[llvm-branch-commits] AlwaysInliner: A new inlining algorithm to interleave alloca promotion with inlines. (PR #145613)

[Nikita Popov via llvm-branch-commits]

================

----------------
nikic wrote:

This looks like a simple missed optimization in InstCombine.

My first thought would be to handle this via foldOpIntoPhi for selects: https://github.com/nikic/llvm-project/commit/c7d9b006453275be94894c6afdae5f6c940c1621 But this isn't quite correct if FMF is involved. Probably better to implement a dedicated fold.

https://github.com/llvm/llvm-project/pull/145613