https://hgpu.org/?p=21741
Autotuning for Automatic Parallelization on Heterogeneous Systems