AArch64: avoid clobbering SP for dead MOVimm pseudos.

We were producing ORR, which actually defines a GPR32sp rather than a GPR32.

Should fix PR23209.

llvm-svn: 265198

diff --git a/llvm/lib/Target/AArch64/AArch64.h b/llvm/lib/Target/AArch64/AArch64.h
index d4c9e6b..c767c75 100644 (file)
--- a/llvm/lib/Target/AArch64/AArch64.h
+++ b/llvm/lib/Target/AArch64/AArch64.h
@@ -45,6 +45,8 @@ FunctionPass *createAArch64A53Fix835769();
 FunctionPass *createAArch64CleanupLocalDynamicTLSPass();
 
 FunctionPass *createAArch64CollectLOHPass();
+
+void initializeAArch64ExpandPseudoPass(PassRegistry&);
 } // end namespace llvm
 
 #endif

diff --git a/llvm/lib/Target/AArch64/AArch64ExpandPseudoInsts.cpp b/llvm/lib/Target/AArch64/AArch64ExpandPseudoInsts.cpp
index d24e42a..610fcc4 100644 (file)
--- a/llvm/lib/Target/AArch64/AArch64ExpandPseudoInsts.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ExpandPseudoInsts.cpp
@@ -403,9 +403,17 @@ bool AArch64ExpandPseudo::expandMOVImm(MachineBasicBlock &MBB,
                                        MachineBasicBlock::iterator MBBI,
                                        unsigned BitSize) {
   MachineInstr &MI = *MBBI;
+  unsigned DstReg = MI.getOperand(0).getReg();
   uint64_t Imm = MI.getOperand(1).getImm();
   const unsigned Mask = 0xFFFF;
 
+  if (DstReg == AArch64::XZR || DstReg == AArch64::WZR) {
+    // Useless def, and we don't want to risk creating an invalid ORR (which
+    // would really write to sp).
+    MI.eraseFromParent();
+    return true;
+  }
+
   // Try a MOVI instruction (aka ORR-immediate with the zero register).
   uint64_t UImm = Imm << (64 - BitSize) >> (64 - BitSize);
   uint64_t Encoding;
@@ -531,7 +539,6 @@ bool AArch64ExpandPseudo::expandMOVImm(MachineBasicBlock &MBB,
     LastShift = (TZ / 16) * 16;
   }
   unsigned Imm16 = (Imm >> Shift) & Mask;
-  unsigned DstReg = MI.getOperand(0).getReg();
   bool DstIsDead = MI.getOperand(0).isDead();
   MachineInstrBuilder MIB1 =
       BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(FirstOpc))

diff --git a/llvm/lib/Target/AArch64/AArch64TargetMachine.cpp b/llvm/lib/Target/AArch64/AArch64TargetMachine.cpp
index 30d8342..a2638aa 100644 (file)
--- a/llvm/lib/Target/AArch64/AArch64TargetMachine.cpp
+++ b/llvm/lib/Target/AArch64/AArch64TargetMachine.cpp
@@ -111,7 +111,9 @@ extern "C" void LLVMInitializeAArch64Target() {
   RegisterTargetMachine<AArch64leTargetMachine> X(TheAArch64leTarget);
   RegisterTargetMachine<AArch64beTargetMachine> Y(TheAArch64beTarget);
   RegisterTargetMachine<AArch64leTargetMachine> Z(TheARM64Target);
-  initializeGlobalISel(*PassRegistry::getPassRegistry());
+  auto PR = PassRegistry::getPassRegistry();
+  initializeGlobalISel(*PR);
+  initializeAArch64ExpandPseudoPass(*PR);
 }
 
 //===----------------------------------------------------------------------===//

diff --git a/llvm/test/CodeGen/AArch64/movimm-wzr.mir b/llvm/test/CodeGen/AArch64/movimm-wzr.mir
new file mode 100644 (file)
index 0000000..32d4676
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/movimm-wzr.mir
@@ -0,0 +1,46 @@
+# RUN: llc -run-pass=aarch64-expand-pseudo %s | FileCheck %s
+
+--- |
+  ; ModuleID = 'simple.ll'
+  source_filename = "simple.ll"
+  target datalayout = "e-m:e-i64:64-i128:128-n32:64-S128"
+  target triple = "aarch64--linux-gnu"
+
+  define i32 @test_mov_0() {
+    ret i32 42
+  }
+
+...
+---
+name:            test_mov_0
+alignment:       2
+exposesReturnsTwice: false
+hasInlineAsm:    false
+allVRegsAllocated: true
+isSSA:           false
+tracksRegLiveness: false
+tracksSubRegLiveness: false
+frameInfo:
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    %wzr = MOVi32imm 42
+    %xzr = MOVi64imm 42
+    RET_ReallyLR implicit killed %w0
+
+...
+
+# CHECK: bb.0
+# CHECK-NEXT: RET %lr