Use CORENEURON_PREFER_OPENMP_OFFLOAD.

Author: olupton

CMake/OpenAccHelper.cmake

@@ -55,7 +55,7 @@ if(CORENRN_ENABLE_GPU)
     if(CORENRN_ENABLE_OPENMP AND CORENRN_ENABLE_OPENMP_OFFLOAD)
       # Enable OpenMP target offload to GPU and if both OpenACC and OpenMP directives are available
       # for a region then prefer OpenMP.
-      add_compile_definitions(CORENRN_PREFER_OPENMP_OFFLOAD)
+      add_compile_definitions(CORENEURON_PREFER_OPENMP_OFFLOAD)
     endif()
   elseif(${CMAKE_CXX_COMPILER_ID} STREQUAL "Clang")
     set(NVHPC_ACC_COMP_FLAGS "-fopenmp -fopenmp-targets=nvptx64-nvidia-cuda -Wno-unknown-cuda-version -I${CUDAToolkit_INCLUDE_DIRS}")

coreneuron/gpu/nrn_acc_manager.cpp

@@ -27,7 +27,7 @@
 #ifdef _OPENACC
 #include <openacc.h>
 #endif
-#ifdef CORENRN_PREFER_OPENMP_OFFLOAD
+#ifdef CORENEURON_PREFER_OPENMP_OFFLOAD
 #include <omp.h>
 #endif
 
@@ -1435,7 +1435,7 @@ void init_gpu() {
 
     int device_num = local_rank % num_devices_per_node;
     acc_set_device_num(device_num, device_type);
-#ifdef CORENRN_PREFER_OPENMP_OFFLOAD
+#ifdef CORENEURON_PREFER_OPENMP_OFFLOAD
     omp_set_default_device(device_num);
 #endif
 

coreneuron/network/partrans.cpp

@@ -114,7 +114,7 @@ void nrnthread_v_transfer(NrnThread* _nt) {
     int* insrc_indices = ttd.insrc_indices.data();
     double* tar_data = _nt->_data;
     // last element in the displacement vector gives total length
-#if defined(_OPENACC) && !defined(CORENRN_PREFER_OPENMP_OFFLOAD)
+#if defined(_OPENACC) && !defined(CORENEURON_PREFER_OPENMP_OFFLOAD)
     int n_insrc_buf = insrcdspl_[nrnmpi_numprocs];
     int ndata = _nt->_ndata;
 #endif

coreneuron/permute/cellorder.cpp

@@ -598,7 +598,7 @@ void solve_interleaved2(int ith) {
         int* strides = ii.stride;           // sum ncycles of these (bad since ncompart/warpsize)
         int* rootbegin = ii.firstnode;      // nwarp+1 of these
         int* nodebegin = ii.lastnode;       // nwarp+1 of these
-#if defined(_OPENACC) && !defined(CORENRN_PREFER_OPENMP_OFFLOAD)
+#if defined(_OPENACC) && !defined(CORENEURON_PREFER_OPENMP_OFFLOAD)
         int nstride = stridedispl[nwarp];
 #endif
         nrn_pragma_acc(parallel loop gang vector vector_length(

coreneuron/utils/offload.hpp

@@ -9,10 +9,10 @@
 #include <cstddef>
 
 #define nrn_pragma_stringify(x) #x
-#if defined(CORENEURON_ENABLE_GPU) && defined(CORENRN_PREFER_OPENMP_OFFLOAD) && defined(_OPENMP)
+#if defined(CORENEURON_ENABLE_GPU) && defined(CORENEURON_PREFER_OPENMP_OFFLOAD) && defined(_OPENMP)
 #define nrn_pragma_acc(x)
 #define nrn_pragma_omp(x) _Pragma(nrn_pragma_stringify(omp x))
-#elif defined(CORENEURON_ENABLE_GPU) && !defined(CORENRN_PREFER_OPENMP_OFFLOAD) && \
+#elif defined(CORENEURON_ENABLE_GPU) && !defined(CORENEURON_PREFER_OPENMP_OFFLOAD) && \
     defined(_OPENACC)
 #define nrn_pragma_acc(x) _Pragma(nrn_pragma_stringify(acc x))
 #define nrn_pragma_omp(x)