[llvm] [SLP] Bail out on store-to-load forwarding hazards (PR #199606)

[Alexey Bataev via llvm-commits]

================
@@ -27318,10 +27324,12 @@ bool SLPVectorizerPass::runImpl(Function &F, ScalarEvolution *SE_,
   DT = DT_;
   AC = AC_;
   DB = DB_;
+  LAIs = LAIs_;
   DL = &F.getDataLayout();
 
   Stores.clear();
   GEPs.clear();
+  StlfConflictCache.clear();
----------------
alexey-bataev wrote:

I think we need to clear it upon each vectorization attempt, because some of the loads might be reconstructed

https://github.com/llvm/llvm-project/pull/199606