[llvm] [InstCombine] add control for SimplifyDemandedVectorElts depth limit (PR #113717)

[Princeton Ferro via llvm-commits]

Prince781 wrote:

Hi @nikic,

> What is the depth cutoff that you plan to use?
1, although we may have other values we'd like to use.

> And do you have a sharable test case demonstrating the compile-time issue?
I'm not able to share anything yet, unfortunately. We have a few internal kernels that see a lot of improvement when this aspect of InstCombine is dialed.

> I'm not opposed to having the option, but at the same time it would be nice to address the issue upstream, as other people may also encounter the issue.
I agree. I'll actually be sending a few more patches to improve InstCombine compile time with vector operations soon.

https://github.com/llvm/llvm-project/pull/113717