https://hgpu.org/?p=2455
Qilin: exploiting parallelism on heterogeneous multiprocessors with adaptive mapping