https://hgpu.org/?p=5616
Improving SIMD efficiency for parallel Monte Carlo light transport on the GPU