I would like to do direct copies from pointers passed from fortran to GPU framebuffer and just completely avoid making instances in sys mem. My current work around is to just write Regent tasks with gpu variants that call cudaMemcpy directly. But we should have support for external attach in Regent.
I would like to do direct copies from pointers passed from fortran to GPU framebuffer and just completely avoid making instances in sys mem. My current work around is to just write Regent tasks with gpu variants that call cudaMemcpy directly. But we should have support for external attach in Regent.