https://hgpu.org/?p=10683
Automatic run-time mapping of polyhedral computations to heterogeneous devices with memory-size restrictions