what changes did they make to make the shader compilation faster?
it's compiled with similar optimizations as the kernel and has access to fast instrinsics (sse, avx)