Revert "[VPlan] Remove unnecessary DomTreeUpdater flush (NFC)."

[aeubanks]

This reverts commit 2e33734.

Causes breakages internally, will post reproducer later.

[llvmbot]

@llvm/pr-subscribers-llvm-transforms

@llvm/pr-subscribers-vectorizers

Author: Arthur Eubanks (aeubanks)

Changes

This reverts commit 2e33734.

Causes breakages internally, will post reproducer later.

---

Full diff: https://github.com/llvm/llvm-project/pull/144758.diff

1 Files Affected:

• (modified) llvm/lib/Transforms/Vectorize/VPlan.cpp (+2)

diff --git a/llvm/lib/Transforms/Vectorize/VPlan.cpp b/llvm/lib/Transforms/Vectorize/VPlan.cpp
index 4332332ef5cc3..773a5a4a829c7 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlan.cpp
@@ -1015,6 +1015,8 @@ void VPlan::execute(VPTransformState *State) {
   for (VPBlockBase *Block : RPOT)
     Block->execute(State);
 
+  State->CFG.DTU.flush();
+
   VPBasicBlock *Header = vputils::getFirstLoopHeader(*this, State->VPDT);
   if (!Header)
     return;

[fhahn]

That's surprising, LGTM thanks

[fhahn]

@aeubanks @nico any chance you already managed to get a reproducer?

[aeubanks]

@fhahn I've been trying to come up with a reproducer, but it's in a ThinLTO build. am continuing to try this week. it might be an earlier pass not properly invalidating DT?

[aeubanks]

@fhahn
opt -p loop-vectorize on the following produces different output
b.ll.txt

hopefully this is enough to debug given the PR said it was NFC