https://hgpu.org/?p=5269
Implementing the PGI Accelerator model