DavidPoliakoff

function_pointer_that_takes_a_string begin_parallel_for_ptr;
function_pointer_that_takes_a_string end_parallel_for_ptr;
void initialize(std::string toolLibrary) {
  void* handle = dlopen(toolLibrary);
  begin_parallel_for_ptr = dlsym(handle, "kokkosp_begin_parallel_for");
  end_parallel_for_ptr = dlsym(handle, "kokkosp_end_parallel_for");
}
void parallel_for(std::string label, boring_parallelism_stuff stuff) {
  if(begin_parallel_for_ptr){
    fence_kokkos(); // this is a _small_ lie

DavidPoliakoff

#pragma omp parallel for
for(int i=0;i<lots;++i){
  // insert psychics here
}

__global__ void cuda(){
  int thread = threadIdx.x +
    blockdim.x * blockIdx.x;
  if(thread < lots){
    // insert psychics here