Merge branch 'develop' of https://github.com/paboyle/Grid into merge2

2025-10-31 20:14:32 +00:00 · 2018-06-12 19:08:06 -04:00
parent 8bb11e5039 f871fb0c6d
commit 855b249f57
67 changed files with 5028 additions and 2406 deletions
--- a/lib/algorithms/iterative/ImplicitlyRestartedLanczos.h
+++ b/lib/algorithms/iterative/ImplicitlyRestartedLanczos.h
@@ -479,15 +479,13 @@ until convergence
 	Field B(grid); B.checkerboard = evec[0].checkerboard;

 	//  power of two search pattern;  not every evalue in eval2 is assessed.
+	int allconv =1;
 	for(int jj = 1; jj<=Nstop; jj*=2){
 	  int j = Nstop-jj;
 	  RealD e = eval2_copy[j]; // Discard the evalue
 	  basisRotateJ(B,evec,Qt,j,0,Nk,Nm);	    
-	  if( _Tester.TestConvergence(j,eresid,B,e,evalMaxApprox) ) {
-	    if ( j > Nconv ) {
-	      Nconv=j+1;
-	      jj=Nstop; // Terminate the scan
-	    }
+	  if( !_Tester.TestConvergence(j,eresid,B,e,evalMaxApprox) ) {
+	    allconv=0;
 	  }
 	}
 	// Do evec[0] for good measure
@@ -495,8 +493,10 @@ until convergence
 	  int j=0;
 	  RealD e = eval2_copy[0]; 
 	  basisRotateJ(B,evec,Qt,j,0,Nk,Nm);	    
-	  _Tester.TestConvergence(j,eresid,B,e,evalMaxApprox);
+	  if( !_Tester.TestConvergence(j,eresid,B,e,evalMaxApprox) ) allconv=0;
 	}
+	if ( allconv ) Nconv = Nstop;
+
 	// test if we converged, if so, terminate
 	std::cout<<GridLogIRL<<" #modes converged: >= "<<Nconv<<"/"<<Nstop<<std::endl;
 	//	if( Nconv>=Nstop || beta_k < betastp){
--- a/lib/algorithms/iterative/LocalCoherenceLanczos.h
+++ b/lib/algorithms/iterative/LocalCoherenceLanczos.h
@@ -48,6 +48,7 @@ struct LanczosParams : Serializable {
 struct LocalCoherenceLanczosParams : Serializable {
 public:
  GRID_SERIALIZABLE_CLASS_MEMBERS(LocalCoherenceLanczosParams,
+				  bool, saveEvecs,
 				  bool, doFine,
 				  bool, doFineRead,
 				  bool, doCoarse,
--- a/lib/allocator/AlignedAllocator.h
+++ b/lib/allocator/AlignedAllocator.h
@@ -277,7 +277,9 @@ public:
    uint8_t *cp = (uint8_t *)ptr;
    if ( ptr ) { 
    // One touch per 4k page, static OMP loop to catch same loop order
+#ifdef GRID_OMP
 #pragma omp parallel for schedule(static)
+#endif
      for(size_type n=0;n<bytes;n+=4096){
 	cp[n]=0;
      }
--- a/lib/communicator/SharedMemoryMPI.cc
+++ b/lib/communicator/SharedMemoryMPI.cc
@@ -27,6 +27,7 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 /*  END LEGAL */

 #include <Grid/GridCore.h>
+#include <pwd.h>

 namespace Grid { 

@@ -113,19 +114,151 @@ void GlobalSharedMemory::Init(Grid_MPI_Comm comm)
  assert(WorldNode!=-1);
  _ShmSetup=1;
 }
-
-void GlobalSharedMemory::OptimalCommunicator(const std::vector<int> &processors,Grid_MPI_Comm & optimal_comm)
+// Gray encode support 
+int BinaryToGray (int  binary) {
+  int gray = (binary>>1)^binary;
+  return gray;
+}
+int Log2Size(int TwoToPower,int MAXLOG2)
 {
-  ////////////////////////////////////////////////////////////////
-  // Assert power of two shm_size.
-  ////////////////////////////////////////////////////////////////
  int log2size = -1;
-  for(int i=0;i<=MAXLOG2RANKSPERNODE;i++){  
-    if ( (0x1<<i) == WorldShmSize ) {
+  for(int i=0;i<=MAXLOG2;i++){
+    if ( (0x1<<i) == TwoToPower ) {
      log2size = i;
      break;
    }
  }
+  return log2size;
+}
+void GlobalSharedMemory::OptimalCommunicator(const std::vector<int> &processors,Grid_MPI_Comm & optimal_comm)
+{
+#undef HYPERCUBE 
+#ifdef HYPERCUBE
+  ////////////////////////////////////////////////////////////////
+  // Assert power of two shm_size.
+  ////////////////////////////////////////////////////////////////
+  int log2size = Log2Size(WorldShmSize,MAXLOG2RANKSPERNODE);
+  assert(log2size != -1);
+
+  ////////////////////////////////////////////////////////////////
+  // Identify the hypercube coordinate of this node using hostname
+  ////////////////////////////////////////////////////////////////
+  // n runs 0...7 9...16 18...25 27...34     (8*4)  5 bits
+  // i runs 0..7                                    3 bits
+  // r runs 0..3                                    2 bits
+  // 2^10 = 1024 nodes
+  const int maxhdim = 10; 
+  std::vector<int> HyperCubeCoords(maxhdim,0);
+  std::vector<int> RootHyperCubeCoords(maxhdim,0);
+  int R;
+  int I;
+  int N;
+  const int namelen = _POSIX_HOST_NAME_MAX;
+  char name[namelen];
+
+  // Parse ICE-XA hostname to get hypercube location
+  gethostname(name,namelen);
+  int nscan = sscanf(name,"r%di%dn%d",&R,&I,&N) ;
+  assert(nscan==3);
+
+  int nlo = N%9;
+  int nhi = N/9;
+  uint32_t hypercoor = (R<<8)|(I<<5)|(nhi<<3)|nlo ;
+  uint32_t rootcoor  = hypercoor;
+
+  //////////////////////////////////////////////////////////////////
+  // Print debug info
+  //////////////////////////////////////////////////////////////////
+  for(int d=0;d<maxhdim;d++){
+    HyperCubeCoords[d] = (hypercoor>>d)&0x1;
+  }
+
+  std::string hname(name);
+  std::cout << "hostname "<<hname<<std::endl;
+  std::cout << "R " << R << " I " << I << " N "<< N<<
+            << " hypercoor 0x"<<std::hex<<hypercoor<<std::dec<<std::endl;
+
+  //////////////////////////////////////////////////////////////////
+  // broadcast node 0's base coordinate for this partition.
+  //////////////////////////////////////////////////////////////////
+  MPI_Bcast(&rootcoor, sizeof(rootcoor), MPI_BYTE, 0, WorldComm); 
+  hypercoor=hypercoor-rootcoor;
+  assert(hypercoor<WorldSize);
+  assert(hypercoor>=0);
+
+  //////////////////////////////////////
+  // Printing
+  //////////////////////////////////////
+  for(int d=0;d<maxhdim;d++){
+    HyperCubeCoords[d] = (hypercoor>>d)&0x1;
+  }
+
+  ////////////////////////////////////////////////////////////////
+  // Identify subblock of ranks on node spreading across dims
+  // in a maximally symmetrical way
+  ////////////////////////////////////////////////////////////////
+  int ndimension              = processors.size();
+  std::vector<int> processor_coor(ndimension);
+  std::vector<int> WorldDims = processors;   std::vector<int> ShmDims  (ndimension,1);  std::vector<int> NodeDims (ndimension);
+  std::vector<int> ShmCoor  (ndimension);    std::vector<int> NodeCoor (ndimension);    std::vector<int> WorldCoor(ndimension);
+  std::vector<int> HyperCoor(ndimension);
+  int dim = 0;
+  for(int l2=0;l2<log2size;l2++){
+    while ( (WorldDims[dim] / ShmDims[dim]) <= 1 ) dim=(dim+1)%ndimension;
+    ShmDims[dim]*=2;
+    dim=(dim+1)%ndimension;
+  }
+
+  ////////////////////////////////////////////////////////////////
+  // Establish torus of processes and nodes with sub-blockings
+  ////////////////////////////////////////////////////////////////
+  for(int d=0;d<ndimension;d++){
+    NodeDims[d] = WorldDims[d]/ShmDims[d];
+  }
+  ////////////////////////////////////////////////////////////////
+  // Map Hcube according to physical lattice 
+  // must partition. Loop over dims and find out who would join.
+  ////////////////////////////////////////////////////////////////
+  int hcoor = hypercoor;
+  for(int d=0;d<ndimension;d++){
+     int bits = Log2Size(NodeDims[d],MAXLOG2RANKSPERNODE);
+     int msk  = (0x1<<bits)-1;
+     HyperCoor[d]=hcoor & msk;  
+     HyperCoor[d]=BinaryToGray(HyperCoor[d]); // Space filling curve magic
+     hcoor = hcoor >> bits;
+  } 
+  ////////////////////////////////////////////////////////////////
+  // Check processor counts match
+  ////////////////////////////////////////////////////////////////
+  int Nprocessors=1;
+  for(int i=0;i<ndimension;i++){
+    Nprocessors*=processors[i];
+  }
+  assert(WorldSize==Nprocessors);
+
+  ////////////////////////////////////////////////////////////////
+  // Establish mapping between lexico physics coord and WorldRank
+  ////////////////////////////////////////////////////////////////
+  int rank;
+
+  Lexicographic::CoorFromIndexReversed(NodeCoor,WorldNode   ,NodeDims);
+
+  for(int d=0;d<ndimension;d++) NodeCoor[d]=HyperCoor[d];
+
+  Lexicographic::CoorFromIndexReversed(ShmCoor ,WorldShmRank,ShmDims);
+  for(int d=0;d<ndimension;d++) WorldCoor[d] = NodeCoor[d]*ShmDims[d]+ShmCoor[d];
+  Lexicographic::IndexFromCoorReversed(WorldCoor,rank,WorldDims);
+
+  /////////////////////////////////////////////////////////////////
+  // Build the new communicator
+  /////////////////////////////////////////////////////////////////
+  int ierr= MPI_Comm_split(WorldComm,0,rank,&optimal_comm);
+  assert(ierr==0);
+#else 
+  ////////////////////////////////////////////////////////////////
+  // Assert power of two shm_size.
+  ////////////////////////////////////////////////////////////////
+  int log2size = Log2Size(WorldShmSize,MAXLOG2RANKSPERNODE);
  assert(log2size != -1);

  ////////////////////////////////////////////////////////////////
@@ -174,15 +307,77 @@ void GlobalSharedMemory::OptimalCommunicator(const std::vector<int> &processors,
  /////////////////////////////////////////////////////////////////
  int ierr= MPI_Comm_split(WorldComm,0,rank,&optimal_comm);
  assert(ierr==0);
+#endif
 }
+////////////////////////////////////////////////////////////////////////////////////////////
+// SHMGET
+////////////////////////////////////////////////////////////////////////////////////////////
+#ifdef GRID_MPI3_SHMGET
+void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
+{
+  std::cout << "SharedMemoryAllocate "<< bytes<< " shmget implementation "<<std::endl;
+  assert(_ShmSetup==1);
+  assert(_ShmAlloc==0);

+  //////////////////////////////////////////////////////////////////////////////////////////////////////////
+  // allocate the shared windows for our group
+  //////////////////////////////////////////////////////////////////////////////////////////////////////////
+  MPI_Barrier(WorldShmComm);
+  WorldShmCommBufs.resize(WorldShmSize);
+  std::vector<int> shmids(WorldShmSize);
+
+  if ( WorldShmRank == 0 ) {
+    for(int r=0;r<WorldShmSize;r++){
+      size_t size = bytes;
+      key_t key   = IPC_PRIVATE;
+      int flags = IPC_CREAT | SHM_R | SHM_W;
+#ifdef SHM_HUGETLB
+      if (Hugepages) flags|=SHM_HUGETLB;
+#endif
+      if ((shmids[r]= shmget(key,size, flags)) ==-1) {
+        int errsv = errno;
+        printf("Errno %d\n",errsv);
+        printf("key   %d\n",key);
+        printf("size  %lld\n",size);
+        printf("flags %d\n",flags);
+        perror("shmget");
+        exit(1);
+      }
+    }
+  }
+  MPI_Barrier(WorldShmComm);
+  MPI_Bcast(&shmids[0],WorldShmSize*sizeof(int),MPI_BYTE,0,WorldShmComm);
+  MPI_Barrier(WorldShmComm);
+
+  for(int r=0;r<WorldShmSize;r++){
+    WorldShmCommBufs[r] = (uint64_t *)shmat(shmids[r], NULL,0);
+    if (WorldShmCommBufs[r] == (uint64_t *)-1) {
+      perror("Shared memory attach failure");
+      shmctl(shmids[r], IPC_RMID, NULL);
+      exit(2);
+    }
+  }
+  MPI_Barrier(WorldShmComm);
+  ///////////////////////////////////
+  // Mark for clean up
+  ///////////////////////////////////
+  for(int r=0;r<WorldShmSize;r++){
+    shmctl(shmids[r], IPC_RMID,(struct shmid_ds *)NULL);
+  }
+  MPI_Barrier(WorldShmComm);
+
+  _ShmAlloc=1;
+  _ShmAllocBytes  = bytes;
+}
+#endif
+ 
 ////////////////////////////////////////////////////////////////////////////////////////////
 // Hugetlbfs mapping intended
 ////////////////////////////////////////////////////////////////////////////////////////////
 #ifdef GRID_MPI3_SHMMMAP
 void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
 {
-  std::cout << "SharedMemoryAllocate "<< bytes<< " MMAP implementation "<<std::endl;
+  std::cout << "SharedMemoryAllocate "<< bytes<< " MMAP implementation "<< GRID_SHM_PATH <<std::endl;
  assert(_ShmSetup==1);
  assert(_ShmAlloc==0);
  //////////////////////////////////////////////////////////////////////////////////////////////////////////
@@ -192,7 +387,7 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
  WorldShmCommBufs.resize(WorldShmSize);
  
  ////////////////////////////////////////////////////////////////////////////////////////////
-  // Hugetlbf and others map filesystems as mappable huge pages
+  // Hugetlbfs and others map filesystems as mappable huge pages
  ////////////////////////////////////////////////////////////////////////////////////////////
  char shm_name [NAME_MAX];
  for(int r=0;r<WorldShmSize;r++){
@@ -226,6 +421,48 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
 };
 #endif // MMAP

+#ifdef GRID_MPI3_SHM_NONE
+void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
+{
+  std::cout << "SharedMemoryAllocate "<< bytes<< " MMAP anonymous implementation "<<std::endl;
+  assert(_ShmSetup==1);
+  assert(_ShmAlloc==0);
+  //////////////////////////////////////////////////////////////////////////////////////////////////////////
+  // allocate the shared windows for our group
+  //////////////////////////////////////////////////////////////////////////////////////////////////////////
+  MPI_Barrier(WorldShmComm);
+  WorldShmCommBufs.resize(WorldShmSize);
+  
+  ////////////////////////////////////////////////////////////////////////////////////////////
+  // Hugetlbf and others map filesystems as mappable huge pages
+  ////////////////////////////////////////////////////////////////////////////////////////////
+  char shm_name [NAME_MAX];
+  assert(WorldShmSize == 1);
+  for(int r=0;r<WorldShmSize;r++){
+    
+    int fd=-1;
+    int mmap_flag = MAP_SHARED |MAP_ANONYMOUS ;
+#ifdef MAP_POPULATE    
+    mmap_flag|=MAP_POPULATE;
+#endif
+#ifdef MAP_HUGETLB
+    if ( flags ) mmap_flag |= MAP_HUGETLB;
+#endif
+    void *ptr = (void *) mmap(NULL, bytes, PROT_READ | PROT_WRITE, mmap_flag,fd, 0); 
+    if ( ptr == (void *)MAP_FAILED ) {    
+      printf("mmap %s failed\n",shm_name);
+      perror("failed mmap");      assert(0);    
+    }
+    assert(((uint64_t)ptr&0x3F)==0);
+    close(fd);
+    WorldShmCommBufs[r] =ptr;
+    std::cout << "Set WorldShmCommBufs["<<r<<"]="<<ptr<< "("<< bytes<< "bytes)"<<std::endl;
+  }
+  _ShmAlloc=1;
+  _ShmAllocBytes  = bytes;
+};
+#endif // MMAP
+
 #ifdef GRID_MPI3_SHMOPEN
 ////////////////////////////////////////////////////////////////////////////////////////////
 // POSIX SHMOPEN ; as far as I know Linux does not allow EXPLICIT HugePages with this case
@@ -246,7 +483,8 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
 	
      size_t size = bytes;
      
-      sprintf(shm_name,"/Grid_mpi3_shm_%d_%d",WorldNode,r);
+      struct passwd *pw = getpwuid (getuid());
+      sprintf(shm_name,"/Grid_%s_mpi3_shm_%d_%d",pw->pw_name,WorldNode,r);
      
      shm_unlink(shm_name);
      int fd=shm_open(shm_name,O_RDWR|O_CREAT,0666);
@@ -281,7 +519,8 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)

      size_t size = bytes ;
      
-      sprintf(shm_name,"/Grid_mpi3_shm_%d_%d",WorldNode,r);
+      struct passwd *pw = getpwuid (getuid());
+      sprintf(shm_name,"/Grid_%s_mpi3_shm_%d_%d",pw->pw_name,WorldNode,r);
      
      int fd=shm_open(shm_name,O_RDWR,0666);
      if ( fd<0 ) {	perror("failed shm_open");	assert(0);      }
@@ -299,6 +538,9 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
 }
 #endif

+
+
+
  ////////////////////////////////////////////////////////
  // Global shared functionality finished
  // Now move to per communicator functionality
--- a/lib/cshift/Cshift_common.h
+++ b/lib/cshift/Cshift_common.h
@@ -45,31 +45,33 @@ Gather_plane_simple (const Lattice<vobj> &rhs,commVector<vobj> &buffer,int dimen
  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 ent = 0;
+
+  static std::vector<std::pair<int,int> > table; table.resize(e1*e2);

  int stride=rhs._grid->_slice_stride[dimension];
  if ( cbmask == 0x3 ) { 
-    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;
 	int bo = n*e2;
-	buffer[off+bo+b]=rhs._odata[so+o+b];
+	table[ent++] = std::pair<int,int>(off+bo+b,so+o+b);
      }
    }
  } else { 
     int bo=0;
-     std::vector<std::pair<int,int> > table;
     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);
 	 if ( ocb &cbmask ) {
-	   table.push_back(std::pair<int,int> (bo++,o+b));
+	   table[ent++]=std::pair<int,int> (off+bo++,so+o+b);
 	 }
       }
     }
-     parallel_for(int i=0;i<table.size();i++){
-       buffer[off+table[i].first]=rhs._odata[so+table[i].second];
-     }
+  }
+  parallel_for(int i=0;i<ent;i++){
+    buffer[table[i].first]=rhs._odata[table[i].second];
  }
 }

@@ -140,31 +142,35 @@ 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];
-  
+
+  static std::vector<std::pair<int,int> > table; table.resize(e1*e2);
+  int ent    =0;
+
  if ( cbmask ==0x3 ) {
-    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*rhs._grid->_slice_stride[dimension];
 	int bo  =n*rhs._grid->_slice_block[dimension];
-	rhs._odata[so+o+b]=buffer[bo+b];
+	table[ent++] = std::pair<int,int>(so+o+b,bo+b);
      }
    }
+
  } 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 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++));
+	  table[ent++]=std::pair<int,int> (so+o+b,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];
-     }
+  }
+
+  parallel_for(int i=0;i<ent;i++){
+    rhs._odata[table[i].first]=buffer[table[i].second];
  }
 }

@@ -228,29 +234,32 @@ template<class vobj> void Copy_plane(Lattice<vobj>& lhs,const Lattice<vobj> &rhs
  int e1=rhs._grid->_slice_nblock[dimension]; // clearly loop invariant for icpc
  int e2=rhs._grid->_slice_block[dimension];
  int stride = rhs._grid->_slice_stride[dimension];
+  static std::vector<std::pair<int,int> > table; table.resize(e1*e2);
+  int ent=0;
+
  if(cbmask == 0x3 ){
-    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+b;
-  	//lhs._odata[lo+o]=rhs._odata[ro+o];
-	vstream(lhs._odata[lo+o],rhs._odata[ro+o]);
+	table[ent++] = std::pair<int,int>(lo+o,ro+o);
      }
    }
  } else { 
-    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+b;
        int ocb=1<<lhs._grid->CheckerBoardFromOindex(o);
        if ( ocb&cbmask ) {
-  	//lhs._odata[lo+o]=rhs._odata[ro+o];
-	  vstream(lhs._odata[lo+o],rhs._odata[ro+o]);
+	  table[ent++] = std::pair<int,int>(lo+o,ro+o);
 	}
      }
    }
  }
-  
+
+  parallel_for(int i=0;i<ent;i++){
+    lhs._odata[table[i].first]=rhs._odata[table[i].second];
+  }
+
 }

 template<class vobj> void Copy_plane_permute(Lattice<vobj>& lhs,const Lattice<vobj> &rhs, int dimension,int lplane,int rplane,int cbmask,int permute_type)
@@ -269,16 +278,28 @@ template<class vobj> void Copy_plane_permute(Lattice<vobj>& lhs,const Lattice<vo
  int e2=rhs._grid->_slice_block [dimension];
  int stride = rhs._grid->_slice_stride[dimension];

-  parallel_for_nest2(int n=0;n<e1;n++){
-  for(int b=0;b<e2;b++){
+  static std::vector<std::pair<int,int> > table;  table.resize(e1*e2);
+  int ent=0;

+  double t_tab,t_perm;
+  if ( cbmask == 0x3 ) {
+    for(int n=0;n<e1;n++){
+    for(int b=0;b<e2;b++){
+      int o  =n*stride;
+      table[ent++] = std::pair<int,int>(lo+o+b,ro+o+b);
+    }}
+  } else {
+    for(int n=0;n<e1;n++){
+    for(int b=0;b<e2;b++){
      int o  =n*stride;
      int ocb=1<<lhs._grid->CheckerBoardFromOindex(o+b);
-      if ( ocb&cbmask ) {
-	permute(lhs._odata[lo+o+b],rhs._odata[ro+o+b],permute_type);
-      }
+      if ( ocb&cbmask ) table[ent++] = std::pair<int,int>(lo+o+b,ro+o+b);
+    }}
+  }

-  }}
+  parallel_for(int i=0;i<ent;i++){
+    permute(lhs._odata[table[i].first],rhs._odata[table[i].second],permute_type);
+  }
 }

 //////////////////////////////////////////////////////
@@ -291,6 +312,8 @@ template<class vobj> void Cshift_local(Lattice<vobj>& ret,const Lattice<vobj> &r
  sshift[0] = rhs._grid->CheckerBoardShiftForCB(rhs.checkerboard,dimension,shift,Even);
  sshift[1] = rhs._grid->CheckerBoardShiftForCB(rhs.checkerboard,dimension,shift,Odd);

+  double t_local;
+  
  if ( sshift[0] == sshift[1] ) {
    Cshift_local(ret,rhs,dimension,shift,0x3);
  } else {
@@ -299,7 +322,7 @@ template<class vobj> void Cshift_local(Lattice<vobj>& ret,const Lattice<vobj> &r
  }
 }

-template<class vobj> Lattice<vobj> Cshift_local(Lattice<vobj> &ret,const Lattice<vobj> &rhs,int dimension,int shift,int cbmask)
+template<class vobj> void Cshift_local(Lattice<vobj> &ret,const Lattice<vobj> &rhs,int dimension,int shift,int cbmask)
 {
  GridBase *grid = rhs._grid;
  int fd = grid->_fdimensions[dimension];
@@ -325,11 +348,7 @@ template<class vobj> Lattice<vobj> Cshift_local(Lattice<vobj> &ret,const Lattice

    int sshift = grid->CheckerBoardShiftForCB(rhs.checkerboard,dimension,shift,cb);
    int sx     = (x+sshift)%rd;
-
-    // FIXME : This must change where we have a 
-    // Rotate slice.
    
-    // Document how this works ; why didn't I do this when I first wrote it...
    // wrap is whether sshift > rd.
    //  num is sshift mod rd.
    // 
@@ -365,10 +384,8 @@ template<class vobj> Lattice<vobj> Cshift_local(Lattice<vobj> &ret,const Lattice

    if ( permute_slice ) Copy_plane_permute(ret,rhs,dimension,x,sx,cbmask,permute_type_dist);
    else                 Copy_plane(ret,rhs,dimension,x,sx,cbmask); 
-
  
  }
-  return ret;
 }
 }
 #endif
--- a/lib/cshift/Cshift_mpi.h
+++ b/lib/cshift/Cshift_mpi.h
@@ -54,13 +54,13 @@ template<class vobj> Lattice<vobj> Cshift(const Lattice<vobj> &rhs,int dimension


  if ( !comm_dim ) {
-    //    std::cout << "Cshift_local" <<std::endl;
+    //std::cout << "CSHIFT: Cshift_local" <<std::endl;
    Cshift_local(ret,rhs,dimension,shift); // Handles checkerboarding
  } else if ( splice_dim ) {
-    //    std::cout << "Cshift_comms_simd" <<std::endl;
+    //std::cout << "CSHIFT: Cshift_comms_simd call - splice_dim = " << splice_dim << " shift " << shift << " dimension = " << dimension << std::endl;
    Cshift_comms_simd(ret,rhs,dimension,shift);
  } else {
-    //    std::cout << "Cshift_comms" <<std::endl;
+    //std::cout << "CSHIFT: Cshift_comms" <<std::endl;
    Cshift_comms(ret,rhs,dimension,shift);
  }
  return ret;
@@ -91,9 +91,12 @@ template<class vobj> void Cshift_comms_simd(Lattice<vobj>& ret,const Lattice<vob
  sshift[0] = rhs._grid->CheckerBoardShiftForCB(rhs.checkerboard,dimension,shift,Even);
  sshift[1] = rhs._grid->CheckerBoardShiftForCB(rhs.checkerboard,dimension,shift,Odd);

+  //std::cout << "Cshift_comms_simd 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_simd(ret,rhs,dimension,shift,0x3);
  } else {
+    //std::cout << "Two pass Cshift_comms" <<std::endl;
    Cshift_comms_simd(ret,rhs,dimension,shift,0x1);// if checkerboard is unfavourable take two passes
    Cshift_comms_simd(ret,rhs,dimension,shift,0x2);// both with block stride loop iteration
  }
@@ -175,6 +178,10 @@ template<class vobj> void  Cshift_comms_simd(Lattice<vobj> &ret,const Lattice<vo
  int simd_layout     = grid->_simd_layout[dimension];
  int comm_dim        = grid->_processors[dimension] >1 ;

+  //std::cout << "Cshift_comms_simd dim "<< dimension << " fd "<<fd<<" rd "<<rd
+  //    << " ld "<<ld<<" pd " << pd<<" simd_layout "<<simd_layout 
+  //    << " comm_dim " << comm_dim << " cbmask " << cbmask <<std::endl;
+
  assert(comm_dim==1);
  assert(simd_layout==2);
  assert(shift>=0);
--- a/lib/lattice/Lattice_base.h
+++ b/lib/lattice/Lattice_base.h
@@ -256,9 +256,42 @@ public:
      _odata[ss]=r._odata[ss];
    }  	
  }
+
+  Lattice(Lattice&& r){ // move constructor
+    _grid = r._grid;
+    checkerboard = r.checkerboard;
+    _odata=std::move(r._odata);
+  }
  
-  
-  
+  inline Lattice<vobj> & operator = (Lattice<vobj> && r)
+  {
+    _grid        = r._grid;
+    checkerboard = r.checkerboard;
+    _odata       =std::move(r._odata);
+    return *this;
+  }
+
+  inline Lattice<vobj> & operator = (const Lattice<vobj> & r){
+    _grid        = r._grid;
+    checkerboard = r.checkerboard;
+    _odata.resize(_grid->oSites());// essential
+    
+    parallel_for(int ss=0;ss<_grid->oSites();ss++){
+      _odata[ss]=r._odata[ss];
+    }  	
+    return *this;
+  }
+
+  template<class robj> strong_inline Lattice<vobj> & operator = (const Lattice<robj> & r){
+    this->checkerboard = r.checkerboard;
+    conformable(*this,r);
+    
+    parallel_for(int ss=0;ss<_grid->oSites();ss++){
+      this->_odata[ss]=r._odata[ss];
+    }
+    return *this;
+  }
+
  virtual ~Lattice(void) = default;
    
  void reset(GridBase* grid) {
@@ -277,15 +310,6 @@ public:
    return *this;
  }
  
-  template<class robj> strong_inline Lattice<vobj> & operator = (const Lattice<robj> & r){
-    this->checkerboard = r.checkerboard;
-    conformable(*this,r);
-    
-    parallel_for(int ss=0;ss<_grid->oSites();ss++){
-      this->_odata[ss]=r._odata[ss];
-    }
-    return *this;
-  }
  
  // *=,+=,-= operators inherit behvour from correspond */+/- operation
  template<class T> strong_inline Lattice<vobj> &operator *=(const T &r) {
--- a/lib/lattice/Lattice_comparison_utils.h
+++ b/lib/lattice/Lattice_comparison_utils.h
@@ -179,7 +179,7 @@ namespace Grid {
      return ret;
    }

-#define DECLARE_RELATIONAL(op,functor) \
+#define DECLARE_RELATIONAL_EQ(op,functor) \
  template<class vsimd,IfSimd<vsimd> = 0>\
    inline vInteger operator op (const vsimd & lhs, const vsimd & rhs)\
    {\
@@ -198,11 +198,6 @@ namespace Grid {
      typedef typename vsimd::scalar_type scalar;\
      return Comparison(functor<scalar,scalar>(),lhs,rhs);\
    }\
-  template<class vsimd,IfSimd<vsimd> = 0>\
-    inline vInteger operator op(const iScalar<vsimd> &lhs,const iScalar<vsimd> &rhs)\
-    {									\
-      return lhs._internal op rhs._internal;				\
-    }									\
  template<class vsimd>\
    inline vInteger operator op(const iScalar<vsimd> &lhs,const typename vsimd::scalar_type &rhs) \
    {									\
@@ -212,14 +207,21 @@ namespace Grid {
    inline vInteger operator op(const typename vsimd::scalar_type &lhs,const iScalar<vsimd> &rhs) \
    {									\
      return lhs op rhs._internal;					\
-    }									
+    }									\

+#define DECLARE_RELATIONAL(op,functor) \
+  DECLARE_RELATIONAL_EQ(op,functor)    \
+  template<class vsimd>\
+    inline vInteger operator op(const iScalar<vsimd> &lhs,const iScalar<vsimd> &rhs)\
+    {									\
+      return lhs._internal op rhs._internal;				\
+    }									

 DECLARE_RELATIONAL(<,slt);
 DECLARE_RELATIONAL(<=,sle);
 DECLARE_RELATIONAL(>,sgt);
 DECLARE_RELATIONAL(>=,sge);
-DECLARE_RELATIONAL(==,seq);
+DECLARE_RELATIONAL_EQ(==,seq);
 DECLARE_RELATIONAL(!=,sne);

 #undef DECLARE_RELATIONAL
--- a/lib/lattice/Lattice_coordinate.h
+++ b/lib/lattice/Lattice_coordinate.h
@@ -52,23 +52,5 @@ namespace Grid {
      }
    };

-    // LatticeCoordinate();
-    // FIXME for debug; deprecate this; made obscelete by 
-    template<class vobj> void lex_sites(Lattice<vobj> &l){
-      Real *v_ptr = (Real *)&l._odata[0];
-      size_t o_len = l._grid->oSites();
-      size_t v_len = sizeof(vobj)/sizeof(vRealF);
-      size_t vec_len = vRealF::Nsimd();
-
-      for(int i=0;i<o_len;i++){
-	for(int j=0;j<v_len;j++){
-          for(int vv=0;vv<vec_len;vv+=2){
-	    v_ptr[i*v_len*vec_len+j*vec_len+vv  ]= i+vv*500;
-	    v_ptr[i*v_len*vec_len+j*vec_len+vv+1]= i+vv*500;
-	  }
-	}}
-    }
-
-
 }
 #endif
--- a/lib/lattice/Lattice_transfer.h
+++ b/lib/lattice/Lattice_transfer.h
@@ -599,6 +599,51 @@ unvectorizeToLexOrdArray(std::vector<sobj> &out, const Lattice<vobj> &in)
    extract1(in_vobj, out_ptrs, 0);
  }
 }
+
+template<typename vobj, typename sobj>
+typename std::enable_if<isSIMDvectorized<vobj>::value && !isSIMDvectorized<sobj>::value, void>::type 
+unvectorizeToRevLexOrdArray(std::vector<sobj> &out, const Lattice<vobj> &in)
+{
+
+  typedef typename vobj::vector_type vtype;
+  
+  GridBase* in_grid = in._grid;
+  out.resize(in_grid->lSites());
+  
+  int ndim = in_grid->Nd();
+  int in_nsimd = vtype::Nsimd();
+
+  std::vector<std::vector<int> > in_icoor(in_nsimd);
+      
+  for(int lane=0; lane < in_nsimd; lane++){
+    in_icoor[lane].resize(ndim);
+    in_grid->iCoorFromIindex(in_icoor[lane], lane);
+  }
+  
+  parallel_for(int in_oidx = 0; in_oidx < in_grid->oSites(); in_oidx++){ //loop over outer index
+    //Assemble vector of pointers to output elements
+    std::vector<sobj*> out_ptrs(in_nsimd);
+
+    std::vector<int> in_ocoor(ndim);
+    in_grid->oCoorFromOindex(in_ocoor, in_oidx);
+
+    std::vector<int> lcoor(in_grid->Nd());
+      
+    for(int lane=0; lane < in_nsimd; lane++){
+      for(int mu=0;mu<ndim;mu++)
+	lcoor[mu] = in_ocoor[mu] + in_grid->_rdimensions[mu]*in_icoor[lane][mu];
+
+      int lex;
+      Lexicographic::IndexFromCoorReversed(lcoor, lex, in_grid->_ldimensions);
+      out_ptrs[lane] = &out[lex];
+    }
+    
+    //Unpack into those ptrs
+    const vobj & in_vobj = in._odata[in_oidx];
+    extract1(in_vobj, out_ptrs, 0);
+  }
+}
+
 //Copy SIMD-vectorized lattice to array of scalar objects in lexicographic order
 template<typename vobj, typename sobj>
 typename std::enable_if<isSIMDvectorized<vobj>::value 
@@ -648,10 +693,59 @@ vectorizeFromLexOrdArray( std::vector<sobj> &in, Lattice<vobj> &out)
  }
 }

+template<typename vobj, typename sobj>
+typename std::enable_if<isSIMDvectorized<vobj>::value 
+                    && !isSIMDvectorized<sobj>::value, void>::type 
+vectorizeFromRevLexOrdArray( std::vector<sobj> &in, Lattice<vobj> &out)
+{
+
+  typedef typename vobj::vector_type vtype;
+  
+  GridBase* grid = out._grid;
+  assert(in.size()==grid->lSites());
+  
+  int ndim     = grid->Nd();
+  int nsimd    = vtype::Nsimd();
+
+  std::vector<std::vector<int> > icoor(nsimd);
+      
+  for(int lane=0; lane < nsimd; lane++){
+    icoor[lane].resize(ndim);
+    grid->iCoorFromIindex(icoor[lane],lane);
+  }
+  
+  parallel_for(uint64_t oidx = 0; oidx < grid->oSites(); oidx++){ //loop over outer index
+    //Assemble vector of pointers to output elements
+    std::vector<sobj*> ptrs(nsimd);
+
+    std::vector<int> ocoor(ndim);
+    grid->oCoorFromOindex(ocoor, oidx);
+
+    std::vector<int> lcoor(grid->Nd());
+      
+    for(int lane=0; lane < nsimd; lane++){
+
+      for(int mu=0;mu<ndim;mu++){
+	lcoor[mu] = ocoor[mu] + grid->_rdimensions[mu]*icoor[lane][mu];
+      }
+
+      int lex;
+      Lexicographic::IndexFromCoorReversed(lcoor, lex, grid->_ldimensions);
+      ptrs[lane] = &in[lex];
+    }
+    
+    //pack from those ptrs
+    vobj vecobj;
+    merge1(vecobj, ptrs, 0);
+    out._odata[oidx] = vecobj; 
+  }
+}
+
 //Convert a Lattice from one precision to another
 template<class VobjOut, class VobjIn>
 void precisionChange(Lattice<VobjOut> &out, const Lattice<VobjIn> &in){
  assert(out._grid->Nd() == in._grid->Nd());
+  assert(out._grid->FullDimensions() == in._grid->FullDimensions());
  out.checkerboard = in.checkerboard;
  GridBase *in_grid=in._grid;
  GridBase *out_grid = out._grid;
--- a/lib/parallelIO/BinaryIO.h
+++ b/lib/parallelIO/BinaryIO.h
@@ -91,7 +91,7 @@ class BinaryIO {
    typedef typename vobj::scalar_object sobj;

    GridBase *grid = lat._grid;
-    int lsites = grid->lSites();
+    uint64_t lsites = grid->lSites();

    std::vector<sobj> scalardata(lsites); 
    unvectorizeToLexOrdArray(scalardata,lat);    
@@ -110,11 +110,11 @@ class BinaryIO {
      lsites = 1;
    }

-    #pragma omp parallel
+PARALLEL_REGION
    {
      uint32_t nersc_csum_thr = 0;

-      #pragma omp for
+PARALLEL_FOR_LOOP_INTERN
      for (uint64_t local_site = 0; local_site < lsites; local_site++)
      {
        uint32_t *site_buf = (uint32_t *)&fbuf[local_site];
@@ -124,7 +124,7 @@ class BinaryIO {
        }
      }

-      #pragma omp critical
+PARALLEL_CRITICAL
      {
        nersc_csum += nersc_csum_thr;
      }
@@ -146,21 +146,23 @@ class BinaryIO {
    std::vector<int> local_start =grid->LocalStarts();
    std::vector<int> global_vol  =grid->FullDimensions();

-#pragma omp parallel
+PARALLEL_REGION
    { 
      std::vector<int> coor(nd);
      uint32_t scidac_csuma_thr=0;
      uint32_t scidac_csumb_thr=0;
      uint32_t site_crc=0;

-#pragma omp for
+PARALLEL_FOR_LOOP_INTERN
      for(uint64_t local_site=0;local_site<lsites;local_site++){

 	uint32_t * site_buf = (uint32_t *)&fbuf[local_site];

 	/* 
 	 * Scidac csum  is rather more heavyweight
+	 * FIXME -- 128^3 x 256 x 16 will overflow.
 	 */
+	
 	int global_site;

 	Lexicographic::CoorFromIndex(coor,local_site,local_vol);
@@ -181,7 +183,7 @@ class BinaryIO {
 	scidac_csumb_thr ^= site_crc<<gsite31 | site_crc>>(32-gsite31);
      }

-#pragma omp critical
+PARALLEL_CRITICAL
      {
 	scidac_csuma^= scidac_csuma_thr;
 	scidac_csumb^= scidac_csumb_thr;
@@ -261,7 +263,7 @@ class BinaryIO {
 			      GridBase *grid,
 			      std::vector<fobj> &iodata,
 			      std::string file,
-			      Integer offset,
+			      uint64_t& offset,
 			      const std::string &format, int control,
 			      uint32_t &nersc_csum,
 			      uint32_t &scidac_csuma,
@@ -370,7 +372,7 @@ class BinaryIO {
 	std::cout << GridLogMessage <<"IOobject: C++ read I/O " << file << " : "
                  << iodata.size() * sizeof(fobj) << " bytes" << std::endl;
        std::ifstream fin;
-        fin.open(file, std::ios::binary | std::ios::in);
+	fin.open(file, std::ios::binary | std::ios::in);
        if (control & BINARYIO_MASTER_APPEND)
        {
          fin.seekg(-sizeof(fobj), fin.end);
@@ -429,14 +431,20 @@ class BinaryIO {
          MPI_Abort(MPI_COMM_WORLD, 1); //assert(ierr == 0);
        }

-        std::cout << GridLogDebug << "MPI read I/O set view " << file << std::endl;
+        std::cout << GridLogDebug << "MPI write I/O set view " << file << std::endl;
        ierr = MPI_File_set_view(fh, disp, mpiObject, fileArray, "native", MPI_INFO_NULL);
        assert(ierr == 0);

-        std::cout << GridLogDebug << "MPI read I/O write all " << file << std::endl;
+        std::cout << GridLogDebug << "MPI write I/O write all " << file << std::endl;
        ierr = MPI_File_write_all(fh, &iodata[0], 1, localArray, &status);
        assert(ierr == 0);

+        MPI_Offset os;
+        MPI_File_get_position(fh, &os);
+        MPI_File_get_byte_offset(fh, os, &disp);
+        offset = disp;
+
+
        MPI_File_close(&fh);
        MPI_Type_free(&fileArray);
        MPI_Type_free(&localArray);
@@ -446,16 +454,20 @@ class BinaryIO {
      } else { 

        std::cout << GridLogMessage << "IOobject: C++ write I/O " << file << " : "
-                  << iodata.size() * sizeof(fobj) << " bytes" << std::endl;
+                  << iodata.size() * sizeof(fobj) << " bytes and offset " << offset << std::endl;
        
 	std::ofstream fout; 
 	fout.exceptions ( std::fstream::failbit | std::fstream::badbit );
 	try {
-	  fout.open(file,std::ios::binary|std::ios::out|std::ios::in);
+	  if (offset) { // Must already exist and contain data
+	    fout.open(file,std::ios::binary|std::ios::out|std::ios::in);
+	  } else {     // Allow create
+	    fout.open(file,std::ios::binary|std::ios::out);
+	  }
 	} catch (const std::fstream::failure& exc) {
 	  std::cout << GridLogError << "Error in opening the file " << file << " for output" <<std::endl;
 	  std::cout << GridLogError << "Exception description: " << exc.what() << std::endl;
-	  std::cout << GridLogError << "Probable cause: wrong path, inaccessible location "<< std::endl;
+	  //	  std::cout << GridLogError << "Probable cause: wrong path, inaccessible location "<< std::endl;
 #ifdef USE_MPI_IO
 	  MPI_Abort(MPI_COMM_WORLD,1);
 #else
@@ -489,6 +501,7 @@ class BinaryIO {
 	  exit(1);
 #endif
 	}
+  offset  = fout.tellp();
 	fout.close();
      }
      timer.Stop();
@@ -523,7 +536,7 @@ class BinaryIO {
  static inline void readLatticeObject(Lattice<vobj> &Umu,
 				       std::string file,
 				       munger munge,
-				       Integer offset,
+				       uint64_t offset,
 				       const std::string &format,
 				       uint32_t &nersc_csum,
 				       uint32_t &scidac_csuma,
@@ -533,7 +546,7 @@ class BinaryIO {
    typedef typename vobj::Realified::scalar_type word;    word w=0;

    GridBase *grid = Umu._grid;
-    int lsites = grid->lSites();
+    uint64_t lsites = grid->lSites();

    std::vector<sobj> scalardata(lsites); 
    std::vector<fobj>     iodata(lsites); // Munge, checksum, byte order in here
@@ -544,7 +557,7 @@ class BinaryIO {
    GridStopWatch timer; 
    timer.Start();

-    parallel_for(int x=0;x<lsites;x++) munge(iodata[x], scalardata[x]);
+    parallel_for(uint64_t x=0;x<lsites;x++) munge(iodata[x], scalardata[x]);

    vectorizeFromLexOrdArray(scalardata,Umu);    
    grid->Barrier();
@@ -560,7 +573,7 @@ class BinaryIO {
    static inline void writeLatticeObject(Lattice<vobj> &Umu,
 					  std::string file,
 					  munger munge,
-					  Integer offset,
+					  uint64_t offset,
 					  const std::string &format,
 					  uint32_t &nersc_csum,
 					  uint32_t &scidac_csuma,
@@ -569,7 +582,7 @@ class BinaryIO {
    typedef typename vobj::scalar_object sobj;
    typedef typename vobj::Realified::scalar_type word;    word w=0;
    GridBase *grid = Umu._grid;
-    int lsites = grid->lSites();
+    uint64_t lsites = grid->lSites();

    std::vector<sobj> scalardata(lsites); 
    std::vector<fobj>     iodata(lsites); // Munge, checksum, byte order in here
@@ -580,7 +593,7 @@ class BinaryIO {
    GridStopWatch timer; timer.Start();
    unvectorizeToLexOrdArray(scalardata,Umu);    

-    parallel_for(int x=0;x<lsites;x++) munge(scalardata[x],iodata[x]);
+    parallel_for(uint64_t x=0;x<lsites;x++) munge(scalardata[x],iodata[x]);

    grid->Barrier();
    timer.Stop();
@@ -597,7 +610,7 @@ class BinaryIO {
  static inline void readRNG(GridSerialRNG &serial,
 			     GridParallelRNG &parallel,
 			     std::string file,
-			     Integer offset,
+			     uint64_t offset,
 			     uint32_t &nersc_csum,
 			     uint32_t &scidac_csuma,
 			     uint32_t &scidac_csumb)
@@ -610,8 +623,8 @@ class BinaryIO {
    std::string format = "IEEE32BIG";

    GridBase *grid = parallel._grid;
-    int gsites = grid->gSites();
-    int lsites = grid->lSites();
+    uint64_t gsites = grid->gSites();
+    uint64_t lsites = grid->lSites();

    uint32_t nersc_csum_tmp   = 0;
    uint32_t scidac_csuma_tmp = 0;
@@ -626,7 +639,7 @@ class BinaryIO {
 	     nersc_csum,scidac_csuma,scidac_csumb);

    timer.Start();
-    parallel_for(int lidx=0;lidx<lsites;lidx++){
+    parallel_for(uint64_t lidx=0;lidx<lsites;lidx++){
      std::vector<RngStateType> tmp(RngStateCount);
      std::copy(iodata[lidx].begin(),iodata[lidx].end(),tmp.begin());
      parallel.SetState(tmp,lidx);
@@ -659,7 +672,7 @@ class BinaryIO {
  static inline void writeRNG(GridSerialRNG &serial,
 			      GridParallelRNG &parallel,
 			      std::string file,
-			      Integer offset,
+			      uint64_t offset,
 			      uint32_t &nersc_csum,
 			      uint32_t &scidac_csuma,
 			      uint32_t &scidac_csumb)
@@ -670,8 +683,8 @@ class BinaryIO {
    typedef std::array<RngStateType,RngStateCount> RNGstate;

    GridBase *grid = parallel._grid;
-    int gsites = grid->gSites();
-    int lsites = grid->lSites();
+    uint64_t gsites = grid->gSites();
+    uint64_t lsites = grid->lSites();

    uint32_t nersc_csum_tmp;
    uint32_t scidac_csuma_tmp;
@@ -684,7 +697,7 @@ class BinaryIO {

    timer.Start();
    std::vector<RNGstate> iodata(lsites);
-    parallel_for(int lidx=0;lidx<lsites;lidx++){
+    parallel_for(uint64_t lidx=0;lidx<lsites;lidx++){
      std::vector<RngStateType> tmp(RngStateCount);
      parallel.GetState(tmp,lidx);
      std::copy(tmp.begin(),tmp.end(),iodata[lidx].begin());
@@ -693,7 +706,6 @@ class BinaryIO {

    IOobject(w,grid,iodata,file,offset,format,BINARYIO_WRITE|BINARYIO_LEXICOGRAPHIC,
 	     nersc_csum,scidac_csuma,scidac_csumb);
-
    iodata.resize(1);
    {
      std::vector<RngStateType> tmp(RngStateCount);
--- a/lib/parallelIO/IldgIO.h
+++ b/lib/parallelIO/IldgIO.h
@@ -182,6 +182,11 @@ class GridLimeReader : public BinaryIO {
   {
     filename= _filename;
     File = fopen(filename.c_str(), "r");
+     if (File == nullptr)
+     {
+       std::cerr << "cannot open file '" << filename << "'" << std::endl;
+       abort();
+     }
     LimeR = limeCreateReader(File);
   }
   /////////////////////////////////////////////
@@ -248,7 +253,6 @@ class GridLimeReader : public BinaryIO {
  template<class serialisable_object>
  void readLimeObject(serialisable_object &object,std::string object_name,std::string record_name)
  {
-    std::string xmlstring;
    // should this be a do while; can we miss a first record??
    while ( limeReaderNextRecord(LimeR) == LIME_SUCCESS ) { 

@@ -262,7 +266,8 @@ class GridLimeReader : public BinaryIO {
 	limeReaderReadData((void *)&xmlc[0], &nbytes, LimeR);    
 	//	std::cout << GridLogMessage<< " readLimeObject matches XML " << &xmlc[0] <<std::endl;

-	XmlReader RD(&xmlc[0],"");
+  std::string xmlstring(&xmlc[0]);
+	XmlReader RD(xmlstring, true, "");
 	read(RD,object_name,object);
 	return;
      }
@@ -272,8 +277,10 @@ class GridLimeReader : public BinaryIO {
  }
 };

-class GridLimeWriter : public BinaryIO {
+class GridLimeWriter : public BinaryIO 
+{
 public:
+
   ///////////////////////////////////////////////////
   // FIXME: format for RNG? Now just binary out instead
   // FIXME: collective calls or not ?
@@ -282,17 +289,24 @@ class GridLimeWriter : public BinaryIO {
   FILE       *File;
   LimeWriter *LimeW;
   std::string filename;
-
+   bool        boss_node;
+   GridLimeWriter( bool isboss = true) {
+     boss_node = isboss;
+   }
   void open(const std::string &_filename) { 
     filename= _filename;
-     File = fopen(filename.c_str(), "w");
-     LimeW = limeCreateWriter(File); assert(LimeW != NULL );
+     if ( boss_node ) {
+       File = fopen(filename.c_str(), "w");
+       LimeW = limeCreateWriter(File); assert(LimeW != NULL );
+     }
   }
   /////////////////////////////////////////////
   // Close the file
   /////////////////////////////////////////////
   void close(void) {
-     fclose(File);
+     if ( boss_node ) {
+       fclose(File);
+     }
     //  limeDestroyWriter(LimeW);
   }
  ///////////////////////////////////////////////////////
@@ -300,10 +314,12 @@ class GridLimeWriter : public BinaryIO {
  ///////////////////////////////////////////////////////
  int createLimeRecordHeader(std::string message, int MB, int ME, size_t PayloadSize)
  {
-    LimeRecordHeader *h;
-    h = limeCreateHeader(MB, ME, const_cast<char *>(message.c_str()), PayloadSize);
-    assert(limeWriteRecordHeader(h, LimeW) >= 0);
-    limeDestroyHeader(h);
+    if ( boss_node ) {
+      LimeRecordHeader *h;
+      h = limeCreateHeader(MB, ME, const_cast<char *>(message.c_str()), PayloadSize);
+      assert(limeWriteRecordHeader(h, LimeW) >= 0);
+      limeDestroyHeader(h);
+    }
    return LIME_SUCCESS;
  }
  ////////////////////////////////////////////
@@ -312,65 +328,99 @@ class GridLimeWriter : public BinaryIO {
  template<class serialisable_object>
  void writeLimeObject(int MB,int ME,serialisable_object &object,std::string object_name,std::string record_name)
  {
-    std::string xmlstring;
-    {
-      XmlWriter WR("","");
-      write(WR,object_name,object);
-      xmlstring = WR.XmlString();
+    if ( boss_node ) {
+      std::string xmlstring;
+      {
+	XmlWriter WR("","");
+	write(WR,object_name,object);
+	xmlstring = WR.XmlString();
+      }
+      //    std::cout << "WriteLimeObject" << record_name <<std::endl;
+      uint64_t nbytes = xmlstring.size();
+      //    std::cout << " xmlstring "<< nbytes<< " " << xmlstring <<std::endl;
+      int err;
+      LimeRecordHeader *h = limeCreateHeader(MB, ME,const_cast<char *>(record_name.c_str()), nbytes); 
+      assert(h!= NULL);
+      
+      err=limeWriteRecordHeader(h, LimeW);                    assert(err>=0);
+      err=limeWriteRecordData(&xmlstring[0], &nbytes, LimeW); assert(err>=0);
+      err=limeWriterCloseRecord(LimeW);                       assert(err>=0);
+      limeDestroyHeader(h);
    }
-    //    std::cout << "WriteLimeObject" << record_name <<std::endl;
-    uint64_t nbytes = xmlstring.size();
-    //    std::cout << " xmlstring "<< nbytes<< " " << xmlstring <<std::endl;
-    int err;
-    LimeRecordHeader *h = limeCreateHeader(MB, ME,const_cast<char *>(record_name.c_str()), nbytes); 
-    assert(h!= NULL);
-
-    err=limeWriteRecordHeader(h, LimeW);                    assert(err>=0);
-    err=limeWriteRecordData(&xmlstring[0], &nbytes, LimeW); assert(err>=0);
-    err=limeWriterCloseRecord(LimeW);                       assert(err>=0);
-    limeDestroyHeader(h);
-    //    std::cout << " File offset is now"<<ftello(File) << std::endl;
  }
-  ////////////////////////////////////////////
+  ////////////////////////////////////////////////////
  // Write a generic lattice field and csum
-  ////////////////////////////////////////////
+  // This routine is Collectively called by all nodes
+  // in communicator used by the field._grid
+  ////////////////////////////////////////////////////
  template<class vobj>
  void writeLimeLatticeBinaryObject(Lattice<vobj> &field,std::string record_name)
  {
-    ////////////////////////////////////////////
-    // Create record header
-    ////////////////////////////////////////////
-    typedef typename vobj::scalar_object sobj;
-    int err;
-    uint32_t nersc_csum,scidac_csuma,scidac_csumb;
-    uint64_t PayloadSize = sizeof(sobj) * field._grid->_gsites;
-    createLimeRecordHeader(record_name, 0, 0, PayloadSize);
-
-    //    std::cout << "W sizeof(sobj)"      <<sizeof(sobj)<<std::endl;
-    //    std::cout << "W Gsites "           <<field._grid->_gsites<<std::endl;
-    //    std::cout << "W Payload expected " <<PayloadSize<<std::endl;
-
    ////////////////////////////////////////////////////////////////////
    // NB: FILE and iostream are jointly writing disjoint sequences in the
    // the same file through different file handles (integer units).
    // 
    // These are both buffered, so why I think this code is right is as follows.
    //
-    // i)  write record header to FILE *File, telegraphing the size. 
-    // ii) ftello reads the offset from FILE *File .
+    // i)  write record header to FILE *File, telegraphing the size; flush
+    // ii) ftello reads the offset from FILE *File . 
    // iii) iostream / MPI Open independently seek this offset. Write sequence direct to disk.
    //      Closes iostream and flushes.
    // iv) fseek on FILE * to end of this disjoint section.
    //  v) Continue writing scidac record.
    ////////////////////////////////////////////////////////////////////
-    uint64_t offset = ftello(File);
-    //    std::cout << " Writing to offset "<<offset << std::endl;
+    
+    GridBase *grid = field._grid;
+    assert(boss_node == field._grid->IsBoss() );
+
+    ////////////////////////////////////////////
+    // Create record header
+    ////////////////////////////////////////////
+    typedef typename vobj::scalar_object sobj;
+    int err;
+    uint32_t nersc_csum,scidac_csuma,scidac_csumb;
+    uint64_t PayloadSize = sizeof(sobj) * grid->_gsites;
+    if ( boss_node ) {
+      createLimeRecordHeader(record_name, 0, 0, PayloadSize);
+      fflush(File);
+    }
+    
+    //    std::cout << "W sizeof(sobj)"      <<sizeof(sobj)<<std::endl;
+    //    std::cout << "W Gsites "           <<field._grid->_gsites<<std::endl;
+    //    std::cout << "W Payload expected " <<PayloadSize<<std::endl;
+
+    ////////////////////////////////////////////////
+    // Check all nodes agree on file position
+    ////////////////////////////////////////////////
+    uint64_t offset1;
+    if ( boss_node ) {
+      offset1 = ftello(File);    
+    }
+    grid->Broadcast(0,(void *)&offset1,sizeof(offset1));
+
+    ///////////////////////////////////////////
+    // The above is collective. Write by other means into the binary record
+    ///////////////////////////////////////////
    std::string format = getFormatString<vobj>();
    BinarySimpleMunger<sobj,sobj> munge;
-    BinaryIO::writeLatticeObject<vobj,sobj>(field, filename, munge, offset, format,nersc_csum,scidac_csuma,scidac_csumb);
-    //    fseek(File,0,SEEK_END);    offset = ftello(File);std::cout << " offset now "<<offset << std::endl;
-    err=limeWriterCloseRecord(LimeW);  assert(err>=0);
+    BinaryIO::writeLatticeObject<vobj,sobj>(field, filename, munge, offset1, format,nersc_csum,scidac_csuma,scidac_csumb);

+    ///////////////////////////////////////////
+    // Wind forward and close the record
+    ///////////////////////////////////////////
+    if ( boss_node ) {
+      fseek(File,0,SEEK_END);             
+      uint64_t offset2 = ftello(File);     //    std::cout << " now at offset "<<offset2 << std::endl;
+      assert( (offset2-offset1) == PayloadSize);
+    }
+
+    /////////////////////////////////////////////////////////////
+    // Check MPI-2 I/O did what we expect to file
+    /////////////////////////////////////////////////////////////
+
+    if ( boss_node ) { 
+      err=limeWriterCloseRecord(LimeW);  assert(err>=0);
+    }
    ////////////////////////////////////////
    // Write checksum element, propagaing forward from the BinaryIO
    // Always pair a checksum with a binary object, and close message
@@ -380,21 +430,26 @@ class GridLimeWriter : public BinaryIO {
    std::stringstream streamb; streamb << std::hex << scidac_csumb;
    checksum.suma= streama.str();
    checksum.sumb= streamb.str();
-    //    std::cout << GridLogMessage<<" writing scidac checksums "<<std::hex<<scidac_csuma<<"/"<<scidac_csumb<<std::dec<<std::endl;
-    writeLimeObject(0,1,checksum,std::string("scidacChecksum"),std::string(SCIDAC_CHECKSUM));
+    if ( boss_node ) { 
+      writeLimeObject(0,1,checksum,std::string("scidacChecksum"),std::string(SCIDAC_CHECKSUM));
+    }
  }
 };

 class ScidacWriter : public GridLimeWriter {
 public:

-   template<class SerialisableUserFile>
-   void writeScidacFileRecord(GridBase *grid,SerialisableUserFile &_userFile)
-   {
-     scidacFile    _scidacFile(grid);
-     writeLimeObject(1,0,_scidacFile,_scidacFile.SerialisableClassName(),std::string(SCIDAC_PRIVATE_FILE_XML));
-     writeLimeObject(0,1,_userFile,_userFile.SerialisableClassName(),std::string(SCIDAC_FILE_XML));
-   }
+  ScidacWriter(bool isboss =true ) : GridLimeWriter(isboss)  { };
+
+  template<class SerialisableUserFile>
+  void writeScidacFileRecord(GridBase *grid,SerialisableUserFile &_userFile)
+  {
+    scidacFile    _scidacFile(grid);
+    if ( this->boss_node ) {
+      writeLimeObject(1,0,_scidacFile,_scidacFile.SerialisableClassName(),std::string(SCIDAC_PRIVATE_FILE_XML));
+      writeLimeObject(0,1,_userFile,_userFile.SerialisableClassName(),std::string(SCIDAC_FILE_XML));
+    }
+  }
  ////////////////////////////////////////////////
  // Write generic lattice field in scidac format
  ////////////////////////////////////////////////
@@ -415,9 +470,12 @@ class ScidacWriter : public GridLimeWriter {
    //////////////////////////////////////////////
    // Fill the Lime file record by record
    //////////////////////////////////////////////
-    writeLimeObject(1,0,header ,std::string("FieldMetaData"),std::string(GRID_FORMAT)); // Open message 
-    writeLimeObject(0,0,_userRecord,_userRecord.SerialisableClassName(),std::string(SCIDAC_RECORD_XML));
-    writeLimeObject(0,0,_scidacRecord,_scidacRecord.SerialisableClassName(),std::string(SCIDAC_PRIVATE_RECORD_XML));
+    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,_scidacRecord,_scidacRecord.SerialisableClassName(),std::string(SCIDAC_PRIVATE_RECORD_XML));
+    }
+    // Collective call
    writeLimeLatticeBinaryObject(field,std::string(ILDG_BINARY_DATA));      // Closes message with checksum
  }
 };
@@ -484,6 +542,8 @@ class ScidacReader : public GridLimeReader {

 class IldgWriter : public ScidacWriter {
 public:
+  
+  IldgWriter(bool isboss) : ScidacWriter(isboss) {};

  ///////////////////////////////////
  // A little helper
@@ -568,7 +628,6 @@ class IldgWriter : public ScidacWriter {
    writeLimeIldgLFN(header.ildg_lfn);                                                 // rec
    writeLimeLatticeBinaryObject(Umu,std::string(ILDG_BINARY_DATA));      // Closes message with checksum
    //    limeDestroyWriter(LimeW);
-    fclose(File);
  }
 };

@@ -644,9 +703,11 @@ class IldgReader : public GridLimeReader {

 	//////////////////////////////////
 	// ILDG format record
+
+  std::string xmlstring(&xmlc[0]);
 	if ( !strncmp(limeReaderType(LimeR), ILDG_FORMAT,strlen(ILDG_FORMAT)) ) { 

-	  XmlReader RD(&xmlc[0],"");
+	  XmlReader RD(xmlstring, true, "");
 	  read(RD,"ildgFormat",ildgFormat_);

 	  if ( ildgFormat_.precision == 64 ) format = std::string("IEEE64BIG");
@@ -661,13 +722,13 @@ class IldgReader : public GridLimeReader {
 	}

 	if ( !strncmp(limeReaderType(LimeR), ILDG_DATA_LFN,strlen(ILDG_DATA_LFN)) ) {
-	  FieldMetaData_.ildg_lfn = std::string(&xmlc[0]);
+	  FieldMetaData_.ildg_lfn = xmlstring;
 	  found_ildgLFN = 1;
 	}

 	if ( !strncmp(limeReaderType(LimeR), GRID_FORMAT,strlen(ILDG_FORMAT)) ) { 

-	  XmlReader RD(&xmlc[0],"");
+	  XmlReader RD(xmlstring, true, "");
 	  read(RD,"FieldMetaData",FieldMetaData_);

 	  format = FieldMetaData_.floating_point;
@@ -681,18 +742,17 @@ class IldgReader : public GridLimeReader {
 	}

 	if ( !strncmp(limeReaderType(LimeR), SCIDAC_RECORD_XML,strlen(SCIDAC_RECORD_XML)) ) { 
-	  std::string xmls(&xmlc[0]);
 	  // is it a USQCD info field
-	  if ( xmls.find(std::string("usqcdInfo")) != std::string::npos ) { 
+	  if ( xmlstring.find(std::string("usqcdInfo")) != std::string::npos ) { 
 	    //	    std::cout << GridLogMessage<<"...found a usqcdInfo field"<<std::endl;
-	    XmlReader RD(&xmlc[0],"");
+	    XmlReader RD(xmlstring, true, "");
 	    read(RD,"usqcdInfo",usqcdInfo_);
 	    found_usqcdInfo = 1;
 	  }
 	}

 	if ( !strncmp(limeReaderType(LimeR), SCIDAC_CHECKSUM,strlen(SCIDAC_CHECKSUM)) ) { 
-	  XmlReader RD(&xmlc[0],"");
+	  XmlReader RD(xmlstring, true, "");
 	  read(RD,"scidacChecksum",scidacChecksum_);
 	  found_scidacChecksum = 1;
 	}
--- a/lib/parallelIO/IldgIOtypes.h
+++ b/lib/parallelIO/IldgIOtypes.h
@@ -136,8 +136,9 @@ struct scidacRecord : Serializable {
 				  int, typesize,
 				  int, datacount);

-  scidacRecord() { version =1.0; }
-
+  scidacRecord()
+  : version(1.0), recordtype(0), colors(0), spins(0), typesize(0), datacount(0)
+  {}
 };

 ////////////////////////
--- a/lib/parallelIO/MetaData.h
+++ b/lib/parallelIO/MetaData.h
@@ -81,18 +81,16 @@ namespace Grid {
 				      std::string, creation_date,
 				      std::string, archive_date,
 				      std::string, floating_point);
-      FieldMetaData(void) { 
-	nd=4;
-	dimension.resize(4);
-	boundary.resize(4);
-	scidac_checksuma=0;
-	scidac_checksumb=0;
-	checksum=0;
-      }
+      // WARNING: non-initialised values might lead to twisted parallel IO
+      // issues, std::string are fine because they initliase to size 0
+      // as per C++ standard.
+      FieldMetaData(void) 
+      : nd(4), dimension(4,0), boundary(4, ""), data_start(0),
+      link_trace(0.), plaquette(0.), checksum(0),
+      scidac_checksuma(0), scidac_checksumb(0), sequence_number(0)
+      {}
    };

-
-
  namespace QCD {

    using namespace Grid;
--- a/lib/parallelIO/NerscIO.h
+++ b/lib/parallelIO/NerscIO.h
@@ -57,7 +57,7 @@ namespace Grid {
      // for the header-reader
      static inline int readHeader(std::string file,GridBase *grid,  FieldMetaData &field)
      {
-      int offset=0;
+      uint64_t offset=0;
      std::map<std::string,std::string> header;
      std::string line;

@@ -139,7 +139,7 @@ namespace Grid {
      typedef Lattice<iLorentzColourMatrix<vsimd> > GaugeField;

      GridBase *grid = Umu._grid;
-      int offset = readHeader(file,Umu._grid,header);
+      uint64_t offset = readHeader(file,Umu._grid,header);

      FieldMetaData clone(header);

@@ -236,21 +236,25 @@ namespace Grid {
 	GaugeStatistics(Umu,header);
 	MachineCharacteristics(header);

-	int offset;
-  
-	truncate(file);
+	uint64_t offset;

 	// Sod it -- always write 3x3 double
 	header.floating_point = std::string("IEEE64BIG");
 	header.data_type      = std::string("4D_SU3_GAUGE_3x3");
 	GaugeSimpleUnmunger<fobj3D,sobj> munge;
-	offset = writeHeader(header,file);
+	if ( grid->IsBoss() ) { 
+	  truncate(file);
+	  offset = writeHeader(header,file);
+	}
+	grid->Broadcast(0,(void *)&offset,sizeof(offset));

 	uint32_t nersc_csum,scidac_csuma,scidac_csumb;
 	BinaryIO::writeLatticeObject<vobj,fobj3D>(Umu,file,munge,offset,header.floating_point,
 								  nersc_csum,scidac_csuma,scidac_csumb);
 	header.checksum = nersc_csum;
-	writeHeader(header,file);
+	if ( grid->IsBoss() ) { 
+	  writeHeader(header,file);
+	}

 	std::cout<<GridLogMessage <<"Written NERSC Configuration on "<< file << " checksum "
 		 <<std::hex<<header.checksum
@@ -278,7 +282,7 @@ namespace Grid {
 	header.plaquette=0.0;
 	MachineCharacteristics(header);

-	int offset;
+	uint64_t offset;
  
 #ifdef RNG_RANLUX
 	header.floating_point = std::string("UINT64");
@@ -293,12 +297,18 @@ namespace Grid {
 	header.data_type      = std::string("SITMO");
 #endif

-	truncate(file);
-	offset = writeHeader(header,file);
+	if ( grid->IsBoss() ) { 
+	  truncate(file);
+	  offset = writeHeader(header,file);
+	}
+	grid->Broadcast(0,(void *)&offset,sizeof(offset));
+	
 	uint32_t nersc_csum,scidac_csuma,scidac_csumb;
 	BinaryIO::writeRNG(serial,parallel,file,offset,nersc_csum,scidac_csuma,scidac_csumb);
 	header.checksum = nersc_csum;
-	offset = writeHeader(header,file);
+	if ( grid->IsBoss() ) { 
+	  offset = writeHeader(header,file);
+	}

 	std::cout<<GridLogMessage 
 		 <<"Written NERSC RNG STATE "<<file<< " checksum "
@@ -313,7 +323,7 @@ namespace Grid {

 	GridBase *grid = parallel._grid;

-	int offset = readHeader(file,grid,header);
+	uint64_t offset = readHeader(file,grid,header);

 	FieldMetaData clone(header);

--- a/lib/pugixml/README.md
+++ b/lib/pugixml/README.md
@@ -1,44 +0,0 @@
-pugixml [![Build Status](https://travis-ci.org/zeux/pugixml.svg?branch=master)](https://travis-ci.org/zeux/pugixml) [![Build status](https://ci.appveyor.com/api/projects/status/9hdks1doqvq8pwe7/branch/master?svg=true)](https://ci.appveyor.com/project/zeux/pugixml)
-=======
-
-pugixml is a C++ XML processing library, which consists of a DOM-like interface with rich traversal/modification
-capabilities, an extremely fast XML parser which constructs the DOM tree from an XML file/buffer, and an XPath 1.0
-implementation for complex data-driven tree queries. Full Unicode support is also available, with Unicode interface
-variants and conversions between different Unicode encodings (which happen automatically during parsing/saving).
-
-pugixml is used by a lot of projects, both open-source and proprietary, for performance and easy-to-use interface.
-
-## Documentation
-
-Documentation for the current release of pugixml is available on-line as two separate documents:
-
-* [Quick-start guide](http://pugixml.org/docs/quickstart.html), that aims to provide enough information to start using the library;
-* [Complete reference manual](http://pugixml.org/docs/manual.html), that describes all features of the library in detail.
-
-You’re advised to start with the quick-start guide; however, many important library features are either not described in it at all or only mentioned briefly; if you require more information you should read the complete manual.
-
-## License
-This library is available to anybody free of charge, under the terms of MIT License:
-
-Copyright (c) 2006-2015 Arseny Kapoulkine
-
-Permission is hereby granted, free of charge, to any person
-obtaining a copy of this software and associated documentation
-files (the "Software"), to deal in the Software without
-restriction, including without limitation the rights to use,
-copy, modify, merge, publish, distribute, sublicense, and/or sell
-copies of the Software, and to permit persons to whom the
-Software is furnished to do so, subject to the following
-conditions:
-
-The above copyright notice and this permission notice shall be
-included in all copies or substantial portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
-EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
-OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
-NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
-HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
-WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
-FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
-OTHER DEALINGS IN THE SOFTWARE.
--- a/lib/pugixml/pugiconfig.hpp
+++ b/lib/pugixml/pugiconfig.hpp
@@ -1,7 +1,7 @@
 /**
- * pugixml parser - version 1.6
+ * pugixml parser - version 1.9
 * --------------------------------------------------------
- * Copyright (C) 2006-2015, by Arseny Kapoulkine (arseny.kapoulkine@gmail.com)
+ * Copyright (C) 2006-2018, by Arseny Kapoulkine (arseny.kapoulkine@gmail.com)
 * Report bugs and download new versions at http://pugixml.org/
 *
 * This library is distributed under the MIT License. See notice at the end
@@ -17,6 +17,9 @@
 // Uncomment this to enable wchar_t mode
 // #define PUGIXML_WCHAR_MODE

+// Uncomment this to enable compact mode
+// #define PUGIXML_COMPACT
+
 // Uncomment this to disable XPath
 // #define PUGIXML_NO_XPATH

@@ -46,7 +49,7 @@
 #endif

 /**
- * Copyright (c) 2006-2015 Arseny Kapoulkine
+ * Copyright (c) 2006-2018 Arseny Kapoulkine
 *
 * Permission is hereby granted, free of charge, to any person
 * obtaining a copy of this software and associated documentation
@@ -59,7 +62,7 @@
 *
 * The above copyright notice and this permission notice shall be
 * included in all copies or substantial portions of the Software.
- * 
+ *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
 * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
--- a/lib/pugixml/pugixml.cc
+++ b/lib/pugixml/pugixml.cc
--- a/lib/pugixml/pugixml.h
+++ b/lib/pugixml/pugixml.h
@@ -1,7 +1,7 @@
 /**
- * pugixml parser - version 1.6
+ * pugixml parser - version 1.9
 * --------------------------------------------------------
- * Copyright (C) 2006-2015, by Arseny Kapoulkine (arseny.kapoulkine@gmail.com)
+ * Copyright (C) 2006-2018, by Arseny Kapoulkine (arseny.kapoulkine@gmail.com)
 * Report bugs and download new versions at http://pugixml.org/
 *
 * This library is distributed under the MIT License. See notice at the end
@@ -13,7 +13,7 @@

 #ifndef PUGIXML_VERSION
 // Define version macro; evaluates to major * 100 + minor so that it's safe to use in less-than comparisons
-#	define PUGIXML_VERSION 160
+#	define PUGIXML_VERSION 190
 #endif

 // Include user configuration file (this can define various configuration macros)
@@ -72,6 +72,44 @@
 #	endif
 #endif

+// If the platform is known to have move semantics support, compile move ctor/operator implementation
+#ifndef PUGIXML_HAS_MOVE
+#	if __cplusplus >= 201103
+#		define PUGIXML_HAS_MOVE
+#	elif defined(_MSC_VER) && _MSC_VER >= 1600
+#		define PUGIXML_HAS_MOVE
+#	endif
+#endif
+
+// If C++ is 2011 or higher, add 'noexcept' specifiers
+#ifndef PUGIXML_NOEXCEPT
+#	if __cplusplus >= 201103
+#		define PUGIXML_NOEXCEPT noexcept
+#	elif defined(_MSC_VER) && _MSC_VER >= 1900
+#		define PUGIXML_NOEXCEPT noexcept
+#	else
+#		define PUGIXML_NOEXCEPT
+#	endif
+#endif
+
+// Some functions can not be noexcept in compact mode
+#ifdef PUGIXML_COMPACT
+#	define PUGIXML_NOEXCEPT_IF_NOT_COMPACT
+#else
+#	define PUGIXML_NOEXCEPT_IF_NOT_COMPACT PUGIXML_NOEXCEPT
+#endif
+
+// If C++ is 2011 or higher, add 'override' qualifiers
+#ifndef PUGIXML_OVERRIDE
+#	if __cplusplus >= 201103
+#		define PUGIXML_OVERRIDE override
+#	elif defined(_MSC_VER) && _MSC_VER >= 1700
+#		define PUGIXML_OVERRIDE override
+#	else
+#		define PUGIXML_OVERRIDE
+#	endif
+#endif
+
 // Character interface macros
 #ifdef PUGIXML_WCHAR_MODE
 #	define PUGIXML_TEXT(t) L ## t
@@ -133,13 +171,13 @@ namespace pugi

 	// This flag determines if EOL characters are normalized (converted to #xA) during parsing. This flag is on by default.
 	const unsigned int parse_eol = 0x0020;
-	
+
 	// This flag determines if attribute values are normalized using CDATA normalization rules during parsing. This flag is on by default.
 	const unsigned int parse_wconv_attribute = 0x0040;

 	// This flag determines if attribute values are normalized using NMTOKENS normalization rules during parsing. This flag is off by default.
 	const unsigned int parse_wnorm_attribute = 0x0080;
-	
+
 	// This flag determines if document declaration (node_declaration) is added to the DOM tree. This flag is off by default.
 	const unsigned int parse_declaration = 0x0100;

@@ -158,6 +196,11 @@ namespace pugi
 	// is a valid document. This flag is off by default.
 	const unsigned int parse_fragment = 0x1000;

+	// This flag determines if plain character data is be stored in the parent element's value. This significantly changes the structure of
+	// the document; this flag is only recommended for parsing documents with many PCDATA nodes in memory-constrained environments.
+	// This flag is off by default.
+	const unsigned int parse_embed_pcdata = 0x2000;
+
 	// The default parsing mode.
 	// Elements, PCDATA and CDATA sections are added to the DOM tree, character/reference entities are expanded,
 	// End-of-Line characters are normalized, attribute values are normalized using CDATA normalization rules.
@@ -184,16 +227,16 @@ namespace pugi
 	};

 	// Formatting flags
-	
+
 	// Indent the nodes that are written to output stream with as many indentation strings as deep the node is in DOM tree. This flag is on by default.
 	const unsigned int format_indent = 0x01;
-	
+
 	// Write encoding-specific BOM to the output stream. This flag is off by default.
 	const unsigned int format_write_bom = 0x02;

 	// Use raw output mode (no indentation and no line breaks are written). This flag is off by default.
 	const unsigned int format_raw = 0x04;
-	
+
 	// Omit default XML declaration even if there is no declaration in the document. This flag is off by default.
 	const unsigned int format_no_declaration = 0x08;

@@ -206,6 +249,9 @@ namespace pugi
 	// Write every attribute on a new line with appropriate indentation. This flag is off by default.
 	const unsigned int format_indent_attributes = 0x40;

+	// Don't output empty element tags, instead writing an explicit start and end tag even if there are no children. This flag is off by default.
+	const unsigned int format_no_empty_element_tags = 0x80;
+
 	// The default set of formatting flags.
 	// Nodes are indented depending on their depth in DOM tree, a default declaration is output if document has none.
 	const unsigned int format_default = format_indent;
@@ -225,7 +271,7 @@ namespace pugi
 	class xml_node;

 	class xml_text;
-	
+
 	#ifndef PUGIXML_NO_XPATH
 	class xpath_node;
 	class xpath_node_set;
@@ -268,7 +314,7 @@ namespace pugi
 		// Construct writer from a FILE* object; void* is used to avoid header dependencies on stdio
 		xml_writer_file(void* file);

-		virtual void write(const void* data, size_t size);
+		virtual void write(const void* data, size_t size) PUGIXML_OVERRIDE;

 	private:
 		void* file;
@@ -283,7 +329,7 @@ namespace pugi
 		xml_writer_stream(std::basic_ostream<char, std::char_traits<char> >& stream);
 		xml_writer_stream(std::basic_ostream<wchar_t, std::char_traits<wchar_t> >& stream);

-		virtual void write(const void* data, size_t size);
+		virtual void write(const void* data, size_t size) PUGIXML_OVERRIDE;

 	private:
 		std::basic_ostream<char, std::char_traits<char> >* narrow_stream;
@@ -299,13 +345,13 @@ namespace pugi

 	private:
 		xml_attribute_struct* _attr;
-	
+
 		typedef void (*unspecified_bool_type)(xml_attribute***);

 	public:
 		// Default constructor. Constructs an empty attribute.
 		xml_attribute();
-		
+
 		// Constructs attribute from internal pointer
 		explicit xml_attribute(xml_attribute_struct* attr);

@@ -354,6 +400,8 @@ namespace pugi
 		// Set attribute value with type conversion (numbers are converted to strings, boolean is converted to "true"/"false")
 		bool set_value(int rhs);
 		bool set_value(unsigned int rhs);
+		bool set_value(long rhs);
+		bool set_value(unsigned long rhs);
 		bool set_value(double rhs);
 		bool set_value(float rhs);
 		bool set_value(bool rhs);
@@ -367,6 +415,8 @@ namespace pugi
 		xml_attribute& operator=(const char_t* rhs);
 		xml_attribute& operator=(int rhs);
 		xml_attribute& operator=(unsigned int rhs);
+		xml_attribute& operator=(long rhs);
+		xml_attribute& operator=(unsigned long rhs);
 		xml_attribute& operator=(double rhs);
 		xml_attribute& operator=(float rhs);
 		xml_attribute& operator=(bool rhs);
@@ -417,7 +467,7 @@ namespace pugi

 		// Borland C++ workaround
 		bool operator!() const;
-	
+
 		// Comparison operators (compares wrapped node pointers)
 		bool operator==(const xml_node& r) const;
 		bool operator!=(const xml_node& r) const;
@@ -438,7 +488,7 @@ namespace pugi
 		// Get node value, or "" if node is empty or it has no value
 		// Note: For <node>text</node> node.value() does not return "text"! Use child_value() or text() methods to access text inside nodes.
 		const char_t* value() const;
-	
+
 		// Get attribute list
 		xml_attribute first_attribute() const;
 		xml_attribute last_attribute() const;
@@ -450,7 +500,7 @@ namespace pugi
 		// Get next/previous sibling in the children list of the parent node
 		xml_node next_sibling() const;
 		xml_node previous_sibling() const;
-		
+
 		// Get parent node
 		xml_node parent() const;

@@ -478,7 +528,7 @@ namespace pugi
 		// Set node name/value (returns false if node is empty, there is not enough memory, or node can not have name/value)
 		bool set_name(const char_t* rhs);
 		bool set_value(const char_t* rhs);
-		
+
 		// Add attribute with specified name. Returns added attribute, or empty attribute on errors.
 		xml_attribute append_attribute(const char_t* name);
 		xml_attribute prepend_attribute(const char_t* name);
@@ -532,11 +582,11 @@ namespace pugi
 		template <typename Predicate> xml_attribute find_attribute(Predicate pred) const
 		{
 			if (!_root) return xml_attribute();
-			
+
 			for (xml_attribute attrib = first_attribute(); attrib; attrib = attrib.next_attribute())
 				if (pred(attrib))
 					return attrib;
-		
+
 			return xml_attribute();
 		}

@@ -544,11 +594,11 @@ namespace pugi
 		template <typename Predicate> xml_node find_child(Predicate pred) const
 		{
 			if (!_root) return xml_node();
-	
+
 			for (xml_node node = first_child(); node; node = node.next_sibling())
 				if (pred(node))
 					return node;
-		
+
 			return xml_node();
 		}

@@ -558,7 +608,7 @@ namespace pugi
 			if (!_root) return xml_node();

 			xml_node cur = first_child();
-			
+
 			while (cur._root && cur._root != _root)
 			{
 				if (pred(cur)) return cur;
@@ -590,7 +640,7 @@ namespace pugi

 		// Recursively traverse subtree with xml_tree_walker
 		bool traverse(xml_tree_walker& walker);
-	
+
 	#ifndef PUGIXML_NO_XPATH
 		// Select single node by evaluating XPath query. Returns first node from the resulting node set.
 		xpath_node select_node(const char_t* query, xpath_variable_set* variables = 0) const;
@@ -601,11 +651,11 @@ namespace pugi
 		xpath_node_set select_nodes(const xpath_query& query) const;

 		// (deprecated: use select_node instead) Select single node by evaluating XPath query.
-		xpath_node select_single_node(const char_t* query, xpath_variable_set* variables = 0) const;
-		xpath_node select_single_node(const xpath_query& query) const;
+		PUGIXML_DEPRECATED xpath_node select_single_node(const char_t* query, xpath_variable_set* variables = 0) const;
+		PUGIXML_DEPRECATED xpath_node select_single_node(const xpath_query& query) const;

 	#endif
-		
+
 		// Print subtree using a writer object
 		void print(xml_writer& writer, const char_t* indent = PUGIXML_TEXT("\t"), unsigned int flags = format_default, xml_encoding encoding = encoding_auto, unsigned int depth = 0) const;

@@ -701,6 +751,8 @@ namespace pugi
 		// Set text with type conversion (numbers are converted to strings, boolean is converted to "true"/"false")
 		bool set(int rhs);
 		bool set(unsigned int rhs);
+		bool set(long rhs);
+		bool set(unsigned long rhs);
 		bool set(double rhs);
 		bool set(float rhs);
 		bool set(bool rhs);
@@ -714,6 +766,8 @@ namespace pugi
 		xml_text& operator=(const char_t* rhs);
 		xml_text& operator=(int rhs);
 		xml_text& operator=(unsigned int rhs);
+		xml_text& operator=(long rhs);
+		xml_text& operator=(unsigned long rhs);
 		xml_text& operator=(double rhs);
 		xml_text& operator=(float rhs);
 		xml_text& operator=(bool rhs);
@@ -867,11 +921,11 @@ namespace pugi

 	private:
 		int _depth;
-	
+
 	protected:
 		// Get current traversal depth
 		int depth() const;
-	
+
 	public:
 		xml_tree_walker();
 		virtual ~xml_tree_walker();
@@ -942,13 +996,14 @@ namespace pugi
 		char_t* _buffer;

 		char _memory[192];
-		
+
 		// Non-copyable semantics
 		xml_document(const xml_document&);
-		const xml_document& operator=(const xml_document&);
+		xml_document& operator=(const xml_document&);

-		void create();
-		void destroy();
+		void _create();
+		void _destroy();
+		void _move(xml_document& rhs) PUGIXML_NOEXCEPT_IF_NOT_COMPACT;

 	public:
 		// Default constructor, makes empty document
@@ -957,6 +1012,12 @@ namespace pugi
 		// Destructor, invalidates all node/attribute handles to this document
 		~xml_document();

+	#ifdef PUGIXML_HAS_MOVE
+		// Move semantics support
+		xml_document(xml_document&& rhs) PUGIXML_NOEXCEPT_IF_NOT_COMPACT;
+		xml_document& operator=(xml_document&& rhs) PUGIXML_NOEXCEPT_IF_NOT_COMPACT;
+	#endif
+
 		// Removes all nodes, leaving the empty document
 		void reset();

@@ -970,7 +1031,7 @@ namespace pugi
 	#endif

 		// (deprecated: use load_string instead) Load document from zero-terminated string. No encoding conversions are applied.
-		xml_parse_result load(const char_t* contents, unsigned int options = parse_default);
+		PUGIXML_DEPRECATED xml_parse_result load(const char_t* contents, unsigned int options = parse_default);

 		// Load document from zero-terminated string. No encoding conversions are applied.
 		xml_parse_result load_string(const char_t* contents, unsigned int options = parse_default);
@@ -1051,7 +1112,7 @@ namespace pugi
 		// Non-copyable semantics
 		xpath_variable(const xpath_variable&);
 		xpath_variable& operator=(const xpath_variable&);
-		
+
 	public:
 		// Get variable name
 		const char_t* name() const;
@@ -1095,10 +1156,10 @@ namespace pugi
 		xpath_variable_set(const xpath_variable_set& rhs);
 		xpath_variable_set& operator=(const xpath_variable_set& rhs);

-	#if __cplusplus >= 201103
+	#ifdef PUGIXML_HAS_MOVE
 		// Move semantics support
-		xpath_variable_set(xpath_variable_set&& rhs);
-		xpath_variable_set& operator=(xpath_variable_set&& rhs);
+		xpath_variable_set(xpath_variable_set&& rhs) PUGIXML_NOEXCEPT;
+		xpath_variable_set& operator=(xpath_variable_set&& rhs) PUGIXML_NOEXCEPT;
 	#endif

 		// Add a new variable or get the existing one, if the types match
@@ -1139,29 +1200,29 @@ namespace pugi
 		// Destructor
 		~xpath_query();

-	#if __cplusplus >= 201103
+	#ifdef PUGIXML_HAS_MOVE
 		// Move semantics support
-		xpath_query(xpath_query&& rhs);
-		xpath_query& operator=(xpath_query&& rhs);
+		xpath_query(xpath_query&& rhs) PUGIXML_NOEXCEPT;
+		xpath_query& operator=(xpath_query&& rhs) PUGIXML_NOEXCEPT;
 	#endif

 		// Get query expression return type
 		xpath_value_type return_type() const;
-		
+
 		// Evaluate expression as boolean value in the specified context; performs type conversion if necessary.
 		// If PUGIXML_NO_EXCEPTIONS is not defined, throws std::bad_alloc on out of memory errors.
 		bool evaluate_boolean(const xpath_node& n) const;
-		
+
 		// Evaluate expression as double value in the specified context; performs type conversion if necessary.
 		// If PUGIXML_NO_EXCEPTIONS is not defined, throws std::bad_alloc on out of memory errors.
 		double evaluate_number(const xpath_node& n) const;
-		
+
 	#ifndef PUGIXML_NO_STL
 		// Evaluate expression as string value in the specified context; performs type conversion if necessary.
 		// If PUGIXML_NO_EXCEPTIONS is not defined, throws std::bad_alloc on out of memory errors.
 		string_t evaluate_string(const xpath_node& n) const;
 	#endif
-		
+
 		// Evaluate expression as string value in the specified context; performs type conversion if necessary.
 		// At most capacity characters are written to the destination buffer, full result size is returned (includes terminating zero).
 		// If PUGIXML_NO_EXCEPTIONS is not defined, throws std::bad_alloc on out of memory errors.
@@ -1188,7 +1249,7 @@ namespace pugi
 		// Borland C++ workaround
 		bool operator!() const;
 	};
-	
+
 	#ifndef PUGIXML_NO_EXCEPTIONS
 	// XPath exception class
 	class PUGIXML_CLASS xpath_exception: public std::exception
@@ -1201,26 +1262,26 @@ namespace pugi
 		explicit xpath_exception(const xpath_parse_result& result);

 		// Get error message
-		virtual const char* what() const throw();
+		virtual const char* what() const throw() PUGIXML_OVERRIDE;

 		// Get parse result
 		const xpath_parse_result& result() const;
 	};
 	#endif
-	
+
 	// XPath node class (either xml_node or xml_attribute)
 	class PUGIXML_CLASS xpath_node
 	{
 	private:
 		xml_node _node;
 		xml_attribute _attribute;
-	
+
 		typedef void (*unspecified_bool_type)(xpath_node***);

 	public:
 		// Default constructor; constructs empty XPath node
 		xpath_node();
-		
+
 		// Construct XPath node from XML node/attribute
 		xpath_node(const xml_node& node);
 		xpath_node(const xml_attribute& attribute, const xml_node& parent);
@@ -1228,13 +1289,13 @@ namespace pugi
 		// Get node/attribute, if any
 		xml_node node() const;
 		xml_attribute attribute() const;
-		
+
 		// Get parent of contained node/attribute
 		xml_node parent() const;

 		// Safe bool conversion operator
 		operator unspecified_bool_type() const;
-		
+
 		// Borland C++ workaround
 		bool operator!() const;

@@ -1260,13 +1321,13 @@ namespace pugi
 			type_sorted,			// Sorted by document order (ascending)
 			type_sorted_reverse		// Sorted by document order (descending)
 		};
-		
+
 		// Constant iterator type
 		typedef const xpath_node* const_iterator;

 		// We define non-constant iterator to be the same as constant iterator so that various generic algorithms (i.e. boost foreach) work
 		typedef const xpath_node* iterator;
-	
+
 		// Default constructor. Constructs empty set.
 		xpath_node_set();

@@ -1275,49 +1336,49 @@ namespace pugi

 		// Destructor
 		~xpath_node_set();
-		
+
 		// Copy constructor/assignment operator
 		xpath_node_set(const xpath_node_set& ns);
 		xpath_node_set& operator=(const xpath_node_set& ns);

-	#if __cplusplus >= 201103
+	#ifdef PUGIXML_HAS_MOVE
 		// Move semantics support
-		xpath_node_set(xpath_node_set&& rhs);
-		xpath_node_set& operator=(xpath_node_set&& rhs);
+		xpath_node_set(xpath_node_set&& rhs) PUGIXML_NOEXCEPT;
+		xpath_node_set& operator=(xpath_node_set&& rhs) PUGIXML_NOEXCEPT;
 	#endif

 		// Get collection type
 		type_t type() const;
-		
+
 		// Get collection size
 		size_t size() const;

 		// Indexing operator
 		const xpath_node& operator[](size_t index) const;
-		
+
 		// Collection iterators
 		const_iterator begin() const;
 		const_iterator end() const;

 		// Sort the collection in ascending/descending order by document order
 		void sort(bool reverse = false);
-		
+
 		// Get first node in the collection by document order
 		xpath_node first() const;
-		
+
 		// Check if collection is empty
 		bool empty() const;
-	
+
 	private:
 		type_t _type;
-		
+
 		xpath_node _storage;
-		
+
 		xpath_node* _begin;
 		xpath_node* _end;

 		void _assign(const_iterator begin, const_iterator end, type_t type);
-		void _move(xpath_node_set& rhs);
+		void _move(xpath_node_set& rhs) PUGIXML_NOEXCEPT;
 	};
 #endif

@@ -1325,7 +1386,7 @@ namespace pugi
 	// Convert wide string to UTF8
 	std::basic_string<char, std::char_traits<char>, std::allocator<char> > PUGIXML_FUNCTION as_utf8(const wchar_t* str);
 	std::basic_string<char, std::char_traits<char>, std::allocator<char> > PUGIXML_FUNCTION as_utf8(const std::basic_string<wchar_t, std::char_traits<wchar_t>, std::allocator<wchar_t> >& str);
-	
+
 	// Convert UTF8 to wide string
 	std::basic_string<wchar_t, std::char_traits<wchar_t>, std::allocator<wchar_t> > PUGIXML_FUNCTION as_wide(const char* str);
 	std::basic_string<wchar_t, std::char_traits<wchar_t>, std::allocator<wchar_t> > PUGIXML_FUNCTION as_wide(const std::basic_string<char, std::char_traits<char>, std::allocator<char> >& str);
@@ -1333,13 +1394,13 @@ namespace pugi

 	// Memory allocation function interface; returns pointer to allocated memory or NULL on failure
 	typedef void* (*allocation_function)(size_t size);
-	
+
 	// Memory deallocation function interface
 	typedef void (*deallocation_function)(void* ptr);

 	// Override default memory management functions. All subsequent allocations/deallocations will be performed via supplied functions.
 	void PUGIXML_FUNCTION set_memory_management_functions(allocation_function allocate, deallocation_function deallocate);
-	
+
 	// Get current memory management functions
 	allocation_function PUGIXML_FUNCTION get_memory_allocation_function();
 	deallocation_function PUGIXML_FUNCTION get_memory_deallocation_function();
@@ -1375,7 +1436,7 @@ namespace std
 #endif

 /**
- * Copyright (c) 2006-2015 Arseny Kapoulkine
+ * Copyright (c) 2006-2018 Arseny Kapoulkine
 *
 * Permission is hereby granted, free of charge, to any person
 * obtaining a copy of this software and associated documentation
@@ -1388,7 +1449,7 @@ namespace std
 *
 * The above copyright notice and this permission notice shall be
 * included in all copies or substantial portions of the Software.
- * 
+ *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
 * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
--- a/lib/pugixml/readme.txt
+++ b/lib/pugixml/readme.txt
@@ -1,6 +1,6 @@
-pugixml 1.6 - an XML processing library
+pugixml 1.9 - an XML processing library

-Copyright (C) 2006-2015, by Arseny Kapoulkine (arseny.kapoulkine@gmail.com)
+Copyright (C) 2006-2018, by Arseny Kapoulkine (arseny.kapoulkine@gmail.com)
 Report bugs and download new versions at http://pugixml.org/

 This is the distribution of pugixml, which is a C++ XML processing library,
@@ -28,7 +28,7 @@ The distribution contains the following folders:

 This library is distributed under the MIT License:

-Copyright (c) 2006-2015 Arseny Kapoulkine
+Copyright (c) 2006-2018 Arseny Kapoulkine

 Permission is hereby granted, free of charge, to any person
 obtaining a copy of this software and associated documentation
--- a/lib/qcd/action/fermion/CayleyFermion5D.cc
+++ b/lib/qcd/action/fermion/CayleyFermion5D.cc
@@ -52,6 +52,35 @@ namespace QCD {
 { 
 }

+///////////////////////////////////////////////////////////////
+// Physical surface field utilities
+///////////////////////////////////////////////////////////////
+template<class Impl>  
+void CayleyFermion5D<Impl>::ExportPhysicalFermionSolution(const FermionField &solution5d,FermionField &exported4d)
+{
+  int Ls = this->Ls;
+  FermionField tmp(this->FermionGrid());
+  tmp = solution5d;
+  conformable(solution5d._grid,this->FermionGrid());
+  conformable(exported4d._grid,this->GaugeGrid());
+  axpby_ssp_pminus(tmp, 0., solution5d, 1., solution5d, 0, 0);
+  axpby_ssp_pplus (tmp, 1., tmp       , 1., solution5d, 0, Ls-1);
+  ExtractSlice(exported4d, tmp, 0, 0);
+}
+template<class Impl>  
+void CayleyFermion5D<Impl>::ImportPhysicalFermionSource(const FermionField &input4d,FermionField &imported5d)
+{
+  int Ls = this->Ls;
+  FermionField tmp(this->FermionGrid());
+  conformable(imported5d._grid,this->FermionGrid());
+  conformable(input4d._grid   ,this->GaugeGrid());
+  tmp = zero;
+  InsertSlice(input4d, tmp, 0   , 0);
+  InsertSlice(input4d, tmp, Ls-1, 0);
+  axpby_ssp_pplus (tmp, 0., tmp, 1., tmp, 0, 0);
+  axpby_ssp_pminus(tmp, 0., tmp, 1., tmp, Ls-1, Ls-1);
+  Dminus(tmp,imported5d);
+}
 template<class Impl>  
 void CayleyFermion5D<Impl>::Dminus(const FermionField &psi, FermionField &chi)
 {
--- a/lib/qcd/action/fermion/CayleyFermion5D.h
+++ b/lib/qcd/action/fermion/CayleyFermion5D.h
@@ -83,8 +83,13 @@ namespace Grid {
      virtual void   M5D   (const FermionField &psi, FermionField &chi);
      virtual void   M5Ddag(const FermionField &psi, FermionField &chi);

+      ///////////////////////////////////////////////////////////////
+      // Physical surface field utilities
+      ///////////////////////////////////////////////////////////////
      virtual void   Dminus(const FermionField &psi, FermionField &chi);
      virtual void   DminusDag(const FermionField &psi, FermionField &chi);
+      virtual void ExportPhysicalFermionSolution(const FermionField &solution5d,FermionField &exported4d);
+      virtual void ImportPhysicalFermionSource(const FermionField &input4d,FermionField &imported5d);

      /////////////////////////////////////////////////////
      // Instantiate different versions depending on Impl
--- a/lib/qcd/action/fermion/ContinuedFractionFermion5D.cc
+++ b/lib/qcd/action/fermion/ContinuedFractionFermion5D.cc
@@ -295,6 +295,27 @@ namespace Grid {
      assert((Ls&0x1)==1); // Odd Ls required
    }

+    template<class Impl>
+    void ContinuedFractionFermion5D<Impl>::ExportPhysicalFermionSolution(const FermionField &solution5d,FermionField &exported4d)
+    {
+      int Ls = this->Ls;
+      conformable(solution5d._grid,this->FermionGrid());
+      conformable(exported4d._grid,this->GaugeGrid());
+      ExtractSlice(exported4d, solution5d, Ls-1, Ls-1);
+    }
+    template<class Impl>
+    void ContinuedFractionFermion5D<Impl>::ImportPhysicalFermionSource(const FermionField &input4d,FermionField &imported5d)
+    {
+      int Ls = this->Ls;
+      conformable(imported5d._grid,this->FermionGrid());
+      conformable(input4d._grid   ,this->GaugeGrid());
+      FermionField tmp(this->FermionGrid());
+      tmp=zero;
+      InsertSlice(input4d, tmp, Ls-1, Ls-1);
+      tmp=Gamma(Gamma::Algebra::Gamma5)*tmp;
+      this->Dminus(tmp,imported5d);
+    }
+
    FermOpTemplateInstantiate(ContinuedFractionFermion5D);

  }
--- a/lib/qcd/action/fermion/ContinuedFractionFermion5D.h
+++ b/lib/qcd/action/fermion/ContinuedFractionFermion5D.h
@@ -65,6 +65,14 @@ namespace Grid {
      // Efficient support for multigrid coarsening
      virtual void  Mdir (const FermionField &in, FermionField &out,int dir,int disp);

+      ///////////////////////////////////////////////////////////////
+      // Physical surface field utilities
+      ///////////////////////////////////////////////////////////////
+      //      virtual void Dminus(const FermionField &psi, FermionField &chi);     // Inherit trivial case
+      //      virtual void DminusDag(const FermionField &psi, FermionField &chi);  // Inherit trivial case
+      virtual void ExportPhysicalFermionSolution(const FermionField &solution5d,FermionField &exported4d);
+      virtual void ImportPhysicalFermionSource  (const FermionField &input4d,FermionField &imported5d);
+
      // Constructors
      ContinuedFractionFermion5D(GaugeField &_Umu,
 				 GridCartesian         &FiveDimGrid,
--- a/lib/qcd/action/fermion/FermionOperator.h
+++ b/lib/qcd/action/fermion/FermionOperator.h
@@ -128,6 +128,19 @@ namespace Grid {
                                       std::vector<Real> mom,
                                       unsigned int tmin, 
                                       unsigned int tmax)=0;
+      ///////////////////////////////////////////////
+      // Physical field import/export
+      ///////////////////////////////////////////////
+      virtual void Dminus(const FermionField &psi, FermionField &chi)    { chi=psi; }
+      virtual void DminusDag(const FermionField &psi, FermionField &chi) { chi=psi; }
+      virtual void ImportPhysicalFermionSource(const FermionField &input,FermionField &imported)
+      {
+	imported = input;
+      };
+      virtual void ExportPhysicalFermionSolution(const FermionField &solution,FermionField &exported)
+      {
+	exported=solution;
+      };
    };

  }
--- a/lib/qcd/action/fermion/PartialFractionFermion5D.cc
+++ b/lib/qcd/action/fermion/PartialFractionFermion5D.cc
@@ -396,6 +396,27 @@ namespace Grid {
      amax=zolo_hi;
    }

+    template<class Impl>
+    void PartialFractionFermion5D<Impl>::ExportPhysicalFermionSolution(const FermionField &solution5d,FermionField &exported4d)
+    {
+      int Ls = this->Ls;
+      conformable(solution5d._grid,this->FermionGrid());
+      conformable(exported4d._grid,this->GaugeGrid());
+      ExtractSlice(exported4d, solution5d, Ls-1, Ls-1);
+    }
+    template<class Impl>
+    void PartialFractionFermion5D<Impl>::ImportPhysicalFermionSource(const FermionField &input4d,FermionField &imported5d)
+    {
+      int Ls = this->Ls;
+      conformable(imported5d._grid,this->FermionGrid());
+      conformable(input4d._grid   ,this->GaugeGrid());
+      FermionField tmp(this->FermionGrid());
+      tmp=zero;
+      InsertSlice(input4d, tmp, Ls-1, Ls-1);
+      tmp=Gamma(Gamma::Algebra::Gamma5)*tmp;
+      this->Dminus(tmp,imported5d);
+    }
+
      // Constructors
    template<class Impl>
    PartialFractionFermion5D<Impl>::PartialFractionFermion5D(GaugeField &_Umu,
--- a/lib/qcd/action/fermion/PartialFractionFermion5D.h
+++ b/lib/qcd/action/fermion/PartialFractionFermion5D.h
@@ -70,6 +70,12 @@ namespace Grid {
      // Efficient support for multigrid coarsening
      virtual void  Mdir (const FermionField &in, FermionField &out,int dir,int disp);

+      ///////////////////////////////////////////////////////////////
+      // Physical surface field utilities
+      ///////////////////////////////////////////////////////////////
+      virtual void ExportPhysicalFermionSolution(const FermionField &solution5d,FermionField &exported4d);
+      virtual void ImportPhysicalFermionSource  (const FermionField &input4d,FermionField &imported5d);
+
      // Constructors
      PartialFractionFermion5D(GaugeField &_Umu,
 			       GridCartesian         &FiveDimGrid,
--- a/lib/qcd/action/fermion/WilsonCompressor.h
+++ b/lib/qcd/action/fermion/WilsonCompressor.h
@@ -69,39 +69,47 @@ class WilsonCompressorTemplate< _HCspinor, _Hspinor, _Spinor, projector,
  /*****************************************************/
  /* Compress includes precision change if mpi data is not same */
  /*****************************************************/
-  inline void Compress(SiteHalfSpinor *buf,Integer o,const SiteSpinor &in) {
-    projector::Proj(buf[o],in,mu,dag);
+  inline void Compress(SiteHalfSpinor * __restrict__ buf,Integer o,const SiteSpinor &in) {
+    SiteHalfSpinor tmp;
+    projector::Proj(tmp,in,mu,dag);
+    vstream(buf[o],tmp);
  }

  /*****************************************************/
  /* Exchange includes precision change if mpi data is not same */
  /*****************************************************/
-  inline void Exchange(SiteHalfSpinor *mp,
-                       SiteHalfSpinor *vp0,
-                       SiteHalfSpinor *vp1,
+  inline void Exchange(SiteHalfSpinor * __restrict__ mp,
+                       const SiteHalfSpinor * __restrict__ vp0,
+                       const SiteHalfSpinor * __restrict__ vp1,
 		       Integer type,Integer o){
-    exchange(mp[2*o],mp[2*o+1],vp0[o],vp1[o],type);
+    SiteHalfSpinor tmp1;
+    SiteHalfSpinor tmp2;
+    exchange(tmp1,tmp2,vp0[o],vp1[o],type);
+    vstream(mp[2*o  ],tmp1);
+    vstream(mp[2*o+1],tmp2);
  }

  /*****************************************************/
  /* Have a decompression step if mpi data is not same */
  /*****************************************************/
-  inline void Decompress(SiteHalfSpinor *out,
-			 SiteHalfSpinor *in, Integer o) {    
+  inline void Decompress(SiteHalfSpinor * __restrict__ out,
+			 SiteHalfSpinor * __restrict__ in, Integer o) {    
    assert(0);
  }

  /*****************************************************/
  /* Compress Exchange                                 */
  /*****************************************************/
-  inline void CompressExchange(SiteHalfSpinor *out0,
-			       SiteHalfSpinor *out1,
-			       const SiteSpinor *in,
+  inline void CompressExchange(SiteHalfSpinor * __restrict__ out0,
+			       SiteHalfSpinor * __restrict__ out1,
+			       const SiteSpinor * __restrict__ in,
 			       Integer j,Integer k, Integer m,Integer type){
    SiteHalfSpinor temp1, temp2,temp3,temp4;
    projector::Proj(temp1,in[k],mu,dag);
    projector::Proj(temp2,in[m],mu,dag);
-    exchange(out0[j],out1[j],temp1,temp2,type);
+    exchange(temp3,temp4,temp1,temp2,type);
+    vstream(out0[j],temp3);
+    vstream(out1[j],temp4);
  }

  /*****************************************************/
--- a/lib/qcd/action/fermion/WilsonKernelsHand.cc
+++ b/lib/qcd/action/fermion/WilsonKernelsHand.cc
@@ -30,181 +30,60 @@ Author: paboyle <paboyle@ph.ed.ac.uk>

 #define REGISTER

-#define LOAD_CHIMU_BODY(F)			\
-  Chimu_00=ref(F)(0)(0);			\
-  Chimu_01=ref(F)(0)(1);			\
-  Chimu_02=ref(F)(0)(2);			\
-  Chimu_10=ref(F)(1)(0);			\
-  Chimu_11=ref(F)(1)(1);			\
-  Chimu_12=ref(F)(1)(2);			\
-  Chimu_20=ref(F)(2)(0);			\
-  Chimu_21=ref(F)(2)(1);			\
-  Chimu_22=ref(F)(2)(2);			\
-  Chimu_30=ref(F)(3)(0);			\
-  Chimu_31=ref(F)(3)(1);			\
-  Chimu_32=ref(F)(3)(2)
+#define LOAD_CHIMU \
+  {const SiteSpinor & ref (in._odata[offset]);	\
+    Chimu_00=ref()(0)(0);\
+    Chimu_01=ref()(0)(1);\
+    Chimu_02=ref()(0)(2);\
+    Chimu_10=ref()(1)(0);\
+    Chimu_11=ref()(1)(1);\
+    Chimu_12=ref()(1)(2);\
+    Chimu_20=ref()(2)(0);\
+    Chimu_21=ref()(2)(1);\
+    Chimu_22=ref()(2)(2);\
+    Chimu_30=ref()(3)(0);\
+    Chimu_31=ref()(3)(1);\
+    Chimu_32=ref()(3)(2);}

-#define LOAD_CHIMU(DIR,F,PERM)						\
-  { const SiteSpinor & ref (in._odata[offset]); LOAD_CHIMU_BODY(F); }
-
-#define LOAD_CHI_BODY(F)				\
-    Chi_00 = ref(F)(0)(0);\
-    Chi_01 = ref(F)(0)(1);\
-    Chi_02 = ref(F)(0)(2);\
-    Chi_10 = ref(F)(1)(0);\
-    Chi_11 = ref(F)(1)(1);\
-    Chi_12 = ref(F)(1)(2)
-
-#define LOAD_CHI(DIR,F,PERM)					\
-  {const SiteHalfSpinor &ref(buf[offset]); LOAD_CHI_BODY(F); }
-
-
-//G-parity implementations using in-place intrinsic ops
-
-//1l 1h -> 1h 1l
-//0l 0h , 1h 1l -> 0l 1h 0h,1l
-//0h,1l -> 1l,0h
-//if( (distance == 1 && !perm_will_occur) || (distance == -1 && perm_will_occur) )
-//Pulled fermion through forwards face, GPBC on upper component
-//Need 0= 0l 1h   1= 1l 0h
-//else if( (distance == -1 && !perm) || (distance == 1 && perm) )
-//Pulled fermion through backwards face, GPBC on lower component
-//Need 0= 1l 0h   1= 0l 1h
-
-//1l 1h -> 1h 1l
-//0l 0h , 1h 1l -> 0l 1h 0h,1l
-#define DO_TWIST_0L_1H(INTO,S,C,F, PERM, tmp1, tmp2, tmp3)			\
-  permute##PERM(tmp1, ref(1)(S)(C));				\
-  exchange##PERM(tmp2,tmp3, ref(0)(S)(C), tmp1);		\
-  INTO = tmp2;
-
-//0l 0h -> 0h 0l
-//1l 1h, 0h 0l -> 1l 0h, 1h 0l
-#define DO_TWIST_1L_0H(INTO,S,C,F, PERM, tmp1, tmp2, tmp3)			\
-  permute##PERM(tmp1, ref(0)(S)(C));				\
-  exchange##PERM(tmp2,tmp3, ref(1)(S)(C), tmp1);		\
-  INTO = tmp2;
-
-
-
-
-#define LOAD_CHI_SETUP(DIR,F)						\
-  g = F;								\
-  direction = st._directions[DIR];				\
-  distance = st._distances[DIR];				\
-  sl = st._grid->_simd_layout[direction];			\
-  inplace_twist = 0;						\
-  if(SE->_around_the_world && this->Params.twists[DIR % 4]){		\
-    if(sl == 1){							\
-      g = (F+1) % 2;							\
-    }else{								\
-      inplace_twist = 1;						\
-    }									\
-  }  
-
-#define LOAD_CHIMU_GPARITY_INPLACE_TWIST(DIR,F,PERM)			\
-  { const SiteSpinor &ref(in._odata[offset]);				\
-    LOAD_CHI_SETUP(DIR,F);						\
-    if(!inplace_twist){							\
-      LOAD_CHIMU_BODY(g);						\
-    }else{								\
-      if(  ( F==0 && ((distance == 1 && !perm) || (distance == -1 && perm)) ) || \
-	   ( F==1 && ((distance == -1 && !perm) || (distance == 1 && perm)) ) ){ \
-	DO_TWIST_0L_1H(Chimu_00,0,0,F,PERM,  U_00,U_01,U_10);		\
-	DO_TWIST_0L_1H(Chimu_01,0,1,F,PERM,  U_11,U_20,U_21);		\
-	DO_TWIST_0L_1H(Chimu_02,0,2,F,PERM,  U_00,U_01,U_10);		\
-	DO_TWIST_0L_1H(Chimu_10,1,0,F,PERM,  U_11,U_20,U_21);		\
-	DO_TWIST_0L_1H(Chimu_11,1,1,F,PERM,  U_00,U_01,U_10);		\
-	DO_TWIST_0L_1H(Chimu_12,1,2,F,PERM,  U_11,U_20,U_21);		\
-	DO_TWIST_0L_1H(Chimu_20,2,0,F,PERM,  U_00,U_01,U_10);		\
-	DO_TWIST_0L_1H(Chimu_21,2,1,F,PERM,  U_11,U_20,U_21);		\
-	DO_TWIST_0L_1H(Chimu_22,2,2,F,PERM,  U_00,U_01,U_10);		\
-	DO_TWIST_0L_1H(Chimu_30,3,0,F,PERM,  U_11,U_20,U_21);		\
-	DO_TWIST_0L_1H(Chimu_31,3,1,F,PERM,  U_00,U_01,U_10);		\
-	DO_TWIST_0L_1H(Chimu_32,3,2,F,PERM,  U_11,U_20,U_21);		\
-      }else{								\
-	DO_TWIST_1L_0H(Chimu_00,0,0,F,PERM,  U_00,U_01,U_10);		\
-	DO_TWIST_1L_0H(Chimu_01,0,1,F,PERM,  U_11,U_20,U_21);		\
-	DO_TWIST_1L_0H(Chimu_02,0,2,F,PERM,  U_00,U_01,U_10);		\
-	DO_TWIST_1L_0H(Chimu_10,1,0,F,PERM,  U_11,U_20,U_21);		\
-	DO_TWIST_1L_0H(Chimu_11,1,1,F,PERM,  U_00,U_01,U_10);		\
-	DO_TWIST_1L_0H(Chimu_12,1,2,F,PERM,  U_11,U_20,U_21);		\
-	DO_TWIST_1L_0H(Chimu_20,2,0,F,PERM,  U_00,U_01,U_10);		\
-	DO_TWIST_1L_0H(Chimu_21,2,1,F,PERM,  U_11,U_20,U_21);		\
-	DO_TWIST_1L_0H(Chimu_22,2,2,F,PERM,  U_00,U_01,U_10);		\
-	DO_TWIST_1L_0H(Chimu_30,3,0,F,PERM,  U_11,U_20,U_21);		\
-	DO_TWIST_1L_0H(Chimu_31,3,1,F,PERM,  U_00,U_01,U_10);		\
-	DO_TWIST_1L_0H(Chimu_32,3,2,F,PERM,  U_11,U_20,U_21);		\
-      } \
-    } \
-  }
-
-
-#define LOAD_CHI_GPARITY_INPLACE_TWIST(DIR,F,PERM)				\
-  { const SiteHalfSpinor &ref(buf[offset]);				\
-    LOAD_CHI_SETUP(DIR,F);						\
-    if(!inplace_twist){							\
-      LOAD_CHI_BODY(g);							\
-    }else{								\
-      if(  ( F==0 && ((distance == 1 && !perm) || (distance == -1 && perm)) ) || \
-	   ( F==1 && ((distance == -1 && !perm) || (distance == 1 && perm)) ) ){ \
-	DO_TWIST_0L_1H(Chi_00,0,0,F,PERM,  U_00,U_01,U_10);			\
-	DO_TWIST_0L_1H(Chi_01,0,1,F,PERM,  U_11,U_20,U_21);			\
-	DO_TWIST_0L_1H(Chi_02,0,2,F,PERM,  UChi_00,UChi_01,UChi_02);		\
-	DO_TWIST_0L_1H(Chi_10,1,0,F,PERM,  UChi_10,UChi_11,UChi_12);		\
-	DO_TWIST_0L_1H(Chi_11,1,1,F,PERM,  U_00,U_01,U_10);			\
-	DO_TWIST_0L_1H(Chi_12,1,2,F,PERM,  U_11,U_20,U_21);			\
-      }else{								\
-	DO_TWIST_1L_0H(Chi_00,0,0,F,PERM,  U_00,U_01,U_10);			\
-	DO_TWIST_1L_0H(Chi_01,0,1,F,PERM,  U_11,U_20,U_21);			\
-	DO_TWIST_1L_0H(Chi_02,0,2,F,PERM,  UChi_00,UChi_01,UChi_02);		\
-	DO_TWIST_1L_0H(Chi_10,1,0,F,PERM,  UChi_10,UChi_11,UChi_12);		\
-	DO_TWIST_1L_0H(Chi_11,1,1,F,PERM,  U_00,U_01,U_10);			\
-	DO_TWIST_1L_0H(Chi_12,1,2,F,PERM,  U_11,U_20,U_21);			\
-      }									\
-    }									\
-  }
-
-
-#define LOAD_CHI_GPARITY(DIR,F,PERM) LOAD_CHI_GPARITY_INPLACE_TWIST(DIR,F,PERM)
-#define LOAD_CHIMU_GPARITY(DIR,F,PERM) LOAD_CHIMU_GPARITY_INPLACE_TWIST(DIR,F,PERM)
+#define LOAD_CHI\
+  {const SiteHalfSpinor &ref(buf[offset]);	\
+    Chi_00 = ref()(0)(0);\
+    Chi_01 = ref()(0)(1);\
+    Chi_02 = ref()(0)(2);\
+    Chi_10 = ref()(1)(0);\
+    Chi_11 = ref()(1)(1);\
+    Chi_12 = ref()(1)(2);}

 // To splat or not to splat depends on the implementation
-#define MULT_2SPIN_BODY \
-  Impl::loadLinkElement(U_00,ref()(0,0));	\
-  Impl::loadLinkElement(U_10,ref()(1,0));	\
-  Impl::loadLinkElement(U_20,ref()(2,0));	\
-  Impl::loadLinkElement(U_01,ref()(0,1));	\
-  Impl::loadLinkElement(U_11,ref()(1,1));	\
-  Impl::loadLinkElement(U_21,ref()(2,1));	\
-  UChi_00 = U_00*Chi_00;			\
-  UChi_10 = U_00*Chi_10;			\
-  UChi_01 = U_10*Chi_00;			\
-  UChi_11 = U_10*Chi_10;			\
-  UChi_02 = U_20*Chi_00;			\
-  UChi_12 = U_20*Chi_10;			\
-  UChi_00+= U_01*Chi_01;			\
-  UChi_10+= U_01*Chi_11;			\
-  UChi_01+= U_11*Chi_01;			\
-  UChi_11+= U_11*Chi_11;			\
-  UChi_02+= U_21*Chi_01;			\
-  UChi_12+= U_21*Chi_11;			\
-  Impl::loadLinkElement(U_00,ref()(0,2));	\
-  Impl::loadLinkElement(U_10,ref()(1,2));	\
-  Impl::loadLinkElement(U_20,ref()(2,2));	\
-  UChi_00+= U_00*Chi_02;			\
-  UChi_10+= U_00*Chi_12;			\
-  UChi_01+= U_10*Chi_02;			\
-  UChi_11+= U_10*Chi_12;			\
-  UChi_02+= U_20*Chi_02;			\
-  UChi_12+= U_20*Chi_12
-
-
-#define MULT_2SPIN(A,F)					\
-  {auto & ref(U._odata[sU](A)); MULT_2SPIN_BODY; }
-
-#define MULT_2SPIN_GPARITY(A,F)				\
-  {auto & ref(U._odata[sU](F)(A)); MULT_2SPIN_BODY; }
+#define MULT_2SPIN(A)\
+  {auto & ref(U._odata[sU](A));			\
+   Impl::loadLinkElement(U_00,ref()(0,0));	\
+   Impl::loadLinkElement(U_10,ref()(1,0));	\
+   Impl::loadLinkElement(U_20,ref()(2,0));	\
+   Impl::loadLinkElement(U_01,ref()(0,1));	\
+   Impl::loadLinkElement(U_11,ref()(1,1));	\
+   Impl::loadLinkElement(U_21,ref()(2,1));	\
+    UChi_00 = U_00*Chi_00;\
+    UChi_10 = U_00*Chi_10;\
+    UChi_01 = U_10*Chi_00;\
+    UChi_11 = U_10*Chi_10;\
+    UChi_02 = U_20*Chi_00;\
+    UChi_12 = U_20*Chi_10;\
+    UChi_00+= U_01*Chi_01;\
+    UChi_10+= U_01*Chi_11;\
+    UChi_01+= U_11*Chi_01;\
+    UChi_11+= U_11*Chi_11;\
+    UChi_02+= U_21*Chi_01;\
+    UChi_12+= U_21*Chi_11;\
+    Impl::loadLinkElement(U_00,ref()(0,2));	\
+    Impl::loadLinkElement(U_10,ref()(1,2));	\
+    Impl::loadLinkElement(U_20,ref()(2,2));	\
+    UChi_00+= U_00*Chi_02;\
+    UChi_10+= U_00*Chi_12;\
+    UChi_01+= U_10*Chi_02;\
+    UChi_11+= U_10*Chi_12;\
+    UChi_02+= U_20*Chi_02;\
+    UChi_12+= U_20*Chi_12;}


 #define PERMUTE_DIR(dir)			\
@@ -428,87 +307,84 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
  result_31-= UChi_11;	\
  result_32-= UChi_12;

-#define HAND_STENCIL_LEG(PROJ,PERM,DIR,RECON,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \
+#define HAND_STENCIL_LEG(PROJ,PERM,DIR,RECON)	\
  SE=st.GetEntry(ptype,DIR,ss);			\
  offset = SE->_offset;				\
  local  = SE->_is_local;			\
  perm   = SE->_permute;			\
  if ( local ) {				\
-    LOAD_CHIMU_IMPL(DIR,F,PERM);			\
+    LOAD_CHIMU;					\
    PROJ;					\
    if ( perm) {				\
      PERMUTE_DIR(PERM);			\
    }						\
  } else {					\
-    LOAD_CHI_IMPL(DIR,F,PERM);			\
+    LOAD_CHI;					\
  }						\
-  MULT_2SPIN_IMPL(DIR,F);			\
+  MULT_2SPIN(DIR);				\
  RECON;					

-
-#define HAND_STENCIL_LEG_INT(PROJ,PERM,DIR,RECON,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL)	\
+#define HAND_STENCIL_LEG_INT(PROJ,PERM,DIR,RECON)	\
  SE=st.GetEntry(ptype,DIR,ss);			\
  offset = SE->_offset;				\
  local  = SE->_is_local;			\
  perm   = SE->_permute;			\
  if ( local ) {				\
-    LOAD_CHIMU_IMPL(DIR,F,PERM);			\
+    LOAD_CHIMU;					\
    PROJ;					\
    if ( perm) {				\
      PERMUTE_DIR(PERM);			\
    }						\
  } else if ( st.same_node[DIR] ) {		\
-    LOAD_CHI_IMPL(DIR,F,PERM);			\
+    LOAD_CHI;					\
  }						\
  if (local || st.same_node[DIR] ) {		\
-    MULT_2SPIN_IMPL(DIR,F);			\
+    MULT_2SPIN(DIR);				\
    RECON;					\
  }

-#define HAND_STENCIL_LEG_EXT(PROJ,PERM,DIR,RECON,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL)	\
+#define HAND_STENCIL_LEG_EXT(PROJ,PERM,DIR,RECON)	\
  SE=st.GetEntry(ptype,DIR,ss);			\
  offset = SE->_offset;				\
-  local  = SE->_is_local;			\
-  perm   = SE->_permute;			\
  if((!SE->_is_local)&&(!st.same_node[DIR]) ) {	\
-    LOAD_CHI_IMPL(DIR,F,PERM);			\
-    MULT_2SPIN_IMPL(DIR,F);			\
+    LOAD_CHI;					\
+    MULT_2SPIN(DIR);				\
    RECON;					\
    nmu++;					\
  }

-#define HAND_RESULT(ss,F)			\
+#define HAND_RESULT(ss)				\
  {						\
    SiteSpinor & ref (out._odata[ss]);		\
-    vstream(ref(F)(0)(0),result_00);		\
-    vstream(ref(F)(0)(1),result_01);		\
-    vstream(ref(F)(0)(2),result_02);		\
-    vstream(ref(F)(1)(0),result_10);		\
-    vstream(ref(F)(1)(1),result_11);		\
-    vstream(ref(F)(1)(2),result_12);		\
-    vstream(ref(F)(2)(0),result_20);		\
-    vstream(ref(F)(2)(1),result_21);		\
-    vstream(ref(F)(2)(2),result_22);		\
-    vstream(ref(F)(3)(0),result_30);		\
-    vstream(ref(F)(3)(1),result_31);		\
-    vstream(ref(F)(3)(2),result_32);		\
+    vstream(ref()(0)(0),result_00);		\
+    vstream(ref()(0)(1),result_01);		\
+    vstream(ref()(0)(2),result_02);		\
+    vstream(ref()(1)(0),result_10);		\
+    vstream(ref()(1)(1),result_11);		\
+    vstream(ref()(1)(2),result_12);		\
+    vstream(ref()(2)(0),result_20);		\
+    vstream(ref()(2)(1),result_21);		\
+    vstream(ref()(2)(2),result_22);		\
+    vstream(ref()(3)(0),result_30);		\
+    vstream(ref()(3)(1),result_31);		\
+    vstream(ref()(3)(2),result_32);		\
  }

-#define HAND_RESULT_EXT(ss,F)			\
+#define HAND_RESULT_EXT(ss)			\
  if (nmu){					\
    SiteSpinor & ref (out._odata[ss]);		\
-    ref(F)(0)(0)+=result_00;		\
-    ref(F)(0)(1)+=result_01;		\
-    ref(F)(0)(2)+=result_02;		\
-    ref(F)(1)(0)+=result_10;		\
-    ref(F)(1)(1)+=result_11;		\
-    ref(F)(1)(2)+=result_12;		\
-    ref(F)(2)(0)+=result_20;		\
-    ref(F)(2)(1)+=result_21;		\
-    ref(F)(2)(2)+=result_22;		\
-    ref(F)(3)(0)+=result_30;		\
-    ref(F)(3)(1)+=result_31;		\
-    ref(F)(3)(2)+=result_32;		\
+    ref()(0)(0)+=result_00;		\
+    ref()(0)(1)+=result_01;		\
+    ref()(0)(2)+=result_02;		\
+    ref()(1)(0)+=result_10;		\
+    ref()(1)(1)+=result_11;		\
+    ref()(1)(2)+=result_12;		\
+    ref()(2)(0)+=result_20;		\
+    ref()(2)(1)+=result_21;		\
+    ref()(2)(2)+=result_22;		\
+    ref()(3)(0)+=result_30;		\
+    ref()(3)(1)+=result_31;		\
+    ref()(3)(2)+=result_32;		\
  }


@@ -587,18 +463,15 @@ WilsonKernels<Impl>::HandDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGauge
  int offset,local,perm, ptype;
  StencilEntry *SE;

-#define HAND_DOP_SITE(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \
-  HAND_STENCIL_LEG(XM_PROJ,3,Xp,XM_RECON,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG(YM_PROJ,2,Yp,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL);	\
-  HAND_STENCIL_LEG(ZM_PROJ,1,Zp,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG(TM_PROJ,0,Tp,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG(XP_PROJ,3,Xm,XP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG(YP_PROJ,2,Ym,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG(ZP_PROJ,1,Zm,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG(TP_PROJ,0,Tm,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_RESULT(ss,F)
-
-  HAND_DOP_SITE(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN);
+  HAND_STENCIL_LEG(XM_PROJ,3,Xp,XM_RECON);
+  HAND_STENCIL_LEG(YM_PROJ,2,Yp,YM_RECON_ACCUM);
+  HAND_STENCIL_LEG(ZM_PROJ,1,Zp,ZM_RECON_ACCUM);
+  HAND_STENCIL_LEG(TM_PROJ,0,Tp,TM_RECON_ACCUM);
+  HAND_STENCIL_LEG(XP_PROJ,3,Xm,XP_RECON_ACCUM);
+  HAND_STENCIL_LEG(YP_PROJ,2,Ym,YP_RECON_ACCUM);
+  HAND_STENCIL_LEG(ZP_PROJ,1,Zm,ZP_RECON_ACCUM);
+  HAND_STENCIL_LEG(TP_PROJ,0,Tm,TP_RECON_ACCUM);
+  HAND_RESULT(ss);
 }

 template<class Impl>
@@ -612,19 +485,16 @@ void WilsonKernels<Impl>::HandDhopSiteDag(StencilImpl &st,LebesgueOrder &lo,Doub

  StencilEntry *SE;
  int offset,local,perm, ptype;
-
-#define HAND_DOP_SITE_DAG(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \
-  HAND_STENCIL_LEG(XP_PROJ,3,Xp,XP_RECON,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG(YP_PROJ,2,Yp,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG(ZP_PROJ,1,Zp,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG(TP_PROJ,0,Tp,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG(XM_PROJ,3,Xm,XM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG(YM_PROJ,2,Ym,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG(ZM_PROJ,1,Zm,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG(TM_PROJ,0,Tm,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_RESULT(ss,F)
-
-  HAND_DOP_SITE_DAG(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN);
+  
+  HAND_STENCIL_LEG(XP_PROJ,3,Xp,XP_RECON);
+  HAND_STENCIL_LEG(YP_PROJ,2,Yp,YP_RECON_ACCUM);
+  HAND_STENCIL_LEG(ZP_PROJ,1,Zp,ZP_RECON_ACCUM);
+  HAND_STENCIL_LEG(TP_PROJ,0,Tp,TP_RECON_ACCUM);
+  HAND_STENCIL_LEG(XM_PROJ,3,Xm,XM_RECON_ACCUM);
+  HAND_STENCIL_LEG(YM_PROJ,2,Ym,YM_RECON_ACCUM);
+  HAND_STENCIL_LEG(ZM_PROJ,1,Zm,ZM_RECON_ACCUM);
+  HAND_STENCIL_LEG(TM_PROJ,0,Tm,TM_RECON_ACCUM);
+  HAND_RESULT(ss);
 }

 template<class Impl> void 
@@ -639,20 +509,16 @@ WilsonKernels<Impl>::HandDhopSiteInt(StencilImpl &st,LebesgueOrder &lo,DoubledGa

  int offset,local,perm, ptype;
  StencilEntry *SE;
-
-#define HAND_DOP_SITE_INT(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \
-  ZERO_RESULT; \
-  HAND_STENCIL_LEG_INT(XM_PROJ,3,Xp,XM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG_INT(YM_PROJ,2,Yp,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG_INT(ZM_PROJ,1,Zp,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG_INT(TM_PROJ,0,Tp,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG_INT(XP_PROJ,3,Xm,XP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG_INT(YP_PROJ,2,Ym,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG_INT(ZP_PROJ,1,Zm,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG_INT(TP_PROJ,0,Tm,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_RESULT(ss,F)
-
-  HAND_DOP_SITE_INT(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN);
+  ZERO_RESULT;
+  HAND_STENCIL_LEG_INT(XM_PROJ,3,Xp,XM_RECON_ACCUM);
+  HAND_STENCIL_LEG_INT(YM_PROJ,2,Yp,YM_RECON_ACCUM);
+  HAND_STENCIL_LEG_INT(ZM_PROJ,1,Zp,ZM_RECON_ACCUM);
+  HAND_STENCIL_LEG_INT(TM_PROJ,0,Tp,TM_RECON_ACCUM);
+  HAND_STENCIL_LEG_INT(XP_PROJ,3,Xm,XP_RECON_ACCUM);
+  HAND_STENCIL_LEG_INT(YP_PROJ,2,Ym,YP_RECON_ACCUM);
+  HAND_STENCIL_LEG_INT(ZP_PROJ,1,Zm,ZP_RECON_ACCUM);
+  HAND_STENCIL_LEG_INT(TP_PROJ,0,Tm,TP_RECON_ACCUM);
+  HAND_RESULT(ss);
 }

 template<class Impl>
@@ -666,20 +532,16 @@ void WilsonKernels<Impl>::HandDhopSiteDagInt(StencilImpl &st,LebesgueOrder &lo,D

  StencilEntry *SE;
  int offset,local,perm, ptype;
-
-#define HAND_DOP_SITE_DAG_INT(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL)				\
-  ZERO_RESULT;							\
-  HAND_STENCIL_LEG_INT(XP_PROJ,3,Xp,XP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL);		\
-  HAND_STENCIL_LEG_INT(YP_PROJ,2,Yp,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL);		\
-  HAND_STENCIL_LEG_INT(ZP_PROJ,1,Zp,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL);		\
-  HAND_STENCIL_LEG_INT(TP_PROJ,0,Tp,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL);		\
-  HAND_STENCIL_LEG_INT(XM_PROJ,3,Xm,XM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL);		\
-  HAND_STENCIL_LEG_INT(YM_PROJ,2,Ym,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL);		\
-  HAND_STENCIL_LEG_INT(ZM_PROJ,1,Zm,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL);		\
-  HAND_STENCIL_LEG_INT(TM_PROJ,0,Tm,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL);		\
-  HAND_RESULT(ss,F)
-  
-  HAND_DOP_SITE_DAG_INT(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN);
+  ZERO_RESULT;
+  HAND_STENCIL_LEG_INT(XP_PROJ,3,Xp,XP_RECON_ACCUM);
+  HAND_STENCIL_LEG_INT(YP_PROJ,2,Yp,YP_RECON_ACCUM);
+  HAND_STENCIL_LEG_INT(ZP_PROJ,1,Zp,ZP_RECON_ACCUM);
+  HAND_STENCIL_LEG_INT(TP_PROJ,0,Tp,TP_RECON_ACCUM);
+  HAND_STENCIL_LEG_INT(XM_PROJ,3,Xm,XM_RECON_ACCUM);
+  HAND_STENCIL_LEG_INT(YM_PROJ,2,Ym,YM_RECON_ACCUM);
+  HAND_STENCIL_LEG_INT(ZM_PROJ,1,Zm,ZM_RECON_ACCUM);
+  HAND_STENCIL_LEG_INT(TM_PROJ,0,Tm,TM_RECON_ACCUM);
+  HAND_RESULT(ss);
 }

 template<class Impl> void 
@@ -695,20 +557,16 @@ WilsonKernels<Impl>::HandDhopSiteExt(StencilImpl &st,LebesgueOrder &lo,DoubledGa
  int offset,local,perm, ptype;
  StencilEntry *SE;
  int nmu=0;
-
-#define HAND_DOP_SITE_EXT(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \
-  ZERO_RESULT; \
-  HAND_STENCIL_LEG_EXT(XM_PROJ,3,Xp,XM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG_EXT(YM_PROJ,2,Yp,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG_EXT(ZM_PROJ,1,Zp,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG_EXT(TM_PROJ,0,Tp,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG_EXT(XP_PROJ,3,Xm,XP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG_EXT(YP_PROJ,2,Ym,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG_EXT(ZP_PROJ,1,Zm,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG_EXT(TP_PROJ,0,Tm,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_RESULT_EXT(ss,F)
-
-  HAND_DOP_SITE_EXT(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN);
+  ZERO_RESULT;
+  HAND_STENCIL_LEG_EXT(XM_PROJ,3,Xp,XM_RECON_ACCUM);
+  HAND_STENCIL_LEG_EXT(YM_PROJ,2,Yp,YM_RECON_ACCUM);
+  HAND_STENCIL_LEG_EXT(ZM_PROJ,1,Zp,ZM_RECON_ACCUM);
+  HAND_STENCIL_LEG_EXT(TM_PROJ,0,Tp,TM_RECON_ACCUM);
+  HAND_STENCIL_LEG_EXT(XP_PROJ,3,Xm,XP_RECON_ACCUM);
+  HAND_STENCIL_LEG_EXT(YP_PROJ,2,Ym,YP_RECON_ACCUM);
+  HAND_STENCIL_LEG_EXT(ZP_PROJ,1,Zm,ZP_RECON_ACCUM);
+  HAND_STENCIL_LEG_EXT(TP_PROJ,0,Tm,TP_RECON_ACCUM);
+  HAND_RESULT_EXT(ss);
 }

 template<class Impl>
@@ -723,193 +581,18 @@ void WilsonKernels<Impl>::HandDhopSiteDagExt(StencilImpl &st,LebesgueOrder &lo,D
  StencilEntry *SE;
  int offset,local,perm, ptype;
  int nmu=0;
-
-#define HAND_DOP_SITE_DAG_EXT(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \
-  ZERO_RESULT; \
-  HAND_STENCIL_LEG_EXT(XP_PROJ,3,Xp,XP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG_EXT(YP_PROJ,2,Yp,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG_EXT(ZP_PROJ,1,Zp,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG_EXT(TP_PROJ,0,Tp,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG_EXT(XM_PROJ,3,Xm,XM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG_EXT(YM_PROJ,2,Ym,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG_EXT(ZM_PROJ,1,Zm,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_STENCIL_LEG_EXT(TM_PROJ,0,Tm,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
-  HAND_RESULT_EXT(ss,F)
-
-  HAND_DOP_SITE_DAG_EXT(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN);
+  ZERO_RESULT;
+  HAND_STENCIL_LEG_EXT(XP_PROJ,3,Xp,XP_RECON_ACCUM);
+  HAND_STENCIL_LEG_EXT(YP_PROJ,2,Yp,YP_RECON_ACCUM);
+  HAND_STENCIL_LEG_EXT(ZP_PROJ,1,Zp,ZP_RECON_ACCUM);
+  HAND_STENCIL_LEG_EXT(TP_PROJ,0,Tp,TP_RECON_ACCUM);
+  HAND_STENCIL_LEG_EXT(XM_PROJ,3,Xm,XM_RECON_ACCUM);
+  HAND_STENCIL_LEG_EXT(YM_PROJ,2,Ym,YM_RECON_ACCUM);
+  HAND_STENCIL_LEG_EXT(ZM_PROJ,1,Zm,ZM_RECON_ACCUM);
+  HAND_STENCIL_LEG_EXT(TM_PROJ,0,Tm,TM_RECON_ACCUM);
+  HAND_RESULT_EXT(ss);
 }

-  ////////////////////////////////////////////////
-  // Specialise Gparity to simple implementation
-  ////////////////////////////////////////////////
-#define HAND_SPECIALISE_EMPTY(IMPL)					\
-  template<> void							\
-  WilsonKernels<IMPL>::HandDhopSite(StencilImpl &st,			\
-				    LebesgueOrder &lo,			\
-				    DoubledGaugeField &U,		\
-				    SiteHalfSpinor *buf,		\
-				    int sF,int sU,			\
-				    const FermionField &in,		\
-				    FermionField &out){ assert(0); }	\
-  template<> void							\
-  WilsonKernels<IMPL>::HandDhopSiteDag(StencilImpl &st,			\
-				    LebesgueOrder &lo,			\
-				    DoubledGaugeField &U,		\
-				    SiteHalfSpinor *buf,		\
-				    int sF,int sU,			\
-				    const FermionField &in,		\
-				    FermionField &out){ assert(0); }	\
-  template<> void							\
-  WilsonKernels<IMPL>::HandDhopSiteInt(StencilImpl &st,			\
-				    LebesgueOrder &lo,			\
-				    DoubledGaugeField &U,		\
-				    SiteHalfSpinor *buf,		\
-				    int sF,int sU,			\
-				    const FermionField &in,		\
-				    FermionField &out){ assert(0); }	\
-  template<> void							\
-  WilsonKernels<IMPL>::HandDhopSiteExt(StencilImpl &st,			\
-				    LebesgueOrder &lo,			\
-				    DoubledGaugeField &U,		\
-				    SiteHalfSpinor *buf,		\
-				    int sF,int sU,			\
-				    const FermionField &in,		\
-				    FermionField &out){ assert(0); }	\
-  template<> void							\
-  WilsonKernels<IMPL>::HandDhopSiteDagInt(StencilImpl &st,	       	\
-				    LebesgueOrder &lo,			\
-				    DoubledGaugeField &U,		\
-				    SiteHalfSpinor *buf,		\
-				    int sF,int sU,			\
-				    const FermionField &in,		\
-				    FermionField &out){ assert(0); }	\
-  template<> void							\
-  WilsonKernels<IMPL>::HandDhopSiteDagExt(StencilImpl &st,	       	\
-				    LebesgueOrder &lo,			\
-				    DoubledGaugeField &U,		\
-				    SiteHalfSpinor *buf,		\
-				    int sF,int sU,			\
-				    const FermionField &in,		\
-				    FermionField &out){ assert(0); }	\
-
-
-
-#define HAND_SPECIALISE_GPARITY(IMPL)					\
-  template<> void							\
-  WilsonKernels<IMPL>::HandDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor  *buf, \
-				    int ss,int sU,const FermionField &in, FermionField &out) \
-  {									\
-    typedef IMPL Impl;							\
-    typedef typename Simd::scalar_type S;				\
-    typedef typename Simd::vector_type V;				\
-									\
-    HAND_DECLARATIONS(ignore);						\
-									\
-    int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist; \
-    StencilEntry *SE;							\
-    HAND_DOP_SITE(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
-    HAND_DOP_SITE(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
-  }									\
-									\
-  template<>								\
-  void WilsonKernels<IMPL>::HandDhopSiteDag(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
-					    int ss,int sU,const FermionField &in, FermionField &out) \
-  {									\
-    typedef IMPL Impl;							\
-    typedef typename Simd::scalar_type S;				\
-    typedef typename Simd::vector_type V;				\
-									\
-    HAND_DECLARATIONS(ignore);						\
-									\
-    StencilEntry *SE;							\
-    int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist;					\
-    HAND_DOP_SITE_DAG(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
-    HAND_DOP_SITE_DAG(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
-  }									\
-									\
-  template<> void							\
-  WilsonKernels<IMPL>::HandDhopSiteInt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor  *buf, \
-						     int ss,int sU,const FermionField &in, FermionField &out) \
-  {									\
-    typedef IMPL Impl;							\
-    typedef typename Simd::scalar_type S;				\
-    typedef typename Simd::vector_type V;				\
-									\
-    HAND_DECLARATIONS(ignore);						\
-									\
-    int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist;					\
-    StencilEntry *SE;							\
-    HAND_DOP_SITE_INT(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
-    HAND_DOP_SITE_INT(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
-  }									\
-									\
-  template<>								\
-  void WilsonKernels<IMPL>::HandDhopSiteDagInt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
-							     int ss,int sU,const FermionField &in, FermionField &out) \
-  {									\
-    typedef IMPL Impl;							\
-    typedef typename Simd::scalar_type S;				\
-    typedef typename Simd::vector_type V;				\
-									\
-    HAND_DECLARATIONS(ignore);						\
-									\
-    StencilEntry *SE;							\
-    int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist; \
-    HAND_DOP_SITE_DAG_INT(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
-    HAND_DOP_SITE_DAG_INT(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
-  }									\
-									\
-  template<> void							\
-  WilsonKernels<IMPL>::HandDhopSiteExt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor  *buf, \
-						     int ss,int sU,const FermionField &in, FermionField &out) \
-  {									\
-    typedef IMPL Impl;							\
-    typedef typename Simd::scalar_type S;				\
-    typedef typename Simd::vector_type V;				\
-									\
-    HAND_DECLARATIONS(ignore);						\
-									\
-    int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist; \
-    StencilEntry *SE;							\
-    int nmu=0;								\
-    HAND_DOP_SITE_EXT(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
-    nmu = 0;								\
-    HAND_DOP_SITE_EXT(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
-  }									\
-  template<>								\
-  void WilsonKernels<IMPL>::HandDhopSiteDagExt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
-							     int ss,int sU,const FermionField &in, FermionField &out) \
-  {									\
-    typedef IMPL Impl;							\
-    typedef typename Simd::scalar_type S;				\
-    typedef typename Simd::vector_type V;				\
-									\
-    HAND_DECLARATIONS(ignore);						\
-									\
-    StencilEntry *SE;							\
-    int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist; \
-    int nmu=0;								\
-    HAND_DOP_SITE_DAG_EXT(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
-    nmu = 0;								\
-    HAND_DOP_SITE_DAG_EXT(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
-  }
-
-
-HAND_SPECIALISE_GPARITY(GparityWilsonImplF);
-HAND_SPECIALISE_GPARITY(GparityWilsonImplD);
-HAND_SPECIALISE_GPARITY(GparityWilsonImplFH);
-HAND_SPECIALISE_GPARITY(GparityWilsonImplDF);
-
-
-
-
-
-
-
-
-
-
-  
 ////////////// Wilson ; uses this implementation /////////////////////

 #define INSTANTIATE_THEM(A) \
@@ -930,8 +613,6 @@ INSTANTIATE_THEM(WilsonImplF);
 INSTANTIATE_THEM(WilsonImplD);
 INSTANTIATE_THEM(ZWilsonImplF);
 INSTANTIATE_THEM(ZWilsonImplD);
-INSTANTIATE_THEM(GparityWilsonImplF);
-INSTANTIATE_THEM(GparityWilsonImplD);
 INSTANTIATE_THEM(DomainWallVec5dImplF);
 INSTANTIATE_THEM(DomainWallVec5dImplD);
 INSTANTIATE_THEM(ZDomainWallVec5dImplF);
@@ -940,12 +621,11 @@ INSTANTIATE_THEM(WilsonImplFH);
 INSTANTIATE_THEM(WilsonImplDF);
 INSTANTIATE_THEM(ZWilsonImplFH);
 INSTANTIATE_THEM(ZWilsonImplDF);
-INSTANTIATE_THEM(GparityWilsonImplFH);
-INSTANTIATE_THEM(GparityWilsonImplDF);
 INSTANTIATE_THEM(DomainWallVec5dImplFH);
 INSTANTIATE_THEM(DomainWallVec5dImplDF);
 INSTANTIATE_THEM(ZDomainWallVec5dImplFH);
 INSTANTIATE_THEM(ZDomainWallVec5dImplDF);
 INSTANTIATE_THEM(WilsonTwoIndexAntiSymmetricImplF);
 INSTANTIATE_THEM(WilsonTwoIndexAntiSymmetricImplD);
+
 }}
--- a/lib/qcd/action/fermion/WilsonKernelsHandGparity.cc
+++ b/lib/qcd/action/fermion/WilsonKernelsHandGparity.cc
@@ -0,0 +1,878 @@
+    /*************************************************************************************
+
+    Grid physics library, www.github.com/paboyle/Grid 
+
+    Source file: ./lib/qcd/action/fermion/WilsonKernelsHand.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/qcd/action/fermion/FermionCore.h>
+
+#define REGISTER
+
+#define LOAD_CHIMU_BODY(F)			\
+  Chimu_00=ref(F)(0)(0);			\
+  Chimu_01=ref(F)(0)(1);			\
+  Chimu_02=ref(F)(0)(2);			\
+  Chimu_10=ref(F)(1)(0);			\
+  Chimu_11=ref(F)(1)(1);			\
+  Chimu_12=ref(F)(1)(2);			\
+  Chimu_20=ref(F)(2)(0);			\
+  Chimu_21=ref(F)(2)(1);			\
+  Chimu_22=ref(F)(2)(2);			\
+  Chimu_30=ref(F)(3)(0);			\
+  Chimu_31=ref(F)(3)(1);			\
+  Chimu_32=ref(F)(3)(2)
+
+#define LOAD_CHIMU(DIR,F,PERM)						\
+  { const SiteSpinor & ref (in._odata[offset]); LOAD_CHIMU_BODY(F); }
+
+#define LOAD_CHI_BODY(F)				\
+    Chi_00 = ref(F)(0)(0);\
+    Chi_01 = ref(F)(0)(1);\
+    Chi_02 = ref(F)(0)(2);\
+    Chi_10 = ref(F)(1)(0);\
+    Chi_11 = ref(F)(1)(1);\
+    Chi_12 = ref(F)(1)(2)
+
+#define LOAD_CHI(DIR,F,PERM)					\
+  {const SiteHalfSpinor &ref(buf[offset]); LOAD_CHI_BODY(F); }
+
+
+//G-parity implementations using in-place intrinsic ops
+
+//1l 1h -> 1h 1l
+//0l 0h , 1h 1l -> 0l 1h 0h,1l
+//0h,1l -> 1l,0h
+//if( (distance == 1 && !perm_will_occur) || (distance == -1 && perm_will_occur) )
+//Pulled fermion through forwards face, GPBC on upper component
+//Need 0= 0l 1h   1= 1l 0h
+//else if( (distance == -1 && !perm) || (distance == 1 && perm) )
+//Pulled fermion through backwards face, GPBC on lower component
+//Need 0= 1l 0h   1= 0l 1h
+
+//1l 1h -> 1h 1l
+//0l 0h , 1h 1l -> 0l 1h 0h,1l
+#define DO_TWIST_0L_1H(INTO,S,C,F, PERM, tmp1, tmp2, tmp3)			\
+  permute##PERM(tmp1, ref(1)(S)(C));				\
+  exchange##PERM(tmp2,tmp3, ref(0)(S)(C), tmp1);		\
+  INTO = tmp2;
+
+//0l 0h -> 0h 0l
+//1l 1h, 0h 0l -> 1l 0h, 1h 0l
+#define DO_TWIST_1L_0H(INTO,S,C,F, PERM, tmp1, tmp2, tmp3)			\
+  permute##PERM(tmp1, ref(0)(S)(C));				\
+  exchange##PERM(tmp2,tmp3, ref(1)(S)(C), tmp1);		\
+  INTO = tmp2;
+
+
+
+
+#define LOAD_CHI_SETUP(DIR,F)						\
+  g = F;								\
+  direction = st._directions[DIR];				\
+  distance = st._distances[DIR];				\
+  sl = st._grid->_simd_layout[direction];			\
+  inplace_twist = 0;						\
+  if(SE->_around_the_world && this->Params.twists[DIR % 4]){		\
+    if(sl == 1){							\
+      g = (F+1) % 2;							\
+    }else{								\
+      inplace_twist = 1;						\
+    }									\
+  }  
+
+#define LOAD_CHIMU_GPARITY_INPLACE_TWIST(DIR,F,PERM)			\
+  { const SiteSpinor &ref(in._odata[offset]);				\
+    LOAD_CHI_SETUP(DIR,F);						\
+    if(!inplace_twist){							\
+      LOAD_CHIMU_BODY(g);						\
+    }else{								\
+      if(  ( F==0 && ((distance == 1 && !perm) || (distance == -1 && perm)) ) || \
+	   ( F==1 && ((distance == -1 && !perm) || (distance == 1 && perm)) ) ){ \
+	DO_TWIST_0L_1H(Chimu_00,0,0,F,PERM,  U_00,U_01,U_10);		\
+	DO_TWIST_0L_1H(Chimu_01,0,1,F,PERM,  U_11,U_20,U_21);		\
+	DO_TWIST_0L_1H(Chimu_02,0,2,F,PERM,  U_00,U_01,U_10);		\
+	DO_TWIST_0L_1H(Chimu_10,1,0,F,PERM,  U_11,U_20,U_21);		\
+	DO_TWIST_0L_1H(Chimu_11,1,1,F,PERM,  U_00,U_01,U_10);		\
+	DO_TWIST_0L_1H(Chimu_12,1,2,F,PERM,  U_11,U_20,U_21);		\
+	DO_TWIST_0L_1H(Chimu_20,2,0,F,PERM,  U_00,U_01,U_10);		\
+	DO_TWIST_0L_1H(Chimu_21,2,1,F,PERM,  U_11,U_20,U_21);		\
+	DO_TWIST_0L_1H(Chimu_22,2,2,F,PERM,  U_00,U_01,U_10);		\
+	DO_TWIST_0L_1H(Chimu_30,3,0,F,PERM,  U_11,U_20,U_21);		\
+	DO_TWIST_0L_1H(Chimu_31,3,1,F,PERM,  U_00,U_01,U_10);		\
+	DO_TWIST_0L_1H(Chimu_32,3,2,F,PERM,  U_11,U_20,U_21);		\
+      }else{								\
+	DO_TWIST_1L_0H(Chimu_00,0,0,F,PERM,  U_00,U_01,U_10);		\
+	DO_TWIST_1L_0H(Chimu_01,0,1,F,PERM,  U_11,U_20,U_21);		\
+	DO_TWIST_1L_0H(Chimu_02,0,2,F,PERM,  U_00,U_01,U_10);		\
+	DO_TWIST_1L_0H(Chimu_10,1,0,F,PERM,  U_11,U_20,U_21);		\
+	DO_TWIST_1L_0H(Chimu_11,1,1,F,PERM,  U_00,U_01,U_10);		\
+	DO_TWIST_1L_0H(Chimu_12,1,2,F,PERM,  U_11,U_20,U_21);		\
+	DO_TWIST_1L_0H(Chimu_20,2,0,F,PERM,  U_00,U_01,U_10);		\
+	DO_TWIST_1L_0H(Chimu_21,2,1,F,PERM,  U_11,U_20,U_21);		\
+	DO_TWIST_1L_0H(Chimu_22,2,2,F,PERM,  U_00,U_01,U_10);		\
+	DO_TWIST_1L_0H(Chimu_30,3,0,F,PERM,  U_11,U_20,U_21);		\
+	DO_TWIST_1L_0H(Chimu_31,3,1,F,PERM,  U_00,U_01,U_10);		\
+	DO_TWIST_1L_0H(Chimu_32,3,2,F,PERM,  U_11,U_20,U_21);		\
+      } \
+    } \
+  }
+
+
+#define LOAD_CHI_GPARITY_INPLACE_TWIST(DIR,F,PERM)				\
+  { const SiteHalfSpinor &ref(buf[offset]);				\
+    LOAD_CHI_SETUP(DIR,F);						\
+    if(!inplace_twist){							\
+      LOAD_CHI_BODY(g);							\
+    }else{								\
+      if(  ( F==0 && ((distance == 1 && !perm) || (distance == -1 && perm)) ) || \
+	   ( F==1 && ((distance == -1 && !perm) || (distance == 1 && perm)) ) ){ \
+	DO_TWIST_0L_1H(Chi_00,0,0,F,PERM,  U_00,U_01,U_10);			\
+	DO_TWIST_0L_1H(Chi_01,0,1,F,PERM,  U_11,U_20,U_21);			\
+	DO_TWIST_0L_1H(Chi_02,0,2,F,PERM,  UChi_00,UChi_01,UChi_02);		\
+	DO_TWIST_0L_1H(Chi_10,1,0,F,PERM,  UChi_10,UChi_11,UChi_12);		\
+	DO_TWIST_0L_1H(Chi_11,1,1,F,PERM,  U_00,U_01,U_10);			\
+	DO_TWIST_0L_1H(Chi_12,1,2,F,PERM,  U_11,U_20,U_21);			\
+      }else{								\
+	DO_TWIST_1L_0H(Chi_00,0,0,F,PERM,  U_00,U_01,U_10);			\
+	DO_TWIST_1L_0H(Chi_01,0,1,F,PERM,  U_11,U_20,U_21);			\
+	DO_TWIST_1L_0H(Chi_02,0,2,F,PERM,  UChi_00,UChi_01,UChi_02);		\
+	DO_TWIST_1L_0H(Chi_10,1,0,F,PERM,  UChi_10,UChi_11,UChi_12);		\
+	DO_TWIST_1L_0H(Chi_11,1,1,F,PERM,  U_00,U_01,U_10);			\
+	DO_TWIST_1L_0H(Chi_12,1,2,F,PERM,  U_11,U_20,U_21);			\
+      }									\
+    }									\
+  }
+
+
+#define LOAD_CHI_GPARITY(DIR,F,PERM) LOAD_CHI_GPARITY_INPLACE_TWIST(DIR,F,PERM)
+#define LOAD_CHIMU_GPARITY(DIR,F,PERM) LOAD_CHIMU_GPARITY_INPLACE_TWIST(DIR,F,PERM)
+
+// To splat or not to splat depends on the implementation
+#define MULT_2SPIN_BODY \
+  Impl::loadLinkElement(U_00,ref()(0,0));	\
+  Impl::loadLinkElement(U_10,ref()(1,0));	\
+  Impl::loadLinkElement(U_20,ref()(2,0));	\
+  Impl::loadLinkElement(U_01,ref()(0,1));	\
+  Impl::loadLinkElement(U_11,ref()(1,1));	\
+  Impl::loadLinkElement(U_21,ref()(2,1));	\
+  UChi_00 = U_00*Chi_00;			\
+  UChi_10 = U_00*Chi_10;			\
+  UChi_01 = U_10*Chi_00;			\
+  UChi_11 = U_10*Chi_10;			\
+  UChi_02 = U_20*Chi_00;			\
+  UChi_12 = U_20*Chi_10;			\
+  UChi_00+= U_01*Chi_01;			\
+  UChi_10+= U_01*Chi_11;			\
+  UChi_01+= U_11*Chi_01;			\
+  UChi_11+= U_11*Chi_11;			\
+  UChi_02+= U_21*Chi_01;			\
+  UChi_12+= U_21*Chi_11;			\
+  Impl::loadLinkElement(U_00,ref()(0,2));	\
+  Impl::loadLinkElement(U_10,ref()(1,2));	\
+  Impl::loadLinkElement(U_20,ref()(2,2));	\
+  UChi_00+= U_00*Chi_02;			\
+  UChi_10+= U_00*Chi_12;			\
+  UChi_01+= U_10*Chi_02;			\
+  UChi_11+= U_10*Chi_12;			\
+  UChi_02+= U_20*Chi_02;			\
+  UChi_12+= U_20*Chi_12
+
+
+#define MULT_2SPIN(A,F)					\
+  {auto & ref(U._odata[sU](A)); MULT_2SPIN_BODY; }
+
+#define MULT_2SPIN_GPARITY(A,F)				\
+  {auto & ref(U._odata[sU](F)(A)); MULT_2SPIN_BODY; }
+
+
+#define PERMUTE_DIR(dir)			\
+      permute##dir(Chi_00,Chi_00);\
+      permute##dir(Chi_01,Chi_01);\
+      permute##dir(Chi_02,Chi_02);\
+      permute##dir(Chi_10,Chi_10);\
+      permute##dir(Chi_11,Chi_11);\
+      permute##dir(Chi_12,Chi_12);
+
+//      hspin(0)=fspin(0)+timesI(fspin(3));
+//      hspin(1)=fspin(1)+timesI(fspin(2));
+#define XP_PROJ \
+    Chi_00 = Chimu_00+timesI(Chimu_30);\
+    Chi_01 = Chimu_01+timesI(Chimu_31);\
+    Chi_02 = Chimu_02+timesI(Chimu_32);\
+    Chi_10 = Chimu_10+timesI(Chimu_20);\
+    Chi_11 = Chimu_11+timesI(Chimu_21);\
+    Chi_12 = Chimu_12+timesI(Chimu_22);
+
+#define YP_PROJ \
+    Chi_00 = Chimu_00-Chimu_30;\
+    Chi_01 = Chimu_01-Chimu_31;\
+    Chi_02 = Chimu_02-Chimu_32;\
+    Chi_10 = Chimu_10+Chimu_20;\
+    Chi_11 = Chimu_11+Chimu_21;\
+    Chi_12 = Chimu_12+Chimu_22;
+
+#define ZP_PROJ \
+  Chi_00 = Chimu_00+timesI(Chimu_20);		\
+  Chi_01 = Chimu_01+timesI(Chimu_21);		\
+  Chi_02 = Chimu_02+timesI(Chimu_22);		\
+  Chi_10 = Chimu_10-timesI(Chimu_30);		\
+  Chi_11 = Chimu_11-timesI(Chimu_31);		\
+  Chi_12 = Chimu_12-timesI(Chimu_32);
+
+#define TP_PROJ \
+  Chi_00 = Chimu_00+Chimu_20;		\
+  Chi_01 = Chimu_01+Chimu_21;		\
+  Chi_02 = Chimu_02+Chimu_22;		\
+  Chi_10 = Chimu_10+Chimu_30;		\
+  Chi_11 = Chimu_11+Chimu_31;		\
+  Chi_12 = Chimu_12+Chimu_32;
+
+
+//      hspin(0)=fspin(0)-timesI(fspin(3));
+//      hspin(1)=fspin(1)-timesI(fspin(2));
+#define XM_PROJ \
+    Chi_00 = Chimu_00-timesI(Chimu_30);\
+    Chi_01 = Chimu_01-timesI(Chimu_31);\
+    Chi_02 = Chimu_02-timesI(Chimu_32);\
+    Chi_10 = Chimu_10-timesI(Chimu_20);\
+    Chi_11 = Chimu_11-timesI(Chimu_21);\
+    Chi_12 = Chimu_12-timesI(Chimu_22);
+
+#define YM_PROJ \
+    Chi_00 = Chimu_00+Chimu_30;\
+    Chi_01 = Chimu_01+Chimu_31;\
+    Chi_02 = Chimu_02+Chimu_32;\
+    Chi_10 = Chimu_10-Chimu_20;\
+    Chi_11 = Chimu_11-Chimu_21;\
+    Chi_12 = Chimu_12-Chimu_22;
+
+#define ZM_PROJ \
+  Chi_00 = Chimu_00-timesI(Chimu_20);		\
+  Chi_01 = Chimu_01-timesI(Chimu_21);		\
+  Chi_02 = Chimu_02-timesI(Chimu_22);		\
+  Chi_10 = Chimu_10+timesI(Chimu_30);		\
+  Chi_11 = Chimu_11+timesI(Chimu_31);		\
+  Chi_12 = Chimu_12+timesI(Chimu_32);
+
+#define TM_PROJ \
+  Chi_00 = Chimu_00-Chimu_20;		\
+  Chi_01 = Chimu_01-Chimu_21;		\
+  Chi_02 = Chimu_02-Chimu_22;		\
+  Chi_10 = Chimu_10-Chimu_30;		\
+  Chi_11 = Chimu_11-Chimu_31;		\
+  Chi_12 = Chimu_12-Chimu_32;
+
+//      fspin(0)=hspin(0);
+//      fspin(1)=hspin(1);
+//      fspin(2)=timesMinusI(hspin(1));
+//      fspin(3)=timesMinusI(hspin(0));
+#define XP_RECON\
+  result_00 = UChi_00;\
+  result_01 = UChi_01;\
+  result_02 = UChi_02;\
+  result_10 = UChi_10;\
+  result_11 = UChi_11;\
+  result_12 = UChi_12;\
+  result_20 = timesMinusI(UChi_10);\
+  result_21 = timesMinusI(UChi_11);\
+  result_22 = timesMinusI(UChi_12);\
+  result_30 = timesMinusI(UChi_00);\
+  result_31 = timesMinusI(UChi_01);\
+  result_32 = timesMinusI(UChi_02);
+
+#define XP_RECON_ACCUM\
+  result_00+=UChi_00;\
+  result_01+=UChi_01;\
+  result_02+=UChi_02;\
+  result_10+=UChi_10;\
+  result_11+=UChi_11;\
+  result_12+=UChi_12;\
+  result_20-=timesI(UChi_10);\
+  result_21-=timesI(UChi_11);\
+  result_22-=timesI(UChi_12);\
+  result_30-=timesI(UChi_00);\
+  result_31-=timesI(UChi_01);\
+  result_32-=timesI(UChi_02);
+
+#define XM_RECON\
+  result_00 = UChi_00;\
+  result_01 = UChi_01;\
+  result_02 = UChi_02;\
+  result_10 = UChi_10;\
+  result_11 = UChi_11;\
+  result_12 = UChi_12;\
+  result_20 = timesI(UChi_10);\
+  result_21 = timesI(UChi_11);\
+  result_22 = timesI(UChi_12);\
+  result_30 = timesI(UChi_00);\
+  result_31 = timesI(UChi_01);\
+  result_32 = timesI(UChi_02);
+
+#define XM_RECON_ACCUM\
+  result_00+= UChi_00;\
+  result_01+= UChi_01;\
+  result_02+= UChi_02;\
+  result_10+= UChi_10;\
+  result_11+= UChi_11;\
+  result_12+= UChi_12;\
+  result_20+= timesI(UChi_10);\
+  result_21+= timesI(UChi_11);\
+  result_22+= timesI(UChi_12);\
+  result_30+= timesI(UChi_00);\
+  result_31+= timesI(UChi_01);\
+  result_32+= timesI(UChi_02);
+
+#define YP_RECON_ACCUM\
+  result_00+= UChi_00;\
+  result_01+= UChi_01;\
+  result_02+= UChi_02;\
+  result_10+= UChi_10;\
+  result_11+= UChi_11;\
+  result_12+= UChi_12;\
+  result_20+= UChi_10;\
+  result_21+= UChi_11;\
+  result_22+= UChi_12;\
+  result_30-= UChi_00;\
+  result_31-= UChi_01;\
+  result_32-= UChi_02;
+
+#define YM_RECON_ACCUM\
+  result_00+= UChi_00;\
+  result_01+= UChi_01;\
+  result_02+= UChi_02;\
+  result_10+= UChi_10;\
+  result_11+= UChi_11;\
+  result_12+= UChi_12;\
+  result_20-= UChi_10;\
+  result_21-= UChi_11;\
+  result_22-= UChi_12;\
+  result_30+= UChi_00;\
+  result_31+= UChi_01;\
+  result_32+= UChi_02;
+
+#define ZP_RECON_ACCUM\
+  result_00+= UChi_00;\
+  result_01+= UChi_01;\
+  result_02+= UChi_02;\
+  result_10+= UChi_10;\
+  result_11+= UChi_11;\
+  result_12+= UChi_12;\
+  result_20-= timesI(UChi_00);			\
+  result_21-= timesI(UChi_01);			\
+  result_22-= timesI(UChi_02);			\
+  result_30+= timesI(UChi_10);			\
+  result_31+= timesI(UChi_11);			\
+  result_32+= timesI(UChi_12);
+
+#define ZM_RECON_ACCUM\
+  result_00+= UChi_00;\
+  result_01+= UChi_01;\
+  result_02+= UChi_02;\
+  result_10+= UChi_10;\
+  result_11+= UChi_11;\
+  result_12+= UChi_12;\
+  result_20+= timesI(UChi_00);			\
+  result_21+= timesI(UChi_01);			\
+  result_22+= timesI(UChi_02);			\
+  result_30-= timesI(UChi_10);			\
+  result_31-= timesI(UChi_11);			\
+  result_32-= timesI(UChi_12);
+
+#define TP_RECON_ACCUM\
+  result_00+= UChi_00;\
+  result_01+= UChi_01;\
+  result_02+= UChi_02;\
+  result_10+= UChi_10;\
+  result_11+= UChi_11;\
+  result_12+= UChi_12;\
+  result_20+= UChi_00;			\
+  result_21+= UChi_01;			\
+  result_22+= UChi_02;			\
+  result_30+= UChi_10;			\
+  result_31+= UChi_11;			\
+  result_32+= UChi_12;
+
+#define TM_RECON_ACCUM\
+  result_00+= UChi_00;\
+  result_01+= UChi_01;\
+  result_02+= UChi_02;\
+  result_10+= UChi_10;\
+  result_11+= UChi_11;\
+  result_12+= UChi_12;\
+  result_20-= UChi_00;	\
+  result_21-= UChi_01;	\
+  result_22-= UChi_02;	\
+  result_30-= UChi_10;	\
+  result_31-= UChi_11;	\
+  result_32-= UChi_12;
+
+#define HAND_STENCIL_LEG(PROJ,PERM,DIR,RECON,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \
+  SE=st.GetEntry(ptype,DIR,ss);			\
+  offset = SE->_offset;				\
+  local  = SE->_is_local;			\
+  perm   = SE->_permute;			\
+  if ( local ) {				\
+    LOAD_CHIMU_IMPL(DIR,F,PERM);			\
+    PROJ;					\
+    if ( perm) {				\
+      PERMUTE_DIR(PERM);			\
+    }						\
+  } else {					\
+    LOAD_CHI_IMPL(DIR,F,PERM);			\
+  }						\
+  MULT_2SPIN_IMPL(DIR,F);			\
+  RECON;					
+
+
+#define HAND_STENCIL_LEG_INT(PROJ,PERM,DIR,RECON,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL)	\
+  SE=st.GetEntry(ptype,DIR,ss);			\
+  offset = SE->_offset;				\
+  local  = SE->_is_local;			\
+  perm   = SE->_permute;			\
+  if ( local ) {				\
+    LOAD_CHIMU_IMPL(DIR,F,PERM);			\
+    PROJ;					\
+    if ( perm) {				\
+      PERMUTE_DIR(PERM);			\
+    }						\
+  } else if ( st.same_node[DIR] ) {		\
+    LOAD_CHI_IMPL(DIR,F,PERM);			\
+  }						\
+  if (local || st.same_node[DIR] ) {		\
+    MULT_2SPIN_IMPL(DIR,F);			\
+    RECON;					\
+  }
+
+#define HAND_STENCIL_LEG_EXT(PROJ,PERM,DIR,RECON,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL)	\
+  SE=st.GetEntry(ptype,DIR,ss);			\
+  offset = SE->_offset;				\
+  local  = SE->_is_local;			\
+  perm   = SE->_permute;			\
+  if((!SE->_is_local)&&(!st.same_node[DIR]) ) {	\
+    LOAD_CHI_IMPL(DIR,F,PERM);			\
+    MULT_2SPIN_IMPL(DIR,F);			\
+    RECON;					\
+    nmu++;					\
+  }
+
+#define HAND_RESULT(ss,F)			\
+  {						\
+    SiteSpinor & ref (out._odata[ss]);		\
+    vstream(ref(F)(0)(0),result_00);		\
+    vstream(ref(F)(0)(1),result_01);		\
+    vstream(ref(F)(0)(2),result_02);		\
+    vstream(ref(F)(1)(0),result_10);		\
+    vstream(ref(F)(1)(1),result_11);		\
+    vstream(ref(F)(1)(2),result_12);		\
+    vstream(ref(F)(2)(0),result_20);		\
+    vstream(ref(F)(2)(1),result_21);		\
+    vstream(ref(F)(2)(2),result_22);		\
+    vstream(ref(F)(3)(0),result_30);		\
+    vstream(ref(F)(3)(1),result_31);		\
+    vstream(ref(F)(3)(2),result_32);		\
+  }
+
+#define HAND_RESULT_EXT(ss,F)			\
+  if (nmu){					\
+    SiteSpinor & ref (out._odata[ss]);		\
+    ref(F)(0)(0)+=result_00;		\
+    ref(F)(0)(1)+=result_01;		\
+    ref(F)(0)(2)+=result_02;		\
+    ref(F)(1)(0)+=result_10;		\
+    ref(F)(1)(1)+=result_11;		\
+    ref(F)(1)(2)+=result_12;		\
+    ref(F)(2)(0)+=result_20;		\
+    ref(F)(2)(1)+=result_21;		\
+    ref(F)(2)(2)+=result_22;		\
+    ref(F)(3)(0)+=result_30;		\
+    ref(F)(3)(1)+=result_31;		\
+    ref(F)(3)(2)+=result_32;		\
+  }
+
+
+#define HAND_DECLARATIONS(a)			\
+  Simd result_00;				\
+  Simd result_01;				\
+  Simd result_02;				\
+  Simd result_10;				\
+  Simd result_11;				\
+  Simd result_12;				\
+  Simd result_20;				\
+  Simd result_21;				\
+  Simd result_22;				\
+  Simd result_30;				\
+  Simd result_31;				\
+  Simd result_32;				\
+  Simd Chi_00;					\
+  Simd Chi_01;					\
+  Simd Chi_02;					\
+  Simd Chi_10;					\
+  Simd Chi_11;					\
+  Simd Chi_12;					\
+  Simd UChi_00;					\
+  Simd UChi_01;					\
+  Simd UChi_02;					\
+  Simd UChi_10;					\
+  Simd UChi_11;					\
+  Simd UChi_12;					\
+  Simd U_00;					\
+  Simd U_10;					\
+  Simd U_20;					\
+  Simd U_01;					\
+  Simd U_11;					\
+  Simd U_21;
+
+#define ZERO_RESULT				\
+  result_00=zero;				\
+  result_01=zero;				\
+  result_02=zero;				\
+  result_10=zero;				\
+  result_11=zero;				\
+  result_12=zero;				\
+  result_20=zero;				\
+  result_21=zero;				\
+  result_22=zero;				\
+  result_30=zero;				\
+  result_31=zero;				\
+  result_32=zero;			
+
+#define Chimu_00 Chi_00
+#define Chimu_01 Chi_01
+#define Chimu_02 Chi_02
+#define Chimu_10 Chi_10
+#define Chimu_11 Chi_11
+#define Chimu_12 Chi_12
+#define Chimu_20 UChi_00
+#define Chimu_21 UChi_01
+#define Chimu_22 UChi_02
+#define Chimu_30 UChi_10
+#define Chimu_31 UChi_11
+#define Chimu_32 UChi_12
+
+namespace Grid {
+namespace QCD {
+
+template<class Impl> void 
+WilsonKernels<Impl>::HandDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor  *buf,
+					  int ss,int sU,const FermionField &in, FermionField &out)
+{
+// T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
+  typedef typename Simd::scalar_type S;
+  typedef typename Simd::vector_type V;
+
+  HAND_DECLARATIONS(ignore);
+
+  int offset,local,perm, ptype;
+  StencilEntry *SE;
+
+#define HAND_DOP_SITE(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \
+  HAND_STENCIL_LEG(XM_PROJ,3,Xp,XM_RECON,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG(YM_PROJ,2,Yp,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL);	\
+  HAND_STENCIL_LEG(ZM_PROJ,1,Zp,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG(TM_PROJ,0,Tp,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG(XP_PROJ,3,Xm,XP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG(YP_PROJ,2,Ym,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG(ZP_PROJ,1,Zm,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG(TP_PROJ,0,Tm,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_RESULT(ss,F)
+
+  HAND_DOP_SITE(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN);
+}
+
+template<class Impl>
+void WilsonKernels<Impl>::HandDhopSiteDag(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
+						  int ss,int sU,const FermionField &in, FermionField &out)
+{
+  typedef typename Simd::scalar_type S;
+  typedef typename Simd::vector_type V;
+
+  HAND_DECLARATIONS(ignore);
+
+  StencilEntry *SE;
+  int offset,local,perm, ptype;
+
+#define HAND_DOP_SITE_DAG(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \
+  HAND_STENCIL_LEG(XP_PROJ,3,Xp,XP_RECON,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG(YP_PROJ,2,Yp,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG(ZP_PROJ,1,Zp,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG(TP_PROJ,0,Tp,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG(XM_PROJ,3,Xm,XM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG(YM_PROJ,2,Ym,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG(ZM_PROJ,1,Zm,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG(TM_PROJ,0,Tm,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_RESULT(ss,F)
+
+  HAND_DOP_SITE_DAG(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN);
+}
+
+template<class Impl> void 
+WilsonKernels<Impl>::HandDhopSiteInt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor  *buf,
+					  int ss,int sU,const FermionField &in, FermionField &out)
+{
+// T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
+  typedef typename Simd::scalar_type S;
+  typedef typename Simd::vector_type V;
+
+  HAND_DECLARATIONS(ignore);
+
+  int offset,local,perm, ptype;
+  StencilEntry *SE;
+
+#define HAND_DOP_SITE_INT(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \
+  ZERO_RESULT; \
+  HAND_STENCIL_LEG_INT(XM_PROJ,3,Xp,XM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG_INT(YM_PROJ,2,Yp,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG_INT(ZM_PROJ,1,Zp,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG_INT(TM_PROJ,0,Tp,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG_INT(XP_PROJ,3,Xm,XP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG_INT(YP_PROJ,2,Ym,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG_INT(ZP_PROJ,1,Zm,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG_INT(TP_PROJ,0,Tm,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_RESULT(ss,F)
+
+  HAND_DOP_SITE_INT(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN);
+}
+
+template<class Impl>
+void WilsonKernels<Impl>::HandDhopSiteDagInt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
+						  int ss,int sU,const FermionField &in, FermionField &out)
+{
+  typedef typename Simd::scalar_type S;
+  typedef typename Simd::vector_type V;
+
+  HAND_DECLARATIONS(ignore);
+
+  StencilEntry *SE;
+  int offset,local,perm, ptype;
+
+#define HAND_DOP_SITE_DAG_INT(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL)				\
+  ZERO_RESULT;							\
+  HAND_STENCIL_LEG_INT(XP_PROJ,3,Xp,XP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL);		\
+  HAND_STENCIL_LEG_INT(YP_PROJ,2,Yp,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL);		\
+  HAND_STENCIL_LEG_INT(ZP_PROJ,1,Zp,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL);		\
+  HAND_STENCIL_LEG_INT(TP_PROJ,0,Tp,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL);		\
+  HAND_STENCIL_LEG_INT(XM_PROJ,3,Xm,XM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL);		\
+  HAND_STENCIL_LEG_INT(YM_PROJ,2,Ym,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL);		\
+  HAND_STENCIL_LEG_INT(ZM_PROJ,1,Zm,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL);		\
+  HAND_STENCIL_LEG_INT(TM_PROJ,0,Tm,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL);		\
+  HAND_RESULT(ss,F)
+  
+  HAND_DOP_SITE_DAG_INT(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN);
+}
+
+template<class Impl> void 
+WilsonKernels<Impl>::HandDhopSiteExt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor  *buf,
+					  int ss,int sU,const FermionField &in, FermionField &out)
+{
+// T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
+  typedef typename Simd::scalar_type S;
+  typedef typename Simd::vector_type V;
+
+  HAND_DECLARATIONS(ignore);
+
+  int offset,local,perm, ptype;
+  StencilEntry *SE;
+  int nmu=0;
+
+#define HAND_DOP_SITE_EXT(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \
+  ZERO_RESULT; \
+  HAND_STENCIL_LEG_EXT(XM_PROJ,3,Xp,XM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG_EXT(YM_PROJ,2,Yp,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG_EXT(ZM_PROJ,1,Zp,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG_EXT(TM_PROJ,0,Tp,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG_EXT(XP_PROJ,3,Xm,XP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG_EXT(YP_PROJ,2,Ym,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG_EXT(ZP_PROJ,1,Zm,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG_EXT(TP_PROJ,0,Tm,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_RESULT_EXT(ss,F)
+
+  HAND_DOP_SITE_EXT(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN);
+}
+
+template<class Impl>
+void WilsonKernels<Impl>::HandDhopSiteDagExt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
+						  int ss,int sU,const FermionField &in, FermionField &out)
+{
+  typedef typename Simd::scalar_type S;
+  typedef typename Simd::vector_type V;
+
+  HAND_DECLARATIONS(ignore);
+
+  StencilEntry *SE;
+  int offset,local,perm, ptype;
+  int nmu=0;
+
+#define HAND_DOP_SITE_DAG_EXT(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \
+  ZERO_RESULT; \
+  HAND_STENCIL_LEG_EXT(XP_PROJ,3,Xp,XP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG_EXT(YP_PROJ,2,Yp,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG_EXT(ZP_PROJ,1,Zp,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG_EXT(TP_PROJ,0,Tp,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG_EXT(XM_PROJ,3,Xm,XM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG_EXT(YM_PROJ,2,Ym,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG_EXT(ZM_PROJ,1,Zm,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_STENCIL_LEG_EXT(TM_PROJ,0,Tm,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
+  HAND_RESULT_EXT(ss,F)
+
+  HAND_DOP_SITE_DAG_EXT(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN);
+}
+
+#define HAND_SPECIALISE_GPARITY(IMPL)					\
+  template<> void							\
+  WilsonKernels<IMPL>::HandDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor  *buf, \
+				    int ss,int sU,const FermionField &in, FermionField &out) \
+  {									\
+    typedef IMPL Impl;							\
+    typedef typename Simd::scalar_type S;				\
+    typedef typename Simd::vector_type V;				\
+									\
+    HAND_DECLARATIONS(ignore);						\
+									\
+    int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist; \
+    StencilEntry *SE;							\
+    HAND_DOP_SITE(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
+    HAND_DOP_SITE(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
+  }									\
+									\
+  template<>								\
+  void WilsonKernels<IMPL>::HandDhopSiteDag(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
+					    int ss,int sU,const FermionField &in, FermionField &out) \
+  {									\
+    typedef IMPL Impl;							\
+    typedef typename Simd::scalar_type S;				\
+    typedef typename Simd::vector_type V;				\
+									\
+    HAND_DECLARATIONS(ignore);						\
+									\
+    StencilEntry *SE;							\
+    int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist;					\
+    HAND_DOP_SITE_DAG(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
+    HAND_DOP_SITE_DAG(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
+  }									\
+									\
+  template<> void							\
+  WilsonKernels<IMPL>::HandDhopSiteInt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor  *buf, \
+						     int ss,int sU,const FermionField &in, FermionField &out) \
+  {									\
+    typedef IMPL Impl;							\
+    typedef typename Simd::scalar_type S;				\
+    typedef typename Simd::vector_type V;				\
+									\
+    HAND_DECLARATIONS(ignore);						\
+									\
+    int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist;					\
+    StencilEntry *SE;							\
+    HAND_DOP_SITE_INT(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
+    HAND_DOP_SITE_INT(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
+  }									\
+									\
+  template<>								\
+  void WilsonKernels<IMPL>::HandDhopSiteDagInt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
+							     int ss,int sU,const FermionField &in, FermionField &out) \
+  {									\
+    typedef IMPL Impl;							\
+    typedef typename Simd::scalar_type S;				\
+    typedef typename Simd::vector_type V;				\
+									\
+    HAND_DECLARATIONS(ignore);						\
+									\
+    StencilEntry *SE;							\
+    int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist; \
+    HAND_DOP_SITE_DAG_INT(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
+    HAND_DOP_SITE_DAG_INT(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
+  }									\
+									\
+  template<> void							\
+  WilsonKernels<IMPL>::HandDhopSiteExt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor  *buf, \
+						     int ss,int sU,const FermionField &in, FermionField &out) \
+  {									\
+    typedef IMPL Impl;							\
+    typedef typename Simd::scalar_type S;				\
+    typedef typename Simd::vector_type V;				\
+									\
+    HAND_DECLARATIONS(ignore);						\
+									\
+    int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist; \
+    StencilEntry *SE;							\
+    int nmu=0;								\
+    HAND_DOP_SITE_EXT(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
+    nmu = 0;								\
+    HAND_DOP_SITE_EXT(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
+  }									\
+  template<>								\
+  void WilsonKernels<IMPL>::HandDhopSiteDagExt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
+							     int ss,int sU,const FermionField &in, FermionField &out) \
+  {									\
+    typedef IMPL Impl;							\
+    typedef typename Simd::scalar_type S;				\
+    typedef typename Simd::vector_type V;				\
+									\
+    HAND_DECLARATIONS(ignore);						\
+									\
+    StencilEntry *SE;							\
+    int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist; \
+    int nmu=0;								\
+    HAND_DOP_SITE_DAG_EXT(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
+    nmu = 0;								\
+    HAND_DOP_SITE_DAG_EXT(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
+  }
+
+
+HAND_SPECIALISE_GPARITY(GparityWilsonImplF);
+HAND_SPECIALISE_GPARITY(GparityWilsonImplD);
+HAND_SPECIALISE_GPARITY(GparityWilsonImplFH);
+HAND_SPECIALISE_GPARITY(GparityWilsonImplDF);
+
+
+
+
+
+
+
+
+
+
+  
+////////////// Wilson ; uses this implementation /////////////////////
+
+#define INSTANTIATE_THEM(A) \
+template void WilsonKernels<A>::HandDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,\
+					     int ss,int sU,const FermionField &in, FermionField &out); \
+template void WilsonKernels<A>::HandDhopSiteDag(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
+						int ss,int sU,const FermionField &in, FermionField &out);\
+template void WilsonKernels<A>::HandDhopSiteInt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,\
+						int ss,int sU,const FermionField &in, FermionField &out); \
+template void WilsonKernels<A>::HandDhopSiteDagInt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
+						   int ss,int sU,const FermionField &in, FermionField &out); \
+template void WilsonKernels<A>::HandDhopSiteExt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,\
+						int ss,int sU,const FermionField &in, FermionField &out); \
+template void WilsonKernels<A>::HandDhopSiteDagExt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
+						   int ss,int sU,const FermionField &in, FermionField &out); 
+
+INSTANTIATE_THEM(GparityWilsonImplF);
+INSTANTIATE_THEM(GparityWilsonImplD);
+INSTANTIATE_THEM(GparityWilsonImplFH);
+INSTANTIATE_THEM(GparityWilsonImplDF);
+}}
--- a/lib/qcd/hmc/HMCResourceManager.h
+++ b/lib/qcd/hmc/HMCResourceManager.h
@@ -48,6 +48,22 @@ with this program; if not, write to the Free Software Foundation, Inc.,
    }                                                                    \
  }

+#define RegisterLoadCheckPointerMetadataFunction(NAME)                   \
+  template < class Metadata >                                            \
+  void Load##NAME##Checkpointer(const CheckpointerParameters& Params_, const Metadata& M_) { \
+    if (!have_CheckPointer) {                                            \
+      std::cout << GridLogDebug << "Loading Metadata Checkpointer " << #NAME      \
+                << std::endl;                                            \
+      CP = std::unique_ptr<CheckpointerBaseModule>(                      \
+        new NAME##CPModule<ImplementationPolicy, Metadata >(Params_, M_));   \
+      have_CheckPointer = true;                                          \
+    } else {                                                             \
+      std::cout << GridLogError << "Checkpointer already loaded "        \
+                << std::endl;                                            \
+      exit(1);                                                           \
+    }                                                                    \
+  }
+
 namespace Grid {
 namespace QCD {

@@ -77,7 +93,7 @@ class HMCResourceManager {
  bool have_CheckPointer;

  // NOTE: operator << is not overloaded for std::vector<string> 
-  // so thsi function is necessary
+  // so this function is necessary
  void output_vector_string(const std::vector<std::string> &vs){
    for (auto &i: vs)
      std::cout << i << " ";
@@ -254,6 +270,7 @@ class HMCResourceManager {
  RegisterLoadCheckPointerFunction(Nersc);
  #ifdef HAVE_LIME
  RegisterLoadCheckPointerFunction(ILDG);
+  RegisterLoadCheckPointerMetadataFunction(Scidac);
  #endif

  ////////////////////////////////////////////////////////
--- a/lib/qcd/hmc/checkpointers/BaseCheckpointer.h
+++ b/lib/qcd/hmc/checkpointers/BaseCheckpointer.h
@@ -76,6 +76,14 @@ class BaseHmcCheckpointer : public HmcObservable<typename Impl::Field> {
    }
 	} 

+  void check_filename(const std::string &filename){
+    std::ifstream f(filename.c_str());
+    if(!f.good()){
+      std::cout << GridLogError << "Filename " << filename << " not found. Aborting. " << std::endl;
+      abort();
+    };
+  }
+
  virtual void initialize(const CheckpointerParameters &Params) = 0;

  virtual void CheckpointRestore(int traj, typename Impl::Field &U,
--- a/lib/qcd/hmc/checkpointers/BinaryCheckpointer.h
+++ b/lib/qcd/hmc/checkpointers/BinaryCheckpointer.h
@@ -93,6 +93,9 @@ class BinaryHmcCheckpointer : public BaseHmcCheckpointer<Impl> {
  void CheckpointRestore(int traj, Field &U, GridSerialRNG &sRNG, GridParallelRNG &pRNG) {
    std::string config, rng;
    this->build_filenames(traj, Params, config, rng);
+    this->check_filename(rng);
+    this->check_filename(config);
+

    BinarySimpleMunger<sobj_double, sobj> munge;

--- a/lib/qcd/hmc/checkpointers/CheckPointerModules.h
+++ b/lib/qcd/hmc/checkpointers/CheckPointerModules.h
@@ -136,6 +136,20 @@ class ILDGCPModule: public CheckPointerModule< ImplementationPolicy> {

 };

+template<class ImplementationPolicy, class Metadata>
+class ScidacCPModule: public CheckPointerModule< ImplementationPolicy> {
+  typedef CheckPointerModule< ImplementationPolicy> CPBase;
+  Metadata M;
+
+  // acquire resource
+  virtual void initialize(){
+     this->CheckPointPtr.reset(new ScidacHmcCheckpointer<ImplementationPolicy, Metadata>(this->Par_, M));
+  }
+public:
+  ScidacCPModule(typename CPBase::APar Par, Metadata M_):M(M_), CPBase(Par) {}
+  template <class ReaderClass>
+  ScidacCPModule(Reader<ReaderClass>& Reader) : Parametrized<typename CPBase::APar>(Reader), M(Reader){};
+};
 #endif


--- a/lib/qcd/hmc/checkpointers/CheckPointers.h
+++ b/lib/qcd/hmc/checkpointers/CheckPointers.h
@@ -34,6 +34,7 @@ directory
 #include <Grid/qcd/hmc/checkpointers/NerscCheckpointer.h>
 #include <Grid/qcd/hmc/checkpointers/BinaryCheckpointer.h>
 #include <Grid/qcd/hmc/checkpointers/ILDGCheckpointer.h>
+#include <Grid/qcd/hmc/checkpointers/ScidacCheckpointer.h>
 //#include <Grid/qcd/hmc/checkpointers/CheckPointerModules.h>


--- a/lib/qcd/hmc/checkpointers/ILDGCheckpointer.h
+++ b/lib/qcd/hmc/checkpointers/ILDGCheckpointer.h
@@ -74,10 +74,10 @@ class ILDGHmcCheckpointer : public BaseHmcCheckpointer<Implementation> {
    if ((traj % Params.saveInterval) == 0) {
      std::string config, rng;
      this->build_filenames(traj, Params, config, rng);
-      
+      GridBase *grid = U._grid;
      uint32_t nersc_csum,scidac_csuma,scidac_csumb;
      BinaryIO::writeRNG(sRNG, pRNG, rng, 0,nersc_csum,scidac_csuma,scidac_csumb);
-      IldgWriter _IldgWriter;
+      IldgWriter _IldgWriter(grid->IsBoss());
      _IldgWriter.open(config);
      _IldgWriter.writeConfiguration(U, traj, config, config);
      _IldgWriter.close();
@@ -95,6 +95,10 @@ class ILDGHmcCheckpointer : public BaseHmcCheckpointer<Implementation> {
                         GridParallelRNG &pRNG) {
    std::string config, rng;
    this->build_filenames(traj, Params, config, rng);
+    this->check_filename(rng);
+    this->check_filename(config);
+
+    

    uint32_t nersc_csum,scidac_csuma,scidac_csumb;
    BinaryIO::readRNG(sRNG, pRNG, rng, 0,nersc_csum,scidac_csuma,scidac_csumb);
--- a/lib/qcd/hmc/checkpointers/NerscCheckpointer.h
+++ b/lib/qcd/hmc/checkpointers/NerscCheckpointer.h
@@ -69,6 +69,9 @@ class NerscHmcCheckpointer : public BaseHmcCheckpointer<Gimpl> {
                         GridParallelRNG &pRNG) {
    std::string config, rng;
    this->build_filenames(traj, Params, config, rng);
+    this->check_filename(rng);
+    this->check_filename(config);
+

    FieldMetaData header;
    NerscIO::readRNGState(sRNG, pRNG, header, rng);
--- a/lib/qcd/hmc/checkpointers/ScidacCheckpointer.h
+++ b/lib/qcd/hmc/checkpointers/ScidacCheckpointer.h
@@ -0,0 +1,125 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid
+
+Source file: ./lib/qcd/hmc/ScidacCheckpointer.h
+
+Copyright (C) 2018
+
+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 SCIDAC_CHECKPOINTER
+#define SCIDAC_CHECKPOINTER
+
+#ifdef HAVE_LIME
+
+#include <iostream>
+#include <sstream>
+#include <string>
+
+namespace Grid {
+namespace QCD {
+
+// For generic fields
+template <class Implementation, class Metadata>
+class ScidacHmcCheckpointer : public BaseHmcCheckpointer<Implementation> {
+ private:
+  CheckpointerParameters Params;
+  Metadata MData;
+
+  typedef typename Implementation::Field Field;
+
+ public:
+  //INHERIT_GIMPL_TYPES(Implementation);
+
+  ScidacHmcCheckpointer(const CheckpointerParameters &Params_) { initialize(Params_); }
+  ScidacHmcCheckpointer(const CheckpointerParameters &Params_, const Metadata& M_):MData(M_) { initialize(Params_); }
+
+  void initialize(const CheckpointerParameters &Params_) {
+    Params = Params_;
+
+    // check here that the format is valid
+    int ieee32big = (Params.format == std::string("IEEE32BIG"));
+    int ieee32    = (Params.format == std::string("IEEE32"));
+    int ieee64big = (Params.format == std::string("IEEE64BIG"));
+    int ieee64    = (Params.format == std::string("IEEE64"));
+
+    if (!(ieee64big || ieee32 || ieee32big || ieee64)) {
+      std::cout << GridLogError << "Unrecognized file format " << Params.format
+                << std::endl;
+      std::cout << GridLogError
+                << "Allowed: IEEE32BIG | IEEE32 | IEEE64BIG | IEEE64"
+                << std::endl;
+
+      exit(1);
+    }
+  }
+
+  void TrajectoryComplete(int traj, Field &U, GridSerialRNG &sRNG,
+                          GridParallelRNG &pRNG) {
+    if ((traj % Params.saveInterval) == 0) {
+      std::string config, rng;
+      this->build_filenames(traj, Params, config, rng);
+      GridBase *grid = U._grid;
+      uint32_t nersc_csum,scidac_csuma,scidac_csumb;
+      BinaryIO::writeRNG(sRNG, pRNG, rng, 0,nersc_csum,scidac_csuma,scidac_csumb);
+      ScidacWriter _ScidacWriter(grid->IsBoss());
+      _ScidacWriter.open(config);
+      _ScidacWriter.writeScidacFieldRecord(U, MData);
+      _ScidacWriter.close();
+
+      std::cout << GridLogMessage << "Written Scidac Configuration on " << config
+                << " checksum " << std::hex << nersc_csum<<"/"
+		            << scidac_csuma<<"/" << scidac_csumb
+		            << std::dec << std::endl;
+    }
+  };
+
+  void CheckpointRestore(int traj, Field &U, GridSerialRNG &sRNG,
+                         GridParallelRNG &pRNG) {
+    std::string config, rng;
+    this->build_filenames(traj, Params, config, rng);
+    this->check_filename(rng);
+    this->check_filename(config);
+
+
+    uint32_t nersc_csum,scidac_csuma,scidac_csumb;
+    BinaryIO::readRNG(sRNG, pRNG, rng, 0,nersc_csum,scidac_csuma,scidac_csumb);
+
+    Metadata md_content;
+    ScidacReader _ScidacReader;
+    _ScidacReader.open(config);
+    _ScidacReader.readScidacFieldRecord(U,md_content);  // format from the header
+    _ScidacReader.close();
+
+    std::cout << GridLogMessage << "Read Scidac Configuration from " << config
+              << " checksum " << std::hex 
+	      << nersc_csum<<"/"
+	      << scidac_csuma<<"/"
+	      << scidac_csumb
+	      << std::dec << std::endl;
+  };
+};
+}
+}
+
+#endif  // HAVE_LIME
+#endif  // ILDG_CHECKPOINTER
--- a/lib/qcd/hmc/integrators/Integrator.h
+++ b/lib/qcd/hmc/integrators/Integrator.h
@@ -114,18 +114,26 @@ class Integrator {
    // input U actually not used in the fundamental case
    // Fundamental updates, include smearing

-   for (int a = 0; a < as[level].actions.size(); ++a) {
+    for (int a = 0; a < as[level].actions.size(); ++a) {
+      double start_full = usecond();
      Field force(U._grid);
      conformable(U._grid, Mom._grid);
+
      Field& Us = Smearer.get_U(as[level].actions.at(a)->is_smeared);
+      double start_force = usecond();
      as[level].actions.at(a)->deriv(Us, force);  // deriv should NOT include Ta

      std::cout << GridLogIntegrator << "Smearing (on/off): " << as[level].actions.at(a)->is_smeared << std::endl;
      if (as[level].actions.at(a)->is_smeared) Smearer.smeared_force(force);
      force = FieldImplementation::projectForce(force); // Ta for gauge fields
+      double end_force = usecond();
      Real force_abs = std::sqrt(norm2(force)/U._grid->gSites());
-      std::cout << GridLogIntegrator << "Force average: " << force_abs << std::endl;
+      std::cout << GridLogIntegrator << "["<<level<<"]["<<a<<"] Force average: " << force_abs << std::endl;
      Mom -= force * ep; 
+      double end_full = usecond();
+      double time_full  = (end_full - start_full) / 1e3;
+      double time_force = (end_force - start_force) / 1e3;
+      std::cout << GridLogIntegrator << "["<<level<<"]["<<a<<"] P update elapsed time: " << time_full << " ms (force: " << time_force << " ms)"  << std::endl;
    }

    // Force from the other representations
--- a/lib/qcd/smearing/GaugeConfiguration.h
+++ b/lib/qcd/smearing/GaugeConfiguration.h
@@ -6,30 +6,33 @@
 #ifndef GAUGE_CONFIG_
 #define GAUGE_CONFIG_

-namespace Grid {
+namespace Grid
+{

-namespace QCD {
+namespace QCD
+{

-  //trivial class for no smearing
-  template< class Impl >
-class NoSmearing {
+//trivial class for no smearing
+template <class Impl>
+class NoSmearing
+{
 public:
  INHERIT_FIELD_TYPES(Impl);

-  Field* ThinField;
+  Field *ThinField;

-  NoSmearing(): ThinField(NULL) {}
+  NoSmearing() : ThinField(NULL) {}

-  void set_Field(Field& U) { ThinField = &U; }
+  void set_Field(Field &U) { ThinField = &U; }

-  void smeared_force(Field&) const {}
+  void smeared_force(Field &) const {}

-  Field& get_SmearedU() { return *ThinField; }
+  Field &get_SmearedU() { return *ThinField; }

-  Field& get_U(bool smeared = false) {
+  Field &get_U(bool smeared = false)
+  {
    return *ThinField;
  }
-
 };

 /*!
@@ -44,32 +47,36 @@ public:
  It stores a list of smeared configurations.
 */
 template <class Gimpl>
-class SmearedConfiguration {
- public:
+class SmearedConfiguration
+{
+public:
  INHERIT_GIMPL_TYPES(Gimpl);

- private:
+private:
  const unsigned int smearingLevels;
  Smear_Stout<Gimpl> StoutSmearing;
  std::vector<GaugeField> SmearedSet;

  // Member functions
  //====================================================================
-  void fill_smearedSet(GaugeField& U) {
-    ThinLinks = &U;  // attach the smearing routine to the field U
+  void fill_smearedSet(GaugeField &U)
+  {
+    ThinLinks = &U; // attach the smearing routine to the field U

    // check the pointer is not null
    if (ThinLinks == NULL)
      std::cout << GridLogError
                << "[SmearedConfiguration] Error in ThinLinks pointer\n";

-    if (smearingLevels > 0) {
+    if (smearingLevels > 0)
+    {
      std::cout << GridLogDebug
                << "[SmearedConfiguration] Filling SmearedSet\n";
      GaugeField previous_u(ThinLinks->_grid);

      previous_u = *ThinLinks;
-      for (int smearLvl = 0; smearLvl < smearingLevels; ++smearLvl) {
+      for (int smearLvl = 0; smearLvl < smearingLevels; ++smearLvl)
+      {
        StoutSmearing.smear(SmearedSet[smearLvl], previous_u);
        previous_u = SmearedSet[smearLvl];

@@ -81,9 +88,10 @@ class SmearedConfiguration {
    }
  }
  //====================================================================
-  GaugeField AnalyticSmearedForce(const GaugeField& SigmaKPrime,
-                                  const GaugeField& GaugeK) const {
-    GridBase* grid = GaugeK._grid;
+  GaugeField AnalyticSmearedForce(const GaugeField &SigmaKPrime,
+                                  const GaugeField &GaugeK) const
+  {
+    GridBase *grid = GaugeK._grid;
    GaugeField C(grid), SigmaK(grid), iLambda(grid);
    GaugeLinkField iLambda_mu(grid);
    GaugeLinkField iQ(grid), e_iQ(grid);
@@ -94,7 +102,8 @@ class SmearedConfiguration {
    SigmaK = zero;
    iLambda = zero;

-    for (int mu = 0; mu < Nd; mu++) {
+    for (int mu = 0; mu < Nd; mu++)
+    {
      Cmu = peekLorentz(C, mu);
      GaugeKmu = peekLorentz(GaugeK, mu);
      SigmaKPrime_mu = peekLorentz(SigmaKPrime, mu);
@@ -104,20 +113,22 @@ class SmearedConfiguration {
      pokeLorentz(iLambda, iLambda_mu, mu);
    }
    StoutSmearing.derivative(SigmaK, iLambda,
-                             GaugeK);  // derivative of SmearBase
+                             GaugeK); // derivative of SmearBase
    return SigmaK;
  }

  /*! @brief Returns smeared configuration at level 'Level' */
-  const GaugeField& get_smeared_conf(int Level) const {
+  const GaugeField &get_smeared_conf(int Level) const
+  {
    return SmearedSet[Level];
  }

  //====================================================================
-  void set_iLambda(GaugeLinkField& iLambda, GaugeLinkField& e_iQ,
-                   const GaugeLinkField& iQ, const GaugeLinkField& Sigmap,
-                   const GaugeLinkField& GaugeK) const {
-    GridBase* grid = iQ._grid;
+  void set_iLambda(GaugeLinkField &iLambda, GaugeLinkField &e_iQ,
+                   const GaugeLinkField &iQ, const GaugeLinkField &Sigmap,
+                   const GaugeLinkField &GaugeK) const
+  {
+    GridBase *grid = iQ._grid;
    GaugeLinkField iQ2(grid), iQ3(grid), B1(grid), B2(grid), USigmap(grid);
    GaugeLinkField unity(grid);
    unity = 1.0;
@@ -206,15 +217,15 @@ class SmearedConfiguration {
  }

  //====================================================================
- public:
-  GaugeField*
-      ThinLinks; /*!< @brief Pointer to the thin
-                                                         links configuration */
+public:
+  GaugeField *
+      ThinLinks; /* Pointer to the thin links configuration */

-  /*! @brief Standard constructor */
-  SmearedConfiguration(GridCartesian* UGrid, unsigned int Nsmear,
-                       Smear_Stout<Gimpl>& Stout)
-      : smearingLevels(Nsmear), StoutSmearing(Stout), ThinLinks(NULL) {
+  /* Standard constructor */
+  SmearedConfiguration(GridCartesian *UGrid, unsigned int Nsmear,
+                       Smear_Stout<Gimpl> &Stout)
+      : smearingLevels(Nsmear), StoutSmearing(Stout), ThinLinks(NULL)
+  {
    for (unsigned int i = 0; i < smearingLevels; ++i)
      SmearedSet.push_back(*(new GaugeField(UGrid)));
  }
@@ -223,21 +234,29 @@ class SmearedConfiguration {
  SmearedConfiguration()
      : smearingLevels(0), StoutSmearing(), SmearedSet(), ThinLinks(NULL) {}

-
-  
  // attach the smeared routines to the thin links U and fill the smeared set
-  void set_Field(GaugeField& U) { fill_smearedSet(U); }
+  void set_Field(GaugeField &U)
+  {
+    double start = usecond();
+    fill_smearedSet(U);
+    double end = usecond();
+    double time = (end - start)/ 1e3;
+    std::cout << GridLogMessage << "Smearing in " << time << " ms" << std::endl;  
+  }

  //====================================================================
-  void smeared_force(GaugeField& SigmaTilde) const {
-    if (smearingLevels > 0) {
+  void smeared_force(GaugeField &SigmaTilde) const
+  {
+    if (smearingLevels > 0)
+    {
+      double start = usecond();
      GaugeField force = SigmaTilde; // actually = U*SigmaTilde
      GaugeLinkField tmp_mu(SigmaTilde._grid);

-      for (int mu = 0; mu < Nd; mu++) {
+      for (int mu = 0; mu < Nd; mu++)
+      {
        // to get just SigmaTilde
-        tmp_mu = adj(peekLorentz(SmearedSet[smearingLevels - 1], mu)) *
-                 peekLorentz(force, mu);
+        tmp_mu = adj(peekLorentz(SmearedSet[smearingLevels - 1], mu)) * peekLorentz(force, mu);
        pokeLorentz(force, tmp_mu, mu);
      }

@@ -246,33 +265,43 @@ class SmearedConfiguration {

      force = AnalyticSmearedForce(force, *ThinLinks);

-      for (int mu = 0; mu < Nd; mu++) {
+      for (int mu = 0; mu < Nd; mu++)
+      {
        tmp_mu = peekLorentz(*ThinLinks, mu) * peekLorentz(force, mu);
        pokeLorentz(SigmaTilde, tmp_mu, mu);
      }
-    }  // if smearingLevels = 0 do nothing
+      double end = usecond();
+      double time = (end - start)/ 1e3;
+      std::cout << GridLogMessage << "Smearing force in " << time << " ms" << std::endl;  
+    } // if smearingLevels = 0 do nothing
  }
  //====================================================================

-  GaugeField& get_SmearedU() { return SmearedSet[smearingLevels - 1]; }
+  GaugeField &get_SmearedU() { return SmearedSet[smearingLevels - 1]; }

-  GaugeField& get_U(bool smeared = false) {
+  GaugeField &get_U(bool smeared = false)
+  {
    // get the config, thin links by default
-    if (smeared) {
-      if (smearingLevels) {
+    if (smeared)
+    {
+      if (smearingLevels)
+      {
        RealD impl_plaq =
            WilsonLoops<Gimpl>::avgPlaquette(SmearedSet[smearingLevels - 1]);
        std::cout << GridLogDebug << "getting Usmr Plaq: " << impl_plaq
                  << std::endl;
        return get_SmearedU();
-
-      } else {
+      }
+      else
+      {
        RealD impl_plaq = WilsonLoops<Gimpl>::avgPlaquette(*ThinLinks);
        std::cout << GridLogDebug << "getting Thin Plaq: " << impl_plaq
                  << std::endl;
        return *ThinLinks;
      }
-    } else {
+    }
+    else
+    {
      RealD impl_plaq = WilsonLoops<Gimpl>::avgPlaquette(*ThinLinks);
      std::cout << GridLogDebug << "getting Thin Plaq: " << impl_plaq
                << std::endl;
--- a/lib/qcd/smearing/WilsonFlow.h
+++ b/lib/qcd/smearing/WilsonFlow.h
@@ -173,8 +173,8 @@ void WilsonFlow<Gimpl>::smear(GaugeField& out, const GaugeField& in) const {
        std::cout << "Time to evolve " << diff.count() << " s\n";
        #endif
        std::cout << GridLogMessage << "[WilsonFlow] Energy density (plaq) : "
-            << step << "  "
-            << energyDensityPlaquette(step,out) << std::endl;
+		  << step << "  " << tau(step) << "  " 
+		  << energyDensityPlaquette(step,out) << std::endl;
         if( step % measure_interval == 0){
         std::cout << GridLogMessage << "[WilsonFlow] Top. charge           : "
            << step << "  " 
@@ -193,8 +193,8 @@ void WilsonFlow<Gimpl>::smear_adaptive(GaugeField& out, const GaugeField& in, Re
        //std::cout << GridLogMessage << "Evolution time :"<< taus << std::endl;
        evolve_step_adaptive(out, maxTau);
        std::cout << GridLogMessage << "[WilsonFlow] Energy density (plaq) : "
-            << step << "  "
-            << energyDensityPlaquette(out) << std::endl;
+		  << step << "  " << taus << "  "
+		  << energyDensityPlaquette(out) << std::endl;
         if( step % measure_interval == 0){
         std::cout << GridLogMessage << "[WilsonFlow] Top. charge           : "
            << step << "  " 
--- a/lib/serialisation/Hdf5IO.cc
+++ b/lib/serialisation/Hdf5IO.cc
@@ -55,6 +55,11 @@ void Hdf5Writer::writeDefault(const std::string &s, const char *x)
  writeDefault(s, sx);
 }

+Group & Hdf5Writer::getGroup(void)
+{
+  return group_;
+}
+
 // Reader implementation ///////////////////////////////////////////////////////
 Hdf5Reader::Hdf5Reader(const std::string &fileName)
 : fileName_(fileName)
@@ -103,3 +108,8 @@ void Hdf5Reader::readDefault(const std::string &s, std::string &x)
  x.resize(strType.getSize());
  attribute.read(strType, &(x[0]));
 }
+
+Group & Hdf5Reader::getGroup(void)
+{
+  return group_;
+}
--- a/lib/serialisation/Hdf5IO.h
+++ b/lib/serialisation/Hdf5IO.h
@@ -38,6 +38,7 @@ namespace Grid
    template <typename U>
    typename std::enable_if<!element<std::vector<U>>::is_number, void>::type
    writeDefault(const std::string &s, const std::vector<U> &x);
+    H5NS::Group & getGroup(void);
  private:
    template <typename U>
    void writeSingleAttribute(const U &x, const std::string &name,
@@ -65,6 +66,7 @@ namespace Grid
    template <typename U>
    typename std::enable_if<!element<std::vector<U>>::is_number, void>::type
    readDefault(const std::string &s, std::vector<U> &x);
+    H5NS::Group & getGroup(void);
  private:
    template <typename U>
    void readSingleAttribute(U &x, const std::string &name,
--- a/lib/serialisation/VectorUtils.h
+++ b/lib/serialisation/VectorUtils.h
@@ -30,6 +30,48 @@ namespace Grid {
    typedef typename std::vector<std::vector<typename TensorToVec<T>::type>> type;
  };

+  template <typename T>
+  void tensorDim(std::vector<size_t> &dim, const T &t, const bool wipe = true)
+  {
+    if (wipe)
+    {
+      dim.clear();
+    }
+  }
+
+  template <typename T>
+  void tensorDim(std::vector<size_t> &dim, const iScalar<T> &t, const bool wipe = true)
+  {
+    if (wipe)
+    {
+      dim.clear();
+    }
+    tensorDim(dim, t._internal, false);
+  }
+
+  template <typename T, int N>
+  void tensorDim(std::vector<size_t> &dim, const iVector<T, N> &t, const bool wipe = true)
+  {
+    if (wipe)
+    {
+      dim.clear();
+    }
+    dim.push_back(N);
+    tensorDim(dim, t._internal[0], false);
+  }
+
+  template <typename T, int N>
+  void tensorDim(std::vector<size_t> &dim, const iMatrix<T, N> &t, const bool wipe = true)
+  {
+    if (wipe)
+    {
+      dim.clear();
+    }
+    dim.push_back(N);
+    dim.push_back(N);
+    tensorDim(dim, t._internal[0][0], false);
+  }
+
  template <typename T>
  typename TensorToVec<T>::type tensorToVec(const T &t)
  {
--- a/lib/serialisation/XmlIO.cc
+++ b/lib/serialisation/XmlIO.cc
@@ -31,6 +31,17 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 using namespace Grid;
 using namespace std;

+void Grid::xmlCheckParse(const pugi::xml_parse_result &result, const std::string name)
+{
+  if (!result) 
+  {
+    std::cerr << "XML parsing error for " << name << std::endl;
+    std::cerr << "XML error description: " << result.description() << std::endl;
+    std::cerr << "XML error offset     : " << result.offset << std::endl;
+    abort();
+  }
+}
+
 // Writer implementation ///////////////////////////////////////////////////////
 XmlWriter::XmlWriter(const string &fileName, string toplev) : fileName_(fileName)
 {
@@ -54,6 +65,19 @@ void XmlWriter::push(const string &s)
  node_ = node_.append_child(s.c_str());
 }

+void XmlWriter::pushXmlString(const std::string &s)
+{
+  pugi::xml_document doc;
+  auto               result = doc.load_buffer(s.c_str(), s.size());
+
+  xmlCheckParse(result, "fragment\n'" + s +"'");
+  for (pugi::xml_node child = doc.first_child(); child; child = child.next_sibling())
+  {
+      node_ = node_.append_copy(child);
+  }
+  pop();
+}
+
 void XmlWriter::pop(void)
 {
  node_ = node_.parent();
@@ -65,40 +89,32 @@ std::string XmlWriter::XmlString(void)
  return oss.str();
 }

-XmlReader::XmlReader(const char *xmlstring,string toplev) : fileName_("")
-{
-  pugi::xml_parse_result result;
-  result = doc_.load_string(xmlstring);
-  if ( !result ) {
-    cerr << "XML error description (from char *): " << result.description() << "\nXML\n"<< xmlstring << "\n";
-    cerr << "XML error offset      (from char *) " << result.offset         << "\nXML\n"<< xmlstring <<"\n";
-    abort();
-  }
-  if ( toplev == std::string("") ) {
-    node_ = doc_;
-  } else { 
-    node_ = doc_.child(toplev.c_str());
-  }
-}
-
 // Reader implementation ///////////////////////////////////////////////////////
-XmlReader::XmlReader(const string &fileName,string toplev) : fileName_(fileName)
+XmlReader::XmlReader(const std::string &s,  const bool isBuffer, 
+                     std::string toplev) 
 {
  pugi::xml_parse_result result;
-  result = doc_.load_file(fileName_.c_str());
-  if ( !result ) {
-    cerr << "XML error description: " << result.description() <<" "<< fileName_ <<"\n";
-    cerr << "XML error offset     : " << result.offset        <<" "<< fileName_ <<"\n";
-    abort();
+  
+  if (isBuffer)
+  {
+    fileName_ = "<string>";
+    result    = doc_.load_string(s.c_str());
+    xmlCheckParse(result, "string\n'" + s + "'");
+  }
+  else
+  {
+    fileName_ = s;
+    result    = doc_.load_file(s.c_str());
+    xmlCheckParse(result, "file '" + fileName_ + "'");
  }
  if ( toplev == std::string("") ) {
-    node_ = doc_;
+  node_ = doc_;
  } else { 
    node_ = doc_.child(toplev.c_str());
  }
 }

-bool XmlReader::push(const string &s)
+bool XmlReader::push(const std::string &s)
 {
  if (node_.child(s.c_str()))
  {
@@ -129,7 +145,6 @@ bool XmlReader::nextElement(const std::string &s)
  {
    return false;
  }
-
 }

 template <>
--- a/lib/serialisation/XmlIO.h
+++ b/lib/serialisation/XmlIO.h
@@ -43,13 +43,15 @@ Author: paboyle <paboyle@ph.ed.ac.uk>

 namespace Grid
 {
+  void xmlCheckParse(const pugi::xml_parse_result &result, const std::string name);
  
  class XmlWriter: public Writer<XmlWriter>
  {    
  public:
-    XmlWriter(const std::string &fileName,std::string toplev = std::string("grid") );
+    XmlWriter(const std::string &fileName, std::string toplev = std::string("grid") );
    virtual ~XmlWriter(void);
    void push(const std::string &s);
+    void pushXmlString(const std::string &s);
    void pop(void);
    template <typename U>
    void writeDefault(const std::string &s, const U &x);
@@ -65,8 +67,8 @@ namespace Grid
  class XmlReader: public Reader<XmlReader>
  {
  public:
-    XmlReader(const char *xmlstring,std::string toplev = std::string("grid") );
-    XmlReader(const std::string &fileName,std::string toplev = std::string("grid") );
+    XmlReader(const std::string &fileName, const bool isBuffer = false, 
+              std::string toplev = std::string("grid") );
    virtual ~XmlReader(void) = default;
    bool push(const std::string &s);
    void pop(void);
@@ -75,6 +77,8 @@ namespace Grid
    void readDefault(const std::string &s, U &output);
    template <typename U>
    void readDefault(const std::string &s, std::vector<U> &output);
+  private:
+    void checkParse(const pugi::xml_parse_result &result, const std::string name);
  private:
    pugi::xml_document doc_;
    pugi::xml_node     node_;
--- a/lib/stencil/Stencil.h
+++ b/lib/stencil/Stencil.h
@@ -66,6 +66,8 @@ void Gather_plane_simple_table (std::vector<std::pair<int,int> >& table,const La
  parallel_for(int i=0;i<num;i++){
    compress.Compress(&buffer[off],table[i].first,rhs._odata[so+table[i].second]);
  }
+// Further optimisatoin: i) streaming store the result
+//                       ii) software prefetch the first element of the next table entry
 }

 ///////////////////////////////////////////////////////////////////
@@ -505,25 +507,24 @@ class CartesianStencil { // Stencil runs along coordinate axes only; NO diagonal
  template<class decompressor>
  void CommsMerge(decompressor decompress,std::vector<Merge> &mm,std::vector<Decompress> &dd) { 

+    mergetime-=usecond();
    for(int i=0;i<mm.size();i++){	
-      mergetime-=usecond();
      parallel_for(int o=0;o<mm[i].buffer_size/2;o++){
 	decompress.Exchange(mm[i].mpointer,
 			    mm[i].vpointers[0],
 			    mm[i].vpointers[1],
 			    mm[i].type,o);
      }
-      mergetime+=usecond();
    }
+    mergetime+=usecond();

+    decompresstime-=usecond();
    for(int i=0;i<dd.size();i++){	
-      decompresstime-=usecond();
      parallel_for(int o=0;o<dd[i].buffer_size;o++){
 	decompress.Decompress(dd[i].kernel_p,dd[i].mpi_p,o);
      }      
-      decompresstime+=usecond();
    }
-
+    decompresstime+=usecond();
  }
  ////////////////////////////////////////
  // Set up routines
--- a/lib/util/Init.cc
+++ b/lib/util/Init.cc
@@ -297,12 +297,7 @@ void Grid_init(int *argc,char ***argv)
    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;
-#ifdef GITHASH
-    std::cout << "Current Grid git commit hash=" << GITHASH << std::endl;
-#else
-    std::cout << "Current Grid git commit hash is undefined. Check makefile." << std::endl;
-#endif
-#undef GITHASH
+    printHash();
    std::cout << std::endl;
  }

@@ -398,6 +393,7 @@ void Grid_init(int *argc,char ***argv)
  if( GridCmdOptionExists(*argv,*argv+*argc,"--comms-threads") ){
    arg= GridCmdOptionPayload(*argv,*argv+*argc,"--comms-threads");
    GridCmdOptionInt(arg,CartesianCommunicator::nCommThreads);
+    assert(CartesianCommunicator::nCommThreads > 0);
  }
  if( GridCmdOptionExists(*argv,*argv+*argc,"--cacheblocking") ){
    arg= GridCmdOptionPayload(*argv,*argv+*argc,"--cacheblocking");
--- a/lib/util/Init.h
+++ b/lib/util/Init.h
@@ -64,6 +64,7 @@ namespace Grid {
 		       std::vector<int> &simd,
 		       std::vector<int> &mpi);

+  void printHash(void);

 };
 #endif
--- a/lib/util/version.cc
+++ b/lib/util/version.cc
@@ -0,0 +1,12 @@
+#include <iostream>
+#include <version.h>
+namespace Grid {
+  void printHash(){
+#ifdef GITHASH
+    std::cout << "Current Grid git commit hash=" << GITHASH << std::endl;
+#else
+    std::cout << "Current Grid git commit hash is undefined. Check makefile." << std::endl;
+#endif
+#undef GITHASH
+}
+}