[ARM64] [Windows] Mark block address as taken when expanding catchrets

[momo5502]

This fixes issue #109250

The issue happens during the MachineBlockPlacement pass. The block, whose address was previously not taken, is deemed redundant by the pass and subsequently replaced using
MachineBasicBlock::ReplaceUsesOfBlockWith in BranchFolding.

ReplaceUsesOfBlockWith only replaces uses in the terminator. However, expandPostRAPseudo introduces new block uses when expanding catchrets. These uses do not get replaced, which results in undefined label errors later on.

Marking the block addresss as taken prevents the replacement of the block, without also replacing non-terminator uses.

[llvmbot]

@llvm/pr-subscribers-backend-aarch64

Author: Maurice Heumann (momo5502)

Changes

This fixes issue #109250

The issue happens during the MachineBlockPlacement pass. The block, whose address was previously not taken, is deemed redundant by the pass and subsequently replaced using
MachineBasicBlock::ReplaceUsesOfBlockWith in BranchFolding.

ReplaceUsesOfBlockWith only replaces uses in the terminator. However, expandPostRAPseudo introduces new block uses when expanding catchrets. These uses do not get replaced, which results in undefined label errors later on.

Marking the block addresss as taken prevents the replacement of the block, without also replacing non-terminator uses.

---

Full diff: https://github.com/llvm/llvm-project/pull/109252.diff

1 Files Affected:

• (modified) llvm/lib/Target/AArch64/AArch64InstrInfo.cpp (+1)

diff --git a/llvm/lib/Target/AArch64/AArch64InstrInfo.cpp b/llvm/lib/Target/AArch64/AArch64InstrInfo.cpp
index 3b38a5f78dee51..32bc0e7d0d6475 100644
--- a/llvm/lib/Target/AArch64/AArch64InstrInfo.cpp
+++ b/llvm/lib/Target/AArch64/AArch64InstrInfo.cpp
@@ -1994,6 +1994,7 @@ bool AArch64InstrInfo::expandPostRAPseudo(MachineInstr &MI) const {
         .addReg(AArch64::X0)
         .addMBB(TargetMBB)
         .addImm(0);
+    TargetMBB->setMachineBlockAddressTaken();
     return true;
   }
 

[efriedma-quic]

Code change makes sense; please add a testcase.

[momo5502]

Code change makes sense; please add a testcase.

I added the sample from the issue as a test case to assert the issue does not occur anymore. I hope this is fine.
Asserting that the block address is taken should by covered by adjusting the two existing tests.

[momo5502]

Is the test fine like this?

[efriedma-quic]

LGTM

[momo5502]

LGTM

Thanks, would you mind merging it for me, please?