//===-- X86InstrXOP.td - XOP 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 XOP (eXtended OPerations)
//
//===----------------------------------------------------------------------===//

multiclass xop2op<bits<8> opc, string OpcodeStr, Intrinsic Int, PatFrag memop> {
  def rr : IXOP<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
           !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
           [(set VR128:$dst, (Int VR128:$src))]>, XOP;
  def rm : IXOP<opc, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
           !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
           [(set VR128:$dst, (Int (bitconvert (memop addr:$src))))]>, XOP;
}

defm VPHSUBWD  : xop2op<0xE2, "vphsubwd", int_x86_xop_vphsubwd, memopv2i64>;
defm VPHSUBDQ  : xop2op<0xE3, "vphsubdq", int_x86_xop_vphsubdq, memopv2i64>;
defm VPHSUBBW  : xop2op<0xE1, "vphsubbw", int_x86_xop_vphsubbw, memopv2i64>;
defm VPHADDWQ  : xop2op<0xC7, "vphaddwq", int_x86_xop_vphaddwq, memopv2i64>;
defm VPHADDWD  : xop2op<0xC6, "vphaddwd", int_x86_xop_vphaddwd, memopv2i64>;
defm VPHADDUWQ : xop2op<0xD7, "vphadduwq", int_x86_xop_vphadduwq, memopv2i64>;
defm VPHADDUWD : xop2op<0xD6, "vphadduwd", int_x86_xop_vphadduwd, memopv2i64>;
defm VPHADDUDQ : xop2op<0xDB, "vphaddudq", int_x86_xop_vphaddudq, memopv2i64>;
defm VPHADDUBW : xop2op<0xD1, "vphaddubw", int_x86_xop_vphaddubw, memopv2i64>;
defm VPHADDUBQ : xop2op<0xD3, "vphaddubq", int_x86_xop_vphaddubq, memopv2i64>;
defm VPHADDUBD : xop2op<0xD2, "vphaddubd", int_x86_xop_vphaddubd, memopv2i64>;
defm VPHADDDQ  : xop2op<0xCB, "vphadddq", int_x86_xop_vphadddq, memopv2i64>;
defm VPHADDBW  : xop2op<0xC1, "vphaddbw", int_x86_xop_vphaddbw, memopv2i64>;
defm VPHADDBQ  : xop2op<0xC3, "vphaddbq", int_x86_xop_vphaddbq, memopv2i64>;
defm VPHADDBD  : xop2op<0xC2, "vphaddbd", int_x86_xop_vphaddbd, memopv2i64>;

// Scalar load 2 addr operand instructions
multiclass xop2opsld<bits<8> opc, string OpcodeStr, Intrinsic Int,
                     Operand memop, ComplexPattern mem_cpat> {
  def rr : IXOP<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
           !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
           [(set VR128:$dst, (Int VR128:$src))]>, XOP;
  def rm : IXOP<opc, MRMSrcMem, (outs VR128:$dst), (ins memop:$src),
           !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
           [(set VR128:$dst, (Int (bitconvert mem_cpat:$src)))]>, XOP;
}

defm VFRCZSS   : xop2opsld<0x82, "vfrczss", int_x86_xop_vfrcz_ss,
                 ssmem, sse_load_f32>;
defm VFRCZSD   : xop2opsld<0x83, "vfrczsd", int_x86_xop_vfrcz_sd,
                 sdmem, sse_load_f64>;

multiclass xop2op128<bits<8> opc, string OpcodeStr, Intrinsic Int,
                     PatFrag memop> {
  def rr : IXOP<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
           !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
           [(set VR128:$dst, (Int VR128:$src))]>, XOP;
  def rm : IXOP<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
           !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
           [(set VR128:$dst, (Int (bitconvert (memop addr:$src))))]>, XOP;
}

defm VFRCZPS : xop2op128<0x80, "vfrczps", int_x86_xop_vfrcz_ps, memopv4f32>;
defm VFRCZPD : xop2op128<0x81, "vfrczpd", int_x86_xop_vfrcz_pd, memopv2f64>;

multiclass xop2op256<bits<8> opc, string OpcodeStr, Intrinsic Int,
                     PatFrag memop> {
  def rrY : IXOP<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
           !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
           [(set VR256:$dst, (Int VR256:$src))]>, XOP, VEX_L;
  def rmY : IXOP<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
           !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
           [(set VR256:$dst, (Int (bitconvert (memop addr:$src))))]>, XOP, VEX_L;
}

defm VFRCZPS : xop2op256<0x80, "vfrczps", int_x86_xop_vfrcz_ps_256, memopv8f32>;
defm VFRCZPD : xop2op256<0x81, "vfrczpd", int_x86_xop_vfrcz_pd_256, memopv4f64>;

multiclass xop3op<bits<8> opc, string OpcodeStr, Intrinsic Int> {
  def rr : IXOP<opc, MRMSrcReg, (outs VR128:$dst),
           (ins VR128:$src1, VR128:$src2),
           !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
           [(set VR128:$dst, (Int VR128:$src1, VR128:$src2))]>, XOP_4VOp3;
  def rm : IXOP<opc, MRMSrcMem, (outs VR128:$dst),
           (ins VR128:$src1, i128mem:$src2),
           !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
           [(set VR128:$dst,
              (Int VR128:$src1, (bitconvert (memopv2i64 addr:$src2))))]>,
           XOP_4V, VEX_W;
  def mr : IXOP<opc, MRMSrcMem, (outs VR128:$dst),
           (ins i128mem:$src1, VR128:$src2),
           !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
           [(set VR128:$dst,
              (Int (bitconvert (memopv2i64 addr:$src1)), VR128:$src2))]>,
             XOP_4VOp3;
}

defm VPSHLW : xop3op<0x95, "vpshlw", int_x86_xop_vpshlw>;
defm VPSHLQ : xop3op<0x97, "vpshlq", int_x86_xop_vpshlq>;
defm VPSHLD : xop3op<0x96, "vpshld", int_x86_xop_vpshld>;
defm VPSHLB : xop3op<0x94, "vpshlb", int_x86_xop_vpshlb>;
defm VPSHAW : xop3op<0x99, "vpshaw", int_x86_xop_vpshaw>;
defm VPSHAQ : xop3op<0x9B, "vpshaq", int_x86_xop_vpshaq>;
defm VPSHAD : xop3op<0x9A, "vpshad", int_x86_xop_vpshad>;
defm VPSHAB : xop3op<0x98, "vpshab", int_x86_xop_vpshab>;
defm VPROTW : xop3op<0x91, "vprotw", int_x86_xop_vprotw>;
defm VPROTQ : xop3op<0x93, "vprotq", int_x86_xop_vprotq>;
defm VPROTD : xop3op<0x92, "vprotd", int_x86_xop_vprotd>;
defm VPROTB : xop3op<0x90, "vprotb", int_x86_xop_vprotb>;

multiclass xop3opimm<bits<8> opc, string OpcodeStr, Intrinsic Int> {
  def ri : IXOPi8<opc, MRMSrcReg, (outs VR128:$dst),
           (ins VR128:$src1, i8imm:$src2),
           !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
           [(set VR128:$dst, (Int VR128:$src1, imm:$src2))]>, XOP;
  def mi : IXOPi8<opc, MRMSrcMem, (outs VR128:$dst),
           (ins i128mem:$src1, i8imm:$src2),
           !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
           [(set VR128:$dst,
             (Int (bitconvert (memopv2i64 addr:$src1)), imm:$src2))]>, XOP;
}

defm VPROTW : xop3opimm<0xC1, "vprotw", int_x86_xop_vprotwi>;
defm VPROTQ : xop3opimm<0xC3, "vprotq", int_x86_xop_vprotqi>;
defm VPROTD : xop3opimm<0xC2, "vprotd", int_x86_xop_vprotdi>;
defm VPROTB : xop3opimm<0xC0, "vprotb", int_x86_xop_vprotbi>;

// Instruction where second source can be memory, but third must be register
multiclass xop4opm2<bits<8> opc, string OpcodeStr, Intrinsic Int> {
  def rr : IXOPi8<opc, MRMSrcReg, (outs VR128:$dst),
           (ins VR128:$src1, VR128:$src2, VR128:$src3),
           !strconcat(OpcodeStr,
           "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
           [(set VR128:$dst,
              (Int VR128:$src1, VR128:$src2, VR128:$src3))]>, XOP_4V, VEX_I8IMM;
  def rm : IXOPi8<opc, MRMSrcMem, (outs VR128:$dst),
           (ins VR128:$src1, i128mem:$src2, VR128:$src3),
           !strconcat(OpcodeStr,
           "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
           [(set VR128:$dst,
              (Int VR128:$src1, (bitconvert (memopv2i64 addr:$src2)),
              VR128:$src3))]>, XOP_4V, VEX_I8IMM;
}

defm VPMADCSWD  : xop4opm2<0xB6, "vpmadcswd", int_x86_xop_vpmadcswd>;
defm VPMADCSSWD : xop4opm2<0xA6, "vpmadcsswd", int_x86_xop_vpmadcsswd>;
defm VPMACSWW   : xop4opm2<0x95, "vpmacsww", int_x86_xop_vpmacsww>;
defm VPMACSWD   : xop4opm2<0x96, "vpmacswd", int_x86_xop_vpmacswd>;
defm VPMACSSWW  : xop4opm2<0x85, "vpmacssww", int_x86_xop_vpmacssww>;
defm VPMACSSWD  : xop4opm2<0x86, "vpmacsswd", int_x86_xop_vpmacsswd>;
defm VPMACSSDQL : xop4opm2<0x87, "vpmacssdql", int_x86_xop_vpmacssdql>;
defm VPMACSSDQH : xop4opm2<0x8F, "vpmacssdqh", int_x86_xop_vpmacssdqh>;
defm VPMACSSDD  : xop4opm2<0x8E, "vpmacssdd", int_x86_xop_vpmacssdd>;
defm VPMACSDQL  : xop4opm2<0x97, "vpmacsdql", int_x86_xop_vpmacsdql>;
defm VPMACSDQH  : xop4opm2<0x9F, "vpmacsdqh", int_x86_xop_vpmacsdqh>;
defm VPMACSDD   : xop4opm2<0x9E, "vpmacsdd", int_x86_xop_vpmacsdd>;

// Instruction where second source can be memory, third must be imm8
multiclass xop4opimm<bits<8> opc, string OpcodeStr, Intrinsic Int> {
  def ri : IXOPi8<opc, MRMSrcReg, (outs VR128:$dst),
           (ins VR128:$src1, VR128:$src2, i8imm:$src3),
           !strconcat(OpcodeStr,
           "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
           [(set VR128:$dst, (Int VR128:$src1, VR128:$src2, imm:$src3))]>,
           XOP_4V;
  def mi : IXOPi8<opc, MRMSrcMem, (outs VR128:$dst),
           (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
           !strconcat(OpcodeStr,
           "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
           [(set VR128:$dst,
             (Int VR128:$src1, (bitconvert (memopv2i64 addr:$src2)),
              imm:$src3))]>, XOP_4V;
}

defm VPCOMB  : xop4opimm<0xCC, "vpcomb", int_x86_xop_vpcomb>;
defm VPCOMW  : xop4opimm<0xCD, "vpcomw", int_x86_xop_vpcomw>;
defm VPCOMD  : xop4opimm<0xCE, "vpcomd", int_x86_xop_vpcomd>;
defm VPCOMQ  : xop4opimm<0xCF, "vpcomq", int_x86_xop_vpcomq>;
defm VPCOMUB : xop4opimm<0xEC, "vpcomub", int_x86_xop_vpcomub>;
defm VPCOMUW : xop4opimm<0xED, "vpcomuw", int_x86_xop_vpcomuw>;
defm VPCOMUD : xop4opimm<0xEE, "vpcomud", int_x86_xop_vpcomud>;
defm VPCOMUQ : xop4opimm<0xEF, "vpcomuq", int_x86_xop_vpcomuq>;

// Instruction where either second or third source can be memory
multiclass xop4op<bits<8> opc, string OpcodeStr, Intrinsic Int> {
  def rr : IXOPi8<opc, MRMSrcReg, (outs VR128:$dst),
           (ins VR128:$src1, VR128:$src2, VR128:$src3),
           !strconcat(OpcodeStr,
           "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
           [(set VR128:$dst, (Int VR128:$src1, VR128:$src2, VR128:$src3))]>,
           XOP_4V, VEX_I8IMM;
  def rm : IXOPi8<opc, MRMSrcMem, (outs VR128:$dst),
           (ins VR128:$src1, VR128:$src2, i128mem:$src3),
           !strconcat(OpcodeStr,
           "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
           [(set VR128:$dst,
             (Int VR128:$src1, VR128:$src2,
              (bitconvert (memopv2i64 addr:$src3))))]>,
           XOP_4V, VEX_I8IMM, VEX_W, MemOp4;
  def mr : IXOPi8<opc, MRMSrcMem, (outs VR128:$dst),
           (ins VR128:$src1, i128mem:$src2, VR128:$src3),
           !strconcat(OpcodeStr,
           "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
           [(set VR128:$dst,
             (Int VR128:$src1, (bitconvert (memopv2i64 addr:$src2)),
              VR128:$src3))]>,
           XOP_4V, VEX_I8IMM;
}

defm VPPERM : xop4op<0xA3, "vpperm", int_x86_xop_vpperm>;
defm VPCMOV : xop4op<0xA2, "vpcmov", int_x86_xop_vpcmov>;

multiclass xop4op256<bits<8> opc, string OpcodeStr, Intrinsic Int> {
  def rrY : IXOPi8<opc, MRMSrcReg, (outs VR256:$dst),
           (ins VR256:$src1, VR256:$src2, VR256:$src3),
           !strconcat(OpcodeStr,
           "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
           [(set VR256:$dst, (Int VR256:$src1, VR256:$src2, VR256:$src3))]>,
           XOP_4V, VEX_I8IMM, VEX_L;
  def rmY : IXOPi8<opc, MRMSrcMem, (outs VR256:$dst),
           (ins VR256:$src1, VR256:$src2, i256mem:$src3),
           !strconcat(OpcodeStr,
           "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
           [(set VR256:$dst,
             (Int VR256:$src1, VR256:$src2,
              (bitconvert (memopv4i64 addr:$src3))))]>,
           XOP_4V, VEX_I8IMM, VEX_W, MemOp4, VEX_L;
  def mrY : IXOPi8<opc, MRMSrcMem, (outs VR256:$dst),
           (ins VR256:$src1, f256mem:$src2, VR256:$src3),
           !strconcat(OpcodeStr,
           "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
           [(set VR256:$dst,
             (Int VR256:$src1, (bitconvert (memopv4i64 addr:$src2)),
              VR256:$src3))]>,
           XOP_4V, VEX_I8IMM, VEX_L;
}

defm VPCMOV : xop4op256<0xA2, "vpcmov", int_x86_xop_vpcmov_256>;

multiclass xop5op<bits<8> opc, string OpcodeStr, Intrinsic Int128,
                  Intrinsic Int256, PatFrag ld_128, PatFrag ld_256> {
  def rr : IXOP5<opc, MRMSrcReg, (outs VR128:$dst),
        (ins VR128:$src1, VR128:$src2, VR128:$src3, i8imm:$src4),
        !strconcat(OpcodeStr,
        "\t{$src4, $src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3, $src4}"),
        [(set VR128:$dst,
           (Int128 VR128:$src1, VR128:$src2, VR128:$src3, imm:$src4))]>;
  def rm : IXOP5<opc, MRMSrcMem, (outs VR128:$dst),
        (ins VR128:$src1, VR128:$src2, f128mem:$src3, i8imm:$src4),
        !strconcat(OpcodeStr,
        "\t{$src4, $src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3, $src4}"),
        [(set VR128:$dst,
           (Int128 VR128:$src1, VR128:$src2, (ld_128 addr:$src3), imm:$src4))]>,
        VEX_W, MemOp4;
  def mr : IXOP5<opc, MRMSrcMem, (outs VR128:$dst),
        (ins VR128:$src1, f128mem:$src2, VR128:$src3, i8imm:$src4),
        !strconcat(OpcodeStr,
        "\t{$src4, $src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3, $src4}"),
        [(set VR128:$dst,
           (Int128 VR128:$src1, (ld_128 addr:$src2), VR128:$src3, imm:$src4))]>;
  def rrY : IXOP5<opc, MRMSrcReg, (outs VR256:$dst),
        (ins VR256:$src1, VR256:$src2, VR256:$src3, i8imm:$src4),
        !strconcat(OpcodeStr,
        "\t{$src4, $src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3, $src4}"),
        [(set VR256:$dst,
          (Int256 VR256:$src1, VR256:$src2, VR256:$src3, imm:$src4))]>, VEX_L;
  def rmY : IXOP5<opc, MRMSrcMem, (outs VR256:$dst),
        (ins VR256:$src1, VR256:$src2, f256mem:$src3, i8imm:$src4),
        !strconcat(OpcodeStr,
        "\t{$src4, $src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3, $src4}"),
        [(set VR256:$dst,
          (Int256 VR256:$src1, VR256:$src2, (ld_256 addr:$src3), imm:$src4))]>,
        VEX_W, MemOp4, VEX_L;
  def mrY : IXOP5<opc, MRMSrcMem, (outs VR256:$dst),
        (ins VR256:$src1, f256mem:$src2, VR256:$src3, i8imm:$src4),
        !strconcat(OpcodeStr,
        "\t{$src4, $src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3, $src4}"),
        [(set VR256:$dst,
           (Int256 VR256:$src1, (ld_256 addr:$src2), VR256:$src3, imm:$src4))]>,
        VEX_L;
}

defm VPERMIL2PD : xop5op<0x49, "vpermil2pd", int_x86_xop_vpermil2pd,
                         int_x86_xop_vpermil2pd_256, memopv2f64, memopv4f64>;
defm VPERMIL2PS : xop5op<0x48, "vpermil2ps", int_x86_xop_vpermil2ps,
                         int_x86_xop_vpermil2ps_256, memopv4f32, memopv8f32>;