Fix issue with Aurora compilers

Grid/communicator/Communicator_none.cc

@@ -27,6 +27,8 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 /*  END LEGAL */
 #include <Grid/GridCore.h>
 
+void GridAbort(void) { abort(); }
+
 NAMESPACE_BEGIN(Grid);
 
 ///////////////////////////////////////////////////////////////////////////////////////////////////
@@ -34,7 +36,6 @@ NAMESPACE_BEGIN(Grid);
 ///////////////////////////////////////////////////////////////////////////////////////////////////
 Grid_MPI_Comm       CartesianCommunicator::communicator_world;
 
-void GridAbort(void) { abort(); }
 
 void CartesianCommunicator::Init(int *argc, char *** arv)
 {

Grid/cshift/Cshift_mpi.h

@@ -124,7 +124,7 @@ template<class vobj> void Cshift_simple(Lattice<vobj>& ret,const Lattice<vobj> &
     void *hsend_buf = (void *)&hrhs[0];
     void *hrecv_buf = (void *)&hret[0];
 
-    acceleratorCopyFromDevice(&send_buf[0],&hsend_buf[0],bytes);
+    acceleratorCopyFromDevice(send_buf,hsend_buf,bytes);
 
     grid->SendToRecvFrom(hsend_buf,
 			 xmit_to_rank,
@@ -132,7 +132,7 @@ template<class vobj> void Cshift_simple(Lattice<vobj>& ret,const Lattice<vobj> &
 			 recv_from_rank,
 			 bytes);
 
-    acceleratorCopyToDevice(&hrecv_buf[0],&recv_buf[0],bytes);
+    acceleratorCopyToDevice(hrecv_buf,recv_buf,bytes);
 #endif
   }
 }

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

@@ -411,7 +411,7 @@ void WilsonKernels<Impl>::DhopDirKernel( StencilImpl &st, DoubledGaugeField &U,S
 #undef LoopBody
 }
 
-#ifdef GRID_SYCL
+#if 0
 extern "C" {
     ulong SYCL_EXTERNAL __attribute__((overloadable)) intel_get_cycle_counter( void );
     uint  SYCL_EXTERNAL __attribute__((overloadable)) intel_get_active_channel_mask( void );

Grid/qcd/action/gauge/GaugeImplTypes.h

@@ -138,10 +138,13 @@ public:
     //auto start = std::chrono::high_resolution_clock::now();
     autoView(U_v,U,AcceleratorWrite);
     autoView(P_v,P,AcceleratorRead);
-    accelerator_for(ss, P.Grid()->oSites(),1,{
+    typedef typename Field::vector_object vobj;
+    const int Nsimd = vobj::Nsimd();
+    accelerator_for(ss, P.Grid()->oSites(),Nsimd,{
       for (int mu = 0; mu < Nd; mu++) {
-          U_v[ss](mu) = Exponentiate(P_v[ss](mu), ep, Nexp) * U_v[ss](mu);
-          U_v[ss](mu) = Group::ProjectOnGeneralGroup(U_v[ss](mu));
+	auto tmp = Exponentiate(P_v(ss)(mu), ep, Nexp) * U_v(ss)(mu);
+	tmp      = Group::ProjectOnGeneralGroup(tmp);
+	coalescedWrite(U_v[ss](mu),tmp);
       }
     });
    //auto end = std::chrono::high_resolution_clock::now();

Grid/qcd/smearing/GaugeConfigurationMasked.h

@@ -291,8 +291,8 @@ public:
     int idx=0;
     for(int mu=0;mu<4;mu++){
     for(int nu=0;nu<4;nu++){
-      if ( mu!=nu) GRID_ASSERT(this->StoutSmearing->SmearRho[idx]==rho);
-      else         GRID_ASSERT(this->StoutSmearing->SmearRho[idx]==0.0);
+      if ( mu!=nu) assert(this->StoutSmearing->SmearRho[idx]==rho);
+      else         assert(this->StoutSmearing->SmearRho[idx]==0.0);
       idx++;
     }}
     //////////////////////////////////////////////////////////////////
@@ -825,6 +825,7 @@ public:
   virtual void fill_smearedSet(GaugeField &U)
   {
     this->ThinLinks = &U;  // attach the smearing routine to the field U
+    std::cout << GridLogMessage << " fill_smearedSet " << WilsonLoops<PeriodicGimplR>::avgPlaquette(U) << std::endl;
 
     // check the pointer is not null
     if (this->ThinLinks == NULL)
@@ -846,6 +847,8 @@ public:
 	ApplyMask(smeared_A,smearLvl);
 	smeared_B = previous_u;
 	ApplyMask(smeared_B,smearLvl);
+	std::cout << GridLogMessage << " smeared_A " << norm2(smeared_A) << std::endl;
+	std::cout << GridLogMessage << " smeared_B " << norm2(smeared_B) << std::endl;
 	// Replace only the masked portion
 	this->SmearedSet[smearLvl] = previous_u-smeared_B + smeared_A;
         previous_u = this->SmearedSet[smearLvl];
@@ -934,10 +937,10 @@ public:
   SmearedConfigurationMasked(GridCartesian* _UGrid, unsigned int Nsmear, Smear_Stout<Gimpl>& Stout)
     : SmearedConfiguration<Gimpl>(_UGrid, Nsmear,Stout)
   {
-    GRID_ASSERT(Nsmear%(2*Nd)==0); // Or multiply by 8??
+    assert(Nsmear%(2*Nd)==0); // Or multiply by 8??
 
     // was resized in base class
-    GRID_ASSERT(this->SmearedSet.size()==Nsmear);
+    assert(this->SmearedSet.size()==Nsmear);
     
     GridRedBlackCartesian * UrbGrid;
     UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(_UGrid);

Grid/qcd/smearing/JacobianAction.h

@@ -54,7 +54,7 @@ public:
   // Usual cases are not used
   //////////////////////////////////
   virtual void refresh(const GaugeField &U, GridSerialRNG &sRNG, GridParallelRNG &pRNG){ GRID_ASSERT(0);};
-  virtual RealD S(const GaugeField &U) { GRID_ASSERT(0); }
+  virtual RealD S(const GaugeField &U) { GRID_ASSERT(0); return 0; }
   virtual void deriv(const GaugeField &U, GaugeField &dSdU) { GRID_ASSERT(0);  }
 
   //////////////////////////////////

Grid/stencil/Stencil.h

@@ -751,7 +751,7 @@ public:
     obj.xbytes      = xbytes;
     obj.rbytes      = rbytes;
     obj.cb          = cb;
-
+    
     for(int i=0;i<CachedTransfers.size();i++){
       if (   (CachedTransfers[i].direction  ==direction)
 	   &&(CachedTransfers[i].OrthogPlane==OrthogPlane)
@@ -763,11 +763,13 @@ public:
 	     ){
 	// FIXME worry about duplicate with partial compression
 	// Wont happen as DWF has no duplicates, but...
-	AddCopy(CachedTransfers[i].recv_buf,recv_buf,rbytes);
-	return 1;
+	//	AddCopy(CachedTransfers[i].recv_buf,recv_buf,rbytes);
+	//	std::cout << "Duplicate dir " <<direction<<" "<<" OrthogPlane "<<OrthogPlane<<" Dest"<<DestProc <<" xbytes " <<xbytes<<" lane "<< lane<<" cb "<<cb<<std::endl;
+	return 0;
+	
+	//	return 1;
       }
     }
-
     CachedTransfers.push_back(obj);
     return 0;
   }

systems/Aurora/config-command

@@ -10,7 +10,7 @@ export HDF5=/opt/cray/pe/hdf5/1.12.2.3/gnu/9.1
 	--disable-gparity \
 	--disable-fermion-reps \
 	--enable-shm=nvlink \
-	--enable-checksum-comms=yes \
+	--enable-checksum-comms=no \
 	--enable-log-views=yes \
 	--enable-accelerator=sycl \
 	--enable-accelerator-aware-mpi=no \

tests/forces/Test_fthmc.cc

@@ -62,6 +62,7 @@ void ForceTest(Action<LatticeGaugeField> &action,ConfigurationBase<LatticeGaugeF
   
   Gimpl::generate_momenta(P,sRNG,RNG4);
   //  Filter.applyFilter(P);
+  std::cout << GridLogMessage << "Initial momenta " << norm2(P) << std::endl;
 
   action.refresh(smU,sRNG,RNG4);
 
@@ -70,6 +71,8 @@ void ForceTest(Action<LatticeGaugeField> &action,ConfigurationBase<LatticeGaugeF
   std::cout << GridLogMessage << "+++++++++++++++++++++++++++++++++++++++++++++++++++++++++"<<std::endl;
 
   RealD S1 = action.S(smU);
+  std::cout << GridLogMessage << "Initial action " << S1 << std::endl;
+
 
   Gimpl::update_field(P,U,eps);
   smU.set_Field(U);
@@ -80,6 +83,7 @@ void ForceTest(Action<LatticeGaugeField> &action,ConfigurationBase<LatticeGaugeF
   action.deriv(smU,UdSdU);
   UdSdU = Ta(UdSdU);
   //  Filter.applyFilter(UdSdU);
+  std::cout << GridLogMessage << "Derivative " << norm2(UdSdU) << std::endl;
 
   DumpSliceNorm("Force",UdSdU,Nd-1);
   
@@ -91,6 +95,7 @@ void ForceTest(Action<LatticeGaugeField> &action,ConfigurationBase<LatticeGaugeF
   std::cout << GridLogMessage << "+++++++++++++++++++++++++++++++++++++++++++++++++++++++++"<<std::endl;
   
   RealD S2 = action.S(smU);
+  std::cout << GridLogMessage << "Final action " << S1 << std::endl;
 
   // Use the derivative
   LatticeComplex dS(UGrid); dS = Zero();
@@ -145,6 +150,8 @@ int main (int argc, char ** argv)
   GridParallelRNG          RNG4(UGrid);  RNG4.SeedFixedIntegers(seeds);
   SU<Nc>::HotConfiguration(RNG4,U);
 #endif
+  std::cout << GridLogMessage << "Initial plaquette: " << WilsonLoops<PeriodicGimplR>::avgPlaquette(U) << std::endl;
+
 
   
   WilsonGaugeActionR  PlaqAction(6.0);