lib/Target/X86/X86InstrMMX.td

//====- X86InstrMMX.td - Describe the MMX Instruction Set --*- tablegen -*-===//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file describes the X86 MMX instruction set, defining the instructions,
// and properties of the instructions which are needed for code generation,
// machine code emission, and analysis.
//
// All instructions that use MMX should be in this file, even if they also use
// SSE.
//
//===----------------------------------------------------------------------===//

//===----------------------------------------------------------------------===//
// MMX Multiclasses
//===----------------------------------------------------------------------===//

let Constraints = "$src1 = $dst" in {
  // MMXI_binop_rm_int - Simple MMX binary operator based on intrinsic.
  // When this is cleaned up, remove the FIXME from X86RecognizableInstr.cpp.
  multiclass MMXI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
                               bit Commutable = 0> {
    def irr : MMXI<opc, MRMSrcReg, (outs VR64:$dst),
                 (ins VR64:$src1, VR64:$src2),
                 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
                 [(set VR64:$dst, (IntId VR64:$src1, VR64:$src2))]> {
      let isCommutable = Commutable;
    }
    def irm : MMXI<opc, MRMSrcMem, (outs VR64:$dst),
                 (ins VR64:$src1, i64mem:$src2),
                 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
                 [(set VR64:$dst, (IntId VR64:$src1,
                                   (bitconvert (load_mmx addr:$src2))))]>;
  }

  multiclass MMXI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
                                string OpcodeStr, Intrinsic IntId,
                                Intrinsic IntId2> {
    def rr : MMXI<opc, MRMSrcReg, (outs VR64:$dst),
                                  (ins VR64:$src1, VR64:$src2),
                  !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
                  [(set VR64:$dst, (IntId VR64:$src1, VR64:$src2))]>;
    def rm : MMXI<opc, MRMSrcMem, (outs VR64:$dst),
                                  (ins VR64:$src1, i64mem:$src2),
                  !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
                  [(set VR64:$dst, (IntId VR64:$src1,
                                    (bitconvert (load_mmx addr:$src2))))]>;
    def ri : MMXIi8<opc2, ImmForm, (outs VR64:$dst),
                                   (ins VR64:$src1, i32i8imm:$src2),
                    !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
           [(set VR64:$dst, (IntId2 VR64:$src1, (i32 imm:$src2)))]>;
  }
}

/// Unary MMX instructions requiring SSSE3.
multiclass SS3I_unop_rm_int_mm<bits<8> opc, string OpcodeStr,
                               Intrinsic IntId64> {
  def rr64 : SS38I<opc, MRMSrcReg, (outs VR64:$dst), (ins VR64:$src),
                   !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
                   [(set VR64:$dst, (IntId64 VR64:$src))]>;

  def rm64 : SS38I<opc, MRMSrcMem, (outs VR64:$dst), (ins i64mem:$src),
                   !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
                   [(set VR64:$dst,
                     (IntId64 (bitconvert (memopmmx addr:$src))))]>;
}

/// Binary MMX instructions requiring SSSE3.
let ImmT = NoImm, Constraints = "$src1 = $dst" in {
multiclass SS3I_binop_rm_int_mm<bits<8> opc, string OpcodeStr,
                             Intrinsic IntId64> {
  let isCommutable = 0 in
  def rr64 : SS38I<opc, MRMSrcReg, (outs VR64:$dst),
       (ins VR64:$src1, VR64:$src2),
        !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
       [(set VR64:$dst, (IntId64 VR64:$src1, VR64:$src2))]>;
  def rm64 : SS38I<opc, MRMSrcMem, (outs VR64:$dst),
       (ins VR64:$src1, i64mem:$src2),
        !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
       [(set VR64:$dst,
         (IntId64 VR64:$src1,
          (bitconvert (memopmmx addr:$src2))))]>;
}
}

/// PALIGN MMX instructions (require SSSE3).
multiclass ssse3_palign_mm<string asm, Intrinsic IntId> {
  def R64irr  : SS3AI<0x0F, MRMSrcReg, (outs VR64:$dst),
      (ins VR64:$src1, VR64:$src2, i8imm:$src3),
      !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"), 
      [(set VR64:$dst, (IntId VR64:$src1, VR64:$src2, (i8 imm:$src3)))]>;
  def R64irm  : SS3AI<0x0F, MRMSrcMem, (outs VR64:$dst),
      (ins VR64:$src1, i64mem:$src2, i8imm:$src3),
      !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
      [(set VR64:$dst, (IntId VR64:$src1,
                       (bitconvert (load_mmx addr:$src2)), (i8 imm:$src3)))]>;
}

multiclass sse12_cvt_pint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
                         Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
                         string asm, Domain d> {
  def irr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
                        [(set DstRC:$dst, (Int SrcRC:$src))], d>;
  def irm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
                        [(set DstRC:$dst, (Int (ld_frag addr:$src)))], d>;
}

multiclass sse12_cvt_pint_3addr<bits<8> opc, RegisterClass SrcRC,
                    RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
                    PatFrag ld_frag, string asm, Domain d> {
  def irr : PI<opc, MRMSrcReg, (outs DstRC:$dst),(ins DstRC:$src1, SrcRC:$src2),
              asm, [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))], d>;
  def irm : PI<opc, MRMSrcMem, (outs DstRC:$dst),
                   (ins DstRC:$src1, x86memop:$src2), asm,
              [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))], d>;
}

//===----------------------------------------------------------------------===//
// MMX EMMS Instruction
//===----------------------------------------------------------------------===//

def MMX_EMMS  : MMXI<0x77, RawFrm, (outs), (ins), "emms",
                     [(int_x86_mmx_emms)]>;

//===----------------------------------------------------------------------===//
// MMX Scalar Instructions
//===----------------------------------------------------------------------===//

// Data Transfer Instructions
def MMX_MOVD64rr : MMXI<0x6E, MRMSrcReg, (outs VR64:$dst), (ins GR32:$src),
                        "movd\t{$src, $dst|$dst, $src}",
                        [(set VR64:$dst, 
                         (x86mmx (scalar_to_vector GR32:$src)))]>;
let canFoldAsLoad = 1 in
def MMX_MOVD64rm : MMXI<0x6E, MRMSrcMem, (outs VR64:$dst), (ins i32mem:$src),
                        "movd\t{$src, $dst|$dst, $src}",
              [(set VR64:$dst,
               (x86mmx (scalar_to_vector (loadi32 addr:$src))))]>;
let mayStore = 1 in
def MMX_MOVD64mr : MMXI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR64:$src),
                        "movd\t{$src, $dst|$dst, $src}", []>;
def MMX_MOVD64grr : MMXI<0x7E, MRMDestReg, (outs), (ins GR32:$dst, VR64:$src),
                        "movd\t{$src, $dst|$dst, $src}", []>;

let neverHasSideEffects = 1 in
def MMX_MOVD64to64rr : MMXRI<0x6E, MRMSrcReg, (outs VR64:$dst), (ins GR64:$src),
                             "movd\t{$src, $dst|$dst, $src}",
                             []>;

// These are 64 bit moves, but since the OS X assembler doesn't
// recognize a register-register movq, we write them as
// movd.
def MMX_MOVD64from64rr : MMXRI<0x7E, MRMDestReg,
                               (outs GR64:$dst), (ins VR64:$src),
                               "movd\t{$src, $dst|$dst, $src}", 
                             [(set GR64:$dst,
                              (bitconvert VR64:$src))]>;
def MMX_MOVD64rrv164 : MMXRI<0x6E, MRMSrcReg, (outs VR64:$dst), (ins GR64:$src),
                             "movd\t{$src, $dst|$dst, $src}",
                             [(set VR64:$dst,
                              (bitconvert GR64:$src))]>;
let neverHasSideEffects = 1 in
def MMX_MOVQ64rr : MMXI<0x6F, MRMSrcReg, (outs VR64:$dst), (ins VR64:$src),
                        "movq\t{$src, $dst|$dst, $src}", []>;
let canFoldAsLoad = 1 in
def MMX_MOVQ64rm : MMXI<0x6F, MRMSrcMem, (outs VR64:$dst), (ins i64mem:$src),
                        "movq\t{$src, $dst|$dst, $src}",
                        [(set VR64:$dst, (load_mmx addr:$src))]>;
def MMX_MOVQ64mr : MMXI<0x7F, MRMDestMem, (outs), (ins i64mem:$dst, VR64:$src),
                        "movq\t{$src, $dst|$dst, $src}",
                        [(store (x86mmx VR64:$src), addr:$dst)]>;

def MMX_MOVDQ2Qrr : SDIi8<0xD6, MRMSrcReg, (outs VR64:$dst), (ins VR128:$src),
                          "movdq2q\t{$src, $dst|$dst, $src}",
                          [(set VR64:$dst,
                            (x86mmx (bitconvert
                            (i64 (vector_extract (v2i64 VR128:$src),
                                  (iPTR 0))))))]>;

def MMX_MOVQ2DQrr : SSDIi8<0xD6, MRMSrcReg, (outs VR128:$dst), (ins VR64:$src),
                           "movq2dq\t{$src, $dst|$dst, $src}",
          [(set VR128:$dst,
            (v2i64 (scalar_to_vector
                              (i64 (bitconvert (x86mmx VR64:$src))))))]>;

let neverHasSideEffects = 1 in
def MMX_MOVQ2FR64rr: SSDIi8<0xD6, MRMSrcReg, (outs FR64:$dst), (ins VR64:$src),
                           "movq2dq\t{$src, $dst|$dst, $src}", []>;

def MMX_MOVFR642Qrr: SDIi8<0xD6, MRMSrcReg, (outs VR64:$dst), (ins FR64:$src),
                           "movdq2q\t{$src, $dst|$dst, $src}", []>;

def MMX_MOVNTQmr  : MMXI<0xE7, MRMDestMem, (outs), (ins i64mem:$dst, VR64:$src),
                         "movntq\t{$src, $dst|$dst, $src}",
                         [(int_x86_mmx_movnt_dq addr:$dst, VR64:$src)]>;

let AddedComplexity = 15 in
// movd to MMX register zero-extends
def MMX_MOVZDI2PDIrr : MMXI<0x6E, MRMSrcReg, (outs VR64:$dst), (ins GR32:$src),
                             "movd\t{$src, $dst|$dst, $src}",
              [(set VR64:$dst,
                    (x86mmx (X86vzmovl (x86mmx (scalar_to_vector GR32:$src)))))]>;
let AddedComplexity = 20 in
def MMX_MOVZDI2PDIrm : MMXI<0x6E, MRMSrcMem, (outs VR64:$dst),
                           (ins i32mem:$src),
                             "movd\t{$src, $dst|$dst, $src}",
          [(set VR64:$dst,
                (x86mmx (X86vzmovl (x86mmx
                                   (scalar_to_vector (loadi32 addr:$src))))))]>;

// Arithmetic Instructions
defm MMX_PABSB : SS3I_unop_rm_int_mm<0x1C, "pabsb", int_x86_ssse3_pabs_b>;
defm MMX_PABSW : SS3I_unop_rm_int_mm<0x1D, "pabsw", int_x86_ssse3_pabs_w>;
defm MMX_PABSD : SS3I_unop_rm_int_mm<0x1E, "pabsd", int_x86_ssse3_pabs_d>;
// -- Addition
defm MMX_PADDB : MMXI_binop_rm_int<0xFC, "paddb", int_x86_mmx_padd_b, 1>;
defm MMX_PADDW : MMXI_binop_rm_int<0xFD, "paddw", int_x86_mmx_padd_w, 1>;
defm MMX_PADDD : MMXI_binop_rm_int<0xFE, "paddd", int_x86_mmx_padd_d, 1>;
defm MMX_PADDQ : MMXI_binop_rm_int<0xD4, "paddq", int_x86_mmx_padd_q, 1>;
defm MMX_PADDSB  : MMXI_binop_rm_int<0xEC, "paddsb" , int_x86_mmx_padds_b, 1>;
defm MMX_PADDSW  : MMXI_binop_rm_int<0xED, "paddsw" , int_x86_mmx_padds_w, 1>;

defm MMX_PADDUSB : MMXI_binop_rm_int<0xDC, "paddusb", int_x86_mmx_paddus_b, 1>;
defm MMX_PADDUSW : MMXI_binop_rm_int<0xDD, "paddusw", int_x86_mmx_paddus_w, 1>;

defm MMX_PHADDW  : SS3I_binop_rm_int_mm<0x01, "phaddw", int_x86_ssse3_phadd_w>;
defm MMX_PHADD   : SS3I_binop_rm_int_mm<0x02, "phaddd", int_x86_ssse3_phadd_d>;
defm MMX_PHADDSW : SS3I_binop_rm_int_mm<0x03, "phaddsw",int_x86_ssse3_phadd_sw>;


// -- Subtraction
defm MMX_PSUBB : MMXI_binop_rm_int<0xF8, "psubb", int_x86_mmx_psub_b>;
defm MMX_PSUBW : MMXI_binop_rm_int<0xF9, "psubw", int_x86_mmx_psub_w>;
defm MMX_PSUBD : MMXI_binop_rm_int<0xFA, "psubd", int_x86_mmx_psub_d>;
defm MMX_PSUBQ : MMXI_binop_rm_int<0xFB, "psubq", int_x86_mmx_psub_q>;

defm MMX_PSUBSB  : MMXI_binop_rm_int<0xE8, "psubsb" , int_x86_mmx_psubs_b>;
defm MMX_PSUBSW  : MMXI_binop_rm_int<0xE9, "psubsw" , int_x86_mmx_psubs_w>;

defm MMX_PSUBUSB : MMXI_binop_rm_int<0xD8, "psubusb", int_x86_mmx_psubus_b>;
defm MMX_PSUBUSW : MMXI_binop_rm_int<0xD9, "psubusw", int_x86_mmx_psubus_w>;

defm MMX_PHSUBW  : SS3I_binop_rm_int_mm<0x05, "phsubw", int_x86_ssse3_phsub_w>;
defm MMX_PHSUBD  : SS3I_binop_rm_int_mm<0x06, "phsubd", int_x86_ssse3_phsub_d>;
defm MMX_PHSUBSW : SS3I_binop_rm_int_mm<0x07, "phsubsw",int_x86_ssse3_phsub_sw>;

// -- Multiplication
defm MMX_PMULLW  : MMXI_binop_rm_int<0xD5, "pmullw", int_x86_mmx_pmull_w, 1>;

defm MMX_PMULHW  : MMXI_binop_rm_int<0xE5, "pmulhw",  int_x86_mmx_pmulh_w,  1>;
defm MMX_PMULHUW : MMXI_binop_rm_int<0xE4, "pmulhuw", int_x86_mmx_pmulhu_w, 1>;
defm MMX_PMULUDQ : MMXI_binop_rm_int<0xF4, "pmuludq", int_x86_mmx_pmulu_dq, 1>;
let isCommutable = 1 in
defm MMX_PMULHRSW : SS3I_binop_rm_int_mm<0x0B, "pmulhrsw",
                                     int_x86_ssse3_pmul_hr_sw>;

// -- Miscellanea
defm MMX_PMADDWD : MMXI_binop_rm_int<0xF5, "pmaddwd", int_x86_mmx_pmadd_wd, 1>;

defm MMX_PMADDUBSW : SS3I_binop_rm_int_mm<0x04, "pmaddubsw",
                                     int_x86_ssse3_pmadd_ub_sw>;
defm MMX_PAVGB   : MMXI_binop_rm_int<0xE0, "pavgb", int_x86_mmx_pavg_b, 1>;
defm MMX_PAVGW   : MMXI_binop_rm_int<0xE3, "pavgw", int_x86_mmx_pavg_w, 1>;

defm MMX_PMINUB  : MMXI_binop_rm_int<0xDA, "pminub", int_x86_mmx_pminu_b, 1>;
defm MMX_PMINSW  : MMXI_binop_rm_int<0xEA, "pminsw", int_x86_mmx_pmins_w, 1>;

defm MMX_PMAXUB  : MMXI_binop_rm_int<0xDE, "pmaxub", int_x86_mmx_pmaxu_b, 1>;
defm MMX_PMAXSW  : MMXI_binop_rm_int<0xEE, "pmaxsw", int_x86_mmx_pmaxs_w, 1>;

defm MMX_PSADBW  : MMXI_binop_rm_int<0xF6, "psadbw", int_x86_mmx_psad_bw, 1>;

defm MMX_PSIGNB :  SS3I_binop_rm_int_mm<0x08, "psignb", int_x86_ssse3_psign_b>;
defm MMX_PSIGNW :  SS3I_binop_rm_int_mm<0x09, "psignw", int_x86_ssse3_psign_w>;
defm MMX_PSIGND :  SS3I_binop_rm_int_mm<0x0A, "psignd", int_x86_ssse3_psign_d>;
let Constraints = "$src1 = $dst" in
  defm MMX_PALIGN : ssse3_palign_mm<"palignr", int_x86_mmx_palignr_b>;

// Logical Instructions
defm MMX_PAND : MMXI_binop_rm_int<0xDB, "pand", int_x86_mmx_pand, 1>;
defm MMX_POR  : MMXI_binop_rm_int<0xEB, "por" , int_x86_mmx_por,  1>;
defm MMX_PXOR : MMXI_binop_rm_int<0xEF, "pxor", int_x86_mmx_pxor, 1>;
defm MMX_PANDN : MMXI_binop_rm_int<0xDF, "pandn", int_x86_mmx_pandn>;

// Shift Instructions
defm MMX_PSRLW : MMXI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
                                    int_x86_mmx_psrl_w, int_x86_mmx_psrli_w>;
defm MMX_PSRLD : MMXI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
                                    int_x86_mmx_psrl_d, int_x86_mmx_psrli_d>;
defm MMX_PSRLQ : MMXI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
                                    int_x86_mmx_psrl_q, int_x86_mmx_psrli_q>;

defm MMX_PSLLW : MMXI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
                                    int_x86_mmx_psll_w, int_x86_mmx_pslli_w>;
defm MMX_PSLLD : MMXI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
                                    int_x86_mmx_psll_d, int_x86_mmx_pslli_d>;
defm MMX_PSLLQ : MMXI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
                                    int_x86_mmx_psll_q, int_x86_mmx_pslli_q>;

defm MMX_PSRAW : MMXI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
                                    int_x86_mmx_psra_w, int_x86_mmx_psrai_w>;
defm MMX_PSRAD : MMXI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
                                    int_x86_mmx_psra_d, int_x86_mmx_psrai_d>;

// Comparison Instructions
defm MMX_PCMPEQB : MMXI_binop_rm_int<0x74, "pcmpeqb", int_x86_mmx_pcmpeq_b>;
defm MMX_PCMPEQW : MMXI_binop_rm_int<0x75, "pcmpeqw", int_x86_mmx_pcmpeq_w>;
defm MMX_PCMPEQD : MMXI_binop_rm_int<0x76, "pcmpeqd", int_x86_mmx_pcmpeq_d>;

defm MMX_PCMPGTB : MMXI_binop_rm_int<0x64, "pcmpgtb", int_x86_mmx_pcmpgt_b>;
defm MMX_PCMPGTW : MMXI_binop_rm_int<0x65, "pcmpgtw", int_x86_mmx_pcmpgt_w>;
defm MMX_PCMPGTD : MMXI_binop_rm_int<0x66, "pcmpgtd", int_x86_mmx_pcmpgt_d>;

// -- Unpack Instructions
defm MMX_PUNPCKHBW : MMXI_binop_rm_int<0x68, "punpckhbw", 
                                       int_x86_mmx_punpckhbw>;
defm MMX_PUNPCKHWD : MMXI_binop_rm_int<0x69, "punpckhwd", 
                                       int_x86_mmx_punpckhwd>;
defm MMX_PUNPCKHDQ : MMXI_binop_rm_int<0x6A, "punpckhdq", 
                                       int_x86_mmx_punpckhdq>;
defm MMX_PUNPCKLBW : MMXI_binop_rm_int<0x60, "punpcklbw", 
                                       int_x86_mmx_punpcklbw>;
defm MMX_PUNPCKLWD : MMXI_binop_rm_int<0x61, "punpcklwd", 
                                       int_x86_mmx_punpcklwd>;
defm MMX_PUNPCKLDQ : MMXI_binop_rm_int<0x62, "punpckldq",
                                       int_x86_mmx_punpckldq>;

// -- Pack Instructions
defm MMX_PACKSSWB : MMXI_binop_rm_int<0x63, "packsswb", int_x86_mmx_packsswb>;
defm MMX_PACKSSDW : MMXI_binop_rm_int<0x6B, "packssdw", int_x86_mmx_packssdw>;
defm MMX_PACKUSWB : MMXI_binop_rm_int<0x67, "packuswb", int_x86_mmx_packuswb>;

// -- Shuffle Instructions
defm MMX_PSHUFB : SS3I_binop_rm_int_mm<0x00, "pshufb", int_x86_ssse3_pshuf_b>;

def MMX_PSHUFWri : MMXIi8<0x70, MRMSrcReg,
                          (outs VR64:$dst), (ins VR64:$src1, i8imm:$src2),
                          "pshufw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
                          [(set VR64:$dst,
                             (int_x86_sse_pshuf_w VR64:$src1, imm:$src2))]>;
def MMX_PSHUFWmi : MMXIi8<0x70, MRMSrcMem,
                          (outs VR64:$dst), (ins i64mem:$src1, i8imm:$src2),
                          "pshufw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
                          [(set VR64:$dst,
                             (int_x86_sse_pshuf_w (load_mmx addr:$src1),
                                                   imm:$src2))]>;





// -- Conversion Instructions
defm MMX_CVTPS2PI : sse12_cvt_pint<0x2D, VR128, VR64, int_x86_sse_cvtps2pi,
                      f64mem, load, "cvtps2pi\t{$src, $dst|$dst, $src}",
                      SSEPackedSingle>, TB;
defm MMX_CVTPD2PI : sse12_cvt_pint<0x2D, VR128, VR64, int_x86_sse_cvtpd2pi,
                      f128mem, memop, "cvtpd2pi\t{$src, $dst|$dst, $src}",
                      SSEPackedDouble>, TB, OpSize;
defm MMX_CVTTPS2PI : sse12_cvt_pint<0x2C, VR128, VR64, int_x86_sse_cvttps2pi,
                       f64mem, load, "cvttps2pi\t{$src, $dst|$dst, $src}",
                       SSEPackedSingle>, TB;
defm MMX_CVTTPD2PI : sse12_cvt_pint<0x2C, VR128, VR64, int_x86_sse_cvttpd2pi,
                       f128mem, memop, "cvttpd2pi\t{$src, $dst|$dst, $src}",
                       SSEPackedDouble>, TB, OpSize;
defm MMX_CVTPI2PD : sse12_cvt_pint<0x2A, VR64, VR128, int_x86_sse_cvtpi2pd,
                         i64mem, load, "cvtpi2pd\t{$src, $dst|$dst, $src}",
                         SSEPackedDouble>, TB, OpSize;
let Constraints = "$src1 = $dst" in {
  defm MMX_CVTPI2PS : sse12_cvt_pint_3addr<0x2A, VR64, VR128,
                         int_x86_sse_cvtpi2ps,
                         i64mem, load, "cvtpi2ps\t{$src2, $dst|$dst, $src2}",
                         SSEPackedSingle>, TB;
}

// Extract / Insert
def MMX_PEXTRWirri: MMXIi8<0xC5, MRMSrcReg,
                           (outs GR32:$dst), (ins VR64:$src1, i32i8imm:$src2),
                           "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
                           [(set GR32:$dst, (int_x86_mmx_pextr_w VR64:$src1,
                                             (iPTR imm:$src2)))]>;
let Constraints = "$src1 = $dst" in {
  def MMX_PINSRWirri : MMXIi8<0xC4, MRMSrcReg,
                      (outs VR64:$dst), 
                      (ins VR64:$src1, GR32:$src2, i32i8imm:$src3),
                      "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
                      [(set VR64:$dst, (int_x86_mmx_pinsr_w VR64:$src1,
                                        GR32:$src2, (iPTR imm:$src3)))]>;

  def MMX_PINSRWirmi : MMXIi8<0xC4, MRMSrcMem,
                     (outs VR64:$dst),
                     (ins VR64:$src1, i16mem:$src2, i32i8imm:$src3),
                     "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
                     [(set VR64:$dst, (int_x86_mmx_pinsr_w VR64:$src1,
                                         (i32 (anyext (loadi16 addr:$src2))),
                                       (iPTR imm:$src3)))]>;
}

// Mask creation
def MMX_PMOVMSKBrr : MMXI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR64:$src),
                          "pmovmskb\t{$src, $dst|$dst, $src}",
                          [(set GR32:$dst, 
                                (int_x86_mmx_pmovmskb VR64:$src))]>;


// MMX to XMM for vector types
def MMX_X86movq2dq : SDNode<"X86ISD::MOVQ2DQ", SDTypeProfile<1, 1,
                            [SDTCisVT<0, v2i64>, SDTCisVT<1, x86mmx>]>>;

def : Pat<(v2i64 (MMX_X86movq2dq VR64:$src)),
          (v2i64 (MMX_MOVQ2DQrr VR64:$src))>;

def : Pat<(v2i64 (MMX_X86movq2dq (load_mmx addr:$src))),
          (v2i64 (MOVQI2PQIrm addr:$src))>;

def : Pat<(v2i64 (MMX_X86movq2dq 
                    (x86mmx (scalar_to_vector (loadi32 addr:$src))))),
          (v2i64 (MOVDI2PDIrm addr:$src))>;

// Low word of XMM to MMX.
def MMX_X86movdq2q : SDNode<"X86ISD::MOVDQ2Q", SDTypeProfile<1, 1,
                            [SDTCisVT<0, x86mmx>, SDTCisVT<1, v2i64>]>>;

def : Pat<(x86mmx (MMX_X86movdq2q VR128:$src)),
          (x86mmx (MMX_MOVDQ2Qrr VR128:$src))>;

def : Pat<(x86mmx (MMX_X86movdq2q (loadv2i64 addr:$src))),
          (x86mmx (MMX_MOVQ64rm addr:$src))>;

// Misc.
let Uses = [EDI] in
def MMX_MASKMOVQ : MMXI<0xF7, MRMSrcReg, (outs), (ins VR64:$src, VR64:$mask),
                        "maskmovq\t{$mask, $src|$src, $mask}",
                        [(int_x86_mmx_maskmovq VR64:$src, VR64:$mask, EDI)]>;
let Uses = [RDI] in
def MMX_MASKMOVQ64: MMXI64<0xF7, MRMSrcReg, (outs), (ins VR64:$src, VR64:$mask),
                           "maskmovq\t{$mask, $src|$src, $mask}",
                           [(int_x86_mmx_maskmovq VR64:$src, VR64:$mask, RDI)]>;

// 64-bit bit convert.
def : Pat<(x86mmx (bitconvert (i64 GR64:$src))),
          (MMX_MOVD64to64rr GR64:$src)>;
def : Pat<(i64 (bitconvert (x86mmx VR64:$src))),
          (MMX_MOVD64from64rr VR64:$src)>;
def : Pat<(f64 (bitconvert (x86mmx VR64:$src))),
          (MMX_MOVQ2FR64rr VR64:$src)>;
def : Pat<(x86mmx (bitconvert (f64 FR64:$src))),
          (MMX_MOVFR642Qrr FR64:$src)>;