Add basic ability to setup call frame, and make procedure calls.

Hello world will compile and execute with this patch.

Patch by Reed Kotler.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@160651 91177308-0d34-0410-b5e6-96231b3b80d8

11 files changed, 338 insertions, 44 deletions

diff --git a/lib/Target/Mips/Mips16InstrFormats.td b/lib/Target/Mips/Mips16InstrFormats.td
index 6babee918d..61602b62fb 100644
--- a/lib/Target/Mips/Mips16InstrFormats.td
+++ b/lib/Target/Mips/Mips16InstrFormats.td
@@ -529,14 +529,16 @@ class FEXT_RRI16<bits<5> _op, dag outs, dag ins, string asmstr,
                  list<dag> pattern, InstrItinClass itin>:
   MipsInst16_EXTEND<outs, ins, asmstr, pattern, itin, FrmEXT_RRI16>
 {
+  bits<5> op;
   bits<16> imm16;
   bits<3> rx;
   bits<3> ry;
 
+  let op=_op;
 
   let Inst{26-21} = imm16{10-5};
   let Inst{20-16} = imm16{15-11};
-  let Inst{15-11} = _op;
+  let Inst{15-11} = op;
   let Inst{10-8} = rx;
   let Inst{7-5} = ry;
   let Inst{4-0} = imm16{4-0};
diff --git a/lib/Target/Mips/Mips16InstrInfo.td b/lib/Target/Mips/Mips16InstrInfo.td
index 71801ca6f2..c85204278b 100644
--- a/lib/Target/Mips/Mips16InstrInfo.td
+++ b/lib/Target/Mips/Mips16InstrInfo.td
@@ -11,35 +11,209 @@
 //
 //===----------------------------------------------------------------------===//
 
+def uimm5      : Operand<i8> {
+  let DecoderMethod= "DecodeSimm16";
+}
+
+//
+// RRR-type instruction format
+//
+
+class FRRR16_ins<bits<2> _f, string asmstr,  InstrItinClass itin> :
+  FRRR16<_f, (outs CPU16Regs:$rz), (ins CPU16Regs:$rx, CPU16Regs:$ry),
+         !strconcat(asmstr, "\t$rz, $rx, $ry"), [], itin>;
+
+//
+// I8_MOV32R instruction format (used only by MOV32R instruction)
+//
+class FI8_MOV32R16_ins<string asmstr, InstrItinClass itin>:
+  FI8_MOV32R16<(outs CPURegs:$r32), (ins CPU16Regs:$rz),
+               !strconcat(asmstr,  "\t$r32, $rz"), [], itin>;
+
+//
+// EXT-RI instruction format
+//
+
+class FEXT_RI16_ins_base<bits<5> _op, string asmstr, string asmstr2,
+                         InstrItinClass itin>:
+  FEXT_RI16<_op, (outs CPU16Regs:$rx), (ins simm16:$imm),
+                  !strconcat(asmstr, asmstr2), [], itin>;
+
+class FEXT_RI16_ins<bits<5> _op, string asmstr,
+                    InstrItinClass itin>:
+  FEXT_RI16_ins_base<_op, asmstr, "\t$rx, $imm", itin>;
+
+class FEXT_RI16_PC_ins<bits<5> _op, string asmstr, InstrItinClass itin>:
+  FEXT_RI16_ins_base<_op, asmstr, "\t$rx, $$pc, $imm", itin>;
+
+//
+// RR-type instruction format
+//
+let rx=0 in
+class FRR16_JALRC_RA_only_ins<bits<1> nd_, bits<1> l_,
+                              string asmstr, InstrItinClass itin>:
+  FRR16_JALRC<nd_, l_, 1, (outs), (ins), !strconcat(asmstr, "\t $$ra"),
+              [], itin> ;
+
+//
+// EXT-RRI instruction format
+//
+
+class FEXT_RRI16_mem_ins<bits<5> op, string asmstr, Operand MemOpnd,
+                         InstrItinClass itin>:
+  FEXT_RRI16<op, (outs CPU16Regs:$ry), (ins  MemOpnd:$addr),
+             !strconcat(asmstr, "\t$ry, $addr"), [], itin>;
+
+//
+// EXT-SHIFT instruction format
+//
+class FEXT_SHIFT16_ins<bits<2> _f, string asmstr, InstrItinClass itin>:
+  FEXT_SHIFT16<_f, (outs CPU16Regs:$rx), (ins CPU16Regs:$ry, uimm5:$sa),
+               !strconcat(asmstr, "\t$rx, $ry, $sa"), [], itin>;
+
+//
+// Address operand
+def mem16 : Operand<i32> {
+  let PrintMethod = "printMemOperand";
+  let MIOperandInfo = (ops CPU16Regs, simm16);
+  let EncoderMethod = "getMemEncoding";
+}
+
+//
+// Format: ADDIU rx, pc, immediate MIPS16e
+// Purpose: Add Immediate Unsigned Word (3-Operand, PC-Relative, Extended)
+// To add a constant to the program counter.
+//
+class AddiuRxPcImmX16_base : FEXT_RI16_PC_ins<0b00001, "addiu", IIAlu>;
+def AddiuRxPcImmX16   : AddiuRxPcImmX16_base;
+//
+// Format: ADDU rz, rx, ry MIPS16e
+// Purpose: Add Unsigned Word (3-Operand)
+// To add 32-bit integers.
+//
+
+class AdduRxRyRz16_base: FRRR16_ins<01, "addu", IIAlu>;
+def AdduRxRyRz16: AdduRxRyRz16_base;
+
+//
+// Format: JR ra MIPS16e
+// Purpose: Jump Register Through Register ra
+// To execute a branch to the instruction address in the return
+// address register.
+//
+
+def JrRa16: FRR16_JALRC_RA_only_ins<0, 0, "jr", IIAlu>;
+
+//
+// Format: LI rx, immediate MIPS16e
+// Purpose: Load Immediate (Extended)
+// To load a constant into a GPR.
+//
+def LiRxImmX16: FEXT_RI16_ins<0b01101, "li", IIAlu>;
+
+//
+// Format: LW ry, offset(rx) MIPS16e
+// Purpose: Load Word (Extended)
+// To load a word from memory as a signed value.
+//
+class LwRxRyOffMemX16_base: FEXT_RRI16_mem_ins<0b10011, "lw", mem16, IIAlu>;
+def LwRxRyOffMemX16: LwRxRyOffMemX16_base;
+
+//
+// Format: MOVE r32, rz MIPS16e
+// Purpose: Move
+// To move the contents of a GPR to a GPR.
+//
+def Mov32R16: FI8_MOV32R16_ins<"move", IIAlu>;
+//
+// Format: RESTORE {ra,}{s0/s1/s0-1,}{framesize}
+// (All args are optional) MIPS16e
+// Purpose: Restore Registers and Deallocate Stack Frame
+// To deallocate a stack frame before exit from a subroutine,
+// restoring return address and static registers, and adjusting
+// stack
+//
+
+// fixed form for restoring RA and the frame
+// for direct object emitter, encoding needs to be adjusted for the
+// frame size
+//
+let ra=1, s=0,s0=0,s1=0 in
+def RestoreRaF16:
+  FI8_SVRS16<0b1, (outs), (ins uimm16:$frame_size),
+             "restore \t$$ra, $frame_size", [], IILoad >;
+
+//
+// Format: SAVE {ra,}{s0/s1/s0-1,}{framesize} (All arguments are optional)
+// MIPS16e
+// Purpose: Save Registers and Set Up Stack Frame
+// To set up a stack frame on entry to a subroutine,
+// saving return address and static registers, and adjusting stack
+//
+let ra=1, s=1,s0=0,s1=0 in
+def SaveRaF16:
+  FI8_SVRS16<0b1, (outs), (ins uimm16:$frame_size),
+             "save \t$$ra, $frame_size", [], IILoad >;
+
+//
+// Format: SLL rx, ry, sa MIPS16e
+// Purpose: Shift Word Left Logical (Extended)
+// To execute a left-shift of a word by a fixed number of bits—0 to 31 bits.
+//
+def SllX16: FEXT_SHIFT16_ins<0b00, "sll", IIAlu>;
+
+//
+// Format: SW ry, offset(rx) MIPS16e
+// Purpose: Store Word (Extended)
+// To store a word to memory.
+//
+class SwRxRyOffMemX16_base: FEXT_RRI16_mem_ins<0b11011, "sw", mem16, IIAlu>;
+def SwRxRyOffMemX16: SwRxRyOffMemX16_base;
+
 class Mips16Pat<dag pattern, dag result> : Pat<pattern, result> {
   let Predicates = [InMips16Mode];
 }
 
-// Mips16 pseudos
-let isReturn=1, isTerminator=1, hasDelaySlot=1, isBarrier=1, hasCtrlDep=1,
-    hasExtraSrcRegAllocReq = 1 in
-def RetRA16 : MipsPseudo16<(outs), (ins), "", [(MipsRet)]>;
+class ArithLogicR16Defs<SDNode OpNode, bit isComm = 0> {
+  dag OutOperandList = (outs CPU16Regs:$rz);
+  dag InOperandList = (ins CPU16Regs:$rx, CPU16Regs:$ry);
+  list<dag> Pattern = [(set CPU16Regs:$rz,
+                       (OpNode CPU16Regs:$rx, CPU16Regs:$ry))];
+}
 
-def LI16E      : FEXT_RI16<0b01101, (outs CPU16Regs:$rx),
-                           (ins uimm16:$amt),
-                           !strconcat("li", "\t$rx, $amt"),
-                           [(set CPU16Regs:$rx, immZExt16:$amt )],IILoad>;
+multiclass ArithLogicR16_base {
+  def _add: AdduRxRyRz16_base, ArithLogicR16Defs<add, 1>;
+}
 
-let isReturn=1, isTerminator=1, hasDelaySlot=1, isCodeGenOnly=1,
-    isBarrier=1, hasCtrlDep=1, rx=0  in
-def RET16 : FRR16_JALRC
-  <0,0,0, (outs), (ins CPURAReg:$target), "jr\t$target", [], IIBranch>;
+defm ArithLogicR16_patt : ArithLogicR16_base;
 
-// As stack alignment is always done with addiu, we need a 16-bit immediate
-let Defs = [SP], Uses = [SP] in {
-def ADJCALLSTACKDOWN16 : MipsPseudo16<(outs), (ins uimm16:$amt),
-                                      "!ADJCALLSTACKDOWN $amt",
-                                      [(callseq_start timm:$amt)]>;
-def ADJCALLSTACKUP16   : MipsPseudo16<(outs), (ins uimm16:$amt1, uimm16:$amt2),
-                                      "!ADJCALLSTACKUP $amt1",
-                                      [(callseq_end timm:$amt1, timm:$amt2)]>;
+class LoadM16Defs<PatFrag OpNode, Operand _MemOpnd, bit Pseudo=0> {
+  bit isPseudo = Pseudo;
+  Operand MemOpnd = _MemOpnd;
+  dag OutOperandList = (outs CPU16Regs:$ry);
+  dag InOperandList = (ins MemOpnd:$addr);
+  list<dag> Pattern = [(set CPU16Regs:$ry, (OpNode addr:$addr))];
+}
+
+multiclass LoadM16_base {
+  def _LwRxRyOffMemX16: LwRxRyOffMemX16_base, LoadM16Defs<load_a, mem16>;
+}
+
+defm LoadM16: LoadM16_base;
+
+class StoreM16Defs<PatFrag OpNode, Operand _MemOpnd, bit Pseudo=0> {
+  bit isPseudo = Pseudo;
+  Operand MemOpnd = _MemOpnd;
+  dag OutOperandList = (outs );
+  dag InOperandList = (ins CPU16Regs:$ry, MemOpnd:$addr);
+  list<dag> Pattern = [(OpNode CPU16Regs:$ry, addr:$addr)];
+}
+
+multiclass StoreM16_base {
+  def _SwRxRyOffMemX16: SwRxRyOffMemX16_base, StoreM16Defs<store_a, mem16>;
 }
 
+defm StoreM16: StoreM16_base;
 
 // Jump and Link (Call)
 let isCall=1, hasDelaySlot=1 in
@@ -47,6 +221,23 @@ def JumpLinkReg16:
   FRR16_JALRC<0, 0, 0, (outs), (ins CPU16Regs:$rs),
               "jalr \t$rs", [(MipsJmpLink CPU16Regs:$rs)], IIBranch>;
 
+// Mips16 pseudos
+let isReturn=1, isTerminator=1, hasDelaySlot=1, isBarrier=1, hasCtrlDep=1,
+  hasExtraSrcRegAllocReq = 1 in
+def RetRA16 : MipsPseudo16<(outs), (ins), "", [(MipsRet)]>;
+
+// As stack alignment is always done with addiu, we need a 16-bit immediate
+// This is basically deprecated code but needs to be there for things
+// to work.
+let Defs = [SP], Uses = [SP] in {
+def ADJCALLSTACKDOWN16 : MipsPseudo16<(outs), (ins uimm16:$amt),
+                                      ";",
+                                      [(callseq_start timm:$amt)]>;
+def ADJCALLSTACKUP16   : MipsPseudo16<(outs), (ins uimm16:$amt1, uimm16:$amt2),
+                                      ";",
+                                      [(callseq_end timm:$amt1, timm:$amt2)]>;
+}
 
 // Small immediates
-def : Mips16Pat<(i32 immZExt16:$in), (LI16E immZExt16:$in)>;
+def : Mips16Pat<(i32 immZExt16:$in), (LiRxImmX16 immZExt16:$in)>;
+def : Mips16Pat<(MipsLo tglobaladdr:$in), (LiRxImmX16 tglobaladdr:$in)>;
diff --git a/lib/Target/Mips/MipsFrameLowering.cpp b/lib/Target/Mips/MipsFrameLowering.cpp
index cb55995404..6338f3c4ea 100644
--- a/lib/Target/Mips/MipsFrameLowering.cpp
+++ b/lib/Target/Mips/MipsFrameLowering.cpp
@@ -114,7 +114,7 @@ void MipsFrameLowering::emitPrologue(MachineFunction &MF) const {
   unsigned StackAlign = getStackAlignment();
   uint64_t StackSize = RoundUpToAlignment(MFI->getStackSize(), StackAlign);
 
-  if (MipsFI->globalBaseRegSet()) 
+  if (MipsFI->globalBaseRegSet())
     StackSize += MFI->getObjectOffset(MipsFI->getGlobalRegFI()) + StackAlign;
   else
     StackSize += RoundUpToAlignment(MipsFI->getMaxCallFrameSize(), StackAlign);
@@ -130,8 +130,13 @@ void MipsFrameLowering::emitPrologue(MachineFunction &MF) const {
   MachineLocation DstML, SrcML;
 
   // Adjust stack.
-  if (isInt<16>(-StackSize)) // addi sp, sp, (-stacksize)
-    BuildMI(MBB, MBBI, dl, TII.get(ADDiu), SP).addReg(SP).addImm(-StackSize);
+  if (isInt<16>(-StackSize)) {// addi sp, sp, (-stacksize)
+    if (STI.inMips16Mode())
+      BuildMI(MBB, MBBI, dl,
+              TII.get(Mips::SaveRaF16)).addImm(StackSize); // cleanup
+    else
+      BuildMI(MBB, MBBI, dl, TII.get(ADDiu), SP).addReg(SP).addImm(-StackSize);
+  }
   else { // Expand immediate that doesn't fit in 16-bit.
     unsigned ATReg = STI.isABI_N64() ? Mips::AT_64 : Mips::AT;
 
@@ -237,8 +242,14 @@ void MipsFrameLowering::emitEpilogue(MachineFunction &MF,
     return;
 
   // Adjust stack.
-  if (isInt<16>(StackSize)) // addi sp, sp, (-stacksize)
-    BuildMI(MBB, MBBI, dl, TII.get(ADDiu), SP).addReg(SP).addImm(StackSize);
+  if (isInt<16>(StackSize)) { // addi sp, sp, (-stacksize)
+    if (STI.inMips16Mode())
+      // assumes stacksize multiple of 8
+      BuildMI(MBB, MBBI, dl,
+              TII.get(Mips::RestoreRaF16)).addImm(StackSize);
+    else
+      BuildMI(MBB, MBBI, dl, TII.get(ADDiu), SP).addReg(SP).addImm(StackSize);
+  }
   else { // Expand immediate that doesn't fit in 16-bit.
     unsigned ATReg = STI.isABI_N64() ? Mips::AT_64 : Mips::AT;
 
diff --git a/lib/Target/Mips/MipsISelDAGToDAG.cpp b/lib/Target/Mips/MipsISelDAGToDAG.cpp
index c409e57021..ea33b74a75 100644
--- a/lib/Target/Mips/MipsISelDAGToDAG.cpp
+++ b/lib/Target/Mips/MipsISelDAGToDAG.cpp
@@ -128,18 +128,21 @@ void MipsDAGToDAGISel::InitGlobalBaseReg(MachineFunction &MF) {
   const MipsInstrInfo *MII = TM.getInstrInfo();
   const TargetInstrInfo &TII = *MF.getTarget().getInstrInfo();
   DebugLoc DL = I != MBB.end() ? I->getDebugLoc() : DebugLoc();
-  unsigned V0, V1, GlobalBaseReg = MipsFI->getGlobalBaseReg();
-  int FI = 0; 
+  unsigned V0, V1, V2, GlobalBaseReg = MipsFI->getGlobalBaseReg();
+  int FI = 0;
 
-  if (!Subtarget.inMips16Mode())
-    FI= MipsFI->initGlobalRegFI();
+  FI= MipsFI->initGlobalRegFI();
 
   const TargetRegisterClass *RC = Subtarget.isABI_N64() ?
     (const TargetRegisterClass*)&Mips::CPU64RegsRegClass :
     (const TargetRegisterClass*)&Mips::CPURegsRegClass;
 
+  if (Subtarget.inMips16Mode())
+    RC=(const TargetRegisterClass*)&Mips::CPU16RegsRegClass;
+
   V0 = RegInfo.createVirtualRegister(RC);
   V1 = RegInfo.createVirtualRegister(RC);
+  V2 = RegInfo.createVirtualRegister(RC);
 
   if (Subtarget.isABI_N64()) {
     MF.getRegInfo().addLiveIn(Mips::T9_64);
@@ -160,6 +163,21 @@ void MipsDAGToDAGISel::InitGlobalBaseReg(MachineFunction &MF) {
     return;
   }
 
+  if (Subtarget.inMips16Mode()) {
+    BuildMI(MBB, I, DL, TII.get(Mips::LiRxImmX16), V0)
+        .addExternalSymbol("_gp_disp", MipsII::MO_ABS_HI);
+    BuildMI(MBB, I, DL, TII.get(Mips::AddiuRxPcImmX16),
+            V1)
+        .addExternalSymbol("_gp_disp", MipsII::MO_ABS_LO);
+    BuildMI(MBB, I, DL, TII.get(Mips::SllX16),
+            V2 ).addReg(V0).addImm(16);
+    BuildMI(MBB, I, DL, TII.get(Mips::AdduRxRyRz16), GlobalBaseReg)
+      .addReg(V1).addReg(V2);
+
+
+    return;
+  }
+
   if (MF.getTarget().getRelocationModel() == Reloc::Static) {
     // Set global register to __gnu_local_gp.
     //
@@ -169,8 +187,6 @@ void MipsDAGToDAGISel::InitGlobalBaseReg(MachineFunction &MF) {
       .addExternalSymbol("__gnu_local_gp", MipsII::MO_ABS_HI);
     BuildMI(MBB, I, DL, TII.get(Mips::ADDiu), GlobalBaseReg).addReg(V0)
       .addExternalSymbol("__gnu_local_gp", MipsII::MO_ABS_LO);
-    MII->storeRegToStackSlot(MBB, I, GlobalBaseReg, false, FI, RC,
-                             TargetRegInfo);
     return;
   }
 
@@ -194,8 +210,10 @@ void MipsDAGToDAGISel::InitGlobalBaseReg(MachineFunction &MF) {
 
   assert(Subtarget.isABI_O32());
 
-  if (Subtarget.inMips16Mode())
-    return; // no need to load GP. It can be calculated anywhere
+
+  //if (Subtarget.inMips16Mode())
+  //  return; // no need to load GP. It can be calculated anywhere
+
 
 
   // For O32 ABI, the following instruction sequence is emitted to initialize
diff --git a/lib/Target/Mips/MipsISelLowering.cpp b/lib/Target/Mips/MipsISelLowering.cpp
index 6b80e3fe3e..7741f9f982 100644
--- a/lib/Target/Mips/MipsISelLowering.cpp
+++ b/lib/Target/Mips/MipsISelLowering.cpp
@@ -2645,6 +2645,7 @@ MipsTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
   SmallVector<ISD::InputArg, 32> &Ins   = CLI.Ins;
   SDValue InChain                       = CLI.Chain;
   SDValue Callee                        = CLI.Callee;
+  SDValue CalleeSave                    = CLI.Callee;
   bool &isTailCall                      = CLI.IsTailCall;
   CallingConv::ID CallConv              = CLI.CallConv;
   bool isVarArg                         = CLI.IsVarArg;
@@ -2901,7 +2902,7 @@ MipsTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
   SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Glue);
   SmallVector<SDValue, 8> Ops;
   Ops.push_back(Chain);
-  Ops.push_back(Callee);
+  Ops.push_back(Subtarget->inMips16Mode()? CalleeSave: Callee);
 
   // Add argument registers to the end of the list so that they are
   // known live into the call.
@@ -2909,6 +2910,8 @@ MipsTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
     Ops.push_back(DAG.getRegister(RegsToPass[i].first,
                                   RegsToPass[i].second.getValueType()));
 
+  if (Subtarget->inMips16Mode())
+    Ops.push_back(Callee);
   // Add a register mask operand representing the call-preserved registers.
   const TargetRegisterInfo *TRI = getTargetMachine().getRegisterInfo();
   const uint32_t *Mask = TRI->getCallPreservedMask(CallConv);
diff --git a/lib/Target/Mips/MipsInstrInfo.cpp b/lib/Target/Mips/MipsInstrInfo.cpp
index 7d8488f0b8..458e4f7d1f 100644
--- a/lib/Target/Mips/MipsInstrInfo.cpp
+++ b/lib/Target/Mips/MipsInstrInfo.cpp
@@ -30,6 +30,7 @@ using namespace llvm;
 MipsInstrInfo::MipsInstrInfo(MipsTargetMachine &tm)
   : MipsGenInstrInfo(Mips::ADJCALLSTACKDOWN, Mips::ADJCALLSTACKUP),
     TM(tm), IsN64(TM.getSubtarget<MipsSubtarget>().isABI_N64()),
+    InMips16Mode(TM.getSubtarget<MipsSubtarget>().inMips16Mode()),
     RI(*TM.getSubtargetImpl(), *this),
     UncondBrOpc(TM.getRelocationModel() == Reloc::PIC_ ? Mips::B : Mips::J) {}
 
@@ -107,8 +108,13 @@ copyPhysReg(MachineBasicBlock &MBB,
   unsigned Opc = 0, ZeroReg = 0;
 
   if (Mips::CPURegsRegClass.contains(DestReg)) { // Copy to CPU Reg.
-    if (Mips::CPURegsRegClass.contains(SrcReg))
-      Opc = Mips::ADDu, ZeroReg = Mips::ZERO;
+    if (Mips::CPURegsRegClass.contains(SrcReg)) {
+      if (InMips16Mode)
+        Opc=Mips::Mov32R16;
+      else {
+        Opc = Mips::ADDu, ZeroReg = Mips::ZERO;
+      }
+    }
     else if (Mips::CCRRegClass.contains(SrcReg))
       Opc = Mips::CFC1;
     else if (Mips::FGR32RegClass.contains(SrcReg))
@@ -240,6 +246,12 @@ void MipsInstrInfo::ExpandRetRA(MachineBasicBlock &MBB,
     .addReg(Mips::RA);
 }
 
+void MipsInstrInfo::ExpandRetRA16(MachineBasicBlock &MBB,
+                                MachineBasicBlock::iterator I,
+                                unsigned Opc) const {
+  BuildMI(MBB, I, I->getDebugLoc(), TM.getInstrInfo()->get(Opc));
+}
+
 void MipsInstrInfo::ExpandExtractElementF64(MachineBasicBlock &MBB,
                                           MachineBasicBlock::iterator I) const {
   const TargetInstrInfo *TII = TM.getInstrInfo();
@@ -283,7 +295,7 @@ bool MipsInstrInfo::expandPostRAPseudo(MachineBasicBlock::iterator MI) const {
     ExpandRetRA(MBB, MI, Mips::RET);
     break;
   case Mips::RetRA16:
-    ExpandRetRA(MBB, MI, Mips::RET16);
+    ExpandRetRA16(MBB, MI, Mips::JrRa16);
     break;
   case Mips::BuildPairF64:
     ExpandBuildPairF64(MBB, MI);
diff --git a/lib/Target/Mips/MipsInstrInfo.h b/lib/Target/Mips/MipsInstrInfo.h
index b4721a7b39..358f8173a0 100644
--- a/lib/Target/Mips/MipsInstrInfo.h
+++ b/lib/Target/Mips/MipsInstrInfo.h
@@ -27,7 +27,7 @@ namespace llvm {
 
 class MipsInstrInfo : public MipsGenInstrInfo {
   MipsTargetMachine &TM;
-  bool IsN64;
+  bool IsN64; bool InMips16Mode;
   const MipsRegisterInfo RI;
   unsigned UncondBrOpc;
 public:
@@ -65,6 +65,9 @@ public:
 private:
   void ExpandRetRA(MachineBasicBlock &MBB, MachineBasicBlock::iterator I,
                    unsigned Opc) const;
+  void ExpandRetRA16(MachineBasicBlock &MBB, MachineBasicBlock::iterator I,
+                   unsigned Opc) const;
+
   void BuildCondBr(MachineBasicBlock &MBB, MachineBasicBlock *TBB, DebugLoc DL,
                    const SmallVectorImpl<MachineOperand>& Cond) const;
   void ExpandExtractElementF64(MachineBasicBlock &MBB,
diff --git a/lib/Target/Mips/MipsMachineFunction.cpp b/lib/Target/Mips/MipsMachineFunction.cpp
index e3746d4a42..362173eda3 100644
--- a/lib/Target/Mips/MipsMachineFunction.cpp
+++ b/lib/Target/Mips/MipsMachineFunction.cpp
@@ -33,10 +33,13 @@ unsigned MipsFunctionInfo::getGlobalBaseReg() {
 
   const MipsSubtarget &ST = MF.getTarget().getSubtarget<MipsSubtarget>();
 
-  const TargetRegisterClass *RC = ST.isABI_N64() ?
-    (const TargetRegisterClass*)&Mips::CPU64RegsRegClass :
-    (const TargetRegisterClass*)&Mips::CPURegsRegClass;
-
+  const TargetRegisterClass *RC;
+  if (ST.inMips16Mode())
+    RC=(const TargetRegisterClass*)&Mips::CPU16RegsRegClass;
+  else
+    RC = ST.isABI_N64() ?
+      (const TargetRegisterClass*)&Mips::CPU64RegsRegClass :
+      (const TargetRegisterClass*)&Mips::CPURegsRegClass;
   return GlobalBaseReg = MF.getRegInfo().createVirtualRegister(RC);
 }
 
diff --git a/test/CodeGen/Mips/helloworld_call.ll b/test/CodeGen/Mips/helloworld_call.ll
new file mode 100644
index 0000000000..61c0446083
--- /dev/null
+++ b/test/CodeGen/Mips/helloworld_call.ll
@@ -0,0 +1,17 @@
+; RUN: llc  -march=mipsel -mcpu=mips16 -relocation-model=pic -O3 < %s | FileCheck %s -check-prefix=16
+
+@.str = private unnamed_addr constant [13 x i8] c"hello world\0A\00", align 1
+
+define i32 @main() nounwind {
+entry:
+  %call = call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([13 x i8]* @.str, i32 0, i32 0))
+  ret i32 0
+
+; 16: 	.set	mips16                  # @main
+
+; 16:	move 	$gp, ${{[0-9]+}}
+; 16:	jalr 	${{[0-9]+}}
+; 16:	li	$2, 0
+}
+
+declare i32 @printf(i8*, ...)
diff --git a/test/CodeGen/Mips/helloworld_pe.ll b/test/CodeGen/Mips/helloworld_pe.ll
new file mode 100644
index 0000000000..ba46662d52
--- /dev/null
+++ b/test/CodeGen/Mips/helloworld_pe.ll
@@ -0,0 +1,18 @@
+; RUN: llc  -march=mipsel -mcpu=mips16 -relocation-model=pic -O3 < %s | FileCheck %s -check-prefix=16
+
+@.str = private unnamed_addr constant [13 x i8] c"hello world\0A\00", align 1
+
+define i32 @main() nounwind {
+entry:
+  %call = call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([13 x i8]* @.str, i32 0, i32 0))
+  ret i32 0
+
+; 16: 	.set	mips16                  # @main
+
+; 16:	li	$[[T1:[0-9]+]], %hi(_gp_disp)
+; 16: 	addiu	$[[T2:[0-9]+]], $pc, %lo(_gp_disp)
+; 16:	sll	$[[T3:[0-9]+]], $[[T1]], 16
+; 16:	li	$2, 0
+}
+
+declare i32 @printf(i8*, ...)
diff --git a/test/CodeGen/Mips/helloworld_sr.ll b/test/CodeGen/Mips/helloworld_sr.ll
new file mode 100644
index 0000000000..641ee274e0
--- /dev/null
+++ b/test/CodeGen/Mips/helloworld_sr.ll
@@ -0,0 +1,16 @@
+; RUN: llc  -march=mipsel -mcpu=mips16 -relocation-model=pic -O3 < %s | FileCheck %s -check-prefix=16
+
+@.str = private unnamed_addr constant [13 x i8] c"hello world\0A\00", align 1
+
+define i32 @main() nounwind {
+entry:
+  %call = call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([13 x i8]* @.str, i32 0, i32 0))
+  ret i32 0
+
+; 16: 	.set	mips16                  # @main
+
+; 16:	save 	$ra, [[FS:[0-9]+]]
+; 16:	restore 	$ra, [[FS]]
+}
+
+declare i32 @printf(i8*, ...)