TwoAddressInstruction: Fix assert on undef operand with -early-live-intervals

[arsenm]

No description provided.

[arsenm]

• TwoAddressInstruction: Fix assert on undef operand with -early-live-intervals #125518 👈 (View in Graphite)
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[llvmbot]

@llvm/pr-subscribers-backend-arm

@llvm/pr-subscribers-llvm-regalloc

Author: Matt Arsenault (arsenm)

Changes

---

Full diff: https://github.com/llvm/llvm-project/pull/125518.diff

2 Files Affected:

• (modified) llvm/lib/CodeGen/TwoAddressInstructionPass.cpp (+3)
• (added) llvm/test/CodeGen/ARM/twoaddress-asserts-liveints-undef-use.mir (+24)

diff --git a/llvm/lib/CodeGen/TwoAddressInstructionPass.cpp b/llvm/lib/CodeGen/TwoAddressInstructionPass.cpp
index fb6274b09919ba..6236268f77ab16 100644
--- a/llvm/lib/CodeGen/TwoAddressInstructionPass.cpp
+++ b/llvm/lib/CodeGen/TwoAddressInstructionPass.cpp
@@ -489,6 +489,9 @@ MachineInstr *TwoAddressInstructionImpl::findOnlyInterestingUse(
     bool &IsDstPhys) const {
   MachineOperand *UseOp = nullptr;
   for (MachineOperand &MO : MRI->use_nodbg_operands(Reg)) {
+    if (MO.isUndef())
+      continue;
+
     MachineInstr *MI = MO.getParent();
     if (MI->getParent() != MBB)
       return nullptr;
diff --git a/llvm/test/CodeGen/ARM/twoaddress-asserts-liveints-undef-use.mir b/llvm/test/CodeGen/ARM/twoaddress-asserts-liveints-undef-use.mir
new file mode 100644
index 00000000000000..6e33d4be9345da
--- /dev/null
+++ b/llvm/test/CodeGen/ARM/twoaddress-asserts-liveints-undef-use.mir
@@ -0,0 +1,24 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5
+# RUN: llc -mtriple=thumbv8.1m.main-none-none-eabi -mattr=+mve -verify-machineinstrs -run-pass=liveintervals,twoaddressinstruction -o - %s | FileCheck %s
+
+---
+name:      undef_use_asserts_lr
+tracksRegLiveness: true
+noPhis:          true
+body:             |
+  bb.0:
+    liveins: $q0
+
+    ; CHECK-LABEL: name: undef_use_asserts_lr
+    ; CHECK: liveins: $q0
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: [[COPY:%[0-9]+]]:mqpr = COPY $q0
+    ; CHECK-NEXT: [[DEF:%[0-9]+]]:spr = IMPLICIT_DEF
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:mqpr = COPY [[COPY]]
+    ; CHECK-NEXT: dead [[COPY1:%[0-9]+]].ssub_2:mqpr = COPY [[DEF]]
+    ; CHECK-NEXT: dead [[COPY2:%[0-9]+]]:spr = COPY undef [[COPY1]].ssub_3
+    %0:mqpr = COPY $q0
+    %1:spr = IMPLICIT_DEF
+    %2:mqpr = INSERT_SUBREG %0, %1, %subreg.ssub_2
+    dead %4:spr = COPY undef %2.ssub_3
+...

[sdesmalen-arm]

Unrelated to your patch, but should there be a break after line 499?

[arsenm]

It probably should, but I'd need to look exactly at all the implementations of isPlainlyKilled. At least the LiveIntervals one checks there's only one value in the interval