https://hgpu.org/?p=7169
Performance Portability with the Chapel Language