Fix the remaining atomic intrinsics to use the right register classes on Thumb2,

and add some basic tests for them.

llvm-svn: 132235

2 files changed, 126 insertions, 10 deletions

diff --git a/llvm/lib/Target/ARM/ARMISelLowering.cpp b/llvm/lib/Target/ARM/ARMISelLowering.cpp
index 682af2f6b81..f23d890d514 100644
--- a/llvm/lib/Target/ARM/ARMISelLowering.cpp
+++ b/llvm/lib/Target/ARM/ARMISelLowering.cpp
@@ -4983,8 +4983,14 @@ ARMTargetLowering::EmitAtomicBinary(MachineInstr *MI, MachineBasicBlock *BB,
   unsigned ptr = MI->getOperand(1).getReg();
   unsigned incr = MI->getOperand(2).getReg();
   DebugLoc dl = MI->getDebugLoc();
-
   bool isThumb2 = Subtarget->isThumb2();
+
+  MachineRegisterInfo &MRI = BB->getParent()->getRegInfo();
+  if (isThumb2) {
+    MRI.constrainRegClass(dest, ARM::rGPRRegisterClass);
+    MRI.constrainRegClass(ptr, ARM::rGPRRegisterClass);
+  }
+
   unsigned ldrOpc, strOpc;
   switch (Size) {
   default: llvm_unreachable("unsupported size for AtomicCmpSwap!");
@@ -5013,10 +5019,10 @@ ARMTargetLowering::EmitAtomicBinary(MachineInstr *MI, MachineBasicBlock *BB,
                   BB->end());
   exitMBB->transferSuccessorsAndUpdatePHIs(BB);
 
-  MachineRegisterInfo &RegInfo = MF->getRegInfo();
-  unsigned scratch = RegInfo.createVirtualRegister(ARM::GPRRegisterClass);
-  unsigned scratch2 = (!BinOpcode) ? incr :
-    RegInfo.createVirtualRegister(ARM::GPRRegisterClass);
+  TargetRegisterClass *TRC =
+    isThumb2 ? ARM::tGPRRegisterClass : ARM::GPRRegisterClass;
+  unsigned scratch = MRI.createVirtualRegister(TRC);
+  unsigned scratch2 = (!BinOpcode) ? incr : MRI.createVirtualRegister(TRC);
 
   //  thisMBB:
   //   ...
@@ -5079,8 +5085,14 @@ ARMTargetLowering::EmitAtomicBinaryMinMax(MachineInstr *MI,
   unsigned incr = MI->getOperand(2).getReg();
   unsigned oldval = dest;
   DebugLoc dl = MI->getDebugLoc();
-
   bool isThumb2 = Subtarget->isThumb2();
+
+  MachineRegisterInfo &MRI = BB->getParent()->getRegInfo();
+  if (isThumb2) {
+    MRI.constrainRegClass(dest, ARM::rGPRRegisterClass);
+    MRI.constrainRegClass(ptr, ARM::rGPRRegisterClass);
+  }
+
   unsigned ldrOpc, strOpc, extendOpc;
   switch (Size) {
   default: llvm_unreachable("unsupported size for AtomicCmpSwap!");
@@ -5112,9 +5124,10 @@ ARMTargetLowering::EmitAtomicBinaryMinMax(MachineInstr *MI,
                   BB->end());
   exitMBB->transferSuccessorsAndUpdatePHIs(BB);
 
-  MachineRegisterInfo &RegInfo = MF->getRegInfo();
-  unsigned scratch = RegInfo.createVirtualRegister(ARM::GPRRegisterClass);
-  unsigned scratch2 = RegInfo.createVirtualRegister(ARM::GPRRegisterClass);
+  TargetRegisterClass *TRC =
+    isThumb2 ? ARM::tGPRRegisterClass : ARM::GPRRegisterClass;
+  unsigned scratch = MRI.createVirtualRegister(TRC);
+  unsigned scratch2 = MRI.createVirtualRegister(TRC);
 
   //  thisMBB:
   //   ...
@@ -5135,7 +5148,7 @@ ARMTargetLowering::EmitAtomicBinaryMinMax(MachineInstr *MI,
 
   // Sign extend the value, if necessary.
   if (signExtend && extendOpc) {
-    oldval = RegInfo.createVirtualRegister(ARM::GPRRegisterClass);
+    oldval = MRI.createVirtualRegister(ARM::GPRRegisterClass);
     AddDefaultPred(BuildMI(BB, dl, TII->get(extendOpc), oldval).addReg(dest));
   }
 
diff --git a/llvm/test/CodeGen/ARM/atomic-op.ll b/llvm/test/CodeGen/ARM/atomic-op.ll
new file mode 100644
index 00000000000..03940e38ce8
--- /dev/null
+++ b/llvm/test/CodeGen/ARM/atomic-op.ll
@@ -0,0 +1,103 @@
+; RUN: llc < %s -mtriple=armv7-apple-darwin10 | FileCheck %s
+; RUN: llc < %s -mtriple=thumbv7-apple-darwin10 | FileCheck %s
+
+define void @func(i32 %argc, i8** %argv) nounwind {
+entry:
+	%argc.addr = alloca i32		; <i32*> [#uses=1]
+	%argv.addr = alloca i8**		; <i8***> [#uses=1]
+	%val1 = alloca i32		; <i32*> [#uses=2]
+	%val2 = alloca i32		; <i32*> [#uses=15]
+	%andt = alloca i32		; <i32*> [#uses=2]
+	%ort = alloca i32		; <i32*> [#uses=2]
+	%xort = alloca i32		; <i32*> [#uses=2]
+	%old = alloca i32		; <i32*> [#uses=18]
+	%temp = alloca i32		; <i32*> [#uses=2]
+	store i32 %argc, i32* %argc.addr
+	store i8** %argv, i8*** %argv.addr
+	store i32 0, i32* %val1
+	store i32 31, i32* %val2
+	store i32 3855, i32* %andt
+	store i32 3855, i32* %ort
+	store i32 3855, i32* %xort
+	store i32 4, i32* %temp
+	%tmp = load i32* %temp
+  ; CHECK: ldrex
+  ; CHECK: add
+  ; CHECK: strex
+	call i32 @llvm.atomic.load.add.i32.p0i32( i32* %val1, i32 %tmp )		; <i32>:0 [#uses=1]
+	store i32 %0, i32* %old
+  ; CHECK: ldrex
+  ; CHECK: sub
+  ; CHECK: strex
+	call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %val2, i32 30 )		; <i32>:1 [#uses=1]
+	store i32 %1, i32* %old
+  ; CHECK: ldrex
+  ; CHECK: add
+  ; CHECK: strex
+	call i32 @llvm.atomic.load.add.i32.p0i32( i32* %val2, i32 1 )		; <i32>:2 [#uses=1]
+	store i32 %2, i32* %old
+  ; CHECK: ldrex
+  ; CHECK: sub
+  ; CHECK: strex
+	call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %val2, i32 1 )		; <i32>:3 [#uses=1]
+	store i32 %3, i32* %old
+  ; CHECK: ldrex
+  ; CHECK: and
+  ; CHECK: strex
+	call i32 @llvm.atomic.load.and.i32.p0i32( i32* %andt, i32 4080 )		; <i32>:4 [#uses=1]
+	store i32 %4, i32* %old
+  ; CHECK: ldrex
+  ; CHECK: or
+  ; CHECK: strex
+	call i32 @llvm.atomic.load.or.i32.p0i32( i32* %ort, i32 4080 )		; <i32>:5 [#uses=1]
+	store i32 %5, i32* %old
+  ; CHECK: ldrex
+  ; CHECK: eor
+  ; CHECK: strex
+	call i32 @llvm.atomic.load.xor.i32.p0i32( i32* %xort, i32 4080 )		; <i32>:6 [#uses=1]
+	store i32 %6, i32* %old
+  ; CHECK: ldrex
+  ; CHECK: cmp
+  ; CHECK: strex
+	call i32 @llvm.atomic.load.min.i32.p0i32( i32* %val2, i32 16 )		; <i32>:7 [#uses=1]
+	store i32 %7, i32* %old
+	%neg = sub i32 0, 1		; <i32> [#uses=1]
+  ; CHECK: ldrex
+  ; CHECK: cmp
+  ; CHECK: strex
+	call i32 @llvm.atomic.load.min.i32.p0i32( i32* %val2, i32 %neg )		; <i32>:8 [#uses=1]
+	store i32 %8, i32* %old
+  ; CHECK: ldrex
+  ; CHECK: cmp
+  ; CHECK: strex
+	call i32 @llvm.atomic.load.max.i32.p0i32( i32* %val2, i32 1 )		; <i32>:9 [#uses=1]
+	store i32 %9, i32* %old
+  ; CHECK: ldrex
+  ; CHECK: cmp
+  ; CHECK: strex
+	call i32 @llvm.atomic.load.max.i32.p0i32( i32* %val2, i32 0 )		; <i32>:10 [#uses=1]
+	store i32 %10, i32* %old
+	ret void
+}
+
+declare i32 @llvm.atomic.load.add.i32.p0i32(i32*, i32) nounwind 
+
+declare i32 @llvm.atomic.load.sub.i32.p0i32(i32*, i32) nounwind 
+
+declare i32 @llvm.atomic.load.and.i32.p0i32(i32*, i32) nounwind 
+
+declare i32 @llvm.atomic.load.or.i32.p0i32(i32*, i32) nounwind 
+
+declare i32 @llvm.atomic.load.xor.i32.p0i32(i32*, i32) nounwind 
+
+declare i32 @llvm.atomic.load.min.i32.p0i32(i32*, i32) nounwind 
+
+declare i32 @llvm.atomic.load.max.i32.p0i32(i32*, i32) nounwind 
+
+declare i32 @llvm.atomic.load.umax.i32.p0i32(i32*, i32) nounwind 
+
+declare i32 @llvm.atomic.load.umin.i32.p0i32(i32*, i32) nounwind 
+
+declare i32 @llvm.atomic.swap.i32.p0i32(i32*, i32) nounwind 
+
+declare i32 @llvm.atomic.cmp.swap.i32.p0i32(i32*, i32, i32) nounwind