The performance of saturating operations or adding intrinsics

Isn’t that already possible using llvmcall?