[llvm] [InstCombine] Relax the conditons of fold of `ucmp`/`scmp` into phi by allowing the phi node to use the result of `ucmp`/`scmp` more than once (PR #109593)

[Nikita Popov via llvm-commits]

================
@@ -1830,7 +1830,7 @@ Instruction *InstCombinerImpl::foldOpIntoPhi(Instruction &I, PHINode *PN) {
     // optimization is profitable, but also to avoid creating a potentially
     // invalid phi node when we have a multi-edge in the CFG.
----------------
nikic wrote:

You can drop the NOTE now.

https://github.com/llvm/llvm-project/pull/109593