[AArch64] Clear kill flags when removing FMOVDr.

[davemgreen]

The uses of OldDef/NewDef may not be killed in the same place they previously were after they are replaced, and so need to be cleared.

[llvmbot]

@llvm/pr-subscribers-backend-aarch64

Author: David Green (davemgreen)

Changes

The uses of OldDef/NewDef may not be killed in the same place they previously were after they are replaced, and so need to be cleared.

---

Full diff: https://github.com/llvm/llvm-project/pull/86308.diff

2 Files Affected:

• (modified) llvm/lib/Target/AArch64/AArch64MIPeepholeOpt.cpp (+3-1)
• (added) llvm/test/CodeGen/AArch64/peephole-movd.mir (+60)

diff --git a/llvm/lib/Target/AArch64/AArch64MIPeepholeOpt.cpp b/llvm/lib/Target/AArch64/AArch64MIPeepholeOpt.cpp
index 6865850cf04feb..22da7ddef98a2a 100644
--- a/llvm/lib/Target/AArch64/AArch64MIPeepholeOpt.cpp
+++ b/llvm/lib/Target/AArch64/AArch64MIPeepholeOpt.cpp
@@ -680,9 +680,11 @@ bool AArch64MIPeepholeOpt::visitFMOVDr(MachineInstr &MI) {
   // Let's remove MIs for high 64-bits.
   Register OldDef = MI.getOperand(0).getReg();
   Register NewDef = MI.getOperand(1).getReg();
+  LLVM_DEBUG(dbgs() << "Removing: " << MI << "\n");
+  MRI->clearKillFlags(OldDef);
+  MRI->clearKillFlags(NewDef);
   MRI->constrainRegClass(NewDef, MRI->getRegClass(OldDef));
   MRI->replaceRegWith(OldDef, NewDef);
-  LLVM_DEBUG(dbgs() << "Removed: " << MI << "\n");
   MI.eraseFromParent();
 
   return true;
diff --git a/llvm/test/CodeGen/AArch64/peephole-movd.mir b/llvm/test/CodeGen/AArch64/peephole-movd.mir
new file mode 100644
index 00000000000000..bd7f0ab3f044e7
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/peephole-movd.mir
@@ -0,0 +1,60 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -run-pass=aarch64-mi-peephole-opt -o - -mtriple=aarch64-unknown-linux -verify-machineinstrs %s | FileCheck %s
+
+---
+name: remove_kill_flags
+tracksRegLiveness: true
+body: |
+  bb.0.entry:
+    liveins: $w0
+    ; CHECK-LABEL: name: remove_kill_flags
+    ; CHECK: liveins: $w0
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: [[MOVIv2d_ns:%[0-9]+]]:fpr128 = MOVIv2d_ns 0
+    ; CHECK-NEXT: [[COPY:%[0-9]+]]:fpr64 = COPY [[MOVIv2d_ns]].dsub
+    ; CHECK-NEXT: [[UQSHLv8i8_shift:%[0-9]+]]:fpr64 = UQSHLv8i8_shift killed [[COPY]], 1
+    ; CHECK-NEXT: [[SUBREG_TO_REG:%[0-9]+]]:fpr128 = SUBREG_TO_REG 0, [[UQSHLv8i8_shift]], %subreg.dsub
+    ; CHECK-NEXT: [[TBLv8i8One:%[0-9]+]]:fpr64 = TBLv8i8One killed [[SUBREG_TO_REG]], [[UQSHLv8i8_shift]]
+    ; CHECK-NEXT: [[DEF:%[0-9]+]]:fpr128 = IMPLICIT_DEF
+    ; CHECK-NEXT: [[INSERT_SUBREG:%[0-9]+]]:fpr128 = INSERT_SUBREG [[DEF]], [[UQSHLv8i8_shift]], %subreg.dsub
+    ; CHECK-NEXT: RET_ReallyLR implicit $w0
+    %0:fpr128 = MOVIv2d_ns 0
+    %1:fpr64 = COPY %0.dsub:fpr128
+    %2:fpr64 = UQSHLv8i8_shift killed %1:fpr64, 1
+    %3:fpr64 = FMOVDr %2:fpr64
+    %4:fpr128 = SUBREG_TO_REG 0, killed %3:fpr64, %subreg.dsub
+    %5:fpr64 = TBLv8i8One killed %4:fpr128, %2:fpr64
+    %7:fpr128 = IMPLICIT_DEF
+    %6:fpr128 = INSERT_SUBREG %7:fpr128, killed %2:fpr64, %subreg.dsub
+    RET_ReallyLR implicit $w0
+...
+---
+name: remove_kill_flags2
+tracksRegLiveness: true
+body: |
+  bb.0.entry:
+    liveins: $w0
+    ; CHECK-LABEL: name: remove_kill_flags2
+    ; CHECK: liveins: $w0
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: [[MOVIv2d_ns:%[0-9]+]]:fpr128 = MOVIv2d_ns 0
+    ; CHECK-NEXT: [[COPY:%[0-9]+]]:fpr64 = COPY [[MOVIv2d_ns]].dsub
+    ; CHECK-NEXT: [[UQSHLv8i8_shift:%[0-9]+]]:fpr64 = UQSHLv8i8_shift killed [[COPY]], 1
+    ; CHECK-NEXT: [[SUBREG_TO_REG:%[0-9]+]]:fpr128 = SUBREG_TO_REG 0, [[UQSHLv8i8_shift]], %subreg.dsub
+    ; CHECK-NEXT: [[DEF:%[0-9]+]]:fpr128 = IMPLICIT_DEF
+    ; CHECK-NEXT: [[INSERT_SUBREG:%[0-9]+]]:fpr128 = INSERT_SUBREG [[DEF]], [[UQSHLv8i8_shift]], %subreg.dsub
+    ; CHECK-NEXT: [[DEF1:%[0-9]+]]:fpr128 = IMPLICIT_DEF
+    ; CHECK-NEXT: [[INSERT_SUBREG1:%[0-9]+]]:fpr128 = INSERT_SUBREG [[DEF1]], [[UQSHLv8i8_shift]], %subreg.dsub
+    ; CHECK-NEXT: RET_ReallyLR implicit $w0
+    %0:fpr128 = MOVIv2d_ns 0
+    %1:fpr64 = COPY %0.dsub:fpr128
+    %2:fpr64 = UQSHLv8i8_shift killed %1:fpr64, 1
+    %3:fpr64 = FMOVDr %2:fpr64
+    %4:fpr128 = SUBREG_TO_REG 0, %3:fpr64, %subreg.dsub
+    %7:fpr128 = IMPLICIT_DEF
+    %6:fpr128 = INSERT_SUBREG %7:fpr128, killed %2:fpr64, %subreg.dsub
+    %9:fpr128 = IMPLICIT_DEF
+    %8:fpr128 = INSERT_SUBREG %9:fpr128, killed %3:fpr64, %subreg.dsub
+    RET_ReallyLR implicit $w0
+...
+