Const correctness patch

by Christoph. Differs from the one by Rudy in BFM since it vectorises the twisted
4D solves in pairs.

.gitignore

@@ -83,6 +83,7 @@ ltmain.sh
 .Trashes
 ehthumbs.db
 Thumbs.db
+.dirstamp
 
 # build directory #
 ###################
@@ -97,11 +98,8 @@ build.sh
 
 # Eigen source #
 ################
-lib/Eigen/*
-
-# FFTW source #
-################
-lib/fftw/*
+Grid/Eigen
+Eigen/*
 
 # libtool macros #
 ##################
@@ -112,21 +110,7 @@ m4/libtool.m4
 ################
 gh-pages/
 
-# Buck files #
-##############
-.buck*
-buck-out
-BUCK
-make-bin-BUCK.sh
-
 # generated sources #
 #####################
-lib/qcd/spin/gamma-gen/*.h
-lib/qcd/spin/gamma-gen/*.cc
-lib/version.h
-
-# vs code editor files #
-########################
-.vscode/
-.vscode/settings.json
-settings.json
+Grid/qcd/spin/gamma-gen/*.h
+Grid/qcd/spin/gamma-gen/*.cc

.travis.yml

@@ -9,6 +9,11 @@ matrix:
     - os:        osx
       osx_image: xcode8.3
       compiler: clang
+      env: PREC=single
+    - os:        osx
+      osx_image: xcode8.3
+      compiler: clang
+      env: PREC=double
       
 before_install:
     - export GRIDDIR=`pwd`
@@ -16,7 +21,7 @@ before_install:
     - if [[ "$TRAVIS_OS_NAME" == "linux" ]] && [[ "$CC" == "clang" ]]; then export PATH="${GRIDDIR}/clang/bin:${PATH}"; fi
     - 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 libmpc openssl; fi
     
 install:
     - export CWD=`pwd`
@@ -33,6 +38,7 @@ install:
     - which $CXX
     - $CXX --version
     - if [[ "$TRAVIS_OS_NAME" == "osx" ]]; then export LDFLAGS='-L/usr/local/lib'; fi
+    - if [[ "$TRAVIS_OS_NAME" == "osx" ]]; then export EXTRACONF='--with-openssl=/usr/local/opt/openssl'; fi
     
 script:
     - ./bootstrap.sh
@@ -49,12 +55,7 @@ script:
     - make -j4
     - make install
     - cd $CWD/build
-    - ../configure --enable-precision=single --enable-simd=SSE4 --enable-comms=none --with-lime=$CWD/build/lime/install
+    - ../configure --enable-precision=$PREC --enable-simd=SSE4 --enable-comms=none --with-lime=$CWD/build/lime/install ${EXTRACONF}
     - make -j4 
     - ./benchmarks/Benchmark_dwf --threads 1 --debug-signals
-    - echo make clean
-    - ../configure --enable-precision=double --enable-simd=SSE4 --enable-comms=none --with-lime=$CWD/build/lime/install
-    - make -j4
-    - ./benchmarks/Benchmark_dwf --threads 1 --debug-signals
     - make check
-

Grid/GridCore.h

@@ -48,6 +48,7 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 #include <Grid/serialisation/Serialisation.h>
 #include <Grid/threads/Threads.h>
 #include <Grid/util/Util.h>
+#include <Grid/util/Sha.h>
 #include <Grid/communicator/Communicator.h> 
 #include <Grid/cartesian/Cartesian.h>    
 #include <Grid/tensors/Tensors.h>      

Grid/Grid_Eigen_Dense.h

@@ -1,4 +1,9 @@
 #pragma once
+// Force Eigen to use MKL if Grid has been configured with --enable-mkl
+#ifdef USE_MKL
+#define EIGEN_USE_MKL_ALL
+#endif
+
 #if defined __GNUC__
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wdeprecated-declarations"

Grid/Makefile.am

@@ -21,6 +21,32 @@ if BUILD_HDF5
   extra_headers+=serialisation/Hdf5Type.h
 endif
 
+all: version-cache
+
+version-cache:
+	@if [ `git status --porcelain | grep -v '??' | wc -l` -gt 0 ]; then\
+		a="uncommited changes";\
+	else\
+		a="clean";\
+	fi;\
+	echo "`git log -n 1 --format=format:"#define GITHASH \\"%H:%d $$a\\"%n" HEAD`" > vertmp;\
+	if [ -e version-cache ]; then\
+		d=`diff vertmp version-cache`;\
+		if [ "$${d}" != "" ]; then\
+			mv vertmp version-cache;\
+			rm -f Version.h;\
+		fi;\
+	else\
+		mv vertmp version-cache;\
+		rm -f Version.h;\
+	fi;\
+	rm -f vertmp
+
+Version.h:
+	cp version-cache Version.h
+
+.PHONY: version-cache
+
 #
 # Libraries
 #
@@ -30,8 +56,8 @@ include Eigen.inc
 lib_LIBRARIES = libGrid.a
 
 CCFILES += $(extra_sources)
-HFILES  += $(extra_headers)
+HFILES  += $(extra_headers) Config.h Version.h
 
 libGrid_a_SOURCES              = $(CCFILES)
-libGrid_adir                   = $(pkgincludedir)
-nobase_dist_pkginclude_HEADERS = $(HFILES) $(eigen_files) Config.h
+libGrid_adir                   = $(includedir)/Grid
+nobase_dist_pkginclude_HEADERS = $(HFILES) $(eigen_files) $(eigen_unsupp_files)

Grid/algorithms/iterative/BlockConjugateGradient.h

@@ -33,7 +33,7 @@ directory
 
 namespace Grid {
 
-enum BlockCGtype { BlockCG, BlockCGrQ, CGmultiRHS, BlockCGVec };
+enum BlockCGtype { BlockCG, BlockCGrQ, CGmultiRHS };
 
 //////////////////////////////////////////////////////////////////////////
 // Block conjugate gradient. Dimension zero should be the block direction
@@ -54,10 +54,9 @@ class BlockConjugateGradient : public OperatorFunction<Field> {
   RealD Tolerance;
   Integer MaxIterations;
   Integer IterationsToComplete; //Number of iterations the CG took to finish. Filled in upon completion
-  Integer PrintInterval; //GridLogMessages or Iterative
   
   BlockConjugateGradient(BlockCGtype cgtype,int _Orthog,RealD tol, Integer maxit, bool err_on_no_conv = true)
-    : Tolerance(tol), CGtype(cgtype),   blockDim(_Orthog),  MaxIterations(maxit), ErrorOnNoConverge(err_on_no_conv),PrintInterval(100)
+    : Tolerance(tol), CGtype(cgtype),   blockDim(_Orthog),  MaxIterations(maxit), ErrorOnNoConverge(err_on_no_conv)
   {};
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////
@@ -128,14 +127,6 @@ void operator()(LinearOperatorBase<Field> &Linop, const Field &Src, Field &Psi)
     assert(0);
   }
 }
-void operator()(LinearOperatorBase<Field> &Linop, const std::vector<Field> &Src, std::vector<Field> &Psi) 
-{
-  if ( CGtype == BlockCGVec ) {
-    BlockCGVecsolve(Linop,Src,Psi);
-  } else {
-    assert(0);
-  }
-}
 
 ////////////////////////////////////////////////////////////////////////////
 // BlockCGrQ implementation:
@@ -609,272 +600,6 @@ void CGmultiRHSsolve(LinearOperatorBase<Field> &Linop, const Field &Src, Field &
   IterationsToComplete = k;
 }
 
-void InnerProductMatrix(Eigen::MatrixXcd &m , const std::vector<Field> &X, std::vector<Field> &Y){
-  for(int b=0;b<Nblock;b++)
-  for(int bp=0;bp<Nblock;bp++) {
-    m(b,bp) = innerProduct(X[b],Y[bp]);  
-    }
-}
-double HermCheck( Eigen::MatrixXcd &m,  const std::string &str, int ForceHerm=1 , int Print = 0) {
-  for(int b=0;b<Nblock;b++)
-  for(int bp=0;bp<=b;bp++) {
-    if(Print)
-    std::cout<<GridLogMessage << "HermCheck "<<str<<" "<<b<<" "<<bp<<" : "<< m(b,bp) <<" "<<conj(m(bp,b))<<" " <<m(b,bp)-conj(m(bp,b)) <<std::endl;
-    if(ForceHerm){
-      if(b==bp) m(b,b) = real(m(b,b));
-      else{
-        auto temp = 0.5*(m(b,bp)+conj(m(bp,b)));
-	m(b,bp) = temp;
-        m(bp,b) = conj(temp);
-      }
-    }
-  }
-}
-
-void BlockCGVecsolve(LinearOperatorBase<Field> &Linop, const std::vector<Field> &Src, std::vector<Field> &Psi) 
-{
-//  int Orthog = blockDim; // First dimension is block dim; this is an assumption
-//  Nblock = Src._grid->_fdimensions[Orthog];
-  Nblock = Src.size();
-  assert(Nblock == Psi.size());
-
-  std::cout<<GridLogMessage<<" Block Conjugate Gradient :  Nblock "<<Nblock<<std::endl;
-
-  for(int b=0;b<Nblock;b++){
-  Psi[b].checkerboard = Src[0].checkerboard;
-  conformable(Psi[b], Src[b]);
-  }
-
-  Field Fake(Src[0]);
-
-  std::vector<Field> P(Nblock,Fake);
-// P.resize(Nblock);
-  std::vector<Field> AP(Nblock,Fake); 
-//AP.resize(Nblock);
-  std::vector<Field> R(Nblock,Fake); 
-  std::vector<Field> TMP(Nblock,Fake); 
-//R.resize(Nblock);
-  
-  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);
-  for(int b=0;b<Nblock;b++){ ssq[b] = norm2(Src[b]);}
-  RealD sssum=0;
-  for(int b=0;b<Nblock;b++) sssum+=ssq[b];
-
-//  sliceNorm(residuals,Src,Orthog);
-  for(int b=0;b<Nblock;b++){ residuals[b] = norm2(Src[b]);}
-  for(int b=0;b<Nblock;b++){ assert(std::isnan(residuals[b])==0); }
-
-//  sliceNorm(residuals,Psi,Orthog);
-  for(int b=0;b<Nblock;b++){ residuals[b] = norm2(Psi[b]);}
-  for(int b=0;b<Nblock;b++){ assert(std::isnan(residuals[b])==0); }
-
-  // Initial search dir is guess
-  for(int b=0;b<Nblock;b++) Linop.HermOp(Psi[b], AP[b]);
-  for(int b=0;b<Nblock;b++) 
-  std::cout << b << " Psi " << norm2(Psi[b]) <<" AP "<<norm2(AP[b])<<std::endl;
-  
-
-  /************************************************************************
-   * 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
-   */
-
-  for(int b=0;b<Nblock;b++){
-  R[b] = Src[b] - AP[b]; //R_0  
-  P[b] = R[b];  // P_1
-  }
-//  sliceInnerProductMatrix(m_rr,R,R,Orthog);
-  InnerProductMatrix(m_rr,R,R);
-  HermCheck(m_rr, "R_0 R_0",1,1);
-  HermCheck(m_rr, "R_0 R_0",0,1);
-#if 0
-  for(int b=0;b<Nblock;b++)
-  for(int bp=0;bp<Nblock;bp++) {
-    m_rr(b,bp) = innerProduct(R[b],R[bp]);  
-    std::cout << 0 <<" : R_0 R_0 "<< b <<" "<<bp<<" "<<innerProduct(R[b],R[bp]) <<std::endl;
-  }
-#endif
-
-  GridStopWatch sliceInnerTimer;
-  GridStopWatch sliceMaddTimer;
-  GridStopWatch MatrixTimer;
-  GridStopWatch SolverTimer;
-  SolverTimer.Start();
-
-  int k;
-  int if_print =0;
-  for (k = 1; k <= MaxIterations; k++) {
-
-    RealD rrsum=0;
-    for(int b=0;b<Nblock;b++) rrsum+=real(m_rr(b,b));
-
-    if(PrintInterval && (k%PrintInterval)==0){  
-	if_print=1;
-       std::cout << GridLogMessage << "\titeration "<<k<<" rr_sum "<<rrsum<<" ssq_sum "<< sssum
-	      <<" / "<<std::sqrt(rrsum/sssum) <<std::endl;
-    } else {
-    if_print=0;
-    std::cout << GridLogIterative << "\titeration "<<k<<" rr_sum "<<rrsum<<" ssq_sum "<< sssum
-	      <<" / "<<std::sqrt(rrsum/sssum) <<std::endl;
-    }
-
-    MatrixTimer.Start();
-    for(int b=0;b<Nblock;b++) Linop.HermOp(P[b], AP[b]);
-    MatrixTimer.Stop();
-
-    // Alpha
-    sliceInnerTimer.Start();
- //   sliceInnerProductMatrix(m_pAp,P,AP,Orthog);
-  InnerProductMatrix(m_pAp,P,AP);
-  HermCheck(m_pAp, "P AP",1,if_print);
-  if(if_print) HermCheck(m_pAp, "P AP",0,if_print);
-#if 0
-  for(int b=0;b<Nblock;b++)
-  for(int bp=0;bp<Nblock;bp++) {
-    m_pAp(b,bp) = innerProduct(P[b],AP[bp]);  
-    std::cout << k <<" : m_pAp "<< b <<" "<<bp<<" "<<innerProduct(P[b],AP[bp]) <<std::endl;
-  }
-#endif
-    sliceInnerTimer.Stop();
-    m_pAp_inv = m_pAp.inverse();
-  HermCheck(m_pAp_inv, "inv (P AP)",1,if_print);
-  if(if_print) HermCheck(m_pAp_inv, "inv (P AP)",0,if_print);
-if(if_print)
-{
-    m_alpha = m_pAp*m_pAp_inv;
-  for(int b=0;b<Nblock;b++){
-  for(int bp=0;bp<Nblock;bp++) {
-    std::cout << k <<" : pAp*pAp_inv "<< b <<" "<<bp<<" "<<m_alpha(b,bp)<<std::endl;
-  }
-  }
-}
-    m_alpha   = m_pAp_inv * m_rr ; //alpha_k+1 = (P_k+1^t A P_k+1)^-1 (R_k^t R_k)
-
-    // Psi, R update
-    sliceMaddTimer.Start();
-//    sliceMaddMatrix(Psi,m_alpha, P,Psi,Orthog);     // X_k+1=X_k+P_k+1 alpha_k+1
-  for(int b=0;b<Nblock;b++)
-  for(int bp=0;bp<Nblock;bp++) {
-    Psi[b] += m_alpha(bp,b)*P[bp];  // X_k+1 = X_k + P_k+1 alpha_k+1
-  }
-
-  for(int b=0;b<Nblock;b++) TMP[b] = R[b];
-//    sliceMaddMatrix(R  ,m_alpha,AP,  R,Orthog,-1.0);// sub alpha * AP to resid
-  for(int b=0;b<Nblock;b++)
-  for(int bp=0;bp<Nblock;bp++) {
-    R[b] -= m_alpha(bp,b)*AP[bp];  // R_k+1 = R_k - AP_k+1 alpha_k+1
-  }
-    sliceMaddTimer.Stop();
-if(if_print)
-{
-//check
-  for(int b=0;b<Nblock;b++){
-  for(int bp=0;bp<Nblock;bp++) {
-    std::cout << k <<" : R_k+1 R_k "<< b <<" "<<bp<<" "<<innerProduct(R[b],TMP[bp]) <<std::endl;
-    std::cout << k <<" : R_k R_k "<< b <<" "<<bp<<" "<<innerProduct(TMP[b],TMP[bp]) <<std::endl;
-  }
-  }
-}
-
-    // Beta
-    m_rr_inv = m_rr.inverse(); //m_rr_inv = (R_k^t R_k)^-1
-    HermCheck(m_rr_inv,"m_rr_inv",1,if_print);
-    if(if_print) HermCheck(m_rr_inv,"m_rr_inv",0,if_print);
-    sliceInnerTimer.Start();
-//    sliceInnerProductMatrix(m_rr,R,R,Orthog);
-    InnerProductMatrix(m_rr,R,R);
-  HermCheck(m_rr,"m_rr",1,if_print);
-  if(if_print) HermCheck(m_rr,"m_rr",0,if_print);
-    sliceInnerTimer.Stop();
-    m_beta = m_rr_inv *m_rr; // beta_k+2 = (R_k^t R_k)^-1 (R_k+1^5 R_k+1)
-//  HermCheck(m_beta,"m_beta");
-
-    // Search update
-    sliceMaddTimer.Start();
-//    sliceMaddMatrix(AP,m_beta,P,R,Orthog);
-  for(int b=0;b<Nblock;b++){
-    AP[b] = R[b];
-  for(int bp=0;bp<Nblock;bp++) {
-    AP[b] += m_beta(bp,b)*P[bp]; //AP = R_k+1 + P_k+1 beta_k+1
-  }
-  }
-if(if_print)
-{
-//check
-    for(int b=0;b<Nblock;b++) Linop.HermOp(P[b], TMP[b]);
-  for(int b=0;b<Nblock;b++){
-  for(int bp=0;bp<Nblock;bp++) {
-    std::cout << k <<" : P_k+2 A P "<< b <<" "<<bp<<" "<<innerProduct(AP[b],TMP[bp]) <<std::endl;
-  }
-  }
-}
-    sliceMaddTimer.Stop();
-  for(int b=0;b<Nblock;b++) P[b]= AP[b]; //P_k+2 = AP
-
-    /*********************
-     * convergence monitor
-     *********************
-     */
-    RealD max_resid=0;
-    RealD rr;
-    for(int b=0;b<Nblock;b++){
-      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<<" computed resid "
-		  << std::sqrt(real(m_rr(b,b))/ssq[b])<<std::endl;
-      }
-	      std::cout << GridLogMessage<<"\tMax residual is "<<std::sqrt(max_resid)<<std::endl;
-
-	  for(int b=0;b<Nblock;b++) { 
-	      Linop.HermOp(Psi[b], AP[b]);
-	AP[b] = AP[b]-Src[b];
-	      std::cout << GridLogMessage <<"\t True residual is " << b<<" "<<std::sqrt(norm2(AP[b])/norm2(Src[b])) <<std::endl;
-}
-
-      std::cout << GridLogMessage << "Time Breakdown "<<std::endl;
-      std::cout << GridLogMessage << "\tElapsed    " << SolverTimer.Elapsed()     <<std::endl;
-      std::cout << GridLogMessage << "\tMatrix     " << MatrixTimer.Elapsed()     <<std::endl;
-      std::cout << GridLogMessage << "\tInnerProd  " << sliceInnerTimer.Elapsed() <<std::endl;
-      std::cout << GridLogMessage << "\tMaddMatrix " << sliceMaddTimer.Elapsed()  <<std::endl;
-	    
-      IterationsToComplete = k;
-      return;
-    }
-
-  }
-  std::cout << GridLogMessage << "BlockConjugateGradient did NOT converge" << std::endl;
-
-  if (ErrorOnNoConverge) assert(0);
-  IterationsToComplete = k;
-}
-
 };
 
 }

Grid/algorithms/iterative/ConjugateGradient.h

@@ -150,13 +150,13 @@ class ConjugateGradient : public OperatorFunction<Field> {
 	std::cout << GridLogMessage << "\tTrue residual " << true_residual<<std::endl;
 	std::cout << GridLogMessage << "\tTarget " << Tolerance << 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 << "\tLinalg     " << LinalgTimer.Elapsed() <<std::endl;
-	std::cout << GridLogMessage << "\tInner      " << InnerTimer.Elapsed() <<std::endl;
-	std::cout << GridLogMessage << "\tAxpyNorm   " << AxpyNormTimer.Elapsed() <<std::endl;
-	std::cout << GridLogMessage << "\tLinearComb " << LinearCombTimer.Elapsed() <<std::endl;
+        std::cout << GridLogPerformance << "Time breakdown "<<std::endl;
+	std::cout << GridLogPerformance << "\tElapsed    " << SolverTimer.Elapsed() <<std::endl;
+	std::cout << GridLogPerformance << "\tMatrix     " << MatrixTimer.Elapsed() <<std::endl;
+	std::cout << GridLogPerformance << "\tLinalg     " << LinalgTimer.Elapsed() <<std::endl;
+	std::cout << GridLogPerformance << "\tInner      " << InnerTimer.Elapsed() <<std::endl;
+	std::cout << GridLogPerformance << "\tAxpyNorm   " << AxpyNormTimer.Elapsed() <<std::endl;
+	std::cout << GridLogPerformance << "\tLinearComb " << LinearCombTimer.Elapsed() <<std::endl;
 
         if (ErrorOnNoConverge) assert(true_residual / Tolerance < 10000.0);
 

Grid/algorithms/iterative/Deflation.h

@@ -30,22 +30,23 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 
 namespace Grid { 
 
-struct ZeroGuesser {
+template<class Field>
+class ZeroGuesser: public LinearFunction<Field> {
 public:
-  template<class Field> 
-  void operator()(const Field &src,Field &guess) { guess = Zero(); };
+  virtual void operator()(const Field &src, Field &guess) { guess = zero; };
 };
-struct SourceGuesser {
+
+template<class Field>
+class SourceGuesser: public LinearFunction<Field> {
 public:
-  template<class Field> 
-  void operator()(const Field &src,Field &guess) { guess = src; };
+  virtual void operator()(const Field &src, Field &guess) { guess = src; };
 };
 
 ////////////////////////////////
 // Fine grid deflation
 ////////////////////////////////
 template<class Field>
-struct DeflatedGuesser {
+class DeflatedGuesser: public LinearFunction<Field> {
 private:
   const std::vector<Field> &evec;
   const std::vector<RealD> &eval;
@@ -54,7 +55,7 @@ public:
 
   DeflatedGuesser(const std::vector<Field> & _evec,const std::vector<RealD> & _eval) : evec(_evec), eval(_eval) {};
 
-  void operator()(const Field &src,Field &guess) { 
+  virtual void operator()(const Field &src,Field &guess) {
     guess = zero;
     assert(evec.size()==eval.size());
     auto N = evec.size();
@@ -62,11 +63,12 @@ public:
       const Field& tmp = evec[i];
       axpy(guess,TensorRemove(innerProduct(tmp,src)) / eval[i],tmp,guess);
     }
+    guess.checkerboard = src.checkerboard;
   }
 };
 
 template<class FineField, class CoarseField>
-class LocalCoherenceDeflatedGuesser {
+class LocalCoherenceDeflatedGuesser: public LinearFunction<FineField> {
 private:
   const std::vector<FineField>   &subspace;
   const std::vector<CoarseField> &evec_coarse;
@@ -92,6 +94,7 @@ public:
       axpy(guess_coarse,TensorRemove(innerProduct(tmp,src_coarse)) / eval_coarse[i],tmp,guess_coarse);
     }
     blockPromote(guess_coarse,guess,subspace);
+    guess.checkerboard = src.checkerboard;
   };
 };
 

Grid/algorithms/iterative/ImplicitlyRestartedLanczos.h

@@ -57,8 +57,9 @@ void basisRotate(std::vector<Field> &basis,Eigen::MatrixXd& Qt,int j0, int j1, i
       
   parallel_region
   {
-    std::vector < vobj > B(Nm); // Thread private
-        
+
+    std::vector < vobj , commAllocator<vobj> > B(Nm); // Thread private
+       
     parallel_for_internal(int ss=0;ss < grid->oSites();ss++){
       for(int j=j0; j<j1; ++j) B[j]=0.;
       

Grid/algorithms/iterative/LocalCoherenceLanczos.h

@@ -285,9 +285,11 @@ public:
   };
 
   void Orthogonalise(void ) {
-    CoarseScalar InnerProd(_CoarseGrid); 
-    blockOrthogonalise(InnerProd,subspace);std::cout << GridLogMessage <<" Gramm-Schmidt pass 1"<<std::endl;
-    blockOrthogonalise(InnerProd,subspace);std::cout << GridLogMessage <<" Gramm-Schmidt pass 2"<<std::endl;
+    CoarseScalar InnerProd(_CoarseGrid);
+    std::cout << GridLogMessage <<" Gramm-Schmidt pass 1"<<std::endl;
+    blockOrthogonalise(InnerProd,subspace);
+    std::cout << GridLogMessage <<" Gramm-Schmidt pass 2"<<std::endl;
+    blockOrthogonalise(InnerProd,subspace);
   };
 
   template<typename T>  static RealD normalise(T& v) 
@@ -333,7 +335,7 @@ public:
     // create a smoother and see if we can get a cheap convergence test and smooth inside the IRL
     //////////////////////////////////////////////////////////////////////////////////////////////////
     Chebyshev<FineField>                          ChebySmooth(cheby_smooth);
-    ProjectedFunctionHermOp<Fobj,CComplex,nbasis> ChebyOp (ChebySmooth,_FineOp,_subspace);
+    ProjectedFunctionHermOp<Fobj,CComplex,nbasis> ChebyOp (ChebySmooth,_FineOp,subspace);
     ImplicitlyRestartedLanczosSmoothedTester<Fobj,CComplex,nbasis> ChebySmoothTester(ChebyOp,ChebySmooth,_FineOp,subspace,relax);
 
     for(int k=0;k<evec_coarse.size();k++){
@@ -374,14 +376,14 @@ public:
 		  RealD MaxIt, RealD betastp, int MinRes)
   {
     Chebyshev<FineField>                          Cheby(cheby_op);
-    ProjectedHermOp<Fobj,CComplex,nbasis>         Op(_FineOp,_subspace);
-    ProjectedFunctionHermOp<Fobj,CComplex,nbasis> ChebyOp (Cheby,_FineOp,_subspace);
+    ProjectedHermOp<Fobj,CComplex,nbasis>         Op(_FineOp,subspace);
+    ProjectedFunctionHermOp<Fobj,CComplex,nbasis> ChebyOp (Cheby,_FineOp,subspace);
     //////////////////////////////////////////////////////////////////////////////////////////////////
     // create a smoother and see if we can get a cheap convergence test and smooth inside the IRL
     //////////////////////////////////////////////////////////////////////////////////////////////////
 
     Chebyshev<FineField>                                           ChebySmooth(cheby_smooth);
-    ImplicitlyRestartedLanczosSmoothedTester<Fobj,CComplex,nbasis> ChebySmoothTester(ChebyOp,ChebySmooth,_FineOp,_subspace,relax);
+    ImplicitlyRestartedLanczosSmoothedTester<Fobj,CComplex,nbasis> ChebySmoothTester(ChebyOp,ChebySmooth,_FineOp,subspace,relax);
 
     evals_coarse.resize(Nm);
     evec_coarse.resize(Nm,_CoarseGrid);

Grid/algorithms/iterative/SchurRedBlack.h

@@ -95,20 +95,30 @@ namespace Grid {
   private:
     OperatorFunction<Field> & _HermitianRBSolver;
     int CBfactorise;
+    bool subGuess;
   public:
 
     /////////////////////////////////////////////////////
     // Wrap the usual normal equations Schur trick
     /////////////////////////////////////////////////////
-  SchurRedBlackStaggeredSolve(OperatorFunction<Field> &HermitianRBSolver)  :
+  SchurRedBlackStaggeredSolve(OperatorFunction<Field> &HermitianRBSolver, const bool initSubGuess = false)  :
      _HermitianRBSolver(HermitianRBSolver) 
     { 
       CBfactorise=0;
+      subtractGuess(initSubGuess);
     };
+    void subtractGuess(const bool initSubGuess)
+    {
+      subGuess = initSubGuess;
+    }
+    bool isSubtractGuess(void)
+    {
+      return subGuess;
+    }
 
     template<class Matrix>
     void operator() (Matrix & _Matrix,const Field &in, Field &out){
-      ZeroGuesser guess;
+      ZeroGuesser<Field> guess;
       (*this)(_Matrix,in,out,guess);
     }
     template<class Matrix, class Guesser>
@@ -150,9 +160,12 @@ namespace Grid {
       // Call the red-black solver
       //////////////////////////////////////////////////////////////
       std::cout<<GridLogMessage << "SchurRedBlackStaggeredSolver calling the Mpc solver" <<std::endl;
-      guess(src_o,sol_o);
+      guess(src_o, sol_o);
+      Mtmp = sol_o;
       _HermitianRBSolver(_HermOpEO,src_o,sol_o);  assert(sol_o.checkerboard==Odd);
       std::cout<<GridLogMessage << "SchurRedBlackStaggeredSolver called  the Mpc solver" <<std::endl;
+      // Fionn A2A boolean behavioural control
+      if (subGuess)        sol_o = sol_o-Mtmp;
 
       ///////////////////////////////////////////////////
       // sol_e = M_ee^-1 * ( src_e - Meo sol_o )...
@@ -167,11 +180,15 @@ namespace Grid {
       std::cout<<GridLogMessage << "SchurRedBlackStaggeredSolver inserted solution" <<std::endl;
 
       // Verify the unprec residual
-      _Matrix.M(out,resid); 
-      resid = resid-in;
-      RealD ns = norm2(in);
-      RealD nr = norm2(resid);
-      std::cout<<GridLogMessage << "SchurRedBlackStaggered solver true unprec resid "<< std::sqrt(nr/ns) <<" nr "<< nr <<" ns "<<ns << std::endl;
+      if ( ! subGuess ) {
+        _Matrix.M(out,resid); 
+        resid = resid-in;
+        RealD ns = norm2(in);
+        RealD nr = norm2(resid);
+        std::cout<<GridLogMessage << "SchurRedBlackStaggered solver true unprec resid "<< std::sqrt(nr/ns) <<" nr "<< nr <<" ns "<<ns << std::endl;
+      } else {
+        std::cout << GridLogMessage << "Guess subtracted after solve." << std::endl;
+      }
     }     
   };
   template<class Field> using SchurRedBlackStagSolve = SchurRedBlackStaggeredSolve<Field>;
@@ -184,18 +201,28 @@ namespace Grid {
   private:
     OperatorFunction<Field> & _HermitianRBSolver;
     int CBfactorise;
+    bool subGuess;
   public:
 
     /////////////////////////////////////////////////////
     // Wrap the usual normal equations Schur trick
     /////////////////////////////////////////////////////
-  SchurRedBlackDiagMooeeSolve(OperatorFunction<Field> &HermitianRBSolver,int cb=0)  :  _HermitianRBSolver(HermitianRBSolver) 
+  SchurRedBlackDiagMooeeSolve(OperatorFunction<Field> &HermitianRBSolver,int cb=0, const bool initSubGuess = false)  :  _HermitianRBSolver(HermitianRBSolver) 
   { 
     CBfactorise=cb;
+    subtractGuess(initSubGuess);
   };
+    void subtractGuess(const bool initSubGuess)
+    {
+      subGuess = initSubGuess;
+    }
+    bool isSubtractGuess(void)
+    {
+      return subGuess;
+    }
     template<class Matrix>
     void operator() (Matrix & _Matrix,const Field &in, Field &out){
-      ZeroGuesser guess;
+      ZeroGuesser<Field> guess;
       (*this)(_Matrix,in,out,guess);
     }
     template<class Matrix, class Guesser>
@@ -236,7 +263,10 @@ namespace Grid {
       //////////////////////////////////////////////////////////////
       std::cout<<GridLogMessage << "SchurRedBlack solver calling the MpcDagMp solver" <<std::endl;
       guess(src_o,sol_o);
+      Mtmp = sol_o;
       _HermitianRBSolver(_HermOpEO,src_o,sol_o);  assert(sol_o.checkerboard==Odd);
+      // Fionn A2A boolean behavioural control
+      if (subGuess)        sol_o = sol_o-Mtmp;
 
       ///////////////////////////////////////////////////
       // sol_e = M_ee^-1 * ( src_e - Meo sol_o )...
@@ -249,12 +279,16 @@ namespace Grid {
       setCheckerboard(out,sol_o); assert(  sol_o.checkerboard ==Odd );
 
       // Verify the unprec residual
-      _Matrix.M(out,resid); 
-      resid = resid-in;
-      RealD ns = norm2(in);
-      RealD nr = norm2(resid);
+      if ( ! subGuess ) {
+        _Matrix.M(out,resid); 
+        resid = resid-in;
+        RealD ns = norm2(in);
+        RealD nr = norm2(resid);
 
-      std::cout<<GridLogMessage << "SchurRedBlackDiagMooee solver true unprec resid "<< std::sqrt(nr/ns) <<" nr "<< nr <<" ns "<<ns << std::endl;
+        std::cout<<GridLogMessage << "SchurRedBlackDiagMooee solver true unprec resid "<< std::sqrt(nr/ns) <<" nr "<< nr <<" ns "<<ns << std::endl;
+      } else {
+        std::cout << GridLogMessage << "Guess subtracted after solve." << std::endl;
+      }
     }     
   };
 
@@ -267,20 +301,30 @@ namespace Grid {
   private:
     OperatorFunction<Field> & _HermitianRBSolver;
     int CBfactorise;
+    bool subGuess;
   public:
 
     /////////////////////////////////////////////////////
     // Wrap the usual normal equations Schur trick
     /////////////////////////////////////////////////////
-  SchurRedBlackDiagTwoSolve(OperatorFunction<Field> &HermitianRBSolver)  :
+  SchurRedBlackDiagTwoSolve(OperatorFunction<Field> &HermitianRBSolver, const bool initSubGuess = false)  :
      _HermitianRBSolver(HermitianRBSolver) 
     { 
-      CBfactorise=0;
+      CBfactorise = 0;
+      subtractGuess(initSubGuess);
     };
+    void subtractGuess(const bool initSubGuess)
+    {
+      subGuess = initSubGuess;
+    }
+    bool isSubtractGuess(void)
+    {
+      return subGuess;
+    }
 
     template<class Matrix>
     void operator() (Matrix & _Matrix,const Field &in, Field &out){
-      ZeroGuesser guess;
+      ZeroGuesser<Field> guess;
       (*this)(_Matrix,in,out,guess);
     }
     template<class Matrix,class Guesser>
@@ -322,8 +366,11 @@ namespace Grid {
       std::cout<<GridLogMessage << "SchurRedBlack solver calling the MpcDagMp solver" <<std::endl;
 //      _HermitianRBSolver(_HermOpEO,src_o,sol_o);  assert(sol_o.checkerboard==Odd);
       guess(src_o,tmp);
+      Mtmp = tmp;
       _HermitianRBSolver(_HermOpEO,src_o,tmp);  assert(tmp.checkerboard==Odd);
-      _Matrix.MooeeInv(tmp,sol_o);        assert(  sol_o.checkerboard   ==Odd);
+      // Fionn A2A boolean behavioural control
+      if (subGuess)      tmp = tmp-Mtmp;
+      _Matrix.MooeeInv(tmp,sol_o);       assert(  sol_o.checkerboard   ==Odd);
 
       ///////////////////////////////////////////////////
       // sol_e = M_ee^-1 * ( src_e - Meo sol_o )...
@@ -336,12 +383,16 @@ namespace Grid {
       setCheckerboard(out,sol_o); assert(  sol_o.checkerboard ==Odd );
 
       // Verify the unprec residual
-      _Matrix.M(out,resid); 
-      resid = resid-in;
-      RealD ns = norm2(in);
-      RealD nr = norm2(resid);
+      if ( ! subGuess ) {
+        _Matrix.M(out,resid); 
+        resid = resid-in;
+        RealD ns = norm2(in);
+        RealD nr = norm2(resid);
 
-      std::cout<<GridLogMessage << "SchurRedBlackDiagTwo solver true unprec resid "<< std::sqrt(nr/ns) <<" nr "<< nr <<" ns "<<ns << std::endl;
+        std::cout<<GridLogMessage << "SchurRedBlackDiagTwo solver true unprec resid "<< std::sqrt(nr/ns) <<" nr "<< nr <<" ns "<<ns << std::endl;
+      } else {
+        std::cout << GridLogMessage << "Guess subtracted after solve." << std::endl;
+      }
     }     
   };
   ///////////////////////////////////////////////////////////////////////////////////////////////////////
@@ -352,20 +403,30 @@ namespace Grid {
   private:
     LinearFunction<Field> & _HermitianRBSolver;
     int CBfactorise;
+    bool subGuess;
   public:
 
     /////////////////////////////////////////////////////
     // Wrap the usual normal equations Schur trick
     /////////////////////////////////////////////////////
-  SchurRedBlackDiagTwoMixed(LinearFunction<Field> &HermitianRBSolver)  :
+  SchurRedBlackDiagTwoMixed(LinearFunction<Field> &HermitianRBSolver, const bool initSubGuess = false)  :
      _HermitianRBSolver(HermitianRBSolver) 
     { 
       CBfactorise=0;
+      subtractGuess(initSubGuess);
     };
+    void subtractGuess(const bool initSubGuess)
+    {
+      subGuess = initSubGuess;
+    }
+    bool isSubtractGuess(void)
+    {
+      return subGuess;
+    }
 
     template<class Matrix>
     void operator() (Matrix & _Matrix,const Field &in, Field &out){
-      ZeroGuesser guess;
+      ZeroGuesser<Field> guess;
       (*this)(_Matrix,in,out,guess);
     }
     template<class Matrix, class Guesser>
@@ -408,7 +469,10 @@ namespace Grid {
 //      _HermitianRBSolver(_HermOpEO,src_o,sol_o);  assert(sol_o.checkerboard==Odd);
 //      _HermitianRBSolver(_HermOpEO,src_o,tmp);  assert(tmp.checkerboard==Odd);
       guess(src_o,tmp);
-      _HermitianRBSolver(src_o,tmp);  assert(tmp.checkerboard==Odd);
+      Mtmp = tmp;
+      _HermitianRBSolver(_HermOpEO,src_o,tmp);  assert(tmp.checkerboard==Odd);
+      // Fionn A2A boolean behavioural control
+      if (subGuess)      tmp = tmp-Mtmp;
       _Matrix.MooeeInv(tmp,sol_o);        assert(  sol_o.checkerboard   ==Odd);
 
       ///////////////////////////////////////////////////
@@ -422,12 +486,16 @@ namespace Grid {
       setCheckerboard(out,sol_o); assert(  sol_o.checkerboard ==Odd );
 
       // Verify the unprec residual
-      _Matrix.M(out,resid); 
-      resid = resid-in;
-      RealD ns = norm2(in);
-      RealD nr = norm2(resid);
+      if ( ! subGuess ) {
+        _Matrix.M(out,resid); 
+        resid = resid-in;
+        RealD ns = norm2(in);
+        RealD nr = norm2(resid);
 
-      std::cout<<GridLogMessage << "SchurRedBlackDiagTwo solver true unprec resid "<< std::sqrt(nr/ns) <<" nr "<< nr <<" ns "<<ns << std::endl;
+        std::cout << GridLogMessage << "SchurRedBlackDiagTwo solver true unprec resid " << std::sqrt(nr / ns) << " nr " << nr << " ns " << ns << std::endl;
+      } else {
+        std::cout << GridLogMessage << "Guess subtracted after solve." << std::endl;
+      }
     }     
   };
 

Grid/cartesian/Cartesian_red_black.h

@@ -179,10 +179,6 @@ public:
       assert(checker_dim_mask.size() == _ndimension);
       assert(processor_grid.size() == _ndimension);
       assert(simd_layout.size() == _ndimension);
-         std::cout <<"dimensions processor_grid simd_layout checker_dim_mask"<<std::endl;
-      for(int i=0;i<_ndimension;i++){
-         std::cout <<i << " "<<dimensions[i]<<" "<<processor_grid[i]<<" "<< simd_layout[i]<<" "<< checker_dim_mask[i]<<std::endl;
-      }
 
       _fdimensions.resize(_ndimension);
       _gdimensions.resize(_ndimension);

Grid/communicator/SharedMemoryMPI.cc

@@ -132,7 +132,6 @@ int Log2Size(int TwoToPower,int MAXLOG2)
 }
 void GlobalSharedMemory::OptimalCommunicator(const std::vector<int> &processors,Grid_MPI_Comm & optimal_comm)
 {
-#undef HYPERCUBE 
 #ifdef HYPERCUBE
   ////////////////////////////////////////////////////////////////
   // Assert power of two shm_size.
@@ -175,7 +174,7 @@ void GlobalSharedMemory::OptimalCommunicator(const std::vector<int> &processors,
 
   std::string hname(name);
   std::cout << "hostname "<<hname<<std::endl;
-  std::cout << "R " << R << " I " << I << " N "<< N<<
+  std::cout << "R " << R << " I " << I << " N "<< N
             << " hypercoor 0x"<<std::hex<<hypercoor<<std::dec<<std::endl;
 
   //////////////////////////////////////////////////////////////////

Grid/lattice/Lattice_reduction.h

@@ -39,7 +39,7 @@ template<class vobj> inline RealD norm2(const Lattice<vobj> &arg){
 
 // Double inner product
 template<class vobj>
-inline ComplexD innerProduct(const Lattice<vobj> &left,const Lattice<vobj> &right) 
+inline ComplexD innerProduct(const Lattice<vobj> &left,const Lattice<vobj> &right)
 {
   typedef typename vobj::scalar_type scalar_type;
   typedef typename vobj::vector_typeD vector_type;
@@ -274,6 +274,115 @@ template<class vobj> inline void sliceSum(const Lattice<vobj> &Data,std::vector<
   }
 }
 
+template<class vobj>
+static void mySliceInnerProductVector( std::vector<ComplexD> & result, const Lattice<vobj> &lhs,const Lattice<vobj> &rhs,int orthogdim)
+{
+  // std::cout << GridLogMessage << "Start mySliceInnerProductVector" << std::endl;
+
+  typedef typename vobj::scalar_type scalar_type;
+  std::vector<scalar_type> lsSum;
+  localSliceInnerProductVector(result, lhs, rhs, lsSum, orthogdim);
+  globalSliceInnerProductVector(result, lhs, lsSum, orthogdim);
+  // std::cout << GridLogMessage << "End mySliceInnerProductVector" << std::endl;
+}
+
+template <class vobj>
+static void localSliceInnerProductVector(std::vector<ComplexD> &result, const Lattice<vobj> &lhs, const Lattice<vobj> &rhs, std::vector<typename vobj::scalar_type> &lsSum, int orthogdim)
+{
+  // std::cout << GridLogMessage << "Start prep" << std::endl;
+  typedef typename vobj::vector_type   vector_type;
+  typedef typename vobj::scalar_type   scalar_type;
+  GridBase  *grid = lhs._grid;
+  assert(grid!=NULL);
+  conformable(grid,rhs._grid);
+
+  const int    Nd = grid->_ndimension;
+  const int Nsimd = grid->Nsimd();
+
+  assert(orthogdim >= 0);
+  assert(orthogdim < Nd);
+
+  int fd=grid->_fdimensions[orthogdim];
+  int ld=grid->_ldimensions[orthogdim];
+  int rd=grid->_rdimensions[orthogdim];
+  // std::cout << GridLogMessage << "Start alloc" << std::endl;
+
+  std::vector<vector_type,alignedAllocator<vector_type> > lvSum(rd); // will locally sum vectors first
+  lsSum.resize(ld,scalar_type(0.0));                    // sum across these down to scalars
+  std::vector<iScalar<scalar_type>> extracted(Nsimd);   // splitting the SIMD  
+  // std::cout << GridLogMessage << "End alloc" << std::endl;
+
+  result.resize(fd); // And then global sum to return the same vector to every node for IO to file
+  for(int r=0;r<rd;r++){
+    lvSum[r]=zero;
+  }
+
+  int e1=    grid->_slice_nblock[orthogdim];
+  int e2=    grid->_slice_block [orthogdim];
+  int stride=grid->_slice_stride[orthogdim];
+  // std::cout << GridLogMessage << "End prep" << std::endl;
+  // std::cout << GridLogMessage << "Start parallel inner product, _rd = " << rd << std::endl;
+  vector_type vv;
+  parallel_for(int r=0;r<rd;r++)
+  {
+
+    int so=r*grid->_ostride[orthogdim]; // base offset for start of plane 
+
+    for(int n=0;n<e1;n++){
+      for(int b=0;b<e2;b++){
+        int ss = so + n * stride + b;
+        vv = TensorRemove(innerProduct(lhs._odata[ss], rhs._odata[ss]));
+        lvSum[r] = lvSum[r] + vv;
+      }
+    }
+  }
+  // std::cout << GridLogMessage << "End parallel inner product" << std::endl;
+
+  // Sum across simd lanes in the plane, breaking out orthog dir.
+  std::vector<int> icoor(Nd);
+  for(int rt=0;rt<rd;rt++){
+
+    iScalar<vector_type> temp; 
+    temp._internal = lvSum[rt];
+    extract(temp,extracted);
+
+    for(int idx=0;idx<Nsimd;idx++){
+
+      grid->iCoorFromIindex(icoor,idx);
+
+      int ldx =rt+icoor[orthogdim]*rd;
+
+      lsSum[ldx]=lsSum[ldx]+extracted[idx]._internal;
+
+    }
+  }
+  // std::cout << GridLogMessage << "End sum over simd lanes" << std::endl;
+}
+template <class vobj>
+static void globalSliceInnerProductVector(std::vector<ComplexD> &result, const Lattice<vobj> &lhs, std::vector<typename vobj::scalar_type> &lsSum, int orthogdim)
+{
+  typedef typename vobj::scalar_type scalar_type;
+  GridBase *grid = lhs._grid;
+  int fd = result.size();
+  int ld = lsSum.size();
+  // sum over nodes.
+  std::vector<scalar_type> gsum;
+  gsum.resize(fd, scalar_type(0.0));
+  // std::cout << GridLogMessage << "Start of gsum[t] creation:" << std::endl;
+  for(int t=0;t<fd;t++){
+    int pt = t/ld; // processor plane
+    int lt = t%ld;
+    if ( pt == grid->_processor_coor[orthogdim] ) {
+      gsum[t]=lsSum[lt];
+    }
+  }
+  // std::cout << GridLogMessage << "End of gsum[t] creation:" << std::endl;
+  // std::cout << GridLogMessage << "Start of GlobalSumVector:" << std::endl;
+  grid->GlobalSumVector(&gsum[0], fd);
+  // std::cout << GridLogMessage << "End of GlobalSumVector:" << std::endl;
+
+  result = gsum;
+}
 template<class vobj>
 static void sliceInnerProductVector( std::vector<ComplexD> & result, const Lattice<vobj> &lhs,const Lattice<vobj> &rhs,int orthogdim) 
 {

Grid/lattice/Lattice_rng.h

@@ -158,10 +158,19 @@ namespace Grid {
       // tens of seconds per trajectory so this is clean in all reasonable cases,
       // and margin of safety is orders of magnitude.
       // We could hack Sitmo to skip in the higher order words of state if necessary
+      //
+      // Replace with 2^30 ; avoid problem on large volumes
+      //
       /////////////////////////////////////////////////////////////////////////////////////
       //      uint64_t skip = site+1;  //   Old init Skipped then drew.  Checked compat with faster init
+      const int shift = 30;
+
       uint64_t skip = site;
-      skip = skip<<40;
+
+      skip = skip<<shift;
+
+      assert((skip >> shift)==site); // check for overflow
+
       eng.discard(skip);
       //      std::cout << " Engine  " <<site << " state " <<eng<<std::endl;
     } 
@@ -242,7 +251,7 @@ namespace Grid {
 
       dist[0].reset();
       for(int idx=0;idx<words;idx++){
-  fillScalar(buf[idx],dist[0],_generators[0]);
+	fillScalar(buf[idx],dist[0],_generators[0]);
       }
 
       CartesianCommunicator::BroadcastWorld(0,(void *)&l,sizeof(l));
@@ -274,7 +283,7 @@ namespace Grid {
       RealF *pointer=(RealF *)&l;
       dist[0].reset();
       for(int i=0;i<2*vComplexF::Nsimd();i++){
-  fillScalar(pointer[i],dist[0],_generators[0]);
+	fillScalar(pointer[i],dist[0],_generators[0]);
       }
       CartesianCommunicator::BroadcastWorld(0,(void *)&l,sizeof(l));
     }
@@ -282,7 +291,7 @@ namespace Grid {
       RealD *pointer=(RealD *)&l;
       dist[0].reset();
       for(int i=0;i<2*vComplexD::Nsimd();i++){
-  fillScalar(pointer[i],dist[0],_generators[0]);
+	fillScalar(pointer[i],dist[0],_generators[0]);
       }
       CartesianCommunicator::BroadcastWorld(0,(void *)&l,sizeof(l));
     }
@@ -290,7 +299,7 @@ namespace Grid {
       RealF *pointer=(RealF *)&l;
       dist[0].reset();
       for(int i=0;i<vRealF::Nsimd();i++){
-  fillScalar(pointer[i],dist[0],_generators[0]);
+	fillScalar(pointer[i],dist[0],_generators[0]);
       }
       CartesianCommunicator::BroadcastWorld(0,(void *)&l,sizeof(l));
     }
@@ -308,6 +317,19 @@ namespace Grid {
       std::seed_seq src(seeds.begin(),seeds.end());
       Seed(src,0);
     }
+
+    void SeedUniqueString(const std::string &s){
+      std::vector<int> seeds;
+      std::stringstream sha;
+      seeds = GridChecksum::sha256_seeds(s);
+      for(int i=0;i<seeds.size();i++) { 
+        sha << std::hex << seeds[i];
+      }
+      std::cout << GridLogMessage << "Intialising serial RNG with unique string '" 
+                << s << "'" << std::endl;
+      std::cout << GridLogMessage << "Seed SHA256: " << sha.str() << std::endl;
+      SeedFixedIntegers(seeds);
+    }
   };
 
   class GridParallelRNG : public GridRNGbase {
@@ -368,6 +390,14 @@ namespace Grid {
       _time_counter += usecond()- inner_time_counter;
     };
 
+    void SeedUniqueString(const std::string &s){
+      std::vector<int> seeds;
+      seeds = GridChecksum::sha256_seeds(s);
+      std::cout << GridLogMessage << "Intialising parallel RNG with unique string '" 
+                << s << "'" << std::endl;
+      std::cout << GridLogMessage << "Seed SHA256: " << GridChecksum::sha256_string(seeds) << std::endl;
+      SeedFixedIntegers(seeds);
+    }
     void SeedFixedIntegers(const std::vector<int> &seeds){
 
       // Everyone generates the same seed_seq based on input seeds

Grid/lattice/Lattice_transfer.h

@@ -485,7 +485,7 @@ void InsertSliceLocal(const Lattice<vobj> &lowDim, Lattice<vobj> & higherDim,int
 
 
 template<class vobj>
-void ExtractSliceLocal(Lattice<vobj> &lowDim, Lattice<vobj> & higherDim,int slice_lo,int slice_hi, int orthog)
+void ExtractSliceLocal(Lattice<vobj> &lowDim, const Lattice<vobj> & higherDim,int slice_lo,int slice_hi, int orthog)
 {
   typedef typename vobj::scalar_object sobj;
 
@@ -520,7 +520,7 @@ void ExtractSliceLocal(Lattice<vobj> &lowDim, Lattice<vobj> & higherDim,int slic
 
 
 template<class vobj>
-void Replicate(Lattice<vobj> &coarse,Lattice<vobj> & fine)
+void Replicate(const Lattice<vobj> &coarse,Lattice<vobj> & fine)
 {
   typedef typename vobj::scalar_object sobj;
 

Grid/log/Log.h

@@ -86,7 +86,7 @@ protected:
   Colours &Painter;
   int active;
   int timing_mode;
-  int topWidth{-1};
+  int topWidth{-1}, chanWidth{-1};
   static int timestamp;
   std::string name, topName;
   std::string COLOUR;
@@ -126,6 +126,7 @@ public:
     }
   }
   void setTopWidth(const int w) {topWidth = w;}
+  void setChanWidth(const int w) {chanWidth = w;}
 
   friend std::ostream& operator<< (std::ostream& stream, Logger& log){
 
@@ -136,7 +137,12 @@ public:
         stream << std::setw(log.topWidth);
       }
       stream << log.topName << log.background()<< " : ";
-      stream << log.colour() <<  std::left << log.name << log.background() << " : ";
+      stream << log.colour() <<  std::left;
+      if (log.chanWidth > 0)
+      {
+        stream << std::setw(log.chanWidth);
+      }
+      stream << log.name << log.background() << " : ";
       if ( log.timestamp ) {
 	log.StopWatch->Stop();
 	GridTime now = log.StopWatch->Elapsed();

Grid/parallelIO/IldgIO.h

@@ -250,8 +250,7 @@ class GridLimeReader : public BinaryIO {
   ////////////////////////////////////////////
   // Read a generic serialisable object
   ////////////////////////////////////////////
-  template<class serialisable_object>
-  void readLimeObject(serialisable_object &object,std::string object_name,std::string record_name)
+  void readLimeObject(std::string &xmlstring,std::string record_name)
   {
     // should this be a do while; can we miss a first record??
     while ( limeReaderNextRecord(LimeR) == LIME_SUCCESS ) { 
@@ -266,15 +265,23 @@ class GridLimeReader : public BinaryIO {
 	limeReaderReadData((void *)&xmlc[0], &nbytes, LimeR);    
 	//	std::cout << GridLogMessage<< " readLimeObject matches XML " << &xmlc[0] <<std::endl;
 
-  std::string xmlstring(&xmlc[0]);
-	XmlReader RD(xmlstring, true, "");
-	read(RD,object_name,object);
+   xmlstring = std::string(&xmlc[0]);
 	return;
       }
 
     }  
     assert(0);
   }
+
+  template<class serialisable_object>
+  void readLimeObject(serialisable_object &object,std::string object_name,std::string record_name)
+  {
+    std::string xmlstring;
+
+    readLimeObject(xmlstring, record_name);
+	  XmlReader RD(xmlstring, true, "");
+	  read(RD,object_name,object);
+  }
 };
 
 class GridLimeWriter : public BinaryIO 
@@ -325,16 +332,11 @@ class GridLimeWriter : public BinaryIO
   ////////////////////////////////////////////
   // Write a generic serialisable object
   ////////////////////////////////////////////
-  template<class serialisable_object>
-  void writeLimeObject(int MB,int ME,serialisable_object &object,std::string object_name,std::string record_name)
+  void writeLimeObject(int MB,int ME,XmlWriter &writer,std::string object_name,std::string record_name)
   {
     if ( boss_node ) {
-      std::string xmlstring;
-      {
-	XmlWriter WR("","");
-	write(WR,object_name,object);
-	xmlstring = WR.XmlString();
-      }
+      std::string xmlstring = writer.docString();
+
       //    std::cout << "WriteLimeObject" << record_name <<std::endl;
       uint64_t nbytes = xmlstring.size();
       //    std::cout << " xmlstring "<< nbytes<< " " << xmlstring <<std::endl;
@@ -348,6 +350,20 @@ class GridLimeWriter : public BinaryIO
       limeDestroyHeader(h);
     }
   }
+
+  template<class serialisable_object>
+  void writeLimeObject(int MB,int ME,serialisable_object &object,std::string object_name,std::string record_name, const unsigned int scientificPrec = 0)
+  {
+    XmlWriter WR("","");
+
+    if (scientificPrec)
+    {
+      WR.scientificFormat(true);
+      WR.setPrecision(scientificPrec);
+    }
+    write(WR,object_name,object);
+    writeLimeObject(MB, ME, WR, object_name, record_name);
+  }
   ////////////////////////////////////////////////////
   // Write a generic lattice field and csum
   // This routine is Collectively called by all nodes
@@ -454,7 +470,8 @@ class ScidacWriter : public GridLimeWriter {
   // Write generic lattice field in scidac format
   ////////////////////////////////////////////////
   template <class vobj, class userRecord>
-  void writeScidacFieldRecord(Lattice<vobj> &field,userRecord _userRecord) 
+  void writeScidacFieldRecord(Lattice<vobj> &field,userRecord _userRecord,
+                              const unsigned int recordScientificPrec = 0) 
   {
     GridBase * grid = field._grid;
 
@@ -472,7 +489,7 @@ class ScidacWriter : public GridLimeWriter {
     //////////////////////////////////////////////
     if ( this->boss_node ) {
       writeLimeObject(1,0,header ,std::string("FieldMetaData"),std::string(GRID_FORMAT)); // Open message 
-      writeLimeObject(0,0,_userRecord,_userRecord.SerialisableClassName(),std::string(SCIDAC_RECORD_XML));
+      writeLimeObject(0,0,_userRecord,_userRecord.SerialisableClassName(),std::string(SCIDAC_RECORD_XML), recordScientificPrec);
       writeLimeObject(0,0,_scidacRecord,_scidacRecord.SerialisableClassName(),std::string(SCIDAC_PRIVATE_RECORD_XML));
     }
     // Collective call

Grid/perfmon/Timer.h

@@ -102,6 +102,9 @@ public:
     assert(running == false);
     return (uint64_t) accumulator.count();
   }
+  bool isRunning(void){
+    return running;
+  }
 };
 
 }

Grid/qcd/QCD.h

@@ -90,17 +90,20 @@ namespace QCD {
     // That probably makes for GridRedBlack4dCartesian grid.
 
     // s,sp,c,spc,lc
-    template<typename vtype> using iSinglet                   = iScalar<iScalar<iScalar<vtype> > >;
-    template<typename vtype> using iSpinMatrix                = iScalar<iMatrix<iScalar<vtype>, Ns> >;
-    template<typename vtype> using iColourMatrix              = iScalar<iScalar<iMatrix<vtype, Nc> > > ;
-    template<typename vtype> using iSpinColourMatrix          = iScalar<iMatrix<iMatrix<vtype, Nc>, Ns> >;
-    template<typename vtype> using iLorentzColourMatrix       = iVector<iScalar<iMatrix<vtype, Nc> >, Nd > ;
-    template<typename vtype> using iDoubleStoredColourMatrix  = iVector<iScalar<iMatrix<vtype, Nc> >, Nds > ;
-    template<typename vtype> using iSpinVector                = iScalar<iVector<iScalar<vtype>, Ns> >;
-    template<typename vtype> using iColourVector              = iScalar<iScalar<iVector<vtype, Nc> > >;
-    template<typename vtype> using iSpinColourVector          = iScalar<iVector<iVector<vtype, Nc>, Ns> >;
-    template<typename vtype> using iHalfSpinVector            = iScalar<iVector<iScalar<vtype>, Nhs> >;
-    template<typename vtype> using iHalfSpinColourVector      = iScalar<iVector<iVector<vtype, Nc>, Nhs> >;
+
+    template<typename vtype> using iSinglet                     = iScalar<iScalar<iScalar<vtype> > >;
+    template<typename vtype> using iSpinMatrix                  = iScalar<iMatrix<iScalar<vtype>, Ns> >;
+    template<typename vtype> using iColourMatrix                = iScalar<iScalar<iMatrix<vtype, Nc> > > ;
+    template<typename vtype> using iSpinColourMatrix            = iScalar<iMatrix<iMatrix<vtype, Nc>, Ns> >;
+    template<typename vtype> using iLorentzColourMatrix         = iVector<iScalar<iMatrix<vtype, Nc> >, Nd > ;
+    template<typename vtype> using iDoubleStoredColourMatrix    = iVector<iScalar<iMatrix<vtype, Nc> >, Nds > ;
+    template<typename vtype> using iSpinVector                  = iScalar<iVector<iScalar<vtype>, Ns> >;
+    template<typename vtype> using iColourVector                = iScalar<iScalar<iVector<vtype, Nc> > >;
+    template<typename vtype> using iSpinColourVector            = iScalar<iVector<iVector<vtype, Nc>, Ns> >;
+    template<typename vtype> using iHalfSpinVector              = iScalar<iVector<iScalar<vtype>, Nhs> >;
+    template<typename vtype> using iHalfSpinColourVector        = iScalar<iVector<iVector<vtype, Nc>, Nhs> >;
+    template<typename vtype> using iSpinColourSpinColourMatrix  = iScalar<iMatrix<iMatrix<iMatrix<iMatrix<vtype, Nc>, Ns>, Nc>, Ns> >;
+
 
     template<typename vtype> using iGparitySpinColourVector       = iVector<iVector<iVector<vtype, Nc>, Ns>, Ngp >;
     template<typename vtype> using iGparityHalfSpinColourVector   = iVector<iVector<iVector<vtype, Nc>, Nhs>, Ngp >;
@@ -127,10 +130,28 @@ namespace QCD {
     typedef iSpinColourMatrix<Complex  >    SpinColourMatrix;
     typedef iSpinColourMatrix<ComplexF >    SpinColourMatrixF;
     typedef iSpinColourMatrix<ComplexD >    SpinColourMatrixD;
-
+    
     typedef iSpinColourMatrix<vComplex >    vSpinColourMatrix;
     typedef iSpinColourMatrix<vComplexF>    vSpinColourMatrixF;
     typedef iSpinColourMatrix<vComplexD>    vSpinColourMatrixD;
+    
+    // SpinColourSpinColour matrix
+    typedef iSpinColourSpinColourMatrix<Complex  >    SpinColourSpinColourMatrix;
+    typedef iSpinColourSpinColourMatrix<ComplexF >    SpinColourSpinColourMatrixF;
+    typedef iSpinColourSpinColourMatrix<ComplexD >    SpinColourSpinColourMatrixD;
+
+    typedef iSpinColourSpinColourMatrix<vComplex >    vSpinColourSpinColourMatrix;
+    typedef iSpinColourSpinColourMatrix<vComplexF>    vSpinColourSpinColourMatrixF;
+    typedef iSpinColourSpinColourMatrix<vComplexD>    vSpinColourSpinColourMatrixD;
+
+    // SpinColourSpinColour matrix
+    typedef iSpinColourSpinColourMatrix<Complex  >    SpinColourSpinColourMatrix;
+    typedef iSpinColourSpinColourMatrix<ComplexF >    SpinColourSpinColourMatrixF;
+    typedef iSpinColourSpinColourMatrix<ComplexD >    SpinColourSpinColourMatrixD;
+
+    typedef iSpinColourSpinColourMatrix<vComplex >    vSpinColourSpinColourMatrix;
+    typedef iSpinColourSpinColourMatrix<vComplexF>    vSpinColourSpinColourMatrixF;
+    typedef iSpinColourSpinColourMatrix<vComplexD>    vSpinColourSpinColourMatrixD;
 
     // LorentzColour
     typedef iLorentzColourMatrix<Complex  > LorentzColourMatrix;
@@ -229,6 +250,9 @@ namespace QCD {
     typedef Lattice<vSpinColourMatrixF>     LatticeSpinColourMatrixF;
     typedef Lattice<vSpinColourMatrixD>     LatticeSpinColourMatrixD;
 
+    typedef Lattice<vSpinColourSpinColourMatrix>      LatticeSpinColourSpinColourMatrix;
+    typedef Lattice<vSpinColourSpinColourMatrixF>     LatticeSpinColourSpinColourMatrixF;
+    typedef Lattice<vSpinColourSpinColourMatrixD>     LatticeSpinColourSpinColourMatrixD;
 
     typedef Lattice<vLorentzColourMatrix>  LatticeLorentzColourMatrix;
     typedef Lattice<vLorentzColourMatrixF> LatticeLorentzColourMatrixF;