https://hgpu.org/?p=6594
Affine Vector Cache for memory bandwidth savings