you don't really need that tbh. you can get pretty good speedups using standard (vector) intrinsics. the new algorithms are (mostly) modular linear algebra (+ some concept of "noise").