https://hgpu.org/?p=19341
Static Analysis and Dynamic Adaptation of Parallelism