[llvm] [NaryReassociate] Check to avoid introducing poison when reusing SCEVs (PR #98156)

[Nikita Popov via llvm-commits]

================
@@ -559,20 +559,31 @@ NaryReassociatePass::findClosestMatchingDominator(const SCEV *CandidateExpr,
     return nullptr;
 
   auto &Candidates = Pos->second;
+  const auto *FullExpr = SE->getSCEV(Dominatee);
----------------
nikic wrote:

Why are we using the SCEV of Dominatee here, rather than CandidateExpr? I'm not familiar with this pass, but from a cursory look what we're actually replacing is CandidateExpr, while Dominatee acts as a context instruction.

https://github.com/llvm/llvm-project/pull/98156