MipsInstrInfo.td

//===- MipsInstrInfo.td - Target Description for Mips Target -*- tablegen -*-=//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file contains the Mips implementation of the TargetInstrInfo class.
//
//===----------------------------------------------------------------------===//

//===----------------------------------------------------------------------===//
// Instruction format superclass
//===----------------------------------------------------------------------===//

include "MipsInstrFormats.td"

//===----------------------------------------------------------------------===//
// Mips profiles and nodes
//===----------------------------------------------------------------------===//

def SDT_MipsRet          : SDTypeProfile<0, 1, [SDTCisInt<0>]>;
def SDT_MipsJmpLink      : SDTypeProfile<0, 1, [SDTCisVT<0, iPTR>]>;
def SDT_MipsCMov         : SDTypeProfile<1, 4, [SDTCisSameAs<0, 1>,
                                                SDTCisSameAs<1, 2>,
                                                SDTCisSameAs<3, 4>,
                                                SDTCisInt<4>]>;
def SDT_MipsCallSeqStart : SDCallSeqStart<[SDTCisVT<0, i32>]>;
def SDT_MipsCallSeqEnd   : SDCallSeqEnd<[SDTCisVT<0, i32>, SDTCisVT<1, i32>]>;
def SDT_MipsMAddMSub     : SDTypeProfile<0, 4,
                                         [SDTCisVT<0, i32>, SDTCisSameAs<0, 1>,
                                          SDTCisSameAs<1, 2>,
                                          SDTCisSameAs<2, 3>]>;
def SDT_MipsDivRem       : SDTypeProfile<0, 2,
                                         [SDTCisVT<0, i32>,
                                          SDTCisSameAs<0, 1>]>;

def SDT_MipsThreadPointer : SDTypeProfile<1, 0, [SDTCisPtrTy<0>]>;

def SDT_MipsDynAlloc    : SDTypeProfile<1, 1, [SDTCisVT<0, i32>,
                                               SDTCisVT<1, iPTR>]>;
def SDT_Sync             : SDTypeProfile<0, 1, [SDTCisVT<0, i32>]>;

def SDT_Ext : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
                                   SDTCisVT<2, i32>, SDTCisSameAs<2, 3>]>;
def SDT_Ins : SDTypeProfile<1, 4, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
                                   SDTCisVT<2, i32>, SDTCisSameAs<2, 3>,
                                   SDTCisSameAs<0, 4>]>;

// Call
def MipsJmpLink : SDNode<"MipsISD::JmpLink",SDT_MipsJmpLink,
                         [SDNPHasChain, SDNPOutGlue, SDNPOptInGlue,
                          SDNPVariadic]>;

// Hi and Lo nodes are used to handle global addresses. Used on
// MipsISelLowering to lower stuff like GlobalAddress, ExternalSymbol
// static model. (nothing to do with Mips Registers Hi and Lo)
def MipsHi    : SDNode<"MipsISD::Hi", SDTIntUnaryOp>;
def MipsLo    : SDNode<"MipsISD::Lo", SDTIntUnaryOp>;
def MipsGPRel : SDNode<"MipsISD::GPRel", SDTIntUnaryOp>;

// TlsGd node is used to handle General Dynamic TLS
def MipsTlsGd : SDNode<"MipsISD::TlsGd", SDTIntUnaryOp>;

// TprelHi and TprelLo nodes are used to handle Local Exec TLS
def MipsTprelHi    : SDNode<"MipsISD::TprelHi", SDTIntUnaryOp>;
def MipsTprelLo    : SDNode<"MipsISD::TprelLo", SDTIntUnaryOp>;

// Thread pointer
def MipsThreadPointer: SDNode<"MipsISD::ThreadPointer", SDT_MipsThreadPointer>;

// Return
def MipsRet : SDNode<"MipsISD::Ret", SDT_MipsRet, [SDNPHasChain,
                     SDNPOptInGlue]>;

// These are target-independent nodes, but have target-specific formats.
def callseq_start : SDNode<"ISD::CALLSEQ_START", SDT_MipsCallSeqStart,
                           [SDNPHasChain, SDNPOutGlue]>;
def callseq_end   : SDNode<"ISD::CALLSEQ_END", SDT_MipsCallSeqEnd,
                           [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]>;

// MAdd*/MSub* nodes
def MipsMAdd      : SDNode<"MipsISD::MAdd", SDT_MipsMAddMSub,
                           [SDNPOptInGlue, SDNPOutGlue]>;
def MipsMAddu     : SDNode<"MipsISD::MAddu", SDT_MipsMAddMSub,
                           [SDNPOptInGlue, SDNPOutGlue]>;
def MipsMSub      : SDNode<"MipsISD::MSub", SDT_MipsMAddMSub,
                           [SDNPOptInGlue, SDNPOutGlue]>;
def MipsMSubu     : SDNode<"MipsISD::MSubu", SDT_MipsMAddMSub,
                           [SDNPOptInGlue, SDNPOutGlue]>;

// DivRem(u) nodes
def MipsDivRem    : SDNode<"MipsISD::DivRem", SDT_MipsDivRem,
                           [SDNPOutGlue]>;
def MipsDivRemU   : SDNode<"MipsISD::DivRemU", SDT_MipsDivRem,
                           [SDNPOutGlue]>;

// Target constant nodes that are not part of any isel patterns and remain
// unchanged can cause instructions with illegal operands to be emitted.
// Wrapper node patterns give the instruction selector a chance to replace
// target constant nodes that would otherwise remain unchanged with ADDiu
// nodes. Without these wrapper node patterns, the following conditional move
// instrucion is emitted when function cmov2 in test/CodeGen/Mips/cmov.ll is
// compiled: 
//  movn  %got(d)($gp), %got(c)($gp), $4
// This instruction is illegal since movn can take only register operands.

def MipsWrapperPIC    : SDNode<"MipsISD::WrapperPIC",  SDTIntUnaryOp>;

// Pointer to dynamically allocated stack area.
def MipsDynAlloc  : SDNode<"MipsISD::DynAlloc", SDT_MipsDynAlloc,
                           [SDNPHasChain, SDNPInGlue]>;

def MipsSync : SDNode<"MipsISD::Sync", SDT_Sync, [SDNPHasChain]>;

def MipsExt :  SDNode<"MipsISD::Ext", SDT_Ext>;
def MipsIns :  SDNode<"MipsISD::Ins", SDT_Ins>;

//===----------------------------------------------------------------------===//
// Mips Instruction Predicate Definitions.
//===----------------------------------------------------------------------===//
def HasSEInReg  : Predicate<"Subtarget.hasSEInReg()">;
def HasBitCount : Predicate<"Subtarget.hasBitCount()">;
def HasSwap     : Predicate<"Subtarget.hasSwap()">;
def HasCondMov  : Predicate<"Subtarget.hasCondMov()">;
def HasMips32    : Predicate<"Subtarget.hasMips32()">;
def HasMips32r2  : Predicate<"Subtarget.hasMips32r2()">;

//===----------------------------------------------------------------------===//
// Mips Operand, Complex Patterns and Transformations Definitions.
//===----------------------------------------------------------------------===//

// Instruction operand types
def brtarget    : Operand<OtherVT>;
def calltarget  : Operand<i32>;
def simm16      : Operand<i32>;
def shamt       : Operand<i32>;

// Unsigned Operand
def uimm16      : Operand<i32> {
  let PrintMethod = "printUnsignedImm";
}

// Address operand
def mem : Operand<i32> {
  let PrintMethod = "printMemOperand";
  let MIOperandInfo = (ops CPURegs, simm16);
}

def mem_ea : Operand<i32> {
  let PrintMethod = "printMemOperandEA";
  let MIOperandInfo = (ops CPURegs, simm16);
}

// Transformation Function - get the lower 16 bits.
def LO16 : SDNodeXForm<imm, [{
  return getI32Imm((unsigned)N->getZExtValue() & 0xFFFF);
}]>;

// Transformation Function - get the higher 16 bits.
def HI16 : SDNodeXForm<imm, [{
  return getI32Imm((unsigned)N->getZExtValue() >> 16);
}]>;

// Node immediate fits as 16-bit sign extended on target immediate.
// e.g. addi, andi
def immSExt16  : PatLeaf<(imm), [{ return isInt<16>(N->getSExtValue()); }]>;

// Node immediate fits as 16-bit zero extended on target immediate.
// The LO16 param means that only the lower 16 bits of the node
// immediate are caught.
// e.g. addiu, sltiu
def immZExt16  : PatLeaf<(imm), [{
  if (N->getValueType(0) == MVT::i32)
    return (uint32_t)N->getZExtValue() == (unsigned short)N->getZExtValue();
  else
    return (uint64_t)N->getZExtValue() == (unsigned short)N->getZExtValue();
}], LO16>;

// shamt field must fit in 5 bits.
def immZExt5 : PatLeaf<(imm), [{
  return N->getZExtValue() == ((N->getZExtValue()) & 0x1f) ;
}]>;

// Mips Address Mode! SDNode frameindex could possibily be a match
// since load and store instructions from stack used it.
def addr : ComplexPattern<iPTR, 2, "SelectAddr", [frameindex], []>;

//===----------------------------------------------------------------------===//
// Instructions specific format
//===----------------------------------------------------------------------===//

// Arithmetic 3 register operands
class ArithR<bits<6> op, bits<6> func, string instr_asm, SDNode OpNode,
             InstrItinClass itin, bit isComm = 0>:
  FR<op, func, (outs CPURegs:$dst), (ins CPURegs:$b, CPURegs:$c),
     !strconcat(instr_asm, "\t$dst, $b, $c"),
     [(set CPURegs:$dst, (OpNode CPURegs:$b, CPURegs:$c))], itin> {
  let isCommutable = isComm;
}

class ArithOverflowR<bits<6> op, bits<6> func, string instr_asm,
                     bit isComm = 0>:
  FR<op, func, (outs CPURegs:$dst), (ins CPURegs:$b, CPURegs:$c),
     !strconcat(instr_asm, "\t$dst, $b, $c"), [], IIAlu> {
  let isCommutable = isComm;
}

// Arithmetic 2 register operands
class ArithI<bits<6> op, string instr_asm, SDNode OpNode,
             Operand Od, PatLeaf imm_type> :
  FI<op, (outs CPURegs:$dst), (ins CPURegs:$b, Od:$c),
     !strconcat(instr_asm, "\t$dst, $b, $c"),
     [(set CPURegs:$dst, (OpNode CPURegs:$b, imm_type:$c))], IIAlu>;

class ArithOverflowI<bits<6> op, string instr_asm, SDNode OpNode,
             Operand Od, PatLeaf imm_type> :
  FI<op, (outs CPURegs:$dst), (ins CPURegs:$b, Od:$c),
     !strconcat(instr_asm, "\t$dst, $b, $c"), [], IIAlu>;

// Arithmetic Multiply ADD/SUB
let rd = 0, shamt = 0, Defs = [HI, LO], Uses = [HI, LO] in
class MArithR<bits<6> func, string instr_asm, SDNode op, bit isComm = 0> :
  FR<0x1c, func, (outs), (ins CPURegs:$rs, CPURegs:$rt),
     !strconcat(instr_asm, "\t$rs, $rt"),
     [(op CPURegs:$rs, CPURegs:$rt, LO, HI)], IIImul> {
  let isCommutable = isComm;
}

//  Logical
let isCommutable = 1 in
class LogicR<bits<6> func, string instr_asm, SDNode OpNode>:
  FR<0x00, func, (outs CPURegs:$dst), (ins CPURegs:$b, CPURegs:$c),
     !strconcat(instr_asm, "\t$dst, $b, $c"),
     [(set CPURegs:$dst, (OpNode CPURegs:$b, CPURegs:$c))], IIAlu>;

class LogicI<bits<6> op, string instr_asm, SDNode OpNode>:
  FI<op, (outs CPURegs:$dst), (ins CPURegs:$b, uimm16:$c),
     !strconcat(instr_asm, "\t$dst, $b, $c"),
     [(set CPURegs:$dst, (OpNode CPURegs:$b, immZExt16:$c))], IIAlu>;

let isCommutable = 1 in
class LogicNOR<bits<6> op, bits<6> func, string instr_asm>:
  FR<op, func, (outs CPURegs:$dst), (ins CPURegs:$b, CPURegs:$c),
     !strconcat(instr_asm, "\t$dst, $b, $c"),
     [(set CPURegs:$dst, (not (or CPURegs:$b, CPURegs:$c)))], IIAlu>;

// Shifts
class LogicR_shift_rotate_imm<bits<6> func, bits<5> _rs, string instr_asm,
                              SDNode OpNode>:
  FR<0x00, func, (outs CPURegs:$dst), (ins CPURegs:$b, shamt:$c),
     !strconcat(instr_asm, "\t$dst, $b, $c"),
     [(set CPURegs:$dst, (OpNode CPURegs:$b, (i32 immZExt5:$c)))], IIAlu> {
  let rs = _rs;
}

class LogicR_shift_rotate_reg<bits<6> func, bits<5> _shamt, string instr_asm,
                              SDNode OpNode>:
  FR<0x00, func, (outs CPURegs:$dst), (ins CPURegs:$c, CPURegs:$b),
     !strconcat(instr_asm, "\t$dst, $b, $c"),
     [(set CPURegs:$dst, (OpNode CPURegs:$b, CPURegs:$c))], IIAlu> {
  let shamt = _shamt;
}

// Load Upper Imediate
class LoadUpper<bits<6> op, string instr_asm>:
  FI< op,
      (outs CPURegs:$dst),
      (ins uimm16:$imm),
      !strconcat(instr_asm, "\t$dst, $imm"),
      [], IIAlu>;

// Memory Load/Store
let canFoldAsLoad = 1 in
class LoadM<bits<6> op, string instr_asm, PatFrag OpNode>:
  FI<op, (outs CPURegs:$dst), (ins mem:$addr),
     !strconcat(instr_asm, "\t$dst, $addr"),
     [(set CPURegs:$dst, (OpNode addr:$addr))], IILoad>;

class StoreM<bits<6> op, string instr_asm, PatFrag OpNode>:
  FI<op, (outs), (ins CPURegs:$dst, mem:$addr),
     !strconcat(instr_asm, "\t$dst, $addr"),
     [(OpNode CPURegs:$dst, addr:$addr)], IIStore>;

// Conditional Branch
let isBranch = 1, isTerminator=1, hasDelaySlot = 1 in {
class CBranch<bits<6> op, string instr_asm, PatFrag cond_op>:
  FI<op, (outs), (ins CPURegs:$a, CPURegs:$b, brtarget:$offset),
     !strconcat(instr_asm, "\t$a, $b, $offset"),
     [(brcond (i32 (cond_op CPURegs:$a, CPURegs:$b)), bb:$offset)],
     IIBranch>;

class CBranchZero<bits<6> op, string instr_asm, PatFrag cond_op>:
  FI<op, (outs), (ins CPURegs:$src, brtarget:$offset),
     !strconcat(instr_asm, "\t$src, $offset"),
     [(brcond (i32 (cond_op CPURegs:$src, 0)), bb:$offset)],
     IIBranch>;
}

// SetCC
class SetCC_R<bits<6> op, bits<6> func, string instr_asm,
      PatFrag cond_op>:
  FR<op, func, (outs CPURegs:$dst), (ins CPURegs:$b, CPURegs:$c),
     !strconcat(instr_asm, "\t$dst, $b, $c"),
     [(set CPURegs:$dst, (cond_op CPURegs:$b, CPURegs:$c))],
     IIAlu>;

class SetCC_I<bits<6> op, string instr_asm, PatFrag cond_op,
      Operand Od, PatLeaf imm_type>:
  FI<op, (outs CPURegs:$dst), (ins CPURegs:$b, Od:$c),
     !strconcat(instr_asm, "\t$dst, $b, $c"),
     [(set CPURegs:$dst, (cond_op CPURegs:$b, imm_type:$c))],
     IIAlu>;

// Unconditional branch
let isBranch=1, isTerminator=1, isBarrier=1, hasDelaySlot = 1 in
class JumpFJ<bits<6> op, string instr_asm>:
  FJ<op, (outs), (ins brtarget:$target),
     !strconcat(instr_asm, "\t$target"), [(br bb:$target)], IIBranch>;

let isBranch=1, isTerminator=1, isBarrier=1, rd=0, hasDelaySlot = 1 in
class JumpFR<bits<6> op, bits<6> func, string instr_asm>:
  FR<op, func, (outs), (ins CPURegs:$target),
     !strconcat(instr_asm, "\t$target"), [(brind CPURegs:$target)], IIBranch>;

// Jump and Link (Call)
let isCall=1, hasDelaySlot=1,
  // All calls clobber the non-callee saved registers...
  Defs = [AT, V0, V1, A0, A1, A2, A3, T0, T1, T2, T3, T4, T5, T6, T7, T8, T9,
          K0, K1, D0, D1, D2, D3, D4, D5, D6, D7, D8, D9], Uses = [GP] in {
  class JumpLink<bits<6> op, string instr_asm>:
    FJ<op, (outs), (ins calltarget:$target, variable_ops),
       !strconcat(instr_asm, "\t$target"), [(MipsJmpLink imm:$target)],
       IIBranch>;

  let rd=31 in
  class JumpLinkReg<bits<6> op, bits<6> func, string instr_asm>:
    FR<op, func, (outs), (ins CPURegs:$rs, variable_ops),
       !strconcat(instr_asm, "\t$rs"), [(MipsJmpLink CPURegs:$rs)], IIBranch>;

  class BranchLink<string instr_asm>:
    FI<0x1, (outs), (ins CPURegs:$rs, brtarget:$target, variable_ops),
       !strconcat(instr_asm, "\t$rs, $target"), [], IIBranch>;
}

// Mul, Div
let Defs = [HI, LO] in {
  let isCommutable = 1 in
  class Mul<bits<6> func, string instr_asm, InstrItinClass itin>:
    FR<0x00, func, (outs), (ins CPURegs:$a, CPURegs:$b),
       !strconcat(instr_asm, "\t$a, $b"), [], itin>;

  class Div<SDNode op, bits<6> func, string instr_asm, InstrItinClass itin>:
            FR<0x00, func, (outs), (ins CPURegs:$a, CPURegs:$b),
            !strconcat(instr_asm, "\t$$zero, $a, $b"),
            [(op CPURegs:$a, CPURegs:$b)], itin>;
}

// Move from Hi/Lo
class MoveFromLOHI<bits<6> func, string instr_asm>:
  FR<0x00, func, (outs CPURegs:$dst), (ins),
     !strconcat(instr_asm, "\t$dst"), [], IIHiLo>;

class MoveToLOHI<bits<6> func, string instr_asm>:
  FR<0x00, func, (outs), (ins CPURegs:$src),
     !strconcat(instr_asm, "\t$src"), [], IIHiLo>;

class EffectiveAddress<string instr_asm> :
  FI<0x09, (outs CPURegs:$dst), (ins mem_ea:$addr),
     instr_asm, [(set CPURegs:$dst, addr:$addr)], IIAlu>;

// Count Leading Ones/Zeros in Word
class CountLeading<bits<6> func, string instr_asm, list<dag> pattern>:
  FR<0x1c, func, (outs CPURegs:$dst), (ins CPURegs:$src),
     !strconcat(instr_asm, "\t$dst, $src"), pattern, IIAlu>,
     Requires<[HasBitCount]> {
  let shamt = 0;
  let rt = rd;
}

// Sign Extend in Register.
class SignExtInReg<bits<6> func, string instr_asm, ValueType vt>:
  FR<0x3f, func, (outs CPURegs:$dst), (ins CPURegs:$src),
     !strconcat(instr_asm, "\t$dst, $src"),
     [(set CPURegs:$dst, (sext_inreg CPURegs:$src, vt))], NoItinerary>;

// Byte Swap
class ByteSwap<bits<6> func, string instr_asm>:
  FR<0x1f, func, (outs CPURegs:$dst), (ins CPURegs:$src),
     !strconcat(instr_asm, "\t$dst, $src"),
     [(set CPURegs:$dst, (bswap CPURegs:$src))], NoItinerary>;

// Conditional Move
class CondMov<bits<6> func, string instr_asm, PatLeaf MovCode>:
  FR<0x00, func, (outs CPURegs:$dst), (ins CPURegs:$F, CPURegs:$T,
     CPURegs:$cond), !strconcat(instr_asm, "\t$dst, $T, $cond"),
     [], NoItinerary>;

// Read Hardware
class ReadHardware: FR<0x1f, 0x3b, (outs CPURegs:$dst), (ins HWRegs:$src),
    "rdhwr\t$dst, $src", [], IIAlu> {
  let rs = 0;
  let shamt = 0;
}

// Ext and Ins
class ExtIns<bits<6> _funct, string instr_asm, dag outs, dag ins,
             list<dag> pattern, InstrItinClass itin>:
  FR<0x1f, _funct, outs, ins, !strconcat(instr_asm, " $rt, $rs, $pos, $sz"),
     pattern, itin>, Requires<[HasMips32r2]> {
  bits<5> pos;
  bits<5> sz;
  let rd = sz;
  let shamt = pos;
}

// Atomic instructions with 2 source operands (ATOMIC_SWAP & ATOMIC_LOAD_*).
class Atomic2Ops<PatFrag Op, string Opstr> :
  MipsPseudo<(outs CPURegs:$dst), (ins CPURegs:$ptr, CPURegs:$incr),
             !strconcat("atomic_", Opstr, "\t$dst, $ptr, $incr"),
             [(set CPURegs:$dst,
              (Op CPURegs:$ptr, CPURegs:$incr))]>;

// Atomic Compare & Swap.
class AtomicCmpSwap<PatFrag Op, string Width> :
  MipsPseudo<(outs CPURegs:$dst), 
             (ins CPURegs:$ptr, CPURegs:$cmp, CPURegs:$swap),
             !strconcat("atomic_cmp_swap_", Width, 
                        "\t$dst, $ptr, $cmp, $swap"),
             [(set CPURegs:$dst,
              (Op CPURegs:$ptr, CPURegs:$cmp, CPURegs:$swap))]>;

//===----------------------------------------------------------------------===//
// Pseudo instructions
//===----------------------------------------------------------------------===//

// As stack alignment is always done with addiu, we need a 16-bit immediate
let Defs = [SP], Uses = [SP] in {
def ADJCALLSTACKDOWN : MipsPseudo<(outs), (ins uimm16:$amt),
                                  "!ADJCALLSTACKDOWN $amt",
                                  [(callseq_start timm:$amt)]>;
def ADJCALLSTACKUP   : MipsPseudo<(outs), (ins uimm16:$amt1, uimm16:$amt2),
                                  "!ADJCALLSTACKUP $amt1",
                                  [(callseq_end timm:$amt1, timm:$amt2)]>;
}

// Some assembly macros need to avoid pseudoinstructions and assembler
// automatic reodering, we should reorder ourselves.
def MACRO     : MipsPseudo<(outs), (ins), ".set\tmacro",     []>;
def REORDER   : MipsPseudo<(outs), (ins), ".set\treorder",   []>;
def NOMACRO   : MipsPseudo<(outs), (ins), ".set\tnomacro",   []>;
def NOREORDER : MipsPseudo<(outs), (ins), ".set\tnoreorder", []>;

// These macros are inserted to prevent GAS from complaining
// when using the AT register.
def NOAT      : MipsPseudo<(outs), (ins), ".set\tnoat", []>;
def ATMACRO   : MipsPseudo<(outs), (ins), ".set\tat", []>;

// When handling PIC code the assembler needs .cpload and .cprestore
// directives. If the real instructions corresponding these directives
// are used, we have the same behavior, but get also a bunch of warnings
// from the assembler.
def CPLOAD : MipsPseudo<(outs), (ins CPURegs:$picreg), ".cpload\t$picreg", []>;
def CPRESTORE : MipsPseudo<(outs), (ins i32imm:$loc), ".cprestore\t$loc", []>;

let usesCustomInserter = 1 in {
  def ATOMIC_LOAD_ADD_I8   : Atomic2Ops<atomic_load_add_8, "load_add_8">;
  def ATOMIC_LOAD_ADD_I16  : Atomic2Ops<atomic_load_add_16, "load_add_16">;
  def ATOMIC_LOAD_ADD_I32  : Atomic2Ops<atomic_load_add_32, "load_add_32">;
  def ATOMIC_LOAD_SUB_I8   : Atomic2Ops<atomic_load_sub_8, "load_sub_8">;
  def ATOMIC_LOAD_SUB_I16  : Atomic2Ops<atomic_load_sub_16, "load_sub_16">;
  def ATOMIC_LOAD_SUB_I32  : Atomic2Ops<atomic_load_sub_32, "load_sub_32">;
  def ATOMIC_LOAD_AND_I8   : Atomic2Ops<atomic_load_and_8, "load_and_8">;
  def ATOMIC_LOAD_AND_I16  : Atomic2Ops<atomic_load_and_16, "load_and_16">;
  def ATOMIC_LOAD_AND_I32  : Atomic2Ops<atomic_load_and_32, "load_and_32">;
  def ATOMIC_LOAD_OR_I8    : Atomic2Ops<atomic_load_or_8, "load_or_8">;
  def ATOMIC_LOAD_OR_I16   : Atomic2Ops<atomic_load_or_16, "load_or_16">;
  def ATOMIC_LOAD_OR_I32   : Atomic2Ops<atomic_load_or_32, "load_or_32">;
  def ATOMIC_LOAD_XOR_I8   : Atomic2Ops<atomic_load_xor_8, "load_xor_8">;
  def ATOMIC_LOAD_XOR_I16  : Atomic2Ops<atomic_load_xor_16, "load_xor_16">;
  def ATOMIC_LOAD_XOR_I32  : Atomic2Ops<atomic_load_xor_32, "load_xor_32">;
  def ATOMIC_LOAD_NAND_I8  : Atomic2Ops<atomic_load_nand_8, "load_nand_8">;
  def ATOMIC_LOAD_NAND_I16 : Atomic2Ops<atomic_load_nand_16, "load_nand_16">;
  def ATOMIC_LOAD_NAND_I32 : Atomic2Ops<atomic_load_nand_32, "load_nand_32">;

  def ATOMIC_SWAP_I8       : Atomic2Ops<atomic_swap_8, "swap_8">;
  def ATOMIC_SWAP_I16      : Atomic2Ops<atomic_swap_16, "swap_16">;
  def ATOMIC_SWAP_I32      : Atomic2Ops<atomic_swap_32, "swap_32">;

  def ATOMIC_CMP_SWAP_I8   : AtomicCmpSwap<atomic_cmp_swap_8, "8">;
  def ATOMIC_CMP_SWAP_I16  : AtomicCmpSwap<atomic_cmp_swap_16, "16">;
  def ATOMIC_CMP_SWAP_I32  : AtomicCmpSwap<atomic_cmp_swap_32, "32">;
}

// Unaligned loads and stores.
// Replaces LW or SW during MCInstLowering if memory access is unaligned.
def ULW :
  MipsPseudo<(outs CPURegs:$dst), (ins mem:$addr), "ulw\t$dst, $addr", []>;
def ULH :
  MipsPseudo<(outs CPURegs:$dst), (ins mem:$addr), "ulh\t$dst, $addr", []>;
def ULHu :
  MipsPseudo<(outs CPURegs:$dst), (ins mem:$addr), "ulhu\t$dst, $addr", []>;
def USW : 
  MipsPseudo<(outs), (ins CPURegs:$dst, mem:$addr), "usw\t$dst, $addr", []>;
def USH : 
  MipsPseudo<(outs), (ins CPURegs:$dst, mem:$addr), "ush\t$dst, $addr", []>;

//===----------------------------------------------------------------------===//
// Instruction definition
//===----------------------------------------------------------------------===//

//===----------------------------------------------------------------------===//
// MipsI Instructions
//===----------------------------------------------------------------------===//

/// Arithmetic Instructions (ALU Immediate)
def ADDiu   : ArithI<0x09, "addiu", add, simm16, immSExt16>;
def ADDi    : ArithOverflowI<0x08, "addi",  add, simm16, immSExt16>;
def SLTi    : SetCC_I<0x0a, "slti", setlt, simm16, immSExt16>;
def SLTiu   : SetCC_I<0x0b, "sltiu", setult, simm16, immSExt16>;
def ANDi    : LogicI<0x0c, "andi", and>;
def ORi     : LogicI<0x0d, "ori",  or>;
def XORi    : LogicI<0x0e, "xori",  xor>;
def LUi     : LoadUpper<0x0f, "lui">;

/// Arithmetic Instructions (3-Operand, R-Type)
def ADDu    : ArithR<0x00, 0x21, "addu", add, IIAlu, 1>;
def SUBu    : ArithR<0x00, 0x23, "subu", sub, IIAlu>;
def ADD     : ArithOverflowR<0x00, 0x20, "add", 1>;
def SUB     : ArithOverflowR<0x00, 0x22, "sub">;
def SLT     : SetCC_R<0x00, 0x2a, "slt", setlt>;
def SLTu    : SetCC_R<0x00, 0x2b, "sltu", setult>;
def AND     : LogicR<0x24, "and", and>;
def OR      : LogicR<0x25, "or",  or>;
def XOR     : LogicR<0x26, "xor", xor>;
def NOR     : LogicNOR<0x00, 0x27, "nor">;

/// Shift Instructions
def SLL     : LogicR_shift_rotate_imm<0x00, 0x00, "sll", shl>;
def SRL     : LogicR_shift_rotate_imm<0x02, 0x00, "srl", srl>;
def SRA     : LogicR_shift_rotate_imm<0x03, 0x00, "sra", sra>;
def SLLV    : LogicR_shift_rotate_reg<0x04, 0x00, "sllv", shl>;
def SRLV    : LogicR_shift_rotate_reg<0x06, 0x00, "srlv", srl>;
def SRAV    : LogicR_shift_rotate_reg<0x07, 0x00, "srav", sra>;

// Rotate Instructions
let Predicates = [HasMips32r2] in {
    def ROTR    : LogicR_shift_rotate_imm<0x02, 0x01, "rotr", rotr>;
    def ROTRV   : LogicR_shift_rotate_reg<0x06, 0x01, "rotrv", rotr>;
}

/// Load and Store Instructions
def LB      : LoadM<0x20, "lb",  sextloadi8>;
def LBu     : LoadM<0x24, "lbu", zextloadi8>;
def LH      : LoadM<0x21, "lh",  sextloadi16>;
def LHu     : LoadM<0x25, "lhu", zextloadi16>;
def LW      : LoadM<0x23, "lw",  load>;
def SB      : StoreM<0x28, "sb", truncstorei8>;
def SH      : StoreM<0x29, "sh", truncstorei16>;
def SW      : StoreM<0x2b, "sw", store>;

let hasSideEffects = 1 in
def SYNC : MipsInst<(outs), (ins i32imm:$stype), "sync $stype",
                    [(MipsSync imm:$stype)], NoItinerary>
{
  let opcode = 0;
  let Inst{25-11} = 0;
  let Inst{5-0} = 15;
}

/// Load-linked, Store-conditional
let mayLoad = 1 in
  def LL    : FI<0x30, (outs CPURegs:$dst), (ins mem:$addr),
              "ll\t$dst, $addr", [], IILoad>;
let mayStore = 1, Constraints = "$src = $dst" in
  def SC    : FI<0x38, (outs CPURegs:$dst), (ins CPURegs:$src, mem:$addr),
              "sc\t$src, $addr", [], IIStore>;

/// Jump and Branch Instructions
def J       : JumpFJ<0x02, "j">;
let isIndirectBranch = 1 in
  def JR      : JumpFR<0x00, 0x08, "jr">;
def JAL     : JumpLink<0x03, "jal">;
def JALR    : JumpLinkReg<0x00, 0x09, "jalr">;
def BEQ     : CBranch<0x04, "beq", seteq>;
def BNE     : CBranch<0x05, "bne", setne>;

let rt=1 in
  def BGEZ  : CBranchZero<0x01, "bgez", setge>;

let rt=0 in {
  def BGTZ  : CBranchZero<0x07, "bgtz", setgt>;
  def BLEZ  : CBranchZero<0x07, "blez", setle>;
  def BLTZ  : CBranchZero<0x01, "bltz", setlt>;
}

def BGEZAL  : BranchLink<"bgezal">;
def BLTZAL  : BranchLink<"bltzal">;

let isReturn=1, isTerminator=1, hasDelaySlot=1,
    isBarrier=1, hasCtrlDep=1, rs=0, rt=0, shamt=0 in
  def RET : FR <0x00, 0x02, (outs), (ins CPURegs:$target),
                "jr\t$target", [(MipsRet CPURegs:$target)], IIBranch>;

/// Multiply and Divide Instructions.
def MULT    : Mul<0x18, "mult", IIImul>;
def MULTu   : Mul<0x19, "multu", IIImul>;
def SDIV    : Div<MipsDivRem, 0x1a, "div", IIIdiv>;
def UDIV    : Div<MipsDivRemU, 0x1b, "divu", IIIdiv>;

let Defs = [HI] in
  def MTHI  : MoveToLOHI<0x11, "mthi">;
let Defs = [LO] in
  def MTLO  : MoveToLOHI<0x13, "mtlo">;

let Uses = [HI] in
  def MFHI  : MoveFromLOHI<0x10, "mfhi">;
let Uses = [LO] in
  def MFLO  : MoveFromLOHI<0x12, "mflo">;

/// Sign Ext In Register Instructions.
let Predicates = [HasSEInReg] in {
  let shamt = 0x10, rs = 0 in
    def SEB : SignExtInReg<0x21, "seb", i8>;

  let shamt = 0x18, rs = 0 in
    def SEH : SignExtInReg<0x20, "seh", i16>;
}

/// Count Leading
def CLZ : CountLeading<0b100000, "clz",
                       [(set CPURegs:$dst, (ctlz CPURegs:$src))]>;
def CLO : CountLeading<0b100001, "clo",
                       [(set CPURegs:$dst, (ctlz (not CPURegs:$src)))]>;

/// Byte Swap
let Predicates = [HasSwap] in {
  let shamt = 0x3, rs = 0 in
    def WSBW : ByteSwap<0x20, "wsbw">;
}

/// Conditional Move
def MIPS_CMOV_ZERO  : PatLeaf<(i32 0)>;
def MIPS_CMOV_NZERO : PatLeaf<(i32 1)>;

// Conditional moves:
// These instructions are expanded in
// MipsISelLowering::EmitInstrWithCustomInserter if target does not have
// conditional move instructions.
// flag:int, data:int
let usesCustomInserter = 1, shamt = 0, Constraints = "$F = $dst" in
  class CondMovIntInt<bits<6> funct, string instr_asm> :
    FR<0, funct, (outs CPURegs:$dst),
       (ins CPURegs:$T, CPURegs:$cond, CPURegs:$F),
       !strconcat(instr_asm, "\t$dst, $T, $cond"), [], NoItinerary>;

def MOVZ_I : CondMovIntInt<0x0a, "movz">;
def MOVN_I : CondMovIntInt<0x0b, "movn">;

/// No operation
let addr=0 in
  def NOP   : FJ<0, (outs), (ins), "nop", [], IIAlu>;

// FrameIndexes are legalized when they are operands from load/store
// instructions. The same not happens for stack address copies, so an
// add op with mem ComplexPattern is used and the stack address copy
// can be matched. It's similar to Sparc LEA_ADDRi
def LEA_ADDiu : EffectiveAddress<"addiu\t$dst, $addr">;

// DynAlloc node points to dynamically allocated stack space.
// $sp is added to the list of implicitly used registers to prevent dead code
// elimination from removing instructions that modify $sp.
let Uses = [SP] in
def DynAlloc : EffectiveAddress<"addiu\t$dst, $addr">;

// MADD*/MSUB*
def MADD  : MArithR<0, "madd", MipsMAdd, 1>;
def MADDU : MArithR<1, "maddu", MipsMAddu, 1>;
def MSUB  : MArithR<4, "msub", MipsMSub>;
def MSUBU : MArithR<5, "msubu", MipsMSubu>;

// MUL is a assembly macro in the current used ISAs. In recent ISA's
// it is a real instruction.
def MUL   : ArithR<0x1c, 0x02, "mul", mul, IIImul, 1>, Requires<[HasMips32]>;

def RDHWR : ReadHardware;

def EXT : ExtIns<0, "ext", (outs CPURegs:$rt),
                 (ins CPURegs:$rs, uimm16:$pos, uimm16:$sz),
                 [(set CPURegs:$rt,
                   (MipsExt CPURegs:$rs, immZExt5:$pos, immZExt5:$sz))],
                 NoItinerary>;

let Constraints = "$src = $rt" in
def INS : ExtIns<4, "ins", (outs CPURegs:$rt),
                 (ins CPURegs:$rs, uimm16:$pos, uimm16:$sz, CPURegs:$src),
                 [(set CPURegs:$rt,
                   (MipsIns CPURegs:$rs, immZExt5:$pos, immZExt5:$sz,
                    CPURegs:$src))],
                 NoItinerary>;

//===----------------------------------------------------------------------===//
//  Arbitrary patterns that map to one or more instructions
//===----------------------------------------------------------------------===//

// Small immediates
def : Pat<(i32 immSExt16:$in),
          (ADDiu ZERO, imm:$in)>;
def : Pat<(i32 immZExt16:$in),
          (ORi ZERO, imm:$in)>;

// Arbitrary immediates
def : Pat<(i32 imm:$imm),
          (ORi (LUi (HI16 imm:$imm)), (LO16 imm:$imm))>;

// Carry patterns
def : Pat<(subc CPURegs:$lhs, CPURegs:$rhs),
          (SUBu CPURegs:$lhs, CPURegs:$rhs)>;
def : Pat<(addc CPURegs:$lhs, CPURegs:$rhs),
          (ADDu CPURegs:$lhs, CPURegs:$rhs)>;
def : Pat<(addc  CPURegs:$src, immSExt16:$imm),
          (ADDiu CPURegs:$src, imm:$imm)>;

// Call
def : Pat<(MipsJmpLink (i32 tglobaladdr:$dst)),
          (JAL tglobaladdr:$dst)>;
def : Pat<(MipsJmpLink (i32 texternalsym:$dst)),
          (JAL texternalsym:$dst)>;
//def : Pat<(MipsJmpLink CPURegs:$dst),
//          (JALR CPURegs:$dst)>;

// hi/lo relocs
def : Pat<(MipsHi tglobaladdr:$in), (LUi tglobaladdr:$in)>;
def : Pat<(MipsHi tblockaddress:$in), (LUi tblockaddress:$in)>;
def : Pat<(MipsLo tglobaladdr:$in), (ADDiu ZERO, tglobaladdr:$in)>;
def : Pat<(MipsLo tblockaddress:$in), (ADDiu ZERO, tblockaddress:$in)>;
def : Pat<(add CPURegs:$hi, (MipsLo tglobaladdr:$lo)),
          (ADDiu CPURegs:$hi, tglobaladdr:$lo)>;
def : Pat<(add CPURegs:$hi, (MipsLo tblockaddress:$lo)),
          (ADDiu CPURegs:$hi, tblockaddress:$lo)>;

def : Pat<(MipsHi tjumptable:$in), (LUi tjumptable:$in)>;
def : Pat<(MipsLo tjumptable:$in), (ADDiu ZERO, tjumptable:$in)>;
def : Pat<(add CPURegs:$hi, (MipsLo tjumptable:$lo)),
          (ADDiu CPURegs:$hi, tjumptable:$lo)>;

def : Pat<(MipsHi tconstpool:$in), (LUi tconstpool:$in)>;
def : Pat<(MipsLo tconstpool:$in), (ADDiu ZERO, tconstpool:$in)>;
def : Pat<(add CPURegs:$hi, (MipsLo tconstpool:$lo)),
          (ADDiu CPURegs:$hi, tconstpool:$lo)>;

// gp_rel relocs
def : Pat<(add CPURegs:$gp, (MipsGPRel tglobaladdr:$in)),
          (ADDiu CPURegs:$gp, tglobaladdr:$in)>;
def : Pat<(add CPURegs:$gp, (MipsGPRel tconstpool:$in)),
          (ADDiu CPURegs:$gp, tconstpool:$in)>;

// tlsgd
def : Pat<(add CPURegs:$gp, (MipsTlsGd tglobaltlsaddr:$in)),
          (ADDiu CPURegs:$gp, tglobaltlsaddr:$in)>;

// tprel hi/lo
def : Pat<(MipsTprelHi tglobaltlsaddr:$in), (LUi tglobaltlsaddr:$in)>;
def : Pat<(MipsTprelLo tglobaltlsaddr:$in), (ADDiu ZERO, tglobaltlsaddr:$in)>;
def : Pat<(add CPURegs:$hi, (MipsTprelLo tglobaltlsaddr:$lo)),
          (ADDiu CPURegs:$hi, tglobaltlsaddr:$lo)>;

// wrapper_pic
class WrapperPICPat<SDNode node>:
      Pat<(MipsWrapperPIC node:$in),
          (ADDiu GP, node:$in)>;

def : WrapperPICPat<tglobaladdr>;
def : WrapperPICPat<tconstpool>;
def : WrapperPICPat<texternalsym>;
def : WrapperPICPat<tblockaddress>;
def : WrapperPICPat<tjumptable>;

// Mips does not have "not", so we expand our way
def : Pat<(not CPURegs:$in),
          (NOR CPURegs:$in, ZERO)>;

// extended load and stores
def : Pat<(extloadi1  addr:$src), (LBu addr:$src)>;
def : Pat<(extloadi8  addr:$src), (LBu addr:$src)>;
def : Pat<(extloadi16 addr:$src), (LHu addr:$src)>;

// peepholes
def : Pat<(store (i32 0), addr:$dst), (SW ZERO, addr:$dst)>;

// brcond patterns
def : Pat<(brcond (i32 (setne CPURegs:$lhs, 0)), bb:$dst),
          (BNE CPURegs:$lhs, ZERO, bb:$dst)>;
def : Pat<(brcond (i32 (seteq CPURegs:$lhs, 0)), bb:$dst),
          (BEQ CPURegs:$lhs, ZERO, bb:$dst)>;

def : Pat<(brcond (i32 (setge CPURegs:$lhs, CPURegs:$rhs)), bb:$dst),
          (BEQ (SLT CPURegs:$lhs, CPURegs:$rhs), ZERO, bb:$dst)>;
def : Pat<(brcond (i32 (setuge CPURegs:$lhs, CPURegs:$rhs)), bb:$dst),
          (BEQ (SLTu CPURegs:$lhs, CPURegs:$rhs), ZERO, bb:$dst)>;
def : Pat<(brcond (i32 (setge CPURegs:$lhs, immSExt16:$rhs)), bb:$dst),
          (BEQ (SLTi CPURegs:$lhs, immSExt16:$rhs), ZERO, bb:$dst)>;
def : Pat<(brcond (i32 (setuge CPURegs:$lhs, immSExt16:$rhs)), bb:$dst),
          (BEQ (SLTiu CPURegs:$lhs, immSExt16:$rhs), ZERO, bb:$dst)>;

def : Pat<(brcond (i32 (setle CPURegs:$lhs, CPURegs:$rhs)), bb:$dst),
          (BEQ (SLT CPURegs:$rhs, CPURegs:$lhs), ZERO, bb:$dst)>;
def : Pat<(brcond (i32 (setule CPURegs:$lhs, CPURegs:$rhs)), bb:$dst),
          (BEQ (SLTu CPURegs:$rhs, CPURegs:$lhs), ZERO, bb:$dst)>;

def : Pat<(brcond CPURegs:$cond, bb:$dst),
          (BNE CPURegs:$cond, ZERO, bb:$dst)>;

// select patterns
multiclass MovzPats<RegisterClass RC, Instruction MOVZInst> {
  def : Pat<(select (i32 (setge CPURegs:$lhs, CPURegs:$rhs)), RC:$T, RC:$F),
            (MOVZInst RC:$T, (SLT CPURegs:$lhs, CPURegs:$rhs), RC:$F)>;
  def : Pat<(select (i32 (setuge CPURegs:$lhs, CPURegs:$rhs)), RC:$T, RC:$F),
            (MOVZInst RC:$T, (SLTu CPURegs:$lhs, CPURegs:$rhs), RC:$F)>;
  def : Pat<(select (i32 (setge CPURegs:$lhs, immSExt16:$rhs)), RC:$T, RC:$F),
            (MOVZInst RC:$T, (SLTi CPURegs:$lhs, immSExt16:$rhs), RC:$F)>;
  def : Pat<(select (i32 (setuge CPURegs:$lh, immSExt16:$rh)), RC:$T, RC:$F),
            (MOVZInst RC:$T, (SLTiu CPURegs:$lh, immSExt16:$rh), RC:$F)>;
  def : Pat<(select (i32 (setle CPURegs:$lhs, CPURegs:$rhs)), RC:$T, RC:$F),
            (MOVZInst RC:$T, (SLT CPURegs:$rhs, CPURegs:$lhs), RC:$F)>;
  def : Pat<(select (i32 (setule CPURegs:$lhs, CPURegs:$rhs)), RC:$T, RC:$F),
            (MOVZInst RC:$T, (SLTu CPURegs:$rhs, CPURegs:$lhs), RC:$F)>;
  def : Pat<(select (i32 (seteq CPURegs:$lhs, CPURegs:$rhs)), RC:$T, RC:$F),
            (MOVZInst RC:$T, (XOR CPURegs:$lhs, CPURegs:$rhs), RC:$F)>;
  def : Pat<(select (i32 (seteq CPURegs:$lhs, 0)), RC:$T, RC:$F),
            (MOVZInst RC:$T, CPURegs:$lhs, RC:$F)>;
}

multiclass MovnPats<RegisterClass RC, Instruction MOVNInst> {
  def : Pat<(select (i32 (setne CPURegs:$lhs, CPURegs:$rhs)), RC:$T, RC:$F),
            (MOVNInst RC:$T, (XOR CPURegs:$lhs, CPURegs:$rhs), RC:$F)>;
  def : Pat<(select CPURegs:$cond, RC:$T, RC:$F),
            (MOVNInst RC:$T, CPURegs:$cond, RC:$F)>;
  def : Pat<(select (i32 (setne CPURegs:$lhs, 0)), RC:$T, RC:$F),
            (MOVNInst RC:$T, CPURegs:$lhs, RC:$F)>;
}

defm : MovzPats<CPURegs, MOVZ_I>;
defm : MovnPats<CPURegs, MOVN_I>;

// setcc patterns
def : Pat<(seteq CPURegs:$lhs, CPURegs:$rhs),
          (SLTu (XOR CPURegs:$lhs, CPURegs:$rhs), 1)>;
def : Pat<(setne CPURegs:$lhs, CPURegs:$rhs),
          (SLTu ZERO, (XOR CPURegs:$lhs, CPURegs:$rhs))>;

def : Pat<(setle CPURegs:$lhs, CPURegs:$rhs),
          (XORi (SLT CPURegs:$rhs, CPURegs:$lhs), 1)>;
def : Pat<(setule CPURegs:$lhs, CPURegs:$rhs),
          (XORi (SLTu CPURegs:$rhs, CPURegs:$lhs), 1)>;

def : Pat<(setgt CPURegs:$lhs, CPURegs:$rhs),
          (SLT CPURegs:$rhs, CPURegs:$lhs)>;
def : Pat<(setugt CPURegs:$lhs, CPURegs:$rhs),
          (SLTu CPURegs:$rhs, CPURegs:$lhs)>;

def : Pat<(setge CPURegs:$lhs, CPURegs:$rhs),
          (XORi (SLT CPURegs:$lhs, CPURegs:$rhs), 1)>;
def : Pat<(setuge CPURegs:$lhs, CPURegs:$rhs),
          (XORi (SLTu CPURegs:$lhs, CPURegs:$rhs), 1)>;

def : Pat<(setge CPURegs:$lhs, immSExt16:$rhs),
          (XORi (SLTi CPURegs:$lhs, immSExt16:$rhs), 1)>;
def : Pat<(setuge CPURegs:$lhs, immSExt16:$rhs),
          (XORi (SLTiu CPURegs:$lhs, immSExt16:$rhs), 1)>;

// select MipsDynAlloc
def : Pat<(MipsDynAlloc addr:$f), (DynAlloc addr:$f)>;

//===----------------------------------------------------------------------===//
// Floating Point Support
//===----------------------------------------------------------------------===//

include "MipsInstrFPU.td"