[llvm] [InstCombine] Avoid computeKnownBits when trying to fold assumes (PR #203550)

[Nikita Popov via llvm-commits]

https://github.com/nikic approved this pull request.

The change looks fine, but I think implementation-wise it would have been cleaner to just change the first condition to `match(IIOperand, m_One())`, rather than combining the cases.

https://github.com/llvm/llvm-project/pull/203550