Skip to content
X86InstrInfo.td 21.5 KiB
Newer Older
//===- X86InstrInfo.td - Describe the X86 Instruction Set -------*- C++ -*-===//
//
// This file describes the X86 instruction set, defining the instructions, and
// properties of the instructions which are needed for code generation, machine
// code emission, and analysis.
//
//===----------------------------------------------------------------------===//

// Format specifies the encoding used by the instruction.  This is part of the
// ad-hoc solution used to emit machine instruction encodings by our machine
// code emitter.
class Format<bits<5> val> {
  bits<5> Value = val;
}

def Pseudo     : Format<0>; def RawFrm     : Format<1>;
def AddRegFrm  : Format<2>; def MRMDestReg : Format<3>;
def MRMDestMem : Format<4>; def MRMSrcReg  : Format<5>;
def MRMSrcMem  : Format<6>;
def MRMS0r : Format<16>; def MRMS1r : Format<17>; def MRMS2r : Format<18>;
def MRMS3r : Format<19>; def MRMS4r : Format<20>; def MRMS5r : Format<21>;
def MRMS6r : Format<22>; def MRMS7r : Format<23>;
def MRMS0m : Format<24>; def MRMS1m : Format<25>; def MRMS2m : Format<26>;
def MRMS3m : Format<27>; def MRMS4m : Format<28>; def MRMS5m : Format<29>;
def MRMS6m : Format<30>; def MRMS7m : Format<31>;

// ArgType - This specifies the argument type used by an instruction. This is
// part of the ad-hoc solution used to emit machine instruction encodings by our
// machine code emitter.
class ArgType<bits<3> val> {
  bits<3> Value = val;
}
def NoArg  : ArgType<0>;
def Arg8   : ArgType<1>;
def Arg16  : ArgType<2>;
def Arg32  : ArgType<3>;
def Arg64  : ArgType<4>;   // 64 bit int argument for FILD64
def ArgF32 : ArgType<5>;
def ArgF64 : ArgType<6>;
def ArgF80 : ArgType<6>;

// FPFormat - This specifies what form this FP instruction has.  This is used by
// the Floating-Point stackifier pass.
class FPFormat<bits<3> val> {
  bits<3> Value = val;
}
def NotFP      : FPFormat<0>;
def ZeroArgFP  : FPFormat<1>;
def OneArgFP   : FPFormat<2>;
def OneArgFPRW : FPFormat<3>;
def TwoArgFP   : FPFormat<4>;
def SpecialFP  : FPFormat<5>;


class X86Inst<string nam, bits<8> opcod, Format f, ArgType a> : Instruction {
  let Namespace = "X86";
  let Name = nam;
  bits<8> Opcode = opcod;
  Format Form = f;
  bits<5> FormBits = Form.Value;
  ArgType Type = a;
  bits<3> TypeBits = Type.Value;

  // Attributes specific to X86 instructions...
  bit hasOpSizePrefix = 0; // Does this inst have a 0x66 prefix?
  bit printImplicitUses = 0; // Should we print implicit uses of this inst?

  bits<4> Prefix = 0;       // Which prefix byte does this inst have?
  FPFormat FPForm;          // What flavor of FP instruction is this?
  bits<3> FPFormBits = 0;
}

class Imp<list<Register> uses, list<Register> defs> {
  list<Register> Uses = uses;
  list<Register> Defs = defs;
}

class Pattern<dag P> {
  dag Pattern = P;
}


// Prefix byte classes which are used to indicate to the ad-hoc machine code
// emitter that various prefix bytes are required.
class OpSize { bit hasOpSizePrefix = 1; }
class TB     { bits<4> Prefix = 1; }
class D8     { bits<4> Prefix = 2; }
class D9     { bits<4> Prefix = 3; }
class DA     { bits<4> Prefix = 4; }
class DB     { bits<4> Prefix = 5; }
class DC     { bits<4> Prefix = 6; }
class DD     { bits<4> Prefix = 7; }
class DE     { bits<4> Prefix = 8; }
class DF     { bits<4> Prefix = 9; }



//===----------------------------------------------------------------------===//
// Instruction list...
//

def PHI : X86Inst<"PHI", 0, Pseudo, NoArg>;          // PHI node...

def NOOP : X86Inst<"nop", 0x90, RawFrm, NoArg>;    // nop

def ADJCALLSTACKDOWN : X86Inst<"ADJCALLSTACKDOWN", 0, Pseudo, NoArg>;
def ADJCALLSTACKUP   : X86Inst<"ADJCALLSTACKUP",   0, Pseudo, NoArg>;
def IMPLICIT_USE     : X86Inst<"IMPLICIT_USE",     0, Pseudo, NoArg>;

//===----------------------------------------------------------------------===//
//  Control Flow Instructions...
//

// Return instruction...
let isTerminator = 1, isReturn = 1 in
  def RET : X86Inst<"ret", 0xC3, RawFrm, NoArg>;

// All branches are RawFrm, Void, Branch, and Terminators
let isBranch = 1, isTerminator = 1 in
  class IBr<string name, bits<8> opcode> : X86Inst<name, opcode, RawFrm, NoArg>;

def JMP : IBr<"jmp", 0xE9>;
def JB  : IBr<"jb" , 0x82>, TB;
def JAE : IBr<"jae", 0x83>, TB;
def JE  : IBr<"je" , 0x84>, TB;
def JNE : IBr<"jne", 0x85>, TB;
def JBE : IBr<"jbe", 0x86>, TB;
def JA  : IBr<"ja" , 0x87>, TB;
def JL  : IBr<"jl" , 0x8C>, TB;
def JGE : IBr<"jge", 0x8D>, TB;
def JLE : IBr<"jle", 0x8E>, TB;
def JG  : IBr<"jg" , 0x8F>, TB;


//===----------------------------------------------------------------------===//
//  Call Instructions...
//
let isCall = 1 in
  // All calls clobber the non-callee saved registers...
  let Defs = [EAX, ECX, EDX, FP0, FP1, FP2, FP3, FP4, FP5, FP6] in {
    def CALLpcrel32 : X86Inst<"call", 0xE8, RawFrm, NoArg>;
    def CALLr32     : X86Inst<"call", 0xFF, MRMS2r, Arg32>;
    def CALLm32     : X86Inst<"call", 0xFF, MRMS2m, Arg32>;
  }

       
//===----------------------------------------------------------------------===//
//  Miscellaneous Instructions...
//
def LEAVE    : X86Inst<"leave", 0xC9, RawFrm, NoArg>, Imp<[EBP], [EBP]>;

let isTwoAddress = 1 in                                      // R32 = bswap R32
  def BSWAPr32 : X86Inst<"bswap", 0xC8, AddRegFrm, Arg32>, TB;

def XCHGrr8  : X86Inst<"xchg", 0x86, MRMDestReg, Arg8>;         // xchg R8, R8
def XCHGrr16 : X86Inst<"xchg", 0x87, MRMDestReg, Arg16>, OpSize;// xchg R16, R16
def XCHGrr32 : X86Inst<"xchg", 0x87, MRMDestReg, Arg32>;        // xchg R32, R32

def LEAr16 : X86Inst<"lea", 0x8D, MRMSrcMem, Arg16>, OpSize; // R16 = lea [mem]
def LEAr32 : X86Inst<"lea", 0x8D, MRMSrcMem, Arg32>;         // R32 = lea [mem]

//===----------------------------------------------------------------------===//
//  Move Instructions...
//
def MOVrr8  : X86Inst<"mov", 0x88, MRMDestReg, Arg8>,          Pattern<(set R8 , R8 )>;
def MOVrr16 : X86Inst<"mov", 0x89, MRMDestReg, Arg16>, OpSize, Pattern<(set R16, R16)>;
def MOVrr32 : X86Inst<"mov", 0x89, MRMDestReg, Arg32>,         Pattern<(set R32, R32)>;
def MOVir8  : X86Inst<"mov", 0xB0, AddRegFrm , Arg8>,          Pattern<(set R8 , imm8 )>;
def MOVir16 : X86Inst<"mov", 0xB8, AddRegFrm , Arg16>, OpSize, Pattern<(set R16, imm16)>;
def MOVir32 : X86Inst<"mov", 0xB8, AddRegFrm , Arg32>,         Pattern<(set R32, imm32)>;
def MOVim8  : X86Inst<"mov", 0xC6, MRMS0m    , Arg8>;             // [mem] = imm8
def MOVim16 : X86Inst<"mov", 0xC7, MRMS0m    , Arg16>, OpSize;    // [mem] = imm16
def MOVim32 : X86Inst<"mov", 0xC7, MRMS0m    , Arg32>;            // [mem] = imm32

def MOVmr8  : X86Inst<"mov", 0x8A, MRMSrcMem , Arg8>;             // R8  = [mem]
def MOVmr16 : X86Inst<"mov", 0x8B, MRMSrcMem , Arg16>, OpSize;    // R16 = [mem]
def MOVmr32 : X86Inst<"mov", 0x8B, MRMSrcMem , Arg32>;            // R32 = [mem]

def MOVrm8  : X86Inst<"mov", 0x88, MRMDestMem, Arg8>;             // [mem] = R8
def MOVrm16 : X86Inst<"mov", 0x89, MRMDestMem, Arg16>, OpSize;    // [mem] = R16
def MOVrm32 : X86Inst<"mov", 0x89, MRMDestMem, Arg32>;            // [mem] = R32

//===----------------------------------------------------------------------===//
//  Fixed-Register Multiplication and Division Instructions...
//

// Extra precision multiplication
def MULr8  : X86Inst<"mul", 0xF6, MRMS4r, Arg8 >, Imp<[AL],[AX]>;               // AL,AH = AL*R8
def MULr16 : X86Inst<"mul", 0xF7, MRMS4r, Arg16>, Imp<[AX],[AX,DX]>, OpSize;    // AX,DX = AX*R16
def MULr32 : X86Inst<"mul", 0xF7, MRMS4r, Arg32>, Imp<[EAX],[EAX,EDX]>;         // EAX,EDX = EAX*R32

// unsigned division/remainder
def DIVr8  : X86Inst<"div", 0xF6, MRMS6r, Arg8 >, Imp<[AX],[AX]>;               // AX/r8 = AL,AH
def DIVr16 : X86Inst<"div", 0xF7, MRMS6r, Arg16>, Imp<[AX,DX],[AX,DX]>, OpSize; // DX:AX/r16 = AX,DX
def DIVr32 : X86Inst<"div", 0xF7, MRMS6r, Arg32>, Imp<[EAX,EDX],[EAX,EDX]>;     // EDX:EAX/r32 = EAX,EDX

// signed division/remainder
def IDIVr8 : X86Inst<"idiv",0xF6, MRMS7r, Arg8 >, Imp<[AX],[AX]>;               // AX/r8 = AL,AH
def IDIVr16: X86Inst<"idiv",0xF7, MRMS7r, Arg16>, Imp<[AX,DX],[AX,DX]>, OpSize; // DX:AX/r16 = AX,DX
def IDIVr32: X86Inst<"idiv",0xF7, MRMS7r, Arg32>, Imp<[EAX,EDX],[EAX,EDX]>;     // EDX:EAX/r32 = EAX,EDX

// Sign-extenders for division
def CBW    : X86Inst<"cbw", 0x98, RawFrm, Arg8 >, Imp<[AL],[AH]>;               // AX = signext(AL)
def CWD    : X86Inst<"cwd", 0x99, RawFrm, Arg8 >, Imp<[AX],[DX]>;               // DX:AX = signext(AX)
def CDQ    : X86Inst<"cdq", 0x99, RawFrm, Arg8 >, Imp<[EAX],[EDX]>;             // EDX:EAX = signext(EAX)


//===----------------------------------------------------------------------===//
//  Two address Instructions...
//
let isTwoAddress = 1 in {  // Define some helper classes to make defs shorter.
  class I2A8 <string n, bits<8> o, Format F> : X86Inst<n, o, F, Arg8>;
  class I2A16<string n, bits<8> o, Format F> : X86Inst<n, o, F, Arg16>;
  class I2A32<string n, bits<8> o, Format F> : X86Inst<n, o, F, Arg32>;
}

// Arithmetic...
def ADDrr8   : I2A8 <"add", 0x00, MRMDestReg>,         Pattern<(set R8 , (plus R8 , R8 ))>;
def ADDrr16  : I2A16<"add", 0x01, MRMDestReg>, OpSize, Pattern<(set R16, (plus R16, R16))>;
def ADDrr32  : I2A32<"add", 0x01, MRMDestReg>,         Pattern<(set R32, (plus R32, R32))>;
def ADDri8   : I2A8 <"add", 0x80, MRMS0r    >,         Pattern<(set R8 , (plus R8 , imm8 ))>;
def ADDri16  : I2A16<"add", 0x81, MRMS0r    >, OpSize, Pattern<(set R16, (plus R16, imm16))>;
def ADDri32  : I2A32<"add", 0x81, MRMS0r    >,         Pattern<(set R32, (plus R32, imm32))>;

def ADCrr32  : I2A32<"adc", 0x11, MRMDestReg>;                // R32 += imm32+Carry

def SUBrr8   : I2A8 <"sub", 0x28, MRMDestReg>,         Pattern<(set R8 , (minus R8 , R8 ))>;
def SUBrr16  : I2A16<"sub", 0x29, MRMDestReg>, OpSize, Pattern<(set R16, (minus R16, R16))>;
def SUBrr32  : I2A32<"sub", 0x29, MRMDestReg>,         Pattern<(set R32, (minus R32, R32))>;
def SUBri8   : I2A8 <"sub", 0x80, MRMS5r    >,         Pattern<(set R8 , (minus R8 , imm8 ))>;
def SUBri16  : I2A16<"sub", 0x81, MRMS5r    >, OpSize, Pattern<(set R16, (minus R16, imm16))>;
def SUBri32  : I2A32<"sub", 0x81, MRMS5r    >,         Pattern<(set R32, (minus R32, imm32))>;

def SBBrr32  : I2A32<"sbb", 0x19, MRMDestReg>;                // R32 -= R32+Carry

def IMULr16  : I2A16<"imul", 0xAF, MRMSrcReg>, TB, OpSize;    // R16 *= R16
def IMULr32  : I2A32<"imul", 0xAF, MRMSrcReg>, TB;            // R32 *= R32

// Logical operators...
def ANDrr8   : I2A8 <"and", 0x20, MRMDestReg>;                // R8  &= R8
def ANDrr16  : I2A16<"and", 0x21, MRMDestReg>, OpSize;        // R16 &= R16
def ANDrr32  : I2A32<"and", 0x21, MRMDestReg>;                // R32 &= R32
def ANDri8   : I2A8 <"and", 0x80, MRMS4r    >;                // R8  &= imm8
def ANDri16  : I2A16<"and", 0x81, MRMS4r    >, OpSize;        // R16 &= imm16
def ANDri32  : I2A32<"and", 0x81, MRMS4r    >;                // R32 &= imm32

def ORrr8    : I2A8 <"or" , 0x08, MRMDestReg>;                // R8  |= R8
def ORrr16   : I2A16<"or" , 0x09, MRMDestReg>, OpSize;        // R16 |= R16
def ORrr32   : I2A32<"or" , 0x09, MRMDestReg>;                // R32 |= R32
def ORri8    : I2A8 <"or" , 0x80, MRMS1r    >;                // R8  |= imm8
def ORri16   : I2A16<"or" , 0x81, MRMS1r    >, OpSize;        // R16 |= imm16
def ORri32   : I2A32<"or" , 0x81, MRMS1r    >;                // R32 |= imm32

def XORrr8   : I2A8 <"xor", 0x30, MRMDestReg>;                // R8  ^= R8
def XORrr16  : I2A16<"xor", 0x31, MRMDestReg>, OpSize;        // R16 ^= R16
def XORrr32  : I2A32<"xor", 0x31, MRMDestReg>;                // R32 ^= R32
def XORri8   : I2A8 <"xor", 0x80, MRMS6r    >;                // R8  ^= imm8
def XORri16  : I2A16<"xor", 0x81, MRMS6r    >, OpSize;        // R16 ^= imm16
def XORri32  : I2A32<"xor", 0x81, MRMS6r    >;                // R32 ^= imm32

// Test instructions are just like AND, except they don't generate a result.
def TESTrr8  : X86Inst<"test", 0x84, MRMDestReg, Arg8 >;          // flags = R8  & R8
def TESTrr16 : X86Inst<"test", 0x85, MRMDestReg, Arg16>, OpSize;  // flags = R16 & R16
def TESTrr32 : X86Inst<"test", 0x85, MRMDestReg, Arg32>;          // flags = R32 & R32
def TESTri8  : X86Inst<"test", 0xF6, MRMS0r    , Arg8 >;          // flags = R8  & imm8
def TESTri16 : X86Inst<"test", 0xF7, MRMS0r    , Arg16>, OpSize;  // flags = R16 & imm16
def TESTri32 : X86Inst<"test", 0xF7, MRMS0r    , Arg32>;          // flags = R32 & imm32

// Shift instructions
class UsesCL { list<Register> Uses = [CL]; bit printImplicitUses = 1; }

def SHLrr8   : I2A8 <"shl", 0xD2, MRMS4r    >        , UsesCL; // R8  <<= cl
def SHLrr16  : I2A8 <"shl", 0xD3, MRMS4r    >, OpSize, UsesCL; // R16 <<= cl
def SHLrr32  : I2A8 <"shl", 0xD3, MRMS4r    >        , UsesCL; // R32 <<= cl
def SHLir8   : I2A8 <"shl", 0xC0, MRMS4r    >;                 // R8  <<= imm8
def SHLir16  : I2A8 <"shl", 0xC1, MRMS4r    >, OpSize;         // R16 <<= imm16
def SHLir32  : I2A8 <"shl", 0xC1, MRMS4r    >;                 // R32 <<= imm32
def SHRrr8   : I2A8 <"shr", 0xD2, MRMS5r    >        , UsesCL; // R8  >>= cl
def SHRrr16  : I2A8 <"shr", 0xD3, MRMS5r    >, OpSize, UsesCL; // R16 >>= cl
def SHRrr32  : I2A8 <"shr", 0xD3, MRMS5r    >        , UsesCL; // R32 >>= cl
def SHRir8   : I2A8 <"shr", 0xC0, MRMS5r    >;                 // R8  >>= imm8
def SHRir16  : I2A8 <"shr", 0xC1, MRMS5r    >, OpSize;         // R16 >>= imm16
def SHRir32  : I2A8 <"shr", 0xC1, MRMS5r    >;                 // R32 >>= imm32
def SARrr8   : I2A8 <"sar", 0xD2, MRMS7r    >        , UsesCL; // R8  >>>= cl
def SARrr16  : I2A8 <"sar", 0xD3, MRMS7r    >, OpSize, UsesCL; // R16 >>>= cl
def SARrr32  : I2A8 <"sar", 0xD3, MRMS7r    >        , UsesCL; // R32 >>>= cl
def SARir8   : I2A8 <"sar", 0xC0, MRMS7r    >;                 // R8  >>>= imm8
def SARir16  : I2A8 <"sar", 0xC1, MRMS7r    >, OpSize;         // R16 >>>= imm16
def SARir32  : I2A8 <"sar", 0xC1, MRMS7r    >;                 // R32 >>>= imm32

def SHLDrr32 : I2A8 <"shld", 0xA5, MRMDestReg>, TB, UsesCL;   // R32 <<= R32,R32 cl
def SHLDir32 : I2A8 <"shld", 0xA4, MRMDestReg>, TB;           // R32 <<= R32,R32 imm8
def SHRDrr32 : I2A8 <"shrd", 0xAD, MRMDestReg>, TB, UsesCL;   // R32 >>= R32,R32 cl
def SHRDir32 : I2A8 <"shrd", 0xAC, MRMDestReg>, TB;           // R32 >>= R32,R32 imm8

// Condition code ops, incl. set if equal/not equal/...
def SAHF     : X86Inst<"sahf" , 0x9E, RawFrm, Arg8>, Imp<[AH],[]>;  // flags = AH
def SETBr    : X86Inst<"setb" , 0x92, MRMS0r, Arg8>, TB;            // R8 = <  unsign
def SETAEr   : X86Inst<"setae", 0x93, MRMS0r, Arg8>, TB;            // R8 = >= unsign
def SETEr    : X86Inst<"sete" , 0x94, MRMS0r, Arg8>, TB;            // R8 = ==
def SETNEr   : X86Inst<"setne", 0x95, MRMS0r, Arg8>, TB;            // R8 = !=
def SETBEr   : X86Inst<"setbe", 0x96, MRMS0r, Arg8>, TB;            // R8 = <= unsign
def SETAr    : X86Inst<"seta" , 0x97, MRMS0r, Arg8>, TB;            // R8 = >  signed
def SETLr    : X86Inst<"setl" , 0x9C, MRMS0r, Arg8>, TB;            // R8 = <  signed
def SETGEr   : X86Inst<"setge", 0x9D, MRMS0r, Arg8>, TB;            // R8 = >= signed
def SETLEr   : X86Inst<"setle", 0x9E, MRMS0r, Arg8>, TB;            // R8 = <= signed
def SETGr    : X86Inst<"setg" , 0x9F, MRMS0r, Arg8>, TB;            // R8 = <  signed

// Conditional moves.  These are modelled as X = cmovXX Y, Z.  Eventually
// register allocated to cmovXX XY, Z
def CMOVErr16 : I2A16<"cmove", 0x44, MRMSrcReg>, TB, OpSize;        // if ==, R16 = R16
def CMOVNErr32: I2A32<"cmovne",0x45, MRMSrcReg>, TB;                // if !=, R32 = R32

// Integer comparisons
def CMPrr8  : X86Inst<"cmp", 0x38, MRMDestReg, Arg8 >;              // compare R8, R8
def CMPrr16 : X86Inst<"cmp", 0x39, MRMDestReg, Arg16>, OpSize;      // compare R16, R16
def CMPrr32 : X86Inst<"cmp", 0x39, MRMDestReg, Arg32>;              // compare R32, R32
def CMPri8  : X86Inst<"cmp", 0x80, MRMS7r    , Arg8 >;              // compare R8, imm8
def CMPri16 : X86Inst<"cmp", 0x81, MRMS7r    , Arg16>, OpSize;      // compare R16, imm16
def CMPri32 : X86Inst<"cmp", 0x81, MRMS7r    , Arg32>;              // compare R32, imm32

// Sign/Zero extenders
def MOVSXr16r8 : X86Inst<"movsx", 0xBE, MRMSrcReg, Arg8>, TB, OpSize; // R16 = signext(R8)
def MOVSXr32r8 : X86Inst<"movsx", 0xBE, MRMSrcReg, Arg8>, TB;         // R32 = signext(R8)
def MOVSXr32r16: X86Inst<"movsx", 0xBF, MRMSrcReg, Arg8>, TB;         // R32 = signext(R16)
def MOVZXr16r8 : X86Inst<"movzx", 0xB6, MRMSrcReg, Arg8>, TB, OpSize; // R16 = zeroext(R8)
def MOVZXr32r8 : X86Inst<"movzx", 0xB6, MRMSrcReg, Arg8>, TB;         // R32 = zeroext(R8)
def MOVZXr32r16: X86Inst<"movzx", 0xB7, MRMSrcReg, Arg8>, TB;         // R32 = zeroext(R16)


//===----------------------------------------------------------------------===//
// Floating point support
//===----------------------------------------------------------------------===//

// FIXME: These need to indicate mod/ref sets for FP regs... & FP 'TOP'

// Floating point pseudo instructions...
class FPInst<string n, bits<8> o, Format F, ArgType t, FPFormat fp>
  : X86Inst<n, o, F, t> { let FPForm = fp; let FPFormBits = FPForm.Value; }

def FpMOV : FPInst<"FMOV", 0, Pseudo, ArgF80, SpecialFP>;   // f1 = fmov f2
def FpADD : FPInst<"FADD", 0, Pseudo, ArgF80, TwoArgFP>;    // f1 = fadd f2, f3
def FpSUB : FPInst<"FSUB", 0, Pseudo, ArgF80, TwoArgFP>;    // f1 = fsub f2, f3
def FpMUL : FPInst<"FMUL", 0, Pseudo, ArgF80, TwoArgFP>;    // f1 = fmul f2, f3
def FpDIV : FPInst<"FDIV", 0, Pseudo, ArgF80, TwoArgFP>;    // f1 = fdiv f2, f3

def FpUCOM : FPInst<"FUCOM", 0, Pseudo, ArgF80, TwoArgFP>;  // FPSW = fucom f1, f2

def FpGETRESULT : FPInst<"FGETRESULT",0, Pseudo, ArgF80, SpecialFP>;  // FPR = ST(0)

def FpSETRESULT : FPInst<"FSETRESULT",0, Pseudo, ArgF80, SpecialFP>;  // ST(0) = FPR

// Floating point loads & stores...
def FLDrr   : FPInst<"fld"   , 0xC0, AddRegFrm, ArgF80, NotFP>, D9;   // push(ST(i))
def FLDr32  : FPInst<"fld"   , 0xD9, MRMS0m   , ArgF32, ZeroArgFP>;        // load float
def FLDr64  : FPInst<"fld"   , 0xDD, MRMS0m   , ArgF64, ZeroArgFP>;        // load double
def FLDr80  : FPInst<"fld"   , 0xDB, MRMS5m   , ArgF80, ZeroArgFP>;        // load extended
def FILDr16 : FPInst<"fild"  , 0xDF, MRMS0m   , Arg16 , ZeroArgFP>;        // load signed short
def FILDr32 : FPInst<"fild"  , 0xDB, MRMS0m   , Arg32 , ZeroArgFP>;        // load signed int
def FILDr64 : FPInst<"fild"  , 0xDF, MRMS5m   , Arg64 , ZeroArgFP>;        // load signed long

def FSTr32   : FPInst<"fst" , 0xD9, MRMS2m   , ArgF32, OneArgFP>;          // store float
def FSTr64   : FPInst<"fst" , 0xDD, MRMS2m   , ArgF64, OneArgFP>;          // store double
def FSTPr32  : FPInst<"fstp", 0xD9, MRMS3m   , ArgF32, OneArgFP>;          // store float, pop
def FSTPr64  : FPInst<"fstp", 0xDD, MRMS3m   , ArgF64, OneArgFP>;          // store double, pop
def FSTPr80  : FPInst<"fstp", 0xDB, MRMS7m   , ArgF80, OneArgFP>;          // store extended, pop
def FSTrr    : FPInst<"fst" , 0xD0, AddRegFrm, ArgF80, NotFP   >, DD;      // ST(i) = ST(0)
def FSTPrr   : FPInst<"fstp", 0xD8, AddRegFrm, ArgF80, NotFP   >, DD;      // ST(i) = ST(0), pop

def FISTr16  : FPInst<"fist",    0xDF, MRMS2m, Arg16 , OneArgFP>;          // store signed short
def FISTr32  : FPInst<"fist",    0xDB, MRMS2m, Arg32 , OneArgFP>;          // store signed int
def FISTPr16 : FPInst<"fistp",   0xDF, MRMS3m, Arg16 , NotFP   >;          // store signed short, pop
def FISTPr32 : FPInst<"fistp",   0xDB, MRMS3m, Arg32 , NotFP   >;          // store signed int, pop
def FISTPr64 : FPInst<"fistpll", 0xDF, MRMS7m, Arg64 , OneArgFP>;          // store signed long, pop

def FXCH     : FPInst<"fxch",    0xC8, AddRegFrm, ArgF80, NotFP>, D9;      // fxch ST(i), ST(0)

// Floating point constant loads...
def FLD0 : FPInst<"fldz", 0xEE, RawFrm, ArgF80, ZeroArgFP>, D9;
def FLD1 : FPInst<"fld1", 0xE8, RawFrm, ArgF80, ZeroArgFP>, D9;

// Binary arithmetic operations...
class FPST0rInst<string n, bits<8> o>
  : X86Inst<n, o, AddRegFrm, ArgF80>, D8 {
  list<Register> Uses = [ST0];
  list<Register> Defs = [ST0];
}
class FPrST0Inst<string n, bits<8> o>
  : X86Inst<n, o, AddRegFrm, ArgF80>, DC {
  bit printImplicitUses = 1;
  list<Register> Uses = [ST0];
}
class FPrST0PInst<string n, bits<8> o>
  : X86Inst<n, o, AddRegFrm, ArgF80>, DE {
  list<Register> Uses = [ST0];
}

def FADDST0r   : FPST0rInst <"fadd",    0xC0>;
def FADDrST0   : FPrST0Inst <"fadd",    0xC0>;
def FADDPrST0  : FPrST0PInst<"faddp",   0xC0>;

def FSUBRST0r  : FPST0rInst <"fsubr",   0xE8>;
def FSUBrST0   : FPrST0Inst <"fsub",    0xE8>;
def FSUBPrST0  : FPrST0PInst<"fsubp",   0xE8>;

def FSUBST0r   : FPST0rInst <"fsub",    0xE0>;
def FSUBRrST0  : FPrST0Inst <"fsubr",   0xE0>;
def FSUBRPrST0 : FPrST0PInst<"fsubrp",  0xE0>;

def FMULST0r   : FPST0rInst <"fmul",    0xC8>;
def FMULrST0   : FPrST0Inst <"fmul",    0xC8>;
def FMULPrST0  : FPrST0PInst<"fmulp",   0xC8>;

def FDIVRST0r  : FPST0rInst <"fdivr",   0xF8>;
def FDIVrST0   : FPrST0Inst <"fdiv",    0xF8>;
def FDIVPrST0  : FPrST0PInst<"fdivp",   0xF8>;

def FDIVST0r   : FPST0rInst <"fdiv",    0xF0>;   // ST(0) = ST(0) / ST(i)
def FDIVRrST0  : FPrST0Inst <"fdivr",   0xF0>;   // ST(i) = ST(0) / ST(i)
def FDIVRPrST0 : FPrST0PInst<"fdivrp",  0xF0>;   // ST(i) = ST(0) / ST(i), pop

// Floating point compares
def FUCOMr    : X86Inst<"fucom"  , 0xE0, AddRegFrm, ArgF80>, DD, Imp<[ST0],[]>;  // FPSW = compare ST(0) with ST(i)
def FUCOMPr   : X86Inst<"fucomp" , 0xE8, AddRegFrm, ArgF80>, DD, Imp<[ST0],[]>;  // FPSW = compare ST(0) with ST(i), pop
def FUCOMPPr  : X86Inst<"fucompp", 0xE9, RawFrm   , ArgF80>, DA, Imp<[ST0],[]>;  // compare ST(0) with ST(1), pop, pop

// Floating point flag ops
def FNSTSWr8  : X86Inst<"fnstsw" , 0xE0, RawFrm   , ArgF80>, DF, Imp<[],[AX]>;   // AX = fp flags
def FNSTCWm16 : X86Inst<"fnstcw" , 0xD9, MRMS7m   , Arg16 >;                     // [mem16] = X87 control world
def FLDCWm16  : X86Inst<"fldcw"  , 0xD9, MRMS5m   , Arg16 >;                     // X87 control world = [mem16]