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
+def MRM0r : Format<16>; def MRM1r : Format<17>; def MRM2r : Format<18>;
+def MRM3r : Format<19>; def MRM4r : Format<20>; def MRM5r : Format<21>;
+def MRM6r : Format<22>; def MRM7r : Format<23>;
+def MRM0m : Format<24>; def MRM1m : Format<25>; def MRM2m : Format<26>;
+def MRM3m : Format<27>; def MRM4m : Format<28>; def MRM5m : Format<29>;
+def MRM6m : Format<30>; def MRM7m : Format<31>;
+
+// ImmType - This specifies the immediate 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> {
+class ImmType<bits<2> val> {
+ bits<2> Value = val;
+}
+def NoImm : ImmType<0>;
+def Imm8 : ImmType<1>;
+def Imm16 : ImmType<2>;
+def Imm32 : ImmType<3>;
+
+// MemType - This specifies the immediate 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 MemType<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>;
+def NoMem : MemType<0>;
+def Mem8 : MemType<1>;
+def Mem16 : MemType<2>;
+def Mem32 : MemType<3>;
+def Mem64 : MemType<4>;
+def Mem80 : MemType<5>;
+def Mem128 : MemType<6>;
// FPFormat - This specifies what form this FP instruction has. This is used by
// the Floating-Point stackifier pass.
def OneArgFP : FPFormat<2>;
def OneArgFPRW : FPFormat<3>;
def TwoArgFP : FPFormat<4>;
-def SpecialFP : FPFormat<5>;
+def CondMovFP : FPFormat<5>;
+def SpecialFP : FPFormat<6>;
-class X86Inst<string nam, bits<8> opcod, Format f, ArgType a> : Instruction {
+class X86Inst<string nam, bits<8> opcod, Format f, MemType m, ImmType i> : 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;
+ MemType MemT = m;
+ bits<3> MemTypeBits = MemT.Value;
+ ImmType ImmT = i;
+ bits<2> ImmTypeBits = ImmT.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?
+ bit printImplicitUsesBefore = 0; // Should we print implicit uses before this inst?
+ bit printImplicitUsesAfter = 0; // Should we print implicit uses after this inst?
bits<4> Prefix = 0; // Which prefix byte does this inst have?
FPFormat FPForm; // What flavor of FP instruction is this?
class DF { bits<4> Prefix = 10; }
+//===----------------------------------------------------------------------===//
+// Instruction templates...
+
+class I<string n, bits<8> o, Format f> : X86Inst<n, o, f, NoMem, NoImm>;
+
+class Im<string n, bits<8> o, Format f, MemType m> : X86Inst<n, o, f, m, NoImm>;
+class Im8 <string n, bits<8> o, Format f> : Im<n, o, f, Mem8 >;
+class Im16<string n, bits<8> o, Format f> : Im<n, o, f, Mem16>;
+class Im32<string n, bits<8> o, Format f> : Im<n, o, f, Mem32>;
+
+class Ii<string n, bits<8> o, Format f, ImmType i> : X86Inst<n, o, f, NoMem, i>;
+class Ii8 <string n, bits<8> o, Format f> : Ii<n, o, f, Imm8 >;
+class Ii16<string n, bits<8> o, Format f> : Ii<n, o, f, Imm16>;
+class Ii32<string n, bits<8> o, Format f> : Ii<n, o, f, Imm32>;
+
+class Im8i8 <string n, bits<8> o, Format f> : X86Inst<n, o, f, Mem8 , Imm8 >;
+class Im16i16<string n, bits<8> o, Format f> : X86Inst<n, o, f, Mem16, Imm16>;
+class Im32i32<string n, bits<8> o, Format f> : X86Inst<n, o, f, Mem32, Imm32>;
+
+class Im16i8<string n, bits<8> o, Format f> : X86Inst<n, o, f, Mem16, Imm8>;
+class Im32i8<string n, bits<8> o, Format f> : X86Inst<n, o, f, Mem32, Imm8>;
+
+// Helper for shift instructions
+class UsesCL { list<Register> Uses = [CL]; bit printImplicitUsesAfter = 1; }
//===----------------------------------------------------------------------===//
// Instruction list...
//
-def PHI : X86Inst<"PHI", 0, Pseudo, NoArg>; // PHI node...
+def PHI : I<"PHI", 0, Pseudo>; // PHI node...
-def NOOP : X86Inst<"nop", 0x90, RawFrm, NoArg>; // nop
+def NOOP : I<"nop", 0x90, RawFrm>; // nop
-def ADJCALLSTACKDOWN : X86Inst<"ADJCALLSTACKDOWN", 0, Pseudo, NoArg>;
-def ADJCALLSTACKUP : X86Inst<"ADJCALLSTACKUP", 0, Pseudo, NoArg>;
-def IMPLICIT_USE : X86Inst<"IMPLICIT_USE", 0, Pseudo, NoArg>;
-def IMPLICIT_DEF : X86Inst<"IMPLICIT_DEF", 0, Pseudo, NoArg>;
+def ADJCALLSTACKDOWN : I<"ADJCALLSTACKDOWN", 0, Pseudo>;
+def ADJCALLSTACKUP : I<"ADJCALLSTACKUP", 0, Pseudo>;
+def IMPLICIT_USE : I<"IMPLICIT_USE", 0, Pseudo>;
+def IMPLICIT_DEF : I<"IMPLICIT_DEF", 0, Pseudo>;
let isTerminator = 1 in
let Defs = [FP0, FP1, FP2, FP3, FP4, FP5, FP6] in
- def FP_REG_KILL : X86Inst<"FP_REG_KILL", 0, Pseudo, NoArg>;
+ def FP_REG_KILL : I<"FP_REG_KILL", 0, Pseudo>;
//===----------------------------------------------------------------------===//
// Control Flow Instructions...
//
// Return instruction...
let isTerminator = 1, isReturn = 1 in
- def RET : X86Inst<"ret", 0xC3, RawFrm, NoArg>, Pattern<(retvoid)>;
+ def RET : I<"ret", 0xC3, RawFrm>, Pattern<(retvoid)>;
// 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>;
+ class IBr<string name, bits<8> opcode> : I<name, opcode, RawFrm>;
def JMP : IBr<"jmp", 0xE9>, Pattern<(br basicblock)>;
def JB : IBr<"jb" , 0x82>, TB;
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>;
+ def CALLpcrel32 : I <"call", 0xE8, RawFrm>;
+ def CALL32r : I <"call", 0xFF, MRM2r>;
+ def CALL32m : Im32<"call", 0xFF, MRM2m>;
}
//===----------------------------------------------------------------------===//
// Miscellaneous Instructions...
//
-def LEAVE : X86Inst<"leave", 0xC9, RawFrm, NoArg>, Imp<[EBP,ESP],[EBP,ESP]>;
-def POPr32 : X86Inst<"pop", 0x58, AddRegFrm, Arg32>, Imp<[ESP],[ESP]>;
+def LEAVE : I<"leave", 0xC9, RawFrm>, Imp<[EBP,ESP],[EBP,ESP]>;
+def POP32r : I<"pop", 0x58, AddRegFrm>, Imp<[ESP],[ESP]>;
-let isTwoAddress = 1 in // R32 = bswap R32
- def BSWAPr32 : X86Inst<"bswap", 0xC8, AddRegFrm, Arg32>, TB;
+let isTwoAddress = 1 in // R32 = bswap R32
+ def BSWAP32r : I<"bswap", 0xC8, AddRegFrm>, 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 XCHG8rr : I <"xchg", 0x86, MRMDestReg>; // xchg R8, R8
+def XCHG16rr : I <"xchg", 0x87, MRMDestReg>, OpSize; // xchg R16, R16
+def XCHG32rr : I <"xchg", 0x87, MRMDestReg>; // xchg R32, R32
+def XCHG8mr : Im8 <"xchg", 0x86, MRMDestMem>; // xchg [mem8], R8
+def XCHG16mr : Im16<"xchg", 0x87, MRMDestMem>, OpSize; // xchg [mem16], R16
+def XCHG32mr : Im32<"xchg", 0x87, MRMDestMem>; // xchg [mem32], R32
+def XCHG8rm : Im8 <"xchg", 0x86, MRMSrcMem >; // xchg R8, [mem8]
+def XCHG16rm : Im16<"xchg", 0x87, MRMSrcMem >, OpSize; // xchg R16, [mem16]
+def XCHG32rm : Im32<"xchg", 0x87, MRMSrcMem >; // xchg R32, [mem32]
-def LEAr16 : X86Inst<"lea", 0x8D, MRMSrcMem, Arg16>, OpSize; // R16 = lea [mem]
-def LEAr32 : X86Inst<"lea", 0x8D, MRMSrcMem, Arg32>; // R32 = lea [mem]
+def LEA16r : Im32<"lea", 0x8D, MRMSrcMem>, OpSize; // R16 = lea [mem]
+def LEA32r : Im32<"lea", 0x8D, MRMSrcMem>; // R32 = lea [mem]
-def REP_MOVSB : X86Inst<"rep movsb", 0xA4, RawFrm, NoArg>, REP,
+def REP_MOVSB : I<"rep movsb", 0xA4, RawFrm>, REP,
Imp<[ECX,EDI,ESI], [ECX,EDI,ESI]>;
-def REP_MOVSW : X86Inst<"rep movsw", 0xA5, RawFrm, NoArg>, REP, OpSize,
+def REP_MOVSW : I<"rep movsw", 0xA5, RawFrm>, REP, OpSize,
Imp<[ECX,EDI,ESI], [ECX,EDI,ESI]>;
-def REP_MOVSD : X86Inst<"rep movsd", 0xA5, RawFrm, NoArg>, REP,
+def REP_MOVSD : I<"rep movsd", 0xA5, RawFrm>, REP,
Imp<[ECX,EDI,ESI], [ECX,EDI,ESI]>;
-def REP_STOSB : X86Inst<"rep stosb", 0xAA, RawFrm, NoArg>, REP,
+def REP_STOSB : I<"rep stosb", 0xAA, RawFrm>, REP,
Imp<[AL,ECX,EDI], [ECX,EDI]>;
-def REP_STOSW : X86Inst<"rep stosw", 0xAB, RawFrm, NoArg>, REP, OpSize,
+def REP_STOSW : I<"rep stosw", 0xAB, RawFrm>, REP, OpSize,
Imp<[AX,ECX,EDI], [ECX,EDI]>;
-def REP_STOSD : X86Inst<"rep stosd", 0xAB, RawFrm, NoArg>, REP,
+def REP_STOSD : I<"rep stosd", 0xAB, RawFrm>, REP,
Imp<[EAX,ECX,EDI], [ECX,EDI]>;
//===----------------------------------------------------------------------===//
// 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 , imm )>;
-def MOVir16 : X86Inst<"mov", 0xB8, AddRegFrm , Arg16>, OpSize, Pattern<(set R16, imm)>;
-def MOVir32 : X86Inst<"mov", 0xB8, AddRegFrm , Arg32>, Pattern<(set R32, imm)>;
-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 MOV8rr : I <"mov", 0x88, MRMDestReg>, Pattern<(set R8 , R8 )>;
+def MOV16rr : I <"mov", 0x89, MRMDestReg>, OpSize, Pattern<(set R16, R16)>;
+def MOV32rr : I <"mov", 0x89, MRMDestReg>, Pattern<(set R32, R32)>;
+def MOV8ri : Ii8 <"mov", 0xB0, AddRegFrm >, Pattern<(set R8 , imm )>;
+def MOV16ri : Ii16 <"mov", 0xB8, AddRegFrm >, OpSize, Pattern<(set R16, imm)>;
+def MOV32ri : Ii32 <"mov", 0xB8, AddRegFrm >, Pattern<(set R32, imm)>;
+def MOV8mi : Im8i8 <"mov", 0xC6, MRM0m >; // [mem8] = imm8
+def MOV16mi : Im16i16<"mov", 0xC7, MRM0m >, OpSize; // [mem16] = imm16
+def MOV32mi : Im32i32<"mov", 0xC7, MRM0m >; // [mem32] = imm32
+
+def MOV8rm : Im8 <"mov", 0x8A, MRMSrcMem>; // R8 = [mem8]
+def MOV16rm : Im16 <"mov", 0x8B, MRMSrcMem>, OpSize, // R16 = [mem16]
Pattern<(set R16, (load (plus R32, (plus (times imm, R32), imm))))>;
-def MOVmr32 : X86Inst<"mov", 0x8B, MRMSrcMem , Arg32>, // R32 = [mem]
+def MOV32rm : Im32 <"mov", 0x8B, MRMSrcMem>, // R32 = [mem32]
Pattern<(set R32, (load (plus R32, (plus (times imm, R32), imm))))>;
-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
+def MOV8mr : Im8 <"mov", 0x88, MRMDestMem>; // [mem8] = R8
+def MOV16mr : Im16 <"mov", 0x89, MRMDestMem>, OpSize; // [mem16] = R16
+def MOV32mr : Im32 <"mov", 0x89, MRMDestMem>; // [mem32] = 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
+def MUL8r : I <"mul", 0xF6, MRM4r>, Imp<[AL],[AX]>; // AL,AH = AL*R8
+def MUL16r : I <"mul", 0xF7, MRM4r>, Imp<[AX],[AX,DX]>, OpSize; // AX,DX = AX*R16
+def MUL32r : I <"mul", 0xF7, MRM4r>, Imp<[EAX],[EAX,EDX]>; // EAX,EDX = EAX*R32
+def MUL8m : Im8 <"mul", 0xF6, MRM4m>, Imp<[AL],[AX]>; // AL,AH = AL*[mem8]
+def MUL16m : Im16<"mul", 0xF7, MRM4m>, Imp<[AX],[AX,DX]>, OpSize; // AX,DX = AX*[mem16]
+def MUL32m : Im32<"mul", 0xF7, MRM4m>, Imp<[EAX],[EAX,EDX]>; // EAX,EDX = EAX*[mem32]
// 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
+def DIV8r : I <"div", 0xF6, MRM6r>, Imp<[AX],[AX]>; // AX/r8 = AL,AH
+def DIV16r : I <"div", 0xF7, MRM6r>, Imp<[AX,DX],[AX,DX]>, OpSize; // DX:AX/r16 = AX,DX
+def DIV32r : I <"div", 0xF7, MRM6r>, Imp<[EAX,EDX],[EAX,EDX]>; // EDX:EAX/r32 = EAX,EDX
+def DIV8m : Im8 <"div", 0xF6, MRM6m>, Imp<[AX],[AX]>; // AX/[mem8] = AL,AH
+def DIV16m : Im16<"div", 0xF7, MRM6m>, Imp<[AX,DX],[AX,DX]>, OpSize; // DX:AX/[mem16] = AX,DX
+def DIV32m : Im32<"div", 0xF7, MRM6m>, Imp<[EAX,EDX],[EAX,EDX]>; // EDX:EAX/[mem32] = 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
+def IDIV8r : I <"idiv",0xF6, MRM7r>, Imp<[AX],[AX]>; // AX/r8 = AL,AH
+def IDIV16r: I <"idiv",0xF7, MRM7r>, Imp<[AX,DX],[AX,DX]>, OpSize; // DX:AX/r16 = AX,DX
+def IDIV32r: I <"idiv",0xF7, MRM7r>, Imp<[EAX,EDX],[EAX,EDX]>; // EDX:EAX/r32 = EAX,EDX
+def IDIV8m : Im8 <"idiv",0xF6, MRM7m>, Imp<[AX],[AX]>; // AX/[mem8] = AL,AH
+def IDIV16m: Im16<"idiv",0xF7, MRM7m>, Imp<[AX,DX],[AX,DX]>, OpSize; // DX:AX/[mem16] = AX,DX
+def IDIV32m: Im32<"idiv",0xF7, MRM7m>, Imp<[EAX,EDX],[EAX,EDX]>; // EDX:EAX/[mem32] = 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)
+def CBW : I<"cbw", 0x98, RawFrm >, Imp<[AL],[AH]>; // AX = signext(AL)
+def CWD : I<"cwd", 0x99, RawFrm >, Imp<[AX],[DX]>; // DX:AX = signext(AX)
+def CDQ : I<"cdq", 0x99, RawFrm >, 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>;
-}
+let isTwoAddress = 1 in {
+
+// Conditional moves
+def CMOVB16rr : I <"cmovb", 0x42, MRMSrcReg>, TB, OpSize; // if <u, R16 = R16
+def CMOVB16rm : Im16<"cmovb", 0x42, MRMSrcMem>, TB, OpSize; // if <u, R16 = [mem16]
+def CMOVB32rr : I <"cmovb", 0x42, MRMSrcReg>, TB; // if <u, R32 = R32
+def CMOVB32rm : Im32<"cmovb", 0x42, MRMSrcMem>, TB; // if <u, R32 = [mem32]
+
+def CMOVAE16rr: I <"cmovae", 0x43, MRMSrcReg>, TB, OpSize; // if >=u, R16 = R16
+def CMOVAE16rm: Im16<"cmovae", 0x43, MRMSrcMem>, TB, OpSize; // if >=u, R16 = [mem16]
+def CMOVAE32rr: I <"cmovae", 0x43, MRMSrcReg>, TB; // if >=u, R32 = R32
+def CMOVAE32rm: Im32<"cmovae", 0x43, MRMSrcMem>, TB; // if >=u, R32 = [mem32]
+
+def CMOVE16rr : I <"cmove", 0x44, MRMSrcReg>, TB, OpSize; // if ==, R16 = R16
+def CMOVE16rm : Im16<"cmove", 0x44, MRMSrcMem>, TB, OpSize; // if ==, R16 = [mem16]
+def CMOVE32rr : I <"cmove", 0x44, MRMSrcReg>, TB; // if ==, R32 = R32
+def CMOVE32rm : Im32<"cmove", 0x44, MRMSrcMem>, TB; // if ==, R32 = [mem32]
+
+def CMOVNE16rr: I <"cmovne",0x45, MRMSrcReg>, TB, OpSize; // if !=, R16 = R16
+def CMOVNE16rm: Im16<"cmovne",0x45, MRMSrcMem>, TB, OpSize; // if !=, R16 = [mem16]
+def CMOVNE32rr: I <"cmovne",0x45, MRMSrcReg>, TB; // if !=, R32 = R32
+def CMOVNE32rm: Im32<"cmovne",0x45, MRMSrcMem>, TB; // if !=, R32 = [mem32]
+
+def CMOVBE16rr: I <"cmovbe",0x46, MRMSrcReg>, TB, OpSize; // if <=u, R16 = R16
+def CMOVBE16rm: Im16<"cmovbe",0x46, MRMSrcMem>, TB, OpSize; // if <=u, R16 = [mem16]
+def CMOVBE32rr: I <"cmovbe",0x46, MRMSrcReg>, TB; // if <=u, R32 = R32
+def CMOVBE32rm: Im32<"cmovbe",0x46, MRMSrcMem>, TB; // if <=u, R32 = [mem32]
+
+def CMOVA16rr : I <"cmova", 0x47, MRMSrcReg>, TB, OpSize; // if >u, R16 = R16
+def CMOVA16rm : Im16<"cmova", 0x47, MRMSrcMem>, TB, OpSize; // if >u, R16 = [mem16]
+def CMOVA32rr : I <"cmova", 0x47, MRMSrcReg>, TB; // if >u, R32 = R32
+def CMOVA32rm : Im32<"cmova", 0x47, MRMSrcMem>, TB; // if >u, R32 = [mem32]
+
+def CMOVS16rr : I <"cmovs", 0x48, MRMSrcReg>, TB, OpSize; // if signed, R16 = R16
+def CMOVS16rm : Im16<"cmovs", 0x48, MRMSrcMem>, TB, OpSize; // if signed, R16 = [mem16]
+def CMOVS32rr : I <"cmovs", 0x48, MRMSrcReg>, TB; // if signed, R32 = R32
+def CMOVS32rm : Im32<"cmovs", 0x48, MRMSrcMem>, TB; // if signed, R32 = [mem32]
+
+def CMOVNS16rr: I <"cmovns",0x49, MRMSrcReg>, TB, OpSize; // if !signed, R16 = R16
+def CMOVNS16rm: Im16<"cmovns",0x49, MRMSrcMem>, TB, OpSize; // if !signed, R16 = [mem16]
+def CMOVNS32rr: I <"cmovns",0x49, MRMSrcReg>, TB; // if !signed, R32 = R32
+def CMOVNS32rm: Im32<"cmovns",0x49, MRMSrcMem>, TB; // if !signed, R32 = [mem32]
+
+def CMOVL16rr : I <"cmovl", 0x4C, MRMSrcReg>, TB, OpSize; // if <s, R16 = R16
+def CMOVL16rm : Im16<"cmovl", 0x4C, MRMSrcMem>, TB, OpSize; // if <s, R16 = [mem16]
+def CMOVL32rr : I <"cmovl", 0x4C, MRMSrcReg>, TB; // if <s, R32 = R32
+def CMOVL32rm : Im32<"cmovl", 0x4C, MRMSrcMem>, TB; // if <s, R32 = [mem32]
+
+def CMOVGE16rr: I <"cmovge",0x4D, MRMSrcReg>, TB, OpSize; // if >=s, R16 = R16
+def CMOVGE16rm: Im16<"cmovge",0x4D, MRMSrcMem>, TB, OpSize; // if >=s, R16 = [mem16]
+def CMOVGE32rr: I <"cmovge",0x4D, MRMSrcReg>, TB; // if >=s, R32 = R32
+def CMOVGE32rm: Im32<"cmovge",0x4D, MRMSrcMem>, TB; // if >=s, R32 = [mem32]
+
+def CMOVLE16rr: I <"cmovle",0x4E, MRMSrcReg>, TB, OpSize; // if <=s, R16 = R16
+def CMOVLE16rm: Im16<"cmovle",0x4E, MRMSrcMem>, TB, OpSize; // if <=s, R16 = [mem16]
+def CMOVLE32rr: I <"cmovle",0x4E, MRMSrcReg>, TB; // if <=s, R32 = R32
+def CMOVLE32rm: Im32<"cmovle",0x4E, MRMSrcMem>, TB; // if <=s, R32 = [mem32]
+
+def CMOVG16rr : I <"cmovg", 0x4F, MRMSrcReg>, TB, OpSize; // if >s, R16 = R16
+def CMOVG16rm : Im16<"cmovg", 0x4F, MRMSrcMem>, TB, OpSize; // if >s, R16 = [mem16]
+def CMOVG32rr : I <"cmovg", 0x4F, MRMSrcReg>, TB; // if >s, R32 = R32
+def CMOVG32rm : Im32<"cmovg", 0x4F, MRMSrcMem>, TB; // if >s, R32 = [mem32]
// unary instructions
-def NEGr8 : I2A8 <"neg", 0xF6, MRMS3r>; // R8 = -R8 = 0-R8
-def NEGr16 : I2A16<"neg", 0xF7, MRMS3r>, OpSize; // R16 = -R16 = 0-R16
-def NEGr32 : I2A32<"neg", 0xF7, MRMS3r>; // R32 = -R32 = 0-R32
-def NOTr8 : I2A8 <"not", 0xF6, MRMS2r>; // R8 = ~R8 = R8^-1
-def NOTr16 : I2A16<"not", 0xF7, MRMS2r>, OpSize; // R16 = ~R16 = R16^-1
-def NOTr32 : I2A32<"not", 0xF7, MRMS2r>; // R32 = ~R32 = R32^-1
+def NEG8r : I <"neg", 0xF6, MRM3r>; // R8 = -R8 = 0-R8
+def NEG16r : I <"neg", 0xF7, MRM3r>, OpSize; // R16 = -R16 = 0-R16
+def NEG32r : I <"neg", 0xF7, MRM3r>; // R32 = -R32 = 0-R32
+def NEG8m : Im8 <"neg", 0xF6, MRM3m>; // [mem8] = -[mem8] = 0-[mem8]
+def NEG16m : Im16<"neg", 0xF7, MRM3m>, OpSize; // [mem16] = -[mem16] = 0-[mem16]
+def NEG32m : Im32<"neg", 0xF7, MRM3m>; // [mem32] = -[mem32] = 0-[mem32]
+
+def NOT8r : I <"not", 0xF6, MRM2r>; // R8 = ~R8 = R8^-1
+def NOT16r : I <"not", 0xF7, MRM2r>, OpSize; // R16 = ~R16 = R16^-1
+def NOT32r : I <"not", 0xF7, MRM2r>; // R32 = ~R32 = R32^-1
+def NOT8m : Im8 <"not", 0xF6, MRM2m>; // [mem8] = ~[mem8] = [mem8^-1]
+def NOT16m : Im16<"not", 0xF7, MRM2m>, OpSize; // [mem16] = ~[mem16] = [mem16^-1]
+def NOT32m : Im32<"not", 0xF7, MRM2m>; // [mem32] = ~[mem32] = [mem32^-1]
+
+def INC8r : I <"inc", 0xFE, MRM0r>; // ++R8
+def INC16r : I <"inc", 0xFF, MRM0r>, OpSize; // ++R16
+def INC32r : I <"inc", 0xFF, MRM0r>; // ++R32
+def INC8m : Im8 <"inc", 0xFE, MRM0m>; // ++R8
+def INC16m : Im16<"inc", 0xFF, MRM0m>, OpSize; // ++R16
+def INC32m : Im32<"inc", 0xFF, MRM0m>; // ++R32
+
+def DEC8r : I <"dec", 0xFE, MRM1r>; // --R8
+def DEC16r : I <"dec", 0xFF, MRM1r>, OpSize; // --R16
+def DEC32r : I <"dec", 0xFF, MRM1r>; // --R32
+def DEC8m : Im8 <"dec", 0xFE, MRM1m>; // --[mem8]
+def DEC16m : Im16<"dec", 0xFF, MRM1m>, OpSize; // --[mem16]
+def DEC32m : Im32<"dec", 0xFF, MRM1m>; // --[mem32]
-def INCr8 : I2A8 <"inc", 0xFE, MRMS0r>; // R8 = R8 +1
-def INCr16 : I2A16<"inc", 0xFF, MRMS0r>, OpSize; // R16 = R16+1
-def INCr32 : I2A32<"inc", 0xFF, MRMS0r>; // R32 = R32+1
-def DECr8 : I2A8 <"dec", 0xFE, MRMS1r>; // R8 = R8 -1
-def DECr16 : I2A16<"dec", 0xFF, MRMS1r>, OpSize; // R16 = R16-1
-def DECr32 : I2A32<"dec", 0xFF, MRMS1r>; // R32 = R32-1
+// Logical operators...
+def AND8rr : I <"and", 0x20, MRMDestReg>, Pattern<(set R8 , (and R8 , R8 ))>;
+def AND16rr : I <"and", 0x21, MRMDestReg>, OpSize, Pattern<(set R16, (and R16, R16))>;
+def AND32rr : I <"and", 0x21, MRMDestReg>, Pattern<(set R32, (and R32, R32))>;
+def AND8mr : Im8 <"and", 0x20, MRMDestMem>; // [mem8] &= R8
+def AND16mr : Im16 <"and", 0x21, MRMDestMem>, OpSize; // [mem16] &= R16
+def AND32mr : Im32 <"and", 0x21, MRMDestMem>; // [mem32] &= R32
+def AND8rm : Im8 <"and", 0x22, MRMSrcMem >; // R8 &= [mem8]
+def AND16rm : Im16 <"and", 0x23, MRMSrcMem >, OpSize; // R16 &= [mem16]
+def AND32rm : Im32 <"and", 0x23, MRMSrcMem >; // R32 &= [mem32]
+
+def AND8ri : Ii8 <"and", 0x80, MRM4r >, Pattern<(set R8 , (and R8 , imm))>;
+def AND16ri : Ii16 <"and", 0x81, MRM4r >, OpSize, Pattern<(set R16, (and R16, imm))>;
+def AND32ri : Ii32 <"and", 0x81, MRM4r >, Pattern<(set R32, (and R32, imm))>;
+def AND8mi : Im8i8 <"and", 0x80, MRM4m >; // [mem8] &= imm8
+def AND16mi : Im16i16<"and", 0x81, MRM4m >, OpSize; // [mem16] &= imm16
+def AND32mi : Im32i32<"and", 0x81, MRM4m >; // [mem32] &= imm32
+
+def AND16ri8 : Ii8 <"and", 0x83, MRM4r >, OpSize; // R16 &= imm8
+def AND32ri8 : Ii8 <"and", 0x83, MRM4r >; // R32 &= imm8
+def AND16mi8 : Im16i8<"and", 0x83, MRM4m >, OpSize; // [mem16] &= imm8
+def AND32mi8 : Im32i8<"and", 0x83, MRM4m >; // [mem32] &= imm8
+
+
+def OR8rr : I <"or" , 0x08, MRMDestReg>, Pattern<(set R8 , (or R8 , R8 ))>;
+def OR16rr : I <"or" , 0x09, MRMDestReg>, OpSize, Pattern<(set R16, (or R16, R16))>;
+def OR32rr : I <"or" , 0x09, MRMDestReg>, Pattern<(set R32, (or R32, R32))>;
+def OR8mr : Im8 <"or" , 0x08, MRMDestMem>; // [mem8] |= R8
+def OR16mr : Im16 <"or" , 0x09, MRMDestMem>, OpSize; // [mem16] |= R16
+def OR32mr : Im32 <"or" , 0x09, MRMDestMem>; // [mem32] |= R32
+def OR8rm : Im8 <"or" , 0x0A, MRMSrcMem >; // R8 |= [mem8]
+def OR16rm : Im16 <"or" , 0x0B, MRMSrcMem >, OpSize; // R16 |= [mem16]
+def OR32rm : Im32 <"or" , 0x0B, MRMSrcMem >; // R32 |= [mem32]
+
+def OR8ri : Ii8 <"or" , 0x80, MRM1r >, Pattern<(set R8 , (or R8 , imm))>;
+def OR16ri : Ii16 <"or" , 0x81, MRM1r >, OpSize, Pattern<(set R16, (or R16, imm))>;
+def OR32ri : Ii32 <"or" , 0x81, MRM1r >, Pattern<(set R32, (or R32, imm))>;
+def OR8mi : Im8i8 <"or" , 0x80, MRM1m >; // [mem8] |= imm8
+def OR16mi : Im16i16<"or" , 0x81, MRM1m >, OpSize; // [mem16] |= imm16
+def OR32mi : Im32i32<"or" , 0x81, MRM1m >; // [mem32] |= imm32
+
+def OR16ri8 : Ii8 <"or" , 0x83, MRM1r >, OpSize; // R16 |= imm8
+def OR32ri8 : Ii8 <"or" , 0x83, MRM1r >; // R32 |= imm8
+def OR16mi8 : Im16i8<"or" , 0x83, MRM1m >, OpSize; // [mem16] |= imm8
+def OR32mi8 : Im32i8<"or" , 0x83, MRM1m >; // [mem32] |= imm8
+
+
+def XOR8rr : I <"xor", 0x30, MRMDestReg>, Pattern<(set R8 , (xor R8 , R8 ))>;
+def XOR16rr : I <"xor", 0x31, MRMDestReg>, OpSize, Pattern<(set R16, (xor R16, R16))>;
+def XOR32rr : I <"xor", 0x31, MRMDestReg>, Pattern<(set R32, (xor R32, R32))>;
+def XOR8mr : Im8 <"xor", 0x30, MRMDestMem>; // [mem8] ^= R8
+def XOR16mr : Im16 <"xor", 0x31, MRMDestMem>, OpSize; // [mem16] ^= R16
+def XOR32mr : Im32 <"xor", 0x31, MRMDestMem>; // [mem32] ^= R32
+def XOR8rm : Im8 <"xor", 0x32, MRMSrcMem >; // R8 ^= [mem8]
+def XOR16rm : Im16 <"xor", 0x33, MRMSrcMem >, OpSize; // R16 ^= [mem16]
+def XOR32rm : Im32 <"xor", 0x33, MRMSrcMem >; // R32 ^= [mem32]
+
+def XOR8ri : Ii8 <"xor", 0x80, MRM6r >, Pattern<(set R8 , (xor R8 , imm))>;
+def XOR16ri : Ii16 <"xor", 0x81, MRM6r >, OpSize, Pattern<(set R16, (xor R16, imm))>;
+def XOR32ri : Ii32 <"xor", 0x81, MRM6r >, Pattern<(set R32, (xor R32, imm))>;
+def XOR8mi : Im8i8 <"xor", 0x80, MRM6m >; // [mem8] ^= R8
+def XOR16mi : Im16i16<"xor", 0x81, MRM6m >, OpSize; // [mem16] ^= R16
+def XOR32mi : Im32i32<"xor", 0x81, MRM6m >; // [mem32] ^= R32
+
+def XOR16ri8 : Ii8 <"xor", 0x83, MRM6r >, OpSize; // R16 ^= imm8
+def XOR32ri8 : Ii8 <"xor", 0x83, MRM6r >; // R32 ^= imm8
+def XOR16mi8 : Im16i8<"xor", 0x83, MRM6m >, OpSize; // [mem16] ^= imm8
+def XOR32mi8 : Im32i8<"xor", 0x83, MRM6m >; // [mem32] ^= imm8
+// Shift instructions
+// FIXME: provide shorter instructions when imm8 == 1
+def SHL8rCL : I <"shl", 0xD2, MRM4r > , UsesCL; // R8 <<= cl
+def SHL16rCL : I <"shl", 0xD3, MRM4r >, OpSize, UsesCL; // R16 <<= cl
+def SHL32rCL : I <"shl", 0xD3, MRM4r > , UsesCL; // R32 <<= cl
+def SHL8mCL : Im8 <"shl", 0xD2, MRM4m > , UsesCL; // [mem8] <<= cl
+def SHL16mCL : Im16 <"shl", 0xD3, MRM4m >, OpSize, UsesCL; // [mem16] <<= cl
+def SHL32mCL : Im32 <"shl", 0xD3, MRM4m > , UsesCL; // [mem32] <<= cl
+
+def SHL8ri : Ii8 <"shl", 0xC0, MRM4r >; // R8 <<= imm8
+def SHL16ri : Ii8 <"shl", 0xC1, MRM4r >, OpSize; // R16 <<= imm8
+def SHL32ri : Ii8 <"shl", 0xC1, MRM4r >; // R32 <<= imm8
+def SHL8mi : Im8i8 <"shl", 0xC0, MRM4m >; // [mem8] <<= imm8
+def SHL16mi : Im16i8<"shl", 0xC1, MRM4m >, OpSize; // [mem16] <<= imm8
+def SHL32mi : Im32i8<"shl", 0xC1, MRM4m >; // [mem32] <<= imm8
+
+def SHR8rCL : I <"shr", 0xD2, MRM5r > , UsesCL; // R8 >>= cl
+def SHR16rCL : I <"shr", 0xD3, MRM5r >, OpSize, UsesCL; // R16 >>= cl
+def SHR32rCL : I <"shr", 0xD3, MRM5r > , UsesCL; // R32 >>= cl
+def SHR8mCL : Im8 <"shr", 0xD2, MRM5m > , UsesCL; // [mem8] >>= cl
+def SHR16mCL : Im16 <"shr", 0xD3, MRM5m >, OpSize, UsesCL; // [mem16] >>= cl
+def SHR32mCL : Im32 <"shr", 0xD3, MRM5m > , UsesCL; // [mem32] >>= cl
+
+def SHR8ri : Ii8 <"shr", 0xC0, MRM5r >; // R8 >>= imm8
+def SHR16ri : Ii8 <"shr", 0xC1, MRM5r >, OpSize; // R16 >>= imm8
+def SHR32ri : Ii8 <"shr", 0xC1, MRM5r >; // R32 >>= imm8
+def SHR8mi : Im8i8 <"shr", 0xC0, MRM5m >; // [mem8] >>= imm8
+def SHR16mi : Im16i8<"shr", 0xC1, MRM5m >, OpSize; // [mem16] >>= imm8
+def SHR32mi : Im32i8<"shr", 0xC1, MRM5m >; // [mem32] >>= imm8
+
+def SAR8rCL : I <"sar", 0xD2, MRM7r > , UsesCL; // R8 >>>= cl
+def SAR16rCL : I <"sar", 0xD3, MRM7r >, OpSize, UsesCL; // R16 >>>= cl
+def SAR32rCL : I <"sar", 0xD3, MRM7r > , UsesCL; // R32 >>>= cl
+def SAR8mCL : Im8 <"sar", 0xD2, MRM7m > , UsesCL; // [mem8] >>>= cl
+def SAR16mCL : Im16 <"sar", 0xD3, MRM7m >, OpSize, UsesCL; // [mem16] >>>= cl
+def SAR32mCL : Im32 <"sar", 0xD3, MRM7m > , UsesCL; // [mem32] >>>= cl
+
+def SAR8ri : Ii8 <"sar", 0xC0, MRM7r >; // R8 >>>= imm8
+def SAR16ri : Ii8 <"sar", 0xC1, MRM7r >, OpSize; // R16 >>>= imm8
+def SAR32ri : Ii8 <"sar", 0xC1, MRM7r >; // R32 >>>= imm8
+def SAR8mi : Im8i8 <"sar", 0xC0, MRM7m >; // [mem8] >>>= imm8
+def SAR16mi : Im16i8<"sar", 0xC1, MRM7m >, OpSize; // [mem16] >>>= imm8
+def SAR32mi : Im32i8<"sar", 0xC1, MRM7m >; // [mem32] >>>= imm8
+
+def SHLD32rrCL : I <"shld", 0xA5, MRMDestReg>, TB, UsesCL; // R32 <<= R32,R32 cl
+def SHLD32mrCL : Im32 <"shld", 0xA5, MRMDestMem>, TB, UsesCL; // [mem32] <<= [mem32],R32 cl
+def SHLD32rri8 : Ii8 <"shld", 0xA4, MRMDestReg>, TB; // R32 <<= R32,R32 imm8
+def SHLD32mri8 : Im32i8<"shld", 0xA4, MRMDestMem>, TB; // [mem32] <<= [mem32],R32 imm8
+
+def SHRD32rrCL : I <"shrd", 0xAD, MRMDestReg>, TB, UsesCL; // R32 >>= R32,R32 cl
+def SHRD32mrCL : Im32 <"shrd", 0xAD, MRMDestMem>, TB, UsesCL; // [mem32] >>= [mem32],R32 cl
+def SHRD32rri8 : Ii8 <"shrd", 0xAC, MRMDestReg>, TB; // R32 >>= R32,R32 imm8
+def SHRD32mri8 : Im32i8<"shrd", 0xAC, MRMDestMem>, TB; // [mem32] >>= [mem32],R32 imm8
// 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 , imm))>;
-def ADDri16 : I2A16<"add", 0x81, MRMS0r >, OpSize, Pattern<(set R16, (plus R16, imm))>;
-def ADDri32 : I2A32<"add", 0x81, MRMS0r >, Pattern<(set R32, (plus R32, imm))>;
-def ADDri16b : I2A8 <"add", 0x83, MRMS0r >, OpSize; // ADDri with sign extended 8 bit imm
-def ADDri32b : I2A8 <"add", 0x83, MRMS0r >;
-
-def ADDmr8 : I2A8 <"add", 0x00, MRMDestMem>; // [mem8] += R8
-def ADDmr16 : I2A16<"add", 0x01, MRMDestMem>, OpSize; // [mem16] += R16
-def ADDmr32 : I2A32<"add", 0x01, MRMDestMem>; // [mem32] += R32
-def ADDrm8 : I2A8 <"add", 0x02, MRMSrcMem >; // R8 += [mem8]
-def ADDrm16 : I2A16<"add", 0x03, MRMSrcMem >, OpSize; // R16 += [mem16]
-def ADDrm32 : I2A32<"add", 0x03, MRMSrcMem >; // R32 += [mem32]
-def ADDmi8 : I2A8 <"add", 0x80, MRMSrcMem >; // [mem8] += I8
-def ADDmi16 : I2A16<"add", 0x81, MRMSrcMem >, OpSize; // [mem16] += I16
-def ADDmi32 : I2A32<"add", 0x81, MRMSrcMem >; // [mem32] += I8
-def ADDmi16b : I2A8 <"add", 0x83, MRMSrcMem >, OpSize; // [mem16] += I8
-def ADDmi32b : I2A8 <"add", 0x83, MRMSrcMem >; // [mem32] += I32
-
-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 , imm))>;
-def SUBri16 : I2A16<"sub", 0x81, MRMS5r >, OpSize, Pattern<(set R16, (minus R16, imm))>;
-def SUBri32 : I2A32<"sub", 0x81, MRMS5r >, Pattern<(set R32, (minus R32, imm))>;
-def SUBri16b : I2A8 <"sub", 0x83, MRMS5r >, OpSize;
-def SUBri32b : I2A8 <"sub", 0x83, MRMS5r >;
-
-def SBBrr32 : I2A32<"sbb", 0x19, MRMDestReg>; // R32 -= R32+Carry
-
-def IMULrr16 : I2A16<"imul", 0xAF, MRMSrcReg>, TB, OpSize, Pattern<(set R16, (times R16, R16))>;
-def IMULrr32 : I2A32<"imul", 0xAF, MRMSrcReg>, TB , Pattern<(set R32, (times R32, R32))>;
-
-// These are suprisingly enough not two addres instructions!
-def IMULri16 : X86Inst<"imul", 0x69, MRMSrcReg, Arg16>, OpSize;
-def IMULri32 : X86Inst<"imul", 0x69, MRMSrcReg, Arg32>;
-def IMULri16b : X86Inst<"imul", 0x6B, MRMSrcReg, Arg8>, OpSize;
-def IMULri32b : X86Inst<"imul", 0x6B, MRMSrcReg, Arg8>;
+def ADD8rr : I <"add", 0x00, MRMDestReg>, Pattern<(set R8 , (plus R8 , R8 ))>;
+def ADD16rr : I <"add", 0x01, MRMDestReg>, OpSize, Pattern<(set R16, (plus R16, R16))>;
+def ADD32rr : I <"add", 0x01, MRMDestReg>, Pattern<(set R32, (plus R32, R32))>;
+def ADD8mr : Im8 <"add", 0x00, MRMDestMem>; // [mem8] += R8
+def ADD16mr : Im16 <"add", 0x01, MRMDestMem>, OpSize; // [mem16] += R16
+def ADD32mr : Im32 <"add", 0x01, MRMDestMem>; // [mem32] += R32
+def ADD8rm : Im8 <"add", 0x02, MRMSrcMem >; // R8 += [mem8]
+def ADD16rm : Im16 <"add", 0x03, MRMSrcMem >, OpSize; // R16 += [mem16]
+def ADD32rm : Im32 <"add", 0x03, MRMSrcMem >; // R32 += [mem32]
+
+def ADD8ri : Ii8 <"add", 0x80, MRM0r >, Pattern<(set R8 , (plus R8 , imm))>;
+def ADD16ri : Ii16 <"add", 0x81, MRM0r >, OpSize, Pattern<(set R16, (plus R16, imm))>;
+def ADD32ri : Ii32 <"add", 0x81, MRM0r >, Pattern<(set R32, (plus R32, imm))>;
+def ADD8mi : Im8i8 <"add", 0x80, MRM0m >; // [mem8] += I8
+def ADD16mi : Im16i16<"add", 0x81, MRM0m >, OpSize; // [mem16] += I16
+def ADD32mi : Im32i32<"add", 0x81, MRM0m >; // [mem32] += I32
+
+def ADD16ri8 : Ii8 <"add", 0x83, MRM0r >, OpSize; // ADDri with sign extended 8 bit imm
+def ADD32ri8 : Ii8 <"add", 0x83, MRM0r >;
+def ADD16mi8 : Im16i8<"add", 0x83, MRM0m >, OpSize; // [mem16] += I8
+def ADD32mi8 : Im32i8<"add", 0x83, MRM0m >; // [mem32] += I8
+
+def ADC32rr : I <"adc", 0x11, MRMDestReg>; // R32 += R32+Carry
+def ADC32rm : Im32 <"adc", 0x11, MRMSrcMem >; // R32 += [mem32]+Carry
+def ADC32mr : Im32 <"adc", 0x13, MRMDestMem>; // [mem32] += R32+Carry
+def ADC32ri : Ii32 <"adc", 0x81, MRM2r >; // R32 += I32+Carry
+def ADC32ri8 : Ii8 <"adc", 0x83, MRM2r >; // R32 += I8+Carry
+def ADC32mi : Im32i32<"adc", 0x81, MRM2m >; // [mem32] += I32+Carry
+def ADC32mi8 : Im32i8 <"adc", 0x83, MRM2m >; // [mem32] += I8+Carry
+
+def SUB8rr : I <"sub", 0x28, MRMDestReg>, Pattern<(set R8 , (minus R8 , R8 ))>;
+def SUB16rr : I <"sub", 0x29, MRMDestReg>, OpSize, Pattern<(set R16, (minus R16, R16))>;
+def SUB32rr : I <"sub", 0x29, MRMDestReg>, Pattern<(set R32, (minus R32, R32))>;
+def SUB8mr : Im8 <"sub", 0x28, MRMDestMem>; // [mem8] -= R8
+def SUB16mr : Im16 <"sub", 0x29, MRMDestMem>, OpSize; // [mem16] -= R16
+def SUB32mr : Im32 <"sub", 0x29, MRMDestMem>; // [mem32] -= R32
+def SUB8rm : Im8 <"sub", 0x2A, MRMSrcMem >; // R8 -= [mem8]
+def SUB16rm : Im16 <"sub", 0x2B, MRMSrcMem >, OpSize; // R16 -= [mem16]
+def SUB32rm : Im32 <"sub", 0x2B, MRMSrcMem >; // R32 -= [mem32]
+
+def SUB8ri : Ii8 <"sub", 0x80, MRM5r >, Pattern<(set R8 , (minus R8 , imm))>;
+def SUB16ri : Ii16 <"sub", 0x81, MRM5r >, OpSize, Pattern<(set R16, (minus R16, imm))>;
+def SUB32ri : Ii32 <"sub", 0x81, MRM5r >, Pattern<(set R32, (minus R32, imm))>;
+def SUB8mi : Im8i8 <"sub", 0x80, MRM5m >; // [mem8] -= I8
+def SUB16mi : Im16i16<"sub", 0x81, MRM5m >, OpSize; // [mem16] -= I16
+def SUB32mi : Im32i32<"sub", 0x81, MRM5m >; // [mem32] -= I32
+
+def SUB16ri8 : Ii8 <"sub", 0x83, MRM5r >, OpSize;
+def SUB32ri8 : Ii8 <"sub", 0x83, MRM5r >;
+def SUB16mi8 : Im16i8<"sub", 0x83, MRM5m >, OpSize; // [mem16] -= I8
+def SUB32mi8 : Im32i8<"sub", 0x83, MRM5m >; // [mem32] -= I8
+
+def SBB32rr : I <"sbb", 0x19, MRMDestReg>; // R32 -= R32+Borrow
+def SBB32rm : Im32 <"sbb", 0x19, MRMSrcMem >; // R32 -= [mem32]+Borrow
+def SBB32mr : Im32 <"sbb", 0x1B, MRMDestMem>; // [mem32] -= R32+Borrow
+def SBB32ri : Ii32 <"adc", 0x81, MRM3r >; // R32 -= I32+Borrow
+def SBB32ri8 : Ii8 <"adc", 0x83, MRM3r >; // R32 -= I8+Borrow
+def SBB32mi : Im32i32<"adc", 0x81, MRM3m >; // [mem32] -= I32+Borrow
+def SBB32mi8 : Im32i8 <"adc", 0x83, MRM3m >; // [mem32] -= I8+Borrow
+
+def IMUL16rr : I <"imul", 0xAF, MRMSrcReg>, TB, OpSize, Pattern<(set R16, (times R16, R16))>;
+def IMUL32rr : I <"imul", 0xAF, MRMSrcReg>, TB , Pattern<(set R32, (times R32, R32))>;
+def IMUL16rm : Im16 <"imul", 0xAF, MRMSrcMem>, TB, OpSize;
+def IMUL32rm : Im32 <"imul", 0xAF, MRMSrcMem>, TB ;
+
+} // end Two Address instructions
+
+// These are suprisingly enough not two address instructions!
+def IMUL16rri : Ii16 <"imul", 0x69, MRMSrcReg>, OpSize; // R16 = R16*I16
+def IMUL32rri : Ii32 <"imul", 0x69, MRMSrcReg>; // R32 = R32*I32
+def IMUL16rri8 : Ii8 <"imul", 0x6B, MRMSrcReg>, OpSize; // R16 = R16*I8
+def IMUL32rri8 : Ii8 <"imul", 0x6B, MRMSrcReg>; // R32 = R32*I8
+def IMUL16rmi : Im16i16<"imul",0x69, MRMSrcMem>, OpSize; // R16 = [mem16]*I16
+def IMUL32rmi : Im32i32<"imul",0x69, MRMSrcMem>; // R32 = [mem32]*I32
+def IMUL16rmi8 : Im16i8<"imul", 0x6B, MRMSrcMem>, OpSize; // R16 = [mem16]*I8
+def IMUL32rmi8 : Im32i8<"imul", 0x6B, MRMSrcMem>; // R32 = [mem32]*I8
+//===----------------------------------------------------------------------===//
+// Test instructions are just like AND, except they don't generate a result.
+def TEST8rr : I <"test", 0x84, MRMDestReg>; // flags = R8 & R8
+def TEST16rr : I <"test", 0x85, MRMDestReg>, OpSize; // flags = R16 & R16
+def TEST32rr : I <"test", 0x85, MRMDestReg>; // flags = R32 & R32
+def TEST8mr : Im8 <"test", 0x84, MRMDestMem>; // flags = [mem8] & R8
+def TEST16mr : Im16 <"test", 0x85, MRMDestMem>, OpSize; // flags = [mem16] & R16
+def TEST32mr : Im32 <"test", 0x85, MRMDestMem>; // flags = [mem32] & R32
+def TEST8rm : Im8 <"test", 0x84, MRMSrcMem >; // flags = R8 & [mem8]
+def TEST16rm : Im16 <"test", 0x85, MRMSrcMem >, OpSize; // flags = R16 & [mem16]
+def TEST32rm : Im32 <"test", 0x85, MRMSrcMem >; // flags = R32 & [mem32]
-// Logical operators...
-def ANDrr8 : I2A8 <"and", 0x20, MRMDestReg>, Pattern<(set R8 , (and R8 , R8 ))>;
-def ANDrr16 : I2A16<"and", 0x21, MRMDestReg>, OpSize, Pattern<(set R16, (and R16, R16))>;
-def ANDrr32 : I2A32<"and", 0x21, MRMDestReg>, Pattern<(set R32, (and R32, R32))>;
-def ANDri8 : I2A8 <"and", 0x80, MRMS4r >, Pattern<(set R8 , (and R8 , imm))>;
-def ANDri16 : I2A16<"and", 0x81, MRMS4r >, OpSize, Pattern<(set R16, (and R16, imm))>;
-def ANDri32 : I2A32<"and", 0x81, MRMS4r >, Pattern<(set R32, (and R32, imm))>;
-def ANDri16b : I2A8 <"and", 0x83, MRMS4r >, OpSize;
-def ANDri32b : I2A8 <"and", 0x83, MRMS4r >;
-
-def ORrr8 : I2A8 <"or" , 0x08, MRMDestReg>, Pattern<(set R8 , (or R8 , R8 ))>;
-def ORrr16 : I2A16<"or" , 0x09, MRMDestReg>, OpSize, Pattern<(set R16, (or R16, R16))>;
-def ORrr32 : I2A32<"or" , 0x09, MRMDestReg>, Pattern<(set R32, (or R32, R32))>;
-def ORri8 : I2A8 <"or" , 0x80, MRMS1r >, Pattern<(set R8 , (or R8 , imm))>;
-def ORri16 : I2A16<"or" , 0x81, MRMS1r >, OpSize, Pattern<(set R16, (or R16, imm))>;
-def ORri32 : I2A32<"or" , 0x81, MRMS1r >, Pattern<(set R32, (or R32, imm))>;
-def ORri16b : I2A8 <"or" , 0x83, MRMS1r >, OpSize;
-def ORri32b : I2A8 <"or" , 0x83, MRMS1r >;
-
-
-def XORrr8 : I2A8 <"xor", 0x30, MRMDestReg>, Pattern<(set R8 , (xor R8 , R8 ))>;
-def XORrr16 : I2A16<"xor", 0x31, MRMDestReg>, OpSize, Pattern<(set R16, (xor R16, R16))>;
-def XORrr32 : I2A32<"xor", 0x31, MRMDestReg>, Pattern<(set R32, (xor R32, R32))>;
-def XORri8 : I2A8 <"xor", 0x80, MRMS6r >, Pattern<(set R8 , (xor R8 , imm))>;
-def XORri16 : I2A16<"xor", 0x81, MRMS6r >, OpSize, Pattern<(set R16, (xor R16, imm))>;
-def XORri32 : I2A32<"xor", 0x81, MRMS6r >, Pattern<(set R32, (xor R32, imm))>;
-def XORri16b : I2A8 <"xor", 0x83, MRMS6r >, OpSize;
-def XORri32b : I2A8 <"xor", 0x83, MRMS6r >;
+def TEST8ri : Ii8 <"test", 0xF6, MRM0r >; // flags = R8 & imm8
+def TEST16ri : Ii16 <"test", 0xF7, MRM0r >, OpSize; // flags = R16 & imm16
+def TEST32ri : Ii32 <"test", 0xF7, MRM0r >; // flags = R32 & imm32
+def TEST8mi : Im8i8 <"test", 0xF6, MRM0m >; // flags = [mem8] & imm8
+def TEST16mi : Im16i16<"test", 0xF7, MRM0m >, OpSize; // flags = [mem16] & imm16
+def TEST32mi : Im32i32<"test", 0xF7, MRM0m >; // flags = [mem32] & 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 SETSr : X86Inst<"sets" , 0x98, MRMS0r, Arg8>, TB; // R8 = <sign bit>
-def SETNSr : X86Inst<"setns", 0x99, MRMS0r, Arg8>, TB; // R8 = !<sign bit>
-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
+def SAHF : I <"sahf" , 0x9E, RawFrm>, Imp<[AH],[]>; // flags = AH
+def LAHF : I <"lahf" , 0x9F, RawFrm>, Imp<[],[AH]>; // AH = flags
+
+def SETBr : I <"setb" , 0x92, MRM0r>, TB; // R8 = < unsign
+def SETBm : Im8<"setb" , 0x92, MRM0m>, TB; // [mem8] = < unsign
+def SETAEr : I <"setae", 0x93, MRM0r>, TB; // R8 = >= unsign
+def SETAEm : Im8<"setae", 0x93, MRM0m>, TB; // [mem8] = >= unsign
+def SETEr : I <"sete" , 0x94, MRM0r>, TB; // R8 = ==
+def SETEm : Im8<"sete" , 0x94, MRM0m>, TB; // [mem8] = ==
+def SETNEr : I <"setne", 0x95, MRM0r>, TB; // R8 = !=
+def SETNEm : Im8<"setne", 0x95, MRM0m>, TB; // [mem8] = !=
+def SETBEr : I <"setbe", 0x96, MRM0r>, TB; // R8 = <= unsign
+def SETBEm : Im8<"setbe", 0x96, MRM0m>, TB; // [mem8] = <= unsign
+def SETAr : I <"seta" , 0x97, MRM0r>, TB; // R8 = > signed
+def SETAm : Im8<"seta" , 0x97, MRM0m>, TB; // [mem8] = > signed
+def SETSr : I <"sets" , 0x98, MRM0r>, TB; // R8 = <sign bit>
+def SETSm : Im8<"sets" , 0x98, MRM0m>, TB; // [mem8] = <sign bit>
+def SETNSr : I <"setns", 0x99, MRM0r>, TB; // R8 = !<sign bit>
+def SETNSm : Im8<"setns", 0x99, MRM0m>, TB; // [mem8] = !<sign bit>
+def SETLr : I <"setl" , 0x9C, MRM0r>, TB; // R8 = < signed
+def SETLm : Im8<"setl" , 0x9C, MRM0m>, TB; // [mem8] = < signed
+def SETGEr : I <"setge", 0x9D, MRM0r>, TB; // R8 = >= signed
+def SETGEm : Im8<"setge", 0x9D, MRM0m>, TB; // [mem8] = >= signed
+def SETLEr : I <"setle", 0x9E, MRM0r>, TB; // R8 = <= signed
+def SETLEm : Im8<"setle", 0x9E, MRM0m>, TB; // [mem8] = <= signed
+def SETGr : I <"setg" , 0x9F, MRM0r>, TB; // R8 = < signed
+def SETGm : Im8<"setg" , 0x9F, MRM0m>, TB; // [mem8] = < signed
// 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 CMP8rr : I <"cmp", 0x38, MRMDestReg>; // compare R8, R8
+def CMP16rr : I <"cmp", 0x39, MRMDestReg>, OpSize; // compare R16, R16
+def CMP32rr : I <"cmp", 0x39, MRMDestReg>, // compare R32, R32
Pattern<(isVoid (unspec2 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
+def CMP8mr : Im8 <"cmp", 0x38, MRMDestMem>; // compare [mem8], R8
+def CMP16mr : Im16 <"cmp", 0x39, MRMDestMem>, OpSize; // compare [mem16], R16
+def CMP32mr : Im32 <"cmp", 0x39, MRMDestMem>; // compare [mem32], R32
+def CMP8rm : Im8 <"cmp", 0x3A, MRMSrcMem >; // compare R8, [mem8]
+def CMP16rm : Im16 <"cmp", 0x3B, MRMSrcMem >, OpSize; // compare R16, [mem16]
+def CMP32rm : Im32 <"cmp", 0x3B, MRMSrcMem >; // compare R32, [mem32]
+def CMP8ri : Ii8 <"cmp", 0x80, MRM7r >; // compare R8, imm8
+def CMP16ri : Ii16 <"cmp", 0x81, MRM7r >, OpSize; // compare R16, imm16
+def CMP32ri : Ii32 <"cmp", 0x81, MRM7r >; // compare R32, imm32
+def CMP8mi : Im8i8 <"cmp", 0x80, MRM7m >; // compare [mem8], imm8
+def CMP16mi : Im16i16<"cmp", 0x81, MRM7m >, OpSize; // compare [mem16], imm16
+def CMP32mi : Im32i32<"cmp", 0x81, MRM7m >; // compare [mem32], 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)
+def MOVSX16rr8 : I <"movsx", 0xBE, MRMSrcReg>, TB, OpSize; // R16 = signext(R8)
+def MOVSX32rr8 : I <"movsx", 0xBE, MRMSrcReg>, TB; // R32 = signext(R8)
+def MOVSX32rr16: I <"movsx", 0xBF, MRMSrcReg>, TB; // R32 = signext(R16)
+def MOVSX16rm8 : Im8 <"movsx", 0xBE, MRMSrcMem>, TB, OpSize; // R16 = signext([mem8])
+def MOVSX32rm8 : Im8 <"movsx", 0xBE, MRMSrcMem>, TB; // R32 = signext([mem8])
+def MOVSX32rm16: Im16<"movsx", 0xBF, MRMSrcMem>, TB; // R32 = signext([mem16])
+
+def MOVZX16rr8 : I <"movzx", 0xB6, MRMSrcReg>, TB, OpSize; // R16 = zeroext(R8)
+def MOVZX32rr8 : I <"movzx", 0xB6, MRMSrcReg>, TB; // R32 = zeroext(R8)
+def MOVZX32rr16: I <"movzx", 0xB7, MRMSrcReg>, TB; // R32 = zeroext(R16)
+def MOVZX16rm8 : Im8 <"movzx", 0xB6, MRMSrcMem>, TB, OpSize; // R16 = zeroext([mem8])
+def MOVZX32rm8 : Im8 <"movzx", 0xB6, MRMSrcMem>, TB; // R32 = zeroext([mem8])
+def MOVZX32rm16: Im16<"movzx", 0xB7, MRMSrcMem>, TB; // R32 = zeroext([mem16])
//===----------------------------------------------------------------------===//
// 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; }
+// Floating point instruction templates
+class FPInst<string n, bits<8> o, Format F, FPFormat fp, MemType m, ImmType i>
+ : X86Inst<n, o, F, m, i> { let FPForm = fp; let FPFormBits = FPForm.Value; }
+
+class FPI<string n, bits<8> o, Format F, FPFormat fp> : FPInst<n, o, F, fp, NoMem, NoImm>;
+
+class FPIM<string n, bits<8> o, Format F, FPFormat fp, MemType m> : FPInst<n, o, F, fp, m, NoImm>;
+
+class FPI16m<string n, bits<8> o, Format F, FPFormat fp> : FPIM<n, o, F, fp, Mem16>;
+class FPI32m<string n, bits<8> o, Format F, FPFormat fp> : FPIM<n, o, F, fp, Mem32>;
+class FPI64m<string n, bits<8> o, Format F, FPFormat fp> : FPIM<n, o, F, fp, Mem64>;
+class FPI80m<string n, bits<8> o, Format F, FPFormat fp> : FPIM<n, o, F, fp, Mem80>;
// Pseudo instructions for floating point. We use these pseudo instructions
// because they can be expanded by the fp spackifier into one of many different
// forms of instructions for doing these operations. Until the stackifier runs,
// we prefer to be abstract.
-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
+def FpMOV : FPI<"FMOV", 0, Pseudo, SpecialFP>; // f1 = fmov f2
+def FpADD : FPI<"FADD", 0, Pseudo, TwoArgFP>; // f1 = fadd f2, f3
+def FpSUB : FPI<"FSUB", 0, Pseudo, TwoArgFP>; // f1 = fsub f2, f3
+def FpMUL : FPI<"FMUL", 0, Pseudo, TwoArgFP>; // f1 = fmul f2, f3
+def FpDIV : FPI<"FDIV", 0, Pseudo, TwoArgFP>; // f1 = fdiv f2, f3
+
+def FpUCOM : FPI<"FUCOM", 0, Pseudo, TwoArgFP>; // FPSW = fucom f1, f2
+def FpGETRESULT : FPI<"FGETRESULT",0, Pseudo, SpecialFP>; // FPR = ST(0)
+def FpSETRESULT : FPI<"FSETRESULT",0, Pseudo, SpecialFP>; // ST(0) = FPR
+
+
+// Floating point cmovs...
+let isTwoAddress = 1, Uses = [ST0], Defs = [ST0], printImplicitUsesBefore = 1 in {
+ def FCMOVB : FPI <"fcmovb" , 0xC0, AddRegFrm, CondMovFP>, DA; // fcmovb ST(i) -> ST(0)
+ def FCMOVBE : FPI <"fcmovbe", 0xD0, AddRegFrm, CondMovFP>, DA; // fcmovbe ST(i) -> ST(0)
+ def FCMOVE : FPI <"fcmove" , 0xC8, AddRegFrm, CondMovFP>, DA; // fcmove ST(i) -> ST(0)
+ def FCMOVAE : FPI <"fcmovae", 0xC0, AddRegFrm, CondMovFP>, DB; // fcmovae ST(i) -> ST(0)
+ def FCMOVA : FPI <"fcmova" , 0xD0, AddRegFrm, CondMovFP>, DB; // fcmova ST(i) -> ST(0)
+ def FCMOVNE : FPI <"fcmovne", 0xC8, AddRegFrm, CondMovFP>, DB; // fcmovne ST(i) -> ST(0)
+}
// 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)
+def FLDrr : FPI <"fld" , 0xC0, AddRegFrm, NotFP>, D9; // push(ST(i))
+def FLD32m : FPI32m <"fld" , 0xD9, MRM0m , ZeroArgFP>; // load float
+def FLD64m : FPI64m <"fld" , 0xDD, MRM0m , ZeroArgFP>; // load double
+def FLD80m : FPI80m <"fld" , 0xDB, MRM5m , ZeroArgFP>; // load extended
+def FILD16m : FPI16m <"fild" , 0xDF, MRM0m , ZeroArgFP>; // load signed short
+def FILD32m : FPI32m <"fild" , 0xDB, MRM0m , ZeroArgFP>; // load signed int
+def FILD64m : FPI64m <"fild" , 0xDF, MRM5m , ZeroArgFP>; // load signed long
+
+def FSTrr : FPI <"fst" , 0xD0, AddRegFrm, NotFP >, DD; // ST(i) = ST(0)
+def FSTPrr : FPI <"fstp", 0xD8, AddRegFrm, NotFP >, DD; // ST(i) = ST(0), pop
+def FST32m : FPI32m <"fst" , 0xD9, MRM2m , OneArgFP>; // store float
+def FST64m : FPI64m <"fst" , 0xDD, MRM2m , OneArgFP>; // store double
+def FSTP32m : FPI32m <"fstp", 0xD9, MRM3m , OneArgFP>; // store float, pop
+def FSTP64m : FPI64m <"fstp", 0xDD, MRM3m , OneArgFP>; // store double, pop
+def FSTP80m : FPI80m <"fstp", 0xDB, MRM7m , OneArgFP>; // store extended, pop
+
+def FIST16m : FPI16m <"fist", 0xDF, MRM2m , OneArgFP>; // store signed short
+def FIST32m : FPI32m <"fist", 0xDB, MRM2m , OneArgFP>; // store signed int
+def FISTP16m : FPI16m <"fistp", 0xDF, MRM3m , NotFP >; // store signed short, pop
+def FISTP32m : FPI32m <"fistp", 0xDB, MRM3m , NotFP >; // store signed int, pop
+def FISTP64m : FPI64m <"fistpll", 0xDF, MRM7m , OneArgFP>; // store signed long, pop
+
+def FXCH : FPI <"fxch", 0xC8, AddRegFrm, 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;
+def FLD0 : FPI<"fldz", 0xEE, RawFrm, ZeroArgFP>, D9;
+def FLD1 : FPI<"fld1", 0xE8, RawFrm, ZeroArgFP>, D9;
// Unary operations...
-def FCHS : FPInst<"fchs", 0xE0, RawFrm, ArgF80, OneArgFPRW>, D9; // f1 = fchs f2
+def FCHS : FPI<"fchs", 0xE0, RawFrm, OneArgFPRW>, D9; // f1 = fchs f2
-def FTST : FPInst<"ftst", 0xE4, RawFrm, ArgF80, OneArgFP>, D9; // ftst ST(0)
+def FTST : FPI<"ftst", 0xE4, RawFrm, OneArgFP>, D9; // ftst ST(0)
// Binary arithmetic operations...
-class FPST0rInst<string n, bits<8> o>
- : X86Inst<n, o, AddRegFrm, ArgF80>, D8 {
+class FPST0rInst<string n, bits<8> o> : I<n, o, AddRegFrm>, 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;
+class FPrST0Inst<string n, bits<8> o> : I<n, o, AddRegFrm>, DC {
+ bit printImplicitUsesAfter = 1;
list<Register> Uses = [ST0];
}
-class FPrST0PInst<string n, bits<8> o>
- : X86Inst<n, o, AddRegFrm, ArgF80>, DE {
+class FPrST0PInst<string n, bits<8> o> : I<n, o, AddRegFrm>, DE {
list<Register> Uses = [ST0];
}
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
+def FUCOMr : I<"fucom" , 0xE0, AddRegFrm>, DD, Imp<[ST0],[]>; // FPSW = compare ST(0) with ST(i)
+def FUCOMPr : I<"fucomp" , 0xE8, AddRegFrm>, DD, Imp<[ST0],[]>; // FPSW = compare ST(0) with ST(i), pop
+def FUCOMPPr : I<"fucompp", 0xE9, RawFrm >, 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]
+def FNSTSW8r : I <"fnstsw" , 0xE0, RawFrm>, DF, Imp<[],[AX]>; // AX = fp flags
+def FNSTCW16m : Im16<"fnstcw" , 0xD9, MRM7m >; // [mem16] = X87 control world
+def FLDCW16m : Im16<"fldcw" , 0xD9, MRM5m >; // X87 control world = [mem16]
//===----------------------------------------------------------------------===//
//
def RET_R32 : Expander<(ret R32:$reg),
- [(MOVrr32 EAX, R32:$reg),
+ [(MOV32rr EAX, R32:$reg),
(RET)]>;
// FIXME: This should eventually just be implemented by defining a frameidx as a
// value address for a load.
def LOAD_FI16 : Expander<(set R16:$dest, (load frameidx:$fi)),
- [(MOVmr16 R16:$dest, frameidx:$fi, 1, 0/*NoReg*/, 0)]>;
+ [(MOV16rm R16:$dest, frameidx:$fi, 1, 0/*NoReg*/, 0)]>;
def LOAD_FI32 : Expander<(set R32:$dest, (load frameidx:$fi)),
- [(MOVmr32 R32:$dest, frameidx:$fi, 1, 0/*NoReg*/, 0)]>;
+ [(MOV32rm R32:$dest, frameidx:$fi, 1, 0/*NoReg*/, 0)]>;
def LOAD_R16 : Expander<(set R16:$dest, (load R32:$src)),
- [(MOVmr16 R16:$dest, R32:$src, 1, 0/*NoReg*/, 0)]>;
+ [(MOV16rm R16:$dest, R32:$src, 1, 0/*NoReg*/, 0)]>;
def LOAD_R32 : Expander<(set R32:$dest, (load R32:$src)),
- [(MOVmr32 R32:$dest, R32:$src, 1, 0/*NoReg*/, 0)]>;
+ [(MOV32rm R32:$dest, R32:$src, 1, 0/*NoReg*/, 0)]>;
def BR_EQ : Expander<(brcond (seteq R32:$a1, R32:$a2),
basicblock:$d1, basicblock:$d2),
- [(CMPrr32 R32:$a1, R32:$a2),
+ [(CMP32rr R32:$a1, R32:$a2),
(JE basicblock:$d1),
(JMP basicblock:$d2)]>;