https://hgpu.org/?p=17973
Efficient SIMD Vectorization for Hashing in OpenCL