[InstSimplify] fold exact divide to poison if it is known to not divide evenly
This is related to the discussion in D140665. I was looking over the demanded
bits implementation in IR and noticed that we just bail out of a potential
fold if a udiv is exact:
https://github.com/llvm/llvm-project/blob/
82be8a1d2b00f6e89096b86f670a8be894c7b9e6/llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp#L799
Also, see tests added with
7f0c11509e8f.
Then, I saw that we could lose a fold to poison if we zap the exact with that
transform, so this patch tries to catch that as a preliminary step.
Alive2 proofs:
https://alive2.llvm.org/ce/z/zCjKM7
https://alive2.llvm.org/ce/z/-tz_RK (trailing zeros must be "less-than")
https://alive2.llvm.org/ce/z/c9CMsJ (general proof and specific example)
Differential Revision: https://reviews.llvm.org/D140733