[llvm] [RFC] [compiler-rt] make MSan up to 20x faster on AMD CPUs (PR #171993)

[Dan Blackwell via llvm-commits]

DanBlackwell wrote:

This is pretty neat! I'd have to think if there's any risks to the change, but seems unlikely. One note is that arm64 has 128 byte cache lines so would want xoring with `80`, but this PR is for x86_64 anyhow.

https://github.com/llvm/llvm-project/pull/171993