[Mlir-commits] [mlir] [mlir][arith] doc updates for ub semantics, and int representations (PR #72932)

[Fehr Mathieu]

math-fehr wrote:

Currently, we are canonicalizing `select` with a DAG of binary operations that propagates poison. The pattern is this one (but for arith rather than LLVM): https://alive2.llvm.org/ce/z/kd787m

With the `select` semantics we are defining here, which are the same one as LLVM, we would have a miscompile since we might introduce poison. So we are in a way overspecializing compared to what we had before, but my take is that the `select` semantics we are defining here are better (and would correspond to the LLVM ones).

https://github.com/llvm/llvm-project/pull/72932