With the new GPU implementation, we no longer offload the whole simulation to the accelerator. Instead we operate on host except when executing a loop in parallel.
This as well as the clever use of pragmas have made the former offloading infrastructure redundant. These will be removed.
With the new GPU implementation, we no longer offload the whole simulation to the accelerator. Instead we operate on host except when executing a loop in parallel.
This as well as the clever use of pragmas have made the former offloading infrastructure redundant. These will be removed.