Clear NewGEPBases after finish using them in CodeGenPrep pass

[yuyichao]

Summary

AFAICT all other set/map are correctly cleared in runOnFunction.

With assertion enabled this causes a crash when the module is freed and potentially if a later pass delete the instruction (not observed in real world though). Without assertion this can potentially cause confusing result when running on a new Function/Module.

[loladiro]

LGTM. Could we have a test for this? There is the -run-twice option to the various tools, which is designed to catch issues like this.

[yuyichao]

Does run-twice free the module? I’ll have to look for the repro again but I was only able to observe a crash when the module is free’d and it doesn’t even require running the pass again...

[loladiro]

Well, the module if freed eventually. Is this issue whether the module if freed before the pass manager? If so I think run-twice will (incidentally also do that). I just though run-twice would catch it, because forgetting to clear such things usually causes incorrect behavior the next time around.

llvm/lib/CodeGen/CodeGenPrepare.cpp

while (MadeChange) {
     EverMadeChange |= MadeChange;
     SeenChainsForSExt.clear();
     ValToSExtendedUses.clear();
     RemovedInsts.clear();
     LargeOffsetGEPMap.clear();
     LargeOffsetGEPID.clear();
   }
 
+  NewGEPBases.clear();
   SunkAddrs.clear();
 
   if (!DisableBranchOpts) {
     MadeChange = false;
     // Use a set vector to get deterministic iteration order. The order the
     // blocks are removed may affect whether or not PHI nodes in successors
     // are removed.
     SmallSetVector<BasicBlock*, 8> WorkList;