Re-apply r110655 with fixes. Epilogue must restore sp from fp if the function stack frame has a var-sized object.

Also added a test case to check for the added benefit of this patch: it's optimizing away the unnecessary restore of sp from fp for some non-leaf functions.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@110707 91177308-0d34-0410-b5e6-96231b3b80d8

7 files changed, 148 insertions, 70 deletions

diff --git a/lib/Target/ARM/ARMBaseRegisterInfo.cpp b/lib/Target/ARM/ARMBaseRegisterInfo.cpp
index ee1a82e..12308c4 100644
--- a/lib/Target/ARM/ARMBaseRegisterInfo.cpp
+++ b/lib/Target/ARM/ARMBaseRegisterInfo.cpp
@@ -177,7 +177,7 @@ getReservedRegs(const MachineFunction &MF) const {
   Reserved.set(ARM::SP);
   Reserved.set(ARM::PC);
   Reserved.set(ARM::FPSCR);
-  if (STI.isTargetDarwin() || hasFP(MF))
+  if (hasFP(MF))
     Reserved.set(FramePtr);
   // Some targets reserve R9.
   if (STI.isR9Reserved())
@@ -194,7 +194,7 @@ bool ARMBaseRegisterInfo::isReservedReg(const MachineFunction &MF,
     return true;
   case ARM::R7:
   case ARM::R11:
-    if (FramePtr == Reg && (STI.isTargetDarwin() || hasFP(MF)))
+    if (FramePtr == Reg && hasFP(MF))
       return true;
     break;
   case ARM::R9:
@@ -511,7 +511,7 @@ ARMBaseRegisterInfo::getAllocationOrder(const TargetRegisterClass *RC,
       return std::make_pair(RC->allocation_order_begin(MF),
                             RC->allocation_order_end(MF));
 
-    if (!STI.isTargetDarwin() && !hasFP(MF)) {
+    if (!hasFP(MF)) {
       if (!STI.isR9Reserved())
         return std::make_pair(GPREven1,
                               GPREven1 + (sizeof(GPREven1)/sizeof(unsigned)));
@@ -540,7 +540,7 @@ ARMBaseRegisterInfo::getAllocationOrder(const TargetRegisterClass *RC,
       return std::make_pair(RC->allocation_order_begin(MF),
                             RC->allocation_order_end(MF));
 
-    if (!STI.isTargetDarwin() && !hasFP(MF)) {
+    if (!hasFP(MF)) {
       if (!STI.isR9Reserved())
         return std::make_pair(GPROdd1,
                               GPROdd1 + (sizeof(GPROdd1)/sizeof(unsigned)));
@@ -610,6 +610,10 @@ ARMBaseRegisterInfo::UpdateRegAllocHint(unsigned Reg, unsigned NewReg,
 /// or if frame pointer elimination is disabled.
 ///
 bool ARMBaseRegisterInfo::hasFP(const MachineFunction &MF) const {
+  // Mac OS X requires FP not to be clobbered for backtracing purpose.
+  if (STI.isTargetDarwin())
+    return true;
+
   const MachineFrameInfo *MFI = MF.getFrameInfo();
   // Always eliminate non-leaf frame pointers.
   return ((DisableFramePointerElim(MF) && MFI->hasCalls()) ||
@@ -683,6 +687,7 @@ static unsigned estimateStackSize(MachineFunction &MF) {
 /// instructions will require a scratch register during their expansion later.
 unsigned
 ARMBaseRegisterInfo::estimateRSStackSizeLimit(MachineFunction &MF) const {
+  const ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
   unsigned Limit = (1 << 12) - 1;
   for (MachineFunction::iterator BB = MF.begin(),E = MF.end(); BB != E; ++BB) {
     for (MachineBasicBlock::iterator I = BB->begin(), E = BB->end();
@@ -708,7 +713,10 @@ ARMBaseRegisterInfo::estimateRSStackSizeLimit(MachineFunction &MF) const {
           Limit = std::min(Limit, ((1U << 8) - 1) * 4);
           break;
         case ARMII::AddrModeT2_i12:
-          if (hasFP(MF)) Limit = std::min(Limit, (1U << 8) - 1);
+          // i12 supports only positive offset so these will be converted to
+          // i8 opcodes. See llvm::rewriteT2FrameIndex.
+          if (hasFP(MF) && AFI->hasStackFrame())
+            Limit = std::min(Limit, (1U << 8) - 1);
           break;
         case ARMII::AddrMode6:
           // Addressing mode 6 (load/store) instructions can't encode an
@@ -860,8 +868,9 @@ ARMBaseRegisterInfo::processFunctionBeforeCalleeSavedScan(MachineFunction &MF,
   //        and which instructions will need a scratch register for them. Is it
   //        worth the effort and added fragility?
   bool BigStack =
-    (RS && (estimateStackSize(MF) + (hasFP(MF) ? 4:0) >=
-            estimateRSStackSizeLimit(MF)))
+    (RS &&
+     (estimateStackSize(MF) + ((hasFP(MF) && AFI->hasStackFrame()) ? 4:0) >=
+      estimateRSStackSizeLimit(MF)))
     || MFI->hasVarSizedObjects()
     || (MFI->adjustsStack() && !canSimplifyCallFramePseudos(MF));
 
@@ -881,9 +890,7 @@ ARMBaseRegisterInfo::processFunctionBeforeCalleeSavedScan(MachineFunction &MF,
       ExtraCSSpill = true;
     }
 
-    // Darwin ABI requires FP to point to the stack slot that contains the
-    // previous FP.
-    if (STI.isTargetDarwin() || hasFP(MF)) {
+    if (hasFP(MF)) {
       MF.getRegInfo().setPhysRegUsed(FramePtr);
       NumGPRSpills++;
     }
@@ -976,7 +983,7 @@ unsigned ARMBaseRegisterInfo::getRARegister() const {
 
 unsigned 
 ARMBaseRegisterInfo::getFrameRegister(const MachineFunction &MF) const {
-  if (STI.isTargetDarwin() || hasFP(MF))
+  if (hasFP(MF))
     return FramePtr;
   return ARM::SP;
 }
@@ -1546,7 +1553,8 @@ emitPrologue(MachineFunction &MF) const {
   // Otherwise, if this is not Darwin, all the callee-saved registers go
   // into spill area 1, including the FP in R11.  In either case, it is
   // now safe to emit this assignment.
-  if (STI.isTargetDarwin() || hasFP(MF)) {
+  bool HasFP = hasFP(MF);
+  if (HasFP) {
     unsigned ADDriOpc = !AFI->isThumbFunction() ? ARM::ADDri : ARM::t2ADDri;
     MachineInstrBuilder MIB =
       BuildMI(MBB, MBBI, dl, TII.get(ADDriOpc), FramePtr)
@@ -1565,7 +1573,7 @@ emitPrologue(MachineFunction &MF) const {
   unsigned DPRCSOffset  = NumBytes - (GPRCS1Size + GPRCS2Size + DPRCSSize);
   unsigned GPRCS2Offset = DPRCSOffset + DPRCSSize;
   unsigned GPRCS1Offset = GPRCS2Offset + GPRCS2Size;
-  if (STI.isTargetDarwin() || hasFP(MF))
+  if (HasFP)
     AFI->setFramePtrSpillOffset(MFI->getObjectOffset(FramePtrSpillFI) +
                                 NumBytes);
   AFI->setGPRCalleeSavedArea1Offset(GPRCS1Offset);
@@ -1577,11 +1585,14 @@ emitPrologue(MachineFunction &MF) const {
   if (NumBytes) {
     // Adjust SP after all the callee-save spills.
     emitSPUpdate(isARM, MBB, MBBI, dl, TII, -NumBytes);
+    if (HasFP)
+      AFI->setShouldRestoreSPFromFP(true);
   }
 
   if (STI.isTargetELF() && hasFP(MF)) {
     MFI->setOffsetAdjustment(MFI->getOffsetAdjustment() -
                              AFI->getFramePtrSpillOffset());
+    AFI->setShouldRestoreSPFromFP(true);
   }
 
   AFI->setGPRCalleeSavedArea1Size(GPRCS1Size);
@@ -1614,7 +1625,14 @@ emitPrologue(MachineFunction &MF) const {
       BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVtgpr2gpr), ARM::SP)
         .addReg(ARM::R4, RegState::Kill);
     }
+
+    AFI->setShouldRestoreSPFromFP(true);
   }
+
+  // If the frame has variable sized objects then the epilogue must restore
+  // the sp from fp.
+  if (!AFI->shouldRestoreSPFromFP() && MFI->hasVarSizedObjects())
+    AFI->setShouldRestoreSPFromFP(true);
 }
 
 static bool isCalleeSavedRegister(unsigned Reg, const unsigned *CSRegs) {
@@ -1669,34 +1687,25 @@ emitEpilogue(MachineFunction &MF, MachineBasicBlock &MBB) const {
                  AFI->getGPRCalleeSavedArea2Size() +
                  AFI->getDPRCalleeSavedAreaSize());
 
-    // Darwin ABI requires FP to point to the stack slot that contains the
-    // previous FP.
-    bool HasFP = hasFP(MF);
-    if ((STI.isTargetDarwin() && NumBytes) || HasFP) {
+    // Reset SP based on frame pointer only if the stack frame extends beyond
+    // frame pointer stack slot or target is ELF and the function has FP.
+    if (AFI->shouldRestoreSPFromFP()) {
       NumBytes = AFI->getFramePtrSpillOffset() - NumBytes;
-      // Reset SP based on frame pointer only if the stack frame extends beyond
-      // frame pointer stack slot or target is ELF and the function has FP.
-      if (HasFP ||
-          AFI->getGPRCalleeSavedArea2Size() ||
-          AFI->getDPRCalleeSavedAreaSize()  ||
-          AFI->getDPRCalleeSavedAreaOffset()) {
-        if (NumBytes) {
-          if (isARM)
-            emitARMRegPlusImmediate(MBB, MBBI, dl, ARM::SP, FramePtr, -NumBytes,
-                                    ARMCC::AL, 0, TII);
-          else
-            emitT2RegPlusImmediate(MBB, MBBI, dl, ARM::SP, FramePtr, -NumBytes,
-                                    ARMCC::AL, 0, TII);
-        } else {
-          // Thumb2 or ARM.
-          if (isARM)
-            BuildMI(MBB, MBBI, dl, TII.get(ARM::MOVr), ARM::SP)
-              .addReg(FramePtr)
-              .addImm((unsigned)ARMCC::AL).addReg(0).addReg(0);
-          else
-            BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVgpr2gpr), ARM::SP)
-              .addReg(FramePtr);
-        }
+      if (NumBytes) {
+        if (isARM)
+          emitARMRegPlusImmediate(MBB, MBBI, dl, ARM::SP, FramePtr, -NumBytes,
+                                  ARMCC::AL, 0, TII);
+        else
+          emitT2RegPlusImmediate(MBB, MBBI, dl, ARM::SP, FramePtr, -NumBytes,
+                                 ARMCC::AL, 0, TII);
+      } else {
+        // Thumb2 or ARM.
+        if (isARM)
+          BuildMI(MBB, MBBI, dl, TII.get(ARM::MOVr), ARM::SP)
+            .addReg(FramePtr).addImm((unsigned)ARMCC::AL).addReg(0).addReg(0);
+        else
+          BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVgpr2gpr), ARM::SP)
+            .addReg(FramePtr);
       }
     } else if (NumBytes)
       emitSPUpdate(isARM, MBB, MBBI, dl, TII, NumBytes);
diff --git a/lib/Target/ARM/ARMISelLowering.cpp b/lib/Target/ARM/ARMISelLowering.cpp
index db550b1..c3d60ad 100644
--- a/lib/Target/ARM/ARMISelLowering.cpp
+++ b/lib/Target/ARM/ARMISelLowering.cpp
@@ -742,14 +742,15 @@ Sched::Preference ARMTargetLowering::getSchedulingPreference(SDNode *N) const {
 unsigned
 ARMTargetLowering::getRegPressureLimit(const TargetRegisterClass *RC,
                                        MachineFunction &MF) const {
-  unsigned FPDiff = RegInfo->hasFP(MF) ? 1 : 0;
   switch (RC->getID()) {
   default:
     return 0;
   case ARM::tGPRRegClassID:
-    return 5 - FPDiff;
-  case ARM::GPRRegClassID:
-    return 10 - FPDiff - (Subtarget->isR9Reserved() ? 1 : 0);
+    return RegInfo->hasFP(MF) ? 4 : 5;
+  case ARM::GPRRegClassID: {
+    unsigned FP = RegInfo->hasFP(MF) ? 1 : 0;
+    return 10 - FP - (Subtarget->isR9Reserved() ? 1 : 0);
+  }
   case ARM::SPRRegClassID:  // Currently not used as 'rep' register class.
   case ARM::DPRRegClassID:
     return 32 - 10;
diff --git a/lib/Target/ARM/ARMMachineFunctionInfo.h b/lib/Target/ARM/ARMMachineFunctionInfo.h
index 7e57a1c..514c26b 100644
--- a/lib/Target/ARM/ARMMachineFunctionInfo.h
+++ b/lib/Target/ARM/ARMMachineFunctionInfo.h
@@ -43,6 +43,10 @@ class ARMFunctionInfo : public MachineFunctionInfo {
   /// processFunctionBeforeCalleeSavedScan().
   bool HasStackFrame;
 
+  /// RestoreSPFromFP - True if epilogue should restore SP from FP. Set by
+  /// emitPrologue.
+  bool RestoreSPFromFP;
+
   /// LRSpilledForFarJump - True if the LR register has been for spilled to
   /// enable far jump.
   bool LRSpilledForFarJump;
@@ -95,7 +99,7 @@ public:
   ARMFunctionInfo() :
     isThumb(false),
     hasThumb2(false),
-    VarArgsRegSaveSize(0), HasStackFrame(false),
+    VarArgsRegSaveSize(0), HasStackFrame(false), RestoreSPFromFP(false),
     LRSpilledForFarJump(false),
     FramePtrSpillOffset(0), GPRCS1Offset(0), GPRCS2Offset(0), DPRCSOffset(0),
     GPRCS1Size(0), GPRCS2Size(0), DPRCSSize(0),
@@ -106,7 +110,7 @@ public:
   explicit ARMFunctionInfo(MachineFunction &MF) :
     isThumb(MF.getTarget().getSubtarget<ARMSubtarget>().isThumb()),
     hasThumb2(MF.getTarget().getSubtarget<ARMSubtarget>().hasThumb2()),
-    VarArgsRegSaveSize(0), HasStackFrame(false),
+    VarArgsRegSaveSize(0), HasStackFrame(false), RestoreSPFromFP(false),
     LRSpilledForFarJump(false),
     FramePtrSpillOffset(0), GPRCS1Offset(0), GPRCS2Offset(0), DPRCSOffset(0),
     GPRCS1Size(0), GPRCS2Size(0), DPRCSSize(0),
@@ -125,6 +129,9 @@ public:
   bool hasStackFrame() const { return HasStackFrame; }
   void setHasStackFrame(bool s) { HasStackFrame = s; }
 
+  bool shouldRestoreSPFromFP() const { return RestoreSPFromFP; }
+  void setShouldRestoreSPFromFP(bool s) { RestoreSPFromFP = s; }
+
   bool isLRSpilledForFarJump() const { return LRSpilledForFarJump; }
   void setLRIsSpilledForFarJump(bool s) { LRSpilledForFarJump = s; }
 
diff --git a/lib/Target/ARM/ARMRegisterInfo.td b/lib/Target/ARM/ARMRegisterInfo.td
index 5e17175..2b72739 100644
--- a/lib/Target/ARM/ARMRegisterInfo.td
+++ b/lib/Target/ARM/ARMRegisterInfo.td
@@ -294,8 +294,7 @@ def GPR : RegisterClass<"ARM", [i32], 32, [R0, R1, R2, R3, R4, R5, R6,
 
       if (Subtarget.isThumb1Only()) {
         I = THUMB_GPR_AO + (sizeof(THUMB_GPR_AO)/sizeof(unsigned));
-        // Mac OS X requires FP not to be clobbered for backtracing purpose.
-        return (Subtarget.isTargetDarwin() || RI->hasFP(MF)) ? I-1 : I;
+        return RI->hasFP(MF) ? I-1 : I;
       }
 
       if (Subtarget.isTargetDarwin()) {
@@ -312,8 +311,7 @@ def GPR : RegisterClass<"ARM", [i32], 32, [R0, R1, R2, R3, R4, R5, R6,
           I = ARM_GPR_AO_1 + (sizeof(ARM_GPR_AO_1)/sizeof(unsigned));
       }
 
-      // Mac OS X requires FP not to be clobbered for backtracing purpose.
-      return (Subtarget.isTargetDarwin() || RI->hasFP(MF)) ? I-1 : I;
+      return RI->hasFP(MF) ? I-1 : I;
     }
   }];
 }
@@ -403,8 +401,7 @@ def rGPR : RegisterClass<"ARM", [i32], 32, [R0, R1, R2, R3, R4, R5, R6,
 
       if (Subtarget.isThumb1Only()) {
         I = THUMB_rGPRAO + (sizeof(THUMB_rGPRAO)/sizeof(unsigned));
-        // Mac OS X requires FP not to be clobbered for backtracing purpose.
-        return (Subtarget.isTargetDarwin() || RI->hasFP(MF)) ? I-1 : I;
+        return RI->hasFP(MF) ? I-1 : I;
       }
 
       if (Subtarget.isTargetDarwin()) {
@@ -421,8 +418,7 @@ def rGPR : RegisterClass<"ARM", [i32], 32, [R0, R1, R2, R3, R4, R5, R6,
           I = ARM_rGPRAO_1 + (sizeof(ARM_rGPRAO_1)/sizeof(unsigned));
       }
 
-      // Mac OS X requires FP not to be clobbered for backtracing purpose.
-      return (Subtarget.isTargetDarwin() || RI->hasFP(MF)) ? I-1 : I;
+      return RI->hasFP(MF) ? I-1 : I;
     }
   }];
 }
@@ -449,11 +445,9 @@ def tGPR : RegisterClass<"ARM", [i32], 32, [R0, R1, R2, R3, R4, R5, R6, R7]> {
     tGPRClass::allocation_order_end(const MachineFunction &MF) const {
       const TargetMachine &TM = MF.getTarget();
       const TargetRegisterInfo *RI = TM.getRegisterInfo();
-      const ARMSubtarget &Subtarget = TM.getSubtarget<ARMSubtarget>();
       tGPRClass::iterator I =
         THUMB_tGPR_AO + (sizeof(THUMB_tGPR_AO)/sizeof(unsigned));
-      // Mac OS X requires FP not to be clobbered for backtracing purpose.
-      return (Subtarget.isTargetDarwin() || RI->hasFP(MF)) ? I-1 : I;
+      return RI->hasFP(MF) ? I-1 : I;
     }
   }];
 }
diff --git a/lib/Target/ARM/Thumb1RegisterInfo.cpp b/lib/Target/ARM/Thumb1RegisterInfo.cpp
index 9ae3145..50328e3 100644
--- a/lib/Target/ARM/Thumb1RegisterInfo.cpp
+++ b/lib/Target/ARM/Thumb1RegisterInfo.cpp
@@ -742,11 +742,11 @@ void Thumb1RegisterInfo::emitPrologue(MachineFunction &MF) const {
       dl = MBBI->getDebugLoc();
   }
 
-  // Darwin ABI requires FP to point to the stack slot that contains the
-  // previous FP.
-  if (STI.isTargetDarwin() || hasFP(MF)) {
+  // Adjust FP so it point to the stack slot that contains the previous FP.
+  if (hasFP(MF)) {
     BuildMI(MBB, MBBI, dl, TII.get(ARM::tADDrSPi), FramePtr)
       .addFrameIndex(FramePtrSpillFI).addImm(0);
+    AFI->setShouldRestoreSPFromFP(true);
   }
 
   // Determine starting offsets of spill areas.
@@ -764,10 +764,9 @@ void Thumb1RegisterInfo::emitPrologue(MachineFunction &MF) const {
     emitSPUpdate(MBB, MBBI, TII, dl, *this, -NumBytes);
   }
 
-  if (STI.isTargetELF() && hasFP(MF)) {
+  if (STI.isTargetELF() && hasFP(MF))
     MFI->setOffsetAdjustment(MFI->getOffsetAdjustment() -
                              AFI->getFramePtrSpillOffset());
-  }
 
   AFI->setGPRCalleeSavedArea1Size(GPRCS1Size);
   AFI->setGPRCalleeSavedArea2Size(GPRCS2Size);
@@ -828,7 +827,7 @@ void Thumb1RegisterInfo::emitEpilogue(MachineFunction &MF,
                  AFI->getGPRCalleeSavedArea2Size() +
                  AFI->getDPRCalleeSavedAreaSize());
 
-    if (hasFP(MF)) {
+    if (AFI->shouldRestoreSPFromFP()) {
       NumBytes = AFI->getFramePtrSpillOffset() - NumBytes;
       // Reset SP based on frame pointer only if the stack frame extends beyond
       // frame pointer stack slot or target is ELF and the function has FP.
diff --git a/test/CodeGen/Thumb/large-stack.ll b/test/CodeGen/Thumb/large-stack.ll
index 02de36a..b05e6bf 100644
--- a/test/CodeGen/Thumb/large-stack.ll
+++ b/test/CodeGen/Thumb/large-stack.ll
@@ -1,20 +1,35 @@
-; RUN: llc < %s -march=thumb | grep {ldr.*LCP} | count 5
+; RUN: llc < %s -mtriple=thumb-apple-darwin | FileCheck %s
 
 define void @test1() {
+; CHECK: test1:
+; CHECK: sub sp, #256
+; CHECK: add sp, #256
     %tmp = alloca [ 64 x i32 ] , align 4
     ret void
 }
 
 define void @test2() {
+; CHECK: test2:
+; CHECK: ldr r0, LCPI
+; CHECK: add sp, r0
+; CHECK: mov sp, r7
+; CHECK: sub sp, #4
     %tmp = alloca [ 4168 x i8 ] , align 4
     ret void
 }
 
 define i32 @test3() {
-	%retval = alloca i32, align 4
-	%tmp = alloca i32, align 4
-	%a = alloca [805306369 x i8], align 16
-	store i32 0, i32* %tmp
-	%tmp1 = load i32* %tmp
-        ret i32 %tmp1
+; CHECK: test3:
+; CHECK: ldr r1, LCPI
+; CHECK: add sp, r1
+; CHECK: ldr r1, LCPI
+; CHECK: add r1, sp
+; CHECK: mov sp, r7
+; CHECK: sub sp, #4
+    %retval = alloca i32, align 4
+    %tmp = alloca i32, align 4
+    %a = alloca [805306369 x i8], align 16
+    store i32 0, i32* %tmp
+    %tmp1 = load i32* %tmp
+    ret i32 %tmp1
 }
diff --git a/test/CodeGen/Thumb2/2010-08-10-VarSizedAllocaBug.ll b/test/CodeGen/Thumb2/2010-08-10-VarSizedAllocaBug.ll
new file mode 100644
index 0000000..abcf13a
--- /dev/null
+++ b/test/CodeGen/Thumb2/2010-08-10-VarSizedAllocaBug.ll
@@ -0,0 +1,53 @@
+; RUN: llc < %s -mtriple=thumbv7-apple-darwin -mcpu=cortex-a8 -O3 | FileCheck %s
+
+@.str = private constant [4 x i8] c"%d\0A\00", align 4 ; <[4 x i8]*> [#uses=1]
+
+define internal fastcc i32 @Callee(i32 %i) nounwind {
+entry:
+; CHECK: Callee:
+  %0 = icmp eq i32 %i, 0                          ; <i1> [#uses=1]
+  br i1 %0, label %bb2, label %bb
+
+bb:                                               ; preds = %entry
+  %1 = alloca [1000 x i8], align 4                ; <[1000 x i8]*> [#uses=1]
+  %.sub = getelementptr inbounds [1000 x i8]* %1, i32 0, i32 0 ; <i8*> [#uses=2]
+  %2 = call i32 (i8*, i32, i32, i8*, ...)* @__sprintf_chk(i8* %.sub, i32 0, i32 1000, i8* getelementptr inbounds ([4 x i8]* @.str, i32 0, i32 0), i32 %i) nounwind ; <i32> [#uses=0]
+  %3 = load i8* %.sub, align 4                    ; <i8> [#uses=1]
+  %4 = sext i8 %3 to i32                          ; <i32> [#uses=1]
+  ret i32 %4
+
+bb2:                                              ; preds = %entry
+; Must restore sp from fp here
+; CHECK: mov sp, r7
+; CHECK: sub sp, #8
+; CHECK: pop
+  ret i32 0
+}
+
+declare i32 @__sprintf_chk(i8*, i32, i32, i8*, ...) nounwind
+
+define i32 @main() nounwind {
+; CHECK: main:
+bb.nph:
+  br label %bb
+
+bb:                                               ; preds = %bb, %bb.nph
+  %0 = phi i32 [ 0, %bb.nph ], [ %3, %bb ]        ; <i32> [#uses=2]
+  %j.01 = phi i32 [ 0, %bb.nph ], [ %2, %bb ]     ; <i32> [#uses=1]
+  %1 = tail call fastcc i32 @Callee(i32 %0) nounwind ; <i32> [#uses=1]
+  %2 = add nsw i32 %1, %j.01                      ; <i32> [#uses=2]
+  %3 = add nsw i32 %0, 1                          ; <i32> [#uses=2]
+  %exitcond = icmp eq i32 %3, 10000               ; <i1> [#uses=1]
+  br i1 %exitcond, label %bb2, label %bb
+
+bb2:                                              ; preds = %bb
+; No need to restore sp from fp here.
+; CHECK: printf
+; CHECK-NOT: mov sp, r7
+; CHECK-NOT: sub sp, #12
+; CHECK: pop
+  %4 = tail call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([4 x i8]* @.str, i32 0, i32 0), i32 %2) nounwind ; <i32> [#uses=0]
+  ret i32 0
+}
+
+declare i32 @printf(i8* nocapture, ...) nounwind