llvm-project

14cc641b - [msan] Strengthen LLVM/NEON floating-point<->int propagation (#196875)

[msan] Strengthen LLVM/NEON floating-point<->int propagation (#196875) This generalizes handleNEONVectorConvertIntrinsic() to apply it to LLVM cross-platform floating point<->int conversion intrinsics. The handler uses an all-or-nothing approach: if any bit of an input element is uninitialized, the corresponding output element is fully uninitialized. This approximates how a single bit flip in an integer can affect multiple bits of the equivalent floating-point (likewise for FP to int). This implements the future work suggested in https://github.com/llvm/llvm-project/pull/196429.