Does the shader automatically get parallelized by the GPU or do you have to choose certain parameters (block size, etc.) and write it in a parallelizable way?