[PATCH] [Xtensa] Implement FrameLowering methods and stack operation lowering. (#92960)

Implement emitPrologue/emitEpilogue methods, determine/spill/restore
callee saved registers functionality with test. Also implement lowering
of the DYNAMIC_STACKALLOC/STACKSAVE/STACKRESTORE stack operations with
tests.

10 files changed, 404 insertions, 14 deletions

diff --git a/llvm/lib/Target/Xtensa/XtensaFrameLowering.cpp b/llvm/lib/Target/Xtensa/XtensaFrameLowering.cpp
index ab37c09..71b1bb6 100644
--- a/llvm/lib/Target/Xtensa/XtensaFrameLowering.cpp
+++ b/llvm/lib/Target/Xtensa/XtensaFrameLowering.cpp
@@ -22,9 +22,10 @@
 
 using namespace llvm;
 
-XtensaFrameLowering::XtensaFrameLowering()
+XtensaFrameLowering::XtensaFrameLowering(const XtensaSubtarget &STI)
     : TargetFrameLowering(TargetFrameLowering::StackGrowsDown, Align(4), 0,
-                          Align(4)) {}
+                          Align(4)),
+      STI(STI), TII(*STI.getInstrInfo()), TRI(STI.getRegisterInfo()) {}
 
 bool XtensaFrameLowering::hasFP(const MachineFunction &MF) const {
   const MachineFrameInfo &MFI = MF.getFrameInfo();
@@ -33,10 +34,196 @@ bool XtensaFrameLowering::hasFP(const MachineFunction &MF) const {
 }
 
 void XtensaFrameLowering::emitPrologue(MachineFunction &MF,
-                                       MachineBasicBlock &MBB) const {}
+                                       MachineBasicBlock &MBB) const {
+  assert(&MBB == &MF.front() && "Shrink-wrapping not yet implemented");
+  MachineFrameInfo &MFI = MF.getFrameInfo();
+  MachineBasicBlock::iterator MBBI = MBB.begin();
+  DebugLoc DL = MBBI != MBB.end() ? MBBI->getDebugLoc() : DebugLoc();
+  MCRegister SP = Xtensa::SP;
+  MCRegister FP = TRI->getFrameRegister(MF);
+  MachineModuleInfo &MMI = MF.getMMI();
+  const MCRegisterInfo *MRI = MMI.getContext().getRegisterInfo();
+
+  // First, compute final stack size.
+  uint64_t StackSize = MFI.getStackSize();
+  uint64_t PrevStackSize = StackSize;
+
+  // Round up StackSize to 16*N
+  StackSize += (16 - StackSize) & 0xf;
+
+  // No need to allocate space on the stack.
+  if (StackSize == 0 && !MFI.adjustsStack())
+    return;
+
+  // Adjust stack.
+  TII.adjustStackPtr(SP, -StackSize, MBB, MBBI);
+
+  // emit ".cfi_def_cfa_offset StackSize"
+  unsigned CFIIndex =
+      MF.addFrameInst(MCCFIInstruction::cfiDefCfaOffset(nullptr, StackSize));
+  BuildMI(MBB, MBBI, DL, TII.get(TargetOpcode::CFI_INSTRUCTION))
+      .addCFIIndex(CFIIndex);
+
+  const std::vector<CalleeSavedInfo> &CSI = MFI.getCalleeSavedInfo();
+
+  if (!CSI.empty()) {
+    // Find the instruction past the last instruction that saves a
+    // callee-saved register to the stack. The callee-saved store
+    // instructions are placed at the begin of basic block, so
+    //  iterate over instruction sequence and check that
+    // save instructions are placed correctly.
+    for (unsigned i = 0, e = CSI.size(); i < e; ++i) {
+#ifndef NDEBUG
+      const CalleeSavedInfo &Info = CSI[i];
+      int FI = Info.getFrameIdx();
+      int StoreFI = 0;
+
+      // Checking that the instruction is exactly as expected
+      bool IsStoreInst = false;
+      if (MBBI->getOpcode() == TargetOpcode::COPY && Info.isSpilledToReg()) {
+        Register DstReg = MBBI->getOperand(0).getReg();
+        Register Reg = MBBI->getOperand(1).getReg();
+        IsStoreInst = (Info.getDstReg() == DstReg) && (Info.getReg() == Reg);
+      } else {
+        Register Reg = TII.isStoreToStackSlot(*MBBI, StoreFI);
+        IsStoreInst = (Reg == Info.getReg()) && (StoreFI == FI);
+      }
+      assert(IsStoreInst &&
+             "Unexpected callee-saved register store instruction");
+#endif
+      ++MBBI;
+    }
+
+    // Iterate over list of callee-saved registers and emit .cfi_offset
+    // directives.
+    for (const auto &I : CSI) {
+      int64_t Offset = MFI.getObjectOffset(I.getFrameIdx());
+      Register Reg = I.getReg();
+
+      unsigned CFIIndex = MF.addFrameInst(MCCFIInstruction::createOffset(
+          nullptr, MRI->getDwarfRegNum(Reg, 1), Offset));
+      BuildMI(MBB, MBBI, DL, TII.get(TargetOpcode::CFI_INSTRUCTION))
+          .addCFIIndex(CFIIndex);
+    }
+  }
+
+  // if framepointer enabled, set it to point to the stack pointer.
+  if (hasFP(MF)) {
+    // Insert instruction "move $fp, $sp" at this location.
+    BuildMI(MBB, MBBI, DL, TII.get(Xtensa::OR), FP)
+        .addReg(SP)
+        .addReg(SP)
+        .setMIFlag(MachineInstr::FrameSetup);
+
+    // emit ".cfi_def_cfa_register $fp"
+    unsigned CFIIndex = MF.addFrameInst(MCCFIInstruction::createDefCfaRegister(
+        nullptr, MRI->getDwarfRegNum(FP, true)));
+    BuildMI(MBB, MBBI, DL, TII.get(TargetOpcode::CFI_INSTRUCTION))
+        .addCFIIndex(CFIIndex);
+  }
+
+  if (StackSize != PrevStackSize) {
+    MFI.setStackSize(StackSize);
+
+    for (int i = MFI.getObjectIndexBegin(); i < MFI.getObjectIndexEnd(); i++) {
+      if (!MFI.isDeadObjectIndex(i)) {
+        int64_t SPOffset = MFI.getObjectOffset(i);
+
+        if (SPOffset < 0)
+          MFI.setObjectOffset(i, SPOffset - StackSize + PrevStackSize);
+      }
+    }
+  }
+}
 
 void XtensaFrameLowering::emitEpilogue(MachineFunction &MF,
-                                       MachineBasicBlock &MBB) const {}
+                                       MachineBasicBlock &MBB) const {
+  MachineBasicBlock::iterator MBBI = MBB.getLastNonDebugInstr();
+  MachineFrameInfo &MFI = MF.getFrameInfo();
+  DebugLoc DL = MBBI->getDebugLoc();
+  MCRegister SP = Xtensa::SP;
+  MCRegister FP = TRI->getFrameRegister(MF);
+
+  // if framepointer enabled, restore the stack pointer.
+  if (hasFP(MF)) {
+    // We should place restore stack pointer instruction just before
+    // sequence of instructions which restores callee-saved registers.
+    // This sequence is placed at the end of the basic block,
+    // so we should find first instruction of the sequence.
+    MachineBasicBlock::iterator I = MBBI;
+
+    const std::vector<CalleeSavedInfo> &CSI = MFI.getCalleeSavedInfo();
+
+    // Find the first instruction at the end that restores a callee-saved
+    // register.
+    for (unsigned i = 0, e = CSI.size(); i < e; ++i) {
+      --I;
+#ifndef NDEBUG
+      const CalleeSavedInfo &Info = CSI[i];
+      int FI = Info.getFrameIdx();
+      int LoadFI = 0;
+
+      // Checking that the instruction is exactly as expected
+      bool IsRestoreInst = false;
+      if (I->getOpcode() == TargetOpcode::COPY && Info.isSpilledToReg()) {
+        Register Reg = I->getOperand(0).getReg();
+        Register DstReg = I->getOperand(1).getReg();
+        IsRestoreInst = (Info.getDstReg() == DstReg) && (Info.getReg() == Reg);
+      } else {
+        Register Reg = TII.isLoadFromStackSlot(*I, LoadFI);
+        IsRestoreInst = (Info.getReg() == Reg) && (LoadFI == FI);
+      }
+      assert(IsRestoreInst &&
+             "Unexpected callee-saved register restore instruction");
+#endif
+    }
+
+    BuildMI(MBB, I, DL, TII.get(Xtensa::OR), SP).addReg(FP).addReg(FP);
+  }
+
+  // Get the number of bytes from FrameInfo
+  uint64_t StackSize = MFI.getStackSize();
+
+  if (!StackSize)
+    return;
+
+  // Adjust stack.
+  TII.adjustStackPtr(SP, StackSize, MBB, MBBI);
+}
+
+bool XtensaFrameLowering::spillCalleeSavedRegisters(
+    MachineBasicBlock &MBB, MachineBasicBlock::iterator MI,
+    ArrayRef<CalleeSavedInfo> CSI, const TargetRegisterInfo *TRI) const {
+  MachineFunction *MF = MBB.getParent();
+  MachineBasicBlock &EntryBlock = *(MF->begin());
+
+  for (unsigned i = 0, e = CSI.size(); i != e; ++i) {
+    // Add the callee-saved register as live-in. Do not add if the register is
+    // A0 and return address is taken, because it will be implemented in
+    // method XtensaTargetLowering::LowerRETURNADDR.
+    // It's killed at the spill, unless the register is RA and return address
+    // is taken.
+    Register Reg = CSI[i].getReg();
+    bool IsA0AndRetAddrIsTaken =
+        (Reg == Xtensa::A0) && MF->getFrameInfo().isReturnAddressTaken();
+    if (!IsA0AndRetAddrIsTaken)
+      EntryBlock.addLiveIn(Reg);
+
+    // Insert the spill to the stack frame.
+    bool IsKill = !IsA0AndRetAddrIsTaken;
+    const TargetRegisterClass *RC = TRI->getMinimalPhysRegClass(Reg);
+    TII.storeRegToStackSlot(EntryBlock, MI, Reg, IsKill, CSI[i].getFrameIdx(),
+                            RC, TRI, Register());
+  }
+
+  return true;
+}
+
+bool XtensaFrameLowering::restoreCalleeSavedRegisters(
+    MachineBasicBlock &MBB, MachineBasicBlock::iterator MI,
+    MutableArrayRef<CalleeSavedInfo> CSI, const TargetRegisterInfo *TRI) const {
+  return TargetFrameLowering::restoreCalleeSavedRegisters(MBB, MI, CSI, TRI);
+}
 
 // Eliminate ADJCALLSTACKDOWN, ADJCALLSTACKUP pseudo instructions
 MachineBasicBlock::iterator XtensaFrameLowering::eliminateCallFramePseudoInstr(
@@ -51,9 +238,37 @@ MachineBasicBlock::iterator XtensaFrameLowering::eliminateCallFramePseudoInstr(
     if (I->getOpcode() == Xtensa::ADJCALLSTACKDOWN)
       Amount = -Amount;
 
-    unsigned SP = Xtensa::SP;
-    TII.adjustStackPtr(SP, Amount, MBB, I);
+    TII.adjustStackPtr(Xtensa::SP, Amount, MBB, I);
   }
 
   return MBB.erase(I);
 }
+
+void XtensaFrameLowering::determineCalleeSaves(MachineFunction &MF,
+                                               BitVector &SavedRegs,
+                                               RegScavenger *RS) const {
+  unsigned FP = TRI->getFrameRegister(MF);
+
+  TargetFrameLowering::determineCalleeSaves(MF, SavedRegs, RS);
+
+  // Mark $fp as used if function has dedicated frame pointer.
+  if (hasFP(MF))
+    SavedRegs.set(FP);
+}
+
+void XtensaFrameLowering::processFunctionBeforeFrameFinalized(
+    MachineFunction &MF, RegScavenger *RS) const {
+  // Set scavenging frame index if necessary.
+  MachineFrameInfo &MFI = MF.getFrameInfo();
+  uint64_t MaxSPOffset = MFI.estimateStackSize(MF);
+
+  if (isInt<12>(MaxSPOffset))
+    return;
+
+  const TargetRegisterClass &RC = Xtensa::ARRegClass;
+  unsigned Size = TRI->getSpillSize(RC);
+  Align Alignment = TRI->getSpillAlign(RC);
+  int FI = MF.getFrameInfo().CreateStackObject(Size, Alignment, false);
+
+  RS->addScavengingFrameIndex(FI);
+}
diff --git a/llvm/lib/Target/Xtensa/XtensaFrameLowering.h b/llvm/lib/Target/Xtensa/XtensaFrameLowering.h
index 2da88ab1..31c496d 100644
--- a/llvm/lib/Target/Xtensa/XtensaFrameLowering.h
+++ b/llvm/lib/Target/Xtensa/XtensaFrameLowering.h
@@ -14,10 +14,16 @@
 namespace llvm {
 class XtensaTargetMachine;
 class XtensaSubtarget;
+class XtensaInstrInfo;
+class XtensaRegisterInfo;
 
 class XtensaFrameLowering : public TargetFrameLowering {
+  const XtensaSubtarget &STI;
+  const XtensaInstrInfo &TII;
+  const XtensaRegisterInfo *TRI;
+
 public:
-  XtensaFrameLowering();
+  XtensaFrameLowering(const XtensaSubtarget &STI);
 
   bool hasFP(const MachineFunction &MF) const override;
 
@@ -29,6 +35,22 @@ public:
   MachineBasicBlock::iterator
   eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
                                 MachineBasicBlock::iterator I) const override;
+
+  bool spillCalleeSavedRegisters(MachineBasicBlock &MBB,
+                                 MachineBasicBlock::iterator MI,
+                                 ArrayRef<CalleeSavedInfo> CSI,
+                                 const TargetRegisterInfo *TRI) const override;
+  bool
+  restoreCalleeSavedRegisters(MachineBasicBlock &MBB,
+                              MachineBasicBlock::iterator MI,
+                              MutableArrayRef<CalleeSavedInfo> CSI,
+                              const TargetRegisterInfo *TRI) const override;
+
+  void determineCalleeSaves(MachineFunction &MF, BitVector &SavedRegs,
+                            RegScavenger *RS) const override;
+
+  void processFunctionBeforeFrameFinalized(MachineFunction &MF,
+                                           RegScavenger *RS) const override;
 };
 
 } // namespace llvm
diff --git a/llvm/lib/Target/Xtensa/XtensaISelLowering.cpp b/llvm/lib/Target/Xtensa/XtensaISelLowering.cpp
index 64b996b..6c3258b 100644
--- a/llvm/lib/Target/Xtensa/XtensaISelLowering.cpp
+++ b/llvm/lib/Target/Xtensa/XtensaISelLowering.cpp
@@ -78,6 +78,12 @@ XtensaTargetLowering::XtensaTargetLowering(const TargetMachine &TM,
 
   setOperationAction(ISD::ConstantPool, PtrVT, Custom);
 
+  // Implement custom stack allocations
+  setOperationAction(ISD::DYNAMIC_STACKALLOC, PtrVT, Custom);
+  // Implement custom stack save and restore
+  setOperationAction(ISD::STACKSAVE, MVT::Other, Custom);
+  setOperationAction(ISD::STACKRESTORE, MVT::Other, Custom);
+
   // Compute derived properties from the register classes
   computeRegisterProperties(STI.getRegisterInfo());
 }
@@ -534,6 +540,43 @@ SDValue XtensaTargetLowering::LowerConstantPool(ConstantPoolSDNode *CP,
   return getAddrPCRel(Result, DAG);
 }
 
+SDValue XtensaTargetLowering::LowerSTACKSAVE(SDValue Op,
+                                             SelectionDAG &DAG) const {
+  return DAG.getCopyFromReg(Op.getOperand(0), SDLoc(Op), Xtensa::SP,
+                            Op.getValueType());
+}
+
+SDValue XtensaTargetLowering::LowerSTACKRESTORE(SDValue Op,
+                                                SelectionDAG &DAG) const {
+  return DAG.getCopyToReg(Op.getOperand(0), SDLoc(Op), Xtensa::SP,
+                          Op.getOperand(1));
+}
+
+SDValue XtensaTargetLowering::LowerDYNAMIC_STACKALLOC(SDValue Op,
+                                                      SelectionDAG &DAG) const {
+  SDValue Chain = Op.getOperand(0); // Legalize the chain.
+  SDValue Size = Op.getOperand(1);  // Legalize the size.
+  EVT VT = Size->getValueType(0);
+  SDLoc DL(Op);
+
+  // Round up Size to 32
+  SDValue SizeTmp =
+      DAG.getNode(ISD::ADD, DL, VT, Size, DAG.getConstant(31, DL, MVT::i32));
+  SDValue SizeRoundUp = DAG.getNode(ISD::AND, DL, VT, SizeTmp,
+                                    DAG.getConstant(~31, DL, MVT::i32));
+
+  unsigned SPReg = Xtensa::SP;
+  SDValue SP = DAG.getCopyFromReg(Chain, DL, SPReg, VT);
+  SDValue NewSP = DAG.getNode(ISD::SUB, DL, VT, SP, SizeRoundUp); // Value
+  Chain = DAG.getCopyToReg(SP.getValue(1), DL, SPReg, NewSP); // Output chain
+
+  SDValue NewVal = DAG.getCopyFromReg(Chain, DL, SPReg, MVT::i32);
+  Chain = NewVal.getValue(1);
+
+  SDValue Ops[2] = {NewVal, Chain};
+  return DAG.getMergeValues(Ops, DL);
+}
+
 SDValue XtensaTargetLowering::LowerOperation(SDValue Op,
                                              SelectionDAG &DAG) const {
   switch (Op.getOpcode()) {
@@ -541,6 +584,12 @@ SDValue XtensaTargetLowering::LowerOperation(SDValue Op,
     return LowerImmediate(Op, DAG);
   case ISD::ConstantPool:
     return LowerConstantPool(cast<ConstantPoolSDNode>(Op), DAG);
+  case ISD::STACKSAVE:
+    return LowerSTACKSAVE(Op, DAG);
+  case ISD::STACKRESTORE:
+    return LowerSTACKRESTORE(Op, DAG);
+  case ISD::DYNAMIC_STACKALLOC:
+    return LowerDYNAMIC_STACKALLOC(Op, DAG);
   default:
     report_fatal_error("Unexpected node to lower");
   }
diff --git a/llvm/lib/Target/Xtensa/XtensaISelLowering.h b/llvm/lib/Target/Xtensa/XtensaISelLowering.h
index 077559e..6f6ec39 100644
--- a/llvm/lib/Target/Xtensa/XtensaISelLowering.h
+++ b/llvm/lib/Target/Xtensa/XtensaISelLowering.h
@@ -75,6 +75,12 @@ private:
 
   SDValue LowerConstantPool(ConstantPoolSDNode *CP, SelectionDAG &DAG) const;
 
+  SDValue LowerDYNAMIC_STACKALLOC(SDValue Op, SelectionDAG &DAG) const;
+
+  SDValue LowerSTACKSAVE(SDValue Op, SelectionDAG &DAG) const;
+
+  SDValue LowerSTACKRESTORE(SDValue Op, SelectionDAG &DAG) const;
+
   SDValue getAddrPCRel(SDValue Op, SelectionDAG &DAG) const;
 
   CCAssignFn *CCAssignFnForCall(CallingConv::ID CC, bool IsVarArg) const;
diff --git a/llvm/lib/Target/Xtensa/XtensaInstrInfo.cpp b/llvm/lib/Target/Xtensa/XtensaInstrInfo.cpp
index 26d8727..27e1985 100644
--- a/llvm/lib/Target/Xtensa/XtensaInstrInfo.cpp
+++ b/llvm/lib/Target/Xtensa/XtensaInstrInfo.cpp
@@ -48,6 +48,30 @@ XtensaInstrInfo::XtensaInstrInfo(const XtensaSubtarget &STI)
     : XtensaGenInstrInfo(Xtensa::ADJCALLSTACKDOWN, Xtensa::ADJCALLSTACKUP),
       RI(STI), STI(STI) {}
 
+Register XtensaInstrInfo::isLoadFromStackSlot(const MachineInstr &MI,
+                                              int &FrameIndex) const {
+  if (MI.getOpcode() == Xtensa::L32I) {
+    if (MI.getOperand(1).isFI() && MI.getOperand(2).isImm() &&
+        MI.getOperand(2).getImm() == 0) {
+      FrameIndex = MI.getOperand(1).getIndex();
+      return MI.getOperand(0).getReg();
+    }
+  }
+  return Register();
+}
+
+Register XtensaInstrInfo::isStoreToStackSlot(const MachineInstr &MI,
+                                             int &FrameIndex) const {
+  if (MI.getOpcode() == Xtensa::S32I) {
+    if (MI.getOperand(1).isFI() && MI.getOperand(2).isImm() &&
+        MI.getOperand(2).getImm() == 0) {
+      FrameIndex = MI.getOperand(1).getIndex();
+      return MI.getOperand(0).getReg();
+    }
+  }
+  return Register();
+}
+
 /// Adjust SP by Amount bytes.
 void XtensaInstrInfo::adjustStackPtr(unsigned SP, int64_t Amount,
                                      MachineBasicBlock &MBB,
diff --git a/llvm/lib/Target/Xtensa/XtensaInstrInfo.h b/llvm/lib/Target/Xtensa/XtensaInstrInfo.h
index 1acd314..37f157f 100644
--- a/llvm/lib/Target/Xtensa/XtensaInstrInfo.h
+++ b/llvm/lib/Target/Xtensa/XtensaInstrInfo.h
@@ -41,6 +41,12 @@ public:
   // Return the XtensaRegisterInfo, which this class owns.
   const XtensaRegisterInfo &getRegisterInfo() const { return RI; }
 
+  Register isLoadFromStackSlot(const MachineInstr &MI,
+                               int &FrameIndex) const override;
+
+  Register isStoreToStackSlot(const MachineInstr &MI,
+                              int &FrameIndex) const override;
+
   void copyPhysReg(MachineBasicBlock &MBB, MachineBasicBlock::iterator MBBI,
                    const DebugLoc &DL, MCRegister DestReg, MCRegister SrcReg,
                    bool KillSrc) const override;
diff --git a/llvm/lib/Target/Xtensa/XtensaSubtarget.cpp b/llvm/lib/Target/Xtensa/XtensaSubtarget.cpp
index 10ab92d..d6b1b4b 100644
--- a/llvm/lib/Target/Xtensa/XtensaSubtarget.cpp
+++ b/llvm/lib/Target/Xtensa/XtensaSubtarget.cpp
@@ -41,4 +41,4 @@ XtensaSubtarget::XtensaSubtarget(const Triple &TT, StringRef CPU, StringRef FS,
                                  const TargetMachine &TM)
     : XtensaGenSubtargetInfo(TT, CPU, /*TuneCPU=*/CPU, FS), TargetTriple(TT),
       InstrInfo(initializeSubtargetDependencies(CPU, FS)), TLInfo(TM, *this),
-      TSInfo(), FrameLowering() {}
+      TSInfo(), FrameLowering(*this) {}
diff --git a/llvm/test/CodeGen/Xtensa/call.ll b/llvm/test/CodeGen/Xtensa/call.ll
index 24c7c4f..921f89a 100644
--- a/llvm/test/CodeGen/Xtensa/call.ll
+++ b/llvm/test/CodeGen/Xtensa/call.ll
@@ -5,11 +5,11 @@ declare i32 @external_function(i32)
 define i32 @test_call_external(i32 %a) nounwind {
 ; CHECK-LABEL: test_call_external:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:  s32i  a0, a1, 0
+; CHECK:       s32i  a0, a1, 0
 ; CHECK-NEXT:  l32r  a8, .LCPI0_0
 ; CHECK-NEXT:  callx0  a8
 ; CHECK-NEXT:  l32i  a0, a1, 0
-; CHECK-NEXT:  ret
+; CHECK:       ret
   %1 = call i32 @external_function(i32 %a)
   ret i32 %1
 }
@@ -26,11 +26,11 @@ define i32 @defined_function(i32 %a) nounwind {
 define i32 @test_call_defined(i32 %a) nounwind {
 ; CHECK-LABEL: test_call_defined:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:  s32i  a0, a1, 0
+; CHECK:       s32i  a0, a1, 0
 ; CHECK-NEXT:  l32r  a8, .LCPI2_0
 ; CHECK-NEXT:  callx0  a8
 ; CHECK-NEXT:  l32i  a0, a1, 0
-; CHECK-NEXT:  ret
+; CHECK:       ret
   %1 = call i32 @defined_function(i32 %a) nounwind
   ret i32 %1
 }
@@ -38,12 +38,12 @@ define i32 @test_call_defined(i32 %a) nounwind {
 define i32 @test_call_indirect(ptr %a, i32 %b) nounwind {
 ; CHECK-LABEL: test_call_indirect:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:  s32i  a0, a1, 0
+; CHECK:       s32i  a0, a1, 0
 ; CHECK-NEXT:  or  a8, a2, a2
 ; CHECK-NEXT:  or  a2, a3, a3
 ; CHECK-NEXT:  callx0  a8
 ; CHECK-NEXT:  l32i  a0, a1, 0
-; CHECK-NEXT:  ret
+; CHECK:       ret
   %1 = call i32 %a(i32 %b)
   ret i32 %1
 }
diff --git a/llvm/test/CodeGen/Xtensa/dynamic-alloc.ll b/llvm/test/CodeGen/Xtensa/dynamic-alloc.ll
new file mode 100644
index 0000000..2b28f35
--- /dev/null
+++ b/llvm/test/CodeGen/Xtensa/dynamic-alloc.ll
@@ -0,0 +1,26 @@
+; RUN: llc -mtriple=xtensa -disable-block-placement -verify-machineinstrs < %s \
+; RUN:   | FileCheck %s
+
+define ptr @test_simple_alloca(i32 %numelts) {
+; CHECK-LABEL: test_simple_alloca
+; CHECK:       addi  a8, a1, -16
+; CHECK:       or  a1, a8, a8
+; CHECK:       s32i  a15, a1, 0
+; CHECK:       or a15, a1, a1
+; CHECK:       addi  a8, a2, 3
+; CHECK-NEXT:  movi  a9, -4
+; CHECK-NEXT:  and  a8, a8, a9
+; CHECK-NEXT:  addi  a8, a8, 31
+; CHECK-NEXT:  movi  a9, -32
+; CHECK-NEXT:  and  a8, a8, a9
+; CHECK-NEXT:  sub  a1, a1, a8
+; CHECK-NEXT:  or  a2, a1, a1
+; CHECK-NEXT:  or  a1, a15, a15
+; CHECK-NEXT:  l32i  a15, a1, 0
+; CHECK-NEXT:  addi  a8, a1, 16
+; CHECK-NEXT:  or  a1, a8, a8
+; CHECK-NEXT:  ret
+
+  %addr = alloca i8, i32 %numelts
+  ret ptr %addr
+}
diff --git a/llvm/test/CodeGen/Xtensa/saverestore.ll b/llvm/test/CodeGen/Xtensa/saverestore.ll
new file mode 100644
index 0000000..69c8b16
--- /dev/null
+++ b/llvm/test/CodeGen/Xtensa/saverestore.ll
@@ -0,0 +1,42 @@
+; RUN: llc --mtriple=xtensa < %s | FileCheck %s
+
+declare ptr @llvm.stacksave()
+
+declare void @llvm.stackrestore(ptr)
+
+declare void @use_addr(ptr)
+
+define void @test_saverestore(i64 %n) {
+; CHECK:       # %bb.0:
+; CHECK-NEXT:  addi a8, a1, -16
+; CHECK-NEXT:  or a1, a8, a8
+; CHECK:       s32i a0, a1, 8
+; CHECK-NEXT:  s32i a12, a1, 4
+; CHECK-NEXT:  s32i a15, a1, 0
+; CHECK:       or a15, a1, a1
+; CHECK:       addi a8, a2, 3
+; CHECK-NEXT:  movi a9, -4
+; CHECK-NEXT:  and a8, a8, a9
+; CHECK-NEXT:  addi a8, a8, 31
+; CHECK-NEXT:  movi a9, -32
+; CHECK-NEXT:  and a8, a8, a9
+; CHECK-NEXT:  or a12, a1, a1
+; CHECK-NEXT:  sub a1, a1, a8
+; CHECK-NEXT:  or a2, a1, a1
+; CHECK-NEXT:  l32r a8, .LCPI0_0
+; CHECK-NEXT:  callx0 a8
+; CHECK-NEXT:  or a1, a12, a12
+; CHECK-NEXT:  or a1, a15, a15
+; CHECK-NEXT:  l32i a15, a1, 0
+; CHECK-NEXT:  l32i a12, a1, 4
+; CHECK-NEXT:  l32i a0, a1, 8
+; CHECK-NEXT:  addi a8, a1, 16
+; CHECK-NEXT:  or a1, a8, a8
+; CHECK-NEXT:  ret
+
+  %sp = call ptr @llvm.stacksave.p0()
+  %addr = alloca i8, i64 %n
+  call void @use_addr(ptr %addr)
+  call void @llvm.stackrestore.p0(ptr %sp)
+  ret void
+}