merged upstream

Grid/DisableWarnings.h

@@ -44,14 +44,22 @@ directory
 #ifdef __NVCC__
  //disables nvcc specific warning in json.hpp
 #pragma clang diagnostic ignored "-Wdeprecated-register"
+
+#if (__CUDACC_VER_MAJOR__ >= 11) && (__CUDACC_VER_MINOR__ >= 5)
+ //disables nvcc specific warning in json.hpp
+#pragma nv_diag_suppress unsigned_compare_with_zero
+#pragma nv_diag_suppress cast_to_qualified_type
+ //disables nvcc specific warning in many files
+#pragma nv_diag_suppress esa_on_defaulted_function_ignored
+#pragma nv_diag_suppress extra_semicolon
+#else
+ //disables nvcc specific warning in json.hpp
 #pragma diag_suppress unsigned_compare_with_zero
 #pragma diag_suppress cast_to_qualified_type
-
  //disables nvcc specific warning in many files
 #pragma diag_suppress esa_on_defaulted_function_ignored
 #pragma diag_suppress extra_semicolon
-
+#endif
-//Eigen only
 #endif
 
 // Disable vectorisation in Eigen on the Power8/9 and PowerPC

Grid/GridStd.h

@@ -16,6 +16,7 @@
 #include <functional>
 #include <stdio.h>
 #include <stdlib.h>
+#include <strings.h>
 #include <stdio.h>
 #include <signal.h>
 #include <ctime>

Grid/Grid_Eigen_Dense.h

@@ -14,7 +14,11 @@
 /* NVCC save and restore compile environment*/
 #ifdef __NVCC__
 #pragma push
+#if (__CUDACC_VER_MAJOR__ >= 11) && (__CUDACC_VER_MINOR__ >= 5)
+#pragma nv_diag_suppress code_is_unreachable
+#else
 #pragma diag_suppress code_is_unreachable
+#endif
 #pragma push_macro("__CUDA_ARCH__")
 #pragma push_macro("__NVCC__")
 #pragma push_macro("__CUDACC__")

Grid/algorithms/CoarsenedMatrix.h

@@ -262,7 +262,7 @@ public:
 	autoView( Tnp_v , (*Tnp), AcceleratorWrite);
 	autoView( Tnm_v , (*Tnm), AcceleratorWrite);
 	const int Nsimd = CComplex::Nsimd();
-	accelerator_forNB(ss, FineGrid->oSites(), Nsimd, {
+	accelerator_for(ss, FineGrid->oSites(), Nsimd, {
 	  coalescedWrite(y_v[ss],xscale*y_v(ss)+mscale*Tn_v(ss));
 	  coalescedWrite(Tnp_v[ss],2.0*y_v(ss)-Tnm_v(ss));
         });

Grid/algorithms/approx/Chebyshev.h

@@ -264,7 +264,7 @@ public:
       auto Tnp_v = Tnp->View();
       auto Tnm_v = Tnm->View();
       constexpr int Nsimd = vector_type::Nsimd();
-      accelerator_forNB(ss, in.Grid()->oSites(), Nsimd, {
+      accelerator_for(ss, in.Grid()->oSites(), Nsimd, {
 	  coalescedWrite(y_v[ss],xscale*y_v(ss)+mscale*Tn_v(ss));
 	  coalescedWrite(Tnp_v[ss],2.0*y_v(ss)-Tnm_v(ss));
       });

Grid/algorithms/iterative/Deflation.h

@@ -113,7 +113,43 @@ public:
     blockPromote(guess_coarse,guess,subspace);
     guess.Checkerboard() = src.Checkerboard();
   };
-};
+
+  void operator()(const std::vector<FineField> &src,std::vector<FineField> &guess) {
+    int Nevec = (int)evec_coarse.size();
+    int Nsrc = (int)src.size();
+    // make temp variables
+    std::vector<CoarseField> src_coarse(Nsrc,evec_coarse[0].Grid());
+    std::vector<CoarseField> guess_coarse(Nsrc,evec_coarse[0].Grid());    
+    //Preporcessing
+    std::cout << GridLogMessage << "Start BlockProject for loop" << std::endl;
+    for (int j=0;j<Nsrc;j++)
+    {
+    guess_coarse[j] = Zero();
+    std::cout << GridLogMessage << "BlockProject iter: " << j << std::endl;
+    blockProject(src_coarse[j],src[j],subspace);
+    }
+    //deflation set up for eigen vector batchsize 1 and source batch size equal number of sources
+    std::cout << GridLogMessage << "Start ProjectAccum for loop" << std::endl;
+    for (int i=0;i<Nevec;i++)
+    {
+      std::cout << GridLogMessage << "ProjectAccum Nvec: " << i << std::endl;
+      const CoarseField & tmp = evec_coarse[i];
+      for (int j=0;j<Nsrc;j++)
+      {
+        axpy(guess_coarse[j],TensorRemove(innerProduct(tmp,src_coarse[j])) / eval_coarse[i],tmp,guess_coarse[j]);
+      }
+    }
+    //postprocessing
+    std::cout << GridLogMessage << "Start BlockPromote for loop" << std::endl;
+    for (int j=0;j<Nsrc;j++)
+    {
+    std::cout << GridLogMessage << "BlockProject iter: " << j << std::endl;
+    blockPromote(guess_coarse[j],guess[j],subspace);
+    guess[j].Checkerboard() = src[j].Checkerboard();
+    }
+  };
+
+  };
 
 
 

Grid/allocator/MemoryManager.cc

@@ -40,7 +40,7 @@ void MemoryManager::PrintBytes(void)
 //////////////////////////////////////////////////////////////////////
 MemoryManager::AllocationCacheEntry MemoryManager::Entries[MemoryManager::NallocType][MemoryManager::NallocCacheMax];
 int MemoryManager::Victim[MemoryManager::NallocType];
-int MemoryManager::Ncache[MemoryManager::NallocType] = { 2, 8, 2, 8, 2, 8 };
+int MemoryManager::Ncache[MemoryManager::NallocType] = { 2, 8, 8, 16, 8, 16 };
 uint64_t MemoryManager::CacheBytes[MemoryManager::NallocType];
 //////////////////////////////////////////////////////////////////////
 // Actual allocation and deallocation utils

Grid/allocator/MemoryManager.h

@@ -36,6 +36,11 @@ NAMESPACE_BEGIN(Grid);
 
 #define GRID_ALLOC_SMALL_LIMIT (4096)
 
+#define STRINGIFY(x) #x
+#define TOSTRING(x) STRINGIFY(x)
+#define FILE_LINE __FILE__ ":" TOSTRING(__LINE__)
+#define AUDIT(a) MemoryManager::Audit(FILE_LINE)
+
 /*Pinning pages is costly*/
 ////////////////////////////////////////////////////////////////////////////
 // Advise the LatticeAccelerator class
@@ -92,8 +97,9 @@ private:
   static void *Insert(void *ptr,size_t bytes,AllocationCacheEntry *entries,int ncache,int &victim,uint64_t &cbytes) ;
   static void *Lookup(size_t bytes,AllocationCacheEntry *entries,int ncache,uint64_t &cbytes) ;
 
-  static void PrintBytes(void);
  public:
+  static void PrintBytes(void);
+  static void Audit(std::string s);
   static void Init(void);
   static void InitMessage(void);
   static void *AcceleratorAllocate(size_t bytes);
@@ -113,6 +119,8 @@ private:
   static uint64_t     DeviceToHostBytes;
   static uint64_t     HostToDeviceXfer;
   static uint64_t     DeviceToHostXfer;
+  static uint64_t     DeviceEvictions;
+  static uint64_t     DeviceDestroy;
  
  private:
 #ifndef GRID_UVM
@@ -170,6 +178,7 @@ private:
 
  public:
   static void Print(void);
+  static void PrintAll(void);
   static void PrintState( void* CpuPtr);
   static int   isOpen   (void* CpuPtr);
   static void  ViewClose(void* CpuPtr,ViewMode mode);

Grid/allocator/MemoryManagerCache.cc

@@ -3,8 +3,13 @@
 
 #warning "Using explicit device memory copies"
 NAMESPACE_BEGIN(Grid);
-//#define dprintf(...) printf ( __VA_ARGS__ ); fflush(stdout);
+
-#define dprintf(...)
+#define MAXLINE 512
+static char print_buffer [ MAXLINE ];
+
+#define mprintf(...) snprintf (print_buffer,MAXLINE, __VA_ARGS__ ); std::cout << GridLogMemory << print_buffer;
+#define dprintf(...) snprintf (print_buffer,MAXLINE, __VA_ARGS__ ); std::cout << GridLogMemory << print_buffer;
+//#define dprintf(...) 
 
 
 ////////////////////////////////////////////////////////////
@@ -23,6 +28,8 @@ uint64_t  MemoryManager::HostToDeviceBytes;
 uint64_t  MemoryManager::DeviceToHostBytes;
 uint64_t  MemoryManager::HostToDeviceXfer;
 uint64_t  MemoryManager::DeviceToHostXfer;
+uint64_t  MemoryManager::DeviceEvictions;
+uint64_t  MemoryManager::DeviceDestroy;
 
 ////////////////////////////////////
 // Priority ordering for unlocked entries
@@ -104,15 +111,17 @@ void MemoryManager::AccDiscard(AcceleratorViewEntry &AccCache)
   ///////////////////////////////////////////////////////////
   assert(AccCache.state!=Empty);
   
-   dprintf("MemoryManager: Discard(%llx) %llx\n",(uint64_t)AccCache.CpuPtr,(uint64_t)AccCache.AccPtr); 
+  mprintf("MemoryManager: Discard(%lx) %lx\n",(uint64_t)AccCache.CpuPtr,(uint64_t)AccCache.AccPtr); 
   assert(AccCache.accLock==0);
   assert(AccCache.cpuLock==0);
   assert(AccCache.CpuPtr!=(uint64_t)NULL);
   if(AccCache.AccPtr) {
     AcceleratorFree((void *)AccCache.AccPtr,AccCache.bytes);
+    DeviceDestroy++;
     DeviceBytes   -=AccCache.bytes;
     LRUremove(AccCache);
-    dprintf("MemoryManager: Free(%llx) LRU %lld Total %lld\n",(uint64_t)AccCache.AccPtr,DeviceLRUBytes,DeviceBytes);  
+    AccCache.AccPtr=(uint64_t) NULL;
+    dprintf("MemoryManager: Free(%lx) LRU %ld Total %ld\n",(uint64_t)AccCache.AccPtr,DeviceLRUBytes,DeviceBytes);  
   }
   uint64_t CpuPtr = AccCache.CpuPtr;
   EntryErase(CpuPtr);
@@ -121,26 +130,36 @@ void MemoryManager::AccDiscard(AcceleratorViewEntry &AccCache)
 void MemoryManager::Evict(AcceleratorViewEntry &AccCache)
 {
   ///////////////////////////////////////////////////////////////////////////
-  // Make CPU consistent, remove from Accelerator, remove entry
+  // Make CPU consistent, remove from Accelerator, remove from LRU, LEAVE CPU only entry
-  // Cannot be locked. If allocated must be in LRU pool.
+  // Cannot be acclocked. If allocated must be in LRU pool.
+  //
+  // Nov 2022... Felix issue: Allocating two CpuPtrs, can have an entry in LRU-q with CPUlock.
+  //                          and require to evict the AccPtr copy. Eviction was a mistake in CpuViewOpen
+  //                          but there is a weakness where CpuLock entries are attempted for erase
+  //                          Take these OUT LRU queue when CPU locked?
+  //                          Cannot take out the table as cpuLock data is important.
   ///////////////////////////////////////////////////////////////////////////
   assert(AccCache.state!=Empty);
   
-  dprintf("MemoryManager: Evict(%llx) %llx\n",(uint64_t)AccCache.CpuPtr,(uint64_t)AccCache.AccPtr); 
+  mprintf("MemoryManager: Evict cpu %lx acc %lx cpuLock %ld accLock %ld\n",
-  assert(AccCache.accLock==0);
+	  (uint64_t)AccCache.CpuPtr,(uint64_t)AccCache.AccPtr,
-  assert(AccCache.cpuLock==0);
+	  (uint64_t)AccCache.cpuLock,(uint64_t)AccCache.accLock); 
+  assert(AccCache.accLock==0); // Cannot evict so logic bomb
+  assert(AccCache.CpuPtr!=(uint64_t)NULL);
   if(AccCache.state==AccDirty) {
     Flush(AccCache);
   }
-  assert(AccCache.CpuPtr!=(uint64_t)NULL);
   if(AccCache.AccPtr) {
     AcceleratorFree((void *)AccCache.AccPtr,AccCache.bytes);
-    DeviceBytes   -=AccCache.bytes;
     LRUremove(AccCache);
-    dprintf("MemoryManager: Free(%llx) footprint now %lld \n",(uint64_t)AccCache.AccPtr,DeviceBytes);  
+    AccCache.AccPtr=(uint64_t)NULL;
+    AccCache.state=CpuDirty; // CPU primary now
+    DeviceBytes   -=AccCache.bytes;
+    dprintf("MemoryManager: Free(%lx) footprint now %ld \n",(uint64_t)AccCache.AccPtr,DeviceBytes);  
   }
-  uint64_t CpuPtr = AccCache.CpuPtr;
+  //  uint64_t CpuPtr = AccCache.CpuPtr;
-  EntryErase(CpuPtr);
+  DeviceEvictions++;
+  //  EntryErase(CpuPtr);
 }
 void MemoryManager::Flush(AcceleratorViewEntry &AccCache)
 {
@@ -150,7 +169,7 @@ void MemoryManager::Flush(AcceleratorViewEntry &AccCache)
   assert(AccCache.AccPtr!=(uint64_t)NULL);
   assert(AccCache.CpuPtr!=(uint64_t)NULL);
   acceleratorCopyFromDevice((void *)AccCache.AccPtr,(void *)AccCache.CpuPtr,AccCache.bytes);
-  dprintf("MemoryManager: Flush  %llx -> %llx\n",(uint64_t)AccCache.AccPtr,(uint64_t)AccCache.CpuPtr); fflush(stdout);
+  mprintf("MemoryManager: Flush  %lx -> %lx\n",(uint64_t)AccCache.AccPtr,(uint64_t)AccCache.CpuPtr); fflush(stdout);
   DeviceToHostBytes+=AccCache.bytes;
   DeviceToHostXfer++;
   AccCache.state=Consistent;
@@ -165,7 +184,7 @@ void MemoryManager::Clone(AcceleratorViewEntry &AccCache)
     AccCache.AccPtr=(uint64_t)AcceleratorAllocate(AccCache.bytes);
     DeviceBytes+=AccCache.bytes;
   }
-  dprintf("MemoryManager: Clone %llx <- %llx\n",(uint64_t)AccCache.AccPtr,(uint64_t)AccCache.CpuPtr); fflush(stdout);
+  mprintf("MemoryManager: Clone %lx <- %lx\n",(uint64_t)AccCache.AccPtr,(uint64_t)AccCache.CpuPtr); fflush(stdout);
   acceleratorCopyToDevice((void *)AccCache.CpuPtr,(void *)AccCache.AccPtr,AccCache.bytes);
   HostToDeviceBytes+=AccCache.bytes;
   HostToDeviceXfer++;
@@ -191,6 +210,7 @@ void MemoryManager::CpuDiscard(AcceleratorViewEntry &AccCache)
 void MemoryManager::ViewClose(void* Ptr,ViewMode mode)
 {
   if( (mode==AcceleratorRead)||(mode==AcceleratorWrite)||(mode==AcceleratorWriteDiscard) ){
+    dprintf("AcceleratorViewClose %lx\n",(uint64_t)Ptr);
     AcceleratorViewClose((uint64_t)Ptr);
   } else if( (mode==CpuRead)||(mode==CpuWrite)){
     CpuViewClose((uint64_t)Ptr);
@@ -202,6 +222,7 @@ void *MemoryManager::ViewOpen(void* _CpuPtr,size_t bytes,ViewMode mode,ViewAdvis
 {
   uint64_t CpuPtr = (uint64_t)_CpuPtr;
   if( (mode==AcceleratorRead)||(mode==AcceleratorWrite)||(mode==AcceleratorWriteDiscard) ){
+    dprintf("AcceleratorViewOpen %lx\n",(uint64_t)CpuPtr);
     return (void *) AcceleratorViewOpen(CpuPtr,bytes,mode,hint);
   } else if( (mode==CpuRead)||(mode==CpuWrite)){
     return (void *)CpuViewOpen(CpuPtr,bytes,mode,hint);
@@ -212,13 +233,16 @@ void *MemoryManager::ViewOpen(void* _CpuPtr,size_t bytes,ViewMode mode,ViewAdvis
 }
 void  MemoryManager::EvictVictims(uint64_t bytes)
 {
+  assert(bytes<DeviceMaxBytes);
   while(bytes+DeviceLRUBytes > DeviceMaxBytes){
     if ( DeviceLRUBytes > 0){
       assert(LRU.size()>0);
-      uint64_t victim = LRU.back();
+      uint64_t victim = LRU.back(); // From the LRU
       auto AccCacheIterator = EntryLookup(victim);
       auto & AccCache = AccCacheIterator->second;
       Evict(AccCache);
+    } else {
+      return;
     }
   }
 }
@@ -241,11 +265,12 @@ uint64_t MemoryManager::AcceleratorViewOpen(uint64_t CpuPtr,size_t bytes,ViewMod
   assert(AccCache.cpuLock==0);  // Programming error
 
   if(AccCache.state!=Empty) {
-    dprintf("ViewOpen found entry %llx %llx : %lld %lld\n",
+    dprintf("ViewOpen found entry %lx %lx : %ld %ld accLock %ld\n",
 		    (uint64_t)AccCache.CpuPtr,
 		    (uint64_t)CpuPtr,
 		    (uint64_t)AccCache.bytes,
-		    (uint64_t)bytes);
+	            (uint64_t)bytes,
+		    (uint64_t)AccCache.accLock);
     assert(AccCache.CpuPtr == CpuPtr);
     assert(AccCache.bytes  ==bytes);
   }
@@ -280,6 +305,7 @@ uint64_t MemoryManager::AcceleratorViewOpen(uint64_t CpuPtr,size_t bytes,ViewMod
       AccCache.state  = Consistent; // Empty + AccRead => Consistent
     }
     AccCache.accLock= 1;
+    dprintf("Copied Empty entry into device accLock= %d\n",AccCache.accLock);
   } else if(AccCache.state==CpuDirty ){
     if(mode==AcceleratorWriteDiscard) {
       CpuDiscard(AccCache);
@@ -292,28 +318,30 @@ uint64_t MemoryManager::AcceleratorViewOpen(uint64_t CpuPtr,size_t bytes,ViewMod
       AccCache.state  = Consistent; // CpuDirty + AccRead => Consistent
     }
     AccCache.accLock++;
-    dprintf("Copied CpuDirty entry into device accLock %d\n",AccCache.accLock);
+    dprintf("CpuDirty entry into device ++accLock= %d\n",AccCache.accLock);
   } else if(AccCache.state==Consistent) {
     if((mode==AcceleratorWrite)||(mode==AcceleratorWriteDiscard))
       AccCache.state  = AccDirty;   // Consistent + AcceleratorWrite=> AccDirty
     else
       AccCache.state  = Consistent; // Consistent + AccRead => Consistent
     AccCache.accLock++;
-    dprintf("Consistent entry into device accLock %d\n",AccCache.accLock);
+    dprintf("Consistent entry into device ++accLock= %d\n",AccCache.accLock);
   } else if(AccCache.state==AccDirty) {
     if((mode==AcceleratorWrite)||(mode==AcceleratorWriteDiscard))
       AccCache.state  = AccDirty; // AccDirty + AcceleratorWrite=> AccDirty
     else
       AccCache.state  = AccDirty; // AccDirty + AccRead => AccDirty
     AccCache.accLock++;
-    dprintf("AccDirty entry into device accLock %d\n",AccCache.accLock);
+    dprintf("AccDirty entry ++accLock= %d\n",AccCache.accLock);
   } else {
     assert(0);
   }
 
-  // If view is opened on device remove from LRU
+  assert(AccCache.accLock>0);
+  // If view is opened on device must remove from LRU
   if(AccCache.LRU_valid==1){
     // must possibly remove from LRU as now locked on GPU
+    dprintf("AccCache entry removed from LRU \n");
     LRUremove(AccCache);
   }
 
@@ -334,10 +362,12 @@ void MemoryManager::AcceleratorViewClose(uint64_t CpuPtr)
   assert(AccCache.accLock>0);
 
   AccCache.accLock--;
-
   // Move to LRU queue if not locked and close on device
   if(AccCache.accLock==0) {
+    dprintf("AccleratorViewClose %lx AccLock decremented to %ld move to LRU queue\n",(uint64_t)CpuPtr,(uint64_t)AccCache.accLock);
     LRUinsert(AccCache);
+  } else {
+    dprintf("AccleratorViewClose %lx AccLock decremented to %ld\n",(uint64_t)CpuPtr,(uint64_t)AccCache.accLock);
   }
 }
 void MemoryManager::CpuViewClose(uint64_t CpuPtr)
@@ -374,9 +404,10 @@ uint64_t MemoryManager::CpuViewOpen(uint64_t CpuPtr,size_t bytes,ViewMode mode,V
   auto AccCacheIterator = EntryLookup(CpuPtr);
   auto & AccCache = AccCacheIterator->second;
 
-  if (!AccCache.AccPtr) {
+  // CPU doesn't need to free space
-     EvictVictims(bytes);
+  //  if (!AccCache.AccPtr) {
-  }
+  //    EvictVictims(bytes);
+  //  }
 
   assert((mode==CpuRead)||(mode==CpuWrite));
   assert(AccCache.accLock==0);  // Programming error
@@ -430,20 +461,28 @@ void  MemoryManager::NotifyDeletion(void *_ptr)
 void  MemoryManager::Print(void)
 {
   PrintBytes();
-  std::cout << GridLogDebug << "--------------------------------------------" << std::endl;
+  std::cout << GridLogMessage << "--------------------------------------------" << std::endl;
-  std::cout << GridLogDebug << "Memory Manager                             " << std::endl;
+  std::cout << GridLogMessage << "Memory Manager                             " << std::endl;
-  std::cout << GridLogDebug << "--------------------------------------------" << std::endl;
+  std::cout << GridLogMessage << "--------------------------------------------" << std::endl;
-  std::cout << GridLogDebug << DeviceBytes   << " bytes allocated on device " << std::endl;
+  std::cout << GridLogMessage << DeviceBytes   << " bytes allocated on device " << std::endl;
-  std::cout << GridLogDebug << DeviceLRUBytes<< " bytes evictable on device " << std::endl;
+  std::cout << GridLogMessage << DeviceLRUBytes<< " bytes evictable on device " << std::endl;
-  std::cout << GridLogDebug << DeviceMaxBytes<< " bytes max on device       " << std::endl;
+  std::cout << GridLogMessage << DeviceMaxBytes<< " bytes max on device       " << std::endl;
-  std::cout << GridLogDebug << HostToDeviceXfer << " transfers        to   device " << std::endl;
+  std::cout << GridLogMessage << HostToDeviceXfer << " transfers        to   device " << std::endl;
-  std::cout << GridLogDebug << DeviceToHostXfer << " transfers        from device " << std::endl;
+  std::cout << GridLogMessage << DeviceToHostXfer << " transfers        from device " << std::endl;
-  std::cout << GridLogDebug << HostToDeviceBytes<< " bytes transfered to   device " << std::endl;
+  std::cout << GridLogMessage << HostToDeviceBytes<< " bytes transfered to   device " << std::endl;
-  std::cout << GridLogDebug << DeviceToHostBytes<< " bytes transfered from device " << std::endl;
+  std::cout << GridLogMessage << DeviceToHostBytes<< " bytes transfered from device " << std::endl;
-  std::cout << GridLogDebug << AccViewTable.size()<< " vectors " << LRU.size()<<" evictable"<< std::endl;
+  std::cout << GridLogMessage << DeviceEvictions  << " Evictions from device " << std::endl;
-  std::cout << GridLogDebug << "--------------------------------------------" << std::endl;
+  std::cout << GridLogMessage << DeviceDestroy    << " Destroyed vectors on device " << std::endl;
-  std::cout << GridLogDebug << "CpuAddr\t\tAccAddr\t\tState\t\tcpuLock\taccLock\tLRU_valid "<<std::endl;
+  std::cout << GridLogMessage << AccViewTable.size()<< " vectors " << LRU.size()<<" evictable"<< std::endl;
-  std::cout << GridLogDebug << "--------------------------------------------" << std::endl;
+  std::cout << GridLogMessage << "--------------------------------------------" << std::endl;
+}
+void  MemoryManager::PrintAll(void)
+{
+  Print();
+  std::cout << GridLogMessage << std::endl;
+  std::cout << GridLogMessage << "--------------------------------------------" << std::endl;
+  std::cout << GridLogMessage << "CpuAddr\t\tAccAddr\t\tState\t\tcpuLock\taccLock\tLRU_valid "<<std::endl;
+  std::cout << GridLogMessage << "--------------------------------------------" << std::endl;
   for(auto it=AccViewTable.begin();it!=AccViewTable.end();it++){
     auto &AccCache = it->second;
     
@@ -453,13 +492,13 @@ void  MemoryManager::Print(void)
     if ( AccCache.state==AccDirty ) str = std::string("AccDirty");
     if ( AccCache.state==Consistent)str = std::string("Consistent");
 
-    std::cout << GridLogDebug << "0x"<<std::hex<<AccCache.CpuPtr<<std::dec
+    std::cout << GridLogMessage << "0x"<<std::hex<<AccCache.CpuPtr<<std::dec
 	      << "\t0x"<<std::hex<<AccCache.AccPtr<<std::dec<<"\t" <<str
 	      << "\t" << AccCache.cpuLock
 	      << "\t" << AccCache.accLock
 	      << "\t" << AccCache.LRU_valid<<std::endl;
   }
-  std::cout << GridLogDebug << "--------------------------------------------" << std::endl;
+  std::cout << GridLogMessage << "--------------------------------------------" << std::endl;
 
 };
 int   MemoryManager::isOpen   (void* _CpuPtr) 
@@ -473,6 +512,61 @@ int   MemoryManager::isOpen   (void* _CpuPtr)
     return 0;
   }
 }
+void MemoryManager::Audit(std::string s)
+{
+  uint64_t CpuBytes=0;
+  uint64_t AccBytes=0;
+  uint64_t LruBytes1=0;
+  uint64_t LruBytes2=0;
+  uint64_t LruCnt=0;
+  uint64_t LockedBytes=0;
+  
+  std::cout << " Memory Manager::Audit() from "<<s<<std::endl;
+  for(auto it=LRU.begin();it!=LRU.end();it++){
+    uint64_t cpuPtr = *it;
+    assert(EntryPresent(cpuPtr));
+    auto AccCacheIterator = EntryLookup(cpuPtr);
+    auto & AccCache = AccCacheIterator->second;
+    LruBytes2+=AccCache.bytes;
+    assert(AccCache.LRU_valid==1);
+    assert(AccCache.LRU_entry==it);
+  }
+  std::cout << " Memory Manager::Audit() LRU queue matches table entries "<<std::endl;
+  for(auto it=AccViewTable.begin();it!=AccViewTable.end();it++){
+    auto &AccCache = it->second;
+    
+    std::string str;
+    if ( AccCache.state==Empty    ) str = std::string("Empty");
+    if ( AccCache.state==CpuDirty ) str = std::string("CpuDirty");
+    if ( AccCache.state==AccDirty ) str = std::string("AccDirty");
+    if ( AccCache.state==Consistent)str = std::string("Consistent");
+
+    CpuBytes+=AccCache.bytes;
+    if( AccCache.AccPtr )    AccBytes+=AccCache.bytes;
+    if( AccCache.LRU_valid ) LruBytes1+=AccCache.bytes;
+    if( AccCache.LRU_valid ) LruCnt++;
+    
+    if ( AccCache.cpuLock || AccCache.accLock ) {
+      assert(AccCache.LRU_valid==0);
+      std::cout << GridLogError << s<< "\n\t 0x"<<std::hex<<AccCache.CpuPtr<<std::dec
+		<< "\t0x"<<std::hex<<AccCache.AccPtr<<std::dec<<"\t" <<str
+		<< "\t cpuLock  " << AccCache.cpuLock
+		<< "\t accLock  " << AccCache.accLock
+		<< "\t LRUvalid " << AccCache.LRU_valid<<std::endl;
+    }
+
+    assert( AccCache.cpuLock== 0 ) ;
+    assert( AccCache.accLock== 0 ) ;
+  }
+  std::cout << " Memory Manager::Audit() no locked table entries "<<std::endl;
+  assert(LruBytes1==LruBytes2);
+  assert(LruBytes1==DeviceLRUBytes);
+  std::cout << " Memory Manager::Audit() evictable bytes matches sum over table "<<std::endl;
+  assert(AccBytes==DeviceBytes);
+  std::cout << " Memory Manager::Audit() device bytes matches sum over table "<<std::endl;
+  assert(LruCnt == LRU.size());
+  std::cout << " Memory Manager::Audit() LRU entry count matches "<<std::endl;
+}
 
 void MemoryManager::PrintState(void* _CpuPtr)
 {
@@ -489,8 +583,8 @@ void MemoryManager::PrintState(void* _CpuPtr)
     if ( AccCache.state==EvictNext) str = std::string("EvictNext");
 
     std::cout << GridLogMessage << "CpuAddr\t\tAccAddr\t\tState\t\tcpuLock\taccLock\tLRU_valid "<<std::endl;
-    std::cout << GridLogMessage << "0x"<<std::hex<<AccCache.CpuPtr<<std::dec
+    std::cout << GridLogMessage << "\tx"<<std::hex<<AccCache.CpuPtr<<std::dec
-    << "\t0x"<<std::hex<<AccCache.AccPtr<<std::dec<<"\t" <<str
+    << "\tx"<<std::hex<<AccCache.AccPtr<<std::dec<<"\t" <<str
     << "\t" << AccCache.cpuLock
     << "\t" << AccCache.accLock
     << "\t" << AccCache.LRU_valid<<std::endl;

Grid/allocator/MemoryManagerShared.cc

@@ -12,7 +12,10 @@ uint64_t  MemoryManager::HostToDeviceBytes;
 uint64_t  MemoryManager::DeviceToHostBytes;
 uint64_t  MemoryManager::HostToDeviceXfer;
 uint64_t  MemoryManager::DeviceToHostXfer;
+uint64_t  MemoryManager::DeviceEvictions;
+uint64_t  MemoryManager::DeviceDestroy;
 
+void  MemoryManager::Audit(std::string s){};
 void  MemoryManager::ViewClose(void* AccPtr,ViewMode mode){};
 void *MemoryManager::ViewOpen(void* CpuPtr,size_t bytes,ViewMode mode,ViewAdvise hint){ return CpuPtr; };
 int   MemoryManager::isOpen   (void* CpuPtr) { return 0;}
@@ -21,6 +24,7 @@ void  MemoryManager::PrintState(void* CpuPtr)
 std::cout << GridLogMessage << "Host<->Device memory movement not currently managed by Grid." << std::endl;
 };
 void  MemoryManager::Print(void){};
+void  MemoryManager::PrintAll(void){};
 void  MemoryManager::NotifyDeletion(void *ptr){};
 
 NAMESPACE_END(Grid);

Grid/communicator/Communicator_mpi3.cc

@@ -392,9 +392,9 @@ double CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsReques
     acceleratorCopyDeviceToDeviceAsynch(xmit,shm,bytes);
   }
 
-  if ( CommunicatorPolicy == CommunicatorPolicySequential ) {
+  //  if ( CommunicatorPolicy == CommunicatorPolicySequential ) {
-    this->StencilSendToRecvFromComplete(list,dir);
+  //    this->StencilSendToRecvFromComplete(list,dir);
-  }
+  //  }
 
   return off_node_bytes;
 }

Grid/lattice/Lattice.h

@@ -46,3 +46,4 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 #include <Grid/lattice/Lattice_unary.h>
 #include <Grid/lattice/Lattice_transfer.h>
 #include <Grid/lattice/Lattice_basis.h>
+#include <Grid/lattice/Lattice_crc.h>

Grid/lattice/Lattice_base.h

@@ -129,7 +129,7 @@ public:
     
     auto exprCopy = expr;
     ExpressionViewOpen(exprCopy);
-    auto me  = View(AcceleratorWriteDiscard);
+    auto me  = View(AcceleratorWrite);
     accelerator_for(ss,me.size(),vobj::Nsimd(),{
       auto tmp = eval(ss,exprCopy);
       coalescedWrite(me[ss],tmp);
@@ -152,7 +152,7 @@ public:
 
     auto exprCopy = expr;
     ExpressionViewOpen(exprCopy);
-    auto me  = View(AcceleratorWriteDiscard);
+    auto me  = View(AcceleratorWrite);
     accelerator_for(ss,me.size(),vobj::Nsimd(),{
       auto tmp = eval(ss,exprCopy);
       coalescedWrite(me[ss],tmp);
@@ -174,7 +174,7 @@ public:
     this->checkerboard=cb;
     auto exprCopy = expr;
     ExpressionViewOpen(exprCopy);
-    auto me  = View(AcceleratorWriteDiscard);
+    auto me  = View(AcceleratorWrite);
     accelerator_for(ss,me.size(),vobj::Nsimd(),{
       auto tmp = eval(ss,exprCopy);
       coalescedWrite(me[ss],tmp);
@@ -245,7 +245,7 @@ public:
   ///////////////////////////////////////////
   // user defined constructor
   ///////////////////////////////////////////
-  Lattice(GridBase *grid,ViewMode mode=AcceleratorWriteDiscard) { 
+  Lattice(GridBase *grid,ViewMode mode=AcceleratorWrite) { 
     this->_grid = grid;
     resize(this->_grid->oSites());
     assert((((uint64_t)&this->_odata[0])&0xF) ==0);
@@ -288,7 +288,7 @@ public:
     typename std::enable_if<!std::is_same<robj,vobj>::value,int>::type i=0;
     conformable(*this,r);
     this->checkerboard = r.Checkerboard();
-    auto me =   View(AcceleratorWriteDiscard);
+    auto me =   View(AcceleratorWrite);
     auto him= r.View(AcceleratorRead);
     accelerator_for(ss,me.size(),vobj::Nsimd(),{
       coalescedWrite(me[ss],him(ss));
@@ -303,7 +303,7 @@ public:
   inline Lattice<vobj> & operator = (const Lattice<vobj> & r){
     this->checkerboard = r.Checkerboard();
     conformable(*this,r);
-    auto me =   View(AcceleratorWriteDiscard);
+    auto me =   View(AcceleratorWrite);
     auto him= r.View(AcceleratorRead);
     accelerator_for(ss,me.size(),vobj::Nsimd(),{
       coalescedWrite(me[ss],him(ss));

Grid/lattice/Lattice_crc.h

@@ -0,0 +1,55 @@
+/*************************************************************************************
+
+    Grid physics library, www.github.com/paboyle/Grid 
+
+    Source file: ./lib/lattice/Lattice_crc.h
+
+    Copyright (C) 2021
+
+Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License along
+    with this program; if not, write to the Free Software Foundation, Inc.,
+    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+    See the full license in the file "LICENSE" in the top level distribution directory
+*************************************************************************************/
+/*  END LEGAL */
+#pragma once
+
+NAMESPACE_BEGIN(Grid);
+
+template<class vobj> void DumpSliceNorm(std::string s,Lattice<vobj> &f,int mu=-1)
+{
+  auto ff = localNorm2(f);
+  if ( mu==-1 ) mu = f.Grid()->Nd()-1;
+  typedef typename vobj::tensor_reduced normtype;
+  typedef typename normtype::scalar_object scalar;
+  std::vector<scalar> sff;
+  sliceSum(ff,sff,mu);
+  for(int t=0;t<sff.size();t++){
+    std::cout << s<<" "<<t<<" "<<sff[t]<<std::endl;
+  }
+}
+
+template<class vobj> uint32_t crc(Lattice<vobj> & buf)
+{
+  autoView( buf_v , buf, CpuRead);
+  return ::crc32(0L,(unsigned char *)&buf_v[0],(size_t)sizeof(vobj)*buf.oSites());
+}
+
+#define CRC(U) std::cout << "FingerPrint "<<__FILE__ <<" "<< __LINE__ <<" "<< #U <<" "<<crc(U)<<std::endl;
+
+NAMESPACE_END(Grid);
+
+

Grid/lattice/Lattice_reduction.h

@@ -28,6 +28,9 @@ Author: Christoph Lehner <christoph@lhnr.de>
 #if defined(GRID_CUDA)||defined(GRID_HIP)
 #include <Grid/lattice/Lattice_reduction_gpu.h>
 #endif
+#if defined(GRID_SYCL)
+#include <Grid/lattice/Lattice_reduction_sycl.h>
+#endif
 
 NAMESPACE_BEGIN(Grid);
 
@@ -127,7 +130,7 @@ inline Double max(const Double *arg, Integer osites)
 template<class vobj>
 inline typename vobj::scalar_object sum(const vobj *arg, Integer osites)
 {
-#if defined(GRID_CUDA)||defined(GRID_HIP)
+#if defined(GRID_CUDA)||defined(GRID_HIP)||defined(GRID_SYCL)
   return sum_gpu(arg,osites);
 #else
   return sum_cpu(arg,osites);
@@ -136,23 +139,31 @@ inline typename vobj::scalar_object sum(const vobj *arg, Integer osites)
 template<class vobj>
 inline typename vobj::scalar_objectD sumD(const vobj *arg, Integer osites)
 {
-#if defined(GRID_CUDA)||defined(GRID_HIP)
+#if defined(GRID_CUDA)||defined(GRID_HIP)||defined(GRID_SYCL)
   return sumD_gpu(arg,osites);
 #else
   return sumD_cpu(arg,osites);
 #endif  
 }
+template<class vobj>
+inline typename vobj::scalar_objectD sumD_large(const vobj *arg, Integer osites)
+{
+#if defined(GRID_CUDA)||defined(GRID_HIP)||defined(GRID_SYCL)
+  return sumD_gpu_large(arg,osites);
+#else
+  return sumD_cpu(arg,osites);
+#endif  
+}
 
 template<class vobj>
 inline typename vobj::scalar_object rankSum(const Lattice<vobj> &arg)
 {
-#if defined(GRID_CUDA)||defined(GRID_HIP)
-  autoView( arg_v, arg, AcceleratorRead);
   Integer osites = arg.Grid()->oSites();
+#if defined(GRID_CUDA)||defined(GRID_HIP)||defined(GRID_SYCL)
+  autoView( arg_v, arg, AcceleratorRead);
   return sum_gpu(&arg_v[0],osites);
 #else
   autoView(arg_v, arg, CpuRead);
-  Integer osites = arg.Grid()->oSites();
   return sum_cpu(&arg_v[0],osites);
 #endif  
 }
@@ -165,6 +176,28 @@ inline typename vobj::scalar_object sum(const Lattice<vobj> &arg)
   return ssum;
 }
 
+template<class vobj>
+inline typename vobj::scalar_object rankSumLarge(const Lattice<vobj> &arg)
+{
+#if defined(GRID_CUDA)||defined(GRID_HIP)||defined(GRID_SYCL)
+  autoView( arg_v, arg, AcceleratorRead);
+  Integer osites = arg.Grid()->oSites();
+  return sum_gpu_large(&arg_v[0],osites);
+#else
+  autoView(arg_v, arg, CpuRead);
+  Integer osites = arg.Grid()->oSites();
+  return sum_cpu(&arg_v[0],osites);
+#endif
+}
+
+template<class vobj>
+inline typename vobj::scalar_object sum_large(const Lattice<vobj> &arg)
+{
+  auto ssum = rankSumLarge(arg);
+  arg.Grid()->GlobalSum(ssum);
+  return ssum;
+}
+
 ////////////////////////////////////////////////////////////////////////////////////////////////////
 // Deterministic Reduction operations
 ////////////////////////////////////////////////////////////////////////////////////////////////////
@@ -216,11 +249,10 @@ inline ComplexD rankInnerProduct(const Lattice<vobj> &left,const Lattice<vobj> &
   typedef decltype(innerProductD(vobj(),vobj())) inner_t;
   Vector<inner_t> inner_tmp(sites);
   auto inner_tmp_v = &inner_tmp[0];
-    
   {
     autoView( left_v , left, AcceleratorRead);
     autoView( right_v,right, AcceleratorRead);
-
+    // This code could read coalesce
     // GPU - SIMT lane compliance...
     accelerator_for( ss, sites, 1,{
 	auto x_l = left_v[ss];

Grid/lattice/Lattice_reduction_gpu.h

@@ -23,7 +23,7 @@ unsigned int nextPow2(Iterator x) {
 }
 
 template <class Iterator>
-void getNumBlocksAndThreads(const Iterator n, const size_t sizeofsobj, Iterator &threads, Iterator &blocks) {
+int getNumBlocksAndThreads(const Iterator n, const size_t sizeofsobj, Iterator &threads, Iterator &blocks) {
   
   int device;
 #ifdef GRID_CUDA
@@ -37,13 +37,13 @@ void getNumBlocksAndThreads(const Iterator n, const size_t sizeofsobj, Iterator
   Iterator sharedMemPerBlock   = gpu_props[device].sharedMemPerBlock;
   Iterator maxThreadsPerBlock  = gpu_props[device].maxThreadsPerBlock;
   Iterator multiProcessorCount = gpu_props[device].multiProcessorCount;
-  
+  /*  
   std::cout << GridLogDebug << "GPU has:" << std::endl;
   std::cout << GridLogDebug << "\twarpSize            = " << warpSize << std::endl;
   std::cout << GridLogDebug << "\tsharedMemPerBlock   = " << sharedMemPerBlock << std::endl;
   std::cout << GridLogDebug << "\tmaxThreadsPerBlock  = " << maxThreadsPerBlock << std::endl;
   std::cout << GridLogDebug << "\tmultiProcessorCount = " << multiProcessorCount << std::endl;
-  
+  */  
   if (warpSize != WARP_SIZE) {
     std::cout << GridLogError << "The warp size of the GPU in use does not match the warp size set when compiling Grid." << std::endl;
     exit(EXIT_FAILURE);
@@ -53,12 +53,12 @@ void getNumBlocksAndThreads(const Iterator n, const size_t sizeofsobj, Iterator
   threads = warpSize;
   if ( threads*sizeofsobj > sharedMemPerBlock ) {
     std::cout << GridLogError << "The object is too large for the shared memory." << std::endl;
-    exit(EXIT_FAILURE);
+    return 0;
   }
   while( 2*threads*sizeofsobj < sharedMemPerBlock && 2*threads <= maxThreadsPerBlock ) threads *= 2;
   // keep all the streaming multiprocessors busy
   blocks = nextPow2(multiProcessorCount);
-  
+  return 1;
 }
 
 template <class sobj, class Iterator>
@@ -198,7 +198,7 @@ __global__ void reduceKernel(const vobj *lat, sobj *buffer, Iterator n) {
 // Possibly promote to double and sum
 /////////////////////////////////////////////////////////////////////////////////////////////////////////
 template <class vobj>
-inline typename vobj::scalar_objectD sumD_gpu(const vobj *lat, Integer osites) 
+inline typename vobj::scalar_objectD sumD_gpu_small(const vobj *lat, Integer osites) 
 {
   typedef typename vobj::scalar_objectD sobj;
   typedef decltype(lat) Iterator;
@@ -207,7 +207,9 @@ inline typename vobj::scalar_objectD sumD_gpu(const vobj *lat, Integer osites)
   Integer size = osites*nsimd;
 
   Integer numThreads, numBlocks;
-  getNumBlocksAndThreads(size, sizeof(sobj), numThreads, numBlocks);
+  int ok = getNumBlocksAndThreads(size, sizeof(sobj), numThreads, numBlocks);
+  assert(ok);
+
   Integer smemSize = numThreads * sizeof(sobj);
 
   Vector<sobj> buffer(numBlocks);
@@ -218,6 +220,54 @@ inline typename vobj::scalar_objectD sumD_gpu(const vobj *lat, Integer osites)
   auto result = buffer_v[0];
   return result;
 }
+
+template <class vobj>
+inline typename vobj::scalar_objectD sumD_gpu_large(const vobj *lat, Integer osites)
+{
+  typedef typename vobj::vector_type  vector;
+  typedef typename vobj::scalar_typeD scalarD;
+  typedef typename vobj::scalar_objectD sobj;
+  sobj ret;
+  scalarD *ret_p = (scalarD *)&ret;
+  
+  const int words = sizeof(vobj)/sizeof(vector);
+
+  Vector<vector> buffer(osites);
+  vector *dat = (vector *)lat;
+  vector *buf = &buffer[0];
+  iScalar<vector> *tbuf =(iScalar<vector> *)  &buffer[0];
+  for(int w=0;w<words;w++) {
+
+    accelerator_for(ss,osites,1,{
+	buf[ss] = dat[ss*words+w];
+      });
+      
+    ret_p[w] = sumD_gpu_small(tbuf,osites);
+  }
+  return ret;
+}
+
+template <class vobj>
+inline typename vobj::scalar_objectD sumD_gpu(const vobj *lat, Integer osites)
+{
+  typedef typename vobj::vector_type  vector;
+  typedef typename vobj::scalar_typeD scalarD;
+  typedef typename vobj::scalar_objectD sobj;
+  sobj ret;
+  
+  Integer nsimd= vobj::Nsimd();
+  Integer size = osites*nsimd;
+  Integer numThreads, numBlocks;
+  int ok = getNumBlocksAndThreads(size, sizeof(sobj), numThreads, numBlocks);
+  
+  if ( ok ) {
+    ret = sumD_gpu_small(lat,osites);
+  } else {
+    ret = sumD_gpu_large(lat,osites);
+  }
+  return ret;
+}
+
 /////////////////////////////////////////////////////////////////////////////////////////////////////////
 // Return as same precision as input performing reduction in double precision though
 /////////////////////////////////////////////////////////////////////////////////////////////////////////
@@ -230,6 +280,13 @@ inline typename vobj::scalar_object sum_gpu(const vobj *lat, Integer osites)
   return result;
 }
 
-
+template <class vobj>
+inline typename vobj::scalar_object sum_gpu_large(const vobj *lat, Integer osites)
+{
+  typedef typename vobj::scalar_object sobj;
+  sobj result;
+  result = sumD_gpu_large(lat,osites);
+  return result;
+}
 
 NAMESPACE_END(Grid);

Grid/lattice/Lattice_reduction_sycl.h

@@ -0,0 +1,125 @@
+NAMESPACE_BEGIN(Grid);
+
+/////////////////////////////////////////////////////////////////////////////////////////////////////////
+// Possibly promote to double and sum
+/////////////////////////////////////////////////////////////////////////////////////////////////////////
+
+template <class vobj>
+inline typename vobj::scalar_objectD sumD_gpu_tensor(const vobj *lat, Integer osites) 
+{
+  typedef typename vobj::scalar_object sobj;
+  typedef typename vobj::scalar_objectD sobjD;
+  sobj *mysum =(sobj *) malloc_shared(sizeof(sobj),*theGridAccelerator);
+  sobj identity; zeroit(identity);
+  sobj ret ; 
+
+  Integer nsimd= vobj::Nsimd();
+  
+  theGridAccelerator->submit([&](cl::sycl::handler &cgh) {
+     auto Reduction = cl::sycl::reduction(mysum,identity,std::plus<>());
+     cgh.parallel_for(cl::sycl::range<1>{osites},
+		      Reduction,
+		      [=] (cl::sycl::id<1> item, auto &sum) {
+      auto osite   = item[0];
+      sum +=Reduce(lat[osite]);
+     });
+   });
+  theGridAccelerator->wait();
+  ret = mysum[0];
+  free(mysum,*theGridAccelerator);
+  sobjD dret; convertType(dret,ret);
+  return dret;
+}
+
+template <class vobj>
+inline typename vobj::scalar_objectD sumD_gpu_large(const vobj *lat, Integer osites)
+{
+  return sumD_gpu_tensor(lat,osites);
+}
+template <class vobj>
+inline typename vobj::scalar_objectD sumD_gpu_small(const vobj *lat, Integer osites)
+{
+  return sumD_gpu_large(lat,osites);
+}
+
+template <class vobj>
+inline typename vobj::scalar_objectD sumD_gpu(const vobj *lat, Integer osites)
+{
+  return sumD_gpu_large(lat,osites);
+}
+
+/////////////////////////////////////////////////////////////////////////////////////////////////////////
+// Return as same precision as input performing reduction in double precision though
+/////////////////////////////////////////////////////////////////////////////////////////////////////////
+template <class vobj>
+inline typename vobj::scalar_object sum_gpu(const vobj *lat, Integer osites) 
+{
+  typedef typename vobj::scalar_object sobj;
+  sobj result;
+  result = sumD_gpu(lat,osites);
+  return result;
+}
+
+template <class vobj>
+inline typename vobj::scalar_object sum_gpu_large(const vobj *lat, Integer osites)
+{
+  typedef typename vobj::scalar_object sobj;
+  sobj result;
+  result = sumD_gpu_large(lat,osites);
+  return result;
+}
+
+NAMESPACE_END(Grid);
+
+/*
+template<class Double> Double svm_reduce(Double *vec,uint64_t L)
+{
+  Double sumResult; zeroit(sumResult);
+  Double *d_sum =(Double *)cl::sycl::malloc_shared(sizeof(Double),*theGridAccelerator);
+  Double identity;  zeroit(identity);
+  theGridAccelerator->submit([&](cl::sycl::handler &cgh) {
+     auto Reduction = cl::sycl::reduction(d_sum,identity,std::plus<>());
+     cgh.parallel_for(cl::sycl::range<1>{L},
+		      Reduction,
+		      [=] (cl::sycl::id<1> index, auto &sum) {
+	 sum +=vec[index];
+     });
+   });
+  theGridAccelerator->wait();
+  Double ret = d_sum[0];
+  free(d_sum,*theGridAccelerator);
+  std::cout << " svm_reduce finished "<<L<<" sites sum = " << ret <<std::endl;
+  return ret;
+}
+
+template <class vobj>
+inline typename vobj::scalar_objectD sumD_gpu_repack(const vobj *lat, Integer osites)
+{
+  typedef typename vobj::vector_type  vector;
+  typedef typename vobj::scalar_type  scalar;
+
+  typedef typename vobj::scalar_typeD scalarD;
+  typedef typename vobj::scalar_objectD sobjD;
+
+  sobjD ret;
+  scalarD *ret_p = (scalarD *)&ret;
+  
+  const int nsimd = vobj::Nsimd();
+  const int words = sizeof(vobj)/sizeof(vector);
+
+  Vector<scalar> buffer(osites*nsimd);
+  scalar *buf = &buffer[0];
+  vector *dat = (vector *)lat;
+
+  for(int w=0;w<words;w++) {
+
+    accelerator_for(ss,osites,nsimd,{
+	int lane = acceleratorSIMTlane(nsimd);
+	buf[ss*nsimd+lane] = dat[ss*words+w].getlane(lane);
+    });
+    //Precision change at this point is to late to gain precision
+    ret_p[w] = svm_reduce(buf,nsimd*osites);
+  }
+  return ret;
+}
+*/

Grid/log/Log.cc

@@ -65,29 +65,40 @@ GridLogger GridLogSolver (1, "Solver", GridLogColours, "NORMAL");
 GridLogger GridLogError  (1, "Error" , GridLogColours, "RED");
 GridLogger GridLogWarning(1, "Warning", GridLogColours, "YELLOW");
 GridLogger GridLogMessage(1, "Message", GridLogColours, "NORMAL");
+GridLogger GridLogMemory (1, "Memory", GridLogColours, "NORMAL");
+GridLogger GridLogTracing(1, "Tracing", GridLogColours, "NORMAL");
 GridLogger GridLogDebug  (1, "Debug", GridLogColours, "PURPLE");
 GridLogger GridLogPerformance(1, "Performance", GridLogColours, "GREEN");
+GridLogger GridLogDslash     (1, "Dslash", GridLogColours, "BLUE");
 GridLogger GridLogIterative  (1, "Iterative", GridLogColours, "BLUE");
 GridLogger GridLogIntegrator (1, "Integrator", GridLogColours, "BLUE");
+GridLogger GridLogHMC (1, "HMC", GridLogColours, "BLUE");
 
 void GridLogConfigure(std::vector<std::string> &logstreams) {
-  GridLogError.Active(0);
+  GridLogError.Active(1);
   GridLogWarning.Active(0);
   GridLogMessage.Active(1); // at least the messages should be always on
+  GridLogMemory.Active(0); 
+  GridLogTracing.Active(0); 
   GridLogIterative.Active(0);
   GridLogDebug.Active(0);
   GridLogPerformance.Active(0);
+  GridLogDslash.Active(0);
   GridLogIntegrator.Active(1);
   GridLogColours.Active(0);
+  GridLogHMC.Active(1);
 
   for (int i = 0; i < logstreams.size(); i++) {
-    if (logstreams[i] == std::string("Error"))       GridLogError.Active(1);
+    if (logstreams[i] == std::string("Tracing"))     GridLogTracing.Active(1);
+    if (logstreams[i] == std::string("Memory"))      GridLogMemory.Active(1);
     if (logstreams[i] == std::string("Warning"))     GridLogWarning.Active(1);
     if (logstreams[i] == std::string("NoMessage"))   GridLogMessage.Active(0);
     if (logstreams[i] == std::string("Iterative"))   GridLogIterative.Active(1);
     if (logstreams[i] == std::string("Debug"))       GridLogDebug.Active(1);
     if (logstreams[i] == std::string("Performance")) GridLogPerformance.Active(1);
-    if (logstreams[i] == std::string("Integrator"))  GridLogIntegrator.Active(1);
+    if (logstreams[i] == std::string("Dslash"))      GridLogDslash.Active(1);
+    if (logstreams[i] == std::string("NoIntegrator"))GridLogIntegrator.Active(0);
+    if (logstreams[i] == std::string("NoHMC"))       GridLogHMC.Active(0);
     if (logstreams[i] == std::string("Colours"))     GridLogColours.Active(1);
   }
 }

Grid/log/Log.h

@@ -138,7 +138,8 @@ public:
         stream << std::setw(log.topWidth);
       }
       stream << log.topName << log.background()<< " : ";
-      stream << log.colour() <<  std::left;
+      //      stream << log.colour() <<  std::left;
+      stream <<  std::left;
       if (log.chanWidth > 0)
       {
         stream << std::setw(log.chanWidth);
@@ -153,9 +154,9 @@ public:
 	stream << log.evidence()
 	       << now	       << log.background() << " : " ;
       }
-      stream << log.colour();
+      //      stream << log.colour();
+      stream <<  std::right;
       stream.flags(f);
-
       return stream;
     } else { 
       return devnull;
@@ -180,8 +181,12 @@ extern GridLogger GridLogWarning;
 extern GridLogger GridLogMessage;
 extern GridLogger GridLogDebug  ;
 extern GridLogger GridLogPerformance;
+extern GridLogger GridLogDslash;
 extern GridLogger GridLogIterative  ;
 extern GridLogger GridLogIntegrator  ;
+extern GridLogger GridLogHMC;
+extern GridLogger GridLogMemory;
+extern GridLogger GridLogTracing;
 extern Colours    GridLogColours;
 
 std::string demangle(const char* name) ;

Grid/parallelIO/IldgIO.h

@@ -31,6 +31,7 @@ directory
 #include <fstream>
 #include <iomanip>
 #include <iostream>
+#include <string>
 #include <map>
 
 #include <pwd.h>
@@ -654,7 +655,8 @@ class IldgWriter : public ScidacWriter {
     // Fill ILDG header data struct
     //////////////////////////////////////////////////////
     ildgFormat ildgfmt ;
-    ildgfmt.field     = std::string("su3gauge");
+    const std::string stNC = std::to_string( Nc ) ;
+    ildgfmt.field          = std::string("su"+stNC+"gauge");
 
     if ( format == std::string("IEEE32BIG") ) { 
       ildgfmt.precision = 32;
@@ -871,7 +873,8 @@ class IldgReader : public GridLimeReader {
     } else { 
 
       assert(found_ildgFormat);
-      assert ( ildgFormat_.field == std::string("su3gauge") );
+      const std::string stNC = std::to_string( Nc ) ;
+      assert ( ildgFormat_.field == std::string("su"+stNC+"gauge") );
 
       ///////////////////////////////////////////////////////////////////////////////////////
       // Populate our Grid metadata as best we can
@@ -879,7 +882,7 @@ class IldgReader : public GridLimeReader {
 
       std::ostringstream vers; vers << ildgFormat_.version;
       FieldMetaData_.hdr_version = vers.str();
-      FieldMetaData_.data_type = std::string("4D_SU3_GAUGE_3X3");
+      FieldMetaData_.data_type = std::string("4D_SU"+stNC+"_GAUGE_"+stNC+"x"+stNC);
 
       FieldMetaData_.nd=4;
       FieldMetaData_.dimension.resize(4);

Grid/parallelIO/MetaData.h

@@ -6,8 +6,8 @@
 
     Copyright (C) 2015
 
-
     Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+    Author: Jamie Hudspith <renwick.james.hudspth@gmail.com>
 
     This program is free software; you can redistribute it and/or modify
     it under the terms of the GNU General Public License as published by
@@ -182,8 +182,8 @@ class GaugeStatistics
 public:
   void operator()(Lattice<vLorentzColourMatrixD> & data,FieldMetaData &header)
   {
-    header.link_trace=WilsonLoops<Impl>::linkTrace(data);
+    header.link_trace = WilsonLoops<Impl>::linkTrace(data);
-    header.plaquette =WilsonLoops<Impl>::avgPlaquette(data);
+    header.plaquette  = WilsonLoops<Impl>::avgPlaquette(data);
   }
 };
 typedef GaugeStatistics<PeriodicGimplD> PeriodicGaugeStatistics;
@@ -203,20 +203,24 @@ template<> inline void PrepareMetaData<vLorentzColourMatrixD>(Lattice<vLorentzCo
 //////////////////////////////////////////////////////////////////////
 inline void reconstruct3(LorentzColourMatrix & cm)
 {
-  const int x=0;
+  assert( Nc < 4 && Nc > 1 ) ;
-  const int y=1;
-  const int z=2;
   for(int mu=0;mu<Nd;mu++){
+    #if Nc == 2
+      cm(mu)()(1,0) = -adj(cm(mu)()(0,y)) ;
+      cm(mu)()(1,1) =  adj(cm(mu)()(0,x)) ;
+    #else
+      const int x=0 , y=1 , z=2 ; // a little disinenuous labelling
       cm(mu)()(2,x) = adj(cm(mu)()(0,y)*cm(mu)()(1,z)-cm(mu)()(0,z)*cm(mu)()(1,y)); //x= yz-zy
       cm(mu)()(2,y) = adj(cm(mu)()(0,z)*cm(mu)()(1,x)-cm(mu)()(0,x)*cm(mu)()(1,z)); //y= zx-xz
       cm(mu)()(2,z) = adj(cm(mu)()(0,x)*cm(mu)()(1,y)-cm(mu)()(0,y)*cm(mu)()(1,x)); //z= xy-yx
+    #endif
   }
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // Some data types for intermediate storage
 ////////////////////////////////////////////////////////////////////////////////
-template<typename vtype> using iLorentzColour2x3 = iVector<iVector<iVector<vtype, Nc>, 2>, Nd >;
+template<typename vtype> using iLorentzColour2x3 = iVector<iVector<iVector<vtype, Nc>, Nc-1>, Nd >;
 
 typedef iLorentzColour2x3<Complex>  LorentzColour2x3;
 typedef iLorentzColour2x3<ComplexF> LorentzColour2x3F;
@@ -278,7 +282,6 @@ struct GaugeSimpleMunger{
 
 template <class fobj, class sobj>
 struct GaugeSimpleUnmunger {
-
   void operator()(sobj &in, fobj &out) {
     for (int mu = 0; mu < Nd; mu++) {
       for (int i = 0; i < Nc; i++) {
@@ -317,8 +320,8 @@ template<class fobj,class sobj>
 struct Gauge3x2munger{
   void operator() (fobj &in,sobj &out){
     for(int mu=0;mu<Nd;mu++){
-      for(int i=0;i<2;i++){
+      for(int i=0;i<Nc-1;i++){
-	for(int j=0;j<3;j++){
+	for(int j=0;j<Nc;j++){
 	  out(mu)()(i,j) = in(mu)(i)(j);
 	}}
     }
@@ -330,8 +333,8 @@ template<class fobj,class sobj>
 struct Gauge3x2unmunger{
   void operator() (sobj &in,fobj &out){
     for(int mu=0;mu<Nd;mu++){
-      for(int i=0;i<2;i++){
+      for(int i=0;i<Nc-1;i++){
-	for(int j=0;j<3;j++){
+	for(int j=0;j<Nc;j++){
 	  out(mu)(i)(j) = in(mu)()(i,j);
 	}}
     }

Grid/parallelIO/NerscIO.h

@@ -9,6 +9,7 @@
     Author: Matt Spraggs <matthew.spraggs@gmail.com>
     Author: Peter Boyle <paboyle@ph.ed.ac.uk>
     Author: paboyle <paboyle@ph.ed.ac.uk>
+    Author: Jamie Hudspith <renwick.james.hudspth@gmail.com>
 
     This program is free software; you can redistribute it and/or modify
     it under the terms of the GNU General Public License as published by
@@ -30,6 +31,8 @@
 #ifndef GRID_NERSC_IO_H
 #define GRID_NERSC_IO_H
 
+#include <string>
+
 NAMESPACE_BEGIN(Grid);
 
 using namespace Grid;
@@ -145,15 +148,17 @@ public:
 
     std::string format(header.floating_point);
 
-    int ieee32big = (format == std::string("IEEE32BIG"));
+    const int ieee32big = (format == std::string("IEEE32BIG"));
-    int ieee32    = (format == std::string("IEEE32"));
+    const int ieee32    = (format == std::string("IEEE32"));
-    int ieee64big = (format == std::string("IEEE64BIG"));
+    const int ieee64big = (format == std::string("IEEE64BIG"));
-    int ieee64    = (format == std::string("IEEE64") || format == std::string("IEEE64LITTLE"));
+    const int ieee64    = (format == std::string("IEEE64") || \
+			   format == std::string("IEEE64LITTLE"));
 
     uint32_t nersc_csum,scidac_csuma,scidac_csumb;
     // depending on datatype, set up munger;
     // munger is a function of <floating point, Real, data_type>
-    if ( header.data_type == std::string("4D_SU3_GAUGE") ) {
+    const std::string stNC = std::to_string( Nc ) ;
+    if ( header.data_type == std::string("4D_SU"+stNC+"_GAUGE") ) {
       if ( ieee32 || ieee32big ) {
 	BinaryIO::readLatticeObject<vLorentzColourMatrixD, LorentzColour2x3F> 
 	  (Umu,file,Gauge3x2munger<LorentzColour2x3F,LorentzColourMatrix>(), offset,format,
@@ -164,7 +169,7 @@ public:
 	  (Umu,file,Gauge3x2munger<LorentzColour2x3D,LorentzColourMatrix>(),offset,format,
 	   nersc_csum,scidac_csuma,scidac_csumb);
       }
-    } else if ( header.data_type == std::string("4D_SU3_GAUGE_3x3") ) {
+    } else if ( header.data_type == std::string("4D_SU"+stNC+"_GAUGE_"+stNC+"x"+stNC) ) {
       if ( ieee32 || ieee32big ) {
 	BinaryIO::readLatticeObject<vLorentzColourMatrixD,LorentzColourMatrixF>
 	  (Umu,file,GaugeSimpleMunger<LorentzColourMatrixF,LorentzColourMatrix>(),offset,format,
@@ -231,6 +236,7 @@ public:
     header.sequence_number = sequence_number;
     header.ensemble_id     = ens_id;
     header.ensemble_label  = ens_label;
+    header.hdr_version     = "1.0" ;
 
     typedef LorentzColourMatrixD fobj3D;
     typedef LorentzColour2x3D    fobj2D;
@@ -244,10 +250,14 @@ public:
 
     uint64_t offset;
 
-    // Sod it -- always write 3x3 double
+    // Sod it -- always write NcxNc double
     header.floating_point  = std::string("IEEE64BIG");
-    header.data_type      = std::string("4D_SU3_GAUGE_3x3");
+    const std::string stNC = std::to_string( Nc ) ;
-    GaugeSimpleUnmunger<fobj3D,sobj> munge;
+    if( two_row ) {
+      header.data_type = std::string("4D_SU" + stNC + "_GAUGE" );
+    } else {
+      header.data_type = std::string("4D_SU" + stNC + "_GAUGE_" + stNC + "x" + stNC );
+    }
     if ( grid->IsBoss() ) { 
       truncate(file);
       offset = writeHeader(header,file);
@@ -255,8 +265,15 @@ public:
     grid->Broadcast(0,(void *)&offset,sizeof(offset));
 
     uint32_t nersc_csum,scidac_csuma,scidac_csumb;
+    if( two_row ) {
+      Gauge3x2unmunger<fobj2D,sobj> munge;
+      BinaryIO::writeLatticeObject<vobj,fobj2D>(Umu,file,munge,offset,header.floating_point,
+						nersc_csum,scidac_csuma,scidac_csumb);
+    } else {
+      GaugeSimpleUnmunger<fobj3D,sobj> munge;
       BinaryIO::writeLatticeObject<vobj,fobj3D>(Umu,file,munge,offset,header.floating_point,
 						nersc_csum,scidac_csuma,scidac_csumb);
+    }
     header.checksum = nersc_csum;
     if ( grid->IsBoss() ) { 
       writeHeader(header,file);
@@ -289,7 +306,6 @@ public:
     MachineCharacteristics(header);
 
     uint64_t offset;
-  
 #ifdef RNG_RANLUX
     header.floating_point = std::string("UINT64");
     header.data_type      = std::string("RANLUX48");

Grid/perfmon/PerfCount.cc

@@ -27,10 +27,13 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 /*  END LEGAL */
 
 #include <Grid/GridCore.h>
-#include <Grid/perfmon/PerfCount.h>
 
+#include <Grid/perfmon/Timer.h>
+#include <Grid/perfmon/PerfCount.h>
 NAMESPACE_BEGIN(Grid);
 
+GridTimePoint theProgramStart = GridClock::now();
+
 #define CacheControl(L,O,R) ((PERF_COUNT_HW_CACHE_##L)|(PERF_COUNT_HW_CACHE_OP_##O<<8)| (PERF_COUNT_HW_CACHE_RESULT_##R<<16))
 #define RawConfig(A,B) (A<<8|B)
 const PerformanceCounter::PerformanceCounterConfig PerformanceCounter::PerformanceCounterConfigs [] = {

Grid/perfmon/PerfCount.h

@@ -30,6 +30,12 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 #ifndef GRID_PERFCOUNT_H
 #define GRID_PERFCOUNT_H
 
+
+#ifndef __SSC_START
+#define __SSC_START
+#define __SSC_STOP
+#endif
+
 #include <sys/time.h>
 #include <ctime>
 #include <chrono>
@@ -72,17 +78,9 @@ static long perf_event_open(struct perf_event_attr *hw_event, pid_t pid,
 inline uint64_t cyclecount(void){ 
   return 0;
 }
-#define __SSC_MARK(mark) __asm__ __volatile__ ("movl %0, %%ebx; .byte 0x64, 0x67, 0x90 " ::"i"(mark):"%ebx")
-#define __SSC_STOP  __SSC_MARK(0x110)
-#define __SSC_START __SSC_MARK(0x111)
-
 
 #else
 
-#define __SSC_MARK(mark) 
-#define __SSC_STOP  
-#define __SSC_START 
-
 /*
  * cycle counters arch dependent
  */

Grid/perfmon/Timer.h

@@ -35,17 +35,8 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 
 NAMESPACE_BEGIN(Grid)
 
-// Dress the output; use std::chrono
+//typedef  std::chrono::system_clock          GridClock;
-// C++11 time facilities better?
+typedef  std::chrono::high_resolution_clock   GridClock;
-inline double usecond(void) {
-  struct timeval tv;
-#ifdef TIMERS_ON
-  gettimeofday(&tv,NULL);
-#endif
-  return 1.0*tv.tv_usec + 1.0e6*tv.tv_sec;
-}
-
-typedef  std::chrono::system_clock          GridClock;
 typedef  std::chrono::time_point<GridClock> GridTimePoint;
 
 typedef  std::chrono::seconds               GridSecs;
@@ -53,6 +44,15 @@ typedef  std::chrono::milliseconds          GridMillisecs;
 typedef  std::chrono::microseconds          GridUsecs;
 typedef  std::chrono::microseconds          GridTime;
 
+extern GridTimePoint theProgramStart;
+// Dress the output; use std::chrono
+// C++11 time facilities better?
+inline double usecond(void) {
+  auto usecs = std::chrono::duration_cast<GridUsecs>(GridClock::now()-theProgramStart); 
+  return 1.0*usecs.count();
+}
+
+
 inline std::ostream& operator<< (std::ostream & stream, const GridSecs & time)
 {
   stream << time.count()<<" s";

Grid/perfmon/Tracing.h

@@ -0,0 +1,70 @@
+#pragma once
+
+NAMESPACE_BEGIN(Grid);
+
+#ifdef GRID_TRACING_NVTX
+#include <nvToolsExt.h>
+class GridTracer {
+public:
+  GridTracer(const char* name) {
+    nvtxRangePushA(name);
+  }
+  ~GridTracer() {
+    nvtxRangePop();
+  }
+};
+inline void tracePush(const char *name) { nvtxRangePushA(name); }
+inline void tracePop(const char *name) { nvtxRangePop(); }
+inline int  traceStart(const char *name) {  }
+inline void traceStop(int ID) {  }
+#endif
+
+#ifdef GRID_TRACING_ROCTX
+#include <roctracer/roctx.h>
+class GridTracer {
+ public:
+  GridTracer(const char* name) {
+    roctxRangePushA(name);
+    std::cout << "roctxRangePush "<<name<<std::endl;
+  }
+  ~GridTracer() {
+    roctxRangePop();
+    std::cout << "roctxRangePop "<<std::endl;
+  }
+};
+inline void tracePush(const char *name) { roctxRangePushA(name); }
+inline void tracePop(const char *name) { roctxRangePop(); }
+inline int  traceStart(const char *name) { roctxRangeStart(name); }
+inline void traceStop(int ID) { roctxRangeStop(ID); }
+#endif
+
+#ifdef GRID_TRACING_TIMER
+class GridTracer {
+ public:
+  const char *name;
+  double elapsed;
+  GridTracer(const char* _name) {
+    name = _name;
+    elapsed=-usecond();
+  }
+  ~GridTracer() {
+    elapsed+=usecond();
+    std::cout << GridLogTracing << name << " took " <<elapsed<< " us" <<std::endl;
+  }
+};
+inline void tracePush(const char *name) {  }
+inline void tracePop(const char *name) {  }
+inline int  traceStart(const char *name) { return 0; }
+inline void traceStop(int ID) {  }
+#endif
+
+#ifdef GRID_TRACING_NONE
+#define GRID_TRACE(name) 
+inline void tracePush(const char *name) {  }
+inline void tracePop(const char *name) {  }
+inline int  traceStart(const char *name) { return 0;  }
+inline void traceStop(int ID) {  }
+#else
+#define GRID_TRACE(name) GridTracer uniq_name_using_macros##__COUNTER__(name);
+#endif
+NAMESPACE_END(Grid);

Grid/pugixml/pugixml.cc

@@ -16,8 +16,12 @@
 
 #ifdef __NVCC__
 #pragma push
+#if (__CUDACC_VER_MAJOR__ >= 11) && (__CUDACC_VER_MINOR__ >= 5)
+#pragma nv_diag_suppress declared_but_not_referenced // suppress "function was declared but never referenced warning"
+#else
 #pragma diag_suppress declared_but_not_referenced // suppress "function was declared but never referenced warning"
 #endif
+#endif
 
 #include "pugixml.h"
 

Grid/qcd/QCD.h

@@ -451,9 +451,20 @@ template<class vobj> void pokeLorentz(vobj &lhs,const decltype(peekIndex<Lorentz
 // Fermion <-> propagator assignements
 //////////////////////////////////////////////
 //template <class Prop, class Ferm>
+#define FAST_FERM_TO_PROP
 template <class Fimpl>
 void FermToProp(typename Fimpl::PropagatorField &p, const typename Fimpl::FermionField &f, const int s, const int c)
 {
+#ifdef FAST_FERM_TO_PROP
+  autoView(p_v,p,CpuWrite);
+  autoView(f_v,f,CpuRead);
+  thread_for(idx,p_v.oSites(),{
+      for(int ss = 0; ss < Ns; ++ss) {
+      for(int cc = 0; cc < Fimpl::Dimension; ++cc) {
+	p_v[idx]()(ss,s)(cc,c) = f_v[idx]()(ss)(cc); // Propagator sink index is LEFT, suitable for left mult by gauge link (e.g.)
+      }}
+    });
+#else
   for(int j = 0; j < Ns; ++j)
     {
       auto pjs = peekSpin(p, j, s);
@@ -465,12 +476,23 @@ void FermToProp(typename Fimpl::PropagatorField &p, const typename Fimpl::Fermio
 	}
       pokeSpin(p, pjs, j, s);
     }
+#endif
 }
     
 //template <class Prop, class Ferm>
 template <class Fimpl>
 void PropToFerm(typename Fimpl::FermionField &f, const typename Fimpl::PropagatorField &p, const int s, const int c)
 {
+#ifdef FAST_FERM_TO_PROP
+  autoView(p_v,p,CpuRead);
+  autoView(f_v,f,CpuWrite);
+  thread_for(idx,p_v.oSites(),{
+      for(int ss = 0; ss < Ns; ++ss) {
+      for(int cc = 0; cc < Fimpl::Dimension; ++cc) {
+	f_v[idx]()(ss)(cc) = p_v[idx]()(ss,s)(cc,c); // LEFT index is copied across for s,c right index
+      }}
+    });
+#else
   for(int j = 0; j < Ns; ++j)
     {
       auto pjs = peekSpin(p, j, s);
@@ -482,6 +504,7 @@ void PropToFerm(typename Fimpl::FermionField &f, const typename Fimpl::Propagato
 	}
       pokeSpin(f, fj, j);
     }
+#endif
 }
     
 //////////////////////////////////////////////

Grid/qcd/action/fermion/CloverHelpers.h

@@ -0,0 +1,333 @@
+/*************************************************************************************
+
+    Grid physics library, www.github.com/paboyle/Grid
+
+    Source file: ./lib/qcd/action/fermion/WilsonCloverFermionImplementation.h
+
+    Copyright (C) 2017 - 2022
+
+    Author: paboyle <paboyle@ph.ed.ac.uk>
+    Author: Daniel Richtmann <daniel.richtmann@gmail.com>
+    Author: Mattia Bruno <mattia.bruno@cern.ch>
+
+    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 */
+
+#pragma once
+
+#include <Grid/Grid.h>
+#include <Grid/qcd/spin/Dirac.h>
+#include <Grid/qcd/action/fermion/WilsonCloverHelpers.h>
+
+////////////////////////////////////////////
+// Standard Clover
+//   (4+m0) + csw * clover_term
+// Exp Clover
+//   (4+m0) * exp(csw/(4+m0) clover_term)
+//   = (4+m0) + csw * clover_term + ...
+////////////////////////////////////////////
+
+NAMESPACE_BEGIN(Grid);
+
+
+//////////////////////////////////
+// Generic Standard Clover
+//////////////////////////////////
+
+template<class Impl>
+class CloverHelpers: public WilsonCloverHelpers<Impl> {
+public:
+
+  INHERIT_IMPL_TYPES(Impl);
+  INHERIT_CLOVER_TYPES(Impl);
+
+  typedef WilsonCloverHelpers<Impl> Helpers;
+
+  static void Instantiate(CloverField& CloverTerm, CloverField& CloverTermInv, RealD csw_t, RealD diag_mass) {
+    GridBase *grid = CloverTerm.Grid();
+    CloverTerm += diag_mass;
+
+    int lvol = grid->lSites();
+    int DimRep = Impl::Dimension;
+    {
+      autoView(CTv,CloverTerm,CpuRead);
+      autoView(CTIv,CloverTermInv,CpuWrite);
+      thread_for(site, lvol, {
+        Coordinate lcoor;
+        grid->LocalIndexToLocalCoor(site, lcoor);
+        Eigen::MatrixXcd EigenCloverOp = Eigen::MatrixXcd::Zero(Ns * DimRep, Ns * DimRep);
+        Eigen::MatrixXcd EigenInvCloverOp = Eigen::MatrixXcd::Zero(Ns * DimRep, Ns * DimRep);
+        typename SiteClover::scalar_object Qx = Zero(), Qxinv = Zero();
+        peekLocalSite(Qx, CTv, lcoor);
+
+        for (int j = 0; j < Ns; j++)
+          for (int k = 0; k < Ns; k++)
+            for (int a = 0; a < DimRep; a++)
+              for (int b = 0; b < DimRep; b++){
+                auto zz =  Qx()(j, k)(a, b);
+                EigenCloverOp(a + j * DimRep, b + k * DimRep) = std::complex<double>(zz);
+              }
+
+        EigenInvCloverOp = EigenCloverOp.inverse();
+        for (int j = 0; j < Ns; j++)
+          for (int k = 0; k < Ns; k++)
+            for (int a = 0; a < DimRep; a++)
+              for (int b = 0; b < DimRep; b++)
+                Qxinv()(j, k)(a, b) = EigenInvCloverOp(a + j * DimRep, b + k * DimRep);
+               pokeLocalSite(Qxinv, CTIv, lcoor);
+      });
+    }
+  }
+
+  static GaugeLinkField Cmunu(std::vector<GaugeLinkField> &U, GaugeLinkField &lambda, int mu, int nu) {
+    return Helpers::Cmunu(U, lambda, mu, nu);
+  }
+
+};
+
+
+//////////////////////////////////
+// Generic Exp Clover
+//////////////////////////////////
+
+template<class Impl>
+class ExpCloverHelpers: public WilsonCloverHelpers<Impl> {
+public:
+
+  INHERIT_IMPL_TYPES(Impl);
+  INHERIT_CLOVER_TYPES(Impl);
+
+  template <typename vtype> using iImplClover = iScalar<iMatrix<iMatrix<vtype, Impl::Dimension>, Ns>>;
+  typedef WilsonCloverHelpers<Impl> Helpers;
+
+  // Can this be avoided?
+  static void IdentityTimesC(const CloverField& in, RealD c) {
+    int DimRep = Impl::Dimension;
+
+    autoView(in_v, in, AcceleratorWrite);
+
+    accelerator_for(ss, in.Grid()->oSites(), 1, {
+      for (int sa=0; sa<Ns; sa++)
+        for (int ca=0; ca<DimRep; ca++)
+          in_v[ss]()(sa,sa)(ca,ca) = c;
+    });
+  }
+
+  static int getNMAX(RealD prec, RealD R) {
+    /* compute stop condition for exponential */
+    int NMAX=1;
+    RealD cond=R*R/2.;
+
+    while (cond*std::exp(R)>prec) {
+      NMAX++;
+      cond*=R/(double)(NMAX+1);
+    }
+    return NMAX;
+  }
+
+  static int getNMAX(Lattice<iImplClover<vComplexD>> &t, RealD R) {return getNMAX(1e-12,R);}
+  static int getNMAX(Lattice<iImplClover<vComplexF>> &t, RealD R) {return getNMAX(1e-6,R);}
+
+  static void Instantiate(CloverField& Clover, CloverField& CloverInv, RealD csw_t, RealD diag_mass) {
+    GridBase* grid = Clover.Grid();
+    CloverField ExpClover(grid);
+
+    int NMAX = getNMAX(Clover, 3.*csw_t/diag_mass);
+
+    Clover *= (1.0/diag_mass);
+
+    // Taylor expansion, slow but generic
+    // Horner scheme: a0 + a1 x + a2 x^2 + .. = a0 + x (a1 + x(...))
+    // qN = cN
+    // qn = cn + qn+1 X
+    std::vector<RealD> cn(NMAX+1);
+    cn[0] = 1.0;
+    for (int i=1; i<=NMAX; i++)
+      cn[i] = cn[i-1] / RealD(i);
+
+    ExpClover = Zero();
+    IdentityTimesC(ExpClover, cn[NMAX]);
+    for (int i=NMAX-1; i>=0; i--)
+      ExpClover = ExpClover * Clover + cn[i];
+
+    // prepare inverse
+    CloverInv = (-1.0)*Clover;
+
+    Clover = ExpClover * diag_mass;
+
+    ExpClover = Zero();
+    IdentityTimesC(ExpClover, cn[NMAX]);
+    for (int i=NMAX-1; i>=0; i--)
+      ExpClover = ExpClover * CloverInv + cn[i];
+
+    CloverInv = ExpClover * (1.0/diag_mass);
+
+  }
+
+  static GaugeLinkField Cmunu(std::vector<GaugeLinkField> &U, GaugeLinkField &lambda, int mu, int nu) {
+    assert(0);
+    return lambda;
+  }
+
+};
+
+
+//////////////////////////////////
+// Compact Standard Clover
+//////////////////////////////////
+
+
+template<class Impl>
+class CompactCloverHelpers: public CompactWilsonCloverHelpers<Impl>,
+                            public WilsonCloverHelpers<Impl> {
+public:
+
+  INHERIT_IMPL_TYPES(Impl);
+  INHERIT_CLOVER_TYPES(Impl);
+  INHERIT_COMPACT_CLOVER_TYPES(Impl);
+
+  typedef WilsonCloverHelpers<Impl> Helpers;
+  typedef CompactWilsonCloverHelpers<Impl> CompactHelpers;
+
+  static void InstantiateClover(CloverField& Clover, CloverField& CloverInv, RealD csw_t, RealD diag_mass) {
+    Clover += diag_mass;
+  }
+
+  static void InvertClover(CloverField& InvClover,
+                            const CloverDiagonalField& diagonal,
+                            const CloverTriangleField& triangle,
+                            CloverDiagonalField&       diagonalInv,
+                            CloverTriangleField&       triangleInv,
+                            bool fixedBoundaries) {
+
+    CompactHelpers::Invert(diagonal, triangle, diagonalInv, triangleInv);
+  }
+
+  // TODO: implement Cmunu for better performances with compact layout, but don't do it
+  // here, but rather in WilsonCloverHelpers.h -> CompactWilsonCloverHelpers
+  static GaugeLinkField Cmunu(std::vector<GaugeLinkField> &U, GaugeLinkField &lambda, int mu, int nu) {
+    return Helpers::Cmunu(U, lambda, mu, nu);
+  }
+};
+
+//////////////////////////////////
+// Compact Exp Clover
+//////////////////////////////////
+
+template<class Impl>
+class CompactExpCloverHelpers: public CompactWilsonCloverHelpers<Impl> {
+public:
+
+  INHERIT_IMPL_TYPES(Impl);
+  INHERIT_CLOVER_TYPES(Impl);
+  INHERIT_COMPACT_CLOVER_TYPES(Impl);
+
+  template <typename vtype> using iImplClover = iScalar<iMatrix<iMatrix<vtype, Impl::Dimension>, Ns>>;
+  typedef CompactWilsonCloverHelpers<Impl> CompactHelpers;
+
+  // Can this be avoided?
+  static void IdentityTimesC(const CloverField& in, RealD c) {
+    int DimRep = Impl::Dimension;
+
+    autoView(in_v, in, AcceleratorWrite);
+
+    accelerator_for(ss, in.Grid()->oSites(), 1, {
+      for (int sa=0; sa<Ns; sa++)
+        for (int ca=0; ca<DimRep; ca++)
+          in_v[ss]()(sa,sa)(ca,ca) = c;
+    });
+  }
+
+  static int getNMAX(RealD prec, RealD R) {
+    /* compute stop condition for exponential */
+    int NMAX=1;
+    RealD cond=R*R/2.;
+
+    while (cond*std::exp(R)>prec) {
+      NMAX++;
+      cond*=R/(double)(NMAX+1);
+    }
+    return NMAX;
+  }
+
+  static int getNMAX(Lattice<iImplClover<vComplexD>> &t, RealD R) {return getNMAX(1e-12,R);}
+  static int getNMAX(Lattice<iImplClover<vComplexF>> &t, RealD R) {return getNMAX(1e-6,R);}
+
+  static void InstantiateClover(CloverField& Clover, CloverField& CloverInv, RealD csw_t, RealD diag_mass) {
+
+    GridBase* grid = Clover.Grid();
+    CloverField ExpClover(grid);
+
+    int NMAX = getNMAX(Clover, 3.*csw_t/diag_mass);
+
+    Clover *= (1.0/diag_mass);
+
+    // Taylor expansion, slow but generic
+    // Horner scheme: a0 + a1 x + a2 x^2 + .. = a0 + x (a1 + x(...))
+    // qN = cN
+    // qn = cn + qn+1 X
+    std::vector<RealD> cn(NMAX+1);
+    cn[0] = 1.0;
+    for (int i=1; i<=NMAX; i++)
+      cn[i] = cn[i-1] / RealD(i);
+
+    ExpClover = Zero();
+    IdentityTimesC(ExpClover, cn[NMAX]);
+    for (int i=NMAX-1; i>=0; i--)
+      ExpClover = ExpClover * Clover + cn[i];
+
+    // prepare inverse
+    CloverInv = (-1.0)*Clover;
+
+    Clover = ExpClover * diag_mass;
+
+    ExpClover = Zero();
+    IdentityTimesC(ExpClover, cn[NMAX]);
+    for (int i=NMAX-1; i>=0; i--)
+      ExpClover = ExpClover * CloverInv + cn[i];
+
+    CloverInv = ExpClover * (1.0/diag_mass);
+
+  }
+
+  static void InvertClover(CloverField& InvClover,
+                            const CloverDiagonalField& diagonal,
+                            const CloverTriangleField& triangle,
+                            CloverDiagonalField&       diagonalInv,
+                            CloverTriangleField&       triangleInv,
+                            bool fixedBoundaries) {
+
+    if (fixedBoundaries)
+    {
+      CompactHelpers::Invert(diagonal, triangle, diagonalInv, triangleInv);
+    }
+    else
+    {
+      CompactHelpers::ConvertLayout(InvClover, diagonalInv, triangleInv);
+    }
+  }
+
+  static GaugeLinkField Cmunu(std::vector<GaugeLinkField> &U, GaugeLinkField &lambda, int mu, int nu) {
+    assert(0);
+    return lambda;
+  }
+
+};
+
+
+NAMESPACE_END(Grid);

Grid/qcd/action/fermion/CompactWilsonCloverFermion.h

@@ -31,6 +31,7 @@
 
 #include <Grid/qcd/action/fermion/WilsonCloverTypes.h>
 #include <Grid/qcd/action/fermion/WilsonCloverHelpers.h>
+#include <Grid/qcd/action/fermion/CloverHelpers.h>
 
 NAMESPACE_BEGIN(Grid);
 
@@ -85,7 +86,7 @@ NAMESPACE_BEGIN(Grid);
 //           + (2 * 1 + 4 * 1/2) triangle parts = 4 triangle parts =  60 complex words per site
 //                                                                 =  84 complex words per site
 
-template<class Impl>
+template<class Impl, class CloverHelpers>
 class CompactWilsonCloverFermion : public WilsonFermion<Impl>,
                                    public WilsonCloverHelpers<Impl>,
                                    public CompactWilsonCloverHelpers<Impl> {
@@ -224,7 +225,7 @@ public:
   RealD csw_t;
   RealD cF;
 
-  bool open_boundaries;
+  bool fixedBoundaries;
 
   CloverDiagonalField Diagonal,    DiagonalEven,    DiagonalOdd;
   CloverDiagonalField DiagonalInv, DiagonalInvEven, DiagonalInvOdd;

Grid/qcd/action/fermion/Fermion.h

@@ -138,38 +138,52 @@ typedef WilsonTMFermion<WilsonImplF> WilsonTMFermionF;
 typedef WilsonTMFermion<WilsonImplD> WilsonTMFermionD;
 
 // Clover fermions
-typedef WilsonCloverFermion<WilsonImplR> WilsonCloverFermionR;
+template <typename WImpl> using WilsonClover = WilsonCloverFermion<WImpl, CloverHelpers<WImpl>>;
-typedef WilsonCloverFermion<WilsonImplF> WilsonCloverFermionF;
+template <typename WImpl> using WilsonExpClover = WilsonCloverFermion<WImpl, ExpCloverHelpers<WImpl>>;
-typedef WilsonCloverFermion<WilsonImplD> WilsonCloverFermionD;
 
-typedef WilsonCloverFermion<WilsonAdjImplR> WilsonCloverAdjFermionR;
+typedef WilsonClover<WilsonImplR> WilsonCloverFermionR;
-typedef WilsonCloverFermion<WilsonAdjImplF> WilsonCloverAdjFermionF;
+typedef WilsonClover<WilsonImplF> WilsonCloverFermionF;
-typedef WilsonCloverFermion<WilsonAdjImplD> WilsonCloverAdjFermionD;
+typedef WilsonClover<WilsonImplD> WilsonCloverFermionD;
 
-typedef WilsonCloverFermion<WilsonTwoIndexSymmetricImplR> WilsonCloverTwoIndexSymmetricFermionR;
+typedef WilsonExpClover<WilsonImplR> WilsonExpCloverFermionR;
-typedef WilsonCloverFermion<WilsonTwoIndexSymmetricImplF> WilsonCloverTwoIndexSymmetricFermionF;
+typedef WilsonExpClover<WilsonImplF> WilsonExpCloverFermionF;
-typedef WilsonCloverFermion<WilsonTwoIndexSymmetricImplD> WilsonCloverTwoIndexSymmetricFermionD;
+typedef WilsonExpClover<WilsonImplD> WilsonExpCloverFermionD;
 
-typedef WilsonCloverFermion<WilsonTwoIndexAntiSymmetricImplR> WilsonCloverTwoIndexAntiSymmetricFermionR;
+typedef WilsonClover<WilsonAdjImplR> WilsonCloverAdjFermionR;
-typedef WilsonCloverFermion<WilsonTwoIndexAntiSymmetricImplF> WilsonCloverTwoIndexAntiSymmetricFermionF;
+typedef WilsonClover<WilsonAdjImplF> WilsonCloverAdjFermionF;
-typedef WilsonCloverFermion<WilsonTwoIndexAntiSymmetricImplD> WilsonCloverTwoIndexAntiSymmetricFermionD;
+typedef WilsonClover<WilsonAdjImplD> WilsonCloverAdjFermionD;
+
+typedef WilsonClover<WilsonTwoIndexSymmetricImplR> WilsonCloverTwoIndexSymmetricFermionR;
+typedef WilsonClover<WilsonTwoIndexSymmetricImplF> WilsonCloverTwoIndexSymmetricFermionF;
+typedef WilsonClover<WilsonTwoIndexSymmetricImplD> WilsonCloverTwoIndexSymmetricFermionD;
+
+typedef WilsonClover<WilsonTwoIndexAntiSymmetricImplR> WilsonCloverTwoIndexAntiSymmetricFermionR;
+typedef WilsonClover<WilsonTwoIndexAntiSymmetricImplF> WilsonCloverTwoIndexAntiSymmetricFermionF;
+typedef WilsonClover<WilsonTwoIndexAntiSymmetricImplD> WilsonCloverTwoIndexAntiSymmetricFermionD;
 
 // Compact Clover fermions
-typedef CompactWilsonCloverFermion<WilsonImplR> CompactWilsonCloverFermionR;
+template <typename WImpl> using CompactWilsonClover = CompactWilsonCloverFermion<WImpl, CompactCloverHelpers<WImpl>>;
-typedef CompactWilsonCloverFermion<WilsonImplF> CompactWilsonCloverFermionF;
+template <typename WImpl> using CompactWilsonExpClover = CompactWilsonCloverFermion<WImpl, CompactExpCloverHelpers<WImpl>>;
-typedef CompactWilsonCloverFermion<WilsonImplD> CompactWilsonCloverFermionD;
 
-typedef CompactWilsonCloverFermion<WilsonAdjImplR> CompactWilsonCloverAdjFermionR;
+typedef CompactWilsonClover<WilsonImplR> CompactWilsonCloverFermionR;
-typedef CompactWilsonCloverFermion<WilsonAdjImplF> CompactWilsonCloverAdjFermionF;
+typedef CompactWilsonClover<WilsonImplF> CompactWilsonCloverFermionF;
-typedef CompactWilsonCloverFermion<WilsonAdjImplD> CompactWilsonCloverAdjFermionD;
+typedef CompactWilsonClover<WilsonImplD> CompactWilsonCloverFermionD;
 
-typedef CompactWilsonCloverFermion<WilsonTwoIndexSymmetricImplR> CompactWilsonCloverTwoIndexSymmetricFermionR;
+typedef CompactWilsonExpClover<WilsonImplR> CompactWilsonExpCloverFermionR;
-typedef CompactWilsonCloverFermion<WilsonTwoIndexSymmetricImplF> CompactWilsonCloverTwoIndexSymmetricFermionF;
+typedef CompactWilsonExpClover<WilsonImplF> CompactWilsonExpCloverFermionF;
-typedef CompactWilsonCloverFermion<WilsonTwoIndexSymmetricImplD> CompactWilsonCloverTwoIndexSymmetricFermionD;
+typedef CompactWilsonExpClover<WilsonImplD> CompactWilsonExpCloverFermionD;
 
-typedef CompactWilsonCloverFermion<WilsonTwoIndexAntiSymmetricImplR> CompactWilsonCloverTwoIndexAntiSymmetricFermionR;
+typedef CompactWilsonClover<WilsonAdjImplR> CompactWilsonCloverAdjFermionR;
-typedef CompactWilsonCloverFermion<WilsonTwoIndexAntiSymmetricImplF> CompactWilsonCloverTwoIndexAntiSymmetricFermionF;
+typedef CompactWilsonClover<WilsonAdjImplF> CompactWilsonCloverAdjFermionF;
-typedef CompactWilsonCloverFermion<WilsonTwoIndexAntiSymmetricImplD> CompactWilsonCloverTwoIndexAntiSymmetricFermionD;
+typedef CompactWilsonClover<WilsonAdjImplD> CompactWilsonCloverAdjFermionD;
+
+typedef CompactWilsonClover<WilsonTwoIndexSymmetricImplR> CompactWilsonCloverTwoIndexSymmetricFermionR;
+typedef CompactWilsonClover<WilsonTwoIndexSymmetricImplF> CompactWilsonCloverTwoIndexSymmetricFermionF;
+typedef CompactWilsonClover<WilsonTwoIndexSymmetricImplD> CompactWilsonCloverTwoIndexSymmetricFermionD;
+
+typedef CompactWilsonClover<WilsonTwoIndexAntiSymmetricImplR> CompactWilsonCloverTwoIndexAntiSymmetricFermionR;
+typedef CompactWilsonClover<WilsonTwoIndexAntiSymmetricImplF> CompactWilsonCloverTwoIndexAntiSymmetricFermionF;
+typedef CompactWilsonClover<WilsonTwoIndexAntiSymmetricImplD> CompactWilsonCloverTwoIndexAntiSymmetricFermionD;
 
 // Domain Wall fermions
 typedef DomainWallFermion<WilsonImplR> DomainWallFermionR;

Grid/qcd/action/fermion/WilsonCloverFermion.h

@@ -32,6 +32,7 @@
 
 #include <Grid/qcd/action/fermion/WilsonCloverTypes.h>
 #include <Grid/qcd/action/fermion/WilsonCloverHelpers.h>
+#include <Grid/qcd/action/fermion/CloverHelpers.h>
 
 NAMESPACE_BEGIN(Grid);
 
@@ -51,7 +52,7 @@ NAMESPACE_BEGIN(Grid);
 // csw_r = csw_t to recover the isotropic version
 //////////////////////////////////////////////////////////////////
 
-template <class Impl>
+template<class Impl, class CloverHelpers>
 class WilsonCloverFermion : public WilsonFermion<Impl>,
                             public WilsonCloverHelpers<Impl>
 {

Grid/qcd/action/fermion/WilsonCloverHelpers.h

@@ -209,6 +209,8 @@ public:
 };
 
 
+////////////////////////////////////////////////////////
+
 template<class Impl> class CompactWilsonCloverHelpers {
 public:
 

Grid/qcd/action/fermion/WilsonCloverTypes.h

@@ -47,8 +47,6 @@ class CompactWilsonCloverTypes {
 public:
   INHERIT_IMPL_TYPES(Impl);
 
-  static_assert(Nd == 4 && Nc == 3 && Ns == 4 && Impl::Dimension == 3, "Wrong dimensions");
-
   static constexpr int Nred      = Nc * Nhs;        // 6
   static constexpr int Nblock    = Nhs;             // 2
   static constexpr int Ndiagonal = Nred;            // 6

Grid/qcd/action/fermion/WilsonCompressor.h

@@ -117,19 +117,19 @@ public:
     typedef decltype(coalescedRead(*in))    sobj;
     typedef decltype(coalescedRead(*out0)) hsobj;
 
-    unsigned int Nsimd = vobj::Nsimd();
+    constexpr unsigned int Nsimd = vobj::Nsimd();
     unsigned int mask = Nsimd >> (type + 1);
     int lane = acceleratorSIMTlane(Nsimd);
     int j0 = lane &(~mask); // inner coor zero
     int j1 = lane |(mask) ; // inner coor one
-    const vobj *vp0 = &in[k];
+    const vobj *vp0 = &in[k];  // out0[j] = merge low bit of type from in[k] and in[m] 
-    const vobj *vp1 = &in[m];
+    const vobj *vp1 = &in[m];  // out1[j] = merge hi  bit of type from in[k] and in[m]
-    const vobj *vp = (lane&mask) ? vp1:vp0;
+    const vobj *vp = (lane&mask) ? vp1:vp0;// if my lane has high bit take vp1, low bit take vp0
-    auto sa = coalescedRead(*vp,j0);
+    auto sa = coalescedRead(*vp,j0); // lane to read for out 0, NB 50% read coalescing
-    auto sb = coalescedRead(*vp,j1);
+    auto sb = coalescedRead(*vp,j1); // lane to read for out 1
     hsobj psa, psb;
-    projector::Proj(psa,sa,mu,dag);
+    projector::Proj(psa,sa,mu,dag);  // spin project the result0
-    projector::Proj(psb,sb,mu,dag);
+    projector::Proj(psb,sb,mu,dag);  // spin project the result1
     coalescedWrite(out0[j],psa);
     coalescedWrite(out1[j],psb);
 #else

Grid/qcd/action/fermion/implementation/CompactWilsonCloverFermionImplementation.h

@@ -32,9 +32,10 @@
 #include <Grid/qcd/spin/Dirac.h>
 #include <Grid/qcd/action/fermion/CompactWilsonCloverFermion.h>
 
+
 NAMESPACE_BEGIN(Grid);
-template<class Impl>
+template<class Impl, class CloverHelpers>
-CompactWilsonCloverFermion<Impl>::CompactWilsonCloverFermion(GaugeField& _Umu,
+CompactWilsonCloverFermion<Impl, CloverHelpers>::CompactWilsonCloverFermion(GaugeField& _Umu,
                                                                             GridCartesian& Fgrid,
                                                                             GridRedBlackCartesian& Hgrid,
                                                                             const RealD _mass,
@@ -47,7 +48,7 @@ CompactWilsonCloverFermion<Impl>::CompactWilsonCloverFermion(GaugeField& _Umu,
   , csw_r(_csw_r)
   , csw_t(_csw_t)
   , cF(_cF)
-  , open_boundaries(impl_p.boundary_phases[Nd-1] == 0.0)
+  , fixedBoundaries(impl_p.boundary_phases[Nd-1] == 0.0)
   , Diagonal(&Fgrid),        Triangle(&Fgrid)
   , DiagonalEven(&Hgrid),    TriangleEven(&Hgrid)
   , DiagonalOdd(&Hgrid),     TriangleOdd(&Hgrid)
@@ -58,83 +59,85 @@ CompactWilsonCloverFermion<Impl>::CompactWilsonCloverFermion(GaugeField& _Umu,
   , BoundaryMask(&Fgrid)
   , BoundaryMaskEven(&Hgrid), BoundaryMaskOdd(&Hgrid)
 {
+  assert(Nd == 4 && Nc == 3 && Ns == 4 && Impl::Dimension == 3);
+
   csw_r *= 0.5;
   csw_t *= 0.5;
   if (clover_anisotropy.isAnisotropic)
     csw_r /= clover_anisotropy.xi_0;
 
   ImportGauge(_Umu);
-  if (open_boundaries) {
+  if (fixedBoundaries) {
     this->BoundaryMaskEven.Checkerboard() = Even;
     this->BoundaryMaskOdd.Checkerboard() = Odd;
     CompactHelpers::SetupMasks(this->BoundaryMask, this->BoundaryMaskEven, this->BoundaryMaskOdd);
   }
 }
 
-template<class Impl>
+template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl>::Dhop(const FermionField& in, FermionField& out, int dag) {
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::Dhop(const FermionField& in, FermionField& out, int dag) {
   WilsonBase::Dhop(in, out, dag);
-  if(open_boundaries) ApplyBoundaryMask(out);
+  if(fixedBoundaries) ApplyBoundaryMask(out);
 }
 
-template<class Impl>
+template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl>::DhopOE(const FermionField& in, FermionField& out, int dag) {
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::DhopOE(const FermionField& in, FermionField& out, int dag) {
   WilsonBase::DhopOE(in, out, dag);
-  if(open_boundaries) ApplyBoundaryMask(out);
+  if(fixedBoundaries) ApplyBoundaryMask(out);
 }
 
-template<class Impl>
+template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl>::DhopEO(const FermionField& in, FermionField& out, int dag) {
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::DhopEO(const FermionField& in, FermionField& out, int dag) {
   WilsonBase::DhopEO(in, out, dag);
-  if(open_boundaries) ApplyBoundaryMask(out);
+  if(fixedBoundaries) ApplyBoundaryMask(out);
 }
 
-template<class Impl>
+template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl>::DhopDir(const FermionField& in, FermionField& out, int dir, int disp) {
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::DhopDir(const FermionField& in, FermionField& out, int dir, int disp) {
   WilsonBase::DhopDir(in, out, dir, disp);
-  if(this->open_boundaries) ApplyBoundaryMask(out);
+  if(this->fixedBoundaries) ApplyBoundaryMask(out);
 }
 
-template<class Impl>
+template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl>::DhopDirAll(const FermionField& in, std::vector<FermionField>& out) {
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::DhopDirAll(const FermionField& in, std::vector<FermionField>& out) {
   WilsonBase::DhopDirAll(in, out);
-  if(this->open_boundaries) {
+  if(this->fixedBoundaries) {
     for(auto& o : out) ApplyBoundaryMask(o);
   }
 }
 
-template<class Impl>
+template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl>::M(const FermionField& in, FermionField& out) {
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::M(const FermionField& in, FermionField& out) {
   out.Checkerboard() = in.Checkerboard();
   WilsonBase::Dhop(in, out, DaggerNo); // call base to save applying bc
   Mooee(in, Tmp);
   axpy(out, 1.0, out, Tmp);
-  if(open_boundaries) ApplyBoundaryMask(out);
+  if(fixedBoundaries) ApplyBoundaryMask(out);
 }
 
-template<class Impl>
+template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl>::Mdag(const FermionField& in, FermionField& out) {
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::Mdag(const FermionField& in, FermionField& out) {
   out.Checkerboard() = in.Checkerboard();
   WilsonBase::Dhop(in, out, DaggerYes);  // call base to save applying bc
   MooeeDag(in, Tmp);
   axpy(out, 1.0, out, Tmp);
-  if(open_boundaries) ApplyBoundaryMask(out);
+  if(fixedBoundaries) ApplyBoundaryMask(out);
 }
 
-template<class Impl>
+template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl>::Meooe(const FermionField& in, FermionField& out) {
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::Meooe(const FermionField& in, FermionField& out) {
   WilsonBase::Meooe(in, out);
-  if(open_boundaries) ApplyBoundaryMask(out);
+  if(fixedBoundaries) ApplyBoundaryMask(out);
 }
 
-template<class Impl>
+template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl>::MeooeDag(const FermionField& in, FermionField& out) {
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::MeooeDag(const FermionField& in, FermionField& out) {
   WilsonBase::MeooeDag(in, out);
-  if(open_boundaries) ApplyBoundaryMask(out);
+  if(fixedBoundaries) ApplyBoundaryMask(out);
 }
 
-template<class Impl>
+template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl>::Mooee(const FermionField& in, FermionField& out) {
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::Mooee(const FermionField& in, FermionField& out) {
   if(in.Grid()->_isCheckerBoarded) {
     if(in.Checkerboard() == Odd) {
       MooeeInternal(in, out, DiagonalOdd, TriangleOdd);
@@ -144,16 +147,16 @@ void CompactWilsonCloverFermion<Impl>::Mooee(const FermionField& in, FermionFiel
   } else {
     MooeeInternal(in, out, Diagonal, Triangle);
   }
-  if(open_boundaries) ApplyBoundaryMask(out);
+  if(fixedBoundaries) ApplyBoundaryMask(out);
 }
 
-template<class Impl>
+template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl>::MooeeDag(const FermionField& in, FermionField& out) {
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::MooeeDag(const FermionField& in, FermionField& out) {
   Mooee(in, out); // blocks are hermitian
 }
 
-template<class Impl>
+template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl>::MooeeInv(const FermionField& in, FermionField& out) {
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::MooeeInv(const FermionField& in, FermionField& out) {
   if(in.Grid()->_isCheckerBoarded) {
     if(in.Checkerboard() == Odd) {
       MooeeInternal(in, out, DiagonalInvOdd, TriangleInvOdd);
@@ -163,27 +166,27 @@ void CompactWilsonCloverFermion<Impl>::MooeeInv(const FermionField& in, FermionF
   } else {
     MooeeInternal(in, out, DiagonalInv, TriangleInv);
   }
-  if(open_boundaries) ApplyBoundaryMask(out);
+  if(fixedBoundaries) ApplyBoundaryMask(out);
 }
 
-template<class Impl>
+template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl>::MooeeInvDag(const FermionField& in, FermionField& out) {
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::MooeeInvDag(const FermionField& in, FermionField& out) {
   MooeeInv(in, out); // blocks are hermitian
 }
 
-template<class Impl>
+template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl>::Mdir(const FermionField& in, FermionField& out, int dir, int disp) {
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::Mdir(const FermionField& in, FermionField& out, int dir, int disp) {
   DhopDir(in, out, dir, disp);
 }
 
-template<class Impl>
+template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl>::MdirAll(const FermionField& in, std::vector<FermionField>& out) {
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::MdirAll(const FermionField& in, std::vector<FermionField>& out) {
   DhopDirAll(in, out);
 }
 
-template<class Impl>
+template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl>::MDeriv(GaugeField& force, const FermionField& X, const FermionField& Y, int dag) {
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::MDeriv(GaugeField& force, const FermionField& X, const FermionField& Y, int dag) {
-  assert(!open_boundaries); // TODO check for changes required for open bc
+  assert(!fixedBoundaries); // TODO check for changes required for open bc
 
   // NOTE: code copied from original clover term
   conformable(X.Grid(), Y.Grid());
@@ -254,7 +257,7 @@ void CompactWilsonCloverFermion<Impl>::MDeriv(GaugeField& force, const FermionFi
       }
       PropagatorField Slambda = Gamma(sigma[count]) * Lambda; // sigma checked
       Impl::TraceSpinImpl(lambda, Slambda);                   // traceSpin ok
-      force_mu -= factor*Helpers::Cmunu(U, lambda, mu, nu);   // checked
+      force_mu -= factor*CloverHelpers::Cmunu(U, lambda, mu, nu);   // checked
       count++;
     }
 
@@ -264,18 +267,18 @@ void CompactWilsonCloverFermion<Impl>::MDeriv(GaugeField& force, const FermionFi
   force += clover_force;
 }
 
-template<class Impl>
+template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl>::MooDeriv(GaugeField& mat, const FermionField& U, const FermionField& V, int dag) {
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::MooDeriv(GaugeField& mat, const FermionField& U, const FermionField& V, int dag) {
   assert(0);
 }
 
-template<class Impl>
+template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl>::MeeDeriv(GaugeField& mat, const FermionField& U, const FermionField& V, int dag) {
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::MeeDeriv(GaugeField& mat, const FermionField& U, const FermionField& V, int dag) {
   assert(0);
 }
 
-template<class Impl>
+template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl>::MooeeInternal(const FermionField&        in,
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::MooeeInternal(const FermionField&        in,
                     FermionField&              out,
                     const CloverDiagonalField& diagonal,
                     const CloverTriangleField& triangle) {
@@ -288,8 +291,8 @@ void CompactWilsonCloverFermion<Impl>::MooeeInternal(const FermionField&
   CompactHelpers::MooeeKernel(diagonal.oSites(), 1, in, out, diagonal, triangle);
 }
 
-template<class Impl>
+template<class Impl, class CloverHelpers>
-void CompactWilsonCloverFermion<Impl>::ImportGauge(const GaugeField& _Umu) {
+void CompactWilsonCloverFermion<Impl, CloverHelpers>::ImportGauge(const GaugeField& _Umu) {
   // NOTE: parts copied from original implementation
 
   // Import gauge into base class
@@ -302,6 +305,7 @@ void CompactWilsonCloverFermion<Impl>::ImportGauge(const GaugeField& _Umu) {
   GridBase* grid = _Umu.Grid();
   typename Impl::GaugeLinkField Bx(grid), By(grid), Bz(grid), Ex(grid), Ey(grid), Ez(grid);
   CloverField TmpOriginal(grid);
+  CloverField TmpInverse(grid);
 
   // Compute the field strength terms mu>nu
   double t2 = usecond();
@@ -321,22 +325,30 @@ void CompactWilsonCloverFermion<Impl>::ImportGauge(const GaugeField& _Umu) {
   TmpOriginal += Helpers::fillCloverXT(Ex) * csw_t;
   TmpOriginal += Helpers::fillCloverYT(Ey) * csw_t;
   TmpOriginal += Helpers::fillCloverZT(Ez) * csw_t;
-  TmpOriginal += this->diag_mass;
+
+  // Instantiate the clover term
+  // - In case of the standard clover the mass term is added
+  // - In case of the exponential clover the clover term is exponentiated
+  double t4 = usecond();
+  CloverHelpers::InstantiateClover(TmpOriginal, TmpInverse, csw_t, this->diag_mass);
 
   // Convert the data layout of the clover term
-  double t4 = usecond();
+  double t5 = usecond();
   CompactHelpers::ConvertLayout(TmpOriginal, Diagonal, Triangle);
 
-  // Possible modify the boundary values
+  // Modify the clover term at the temporal boundaries in case of open boundary conditions
-  double t5 = usecond();
-  if(open_boundaries) CompactHelpers::ModifyBoundaries(Diagonal, Triangle, csw_t, cF, this->diag_mass);
-
-  // Invert the clover term in the improved layout
   double t6 = usecond();
-  CompactHelpers::Invert(Diagonal, Triangle, DiagonalInv, TriangleInv);
+  if(fixedBoundaries) CompactHelpers::ModifyBoundaries(Diagonal, Triangle, csw_t, cF, this->diag_mass);
+
+  // Invert the Clover term
+  // In case of the exponential clover with (anti-)periodic boundary conditions exp(-Clover) saved
+  // in TmpInverse can be used. In all other cases the clover term has to be explictly inverted.
+  // TODO: For now this inversion is explictly done on the CPU
+  double t7 = usecond();
+  CloverHelpers::InvertClover(TmpInverse, Diagonal, Triangle, DiagonalInv, TriangleInv, fixedBoundaries);
 
   // Fill the remaining clover fields
-  double t7 = usecond();
+  double t8 = usecond();
   pickCheckerboard(Even, DiagonalEven,    Diagonal);
   pickCheckerboard(Even, TriangleEven,    Triangle);
   pickCheckerboard(Odd,  DiagonalOdd,     Diagonal);
@@ -347,20 +359,19 @@ void CompactWilsonCloverFermion<Impl>::ImportGauge(const GaugeField& _Umu) {
   pickCheckerboard(Odd,  TriangleInvOdd,  TriangleInv);
 
   // Report timings
-  double t8 = usecond();
+  double t9 = usecond();
-#if 0
+
-  std::cout << GridLogMessage << "CompactWilsonCloverFermion::ImportGauge timings:"
+  std::cout << GridLogDebug << "CompactWilsonCloverFermion::ImportGauge timings:" << std::endl;
-            << " WilsonFermion::Importgauge = " << (t1 - t0) / 1e6
+  std::cout << GridLogDebug << "WilsonFermion::Importgauge = " << (t1 - t0) / 1e6 << std::endl;
-            << ", allocations = "               << (t2 - t1) / 1e6
+  std::cout << GridLogDebug << "allocations =                " << (t2 - t1) / 1e6 << std::endl;
-            << ", field strength = "            << (t3 - t2) / 1e6
+  std::cout << GridLogDebug << "field strength =             " << (t3 - t2) / 1e6 << std::endl;
-            << ", fill clover = "               << (t4 - t3) / 1e6
+  std::cout << GridLogDebug << "fill clover =                " << (t4 - t3) / 1e6 << std::endl;
-            << ", convert = "                   << (t5 - t4) / 1e6
+  std::cout << GridLogDebug << "instantiate clover =         " << (t5 - t4) / 1e6 << std::endl;
-            << ", boundaries = "                << (t6 - t5) / 1e6
+  std::cout << GridLogDebug << "convert layout =             " << (t6 - t5) / 1e6 << std::endl;
-            << ", inversions = "                << (t7 - t6) / 1e6
+  std::cout << GridLogDebug << "modify boundaries =          " << (t7 - t6) / 1e6 << std::endl;
-            << ", pick cbs = "                  << (t8 - t7) / 1e6
+  std::cout << GridLogDebug << "invert clover =              " << (t8 - t7) / 1e6 << std::endl;
-            << ", total = "                     << (t8 - t0) / 1e6
+  std::cout << GridLogDebug << "pick cbs =                   " << (t9 - t8) / 1e6 << std::endl;
-            << std::endl;
+  std::cout << GridLogDebug << "total =                      " << (t9 - t0) / 1e6 << std::endl;
-#endif
 }
 
 NAMESPACE_END(Grid);

Grid/qcd/action/fermion/implementation/WilsonCloverFermionImplementation.h

@@ -34,8 +34,8 @@
 
 NAMESPACE_BEGIN(Grid);
 
-template<class Impl>
+template<class Impl, class CloverHelpers>
-WilsonCloverFermion<Impl>::WilsonCloverFermion(GaugeField&                         _Umu,
+WilsonCloverFermion<Impl, CloverHelpers>::WilsonCloverFermion(GaugeField&                         _Umu,
                                                GridCartesian&                      Fgrid,
                                                GridRedBlackCartesian&              Hgrid,
                                                const RealD                         _mass,
@@ -74,8 +74,8 @@ WilsonCloverFermion<Impl>::WilsonCloverFermion(GaugeField&
 }
 
 // *NOT* EO
-template <class Impl>
+template<class Impl, class CloverHelpers>
-void WilsonCloverFermion<Impl>::M(const FermionField &in, FermionField &out)
+void WilsonCloverFermion<Impl, CloverHelpers>::M(const FermionField &in, FermionField &out)
 {
   FermionField temp(out.Grid());
 
@@ -89,8 +89,8 @@ void WilsonCloverFermion<Impl>::M(const FermionField &in, FermionField &out)
   out += temp;
 }
 
-template <class Impl>
+template<class Impl, class CloverHelpers>
-void WilsonCloverFermion<Impl>::Mdag(const FermionField &in, FermionField &out)
+void WilsonCloverFermion<Impl, CloverHelpers>::Mdag(const FermionField &in, FermionField &out)
 {
   FermionField temp(out.Grid());
 
@@ -104,8 +104,8 @@ void WilsonCloverFermion<Impl>::Mdag(const FermionField &in, FermionField &out)
   out += temp;
 }
 
-template <class Impl>
+template<class Impl, class CloverHelpers>
-void WilsonCloverFermion<Impl>::ImportGauge(const GaugeField &_Umu)
+void WilsonCloverFermion<Impl, CloverHelpers>::ImportGauge(const GaugeField &_Umu)
 {
   double t0 = usecond();
   WilsonFermion<Impl>::ImportGauge(_Umu);
@@ -131,47 +131,11 @@ void WilsonCloverFermion<Impl>::ImportGauge(const GaugeField &_Umu)
   CloverTerm += Helpers::fillCloverXT(Ex) * csw_t;
   CloverTerm += Helpers::fillCloverYT(Ey) * csw_t;
   CloverTerm += Helpers::fillCloverZT(Ez) * csw_t;
-  CloverTerm += diag_mass;
    
   double t4 = usecond();
-  int lvol = _Umu.Grid()->lSites();
+  CloverHelpers::Instantiate(CloverTerm, CloverTermInv, csw_t, this->diag_mass);
-  int DimRep = Impl::Dimension;
 
   double t5 = usecond();
-  {
-    autoView(CTv,CloverTerm,CpuRead);
-    autoView(CTIv,CloverTermInv,CpuWrite);
-    thread_for(site, lvol, {
-      Coordinate lcoor;
-      grid->LocalIndexToLocalCoor(site, lcoor);
-      Eigen::MatrixXcd EigenCloverOp = Eigen::MatrixXcd::Zero(Ns * DimRep, Ns * DimRep);
-      Eigen::MatrixXcd EigenInvCloverOp = Eigen::MatrixXcd::Zero(Ns * DimRep, Ns * DimRep);
-      typename SiteClover::scalar_object Qx = Zero(), Qxinv = Zero();
-      peekLocalSite(Qx, CTv, lcoor);
-      //if (csw!=0){
-      for (int j = 0; j < Ns; j++)
-	for (int k = 0; k < Ns; k++)
-	  for (int a = 0; a < DimRep; a++)
-	    for (int b = 0; b < DimRep; b++){
-	      auto zz =  Qx()(j, k)(a, b);
-	      EigenCloverOp(a + j * DimRep, b + k * DimRep) = std::complex<double>(zz);
-	    }
-      //   if (site==0) std::cout << "site =" << site << "\n" << EigenCloverOp << std::endl;
-      
-      EigenInvCloverOp = EigenCloverOp.inverse();
-      //std::cout << EigenInvCloverOp << std::endl;
-      for (int j = 0; j < Ns; j++)
-	for (int k = 0; k < Ns; k++)
-	  for (int a = 0; a < DimRep; a++)
-	    for (int b = 0; b < DimRep; b++)
-	      Qxinv()(j, k)(a, b) = EigenInvCloverOp(a + j * DimRep, b + k * DimRep);
-      //    if (site==0) std::cout << "site =" << site << "\n" << EigenInvCloverOp << std::endl;
-      //  }
-      pokeLocalSite(Qxinv, CTIv, lcoor);
-    });
-  }
-
-  double t6 = usecond();
   // Separate the even and odd parts
   pickCheckerboard(Even, CloverTermEven, CloverTerm);
   pickCheckerboard(Odd, CloverTermOdd, CloverTerm);
@@ -184,48 +148,44 @@ void WilsonCloverFermion<Impl>::ImportGauge(const GaugeField &_Umu)
 
   pickCheckerboard(Even, CloverTermInvDagEven, adj(CloverTermInv));
   pickCheckerboard(Odd, CloverTermInvDagOdd, adj(CloverTermInv));
-  double t7 = usecond();
+  double t6 = usecond();
 
-#if 0
+  std::cout << GridLogDebug << "WilsonCloverFermion::ImportGauge timings:" << std::endl;
-  std::cout << GridLogMessage << "WilsonCloverFermion::ImportGauge timings:"
+  std::cout << GridLogDebug << "WilsonFermion::Importgauge = " << (t1 - t0) / 1e6 << std::endl;
-            << " WilsonFermion::Importgauge = " << (t1 - t0) / 1e6
+  std::cout << GridLogDebug << "allocations =                " << (t2 - t1) / 1e6 << std::endl;
-            << ", allocations = "               << (t2 - t1) / 1e6
+  std::cout << GridLogDebug << "field strength =             " << (t3 - t2) / 1e6 << std::endl;
-            << ", field strength = "            << (t3 - t2) / 1e6
+  std::cout << GridLogDebug << "fill clover =                " << (t4 - t3) / 1e6 << std::endl;
-            << ", fill clover = "               << (t4 - t3) / 1e6
+  std::cout << GridLogDebug << "instantiation =              " << (t5 - t4) / 1e6 << std::endl;
-            << ", misc = "                      << (t5 - t4) / 1e6
+  std::cout << GridLogDebug << "pick cbs =                   " << (t6 - t5) / 1e6 << std::endl;
-            << ", inversions = "                << (t6 - t5) / 1e6
+  std::cout << GridLogDebug << "total =                      " << (t6 - t0) / 1e6 << std::endl;
-            << ", pick cbs = "                  << (t7 - t6) / 1e6
-            << ", total = "                     << (t7 - t0) / 1e6
-            << std::endl;
-#endif
 }
 
-template <class Impl>
+template<class Impl, class CloverHelpers>
-void WilsonCloverFermion<Impl>::Mooee(const FermionField &in, FermionField &out)
+void WilsonCloverFermion<Impl, CloverHelpers>::Mooee(const FermionField &in, FermionField &out)
 {
   this->MooeeInternal(in, out, DaggerNo, InverseNo);
 }
 
-template <class Impl>
+template<class Impl, class CloverHelpers>
-void WilsonCloverFermion<Impl>::MooeeDag(const FermionField &in, FermionField &out)
+void WilsonCloverFermion<Impl, CloverHelpers>::MooeeDag(const FermionField &in, FermionField &out)
 {
   this->MooeeInternal(in, out, DaggerYes, InverseNo);
 }
 
-template <class Impl>
+template<class Impl, class CloverHelpers>
-void WilsonCloverFermion<Impl>::MooeeInv(const FermionField &in, FermionField &out)
+void WilsonCloverFermion<Impl, CloverHelpers>::MooeeInv(const FermionField &in, FermionField &out)
 {
   this->MooeeInternal(in, out, DaggerNo, InverseYes);
 }
 
-template <class Impl>
+template<class Impl, class CloverHelpers>
-void WilsonCloverFermion<Impl>::MooeeInvDag(const FermionField &in, FermionField &out)
+void WilsonCloverFermion<Impl, CloverHelpers>::MooeeInvDag(const FermionField &in, FermionField &out)
 {
   this->MooeeInternal(in, out, DaggerYes, InverseYes);
 }
 
-template <class Impl>
+template<class Impl, class CloverHelpers>
-void WilsonCloverFermion<Impl>::MooeeInternal(const FermionField &in, FermionField &out, int dag, int inv)
+void WilsonCloverFermion<Impl, CloverHelpers>::MooeeInternal(const FermionField &in, FermionField &out, int dag, int inv)
 {
   out.Checkerboard() = in.Checkerboard();
   CloverField *Clover;
@@ -278,8 +238,8 @@ void WilsonCloverFermion<Impl>::MooeeInternal(const FermionField &in, FermionFie
 } // MooeeInternal
 
 // Derivative parts unpreconditioned pseudofermions
-template <class Impl>
+template<class Impl, class CloverHelpers>
-void WilsonCloverFermion<Impl>::MDeriv(GaugeField &force, const FermionField &X, const FermionField &Y, int dag)
+void WilsonCloverFermion<Impl, CloverHelpers>::MDeriv(GaugeField &force, const FermionField &X, const FermionField &Y, int dag)
 {
   conformable(X.Grid(), Y.Grid());
   conformable(X.Grid(), force.Grid());
@@ -349,7 +309,7 @@ void WilsonCloverFermion<Impl>::MDeriv(GaugeField &force, const FermionField &X,
       }
       PropagatorField Slambda = Gamma(sigma[count]) * Lambda; // sigma checked
       Impl::TraceSpinImpl(lambda, Slambda);                   // traceSpin ok
-      force_mu -= factor*Helpers::Cmunu(U, lambda, mu, nu);                   // checked
+      force_mu -= factor*CloverHelpers::Cmunu(U, lambda, mu, nu);                   // checked
       count++;
     }
 
@@ -360,15 +320,15 @@ void WilsonCloverFermion<Impl>::MDeriv(GaugeField &force, const FermionField &X,
 }
 
 // Derivative parts
-template <class Impl>
+template<class Impl, class CloverHelpers>
-void WilsonCloverFermion<Impl>::MooDeriv(GaugeField &mat, const FermionField &X, const FermionField &Y, int dag)
+void WilsonCloverFermion<Impl, CloverHelpers>::MooDeriv(GaugeField &mat, const FermionField &X, const FermionField &Y, int dag)
 {
   assert(0);
 }
 
 // Derivative parts
-template <class Impl>
+template<class Impl, class CloverHelpers>
-void WilsonCloverFermion<Impl>::MeeDeriv(GaugeField &mat, const FermionField &U, const FermionField &V, int dag)
+void WilsonCloverFermion<Impl, CloverHelpers>::MeeDeriv(GaugeField &mat, const FermionField &U, const FermionField &V, int dag)
 {
   assert(0); // not implemented yet
 }

Grid/qcd/action/fermion/implementation/WilsonFermionImplementation.h

@@ -4,12 +4,13 @@ Grid physics library, www.github.com/paboyle/Grid
 
 Source file: ./lib/qcd/action/fermion/WilsonFermion.cc
 
-Copyright (C) 2015
+Copyright (C) 2022
 
 Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
 Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
 Author: paboyle <paboyle@ph.ed.ac.uk>
+Author: Fabian Joswig <fabian.joswig@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
@@ -599,11 +600,47 @@ void WilsonFermion<Impl>::ContractConservedCurrent(PropagatorField &q_in_1,
                                                    Current curr_type,
                                                    unsigned int mu)
 {
+  if(curr_type != Current::Vector)
+  {
+    std::cout << GridLogError << "Only the conserved vector current is implemented so far." << std::endl;
+    exit(1);
+  }
+
   Gamma g5(Gamma::Algebra::Gamma5);
   conformable(_grid, q_in_1.Grid());
   conformable(_grid, q_in_2.Grid());
   conformable(_grid, q_out.Grid());
-  assert(0);
+  auto UGrid= this->GaugeGrid();
+
+  PropagatorField tmp_shifted(UGrid);
+  PropagatorField g5Lg5(UGrid);
+  PropagatorField R(UGrid);
+  PropagatorField gmuR(UGrid);
+
+    Gamma::Algebra Gmu [] = {
+    Gamma::Algebra::GammaX,
+    Gamma::Algebra::GammaY,
+    Gamma::Algebra::GammaZ,
+    Gamma::Algebra::GammaT,
+  };
+  Gamma gmu=Gamma(Gmu[mu]);
+
+  g5Lg5=g5*q_in_1*g5;
+  tmp_shifted=Cshift(q_in_2,mu,1);
+  Impl::multLinkField(R,this->Umu,tmp_shifted,mu);
+  gmuR=gmu*R;
+
+  q_out=adj(g5Lg5)*R;
+  q_out-=adj(g5Lg5)*gmuR;
+
+  tmp_shifted=Cshift(q_in_1,mu,1);
+  Impl::multLinkField(g5Lg5,this->Umu,tmp_shifted,mu);
+  g5Lg5=g5*g5Lg5*g5;
+  R=q_in_2;
+  gmuR=gmu*R;
+
+  q_out-=adj(g5Lg5)*R;
+  q_out-=adj(g5Lg5)*gmuR;
 }
 
 
@@ -617,9 +654,51 @@ void WilsonFermion<Impl>::SeqConservedCurrent(PropagatorField &q_in,
                                               unsigned int tmax,
 					      ComplexField &lattice_cmplx)
 {
+  if(curr_type != Current::Vector)
+  {
+    std::cout << GridLogError << "Only the conserved vector current is implemented so far." << std::endl;
+    exit(1);
+  }
+
+  int tshift = (mu == Nd-1) ? 1 : 0;
+  unsigned int LLt    = GridDefaultLatt()[Tp];
   conformable(_grid, q_in.Grid());
   conformable(_grid, q_out.Grid());
-  assert(0);
+  auto UGrid= this->GaugeGrid();
+
+  PropagatorField tmp(UGrid);
+  PropagatorField Utmp(UGrid);
+  PropagatorField L(UGrid);
+  PropagatorField zz (UGrid);
+  zz=Zero();
+  LatticeInteger lcoor(UGrid); LatticeCoordinate(lcoor,Nd-1);
+
+    Gamma::Algebra Gmu [] = {
+    Gamma::Algebra::GammaX,
+    Gamma::Algebra::GammaY,
+    Gamma::Algebra::GammaZ,
+    Gamma::Algebra::GammaT,
+  };
+  Gamma gmu=Gamma(Gmu[mu]);
+
+  tmp = Cshift(q_in,mu,1);
+  Impl::multLinkField(Utmp,this->Umu,tmp,mu);
+  tmp = ( Utmp*lattice_cmplx - gmu*Utmp*lattice_cmplx ); // Forward hop
+  tmp = where((lcoor>=tmin),tmp,zz); // Mask the time
+  q_out = where((lcoor<=tmax),tmp,zz); // Position of current complicated
+
+  tmp = q_in *lattice_cmplx;
+  tmp = Cshift(tmp,mu,-1);
+  Impl::multLinkField(Utmp,this->Umu,tmp,mu+Nd); // Adjoint link
+  tmp = -( Utmp + gmu*Utmp );
+  // Mask the time
+  if (tmax == LLt - 1 && tshift == 1){ // quick fix to include timeslice 0 if tmax + tshift is over the last timeslice
+    unsigned int t0 = 0;
+    tmp = where(((lcoor==t0) || (lcoor>=tmin+tshift)),tmp,zz);
+  } else {
+    tmp = where((lcoor>=tmin+tshift),tmp,zz);
+  }
+  q_out+= where((lcoor<=tmax+tshift),tmp,zz); // Position of current complicated
 }
 
 NAMESPACE_END(Grid);

Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h

@@ -498,6 +498,7 @@ void WilsonKernels<Impl>::DhopKernel(int Opt,StencilImpl &st,  DoubledGaugeField
 #ifndef GRID_CUDA
      if (Opt == WilsonKernelsStatic::OptInlineAsm  ) {  ASM_CALL(AsmDhopSiteDag);     return;}
 #endif
+     acceleratorFenceComputeStream();
    } else if( interior ) {
      if (Opt == WilsonKernelsStatic::OptGeneric    ) { KERNEL_CALL(GenericDhopSiteDagInt); return;}
      if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSiteDagInt);    return;}
@@ -505,11 +506,13 @@ void WilsonKernels<Impl>::DhopKernel(int Opt,StencilImpl &st,  DoubledGaugeField
      if (Opt == WilsonKernelsStatic::OptInlineAsm  ) {  ASM_CALL(AsmDhopSiteDagInt);     return;}
 #endif
    } else if( exterior ) {
+     acceleratorFenceComputeStream();
      if (Opt == WilsonKernelsStatic::OptGeneric    ) { KERNEL_CALL(GenericDhopSiteDagExt); return;}
      if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSiteDagExt);    return;}
 #ifndef GRID_CUDA
      if (Opt == WilsonKernelsStatic::OptInlineAsm  ) {  ASM_CALL(AsmDhopSiteDagExt);     return;}
 #endif
+     acceleratorFenceComputeStream();
    }
    assert(0 && " Kernel optimisation case not covered ");
   }

Grid/qcd/action/fermion/instantiation/CompactWilsonCloverFermionInstantiation.cc.master

@@ -9,6 +9,7 @@
     Author: paboyle <paboyle@ph.ed.ac.uk>
     Author: Guido Cossu <guido.cossu@ed.ac.uk>
     Author: Daniel Richtmann <daniel.richtmann@gmail.com>
+    Author: Mattia Bruno <mattia.bruno@cern.ch>
 
     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
@@ -32,10 +33,12 @@
 #include <Grid/qcd/spin/Dirac.h>
 #include <Grid/qcd/action/fermion/CompactWilsonCloverFermion.h>
 #include <Grid/qcd/action/fermion/implementation/CompactWilsonCloverFermionImplementation.h>
+#include <Grid/qcd/action/fermion/CloverHelpers.h>
 
 NAMESPACE_BEGIN(Grid);
 
 #include "impl.h"
-template class CompactWilsonCloverFermion<IMPLEMENTATION>; 
+template class CompactWilsonCloverFermion<IMPLEMENTATION, CompactCloverHelpers<IMPLEMENTATION>>; 
+template class CompactWilsonCloverFermion<IMPLEMENTATION, CompactExpCloverHelpers<IMPLEMENTATION>>; 
 
 NAMESPACE_END(Grid);

Grid/qcd/action/fermion/instantiation/WilsonAdjImplD/WilsonKernelsInstantiationWilsonAdjImplD.cc

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-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>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/instantiation/WilsonAdjImplD/WilsonKernelsInstantiationWilsonAdjImplD.cc

@@ -0,0 +1 @@
+../WilsonKernelsInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/WilsonAdjImplF/WilsonKernelsInstantiationWilsonAdjImplF.cc

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-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>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/instantiation/WilsonAdjImplF/WilsonKernelsInstantiationWilsonAdjImplF.cc

@@ -0,0 +1 @@
+../WilsonKernelsInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/WilsonCloverFermionInstantiation.cc.master

@@ -8,6 +8,7 @@
 
     Author: paboyle <paboyle@ph.ed.ac.uk>
     Author: Guido Cossu <guido.cossu@ed.ac.uk>
+    Author: Mattia Bruno <mattia.bruno@cern.ch>
     
     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
@@ -31,10 +32,12 @@
 #include <Grid/qcd/spin/Dirac.h>
 #include <Grid/qcd/action/fermion/WilsonCloverFermion.h>
 #include <Grid/qcd/action/fermion/implementation/WilsonCloverFermionImplementation.h>
+#include <Grid/qcd/action/fermion/CloverHelpers.h>
 
 NAMESPACE_BEGIN(Grid);
 
 #include "impl.h"
-template class WilsonCloverFermion<IMPLEMENTATION>; 
+template class WilsonCloverFermion<IMPLEMENTATION, CloverHelpers<IMPLEMENTATION>>; 
+template class WilsonCloverFermion<IMPLEMENTATION, ExpCloverHelpers<IMPLEMENTATION>>; 
 
 NAMESPACE_END(Grid);

Grid/qcd/action/fermion/instantiation/WilsonImplD/WilsonKernelsInstantiationWilsonImplD.cc

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-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>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/instantiation/WilsonImplD/WilsonKernelsInstantiationWilsonImplD.cc

@@ -0,0 +1 @@
+../WilsonKernelsInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/WilsonImplF/WilsonKernelsInstantiationWilsonImplF.cc

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-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>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/instantiation/WilsonImplF/WilsonKernelsInstantiationWilsonImplF.cc

@@ -0,0 +1 @@
+../WilsonKernelsInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplD/WilsonKernelsInstantiationWilsonTwoIndexAntiSymmetricImplD.cc

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-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>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplD/WilsonKernelsInstantiationWilsonTwoIndexAntiSymmetricImplD.cc

@@ -0,0 +1 @@
+../WilsonKernelsInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplF/WilsonKernelsInstantiationWilsonTwoIndexAntiSymmetricImplF.cc

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-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>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplF/WilsonKernelsInstantiationWilsonTwoIndexAntiSymmetricImplF.cc

@@ -0,0 +1 @@
+../WilsonKernelsInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplD/WilsonKernelsInstantiationWilsonTwoIndexSymmetricImplD.cc

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-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>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplD/WilsonKernelsInstantiationWilsonTwoIndexSymmetricImplD.cc

@@ -0,0 +1 @@
+../WilsonKernelsInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplF/WilsonKernelsInstantiationWilsonTwoIndexSymmetricImplF.cc

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-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>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplF/WilsonKernelsInstantiationWilsonTwoIndexSymmetricImplF.cc

@@ -0,0 +1 @@
+../WilsonKernelsInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/ZWilsonImplD/WilsonKernelsInstantiationZWilsonImplD.cc

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-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>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/instantiation/ZWilsonImplD/WilsonKernelsInstantiationZWilsonImplD.cc

@@ -0,0 +1 @@
+../WilsonKernelsInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/ZWilsonImplF/WilsonKernelsInstantiationZWilsonImplF.cc

@@ -1,51 +0,0 @@
-/*************************************************************************************
-
-Grid physics library, www.github.com/paboyle/Grid
-
-Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
-
-Copyright (C) 2015, 2020
-
-Author: Peter Boyle <paboyle@ph.ed.ac.uk>
-Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
-Author: paboyle <paboyle@ph.ed.ac.uk>
-Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
-
-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>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
-
-#ifndef AVX512
-#ifndef QPX
-#ifndef A64FX
-#ifndef A64FXFIXEDSIZE
-#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
-#endif
-#endif
-#endif
-#endif
-
-NAMESPACE_BEGIN(Grid);
-
-#include "impl.h"
-template class WilsonKernels<IMPLEMENTATION>;
-
-NAMESPACE_END(Grid);

Grid/qcd/action/fermion/instantiation/ZWilsonImplF/WilsonKernelsInstantiationZWilsonImplF.cc

@@ -0,0 +1 @@
+../WilsonKernelsInstantiation.cc.master

Grid/qcd/action/fermion/instantiation/generate_instantiations.sh

@@ -18,6 +18,10 @@ WILSON_IMPL_LIST=" \
 	   GparityWilsonImplF \
 	   GparityWilsonImplD "
 
+COMPACT_WILSON_IMPL_LIST=" \
+	   WilsonImplF \
+	   WilsonImplD "
+
 DWF_IMPL_LIST=" \
 	   WilsonImplF \
 	   WilsonImplD \
@@ -40,7 +44,7 @@ EOF
 
 done
 
-CC_LIST="WilsonCloverFermionInstantiation CompactWilsonCloverFermionInstantiation WilsonFermionInstantiation WilsonKernelsInstantiation WilsonTMFermionInstantiation"
+CC_LIST="WilsonCloverFermionInstantiation WilsonFermionInstantiation WilsonKernelsInstantiation WilsonTMFermionInstantiation"
 
 for impl in $WILSON_IMPL_LIST
 do
@@ -50,6 +54,16 @@ do
 done
 done
 
+CC_LIST="CompactWilsonCloverFermionInstantiation"
+
+for impl in $COMPACT_WILSON_IMPL_LIST
+do
+for f in $CC_LIST
+do
+  ln -f -s ../$f.cc.master $impl/$f$impl.cc
+done
+done
+
 CC_LIST=" \
   CayleyFermion5DInstantiation \
   ContinuedFractionFermion5DInstantiation \

Grid/qcd/action/gauge/Photon.h

@@ -49,7 +49,7 @@ NAMESPACE_BEGIN(Grid);
     
     typedef Lattice<SiteLink>  LinkField;
     typedef Lattice<SiteField> Field;
-    typedef Field              ComplexField;
+    typedef LinkField          ComplexField;
   };
   
   typedef QedGImpl<vComplex> QedGImplR;

Grid/qcd/utils/GaugeFix.h

@@ -55,12 +55,12 @@ public:
     }
   }  
 
-  static void SteepestDescentGaugeFix(GaugeLorentz &Umu,Real & alpha,int maxiter,Real Omega_tol, Real Phi_tol,bool Fourier=false,int orthog=-1) {
+  static void SteepestDescentGaugeFix(GaugeLorentz &Umu,Real & alpha,int maxiter,Real Omega_tol, Real Phi_tol,bool Fourier=false,int orthog=-1,bool err_on_no_converge=true) {
     GridBase *grid = Umu.Grid();
     GaugeMat xform(grid);
-    SteepestDescentGaugeFix(Umu,xform,alpha,maxiter,Omega_tol,Phi_tol,Fourier,orthog);
+    SteepestDescentGaugeFix(Umu,xform,alpha,maxiter,Omega_tol,Phi_tol,Fourier,orthog,err_on_no_converge);
   }
-  static void SteepestDescentGaugeFix(GaugeLorentz &Umu,GaugeMat &xform,Real & alpha,int maxiter,Real Omega_tol, Real Phi_tol,bool Fourier=false,int orthog=-1) {
+  static void SteepestDescentGaugeFix(GaugeLorentz &Umu,GaugeMat &xform,Real & alpha,int maxiter,Real Omega_tol, Real Phi_tol,bool Fourier=false,int orthog=-1,bool err_on_no_converge=true) {
 
     GridBase *grid = Umu.Grid();
 
@@ -122,6 +122,8 @@ public:
 
       }
     }
+    std::cout << GridLogError << "Gauge fixing did not converge in " << maxiter << " iterations." << std::endl;
+    if (err_on_no_converge) assert(0);
   };
   static Real SteepestDescentStep(std::vector<GaugeMat> &U,GaugeMat &xform,Real & alpha, GaugeMat & dmuAmu,int orthog) {
     GridBase *grid = U[0].Grid();

Grid/qcd/utils/WilsonLoops.h

@@ -125,7 +125,6 @@ public:
     return sumplaq / vol / faces / Nc; // Nd , Nc dependent... FIXME
   }
 
-
   //////////////////////////////////////////////////
   // average over all x,y,z the temporal loop
   //////////////////////////////////////////////////
@@ -165,7 +164,7 @@ public:
 
     double vol = Umu.Grid()->gSites();
 
-    return p.real() / vol / 4.0 / 3.0;
+    return p.real() / vol / (4.0 * Nc ) ;
   };
 
   //////////////////////////////////////////////////

Grid/serialisation/JSON_IO.cc

@@ -26,7 +26,7 @@
     *************************************************************************************/
     /*  END LEGAL */
 #include <Grid/Grid.h>
-#if (!defined(GRID_CUDA)) && (!defined(GRID_HIP))
+#ifndef GRID_HIP
 
 NAMESPACE_BEGIN(Grid);
 
@@ -82,7 +82,7 @@ void JSONWriter::writeDefault(const std::string &s,	const std::string &x)
   if (s.size())
     ss_ << "\""<< s << "\" : \"" << os.str() << "\" ," ;
   else
-    ss_ << os.str() << " ," ;
+    ss_ << "\""<< os.str() << "\" ," ;
 }
 
 // Reader implementation ///////////////////////////////////////////////////////

Grid/serialisation/JSON_IO.h

@@ -54,7 +54,7 @@ namespace Grid
     void pop(void);
     template <typename U>
     void writeDefault(const std::string &s, const U &x);
-#ifdef __NVCC__
+#if defined(GRID_CUDA) || defined(GRID_HIP)
     void writeDefault(const std::string &s, const Grid::ComplexD &x) 
     { 
       std::complex<double> z(real(x),imag(x));
@@ -101,7 +101,7 @@ namespace Grid
     void readDefault(const std::string &s, std::vector<U> &output);
     template <typename U, typename P>
     void readDefault(const std::string &s, std::pair<U,P> &output);
-#ifdef __NVCC__
+#if defined(GRID_CUDA) || defined(GRID_HIP)
     void readDefault(const std::string &s, ComplexD &output)
     { 
       std::complex<double> z;

Grid/serialisation/Serialisation.h

@@ -36,7 +36,7 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 #include "BinaryIO.h"
 #include "TextIO.h"
 #include "XmlIO.h"
-#if (!defined(GRID_CUDA)) && (!defined(GRID_HIP))
+#ifndef GRID_HIP
 #include "JSON_IO.h"
 #endif
 

Grid/stencil/Stencil.h

@@ -80,11 +80,14 @@ void Gather_plane_simple_table (commVector<std::pair<int,int> >& table,const Lat
 ///////////////////////////////////////////////////////////////////
 template<class cobj,class vobj,class compressor>
 void Gather_plane_exchange_table(const Lattice<vobj> &rhs,
-				 commVector<cobj *> pointers,int dimension,int plane,int cbmask,compressor &compress,int type) __attribute__((noinline));
+				 commVector<cobj *> pointers,
+				 int dimension,int plane,
+				 int cbmask,compressor &compress,int type) __attribute__((noinline));
 
 template<class cobj,class vobj,class compressor>
-void Gather_plane_exchange_table(commVector<std::pair<int,int> >& table,const Lattice<vobj> &rhs,
+void Gather_plane_exchange_table(commVector<std::pair<int,int> >& table,
-				 Vector<cobj *> pointers,int dimension,int plane,int cbmask,
+				 const Lattice<vobj> &rhs,
+				 std::vector<cobj *> &pointers,int dimension,int plane,int cbmask,
 				 compressor &compress,int type)
 {
   assert( (table.size()&0x1)==0);
@@ -92,11 +95,12 @@ void Gather_plane_exchange_table(commVector<std::pair<int,int> >& table,const La
   int so  = plane*rhs.Grid()->_ostride[dimension]; // base offset for start of plane
 
   auto rhs_v = rhs.View(AcceleratorRead);
+  auto rhs_p = &rhs_v[0];
   auto p0=&pointers[0][0];
   auto p1=&pointers[1][0];
   auto tp=&table[0];
   accelerator_forNB(j, num, vobj::Nsimd(), {
-      compress.CompressExchange(p0,p1, &rhs_v[0], j,
+      compress.CompressExchange(p0,p1, rhs_p, j,
 				so+tp[2*j  ].second,
 				so+tp[2*j+1].second,
 				type);
@@ -230,8 +234,8 @@ public:
   };
   struct Merge {
     cobj * mpointer;
-    Vector<scalar_object *> rpointers;
+    //    std::vector<scalar_object *> rpointers;
-    Vector<cobj *> vpointers;
+    std::vector<cobj *> vpointers;
     Integer buffer_size;
     Integer type;
   };
@@ -406,6 +410,7 @@ public:
       comms_bytes+=bytes;
       shm_bytes  +=2*Packets[i].bytes-bytes;
     }
+    _grid->StencilBarrier();// Synch shared memory on a single nodes
   }
 
   void CommunicateComplete(std::vector<std::vector<CommsRequest_t> > &reqs)
@@ -420,7 +425,7 @@ public:
   ////////////////////////////////////////////////////////////////////////
   void Communicate(void)
   {
-    if ( CartesianCommunicator::CommunicatorPolicy == CartesianCommunicator::CommunicatorPolicySequential ){
+    if ( 0 ){
       thread_region {
 	// must be called in parallel region
 	int mythread  = thread_num();
@@ -569,7 +574,7 @@ public:
     d.buffer_size = buffer_size;
     dv.push_back(d);
   }
-  void AddMerge(cobj *merge_p,Vector<cobj *> &rpointers,Integer buffer_size,Integer type,std::vector<Merge> &mv) {
+  void AddMerge(cobj *merge_p,std::vector<cobj *> &rpointers,Integer buffer_size,Integer type,std::vector<Merge> &mv) {
     Merge m;
     m.type     = type;
     m.mpointer = merge_p;
@@ -582,6 +587,7 @@ public:
   }
   template<class decompressor>  void CommsMergeSHM(decompressor decompress) {
     mpi3synctime-=usecond();
+    accelerator_barrier();
     _grid->StencilBarrier();// Synch shared memory on a single nodes
     mpi3synctime+=usecond();
     shmmergetime-=usecond();
@@ -1114,8 +1120,8 @@ public:
     int bytes = (reduced_buffer_size*datum_bytes)/simd_layout;
     assert(bytes*simd_layout == reduced_buffer_size*datum_bytes);
 
-    Vector<cobj *> rpointers(maxl);
+    std::vector<cobj *> rpointers(maxl);
-    Vector<cobj *> spointers(maxl);
+    std::vector<cobj *> spointers(maxl);
 
     ///////////////////////////////////////////
     // Work out what to send where

Grid/threads/Accelerator.cc

@@ -195,12 +195,15 @@ void acceleratorInit(void)
 #ifdef GRID_SYCL
 
 cl::sycl::queue *theGridAccelerator;
+cl::sycl::queue *theCopyAccelerator;
 void acceleratorInit(void)
 {
   int nDevices = 1;
   cl::sycl::gpu_selector selector;
   cl::sycl::device selectedDevice { selector };
   theGridAccelerator = new sycl::queue (selectedDevice);
+  //  theCopyAccelerator = new sycl::queue (selectedDevice);
+  theCopyAccelerator = theGridAccelerator; // Should proceed concurrenlty anyway.
 
 #ifdef GRID_SYCL_LEVEL_ZERO_IPC
   zeInit(0);

Grid/threads/Accelerator.h

@@ -247,7 +247,6 @@ inline int  acceleratorIsCommunicable(void *ptr)
 //////////////////////////////////////////////
 // SyCL acceleration
 //////////////////////////////////////////////
-
 #ifdef GRID_SYCL
 NAMESPACE_END(Grid);
 #include <CL/sycl.hpp>
@@ -262,6 +261,7 @@ NAMESPACE_END(Grid);
 NAMESPACE_BEGIN(Grid);
 
 extern cl::sycl::queue *theGridAccelerator;
+extern cl::sycl::queue *theCopyAccelerator;
 
 #ifdef __SYCL_DEVICE_ONLY__
 #define GRID_SIMT
@@ -289,7 +289,7 @@ accelerator_inline int acceleratorSIMTlane(int Nsimd) {
       cgh.parallel_for(					\
       cl::sycl::nd_range<3>(global,local), \
       [=] (cl::sycl::nd_item<3> item) /*mutable*/     \
-      [[intel::reqd_sub_group_size(8)]]	      \
+      [[intel::reqd_sub_group_size(16)]]	      \
       {						      \
       auto iter1    = item.get_global_id(0);	      \
       auto iter2    = item.get_global_id(1);	      \
@@ -298,19 +298,19 @@ accelerator_inline int acceleratorSIMTlane(int Nsimd) {
      });	   			              \
     });
 
-#define accelerator_barrier(dummy) theGridAccelerator->wait();
+#define accelerator_barrier(dummy) { theGridAccelerator->wait(); }
 
 inline void *acceleratorAllocShared(size_t bytes){ return malloc_shared(bytes,*theGridAccelerator);};
 inline void *acceleratorAllocDevice(size_t bytes){ return malloc_device(bytes,*theGridAccelerator);};
 inline void acceleratorFreeShared(void *ptr){free(ptr,*theGridAccelerator);};
 inline void acceleratorFreeDevice(void *ptr){free(ptr,*theGridAccelerator);};
-inline void acceleratorCopyDeviceToDeviceAsynch(void *from,void *to,size_t bytes)  {
+
-  theGridAccelerator->memcpy(to,from,bytes);
+inline void acceleratorCopySynchronise(void) {  theCopyAccelerator->wait(); }
-}
+inline void acceleratorCopyDeviceToDeviceAsynch(void *from,void *to,size_t bytes)  {  theCopyAccelerator->memcpy(to,from,bytes);}
-inline void acceleratorCopySynchronise(void) {  theGridAccelerator->wait(); std::cout<<"acceleratorCopySynchronise() wait "<<std::endl; }
+inline void acceleratorCopyToDevice(void *from,void *to,size_t bytes)  { theCopyAccelerator->memcpy(to,from,bytes); theCopyAccelerator->wait();}
-inline void acceleratorCopyToDevice(void *from,void *to,size_t bytes)  { theGridAccelerator->memcpy(to,from,bytes); theGridAccelerator->wait();}
+inline void acceleratorCopyFromDevice(void *from,void *to,size_t bytes){ theCopyAccelerator->memcpy(to,from,bytes); theCopyAccelerator->wait();}
-inline void acceleratorCopyFromDevice(void *from,void *to,size_t bytes){ theGridAccelerator->memcpy(to,from,bytes); theGridAccelerator->wait();}
+inline void acceleratorMemSet(void *base,int value,size_t bytes) { theCopyAccelerator->memset(base,value,bytes); theCopyAccelerator->wait();}
-inline void acceleratorMemSet(void *base,int value,size_t bytes) { theGridAccelerator->memset(base,value,bytes); theGridAccelerator->wait();}
+
 inline int  acceleratorIsCommunicable(void *ptr)
 {
 #if 0
@@ -481,9 +481,10 @@ inline void acceleratorCopySynchronise(void) { hipStreamSynchronize(copyStream);
 #define accelerator_for2d(iter1, num1, iter2, num2, nsimd, ... ) thread_for2d(iter1,num1,iter2,num2,{ __VA_ARGS__ });
 
 accelerator_inline int acceleratorSIMTlane(int Nsimd) { return 0; } // CUDA specific
-inline void acceleratorCopyToDevice(void *from,void *to,size_t bytes)  { memcpy(to,from,bytes);}
+
-inline void acceleratorCopyFromDevice(void *from,void *to,size_t bytes){ memcpy(to,from,bytes);}
+inline void acceleratorCopyToDevice(void *from,void *to,size_t bytes)  { thread_bcopy(from,to,bytes); }
-inline void acceleratorCopyDeviceToDeviceAsynch(void *from,void *to,size_t bytes)  { memcpy(to,from,bytes);}
+inline void acceleratorCopyFromDevice(void *from,void *to,size_t bytes){ thread_bcopy(from,to,bytes);}
+inline void acceleratorCopyDeviceToDeviceAsynch(void *from,void *to,size_t bytes)  { thread_bcopy(from,to,bytes);}
 inline void acceleratorCopySynchronise(void) {};
 
 inline int  acceleratorIsCommunicable(void *ptr){ return 1; }
@@ -510,7 +511,16 @@ inline void *acceleratorAllocCpu(size_t bytes){return memalign(GRID_ALLOC_ALIGN,
 inline void acceleratorFreeCpu  (void *ptr){free(ptr);};
 #endif
 
+//////////////////////////////////////////////
+// Fencing needed ONLY for SYCL
+//////////////////////////////////////////////
 
+#ifdef GRID_SYCL
+inline void acceleratorFenceComputeStream(void){ accelerator_barrier();};
+#else
+// Ordering within a stream guaranteed on Nvidia & AMD
+inline void acceleratorFenceComputeStream(void){ };
+#endif
 
 ///////////////////////////////////////////////////
 // Synchronise across local threads for divergence resynch

Grid/threads/Threads.h

@@ -72,3 +72,20 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
 #define thread_region                                       DO_PRAGMA(omp parallel)
 #define thread_critical                                     DO_PRAGMA(omp critical)
 
+#ifdef GRID_OMP
+inline void thread_bcopy(void *from, void *to,size_t bytes)
+{
+  uint64_t *ufrom = (uint64_t *)from;
+  uint64_t *uto   = (uint64_t *)to;
+  assert(bytes%8==0);
+  uint64_t words=bytes/8;
+  thread_for(w,words,{
+      uto[w] = ufrom[w];
+  });
+}
+#else
+inline void thread_bcopy(void *from, void *to,size_t bytes)
+{
+  bcopy(from,to,bytes);
+}
+#endif

Grid/util/Sha.h

@@ -27,6 +27,7 @@
     /*  END LEGAL */
 extern "C" {
 #include <openssl/sha.h>
+#include <openssl/evp.h>
 }
 #ifdef USE_IPP
 #include "ipp.h"
@@ -70,10 +71,8 @@ public:
   static inline std::vector<unsigned char> sha256(const void *data,size_t bytes)
   {
     std::vector<unsigned char> hash(SHA256_DIGEST_LENGTH);
-    SHA256_CTX sha256;
+    auto digest = EVP_get_digestbyname("SHA256");
-    SHA256_Init  (&sha256);
+    EVP_Digest(data, bytes, &hash[0], NULL, digest, NULL);
-    SHA256_Update(&sha256, data,bytes);
-    SHA256_Final (&hash[0], &sha256);
     return hash;
   }
   static inline std::vector<int> sha256_seeds(const std::string &s)

README.md

@@ -148,7 +148,7 @@ If you want to build all the tests at once just use `make tests`.
 - `--enable-mkl[=<path>]`: use Intel MKL for FFT (and LAPACK if enabled) routines. A UNIX prefix containing the library can be specified (optional).
 - `--enable-numa`: enable NUMA first touch optimisation
 - `--enable-simd=<code>`: setup Grid for the SIMD target `<code>` (default: `GEN`). A list of possible SIMD targets is detailed in a section below.
-- `--enable-gen-simd-width=<size>`: select the size (in bytes) of the generic SIMD vector type (default: 32 bytes).
+- `--enable-gen-simd-width=<size>`: select the size (in bytes) of the generic SIMD vector type (default: 64 bytes).
 - `--enable-comms=<comm>`: Use `<comm>` for message passing (default: `none`). A list of possible SIMD targets is detailed in a section below.
 - `--enable-rng={sitmo|ranlux48|mt19937}`: choose the RNG (default: `sitmo `).
 - `--disable-timers`: disable system dependent high-resolution timers.

benchmarks/Benchmark_halo.cc

@@ -0,0 +1,131 @@
+ /*************************************************************************************
+    Grid physics library, www.github.com/paboyle/Grid
+    Source file: ./benchmarks/Benchmark_dwf.cc
+    Copyright (C) 2015
+
+    Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+    Author: paboyle <paboyle@ph.ed.ac.uk>
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+    You should have received a copy of the GNU General Public License along
+    with this program; if not, write to the Free Software Foundation, Inc.,
+    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+    See the full license in the file "LICENSE" in the top level distribution directory
+    *************************************************************************************/
+    /*  END LEGAL */
+#include <Grid/Grid.h>
+#ifdef GRID_CUDA
+#define CUDA_PROFILE
+#endif
+
+#ifdef CUDA_PROFILE
+#include <cuda_profiler_api.h>
+#endif
+
+using namespace std;
+using namespace Grid;
+
+template<class d>
+struct scal {
+  d internal;
+};
+
+  Gamma::Algebra Gmu [] = {
+    Gamma::Algebra::GammaX,
+    Gamma::Algebra::GammaY,
+    Gamma::Algebra::GammaZ,
+    Gamma::Algebra::GammaT
+  };
+
+
+int main (int argc, char ** argv)
+{
+  Grid_init(&argc,&argv);
+
+  Coordinate latt4= GridDefaultLatt();
+  Coordinate mpi  = GridDefaultMpi();
+  Coordinate simd = GridDefaultSimd(Nd,vComplexF::Nsimd());
+
+  GridLogLayout();
+
+  int Ls=16;
+  for(int i=0;i<argc;i++)
+    if(std::string(argv[i]) == "-Ls"){
+      std::stringstream ss(argv[i+1]); ss >> Ls;
+    }
+
+  
+  GridCartesian         * UGrid   = SpaceTimeGrid::makeFourDimGrid(latt4,simd ,mpi);
+  GridRedBlackCartesian * UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
+  GridCartesian         * FGrid   = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
+  GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid);
+
+  std::cout << GridLogMessage << "Making s innermost grids"<<std::endl;
+  GridCartesian         * sUGrid   = SpaceTimeGrid::makeFourDimDWFGrid(GridDefaultLatt(),GridDefaultMpi());
+  GridRedBlackCartesian * sUrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(sUGrid);
+  GridCartesian         * sFGrid   = SpaceTimeGrid::makeFiveDimDWFGrid(Ls,UGrid);
+  GridRedBlackCartesian * sFrbGrid = SpaceTimeGrid::makeFiveDimDWFRedBlackGrid(Ls,UGrid);
+
+  std::vector<int> seeds4({1,2,3,4});
+  std::vector<int> seeds5({5,6,7,8});
+
+  std::cout << GridLogMessage << "Initialising 4d RNG" << std::endl;
+  GridParallelRNG          RNG4(UGrid);  RNG4.SeedUniqueString(std::string("The 4D RNG"));
+  std::cout << GridLogMessage << "Initialising 5d RNG" << std::endl;
+  GridParallelRNG          RNG5(FGrid);  RNG5.SeedUniqueString(std::string("The 5D RNG"));
+  std::cout << GridLogMessage << "Initialised RNGs" << std::endl;
+
+  LatticeFermionF src   (FGrid); random(RNG5,src);
+  RealD N2 = 1.0/::sqrt(norm2(src));
+  src = src*N2;
+
+  std::cout << GridLogMessage << "Drawing gauge field" << std::endl;
+  LatticeGaugeFieldF Umu(UGrid);
+  SU<Nc>::HotConfiguration(RNG4,Umu);
+  std::cout << GridLogMessage << "Random gauge initialised " << std::endl;
+
+  RealD mass=0.1;
+  RealD M5  =1.8;
+
+  RealD NP = UGrid->_Nprocessors;
+  RealD NN = UGrid->NodeCount();
+
+  DomainWallFermionF Dw(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
+
+  const int ncall = 500;
+  std::cout << GridLogMessage<< "*********************************************************" <<std::endl;
+  std::cout << GridLogMessage<< "* Benchmarking DomainWallFermionF::HaloGatherOpt         "<<std::endl;
+  std::cout << GridLogMessage<< "*********************************************************" <<std::endl;
+  {
+    typename DomainWallFermionF::Compressor compressor(0);
+    FGrid->Barrier();
+    Dw.Stencil.HaloExchangeOptGather(src,compressor);
+    double t0=usecond();
+    for(int i=0;i<ncall;i++){
+      Dw.Stencil.HaloExchangeOptGather(src,compressor);
+    }
+    double t1=usecond();
+    FGrid->Barrier();
+
+    double bytes=0.0;
+    if(mpi[0]) bytes+=latt4[1]*latt4[2]*latt4[3];
+    if(mpi[1]) bytes+=latt4[0]*latt4[2]*latt4[3];
+    if(mpi[2]) bytes+=latt4[0]*latt4[1]*latt4[3];
+    if(mpi[3]) bytes+=latt4[0]*latt4[1]*latt4[2];
+    bytes = bytes * Ls * 8.* (24.+12.)* 2.0;
+
+    std::cout<<GridLogMessage << "Gather us /call =   "<< (t1-t0)/ncall<<std::endl;
+    std::cout<<GridLogMessage << "Gather MBs /call =   "<< bytes*ncall/(t1-t0)<<std::endl;
+
+  }
+
+  Grid_finalize();
+  exit(0);
+}

benchmarks/Benchmark_wilson_sweep.cc

@@ -44,6 +44,13 @@ void bench_wilson (
 		   double const     volume,
 		   int const           dag );
 
+void bench_wilson_eo (
+       LatticeFermion &    src,
+       LatticeFermion & result,
+       WilsonFermionR &     Dw,
+       double const     volume,
+       int const           dag );
+
 int main (int argc, char ** argv)
 {
   Grid_init(&argc,&argv);
@@ -110,8 +117,8 @@ int main (int argc, char ** argv)
 	  bench_wilson(src,result,Dw,volume,DaggerYes);
 	  std::cout << "\t";
     // EO
-	  bench_wilson(src,result,Dw,volume,DaggerNo);
+    bench_wilson_eo(src_o,result_e,Dw,volume,DaggerNo);
-	  bench_wilson(src,result,Dw,volume,DaggerYes);
+    bench_wilson_eo(src_o,result_e,Dw,volume,DaggerYes);
 	  std::cout << std::endl;
 	}
     }

configure.ac

@@ -159,7 +159,7 @@ case ${ac_ZMOBIUS} in
 esac
 ############### Nc
 AC_ARG_ENABLE([Nc],
-    [AC_HELP_STRING([--enable-Nc=2|3|4], [enable number of colours])],
+    [AC_HELP_STRING([--enable-Nc=2|3|4|5], [enable number of colours])],
     [ac_Nc=${enable_Nc}], [ac_Nc=3])
 
 case ${ac_Nc} in

systems/PVC/benchmarks/run-1tile.sh

@@ -0,0 +1,62 @@
+#!/bin/sh
+##SBATCH -p PVC-SPR-QZEH 
+##SBATCH -p PVC-ICX-QZNW
+#SBATCH -p QZ1J-ICX-PVC
+##SBATCH -p QZ1J-SPR-PVC-2C
+
+source /nfs/site/home/paboylex/ATS/GridNew/Grid/systems/PVC-nightly/setup.sh
+
+export NT=8
+
+export I_MPI_OFFLOAD=1
+export I_MPI_OFFLOAD_TOPOLIB=level_zero
+export I_MPI_OFFLOAD_DOMAIN_SIZE=-1
+
+# export IGC_EnableLSCFenceUGMBeforeEOT=0
+# export SYCL_PROGRAM_COMPILE_OPTIONS="-ze-opt-large-register-file=False"
+export SYCL_DEVICE_FILTER=gpu,level_zero
+#export IGC_ShaderDumpEnable=1 
+#export IGC_DumpToCurrentDir=1
+export I_MPI_OFFLOAD_CELL=tile
+export EnableImplicitScaling=0
+export EnableWalkerPartition=0
+export ZE_AFFINITY_MASK=0.0
+mpiexec -launcher ssh -n 1 -host localhost  ./Benchmark_dwf_fp32 --mpi 1.1.1.1 --grid 32.32.32.32 --accelerator-threads $NT --comms-sequential --shm-mpi 1 --device-mem 32768
+
+export ZE_AFFINITY_MASK=0
+export I_MPI_OFFLOAD_CELL=device
+export EnableImplicitScaling=1
+export EnableWalkerPartition=1
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+#mpiexec -launcher ssh -n 2 -host localhost  vtune -collect gpu-hotspots -knob gpu-sampling-interval=1 -data-limit=0 -r ./vtune_run4 -- ./wrap.sh ./Benchmark_dwf_fp32 --mpi 2.1.1.1 --grid 64.32.32.32 --accelerator-threads $NT --comms-overlap --shm-mpi 1
+
+#mpiexec  -launcher ssh -n 1 -host localhost ./wrap.sh ./Benchmark_dwf_fp32 --mpi 1.1.1.1 --grid 64.32.32.32 --accelerator-threads $NT --comms-overlap --shm-mpi 1
+
+#mpiexec  -launcher ssh -n 2 -host localhost ./wrap.sh ./Benchmark_dwf_fp32 --mpi 2.1.1.1 --grid 64.32.32.32 --accelerator-threads $NT --comms-sequential --shm-mpi 1
+
+#mpiexec  -launcher ssh -n 2 -host localhost ./wrap.sh ./Benchmark_dwf_fp32 --mpi 2.1.1.1 --grid 64.32.32.32 --accelerator-threads $NT --comms-overlap --shm-mpi 1
+
+#mpiexec  -launcher ssh -n 2 -host localhost ./wrap.sh ./Benchmark_dwf_fp32 --mpi 2.1.1.1 --grid 64.32.32.32 --accelerator-threads $NT --comms-sequential --shm-mpi 0
+
+#mpirun -np 2 ./wrap.sh ./Benchmark_dwf_fp32 --mpi 1.1.1.2 --grid 16.32.32.64 --accelerator-threads $NT --comms-sequential --shm-mpi 0
+#mpirun -np 2 ./wrap.sh ./Benchmark_dwf_fp32 --mpi 1.1.1.2 --grid 32.32.32.64 --accelerator-threads $NT --comms-sequential --shm-mpi 1
+

systems/PVC/benchmarks/run-2tile-mpi.sh

@@ -0,0 +1,34 @@
+#!/bin/bash
+##SBATCH -p PVC-SPR-QZEH 
+##SBATCH -p PVC-ICX-QZNW
+
+#SBATCH -p QZ1J-ICX-PVC
+
+source /nfs/site/home/paboylex/ATS/GridNew/Grid/systems/PVC-nightly/setup.sh
+
+export NT=16
+
+# export IGC_EnableLSCFenceUGMBeforeEOT=0
+# export SYCL_PROGRAM_COMPILE_OPTIONS="-ze-opt-large-register-file=False"
+#export IGC_ShaderDumpEnable=1 
+#export IGC_DumpToCurrentDir=1
+export I_MPI_OFFLOAD=1
+export I_MPI_OFFLOAD_TOPOLIB=level_zero
+export I_MPI_OFFLOAD_DOMAIN_SIZE=-1
+export SYCL_DEVICE_FILTER=gpu,level_zero
+export I_MPI_OFFLOAD_CELL=tile
+export EnableImplicitScaling=0
+export EnableWalkerPartition=0
+export SYCL_PI_LEVEL_ZERO_DEVICE_SCOPE_EVENTS=1
+export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1
+export SYCL_PI_LEVEL_ZERO_USE_COPY_ENGINE=0
+
+for i in 0 
+do
+mpiexec -launcher ssh -n 2 -host localhost  ./wrap4gpu.sh ./Benchmark_dwf_fp32 --mpi 1.1.1.2 --grid 32.32.32.64 --accelerator-threads $NT  --shm-mpi 1  --device-mem 32768
+mpiexec -launcher ssh -n 2 -host localhost  ./wrap4gpu.sh ./Benchmark_dwf_fp32 --mpi 2.1.1.1 --grid 64.32.32.32 --accelerator-threads $NT  --shm-mpi 1  --device-mem 32768
+done
+#mpiexec -launcher ssh -n 2 -host localhost  ./wrap4gpu.sh ./Benchmark_halo --mpi 1.1.1.2 --grid 32.32.32.64 --accelerator-threads $NT  --shm-mpi 1 > halo.2tile.1x2.log
+#mpiexec -launcher ssh -n 2 -host localhost  ./wrap4gpu.sh ./Benchmark_halo --mpi 2.1.1.1 --grid 64.32.32.32 --accelerator-threads $NT  --shm-mpi 1 > halo.2tile.2x1.log
+
+

systems/PVC/benchmarks/wrap.sh

@@ -0,0 +1,14 @@
+#!/bin/sh
+
+export ZE_AFFINITY_MASK=0.$MPI_LOCALRANKID
+
+echo Ranke $MPI_LOCALRANKID ZE_AFFINITY_MASK is $ZE_AFFINITY_MASK
+
+
+if [ $MPI_LOCALRANKID = "0" ] 
+then
+#  ~psteinbr/build_pti/ze_tracer -h $@
+  onetrace --chrome-device-timeline $@
+else
+  $@
+fi

systems/PVC/config-command

@@ -0,0 +1,16 @@
+INSTALL=/nfs/site/home/azusayax/install
+../../configure \
+	--enable-simd=GPU \
+	--enable-gen-simd-width=64 \
+	--enable-comms=mpi-auto \
+	--disable-accelerator-cshift \
+	--disable-gparity \
+	--disable-fermion-reps \
+	--enable-shm=nvlink \
+	--enable-accelerator=sycl \
+	--enable-unified=no \
+	MPICXX=mpicxx \
+	CXX=dpcpp \
+	LDFLAGS="-fsycl-device-code-split=per_kernel -fsycl-device-lib=all -lze_loader -L$INSTALL/lib" \
+	CXXFLAGS="-fsycl-unnamed-lambda -fsycl -no-fma -I$INSTALL/include -Wno-tautological-compare"
+

systems/PVC/setup.sh

@@ -0,0 +1,11 @@
+export https_proxy=http://proxy-chain.intel.com:911
+export LD_LIBRARY_PATH=/nfs/site/home/azusayax/install/lib:$LD_LIBRARY_PATH
+
+module load intel-release
+source /opt/intel/oneapi/PVC_setup.sh
+#source /opt/intel/oneapi/ATS_setup.sh
+module load intel/mpich/pvc45.3
+export PATH=~/ATS/pti-gpu/tools/onetrace/:$PATH
+
+#clsh embargo-ci-neo-022845
+#source /opt/intel/vtune_amplifier/amplxe-vars.sh

systems/mac-arm/config-command-mpi

@@ -1 +1 @@
-CXX=mpicxx-openmpi-mp CXXFLAGS=-I/opt/local/include/ LDFLAGS=-L/opt/local/lib/ ../../configure --enable-simd=GEN --enable-debug --enable-comms=mpi
+CXX=mpicxx-openmpi-mp CXXFLAGS=-I/opt/local/include/ LDFLAGS=-L/opt/local/lib/ ../../configure --enable-simd=GEN --enable-debug --enable-comms=mpi --enable-unified=no

tests/IO/Test_nersc_io.cc

@@ -147,7 +147,7 @@ int main (int argc, char ** argv)
   Complex p  = TensorRemove(Tp);
   std::cout<<GridLogMessage << "calculated plaquettes " <<p*PlaqScale<<std::endl;
 
-  Complex LinkTraceScale(1.0/vol/4.0/3.0);
+  Complex LinkTraceScale(1.0/vol/4.0/(Real)Nc);
   TComplex Tl = sum(LinkTrace);
   Complex l  = TensorRemove(Tl);
   std::cout<<GridLogMessage << "calculated link trace " <<l*LinkTraceScale<<std::endl;
@@ -157,8 +157,10 @@ int main (int argc, char ** argv)
   Complex ll= TensorRemove(TcP);
   std::cout<<GridLogMessage << "coarsened plaquettes sum to " <<ll*PlaqScale<<std::endl;
 
-  std::string clone2x3("./ckpoint_clone2x3.4000");
+  const string stNc   = to_string( Nc   ) ;
-  std::string clone3x3("./ckpoint_clone3x3.4000");
+  const string stNcM1 = to_string( Nc-1 ) ;
+  std::string clone2x3("./ckpoint_clone"+stNcM1+"x"+stNc+".4000");
+  std::string clone3x3("./ckpoint_clone"+stNc+"x"+stNc+".4000");
 
   NerscIO::writeConfiguration(Umu,clone3x3,0,precision32);
   NerscIO::writeConfiguration(Umu,clone2x3,1,precision32);

tests/Test_simd.cc

@@ -793,6 +793,7 @@ int main (int argc, char ** argv)
     }
     std::cout <<" OK ! "<<std::endl;
 
+#ifdef USE_FP16
     // Double to Half
     std::cout << GridLogMessage<< "Double to half" ;
     precisionChange(&H[0],&D[0],Ndp);
@@ -822,6 +823,7 @@ int main (int argc, char ** argv)
       assert( tmp < 1.0e-3 );
     }
     std::cout <<" OK ! "<<std::endl;
+#endif
 
   }
   Grid_finalize();

tests/core/Test_compact_wilson_clover_speedup.cc

@@ -117,8 +117,8 @@ void runBenchmark(int* argc, char*** argv) {
 
   // type definitions
   typedef WilsonImpl<vCoeff_t, FundamentalRepresentation, CoeffReal> WImpl;
-  typedef WilsonCloverFermion<WImpl> WilsonCloverOperator;
+  typedef WilsonCloverFermion<WImpl, CloverHelpers<WImpl>> WilsonCloverOperator;
-  typedef CompactWilsonCloverFermion<WImpl> CompactWilsonCloverOperator;
+  typedef CompactWilsonCloverFermion<WImpl, CompactCloverHelpers<WImpl>> CompactWilsonCloverOperator;
   typedef typename WilsonCloverOperator::FermionField Fermion;
   typedef typename WilsonCloverOperator::GaugeField Gauge;
 

tests/core/Test_fft_matt.cc

@@ -0,0 +1,270 @@
+    /*************************************************************************************
+    grid` physics library, www.github.com/paboyle/Grid 
+
+    Source file: ./tests/Test_cshift.cc
+
+    Copyright (C) 2015
+
+Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
+Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License along
+    with this program; if not, write to the Free Software Foundation, Inc.,
+    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+    See the full license in the file "LICENSE" in the top level distribution directory
+    *************************************************************************************/
+    /*  END LEGAL */
+#include <Grid/Grid.h>
+
+using namespace Grid;
+
+Gamma::Algebra Gmu [] = {
+  Gamma::Algebra::GammaX,
+  Gamma::Algebra::GammaY,
+  Gamma::Algebra::GammaZ,
+  Gamma::Algebra::GammaT,
+  Gamma::Algebra::Gamma5
+};
+
+int main (int argc, char ** argv)
+{
+  Grid_init(&argc,&argv);
+
+  int threads = GridThread::GetThreads();
+  std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
+
+  Coordinate latt_size   = GridDefaultLatt();
+  Coordinate simd_layout = GridDefaultSimd(Nd,vComplexD::Nsimd());
+  Coordinate mpi_layout  = GridDefaultMpi();
+
+  int vol = 1;
+  for(int d=0;d<latt_size.size();d++){
+    vol = vol * latt_size[d];
+  }
+  GridCartesian         GRID(latt_size,simd_layout,mpi_layout);
+  GridRedBlackCartesian RBGRID(&GRID);
+
+  LatticeComplexD    coor(&GRID);
+  ComplexD ci(0.0,1.0);
+
+  std::vector<int> seeds({1,2,3,4});
+  GridSerialRNG          sRNG;  sRNG.SeedFixedIntegers(seeds); // naughty seeding
+  GridParallelRNG          pRNG(&GRID);
+  pRNG.SeedFixedIntegers(seeds);
+
+  LatticeGaugeFieldD Umu(&GRID);
+  SU<Nc>::ColdConfiguration(pRNG,Umu); // Unit gauge
+
+  ////////////////////////////////////////////////////
+  // Wilson test
+  ////////////////////////////////////////////////////
+  {
+    LatticeFermionD    src(&GRID); gaussian(pRNG,src);
+    LatticeFermionD    src_p(&GRID);
+    LatticeFermionD    tmp(&GRID);
+    LatticeFermionD    ref(&GRID);
+    LatticeFermionD    result(&GRID);
+    
+    RealD mass=0.1;
+    WilsonFermionD Dw(Umu,GRID,RBGRID,mass);
+    
+    Dw.M(src,ref);
+    std::cout << "Norm src "<<norm2(src)<<std::endl;
+    std::cout << "Norm Dw x src "<<norm2(ref)<<std::endl;
+    {
+      FFT theFFT(&GRID);
+
+      ////////////////
+      // operator in Fourier space
+      ////////////////
+      tmp =ref;
+      theFFT.FFT_all_dim(result,tmp,FFT::forward);
+      std::cout<<"FFT[ Dw x src ]  "<< norm2(result)<<std::endl;    
+
+      tmp = src;
+      theFFT.FFT_all_dim(src_p,tmp,FFT::forward);
+      std::cout<<"FFT[ src      ]  "<< norm2(src_p)<<std::endl;
+      
+      /////////////////////////////////////////////////////////////////
+      // work out the predicted FT from Fourier
+      /////////////////////////////////////////////////////////////////
+      auto FGrid = &GRID;
+      LatticeFermionD    Kinetic(FGrid); Kinetic = Zero();
+      LatticeComplexD    kmu(FGrid); 
+      LatticeInteger     scoor(FGrid); 
+      LatticeComplexD    sk (FGrid); sk = Zero();
+      LatticeComplexD    sk2(FGrid); sk2= Zero();
+      LatticeComplexD    W(FGrid); W= Zero();
+      LatticeComplexD    one(FGrid); one =ComplexD(1.0,0.0);
+      ComplexD ci(0.0,1.0);
+    
+      for(int mu=0;mu<Nd;mu++) {
+	
+	RealD TwoPiL =  M_PI * 2.0/ latt_size[mu];
+
+	LatticeCoordinate(kmu,mu);
+
+	kmu = TwoPiL * kmu;
+      
+	sk2 = sk2 + 2.0*sin(kmu*0.5)*sin(kmu*0.5);
+	sk  = sk  +     sin(kmu)    *sin(kmu); 
+      
+	// -1/2 Dw ->  1/2 gmu (eip - emip) = i sinp gmu
+	Kinetic = Kinetic + sin(kmu)*ci*(Gamma(Gmu[mu])*src_p);
+	
+      }
+    
+      W = mass + sk2; 
+      Kinetic = Kinetic + W * src_p;
+    
+      std::cout<<"Momentum space src         "<< norm2(src_p)<<std::endl;
+      std::cout<<"Momentum space Dw x src    "<< norm2(Kinetic)<<std::endl;
+      std::cout<<"FT[Coordinate space Dw]    "<< norm2(result)<<std::endl;
+    
+      result = result - Kinetic;
+      std::cout<<"diff "<< norm2(result)<<std::endl;
+      
+    }
+
+    std::cout << " =======================================" <<std::endl;
+    std::cout << " Checking FourierFreePropagator x Dw = 1" <<std::endl;
+    std::cout << " =======================================" <<std::endl;
+    std::cout << "Dw src = " <<norm2(src)<<std::endl;
+    std::cout << "Dw tmp = " <<norm2(tmp)<<std::endl;
+    Dw.M(src,tmp);
+    Dw.FreePropagator(tmp,ref,mass);
+
+    std::cout << "Dw ref = " <<norm2(ref)<<std::endl;
+    
+    ref = ref - src;
+    
+    std::cout << "Dw ref-src = " <<norm2(ref)<<std::endl;
+  }
+
+
+  ////////////////////////////////////////////////////
+  // Wilson prop
+  ////////////////////////////////////////////////////
+  {
+    std::cout<<"****************************************"<<std::endl;
+    std::cout << "Wilson Mom space 4d propagator \n";
+    std::cout<<"****************************************"<<std::endl;
+
+    LatticeFermionD    src(&GRID); gaussian(pRNG,src);
+    LatticeFermionD    tmp(&GRID);
+    LatticeFermionD    ref(&GRID);
+    LatticeFermionD    diff(&GRID);
+
+    src=Zero();
+    Coordinate point(4,0); // 0,0,0,0
+    SpinColourVectorD ferm;
+    ferm=Zero();
+    ferm()(0)(0) = ComplexD(1.0);
+    pokeSite(ferm,src,point);
+
+    RealD mass=0.1;
+    WilsonFermionD Dw(Umu,GRID,RBGRID,mass);
+
+    // Momentum space prop
+    std::cout << " Solving by FFT and Feynman rules" <<std::endl;
+    Dw.FreePropagator(src,ref,mass) ;
+
+    Gamma G5(Gamma::Algebra::Gamma5);
+
+    LatticeFermionD    result(&GRID); 
+    const int sdir=0;
+    
+    ////////////////////////////////////////////////////////////////////////
+    // Conjugate gradient on normal equations system
+    ////////////////////////////////////////////////////////////////////////
+    std::cout << " Solving by Conjugate Gradient (CGNE)" <<std::endl;
+    Dw.Mdag(src,tmp);
+    src=tmp;
+    MdagMLinearOperator<WilsonFermionD,LatticeFermionD> HermOp(Dw);
+    ConjugateGradient<LatticeFermionD> CG(1.0e-10,10000);
+    CG(HermOp,src,result);
+    
+    ////////////////////////////////////////////////////////////////////////
+    std::cout << " Taking difference" <<std::endl;
+    std::cout << "Dw result "<<norm2(result)<<std::endl;
+    std::cout << "Dw ref     "<<norm2(ref)<<std::endl;
+    
+    diff = ref - result;
+    std::cout << "result - ref     "<<norm2(diff)<<std::endl;
+
+    DumpSliceNorm("Slice Norm Solution ",result,Nd-1);
+  }
+
+  ////////////////////////////////////////////////////
+  //Gauge invariance test
+  ////////////////////////////////////////////////////
+  {
+    std::cout<<"****************************************"<<std::endl;
+    std::cout << "Gauge invariance test \n";
+    std::cout<<"****************************************"<<std::endl;
+    LatticeGaugeField     U_GT(&GRID); // Gauge transformed field
+    LatticeColourMatrix   g(&GRID);    // local Gauge xform matrix
+    U_GT = Umu;
+    // Make a random xform to teh gauge field
+    SU<Nc>::RandomGaugeTransform(pRNG,U_GT,g); // Unit gauge
+
+    LatticeFermionD    src(&GRID);
+    LatticeFermionD    tmp(&GRID);
+    LatticeFermionD    ref(&GRID);
+    LatticeFermionD    diff(&GRID);
+
+    // could loop over colors
+    src=Zero();
+    Coordinate point(4,0); // 0,0,0,0
+    SpinColourVectorD ferm;
+    ferm=Zero();
+    ferm()(0)(0) = ComplexD(1.0);
+    pokeSite(ferm,src,point);
+
+    RealD mass=0.1;
+    WilsonFermionD Dw(U_GT,GRID,RBGRID,mass);
+
+    // Momentum space prop
+    std::cout << " Solving by FFT and Feynman rules" <<std::endl;
+    Dw.FreePropagator(src,ref,mass) ;
+
+    Gamma G5(Gamma::Algebra::Gamma5);
+
+    LatticeFermionD    result(&GRID); 
+    const int sdir=0;
+    
+    ////////////////////////////////////////////////////////////////////////
+    // Conjugate gradient on normal equations system
+    ////////////////////////////////////////////////////////////////////////
+    std::cout << " Solving by Conjugate Gradient (CGNE)" <<std::endl;
+    Dw.Mdag(src,tmp);
+    src=tmp;
+    MdagMLinearOperator<WilsonFermionD,LatticeFermionD> HermOp(Dw);
+    ConjugateGradient<LatticeFermionD> CG(1.0e-10,10000);
+    CG(HermOp,src,result);
+    
+    ////////////////////////////////////////////////////////////////////////
+    std::cout << " Taking difference" <<std::endl;
+    std::cout << "Dw result "<<norm2(result)<<std::endl;
+    std::cout << "Dw ref     "<<norm2(ref)<<std::endl;
+    
+    diff = ref - result;
+    std::cout << "result - ref     "<<norm2(diff)<<std::endl;
+
+    DumpSliceNorm("Slice Norm Solution ",result,Nd-1);
+  }
+  
+  
+  Grid_finalize();
+}

tests/core/Test_lie_generators.cc

@@ -9,6 +9,7 @@ Copyright (C) 2015
 Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
 Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 Author: Guido Cossu <guido.cossu@ed.ac.uk>
+Author: Jamie Hudspith <renwick.james.hudspth@gmail.com>
 
 This program is free software; you can redistribute it and/or modify
 it under the terms of the GNU General Public License as published by
@@ -42,7 +43,7 @@ directory
 
 using namespace std;
 using namespace Grid;
- ;
+;
 
 int main(int argc, char** argv) {
   Grid_init(&argc, &argv);
@@ -60,6 +61,10 @@ int main(int argc, char** argv) {
             << std::endl;
   SU2::printGenerators();
   std::cout << "Dimension of adjoint representation: "<< SU2Adjoint::Dimension << std::endl;
+
+  // guard as this code fails to compile for Nc != 3
+#if (Nc == 3)
+    
   SU2Adjoint::printGenerators();
   SU2::testGenerators();
   SU2Adjoint::testGenerators();
@@ -111,8 +116,6 @@ int main(int argc, char** argv) {
 
   // AdjointRepresentation has the predefined number of colours Nc
   //  Representations<FundamentalRepresentation, AdjointRepresentation, TwoIndexSymmetricRepresentation> RepresentationTypes(grid);  
-
-  
   LatticeGaugeField U(grid), V(grid);
   SU3::HotConfiguration<LatticeGaugeField>(gridRNG, U);
   SU3::HotConfiguration<LatticeGaugeField>(gridRNG, V);
@@ -181,7 +184,6 @@ int main(int argc, char** argv) {
   std::cout << GridLogMessage << "orthogonality check 2: " << norm2(Ucheck)
 	    << std::endl;
     
-
   // Construct the fundamental matrix in the group
   SU3::LatticeMatrix Af(grid);
   SU3::FundamentalLieAlgebraMatrix(h_adj,Af);
@@ -198,7 +200,6 @@ int main(int argc, char** argv) {
   std::cout << GridLogMessage << "unitarity check 2: " << norm2(UnitCheck)
 	    << std::endl;
 
-
   // Tranform to the adjoint representation
   U = Zero(); // fill this with only one direction
   pokeLorentz(U,Ufund,0); // the representation transf acts on full gauge fields
@@ -210,11 +211,7 @@ int main(int argc, char** argv) {
   typename AdjointRep<Nc>::LatticeMatrix Diff_check_mat = Ur0 - Uadj;
   std::cout << GridLogMessage << "Projections structure check group difference : " << norm2(Diff_check_mat) << std::endl;
 
-
-
-
   // TwoIndexRep tests
-
   std::cout << GridLogMessage << "*********************************************"
 	    << std::endl;
   std::cout << GridLogMessage << "*********************************************"
@@ -253,8 +250,6 @@ int main(int argc, char** argv) {
   std::cout << GridLogMessage << "Test of Two Index anti-Symmetric Generators: "<< SU2TwoIndexAntiSymm::Dimension << std::endl;
   SU2TwoIndexAntiSymm::testGenerators();
   
-  
-  
   std::cout << GridLogMessage << "*********************************************"
 	    << std::endl;
   std::cout << GridLogMessage << "Test for the Two Index Symmetric projectors"
@@ -357,7 +352,6 @@ int main(int argc, char** argv) {
   SU3::LatticeAlgebraVector h_diff_sym = h_sym - h_sym2;
   std::cout << GridLogMessage << "Projections structure check vector difference (Two Index Symmetric): " << norm2(h_diff_sym) << std::endl;
     
-  
   // Exponentiate
   typename TwoIndexRep< Nc, Symmetric>::LatticeMatrix U2iS(grid);
   U2iS  = expMat(Ar_sym, 1.0, 16);
@@ -376,8 +370,6 @@ int main(int argc, char** argv) {
   std::cout << GridLogMessage << "orthogonality check 2: " << norm2(Ucheck2iS)
 	    << std::endl;
     
-  
-  
   // Construct the fundamental matrix in the group
   SU3::LatticeMatrix Af_sym(grid);
   SU3::FundamentalLieAlgebraMatrix(h_sym,Af_sym);
@@ -403,9 +395,6 @@ int main(int argc, char** argv) {
   typename TwoIndexRep< Nc, Symmetric>::LatticeMatrix Diff_check_mat2 = Ur02 - U2iS;
   std::cout << GridLogMessage << "Projections structure check group difference (Two Index Symmetric): " << norm2(Diff_check_mat2) << std::endl;
     
-
-
-
   if (TwoIndexRep<Nc, AntiSymmetric >::Dimension != 1){
 
     std::cout << GridLogMessage << "*********************************************"
@@ -513,20 +502,13 @@ int main(int argc, char** argv) {
     typename TwoIndexRep< Nc, AntiSymmetric>::LatticeMatrix Diff_check_mat2A = Ur02A - U2iAS;
     std::cout << GridLogMessage << "Projections structure check group difference (Two Index anti-Symmetric): " << norm2(Diff_check_mat2A) << std::endl;
   
-} else  {
+  } else  {
     std::cout << GridLogMessage << "Skipping Two Index anti-Symmetric tests "
       "because representation is trivial (dim = 1)"
 	      << std::endl;
-}
+  }
-
-
-
-
-
-
-
-
 
+#endif
   
   Grid_finalize();
 }

tests/core/Test_memory_manager.cc

@@ -0,0 +1,110 @@
+    /*************************************************************************************
+
+    Grid physics library, www.github.com/paboyle/Grid 
+
+    Source file: ./tests/Test_memory_manager.cc
+
+    Copyright (C) 2022
+
+Author: Peter Boyle <pboyle@bnl.gov>
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License along
+    with this program; if not, write to the Free Software Foundation, Inc.,
+    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+    See the full license in the file "LICENSE" in the top level distribution directory
+    *************************************************************************************/
+    /*  END LEGAL */
+#include <Grid/Grid.h>
+
+using namespace std;
+using namespace Grid;
+
+void  MemoryTest(GridCartesian         * FGrid,int N);
+
+int main (int argc, char ** argv)
+{
+  Grid_init(&argc,&argv);
+
+  GridCartesian         * UGrid   = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());
+
+  int N=100;
+  for(int i=0;i<N;i++){
+    std::cout << "============================"<<std::endl;
+    std::cout << "Epoch "<<i<<"/"<<N<<std::endl;
+    std::cout << "============================"<<std::endl;
+    MemoryTest(UGrid,256);
+    MemoryManager::Print();
+    AUDIT();
+  }
+  Grid_finalize();
+}
+
+void  MemoryTest(GridCartesian         * FGrid, int N)
+{
+  LatticeComplexD zero(FGrid); zero=Zero();
+  std::vector<LatticeComplexD> A(N,zero);//FGrid);
+
+  std::vector<ComplexD> B(N,ComplexD(0.0)); // Update sequentially on host
+
+  for(int v=0;v<N;v++) A[v] = Zero();
+
+  uint64_t counter = 0;
+  for(int epoch = 0;epoch<10000;epoch++){
+
+    int v  = random() %N; // Which vec
+    int w  = random() %2; // Write or read
+    int e  = random() %3; // expression or for loop
+    int dev= random() %2; // On device?
+    //    int e=1;
+    ComplexD zc = counter++;
+    
+    if ( w ) {
+      B[v] = B[v] + zc;
+      if ( e == 0 ) {
+	A[v] = A[v] + zc - A[v] + A[v];
+      } else {
+	if ( dev ) { 
+	  autoView(A_v,A[v],AcceleratorWrite);
+	  accelerator_for(ss,FGrid->oSites(),1,{
+	    A_v[ss] = A_v[ss] + zc;
+	    });
+	} else {
+	  autoView(A_v,A[v],CpuWrite);
+	  thread_for(ss,FGrid->oSites(),{
+	      A_v[ss] = A_v[ss] + zc;
+	    });
+	}
+      }
+    } else {
+      if ( e == 0 ) {
+	A[v] = A[v] + A[v] - A[v];
+      } else { 
+	if ( dev ) { 
+	  autoView(A_v,A[v],AcceleratorRead);
+	  accelerator_for(ss,FGrid->oSites(),1,{
+	      assert(B[v]==A_v[ss]()()().getlane(0));
+	    });
+	  //	std::cout << "["<<v<<"] checked on GPU"<<B[v]<<std::endl;
+	} else {
+	  autoView(A_v,A[v],CpuRead);
+	  thread_for(ss,FGrid->oSites(),{
+	      assert(B[v]==A_v[ss]()()().getlane(0));
+	    });
+	  //	std::cout << "["<<v<<"] checked on CPU"<<B[v]<<std::endl;
+	}
+      }    
+    }
+  }
+
+}

tests/core/Test_reunitarise.cc

@@ -122,13 +122,14 @@ int main (int argc, char ** argv)
   std::cout << "Determinant defect before projection " <<norm2(detU)<<std::endl;
   tmp = U*adj(U) - ident;
   std::cout << "Unitarity check before projection    " << norm2(tmp)<<std::endl; 
-  
+#if (Nc == 3)
   ProjectSU3(U);
   detU= Determinant(U) ;
   detU= detU -1.0;
   std::cout << "Determinant ProjectSU3 defect " <<norm2(detU)<<std::endl;
   tmp = U*adj(U) - ident;
   std::cout << "Unitarity check after projection    " << norm2(tmp)<<std::endl; 
+#endif
   
   ProjectSUn(UU);
   detUU= Determinant(UU);

tests/core/Test_wilson_clover.cc

@@ -2,11 +2,12 @@
 
     Grid physics library, www.github.com/paboyle/Grid 
 
-    Source file: ./benchmarks/Benchmark_wilson.cc
+    Source file: ./tests/core/Test_wilson_clover.cc
 
     Copyright (C) 2015
 
     Author: Guido Cossu <guido.cossu@ed.ac.uk>
+            Fabian Joswig <fabian.joswig@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
@@ -67,8 +68,6 @@ int main(int argc, char **argv)
   tmp = Zero();
   FermionField err(&Grid);
   err = Zero();
-  FermionField err2(&Grid);
-  err2 = Zero();
   FermionField phi(&Grid);
   random(pRNG, phi);
   FermionField chi(&Grid);
@@ -77,6 +76,8 @@ int main(int argc, char **argv)
   SU<Nc>::HotConfiguration(pRNG, Umu);
   std::vector<LatticeColourMatrix> U(4, &Grid);
 
+  double tolerance = 1e-4;
+
   double volume = 1;
   for (int mu = 0; mu < Nd; mu++)
   {
@@ -88,7 +89,7 @@ int main(int argc, char **argv)
   RealD csw_t = 1.0;
 
   WilsonCloverFermionR Dwc(Umu, Grid, RBGrid, mass, csw_r, csw_t, anis, params);
-  //Dwc.ImportGauge(Umu); // not necessary, included in the constructor
+  CompactWilsonCloverFermionR Dwc_compact(Umu, Grid, RBGrid, mass, csw_r, csw_t, 1.0, anis, params);
 
   std::cout << GridLogMessage << "==========================================================" << std::endl;
   std::cout << GridLogMessage << "= Testing that Deo + Doe = Dunprec " << std::endl;
@@ -112,7 +113,24 @@ int main(int argc, char **argv)
   setCheckerboard(r_eo, r_e);
 
   err = ref - r_eo;
-  std::cout << GridLogMessage << "EO norm diff   " << norm2(err) << " " << norm2(ref) << " " << norm2(r_eo) << std::endl;
+  std::cout << GridLogMessage << "EO norm diff\t" << norm2(err) << " (" << norm2(ref) << " - " << norm2(r_eo) << ")" << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
+
+
+
+  Dwc_compact.Meooe(src_e, r_o);
+  std::cout << GridLogMessage << "Applied Meo" << std::endl;
+  Dwc_compact.Meooe(src_o, r_e);
+  std::cout << GridLogMessage << "Applied Moe" << std::endl;
+  Dwc_compact.Dhop(src, ref, DaggerNo);
+
+  setCheckerboard(r_eo, r_o);
+  setCheckerboard(r_eo, r_e);
+
+  err = ref - r_eo;
+  std::cout << GridLogMessage << "EO norm diff compact\t" << norm2(err) << " (" << norm2(ref) << " - " << norm2(r_eo) << ")" << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
+
 
   std::cout << GridLogMessage << "==============================================================" << std::endl;
   std::cout << GridLogMessage << "= Test Ddagger is the dagger of D by requiring                " << std::endl;
@@ -152,6 +170,22 @@ int main(int argc, char **argv)
   std::cout << GridLogMessage << "pDce - conj(cDpo) " << pDce - conj(cDpo) << std::endl;
   std::cout << GridLogMessage << "pDco - conj(cDpe) " << pDco - conj(cDpe) << std::endl;
 
+  Dwc_compact.Meooe(chi_e, dchi_o);
+  Dwc_compact.Meooe(chi_o, dchi_e);
+  Dwc_compact.MeooeDag(phi_e, dphi_o);
+  Dwc_compact.MeooeDag(phi_o, dphi_e);
+
+  pDce = innerProduct(phi_e, dchi_e);
+  pDco = innerProduct(phi_o, dchi_o);
+  cDpe = innerProduct(chi_e, dphi_e);
+  cDpo = innerProduct(chi_o, dphi_o);
+
+  std::cout << GridLogMessage << "e compact " << pDce << " " << cDpe << std::endl;
+  std::cout << GridLogMessage << "o compact " << pDco << " " << cDpo << std::endl;
+
+  std::cout << GridLogMessage << "pDce - conj(cDpo) compact " << pDce - conj(cDpo) << std::endl;
+  std::cout << GridLogMessage << "pDco - conj(cDpe) compact " << pDco - conj(cDpe) << std::endl;
+
   std::cout << GridLogMessage << "==============================================================" << std::endl;
   std::cout << GridLogMessage << "= Test MeeInv Mee = 1   (if csw!=0)                           " << std::endl;
   std::cout << GridLogMessage << "==============================================================" << std::endl;
@@ -170,6 +204,20 @@ int main(int argc, char **argv)
 
   err = phi - chi;
   std::cout << GridLogMessage << "norm diff " << norm2(err) << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
+
+  Dwc_compact.Mooee(chi_e, src_e);
+  Dwc_compact.MooeeInv(src_e, phi_e);
+
+  Dwc_compact.Mooee(chi_o, src_o);
+  Dwc_compact.MooeeInv(src_o, phi_o);
+
+  setCheckerboard(phi, phi_e);
+  setCheckerboard(phi, phi_o);
+
+  err = phi - chi;
+  std::cout << GridLogMessage << "norm diff compact " << norm2(err) << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
 
   std::cout << GridLogMessage << "==============================================================" << std::endl;
   std::cout << GridLogMessage << "= Test MeeDag MeeInvDag = 1    (if csw!=0)                    " << std::endl;
@@ -189,6 +237,20 @@ int main(int argc, char **argv)
 
   err = phi - chi;
   std::cout << GridLogMessage << "norm diff " << norm2(err) << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
+
+  Dwc_compact.MooeeDag(chi_e, src_e);
+  Dwc_compact.MooeeInvDag(src_e, phi_e);
+
+  Dwc_compact.MooeeDag(chi_o, src_o);
+  Dwc_compact.MooeeInvDag(src_o, phi_o);
+
+  setCheckerboard(phi, phi_e);
+  setCheckerboard(phi, phi_o);
+
+  err = phi - chi;
+  std::cout << GridLogMessage << "norm diff compact " << norm2(err) << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
 
   std::cout << GridLogMessage << "==============================================================" << std::endl;
   std::cout << GridLogMessage << "= Test MeeInv MeeDag = 1      (if csw!=0)                     " << std::endl;
@@ -208,6 +270,20 @@ int main(int argc, char **argv)
 
   err = phi - chi;
   std::cout << GridLogMessage << "norm diff " << norm2(err) << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
+
+  Dwc_compact.MooeeDag(chi_e, src_e);
+  Dwc_compact.MooeeInv(src_e, phi_e);
+
+  Dwc_compact.MooeeDag(chi_o, src_o);
+  Dwc_compact.MooeeInv(src_o, phi_o);
+
+  setCheckerboard(phi, phi_e);
+  setCheckerboard(phi, phi_o);
+
+  err = phi - chi;
+  std::cout << GridLogMessage << "norm diff compact " << norm2(err) << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
 
   std::cout << GridLogMessage << "================================================================" << std::endl;
   std::cout << GridLogMessage << "= Testing gauge covariance Clover term with EO preconditioning  " << std::endl;
@@ -249,7 +325,7 @@ int main(int argc, char **argv)
   /////////////////
 
   WilsonCloverFermionR Dwc_prime(U_prime, Grid, RBGrid, mass, csw_r, csw_t, anis, params);
-  Dwc_prime.ImportGauge(U_prime);
+  CompactWilsonCloverFermionR Dwc_compact_prime(U_prime, Grid, RBGrid, mass, csw_r, csw_t, 1.0, anis, params);
 
   tmp = Omega * src;
   pickCheckerboard(Even, src_e, tmp);
@@ -263,6 +339,36 @@ int main(int argc, char **argv)
 
   err = chi - adj(Omega) * phi;
   std::cout << GridLogMessage << "norm diff " << norm2(err) << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
+
+  chi = Zero();
+  phi = Zero();
+  tmp = Zero();
+  pickCheckerboard(Even, chi_e, chi);
+  pickCheckerboard(Odd, chi_o, chi);
+  pickCheckerboard(Even, phi_e, phi);
+  pickCheckerboard(Odd, phi_o, phi);
+
+  Dwc_compact.Mooee(src_e, chi_e);
+  Dwc_compact.Mooee(src_o, chi_o);
+  setCheckerboard(chi, chi_e);
+  setCheckerboard(chi, chi_o);
+  setCheckerboard(src, src_e);
+  setCheckerboard(src, src_o);
+
+  tmp = Omega * src;
+  pickCheckerboard(Even, src_e, tmp);
+  pickCheckerboard(Odd, src_o, tmp);
+
+  Dwc_compact_prime.Mooee(src_e, phi_e);
+  Dwc_compact_prime.Mooee(src_o, phi_o);
+
+  setCheckerboard(phi, phi_e);
+  setCheckerboard(phi, phi_o);
+
+  err = chi - adj(Omega) * phi;
+  std::cout << GridLogMessage << "norm diff compact " << norm2(err) << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
 
   std::cout << GridLogMessage << "=================================================================" << std::endl;
   std::cout << GridLogMessage << "= Testing gauge covariance Clover term w/o EO preconditioning  " << std::endl;
@@ -272,7 +378,6 @@ int main(int argc, char **argv)
   phi = Zero();
 
   WilsonFermionR Dw(Umu, Grid, RBGrid, mass, params);
-  Dw.ImportGauge(Umu);
 
   Dw.M(src, result);
   Dwc.M(src, chi);
@@ -280,13 +385,24 @@ int main(int argc, char **argv)
   Dwc_prime.M(Omega * src, phi);
 
   WilsonFermionR Dw_prime(U_prime, Grid, RBGrid, mass, params);
-  Dw_prime.ImportGauge(U_prime);
   Dw_prime.M(Omega * src, result2);
 
-  err = chi - adj(Omega) * phi;
+  err = result - adj(Omega) * result2;
-  err2 = result - adj(Omega) * result2;
   std::cout << GridLogMessage << "norm diff Wilson              " << norm2(err) << std::endl;
-  std::cout << GridLogMessage << "norm diff WilsonClover  " << norm2(err2) << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
+  err = chi - adj(Omega) * phi;
+  std::cout << GridLogMessage << "norm diff WilsonClover        " << norm2(err) << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
+
+  chi = Zero();
+  phi = Zero();
+
+  Dwc_compact.M(src, chi);
+  Dwc_compact_prime.M(Omega * src, phi);
+
+  err = chi - adj(Omega) * phi;
+  std::cout << GridLogMessage << "norm diff CompactWilsonClover " << norm2(err) << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
 
   std::cout << GridLogMessage << "==========================================================" << std::endl;
   std::cout << GridLogMessage << "= Testing Mooee(csw=0) Clover to reproduce Mooee Wilson   " << std::endl;
@@ -296,7 +412,6 @@ int main(int argc, char **argv)
   phi = Zero();
   err = Zero();
   WilsonCloverFermionR Dwc_csw0(Umu, Grid, RBGrid, mass, 0.0, 0.0, anis, params); //  <-- Notice: csw=0
-  Dwc_csw0.ImportGauge(Umu);
 
   pickCheckerboard(Even, phi_e, phi);
   pickCheckerboard(Odd, phi_o, phi);
@@ -317,6 +432,33 @@ int main(int argc, char **argv)
 
   err = chi - phi;
   std::cout << GridLogMessage << "norm diff " << norm2(err) << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
+
+  chi = Zero();
+  phi = Zero();
+  err = Zero();
+  CompactWilsonCloverFermionR Dwc_compact_csw0(Umu, Grid, RBGrid, mass, 0.0, 0.0, 1.0, anis, params); //  <-- Notice: csw=0
+
+  pickCheckerboard(Even, phi_e, phi);
+  pickCheckerboard(Odd, phi_o, phi);
+  pickCheckerboard(Even, chi_e, chi);
+  pickCheckerboard(Odd, chi_o, chi);
+
+  Dw.Mooee(src_e, chi_e);
+  Dw.Mooee(src_o, chi_o);
+  Dwc_compact_csw0.Mooee(src_e, phi_e);
+  Dwc_compact_csw0.Mooee(src_o, phi_o);
+
+  setCheckerboard(chi, chi_e);
+  setCheckerboard(chi, chi_o);
+  setCheckerboard(phi, phi_e);
+  setCheckerboard(phi, phi_o);
+  setCheckerboard(src, src_e);
+  setCheckerboard(src, src_o);
+
+  err = chi - phi;
+  std::cout << GridLogMessage << "norm diff compact " << norm2(err) << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
 
   std::cout << GridLogMessage << "==========================================================" << std::endl;
   std::cout << GridLogMessage << "= Testing EO operator is equal to the unprec              " << std::endl;
@@ -348,9 +490,41 @@ int main(int argc, char **argv)
   setCheckerboard(phi, phi_o);
 
   err = ref - phi;
-  std::cout << GridLogMessage << "ref (unpreconditioned operator) diff  :" << norm2(ref) << std::endl;
+  std::cout << GridLogMessage << "ref (unpreconditioned operator) diff         : " << norm2(ref) << std::endl;
-  std::cout << GridLogMessage << "phi (EO decomposition)          diff  :" << norm2(phi) << std::endl;
+  std::cout << GridLogMessage << "phi (EO decomposition)          diff         : " << norm2(phi) << std::endl;
-  std::cout << GridLogMessage << "norm diff                             :" << norm2(err) << std::endl;
+  std::cout << GridLogMessage << "norm diff                                    : " << norm2(err) << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
+
+  chi = Zero();
+  phi = Zero();
+  err = Zero();
+
+  pickCheckerboard(Even, phi_e, phi);
+  pickCheckerboard(Odd, phi_o, phi);
+  pickCheckerboard(Even, chi_e, chi);
+  pickCheckerboard(Odd, chi_o, chi);
+
+  // M phi = (Mooee src_e + Meooe src_o , Meooe src_e + Mooee src_o)
+
+  Dwc_compact.M(src, ref); // Reference result from the unpreconditioned operator
+
+  // EO matrix
+  Dwc_compact.Mooee(src_e, chi_e); 
+  Dwc_compact.Mooee(src_o, chi_o);
+  Dwc_compact.Meooe(src_o, phi_e);
+  Dwc_compact.Meooe(src_e, phi_o);
+
+  phi_o += chi_o;
+  phi_e += chi_e;
+
+  setCheckerboard(phi, phi_e);
+  setCheckerboard(phi, phi_o);
+
+  err = ref - phi;
+  std::cout << GridLogMessage << "ref (unpreconditioned operator) diff compact : " << norm2(ref) << std::endl;
+  std::cout << GridLogMessage << "phi (EO decomposition)          diff compact : " << norm2(phi) << std::endl;
+  std::cout << GridLogMessage << "norm diff compact                            : " << norm2(err) << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
 
   Grid_finalize();
 }

tests/core/Test_wilson_conserved_current.cc

@@ -0,0 +1,253 @@
+/*************************************************************************************
+Grid physics library, www.github.com/paboyle/Grid
+
+Source file: ./tests/Test_cayley_cg.cc
+
+Copyright (C) 2022
+
+Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+Author: Fabian Joswig <fabian.joswig@ed.ac.uk>
+
+This program is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 2 of the License, or
+(at your option) any later version.
+
+This program is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along
+with this program; if not, write to the Free Software Foundation, Inc.,
+51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+See the full license in the file "LICENSE" in the top level distribution directory
+*************************************************************************************/
+/*  END LEGAL */
+#include <Grid/Grid.h>
+
+using namespace std;
+using namespace Grid;
+
+
+template<class What>
+void  TestConserved(What & Dw,
+		    LatticeGaugeField &Umu,
+		    GridCartesian         * UGrid,	       GridRedBlackCartesian * UrbGrid,
+		    GridParallelRNG *RNG4);
+
+  Gamma::Algebra Gmu [] = {
+    Gamma::Algebra::GammaX,
+    Gamma::Algebra::GammaY,
+    Gamma::Algebra::GammaZ,
+    Gamma::Algebra::GammaT,
+    Gamma::Algebra::Gamma5
+  };
+
+int main (int argc, char ** argv)
+{
+  Grid_init(&argc,&argv);
+
+  int threads = GridThread::GetThreads();
+  std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
+
+  GridCartesian         * UGrid   = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(),
+								   GridDefaultSimd(Nd,vComplex::Nsimd()),
+								   GridDefaultMpi());
+  GridRedBlackCartesian * UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
+
+  std::vector<int> seeds5({5,6,7,8});
+  GridParallelRNG          RNG4(UGrid);
+  std::vector<int> seeds4({1,2,3,4}); RNG4.SeedFixedIntegers(seeds4);
+
+  LatticeGaugeField Umu(UGrid);
+  if( argc > 1 && argv[1][0] != '-' )
+  {
+    std::cout<<GridLogMessage <<"Loading configuration from "<<argv[1]<<std::endl;
+    FieldMetaData header;
+    NerscIO::readConfiguration(Umu, header, argv[1]);
+  }
+  else
+  {
+    std::cout<<GridLogMessage <<"Using hot configuration"<<std::endl;
+    SU<Nc>::HotConfiguration(RNG4,Umu);
+  }
+
+  typename WilsonCloverFermionR::ImplParams params;
+  WilsonAnisotropyCoefficients anis;
+  RealD mass = 0.1;
+  RealD csw_r = 1.0;
+  RealD csw_t = 1.0;
+
+  std::cout<<GridLogMessage <<"=================================="<<std::endl;
+  std::cout<<GridLogMessage <<"WilsonFermion test"<<std::endl;
+  std::cout<<GridLogMessage <<"=================================="<<std::endl;
+  WilsonFermionR Dw(Umu,*UGrid,*UrbGrid,mass,params);
+  TestConserved<WilsonFermionR>(Dw,Umu,UGrid,UrbGrid,&RNG4);
+
+  std::cout<<GridLogMessage <<"=================================="<<std::endl;
+  std::cout<<GridLogMessage <<"WilsonCloverFermion test"<<std::endl;
+  std::cout<<GridLogMessage <<"=================================="<<std::endl;
+  WilsonCloverFermionR Dwc(Umu, *UGrid, *UrbGrid, mass, csw_r, csw_t, anis, params);
+  TestConserved<WilsonCloverFermionR>(Dwc,Umu,UGrid,UrbGrid,&RNG4);
+
+  std::cout<<GridLogMessage <<"=================================="<<std::endl;
+  std::cout<<GridLogMessage <<"CompactWilsonCloverFermion test"<<std::endl;
+  std::cout<<GridLogMessage <<"=================================="<<std::endl;
+  CompactWilsonCloverFermionR Dwcc(Umu, *UGrid, *UrbGrid, mass, csw_r, csw_t, 1.0, anis, params);
+  TestConserved<CompactWilsonCloverFermionR>(Dwcc,Umu,UGrid,UrbGrid,&RNG4);
+
+  std::cout<<GridLogMessage <<"=================================="<<std::endl;
+  std::cout<<GridLogMessage <<"WilsonExpCloverFermion test"<<std::endl;
+  std::cout<<GridLogMessage <<"=================================="<<std::endl;
+  WilsonExpCloverFermionR Dewc(Umu, *UGrid, *UrbGrid, mass, csw_r, csw_t, anis, params);
+  TestConserved<WilsonExpCloverFermionR>(Dewc,Umu,UGrid,UrbGrid,&RNG4);
+
+  std::cout<<GridLogMessage <<"=================================="<<std::endl;
+  std::cout<<GridLogMessage <<"CompactWilsonExpCloverFermion test"<<std::endl;
+  std::cout<<GridLogMessage <<"=================================="<<std::endl;
+  CompactWilsonExpCloverFermionR Dewcc(Umu, *UGrid, *UrbGrid, mass, csw_r, csw_t, 1.0, anis, params);
+  TestConserved<CompactWilsonExpCloverFermionR>(Dewcc,Umu,UGrid,UrbGrid,&RNG4);
+
+  Grid_finalize();
+}
+
+
+
+template<class Action>
+void  TestConserved(Action & Dw,
+		    LatticeGaugeField &Umu,
+		    GridCartesian         * UGrid,	       GridRedBlackCartesian * UrbGrid,
+		    GridParallelRNG *RNG4)
+{
+  LatticePropagator phys_src(UGrid);
+  LatticePropagator seqsrc(UGrid);
+  LatticePropagator prop4(UGrid);
+  LatticePropagator Vector_mu(UGrid);
+  LatticeComplex    SV (UGrid);
+  LatticeComplex    VV (UGrid);
+  LatticePropagator seqprop(UGrid);
+
+  SpinColourMatrix kronecker; kronecker=1.0;
+  Coordinate coor({0,0,0,0});
+  phys_src=Zero();
+  pokeSite(kronecker,phys_src,coor);
+
+  ConjugateGradient<LatticeFermion> CG(1.0e-16,100000);
+  SchurRedBlackDiagTwoSolve<LatticeFermion> schur(CG);
+  ZeroGuesser<LatticeFermion> zpg;
+  for(int s=0;s<Nd;s++){
+    for(int c=0;c<Nc;c++){
+      LatticeFermion src4  (UGrid);
+      PropToFerm<Action>(src4,phys_src,s,c);
+
+      LatticeFermion result4(UGrid); result4=Zero();
+      schur(Dw,src4,result4,zpg);
+      std::cout<<GridLogMessage<<"spin "<<s<<" color "<<c<<" norm2(sourc4d) "<<norm2(src4)
+               <<" norm2(result4d) "<<norm2(result4)<<std::endl;
+      FermToProp<Action>(prop4,result4,s,c);
+    }
+  }
+
+  auto curr = Current::Vector;
+  const int mu_J=0;
+  const int t_J=0;
+
+  LatticeComplex    ph (UGrid); ph=1.0;
+
+  Dw.SeqConservedCurrent(prop4,
+			   seqsrc,
+			   phys_src,
+			   curr,
+			   mu_J,
+			   t_J,
+			   t_J,// whole lattice
+			   ph);
+
+  for(int s=0;s<Nd;s++){
+    for(int c=0;c<Nc;c++){
+
+      LatticeFermion src4  (UGrid);
+      PropToFerm<Action>(src4,seqsrc,s,c);
+
+      LatticeFermion result4(UGrid); result4=Zero();
+      schur(Dw,src4,result4,zpg);
+
+      FermToProp<Action>(seqprop,result4,s,c);
+    }
+  }
+
+  Gamma g5(Gamma::Algebra::Gamma5);
+  Gamma gT(Gamma::Algebra::GammaT);
+
+  std::vector<TComplex> sumSV;
+  std::vector<TComplex> sumVV;
+
+  Dw.ContractConservedCurrent(prop4,prop4,Vector_mu,phys_src,Current::Vector,Tdir);
+
+  SV       = trace(Vector_mu);        // Scalar-Vector conserved current
+  VV       = trace(gT*Vector_mu);     // (local) Vector-Vector conserved current
+
+  // Spatial sum
+  sliceSum(SV,sumSV,Tdir);
+  sliceSum(VV,sumVV,Tdir);
+
+  const int Nt{static_cast<int>(sumSV.size())};
+
+  std::cout<<GridLogMessage<<"Vector Ward identity by timeslice (~ 0)"<<std::endl;
+  for(int t=0;t<Nt;t++){
+    std::cout<<GridLogMessage <<" t "<<t<<" SV "<<real(TensorRemove(sumSV[t]))<<" VV "<<real(TensorRemove(sumVV[t]))<<std::endl;
+    assert(abs(real(TensorRemove(sumSV[t]))) < 1e-10);
+    assert(abs(real(TensorRemove(sumVV[t]))) < 1e-2);
+  }
+
+  ///////////////////////////////
+  // 3pt vs 2pt check
+  ///////////////////////////////
+  {
+    Gamma::Algebra        gA = Gamma::Algebra::Identity;
+    Gamma                 g(gA);
+
+    LatticePropagator cur(UGrid);
+    LatticePropagator tmp(UGrid);
+    LatticeComplex c(UGrid);
+    SpinColourMatrix qSite;
+    peekSite(qSite, seqprop, coor);
+
+    Complex               test_S, test_V, check_S, check_V;
+
+    std::vector<TComplex> check_buf;
+
+    test_S = trace(qSite*g);
+    test_V = trace(qSite*g*Gamma::gmu[mu_J]);
+
+    Dw.ContractConservedCurrent(prop4,prop4,cur,phys_src,curr,mu_J);
+
+    c = trace(cur*g);
+    sliceSum(c, check_buf, Tp);
+    check_S = TensorRemove(check_buf[t_J]);
+
+    auto gmu=Gamma::gmu[mu_J];
+    c = trace(cur*g*gmu);
+    sliceSum(c, check_buf, Tp);
+    check_V = TensorRemove(check_buf[t_J]);
+
+
+    std::cout<<GridLogMessage << std::setprecision(14)<<"Test S  = " << abs(test_S)   << std::endl;
+    std::cout<<GridLogMessage << "Test V  = " << abs(test_V) << std::endl;
+    std::cout<<GridLogMessage << "Check S = " << abs(check_S) << std::endl;
+    std::cout<<GridLogMessage << "Check V = " << abs(check_V) << std::endl;
+
+    // Check difference = 0
+    check_S = check_S - test_S;
+    check_V = check_V - test_V;
+
+    std::cout<<GridLogMessage << "Consistency check for sequential conserved " <<std::endl;
+    std::cout<<GridLogMessage << "Diff S  = " << abs(check_S) << std::endl;
+    assert(abs(check_S) < 1e-8);
+    std::cout<<GridLogMessage << "Diff V  = " << abs(check_V) << std::endl;
+    assert(abs(check_V) < 1e-8);
+  }
+
+}

tests/core/Test_wilson_exp_clover.cc

@@ -0,0 +1,530 @@
+/*************************************************************************************
+
+    Grid physics library, www.github.com/paboyle/Grid 
+
+    Source file: ./tests/core/Test_wilson_exp_clover.cc
+
+    Copyright (C) 2022
+
+    Author: Guido Cossu <guido.cossu@ed.ac.uk>
+            Fabian Joswig <fabian.joswig@ed.ac.uk>
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License along
+    with this program; if not, write to the Free Software Foundation, Inc.,
+    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+    See the full license in the file "LICENSE" in the top level distribution directory
+    *************************************************************************************/
+/*  END LEGAL */
+#include <Grid/Grid.h>
+
+using namespace std;
+using namespace Grid;
+
+int main(int argc, char **argv)
+{
+  Grid_init(&argc, &argv);
+
+  auto latt_size = GridDefaultLatt();
+  auto simd_layout = GridDefaultSimd(Nd, vComplex::Nsimd());
+  auto mpi_layout = GridDefaultMpi();
+  GridCartesian Grid(latt_size, simd_layout, mpi_layout);
+  GridRedBlackCartesian RBGrid(&Grid);
+
+  int threads = GridThread::GetThreads();
+  std::cout << GridLogMessage << "Grid is setup to use " << threads << " threads" << std::endl;
+  std::cout << GridLogMessage << "Grid floating point word size is REALF" << sizeof(RealF) << std::endl;
+  std::cout << GridLogMessage << "Grid floating point word size is REALD" << sizeof(RealD) << std::endl;
+  std::cout << GridLogMessage << "Grid floating point word size is REAL" << sizeof(Real) << std::endl;
+
+  std::vector<int> seeds({1, 2, 3, 4});
+  GridParallelRNG pRNG(&Grid);
+  pRNG.SeedFixedIntegers(seeds);
+  //  pRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9});
+
+  typedef typename WilsonExpCloverFermionR::FermionField FermionField;
+  typename WilsonExpCloverFermionR::ImplParams params;
+  WilsonAnisotropyCoefficients anis;
+
+  FermionField src(&Grid);
+  random(pRNG, src);
+  FermionField result(&Grid);
+  result = Zero();
+  FermionField result2(&Grid);
+  result2 = Zero();
+  FermionField ref(&Grid);
+  ref = Zero();
+  FermionField tmp(&Grid);
+  tmp = Zero();
+  FermionField err(&Grid);
+  err = Zero();
+  FermionField phi(&Grid);
+  random(pRNG, phi);
+  FermionField chi(&Grid);
+  random(pRNG, chi);
+  LatticeGaugeField Umu(&Grid);
+  SU<Nc>::HotConfiguration(pRNG, Umu);
+  std::vector<LatticeColourMatrix> U(4, &Grid);
+
+  double tolerance = 1e-4;
+
+  double volume = 1;
+  for (int mu = 0; mu < Nd; mu++)
+  {
+    volume = volume * latt_size[mu];
+  }
+
+  RealD mass = 0.1;
+  RealD csw_r = 1.0;
+  RealD csw_t = 1.0;
+
+  WilsonExpCloverFermionR Dwc(Umu, Grid, RBGrid, mass, csw_r, csw_t, anis, params);
+  CompactWilsonExpCloverFermionR Dwc_compact(Umu, Grid, RBGrid, mass, csw_r, csw_t, 1.0, anis, params);
+
+  std::cout << GridLogMessage << "==========================================================" << std::endl;
+  std::cout << GridLogMessage << "= Testing that Deo + Doe = Dunprec " << std::endl;
+  std::cout << GridLogMessage << "==========================================================" << std::endl;
+
+  FermionField src_e(&RBGrid);
+  FermionField src_o(&RBGrid);
+  FermionField r_e(&RBGrid);
+  FermionField r_o(&RBGrid);
+  FermionField r_eo(&Grid);
+  pickCheckerboard(Even, src_e, src);
+  pickCheckerboard(Odd, src_o, src);
+
+  Dwc.Meooe(src_e, r_o);
+  std::cout << GridLogMessage << "Applied Meo" << std::endl;
+  Dwc.Meooe(src_o, r_e);
+  std::cout << GridLogMessage << "Applied Moe" << std::endl;
+  Dwc.Dhop(src, ref, DaggerNo);
+
+  setCheckerboard(r_eo, r_o);
+  setCheckerboard(r_eo, r_e);
+
+  err = ref - r_eo;
+  std::cout << GridLogMessage << "EO norm diff\t" << norm2(err) << " (" << norm2(ref) << " - " << norm2(r_eo) << ")" << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
+
+
+
+  Dwc_compact.Meooe(src_e, r_o);
+  std::cout << GridLogMessage << "Applied Meo" << std::endl;
+  Dwc_compact.Meooe(src_o, r_e);
+  std::cout << GridLogMessage << "Applied Moe" << std::endl;
+  Dwc_compact.Dhop(src, ref, DaggerNo);
+
+  setCheckerboard(r_eo, r_o);
+  setCheckerboard(r_eo, r_e);
+
+  err = ref - r_eo;
+  std::cout << GridLogMessage << "EO norm diff compact\t" << norm2(err) << " (" << norm2(ref) << " - " << norm2(r_eo) << ")" << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
+
+
+  std::cout << GridLogMessage << "==============================================================" << std::endl;
+  std::cout << GridLogMessage << "= Test Ddagger is the dagger of D by requiring                " << std::endl;
+  std::cout << GridLogMessage << "=  < phi | Deo | chi > * = < chi | Deo^dag| phi>  " << std::endl;
+  std::cout << GridLogMessage << "==============================================================" << std::endl;
+
+  FermionField chi_e(&RBGrid);
+  FermionField chi_o(&RBGrid);
+
+  FermionField dchi_e(&RBGrid);
+  FermionField dchi_o(&RBGrid);
+
+  FermionField phi_e(&RBGrid);
+  FermionField phi_o(&RBGrid);
+
+  FermionField dphi_e(&RBGrid);
+  FermionField dphi_o(&RBGrid);
+
+  pickCheckerboard(Even, chi_e, chi);
+  pickCheckerboard(Odd, chi_o, chi);
+  pickCheckerboard(Even, phi_e, phi);
+  pickCheckerboard(Odd, phi_o, phi);
+
+  Dwc.Meooe(chi_e, dchi_o);
+  Dwc.Meooe(chi_o, dchi_e);
+  Dwc.MeooeDag(phi_e, dphi_o);
+  Dwc.MeooeDag(phi_o, dphi_e);
+
+  ComplexD pDce = innerProduct(phi_e, dchi_e);
+  ComplexD pDco = innerProduct(phi_o, dchi_o);
+  ComplexD cDpe = innerProduct(chi_e, dphi_e);
+  ComplexD cDpo = innerProduct(chi_o, dphi_o);
+
+  std::cout << GridLogMessage << "e " << pDce << " " << cDpe << std::endl;
+  std::cout << GridLogMessage << "o " << pDco << " " << cDpo << std::endl;
+
+  std::cout << GridLogMessage << "pDce - conj(cDpo) " << pDce - conj(cDpo) << std::endl;
+  std::cout << GridLogMessage << "pDco - conj(cDpe) " << pDco - conj(cDpe) << std::endl;
+
+  Dwc_compact.Meooe(chi_e, dchi_o);
+  Dwc_compact.Meooe(chi_o, dchi_e);
+  Dwc_compact.MeooeDag(phi_e, dphi_o);
+  Dwc_compact.MeooeDag(phi_o, dphi_e);
+
+  pDce = innerProduct(phi_e, dchi_e);
+  pDco = innerProduct(phi_o, dchi_o);
+  cDpe = innerProduct(chi_e, dphi_e);
+  cDpo = innerProduct(chi_o, dphi_o);
+
+  std::cout << GridLogMessage << "e compact " << pDce << " " << cDpe << std::endl;
+  std::cout << GridLogMessage << "o compact " << pDco << " " << cDpo << std::endl;
+
+  std::cout << GridLogMessage << "pDce - conj(cDpo) compact " << pDce - conj(cDpo) << std::endl;
+  std::cout << GridLogMessage << "pDco - conj(cDpe) compact " << pDco - conj(cDpe) << std::endl;
+
+  std::cout << GridLogMessage << "==============================================================" << std::endl;
+  std::cout << GridLogMessage << "= Test MeeInv Mee = 1   (if csw!=0)                           " << std::endl;
+  std::cout << GridLogMessage << "==============================================================" << std::endl;
+
+  pickCheckerboard(Even, chi_e, chi);
+  pickCheckerboard(Odd, chi_o, chi);
+
+  Dwc.Mooee(chi_e, src_e);
+  Dwc.MooeeInv(src_e, phi_e);
+
+  Dwc.Mooee(chi_o, src_o);
+  Dwc.MooeeInv(src_o, phi_o);
+
+  setCheckerboard(phi, phi_e);
+  setCheckerboard(phi, phi_o);
+
+  err = phi - chi;
+  std::cout << GridLogMessage << "norm diff " << norm2(err) << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
+
+  Dwc_compact.Mooee(chi_e, src_e);
+  Dwc_compact.MooeeInv(src_e, phi_e);
+
+  Dwc_compact.Mooee(chi_o, src_o);
+  Dwc_compact.MooeeInv(src_o, phi_o);
+
+  setCheckerboard(phi, phi_e);
+  setCheckerboard(phi, phi_o);
+
+  err = phi - chi;
+  std::cout << GridLogMessage << "norm diff compact " << norm2(err) << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
+
+  std::cout << GridLogMessage << "==============================================================" << std::endl;
+  std::cout << GridLogMessage << "= Test MeeDag MeeInvDag = 1    (if csw!=0)                    " << std::endl;
+  std::cout << GridLogMessage << "==============================================================" << std::endl;
+
+  pickCheckerboard(Even, chi_e, chi);
+  pickCheckerboard(Odd, chi_o, chi);
+
+  Dwc.MooeeDag(chi_e, src_e);
+  Dwc.MooeeInvDag(src_e, phi_e);
+
+  Dwc.MooeeDag(chi_o, src_o);
+  Dwc.MooeeInvDag(src_o, phi_o);
+
+  setCheckerboard(phi, phi_e);
+  setCheckerboard(phi, phi_o);
+
+  err = phi - chi;
+  std::cout << GridLogMessage << "norm diff " << norm2(err) << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
+
+  Dwc_compact.MooeeDag(chi_e, src_e);
+  Dwc_compact.MooeeInvDag(src_e, phi_e);
+
+  Dwc_compact.MooeeDag(chi_o, src_o);
+  Dwc_compact.MooeeInvDag(src_o, phi_o);
+
+  setCheckerboard(phi, phi_e);
+  setCheckerboard(phi, phi_o);
+
+  err = phi - chi;
+  std::cout << GridLogMessage << "norm diff compact " << norm2(err) << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
+
+  std::cout << GridLogMessage << "==============================================================" << std::endl;
+  std::cout << GridLogMessage << "= Test MeeInv MeeDag = 1      (if csw!=0)                     " << std::endl;
+  std::cout << GridLogMessage << "==============================================================" << std::endl;
+
+  pickCheckerboard(Even, chi_e, chi);
+  pickCheckerboard(Odd, chi_o, chi);
+
+  Dwc.MooeeDag(chi_e, src_e);
+  Dwc.MooeeInv(src_e, phi_e);
+
+  Dwc.MooeeDag(chi_o, src_o);
+  Dwc.MooeeInv(src_o, phi_o);
+
+  setCheckerboard(phi, phi_e);
+  setCheckerboard(phi, phi_o);
+
+  err = phi - chi;
+  std::cout << GridLogMessage << "norm diff " << norm2(err) << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
+
+  Dwc_compact.MooeeDag(chi_e, src_e);
+  Dwc_compact.MooeeInv(src_e, phi_e);
+
+  Dwc_compact.MooeeDag(chi_o, src_o);
+  Dwc_compact.MooeeInv(src_o, phi_o);
+
+  setCheckerboard(phi, phi_e);
+  setCheckerboard(phi, phi_o);
+
+  err = phi - chi;
+  std::cout << GridLogMessage << "norm diff compact " << norm2(err) << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
+
+  std::cout << GridLogMessage << "================================================================" << std::endl;
+  std::cout << GridLogMessage << "= Testing gauge covariance Clover term with EO preconditioning  " << std::endl;
+  std::cout << GridLogMessage << "================================================================" << std::endl;
+
+  chi = Zero();
+  phi = Zero();
+  tmp = Zero();
+  pickCheckerboard(Even, chi_e, chi);
+  pickCheckerboard(Odd, chi_o, chi);
+  pickCheckerboard(Even, phi_e, phi);
+  pickCheckerboard(Odd, phi_o, phi);
+
+  Dwc.Mooee(src_e, chi_e);
+  Dwc.Mooee(src_o, chi_o);
+  setCheckerboard(chi, chi_e);
+  setCheckerboard(chi, chi_o);
+  setCheckerboard(src, src_e);
+  setCheckerboard(src, src_o);
+
+  ////////////////////// Gauge Transformation
+  std::vector<int> seeds2({5, 6, 7, 8});
+  GridParallelRNG pRNG2(&Grid);
+  pRNG2.SeedFixedIntegers(seeds2);
+  LatticeColourMatrix Omega(&Grid);
+  LatticeColourMatrix ShiftedOmega(&Grid);
+  LatticeGaugeField U_prime(&Grid);
+  U_prime = Zero();
+  LatticeColourMatrix U_prime_mu(&Grid);
+  U_prime_mu = Zero();
+  SU<Nc>::LieRandomize(pRNG2, Omega, 1.0);
+  for (int mu = 0; mu < Nd; mu++)
+  {
+    U[mu] = peekLorentz(Umu, mu);
+    ShiftedOmega = Cshift(Omega, mu, 1);
+    U_prime_mu = Omega * U[mu] * adj(ShiftedOmega);
+    pokeLorentz(U_prime, U_prime_mu, mu);
+  }
+  /////////////////
+
+  WilsonExpCloverFermionR Dwc_prime(U_prime, Grid, RBGrid, mass, csw_r, csw_t, anis, params);
+  CompactWilsonExpCloverFermionR Dwc_compact_prime(U_prime, Grid, RBGrid, mass, csw_r, csw_t, 1.0, anis, params);
+
+  tmp = Omega * src;
+  pickCheckerboard(Even, src_e, tmp);
+  pickCheckerboard(Odd, src_o, tmp);
+
+  Dwc_prime.Mooee(src_e, phi_e);
+  Dwc_prime.Mooee(src_o, phi_o);
+
+  setCheckerboard(phi, phi_e);
+  setCheckerboard(phi, phi_o);
+
+  err = chi - adj(Omega) * phi;
+  std::cout << GridLogMessage << "norm diff " << norm2(err) << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
+
+  chi = Zero();
+  phi = Zero();
+  tmp = Zero();
+  pickCheckerboard(Even, chi_e, chi);
+  pickCheckerboard(Odd, chi_o, chi);
+  pickCheckerboard(Even, phi_e, phi);
+  pickCheckerboard(Odd, phi_o, phi);
+
+  Dwc_compact.Mooee(src_e, chi_e);
+  Dwc_compact.Mooee(src_o, chi_o);
+  setCheckerboard(chi, chi_e);
+  setCheckerboard(chi, chi_o);
+  setCheckerboard(src, src_e);
+  setCheckerboard(src, src_o);
+
+  tmp = Omega * src;
+  pickCheckerboard(Even, src_e, tmp);
+  pickCheckerboard(Odd, src_o, tmp);
+
+  Dwc_compact_prime.Mooee(src_e, phi_e);
+  Dwc_compact_prime.Mooee(src_o, phi_o);
+
+  setCheckerboard(phi, phi_e);
+  setCheckerboard(phi, phi_o);
+
+  err = chi - adj(Omega) * phi;
+  std::cout << GridLogMessage << "norm diff compact " << norm2(err) << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
+
+  std::cout << GridLogMessage << "=================================================================" << std::endl;
+  std::cout << GridLogMessage << "= Testing gauge covariance Clover term w/o EO preconditioning  " << std::endl;
+  std::cout << GridLogMessage << "================================================================" << std::endl;
+
+  chi = Zero();
+  phi = Zero();
+
+  WilsonFermionR Dw(Umu, Grid, RBGrid, mass, params);
+
+  Dw.M(src, result);
+  Dwc.M(src, chi);
+
+  Dwc_prime.M(Omega * src, phi);
+
+  WilsonFermionR Dw_prime(U_prime, Grid, RBGrid, mass, params);
+  Dw_prime.M(Omega * src, result2);
+
+  err = result - adj(Omega) * result2;
+  std::cout << GridLogMessage << "norm diff Wilson                 " << norm2(err) << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
+  err = chi - adj(Omega) * phi;
+  std::cout << GridLogMessage << "norm diff WilsonExpClover        " << norm2(err) << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
+
+  chi = Zero();
+  phi = Zero();
+
+  Dwc_compact.M(src, chi);
+  Dwc_compact_prime.M(Omega * src, phi);
+
+  err = chi - adj(Omega) * phi;
+  std::cout << GridLogMessage << "norm diff CompactWilsonExpClover " << norm2(err) << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
+
+  std::cout << GridLogMessage << "==========================================================" << std::endl;
+  std::cout << GridLogMessage << "= Testing Mooee(csw=0) Clover to reproduce Mooee Wilson   " << std::endl;
+  std::cout << GridLogMessage << "==========================================================" << std::endl;
+
+  chi = Zero();
+  phi = Zero();
+  err = Zero();
+  WilsonExpCloverFermionR Dwc_csw0(Umu, Grid, RBGrid, mass, 0.0, 0.0, anis, params); //  <-- Notice: csw=0
+
+  pickCheckerboard(Even, phi_e, phi);
+  pickCheckerboard(Odd, phi_o, phi);
+  pickCheckerboard(Even, chi_e, chi);
+  pickCheckerboard(Odd, chi_o, chi);
+
+  Dw.Mooee(src_e, chi_e);
+  Dw.Mooee(src_o, chi_o);
+  Dwc_csw0.Mooee(src_e, phi_e);
+  Dwc_csw0.Mooee(src_o, phi_o);
+
+  setCheckerboard(chi, chi_e);
+  setCheckerboard(chi, chi_o);
+  setCheckerboard(phi, phi_e);
+  setCheckerboard(phi, phi_o);
+  setCheckerboard(src, src_e);
+  setCheckerboard(src, src_o);
+
+  err = chi - phi;
+  std::cout << GridLogMessage << "norm diff " << norm2(err) << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
+
+  chi = Zero();
+  phi = Zero();
+  err = Zero();
+  CompactWilsonExpCloverFermionR Dwc_compact_csw0(Umu, Grid, RBGrid, mass, 0.0, 0.0, 1.0, anis, params); //  <-- Notice: csw=0
+
+  pickCheckerboard(Even, phi_e, phi);
+  pickCheckerboard(Odd, phi_o, phi);
+  pickCheckerboard(Even, chi_e, chi);
+  pickCheckerboard(Odd, chi_o, chi);
+
+  Dw.Mooee(src_e, chi_e);
+  Dw.Mooee(src_o, chi_o);
+  Dwc_compact_csw0.Mooee(src_e, phi_e);
+  Dwc_compact_csw0.Mooee(src_o, phi_o);
+
+  setCheckerboard(chi, chi_e);
+  setCheckerboard(chi, chi_o);
+  setCheckerboard(phi, phi_e);
+  setCheckerboard(phi, phi_o);
+  setCheckerboard(src, src_e);
+  setCheckerboard(src, src_o);
+
+  err = chi - phi;
+  std::cout << GridLogMessage << "norm diff compact " << norm2(err) << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
+
+  std::cout << GridLogMessage << "==========================================================" << std::endl;
+  std::cout << GridLogMessage << "= Testing EO operator is equal to the unprec              " << std::endl;
+  std::cout << GridLogMessage << "==========================================================" << std::endl;
+
+  chi = Zero();
+  phi = Zero();
+  err = Zero();
+
+  pickCheckerboard(Even, phi_e, phi);
+  pickCheckerboard(Odd, phi_o, phi);
+  pickCheckerboard(Even, chi_e, chi);
+  pickCheckerboard(Odd, chi_o, chi);
+
+  // M phi = (Mooee src_e + Meooe src_o , Meooe src_e + Mooee src_o)
+
+  Dwc.M(src, ref); // Reference result from the unpreconditioned operator
+
+  // EO matrix
+  Dwc.Mooee(src_e, chi_e); 
+  Dwc.Mooee(src_o, chi_o);
+  Dwc.Meooe(src_o, phi_e);
+  Dwc.Meooe(src_e, phi_o);
+
+  phi_o += chi_o;
+  phi_e += chi_e;
+
+  setCheckerboard(phi, phi_e);
+  setCheckerboard(phi, phi_o);
+
+  err = ref - phi;
+  std::cout << GridLogMessage << "ref (unpreconditioned operator) diff         : " << norm2(ref) << std::endl;
+  std::cout << GridLogMessage << "phi (EO decomposition)          diff         : " << norm2(phi) << std::endl;
+  std::cout << GridLogMessage << "norm diff                                    : " << norm2(err) << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
+
+  chi = Zero();
+  phi = Zero();
+  err = Zero();
+
+  pickCheckerboard(Even, phi_e, phi);
+  pickCheckerboard(Odd, phi_o, phi);
+  pickCheckerboard(Even, chi_e, chi);
+  pickCheckerboard(Odd, chi_o, chi);
+
+  // M phi = (Mooee src_e + Meooe src_o , Meooe src_e + Mooee src_o)
+
+  Dwc_compact.M(src, ref); // Reference result from the unpreconditioned operator
+
+  // EO matrix
+  Dwc_compact.Mooee(src_e, chi_e); 
+  Dwc_compact.Mooee(src_o, chi_o);
+  Dwc_compact.Meooe(src_o, phi_e);
+  Dwc_compact.Meooe(src_e, phi_o);
+
+  phi_o += chi_o;
+  phi_e += chi_e;
+
+  setCheckerboard(phi, phi_e);
+  setCheckerboard(phi, phi_o);
+
+  err = ref - phi;
+  std::cout << GridLogMessage << "ref (unpreconditioned operator) diff compact : " << norm2(ref) << std::endl;
+  std::cout << GridLogMessage << "phi (EO decomposition)          diff compact : " << norm2(phi) << std::endl;
+  std::cout << GridLogMessage << "norm diff compact                            : " << norm2(err) << std::endl;
+  assert(fabs(norm2(err)) < tolerance);
+
+  Grid_finalize();
+}

tests/hmc/Test_hmc_ScalarActionNxN.cc

@@ -132,8 +132,8 @@ int main(int argc, char **argv) {
 
   // Checkpointer definition
   CheckpointerParameters CPparams(Reader);
-  //TheHMC.Resources.LoadBinaryCheckpointer(CPparams);
+  TheHMC.Resources.LoadBinaryCheckpointer(CPparams);
-  TheHMC.Resources.LoadScidacCheckpointer(CPparams, SPar);
+  //TheHMC.Resources.LoadScidacCheckpointer(CPparams, SPar); this breaks for compilation without lime
 
   RNGModuleParameters RNGpar(Reader);
   TheHMC.Resources.SetRNGSeeds(RNGpar);

tests/hmc/Test_hmc_WG_Production.cc

@@ -74,10 +74,10 @@ int main(int argc, char **argv) {
 
   // Checkpointer definition
   CheckpointerParameters CPparams(Reader);
-  //TheHMC.Resources.LoadNerscCheckpointer(CPparams);
+  TheHMC.Resources.LoadNerscCheckpointer(CPparams);
 
-  // Store metadata in the Scidac checkpointer
+  // Store metadata in the Scidac checkpointer - obviously breaks without LIME
-  TheHMC.Resources.LoadScidacCheckpointer(CPparams, WilsonPar);
+  //TheHMC.Resources.LoadScidacCheckpointer(CPparams, WilsonPar);
 
   RNGModuleParameters RNGpar(Reader);
   TheHMC.Resources.SetRNGSeeds(RNGpar);

tests/lanczos/Test_compressed_lanczos.cc

@@ -37,6 +37,8 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 using namespace std;
 using namespace Grid;
 
+#ifdef HAVE_LIME
+
 template<class Fobj,class CComplex,int nbasis>
 class LocalCoherenceLanczosScidac : public LocalCoherenceLanczos<Fobj,CComplex,nbasis>
 { 
@@ -249,3 +251,11 @@ int main (int argc, char ** argv) {
   Grid_finalize();
 }
 
+#else
+
+int main( void )
+{
+  return 0 ;
+}
+
+#endif // HAVE_LIME_H

tests/lanczos/Test_dwf_compressed_lanczos_reorg.cc

@@ -36,7 +36,8 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 
 using namespace std;
 using namespace Grid;
- ;
+
+#ifdef HAVE_LIME
 
 template<class Fobj,class CComplex,int nbasis>
 class LocalCoherenceLanczosScidac : public LocalCoherenceLanczos<Fobj,CComplex,nbasis>
@@ -250,3 +251,11 @@ int main (int argc, char ** argv) {
   Grid_finalize();
 }
 
+#else
+
+int main( void )
+{
+  return 0 ;
+}
+
+#endif

tests/solver/Test_coarse_even_odd.cc

@@ -93,8 +93,16 @@ int main(int argc, char** argv) {
   //                    Setup of Dirac Matrix and Operator                   //
   /////////////////////////////////////////////////////////////////////////////
 
-  LatticeGaugeField Umu(Grid_f); SU3::HotConfiguration(pRNG_f, Umu);
+  LatticeGaugeField Umu(Grid_f);
-
+#if (Nc==2)
+  SU2::HotConfiguration(pRNG_f, Umu);
+#elif (defined Nc==3)
+  SU3::HotConfiguration(pRNG_f, Umu);
+#elif (defined Nc==4)
+  SU4::HotConfiguration(pRNG_f, Umu);
+#elif (defined Nc==5)
+  SU5::HotConfiguration(pRNG_f, Umu);
+#endif
   RealD checkTolerance = (getPrecision<LatticeFermion>::value == 1) ? 1e-7 : 1e-15;
 
   RealD mass = -0.30;

tests/solver/Test_dwf_mrhs_cg.cc

@@ -34,6 +34,7 @@ using namespace Grid;
 
 int main (int argc, char ** argv)
 {
+#ifdef HAVE_LIME 
   typedef typename DomainWallFermionR::FermionField FermionField; 
   typedef typename DomainWallFermionR::ComplexField ComplexField; 
   typename DomainWallFermionR::ImplParams params; 
@@ -237,4 +238,5 @@ int main (int argc, char ** argv)
   }
 
   Grid_finalize();
+#endif // HAVE_LIME
 }