[llvm] IR: Allow vector type in atomic load and store (PR #117625)

[Nikita Popov via llvm-commits]

nikic wrote:

I agree with @jyknight that this needs to come with support for all targets, not just X86. Preferably implemented in a way that does not require defining a hook in every single backend.

> > We shouldn't have this work only on some platforms, but not others.
> 
> There's a ton of IR that will not work on all platforms with varying failure modes, this case isn't special (e.g. 128-bit atomics)

We should always aspire to make generic IR constructs work target-independently. If we have fallen short of that in some area, that's not an excuse to repeat the mistake.

https://github.com/llvm/llvm-project/pull/117625