CPU compile ordering is important

Grid/lattice/Lattice_base.h

@@ -236,11 +236,18 @@ public:
   template<class sobj> inline Lattice<vobj> & operator = (const sobj & r){
     vobj vtmp;
     vtmp = r;
+#if 1
+    auto me  = View(CpuWrite);
+    thread_for(ss,me.size(),{
+       me[ss]= r;
+      });
+#else    
     auto me  = View(AcceleratorWrite);
     accelerator_for(ss,me.size(),vobj::Nsimd(),{
 	auto stmp=coalescedRead(vtmp);
 	coalescedWrite(me[ss],stmp);
     });
+#endif    
     me.ViewClose();
     return *this;
   }

Grid/threads/Accelerator.h

@@ -539,12 +539,6 @@ inline void acceleratorCopySynchronise(void) { auto discard=hipStreamSynchronize
 
 #endif
 
-inline void acceleratorCopyDeviceToDevice(void *from,void *to,size_t bytes)
-{
-  acceleratorCopyDeviceToDeviceAsynch(from,to,bytes);
-  acceleratorCopySynchronise();
-}
-
 //////////////////////////////////////////////
 // CPU Target - No accelerator just thread instead
 //////////////////////////////////////////////
@@ -553,7 +547,6 @@ inline void acceleratorCopyDeviceToDevice(void *from,void *to,size_t bytes)
 
 #undef GRID_SIMT
 
-
 inline void acceleratorMem(void)
 {
   /*
@@ -656,6 +649,12 @@ accelerator_inline void acceleratorFence(void)
   return;
 }
 
+inline void acceleratorCopyDeviceToDevice(void *from,void *to,size_t bytes)
+{
+  acceleratorCopyDeviceToDeviceAsynch(from,to,bytes);
+  acceleratorCopySynchronise();
+}
+
 template<class T> void acceleratorPut(T& dev,T&host)
 {
   acceleratorCopyToDevice(&host,&dev,sizeof(T));

systems/Tursa/config-command

@@ -2,11 +2,11 @@
     --enable-comms=mpi \
     --enable-simd=GPU \
     --enable-shm=nvlink \
-    --enable-gen-simd-width=64 \
     --enable-accelerator=cuda \
+    --enable-gen-simd-width=64 \
+    --disable-gparity \
     --with-lime=/mnt/lustre/tursafs1/home/tc002/tc002/dc-boyl1/spack/spack/opt/spack/linux-rhel8-zen/gcc-8.4.1/c-lime-2-3-9-e6wxqrid6rqmd45z7n32dxkvkykpvyez \
-    --enable-accelerator-cshift \
     --disable-unified \
     CXX=nvcc \
-    LDFLAGS="-cudart shared " \
+    LDFLAGS="-cudart shared -lcublas " \
-    CXXFLAGS="-ccbin mpicxx -gencode arch=compute_80,code=sm_80 -std=c++14 -cudart shared"
+    CXXFLAGS="-ccbin mpicxx -gencode arch=compute_80,code=sm_80 -std=c++17 -cudart shared --diag-suppress 177,550,611"

systems/Tursa/sourceme.sh

@@ -1,6 +1,7 @@
-module load cuda/11.4.1  openmpi/4.1.1-cuda11.4.1  ucx/1.12.0-cuda11.4.1  
+module load cuda/12.3 
-#module load cuda/11.4.1 openmpi/4.1.1 ucx/1.10.1
+module load ucx/1.15.0-cuda12.3  
-export PREFIX=/home/tc002/tc002/shared/env/prefix/
+module load openmpi/4.1.5-cuda12.3
-export LD_LIBRARY_PATH=$PREFIX/lib/:$LD_LIBRARY_PATH
+source /home/dp207/dp207/shared/env/production/env-base.sh 
+source /home/dp207/dp207/shared/env/production/env-gpu.sh 
 unset SBATCH_EXPORT