[IndVarSimplify][NFC] Removed mayThrow from if-condition in predicateLoopExits of IndVarSimplify

[vdsered]

Summary

Instruction has mayHaveSideEffects method that returns true if mayThrow return true because this is called internally in the first method

[xbolva00]

llvm/lib/Transforms/Scalar/IndVarSimplify.cpp

1673

Or better, implement TODO?

[xbolva00]

llvm/lib/Transforms/Scalar/IndVarSimplify.cpp

1672

Maybe we should have some scanlimit for huge BBs?

[nikic]

llvm/lib/Transforms/Scalar/IndVarSimplify.cpp

1673

Ideally we should be using SCEV loop properties here, which are cached. Unfortunately, the SCEV notion of "side effect" differs from what is needed here (SCEV ignores simple writes).

We could at least use loopHasNoAbnormalExits() though, which uses isGuaranteedToTransfer(). And I do believe that one is needed here -- I don't think the transform is valid if the loop contains a non-willreturn readonly call. It would be good to add a test for that case.

[reames]

LGTM.

(I'm ignoring the side discussion as it doesn't appear to relate to the patch which is simply a small cleanup.)

[vdsered]

@reames, I don't have commit access. Can you commit the patch?

llvm/lib/Transforms/Scalar/IndVarSimplify.cpp

bool IndVarSimplify::predicateLoopExits(Loop *L, SCEVExpander &Rewriter) {
   // predicatable.  Given that, we know we have at least one exit we can
   // predicate if the loop is doesn't have side effects and doesn't have any
   // implicit exits (because then our exact BTC isn't actually exact).
   // @Reviewers - As structured, this is O(I^2) for loop nests.  Any
   // suggestions on how to improve this?  I can obviously bail out for outer
   // loops, but that seems less than ideal.  MemorySSA can find memory writes,
   // is that enough for *all* side effects?
   for (BasicBlock *BB : L->blocks())
     for (auto &I : *BB)
       // TODO:isGuaranteedToTransfer
-      if (I.mayHaveSideEffects() || I.mayThrow())
+      if (I.mayHaveSideEffects())
         return false;
 
   bool Changed = false;
   // Finally, do the actual predication for all predicatable blocks.  A couple
   // of notes here:
   // 1) We don't bother to constant fold dominated exits with identical exit
   //    counts; that's simply a form of CSE/equality propagation and we leave
   //    it for dedicated passes.