[RemoveDIs] Replicate dbg intrinsic movement pattern in SelectOptimize (#81737)

Fix crash mentioned in comments on
d759618df76361a8e490eeae5c5399e0738cbfd0.

The assertion being hit was complaining that we had dangling DPValues;
the DPValues attached to the terminator of StartBlock become dangling
after the terminator is erased, and they're never "flushed" back onto
the new terminator once it's added. Doing that makes the crash go away,
but doesn't replicate existing dbg.* behaviour. See the comment in the
patch.

This change both fixes the crash (because there are now no DPValues left
on the terminator to dangle) and replicates existing behaviour (moves
those DPValues down to the new block).

1 files changed, 6 insertions, 0 deletions

diff --git a/llvm/lib/CodeGen/SelectOptimize.cpp b/llvm/lib/CodeGen/SelectOptimize.cpp
index 31c4b63..5609f48 100644
--- a/llvm/lib/CodeGen/SelectOptimize.cpp
+++ b/llvm/lib/CodeGen/SelectOptimize.cpp
@@ -621,6 +621,12 @@ void SelectOptimizeImpl::convertProfitableSIGroups(SelectGroups &ProfSIGroups) {
     SelectLike LastSI = ASI.back();
     BasicBlock *StartBlock = SI.getI()->getParent();
     BasicBlock::iterator SplitPt = ++(BasicBlock::iterator(LastSI.getI()));
+    // With RemoveDIs turned off, SplitPt can be a dbg.* intrinsic. With
+    // RemoveDIs turned on, SplitPt would instead point to the next
+    // instruction. To match existing dbg.* intrinsic behaviour with RemoveDIs,
+    // tell splitBasicBlock that we want to include any DPValues attached to
+    // SplitPt in the splice.
+    SplitPt.setHeadBit(true);
     BasicBlock *EndBlock = StartBlock->splitBasicBlock(SplitPt, "select.end");
     BFI->setBlockFreq(EndBlock, BFI->getBlockFreq(StartBlock));
     // Delete the unconditional branch that was just created by the split.