[llvm] [ModuloSchedule] Implement modulo variable expansion for pipelining (PR #65609)

[James Nagurne via llvm-commits]

DragonDisciple wrote:

I am by no means an expert in AArch64, so someone with a better handle on it than I should review the implementations and tests.

However, this looks fine, and the results seem generally positive. I'd like to get this upstream so my team can pull it down and play with it. One thing I think we'd like to improve (and share upstream, of course!) is handling loop-carried dependences such as accumulations.

We have a pre-pipeliner unroller that does this, but think that using this method is likely cleaner.

tl;dr: LGTM

https://github.com/llvm/llvm-project/pull/65609