llvm-project/llvm/lib/Target/X86/X86InstrMMX.td

669 lines
31 KiB
TableGen
Raw Normal View History

//===-- 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 Sched = WriteVecALU in {
def MMX_INTALU_ITINS : OpndItins<
IIC_MMX_ALU_RR, IIC_MMX_ALU_RM
>;
def MMX_INTALUQ_ITINS : OpndItins<
IIC_MMX_ALUQ_RR, IIC_MMX_ALUQ_RM
>;
def MMX_PHADDSUBW : OpndItins<
IIC_MMX_PHADDSUBW_RR, IIC_MMX_PHADDSUBW_RM
>;
def MMX_PHADDSUBD : OpndItins<
IIC_MMX_PHADDSUBD_RR, IIC_MMX_PHADDSUBD_RM
>;
}
let Sched = WriteVecLogic in
def MMX_INTALU_ITINS_VECLOGICSCHED : OpndItins<
IIC_MMX_ALU_RR, IIC_MMX_ALU_RM
>;
let Sched = WriteVecIMul in
def MMX_PMUL_ITINS : OpndItins<
IIC_MMX_PMUL, IIC_MMX_PMUL
>;
let Sched = WriteVecIMul in {
def MMX_PSADBW_ITINS : OpndItins<
IIC_MMX_PSADBW, IIC_MMX_PSADBW
>;
def MMX_MISC_FUNC_ITINS : OpndItins<
IIC_MMX_MISC_FUNC_MEM, IIC_MMX_MISC_FUNC_REG
>;
}
def MMX_SHIFT_ITINS : ShiftOpndItins<
IIC_MMX_SHIFT_RR, IIC_MMX_SHIFT_RM, IIC_MMX_SHIFT_RI
>;
let Sched = WriteShuffle in {
def MMX_UNPCK_H_ITINS : OpndItins<
IIC_MMX_UNPCK_H_RR, IIC_MMX_UNPCK_H_RM
>;
def MMX_UNPCK_L_ITINS : OpndItins<
IIC_MMX_UNPCK_L, IIC_MMX_UNPCK_L
>;
def MMX_PCK_ITINS : OpndItins<
IIC_MMX_PCK_RR, IIC_MMX_PCK_RM
>;
def MMX_PSHUF_ITINS : OpndItins<
IIC_MMX_PSHUF, IIC_MMX_PSHUF
>;
} // Sched
let Sched = WriteCvtF2I in {
def MMX_CVT_PD_ITINS : OpndItins<
IIC_MMX_CVT_PD_RR, IIC_MMX_CVT_PD_RM
>;
def MMX_CVT_PS_ITINS : OpndItins<
IIC_MMX_CVT_PS_RR, IIC_MMX_CVT_PS_RM
>;
}
let Constraints = "$src1 = $dst" in {
// MMXI_binop_rm_int - Simple MMX binary operator based on intrinsic.
// When this is cleaned up, remove the FIXME from X86RecognizableInstr.cpp.
multiclass MMXI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
OpndItins itins, bit Commutable = 0,
X86MemOperand OType = i64mem> {
def irr : MMXI<opc, MRMSrcReg, (outs VR64:$dst),
(ins VR64:$src1, VR64:$src2),
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
[(set VR64:$dst, (IntId VR64:$src1, VR64:$src2))], itins.rr>,
Sched<[itins.Sched]> {
let isCommutable = Commutable;
}
def irm : MMXI<opc, MRMSrcMem, (outs VR64:$dst),
(ins VR64:$src1, OType:$src2),
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
[(set VR64:$dst, (IntId VR64:$src1,
(bitconvert (load_mmx addr:$src2))))],
itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
}
multiclass MMXI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
string OpcodeStr, Intrinsic IntId,
Intrinsic IntId2, ShiftOpndItins itins> {
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))], itins.rr>,
Sched<[WriteVecShift]>;
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))))],
itins.rm>, Sched<[WriteVecShiftLd, ReadAfterLd]>;
def ri : MMXIi8<opc2, ImmForm, (outs VR64:$dst),
(ins VR64:$src1, i32u8imm:$src2),
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
[(set VR64:$dst, (IntId2 VR64:$src1, imm:$src2))], itins.ri>,
Sched<[WriteVecShift]>;
}
}
/// Unary MMX instructions requiring SSSE3.
multiclass SS3I_unop_rm_int_mm<bits<8> opc, string OpcodeStr,
Intrinsic IntId64, OpndItins itins> {
def rr64 : MMXSS38I<opc, MRMSrcReg, (outs VR64:$dst), (ins VR64:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[(set VR64:$dst, (IntId64 VR64:$src))], itins.rr>,
Sched<[itins.Sched]>;
def rm64 : MMXSS38I<opc, MRMSrcMem, (outs VR64:$dst), (ins i64mem:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[(set VR64:$dst,
(IntId64 (bitconvert (load_mmx addr:$src))))],
itins.rm>, Sched<[itins.Sched.Folded]>;
}
/// Binary MMX instructions requiring SSSE3.
let ImmT = NoImm, Constraints = "$src1 = $dst" in {
multiclass SS3I_binop_rm_int_mm<bits<8> opc, string OpcodeStr,
Intrinsic IntId64, OpndItins itins,
bit Commutable = 0> {
let isCommutable = Commutable in
def rr64 : MMXSS38I<opc, MRMSrcReg, (outs VR64:$dst),
(ins VR64:$src1, VR64:$src2),
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
[(set VR64:$dst, (IntId64 VR64:$src1, VR64:$src2))], itins.rr>,
Sched<[itins.Sched]>;
def rm64 : MMXSS38I<opc, MRMSrcMem, (outs VR64:$dst),
(ins VR64:$src1, i64mem:$src2),
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
[(set VR64:$dst,
(IntId64 VR64:$src1,
(bitconvert (load_mmx addr:$src2))))], itins.rm>,
Sched<[itins.Sched.Folded, ReadAfterLd]>;
}
}
/// PALIGN MMX instructions (require SSSE3).
multiclass ssse3_palign_mm<string asm, Intrinsic IntId> {
def R64irr : MMXSS3AI<0x0F, MRMSrcReg, (outs VR64:$dst),
(ins VR64:$src1, VR64:$src2, u8imm:$src3),
!strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
[(set VR64:$dst, (IntId VR64:$src1, VR64:$src2, (i8 imm:$src3)))]>,
Sched<[WriteShuffle]>;
def R64irm : MMXSS3AI<0x0F, MRMSrcMem, (outs VR64:$dst),
(ins VR64:$src1, i64mem:$src2, u8imm:$src3),
!strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
[(set VR64:$dst, (IntId VR64:$src1,
(bitconvert (load_mmx addr:$src2)), (i8 imm:$src3)))]>,
Sched<[WriteShuffleLd, ReadAfterLd]>;
}
multiclass sse12_cvt_pint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
string asm, OpndItins itins, Domain d> {
def irr : MMXPI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
[(set DstRC:$dst, (Int SrcRC:$src))], itins.rr, d>,
Sched<[itins.Sched]>;
def irm : MMXPI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
[(set DstRC:$dst, (Int (ld_frag addr:$src)))], itins.rm, d>,
Sched<[itins.Sched.Folded]>;
}
multiclass sse12_cvt_pint_3addr<bits<8> opc, RegisterClass SrcRC,
RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
PatFrag ld_frag, string asm, Domain d> {
def irr : MMXPI<opc, MRMSrcReg, (outs DstRC:$dst),
(ins DstRC:$src1, SrcRC:$src2), asm,
[(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))],
NoItinerary, d>, Sched<[WriteCvtI2F]>;
def irm : MMXPI<opc, MRMSrcMem, (outs DstRC:$dst),
(ins DstRC:$src1, x86memop:$src2), asm,
[(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))],
NoItinerary, d>, Sched<[WriteCvtI2FLd]>;
}
//===----------------------------------------------------------------------===//
// MMX EMMS Instruction
//===----------------------------------------------------------------------===//
def MMX_EMMS : MMXI<0x77, RawFrm, (outs), (ins), "emms",
[(int_x86_mmx_emms)], IIC_MMX_EMMS>;
//===----------------------------------------------------------------------===//
// MMX Scalar Instructions
//===----------------------------------------------------------------------===//
// Data Transfer Instructions
def MMX_MOVD64rr : MMXI<0x6E, MRMSrcReg, (outs VR64:$dst), (ins GR32:$src),
"movd\t{$src, $dst|$dst, $src}",
[(set VR64:$dst,
(x86mmx (scalar_to_vector GR32:$src)))],
IIC_MMX_MOV_MM_RM>, Sched<[WriteMove]>;
def MMX_MOVD64rm : MMXI<0x6E, MRMSrcMem, (outs VR64:$dst), (ins i32mem:$src),
"movd\t{$src, $dst|$dst, $src}",
[(set VR64:$dst,
(x86mmx (scalar_to_vector (loadi32 addr:$src))))],
IIC_MMX_MOV_MM_RM>, Sched<[WriteLoad]>;
let Predicates = [HasMMX] in {
let AddedComplexity = 15 in
def : Pat<(x86mmx (MMX_X86movw2d GR32:$src)),
(MMX_MOVD64rr GR32:$src)>;
let AddedComplexity = 20 in
def : Pat<(x86mmx (MMX_X86movw2d (loadi32 addr:$src))),
(MMX_MOVD64rm addr:$src)>;
}
let mayStore = 1 in
def MMX_MOVD64mr : MMXI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR64:$src),
"movd\t{$src, $dst|$dst, $src}", [], IIC_MMX_MOV_MM_RM>,
Sched<[WriteStore]>;
def MMX_MOVD64grr : MMXI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR64:$src),
"movd\t{$src, $dst|$dst, $src}",
[(set GR32:$dst,
(MMX_X86movd2w (x86mmx VR64:$src)))],
[X86] Adding FoldGenRegForm helper field (for memory folding tables tableGen backend) to X86Inst class and set its value for the relevant instructions. Some register-register instructions can be encoded in 2 different ways, this happens when 2 register operands can be folded (separately). For example if we look at the MOV8rr and MOV8rr_REV, both instructions perform exactly the same operation, but are encoded differently. Here is the relevant information about these instructions from Intel's 64-ia-32-architectures-software-developer-manual: Opcode Instruction Op/En 64-Bit Mode Compat/Leg Mode Description 8A /r MOV r8,r/m8 RM Valid Valid Move r/m8 to r8. 88 /r MOV r/m8,r8 MR Valid Valid Move r8 to r/m8. Here we can see that in order to enable the folding of the output and input registers, we had to define 2 "encodings", and as a result we got 2 move 8-bit register-register instructions. In the X86 backend, we define both of these instructions, usually one has a regular name (MOV8rr) while the other has "_REV" suffix (MOV8rr_REV), must be marked with isCodeGenOnly flag and is not emitted from CodeGen. Automatically generating the memory folding tables relies on matching encodings of instructions, but in these cases where we want to map both memory forms of the mov 8-bit (MOV8rm & MOV8mr) to MOV8rr (not to MOV8rr_REV) we have to somehow point from the MOV8rr_REV to the "regular" appropriate instruction which in this case is MOV8rr. This field enable this "pointing" mechanism - which is used in the TableGen backend for generating memory folding tables. Differential Revision: https://reviews.llvm.org/D32683 llvm-svn: 304087
2017-05-28 20:39:37 +08:00
IIC_MMX_MOV_REG_MM>, Sched<[WriteMove]>,
FoldGenData<"MMX_MOVD64rr">;
let isBitcast = 1 in
def MMX_MOVD64to64rr : MMXRI<0x6E, MRMSrcReg, (outs VR64:$dst), (ins GR64:$src),
"movq\t{$src, $dst|$dst, $src}",
[(set VR64:$dst, (bitconvert GR64:$src))],
IIC_MMX_MOV_MM_RM>, Sched<[WriteMove]>;
let isCodeGenOnly = 1, ForceDisassemble = 1, hasSideEffects = 0, mayLoad = 1 in
def MMX_MOVD64to64rm : MMXRI<0x6E, MRMSrcMem, (outs VR64:$dst),
(ins i64mem:$src), "movq\t{$src, $dst|$dst, $src}",
[], IIC_MMX_MOVQ_RM>, Sched<[WriteLoad]>;
// These are 64 bit moves, but since the OS X assembler doesn't
// recognize a register-register movq, we write them as
// movd.
let SchedRW = [WriteMove], isBitcast = 1 in {
def MMX_MOVD64from64rr : MMXRI<0x7E, MRMDestReg,
(outs GR64:$dst), (ins VR64:$src),
"movq\t{$src, $dst|$dst, $src}",
[(set GR64:$dst,
(bitconvert VR64:$src))], IIC_MMX_MOV_REG_MM>;
let hasSideEffects = 0 in
def MMX_MOVQ64rr : MMXI<0x6F, MRMSrcReg, (outs VR64:$dst), (ins VR64:$src),
"movq\t{$src, $dst|$dst, $src}", [],
IIC_MMX_MOVQ_RR>;
let isCodeGenOnly = 1, ForceDisassemble = 1, hasSideEffects = 0 in {
def MMX_MOVQ64rr_REV : MMXI<0x7F, MRMDestReg, (outs VR64:$dst), (ins VR64:$src),
"movq\t{$src, $dst|$dst, $src}", [],
[X86] Adding FoldGenRegForm helper field (for memory folding tables tableGen backend) to X86Inst class and set its value for the relevant instructions. Some register-register instructions can be encoded in 2 different ways, this happens when 2 register operands can be folded (separately). For example if we look at the MOV8rr and MOV8rr_REV, both instructions perform exactly the same operation, but are encoded differently. Here is the relevant information about these instructions from Intel's 64-ia-32-architectures-software-developer-manual: Opcode Instruction Op/En 64-Bit Mode Compat/Leg Mode Description 8A /r MOV r8,r/m8 RM Valid Valid Move r/m8 to r8. 88 /r MOV r/m8,r8 MR Valid Valid Move r8 to r/m8. Here we can see that in order to enable the folding of the output and input registers, we had to define 2 "encodings", and as a result we got 2 move 8-bit register-register instructions. In the X86 backend, we define both of these instructions, usually one has a regular name (MOV8rr) while the other has "_REV" suffix (MOV8rr_REV), must be marked with isCodeGenOnly flag and is not emitted from CodeGen. Automatically generating the memory folding tables relies on matching encodings of instructions, but in these cases where we want to map both memory forms of the mov 8-bit (MOV8rm & MOV8mr) to MOV8rr (not to MOV8rr_REV) we have to somehow point from the MOV8rr_REV to the "regular" appropriate instruction which in this case is MOV8rr. This field enable this "pointing" mechanism - which is used in the TableGen backend for generating memory folding tables. Differential Revision: https://reviews.llvm.org/D32683 llvm-svn: 304087
2017-05-28 20:39:37 +08:00
IIC_MMX_MOVQ_RR>, FoldGenData<"MMX_MOVQ64rr">;
}
} // SchedRW
let isCodeGenOnly = 1, ForceDisassemble = 1, hasSideEffects = 0, mayStore = 1 in
def MMX_MOVD64from64rm : MMXRI<0x7E, MRMDestMem,
(outs), (ins i64mem:$dst, VR64:$src),
"movq\t{$src, $dst|$dst, $src}",
[], IIC_MMX_MOV_REG_MM>, Sched<[WriteStore]>;
let SchedRW = [WriteLoad] in {
let canFoldAsLoad = 1 in
def MMX_MOVQ64rm : MMXI<0x6F, MRMSrcMem, (outs VR64:$dst), (ins i64mem:$src),
"movq\t{$src, $dst|$dst, $src}",
[(set VR64:$dst, (load_mmx addr:$src))],
IIC_MMX_MOVQ_RM>;
} // SchedRW
let SchedRW = [WriteStore] in
def MMX_MOVQ64mr : MMXI<0x7F, MRMDestMem, (outs), (ins i64mem:$dst, VR64:$src),
"movq\t{$src, $dst|$dst, $src}",
[(store (x86mmx VR64:$src), addr:$dst)],
IIC_MMX_MOVQ_RM>;
let SchedRW = [WriteMove] in {
def MMX_MOVDQ2Qrr : MMXSDIi8<0xD6, MRMSrcReg, (outs VR64:$dst),
(ins VR128:$src), "movdq2q\t{$src, $dst|$dst, $src}",
[(set VR64:$dst,
(x86mmx (bitconvert
(i64 (extractelt (v2i64 VR128:$src),
(iPTR 0))))))],
IIC_MMX_MOVQ_RR>;
def MMX_MOVQ2DQrr : MMXS2SIi8<0xD6, MRMSrcReg, (outs VR128:$dst),
(ins VR64:$src), "movq2dq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
(v2i64
(scalar_to_vector
(i64 (bitconvert (x86mmx VR64:$src))))))],
IIC_MMX_MOVQ_RR>;
let isCodeGenOnly = 1, hasSideEffects = 1 in {
def MMX_MOVQ2FR64rr: MMXS2SIi8<0xD6, MRMSrcReg, (outs FR64:$dst),
(ins VR64:$src), "movq2dq\t{$src, $dst|$dst, $src}",
[], IIC_MMX_MOVQ_RR>;
def MMX_MOVFR642Qrr: MMXSDIi8<0xD6, MRMSrcReg, (outs VR64:$dst),
(ins FR64:$src), "movdq2q\t{$src, $dst|$dst, $src}",
[], IIC_MMX_MOVQ_RR>;
}
} // SchedRW
let Predicates = [HasSSE1] in
def MMX_MOVNTQmr : MMXI<0xE7, MRMDestMem, (outs), (ins i64mem:$dst, VR64:$src),
"movntq\t{$src, $dst|$dst, $src}",
[(int_x86_mmx_movnt_dq addr:$dst, VR64:$src)],
IIC_MMX_MOVQ_RM>, Sched<[WriteStore]>;
let Predicates = [HasMMX] in {
let AddedComplexity = 15 in
// movd to MMX register zero-extends
def : Pat<(x86mmx (X86vzmovl (x86mmx (scalar_to_vector GR32:$src)))),
(MMX_MOVD64rr GR32:$src)>;
let AddedComplexity = 20 in
def : Pat<(x86mmx (X86vzmovl (x86mmx (scalar_to_vector (loadi32 addr:$src))))),
(MMX_MOVD64rm addr:$src)>;
}
// Arithmetic Instructions
defm MMX_PABSB : SS3I_unop_rm_int_mm<0x1C, "pabsb", int_x86_ssse3_pabs_b,
MMX_INTALU_ITINS>;
defm MMX_PABSW : SS3I_unop_rm_int_mm<0x1D, "pabsw", int_x86_ssse3_pabs_w,
MMX_INTALU_ITINS>;
defm MMX_PABSD : SS3I_unop_rm_int_mm<0x1E, "pabsd", int_x86_ssse3_pabs_d,
MMX_INTALU_ITINS>;
// -- Addition
defm MMX_PADDB : MMXI_binop_rm_int<0xFC, "paddb", int_x86_mmx_padd_b,
MMX_INTALU_ITINS, 1>;
defm MMX_PADDW : MMXI_binop_rm_int<0xFD, "paddw", int_x86_mmx_padd_w,
MMX_INTALU_ITINS, 1>;
defm MMX_PADDD : MMXI_binop_rm_int<0xFE, "paddd", int_x86_mmx_padd_d,
MMX_INTALU_ITINS, 1>;
let Predicates = [HasSSE2] in
defm MMX_PADDQ : MMXI_binop_rm_int<0xD4, "paddq", int_x86_mmx_padd_q,
MMX_INTALUQ_ITINS, 1>;
defm MMX_PADDSB : MMXI_binop_rm_int<0xEC, "paddsb" , int_x86_mmx_padds_b,
MMX_INTALU_ITINS, 1>;
defm MMX_PADDSW : MMXI_binop_rm_int<0xED, "paddsw" , int_x86_mmx_padds_w,
MMX_INTALU_ITINS, 1>;
defm MMX_PADDUSB : MMXI_binop_rm_int<0xDC, "paddusb", int_x86_mmx_paddus_b,
MMX_INTALU_ITINS, 1>;
defm MMX_PADDUSW : MMXI_binop_rm_int<0xDD, "paddusw", int_x86_mmx_paddus_w,
MMX_INTALU_ITINS, 1>;
defm MMX_PHADDW : SS3I_binop_rm_int_mm<0x01, "phaddw", int_x86_ssse3_phadd_w,
MMX_PHADDSUBW>;
defm MMX_PHADD : SS3I_binop_rm_int_mm<0x02, "phaddd", int_x86_ssse3_phadd_d,
MMX_PHADDSUBD>;
defm MMX_PHADDSW : SS3I_binop_rm_int_mm<0x03, "phaddsw",int_x86_ssse3_phadd_sw,
MMX_PHADDSUBW>;
// -- Subtraction
defm MMX_PSUBB : MMXI_binop_rm_int<0xF8, "psubb", int_x86_mmx_psub_b,
MMX_INTALU_ITINS>;
defm MMX_PSUBW : MMXI_binop_rm_int<0xF9, "psubw", int_x86_mmx_psub_w,
MMX_INTALU_ITINS>;
defm MMX_PSUBD : MMXI_binop_rm_int<0xFA, "psubd", int_x86_mmx_psub_d,
MMX_INTALU_ITINS>;
let Predicates = [HasSSE2] in
defm MMX_PSUBQ : MMXI_binop_rm_int<0xFB, "psubq", int_x86_mmx_psub_q,
MMX_INTALUQ_ITINS>;
defm MMX_PSUBSB : MMXI_binop_rm_int<0xE8, "psubsb" , int_x86_mmx_psubs_b,
MMX_INTALU_ITINS>;
defm MMX_PSUBSW : MMXI_binop_rm_int<0xE9, "psubsw" , int_x86_mmx_psubs_w,
MMX_INTALU_ITINS>;
defm MMX_PSUBUSB : MMXI_binop_rm_int<0xD8, "psubusb", int_x86_mmx_psubus_b,
MMX_INTALU_ITINS>;
defm MMX_PSUBUSW : MMXI_binop_rm_int<0xD9, "psubusw", int_x86_mmx_psubus_w,
MMX_INTALU_ITINS>;
defm MMX_PHSUBW : SS3I_binop_rm_int_mm<0x05, "phsubw", int_x86_ssse3_phsub_w,
MMX_PHADDSUBW>;
defm MMX_PHSUBD : SS3I_binop_rm_int_mm<0x06, "phsubd", int_x86_ssse3_phsub_d,
MMX_PHADDSUBD>;
defm MMX_PHSUBSW : SS3I_binop_rm_int_mm<0x07, "phsubsw",int_x86_ssse3_phsub_sw,
MMX_PHADDSUBW>;
// -- Multiplication
defm MMX_PMULLW : MMXI_binop_rm_int<0xD5, "pmullw", int_x86_mmx_pmull_w,
MMX_PMUL_ITINS, 1>;
defm MMX_PMULHW : MMXI_binop_rm_int<0xE5, "pmulhw", int_x86_mmx_pmulh_w,
MMX_PMUL_ITINS, 1>;
let Predicates = [HasSSE1] in
defm MMX_PMULHUW : MMXI_binop_rm_int<0xE4, "pmulhuw", int_x86_mmx_pmulhu_w,
MMX_PMUL_ITINS, 1>;
let Predicates = [HasSSE2] in
defm MMX_PMULUDQ : MMXI_binop_rm_int<0xF4, "pmuludq", int_x86_mmx_pmulu_dq,
MMX_PMUL_ITINS, 1>;
defm MMX_PMULHRSW : SS3I_binop_rm_int_mm<0x0B, "pmulhrsw",
int_x86_ssse3_pmul_hr_sw,
MMX_PMUL_ITINS, 1>;
// -- Miscellanea
defm MMX_PMADDWD : MMXI_binop_rm_int<0xF5, "pmaddwd", int_x86_mmx_pmadd_wd,
MMX_PMUL_ITINS, 1>;
defm MMX_PMADDUBSW : SS3I_binop_rm_int_mm<0x04, "pmaddubsw",
int_x86_ssse3_pmadd_ub_sw, MMX_PMUL_ITINS>;
let Predicates = [HasSSE1] in {
defm MMX_PAVGB : MMXI_binop_rm_int<0xE0, "pavgb", int_x86_mmx_pavg_b,
MMX_MISC_FUNC_ITINS, 1>;
defm MMX_PAVGW : MMXI_binop_rm_int<0xE3, "pavgw", int_x86_mmx_pavg_w,
MMX_MISC_FUNC_ITINS, 1>;
defm MMX_PMINUB : MMXI_binop_rm_int<0xDA, "pminub", int_x86_mmx_pminu_b,
MMX_MISC_FUNC_ITINS, 1>;
defm MMX_PMINSW : MMXI_binop_rm_int<0xEA, "pminsw", int_x86_mmx_pmins_w,
MMX_MISC_FUNC_ITINS, 1>;
defm MMX_PMAXUB : MMXI_binop_rm_int<0xDE, "pmaxub", int_x86_mmx_pmaxu_b,
MMX_MISC_FUNC_ITINS, 1>;
defm MMX_PMAXSW : MMXI_binop_rm_int<0xEE, "pmaxsw", int_x86_mmx_pmaxs_w,
MMX_MISC_FUNC_ITINS, 1>;
defm MMX_PSADBW : MMXI_binop_rm_int<0xF6, "psadbw", int_x86_mmx_psad_bw,
MMX_PSADBW_ITINS, 1>;
}
defm MMX_PSIGNB : SS3I_binop_rm_int_mm<0x08, "psignb", int_x86_ssse3_psign_b,
MMX_MISC_FUNC_ITINS>;
defm MMX_PSIGNW : SS3I_binop_rm_int_mm<0x09, "psignw", int_x86_ssse3_psign_w,
MMX_MISC_FUNC_ITINS>;
defm MMX_PSIGND : SS3I_binop_rm_int_mm<0x0A, "psignd", int_x86_ssse3_psign_d,
MMX_MISC_FUNC_ITINS>;
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,
MMX_INTALU_ITINS_VECLOGICSCHED, 1>;
defm MMX_POR : MMXI_binop_rm_int<0xEB, "por" , int_x86_mmx_por,
MMX_INTALU_ITINS_VECLOGICSCHED, 1>;
defm MMX_PXOR : MMXI_binop_rm_int<0xEF, "pxor", int_x86_mmx_pxor,
MMX_INTALU_ITINS_VECLOGICSCHED, 1>;
defm MMX_PANDN : MMXI_binop_rm_int<0xDF, "pandn", int_x86_mmx_pandn,
MMX_INTALU_ITINS_VECLOGICSCHED>;
// Shift Instructions
defm MMX_PSRLW : MMXI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
int_x86_mmx_psrl_w, int_x86_mmx_psrli_w,
MMX_SHIFT_ITINS>;
defm MMX_PSRLD : MMXI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
int_x86_mmx_psrl_d, int_x86_mmx_psrli_d,
MMX_SHIFT_ITINS>;
defm MMX_PSRLQ : MMXI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
int_x86_mmx_psrl_q, int_x86_mmx_psrli_q,
MMX_SHIFT_ITINS>;
defm MMX_PSLLW : MMXI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
int_x86_mmx_psll_w, int_x86_mmx_pslli_w,
MMX_SHIFT_ITINS>;
defm MMX_PSLLD : MMXI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
int_x86_mmx_psll_d, int_x86_mmx_pslli_d,
MMX_SHIFT_ITINS>;
defm MMX_PSLLQ : MMXI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
int_x86_mmx_psll_q, int_x86_mmx_pslli_q,
MMX_SHIFT_ITINS>;
defm MMX_PSRAW : MMXI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
int_x86_mmx_psra_w, int_x86_mmx_psrai_w,
MMX_SHIFT_ITINS>;
defm MMX_PSRAD : MMXI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
int_x86_mmx_psra_d, int_x86_mmx_psrai_d,
MMX_SHIFT_ITINS>;
// Comparison Instructions
defm MMX_PCMPEQB : MMXI_binop_rm_int<0x74, "pcmpeqb", int_x86_mmx_pcmpeq_b,
MMX_INTALU_ITINS>;
defm MMX_PCMPEQW : MMXI_binop_rm_int<0x75, "pcmpeqw", int_x86_mmx_pcmpeq_w,
MMX_INTALU_ITINS>;
defm MMX_PCMPEQD : MMXI_binop_rm_int<0x76, "pcmpeqd", int_x86_mmx_pcmpeq_d,
MMX_INTALU_ITINS>;
defm MMX_PCMPGTB : MMXI_binop_rm_int<0x64, "pcmpgtb", int_x86_mmx_pcmpgt_b,
MMX_INTALU_ITINS>;
defm MMX_PCMPGTW : MMXI_binop_rm_int<0x65, "pcmpgtw", int_x86_mmx_pcmpgt_w,
MMX_INTALU_ITINS>;
defm MMX_PCMPGTD : MMXI_binop_rm_int<0x66, "pcmpgtd", int_x86_mmx_pcmpgt_d,
MMX_INTALU_ITINS>;
// -- Unpack Instructions
defm MMX_PUNPCKHBW : MMXI_binop_rm_int<0x68, "punpckhbw",
int_x86_mmx_punpckhbw,
MMX_UNPCK_H_ITINS>;
defm MMX_PUNPCKHWD : MMXI_binop_rm_int<0x69, "punpckhwd",
int_x86_mmx_punpckhwd,
MMX_UNPCK_H_ITINS>;
defm MMX_PUNPCKHDQ : MMXI_binop_rm_int<0x6A, "punpckhdq",
int_x86_mmx_punpckhdq,
MMX_UNPCK_H_ITINS>;
defm MMX_PUNPCKLBW : MMXI_binop_rm_int<0x60, "punpcklbw",
int_x86_mmx_punpcklbw,
MMX_UNPCK_L_ITINS,
0, i32mem>;
defm MMX_PUNPCKLWD : MMXI_binop_rm_int<0x61, "punpcklwd",
int_x86_mmx_punpcklwd,
MMX_UNPCK_L_ITINS,
0, i32mem>;
defm MMX_PUNPCKLDQ : MMXI_binop_rm_int<0x62, "punpckldq",
int_x86_mmx_punpckldq,
MMX_UNPCK_L_ITINS,
0, i32mem>;
// -- Pack Instructions
defm MMX_PACKSSWB : MMXI_binop_rm_int<0x63, "packsswb", int_x86_mmx_packsswb,
MMX_PCK_ITINS>;
defm MMX_PACKSSDW : MMXI_binop_rm_int<0x6B, "packssdw", int_x86_mmx_packssdw,
MMX_PCK_ITINS>;
defm MMX_PACKUSWB : MMXI_binop_rm_int<0x67, "packuswb", int_x86_mmx_packuswb,
MMX_PCK_ITINS>;
// -- Shuffle Instructions
defm MMX_PSHUFB : SS3I_binop_rm_int_mm<0x00, "pshufb", int_x86_ssse3_pshuf_b,
MMX_PSHUF_ITINS>;
def MMX_PSHUFWri : MMXIi8<0x70, MRMSrcReg,
(outs VR64:$dst), (ins VR64:$src1, u8imm:$src2),
"pshufw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set VR64:$dst,
(int_x86_sse_pshuf_w VR64:$src1, imm:$src2))],
IIC_MMX_PSHUF>, Sched<[WriteShuffle]>;
def MMX_PSHUFWmi : MMXIi8<0x70, MRMSrcMem,
(outs VR64:$dst), (ins i64mem:$src1, u8imm:$src2),
"pshufw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set VR64:$dst,
(int_x86_sse_pshuf_w (load_mmx addr:$src1),
imm:$src2))],
IIC_MMX_PSHUF>, Sched<[WriteShuffleLd]>;
// -- Conversion Instructions
defm MMX_CVTPS2PI : sse12_cvt_pint<0x2D, VR128, VR64, int_x86_sse_cvtps2pi,
f64mem, load, "cvtps2pi\t{$src, $dst|$dst, $src}",
MMX_CVT_PS_ITINS, SSEPackedSingle>, PS;
defm MMX_CVTPD2PI : sse12_cvt_pint<0x2D, VR128, VR64, int_x86_sse_cvtpd2pi,
f128mem, memop, "cvtpd2pi\t{$src, $dst|$dst, $src}",
MMX_CVT_PD_ITINS, SSEPackedDouble>, PD;
defm MMX_CVTTPS2PI : sse12_cvt_pint<0x2C, VR128, VR64, int_x86_sse_cvttps2pi,
f64mem, load, "cvttps2pi\t{$src, $dst|$dst, $src}",
MMX_CVT_PS_ITINS, SSEPackedSingle>, PS;
defm MMX_CVTTPD2PI : sse12_cvt_pint<0x2C, VR128, VR64, int_x86_sse_cvttpd2pi,
f128mem, memop, "cvttpd2pi\t{$src, $dst|$dst, $src}",
MMX_CVT_PD_ITINS, SSEPackedDouble>, PD;
defm MMX_CVTPI2PD : sse12_cvt_pint<0x2A, VR64, VR128, int_x86_sse_cvtpi2pd,
i64mem, load, "cvtpi2pd\t{$src, $dst|$dst, $src}",
MMX_CVT_PD_ITINS, SSEPackedDouble>, PD;
let Constraints = "$src1 = $dst" in {
defm MMX_CVTPI2PS : sse12_cvt_pint_3addr<0x2A, VR64, VR128,
int_x86_sse_cvtpi2ps,
i64mem, load, "cvtpi2ps\t{$src2, $dst|$dst, $src2}",
SSEPackedSingle>, PS;
}
// Extract / Insert
let Predicates = [HasSSE1] in
def MMX_PEXTRWirri: MMXIi8<0xC5, MRMSrcReg,
(outs GR32orGR64:$dst), (ins VR64:$src1, i32u8imm:$src2),
"pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set GR32orGR64:$dst, (int_x86_mmx_pextr_w VR64:$src1,
imm:$src2))],
IIC_MMX_PEXTR>, Sched<[WriteShuffle]>;
let Constraints = "$src1 = $dst" in {
let Predicates = [HasSSE1] in {
def MMX_PINSRWirri : MMXIi8<0xC4, MRMSrcReg,
(outs VR64:$dst),
(ins VR64:$src1, GR32orGR64:$src2, i32u8imm:$src3),
"pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
[(set VR64:$dst, (int_x86_mmx_pinsr_w VR64:$src1,
GR32orGR64:$src2, imm:$src3))],
IIC_MMX_PINSRW>, Sched<[WriteShuffle]>;
def MMX_PINSRWirmi : MMXIi8<0xC4, MRMSrcMem,
(outs VR64:$dst),
(ins VR64:$src1, i16mem:$src2, i32u8imm:$src3),
"pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
[(set VR64:$dst, (int_x86_mmx_pinsr_w VR64:$src1,
(i32 (anyext (loadi16 addr:$src2))),
imm:$src3))],
IIC_MMX_PINSRW>, Sched<[WriteShuffleLd, ReadAfterLd]>;
}
}
// Mask creation
let Predicates = [HasSSE1] in
def MMX_PMOVMSKBrr : MMXI<0xD7, MRMSrcReg, (outs GR32orGR64:$dst),
(ins VR64:$src),
"pmovmskb\t{$src, $dst|$dst, $src}",
[(set GR32orGR64:$dst,
(int_x86_mmx_pmovmskb VR64:$src))],
IIC_MMX_MOVMSK>, Sched<[WriteVecLogic]>;
// Low word of XMM to MMX.
def MMX_X86movdq2q : SDNode<"X86ISD::MOVDQ2Q", SDTypeProfile<1, 1,
[SDTCisVT<0, x86mmx>, SDTCisVT<1, v2i64>]>>;
def : Pat<(x86mmx (MMX_X86movdq2q VR128:$src)),
(x86mmx (MMX_MOVDQ2Qrr VR128:$src))>;
def : Pat<(x86mmx (MMX_X86movdq2q (loadv2i64 addr:$src))),
(x86mmx (MMX_MOVQ64rm addr:$src))>;
// Misc.
let SchedRW = [WriteShuffle] in {
let Uses = [EDI], Predicates = [HasSSE1,Not64BitMode] in
def MMX_MASKMOVQ : MMXI32<0xF7, MRMSrcReg, (outs), (ins VR64:$src, VR64:$mask),
"maskmovq\t{$mask, $src|$src, $mask}",
[(int_x86_mmx_maskmovq VR64:$src, VR64:$mask, EDI)],
IIC_MMX_MASKMOV>;
let Uses = [RDI], Predicates = [HasSSE1,In64BitMode] in
def MMX_MASKMOVQ64: MMXI64<0xF7, MRMSrcReg, (outs), (ins VR64:$src, VR64:$mask),
"maskmovq\t{$mask, $src|$src, $mask}",
[(int_x86_mmx_maskmovq VR64:$src, VR64:$mask, RDI)],
IIC_MMX_MASKMOV>;
}
// 64-bit bit convert.
let Predicates = [HasSSE2] in {
def : Pat<(f64 (bitconvert (x86mmx VR64:$src))),
(MMX_MOVQ2FR64rr VR64:$src)>;
def : Pat<(x86mmx (bitconvert (f64 FR64:$src))),
(MMX_MOVFR642Qrr FR64:$src)>;
def : Pat<(x86mmx (MMX_X86movdq2q
(bc_v2i64 (v4i32 (int_x86_sse2_cvtps2dq VR128:$src))))),
(MMX_CVTPS2PIirr VR128:$src)>;
def : Pat<(x86mmx (MMX_X86movdq2q
(bc_v2i64 (v4i32 (fp_to_sint (v4f32 VR128:$src)))))),
(MMX_CVTTPS2PIirr VR128:$src)>;
def : Pat<(x86mmx (MMX_X86movdq2q
(bc_v2i64 (v4i32 (X86cvtp2Int (v2f64 VR128:$src)))))),
(MMX_CVTPD2PIirr VR128:$src)>;
def : Pat<(x86mmx (MMX_X86movdq2q
(bc_v2i64 (v4i32 (X86cvttp2si (v2f64 VR128:$src)))))),
(MMX_CVTTPD2PIirr VR128:$src)>;
}