
+	  
+	  SolverTimer.Stop();
+	  Linop.HermOpAndNorm(psi,mmp,d,qq);
+	  if(shift) mmp = mmp + (*shift) * psi;
+	  p=mmp-src;
+	  
+	  RealD mmpnorm = sqrt(norm2(mmp));
+	  RealD psinorm = sqrt(norm2(psi));
+	  RealD srcnorm = sqrt(norm2(src));
+	  RealD resnorm = sqrt(norm2(p));
+	  RealD true_residual = resnorm/srcnorm;
+
+	  std::cout<<GridLogMessage<<"ConjugateGradient: Converged on iteration " <<k
+		   <<" computed residual "<<sqrt(cp/ssq)
+		   <<" true residual "    <<true_residual
+		   <<" target "<<Tolerance<<std::endl;
+	  std::cout<<GridLogMessage<<"Time elapsed: Total "<< SolverTimer.Elapsed() << " Matrix  "<<MatrixTimer.Elapsed() << " Linalg "<<LinalgTimer.Elapsed();
+	  std::cout<<std::endl;
+	  
+	if(ErrorOnNoConverge)
+	  assert(true_residual/Tolerance < 1000.0);
+
+	  return;
+	}
+      }
+      std::cout<<GridLogMessage<<"ConjugateGradient did NOT converge"<<std::endl;
+//      assert(0);
+    }
+  };
+}
+#endif

lib/algorithms/iterative/ImplicitlyRestartedLanczos.h

@@ -31,11 +31,16 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 
 #include <string.h> //memset
 #ifdef USE_LAPACK
+#ifdef USE_MKL
+#include<mkl_lapack.h>
+#else
 void LAPACK_dstegr(char *jobz, char *range, int *n, double *d, double *e,
                    double *vl, double *vu, int *il, int *iu, double *abstol,
                    int *m, double *w, double *z, int *ldz, int *isuppz,
                    double *work, int *lwork, int *iwork, int *liwork,
                    int *info);
+//#include <lapacke/lapacke.h>
+#endif
 #endif
 #include "DenseMatrix.h"
 #include "EigenSort.h"
@@ -59,15 +64,16 @@ public:
 
     int Nstop;   // Number of evecs checked for convergence
     int Nk;      // Number of converged sought
-    int Np;      // Np -- Number of spare vecs in krylov space
+    int Np;      // Np -- Number of spare vecs in kryloc space
     int Nm;      // Nm -- total number of vectors
 
+
+    RealD OrthoTime;
+
     RealD eresid;
 
     SortEigen<Field> _sort;
 
-//    GridCartesian &_fgrid;
-
     LinearOperatorBase<Field> &_Linop;
 
     OperatorFunction<Field>   &_poly;
@@ -124,23 +130,23 @@ public:
 
       GridBase *grid = evec[0]._grid;
       Field w(grid);
-      std::cout << "RitzMatrix "<<std::endl;
+      std::cout<<GridLogMessage << "RitzMatrix "<<std::endl;
       for(int i=0;i<k;i++){
 	_poly(_Linop,evec[i],w);
-	std::cout << "["<<i<<"] ";
+	std::cout<<GridLogMessage << "["<<i<<"] ";
 	for(int j=0;j<k;j++){
 	  ComplexD in = innerProduct(evec[j],w);
 	  if ( fabs((double)i-j)>1 ) { 
 	    if (abs(in) >1.0e-9 )  { 
-	      std::cout<<"oops"<<std::endl;
+	      std::cout<<GridLogMessage<<"oops"<<std::endl;
 	      abort();
 	    } else 
-	      std::cout << " 0 ";
+	      std::cout<<GridLogMessage << " 0 ";
 	  } else { 
-	    std::cout << " "<<in<<" ";
+	    std::cout<<GridLogMessage << " "<<in<<" ";
 	  }
 	}
-	std::cout << std::endl;
+	std::cout<<GridLogMessage << std::endl;
       }
     }
 
@@ -174,10 +180,10 @@ public:
       RealD beta = normalise(w); // 6. βk+1 := ∥wk∥2. If βk+1 = 0 then Stop
                                  // 7. vk+1 := wk/βk+1
 
-//	std::cout << "alpha = " << zalph << " beta "<<beta<<std::endl;
+	std::cout<<GridLogMessage << "alpha = " << zalph << " beta "<<beta<<std::endl;
       const RealD tiny = 1.0e-20;
       if ( beta < tiny ) { 
-	std::cout << " beta is tiny "<<beta<<std::endl;
+	std::cout<<GridLogMessage << " beta is tiny "<<beta<<std::endl;
      }
       lmd[k] = alph;
       lme[k]  = beta;
@@ -253,6 +259,7 @@ public:
     }
 
 #ifdef USE_LAPACK
+#define LAPACK_INT long long
     void diagonalize_lapack(DenseVector<RealD>& lmd,
 		     DenseVector<RealD>& lme, 
 		     int N1,
@@ -262,7 +269,7 @@ public:
   const int size = Nm;
 //  tevals.resize(size);
 //  tevecs.resize(size);
-  int NN = N1;
+  LAPACK_INT NN = N1;
   double evals_tmp[NN];
   double evec_tmp[NN][NN];
   memset(evec_tmp[0],0,sizeof(double)*NN*NN);
@@ -276,19 +283,19 @@ public:
         if (i==j) evals_tmp[i] = lmd[i];
         if (j==(i-1)) EE[j] = lme[j];
       }
-  int evals_found;
-  int lwork = ( (18*NN) > (1+4*NN+NN*NN)? (18*NN):(1+4*NN+NN*NN)) ;
-  int liwork =  3+NN*10 ;
-  int iwork[liwork];
+  LAPACK_INT evals_found;
+  LAPACK_INT lwork = ( (18*NN) > (1+4*NN+NN*NN)? (18*NN):(1+4*NN+NN*NN)) ;
+  LAPACK_INT liwork =  3+NN*10 ;
+  LAPACK_INT iwork[liwork];
   double work[lwork];
-  int isuppz[2*NN];
+  LAPACK_INT isuppz[2*NN];
   char jobz = 'V'; // calculate evals & evecs
   char range = 'I'; // calculate all evals
   //    char range = 'A'; // calculate all evals
   char uplo = 'U'; // refer to upper half of original matrix
   char compz = 'I'; // Compute eigenvectors of tridiagonal matrix
   int ifail[NN];
-  int info;
+  long long info;
 //  int total = QMP_get_number_of_nodes();
 //  int node = QMP_get_node_number();
 //  GridBase *grid = evec[0]._grid;
@@ -296,14 +303,18 @@ public:
   int node = grid->_processor;
   int interval = (NN/total)+1;
   double vl = 0.0, vu = 0.0;
-  int il = interval*node+1 , iu = interval*(node+1);
+  LAPACK_INT il = interval*node+1 , iu = interval*(node+1);
   if (iu > NN)  iu=NN;
   double tol = 0.0;
     if (1) {
       memset(evals_tmp,0,sizeof(double)*NN);
       if ( il <= NN){
         printf("total=%d node=%d il=%d iu=%d\n",total,node,il,iu);
+#ifdef USE_MKL
+        dstegr(&jobz, &range, &NN,
+#else
         LAPACK_dstegr(&jobz, &range, &NN,
+#endif
             (double*)DD, (double*)EE,
             &vl, &vu, &il, &iu, // these four are ignored if second parameteris 'A'
             &tol, // tolerance
@@ -335,6 +346,7 @@ public:
       lmd [NN-1-i]=evals_tmp[i];
   }
 }
+#undef LAPACK_INT 
 #endif
 
 
@@ -365,12 +377,14 @@ public:
 //	diagonalize_lapack(lmd2,lme2,Nm2,Nm,Qt,grid);
 #endif
 
-      int Niter = 100*N1;
+      int Niter = 10000*N1;
       int kmin = 1;
       int kmax = N2;
       // (this should be more sophisticated)
 
-      for(int iter=0; iter<Niter; ++iter){
+      for(int iter=0; ; ++iter){
+      if ( (iter+1)%(100*N1)==0) 
+      std::cout<<GridLogMessage << "[QL method] Not converged - iteration "<<iter+1<<"\n";
 
 	// determination of 2x2 leading submatrix
 	RealD dsub = lmd[kmax-1]-lmd[kmax-2];
@@ -399,11 +413,11 @@ public:
         _sort.push(lmd3,N2);
         _sort.push(lmd2,N2);
          for(int k=0; k<N2; ++k){
-	    if (fabs(lmd2[k] - lmd3[k]) >SMALL)  std::cout <<"lmd(qr) lmd(lapack) "<< k << ": " << lmd2[k] <<" "<< lmd3[k] <<std::endl;
-//	    if (fabs(lme2[k] - lme[k]) >SMALL)  std::cout <<"lme(qr)-lme(lapack) "<< k << ": " << lme2[k] - lme[k] <<std::endl;
+	    if (fabs(lmd2[k] - lmd3[k]) >SMALL)  std::cout<<GridLogMessage <<"lmd(qr) lmd(lapack) "<< k << ": " << lmd2[k] <<" "<< lmd3[k] <<std::endl;
+//	    if (fabs(lme2[k] - lme[k]) >SMALL)  std::cout<<GridLogMessage <<"lme(qr)-lme(lapack) "<< k << ": " << lme2[k] - lme[k] <<std::endl;
 	  }
          for(int k=0; k<N1*N1; ++k){
-//	    if (fabs(Qt2[k] - Qt[k]) >SMALL)  std::cout <<"Qt(qr)-Qt(lapack) "<< k << ": " << Qt2[k] - Qt[k] <<std::endl;
+//	    if (fabs(Qt2[k] - Qt[k]) >SMALL)  std::cout<<GridLogMessage <<"Qt(qr)-Qt(lapack) "<< k << ": " << Qt2[k] - Qt[k] <<std::endl;
 	}
     }
 #endif
@@ -418,7 +432,7 @@ public:
 	  }
 	}
       }
-      std::cout << "[QL method] Error - Too many iteration: "<<Niter<<"\n";
+      std::cout<<GridLogMessage << "[QL method] Error - Too many iteration: "<<Niter<<"\n";
       abort();
     }
 
@@ -435,6 +449,7 @@ public:
 		       DenseVector<Field>& evec,
 		       int k)
     {
+      double t0=-usecond()/1e6;
       typedef typename Field::scalar_type MyComplex;
       MyComplex ip;
 
@@ -453,6 +468,8 @@ public:
 	w = w - ip * evec[j];
       }
       normalise(w);
+      t0+=usecond()/1e6;
+      OrthoTime +=t0;
     }
 
     void setUnit_Qt(int Nm, DenseVector<RealD> &Qt) {
@@ -486,10 +503,10 @@ until convergence
 	GridBase *grid = evec[0]._grid;
 	assert(grid == src._grid);
 
-	std::cout << " -- Nk = " << Nk << " Np = "<< Np << std::endl;
-	std::cout << " -- Nm = " << Nm << std::endl;
-	std::cout << " -- size of eval   = " << eval.size() << std::endl;
-	std::cout << " -- size of evec  = " << evec.size() << std::endl;
+	std::cout<<GridLogMessage << " -- Nk = " << Nk << " Np = "<< Np << std::endl;
+	std::cout<<GridLogMessage << " -- Nm = " << Nm << std::endl;
+	std::cout<<GridLogMessage << " -- size of eval   = " << eval.size() << std::endl;
+	std::cout<<GridLogMessage << " -- size of evec  = " << evec.size() << std::endl;
 	
 	assert(Nm == evec.size() && Nm == eval.size());
 	
@@ -500,6 +517,7 @@ until convergence
 	DenseVector<int>   Iconv(Nm);
 
 	DenseVector<Field>  B(Nm,grid); // waste of space replicating
+//	DenseVector<Field>  Btemp(Nm,grid); // waste of space replicating
 	
 	Field f(grid);
 	Field v(grid);
@@ -515,35 +533,48 @@ until convergence
 	// (uniform vector) Why not src??
 	//	evec[0] = 1.0;
 	evec[0] = src;
-	std:: cout <<"norm2(src)= " << norm2(src)<<std::endl;
+	std:: cout<<GridLogMessage <<"norm2(src)= " << norm2(src)<<std::endl;
 // << src._grid  << std::endl;
 	normalise(evec[0]);
-	std:: cout <<"norm2(evec[0])= " << norm2(evec[0]) <<std::endl;
+	std:: cout<<GridLogMessage <<"norm2(evec[0])= " << norm2(evec[0]) <<std::endl;
 // << evec[0]._grid << std::endl;
 	
 	// Initial Nk steps
+	OrthoTime=0.;
+	double t0=usecond()/1e6;
 	for(int k=0; k<Nk; ++k) step(eval,lme,evec,f,Nm,k);
-//	std:: cout <<"norm2(evec[1])= " << norm2(evec[1]) << std::endl;
-//	std:: cout <<"norm2(evec[2])= " << norm2(evec[2]) << std::endl;
+	double t1=usecond()/1e6;
+	std::cout<<GridLogMessage <<"IRL::Initial steps: "<<t1-t0<< "seconds"<<std::endl; t0=t1;
+	std::cout<<GridLogMessage <<"IRL::Initial steps:OrthoTime "<<OrthoTime<< "seconds"<<std::endl;
+//	std:: cout<<GridLogMessage <<"norm2(evec[1])= " << norm2(evec[1]) << std::endl;
+//	std:: cout<<GridLogMessage <<"norm2(evec[2])= " << norm2(evec[2]) << std::endl;
 	RitzMatrix(evec,Nk);
+	t1=usecond()/1e6;
+	std::cout<<GridLogMessage <<"IRL::RitzMatrix: "<<t1-t0<< "seconds"<<std::endl; t0=t1;
 	for(int k=0; k<Nk; ++k){
-//	std:: cout <<"eval " << k << " " <<eval[k] << std::endl;
-//	std:: cout <<"lme " << k << " " << lme[k] << std::endl;
+//	std:: cout<<GridLogMessage <<"eval " << k << " " <<eval[k] << std::endl;
+//	std:: cout<<GridLogMessage <<"lme " << k << " " << lme[k] << std::endl;
 	}
 
 	// Restarting loop begins
 	for(int iter = 0; iter<Niter; ++iter){
 
-	  std::cout<<"\n Restart iteration = "<< iter << std::endl;
+	  std::cout<<GridLogMessage<<"\n Restart iteration = "<< iter << std::endl;
 
 	  // 
 	  // Rudy does a sort first which looks very different. Getting fed up with sorting out the algo defs.
 	  // We loop over 
 	  //
+	OrthoTime=0.;
 	  for(int k=Nk; k<Nm; ++k) step(eval,lme,evec,f,Nm,k);
+	t1=usecond()/1e6;
+	std::cout<<GridLogMessage <<"IRL:: "<<Np <<" steps: "<<t1-t0<< "seconds"<<std::endl; t0=t1;
+	std::cout<<GridLogMessage <<"IRL::Initial steps:OrthoTime "<<OrthoTime<< "seconds"<<std::endl;
 	  f *= lme[Nm-1];
 
 	  RitzMatrix(evec,k2);
+	t1=usecond()/1e6;
+	std::cout<<GridLogMessage <<"IRL:: RitzMatrix: "<<t1-t0<< "seconds"<<std::endl; t0=t1;
 	  
 	  // getting eigenvalues
 	  for(int k=0; k<Nm; ++k){
@@ -552,18 +583,27 @@ until convergence
 	  }
 	  setUnit_Qt(Nm,Qt);
 	  diagonalize(eval2,lme2,Nm,Nm,Qt,grid);
+	t1=usecond()/1e6;
+	std::cout<<GridLogMessage <<"IRL:: diagonalize: "<<t1-t0<< "seconds"<<std::endl; t0=t1;
 
 	  // sorting
 	  _sort.push(eval2,Nm);
+	t1=usecond()/1e6;
+	std::cout<<GridLogMessage <<"IRL:: eval sorting: "<<t1-t0<< "seconds"<<std::endl; t0=t1;
 	  
 	  // Implicitly shifted QR transformations
 	  setUnit_Qt(Nm,Qt);
+	  for(int ip=0; ip<k2; ++ip){
+	std::cout<<GridLogMessage << "eval "<< ip << " "<< eval2[ip] << std::endl;
+	}
 	  for(int ip=k2; ip<Nm; ++ip){ 
-	std::cout << "qr_decomp "<< ip << " "<< eval2[ip] << std::endl;
+	std::cout<<GridLogMessage << "qr_decomp "<< ip << " "<< eval2[ip] << std::endl;
 	    qr_decomp(eval,lme,Nm,Nm,Qt,eval2[ip],k1,Nm);
 		
 	}
-    
+	t1=usecond()/1e6;
+	std::cout<<GridLogMessage <<"IRL::qr_decomp: "<<t1-t0<< "seconds"<<std::endl; t0=t1;
+if (0) {  
 	  for(int i=0; i<(Nk+1); ++i) B[i] = 0.0;
 	  
 	  for(int j=k1-1; j<k2+1; ++j){
@@ -572,14 +612,38 @@ until convergence
 	      B[j] += Qt[k+Nm*j] * evec[k];
 	    }
 	  }
-	  for(int j=k1-1; j<k2+1; ++j) evec[j] = B[j];
+	t1=usecond()/1e6;
+	std::cout<<GridLogMessage <<"IRL::QR Rotate: "<<t1-t0<< "seconds"<<std::endl; t0=t1;
+}
+
+if (1) {
+	for(int i=0; i<(Nk+1); ++i) {
+		B[i] = 0.0;
+	  	B[i].checkerboard = evec[0].checkerboard;
+	}
+
+	int j_block = 24; int k_block=24;
+PARALLEL_FOR_LOOP
+	for(int ss=0;ss < grid->oSites();ss++){
+	for(int jj=k1-1; jj<k2+1; jj += j_block)
+	for(int kk=0; kk<Nm; kk += k_block)
+	for(int j=jj; (j<(k2+1)) && j<(jj+j_block); ++j){
+	for(int k=kk; (k<Nm) && k<(kk+k_block) ; ++k){
+	    B[j]._odata[ss] +=Qt[k+Nm*j] * evec[k]._odata[ss]; 
+	}
+	}
+	}
+	t1=usecond()/1e6;
+	std::cout<<GridLogMessage <<"IRL::QR rotation: "<<t1-t0<< "seconds"<<std::endl; t0=t1;
+}
+	for(int j=k1-1; j<k2+1; ++j) evec[j] = B[j];
 
 	  // Compressed vector f and beta(k2)
 	  f *= Qt[Nm-1+Nm*(k2-1)];
 	  f += lme[k2-1] * evec[k2];
 	  beta_k = norm2(f);
 	  beta_k = sqrt(beta_k);
-	  std::cout<<" beta(k) = "<<beta_k<<std::endl;
+	  std::cout<<GridLogMessage<<" beta(k) = "<<beta_k<<std::endl;
 
 	  RealD betar = 1.0/beta_k;
 	  evec[k2] = betar * f;
@@ -592,7 +656,10 @@ until convergence
 	  }
 	  setUnit_Qt(Nm,Qt);
 	  diagonalize(eval2,lme2,Nk,Nm,Qt,grid);
+	t1=usecond()/1e6;
+	std::cout<<GridLogMessage <<"IRL::diagonalize: "<<t1-t0<< "seconds"<<std::endl; t0=t1;
 	  
+if (0) {
 	  for(int k = 0; k<Nk; ++k) B[k]=0.0;
 	  
 	  for(int j = 0; j<Nk; ++j){
@@ -600,12 +667,34 @@ until convergence
 	    B[j].checkerboard = evec[k].checkerboard;
 	      B[j] += Qt[k+j*Nm] * evec[k];
 	    }
-//	    std::cout << "norm(B["<<j<<"])="<<norm2(B[j])<<std::endl;
+	    std::cout<<GridLogMessage << "norm(B["<<j<<"])="<<norm2(B[j])<<std::endl;
 	  }
-//	_sort.push(eval2,B,Nk);
+	t1=usecond()/1e6;
+	std::cout<<GridLogMessage <<"IRL::Convergence rotation: "<<t1-t0<< "seconds"<<std::endl; t0=t1;
+}
+if (1) {
+	for(int i=0; i<(Nk+1); ++i) {
+		B[i] = 0.0;
+	  	B[i].checkerboard = evec[0].checkerboard;
+	}
+
+	int j_block = 24; int k_block=24;
+PARALLEL_FOR_LOOP
+	for(int ss=0;ss < grid->oSites();ss++){
+	for(int jj=0; jj<Nk; jj += j_block)
+	for(int kk=0; kk<Nk; kk += k_block)
+	for(int j=jj; (j<Nk) && j<(jj+j_block); ++j){
+	for(int k=kk; (k<Nk) && k<(kk+k_block) ; ++k){
+	    B[j]._odata[ss] +=Qt[k+Nm*j] * evec[k]._odata[ss]; 
+	}
+	}
+	}
+	t1=usecond()/1e6;
+	std::cout<<GridLogMessage <<"IRL::convergence rotation : "<<t1-t0<< "seconds"<<std::endl; t0=t1;
+}
 
 	  Nconv = 0;
-	  //	  std::cout << std::setiosflags(std::ios_base::scientific);
+	  //	  std::cout<<GridLogMessage << std::setiosflags(std::ios_base::scientific);
 	  for(int i=0; i<Nk; ++i){
 
 //	    _poly(_Linop,B[i],v);
@@ -613,14 +702,16 @@ until convergence
 	    
 	    RealD vnum = real(innerProduct(B[i],v)); // HermOp.
 	    RealD vden = norm2(B[i]);
+	    RealD vv0 = norm2(v);
 	    eval2[i] = vnum/vden;
 	    v -= eval2[i]*B[i];
 	    RealD vv = norm2(v);
 	    
 	    std::cout.precision(13);
-	    std::cout << "[" << std::setw(3)<< std::setiosflags(std::ios_base::right) <<i<<"] ";
-	    std::cout << "eval = "<<std::setw(25)<< std::setiosflags(std::ios_base::left)<< eval2[i];
-	    std::cout <<" |H B[i] - eval[i]B[i]|^2 "<< std::setw(25)<< std::setiosflags(std::ios_base::right)<< vv<< std::endl;
+	    std::cout<<GridLogMessage << "[" << std::setw(3)<< std::setiosflags(std::ios_base::right) <<i<<"] ";
+	    std::cout<<"eval = "<<std::setw(25)<< std::setiosflags(std::ios_base::left)<< eval2[i];
+	    std::cout<<"|H B[i] - eval[i]B[i]|^2 "<< std::setw(25)<< std::setiosflags(std::ios_base::right)<< vv;
+	    std::cout<<" "<< vnum/(sqrt(vden)*sqrt(vv0)) << std::endl;
 	    
 	// change the criteria as evals are supposed to be sorted, all evals smaller(larger) than Nstop should have converged
 	    if((vv<eresid*eresid) && (i == Nconv) ){
@@ -629,17 +720,19 @@ until convergence
 	    }
 
 	  }  // i-loop end
-	  //	  std::cout << std::resetiosflags(std::ios_base::scientific);
+	  //	  std::cout<<GridLogMessage << std::resetiosflags(std::ios_base::scientific);
+	t1=usecond()/1e6;
+	std::cout<<GridLogMessage <<"IRL::convergence testing: "<<t1-t0<< "seconds"<<std::endl; t0=t1;
 
 
-	  std::cout<<" #modes converged: "<<Nconv<<std::endl;
+	  std::cout<<GridLogMessage<<" #modes converged: "<<Nconv<<std::endl;
 
 	  if( Nconv>=Nstop ){
 	    goto converged;
 	  }
 	} // end of iter loop
 	
-	std::cout<<"\n NOT converged.\n";
+	std::cout<<GridLogMessage<<"\n NOT converged.\n";
 	abort();
 	
       converged:
@@ -652,10 +745,10 @@ until convergence
        }
       _sort.push(eval,evec,Nconv);
 
-      std::cout << "\n Converged\n Summary :\n";
-      std::cout << " -- Iterations  = "<< Nconv  << "\n";
-      std::cout << " -- beta(k)     = "<< beta_k << "\n";
-      std::cout << " -- Nconv       = "<< Nconv  << "\n";
+      std::cout<<GridLogMessage << "\n Converged\n Summary :\n";
+      std::cout<<GridLogMessage << " -- Iterations  = "<< Nconv  << "\n";
+      std::cout<<GridLogMessage << " -- beta(k)     = "<< beta_k << "\n";
+      std::cout<<GridLogMessage << " -- Nconv       = "<< Nconv  << "\n";
      }
 
     /////////////////////////////////////////////////
@@ -678,25 +771,25 @@ until convergence
 	}
       }
 
-      std::cout<<"Lanczos_Factor start/end " <<start <<"/"<<end<<std::endl;
+      std::cout<<GridLogMessage<<"Lanczos_Factor start/end " <<start <<"/"<<end<<std::endl;
 
       // Starting from scratch, bq[0] contains a random vector and |bq[0]| = 1
       int first;
       if(start == 0){
 
-	std::cout << "start == 0\n"; //TESTING
+	std::cout<<GridLogMessage << "start == 0\n"; //TESTING
 
 	_poly(_Linop,bq[0],bf);
 
 	alpha = real(innerProduct(bq[0],bf));//alpha =  bq[0]^dag A bq[0]
 
-	std::cout << "alpha = " << alpha << std::endl;
+	std::cout<<GridLogMessage << "alpha = " << alpha << std::endl;
 	
 	bf = bf - alpha * bq[0];  //bf =  A bq[0] - alpha bq[0]
 
 	H[0][0]=alpha;
 
-	std::cout << "Set H(0,0) to " << H[0][0] << std::endl;
+	std::cout<<GridLogMessage << "Set H(0,0) to " << H[0][0] << std::endl;
 
 	first = 1;
 
@@ -716,19 +809,19 @@ until convergence
 
 	beta = 0;sqbt = 0;
 
-	std::cout << "cont is true so setting beta to zero\n";
+	std::cout<<GridLogMessage << "cont is true so setting beta to zero\n";
 
       }	else {
 
 	beta = norm2(bf);
 	sqbt = sqrt(beta);
 
-	std::cout << "beta = " << beta << std::endl;
+	std::cout<<GridLogMessage << "beta = " << beta << std::endl;
       }
 
       for(int j=first;j<end;j++){
 
-	std::cout << "Factor j " << j <<std::endl;
+	std::cout<<GridLogMessage << "Factor j " << j <<std::endl;
 
 	if(cont){ // switches to factoring; understand start!=0 and initial bf value is right.
 	  bq[j] = bf; cont = false;
@@ -751,7 +844,7 @@ until convergence
 
 	beta = fnorm;
 	sqbt = sqrt(beta);
-	std::cout << "alpha = " << alpha << " fnorm = " << fnorm << '\n';
+	std::cout<<GridLogMessage << "alpha = " << alpha << " fnorm = " << fnorm << '\n';
 
 	///Iterative refinement of orthogonality V = [ bq[0]  bq[1]  ...  bq[M] ]
 	int re = 0;
@@ -786,8 +879,8 @@ until convergence
 	  bck = sqrt( nmbex );
 	  re++;
 	}
-	std::cout << "Iteratively refined orthogonality, changes alpha\n";
-	if(re > 1) std::cout << "orthagonality refined " << re << " times" <<std::endl;
+	std::cout<<GridLogMessage << "Iteratively refined orthogonality, changes alpha\n";
+	if(re > 1) std::cout<<GridLogMessage << "orthagonality refined " << re << " times" <<std::endl;
 	H[j][j]=alpha;
       }
 
@@ -802,11 +895,13 @@ until convergence
 
     void ImplicitRestart(int TM, DenseVector<RealD> &evals,  DenseVector<DenseVector<RealD> > &evecs, DenseVector<Field> &bq, Field &bf, int cont)
     {
-      std::cout << "ImplicitRestart begin. Eigensort starting\n";
+      std::cout<<GridLogMessage << "ImplicitRestart begin. Eigensort starting\n";
 
       DenseMatrix<RealD> H; Resize(H,Nm,Nm);
 
+#ifndef USE_LAPACK
       EigenSort(evals, evecs);
+#endif
 
       ///Assign shifts
       int K=Nk;
@@ -829,15 +924,15 @@ until convergence
       /// Shifted H defines a new K step Arnoldi factorization
       RealD  beta = H[ff][ff-1]; 
       RealD  sig  = Q[TM - 1][ff - 1];
-      std::cout << "beta = " << beta << " sig = " << real(sig) <<std::endl;
+      std::cout<<GridLogMessage << "beta = " << beta << " sig = " << real(sig) <<std::endl;
 
-      std::cout << "TM = " << TM << " ";
-      std::cout << norm2(bq[0]) << " -- before" <<std::endl;
+      std::cout<<GridLogMessage << "TM = " << TM << " ";
+      std::cout<<GridLogMessage << norm2(bq[0]) << " -- before" <<std::endl;
 
       /// q -> q Q
       times_real(bq, Q, TM);
 
-      std::cout << norm2(bq[0]) << " -- after " << ff <<std::endl;
+      std::cout<<GridLogMessage << norm2(bq[0]) << " -- after " << ff <<std::endl;
       bf =  beta* bq[ff] + sig* bf;
 
       /// Do the rest of the factorization
@@ -861,7 +956,7 @@ until convergence
       int ff = Lanczos_Factor(0, M, cont, bq,bf,H); // 0--M to begin with
 
       if(ff < M) {
-	std::cout << "Krylov: aborting ff "<<ff <<" "<<M<<std::endl;
+	std::cout<<GridLogMessage << "Krylov: aborting ff "<<ff <<" "<<M<<std::endl;
 	abort(); // Why would this happen?
       }
 
@@ -870,7 +965,7 @@ until convergence
 
       for(int it = 0; it < Niter && (converged < Nk); ++it) {
 
-	std::cout << "Krylov: Iteration --> " << it << std::endl;
+	std::cout<<GridLogMessage << "Krylov: Iteration --> " << it << std::endl;
 	int lock_num = lock ? converged : 0;
 	DenseVector<RealD> tevals(M - lock_num );
 	DenseMatrix<RealD> tevecs; Resize(tevecs,M - lock_num,M - lock_num);
@@ -886,7 +981,7 @@ until convergence
       Wilkinson<RealD>(H, evals, evecs, small); 
       //      Check();
 
-      std::cout << "Done  "<<std::endl;
+      std::cout<<GridLogMessage << "Done  "<<std::endl;
 
     }
 
@@ -951,7 +1046,7 @@ until convergence
 		  DenseVector<RealD> &tevals, DenseVector<DenseVector<RealD> > &tevecs, 
 		  int lock, int converged)
     {
-      std::cout << "Converged " << converged << " so far." << std::endl;
+      std::cout<<GridLogMessage << "Converged " << converged << " so far." << std::endl;
       int lock_num = lock ? converged : 0;
       int M = Nm;
 
@@ -966,7 +1061,9 @@ until convergence
       RealD small=1.0e-16;
       Wilkinson<RealD>(AH, tevals, tevecs, small);
 
+#ifndef USE_LAPACK
       EigenSort(tevals, tevecs);
+#endif
 
       RealD resid_nrm=  norm2(bf);
 
@@ -977,7 +1074,7 @@ until convergence
 	RealD diff = 0;
 	diff = abs( tevecs[i][Nm - 1 - lock_num] ) * resid_nrm;
 
-	std::cout << "residual estimate " << SS-1-i << " " << diff << " of (" << tevals[i] << ")" << std::endl;
+	std::cout<<GridLogMessage << "residual estimate " << SS-1-i << " " << diff << " of (" << tevals[i] << ")" << std::endl;
 
 	if(diff < converged) {
 
@@ -993,13 +1090,13 @@ until convergence
 	    lock_num++;
 	  }
 	  converged++;
-	  std::cout << " converged on eval " << converged << " of " << Nk << std::endl;
+	  std::cout<<GridLogMessage << " converged on eval " << converged << " of " << Nk << std::endl;
 	} else {
 	  break;
 	}
       }
 #endif
-      std::cout << "Got " << converged << " so far " <<std::endl;	
+      std::cout<<GridLogMessage << "Got " << converged << " so far " <<std::endl;	
     }
 
     ///Check
@@ -1008,7 +1105,9 @@ until convergence
 
       DenseVector<RealD> goodval(this->get);
 
+#ifndef USE_LAPACK
       EigenSort(evals,evecs);
+#endif
 
       int NM = Nm;
 
@@ -1080,10 +1179,10 @@ 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/cartesian/Cartesian_base.h

@@ -77,7 +77,7 @@ public:
     // GridCartesian / GridRedBlackCartesian
     ////////////////////////////////////////////////////////////////
     virtual int CheckerBoarded(int dim)=0;
-    virtual int CheckerBoard(std::vector<int> &site)=0;
+    virtual int CheckerBoard(const std::vector<int> &site)=0;
     virtual int CheckerBoardDestination(int source_cb,int shift,int dim)=0;
     virtual int CheckerBoardShift(int source_cb,int dim,int shift,int osite)=0;
     virtual int CheckerBoardShiftForCB(int source_cb,int dim,int shift,int cb)=0;

lib/cartesian/Cartesian_full.h

@@ -49,7 +49,7 @@ public:
     virtual int CheckerBoarded(int dim){
       return 0;
     }
-    virtual int CheckerBoard(std::vector<int> &site){
+    virtual int CheckerBoard(const std::vector<int> &site){
         return 0;
     }
     virtual int CheckerBoardDestination(int cb,int shift,int dim){

lib/cartesian/Cartesian_red_black.h

@@ -49,7 +49,7 @@ public:
       if( dim==_checker_dim) return 1;
       else return 0;
     }
-    virtual int CheckerBoard(std::vector<int> &site){
+    virtual int CheckerBoard(const std::vector<int> &site){
       int linear=0;
       assert(site.size()==_ndimension);
       for(int d=0;d<_ndimension;d++){ 

lib/communicator/Communicator_base.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>
 namespace Grid {
 
 ///////////////////////////////////////////////////////////////
@@ -65,7 +65,6 @@ const std::vector<int> & CartesianCommunicator::ThisProcessorCoor(void) { return
 const std::vector<int> & CartesianCommunicator::ProcessorGrid(void)     { return _processors; };
 int                      CartesianCommunicator::ProcessorCount(void)    { return _Nprocessors; };
 
-
 ////////////////////////////////////////////////////////////////////////////////
 // very VERY rarely (Log, serial RNG) we need world without a grid
 ////////////////////////////////////////////////////////////////////////////////
@@ -90,11 +89,11 @@ void CartesianCommunicator::GlobalSumVector(ComplexD *c,int N)
 #if !defined( GRID_COMMS_MPI3) && !defined (GRID_COMMS_MPI3L)
 
 void CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsRequest_t> &list,
-                                                       void *xmit,
-                                                       int xmit_to_rank,
-                                                       void *recv,
-                                                       int recv_from_rank,
-                                                       int bytes)
+						       void *xmit,
+						       int xmit_to_rank,
+						       void *recv,
+						       int recv_from_rank,
+						       int bytes)
 {
   SendToRecvFromBegin(list,xmit,xmit_to_rank,recv,recv_from_rank,bytes);
 }

lib/communicator/Communicator_base.h

@@ -68,8 +68,6 @@ class CartesianCommunicator {
   static MPI_Comm communicator_world;
          MPI_Comm communicator;
   typedef MPI_Request CommsRequest_t;
-  static char name[MPI_MAX_PROCESSOR_NAME]; // processing node physical name
-  static int length; 
 #else 
   typedef int CommsRequest_t;
 #endif
@@ -151,7 +149,6 @@ class CartesianCommunicator {
   const std::vector<int> & ProcessorGrid(void)     ;
   int                      ProcessorCount(void)    ;
 
-  void        		 PrintRankInfo(void)     ;
   ////////////////////////////////////////////////////////////////////////////////
   // very VERY rarely (Log, serial RNG) we need world without a grid
   ////////////////////////////////////////////////////////////////////////////////

lib/communicator/Communicator_mpi.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>
 #include <mpi.h>
 
 namespace Grid {
@@ -35,8 +35,6 @@ namespace Grid {
 // Info that is setup once and indept of cartesian layout
 ///////////////////////////////////////////////////////////////////////////////////////////////////
 MPI_Comm CartesianCommunicator::communicator_world;
-char CartesianCommunicator::name[MPI_MAX_PROCESSOR_NAME]; // processing node physical name
-int CartesianCommunicator::length; 
 
 // Should error check all MPI calls.
 void CartesianCommunicator::Init(int *argc, char ***argv) {
@@ -46,8 +44,6 @@ void CartesianCommunicator::Init(int *argc, char ***argv) {
     MPI_Init(argc,argv);
   }
   MPI_Comm_dup (MPI_COMM_WORLD,&communicator_world);
-
-  MPI_Get_processor_name(name, &length);
   ShmInitGeneric();
 }
 
@@ -210,10 +206,5 @@ void CartesianCommunicator::BroadcastWorld(int root,void* data, int bytes)
   assert(ierr==0);
 }
 
-void CartesianCommunicator::PrintRankInfo(){
-    std::cout << "Grid: Rank "<< _processor << "  -  Physical node name: " << name << std::endl;
 }
 
-
-}// end of namespace
-

lib/communicator/Communicator_mpi3.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>
 #include <mpi.h>
 
 namespace Grid {
@@ -576,10 +576,5 @@ void CartesianCommunicator::BroadcastWorld(int root,void* data, int bytes)
   assert(ierr==0);
 }
 
-  void CartesianCommunicator::PrintRankInfo(){
-    std::cout << "Grid: Rank "<< _processor << "  -  Physical node name: " << name << std::endl;
-  }
-
-  
 }
 

lib/communicator/Communicator_mpi3_leader.cc

@@ -869,9 +869,6 @@ void *CartesianCommunicator::ShmBufferTranslate(int rank,void * local_p) {
   return NULL;
 }
 
-  void CartesianCommunicator::PrintRankInfo(){
-    std::cout << "Grid: Rank "<< _processor << "  -  Physical node name: " << name << std::endl;
-  }
 
 };
 

lib/communicator/Communicator_none.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>
 namespace Grid {
 
 ///////////////////////////////////////////////////////////////////////////////////////////////////
@@ -37,11 +37,6 @@ void CartesianCommunicator::Init(int *argc, char *** arv)
   ShmInitGeneric();
 }
 
-void CartesianCommunicator::PrintRankInfo(){
-    std::cout << GridLogMessage << "No Rank Info available" << std::endl;
-}
-
-
 CartesianCommunicator::CartesianCommunicator(const std::vector<int> &processors)
 {
   _processors = processors;
@@ -65,10 +60,10 @@ void CartesianCommunicator::GlobalSum(uint64_t &){}
 void CartesianCommunicator::GlobalSumVector(double *,int N){}
 
 void CartesianCommunicator::SendRecvPacket(void *xmit,
-                                           void *recv,
-                                           int xmit_to_rank,
-                                           int recv_from_rank,
-                                           int bytes)
+					   void *recv,
+					   int xmit_to_rank,
+					   int recv_from_rank,
+					   int bytes)
 {
   assert(0);
 }
@@ -76,19 +71,19 @@ void CartesianCommunicator::SendRecvPacket(void *xmit,
 
 // Basic Halo comms primitive -- should never call in single node
 void CartesianCommunicator::SendToRecvFrom(void *xmit,
-                                           int dest,
-                                           void *recv,
-                                           int from,
-                                           int bytes)
+					   int dest,
+					   void *recv,
+					   int from,
+					   int bytes)
 {
   assert(0);
 }
 void CartesianCommunicator::SendToRecvFromBegin(std::vector<CommsRequest_t> &list,
-                                                void *xmit,
-                                                int dest,
-                                                void *recv,
-                                                int from,
-                                                int bytes)
+						void *xmit,
+						int dest,
+						void *recv,
+						int from,
+						int bytes)
 {
   assert(0);
 }

lib/communicator/Communicator_shmem.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>
 #include <mpp/shmem.h>
 
 namespace Grid {
@@ -333,9 +333,5 @@ void CartesianCommunicator::BroadcastWorld(int root,void* data, int bytes)
   }
 }
 
-  void CartesianCommunicator::PrintRankInfo(){
-    std::cout << GridLogMessage << "SHMEM: Rank Info not implemented yet" << std::endl;
-  }
-
 }
 

lib/lattice/Lattice_peekpoke.h

@@ -86,7 +86,7 @@ PARALLEL_FOR_LOOP
     // Poke a scalar object into the SIMD array
     //////////////////////////////////////////////////////
     template<class vobj,class sobj>
-    void pokeSite(const sobj &s,Lattice<vobj> &l,std::vector<int> &site){
+    void pokeSite(const sobj &s,Lattice<vobj> &l,const std::vector<int> &site){
 
       GridBase *grid=l._grid;
 
@@ -120,7 +120,7 @@ PARALLEL_FOR_LOOP
     // Peek a scalar object from the SIMD array
     //////////////////////////////////////////////////////////
     template<class vobj,class sobj>
-      void peekSite(sobj &s,const Lattice<vobj> &l,std::vector<int> &site){
+      void peekSite(sobj &s,const Lattice<vobj> &l,const std::vector<int> &site){
         
       GridBase *grid=l._grid;
 

lib/lattice/Lattice_reduction.h

@@ -31,87 +31,6 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 #define GRID_LATTICE_REDUCTION_H
 
 namespace Grid {
-
-template <class T>
-class ReproducibilityState {
- public:
-  typedef typename T::vector_type vector_type;
-  typedef std::vector<vector_type, alignedAllocator<vector_type> >  sum_type;
-  unsigned int  n_call;
-  bool          do_check;
-  bool          enable_reprocheck;
-  bool          success;
-  unsigned int  interval; 
-  std::vector<sum_type> th_states;
-
-  void reset_counter() { n_call = 0; }
-  void reset() {
-    th_states.clear();
-    do_check          = false;
-    enable_reprocheck = false;
-    success           = true;
-    n_call            = 0;
-  };
-
-  ReproducibilityState():interval(1) { 
-    reset(); 
-  }
-
-  void check(GridBase* grid, sum_type &sumarray){
-    ///////////////////////  Reproducibility section, not threaded on purpouse
-    if (enable_reprocheck) {
-      if (do_check && (n_call % interval) == 0) {
-        for (int thread = 0; thread < sumarray.size(); thread++) {
-          int words = sizeof(sumarray[thread])/sizeof(unsigned char);
-          unsigned char xors[words];
-          bitwise_xor(sumarray[thread], th_states[n_call][thread],xors);
-          // OR all words
-          unsigned char res = 0;
-          for (int w = 0; w < words; w++) res = res | xors[w];
-
-          Grid_unquiesce_nodes();
-          int rank = 0;
-          while (rank < grid->_Nprocessors){
-            if (rank == grid->ThisRank() ){
-              if ( res ) {
-                std::cout << "Reproducibility failure report" << std::endl;
-                grid->PrintRankInfo();
-                std::cout << "Call: "<< n_call << " Thread: " << thread << std::endl;
-                std::cout << "Size of states: " << th_states.size() << std::endl;
-                std::cout << std::setprecision(GRID_REAL_DIGITS+1) << std::scientific;
-                std::cout << "Saved partial sum  : " << th_states[n_call][thread] << std::endl;
-                std::cout << "Current partial sum: " << sumarray[thread] << std::endl;
-                std::cout << "Saved state "  << std::endl; show_binaryrep(th_states[n_call][thread]);
-                std::cout << "Current state" << std::endl; show_binaryrep(sumarray[thread]);
-                std::cout << "XOR result"    << std::endl; show_binaryrep(xors, words);
-              		//std::cout << std::defaultfloat;  //not supported by some compilers
-                std::cout << std::setprecision(6);
-                success = false; 
-              }
-            }
-            rank++;
-            grid->Barrier();
-          }
-          Grid_quiesce_nodes();                         
-        }
-
-      } else if (!do_check)
-      {
-        std::cout << GridLogDebug << "Saving thread state for inner product. Call n. " 
-        << n_call << std::endl;
-        th_states.resize(n_call+1);
-        th_states[n_call].resize(grid->SumArraySize());
-          th_states[n_call] = sumarray;  // save threads state
-          //n_call++;
-        }
-        n_call++;
-      }
-    }
-
-};
-
-
-
 #ifdef GRID_WARN_SUBOPTIMAL
 #warning "Optimisation alert all these reduction loops are NOT threaded "
 #endif     
@@ -120,25 +39,12 @@ class ReproducibilityState {
     // Deterministic Reduction operations
     ////////////////////////////////////////////////////////////////////////////////////////////////////
   template<class vobj> inline RealD norm2(const Lattice<vobj> &arg){
-        ReproducibilityState<vobj> repr;
-        ComplexD nrm = innerProduct(arg, arg, repr);
-        return std::real(nrm); 
-    }
-
-    template <class vobj>
-    inline RealD norm2(const Lattice<vobj> &arg, ReproducibilityState<vobj>& rep) {
-      ComplexD nrm = innerProduct(arg, arg, rep);
-      return std::real(nrm);
-    }
+    ComplexD nrm = innerProduct(arg,arg);
+    return std::real(nrm); 
+  }
 
     template<class vobj>
-    inline ComplexD innerProduct(const Lattice<vobj> &left,const Lattice<vobj> &right){
-      ReproducibilityState<vobj> repr;
-      return innerProduct(left, right, repr);
-    } 
-    
-    template<class vobj>
-    inline ComplexD innerProduct(const Lattice<vobj> &left,const Lattice<vobj> &right, ReproducibilityState<vobj>& repr) 
+    inline ComplexD innerProduct(const Lattice<vobj> &left,const Lattice<vobj> &right) 
     {
       typedef typename vobj::scalar_type scalar_type;
       typedef typename vobj::vector_type vector_type;
@@ -148,28 +54,24 @@ class ReproducibilityState {
 
       std::vector<vector_type,alignedAllocator<vector_type> > sumarray(grid->SumArraySize());
       for(int i=0;i<grid->SumArraySize();i++){
-        sumarray[i]=zero;
+	sumarray[i]=zero;
       }
 
-      PARALLEL_FOR_LOOP_STATIC //request statically scheduled threads for reproducibility
+PARALLEL_FOR_LOOP
       for(int thr=0;thr<grid->SumArraySize();thr++){
-        int nwork, mywork, myoff;
-        GridThread::GetWork(left._grid->oSites(),thr,mywork,myoff);
-        
-        decltype(innerProduct(left._odata[0],right._odata[0])) vnrm = zero; // private to thread; sub summation
+	int nwork, mywork, myoff;
+	GridThread::GetWork(left._grid->oSites(),thr,mywork,myoff);
+	
+	decltype(innerProduct(left._odata[0],right._odata[0])) vnrm=zero; // private to thread; sub summation
         for(int ss=myoff;ss<mywork+myoff; ss++){
-          vnrm = vnrm + innerProduct(left._odata[ss],right._odata[ss]);
-        }
-        sumarray[thr]=TensorRemove(vnrm) ;
+	  vnrm = vnrm + innerProduct(left._odata[ss],right._odata[ss]);
+	}
+	sumarray[thr]=TensorRemove(vnrm) ;
       }
-      
-      /////////// Reproducibility
-      repr.check(grid, sumarray);
-      ///////////////////////////
-
+    
       vector_type vvnrm; vvnrm=zero;  // sum across threads
       for(int i=0;i<grid->SumArraySize();i++){
-        vvnrm = vvnrm+sumarray[i];
+	vvnrm = vvnrm+sumarray[i];
       } 
       nrm = Reduce(vvnrm);// sum across simd
       right._grid->GlobalSum(nrm);
@@ -177,26 +79,26 @@ class ReproducibilityState {
     }
 
     template<class Op,class T1>
-    inline auto sum(const LatticeUnaryExpression<Op,T1> & expr)
-    ->typename decltype(expr.first.func(eval(0,std::get<0>(expr.second))))::scalar_object
+      inline auto sum(const LatticeUnaryExpression<Op,T1> & expr)
+      ->typename decltype(expr.first.func(eval(0,std::get<0>(expr.second))))::scalar_object
     {
       return sum(closure(expr));
     }
 
     template<class Op,class T1,class T2>
-    inline auto sum(const LatticeBinaryExpression<Op,T1,T2> & expr)
-    ->typename decltype(expr.first.func(eval(0,std::get<0>(expr.second)),eval(0,std::get<1>(expr.second))))::scalar_object
+      inline auto sum(const LatticeBinaryExpression<Op,T1,T2> & expr)
+      ->typename decltype(expr.first.func(eval(0,std::get<0>(expr.second)),eval(0,std::get<1>(expr.second))))::scalar_object
     {
       return sum(closure(expr));
     }
 
 
     template<class Op,class T1,class T2,class T3>
-    inline auto sum(const LatticeTrinaryExpression<Op,T1,T2,T3> & expr)
-    ->typename decltype(expr.first.func(eval(0,std::get<0>(expr.second)),
-     eval(0,std::get<1>(expr.second)),
-     eval(0,std::get<2>(expr.second))
-     ))::scalar_object
+      inline auto sum(const LatticeTrinaryExpression<Op,T1,T2,T3> & expr)
+      ->typename decltype(expr.first.func(eval(0,std::get<0>(expr.second)),
+				 eval(0,std::get<1>(expr.second)),
+				 eval(0,std::get<2>(expr.second))
+				 ))::scalar_object
     {
       return sum(closure(expr));
     }
@@ -209,24 +111,24 @@ class ReproducibilityState {
 
       std::vector<vobj,alignedAllocator<vobj> > sumarray(grid->SumArraySize());
       for(int i=0;i<grid->SumArraySize();i++){
-        sumarray[i]=zero;
+	sumarray[i]=zero;
       }
 
-      PARALLEL_FOR_LOOP
+PARALLEL_FOR_LOOP
       for(int thr=0;thr<grid->SumArraySize();thr++){
-        int nwork, mywork, myoff;
-        GridThread::GetWork(grid->oSites(),thr,mywork,myoff);
+	int nwork, mywork, myoff;
+	GridThread::GetWork(grid->oSites(),thr,mywork,myoff);
 
-        vobj vvsum=zero;
+	vobj vvsum=zero;
         for(int ss=myoff;ss<mywork+myoff; ss++){
-          vvsum = vvsum + arg._odata[ss];
-        }
-        sumarray[thr]=vvsum;
+	  vvsum = vvsum + arg._odata[ss];
+	}
+	sumarray[thr]=vvsum;
       }
 
       vobj vsum=zero;  // sum across threads
       for(int i=0;i<grid->SumArraySize();i++){
-        vsum = vsum+sumarray[i];
+	vsum = vsum+sumarray[i];
       } 
 
       typedef typename vobj::scalar_object sobj;
@@ -236,7 +138,7 @@ class ReproducibilityState {
       extract(vsum,buf);
 
       for(int i=0;i<Nsimd;i++) ssum = ssum + buf[i];
-        arg._grid->GlobalSum(ssum);
+      arg._grid->GlobalSum(ssum);
 
       return ssum;
     }
@@ -244,23 +146,23 @@ class ReproducibilityState {
 
 
 template<class vobj> inline void sliceSum(const Lattice<vobj> &Data,std::vector<typename vobj::scalar_object> &result,int orthogdim)
-    {
-      typedef typename vobj::scalar_object sobj;
-      GridBase  *grid = Data._grid;
-      assert(grid!=NULL);
+{
+  typedef typename vobj::scalar_object sobj;
+  GridBase  *grid = Data._grid;
+  assert(grid!=NULL);
 
   // FIXME
   // std::cout<<GridLogMessage<<"WARNING ! SliceSum is unthreaded "<<grid->SumArraySize()<<" threads "<<std::endl;
 
-      const int    Nd = grid->_ndimension;
-      const int Nsimd = grid->Nsimd();
+  const int    Nd = grid->_ndimension;
+  const int Nsimd = grid->Nsimd();
 
-      assert(orthogdim >= 0);
-      assert(orthogdim < Nd);
+  assert(orthogdim >= 0);
+  assert(orthogdim < Nd);
 
-      int fd=grid->_fdimensions[orthogdim];
-      int ld=grid->_ldimensions[orthogdim];
-      int rd=grid->_rdimensions[orthogdim];
+  int fd=grid->_fdimensions[orthogdim];
+  int ld=grid->_ldimensions[orthogdim];
+  int rd=grid->_rdimensions[orthogdim];
 
   std::vector<vobj,alignedAllocator<vobj> > lvSum(rd); // will locally sum vectors first
   std::vector<sobj> lsSum(ld,zero); // sum across these down to scalars

lib/lattice/Lattice_rng.h

@@ -30,6 +30,7 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 #define GRID_LATTICE_RNG_H
 
 #include <random>
+#include <Grid/sitmo_rng/sitmo_prng_engine.hpp>
 
 namespace Grid {
 
@@ -114,10 +115,14 @@ namespace Grid {
     typedef uint64_t      RngStateType;
     typedef std::ranlux48 RngEngine;
     static const int RngStateCount = 15;
-#else
+#elif RNG_MT19937 
     typedef std::mt19937 RngEngine;
     typedef uint32_t     RngStateType;
     static const int     RngStateCount = std::mt19937::state_size;
+#elif RNG_SITMO
+    typedef sitmo::prng_engine 	RngEngine;
+    typedef uint64_t    	RngStateType;
+    static const int    	RngStateCount = 4;
 #endif
     std::vector<RngEngine>                             _generators;
     std::vector<std::uniform_real_distribution<RealD>> _uniform;

lib/lattice/Lattice_transfer.h

@@ -386,6 +386,7 @@ void InsertSlice(Lattice<vobj> &lowDim,Lattice<vobj> & higherDim,int slice, int
   }
 
   // the above should guarantee that the operations are local
+  // Guido: check the threading here
   //PARALLEL_FOR_LOOP
   for(int idx=0;idx<lg->lSites();idx++){
     std::vector<int> lcoor(nl);

lib/pugixml/pugixml.cc

@@ -14,7 +14,7 @@
 #ifndef SOURCE_PUGIXML_CPP
 #define SOURCE_PUGIXML_CPP
 
-#include <pugixml/pugixml.h>
+#include <Grid/pugixml/pugixml.h>
 
 #include <stdlib.h>
 #include <stdio.h>

lib/qcd/action/fermion/CayleyFermion5D.cc

@@ -29,7 +29,8 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
     *************************************************************************************/
     /*  END LEGAL */
 
-#include <Grid.h>
+#include <Grid/Eigen/Dense>
+#include <Grid/Grid.h>
 
 
 namespace Grid {
@@ -48,18 +49,19 @@ namespace QCD {
 		   FourDimGrid,
  	 	   FourDimRedBlackGrid,_M5,p),
    mass(_mass)
- { }
+ { 
+ }
 
 template<class Impl>  
 void CayleyFermion5D<Impl>::Dminus(const FermionField &psi, FermionField &chi)
 {
   int Ls=this->Ls;
-  FermionField tmp(psi._grid);
 
-  this->DW(psi,tmp,DaggerNo);
+  FermionField tmp_f(this->FermionGrid());
+  this->DW(psi,tmp_f,DaggerNo);
 
   for(int s=0;s<Ls;s++){
-    axpby_ssp(chi,Coeff_t(1.0),psi,-cs[s],tmp,s,s);// chi = (1-c[s] D_W) psi
+    axpby_ssp(chi,Coeff_t(1.0),psi,-cs[s],tmp_f,s,s);// chi = (1-c[s] D_W) psi
   }
 }
 
@@ -87,8 +89,8 @@ template<class Impl> void CayleyFermion5D<Impl>::CayleyReport(void)
     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
+    // Flops = MADD * Ls *Ls *4dvol * spin/colour/complex
+    RealD mflops = 2.0*24*this->Ls*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;
   }
@@ -110,12 +112,12 @@ template<class Impl>
 void CayleyFermion5D<Impl>::DminusDag(const FermionField &psi, FermionField &chi)
 {
   int Ls=this->Ls;
-  FermionField tmp(psi._grid);
 
-  this->DW(psi,tmp,DaggerYes);
+  FermionField tmp_f(this->FermionGrid());
+  this->DW(psi,tmp_f,DaggerYes);
 
   for(int s=0;s<Ls;s++){
-    axpby_ssp(chi,Coeff_t(1.0),psi,-cs[s],tmp,s,s);// chi = (1-c[s] D_W) psi
+    axpby_ssp(chi,Coeff_t(1.0),psi,-cs[s],tmp_f,s,s);// chi = (1-c[s] D_W) psi
   }
 }
 template<class Impl>  
@@ -138,6 +140,7 @@ void CayleyFermion5D<Impl>::Meooe5D    (const FermionField &psi, FermionField &D
   lower[0]   =-mass*lower[0];
   M5D(psi,psi,Din,lower,diag,upper);
 }
+// FIXME Redunant with the above routine; check this and eliminate
 template<class Impl> void CayleyFermion5D<Impl>::Meo5D     (const FermionField &psi, FermionField &chi)
 {
   int Ls=this->Ls;
@@ -259,36 +262,33 @@ template<class Impl>
 void CayleyFermion5D<Impl>::Meooe       (const FermionField &psi, FermionField &chi)
 {
   int Ls=this->Ls;
-  FermionField tmp(psi._grid);
 
-  Meooe5D(psi,tmp); 
+  Meooe5D(psi,this->tmp()); 
 
   if ( psi.checkerboard == Odd ) {
-    this->DhopEO(tmp,chi,DaggerNo);
+    this->DhopEO(this->tmp(),chi,DaggerNo);
   } else {
-    this->DhopOE(tmp,chi,DaggerNo);
+    this->DhopOE(this->tmp(),chi,DaggerNo);
   }
 }
 
 template<class Impl>
 void CayleyFermion5D<Impl>::MeooeDag    (const FermionField &psi, FermionField &chi)
 {
-  FermionField tmp(psi._grid);
   // Apply 4d dslash
   if ( psi.checkerboard == Odd ) {
-    this->DhopEO(psi,tmp,DaggerYes);
+    this->DhopEO(psi,this->tmp(),DaggerYes);
   } else {
-    this->DhopOE(psi,tmp,DaggerYes);
+    this->DhopOE(psi,this->tmp(),DaggerYes);
   }
-  MeooeDag5D(tmp,chi); 
+  MeooeDag5D(this->tmp(),chi); 
 }
 
 template<class Impl>
 void  CayleyFermion5D<Impl>::Mdir (const FermionField &psi, FermionField &chi,int dir,int disp){
-  FermionField tmp(psi._grid);
-  Meo5D(psi,tmp);
+  Meo5D(psi,this->tmp());
   // Apply 4d dslash fragment
-  this->DhopDir(tmp,chi,dir,disp);
+  this->DhopDir(this->tmp(),chi,dir,disp);
 }
 // force terms; five routines; default to Dhop on diagonal
 template<class Impl>
@@ -459,9 +459,91 @@ void CayleyFermion5D<Impl>::SetCoefficientsInternal(RealD zolo_hi,std::vector<Co
     for(int j=0;j<Ls-1;j++) delta_d *= cee[j]/bee[j];
     dee[Ls-1] += delta_d;
   }  
+
+  int inv=1;
+  this->MooeeInternalCompute(0,inv,MatpInv,MatmInv);
+  this->MooeeInternalCompute(1,inv,MatpInvDag,MatmInvDag);
+
 }
 
 
+template<class Impl>
+void CayleyFermion5D<Impl>::MooeeInternalCompute(int dag, int inv,
+						 Vector<iSinglet<Simd> > & Matp,
+						 Vector<iSinglet<Simd> > & Matm)
+{
+  int Ls=this->Ls;
+
+  GridBase *grid = this->FermionRedBlackGrid();
+  int LLs = grid->_rdimensions[0];
+
+  if ( LLs == Ls ) return; // Not vectorised in 5th direction
+
+  Eigen::MatrixXcd Pplus  = Eigen::MatrixXcd::Zero(Ls,Ls);
+  Eigen::MatrixXcd Pminus = Eigen::MatrixXcd::Zero(Ls,Ls);
+  
+  for(int s=0;s<Ls;s++){
+    Pplus(s,s) = bee[s];
+    Pminus(s,s)= bee[s];
+  }
+  
+  for(int s=0;s<Ls-1;s++){
+    Pminus(s,s+1) = -cee[s];
+  }
+  
+  for(int s=0;s<Ls-1;s++){
+    Pplus(s+1,s) = -cee[s+1];
+  }
+  Pplus (0,Ls-1) = mass*cee[0];
+  Pminus(Ls-1,0) = mass*cee[Ls-1];
+  
+  Eigen::MatrixXcd PplusMat ;
+  Eigen::MatrixXcd PminusMat;
+  
+  if ( inv ) {
+    PplusMat =Pplus.inverse();
+    PminusMat=Pminus.inverse();
+  } else { 
+    PplusMat =Pplus;
+    PminusMat=Pminus;
+  }
+  
+  if(dag){
+    PplusMat.adjointInPlace();
+    PminusMat.adjointInPlace();
+  }
+  
+  typedef typename SiteHalfSpinor::scalar_type scalar_type;
+  const int Nsimd=Simd::Nsimd();
+  Matp.resize(Ls*LLs);
+  Matm.resize(Ls*LLs);
+
+  for(int s2=0;s2<Ls;s2++){
+  for(int s1=0;s1<LLs;s1++){
+    int istride = LLs;
+    int ostride = 1;
+    Simd Vp;
+    Simd Vm;
+    scalar_type *sp = (scalar_type *)&Vp;
+    scalar_type *sm = (scalar_type *)&Vm;
+    for(int l=0;l<Nsimd;l++){
+      if ( switcheroo<Coeff_t>::iscomplex() ) {
+	sp[l] = PplusMat (l*istride+s1*ostride,s2);
+	sm[l] = PminusMat(l*istride+s1*ostride,s2);
+      } else { 
+      // if real
+	scalar_type tmp;
+	tmp = PplusMat (l*istride+s1*ostride,s2);
+	sp[l] = scalar_type(tmp.real(),tmp.real());
+	tmp = PminusMat(l*istride+s1*ostride,s2);
+	sm[l] = scalar_type(tmp.real(),tmp.real());
+      }
+    }
+    Matp[LLs*s2+s1] = Vp;
+    Matm[LLs*s2+s1] = Vm;
+  }}
+}
+
 
   FermOpTemplateInstantiate(CayleyFermion5D);
   GparityFermOpTemplateInstantiate(CayleyFermion5D);

lib/qcd/action/fermion/CayleyFermion5D.h

@@ -33,6 +33,31 @@ namespace Grid {
 
   namespace QCD {
 
+     template<typename T> struct switcheroo   {  
+       static inline int iscomplex()  { return 0; } 
+
+       template<class vec>
+       static inline vec mult(vec a, vec b) {
+	 return real_mult(a,b);
+       }
+     };
+     template<> struct switcheroo<ComplexD> {  
+       static inline int iscomplex()  { return 1; } 
+
+       template<class vec>
+       static inline vec mult(vec a, vec b) {
+	 return a*b;
+       }
+     };
+     template<> struct switcheroo<ComplexF> {  
+       static inline int iscomplex()  { return 1; } 
+       template<class vec>
+       static inline vec mult(vec a, vec b) {
+	 return a*b;
+       }
+     };
+
+
     template<class Impl>
     class CayleyFermion5D : public WilsonFermion5D<Impl>
     {
@@ -75,7 +100,19 @@ namespace Grid {
 		  std::vector<Coeff_t> &lower,
 		  std::vector<Coeff_t> &diag,
 		  std::vector<Coeff_t> &upper);
+
       void MooeeInternal(const FermionField &in, FermionField &out,int dag,int inv);
+      void MooeeInternalCompute(int dag, int inv, Vector<iSinglet<Simd> > & Matp, Vector<iSinglet<Simd> > & Matm);
+
+      void MooeeInternalAsm(const FermionField &in, FermionField &out,
+			    int LLs, int site,
+			    Vector<iSinglet<Simd> > &Matp,
+			    Vector<iSinglet<Simd> > &Matm);
+      void MooeeInternalZAsm(const FermionField &in, FermionField &out,
+			    int LLs, int site,
+			    Vector<iSinglet<Simd> > &Matp,
+			    Vector<iSinglet<Simd> > &Matm);
+
 
       virtual void   Instantiatable(void)=0;
 
@@ -112,6 +149,12 @@ namespace Grid {
       std::vector<Coeff_t> ueem;    
       std::vector<Coeff_t> dee;    
 
+      // Matrices of 5d ee inverse params
+      Vector<iSinglet<Simd> >  MatpInv;
+      Vector<iSinglet<Simd> >  MatmInv;
+      Vector<iSinglet<Simd> >  MatpInvDag;
+      Vector<iSinglet<Simd> >  MatmInvDag;
+
       // Constructors
       CayleyFermion5D(GaugeField &_Umu,
 		      GridCartesian         &FiveDimGrid,

lib/qcd/action/fermion/CayleyFermion5Dcache.cc

@@ -29,7 +29,7 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
     *************************************************************************************/
     /*  END LEGAL */
 
-#include <Grid.h>
+#include <Grid/Grid.h>
 
 
 namespace Grid {

lib/qcd/action/fermion/CayleyFermion5Ddense.cc

@@ -30,7 +30,7 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
     /*  END LEGAL */
 
 #include <Grid/Eigen/Dense>
-#include <Grid.h>
+#include <Grid/Grid.h>
 
 
 namespace Grid {

lib/qcd/action/fermion/CayleyFermion5Dssp.cc

@@ -29,7 +29,7 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
     *************************************************************************************/
     /*  END LEGAL */
 
-#include <Grid.h>
+#include <Grid/Grid.h>
 
 
 namespace Grid {

lib/qcd/action/fermion/CayleyFermion5Dvec.cc

@@ -29,13 +29,12 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
     *************************************************************************************/
     /*  END LEGAL */
 
-#include <Grid/Eigen/Dense>
-#include <Grid.h>
+
+#include <Grid/Grid.h>
 
 
 namespace Grid {
-namespace QCD {
-  /*
+namespace QCD {  /*
    * Dense matrix versions of routines
    */
 template<class Impl>
@@ -126,7 +125,6 @@ PARALLEL_FOR_LOOP
       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;
@@ -145,9 +143,6 @@ PARALLEL_FOR_LOOP
 	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);
@@ -165,42 +160,20 @@ PARALLEL_FOR_LOOP
 	  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; 
+	// Can force these to real arithmetic and save 2x.
+	Simd p_00  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(0)(0))  + switcheroo<Coeff_t>::mult(l[v]()()(),hm_00); 
+	Simd p_01  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(0)(1))  + switcheroo<Coeff_t>::mult(l[v]()()(),hm_01); 
+	Simd p_02  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(0)(2))  + switcheroo<Coeff_t>::mult(l[v]()()(),hm_02); 
+	Simd p_10  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(1)(0))  + switcheroo<Coeff_t>::mult(l[v]()()(),hm_10); 
+	Simd p_11  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(1)(1))  + switcheroo<Coeff_t>::mult(l[v]()()(),hm_11); 
+	Simd p_12  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(1)(2))  + switcheroo<Coeff_t>::mult(l[v]()()(),hm_12); 
+	Simd p_20  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(2)(0))  + switcheroo<Coeff_t>::mult(u[v]()()(),hp_00); 
+	Simd p_21  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(2)(1))  + switcheroo<Coeff_t>::mult(u[v]()()(),hp_01); 
+	Simd p_22  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(2)(2))  + switcheroo<Coeff_t>::mult(u[v]()()(),hp_02);  
+	Simd p_30  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(3)(0))  + switcheroo<Coeff_t>::mult(u[v]()()(),hp_10); 
+	Simd p_31  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(3)(1))  + switcheroo<Coeff_t>::mult(u[v]()()(),hp_11); 
+	Simd p_32  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(3)(2))  + switcheroo<Coeff_t>::mult(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);
@@ -261,7 +234,7 @@ void CayleyFermion5D<Impl>::M5Ddag(const FermionField &psi,
   M5Dtime-=usecond();
 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;
@@ -287,9 +260,504 @@ PARALLEL_FOR_LOOP
       chi[ss+v] = chi[ss+v]     +l[v]*fm;
 
     }
+#else
+      for(int v=0;v<LLs;v++){
+
+	vprefetch(psi[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]()(0)(0); 
+	Simd hp_01 = psi[ss+vp]()(0)(1); 
+	Simd hp_02 = psi[ss+vp]()(0)(2); 
+	Simd hp_10 = psi[ss+vp]()(1)(0); 
+	Simd hp_11 = psi[ss+vp]()(1)(1); 
+	Simd hp_12 = psi[ss+vp]()(1)(2); 
+	
+	Simd hm_00 = psi[ss+vm]()(2)(0); 
+	Simd hm_01 = psi[ss+vm]()(2)(1); 
+	Simd hm_02 = psi[ss+vm]()(2)(2); 
+	Simd hm_10 = psi[ss+vm]()(3)(0); 
+	Simd hm_11 = psi[ss+vm]()(3)(1); 
+	Simd hm_12 = psi[ss+vm]()(3)(2); 
+
+	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);
+	}
+
+	Simd p_00  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(0)(0))  + switcheroo<Coeff_t>::mult(u[v]()()(),hp_00); 
+	Simd p_01  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(0)(1))  + switcheroo<Coeff_t>::mult(u[v]()()(),hp_01); 
+	Simd p_02  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(0)(2))  + switcheroo<Coeff_t>::mult(u[v]()()(),hp_02); 
+	Simd p_10  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(1)(0))  + switcheroo<Coeff_t>::mult(u[v]()()(),hp_10); 
+	Simd p_11  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(1)(1))  + switcheroo<Coeff_t>::mult(u[v]()()(),hp_11); 
+	Simd p_12  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(1)(2))  + switcheroo<Coeff_t>::mult(u[v]()()(),hp_12); 
+
+	Simd p_20  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(2)(0))  + switcheroo<Coeff_t>::mult(l[v]()()(),hm_00); 
+	Simd p_21  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(2)(1))  + switcheroo<Coeff_t>::mult(l[v]()()(),hm_01); 
+	Simd p_22  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(2)(2))  + switcheroo<Coeff_t>::mult(l[v]()()(),hm_02);  
+	Simd p_30  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(3)(0))  + switcheroo<Coeff_t>::mult(l[v]()()(),hm_10); 
+	Simd p_31  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(3)(1))  + switcheroo<Coeff_t>::mult(l[v]()()(),hm_11); 
+	Simd p_32  = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(3)(2))  + switcheroo<Coeff_t>::mult(l[v]()()(),hm_12); 
+
+	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();
 }
+
+
+#ifdef AVX512 
+#include <simd/Intel512common.h>
+#include <simd/Intel512avx.h>
+#include <simd/Intel512single.h>
+#endif 
+
+template<class Impl>
+void CayleyFermion5D<Impl>::MooeeInternalAsm(const FermionField &psi, FermionField &chi,
+					     int LLs, int site,
+					     Vector<iSinglet<Simd> > &Matp,
+					     Vector<iSinglet<Simd> > &Matm)
+{
+#ifndef AVX512
+  {
+  SiteHalfSpinor BcastP;
+  SiteHalfSpinor BcastM;
+  SiteHalfSpinor SiteChiP;
+  SiteHalfSpinor SiteChiM;
+
+  // Ls*Ls * 2 * 12 * vol flops
+  for(int s1=0;s1<LLs;s1++){ 
+    for(int s2=0;s2<LLs;s2++){ 
+      for(int  l=0; l<Simd::Nsimd();l++){ // simd lane
+
+        int s=s2+l*LLs;
+	int lex=s2+LLs*site;
+	
+	if ( s2==0 && l==0) {
+	  SiteChiP=zero;
+	  SiteChiM=zero;
+	}
+	
+	for(int sp=0;sp<2;sp++){
+        for(int co=0;co<Nc;co++){
+	  vbroadcast(BcastP()(sp  )(co),psi[lex]()(sp)(co),l);
+	}}
+	for(int sp=0;sp<2;sp++){
+        for(int co=0;co<Nc;co++){
+	  vbroadcast(BcastM()(sp  )(co),psi[lex]()(sp+2)(co),l);
+	}}
+
+	for(int sp=0;sp<2;sp++){
+        for(int co=0;co<Nc;co++){
+	  SiteChiP()(sp)(co)=real_madd(Matp[LLs*s+s1]()()(),BcastP()(sp)(co),SiteChiP()(sp)(co)); // 1100 us.
+	  SiteChiM()(sp)(co)=real_madd(Matm[LLs*s+s1]()()(),BcastM()(sp)(co),SiteChiM()(sp)(co)); // each found by commenting out
+	}}
+
+    }}
+    {
+      int lex = s1+LLs*site;
+      for(int sp=0;sp<2;sp++){
+      for(int co=0;co<Nc;co++){
+	vstream(chi[lex]()(sp)(co), SiteChiP()(sp)(co));
+	vstream(chi[lex]()(sp+2)(co), SiteChiM()(sp)(co));
+      }}
+    }
+  }
+
+  }
+#else
+  {
+  // pointers
+    //  MASK_REGS;
+#define Chi_00 %%zmm1
+#define Chi_01 %%zmm2
+#define Chi_02 %%zmm3
+#define Chi_10 %%zmm4
+#define Chi_11 %%zmm5
+#define Chi_12 %%zmm6
+#define Chi_20 %%zmm7
+#define Chi_21 %%zmm8
+#define Chi_22 %%zmm9
+#define Chi_30 %%zmm10
+#define Chi_31 %%zmm11
+#define Chi_32 %%zmm12
+
+#define BCAST0   %%zmm13
+#define BCAST1   %%zmm14
+#define BCAST2   %%zmm15
+#define BCAST3   %%zmm16
+#define BCAST4   %%zmm17
+#define BCAST5   %%zmm18
+#define BCAST6   %%zmm19
+#define BCAST7   %%zmm20
+#define BCAST8   %%zmm21
+#define BCAST9   %%zmm22
+#define BCAST10  %%zmm23
+#define BCAST11  %%zmm24
+
+  int incr=LLs*LLs*sizeof(iSinglet<Simd>);
+  for(int s1=0;s1<LLs;s1++){ 
+    for(int s2=0;s2<LLs;s2++){ 
+      int lex=s2+LLs*site;
+      uint64_t a0 = (uint64_t)&Matp[LLs*s2+s1]; // should be cacheable
+      uint64_t a1 = (uint64_t)&Matm[LLs*s2+s1];
+      uint64_t a2 = (uint64_t)&psi[lex];
+      for(int  l=0; l<Simd::Nsimd();l++){ // simd lane
+	if ( (s2+l)==0 ) {
+	  asm (
+  	           VPREFETCH1(0,%2)  	     VPREFETCH1(0,%1)
+  	           VPREFETCH1(12,%2)  	     VPREFETCH1(13,%2)
+  	           VPREFETCH1(14,%2)  	     VPREFETCH1(15,%2)         
+		   VBCASTCDUP(0,%2,BCAST0)   
+		   VBCASTCDUP(1,%2,BCAST1)   
+		   VBCASTCDUP(2,%2,BCAST2)   
+		   VBCASTCDUP(3,%2,BCAST3)   
+		   VBCASTCDUP(4,%2,BCAST4)     VMULMEM (0,%0,BCAST0,Chi_00)
+		   VBCASTCDUP(5,%2,BCAST5)     VMULMEM (0,%0,BCAST1,Chi_01)
+		   VBCASTCDUP(6,%2,BCAST6)     VMULMEM (0,%0,BCAST2,Chi_02)
+		   VBCASTCDUP(7,%2,BCAST7)     VMULMEM (0,%0,BCAST3,Chi_10)
+		   VBCASTCDUP(8,%2,BCAST8)     VMULMEM (0,%0,BCAST4,Chi_11)
+		   VBCASTCDUP(9,%2,BCAST9)     VMULMEM (0,%0,BCAST5,Chi_12)
+		   VBCASTCDUP(10,%2,BCAST10)   VMULMEM (0,%1,BCAST6,Chi_20)
+		   VBCASTCDUP(11,%2,BCAST11)   VMULMEM (0,%1,BCAST7,Chi_21)
+		   VMULMEM (0,%1,BCAST8,Chi_22)         
+		   VMULMEM (0,%1,BCAST9,Chi_30)
+		   VMULMEM (0,%1,BCAST10,Chi_31)       
+		   VMULMEM (0,%1,BCAST11,Chi_32)
+		   : : "r" (a0), "r" (a1), "r" (a2)  );
+	} else { 
+	  asm (
+		   VBCASTCDUP(0,%2,BCAST0)   VMADDMEM (0,%0,BCAST0,Chi_00)
+		   VBCASTCDUP(1,%2,BCAST1)   VMADDMEM (0,%0,BCAST1,Chi_01)
+		   VBCASTCDUP(2,%2,BCAST2)   VMADDMEM (0,%0,BCAST2,Chi_02)
+		   VBCASTCDUP(3,%2,BCAST3)   VMADDMEM (0,%0,BCAST3,Chi_10)
+		   VBCASTCDUP(4,%2,BCAST4)   VMADDMEM (0,%0,BCAST4,Chi_11)
+		   VBCASTCDUP(5,%2,BCAST5)   VMADDMEM (0,%0,BCAST5,Chi_12)
+		   VBCASTCDUP(6,%2,BCAST6)   VMADDMEM (0,%1,BCAST6,Chi_20)
+		   VBCASTCDUP(7,%2,BCAST7)   VMADDMEM (0,%1,BCAST7,Chi_21)
+		   VBCASTCDUP(8,%2,BCAST8)   VMADDMEM (0,%1,BCAST8,Chi_22)
+		   VBCASTCDUP(9,%2,BCAST9)   VMADDMEM (0,%1,BCAST9,Chi_30)
+		   VBCASTCDUP(10,%2,BCAST10)  VMADDMEM (0,%1,BCAST10,Chi_31)
+		   VBCASTCDUP(11,%2,BCAST11)  VMADDMEM (0,%1,BCAST11,Chi_32) 
+		   : : "r" (a0), "r" (a1), "r" (a2)  );
+	}
+	a0 = a0+incr;
+	a1 = a1+incr;
+	a2 = a2+sizeof(Simd::scalar_type);
+      }}
+    {
+      int lexa = s1+LLs*site;
+      asm (
+	       VSTORE(0,%0,Chi_00) VSTORE(1 ,%0,Chi_01)  VSTORE(2 ,%0,Chi_02)		
+	       VSTORE(3,%0,Chi_10) VSTORE(4 ,%0,Chi_11)  VSTORE(5 ,%0,Chi_12)		
+	       VSTORE(6,%0,Chi_20) VSTORE(7 ,%0,Chi_21)  VSTORE(8 ,%0,Chi_22)		
+	       VSTORE(9,%0,Chi_30) VSTORE(10,%0,Chi_31)  VSTORE(11,%0,Chi_32)		
+	       : : "r" ((uint64_t)&chi[lexa]) : "memory" );
+
+    }
+  }
+  }
+#undef Chi_00
+#undef Chi_01
+#undef Chi_02
+#undef Chi_10
+#undef Chi_11
+#undef Chi_12
+#undef Chi_20
+#undef Chi_21
+#undef Chi_22
+#undef Chi_30
+#undef Chi_31
+#undef Chi_32
+
+#undef BCAST0
+#undef BCAST1
+#undef BCAST2
+#undef BCAST3
+#undef BCAST4
+#undef BCAST5
+#undef BCAST6
+#undef BCAST7
+#undef BCAST8
+#undef BCAST9
+#undef BCAST10
+#undef BCAST11
+#endif
+};
+
+  // Z-mobius version
+template<class Impl>
+void CayleyFermion5D<Impl>::MooeeInternalZAsm(const FermionField &psi, FermionField &chi,
+					     int LLs, int site, Vector<iSinglet<Simd> > &Matp, Vector<iSinglet<Simd> > &Matm)
+{
+#ifndef AVX512
+  {
+  SiteHalfSpinor BcastP;
+  SiteHalfSpinor BcastM;
+  SiteHalfSpinor SiteChiP;
+  SiteHalfSpinor SiteChiM;
+
+  // Ls*Ls * 2 * 12 * vol flops
+  for(int s1=0;s1<LLs;s1++){ 
+    for(int s2=0;s2<LLs;s2++){ 
+      for(int  l=0; l<Simd::Nsimd();l++){ // simd lane
+
+        int s=s2+l*LLs;
+	int lex=s2+LLs*site;
+	
+	if ( s2==0 && l==0) {
+	  SiteChiP=zero;
+	  SiteChiM=zero;
+	}
+	
+	for(int sp=0;sp<2;sp++){
+        for(int co=0;co<Nc;co++){
+	  vbroadcast(BcastP()(sp  )(co),psi[lex]()(sp)(co),l);
+	}}
+	for(int sp=0;sp<2;sp++){
+        for(int co=0;co<Nc;co++){
+	  vbroadcast(BcastM()(sp  )(co),psi[lex]()(sp+2)(co),l);
+	}}
+
+	for(int sp=0;sp<2;sp++){
+        for(int co=0;co<Nc;co++){
+	  SiteChiP()(sp)(co)=SiteChiP()(sp)(co)+ Matp[LLs*s+s1]()()()*BcastP()(sp)(co); 
+	  SiteChiM()(sp)(co)=SiteChiM()(sp)(co)+ Matm[LLs*s+s1]()()()*BcastM()(sp)(co); 
+	}}
+
+
+    }}
+    {
+      int lex = s1+LLs*site;
+      for(int sp=0;sp<2;sp++){
+      for(int co=0;co<Nc;co++){
+	vstream(chi[lex]()(sp)(co), SiteChiP()(sp)(co));
+	vstream(chi[lex]()(sp+2)(co), SiteChiM()(sp)(co));
+      }}
+    }
+  }
+
+  }
+#else
+  {
+  // pointers
+  //  MASK_REGS;
+#define Chi_00 %zmm0
+#define Chi_01 %zmm1
+#define Chi_02 %zmm2
+#define Chi_10 %zmm3
+#define Chi_11 %zmm4
+#define Chi_12 %zmm5
+#define Chi_20 %zmm6
+#define Chi_21 %zmm7
+#define Chi_22 %zmm8
+#define Chi_30 %zmm9
+#define Chi_31 %zmm10
+#define Chi_32 %zmm11
+#define pChi_00 %%zmm0
+#define pChi_01 %%zmm1
+#define pChi_02 %%zmm2
+#define pChi_10 %%zmm3
+#define pChi_11 %%zmm4
+#define pChi_12 %%zmm5
+#define pChi_20 %%zmm6
+#define pChi_21 %%zmm7
+#define pChi_22 %%zmm8
+#define pChi_30 %%zmm9
+#define pChi_31 %%zmm10
+#define pChi_32 %%zmm11
+
+#define BCAST_00   %zmm12
+#define  SHUF_00   %zmm13
+#define BCAST_01   %zmm14
+#define  SHUF_01   %zmm15
+#define BCAST_02   %zmm16
+#define  SHUF_02   %zmm17
+#define BCAST_10   %zmm18
+#define  SHUF_10   %zmm19
+#define BCAST_11   %zmm20
+#define  SHUF_11   %zmm21
+#define BCAST_12   %zmm22
+#define  SHUF_12   %zmm23
+
+#define Mp  %zmm24
+#define Mps %zmm25
+#define Mm  %zmm26
+#define Mms %zmm27
+#define N 8
+  int incr=LLs*LLs*sizeof(iSinglet<Simd>);
+  for(int s1=0;s1<LLs;s1++){ 
+    for(int s2=0;s2<LLs;s2++){ 
+      int lex=s2+LLs*site;
+      uint64_t a0 = (uint64_t)&Matp[LLs*s2+s1]; // should be cacheable
+      uint64_t a1 = (uint64_t)&Matm[LLs*s2+s1];
+      uint64_t a2 = (uint64_t)&psi[lex];
+      for(int  l=0; l<Simd::Nsimd();l++){ // simd lane
+	if ( (s2+l)==0 ) {
+	  LOAD64(%r8,a0);
+	  LOAD64(%r9,a1);
+	  LOAD64(%r10,a2);
+	  asm (
+	       VLOAD(0,%r8,Mp)// i r
+	       VLOAD(0,%r9,Mm)
+	       VSHUF(Mp,Mps)  // r i 
+	       VSHUF(Mm,Mms)
+	       VPREFETCH1(12,%r10)  	     VPREFETCH1(13,%r10)
+	       VPREFETCH1(14,%r10)  	     VPREFETCH1(15,%r10)         
+
+	       VMULIDUP(0*N,%r10,Mps,Chi_00)
+	       VMULIDUP(1*N,%r10,Mps,Chi_01)
+	       VMULIDUP(2*N,%r10,Mps,Chi_02)
+	       VMULIDUP(3*N,%r10,Mps,Chi_10)
+	       VMULIDUP(4*N,%r10,Mps,Chi_11)
+	       VMULIDUP(5*N,%r10,Mps,Chi_12)
+
+	       VMULIDUP(6*N ,%r10,Mms,Chi_20)
+	       VMULIDUP(7*N ,%r10,Mms,Chi_21)
+	       VMULIDUP(8*N ,%r10,Mms,Chi_22)
+	       VMULIDUP(9*N ,%r10,Mms,Chi_30)
+	       VMULIDUP(10*N,%r10,Mms,Chi_31)
+	       VMULIDUP(11*N,%r10,Mms,Chi_32)
+
+	       VMADDSUBRDUP(0*N,%r10,Mp,Chi_00)
+	       VMADDSUBRDUP(1*N,%r10,Mp,Chi_01)
+	       VMADDSUBRDUP(2*N,%r10,Mp,Chi_02)
+	       VMADDSUBRDUP(3*N,%r10,Mp,Chi_10)
+	       VMADDSUBRDUP(4*N,%r10,Mp,Chi_11)
+	       VMADDSUBRDUP(5*N,%r10,Mp,Chi_12)
+
+	       VMADDSUBRDUP(6*N ,%r10,Mm,Chi_20)
+	       VMADDSUBRDUP(7*N ,%r10,Mm,Chi_21)
+	       VMADDSUBRDUP(8*N ,%r10,Mm,Chi_22)
+	       VMADDSUBRDUP(9*N ,%r10,Mm,Chi_30)
+	       VMADDSUBRDUP(10*N,%r10,Mm,Chi_31)
+	       VMADDSUBRDUP(11*N,%r10,Mm,Chi_32)
+	       );
+	} else { 
+	  LOAD64(%r8,a0);
+	  LOAD64(%r9,a1);
+	  LOAD64(%r10,a2);
+	  asm (
+	       VLOAD(0,%r8,Mp)
+	       VSHUF(Mp,Mps)
+
+	       VLOAD(0,%r9,Mm)
+	       VSHUF(Mm,Mms)
+
+	       VMADDSUBIDUP(0*N,%r10,Mps,Chi_00) //  Mri * Pii +- Cir
+	       VMADDSUBIDUP(1*N,%r10,Mps,Chi_01)
+	       VMADDSUBIDUP(2*N,%r10,Mps,Chi_02)
+	       VMADDSUBIDUP(3*N,%r10,Mps,Chi_10)
+	       VMADDSUBIDUP(4*N,%r10,Mps,Chi_11)
+	       VMADDSUBIDUP(5*N,%r10,Mps,Chi_12)
+
+	       VMADDSUBIDUP(6 *N,%r10,Mms,Chi_20)
+	       VMADDSUBIDUP(7 *N,%r10,Mms,Chi_21)
+	       VMADDSUBIDUP(8 *N,%r10,Mms,Chi_22)
+	       VMADDSUBIDUP(9 *N,%r10,Mms,Chi_30)
+	       VMADDSUBIDUP(10*N,%r10,Mms,Chi_31)
+	       VMADDSUBIDUP(11*N,%r10,Mms,Chi_32)
+
+	       VMADDSUBRDUP(0*N,%r10,Mp,Chi_00) //  Cir = Mir * Prr +- ( Mri * Pii +- Cir) 
+	       VMADDSUBRDUP(1*N,%r10,Mp,Chi_01) //  Ci = MiPr + Ci + MrPi ;    Cr = MrPr - ( MiPi - Cr)
+	       VMADDSUBRDUP(2*N,%r10,Mp,Chi_02)
+	       VMADDSUBRDUP(3*N,%r10,Mp,Chi_10)
+	       VMADDSUBRDUP(4*N,%r10,Mp,Chi_11)
+	       VMADDSUBRDUP(5*N,%r10,Mp,Chi_12)
+
+	       VMADDSUBRDUP(6 *N,%r10,Mm,Chi_20)
+	       VMADDSUBRDUP(7 *N,%r10,Mm,Chi_21)
+	       VMADDSUBRDUP(8 *N,%r10,Mm,Chi_22)
+	       VMADDSUBRDUP(9 *N,%r10,Mm,Chi_30)
+	       VMADDSUBRDUP(10*N,%r10,Mm,Chi_31)
+	       VMADDSUBRDUP(11*N,%r10,Mm,Chi_32)
+	       );
+	}
+	a0 = a0+incr;
+	a1 = a1+incr;
+	a2 = a2+sizeof(Simd::scalar_type);
+      }}
+    {
+      int lexa = s1+LLs*site;
+      /*
+      SiteSpinor tmp;
+      asm (
+	       VSTORE(0,%0,pChi_00) VSTORE(1 ,%0,pChi_01)  VSTORE(2 ,%0,pChi_02)		
+	       VSTORE(3,%0,pChi_10) VSTORE(4 ,%0,pChi_11)  VSTORE(5 ,%0,pChi_12)		
+	       VSTORE(6,%0,pChi_20) VSTORE(7 ,%0,pChi_21)  VSTORE(8 ,%0,pChi_22)		
+	       VSTORE(9,%0,pChi_30) VSTORE(10,%0,pChi_31)  VSTORE(11,%0,pChi_32)		
+	       : : "r" ((uint64_t)&tmp) : "memory" );
+      */
+
+      asm (
+	       VSTORE(0,%0,pChi_00) VSTORE(1 ,%0,pChi_01)  VSTORE(2 ,%0,pChi_02)		
+	       VSTORE(3,%0,pChi_10) VSTORE(4 ,%0,pChi_11)  VSTORE(5 ,%0,pChi_12)		
+	       VSTORE(6,%0,pChi_20) VSTORE(7 ,%0,pChi_21)  VSTORE(8 ,%0,pChi_22)		
+	       VSTORE(9,%0,pChi_30) VSTORE(10,%0,pChi_31)  VSTORE(11,%0,pChi_32)		
+	       : : "r" ((uint64_t)&chi[lexa]) : "memory" );
+
+      //      if ( 1 || (site==0) ) { 
+      //	std::cout<<site << " s1 "<<s1<<"\n\t"<<tmp << "\n't" << chi[lexa] <<"\n\t"<<tmp-chi[lexa]<<std::endl;
+      //      }
+    }
+  }
+  }
+#undef Chi_00
+#undef Chi_01
+#undef Chi_02
+#undef Chi_10
+#undef Chi_11
+#undef Chi_12
+#undef Chi_20
+#undef Chi_21
+#undef Chi_22
+#undef Chi_30
+#undef Chi_31
+#undef Chi_32
+
+#undef BCAST0
+#undef BCAST1
+#undef BCAST2
+#undef BCAST3
+#undef BCAST4
+#undef BCAST5
+#undef BCAST6
+#undef BCAST7
+#undef BCAST8
+#undef BCAST9
+#undef BCAST10
+#undef BCAST11
+
+#endif
+};
+
+
 template<class Impl>
 void CayleyFermion5D<Impl>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv)
 {
@@ -299,108 +767,41 @@ void CayleyFermion5D<Impl>::MooeeInternal(const FermionField &psi, FermionField
 
   chi.checkerboard=psi.checkerboard;
   
-  Eigen::MatrixXcd Pplus  = Eigen::MatrixXcd::Zero(Ls,Ls);
-  Eigen::MatrixXcd Pminus = Eigen::MatrixXcd::Zero(Ls,Ls);
+  Vector<iSinglet<Simd> >  Matp;
+  Vector<iSinglet<Simd> >  Matm;
+  Vector<iSinglet<Simd> >  *_Matp;
+  Vector<iSinglet<Simd> >  *_Matm;
   
-  for(int s=0;s<Ls;s++){
-    Pplus(s,s) = bee[s];
-    Pminus(s,s)= bee[s];
+  //  MooeeInternalCompute(dag,inv,Matp,Matm);
+  if ( inv && dag ) { 
+    _Matp = &MatpInvDag;
+    _Matm = &MatmInvDag;
   }
-  
-  for(int s=0;s<Ls-1;s++){
-    Pminus(s,s+1) = -cee[s];
+  if ( inv && (!dag) ) { 
+    _Matp = &MatpInv;
+    _Matm = &MatmInv;
+  } 
+  if ( !inv ) {
+    MooeeInternalCompute(dag,inv,Matp,Matm);
+    _Matp = &Matp;
+    _Matm = &Matm;
   }
-  
-  for(int s=0;s<Ls-1;s++){
-    Pplus(s+1,s) = -cee[s+1];
-  }
-  Pplus (0,Ls-1) = mass*cee[0];
-  Pminus(Ls-1,0) = mass*cee[Ls-1];
-  
-  Eigen::MatrixXcd PplusMat ;
-  Eigen::MatrixXcd PminusMat;
-  
-  if ( inv ) {
-    PplusMat =Pplus.inverse();
-    PminusMat=Pminus.inverse();
-  } else { 
-    PplusMat =Pplus;
-    PminusMat=Pminus;
-  }
-  
-  if(dag){
-    PplusMat.adjointInPlace();
-    PminusMat.adjointInPlace();
-  }
-  
-  typedef typename SiteHalfSpinor::scalar_type scalar_type;
-  const int Nsimd=Simd::Nsimd();
-  Vector<iSinglet<Simd> > Matp(Ls*LLs);
-  Vector<iSinglet<Simd> > Matm(Ls*LLs);
+  assert(_Matp->size()==Ls*LLs);
 
-  for(int s2=0;s2<Ls;s2++){
-  for(int s1=0;s1<LLs;s1++){
-    int istride = LLs;
-    int ostride = 1;
-      Simd Vp;
-      Simd Vm;
-      scalar_type *sp = (scalar_type *)&Vp;
-      scalar_type *sm = (scalar_type *)&Vm;
-      for(int l=0;l<Nsimd;l++){
-	sp[l] = PplusMat (l*istride+s1*ostride ,s2);
-	sm[l] = PminusMat(l*istride+s1*ostride,s2);
-      }
-      Matp[LLs*s2+s1] = Vp;
-      Matm[LLs*s2+s1] = Vm;
-    }
-  }
-  
   MooeeInvCalls++;
   MooeeInvTime-=usecond();
-  // Dynamic allocate on stack to get per thread without serialised heap acces
-#pragma omp parallel  
-  {
 
-    Vector<SiteHalfSpinor> SitePplus(LLs);
-    Vector<SiteHalfSpinor> SitePminus(LLs);
-    Vector<SiteHalfSpinor> SiteChiP(LLs);
-    Vector<SiteHalfSpinor> SiteChiM(LLs);
-    Vector<SiteSpinor>     SiteChi(LLs);
-
-    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]);
-      spProj5m(SitePminus[s],psi[lex]);
-      SiteChiP[s]=zero;
-      SiteChiM[s]=zero;
+  if ( switcheroo<Coeff_t>::iscomplex() ) {
+  PARALLEL_FOR_LOOP
+    for(auto site=0;site<vol;site++){
+      MooeeInternalZAsm(psi,chi,LLs,site,*_Matp,*_Matm);
     }
-      
-    int s=0;
-    for(int  l=0; l<Simd::Nsimd();l++){ // simd lane
-      for(int s2=0;s2<LLs;s2++){ // Column loop of right hand side
-	vbroadcast(BcastP,SitePplus [s2],l);
-	vbroadcast(BcastM,SitePminus[s2],l);
-	for(int s1=0;s1<LLs;s1++){ // Column loop of reduction variables
-	  SiteChiP[s1]=SiteChiP[s1]+Matp[LLs*s+s1]*BcastP;
-	  SiteChiM[s1]=SiteChiM[s1]+Matm[LLs*s+s1]*BcastM;
-	}
-      s++;
-    }}
-
-    for(int s=0;s<LLs;s++){
-      int lex = s+LLs*site;
-      spRecon5p(SiteChi[s],SiteChiP[s]);
-      accumRecon5m(SiteChi[s],SiteChiM[s]);
-      chi[lex] = SiteChi[s]*0.5;
+  } else { 
+  PARALLEL_FOR_LOOP
+    for(auto site=0;site<vol;site++){
+      MooeeInternalAsm(psi,chi,LLs,site,*_Matp,*_Matm);
     }
   }
-  }
   MooeeInvTime+=usecond();
 }
 
@@ -414,4 +815,5 @@ template void CayleyFermion5D<DomainWallVec5dImplD>::MooeeInternal(const Fermion
 template void CayleyFermion5D<ZDomainWallVec5dImplF>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
 template void CayleyFermion5D<ZDomainWallVec5dImplD>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
 
+
 }}

lib/qcd/action/fermion/ContinuedFractionFermion5D.cc

@@ -26,7 +26,7 @@ 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>
 
 namespace Grid {
   namespace QCD {

lib/qcd/action/fermion/FermionOperator.h

@@ -48,6 +48,8 @@ namespace Grid {
 
       FermionOperator(const ImplParams &p= ImplParams()) : Impl(p) {};
 
+      virtual FermionField &tmp(void) = 0;
+
       GridBase * Grid(void)   { return FermionGrid(); };   // this is all the linalg routines need to know
       GridBase * RedBlackGrid(void) { return FermionRedBlackGrid(); };
 

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/PartialFractionFermion5D.cc

@@ -26,7 +26,7 @@ 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>
 namespace Grid {
   namespace QCD {
 

lib/qcd/action/fermion/WilsonFermion.cc

@@ -29,7 +29,7 @@ See the full license in the file "LICENSE" in the top level distribution
 directory
 *************************************************************************************/
 /*  END LEGAL */
-#include <Grid.h>
+#include <Grid/Grid.h>
 
 namespace Grid {
 namespace QCD {
@@ -61,7 +61,9 @@ WilsonFermion<Impl>::WilsonFermion(GaugeField &_Umu, GridCartesian &Fgrid,
       LebesgueEvenOdd(_cbgrid),
       Umu(&Fgrid),
       UmuEven(&Hgrid),
-      UmuOdd(&Hgrid) {
+      UmuOdd(&Hgrid),
+      _tmp(&Hgrid)
+{
   // Allocate the required comms buffer
   ImportGauge(_Umu);
 }
@@ -147,11 +149,11 @@ void WilsonFermion<Impl>::MeooeDag(const FermionField &in, FermionField &out) {
 
     typedef Lattice<iSinglet<vector_type> > LatComplex;
 
-    Gamma::GammaMatrix Gmu [] = {
-      Gamma::GammaX,
-      Gamma::GammaY,
-      Gamma::GammaZ,
-      Gamma::GammaT
+    Gamma::Algebra Gmu [] = {
+      Gamma::Algebra::GammaX,
+      Gamma::Algebra::GammaY,
+      Gamma::Algebra::GammaZ,
+      Gamma::Algebra::GammaT
     };
 
     std::vector<int> latt_size   = _grid->_fdimensions;

lib/qcd/action/fermion/WilsonFermion.h

@@ -58,6 +58,9 @@ class WilsonFermion : public WilsonKernels<Impl>, public WilsonFermionStatic {
   GridBase *FermionGrid(void) { return _grid; }
   GridBase *FermionRedBlackGrid(void) { return _cbgrid; }
 
+  FermionField _tmp;
+  FermionField &tmp(void) { return _tmp; }
+
   //////////////////////////////////////////////////////////////////
   // override multiply; cut number routines if pass dagger argument
   // and also make interface more uniformly consistent