diff --git a/llvm/lib/Target/Hexagon/HexagonFrameLowering.h b/llvm/lib/Target/Hexagon/HexagonFrameLowering.h
--- a/llvm/lib/Target/Hexagon/HexagonFrameLowering.h
+++ b/llvm/lib/Target/Hexagon/HexagonFrameLowering.h
@@ -29,6 +29,8 @@
 
 class HexagonFrameLowering : public TargetFrameLowering {
 public:
+  // First register which could possibly hold a variable argument.
+  int FirstVarArgSavedReg;
   explicit HexagonFrameLowering()
       : TargetFrameLowering(StackGrowsDown, Align(8), 0, Align::None(), true) {}
 
diff --git a/llvm/lib/Target/Hexagon/HexagonFrameLowering.cpp b/llvm/lib/Target/Hexagon/HexagonFrameLowering.cpp
--- a/llvm/lib/Target/Hexagon/HexagonFrameLowering.cpp
+++ b/llvm/lib/Target/Hexagon/HexagonFrameLowering.cpp
@@ -395,6 +395,9 @@
       MachineBasicBlock *&PrologB, MachineBasicBlock *&EpilogB) const {
   static unsigned ShrinkCounter = 0;
 
+  if (MF.getSubtarget<HexagonSubtarget>().isEnvironmentMusl() &&
+      MF.getFunction().isVarArg())
+    return;
   if (ShrinkLimit.getPosition()) {
     if (ShrinkCounter >= ShrinkLimit)
       return;
@@ -622,6 +625,118 @@
 
   DebugLoc dl = MBB.findDebugLoc(InsertPt);
 
+  if (MF.getFunction().isVarArg() &&
+      MF.getSubtarget<HexagonSubtarget>().isEnvironmentMusl()) {
+    // Calculate the size of register saved area.
+    int NumVarArgRegs = 6 - FirstVarArgSavedReg;
+    int RegisterSavedAreaSizePlusPadding = (NumVarArgRegs % 2 == 0)
+                                              ? NumVarArgRegs * 4
+                                              : NumVarArgRegs * 4 + 4;
+    if (RegisterSavedAreaSizePlusPadding > 0) {
+      // Decrement the stack pointer by size of register saved area plus
+      // padding if any.
+      BuildMI(MBB, InsertPt, dl, HII.get(Hexagon::A2_addi), SP)
+        .addReg(SP)
+        .addImm(-RegisterSavedAreaSizePlusPadding)
+        .setMIFlag(MachineInstr::FrameSetup);
+
+      int NumBytes = 0;
+      // Copy all the named arguments below register saved area.
+      auto &HMFI = *MF.getInfo<HexagonMachineFunctionInfo>();
+      for (int i = HMFI.getFirstNamedArgFrameIndex(),
+               e = HMFI.getLastNamedArgFrameIndex(); i >= e; --i) {
+        int ObjSize = MFI.getObjectSize(i);
+        int ObjAlign = MFI.getObjectAlignment(i);
+
+        // Determine the kind of load/store that should be used.
+        unsigned LDOpc, STOpc;
+        int OpcodeChecker = ObjAlign;
+
+        // Handle cases where alignment of an object is > its size.
+        if (ObjSize < ObjAlign) {
+          if (ObjSize <= 1)
+            OpcodeChecker = 1;
+          else if (ObjSize <= 2)
+            OpcodeChecker = 2;
+          else if (ObjSize <= 4)
+            OpcodeChecker = 4;
+          else if (ObjSize > 4)
+            OpcodeChecker = 8;
+        }
+
+        switch (OpcodeChecker) {
+          case 1:
+            LDOpc = Hexagon::L2_loadrb_io;
+            STOpc = Hexagon::S2_storerb_io;
+            break;
+          case 2:
+            LDOpc = Hexagon::L2_loadrh_io;
+            STOpc = Hexagon::S2_storerh_io;
+            break;
+          case 4:
+            LDOpc = Hexagon::L2_loadri_io;
+            STOpc = Hexagon::S2_storeri_io;
+            break;
+          case 8:
+          default:
+            LDOpc = Hexagon::L2_loadrd_io;
+            STOpc = Hexagon::S2_storerd_io;
+            break;
+        }
+
+        unsigned RegUsed = LDOpc == Hexagon::L2_loadrd_io ? Hexagon::D3
+                                                          : Hexagon::R6;
+        int LoadStoreCount = ObjSize / OpcodeChecker;
+
+        if (ObjSize % OpcodeChecker)
+          ++LoadStoreCount;
+
+        // Get the start location of the load. NumBytes is basically the
+        // offset from the stack pointer of previous function, which would be
+        // the caller in this case, as this function has variable argument
+        // list.
+        if (NumBytes != 0)
+          NumBytes = alignTo(NumBytes, ObjAlign);
+
+        int Count = 0;
+        while (Count < LoadStoreCount) {
+          // Load the value of the named argument on stack.
+          BuildMI(MBB, InsertPt, dl, HII.get(LDOpc), RegUsed)
+            .addReg(SP)
+            .addImm(RegisterSavedAreaSizePlusPadding +
+                    ObjAlign * Count + NumBytes)
+            .setMIFlag(MachineInstr::FrameSetup);
+
+          // Store it below the register saved area plus padding.
+          BuildMI(MBB, InsertPt, dl, HII.get(STOpc))
+            .addReg(SP)
+            .addImm(ObjAlign * Count + NumBytes)
+            .addReg(RegUsed)
+            .setMIFlag(MachineInstr::FrameSetup);
+
+          Count++;
+        }
+        NumBytes += MFI.getObjectSize(i);
+      }
+
+      // Make NumBytes 8 byte aligned
+      NumBytes = alignTo(NumBytes, 8);
+
+      // If the number of registers having variable arguments is odd,
+      // leave 4 bytes of padding to get to the location where first
+      // variable argument which was passed through register was copied.
+      NumBytes = (NumVarArgRegs % 2 == 0) ? NumBytes : NumBytes + 4;
+
+      for (int j = FirstVarArgSavedReg, i = 0; j < 6; ++j, ++i) {
+        BuildMI(MBB, InsertPt, dl, HII.get(Hexagon::S2_storeri_io))
+          .addReg(SP)
+          .addImm(NumBytes + 4 * i)
+          .addReg(Hexagon::R0 + j)
+          .setMIFlag(MachineInstr::FrameSetup);
+      }
+    }
+  }
+
   if (hasFP(MF)) {
     insertAllocframe(MBB, InsertPt, NumBytes);
     if (AlignStack) {
@@ -655,7 +770,16 @@
 
   if (!hasFP(MF)) {
     MachineFrameInfo &MFI = MF.getFrameInfo();
-    if (unsigned NumBytes = MFI.getStackSize()) {
+    unsigned NumBytes = MFI.getStackSize();
+    if (MF.getFunction().isVarArg() &&
+        MF.getSubtarget<HexagonSubtarget>().isEnvironmentMusl()) {
+      // On Hexagon Linux, deallocate the stack for the register saved area.
+      int NumVarArgRegs = 6 - FirstVarArgSavedReg;
+      int RegisterSavedAreaSizePlusPadding = (NumVarArgRegs % 2 == 0) ?
+        (NumVarArgRegs * 4) : (NumVarArgRegs * 4 + 4);
+      NumBytes += RegisterSavedAreaSizePlusPadding;
+    }
+    if (NumBytes) {
       BuildMI(MBB, InsertPt, dl, HII.get(Hexagon::A2_addi), SP)
         .addReg(SP)
         .addImm(NumBytes);
@@ -710,24 +834,49 @@
       NeedsDeallocframe = false;
   }
 
-  if (!NeedsDeallocframe)
-    return;
-  // If the returning instruction is PS_jmpret, replace it with dealloc_return,
-  // otherwise just add deallocframe. The function could be returning via a
-  // tail call.
-  if (RetOpc != Hexagon::PS_jmpret || DisableDeallocRet) {
-    BuildMI(MBB, InsertPt, dl, HII.get(Hexagon::L2_deallocframe))
+  if (!MF.getSubtarget<HexagonSubtarget>().isEnvironmentMusl() ||
+      !MF.getFunction().isVarArg()) {
+    if (!NeedsDeallocframe)
+      return;
+    // If the returning instruction is PS_jmpret, replace it with
+    // dealloc_return, otherwise just add deallocframe. The function
+    // could be returning via a tail call.
+    if (RetOpc != Hexagon::PS_jmpret || DisableDeallocRet) {
+      BuildMI(MBB, InsertPt, dl, HII.get(Hexagon::L2_deallocframe))
       .addDef(Hexagon::D15)
       .addReg(Hexagon::R30);
-    return;
-  }
-  unsigned NewOpc = Hexagon::L4_return;
-  MachineInstr *NewI = BuildMI(MBB, RetI, dl, HII.get(NewOpc))
+      return;
+    }
+    unsigned NewOpc = Hexagon::L4_return;
+    MachineInstr *NewI = BuildMI(MBB, RetI, dl, HII.get(NewOpc))
       .addDef(Hexagon::D15)
       .addReg(Hexagon::R30);
-  // Transfer the function live-out registers.
-  NewI->copyImplicitOps(MF, *RetI);
-  MBB.erase(RetI);
+    // Transfer the function live-out registers.
+    NewI->copyImplicitOps(MF, *RetI);
+    MBB.erase(RetI);
+  } else {
+    // L2_deallocframe instruction after it.
+    // Calculate the size of register saved area.
+    int NumVarArgRegs = 6 - FirstVarArgSavedReg;
+    int RegisterSavedAreaSizePlusPadding = (NumVarArgRegs % 2 == 0) ?
+      (NumVarArgRegs * 4) : (NumVarArgRegs * 4 + 4);
+
+    MachineBasicBlock::iterator Term = MBB.getFirstTerminator();
+    MachineBasicBlock::iterator I = (Term == MBB.begin()) ? MBB.end()
+                                                          : std::prev(Term);
+    if (I == MBB.end() ||
+       (I->getOpcode() != Hexagon::RESTORE_DEALLOC_BEFORE_TAILCALL_V4_EXT &&
+        I->getOpcode() != Hexagon::RESTORE_DEALLOC_BEFORE_TAILCALL_V4_EXT_PIC &&
+        I->getOpcode() != Hexagon::RESTORE_DEALLOC_BEFORE_TAILCALL_V4 &&
+        I->getOpcode() != Hexagon::RESTORE_DEALLOC_BEFORE_TAILCALL_V4_PIC))
+      BuildMI(MBB, InsertPt, dl, HII.get(Hexagon::L2_deallocframe))
+        .addDef(Hexagon::D15)
+        .addReg(Hexagon::R30);
+    if (RegisterSavedAreaSizePlusPadding != 0)
+      BuildMI(MBB, InsertPt, dl, HII.get(Hexagon::A2_addi), SP)
+        .addReg(SP)
+        .addImm(RegisterSavedAreaSizePlusPadding);
+  }
 }
 
 void HexagonFrameLowering::insertAllocframe(MachineBasicBlock &MBB,
@@ -2473,6 +2622,8 @@
 /// checks are performed, which may still lead to the inline code.
 bool HexagonFrameLowering::shouldInlineCSR(const MachineFunction &MF,
       const CSIVect &CSI) const {
+  if (MF.getSubtarget<HexagonSubtarget>().isEnvironmentMusl())
+    return true;
   if (MF.getInfo<HexagonMachineFunctionInfo>()->hasEHReturn())
     return true;
   if (!hasFP(MF))
diff --git a/llvm/lib/Target/Hexagon/HexagonISelLowering.h b/llvm/lib/Target/Hexagon/HexagonISelLowering.h
--- a/llvm/lib/Target/Hexagon/HexagonISelLowering.h
+++ b/llvm/lib/Target/Hexagon/HexagonISelLowering.h
@@ -248,6 +248,7 @@
     }
 
     SDValue LowerVASTART(SDValue Op, SelectionDAG &DAG) const;
+    SDValue LowerVACOPY(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerConstantPool(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerJumpTable(SDValue Op, SelectionDAG &DAG) const;
 
diff --git a/llvm/lib/Target/Hexagon/HexagonISelLowering.cpp b/llvm/lib/Target/Hexagon/HexagonISelLowering.cpp
--- a/llvm/lib/Target/Hexagon/HexagonISelLowering.cpp
+++ b/llvm/lib/Target/Hexagon/HexagonISelLowering.cpp
@@ -393,9 +393,12 @@
   if (GlobalAddressSDNode *GAN = dyn_cast<GlobalAddressSDNode>(Callee))
     Callee = DAG.getTargetGlobalAddress(GAN->getGlobal(), dl, MVT::i32);
 
+  // Linux ABI treats var-arg calls the same way as regular ones.
+  bool TreatAsVarArg = !Subtarget.isEnvironmentMusl() && IsVarArg;
+
   // Analyze operands of the call, assigning locations to each operand.
   SmallVector<CCValAssign, 16> ArgLocs;
-  HexagonCCState CCInfo(CallConv, IsVarArg, MF, ArgLocs, *DAG.getContext(),
+  HexagonCCState CCInfo(CallConv, TreatAsVarArg, MF, ArgLocs, *DAG.getContext(),
                         NumParams);
 
   if (Subtarget.useHVXOps())
@@ -750,9 +753,13 @@
   MachineFrameInfo &MFI = MF.getFrameInfo();
   MachineRegisterInfo &MRI = MF.getRegInfo();
 
+  // Linux ABI treats var-arg calls the same way as regular ones.
+  bool TreatAsVarArg = !Subtarget.isEnvironmentMusl() && IsVarArg;
+
   // Assign locations to all of the incoming arguments.
   SmallVector<CCValAssign, 16> ArgLocs;
-  HexagonCCState CCInfo(CallConv, IsVarArg, MF, ArgLocs, *DAG.getContext(),
+  HexagonCCState CCInfo(CallConv, TreatAsVarArg, MF, ArgLocs,
+                        *DAG.getContext(),
                         MF.getFunction().getFunctionType()->getNumParams());
 
   if (Subtarget.useHVXOps())
@@ -766,8 +773,24 @@
   // caller's stack is passed only when the struct size is smaller than (and
   // equal to) 8 bytes. If not, no address will be passed into callee and
   // callee return the result direclty through R0/R1.
+  auto NextSingleReg = [] (const TargetRegisterClass &RC, unsigned Reg) {
+    switch (RC.getID()) {
+    case Hexagon::IntRegsRegClassID:
+      return Reg - Hexagon::R0 + 1;
+    case Hexagon::DoubleRegsRegClassID:
+      return (Reg - Hexagon::D0 + 1) * 2;
+    case Hexagon::HvxVRRegClassID:
+      return Reg - Hexagon::V0 + 1;
+    case Hexagon::HvxWRRegClassID:
+      return (Reg - Hexagon::W0 + 1) * 2;
+    }
+    llvm_unreachable("Unexpected register class");
+  };
 
+  auto &HFL = const_cast<HexagonFrameLowering&>(*Subtarget.getFrameLowering());
   auto &HMFI = *MF.getInfo<HexagonMachineFunctionInfo>();
+  HFL.FirstVarArgSavedReg = 0;
+  HMFI.setFirstNamedArgFrameIndex(-int(MFI.getNumFixedObjects()));
 
   for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
     CCValAssign &VA = ArgLocs[i];
@@ -811,6 +834,7 @@
       }
       InVals.push_back(Copy);
       MRI.addLiveIn(VA.getLocReg(), VReg);
+      HFL.FirstVarArgSavedReg = NextSingleReg(*RC, VA.getLocReg());
     } else {
       assert(VA.isMemLoc() && "Argument should be passed in memory");
 
@@ -838,8 +862,48 @@
     }
   }
 
+  if (IsVarArg && Subtarget.isEnvironmentMusl()) {
+    for (int i = HFL.FirstVarArgSavedReg; i < 6; i++)
+      MRI.addLiveIn(Hexagon::R0+i);
+  }
+
+  if (IsVarArg && Subtarget.isEnvironmentMusl()) {
+    HMFI.setFirstNamedArgFrameIndex(HMFI.getFirstNamedArgFrameIndex() - 1);
+    HMFI.setLastNamedArgFrameIndex(-int(MFI.getNumFixedObjects()));
+
+    // Create Frame index for the start of register saved area.
+    int NumVarArgRegs = 6 - HFL.FirstVarArgSavedReg;
+    bool RequiresPadding = (NumVarArgRegs & 1);
+    int RegSaveAreaSizePlusPadding = RequiresPadding
+                                        ? (NumVarArgRegs + 1) * 4
+                                        : NumVarArgRegs * 4;
+
+    if (RegSaveAreaSizePlusPadding > 0) {
+      // The offset to saved register area should be 8 byte aligned.
+      int RegAreaStart = HEXAGON_LRFP_SIZE + CCInfo.getNextStackOffset();
+      if (!(RegAreaStart % 8))
+        RegAreaStart = (RegAreaStart + 7) & -8;
 
-  if (IsVarArg) {
+      int RegSaveAreaFrameIndex =
+        MFI.CreateFixedObject(RegSaveAreaSizePlusPadding, RegAreaStart, true);
+      HMFI.setRegSavedAreaStartFrameIndex(RegSaveAreaFrameIndex);
+
+      // This will point to the next argument passed via stack.
+      int Offset = RegAreaStart + RegSaveAreaSizePlusPadding;
+      int FI = MFI.CreateFixedObject(Hexagon_PointerSize, Offset, true);
+      HMFI.setVarArgsFrameIndex(FI);
+    } else {
+      // This will point to the next argument passed via stack, when
+      // there is no saved register area.
+      int Offset = HEXAGON_LRFP_SIZE + CCInfo.getNextStackOffset();
+      int FI = MFI.CreateFixedObject(Hexagon_PointerSize, Offset, true);
+      HMFI.setRegSavedAreaStartFrameIndex(FI);
+      HMFI.setVarArgsFrameIndex(FI);
+    }
+  }
+
+
+  if (IsVarArg && !Subtarget.isEnvironmentMusl()) {
     // This will point to the next argument passed via stack.
     int Offset = HEXAGON_LRFP_SIZE + CCInfo.getNextStackOffset();
     int FI = MFI.CreateFixedObject(Hexagon_PointerSize, Offset, true);
@@ -857,8 +921,82 @@
   HexagonMachineFunctionInfo *QFI = MF.getInfo<HexagonMachineFunctionInfo>();
   SDValue Addr = DAG.getFrameIndex(QFI->getVarArgsFrameIndex(), MVT::i32);
   const Value *SV = cast<SrcValueSDNode>(Op.getOperand(2))->getValue();
-  return DAG.getStore(Op.getOperand(0), SDLoc(Op), Addr, Op.getOperand(1),
-                      MachinePointerInfo(SV));
+
+  if (!Subtarget.isEnvironmentMusl()) {
+    return DAG.getStore(Op.getOperand(0), SDLoc(Op), Addr, Op.getOperand(1),
+                        MachinePointerInfo(SV));
+  }
+  auto &FuncInfo = *MF.getInfo<HexagonMachineFunctionInfo>();
+  auto &HFL = *Subtarget.getFrameLowering();
+  SDLoc DL(Op);
+  SmallVector<SDValue, 8> MemOps;
+
+  // Get frame index of va_list.
+  SDValue FIN = Op.getOperand(1);
+
+  // If first Vararg register is odd, add 4 bytes to start of
+  // saved register area to point to the first register location.
+  // This is because the saved register area has to be 8 byte aligned.
+  // Incase of an odd start register, there will be 4 bytes of padding in
+  // the beginning of saved register area. If all registers area used up,
+  // the following condition will handle it correctly.
+  SDValue SavedRegAreaStartFrameIndex =
+    DAG.getFrameIndex(FuncInfo.getRegSavedAreaStartFrameIndex(), MVT::i32);
+
+  auto PtrVT = getPointerTy(DAG.getDataLayout());
+
+  if (HFL.FirstVarArgSavedReg & 1)
+    SavedRegAreaStartFrameIndex =
+      DAG.getNode(ISD::ADD, DL, PtrVT,
+                  DAG.getFrameIndex(FuncInfo.getRegSavedAreaStartFrameIndex(),
+                                    MVT::i32),
+                  DAG.getIntPtrConstant(4, DL));
+
+  // Store the saved register area start pointer.
+  SDValue Store =
+    DAG.getStore(Op.getOperand(0), DL,
+                 SavedRegAreaStartFrameIndex,
+                 FIN, MachinePointerInfo(SV));
+  MemOps.push_back(Store);
+
+  // Store saved register area end pointer.
+  FIN = DAG.getNode(ISD::ADD, DL, PtrVT,
+                    FIN, DAG.getIntPtrConstant(4, DL));
+  Store = DAG.getStore(Op.getOperand(0), DL,
+                       DAG.getFrameIndex(FuncInfo.getVarArgsFrameIndex(),
+                                         PtrVT),
+                       FIN, MachinePointerInfo(SV, 4));
+  MemOps.push_back(Store);
+
+  // Store overflow area pointer.
+  FIN = DAG.getNode(ISD::ADD, DL, PtrVT,
+                    FIN, DAG.getIntPtrConstant(4, DL));
+  Store = DAG.getStore(Op.getOperand(0), DL,
+                       DAG.getFrameIndex(FuncInfo.getVarArgsFrameIndex(),
+                                         PtrVT),
+                       FIN, MachinePointerInfo(SV, 8));
+  MemOps.push_back(Store);
+
+  return DAG.getNode(ISD::TokenFactor, DL, MVT::Other, MemOps);
+}
+
+SDValue
+HexagonTargetLowering::LowerVACOPY(SDValue Op, SelectionDAG &DAG) const {
+  // Assert that the linux ABI is enabled for the current compilation.
+  assert(Subtarget.isEnvironmentMusl() && "Linux ABI should be enabled");
+  SDValue Chain = Op.getOperand(0);
+  SDValue DestPtr = Op.getOperand(1);
+  SDValue SrcPtr = Op.getOperand(2);
+  const Value *DestSV = cast<SrcValueSDNode>(Op.getOperand(3))->getValue();
+  const Value *SrcSV = cast<SrcValueSDNode>(Op.getOperand(4))->getValue();
+  SDLoc DL(Op);
+  // Size of the va_list is 12 bytes as it has 3 pointers. Therefore,
+  // we need to memcopy 12 bytes from va_list to another similar list.
+  return DAG.getMemcpy(Chain, DL, DestPtr, SrcPtr,
+                       DAG.getIntPtrConstant(12, DL), 4, /*isVolatile*/false,
+                       false, false,
+                       MachinePointerInfo(DestSV), MachinePointerInfo(SrcSV));
+
 }
 
 SDValue HexagonTargetLowering::LowerSETCC(SDValue Op, SelectionDAG &DAG) const {
@@ -1375,7 +1513,10 @@
   setOperationAction(ISD::VASTART, MVT::Other, Custom);
   setOperationAction(ISD::VAEND,   MVT::Other, Expand);
   setOperationAction(ISD::VAARG,   MVT::Other, Expand);
-  setOperationAction(ISD::VACOPY,  MVT::Other, Expand);
+  if (Subtarget.isEnvironmentMusl())
+    setOperationAction(ISD::VACOPY, MVT::Other, Custom);
+  else
+    setOperationAction(ISD::VACOPY,  MVT::Other, Expand);
 
   setOperationAction(ISD::STACKSAVE, MVT::Other, Expand);
   setOperationAction(ISD::STACKRESTORE, MVT::Other, Expand);
@@ -2928,6 +3069,7 @@
     case ISD::GlobalAddress:        return LowerGLOBALADDRESS(Op, DAG);
     case ISD::BlockAddress:         return LowerBlockAddress(Op, DAG);
     case ISD::GLOBAL_OFFSET_TABLE:  return LowerGLOBAL_OFFSET_TABLE(Op, DAG);
+    case ISD::VACOPY:              return LowerVACOPY(Op, DAG);
     case ISD::VASTART:              return LowerVASTART(Op, DAG);
     case ISD::DYNAMIC_STACKALLOC:   return LowerDYNAMIC_STACKALLOC(Op, DAG);
     case ISD::SETCC:                return LowerSETCC(Op, DAG);
diff --git a/llvm/lib/Target/Hexagon/HexagonMachineFunctionInfo.h b/llvm/lib/Target/Hexagon/HexagonMachineFunctionInfo.h
--- a/llvm/lib/Target/Hexagon/HexagonMachineFunctionInfo.h
+++ b/llvm/lib/Target/Hexagon/HexagonMachineFunctionInfo.h
@@ -30,6 +30,9 @@
   unsigned StackAlignBaseVReg = 0;    // Aligned-stack base register (virtual)
   unsigned StackAlignBasePhysReg = 0; //                             (physical)
   int VarArgsFrameIndex;
+  int RegSavedAreaStartFrameIndex;
+  int FirstNamedArgFrameIndex;
+  int LastNamedArgFrameIndex;
   bool HasClobberLR = false;
   bool HasEHReturn = false;
   std::map<const MachineInstr*, unsigned> PacketInfo;
@@ -46,6 +49,15 @@
   void setVarArgsFrameIndex(int v) { VarArgsFrameIndex = v; }
   int getVarArgsFrameIndex() { return VarArgsFrameIndex; }
 
+  void setRegSavedAreaStartFrameIndex(int v) { RegSavedAreaStartFrameIndex = v;}
+  int getRegSavedAreaStartFrameIndex() { return RegSavedAreaStartFrameIndex; }
+
+  void setFirstNamedArgFrameIndex(int v) { FirstNamedArgFrameIndex = v; }
+  int getFirstNamedArgFrameIndex() { return FirstNamedArgFrameIndex; }
+
+  void setLastNamedArgFrameIndex(int v) { LastNamedArgFrameIndex = v; }
+  int getLastNamedArgFrameIndex() { return LastNamedArgFrameIndex; }
+
   void setStartPacket(MachineInstr* MI) {
     PacketInfo[MI] |= Hexagon::StartPacket;
   }
diff --git a/llvm/lib/Target/Hexagon/HexagonSubtarget.h b/llvm/lib/Target/Hexagon/HexagonSubtarget.h
--- a/llvm/lib/Target/Hexagon/HexagonSubtarget.h
+++ b/llvm/lib/Target/Hexagon/HexagonSubtarget.h
@@ -85,6 +85,7 @@
 
 private:
   std::string CPUString;
+  Triple TargetTriple;
   HexagonInstrInfo InstrInfo;
   HexagonRegisterInfo RegInfo;
   HexagonTargetLowering TLInfo;
@@ -96,6 +97,11 @@
   HexagonSubtarget(const Triple &TT, StringRef CPU, StringRef FS,
                    const TargetMachine &TM);
 
+  const Triple &getTargetTriple() const { return TargetTriple; }
+  bool isEnvironmentMusl() const {
+    return TargetTriple.getEnvironment() == Triple::Musl;
+  }
+
   /// getInstrItins - Return the instruction itineraries based on subtarget
   /// selection.
   const InstrItineraryData *getInstrItineraryData() const override {
diff --git a/llvm/lib/Target/Hexagon/HexagonSubtarget.cpp b/llvm/lib/Target/Hexagon/HexagonSubtarget.cpp
--- a/llvm/lib/Target/Hexagon/HexagonSubtarget.cpp
+++ b/llvm/lib/Target/Hexagon/HexagonSubtarget.cpp
@@ -79,7 +79,7 @@
 HexagonSubtarget::HexagonSubtarget(const Triple &TT, StringRef CPU,
                                    StringRef FS, const TargetMachine &TM)
     : HexagonGenSubtargetInfo(TT, CPU, FS), OptLevel(TM.getOptLevel()),
-      CPUString(Hexagon_MC::selectHexagonCPU(CPU)),
+      CPUString(Hexagon_MC::selectHexagonCPU(CPU)), TargetTriple(TT),
       InstrInfo(initializeSubtargetDependencies(CPU, FS)),
       RegInfo(getHwMode()), TLInfo(TM, *this),
       InstrItins(getInstrItineraryForCPU(CPUString)) {
diff --git a/llvm/test/CodeGen/Hexagon/vacopy.ll b/llvm/test/CodeGen/Hexagon/vacopy.ll
new file mode 100644
--- /dev/null
+++ b/llvm/test/CodeGen/Hexagon/vacopy.ll
@@ -0,0 +1,39 @@
+; RUN: llc -march=hexagon -mcpu=hexagonv62 -mtriple=hexagon-unknown-linux-musl < %s | FileCheck %s
+; CHECK-LABEL: PrintInts:
+; CHECK-DAG: memw{{.*}} = r{{[0-9]+}}
+; CHECK-DAG: memw{{.*}} = r{{[0-9]+}}
+; CHECK-DAG: r{{[0-9]+}}:{{[0-9]+}} = memd{{.*}}
+; CHECK-DAG: memd{{.*}} = r{{[0-9]+}}:{{[0-9]+}}
+
+%struct.__va_list_tag = type { i8*, i8*, i8* }
+
+; Function Attrs: nounwind
+define void @PrintInts(i32 %first, ...) #0 {
+entry:
+  %vl = alloca [1 x %struct.__va_list_tag], align 8
+  %vl_count = alloca [1 x %struct.__va_list_tag], align 8
+  %arraydecay1 = bitcast [1 x %struct.__va_list_tag]* %vl to i8*
+  call void @llvm.va_start(i8* %arraydecay1)
+  %0 = bitcast [1 x %struct.__va_list_tag]* %vl_count to i8*
+  call void @llvm.va_copy(i8* %0, i8* %arraydecay1)
+  ret void
+}
+
+; Function Attrs: nounwind
+declare void @llvm.va_start(i8*) #1
+
+; Function Attrs: nounwind
+declare void @llvm.va_copy(i8*, i8*) #1
+
+; Function Attrs: nounwind
+define i32 @main() #0 {
+entry:
+  tail call void (i32, ...) @PrintInts(i32 undef, i32 20, i32 30, i32 40, i32 50, i32 0)
+  ret i32 0
+}
+
+attributes #0 = { nounwind }
+
+!llvm.ident = !{!0}
+
+!0 = !{!"Clang 3.1"}
diff --git a/llvm/test/CodeGen/Hexagon/vararg-deallocate-sp.ll b/llvm/test/CodeGen/Hexagon/vararg-deallocate-sp.ll
new file mode 100644
--- /dev/null
+++ b/llvm/test/CodeGen/Hexagon/vararg-deallocate-sp.ll
@@ -0,0 +1,13 @@
+; RUN: llc -march=hexagon -mtriple=hexagon-unknown-linux-musl < %s | FileCheck %s
+
+; Test that the compiler deallocates the register saved area on Linux
+; for functions that do not need a frame pointer.
+
+; CHECK: r29 = add(r29,#-[[SIZE:[0-9]+]]
+; CHECK: r29 = add(r29,#[[SIZE]])
+
+define void @test(...) {
+entry:
+  ret void
+}
+
diff --git a/llvm/test/CodeGen/Hexagon/vararg-linux-abi.ll b/llvm/test/CodeGen/Hexagon/vararg-linux-abi.ll
new file mode 100644
--- /dev/null
+++ b/llvm/test/CodeGen/Hexagon/vararg-linux-abi.ll
@@ -0,0 +1,93 @@
+; RUN: llc -march=hexagon -mtriple=hexagon-unknown-linux-musl < %s | FileCheck %s
+
+; Check that we update the stack pointer before we do allocframe, so that
+; the LR/FP are stored in the location required by the Linux ABI.
+; CHECK: r29 = add(r29,#-24)
+; CHECK: allocframe
+
+target triple = "hexagon-unknown-linux"
+
+%s.0 = type { i8*, i8*, i8* }
+
+define dso_local i32 @f0(i32 %a0, ...) local_unnamed_addr #0 {
+b0:
+  %v0 = alloca [1 x %s.0], align 8
+  %v1 = bitcast [1 x %s.0]* %v0 to i8*
+  call void @llvm.lifetime.start.p0i8(i64 12, i8* nonnull %v1) #2
+  call void @llvm.va_start(i8* nonnull %v1)
+  %v2 = getelementptr inbounds [1 x %s.0], [1 x %s.0]* %v0, i32 0, i32 0, i32 0
+  %v3 = load i8*, i8** %v2, align 8
+  %v4 = getelementptr inbounds [1 x %s.0], [1 x %s.0]* %v0, i32 0, i32 0, i32 1
+  %v5 = load i8*, i8** %v4, align 4
+  %v6 = getelementptr i8, i8* %v3, i32 4
+  %v7 = icmp sgt i8* %v6, %v5
+  br i1 %v7, label %b1, label %b2
+
+b1:                                               ; preds = %b0
+  %v8 = getelementptr inbounds [1 x %s.0], [1 x %s.0]* %v0, i32 0, i32 0, i32 2
+  %v9 = load i8*, i8** %v8, align 8
+  %v10 = getelementptr i8, i8* %v9, i32 4
+  store i8* %v10, i8** %v8, align 8
+  br label %b2
+
+b2:                                               ; preds = %b1, %b0
+  %v11 = phi i8* [ %v10, %b1 ], [ %v6, %b0 ]
+  %v12 = phi i8* [ %v9, %b1 ], [ %v3, %b0 ]
+  %v13 = bitcast i8* %v12 to i32*
+  store i8* %v11, i8** %v2, align 8
+  %v14 = load i32, i32* %v13, align 4
+  %v15 = icmp eq i32 %v14, 0
+  br i1 %v15, label %b7, label %b3
+
+b3:                                               ; preds = %b2
+  %v16 = getelementptr inbounds [1 x %s.0], [1 x %s.0]* %v0, i32 0, i32 0, i32 2
+  br label %b4
+
+b4:                                               ; preds = %b6, %b3
+  %v17 = phi i32 [ %v14, %b3 ], [ %v28, %b6 ]
+  %v18 = phi i32 [ %a0, %b3 ], [ %v20, %b6 ]
+  %v19 = phi i8* [ %v11, %b3 ], [ %v25, %b6 ]
+  %v20 = add nsw i32 %v17, %v18
+  %v21 = getelementptr i8, i8* %v19, i32 4
+  %v22 = icmp sgt i8* %v21, %v5
+  br i1 %v22, label %b5, label %b6
+
+b5:                                               ; preds = %b4
+  %v23 = load i8*, i8** %v16, align 8
+  %v24 = getelementptr i8, i8* %v23, i32 4
+  store i8* %v24, i8** %v16, align 8
+  br label %b6
+
+b6:                                               ; preds = %b5, %b4
+  %v25 = phi i8* [ %v24, %b5 ], [ %v21, %b4 ]
+  %v26 = phi i8* [ %v23, %b5 ], [ %v19, %b4 ]
+  %v27 = bitcast i8* %v26 to i32*
+  store i8* %v25, i8** %v2, align 8
+  %v28 = load i32, i32* %v27, align 4
+  %v29 = icmp eq i32 %v28, 0
+  br i1 %v29, label %b7, label %b4
+
+b7:                                               ; preds = %b6, %b2
+  %v30 = phi i32 [ %a0, %b2 ], [ %v20, %b6 ]
+  call void @llvm.va_end(i8* nonnull %v1)
+  call void @llvm.lifetime.end.p0i8(i64 12, i8* nonnull %v1) #2
+  ret i32 %v30
+}
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.start.p0i8(i64, i8* nocapture) #1
+
+; Function Attrs: nounwind
+declare void @llvm.va_start(i8*) #2
+
+; Function Attrs: nounwind
+declare void @llvm.va_end(i8*) #2
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.end.p0i8(i64, i8* nocapture) #1
+
+attributes #0 = { argmemonly nounwind "frame-pointer"="all" }
+
+!llvm.module.flags = !{!0}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
diff --git a/llvm/test/CodeGen/Hexagon/vararg.ll b/llvm/test/CodeGen/Hexagon/vararg.ll
new file mode 100644
--- /dev/null
+++ b/llvm/test/CodeGen/Hexagon/vararg.ll
@@ -0,0 +1,97 @@
+; RUN: llc -march=hexagon -mcpu=hexagonv62 -mtriple=hexagon-unknown-linux-musl  -O0 < %s | FileCheck %s
+
+; CHECK-LABEL: foo:
+
+; Check function prologue generation
+; CHECK: r29 = add(r29,#-24)
+; CHECK: memw(r29+#4) = r1
+; CHECK: memw(r29+#8) = r2
+; CHECK: memw(r29+#12) = r3
+; CHECK: memw(r29+#16) = r4
+; CHECK: memw(r29+#20) = r5
+; CHECK: r29 = add(r29,#24)
+
+
+%struct.AAA = type { i32, i32, i32, i32 }
+%struct.__va_list_tag = type { i8*, i8*, i8* }
+
+@aaa = global %struct.AAA { i32 100, i32 200, i32 300, i32 400 }, align 4
+@.str = private unnamed_addr constant [13 x i8] c"result = %d\0A\00", align 1
+
+; Function Attrs: nounwind
+define i32 @foo(i32 %xx, ...) #0 {
+entry:
+  %ap = alloca [1 x %struct.__va_list_tag], align 8
+  %arraydecay1 = bitcast [1 x %struct.__va_list_tag]* %ap to i8*
+  call void @llvm.va_start(i8* %arraydecay1)
+  %__current_saved_reg_area_pointer_p = getelementptr inbounds [1 x %struct.__va_list_tag], [1 x %struct.__va_list_tag]* %ap, i32 0, i32 0, i32 0
+  %__current_saved_reg_area_pointer = load i8*, i8** %__current_saved_reg_area_pointer_p, align 8
+  %__saved_reg_area_end_pointer_p = getelementptr inbounds [1 x %struct.__va_list_tag], [1 x %struct.__va_list_tag]* %ap, i32 0, i32 0, i32 1
+  %__saved_reg_area_end_pointer = load i8*, i8** %__saved_reg_area_end_pointer_p, align 4
+  %__new_saved_reg_area_pointer = getelementptr i8, i8* %__current_saved_reg_area_pointer, i32 4
+  %0 = icmp sgt i8* %__new_saved_reg_area_pointer, %__saved_reg_area_end_pointer
+  %__overflow_area_pointer_p = getelementptr inbounds [1 x %struct.__va_list_tag], [1 x %struct.__va_list_tag]* %ap, i32 0, i32 0, i32 2
+  %__overflow_area_pointer = load i8*, i8** %__overflow_area_pointer_p, align 8
+  br i1 %0, label %vaarg.on_stack, label %vaarg.end
+
+vaarg.on_stack:                                   ; preds = %entry
+  %__overflow_area_pointer.next = getelementptr i8, i8* %__overflow_area_pointer, i32 4
+  store i8* %__overflow_area_pointer.next, i8** %__overflow_area_pointer_p, align 8
+  br label %vaarg.end
+
+vaarg.end:                                        ; preds = %entry, %vaarg.on_stack
+  %__overflow_area_pointer5 = phi i8* [ %__overflow_area_pointer.next, %vaarg.on_stack ], [ %__overflow_area_pointer, %entry ]
+  %storemerge32 = phi i8* [ %__overflow_area_pointer.next, %vaarg.on_stack ], [ %__new_saved_reg_area_pointer, %entry ]
+  %vaarg.addr.in = phi i8* [ %__overflow_area_pointer, %vaarg.on_stack ], [ %__current_saved_reg_area_pointer, %entry ]
+  store i8* %storemerge32, i8** %__current_saved_reg_area_pointer_p, align 8
+  %vaarg.addr = bitcast i8* %vaarg.addr.in to i32*
+  %1 = load i32, i32* %vaarg.addr, align 4
+  %__overflow_area_pointer_p4 = getelementptr inbounds [1 x %struct.__va_list_tag], [1 x %struct.__va_list_tag]* %ap, i32 0, i32 0, i32 2
+  %__overflow_area_pointer.next6 = getelementptr i8, i8* %__overflow_area_pointer5, i32 16
+  store i8* %__overflow_area_pointer.next6, i8** %__overflow_area_pointer_p4, align 8
+  %bbb.sroa.1.0.idx27 = getelementptr inbounds i8, i8* %__overflow_area_pointer5, i32 12
+  %2 = bitcast i8* %bbb.sroa.1.0.idx27 to i32*
+  %bbb.sroa.1.0.copyload = load i32, i32* %2, align 4
+  %add8 = add nsw i32 %bbb.sroa.1.0.copyload, %1
+  %__new_saved_reg_area_pointer15 = getelementptr i8, i8* %storemerge32, i32 4
+  %3 = icmp sgt i8* %__new_saved_reg_area_pointer15, %__saved_reg_area_end_pointer
+  br i1 %3, label %vaarg.on_stack17, label %vaarg.end21
+
+vaarg.on_stack17:                                 ; preds = %vaarg.end
+  %__overflow_area_pointer.next20 = getelementptr i8, i8* %__overflow_area_pointer5, i32 20
+  store i8* %__overflow_area_pointer.next20, i8** %__overflow_area_pointer_p4, align 8
+  br label %vaarg.end21
+
+vaarg.end21:                                      ; preds = %vaarg.end, %vaarg.on_stack17
+  %storemerge = phi i8* [ %__overflow_area_pointer.next20, %vaarg.on_stack17 ], [ %__new_saved_reg_area_pointer15, %vaarg.end ]
+  %vaarg.addr22.in = phi i8* [ %__overflow_area_pointer.next6, %vaarg.on_stack17 ], [ %storemerge32, %vaarg.end ]
+  store i8* %storemerge, i8** %__current_saved_reg_area_pointer_p, align 8
+  %vaarg.addr22 = bitcast i8* %vaarg.addr22.in to i32*
+  %4 = load i32, i32* %vaarg.addr22, align 4
+  %add23 = add nsw i32 %add8, %4
+  call void @llvm.va_end(i8* %arraydecay1)
+  ret i32 %add23
+}
+
+; Function Attrs: nounwind
+declare void @llvm.va_start(i8*) #1
+
+; Function Attrs: nounwind
+declare void @llvm.va_end(i8*) #1
+
+; Function Attrs: nounwind
+define i32 @main() #0 {
+entry:
+  %call = tail call i32 (i32, ...) @foo(i32 undef, i32 2, %struct.AAA* byval align 4 @aaa, i32 4)
+  %call1 = tail call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([13 x i8], [13 x i8]* @.str, i32 0, i32 0), i32 %call) #1
+  ret i32 %call
+}
+
+; Function Attrs: nounwind
+declare i32 @printf(i8* nocapture readonly, ...) #0
+
+attributes #0 = { nounwind }
+
+!llvm.ident = !{!0}
+
+!0 = !{!"Clang 3.1"}
diff --git a/llvm/test/CodeGen/Hexagon/vararg_align_check.ll b/llvm/test/CodeGen/Hexagon/vararg_align_check.ll
new file mode 100644
--- /dev/null
+++ b/llvm/test/CodeGen/Hexagon/vararg_align_check.ll
@@ -0,0 +1,186 @@
+; RUN: llc -march=hexagon -mcpu=hexagonv62  -mtriple=hexagon-unknown-linux-musl -O0 < %s | FileCheck %s
+
+; CHECK-LABEL: foo:
+
+; Check Function prologue.
+; Note. All register numbers and offset are fixed.
+; Hence, no need of regular expression.
+
+; CHECK: r29 = add(r29,#-24)
+; CHECK: r7:6 = memd(r29+#24)
+; CHECK: memd(r29+#0) = r7:6
+; CHECK: r7:6 = memd(r29+#32)
+; CHECK: memd(r29+#8) = r7:6
+; CHECK: r7:6 = memd(r29+#40)
+; CHECK: memd(r29+#16) = r7:6
+; CHECK: memw(r29+#28) = r1
+; CHECK: memw(r29+#32) = r2
+; CHECK: memw(r29+#36) = r3
+; CHECK: memw(r29+#40) = r4
+; CHECK: memw(r29+#44) = r5
+; CHECK: r29 = add(r29,#24)
+
+%struct.AAA = type { i32, i32, i32, i32 }
+%struct.BBB = type { i8, i64, i32 }
+%struct.__va_list_tag = type { i8*, i8*, i8* }
+
+@aaa = global %struct.AAA { i32 100, i32 200, i32 300, i32 400 }, align 4
+@ddd = global { i8, i64, i32, [4 x i8] } { i8 1, i64 1000000, i32 5, [4 x i8] undef }, align 8
+@.str = private unnamed_addr constant [13 x i8] c"result = %d\0A\00", align 1
+
+; Function Attrs: nounwind
+define i32 @foo(i32 %xx, %struct.BBB* byval align 8 %eee, ...) #0 {
+entry:
+  %xx.addr = alloca i32, align 4
+  %ap = alloca [1 x %struct.__va_list_tag], align 8
+  %d = alloca i32, align 4
+  %k = alloca i64, align 8
+  %ret = alloca i32, align 4
+  %bbb = alloca %struct.AAA, align 4
+  store i32 %xx, i32* %xx.addr, align 4
+  store i32 0, i32* %ret, align 4
+  %x = getelementptr inbounds %struct.BBB, %struct.BBB* %eee, i32 0, i32 0
+  %0 = load i8, i8* %x, align 1
+  %tobool = trunc i8 %0 to i1
+  br i1 %tobool, label %if.then, label %if.end
+
+if.then:                                          ; preds = %entry
+  store i32 1, i32* %ret, align 4
+  br label %if.end
+
+if.end:                                           ; preds = %if.then, %entry
+  %arraydecay = getelementptr inbounds [1 x %struct.__va_list_tag], [1 x %struct.__va_list_tag]* %ap, i32 0, i32 0
+  %arraydecay1 = bitcast %struct.__va_list_tag* %arraydecay to i8*
+  call void @llvm.va_start(i8* %arraydecay1)
+  %arraydecay2 = getelementptr inbounds [1 x %struct.__va_list_tag], [1 x %struct.__va_list_tag]* %ap, i32 0, i32 0
+  br label %vaarg.maybe_reg
+
+vaarg.maybe_reg:                                  ; preds = %if.end
+  %__current_saved_reg_area_pointer_p = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %arraydecay2, i32 0, i32 0
+  %__current_saved_reg_area_pointer = load i8*, i8** %__current_saved_reg_area_pointer_p
+  %__saved_reg_area_end_pointer_p = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %arraydecay2, i32 0, i32 1
+  %__saved_reg_area_end_pointer = load i8*, i8** %__saved_reg_area_end_pointer_p
+  %1 = ptrtoint i8* %__current_saved_reg_area_pointer to i32
+  %align_current_saved_reg_area_pointer = add i32 %1, 7
+  %align_current_saved_reg_area_pointer3 = and i32 %align_current_saved_reg_area_pointer, -8
+  %align_current_saved_reg_area_pointer4 = inttoptr i32 %align_current_saved_reg_area_pointer3 to i8*
+  %__new_saved_reg_area_pointer = getelementptr i8, i8* %align_current_saved_reg_area_pointer4, i32 8
+  %2 = icmp sgt i8* %__new_saved_reg_area_pointer, %__saved_reg_area_end_pointer
+  br i1 %2, label %vaarg.on_stack, label %vaarg.in_reg
+
+vaarg.in_reg:                                     ; preds = %vaarg.maybe_reg
+  %3 = bitcast i8* %align_current_saved_reg_area_pointer4 to i64*
+  store i8* %__new_saved_reg_area_pointer, i8** %__current_saved_reg_area_pointer_p
+  br label %vaarg.end
+
+vaarg.on_stack:                                   ; preds = %vaarg.maybe_reg
+  %__overflow_area_pointer_p = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %arraydecay2, i32 0, i32 2
+  %__overflow_area_pointer = load i8*, i8** %__overflow_area_pointer_p
+  %4 = ptrtoint i8* %__overflow_area_pointer to i32
+  %align_overflow_area_pointer = add i32 %4, 7
+  %align_overflow_area_pointer5 = and i32 %align_overflow_area_pointer, -8
+  %align_overflow_area_pointer6 = inttoptr i32 %align_overflow_area_pointer5 to i8*
+  %__overflow_area_pointer.next = getelementptr i8, i8* %align_overflow_area_pointer6, i32 8
+  store i8* %__overflow_area_pointer.next, i8** %__overflow_area_pointer_p
+  store i8* %__overflow_area_pointer.next, i8** %__current_saved_reg_area_pointer_p
+  %5 = bitcast i8* %align_overflow_area_pointer6 to i64*
+  br label %vaarg.end
+
+vaarg.end:                                        ; preds = %vaarg.on_stack, %vaarg.in_reg
+  %vaarg.addr = phi i64* [ %3, %vaarg.in_reg ], [ %5, %vaarg.on_stack ]
+  %6 = load i64, i64* %vaarg.addr
+  store i64 %6, i64* %k, align 8
+  %7 = load i64, i64* %k, align 8
+  %conv = trunc i64 %7 to i32
+  %div = sdiv i32 %conv, 1000
+  %8 = load i32, i32* %ret, align 4
+  %add = add nsw i32 %8, %div
+  store i32 %add, i32* %ret, align 4
+  %arraydecay7 = getelementptr inbounds [1 x %struct.__va_list_tag], [1 x %struct.__va_list_tag]* %ap, i32 0, i32 0
+  %__overflow_area_pointer_p8 = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %arraydecay7, i32 0, i32 2
+  %__overflow_area_pointer9 = load i8*, i8** %__overflow_area_pointer_p8
+  %9 = bitcast i8* %__overflow_area_pointer9 to %struct.AAA*
+  %__overflow_area_pointer.next10 = getelementptr i8, i8* %__overflow_area_pointer9, i32 16
+  store i8* %__overflow_area_pointer.next10, i8** %__overflow_area_pointer_p8
+  %10 = bitcast %struct.AAA* %bbb to i8*
+  %11 = bitcast %struct.AAA* %9 to i8*
+  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %10, i8* %11, i32 16, i32 4, i1 false)
+  %d11 = getelementptr inbounds %struct.AAA, %struct.AAA* %bbb, i32 0, i32 3
+  %12 = load i32, i32* %d11, align 4
+  %13 = load i32, i32* %ret, align 4
+  %add12 = add nsw i32 %13, %12
+  store i32 %add12, i32* %ret, align 4
+  %arraydecay13 = getelementptr inbounds [1 x %struct.__va_list_tag], [1 x %struct.__va_list_tag]* %ap, i32 0, i32 0
+  br label %vaarg.maybe_reg14
+
+vaarg.maybe_reg14:                                ; preds = %vaarg.end
+  %__current_saved_reg_area_pointer_p15 = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %arraydecay13, i32 0, i32 0
+  %__current_saved_reg_area_pointer16 = load i8*, i8** %__current_saved_reg_area_pointer_p15
+  %__saved_reg_area_end_pointer_p17 = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %arraydecay13, i32 0, i32 1
+  %__saved_reg_area_end_pointer18 = load i8*, i8** %__saved_reg_area_end_pointer_p17
+  %__new_saved_reg_area_pointer19 = getelementptr i8, i8* %__current_saved_reg_area_pointer16, i32 4
+  %14 = icmp sgt i8* %__new_saved_reg_area_pointer19, %__saved_reg_area_end_pointer18
+  br i1 %14, label %vaarg.on_stack21, label %vaarg.in_reg20
+
+vaarg.in_reg20:                                   ; preds = %vaarg.maybe_reg14
+  %15 = bitcast i8* %__current_saved_reg_area_pointer16 to i32*
+  store i8* %__new_saved_reg_area_pointer19, i8** %__current_saved_reg_area_pointer_p15
+  br label %vaarg.end25
+
+vaarg.on_stack21:                                 ; preds = %vaarg.maybe_reg14
+  %__overflow_area_pointer_p22 = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %arraydecay13, i32 0, i32 2
+  %__overflow_area_pointer23 = load i8*, i8** %__overflow_area_pointer_p22
+  %__overflow_area_pointer.next24 = getelementptr i8, i8* %__overflow_area_pointer23, i32 4
+  store i8* %__overflow_area_pointer.next24, i8** %__overflow_area_pointer_p22
+  store i8* %__overflow_area_pointer.next24, i8** %__current_saved_reg_area_pointer_p15
+  %16 = bitcast i8* %__overflow_area_pointer23 to i32*
+  br label %vaarg.end25
+
+vaarg.end25:                                      ; preds = %vaarg.on_stack21, %vaarg.in_reg20
+  %vaarg.addr26 = phi i32* [ %15, %vaarg.in_reg20 ], [ %16, %vaarg.on_stack21 ]
+  %17 = load i32, i32* %vaarg.addr26
+  store i32 %17, i32* %d, align 4
+  %18 = load i32, i32* %d, align 4
+  %19 = load i32, i32* %ret, align 4
+  %add27 = add nsw i32 %19, %18
+  store i32 %add27, i32* %ret, align 4
+  %arraydecay28 = getelementptr inbounds [1 x %struct.__va_list_tag], [1 x %struct.__va_list_tag]* %ap, i32 0, i32 0
+  %arraydecay2829 = bitcast %struct.__va_list_tag* %arraydecay28 to i8*
+  call void @llvm.va_end(i8* %arraydecay2829)
+  %20 = load i32, i32* %ret, align 4
+  ret i32 %20
+}
+
+; Function Attrs: nounwind
+declare void @llvm.va_start(i8*) #1
+
+; Function Attrs: nounwind
+declare void @llvm.memcpy.p0i8.p0i8.i32(i8* nocapture, i8* nocapture readonly, i32, i32, i1) #1
+
+; Function Attrs: nounwind
+declare void @llvm.va_end(i8*) #1
+
+; Function Attrs: nounwind
+define i32 @main() #0 {
+entry:
+  %retval = alloca i32, align 4
+  %x = alloca i32, align 4
+  %m = alloca i64, align 8
+  store i32 0, i32* %retval
+  store i64 1000000, i64* %m, align 8
+  %0 = load i64, i64* %m, align 8
+  %call = call i32 (i32, %struct.BBB*, ...) @foo(i32 1, %struct.BBB* byval align 8 bitcast ({ i8, i64, i32, [4 x i8] }* @ddd to %struct.BBB*), i64 %0, %struct.AAA* byval align 4 @aaa, i32 4)
+  store i32 %call, i32* %x, align 4
+  %1 = load i32, i32* %x, align 4
+  %call1 = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([13 x i8], [13 x i8]* @.str, i32 0, i32 0), i32 %1)
+  %2 = load i32, i32* %x, align 4
+  ret i32 %2
+}
+
+declare i32 @printf(i8*, ...) #2
+
+attributes #1 = { nounwind }
+
+!llvm.ident = !{!0}
+
+!0 = !{!"Clang 3.1"}
diff --git a/llvm/test/CodeGen/Hexagon/vararg_double_onstack.ll b/llvm/test/CodeGen/Hexagon/vararg_double_onstack.ll
new file mode 100644
--- /dev/null
+++ b/llvm/test/CodeGen/Hexagon/vararg_double_onstack.ll
@@ -0,0 +1,214 @@
+; RUN: llc -march=hexagon -mcpu=hexagonv62  -mtriple=hexagon-unknown-linux-musl -O0 < %s | FileCheck %s
+
+; CHECK-LABEL: foo:
+
+; Check Function prologue.
+; Note. All register numbers and offset are fixed.
+; Hence, no need of regular expression.
+
+; CHECK: r29 = add(r29,#-8)
+; CHECK: memw(r29+#4) = r5
+; CHECK: r29 = add(r29,#8)
+
+%struct.AAA = type { i32, i32, i32, i32 }
+%struct.__va_list_tag = type { i8*, i8*, i8* }
+
+@aaa = global %struct.AAA { i32 100, i32 200, i32 300, i32 400 }, align 4
+@.str = private unnamed_addr constant [13 x i8] c"result = %d\0A\00", align 1
+
+; Function Attrs: nounwind
+define i32 @foo(i32 %xx, i32 %a, i32 %b, i32 %c, i32 %x, ...) #0 {
+entry:
+  %xx.addr = alloca i32, align 4
+  %a.addr = alloca i32, align 4
+  %b.addr = alloca i32, align 4
+  %c.addr = alloca i32, align 4
+  %x.addr = alloca i32, align 4
+  %ap = alloca [1 x %struct.__va_list_tag], align 8
+  %d = alloca i32, align 4
+  %ret = alloca i32, align 4
+  %bbb = alloca %struct.AAA, align 4
+  store i32 %xx, i32* %xx.addr, align 4
+  store i32 %a, i32* %a.addr, align 4
+  store i32 %b, i32* %b.addr, align 4
+  store i32 %c, i32* %c.addr, align 4
+  store i32 %x, i32* %x.addr, align 4
+  store i32 0, i32* %ret, align 4
+  %arraydecay = getelementptr inbounds [1 x %struct.__va_list_tag], [1 x %struct.__va_list_tag]* %ap, i32 0, i32 0
+  %arraydecay1 = bitcast %struct.__va_list_tag* %arraydecay to i8*
+  call void @llvm.va_start(i8* %arraydecay1)
+  %arraydecay2 = getelementptr inbounds [1 x %struct.__va_list_tag], [1 x %struct.__va_list_tag]* %ap, i32 0, i32 0
+  br label %vaarg.maybe_reg
+
+vaarg.maybe_reg:                                  ; preds = %entry
+  %__current_saved_reg_area_pointer_p = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %arraydecay2, i32 0, i32 0
+  %__current_saved_reg_area_pointer = load i8*, i8** %__current_saved_reg_area_pointer_p
+  %__saved_reg_area_end_pointer_p = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %arraydecay2, i32 0, i32 1
+  %__saved_reg_area_end_pointer = load i8*, i8** %__saved_reg_area_end_pointer_p
+  %0 = ptrtoint i8* %__current_saved_reg_area_pointer to i32
+  %align_current_saved_reg_area_pointer = add i32 %0, 7
+  %align_current_saved_reg_area_pointer3 = and i32 %align_current_saved_reg_area_pointer, -8
+  %align_current_saved_reg_area_pointer4 = inttoptr i32 %align_current_saved_reg_area_pointer3 to i8*
+  %__new_saved_reg_area_pointer = getelementptr i8, i8* %align_current_saved_reg_area_pointer4, i32 8
+  %1 = icmp sgt i8* %__new_saved_reg_area_pointer, %__saved_reg_area_end_pointer
+  br i1 %1, label %vaarg.on_stack, label %vaarg.in_reg
+
+vaarg.in_reg:                                     ; preds = %vaarg.maybe_reg
+  %2 = bitcast i8* %align_current_saved_reg_area_pointer4 to i64*
+  store i8* %__new_saved_reg_area_pointer, i8** %__current_saved_reg_area_pointer_p
+  br label %vaarg.end
+
+vaarg.on_stack:                                   ; preds = %vaarg.maybe_reg
+  %__overflow_area_pointer_p = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %arraydecay2, i32 0, i32 2
+  %__overflow_area_pointer = load i8*, i8** %__overflow_area_pointer_p
+  %3 = ptrtoint i8* %__overflow_area_pointer to i32
+  %align_overflow_area_pointer = add i32 %3, 7
+  %align_overflow_area_pointer5 = and i32 %align_overflow_area_pointer, -8
+  %align_overflow_area_pointer6 = inttoptr i32 %align_overflow_area_pointer5 to i8*
+  %__overflow_area_pointer.next = getelementptr i8, i8* %align_overflow_area_pointer6, i32 8
+  store i8* %__overflow_area_pointer.next, i8** %__overflow_area_pointer_p
+  store i8* %__overflow_area_pointer.next, i8** %__current_saved_reg_area_pointer_p
+  %4 = bitcast i8* %align_overflow_area_pointer6 to i64*
+  br label %vaarg.end
+
+vaarg.end:                                        ; preds = %vaarg.on_stack, %vaarg.in_reg
+  %vaarg.addr = phi i64* [ %2, %vaarg.in_reg ], [ %4, %vaarg.on_stack ]
+  %5 = load i64, i64* %vaarg.addr
+  %conv = trunc i64 %5 to i32
+  store i32 %conv, i32* %d, align 4
+  %6 = load i32, i32* %d, align 4
+  %7 = load i32, i32* %ret, align 4
+  %add = add nsw i32 %7, %6
+  store i32 %add, i32* %ret, align 4
+  %arraydecay7 = getelementptr inbounds [1 x %struct.__va_list_tag], [1 x %struct.__va_list_tag]* %ap, i32 0, i32 0
+  %__overflow_area_pointer_p8 = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %arraydecay7, i32 0, i32 2
+  %__overflow_area_pointer9 = load i8*, i8** %__overflow_area_pointer_p8
+  %8 = bitcast i8* %__overflow_area_pointer9 to %struct.AAA*
+  %__overflow_area_pointer.next10 = getelementptr i8, i8* %__overflow_area_pointer9, i32 16
+  store i8* %__overflow_area_pointer.next10, i8** %__overflow_area_pointer_p8
+  %9 = bitcast %struct.AAA* %bbb to i8*
+  %10 = bitcast %struct.AAA* %8 to i8*
+  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %9, i8* %10, i32 16, i32 4, i1 false)
+  %d11 = getelementptr inbounds %struct.AAA, %struct.AAA* %bbb, i32 0, i32 3
+  %11 = load i32, i32* %d11, align 4
+  %12 = load i32, i32* %ret, align 4
+  %add12 = add nsw i32 %12, %11
+  store i32 %add12, i32* %ret, align 4
+  %arraydecay13 = getelementptr inbounds [1 x %struct.__va_list_tag], [1 x %struct.__va_list_tag]* %ap, i32 0, i32 0
+  br label %vaarg.maybe_reg14
+
+vaarg.maybe_reg14:                                ; preds = %vaarg.end
+  %__current_saved_reg_area_pointer_p15 = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %arraydecay13, i32 0, i32 0
+  %__current_saved_reg_area_pointer16 = load i8*, i8** %__current_saved_reg_area_pointer_p15
+  %__saved_reg_area_end_pointer_p17 = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %arraydecay13, i32 0, i32 1
+  %__saved_reg_area_end_pointer18 = load i8*, i8** %__saved_reg_area_end_pointer_p17
+  %__new_saved_reg_area_pointer19 = getelementptr i8, i8* %__current_saved_reg_area_pointer16, i32 4
+  %13 = icmp sgt i8* %__new_saved_reg_area_pointer19, %__saved_reg_area_end_pointer18
+  br i1 %13, label %vaarg.on_stack21, label %vaarg.in_reg20
+
+vaarg.in_reg20:                                   ; preds = %vaarg.maybe_reg14
+  %14 = bitcast i8* %__current_saved_reg_area_pointer16 to i32*
+  store i8* %__new_saved_reg_area_pointer19, i8** %__current_saved_reg_area_pointer_p15
+  br label %vaarg.end25
+
+vaarg.on_stack21:                                 ; preds = %vaarg.maybe_reg14
+  %__overflow_area_pointer_p22 = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %arraydecay13, i32 0, i32 2
+  %__overflow_area_pointer23 = load i8*, i8** %__overflow_area_pointer_p22
+  %__overflow_area_pointer.next24 = getelementptr i8, i8* %__overflow_area_pointer23, i32 4
+  store i8* %__overflow_area_pointer.next24, i8** %__overflow_area_pointer_p22
+  store i8* %__overflow_area_pointer.next24, i8** %__current_saved_reg_area_pointer_p15
+  %15 = bitcast i8* %__overflow_area_pointer23 to i32*
+  br label %vaarg.end25
+
+vaarg.end25:                                      ; preds = %vaarg.on_stack21, %vaarg.in_reg20
+  %vaarg.addr26 = phi i32* [ %14, %vaarg.in_reg20 ], [ %15, %vaarg.on_stack21 ]
+  %16 = load i32, i32* %vaarg.addr26
+  store i32 %16, i32* %d, align 4
+  %17 = load i32, i32* %d, align 4
+  %18 = load i32, i32* %ret, align 4
+  %add27 = add nsw i32 %18, %17
+  store i32 %add27, i32* %ret, align 4
+  %arraydecay28 = getelementptr inbounds [1 x %struct.__va_list_tag], [1 x %struct.__va_list_tag]* %ap, i32 0, i32 0
+  br label %vaarg.maybe_reg29
+
+vaarg.maybe_reg29:                                ; preds = %vaarg.end25
+  %__current_saved_reg_area_pointer_p30 = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %arraydecay28, i32 0, i32 0
+  %__current_saved_reg_area_pointer31 = load i8*, i8** %__current_saved_reg_area_pointer_p30
+  %__saved_reg_area_end_pointer_p32 = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %arraydecay28, i32 0, i32 1
+  %__saved_reg_area_end_pointer33 = load i8*, i8** %__saved_reg_area_end_pointer_p32
+  %19 = ptrtoint i8* %__current_saved_reg_area_pointer31 to i32
+  %align_current_saved_reg_area_pointer34 = add i32 %19, 7
+  %align_current_saved_reg_area_pointer35 = and i32 %align_current_saved_reg_area_pointer34, -8
+  %align_current_saved_reg_area_pointer36 = inttoptr i32 %align_current_saved_reg_area_pointer35 to i8*
+  %__new_saved_reg_area_pointer37 = getelementptr i8, i8* %align_current_saved_reg_area_pointer36, i32 8
+  %20 = icmp sgt i8* %__new_saved_reg_area_pointer37, %__saved_reg_area_end_pointer33
+  br i1 %20, label %vaarg.on_stack39, label %vaarg.in_reg38
+
+vaarg.in_reg38:                                   ; preds = %vaarg.maybe_reg29
+  %21 = bitcast i8* %align_current_saved_reg_area_pointer36 to i64*
+  store i8* %__new_saved_reg_area_pointer37, i8** %__current_saved_reg_area_pointer_p30
+  br label %vaarg.end46
+
+vaarg.on_stack39:                                 ; preds = %vaarg.maybe_reg29
+  %__overflow_area_pointer_p40 = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %arraydecay28, i32 0, i32 2
+  %__overflow_area_pointer41 = load i8*, i8** %__overflow_area_pointer_p40
+  %22 = ptrtoint i8* %__overflow_area_pointer41 to i32
+  %align_overflow_area_pointer42 = add i32 %22, 7
+  %align_overflow_area_pointer43 = and i32 %align_overflow_area_pointer42, -8
+  %align_overflow_area_pointer44 = inttoptr i32 %align_overflow_area_pointer43 to i8*
+  %__overflow_area_pointer.next45 = getelementptr i8, i8* %align_overflow_area_pointer44, i32 8
+  store i8* %__overflow_area_pointer.next45, i8** %__overflow_area_pointer_p40
+  store i8* %__overflow_area_pointer.next45, i8** %__current_saved_reg_area_pointer_p30
+  %23 = bitcast i8* %align_overflow_area_pointer44 to i64*
+  br label %vaarg.end46
+
+vaarg.end46:                                      ; preds = %vaarg.on_stack39, %vaarg.in_reg38
+  %vaarg.addr47 = phi i64* [ %21, %vaarg.in_reg38 ], [ %23, %vaarg.on_stack39 ]
+  %24 = load i64, i64* %vaarg.addr47
+  %conv48 = trunc i64 %24 to i32
+  store i32 %conv48, i32* %d, align 4
+  %25 = load i32, i32* %d, align 4
+  %26 = load i32, i32* %ret, align 4
+  %add49 = add nsw i32 %26, %25
+  store i32 %add49, i32* %ret, align 4
+  %arraydecay50 = getelementptr inbounds [1 x %struct.__va_list_tag], [1 x %struct.__va_list_tag]* %ap, i32 0, i32 0
+  %arraydecay5051 = bitcast %struct.__va_list_tag* %arraydecay50 to i8*
+  call void @llvm.va_end(i8* %arraydecay5051)
+  %27 = load i32, i32* %ret, align 4
+  ret i32 %27
+}
+
+; Function Attrs: nounwind
+declare void @llvm.va_start(i8*) #1
+
+; Function Attrs: nounwind
+declare void @llvm.memcpy.p0i8.p0i8.i32(i8* nocapture, i8* nocapture readonly, i32, i32, i1) #1
+
+; Function Attrs: nounwind
+declare void @llvm.va_end(i8*) #1
+
+; Function Attrs: nounwind
+define i32 @main() #0 {
+entry:
+  %retval = alloca i32, align 4
+  %x = alloca i32, align 4
+  %y = alloca i64, align 8
+  store i32 0, i32* %retval
+  store i64 1000000, i64* %y, align 8
+  %0 = load i64, i64* %y, align 8
+  %1 = load i64, i64* %y, align 8
+  %call = call i32 (i32, i32, i32, i32, i32, ...) @foo(i32 1, i32 2, i32 3, i32 4, i32 5, i64 %0, %struct.AAA* byval align 4 @aaa, i32 4, i64 %1)
+  store i32 %call, i32* %x, align 4
+  %2 = load i32, i32* %x, align 4
+  %call1 = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([13 x i8], [13 x i8]* @.str, i32 0, i32 0), i32 %2)
+  %3 = load i32, i32* %x, align 4
+  ret i32 %3
+}
+
+declare i32 @printf(i8*, ...) #2
+
+attributes #0 = { nounwind }
+
+!llvm.ident = !{!0}
+
+!0 = !{!"Clang 3.1"}
diff --git a/llvm/test/CodeGen/Hexagon/vararg_named.ll b/llvm/test/CodeGen/Hexagon/vararg_named.ll
new file mode 100644
--- /dev/null
+++ b/llvm/test/CodeGen/Hexagon/vararg_named.ll
@@ -0,0 +1,211 @@
+; RUN: llc -march=hexagon -mcpu=hexagonv62 -mtriple=hexagon-unknown-linux-musl -O0 < %s | FileCheck %s
+
+; CHECK-LABEL: foo:
+
+; Check Function prologue.
+; Note. All register numbers and offset are fixed.
+; Hence, no need of regular expression.
+
+; CHECK: r29 = add(r29,#-16)
+; CHECK: r7:6 = memd(r29+#16)
+; CHECK: memd(r29+#0) = r7:6
+; CHECK: r7:6 = memd(r29+#24)
+; CHECK: memd(r29+#8) = r7:6
+; CHECK: r7:6 = memd(r29+#32)
+; CHECK: memd(r29+#16) = r7:6
+; CHECK: r7:6 = memd(r29+#40)
+; CHECK: memd(r29+#24) = r7:6
+; CHECK: memw(r29+#36) = r3
+; CHECK: memw(r29+#40) = r4
+; CHECK: memw(r29+#44) = r5
+; CHECK: r29 = add(r29,#16)
+
+%struct.AAA = type { i32, i32, i32, i32 }
+%struct.__va_list_tag = type { i8*, i8*, i8* }
+
+@aaa = global %struct.AAA { i32 100, i32 200, i32 300, i32 400 }, align 4
+@xxx = global %struct.AAA { i32 100, i32 200, i32 300, i32 400 }, align 4
+@yyy = global %struct.AAA { i32 100, i32 200, i32 300, i32 400 }, align 4
+@ccc = global %struct.AAA { i32 10, i32 20, i32 30, i32 40 }, align 4
+@fff = global %struct.AAA { i32 1, i32 2, i32 3, i32 4 }, align 4
+@.str = private unnamed_addr constant [13 x i8] c"result = %d\0A\00", align 1
+
+; Function Attrs: nounwind
+define i32 @foo(i32 %xx, i32 %z, i32 %m, %struct.AAA* byval align 4 %bbb, %struct.AAA* byval align 4 %GGG, ...) #0 {
+entry:
+  %xx.addr = alloca i32, align 4
+  %z.addr = alloca i32, align 4
+  %m.addr = alloca i32, align 4
+  %ap = alloca [1 x %struct.__va_list_tag], align 8
+  %d = alloca i32, align 4
+  %ret = alloca i32, align 4
+  %ddd = alloca %struct.AAA, align 4
+  %ggg = alloca %struct.AAA, align 4
+  %nnn = alloca %struct.AAA, align 4
+  store i32 %xx, i32* %xx.addr, align 4
+  store i32 %z, i32* %z.addr, align 4
+  store i32 %m, i32* %m.addr, align 4
+  store i32 0, i32* %ret, align 4
+  %arraydecay = getelementptr inbounds [1 x %struct.__va_list_tag], [1 x %struct.__va_list_tag]* %ap, i32 0, i32 0
+  %arraydecay1 = bitcast %struct.__va_list_tag* %arraydecay to i8*
+  call void @llvm.va_start(i8* %arraydecay1)
+  %d2 = getelementptr inbounds %struct.AAA, %struct.AAA* %bbb, i32 0, i32 3
+  %0 = load i32, i32* %d2, align 4
+  %1 = load i32, i32* %ret, align 4
+  %add = add nsw i32 %1, %0
+  store i32 %add, i32* %ret, align 4
+  %2 = load i32, i32* %z.addr, align 4
+  %3 = load i32, i32* %ret, align 4
+  %add3 = add nsw i32 %3, %2
+  store i32 %add3, i32* %ret, align 4
+  %arraydecay4 = getelementptr inbounds [1 x %struct.__va_list_tag], [1 x %struct.__va_list_tag]* %ap, i32 0, i32 0
+  br label %vaarg.maybe_reg
+
+vaarg.maybe_reg:                                  ; preds = %entry
+  %__current_saved_reg_area_pointer_p = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %arraydecay4, i32 0, i32 0
+  %__current_saved_reg_area_pointer = load i8*, i8** %__current_saved_reg_area_pointer_p
+  %__saved_reg_area_end_pointer_p = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %arraydecay4, i32 0, i32 1
+  %__saved_reg_area_end_pointer = load i8*, i8** %__saved_reg_area_end_pointer_p
+  %__new_saved_reg_area_pointer = getelementptr i8, i8* %__current_saved_reg_area_pointer, i32 4
+  %4 = icmp sgt i8* %__new_saved_reg_area_pointer, %__saved_reg_area_end_pointer
+  br i1 %4, label %vaarg.on_stack, label %vaarg.in_reg
+
+vaarg.in_reg:                                     ; preds = %vaarg.maybe_reg
+  %5 = bitcast i8* %__current_saved_reg_area_pointer to i32*
+  store i8* %__new_saved_reg_area_pointer, i8** %__current_saved_reg_area_pointer_p
+  br label %vaarg.end
+
+vaarg.on_stack:                                   ; preds = %vaarg.maybe_reg
+  %__overflow_area_pointer_p = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %arraydecay4, i32 0, i32 2
+  %__overflow_area_pointer = load i8*, i8** %__overflow_area_pointer_p
+  %__overflow_area_pointer.next = getelementptr i8, i8* %__overflow_area_pointer, i32 4
+  store i8* %__overflow_area_pointer.next, i8** %__overflow_area_pointer_p
+  store i8* %__overflow_area_pointer.next, i8** %__current_saved_reg_area_pointer_p
+  %6 = bitcast i8* %__overflow_area_pointer to i32*
+  br label %vaarg.end
+
+vaarg.end:                                        ; preds = %vaarg.on_stack, %vaarg.in_reg
+  %vaarg.addr = phi i32* [ %5, %vaarg.in_reg ], [ %6, %vaarg.on_stack ]
+  %7 = load i32, i32* %vaarg.addr
+  store i32 %7, i32* %d, align 4
+  %8 = load i32, i32* %d, align 4
+  %9 = load i32, i32* %ret, align 4
+  %add5 = add nsw i32 %9, %8
+  store i32 %add5, i32* %ret, align 4
+  %arraydecay6 = getelementptr inbounds [1 x %struct.__va_list_tag], [1 x %struct.__va_list_tag]* %ap, i32 0, i32 0
+  %__overflow_area_pointer_p7 = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %arraydecay6, i32 0, i32 2
+  %__overflow_area_pointer8 = load i8*, i8** %__overflow_area_pointer_p7
+  %10 = bitcast i8* %__overflow_area_pointer8 to %struct.AAA*
+  %__overflow_area_pointer.next9 = getelementptr i8, i8* %__overflow_area_pointer8, i32 16
+  store i8* %__overflow_area_pointer.next9, i8** %__overflow_area_pointer_p7
+  %11 = bitcast %struct.AAA* %ddd to i8*
+  %12 = bitcast %struct.AAA* %10 to i8*
+  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %11, i8* %12, i32 16, i32 4, i1 false)
+  %d10 = getelementptr inbounds %struct.AAA, %struct.AAA* %ddd, i32 0, i32 3
+  %13 = load i32, i32* %d10, align 4
+  %14 = load i32, i32* %ret, align 4
+  %add11 = add nsw i32 %14, %13
+  store i32 %add11, i32* %ret, align 4
+  %arraydecay12 = getelementptr inbounds [1 x %struct.__va_list_tag], [1 x %struct.__va_list_tag]* %ap, i32 0, i32 0
+  %__overflow_area_pointer_p13 = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %arraydecay12, i32 0, i32 2
+  %__overflow_area_pointer14 = load i8*, i8** %__overflow_area_pointer_p13
+  %15 = bitcast i8* %__overflow_area_pointer14 to %struct.AAA*
+  %__overflow_area_pointer.next15 = getelementptr i8, i8* %__overflow_area_pointer14, i32 16
+  store i8* %__overflow_area_pointer.next15, i8** %__overflow_area_pointer_p13
+  %16 = bitcast %struct.AAA* %ggg to i8*
+  %17 = bitcast %struct.AAA* %15 to i8*
+  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %16, i8* %17, i32 16, i32 4, i1 false)
+  %d16 = getelementptr inbounds %struct.AAA, %struct.AAA* %ggg, i32 0, i32 3
+  %18 = load i32, i32* %d16, align 4
+  %19 = load i32, i32* %ret, align 4
+  %add17 = add nsw i32 %19, %18
+  store i32 %add17, i32* %ret, align 4
+  %arraydecay18 = getelementptr inbounds [1 x %struct.__va_list_tag], [1 x %struct.__va_list_tag]* %ap, i32 0, i32 0
+  %__overflow_area_pointer_p19 = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %arraydecay18, i32 0, i32 2
+  %__overflow_area_pointer20 = load i8*, i8** %__overflow_area_pointer_p19
+  %20 = bitcast i8* %__overflow_area_pointer20 to %struct.AAA*
+  %__overflow_area_pointer.next21 = getelementptr i8, i8* %__overflow_area_pointer20, i32 16
+  store i8* %__overflow_area_pointer.next21, i8** %__overflow_area_pointer_p19
+  %21 = bitcast %struct.AAA* %nnn to i8*
+  %22 = bitcast %struct.AAA* %20 to i8*
+  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %21, i8* %22, i32 16, i32 4, i1 false)
+  %d22 = getelementptr inbounds %struct.AAA, %struct.AAA* %nnn, i32 0, i32 3
+  %23 = load i32, i32* %d22, align 4
+  %24 = load i32, i32* %ret, align 4
+  %add23 = add nsw i32 %24, %23
+  store i32 %add23, i32* %ret, align 4
+  %arraydecay24 = getelementptr inbounds [1 x %struct.__va_list_tag], [1 x %struct.__va_list_tag]* %ap, i32 0, i32 0
+  br label %vaarg.maybe_reg25
+
+vaarg.maybe_reg25:                                ; preds = %vaarg.end
+  %__current_saved_reg_area_pointer_p26 = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %arraydecay24, i32 0, i32 0
+  %__current_saved_reg_area_pointer27 = load i8*, i8** %__current_saved_reg_area_pointer_p26
+  %__saved_reg_area_end_pointer_p28 = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %arraydecay24, i32 0, i32 1
+  %__saved_reg_area_end_pointer29 = load i8*, i8** %__saved_reg_area_end_pointer_p28
+  %__new_saved_reg_area_pointer30 = getelementptr i8, i8* %__current_saved_reg_area_pointer27, i32 4
+  %25 = icmp sgt i8* %__new_saved_reg_area_pointer30, %__saved_reg_area_end_pointer29
+  br i1 %25, label %vaarg.on_stack32, label %vaarg.in_reg31
+
+vaarg.in_reg31:                                   ; preds = %vaarg.maybe_reg25
+  %26 = bitcast i8* %__current_saved_reg_area_pointer27 to i32*
+  store i8* %__new_saved_reg_area_pointer30, i8** %__current_saved_reg_area_pointer_p26
+  br label %vaarg.end36
+
+vaarg.on_stack32:                                 ; preds = %vaarg.maybe_reg25
+  %__overflow_area_pointer_p33 = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %arraydecay24, i32 0, i32 2
+  %__overflow_area_pointer34 = load i8*, i8** %__overflow_area_pointer_p33
+  %__overflow_area_pointer.next35 = getelementptr i8, i8* %__overflow_area_pointer34, i32 4
+  store i8* %__overflow_area_pointer.next35, i8** %__overflow_area_pointer_p33
+  store i8* %__overflow_area_pointer.next35, i8** %__current_saved_reg_area_pointer_p26
+  %27 = bitcast i8* %__overflow_area_pointer34 to i32*
+  br label %vaarg.end36
+
+vaarg.end36:                                      ; preds = %vaarg.on_stack32, %vaarg.in_reg31
+  %vaarg.addr37 = phi i32* [ %26, %vaarg.in_reg31 ], [ %27, %vaarg.on_stack32 ]
+  %28 = load i32, i32* %vaarg.addr37
+  store i32 %28, i32* %d, align 4
+  %29 = load i32, i32* %d, align 4
+  %30 = load i32, i32* %ret, align 4
+  %add38 = add nsw i32 %30, %29
+  store i32 %add38, i32* %ret, align 4
+  %31 = load i32, i32* %m.addr, align 4
+  %32 = load i32, i32* %ret, align 4
+  %add39 = add nsw i32 %32, %31
+  store i32 %add39, i32* %ret, align 4
+  %arraydecay40 = getelementptr inbounds [1 x %struct.__va_list_tag], [1 x %struct.__va_list_tag]* %ap, i32 0, i32 0
+  %arraydecay4041 = bitcast %struct.__va_list_tag* %arraydecay40 to i8*
+  call void @llvm.va_end(i8* %arraydecay4041)
+  %33 = load i32, i32* %ret, align 4
+  ret i32 %33
+}
+
+; Function Attrs: nounwind
+declare void @llvm.va_start(i8*) #1
+
+; Function Attrs: nounwind
+declare void @llvm.memcpy.p0i8.p0i8.i32(i8* nocapture, i8* nocapture readonly, i32, i32, i1) #1
+
+; Function Attrs: nounwind
+declare void @llvm.va_end(i8*) #1
+
+; Function Attrs: nounwind
+define i32 @main() #0 {
+entry:
+  %retval = alloca i32, align 4
+  %x = alloca i32, align 4
+  store i32 0, i32* %retval
+  %call = call i32 (i32, i32, i32, %struct.AAA*, %struct.AAA*, ...) @foo(i32 1, i32 3, i32 5, %struct.AAA* byval align 4 @aaa, %struct.AAA* byval align 4 @fff, i32 2, %struct.AAA* byval align 4 @xxx, %struct.AAA* byval align 4 @yyy, %struct.AAA* byval align 4 @ccc, i32 4)
+  store i32 %call, i32* %x, align 4
+  %0 = load i32, i32* %x, align 4
+  %call1 = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([13 x i8], [13 x i8]* @.str, i32 0, i32 0), i32 %0)
+  %1 = load i32, i32* %x, align 4
+  ret i32 %1
+}
+
+declare i32 @printf(i8*, ...) #2
+
+attributes #0 = { nounwind }
+
+!llvm.ident = !{!0}
+
+!0 = !{!"Clang 3.1"}