https://hgpu.org/?p=28357
Reducing branch divergence to speed up parallel execution of unit testing on GPUs