Checking in working version of Lanczos.

.gitignore

@@ -92,7 +92,6 @@ build*/*
 #####################
 *.xcodeproj/*
 build.sh
-.vscode
 
 # Eigen source #
 ################
@@ -107,10 +106,6 @@ lib/fftw/*
 m4/lt*
 m4/libtool.m4
 
-# github pages #
-################
-gh-pages/
-
 # Buck files #
 ##############
 .buck*
@@ -121,5 +116,4 @@ make-bin-BUCK.sh
 # generated sources #
 #####################
 lib/qcd/spin/gamma-gen/*.h
-lib/qcd/spin/gamma-gen/*.cc
+lib/qcd/spin/gamma-gen/*.cc
-

.travis.yml

@@ -7,11 +7,9 @@ cache:
 matrix:
   include:
     - os:        osx
-      osx_image: xcode8.3
+      osx_image: xcode7.2
       compiler: clang
     - compiler: gcc
-      dist: trusty
-      sudo: required
       addons:
         apt:
           sources:
@@ -26,8 +24,6 @@ matrix:
             - binutils-dev
       env: VERSION=-4.9
     - compiler: gcc
-      dist: trusty
-      sudo: required
       addons:
         apt:
           sources:
@@ -42,7 +38,6 @@ matrix:
             - binutils-dev
       env: VERSION=-5
     - compiler: clang
-      dist: trusty
       addons:
         apt:
           sources:
@@ -57,7 +52,6 @@ matrix:
             - binutils-dev
       env: CLANG_LINK=http://llvm.org/releases/3.8.0/clang+llvm-3.8.0-x86_64-linux-gnu-ubuntu-14.04.tar.xz
     - compiler: clang
-      dist: trusty
       addons:
         apt:
           sources:
@@ -79,15 +73,13 @@ before_install:
     - if [[ "$TRAVIS_OS_NAME" == "linux" ]] && [[ "$CC" == "clang" ]]; then export LD_LIBRARY_PATH="${GRIDDIR}/clang/lib:${LD_LIBRARY_PATH}"; fi
     - if [[ "$TRAVIS_OS_NAME" == "osx" ]]; then brew update; fi
     - if [[ "$TRAVIS_OS_NAME" == "osx" ]]; then brew install libmpc; fi
+    - if [[ "$TRAVIS_OS_NAME" == "osx" ]]; then brew install openmpi; fi
+    - if [[ "$TRAVIS_OS_NAME" == "osx" ]] && [[ "$CC" == "gcc" ]]; then brew install gcc5; fi
     
 install:
     - export CC=$CC$VERSION
     - export CXX=$CXX$VERSION
     - echo $PATH
-    - which autoconf
-    - autoconf  --version
-    - which automake
-    - automake  --version
     - which $CC
     - $CC  --version
     - which $CXX
@@ -100,15 +92,15 @@ script:
     - cd build
     - ../configure --enable-precision=single --enable-simd=SSE4 --enable-comms=none
     - make -j4 
-    - ./benchmarks/Benchmark_dwf --threads 1 --debug-signals
+    - ./benchmarks/Benchmark_dwf --threads 1
     - echo make clean
     - ../configure --enable-precision=double --enable-simd=SSE4 --enable-comms=none
     - make -j4
-    - ./benchmarks/Benchmark_dwf --threads 1 --debug-signals
+    - ./benchmarks/Benchmark_dwf --threads 1
-    - make check
     - echo make clean
-    - if [[ "$TRAVIS_OS_NAME" == "linux" ]] && [[ "$CC" == "clang" ]]; then ../configure --enable-precision=single --enable-simd=SSE4 --enable-comms=mpi-auto ; fi
+    - if [[ "$TRAVIS_OS_NAME" == "linux" ]]; then export CXXFLAGS='-DMPI_UINT32_T=MPI_UNSIGNED -DMPI_UINT64_T=MPI_UNSIGNED_LONG'; fi
-    - if [[ "$TRAVIS_OS_NAME" == "linux" ]] && [[ "$CC" == "clang" ]]; then make -j4; fi
+    - ../configure --enable-precision=single --enable-simd=SSE4 --enable-comms=mpi-auto
-    - if [[ "$TRAVIS_OS_NAME" == "linux" ]] && [[ "$CC" == "clang" ]]; then mpirun.openmpi -n 2 ./benchmarks/Benchmark_dwf --threads 1 --mpi 2.1.1.1; fi
+    - make -j4
-
+    - if [[ "$TRAVIS_OS_NAME" == "linux" ]]; then mpirun.openmpi -n 2 ./benchmarks/Benchmark_dwf --threads 1 --mpi 2.1.1.1; fi
+    - if [[ "$TRAVIS_OS_NAME" == "osx" ]]; then mpirun -n 2 ./benchmarks/Benchmark_dwf --threads 1 --mpi 2.1.1.1; fi
 

Makefile.am

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

README.md

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

TODO

@@ -1,26 +1,6 @@
 TODO:
 ---------------
 
-Peter's work list:
-2)- Precision conversion and sort out localConvert      <-- 
-3)- Remove DenseVector, DenseMatrix; Use Eigen instead. <-- started 
-4)- Binary I/O speed up & x-strips
--- Profile CG, BlockCG, etc... Flop count/rate -- PARTIAL, time but no flop/s yet
--- Physical propagator interface
--- Conserved currents
--- GaugeFix into central location
--- Multigrid Wilson and DWF, compare to other Multigrid implementations
--- HDCR resume
-
-Recent DONE 
--- Cut down the exterior overhead                      <-- DONE
--- Interior legs from SHM comms                        <-- DONE
--- Half-precision comms                                <-- DONE
--- Merge high precision reduction into develop        
--- multiRHS DWF; benchmark on Cori/BNL for comms elimination
-   -- slice* linalg routines for multiRHS, BlockCG    
-
------
 * Forces; the UdSdU  term in gauge force term is half of what I think it should
   be. This is a consequence of taking ONLY the first term in:
 
@@ -41,8 +21,16 @@ Recent DONE
   This means we must double the force in the Test_xxx_force routines, and is the origin of the factor of two.
   This 2x is applied by hand in the fermion routines and in the Test_rect_force routine.
 
+
+Policies:
+
+* Link smearing/boundary conds; Policy class based implementation ; framework more in place
+
 * Support different boundary conditions (finite temp, chem. potential ... )
 
+* Support different fermion representations? 
+  - contained entirely within the integrator presently
+
 - Sign of force term.
 
 - Reversibility test.
@@ -53,6 +41,11 @@ Recent DONE
 
 - Audit oIndex usage for cb behaviour
 
+- Rectangle gauge actions.
+  Iwasaki,
+  Symanzik,
+  ... etc...
+
 - Prepare multigrid for HMC. - Alternate setup schemes.
 
 - Support for ILDG --- ugly, not done
@@ -62,11 +55,9 @@ Recent DONE
 - FFTnD ?
 
 - Gparity; hand opt use template specialisation elegance to enable the optimised paths ?
-
 - Gparity force term; Gparity (R)HMC.
-
+- Random number state save restore
 - Mobius implementation clean up to rmove #if 0 stale code sequences
-
 - CG -- profile carefully, kernel fusion, whole CG performance measurements.
 
 ================================================================
@@ -99,7 +90,6 @@ Insert/Extract
 Not sure of status of this -- reverify. Things are working nicely now though.
 
 * Make the Tensor types and Complex etc... play more nicely.
-
   - TensorRemove is a hack, come up with a long term rationalised approach to Complex vs. Scalar<Scalar<Scalar<Complex > > >
     QDP forces use of "toDouble" to get back to non tensor scalar. This role is presently taken TensorRemove, but I
     want to introduce a syntax that does not require this.
@@ -122,8 +112,6 @@ Not sure of status of this -- reverify. Things are working nicely now though.
 RECENT
 ---------------
 
-  - Support different fermion representations? -- DONE
-  - contained entirely within the integrator presently
   - Clean up HMC                                                             -- DONE
   - LorentzScalar<GaugeField> gets Gauge link type (cleaner).                -- DONE
   - Simplified the integrators a bit.                                        -- DONE
@@ -135,26 +123,6 @@ RECENT
   - Parallel io improvements                                  -- DONE
   - Plaquette and link trace checks into nersc reader from the Grid_nersc_io.cc test. -- DONE
 
-
-DONE:
-- MultiArray -- MultiRHS done
-- ConjugateGradientMultiShift -- DONE
-- MCR                         -- DONE
-- Remez -- Mike or Boost?     -- DONE
-- Proto (ET)                  -- DONE
-- uBlas                       -- DONE ; Eigen
-- Potentially Useful Boost libraries -- DONE ; Eigen
-- Aligned allocator; memory pool -- DONE
-- Multiprecision              -- DONE
-- Serialization               -- DONE
-- Regex -- Not needed
-- Tokenize -- Why?
-
-- Random number state save restore -- DONE
-- Rectangle gauge actions. -- DONE
-  Iwasaki,
-  Symanzik,
-  ... etc...
 Done: Cayley, Partial , ContFrac force terms.
 
 DONE
@@ -239,7 +207,6 @@ Done
 FUNCTIONALITY: it pleases me to keep track of things I have done (keeps me arguably sane)
 ======================================================================================================
 
-* Link smearing/boundary conds; Policy class based implementation ; framework more in place -- DONE
 * Command line args for geometry, simd, etc. layout. Is it necessary to have -- DONE
   user pass these? Is this a QCD specific?
 

VERSION

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

benchmarks/Benchmark_comms.cc

@@ -48,9 +48,9 @@ int main (int argc, char ** argv)
   std::cout<<GridLogMessage << "= Benchmarking concurrent halo exchange in "<<nmu<<" dimensions"<<std::endl;
   std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
   std::cout<<GridLogMessage << "  L  "<<"\t\t"<<" Ls  "<<"\t\t"<<"bytes"<<"\t\t"<<"MB/s uni"<<"\t\t"<<"MB/s bidi"<<std::endl;
-  int maxlat=24;
+  int maxlat=16;
-  for(int lat=4;lat<=maxlat;lat+=4){
+  for(int lat=4;lat<=maxlat;lat+=2){
-    for(int Ls=8;Ls<=32;Ls*=2){
+    for(int Ls=1;Ls<=16;Ls*=2){
 
       std::vector<int> latt_size  ({lat*mpi_layout[0],
       				    lat*mpi_layout[1],
@@ -124,8 +124,8 @@ int main (int argc, char ** argv)
   std::cout<<GridLogMessage << "  L  "<<"\t\t"<<" Ls  "<<"\t\t"<<"bytes"<<"\t\t"<<"MB/s uni"<<"\t\t"<<"MB/s bidi"<<std::endl;
 
 
-  for(int lat=4;lat<=maxlat;lat+=4){
+  for(int lat=4;lat<=maxlat;lat+=2){
-    for(int Ls=8;Ls<=32;Ls*=2){
+    for(int Ls=1;Ls<=16;Ls*=2){
 
       std::vector<int> latt_size  ({lat,lat,lat,lat});
 
@@ -194,14 +194,14 @@ int main (int argc, char ** argv)
   }  
 
 
-  Nloop=10;
+  Nloop=100;
   std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
   std::cout<<GridLogMessage << "= Benchmarking concurrent STENCIL halo exchange in "<<nmu<<" dimensions"<<std::endl;
   std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
   std::cout<<GridLogMessage << "  L  "<<"\t\t"<<" Ls  "<<"\t\t"<<"bytes"<<"\t\t"<<"MB/s uni"<<"\t\t"<<"MB/s bidi"<<std::endl;
 
-  for(int lat=4;lat<=maxlat;lat+=4){
+  for(int lat=4;lat<=maxlat;lat+=2){
-    for(int Ls=8;Ls<=32;Ls*=2){
+    for(int Ls=1;Ls<=16;Ls*=2){
 
       std::vector<int> latt_size  ({lat*mpi_layout[0],
       				    lat*mpi_layout[1],
@@ -281,8 +281,8 @@ int main (int argc, char ** argv)
   std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
   std::cout<<GridLogMessage << "  L  "<<"\t\t"<<" Ls  "<<"\t\t"<<"bytes"<<"\t\t"<<"MB/s uni"<<"\t\t"<<"MB/s bidi"<<std::endl;
 
-  for(int lat=4;lat<=maxlat;lat+=4){
+  for(int lat=4;lat<=maxlat;lat+=2){
-    for(int Ls=8;Ls<=32;Ls*=2){
+    for(int Ls=1;Ls<=16;Ls*=2){
 
       std::vector<int> latt_size  ({lat*mpi_layout[0],
       				    lat*mpi_layout[1],
@@ -324,8 +324,8 @@ int main (int argc, char ** argv)
 					    (void *)&rbuf[mu][0],
 					    recv_from_rank,
 					    bytes);
-	    Grid.StencilSendToRecvFromComplete(requests);
+	    //	    Grid.StencilSendToRecvFromComplete(requests);
-	    requests.resize(0);
+	    //	    requests.resize(0);
 
 	    comm_proc = mpi_layout[mu]-1;
 	  

benchmarks/Benchmark_dwf.cc

@@ -1,22 +1,28 @@
- /*************************************************************************************
+    /*************************************************************************************
+
     Grid physics library, www.github.com/paboyle/Grid 
+
     Source file: ./benchmarks/Benchmark_dwf.cc
+
     Copyright (C) 2015
 
-    Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-    Author: paboyle <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 */
@@ -42,16 +48,16 @@ typedef WilsonFermion5D<DomainWallVec5dImplR> WilsonFermion5DR;
 typedef WilsonFermion5D<DomainWallVec5dImplF> WilsonFermion5DF;
 typedef WilsonFermion5D<DomainWallVec5dImplD> WilsonFermion5DD;
 
+
 int main (int argc, char ** argv)
 {
   Grid_init(&argc,&argv);
 
-
   int threads = GridThread::GetThreads();
   std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
 
   std::vector<int> latt4 = GridDefaultLatt();
-  const int Ls=16;
+  const int Ls=8;
   GridCartesian         * UGrid   = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());
   GridRedBlackCartesian * UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
   GridCartesian         * FGrid   = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
@@ -65,66 +71,35 @@ int main (int argc, char ** argv)
 
   std::vector<int> seeds4({1,2,3,4});
   std::vector<int> seeds5({5,6,7,8});
-  
+
-  std::cout << GridLogMessage << "Initialising 4d RNG" << std::endl;
   GridParallelRNG          RNG4(UGrid);  RNG4.SeedFixedIntegers(seeds4);
-  std::cout << GridLogMessage << "Initialising 5d RNG" << std::endl;
   GridParallelRNG          RNG5(FGrid);  RNG5.SeedFixedIntegers(seeds5);
-  std::cout << GridLogMessage << "Initialised RNGs" << std::endl;
 
   LatticeFermion src   (FGrid); random(RNG5,src);
-#if 0
-  src = zero;
-  {
-    std::vector<int> origin({0,0,0,latt4[2]-1,0});
-    SpinColourVectorF tmp;
-    tmp=zero;
-    tmp()(0)(0)=Complex(-2.0,0.0);
-    std::cout << " source site 0 " << tmp<<std::endl;
-    pokeSite(tmp,src,origin);
-  }
-#else
-  RealD N2 = 1.0/::sqrt(norm2(src));
-  src = src*N2;
-#endif
-
-
   LatticeFermion result(FGrid); result=zero;
   LatticeFermion    ref(FGrid);    ref=zero;
   LatticeFermion    tmp(FGrid);
   LatticeFermion    err(FGrid);
 
-  std::cout << GridLogMessage << "Drawing gauge field" << std::endl;
   LatticeGaugeField Umu(UGrid); 
-  SU3::HotConfiguration(RNG4,Umu); 
+  random(RNG4,Umu);
-  std::cout << GridLogMessage << "Random gauge initialised " << std::endl;
-#if 0
-  Umu=1.0;
-  for(int mu=0;mu<Nd;mu++){
-    LatticeColourMatrix ttmp(UGrid);
-    ttmp = PeekIndex<LorentzIndex>(Umu,mu);
-    //    if (mu !=2 ) ttmp = 0;
-    //    ttmp = ttmp* pow(10.0,mu);
-    PokeIndex<LorentzIndex>(Umu,ttmp,mu);
-  }
-  std::cout << GridLogMessage << "Forced to diagonal " << std::endl;
-#endif
 
-  ////////////////////////////////////
-  // Naive wilson implementation
-  ////////////////////////////////////
-  // replicate across fifth dimension
   LatticeGaugeField Umu5d(FGrid); 
-  std::vector<LatticeColourMatrix> U(4,FGrid);
+
+  // replicate across fifth dimension
   for(int ss=0;ss<Umu._grid->oSites();ss++){
     for(int s=0;s<Ls;s++){
       Umu5d._odata[Ls*ss+s] = Umu._odata[ss];
     }
   }
+
+  ////////////////////////////////////
+  // Naive wilson implementation
+  ////////////////////////////////////
+  std::vector<LatticeColourMatrix> U(4,FGrid);
   for(int mu=0;mu<Nd;mu++){
     U[mu] = PeekIndex<LorentzIndex>(Umu5d,mu);
   }
-  std::cout << GridLogMessage << "Setting up Cshift based reference " << std::endl;
 
   if (1)
   {
@@ -145,7 +120,8 @@ int main (int argc, char ** argv)
   RealD M5  =1.8;
 
   RealD NP = UGrid->_Nprocessors;
-  RealD NN = UGrid->NodeCount();
+
+  DomainWallFermionR Dw(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
 
   std::cout << GridLogMessage<< "*****************************************************************" <<std::endl;
   std::cout << GridLogMessage<< "* Kernel options --dslash-generic, --dslash-unroll, --dslash-asm" <<std::endl;
@@ -155,22 +131,15 @@ int main (int argc, char ** argv)
   std::cout << GridLogMessage<< "* Vectorising space-time by "<<vComplex::Nsimd()<<std::endl;
   if ( sizeof(Real)==4 )   std::cout << GridLogMessage<< "* SINGLE precision "<<std::endl;
   if ( sizeof(Real)==8 )   std::cout << GridLogMessage<< "* DOUBLE precision "<<std::endl;
-#ifdef GRID_OMP
-  if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsAndCompute ) std::cout << GridLogMessage<< "* Using Overlapped Comms/Compute" <<std::endl;
-  if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsThenCompute) std::cout << GridLogMessage<< "* Using sequential comms compute" <<std::endl;
-#endif
   if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptGeneric   ) std::cout << GridLogMessage<< "* Using GENERIC Nc WilsonKernels" <<std::endl;
   if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptHandUnroll) std::cout << GridLogMessage<< "* Using Nc=3       WilsonKernels" <<std::endl;
   if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptInlineAsm ) std::cout << GridLogMessage<< "* Using Asm Nc=3   WilsonKernels" <<std::endl;
   std::cout << GridLogMessage<< "*****************************************************************" <<std::endl;
 
-  DomainWallFermionR Dw(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
+  int ncall =100;
-  int ncall =1000;
   if (1) {
     FGrid->Barrier();
     Dw.ZeroCounters();
-    Dw.Dhop(src,result,0);
-    std::cout<<GridLogMessage<<"Called warmup"<<std::endl;
     double t0=usecond();
     for(int i=0;i<ncall;i++){
       __SSC_START;
@@ -184,55 +153,16 @@ int main (int argc, char ** argv)
     double flops=1344*volume*ncall;
 
     std::cout<<GridLogMessage << "Called Dw "<<ncall<<" times in "<<t1-t0<<" us"<<std::endl;
-    //    std::cout<<GridLogMessage << "norm result "<< norm2(result)<<std::endl;
+    std::cout<<GridLogMessage << "norm result "<< norm2(result)<<std::endl;
-    //    std::cout<<GridLogMessage << "norm ref    "<< norm2(ref)<<std::endl;
+    std::cout<<GridLogMessage << "norm ref    "<< norm2(ref)<<std::endl;
     std::cout<<GridLogMessage << "mflop/s =   "<< flops/(t1-t0)<<std::endl;
     std::cout<<GridLogMessage << "mflop/s per rank =  "<< flops/(t1-t0)/NP<<std::endl;
-    std::cout<<GridLogMessage << "mflop/s per node =  "<< flops/(t1-t0)/NN<<std::endl;
     err = ref-result; 
     std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<<std::endl;
-
-    /*
-    if(( norm2(err)>1.0e-4) ) { 
-      std::cout << "RESULT\n " << result<<std::endl;
-      std::cout << "REF   \n " << ref   <<std::endl;
-      std::cout << "ERR   \n " << err   <<std::endl;
-      FGrid->Barrier();
-      exit(-1);
-    }
-    */
     assert (norm2(err)< 1.0e-4 );
     Dw.Report();
   }
 
-  DomainWallFermionRL DwH(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
-  if (1) {
-    FGrid->Barrier();
-    DwH.ZeroCounters();
-    DwH.Dhop(src,result,0);
-    double t0=usecond();
-    for(int i=0;i<ncall;i++){
-      __SSC_START;
-      DwH.Dhop(src,result,0);
-      __SSC_STOP;
-    }
-    double t1=usecond();
-    FGrid->Barrier();
-    
-    double volume=Ls;  for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
-    double flops=1344*volume*ncall;
-
-    std::cout<<GridLogMessage << "Called half prec comms Dw "<<ncall<<" times in "<<t1-t0<<" us"<<std::endl;
-    std::cout<<GridLogMessage << "mflop/s =   "<< flops/(t1-t0)<<std::endl;
-    std::cout<<GridLogMessage << "mflop/s per rank =  "<< flops/(t1-t0)/NP<<std::endl;
-    std::cout<<GridLogMessage << "mflop/s per node =  "<< flops/(t1-t0)/NN<<std::endl;
-    err = ref-result; 
-    std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<<std::endl;
-
-    assert (norm2(err)< 1.0e-3 );
-    DwH.Report();
-  }
-
   if (1)
   {
 
@@ -241,10 +171,6 @@ int main (int argc, char ** argv)
     std::cout << GridLogMessage<< "* Vectorising fifth dimension by "<<vComplex::Nsimd()<<std::endl;
     if ( sizeof(Real)==4 )   std::cout << GridLogMessage<< "* SINGLE precision "<<std::endl;
     if ( sizeof(Real)==8 )   std::cout << GridLogMessage<< "* DOUBLE precision "<<std::endl;
-#ifdef GRID_OMP
-  if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsAndCompute ) std::cout << GridLogMessage<< "* Using Overlapped Comms/Compute" <<std::endl;
-  if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsThenCompute) std::cout << GridLogMessage<< "* Using sequential comms compute" <<std::endl;
-#endif
     if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptGeneric   ) std::cout << GridLogMessage<< "* Using GENERIC Nc WilsonKernels" <<std::endl;
     if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptHandUnroll) std::cout << GridLogMessage<< "* Using Nc=3       WilsonKernels" <<std::endl;
     if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptInlineAsm ) std::cout << GridLogMessage<< "* Using Asm Nc=3   WilsonKernels" <<std::endl;
@@ -256,13 +182,21 @@ int main (int argc, char ** argv)
     LatticeFermion sresult(sFGrid);
 
     WilsonFermion5DR sDw(Umu,*sFGrid,*sFrbGrid,*sUGrid,*sUrbGrid,M5);
-
+  
-    localConvert(src,ssrc);
+    for(int x=0;x<latt4[0];x++){
+    for(int y=0;y<latt4[1];y++){
+    for(int z=0;z<latt4[2];z++){
+    for(int t=0;t<latt4[3];t++){
+    for(int s=0;s<Ls;s++){
+      std::vector<int> site({s,x,y,z,t});
+      SpinColourVector tmp;
+      peekSite(tmp,src,site);
+      pokeSite(tmp,ssrc,site);
+    }}}}}
     std::cout<<GridLogMessage<< "src norms "<< norm2(src)<<" " <<norm2(ssrc)<<std::endl;
     FGrid->Barrier();
-    sDw.Dhop(ssrc,sresult,0);
-    sDw.ZeroCounters();
     double t0=usecond();
+    sDw.ZeroCounters();
     for(int i=0;i<ncall;i++){
       __SSC_START;
       sDw.Dhop(ssrc,sresult,0);
@@ -276,52 +210,46 @@ int main (int argc, char ** argv)
     std::cout<<GridLogMessage << "Called Dw s_inner "<<ncall<<" times in "<<t1-t0<<" us"<<std::endl;
     std::cout<<GridLogMessage << "mflop/s =   "<< flops/(t1-t0)<<std::endl;
     std::cout<<GridLogMessage << "mflop/s per rank =  "<< flops/(t1-t0)/NP<<std::endl;
-    std::cout<<GridLogMessage << "mflop/s per node =  "<< flops/(t1-t0)/NN<<std::endl;
-    //    std::cout<<GridLogMessage<< "res norms "<< norm2(result)<<" " <<norm2(sresult)<<std::endl;
     sDw.Report();
+  
+    if(0){
+      for(int i=0;i< PerformanceCounter::NumTypes(); i++ ){
+	sDw.Dhop(ssrc,sresult,0);
+	PerformanceCounter Counter(i);
+	Counter.Start();
+	sDw.Dhop(ssrc,sresult,0);
+	Counter.Stop();
+	Counter.Report();
+      }
+    }
+
+    std::cout<<GridLogMessage<< "res norms "<< norm2(result)<<" " <<norm2(sresult)<<std::endl;
+
     RealD sum=0;
+    for(int x=0;x<latt4[0];x++){
+    for(int y=0;y<latt4[1];y++){
+    for(int z=0;z<latt4[2];z++){
+    for(int t=0;t<latt4[3];t++){
+    for(int s=0;s<Ls;s++){
+      std::vector<int> site({s,x,y,z,t});
+      SpinColourVector normal, simd;
+      peekSite(normal,result,site);
+      peekSite(simd,sresult,site);
+      sum=sum+norm2(normal-simd);
+      if (norm2(normal-simd) > 1.0e-6 ) {
+	std::cout << "site "<<x<<","<<y<<","<<z<<","<<t<<","<<s<<" "<<norm2(normal-simd)<<std::endl;
+	std::cout << "site "<<x<<","<<y<<","<<z<<","<<t<<","<<s<<" normal "<<normal<<std::endl;
+	std::cout << "site "<<x<<","<<y<<","<<z<<","<<t<<","<<s<<" simd   "<<simd<<std::endl;
+      }
+    }}}}}
+    std::cout<<GridLogMessage<<" difference between normal and simd is "<<sum<<std::endl;
+    assert (sum< 1.0e-4 );
 
-    err=zero;
-    localConvert(sresult,err);
-    err = err - ref;
-    sum = norm2(err);
-    std::cout<<GridLogMessage<<" difference between normal ref and simd is "<<sum<<std::endl;
-    if(sum > 1.0e-4 ){
-      std::cout<< "sD REF\n " <<ref << std::endl;
-      std::cout<< "sD ERR   \n " <<err  <<std::endl;
-    }
-    //    assert(sum < 1.0e-4);
 
-    err=zero;
+    if (1) {
-    localConvert(sresult,err);
-    err = err - result;
-    sum = norm2(err);
-    std::cout<<GridLogMessage<<" difference between normal result and simd is "<<sum<<std::endl;
-    if(sum > 1.0e-4 ){
-      std::cout<< "sD REF\n " <<result << std::endl;
-      std::cout<< "sD ERR   \n " << err  <<std::endl;
-    }
-    assert(sum < 1.0e-4);
-    
-    if(1){
-      std::cout << GridLogMessage<< "*********************************************************" <<std::endl;
-      std::cout << GridLogMessage<< "* Benchmarking WilsonFermion5D<DomainWallVec5dImplR>::DhopEO "<<std::endl;
-      std::cout << GridLogMessage<< "* Vectorising fifth dimension by "<<vComplex::Nsimd()<<std::endl;
-      if ( sizeof(Real)==4 )   std::cout << GridLogMessage<< "* SINGLE precision "<<std::endl;
-      if ( sizeof(Real)==8 )   std::cout << GridLogMessage<< "* DOUBLE precision "<<std::endl;
-#ifdef GRID_OMP
-  if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsAndCompute ) std::cout << GridLogMessage<< "* Using Overlapped Comms/Compute" <<std::endl;
-  if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsThenCompute) std::cout << GridLogMessage<< "* Using sequential comms compute" <<std::endl;
-#endif
-      if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptGeneric   ) 
-	std::cout << GridLogMessage<< "* Using GENERIC Nc WilsonKernels" <<std::endl;
-      if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptHandUnroll) 
-	std::cout << GridLogMessage<< "* Using Nc=3       WilsonKernels" <<std::endl;
-      if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptInlineAsm ) 
-	std::cout << GridLogMessage<< "* Using Asm Nc=3   WilsonKernels" <<std::endl;
-      std::cout << GridLogMessage<< "*********************************************************" <<std::endl;
 
       LatticeFermion sr_eo(sFGrid);
+
       LatticeFermion ssrc_e (sFrbGrid);
       LatticeFermion ssrc_o (sFrbGrid);
       LatticeFermion sr_e   (sFrbGrid);
@@ -329,30 +257,39 @@ int main (int argc, char ** argv)
 
       pickCheckerboard(Even,ssrc_e,ssrc);
       pickCheckerboard(Odd,ssrc_o,ssrc);
-      //      setCheckerboard(sr_eo,ssrc_o);
+
-      //      setCheckerboard(sr_eo,ssrc_e);
+      setCheckerboard(sr_eo,ssrc_o);
+      setCheckerboard(sr_eo,ssrc_e);
 
       sr_e = zero;
       sr_o = zero;
 
+      std::cout << GridLogMessage<< "*********************************************************" <<std::endl;
+      std::cout << GridLogMessage<< "* Benchmarking WilsonFermion5D<DomainWallVec5dImplR>::DhopEO "<<std::endl;
+      std::cout << GridLogMessage<< "* Vectorising fifth dimension by "<<vComplex::Nsimd()<<std::endl;
+      if ( sizeof(Real)==4 )   std::cout << GridLogMessage<< "* SINGLE precision "<<std::endl;
+      if ( sizeof(Real)==8 )   std::cout << GridLogMessage<< "* DOUBLE precision "<<std::endl;
+      if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptGeneric   ) std::cout << GridLogMessage<< "* Using GENERIC Nc WilsonKernels" <<std::endl;
+      if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptHandUnroll) std::cout << GridLogMessage<< "* Using Nc=3       WilsonKernels" <<std::endl;
+      if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptInlineAsm ) std::cout << GridLogMessage<< "* Using Asm Nc=3   WilsonKernels" <<std::endl;
+      std::cout << GridLogMessage<< "*********************************************************" <<std::endl;
+
       FGrid->Barrier();
-      sDw.DhopEO(ssrc_o, sr_e, DaggerNo);
       sDw.ZeroCounters();
-      //      sDw.stat.init("DhopEO");
+      sDw.stat.init("DhopEO");
       double t0=usecond();
       for (int i = 0; i < ncall; i++) {
         sDw.DhopEO(ssrc_o, sr_e, DaggerNo);
       }
       double t1=usecond();
       FGrid->Barrier();
-      //      sDw.stat.print();
+      sDw.stat.print();
 
       double volume=Ls;  for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
       double flops=(1344.0*volume*ncall)/2;
 
       std::cout<<GridLogMessage << "sDeo mflop/s =   "<< flops/(t1-t0)<<std::endl;
       std::cout<<GridLogMessage << "sDeo mflop/s per rank   "<< flops/(t1-t0)/NP<<std::endl;
-      std::cout<<GridLogMessage << "sDeo mflop/s per node   "<< flops/(t1-t0)/NN<<std::endl;
       sDw.Report();
 
       sDw.DhopEO(ssrc_o,sr_e,DaggerNo);
@@ -361,26 +298,22 @@ int main (int argc, char ** argv)
 
       pickCheckerboard(Even,ssrc_e,sresult);
       pickCheckerboard(Odd ,ssrc_o,sresult);
-
       ssrc_e = ssrc_e - sr_e;
       RealD error = norm2(ssrc_e);
-      std::cout<<GridLogMessage << "sE norm diff   "<< norm2(ssrc_e)<< "  vec nrm"<<norm2(sr_e) <<std::endl;
 
+      std::cout<<GridLogMessage << "sE norm diff   "<< norm2(ssrc_e)<< "  vec nrm"<<norm2(sr_e) <<std::endl;
       ssrc_o = ssrc_o - sr_o;
+
       error+= norm2(ssrc_o);
       std::cout<<GridLogMessage << "sO norm diff   "<< norm2(ssrc_o)<< "  vec nrm"<<norm2(sr_o) <<std::endl;
-
+      if(error>1.0e-4) { 
-      if(( error>1.0e-4) ) { 
 	setCheckerboard(ssrc,ssrc_o);
 	setCheckerboard(ssrc,ssrc_e);
-	std::cout<< "DIFF\n " <<ssrc << std::endl;
+	std::cout<< ssrc << std::endl;
-	setCheckerboard(ssrc,sr_o);
-	setCheckerboard(ssrc,sr_e);
-	std::cout<< "CBRESULT\n " <<ssrc << std::endl;
-	std::cout<< "RESULT\n " <<sresult<< std::endl;
       }
-      assert(error<1.0e-4);
     }
+
+
   }
 
   if (1)
@@ -391,30 +324,25 @@ int main (int argc, char ** argv)
       //    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]; ;
+  ref._odata[i]+= tmp._odata[i] + Gamma(Gmu[mu])*tmp._odata[i]; ;
       }
 
       tmp =adj(U[mu])*src;
       tmp =Cshift(tmp,mu+1,-1);
       for(int i=0;i<ref._odata.size();i++){
-	ref._odata[i]+= tmp._odata[i] - Gamma(Gmu[mu])*tmp._odata[i]; ;
+  ref._odata[i]+= tmp._odata[i] - Gamma(Gmu[mu])*tmp._odata[i]; ;
       }
     }
     ref = -0.5*ref;
   }
-  //  dump=1;
   Dw.Dhop(src,result,1);
   std::cout << GridLogMessage << "Compare to naive wilson implementation Dag to verify correctness" << std::endl;
   std::cout<<GridLogMessage << "Called DwDag"<<std::endl;
-  std::cout<<GridLogMessage << "norm dag result "<< norm2(result)<<std::endl;
+  std::cout<<GridLogMessage << "norm result "<< norm2(result)<<std::endl;
-  std::cout<<GridLogMessage << "norm dag ref    "<< norm2(ref)<<std::endl;
+  std::cout<<GridLogMessage << "norm ref    "<< norm2(ref)<<std::endl;
   err = ref-result; 
-  std::cout<<GridLogMessage << "norm dag diff   "<< norm2(err)<<std::endl;
+  std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<<std::endl;
-  if((norm2(err)>1.0e-4)){
+  assert(norm2(err)<1.0e-4);
-	std::cout<< "DAG RESULT\n "  <<ref     << std::endl;
-	std::cout<< "DAG sRESULT\n " <<result  << std::endl;
-	std::cout<< "DAG ERR   \n "  << err    <<std::endl;
-  }
   LatticeFermion src_e (FrbGrid);
   LatticeFermion src_o (FrbGrid);
   LatticeFermion r_e   (FrbGrid);
@@ -422,24 +350,18 @@ int main (int argc, char ** argv)
   LatticeFermion r_eo  (FGrid);
 
 
-  std::cout<<GridLogMessage << "Calling Deo and Doe and //assert Deo+Doe == Dunprec"<<std::endl;
+  std::cout<<GridLogMessage << "Calling Deo and Doe and assert Deo+Doe == Dunprec"<<std::endl;
   pickCheckerboard(Even,src_e,src);
   pickCheckerboard(Odd,src_o,src);
 
   std::cout<<GridLogMessage << "src_e"<<norm2(src_e)<<std::endl;
   std::cout<<GridLogMessage << "src_o"<<norm2(src_o)<<std::endl;
 
-
-  // S-direction is INNERMOST and takes no part in the parity.
   std::cout << GridLogMessage<< "*********************************************************" <<std::endl;
   std::cout << GridLogMessage<< "* Benchmarking DomainWallFermionR::DhopEO                "<<std::endl;
   std::cout << GridLogMessage<< "* Vectorising space-time by "<<vComplex::Nsimd()<<std::endl;
   if ( sizeof(Real)==4 )   std::cout << GridLogMessage<< "* SINGLE precision "<<std::endl;
   if ( sizeof(Real)==8 )   std::cout << GridLogMessage<< "* DOUBLE precision "<<std::endl;
-#ifdef GRID_OMP
-  if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsAndCompute ) std::cout << GridLogMessage<< "* Using Overlapped Comms/Compute" <<std::endl;
-  if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsThenCompute) std::cout << GridLogMessage<< "* Using sequential comms compute" <<std::endl;
-#endif
   if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptGeneric   ) std::cout << GridLogMessage<< "* Using GENERIC Nc WilsonKernels" <<std::endl;
   if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptHandUnroll) std::cout << GridLogMessage<< "* Using Nc=3       WilsonKernels" <<std::endl;
   if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptInlineAsm ) std::cout << GridLogMessage<< "* Using Asm Nc=3   WilsonKernels" <<std::endl;
@@ -447,7 +369,6 @@ int main (int argc, char ** argv)
   {
     Dw.ZeroCounters();
     FGrid->Barrier();
-    Dw.DhopEO(src_o,r_e,DaggerNo);
     double t0=usecond();
     for(int i=0;i<ncall;i++){
       Dw.DhopEO(src_o,r_e,DaggerNo);
@@ -460,7 +381,6 @@ int main (int argc, char ** argv)
 
     std::cout<<GridLogMessage << "Deo mflop/s =   "<< flops/(t1-t0)<<std::endl;
     std::cout<<GridLogMessage << "Deo mflop/s per rank   "<< flops/(t1-t0)/NP<<std::endl;
-    std::cout<<GridLogMessage << "Deo mflop/s per node   "<< flops/(t1-t0)/NN<<std::endl;
     Dw.Report();
   }
   Dw.DhopEO(src_o,r_e,DaggerNo);
@@ -476,20 +396,14 @@ int main (int argc, char ** argv)
 
   err = r_eo-result; 
   std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<<std::endl;
-  if((norm2(err)>1.0e-4)){
+  assert(norm2(err)<1.0e-4);
-	std::cout<< "Deo RESULT\n " <<r_eo << std::endl;
-	std::cout<< "Deo REF\n " <<result  << std::endl;
-	std::cout<< "Deo ERR   \n " << err <<std::endl;
-  }
 
   pickCheckerboard(Even,src_e,err);
   pickCheckerboard(Odd,src_o,err);
   std::cout<<GridLogMessage << "norm diff even  "<< norm2(src_e)<<std::endl;
   std::cout<<GridLogMessage << "norm diff odd   "<< norm2(src_o)<<std::endl;
-
+  assert(norm2(src_e)<1.0e-4);
-  //assert(norm2(src_e)<1.0e-4);
+  assert(norm2(src_o)<1.0e-4);
-  //assert(norm2(src_o)<1.0e-4);
 
   Grid_finalize();
 }
-

benchmarks/Benchmark_memory_asynch.cc

@@ -66,8 +66,7 @@ int main (int argc, char ** argv)
 
     Vec tsum; tsum = zero;
 
-    GridParallelRNG          pRNG(&Grid);      
+    GridParallelRNG          pRNG(&Grid);      pRNG.SeedRandomDevice();
-    pRNG.SeedFixedIntegers(std::vector<int>({56,17,89,101}));
 
     std::vector<double> stop(threads);
     Vector<Vec> sum(threads);
@@ -78,7 +77,8 @@ int main (int argc, char ** argv)
     }
 
     double start=usecond();
-    parallel_for(int t=0;t<threads;t++){
+PARALLEL_FOR_LOOP
+    for(int t=0;t<threads;t++){
 
       sum[t] = x[t]._odata[0];
       for(int i=0;i<Nloop;i++){

benchmarks/Benchmark_memory_bandwidth.cc

@@ -65,7 +65,7 @@ int main (int argc, char ** argv)
 
       uint64_t Nloop=NLOOP;
 
-      //      GridParallelRNG          pRNG(&Grid);      pRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9});
+      //      GridParallelRNG          pRNG(&Grid);      pRNG.SeedRandomDevice();
 
       LatticeVec z(&Grid); //random(pRNG,z);
       LatticeVec x(&Grid); //random(pRNG,x);
@@ -100,7 +100,7 @@ int main (int argc, char ** argv)
       int vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
       GridCartesian     Grid(latt_size,simd_layout,mpi_layout);
 
-      //      GridParallelRNG          pRNG(&Grid);      pRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9});
+      //      GridParallelRNG          pRNG(&Grid);      pRNG.SeedRandomDevice();
 
       LatticeVec z(&Grid); //random(pRNG,z);
       LatticeVec x(&Grid); //random(pRNG,x);
@@ -138,7 +138,7 @@ int main (int argc, char ** argv)
 
       GridCartesian     Grid(latt_size,simd_layout,mpi_layout);
 
-      //      GridParallelRNG          pRNG(&Grid);      pRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9});
+      //      GridParallelRNG          pRNG(&Grid);      pRNG.SeedRandomDevice();
 
       LatticeVec z(&Grid); //random(pRNG,z);
       LatticeVec x(&Grid); //random(pRNG,x);
@@ -173,7 +173,7 @@ int main (int argc, char ** argv)
       uint64_t Nloop=NLOOP;
       GridCartesian     Grid(latt_size,simd_layout,mpi_layout);
 
-      //      GridParallelRNG          pRNG(&Grid);      pRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9});
+      //      GridParallelRNG          pRNG(&Grid);      pRNG.SeedRandomDevice();
       LatticeVec z(&Grid); //random(pRNG,z);
       LatticeVec x(&Grid); //random(pRNG,x);
       LatticeVec y(&Grid); //random(pRNG,y);

benchmarks/Benchmark_staggered.cc

@@ -1,134 +0,0 @@
-    /*************************************************************************************
-
-    Grid physics library, www.github.com/paboyle/Grid 
-
-    Source file: ./benchmarks/Benchmark_staggered.cc
-
-    Copyright (C) 2015
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License along
-    with this program; if not, write to the Free Software Foundation, Inc.,
-    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-    See the full license in the file "LICENSE" in the top level distribution directory
-    *************************************************************************************/
-    /*  END LEGAL */
-#include <Grid/Grid.h>
-
-using namespace std;
-using namespace Grid;
-using namespace Grid::QCD;
-
-int main (int argc, char ** argv)
-{
-  Grid_init(&argc,&argv);
-
-  std::vector<int> latt_size   = GridDefaultLatt();
-  std::vector<int> simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd());
-  std::vector<int> mpi_layout  = GridDefaultMpi();
-  GridCartesian               Grid(latt_size,simd_layout,mpi_layout);
-  GridRedBlackCartesian     RBGrid(latt_size,simd_layout,mpi_layout);
-
-  int threads = GridThread::GetThreads();
-  std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
-  std::cout<<GridLogMessage << "Grid floating point word size is REALF"<< sizeof(RealF)<<std::endl;
-  std::cout<<GridLogMessage << "Grid floating point word size is REALD"<< sizeof(RealD)<<std::endl;
-  std::cout<<GridLogMessage << "Grid floating point word size is REAL"<< sizeof(Real)<<std::endl;
-
-  std::vector<int> seeds({1,2,3,4});
-  GridParallelRNG          pRNG(&Grid);
-  pRNG.SeedFixedIntegers(seeds);
-  //  pRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9});
-
-  typedef typename ImprovedStaggeredFermionR::FermionField FermionField; 
-  typename ImprovedStaggeredFermionR::ImplParams params; 
-
-  FermionField src   (&Grid); random(pRNG,src);
-  FermionField result(&Grid); result=zero;
-  FermionField    ref(&Grid);    ref=zero;
-  FermionField    tmp(&Grid);    tmp=zero;
-  FermionField    err(&Grid);    tmp=zero;
-  LatticeGaugeField Umu(&Grid); random(pRNG,Umu);
-  std::vector<LatticeColourMatrix> U(4,&Grid);
-
-  double volume=1;
-  for(int mu=0;mu<Nd;mu++){
-    volume=volume*latt_size[mu];
-  }  
-
-  // Only one non-zero (y)
-#if 0
-  Umu=zero;
-  Complex cone(1.0,0.0);
-  for(int nn=0;nn<Nd;nn++){
-    random(pRNG,U[nn]);
-    if(1) {
-      if (nn!=2) { U[nn]=zero; std::cout<<GridLogMessage << "zeroing gauge field in dir "<<nn<<std::endl; }
-      //      else       { U[nn]= cone;std::cout<<GridLogMessage << "unit gauge field in dir "<<nn<<std::endl; }
-      else       { std::cout<<GridLogMessage << "random gauge field in dir "<<nn<<std::endl; }
-    }
-    PokeIndex<LorentzIndex>(Umu,U[nn],nn);
-  }
-#endif
-
-  for(int mu=0;mu<Nd;mu++){
-    U[mu] = PeekIndex<LorentzIndex>(Umu,mu);
-  }
-  ref = zero;
-  /*  
-  { // Naive wilson implementation
-    ref = zero;
-    for(int mu=0;mu<Nd;mu++){
-      //    ref =  src + Gamma(Gamma::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]; ;
-      }
-
-      tmp =adj(U[mu])*src;
-      tmp =Cshift(tmp,mu,-1);
-      for(int i=0;i<ref._odata.size();i++){
-	ref._odata[i]+= tmp._odata[i] + Gamma(Gmu[mu])*tmp._odata[i]; ;
-      }
-    }
-  }
-  ref = -0.5*ref;
-  */
-
-  RealD mass=0.1;
-  RealD c1=9.0/8.0;
-  RealD c2=-1.0/24.0;
-  RealD u0=1.0;
-  ImprovedStaggeredFermionR Ds(Umu,Umu,Grid,RBGrid,mass,c1,c2,u0,params);
-  
-  std::cout<<GridLogMessage << "Calling Ds"<<std::endl;
-  int ncall=1000;
-  double t0=usecond();
-  for(int i=0;i<ncall;i++){
-    Ds.Dhop(src,result,0);
-  }
-  double t1=usecond();
-  double flops=(16*(3*(6+8+8)) + 15*3*2)*volume*ncall; // == 66*16 +  == 1146
-  
-  std::cout<<GridLogMessage << "Called Ds"<<std::endl;
-  std::cout<<GridLogMessage << "norm result "<< norm2(result)<<std::endl;
-  std::cout<<GridLogMessage << "norm ref    "<< norm2(ref)<<std::endl;
-  std::cout<<GridLogMessage << "mflop/s =   "<< flops/(t1-t0)<<std::endl;
-  err = ref-result; 
-  std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<<std::endl;
-
-  Grid_finalize();
-}

benchmarks/Benchmark_su3.cc

@@ -35,9 +35,8 @@ using namespace Grid::QCD;
 int main (int argc, char ** argv)
 {
   Grid_init(&argc,&argv);
-#define LMAX (32)
 
-  int Nloop=200;
+  int Nloop=1000;
 
   std::vector<int> simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd());
   std::vector<int> mpi_layout  = GridDefaultMpi();
@@ -51,12 +50,12 @@ int main (int argc, char ** argv)
   std::cout<<GridLogMessage << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
   std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
 
-  for(int lat=2;lat<=LMAX;lat+=2){
+  for(int lat=2;lat<=32;lat+=2){
 
       std::vector<int> latt_size  ({lat*mpi_layout[0],lat*mpi_layout[1],lat*mpi_layout[2],lat*mpi_layout[3]});
       int vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
       GridCartesian     Grid(latt_size,simd_layout,mpi_layout);
-      //      GridParallelRNG          pRNG(&Grid);      pRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9});
+      //      GridParallelRNG          pRNG(&Grid);      pRNG.SeedRandomDevice();
 
       LatticeColourMatrix z(&Grid);// random(pRNG,z);
       LatticeColourMatrix x(&Grid);// random(pRNG,x);
@@ -83,13 +82,13 @@ int main (int argc, char ** argv)
   std::cout<<GridLogMessage << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
   std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
 
-  for(int lat=2;lat<=LMAX;lat+=2){
+  for(int lat=2;lat<=32;lat+=2){
 
       std::vector<int> latt_size  ({lat*mpi_layout[0],lat*mpi_layout[1],lat*mpi_layout[2],lat*mpi_layout[3]});
       int vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
 
       GridCartesian     Grid(latt_size,simd_layout,mpi_layout);
-      //      GridParallelRNG          pRNG(&Grid);      pRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9});
+      //      GridParallelRNG          pRNG(&Grid);      pRNG.SeedRandomDevice();
 
       LatticeColourMatrix z(&Grid); //random(pRNG,z);
       LatticeColourMatrix x(&Grid); //random(pRNG,x);
@@ -114,13 +113,13 @@ int main (int argc, char ** argv)
   std::cout<<GridLogMessage << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
   std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
 
-  for(int lat=2;lat<=LMAX;lat+=2){
+  for(int lat=2;lat<=32;lat+=2){
 
       std::vector<int> latt_size  ({lat*mpi_layout[0],lat*mpi_layout[1],lat*mpi_layout[2],lat*mpi_layout[3]});
       int vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
 
       GridCartesian     Grid(latt_size,simd_layout,mpi_layout);
-      //      GridParallelRNG          pRNG(&Grid);      pRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9});
+      //      GridParallelRNG          pRNG(&Grid);      pRNG.SeedRandomDevice();
 
       LatticeColourMatrix z(&Grid); //random(pRNG,z);
       LatticeColourMatrix x(&Grid); //random(pRNG,x);
@@ -145,13 +144,13 @@ int main (int argc, char ** argv)
   std::cout<<GridLogMessage << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
   std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
 
-  for(int lat=2;lat<=LMAX;lat+=2){
+  for(int lat=2;lat<=32;lat+=2){
 
       std::vector<int> latt_size  ({lat*mpi_layout[0],lat*mpi_layout[1],lat*mpi_layout[2],lat*mpi_layout[3]});
       int vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
 
       GridCartesian     Grid(latt_size,simd_layout,mpi_layout);
-      //      GridParallelRNG          pRNG(&Grid);      pRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9});
+      //      GridParallelRNG          pRNG(&Grid);      pRNG.SeedRandomDevice();
 
       LatticeColourMatrix z(&Grid); //random(pRNG,z);
       LatticeColourMatrix x(&Grid); //random(pRNG,x);

benchmarks/Benchmark_wilson.cc

@@ -69,7 +69,7 @@ int main (int argc, char ** argv)
   std::vector<int> seeds({1,2,3,4});
   GridParallelRNG          pRNG(&Grid);
   pRNG.SeedFixedIntegers(seeds);
-  //  pRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9});
+  //  pRNG.SeedRandomDevice();
 
   LatticeFermion src   (&Grid); random(pRNG,src);
   LatticeFermion result(&Grid); result=zero;

benchmarks/Makefile.am

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

bootstrap.sh

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

configure.ac

@@ -1,19 +1,16 @@
 AC_PREREQ([2.63])
-AC_INIT([Grid], [0.7.0], [https://github.com/paboyle/Grid], [Grid])
+AC_INIT([Grid], [0.6.0], [https://github.com/paboyle/Grid], [Grid])
 AC_CANONICAL_BUILD
 AC_CANONICAL_HOST
 AC_CANONICAL_TARGET
-AM_INIT_AUTOMAKE([subdir-objects 1.13])
+AM_INIT_AUTOMAKE(subdir-objects)
-AM_EXTRA_RECURSIVE_TARGETS([tests bench])
 AC_CONFIG_MACRO_DIR([m4])
 AC_CONFIG_SRCDIR([lib/Grid.h])
 AC_CONFIG_HEADERS([lib/Config.h],[sed -i 's|PACKAGE_|GRID_|' lib/Config.h])
 m4_ifdef([AM_SILENT_RULES], [AM_SILENT_RULES([yes])])
 
-################ Get git info
-#AC_REVISION([m4_esyscmd_s([./scripts/configure.commit])])
-
 ############### Checks for programs
+CXXFLAGS="-O3 $CXXFLAGS"
 AC_PROG_CXX
 AC_PROG_RANLIB
 
@@ -27,15 +24,12 @@ AX_GXX_VERSION
 AC_DEFINE_UNQUOTED([GXX_VERSION],["$GXX_VERSION"],
       [version of g++ that will compile the code])
 
-CXXFLAGS="-O3 $CXXFLAGS"
-
-
 ############### Checks for typedefs, structures, and compiler characteristics
 AC_TYPE_SIZE_T
 AC_TYPE_UINT32_T
 AC_TYPE_UINT64_T
 
-############### OpenMP
+############### OpenMP 
 AC_OPENMP
 ac_openmp=no
 if test "${OPENMP_CXXFLAGS}X" != "X"; then
@@ -66,23 +60,16 @@ AC_ARG_WITH([mpfr],
     [AM_CXXFLAGS="-I$with_mpfr/include $AM_CXXFLAGS"]
     [AM_LDFLAGS="-L$with_mpfr/lib $AM_LDFLAGS"])
 
-############### FFTW3
+############### FFTW3 
-AC_ARG_WITH([fftw],
+AC_ARG_WITH([fftw],    
             [AS_HELP_STRING([--with-fftw=prefix],
             [try this for a non-standard install prefix of the FFTW3 library])],
             [AM_CXXFLAGS="-I$with_fftw/include $AM_CXXFLAGS"]
             [AM_LDFLAGS="-L$with_fftw/lib $AM_LDFLAGS"])
 
-############### LIME
+############### lapack 
-AC_ARG_WITH([lime],
-            [AS_HELP_STRING([--with-lime=prefix],
-            [try this for a non-standard install prefix of the LIME library])],
-            [AM_CXXFLAGS="-I$with_lime/include $AM_CXXFLAGS"]
-            [AM_LDFLAGS="-L$with_lime/lib $AM_LDFLAGS"])
-
-############### lapack
 AC_ARG_ENABLE([lapack],
-    [AC_HELP_STRING([--enable-lapack=yes|no|prefix], [enable LAPACK])],
+    [AC_HELP_STRING([--enable-lapack=yes|no|prefix], [enable LAPACK])], 
     [ac_LAPACK=${enable_lapack}], [ac_LAPACK=no])
 
 case ${ac_LAPACK} in
@@ -96,18 +83,6 @@ case ${ac_LAPACK} in
         AC_DEFINE([USE_LAPACK],[1],[use LAPACK]);;
 esac
 
-############### FP16 conversions
-AC_ARG_ENABLE([sfw-fp16],
-    [AC_HELP_STRING([--enable-sfw-fp16=yes|no], [enable software fp16 comms])],
-    [ac_SFW_FP16=${enable_sfw_fp16}], [ac_SFW_FP16=yes])
-case ${ac_SFW_FP16} in
-    yes)
-      AC_DEFINE([SFW_FP16],[1],[software conversion to fp16]);;
-    no);;
-    *)
-      AC_MSG_ERROR(["SFW FP16 option not supported ${ac_SFW_FP16}"]);;
-esac
-
 ############### MKL
 AC_ARG_ENABLE([mkl],
     [AC_HELP_STRING([--enable-mkl=yes|no|prefix], [enable Intel MKL for LAPACK & FFTW])],
@@ -133,7 +108,7 @@ AC_ARG_WITH([hdf5],
 
 ############### first-touch
 AC_ARG_ENABLE([numa],
-    [AC_HELP_STRING([--enable-numa=yes|no|prefix], [enable first touch numa opt])],
+    [AC_HELP_STRING([--enable-numa=yes|no|prefix], [enable first touch numa opt])], 
     [ac_NUMA=${enable_NUMA}],[ac_NUMA=no])
 
 case ${ac_NUMA} in
@@ -159,8 +134,8 @@ if test "${ac_MKL}x" != "nox"; then
 fi
 
 AC_SEARCH_LIBS([__gmpf_init], [gmp],
-               [AC_SEARCH_LIBS([mpfr_init], [mpfr],
+               [AC_SEARCH_LIBS([mpfr_init], [mpfr], 
-                               [AC_DEFINE([HAVE_LIBMPFR], [1],
+                               [AC_DEFINE([HAVE_LIBMPFR], [1], 
                                           [Define to 1 if you have the `MPFR' library])]
                                [have_mpfr=true], [AC_MSG_ERROR([MPFR library not found])])]
                [AC_DEFINE([HAVE_LIBGMP], [1], [Define to 1 if you have the `GMP' library])]
@@ -169,7 +144,7 @@ AC_SEARCH_LIBS([__gmpf_init], [gmp],
 if test "${ac_LAPACK}x" != "nox"; then
     AC_SEARCH_LIBS([LAPACKE_sbdsdc], [lapack], [],
                    [AC_MSG_ERROR("LAPACK enabled but library not found")])
-fi
+fi   
 
 AC_SEARCH_LIBS([fftw_execute], [fftw3],
                [AC_SEARCH_LIBS([fftwf_execute], [fftw3f], [],
@@ -177,14 +152,6 @@ AC_SEARCH_LIBS([fftw_execute], [fftw3],
                [AC_DEFINE([HAVE_FFTW], [1], [Define to 1 if you have the `FFTW' library])]
                [have_fftw=true])
 
-AC_SEARCH_LIBS([limeCreateReader], [lime],
-               [AC_DEFINE([HAVE_LIME], [1], [Define to 1 if you have the `LIME' library])]
-               [have_lime=true],
-	       [AC_MSG_WARN(C-LIME library was not found in your system.
-In order to use ILGG file format please install or provide the correct path to your installation
-Info at: http://usqcd.jlab.org/usqcd-docs/c-lime/)])
-
-
 AC_SEARCH_LIBS([H5Fopen], [hdf5_cpp],
                [AC_DEFINE([HAVE_HDF5], [1], [Define to 1 if you have the `HDF5' library])]
                [have_hdf5=true]
@@ -209,26 +176,19 @@ case ${ax_cv_cxx_compiler_vendor} in
     case ${ac_SIMD} in
       SSE4)
         AC_DEFINE([SSE4],[1],[SSE4 intrinsics])
-	case ${ac_SFW_FP16} in
+        SIMD_FLAGS='-msse4.2';;
-	  yes)
-	  SIMD_FLAGS='-msse4.2';;
-	  no)
-	  SIMD_FLAGS='-msse4.2 -mf16c';;
-	  *)
-          AC_MSG_ERROR(["SFW_FP16 must be either yes or no value ${ac_SFW_FP16} "]);;
-	esac;;
       AVX)
         AC_DEFINE([AVX1],[1],[AVX intrinsics])
-        SIMD_FLAGS='-mavx -mf16c';;
+        SIMD_FLAGS='-mavx';;
       AVXFMA4)
         AC_DEFINE([AVXFMA4],[1],[AVX intrinsics with FMA4])
-        SIMD_FLAGS='-mavx -mfma4 -mf16c';;
+        SIMD_FLAGS='-mavx -mfma4';;
       AVXFMA)
         AC_DEFINE([AVXFMA],[1],[AVX intrinsics with FMA3])
-        SIMD_FLAGS='-mavx -mfma -mf16c';;
+        SIMD_FLAGS='-mavx -mfma';;
       AVX2)
         AC_DEFINE([AVX2],[1],[AVX2 intrinsics])
-        SIMD_FLAGS='-mavx2 -mfma -mf16c';;
+        SIMD_FLAGS='-mavx2 -mfma';;
       AVX512)
         AC_DEFINE([AVX512],[1],[AVX512 intrinsics])
         SIMD_FLAGS='-mavx512f -mavx512pf -mavx512er -mavx512cd';;
@@ -337,7 +297,7 @@ case ${ac_COMMS} in
         comms_type='shmem'
      ;;
      *)
-        AC_MSG_ERROR([${ac_COMMS} unsupported --enable-comms option]);
+        AC_MSG_ERROR([${ac_COMMS} unsupported --enable-comms option]); 
      ;;
 esac
 case ${ac_COMMS} in
@@ -361,7 +321,7 @@ AM_CONDITIONAL(BUILD_COMMS_NONE,  [ test "${comms_type}X" == "noneX" ])
 ############### RNG selection
 AC_ARG_ENABLE([rng],[AC_HELP_STRING([--enable-rng=ranlux48|mt19937|sitmo],\
 	            [Select Random Number Generator to be used])],\
-	            [ac_RNG=${enable_rng}],[ac_RNG=sitmo])
+	            [ac_RNG=${enable_rng}],[ac_RNG=ranlux48])
 
 case ${ac_RNG} in
      ranlux48)
@@ -374,7 +334,7 @@ case ${ac_RNG} in
       AC_DEFINE([RNG_SITMO],[1],[RNG_SITMO] )
      ;;
      *)
-      AC_MSG_ERROR([${ac_RNG} unsupported --enable-rng option]);
+      AC_MSG_ERROR([${ac_RNG} unsupported --enable-rng option]); 
      ;;
 esac
 
@@ -391,7 +351,7 @@ case ${ac_TIMERS} in
       AC_DEFINE([TIMERS_OFF],[1],[TIMERS_OFF] )
      ;;
      *)
-      AC_MSG_ERROR([${ac_TIMERS} unsupported --enable-timers option]);
+      AC_MSG_ERROR([${ac_TIMERS} unsupported --enable-timers option]); 
      ;;
 esac
 
@@ -403,7 +363,7 @@ case ${ac_CHROMA} in
      yes|no)
      ;;
      *)
-       AC_MSG_ERROR([${ac_CHROMA} unsupported --enable-chroma option]);
+       AC_MSG_ERROR([${ac_CHROMA} unsupported --enable-chroma option]); 
      ;;
 esac
 
@@ -424,65 +384,12 @@ DX_INIT_DOXYGEN([$PACKAGE_NAME], [doxygen.cfg])
 
 ############### Ouput
 cwd=`pwd -P`; cd ${srcdir}; abs_srcdir=`pwd -P`; cd ${cwd}
-GRID_CXXFLAGS="$AM_CXXFLAGS $CXXFLAGS"
-GRID_LDFLAGS="$AM_LDFLAGS $LDFLAGS"
-GRID_LIBS=$LIBS
-GRID_SHORT_SHA=`git rev-parse --short HEAD`
-GRID_SHA=`git rev-parse HEAD`
-GRID_BRANCH=`git rev-parse --abbrev-ref HEAD`
 AM_CXXFLAGS="-I${abs_srcdir}/include $AM_CXXFLAGS"
 AM_CFLAGS="-I${abs_srcdir}/include $AM_CFLAGS"
 AM_LDFLAGS="-L${cwd}/lib $AM_LDFLAGS"
 AC_SUBST([AM_CFLAGS])
 AC_SUBST([AM_CXXFLAGS])
 AC_SUBST([AM_LDFLAGS])
-AC_SUBST([GRID_CXXFLAGS])
-AC_SUBST([GRID_LDFLAGS])
-AC_SUBST([GRID_LIBS])
-AC_SUBST([GRID_SHA])
-AC_SUBST([GRID_BRANCH])
-
-git_commit=`cd $srcdir && ./scripts/configure.commit`
-
-echo "~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-Summary of configuration for $PACKAGE v$VERSION
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
------ GIT VERSION -------------------------------------
-$git_commit
------ PLATFORM ----------------------------------------
-architecture (build)        : $build_cpu
-os (build)                  : $build_os
-architecture (target)       : $target_cpu
-os (target)                 : $target_os
-compiler vendor             : ${ax_cv_cxx_compiler_vendor}
-compiler version            : ${ax_cv_gxx_version}
------ BUILD OPTIONS -----------------------------------
-SIMD                        : ${ac_SIMD}${SIMD_GEN_WIDTH_MSG}
-Threading                   : ${ac_openmp}
-Communications type         : ${comms_type}
-Default precision           : ${ac_PRECISION}
-Software FP16 conversion    : ${ac_SFW_FP16}
-RNG choice                  : ${ac_RNG}
-GMP                         : `if test "x$have_gmp" = xtrue; then echo yes; else echo no; fi`
-LAPACK                      : ${ac_LAPACK}
-FFTW                        : `if test "x$have_fftw" = xtrue; then echo yes; else echo no; fi`
-LIME (ILDG support)         : `if test "x$have_lime" = xtrue; then echo yes; else echo no; fi`
-HDF5                        : `if test "x$have_hdf5" = xtrue; then echo yes; else echo no; fi`
-build DOXYGEN documentation : `if test "$DX_FLAG_doc" = '1'; then echo yes; else echo no; fi`
------ BUILD FLAGS -------------------------------------
-CXXFLAGS:
-`echo ${AM_CXXFLAGS} ${CXXFLAGS} | tr ' ' '\n' | sed 's/^-/    -/g'`
-LDFLAGS:
-`echo ${AM_LDFLAGS} ${LDFLAGS} | tr ' ' '\n' | sed 's/^-/    -/g'`
-LIBS:
-`echo ${LIBS} | tr ' ' '\n' | sed 's/^-/    -/g'`
--------------------------------------------------------" > grid.configure.summary
-
-GRID_SUMMARY="`cat grid.configure.summary`"
-AM_SUBST_NOTMAKE([GRID_SUMMARY])
-AC_SUBST([GRID_SUMMARY])
-
-AC_CONFIG_FILES([grid-config], [chmod +x grid-config])
 AC_CONFIG_FILES(Makefile)
 AC_CONFIG_FILES(lib/Makefile)
 AC_CONFIG_FILES(tests/Makefile)
@@ -493,15 +400,42 @@ AC_CONFIG_FILES(tests/forces/Makefile)
 AC_CONFIG_FILES(tests/hadrons/Makefile)
 AC_CONFIG_FILES(tests/hmc/Makefile)
 AC_CONFIG_FILES(tests/solver/Makefile)
-AC_CONFIG_FILES(tests/smearing/Makefile)
 AC_CONFIG_FILES(tests/qdpxx/Makefile)
-AC_CONFIG_FILES(tests/testu01/Makefile)
 AC_CONFIG_FILES(benchmarks/Makefile)
 AC_CONFIG_FILES(extras/Makefile)
 AC_CONFIG_FILES(extras/Hadrons/Makefile)
 AC_OUTPUT
 
-echo ""
+echo "~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-cat grid.configure.summary
+Summary of configuration for $PACKAGE v$VERSION
-echo ""
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
+----- PLATFORM ----------------------------------------
+architecture (build)        : $build_cpu
+os (build)                  : $build_os
+architecture (target)       : $target_cpu
+os (target)                 : $target_os
+compiler vendor             : ${ax_cv_cxx_compiler_vendor}
+compiler version            : ${ax_cv_gxx_version}
+----- BUILD OPTIONS -----------------------------------
+SIMD                        : ${ac_SIMD}${SIMD_GEN_WIDTH_MSG}
+Threading                   : ${ac_openmp} 
+Communications type         : ${comms_type}
+Default precision           : ${ac_PRECISION}
+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:
+`echo ${AM_CXXFLAGS} ${CXXFLAGS} | tr ' ' '\n' | sed 's/^-/    -/g'`
+LDFLAGS:
+`echo ${AM_LDFLAGS} ${LDFLAGS} | tr ' ' '\n' | sed 's/^-/    -/g'`
+LIBS:
+`echo ${LIBS} | tr ' ' '\n' | sed 's/^-/    -/g'`
+-------------------------------------------------------" > config.summary
+echo ""
+cat config.summary
+echo ""

extras/Hadrons/Application.cc

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

extras/Hadrons/Global.hpp

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

extras/Hadrons/Modules.hpp

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

extras/Hadrons/Modules/MContraction/Baryon.hpp

@@ -112,7 +112,7 @@ void TBaryon<FImpl1, FImpl2, FImpl3>::execute(void)
                  << " quarks '" << par().q1 << "', '" << par().q2 << "', and '"
                  << par().q3 << "'" << std::endl;
     
-    CorrWriter             writer(par().output);
+    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);

extras/Hadrons/Modules/MContraction/DiscLoop.hpp

@@ -1,144 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid 
-
-Source file: extras/Hadrons/Modules/MContraction/DiscLoop.hpp
-
-Copyright (C) 2017
-
-Author: Andrew Lawson    <andrew.lawson1991@gmail.com>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-
-#ifndef Hadrons_DiscLoop_hpp_
-#define Hadrons_DiscLoop_hpp_
-
-#include <Grid/Hadrons/Global.hpp>
-#include <Grid/Hadrons/Module.hpp>
-#include <Grid/Hadrons/ModuleFactory.hpp>
-
-BEGIN_HADRONS_NAMESPACE
-
-/******************************************************************************
- *                                DiscLoop                                    *
- ******************************************************************************/
-BEGIN_MODULE_NAMESPACE(MContraction)
-
-class DiscLoopPar: Serializable
-{
-public:
-    GRID_SERIALIZABLE_CLASS_MEMBERS(DiscLoopPar,
-                                    std::string,    q_loop,
-                                    Gamma::Algebra, gamma,
-                                    std::string,    output);
-};
-
-template <typename FImpl>
-class TDiscLoop: public Module<DiscLoopPar>
-{
-    TYPE_ALIASES(FImpl,);
-    class Result: Serializable
-    {
-    public:
-        GRID_SERIALIZABLE_CLASS_MEMBERS(Result,
-                                        Gamma::Algebra, gamma,
-                                        std::vector<Complex>, corr);
-    };
-public:
-    // constructor
-    TDiscLoop(const std::string name);
-    // destructor
-    virtual ~TDiscLoop(void) = default;
-    // dependency relation
-    virtual std::vector<std::string> getInput(void);
-    virtual std::vector<std::string> getOutput(void);
-    // setup
-    virtual void setup(void);
-    // execution
-    virtual void execute(void);
-};
-
-MODULE_REGISTER_NS(DiscLoop, TDiscLoop<FIMPL>, MContraction);
-
-/******************************************************************************
- *                       TDiscLoop implementation                             *
- ******************************************************************************/
-// constructor /////////////////////////////////////////////////////////////////
-template <typename FImpl>
-TDiscLoop<FImpl>::TDiscLoop(const std::string name)
-: Module<DiscLoopPar>(name)
-{}
-
-// dependencies/products ///////////////////////////////////////////////////////
-template <typename FImpl>
-std::vector<std::string> TDiscLoop<FImpl>::getInput(void)
-{
-    std::vector<std::string> in = {par().q_loop};
-    
-    return in;
-}
-
-template <typename FImpl>
-std::vector<std::string> TDiscLoop<FImpl>::getOutput(void)
-{
-    std::vector<std::string> out = {getName()};
-    
-    return out;
-}
-
-// setup ///////////////////////////////////////////////////////////////////////
-template <typename FImpl>
-void TDiscLoop<FImpl>::setup(void)
-{
-    
-}
-
-// execution ///////////////////////////////////////////////////////////////////
-template <typename FImpl>
-void TDiscLoop<FImpl>::execute(void)
-{
-    LOG(Message) << "Computing disconnected loop contraction '" << getName() 
-                 << "' using '" << par().q_loop << "' with " << par().gamma 
-                 << " insertion." << std::endl;
-
-    CorrWriter            writer(par().output);
-    PropagatorField       &q_loop = *env().template getObject<PropagatorField>(par().q_loop);
-    LatticeComplex        c(env().getGrid());
-    Gamma                 gamma(par().gamma);
-    std::vector<TComplex> buf;
-    Result                result;
-
-    c = trace(gamma*q_loop);
-    sliceSum(c, buf, Tp);
-
-    result.gamma = par().gamma;
-    result.corr.resize(buf.size());
-    for (unsigned int t = 0; t < buf.size(); ++t)
-    {
-        result.corr[t] = TensorRemove(buf[t]);
-    }
-
-    write(writer, "disc", result);
-}
-
-END_MODULE_NAMESPACE
-
-END_HADRONS_NAMESPACE
-
-#endif // Hadrons_DiscLoop_hpp_

extras/Hadrons/Modules/MContraction/Gamma3pt.hpp

@@ -1,170 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid 
-
-Source file: extras/Hadrons/Modules/MContraction/Gamma3pt.hpp
-
-Copyright (C) 2017
-
-Author: Andrew Lawson    <andrew.lawson1991@gmail.com>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-
-#ifndef Hadrons_Gamma3pt_hpp_
-#define Hadrons_Gamma3pt_hpp_
-
-#include <Grid/Hadrons/Global.hpp>
-#include <Grid/Hadrons/Module.hpp>
-#include <Grid/Hadrons/ModuleFactory.hpp>
-
-BEGIN_HADRONS_NAMESPACE
-
-/*
- * 3pt contraction with gamma matrix insertion.
- *
- * Schematic:
- *
- *             q2           q3
- *        /----<------*------<----¬
- *       /          gamma          \
- *      /                           \
- *   i *                            * f
- *      \                          /
- *       \                        /
- *        \----------->----------/
- *                   q1
- *
- *      trace(g5*q1*adj(q2)*g5*gamma*q3)
- */
-
-/******************************************************************************
- *                               Gamma3pt                                     *
- ******************************************************************************/
-BEGIN_MODULE_NAMESPACE(MContraction)
-
-class Gamma3ptPar: Serializable
-{
-public:
-    GRID_SERIALIZABLE_CLASS_MEMBERS(Gamma3ptPar,
-                                    std::string,    q1,
-                                    std::string,    q2,
-                                    std::string,    q3,
-                                    Gamma::Algebra, gamma,
-                                    std::string,    output);
-};
-
-template <typename FImpl1, typename FImpl2, typename FImpl3>
-class TGamma3pt: public Module<Gamma3ptPar>
-{
-    TYPE_ALIASES(FImpl1, 1);
-    TYPE_ALIASES(FImpl2, 2);
-    TYPE_ALIASES(FImpl3, 3);
-    class Result: Serializable
-    {
-    public:
-        GRID_SERIALIZABLE_CLASS_MEMBERS(Result,
-                                        Gamma::Algebra, gamma,
-                                        std::vector<Complex>, corr);
-    };
-public:
-    // constructor
-    TGamma3pt(const std::string name);
-    // destructor
-    virtual ~TGamma3pt(void) = default;
-    // dependency relation
-    virtual std::vector<std::string> getInput(void);
-    virtual std::vector<std::string> getOutput(void);
-    // setup
-    virtual void setup(void);
-    // execution
-    virtual void execute(void);
-};
-
-MODULE_REGISTER_NS(Gamma3pt, ARG(TGamma3pt<FIMPL, FIMPL, FIMPL>), MContraction);
-
-/******************************************************************************
- *                       TGamma3pt implementation                             *
- ******************************************************************************/
-// constructor /////////////////////////////////////////////////////////////////
-template <typename FImpl1, typename FImpl2, typename FImpl3>
-TGamma3pt<FImpl1, FImpl2, FImpl3>::TGamma3pt(const std::string name)
-: Module<Gamma3ptPar>(name)
-{}
-
-// dependencies/products ///////////////////////////////////////////////////////
-template <typename FImpl1, typename FImpl2, typename FImpl3>
-std::vector<std::string> TGamma3pt<FImpl1, FImpl2, FImpl3>::getInput(void)
-{
-    std::vector<std::string> in = {par().q1, par().q2, par().q3};
-    
-    return in;
-}
-
-template <typename FImpl1, typename FImpl2, typename FImpl3>
-std::vector<std::string> TGamma3pt<FImpl1, FImpl2, FImpl3>::getOutput(void)
-{
-    std::vector<std::string> out = {getName()};
-    
-    return out;
-}
-
-// setup ///////////////////////////////////////////////////////////////////////
-template <typename FImpl1, typename FImpl2, typename FImpl3>
-void TGamma3pt<FImpl1, FImpl2, FImpl3>::setup(void)
-{
-    
-}
-
-// execution ///////////////////////////////////////////////////////////////////
-template <typename FImpl1, typename FImpl2, typename FImpl3>
-void TGamma3pt<FImpl1, FImpl2, FImpl3>::execute(void)
-{
-    LOG(Message) << "Computing 3pt contractions '" << getName() << "' using"
-                 << " quarks '" << par().q1 << "', '" << par().q2 << "' and '"
-                 << par().q3 << "', with " << par().gamma << " insertion." 
-                 << std::endl;
-
-    CorrWriter            writer(par().output);
-    PropagatorField1      &q1 = *env().template getObject<PropagatorField1>(par().q1);
-    PropagatorField2      &q2 = *env().template getObject<PropagatorField2>(par().q2);
-    PropagatorField3      &q3 = *env().template getObject<PropagatorField3>(par().q3);
-    LatticeComplex        c(env().getGrid());
-    Gamma                 g5(Gamma::Algebra::Gamma5);
-    Gamma                 gamma(par().gamma);
-    std::vector<TComplex> buf;
-    Result                result;
-
-    c = trace(g5*q1*adj(q2)*(g5*gamma)*q3);
-    sliceSum(c, buf, Tp);
-
-    result.gamma = par().gamma;
-    result.corr.resize(buf.size());
-    for (unsigned int t = 0; t < buf.size(); ++t)
-    {
-        result.corr[t] = TensorRemove(buf[t]);
-    }
-
-    write(writer, "gamma3pt", result);
-}
-
-END_MODULE_NAMESPACE
-
-END_HADRONS_NAMESPACE
-
-#endif // Hadrons_Gamma3pt_hpp_

extras/Hadrons/Modules/MContraction/Meson.hpp

@@ -6,10 +6,8 @@ Source file: extras/Hadrons/Modules/MContraction/Meson.hpp
 
 Copyright (C) 2015
 Copyright (C) 2016
-Copyright (C) 2017
 
 Author: Antonin Portelli <antonin.portelli@me.com>
-        Andrew Lawson    <andrew.lawson1991@gmail.com>
 
 This program is free software; you can redistribute it and/or modify
 it under the terms of the GNU General Public License as published by
@@ -38,39 +36,20 @@ See the full license in the file "LICENSE" in the top level distribution directo
 
 BEGIN_HADRONS_NAMESPACE
 
-/*
- 
- Meson contractions
- -----------------------------
- 
- * options:
- - q1: input propagator 1 (string)
- - q2: input propagator 2 (string)
- - gammas: gamma products to insert at sink & source, pairs of gamma matrices 
-           (space-separated strings) in angled brackets (i.e. <g_sink g_src>),
-           in a sequence (e.g. "<Gamma5 Gamma5><Gamma5 GammaT>").
-
-           Special values: "all" - perform all possible contractions.
- - mom: momentum insertion, space-separated float sequence (e.g ".1 .2 1. 0."),
-        given as multiples of (2*pi) / L.
-*/
-
 /******************************************************************************
  *                                TMeson                                       *
  ******************************************************************************/
 BEGIN_MODULE_NAMESPACE(MContraction)
 
-typedef std::pair<Gamma::Algebra, Gamma::Algebra> GammaPair;
-
 class MesonPar: Serializable
 {
 public:
     GRID_SERIALIZABLE_CLASS_MEMBERS(MesonPar,
-                                    std::string, q1,
+                                    std::string,    q1,
-                                    std::string, q2,
+                                    std::string,    q2,
-                                    std::string, gammas,
+                                    std::string,    output,
-                                    std::string, mom,
+                                    Gamma::Algebra, gammaSource,
-                                    std::string, output);
+                                    Gamma::Algebra, gammaSink);
 };
 
 template <typename FImpl1, typename FImpl2>
@@ -82,10 +61,7 @@ public:
     class Result: Serializable
     {
     public:
-        GRID_SERIALIZABLE_CLASS_MEMBERS(Result,
+        GRID_SERIALIZABLE_CLASS_MEMBERS(Result, std::vector<Complex>, corr);
-                                        Gamma::Algebra, gamma_snk,
-                                        Gamma::Algebra, gamma_src,
-                                        std::vector<Complex>, corr);
     };
 public:
     // constructor
@@ -95,7 +71,6 @@ public:
     // dependencies/products
     virtual std::vector<std::string> getInput(void);
     virtual std::vector<std::string> getOutput(void);
-    virtual void parseGammaString(std::vector<GammaPair> &gammaList);
     // execution
     virtual void execute(void);
 };
@@ -128,32 +103,6 @@ std::vector<std::string> TMeson<FImpl1, FImpl2>::getOutput(void)
     return output;
 }
 
-template <typename FImpl1, typename FImpl2>
-void TMeson<FImpl1, FImpl2>::parseGammaString(std::vector<GammaPair> &gammaList)
-{
-    // Determine gamma matrices to insert at source/sink.
-    if (par().gammas.compare("all") == 0)
-    {
-        // Do all contractions.
-        unsigned int n_gam = Ns * Ns;
-        gammaList.resize(n_gam*n_gam);
-        for (unsigned int i = 1; i < Gamma::nGamma; i += 2)
-        {
-            for (unsigned int j = 1; j < Gamma::nGamma; j += 2)
-            {
-                gammaList.push_back(std::make_pair((Gamma::Algebra)i, 
-                                                   (Gamma::Algebra)j));
-            }
-        }
-    }
-    else
-    {
-        // Parse individual contractions from input string.
-        gammaList = strToVec<GammaPair>(par().gammas);
-    }
-}
-
-
 // execution ///////////////////////////////////////////////////////////////////
 template <typename FImpl1, typename FImpl2>
 void TMeson<FImpl1, FImpl2>::execute(void)
@@ -162,44 +111,21 @@ void TMeson<FImpl1, FImpl2>::execute(void)
                  << " quarks '" << par().q1 << "' and '" << par().q2 << "'"
                  << std::endl;
     
-    CorrWriter              writer(par().output);
+    XmlWriter             writer(par().output);
-    PropagatorField1       &q1 = *env().template getObject<PropagatorField1>(par().q1);
+    PropagatorField1      &q1 = *env().template getObject<PropagatorField1>(par().q1);
-    PropagatorField2       &q2 = *env().template getObject<PropagatorField2>(par().q2);
+    PropagatorField2      &q2 = *env().template getObject<PropagatorField2>(par().q2);
-    LatticeComplex         c(env().getGrid());
+    LatticeComplex        c(env().getGrid());
-    Gamma                  g5(Gamma::Algebra::Gamma5);
+    Gamma                 gSrc(par().gammaSource), gSnk(par().gammaSink);
-    std::vector<GammaPair> gammaList;
+    Gamma                 g5(Gamma::Algebra::Gamma5);
-    std::vector<TComplex>  buf;
+    std::vector<TComplex> buf;
-    std::vector<Result>    result;
+    Result                result;
-    std::vector<Real>      p;
-
-    p  = strToVec<Real>(par().mom);
-    LatticeComplex         ph(env().getGrid()), coor(env().getGrid());
-    Complex                i(0.0,1.0);
-    ph = zero;
-    for(unsigned int mu = 0; mu < env().getNd(); mu++)
-    {
-        LatticeCoordinate(coor, mu);
-        ph = ph + p[mu]*coor*((1./(env().getGrid()->_fdimensions[mu])));
-    }
-    ph = exp((Real)(2*M_PI)*i*ph);
     
-    parseGammaString(gammaList);
+    c = trace(gSnk*q1*adj(gSrc)*g5*adj(q2)*g5);
-
+    sliceSum(c, buf, Tp);
-    result.resize(gammaList.size());
+    result.corr.resize(buf.size());
-    for (unsigned int i = 0; i < result.size(); ++i)
+    for (unsigned int t = 0; t < buf.size(); ++t)
     {
-        Gamma gSnk(gammaList[i].first);
+        result.corr[t] = TensorRemove(buf[t]);
-        Gamma gSrc(gammaList[i].second);
-        c = trace((g5*gSnk)*q1*(adj(gSrc)*g5)*adj(q2))*ph;
-        sliceSum(c, buf, Tp);
-
-        result[i].gamma_snk = gammaList[i].first;
-        result[i].gamma_src = gammaList[i].second;
-        result[i].corr.resize(buf.size());
-        for (unsigned int t = 0; t < buf.size(); ++t)
-        {
-            result[i].corr[t] = TensorRemove(buf[t]);
-        }
     }
     write(writer, "meson", result);
 }

extras/Hadrons/Modules/MContraction/WeakHamiltonian.hpp

@@ -1,114 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid 
-
-Source file: extras/Hadrons/Modules/MContraction/WeakHamiltonian.hpp
-
-Copyright (C) 2017
-
-Author: Andrew Lawson    <andrew.lawson1991@gmail.com>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-
-#ifndef Hadrons_WeakHamiltonian_hpp_
-#define Hadrons_WeakHamiltonian_hpp_
-
-#include <Grid/Hadrons/Global.hpp>
-#include <Grid/Hadrons/Module.hpp>
-#include <Grid/Hadrons/ModuleFactory.hpp>
-
-BEGIN_HADRONS_NAMESPACE
-
-/******************************************************************************
- *                         WeakHamiltonian                                    *
- ******************************************************************************/
-BEGIN_MODULE_NAMESPACE(MContraction)
-
-/*******************************************************************************
- * Utilities for contractions involving the Weak Hamiltonian.
- ******************************************************************************/
-//// Sum and store correlator.
-#define MAKE_DIAG(exp, buf, res, n)\
-sliceSum(exp, buf, Tp);\
-res.name = (n);\
-res.corr.resize(buf.size());\
-for (unsigned int t = 0; t < buf.size(); ++t)\
-{\
-    res.corr[t] = TensorRemove(buf[t]);\
-}
-
-//// Contraction of mu index: use 'mu' variable in exp.
-#define SUM_MU(buf,exp)\
-buf = zero;\
-for (unsigned int mu = 0; mu < ndim; ++mu)\
-{\
-    buf += exp;\
-}
-
-enum 
-{
-  i_V = 0,
-  i_A = 1,
-  n_i = 2
-};
-
-class WeakHamiltonianPar: Serializable
-{
-public:
-    GRID_SERIALIZABLE_CLASS_MEMBERS(WeakHamiltonianPar,
-                                    std::string, q1,
-                                    std::string, q2,
-                                    std::string, q3,
-                                    std::string, q4,
-                                    std::string, output);
-};
-
-#define MAKE_WEAK_MODULE(modname)\
-class T##modname: public Module<WeakHamiltonianPar>\
-{\
-public:\
-    TYPE_ALIASES(FIMPL,)\
-    class Result: Serializable\
-    {\
-    public:\
-        GRID_SERIALIZABLE_CLASS_MEMBERS(Result,\
-                                        std::string, name,\
-                                        std::vector<Complex>, corr);\
-    };\
-public:\
-    /* constructor */ \
-    T##modname(const std::string name);\
-    /* destructor */ \
-    virtual ~T##modname(void) = default;\
-    /* dependency relation */ \
-    virtual std::vector<std::string> getInput(void);\
-    virtual std::vector<std::string> getOutput(void);\
-    /* setup */ \
-    virtual void setup(void);\
-    /* execution */ \
-    virtual void execute(void);\
-    std::vector<std::string> VA_label = {"V", "A"};\
-};\
-MODULE_REGISTER_NS(modname, T##modname, MContraction);
-
-END_MODULE_NAMESPACE
-
-END_HADRONS_NAMESPACE
-
-#endif // Hadrons_WeakHamiltonian_hpp_

extras/Hadrons/Modules/MContraction/WeakHamiltonianEye.cc

@@ -1,137 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid 
-
-Source file: extras/Hadrons/Modules/MContraction/WeakHamiltonianEye.cc
-
-Copyright (C) 2017
-
-Author: Andrew Lawson    <andrew.lawson1991@gmail.com>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-
-#include <Grid/Hadrons/Modules/MContraction/WeakHamiltonianEye.hpp>
-
-using namespace Grid;
-using namespace Hadrons;
-using namespace MContraction;
-
-/*
- * Weak Hamiltonian current-current contractions, Eye-type.
- * 
- * These contractions are generated by the Q1 and Q2 operators in the physical
- * basis (see e.g. Fig 3 of arXiv:1507.03094).
- * 
- * Schematics:        q4                 |                  
- *                  /-<-¬                |                             
- *                 /     \               |             q2           q3
- *                 \     /               |        /----<------*------<----¬                        
- *            q2    \   /    q3          |       /          /-*-¬          \
- *       /-----<-----* *-----<----¬      |      /          /     \          \
- *    i *            H_W           * f   |   i *           \     /  q4      * f
- *       \                        /      |      \           \->-/          /   
- *        \                      /       |       \                        /       
- *         \---------->---------/        |        \----------->----------/        
- *                   q1                  |                   q1                  
- *                                       |
- *                Saucer (S)             |                  Eye (E)
- * 
- * S: trace(q3*g5*q1*adj(q2)*g5*gL[mu][p_1]*q4*gL[mu][p_2])
- * E: trace(q3*g5*q1*adj(q2)*g5*gL[mu][p_1])*trace(q4*gL[mu][p_2])
- */
-
-/******************************************************************************
- *                  TWeakHamiltonianEye implementation                        *
- ******************************************************************************/
-// constructor /////////////////////////////////////////////////////////////////
-TWeakHamiltonianEye::TWeakHamiltonianEye(const std::string name)
-: Module<WeakHamiltonianPar>(name)
-{}
-
-// dependencies/products ///////////////////////////////////////////////////////
-std::vector<std::string> TWeakHamiltonianEye::getInput(void)
-{
-    std::vector<std::string> in = {par().q1, par().q2, par().q3, par().q4};
-    
-    return in;
-}
-
-std::vector<std::string> TWeakHamiltonianEye::getOutput(void)
-{
-    std::vector<std::string> out = {getName()};
-    
-    return out;
-}
-
-// setup ///////////////////////////////////////////////////////////////////////
-void TWeakHamiltonianEye::setup(void)
-{
-
-}
-
-// execution ///////////////////////////////////////////////////////////////////
-void TWeakHamiltonianEye::execute(void)
-{
-    LOG(Message) << "Computing Weak Hamiltonian (Eye type) contractions '" 
-                 << getName() << "' using quarks '" << par().q1 << "', '" 
-                 << par().q2 << ", '" << par().q3 << "' and '" << par().q4 
-                 << "'." << std::endl;
-
-    CorrWriter             writer(par().output);
-    PropagatorField &q1 = *env().template getObject<PropagatorField>(par().q1);
-    PropagatorField &q2 = *env().template getObject<PropagatorField>(par().q2);
-    PropagatorField &q3 = *env().template getObject<PropagatorField>(par().q3);
-    PropagatorField &q4 = *env().template getObject<PropagatorField>(par().q4);
-    Gamma g5            = Gamma(Gamma::Algebra::Gamma5);
-    LatticeComplex        expbuf(env().getGrid());
-    std::vector<TComplex> corrbuf;
-    std::vector<Result>   result(n_eye_diag);
-    unsigned int ndim   = env().getNd();
-
-    PropagatorField              tmp1(env().getGrid());
-    LatticeComplex               tmp2(env().getGrid());
-    std::vector<PropagatorField> S_body(ndim, tmp1);
-    std::vector<PropagatorField> S_loop(ndim, tmp1);
-    std::vector<LatticeComplex>  E_body(ndim, tmp2);
-    std::vector<LatticeComplex>  E_loop(ndim, tmp2);
-
-    // Setup for S-type contractions.
-    for (int mu = 0; mu < ndim; ++mu)
-    {
-        S_body[mu] = MAKE_SE_BODY(q1, q2, q3, GammaL(Gamma::gmu[mu]));
-        S_loop[mu] = MAKE_SE_LOOP(q4, GammaL(Gamma::gmu[mu]));
-    }
-
-    // Perform S-type contractions.    
-    SUM_MU(expbuf, trace(S_body[mu]*S_loop[mu]))
-    MAKE_DIAG(expbuf, corrbuf, result[S_diag], "HW_S")
-
-    // Recycle sub-expressions for E-type contractions.
-    for (unsigned int mu = 0; mu < ndim; ++mu)
-    {
-        E_body[mu] = trace(S_body[mu]);
-        E_loop[mu] = trace(S_loop[mu]);
-    }
-
-    // Perform E-type contractions.
-    SUM_MU(expbuf, E_body[mu]*E_loop[mu])
-    MAKE_DIAG(expbuf, corrbuf, result[E_diag], "HW_E")
-
-    write(writer, "HW_Eye", result);
-}

extras/Hadrons/Modules/MContraction/WeakHamiltonianEye.hpp

@@ -1,58 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid 
-
-Source file: extras/Hadrons/Modules/MContraction/WeakHamiltonianEye.hpp
-
-Copyright (C) 2017
-
-Author: Andrew Lawson    <andrew.lawson1991@gmail.com>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-
-#ifndef Hadrons_WeakHamiltonianEye_hpp_
-#define Hadrons_WeakHamiltonianEye_hpp_
-
-#include <Grid/Hadrons/Modules/MContraction/WeakHamiltonian.hpp>
-
-BEGIN_HADRONS_NAMESPACE
-
-/******************************************************************************
- *                         WeakHamiltonianEye                                 *
- ******************************************************************************/
-BEGIN_MODULE_NAMESPACE(MContraction)
-
-enum
-{
-    S_diag = 0,
-    E_diag = 1,
-    n_eye_diag = 2
-};
-
-// Saucer and Eye subdiagram contractions.
-#define MAKE_SE_BODY(Q_1, Q_2, Q_3, gamma) (Q_3*g5*Q_1*adj(Q_2)*g5*gamma)
-#define MAKE_SE_LOOP(Q_loop, gamma) (Q_loop*gamma)
-
-MAKE_WEAK_MODULE(WeakHamiltonianEye)
-
-END_MODULE_NAMESPACE
-
-END_HADRONS_NAMESPACE
-
-#endif // Hadrons_WeakHamiltonianEye_hpp_

extras/Hadrons/Modules/MContraction/WeakHamiltonianNonEye.cc

@@ -1,139 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid 
-
-Source file: extras/Hadrons/Modules/MContraction/WeakHamiltonianNonEye.cc
-
-Copyright (C) 2017
-
-Author: Andrew Lawson    <andrew.lawson1991@gmail.com>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-
-#include <Grid/Hadrons/Modules/MContraction/WeakHamiltonianNonEye.hpp>
-
-using namespace Grid;
-using namespace Hadrons;
-using namespace MContraction;
-
-/*
- * Weak Hamiltonian current-current contractions, Non-Eye-type.
- * 
- * These contractions are generated by the Q1 and Q2 operators in the physical
- * basis (see e.g. Fig 3 of arXiv:1507.03094).
- * 
- * Schematic:     
- *            q2             q3          |           q2              q3
- *          /--<--¬       /--<--¬        |        /--<--¬         /--<--¬       
- *         /       \     /       \       |       /       \       /       \      
- *        /         \   /         \      |      /         \     /         \     
- *       /           \ /           \     |     /           \   /           \    
- *    i *             * H_W         *  f |  i *             * * H_W         * f 
- *      \             *             |    |     \           /   \           /
- *       \           / \           /     |      \         /     \         /    
- *        \         /   \         /      |       \       /       \       /  
- *         \       /     \       /       |        \-->--/         \-->--/      
- *          \-->--/       \-->--/        |          q1               q4 
- *            q1             q4          |
- *                Connected (C)          |                 Wing (W)
- *
- * C: trace(q1*adj(q2)*g5*gL[mu]*q3*adj(q4)*g5*gL[mu])
- * W: trace(q1*adj(q2)*g5*gL[mu])*trace(q3*adj(q4)*g5*gL[mu])
- * 
- */
-
-/******************************************************************************
- *                  TWeakHamiltonianNonEye implementation                     *
- ******************************************************************************/
-// constructor /////////////////////////////////////////////////////////////////
-TWeakHamiltonianNonEye::TWeakHamiltonianNonEye(const std::string name)
-: Module<WeakHamiltonianPar>(name)
-{}
-
-// dependencies/products ///////////////////////////////////////////////////////
-std::vector<std::string> TWeakHamiltonianNonEye::getInput(void)
-{
-    std::vector<std::string> in = {par().q1, par().q2, par().q3, par().q4};
-    
-    return in;
-}
-
-std::vector<std::string> TWeakHamiltonianNonEye::getOutput(void)
-{
-    std::vector<std::string> out = {getName()};
-    
-    return out;
-}
-
-// setup ///////////////////////////////////////////////////////////////////////
-void TWeakHamiltonianNonEye::setup(void)
-{
-
-}
-
-// execution ///////////////////////////////////////////////////////////////////
-void TWeakHamiltonianNonEye::execute(void)
-{
-    LOG(Message) << "Computing Weak Hamiltonian (Non-Eye type) contractions '" 
-                 << getName() << "' using quarks '" << par().q1 << "', '" 
-                 << par().q2 << ", '" << par().q3 << "' and '" << par().q4 
-                 << "'." << std::endl;
-    
-    CorrWriter             writer(par().output);
-    PropagatorField &q1 = *env().template getObject<PropagatorField>(par().q1);
-    PropagatorField &q2 = *env().template getObject<PropagatorField>(par().q2);
-    PropagatorField &q3 = *env().template getObject<PropagatorField>(par().q3);
-    PropagatorField &q4 = *env().template getObject<PropagatorField>(par().q4);
-    Gamma g5            = Gamma(Gamma::Algebra::Gamma5);
-    LatticeComplex        expbuf(env().getGrid());
-    std::vector<TComplex> corrbuf;
-    std::vector<Result>   result(n_noneye_diag); 
-    unsigned int ndim   = env().getNd();
-
-    PropagatorField              tmp1(env().getGrid());
-    LatticeComplex               tmp2(env().getGrid());
-    std::vector<PropagatorField> C_i_side_loop(ndim, tmp1);
-    std::vector<PropagatorField> C_f_side_loop(ndim, tmp1);
-    std::vector<LatticeComplex>  W_i_side_loop(ndim, tmp2);
-    std::vector<LatticeComplex>  W_f_side_loop(ndim, tmp2);
-
-    // Setup for C-type contractions.
-    for (int mu = 0; mu < ndim; ++mu)
-    {
-        C_i_side_loop[mu] = MAKE_CW_SUBDIAG(q1, q2, GammaL(Gamma::gmu[mu]));
-        C_f_side_loop[mu] = MAKE_CW_SUBDIAG(q3, q4, GammaL(Gamma::gmu[mu]));
-    }
-
-    // Perform C-type contractions.    
-    SUM_MU(expbuf, trace(C_i_side_loop[mu]*C_f_side_loop[mu]))
-    MAKE_DIAG(expbuf, corrbuf, result[C_diag], "HW_C")
-
-    // Recycle sub-expressions for W-type contractions.
-    for (unsigned int mu = 0; mu < ndim; ++mu)
-    {
-        W_i_side_loop[mu] = trace(C_i_side_loop[mu]);
-        W_f_side_loop[mu] = trace(C_f_side_loop[mu]);
-    }
-
-    // Perform W-type contractions.
-    SUM_MU(expbuf, W_i_side_loop[mu]*W_f_side_loop[mu])
-    MAKE_DIAG(expbuf, corrbuf, result[W_diag], "HW_W")
-
-    write(writer, "HW_NonEye", result);
-}

extras/Hadrons/Modules/MContraction/WeakHamiltonianNonEye.hpp

@@ -1,57 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid 
-
-Source file: extras/Hadrons/Modules/MContraction/WeakHamiltonianNonEye.hpp
-
-Copyright (C) 2017
-
-Author: Andrew Lawson    <andrew.lawson1991@gmail.com>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-
-#ifndef Hadrons_WeakHamiltonianNonEye_hpp_
-#define Hadrons_WeakHamiltonianNonEye_hpp_
-
-#include <Grid/Hadrons/Modules/MContraction/WeakHamiltonian.hpp>
-
-BEGIN_HADRONS_NAMESPACE
-
-/******************************************************************************
- *                         WeakHamiltonianNonEye                              *
- ******************************************************************************/
-BEGIN_MODULE_NAMESPACE(MContraction)
-
-enum
-{
-    W_diag = 0,
-    C_diag = 1,
-    n_noneye_diag = 2
-};
-
-// Wing and Connected subdiagram contractions
-#define MAKE_CW_SUBDIAG(Q_1, Q_2, gamma) (Q_1*adj(Q_2)*g5*gamma)
-
-MAKE_WEAK_MODULE(WeakHamiltonianNonEye)
-
-END_MODULE_NAMESPACE
-
-END_HADRONS_NAMESPACE
-
-#endif // Hadrons_WeakHamiltonianNonEye_hpp_

extras/Hadrons/Modules/MContraction/WeakNeutral4ptDisc.cc

@@ -1,135 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid 
-
-Source file: extras/Hadrons/Modules/MContraction/WeakNeutral4ptDisc.cc
-
-Copyright (C) 2017
-
-Author: Andrew Lawson    <andrew.lawson1991@gmail.com>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-
-#include <Grid/Hadrons/Modules/MContraction/WeakNeutral4ptDisc.hpp>
-
-using namespace Grid;
-using namespace Hadrons;
-using namespace MContraction;
-
-/*
- * Weak Hamiltonian + current contractions, disconnected topology for neutral 
- * mesons.
- * 
- * These contractions are generated by operators Q_1,...,10 of the dS=1 Weak
- * Hamiltonian in the physical basis and an additional current J (see e.g. 
- * Fig 11 of arXiv:1507.03094).
- * 
- * Schematic:
- *                        
- *           q2          q4             q3
- *       /--<--¬     /---<--¬       /---<--¬
- *     /         \ /         \     /        \
- *  i *           * H_W      |  J *          * f
- *     \         / \         /     \        /
- *      \--->---/   \-------/       \------/
- *          q1 
- * 
- * options
- * - q1: input propagator 1 (string)
- * - q2: input propagator 2 (string)
- * - q3: input propagator 3 (string), assumed to be sequential propagator 
- * - q4: input propagator 4 (string), assumed to be a loop
- * 
- * type 1: trace(q1*adj(q2)*g5*gL[mu])*trace(loop*gL[mu])*trace(q3*g5)
- * type 2: trace(q1*adj(q2)*g5*gL[mu]*loop*gL[mu])*trace(q3*g5)
- */
-
-/*******************************************************************************
- *                  TWeakNeutral4ptDisc implementation                         *
- ******************************************************************************/
-// constructor /////////////////////////////////////////////////////////////////
-TWeakNeutral4ptDisc::TWeakNeutral4ptDisc(const std::string name)
-: Module<WeakHamiltonianPar>(name)
-{}
-
-// dependencies/products ///////////////////////////////////////////////////////
-std::vector<std::string> TWeakNeutral4ptDisc::getInput(void)
-{
-    std::vector<std::string> in = {par().q1, par().q2, par().q3, par().q4};
-    
-    return in;
-}
-
-std::vector<std::string> TWeakNeutral4ptDisc::getOutput(void)
-{
-    std::vector<std::string> out = {getName()};
-    
-    return out;
-}
-
-// setup ///////////////////////////////////////////////////////////////////////
-void TWeakNeutral4ptDisc::setup(void)
-{
-
-}
-
-// execution ///////////////////////////////////////////////////////////////////
-void TWeakNeutral4ptDisc::execute(void)
-{
-    LOG(Message) << "Computing Weak Hamiltonian neutral disconnected contractions '" 
-                 << getName() << "' using quarks '" << par().q1 << "', '" 
-                 << par().q2 << ", '" << par().q3 << "' and '" << par().q4 
-                 << "'." << std::endl;
-
-    CorrWriter             writer(par().output);
-    PropagatorField &q1 = *env().template getObject<PropagatorField>(par().q1);
-    PropagatorField &q2 = *env().template getObject<PropagatorField>(par().q2);
-    PropagatorField &q3 = *env().template getObject<PropagatorField>(par().q3);
-    PropagatorField &q4 = *env().template getObject<PropagatorField>(par().q4);
-    Gamma g5            = Gamma(Gamma::Algebra::Gamma5);
-    LatticeComplex        expbuf(env().getGrid());
-    std::vector<TComplex> corrbuf;
-    std::vector<Result>   result(n_neut_disc_diag);
-    unsigned int ndim   = env().getNd();
-
-    PropagatorField              tmp(env().getGrid());
-    std::vector<PropagatorField> meson(ndim, tmp);
-    std::vector<PropagatorField> loop(ndim, tmp);
-    LatticeComplex               curr(env().getGrid());
-
-    // Setup for type 1 contractions.
-    for (int mu = 0; mu < ndim; ++mu)
-    {
-        meson[mu] = MAKE_DISC_MESON(q1, q2, GammaL(Gamma::gmu[mu]));
-        loop[mu] = MAKE_DISC_LOOP(q4, GammaL(Gamma::gmu[mu]));
-    }
-    curr = MAKE_DISC_CURR(q3, GammaL(Gamma::Algebra::Gamma5));
-
-    // Perform type 1 contractions.    
-    SUM_MU(expbuf, trace(meson[mu]*loop[mu]))
-    expbuf *= curr;
-    MAKE_DIAG(expbuf, corrbuf, result[neut_disc_1_diag], "HW_disc0_1")
-
-    // Perform type 2 contractions.
-    SUM_MU(expbuf, trace(meson[mu])*trace(loop[mu]))
-    expbuf *= curr;
-    MAKE_DIAG(expbuf, corrbuf, result[neut_disc_2_diag], "HW_disc0_2")
-
-    write(writer, "HW_disc0", result);
-}

extras/Hadrons/Modules/MContraction/WeakNeutral4ptDisc.hpp

@@ -1,59 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid 
-
-Source file: extras/Hadrons/Modules/MContraction/WeakNeutral4ptDisc.hpp
-
-Copyright (C) 2017
-
-Author: Andrew Lawson    <andrew.lawson1991@gmail.com>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-
-#ifndef Hadrons_WeakNeutral4ptDisc_hpp_
-#define Hadrons_WeakNeutral4ptDisc_hpp_
-
-#include <Grid/Hadrons/Modules/MContraction/WeakHamiltonian.hpp>
-
-BEGIN_HADRONS_NAMESPACE
-
-/******************************************************************************
- *                         WeakNeutral4ptDisc                                 *
- ******************************************************************************/
-BEGIN_MODULE_NAMESPACE(MContraction)
-
-enum
-{
-    neut_disc_1_diag = 0,
-    neut_disc_2_diag = 1,
-    n_neut_disc_diag = 2
-};
-
-// Neutral 4pt disconnected subdiagram contractions.
-#define MAKE_DISC_MESON(Q_1, Q_2, gamma) (Q_1*adj(Q_2)*g5*gamma)
-#define MAKE_DISC_LOOP(Q_LOOP, gamma) (Q_LOOP*gamma)
-#define MAKE_DISC_CURR(Q_c, gamma) (trace(Q_c*gamma))
-
-MAKE_WEAK_MODULE(WeakNeutral4ptDisc)
-
-END_MODULE_NAMESPACE
-
-END_HADRONS_NAMESPACE
-
-#endif // Hadrons_WeakNeutral4ptDisc_hpp_

extras/Hadrons/Modules/MLoop/NoiseLoop.hpp

@@ -1,132 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid 
-
-Source file: extras/Hadrons/Modules/MLoop/NoiseLoop.hpp
-
-Copyright (C) 2016
-
-Author: Andrew Lawson <andrew.lawson1991@gmail.com>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-
-#ifndef Hadrons_NoiseLoop_hpp_
-#define Hadrons_NoiseLoop_hpp_
-
-#include <Grid/Hadrons/Global.hpp>
-#include <Grid/Hadrons/Module.hpp>
-#include <Grid/Hadrons/ModuleFactory.hpp>
-
-BEGIN_HADRONS_NAMESPACE
-
-/*
- 
- Noise loop propagator
- -----------------------------
- * loop_x = q_x * adj(eta_x)
- 
- * options:
- - q = Result of inversion on noise source.
- - eta = noise source.
-
- */
-
-
-/******************************************************************************
- *                         NoiseLoop                                          *
- ******************************************************************************/
-BEGIN_MODULE_NAMESPACE(MLoop)
-
-class NoiseLoopPar: Serializable
-{
-public:
-    GRID_SERIALIZABLE_CLASS_MEMBERS(NoiseLoopPar,
-                                    std::string, q,
-                                    std::string, eta);
-};
-
-template <typename FImpl>
-class TNoiseLoop: public Module<NoiseLoopPar>
-{
-public:
-    TYPE_ALIASES(FImpl,);
-public:
-    // constructor
-    TNoiseLoop(const std::string name);
-    // destructor
-    virtual ~TNoiseLoop(void) = default;
-    // dependency relation
-    virtual std::vector<std::string> getInput(void);
-    virtual std::vector<std::string> getOutput(void);
-    // setup
-    virtual void setup(void);
-    // execution
-    virtual void execute(void);
-};
-
-MODULE_REGISTER_NS(NoiseLoop, TNoiseLoop<FIMPL>, MLoop);
-
-/******************************************************************************
- *                 TNoiseLoop implementation                                  *
- ******************************************************************************/
-// constructor /////////////////////////////////////////////////////////////////
-template <typename FImpl>
-TNoiseLoop<FImpl>::TNoiseLoop(const std::string name)
-: Module<NoiseLoopPar>(name)
-{}
-
-// dependencies/products ///////////////////////////////////////////////////////
-template <typename FImpl>
-std::vector<std::string> TNoiseLoop<FImpl>::getInput(void)
-{
-    std::vector<std::string> in = {par().q, par().eta};
-    
-    return in;
-}
-
-template <typename FImpl>
-std::vector<std::string> TNoiseLoop<FImpl>::getOutput(void)
-{
-    std::vector<std::string> out = {getName()};
-    
-    return out;
-}
-
-// setup ///////////////////////////////////////////////////////////////////////
-template <typename FImpl>
-void TNoiseLoop<FImpl>::setup(void)
-{
-    env().template registerLattice<PropagatorField>(getName());
-}
-
-// execution ///////////////////////////////////////////////////////////////////
-template <typename FImpl>
-void TNoiseLoop<FImpl>::execute(void)
-{
-    PropagatorField &loop = *env().template createLattice<PropagatorField>(getName());
-    PropagatorField &q    = *env().template getObject<PropagatorField>(par().q);
-    PropagatorField &eta  = *env().template getObject<PropagatorField>(par().eta);
-    loop = q*adj(eta);
-}
-
-END_MODULE_NAMESPACE
-
-END_HADRONS_NAMESPACE
-
-#endif // Hadrons_NoiseLoop_hpp_

extras/Hadrons/Modules/MSource/SeqGamma.hpp

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

extras/Hadrons/Modules/MSource/Wall.hpp

@@ -1,147 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid 
-
-Source file: extras/Hadrons/Modules/MSource/Wall.hpp
-
-Copyright (C) 2017
-
-Author: Andrew Lawson <andrew.lawson1991@gmail.com>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution directory
-*************************************************************************************/
-/*  END LEGAL */
-
-#ifndef Hadrons_WallSource_hpp_
-#define Hadrons_WallSource_hpp_
-
-#include <Grid/Hadrons/Global.hpp>
-#include <Grid/Hadrons/Module.hpp>
-#include <Grid/Hadrons/ModuleFactory.hpp>
-
-BEGIN_HADRONS_NAMESPACE
-
-/*
- 
- Wall source
- -----------------------------
- * src_x = delta(x_3 - tW) * exp(i x.mom)
- 
- * options:
- - tW: source timeslice (integer)
- - mom: momentum insertion, space-separated float sequence (e.g ".1 .2 1. 0.")
- 
- */
-
-/******************************************************************************
- *                         Wall                                               *
- ******************************************************************************/
-BEGIN_MODULE_NAMESPACE(MSource)
-
-class WallPar: Serializable
-{
-public:
-    GRID_SERIALIZABLE_CLASS_MEMBERS(WallPar,
-                                    unsigned int, tW,
-                                    std::string, mom);
-};
-
-template <typename FImpl>
-class TWall: public Module<WallPar>
-{
-public:
-    TYPE_ALIASES(FImpl,);
-public:
-    // constructor
-    TWall(const std::string name);
-    // destructor
-    virtual ~TWall(void) = default;
-    // dependency relation
-    virtual std::vector<std::string> getInput(void);
-    virtual std::vector<std::string> getOutput(void);
-    // setup
-    virtual void setup(void);
-    // execution
-    virtual void execute(void);
-};
-
-MODULE_REGISTER_NS(Wall, TWall<FIMPL>, MSource);
-
-/******************************************************************************
- *                 TWall implementation                                       *
- ******************************************************************************/
-// constructor /////////////////////////////////////////////////////////////////
-template <typename FImpl>
-TWall<FImpl>::TWall(const std::string name)
-: Module<WallPar>(name)
-{}
-
-// dependencies/products ///////////////////////////////////////////////////////
-template <typename FImpl>
-std::vector<std::string> TWall<FImpl>::getInput(void)
-{
-    std::vector<std::string> in;
-    
-    return in;
-}
-
-template <typename FImpl>
-std::vector<std::string> TWall<FImpl>::getOutput(void)
-{
-    std::vector<std::string> out = {getName()};
-    
-    return out;
-}
-
-// setup ///////////////////////////////////////////////////////////////////////
-template <typename FImpl>
-void TWall<FImpl>::setup(void)
-{
-    env().template registerLattice<PropagatorField>(getName());
-}
-
-// execution ///////////////////////////////////////////////////////////////////
-template <typename FImpl>
-void TWall<FImpl>::execute(void)
-{    
-    LOG(Message) << "Generating wall source at t = " << par().tW 
-                 << " with momentum " << par().mom << std::endl;
-    
-    PropagatorField &src = *env().template createLattice<PropagatorField>(getName());
-    Lattice<iScalar<vInteger>> t(env().getGrid());
-    LatticeComplex             ph(env().getGrid()), coor(env().getGrid());
-    std::vector<Real>          p;
-    Complex                    i(0.0,1.0);
-    
-    p  = strToVec<Real>(par().mom);
-    ph = zero;
-    for(unsigned int mu = 0; mu < Nd; mu++)
-    {
-        LatticeCoordinate(coor, mu);
-        ph = ph + p[mu]*coor*((1./(env().getGrid()->_fdimensions[mu])));
-    }
-    ph = exp((Real)(2*M_PI)*i*ph);
-    LatticeCoordinate(t, Tp);
-    src = 1.;
-    src = where((t == par().tW), src*ph, 0.*src);
-}
-
-END_MODULE_NAMESPACE
-
-END_HADRONS_NAMESPACE
-
-#endif // Hadrons_WallSource_hpp_

extras/Hadrons/Modules/Quark.hpp

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

extras/Hadrons/modules.inc

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

gcc-bug-report/README

@@ -20,17 +20,4 @@ The simple testcase in this directory is the submitted bug report that encapsula
 problem. The test case works with icpc and with clang++, but fails consistently on g++
 current variants.
 
-Peter
+Peter
-
-
-************
-
-Second GCC bug reported, see Issue 100.
-
-https://wandbox.org/permlink/tzssJza6R9XnqANw
-https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80652
-
-Getting Travis fails under gcc-5 for Test_simd, now that I added more comprehensive testing to the
-CI test suite. The limitations of Travis runtime limits & weak cores are being shown.
-
-Travis uses 5.4.1 for g++-5.

grid-config.in

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

lib/Algorithms.h

@@ -39,17 +39,19 @@ 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>
+
 #include <Grid/algorithms/iterative/ConjugateGradientMultiShift.h>
 #include <Grid/algorithms/iterative/ConjugateGradientMixedPrec.h>
 
 // Lanczos support
-//#include <Grid/algorithms/iterative/MatrixUtils.h>
+#include <Grid/algorithms/iterative/MatrixUtils.h>
 #include <Grid/algorithms/iterative/ImplicitlyRestartedLanczos.h>
+
 #include <Grid/algorithms/CoarsenedMatrix.h>
-#include <Grid/algorithms/FFT.h>
 
 // Eigen/lanczos
 // EigCg

lib/AlignedAllocator.cc

@@ -1,7 +1,7 @@
 
 
 
-#include <Grid/GridCore.h>
+#include <Grid/Grid.h>
 
 namespace Grid {
 
@@ -13,10 +13,9 @@ void *PointerCache::Insert(void *ptr,size_t bytes) {
 
   if (bytes < 4096 ) return NULL;
 
-#ifdef GRID_OMP
+#ifdef _OPENMP
   assert(omp_in_parallel()==0);
 #endif 
-
   void * ret = NULL;
   int v = -1;
 

lib/DisableWarnings.h

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

lib/Grid.h

@@ -38,10 +38,52 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 #ifndef GRID_H
 #define GRID_H
 
-#include <Grid/GridCore.h>
+///////////////////
-#include <Grid/GridQCDcore.h>
+// Std C++ dependencies
-#include <Grid/qcd/action/Action.h>
+///////////////////
-#include <Grid/qcd/smearing/Smearing.h>
+#include <cassert>
-#include <Grid/qcd/hmc/HMC_aggregate.h>
+#include <complex>
+#include <vector>
+#include <iostream>
+#include <iomanip>
+#include <random>
+#include <functional>
+#include <stdio.h>
+#include <stdlib.h>
+#include <stdio.h>
+#include <signal.h>
+#include <ctime>
+#include <sys/time.h>
+#include <chrono>
+
+///////////////////
+// Grid headers
+///////////////////
+#include "Config.h"
+#include <Grid/Timer.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>
+#include <Grid/Communicator.h> 
+#include <Grid/Cartesian.h>    
+#include <Grid/Tensors.h>      
+#include <Grid/Lattice.h>      
+#include <Grid/Cshift.h>       
+#include <Grid/Stencil.h>      
+#include <Grid/Algorithms.h>   
+#include <Grid/parallelIO/BinaryIO.h>
+#include <Grid/FFT.h>
+
+#include <Grid/qcd/QCD.h>
+#include <Grid/parallelIO/NerscIO.h>
+#include <Grid/qcd/hmc/NerscCheckpointer.h>
+#include <Grid/qcd/hmc/HmcRunner.h>
+
+
 
 #endif

lib/GridStd.h

@@ -1,27 +0,0 @@
-#ifndef GRID_STD_H
-#define GRID_STD_H
-
-///////////////////
-// Std C++ dependencies
-///////////////////
-#include <cassert>
-#include <complex>
-#include <vector>
-#include <iostream>
-#include <iomanip>
-#include <random>
-#include <functional>
-#include <stdio.h>
-#include <stdlib.h>
-#include <stdio.h>
-#include <signal.h>
-#include <ctime>
-#include <sys/time.h>
-#include <chrono>
-
-///////////////////
-// Grid config
-///////////////////
-#include "Config.h"
-
-#endif /* GRID_STD_H */

lib/Grid_Eigen_Dense.h

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

lib/Init.cc

@@ -1,6 +1,6 @@
-/*************************************************************************************
+    /*************************************************************************************
 
-    Grid physics library, www.github.com/paboyle/Grid
+    Grid physics library, www.github.com/paboyle/Grid 
 
     Source file: ./lib/Init.cc
 
@@ -36,20 +36,17 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 #include <stdint.h>
 #include <unistd.h>
 #include <sys/mman.h>
-#include <sys/stat.h>
+#include <sys/stat.h> 
 #include <sys/time.h>
 #include <signal.h>
 #include <iostream>
 #include <iterator>
+#include <Grid/Grid.h>
 #include <algorithm>
 #include <iterator>
 #include <cstdlib>
 #include <memory>
 
-#include <Grid/Grid.h>
-
-#include <Grid/util/CompilerCompatible.h>
-
 
 #include <fenv.h>
 #ifdef __APPLE__
@@ -95,14 +92,14 @@ const std::vector<int> GridDefaultSimd(int dims,int nsimd)
       if ( nn>=2) {
 	layout[d]=2;
 	nn/=2;
-      } else {
+      } else { 
 	layout[d]=1;
       }
     }
     assert(nn==1);
     return layout;
 }
-
+  
 ////////////////////////////////////////////////////////////
 // Command line parsing assist for stock controls
 ////////////////////////////////////////////////////////////
@@ -146,7 +143,7 @@ void GridCmdOptionIntVector(std::string &str,std::vector<int> & vec)
     vec.push_back(i);
     if(std::ispunct(ss.peek()))
       ss.ignore();
-  }
+  }    
   return;
 }
 
@@ -222,59 +219,8 @@ void Grid_init(int *argc,char ***argv)
     CartesianCommunicator::MAX_MPI_SHM_BYTES = MB*1024*1024;
   }
 
-  if( GridCmdOptionExists(*argv,*argv+*argc,"--debug-signals") ){
-    Grid_debug_handler_init();
-  }
-
   CartesianCommunicator::Init(argc,argv);
 
-  if( !GridCmdOptionExists(*argv,*argv+*argc,"--debug-stdout") ){
-    Grid_quiesce_nodes();
-  } else { 
-    FILE *fp;
-    std::ostringstream fname;
-    fname<<"Grid.stdout.";
-    fname<<CartesianCommunicator::RankWorld();
-    fp=freopen(fname.str().c_str(),"w",stdout);
-    assert(fp!=(FILE *)NULL);
-  }
-
-  ////////////////////////////////////
-  // Banner
-  ////////////////////////////////////
-  if ( CartesianCommunicator::RankWorld() == 0 ) { 
-    std::cout <<std::endl;
-    std::cout  << "__|__|__|__|__|__|__|__|__|__|__|__|__|__|__"<<std::endl; 
-    std::cout  << "__|__|__|__|__|__|__|__|__|__|__|__|__|__|__"<<std::endl; 
-    std::cout  << "__|_ |  |  |  |  |  |  |  |  |  |  |  | _|__"<<std::endl; 
-    std::cout  << "__|_                                    _|__"<<std::endl; 
-    std::cout  << "__|_   GGGG    RRRR    III    DDDD      _|__"<<std::endl;
-    std::cout  << "__|_  G        R   R    I     D   D     _|__"<<std::endl;
-    std::cout  << "__|_  G        R   R    I     D    D    _|__"<<std::endl;
-    std::cout  << "__|_  G  GG    RRRR     I     D    D    _|__"<<std::endl;
-    std::cout  << "__|_  G   G    R  R     I     D   D     _|__"<<std::endl;
-    std::cout  << "__|_   GGGG    R   R   III    DDDD      _|__"<<std::endl;
-    std::cout  << "__|_                                    _|__"<<std::endl; 
-    std::cout  << "__|__|__|__|__|__|__|__|__|__|__|__|__|__|__"<<std::endl; 
-    std::cout  << "__|__|__|__|__|__|__|__|__|__|__|__|__|__|__"<<std::endl; 
-    std::cout  << "  |  |  |  |  |  |  |  |  |  |  |  |  |  |  "<<std::endl; 
-    std::cout << std::endl;
-    std::cout << std::endl;
-    std::cout << "Copyright (C) 2015 Peter Boyle, Azusa Yamaguchi, Guido Cossu, Antonin Portelli and other authors"<<std::endl;
-    std::cout << std::endl;
-    std::cout << "This program is free software; you can redistribute it and/or modify"<<std::endl;
-    std::cout << "it under the terms of the GNU General Public License as published by"<<std::endl;
-    std::cout << "the Free Software Foundation; either version 2 of the License, or"<<std::endl;
-    std::cout << "(at your option) any later version."<<std::endl;
-    std::cout << std::endl;
-    std::cout << "This program is distributed in the hope that it will be useful,"<<std::endl;
-    std::cout << "but WITHOUT ANY WARRANTY; without even the implied warranty of"<<std::endl;
-    std::cout << "MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the"<<std::endl;
-    std::cout << "GNU General Public License for more details."<<std::endl;
-    std::cout << std::endl;
-  }
-
-
   ////////////////////////////////////
   // Logging
   ////////////////////////////////////
@@ -284,6 +230,9 @@ void Grid_init(int *argc,char ***argv)
   GridCmdOptionCSL(defaultLog,logstreams);
   GridLogConfigure(logstreams);
 
+  if( !GridCmdOptionExists(*argv,*argv+*argc,"--debug-stdout") ){
+    Grid_quiesce_nodes();
+  }
 
   if( GridCmdOptionExists(*argv,*argv+*argc,"--log") ){
     arg = GridCmdOptionPayload(*argv,*argv+*argc,"--log");
@@ -299,74 +248,101 @@ void Grid_init(int *argc,char ***argv)
     std::cout<<GridLogMessage<<"  --help : this message"<<std::endl;
     std::cout<<GridLogMessage<<std::endl;
     std::cout<<GridLogMessage<<"Geometry:"<<std::endl;
-    std::cout<<GridLogMessage<<std::endl;
     std::cout<<GridLogMessage<<"  --mpi n.n.n.n   : default MPI decomposition"<<std::endl;    
     std::cout<<GridLogMessage<<"  --threads n     : default number of OMP threads"<<std::endl;
-    std::cout<<GridLogMessage<<"  --grid n.n.n.n  : default Grid size"<<std::endl;
+    std::cout<<GridLogMessage<<"  --grid n.n.n.n  : default Grid size"<<std::endl;    
-    std::cout<<GridLogMessage<<"  --shm  M        : allocate M megabytes of shared memory for comms"<<std::endl;
+    std::cout<<GridLogMessage<<"  --shm  M        : allocate M megabytes of shared memory for comms"<<std::endl;    
     std::cout<<GridLogMessage<<std::endl;
     std::cout<<GridLogMessage<<"Verbose and debug:"<<std::endl;
-    std::cout<<GridLogMessage<<std::endl;
+    std::cout<<GridLogMessage<<"  --log list      : comma separted list of streams from Error,Warning,Message,Performance,Iterative,Integrator,Debug,Colours"<<std::endl;
-    std::cout<<GridLogMessage<<"  --log list      : comma separated list from Error,Warning,Message,Performance,Iterative,Integrator,Debug,Colours"<<std::endl;
     std::cout<<GridLogMessage<<"  --decomposition : report on default omp,mpi and simd decomposition"<<std::endl;    
     std::cout<<GridLogMessage<<"  --debug-signals : catch sigsegv and print a blame report"<<std::endl;
-    std::cout<<GridLogMessage<<"  --debug-stdout  : print stdout from EVERY node"<<std::endl;
+    std::cout<<GridLogMessage<<"  --debug-stdout  : print stdout from EVERY node"<<std::endl;    
-    std::cout<<GridLogMessage<<"  --notimestamp   : suppress millisecond resolution stamps"<<std::endl;
+    std::cout<<GridLogMessage<<"  --notimestamp   : suppress millisecond resolution stamps"<<std::endl;    
     std::cout<<GridLogMessage<<std::endl;
     std::cout<<GridLogMessage<<"Performance:"<<std::endl;
-    std::cout<<GridLogMessage<<std::endl;
-    std::cout<<GridLogMessage<<"  --comms-concurrent : Asynchronous MPI calls; several dirs at a time "<<std::endl;    
-    std::cout<<GridLogMessage<<"  --comms-sequential : Synchronous MPI calls; one dirs at a time "<<std::endl;    
-    std::cout<<GridLogMessage<<"  --comms-overlap : Overlap comms with compute "<<std::endl;    
-    std::cout<<GridLogMessage<<std::endl;
     std::cout<<GridLogMessage<<"  --dslash-generic: Wilson kernel for generic Nc"<<std::endl;    
     std::cout<<GridLogMessage<<"  --dslash-unroll : Wilson kernel for Nc=3"<<std::endl;    
     std::cout<<GridLogMessage<<"  --dslash-asm    : Wilson kernel for AVX512"<<std::endl;    
-    std::cout<<GridLogMessage<<std::endl;
     std::cout<<GridLogMessage<<"  --lebesgue      : Cache oblivious Lebesgue curve/Morton order/Z-graph stencil looping"<<std::endl;    
     std::cout<<GridLogMessage<<"  --cacheblocking n.m.o.p : Hypercuboidal cache blocking"<<std::endl;    
     std::cout<<GridLogMessage<<std::endl;
     exit(EXIT_SUCCESS);
   }
 
+  ////////////////////////////////////
+  // Banner
+  ////////////////////////////////////
+
+  std::string COL_RED    = GridLogColours.colour["RED"];
+  std::string COL_PURPLE = GridLogColours.colour["PURPLE"];
+  std::string COL_BLACK  = GridLogColours.colour["BLACK"];
+  std::string COL_GREEN  = GridLogColours.colour["GREEN"];
+  std::string COL_BLUE   = GridLogColours.colour["BLUE"];
+  std::string COL_YELLOW = GridLogColours.colour["YELLOW"];
+  std::string COL_BACKGROUND = GridLogColours.colour["NORMAL"];
+  
+  std::cout <<std::endl;
+  std::cout <<COL_RED  << "__|__|__|__|__"<<             "|__|__|_"<<COL_PURPLE<<"_|__|__|"<<                "__|__|__|__|__"<<std::endl; 
+  std::cout <<COL_RED  << "__|__|__|__|__"<<             "|__|__|_"<<COL_PURPLE<<"_|__|__|"<<                "__|__|__|__|__"<<std::endl; 
+  std::cout <<COL_RED  << "__|_ |  |  |  "<<             "|  |  | "<<COL_PURPLE<<" |  |  |"<<                "  |  |  | _|__"<<std::endl; 
+  std::cout <<COL_RED  << "__|_          "<<             "        "<<COL_PURPLE<<"        "<<                "          _|__"<<std::endl; 
+  std::cout <<COL_RED  << "__|_  "<<COL_GREEN<<" GGGG   "<<COL_RED<<" RRRR   "<<COL_BLUE  <<" III    "<<COL_PURPLE<<"DDDD  "<<COL_PURPLE<<"    _|__"<<std::endl;
+  std::cout <<COL_RED  << "__|_  "<<COL_GREEN<<"G       "<<COL_RED<<" R   R  "<<COL_BLUE  <<"  I     "<<COL_PURPLE<<"D   D "<<COL_PURPLE<<"    _|__"<<std::endl;
+  std::cout <<COL_RED  << "__|_  "<<COL_GREEN<<"G       "<<COL_RED<<" R   R  "<<COL_BLUE  <<"  I     "<<COL_PURPLE<<"D    D"<<COL_PURPLE<<"    _|__"<<std::endl;
+  std::cout <<COL_BLUE << "__|_  "<<COL_GREEN<<"G  GG   "<<COL_RED<<" RRRR   "<<COL_BLUE  <<"  I     "<<COL_PURPLE<<"D    D"<<COL_GREEN <<"    _|__"<<std::endl;
+  std::cout <<COL_BLUE << "__|_  "<<COL_GREEN<<"G   G   "<<COL_RED<<" R  R   "<<COL_BLUE  <<"  I     "<<COL_PURPLE<<"D   D "<<COL_GREEN <<"    _|__"<<std::endl;
+  std::cout <<COL_BLUE << "__|_  "<<COL_GREEN<<" GGGG   "<<COL_RED<<" R   R  "<<COL_BLUE  <<" III    "<<COL_PURPLE<<"DDDD  "<<COL_GREEN <<"    _|__"<<std::endl;
+  std::cout <<COL_BLUE << "__|_          "<<             "        "<<COL_GREEN <<"        "<<                "          _|__"<<std::endl; 
+  std::cout <<COL_BLUE << "__|__|__|__|__"<<             "|__|__|_"<<COL_GREEN <<"_|__|__|"<<                "__|__|__|__|__"<<std::endl; 
+  std::cout <<COL_BLUE << "__|__|__|__|__"<<             "|__|__|_"<<COL_GREEN <<"_|__|__|"<<                "__|__|__|__|__"<<std::endl; 
+  std::cout <<COL_BLUE << "  |  |  |  |  "<<             "|  |  | "<<COL_GREEN <<" |  |  |"<<                "  |  |  |  |  "<<std::endl; 
+  std::cout << std::endl;
+  std::cout << std::endl;
+  std::cout <<COL_YELLOW<< std::endl;
+  std::cout << "Copyright (C) 2015 Peter Boyle, Azusa Yamaguchi, Guido Cossu, Antonin Portelli and other authors"<<std::endl;
+  std::cout << std::endl;
+  std::cout << "This program is free software; you can redistribute it and/or modify"<<std::endl;
+  std::cout << "it under the terms of the GNU General Public License as published by"<<std::endl;
+  std::cout << "the Free Software Foundation; either version 2 of the License, or"<<std::endl;
+  std::cout << "(at your option) any later version."<<std::endl;
+  std::cout << std::endl;
+  std::cout << "This program is distributed in the hope that it will be useful,"<<std::endl;
+  std::cout << "but WITHOUT ANY WARRANTY; without even the implied warranty of"<<std::endl;
+  std::cout << "MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the"<<std::endl;
+  std::cout << "GNU General Public License for more details."<<std::endl;
+  std::cout << COL_BACKGROUND <<std::endl;
+  std::cout << std::endl;
+
   ////////////////////////////////////
   // Debug and performance options
   ////////////////////////////////////
 
+
+  if( GridCmdOptionExists(*argv,*argv+*argc,"--debug-signals") ){
+    Grid_debug_handler_init();
+  }
   if( GridCmdOptionExists(*argv,*argv+*argc,"--dslash-unroll") ){
     QCD::WilsonKernelsStatic::Opt=QCD::WilsonKernelsStatic::OptHandUnroll;
-    QCD::StaggeredKernelsStatic::Opt=QCD::StaggeredKernelsStatic::OptHandUnroll;
   }
   if( GridCmdOptionExists(*argv,*argv+*argc,"--dslash-asm") ){
     QCD::WilsonKernelsStatic::Opt=QCD::WilsonKernelsStatic::OptInlineAsm;
-    QCD::StaggeredKernelsStatic::Opt=QCD::StaggeredKernelsStatic::OptInlineAsm;
   }
   if( GridCmdOptionExists(*argv,*argv+*argc,"--dslash-generic") ){
     QCD::WilsonKernelsStatic::Opt=QCD::WilsonKernelsStatic::OptGeneric;
-    QCD::StaggeredKernelsStatic::Opt=QCD::StaggeredKernelsStatic::OptGeneric;
-  }
-  if( GridCmdOptionExists(*argv,*argv+*argc,"--comms-overlap") ){
-    QCD::WilsonKernelsStatic::Comms = QCD::WilsonKernelsStatic::CommsAndCompute;
-  } else {
-    QCD::WilsonKernelsStatic::Comms = QCD::WilsonKernelsStatic::CommsThenCompute;
-  }
-  if( GridCmdOptionExists(*argv,*argv+*argc,"--comms-concurrent") ){
-    CartesianCommunicator::SetCommunicatorPolicy(CartesianCommunicator::CommunicatorPolicyConcurrent);
-  }
-  if( GridCmdOptionExists(*argv,*argv+*argc,"--comms-sequential") ){
-    CartesianCommunicator::SetCommunicatorPolicy(CartesianCommunicator::CommunicatorPolicySequential);
   }
   if( GridCmdOptionExists(*argv,*argv+*argc,"--lebesgue") ){
     LebesgueOrder::UseLebesgueOrder=1;
   }
 
+
   if( GridCmdOptionExists(*argv,*argv+*argc,"--cacheblocking") ){
     arg= GridCmdOptionPayload(*argv,*argv+*argc,"--cacheblocking");
     GridCmdOptionIntVector(arg,LebesgueOrder::Block);
   }
   if( GridCmdOptionExists(*argv,*argv+*argc,"--notimestamp") ){
     GridLogTimestamp(0);
-  } else {
+  } else { 
     GridLogTimestamp(1);
   }
 
@@ -374,7 +350,7 @@ void Grid_init(int *argc,char ***argv)
 		  Grid_default_latt,
 		  Grid_default_mpi);
 
-  std::cout << GridLogDebug << "Requesting "<< CartesianCommunicator::MAX_MPI_SHM_BYTES <<" byte stencil comms buffers "<<std::endl;
+  std::cout << GridLogMessage << "Requesting "<< CartesianCommunicator::MAX_MPI_SHM_BYTES <<" byte stencil comms buffers "<<std::endl;
 
   if( GridCmdOptionExists(*argv,*argv+*argc,"--decomposition") ){
     std::cout<<GridLogMessage<<"Grid Decomposition\n";
@@ -390,39 +366,30 @@ void Grid_init(int *argc,char ***argv)
   Grid_is_initialised = 1;
 }
 
-
+  
 void Grid_finalize(void)
 {
-#if defined (GRID_COMMS_MPI) || defined (GRID_COMMS_MPI3)
+#if defined (GRID_COMMS_MPI) || defined (GRID_COMMS_MPI3) 
   MPI_Finalize();
   Grid_unquiesce_nodes();
 #endif
-#if defined (GRID_COMMS_SHMEM)
-  shmem_finalize();
-#endif
-}
-
-void GridLogLayout() {
-    std::cout << GridLogMessage << "Grid Layout\n";
-    std::cout << GridLogMessage << "\tGlobal lattice size  : "<< GridCmdVectorIntToString(GridDefaultLatt()) << std::endl;
-    std::cout << GridLogMessage << "\tOpenMP threads       : "<< GridThread::GetThreads() <<std::endl;
-    std::cout << GridLogMessage << "\tMPI tasks            : "<< GridCmdVectorIntToString(GridDefaultMpi()) << std::endl;
 }
 
 void * Grid_backtrace_buffer[_NBACKTRACE];
 
 void Grid_sa_signal_handler(int sig,siginfo_t *si,void * ptr)
 {
-  fprintf(stderr,"Caught signal %d\n",si->si_signo);
+  printf("Caught signal %d\n",si->si_signo);
-  fprintf(stderr,"  mem address %llx\n",(unsigned long long)si->si_addr);
+  printf("  mem address %llx\n",(unsigned long long)si->si_addr);
-  fprintf(stderr,"         code %d\n",si->si_code);
+  printf("         code %d\n",si->si_code);
+
   // Linux/Posix
 #ifdef __linux__
   // And x86 64bit
 #ifdef __x86_64__
   ucontext_t * uc= (ucontext_t *)ptr;
   struct sigcontext *sc = (struct sigcontext *)&uc->uc_mcontext;
-  fprintf(stderr,"  instruction %llx\n",(unsigned long long)sc->rip);
+  printf("  instruction %llx\n",(unsigned long long)sc->rip);
 #define REG(A)  printf("  %s %lx\n",#A,sc-> A);
   REG(rdi);
   REG(rsi);
@@ -445,11 +412,7 @@ void Grid_sa_signal_handler(int sig,siginfo_t *si,void * ptr)
   REG(r15);
 #endif
 #endif
-  fflush(stderr);
+  BACKTRACE();
-  BACKTRACEFP(stderr);
-  fprintf(stderr,"Called backtrace\n");
-  fflush(stdout);
-  fflush(stderr);
   exit(0);
   return;
 };
@@ -462,12 +425,9 @@ void Grid_debug_handler_init(void)
   sa.sa_flags    = SA_SIGINFO;
   sigaction(SIGSEGV,&sa,NULL);
   sigaction(SIGTRAP,&sa,NULL);
-  sigaction(SIGBUS,&sa,NULL);
 
   feenableexcept( FE_INVALID|FE_OVERFLOW|FE_DIVBYZERO);
 
   sigaction(SIGFPE,&sa,NULL);
-  sigaction(SIGKILL,&sa,NULL);
-  sigaction(SIGILL,&sa,NULL);
 }
 }

lib/Init.h

@@ -1,6 +1,6 @@
     /*************************************************************************************
 
-    Grid physics library, www.github.com/paboyle/Grid
+    Grid physics library, www.github.com/paboyle/Grid 
 
     Source file: ./lib/Init.h
 
@@ -46,7 +46,6 @@ namespace Grid {
   const int              &GridThreads(void)  ;
   void                    GridSetThreads(int t) ;
   void GridLogTimestamp(int);
-  void GridLogLayout();
 
   // Common parsing chores
   std::string GridCmdOptionPayload(char ** begin, char ** end, const std::string & option);

lib/Log.cc

@@ -29,11 +29,9 @@ See the full license in the file "LICENSE" in the top level distribution
 directory
 *************************************************************************************/
 /*  END LEGAL */
-#include <Grid/GridCore.h>
+#include <Grid/Grid.h>
-#include <Grid/util/CompilerCompatible.h>
 
 #include <cxxabi.h>
-#include <memory>
 
 namespace Grid {
 

lib/Log.h

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

lib/Old/Tensor_peek.h

@@ -0,0 +1,154 @@
+    /*************************************************************************************
+
+    Grid physics library, www.github.com/paboyle/Grid 
+
+    Source file: ./lib/Old/Tensor_peek.h
+
+    Copyright (C) 2015
+
+Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License along
+    with this program; if not, write to the Free Software Foundation, Inc.,
+    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+    See the full license in the file "LICENSE" in the top level distribution directory
+    *************************************************************************************/
+    /*  END LEGAL */
+#ifndef GRID_MATH_PEEK_H
+#define GRID_MATH_PEEK_H
+namespace Grid {
+
+//////////////////////////////////////////////////////////////////////////////
+// Peek on a specific index; returns a scalar in that index, tensor inherits rest
+//////////////////////////////////////////////////////////////////////////////
+// If we hit the right index, return scalar with no further recursion
+
+//template<int Level> inline ComplexF peekIndex(const ComplexF arg) { return arg;}
+//template<int Level> inline ComplexD peekIndex(const ComplexD arg) { return arg;}
+//template<int Level> inline RealF peekIndex(const RealF arg) { return arg;}
+//template<int Level> inline RealD peekIndex(const RealD arg) { return arg;}
+#if 0
+// Scalar peek, no indices
+template<int Level,class vtype,typename std::enable_if< iScalar<vtype>::TensorLevel == Level >::type * =nullptr> inline 
+  auto peekIndex(const iScalar<vtype> &arg) ->  iScalar<vtype> 
+{
+  return arg;
+}
+// Vector peek, one index
+template<int Level,class vtype,int N,typename std::enable_if< iScalar<vtype>::TensorLevel == Level >::type * =nullptr> inline 
+  auto peekIndex(const iVector<vtype,N> &arg,int i) -> iScalar<vtype> // Index matches
+{
+  iScalar<vtype> ret;                              // return scalar
+  ret._internal = arg._internal[i];
+  return ret;
+}
+// Matrix peek, two indices
+template<int Level,class vtype,int N,typename std::enable_if< iScalar<vtype>::TensorLevel == Level >::type * =nullptr> inline 
+  auto peekIndex(const iMatrix<vtype,N> &arg,int i,int j) ->  iScalar<vtype>
+{
+  iScalar<vtype> ret;                              // return scalar
+  ret._internal = arg._internal[i][j];
+  return ret;
+}
+
+/////////////
+// No match peek for scalar,vector,matrix must forward on either 0,1,2 args. Must have 9 routines with notvalue
+/////////////
+// scalar
+template<int Level,class vtype,typename std::enable_if< iScalar<vtype>::TensorLevel != Level >::type * =nullptr> inline 
+  auto peekIndex(const iScalar<vtype> &arg) -> iScalar<decltype(peekIndex<Level>(arg._internal))>
+{
+  iScalar<decltype(peekIndex<Level>(arg._internal))> ret;
+  ret._internal= peekIndex<Level>(arg._internal);
+  return ret;
+}
+template<int Level,class vtype, typename std::enable_if< iScalar<vtype>::TensorLevel != Level >::type * =nullptr> inline 
+  auto peekIndex(const iScalar<vtype> &arg,int i) ->  iScalar<decltype(peekIndex<Level>(arg._internal,i))> 
+{
+  iScalar<decltype(peekIndex<Level>(arg._internal,i))> ret;
+  ret._internal=peekIndex<Level>(arg._internal,i);
+  return ret;
+}
+template<int Level,class vtype, typename std::enable_if< iScalar<vtype>::TensorLevel != Level >::type * =nullptr> inline 
+  auto peekIndex(const iScalar<vtype> &arg,int i,int j) ->  iScalar<decltype(peekIndex<Level>(arg._internal,i,j))>
+{
+  iScalar<decltype(peekIndex<Level>(arg._internal,i,j))> ret;
+  ret._internal=peekIndex<Level>(arg._internal,i,j);
+  return ret;
+}
+// vector
+template<int Level,class vtype,int N, typename std::enable_if< iScalar<vtype>::TensorLevel != Level >::type * =nullptr> inline 
+auto peekIndex(const iVector<vtype,N> &arg) ->   iVector<decltype(peekIndex<Level>(arg._internal[0])),N>
+{
+  iVector<decltype(peekIndex<Level>(arg._internal[0])),N> ret;
+  for(int ii=0;ii<N;ii++){
+    ret._internal[ii]=peekIndex<Level>(arg._internal[ii]);
+  }
+  return ret;
+}
+template<int Level,class vtype,int N, typename std::enable_if< iScalar<vtype>::TensorLevel != Level >::type * =nullptr> inline 
+  auto peekIndex(const iVector<vtype,N> &arg,int i) ->  iVector<decltype(peekIndex<Level>(arg._internal[0],i)),N>
+{
+  iVector<decltype(peekIndex<Level>(arg._internal[0],i)),N> ret;
+  for(int ii=0;ii<N;ii++){
+    ret._internal[ii]=peekIndex<Level>(arg._internal[ii],i);
+  }
+  return ret;
+}
+template<int Level,class vtype,int N, typename std::enable_if< iScalar<vtype>::TensorLevel != Level >::type * =nullptr> inline 
+  auto peekIndex(const iVector<vtype,N> &arg,int i,int j) ->  iVector<decltype(peekIndex<Level>(arg._internal[0],i,j)),N> 
+{
+  iVector<decltype(peekIndex<Level>(arg._internal[0],i,j)),N> ret;
+  for(int ii=0;ii<N;ii++){
+    ret._internal[ii]=peekIndex<Level>(arg._internal[ii],i,j);
+  }
+  return ret;
+}
+
+// matrix
+template<int Level,class vtype,int N, typename std::enable_if< iScalar<vtype>::TensorLevel != Level >::type * =nullptr> inline 
+auto peekIndex(const iMatrix<vtype,N> &arg) ->   iMatrix<decltype(peekIndex<Level>(arg._internal[0][0])),N> 
+{
+  iMatrix<decltype(peekIndex<Level>(arg._internal[0][0])),N> ret;
+  for(int ii=0;ii<N;ii++){
+  for(int jj=0;jj<N;jj++){
+    ret._internal[ii][jj]=peekIndex<Level>(arg._internal[ii][jj]);// Could avoid this because peeking a scalar is dumb
+  }}
+  return ret;
+}
+template<int Level,class vtype,int N, typename std::enable_if< iScalar<vtype>::TensorLevel != Level >::type * =nullptr> inline 
+  auto peekIndex(const iMatrix<vtype,N> &arg,int i) ->   iMatrix<decltype(peekIndex<Level>(arg._internal[0][0],i)),N>
+{
+  iMatrix<decltype(peekIndex<Level>(arg._internal[0][0],i)),N> ret;
+  for(int ii=0;ii<N;ii++){
+  for(int jj=0;jj<N;jj++){
+    ret._internal[ii][jj]=peekIndex<Level>(arg._internal[ii][jj],i);
+  }}
+  return ret;
+}
+template<int Level,class vtype,int N, typename std::enable_if< iScalar<vtype>::TensorLevel != Level >::type * =nullptr> inline 
+  auto peekIndex(const iMatrix<vtype,N> &arg,int i,int j) ->   iMatrix<decltype(peekIndex<Level>(arg._internal[0][0],i,j)),N>
+{
+  iMatrix<decltype(peekIndex<Level>(arg._internal[0][0],i,j)),N> ret;
+  for(int ii=0;ii<N;ii++){
+  for(int jj=0;jj<N;jj++){
+    ret._internal[ii][jj]=peekIndex<Level>(arg._internal[ii][jj],i,j);
+  }}
+  return ret;
+}
+#endif
+
+
+}
+#endif

lib/Old/Tensor_poke.h

@@ -0,0 +1,127 @@
+    /*************************************************************************************
+
+    Grid physics library, www.github.com/paboyle/Grid 
+
+    Source file: ./lib/Old/Tensor_poke.h
+
+    Copyright (C) 2015
+
+Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License along
+    with this program; if not, write to the Free Software Foundation, Inc.,
+    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+    See the full license in the file "LICENSE" in the top level distribution directory
+    *************************************************************************************/
+    /*  END LEGAL */
+#ifndef GRID_MATH_POKE_H
+#define GRID_MATH_POKE_H
+namespace Grid {
+
+//////////////////////////////////////////////////////////////////////////////
+// Poke a specific index; 
+//////////////////////////////////////////////////////////////////////////////
+#if 0
+// Scalar poke
+template<int Level,class vtype,typename std::enable_if< iScalar<vtype>::TensorLevel == Level >::type * =nullptr> inline 
+  void pokeIndex(iScalar<vtype> &ret, const iScalar<vtype> &arg)
+{
+  ret._internal = arg._internal;
+}
+// Vector poke, one index
+template<int Level,class vtype,int N,typename std::enable_if< iScalar<vtype>::TensorLevel == Level >::type * =nullptr> inline 
+  void pokeIndex(iVector<vtype,N> &ret, const iScalar<vtype> &arg,int i)
+{
+  ret._internal[i] = arg._internal;
+}
+//Matrix poke, two indices
+template<int Level,class vtype,int N,typename std::enable_if< iScalar<vtype>::TensorLevel == Level >::type * =nullptr> inline 
+  void pokeIndex(iMatrix<vtype,N> &ret, const iScalar<vtype> &arg,int i,int j)
+{
+  ret._internal[i][j] = arg._internal;
+}
+
+/////////////
+// No match poke for scalar,vector,matrix must forward on either 0,1,2 args. Must have 9 routines with notvalue
+/////////////
+// scalar
+template<int Level,class vtype,typename std::enable_if< iScalar<vtype>::TensorLevel != Level >::type * =nullptr> inline 
+void pokeIndex(iScalar<vtype> &ret, const iScalar<decltype(peekIndex<Level>(ret._internal))>  &arg)
+{
+  pokeIndex<Level>(ret._internal,arg._internal);
+}
+template<int Level,class vtype,typename std::enable_if< iScalar<vtype>::TensorLevel != Level >::type * =nullptr> inline 
+  void pokeIndex(iScalar<vtype> &ret, const iScalar<decltype(peekIndex<Level>(ret._internal,0))> &arg, int i)
+		 
+{
+  pokeIndex<Level>(ret._internal,arg._internal,i);
+}
+template<int Level,class vtype,typename std::enable_if< iScalar<vtype>::TensorLevel != Level >::type * =nullptr> inline 
+  void pokeIndex(iScalar<vtype> &ret, const iScalar<decltype(peekIndex<Level>(ret._internal,0,0))> &arg,int i,int j)
+{
+  pokeIndex<Level>(ret._internal,arg._internal,i,j);
+}
+
+// Vector
+template<int Level,class vtype,int N,typename std::enable_if< iScalar<vtype>::TensorLevel != Level >::type * =nullptr> inline 
+  void pokeIndex(iVector<vtype,N> &ret, iVector<decltype(peekIndex<Level>(ret._internal)),N>  &arg)
+{
+  for(int ii=0;ii<N;ii++){
+    pokeIndex<Level>(ret._internal[ii],arg._internal[ii]);
+  }
+}
+template<int Level,class vtype,int N,typename std::enable_if< iScalar<vtype>::TensorLevel != Level >::type * =nullptr> inline 
+  void pokeIndex(iVector<vtype,N> &ret, const iVector<decltype(peekIndex<Level>(ret._internal,0)),N> &arg,int i)
+{
+  for(int ii=0;ii<N;ii++){
+    pokeIndex<Level>(ret._internal[ii],arg._internal[ii],i);
+  }
+}
+template<int Level,class vtype,int N,typename std::enable_if< iScalar<vtype>::TensorLevel != Level >::type * =nullptr> inline 
+  void pokeIndex(iVector<vtype,N> &ret, const iVector<decltype(peekIndex<Level>(ret._internal,0,0)),N> &arg,int i,int j)
+{
+  for(int ii=0;ii<N;ii++){
+    pokeIndex<Level>(ret._internal[ii],arg._internal[ii],i,j);
+  }
+}
+
+// Matrix
+template<int Level,class vtype,int N,typename std::enable_if< iScalar<vtype>::TensorLevel != Level >::type * =nullptr> inline 
+  void pokeIndex(iMatrix<vtype,N> &ret, const iMatrix<decltype(peekIndex<Level>(ret._internal)),N> &arg)		 
+{
+  for(int ii=0;ii<N;ii++){
+  for(int jj=0;jj<N;jj++){
+    pokeIndex<Level>(ret._internal[ii][jj],arg._internal[ii][jj]);
+  }}
+}
+template<int Level,class vtype,int N,typename std::enable_if< iScalar<vtype>::TensorLevel != Level >::type * =nullptr> inline 
+  void pokeIndex(iMatrix<vtype,N> &ret, const iMatrix<decltype(peekIndex<Level>(ret._internal,0)),N> &arg,int i)
+{
+  for(int ii=0;ii<N;ii++){
+  for(int jj=0;jj<N;jj++){
+    pokeIndex<Level>(ret._internal[ii][jj],arg._internal[ii][jj],i);
+  }}
+}
+template<int Level,class vtype,int N,typename std::enable_if< iScalar<vtype>::TensorLevel != Level >::type * =nullptr> inline 
+  void pokeIndex(iMatrix<vtype,N> &ret, const iMatrix<decltype(peekIndex<Level>(ret._internal,0,0)),N> &arg, int i,int j)
+{
+  for(int ii=0;ii<N;ii++){
+  for(int jj=0;jj<N;jj++){
+    pokeIndex<Level>(ret._internal[ii][jj],arg._internal[ii][jj],i,j);
+  }}
+}
+#endif
+
+}
+#endif

lib/PerfCount.cc

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

lib/PerfCount.h

@@ -172,7 +172,7 @@ public:
     const char * name = PerformanceCounterConfigs[PCT].name;
     fd = perf_event_open(&pe, 0, -1, -1, 0); // pid 0, cpu -1 current process any cpu. group -1
     if (fd == -1) {
-      fprintf(stderr, "Error opening leader %llx for event %s\n",(long long) pe.config,name);
+      fprintf(stderr, "Error opening leader %llx for event %s\n", pe.config,name);
       perror("Error is");
     }
     int norm = PerformanceCounterConfigs[PCT].normalisation;
@@ -181,7 +181,7 @@ public:
     name = PerformanceCounterConfigs[norm].name;
     cyclefd = perf_event_open(&pe, 0, -1, -1, 0); // pid 0, cpu -1 current process any cpu. group -1
     if (cyclefd == -1) {
-      fprintf(stderr, "Error opening leader %llx for event %s\n",(long long) pe.config,name);
+      fprintf(stderr, "Error opening leader %llx for event %s\n", pe.config,name);
       perror("Error is");
     }
 #endif
@@ -205,14 +205,13 @@ public:
   void Stop(void) {
     count=0;
     cycles=0;
+    size_t ign;
 #ifdef __linux__
-    ssize_t ign;
     if ( fd!= -1) {
       ::ioctl(fd, PERF_EVENT_IOC_DISABLE, 0);
       ::ioctl(cyclefd, PERF_EVENT_IOC_DISABLE, 0);
       ign=::read(fd, &count, sizeof(long long));
-      ign+=::read(cyclefd, &cycles, sizeof(long long));
+      ign=::read(cyclefd, &cycles, sizeof(long long));
-      assert(ign=2*sizeof(long long));
     }
     elapsed = cyclecount() - begin;
 #else

lib/Simd.h

@@ -172,8 +172,8 @@ namespace Grid {
 
 };
 
-#include <Grid/simd/Grid_vector_types.h>
+#include "simd/Grid_vector_types.h"
-#include <Grid/simd/Grid_vector_unops.h>
+#include "simd/Grid_vector_unops.h"
 
 namespace Grid {
   // Default precision

lib/Stat.cc

@@ -1,9 +1,11 @@
-#include <Grid/GridCore.h>
+#include <Grid/Grid.h>
-#include <Grid/perfmon/PerfCount.h>
+#include <Grid/PerfCount.h>
-#include <Grid/perfmon/Stat.h>
+#include <Grid/Stat.h>
+
 
 namespace Grid { 
 
+
 bool PmuStat::pmu_initialized=false;
 
 

lib/Threads.h

@@ -37,9 +37,13 @@ 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)")
+#endif
 #define PARALLEL_NESTED_LOOP2 _Pragma("omp parallel for collapse(2)")
 #define PARALLEL_REGION       _Pragma("omp parallel")
 #define PARALLEL_CRITICAL     _Pragma("omp critical")
@@ -51,9 +55,6 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 #define PARALLEL_CRITICAL
 #endif
 
-#define parallel_for       PARALLEL_FOR_LOOP for
-#define parallel_for_nest2 PARALLEL_NESTED_LOOP2 for
-
 namespace Grid {
 
   // Introduce a class to gain deterministic bit reproducible reduction.

lib/algorithms/CoarsenedMatrix.h

@@ -267,7 +267,8 @@ namespace Grid {
       SimpleCompressor<siteVector> compressor;
       Stencil.HaloExchange(in,compressor);
 
-      parallel_for(int ss=0;ss<Grid()->oSites();ss++){
+PARALLEL_FOR_LOOP
+      for(int ss=0;ss<Grid()->oSites();ss++){
         siteVector res = zero;
 	siteVector nbr;
 	int ptype;
@@ -379,7 +380,8 @@ namespace Grid {
 	  Subspace.ProjectToSubspace(oProj,oblock);
 	  //	  blockProject(iProj,iblock,Subspace.subspace);
 	  //	  blockProject(oProj,oblock,Subspace.subspace);
-	  parallel_for(int ss=0;ss<Grid()->oSites();ss++){
+PARALLEL_FOR_LOOP
+	  for(int ss=0;ss<Grid()->oSites();ss++){
 	    for(int j=0;j<nbasis;j++){
 	      if( disp!= 0 ) {
 		A[p]._odata[ss](j,i) = oProj._odata[ss](j);
@@ -425,7 +427,7 @@ namespace Grid {
 	A[p]=zero;
       }
 
-      GridParallelRNG  RNG(Grid()); RNG.SeedFixedIntegers(std::vector<int>({55,72,19,17,34}));
+      GridParallelRNG  RNG(Grid()); RNG.SeedRandomDevice();
       Lattice<iScalar<CComplex> > val(Grid()); random(RNG,val);
 
       Complex one(1.0);

lib/algorithms/LinearOperator.h

@@ -235,7 +235,7 @@ namespace Grid {
 	Field tmp(in._grid);
 
 	_Mat.MeooeDag(in,tmp);
-        _Mat.MooeeInvDag(tmp,out);
+	_Mat.MooeeInvDag(tmp,out);
 	_Mat.MeooeDag(out,tmp);
 
 	_Mat.MooeeDag(in,out);

lib/algorithms/approx/Chebyshev.h

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

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/GridCore.h>
+#include <Grid/Grid.h>
 
 namespace Grid {
 double MultiShiftFunction::approx(double x)

lib/algorithms/approx/Remez.h

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

lib/algorithms/iterative/BlockConjugateGradient.h

@@ -1,366 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/algorithms/iterative/BlockConjugateGradient.h
-
-Copyright (C) 2017
-
-Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
-
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
-
-See the full license in the file "LICENSE" in the top level distribution
-directory
-*************************************************************************************/
-/*  END LEGAL */
-#ifndef GRID_BLOCK_CONJUGATE_GRADIENT_H
-#define GRID_BLOCK_CONJUGATE_GRADIENT_H
-
-
-namespace Grid {
-
-//////////////////////////////////////////////////////////////////////////
-// Block conjugate gradient. Dimension zero should be the block direction
-//////////////////////////////////////////////////////////////////////////
-template <class Field>
-class BlockConjugateGradient : public OperatorFunction<Field> {
- public:
-
-  typedef typename Field::scalar_type scomplex;
-
-  const int blockDim = 0;
-
-  int Nblock;
-  bool ErrorOnNoConverge;  // throw an assert when the CG fails to converge.
-                           // Defaults true.
-  RealD Tolerance;
-  Integer MaxIterations;
-  Integer IterationsToComplete; //Number of iterations the CG took to finish. Filled in upon completion
-  
-  BlockConjugateGradient(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) 
-{
-  int Orthog = 0; // First dimension is block dim
-  Nblock = Src._grid->_fdimensions[Orthog];
-
-  std::cout<<GridLogMessage<<" Block Conjugate Gradient : Orthog "<<Orthog<<" Nblock "<<Nblock<<std::endl;
-
-  Psi.checkerboard = Src.checkerboard;
-  conformable(Psi, Src);
-
-  Field P(Src);
-  Field AP(Src);
-  Field R(Src);
-  
-  Eigen::MatrixXcd m_pAp    = Eigen::MatrixXcd::Identity(Nblock,Nblock);
-  Eigen::MatrixXcd m_pAp_inv= Eigen::MatrixXcd::Identity(Nblock,Nblock);
-  Eigen::MatrixXcd m_rr     = Eigen::MatrixXcd::Zero(Nblock,Nblock);
-  Eigen::MatrixXcd m_rr_inv = Eigen::MatrixXcd::Zero(Nblock,Nblock);
-
-  Eigen::MatrixXcd m_alpha      = Eigen::MatrixXcd::Zero(Nblock,Nblock);
-  Eigen::MatrixXcd m_beta   = Eigen::MatrixXcd::Zero(Nblock,Nblock);
-
-  // Initial residual computation & set up
-  std::vector<RealD> residuals(Nblock);
-  std::vector<RealD> ssq(Nblock);
-
-  sliceNorm(ssq,Src,Orthog);
-  RealD sssum=0;
-  for(int b=0;b<Nblock;b++) sssum+=ssq[b];
-
-  sliceNorm(residuals,Src,Orthog);
-  for(int b=0;b<Nblock;b++){ assert(std::isnan(residuals[b])==0); }
-
-  sliceNorm(residuals,Psi,Orthog);
-  for(int b=0;b<Nblock;b++){ assert(std::isnan(residuals[b])==0); }
-
-  // Initial search dir is guess
-  Linop.HermOp(Psi, AP);
-  
-
-  /************************************************************************
-   * Block conjugate gradient (Stephen Pickles, thesis 1995, pp 71, O Leary 1980)
-   ************************************************************************
-   * O'Leary : R = B - A X
-   * O'Leary : P = M R ; preconditioner M = 1
-   * O'Leary : alpha = PAP^{-1} RMR
-   * O'Leary : beta  = RMR^{-1}_old RMR_new
-   * O'Leary : X=X+Palpha
-   * O'Leary : R_new=R_old-AP alpha
-   * O'Leary : P=MR_new+P beta
-   */
-
-  R = Src - AP;  
-  P = R;
-  sliceInnerProductMatrix(m_rr,R,R,Orthog);
-
-  GridStopWatch sliceInnerTimer;
-  GridStopWatch sliceMaddTimer;
-  GridStopWatch MatrixTimer;
-  GridStopWatch SolverTimer;
-  SolverTimer.Start();
-
-  int k;
-  for (k = 1; k <= MaxIterations; k++) {
-
-    RealD rrsum=0;
-    for(int b=0;b<Nblock;b++) rrsum+=real(m_rr(b,b));
-
-    std::cout << GridLogIterative << "\titeration "<<k<<" rr_sum "<<rrsum<<" ssq_sum "<< sssum
-	      <<" / "<<std::sqrt(rrsum/sssum) <<std::endl;
-
-    MatrixTimer.Start();
-    Linop.HermOp(P, AP);
-    MatrixTimer.Stop();
-
-    // Alpha
-    sliceInnerTimer.Start();
-    sliceInnerProductMatrix(m_pAp,P,AP,Orthog);
-    sliceInnerTimer.Stop();
-    m_pAp_inv = m_pAp.inverse();
-    m_alpha   = m_pAp_inv * m_rr ;
-
-    // Psi, R update
-    sliceMaddTimer.Start();
-    sliceMaddMatrix(Psi,m_alpha, P,Psi,Orthog);     // add alpha *  P to psi
-    sliceMaddMatrix(R  ,m_alpha,AP,  R,Orthog,-1.0);// sub alpha * AP to resid
-    sliceMaddTimer.Stop();
-
-    // Beta
-    m_rr_inv = m_rr.inverse();
-    sliceInnerTimer.Start();
-    sliceInnerProductMatrix(m_rr,R,R,Orthog);
-    sliceInnerTimer.Stop();
-    m_beta = m_rr_inv *m_rr;
-
-    // Search update
-    sliceMaddTimer.Start();
-    sliceMaddMatrix(AP,m_beta,P,R,Orthog);
-    sliceMaddTimer.Stop();
-    P= AP;
-
-    /*********************
-     * convergence monitor
-     *********************
-     */
-    RealD max_resid=0;
-    for(int b=0;b<Nblock;b++){
-      RealD rr = real(m_rr(b,b))/ssq[b];
-      if ( rr > max_resid ) max_resid = rr;
-    }
-    
-    if ( max_resid < Tolerance*Tolerance ) { 
-
-      SolverTimer.Stop();
-
-      std::cout << GridLogMessage<<"BlockCG converged in "<<k<<" iterations"<<std::endl;
-      for(int b=0;b<Nblock;b++){
-	std::cout << GridLogMessage<< "\t\tblock "<<b<<" resid "<< std::sqrt(real(m_rr(b,b))/ssq[b])<<std::endl;
-      }
-      std::cout << GridLogMessage<<"\tMax residual is "<<std::sqrt(max_resid)<<std::endl;
-
-      Linop.HermOp(Psi, AP);
-      AP = AP-Src;
-      std::cout << GridLogMessage <<"\tTrue residual is " << std::sqrt(norm2(AP)/norm2(Src)) <<std::endl;
-
-      std::cout << GridLogMessage << "Time Breakdown "<<std::endl;
-      std::cout << GridLogMessage << "\tElapsed    " << SolverTimer.Elapsed()     <<std::endl;
-      std::cout << GridLogMessage << "\tMatrix     " << MatrixTimer.Elapsed()     <<std::endl;
-      std::cout << GridLogMessage << "\tInnerProd  " << sliceInnerTimer.Elapsed() <<std::endl;
-      std::cout << GridLogMessage << "\tMaddMatrix " << sliceMaddTimer.Elapsed()  <<std::endl;
-	    
-      IterationsToComplete = k;
-      return;
-    }
-
-  }
-  std::cout << GridLogMessage << "BlockConjugateGradient did NOT converge" << std::endl;
-
-  if (ErrorOnNoConverge) assert(0);
-  IterationsToComplete = k;
-}
-};
-
-
-//////////////////////////////////////////////////////////////////////////
-// multiRHS conjugate gradient. Dimension zero should be the block direction
-//////////////////////////////////////////////////////////////////////////
-template <class Field>
-class MultiRHSConjugateGradient : public OperatorFunction<Field> {
- public:
-
-  typedef typename Field::scalar_type scomplex;
-
-  const int blockDim = 0;
-
-  int Nblock;
-  bool ErrorOnNoConverge;  // throw an assert when the CG fails to converge.
-                           // Defaults true.
-  RealD Tolerance;
-  Integer MaxIterations;
-  Integer IterationsToComplete; //Number of iterations the CG took to finish. Filled in upon completion
-  
-   MultiRHSConjugateGradient(RealD tol, Integer maxit, bool err_on_no_conv = true)
-    : Tolerance(tol),
-    MaxIterations(maxit),
-    ErrorOnNoConverge(err_on_no_conv){};
-
-void operator()(LinearOperatorBase<Field> &Linop, const Field &Src, Field &Psi) 
-{
-  int Orthog = 0; // First dimension is block dim
-  Nblock = Src._grid->_fdimensions[Orthog];
-
-  std::cout<<GridLogMessage<<"MultiRHS Conjugate Gradient : Orthog "<<Orthog<<" Nblock "<<Nblock<<std::endl;
-
-  Psi.checkerboard = Src.checkerboard;
-  conformable(Psi, Src);
-
-  Field P(Src);
-  Field AP(Src);
-  Field R(Src);
-  
-  std::vector<ComplexD> v_pAp(Nblock);
-  std::vector<RealD> v_rr (Nblock);
-  std::vector<RealD> v_rr_inv(Nblock);
-  std::vector<RealD> v_alpha(Nblock);
-  std::vector<RealD> v_beta(Nblock);
-
-  // Initial residual computation & set up
-  std::vector<RealD> residuals(Nblock);
-  std::vector<RealD> ssq(Nblock);
-
-  sliceNorm(ssq,Src,Orthog);
-  RealD sssum=0;
-  for(int b=0;b<Nblock;b++) sssum+=ssq[b];
-
-  sliceNorm(residuals,Src,Orthog);
-  for(int b=0;b<Nblock;b++){ assert(std::isnan(residuals[b])==0); }
-
-  sliceNorm(residuals,Psi,Orthog);
-  for(int b=0;b<Nblock;b++){ assert(std::isnan(residuals[b])==0); }
-
-  // Initial search dir is guess
-  Linop.HermOp(Psi, AP);
-
-  R = Src - AP;  
-  P = R;
-  sliceNorm(v_rr,R,Orthog);
-
-  GridStopWatch sliceInnerTimer;
-  GridStopWatch sliceMaddTimer;
-  GridStopWatch sliceNormTimer;
-  GridStopWatch MatrixTimer;
-  GridStopWatch SolverTimer;
-
-  SolverTimer.Start();
-  int k;
-  for (k = 1; k <= MaxIterations; k++) {
-
-    RealD rrsum=0;
-    for(int b=0;b<Nblock;b++) rrsum+=real(v_rr[b]);
-
-    std::cout << GridLogIterative << "\titeration "<<k<<" rr_sum "<<rrsum<<" ssq_sum "<< sssum
-	      <<" / "<<std::sqrt(rrsum/sssum) <<std::endl;
-
-    MatrixTimer.Start();
-    Linop.HermOp(P, AP);
-    MatrixTimer.Stop();
-
-    // Alpha
-    //    sliceInnerProductVectorTest(v_pAp_test,P,AP,Orthog);
-    sliceInnerTimer.Start();
-    sliceInnerProductVector(v_pAp,P,AP,Orthog);
-    sliceInnerTimer.Stop();
-    for(int b=0;b<Nblock;b++){
-      //      std::cout << " "<< v_pAp[b]<<" "<< v_pAp_test[b]<<std::endl;
-      v_alpha[b] = v_rr[b]/real(v_pAp[b]);
-    }
-
-    // Psi, R update
-    sliceMaddTimer.Start();
-    sliceMaddVector(Psi,v_alpha, P,Psi,Orthog);     // add alpha *  P to psi
-    sliceMaddVector(R  ,v_alpha,AP,  R,Orthog,-1.0);// sub alpha * AP to resid
-    sliceMaddTimer.Stop();
-
-    // Beta
-    for(int b=0;b<Nblock;b++){
-      v_rr_inv[b] = 1.0/v_rr[b];
-    }
-    sliceNormTimer.Start();
-    sliceNorm(v_rr,R,Orthog);
-    sliceNormTimer.Stop();
-    for(int b=0;b<Nblock;b++){
-      v_beta[b] = v_rr_inv[b] *v_rr[b];
-    }
-
-    // Search update
-    sliceMaddTimer.Start();
-    sliceMaddVector(P,v_beta,P,R,Orthog);
-    sliceMaddTimer.Stop();
-
-    /*********************
-     * convergence monitor
-     *********************
-     */
-    RealD max_resid=0;
-    for(int b=0;b<Nblock;b++){
-      RealD rr = v_rr[b]/ssq[b];
-      if ( rr > max_resid ) max_resid = rr;
-    }
-    
-    if ( max_resid < Tolerance*Tolerance ) { 
-
-      SolverTimer.Stop();
-
-      std::cout << GridLogMessage<<"MultiRHS solver converged in " <<k<<" iterations"<<std::endl;
-      for(int b=0;b<Nblock;b++){
-	std::cout << GridLogMessage<< "\t\tBlock "<<b<<" resid "<< std::sqrt(v_rr[b]/ssq[b])<<std::endl;
-      }
-      std::cout << GridLogMessage<<"\tMax residual is "<<std::sqrt(max_resid)<<std::endl;
-
-      Linop.HermOp(Psi, AP);
-      AP = AP-Src;
-      std::cout <<GridLogMessage << "\tTrue residual is " << std::sqrt(norm2(AP)/norm2(Src)) <<std::endl;
-
-      std::cout << GridLogMessage << "Time Breakdown "<<std::endl;
-      std::cout << GridLogMessage << "\tElapsed    " << SolverTimer.Elapsed()     <<std::endl;
-      std::cout << GridLogMessage << "\tMatrix     " << MatrixTimer.Elapsed()     <<std::endl;
-      std::cout << GridLogMessage << "\tInnerProd  " << sliceInnerTimer.Elapsed() <<std::endl;
-      std::cout << GridLogMessage << "\tNorm       " << sliceNormTimer.Elapsed() <<std::endl;
-      std::cout << GridLogMessage << "\tMaddMatrix " << sliceMaddTimer.Elapsed()  <<std::endl;
-
-
-      IterationsToComplete = k;
-      return;
-    }
-
-  }
-  std::cout << GridLogMessage << "MultiRHSConjugateGradient did NOT converge" << std::endl;
-
-  if (ErrorOnNoConverge) assert(0);
-  IterationsToComplete = k;
-}
-};
-
-
-
-}
-#endif

lib/algorithms/iterative/ConjugateGradient.h

@@ -45,8 +45,6 @@ class ConjugateGradient : public OperatorFunction<Field> {
                            // Defaults true.
   RealD Tolerance;
   Integer MaxIterations;
-  Integer IterationsToComplete; //Number of iterations the CG took to finish. Filled in upon completion
-  
   ConjugateGradient(RealD tol, Integer maxit, bool err_on_no_conv = true)
       : Tolerance(tol),
         MaxIterations(maxit),
@@ -78,12 +76,18 @@ class ConjugateGradient : public OperatorFunction<Field> {
     cp = a;
     ssq = norm2(src);
 
-    std::cout << GridLogIterative << std::setprecision(4) << "ConjugateGradient: guess " << guess << std::endl;
+    std::cout << GridLogIterative << std::setprecision(4)
-    std::cout << GridLogIterative << std::setprecision(4) << "ConjugateGradient:   src " << ssq << std::endl;
+              << "ConjugateGradient: guess " << guess << std::endl;
-    std::cout << GridLogIterative << std::setprecision(4) << "ConjugateGradient:    mp " << d << std::endl;
+    std::cout << GridLogIterative << std::setprecision(4)
-    std::cout << GridLogIterative << std::setprecision(4) << "ConjugateGradient:   mmp " << b << std::endl;
+              << "ConjugateGradient:   src " << ssq << std::endl;
-    std::cout << GridLogIterative << std::setprecision(4) << "ConjugateGradient:  cp,r " << cp << std::endl;
+    std::cout << GridLogIterative << std::setprecision(4)
-    std::cout << GridLogIterative << std::setprecision(4) << "ConjugateGradient:     p " << a << std::endl;
+              << "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;
 
@@ -93,7 +97,8 @@ class ConjugateGradient : public OperatorFunction<Field> {
     }
 
     std::cout << GridLogIterative << std::setprecision(4)
-              << "ConjugateGradient: k=0 residual " << cp << " target " << rsq << std::endl;
+              << "ConjugateGradient: k=0 residual " << cp << " target " << rsq
+              << std::endl;
 
     GridStopWatch LinalgTimer;
     GridStopWatch MatrixTimer;
@@ -123,11 +128,8 @@ class ConjugateGradient : public OperatorFunction<Field> {
       p = p * b + r;
 
       LinalgTimer.Stop();
-
       std::cout << GridLogIterative << "ConjugateGradient: Iteration " << k
                 << " residual " << cp << " target " << rsq << std::endl;
-      std::cout << GridLogDebug << "a = "<< a << " b_pred = "<< b_pred << "  b = "<< b << std::endl;
-      std::cout << GridLogDebug << "qq = "<< qq << " d = "<< d << "  c = "<< c << std::endl;
 
       // Stopping condition
       if (cp <= rsq) {
@@ -135,33 +137,31 @@ class ConjugateGradient : public OperatorFunction<Field> {
         Linop.HermOpAndNorm(psi, mmp, d, qq);
         p = mmp - src;
 
+        RealD mmpnorm = sqrt(norm2(mmp));
+        RealD psinorm = sqrt(norm2(psi));
         RealD srcnorm = sqrt(norm2(src));
         RealD resnorm = sqrt(norm2(p));
         RealD true_residual = resnorm / srcnorm;
 
-        std::cout << GridLogMessage << "ConjugateGradient Converged on iteration " << k << std::endl;
+        std::cout << GridLogMessage
-        std::cout << GridLogMessage << "\tComputed residual " << sqrt(cp / ssq)<<std::endl;
+                  << "ConjugateGradient: Converged on iteration " << k << std::endl;
-	std::cout << GridLogMessage << "\tTrue residual " << true_residual<<std::endl;
+        std::cout << GridLogMessage << "Computed residual " << sqrt(cp / ssq)
-	std::cout << GridLogMessage << "\tTarget " << Tolerance << std::endl;
+                  << " true residual " << true_residual << " target "
-
+                  << Tolerance << std::endl;
-        std::cout << GridLogMessage << "Time breakdown "<<std::endl;
+        std::cout << GridLogMessage << "Time elapsed: Iterations "
-	std::cout << GridLogMessage << "\tElapsed    " << SolverTimer.Elapsed() <<std::endl;
+                  << SolverTimer.Elapsed() << " Matrix  "
-	std::cout << GridLogMessage << "\tMatrix     " << MatrixTimer.Elapsed() <<std::endl;
+                  << MatrixTimer.Elapsed() << " Linalg "
-	std::cout << GridLogMessage << "\tLinalg     " << LinalgTimer.Elapsed() <<std::endl;
+                  << LinalgTimer.Elapsed();
+        std::cout << std::endl;
 
         if (ErrorOnNoConverge) assert(true_residual / Tolerance < 10000.0);
 
-	IterationsToComplete = k;	
-
         return;
       }
     }
     std::cout << GridLogMessage << "ConjugateGradient did NOT converge"
               << std::endl;
-
     if (ErrorOnNoConverge) assert(0);
-    IterationsToComplete = k;
-
   }
 };
 }

lib/algorithms/iterative/ConjugateGradientMixedPrec.h

@@ -35,7 +35,6 @@ namespace Grid {
   class MixedPrecisionConjugateGradient : public LinearFunction<FieldD> {
   public:                                                
     RealD   Tolerance;
-    RealD   InnerTolerance; //Initial tolerance for inner CG. Defaults to Tolerance but can be changed
     Integer MaxInnerIterations;
     Integer MaxOuterIterations;
     GridBase* SinglePrecGrid; //Grid for single-precision fields
@@ -43,16 +42,12 @@ namespace Grid {
     LinearOperatorBase<FieldF> &Linop_f;
     LinearOperatorBase<FieldD> &Linop_d;
 
-    Integer TotalInnerIterations; //Number of inner CG iterations
-    Integer TotalOuterIterations; //Number of restarts
-    Integer TotalFinalStepIterations; //Number of CG iterations in final patch-up step
-
     //Option to speed up *inner single precision* solves using a LinearFunction that produces a guess
     LinearFunction<FieldF> *guesser;
     
     MixedPrecisionConjugateGradient(RealD tol, Integer maxinnerit, Integer maxouterit, GridBase* _sp_grid, LinearOperatorBase<FieldF> &_Linop_f, LinearOperatorBase<FieldD> &_Linop_d) :
       Linop_f(_Linop_f), Linop_d(_Linop_d),
-      Tolerance(tol), InnerTolerance(tol), MaxInnerIterations(maxinnerit), MaxOuterIterations(maxouterit), SinglePrecGrid(_sp_grid),
+      Tolerance(tol), MaxInnerIterations(maxinnerit), MaxOuterIterations(maxouterit), SinglePrecGrid(_sp_grid),
       OuterLoopNormMult(100.), guesser(NULL){ };
 
     void useGuesser(LinearFunction<FieldF> &g){
@@ -60,8 +55,9 @@ namespace Grid {
     }
   
     void operator() (const FieldD &src_d_in, FieldD &sol_d){
-      TotalInnerIterations = 0;
+	(*this)(src_d_in,sol_d,NULL);
-	
+    }
+    void operator() (const FieldD &src_d_in, FieldD &sol_d, RealD *shift){
       GridStopWatch TotalTimer;
       TotalTimer.Start();
     
@@ -81,7 +77,7 @@ namespace Grid {
       FieldD src_d(DoublePrecGrid);
       src_d = src_d_in; //source for next inner iteration, computed from residual during operation
     
-      RealD inner_tol = InnerTolerance;
+      RealD inner_tol = Tolerance;
     
       FieldF src_f(SinglePrecGrid);
       src_f.checkerboard = cb;
@@ -89,18 +85,17 @@ 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;
 
       GridStopWatch PrecChangeTimer;
     
-      Integer &outer_iter = TotalOuterIterations; //so it will be equal to the final iteration count
+      for(Integer outer_iter = 0; outer_iter < MaxOuterIterations; outer_iter++){
-      
-      for(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;
@@ -124,9 +119,8 @@ 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();
-	TotalInnerIterations += CG_f.IterationsToComplete;
       
 	//Convert sol back to double and add to double prec solution
 	PrecChangeTimer.Start();
@@ -139,13 +133,11 @@ namespace Grid {
       //Final trial CG
       std::cout<<GridLogMessage<<"MixedPrecisionConjugateGradient: Starting final patch-up double-precision solve"<<std::endl;
     
-      ConjugateGradient<FieldD> CG_d(Tolerance, MaxInnerIterations);
+      ConjugateGradientShifted<FieldD> CG_d(Tolerance, MaxInnerIterations);
-      CG_d(Linop_d, src_d_in, sol_d);
+      CG_d(Linop_d, src_d_in, sol_d,shift);
-      TotalFinalStepIterations = CG_d.IterationsToComplete;
 
       TotalTimer.Stop();
-      std::cout<<GridLogMessage<<"MixedPrecisionConjugateGradient: Inner CG iterations " << TotalInnerIterations << " Restarts " << TotalOuterIterations << " Final CG iterations " << TotalFinalStepIterations << std::endl;
+      std::cout<<GridLogMessage<<"MixedPrecisionConjugateGradient: Total " << TotalTimer.Elapsed() << " Precision change " << PrecChangeTimer.Elapsed() << " Inner CG total " << InnerCGtimer.Elapsed() << std::endl;
-      std::cout<<GridLogMessage<<"MixedPrecisionConjugateGradient: Total time " << 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

@@ -30,17 +30,20 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 #define GRID_IRL_H
 
 #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 <Grid/algorithms/densematrix/DenseMatrix.h>
+#include "DenseMatrix.h"
-#include <Grid/algorithms/iterative/EigenSort.h>
+#include "EigenSort.h"
 
 namespace Grid {
 
@@ -64,12 +67,13 @@ public:
     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;
@@ -126,23 +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;
       }
     }
 
@@ -176,10 +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;
@@ -255,6 +259,7 @@ public:
     }
 
 #ifdef USE_LAPACK
+#define LAPACK_INT long long
     void diagonalize_lapack(DenseVector<RealD>& lmd,
 		     DenseVector<RealD>& lme, 
 		     int N1,
@@ -264,7 +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);
@@ -278,19 +283,19 @@ public:
         if (i==j) evals_tmp[i] = lmd[i];
         if (j==(i-1)) EE[j] = lme[j];
       }
-  int evals_found;
+  LAPACK_INT evals_found;
-  int lwork = ( (18*NN) > (1+4*NN+NN*NN)? (18*NN):(1+4*NN+NN*NN)) ;
+  LAPACK_INT lwork = ( (18*NN) > (1+4*NN+NN*NN)? (18*NN):(1+4*NN+NN*NN)) ;
-  int liwork =  3+NN*10 ;
+  LAPACK_INT liwork =  3+NN*10 ;
-  int iwork[liwork];
+  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;
@@ -298,14 +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
@@ -337,6 +346,7 @@ public:
       lmd [NN-1-i]=evals_tmp[i];
   }
 }
+#undef LAPACK_INT 
 #endif
 
 
@@ -367,12 +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];
@@ -401,11 +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(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 <<"lme(qr)-lme(lapack) "<< k << ": " << lme2[k] - lme[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
@@ -420,7 +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();
     }
 
@@ -437,6 +449,7 @@ public:
 		       DenseVector<Field>& evec,
 		       int k)
     {
+      double t0=-usecond()/1e6;
       typedef typename Field::scalar_type MyComplex;
       MyComplex ip;
 
@@ -455,6 +468,8 @@ public:
 	w = w - ip * evec[j];
       }
       normalise(w);
+      t0+=usecond()/1e6;
+      OrthoTime +=t0;
     }
 
     void setUnit_Qt(int Nm, DenseVector<RealD> &Qt) {
@@ -488,10 +503,10 @@ until convergence
 	GridBase *grid = evec[0]._grid;
 	assert(grid == src._grid);
 
-	std::cout << " -- Nk = " << Nk << " Np = "<< Np << std::endl;
+	std::cout<<GridLogMessage << " -- Nk = " << Nk << " Np = "<< Np << std::endl;
-	std::cout << " -- Nm = " << Nm << std::endl;
+	std::cout<<GridLogMessage << " -- Nm = " << Nm << std::endl;
-	std::cout << " -- size of eval   = " << eval.size() << std::endl;
+	std::cout<<GridLogMessage << " -- size of eval   = " << eval.size() << std::endl;
-	std::cout << " -- size of evec  = " << evec.size() << std::endl;
+	std::cout<<GridLogMessage << " -- size of evec  = " << evec.size() << std::endl;
 	
 	assert(Nm == evec.size() && Nm == eval.size());
 	
@@ -502,6 +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);
@@ -517,35 +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;
+	double t1=usecond()/1e6;
-//	std:: cout <<"norm2(evec[2])= " << norm2(evec[2]) << std::endl;
+	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<<GridLogMessage <<"eval " << k << " " <<eval[k] << std::endl;
-//	std:: cout <<"lme " << k << " " << lme[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){
@@ -554,18 +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){
@@ -574,14 +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;
@@ -594,7 +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){
@@ -602,12 +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);
@@ -615,14 +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<<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<<"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<<"|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) ){
@@ -631,17 +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:
@@ -654,10 +745,10 @@ until convergence
        }
       _sort.push(eval,evec,Nconv);
 
-      std::cout << "\n Converged\n Summary :\n";
+      std::cout<<GridLogMessage << "\n Converged\n Summary :\n";
-      std::cout << " -- Iterations  = "<< Nconv  << "\n";
+      std::cout<<GridLogMessage << " -- Iterations  = "<< Nconv  << "\n";
-      std::cout << " -- beta(k)     = "<< beta_k << "\n";
+      std::cout<<GridLogMessage << " -- beta(k)     = "<< beta_k << "\n";
-      std::cout << " -- Nconv       = "<< Nconv  << "\n";
+      std::cout<<GridLogMessage << " -- Nconv       = "<< Nconv  << "\n";
      }
 
     /////////////////////////////////////////////////
@@ -680,25 +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;
 
@@ -718,19 +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;
@@ -753,7 +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;
@@ -788,8 +879,8 @@ until convergence
 	  bck = sqrt( nmbex );
 	  re++;
 	}
-	std::cout << "Iteratively refined orthogonality, changes alpha\n";
+	std::cout<<GridLogMessage << "Iteratively refined orthogonality, changes alpha\n";
-	if(re > 1) std::cout << "orthagonality refined " << re << " times" <<std::endl;
+	if(re > 1) std::cout<<GridLogMessage << "orthagonality refined " << re << " times" <<std::endl;
 	H[j][j]=alpha;
       }
 
@@ -804,11 +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;
@@ -831,15 +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<<GridLogMessage << "TM = " << TM << " ";
-      std::cout << norm2(bq[0]) << " -- before" <<std::endl;
+      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
@@ -863,7 +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?
       }
 
@@ -872,7 +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);
@@ -888,7 +981,7 @@ until convergence
       Wilkinson<RealD>(H, evals, evecs, small); 
       //      Check();
 
-      std::cout << "Done  "<<std::endl;
+      std::cout<<GridLogMessage << "Done  "<<std::endl;
 
     }
 
@@ -953,7 +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;
 
@@ -968,7 +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);
 
@@ -979,7 +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) {
 
@@ -995,13 +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
@@ -1010,7 +1105,9 @@ until convergence
 
       DenseVector<RealD> goodval(this->get);
 
+#ifndef USE_LAPACK
       EigenSort(evals,evecs);
+#endif
 
       int NM = Nm;
 
@@ -1082,14 +1179,16 @@ say con = 2
 **/
 
 template<class T>
-static void Lock(DenseMatrix<T> &H, 	// Hess mtx	
+static void Lock(DenseMatrix<T> &H, 	///Hess mtx	
-		 DenseMatrix<T> &Q, 	// Lock Transform
+		 DenseMatrix<T> &Q, 	///Lock Transform
-		 T val, 		// value to be locked
+		 T val, 		///value to be locked
-		 int con, 	// number already locked
+		 int con, 	///number already locked
 		 RealD small,
 		 int dfg,
 		 bool herm)
 {	
+
+
   //ForceTridiagonal(H);
 
   int M = H.dim;
@@ -1121,6 +1220,7 @@ static void Lock(DenseMatrix<T> &H, 	// Hess mtx
 
   AH = Hermitian(QQ)*AH;
   AH = AH*QQ;
+	
 
   for(int i=con;i<M;i++){
     for(int j=con;j<M;j++){

lib/algorithms/iterative/Matrix.h

@@ -0,0 +1,453 @@
+    /*************************************************************************************
+
+    Grid physics library, www.github.com/paboyle/Grid 
+
+    Source file: ./lib/algorithms/iterative/Matrix.h
+
+    Copyright (C) 2015
+
+Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License along
+    with this program; if not, write to the Free Software Foundation, Inc.,
+    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+    See the full license in the file "LICENSE" in the top level distribution directory
+    *************************************************************************************/
+    /*  END LEGAL */
+#ifndef MATRIX_H
+#define MATRIX_H
+
+#include <cstdlib>
+#include <string>
+#include <cmath>
+#include <vector>
+#include <iostream>
+#include <iomanip>
+#include <complex>
+#include <typeinfo>
+#include <Grid.h>
+
+
+/** Sign function **/
+template <class T> T sign(T p){return ( p/abs(p) );}
+
+/////////////////////////////////////////////////////////////////////////////////////////////////////////
+///////////////////// Hijack STL containers for our wicked means /////////////////////////////////////////
+/////////////////////////////////////////////////////////////////////////////////////////////////////////
+template<class T> using Vector = Vector<T>;
+template<class T> using Matrix = Vector<Vector<T> >;
+
+template<class T> void Resize(Vector<T > & vec, int N) { vec.resize(N); }
+
+template<class T> void Resize(Matrix<T > & mat, int N, int M) { 
+  mat.resize(N);
+  for(int i=0;i<N;i++){
+    mat[i].resize(M);
+  }
+}
+template<class T> void Size(Vector<T> & vec, int &N) 
+{ 
+  N= vec.size();
+}
+template<class T> void Size(Matrix<T> & mat, int &N,int &M) 
+{ 
+  N= mat.size();
+  M= mat[0].size();
+}
+template<class T> void SizeSquare(Matrix<T> & mat, int &N) 
+{ 
+  int M; Size(mat,N,M);
+  assert(N==M);
+}
+template<class T> void SizeSame(Matrix<T> & mat1,Matrix<T> &mat2, int &N1,int &M1) 
+{ 
+  int N2,M2;
+  Size(mat1,N1,M1);
+  Size(mat2,N2,M2);
+  assert(N1==N2);
+  assert(M1==M2);
+}
+
+//*****************************************
+//*	(Complex) Vector operations	*
+//*****************************************
+
+/**Conj of a Vector **/
+template <class T> Vector<T> conj(Vector<T> p){
+	Vector<T> q(p.size());
+	for(int i=0;i<p.size();i++){q[i] = conj(p[i]);}
+	return q;
+}
+
+/** Norm of a Vector**/
+template <class T> T norm(Vector<T> p){
+	T sum = 0;
+	for(int i=0;i<p.size();i++){sum = sum + p[i]*conj(p[i]);}
+	return abs(sqrt(sum));
+}
+
+/** Norm squared of a Vector **/
+template <class T> T norm2(Vector<T> p){
+	T sum = 0;
+	for(int i=0;i<p.size();i++){sum = sum + p[i]*conj(p[i]);}
+	return abs((sum));
+}
+
+/** Sum elements of a Vector **/
+template <class T> T trace(Vector<T> p){
+	T sum = 0;
+	for(int i=0;i<p.size();i++){sum = sum + p[i];}
+	return sum;
+}
+
+/** Fill a Vector with constant c **/
+template <class T> void Fill(Vector<T> &p, T c){
+	for(int i=0;i<p.size();i++){p[i] = c;}
+}
+/** Normalize a Vector **/
+template <class T> void normalize(Vector<T> &p){
+	T m = norm(p);
+	if( abs(m) > 0.0) for(int i=0;i<p.size();i++){p[i] /= m;}
+}
+/** Vector by scalar **/
+template <class T, class U> Vector<T> times(Vector<T> p, U s){
+	for(int i=0;i<p.size();i++){p[i] *= s;}
+	return p;
+}
+template <class T, class U> Vector<T> times(U s, Vector<T> p){
+	for(int i=0;i<p.size();i++){p[i] *= s;}
+	return p;
+}
+/** inner product of a and b = conj(a) . b **/
+template <class T> T inner(Vector<T> a, Vector<T> b){
+	T m = 0.;
+	for(int i=0;i<a.size();i++){m = m + conj(a[i])*b[i];}
+	return m;
+}
+/** sum of a and b = a + b **/
+template <class T> Vector<T> add(Vector<T> a, Vector<T> b){
+	Vector<T> m(a.size());
+	for(int i=0;i<a.size();i++){m[i] = a[i] + b[i];}
+	return m;
+}
+/** sum of a and b = a - b **/
+template <class T> Vector<T> sub(Vector<T> a, Vector<T> b){
+	Vector<T> m(a.size());
+	for(int i=0;i<a.size();i++){m[i] = a[i] - b[i];}
+	return m;
+}
+
+/** 
+ *********************************
+ *	Matrices	         *
+ *********************************
+ **/
+
+template<class T> void Fill(Matrix<T> & mat, T&val) { 
+  int N,M;
+  Size(mat,N,M);
+  for(int i=0;i<N;i++){
+  for(int j=0;j<M;j++){
+    mat[i][j] = val;
+  }}
+}
+
+/** Transpose of a matrix **/
+Matrix<T> Transpose(Matrix<T> & mat){
+  int N,M;
+  Size(mat,N,M);
+  Matrix C; Resize(C,M,N);
+  for(int i=0;i<M;i++){
+  for(int j=0;j<N;j++){
+    C[i][j] = mat[j][i];
+  }} 
+  return C;
+}
+/** Set Matrix to unit matrix **/
+template<class T> void Unity(Matrix<T> &mat){
+  int N;  SizeSquare(mat,N);
+  for(int i=0;i<N;i++){
+    for(int j=0;j<N;j++){
+      if ( i==j ) A[i][j] = 1;
+      else        A[i][j] = 0;
+    } 
+  } 
+}
+/** Add C * I to matrix **/
+template<class T>
+void PlusUnit(Matrix<T> & A,T c){
+  int dim;  SizeSquare(A,dim);
+  for(int i=0;i<dim;i++){A[i][i] = A[i][i] + c;} 
+}
+
+/** return the Hermitian conjugate of matrix **/
+Matrix<T> HermitianConj(Matrix<T> &mat){
+
+  int dim; SizeSquare(mat,dim);
+
+  Matrix<T> C; Resize(C,dim,dim);
+
+  for(int i=0;i<dim;i++){
+    for(int j=0;j<dim;j++){
+      C[i][j] = conj(mat[j][i]);
+    } 
+  } 
+  return C;
+}
+
+/** return diagonal entries as a Vector **/
+Vector<T> diag(Matrix<T> &A)
+{
+  int dim; SizeSquare(A,dim);
+  Vector<T> d; Resize(d,dim);
+
+  for(int i=0;i<dim;i++){
+    d[i] = A[i][i];
+  }
+  return d;
+}
+
+/** Left multiply by a Vector **/
+Vector<T> operator *(Vector<T> &B,Matrix<T> &A)
+{
+  int K,M,N; 
+  Size(B,K);
+  Size(A,M,N);
+  assert(K==M);
+  
+  Vector<T> C; Resize(C,N);
+
+  for(int j=0;j<N;j++){
+    T sum = 0.0;
+    for(int i=0;i<M;i++){
+      sum += B[i] * A[i][j];
+    }
+    C[j] =  sum;
+  }
+  return C; 
+}
+
+/** return 1/diagonal entries as a Vector **/
+Vector<T> inv_diag(Matrix<T> & A){
+  int dim; SizeSquare(A,dim);
+  Vector<T> d; Resize(d,dim);
+  for(int i=0;i<dim;i++){
+    d[i] = 1.0/A[i][i];
+  }
+  return d;
+}
+/** Matrix Addition **/
+inline Matrix<T> operator + (Matrix<T> &A,Matrix<T> &B)
+{
+  int N,M  ; SizeSame(A,B,N,M);
+  Matrix C; Resize(C,N,M);
+  for(int i=0;i<N;i++){
+    for(int j=0;j<M;j++){
+      C[i][j] = A[i][j] +  B[i][j];
+    } 
+  } 
+  return C;
+} 
+/** Matrix Subtraction **/
+inline Matrix<T> operator- (Matrix<T> & A,Matrix<T> &B){
+  int N,M  ; SizeSame(A,B,N,M);
+  Matrix C; Resize(C,N,M);
+  for(int i=0;i<N;i++){
+  for(int j=0;j<M;j++){
+    C[i][j] = A[i][j] -  B[i][j];
+  }}
+  return C;
+} 
+
+/** Matrix scalar multiplication **/
+inline Matrix<T> operator* (Matrix<T> & A,T c){
+  int N,M; Size(A,N,M);
+  Matrix C; Resize(C,N,M);
+  for(int i=0;i<N;i++){
+  for(int j=0;j<M;j++){
+    C[i][j] = A[i][j]*c;
+  }} 
+  return C;
+} 
+/** Matrix Matrix multiplication **/
+inline Matrix<T> operator* (Matrix<T> &A,Matrix<T> &B){
+  int K,L,N,M;
+  Size(A,K,L);
+  Size(B,N,M); assert(L==N);
+  Matrix C; Resize(C,K,M);
+
+  for(int i=0;i<K;i++){
+    for(int j=0;j<M;j++){
+      T sum = 0.0;
+      for(int k=0;k<N;k++) sum += A[i][k]*B[k][j];
+      C[i][j] =sum;
+    }
+  }
+  return C; 
+} 
+/** Matrix Vector multiplication **/
+inline Vector<T> operator* (Matrix<T> &A,Vector<T> &B){
+  int M,N,K;
+  Size(A,N,M);
+  Size(B,K); assert(K==M);
+  Vector<T> C; Resize(C,N);
+  for(int i=0;i<N;i++){
+    T sum = 0.0;
+    for(int j=0;j<M;j++) sum += A[i][j]*B[j];
+    C[i] =  sum;
+  }
+  return C; 
+} 
+
+/** Some version of Matrix norm **/
+/*
+inline T Norm(){ // this is not a usual L2 norm
+    T norm = 0;
+    for(int i=0;i<dim;i++){
+      for(int j=0;j<dim;j++){
+	norm += abs(A[i][j]);
+    }}
+    return norm;
+  }
+*/
+
+/** Some version of Matrix norm **/
+template<class T> T LargestDiag(Matrix<T> &A)
+{
+  int dim ; SizeSquare(A,dim); 
+
+  T ld = abs(A[0][0]);
+  for(int i=1;i<dim;i++){
+    T cf = abs(A[i][i]);
+    if(abs(cf) > abs(ld) ){ld = cf;}
+  }
+  return ld;
+}
+
+/** Look for entries on the leading subdiagonal that are smaller than 'small' **/
+template <class T,class U> int Chop_subdiag(Matrix<T> &A,T norm, int offset, U small)
+{
+  int dim; SizeSquare(A,dim);
+  for(int l = dim - 1 - offset; l >= 1; l--) {             		
+    if((U)abs(A[l][l - 1]) < (U)small) {
+      A[l][l-1]=(U)0.0;
+      return l;
+    }
+  }
+  return 0;
+}
+
+/** Look for entries on the leading subdiagonal that are smaller than 'small' **/
+template <class T,class U> int Chop_symm_subdiag(Matrix<T> & A,T norm, int offset, U small) 
+{
+  int dim; SizeSquare(A,dim);
+  for(int l = dim - 1 - offset; l >= 1; l--) {
+    if((U)abs(A[l][l - 1]) < (U)small) {
+      A[l][l - 1] = (U)0.0;
+      A[l - 1][l] = (U)0.0;
+      return l;
+    }
+  }
+  return 0;
+}
+/**Assign a submatrix to a larger one**/
+template<class T>
+void AssignSubMtx(Matrix<T> & A,int row_st, int row_end, int col_st, int col_end, Matrix<T> &S)
+{
+  for(int i = row_st; i<row_end; i++){
+    for(int j = col_st; j<col_end; j++){
+      A[i][j] = S[i - row_st][j - col_st];
+    }
+  }
+}
+
+/**Get a square submatrix**/
+template <class T>
+Matrix<T> GetSubMtx(Matrix<T> &A,int row_st, int row_end, int col_st, int col_end)
+{
+  Matrix<T> H; Resize(row_end - row_st,col_end-col_st);
+
+  for(int i = row_st; i<row_end; i++){
+  for(int j = col_st; j<col_end; j++){
+    H[i-row_st][j-col_st]=A[i][j];
+  }}
+  return H;
+}
+  
+ /**Assign a submatrix to a larger one NB remember Vector Vectors are transposes of the matricies they represent**/
+template<class T>
+void AssignSubMtx(Matrix<T> & A,int row_st, int row_end, int col_st, int col_end, Matrix<T> &S)
+{
+  for(int i = row_st; i<row_end; i++){
+  for(int j = col_st; j<col_end; j++){
+    A[i][j] = S[i - row_st][j - col_st];
+  }}
+}
+  
+/** compute b_i A_ij b_j **/ // surprised no Conj
+template<class T> T proj(Matrix<T> A, Vector<T> B){
+  int dim; SizeSquare(A,dim);
+  int dimB; Size(B,dimB);
+  assert(dimB==dim);
+  T C = 0;
+  for(int i=0;i<dim;i++){
+    T sum = 0.0;
+    for(int j=0;j<dim;j++){
+      sum += A[i][j]*B[j];
+    }
+    C +=  B[i]*sum; // No conj?
+  }
+  return C; 
+}
+
+
+/*
+ *************************************************************
+ *
+ * Matrix Vector products
+ *
+ *************************************************************
+ */
+// Instead make a linop and call my CG;
+
+/// q -> q Q
+template <class T,class Fermion> void times(Vector<Fermion> &q, Matrix<T> &Q)
+{
+  int M; SizeSquare(Q,M);
+  int N; Size(q,N); 
+  assert(M==N);
+
+  times(q,Q,N);
+}
+
+/// q -> q Q
+template <class T> void times(multi1d<LatticeFermion> &q, Matrix<T> &Q, int N)
+{
+  GridBase *grid = q[0]._grid;
+  int M; SizeSquare(Q,M);
+  int K; Size(q,K); 
+  assert(N<M);
+  assert(N<K);
+  Vector<Fermion> S(N,grid );
+  for(int j=0;j<N;j++){
+    S[j] = zero;
+    for(int k=0;k<N;k++){
+      S[j] = S[j] +  q[k]* Q[k][j]; 
+    }
+  }
+  for(int j=0;j<q.size();j++){
+    q[j] = S[j];
+  }
+}
+#endif

lib/algorithms/iterative/MatrixUtils.h

@@ -2,13 +2,11 @@
 
     Grid physics library, www.github.com/paboyle/Grid 
 
-    Source file: ./lib/Grid.h
+    Source file: ./lib/algorithms/iterative/MatrixUtils.h
 
     Copyright (C) 2015
 
 Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: azusayamaguchi <ayamaguc@YAMAKAZE.local>
-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
@@ -27,34 +25,51 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
     See the full license in the file "LICENSE" in the top level distribution directory
     *************************************************************************************/
     /*  END LEGAL */
-//
+#ifndef GRID_MATRIX_UTILS_H
-//  Grid.h
+#define GRID_MATRIX_UTILS_H
-//  simd
-//
-//  Created by Peter Boyle on 09/05/2014.
-//  Copyright (c) 2014 University of Edinburgh. All rights reserved.
-//
 
-#ifndef GRID_BASE_H
+namespace Grid {
-#define GRID_BASE_H
 
-#include <Grid/GridStd.h>
+  namespace MatrixUtils { 
 
-#include <Grid/perfmon/Timer.h>
+    template<class T> inline void Size(Matrix<T>& A,int &N,int &M){
-#include <Grid/perfmon/PerfCount.h>
+      N=A.size(); assert(N>0);
-#include <Grid/log/Log.h>
+      M=A[0].size();
-#include <Grid/allocator/AlignedAllocator.h>
+      for(int i=0;i<N;i++){
-#include <Grid/simd/Simd.h>
+	assert(A[i].size()==M);
-#include <Grid/serialisation/Serialisation.h>
+      }
-#include <Grid/threads/Threads.h>
+    }
-#include <Grid/util/Util.h>
-#include <Grid/communicator/Communicator.h> 
-#include <Grid/cartesian/Cartesian.h>    
-#include <Grid/tensors/Tensors.h>      
-#include <Grid/lattice/Lattice.h>      
-#include <Grid/cshift/Cshift.h>       
-#include <Grid/stencil/Stencil.h>      
-#include <Grid/parallelIO/BinaryIO.h>
-#include <Grid/algorithms/Algorithms.h>   
 
+    template<class T> inline void SizeSquare(Matrix<T>& A,int &N)
+    {
+      int M;
+      Size(A,N,M);
+      assert(N==M);
+    }
+
+    template<class T> inline void Fill(Matrix<T>& A,T & val)
+    { 
+      int N,M;
+      Size(A,N,M);
+      for(int i=0;i<N;i++){
+      for(int j=0;j<M;j++){
+	A[i][j]=val;
+      }}
+    }
+    template<class T> inline void Diagonal(Matrix<T>& A,T & val)
+    { 
+      int N;
+      SizeSquare(A,N);
+      for(int i=0;i<N;i++){
+	A[i][i]=val;
+      }
+    }
+    template<class T> inline void Identity(Matrix<T>& A)
+    {
+      Fill(A,0.0);
+      Diagonal(A,1.0);
+    }
+
+  };
+}
 #endif

lib/algorithms/iterative/TODO

@@ -0,0 +1,15 @@
+- ConjugateGradientMultiShift
+- MCR
+
+- Potentially Useful Boost libraries
+
+- MultiArray
+- Aligned allocator; memory pool
+- Remez -- Mike or Boost?
+- Multiprecision
+- quaternians
+- Tokenize
+- Serialization
+- Regex
+- Proto (ET)
+- uBlas

lib/algorithms/iterative/bisec.c

@@ -0,0 +1,122 @@
+#include <math.h>
+#include <stdlib.h>
+#include <vector>
+
+struct Bisection {
+
+static void get_eig2(int row_num,std::vector<RealD> &ALPHA,std::vector<RealD> &BETA, std::vector<RealD> & eig)
+{
+  int i,j;
+  std::vector<RealD> evec1(row_num+3);
+  std::vector<RealD> evec2(row_num+3);
+  RealD eps2;
+  ALPHA[1]=0.;
+  BETHA[1]=0.;
+  for(i=0;i<row_num-1;i++) {
+    ALPHA[i+1] = A[i*(row_num+1)].real();
+    BETHA[i+2] = A[i*(row_num+1)+1].real();
+  }
+  ALPHA[row_num] = A[(row_num-1)*(row_num+1)].real();
+  bisec(ALPHA,BETHA,row_num,1,row_num,1e-10,1e-10,evec1,eps2);
+  bisec(ALPHA,BETHA,row_num,1,row_num,1e-16,1e-16,evec2,eps2);
+
+  // Do we really need to sort here?
+  int begin=1;
+  int end = row_num;
+  int swapped=1;
+  while(swapped) {
+    swapped=0;
+    for(i=begin;i<end;i++){
+      if(mag(evec2[i])>mag(evec2[i+1]))	{
+	swap(evec2+i,evec2+i+1);
+	swapped=1;
+      }
+    }
+    end--;
+    for(i=end-1;i>=begin;i--){
+      if(mag(evec2[i])>mag(evec2[i+1]))	{
+	swap(evec2+i,evec2+i+1);
+	swapped=1;
+      }
+    }
+    begin++;
+  }
+
+  for(i=0;i<row_num;i++){
+    for(j=0;j<row_num;j++) {
+      if(i==j) H[i*row_num+j]=evec2[i+1];
+      else H[i*row_num+j]=0.;
+    }
+  }
+}
+
+static void bisec(std::vector<RealD> &c,   
+		  std::vector<RealD> &b,
+		  int n,
+		  int m1,
+		  int m2,
+		  RealD eps1,
+		  RealD relfeh,
+		  std::vector<RealD> &x,
+		  RealD &eps2)
+{
+  std::vector<RealD> wu(n+2);
+
+  RealD h,q,x1,xu,x0,xmin,xmax; 
+  int i,a,k;
+
+  b[1]=0.0;
+  xmin=c[n]-fabs(b[n]);
+  xmax=c[n]+fabs(b[n]);
+  for(i=1;i<n;i++){
+    h=fabs(b[i])+fabs(b[i+1]);
+    if(c[i]+h>xmax) xmax= c[i]+h;
+    if(c[i]-h<xmin) xmin= c[i]-h;
+  }
+  xmax *=2.;
+
+  eps2=relfeh*((xmin+xmax)>0.0 ? xmax : -xmin);
+  if(eps1<=0.0) eps1=eps2;
+  eps2=0.5*eps1+7.0*(eps2);
+  x0=xmax;
+  for(i=m1;i<=m2;i++){
+    x[i]=xmax;
+    wu[i]=xmin;
+  }
+
+  for(k=m2;k>=m1;k--){
+    xu=xmin;
+    i=k;
+    do{
+      if(xu<wu[i]){
+	xu=wu[i];
+	i=m1-1;
+      }
+      i--;
+    }while(i>=m1);
+    if(x0>x[k]) x0=x[k];
+    while((x0-xu)>2*relfeh*(fabs(xu)+fabs(x0))+eps1){
+      x1=(xu+x0)/2;
+
+      a=0;
+      q=1.0;
+      for(i=1;i<=n;i++){
+	q=c[i]-x1-((q!=0.0)? b[i]*b[i]/q:fabs(b[i])/relfeh);
+	if(q<0) a++;
+      }
+      //			printf("x1=%e a=%d\n",x1,a);
+      if(a<k){
+	if(a<m1){
+	  xu=x1;
+	  wu[m1]=x1;
+	}else {
+	  xu=x1;
+	  wu[a+1]=x1;
+	  if(x[a]>x1) x[a]=x1;
+	}
+      }else x0=x1;
+    }
+    x[k]=(x0+xu)/2;
+  }
+}
+}

lib/algorithms/iterative/get_eig.c

@@ -0,0 +1 @@
+

lib/cartesian/Cartesian_base.h

@@ -6,9 +6,8 @@
 
     Copyright (C) 2015
 
-    Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-    Author: paboyle <paboyle@ph.ed.ac.uk>
+Author: paboyle <paboyle@ph.ed.ac.uk>
-    Author: Guido Cossu <guido.cossu@ed.ac.uk>
 
     This program is free software; you can redistribute it and/or modify
     it under the terms of the GNU General Public License as published by
@@ -53,7 +52,7 @@ public:
 
     // Physics Grid information.
     std::vector<int> _simd_layout;// Which dimensions get relayed out over simd lanes.
-    std::vector<int> _fdimensions;// (full) Global dimensions of array prior to cb removal
+    std::vector<int> _fdimensions;// Global dimensions of array prior to cb removal
     std::vector<int> _gdimensions;// Global dimensions of array after cb removal
     std::vector<int> _ldimensions;// local dimensions of array with processor images removed
     std::vector<int> _rdimensions;// Reduced local dimensions with simd lane images and processor images removed 
@@ -100,7 +99,7 @@ public:
     virtual int oIndex(std::vector<int> &coor)
     {
         int idx=0;
-        // Works with either global or local coordinates
+	// Works with either global or local coordinates
         for(int d=0;d<_ndimension;d++) idx+=_ostride[d]*(coor[d]%_rdimensions[d]);
         return idx;
     }
@@ -122,11 +121,6 @@ public:
       Lexicographic::CoorFromIndex(coor,Oindex,_rdimensions);
     }
 
-    inline void InOutCoorToLocalCoor (std::vector<int> &ocoor, std::vector<int> &icoor, std::vector<int> &lcoor) {
-      lcoor.resize(_ndimension);
-      for (int d = 0; d < _ndimension; d++)
-        lcoor[d] = ocoor[d] + _rdimensions[d] * icoor[d];
-    }
 
     //////////////////////////////////////////////////////////
     // SIMD lane addressing
@@ -135,7 +129,6 @@ public:
     {
       Lexicographic::CoorFromIndex(coor,lane,_simd_layout);
     }
-
     inline int PermuteDim(int dimension){
       return _simd_layout[dimension]>1;
     }
@@ -153,15 +146,15 @@ public:
       // Distance should be either 0,1,2..
       //
       if ( _simd_layout[dimension] > 2 ) { 
-        for(int d=0;d<_ndimension;d++){
+	for(int d=0;d<_ndimension;d++){
-          if ( d != dimension ) assert ( (_simd_layout[d]==1)  );
+	  if ( d != dimension ) assert ( (_simd_layout[d]==1)  );
-        }
+	}
-        permute_type = RotateBit; // How to specify distance; this is not just direction.
+	permute_type = RotateBit; // How to specify distance; this is not just direction.
-        return permute_type;
+	return permute_type;
       }
 
       for(int d=_ndimension-1;d>dimension;d--){
-        if (_simd_layout[d]>1 ) permute_type++;
+	if (_simd_layout[d]>1 ) permute_type++;
       }
       return permute_type;
     }
@@ -181,44 +174,21 @@ public:
     inline const std::vector<int> &LocalDimensions(void)        { return _ldimensions;};
     inline const std::vector<int> &VirtualLocalDimensions(void) { return _ldimensions;};
 
-    ////////////////////////////////////////////////////////////////
-    // Utility to print the full decomposition details 
-    ////////////////////////////////////////////////////////////////
-
-    void show_decomposition(){
-      std::cout << GridLogMessage << "Full Dimensions    : " << _fdimensions << std::endl;
-      std::cout << GridLogMessage << "Global Dimensions  : " << _gdimensions << std::endl;
-      std::cout << GridLogMessage << "Local Dimensions   : " << _ldimensions << std::endl;
-      std::cout << GridLogMessage << "Reduced Dimensions : " << _rdimensions << std::endl;
-      std::cout << GridLogMessage << "Outer strides      : " << _ostride << std::endl;
-      std::cout << GridLogMessage << "Inner strides      : " << _istride << std::endl;
-      std::cout << GridLogMessage << "iSites             : " << _isites << std::endl;
-      std::cout << GridLogMessage << "oSites             : " << _osites << std::endl;
-      std::cout << GridLogMessage << "lSites             : " << lSites() << std::endl;        
-      std::cout << GridLogMessage << "gSites             : " << gSites() << std::endl;
-      std::cout << GridLogMessage << "Nd                 : " << _ndimension << std::endl;             
-    } 
-
     ////////////////////////////////////////////////////////////////
     // Global addressing
     ////////////////////////////////////////////////////////////////
     void GlobalIndexToGlobalCoor(int gidx,std::vector<int> &gcoor){
-      assert(gidx< gSites());
       Lexicographic::CoorFromIndex(gcoor,gidx,_gdimensions);
     }
     void LocalIndexToLocalCoor(int lidx,std::vector<int> &lcoor){
-      assert(lidx<lSites());
       Lexicographic::CoorFromIndex(lcoor,lidx,_ldimensions);
     }
-
-
-
     void GlobalCoorToGlobalIndex(const std::vector<int> & gcoor,int & gidx){
       gidx=0;
       int mult=1;
       for(int mu=0;mu<_ndimension;mu++) {
-        gidx+=mult*gcoor[mu];
+	gidx+=mult*gcoor[mu];
-        mult*=_gdimensions[mu];
+	mult*=_gdimensions[mu];
       }
     }
     void GlobalCoorToProcessorCoorLocalCoor(std::vector<int> &pcoor,std::vector<int> &lcoor,const std::vector<int> &gcoor)
@@ -226,9 +196,9 @@ public:
       pcoor.resize(_ndimension);
       lcoor.resize(_ndimension);
       for(int mu=0;mu<_ndimension;mu++){
-        int _fld  = _fdimensions[mu]/_processors[mu];
+	int _fld  = _fdimensions[mu]/_processors[mu];
-        pcoor[mu] = gcoor[mu]/_fld;
+	pcoor[mu] = gcoor[mu]/_fld;
-        lcoor[mu] = gcoor[mu]%_fld;
+	lcoor[mu] = gcoor[mu]%_fld;
       }
     }
     void GlobalCoorToRankIndex(int &rank, int &o_idx, int &i_idx ,const std::vector<int> &gcoor)
@@ -237,16 +207,16 @@ public:
       std::vector<int> lcoor;
       GlobalCoorToProcessorCoorLocalCoor(pcoor,lcoor,gcoor);
       rank = RankFromProcessorCoor(pcoor);
-      /*
+
       std::vector<int> cblcoor(lcoor);
       for(int d=0;d<cblcoor.size();d++){
-        if( this->CheckerBoarded(d) ) {
+	if( this->CheckerBoarded(d) ) {
-          cblcoor[d] = lcoor[d]/2;
+	  cblcoor[d] = lcoor[d]/2;
-        }
+	}
       }
-      */
+
-      i_idx= iIndex(lcoor);
+      i_idx= iIndex(cblcoor);// this does not imply divide by 2 on checker dim
-      o_idx= oIndex(lcoor);
+      o_idx= oIndex(lcoor);  // this implies divide by 2 on checkerdim
     }
 
     void RankIndexToGlobalCoor(int rank, int o_idx, int i_idx , std::vector<int> &gcoor)
@@ -268,7 +238,7 @@ public:
     {
       RankIndexToGlobalCoor(rank,o_idx,i_idx ,fcoor);
       if(CheckerBoarded(0)){
-        fcoor[0] = fcoor[0]*2+cb;
+	fcoor[0] = fcoor[0]*2+cb;
       }
     }
     void ProcessorCoorLocalCoorToGlobalCoor(std::vector<int> &Pcoor,std::vector<int> &Lcoor,std::vector<int> &gcoor)

lib/communicator/Communicator_base.cc

@@ -25,8 +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/GridCore.h>
+#include <Grid/Grid.h>
-
 namespace Grid {
 
 ///////////////////////////////////////////////////////////////
@@ -34,7 +33,6 @@ namespace Grid {
 ///////////////////////////////////////////////////////////////
 void *              CartesianCommunicator::ShmCommBuf;
 uint64_t            CartesianCommunicator::MAX_MPI_SHM_BYTES   = 128*1024*1024; 
-CartesianCommunicator::CommunicatorPolicy_t  CartesianCommunicator::CommunicatorPolicy= CartesianCommunicator::CommunicatorPolicyConcurrent;
 
 /////////////////////////////////
 // Alloc, free shmem region
@@ -90,9 +88,7 @@ void CartesianCommunicator::GlobalSumVector(ComplexD *c,int N)
 
 #if !defined( GRID_COMMS_MPI3) && !defined (GRID_COMMS_MPI3L)
 
-int                      CartesianCommunicator::NodeCount(void)    { return ProcessorCount();};
+void CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsRequest_t> &list,
-
-double CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsRequest_t> &list,
 						       void *xmit,
 						       int xmit_to_rank,
 						       void *recv,
@@ -100,7 +96,6 @@ double CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsReques
 						       int bytes)
 {
   SendToRecvFromBegin(list,xmit,xmit_to_rank,recv,recv_from_rank,bytes);
-  return 2.0*bytes;
 }
 void CartesianCommunicator::StencilSendToRecvFromComplete(std::vector<CommsRequest_t> &waitall)
 {

lib/communicator/Communicator_base.h

@@ -116,12 +116,6 @@ class CartesianCommunicator {
   // Implemented in Communicator_base.C
   /////////////////////////////////
   static void * ShmCommBuf;
-
-  // Isend/Irecv/Wait, or Sendrecv blocking
-  enum CommunicatorPolicy_t { CommunicatorPolicyConcurrent, CommunicatorPolicySequential };
-  static CommunicatorPolicy_t CommunicatorPolicy;
-  static void SetCommunicatorPolicy(CommunicatorPolicy_t policy ) { CommunicatorPolicy = policy; }
-
   size_t heap_top;
   size_t heap_bytes;
 
@@ -154,7 +148,6 @@ class CartesianCommunicator {
   const std::vector<int> & ThisProcessorCoor(void) ;
   const std::vector<int> & ProcessorGrid(void)     ;
   int                      ProcessorCount(void)    ;
-  int                      NodeCount(void)    ;
 
   ////////////////////////////////////////////////////////////////////////////////
   // very VERY rarely (Log, serial RNG) we need world without a grid
@@ -207,7 +200,7 @@ class CartesianCommunicator {
   
   void SendToRecvFromComplete(std::vector<CommsRequest_t> &waitall);
 
-  double StencilSendToRecvFromBegin(std::vector<CommsRequest_t> &list,
+  void StencilSendToRecvFromBegin(std::vector<CommsRequest_t> &list,
 				  void *xmit,
 				  int xmit_to_rank,
 				  void *recv,

lib/communicator/Communicator_mpi.cc

@@ -25,9 +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/GridCore.h>
+#include <Grid/Grid.h>
-#include <Grid/GridQCDcore.h>
-#include <Grid/qcd/action/ActionCore.h>
 #include <mpi.h>
 
 namespace Grid {
@@ -41,13 +39,9 @@ MPI_Comm CartesianCommunicator::communicator_world;
 // Should error check all MPI calls.
 void CartesianCommunicator::Init(int *argc, char ***argv) {
   int flag;
-  int provided;
   MPI_Initialized(&flag); // needed to coexist with other libs apparently
   if ( !flag ) {
-    MPI_Init_thread(argc,argv,MPI_THREAD_MULTIPLE,&provided);
+    MPI_Init(argc,argv);
-    if ( provided != MPI_THREAD_MULTIPLE ) {
-      QCD::WilsonKernelsStatic::Comms = QCD::WilsonKernelsStatic::CommsThenCompute;
-    }
   }
   MPI_Comm_dup (MPI_COMM_WORLD,&communicator_world);
   ShmInitGeneric();
@@ -158,34 +152,24 @@ void CartesianCommunicator::SendToRecvFromBegin(std::vector<CommsRequest_t> &lis
 						int from,
 						int bytes)
 {
-  int myrank = _processor;
+  MPI_Request xrq;
+  MPI_Request rrq;
+  int rank = _processor;
   int ierr;
-  if ( CommunicatorPolicy == CommunicatorPolicyConcurrent ) { 
+  ierr =MPI_Isend(xmit, bytes, MPI_CHAR,dest,_processor,communicator,&xrq);
-    MPI_Request xrq;
+  ierr|=MPI_Irecv(recv, bytes, MPI_CHAR,from,from,communicator,&rrq);
-    MPI_Request rrq;
+  
+  assert(ierr==0);
 
-    ierr =MPI_Irecv(recv, bytes, MPI_CHAR,from,from,communicator,&rrq);
+  list.push_back(xrq);
-    ierr|=MPI_Isend(xmit, bytes, MPI_CHAR,dest,_processor,communicator,&xrq);
+  list.push_back(rrq);
-    
-    assert(ierr==0);
-    list.push_back(xrq);
-    list.push_back(rrq);
-  } else { 
-    // Give the CPU to MPI immediately; can use threads to overlap optionally
-    ierr=MPI_Sendrecv(xmit,bytes,MPI_CHAR,dest,myrank,
-		      recv,bytes,MPI_CHAR,from, from,
-		      communicator,MPI_STATUS_IGNORE);
-    assert(ierr==0);
-  }
 }
 void CartesianCommunicator::SendToRecvFromComplete(std::vector<CommsRequest_t> &list)
 {
-  if ( CommunicatorPolicy == CommunicatorPolicyConcurrent ) { 
+  int nreq=list.size();
-    int nreq=list.size();
+  std::vector<MPI_Status> status(nreq);
-    std::vector<MPI_Status> status(nreq);
+  int ierr = MPI_Waitall(nreq,&list[0],&status[0]);
-    int ierr = MPI_Waitall(nreq,&list[0],&status[0]);
+  assert(ierr==0);
-    assert(ierr==0);
-  }
 }
 
 void CartesianCommunicator::Barrier(void)

lib/communicator/Communicator_mpi3.cc

@@ -1,4 +1,4 @@
-/*************************************************************************************
+    /*************************************************************************************
 
     Grid physics library, www.github.com/paboyle/Grid 
 
@@ -25,23 +25,9 @@ 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/GridCore.h>
+#include <Grid/Grid.h>
-
 #include <mpi.h>
 
-#include <semaphore.h>
-#include <fcntl.h>
-#include <unistd.h>
-#include <limits.h>
-#include <sys/types.h>
-#include <sys/ipc.h>
-#include <sys/shm.h>
-#include <sys/mman.h>
-//#include <zlib.h>
-#ifndef SHM_HUGETLB
-#define SHM_HUGETLB 04000
-#endif
-
 namespace Grid {
 
 ///////////////////////////////////////////////////////////////////////////////////////////////////
@@ -64,10 +50,6 @@ std::vector<int> CartesianCommunicator::GroupRanks;
 std::vector<int> CartesianCommunicator::MyGroup;
 std::vector<void *> CartesianCommunicator::ShmCommBufs;
 
-int CartesianCommunicator::NodeCount(void)    { return GroupSize;};
-
-
-#undef FORCE_COMMS
 void *CartesianCommunicator::ShmBufferSelf(void)
 {
   return ShmCommBufs[ShmRank];
@@ -75,9 +57,6 @@ void *CartesianCommunicator::ShmBufferSelf(void)
 void *CartesianCommunicator::ShmBuffer(int rank)
 {
   int gpeer = GroupRanks[rank];
-#ifdef FORCE_COMMS
-  return NULL;
-#endif
   if (gpeer == MPI_UNDEFINED){
     return NULL;
   } else { 
@@ -86,13 +65,7 @@ void *CartesianCommunicator::ShmBuffer(int rank)
 }
 void *CartesianCommunicator::ShmBufferTranslate(int rank,void * local_p)
 {
-  static int count =0;
   int gpeer = GroupRanks[rank];
-  assert(gpeer!=ShmRank); // never send to self
-  assert(rank!=WorldRank);// never send to self
-#ifdef FORCE_COMMS
-  return NULL;
-#endif
   if (gpeer == MPI_UNDEFINED){
     return NULL;
   } else { 
@@ -103,27 +76,16 @@ void *CartesianCommunicator::ShmBufferTranslate(int rank,void * local_p)
 }
 
 void CartesianCommunicator::Init(int *argc, char ***argv) {
-
   int flag;
-  int provided;
-  //  mtrace();
-
   MPI_Initialized(&flag); // needed to coexist with other libs apparently
   if ( !flag ) {
-    MPI_Init_thread(argc,argv,MPI_THREAD_MULTIPLE,&provided);
+    MPI_Init(argc,argv);
-    assert (provided == MPI_THREAD_MULTIPLE);
   }
 
-  Grid_quiesce_nodes();
-
   MPI_Comm_dup (MPI_COMM_WORLD,&communicator_world);
   MPI_Comm_rank(communicator_world,&WorldRank);
   MPI_Comm_size(communicator_world,&WorldSize);
 
-  if ( WorldRank == 0 ) {
-    std::cout << GridLogMessage<< "Initialising MPI "<< WorldRank <<"/"<<WorldSize <<std::endl;
-  }
-
   /////////////////////////////////////////////////////////////////////
   // Split into groups that can share memory
   /////////////////////////////////////////////////////////////////////
@@ -169,6 +131,7 @@ void CartesianCommunicator::Init(int *argc, char ***argv) {
   ///////////////////////////////////////////////////////////////////
   int ierr=MPI_Allreduce(MPI_IN_PLACE,&leaders_1hot[0],WorldSize,MPI_INT,MPI_SUM,communicator_world);
   assert(ierr==0);
+  
   ///////////////////////////////////////////////////////////////////
   // find the group leaders world rank
   ///////////////////////////////////////////////////////////////////
@@ -178,6 +141,7 @@ void CartesianCommunicator::Init(int *argc, char ***argv) {
       leaders_group[group++] = l;
     }
   }
+  
   ///////////////////////////////////////////////////////////////////
   // Identify the rank of the group in which I (and my leader) live
   ///////////////////////////////////////////////////////////////////
@@ -188,113 +152,38 @@ void CartesianCommunicator::Init(int *argc, char ***argv) {
     }
   }
   assert(GroupRank!=-1);
+  
   //////////////////////////////////////////////////////////////////////////////////////////////////////////
   // allocate the shared window for our group
   //////////////////////////////////////////////////////////////////////////////////////////////////////////
-  MPI_Barrier(ShmComm);
+  
-
   ShmCommBuf = 0;
-  ShmCommBufs.resize(ShmSize);
+  ierr = MPI_Win_allocate_shared(MAX_MPI_SHM_BYTES,1,MPI_INFO_NULL,ShmComm,&ShmCommBuf,&ShmWindow);
-
+  assert(ierr==0);
-#if 1
+  // KNL hack -- force to numa-domain 1 in flat
-  char shm_name [NAME_MAX];
+#if 0
-  if ( ShmRank == 0 ) {
+  //#include <numaif.h>
-    for(int r=0;r<ShmSize;r++){
+  for(uint64_t page=0;page<MAX_MPI_SHM_BYTES;page+=4096){
-
+    void *pages = (void *) ( page + ShmCommBuf );
-      size_t size = CartesianCommunicator::MAX_MPI_SHM_BYTES;
+    int status;
-
+    int flags=MPOL_MF_MOVE_ALL;
-      sprintf(shm_name,"/Grid_mpi3_shm_%d_%d",GroupRank,r);
+    int nodes=1; // numa domain == MCDRAM
-
+    unsigned long count=1;
-      shm_unlink(shm_name);
+    ierr= move_pages(0,count, &pages,&nodes,&status,flags);
-      int fd=shm_open(shm_name,O_RDWR|O_CREAT,0666);
+    if (ierr && (page==0)) perror("numa relocate command failed");
-      if ( fd < 0 ) {	perror("failed shm_open");	assert(0);      }
-      ftruncate(fd, size);
-
-      void * ptr =  mmap(NULL,size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
-      if ( ptr == MAP_FAILED ) {       perror("failed mmap");      assert(0);    }
-      assert(((uint64_t)ptr&0x3F)==0);
-      ShmCommBufs[r] =ptr;
-      
-    }
   }
-
-  MPI_Barrier(ShmComm);
-
-  if ( ShmRank != 0 ) { 
-    for(int r=0;r<ShmSize;r++){
-      size_t size = CartesianCommunicator::MAX_MPI_SHM_BYTES ;
-    
-      sprintf(shm_name,"/Grid_mpi3_shm_%d_%d",GroupRank,r);
-
-      int fd=shm_open(shm_name,O_RDWR,0666);
-      if ( fd<0 ) {	perror("failed shm_open");	assert(0);      }
-
-      void * ptr =  mmap(NULL,size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
-      if ( ptr == MAP_FAILED ) {       perror("failed mmap");      assert(0);    }
-      assert(((uint64_t)ptr&0x3F)==0);
-      ShmCommBufs[r] =ptr;
-    }
-  }
-
-#else
-  std::vector<int> shmids(ShmSize);
-
-  if ( ShmRank == 0 ) {
-    for(int r=0;r<ShmSize;r++){
-      size_t size = CartesianCommunicator::MAX_MPI_SHM_BYTES;
-      key_t key   = 0x4545 + r;
-      if ((shmids[r]= shmget(key,size, SHM_HUGETLB | IPC_CREAT | SHM_R | SHM_W)) < 0) {
-	int errsv = errno;
-	printf("Errno %d\n",errsv);
-	perror("shmget");
-	exit(1);
-      }
-      printf("shmid: 0x%x\n", shmids[r]);
-    }
-  }
-  MPI_Barrier(ShmComm);
-  MPI_Bcast(&shmids[0],ShmSize*sizeof(int),MPI_BYTE,0,ShmComm);
-  MPI_Barrier(ShmComm);
-
-  for(int r=0;r<ShmSize;r++){
-    ShmCommBufs[r] = (uint64_t *)shmat(shmids[r], NULL,0);
-    if (ShmCommBufs[r] == (uint64_t *)-1) {
-      perror("Shared memory attach failure");
-      shmctl(shmids[r], IPC_RMID, NULL);
-      exit(2);
-    }
-    printf("shmaddr: %p\n", ShmCommBufs[r]);
-  }
-  MPI_Barrier(ShmComm);
-  // Mark for clean up
-  for(int r=0;r<ShmSize;r++){
-    shmctl(shmids[r], IPC_RMID,(struct shmid_ds *)NULL);
-  }
-  MPI_Barrier(ShmComm);
-
 #endif
-  ShmCommBuf         = ShmCommBufs[ShmRank];
+  MPI_Win_lock_all (MPI_MODE_NOCHECK, ShmWindow);
-
+  
-  MPI_Barrier(ShmComm);
+  /////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-  if ( ShmRank == 0 ) {
+  // Plan: allocate a fixed SHM region. Scratch that is just used via some scheme during stencil comms, with no allocate free.
-    for(int r=0;r<ShmSize;r++){
+  /////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-      uint64_t * check = (uint64_t *) ShmCommBufs[r];
+  ShmCommBufs.resize(ShmSize);
-      check[0] = GroupRank;
-      check[1] = r;
-      check[2] = 0x5A5A5A;
-    }
-  }
-
-  MPI_Barrier(ShmComm);
   for(int r=0;r<ShmSize;r++){
-    uint64_t * check = (uint64_t *) ShmCommBufs[r];
+    MPI_Aint sz;
-    
+    int dsp_unit;
-    assert(check[0]==GroupRank);
+    MPI_Win_shared_query (ShmWindow, r, &sz, &dsp_unit, &ShmCommBufs[r]);
-    assert(check[1]==r);
-    assert(check[2]==0x5A5A5A);
-
   }
-  MPI_Barrier(ShmComm);
 
   //////////////////////////////////////////////////////////////////////////////////////////////////////////
   // Verbose for now
@@ -303,7 +192,7 @@ void CartesianCommunicator::Init(int *argc, char ***argv) {
     std::cout<<GridLogMessage<< "Grid MPI-3 configuration: detected ";
     std::cout<< WorldSize << " Ranks " ;
     std::cout<< GroupSize << " Nodes " ;
-    std::cout<< " with "<< ShmSize  << " ranks-per-node "<<std::endl;
+    std::cout<<  ShmSize  << " with ranks-per-node "<<std::endl;
     
     std::cout<<GridLogMessage     <<"Grid MPI-3 configuration: allocated shared memory region of size ";
     std::cout<<std::hex << MAX_MPI_SHM_BYTES <<" ShmCommBuf address = "<<ShmCommBuf << std::dec<<std::endl;
@@ -318,6 +207,7 @@ void CartesianCommunicator::Init(int *argc, char ***argv) {
       if(g!=ShmSize-1) std::cout<<",";
       else std::cout<<"}"<<std::endl;
     }
+
   }
   
   for(int g=0;g<GroupSize;g++){
@@ -326,21 +216,23 @@ void CartesianCommunicator::Init(int *argc, char ***argv) {
       if ( (ShmRank == 0) && (GroupRank==g) ) {
 	std::cout<<MyGroup[r];
 	if(r<ShmSize-1) std::cout<<",";
-	else std::cout<<"}"<<std::endl<<std::flush;
+	else std::cout<<"}"<<std::endl;
       }
       MPI_Barrier(communicator_world);
     }
   }
-
+  
   assert(ShmSetup==0);  ShmSetup=1;
 }
 
+
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////
 // Want to implement some magic ... Group sub-cubes into those on same node
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////
-void CartesianCommunicator::ShiftedRanks(int dim,int shift,int &dest,int &source)
+void CartesianCommunicator::ShiftedRanks(int dim,int shift,int &source,int &dest)
 {
-  std::vector<int> coor = _processor_coor; // my coord
+  std::vector<int> coor = _processor_coor;
+
   assert(std::abs(shift) <_processors[dim]);
 
   coor[dim] = (_processor_coor[dim] + shift + _processors[dim])%_processors[dim];
@@ -350,32 +242,28 @@ void CartesianCommunicator::ShiftedRanks(int dim,int shift,int &dest,int &source
   coor[dim] = (_processor_coor[dim] - shift + _processors[dim])%_processors[dim];
   Lexicographic::IndexFromCoor(coor,dest,_processors);
   dest = LexicographicToWorldRank[dest];
-
+}
-}// rank is world rank.
-
 int CartesianCommunicator::RankFromProcessorCoor(std::vector<int> &coor)
 {
   int rank;
   Lexicographic::IndexFromCoor(coor,rank,_processors);
   rank = LexicographicToWorldRank[rank];
   return rank;
-}// rank is world rank
+}
-
 void  CartesianCommunicator::ProcessorCoorFromRank(int rank, std::vector<int> &coor)
 {
-  int lr=-1;
+  Lexicographic::CoorFromIndex(coor,rank,_processors);
-  for(int r=0;r<WorldSize;r++){// map world Rank to lexico and then to coor
+  rank = LexicographicToWorldRank[rank];
-    if( LexicographicToWorldRank[r]==rank) lr = r;
-  }
-  assert(lr!=-1);
-  Lexicographic::CoorFromIndex(coor,lr,_processors);
 }
+
 CartesianCommunicator::CartesianCommunicator(const std::vector<int> &processors)
 { 
   int ierr;
-  communicator=communicator_world;
-  _ndimension = processors.size();
 
+  communicator=communicator_world;
+
+  _ndimension = processors.size();
+  
   ////////////////////////////////////////////////////////////////
   // Assert power of two shm_size.
   ////////////////////////////////////////////////////////////////
@@ -387,22 +275,24 @@ CartesianCommunicator::CartesianCommunicator(const std::vector<int> &processors)
     }
   }
   assert(log2size != -1);
-
+  
   ////////////////////////////////////////////////////////////////
   // Identify subblock of ranks on node spreading across dims
   // in a maximally symmetrical way
   ////////////////////////////////////////////////////////////////
+  int dim = 0;
+  
   std::vector<int> WorldDims = processors;
 
-  ShmDims.resize  (_ndimension,1);
+  ShmDims.resize(_ndimension,1);
   GroupDims.resize(_ndimension);
-  ShmCoor.resize  (_ndimension);
+    
+  ShmCoor.resize(_ndimension);
   GroupCoor.resize(_ndimension);
   WorldCoor.resize(_ndimension);
 
-  int dim = 0;
   for(int l2=0;l2<log2size;l2++){
-    while ( (WorldDims[dim] / ShmDims[dim]) <= 1 ) dim=(dim+1)%_ndimension;
+    while ( WorldDims[dim] / ShmDims[dim] <= 1 ) dim=(dim+1)%_ndimension;
     ShmDims[dim]*=2;
     dim=(dim+1)%_ndimension;
   }
@@ -414,29 +304,6 @@ CartesianCommunicator::CartesianCommunicator(const std::vector<int> &processors)
     GroupDims[d] = WorldDims[d]/ShmDims[d];
   }
 
-  ////////////////////////////////////////////////////////////////
-  // Verbose
-  ////////////////////////////////////////////////////////////////
-#if 0
-  std::cout<< GridLogMessage << "MPI-3 usage "<<std::endl;
-  std::cout<< GridLogMessage << "SHM   ";
-  for(int d=0;d<_ndimension;d++){
-    std::cout<< ShmDims[d] <<" ";
-  }
-  std::cout<< std::endl;
-
-  std::cout<< GridLogMessage << "Group ";
-  for(int d=0;d<_ndimension;d++){
-    std::cout<< GroupDims[d] <<" ";
-  }
-  std::cout<< std::endl;
-
-  std::cout<< GridLogMessage<<"World ";
-  for(int d=0;d<_ndimension;d++){
-    std::cout<< WorldDims[d] <<" ";
-  }
-  std::cout<< std::endl;
-#endif
   ////////////////////////////////////////////////////////////////
   // Check processor counts match
   ////////////////////////////////////////////////////////////////
@@ -450,57 +317,29 @@ CartesianCommunicator::CartesianCommunicator(const std::vector<int> &processors)
       
   ////////////////////////////////////////////////////////////////
   // Establish mapping between lexico physics coord and WorldRank
+  // 
   ////////////////////////////////////////////////////////////////
+  LexicographicToWorldRank.resize(WorldSize,0);
   Lexicographic::CoorFromIndex(GroupCoor,GroupRank,GroupDims);
   Lexicographic::CoorFromIndex(ShmCoor,ShmRank,ShmDims);
   for(int d=0;d<_ndimension;d++){
     WorldCoor[d] = GroupCoor[d]*ShmDims[d]+ShmCoor[d];
   }
   _processor_coor = WorldCoor;
-  _processor      = WorldRank;
+
+  int lexico;
+  Lexicographic::IndexFromCoor(WorldCoor,lexico,WorldDims);
+  LexicographicToWorldRank[lexico]=WorldRank;
+  _processor = lexico;
 
   ///////////////////////////////////////////////////////////////////
   // global sum Lexico to World mapping
   ///////////////////////////////////////////////////////////////////
-  int lexico;
-  LexicographicToWorldRank.resize(WorldSize,0);
-  Lexicographic::IndexFromCoor(WorldCoor,lexico,WorldDims);
-  LexicographicToWorldRank[lexico] = WorldRank;
   ierr=MPI_Allreduce(MPI_IN_PLACE,&LexicographicToWorldRank[0],WorldSize,MPI_INT,MPI_SUM,communicator);
   assert(ierr==0);
-
+  
-  for(int i=0;i<WorldSize;i++){
-
-    int wr = LexicographicToWorldRank[i];
-    //    int wr = i;
-
-    std::vector<int> coor(_ndimension);
-    ProcessorCoorFromRank(wr,coor); // from world rank
-    int ck = RankFromProcessorCoor(coor);
-    assert(ck==wr);
-
-    if ( wr == WorldRank ) { 
-      for(int j=0;j<coor.size();j++) {
-	assert(coor[j] == _processor_coor[j]);
-      }
-    }
-    /*
-    std::cout << GridLogMessage<< " Lexicographic "<<i;
-    std::cout << " MPI rank      "<<wr;
-    std::cout << " Coor          ";
-    for(int j=0;j<coor.size();j++) std::cout << coor[j];
-    std::cout<< std::endl;
-    */
-    /////////////////////////////////////////////////////
-    // Check everyone agrees on everyone elses coords
-    /////////////////////////////////////////////////////
-    std::vector<int> mcoor = coor;
-    this->Broadcast(0,(void *)&mcoor[0],mcoor.size()*sizeof(int));
-    for(int d = 0 ; d< _ndimension; d++) {
-      assert(coor[d] == mcoor[d]);
-    }
-  }
 };
+
 void CartesianCommunicator::GlobalSum(uint32_t &u){
   int ierr=MPI_Allreduce(MPI_IN_PLACE,&u,1,MPI_UINT32_T,MPI_SUM,communicator);
   assert(ierr==0);
@@ -528,6 +367,8 @@ void CartesianCommunicator::GlobalSumVector(double *d,int N)
   int ierr = MPI_Allreduce(MPI_IN_PLACE,d,N,MPI_DOUBLE,MPI_SUM,communicator);
   assert(ierr==0);
 }
+
+
 // Basic Halo comms primitive
 void CartesianCommunicator::SendToRecvFrom(void *xmit,
 					   int dest,
@@ -536,14 +377,10 @@ void CartesianCommunicator::SendToRecvFrom(void *xmit,
 					   int bytes)
 {
   std::vector<CommsRequest_t> reqs(0);
-  //    unsigned long  xcrc = crc32(0L, Z_NULL, 0);
-  //    unsigned long  rcrc = crc32(0L, Z_NULL, 0);
-  //    xcrc = crc32(xcrc,(unsigned char *)xmit,bytes);
   SendToRecvFromBegin(reqs,xmit,dest,recv,from,bytes);
   SendToRecvFromComplete(reqs);
-  //    rcrc = crc32(rcrc,(unsigned char *)recv,bytes);
-  //    printf("proc %d SendToRecvFrom %d bytes %lx %lx\n",_processor,bytes,xcrc,rcrc);
 }
+
 void CartesianCommunicator::SendRecvPacket(void *xmit,
 					   void *recv,
 					   int sender,
@@ -560,6 +397,7 @@ void CartesianCommunicator::SendRecvPacket(void *xmit,
     MPI_Recv(recv, bytes, MPI_CHAR,sender,tag,communicator,&stat);
   }
 }
+
 // Basic Halo comms primitive
 void CartesianCommunicator::SendToRecvFromBegin(std::vector<CommsRequest_t> &list,
 						void *xmit,
@@ -568,29 +406,95 @@ void CartesianCommunicator::SendToRecvFromBegin(std::vector<CommsRequest_t> &lis
 						int from,
 						int bytes)
 {
-  int myrank = _processor;
+#if 0
+  this->StencilBarrier();
+
+  MPI_Request xrq;
+  MPI_Request rrq;
+  
+  static int sequence;
+
   int ierr;
+  int tag;
+  int check;
 
-  if ( CommunicatorPolicy == CommunicatorPolicyConcurrent ) { 
+  assert(dest != _processor);
-    MPI_Request xrq;
+  assert(from != _processor);
-    MPI_Request rrq;
+  
+  int gdest = GroupRanks[dest];
+  int gfrom = GroupRanks[from];
+  int gme   = GroupRanks[_processor];
 
-    ierr =MPI_Irecv(recv, bytes, MPI_CHAR,from,from,communicator,&rrq);
+  sequence++;
-    ierr|=MPI_Isend(xmit, bytes, MPI_CHAR,dest,_processor,communicator,&xrq);
+  
-    
+  char *from_ptr = (char *)ShmCommBufs[ShmRank];
+
+  int small = (bytes<MAX_MPI_SHM_BYTES);
+
+  typedef uint64_t T;
+  int words = bytes/sizeof(T);
+
+  assert(((size_t)bytes &(sizeof(T)-1))==0);
+  assert(gme == ShmRank);
+
+  if ( small && (gdest !=MPI_UNDEFINED) ) {
+
+    char *to_ptr   = (char *)ShmCommBufs[gdest];
+
+    assert(gme != gdest);
+
+    T *ip = (T *)xmit;
+    T *op = (T *)to_ptr;
+PARALLEL_FOR_LOOP 
+    for(int w=0;w<words;w++) {
+      op[w]=ip[w];
+    }
+
+    bcopy(&_processor,&to_ptr[bytes],sizeof(_processor));
+    bcopy(&  sequence,&to_ptr[bytes+4],sizeof(sequence));
+  } else { 
+    ierr =MPI_Isend(xmit, bytes, MPI_CHAR,dest,_processor,communicator,&xrq);
     assert(ierr==0);
     list.push_back(xrq);
-    list.push_back(rrq);
-  } else { 
-    // Give the CPU to MPI immediately; can use threads to overlap optionally
-    ierr=MPI_Sendrecv(xmit,bytes,MPI_CHAR,dest,myrank,
-		      recv,bytes,MPI_CHAR,from, from,
-		      communicator,MPI_STATUS_IGNORE);
-    assert(ierr==0);
   }
+
+  this->StencilBarrier();
+  
+  if (small && (gfrom !=MPI_UNDEFINED) ) {
+    T *ip = (T *)from_ptr;
+    T *op = (T *)recv;
+PARALLEL_FOR_LOOP 
+    for(int w=0;w<words;w++) {
+      op[w]=ip[w];
+    }
+    bcopy(&from_ptr[bytes]  ,&tag  ,sizeof(tag));
+    bcopy(&from_ptr[bytes+4],&check,sizeof(check));
+    assert(check==sequence);
+    assert(tag==from);
+  } else { 
+    ierr=MPI_Irecv(recv, bytes, MPI_CHAR,from,from,communicator,&rrq);
+    assert(ierr==0);
+    list.push_back(rrq);
+  }
+
+  this->StencilBarrier();
+
+#else
+  MPI_Request xrq;
+  MPI_Request rrq;
+  int rank = _processor;
+  int ierr;
+  ierr =MPI_Isend(xmit, bytes, MPI_CHAR,dest,_processor,communicator,&xrq);
+  ierr|=MPI_Irecv(recv, bytes, MPI_CHAR,from,from,communicator,&rrq);
+  
+  assert(ierr==0);
+
+  list.push_back(xrq);
+  list.push_back(rrq);
+#endif
 }
 
-double CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsRequest_t> &list,
+void CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsRequest_t> &list,
 						       void *xmit,
 						       int dest,
 						       void *recv,
@@ -601,63 +505,57 @@ double CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsReques
   MPI_Request rrq;
 
   int ierr;
+
+  assert(dest != _processor);
+  assert(from != _processor);
+  
   int gdest = GroupRanks[dest];
   int gfrom = GroupRanks[from];
   int gme   = GroupRanks[_processor];
 
-  assert(dest != _processor);
+  assert(gme == ShmRank);
-  assert(from != _processor);
-  assert(gme  == ShmRank);
-  double off_node_bytes=0.0;
-
-#ifdef FORCE_COMMS
-  gdest = MPI_UNDEFINED;
-  gfrom = MPI_UNDEFINED;
-#endif
-  if ( gfrom ==MPI_UNDEFINED) {
-    ierr=MPI_Irecv(recv, bytes, MPI_CHAR,from,from,communicator,&rrq);
-    assert(ierr==0);
-    list.push_back(rrq);
-    off_node_bytes+=bytes;
-  }
 
   if ( gdest == MPI_UNDEFINED ) {
     ierr =MPI_Isend(xmit, bytes, MPI_CHAR,dest,_processor,communicator,&xrq);
     assert(ierr==0);
     list.push_back(xrq);
-    off_node_bytes+=bytes;
+  }
+  
+  if ( gfrom ==MPI_UNDEFINED) {
+    ierr=MPI_Irecv(recv, bytes, MPI_CHAR,from,from,communicator,&rrq);
+    assert(ierr==0);
+    list.push_back(rrq);
   }
 
-  if ( CommunicatorPolicy == CommunicatorPolicySequential ) { 
-    this->StencilSendToRecvFromComplete(list);
-  }
-
-  return off_node_bytes;
 }
-void CartesianCommunicator::StencilSendToRecvFromComplete(std::vector<CommsRequest_t> &waitall)
+
+
+void CartesianCommunicator::StencilSendToRecvFromComplete(std::vector<CommsRequest_t> &list)
 {
-  SendToRecvFromComplete(waitall);
+  SendToRecvFromComplete(list);
 }
+
 void CartesianCommunicator::StencilBarrier(void)
 {
+  MPI_Win_sync (ShmWindow);   
   MPI_Barrier  (ShmComm);
+  MPI_Win_sync (ShmWindow);   
 }
+
 void CartesianCommunicator::SendToRecvFromComplete(std::vector<CommsRequest_t> &list)
 {
   int nreq=list.size();
-
-  if (nreq==0) return;
-
   std::vector<MPI_Status> status(nreq);
   int ierr = MPI_Waitall(nreq,&list[0],&status[0]);
   assert(ierr==0);
-  list.resize(0);
 }
+
 void CartesianCommunicator::Barrier(void)
 {
   int ierr = MPI_Barrier(communicator);
   assert(ierr==0);
 }
+
 void CartesianCommunicator::Broadcast(int root,void* data, int bytes)
 {
   int ierr=MPI_Bcast(data,
@@ -667,11 +565,7 @@ void CartesianCommunicator::Broadcast(int root,void* data, int bytes)
 		     communicator);
   assert(ierr==0);
 }
-int CartesianCommunicator::RankWorld(void){ 
+
-  int r; 
-  MPI_Comm_rank(communicator_world,&r);
-  return r;
-}
 void CartesianCommunicator::BroadcastWorld(int root,void* data, int bytes)
 {
   int ierr= MPI_Bcast(data,

lib/communicator/Communicator_mpi3_leader.cc

@@ -27,7 +27,6 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
     /*  END LEGAL */
 #include "Grid.h"
 #include <mpi.h>
-//#include <numaif.h>
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 /// Workarounds:
@@ -43,27 +42,19 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 #include <fcntl.h>
 #include <unistd.h>
 #include <limits.h>
+
 typedef sem_t *Grid_semaphore;
 
-
-#error  /*THis is deprecated*/
-
-#if 0 
 #define SEM_INIT(S)      S = sem_open(sem_name,0,0600,0); assert ( S != SEM_FAILED );
 #define SEM_INIT_EXCL(S) sem_unlink(sem_name); S = sem_open(sem_name,O_CREAT|O_EXCL,0600,0); assert ( S != SEM_FAILED );
 #define SEM_POST(S) assert ( sem_post(S) == 0 ); 
 #define SEM_WAIT(S) assert ( sem_wait(S) == 0 );
-#else
+
-#define SEM_INIT(S)      ;
-#define SEM_INIT_EXCL(S) ;
-#define SEM_POST(S) ;
-#define SEM_WAIT(S) ;
-#endif
 #include <sys/mman.h>
 
 namespace Grid {
 
-enum { COMMAND_ISEND, COMMAND_IRECV, COMMAND_WAITALL, COMMAND_SENDRECV };
+enum { COMMAND_ISEND, COMMAND_IRECV, COMMAND_WAITALL };
 
 struct Descriptor {
   uint64_t buf;
@@ -71,12 +62,6 @@ struct Descriptor {
   int rank;
   int tag;
   int command;
-  uint64_t xbuf;
-  uint64_t rbuf;
-  int xtag;
-  int rtag;
-  int src;
-  int dest;
   MPI_Request request;
 };
 
@@ -109,14 +94,18 @@ public:
 
   void SemInit(void) {
     sprintf(sem_name,"/Grid_mpi3_sem_head_%d",universe_rank);
+    //    printf("SEM_NAME: %s \n",sem_name);
     SEM_INIT(sem_head);
     sprintf(sem_name,"/Grid_mpi3_sem_tail_%d",universe_rank);
+    //    printf("SEM_NAME: %s \n",sem_name);
     SEM_INIT(sem_tail);
   }  
   void SemInitExcl(void) {
     sprintf(sem_name,"/Grid_mpi3_sem_head_%d",universe_rank);
+    //    printf("SEM_INIT_EXCL: %s \n",sem_name);
     SEM_INIT_EXCL(sem_head);
     sprintf(sem_name,"/Grid_mpi3_sem_tail_%d",universe_rank);
+    //    printf("SEM_INIT_EXCL: %s \n",sem_name);
     SEM_INIT_EXCL(sem_tail);
   }  
   void WakeUpDMA(void) { 
@@ -136,13 +125,6 @@ public:
     while(1){
       WaitForCommand();
       //      std::cout << "Getting command "<<std::endl;
-#if 0
-      _mm_monitor((void *)&state->head,0,0);
-      int s=state->start;
-      if ( s != state->head ) {
-	_mm_mwait(0,0);
-      }
-#endif
       Event();
     }
   }
@@ -150,7 +132,6 @@ public:
   int Event (void) ;
 
   uint64_t QueueCommand(int command,void *buf, int bytes, int hashtag, MPI_Comm comm,int u_rank) ;
-  void QueueSendRecv(void *xbuf, void *rbuf, int bytes, int xtag, int rtag, MPI_Comm comm,int dest,int src) ;
 
   void WaitAll() {
     //    std::cout << "Queueing WAIT command  "<<std::endl;
@@ -160,7 +141,7 @@ public:
     //    std::cout << "Waiting from semaphore "<<std::endl;
     WaitForComplete();
     //    std::cout << "Checking FIFO is empty "<<std::endl;
-    while ( state->tail != state->head );
+    assert ( state->tail == state->head );
   }
 };
 
@@ -215,12 +196,6 @@ public:
     //    std::cout << "Waking up DMA "<< slave<<std::endl;
   };
 
-  static void QueueSendRecv(int slave,void *xbuf, void *rbuf, int bytes, int xtag, int rtag, MPI_Comm comm,int dest,int src) 
-  {
-    Slaves[slave].QueueSendRecv(xbuf,rbuf,bytes,xtag,rtag,comm,dest,src);
-    Slaves[slave].WakeUpDMA();
-  }
-
   static void QueueRecv(int slave, void *buf, int bytes, int tag, MPI_Comm comm,int rank) {
     //    std::cout<< " Queueing recv "<< bytes<< " slave "<< slave << " from comm "<<rank  <<std::endl;
     Slaves[slave].QueueCommand(COMMAND_IRECV,buf,bytes,tag,comm,rank);
@@ -251,28 +226,6 @@ public:
     return;
   };
 
-  static void QueueRoundRobinSendRecv(void *xbuf, void *rbuf, int bytes, int xtag, int rtag, MPI_Comm comm,int dest,int src) {
-    uint8_t * cxbuf = (uint8_t *) xbuf;
-    uint8_t * crbuf = (uint8_t *) rbuf;
-    static int rrp=0;
-    int procs = VerticalSize-1;
-    int myoff=0;
-    int mywork=bytes;
-    QueueSendRecv(rrp+1,&cxbuf[myoff],&crbuf[myoff],mywork,xtag,rtag,comm,dest,src);
-    rrp = rrp+1;
-    if ( rrp == (VerticalSize-1) ) rrp = 0;
-  }
-
-  static void QueueMultiplexedSendRecv(void *xbuf, void *rbuf, int bytes, int xtag, int rtag, MPI_Comm comm,int dest,int src) {
-    uint8_t * cxbuf = (uint8_t *) xbuf;
-    uint8_t * crbuf = (uint8_t *) rbuf;
-    int mywork, myoff, procs;
-    procs = VerticalSize-1;
-    for(int s=0;s<procs;s++) {
-      GetWork(bytes,s,mywork,myoff,procs);
-      QueueSendRecv(s+1,&cxbuf[myoff],&crbuf[myoff],mywork,xtag,rtag,comm,dest,src);
-    }
-  };
   static void QueueMultiplexedSend(void *buf, int bytes, int tag, MPI_Comm comm,int rank) {
     uint8_t * cbuf = (uint8_t *) buf;
     int mywork, myoff, procs;
@@ -322,7 +275,6 @@ std::vector<void *>            MPIoffloadEngine::VerticalShmBufs;
 std::vector<std::vector<int> > MPIoffloadEngine::UniverseRanks;
 std::vector<int>               MPIoffloadEngine::UserCommunicatorToWorldRanks; 
 
-int CartesianCommunicator::NodeCount(void)    { return HorizontalSize;};
 int MPIoffloadEngine::ShmSetup = 0;
 
 void MPIoffloadEngine::CommunicatorInit (MPI_Comm &communicator_world,
@@ -418,22 +370,12 @@ void MPIoffloadEngine::CommunicatorInit (MPI_Comm &communicator_world,
       ftruncate(fd, size);
 
       VerticalShmBufs[r] = mmap(NULL,size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
+
       if ( VerticalShmBufs[r] == MAP_FAILED ) { 
 	perror("failed mmap");
 	assert(0);
       }
 
-      /*
-      for(uint64_t page=0;page<size;page+=4096){
-	void *pages = (void *) ( page + (uint64_t)VerticalShmBufs[r] );
-	int status;
-	int flags=MPOL_MF_MOVE_ALL;
-	int nodes=1; // numa domain == MCDRAM
-	unsigned long count=1;
-	ierr= move_pages(0,count, &pages,&nodes,&status,flags);
-	if (ierr && (page==0)) perror("numa relocate command failed");
-      }
-      */
       uint64_t * check = (uint64_t *) VerticalShmBufs[r];
       check[0] = WorldRank;
       check[1] = r;
@@ -462,7 +404,7 @@ void MPIoffloadEngine::CommunicatorInit (MPI_Comm &communicator_world,
     uint64_t * check = (uint64_t *) VerticalShmBufs[r];
     assert(check[0]== WorldRank);
     assert(check[1]== r);
-    //    std::cerr<<"SHM "<<r<<" " <<VerticalShmBufs[r]<<std::endl;
+    std::cerr<<"SHM "<<r<<" " <<VerticalShmBufs[r]<<std::endl;
   }
   }
 #endif
@@ -600,8 +542,6 @@ int Slave::Event (void) {
   static int head_last;
   static int start_last;
   int ierr;
-  MPI_Status stat;
-  static int i=0;
 
   ////////////////////////////////////////////////////
   // Try to advance the start pointers
@@ -610,6 +550,11 @@ int Slave::Event (void) {
   if ( s != state->head ) {
     switch ( state->Descrs[s].command ) {
     case COMMAND_ISEND:
+      /*
+            std::cout<< " Send "<<s << " ptr "<< state<<" "<< state->Descrs[s].buf<< "["<<state->Descrs[s].bytes<<"]"
+      	       << " to " << state->Descrs[s].rank<< " tag" << state->Descrs[s].tag
+       << " Comm " << MPIoffloadEngine::communicator_universe<< " me " <<universe_rank<< std::endl;
+      */
       ierr = MPI_Isend((void *)(state->Descrs[s].buf+base), 
 		       state->Descrs[s].bytes, 
 		       MPI_CHAR,
@@ -623,6 +568,11 @@ int Slave::Event (void) {
       break;
 
     case COMMAND_IRECV:
+      /*
+      std::cout<< " Recv "<<s << " ptr "<< state<<" "<< state->Descrs[s].buf<< "["<<state->Descrs[s].bytes<<"]"
+	       << " from " << state->Descrs[s].rank<< " tag" << state->Descrs[s].tag
+	       << " Comm " << MPIoffloadEngine::communicator_universe<< " me "<< universe_rank<< std::endl;
+      */
       ierr=MPI_Irecv((void *)(state->Descrs[s].buf+base), 
 		     state->Descrs[s].bytes, 
 		     MPI_CHAR,
@@ -638,32 +588,10 @@ int Slave::Event (void) {
       return 1;
       break;
 
-    case COMMAND_SENDRECV:
-
-      //      fprintf(stderr,"Sendrecv ->%d %d : <-%d %d \n",state->Descrs[s].dest, state->Descrs[s].xtag+i*10,state->Descrs[s].src, state->Descrs[s].rtag+i*10);
-
-      ierr=MPI_Sendrecv((void *)(state->Descrs[s].xbuf+base), state->Descrs[s].bytes, MPI_CHAR, state->Descrs[s].dest, state->Descrs[s].xtag+i*10,
-			(void *)(state->Descrs[s].rbuf+base), state->Descrs[s].bytes, MPI_CHAR, state->Descrs[s].src , state->Descrs[s].rtag+i*10,
-			MPIoffloadEngine::communicator_universe,MPI_STATUS_IGNORE);
-
-      assert(ierr==0);
-
-      //      fprintf(stderr,"Sendrecv done %d %d\n",ierr,i);
-      //      MPI_Barrier(MPIoffloadEngine::HorizontalComm);
-      //      fprintf(stderr,"Barrier\n");
-      i++;
-
-      state->start = PERI_PLUS(s);
-
-      return 1;
-      break;
-
     case COMMAND_WAITALL:
 
       for(int t=state->tail;t!=s; t=PERI_PLUS(t) ){
-	if ( state->Descrs[t].command != COMMAND_SENDRECV ) {
+	MPI_Wait((MPI_Request *)&state->Descrs[t].request,MPI_STATUS_IGNORE);
-	  MPI_Wait((MPI_Request *)&state->Descrs[t].request,MPI_STATUS_IGNORE);
-	}
       };
       s=PERI_PLUS(s);
       state->start = s;
@@ -685,45 +613,6 @@ int Slave::Event (void) {
   // External interaction with the queue
   //////////////////////////////////////////////////////////////////////////////
   
-void Slave::QueueSendRecv(void *xbuf, void *rbuf, int bytes, int xtag, int rtag, MPI_Comm comm,int dest,int src) 
-{
-  int head =state->head;
-  int next = PERI_PLUS(head);
-  
-  // Set up descriptor
-  int worldrank;
-  int hashtag;
-  MPI_Comm    communicator;
-  MPI_Request request;
-  uint64_t relative;
-  
-  relative = (uint64_t)xbuf - base;
-  state->Descrs[head].xbuf    = relative;
-  
-  relative= (uint64_t)rbuf - base;
-  state->Descrs[head].rbuf    = relative;
-  
-  state->Descrs[head].bytes  = bytes;
-  
-  MPIoffloadEngine::MapCommRankToWorldRank(hashtag,worldrank,xtag,comm,dest);
-  state->Descrs[head].dest   = MPIoffloadEngine::UniverseRanks[worldrank][vertical_rank];
-  state->Descrs[head].xtag    = hashtag;
-  
-  MPIoffloadEngine::MapCommRankToWorldRank(hashtag,worldrank,rtag,comm,src);
-  state->Descrs[head].src    = MPIoffloadEngine::UniverseRanks[worldrank][vertical_rank];
-  state->Descrs[head].rtag    = hashtag;
-  
-  state->Descrs[head].command= COMMAND_SENDRECV;
-  
-  // Block until FIFO has space
-  while( state->tail==next );
-  
-  // Msync on weak order architectures
-  
-  // Advance pointer
-  state->head = next;
-  
-};
 uint64_t Slave::QueueCommand(int command,void *buf, int bytes, int tag, MPI_Comm comm,int commrank) 
 {
   /////////////////////////////////////////
@@ -923,22 +812,19 @@ void CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsRequest_
   assert( (recv_i >= shm) && (recv_i+bytes <= shm+MAX_MPI_SHM_BYTES) );
   assert(from!=_processor);
   assert(dest!=_processor);
-
+  MPIoffloadEngine::QueueMultiplexedSend(xmit,bytes,_processor,communicator,dest);
-  MPIoffloadEngine::QueueMultiplexedSendRecv(xmit,recv,bytes,_processor,from,communicator,dest,from);
+  MPIoffloadEngine::QueueMultiplexedRecv(recv,bytes,from,communicator,from);
-
-  //MPIoffloadEngine::QueueRoundRobinSendRecv(xmit,recv,bytes,_processor,from,communicator,dest,from);
-
-  //MPIoffloadEngine::QueueMultiplexedSend(xmit,bytes,_processor,communicator,dest);
-  //MPIoffloadEngine::QueueMultiplexedRecv(recv,bytes,from,communicator,from);
 }
 
+
 void CartesianCommunicator::StencilSendToRecvFromComplete(std::vector<CommsRequest_t> &list)
 {
   MPIoffloadEngine::WaitAll();
-  //this->Barrier();
 }
 
-void CartesianCommunicator::StencilBarrier(void) { }
+void CartesianCommunicator::StencilBarrier(void)
+{
+}
 
 void CartesianCommunicator::SendToRecvFromComplete(std::vector<CommsRequest_t> &list)
 {

lib/communicator/Communicator_none.cc

@@ -25,8 +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/GridCore.h>
+#include <Grid/Grid.h>
-
 namespace Grid {
 
 ///////////////////////////////////////////////////////////////////////////////////////////////////
@@ -88,7 +87,6 @@ void CartesianCommunicator::SendToRecvFromBegin(std::vector<CommsRequest_t> &lis
 {
   assert(0);
 }
-
 void CartesianCommunicator::SendToRecvFromComplete(std::vector<CommsRequest_t> &list)
 {
   assert(0);
@@ -99,7 +97,7 @@ void CartesianCommunicator::Barrier(void){}
 void CartesianCommunicator::Broadcast(int root,void* data, int bytes) {}
 void CartesianCommunicator::BroadcastWorld(int root,void* data, int bytes) { }
 int  CartesianCommunicator::RankFromProcessorCoor(std::vector<int> &coor) {  return 0;}
-void CartesianCommunicator::ProcessorCoorFromRank(int rank, std::vector<int> &coor){  coor = _processor_coor; }
+void CartesianCommunicator::ProcessorCoorFromRank(int rank, std::vector<int> &coor){ coor = _processor_coor ;}
 void CartesianCommunicator::ShiftedRanks(int dim,int shift,int &source,int &dest)
 {
   source =0;

lib/communicator/Communicator_shmem.cc

@@ -27,7 +27,6 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
     /*  END LEGAL */
 #include <Grid/Grid.h>
 #include <mpp/shmem.h>
-#include <array>
 
 namespace Grid {
 
@@ -52,7 +51,7 @@ typedef struct HandShake_t {
 } HandShake;
 
 std::array<long,_SHMEM_REDUCE_SYNC_SIZE> make_psync_init(void) {
-  std::array<long,_SHMEM_REDUCE_SYNC_SIZE> ret;
+  array<long,_SHMEM_REDUCE_SYNC_SIZE> ret;
   ret.fill(SHMEM_SYNC_VALUE);
   return ret;
 }
@@ -110,7 +109,7 @@ void CartesianCommunicator::GlobalSum(uint32_t &u){
 
   source = u;
   dest   = 0;
-  shmem_longlong_sum_to_all(&dest,&source,1,0,0,_Nprocessors,llwrk,psync.data());
+  shmem_longlong_sum_to_all(&dest,&source,1,0,0,_Nprocessors,llwrk,psync);
   shmem_barrier_all(); // necessary?
   u = dest;
 }
@@ -126,7 +125,7 @@ void CartesianCommunicator::GlobalSum(uint64_t &u){
 
   source = u;
   dest   = 0;
-  shmem_longlong_sum_to_all(&dest,&source,1,0,0,_Nprocessors,llwrk,psync.data());
+  shmem_longlong_sum_to_all(&dest,&source,1,0,0,_Nprocessors,llwrk,psync);
   shmem_barrier_all(); // necessary?
   u = dest;
 }
@@ -138,8 +137,7 @@ void CartesianCommunicator::GlobalSum(float &f){
 
   source = f;
   dest   =0.0;
-  shmem_float_sum_to_all(&dest,&source,1,0,0,_Nprocessors,llwrk,psync.data());
+  shmem_float_sum_to_all(&dest,&source,1,0,0,_Nprocessors,llwrk,psync);
-  shmem_barrier_all();
   f = dest;
 }
 void CartesianCommunicator::GlobalSumVector(float *f,int N)
@@ -150,16 +148,14 @@ void CartesianCommunicator::GlobalSumVector(float *f,int N)
   static std::array<long,_SHMEM_REDUCE_SYNC_SIZE> psync =  psync_init;
 
   if ( shmem_addr_accessible(f,_processor)  ){
-    shmem_float_sum_to_all(f,f,N,0,0,_Nprocessors,llwrk,psync.data());
+    shmem_float_sum_to_all(f,f,N,0,0,_Nprocessors,llwrk,psync);
-    shmem_barrier_all();
     return;
   }
 
   for(int i=0;i<N;i++){
     dest   =0.0;
     source = f[i];
-    shmem_float_sum_to_all(&dest,&source,1,0,0,_Nprocessors,llwrk,psync.data());
+    shmem_float_sum_to_all(&dest,&source,1,0,0,_Nprocessors,llwrk,psync);
-    shmem_barrier_all();
     f[i] = dest;
   }
 }
@@ -172,8 +168,7 @@ void CartesianCommunicator::GlobalSum(double &d)
 
   source = d;
   dest   = 0;
-  shmem_double_sum_to_all(&dest,&source,1,0,0,_Nprocessors,llwrk,psync.data());
+  shmem_double_sum_to_all(&dest,&source,1,0,0,_Nprocessors,llwrk,psync);
-  shmem_barrier_all();
   d = dest;
 }
 void CartesianCommunicator::GlobalSumVector(double *d,int N)
@@ -185,16 +180,14 @@ void CartesianCommunicator::GlobalSumVector(double *d,int N)
 
 
   if ( shmem_addr_accessible(d,_processor)  ){
-    shmem_double_sum_to_all(d,d,N,0,0,_Nprocessors,llwrk,psync.data());
+    shmem_double_sum_to_all(d,d,N,0,0,_Nprocessors,llwrk,psync);
-    shmem_barrier_all();
     return;
   }
 
   for(int i=0;i<N;i++){
     source = d[i];
     dest   =0.0;
-    shmem_double_sum_to_all(&dest,&source,1,0,0,_Nprocessors,llwrk,psync.data());
+    shmem_double_sum_to_all(&dest,&source,1,0,0,_Nprocessors,llwrk,psync);
-    shmem_barrier_all();
     d[i] = dest;
   }
 }
@@ -289,13 +282,11 @@ void CartesianCommunicator::SendToRecvFromBegin(std::vector<CommsRequest_t> &lis
   SHMEM_VET(recv);
   //  shmem_putmem_nb(recv,xmit,bytes,dest,NULL);
   shmem_putmem(recv,xmit,bytes,dest);
-
-  if ( CommunicatorPolicy == CommunicatorPolicySequential ) shmem_barrier_all(); 
 }
 void CartesianCommunicator::SendToRecvFromComplete(std::vector<CommsRequest_t> &list)
 {
   //  shmem_quiet();      // I'm done
-  if( CommunicatorPolicy == CommunicatorPolicyConcurrent ) shmem_barrier_all();// He's done too
+  shmem_barrier_all();// He's done too
 }
 void CartesianCommunicator::Barrier(void)
 {
@@ -310,13 +301,13 @@ void CartesianCommunicator::Broadcast(int root,void* data, int bytes)
   int words = bytes/4;
 
   if ( shmem_addr_accessible(data,_processor)  ){
-    shmem_broadcast32(data,data,words,root,0,0,shmem_n_pes(),psync.data());
+    shmem_broadcast32(data,data,words,root,0,0,shmem_n_pes(),psync);
     return;
   }
 
   for(int w=0;w<words;w++){
     word = array[w];
-    shmem_broadcast32((void *)&word,(void *)&word,1,root,0,0,shmem_n_pes(),psync.data());
+    shmem_broadcast32((void *)&word,(void *)&word,1,root,0,0,shmem_n_pes(),psync);
     if ( shmem_my_pe() != root ) {
       array[w] = word;
     }
@@ -334,17 +325,13 @@ void CartesianCommunicator::BroadcastWorld(int root,void* data, int bytes)
 
   for(int w=0;w<words;w++){
     word = array[w];
-    shmem_broadcast32((void *)&word,(void *)&word,1,root,0,0,shmem_n_pes(),psync.data());
+    shmem_broadcast32((void *)&word,(void *)&word,1,root,0,0,shmem_n_pes(),psync);
     if ( shmem_my_pe() != root ) {
       array[w]= word;
     }
     shmem_barrier_all();
   }
 }
-  
-int CartesianCommunicator::RankWorld(void){ 
-  return shmem_my_pe();
-}
 
 }
 

lib/cshift/Cshift_common.h

@@ -1,4 +1,5 @@
-/*************************************************************************************
+
+    /*************************************************************************************
 
     Grid physics library, www.github.com/paboyle/Grid 
 
@@ -30,11 +31,21 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 
 namespace Grid {
 
+template<class vobj>
+class SimpleCompressor {
+public:
+  void Point(int) {};
+
+  vobj operator() (const vobj &arg) {
+    return arg;
+  }
+};
+
 ///////////////////////////////////////////////////////////////////
-// Gather for when there is no need to SIMD split 
+// Gather for when there is no need to SIMD split with compression
 ///////////////////////////////////////////////////////////////////
-template<class vobj> void 
+template<class vobj,class cobj,class compressor> void 
-Gather_plane_simple (const Lattice<vobj> &rhs,commVector<vobj> &buffer,int dimension,int plane,int cbmask, int off=0)
+Gather_plane_simple (const Lattice<vobj> &rhs,commVector<cobj> &buffer,int dimension,int plane,int cbmask,compressor &compress, int off=0)
 {
   int rd = rhs._grid->_rdimensions[dimension];
 
@@ -42,17 +53,19 @@ Gather_plane_simple (const Lattice<vobj> &rhs,commVector<vobj> &buffer,int dimen
     cbmask = 0x3;
   }
   
-  int so=plane*rhs._grid->_ostride[dimension]; // base offset for start of plane 
+  int so  = plane*rhs._grid->_ostride[dimension]; // base offset for start of plane 
+  
   int e1=rhs._grid->_slice_nblock[dimension];
   int e2=rhs._grid->_slice_block[dimension];
 
   int stride=rhs._grid->_slice_stride[dimension];
   if ( cbmask == 0x3 ) { 
-    parallel_for_nest2(int n=0;n<e1;n++){
+PARALLEL_NESTED_LOOP2
+    for(int n=0;n<e1;n++){
       for(int b=0;b<e2;b++){
 	int o  = n*stride;
 	int bo = n*e2;
-	buffer[off+bo+b]=rhs._odata[so+o+b];
+	buffer[off+bo+b]=compress(rhs._odata[so+o+b]);
       }
     }
   } else { 
@@ -61,23 +74,25 @@ Gather_plane_simple (const Lattice<vobj> &rhs,commVector<vobj> &buffer,int dimen
      for(int n=0;n<e1;n++){
        for(int b=0;b<e2;b++){
 	 int o  = n*stride;
-	 int ocb=1<<rhs._grid->CheckerBoardFromOindex(o+b);
+	 int ocb=1<<rhs._grid->CheckerBoardFromOindexTable(o+b);
 	 if ( ocb &cbmask ) {
 	   table.push_back(std::pair<int,int> (bo++,o+b));
 	 }
        }
      }
-     parallel_for(int i=0;i<table.size();i++){
+PARALLEL_FOR_LOOP     
-       buffer[off+table[i].first]=rhs._odata[so+table[i].second];
+     for(int i=0;i<table.size();i++){
+       buffer[off+table[i].first]=compress(rhs._odata[so+table[i].second]);
      }
   }
 }
 
+
 ///////////////////////////////////////////////////////////////////
-// Gather for when there *is* need to SIMD split 
+// Gather for when there *is* need to SIMD split with compression
 ///////////////////////////////////////////////////////////////////
-template<class vobj> void 
+template<class cobj,class vobj,class compressor> void 
-Gather_plane_extract(const Lattice<vobj> &rhs,std::vector<typename vobj::scalar_object *> pointers,int dimension,int plane,int cbmask)
+Gather_plane_extract(const Lattice<vobj> &rhs,std::vector<typename cobj::scalar_object *> pointers,int dimension,int plane,int cbmask,compressor &compress)
 {
   int rd = rhs._grid->_rdimensions[dimension];
 
@@ -90,40 +105,57 @@ Gather_plane_extract(const Lattice<vobj> &rhs,std::vector<typename vobj::scalar_
   int e1=rhs._grid->_slice_nblock[dimension];
   int e2=rhs._grid->_slice_block[dimension];
   int n1=rhs._grid->_slice_stride[dimension];
-
+  int n2=rhs._grid->_slice_block[dimension];
   if ( cbmask ==0x3){
-    parallel_for_nest2(int n=0;n<e1;n++){
+PARALLEL_NESTED_LOOP2
+    for(int n=0;n<e1;n++){
       for(int b=0;b<e2;b++){
 
 	int o      =   n*n1;
-	int offset = b+n*e2;
+	int offset = b+n*n2;
-	
+	cobj temp =compress(rhs._odata[so+o+b]);
-	vobj temp =rhs._odata[so+o+b];
+
-	extract<vobj>(temp,pointers,offset);
+	extract<cobj>(temp,pointers,offset);
 
       }
     }
   } else { 
 
-    // Case of SIMD split AND checker dim cannot currently be hit, except in 
+    assert(0); //Fixme think this is buggy
-    // Test_cshift_red_black code.
-    std::cout << " Dense packed buffer WARNING " <<std::endl;
-    parallel_for_nest2(int n=0;n<e1;n++){
-      for(int b=0;b<e2;b++){
 
-	int o=n*n1;
+    for(int n=0;n<e1;n++){
+      for(int b=0;b<e2;b++){
+	int o=n*rhs._grid->_slice_stride[dimension];
 	int ocb=1<<rhs._grid->CheckerBoardFromOindex(o+b);
-	int offset = b+n*e2;
+	int offset = b+n*rhs._grid->_slice_block[dimension];
 
 	if ( ocb & cbmask ) {
-	  vobj temp =rhs._odata[so+o+b];
+	  cobj temp =compress(rhs._odata[so+o+b]);
-	  extract<vobj>(temp,pointers,offset);
+	  extract<cobj>(temp,pointers,offset);
 	}
       }
     }
   }
 }
 
+//////////////////////////////////////////////////////
+// Gather for when there is no need to SIMD split
+//////////////////////////////////////////////////////
+template<class vobj> void Gather_plane_simple (const Lattice<vobj> &rhs,commVector<vobj> &buffer, int dimension,int plane,int cbmask)
+{
+  SimpleCompressor<vobj> dontcompress;
+  Gather_plane_simple (rhs,buffer,dimension,plane,cbmask,dontcompress);
+}
+
+//////////////////////////////////////////////////////
+// Gather for when there *is* need to SIMD split
+//////////////////////////////////////////////////////
+template<class vobj> void Gather_plane_extract(const Lattice<vobj> &rhs,std::vector<typename vobj::scalar_object *> pointers,int dimension,int plane,int cbmask)
+{
+  SimpleCompressor<vobj> dontcompress;
+  Gather_plane_extract<vobj,vobj,decltype(dontcompress)>(rhs,pointers,dimension,plane,cbmask,dontcompress);
+}
+
 //////////////////////////////////////////////////////
 // Scatter for when there is no need to SIMD split
 //////////////////////////////////////////////////////
@@ -139,10 +171,10 @@ template<class vobj> void Scatter_plane_simple (Lattice<vobj> &rhs,commVector<vo
     
   int e1=rhs._grid->_slice_nblock[dimension];
   int e2=rhs._grid->_slice_block[dimension];
-  int stride=rhs._grid->_slice_stride[dimension];
   
   if ( cbmask ==0x3 ) {
-    parallel_for_nest2(int n=0;n<e1;n++){
+PARALLEL_NESTED_LOOP2
+    for(int n=0;n<e1;n++){
       for(int b=0;b<e2;b++){
 	int o   =n*rhs._grid->_slice_stride[dimension];
 	int bo  =n*rhs._grid->_slice_block[dimension];
@@ -150,28 +182,24 @@ template<class vobj> void Scatter_plane_simple (Lattice<vobj> &rhs,commVector<vo
       }
     }
   } else { 
-    std::vector<std::pair<int,int> > table;
     int bo=0;
     for(int n=0;n<e1;n++){
       for(int b=0;b<e2;b++){
 	int o   =n*rhs._grid->_slice_stride[dimension];
+	int bo  =n*rhs._grid->_slice_block[dimension];
 	int ocb=1<<rhs._grid->CheckerBoardFromOindex(o+b);// Could easily be a table lookup
 	if ( ocb & cbmask ) {
-	  table.push_back(std::pair<int,int> (so+o+b,bo++));
+	  rhs._odata[so+o+b]=buffer[bo++];
 	}
       }
     }
-    parallel_for(int i=0;i<table.size();i++){
-       //       std::cout << "Rcv"<< table[i].first << " " << table[i].second << " " <<buffer[table[i].second]<<std::endl;
-       rhs._odata[table[i].first]=buffer[table[i].second];
-     }
   }
 }
 
 //////////////////////////////////////////////////////
 // Scatter for when there *is* need to SIMD split
 //////////////////////////////////////////////////////
-template<class vobj> void Scatter_plane_merge(Lattice<vobj> &rhs,std::vector<typename vobj::scalar_object *> pointers,int dimension,int plane,int cbmask)
+ template<class vobj,class cobj> void Scatter_plane_merge(Lattice<vobj> &rhs,std::vector<cobj *> pointers,int dimension,int plane,int cbmask)
 {
   int rd = rhs._grid->_rdimensions[dimension];
 
@@ -185,7 +213,8 @@ template<class vobj> void Scatter_plane_merge(Lattice<vobj> &rhs,std::vector<typ
   int e2=rhs._grid->_slice_block[dimension];
 
   if(cbmask ==0x3 ) {
-    parallel_for_nest2(int n=0;n<e1;n++){
+PARALLEL_NESTED_LOOP2
+    for(int n=0;n<e1;n++){
       for(int b=0;b<e2;b++){
 	int o      = n*rhs._grid->_slice_stride[dimension];
 	int offset = b+n*rhs._grid->_slice_block[dimension];
@@ -193,11 +222,7 @@ template<class vobj> void Scatter_plane_merge(Lattice<vobj> &rhs,std::vector<typ
       }
     }
   } else { 
-
+    assert(0); // think this is buggy FIXME
-    // Case of SIMD split AND checker dim cannot currently be hit, except in 
-    // Test_cshift_red_black code.
-    //    std::cout << "Scatter_plane merge assert(0); think this is buggy FIXME "<< std::endl;// think this is buggy FIXME
-    std::cout<<" Unthreaded warning -- buffer is not densely packed ??"<<std::endl;
     for(int n=0;n<e1;n++){
       for(int b=0;b<e2;b++){
 	int o      = n*rhs._grid->_slice_stride[dimension];
@@ -229,7 +254,8 @@ template<class vobj> void Copy_plane(Lattice<vobj>& lhs,const Lattice<vobj> &rhs
   int e2=rhs._grid->_slice_block[dimension];
   int stride = rhs._grid->_slice_stride[dimension];
   if(cbmask == 0x3 ){
-    parallel_for_nest2(int n=0;n<e1;n++){
+PARALLEL_NESTED_LOOP2
+    for(int n=0;n<e1;n++){
       for(int b=0;b<e2;b++){
  
         int o =n*stride+b;
@@ -238,7 +264,8 @@ template<class vobj> void Copy_plane(Lattice<vobj>& lhs,const Lattice<vobj> &rhs
       }
     }
   } else { 
-    parallel_for_nest2(int n=0;n<e1;n++){
+PARALLEL_NESTED_LOOP2
+    for(int n=0;n<e1;n++){
       for(int b=0;b<e2;b++){
  
         int o =n*stride+b;
@@ -268,8 +295,8 @@ template<class vobj> void Copy_plane_permute(Lattice<vobj>& lhs,const Lattice<vo
   int e1=rhs._grid->_slice_nblock[dimension];
   int e2=rhs._grid->_slice_block [dimension];
   int stride = rhs._grid->_slice_stride[dimension];
-
+PARALLEL_NESTED_LOOP2
-  parallel_for_nest2(int n=0;n<e1;n++){
+  for(int n=0;n<e1;n++){
   for(int b=0;b<e2;b++){
 
       int o  =n*stride;
@@ -311,8 +338,8 @@ template<class vobj> Lattice<vobj> Cshift_local(Lattice<vobj> &ret,const Lattice
   // Map to always positive shift modulo global full dimension.
   shift = (shift+fd)%fd;
 
-  // the permute type
   ret.checkerboard = grid->CheckerBoardDestination(rhs.checkerboard,shift,dimension);
+  // the permute type
   int permute_dim =grid->PermuteDim(dimension);
   int permute_type=grid->PermuteType(dimension);
   int permute_type_dist;
@@ -321,6 +348,7 @@ template<class vobj> Lattice<vobj> Cshift_local(Lattice<vobj> &ret,const Lattice
 
     int o   = 0;
     int bo  = x * grid->_ostride[dimension];
+    
     int cb= (cbmask==0x2)? Odd : Even;
 
     int sshift = grid->CheckerBoardShiftForCB(rhs.checkerboard,dimension,shift,cb);
@@ -333,23 +361,9 @@ template<class vobj> Lattice<vobj> Cshift_local(Lattice<vobj> &ret,const Lattice
     // wrap is whether sshift > rd.
     //  num is sshift mod rd.
     // 
-    //  shift 7
-    //
-    //  XoXo YcYc 
-    //  oXoX cYcY
-    //  XoXo YcYc
-    //  oXoX cYcY
-    //
-    //  sshift -- 
-    //
-    //  XX YY ; 3
-    //  XX YY ; 0
-    //  XX YY ; 3
-    //  XX YY ; 0
-    //
     int permute_slice=0;
     if(permute_dim){
-      int wrap = sshift/rd; wrap=wrap % ly;
+      int wrap = sshift/rd;
       int  num = sshift%rd;
 
       if ( x< rd-num ) permute_slice=wrap;
@@ -361,6 +375,7 @@ template<class vobj> Lattice<vobj> Cshift_local(Lattice<vobj> &ret,const Lattice
       } else {
 	permute_type_dist = permute_type;
       }
+      
     }
 
     if ( permute_slice ) Copy_plane_permute(ret,rhs,dimension,x,sx,cbmask,permute_type_dist);

lib/cshift/Cshift_mpi.h

@@ -74,6 +74,7 @@ template<class vobj> void Cshift_comms(Lattice<vobj>& ret,const Lattice<vobj> &r
   sshift[1] = rhs._grid->CheckerBoardShiftForCB(rhs.checkerboard,dimension,shift,Odd);
 
   //  std::cout << "Cshift_comms dim "<<dimension<<"cb "<<rhs.checkerboard<<"shift "<<shift<<" sshift " << sshift[0]<<" "<<sshift[1]<<std::endl;
+
   if ( sshift[0] == sshift[1] ) {
     //    std::cout << "Single pass Cshift_comms" <<std::endl;
     Cshift_comms(ret,rhs,dimension,shift,0x3);
@@ -153,8 +154,10 @@ template<class vobj> void Cshift_comms(Lattice<vobj> &ret,const Lattice<vobj> &r
 			   (void *)&recv_buf[0],
 			   recv_from_rank,
 			   bytes);
-      grid->Barrier();
 
+      //      for(int i=0;i<words;i++){
+      //	std::cout << "SendRecv ["<<i<<"] snd "<<send_buf[i]<<" rcv " << recv_buf[i] << "  0x" << cbmask<<std::endl;
+      //      }
       Scatter_plane_simple (ret,recv_buf,dimension,x,cbmask);
     }
   }
@@ -240,7 +243,7 @@ template<class vobj> void  Cshift_comms_simd(Lattice<vobj> &ret,const Lattice<vo
 			     (void *)&recv_buf_extract[i][0],
 			     recv_from_rank,
 			     bytes);
-	grid->Barrier();
+
 	rpointers[i] = &recv_buf_extract[i][0];
       } else { 
 	rpointers[i] = &send_buf_extract[nbr_lane][0];

lib/lattice/Lattice_arith.h

@@ -39,7 +39,8 @@ namespace Grid {
     ret.checkerboard = lhs.checkerboard;
     conformable(ret,rhs);
     conformable(lhs,rhs);
-    parallel_for(int ss=0;ss<lhs._grid->oSites();ss++){
+PARALLEL_FOR_LOOP
+    for(int ss=0;ss<lhs._grid->oSites();ss++){
 #ifdef STREAMING_STORES
       obj1 tmp;
       mult(&tmp,&lhs._odata[ss],&rhs._odata[ss]);
@@ -55,7 +56,8 @@ namespace Grid {
     ret.checkerboard = lhs.checkerboard;
     conformable(ret,rhs);
     conformable(lhs,rhs);
-    parallel_for(int ss=0;ss<lhs._grid->oSites();ss++){
+PARALLEL_FOR_LOOP
+    for(int ss=0;ss<lhs._grid->oSites();ss++){
 #ifdef STREAMING_STORES
       obj1 tmp;
       mac(&tmp,&lhs._odata[ss],&rhs._odata[ss]);
@@ -71,7 +73,8 @@ namespace Grid {
     ret.checkerboard = lhs.checkerboard;
     conformable(ret,rhs);
     conformable(lhs,rhs);
-    parallel_for(int ss=0;ss<lhs._grid->oSites();ss++){
+PARALLEL_FOR_LOOP
+    for(int ss=0;ss<lhs._grid->oSites();ss++){
 #ifdef STREAMING_STORES
       obj1 tmp;
       sub(&tmp,&lhs._odata[ss],&rhs._odata[ss]);
@@ -86,7 +89,8 @@ namespace Grid {
     ret.checkerboard = lhs.checkerboard;
     conformable(ret,rhs);
     conformable(lhs,rhs);
-    parallel_for(int ss=0;ss<lhs._grid->oSites();ss++){
+PARALLEL_FOR_LOOP
+    for(int ss=0;ss<lhs._grid->oSites();ss++){
 #ifdef STREAMING_STORES
       obj1 tmp;
       add(&tmp,&lhs._odata[ss],&rhs._odata[ss]);
@@ -104,7 +108,8 @@ namespace Grid {
     void mult(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
     ret.checkerboard = lhs.checkerboard;
     conformable(lhs,ret);
-    parallel_for(int ss=0;ss<lhs._grid->oSites();ss++){
+PARALLEL_FOR_LOOP
+    for(int ss=0;ss<lhs._grid->oSites();ss++){
       obj1 tmp;
       mult(&tmp,&lhs._odata[ss],&rhs);
       vstream(ret._odata[ss],tmp);
@@ -115,7 +120,8 @@ namespace Grid {
     void mac(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
     ret.checkerboard = lhs.checkerboard;
     conformable(ret,lhs);
-    parallel_for(int ss=0;ss<lhs._grid->oSites();ss++){
+PARALLEL_FOR_LOOP
+    for(int ss=0;ss<lhs._grid->oSites();ss++){
       obj1 tmp;
       mac(&tmp,&lhs._odata[ss],&rhs);
       vstream(ret._odata[ss],tmp);
@@ -126,7 +132,8 @@ namespace Grid {
     void sub(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
     ret.checkerboard = lhs.checkerboard;
     conformable(ret,lhs);
-    parallel_for(int ss=0;ss<lhs._grid->oSites();ss++){
+PARALLEL_FOR_LOOP
+    for(int ss=0;ss<lhs._grid->oSites();ss++){
 #ifdef STREAMING_STORES
       obj1 tmp;
       sub(&tmp,&lhs._odata[ss],&rhs);
@@ -140,7 +147,8 @@ namespace Grid {
     void add(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
     ret.checkerboard = lhs.checkerboard;
     conformable(lhs,ret);
-    parallel_for(int ss=0;ss<lhs._grid->oSites();ss++){
+PARALLEL_FOR_LOOP
+    for(int ss=0;ss<lhs._grid->oSites();ss++){
 #ifdef STREAMING_STORES
       obj1 tmp;
       add(&tmp,&lhs._odata[ss],&rhs);
@@ -158,7 +166,8 @@ namespace Grid {
     void mult(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
     ret.checkerboard = rhs.checkerboard;
     conformable(ret,rhs);
-    parallel_for(int ss=0;ss<rhs._grid->oSites();ss++){
+PARALLEL_FOR_LOOP
+    for(int ss=0;ss<rhs._grid->oSites();ss++){
 #ifdef STREAMING_STORES
       obj1 tmp;
       mult(&tmp,&lhs,&rhs._odata[ss]);
@@ -173,7 +182,8 @@ namespace Grid {
     void mac(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
     ret.checkerboard = rhs.checkerboard;
     conformable(ret,rhs);
-    parallel_for(int ss=0;ss<rhs._grid->oSites();ss++){
+PARALLEL_FOR_LOOP
+    for(int ss=0;ss<rhs._grid->oSites();ss++){
 #ifdef STREAMING_STORES
       obj1 tmp;
       mac(&tmp,&lhs,&rhs._odata[ss]);
@@ -188,7 +198,8 @@ namespace Grid {
     void sub(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
     ret.checkerboard = rhs.checkerboard;
     conformable(ret,rhs);
-    parallel_for(int ss=0;ss<rhs._grid->oSites();ss++){
+PARALLEL_FOR_LOOP
+    for(int ss=0;ss<rhs._grid->oSites();ss++){
 #ifdef STREAMING_STORES
       obj1 tmp;
       sub(&tmp,&lhs,&rhs._odata[ss]);
@@ -202,7 +213,8 @@ namespace Grid {
     void add(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
     ret.checkerboard = rhs.checkerboard;
     conformable(ret,rhs);
-    parallel_for(int ss=0;ss<rhs._grid->oSites();ss++){
+PARALLEL_FOR_LOOP
+    for(int ss=0;ss<rhs._grid->oSites();ss++){
 #ifdef STREAMING_STORES
       obj1 tmp;
       add(&tmp,&lhs,&rhs._odata[ss]);
@@ -218,7 +230,8 @@ namespace Grid {
     ret.checkerboard = x.checkerboard;
     conformable(ret,x);
     conformable(x,y);
-    parallel_for(int ss=0;ss<x._grid->oSites();ss++){
+PARALLEL_FOR_LOOP
+    for(int ss=0;ss<x._grid->oSites();ss++){
 #ifdef STREAMING_STORES
       vobj tmp = a*x._odata[ss]+y._odata[ss];
       vstream(ret._odata[ss],tmp);
@@ -232,7 +245,8 @@ namespace Grid {
     ret.checkerboard = x.checkerboard;
     conformable(ret,x);
     conformable(x,y);
-    parallel_for(int ss=0;ss<x._grid->oSites();ss++){
+PARALLEL_FOR_LOOP
+    for(int ss=0;ss<x._grid->oSites();ss++){
 #ifdef STREAMING_STORES
       vobj tmp = a*x._odata[ss]+b*y._odata[ss];
       vstream(ret._odata[ss],tmp);