4 ms·

if this produced SIMD optimised code, better inlining, and just more LLVM features, I would use this!