[PowerPC] Implement probing for dynamic stack allocation

[lkail]

Summary

This patch is part of supporting -fstack-clash-protection. Mainly do such things compared to existing lowerDynamicAlloc

• Added a new pseudo instruction PPC::PREPARE_PROBED_ALLOC to get actual frame pointer and final stack pointer.
• Synthesize a loop to probe by blocks.
• Use DYNAREAOFFSET to get MaxCallFrameSize which is calculated in prologepilog.

[steven.zhang]

Some minor comments.

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

11746

Can we make it as unsigned integer ?

11775

Adding some comments to indicate what the CFG looks like will make the code more clear.

11829

Can we use the CTR loop ?

llvm/test/CodeGen/PowerPC/stack-clash-dynamic-alloca.ll

4

Check the run for pwr9

[lkail]

Rebased and address steven's comments.

[lkail]

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

11746

Looks getAsInteger already does it.

template <typename T>
std::enable_if_t<!std::numeric_limits<T>::is_signed, bool>
getAsInteger(unsigned Radix, T &Result) const;

[lkail]

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

11829

Yes, we can, but it looks not worth the effort, since the trip count can be zero, we have to add additional code to check this case.

[jsji]

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

7921

Meaningless comment.

7923

Meaningless comment.

11741

Can we set the size to sys::Process::getPageSizeEstimate()? Typical systems are all giving me 65536, are we probing too much if we set it to 4096 here?

11749

What happen with "stack-probe-size"="65536"?

The imm is known value, if we can NOT load an imm > 32768 with one instruction, we should be able to load it in two instructions. Why we have to assert here?

11750

What about "stack-probe-size"="0"? When user would like to opt-out probe for some specific function? See https://reviews.llvm.org/rL225360`. We will now generate code divided by 0...

Also what about "stack-probe-size"="1"? I think we should also round the size to stack alignment similar to Z.

11754

I don't quite understand this. Why we can NOT do this similar to x86 or SystemZ? using PHI nodes.

11770

LLVM_BB ? Why LLVM? Can we use more meaningful name?

11788

calculate?

[lkail]

Address @jsji 's comments. Add test case for large probe size.

[lkail]

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

11741

Yes, sysconf gives 65536 on my rhel POWER machine and 4096 on my rhel X86 machine. I'm quite concerned if users do cross compilation if the host has large page size than the target.

11754

The comment to compare with PPCRegisterInfo::lowerDynamicAlloc might be a little bit misleading. My point is PPCTargetLowering::emitProbedAlloca is unable to expaned in prologepilog like PPCRegisterInfo::lowerDynamicAlloc. Should I delete this comparison?

Looks we don't need phi nodes in the loop, only r1, which is a physical register, is updated.

[jsji]

LGTM.

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

11741

How about we introduce an option for user to switch the default value here, eg: default is still the smallest, 4096, but we provide another choice to use system default (65536 most of the time) .

11754

yeah, maybe remove the comments comparing to lowerDynamicAlloc.

llvm/lib/Target/PowerPC/PPCRegisterInfo.cpp

614

Meaningless comments.

llvm/lib/Target/PowerPC/PPCISelLowering.h

enum NodeType : unsigned {
     /// compute an allocation on the stack.
     DYNALLOC,
 
     /// This instruction is lowered in PPCRegisterInfo::eliminateFrameIndex to
     /// compute an offset from native SP to the address  of the most recent
     /// dynamic alloca.
     DYNAREAOFFSET,
 
+    /// To avoid stack clash, allocation is performed by block and each block is
+    /// probed.
+    PROBED_ALLOCA,
+
     /// GlobalBaseReg - On Darwin, this node represents the result of the mflr
     /// at function entry, used for PIC code.
     GlobalBaseReg,
 
     /// These nodes represent PPC shifts.
     ///
     /// For scalar types, only the last `n + 1` bits of the shift amounts
     /// are used, where n is log2(sizeof(element) * 8). See sld/slw, etc.
MachineBasicBlock *EmitPartwordAtomicBinary(MachineInstr &MI,
                                                 unsigned CmpPred = 0) const;
 
     MachineBasicBlock *emitEHSjLjSetJmp(MachineInstr &MI,
                                         MachineBasicBlock *MBB) const;
 
     MachineBasicBlock *emitEHSjLjLongJmp(MachineInstr &MI,
                                          MachineBasicBlock *MBB) const;
 
+    MachineBasicBlock *emitProbedAlloca(MachineInstr &MI,
+                                        MachineBasicBlock *MBB) const;
+
+    bool hasInlineStackProbe(MachineFunction &MF) const override;
+
+    unsigned getStackProbeSize(MachineFunction &MF) const;
+
     ConstraintType getConstraintType(StringRef Constraint) const override;
 
     /// Examine constraint string and operand type and determine a weight value.
     /// The operand object must already have been set up with the operand type.
     ConstraintWeight getSingleConstraintMatchWeight(
       AsmOperandInfo &info, const char *constraint) const override;
 
     std::pair<unsigned, const TargetRegisterClass *>

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

 cl::desc("enable quad precision float support on ppc"), cl::Hidden);
 
 static cl::opt<bool> UseAbsoluteJumpTables("ppc-use-absolute-jumptables",
 cl::desc("use absolute jump tables on ppc"), cl::Hidden);
 
 STATISTIC(NumTailCalls, "Number of tail calls");
 STATISTIC(NumSiblingCalls, "Number of sibling calls");
 STATISTIC(ShufflesHandledWithVPERM, "Number of shuffles lowered to a VPERM");
+STATISTIC(NumDynamicAllocaProbed, "Number of dynamic stack allocation probed");
 
 static bool isNByteElemShuffleMask(ShuffleVectorSDNode *, unsigned, int);
 
 static SDValue widenVec(SelectionDAG &DAG, SDValue Vec, const SDLoc &dl);
 
 // FIXME: Remove this once the bug has been fixed!
 extern cl::opt<bool> ANDIGlueBug;
 
const char *PPCTargetLowering::getTargetNodeName(unsigned Opcode) const {
   case PPCISD::CMPB:            return "PPCISD::CMPB";
   case PPCISD::Hi:              return "PPCISD::Hi";
   case PPCISD::Lo:              return "PPCISD::Lo";
   case PPCISD::TOC_ENTRY:       return "PPCISD::TOC_ENTRY";
   case PPCISD::ATOMIC_CMP_SWAP_8: return "PPCISD::ATOMIC_CMP_SWAP_8";
   case PPCISD::ATOMIC_CMP_SWAP_16: return "PPCISD::ATOMIC_CMP_SWAP_16";
   case PPCISD::DYNALLOC:        return "PPCISD::DYNALLOC";
   case PPCISD::DYNAREAOFFSET:   return "PPCISD::DYNAREAOFFSET";
+  case PPCISD::PROBED_ALLOCA:   return "PPCISD::PROBED_ALLOCA";
   case PPCISD::GlobalBaseReg:   return "PPCISD::GlobalBaseReg";
   case PPCISD::SRL:             return "PPCISD::SRL";
   case PPCISD::SRA:             return "PPCISD::SRA";
   case PPCISD::SHL:             return "PPCISD::SHL";
   case PPCISD::SRA_ADDZE:       return "PPCISD::SRA_ADDZE";
   case PPCISD::CALL:            return "PPCISD::CALL";
   case PPCISD::CALL_NOP:        return "PPCISD::CALL_NOP";
   case PPCISD::CALL_NOTOC:      return "PPCISD::CALL_NOTOC";
if (!FPSI) {
     // Save the result.
     FI->setFramePointerSaveIndex(FPSI);
   }
   return DAG.getFrameIndex(FPSI, PtrVT);
 }
 
 SDValue PPCTargetLowering::LowerDYNAMIC_STACKALLOC(SDValue Op,
                                                    SelectionDAG &DAG) const {
+  MachineFunction &MF = DAG.getMachineFunction();
   // Get the inputs.
   SDValue Chain = Op.getOperand(0);
   SDValue Size  = Op.getOperand(1);
   SDLoc dl(Op);
 
   // Get the correct type for pointers.
   EVT PtrVT = getPointerTy(DAG.getDataLayout());
   // Negate the size.
   SDValue NegSize = DAG.getNode(ISD::SUB, dl, PtrVT,
                                 DAG.getConstant(0, dl, PtrVT), Size);
   // Construct a node for the frame pointer save index.
   SDValue FPSIdx = getFramePointerFrameIndex(DAG);
-  // Build a DYNALLOC node.
   SDValue Ops[3] = { Chain, NegSize, FPSIdx };
   SDVTList VTs = DAG.getVTList(PtrVT, MVT::Other);
+  if (hasInlineStackProbe(MF))
+    return DAG.getNode(PPCISD::PROBED_ALLOCA, dl, VTs, Ops);
   return DAG.getNode(PPCISD::DYNALLOC, dl, VTs, Ops);
 }
 
 SDValue PPCTargetLowering::LowerEH_DWARF_CFA(SDValue Op,
                                                      SelectionDAG &DAG) const {
   MachineFunction &MF = DAG.getMachineFunction();
 
   bool isPPC64 = Subtarget.isPPC64();
   EVT PtrVT = getPointerTy(DAG.getDataLayout());
 
PPCTargetLowering::emitEHSjLjLongJmp(MachineInstr &MI,
   BuildMI(*MBB, MI, DL,
           TII->get(PVT == MVT::i64 ? PPC::MTCTR8 : PPC::MTCTR)).addReg(Tmp);
   BuildMI(*MBB, MI, DL, TII->get(PVT == MVT::i64 ? PPC::BCTR8 : PPC::BCTR));
 
   MI.eraseFromParent();
   return MBB;
 }
 
+bool PPCTargetLowering::hasInlineStackProbe(MachineFunction &MF) const {
+  // If the function specifically requests inline stack probes, emit them.
+  if (MF.getFunction().hasFnAttribute("probe-stack"))
+    return MF.getFunction().getFnAttribute("probe-stack").getValueAsString() ==
+           "inline-asm";
+  return false;
+}
+
+unsigned PPCTargetLowering::getStackProbeSize(MachineFunction &MF) const {
+  const TargetFrameLowering *TFI = Subtarget.getFrameLowering();
+  unsigned StackAlign = TFI->getStackAlignment();
+  assert(StackAlign >= 1 && isPowerOf2_32(StackAlign) &&
+         "Unexpected stack alignment");
+  // The default stack probe size is 4096 if the function has no
+  // stack-probe-size attribute.
+  unsigned StackProbeSize = 4096;
+  const Function &Fn = MF.getFunction();
+  if (Fn.hasFnAttribute("stack-probe-size"))
+    Fn.getFnAttribute("stack-probe-size")
+        .getValueAsString()
+        .getAsInteger(0, StackProbeSize);
+  // Round down to the stack alignment.
+  StackProbeSize &= ~(StackAlign - 1);
+  return StackProbeSize ? StackProbeSize : StackAlign;
+}
+
+// Lower dynamic stack allocation with probing. It's not implemented like
+// PPCRegisterInfo::lowerDynamicAlloc which is expanded in prologepilog pass,
+// since it involves loop and RegScavenger is unable to allocate registers for
+// loops in prologepilog. emitProbedAlloca is splitted into three phases. In the
+// first phase, it uses pseudo instruction PREPARE_PROBED_ALLOCA to get the
+// future result of actual FramePointer and FinalStackPtr. In the second phase,
+// it generates a loop for probing blocks. At last, it uses pseudo instruction
+// DYNAREAOFFSET to get the future result of MaxCallFrameSize so that it can
+// calculate correct data area pointer.
+MachineBasicBlock *
+PPCTargetLowering::emitProbedAlloca(MachineInstr &MI,
+                                    MachineBasicBlock *MBB) const {
+  const bool isPPC64 = Subtarget.isPPC64();
+  MachineFunction *MF = MBB->getParent();
+  const TargetInstrInfo *TII = Subtarget.getInstrInfo();
+  DebugLoc DL = MI.getDebugLoc();
+  const unsigned ProbeSize = getStackProbeSize(*MF);
+  const BasicBlock *ProbedBB = MBB->getBasicBlock();
+  MachineRegisterInfo &MRI = MF->getRegInfo();
+  // The CFG of probing stack looks as
+  //         +-----+
+  //         | MBB |
+  //         +--+--+
+  //            |
+  //       +----v----+
+  //  +--->+ TestMBB +---+
+  //  |    +----+----+   |
+  //  |         |        |
+  //  |   +-----v----+   |
+  //  +---+ BlockMBB |   |
+  //      +----------+   |
+  //                     |
+  //       +---------+   |
+  //       | TailMBB +<--+
+  //       +---------+
+  // In MBB, calculate previous frame pointer and final stack pointer.
+  // In TestMBB, test if sp is equal to final stack pointer, if so, jump to
+  // TailMBB. In BlockMBB, update the sp atomically and jump back to TestMBB.
+  // TailMBB is spliced via \p MI.
+  MachineBasicBlock *TestMBB = MF->CreateMachineBasicBlock(ProbedBB);
+  MachineBasicBlock *TailMBB = MF->CreateMachineBasicBlock(ProbedBB);
+  MachineBasicBlock *BlockMBB = MF->CreateMachineBasicBlock(ProbedBB);
+
+  MachineFunction::iterator MBBIter = ++MBB->getIterator();
+  MF->insert(MBBIter, TestMBB);
+  MF->insert(MBBIter, BlockMBB);
+  MF->insert(MBBIter, TailMBB);
+
+  const TargetRegisterClass *G8RC = &PPC::G8RCRegClass;
+  const TargetRegisterClass *GPRC = &PPC::GPRCRegClass;
+
+  Register DstReg = MI.getOperand(0).getReg();
+  Register NegSizeReg = MI.getOperand(1).getReg();
+  Register SPReg = isPPC64 ? PPC::X1 : PPC::R1;
+  Register FinalStackPtr = MRI.createVirtualRegister(isPPC64 ? G8RC : GPRC);
+  Register FramePointer = MRI.createVirtualRegister(isPPC64 ? G8RC : GPRC);
+
+  // Get the canonical FinalStackPtr like what
+  // PPCRegisterInfo::lowerDynamicAlloc does.
+  BuildMI(*MBB, {MI}, DL,
+          TII->get(isPPC64 ? PPC::PREPARE_PROBED_ALLOCA_64
+                           : PPC::PREPARE_PROBED_ALLOCA_32),
+          FramePointer)
+      .addDef(FinalStackPtr)
+      .addReg(NegSizeReg)
+      .add(MI.getOperand(2))
+      .add(MI.getOperand(3));
+
+  // Materialize a scratch register for update.
+  int64_t NegProbeSize = -(int64_t)ProbeSize;
+  assert(isInt<32>(NegProbeSize) && "Unhandled probe size!");
+  Register ScratchReg = MRI.createVirtualRegister(isPPC64 ? G8RC : GPRC);
+  if (!isInt<16>(NegProbeSize)) {
+    Register TempReg = MRI.createVirtualRegister(isPPC64 ? G8RC : GPRC);
+    BuildMI(*MBB, {MI}, DL, TII->get(isPPC64 ? PPC::LIS8 : PPC::LIS), TempReg)
+        .addImm(NegProbeSize >> 16);
+    BuildMI(*MBB, {MI}, DL, TII->get(isPPC64 ? PPC::ORI8 : PPC::ORI),
+            ScratchReg)
+        .addReg(TempReg)
+        .addImm(NegProbeSize & 0xFFFF);
+  } else
+    BuildMI(*MBB, {MI}, DL, TII->get(isPPC64 ? PPC::LI8 : PPC::LI), ScratchReg)
+        .addImm(NegProbeSize);
+
+  {
+    // Probing leading residual part.
+    Register Div = MRI.createVirtualRegister(isPPC64 ? G8RC : GPRC);
+    BuildMI(*MBB, {MI}, DL, TII->get(isPPC64 ? PPC::DIVD : PPC::DIVW), Div)
+        .addReg(NegSizeReg)
+        .addReg(ScratchReg);
+    Register Mul = MRI.createVirtualRegister(isPPC64 ? G8RC : GPRC);
+    BuildMI(*MBB, {MI}, DL, TII->get(isPPC64 ? PPC::MULLD : PPC::MULLW), Mul)
+        .addReg(Div)
+        .addReg(ScratchReg);
+    Register NegMod = MRI.createVirtualRegister(isPPC64 ? G8RC : GPRC);
+    BuildMI(*MBB, {MI}, DL, TII->get(isPPC64 ? PPC::SUBF8 : PPC::SUBF), NegMod)
+        .addReg(Mul)
+        .addReg(NegSizeReg);
+    BuildMI(*MBB, {MI}, DL, TII->get(isPPC64 ? PPC::STDUX : PPC::STWUX), SPReg)
+        .addReg(FramePointer)
+        .addReg(SPReg)
+        .addReg(NegMod);
+  }
+
+  {
+    // Remaining part should be multiple of ProbeSize.
+    Register CmpResult = MRI.createVirtualRegister(&PPC::CRRCRegClass);
+    BuildMI(TestMBB, DL, TII->get(isPPC64 ? PPC::CMPD : PPC::CMPW), CmpResult)
+        .addReg(SPReg)
+        .addReg(FinalStackPtr);
+    BuildMI(TestMBB, DL, TII->get(PPC::BCC))
+        .addImm(PPC::PRED_EQ)
+        .addReg(CmpResult)
+        .addMBB(TailMBB);
+    TestMBB->addSuccessor(BlockMBB);
+    TestMBB->addSuccessor(TailMBB);
+  }
+
+  {
+    // Touch the block.
+    // |P...|P...|P...
+    BuildMI(BlockMBB, DL, TII->get(isPPC64 ? PPC::STDUX : PPC::STWUX), SPReg)
+        .addReg(FramePointer)
+        .addReg(SPReg)
+        .addReg(ScratchReg);
+    BuildMI(BlockMBB, DL, TII->get(PPC::B)).addMBB(TestMBB);
+    BlockMBB->addSuccessor(TestMBB);
+  }
+
+  // Calculation of MaxCallFrameSize is deferred to prologepilog, use
+  // DYNAREAOFFSET pseudo instruction to get the future result.
+  Register MaxCallFrameSizeReg =
+      MRI.createVirtualRegister(isPPC64 ? G8RC : GPRC);
+  BuildMI(TailMBB, DL,
+          TII->get(isPPC64 ? PPC::DYNAREAOFFSET8 : PPC::DYNAREAOFFSET),
+          MaxCallFrameSizeReg)
+      .add(MI.getOperand(2))
+      .add(MI.getOperand(3));
+  BuildMI(TailMBB, DL, TII->get(isPPC64 ? PPC::ADD8 : PPC::ADD4), DstReg)
+      .addReg(SPReg)
+      .addReg(MaxCallFrameSizeReg);
+
+  // Splice instructions after MI to TailMBB.
+  TailMBB->splice(TailMBB->end(), MBB,
+                  std::next(MachineBasicBlock::iterator(MI)), MBB->end());
+  TailMBB->transferSuccessorsAndUpdatePHIs(MBB);
+  MBB->addSuccessor(TestMBB);
+
+  // Delete the pseudo instruction.
+  MI.eraseFromParent();
+
+  ++NumDynamicAllocaProbed;
+  return TailMBB;
+}
+
 MachineBasicBlock *
 PPCTargetLowering::EmitInstrWithCustomInserter(MachineInstr &MI,
                                                MachineBasicBlock *BB) const {
   if (MI.getOpcode() == TargetOpcode::STACKMAP ||
       MI.getOpcode() == TargetOpcode::PATCHPOINT) {
     if (Subtarget.is64BitELFABI() &&
         MI.getOpcode() == TargetOpcode::PATCHPOINT &&
         !Subtarget.isUsingPCRelativeCalls()) {
else if (MI.getOpcode() == PPC::ATOMIC_CMP_SWAP_I32 ||
 
     // The mask 255 means that put the 32:63 bits of NewFPSCRReg to the 32:63
     // bits of FPSCR.
     BuildMI(*BB, MI, dl, TII->get(PPC::MTFSF))
       .addImm(255)
       .addReg(NewFPSCRReg)
       .addImm(0)
       .addImm(0);
+  } else if (MI.getOpcode() == PPC::PROBED_ALLOCA_32 ||
+             MI.getOpcode() == PPC::PROBED_ALLOCA_64) {
+    return emitProbedAlloca(MI, BB);
   } else {
     llvm_unreachable("Unexpected instr type to insert");
   }
 
   MI.eraseFromParent(); // The pseudo instruction is gone now.
   return BB;
 }
 

llvm/lib/Target/PowerPC/PPCInstr64Bit.td

 // the POWER3.
 
 let Defs = [X1], Uses = [X1] in
 def DYNALLOC8 : PPCEmitTimePseudo<(outs g8rc:$result), (ins g8rc:$negsize, memri:$fpsi),"#DYNALLOC8",
                        [(set i64:$result,
                              (PPCdynalloc i64:$negsize, iaddr:$fpsi))]>;
 def DYNAREAOFFSET8 : PPCEmitTimePseudo<(outs i64imm:$result), (ins memri:$fpsi), "#DYNAREAOFFSET8",
                        [(set i64:$result, (PPCdynareaoffset iaddr:$fpsi))]>;
+// Probed alloca to support stack clash protection.
+let Defs = [X1], Uses = [X1], hasNoSchedulingInfo = 1 in {
+def PROBED_ALLOCA_64 : PPCCustomInserterPseudo<(outs g8rc:$result),
+                         (ins g8rc:$negsize, memri:$fpsi), "#PROBED_ALLOCA_64",
+                           [(set i64:$result,
+                             (PPCprobedalloca i64:$negsize, iaddr:$fpsi))]>;
+def PREPARE_PROBED_ALLOCA_64 : PPCEmitTimePseudo<(outs g8rc:$fp,
+    g8rc:$sp),
+    (ins g8rc:$negsize, memri:$fpsi), "#PREPARE_PROBED_ALLOCA_64", []>;
+}
 
 let hasSideEffects = 0 in {
 let Defs = [LR8] in {
 def MTLR8  : XFXForm_7_ext<31, 467, 8, (outs), (ins g8rc:$rS),
                            "mtlr $rS", IIC_SprMTSPR>,
              PPC970_DGroup_First, PPC970_Unit_FXU;
 }
 let Uses = [LR8] in {

llvm/lib/Target/PowerPC/PPCInstrInfo.td

 def PPCcr6unset : SDNode<"PPCISD::CR6UNSET", SDTNone,
                          [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]>;
 
 // Instructions to support dynamic alloca.
 def SDTDynOp  : SDTypeProfile<1, 2, []>;
 def SDTDynAreaOp  : SDTypeProfile<1, 1, []>;
 def PPCdynalloc   : SDNode<"PPCISD::DYNALLOC", SDTDynOp, [SDNPHasChain]>;
 def PPCdynareaoffset   : SDNode<"PPCISD::DYNAREAOFFSET", SDTDynAreaOp, [SDNPHasChain]>;
+def PPCprobedalloca : SDNode<"PPCISD::PROBED_ALLOCA", SDTDynOp, [SDNPHasChain]>;
 
 // PC Relative Specific Nodes
 def PPCmatpcreladdr : SDNode<"PPCISD::MAT_PCREL_ADDR", SDTIntUnaryOp, []>;
 
 //===----------------------------------------------------------------------===//
 // PowerPC specific transformation functions and pattern fragments.
 //
 

 }
 
 let Defs = [R1], Uses = [R1] in
 def DYNALLOC : PPCEmitTimePseudo<(outs gprc:$result), (ins gprc:$negsize, memri:$fpsi), "#DYNALLOC",
                        [(set i32:$result,
                              (PPCdynalloc i32:$negsize, iaddr:$fpsi))]>;
 def DYNAREAOFFSET : PPCEmitTimePseudo<(outs i32imm:$result), (ins memri:$fpsi), "#DYNAREAOFFSET",
                        [(set i32:$result, (PPCdynareaoffset iaddr:$fpsi))]>;
+// Probed alloca to support stack clash protection.
+let Defs = [R1], Uses = [R1], hasNoSchedulingInfo = 1 in {
+def PROBED_ALLOCA_32 : PPCCustomInserterPseudo<(outs gprc:$result),
+                         (ins gprc:$negsize, memri:$fpsi), "#PROBED_ALLOCA_32",
+                           [(set i32:$result,
+                             (PPCprobedalloca i32:$negsize, iaddr:$fpsi))]>;
+def PREPARE_PROBED_ALLOCA_32 : PPCEmitTimePseudo<(outs gprc:$fp,
+    gprc:$sp),
+    (ins gprc:$negsize, memri:$fpsi), "#PREPARE_PROBED_ALLOCA_32", []>;
+}
                          
 // SELECT_CC_* - Used to implement the SELECT_CC DAG operation.  Expanded after
 // instruction selection into a branch sequence.
 let PPC970_Single = 1 in {
   // Note that SELECT_CC_I4 and SELECT_CC_I8 use the no-r0 register classes
   // because either operand might become the first operand in an isel, and
   // that operand cannot be r0.
   def SELECT_CC_I4 : PPCCustomInserterPseudo<(outs gprc:$dst), (ins crrc:$cond,
                               gprc_nor0:$T, gprc_nor0:$F,

llvm/lib/Target/PowerPC/PPCRegisterInfo.h

bool requiresVirtualBaseRegisters(const MachineFunction &MF) const override {
     return true;
   }
 
   void lowerDynamicAlloc(MachineBasicBlock::iterator II) const;
   void lowerDynamicAreaOffset(MachineBasicBlock::iterator II) const;
   void prepareDynamicAlloca(MachineBasicBlock::iterator II,
                             Register &NegSizeReg, bool &KillNegSizeReg,
                             Register &FramePointer) const;
+  void lowerPrepareProbedAlloca(MachineBasicBlock::iterator II) const;
   void lowerCRSpilling(MachineBasicBlock::iterator II,
                        unsigned FrameIndex) const;
   void lowerCRRestore(MachineBasicBlock::iterator II,
                       unsigned FrameIndex) const;
   void lowerCRBitSpilling(MachineBasicBlock::iterator II,
                           unsigned FrameIndex) const;
   void lowerCRBitRestore(MachineBasicBlock::iterator II,
                          unsigned FrameIndex) const;

llvm/lib/Target/PowerPC/PPCRegisterInfo.cpp

if (MaxAlign > TargetAlign) {
       BuildMI(MBB, II, dl, TII.get(PPC::AND), NegSizeReg)
         .addReg(UnalNegSizeReg, getKillRegState(KillNegSizeReg))
         .addReg(NegSizeReg1, RegState::Kill);
       KillNegSizeReg = true;
     }
   }
 }
 
+void PPCRegisterInfo::lowerPrepareProbedAlloca(
+    MachineBasicBlock::iterator II) const {
+  // Get the instruction.
+  MachineInstr &MI = *II;
+  // Get the instruction's basic block.
+  MachineBasicBlock &MBB = *MI.getParent();
+  // Get the basic block's function.
+  MachineFunction &MF = *MBB.getParent();
+  const PPCSubtarget &Subtarget = MF.getSubtarget<PPCSubtarget>();
+  // Get the instruction info.
+  const TargetInstrInfo &TII = *Subtarget.getInstrInfo();
+  // Determine whether 64-bit pointers are used.
+  bool LP64 = TM.isPPC64();
+  DebugLoc dl = MI.getDebugLoc();
+  Register FramePointer = MI.getOperand(0).getReg();
+  Register FinalStackPtr = MI.getOperand(1).getReg();
+  bool KillNegSizeReg = MI.getOperand(2).isKill();
+  Register NegSizeReg = MI.getOperand(2).getReg();
+  prepareDynamicAlloca(II, NegSizeReg, KillNegSizeReg, FramePointer);
+  if (LP64) {
+    BuildMI(MBB, II, dl, TII.get(PPC::ADD8), FinalStackPtr)
+        .addReg(PPC::X1)
+        .addReg(NegSizeReg, getKillRegState(KillNegSizeReg));
+
+  } else {
+    BuildMI(MBB, II, dl, TII.get(PPC::ADD4), FinalStackPtr)
+        .addReg(PPC::R1)
+        .addReg(NegSizeReg, getKillRegState(KillNegSizeReg));
+  }
+
+  MBB.erase(II);
+}
+
 void PPCRegisterInfo::lowerDynamicAreaOffset(
     MachineBasicBlock::iterator II) const {
   // Get the instruction.
   MachineInstr &MI = *II;
   // Get the instruction's basic block.
   MachineBasicBlock &MBB = *MI.getParent();
   // Get the basic block's function.
   MachineFunction &MF = *MBB.getParent();
PPCRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
 
   // Special case for dynamic alloca.
   if (FPSI && FrameIndex == FPSI &&
       (OpC == PPC::DYNALLOC || OpC == PPC::DYNALLOC8)) {
     lowerDynamicAlloc(II);
     return;
   }
 
+  if (FPSI && FrameIndex == FPSI &&
+      (OpC == PPC::PREPARE_PROBED_ALLOCA_64 ||
+       OpC == PPC::PREPARE_PROBED_ALLOCA_32)) {
+    lowerPrepareProbedAlloca(II);
+    return;
+  }
+
   // Special case for pseudo-ops SPILL_CR and RESTORE_CR, etc.
   if (OpC == PPC::SPILL_CR) {
     lowerCRSpilling(II, FrameIndex);
     return;
   } else if (OpC == PPC::RESTORE_CR) {
     lowerCRRestore(II, FrameIndex);
     return;
   } else if (OpC == PPC::SPILL_CRBIT) {

llvm/test/CodeGen/PowerPC/stack-clash-dynamic-alloca.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -ppc-asm-full-reg-names -verify-machineinstrs \
+; RUN:   -mtriple=powerpc64le-linux-gnu < %s | FileCheck \
+; RUN:   -check-prefix=CHECK-LE %s
+; RUN: llc -ppc-asm-full-reg-names -verify-machineinstrs \
+; RUN:   -mtriple=powerpc64le-linux-gnu -mcpu=pwr9 < %s | FileCheck \
+; RUN:   -check-prefix=CHECK-P9-LE %s
+; RUN: llc -ppc-asm-full-reg-names -verify-machineinstrs \
+; RUN:   -mtriple=powerpc64-linux-gnu < %s | FileCheck \
+; RUN:   -check-prefix=CHECK-BE %s
+; RUN: llc -ppc-asm-full-reg-names -verify-machineinstrs \
+; RUN:   -mtriple=powerpc-linux-gnu < %s | FileCheck \
+; RUN:   -check-prefix=CHECK-32 %s
+
+define i32 @foo(i32 %n) local_unnamed_addr #0 "stack-probe-size"="32768" nounwind {
+; CHECK-LE-LABEL: foo:
+; CHECK-LE:       # %bb.0:
+; CHECK-LE-NEXT:    std r31, -8(r1)
+; CHECK-LE-NEXT:    stdu r1, -48(r1)
+; CHECK-LE-NEXT:    rldic r3, r3, 2, 30
+; CHECK-LE-NEXT:    li r6, -32768
+; CHECK-LE-NEXT:    mr r31, r1
+; CHECK-LE-NEXT:    addi r3, r3, 15
+; CHECK-LE-NEXT:    addi r4, r31, 48
+; CHECK-LE-NEXT:    rldicl r3, r3, 60, 4
+; CHECK-LE-NEXT:    rldicl r3, r3, 4, 29
+; CHECK-LE-NEXT:    neg r5, r3
+; CHECK-LE-NEXT:    divd r7, r5, r6
+; CHECK-LE-NEXT:    add r3, r1, r5
+; CHECK-LE-NEXT:    mulld r6, r7, r6
+; CHECK-LE-NEXT:    sub r5, r5, r6
+; CHECK-LE-NEXT:    stdux r4, r1, r5
+; CHECK-LE-NEXT:    cmpd r1, r3
+; CHECK-LE-NEXT:    beq cr0, .LBB0_2
+; CHECK-LE-NEXT:  .LBB0_1:
+; CHECK-LE-NEXT:    stdu r4, -32768(r1)
+; CHECK-LE-NEXT:    cmpd r1, r3
+; CHECK-LE-NEXT:    bne cr0, .LBB0_1
+; CHECK-LE-NEXT:  .LBB0_2:
+; CHECK-LE-NEXT:    li r4, 1
+; CHECK-LE-NEXT:    addi r3, r1, 32
+; CHECK-LE-NEXT:    stw r4, 4792(r3)
+; CHECK-LE-NEXT:    lwz r3, 0(r3)
+; CHECK-LE-NEXT:    ld r1, 0(r1)
+; CHECK-LE-NEXT:    ld r31, -8(r1)
+; CHECK-LE-NEXT:    blr
+;
+; CHECK-P9-LE-LABEL: foo:
+; CHECK-P9-LE:       # %bb.0:
+; CHECK-P9-LE-NEXT:    std r31, -8(r1)
+; CHECK-P9-LE-NEXT:    stdu r1, -48(r1)
+; CHECK-P9-LE-NEXT:    rldic r3, r3, 2, 30
+; CHECK-P9-LE-NEXT:    addi r3, r3, 15
+; CHECK-P9-LE-NEXT:    rldicl r3, r3, 60, 4
+; CHECK-P9-LE-NEXT:    rldicl r3, r3, 4, 29
+; CHECK-P9-LE-NEXT:    neg r5, r3
+; CHECK-P9-LE-NEXT:    li r6, -32768
+; CHECK-P9-LE-NEXT:    divd r7, r5, r6
+; CHECK-P9-LE-NEXT:    mulld r6, r7, r6
+; CHECK-P9-LE-NEXT:    mr r31, r1
+; CHECK-P9-LE-NEXT:    addi r4, r31, 48
+; CHECK-P9-LE-NEXT:    add r3, r1, r5
+; CHECK-P9-LE-NEXT:    sub r5, r5, r6
+; CHECK-P9-LE-NEXT:    stdux r4, r1, r5
+; CHECK-P9-LE-NEXT:    cmpd r1, r3
+; CHECK-P9-LE-NEXT:    beq cr0, .LBB0_2
+; CHECK-P9-LE-NEXT:  .LBB0_1:
+; CHECK-P9-LE-NEXT:    stdu r4, -32768(r1)
+; CHECK-P9-LE-NEXT:    cmpd r1, r3
+; CHECK-P9-LE-NEXT:    bne cr0, .LBB0_1
+; CHECK-P9-LE-NEXT:  .LBB0_2:
+; CHECK-P9-LE-NEXT:    addi r3, r1, 32
+; CHECK-P9-LE-NEXT:    li r4, 1
+; CHECK-P9-LE-NEXT:    stw r4, 4792(r3)
+; CHECK-P9-LE-NEXT:    lwz r3, 0(r3)
+; CHECK-P9-LE-NEXT:    ld r1, 0(r1)
+; CHECK-P9-LE-NEXT:    ld r31, -8(r1)
+; CHECK-P9-LE-NEXT:    blr
+;
+; CHECK-BE-LABEL: foo:
+; CHECK-BE:       # %bb.0:
+; CHECK-BE-NEXT:    std r31, -8(r1)
+; CHECK-BE-NEXT:    stdu r1, -64(r1)
+; CHECK-BE-NEXT:    rldic r3, r3, 2, 30
+; CHECK-BE-NEXT:    li r6, -32768
+; CHECK-BE-NEXT:    addi r3, r3, 15
+; CHECK-BE-NEXT:    rldicl r3, r3, 60, 4
+; CHECK-BE-NEXT:    mr r31, r1
+; CHECK-BE-NEXT:    rldicl r3, r3, 4, 29
+; CHECK-BE-NEXT:    addi r4, r31, 64
+; CHECK-BE-NEXT:    neg r5, r3
+; CHECK-BE-NEXT:    divd r7, r5, r6
+; CHECK-BE-NEXT:    add r3, r1, r5
+; CHECK-BE-NEXT:    mulld r6, r7, r6
+; CHECK-BE-NEXT:    sub r5, r5, r6
+; CHECK-BE-NEXT:    stdux r4, r1, r5
+; CHECK-BE-NEXT:    cmpd r1, r3
+; CHECK-BE-NEXT:    beq cr0, .LBB0_2
+; CHECK-BE-NEXT:  .LBB0_1:
+; CHECK-BE-NEXT:    stdu r4, -32768(r1)
+; CHECK-BE-NEXT:    cmpd r1, r3
+; CHECK-BE-NEXT:    bne cr0, .LBB0_1
+; CHECK-BE-NEXT:  .LBB0_2:
+; CHECK-BE-NEXT:    li r4, 1
+; CHECK-BE-NEXT:    addi r3, r1, 48
+; CHECK-BE-NEXT:    stw r4, 4792(r3)
+; CHECK-BE-NEXT:    lwz r3, 0(r3)
+; CHECK-BE-NEXT:    ld r1, 0(r1)
+; CHECK-BE-NEXT:    ld r31, -8(r1)
+; CHECK-BE-NEXT:    blr
+;
+; CHECK-32-LABEL: foo:
+; CHECK-32:       # %bb.0:
+; CHECK-32-NEXT:    stwu r1, -32(r1)
+; CHECK-32-NEXT:    slwi r3, r3, 2
+; CHECK-32-NEXT:    addi r3, r3, 15
+; CHECK-32-NEXT:    rlwinm r3, r3, 0, 0, 27
+; CHECK-32-NEXT:    neg r5, r3
+; CHECK-32-NEXT:    li r6, -32768
+; CHECK-32-NEXT:    divw r7, r5, r6
+; CHECK-32-NEXT:    stw r31, 28(r1)
+; CHECK-32-NEXT:    mr r31, r1
+; CHECK-32-NEXT:    addi r4, r31, 32
+; CHECK-32-NEXT:    add r3, r1, r5
+; CHECK-32-NEXT:    mullw r6, r7, r6
+; CHECK-32-NEXT:    sub r5, r5, r6
+; CHECK-32-NEXT:    stwux r4, r1, r5
+; CHECK-32-NEXT:    cmpw r1, r3
+; CHECK-32-NEXT:    beq cr0, .LBB0_2
+; CHECK-32-NEXT:  .LBB0_1:
+; CHECK-32-NEXT:    stwu r4, -32768(r1)
+; CHECK-32-NEXT:    cmpw r1, r3
+; CHECK-32-NEXT:    bne cr0, .LBB0_1
+; CHECK-32-NEXT:  .LBB0_2:
+; CHECK-32-NEXT:    li r4, 1
+; CHECK-32-NEXT:    addi r3, r1, 16
+; CHECK-32-NEXT:    stw r4, 4792(r3)
+; CHECK-32-NEXT:    lwz r3, 0(r3)
+; CHECK-32-NEXT:    lwz r31, 0(r1)
+; CHECK-32-NEXT:    lwz r0, -4(r31)
+; CHECK-32-NEXT:    mr r1, r31
+; CHECK-32-NEXT:    mr r31, r0
+; CHECK-32-NEXT:    blr
+  %a = alloca i32, i32 %n, align 16
+  %b = getelementptr inbounds i32, i32* %a, i64 1198
+  store volatile i32 1, i32* %b
+  %c = load volatile i32, i32* %a
+  ret i32 %c
+}
+
+define i32 @bar(i32 %n) local_unnamed_addr #0 nounwind {
+; CHECK-LE-LABEL: bar:
+; CHECK-LE:       # %bb.0:
+; CHECK-LE-NEXT:    std r31, -8(r1)
+; CHECK-LE-NEXT:    stdu r1, -48(r1)
+; CHECK-LE-NEXT:    rldic r4, r3, 2, 30
+; CHECK-LE-NEXT:    li r7, -4096
+; CHECK-LE-NEXT:    mr r31, r1
+; CHECK-LE-NEXT:    addi r4, r4, 15
+; CHECK-LE-NEXT:    addi r5, r31, 48
+; CHECK-LE-NEXT:    rldicl r4, r4, 60, 4
+; CHECK-LE-NEXT:    rldicl r4, r4, 4, 29
+; CHECK-LE-NEXT:    neg r6, r4
+; CHECK-LE-NEXT:    divd r8, r6, r7
+; CHECK-LE-NEXT:    add r4, r1, r6
+; CHECK-LE-NEXT:    mulld r7, r8, r7
+; CHECK-LE-NEXT:    sub r6, r6, r7
+; CHECK-LE-NEXT:    stdux r5, r1, r6
+; CHECK-LE-NEXT:    cmpd r1, r4
+; CHECK-LE-NEXT:    beq cr0, .LBB1_2
+; CHECK-LE-NEXT:  .LBB1_1:
+; CHECK-LE-NEXT:    stdu r5, -4096(r1)
+; CHECK-LE-NEXT:    cmpd r1, r4
+; CHECK-LE-NEXT:    bne cr0, .LBB1_1
+; CHECK-LE-NEXT:  .LBB1_2:
+; CHECK-LE-NEXT:    extsw r3, r3
+; CHECK-LE-NEXT:    li r5, 1
+; CHECK-LE-NEXT:    addi r4, r1, 32
+; CHECK-LE-NEXT:    sldi r3, r3, 2
+; CHECK-LE-NEXT:    add r3, r4, r3
+; CHECK-LE-NEXT:    stw r5, 4096(r3)
+; CHECK-LE-NEXT:    lwz r3, 0(r4)
+; CHECK-LE-NEXT:    ld r1, 0(r1)
+; CHECK-LE-NEXT:    ld r31, -8(r1)
+; CHECK-LE-NEXT:    blr
+;
+; CHECK-P9-LE-LABEL: bar:
+; CHECK-P9-LE:       # %bb.0:
+; CHECK-P9-LE-NEXT:    std r31, -8(r1)
+; CHECK-P9-LE-NEXT:    stdu r1, -48(r1)
+; CHECK-P9-LE-NEXT:    rldic r4, r3, 2, 30
+; CHECK-P9-LE-NEXT:    addi r4, r4, 15
+; CHECK-P9-LE-NEXT:    rldicl r4, r4, 60, 4
+; CHECK-P9-LE-NEXT:    rldicl r4, r4, 4, 29
+; CHECK-P9-LE-NEXT:    neg r6, r4
+; CHECK-P9-LE-NEXT:    li r7, -4096
+; CHECK-P9-LE-NEXT:    divd r8, r6, r7
+; CHECK-P9-LE-NEXT:    mulld r7, r8, r7
+; CHECK-P9-LE-NEXT:    mr r31, r1
+; CHECK-P9-LE-NEXT:    addi r5, r31, 48
+; CHECK-P9-LE-NEXT:    add r4, r1, r6
+; CHECK-P9-LE-NEXT:    sub r6, r6, r7
+; CHECK-P9-LE-NEXT:    stdux r5, r1, r6
+; CHECK-P9-LE-NEXT:    cmpd r1, r4
+; CHECK-P9-LE-NEXT:    beq cr0, .LBB1_2
+; CHECK-P9-LE-NEXT:  .LBB1_1:
+; CHECK-P9-LE-NEXT:    stdu r5, -4096(r1)
+; CHECK-P9-LE-NEXT:    cmpd r1, r4
+; CHECK-P9-LE-NEXT:    bne cr0, .LBB1_1
+; CHECK-P9-LE-NEXT:  .LBB1_2:
+; CHECK-P9-LE-NEXT:    addi r4, r1, 32
+; CHECK-P9-LE-NEXT:    extswsli r3, r3, 2
+; CHECK-P9-LE-NEXT:    add r3, r4, r3
+; CHECK-P9-LE-NEXT:    li r5, 1
+; CHECK-P9-LE-NEXT:    stw r5, 4096(r3)
+; CHECK-P9-LE-NEXT:    lwz r3, 0(r4)
+; CHECK-P9-LE-NEXT:    ld r1, 0(r1)
+; CHECK-P9-LE-NEXT:    ld r31, -8(r1)
+; CHECK-P9-LE-NEXT:    blr
+;
+; CHECK-BE-LABEL: bar:
+; CHECK-BE:       # %bb.0:
+; CHECK-BE-NEXT:    std r31, -8(r1)
+; CHECK-BE-NEXT:    stdu r1, -64(r1)
+; CHECK-BE-NEXT:    rldic r4, r3, 2, 30
+; CHECK-BE-NEXT:    li r7, -4096
+; CHECK-BE-NEXT:    addi r4, r4, 15
+; CHECK-BE-NEXT:    rldicl r4, r4, 60, 4
+; CHECK-BE-NEXT:    mr r31, r1
+; CHECK-BE-NEXT:    rldicl r4, r4, 4, 29
+; CHECK-BE-NEXT:    addi r5, r31, 64
+; CHECK-BE-NEXT:    neg r6, r4
+; CHECK-BE-NEXT:    divd r8, r6, r7
+; CHECK-BE-NEXT:    add r4, r1, r6
+; CHECK-BE-NEXT:    mulld r7, r8, r7
+; CHECK-BE-NEXT:    sub r6, r6, r7
+; CHECK-BE-NEXT:    stdux r5, r1, r6
+; CHECK-BE-NEXT:    cmpd r1, r4
+; CHECK-BE-NEXT:    beq cr0, .LBB1_2
+; CHECK-BE-NEXT:  .LBB1_1:
+; CHECK-BE-NEXT:    stdu r5, -4096(r1)
+; CHECK-BE-NEXT:    cmpd r1, r4
+; CHECK-BE-NEXT:    bne cr0, .LBB1_1
+; CHECK-BE-NEXT:  .LBB1_2:
+; CHECK-BE-NEXT:    extsw r3, r3
+; CHECK-BE-NEXT:    addi r4, r1, 48
+; CHECK-BE-NEXT:    sldi r3, r3, 2
+; CHECK-BE-NEXT:    li r5, 1
+; CHECK-BE-NEXT:    add r3, r4, r3
+; CHECK-BE-NEXT:    stw r5, 4096(r3)
+; CHECK-BE-NEXT:    lwz r3, 0(r4)
+; CHECK-BE-NEXT:    ld r1, 0(r1)
+; CHECK-BE-NEXT:    ld r31, -8(r1)
+; CHECK-BE-NEXT:    blr
+;
+; CHECK-32-LABEL: bar:
+; CHECK-32:       # %bb.0:
+; CHECK-32-NEXT:    stwu r1, -32(r1)
+; CHECK-32-NEXT:    slwi r3, r3, 2
+; CHECK-32-NEXT:    addi r4, r3, 15
+; CHECK-32-NEXT:    rlwinm r4, r4, 0, 0, 27
+; CHECK-32-NEXT:    neg r6, r4
+; CHECK-32-NEXT:    li r7, -4096
+; CHECK-32-NEXT:    divw r8, r6, r7
+; CHECK-32-NEXT:    stw r31, 28(r1)
+; CHECK-32-NEXT:    mr r31, r1
+; CHECK-32-NEXT:    addi r5, r31, 32
+; CHECK-32-NEXT:    add r4, r1, r6
+; CHECK-32-NEXT:    mullw r7, r8, r7
+; CHECK-32-NEXT:    sub r6, r6, r7
+; CHECK-32-NEXT:    stwux r5, r1, r6
+; CHECK-32-NEXT:    cmpw r1, r4
+; CHECK-32-NEXT:    beq cr0, .LBB1_2
+; CHECK-32-NEXT:  .LBB1_1:
+; CHECK-32-NEXT:    stwu r5, -4096(r1)
+; CHECK-32-NEXT:    cmpw r1, r4
+; CHECK-32-NEXT:    bne cr0, .LBB1_1
+; CHECK-32-NEXT:  .LBB1_2:
+; CHECK-32-NEXT:    addi r4, r1, 16
+; CHECK-32-NEXT:    li r5, 1
+; CHECK-32-NEXT:    add r3, r4, r3
+; CHECK-32-NEXT:    stw r5, 4096(r3)
+; CHECK-32-NEXT:    lwz r3, 0(r4)
+; CHECK-32-NEXT:    lwz r31, 0(r1)
+; CHECK-32-NEXT:    lwz r0, -4(r31)
+; CHECK-32-NEXT:    mr r1, r31
+; CHECK-32-NEXT:    mr r31, r0
+; CHECK-32-NEXT:    blr
+  %a = alloca i32, i32 %n, align 16
+  %i = add i32 %n, 1024
+  %b = getelementptr inbounds i32, i32* %a, i32 %i
+  store volatile i32 1, i32* %b
+  %c = load volatile i32, i32* %a
+  ret i32 %c
+}
+
+define i32 @f(i32 %n) local_unnamed_addr #0 "stack-probe-size"="65536" nounwind {
+; CHECK-LE-LABEL: f:
+; CHECK-LE:       # %bb.0:
+; CHECK-LE-NEXT:    std r31, -8(r1)
+; CHECK-LE-NEXT:    stdu r1, -48(r1)
+; CHECK-LE-NEXT:    rldic r3, r3, 2, 30
+; CHECK-LE-NEXT:    lis r5, -1
+; CHECK-LE-NEXT:    mr r31, r1
+; CHECK-LE-NEXT:    addi r3, r3, 15
+; CHECK-LE-NEXT:    ori r5, r5, 0
+; CHECK-LE-NEXT:    addi r4, r31, 48
+; CHECK-LE-NEXT:    rldicl r3, r3, 60, 4
+; CHECK-LE-NEXT:    rldicl r3, r3, 4, 29
+; CHECK-LE-NEXT:    neg r6, r3
+; CHECK-LE-NEXT:    divd r7, r6, r5
+; CHECK-LE-NEXT:    add r3, r1, r6
+; CHECK-LE-NEXT:    mulld r7, r7, r5
+; CHECK-LE-NEXT:    sub r6, r6, r7
+; CHECK-LE-NEXT:    stdux r4, r1, r6
+; CHECK-LE-NEXT:    cmpd r1, r3
+; CHECK-LE-NEXT:    beq cr0, .LBB2_2
+; CHECK-LE-NEXT:  .LBB2_1:
+; CHECK-LE-NEXT:    stdux r4, r1, r5
+; CHECK-LE-NEXT:    cmpd r1, r3
+; CHECK-LE-NEXT:    bne cr0, .LBB2_1
+; CHECK-LE-NEXT:  .LBB2_2:
+; CHECK-LE-NEXT:    li r4, 1
+; CHECK-LE-NEXT:    addi r3, r1, 32
+; CHECK-LE-NEXT:    stw r4, 4792(r3)
+; CHECK-LE-NEXT:    lwz r3, 0(r3)
+; CHECK-LE-NEXT:    ld r1, 0(r1)
+; CHECK-LE-NEXT:    ld r31, -8(r1)
+; CHECK-LE-NEXT:    blr
+;
+; CHECK-P9-LE-LABEL: f:
+; CHECK-P9-LE:       # %bb.0:
+; CHECK-P9-LE-NEXT:    std r31, -8(r1)
+; CHECK-P9-LE-NEXT:    stdu r1, -48(r1)
+; CHECK-P9-LE-NEXT:    rldic r3, r3, 2, 30
+; CHECK-P9-LE-NEXT:    addi r3, r3, 15
+; CHECK-P9-LE-NEXT:    rldicl r3, r3, 60, 4
+; CHECK-P9-LE-NEXT:    rldicl r3, r3, 4, 29
+; CHECK-P9-LE-NEXT:    lis r5, -1
+; CHECK-P9-LE-NEXT:    ori r5, r5, 0
+; CHECK-P9-LE-NEXT:    neg r6, r3
+; CHECK-P9-LE-NEXT:    divd r7, r6, r5
+; CHECK-P9-LE-NEXT:    mulld r7, r7, r5
+; CHECK-P9-LE-NEXT:    mr r31, r1
+; CHECK-P9-LE-NEXT:    addi r4, r31, 48
+; CHECK-P9-LE-NEXT:    add r3, r1, r6
+; CHECK-P9-LE-NEXT:    sub r6, r6, r7
+; CHECK-P9-LE-NEXT:    stdux r4, r1, r6
+; CHECK-P9-LE-NEXT:    cmpd r1, r3
+; CHECK-P9-LE-NEXT:    beq cr0, .LBB2_2
+; CHECK-P9-LE-NEXT:  .LBB2_1:
+; CHECK-P9-LE-NEXT:    stdux r4, r1, r5
+; CHECK-P9-LE-NEXT:    cmpd r1, r3
+; CHECK-P9-LE-NEXT:    bne cr0, .LBB2_1
+; CHECK-P9-LE-NEXT:  .LBB2_2:
+; CHECK-P9-LE-NEXT:    addi r3, r1, 32
+; CHECK-P9-LE-NEXT:    li r4, 1
+; CHECK-P9-LE-NEXT:    stw r4, 4792(r3)
+; CHECK-P9-LE-NEXT:    lwz r3, 0(r3)
+; CHECK-P9-LE-NEXT:    ld r1, 0(r1)
+; CHECK-P9-LE-NEXT:    ld r31, -8(r1)
+; CHECK-P9-LE-NEXT:    blr
+;
+; CHECK-BE-LABEL: f:
+; CHECK-BE:       # %bb.0:
+; CHECK-BE-NEXT:    std r31, -8(r1)
+; CHECK-BE-NEXT:    stdu r1, -64(r1)
+; CHECK-BE-NEXT:    rldic r3, r3, 2, 30
+; CHECK-BE-NEXT:    lis r5, -1
+; CHECK-BE-NEXT:    addi r3, r3, 15
+; CHECK-BE-NEXT:    rldicl r3, r3, 60, 4
+; CHECK-BE-NEXT:    ori r5, r5, 0
+; CHECK-BE-NEXT:    rldicl r3, r3, 4, 29
+; CHECK-BE-NEXT:    mr r31, r1
+; CHECK-BE-NEXT:    neg r6, r3
+; CHECK-BE-NEXT:    divd r7, r6, r5
+; CHECK-BE-NEXT:    addi r4, r31, 64
+; CHECK-BE-NEXT:    mulld r7, r7, r5
+; CHECK-BE-NEXT:    add r3, r1, r6
+; CHECK-BE-NEXT:    sub r6, r6, r7
+; CHECK-BE-NEXT:    stdux r4, r1, r6
+; CHECK-BE-NEXT:    cmpd r1, r3
+; CHECK-BE-NEXT:    beq cr0, .LBB2_2
+; CHECK-BE-NEXT:  .LBB2_1:
+; CHECK-BE-NEXT:    stdux r4, r1, r5
+; CHECK-BE-NEXT:    cmpd r1, r3
+; CHECK-BE-NEXT:    bne cr0, .LBB2_1
+; CHECK-BE-NEXT:  .LBB2_2:
+; CHECK-BE-NEXT:    li r4, 1
+; CHECK-BE-NEXT:    addi r3, r1, 48
+; CHECK-BE-NEXT:    stw r4, 4792(r3)
+; CHECK-BE-NEXT:    lwz r3, 0(r3)
+; CHECK-BE-NEXT:    ld r1, 0(r1)
+; CHECK-BE-NEXT:    ld r31, -8(r1)
+; CHECK-BE-NEXT:    blr
+;
+; CHECK-32-LABEL: f:
+; CHECK-32:       # %bb.0:
+; CHECK-32-NEXT:    stwu r1, -32(r1)
+; CHECK-32-NEXT:    slwi r3, r3, 2
+; CHECK-32-NEXT:    addi r3, r3, 15
+; CHECK-32-NEXT:    rlwinm r3, r3, 0, 0, 27
+; CHECK-32-NEXT:    lis r5, -1
+; CHECK-32-NEXT:    neg r6, r3
+; CHECK-32-NEXT:    ori r5, r5, 0
+; CHECK-32-NEXT:    divw r7, r6, r5
+; CHECK-32-NEXT:    stw r31, 28(r1)
+; CHECK-32-NEXT:    mr r31, r1
+; CHECK-32-NEXT:    addi r4, r31, 32
+; CHECK-32-NEXT:    add r3, r1, r6
+; CHECK-32-NEXT:    mullw r7, r7, r5
+; CHECK-32-NEXT:    sub r6, r6, r7
+; CHECK-32-NEXT:    stwux r4, r1, r6
+; CHECK-32-NEXT:    cmpw r1, r3
+; CHECK-32-NEXT:    beq cr0, .LBB2_2
+; CHECK-32-NEXT:  .LBB2_1:
+; CHECK-32-NEXT:    stwux r4, r1, r5
+; CHECK-32-NEXT:    cmpw r1, r3
+; CHECK-32-NEXT:    bne cr0, .LBB2_1
+; CHECK-32-NEXT:  .LBB2_2:
+; CHECK-32-NEXT:    li r4, 1
+; CHECK-32-NEXT:    addi r3, r1, 16
+; CHECK-32-NEXT:    stw r4, 4792(r3)
+; CHECK-32-NEXT:    lwz r3, 0(r3)
+; CHECK-32-NEXT:    lwz r31, 0(r1)
+; CHECK-32-NEXT:    lwz r0, -4(r31)
+; CHECK-32-NEXT:    mr r1, r31
+; CHECK-32-NEXT:    mr r31, r0
+; CHECK-32-NEXT:    blr
+  %a = alloca i32, i32 %n, align 16
+  %b = getelementptr inbounds i32, i32* %a, i64 1198
+  store volatile i32 1, i32* %b
+  %c = load volatile i32, i32* %a
+  ret i32 %c
+}
+
+attributes #0 =  {"probe-stack"="inline-asm"}