[llvm] [X86] Enable unaligned loads on x86 using cmpxchg (PR #142645)

[Craig Topper via llvm-commits]

topperc wrote:

> the actual condition is that the value must fit in a single cache line which I think means no straddling 64-byte boundaries (aligned values always fit in a single cache line)
> 
> How can I ensure this?

LLVM has no way to know that other than emitting a runtime branch to check the alignment.

https://github.com/llvm/llvm-project/pull/142645