[llvm] [VPlan] Recalculate VPDT in handleUncountableExitsWithSideEffects (PR #203233)

[via llvm-commits]

llvmorg-github-actions[bot] wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-llvm-transforms

Author: Luke Lau (lukel97)

<details>
<summary>Changes</summary>

In the loop before we're modifying the CFG, but this invalidates the
dominator tree. We need to recalculate since we query it later on. I
can't really think of a test case for this, if anything using the
stale dominator tree with the extra branch will make the dominance
queries more conservative. But it's probably something we should fix.


---
Full diff: https://github.com/llvm/llvm-project/pull/203233.diff


1 Files Affected:

- (modified) llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp (+2) 


``````````diff
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
index 3fe8c84658c8f..cd0fddd4489d8 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
@@ -4250,6 +4250,8 @@ static bool handleUncountableExitsWithSideEffects(
     VPBlockUtils::disconnectBlocks(Exit.EarlyExitingVPBB, Exit.EarlyExitVPBB);
   }
 
+  VPDT.recalculate(Plan);
+
   // We can abandon a VPlan entirely if we return false here, so we shouldn't
   // crash if some earlier assumptions on scalar IR don't hold for the vplan
   // version of the loop.

``````````

</details>


https://github.com/llvm/llvm-project/pull/203233